]> CyberLeo.Net >> Repos - FreeBSD/stable/10.git/blob - sys/cam/ctl/ctl.c
MFC r271949:
[FreeBSD/stable/10.git] / sys / cam / ctl / ctl.c
1 /*-
2  * Copyright (c) 2003-2009 Silicon Graphics International Corp.
3  * Copyright (c) 2012 The FreeBSD Foundation
4  * All rights reserved.
5  *
6  * Portions of this software were developed by Edward Tomasz Napierala
7  * under sponsorship from the FreeBSD Foundation.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions, and the following disclaimer,
14  *    without modification.
15  * 2. Redistributions in binary form must reproduce at minimum a disclaimer
16  *    substantially similar to the "NO WARRANTY" disclaimer below
17  *    ("Disclaimer") and any redistribution must be conditioned upon
18  *    including a substantially similar Disclaimer requirement for further
19  *    binary redistribution.
20  *
21  * NO WARRANTY
22  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
23  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
24  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR
25  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
26  * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
30  * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
31  * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
32  * POSSIBILITY OF SUCH DAMAGES.
33  *
34  * $Id: //depot/users/kenm/FreeBSD-test2/sys/cam/ctl/ctl.c#8 $
35  */
36 /*
37  * CAM Target Layer, a SCSI device emulation subsystem.
38  *
39  * Author: Ken Merry <ken@FreeBSD.org>
40  */
41
42 #define _CTL_C
43
44 #include <sys/cdefs.h>
45 __FBSDID("$FreeBSD$");
46
47 #include <sys/param.h>
48 #include <sys/systm.h>
49 #include <sys/kernel.h>
50 #include <sys/types.h>
51 #include <sys/kthread.h>
52 #include <sys/bio.h>
53 #include <sys/fcntl.h>
54 #include <sys/lock.h>
55 #include <sys/module.h>
56 #include <sys/mutex.h>
57 #include <sys/condvar.h>
58 #include <sys/malloc.h>
59 #include <sys/conf.h>
60 #include <sys/ioccom.h>
61 #include <sys/queue.h>
62 #include <sys/sbuf.h>
63 #include <sys/smp.h>
64 #include <sys/endian.h>
65 #include <sys/sysctl.h>
66
67 #include <cam/cam.h>
68 #include <cam/scsi/scsi_all.h>
69 #include <cam/scsi/scsi_da.h>
70 #include <cam/ctl/ctl_io.h>
71 #include <cam/ctl/ctl.h>
72 #include <cam/ctl/ctl_frontend.h>
73 #include <cam/ctl/ctl_frontend_internal.h>
74 #include <cam/ctl/ctl_util.h>
75 #include <cam/ctl/ctl_backend.h>
76 #include <cam/ctl/ctl_ioctl.h>
77 #include <cam/ctl/ctl_ha.h>
78 #include <cam/ctl/ctl_private.h>
79 #include <cam/ctl/ctl_debug.h>
80 #include <cam/ctl/ctl_scsi_all.h>
81 #include <cam/ctl/ctl_error.h>
82
83 struct ctl_softc *control_softc = NULL;
84
85 /*
86  * Size and alignment macros needed for Copan-specific HA hardware.  These
87  * can go away when the HA code is re-written, and uses busdma for any
88  * hardware.
89  */
90 #define CTL_ALIGN_8B(target, source, type)                              \
91         if (((uint32_t)source & 0x7) != 0)                              \
92                 target = (type)(source + (0x8 - ((uint32_t)source & 0x7)));\
93         else                                                            \
94                 target = (type)source;
95
96 #define CTL_SIZE_8B(target, size)                                       \
97         if ((size & 0x7) != 0)                                          \
98                 target = size + (0x8 - (size & 0x7));                   \
99         else                                                            \
100                 target = size;
101
102 #define CTL_ALIGN_8B_MARGIN     16
103
104 /*
105  * Template mode pages.
106  */
107
108 /*
109  * Note that these are default values only.  The actual values will be
110  * filled in when the user does a mode sense.
111  */
112 static struct copan_power_subpage power_page_default = {
113         /*page_code*/ PWR_PAGE_CODE | SMPH_SPF,
114         /*subpage*/ PWR_SUBPAGE_CODE,
115         /*page_length*/ {(sizeof(struct copan_power_subpage) - 4) & 0xff00,
116                          (sizeof(struct copan_power_subpage) - 4) & 0x00ff},
117         /*page_version*/ PWR_VERSION,
118         /* total_luns */ 26,
119         /* max_active_luns*/ PWR_DFLT_MAX_LUNS,
120         /*reserved*/ {0, 0, 0, 0, 0, 0, 0, 0, 0,
121                       0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
122                       0, 0, 0, 0, 0, 0}
123 };
124
125 static struct copan_power_subpage power_page_changeable = {
126         /*page_code*/ PWR_PAGE_CODE | SMPH_SPF,
127         /*subpage*/ PWR_SUBPAGE_CODE,
128         /*page_length*/ {(sizeof(struct copan_power_subpage) - 4) & 0xff00,
129                          (sizeof(struct copan_power_subpage) - 4) & 0x00ff},
130         /*page_version*/ 0,
131         /* total_luns */ 0,
132         /* max_active_luns*/ 0,
133         /*reserved*/ {0, 0, 0, 0, 0, 0, 0, 0, 0,
134                       0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
135                       0, 0, 0, 0, 0, 0}
136 };
137
138 static struct copan_aps_subpage aps_page_default = {
139         APS_PAGE_CODE | SMPH_SPF, //page_code
140         APS_SUBPAGE_CODE, //subpage
141         {(sizeof(struct copan_aps_subpage) - 4) & 0xff00,
142          (sizeof(struct copan_aps_subpage) - 4) & 0x00ff}, //page_length
143         APS_VERSION, //page_version
144         0, //lock_active
145         {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
146         0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
147         0, 0, 0, 0, 0} //reserved
148 };
149
150 static struct copan_aps_subpage aps_page_changeable = {
151         APS_PAGE_CODE | SMPH_SPF, //page_code
152         APS_SUBPAGE_CODE, //subpage
153         {(sizeof(struct copan_aps_subpage) - 4) & 0xff00,
154          (sizeof(struct copan_aps_subpage) - 4) & 0x00ff}, //page_length
155         0, //page_version
156         0, //lock_active
157         {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
158         0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
159         0, 0, 0, 0, 0} //reserved
160 };
161
162 static struct copan_debugconf_subpage debugconf_page_default = {
163         DBGCNF_PAGE_CODE | SMPH_SPF,    /* page_code */
164         DBGCNF_SUBPAGE_CODE,            /* subpage */
165         {(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
166          (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
167         DBGCNF_VERSION,                 /* page_version */
168         {CTL_TIME_IO_DEFAULT_SECS>>8,
169          CTL_TIME_IO_DEFAULT_SECS>>0},  /* ctl_time_io_secs */
170 };
171
172 static struct copan_debugconf_subpage debugconf_page_changeable = {
173         DBGCNF_PAGE_CODE | SMPH_SPF,    /* page_code */
174         DBGCNF_SUBPAGE_CODE,            /* subpage */
175         {(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
176          (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
177         0,                              /* page_version */
178         {0xff,0xff},                    /* ctl_time_io_secs */
179 };
180
181 static struct scsi_format_page format_page_default = {
182         /*page_code*/SMS_FORMAT_DEVICE_PAGE,
183         /*page_length*/sizeof(struct scsi_format_page) - 2,
184         /*tracks_per_zone*/ {0, 0},
185         /*alt_sectors_per_zone*/ {0, 0},
186         /*alt_tracks_per_zone*/ {0, 0},
187         /*alt_tracks_per_lun*/ {0, 0},
188         /*sectors_per_track*/ {(CTL_DEFAULT_SECTORS_PER_TRACK >> 8) & 0xff,
189                                 CTL_DEFAULT_SECTORS_PER_TRACK & 0xff},
190         /*bytes_per_sector*/ {0, 0},
191         /*interleave*/ {0, 0},
192         /*track_skew*/ {0, 0},
193         /*cylinder_skew*/ {0, 0},
194         /*flags*/ SFP_HSEC,
195         /*reserved*/ {0, 0, 0}
196 };
197
198 static struct scsi_format_page format_page_changeable = {
199         /*page_code*/SMS_FORMAT_DEVICE_PAGE,
200         /*page_length*/sizeof(struct scsi_format_page) - 2,
201         /*tracks_per_zone*/ {0, 0},
202         /*alt_sectors_per_zone*/ {0, 0},
203         /*alt_tracks_per_zone*/ {0, 0},
204         /*alt_tracks_per_lun*/ {0, 0},
205         /*sectors_per_track*/ {0, 0},
206         /*bytes_per_sector*/ {0, 0},
207         /*interleave*/ {0, 0},
208         /*track_skew*/ {0, 0},
209         /*cylinder_skew*/ {0, 0},
210         /*flags*/ 0,
211         /*reserved*/ {0, 0, 0}
212 };
213
214 static struct scsi_rigid_disk_page rigid_disk_page_default = {
215         /*page_code*/SMS_RIGID_DISK_PAGE,
216         /*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
217         /*cylinders*/ {0, 0, 0},
218         /*heads*/ CTL_DEFAULT_HEADS,
219         /*start_write_precomp*/ {0, 0, 0},
220         /*start_reduced_current*/ {0, 0, 0},
221         /*step_rate*/ {0, 0},
222         /*landing_zone_cylinder*/ {0, 0, 0},
223         /*rpl*/ SRDP_RPL_DISABLED,
224         /*rotational_offset*/ 0,
225         /*reserved1*/ 0,
226         /*rotation_rate*/ {(CTL_DEFAULT_ROTATION_RATE >> 8) & 0xff,
227                            CTL_DEFAULT_ROTATION_RATE & 0xff},
228         /*reserved2*/ {0, 0}
229 };
230
231 static struct scsi_rigid_disk_page rigid_disk_page_changeable = {
232         /*page_code*/SMS_RIGID_DISK_PAGE,
233         /*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
234         /*cylinders*/ {0, 0, 0},
235         /*heads*/ 0,
236         /*start_write_precomp*/ {0, 0, 0},
237         /*start_reduced_current*/ {0, 0, 0},
238         /*step_rate*/ {0, 0},
239         /*landing_zone_cylinder*/ {0, 0, 0},
240         /*rpl*/ 0,
241         /*rotational_offset*/ 0,
242         /*reserved1*/ 0,
243         /*rotation_rate*/ {0, 0},
244         /*reserved2*/ {0, 0}
245 };
246
247 static struct scsi_caching_page caching_page_default = {
248         /*page_code*/SMS_CACHING_PAGE,
249         /*page_length*/sizeof(struct scsi_caching_page) - 2,
250         /*flags1*/ SCP_DISC | SCP_WCE,
251         /*ret_priority*/ 0,
252         /*disable_pf_transfer_len*/ {0xff, 0xff},
253         /*min_prefetch*/ {0, 0},
254         /*max_prefetch*/ {0xff, 0xff},
255         /*max_pf_ceiling*/ {0xff, 0xff},
256         /*flags2*/ 0,
257         /*cache_segments*/ 0,
258         /*cache_seg_size*/ {0, 0},
259         /*reserved*/ 0,
260         /*non_cache_seg_size*/ {0, 0, 0}
261 };
262
263 static struct scsi_caching_page caching_page_changeable = {
264         /*page_code*/SMS_CACHING_PAGE,
265         /*page_length*/sizeof(struct scsi_caching_page) - 2,
266         /*flags1*/ 0,
267         /*ret_priority*/ 0,
268         /*disable_pf_transfer_len*/ {0, 0},
269         /*min_prefetch*/ {0, 0},
270         /*max_prefetch*/ {0, 0},
271         /*max_pf_ceiling*/ {0, 0},
272         /*flags2*/ 0,
273         /*cache_segments*/ 0,
274         /*cache_seg_size*/ {0, 0},
275         /*reserved*/ 0,
276         /*non_cache_seg_size*/ {0, 0, 0}
277 };
278
279 static struct scsi_control_page control_page_default = {
280         /*page_code*/SMS_CONTROL_MODE_PAGE,
281         /*page_length*/sizeof(struct scsi_control_page) - 2,
282         /*rlec*/0,
283         /*queue_flags*/0,
284         /*eca_and_aen*/0,
285         /*flags4*/SCP_TAS,
286         /*aen_holdoff_period*/{0, 0},
287         /*busy_timeout_period*/{0, 0},
288         /*extended_selftest_completion_time*/{0, 0}
289 };
290
291 static struct scsi_control_page control_page_changeable = {
292         /*page_code*/SMS_CONTROL_MODE_PAGE,
293         /*page_length*/sizeof(struct scsi_control_page) - 2,
294         /*rlec*/SCP_DSENSE,
295         /*queue_flags*/0,
296         /*eca_and_aen*/0,
297         /*flags4*/0,
298         /*aen_holdoff_period*/{0, 0},
299         /*busy_timeout_period*/{0, 0},
300         /*extended_selftest_completion_time*/{0, 0}
301 };
302
303
304 /*
305  * XXX KDM move these into the softc.
306  */
307 static int rcv_sync_msg;
308 static int persis_offset;
309 static uint8_t ctl_pause_rtr;
310 static int     ctl_is_single = 1;
311 static int     index_to_aps_page;
312
313 SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD, 0, "CAM Target Layer");
314 static int worker_threads = -1;
315 TUNABLE_INT("kern.cam.ctl.worker_threads", &worker_threads);
316 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, worker_threads, CTLFLAG_RDTUN,
317     &worker_threads, 1, "Number of worker threads");
318 static int verbose = 0;
319 TUNABLE_INT("kern.cam.ctl.verbose", &verbose);
320 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, verbose, CTLFLAG_RWTUN,
321     &verbose, 0, "Show SCSI errors returned to initiator");
322
323 /*
324  * Supported pages (0x00), Serial number (0x80), Device ID (0x83),
325  * SCSI Ports (0x88), Third-party Copy (0x8F), Block limits (0xB0),
326  * Block Device Characteristics (0xB1) and Logical Block Provisioning (0xB2)
327  */
328 #define SCSI_EVPD_NUM_SUPPORTED_PAGES   8
329
330 static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
331                                   int param);
332 static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
333 static int ctl_init(void);
334 void ctl_shutdown(void);
335 static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
336 static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
337 static void ctl_ioctl_online(void *arg);
338 static void ctl_ioctl_offline(void *arg);
339 static int ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id);
340 static int ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id);
341 static int ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio);
342 static int ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio);
343 static int ctl_ioctl_submit_wait(union ctl_io *io);
344 static void ctl_ioctl_datamove(union ctl_io *io);
345 static void ctl_ioctl_done(union ctl_io *io);
346 static void ctl_ioctl_hard_startstop_callback(void *arg,
347                                               struct cfi_metatask *metatask);
348 static void ctl_ioctl_bbrread_callback(void *arg,struct cfi_metatask *metatask);
349 static int ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
350                               struct ctl_ooa *ooa_hdr,
351                               struct ctl_ooa_entry *kern_entries);
352 static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
353                      struct thread *td);
354 static uint32_t ctl_map_lun(int port_num, uint32_t lun);
355 static uint32_t ctl_map_lun_back(int port_num, uint32_t lun);
356 #ifdef unused
357 static union ctl_io *ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port,
358                                    uint32_t targ_target, uint32_t targ_lun,
359                                    int can_wait);
360 static void ctl_kfree_io(union ctl_io *io);
361 #endif /* unused */
362 static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
363                          struct ctl_be_lun *be_lun, struct ctl_id target_id);
364 static int ctl_free_lun(struct ctl_lun *lun);
365 static void ctl_create_lun(struct ctl_be_lun *be_lun);
366 /**
367 static void ctl_failover_change_pages(struct ctl_softc *softc,
368                                       struct ctl_scsiio *ctsio, int master);
369 **/
370
371 static int ctl_do_mode_select(union ctl_io *io);
372 static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
373                            uint64_t res_key, uint64_t sa_res_key,
374                            uint8_t type, uint32_t residx,
375                            struct ctl_scsiio *ctsio,
376                            struct scsi_per_res_out *cdb,
377                            struct scsi_per_res_out_parms* param);
378 static void ctl_pro_preempt_other(struct ctl_lun *lun,
379                                   union ctl_ha_msg *msg);
380 static void ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg);
381 static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
382 static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
383 static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
384 static int ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio,
385                                          int alloc_len);
386 static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
387                                          int alloc_len);
388 static int ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len);
389 static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
390 static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
391 static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
392 static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint32_t *len);
393 static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2);
394 static ctl_action ctl_check_for_blockage(union ctl_io *pending_io,
395                                          union ctl_io *ooa_io);
396 static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
397                                 union ctl_io *starting_io);
398 static int ctl_check_blocked(struct ctl_lun *lun);
399 static int ctl_scsiio_lun_check(struct ctl_softc *ctl_softc,
400                                 struct ctl_lun *lun,
401                                 const struct ctl_cmd_entry *entry,
402                                 struct ctl_scsiio *ctsio);
403 //static int ctl_check_rtr(union ctl_io *pending_io, struct ctl_softc *softc);
404 static void ctl_failover(void);
405 static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
406                                struct ctl_scsiio *ctsio);
407 static int ctl_scsiio(struct ctl_scsiio *ctsio);
408
409 static int ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
410 static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
411                             ctl_ua_type ua_type);
412 static int ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io,
413                          ctl_ua_type ua_type);
414 static int ctl_abort_task(union ctl_io *io);
415 static int ctl_abort_task_set(union ctl_io *io);
416 static int ctl_i_t_nexus_reset(union ctl_io *io);
417 static void ctl_run_task(union ctl_io *io);
418 #ifdef CTL_IO_DELAY
419 static void ctl_datamove_timer_wakeup(void *arg);
420 static void ctl_done_timer_wakeup(void *arg);
421 #endif /* CTL_IO_DELAY */
422
423 static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
424 static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
425 static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
426 static void ctl_datamove_remote_write(union ctl_io *io);
427 static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
428 static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
429 static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
430 static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
431                                     ctl_ha_dt_cb callback);
432 static void ctl_datamove_remote_read(union ctl_io *io);
433 static void ctl_datamove_remote(union ctl_io *io);
434 static int ctl_process_done(union ctl_io *io);
435 static void ctl_lun_thread(void *arg);
436 static void ctl_work_thread(void *arg);
437 static void ctl_enqueue_incoming(union ctl_io *io);
438 static void ctl_enqueue_rtr(union ctl_io *io);
439 static void ctl_enqueue_done(union ctl_io *io);
440 static void ctl_enqueue_isc(union ctl_io *io);
441 static const struct ctl_cmd_entry *
442     ctl_get_cmd_entry(struct ctl_scsiio *ctsio);
443 static const struct ctl_cmd_entry *
444     ctl_validate_command(struct ctl_scsiio *ctsio);
445 static int ctl_cmd_applicable(uint8_t lun_type,
446     const struct ctl_cmd_entry *entry);
447
448 /*
449  * Load the serialization table.  This isn't very pretty, but is probably
450  * the easiest way to do it.
451  */
452 #include "ctl_ser_table.c"
453
454 /*
455  * We only need to define open, close and ioctl routines for this driver.
456  */
457 static struct cdevsw ctl_cdevsw = {
458         .d_version =    D_VERSION,
459         .d_flags =      0,
460         .d_open =       ctl_open,
461         .d_close =      ctl_close,
462         .d_ioctl =      ctl_ioctl,
463         .d_name =       "ctl",
464 };
465
466
467 MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
468 MALLOC_DEFINE(M_CTLIO, "ctlio", "Memory used for CTL requests");
469
470 static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
471
472 static moduledata_t ctl_moduledata = {
473         "ctl",
474         ctl_module_event_handler,
475         NULL
476 };
477
478 DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
479 MODULE_VERSION(ctl, 1);
480
481 static struct ctl_frontend ioctl_frontend =
482 {
483         .name = "ioctl",
484 };
485
486 static void
487 ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
488                             union ctl_ha_msg *msg_info)
489 {
490         struct ctl_scsiio *ctsio;
491
492         if (msg_info->hdr.original_sc == NULL) {
493                 printf("%s: original_sc == NULL!\n", __func__);
494                 /* XXX KDM now what? */
495                 return;
496         }
497
498         ctsio = &msg_info->hdr.original_sc->scsiio;
499         ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
500         ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
501         ctsio->io_hdr.status = msg_info->hdr.status;
502         ctsio->scsi_status = msg_info->scsi.scsi_status;
503         ctsio->sense_len = msg_info->scsi.sense_len;
504         ctsio->sense_residual = msg_info->scsi.sense_residual;
505         ctsio->residual = msg_info->scsi.residual;
506         memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
507                sizeof(ctsio->sense_data));
508         memcpy(&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
509                &msg_info->scsi.lbalen, sizeof(msg_info->scsi.lbalen));
510         ctl_enqueue_isc((union ctl_io *)ctsio);
511 }
512
513 static void
514 ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
515                                 union ctl_ha_msg *msg_info)
516 {
517         struct ctl_scsiio *ctsio;
518
519         if (msg_info->hdr.serializing_sc == NULL) {
520                 printf("%s: serializing_sc == NULL!\n", __func__);
521                 /* XXX KDM now what? */
522                 return;
523         }
524
525         ctsio = &msg_info->hdr.serializing_sc->scsiio;
526 #if 0
527         /*
528          * Attempt to catch the situation where an I/O has
529          * been freed, and we're using it again.
530          */
531         if (ctsio->io_hdr.io_type == 0xff) {
532                 union ctl_io *tmp_io;
533                 tmp_io = (union ctl_io *)ctsio;
534                 printf("%s: %p use after free!\n", __func__,
535                        ctsio);
536                 printf("%s: type %d msg %d cdb %x iptl: "
537                        "%d:%d:%d:%d tag 0x%04x "
538                        "flag %#x status %x\n",
539                         __func__,
540                         tmp_io->io_hdr.io_type,
541                         tmp_io->io_hdr.msg_type,
542                         tmp_io->scsiio.cdb[0],
543                         tmp_io->io_hdr.nexus.initid.id,
544                         tmp_io->io_hdr.nexus.targ_port,
545                         tmp_io->io_hdr.nexus.targ_target.id,
546                         tmp_io->io_hdr.nexus.targ_lun,
547                         (tmp_io->io_hdr.io_type ==
548                         CTL_IO_TASK) ?
549                         tmp_io->taskio.tag_num :
550                         tmp_io->scsiio.tag_num,
551                         tmp_io->io_hdr.flags,
552                         tmp_io->io_hdr.status);
553         }
554 #endif
555         ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
556         ctl_enqueue_isc((union ctl_io *)ctsio);
557 }
558
559 /*
560  * ISC (Inter Shelf Communication) event handler.  Events from the HA
561  * subsystem come in here.
562  */
563 static void
564 ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
565 {
566         struct ctl_softc *ctl_softc;
567         union ctl_io *io;
568         struct ctl_prio *presio;
569         ctl_ha_status isc_status;
570
571         ctl_softc = control_softc;
572         io = NULL;
573
574
575 #if 0
576         printf("CTL: Isc Msg event %d\n", event);
577 #endif
578         if (event == CTL_HA_EVT_MSG_RECV) {
579                 union ctl_ha_msg msg_info;
580
581                 isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
582                                              sizeof(msg_info), /*wait*/ 0);
583 #if 0
584                 printf("CTL: msg_type %d\n", msg_info.msg_type);
585 #endif
586                 if (isc_status != 0) {
587                         printf("Error receiving message, status = %d\n",
588                                isc_status);
589                         return;
590                 }
591
592                 switch (msg_info.hdr.msg_type) {
593                 case CTL_MSG_SERIALIZE:
594 #if 0
595                         printf("Serialize\n");
596 #endif
597                         io = ctl_alloc_io((void *)ctl_softc->othersc_pool);
598                         if (io == NULL) {
599                                 printf("ctl_isc_event_handler: can't allocate "
600                                        "ctl_io!\n");
601                                 /* Bad Juju */
602                                 /* Need to set busy and send msg back */
603                                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
604                                 msg_info.hdr.status = CTL_SCSI_ERROR;
605                                 msg_info.scsi.scsi_status = SCSI_STATUS_BUSY;
606                                 msg_info.scsi.sense_len = 0;
607                                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
608                                     sizeof(msg_info), 0) > CTL_HA_STATUS_SUCCESS){
609                                 }
610                                 goto bailout;
611                         }
612                         ctl_zero_io(io);
613                         // populate ctsio from msg_info
614                         io->io_hdr.io_type = CTL_IO_SCSI;
615                         io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
616                         io->io_hdr.original_sc = msg_info.hdr.original_sc;
617 #if 0
618                         printf("pOrig %x\n", (int)msg_info.original_sc);
619 #endif
620                         io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
621                                             CTL_FLAG_IO_ACTIVE;
622                         /*
623                          * If we're in serialization-only mode, we don't
624                          * want to go through full done processing.  Thus
625                          * the COPY flag.
626                          *
627                          * XXX KDM add another flag that is more specific.
628                          */
629                         if (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)
630                                 io->io_hdr.flags |= CTL_FLAG_INT_COPY;
631                         io->io_hdr.nexus = msg_info.hdr.nexus;
632 #if 0
633                         printf("targ %d, port %d, iid %d, lun %d\n",
634                                io->io_hdr.nexus.targ_target.id,
635                                io->io_hdr.nexus.targ_port,
636                                io->io_hdr.nexus.initid.id,
637                                io->io_hdr.nexus.targ_lun);
638 #endif
639                         io->scsiio.tag_num = msg_info.scsi.tag_num;
640                         io->scsiio.tag_type = msg_info.scsi.tag_type;
641                         memcpy(io->scsiio.cdb, msg_info.scsi.cdb,
642                                CTL_MAX_CDBLEN);
643                         if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
644                                 const struct ctl_cmd_entry *entry;
645
646                                 entry = ctl_get_cmd_entry(&io->scsiio);
647                                 io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
648                                 io->io_hdr.flags |=
649                                         entry->flags & CTL_FLAG_DATA_MASK;
650                         }
651                         ctl_enqueue_isc(io);
652                         break;
653
654                 /* Performed on the Originating SC, XFER mode only */
655                 case CTL_MSG_DATAMOVE: {
656                         struct ctl_sg_entry *sgl;
657                         int i, j;
658
659                         io = msg_info.hdr.original_sc;
660                         if (io == NULL) {
661                                 printf("%s: original_sc == NULL!\n", __func__);
662                                 /* XXX KDM do something here */
663                                 break;
664                         }
665                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
666                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
667                         /*
668                          * Keep track of this, we need to send it back over
669                          * when the datamove is complete.
670                          */
671                         io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
672
673                         if (msg_info.dt.sg_sequence == 0) {
674                                 /*
675                                  * XXX KDM we use the preallocated S/G list
676                                  * here, but we'll need to change this to
677                                  * dynamic allocation if we need larger S/G
678                                  * lists.
679                                  */
680                                 if (msg_info.dt.kern_sg_entries >
681                                     sizeof(io->io_hdr.remote_sglist) /
682                                     sizeof(io->io_hdr.remote_sglist[0])) {
683                                         printf("%s: number of S/G entries "
684                                             "needed %u > allocated num %zd\n",
685                                             __func__,
686                                             msg_info.dt.kern_sg_entries,
687                                             sizeof(io->io_hdr.remote_sglist)/
688                                             sizeof(io->io_hdr.remote_sglist[0]));
689                                 
690                                         /*
691                                          * XXX KDM send a message back to
692                                          * the other side to shut down the
693                                          * DMA.  The error will come back
694                                          * through via the normal channel.
695                                          */
696                                         break;
697                                 }
698                                 sgl = io->io_hdr.remote_sglist;
699                                 memset(sgl, 0,
700                                        sizeof(io->io_hdr.remote_sglist));
701
702                                 io->scsiio.kern_data_ptr = (uint8_t *)sgl;
703
704                                 io->scsiio.kern_sg_entries =
705                                         msg_info.dt.kern_sg_entries;
706                                 io->scsiio.rem_sg_entries =
707                                         msg_info.dt.kern_sg_entries;
708                                 io->scsiio.kern_data_len =
709                                         msg_info.dt.kern_data_len;
710                                 io->scsiio.kern_total_len =
711                                         msg_info.dt.kern_total_len;
712                                 io->scsiio.kern_data_resid =
713                                         msg_info.dt.kern_data_resid;
714                                 io->scsiio.kern_rel_offset =
715                                         msg_info.dt.kern_rel_offset;
716                                 /*
717                                  * Clear out per-DMA flags.
718                                  */
719                                 io->io_hdr.flags &= ~CTL_FLAG_RDMA_MASK;
720                                 /*
721                                  * Add per-DMA flags that are set for this
722                                  * particular DMA request.
723                                  */
724                                 io->io_hdr.flags |= msg_info.dt.flags &
725                                                     CTL_FLAG_RDMA_MASK;
726                         } else
727                                 sgl = (struct ctl_sg_entry *)
728                                         io->scsiio.kern_data_ptr;
729
730                         for (i = msg_info.dt.sent_sg_entries, j = 0;
731                              i < (msg_info.dt.sent_sg_entries +
732                              msg_info.dt.cur_sg_entries); i++, j++) {
733                                 sgl[i].addr = msg_info.dt.sg_list[j].addr;
734                                 sgl[i].len = msg_info.dt.sg_list[j].len;
735
736 #if 0
737                                 printf("%s: L: %p,%d -> %p,%d j=%d, i=%d\n",
738                                        __func__,
739                                        msg_info.dt.sg_list[j].addr,
740                                        msg_info.dt.sg_list[j].len,
741                                        sgl[i].addr, sgl[i].len, j, i);
742 #endif
743                         }
744 #if 0
745                         memcpy(&sgl[msg_info.dt.sent_sg_entries],
746                                msg_info.dt.sg_list,
747                                sizeof(*sgl) * msg_info.dt.cur_sg_entries);
748 #endif
749
750                         /*
751                          * If this is the last piece of the I/O, we've got
752                          * the full S/G list.  Queue processing in the thread.
753                          * Otherwise wait for the next piece.
754                          */
755                         if (msg_info.dt.sg_last != 0)
756                                 ctl_enqueue_isc(io);
757                         break;
758                 }
759                 /* Performed on the Serializing (primary) SC, XFER mode only */
760                 case CTL_MSG_DATAMOVE_DONE: {
761                         if (msg_info.hdr.serializing_sc == NULL) {
762                                 printf("%s: serializing_sc == NULL!\n",
763                                        __func__);
764                                 /* XXX KDM now what? */
765                                 break;
766                         }
767                         /*
768                          * We grab the sense information here in case
769                          * there was a failure, so we can return status
770                          * back to the initiator.
771                          */
772                         io = msg_info.hdr.serializing_sc;
773                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
774                         io->io_hdr.status = msg_info.hdr.status;
775                         io->scsiio.scsi_status = msg_info.scsi.scsi_status;
776                         io->scsiio.sense_len = msg_info.scsi.sense_len;
777                         io->scsiio.sense_residual =msg_info.scsi.sense_residual;
778                         io->io_hdr.port_status = msg_info.scsi.fetd_status;
779                         io->scsiio.residual = msg_info.scsi.residual;
780                         memcpy(&io->scsiio.sense_data,&msg_info.scsi.sense_data,
781                                sizeof(io->scsiio.sense_data));
782                         ctl_enqueue_isc(io);
783                         break;
784                 }
785
786                 /* Preformed on Originating SC, SER_ONLY mode */
787                 case CTL_MSG_R2R:
788                         io = msg_info.hdr.original_sc;
789                         if (io == NULL) {
790                                 printf("%s: Major Bummer\n", __func__);
791                                 return;
792                         } else {
793 #if 0
794                                 printf("pOrig %x\n",(int) ctsio);
795 #endif
796                         }
797                         io->io_hdr.msg_type = CTL_MSG_R2R;
798                         io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
799                         ctl_enqueue_isc(io);
800                         break;
801
802                 /*
803                  * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
804                  * mode.
805                  * Performed on the Originating (i.e. secondary) SC in XFER
806                  * mode
807                  */
808                 case CTL_MSG_FINISH_IO:
809                         if (ctl_softc->ha_mode == CTL_HA_MODE_XFER)
810                                 ctl_isc_handler_finish_xfer(ctl_softc,
811                                                             &msg_info);
812                         else
813                                 ctl_isc_handler_finish_ser_only(ctl_softc,
814                                                                 &msg_info);
815                         break;
816
817                 /* Preformed on Originating SC */
818                 case CTL_MSG_BAD_JUJU:
819                         io = msg_info.hdr.original_sc;
820                         if (io == NULL) {
821                                 printf("%s: Bad JUJU!, original_sc is NULL!\n",
822                                        __func__);
823                                 break;
824                         }
825                         ctl_copy_sense_data(&msg_info, io);
826                         /*
827                          * IO should have already been cleaned up on other
828                          * SC so clear this flag so we won't send a message
829                          * back to finish the IO there.
830                          */
831                         io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
832                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
833
834                         /* io = msg_info.hdr.serializing_sc; */
835                         io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
836                         ctl_enqueue_isc(io);
837                         break;
838
839                 /* Handle resets sent from the other side */
840                 case CTL_MSG_MANAGE_TASKS: {
841                         struct ctl_taskio *taskio;
842                         taskio = (struct ctl_taskio *)ctl_alloc_io(
843                                 (void *)ctl_softc->othersc_pool);
844                         if (taskio == NULL) {
845                                 printf("ctl_isc_event_handler: can't allocate "
846                                        "ctl_io!\n");
847                                 /* Bad Juju */
848                                 /* should I just call the proper reset func
849                                    here??? */
850                                 goto bailout;
851                         }
852                         ctl_zero_io((union ctl_io *)taskio);
853                         taskio->io_hdr.io_type = CTL_IO_TASK;
854                         taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
855                         taskio->io_hdr.nexus = msg_info.hdr.nexus;
856                         taskio->task_action = msg_info.task.task_action;
857                         taskio->tag_num = msg_info.task.tag_num;
858                         taskio->tag_type = msg_info.task.tag_type;
859 #ifdef CTL_TIME_IO
860                         taskio->io_hdr.start_time = time_uptime;
861                         getbintime(&taskio->io_hdr.start_bt);
862 #if 0
863                         cs_prof_gettime(&taskio->io_hdr.start_ticks);
864 #endif
865 #endif /* CTL_TIME_IO */
866                         ctl_run_task((union ctl_io *)taskio);
867                         break;
868                 }
869                 /* Persistent Reserve action which needs attention */
870                 case CTL_MSG_PERS_ACTION:
871                         presio = (struct ctl_prio *)ctl_alloc_io(
872                                 (void *)ctl_softc->othersc_pool);
873                         if (presio == NULL) {
874                                 printf("ctl_isc_event_handler: can't allocate "
875                                        "ctl_io!\n");
876                                 /* Bad Juju */
877                                 /* Need to set busy and send msg back */
878                                 goto bailout;
879                         }
880                         ctl_zero_io((union ctl_io *)presio);
881                         presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
882                         presio->pr_msg = msg_info.pr;
883                         ctl_enqueue_isc((union ctl_io *)presio);
884                         break;
885                 case CTL_MSG_SYNC_FE:
886                         rcv_sync_msg = 1;
887                         break;
888                 case CTL_MSG_APS_LOCK: {
889                         // It's quicker to execute this then to
890                         // queue it.
891                         struct ctl_lun *lun;
892                         struct ctl_page_index *page_index;
893                         struct copan_aps_subpage *current_sp;
894                         uint32_t targ_lun;
895
896                         targ_lun = msg_info.hdr.nexus.targ_mapped_lun;
897                         lun = ctl_softc->ctl_luns[targ_lun];
898                         mtx_lock(&lun->lun_lock);
899                         page_index = &lun->mode_pages.index[index_to_aps_page];
900                         current_sp = (struct copan_aps_subpage *)
901                                      (page_index->page_data +
902                                      (page_index->page_len * CTL_PAGE_CURRENT));
903
904                         current_sp->lock_active = msg_info.aps.lock_flag;
905                         mtx_unlock(&lun->lun_lock);
906                         break;
907                 }
908                 default:
909                         printf("How did I get here?\n");
910                 }
911         } else if (event == CTL_HA_EVT_MSG_SENT) {
912                 if (param != CTL_HA_STATUS_SUCCESS) {
913                         printf("Bad status from ctl_ha_msg_send status %d\n",
914                                param);
915                 }
916                 return;
917         } else if (event == CTL_HA_EVT_DISCONNECT) {
918                 printf("CTL: Got a disconnect from Isc\n");
919                 return;
920         } else {
921                 printf("ctl_isc_event_handler: Unknown event %d\n", event);
922                 return;
923         }
924
925 bailout:
926         return;
927 }
928
929 static void
930 ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
931 {
932         struct scsi_sense_data *sense;
933
934         sense = &dest->scsiio.sense_data;
935         bcopy(&src->scsi.sense_data, sense, sizeof(*sense));
936         dest->scsiio.scsi_status = src->scsi.scsi_status;
937         dest->scsiio.sense_len = src->scsi.sense_len;
938         dest->io_hdr.status = src->hdr.status;
939 }
940
941 static int
942 ctl_init(void)
943 {
944         struct ctl_softc *softc;
945         struct ctl_io_pool *internal_pool, *emergency_pool, *other_pool;
946         struct ctl_port *port;
947         uint8_t sc_id =0;
948         int i, error, retval;
949         //int isc_retval;
950
951         retval = 0;
952         ctl_pause_rtr = 0;
953         rcv_sync_msg = 0;
954
955         control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
956                                M_WAITOK | M_ZERO);
957         softc = control_softc;
958
959         softc->dev = make_dev(&ctl_cdevsw, 0, UID_ROOT, GID_OPERATOR, 0600,
960                               "cam/ctl");
961
962         softc->dev->si_drv1 = softc;
963
964         /*
965          * By default, return a "bad LUN" peripheral qualifier for unknown
966          * LUNs.  The user can override this default using the tunable or
967          * sysctl.  See the comment in ctl_inquiry_std() for more details.
968          */
969         softc->inquiry_pq_no_lun = 1;
970         TUNABLE_INT_FETCH("kern.cam.ctl.inquiry_pq_no_lun",
971                           &softc->inquiry_pq_no_lun);
972         sysctl_ctx_init(&softc->sysctl_ctx);
973         softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
974                 SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
975                 CTLFLAG_RD, 0, "CAM Target Layer");
976
977         if (softc->sysctl_tree == NULL) {
978                 printf("%s: unable to allocate sysctl tree\n", __func__);
979                 destroy_dev(softc->dev);
980                 free(control_softc, M_DEVBUF);
981                 control_softc = NULL;
982                 return (ENOMEM);
983         }
984
985         SYSCTL_ADD_INT(&softc->sysctl_ctx,
986                        SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO,
987                        "inquiry_pq_no_lun", CTLFLAG_RW,
988                        &softc->inquiry_pq_no_lun, 0,
989                        "Report no lun possible for invalid LUNs");
990
991         mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
992         mtx_init(&softc->pool_lock, "CTL pool mutex", NULL, MTX_DEF);
993         softc->open_count = 0;
994
995         /*
996          * Default to actually sending a SYNCHRONIZE CACHE command down to
997          * the drive.
998          */
999         softc->flags = CTL_FLAG_REAL_SYNC;
1000
1001         /*
1002          * In Copan's HA scheme, the "master" and "slave" roles are
1003          * figured out through the slot the controller is in.  Although it
1004          * is an active/active system, someone has to be in charge.
1005          */
1006 #ifdef NEEDTOPORT
1007         scmicro_rw(SCMICRO_GET_SHELF_ID, &sc_id);
1008 #endif
1009
1010         if (sc_id == 0) {
1011                 softc->flags |= CTL_FLAG_MASTER_SHELF;
1012                 persis_offset = 0;
1013         } else
1014                 persis_offset = CTL_MAX_INITIATORS;
1015
1016         /*
1017          * XXX KDM need to figure out where we want to get our target ID
1018          * and WWID.  Is it different on each port?
1019          */
1020         softc->target.id = 0;
1021         softc->target.wwid[0] = 0x12345678;
1022         softc->target.wwid[1] = 0x87654321;
1023         STAILQ_INIT(&softc->lun_list);
1024         STAILQ_INIT(&softc->pending_lun_queue);
1025         STAILQ_INIT(&softc->fe_list);
1026         STAILQ_INIT(&softc->port_list);
1027         STAILQ_INIT(&softc->be_list);
1028         STAILQ_INIT(&softc->io_pools);
1029         ctl_tpc_init(softc);
1030
1031         if (ctl_pool_create(softc, CTL_POOL_INTERNAL, CTL_POOL_ENTRIES_INTERNAL,
1032                             &internal_pool)!= 0){
1033                 printf("ctl: can't allocate %d entry internal pool, "
1034                        "exiting\n", CTL_POOL_ENTRIES_INTERNAL);
1035                 return (ENOMEM);
1036         }
1037
1038         if (ctl_pool_create(softc, CTL_POOL_EMERGENCY,
1039                             CTL_POOL_ENTRIES_EMERGENCY, &emergency_pool) != 0) {
1040                 printf("ctl: can't allocate %d entry emergency pool, "
1041                        "exiting\n", CTL_POOL_ENTRIES_EMERGENCY);
1042                 ctl_pool_free(internal_pool);
1043                 return (ENOMEM);
1044         }
1045
1046         if (ctl_pool_create(softc, CTL_POOL_4OTHERSC, CTL_POOL_ENTRIES_OTHER_SC,
1047                             &other_pool) != 0)
1048         {
1049                 printf("ctl: can't allocate %d entry other SC pool, "
1050                        "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1051                 ctl_pool_free(internal_pool);
1052                 ctl_pool_free(emergency_pool);
1053                 return (ENOMEM);
1054         }
1055
1056         softc->internal_pool = internal_pool;
1057         softc->emergency_pool = emergency_pool;
1058         softc->othersc_pool = other_pool;
1059
1060         if (worker_threads <= 0)
1061                 worker_threads = max(1, mp_ncpus / 4);
1062         if (worker_threads > CTL_MAX_THREADS)
1063                 worker_threads = CTL_MAX_THREADS;
1064
1065         for (i = 0; i < worker_threads; i++) {
1066                 struct ctl_thread *thr = &softc->threads[i];
1067
1068                 mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1069                 thr->ctl_softc = softc;
1070                 STAILQ_INIT(&thr->incoming_queue);
1071                 STAILQ_INIT(&thr->rtr_queue);
1072                 STAILQ_INIT(&thr->done_queue);
1073                 STAILQ_INIT(&thr->isc_queue);
1074
1075                 error = kproc_kthread_add(ctl_work_thread, thr,
1076                     &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1077                 if (error != 0) {
1078                         printf("error creating CTL work thread!\n");
1079                         ctl_pool_free(internal_pool);
1080                         ctl_pool_free(emergency_pool);
1081                         ctl_pool_free(other_pool);
1082                         return (error);
1083                 }
1084         }
1085         error = kproc_kthread_add(ctl_lun_thread, softc,
1086             &softc->ctl_proc, NULL, 0, 0, "ctl", "lun");
1087         if (error != 0) {
1088                 printf("error creating CTL lun thread!\n");
1089                 ctl_pool_free(internal_pool);
1090                 ctl_pool_free(emergency_pool);
1091                 ctl_pool_free(other_pool);
1092                 return (error);
1093         }
1094         if (bootverbose)
1095                 printf("ctl: CAM Target Layer loaded\n");
1096
1097         /*
1098          * Initialize the ioctl front end.
1099          */
1100         ctl_frontend_register(&ioctl_frontend);
1101         port = &softc->ioctl_info.port;
1102         port->frontend = &ioctl_frontend;
1103         sprintf(softc->ioctl_info.port_name, "ioctl");
1104         port->port_type = CTL_PORT_IOCTL;
1105         port->num_requested_ctl_io = 100;
1106         port->port_name = softc->ioctl_info.port_name;
1107         port->port_online = ctl_ioctl_online;
1108         port->port_offline = ctl_ioctl_offline;
1109         port->onoff_arg = &softc->ioctl_info;
1110         port->lun_enable = ctl_ioctl_lun_enable;
1111         port->lun_disable = ctl_ioctl_lun_disable;
1112         port->targ_lun_arg = &softc->ioctl_info;
1113         port->fe_datamove = ctl_ioctl_datamove;
1114         port->fe_done = ctl_ioctl_done;
1115         port->max_targets = 15;
1116         port->max_target_id = 15;
1117
1118         if (ctl_port_register(&softc->ioctl_info.port,
1119                           (softc->flags & CTL_FLAG_MASTER_SHELF)) != 0) {
1120                 printf("ctl: ioctl front end registration failed, will "
1121                        "continue anyway\n");
1122         }
1123
1124 #ifdef CTL_IO_DELAY
1125         if (sizeof(struct callout) > CTL_TIMER_BYTES) {
1126                 printf("sizeof(struct callout) %zd > CTL_TIMER_BYTES %zd\n",
1127                        sizeof(struct callout), CTL_TIMER_BYTES);
1128                 return (EINVAL);
1129         }
1130 #endif /* CTL_IO_DELAY */
1131
1132         return (0);
1133 }
1134
1135 void
1136 ctl_shutdown(void)
1137 {
1138         struct ctl_softc *softc;
1139         struct ctl_lun *lun, *next_lun;
1140         struct ctl_io_pool *pool;
1141
1142         softc = (struct ctl_softc *)control_softc;
1143
1144         if (ctl_port_deregister(&softc->ioctl_info.port) != 0)
1145                 printf("ctl: ioctl front end deregistration failed\n");
1146
1147         mtx_lock(&softc->ctl_lock);
1148
1149         /*
1150          * Free up each LUN.
1151          */
1152         for (lun = STAILQ_FIRST(&softc->lun_list); lun != NULL; lun = next_lun){
1153                 next_lun = STAILQ_NEXT(lun, links);
1154                 ctl_free_lun(lun);
1155         }
1156
1157         mtx_unlock(&softc->ctl_lock);
1158
1159         ctl_frontend_deregister(&ioctl_frontend);
1160
1161         /*
1162          * This will rip the rug out from under any FETDs or anyone else
1163          * that has a pool allocated.  Since we increment our module
1164          * refcount any time someone outside the main CTL module allocates
1165          * a pool, we shouldn't have any problems here.  The user won't be
1166          * able to unload the CTL module until client modules have
1167          * successfully unloaded.
1168          */
1169         while ((pool = STAILQ_FIRST(&softc->io_pools)) != NULL)
1170                 ctl_pool_free(pool);
1171
1172 #if 0
1173         ctl_shutdown_thread(softc->work_thread);
1174         mtx_destroy(&softc->queue_lock);
1175 #endif
1176
1177         ctl_tpc_shutdown(softc);
1178         mtx_destroy(&softc->pool_lock);
1179         mtx_destroy(&softc->ctl_lock);
1180
1181         destroy_dev(softc->dev);
1182
1183         sysctl_ctx_free(&softc->sysctl_ctx);
1184
1185         free(control_softc, M_DEVBUF);
1186         control_softc = NULL;
1187
1188         if (bootverbose)
1189                 printf("ctl: CAM Target Layer unloaded\n");
1190 }
1191
1192 static int
1193 ctl_module_event_handler(module_t mod, int what, void *arg)
1194 {
1195
1196         switch (what) {
1197         case MOD_LOAD:
1198                 return (ctl_init());
1199         case MOD_UNLOAD:
1200                 return (EBUSY);
1201         default:
1202                 return (EOPNOTSUPP);
1203         }
1204 }
1205
1206 /*
1207  * XXX KDM should we do some access checks here?  Bump a reference count to
1208  * prevent a CTL module from being unloaded while someone has it open?
1209  */
1210 static int
1211 ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1212 {
1213         return (0);
1214 }
1215
1216 static int
1217 ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1218 {
1219         return (0);
1220 }
1221
1222 int
1223 ctl_port_enable(ctl_port_type port_type)
1224 {
1225         struct ctl_softc *softc;
1226         struct ctl_port *port;
1227
1228         if (ctl_is_single == 0) {
1229                 union ctl_ha_msg msg_info;
1230                 int isc_retval;
1231
1232 #if 0
1233                 printf("%s: HA mode, synchronizing frontend enable\n",
1234                         __func__);
1235 #endif
1236                 msg_info.hdr.msg_type = CTL_MSG_SYNC_FE;
1237                 if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1238                         sizeof(msg_info), 1 )) > CTL_HA_STATUS_SUCCESS) {
1239                         printf("Sync msg send error retval %d\n", isc_retval);
1240                 }
1241                 if (!rcv_sync_msg) {
1242                         isc_retval=ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
1243                                 sizeof(msg_info), 1);
1244                 }
1245 #if 0
1246                 printf("CTL:Frontend Enable\n");
1247         } else {
1248                 printf("%s: single mode, skipping frontend synchronization\n",
1249                         __func__);
1250 #endif
1251         }
1252
1253         softc = control_softc;
1254
1255         STAILQ_FOREACH(port, &softc->port_list, links) {
1256                 if (port_type & port->port_type)
1257                 {
1258 #if 0
1259                         printf("port %d\n", port->targ_port);
1260 #endif
1261                         ctl_port_online(port);
1262                 }
1263         }
1264
1265         return (0);
1266 }
1267
1268 int
1269 ctl_port_disable(ctl_port_type port_type)
1270 {
1271         struct ctl_softc *softc;
1272         struct ctl_port *port;
1273
1274         softc = control_softc;
1275
1276         STAILQ_FOREACH(port, &softc->port_list, links) {
1277                 if (port_type & port->port_type)
1278                         ctl_port_offline(port);
1279         }
1280
1281         return (0);
1282 }
1283
1284 /*
1285  * Returns 0 for success, 1 for failure.
1286  * Currently the only failure mode is if there aren't enough entries
1287  * allocated.  So, in case of a failure, look at num_entries_dropped,
1288  * reallocate and try again.
1289  */
1290 int
1291 ctl_port_list(struct ctl_port_entry *entries, int num_entries_alloced,
1292               int *num_entries_filled, int *num_entries_dropped,
1293               ctl_port_type port_type, int no_virtual)
1294 {
1295         struct ctl_softc *softc;
1296         struct ctl_port *port;
1297         int entries_dropped, entries_filled;
1298         int retval;
1299         int i;
1300
1301         softc = control_softc;
1302
1303         retval = 0;
1304         entries_filled = 0;
1305         entries_dropped = 0;
1306
1307         i = 0;
1308         mtx_lock(&softc->ctl_lock);
1309         STAILQ_FOREACH(port, &softc->port_list, links) {
1310                 struct ctl_port_entry *entry;
1311
1312                 if ((port->port_type & port_type) == 0)
1313                         continue;
1314
1315                 if ((no_virtual != 0)
1316                  && (port->virtual_port != 0))
1317                         continue;
1318
1319                 if (entries_filled >= num_entries_alloced) {
1320                         entries_dropped++;
1321                         continue;
1322                 }
1323                 entry = &entries[i];
1324
1325                 entry->port_type = port->port_type;
1326                 strlcpy(entry->port_name, port->port_name,
1327                         sizeof(entry->port_name));
1328                 entry->physical_port = port->physical_port;
1329                 entry->virtual_port = port->virtual_port;
1330                 entry->wwnn = port->wwnn;
1331                 entry->wwpn = port->wwpn;
1332
1333                 i++;
1334                 entries_filled++;
1335         }
1336
1337         mtx_unlock(&softc->ctl_lock);
1338
1339         if (entries_dropped > 0)
1340                 retval = 1;
1341
1342         *num_entries_dropped = entries_dropped;
1343         *num_entries_filled = entries_filled;
1344
1345         return (retval);
1346 }
1347
1348 static void
1349 ctl_ioctl_online(void *arg)
1350 {
1351         struct ctl_ioctl_info *ioctl_info;
1352
1353         ioctl_info = (struct ctl_ioctl_info *)arg;
1354
1355         ioctl_info->flags |= CTL_IOCTL_FLAG_ENABLED;
1356 }
1357
1358 static void
1359 ctl_ioctl_offline(void *arg)
1360 {
1361         struct ctl_ioctl_info *ioctl_info;
1362
1363         ioctl_info = (struct ctl_ioctl_info *)arg;
1364
1365         ioctl_info->flags &= ~CTL_IOCTL_FLAG_ENABLED;
1366 }
1367
1368 /*
1369  * Remove an initiator by port number and initiator ID.
1370  * Returns 0 for success, -1 for failure.
1371  */
1372 int
1373 ctl_remove_initiator(struct ctl_port *port, int iid)
1374 {
1375         struct ctl_softc *softc = control_softc;
1376
1377         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1378
1379         if (iid > CTL_MAX_INIT_PER_PORT) {
1380                 printf("%s: initiator ID %u > maximun %u!\n",
1381                        __func__, iid, CTL_MAX_INIT_PER_PORT);
1382                 return (-1);
1383         }
1384
1385         mtx_lock(&softc->ctl_lock);
1386         port->wwpn_iid[iid].in_use--;
1387         port->wwpn_iid[iid].last_use = time_uptime;
1388         mtx_unlock(&softc->ctl_lock);
1389
1390         return (0);
1391 }
1392
1393 /*
1394  * Add an initiator to the initiator map.
1395  * Returns iid for success, < 0 for failure.
1396  */
1397 int
1398 ctl_add_initiator(struct ctl_port *port, int iid, uint64_t wwpn, char *name)
1399 {
1400         struct ctl_softc *softc = control_softc;
1401         time_t best_time;
1402         int i, best;
1403
1404         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1405
1406         if (iid >= CTL_MAX_INIT_PER_PORT) {
1407                 printf("%s: WWPN %#jx initiator ID %u > maximum %u!\n",
1408                        __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
1409                 free(name, M_CTL);
1410                 return (-1);
1411         }
1412
1413         mtx_lock(&softc->ctl_lock);
1414
1415         if (iid < 0 && (wwpn != 0 || name != NULL)) {
1416                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1417                         if (wwpn != 0 && wwpn == port->wwpn_iid[i].wwpn) {
1418                                 iid = i;
1419                                 break;
1420                         }
1421                         if (name != NULL && port->wwpn_iid[i].name != NULL &&
1422                             strcmp(name, port->wwpn_iid[i].name) == 0) {
1423                                 iid = i;
1424                                 break;
1425                         }
1426                 }
1427         }
1428
1429         if (iid < 0) {
1430                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1431                         if (port->wwpn_iid[i].in_use == 0 &&
1432                             port->wwpn_iid[i].wwpn == 0 &&
1433                             port->wwpn_iid[i].name == NULL) {
1434                                 iid = i;
1435                                 break;
1436                         }
1437                 }
1438         }
1439
1440         if (iid < 0) {
1441                 best = -1;
1442                 best_time = INT32_MAX;
1443                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1444                         if (port->wwpn_iid[i].in_use == 0) {
1445                                 if (port->wwpn_iid[i].last_use < best_time) {
1446                                         best = i;
1447                                         best_time = port->wwpn_iid[i].last_use;
1448                                 }
1449                         }
1450                 }
1451                 iid = best;
1452         }
1453
1454         if (iid < 0) {
1455                 mtx_unlock(&softc->ctl_lock);
1456                 free(name, M_CTL);
1457                 return (-2);
1458         }
1459
1460         if (port->wwpn_iid[iid].in_use > 0 && (wwpn != 0 || name != NULL)) {
1461                 /*
1462                  * This is not an error yet.
1463                  */
1464                 if (wwpn != 0 && wwpn == port->wwpn_iid[iid].wwpn) {
1465 #if 0
1466                         printf("%s: port %d iid %u WWPN %#jx arrived"
1467                             " again\n", __func__, port->targ_port,
1468                             iid, (uintmax_t)wwpn);
1469 #endif
1470                         goto take;
1471                 }
1472                 if (name != NULL && port->wwpn_iid[iid].name != NULL &&
1473                     strcmp(name, port->wwpn_iid[iid].name) == 0) {
1474 #if 0
1475                         printf("%s: port %d iid %u name '%s' arrived"
1476                             " again\n", __func__, port->targ_port,
1477                             iid, name);
1478 #endif
1479                         goto take;
1480                 }
1481
1482                 /*
1483                  * This is an error, but what do we do about it?  The
1484                  * driver is telling us we have a new WWPN for this
1485                  * initiator ID, so we pretty much need to use it.
1486                  */
1487                 printf("%s: port %d iid %u WWPN %#jx '%s' arrived,"
1488                     " but WWPN %#jx '%s' is still at that address\n",
1489                     __func__, port->targ_port, iid, wwpn, name,
1490                     (uintmax_t)port->wwpn_iid[iid].wwpn,
1491                     port->wwpn_iid[iid].name);
1492
1493                 /*
1494                  * XXX KDM clear have_ca and ua_pending on each LUN for
1495                  * this initiator.
1496                  */
1497         }
1498 take:
1499         free(port->wwpn_iid[iid].name, M_CTL);
1500         port->wwpn_iid[iid].name = name;
1501         port->wwpn_iid[iid].wwpn = wwpn;
1502         port->wwpn_iid[iid].in_use++;
1503         mtx_unlock(&softc->ctl_lock);
1504
1505         return (iid);
1506 }
1507
1508 static int
1509 ctl_create_iid(struct ctl_port *port, int iid, uint8_t *buf)
1510 {
1511         int len;
1512
1513         switch (port->port_type) {
1514         case CTL_PORT_FC:
1515         {
1516                 struct scsi_transportid_fcp *id =
1517                     (struct scsi_transportid_fcp *)buf;
1518                 if (port->wwpn_iid[iid].wwpn == 0)
1519                         return (0);
1520                 memset(id, 0, sizeof(*id));
1521                 id->format_protocol = SCSI_PROTO_FC;
1522                 scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->n_port_name);
1523                 return (sizeof(*id));
1524         }
1525         case CTL_PORT_ISCSI:
1526         {
1527                 struct scsi_transportid_iscsi_port *id =
1528                     (struct scsi_transportid_iscsi_port *)buf;
1529                 if (port->wwpn_iid[iid].name == NULL)
1530                         return (0);
1531                 memset(id, 0, 256);
1532                 id->format_protocol = SCSI_TRN_ISCSI_FORMAT_PORT |
1533                     SCSI_PROTO_ISCSI;
1534                 len = strlcpy(id->iscsi_name, port->wwpn_iid[iid].name, 252) + 1;
1535                 len = roundup2(min(len, 252), 4);
1536                 scsi_ulto2b(len, id->additional_length);
1537                 return (sizeof(*id) + len);
1538         }
1539         case CTL_PORT_SAS:
1540         {
1541                 struct scsi_transportid_sas *id =
1542                     (struct scsi_transportid_sas *)buf;
1543                 if (port->wwpn_iid[iid].wwpn == 0)
1544                         return (0);
1545                 memset(id, 0, sizeof(*id));
1546                 id->format_protocol = SCSI_PROTO_SAS;
1547                 scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->sas_address);
1548                 return (sizeof(*id));
1549         }
1550         default:
1551         {
1552                 struct scsi_transportid_spi *id =
1553                     (struct scsi_transportid_spi *)buf;
1554                 memset(id, 0, sizeof(*id));
1555                 id->format_protocol = SCSI_PROTO_SPI;
1556                 scsi_ulto2b(iid, id->scsi_addr);
1557                 scsi_ulto2b(port->targ_port, id->rel_trgt_port_id);
1558                 return (sizeof(*id));
1559         }
1560         }
1561 }
1562
1563 static int
1564 ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id)
1565 {
1566         return (0);
1567 }
1568
1569 static int
1570 ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id)
1571 {
1572         return (0);
1573 }
1574
1575 /*
1576  * Data movement routine for the CTL ioctl frontend port.
1577  */
1578 static int
1579 ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio)
1580 {
1581         struct ctl_sg_entry *ext_sglist, *kern_sglist;
1582         struct ctl_sg_entry ext_entry, kern_entry;
1583         int ext_sglen, ext_sg_entries, kern_sg_entries;
1584         int ext_sg_start, ext_offset;
1585         int len_to_copy, len_copied;
1586         int kern_watermark, ext_watermark;
1587         int ext_sglist_malloced;
1588         int i, j;
1589
1590         ext_sglist_malloced = 0;
1591         ext_sg_start = 0;
1592         ext_offset = 0;
1593
1594         CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove\n"));
1595
1596         /*
1597          * If this flag is set, fake the data transfer.
1598          */
1599         if (ctsio->io_hdr.flags & CTL_FLAG_NO_DATAMOVE) {
1600                 ctsio->ext_data_filled = ctsio->ext_data_len;
1601                 goto bailout;
1602         }
1603
1604         /*
1605          * To simplify things here, if we have a single buffer, stick it in
1606          * a S/G entry and just make it a single entry S/G list.
1607          */
1608         if (ctsio->io_hdr.flags & CTL_FLAG_EDPTR_SGLIST) {
1609                 int len_seen;
1610
1611                 ext_sglen = ctsio->ext_sg_entries * sizeof(*ext_sglist);
1612
1613                 ext_sglist = (struct ctl_sg_entry *)malloc(ext_sglen, M_CTL,
1614                                                            M_WAITOK);
1615                 ext_sglist_malloced = 1;
1616                 if (copyin(ctsio->ext_data_ptr, ext_sglist,
1617                                    ext_sglen) != 0) {
1618                         ctl_set_internal_failure(ctsio,
1619                                                  /*sks_valid*/ 0,
1620                                                  /*retry_count*/ 0);
1621                         goto bailout;
1622                 }
1623                 ext_sg_entries = ctsio->ext_sg_entries;
1624                 len_seen = 0;
1625                 for (i = 0; i < ext_sg_entries; i++) {
1626                         if ((len_seen + ext_sglist[i].len) >=
1627                              ctsio->ext_data_filled) {
1628                                 ext_sg_start = i;
1629                                 ext_offset = ctsio->ext_data_filled - len_seen;
1630                                 break;
1631                         }
1632                         len_seen += ext_sglist[i].len;
1633                 }
1634         } else {
1635                 ext_sglist = &ext_entry;
1636                 ext_sglist->addr = ctsio->ext_data_ptr;
1637                 ext_sglist->len = ctsio->ext_data_len;
1638                 ext_sg_entries = 1;
1639                 ext_sg_start = 0;
1640                 ext_offset = ctsio->ext_data_filled;
1641         }
1642
1643         if (ctsio->kern_sg_entries > 0) {
1644                 kern_sglist = (struct ctl_sg_entry *)ctsio->kern_data_ptr;
1645                 kern_sg_entries = ctsio->kern_sg_entries;
1646         } else {
1647                 kern_sglist = &kern_entry;
1648                 kern_sglist->addr = ctsio->kern_data_ptr;
1649                 kern_sglist->len = ctsio->kern_data_len;
1650                 kern_sg_entries = 1;
1651         }
1652
1653
1654         kern_watermark = 0;
1655         ext_watermark = ext_offset;
1656         len_copied = 0;
1657         for (i = ext_sg_start, j = 0;
1658              i < ext_sg_entries && j < kern_sg_entries;) {
1659                 uint8_t *ext_ptr, *kern_ptr;
1660
1661                 len_to_copy = ctl_min(ext_sglist[i].len - ext_watermark,
1662                                       kern_sglist[j].len - kern_watermark);
1663
1664                 ext_ptr = (uint8_t *)ext_sglist[i].addr;
1665                 ext_ptr = ext_ptr + ext_watermark;
1666                 if (ctsio->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
1667                         /*
1668                          * XXX KDM fix this!
1669                          */
1670                         panic("need to implement bus address support");
1671 #if 0
1672                         kern_ptr = bus_to_virt(kern_sglist[j].addr);
1673 #endif
1674                 } else
1675                         kern_ptr = (uint8_t *)kern_sglist[j].addr;
1676                 kern_ptr = kern_ptr + kern_watermark;
1677
1678                 kern_watermark += len_to_copy;
1679                 ext_watermark += len_to_copy;
1680
1681                 if ((ctsio->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
1682                      CTL_FLAG_DATA_IN) {
1683                         CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1684                                          "bytes to user\n", len_to_copy));
1685                         CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1686                                          "to %p\n", kern_ptr, ext_ptr));
1687                         if (copyout(kern_ptr, ext_ptr, len_to_copy) != 0) {
1688                                 ctl_set_internal_failure(ctsio,
1689                                                          /*sks_valid*/ 0,
1690                                                          /*retry_count*/ 0);
1691                                 goto bailout;
1692                         }
1693                 } else {
1694                         CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1695                                          "bytes from user\n", len_to_copy));
1696                         CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1697                                          "to %p\n", ext_ptr, kern_ptr));
1698                         if (copyin(ext_ptr, kern_ptr, len_to_copy)!= 0){
1699                                 ctl_set_internal_failure(ctsio,
1700                                                          /*sks_valid*/ 0,
1701                                                          /*retry_count*/0);
1702                                 goto bailout;
1703                         }
1704                 }
1705
1706                 len_copied += len_to_copy;
1707
1708                 if (ext_sglist[i].len == ext_watermark) {
1709                         i++;
1710                         ext_watermark = 0;
1711                 }
1712
1713                 if (kern_sglist[j].len == kern_watermark) {
1714                         j++;
1715                         kern_watermark = 0;
1716                 }
1717         }
1718
1719         ctsio->ext_data_filled += len_copied;
1720
1721         CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_sg_entries: %d, "
1722                          "kern_sg_entries: %d\n", ext_sg_entries,
1723                          kern_sg_entries));
1724         CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_data_len = %d, "
1725                          "kern_data_len = %d\n", ctsio->ext_data_len,
1726                          ctsio->kern_data_len));
1727
1728
1729         /* XXX KDM set residual?? */
1730 bailout:
1731
1732         if (ext_sglist_malloced != 0)
1733                 free(ext_sglist, M_CTL);
1734
1735         return (CTL_RETVAL_COMPLETE);
1736 }
1737
1738 /*
1739  * Serialize a command that went down the "wrong" side, and so was sent to
1740  * this controller for execution.  The logic is a little different than the
1741  * standard case in ctl_scsiio_precheck().  Errors in this case need to get
1742  * sent back to the other side, but in the success case, we execute the
1743  * command on this side (XFER mode) or tell the other side to execute it
1744  * (SER_ONLY mode).
1745  */
1746 static int
1747 ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
1748 {
1749         struct ctl_softc *ctl_softc;
1750         union ctl_ha_msg msg_info;
1751         struct ctl_lun *lun;
1752         int retval = 0;
1753         uint32_t targ_lun;
1754
1755         ctl_softc = control_softc;
1756
1757         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
1758         lun = ctl_softc->ctl_luns[targ_lun];
1759         if (lun==NULL)
1760         {
1761                 /*
1762                  * Why isn't LUN defined? The other side wouldn't
1763                  * send a cmd if the LUN is undefined.
1764                  */
1765                 printf("%s: Bad JUJU!, LUN is NULL!\n", __func__);
1766
1767                 /* "Logical unit not supported" */
1768                 ctl_set_sense_data(&msg_info.scsi.sense_data,
1769                                    lun,
1770                                    /*sense_format*/SSD_TYPE_NONE,
1771                                    /*current_error*/ 1,
1772                                    /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1773                                    /*asc*/ 0x25,
1774                                    /*ascq*/ 0x00,
1775                                    SSD_ELEM_NONE);
1776
1777                 msg_info.scsi.sense_len = SSD_FULL_SIZE;
1778                 msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1779                 msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1780                 msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1781                 msg_info.hdr.serializing_sc = NULL;
1782                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1783                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1784                                 sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1785                 }
1786                 return(1);
1787
1788         }
1789
1790         mtx_lock(&lun->lun_lock);
1791         TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1792
1793         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
1794                 (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
1795                  ooa_links))) {
1796         case CTL_ACTION_BLOCK:
1797                 ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
1798                 TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
1799                                   blocked_links);
1800                 break;
1801         case CTL_ACTION_PASS:
1802         case CTL_ACTION_SKIP:
1803                 if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
1804                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
1805                         ctl_enqueue_rtr((union ctl_io *)ctsio);
1806                 } else {
1807
1808                         /* send msg back to other side */
1809                         msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1810                         msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
1811                         msg_info.hdr.msg_type = CTL_MSG_R2R;
1812 #if 0
1813                         printf("2. pOrig %x\n", (int)msg_info.hdr.original_sc);
1814 #endif
1815                         if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1816                             sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1817                         }
1818                 }
1819                 break;
1820         case CTL_ACTION_OVERLAP:
1821                 /* OVERLAPPED COMMANDS ATTEMPTED */
1822                 ctl_set_sense_data(&msg_info.scsi.sense_data,
1823                                    lun,
1824                                    /*sense_format*/SSD_TYPE_NONE,
1825                                    /*current_error*/ 1,
1826                                    /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1827                                    /*asc*/ 0x4E,
1828                                    /*ascq*/ 0x00,
1829                                    SSD_ELEM_NONE);
1830
1831                 msg_info.scsi.sense_len = SSD_FULL_SIZE;
1832                 msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1833                 msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1834                 msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1835                 msg_info.hdr.serializing_sc = NULL;
1836                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1837 #if 0
1838                 printf("BAD JUJU:Major Bummer Overlap\n");
1839 #endif
1840                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1841                 retval = 1;
1842                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1843                     sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1844                 }
1845                 break;
1846         case CTL_ACTION_OVERLAP_TAG:
1847                 /* TAGGED OVERLAPPED COMMANDS (NN = QUEUE TAG) */
1848                 ctl_set_sense_data(&msg_info.scsi.sense_data,
1849                                    lun,
1850                                    /*sense_format*/SSD_TYPE_NONE,
1851                                    /*current_error*/ 1,
1852                                    /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1853                                    /*asc*/ 0x4D,
1854                                    /*ascq*/ ctsio->tag_num & 0xff,
1855                                    SSD_ELEM_NONE);
1856
1857                 msg_info.scsi.sense_len = SSD_FULL_SIZE;
1858                 msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1859                 msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1860                 msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1861                 msg_info.hdr.serializing_sc = NULL;
1862                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1863 #if 0
1864                 printf("BAD JUJU:Major Bummer Overlap Tag\n");
1865 #endif
1866                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1867                 retval = 1;
1868                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1869                     sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1870                 }
1871                 break;
1872         case CTL_ACTION_ERROR:
1873         default:
1874                 /* "Internal target failure" */
1875                 ctl_set_sense_data(&msg_info.scsi.sense_data,
1876                                    lun,
1877                                    /*sense_format*/SSD_TYPE_NONE,
1878                                    /*current_error*/ 1,
1879                                    /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
1880                                    /*asc*/ 0x44,
1881                                    /*ascq*/ 0x00,
1882                                    SSD_ELEM_NONE);
1883
1884                 msg_info.scsi.sense_len = SSD_FULL_SIZE;
1885                 msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1886                 msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1887                 msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1888                 msg_info.hdr.serializing_sc = NULL;
1889                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1890 #if 0
1891                 printf("BAD JUJU:Major Bummer HW Error\n");
1892 #endif
1893                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1894                 retval = 1;
1895                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1896                     sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1897                 }
1898                 break;
1899         }
1900         mtx_unlock(&lun->lun_lock);
1901         return (retval);
1902 }
1903
1904 static int
1905 ctl_ioctl_submit_wait(union ctl_io *io)
1906 {
1907         struct ctl_fe_ioctl_params params;
1908         ctl_fe_ioctl_state last_state;
1909         int done, retval;
1910
1911         retval = 0;
1912
1913         bzero(&params, sizeof(params));
1914
1915         mtx_init(&params.ioctl_mtx, "ctliocmtx", NULL, MTX_DEF);
1916         cv_init(&params.sem, "ctlioccv");
1917         params.state = CTL_IOCTL_INPROG;
1918         last_state = params.state;
1919
1920         io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = &params;
1921
1922         CTL_DEBUG_PRINT(("ctl_ioctl_submit_wait\n"));
1923
1924         /* This shouldn't happen */
1925         if ((retval = ctl_queue(io)) != CTL_RETVAL_COMPLETE)
1926                 return (retval);
1927
1928         done = 0;
1929
1930         do {
1931                 mtx_lock(&params.ioctl_mtx);
1932                 /*
1933                  * Check the state here, and don't sleep if the state has
1934                  * already changed (i.e. wakeup has already occured, but we
1935                  * weren't waiting yet).
1936                  */
1937                 if (params.state == last_state) {
1938                         /* XXX KDM cv_wait_sig instead? */
1939                         cv_wait(&params.sem, &params.ioctl_mtx);
1940                 }
1941                 last_state = params.state;
1942
1943                 switch (params.state) {
1944                 case CTL_IOCTL_INPROG:
1945                         /* Why did we wake up? */
1946                         /* XXX KDM error here? */
1947                         mtx_unlock(&params.ioctl_mtx);
1948                         break;
1949                 case CTL_IOCTL_DATAMOVE:
1950                         CTL_DEBUG_PRINT(("got CTL_IOCTL_DATAMOVE\n"));
1951
1952                         /*
1953                          * change last_state back to INPROG to avoid
1954                          * deadlock on subsequent data moves.
1955                          */
1956                         params.state = last_state = CTL_IOCTL_INPROG;
1957
1958                         mtx_unlock(&params.ioctl_mtx);
1959                         ctl_ioctl_do_datamove(&io->scsiio);
1960                         /*
1961                          * Note that in some cases, most notably writes,
1962                          * this will queue the I/O and call us back later.
1963                          * In other cases, generally reads, this routine
1964                          * will immediately call back and wake us up,
1965                          * probably using our own context.
1966                          */
1967                         io->scsiio.be_move_done(io);
1968                         break;
1969                 case CTL_IOCTL_DONE:
1970                         mtx_unlock(&params.ioctl_mtx);
1971                         CTL_DEBUG_PRINT(("got CTL_IOCTL_DONE\n"));
1972                         done = 1;
1973                         break;
1974                 default:
1975                         mtx_unlock(&params.ioctl_mtx);
1976                         /* XXX KDM error here? */
1977                         break;
1978                 }
1979         } while (done == 0);
1980
1981         mtx_destroy(&params.ioctl_mtx);
1982         cv_destroy(&params.sem);
1983
1984         return (CTL_RETVAL_COMPLETE);
1985 }
1986
1987 static void
1988 ctl_ioctl_datamove(union ctl_io *io)
1989 {
1990         struct ctl_fe_ioctl_params *params;
1991
1992         params = (struct ctl_fe_ioctl_params *)
1993                 io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
1994
1995         mtx_lock(&params->ioctl_mtx);
1996         params->state = CTL_IOCTL_DATAMOVE;
1997         cv_broadcast(&params->sem);
1998         mtx_unlock(&params->ioctl_mtx);
1999 }
2000
2001 static void
2002 ctl_ioctl_done(union ctl_io *io)
2003 {
2004         struct ctl_fe_ioctl_params *params;
2005
2006         params = (struct ctl_fe_ioctl_params *)
2007                 io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
2008
2009         mtx_lock(&params->ioctl_mtx);
2010         params->state = CTL_IOCTL_DONE;
2011         cv_broadcast(&params->sem);
2012         mtx_unlock(&params->ioctl_mtx);
2013 }
2014
2015 static void
2016 ctl_ioctl_hard_startstop_callback(void *arg, struct cfi_metatask *metatask)
2017 {
2018         struct ctl_fe_ioctl_startstop_info *sd_info;
2019
2020         sd_info = (struct ctl_fe_ioctl_startstop_info *)arg;
2021
2022         sd_info->hs_info.status = metatask->status;
2023         sd_info->hs_info.total_luns = metatask->taskinfo.startstop.total_luns;
2024         sd_info->hs_info.luns_complete =
2025                 metatask->taskinfo.startstop.luns_complete;
2026         sd_info->hs_info.luns_failed = metatask->taskinfo.startstop.luns_failed;
2027
2028         cv_broadcast(&sd_info->sem);
2029 }
2030
2031 static void
2032 ctl_ioctl_bbrread_callback(void *arg, struct cfi_metatask *metatask)
2033 {
2034         struct ctl_fe_ioctl_bbrread_info *fe_bbr_info;
2035
2036         fe_bbr_info = (struct ctl_fe_ioctl_bbrread_info *)arg;
2037
2038         mtx_lock(fe_bbr_info->lock);
2039         fe_bbr_info->bbr_info->status = metatask->status;
2040         fe_bbr_info->bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2041         fe_bbr_info->wakeup_done = 1;
2042         mtx_unlock(fe_bbr_info->lock);
2043
2044         cv_broadcast(&fe_bbr_info->sem);
2045 }
2046
2047 /*
2048  * Returns 0 for success, errno for failure.
2049  */
2050 static int
2051 ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
2052                    struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
2053 {
2054         union ctl_io *io;
2055         int retval;
2056
2057         retval = 0;
2058
2059         mtx_lock(&lun->lun_lock);
2060         for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
2061              (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2062              ooa_links)) {
2063                 struct ctl_ooa_entry *entry;
2064
2065                 /*
2066                  * If we've got more than we can fit, just count the
2067                  * remaining entries.
2068                  */
2069                 if (*cur_fill_num >= ooa_hdr->alloc_num)
2070                         continue;
2071
2072                 entry = &kern_entries[*cur_fill_num];
2073
2074                 entry->tag_num = io->scsiio.tag_num;
2075                 entry->lun_num = lun->lun;
2076 #ifdef CTL_TIME_IO
2077                 entry->start_bt = io->io_hdr.start_bt;
2078 #endif
2079                 bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2080                 entry->cdb_len = io->scsiio.cdb_len;
2081                 if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
2082                         entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2083
2084                 if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2085                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2086
2087                 if (io->io_hdr.flags & CTL_FLAG_ABORT)
2088                         entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2089
2090                 if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2091                         entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2092
2093                 if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2094                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2095         }
2096         mtx_unlock(&lun->lun_lock);
2097
2098         return (retval);
2099 }
2100
2101 static void *
2102 ctl_copyin_alloc(void *user_addr, int len, char *error_str,
2103                  size_t error_str_len)
2104 {
2105         void *kptr;
2106
2107         kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2108
2109         if (copyin(user_addr, kptr, len) != 0) {
2110                 snprintf(error_str, error_str_len, "Error copying %d bytes "
2111                          "from user address %p to kernel address %p", len,
2112                          user_addr, kptr);
2113                 free(kptr, M_CTL);
2114                 return (NULL);
2115         }
2116
2117         return (kptr);
2118 }
2119
2120 static void
2121 ctl_free_args(int num_args, struct ctl_be_arg *args)
2122 {
2123         int i;
2124
2125         if (args == NULL)
2126                 return;
2127
2128         for (i = 0; i < num_args; i++) {
2129                 free(args[i].kname, M_CTL);
2130                 free(args[i].kvalue, M_CTL);
2131         }
2132
2133         free(args, M_CTL);
2134 }
2135
2136 static struct ctl_be_arg *
2137 ctl_copyin_args(int num_args, struct ctl_be_arg *uargs,
2138                 char *error_str, size_t error_str_len)
2139 {
2140         struct ctl_be_arg *args;
2141         int i;
2142
2143         args = ctl_copyin_alloc(uargs, num_args * sizeof(*args),
2144                                 error_str, error_str_len);
2145
2146         if (args == NULL)
2147                 goto bailout;
2148
2149         for (i = 0; i < num_args; i++) {
2150                 args[i].kname = NULL;
2151                 args[i].kvalue = NULL;
2152         }
2153
2154         for (i = 0; i < num_args; i++) {
2155                 uint8_t *tmpptr;
2156
2157                 args[i].kname = ctl_copyin_alloc(args[i].name,
2158                         args[i].namelen, error_str, error_str_len);
2159                 if (args[i].kname == NULL)
2160                         goto bailout;
2161
2162                 if (args[i].kname[args[i].namelen - 1] != '\0') {
2163                         snprintf(error_str, error_str_len, "Argument %d "
2164                                  "name is not NUL-terminated", i);
2165                         goto bailout;
2166                 }
2167
2168                 if (args[i].flags & CTL_BEARG_RD) {
2169                         tmpptr = ctl_copyin_alloc(args[i].value,
2170                                 args[i].vallen, error_str, error_str_len);
2171                         if (tmpptr == NULL)
2172                                 goto bailout;
2173                         if ((args[i].flags & CTL_BEARG_ASCII)
2174                          && (tmpptr[args[i].vallen - 1] != '\0')) {
2175                                 snprintf(error_str, error_str_len, "Argument "
2176                                     "%d value is not NUL-terminated", i);
2177                                 goto bailout;
2178                         }
2179                         args[i].kvalue = tmpptr;
2180                 } else {
2181                         args[i].kvalue = malloc(args[i].vallen,
2182                             M_CTL, M_WAITOK | M_ZERO);
2183                 }
2184         }
2185
2186         return (args);
2187 bailout:
2188
2189         ctl_free_args(num_args, args);
2190
2191         return (NULL);
2192 }
2193
2194 static void
2195 ctl_copyout_args(int num_args, struct ctl_be_arg *args)
2196 {
2197         int i;
2198
2199         for (i = 0; i < num_args; i++) {
2200                 if (args[i].flags & CTL_BEARG_WR)
2201                         copyout(args[i].kvalue, args[i].value, args[i].vallen);
2202         }
2203 }
2204
2205 /*
2206  * Escape characters that are illegal or not recommended in XML.
2207  */
2208 int
2209 ctl_sbuf_printf_esc(struct sbuf *sb, char *str)
2210 {
2211         int retval;
2212
2213         retval = 0;
2214
2215         for (; *str; str++) {
2216                 switch (*str) {
2217                 case '&':
2218                         retval = sbuf_printf(sb, "&amp;");
2219                         break;
2220                 case '>':
2221                         retval = sbuf_printf(sb, "&gt;");
2222                         break;
2223                 case '<':
2224                         retval = sbuf_printf(sb, "&lt;");
2225                         break;
2226                 default:
2227                         retval = sbuf_putc(sb, *str);
2228                         break;
2229                 }
2230
2231                 if (retval != 0)
2232                         break;
2233
2234         }
2235
2236         return (retval);
2237 }
2238
2239 static int
2240 ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2241           struct thread *td)
2242 {
2243         struct ctl_softc *softc;
2244         int retval;
2245
2246         softc = control_softc;
2247
2248         retval = 0;
2249
2250         switch (cmd) {
2251         case CTL_IO: {
2252                 union ctl_io *io;
2253                 void *pool_tmp;
2254
2255                 /*
2256                  * If we haven't been "enabled", don't allow any SCSI I/O
2257                  * to this FETD.
2258                  */
2259                 if ((softc->ioctl_info.flags & CTL_IOCTL_FLAG_ENABLED) == 0) {
2260                         retval = EPERM;
2261                         break;
2262                 }
2263
2264                 io = ctl_alloc_io(softc->ioctl_info.port.ctl_pool_ref);
2265                 if (io == NULL) {
2266                         printf("ctl_ioctl: can't allocate ctl_io!\n");
2267                         retval = ENOSPC;
2268                         break;
2269                 }
2270
2271                 /*
2272                  * Need to save the pool reference so it doesn't get
2273                  * spammed by the user's ctl_io.
2274                  */
2275                 pool_tmp = io->io_hdr.pool;
2276
2277                 memcpy(io, (void *)addr, sizeof(*io));
2278
2279                 io->io_hdr.pool = pool_tmp;
2280                 /*
2281                  * No status yet, so make sure the status is set properly.
2282                  */
2283                 io->io_hdr.status = CTL_STATUS_NONE;
2284
2285                 /*
2286                  * The user sets the initiator ID, target and LUN IDs.
2287                  */
2288                 io->io_hdr.nexus.targ_port = softc->ioctl_info.port.targ_port;
2289                 io->io_hdr.flags |= CTL_FLAG_USER_REQ;
2290                 if ((io->io_hdr.io_type == CTL_IO_SCSI)
2291                  && (io->scsiio.tag_type != CTL_TAG_UNTAGGED))
2292                         io->scsiio.tag_num = softc->ioctl_info.cur_tag_num++;
2293
2294                 retval = ctl_ioctl_submit_wait(io);
2295
2296                 if (retval != 0) {
2297                         ctl_free_io(io);
2298                         break;
2299                 }
2300
2301                 memcpy((void *)addr, io, sizeof(*io));
2302
2303                 /* return this to our pool */
2304                 ctl_free_io(io);
2305
2306                 break;
2307         }
2308         case CTL_ENABLE_PORT:
2309         case CTL_DISABLE_PORT:
2310         case CTL_SET_PORT_WWNS: {
2311                 struct ctl_port *port;
2312                 struct ctl_port_entry *entry;
2313
2314                 entry = (struct ctl_port_entry *)addr;
2315                 
2316                 mtx_lock(&softc->ctl_lock);
2317                 STAILQ_FOREACH(port, &softc->port_list, links) {
2318                         int action, done;
2319
2320                         action = 0;
2321                         done = 0;
2322
2323                         if ((entry->port_type == CTL_PORT_NONE)
2324                          && (entry->targ_port == port->targ_port)) {
2325                                 /*
2326                                  * If the user only wants to enable or
2327                                  * disable or set WWNs on a specific port,
2328                                  * do the operation and we're done.
2329                                  */
2330                                 action = 1;
2331                                 done = 1;
2332                         } else if (entry->port_type & port->port_type) {
2333                                 /*
2334                                  * Compare the user's type mask with the
2335                                  * particular frontend type to see if we
2336                                  * have a match.
2337                                  */
2338                                 action = 1;
2339                                 done = 0;
2340
2341                                 /*
2342                                  * Make sure the user isn't trying to set
2343                                  * WWNs on multiple ports at the same time.
2344                                  */
2345                                 if (cmd == CTL_SET_PORT_WWNS) {
2346                                         printf("%s: Can't set WWNs on "
2347                                                "multiple ports\n", __func__);
2348                                         retval = EINVAL;
2349                                         break;
2350                                 }
2351                         }
2352                         if (action != 0) {
2353                                 /*
2354                                  * XXX KDM we have to drop the lock here,
2355                                  * because the online/offline operations
2356                                  * can potentially block.  We need to
2357                                  * reference count the frontends so they
2358                                  * can't go away,
2359                                  */
2360                                 mtx_unlock(&softc->ctl_lock);
2361
2362                                 if (cmd == CTL_ENABLE_PORT) {
2363                                         struct ctl_lun *lun;
2364
2365                                         STAILQ_FOREACH(lun, &softc->lun_list,
2366                                                        links) {
2367                                                 port->lun_enable(port->targ_lun_arg,
2368                                                     lun->target,
2369                                                     lun->lun);
2370                                         }
2371
2372                                         ctl_port_online(port);
2373                                 } else if (cmd == CTL_DISABLE_PORT) {
2374                                         struct ctl_lun *lun;
2375
2376                                         ctl_port_offline(port);
2377
2378                                         STAILQ_FOREACH(lun, &softc->lun_list,
2379                                                        links) {
2380                                                 port->lun_disable(
2381                                                     port->targ_lun_arg,
2382                                                     lun->target,
2383                                                     lun->lun);
2384                                         }
2385                                 }
2386
2387                                 mtx_lock(&softc->ctl_lock);
2388
2389                                 if (cmd == CTL_SET_PORT_WWNS)
2390                                         ctl_port_set_wwns(port,
2391                                             (entry->flags & CTL_PORT_WWNN_VALID) ?
2392                                             1 : 0, entry->wwnn,
2393                                             (entry->flags & CTL_PORT_WWPN_VALID) ?
2394                                             1 : 0, entry->wwpn);
2395                         }
2396                         if (done != 0)
2397                                 break;
2398                 }
2399                 mtx_unlock(&softc->ctl_lock);
2400                 break;
2401         }
2402         case CTL_GET_PORT_LIST: {
2403                 struct ctl_port *port;
2404                 struct ctl_port_list *list;
2405                 int i;
2406
2407                 list = (struct ctl_port_list *)addr;
2408
2409                 if (list->alloc_len != (list->alloc_num *
2410                     sizeof(struct ctl_port_entry))) {
2411                         printf("%s: CTL_GET_PORT_LIST: alloc_len %u != "
2412                                "alloc_num %u * sizeof(struct ctl_port_entry) "
2413                                "%zu\n", __func__, list->alloc_len,
2414                                list->alloc_num, sizeof(struct ctl_port_entry));
2415                         retval = EINVAL;
2416                         break;
2417                 }
2418                 list->fill_len = 0;
2419                 list->fill_num = 0;
2420                 list->dropped_num = 0;
2421                 i = 0;
2422                 mtx_lock(&softc->ctl_lock);
2423                 STAILQ_FOREACH(port, &softc->port_list, links) {
2424                         struct ctl_port_entry entry, *list_entry;
2425
2426                         if (list->fill_num >= list->alloc_num) {
2427                                 list->dropped_num++;
2428                                 continue;
2429                         }
2430
2431                         entry.port_type = port->port_type;
2432                         strlcpy(entry.port_name, port->port_name,
2433                                 sizeof(entry.port_name));
2434                         entry.targ_port = port->targ_port;
2435                         entry.physical_port = port->physical_port;
2436                         entry.virtual_port = port->virtual_port;
2437                         entry.wwnn = port->wwnn;
2438                         entry.wwpn = port->wwpn;
2439                         if (port->status & CTL_PORT_STATUS_ONLINE)
2440                                 entry.online = 1;
2441                         else
2442                                 entry.online = 0;
2443
2444                         list_entry = &list->entries[i];
2445
2446                         retval = copyout(&entry, list_entry, sizeof(entry));
2447                         if (retval != 0) {
2448                                 printf("%s: CTL_GET_PORT_LIST: copyout "
2449                                        "returned %d\n", __func__, retval);
2450                                 break;
2451                         }
2452                         i++;
2453                         list->fill_num++;
2454                         list->fill_len += sizeof(entry);
2455                 }
2456                 mtx_unlock(&softc->ctl_lock);
2457
2458                 /*
2459                  * If this is non-zero, we had a copyout fault, so there's
2460                  * probably no point in attempting to set the status inside
2461                  * the structure.
2462                  */
2463                 if (retval != 0)
2464                         break;
2465
2466                 if (list->dropped_num > 0)
2467                         list->status = CTL_PORT_LIST_NEED_MORE_SPACE;
2468                 else
2469                         list->status = CTL_PORT_LIST_OK;
2470                 break;
2471         }
2472         case CTL_DUMP_OOA: {
2473                 struct ctl_lun *lun;
2474                 union ctl_io *io;
2475                 char printbuf[128];
2476                 struct sbuf sb;
2477
2478                 mtx_lock(&softc->ctl_lock);
2479                 printf("Dumping OOA queues:\n");
2480                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
2481                         mtx_lock(&lun->lun_lock);
2482                         for (io = (union ctl_io *)TAILQ_FIRST(
2483                              &lun->ooa_queue); io != NULL;
2484                              io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2485                              ooa_links)) {
2486                                 sbuf_new(&sb, printbuf, sizeof(printbuf),
2487                                          SBUF_FIXEDLEN);
2488                                 sbuf_printf(&sb, "LUN %jd tag 0x%04x%s%s%s%s: ",
2489                                             (intmax_t)lun->lun,
2490                                             io->scsiio.tag_num,
2491                                             (io->io_hdr.flags &
2492                                             CTL_FLAG_BLOCKED) ? "" : " BLOCKED",
2493                                             (io->io_hdr.flags &
2494                                             CTL_FLAG_DMA_INPROG) ? " DMA" : "",
2495                                             (io->io_hdr.flags &
2496                                             CTL_FLAG_ABORT) ? " ABORT" : "",
2497                                             (io->io_hdr.flags &
2498                                         CTL_FLAG_IS_WAS_ON_RTR) ? " RTR" : "");
2499                                 ctl_scsi_command_string(&io->scsiio, NULL, &sb);
2500                                 sbuf_finish(&sb);
2501                                 printf("%s\n", sbuf_data(&sb));
2502                         }
2503                         mtx_unlock(&lun->lun_lock);
2504                 }
2505                 printf("OOA queues dump done\n");
2506                 mtx_unlock(&softc->ctl_lock);
2507                 break;
2508         }
2509         case CTL_GET_OOA: {
2510                 struct ctl_lun *lun;
2511                 struct ctl_ooa *ooa_hdr;
2512                 struct ctl_ooa_entry *entries;
2513                 uint32_t cur_fill_num;
2514
2515                 ooa_hdr = (struct ctl_ooa *)addr;
2516
2517                 if ((ooa_hdr->alloc_len == 0)
2518                  || (ooa_hdr->alloc_num == 0)) {
2519                         printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2520                                "must be non-zero\n", __func__,
2521                                ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2522                         retval = EINVAL;
2523                         break;
2524                 }
2525
2526                 if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2527                     sizeof(struct ctl_ooa_entry))) {
2528                         printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2529                                "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2530                                __func__, ooa_hdr->alloc_len,
2531                                ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2532                         retval = EINVAL;
2533                         break;
2534                 }
2535
2536                 entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2537                 if (entries == NULL) {
2538                         printf("%s: could not allocate %d bytes for OOA "
2539                                "dump\n", __func__, ooa_hdr->alloc_len);
2540                         retval = ENOMEM;
2541                         break;
2542                 }
2543
2544                 mtx_lock(&softc->ctl_lock);
2545                 if (((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0)
2546                  && ((ooa_hdr->lun_num > CTL_MAX_LUNS)
2547                   || (softc->ctl_luns[ooa_hdr->lun_num] == NULL))) {
2548                         mtx_unlock(&softc->ctl_lock);
2549                         free(entries, M_CTL);
2550                         printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2551                                __func__, (uintmax_t)ooa_hdr->lun_num);
2552                         retval = EINVAL;
2553                         break;
2554                 }
2555
2556                 cur_fill_num = 0;
2557
2558                 if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2559                         STAILQ_FOREACH(lun, &softc->lun_list, links) {
2560                                 retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2561                                         ooa_hdr, entries);
2562                                 if (retval != 0)
2563                                         break;
2564                         }
2565                         if (retval != 0) {
2566                                 mtx_unlock(&softc->ctl_lock);
2567                                 free(entries, M_CTL);
2568                                 break;
2569                         }
2570                 } else {
2571                         lun = softc->ctl_luns[ooa_hdr->lun_num];
2572
2573                         retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,ooa_hdr,
2574                                                     entries);
2575                 }
2576                 mtx_unlock(&softc->ctl_lock);
2577
2578                 ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2579                 ooa_hdr->fill_len = ooa_hdr->fill_num *
2580                         sizeof(struct ctl_ooa_entry);
2581                 retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2582                 if (retval != 0) {
2583                         printf("%s: error copying out %d bytes for OOA dump\n", 
2584                                __func__, ooa_hdr->fill_len);
2585                 }
2586
2587                 getbintime(&ooa_hdr->cur_bt);
2588
2589                 if (cur_fill_num > ooa_hdr->alloc_num) {
2590                         ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2591                         ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2592                 } else {
2593                         ooa_hdr->dropped_num = 0;
2594                         ooa_hdr->status = CTL_OOA_OK;
2595                 }
2596
2597                 free(entries, M_CTL);
2598                 break;
2599         }
2600         case CTL_CHECK_OOA: {
2601                 union ctl_io *io;
2602                 struct ctl_lun *lun;
2603                 struct ctl_ooa_info *ooa_info;
2604
2605
2606                 ooa_info = (struct ctl_ooa_info *)addr;
2607
2608                 if (ooa_info->lun_id >= CTL_MAX_LUNS) {
2609                         ooa_info->status = CTL_OOA_INVALID_LUN;
2610                         break;
2611                 }
2612                 mtx_lock(&softc->ctl_lock);
2613                 lun = softc->ctl_luns[ooa_info->lun_id];
2614                 if (lun == NULL) {
2615                         mtx_unlock(&softc->ctl_lock);
2616                         ooa_info->status = CTL_OOA_INVALID_LUN;
2617                         break;
2618                 }
2619                 mtx_lock(&lun->lun_lock);
2620                 mtx_unlock(&softc->ctl_lock);
2621                 ooa_info->num_entries = 0;
2622                 for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
2623                      io != NULL; io = (union ctl_io *)TAILQ_NEXT(
2624                      &io->io_hdr, ooa_links)) {
2625                         ooa_info->num_entries++;
2626                 }
2627                 mtx_unlock(&lun->lun_lock);
2628
2629                 ooa_info->status = CTL_OOA_SUCCESS;
2630
2631                 break;
2632         }
2633         case CTL_HARD_START:
2634         case CTL_HARD_STOP: {
2635                 struct ctl_fe_ioctl_startstop_info ss_info;
2636                 struct cfi_metatask *metatask;
2637                 struct mtx hs_mtx;
2638
2639                 mtx_init(&hs_mtx, "HS Mutex", NULL, MTX_DEF);
2640
2641                 cv_init(&ss_info.sem, "hard start/stop cv" );
2642
2643                 metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2644                 if (metatask == NULL) {
2645                         retval = ENOMEM;
2646                         mtx_destroy(&hs_mtx);
2647                         break;
2648                 }
2649
2650                 if (cmd == CTL_HARD_START)
2651                         metatask->tasktype = CFI_TASK_STARTUP;
2652                 else
2653                         metatask->tasktype = CFI_TASK_SHUTDOWN;
2654
2655                 metatask->callback = ctl_ioctl_hard_startstop_callback;
2656                 metatask->callback_arg = &ss_info;
2657
2658                 cfi_action(metatask);
2659
2660                 /* Wait for the callback */
2661                 mtx_lock(&hs_mtx);
2662                 cv_wait_sig(&ss_info.sem, &hs_mtx);
2663                 mtx_unlock(&hs_mtx);
2664
2665                 /*
2666                  * All information has been copied from the metatask by the
2667                  * time cv_broadcast() is called, so we free the metatask here.
2668                  */
2669                 cfi_free_metatask(metatask);
2670
2671                 memcpy((void *)addr, &ss_info.hs_info, sizeof(ss_info.hs_info));
2672
2673                 mtx_destroy(&hs_mtx);
2674                 break;
2675         }
2676         case CTL_BBRREAD: {
2677                 struct ctl_bbrread_info *bbr_info;
2678                 struct ctl_fe_ioctl_bbrread_info fe_bbr_info;
2679                 struct mtx bbr_mtx;
2680                 struct cfi_metatask *metatask;
2681
2682                 bbr_info = (struct ctl_bbrread_info *)addr;
2683
2684                 bzero(&fe_bbr_info, sizeof(fe_bbr_info));
2685
2686                 bzero(&bbr_mtx, sizeof(bbr_mtx));
2687                 mtx_init(&bbr_mtx, "BBR Mutex", NULL, MTX_DEF);
2688
2689                 fe_bbr_info.bbr_info = bbr_info;
2690                 fe_bbr_info.lock = &bbr_mtx;
2691
2692                 cv_init(&fe_bbr_info.sem, "BBR read cv");
2693                 metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2694
2695                 if (metatask == NULL) {
2696                         mtx_destroy(&bbr_mtx);
2697                         cv_destroy(&fe_bbr_info.sem);
2698                         retval = ENOMEM;
2699                         break;
2700                 }
2701                 metatask->tasktype = CFI_TASK_BBRREAD;
2702                 metatask->callback = ctl_ioctl_bbrread_callback;
2703                 metatask->callback_arg = &fe_bbr_info;
2704                 metatask->taskinfo.bbrread.lun_num = bbr_info->lun_num;
2705                 metatask->taskinfo.bbrread.lba = bbr_info->lba;
2706                 metatask->taskinfo.bbrread.len = bbr_info->len;
2707
2708                 cfi_action(metatask);
2709
2710                 mtx_lock(&bbr_mtx);
2711                 while (fe_bbr_info.wakeup_done == 0)
2712                         cv_wait_sig(&fe_bbr_info.sem, &bbr_mtx);
2713                 mtx_unlock(&bbr_mtx);
2714
2715                 bbr_info->status = metatask->status;
2716                 bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2717                 bbr_info->scsi_status = metatask->taskinfo.bbrread.scsi_status;
2718                 memcpy(&bbr_info->sense_data,
2719                        &metatask->taskinfo.bbrread.sense_data,
2720                        ctl_min(sizeof(bbr_info->sense_data),
2721                                sizeof(metatask->taskinfo.bbrread.sense_data)));
2722
2723                 cfi_free_metatask(metatask);
2724
2725                 mtx_destroy(&bbr_mtx);
2726                 cv_destroy(&fe_bbr_info.sem);
2727
2728                 break;
2729         }
2730         case CTL_DELAY_IO: {
2731                 struct ctl_io_delay_info *delay_info;
2732 #ifdef CTL_IO_DELAY
2733                 struct ctl_lun *lun;
2734 #endif /* CTL_IO_DELAY */
2735
2736                 delay_info = (struct ctl_io_delay_info *)addr;
2737
2738 #ifdef CTL_IO_DELAY
2739                 mtx_lock(&softc->ctl_lock);
2740
2741                 if ((delay_info->lun_id > CTL_MAX_LUNS)
2742                  || (softc->ctl_luns[delay_info->lun_id] == NULL)) {
2743                         delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2744                 } else {
2745                         lun = softc->ctl_luns[delay_info->lun_id];
2746                         mtx_lock(&lun->lun_lock);
2747
2748                         delay_info->status = CTL_DELAY_STATUS_OK;
2749
2750                         switch (delay_info->delay_type) {
2751                         case CTL_DELAY_TYPE_CONT:
2752                                 break;
2753                         case CTL_DELAY_TYPE_ONESHOT:
2754                                 break;
2755                         default:
2756                                 delay_info->status =
2757                                         CTL_DELAY_STATUS_INVALID_TYPE;
2758                                 break;
2759                         }
2760
2761                         switch (delay_info->delay_loc) {
2762                         case CTL_DELAY_LOC_DATAMOVE:
2763                                 lun->delay_info.datamove_type =
2764                                         delay_info->delay_type;
2765                                 lun->delay_info.datamove_delay =
2766                                         delay_info->delay_secs;
2767                                 break;
2768                         case CTL_DELAY_LOC_DONE:
2769                                 lun->delay_info.done_type =
2770                                         delay_info->delay_type;
2771                                 lun->delay_info.done_delay =
2772                                         delay_info->delay_secs;
2773                                 break;
2774                         default:
2775                                 delay_info->status =
2776                                         CTL_DELAY_STATUS_INVALID_LOC;
2777                                 break;
2778                         }
2779                         mtx_unlock(&lun->lun_lock);
2780                 }
2781
2782                 mtx_unlock(&softc->ctl_lock);
2783 #else
2784                 delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2785 #endif /* CTL_IO_DELAY */
2786                 break;
2787         }
2788         case CTL_REALSYNC_SET: {
2789                 int *syncstate;
2790
2791                 syncstate = (int *)addr;
2792
2793                 mtx_lock(&softc->ctl_lock);
2794                 switch (*syncstate) {
2795                 case 0:
2796                         softc->flags &= ~CTL_FLAG_REAL_SYNC;
2797                         break;
2798                 case 1:
2799                         softc->flags |= CTL_FLAG_REAL_SYNC;
2800                         break;
2801                 default:
2802                         retval = EINVAL;
2803                         break;
2804                 }
2805                 mtx_unlock(&softc->ctl_lock);
2806                 break;
2807         }
2808         case CTL_REALSYNC_GET: {
2809                 int *syncstate;
2810
2811                 syncstate = (int*)addr;
2812
2813                 mtx_lock(&softc->ctl_lock);
2814                 if (softc->flags & CTL_FLAG_REAL_SYNC)
2815                         *syncstate = 1;
2816                 else
2817                         *syncstate = 0;
2818                 mtx_unlock(&softc->ctl_lock);
2819
2820                 break;
2821         }
2822         case CTL_SETSYNC:
2823         case CTL_GETSYNC: {
2824                 struct ctl_sync_info *sync_info;
2825                 struct ctl_lun *lun;
2826
2827                 sync_info = (struct ctl_sync_info *)addr;
2828
2829                 mtx_lock(&softc->ctl_lock);
2830                 lun = softc->ctl_luns[sync_info->lun_id];
2831                 if (lun == NULL) {
2832                         mtx_unlock(&softc->ctl_lock);
2833                         sync_info->status = CTL_GS_SYNC_NO_LUN;
2834                 }
2835                 /*
2836                  * Get or set the sync interval.  We're not bounds checking
2837                  * in the set case, hopefully the user won't do something
2838                  * silly.
2839                  */
2840                 mtx_lock(&lun->lun_lock);
2841                 mtx_unlock(&softc->ctl_lock);
2842                 if (cmd == CTL_GETSYNC)
2843                         sync_info->sync_interval = lun->sync_interval;
2844                 else
2845                         lun->sync_interval = sync_info->sync_interval;
2846                 mtx_unlock(&lun->lun_lock);
2847
2848                 sync_info->status = CTL_GS_SYNC_OK;
2849
2850                 break;
2851         }
2852         case CTL_GETSTATS: {
2853                 struct ctl_stats *stats;
2854                 struct ctl_lun *lun;
2855                 int i;
2856
2857                 stats = (struct ctl_stats *)addr;
2858
2859                 if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) >
2860                      stats->alloc_len) {
2861                         stats->status = CTL_SS_NEED_MORE_SPACE;
2862                         stats->num_luns = softc->num_luns;
2863                         break;
2864                 }
2865                 /*
2866                  * XXX KDM no locking here.  If the LUN list changes,
2867                  * things can blow up.
2868                  */
2869                 for (i = 0, lun = STAILQ_FIRST(&softc->lun_list); lun != NULL;
2870                      i++, lun = STAILQ_NEXT(lun, links)) {
2871                         retval = copyout(&lun->stats, &stats->lun_stats[i],
2872                                          sizeof(lun->stats));
2873                         if (retval != 0)
2874                                 break;
2875                 }
2876                 stats->num_luns = softc->num_luns;
2877                 stats->fill_len = sizeof(struct ctl_lun_io_stats) *
2878                                  softc->num_luns;
2879                 stats->status = CTL_SS_OK;
2880 #ifdef CTL_TIME_IO
2881                 stats->flags = CTL_STATS_FLAG_TIME_VALID;
2882 #else
2883                 stats->flags = CTL_STATS_FLAG_NONE;
2884 #endif
2885                 getnanouptime(&stats->timestamp);
2886                 break;
2887         }
2888         case CTL_ERROR_INJECT: {
2889                 struct ctl_error_desc *err_desc, *new_err_desc;
2890                 struct ctl_lun *lun;
2891
2892                 err_desc = (struct ctl_error_desc *)addr;
2893
2894                 new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2895                                       M_WAITOK | M_ZERO);
2896                 bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2897
2898                 mtx_lock(&softc->ctl_lock);
2899                 lun = softc->ctl_luns[err_desc->lun_id];
2900                 if (lun == NULL) {
2901                         mtx_unlock(&softc->ctl_lock);
2902                         printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2903                                __func__, (uintmax_t)err_desc->lun_id);
2904                         retval = EINVAL;
2905                         break;
2906                 }
2907                 mtx_lock(&lun->lun_lock);
2908                 mtx_unlock(&softc->ctl_lock);
2909
2910                 /*
2911                  * We could do some checking here to verify the validity
2912                  * of the request, but given the complexity of error
2913                  * injection requests, the checking logic would be fairly
2914                  * complex.
2915                  *
2916                  * For now, if the request is invalid, it just won't get
2917                  * executed and might get deleted.
2918                  */
2919                 STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2920
2921                 /*
2922                  * XXX KDM check to make sure the serial number is unique,
2923                  * in case we somehow manage to wrap.  That shouldn't
2924                  * happen for a very long time, but it's the right thing to
2925                  * do.
2926                  */
2927                 new_err_desc->serial = lun->error_serial;
2928                 err_desc->serial = lun->error_serial;
2929                 lun->error_serial++;
2930
2931                 mtx_unlock(&lun->lun_lock);
2932                 break;
2933         }
2934         case CTL_ERROR_INJECT_DELETE: {
2935                 struct ctl_error_desc *delete_desc, *desc, *desc2;
2936                 struct ctl_lun *lun;
2937                 int delete_done;
2938
2939                 delete_desc = (struct ctl_error_desc *)addr;
2940                 delete_done = 0;
2941
2942                 mtx_lock(&softc->ctl_lock);
2943                 lun = softc->ctl_luns[delete_desc->lun_id];
2944                 if (lun == NULL) {
2945                         mtx_unlock(&softc->ctl_lock);
2946                         printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2947                                __func__, (uintmax_t)delete_desc->lun_id);
2948                         retval = EINVAL;
2949                         break;
2950                 }
2951                 mtx_lock(&lun->lun_lock);
2952                 mtx_unlock(&softc->ctl_lock);
2953                 STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2954                         if (desc->serial != delete_desc->serial)
2955                                 continue;
2956
2957                         STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2958                                       links);
2959                         free(desc, M_CTL);
2960                         delete_done = 1;
2961                 }
2962                 mtx_unlock(&lun->lun_lock);
2963                 if (delete_done == 0) {
2964                         printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2965                                "error serial %ju on LUN %u\n", __func__, 
2966                                delete_desc->serial, delete_desc->lun_id);
2967                         retval = EINVAL;
2968                         break;
2969                 }
2970                 break;
2971         }
2972         case CTL_DUMP_STRUCTS: {
2973                 int i, j, k, idx;
2974                 struct ctl_port *port;
2975                 struct ctl_frontend *fe;
2976
2977                 mtx_lock(&softc->ctl_lock);
2978                 printf("CTL Persistent Reservation information start:\n");
2979                 for (i = 0; i < CTL_MAX_LUNS; i++) {
2980                         struct ctl_lun *lun;
2981
2982                         lun = softc->ctl_luns[i];
2983
2984                         if ((lun == NULL)
2985                          || ((lun->flags & CTL_LUN_DISABLED) != 0))
2986                                 continue;
2987
2988                         for (j = 0; j < (CTL_MAX_PORTS * 2); j++) {
2989                                 for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2990                                         idx = j * CTL_MAX_INIT_PER_PORT + k;
2991                                         if (lun->per_res[idx].registered == 0)
2992                                                 continue;
2993                                         printf("  LUN %d port %d iid %d key "
2994                                                "%#jx\n", i, j, k,
2995                                                (uintmax_t)scsi_8btou64(
2996                                                lun->per_res[idx].res_key.key));
2997                                 }
2998                         }
2999                 }
3000                 printf("CTL Persistent Reservation information end\n");
3001                 printf("CTL Ports:\n");
3002                 STAILQ_FOREACH(port, &softc->port_list, links) {
3003                         printf("  Port %d '%s' Frontend '%s' Type %u pp %d vp %d WWNN "
3004                                "%#jx WWPN %#jx\n", port->targ_port, port->port_name,
3005                                port->frontend->name, port->port_type,
3006                                port->physical_port, port->virtual_port,
3007                                (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
3008                         for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3009                                 if (port->wwpn_iid[j].in_use == 0 &&
3010                                     port->wwpn_iid[j].wwpn == 0 &&
3011                                     port->wwpn_iid[j].name == NULL)
3012                                         continue;
3013
3014                                 printf("    iid %u use %d WWPN %#jx '%s'\n",
3015                                     j, port->wwpn_iid[j].in_use,
3016                                     (uintmax_t)port->wwpn_iid[j].wwpn,
3017                                     port->wwpn_iid[j].name);
3018                         }
3019                 }
3020                 printf("CTL Port information end\n");
3021                 mtx_unlock(&softc->ctl_lock);
3022                 /*
3023                  * XXX KDM calling this without a lock.  We'd likely want
3024                  * to drop the lock before calling the frontend's dump
3025                  * routine anyway.
3026                  */
3027                 printf("CTL Frontends:\n");
3028                 STAILQ_FOREACH(fe, &softc->fe_list, links) {
3029                         printf("  Frontend '%s'\n", fe->name);
3030                         if (fe->fe_dump != NULL)
3031                                 fe->fe_dump();
3032                 }
3033                 printf("CTL Frontend information end\n");
3034                 break;
3035         }
3036         case CTL_LUN_REQ: {
3037                 struct ctl_lun_req *lun_req;
3038                 struct ctl_backend_driver *backend;
3039
3040                 lun_req = (struct ctl_lun_req *)addr;
3041
3042                 backend = ctl_backend_find(lun_req->backend);
3043                 if (backend == NULL) {
3044                         lun_req->status = CTL_LUN_ERROR;
3045                         snprintf(lun_req->error_str,
3046                                  sizeof(lun_req->error_str),
3047                                  "Backend \"%s\" not found.",
3048                                  lun_req->backend);
3049                         break;
3050                 }
3051                 if (lun_req->num_be_args > 0) {
3052                         lun_req->kern_be_args = ctl_copyin_args(
3053                                 lun_req->num_be_args,
3054                                 lun_req->be_args,
3055                                 lun_req->error_str,
3056                                 sizeof(lun_req->error_str));
3057                         if (lun_req->kern_be_args == NULL) {
3058                                 lun_req->status = CTL_LUN_ERROR;
3059                                 break;
3060                         }
3061                 }
3062
3063                 retval = backend->ioctl(dev, cmd, addr, flag, td);
3064
3065                 if (lun_req->num_be_args > 0) {
3066                         ctl_copyout_args(lun_req->num_be_args,
3067                                       lun_req->kern_be_args);
3068                         ctl_free_args(lun_req->num_be_args,
3069                                       lun_req->kern_be_args);
3070                 }
3071                 break;
3072         }
3073         case CTL_LUN_LIST: {
3074                 struct sbuf *sb;
3075                 struct ctl_lun *lun;
3076                 struct ctl_lun_list *list;
3077                 struct ctl_option *opt;
3078
3079                 list = (struct ctl_lun_list *)addr;
3080
3081                 /*
3082                  * Allocate a fixed length sbuf here, based on the length
3083                  * of the user's buffer.  We could allocate an auto-extending
3084                  * buffer, and then tell the user how much larger our
3085                  * amount of data is than his buffer, but that presents
3086                  * some problems:
3087                  *
3088                  * 1.  The sbuf(9) routines use a blocking malloc, and so
3089                  *     we can't hold a lock while calling them with an
3090                  *     auto-extending buffer.
3091                  *
3092                  * 2.  There is not currently a LUN reference counting
3093                  *     mechanism, outside of outstanding transactions on
3094                  *     the LUN's OOA queue.  So a LUN could go away on us
3095                  *     while we're getting the LUN number, backend-specific
3096                  *     information, etc.  Thus, given the way things
3097                  *     currently work, we need to hold the CTL lock while
3098                  *     grabbing LUN information.
3099                  *
3100                  * So, from the user's standpoint, the best thing to do is
3101                  * allocate what he thinks is a reasonable buffer length,
3102                  * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
3103                  * double the buffer length and try again.  (And repeat
3104                  * that until he succeeds.)
3105                  */
3106                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3107                 if (sb == NULL) {
3108                         list->status = CTL_LUN_LIST_ERROR;
3109                         snprintf(list->error_str, sizeof(list->error_str),
3110                                  "Unable to allocate %d bytes for LUN list",
3111                                  list->alloc_len);
3112                         break;
3113                 }
3114
3115                 sbuf_printf(sb, "<ctllunlist>\n");
3116
3117                 mtx_lock(&softc->ctl_lock);
3118                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
3119                         mtx_lock(&lun->lun_lock);
3120                         retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3121                                              (uintmax_t)lun->lun);
3122
3123                         /*
3124                          * Bail out as soon as we see that we've overfilled
3125                          * the buffer.
3126                          */
3127                         if (retval != 0)
3128                                 break;
3129
3130                         retval = sbuf_printf(sb, "\t<backend_type>%s"
3131                                              "</backend_type>\n",
3132                                              (lun->backend == NULL) ?  "none" :
3133                                              lun->backend->name);
3134
3135                         if (retval != 0)
3136                                 break;
3137
3138                         retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
3139                                              lun->be_lun->lun_type);
3140
3141                         if (retval != 0)
3142                                 break;
3143
3144                         if (lun->backend == NULL) {
3145                                 retval = sbuf_printf(sb, "</lun>\n");
3146                                 if (retval != 0)
3147                                         break;
3148                                 continue;
3149                         }
3150
3151                         retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
3152                                              (lun->be_lun->maxlba > 0) ?
3153                                              lun->be_lun->maxlba + 1 : 0);
3154
3155                         if (retval != 0)
3156                                 break;
3157
3158                         retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
3159                                              lun->be_lun->blocksize);
3160
3161                         if (retval != 0)
3162                                 break;
3163
3164                         retval = sbuf_printf(sb, "\t<serial_number>");
3165
3166                         if (retval != 0)
3167                                 break;
3168
3169                         retval = ctl_sbuf_printf_esc(sb,
3170                                                      lun->be_lun->serial_num);
3171
3172                         if (retval != 0)
3173                                 break;
3174
3175                         retval = sbuf_printf(sb, "</serial_number>\n");
3176                 
3177                         if (retval != 0)
3178                                 break;
3179
3180                         retval = sbuf_printf(sb, "\t<device_id>");
3181
3182                         if (retval != 0)
3183                                 break;
3184
3185                         retval = ctl_sbuf_printf_esc(sb,lun->be_lun->device_id);
3186
3187                         if (retval != 0)
3188                                 break;
3189
3190                         retval = sbuf_printf(sb, "</device_id>\n");
3191
3192                         if (retval != 0)
3193                                 break;
3194
3195                         if (lun->backend->lun_info != NULL) {
3196                                 retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3197                                 if (retval != 0)
3198                                         break;
3199                         }
3200                         STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
3201                                 retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3202                                     opt->name, opt->value, opt->name);
3203                                 if (retval != 0)
3204                                         break;
3205                         }
3206
3207                         retval = sbuf_printf(sb, "</lun>\n");
3208
3209                         if (retval != 0)
3210                                 break;
3211                         mtx_unlock(&lun->lun_lock);
3212                 }
3213                 if (lun != NULL)
3214                         mtx_unlock(&lun->lun_lock);
3215                 mtx_unlock(&softc->ctl_lock);
3216
3217                 if ((retval != 0)
3218                  || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3219                         retval = 0;
3220                         sbuf_delete(sb);
3221                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3222                         snprintf(list->error_str, sizeof(list->error_str),
3223                                  "Out of space, %d bytes is too small",
3224                                  list->alloc_len);
3225                         break;
3226                 }
3227
3228                 sbuf_finish(sb);
3229
3230                 retval = copyout(sbuf_data(sb), list->lun_xml,
3231                                  sbuf_len(sb) + 1);
3232
3233                 list->fill_len = sbuf_len(sb) + 1;
3234                 list->status = CTL_LUN_LIST_OK;
3235                 sbuf_delete(sb);
3236                 break;
3237         }
3238         case CTL_ISCSI: {
3239                 struct ctl_iscsi *ci;
3240                 struct ctl_frontend *fe;
3241
3242                 ci = (struct ctl_iscsi *)addr;
3243
3244                 fe = ctl_frontend_find("iscsi");
3245                 if (fe == NULL) {
3246                         ci->status = CTL_ISCSI_ERROR;
3247                         snprintf(ci->error_str, sizeof(ci->error_str),
3248                             "Frontend \"iscsi\" not found.");
3249                         break;
3250                 }
3251
3252                 retval = fe->ioctl(dev, cmd, addr, flag, td);
3253                 break;
3254         }
3255         case CTL_PORT_REQ: {
3256                 struct ctl_req *req;
3257                 struct ctl_frontend *fe;
3258
3259                 req = (struct ctl_req *)addr;
3260
3261                 fe = ctl_frontend_find(req->driver);
3262                 if (fe == NULL) {
3263                         req->status = CTL_LUN_ERROR;
3264                         snprintf(req->error_str, sizeof(req->error_str),
3265                             "Frontend \"%s\" not found.", req->driver);
3266                         break;
3267                 }
3268                 if (req->num_args > 0) {
3269                         req->kern_args = ctl_copyin_args(req->num_args,
3270                             req->args, req->error_str, sizeof(req->error_str));
3271                         if (req->kern_args == NULL) {
3272                                 req->status = CTL_LUN_ERROR;
3273                                 break;
3274                         }
3275                 }
3276
3277                 retval = fe->ioctl(dev, cmd, addr, flag, td);
3278
3279                 if (req->num_args > 0) {
3280                         ctl_copyout_args(req->num_args, req->kern_args);
3281                         ctl_free_args(req->num_args, req->kern_args);
3282                 }
3283                 break;
3284         }
3285         case CTL_PORT_LIST: {
3286                 struct sbuf *sb;
3287                 struct ctl_port *port;
3288                 struct ctl_lun_list *list;
3289                 struct ctl_option *opt;
3290
3291                 list = (struct ctl_lun_list *)addr;
3292
3293                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3294                 if (sb == NULL) {
3295                         list->status = CTL_LUN_LIST_ERROR;
3296                         snprintf(list->error_str, sizeof(list->error_str),
3297                                  "Unable to allocate %d bytes for LUN list",
3298                                  list->alloc_len);
3299                         break;
3300                 }
3301
3302                 sbuf_printf(sb, "<ctlportlist>\n");
3303
3304                 mtx_lock(&softc->ctl_lock);
3305                 STAILQ_FOREACH(port, &softc->port_list, links) {
3306                         retval = sbuf_printf(sb, "<targ_port id=\"%ju\">\n",
3307                                              (uintmax_t)port->targ_port);
3308
3309                         /*
3310                          * Bail out as soon as we see that we've overfilled
3311                          * the buffer.
3312                          */
3313                         if (retval != 0)
3314                                 break;
3315
3316                         retval = sbuf_printf(sb, "\t<frontend_type>%s"
3317                             "</frontend_type>\n", port->frontend->name);
3318                         if (retval != 0)
3319                                 break;
3320
3321                         retval = sbuf_printf(sb, "\t<port_type>%d</port_type>\n",
3322                                              port->port_type);
3323                         if (retval != 0)
3324                                 break;
3325
3326                         retval = sbuf_printf(sb, "\t<online>%s</online>\n",
3327                             (port->status & CTL_PORT_STATUS_ONLINE) ? "YES" : "NO");
3328                         if (retval != 0)
3329                                 break;
3330
3331                         retval = sbuf_printf(sb, "\t<port_name>%s</port_name>\n",
3332                             port->port_name);
3333                         if (retval != 0)
3334                                 break;
3335
3336                         retval = sbuf_printf(sb, "\t<physical_port>%d</physical_port>\n",
3337                             port->physical_port);
3338                         if (retval != 0)
3339                                 break;
3340
3341                         retval = sbuf_printf(sb, "\t<virtual_port>%d</virtual_port>\n",
3342                             port->virtual_port);
3343                         if (retval != 0)
3344                                 break;
3345
3346                         retval = sbuf_printf(sb, "\t<wwnn>%#jx</wwnn>\n",
3347                             (uintmax_t)port->wwnn);
3348                         if (retval != 0)
3349                                 break;
3350
3351                         retval = sbuf_printf(sb, "\t<wwpn>%#jx</wwpn>\n",
3352                             (uintmax_t)port->wwpn);
3353                         if (retval != 0)
3354                                 break;
3355
3356                         if (port->port_info != NULL) {
3357                                 retval = port->port_info(port->onoff_arg, sb);
3358                                 if (retval != 0)
3359                                         break;
3360                         }
3361                         STAILQ_FOREACH(opt, &port->options, links) {
3362                                 retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3363                                     opt->name, opt->value, opt->name);
3364                                 if (retval != 0)
3365                                         break;
3366                         }
3367
3368                         retval = sbuf_printf(sb, "</targ_port>\n");
3369                         if (retval != 0)
3370                                 break;
3371                 }
3372                 mtx_unlock(&softc->ctl_lock);
3373
3374                 if ((retval != 0)
3375                  || ((retval = sbuf_printf(sb, "</ctlportlist>\n")) != 0)) {
3376                         retval = 0;
3377                         sbuf_delete(sb);
3378                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3379                         snprintf(list->error_str, sizeof(list->error_str),
3380                                  "Out of space, %d bytes is too small",
3381                                  list->alloc_len);
3382                         break;
3383                 }
3384
3385                 sbuf_finish(sb);
3386
3387                 retval = copyout(sbuf_data(sb), list->lun_xml,
3388                                  sbuf_len(sb) + 1);
3389
3390                 list->fill_len = sbuf_len(sb) + 1;
3391                 list->status = CTL_LUN_LIST_OK;
3392                 sbuf_delete(sb);
3393                 break;
3394         }
3395         default: {
3396                 /* XXX KDM should we fix this? */
3397 #if 0
3398                 struct ctl_backend_driver *backend;
3399                 unsigned int type;
3400                 int found;
3401
3402                 found = 0;
3403
3404                 /*
3405                  * We encode the backend type as the ioctl type for backend
3406                  * ioctls.  So parse it out here, and then search for a
3407                  * backend of this type.
3408                  */
3409                 type = _IOC_TYPE(cmd);
3410
3411                 STAILQ_FOREACH(backend, &softc->be_list, links) {
3412                         if (backend->type == type) {
3413                                 found = 1;
3414                                 break;
3415                         }
3416                 }
3417                 if (found == 0) {
3418                         printf("ctl: unknown ioctl command %#lx or backend "
3419                                "%d\n", cmd, type);
3420                         retval = EINVAL;
3421                         break;
3422                 }
3423                 retval = backend->ioctl(dev, cmd, addr, flag, td);
3424 #endif
3425                 retval = ENOTTY;
3426                 break;
3427         }
3428         }
3429         return (retval);
3430 }
3431
3432 uint32_t
3433 ctl_get_initindex(struct ctl_nexus *nexus)
3434 {
3435         if (nexus->targ_port < CTL_MAX_PORTS)
3436                 return (nexus->initid.id +
3437                         (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3438         else
3439                 return (nexus->initid.id +
3440                        ((nexus->targ_port - CTL_MAX_PORTS) *
3441                         CTL_MAX_INIT_PER_PORT));
3442 }
3443
3444 uint32_t
3445 ctl_get_resindex(struct ctl_nexus *nexus)
3446 {
3447         return (nexus->initid.id + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3448 }
3449
3450 uint32_t
3451 ctl_port_idx(int port_num)
3452 {
3453         if (port_num < CTL_MAX_PORTS)
3454                 return(port_num);
3455         else
3456                 return(port_num - CTL_MAX_PORTS);
3457 }
3458
3459 static uint32_t
3460 ctl_map_lun(int port_num, uint32_t lun_id)
3461 {
3462         struct ctl_port *port;
3463
3464         port = control_softc->ctl_ports[ctl_port_idx(port_num)];
3465         if (port == NULL)
3466                 return (UINT32_MAX);
3467         if (port->lun_map == NULL)
3468                 return (lun_id);
3469         return (port->lun_map(port->targ_lun_arg, lun_id));
3470 }
3471
3472 static uint32_t
3473 ctl_map_lun_back(int port_num, uint32_t lun_id)
3474 {
3475         struct ctl_port *port;
3476         uint32_t i;
3477
3478         port = control_softc->ctl_ports[ctl_port_idx(port_num)];
3479         if (port->lun_map == NULL)
3480                 return (lun_id);
3481         for (i = 0; i < CTL_MAX_LUNS; i++) {
3482                 if (port->lun_map(port->targ_lun_arg, i) == lun_id)
3483                         return (i);
3484         }
3485         return (UINT32_MAX);
3486 }
3487
3488 /*
3489  * Note:  This only works for bitmask sizes that are at least 32 bits, and
3490  * that are a power of 2.
3491  */
3492 int
3493 ctl_ffz(uint32_t *mask, uint32_t size)
3494 {
3495         uint32_t num_chunks, num_pieces;
3496         int i, j;
3497
3498         num_chunks = (size >> 5);
3499         if (num_chunks == 0)
3500                 num_chunks++;
3501         num_pieces = ctl_min((sizeof(uint32_t) * 8), size);
3502
3503         for (i = 0; i < num_chunks; i++) {
3504                 for (j = 0; j < num_pieces; j++) {
3505                         if ((mask[i] & (1 << j)) == 0)
3506                                 return ((i << 5) + j);
3507                 }
3508         }
3509
3510         return (-1);
3511 }
3512
3513 int
3514 ctl_set_mask(uint32_t *mask, uint32_t bit)
3515 {
3516         uint32_t chunk, piece;
3517
3518         chunk = bit >> 5;
3519         piece = bit % (sizeof(uint32_t) * 8);
3520
3521         if ((mask[chunk] & (1 << piece)) != 0)
3522                 return (-1);
3523         else
3524                 mask[chunk] |= (1 << piece);
3525
3526         return (0);
3527 }
3528
3529 int
3530 ctl_clear_mask(uint32_t *mask, uint32_t bit)
3531 {
3532         uint32_t chunk, piece;
3533
3534         chunk = bit >> 5;
3535         piece = bit % (sizeof(uint32_t) * 8);
3536
3537         if ((mask[chunk] & (1 << piece)) == 0)
3538                 return (-1);
3539         else
3540                 mask[chunk] &= ~(1 << piece);
3541
3542         return (0);
3543 }
3544
3545 int
3546 ctl_is_set(uint32_t *mask, uint32_t bit)
3547 {
3548         uint32_t chunk, piece;
3549
3550         chunk = bit >> 5;
3551         piece = bit % (sizeof(uint32_t) * 8);
3552
3553         if ((mask[chunk] & (1 << piece)) == 0)
3554                 return (0);
3555         else
3556                 return (1);
3557 }
3558
3559 #ifdef unused
3560 /*
3561  * The bus, target and lun are optional, they can be filled in later.
3562  * can_wait is used to determine whether we can wait on the malloc or not.
3563  */
3564 union ctl_io*
3565 ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port, uint32_t targ_target,
3566               uint32_t targ_lun, int can_wait)
3567 {
3568         union ctl_io *io;
3569
3570         if (can_wait)
3571                 io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_WAITOK);
3572         else
3573                 io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_NOWAIT);
3574
3575         if (io != NULL) {
3576                 io->io_hdr.io_type = io_type;
3577                 io->io_hdr.targ_port = targ_port;
3578                 /*
3579                  * XXX KDM this needs to change/go away.  We need to move
3580                  * to a preallocated pool of ctl_scsiio structures.
3581                  */
3582                 io->io_hdr.nexus.targ_target.id = targ_target;
3583                 io->io_hdr.nexus.targ_lun = targ_lun;
3584         }
3585
3586         return (io);
3587 }
3588
3589 void
3590 ctl_kfree_io(union ctl_io *io)
3591 {
3592         free(io, M_CTL);
3593 }
3594 #endif /* unused */
3595
3596 /*
3597  * ctl_softc, pool_type, total_ctl_io are passed in.
3598  * npool is passed out.
3599  */
3600 int
3601 ctl_pool_create(struct ctl_softc *ctl_softc, ctl_pool_type pool_type,
3602                 uint32_t total_ctl_io, struct ctl_io_pool **npool)
3603 {
3604         uint32_t i;
3605         union ctl_io *cur_io, *next_io;
3606         struct ctl_io_pool *pool;
3607         int retval;
3608
3609         retval = 0;
3610
3611         pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3612                                             M_NOWAIT | M_ZERO);
3613         if (pool == NULL) {
3614                 retval = ENOMEM;
3615                 goto bailout;
3616         }
3617
3618         pool->type = pool_type;
3619         pool->ctl_softc = ctl_softc;
3620
3621         mtx_lock(&ctl_softc->pool_lock);
3622         pool->id = ctl_softc->cur_pool_id++;
3623         mtx_unlock(&ctl_softc->pool_lock);
3624
3625         pool->flags = CTL_POOL_FLAG_NONE;
3626         pool->refcount = 1;             /* Reference for validity. */
3627         STAILQ_INIT(&pool->free_queue);
3628
3629         /*
3630          * XXX KDM other options here:
3631          * - allocate a page at a time
3632          * - allocate one big chunk of memory.
3633          * Page allocation might work well, but would take a little more
3634          * tracking.
3635          */
3636         for (i = 0; i < total_ctl_io; i++) {
3637                 cur_io = (union ctl_io *)malloc(sizeof(*cur_io), M_CTLIO,
3638                                                 M_NOWAIT);
3639                 if (cur_io == NULL) {
3640                         retval = ENOMEM;
3641                         break;
3642                 }
3643                 cur_io->io_hdr.pool = pool;
3644                 STAILQ_INSERT_TAIL(&pool->free_queue, &cur_io->io_hdr, links);
3645                 pool->total_ctl_io++;
3646                 pool->free_ctl_io++;
3647         }
3648
3649         if (retval != 0) {
3650                 for (cur_io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3651                      cur_io != NULL; cur_io = next_io) {
3652                         next_io = (union ctl_io *)STAILQ_NEXT(&cur_io->io_hdr,
3653                                                               links);
3654                         STAILQ_REMOVE(&pool->free_queue, &cur_io->io_hdr,
3655                                       ctl_io_hdr, links);
3656                         free(cur_io, M_CTLIO);
3657                 }
3658
3659                 free(pool, M_CTL);
3660                 goto bailout;
3661         }
3662         mtx_lock(&ctl_softc->pool_lock);
3663         ctl_softc->num_pools++;
3664         STAILQ_INSERT_TAIL(&ctl_softc->io_pools, pool, links);
3665         /*
3666          * Increment our usage count if this is an external consumer, so we
3667          * can't get unloaded until the external consumer (most likely a
3668          * FETD) unloads and frees his pool.
3669          *
3670          * XXX KDM will this increment the caller's module use count, or
3671          * mine?
3672          */
3673 #if 0
3674         if ((pool_type != CTL_POOL_EMERGENCY)
3675          && (pool_type != CTL_POOL_INTERNAL)
3676          && (pool_type != CTL_POOL_4OTHERSC))
3677                 MOD_INC_USE_COUNT;
3678 #endif
3679
3680         mtx_unlock(&ctl_softc->pool_lock);
3681
3682         *npool = pool;
3683
3684 bailout:
3685
3686         return (retval);
3687 }
3688
3689 static int
3690 ctl_pool_acquire(struct ctl_io_pool *pool)
3691 {
3692
3693         mtx_assert(&pool->ctl_softc->pool_lock, MA_OWNED);
3694
3695         if (pool->flags & CTL_POOL_FLAG_INVALID)
3696                 return (EINVAL);
3697
3698         pool->refcount++;
3699
3700         return (0);
3701 }
3702
3703 static void
3704 ctl_pool_release(struct ctl_io_pool *pool)
3705 {
3706         struct ctl_softc *ctl_softc = pool->ctl_softc;
3707         union ctl_io *io;
3708
3709         mtx_assert(&ctl_softc->pool_lock, MA_OWNED);
3710
3711         if (--pool->refcount != 0)
3712                 return;
3713
3714         while ((io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue)) != NULL) {
3715                 STAILQ_REMOVE(&pool->free_queue, &io->io_hdr, ctl_io_hdr,
3716                               links);
3717                 free(io, M_CTLIO);
3718         }
3719
3720         STAILQ_REMOVE(&ctl_softc->io_pools, pool, ctl_io_pool, links);
3721         ctl_softc->num_pools--;
3722
3723         /*
3724          * XXX KDM will this decrement the caller's usage count or mine?
3725          */
3726 #if 0
3727         if ((pool->type != CTL_POOL_EMERGENCY)
3728          && (pool->type != CTL_POOL_INTERNAL)
3729          && (pool->type != CTL_POOL_4OTHERSC))
3730                 MOD_DEC_USE_COUNT;
3731 #endif
3732
3733         free(pool, M_CTL);
3734 }
3735
3736 void
3737 ctl_pool_free(struct ctl_io_pool *pool)
3738 {
3739         struct ctl_softc *ctl_softc;
3740
3741         if (pool == NULL)
3742                 return;
3743
3744         ctl_softc = pool->ctl_softc;
3745         mtx_lock(&ctl_softc->pool_lock);
3746         pool->flags |= CTL_POOL_FLAG_INVALID;
3747         ctl_pool_release(pool);
3748         mtx_unlock(&ctl_softc->pool_lock);
3749 }
3750
3751 /*
3752  * This routine does not block (except for spinlocks of course).
3753  * It tries to allocate a ctl_io union from the caller's pool as quickly as
3754  * possible.
3755  */
3756 union ctl_io *
3757 ctl_alloc_io(void *pool_ref)
3758 {
3759         union ctl_io *io;
3760         struct ctl_softc *ctl_softc;
3761         struct ctl_io_pool *pool, *npool;
3762         struct ctl_io_pool *emergency_pool;
3763
3764         pool = (struct ctl_io_pool *)pool_ref;
3765
3766         if (pool == NULL) {
3767                 printf("%s: pool is NULL\n", __func__);
3768                 return (NULL);
3769         }
3770
3771         emergency_pool = NULL;
3772
3773         ctl_softc = pool->ctl_softc;
3774
3775         mtx_lock(&ctl_softc->pool_lock);
3776         /*
3777          * First, try to get the io structure from the user's pool.
3778          */
3779         if (ctl_pool_acquire(pool) == 0) {
3780                 io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3781                 if (io != NULL) {
3782                         STAILQ_REMOVE_HEAD(&pool->free_queue, links);
3783                         pool->total_allocated++;
3784                         pool->free_ctl_io--;
3785                         mtx_unlock(&ctl_softc->pool_lock);
3786                         return (io);
3787                 } else
3788                         ctl_pool_release(pool);
3789         }
3790         /*
3791          * If he doesn't have any io structures left, search for an
3792          * emergency pool and grab one from there.
3793          */
3794         STAILQ_FOREACH(npool, &ctl_softc->io_pools, links) {
3795                 if (npool->type != CTL_POOL_EMERGENCY)
3796                         continue;
3797
3798                 if (ctl_pool_acquire(npool) != 0)
3799                         continue;
3800
3801                 emergency_pool = npool;
3802
3803                 io = (union ctl_io *)STAILQ_FIRST(&npool->free_queue);
3804                 if (io != NULL) {
3805                         STAILQ_REMOVE_HEAD(&npool->free_queue, links);
3806                         npool->total_allocated++;
3807                         npool->free_ctl_io--;
3808                         mtx_unlock(&ctl_softc->pool_lock);
3809                         return (io);
3810                 } else
3811                         ctl_pool_release(npool);
3812         }
3813
3814         /* Drop the spinlock before we malloc */
3815         mtx_unlock(&ctl_softc->pool_lock);
3816
3817         /*
3818          * The emergency pool (if it exists) didn't have one, so try an
3819          * atomic (i.e. nonblocking) malloc and see if we get lucky.
3820          */
3821         io = (union ctl_io *)malloc(sizeof(*io), M_CTLIO, M_NOWAIT);
3822         if (io != NULL) {
3823                 /*
3824                  * If the emergency pool exists but is empty, add this
3825                  * ctl_io to its list when it gets freed.
3826                  */
3827                 if (emergency_pool != NULL) {
3828                         mtx_lock(&ctl_softc->pool_lock);
3829                         if (ctl_pool_acquire(emergency_pool) == 0) {
3830                                 io->io_hdr.pool = emergency_pool;
3831                                 emergency_pool->total_ctl_io++;
3832                                 /*
3833                                  * Need to bump this, otherwise
3834                                  * total_allocated and total_freed won't
3835                                  * match when we no longer have anything
3836                                  * outstanding.
3837                                  */
3838                                 emergency_pool->total_allocated++;
3839                         }
3840                         mtx_unlock(&ctl_softc->pool_lock);
3841                 } else
3842                         io->io_hdr.pool = NULL;
3843         }
3844
3845         return (io);
3846 }
3847
3848 void
3849 ctl_free_io(union ctl_io *io)
3850 {
3851         if (io == NULL)
3852                 return;
3853
3854         /*
3855          * If this ctl_io has a pool, return it to that pool.
3856          */
3857         if (io->io_hdr.pool != NULL) {
3858                 struct ctl_io_pool *pool;
3859
3860                 pool = (struct ctl_io_pool *)io->io_hdr.pool;
3861                 mtx_lock(&pool->ctl_softc->pool_lock);
3862                 io->io_hdr.io_type = 0xff;
3863                 STAILQ_INSERT_TAIL(&pool->free_queue, &io->io_hdr, links);
3864                 pool->total_freed++;
3865                 pool->free_ctl_io++;
3866                 ctl_pool_release(pool);
3867                 mtx_unlock(&pool->ctl_softc->pool_lock);
3868         } else {
3869                 /*
3870                  * Otherwise, just free it.  We probably malloced it and
3871                  * the emergency pool wasn't available.
3872                  */
3873                 free(io, M_CTLIO);
3874         }
3875
3876 }
3877
3878 void
3879 ctl_zero_io(union ctl_io *io)
3880 {
3881         void *pool_ref;
3882
3883         if (io == NULL)
3884                 return;
3885
3886         /*
3887          * May need to preserve linked list pointers at some point too.
3888          */
3889         pool_ref = io->io_hdr.pool;
3890
3891         memset(io, 0, sizeof(*io));
3892
3893         io->io_hdr.pool = pool_ref;
3894 }
3895
3896 /*
3897  * This routine is currently used for internal copies of ctl_ios that need
3898  * to persist for some reason after we've already returned status to the
3899  * FETD.  (Thus the flag set.)
3900  *
3901  * XXX XXX
3902  * Note that this makes a blind copy of all fields in the ctl_io, except
3903  * for the pool reference.  This includes any memory that has been
3904  * allocated!  That memory will no longer be valid after done has been
3905  * called, so this would be VERY DANGEROUS for command that actually does
3906  * any reads or writes.  Right now (11/7/2005), this is only used for immediate
3907  * start and stop commands, which don't transfer any data, so this is not a
3908  * problem.  If it is used for anything else, the caller would also need to
3909  * allocate data buffer space and this routine would need to be modified to
3910  * copy the data buffer(s) as well.
3911  */
3912 void
3913 ctl_copy_io(union ctl_io *src, union ctl_io *dest)
3914 {
3915         void *pool_ref;
3916
3917         if ((src == NULL)
3918          || (dest == NULL))
3919                 return;
3920
3921         /*
3922          * May need to preserve linked list pointers at some point too.
3923          */
3924         pool_ref = dest->io_hdr.pool;
3925
3926         memcpy(dest, src, ctl_min(sizeof(*src), sizeof(*dest)));
3927
3928         dest->io_hdr.pool = pool_ref;
3929         /*
3930          * We need to know that this is an internal copy, and doesn't need
3931          * to get passed back to the FETD that allocated it.
3932          */
3933         dest->io_hdr.flags |= CTL_FLAG_INT_COPY;
3934 }
3935
3936 #ifdef NEEDTOPORT
3937 static void
3938 ctl_update_power_subpage(struct copan_power_subpage *page)
3939 {
3940         int num_luns, num_partitions, config_type;
3941         struct ctl_softc *softc;
3942         cs_BOOL_t aor_present, shelf_50pct_power;
3943         cs_raidset_personality_t rs_type;
3944         int max_active_luns;
3945
3946         softc = control_softc;
3947
3948         /* subtract out the processor LUN */
3949         num_luns = softc->num_luns - 1;
3950         /*
3951          * Default to 7 LUNs active, which was the only number we allowed
3952          * in the past.
3953          */
3954         max_active_luns = 7;
3955
3956         num_partitions = config_GetRsPartitionInfo();
3957         config_type = config_GetConfigType();
3958         shelf_50pct_power = config_GetShelfPowerMode();
3959         aor_present = config_IsAorRsPresent();
3960
3961         rs_type = ddb_GetRsRaidType(1);
3962         if ((rs_type != CS_RAIDSET_PERSONALITY_RAID5)
3963          && (rs_type != CS_RAIDSET_PERSONALITY_RAID1)) {
3964                 EPRINT(0, "Unsupported RS type %d!", rs_type);
3965         }
3966
3967
3968         page->total_luns = num_luns;
3969
3970         switch (config_type) {
3971         case 40:
3972                 /*
3973                  * In a 40 drive configuration, it doesn't matter what DC
3974                  * cards we have, whether we have AOR enabled or not,
3975                  * partitioning or not, or what type of RAIDset we have.
3976                  * In that scenario, we can power up every LUN we present
3977                  * to the user.
3978                  */
3979                 max_active_luns = num_luns;
3980
3981                 break;
3982         case 64:
3983                 if (shelf_50pct_power == CS_FALSE) {
3984                         /* 25% power */
3985                         if (aor_present == CS_TRUE) {
3986                                 if (rs_type ==
3987                                      CS_RAIDSET_PERSONALITY_RAID5) {
3988                                         max_active_luns = 7;
3989                                 } else if (rs_type ==
3990                                          CS_RAIDSET_PERSONALITY_RAID1){
3991                                         max_active_luns = 14;
3992                                 } else {
3993                                         /* XXX KDM now what?? */
3994                                 }
3995                         } else {
3996                                 if (rs_type ==
3997                                      CS_RAIDSET_PERSONALITY_RAID5) {
3998                                         max_active_luns = 8;
3999                                 } else if (rs_type ==
4000                                          CS_RAIDSET_PERSONALITY_RAID1){
4001                                         max_active_luns = 16;
4002                                 } else {
4003                                         /* XXX KDM now what?? */
4004                                 }
4005                         }
4006                 } else {
4007                         /* 50% power */
4008                         /*
4009                          * With 50% power in a 64 drive configuration, we
4010                          * can power all LUNs we present.
4011                          */
4012                         max_active_luns = num_luns;
4013                 }
4014                 break;
4015         case 112:
4016                 if (shelf_50pct_power == CS_FALSE) {
4017                         /* 25% power */
4018                         if (aor_present == CS_TRUE) {
4019                                 if (rs_type ==
4020                                      CS_RAIDSET_PERSONALITY_RAID5) {
4021                                         max_active_luns = 7;
4022                                 } else if (rs_type ==
4023                                          CS_RAIDSET_PERSONALITY_RAID1){
4024                                         max_active_luns = 14;
4025                                 } else {
4026                                         /* XXX KDM now what?? */
4027                                 }
4028                         } else {
4029                                 if (rs_type ==
4030                                      CS_RAIDSET_PERSONALITY_RAID5) {
4031                                         max_active_luns = 8;
4032                                 } else if (rs_type ==
4033                                          CS_RAIDSET_PERSONALITY_RAID1){
4034                                         max_active_luns = 16;
4035                                 } else {
4036                                         /* XXX KDM now what?? */
4037                                 }
4038                         }
4039                 } else {
4040                         /* 50% power */
4041                         if (aor_present == CS_TRUE) {
4042                                 if (rs_type ==
4043                                      CS_RAIDSET_PERSONALITY_RAID5) {
4044                                         max_active_luns = 14;
4045                                 } else if (rs_type ==
4046                                          CS_RAIDSET_PERSONALITY_RAID1){
4047                                         /*
4048                                          * We're assuming here that disk
4049                                          * caching is enabled, and so we're
4050                                          * able to power up half of each
4051                                          * LUN, and cache all writes.
4052                                          */
4053                                         max_active_luns = num_luns;
4054                                 } else {
4055                                         /* XXX KDM now what?? */
4056                                 }
4057                         } else {
4058                                 if (rs_type ==
4059                                      CS_RAIDSET_PERSONALITY_RAID5) {
4060                                         max_active_luns = 15;
4061                                 } else if (rs_type ==
4062                                          CS_RAIDSET_PERSONALITY_RAID1){
4063                                         max_active_luns = 30;
4064                                 } else {
4065                                         /* XXX KDM now what?? */
4066                                 }
4067                         }
4068                 }
4069                 break;
4070         default:
4071                 /*
4072                  * In this case, we have an unknown configuration, so we
4073                  * just use the default from above.
4074                  */
4075                 break;
4076         }
4077
4078         page->max_active_luns = max_active_luns;
4079 #if 0
4080         printk("%s: total_luns = %d, max_active_luns = %d\n", __func__,
4081                page->total_luns, page->max_active_luns);
4082 #endif
4083 }
4084 #endif /* NEEDTOPORT */
4085
4086 /*
4087  * This routine could be used in the future to load default and/or saved
4088  * mode page parameters for a particuar lun.
4089  */
4090 static int
4091 ctl_init_page_index(struct ctl_lun *lun)
4092 {
4093         int i;
4094         struct ctl_page_index *page_index;
4095         struct ctl_softc *softc;
4096
4097         memcpy(&lun->mode_pages.index, page_index_template,
4098                sizeof(page_index_template));
4099
4100         softc = lun->ctl_softc;
4101
4102         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
4103
4104                 page_index = &lun->mode_pages.index[i];
4105                 /*
4106                  * If this is a disk-only mode page, there's no point in
4107                  * setting it up.  For some pages, we have to have some
4108                  * basic information about the disk in order to calculate the
4109                  * mode page data.
4110                  */
4111                 if ((lun->be_lun->lun_type != T_DIRECT)
4112                  && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
4113                         continue;
4114
4115                 switch (page_index->page_code & SMPH_PC_MASK) {
4116                 case SMS_FORMAT_DEVICE_PAGE: {
4117                         struct scsi_format_page *format_page;
4118
4119                         if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4120                                 panic("subpage is incorrect!");
4121
4122                         /*
4123                          * Sectors per track are set above.  Bytes per
4124                          * sector need to be set here on a per-LUN basis.
4125                          */
4126                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
4127                                &format_page_default,
4128                                sizeof(format_page_default));
4129                         memcpy(&lun->mode_pages.format_page[
4130                                CTL_PAGE_CHANGEABLE], &format_page_changeable,
4131                                sizeof(format_page_changeable));
4132                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
4133                                &format_page_default,
4134                                sizeof(format_page_default));
4135                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
4136                                &format_page_default,
4137                                sizeof(format_page_default));
4138
4139                         format_page = &lun->mode_pages.format_page[
4140                                 CTL_PAGE_CURRENT];
4141                         scsi_ulto2b(lun->be_lun->blocksize,
4142                                     format_page->bytes_per_sector);
4143
4144                         format_page = &lun->mode_pages.format_page[
4145                                 CTL_PAGE_DEFAULT];
4146                         scsi_ulto2b(lun->be_lun->blocksize,
4147                                     format_page->bytes_per_sector);
4148
4149                         format_page = &lun->mode_pages.format_page[
4150                                 CTL_PAGE_SAVED];
4151                         scsi_ulto2b(lun->be_lun->blocksize,
4152                                     format_page->bytes_per_sector);
4153
4154                         page_index->page_data =
4155                                 (uint8_t *)lun->mode_pages.format_page;
4156                         break;
4157                 }
4158                 case SMS_RIGID_DISK_PAGE: {
4159                         struct scsi_rigid_disk_page *rigid_disk_page;
4160                         uint32_t sectors_per_cylinder;
4161                         uint64_t cylinders;
4162 #ifndef __XSCALE__
4163                         int shift;
4164 #endif /* !__XSCALE__ */
4165
4166                         if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4167                                 panic("invalid subpage value %d",
4168                                       page_index->subpage);
4169
4170                         /*
4171                          * Rotation rate and sectors per track are set
4172                          * above.  We calculate the cylinders here based on
4173                          * capacity.  Due to the number of heads and
4174                          * sectors per track we're using, smaller arrays
4175                          * may turn out to have 0 cylinders.  Linux and
4176                          * FreeBSD don't pay attention to these mode pages
4177                          * to figure out capacity, but Solaris does.  It
4178                          * seems to deal with 0 cylinders just fine, and
4179                          * works out a fake geometry based on the capacity.
4180                          */
4181                         memcpy(&lun->mode_pages.rigid_disk_page[
4182                                CTL_PAGE_CURRENT], &rigid_disk_page_default,
4183                                sizeof(rigid_disk_page_default));
4184                         memcpy(&lun->mode_pages.rigid_disk_page[
4185                                CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
4186                                sizeof(rigid_disk_page_changeable));
4187                         memcpy(&lun->mode_pages.rigid_disk_page[
4188                                CTL_PAGE_DEFAULT], &rigid_disk_page_default,
4189                                sizeof(rigid_disk_page_default));
4190                         memcpy(&lun->mode_pages.rigid_disk_page[
4191                                CTL_PAGE_SAVED], &rigid_disk_page_default,
4192                                sizeof(rigid_disk_page_default));
4193
4194                         sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
4195                                 CTL_DEFAULT_HEADS;
4196
4197                         /*
4198                          * The divide method here will be more accurate,
4199                          * probably, but results in floating point being
4200                          * used in the kernel on i386 (__udivdi3()).  On the
4201                          * XScale, though, __udivdi3() is implemented in
4202                          * software.
4203                          *
4204                          * The shift method for cylinder calculation is
4205                          * accurate if sectors_per_cylinder is a power of
4206                          * 2.  Otherwise it might be slightly off -- you
4207                          * might have a bit of a truncation problem.
4208                          */
4209 #ifdef  __XSCALE__
4210                         cylinders = (lun->be_lun->maxlba + 1) /
4211                                 sectors_per_cylinder;
4212 #else
4213                         for (shift = 31; shift > 0; shift--) {
4214                                 if (sectors_per_cylinder & (1 << shift))
4215                                         break;
4216                         }
4217                         cylinders = (lun->be_lun->maxlba + 1) >> shift;
4218 #endif
4219
4220                         /*
4221                          * We've basically got 3 bytes, or 24 bits for the
4222                          * cylinder size in the mode page.  If we're over,
4223                          * just round down to 2^24.
4224                          */
4225                         if (cylinders > 0xffffff)
4226                                 cylinders = 0xffffff;
4227
4228                         rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4229                                 CTL_PAGE_CURRENT];
4230                         scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4231
4232                         rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4233                                 CTL_PAGE_DEFAULT];
4234                         scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4235
4236                         rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4237                                 CTL_PAGE_SAVED];
4238                         scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4239
4240                         page_index->page_data =
4241                                 (uint8_t *)lun->mode_pages.rigid_disk_page;
4242                         break;
4243                 }
4244                 case SMS_CACHING_PAGE: {
4245
4246                         if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4247                                 panic("invalid subpage value %d",
4248                                       page_index->subpage);
4249                         /*
4250                          * Defaults should be okay here, no calculations
4251                          * needed.
4252                          */
4253                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4254                                &caching_page_default,
4255                                sizeof(caching_page_default));
4256                         memcpy(&lun->mode_pages.caching_page[
4257                                CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4258                                sizeof(caching_page_changeable));
4259                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4260                                &caching_page_default,
4261                                sizeof(caching_page_default));
4262                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4263                                &caching_page_default,
4264                                sizeof(caching_page_default));
4265                         page_index->page_data =
4266                                 (uint8_t *)lun->mode_pages.caching_page;
4267                         break;
4268                 }
4269                 case SMS_CONTROL_MODE_PAGE: {
4270
4271                         if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4272                                 panic("invalid subpage value %d",
4273                                       page_index->subpage);
4274
4275                         /*
4276                          * Defaults should be okay here, no calculations
4277                          * needed.
4278                          */
4279                         memcpy(&lun->mode_pages.control_page[CTL_PAGE_CURRENT],
4280                                &control_page_default,
4281                                sizeof(control_page_default));
4282                         memcpy(&lun->mode_pages.control_page[
4283                                CTL_PAGE_CHANGEABLE], &control_page_changeable,
4284                                sizeof(control_page_changeable));
4285                         memcpy(&lun->mode_pages.control_page[CTL_PAGE_DEFAULT],
4286                                &control_page_default,
4287                                sizeof(control_page_default));
4288                         memcpy(&lun->mode_pages.control_page[CTL_PAGE_SAVED],
4289                                &control_page_default,
4290                                sizeof(control_page_default));
4291                         page_index->page_data =
4292                                 (uint8_t *)lun->mode_pages.control_page;
4293                         break;
4294
4295                 }
4296                 case SMS_VENDOR_SPECIFIC_PAGE:{
4297                         switch (page_index->subpage) {
4298                         case PWR_SUBPAGE_CODE: {
4299                                 struct copan_power_subpage *current_page,
4300                                                            *saved_page;
4301
4302                                 memcpy(&lun->mode_pages.power_subpage[
4303                                        CTL_PAGE_CURRENT],
4304                                        &power_page_default,
4305                                        sizeof(power_page_default));
4306                                 memcpy(&lun->mode_pages.power_subpage[
4307                                        CTL_PAGE_CHANGEABLE],
4308                                        &power_page_changeable,
4309                                        sizeof(power_page_changeable));
4310                                 memcpy(&lun->mode_pages.power_subpage[
4311                                        CTL_PAGE_DEFAULT],
4312                                        &power_page_default,
4313                                        sizeof(power_page_default));
4314                                 memcpy(&lun->mode_pages.power_subpage[
4315                                        CTL_PAGE_SAVED],
4316                                        &power_page_default,
4317                                        sizeof(power_page_default));
4318                                 page_index->page_data =
4319                                     (uint8_t *)lun->mode_pages.power_subpage;
4320
4321                                 current_page = (struct copan_power_subpage *)
4322                                         (page_index->page_data +
4323                                          (page_index->page_len *
4324                                           CTL_PAGE_CURRENT));
4325                                 saved_page = (struct copan_power_subpage *)
4326                                         (page_index->page_data +
4327                                          (page_index->page_len *
4328                                           CTL_PAGE_SAVED));
4329                                 break;
4330                         }
4331                         case APS_SUBPAGE_CODE: {
4332                                 struct copan_aps_subpage *current_page,
4333                                                          *saved_page;
4334
4335                                 // This gets set multiple times but
4336                                 // it should always be the same. It's
4337                                 // only done during init so who cares.
4338                                 index_to_aps_page = i;
4339
4340                                 memcpy(&lun->mode_pages.aps_subpage[
4341                                        CTL_PAGE_CURRENT],
4342                                        &aps_page_default,
4343                                        sizeof(aps_page_default));
4344                                 memcpy(&lun->mode_pages.aps_subpage[
4345                                        CTL_PAGE_CHANGEABLE],
4346                                        &aps_page_changeable,
4347                                        sizeof(aps_page_changeable));
4348                                 memcpy(&lun->mode_pages.aps_subpage[
4349                                        CTL_PAGE_DEFAULT],
4350                                        &aps_page_default,
4351                                        sizeof(aps_page_default));
4352                                 memcpy(&lun->mode_pages.aps_subpage[
4353                                        CTL_PAGE_SAVED],
4354                                        &aps_page_default,
4355                                        sizeof(aps_page_default));
4356                                 page_index->page_data =
4357                                         (uint8_t *)lun->mode_pages.aps_subpage;
4358
4359                                 current_page = (struct copan_aps_subpage *)
4360                                         (page_index->page_data +
4361                                          (page_index->page_len *
4362                                           CTL_PAGE_CURRENT));
4363                                 saved_page = (struct copan_aps_subpage *)
4364                                         (page_index->page_data +
4365                                          (page_index->page_len *
4366                                           CTL_PAGE_SAVED));
4367                                 break;
4368                         }
4369                         case DBGCNF_SUBPAGE_CODE: {
4370                                 struct copan_debugconf_subpage *current_page,
4371                                                                *saved_page;
4372
4373                                 memcpy(&lun->mode_pages.debugconf_subpage[
4374                                        CTL_PAGE_CURRENT],
4375                                        &debugconf_page_default,
4376                                        sizeof(debugconf_page_default));
4377                                 memcpy(&lun->mode_pages.debugconf_subpage[
4378                                        CTL_PAGE_CHANGEABLE],
4379                                        &debugconf_page_changeable,
4380                                        sizeof(debugconf_page_changeable));
4381                                 memcpy(&lun->mode_pages.debugconf_subpage[
4382                                        CTL_PAGE_DEFAULT],
4383                                        &debugconf_page_default,
4384                                        sizeof(debugconf_page_default));
4385                                 memcpy(&lun->mode_pages.debugconf_subpage[
4386                                        CTL_PAGE_SAVED],
4387                                        &debugconf_page_default,
4388                                        sizeof(debugconf_page_default));
4389                                 page_index->page_data =
4390                                         (uint8_t *)lun->mode_pages.debugconf_subpage;
4391
4392                                 current_page = (struct copan_debugconf_subpage *)
4393                                         (page_index->page_data +
4394                                          (page_index->page_len *
4395                                           CTL_PAGE_CURRENT));
4396                                 saved_page = (struct copan_debugconf_subpage *)
4397                                         (page_index->page_data +
4398                                          (page_index->page_len *
4399                                           CTL_PAGE_SAVED));
4400                                 break;
4401                         }
4402                         default:
4403                                 panic("invalid subpage value %d",
4404                                       page_index->subpage);
4405                                 break;
4406                         }
4407                         break;
4408                 }
4409                 default:
4410                         panic("invalid page value %d",
4411                               page_index->page_code & SMPH_PC_MASK);
4412                         break;
4413         }
4414         }
4415
4416         return (CTL_RETVAL_COMPLETE);
4417 }
4418
4419 /*
4420  * LUN allocation.
4421  *
4422  * Requirements:
4423  * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4424  *   wants us to allocate the LUN and he can block.
4425  * - ctl_softc is always set
4426  * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4427  *
4428  * Returns 0 for success, non-zero (errno) for failure.
4429  */
4430 static int
4431 ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4432               struct ctl_be_lun *const be_lun, struct ctl_id target_id)
4433 {
4434         struct ctl_lun *nlun, *lun;
4435         struct ctl_port *port;
4436         struct scsi_vpd_id_descriptor *desc;
4437         struct scsi_vpd_id_t10 *t10id;
4438         const char *eui, *naa, *scsiname, *vendor;
4439         int lun_number, i, lun_malloced;
4440         int devidlen, idlen1, idlen2 = 0, len;
4441
4442         if (be_lun == NULL)
4443                 return (EINVAL);
4444
4445         /*
4446          * We currently only support Direct Access or Processor LUN types.
4447          */
4448         switch (be_lun->lun_type) {
4449         case T_DIRECT:
4450                 break;
4451         case T_PROCESSOR:
4452                 break;
4453         case T_SEQUENTIAL:
4454         case T_CHANGER:
4455         default:
4456                 be_lun->lun_config_status(be_lun->be_lun,
4457                                           CTL_LUN_CONFIG_FAILURE);
4458                 break;
4459         }
4460         if (ctl_lun == NULL) {
4461                 lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4462                 lun_malloced = 1;
4463         } else {
4464                 lun_malloced = 0;
4465                 lun = ctl_lun;
4466         }
4467
4468         memset(lun, 0, sizeof(*lun));
4469         if (lun_malloced)
4470                 lun->flags = CTL_LUN_MALLOCED;
4471
4472         /* Generate LUN ID. */
4473         devidlen = max(CTL_DEVID_MIN_LEN,
4474             strnlen(be_lun->device_id, CTL_DEVID_LEN));
4475         idlen1 = sizeof(*t10id) + devidlen;
4476         len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4477         scsiname = ctl_get_opt(&be_lun->options, "scsiname");
4478         if (scsiname != NULL) {
4479                 idlen2 = roundup2(strlen(scsiname) + 1, 4);
4480                 len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4481         }
4482         eui = ctl_get_opt(&be_lun->options, "eui");
4483         if (eui != NULL) {
4484                 len += sizeof(struct scsi_vpd_id_descriptor) + 8;
4485         }
4486         naa = ctl_get_opt(&be_lun->options, "naa");
4487         if (naa != NULL) {
4488                 len += sizeof(struct scsi_vpd_id_descriptor) + 8;
4489         }
4490         lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4491             M_CTL, M_WAITOK | M_ZERO);
4492         lun->lun_devid->len = len;
4493         desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4494         desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4495         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4496         desc->length = idlen1;
4497         t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4498         memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4499         if ((vendor = ctl_get_opt(&be_lun->options, "vendor")) == NULL) {
4500                 strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4501         } else {
4502                 strncpy(t10id->vendor, vendor,
4503                     min(sizeof(t10id->vendor), strlen(vendor)));
4504         }
4505         strncpy((char *)t10id->vendor_spec_id,
4506             (char *)be_lun->device_id, devidlen);
4507         if (scsiname != NULL) {
4508                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4509                     desc->length);
4510                 desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4511                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4512                     SVPD_ID_TYPE_SCSI_NAME;
4513                 desc->length = idlen2;
4514                 strlcpy(desc->identifier, scsiname, idlen2);
4515         }
4516         if (eui != NULL) {
4517                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4518                     desc->length);
4519                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4520                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4521                     SVPD_ID_TYPE_EUI64;
4522                 desc->length = 8;
4523                 scsi_u64to8b(strtouq(eui, NULL, 0), desc->identifier);
4524         }
4525         if (naa != NULL) {
4526                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4527                     desc->length);
4528                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4529                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4530                     SVPD_ID_TYPE_NAA;
4531                 desc->length = 8;
4532                 scsi_u64to8b(strtouq(naa, NULL, 0), desc->identifier);
4533         }
4534
4535         mtx_lock(&ctl_softc->ctl_lock);
4536         /*
4537          * See if the caller requested a particular LUN number.  If so, see
4538          * if it is available.  Otherwise, allocate the first available LUN.
4539          */
4540         if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4541                 if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4542                  || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4543                         mtx_unlock(&ctl_softc->ctl_lock);
4544                         if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4545                                 printf("ctl: requested LUN ID %d is higher "
4546                                        "than CTL_MAX_LUNS - 1 (%d)\n",
4547                                        be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4548                         } else {
4549                                 /*
4550                                  * XXX KDM return an error, or just assign
4551                                  * another LUN ID in this case??
4552                                  */
4553                                 printf("ctl: requested LUN ID %d is already "
4554                                        "in use\n", be_lun->req_lun_id);
4555                         }
4556                         if (lun->flags & CTL_LUN_MALLOCED)
4557                                 free(lun, M_CTL);
4558                         be_lun->lun_config_status(be_lun->be_lun,
4559                                                   CTL_LUN_CONFIG_FAILURE);
4560                         return (ENOSPC);
4561                 }
4562                 lun_number = be_lun->req_lun_id;
4563         } else {
4564                 lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, CTL_MAX_LUNS);
4565                 if (lun_number == -1) {
4566                         mtx_unlock(&ctl_softc->ctl_lock);
4567                         printf("ctl: can't allocate LUN on target %ju, out of "
4568                                "LUNs\n", (uintmax_t)target_id.id);
4569                         if (lun->flags & CTL_LUN_MALLOCED)
4570                                 free(lun, M_CTL);
4571                         be_lun->lun_config_status(be_lun->be_lun,
4572                                                   CTL_LUN_CONFIG_FAILURE);
4573                         return (ENOSPC);
4574                 }
4575         }
4576         ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4577
4578         mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4579         lun->target = target_id;
4580         lun->lun = lun_number;
4581         lun->be_lun = be_lun;
4582         /*
4583          * The processor LUN is always enabled.  Disk LUNs come on line
4584          * disabled, and must be enabled by the backend.
4585          */
4586         lun->flags |= CTL_LUN_DISABLED;
4587         lun->backend = be_lun->be;
4588         be_lun->ctl_lun = lun;
4589         be_lun->lun_id = lun_number;
4590         atomic_add_int(&be_lun->be->num_luns, 1);
4591         if (be_lun->flags & CTL_LUN_FLAG_POWERED_OFF)
4592                 lun->flags |= CTL_LUN_STOPPED;
4593
4594         if (be_lun->flags & CTL_LUN_FLAG_INOPERABLE)
4595                 lun->flags |= CTL_LUN_INOPERABLE;
4596
4597         if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4598                 lun->flags |= CTL_LUN_PRIMARY_SC;
4599
4600         lun->ctl_softc = ctl_softc;
4601         TAILQ_INIT(&lun->ooa_queue);
4602         TAILQ_INIT(&lun->blocked_queue);
4603         STAILQ_INIT(&lun->error_list);
4604         ctl_tpc_lun_init(lun);
4605
4606         /*
4607          * Initialize the mode page index.
4608          */
4609         ctl_init_page_index(lun);
4610
4611         /*
4612          * Set the poweron UA for all initiators on this LUN only.
4613          */
4614         for (i = 0; i < CTL_MAX_INITIATORS; i++)
4615                 lun->pending_ua[i] = CTL_UA_POWERON;
4616
4617         /*
4618          * Now, before we insert this lun on the lun list, set the lun
4619          * inventory changed UA for all other luns.
4620          */
4621         STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4622                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4623                         nlun->pending_ua[i] |= CTL_UA_LUN_CHANGE;
4624                 }
4625         }
4626
4627         STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4628
4629         ctl_softc->ctl_luns[lun_number] = lun;
4630
4631         ctl_softc->num_luns++;
4632
4633         /* Setup statistics gathering */
4634         lun->stats.device_type = be_lun->lun_type;
4635         lun->stats.lun_number = lun_number;
4636         if (lun->stats.device_type == T_DIRECT)
4637                 lun->stats.blocksize = be_lun->blocksize;
4638         else
4639                 lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4640         for (i = 0;i < CTL_MAX_PORTS;i++)
4641                 lun->stats.ports[i].targ_port = i;
4642
4643         mtx_unlock(&ctl_softc->ctl_lock);
4644
4645         lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4646
4647         /*
4648          * Run through each registered FETD and bring it online if it isn't
4649          * already.  Enable the target ID if it hasn't been enabled, and
4650          * enable this particular LUN.
4651          */
4652         STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4653                 int retval;
4654
4655                 retval = port->lun_enable(port->targ_lun_arg, target_id,lun_number);
4656                 if (retval != 0) {
4657                         printf("ctl_alloc_lun: FETD %s port %d returned error "
4658                                "%d for lun_enable on target %ju lun %d\n",
4659                                port->port_name, port->targ_port, retval,
4660                                (uintmax_t)target_id.id, lun_number);
4661                 } else
4662                         port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4663         }
4664         return (0);
4665 }
4666
4667 /*
4668  * Delete a LUN.
4669  * Assumptions:
4670  * - LUN has already been marked invalid and any pending I/O has been taken
4671  *   care of.
4672  */
4673 static int
4674 ctl_free_lun(struct ctl_lun *lun)
4675 {
4676         struct ctl_softc *softc;
4677 #if 0
4678         struct ctl_port *port;
4679 #endif
4680         struct ctl_lun *nlun;
4681         int i;
4682
4683         softc = lun->ctl_softc;
4684
4685         mtx_assert(&softc->ctl_lock, MA_OWNED);
4686
4687         STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4688
4689         ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4690
4691         softc->ctl_luns[lun->lun] = NULL;
4692
4693         if (!TAILQ_EMPTY(&lun->ooa_queue))
4694                 panic("Freeing a LUN %p with outstanding I/O!!\n", lun);
4695
4696         softc->num_luns--;
4697
4698         /*
4699          * XXX KDM this scheme only works for a single target/multiple LUN
4700          * setup.  It needs to be revamped for a multiple target scheme.
4701          *
4702          * XXX KDM this results in port->lun_disable() getting called twice,
4703          * once when ctl_disable_lun() is called, and a second time here.
4704          * We really need to re-think the LUN disable semantics.  There
4705          * should probably be several steps/levels to LUN removal:
4706          *  - disable
4707          *  - invalidate
4708          *  - free
4709          *
4710          * Right now we only have a disable method when communicating to
4711          * the front end ports, at least for individual LUNs.
4712          */
4713 #if 0
4714         STAILQ_FOREACH(port, &softc->port_list, links) {
4715                 int retval;
4716
4717                 retval = port->lun_disable(port->targ_lun_arg, lun->target,
4718                                          lun->lun);
4719                 if (retval != 0) {
4720                         printf("ctl_free_lun: FETD %s port %d returned error "
4721                                "%d for lun_disable on target %ju lun %jd\n",
4722                                port->port_name, port->targ_port, retval,
4723                                (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4724                 }
4725
4726                 if (STAILQ_FIRST(&softc->lun_list) == NULL) {
4727                         port->status &= ~CTL_PORT_STATUS_LUN_ONLINE;
4728
4729                         retval = port->targ_disable(port->targ_lun_arg,lun->target);
4730                         if (retval != 0) {
4731                                 printf("ctl_free_lun: FETD %s port %d "
4732                                        "returned error %d for targ_disable on "
4733                                        "target %ju\n", port->port_name,
4734                                        port->targ_port, retval,
4735                                        (uintmax_t)lun->target.id);
4736                         } else
4737                                 port->status &= ~CTL_PORT_STATUS_TARG_ONLINE;
4738
4739                         if ((port->status & CTL_PORT_STATUS_TARG_ONLINE) != 0)
4740                                 continue;
4741
4742 #if 0
4743                         port->port_offline(port->onoff_arg);
4744                         port->status &= ~CTL_PORT_STATUS_ONLINE;
4745 #endif
4746                 }
4747         }
4748 #endif
4749
4750         /*
4751          * Tell the backend to free resources, if this LUN has a backend.
4752          */
4753         atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4754         lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4755
4756         ctl_tpc_lun_shutdown(lun);
4757         mtx_destroy(&lun->lun_lock);
4758         free(lun->lun_devid, M_CTL);
4759         if (lun->flags & CTL_LUN_MALLOCED)
4760                 free(lun, M_CTL);
4761
4762         STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4763                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4764                         nlun->pending_ua[i] |= CTL_UA_LUN_CHANGE;
4765                 }
4766         }
4767
4768         return (0);
4769 }
4770
4771 static void
4772 ctl_create_lun(struct ctl_be_lun *be_lun)
4773 {
4774         struct ctl_softc *ctl_softc;
4775
4776         ctl_softc = control_softc;
4777
4778         /*
4779          * ctl_alloc_lun() should handle all potential failure cases.
4780          */
4781         ctl_alloc_lun(ctl_softc, NULL, be_lun, ctl_softc->target);
4782 }
4783
4784 int
4785 ctl_add_lun(struct ctl_be_lun *be_lun)
4786 {
4787         struct ctl_softc *ctl_softc = control_softc;
4788
4789         mtx_lock(&ctl_softc->ctl_lock);
4790         STAILQ_INSERT_TAIL(&ctl_softc->pending_lun_queue, be_lun, links);
4791         mtx_unlock(&ctl_softc->ctl_lock);
4792         wakeup(&ctl_softc->pending_lun_queue);
4793
4794         return (0);
4795 }
4796
4797 int
4798 ctl_enable_lun(struct ctl_be_lun *be_lun)
4799 {
4800         struct ctl_softc *ctl_softc;
4801         struct ctl_port *port, *nport;
4802         struct ctl_lun *lun;
4803         int retval;
4804
4805         ctl_softc = control_softc;
4806
4807         lun = (struct ctl_lun *)be_lun->ctl_lun;
4808
4809         mtx_lock(&ctl_softc->ctl_lock);
4810         mtx_lock(&lun->lun_lock);
4811         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4812                 /*
4813                  * eh?  Why did we get called if the LUN is already
4814                  * enabled?
4815                  */
4816                 mtx_unlock(&lun->lun_lock);
4817                 mtx_unlock(&ctl_softc->ctl_lock);
4818                 return (0);
4819         }
4820         lun->flags &= ~CTL_LUN_DISABLED;
4821         mtx_unlock(&lun->lun_lock);
4822
4823         for (port = STAILQ_FIRST(&ctl_softc->port_list); port != NULL; port = nport) {
4824                 nport = STAILQ_NEXT(port, links);
4825
4826                 /*
4827                  * Drop the lock while we call the FETD's enable routine.
4828                  * This can lead to a callback into CTL (at least in the
4829                  * case of the internal initiator frontend.
4830                  */
4831                 mtx_unlock(&ctl_softc->ctl_lock);
4832                 retval = port->lun_enable(port->targ_lun_arg, lun->target,lun->lun);
4833                 mtx_lock(&ctl_softc->ctl_lock);
4834                 if (retval != 0) {
4835                         printf("%s: FETD %s port %d returned error "
4836                                "%d for lun_enable on target %ju lun %jd\n",
4837                                __func__, port->port_name, port->targ_port, retval,
4838                                (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4839                 }
4840 #if 0
4841                  else {
4842             /* NOTE:  TODO:  why does lun enable affect port status? */
4843                         port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4844                 }
4845 #endif
4846         }
4847
4848         mtx_unlock(&ctl_softc->ctl_lock);
4849
4850         return (0);
4851 }
4852
4853 int
4854 ctl_disable_lun(struct ctl_be_lun *be_lun)
4855 {
4856         struct ctl_softc *ctl_softc;
4857         struct ctl_port *port;
4858         struct ctl_lun *lun;
4859         int retval;
4860
4861         ctl_softc = control_softc;
4862
4863         lun = (struct ctl_lun *)be_lun->ctl_lun;
4864
4865         mtx_lock(&ctl_softc->ctl_lock);
4866         mtx_lock(&lun->lun_lock);
4867         if (lun->flags & CTL_LUN_DISABLED) {
4868                 mtx_unlock(&lun->lun_lock);
4869                 mtx_unlock(&ctl_softc->ctl_lock);
4870                 return (0);
4871         }
4872         lun->flags |= CTL_LUN_DISABLED;
4873         mtx_unlock(&lun->lun_lock);
4874
4875         STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4876                 mtx_unlock(&ctl_softc->ctl_lock);
4877                 /*
4878                  * Drop the lock before we call the frontend's disable
4879                  * routine, to avoid lock order reversals.
4880                  *
4881                  * XXX KDM what happens if the frontend list changes while
4882                  * we're traversing it?  It's unlikely, but should be handled.
4883                  */
4884                 retval = port->lun_disable(port->targ_lun_arg, lun->target,
4885                                          lun->lun);
4886                 mtx_lock(&ctl_softc->ctl_lock);
4887                 if (retval != 0) {
4888                         printf("ctl_alloc_lun: FETD %s port %d returned error "
4889                                "%d for lun_disable on target %ju lun %jd\n",
4890                                port->port_name, port->targ_port, retval,
4891                                (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4892                 }
4893         }
4894
4895         mtx_unlock(&ctl_softc->ctl_lock);
4896
4897         return (0);
4898 }
4899
4900 int
4901 ctl_start_lun(struct ctl_be_lun *be_lun)
4902 {
4903         struct ctl_softc *ctl_softc;
4904         struct ctl_lun *lun;
4905
4906         ctl_softc = control_softc;
4907
4908         lun = (struct ctl_lun *)be_lun->ctl_lun;
4909
4910         mtx_lock(&lun->lun_lock);
4911         lun->flags &= ~CTL_LUN_STOPPED;
4912         mtx_unlock(&lun->lun_lock);
4913
4914         return (0);
4915 }
4916
4917 int
4918 ctl_stop_lun(struct ctl_be_lun *be_lun)
4919 {
4920         struct ctl_softc *ctl_softc;
4921         struct ctl_lun *lun;
4922
4923         ctl_softc = control_softc;
4924
4925         lun = (struct ctl_lun *)be_lun->ctl_lun;
4926
4927         mtx_lock(&lun->lun_lock);
4928         lun->flags |= CTL_LUN_STOPPED;
4929         mtx_unlock(&lun->lun_lock);
4930
4931         return (0);
4932 }
4933
4934 int
4935 ctl_lun_offline(struct ctl_be_lun *be_lun)
4936 {
4937         struct ctl_softc *ctl_softc;
4938         struct ctl_lun *lun;
4939
4940         ctl_softc = control_softc;
4941
4942         lun = (struct ctl_lun *)be_lun->ctl_lun;
4943
4944         mtx_lock(&lun->lun_lock);
4945         lun->flags |= CTL_LUN_OFFLINE;
4946         mtx_unlock(&lun->lun_lock);
4947
4948         return (0);
4949 }
4950
4951 int
4952 ctl_lun_online(struct ctl_be_lun *be_lun)
4953 {
4954         struct ctl_softc *ctl_softc;
4955         struct ctl_lun *lun;
4956
4957         ctl_softc = control_softc;
4958
4959         lun = (struct ctl_lun *)be_lun->ctl_lun;
4960
4961         mtx_lock(&lun->lun_lock);
4962         lun->flags &= ~CTL_LUN_OFFLINE;
4963         mtx_unlock(&lun->lun_lock);
4964
4965         return (0);
4966 }
4967
4968 int
4969 ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4970 {
4971         struct ctl_softc *ctl_softc;
4972         struct ctl_lun *lun;
4973
4974         ctl_softc = control_softc;
4975
4976         lun = (struct ctl_lun *)be_lun->ctl_lun;
4977
4978         mtx_lock(&lun->lun_lock);
4979
4980         /*
4981          * The LUN needs to be disabled before it can be marked invalid.
4982          */
4983         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4984                 mtx_unlock(&lun->lun_lock);
4985                 return (-1);
4986         }
4987         /*
4988          * Mark the LUN invalid.
4989          */
4990         lun->flags |= CTL_LUN_INVALID;
4991
4992         /*
4993          * If there is nothing in the OOA queue, go ahead and free the LUN.
4994          * If we have something in the OOA queue, we'll free it when the
4995          * last I/O completes.
4996          */
4997         if (TAILQ_EMPTY(&lun->ooa_queue)) {
4998                 mtx_unlock(&lun->lun_lock);
4999                 mtx_lock(&ctl_softc->ctl_lock);
5000                 ctl_free_lun(lun);
5001                 mtx_unlock(&ctl_softc->ctl_lock);
5002         } else
5003                 mtx_unlock(&lun->lun_lock);
5004
5005         return (0);
5006 }
5007
5008 int
5009 ctl_lun_inoperable(struct ctl_be_lun *be_lun)
5010 {
5011         struct ctl_softc *ctl_softc;
5012         struct ctl_lun *lun;
5013
5014         ctl_softc = control_softc;
5015         lun = (struct ctl_lun *)be_lun->ctl_lun;
5016
5017         mtx_lock(&lun->lun_lock);
5018         lun->flags |= CTL_LUN_INOPERABLE;
5019         mtx_unlock(&lun->lun_lock);
5020
5021         return (0);
5022 }
5023
5024 int
5025 ctl_lun_operable(struct ctl_be_lun *be_lun)
5026 {
5027         struct ctl_softc *ctl_softc;
5028         struct ctl_lun *lun;
5029
5030         ctl_softc = control_softc;
5031         lun = (struct ctl_lun *)be_lun->ctl_lun;
5032
5033         mtx_lock(&lun->lun_lock);
5034         lun->flags &= ~CTL_LUN_INOPERABLE;
5035         mtx_unlock(&lun->lun_lock);
5036
5037         return (0);
5038 }
5039
5040 int
5041 ctl_lun_power_lock(struct ctl_be_lun *be_lun, struct ctl_nexus *nexus,
5042                    int lock)
5043 {
5044         struct ctl_softc *softc;
5045         struct ctl_lun *lun;
5046         struct copan_aps_subpage *current_sp;
5047         struct ctl_page_index *page_index;
5048         int i;
5049
5050         softc = control_softc;
5051
5052         mtx_lock(&softc->ctl_lock);
5053
5054         lun = (struct ctl_lun *)be_lun->ctl_lun;
5055         mtx_lock(&lun->lun_lock);
5056
5057         page_index = NULL;
5058         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
5059                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
5060                      APS_PAGE_CODE)
5061                         continue;
5062
5063                 if (lun->mode_pages.index[i].subpage != APS_SUBPAGE_CODE)
5064                         continue;
5065                 page_index = &lun->mode_pages.index[i];
5066         }
5067
5068         if (page_index == NULL) {
5069                 mtx_unlock(&lun->lun_lock);
5070                 mtx_unlock(&softc->ctl_lock);
5071                 printf("%s: APS subpage not found for lun %ju!\n", __func__,
5072                        (uintmax_t)lun->lun);
5073                 return (1);
5074         }
5075 #if 0
5076         if ((softc->aps_locked_lun != 0)
5077          && (softc->aps_locked_lun != lun->lun)) {
5078                 printf("%s: attempt to lock LUN %llu when %llu is already "
5079                        "locked\n");
5080                 mtx_unlock(&lun->lun_lock);
5081                 mtx_unlock(&softc->ctl_lock);
5082                 return (1);
5083         }
5084 #endif
5085
5086         current_sp = (struct copan_aps_subpage *)(page_index->page_data +
5087                 (page_index->page_len * CTL_PAGE_CURRENT));
5088
5089         if (lock != 0) {
5090                 current_sp->lock_active = APS_LOCK_ACTIVE;
5091                 softc->aps_locked_lun = lun->lun;
5092         } else {
5093                 current_sp->lock_active = 0;
5094                 softc->aps_locked_lun = 0;
5095         }
5096
5097
5098         /*
5099          * If we're in HA mode, try to send the lock message to the other
5100          * side.
5101          */
5102         if (ctl_is_single == 0) {
5103                 int isc_retval;
5104                 union ctl_ha_msg lock_msg;
5105
5106                 lock_msg.hdr.nexus = *nexus;
5107                 lock_msg.hdr.msg_type = CTL_MSG_APS_LOCK;
5108                 if (lock != 0)
5109                         lock_msg.aps.lock_flag = 1;
5110                 else
5111                         lock_msg.aps.lock_flag = 0;
5112                 isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &lock_msg,
5113                                          sizeof(lock_msg), 0);
5114                 if (isc_retval > CTL_HA_STATUS_SUCCESS) {
5115                         printf("%s: APS (lock=%d) error returned from "
5116                                "ctl_ha_msg_send: %d\n", __func__, lock, isc_retval);
5117                         mtx_unlock(&lun->lun_lock);
5118                         mtx_unlock(&softc->ctl_lock);
5119                         return (1);
5120                 }
5121         }
5122
5123         mtx_unlock(&lun->lun_lock);
5124         mtx_unlock(&softc->ctl_lock);
5125
5126         return (0);
5127 }
5128
5129 void
5130 ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
5131 {
5132         struct ctl_lun *lun;
5133         struct ctl_softc *softc;
5134         int i;
5135
5136         softc = control_softc;
5137
5138         lun = (struct ctl_lun *)be_lun->ctl_lun;
5139
5140         mtx_lock(&lun->lun_lock);
5141
5142         for (i = 0; i < CTL_MAX_INITIATORS; i++) 
5143                 lun->pending_ua[i] |= CTL_UA_CAPACITY_CHANGED;
5144
5145         mtx_unlock(&lun->lun_lock);
5146 }
5147
5148 /*
5149  * Backend "memory move is complete" callback for requests that never
5150  * make it down to say RAIDCore's configuration code.
5151  */
5152 int
5153 ctl_config_move_done(union ctl_io *io)
5154 {
5155         int retval;
5156
5157         retval = CTL_RETVAL_COMPLETE;
5158
5159
5160         CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
5161         /*
5162          * XXX KDM this shouldn't happen, but what if it does?
5163          */
5164         if (io->io_hdr.io_type != CTL_IO_SCSI)
5165                 panic("I/O type isn't CTL_IO_SCSI!");
5166
5167         if ((io->io_hdr.port_status == 0)
5168          && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
5169          && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE))
5170                 io->io_hdr.status = CTL_SUCCESS;
5171         else if ((io->io_hdr.port_status != 0)
5172               && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
5173               && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)){
5174                 /*
5175                  * For hardware error sense keys, the sense key
5176                  * specific value is defined to be a retry count,
5177                  * but we use it to pass back an internal FETD
5178                  * error code.  XXX KDM  Hopefully the FETD is only
5179                  * using 16 bits for an error code, since that's
5180                  * all the space we have in the sks field.
5181                  */
5182                 ctl_set_internal_failure(&io->scsiio,
5183                                          /*sks_valid*/ 1,
5184                                          /*retry_count*/
5185                                          io->io_hdr.port_status);
5186                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5187                         free(io->scsiio.kern_data_ptr, M_CTL);
5188                 ctl_done(io);
5189                 goto bailout;
5190         }
5191
5192         if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
5193          || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
5194          || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
5195                 /*
5196                  * XXX KDM just assuming a single pointer here, and not a
5197                  * S/G list.  If we start using S/G lists for config data,
5198                  * we'll need to know how to clean them up here as well.
5199                  */
5200                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5201                         free(io->scsiio.kern_data_ptr, M_CTL);
5202                 /* Hopefully the user has already set the status... */
5203                 ctl_done(io);
5204         } else {
5205                 /*
5206                  * XXX KDM now we need to continue data movement.  Some
5207                  * options:
5208                  * - call ctl_scsiio() again?  We don't do this for data
5209                  *   writes, because for those at least we know ahead of
5210                  *   time where the write will go and how long it is.  For
5211                  *   config writes, though, that information is largely
5212                  *   contained within the write itself, thus we need to
5213                  *   parse out the data again.
5214                  *
5215                  * - Call some other function once the data is in?
5216                  */
5217
5218                 /*
5219                  * XXX KDM call ctl_scsiio() again for now, and check flag
5220                  * bits to see whether we're allocated or not.
5221                  */
5222                 retval = ctl_scsiio(&io->scsiio);
5223         }
5224 bailout:
5225         return (retval);
5226 }
5227
5228 /*
5229  * This gets called by a backend driver when it is done with a
5230  * data_submit method.
5231  */
5232 void
5233 ctl_data_submit_done(union ctl_io *io)
5234 {
5235         /*
5236          * If the IO_CONT flag is set, we need to call the supplied
5237          * function to continue processing the I/O, instead of completing
5238          * the I/O just yet.
5239          *
5240          * If there is an error, though, we don't want to keep processing.
5241          * Instead, just send status back to the initiator.
5242          */
5243         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5244             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5245             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5246              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5247                 io->scsiio.io_cont(io);
5248                 return;
5249         }
5250         ctl_done(io);
5251 }
5252
5253 /*
5254  * This gets called by a backend driver when it is done with a
5255  * configuration write.
5256  */
5257 void
5258 ctl_config_write_done(union ctl_io *io)
5259 {
5260         /*
5261          * If the IO_CONT flag is set, we need to call the supplied
5262          * function to continue processing the I/O, instead of completing
5263          * the I/O just yet.
5264          *
5265          * If there is an error, though, we don't want to keep processing.
5266          * Instead, just send status back to the initiator.
5267          */
5268         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT)
5269          && (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)
5270           || ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))) {
5271                 io->scsiio.io_cont(io);
5272                 return;
5273         }
5274         /*
5275          * Since a configuration write can be done for commands that actually
5276          * have data allocated, like write buffer, and commands that have
5277          * no data, like start/stop unit, we need to check here.
5278          */
5279         if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT)
5280                 free(io->scsiio.kern_data_ptr, M_CTL);
5281         ctl_done(io);
5282 }
5283
5284 /*
5285  * SCSI release command.
5286  */
5287 int
5288 ctl_scsi_release(struct ctl_scsiio *ctsio)
5289 {
5290         int length, longid, thirdparty_id, resv_id;
5291         struct ctl_softc *ctl_softc;
5292         struct ctl_lun *lun;
5293
5294         length = 0;
5295         resv_id = 0;
5296
5297         CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5298
5299         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5300         ctl_softc = control_softc;
5301
5302         switch (ctsio->cdb[0]) {
5303         case RELEASE_10: {
5304                 struct scsi_release_10 *cdb;
5305
5306                 cdb = (struct scsi_release_10 *)ctsio->cdb;
5307
5308                 if (cdb->byte2 & SR10_LONGID)
5309                         longid = 1;
5310                 else
5311                         thirdparty_id = cdb->thirdparty_id;
5312
5313                 resv_id = cdb->resv_id;
5314                 length = scsi_2btoul(cdb->length);
5315                 break;
5316         }
5317         }
5318
5319
5320         /*
5321          * XXX KDM right now, we only support LUN reservation.  We don't
5322          * support 3rd party reservations, or extent reservations, which
5323          * might actually need the parameter list.  If we've gotten this
5324          * far, we've got a LUN reservation.  Anything else got kicked out
5325          * above.  So, according to SPC, ignore the length.
5326          */
5327         length = 0;
5328
5329         if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5330          && (length > 0)) {
5331                 ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5332                 ctsio->kern_data_len = length;
5333                 ctsio->kern_total_len = length;
5334                 ctsio->kern_data_resid = 0;
5335                 ctsio->kern_rel_offset = 0;
5336                 ctsio->kern_sg_entries = 0;
5337                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5338                 ctsio->be_move_done = ctl_config_move_done;
5339                 ctl_datamove((union ctl_io *)ctsio);
5340
5341                 return (CTL_RETVAL_COMPLETE);
5342         }
5343
5344         if (length > 0)
5345                 thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5346
5347         mtx_lock(&lun->lun_lock);
5348
5349         /*
5350          * According to SPC, it is not an error for an intiator to attempt
5351          * to release a reservation on a LUN that isn't reserved, or that
5352          * is reserved by another initiator.  The reservation can only be
5353          * released, though, by the initiator who made it or by one of
5354          * several reset type events.
5355          */
5356         if (lun->flags & CTL_LUN_RESERVED) {
5357                 if ((ctsio->io_hdr.nexus.initid.id == lun->rsv_nexus.initid.id)
5358                  && (ctsio->io_hdr.nexus.targ_port == lun->rsv_nexus.targ_port)
5359                  && (ctsio->io_hdr.nexus.targ_target.id ==
5360                      lun->rsv_nexus.targ_target.id)) {
5361                         lun->flags &= ~CTL_LUN_RESERVED;
5362                 }
5363         }
5364
5365         mtx_unlock(&lun->lun_lock);
5366
5367         ctsio->scsi_status = SCSI_STATUS_OK;
5368         ctsio->io_hdr.status = CTL_SUCCESS;
5369
5370         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5371                 free(ctsio->kern_data_ptr, M_CTL);
5372                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5373         }
5374
5375         ctl_done((union ctl_io *)ctsio);
5376         return (CTL_RETVAL_COMPLETE);
5377 }
5378
5379 int
5380 ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5381 {
5382         int extent, thirdparty, longid;
5383         int resv_id, length;
5384         uint64_t thirdparty_id;
5385         struct ctl_softc *ctl_softc;
5386         struct ctl_lun *lun;
5387
5388         extent = 0;
5389         thirdparty = 0;
5390         longid = 0;
5391         resv_id = 0;
5392         length = 0;
5393         thirdparty_id = 0;
5394
5395         CTL_DEBUG_PRINT(("ctl_reserve\n"));
5396
5397         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5398         ctl_softc = control_softc;
5399
5400         switch (ctsio->cdb[0]) {
5401         case RESERVE_10: {
5402                 struct scsi_reserve_10 *cdb;
5403
5404                 cdb = (struct scsi_reserve_10 *)ctsio->cdb;
5405
5406                 if (cdb->byte2 & SR10_LONGID)
5407                         longid = 1;
5408                 else
5409                         thirdparty_id = cdb->thirdparty_id;
5410
5411                 resv_id = cdb->resv_id;
5412                 length = scsi_2btoul(cdb->length);
5413                 break;
5414         }
5415         }
5416
5417         /*
5418          * XXX KDM right now, we only support LUN reservation.  We don't
5419          * support 3rd party reservations, or extent reservations, which
5420          * might actually need the parameter list.  If we've gotten this
5421          * far, we've got a LUN reservation.  Anything else got kicked out
5422          * above.  So, according to SPC, ignore the length.
5423          */
5424         length = 0;
5425
5426         if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5427          && (length > 0)) {
5428                 ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5429                 ctsio->kern_data_len = length;
5430                 ctsio->kern_total_len = length;
5431                 ctsio->kern_data_resid = 0;
5432                 ctsio->kern_rel_offset = 0;
5433                 ctsio->kern_sg_entries = 0;
5434                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5435                 ctsio->be_move_done = ctl_config_move_done;
5436                 ctl_datamove((union ctl_io *)ctsio);
5437
5438                 return (CTL_RETVAL_COMPLETE);
5439         }
5440
5441         if (length > 0)
5442                 thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5443
5444         mtx_lock(&lun->lun_lock);
5445         if (lun->flags & CTL_LUN_RESERVED) {
5446                 if ((ctsio->io_hdr.nexus.initid.id != lun->rsv_nexus.initid.id)
5447                  || (ctsio->io_hdr.nexus.targ_port != lun->rsv_nexus.targ_port)
5448                  || (ctsio->io_hdr.nexus.targ_target.id !=
5449                      lun->rsv_nexus.targ_target.id)) {
5450                         ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
5451                         ctsio->io_hdr.status = CTL_SCSI_ERROR;
5452                         goto bailout;
5453                 }
5454         }
5455
5456         lun->flags |= CTL_LUN_RESERVED;
5457         lun->rsv_nexus = ctsio->io_hdr.nexus;
5458
5459         ctsio->scsi_status = SCSI_STATUS_OK;
5460         ctsio->io_hdr.status = CTL_SUCCESS;
5461
5462 bailout:
5463         mtx_unlock(&lun->lun_lock);
5464
5465         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5466                 free(ctsio->kern_data_ptr, M_CTL);
5467                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5468         }
5469
5470         ctl_done((union ctl_io *)ctsio);
5471         return (CTL_RETVAL_COMPLETE);
5472 }
5473
5474 int
5475 ctl_start_stop(struct ctl_scsiio *ctsio)
5476 {
5477         struct scsi_start_stop_unit *cdb;
5478         struct ctl_lun *lun;
5479         struct ctl_softc *ctl_softc;
5480         int retval;
5481
5482         CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5483
5484         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5485         ctl_softc = control_softc;
5486         retval = 0;
5487
5488         cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5489
5490         /*
5491          * XXX KDM
5492          * We don't support the immediate bit on a stop unit.  In order to
5493          * do that, we would need to code up a way to know that a stop is
5494          * pending, and hold off any new commands until it completes, one
5495          * way or another.  Then we could accept or reject those commands
5496          * depending on its status.  We would almost need to do the reverse
5497          * of what we do below for an immediate start -- return the copy of
5498          * the ctl_io to the FETD with status to send to the host (and to
5499          * free the copy!) and then free the original I/O once the stop
5500          * actually completes.  That way, the OOA queue mechanism can work
5501          * to block commands that shouldn't proceed.  Another alternative
5502          * would be to put the copy in the queue in place of the original,
5503          * and return the original back to the caller.  That could be
5504          * slightly safer..
5505          */
5506         if ((cdb->byte2 & SSS_IMMED)
5507          && ((cdb->how & SSS_START) == 0)) {
5508                 ctl_set_invalid_field(ctsio,
5509                                       /*sks_valid*/ 1,
5510                                       /*command*/ 1,
5511                                       /*field*/ 1,
5512                                       /*bit_valid*/ 1,
5513                                       /*bit*/ 0);
5514                 ctl_done((union ctl_io *)ctsio);
5515                 return (CTL_RETVAL_COMPLETE);
5516         }
5517
5518         if ((lun->flags & CTL_LUN_PR_RESERVED)
5519          && ((cdb->how & SSS_START)==0)) {
5520                 uint32_t residx;
5521
5522                 residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5523                 if (!lun->per_res[residx].registered
5524                  || (lun->pr_res_idx!=residx && lun->res_type < 4)) {
5525
5526                         ctl_set_reservation_conflict(ctsio);
5527                         ctl_done((union ctl_io *)ctsio);
5528                         return (CTL_RETVAL_COMPLETE);
5529                 }
5530         }
5531
5532         /*
5533          * If there is no backend on this device, we can't start or stop
5534          * it.  In theory we shouldn't get any start/stop commands in the
5535          * first place at this level if the LUN doesn't have a backend.
5536          * That should get stopped by the command decode code.
5537          */
5538         if (lun->backend == NULL) {
5539                 ctl_set_invalid_opcode(ctsio);
5540                 ctl_done((union ctl_io *)ctsio);
5541                 return (CTL_RETVAL_COMPLETE);
5542         }
5543
5544         /*
5545          * XXX KDM Copan-specific offline behavior.
5546          * Figure out a reasonable way to port this?
5547          */
5548 #ifdef NEEDTOPORT
5549         mtx_lock(&lun->lun_lock);
5550
5551         if (((cdb->byte2 & SSS_ONOFFLINE) == 0)
5552          && (lun->flags & CTL_LUN_OFFLINE)) {
5553                 /*
5554                  * If the LUN is offline, and the on/offline bit isn't set,
5555                  * reject the start or stop.  Otherwise, let it through.
5556                  */
5557                 mtx_unlock(&lun->lun_lock);
5558                 ctl_set_lun_not_ready(ctsio);
5559                 ctl_done((union ctl_io *)ctsio);
5560         } else {
5561                 mtx_unlock(&lun->lun_lock);
5562 #endif /* NEEDTOPORT */
5563                 /*
5564                  * This could be a start or a stop when we're online,
5565                  * or a stop/offline or start/online.  A start or stop when
5566                  * we're offline is covered in the case above.
5567                  */
5568                 /*
5569                  * In the non-immediate case, we send the request to
5570                  * the backend and return status to the user when
5571                  * it is done.
5572                  *
5573                  * In the immediate case, we allocate a new ctl_io
5574                  * to hold a copy of the request, and send that to
5575                  * the backend.  We then set good status on the
5576                  * user's request and return it immediately.
5577                  */
5578                 if (cdb->byte2 & SSS_IMMED) {
5579                         union ctl_io *new_io;
5580
5581                         new_io = ctl_alloc_io(ctsio->io_hdr.pool);
5582                         if (new_io == NULL) {
5583                                 ctl_set_busy(ctsio);
5584                                 ctl_done((union ctl_io *)ctsio);
5585                         } else {
5586                                 ctl_copy_io((union ctl_io *)ctsio,
5587                                             new_io);
5588                                 retval = lun->backend->config_write(new_io);
5589                                 ctl_set_success(ctsio);
5590                                 ctl_done((union ctl_io *)ctsio);
5591                         }
5592                 } else {
5593                         retval = lun->backend->config_write(
5594                                 (union ctl_io *)ctsio);
5595                 }
5596 #ifdef NEEDTOPORT
5597         }
5598 #endif
5599         return (retval);
5600 }
5601
5602 /*
5603  * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5604  * we don't really do anything with the LBA and length fields if the user
5605  * passes them in.  Instead we'll just flush out the cache for the entire
5606  * LUN.
5607  */
5608 int
5609 ctl_sync_cache(struct ctl_scsiio *ctsio)
5610 {
5611         struct ctl_lun *lun;
5612         struct ctl_softc *ctl_softc;
5613         uint64_t starting_lba;
5614         uint32_t block_count;
5615         int retval;
5616
5617         CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5618
5619         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5620         ctl_softc = control_softc;
5621         retval = 0;
5622
5623         switch (ctsio->cdb[0]) {
5624         case SYNCHRONIZE_CACHE: {
5625                 struct scsi_sync_cache *cdb;
5626                 cdb = (struct scsi_sync_cache *)ctsio->cdb;
5627
5628                 starting_lba = scsi_4btoul(cdb->begin_lba);
5629                 block_count = scsi_2btoul(cdb->lb_count);
5630                 break;
5631         }
5632         case SYNCHRONIZE_CACHE_16: {
5633                 struct scsi_sync_cache_16 *cdb;
5634                 cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5635
5636                 starting_lba = scsi_8btou64(cdb->begin_lba);
5637                 block_count = scsi_4btoul(cdb->lb_count);
5638                 break;
5639         }
5640         default:
5641                 ctl_set_invalid_opcode(ctsio);
5642                 ctl_done((union ctl_io *)ctsio);
5643                 goto bailout;
5644                 break; /* NOTREACHED */
5645         }
5646
5647         /*
5648          * We check the LBA and length, but don't do anything with them.
5649          * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5650          * get flushed.  This check will just help satisfy anyone who wants
5651          * to see an error for an out of range LBA.
5652          */
5653         if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5654                 ctl_set_lba_out_of_range(ctsio);
5655                 ctl_done((union ctl_io *)ctsio);
5656                 goto bailout;
5657         }
5658
5659         /*
5660          * If this LUN has no backend, we can't flush the cache anyway.
5661          */
5662         if (lun->backend == NULL) {
5663                 ctl_set_invalid_opcode(ctsio);
5664                 ctl_done((union ctl_io *)ctsio);
5665                 goto bailout;
5666         }
5667
5668         /*
5669          * Check to see whether we're configured to send the SYNCHRONIZE
5670          * CACHE command directly to the back end.
5671          */
5672         mtx_lock(&lun->lun_lock);
5673         if ((ctl_softc->flags & CTL_FLAG_REAL_SYNC)
5674          && (++(lun->sync_count) >= lun->sync_interval)) {
5675                 lun->sync_count = 0;
5676                 mtx_unlock(&lun->lun_lock);
5677                 retval = lun->backend->config_write((union ctl_io *)ctsio);
5678         } else {
5679                 mtx_unlock(&lun->lun_lock);
5680                 ctl_set_success(ctsio);
5681                 ctl_done((union ctl_io *)ctsio);
5682         }
5683
5684 bailout:
5685
5686         return (retval);
5687 }
5688
5689 int
5690 ctl_format(struct ctl_scsiio *ctsio)
5691 {
5692         struct scsi_format *cdb;
5693         struct ctl_lun *lun;
5694         struct ctl_softc *ctl_softc;
5695         int length, defect_list_len;
5696
5697         CTL_DEBUG_PRINT(("ctl_format\n"));
5698
5699         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5700         ctl_softc = control_softc;
5701
5702         cdb = (struct scsi_format *)ctsio->cdb;
5703
5704         length = 0;
5705         if (cdb->byte2 & SF_FMTDATA) {
5706                 if (cdb->byte2 & SF_LONGLIST)
5707                         length = sizeof(struct scsi_format_header_long);
5708                 else
5709                         length = sizeof(struct scsi_format_header_short);
5710         }
5711
5712         if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5713          && (length > 0)) {
5714                 ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5715                 ctsio->kern_data_len = length;
5716                 ctsio->kern_total_len = length;
5717                 ctsio->kern_data_resid = 0;
5718                 ctsio->kern_rel_offset = 0;
5719                 ctsio->kern_sg_entries = 0;
5720                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5721                 ctsio->be_move_done = ctl_config_move_done;
5722                 ctl_datamove((union ctl_io *)ctsio);
5723
5724                 return (CTL_RETVAL_COMPLETE);
5725         }
5726
5727         defect_list_len = 0;
5728
5729         if (cdb->byte2 & SF_FMTDATA) {
5730                 if (cdb->byte2 & SF_LONGLIST) {
5731                         struct scsi_format_header_long *header;
5732
5733                         header = (struct scsi_format_header_long *)
5734                                 ctsio->kern_data_ptr;
5735
5736                         defect_list_len = scsi_4btoul(header->defect_list_len);
5737                         if (defect_list_len != 0) {
5738                                 ctl_set_invalid_field(ctsio,
5739                                                       /*sks_valid*/ 1,
5740                                                       /*command*/ 0,
5741                                                       /*field*/ 2,
5742                                                       /*bit_valid*/ 0,
5743                                                       /*bit*/ 0);
5744                                 goto bailout;
5745                         }
5746                 } else {
5747                         struct scsi_format_header_short *header;
5748
5749                         header = (struct scsi_format_header_short *)
5750                                 ctsio->kern_data_ptr;
5751
5752                         defect_list_len = scsi_2btoul(header->defect_list_len);
5753                         if (defect_list_len != 0) {
5754                                 ctl_set_invalid_field(ctsio,
5755                                                       /*sks_valid*/ 1,
5756                                                       /*command*/ 0,
5757                                                       /*field*/ 2,
5758                                                       /*bit_valid*/ 0,
5759                                                       /*bit*/ 0);
5760                                 goto bailout;
5761                         }
5762                 }
5763         }
5764
5765         /*
5766          * The format command will clear out the "Medium format corrupted"
5767          * status if set by the configuration code.  That status is really
5768          * just a way to notify the host that we have lost the media, and
5769          * get them to issue a command that will basically make them think
5770          * they're blowing away the media.
5771          */
5772         mtx_lock(&lun->lun_lock);
5773         lun->flags &= ~CTL_LUN_INOPERABLE;
5774         mtx_unlock(&lun->lun_lock);
5775
5776         ctsio->scsi_status = SCSI_STATUS_OK;
5777         ctsio->io_hdr.status = CTL_SUCCESS;
5778 bailout:
5779
5780         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5781                 free(ctsio->kern_data_ptr, M_CTL);
5782                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5783         }
5784
5785         ctl_done((union ctl_io *)ctsio);
5786         return (CTL_RETVAL_COMPLETE);
5787 }
5788
5789 int
5790 ctl_read_buffer(struct ctl_scsiio *ctsio)
5791 {
5792         struct scsi_read_buffer *cdb;
5793         struct ctl_lun *lun;
5794         int buffer_offset, len;
5795         static uint8_t descr[4];
5796         static uint8_t echo_descr[4] = { 0 };
5797
5798         CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5799
5800         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5801         cdb = (struct scsi_read_buffer *)ctsio->cdb;
5802
5803         if (lun->flags & CTL_LUN_PR_RESERVED) {
5804                 uint32_t residx;
5805
5806                 /*
5807                  * XXX KDM need a lock here.
5808                  */
5809                 residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5810                 if ((lun->res_type == SPR_TYPE_EX_AC
5811                   && residx != lun->pr_res_idx)
5812                  || ((lun->res_type == SPR_TYPE_EX_AC_RO
5813                    || lun->res_type == SPR_TYPE_EX_AC_AR)
5814                   && !lun->per_res[residx].registered)) {
5815                         ctl_set_reservation_conflict(ctsio);
5816                         ctl_done((union ctl_io *)ctsio);
5817                         return (CTL_RETVAL_COMPLETE);
5818                 }
5819         }
5820
5821         if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA &&
5822             (cdb->byte2 & RWB_MODE) != RWB_MODE_ECHO_DESCR &&
5823             (cdb->byte2 & RWB_MODE) != RWB_MODE_DESCR) {
5824                 ctl_set_invalid_field(ctsio,
5825                                       /*sks_valid*/ 1,
5826                                       /*command*/ 1,
5827                                       /*field*/ 1,
5828                                       /*bit_valid*/ 1,
5829                                       /*bit*/ 4);
5830                 ctl_done((union ctl_io *)ctsio);
5831                 return (CTL_RETVAL_COMPLETE);
5832         }
5833
5834         len = scsi_3btoul(cdb->length);
5835         buffer_offset = scsi_3btoul(cdb->offset);
5836
5837         if (buffer_offset + len > sizeof(lun->write_buffer)) {
5838                 ctl_set_invalid_field(ctsio,
5839                                       /*sks_valid*/ 1,
5840                                       /*command*/ 1,
5841                                       /*field*/ 6,
5842                                       /*bit_valid*/ 0,
5843                                       /*bit*/ 0);
5844                 ctl_done((union ctl_io *)ctsio);
5845                 return (CTL_RETVAL_COMPLETE);
5846         }
5847
5848         if ((cdb->byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5849                 descr[0] = 0;
5850                 scsi_ulto3b(sizeof(lun->write_buffer), &descr[1]);
5851                 ctsio->kern_data_ptr = descr;
5852                 len = min(len, sizeof(descr));
5853         } else if ((cdb->byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5854                 ctsio->kern_data_ptr = echo_descr;
5855                 len = min(len, sizeof(echo_descr));
5856         } else
5857                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5858         ctsio->kern_data_len = len;
5859         ctsio->kern_total_len = len;
5860         ctsio->kern_data_resid = 0;
5861         ctsio->kern_rel_offset = 0;
5862         ctsio->kern_sg_entries = 0;
5863         ctsio->be_move_done = ctl_config_move_done;
5864         ctl_datamove((union ctl_io *)ctsio);
5865
5866         return (CTL_RETVAL_COMPLETE);
5867 }
5868
5869 int
5870 ctl_write_buffer(struct ctl_scsiio *ctsio)
5871 {
5872         struct scsi_write_buffer *cdb;
5873         struct ctl_lun *lun;
5874         int buffer_offset, len;
5875
5876         CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5877
5878         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5879         cdb = (struct scsi_write_buffer *)ctsio->cdb;
5880
5881         if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5882                 ctl_set_invalid_field(ctsio,
5883                                       /*sks_valid*/ 1,
5884                                       /*command*/ 1,
5885                                       /*field*/ 1,
5886                                       /*bit_valid*/ 1,
5887                                       /*bit*/ 4);
5888                 ctl_done((union ctl_io *)ctsio);
5889                 return (CTL_RETVAL_COMPLETE);
5890         }
5891
5892         len = scsi_3btoul(cdb->length);
5893         buffer_offset = scsi_3btoul(cdb->offset);
5894
5895         if (buffer_offset + len > sizeof(lun->write_buffer)) {
5896                 ctl_set_invalid_field(ctsio,
5897                                       /*sks_valid*/ 1,
5898                                       /*command*/ 1,
5899                                       /*field*/ 6,
5900                                       /*bit_valid*/ 0,
5901                                       /*bit*/ 0);
5902                 ctl_done((union ctl_io *)ctsio);
5903                 return (CTL_RETVAL_COMPLETE);
5904         }
5905
5906         /*
5907          * If we've got a kernel request that hasn't been malloced yet,
5908          * malloc it and tell the caller the data buffer is here.
5909          */
5910         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5911                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5912                 ctsio->kern_data_len = len;
5913                 ctsio->kern_total_len = len;
5914                 ctsio->kern_data_resid = 0;
5915                 ctsio->kern_rel_offset = 0;
5916                 ctsio->kern_sg_entries = 0;
5917                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5918                 ctsio->be_move_done = ctl_config_move_done;
5919                 ctl_datamove((union ctl_io *)ctsio);
5920
5921                 return (CTL_RETVAL_COMPLETE);
5922         }
5923
5924         ctl_done((union ctl_io *)ctsio);
5925
5926         return (CTL_RETVAL_COMPLETE);
5927 }
5928
5929 int
5930 ctl_write_same(struct ctl_scsiio *ctsio)
5931 {
5932         struct ctl_lun *lun;
5933         struct ctl_lba_len_flags *lbalen;
5934         uint64_t lba;
5935         uint32_t num_blocks;
5936         int len, retval;
5937         uint8_t byte2;
5938
5939         retval = CTL_RETVAL_COMPLETE;
5940
5941         CTL_DEBUG_PRINT(("ctl_write_same\n"));
5942
5943         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5944
5945         switch (ctsio->cdb[0]) {
5946         case WRITE_SAME_10: {
5947                 struct scsi_write_same_10 *cdb;
5948
5949                 cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5950
5951                 lba = scsi_4btoul(cdb->addr);
5952                 num_blocks = scsi_2btoul(cdb->length);
5953                 byte2 = cdb->byte2;
5954                 break;
5955         }
5956         case WRITE_SAME_16: {
5957                 struct scsi_write_same_16 *cdb;
5958
5959                 cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5960
5961                 lba = scsi_8btou64(cdb->addr);
5962                 num_blocks = scsi_4btoul(cdb->length);
5963                 byte2 = cdb->byte2;
5964                 break;
5965         }
5966         default:
5967                 /*
5968                  * We got a command we don't support.  This shouldn't
5969                  * happen, commands should be filtered out above us.
5970                  */
5971                 ctl_set_invalid_opcode(ctsio);
5972                 ctl_done((union ctl_io *)ctsio);
5973
5974                 return (CTL_RETVAL_COMPLETE);
5975                 break; /* NOTREACHED */
5976         }
5977
5978         /*
5979          * The first check is to make sure we're in bounds, the second
5980          * check is to catch wrap-around problems.  If the lba + num blocks
5981          * is less than the lba, then we've wrapped around and the block
5982          * range is invalid anyway.
5983          */
5984         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5985          || ((lba + num_blocks) < lba)) {
5986                 ctl_set_lba_out_of_range(ctsio);
5987                 ctl_done((union ctl_io *)ctsio);
5988                 return (CTL_RETVAL_COMPLETE);
5989         }
5990
5991         /* Zero number of blocks means "to the last logical block" */
5992         if (num_blocks == 0) {
5993                 if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5994                         ctl_set_invalid_field(ctsio,
5995                                               /*sks_valid*/ 0,
5996                                               /*command*/ 1,
5997                                               /*field*/ 0,
5998                                               /*bit_valid*/ 0,
5999                                               /*bit*/ 0);
6000                         ctl_done((union ctl_io *)ctsio);
6001                         return (CTL_RETVAL_COMPLETE);
6002                 }
6003                 num_blocks = (lun->be_lun->maxlba + 1) - lba;
6004         }
6005
6006         len = lun->be_lun->blocksize;
6007
6008         /*
6009          * If we've got a kernel request that hasn't been malloced yet,
6010          * malloc it and tell the caller the data buffer is here.
6011          */
6012         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6013                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
6014                 ctsio->kern_data_len = len;
6015                 ctsio->kern_total_len = len;
6016                 ctsio->kern_data_resid = 0;
6017                 ctsio->kern_rel_offset = 0;
6018                 ctsio->kern_sg_entries = 0;
6019                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6020                 ctsio->be_move_done = ctl_config_move_done;
6021                 ctl_datamove((union ctl_io *)ctsio);
6022
6023                 return (CTL_RETVAL_COMPLETE);
6024         }
6025
6026         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
6027         lbalen->lba = lba;
6028         lbalen->len = num_blocks;
6029         lbalen->flags = byte2;
6030         retval = lun->backend->config_write((union ctl_io *)ctsio);
6031
6032         return (retval);
6033 }
6034
6035 int
6036 ctl_unmap(struct ctl_scsiio *ctsio)
6037 {
6038         struct ctl_lun *lun;
6039         struct scsi_unmap *cdb;
6040         struct ctl_ptr_len_flags *ptrlen;
6041         struct scsi_unmap_header *hdr;
6042         struct scsi_unmap_desc *buf, *end, *endnz, *range;
6043         uint64_t lba;
6044         uint32_t num_blocks;
6045         int len, retval;
6046         uint8_t byte2;
6047
6048         retval = CTL_RETVAL_COMPLETE;
6049
6050         CTL_DEBUG_PRINT(("ctl_unmap\n"));
6051
6052         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6053         cdb = (struct scsi_unmap *)ctsio->cdb;
6054
6055         len = scsi_2btoul(cdb->length);
6056         byte2 = cdb->byte2;
6057
6058         /*
6059          * If we've got a kernel request that hasn't been malloced yet,
6060          * malloc it and tell the caller the data buffer is here.
6061          */
6062         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6063                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
6064                 ctsio->kern_data_len = len;
6065                 ctsio->kern_total_len = len;
6066                 ctsio->kern_data_resid = 0;
6067                 ctsio->kern_rel_offset = 0;
6068                 ctsio->kern_sg_entries = 0;
6069                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6070                 ctsio->be_move_done = ctl_config_move_done;
6071                 ctl_datamove((union ctl_io *)ctsio);
6072
6073                 return (CTL_RETVAL_COMPLETE);
6074         }
6075
6076         len = ctsio->kern_total_len - ctsio->kern_data_resid;
6077         hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
6078         if (len < sizeof (*hdr) ||
6079             len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
6080             len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
6081             scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
6082                 ctl_set_invalid_field(ctsio,
6083                                       /*sks_valid*/ 0,
6084                                       /*command*/ 0,
6085                                       /*field*/ 0,
6086                                       /*bit_valid*/ 0,
6087                                       /*bit*/ 0);
6088                 ctl_done((union ctl_io *)ctsio);
6089                 return (CTL_RETVAL_COMPLETE);
6090         }
6091         len = scsi_2btoul(hdr->desc_length);
6092         buf = (struct scsi_unmap_desc *)(hdr + 1);
6093         end = buf + len / sizeof(*buf);
6094
6095         endnz = buf;
6096         for (range = buf; range < end; range++) {
6097                 lba = scsi_8btou64(range->lba);
6098                 num_blocks = scsi_4btoul(range->length);
6099                 if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
6100                  || ((lba + num_blocks) < lba)) {
6101                         ctl_set_lba_out_of_range(ctsio);
6102                         ctl_done((union ctl_io *)ctsio);
6103                         return (CTL_RETVAL_COMPLETE);
6104                 }
6105                 if (num_blocks != 0)
6106                         endnz = range + 1;
6107         }
6108
6109         /*
6110          * Block backend can not handle zero last range.
6111          * Filter it out and return if there is nothing left.
6112          */
6113         len = (uint8_t *)endnz - (uint8_t *)buf;
6114         if (len == 0) {
6115                 ctl_set_success(ctsio);
6116                 ctl_done((union ctl_io *)ctsio);
6117                 return (CTL_RETVAL_COMPLETE);
6118         }
6119
6120         ptrlen = (struct ctl_ptr_len_flags *)
6121             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
6122         ptrlen->ptr = (void *)buf;
6123         ptrlen->len = len;
6124         ptrlen->flags = byte2;
6125
6126         retval = lun->backend->config_write((union ctl_io *)ctsio);
6127         return (retval);
6128 }
6129
6130 /*
6131  * Note that this function currently doesn't actually do anything inside
6132  * CTL to enforce things if the DQue bit is turned on.
6133  *
6134  * Also note that this function can't be used in the default case, because
6135  * the DQue bit isn't set in the changeable mask for the control mode page
6136  * anyway.  This is just here as an example for how to implement a page
6137  * handler, and a placeholder in case we want to allow the user to turn
6138  * tagged queueing on and off.
6139  *
6140  * The D_SENSE bit handling is functional, however, and will turn
6141  * descriptor sense on and off for a given LUN.
6142  */
6143 int
6144 ctl_control_page_handler(struct ctl_scsiio *ctsio,
6145                          struct ctl_page_index *page_index, uint8_t *page_ptr)
6146 {
6147         struct scsi_control_page *current_cp, *saved_cp, *user_cp;
6148         struct ctl_lun *lun;
6149         struct ctl_softc *softc;
6150         int set_ua;
6151         uint32_t initidx;
6152
6153         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6154         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6155         set_ua = 0;
6156
6157         user_cp = (struct scsi_control_page *)page_ptr;
6158         current_cp = (struct scsi_control_page *)
6159                 (page_index->page_data + (page_index->page_len *
6160                 CTL_PAGE_CURRENT));
6161         saved_cp = (struct scsi_control_page *)
6162                 (page_index->page_data + (page_index->page_len *
6163                 CTL_PAGE_SAVED));
6164
6165         softc = control_softc;
6166
6167         mtx_lock(&lun->lun_lock);
6168         if (((current_cp->rlec & SCP_DSENSE) == 0)
6169          && ((user_cp->rlec & SCP_DSENSE) != 0)) {
6170                 /*
6171                  * Descriptor sense is currently turned off and the user
6172                  * wants to turn it on.
6173                  */
6174                 current_cp->rlec |= SCP_DSENSE;
6175                 saved_cp->rlec |= SCP_DSENSE;
6176                 lun->flags |= CTL_LUN_SENSE_DESC;
6177                 set_ua = 1;
6178         } else if (((current_cp->rlec & SCP_DSENSE) != 0)
6179                 && ((user_cp->rlec & SCP_DSENSE) == 0)) {
6180                 /*
6181                  * Descriptor sense is currently turned on, and the user
6182                  * wants to turn it off.
6183                  */
6184                 current_cp->rlec &= ~SCP_DSENSE;
6185                 saved_cp->rlec &= ~SCP_DSENSE;
6186                 lun->flags &= ~CTL_LUN_SENSE_DESC;
6187                 set_ua = 1;
6188         }
6189         if (current_cp->queue_flags & SCP_QUEUE_DQUE) {
6190                 if (user_cp->queue_flags & SCP_QUEUE_DQUE) {
6191 #ifdef NEEDTOPORT
6192                         csevent_log(CSC_CTL | CSC_SHELF_SW |
6193                                     CTL_UNTAG_TO_UNTAG,
6194                                     csevent_LogType_Trace,
6195                                     csevent_Severity_Information,
6196                                     csevent_AlertLevel_Green,
6197                                     csevent_FRU_Firmware,
6198                                     csevent_FRU_Unknown,
6199                                     "Received untagged to untagged transition");
6200 #endif /* NEEDTOPORT */
6201                 } else {
6202 #ifdef NEEDTOPORT
6203                         csevent_log(CSC_CTL | CSC_SHELF_SW |
6204                                     CTL_UNTAG_TO_TAG,
6205                                     csevent_LogType_ConfigChange,
6206                                     csevent_Severity_Information,
6207                                     csevent_AlertLevel_Green,
6208                                     csevent_FRU_Firmware,
6209                                     csevent_FRU_Unknown,
6210                                     "Received untagged to tagged "
6211                                     "queueing transition");
6212 #endif /* NEEDTOPORT */
6213
6214                         current_cp->queue_flags &= ~SCP_QUEUE_DQUE;
6215                         saved_cp->queue_flags &= ~SCP_QUEUE_DQUE;
6216                         set_ua = 1;
6217                 }
6218         } else {
6219                 if (user_cp->queue_flags & SCP_QUEUE_DQUE) {
6220 #ifdef NEEDTOPORT
6221                         csevent_log(CSC_CTL | CSC_SHELF_SW |
6222                                     CTL_TAG_TO_UNTAG,
6223                                     csevent_LogType_ConfigChange,
6224                                     csevent_Severity_Warning,
6225                                     csevent_AlertLevel_Yellow,
6226                                     csevent_FRU_Firmware,
6227                                     csevent_FRU_Unknown,
6228                                     "Received tagged queueing to untagged "
6229                                     "transition");
6230 #endif /* NEEDTOPORT */
6231
6232                         current_cp->queue_flags |= SCP_QUEUE_DQUE;
6233                         saved_cp->queue_flags |= SCP_QUEUE_DQUE;
6234                         set_ua = 1;
6235                 } else {
6236 #ifdef NEEDTOPORT
6237                         csevent_log(CSC_CTL | CSC_SHELF_SW |
6238                                     CTL_TAG_TO_TAG,
6239                                     csevent_LogType_Trace,
6240                                     csevent_Severity_Information,
6241                                     csevent_AlertLevel_Green,
6242                                     csevent_FRU_Firmware,
6243                                     csevent_FRU_Unknown,
6244                                     "Received tagged queueing to tagged "
6245                                     "queueing transition");
6246 #endif /* NEEDTOPORT */
6247                 }
6248         }
6249         if (set_ua != 0) {
6250                 int i;
6251                 /*
6252                  * Let other initiators know that the mode
6253                  * parameters for this LUN have changed.
6254                  */
6255                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
6256                         if (i == initidx)
6257                                 continue;
6258
6259                         lun->pending_ua[i] |= CTL_UA_MODE_CHANGE;
6260                 }
6261         }
6262         mtx_unlock(&lun->lun_lock);
6263
6264         return (0);
6265 }
6266
6267 int
6268 ctl_power_sp_handler(struct ctl_scsiio *ctsio,
6269                      struct ctl_page_index *page_index, uint8_t *page_ptr)
6270 {
6271         return (0);
6272 }
6273
6274 int
6275 ctl_power_sp_sense_handler(struct ctl_scsiio *ctsio,
6276                            struct ctl_page_index *page_index, int pc)
6277 {
6278         struct copan_power_subpage *page;
6279
6280         page = (struct copan_power_subpage *)page_index->page_data +
6281                 (page_index->page_len * pc);
6282
6283         switch (pc) {
6284         case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6285                 /*
6286                  * We don't update the changable bits for this page.
6287                  */
6288                 break;
6289         case SMS_PAGE_CTRL_CURRENT >> 6:
6290         case SMS_PAGE_CTRL_DEFAULT >> 6:
6291         case SMS_PAGE_CTRL_SAVED >> 6:
6292 #ifdef NEEDTOPORT
6293                 ctl_update_power_subpage(page);
6294 #endif
6295                 break;
6296         default:
6297 #ifdef NEEDTOPORT
6298                 EPRINT(0, "Invalid PC %d!!", pc);
6299 #endif
6300                 break;
6301         }
6302         return (0);
6303 }
6304
6305
6306 int
6307 ctl_aps_sp_handler(struct ctl_scsiio *ctsio,
6308                    struct ctl_page_index *page_index, uint8_t *page_ptr)
6309 {
6310         struct copan_aps_subpage *user_sp;
6311         struct copan_aps_subpage *current_sp;
6312         union ctl_modepage_info *modepage_info;
6313         struct ctl_softc *softc;
6314         struct ctl_lun *lun;
6315         int retval;
6316
6317         retval = CTL_RETVAL_COMPLETE;
6318         current_sp = (struct copan_aps_subpage *)(page_index->page_data +
6319                      (page_index->page_len * CTL_PAGE_CURRENT));
6320         softc = control_softc;
6321         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6322
6323         user_sp = (struct copan_aps_subpage *)page_ptr;
6324
6325         modepage_info = (union ctl_modepage_info *)
6326                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6327
6328         modepage_info->header.page_code = page_index->page_code & SMPH_PC_MASK;
6329         modepage_info->header.subpage = page_index->subpage;
6330         modepage_info->aps.lock_active = user_sp->lock_active;
6331
6332         mtx_lock(&softc->ctl_lock);
6333
6334         /*
6335          * If there is a request to lock the LUN and another LUN is locked
6336          * this is an error. If the requested LUN is already locked ignore
6337          * the request. If no LUN is locked attempt to lock it.
6338          * if there is a request to unlock the LUN and the LUN is currently
6339          * locked attempt to unlock it. Otherwise ignore the request. i.e.
6340          * if another LUN is locked or no LUN is locked.
6341          */
6342         if (user_sp->lock_active & APS_LOCK_ACTIVE) {
6343                 if (softc->aps_locked_lun == lun->lun) {
6344                         /*
6345                          * This LUN is already locked, so we're done.
6346                          */
6347                         retval = CTL_RETVAL_COMPLETE;
6348                 } else if (softc->aps_locked_lun == 0) {
6349                         /*
6350                          * No one has the lock, pass the request to the
6351                          * backend.
6352                          */
6353                         retval = lun->backend->config_write(
6354                                 (union ctl_io *)ctsio);
6355                 } else {
6356                         /*
6357                          * Someone else has the lock, throw out the request.
6358                          */
6359                         ctl_set_already_locked(ctsio);
6360                         free(ctsio->kern_data_ptr, M_CTL);
6361                         ctl_done((union ctl_io *)ctsio);
6362
6363                         /*
6364                          * Set the return value so that ctl_do_mode_select()
6365                          * won't try to complete the command.  We already
6366                          * completed it here.
6367                          */
6368                         retval = CTL_RETVAL_ERROR;
6369                 }
6370         } else if (softc->aps_locked_lun == lun->lun) {
6371                 /*
6372                  * This LUN is locked, so pass the unlock request to the
6373                  * backend.
6374                  */
6375                 retval = lun->backend->config_write((union ctl_io *)ctsio);
6376         }
6377         mtx_unlock(&softc->ctl_lock);
6378
6379         return (retval);
6380 }
6381
6382 int
6383 ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio,
6384                                 struct ctl_page_index *page_index,
6385                                 uint8_t *page_ptr)
6386 {
6387         uint8_t *c;
6388         int i;
6389
6390         c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs;
6391         ctl_time_io_secs =
6392                 (c[0] << 8) |
6393                 (c[1] << 0) |
6394                 0;
6395         CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs));
6396         printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs);
6397         printf("page data:");
6398         for (i=0; i<8; i++)
6399                 printf(" %.2x",page_ptr[i]);
6400         printf("\n");
6401         return (0);
6402 }
6403
6404 int
6405 ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio,
6406                                struct ctl_page_index *page_index,
6407                                int pc)
6408 {
6409         struct copan_debugconf_subpage *page;
6410
6411         page = (struct copan_debugconf_subpage *)page_index->page_data +
6412                 (page_index->page_len * pc);
6413
6414         switch (pc) {
6415         case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6416         case SMS_PAGE_CTRL_DEFAULT >> 6:
6417         case SMS_PAGE_CTRL_SAVED >> 6:
6418                 /*
6419                  * We don't update the changable or default bits for this page.
6420                  */
6421                 break;
6422         case SMS_PAGE_CTRL_CURRENT >> 6:
6423                 page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8;
6424                 page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0;
6425                 break;
6426         default:
6427 #ifdef NEEDTOPORT
6428                 EPRINT(0, "Invalid PC %d!!", pc);
6429 #endif /* NEEDTOPORT */
6430                 break;
6431         }
6432         return (0);
6433 }
6434
6435
6436 static int
6437 ctl_do_mode_select(union ctl_io *io)
6438 {
6439         struct scsi_mode_page_header *page_header;
6440         struct ctl_page_index *page_index;
6441         struct ctl_scsiio *ctsio;
6442         int control_dev, page_len;
6443         int page_len_offset, page_len_size;
6444         union ctl_modepage_info *modepage_info;
6445         struct ctl_lun *lun;
6446         int *len_left, *len_used;
6447         int retval, i;
6448
6449         ctsio = &io->scsiio;
6450         page_index = NULL;
6451         page_len = 0;
6452         retval = CTL_RETVAL_COMPLETE;
6453
6454         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6455
6456         if (lun->be_lun->lun_type != T_DIRECT)
6457                 control_dev = 1;
6458         else
6459                 control_dev = 0;
6460
6461         modepage_info = (union ctl_modepage_info *)
6462                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6463         len_left = &modepage_info->header.len_left;
6464         len_used = &modepage_info->header.len_used;
6465
6466 do_next_page:
6467
6468         page_header = (struct scsi_mode_page_header *)
6469                 (ctsio->kern_data_ptr + *len_used);
6470
6471         if (*len_left == 0) {
6472                 free(ctsio->kern_data_ptr, M_CTL);
6473                 ctl_set_success(ctsio);
6474                 ctl_done((union ctl_io *)ctsio);
6475                 return (CTL_RETVAL_COMPLETE);
6476         } else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6477
6478                 free(ctsio->kern_data_ptr, M_CTL);
6479                 ctl_set_param_len_error(ctsio);
6480                 ctl_done((union ctl_io *)ctsio);
6481                 return (CTL_RETVAL_COMPLETE);
6482
6483         } else if ((page_header->page_code & SMPH_SPF)
6484                 && (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6485
6486                 free(ctsio->kern_data_ptr, M_CTL);
6487                 ctl_set_param_len_error(ctsio);
6488                 ctl_done((union ctl_io *)ctsio);
6489                 return (CTL_RETVAL_COMPLETE);
6490         }
6491
6492
6493         /*
6494          * XXX KDM should we do something with the block descriptor?
6495          */
6496         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6497
6498                 if ((control_dev != 0)
6499                  && (lun->mode_pages.index[i].page_flags &
6500                      CTL_PAGE_FLAG_DISK_ONLY))
6501                         continue;
6502
6503                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
6504                     (page_header->page_code & SMPH_PC_MASK))
6505                         continue;
6506
6507                 /*
6508                  * If neither page has a subpage code, then we've got a
6509                  * match.
6510                  */
6511                 if (((lun->mode_pages.index[i].page_code & SMPH_SPF) == 0)
6512                  && ((page_header->page_code & SMPH_SPF) == 0)) {
6513                         page_index = &lun->mode_pages.index[i];
6514                         page_len = page_header->page_length;
6515                         break;
6516                 }
6517
6518                 /*
6519                  * If both pages have subpages, then the subpage numbers
6520                  * have to match.
6521                  */
6522                 if ((lun->mode_pages.index[i].page_code & SMPH_SPF)
6523                   && (page_header->page_code & SMPH_SPF)) {
6524                         struct scsi_mode_page_header_sp *sph;
6525
6526                         sph = (struct scsi_mode_page_header_sp *)page_header;
6527
6528                         if (lun->mode_pages.index[i].subpage ==
6529                             sph->subpage) {
6530                                 page_index = &lun->mode_pages.index[i];
6531                                 page_len = scsi_2btoul(sph->page_length);
6532                                 break;
6533                         }
6534                 }
6535         }
6536
6537         /*
6538          * If we couldn't find the page, or if we don't have a mode select
6539          * handler for it, send back an error to the user.
6540          */
6541         if ((page_index == NULL)
6542          || (page_index->select_handler == NULL)) {
6543                 ctl_set_invalid_field(ctsio,
6544                                       /*sks_valid*/ 1,
6545                                       /*command*/ 0,
6546                                       /*field*/ *len_used,
6547                                       /*bit_valid*/ 0,
6548                                       /*bit*/ 0);
6549                 free(ctsio->kern_data_ptr, M_CTL);
6550                 ctl_done((union ctl_io *)ctsio);
6551                 return (CTL_RETVAL_COMPLETE);
6552         }
6553
6554         if (page_index->page_code & SMPH_SPF) {
6555                 page_len_offset = 2;
6556                 page_len_size = 2;
6557         } else {
6558                 page_len_size = 1;
6559                 page_len_offset = 1;
6560         }
6561
6562         /*
6563          * If the length the initiator gives us isn't the one we specify in
6564          * the mode page header, or if they didn't specify enough data in
6565          * the CDB to avoid truncating this page, kick out the request.
6566          */
6567         if ((page_len != (page_index->page_len - page_len_offset -
6568                           page_len_size))
6569          || (*len_left < page_index->page_len)) {
6570
6571
6572                 ctl_set_invalid_field(ctsio,
6573                                       /*sks_valid*/ 1,
6574                                       /*command*/ 0,
6575                                       /*field*/ *len_used + page_len_offset,
6576                                       /*bit_valid*/ 0,
6577                                       /*bit*/ 0);
6578                 free(ctsio->kern_data_ptr, M_CTL);
6579                 ctl_done((union ctl_io *)ctsio);
6580                 return (CTL_RETVAL_COMPLETE);
6581         }
6582
6583         /*
6584          * Run through the mode page, checking to make sure that the bits
6585          * the user changed are actually legal for him to change.
6586          */
6587         for (i = 0; i < page_index->page_len; i++) {
6588                 uint8_t *user_byte, *change_mask, *current_byte;
6589                 int bad_bit;
6590                 int j;
6591
6592                 user_byte = (uint8_t *)page_header + i;
6593                 change_mask = page_index->page_data +
6594                               (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6595                 current_byte = page_index->page_data +
6596                                (page_index->page_len * CTL_PAGE_CURRENT) + i;
6597
6598                 /*
6599                  * Check to see whether the user set any bits in this byte
6600                  * that he is not allowed to set.
6601                  */
6602                 if ((*user_byte & ~(*change_mask)) ==
6603                     (*current_byte & ~(*change_mask)))
6604                         continue;
6605
6606                 /*
6607                  * Go through bit by bit to determine which one is illegal.
6608                  */
6609                 bad_bit = 0;
6610                 for (j = 7; j >= 0; j--) {
6611                         if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6612                             (((1 << i) & ~(*change_mask)) & *current_byte)) {
6613                                 bad_bit = i;
6614                                 break;
6615                         }
6616                 }
6617                 ctl_set_invalid_field(ctsio,
6618                                       /*sks_valid*/ 1,
6619                                       /*command*/ 0,
6620                                       /*field*/ *len_used + i,
6621                                       /*bit_valid*/ 1,
6622                                       /*bit*/ bad_bit);
6623                 free(ctsio->kern_data_ptr, M_CTL);
6624                 ctl_done((union ctl_io *)ctsio);
6625                 return (CTL_RETVAL_COMPLETE);
6626         }
6627
6628         /*
6629          * Decrement these before we call the page handler, since we may
6630          * end up getting called back one way or another before the handler
6631          * returns to this context.
6632          */
6633         *len_left -= page_index->page_len;
6634         *len_used += page_index->page_len;
6635
6636         retval = page_index->select_handler(ctsio, page_index,
6637                                             (uint8_t *)page_header);
6638
6639         /*
6640          * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6641          * wait until this queued command completes to finish processing
6642          * the mode page.  If it returns anything other than
6643          * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6644          * already set the sense information, freed the data pointer, and
6645          * completed the io for us.
6646          */
6647         if (retval != CTL_RETVAL_COMPLETE)
6648                 goto bailout_no_done;
6649
6650         /*
6651          * If the initiator sent us more than one page, parse the next one.
6652          */
6653         if (*len_left > 0)
6654                 goto do_next_page;
6655
6656         ctl_set_success(ctsio);
6657         free(ctsio->kern_data_ptr, M_CTL);
6658         ctl_done((union ctl_io *)ctsio);
6659
6660 bailout_no_done:
6661
6662         return (CTL_RETVAL_COMPLETE);
6663
6664 }
6665
6666 int
6667 ctl_mode_select(struct ctl_scsiio *ctsio)
6668 {
6669         int param_len, pf, sp;
6670         int header_size, bd_len;
6671         int len_left, len_used;
6672         struct ctl_page_index *page_index;
6673         struct ctl_lun *lun;
6674         int control_dev, page_len;
6675         union ctl_modepage_info *modepage_info;
6676         int retval;
6677
6678         pf = 0;
6679         sp = 0;
6680         page_len = 0;
6681         len_used = 0;
6682         len_left = 0;
6683         retval = 0;
6684         bd_len = 0;
6685         page_index = NULL;
6686
6687         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6688
6689         if (lun->be_lun->lun_type != T_DIRECT)
6690                 control_dev = 1;
6691         else
6692                 control_dev = 0;
6693
6694         switch (ctsio->cdb[0]) {
6695         case MODE_SELECT_6: {
6696                 struct scsi_mode_select_6 *cdb;
6697
6698                 cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6699
6700                 pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6701                 sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6702
6703                 param_len = cdb->length;
6704                 header_size = sizeof(struct scsi_mode_header_6);
6705                 break;
6706         }
6707         case MODE_SELECT_10: {
6708                 struct scsi_mode_select_10 *cdb;
6709
6710                 cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6711
6712                 pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6713                 sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6714
6715                 param_len = scsi_2btoul(cdb->length);
6716                 header_size = sizeof(struct scsi_mode_header_10);
6717                 break;
6718         }
6719         default:
6720                 ctl_set_invalid_opcode(ctsio);
6721                 ctl_done((union ctl_io *)ctsio);
6722                 return (CTL_RETVAL_COMPLETE);
6723                 break; /* NOTREACHED */
6724         }
6725
6726         /*
6727          * From SPC-3:
6728          * "A parameter list length of zero indicates that the Data-Out Buffer
6729          * shall be empty. This condition shall not be considered as an error."
6730          */
6731         if (param_len == 0) {
6732                 ctl_set_success(ctsio);
6733                 ctl_done((union ctl_io *)ctsio);
6734                 return (CTL_RETVAL_COMPLETE);
6735         }
6736
6737         /*
6738          * Since we'll hit this the first time through, prior to
6739          * allocation, we don't need to free a data buffer here.
6740          */
6741         if (param_len < header_size) {
6742                 ctl_set_param_len_error(ctsio);
6743                 ctl_done((union ctl_io *)ctsio);
6744                 return (CTL_RETVAL_COMPLETE);
6745         }
6746
6747         /*
6748          * Allocate the data buffer and grab the user's data.  In theory,
6749          * we shouldn't have to sanity check the parameter list length here
6750          * because the maximum size is 64K.  We should be able to malloc
6751          * that much without too many problems.
6752          */
6753         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6754                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6755                 ctsio->kern_data_len = param_len;
6756                 ctsio->kern_total_len = param_len;
6757                 ctsio->kern_data_resid = 0;
6758                 ctsio->kern_rel_offset = 0;
6759                 ctsio->kern_sg_entries = 0;
6760                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6761                 ctsio->be_move_done = ctl_config_move_done;
6762                 ctl_datamove((union ctl_io *)ctsio);
6763
6764                 return (CTL_RETVAL_COMPLETE);
6765         }
6766
6767         switch (ctsio->cdb[0]) {
6768         case MODE_SELECT_6: {
6769                 struct scsi_mode_header_6 *mh6;
6770
6771                 mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6772                 bd_len = mh6->blk_desc_len;
6773                 break;
6774         }
6775         case MODE_SELECT_10: {
6776                 struct scsi_mode_header_10 *mh10;
6777
6778                 mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6779                 bd_len = scsi_2btoul(mh10->blk_desc_len);
6780                 break;
6781         }
6782         default:
6783                 panic("Invalid CDB type %#x", ctsio->cdb[0]);
6784                 break;
6785         }
6786
6787         if (param_len < (header_size + bd_len)) {
6788                 free(ctsio->kern_data_ptr, M_CTL);
6789                 ctl_set_param_len_error(ctsio);
6790                 ctl_done((union ctl_io *)ctsio);
6791                 return (CTL_RETVAL_COMPLETE);
6792         }
6793
6794         /*
6795          * Set the IO_CONT flag, so that if this I/O gets passed to
6796          * ctl_config_write_done(), it'll get passed back to
6797          * ctl_do_mode_select() for further processing, or completion if
6798          * we're all done.
6799          */
6800         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6801         ctsio->io_cont = ctl_do_mode_select;
6802
6803         modepage_info = (union ctl_modepage_info *)
6804                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6805
6806         memset(modepage_info, 0, sizeof(*modepage_info));
6807
6808         len_left = param_len - header_size - bd_len;
6809         len_used = header_size + bd_len;
6810
6811         modepage_info->header.len_left = len_left;
6812         modepage_info->header.len_used = len_used;
6813
6814         return (ctl_do_mode_select((union ctl_io *)ctsio));
6815 }
6816
6817 int
6818 ctl_mode_sense(struct ctl_scsiio *ctsio)
6819 {
6820         struct ctl_lun *lun;
6821         int pc, page_code, dbd, llba, subpage;
6822         int alloc_len, page_len, header_len, total_len;
6823         struct scsi_mode_block_descr *block_desc;
6824         struct ctl_page_index *page_index;
6825         int control_dev;
6826
6827         dbd = 0;
6828         llba = 0;
6829         block_desc = NULL;
6830         page_index = NULL;
6831
6832         CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6833
6834         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6835
6836         if (lun->be_lun->lun_type != T_DIRECT)
6837                 control_dev = 1;
6838         else
6839                 control_dev = 0;
6840
6841         if (lun->flags & CTL_LUN_PR_RESERVED) {
6842                 uint32_t residx;
6843
6844                 /*
6845                  * XXX KDM need a lock here.
6846                  */
6847                 residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
6848                 if ((lun->res_type == SPR_TYPE_EX_AC
6849                   && residx != lun->pr_res_idx)
6850                  || ((lun->res_type == SPR_TYPE_EX_AC_RO
6851                    || lun->res_type == SPR_TYPE_EX_AC_AR)
6852                   && !lun->per_res[residx].registered)) {
6853                         ctl_set_reservation_conflict(ctsio);
6854                         ctl_done((union ctl_io *)ctsio);
6855                         return (CTL_RETVAL_COMPLETE);
6856                 }
6857         }
6858
6859         switch (ctsio->cdb[0]) {
6860         case MODE_SENSE_6: {
6861                 struct scsi_mode_sense_6 *cdb;
6862
6863                 cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6864
6865                 header_len = sizeof(struct scsi_mode_hdr_6);
6866                 if (cdb->byte2 & SMS_DBD)
6867                         dbd = 1;
6868                 else
6869                         header_len += sizeof(struct scsi_mode_block_descr);
6870
6871                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6872                 page_code = cdb->page & SMS_PAGE_CODE;
6873                 subpage = cdb->subpage;
6874                 alloc_len = cdb->length;
6875                 break;
6876         }
6877         case MODE_SENSE_10: {
6878                 struct scsi_mode_sense_10 *cdb;
6879
6880                 cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6881
6882                 header_len = sizeof(struct scsi_mode_hdr_10);
6883
6884                 if (cdb->byte2 & SMS_DBD)
6885                         dbd = 1;
6886                 else
6887                         header_len += sizeof(struct scsi_mode_block_descr);
6888                 if (cdb->byte2 & SMS10_LLBAA)
6889                         llba = 1;
6890                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6891                 page_code = cdb->page & SMS_PAGE_CODE;
6892                 subpage = cdb->subpage;
6893                 alloc_len = scsi_2btoul(cdb->length);
6894                 break;
6895         }
6896         default:
6897                 ctl_set_invalid_opcode(ctsio);
6898                 ctl_done((union ctl_io *)ctsio);
6899                 return (CTL_RETVAL_COMPLETE);
6900                 break; /* NOTREACHED */
6901         }
6902
6903         /*
6904          * We have to make a first pass through to calculate the size of
6905          * the pages that match the user's query.  Then we allocate enough
6906          * memory to hold it, and actually copy the data into the buffer.
6907          */
6908         switch (page_code) {
6909         case SMS_ALL_PAGES_PAGE: {
6910                 int i;
6911
6912                 page_len = 0;
6913
6914                 /*
6915                  * At the moment, values other than 0 and 0xff here are
6916                  * reserved according to SPC-3.
6917                  */
6918                 if ((subpage != SMS_SUBPAGE_PAGE_0)
6919                  && (subpage != SMS_SUBPAGE_ALL)) {
6920                         ctl_set_invalid_field(ctsio,
6921                                               /*sks_valid*/ 1,
6922                                               /*command*/ 1,
6923                                               /*field*/ 3,
6924                                               /*bit_valid*/ 0,
6925                                               /*bit*/ 0);
6926                         ctl_done((union ctl_io *)ctsio);
6927                         return (CTL_RETVAL_COMPLETE);
6928                 }
6929
6930                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6931                         if ((control_dev != 0)
6932                          && (lun->mode_pages.index[i].page_flags &
6933                              CTL_PAGE_FLAG_DISK_ONLY))
6934                                 continue;
6935
6936                         /*
6937                          * We don't use this subpage if the user didn't
6938                          * request all subpages.
6939                          */
6940                         if ((lun->mode_pages.index[i].subpage != 0)
6941                          && (subpage == SMS_SUBPAGE_PAGE_0))
6942                                 continue;
6943
6944 #if 0
6945                         printf("found page %#x len %d\n",
6946                                lun->mode_pages.index[i].page_code &
6947                                SMPH_PC_MASK,
6948                                lun->mode_pages.index[i].page_len);
6949 #endif
6950                         page_len += lun->mode_pages.index[i].page_len;
6951                 }
6952                 break;
6953         }
6954         default: {
6955                 int i;
6956
6957                 page_len = 0;
6958
6959                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6960                         /* Look for the right page code */
6961                         if ((lun->mode_pages.index[i].page_code &
6962                              SMPH_PC_MASK) != page_code)
6963                                 continue;
6964
6965                         /* Look for the right subpage or the subpage wildcard*/
6966                         if ((lun->mode_pages.index[i].subpage != subpage)
6967                          && (subpage != SMS_SUBPAGE_ALL))
6968                                 continue;
6969
6970                         /* Make sure the page is supported for this dev type */
6971                         if ((control_dev != 0)
6972                          && (lun->mode_pages.index[i].page_flags &
6973                              CTL_PAGE_FLAG_DISK_ONLY))
6974                                 continue;
6975
6976 #if 0
6977                         printf("found page %#x len %d\n",
6978                                lun->mode_pages.index[i].page_code &
6979                                SMPH_PC_MASK,
6980                                lun->mode_pages.index[i].page_len);
6981 #endif
6982
6983                         page_len += lun->mode_pages.index[i].page_len;
6984                 }
6985
6986                 if (page_len == 0) {
6987                         ctl_set_invalid_field(ctsio,
6988                                               /*sks_valid*/ 1,
6989                                               /*command*/ 1,
6990                                               /*field*/ 2,
6991                                               /*bit_valid*/ 1,
6992                                               /*bit*/ 5);
6993                         ctl_done((union ctl_io *)ctsio);
6994                         return (CTL_RETVAL_COMPLETE);
6995                 }
6996                 break;
6997         }
6998         }
6999
7000         total_len = header_len + page_len;
7001 #if 0
7002         printf("header_len = %d, page_len = %d, total_len = %d\n",
7003                header_len, page_len, total_len);
7004 #endif
7005
7006         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7007         ctsio->kern_sg_entries = 0;
7008         ctsio->kern_data_resid = 0;
7009         ctsio->kern_rel_offset = 0;
7010         if (total_len < alloc_len) {
7011                 ctsio->residual = alloc_len - total_len;
7012                 ctsio->kern_data_len = total_len;
7013                 ctsio->kern_total_len = total_len;
7014         } else {
7015                 ctsio->residual = 0;
7016                 ctsio->kern_data_len = alloc_len;
7017                 ctsio->kern_total_len = alloc_len;
7018         }
7019
7020         switch (ctsio->cdb[0]) {
7021         case MODE_SENSE_6: {
7022                 struct scsi_mode_hdr_6 *header;
7023
7024                 header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
7025
7026                 header->datalen = ctl_min(total_len - 1, 254);
7027
7028                 if (dbd)
7029                         header->block_descr_len = 0;
7030                 else
7031                         header->block_descr_len =
7032                                 sizeof(struct scsi_mode_block_descr);
7033                 block_desc = (struct scsi_mode_block_descr *)&header[1];
7034                 break;
7035         }
7036         case MODE_SENSE_10: {
7037                 struct scsi_mode_hdr_10 *header;
7038                 int datalen;
7039
7040                 header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
7041
7042                 datalen = ctl_min(total_len - 2, 65533);
7043                 scsi_ulto2b(datalen, header->datalen);
7044                 if (dbd)
7045                         scsi_ulto2b(0, header->block_descr_len);
7046                 else
7047                         scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
7048                                     header->block_descr_len);
7049                 block_desc = (struct scsi_mode_block_descr *)&header[1];
7050                 break;
7051         }
7052         default:
7053                 panic("invalid CDB type %#x", ctsio->cdb[0]);
7054                 break; /* NOTREACHED */
7055         }
7056
7057         /*
7058          * If we've got a disk, use its blocksize in the block
7059          * descriptor.  Otherwise, just set it to 0.
7060          */
7061         if (dbd == 0) {
7062                 if (control_dev != 0)
7063                         scsi_ulto3b(lun->be_lun->blocksize,
7064                                     block_desc->block_len);
7065                 else
7066                         scsi_ulto3b(0, block_desc->block_len);
7067         }
7068
7069         switch (page_code) {
7070         case SMS_ALL_PAGES_PAGE: {
7071                 int i, data_used;
7072
7073                 data_used = header_len;
7074                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
7075                         struct ctl_page_index *page_index;
7076
7077                         page_index = &lun->mode_pages.index[i];
7078
7079                         if ((control_dev != 0)
7080                          && (page_index->page_flags &
7081                             CTL_PAGE_FLAG_DISK_ONLY))
7082                                 continue;
7083
7084                         /*
7085                          * We don't use this subpage if the user didn't
7086                          * request all subpages.  We already checked (above)
7087                          * to make sure the user only specified a subpage
7088                          * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
7089                          */
7090                         if ((page_index->subpage != 0)
7091                          && (subpage == SMS_SUBPAGE_PAGE_0))
7092                                 continue;
7093
7094                         /*
7095                          * Call the handler, if it exists, to update the
7096                          * page to the latest values.
7097                          */
7098                         if (page_index->sense_handler != NULL)
7099                                 page_index->sense_handler(ctsio, page_index,pc);
7100
7101                         memcpy(ctsio->kern_data_ptr + data_used,
7102                                page_index->page_data +
7103                                (page_index->page_len * pc),
7104                                page_index->page_len);
7105                         data_used += page_index->page_len;
7106                 }
7107                 break;
7108         }
7109         default: {
7110                 int i, data_used;
7111
7112                 data_used = header_len;
7113
7114                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
7115                         struct ctl_page_index *page_index;
7116
7117                         page_index = &lun->mode_pages.index[i];
7118
7119                         /* Look for the right page code */
7120                         if ((page_index->page_code & SMPH_PC_MASK) != page_code)
7121                                 continue;
7122
7123                         /* Look for the right subpage or the subpage wildcard*/
7124                         if ((page_index->subpage != subpage)
7125                          && (subpage != SMS_SUBPAGE_ALL))
7126                                 continue;
7127
7128                         /* Make sure the page is supported for this dev type */
7129                         if ((control_dev != 0)
7130                          && (page_index->page_flags &
7131                              CTL_PAGE_FLAG_DISK_ONLY))
7132                                 continue;
7133
7134                         /*
7135                          * Call the handler, if it exists, to update the
7136                          * page to the latest values.
7137                          */
7138                         if (page_index->sense_handler != NULL)
7139                                 page_index->sense_handler(ctsio, page_index,pc);
7140
7141                         memcpy(ctsio->kern_data_ptr + data_used,
7142                                page_index->page_data +
7143                                (page_index->page_len * pc),
7144                                page_index->page_len);
7145                         data_used += page_index->page_len;
7146                 }
7147                 break;
7148         }
7149         }
7150
7151         ctsio->scsi_status = SCSI_STATUS_OK;
7152
7153         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7154         ctsio->be_move_done = ctl_config_move_done;
7155         ctl_datamove((union ctl_io *)ctsio);
7156
7157         return (CTL_RETVAL_COMPLETE);
7158 }
7159
7160 int
7161 ctl_read_capacity(struct ctl_scsiio *ctsio)
7162 {
7163         struct scsi_read_capacity *cdb;
7164         struct scsi_read_capacity_data *data;
7165         struct ctl_lun *lun;
7166         uint32_t lba;
7167
7168         CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
7169
7170         cdb = (struct scsi_read_capacity *)ctsio->cdb;
7171
7172         lba = scsi_4btoul(cdb->addr);
7173         if (((cdb->pmi & SRC_PMI) == 0)
7174          && (lba != 0)) {
7175                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7176                                       /*sks_valid*/ 1,
7177                                       /*command*/ 1,
7178                                       /*field*/ 2,
7179                                       /*bit_valid*/ 0,
7180                                       /*bit*/ 0);
7181                 ctl_done((union ctl_io *)ctsio);
7182                 return (CTL_RETVAL_COMPLETE);
7183         }
7184
7185         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7186
7187         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7188         data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
7189         ctsio->residual = 0;
7190         ctsio->kern_data_len = sizeof(*data);
7191         ctsio->kern_total_len = sizeof(*data);
7192         ctsio->kern_data_resid = 0;
7193         ctsio->kern_rel_offset = 0;
7194         ctsio->kern_sg_entries = 0;
7195
7196         /*
7197          * If the maximum LBA is greater than 0xfffffffe, the user must
7198          * issue a SERVICE ACTION IN (16) command, with the read capacity
7199          * serivce action set.
7200          */
7201         if (lun->be_lun->maxlba > 0xfffffffe)
7202                 scsi_ulto4b(0xffffffff, data->addr);
7203         else
7204                 scsi_ulto4b(lun->be_lun->maxlba, data->addr);
7205
7206         /*
7207          * XXX KDM this may not be 512 bytes...
7208          */
7209         scsi_ulto4b(lun->be_lun->blocksize, data->length);
7210
7211         ctsio->scsi_status = SCSI_STATUS_OK;
7212
7213         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7214         ctsio->be_move_done = ctl_config_move_done;
7215         ctl_datamove((union ctl_io *)ctsio);
7216
7217         return (CTL_RETVAL_COMPLETE);
7218 }
7219
7220 int
7221 ctl_read_capacity_16(struct ctl_scsiio *ctsio)
7222 {
7223         struct scsi_read_capacity_16 *cdb;
7224         struct scsi_read_capacity_data_long *data;
7225         struct ctl_lun *lun;
7226         uint64_t lba;
7227         uint32_t alloc_len;
7228
7229         CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
7230
7231         cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
7232
7233         alloc_len = scsi_4btoul(cdb->alloc_len);
7234         lba = scsi_8btou64(cdb->addr);
7235
7236         if ((cdb->reladr & SRC16_PMI)
7237          && (lba != 0)) {
7238                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7239                                       /*sks_valid*/ 1,
7240                                       /*command*/ 1,
7241                                       /*field*/ 2,
7242                                       /*bit_valid*/ 0,
7243                                       /*bit*/ 0);
7244                 ctl_done((union ctl_io *)ctsio);
7245                 return (CTL_RETVAL_COMPLETE);
7246         }
7247
7248         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7249
7250         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7251         data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
7252
7253         if (sizeof(*data) < alloc_len) {
7254                 ctsio->residual = alloc_len - sizeof(*data);
7255                 ctsio->kern_data_len = sizeof(*data);
7256                 ctsio->kern_total_len = sizeof(*data);
7257         } else {
7258                 ctsio->residual = 0;
7259                 ctsio->kern_data_len = alloc_len;
7260                 ctsio->kern_total_len = alloc_len;
7261         }
7262         ctsio->kern_data_resid = 0;
7263         ctsio->kern_rel_offset = 0;
7264         ctsio->kern_sg_entries = 0;
7265
7266         scsi_u64to8b(lun->be_lun->maxlba, data->addr);
7267         /* XXX KDM this may not be 512 bytes... */
7268         scsi_ulto4b(lun->be_lun->blocksize, data->length);
7269         data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
7270         scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
7271         if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
7272                 data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ;
7273
7274         ctsio->scsi_status = SCSI_STATUS_OK;
7275
7276         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7277         ctsio->be_move_done = ctl_config_move_done;
7278         ctl_datamove((union ctl_io *)ctsio);
7279
7280         return (CTL_RETVAL_COMPLETE);
7281 }
7282
7283 int
7284 ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7285 {
7286         struct scsi_maintenance_in *cdb;
7287         int retval;
7288         int alloc_len, ext, total_len = 0, g, p, pc, pg;
7289         int num_target_port_groups, num_target_ports, single;
7290         struct ctl_lun *lun;
7291         struct ctl_softc *softc;
7292         struct ctl_port *port;
7293         struct scsi_target_group_data *rtg_ptr;
7294         struct scsi_target_group_data_extended *rtg_ext_ptr;
7295         struct scsi_target_port_group_descriptor *tpg_desc;
7296
7297         CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7298
7299         cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7300         softc = control_softc;
7301         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7302
7303         retval = CTL_RETVAL_COMPLETE;
7304
7305         switch (cdb->byte2 & STG_PDF_MASK) {
7306         case STG_PDF_LENGTH:
7307                 ext = 0;
7308                 break;
7309         case STG_PDF_EXTENDED:
7310                 ext = 1;
7311                 break;
7312         default:
7313                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7314                                       /*sks_valid*/ 1,
7315                                       /*command*/ 1,
7316                                       /*field*/ 2,
7317                                       /*bit_valid*/ 1,
7318                                       /*bit*/ 5);
7319                 ctl_done((union ctl_io *)ctsio);
7320                 return(retval);
7321         }
7322
7323         single = ctl_is_single;
7324         if (single)
7325                 num_target_port_groups = 1;
7326         else
7327                 num_target_port_groups = NUM_TARGET_PORT_GROUPS;
7328         num_target_ports = 0;
7329         mtx_lock(&softc->ctl_lock);
7330         STAILQ_FOREACH(port, &softc->port_list, links) {
7331                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7332                         continue;
7333                 if (ctl_map_lun_back(port->targ_port, lun->lun) >= CTL_MAX_LUNS)
7334                         continue;
7335                 num_target_ports++;
7336         }
7337         mtx_unlock(&softc->ctl_lock);
7338
7339         if (ext)
7340                 total_len = sizeof(struct scsi_target_group_data_extended);
7341         else
7342                 total_len = sizeof(struct scsi_target_group_data);
7343         total_len += sizeof(struct scsi_target_port_group_descriptor) *
7344                 num_target_port_groups +
7345             sizeof(struct scsi_target_port_descriptor) *
7346                 num_target_ports * num_target_port_groups;
7347
7348         alloc_len = scsi_4btoul(cdb->length);
7349
7350         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7351
7352         ctsio->kern_sg_entries = 0;
7353
7354         if (total_len < alloc_len) {
7355                 ctsio->residual = alloc_len - total_len;
7356                 ctsio->kern_data_len = total_len;
7357                 ctsio->kern_total_len = total_len;
7358         } else {
7359                 ctsio->residual = 0;
7360                 ctsio->kern_data_len = alloc_len;
7361                 ctsio->kern_total_len = alloc_len;
7362         }
7363         ctsio->kern_data_resid = 0;
7364         ctsio->kern_rel_offset = 0;
7365
7366         if (ext) {
7367                 rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7368                     ctsio->kern_data_ptr;
7369                 scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7370                 rtg_ext_ptr->format_type = 0x10;
7371                 rtg_ext_ptr->implicit_transition_time = 0;
7372                 tpg_desc = &rtg_ext_ptr->groups[0];
7373         } else {
7374                 rtg_ptr = (struct scsi_target_group_data *)
7375                     ctsio->kern_data_ptr;
7376                 scsi_ulto4b(total_len - 4, rtg_ptr->length);
7377                 tpg_desc = &rtg_ptr->groups[0];
7378         }
7379
7380         pg = ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS;
7381         mtx_lock(&softc->ctl_lock);
7382         for (g = 0; g < num_target_port_groups; g++) {
7383                 if (g == pg)
7384                         tpg_desc->pref_state = TPG_PRIMARY |
7385                             TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7386                 else
7387                         tpg_desc->pref_state =
7388                             TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7389                 tpg_desc->support = TPG_AO_SUP;
7390                 if (!single)
7391                         tpg_desc->support |= TPG_AN_SUP;
7392                 scsi_ulto2b(g + 1, tpg_desc->target_port_group);
7393                 tpg_desc->status = TPG_IMPLICIT;
7394                 pc = 0;
7395                 STAILQ_FOREACH(port, &softc->port_list, links) {
7396                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7397                                 continue;
7398                         if (ctl_map_lun_back(port->targ_port, lun->lun) >=
7399                             CTL_MAX_LUNS)
7400                                 continue;
7401                         p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
7402                         scsi_ulto2b(p, tpg_desc->descriptors[pc].
7403                             relative_target_port_identifier);
7404                         pc++;
7405                 }
7406                 tpg_desc->target_port_count = pc;
7407                 tpg_desc = (struct scsi_target_port_group_descriptor *)
7408                     &tpg_desc->descriptors[pc];
7409         }
7410         mtx_unlock(&softc->ctl_lock);
7411
7412         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7413         ctsio->be_move_done = ctl_config_move_done;
7414
7415         CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7416                          ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7417                          ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7418                          ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7419                          ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7420
7421         ctl_datamove((union ctl_io *)ctsio);
7422         return(retval);
7423 }
7424
7425 int
7426 ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7427 {
7428         struct ctl_lun *lun;
7429         struct scsi_report_supported_opcodes *cdb;
7430         const struct ctl_cmd_entry *entry, *sentry;
7431         struct scsi_report_supported_opcodes_all *all;
7432         struct scsi_report_supported_opcodes_descr *descr;
7433         struct scsi_report_supported_opcodes_one *one;
7434         int retval;
7435         int alloc_len, total_len;
7436         int opcode, service_action, i, j, num;
7437
7438         CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7439
7440         cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7441         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7442
7443         retval = CTL_RETVAL_COMPLETE;
7444
7445         opcode = cdb->requested_opcode;
7446         service_action = scsi_2btoul(cdb->requested_service_action);
7447         switch (cdb->options & RSO_OPTIONS_MASK) {
7448         case RSO_OPTIONS_ALL:
7449                 num = 0;
7450                 for (i = 0; i < 256; i++) {
7451                         entry = &ctl_cmd_table[i];
7452                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7453                                 for (j = 0; j < 32; j++) {
7454                                         sentry = &((const struct ctl_cmd_entry *)
7455                                             entry->execute)[j];
7456                                         if (ctl_cmd_applicable(
7457                                             lun->be_lun->lun_type, sentry))
7458                                                 num++;
7459                                 }
7460                         } else {
7461                                 if (ctl_cmd_applicable(lun->be_lun->lun_type,
7462                                     entry))
7463                                         num++;
7464                         }
7465                 }
7466                 total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7467                     num * sizeof(struct scsi_report_supported_opcodes_descr);
7468                 break;
7469         case RSO_OPTIONS_OC:
7470                 if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7471                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7472                                               /*sks_valid*/ 1,
7473                                               /*command*/ 1,
7474                                               /*field*/ 2,
7475                                               /*bit_valid*/ 1,
7476                                               /*bit*/ 2);
7477                         ctl_done((union ctl_io *)ctsio);
7478                         return (CTL_RETVAL_COMPLETE);
7479                 }
7480                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7481                 break;
7482         case RSO_OPTIONS_OC_SA:
7483                 if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7484                     service_action >= 32) {
7485                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7486                                               /*sks_valid*/ 1,
7487                                               /*command*/ 1,
7488                                               /*field*/ 2,
7489                                               /*bit_valid*/ 1,
7490                                               /*bit*/ 2);
7491                         ctl_done((union ctl_io *)ctsio);
7492                         return (CTL_RETVAL_COMPLETE);
7493                 }
7494                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7495                 break;
7496         default:
7497                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7498                                       /*sks_valid*/ 1,
7499                                       /*command*/ 1,
7500                                       /*field*/ 2,
7501                                       /*bit_valid*/ 1,
7502                                       /*bit*/ 2);
7503                 ctl_done((union ctl_io *)ctsio);
7504                 return (CTL_RETVAL_COMPLETE);
7505         }
7506
7507         alloc_len = scsi_4btoul(cdb->length);
7508
7509         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7510
7511         ctsio->kern_sg_entries = 0;
7512
7513         if (total_len < alloc_len) {
7514                 ctsio->residual = alloc_len - total_len;
7515                 ctsio->kern_data_len = total_len;
7516                 ctsio->kern_total_len = total_len;
7517         } else {
7518                 ctsio->residual = 0;
7519                 ctsio->kern_data_len = alloc_len;
7520                 ctsio->kern_total_len = alloc_len;
7521         }
7522         ctsio->kern_data_resid = 0;
7523         ctsio->kern_rel_offset = 0;
7524
7525         switch (cdb->options & RSO_OPTIONS_MASK) {
7526         case RSO_OPTIONS_ALL:
7527                 all = (struct scsi_report_supported_opcodes_all *)
7528                     ctsio->kern_data_ptr;
7529                 num = 0;
7530                 for (i = 0; i < 256; i++) {
7531                         entry = &ctl_cmd_table[i];
7532                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7533                                 for (j = 0; j < 32; j++) {
7534                                         sentry = &((const struct ctl_cmd_entry *)
7535                                             entry->execute)[j];
7536                                         if (!ctl_cmd_applicable(
7537                                             lun->be_lun->lun_type, sentry))
7538                                                 continue;
7539                                         descr = &all->descr[num++];
7540                                         descr->opcode = i;
7541                                         scsi_ulto2b(j, descr->service_action);
7542                                         descr->flags = RSO_SERVACTV;
7543                                         scsi_ulto2b(sentry->length,
7544                                             descr->cdb_length);
7545                                 }
7546                         } else {
7547                                 if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7548                                     entry))
7549                                         continue;
7550                                 descr = &all->descr[num++];
7551                                 descr->opcode = i;
7552                                 scsi_ulto2b(0, descr->service_action);
7553                                 descr->flags = 0;
7554                                 scsi_ulto2b(entry->length, descr->cdb_length);
7555                         }
7556                 }
7557                 scsi_ulto4b(
7558                     num * sizeof(struct scsi_report_supported_opcodes_descr),
7559                     all->length);
7560                 break;
7561         case RSO_OPTIONS_OC:
7562                 one = (struct scsi_report_supported_opcodes_one *)
7563                     ctsio->kern_data_ptr;
7564                 entry = &ctl_cmd_table[opcode];
7565                 goto fill_one;
7566         case RSO_OPTIONS_OC_SA:
7567                 one = (struct scsi_report_supported_opcodes_one *)
7568                     ctsio->kern_data_ptr;
7569                 entry = &ctl_cmd_table[opcode];
7570                 entry = &((const struct ctl_cmd_entry *)
7571                     entry->execute)[service_action];
7572 fill_one:
7573                 if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7574                         one->support = 3;
7575                         scsi_ulto2b(entry->length, one->cdb_length);
7576                         one->cdb_usage[0] = opcode;
7577                         memcpy(&one->cdb_usage[1], entry->usage,
7578                             entry->length - 1);
7579                 } else
7580                         one->support = 1;
7581                 break;
7582         }
7583
7584         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7585         ctsio->be_move_done = ctl_config_move_done;
7586
7587         ctl_datamove((union ctl_io *)ctsio);
7588         return(retval);
7589 }
7590
7591 int
7592 ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7593 {
7594         struct ctl_lun *lun;
7595         struct scsi_report_supported_tmf *cdb;
7596         struct scsi_report_supported_tmf_data *data;
7597         int retval;
7598         int alloc_len, total_len;
7599
7600         CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7601
7602         cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7603         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7604
7605         retval = CTL_RETVAL_COMPLETE;
7606
7607         total_len = sizeof(struct scsi_report_supported_tmf_data);
7608         alloc_len = scsi_4btoul(cdb->length);
7609
7610         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7611
7612         ctsio->kern_sg_entries = 0;
7613
7614         if (total_len < alloc_len) {
7615                 ctsio->residual = alloc_len - total_len;
7616                 ctsio->kern_data_len = total_len;
7617                 ctsio->kern_total_len = total_len;
7618         } else {
7619                 ctsio->residual = 0;
7620                 ctsio->kern_data_len = alloc_len;
7621                 ctsio->kern_total_len = alloc_len;
7622         }
7623         ctsio->kern_data_resid = 0;
7624         ctsio->kern_rel_offset = 0;
7625
7626         data = (struct scsi_report_supported_tmf_data *)ctsio->kern_data_ptr;
7627         data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_TRS;
7628         data->byte2 |= RST_ITNRS;
7629
7630         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7631         ctsio->be_move_done = ctl_config_move_done;
7632
7633         ctl_datamove((union ctl_io *)ctsio);
7634         return (retval);
7635 }
7636
7637 int
7638 ctl_report_timestamp(struct ctl_scsiio *ctsio)
7639 {
7640         struct ctl_lun *lun;
7641         struct scsi_report_timestamp *cdb;
7642         struct scsi_report_timestamp_data *data;
7643         struct timeval tv;
7644         int64_t timestamp;
7645         int retval;
7646         int alloc_len, total_len;
7647
7648         CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7649
7650         cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7651         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7652
7653         retval = CTL_RETVAL_COMPLETE;
7654
7655         total_len = sizeof(struct scsi_report_timestamp_data);
7656         alloc_len = scsi_4btoul(cdb->length);
7657
7658         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7659
7660         ctsio->kern_sg_entries = 0;
7661
7662         if (total_len < alloc_len) {
7663                 ctsio->residual = alloc_len - total_len;
7664                 ctsio->kern_data_len = total_len;
7665                 ctsio->kern_total_len = total_len;
7666         } else {
7667                 ctsio->residual = 0;
7668                 ctsio->kern_data_len = alloc_len;
7669                 ctsio->kern_total_len = alloc_len;
7670         }
7671         ctsio->kern_data_resid = 0;
7672         ctsio->kern_rel_offset = 0;
7673
7674         data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7675         scsi_ulto2b(sizeof(*data) - 2, data->length);
7676         data->origin = RTS_ORIG_OUTSIDE;
7677         getmicrotime(&tv);
7678         timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7679         scsi_ulto4b(timestamp >> 16, data->timestamp);
7680         scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7681
7682         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7683         ctsio->be_move_done = ctl_config_move_done;
7684
7685         ctl_datamove((union ctl_io *)ctsio);
7686         return (retval);
7687 }
7688
7689 int
7690 ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7691 {
7692         struct scsi_per_res_in *cdb;
7693         int alloc_len, total_len = 0;
7694         /* struct scsi_per_res_in_rsrv in_data; */
7695         struct ctl_lun *lun;
7696         struct ctl_softc *softc;
7697
7698         CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7699
7700         softc = control_softc;
7701
7702         cdb = (struct scsi_per_res_in *)ctsio->cdb;
7703
7704         alloc_len = scsi_2btoul(cdb->length);
7705
7706         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7707
7708 retry:
7709         mtx_lock(&lun->lun_lock);
7710         switch (cdb->action) {
7711         case SPRI_RK: /* read keys */
7712                 total_len = sizeof(struct scsi_per_res_in_keys) +
7713                         lun->pr_key_count *
7714                         sizeof(struct scsi_per_res_key);
7715                 break;
7716         case SPRI_RR: /* read reservation */
7717                 if (lun->flags & CTL_LUN_PR_RESERVED)
7718                         total_len = sizeof(struct scsi_per_res_in_rsrv);
7719                 else
7720                         total_len = sizeof(struct scsi_per_res_in_header);
7721                 break;
7722         case SPRI_RC: /* report capabilities */
7723                 total_len = sizeof(struct scsi_per_res_cap);
7724                 break;
7725         case SPRI_RS: /* read full status */
7726                 total_len = sizeof(struct scsi_per_res_in_header) +
7727                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7728                     lun->pr_key_count;
7729                 break;
7730         default:
7731                 panic("Invalid PR type %x", cdb->action);
7732         }
7733         mtx_unlock(&lun->lun_lock);
7734
7735         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7736
7737         if (total_len < alloc_len) {
7738                 ctsio->residual = alloc_len - total_len;
7739                 ctsio->kern_data_len = total_len;
7740                 ctsio->kern_total_len = total_len;
7741         } else {
7742                 ctsio->residual = 0;
7743                 ctsio->kern_data_len = alloc_len;
7744                 ctsio->kern_total_len = alloc_len;
7745         }
7746
7747         ctsio->kern_data_resid = 0;
7748         ctsio->kern_rel_offset = 0;
7749         ctsio->kern_sg_entries = 0;
7750
7751         mtx_lock(&lun->lun_lock);
7752         switch (cdb->action) {
7753         case SPRI_RK: { // read keys
7754         struct scsi_per_res_in_keys *res_keys;
7755                 int i, key_count;
7756
7757                 res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7758
7759                 /*
7760                  * We had to drop the lock to allocate our buffer, which
7761                  * leaves time for someone to come in with another
7762                  * persistent reservation.  (That is unlikely, though,
7763                  * since this should be the only persistent reservation
7764                  * command active right now.)
7765                  */
7766                 if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7767                     (lun->pr_key_count *
7768                      sizeof(struct scsi_per_res_key)))){
7769                         mtx_unlock(&lun->lun_lock);
7770                         free(ctsio->kern_data_ptr, M_CTL);
7771                         printf("%s: reservation length changed, retrying\n",
7772                                __func__);
7773                         goto retry;
7774                 }
7775
7776                 scsi_ulto4b(lun->PRGeneration, res_keys->header.generation);
7777
7778                 scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7779                              lun->pr_key_count, res_keys->header.length);
7780
7781                 for (i = 0, key_count = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7782                         if (!lun->per_res[i].registered)
7783                                 continue;
7784
7785                         /*
7786                          * We used lun->pr_key_count to calculate the
7787                          * size to allocate.  If it turns out the number of
7788                          * initiators with the registered flag set is
7789                          * larger than that (i.e. they haven't been kept in
7790                          * sync), we've got a problem.
7791                          */
7792                         if (key_count >= lun->pr_key_count) {
7793 #ifdef NEEDTOPORT
7794                                 csevent_log(CSC_CTL | CSC_SHELF_SW |
7795                                             CTL_PR_ERROR,
7796                                             csevent_LogType_Fault,
7797                                             csevent_AlertLevel_Yellow,
7798                                             csevent_FRU_ShelfController,
7799                                             csevent_FRU_Firmware,
7800                                         csevent_FRU_Unknown,
7801                                             "registered keys %d >= key "
7802                                             "count %d", key_count,
7803                                             lun->pr_key_count);
7804 #endif
7805                                 key_count++;
7806                                 continue;
7807                         }
7808                         memcpy(res_keys->keys[key_count].key,
7809                                lun->per_res[i].res_key.key,
7810                                ctl_min(sizeof(res_keys->keys[key_count].key),
7811                                sizeof(lun->per_res[i].res_key)));
7812                         key_count++;
7813                 }
7814                 break;
7815         }
7816         case SPRI_RR: { // read reservation
7817                 struct scsi_per_res_in_rsrv *res;
7818                 int tmp_len, header_only;
7819
7820                 res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7821
7822                 scsi_ulto4b(lun->PRGeneration, res->header.generation);
7823
7824                 if (lun->flags & CTL_LUN_PR_RESERVED)
7825                 {
7826                         tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7827                         scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7828                                     res->header.length);
7829                         header_only = 0;
7830                 } else {
7831                         tmp_len = sizeof(struct scsi_per_res_in_header);
7832                         scsi_ulto4b(0, res->header.length);
7833                         header_only = 1;
7834                 }
7835
7836                 /*
7837                  * We had to drop the lock to allocate our buffer, which
7838                  * leaves time for someone to come in with another
7839                  * persistent reservation.  (That is unlikely, though,
7840                  * since this should be the only persistent reservation
7841                  * command active right now.)
7842                  */
7843                 if (tmp_len != total_len) {
7844                         mtx_unlock(&lun->lun_lock);
7845                         free(ctsio->kern_data_ptr, M_CTL);
7846                         printf("%s: reservation status changed, retrying\n",
7847                                __func__);
7848                         goto retry;
7849                 }
7850
7851                 /*
7852                  * No reservation held, so we're done.
7853                  */
7854                 if (header_only != 0)
7855                         break;
7856
7857                 /*
7858                  * If the registration is an All Registrants type, the key
7859                  * is 0, since it doesn't really matter.
7860                  */
7861                 if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7862                         memcpy(res->data.reservation,
7863                                &lun->per_res[lun->pr_res_idx].res_key,
7864                                sizeof(struct scsi_per_res_key));
7865                 }
7866                 res->data.scopetype = lun->res_type;
7867                 break;
7868         }
7869         case SPRI_RC:     //report capabilities
7870         {
7871                 struct scsi_per_res_cap *res_cap;
7872                 uint16_t type_mask;
7873
7874                 res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7875                 scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7876                 res_cap->flags2 |= SPRI_TMV | SPRI_ALLOW_3;
7877                 type_mask = SPRI_TM_WR_EX_AR |
7878                             SPRI_TM_EX_AC_RO |
7879                             SPRI_TM_WR_EX_RO |
7880                             SPRI_TM_EX_AC |
7881                             SPRI_TM_WR_EX |
7882                             SPRI_TM_EX_AC_AR;
7883                 scsi_ulto2b(type_mask, res_cap->type_mask);
7884                 break;
7885         }
7886         case SPRI_RS: { // read full status
7887                 struct scsi_per_res_in_full *res_status;
7888                 struct scsi_per_res_in_full_desc *res_desc;
7889                 struct ctl_port *port;
7890                 int i, len;
7891
7892                 res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7893
7894                 /*
7895                  * We had to drop the lock to allocate our buffer, which
7896                  * leaves time for someone to come in with another
7897                  * persistent reservation.  (That is unlikely, though,
7898                  * since this should be the only persistent reservation
7899                  * command active right now.)
7900                  */
7901                 if (total_len < (sizeof(struct scsi_per_res_in_header) +
7902                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7903                      lun->pr_key_count)){
7904                         mtx_unlock(&lun->lun_lock);
7905                         free(ctsio->kern_data_ptr, M_CTL);
7906                         printf("%s: reservation length changed, retrying\n",
7907                                __func__);
7908                         goto retry;
7909                 }
7910
7911                 scsi_ulto4b(lun->PRGeneration, res_status->header.generation);
7912
7913                 res_desc = &res_status->desc[0];
7914                 for (i = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7915                         if (!lun->per_res[i].registered)
7916                                 continue;
7917
7918                         memcpy(&res_desc->res_key, &lun->per_res[i].res_key.key,
7919                             sizeof(res_desc->res_key));
7920                         if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7921                             (lun->pr_res_idx == i ||
7922                              lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7923                                 res_desc->flags = SPRI_FULL_R_HOLDER;
7924                                 res_desc->scopetype = lun->res_type;
7925                         }
7926                         scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7927                             res_desc->rel_trgt_port_id);
7928                         len = 0;
7929                         port = softc->ctl_ports[i / CTL_MAX_INIT_PER_PORT];
7930                         if (port != NULL)
7931                                 len = ctl_create_iid(port,
7932                                     i % CTL_MAX_INIT_PER_PORT,
7933                                     res_desc->transport_id);
7934                         scsi_ulto4b(len, res_desc->additional_length);
7935                         res_desc = (struct scsi_per_res_in_full_desc *)
7936                             &res_desc->transport_id[len];
7937                 }
7938                 scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7939                     res_status->header.length);
7940                 break;
7941         }
7942         default:
7943                 /*
7944                  * This is a bug, because we just checked for this above,
7945                  * and should have returned an error.
7946                  */
7947                 panic("Invalid PR type %x", cdb->action);
7948                 break; /* NOTREACHED */
7949         }
7950         mtx_unlock(&lun->lun_lock);
7951
7952         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7953         ctsio->be_move_done = ctl_config_move_done;
7954
7955         CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7956                          ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7957                          ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7958                          ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7959                          ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7960
7961         ctl_datamove((union ctl_io *)ctsio);
7962
7963         return (CTL_RETVAL_COMPLETE);
7964 }
7965
7966 /*
7967  * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7968  * it should return.
7969  */
7970 static int
7971 ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7972                 uint64_t sa_res_key, uint8_t type, uint32_t residx,
7973                 struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7974                 struct scsi_per_res_out_parms* param)
7975 {
7976         union ctl_ha_msg persis_io;
7977         int retval, i;
7978         int isc_retval;
7979
7980         retval = 0;
7981
7982         mtx_lock(&lun->lun_lock);
7983         if (sa_res_key == 0) {
7984                 if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7985                         /* validate scope and type */
7986                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7987                              SPR_LU_SCOPE) {
7988                                 mtx_unlock(&lun->lun_lock);
7989                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7990                                                       /*sks_valid*/ 1,
7991                                                       /*command*/ 1,
7992                                                       /*field*/ 2,
7993                                                       /*bit_valid*/ 1,
7994                                                       /*bit*/ 4);
7995                                 ctl_done((union ctl_io *)ctsio);
7996                                 return (1);
7997                         }
7998
7999                         if (type>8 || type==2 || type==4 || type==0) {
8000                                 mtx_unlock(&lun->lun_lock);
8001                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
8002                                                       /*sks_valid*/ 1,
8003                                                       /*command*/ 1,
8004                                                       /*field*/ 2,
8005                                                       /*bit_valid*/ 1,
8006                                                       /*bit*/ 0);
8007                                 ctl_done((union ctl_io *)ctsio);
8008                                 return (1);
8009                         }
8010
8011                         /* temporarily unregister this nexus */
8012                         lun->per_res[residx].registered = 0;
8013
8014                         /*
8015                          * Unregister everybody else and build UA for
8016                          * them
8017                          */
8018                         for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8019                                 if (lun->per_res[i].registered == 0)
8020                                         continue;
8021
8022                                 if (!persis_offset
8023                                  && i <CTL_MAX_INITIATORS)
8024                                         lun->pending_ua[i] |=
8025                                                 CTL_UA_REG_PREEMPT;
8026                                 else if (persis_offset
8027                                       && i >= persis_offset)
8028                                         lun->pending_ua[i-persis_offset] |=
8029                                                 CTL_UA_REG_PREEMPT;
8030                                 lun->per_res[i].registered = 0;
8031                                 memset(&lun->per_res[i].res_key, 0,
8032                                        sizeof(struct scsi_per_res_key));
8033                         }
8034                         lun->per_res[residx].registered = 1;
8035                         lun->pr_key_count = 1;
8036                         lun->res_type = type;
8037                         if (lun->res_type != SPR_TYPE_WR_EX_AR
8038                          && lun->res_type != SPR_TYPE_EX_AC_AR)
8039                                 lun->pr_res_idx = residx;
8040
8041                         /* send msg to other side */
8042                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8043                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8044                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8045                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8046                         persis_io.pr.pr_info.res_type = type;
8047                         memcpy(persis_io.pr.pr_info.sa_res_key,
8048                                param->serv_act_res_key,
8049                                sizeof(param->serv_act_res_key));
8050                         if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8051                              &persis_io, sizeof(persis_io), 0)) >
8052                              CTL_HA_STATUS_SUCCESS) {
8053                                 printf("CTL:Persis Out error returned "
8054                                        "from ctl_ha_msg_send %d\n",
8055                                        isc_retval);
8056                         }
8057                 } else {
8058                         /* not all registrants */
8059                         mtx_unlock(&lun->lun_lock);
8060                         free(ctsio->kern_data_ptr, M_CTL);
8061                         ctl_set_invalid_field(ctsio,
8062                                               /*sks_valid*/ 1,
8063                                               /*command*/ 0,
8064                                               /*field*/ 8,
8065                                               /*bit_valid*/ 0,
8066                                               /*bit*/ 0);
8067                         ctl_done((union ctl_io *)ctsio);
8068                         return (1);
8069                 }
8070         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8071                 || !(lun->flags & CTL_LUN_PR_RESERVED)) {
8072                 int found = 0;
8073
8074                 if (res_key == sa_res_key) {
8075                         /* special case */
8076                         /*
8077                          * The spec implies this is not good but doesn't
8078                          * say what to do. There are two choices either
8079                          * generate a res conflict or check condition
8080                          * with illegal field in parameter data. Since
8081                          * that is what is done when the sa_res_key is
8082                          * zero I'll take that approach since this has
8083                          * to do with the sa_res_key.
8084                          */
8085                         mtx_unlock(&lun->lun_lock);
8086                         free(ctsio->kern_data_ptr, M_CTL);
8087                         ctl_set_invalid_field(ctsio,
8088                                               /*sks_valid*/ 1,
8089                                               /*command*/ 0,
8090                                               /*field*/ 8,
8091                                               /*bit_valid*/ 0,
8092                                               /*bit*/ 0);
8093                         ctl_done((union ctl_io *)ctsio);
8094                         return (1);
8095                 }
8096
8097                 for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8098                         if (lun->per_res[i].registered
8099                          && memcmp(param->serv_act_res_key,
8100                             lun->per_res[i].res_key.key,
8101                             sizeof(struct scsi_per_res_key)) != 0)
8102                                 continue;
8103
8104                         found = 1;
8105                         lun->per_res[i].registered = 0;
8106                         memset(&lun->per_res[i].res_key, 0,
8107                                sizeof(struct scsi_per_res_key));
8108                         lun->pr_key_count--;
8109
8110                         if (!persis_offset && i < CTL_MAX_INITIATORS)
8111                                 lun->pending_ua[i] |= CTL_UA_REG_PREEMPT;
8112                         else if (persis_offset && i >= persis_offset)
8113                                 lun->pending_ua[i-persis_offset] |=
8114                                         CTL_UA_REG_PREEMPT;
8115                 }
8116                 if (!found) {
8117                         mtx_unlock(&lun->lun_lock);
8118                         free(ctsio->kern_data_ptr, M_CTL);
8119                         ctl_set_reservation_conflict(ctsio);
8120                         ctl_done((union ctl_io *)ctsio);
8121                         return (CTL_RETVAL_COMPLETE);
8122                 }
8123                 /* send msg to other side */
8124                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8125                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8126                 persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8127                 persis_io.pr.pr_info.residx = lun->pr_res_idx;
8128                 persis_io.pr.pr_info.res_type = type;
8129                 memcpy(persis_io.pr.pr_info.sa_res_key,
8130                        param->serv_act_res_key,
8131                        sizeof(param->serv_act_res_key));
8132                 if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8133                      &persis_io, sizeof(persis_io), 0)) >
8134                      CTL_HA_STATUS_SUCCESS) {
8135                         printf("CTL:Persis Out error returned from "
8136                                "ctl_ha_msg_send %d\n", isc_retval);
8137                 }
8138         } else {
8139                 /* Reserved but not all registrants */
8140                 /* sa_res_key is res holder */
8141                 if (memcmp(param->serv_act_res_key,
8142                    lun->per_res[lun->pr_res_idx].res_key.key,
8143                    sizeof(struct scsi_per_res_key)) == 0) {
8144                         /* validate scope and type */
8145                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
8146                              SPR_LU_SCOPE) {
8147                                 mtx_unlock(&lun->lun_lock);
8148                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
8149                                                       /*sks_valid*/ 1,
8150                                                       /*command*/ 1,
8151                                                       /*field*/ 2,
8152                                                       /*bit_valid*/ 1,
8153                                                       /*bit*/ 4);
8154                                 ctl_done((union ctl_io *)ctsio);
8155                                 return (1);
8156                         }
8157
8158                         if (type>8 || type==2 || type==4 || type==0) {
8159                                 mtx_unlock(&lun->lun_lock);
8160                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
8161                                                       /*sks_valid*/ 1,
8162                                                       /*command*/ 1,
8163                                                       /*field*/ 2,
8164                                                       /*bit_valid*/ 1,
8165                                                       /*bit*/ 0);
8166                                 ctl_done((union ctl_io *)ctsio);
8167                                 return (1);
8168                         }
8169
8170                         /*
8171                          * Do the following:
8172                          * if sa_res_key != res_key remove all
8173                          * registrants w/sa_res_key and generate UA
8174                          * for these registrants(Registrations
8175                          * Preempted) if it wasn't an exclusive
8176                          * reservation generate UA(Reservations
8177                          * Preempted) for all other registered nexuses
8178                          * if the type has changed. Establish the new
8179                          * reservation and holder. If res_key and
8180                          * sa_res_key are the same do the above
8181                          * except don't unregister the res holder.
8182                          */
8183
8184                         /*
8185                          * Temporarily unregister so it won't get
8186                          * removed or UA generated
8187                          */
8188                         lun->per_res[residx].registered = 0;
8189                         for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8190                                 if (lun->per_res[i].registered == 0)
8191                                         continue;
8192
8193                                 if (memcmp(param->serv_act_res_key,
8194                                     lun->per_res[i].res_key.key,
8195                                     sizeof(struct scsi_per_res_key)) == 0) {
8196                                         lun->per_res[i].registered = 0;
8197                                         memset(&lun->per_res[i].res_key,
8198                                                0,
8199                                                sizeof(struct scsi_per_res_key));
8200                                         lun->pr_key_count--;
8201
8202                                         if (!persis_offset
8203                                          && i < CTL_MAX_INITIATORS)
8204                                                 lun->pending_ua[i] |=
8205                                                         CTL_UA_REG_PREEMPT;
8206                                         else if (persis_offset
8207                                               && i >= persis_offset)
8208                                                 lun->pending_ua[i-persis_offset] |=
8209                                                   CTL_UA_REG_PREEMPT;
8210                                 } else if (type != lun->res_type
8211                                         && (lun->res_type == SPR_TYPE_WR_EX_RO
8212                                          || lun->res_type ==SPR_TYPE_EX_AC_RO)){
8213                                                 if (!persis_offset
8214                                                  && i < CTL_MAX_INITIATORS)
8215                                                         lun->pending_ua[i] |=
8216                                                         CTL_UA_RES_RELEASE;
8217                                                 else if (persis_offset
8218                                                       && i >= persis_offset)
8219                                                         lun->pending_ua[
8220                                                         i-persis_offset] |=
8221                                                         CTL_UA_RES_RELEASE;
8222                                 }
8223                         }
8224                         lun->per_res[residx].registered = 1;
8225                         lun->res_type = type;
8226                         if (lun->res_type != SPR_TYPE_WR_EX_AR
8227                          && lun->res_type != SPR_TYPE_EX_AC_AR)
8228                                 lun->pr_res_idx = residx;
8229                         else
8230                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8231
8232                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8233                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8234                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8235                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8236                         persis_io.pr.pr_info.res_type = type;
8237                         memcpy(persis_io.pr.pr_info.sa_res_key,
8238                                param->serv_act_res_key,
8239                                sizeof(param->serv_act_res_key));
8240                         if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8241                              &persis_io, sizeof(persis_io), 0)) >
8242                              CTL_HA_STATUS_SUCCESS) {
8243                                 printf("CTL:Persis Out error returned "
8244                                        "from ctl_ha_msg_send %d\n",
8245                                        isc_retval);
8246                         }
8247                 } else {
8248                         /*
8249                          * sa_res_key is not the res holder just
8250                          * remove registrants
8251                          */
8252                         int found=0;
8253
8254                         for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8255                                 if (memcmp(param->serv_act_res_key,
8256                                     lun->per_res[i].res_key.key,
8257                                     sizeof(struct scsi_per_res_key)) != 0)
8258                                         continue;
8259
8260                                 found = 1;
8261                                 lun->per_res[i].registered = 0;
8262                                 memset(&lun->per_res[i].res_key, 0,
8263                                        sizeof(struct scsi_per_res_key));
8264                                 lun->pr_key_count--;
8265
8266                                 if (!persis_offset
8267                                  && i < CTL_MAX_INITIATORS)
8268                                         lun->pending_ua[i] |=
8269                                                 CTL_UA_REG_PREEMPT;
8270                                 else if (persis_offset
8271                                       && i >= persis_offset)
8272                                         lun->pending_ua[i-persis_offset] |=
8273                                                 CTL_UA_REG_PREEMPT;
8274                         }
8275
8276                         if (!found) {
8277                                 mtx_unlock(&lun->lun_lock);
8278                                 free(ctsio->kern_data_ptr, M_CTL);
8279                                 ctl_set_reservation_conflict(ctsio);
8280                                 ctl_done((union ctl_io *)ctsio);
8281                                 return (1);
8282                         }
8283                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8284                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8285                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8286                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8287                         persis_io.pr.pr_info.res_type = type;
8288                         memcpy(persis_io.pr.pr_info.sa_res_key,
8289                                param->serv_act_res_key,
8290                                sizeof(param->serv_act_res_key));
8291                         if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8292                              &persis_io, sizeof(persis_io), 0)) >
8293                              CTL_HA_STATUS_SUCCESS) {
8294                                 printf("CTL:Persis Out error returned "
8295                                        "from ctl_ha_msg_send %d\n",
8296                                 isc_retval);
8297                         }
8298                 }
8299         }
8300
8301         lun->PRGeneration++;
8302         mtx_unlock(&lun->lun_lock);
8303
8304         return (retval);
8305 }
8306
8307 static void
8308 ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
8309 {
8310         int i;
8311
8312         if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8313          || lun->pr_res_idx == CTL_PR_NO_RESERVATION
8314          || memcmp(&lun->per_res[lun->pr_res_idx].res_key,
8315                    msg->pr.pr_info.sa_res_key,
8316                    sizeof(struct scsi_per_res_key)) != 0) {
8317                 uint64_t sa_res_key;
8318                 sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
8319
8320                 if (sa_res_key == 0) {
8321                         /* temporarily unregister this nexus */
8322                         lun->per_res[msg->pr.pr_info.residx].registered = 0;
8323
8324                         /*
8325                          * Unregister everybody else and build UA for
8326                          * them
8327                          */
8328                         for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8329                                 if (lun->per_res[i].registered == 0)
8330                                         continue;
8331
8332                                 if (!persis_offset
8333                                  && i < CTL_MAX_INITIATORS)
8334                                         lun->pending_ua[i] |=
8335                                                 CTL_UA_REG_PREEMPT;
8336                                 else if (persis_offset && i >= persis_offset)
8337                                         lun->pending_ua[i - persis_offset] |=
8338                                                 CTL_UA_REG_PREEMPT;
8339                                 lun->per_res[i].registered = 0;
8340                                 memset(&lun->per_res[i].res_key, 0,
8341                                        sizeof(struct scsi_per_res_key));
8342                         }
8343
8344                         lun->per_res[msg->pr.pr_info.residx].registered = 1;
8345                         lun->pr_key_count = 1;
8346                         lun->res_type = msg->pr.pr_info.res_type;
8347                         if (lun->res_type != SPR_TYPE_WR_EX_AR
8348                          && lun->res_type != SPR_TYPE_EX_AC_AR)
8349                                 lun->pr_res_idx = msg->pr.pr_info.residx;
8350                 } else {
8351                         for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8352                                 if (memcmp(msg->pr.pr_info.sa_res_key,
8353                                    lun->per_res[i].res_key.key,
8354                                    sizeof(struct scsi_per_res_key)) != 0)
8355                                         continue;
8356
8357                                 lun->per_res[i].registered = 0;
8358                                 memset(&lun->per_res[i].res_key, 0,
8359                                        sizeof(struct scsi_per_res_key));
8360                                 lun->pr_key_count--;
8361
8362                                 if (!persis_offset
8363                                  && i < persis_offset)
8364                                         lun->pending_ua[i] |=
8365                                                 CTL_UA_REG_PREEMPT;
8366                                 else if (persis_offset
8367                                       && i >= persis_offset)
8368                                         lun->pending_ua[i - persis_offset] |=
8369                                                 CTL_UA_REG_PREEMPT;
8370                         }
8371                 }
8372         } else {
8373                 /*
8374                  * Temporarily unregister so it won't get removed
8375                  * or UA generated
8376                  */
8377                 lun->per_res[msg->pr.pr_info.residx].registered = 0;
8378                 for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8379                         if (lun->per_res[i].registered == 0)
8380                                 continue;
8381
8382                         if (memcmp(msg->pr.pr_info.sa_res_key,
8383                            lun->per_res[i].res_key.key,
8384                            sizeof(struct scsi_per_res_key)) == 0) {
8385                                 lun->per_res[i].registered = 0;
8386                                 memset(&lun->per_res[i].res_key, 0,
8387                                        sizeof(struct scsi_per_res_key));
8388                                 lun->pr_key_count--;
8389                                 if (!persis_offset
8390                                  && i < CTL_MAX_INITIATORS)
8391                                         lun->pending_ua[i] |=
8392                                                 CTL_UA_REG_PREEMPT;
8393                                 else if (persis_offset
8394                                       && i >= persis_offset)
8395                                         lun->pending_ua[i - persis_offset] |=
8396                                                 CTL_UA_REG_PREEMPT;
8397                         } else if (msg->pr.pr_info.res_type != lun->res_type
8398                                 && (lun->res_type == SPR_TYPE_WR_EX_RO
8399                                  || lun->res_type == SPR_TYPE_EX_AC_RO)) {
8400                                         if (!persis_offset
8401                                          && i < persis_offset)
8402                                                 lun->pending_ua[i] |=
8403                                                         CTL_UA_RES_RELEASE;
8404                                         else if (persis_offset
8405                                               && i >= persis_offset)
8406                                         lun->pending_ua[i - persis_offset] |=
8407                                                 CTL_UA_RES_RELEASE;
8408                         }
8409                 }
8410                 lun->per_res[msg->pr.pr_info.residx].registered = 1;
8411                 lun->res_type = msg->pr.pr_info.res_type;
8412                 if (lun->res_type != SPR_TYPE_WR_EX_AR
8413                  && lun->res_type != SPR_TYPE_EX_AC_AR)
8414                         lun->pr_res_idx = msg->pr.pr_info.residx;
8415                 else
8416                         lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8417         }
8418         lun->PRGeneration++;
8419
8420 }
8421
8422
8423 int
8424 ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8425 {
8426         int retval;
8427         int isc_retval;
8428         u_int32_t param_len;
8429         struct scsi_per_res_out *cdb;
8430         struct ctl_lun *lun;
8431         struct scsi_per_res_out_parms* param;
8432         struct ctl_softc *softc;
8433         uint32_t residx;
8434         uint64_t res_key, sa_res_key;
8435         uint8_t type;
8436         union ctl_ha_msg persis_io;
8437         int    i;
8438
8439         CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8440
8441         retval = CTL_RETVAL_COMPLETE;
8442
8443         softc = control_softc;
8444
8445         cdb = (struct scsi_per_res_out *)ctsio->cdb;
8446         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8447
8448         /*
8449          * We only support whole-LUN scope.  The scope & type are ignored for
8450          * register, register and ignore existing key and clear.
8451          * We sometimes ignore scope and type on preempts too!!
8452          * Verify reservation type here as well.
8453          */
8454         type = cdb->scope_type & SPR_TYPE_MASK;
8455         if ((cdb->action == SPRO_RESERVE)
8456          || (cdb->action == SPRO_RELEASE)) {
8457                 if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8458                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8459                                               /*sks_valid*/ 1,
8460                                               /*command*/ 1,
8461                                               /*field*/ 2,
8462                                               /*bit_valid*/ 1,
8463                                               /*bit*/ 4);
8464                         ctl_done((union ctl_io *)ctsio);
8465                         return (CTL_RETVAL_COMPLETE);
8466                 }
8467
8468                 if (type>8 || type==2 || type==4 || type==0) {
8469                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8470                                               /*sks_valid*/ 1,
8471                                               /*command*/ 1,
8472                                               /*field*/ 2,
8473                                               /*bit_valid*/ 1,
8474                                               /*bit*/ 0);
8475                         ctl_done((union ctl_io *)ctsio);
8476                         return (CTL_RETVAL_COMPLETE);
8477                 }
8478         }
8479
8480         param_len = scsi_4btoul(cdb->length);
8481
8482         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8483                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8484                 ctsio->kern_data_len = param_len;
8485                 ctsio->kern_total_len = param_len;
8486                 ctsio->kern_data_resid = 0;
8487                 ctsio->kern_rel_offset = 0;
8488                 ctsio->kern_sg_entries = 0;
8489                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8490                 ctsio->be_move_done = ctl_config_move_done;
8491                 ctl_datamove((union ctl_io *)ctsio);
8492
8493                 return (CTL_RETVAL_COMPLETE);
8494         }
8495
8496         param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8497
8498         residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8499         res_key = scsi_8btou64(param->res_key.key);
8500         sa_res_key = scsi_8btou64(param->serv_act_res_key);
8501
8502         /*
8503          * Validate the reservation key here except for SPRO_REG_IGNO
8504          * This must be done for all other service actions
8505          */
8506         if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8507                 mtx_lock(&lun->lun_lock);
8508                 if (lun->per_res[residx].registered) {
8509                     if (memcmp(param->res_key.key,
8510                                lun->per_res[residx].res_key.key,
8511                                ctl_min(sizeof(param->res_key),
8512                                sizeof(lun->per_res[residx].res_key))) != 0) {
8513                                 /*
8514                                  * The current key passed in doesn't match
8515                                  * the one the initiator previously
8516                                  * registered.
8517                                  */
8518                                 mtx_unlock(&lun->lun_lock);
8519                                 free(ctsio->kern_data_ptr, M_CTL);
8520                                 ctl_set_reservation_conflict(ctsio);
8521                                 ctl_done((union ctl_io *)ctsio);
8522                                 return (CTL_RETVAL_COMPLETE);
8523                         }
8524                 } else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8525                         /*
8526                          * We are not registered
8527                          */
8528                         mtx_unlock(&lun->lun_lock);
8529                         free(ctsio->kern_data_ptr, M_CTL);
8530                         ctl_set_reservation_conflict(ctsio);
8531                         ctl_done((union ctl_io *)ctsio);
8532                         return (CTL_RETVAL_COMPLETE);
8533                 } else if (res_key != 0) {
8534                         /*
8535                          * We are not registered and trying to register but
8536                          * the register key isn't zero.
8537                          */
8538                         mtx_unlock(&lun->lun_lock);
8539                         free(ctsio->kern_data_ptr, M_CTL);
8540                         ctl_set_reservation_conflict(ctsio);
8541                         ctl_done((union ctl_io *)ctsio);
8542                         return (CTL_RETVAL_COMPLETE);
8543                 }
8544                 mtx_unlock(&lun->lun_lock);
8545         }
8546
8547         switch (cdb->action & SPRO_ACTION_MASK) {
8548         case SPRO_REGISTER:
8549         case SPRO_REG_IGNO: {
8550
8551 #if 0
8552                 printf("Registration received\n");
8553 #endif
8554
8555                 /*
8556                  * We don't support any of these options, as we report in
8557                  * the read capabilities request (see
8558                  * ctl_persistent_reserve_in(), above).
8559                  */
8560                 if ((param->flags & SPR_SPEC_I_PT)
8561                  || (param->flags & SPR_ALL_TG_PT)
8562                  || (param->flags & SPR_APTPL)) {
8563                         int bit_ptr;
8564
8565                         if (param->flags & SPR_APTPL)
8566                                 bit_ptr = 0;
8567                         else if (param->flags & SPR_ALL_TG_PT)
8568                                 bit_ptr = 2;
8569                         else /* SPR_SPEC_I_PT */
8570                                 bit_ptr = 3;
8571
8572                         free(ctsio->kern_data_ptr, M_CTL);
8573                         ctl_set_invalid_field(ctsio,
8574                                               /*sks_valid*/ 1,
8575                                               /*command*/ 0,
8576                                               /*field*/ 20,
8577                                               /*bit_valid*/ 1,
8578                                               /*bit*/ bit_ptr);
8579                         ctl_done((union ctl_io *)ctsio);
8580                         return (CTL_RETVAL_COMPLETE);
8581                 }
8582
8583                 mtx_lock(&lun->lun_lock);
8584
8585                 /*
8586                  * The initiator wants to clear the
8587                  * key/unregister.
8588                  */
8589                 if (sa_res_key == 0) {
8590                         if ((res_key == 0
8591                           && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8592                          || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8593                           && !lun->per_res[residx].registered)) {
8594                                 mtx_unlock(&lun->lun_lock);
8595                                 goto done;
8596                         }
8597
8598                         lun->per_res[residx].registered = 0;
8599                         memset(&lun->per_res[residx].res_key,
8600                                0, sizeof(lun->per_res[residx].res_key));
8601                         lun->pr_key_count--;
8602
8603                         if (residx == lun->pr_res_idx) {
8604                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8605                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8606
8607                                 if ((lun->res_type == SPR_TYPE_WR_EX_RO
8608                                   || lun->res_type == SPR_TYPE_EX_AC_RO)
8609                                  && lun->pr_key_count) {
8610                                         /*
8611                                          * If the reservation is a registrants
8612                                          * only type we need to generate a UA
8613                                          * for other registered inits.  The
8614                                          * sense code should be RESERVATIONS
8615                                          * RELEASED
8616                                          */
8617
8618                                         for (i = 0; i < CTL_MAX_INITIATORS;i++){
8619                                                 if (lun->per_res[
8620                                                     i+persis_offset].registered
8621                                                     == 0)
8622                                                         continue;
8623                                                 lun->pending_ua[i] |=
8624                                                         CTL_UA_RES_RELEASE;
8625                                         }
8626                                 }
8627                                 lun->res_type = 0;
8628                         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8629                                 if (lun->pr_key_count==0) {
8630                                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8631                                         lun->res_type = 0;
8632                                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8633                                 }
8634                         }
8635                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8636                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8637                         persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8638                         persis_io.pr.pr_info.residx = residx;
8639                         if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8640                              &persis_io, sizeof(persis_io), 0 )) >
8641                              CTL_HA_STATUS_SUCCESS) {
8642                                 printf("CTL:Persis Out error returned from "
8643                                        "ctl_ha_msg_send %d\n", isc_retval);
8644                         }
8645                 } else /* sa_res_key != 0 */ {
8646
8647                         /*
8648                          * If we aren't registered currently then increment
8649                          * the key count and set the registered flag.
8650                          */
8651                         if (!lun->per_res[residx].registered) {
8652                                 lun->pr_key_count++;
8653                                 lun->per_res[residx].registered = 1;
8654                         }
8655
8656                         memcpy(&lun->per_res[residx].res_key,
8657                                param->serv_act_res_key,
8658                                ctl_min(sizeof(param->serv_act_res_key),
8659                                sizeof(lun->per_res[residx].res_key)));
8660
8661                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8662                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8663                         persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8664                         persis_io.pr.pr_info.residx = residx;
8665                         memcpy(persis_io.pr.pr_info.sa_res_key,
8666                                param->serv_act_res_key,
8667                                sizeof(param->serv_act_res_key));
8668                         if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8669                              &persis_io, sizeof(persis_io), 0)) >
8670                              CTL_HA_STATUS_SUCCESS) {
8671                                 printf("CTL:Persis Out error returned from "
8672                                        "ctl_ha_msg_send %d\n", isc_retval);
8673                         }
8674                 }
8675                 lun->PRGeneration++;
8676                 mtx_unlock(&lun->lun_lock);
8677
8678                 break;
8679         }
8680         case SPRO_RESERVE:
8681 #if 0
8682                 printf("Reserve executed type %d\n", type);
8683 #endif
8684                 mtx_lock(&lun->lun_lock);
8685                 if (lun->flags & CTL_LUN_PR_RESERVED) {
8686                         /*
8687                          * if this isn't the reservation holder and it's
8688                          * not a "all registrants" type or if the type is
8689                          * different then we have a conflict
8690                          */
8691                         if ((lun->pr_res_idx != residx
8692                           && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8693                          || lun->res_type != type) {
8694                                 mtx_unlock(&lun->lun_lock);
8695                                 free(ctsio->kern_data_ptr, M_CTL);
8696                                 ctl_set_reservation_conflict(ctsio);
8697                                 ctl_done((union ctl_io *)ctsio);
8698                                 return (CTL_RETVAL_COMPLETE);
8699                         }
8700                         mtx_unlock(&lun->lun_lock);
8701                 } else /* create a reservation */ {
8702                         /*
8703                          * If it's not an "all registrants" type record
8704                          * reservation holder
8705                          */
8706                         if (type != SPR_TYPE_WR_EX_AR
8707                          && type != SPR_TYPE_EX_AC_AR)
8708                                 lun->pr_res_idx = residx; /* Res holder */
8709                         else
8710                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8711
8712                         lun->flags |= CTL_LUN_PR_RESERVED;
8713                         lun->res_type = type;
8714
8715                         mtx_unlock(&lun->lun_lock);
8716
8717                         /* send msg to other side */
8718                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8719                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8720                         persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8721                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8722                         persis_io.pr.pr_info.res_type = type;
8723                         if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8724                              &persis_io, sizeof(persis_io), 0)) >
8725                              CTL_HA_STATUS_SUCCESS) {
8726                                 printf("CTL:Persis Out error returned from "
8727                                        "ctl_ha_msg_send %d\n", isc_retval);
8728                         }
8729                 }
8730                 break;
8731
8732         case SPRO_RELEASE:
8733                 mtx_lock(&lun->lun_lock);
8734                 if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8735                         /* No reservation exists return good status */
8736                         mtx_unlock(&lun->lun_lock);
8737                         goto done;
8738                 }
8739                 /*
8740                  * Is this nexus a reservation holder?
8741                  */
8742                 if (lun->pr_res_idx != residx
8743                  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8744                         /*
8745                          * not a res holder return good status but
8746                          * do nothing
8747                          */
8748                         mtx_unlock(&lun->lun_lock);
8749                         goto done;
8750                 }
8751
8752                 if (lun->res_type != type) {
8753                         mtx_unlock(&lun->lun_lock);
8754                         free(ctsio->kern_data_ptr, M_CTL);
8755                         ctl_set_illegal_pr_release(ctsio);
8756                         ctl_done((union ctl_io *)ctsio);
8757                         return (CTL_RETVAL_COMPLETE);
8758                 }
8759
8760                 /* okay to release */
8761                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8762                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8763                 lun->res_type = 0;
8764
8765                 /*
8766                  * if this isn't an exclusive access
8767                  * res generate UA for all other
8768                  * registrants.
8769                  */
8770                 if (type != SPR_TYPE_EX_AC
8771                  && type != SPR_TYPE_WR_EX) {
8772                         /*
8773                          * temporarily unregister so we don't generate UA
8774                          */
8775                         lun->per_res[residx].registered = 0;
8776
8777                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8778                                 if (lun->per_res[i+persis_offset].registered
8779                                     == 0)
8780                                         continue;
8781                                 lun->pending_ua[i] |=
8782                                         CTL_UA_RES_RELEASE;
8783                         }
8784
8785                         lun->per_res[residx].registered = 1;
8786                 }
8787                 mtx_unlock(&lun->lun_lock);
8788                 /* Send msg to other side */
8789                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8790                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8791                 persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8792                 if ((isc_retval=ctl_ha_msg_send( CTL_HA_CHAN_CTL, &persis_io,
8793                      sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8794                         printf("CTL:Persis Out error returned from "
8795                                "ctl_ha_msg_send %d\n", isc_retval);
8796                 }
8797                 break;
8798
8799         case SPRO_CLEAR:
8800                 /* send msg to other side */
8801
8802                 mtx_lock(&lun->lun_lock);
8803                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8804                 lun->res_type = 0;
8805                 lun->pr_key_count = 0;
8806                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8807
8808
8809                 memset(&lun->per_res[residx].res_key,
8810                        0, sizeof(lun->per_res[residx].res_key));
8811                 lun->per_res[residx].registered = 0;
8812
8813                 for (i=0; i < 2*CTL_MAX_INITIATORS; i++)
8814                         if (lun->per_res[i].registered) {
8815                                 if (!persis_offset && i < CTL_MAX_INITIATORS)
8816                                         lun->pending_ua[i] |=
8817                                                 CTL_UA_RES_PREEMPT;
8818                                 else if (persis_offset && i >= persis_offset)
8819                                         lun->pending_ua[i-persis_offset] |=
8820                                             CTL_UA_RES_PREEMPT;
8821
8822                                 memset(&lun->per_res[i].res_key,
8823                                        0, sizeof(struct scsi_per_res_key));
8824                                 lun->per_res[i].registered = 0;
8825                         }
8826                 lun->PRGeneration++;
8827                 mtx_unlock(&lun->lun_lock);
8828                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8829                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8830                 persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8831                 if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8832                      sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8833                         printf("CTL:Persis Out error returned from "
8834                                "ctl_ha_msg_send %d\n", isc_retval);
8835                 }
8836                 break;
8837
8838         case SPRO_PREEMPT: {
8839                 int nretval;
8840
8841                 nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8842                                           residx, ctsio, cdb, param);
8843                 if (nretval != 0)
8844                         return (CTL_RETVAL_COMPLETE);
8845                 break;
8846         }
8847         default:
8848                 panic("Invalid PR type %x", cdb->action);
8849         }
8850
8851 done:
8852         free(ctsio->kern_data_ptr, M_CTL);
8853         ctl_set_success(ctsio);
8854         ctl_done((union ctl_io *)ctsio);
8855
8856         return (retval);
8857 }
8858
8859 /*
8860  * This routine is for handling a message from the other SC pertaining to
8861  * persistent reserve out. All the error checking will have been done
8862  * so only perorming the action need be done here to keep the two
8863  * in sync.
8864  */
8865 static void
8866 ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8867 {
8868         struct ctl_lun *lun;
8869         struct ctl_softc *softc;
8870         int i;
8871         uint32_t targ_lun;
8872
8873         softc = control_softc;
8874
8875         targ_lun = msg->hdr.nexus.targ_mapped_lun;
8876         lun = softc->ctl_luns[targ_lun];
8877         mtx_lock(&lun->lun_lock);
8878         switch(msg->pr.pr_info.action) {
8879         case CTL_PR_REG_KEY:
8880                 if (!lun->per_res[msg->pr.pr_info.residx].registered) {
8881                         lun->per_res[msg->pr.pr_info.residx].registered = 1;
8882                         lun->pr_key_count++;
8883                 }
8884                 lun->PRGeneration++;
8885                 memcpy(&lun->per_res[msg->pr.pr_info.residx].res_key,
8886                        msg->pr.pr_info.sa_res_key,
8887                        sizeof(struct scsi_per_res_key));
8888                 break;
8889
8890         case CTL_PR_UNREG_KEY:
8891                 lun->per_res[msg->pr.pr_info.residx].registered = 0;
8892                 memset(&lun->per_res[msg->pr.pr_info.residx].res_key,
8893                        0, sizeof(struct scsi_per_res_key));
8894                 lun->pr_key_count--;
8895
8896                 /* XXX Need to see if the reservation has been released */
8897                 /* if so do we need to generate UA? */
8898                 if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8899                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8900                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8901
8902                         if ((lun->res_type == SPR_TYPE_WR_EX_RO
8903                           || lun->res_type == SPR_TYPE_EX_AC_RO)
8904                          && lun->pr_key_count) {
8905                                 /*
8906                                  * If the reservation is a registrants
8907                                  * only type we need to generate a UA
8908                                  * for other registered inits.  The
8909                                  * sense code should be RESERVATIONS
8910                                  * RELEASED
8911                                  */
8912
8913                                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8914                                         if (lun->per_res[i+
8915                                             persis_offset].registered == 0)
8916                                                 continue;
8917
8918                                         lun->pending_ua[i] |=
8919                                                 CTL_UA_RES_RELEASE;
8920                                 }
8921                         }
8922                         lun->res_type = 0;
8923                 } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8924                         if (lun->pr_key_count==0) {
8925                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8926                                 lun->res_type = 0;
8927                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8928                         }
8929                 }
8930                 lun->PRGeneration++;
8931                 break;
8932
8933         case CTL_PR_RESERVE:
8934                 lun->flags |= CTL_LUN_PR_RESERVED;
8935                 lun->res_type = msg->pr.pr_info.res_type;
8936                 lun->pr_res_idx = msg->pr.pr_info.residx;
8937
8938                 break;
8939
8940         case CTL_PR_RELEASE:
8941                 /*
8942                  * if this isn't an exclusive access res generate UA for all
8943                  * other registrants.
8944                  */
8945                 if (lun->res_type != SPR_TYPE_EX_AC
8946                  && lun->res_type != SPR_TYPE_WR_EX) {
8947                         for (i = 0; i < CTL_MAX_INITIATORS; i++)
8948                                 if (lun->per_res[i+persis_offset].registered)
8949                                         lun->pending_ua[i] |=
8950                                                 CTL_UA_RES_RELEASE;
8951                 }
8952
8953                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8954                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8955                 lun->res_type = 0;
8956                 break;
8957
8958         case CTL_PR_PREEMPT:
8959                 ctl_pro_preempt_other(lun, msg);
8960                 break;
8961         case CTL_PR_CLEAR:
8962                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8963                 lun->res_type = 0;
8964                 lun->pr_key_count = 0;
8965                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8966
8967                 for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8968                         if (lun->per_res[i].registered == 0)
8969                                 continue;
8970                         if (!persis_offset
8971                          && i < CTL_MAX_INITIATORS)
8972                                 lun->pending_ua[i] |= CTL_UA_RES_PREEMPT;
8973                         else if (persis_offset
8974                               && i >= persis_offset)
8975                                 lun->pending_ua[i-persis_offset] |=
8976                                         CTL_UA_RES_PREEMPT;
8977                         memset(&lun->per_res[i].res_key, 0,
8978                                sizeof(struct scsi_per_res_key));
8979                         lun->per_res[i].registered = 0;
8980                 }
8981                 lun->PRGeneration++;
8982                 break;
8983         }
8984
8985         mtx_unlock(&lun->lun_lock);
8986 }
8987
8988 int
8989 ctl_read_write(struct ctl_scsiio *ctsio)
8990 {
8991         struct ctl_lun *lun;
8992         struct ctl_lba_len_flags *lbalen;
8993         uint64_t lba;
8994         uint32_t num_blocks;
8995         int fua, dpo;
8996         int retval;
8997         int isread;
8998
8999         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9000
9001         CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
9002
9003         fua = 0;
9004         dpo = 0;
9005
9006         retval = CTL_RETVAL_COMPLETE;
9007
9008         isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
9009               || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
9010         if (lun->flags & CTL_LUN_PR_RESERVED && isread) {
9011                 uint32_t residx;
9012
9013                 /*
9014                  * XXX KDM need a lock here.
9015                  */
9016                 residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
9017                 if ((lun->res_type == SPR_TYPE_EX_AC
9018                   && residx != lun->pr_res_idx)
9019                  || ((lun->res_type == SPR_TYPE_EX_AC_RO
9020                    || lun->res_type == SPR_TYPE_EX_AC_AR)
9021                   && !lun->per_res[residx].registered)) {
9022                         ctl_set_reservation_conflict(ctsio);
9023                         ctl_done((union ctl_io *)ctsio);
9024                         return (CTL_RETVAL_COMPLETE);
9025                 }
9026         }
9027
9028         switch (ctsio->cdb[0]) {
9029         case READ_6:
9030         case WRITE_6: {
9031                 struct scsi_rw_6 *cdb;
9032
9033                 cdb = (struct scsi_rw_6 *)ctsio->cdb;
9034
9035                 lba = scsi_3btoul(cdb->addr);
9036                 /* only 5 bits are valid in the most significant address byte */
9037                 lba &= 0x1fffff;
9038                 num_blocks = cdb->length;
9039                 /*
9040                  * This is correct according to SBC-2.
9041                  */
9042                 if (num_blocks == 0)
9043                         num_blocks = 256;
9044                 break;
9045         }
9046         case READ_10:
9047         case WRITE_10: {
9048                 struct scsi_rw_10 *cdb;
9049
9050                 cdb = (struct scsi_rw_10 *)ctsio->cdb;
9051
9052                 if (cdb->byte2 & SRW10_FUA)
9053                         fua = 1;
9054                 if (cdb->byte2 & SRW10_DPO)
9055                         dpo = 1;
9056
9057                 lba = scsi_4btoul(cdb->addr);
9058                 num_blocks = scsi_2btoul(cdb->length);
9059                 break;
9060         }
9061         case WRITE_VERIFY_10: {
9062                 struct scsi_write_verify_10 *cdb;
9063
9064                 cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
9065
9066                 /*
9067                  * XXX KDM we should do actual write verify support at some
9068                  * point.  This is obviously fake, we're just translating
9069                  * things to a write.  So we don't even bother checking the
9070                  * BYTCHK field, since we don't do any verification.  If
9071                  * the user asks for it, we'll just pretend we did it.
9072                  */
9073                 if (cdb->byte2 & SWV_DPO)
9074                         dpo = 1;
9075
9076                 lba = scsi_4btoul(cdb->addr);
9077                 num_blocks = scsi_2btoul(cdb->length);
9078                 break;
9079         }
9080         case READ_12:
9081         case WRITE_12: {
9082                 struct scsi_rw_12 *cdb;
9083
9084                 cdb = (struct scsi_rw_12 *)ctsio->cdb;
9085
9086                 if (cdb->byte2 & SRW12_FUA)
9087                         fua = 1;
9088                 if (cdb->byte2 & SRW12_DPO)
9089                         dpo = 1;
9090                 lba = scsi_4btoul(cdb->addr);
9091                 num_blocks = scsi_4btoul(cdb->length);
9092                 break;
9093         }
9094         case WRITE_VERIFY_12: {
9095                 struct scsi_write_verify_12 *cdb;
9096
9097                 cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
9098
9099                 if (cdb->byte2 & SWV_DPO)
9100                         dpo = 1;
9101                 
9102                 lba = scsi_4btoul(cdb->addr);
9103                 num_blocks = scsi_4btoul(cdb->length);
9104
9105                 break;
9106         }
9107         case READ_16:
9108         case WRITE_16: {
9109                 struct scsi_rw_16 *cdb;
9110
9111                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
9112
9113                 if (cdb->byte2 & SRW12_FUA)
9114                         fua = 1;
9115                 if (cdb->byte2 & SRW12_DPO)
9116                         dpo = 1;
9117
9118                 lba = scsi_8btou64(cdb->addr);
9119                 num_blocks = scsi_4btoul(cdb->length);
9120                 break;
9121         }
9122         case WRITE_VERIFY_16: {
9123                 struct scsi_write_verify_16 *cdb;
9124
9125                 cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
9126
9127                 if (cdb->byte2 & SWV_DPO)
9128                         dpo = 1;
9129
9130                 lba = scsi_8btou64(cdb->addr);
9131                 num_blocks = scsi_4btoul(cdb->length);
9132                 break;
9133         }
9134         default:
9135                 /*
9136                  * We got a command we don't support.  This shouldn't
9137                  * happen, commands should be filtered out above us.
9138                  */
9139                 ctl_set_invalid_opcode(ctsio);
9140                 ctl_done((union ctl_io *)ctsio);
9141
9142                 return (CTL_RETVAL_COMPLETE);
9143                 break; /* NOTREACHED */
9144         }
9145
9146         /*
9147          * XXX KDM what do we do with the DPO and FUA bits?  FUA might be
9148          * interesting for us, but if RAIDCore is in write-back mode,
9149          * getting it to do write-through for a particular transaction may
9150          * not be possible.
9151          */
9152
9153         /*
9154          * The first check is to make sure we're in bounds, the second
9155          * check is to catch wrap-around problems.  If the lba + num blocks
9156          * is less than the lba, then we've wrapped around and the block
9157          * range is invalid anyway.
9158          */
9159         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9160          || ((lba + num_blocks) < lba)) {
9161                 ctl_set_lba_out_of_range(ctsio);
9162                 ctl_done((union ctl_io *)ctsio);
9163                 return (CTL_RETVAL_COMPLETE);
9164         }
9165
9166         /*
9167          * According to SBC-3, a transfer length of 0 is not an error.
9168          * Note that this cannot happen with WRITE(6) or READ(6), since 0
9169          * translates to 256 blocks for those commands.
9170          */
9171         if (num_blocks == 0) {
9172                 ctl_set_success(ctsio);
9173                 ctl_done((union ctl_io *)ctsio);
9174                 return (CTL_RETVAL_COMPLETE);
9175         }
9176
9177         lbalen = (struct ctl_lba_len_flags *)
9178             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9179         lbalen->lba = lba;
9180         lbalen->len = num_blocks;
9181         lbalen->flags = isread ? CTL_LLF_READ : CTL_LLF_WRITE;
9182
9183         ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9184         ctsio->kern_rel_offset = 0;
9185
9186         CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
9187
9188         retval = lun->backend->data_submit((union ctl_io *)ctsio);
9189
9190         return (retval);
9191 }
9192
9193 static int
9194 ctl_cnw_cont(union ctl_io *io)
9195 {
9196         struct ctl_scsiio *ctsio;
9197         struct ctl_lun *lun;
9198         struct ctl_lba_len_flags *lbalen;
9199         int retval;
9200
9201         ctsio = &io->scsiio;
9202         ctsio->io_hdr.status = CTL_STATUS_NONE;
9203         ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
9204         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9205         lbalen = (struct ctl_lba_len_flags *)
9206             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9207         lbalen->flags = CTL_LLF_WRITE;
9208
9209         CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
9210         retval = lun->backend->data_submit((union ctl_io *)ctsio);
9211         return (retval);
9212 }
9213
9214 int
9215 ctl_cnw(struct ctl_scsiio *ctsio)
9216 {
9217         struct ctl_lun *lun;
9218         struct ctl_lba_len_flags *lbalen;
9219         uint64_t lba;
9220         uint32_t num_blocks;
9221         int fua, dpo;
9222         int retval;
9223
9224         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9225
9226         CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
9227
9228         fua = 0;
9229         dpo = 0;
9230
9231         retval = CTL_RETVAL_COMPLETE;
9232
9233         switch (ctsio->cdb[0]) {
9234         case COMPARE_AND_WRITE: {
9235                 struct scsi_compare_and_write *cdb;
9236
9237                 cdb = (struct scsi_compare_and_write *)ctsio->cdb;
9238
9239                 if (cdb->byte2 & SRW10_FUA)
9240                         fua = 1;
9241                 if (cdb->byte2 & SRW10_DPO)
9242                         dpo = 1;
9243                 lba = scsi_8btou64(cdb->addr);
9244                 num_blocks = cdb->length;
9245                 break;
9246         }
9247         default:
9248                 /*
9249                  * We got a command we don't support.  This shouldn't
9250                  * happen, commands should be filtered out above us.
9251                  */
9252                 ctl_set_invalid_opcode(ctsio);
9253                 ctl_done((union ctl_io *)ctsio);
9254
9255                 return (CTL_RETVAL_COMPLETE);
9256                 break; /* NOTREACHED */
9257         }
9258
9259         /*
9260          * XXX KDM what do we do with the DPO and FUA bits?  FUA might be
9261          * interesting for us, but if RAIDCore is in write-back mode,
9262          * getting it to do write-through for a particular transaction may
9263          * not be possible.
9264          */
9265
9266         /*
9267          * The first check is to make sure we're in bounds, the second
9268          * check is to catch wrap-around problems.  If the lba + num blocks
9269          * is less than the lba, then we've wrapped around and the block
9270          * range is invalid anyway.
9271          */
9272         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9273          || ((lba + num_blocks) < lba)) {
9274                 ctl_set_lba_out_of_range(ctsio);
9275                 ctl_done((union ctl_io *)ctsio);
9276                 return (CTL_RETVAL_COMPLETE);
9277         }
9278
9279         /*
9280          * According to SBC-3, a transfer length of 0 is not an error.
9281          */
9282         if (num_blocks == 0) {
9283                 ctl_set_success(ctsio);
9284                 ctl_done((union ctl_io *)ctsio);
9285                 return (CTL_RETVAL_COMPLETE);
9286         }
9287
9288         ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
9289         ctsio->kern_rel_offset = 0;
9290
9291         /*
9292          * Set the IO_CONT flag, so that if this I/O gets passed to
9293          * ctl_data_submit_done(), it'll get passed back to
9294          * ctl_ctl_cnw_cont() for further processing.
9295          */
9296         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
9297         ctsio->io_cont = ctl_cnw_cont;
9298
9299         lbalen = (struct ctl_lba_len_flags *)
9300             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9301         lbalen->lba = lba;
9302         lbalen->len = num_blocks;
9303         lbalen->flags = CTL_LLF_COMPARE;
9304
9305         CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
9306         retval = lun->backend->data_submit((union ctl_io *)ctsio);
9307         return (retval);
9308 }
9309
9310 int
9311 ctl_verify(struct ctl_scsiio *ctsio)
9312 {
9313         struct ctl_lun *lun;
9314         struct ctl_lba_len_flags *lbalen;
9315         uint64_t lba;
9316         uint32_t num_blocks;
9317         int bytchk, dpo;
9318         int retval;
9319
9320         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9321
9322         CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
9323
9324         bytchk = 0;
9325         dpo = 0;
9326         retval = CTL_RETVAL_COMPLETE;
9327
9328         switch (ctsio->cdb[0]) {
9329         case VERIFY_10: {
9330                 struct scsi_verify_10 *cdb;
9331
9332                 cdb = (struct scsi_verify_10 *)ctsio->cdb;
9333                 if (cdb->byte2 & SVFY_BYTCHK)
9334                         bytchk = 1;
9335                 if (cdb->byte2 & SVFY_DPO)
9336                         dpo = 1;
9337                 lba = scsi_4btoul(cdb->addr);
9338                 num_blocks = scsi_2btoul(cdb->length);
9339                 break;
9340         }
9341         case VERIFY_12: {
9342                 struct scsi_verify_12 *cdb;
9343
9344                 cdb = (struct scsi_verify_12 *)ctsio->cdb;
9345                 if (cdb->byte2 & SVFY_BYTCHK)
9346                         bytchk = 1;
9347                 if (cdb->byte2 & SVFY_DPO)
9348                         dpo = 1;
9349                 lba = scsi_4btoul(cdb->addr);
9350                 num_blocks = scsi_4btoul(cdb->length);
9351                 break;
9352         }
9353         case VERIFY_16: {
9354                 struct scsi_rw_16 *cdb;
9355
9356                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
9357                 if (cdb->byte2 & SVFY_BYTCHK)
9358                         bytchk = 1;
9359                 if (cdb->byte2 & SVFY_DPO)
9360                         dpo = 1;
9361                 lba = scsi_8btou64(cdb->addr);
9362                 num_blocks = scsi_4btoul(cdb->length);
9363                 break;
9364         }
9365         default:
9366                 /*
9367                  * We got a command we don't support.  This shouldn't
9368                  * happen, commands should be filtered out above us.
9369                  */
9370                 ctl_set_invalid_opcode(ctsio);
9371                 ctl_done((union ctl_io *)ctsio);
9372                 return (CTL_RETVAL_COMPLETE);
9373         }
9374
9375         /*
9376          * The first check is to make sure we're in bounds, the second
9377          * check is to catch wrap-around problems.  If the lba + num blocks
9378          * is less than the lba, then we've wrapped around and the block
9379          * range is invalid anyway.
9380          */
9381         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9382          || ((lba + num_blocks) < lba)) {
9383                 ctl_set_lba_out_of_range(ctsio);
9384                 ctl_done((union ctl_io *)ctsio);
9385                 return (CTL_RETVAL_COMPLETE);
9386         }
9387
9388         /*
9389          * According to SBC-3, a transfer length of 0 is not an error.
9390          */
9391         if (num_blocks == 0) {
9392                 ctl_set_success(ctsio);
9393                 ctl_done((union ctl_io *)ctsio);
9394                 return (CTL_RETVAL_COMPLETE);
9395         }
9396
9397         lbalen = (struct ctl_lba_len_flags *)
9398             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9399         lbalen->lba = lba;
9400         lbalen->len = num_blocks;
9401         if (bytchk) {
9402                 lbalen->flags = CTL_LLF_COMPARE;
9403                 ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9404         } else {
9405                 lbalen->flags = CTL_LLF_VERIFY;
9406                 ctsio->kern_total_len = 0;
9407         }
9408         ctsio->kern_rel_offset = 0;
9409
9410         CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9411         retval = lun->backend->data_submit((union ctl_io *)ctsio);
9412         return (retval);
9413 }
9414
9415 int
9416 ctl_report_luns(struct ctl_scsiio *ctsio)
9417 {
9418         struct scsi_report_luns *cdb;
9419         struct scsi_report_luns_data *lun_data;
9420         struct ctl_lun *lun, *request_lun;
9421         int num_luns, retval;
9422         uint32_t alloc_len, lun_datalen;
9423         int num_filled, well_known;
9424         uint32_t initidx, targ_lun_id, lun_id;
9425
9426         retval = CTL_RETVAL_COMPLETE;
9427         well_known = 0;
9428
9429         cdb = (struct scsi_report_luns *)ctsio->cdb;
9430
9431         CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9432
9433         mtx_lock(&control_softc->ctl_lock);
9434         num_luns = control_softc->num_luns;
9435         mtx_unlock(&control_softc->ctl_lock);
9436
9437         switch (cdb->select_report) {
9438         case RPL_REPORT_DEFAULT:
9439         case RPL_REPORT_ALL:
9440                 break;
9441         case RPL_REPORT_WELLKNOWN:
9442                 well_known = 1;
9443                 num_luns = 0;
9444                 break;
9445         default:
9446                 ctl_set_invalid_field(ctsio,
9447                                       /*sks_valid*/ 1,
9448                                       /*command*/ 1,
9449                                       /*field*/ 2,
9450                                       /*bit_valid*/ 0,
9451                                       /*bit*/ 0);
9452                 ctl_done((union ctl_io *)ctsio);
9453                 return (retval);
9454                 break; /* NOTREACHED */
9455         }
9456
9457         alloc_len = scsi_4btoul(cdb->length);
9458         /*
9459          * The initiator has to allocate at least 16 bytes for this request,
9460          * so he can at least get the header and the first LUN.  Otherwise
9461          * we reject the request (per SPC-3 rev 14, section 6.21).
9462          */
9463         if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9464             sizeof(struct scsi_report_luns_lundata))) {
9465                 ctl_set_invalid_field(ctsio,
9466                                       /*sks_valid*/ 1,
9467                                       /*command*/ 1,
9468                                       /*field*/ 6,
9469                                       /*bit_valid*/ 0,
9470                                       /*bit*/ 0);
9471                 ctl_done((union ctl_io *)ctsio);
9472                 return (retval);
9473         }
9474
9475         request_lun = (struct ctl_lun *)
9476                 ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9477
9478         lun_datalen = sizeof(*lun_data) +
9479                 (num_luns * sizeof(struct scsi_report_luns_lundata));
9480
9481         ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9482         lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9483         ctsio->kern_sg_entries = 0;
9484
9485         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9486
9487         mtx_lock(&control_softc->ctl_lock);
9488         for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
9489                 lun_id = ctl_map_lun(ctsio->io_hdr.nexus.targ_port, targ_lun_id);
9490                 if (lun_id >= CTL_MAX_LUNS)
9491                         continue;
9492                 lun = control_softc->ctl_luns[lun_id];
9493                 if (lun == NULL)
9494                         continue;
9495
9496                 if (targ_lun_id <= 0xff) {
9497                         /*
9498                          * Peripheral addressing method, bus number 0.
9499                          */
9500                         lun_data->luns[num_filled].lundata[0] =
9501                                 RPL_LUNDATA_ATYP_PERIPH;
9502                         lun_data->luns[num_filled].lundata[1] = targ_lun_id;
9503                         num_filled++;
9504                 } else if (targ_lun_id <= 0x3fff) {
9505                         /*
9506                          * Flat addressing method.
9507                          */
9508                         lun_data->luns[num_filled].lundata[0] =
9509                                 RPL_LUNDATA_ATYP_FLAT |
9510                                 (targ_lun_id & RPL_LUNDATA_FLAT_LUN_MASK);
9511 #ifdef OLDCTLHEADERS
9512                                 (SRLD_ADDR_FLAT << SRLD_ADDR_SHIFT) |
9513                                 (targ_lun_id & SRLD_BUS_LUN_MASK);
9514 #endif
9515                         lun_data->luns[num_filled].lundata[1] =
9516 #ifdef OLDCTLHEADERS
9517                                 targ_lun_id >> SRLD_BUS_LUN_BITS;
9518 #endif
9519                                 targ_lun_id >> RPL_LUNDATA_FLAT_LUN_BITS;
9520                         num_filled++;
9521                 } else {
9522                         printf("ctl_report_luns: bogus LUN number %jd, "
9523                                "skipping\n", (intmax_t)targ_lun_id);
9524                 }
9525                 /*
9526                  * According to SPC-3, rev 14 section 6.21:
9527                  *
9528                  * "The execution of a REPORT LUNS command to any valid and
9529                  * installed logical unit shall clear the REPORTED LUNS DATA
9530                  * HAS CHANGED unit attention condition for all logical
9531                  * units of that target with respect to the requesting
9532                  * initiator. A valid and installed logical unit is one
9533                  * having a PERIPHERAL QUALIFIER of 000b in the standard
9534                  * INQUIRY data (see 6.4.2)."
9535                  *
9536                  * If request_lun is NULL, the LUN this report luns command
9537                  * was issued to is either disabled or doesn't exist. In that
9538                  * case, we shouldn't clear any pending lun change unit
9539                  * attention.
9540                  */
9541                 if (request_lun != NULL) {
9542                         mtx_lock(&lun->lun_lock);
9543                         lun->pending_ua[initidx] &= ~CTL_UA_LUN_CHANGE;
9544                         mtx_unlock(&lun->lun_lock);
9545                 }
9546         }
9547         mtx_unlock(&control_softc->ctl_lock);
9548
9549         /*
9550          * It's quite possible that we've returned fewer LUNs than we allocated
9551          * space for.  Trim it.
9552          */
9553         lun_datalen = sizeof(*lun_data) +
9554                 (num_filled * sizeof(struct scsi_report_luns_lundata));
9555
9556         if (lun_datalen < alloc_len) {
9557                 ctsio->residual = alloc_len - lun_datalen;
9558                 ctsio->kern_data_len = lun_datalen;
9559                 ctsio->kern_total_len = lun_datalen;
9560         } else {
9561                 ctsio->residual = 0;
9562                 ctsio->kern_data_len = alloc_len;
9563                 ctsio->kern_total_len = alloc_len;
9564         }
9565         ctsio->kern_data_resid = 0;
9566         ctsio->kern_rel_offset = 0;
9567         ctsio->kern_sg_entries = 0;
9568
9569         /*
9570          * We set this to the actual data length, regardless of how much
9571          * space we actually have to return results.  If the user looks at
9572          * this value, he'll know whether or not he allocated enough space
9573          * and reissue the command if necessary.  We don't support well
9574          * known logical units, so if the user asks for that, return none.
9575          */
9576         scsi_ulto4b(lun_datalen - 8, lun_data->length);
9577
9578         /*
9579          * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9580          * this request.
9581          */
9582         ctsio->scsi_status = SCSI_STATUS_OK;
9583
9584         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9585         ctsio->be_move_done = ctl_config_move_done;
9586         ctl_datamove((union ctl_io *)ctsio);
9587
9588         return (retval);
9589 }
9590
9591 int
9592 ctl_request_sense(struct ctl_scsiio *ctsio)
9593 {
9594         struct scsi_request_sense *cdb;
9595         struct scsi_sense_data *sense_ptr;
9596         struct ctl_lun *lun;
9597         uint32_t initidx;
9598         int have_error;
9599         scsi_sense_data_type sense_format;
9600
9601         cdb = (struct scsi_request_sense *)ctsio->cdb;
9602
9603         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9604
9605         CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9606
9607         /*
9608          * Determine which sense format the user wants.
9609          */
9610         if (cdb->byte2 & SRS_DESC)
9611                 sense_format = SSD_TYPE_DESC;
9612         else
9613                 sense_format = SSD_TYPE_FIXED;
9614
9615         ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9616         sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9617         ctsio->kern_sg_entries = 0;
9618
9619         /*
9620          * struct scsi_sense_data, which is currently set to 256 bytes, is
9621          * larger than the largest allowed value for the length field in the
9622          * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9623          */
9624         ctsio->residual = 0;
9625         ctsio->kern_data_len = cdb->length;
9626         ctsio->kern_total_len = cdb->length;
9627
9628         ctsio->kern_data_resid = 0;
9629         ctsio->kern_rel_offset = 0;
9630         ctsio->kern_sg_entries = 0;
9631
9632         /*
9633          * If we don't have a LUN, we don't have any pending sense.
9634          */
9635         if (lun == NULL)
9636                 goto no_sense;
9637
9638         have_error = 0;
9639         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9640         /*
9641          * Check for pending sense, and then for pending unit attentions.
9642          * Pending sense gets returned first, then pending unit attentions.
9643          */
9644         mtx_lock(&lun->lun_lock);
9645 #ifdef CTL_WITH_CA
9646         if (ctl_is_set(lun->have_ca, initidx)) {
9647                 scsi_sense_data_type stored_format;
9648
9649                 /*
9650                  * Check to see which sense format was used for the stored
9651                  * sense data.
9652                  */
9653                 stored_format = scsi_sense_type(&lun->pending_sense[initidx]);
9654
9655                 /*
9656                  * If the user requested a different sense format than the
9657                  * one we stored, then we need to convert it to the other
9658                  * format.  If we're going from descriptor to fixed format
9659                  * sense data, we may lose things in translation, depending
9660                  * on what options were used.
9661                  *
9662                  * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9663                  * for some reason we'll just copy it out as-is.
9664                  */
9665                 if ((stored_format == SSD_TYPE_FIXED)
9666                  && (sense_format == SSD_TYPE_DESC))
9667                         ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9668                             &lun->pending_sense[initidx],
9669                             (struct scsi_sense_data_desc *)sense_ptr);
9670                 else if ((stored_format == SSD_TYPE_DESC)
9671                       && (sense_format == SSD_TYPE_FIXED))
9672                         ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9673                             &lun->pending_sense[initidx],
9674                             (struct scsi_sense_data_fixed *)sense_ptr);
9675                 else
9676                         memcpy(sense_ptr, &lun->pending_sense[initidx],
9677                                ctl_min(sizeof(*sense_ptr),
9678                                sizeof(lun->pending_sense[initidx])));
9679
9680                 ctl_clear_mask(lun->have_ca, initidx);
9681                 have_error = 1;
9682         } else
9683 #endif
9684         if (lun->pending_ua[initidx] != CTL_UA_NONE) {
9685                 ctl_ua_type ua_type;
9686
9687                 ua_type = ctl_build_ua(lun->pending_ua[initidx],
9688                                        sense_ptr, sense_format);
9689                 if (ua_type != CTL_UA_NONE) {
9690                         have_error = 1;
9691                         /* We're reporting this UA, so clear it */
9692                         lun->pending_ua[initidx] &= ~ua_type;
9693                 }
9694         }
9695         mtx_unlock(&lun->lun_lock);
9696
9697         /*
9698          * We already have a pending error, return it.
9699          */
9700         if (have_error != 0) {
9701                 /*
9702                  * We report the SCSI status as OK, since the status of the
9703                  * request sense command itself is OK.
9704                  */
9705                 ctsio->scsi_status = SCSI_STATUS_OK;
9706
9707                 /*
9708                  * We report 0 for the sense length, because we aren't doing
9709                  * autosense in this case.  We're reporting sense as
9710                  * parameter data.
9711                  */
9712                 ctsio->sense_len = 0;
9713                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9714                 ctsio->be_move_done = ctl_config_move_done;
9715                 ctl_datamove((union ctl_io *)ctsio);
9716
9717                 return (CTL_RETVAL_COMPLETE);
9718         }
9719
9720 no_sense:
9721
9722         /*
9723          * No sense information to report, so we report that everything is
9724          * okay.
9725          */
9726         ctl_set_sense_data(sense_ptr,
9727                            lun,
9728                            sense_format,
9729                            /*current_error*/ 1,
9730                            /*sense_key*/ SSD_KEY_NO_SENSE,
9731                            /*asc*/ 0x00,
9732                            /*ascq*/ 0x00,
9733                            SSD_ELEM_NONE);
9734
9735         ctsio->scsi_status = SCSI_STATUS_OK;
9736
9737         /*
9738          * We report 0 for the sense length, because we aren't doing
9739          * autosense in this case.  We're reporting sense as parameter data.
9740          */
9741         ctsio->sense_len = 0;
9742         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9743         ctsio->be_move_done = ctl_config_move_done;
9744         ctl_datamove((union ctl_io *)ctsio);
9745
9746         return (CTL_RETVAL_COMPLETE);
9747 }
9748
9749 int
9750 ctl_tur(struct ctl_scsiio *ctsio)
9751 {
9752         struct ctl_lun *lun;
9753
9754         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9755
9756         CTL_DEBUG_PRINT(("ctl_tur\n"));
9757
9758         if (lun == NULL)
9759                 return (EINVAL);
9760
9761         ctsio->scsi_status = SCSI_STATUS_OK;
9762         ctsio->io_hdr.status = CTL_SUCCESS;
9763
9764         ctl_done((union ctl_io *)ctsio);
9765
9766         return (CTL_RETVAL_COMPLETE);
9767 }
9768
9769 #ifdef notyet
9770 static int
9771 ctl_cmddt_inquiry(struct ctl_scsiio *ctsio)
9772 {
9773
9774 }
9775 #endif
9776
9777 static int
9778 ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9779 {
9780         struct scsi_vpd_supported_pages *pages;
9781         int sup_page_size;
9782         struct ctl_lun *lun;
9783
9784         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9785
9786         sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9787             SCSI_EVPD_NUM_SUPPORTED_PAGES;
9788         ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9789         pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9790         ctsio->kern_sg_entries = 0;
9791
9792         if (sup_page_size < alloc_len) {
9793                 ctsio->residual = alloc_len - sup_page_size;
9794                 ctsio->kern_data_len = sup_page_size;
9795                 ctsio->kern_total_len = sup_page_size;
9796         } else {
9797                 ctsio->residual = 0;
9798                 ctsio->kern_data_len = alloc_len;
9799                 ctsio->kern_total_len = alloc_len;
9800         }
9801         ctsio->kern_data_resid = 0;
9802         ctsio->kern_rel_offset = 0;
9803         ctsio->kern_sg_entries = 0;
9804
9805         /*
9806          * The control device is always connected.  The disk device, on the
9807          * other hand, may not be online all the time.  Need to change this
9808          * to figure out whether the disk device is actually online or not.
9809          */
9810         if (lun != NULL)
9811                 pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9812                                 lun->be_lun->lun_type;
9813         else
9814                 pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9815
9816         pages->length = SCSI_EVPD_NUM_SUPPORTED_PAGES;
9817         /* Supported VPD pages */
9818         pages->page_list[0] = SVPD_SUPPORTED_PAGES;
9819         /* Serial Number */
9820         pages->page_list[1] = SVPD_UNIT_SERIAL_NUMBER;
9821         /* Device Identification */
9822         pages->page_list[2] = SVPD_DEVICE_ID;
9823         /* SCSI Ports */
9824         pages->page_list[3] = SVPD_SCSI_PORTS;
9825         /* Third-party Copy */
9826         pages->page_list[4] = SVPD_SCSI_TPC;
9827         /* Block limits */
9828         pages->page_list[5] = SVPD_BLOCK_LIMITS;
9829         /* Block Device Characteristics */
9830         pages->page_list[6] = SVPD_BDC;
9831         /* Logical Block Provisioning */
9832         pages->page_list[7] = SVPD_LBP;
9833
9834         ctsio->scsi_status = SCSI_STATUS_OK;
9835
9836         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9837         ctsio->be_move_done = ctl_config_move_done;
9838         ctl_datamove((union ctl_io *)ctsio);
9839
9840         return (CTL_RETVAL_COMPLETE);
9841 }
9842
9843 static int
9844 ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9845 {
9846         struct scsi_vpd_unit_serial_number *sn_ptr;
9847         struct ctl_lun *lun;
9848
9849         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9850
9851         ctsio->kern_data_ptr = malloc(sizeof(*sn_ptr), M_CTL, M_WAITOK | M_ZERO);
9852         sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9853         ctsio->kern_sg_entries = 0;
9854
9855         if (sizeof(*sn_ptr) < alloc_len) {
9856                 ctsio->residual = alloc_len - sizeof(*sn_ptr);
9857                 ctsio->kern_data_len = sizeof(*sn_ptr);
9858                 ctsio->kern_total_len = sizeof(*sn_ptr);
9859         } else {
9860                 ctsio->residual = 0;
9861                 ctsio->kern_data_len = alloc_len;
9862                 ctsio->kern_total_len = alloc_len;
9863         }
9864         ctsio->kern_data_resid = 0;
9865         ctsio->kern_rel_offset = 0;
9866         ctsio->kern_sg_entries = 0;
9867
9868         /*
9869          * The control device is always connected.  The disk device, on the
9870          * other hand, may not be online all the time.  Need to change this
9871          * to figure out whether the disk device is actually online or not.
9872          */
9873         if (lun != NULL)
9874                 sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9875                                   lun->be_lun->lun_type;
9876         else
9877                 sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9878
9879         sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9880         sn_ptr->length = ctl_min(sizeof(*sn_ptr) - 4, CTL_SN_LEN);
9881         /*
9882          * If we don't have a LUN, we just leave the serial number as
9883          * all spaces.
9884          */
9885         memset(sn_ptr->serial_num, 0x20, sizeof(sn_ptr->serial_num));
9886         if (lun != NULL) {
9887                 strncpy((char *)sn_ptr->serial_num,
9888                         (char *)lun->be_lun->serial_num, CTL_SN_LEN);
9889         }
9890         ctsio->scsi_status = SCSI_STATUS_OK;
9891
9892         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9893         ctsio->be_move_done = ctl_config_move_done;
9894         ctl_datamove((union ctl_io *)ctsio);
9895
9896         return (CTL_RETVAL_COMPLETE);
9897 }
9898
9899
9900 static int
9901 ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9902 {
9903         struct scsi_vpd_device_id *devid_ptr;
9904         struct scsi_vpd_id_descriptor *desc;
9905         struct ctl_softc *ctl_softc;
9906         struct ctl_lun *lun;
9907         struct ctl_port *port;
9908         int data_len;
9909         uint8_t proto;
9910
9911         ctl_softc = control_softc;
9912
9913         port = ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)];
9914         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9915
9916         data_len = sizeof(struct scsi_vpd_device_id) +
9917             sizeof(struct scsi_vpd_id_descriptor) +
9918                 sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9919             sizeof(struct scsi_vpd_id_descriptor) +
9920                 sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9921         if (lun && lun->lun_devid)
9922                 data_len += lun->lun_devid->len;
9923         if (port->port_devid)
9924                 data_len += port->port_devid->len;
9925         if (port->target_devid)
9926                 data_len += port->target_devid->len;
9927
9928         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9929         devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9930         ctsio->kern_sg_entries = 0;
9931
9932         if (data_len < alloc_len) {
9933                 ctsio->residual = alloc_len - data_len;
9934                 ctsio->kern_data_len = data_len;
9935                 ctsio->kern_total_len = data_len;
9936         } else {
9937                 ctsio->residual = 0;
9938                 ctsio->kern_data_len = alloc_len;
9939                 ctsio->kern_total_len = alloc_len;
9940         }
9941         ctsio->kern_data_resid = 0;
9942         ctsio->kern_rel_offset = 0;
9943         ctsio->kern_sg_entries = 0;
9944
9945         /*
9946          * The control device is always connected.  The disk device, on the
9947          * other hand, may not be online all the time.
9948          */
9949         if (lun != NULL)
9950                 devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9951                                      lun->be_lun->lun_type;
9952         else
9953                 devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9954         devid_ptr->page_code = SVPD_DEVICE_ID;
9955         scsi_ulto2b(data_len - 4, devid_ptr->length);
9956
9957         if (port->port_type == CTL_PORT_FC)
9958                 proto = SCSI_PROTO_FC << 4;
9959         else if (port->port_type == CTL_PORT_ISCSI)
9960                 proto = SCSI_PROTO_ISCSI << 4;
9961         else
9962                 proto = SCSI_PROTO_SPI << 4;
9963         desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9964
9965         /*
9966          * We're using a LUN association here.  i.e., this device ID is a
9967          * per-LUN identifier.
9968          */
9969         if (lun && lun->lun_devid) {
9970                 memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
9971                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9972                     lun->lun_devid->len);
9973         }
9974
9975         /*
9976          * This is for the WWPN which is a port association.
9977          */
9978         if (port->port_devid) {
9979                 memcpy(desc, port->port_devid->data, port->port_devid->len);
9980                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9981                     port->port_devid->len);
9982         }
9983
9984         /*
9985          * This is for the Relative Target Port(type 4h) identifier
9986          */
9987         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9988         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9989             SVPD_ID_TYPE_RELTARG;
9990         desc->length = 4;
9991         scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
9992         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9993             sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9994
9995         /*
9996          * This is for the Target Port Group(type 5h) identifier
9997          */
9998         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9999         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
10000             SVPD_ID_TYPE_TPORTGRP;
10001         desc->length = 4;
10002         scsi_ulto2b(ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS + 1,
10003             &desc->identifier[2]);
10004         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
10005             sizeof(struct scsi_vpd_id_trgt_port_grp_id));
10006
10007         /*
10008          * This is for the Target identifier
10009          */
10010         if (port->target_devid) {
10011                 memcpy(desc, port->target_devid->data, port->target_devid->len);
10012         }
10013
10014         ctsio->scsi_status = SCSI_STATUS_OK;
10015         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10016         ctsio->be_move_done = ctl_config_move_done;
10017         ctl_datamove((union ctl_io *)ctsio);
10018
10019         return (CTL_RETVAL_COMPLETE);
10020 }
10021
10022 static int
10023 ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
10024 {
10025         struct ctl_softc *softc = control_softc;
10026         struct scsi_vpd_scsi_ports *sp;
10027         struct scsi_vpd_port_designation *pd;
10028         struct scsi_vpd_port_designation_cont *pdc;
10029         struct ctl_lun *lun;
10030         struct ctl_port *port;
10031         int data_len, num_target_ports, iid_len, id_len, g, pg, p;
10032         int num_target_port_groups, single;
10033
10034         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10035
10036         single = ctl_is_single;
10037         if (single)
10038                 num_target_port_groups = 1;
10039         else
10040                 num_target_port_groups = NUM_TARGET_PORT_GROUPS;
10041         num_target_ports = 0;
10042         iid_len = 0;
10043         id_len = 0;
10044         mtx_lock(&softc->ctl_lock);
10045         STAILQ_FOREACH(port, &softc->port_list, links) {
10046                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
10047                         continue;
10048                 if (lun != NULL &&
10049                     ctl_map_lun_back(port->targ_port, lun->lun) >=
10050                     CTL_MAX_LUNS)
10051                         continue;
10052                 num_target_ports++;
10053                 if (port->init_devid)
10054                         iid_len += port->init_devid->len;
10055                 if (port->port_devid)
10056                         id_len += port->port_devid->len;
10057         }
10058         mtx_unlock(&softc->ctl_lock);
10059
10060         data_len = sizeof(struct scsi_vpd_scsi_ports) + num_target_port_groups *
10061             num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
10062              sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
10063         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10064         sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
10065         ctsio->kern_sg_entries = 0;
10066
10067         if (data_len < alloc_len) {
10068                 ctsio->residual = alloc_len - data_len;
10069                 ctsio->kern_data_len = data_len;
10070                 ctsio->kern_total_len = data_len;
10071         } else {
10072                 ctsio->residual = 0;
10073                 ctsio->kern_data_len = alloc_len;
10074                 ctsio->kern_total_len = alloc_len;
10075         }
10076         ctsio->kern_data_resid = 0;
10077         ctsio->kern_rel_offset = 0;
10078         ctsio->kern_sg_entries = 0;
10079
10080         /*
10081          * The control device is always connected.  The disk device, on the
10082          * other hand, may not be online all the time.  Need to change this
10083          * to figure out whether the disk device is actually online or not.
10084          */
10085         if (lun != NULL)
10086                 sp->device = (SID_QUAL_LU_CONNECTED << 5) |
10087                                   lun->be_lun->lun_type;
10088         else
10089                 sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10090
10091         sp->page_code = SVPD_SCSI_PORTS;
10092         scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
10093             sp->page_length);
10094         pd = &sp->design[0];
10095
10096         mtx_lock(&softc->ctl_lock);
10097         if (softc->flags & CTL_FLAG_MASTER_SHELF)
10098                 pg = 0;
10099         else
10100                 pg = 1;
10101         for (g = 0; g < num_target_port_groups; g++) {
10102                 STAILQ_FOREACH(port, &softc->port_list, links) {
10103                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
10104                                 continue;
10105                         if (lun != NULL &&
10106                             ctl_map_lun_back(port->targ_port, lun->lun) >=
10107                             CTL_MAX_LUNS)
10108                                 continue;
10109                         p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
10110                         scsi_ulto2b(p, pd->relative_port_id);
10111                         if (port->init_devid && g == pg) {
10112                                 iid_len = port->init_devid->len;
10113                                 memcpy(pd->initiator_transportid,
10114                                     port->init_devid->data, port->init_devid->len);
10115                         } else
10116                                 iid_len = 0;
10117                         scsi_ulto2b(iid_len, pd->initiator_transportid_length);
10118                         pdc = (struct scsi_vpd_port_designation_cont *)
10119                             (&pd->initiator_transportid[iid_len]);
10120                         if (port->port_devid && g == pg) {
10121                                 id_len = port->port_devid->len;
10122                                 memcpy(pdc->target_port_descriptors,
10123                                     port->port_devid->data, port->port_devid->len);
10124                         } else
10125                                 id_len = 0;
10126                         scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
10127                         pd = (struct scsi_vpd_port_designation *)
10128                             ((uint8_t *)pdc->target_port_descriptors + id_len);
10129                 }
10130         }
10131         mtx_unlock(&softc->ctl_lock);
10132
10133         ctsio->scsi_status = SCSI_STATUS_OK;
10134         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10135         ctsio->be_move_done = ctl_config_move_done;
10136         ctl_datamove((union ctl_io *)ctsio);
10137
10138         return (CTL_RETVAL_COMPLETE);
10139 }
10140
10141 static int
10142 ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
10143 {
10144         struct scsi_vpd_block_limits *bl_ptr;
10145         struct ctl_lun *lun;
10146         int bs;
10147
10148         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10149
10150         ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
10151         bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
10152         ctsio->kern_sg_entries = 0;
10153
10154         if (sizeof(*bl_ptr) < alloc_len) {
10155                 ctsio->residual = alloc_len - sizeof(*bl_ptr);
10156                 ctsio->kern_data_len = sizeof(*bl_ptr);
10157                 ctsio->kern_total_len = sizeof(*bl_ptr);
10158         } else {
10159                 ctsio->residual = 0;
10160                 ctsio->kern_data_len = alloc_len;
10161                 ctsio->kern_total_len = alloc_len;
10162         }
10163         ctsio->kern_data_resid = 0;
10164         ctsio->kern_rel_offset = 0;
10165         ctsio->kern_sg_entries = 0;
10166
10167         /*
10168          * The control device is always connected.  The disk device, on the
10169          * other hand, may not be online all the time.  Need to change this
10170          * to figure out whether the disk device is actually online or not.
10171          */
10172         if (lun != NULL)
10173                 bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10174                                   lun->be_lun->lun_type;
10175         else
10176                 bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10177
10178         bl_ptr->page_code = SVPD_BLOCK_LIMITS;
10179         scsi_ulto2b(sizeof(*bl_ptr), bl_ptr->page_length);
10180         bl_ptr->max_cmp_write_len = 0xff;
10181         scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
10182         if (lun != NULL) {
10183                 bs = lun->be_lun->blocksize;
10184                 scsi_ulto4b(MAXPHYS / bs, bl_ptr->opt_txfer_len);
10185                 if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10186                         scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_lba_cnt);
10187                         scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_blk_cnt);
10188                         if (lun->be_lun->pblockexp != 0) {
10189                                 scsi_ulto4b((1 << lun->be_lun->pblockexp),
10190                                     bl_ptr->opt_unmap_grain);
10191                                 scsi_ulto4b(0x80000000 | lun->be_lun->pblockoff,
10192                                     bl_ptr->unmap_grain_align);
10193                         }
10194                 }
10195         }
10196         scsi_u64to8b(UINT64_MAX, bl_ptr->max_write_same_length);
10197
10198         ctsio->scsi_status = SCSI_STATUS_OK;
10199         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10200         ctsio->be_move_done = ctl_config_move_done;
10201         ctl_datamove((union ctl_io *)ctsio);
10202
10203         return (CTL_RETVAL_COMPLETE);
10204 }
10205
10206 static int
10207 ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len)
10208 {
10209         struct scsi_vpd_block_device_characteristics *bdc_ptr;
10210         struct ctl_lun *lun;
10211
10212         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10213
10214         ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO);
10215         bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr;
10216         ctsio->kern_sg_entries = 0;
10217
10218         if (sizeof(*bdc_ptr) < alloc_len) {
10219                 ctsio->residual = alloc_len - sizeof(*bdc_ptr);
10220                 ctsio->kern_data_len = sizeof(*bdc_ptr);
10221                 ctsio->kern_total_len = sizeof(*bdc_ptr);
10222         } else {
10223                 ctsio->residual = 0;
10224                 ctsio->kern_data_len = alloc_len;
10225                 ctsio->kern_total_len = alloc_len;
10226         }
10227         ctsio->kern_data_resid = 0;
10228         ctsio->kern_rel_offset = 0;
10229         ctsio->kern_sg_entries = 0;
10230
10231         /*
10232          * The control device is always connected.  The disk device, on the
10233          * other hand, may not be online all the time.  Need to change this
10234          * to figure out whether the disk device is actually online or not.
10235          */
10236         if (lun != NULL)
10237                 bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10238                                   lun->be_lun->lun_type;
10239         else
10240                 bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10241         bdc_ptr->page_code = SVPD_BDC;
10242         scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length);
10243         scsi_ulto2b(SVPD_NON_ROTATING, bdc_ptr->medium_rotation_rate);
10244         bdc_ptr->flags = SVPD_FUAB | SVPD_VBULS;
10245
10246         ctsio->scsi_status = SCSI_STATUS_OK;
10247         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10248         ctsio->be_move_done = ctl_config_move_done;
10249         ctl_datamove((union ctl_io *)ctsio);
10250
10251         return (CTL_RETVAL_COMPLETE);
10252 }
10253
10254 static int
10255 ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
10256 {
10257         struct scsi_vpd_logical_block_prov *lbp_ptr;
10258         struct ctl_lun *lun;
10259
10260         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10261
10262         ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
10263         lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
10264         ctsio->kern_sg_entries = 0;
10265
10266         if (sizeof(*lbp_ptr) < alloc_len) {
10267                 ctsio->residual = alloc_len - sizeof(*lbp_ptr);
10268                 ctsio->kern_data_len = sizeof(*lbp_ptr);
10269                 ctsio->kern_total_len = sizeof(*lbp_ptr);
10270         } else {
10271                 ctsio->residual = 0;
10272                 ctsio->kern_data_len = alloc_len;
10273                 ctsio->kern_total_len = alloc_len;
10274         }
10275         ctsio->kern_data_resid = 0;
10276         ctsio->kern_rel_offset = 0;
10277         ctsio->kern_sg_entries = 0;
10278
10279         /*
10280          * The control device is always connected.  The disk device, on the
10281          * other hand, may not be online all the time.  Need to change this
10282          * to figure out whether the disk device is actually online or not.
10283          */
10284         if (lun != NULL)
10285                 lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10286                                   lun->be_lun->lun_type;
10287         else
10288                 lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10289
10290         lbp_ptr->page_code = SVPD_LBP;
10291         scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length);
10292         if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10293                 lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 |
10294                     SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP;
10295                 lbp_ptr->prov_type = SVPD_LBP_RESOURCE;
10296         }
10297
10298         ctsio->scsi_status = SCSI_STATUS_OK;
10299         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10300         ctsio->be_move_done = ctl_config_move_done;
10301         ctl_datamove((union ctl_io *)ctsio);
10302
10303         return (CTL_RETVAL_COMPLETE);
10304 }
10305
10306 static int
10307 ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
10308 {
10309         struct scsi_inquiry *cdb;
10310         struct ctl_lun *lun;
10311         int alloc_len, retval;
10312
10313         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10314         cdb = (struct scsi_inquiry *)ctsio->cdb;
10315
10316         retval = CTL_RETVAL_COMPLETE;
10317
10318         alloc_len = scsi_2btoul(cdb->length);
10319
10320         switch (cdb->page_code) {
10321         case SVPD_SUPPORTED_PAGES:
10322                 retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
10323                 break;
10324         case SVPD_UNIT_SERIAL_NUMBER:
10325                 retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
10326                 break;
10327         case SVPD_DEVICE_ID:
10328                 retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
10329                 break;
10330         case SVPD_SCSI_PORTS:
10331                 retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
10332                 break;
10333         case SVPD_SCSI_TPC:
10334                 retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
10335                 break;
10336         case SVPD_BLOCK_LIMITS:
10337                 retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
10338                 break;
10339         case SVPD_BDC:
10340                 retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len);
10341                 break;
10342         case SVPD_LBP:
10343                 retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
10344                 break;
10345         default:
10346                 ctl_set_invalid_field(ctsio,
10347                                       /*sks_valid*/ 1,
10348                                       /*command*/ 1,
10349                                       /*field*/ 2,
10350                                       /*bit_valid*/ 0,
10351                                       /*bit*/ 0);
10352                 ctl_done((union ctl_io *)ctsio);
10353                 retval = CTL_RETVAL_COMPLETE;
10354                 break;
10355         }
10356
10357         return (retval);
10358 }
10359
10360 static int
10361 ctl_inquiry_std(struct ctl_scsiio *ctsio)
10362 {
10363         struct scsi_inquiry_data *inq_ptr;
10364         struct scsi_inquiry *cdb;
10365         struct ctl_softc *ctl_softc;
10366         struct ctl_lun *lun;
10367         char *val;
10368         uint32_t alloc_len;
10369         ctl_port_type port_type;
10370
10371         ctl_softc = control_softc;
10372
10373         /*
10374          * Figure out whether we're talking to a Fibre Channel port or not.
10375          * We treat the ioctl front end, and any SCSI adapters, as packetized
10376          * SCSI front ends.
10377          */
10378         port_type = ctl_softc->ctl_ports[
10379             ctl_port_idx(ctsio->io_hdr.nexus.targ_port)]->port_type;
10380         if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
10381                 port_type = CTL_PORT_SCSI;
10382
10383         lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10384         cdb = (struct scsi_inquiry *)ctsio->cdb;
10385         alloc_len = scsi_2btoul(cdb->length);
10386
10387         /*
10388          * We malloc the full inquiry data size here and fill it
10389          * in.  If the user only asks for less, we'll give him
10390          * that much.
10391          */
10392         ctsio->kern_data_ptr = malloc(sizeof(*inq_ptr), M_CTL, M_WAITOK | M_ZERO);
10393         inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
10394         ctsio->kern_sg_entries = 0;
10395         ctsio->kern_data_resid = 0;
10396         ctsio->kern_rel_offset = 0;
10397
10398         if (sizeof(*inq_ptr) < alloc_len) {
10399                 ctsio->residual = alloc_len - sizeof(*inq_ptr);
10400                 ctsio->kern_data_len = sizeof(*inq_ptr);
10401                 ctsio->kern_total_len = sizeof(*inq_ptr);
10402         } else {
10403                 ctsio->residual = 0;
10404                 ctsio->kern_data_len = alloc_len;
10405                 ctsio->kern_total_len = alloc_len;
10406         }
10407
10408         /*
10409          * If we have a LUN configured, report it as connected.  Otherwise,
10410          * report that it is offline or no device is supported, depending 
10411          * on the value of inquiry_pq_no_lun.
10412          *
10413          * According to the spec (SPC-4 r34), the peripheral qualifier
10414          * SID_QUAL_LU_OFFLINE (001b) is used in the following scenario:
10415          *
10416          * "A peripheral device having the specified peripheral device type 
10417          * is not connected to this logical unit. However, the device
10418          * server is capable of supporting the specified peripheral device
10419          * type on this logical unit."
10420          *
10421          * According to the same spec, the peripheral qualifier
10422          * SID_QUAL_BAD_LU (011b) is used in this scenario:
10423          *
10424          * "The device server is not capable of supporting a peripheral
10425          * device on this logical unit. For this peripheral qualifier the
10426          * peripheral device type shall be set to 1Fh. All other peripheral
10427          * device type values are reserved for this peripheral qualifier."
10428          *
10429          * Given the text, it would seem that we probably want to report that
10430          * the LUN is offline here.  There is no LUN connected, but we can
10431          * support a LUN at the given LUN number.
10432          *
10433          * In the real world, though, it sounds like things are a little
10434          * different:
10435          *
10436          * - Linux, when presented with a LUN with the offline peripheral
10437          *   qualifier, will create an sg driver instance for it.  So when
10438          *   you attach it to CTL, you wind up with a ton of sg driver
10439          *   instances.  (One for every LUN that Linux bothered to probe.)
10440          *   Linux does this despite the fact that it issues a REPORT LUNs
10441          *   to LUN 0 to get the inventory of supported LUNs.
10442          *
10443          * - There is other anecdotal evidence (from Emulex folks) about
10444          *   arrays that use the offline peripheral qualifier for LUNs that
10445          *   are on the "passive" path in an active/passive array.
10446          *
10447          * So the solution is provide a hopefully reasonable default
10448          * (return bad/no LUN) and allow the user to change the behavior
10449          * with a tunable/sysctl variable.
10450          */
10451         if (lun != NULL)
10452                 inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10453                                   lun->be_lun->lun_type;
10454         else if (ctl_softc->inquiry_pq_no_lun == 0)
10455                 inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10456         else
10457                 inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10458
10459         /* RMB in byte 2 is 0 */
10460         inq_ptr->version = SCSI_REV_SPC4;
10461
10462         /*
10463          * According to SAM-3, even if a device only supports a single
10464          * level of LUN addressing, it should still set the HISUP bit:
10465          *
10466          * 4.9.1 Logical unit numbers overview
10467          *
10468          * All logical unit number formats described in this standard are
10469          * hierarchical in structure even when only a single level in that
10470          * hierarchy is used. The HISUP bit shall be set to one in the
10471          * standard INQUIRY data (see SPC-2) when any logical unit number
10472          * format described in this standard is used.  Non-hierarchical
10473          * formats are outside the scope of this standard.
10474          *
10475          * Therefore we set the HiSup bit here.
10476          *
10477          * The reponse format is 2, per SPC-3.
10478          */
10479         inq_ptr->response_format = SID_HiSup | 2;
10480
10481         inq_ptr->additional_length =
10482             offsetof(struct scsi_inquiry_data, vendor_specific1) -
10483             (offsetof(struct scsi_inquiry_data, additional_length) + 1);
10484         CTL_DEBUG_PRINT(("additional_length = %d\n",
10485                          inq_ptr->additional_length));
10486
10487         inq_ptr->spc3_flags = SPC3_SID_3PC;
10488         if (!ctl_is_single)
10489                 inq_ptr->spc3_flags |= SPC3_SID_TPGS_IMPLICIT;
10490         /* 16 bit addressing */
10491         if (port_type == CTL_PORT_SCSI)
10492                 inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10493         /* XXX set the SID_MultiP bit here if we're actually going to
10494            respond on multiple ports */
10495         inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10496
10497         /* 16 bit data bus, synchronous transfers */
10498         if (port_type == CTL_PORT_SCSI)
10499                 inq_ptr->flags = SID_WBus16 | SID_Sync;
10500         /*
10501          * XXX KDM do we want to support tagged queueing on the control
10502          * device at all?
10503          */
10504         if ((lun == NULL)
10505          || (lun->be_lun->lun_type != T_PROCESSOR))
10506                 inq_ptr->flags |= SID_CmdQue;
10507         /*
10508          * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10509          * We have 8 bytes for the vendor name, and 16 bytes for the device
10510          * name and 4 bytes for the revision.
10511          */
10512         if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10513             "vendor")) == NULL) {
10514                 strcpy(inq_ptr->vendor, CTL_VENDOR);
10515         } else {
10516                 memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10517                 strncpy(inq_ptr->vendor, val,
10518                     min(sizeof(inq_ptr->vendor), strlen(val)));
10519         }
10520         if (lun == NULL) {
10521                 strcpy(inq_ptr->product, CTL_DIRECT_PRODUCT);
10522         } else if ((val = ctl_get_opt(&lun->be_lun->options, "product")) == NULL) {
10523                 switch (lun->be_lun->lun_type) {
10524                 case T_DIRECT:
10525                         strcpy(inq_ptr->product, CTL_DIRECT_PRODUCT);
10526                         break;
10527                 case T_PROCESSOR:
10528                         strcpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT);
10529                         break;
10530                 default:
10531                         strcpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT);
10532                         break;
10533                 }
10534         } else {
10535                 memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10536                 strncpy(inq_ptr->product, val,
10537                     min(sizeof(inq_ptr->product), strlen(val)));
10538         }
10539
10540         /*
10541          * XXX make this a macro somewhere so it automatically gets
10542          * incremented when we make changes.
10543          */
10544         if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10545             "revision")) == NULL) {
10546                 strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10547         } else {
10548                 memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10549                 strncpy(inq_ptr->revision, val,
10550                     min(sizeof(inq_ptr->revision), strlen(val)));
10551         }
10552
10553         /*
10554          * For parallel SCSI, we support double transition and single
10555          * transition clocking.  We also support QAS (Quick Arbitration
10556          * and Selection) and Information Unit transfers on both the
10557          * control and array devices.
10558          */
10559         if (port_type == CTL_PORT_SCSI)
10560                 inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10561                                     SID_SPI_IUS;
10562
10563         /* SAM-5 (no version claimed) */
10564         scsi_ulto2b(0x00A0, inq_ptr->version1);
10565         /* SPC-4 (no version claimed) */
10566         scsi_ulto2b(0x0460, inq_ptr->version2);
10567         if (port_type == CTL_PORT_FC) {
10568                 /* FCP-2 ANSI INCITS.350:2003 */
10569                 scsi_ulto2b(0x0917, inq_ptr->version3);
10570         } else if (port_type == CTL_PORT_SCSI) {
10571                 /* SPI-4 ANSI INCITS.362:200x */
10572                 scsi_ulto2b(0x0B56, inq_ptr->version3);
10573         } else if (port_type == CTL_PORT_ISCSI) {
10574                 /* iSCSI (no version claimed) */
10575                 scsi_ulto2b(0x0960, inq_ptr->version3);
10576         } else if (port_type == CTL_PORT_SAS) {
10577                 /* SAS (no version claimed) */
10578                 scsi_ulto2b(0x0BE0, inq_ptr->version3);
10579         }
10580
10581         if (lun == NULL) {
10582                 /* SBC-3 (no version claimed) */
10583                 scsi_ulto2b(0x04C0, inq_ptr->version4);
10584         } else {
10585                 switch (lun->be_lun->lun_type) {
10586                 case T_DIRECT:
10587                         /* SBC-3 (no version claimed) */
10588                         scsi_ulto2b(0x04C0, inq_ptr->version4);
10589                         break;
10590                 case T_PROCESSOR:
10591                 default:
10592                         break;
10593                 }
10594         }
10595
10596         ctsio->scsi_status = SCSI_STATUS_OK;
10597         if (ctsio->kern_data_len > 0) {
10598                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10599                 ctsio->be_move_done = ctl_config_move_done;
10600                 ctl_datamove((union ctl_io *)ctsio);
10601         } else {
10602                 ctsio->io_hdr.status = CTL_SUCCESS;
10603                 ctl_done((union ctl_io *)ctsio);
10604         }
10605
10606         return (CTL_RETVAL_COMPLETE);
10607 }
10608
10609 int
10610 ctl_inquiry(struct ctl_scsiio *ctsio)
10611 {
10612         struct scsi_inquiry *cdb;
10613         int retval;
10614
10615         cdb = (struct scsi_inquiry *)ctsio->cdb;
10616
10617         retval = 0;
10618
10619         CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10620
10621         /*
10622          * Right now, we don't support the CmdDt inquiry information.
10623          * This would be nice to support in the future.  When we do
10624          * support it, we should change this test so that it checks to make
10625          * sure SI_EVPD and SI_CMDDT aren't both set at the same time.
10626          */
10627 #ifdef notyet
10628         if (((cdb->byte2 & SI_EVPD)
10629          && (cdb->byte2 & SI_CMDDT)))
10630 #endif
10631         if (cdb->byte2 & SI_CMDDT) {
10632                 /*
10633                  * Point to the SI_CMDDT bit.  We might change this
10634                  * when we support SI_CMDDT, but since both bits would be
10635                  * "wrong", this should probably just stay as-is then.
10636                  */
10637                 ctl_set_invalid_field(ctsio,
10638                                       /*sks_valid*/ 1,
10639                                       /*command*/ 1,
10640                                       /*field*/ 1,
10641                                       /*bit_valid*/ 1,
10642                                       /*bit*/ 1);
10643                 ctl_done((union ctl_io *)ctsio);
10644                 return (CTL_RETVAL_COMPLETE);
10645         }
10646         if (cdb->byte2 & SI_EVPD)
10647                 retval = ctl_inquiry_evpd(ctsio);
10648 #ifdef notyet
10649         else if (cdb->byte2 & SI_CMDDT)
10650                 retval = ctl_inquiry_cmddt(ctsio);
10651 #endif
10652         else
10653                 retval = ctl_inquiry_std(ctsio);
10654
10655         return (retval);
10656 }
10657
10658 /*
10659  * For known CDB types, parse the LBA and length.
10660  */
10661 static int
10662 ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint32_t *len)
10663 {
10664         if (io->io_hdr.io_type != CTL_IO_SCSI)
10665                 return (1);
10666
10667         switch (io->scsiio.cdb[0]) {
10668         case COMPARE_AND_WRITE: {
10669                 struct scsi_compare_and_write *cdb;
10670
10671                 cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10672
10673                 *lba = scsi_8btou64(cdb->addr);
10674                 *len = cdb->length;
10675                 break;
10676         }
10677         case READ_6:
10678         case WRITE_6: {
10679                 struct scsi_rw_6 *cdb;
10680
10681                 cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10682
10683                 *lba = scsi_3btoul(cdb->addr);
10684                 /* only 5 bits are valid in the most significant address byte */
10685                 *lba &= 0x1fffff;
10686                 *len = cdb->length;
10687                 break;
10688         }
10689         case READ_10:
10690         case WRITE_10: {
10691                 struct scsi_rw_10 *cdb;
10692
10693                 cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10694
10695                 *lba = scsi_4btoul(cdb->addr);
10696                 *len = scsi_2btoul(cdb->length);
10697                 break;
10698         }
10699         case WRITE_VERIFY_10: {
10700                 struct scsi_write_verify_10 *cdb;
10701
10702                 cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10703
10704                 *lba = scsi_4btoul(cdb->addr);
10705                 *len = scsi_2btoul(cdb->length);
10706                 break;
10707         }
10708         case READ_12:
10709         case WRITE_12: {
10710                 struct scsi_rw_12 *cdb;
10711
10712                 cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10713
10714                 *lba = scsi_4btoul(cdb->addr);
10715                 *len = scsi_4btoul(cdb->length);
10716                 break;
10717         }
10718         case WRITE_VERIFY_12: {
10719                 struct scsi_write_verify_12 *cdb;
10720
10721                 cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10722
10723                 *lba = scsi_4btoul(cdb->addr);
10724                 *len = scsi_4btoul(cdb->length);
10725                 break;
10726         }
10727         case READ_16:
10728         case WRITE_16: {
10729                 struct scsi_rw_16 *cdb;
10730
10731                 cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10732
10733                 *lba = scsi_8btou64(cdb->addr);
10734                 *len = scsi_4btoul(cdb->length);
10735                 break;
10736         }
10737         case WRITE_VERIFY_16: {
10738                 struct scsi_write_verify_16 *cdb;
10739
10740                 cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10741
10742                 
10743                 *lba = scsi_8btou64(cdb->addr);
10744                 *len = scsi_4btoul(cdb->length);
10745                 break;
10746         }
10747         case WRITE_SAME_10: {
10748                 struct scsi_write_same_10 *cdb;
10749
10750                 cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10751
10752                 *lba = scsi_4btoul(cdb->addr);
10753                 *len = scsi_2btoul(cdb->length);
10754                 break;
10755         }
10756         case WRITE_SAME_16: {
10757                 struct scsi_write_same_16 *cdb;
10758
10759                 cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10760
10761                 *lba = scsi_8btou64(cdb->addr);
10762                 *len = scsi_4btoul(cdb->length);
10763                 break;
10764         }
10765         case VERIFY_10: {
10766                 struct scsi_verify_10 *cdb;
10767
10768                 cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10769
10770                 *lba = scsi_4btoul(cdb->addr);
10771                 *len = scsi_2btoul(cdb->length);
10772                 break;
10773         }
10774         case VERIFY_12: {
10775                 struct scsi_verify_12 *cdb;
10776
10777                 cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10778
10779                 *lba = scsi_4btoul(cdb->addr);
10780                 *len = scsi_4btoul(cdb->length);
10781                 break;
10782         }
10783         case VERIFY_16: {
10784                 struct scsi_verify_16 *cdb;
10785
10786                 cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10787
10788                 *lba = scsi_8btou64(cdb->addr);
10789                 *len = scsi_4btoul(cdb->length);
10790                 break;
10791         }
10792         default:
10793                 return (1);
10794                 break; /* NOTREACHED */
10795         }
10796
10797         return (0);
10798 }
10799
10800 static ctl_action
10801 ctl_extent_check_lba(uint64_t lba1, uint32_t len1, uint64_t lba2, uint32_t len2)
10802 {
10803         uint64_t endlba1, endlba2;
10804
10805         endlba1 = lba1 + len1 - 1;
10806         endlba2 = lba2 + len2 - 1;
10807
10808         if ((endlba1 < lba2)
10809          || (endlba2 < lba1))
10810                 return (CTL_ACTION_PASS);
10811         else
10812                 return (CTL_ACTION_BLOCK);
10813 }
10814
10815 static ctl_action
10816 ctl_extent_check(union ctl_io *io1, union ctl_io *io2)
10817 {
10818         uint64_t lba1, lba2;
10819         uint32_t len1, len2;
10820         int retval;
10821
10822         retval = ctl_get_lba_len(io1, &lba1, &len1);
10823         if (retval != 0)
10824                 return (CTL_ACTION_ERROR);
10825
10826         retval = ctl_get_lba_len(io2, &lba2, &len2);
10827         if (retval != 0)
10828                 return (CTL_ACTION_ERROR);
10829
10830         return (ctl_extent_check_lba(lba1, len1, lba2, len2));
10831 }
10832
10833 static ctl_action
10834 ctl_check_for_blockage(union ctl_io *pending_io, union ctl_io *ooa_io)
10835 {
10836         const struct ctl_cmd_entry *pending_entry, *ooa_entry;
10837         ctl_serialize_action *serialize_row;
10838
10839         /*
10840          * The initiator attempted multiple untagged commands at the same
10841          * time.  Can't do that.
10842          */
10843         if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10844          && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10845          && ((pending_io->io_hdr.nexus.targ_port ==
10846               ooa_io->io_hdr.nexus.targ_port)
10847           && (pending_io->io_hdr.nexus.initid.id ==
10848               ooa_io->io_hdr.nexus.initid.id))
10849          && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10850                 return (CTL_ACTION_OVERLAP);
10851
10852         /*
10853          * The initiator attempted to send multiple tagged commands with
10854          * the same ID.  (It's fine if different initiators have the same
10855          * tag ID.)
10856          *
10857          * Even if all of those conditions are true, we don't kill the I/O
10858          * if the command ahead of us has been aborted.  We won't end up
10859          * sending it to the FETD, and it's perfectly legal to resend a
10860          * command with the same tag number as long as the previous
10861          * instance of this tag number has been aborted somehow.
10862          */
10863         if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10864          && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10865          && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
10866          && ((pending_io->io_hdr.nexus.targ_port ==
10867               ooa_io->io_hdr.nexus.targ_port)
10868           && (pending_io->io_hdr.nexus.initid.id ==
10869               ooa_io->io_hdr.nexus.initid.id))
10870          && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10871                 return (CTL_ACTION_OVERLAP_TAG);
10872
10873         /*
10874          * If we get a head of queue tag, SAM-3 says that we should
10875          * immediately execute it.
10876          *
10877          * What happens if this command would normally block for some other
10878          * reason?  e.g. a request sense with a head of queue tag
10879          * immediately after a write.  Normally that would block, but this
10880          * will result in its getting executed immediately...
10881          *
10882          * We currently return "pass" instead of "skip", so we'll end up
10883          * going through the rest of the queue to check for overlapped tags.
10884          *
10885          * XXX KDM check for other types of blockage first??
10886          */
10887         if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10888                 return (CTL_ACTION_PASS);
10889
10890         /*
10891          * Ordered tags have to block until all items ahead of them
10892          * have completed.  If we get called with an ordered tag, we always
10893          * block, if something else is ahead of us in the queue.
10894          */
10895         if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
10896                 return (CTL_ACTION_BLOCK);
10897
10898         /*
10899          * Simple tags get blocked until all head of queue and ordered tags
10900          * ahead of them have completed.  I'm lumping untagged commands in
10901          * with simple tags here.  XXX KDM is that the right thing to do?
10902          */
10903         if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10904           || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
10905          && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10906           || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
10907                 return (CTL_ACTION_BLOCK);
10908
10909         pending_entry = ctl_get_cmd_entry(&pending_io->scsiio);
10910         ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio);
10911
10912         serialize_row = ctl_serialize_table[ooa_entry->seridx];
10913
10914         switch (serialize_row[pending_entry->seridx]) {
10915         case CTL_SER_BLOCK:
10916                 return (CTL_ACTION_BLOCK);
10917                 break; /* NOTREACHED */
10918         case CTL_SER_EXTENT:
10919                 return (ctl_extent_check(pending_io, ooa_io));
10920                 break; /* NOTREACHED */
10921         case CTL_SER_PASS:
10922                 return (CTL_ACTION_PASS);
10923                 break; /* NOTREACHED */
10924         case CTL_SER_SKIP:
10925                 return (CTL_ACTION_SKIP);
10926                 break;
10927         default:
10928                 panic("invalid serialization value %d",
10929                       serialize_row[pending_entry->seridx]);
10930                 break; /* NOTREACHED */
10931         }
10932
10933         return (CTL_ACTION_ERROR);
10934 }
10935
10936 /*
10937  * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
10938  * Assumptions:
10939  * - pending_io is generally either incoming, or on the blocked queue
10940  * - starting I/O is the I/O we want to start the check with.
10941  */
10942 static ctl_action
10943 ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
10944               union ctl_io *starting_io)
10945 {
10946         union ctl_io *ooa_io;
10947         ctl_action action;
10948
10949         mtx_assert(&lun->lun_lock, MA_OWNED);
10950
10951         /*
10952          * Run back along the OOA queue, starting with the current
10953          * blocked I/O and going through every I/O before it on the
10954          * queue.  If starting_io is NULL, we'll just end up returning
10955          * CTL_ACTION_PASS.
10956          */
10957         for (ooa_io = starting_io; ooa_io != NULL;
10958              ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
10959              ooa_links)){
10960
10961                 /*
10962                  * This routine just checks to see whether
10963                  * cur_blocked is blocked by ooa_io, which is ahead
10964                  * of it in the queue.  It doesn't queue/dequeue
10965                  * cur_blocked.
10966                  */
10967                 action = ctl_check_for_blockage(pending_io, ooa_io);
10968                 switch (action) {
10969                 case CTL_ACTION_BLOCK:
10970                 case CTL_ACTION_OVERLAP:
10971                 case CTL_ACTION_OVERLAP_TAG:
10972                 case CTL_ACTION_SKIP:
10973                 case CTL_ACTION_ERROR:
10974                         return (action);
10975                         break; /* NOTREACHED */
10976                 case CTL_ACTION_PASS:
10977                         break;
10978                 default:
10979                         panic("invalid action %d", action);
10980                         break;  /* NOTREACHED */
10981                 }
10982         }
10983
10984         return (CTL_ACTION_PASS);
10985 }
10986
10987 /*
10988  * Assumptions:
10989  * - An I/O has just completed, and has been removed from the per-LUN OOA
10990  *   queue, so some items on the blocked queue may now be unblocked.
10991  */
10992 static int
10993 ctl_check_blocked(struct ctl_lun *lun)
10994 {
10995         union ctl_io *cur_blocked, *next_blocked;
10996
10997         mtx_assert(&lun->lun_lock, MA_OWNED);
10998
10999         /*
11000          * Run forward from the head of the blocked queue, checking each
11001          * entry against the I/Os prior to it on the OOA queue to see if
11002          * there is still any blockage.
11003          *
11004          * We cannot use the TAILQ_FOREACH() macro, because it can't deal
11005          * with our removing a variable on it while it is traversing the
11006          * list.
11007          */
11008         for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
11009              cur_blocked != NULL; cur_blocked = next_blocked) {
11010                 union ctl_io *prev_ooa;
11011                 ctl_action action;
11012
11013                 next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
11014                                                           blocked_links);
11015
11016                 prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
11017                                                       ctl_ooaq, ooa_links);
11018
11019                 /*
11020                  * If cur_blocked happens to be the first item in the OOA
11021                  * queue now, prev_ooa will be NULL, and the action
11022                  * returned will just be CTL_ACTION_PASS.
11023                  */
11024                 action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
11025
11026                 switch (action) {
11027                 case CTL_ACTION_BLOCK:
11028                         /* Nothing to do here, still blocked */
11029                         break;
11030                 case CTL_ACTION_OVERLAP:
11031                 case CTL_ACTION_OVERLAP_TAG:
11032                         /*
11033                          * This shouldn't happen!  In theory we've already
11034                          * checked this command for overlap...
11035                          */
11036                         break;
11037                 case CTL_ACTION_PASS:
11038                 case CTL_ACTION_SKIP: {
11039                         struct ctl_softc *softc;
11040                         const struct ctl_cmd_entry *entry;
11041                         uint32_t initidx;
11042                         int isc_retval;
11043
11044                         /*
11045                          * The skip case shouldn't happen, this transaction
11046                          * should have never made it onto the blocked queue.
11047                          */
11048                         /*
11049                          * This I/O is no longer blocked, we can remove it
11050                          * from the blocked queue.  Since this is a TAILQ
11051                          * (doubly linked list), we can do O(1) removals
11052                          * from any place on the list.
11053                          */
11054                         TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
11055                                      blocked_links);
11056                         cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11057
11058                         if (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC){
11059                                 /*
11060                                  * Need to send IO back to original side to
11061                                  * run
11062                                  */
11063                                 union ctl_ha_msg msg_info;
11064
11065                                 msg_info.hdr.original_sc =
11066                                         cur_blocked->io_hdr.original_sc;
11067                                 msg_info.hdr.serializing_sc = cur_blocked;
11068                                 msg_info.hdr.msg_type = CTL_MSG_R2R;
11069                                 if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11070                                      &msg_info, sizeof(msg_info), 0)) >
11071                                      CTL_HA_STATUS_SUCCESS) {
11072                                         printf("CTL:Check Blocked error from "
11073                                                "ctl_ha_msg_send %d\n",
11074                                                isc_retval);
11075                                 }
11076                                 break;
11077                         }
11078                         entry = ctl_get_cmd_entry(&cur_blocked->scsiio);
11079                         softc = control_softc;
11080
11081                         initidx = ctl_get_initindex(&cur_blocked->io_hdr.nexus);
11082
11083                         /*
11084                          * Check this I/O for LUN state changes that may
11085                          * have happened while this command was blocked.
11086                          * The LUN state may have been changed by a command
11087                          * ahead of us in the queue, so we need to re-check
11088                          * for any states that can be caused by SCSI
11089                          * commands.
11090                          */
11091                         if (ctl_scsiio_lun_check(softc, lun, entry,
11092                                                  &cur_blocked->scsiio) == 0) {
11093                                 cur_blocked->io_hdr.flags |=
11094                                                       CTL_FLAG_IS_WAS_ON_RTR;
11095                                 ctl_enqueue_rtr(cur_blocked);
11096                         } else
11097                                 ctl_done(cur_blocked);
11098                         break;
11099                 }
11100                 default:
11101                         /*
11102                          * This probably shouldn't happen -- we shouldn't
11103                          * get CTL_ACTION_ERROR, or anything else.
11104                          */
11105                         break;
11106                 }
11107         }
11108
11109         return (CTL_RETVAL_COMPLETE);
11110 }
11111
11112 /*
11113  * This routine (with one exception) checks LUN flags that can be set by
11114  * commands ahead of us in the OOA queue.  These flags have to be checked
11115  * when a command initially comes in, and when we pull a command off the
11116  * blocked queue and are preparing to execute it.  The reason we have to
11117  * check these flags for commands on the blocked queue is that the LUN
11118  * state may have been changed by a command ahead of us while we're on the
11119  * blocked queue.
11120  *
11121  * Ordering is somewhat important with these checks, so please pay
11122  * careful attention to the placement of any new checks.
11123  */
11124 static int
11125 ctl_scsiio_lun_check(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
11126     const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
11127 {
11128         int retval;
11129
11130         retval = 0;
11131
11132         mtx_assert(&lun->lun_lock, MA_OWNED);
11133
11134         /*
11135          * If this shelf is a secondary shelf controller, we have to reject
11136          * any media access commands.
11137          */
11138 #if 0
11139         /* No longer needed for HA */
11140         if (((ctl_softc->flags & CTL_FLAG_MASTER_SHELF) == 0)
11141          && ((entry->flags & CTL_CMD_FLAG_OK_ON_SECONDARY) == 0)) {
11142                 ctl_set_lun_standby(ctsio);
11143                 retval = 1;
11144                 goto bailout;
11145         }
11146 #endif
11147
11148         /*
11149          * Check for a reservation conflict.  If this command isn't allowed
11150          * even on reserved LUNs, and if this initiator isn't the one who
11151          * reserved us, reject the command with a reservation conflict.
11152          */
11153         if ((lun->flags & CTL_LUN_RESERVED)
11154          && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
11155                 if ((ctsio->io_hdr.nexus.initid.id != lun->rsv_nexus.initid.id)
11156                  || (ctsio->io_hdr.nexus.targ_port != lun->rsv_nexus.targ_port)
11157                  || (ctsio->io_hdr.nexus.targ_target.id !=
11158                      lun->rsv_nexus.targ_target.id)) {
11159                         ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
11160                         ctsio->io_hdr.status = CTL_SCSI_ERROR;
11161                         retval = 1;
11162                         goto bailout;
11163                 }
11164         }
11165
11166         if ( (lun->flags & CTL_LUN_PR_RESERVED)
11167          && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV) == 0)) {
11168                 uint32_t residx;
11169
11170                 residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
11171                 /*
11172                  * if we aren't registered or it's a res holder type
11173                  * reservation and this isn't the res holder then set a
11174                  * conflict.
11175                  * NOTE: Commands which might be allowed on write exclusive
11176                  * type reservations are checked in the particular command
11177                  * for a conflict. Read and SSU are the only ones.
11178                  */
11179                 if (!lun->per_res[residx].registered
11180                  || (residx != lun->pr_res_idx && lun->res_type < 4)) {
11181                         ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
11182                         ctsio->io_hdr.status = CTL_SCSI_ERROR;
11183                         retval = 1;
11184                         goto bailout;
11185                 }
11186
11187         }
11188
11189         if ((lun->flags & CTL_LUN_OFFLINE)
11190          && ((entry->flags & CTL_CMD_FLAG_OK_ON_OFFLINE) == 0)) {
11191                 ctl_set_lun_not_ready(ctsio);
11192                 retval = 1;
11193                 goto bailout;
11194         }
11195
11196         /*
11197          * If the LUN is stopped, see if this particular command is allowed
11198          * for a stopped lun.  Otherwise, reject it with 0x04,0x02.
11199          */
11200         if ((lun->flags & CTL_LUN_STOPPED)
11201          && ((entry->flags & CTL_CMD_FLAG_OK_ON_STOPPED) == 0)) {
11202                 /* "Logical unit not ready, initializing cmd. required" */
11203                 ctl_set_lun_stopped(ctsio);
11204                 retval = 1;
11205                 goto bailout;
11206         }
11207
11208         if ((lun->flags & CTL_LUN_INOPERABLE)
11209          && ((entry->flags & CTL_CMD_FLAG_OK_ON_INOPERABLE) == 0)) {
11210                 /* "Medium format corrupted" */
11211                 ctl_set_medium_format_corrupted(ctsio);
11212                 retval = 1;
11213                 goto bailout;
11214         }
11215
11216 bailout:
11217         return (retval);
11218
11219 }
11220
11221 static void
11222 ctl_failover_io(union ctl_io *io, int have_lock)
11223 {
11224         ctl_set_busy(&io->scsiio);
11225         ctl_done(io);
11226 }
11227
11228 static void
11229 ctl_failover(void)
11230 {
11231         struct ctl_lun *lun;
11232         struct ctl_softc *ctl_softc;
11233         union ctl_io *next_io, *pending_io;
11234         union ctl_io *io;
11235         int lun_idx;
11236         int i;
11237
11238         ctl_softc = control_softc;
11239
11240         mtx_lock(&ctl_softc->ctl_lock);
11241         /*
11242          * Remove any cmds from the other SC from the rtr queue.  These
11243          * will obviously only be for LUNs for which we're the primary.
11244          * We can't send status or get/send data for these commands.
11245          * Since they haven't been executed yet, we can just remove them.
11246          * We'll either abort them or delete them below, depending on
11247          * which HA mode we're in.
11248          */
11249 #ifdef notyet
11250         mtx_lock(&ctl_softc->queue_lock);
11251         for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->rtr_queue);
11252              io != NULL; io = next_io) {
11253                 next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
11254                 if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
11255                         STAILQ_REMOVE(&ctl_softc->rtr_queue, &io->io_hdr,
11256                                       ctl_io_hdr, links);
11257         }
11258         mtx_unlock(&ctl_softc->queue_lock);
11259 #endif
11260
11261         for (lun_idx=0; lun_idx < ctl_softc->num_luns; lun_idx++) {
11262                 lun = ctl_softc->ctl_luns[lun_idx];
11263                 if (lun==NULL)
11264                         continue;
11265
11266                 /*
11267                  * Processor LUNs are primary on both sides.
11268                  * XXX will this always be true?
11269                  */
11270                 if (lun->be_lun->lun_type == T_PROCESSOR)
11271                         continue;
11272
11273                 if ((lun->flags & CTL_LUN_PRIMARY_SC)
11274                  && (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11275                         printf("FAILOVER: primary lun %d\n", lun_idx);
11276                         /*
11277                          * Remove all commands from the other SC. First from the
11278                          * blocked queue then from the ooa queue. Once we have
11279                          * removed them. Call ctl_check_blocked to see if there
11280                          * is anything that can run.
11281                          */
11282                         for (io = (union ctl_io *)TAILQ_FIRST(
11283                              &lun->blocked_queue); io != NULL; io = next_io) {
11284
11285                                 next_io = (union ctl_io *)TAILQ_NEXT(
11286                                     &io->io_hdr, blocked_links);
11287
11288                                 if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11289                                         TAILQ_REMOVE(&lun->blocked_queue,
11290                                                      &io->io_hdr,blocked_links);
11291                                         io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11292                                         TAILQ_REMOVE(&lun->ooa_queue,
11293                                                      &io->io_hdr, ooa_links);
11294
11295                                         ctl_free_io(io);
11296                                 }
11297                         }
11298
11299                         for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11300                              io != NULL; io = next_io) {
11301
11302                                 next_io = (union ctl_io *)TAILQ_NEXT(
11303                                     &io->io_hdr, ooa_links);
11304
11305                                 if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11306
11307                                         TAILQ_REMOVE(&lun->ooa_queue,
11308                                                 &io->io_hdr,
11309                                                 ooa_links);
11310
11311                                         ctl_free_io(io);
11312                                 }
11313                         }
11314                         ctl_check_blocked(lun);
11315                 } else if ((lun->flags & CTL_LUN_PRIMARY_SC)
11316                         && (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
11317
11318                         printf("FAILOVER: primary lun %d\n", lun_idx);
11319                         /*
11320                          * Abort all commands from the other SC.  We can't
11321                          * send status back for them now.  These should get
11322                          * cleaned up when they are completed or come out
11323                          * for a datamove operation.
11324                          */
11325                         for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11326                              io != NULL; io = next_io) {
11327                                 next_io = (union ctl_io *)TAILQ_NEXT(
11328                                         &io->io_hdr, ooa_links);
11329
11330                                 if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
11331                                         io->io_hdr.flags |= CTL_FLAG_ABORT;
11332                         }
11333                 } else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11334                         && (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
11335
11336                         printf("FAILOVER: secondary lun %d\n", lun_idx);
11337
11338                         lun->flags |= CTL_LUN_PRIMARY_SC;
11339
11340                         /*
11341                          * We send all I/O that was sent to this controller
11342                          * and redirected to the other side back with
11343                          * busy status, and have the initiator retry it.
11344                          * Figuring out how much data has been transferred,
11345                          * etc. and picking up where we left off would be 
11346                          * very tricky.
11347                          *
11348                          * XXX KDM need to remove I/O from the blocked
11349                          * queue as well!
11350                          */
11351                         for (pending_io = (union ctl_io *)TAILQ_FIRST(
11352                              &lun->ooa_queue); pending_io != NULL;
11353                              pending_io = next_io) {
11354
11355                                 next_io =  (union ctl_io *)TAILQ_NEXT(
11356                                         &pending_io->io_hdr, ooa_links);
11357
11358                                 pending_io->io_hdr.flags &=
11359                                         ~CTL_FLAG_SENT_2OTHER_SC;
11360
11361                                 if (pending_io->io_hdr.flags &
11362                                     CTL_FLAG_IO_ACTIVE) {
11363                                         pending_io->io_hdr.flags |=
11364                                                 CTL_FLAG_FAILOVER;
11365                                 } else {
11366                                         ctl_set_busy(&pending_io->scsiio);
11367                                         ctl_done(pending_io);
11368                                 }
11369                         }
11370
11371                         /*
11372                          * Build Unit Attention
11373                          */
11374                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11375                                 lun->pending_ua[i] |=
11376                                                      CTL_UA_ASYM_ACC_CHANGE;
11377                         }
11378                 } else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11379                         && (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11380                         printf("FAILOVER: secondary lun %d\n", lun_idx);
11381                         /*
11382                          * if the first io on the OOA is not on the RtR queue
11383                          * add it.
11384                          */
11385                         lun->flags |= CTL_LUN_PRIMARY_SC;
11386
11387                         pending_io = (union ctl_io *)TAILQ_FIRST(
11388                             &lun->ooa_queue);
11389                         if (pending_io==NULL) {
11390                                 printf("Nothing on OOA queue\n");
11391                                 continue;
11392                         }
11393
11394                         pending_io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11395                         if ((pending_io->io_hdr.flags &
11396                              CTL_FLAG_IS_WAS_ON_RTR) == 0) {
11397                                 pending_io->io_hdr.flags |=
11398                                     CTL_FLAG_IS_WAS_ON_RTR;
11399                                 ctl_enqueue_rtr(pending_io);
11400                         }
11401 #if 0
11402                         else
11403                         {
11404                                 printf("Tag 0x%04x is running\n",
11405                                       pending_io->scsiio.tag_num);
11406                         }
11407 #endif
11408
11409                         next_io = (union ctl_io *)TAILQ_NEXT(
11410                             &pending_io->io_hdr, ooa_links);
11411                         for (pending_io=next_io; pending_io != NULL;
11412                              pending_io = next_io) {
11413                                 pending_io->io_hdr.flags &=
11414                                     ~CTL_FLAG_SENT_2OTHER_SC;
11415                                 next_io = (union ctl_io *)TAILQ_NEXT(
11416                                         &pending_io->io_hdr, ooa_links);
11417                                 if (pending_io->io_hdr.flags &
11418                                     CTL_FLAG_IS_WAS_ON_RTR) {
11419 #if 0
11420                                         printf("Tag 0x%04x is running\n",
11421                                                 pending_io->scsiio.tag_num);
11422 #endif
11423                                         continue;
11424                                 }
11425
11426                                 switch (ctl_check_ooa(lun, pending_io,
11427                                     (union ctl_io *)TAILQ_PREV(
11428                                     &pending_io->io_hdr, ctl_ooaq,
11429                                     ooa_links))) {
11430
11431                                 case CTL_ACTION_BLOCK:
11432                                         TAILQ_INSERT_TAIL(&lun->blocked_queue,
11433                                                           &pending_io->io_hdr,
11434                                                           blocked_links);
11435                                         pending_io->io_hdr.flags |=
11436                                             CTL_FLAG_BLOCKED;
11437                                         break;
11438                                 case CTL_ACTION_PASS:
11439                                 case CTL_ACTION_SKIP:
11440                                         pending_io->io_hdr.flags |=
11441                                             CTL_FLAG_IS_WAS_ON_RTR;
11442                                         ctl_enqueue_rtr(pending_io);
11443                                         break;
11444                                 case CTL_ACTION_OVERLAP:
11445                                         ctl_set_overlapped_cmd(
11446                                             (struct ctl_scsiio *)pending_io);
11447                                         ctl_done(pending_io);
11448                                         break;
11449                                 case CTL_ACTION_OVERLAP_TAG:
11450                                         ctl_set_overlapped_tag(
11451                                             (struct ctl_scsiio *)pending_io,
11452                                             pending_io->scsiio.tag_num & 0xff);
11453                                         ctl_done(pending_io);
11454                                         break;
11455                                 case CTL_ACTION_ERROR:
11456                                 default:
11457                                         ctl_set_internal_failure(
11458                                                 (struct ctl_scsiio *)pending_io,
11459                                                 0,  // sks_valid
11460                                                 0); //retry count
11461                                         ctl_done(pending_io);
11462                                         break;
11463                                 }
11464                         }
11465
11466                         /*
11467                          * Build Unit Attention
11468                          */
11469                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11470                                 lun->pending_ua[i] |=
11471                                                      CTL_UA_ASYM_ACC_CHANGE;
11472                         }
11473                 } else {
11474                         panic("Unhandled HA mode failover, LUN flags = %#x, "
11475                               "ha_mode = #%x", lun->flags, ctl_softc->ha_mode);
11476                 }
11477         }
11478         ctl_pause_rtr = 0;
11479         mtx_unlock(&ctl_softc->ctl_lock);
11480 }
11481
11482 static int
11483 ctl_scsiio_precheck(struct ctl_softc *ctl_softc, struct ctl_scsiio *ctsio)
11484 {
11485         struct ctl_lun *lun;
11486         const struct ctl_cmd_entry *entry;
11487         uint32_t initidx, targ_lun;
11488         int retval;
11489
11490         retval = 0;
11491
11492         lun = NULL;
11493
11494         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11495         if ((targ_lun < CTL_MAX_LUNS)
11496          && (ctl_softc->ctl_luns[targ_lun] != NULL)) {
11497                 lun = ctl_softc->ctl_luns[targ_lun];
11498                 /*
11499                  * If the LUN is invalid, pretend that it doesn't exist.
11500                  * It will go away as soon as all pending I/O has been
11501                  * completed.
11502                  */
11503                 if (lun->flags & CTL_LUN_DISABLED) {
11504                         lun = NULL;
11505                 } else {
11506                         ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
11507                         ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
11508                                 lun->be_lun;
11509                         if (lun->be_lun->lun_type == T_PROCESSOR) {
11510                                 ctsio->io_hdr.flags |= CTL_FLAG_CONTROL_DEV;
11511                         }
11512
11513                         /*
11514                          * Every I/O goes into the OOA queue for a
11515                          * particular LUN, and stays there until completion.
11516                          */
11517                         mtx_lock(&lun->lun_lock);
11518                         TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr,
11519                             ooa_links);
11520                 }
11521         } else {
11522                 ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11523                 ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11524         }
11525
11526         /* Get command entry and return error if it is unsuppotyed. */
11527         entry = ctl_validate_command(ctsio);
11528         if (entry == NULL) {
11529                 if (lun)
11530                         mtx_unlock(&lun->lun_lock);
11531                 return (retval);
11532         }
11533
11534         ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11535         ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11536
11537         /*
11538          * Check to see whether we can send this command to LUNs that don't
11539          * exist.  This should pretty much only be the case for inquiry
11540          * and request sense.  Further checks, below, really require having
11541          * a LUN, so we can't really check the command anymore.  Just put
11542          * it on the rtr queue.
11543          */
11544         if (lun == NULL) {
11545                 if (entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) {
11546                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11547                         ctl_enqueue_rtr((union ctl_io *)ctsio);
11548                         return (retval);
11549                 }
11550
11551                 ctl_set_unsupported_lun(ctsio);
11552                 ctl_done((union ctl_io *)ctsio);
11553                 CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11554                 return (retval);
11555         } else {
11556                 /*
11557                  * Make sure we support this particular command on this LUN.
11558                  * e.g., we don't support writes to the control LUN.
11559                  */
11560                 if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11561                         mtx_unlock(&lun->lun_lock);
11562                         ctl_set_invalid_opcode(ctsio);
11563                         ctl_done((union ctl_io *)ctsio);
11564                         return (retval);
11565                 }
11566         }
11567
11568         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11569
11570 #ifdef CTL_WITH_CA
11571         /*
11572          * If we've got a request sense, it'll clear the contingent
11573          * allegiance condition.  Otherwise, if we have a CA condition for
11574          * this initiator, clear it, because it sent down a command other
11575          * than request sense.
11576          */
11577         if ((ctsio->cdb[0] != REQUEST_SENSE)
11578          && (ctl_is_set(lun->have_ca, initidx)))
11579                 ctl_clear_mask(lun->have_ca, initidx);
11580 #endif
11581
11582         /*
11583          * If the command has this flag set, it handles its own unit
11584          * attention reporting, we shouldn't do anything.  Otherwise we
11585          * check for any pending unit attentions, and send them back to the
11586          * initiator.  We only do this when a command initially comes in,
11587          * not when we pull it off the blocked queue.
11588          *
11589          * According to SAM-3, section 5.3.2, the order that things get
11590          * presented back to the host is basically unit attentions caused
11591          * by some sort of reset event, busy status, reservation conflicts
11592          * or task set full, and finally any other status.
11593          *
11594          * One issue here is that some of the unit attentions we report
11595          * don't fall into the "reset" category (e.g. "reported luns data
11596          * has changed").  So reporting it here, before the reservation
11597          * check, may be technically wrong.  I guess the only thing to do
11598          * would be to check for and report the reset events here, and then
11599          * check for the other unit attention types after we check for a
11600          * reservation conflict.
11601          *
11602          * XXX KDM need to fix this
11603          */
11604         if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11605                 ctl_ua_type ua_type;
11606
11607                 ua_type = lun->pending_ua[initidx];
11608                 if (ua_type != CTL_UA_NONE) {
11609                         scsi_sense_data_type sense_format;
11610
11611                         if (lun != NULL)
11612                                 sense_format = (lun->flags &
11613                                     CTL_LUN_SENSE_DESC) ? SSD_TYPE_DESC :
11614                                     SSD_TYPE_FIXED;
11615                         else
11616                                 sense_format = SSD_TYPE_FIXED;
11617
11618                         ua_type = ctl_build_ua(ua_type, &ctsio->sense_data,
11619                                                sense_format);
11620                         if (ua_type != CTL_UA_NONE) {
11621                                 ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11622                                 ctsio->io_hdr.status = CTL_SCSI_ERROR |
11623                                                        CTL_AUTOSENSE;
11624                                 ctsio->sense_len = SSD_FULL_SIZE;
11625                                 lun->pending_ua[initidx] &= ~ua_type;
11626                                 mtx_unlock(&lun->lun_lock);
11627                                 ctl_done((union ctl_io *)ctsio);
11628                                 return (retval);
11629                         }
11630                 }
11631         }
11632
11633
11634         if (ctl_scsiio_lun_check(ctl_softc, lun, entry, ctsio) != 0) {
11635                 mtx_unlock(&lun->lun_lock);
11636                 ctl_done((union ctl_io *)ctsio);
11637                 return (retval);
11638         }
11639
11640         /*
11641          * XXX CHD this is where we want to send IO to other side if
11642          * this LUN is secondary on this SC. We will need to make a copy
11643          * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11644          * the copy we send as FROM_OTHER.
11645          * We also need to stuff the address of the original IO so we can
11646          * find it easily. Something similar will need be done on the other
11647          * side so when we are done we can find the copy.
11648          */
11649         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11650                 union ctl_ha_msg msg_info;
11651                 int isc_retval;
11652
11653                 ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11654
11655                 msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11656                 msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11657 #if 0
11658                 printf("1. ctsio %p\n", ctsio);
11659 #endif
11660                 msg_info.hdr.serializing_sc = NULL;
11661                 msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11662                 msg_info.scsi.tag_num = ctsio->tag_num;
11663                 msg_info.scsi.tag_type = ctsio->tag_type;
11664                 memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11665
11666                 ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11667
11668                 if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11669                     (void *)&msg_info, sizeof(msg_info), 0)) >
11670                     CTL_HA_STATUS_SUCCESS) {
11671                         printf("CTL:precheck, ctl_ha_msg_send returned %d\n",
11672                                isc_retval);
11673                         printf("CTL:opcode is %x\n", ctsio->cdb[0]);
11674                 } else {
11675 #if 0
11676                         printf("CTL:Precheck sent msg, opcode is %x\n",opcode);
11677 #endif
11678                 }
11679
11680                 /*
11681                  * XXX KDM this I/O is off the incoming queue, but hasn't
11682                  * been inserted on any other queue.  We may need to come
11683                  * up with a holding queue while we wait for serialization
11684                  * so that we have an idea of what we're waiting for from
11685                  * the other side.
11686                  */
11687                 mtx_unlock(&lun->lun_lock);
11688                 return (retval);
11689         }
11690
11691         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11692                               (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11693                               ctl_ooaq, ooa_links))) {
11694         case CTL_ACTION_BLOCK:
11695                 ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11696                 TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11697                                   blocked_links);
11698                 mtx_unlock(&lun->lun_lock);
11699                 return (retval);
11700         case CTL_ACTION_PASS:
11701         case CTL_ACTION_SKIP:
11702                 ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11703                 mtx_unlock(&lun->lun_lock);
11704                 ctl_enqueue_rtr((union ctl_io *)ctsio);
11705                 break;
11706         case CTL_ACTION_OVERLAP:
11707                 mtx_unlock(&lun->lun_lock);
11708                 ctl_set_overlapped_cmd(ctsio);
11709                 ctl_done((union ctl_io *)ctsio);
11710                 break;
11711         case CTL_ACTION_OVERLAP_TAG:
11712                 mtx_unlock(&lun->lun_lock);
11713                 ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11714                 ctl_done((union ctl_io *)ctsio);
11715                 break;
11716         case CTL_ACTION_ERROR:
11717         default:
11718                 mtx_unlock(&lun->lun_lock);
11719                 ctl_set_internal_failure(ctsio,
11720                                          /*sks_valid*/ 0,
11721                                          /*retry_count*/ 0);
11722                 ctl_done((union ctl_io *)ctsio);
11723                 break;
11724         }
11725         return (retval);
11726 }
11727
11728 const struct ctl_cmd_entry *
11729 ctl_get_cmd_entry(struct ctl_scsiio *ctsio)
11730 {
11731         const struct ctl_cmd_entry *entry;
11732         int service_action;
11733
11734         entry = &ctl_cmd_table[ctsio->cdb[0]];
11735         if (entry->flags & CTL_CMD_FLAG_SA5) {
11736                 service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11737                 entry = &((const struct ctl_cmd_entry *)
11738                     entry->execute)[service_action];
11739         }
11740         return (entry);
11741 }
11742
11743 const struct ctl_cmd_entry *
11744 ctl_validate_command(struct ctl_scsiio *ctsio)
11745 {
11746         const struct ctl_cmd_entry *entry;
11747         int i;
11748         uint8_t diff;
11749
11750         entry = ctl_get_cmd_entry(ctsio);
11751         if (entry->execute == NULL) {
11752                 ctl_set_invalid_opcode(ctsio);
11753                 ctl_done((union ctl_io *)ctsio);
11754                 return (NULL);
11755         }
11756         KASSERT(entry->length > 0,
11757             ("Not defined length for command 0x%02x/0x%02x",
11758              ctsio->cdb[0], ctsio->cdb[1]));
11759         for (i = 1; i < entry->length; i++) {
11760                 diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11761                 if (diff == 0)
11762                         continue;
11763                 ctl_set_invalid_field(ctsio,
11764                                       /*sks_valid*/ 1,
11765                                       /*command*/ 1,
11766                                       /*field*/ i,
11767                                       /*bit_valid*/ 1,
11768                                       /*bit*/ fls(diff) - 1);
11769                 ctl_done((union ctl_io *)ctsio);
11770                 return (NULL);
11771         }
11772         return (entry);
11773 }
11774
11775 static int
11776 ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11777 {
11778
11779         switch (lun_type) {
11780         case T_PROCESSOR:
11781                 if (((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0) &&
11782                     ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11783                         return (0);
11784                 break;
11785         case T_DIRECT:
11786                 if (((entry->flags & CTL_CMD_FLAG_OK_ON_SLUN) == 0) &&
11787                     ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11788                         return (0);
11789                 break;
11790         default:
11791                 return (0);
11792         }
11793         return (1);
11794 }
11795
11796 static int
11797 ctl_scsiio(struct ctl_scsiio *ctsio)
11798 {
11799         int retval;
11800         const struct ctl_cmd_entry *entry;
11801
11802         retval = CTL_RETVAL_COMPLETE;
11803
11804         CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11805
11806         entry = ctl_get_cmd_entry(ctsio);
11807
11808         /*
11809          * If this I/O has been aborted, just send it straight to
11810          * ctl_done() without executing it.
11811          */
11812         if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11813                 ctl_done((union ctl_io *)ctsio);
11814                 goto bailout;
11815         }
11816
11817         /*
11818          * All the checks should have been handled by ctl_scsiio_precheck().
11819          * We should be clear now to just execute the I/O.
11820          */
11821         retval = entry->execute(ctsio);
11822
11823 bailout:
11824         return (retval);
11825 }
11826
11827 /*
11828  * Since we only implement one target right now, a bus reset simply resets
11829  * our single target.
11830  */
11831 static int
11832 ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io)
11833 {
11834         return(ctl_target_reset(ctl_softc, io, CTL_UA_BUS_RESET));
11835 }
11836
11837 static int
11838 ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
11839                  ctl_ua_type ua_type)
11840 {
11841         struct ctl_lun *lun;
11842         int retval;
11843
11844         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11845                 union ctl_ha_msg msg_info;
11846
11847                 io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11848                 msg_info.hdr.nexus = io->io_hdr.nexus;
11849                 if (ua_type==CTL_UA_TARG_RESET)
11850                         msg_info.task.task_action = CTL_TASK_TARGET_RESET;
11851                 else
11852                         msg_info.task.task_action = CTL_TASK_BUS_RESET;
11853                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11854                 msg_info.hdr.original_sc = NULL;
11855                 msg_info.hdr.serializing_sc = NULL;
11856                 if (CTL_HA_STATUS_SUCCESS != ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11857                     (void *)&msg_info, sizeof(msg_info), 0)) {
11858                 }
11859         }
11860         retval = 0;
11861
11862         mtx_lock(&ctl_softc->ctl_lock);
11863         STAILQ_FOREACH(lun, &ctl_softc->lun_list, links)
11864                 retval += ctl_lun_reset(lun, io, ua_type);
11865         mtx_unlock(&ctl_softc->ctl_lock);
11866
11867         return (retval);
11868 }
11869
11870 /*
11871  * The LUN should always be set.  The I/O is optional, and is used to
11872  * distinguish between I/Os sent by this initiator, and by other
11873  * initiators.  We set unit attention for initiators other than this one.
11874  * SAM-3 is vague on this point.  It does say that a unit attention should
11875  * be established for other initiators when a LUN is reset (see section
11876  * 5.7.3), but it doesn't specifically say that the unit attention should
11877  * be established for this particular initiator when a LUN is reset.  Here
11878  * is the relevant text, from SAM-3 rev 8:
11879  *
11880  * 5.7.2 When a SCSI initiator port aborts its own tasks
11881  *
11882  * When a SCSI initiator port causes its own task(s) to be aborted, no
11883  * notification that the task(s) have been aborted shall be returned to
11884  * the SCSI initiator port other than the completion response for the
11885  * command or task management function action that caused the task(s) to
11886  * be aborted and notification(s) associated with related effects of the
11887  * action (e.g., a reset unit attention condition).
11888  *
11889  * XXX KDM for now, we're setting unit attention for all initiators.
11890  */
11891 static int
11892 ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
11893 {
11894         union ctl_io *xio;
11895 #if 0
11896         uint32_t initindex;
11897 #endif
11898         int i;
11899
11900         mtx_lock(&lun->lun_lock);
11901         /*
11902          * Run through the OOA queue and abort each I/O.
11903          */
11904 #if 0
11905         TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
11906 #endif
11907         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11908              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11909                 xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
11910         }
11911
11912         /*
11913          * This version sets unit attention for every
11914          */
11915 #if 0
11916         initindex = ctl_get_initindex(&io->io_hdr.nexus);
11917         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11918                 if (initindex == i)
11919                         continue;
11920                 lun->pending_ua[i] |= ua_type;
11921         }
11922 #endif
11923
11924         /*
11925          * A reset (any kind, really) clears reservations established with
11926          * RESERVE/RELEASE.  It does not clear reservations established
11927          * with PERSISTENT RESERVE OUT, but we don't support that at the
11928          * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
11929          * reservations made with the RESERVE/RELEASE commands, because
11930          * those commands are obsolete in SPC-3.
11931          */
11932         lun->flags &= ~CTL_LUN_RESERVED;
11933
11934         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11935 #ifdef CTL_WITH_CA
11936                 ctl_clear_mask(lun->have_ca, i);
11937 #endif
11938                 lun->pending_ua[i] |= ua_type;
11939         }
11940         mtx_unlock(&lun->lun_lock);
11941
11942         return (0);
11943 }
11944
11945 static int
11946 ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
11947     int other_sc)
11948 {
11949         union ctl_io *xio;
11950         int found;
11951
11952         mtx_assert(&lun->lun_lock, MA_OWNED);
11953
11954         /*
11955          * Run through the OOA queue and attempt to find the given I/O.
11956          * The target port, initiator ID, tag type and tag number have to
11957          * match the values that we got from the initiator.  If we have an
11958          * untagged command to abort, simply abort the first untagged command
11959          * we come to.  We only allow one untagged command at a time of course.
11960          */
11961         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11962              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11963
11964                 if ((targ_port == UINT32_MAX ||
11965                      targ_port == xio->io_hdr.nexus.targ_port) &&
11966                     (init_id == UINT32_MAX ||
11967                      init_id == xio->io_hdr.nexus.initid.id)) {
11968                         if (targ_port != xio->io_hdr.nexus.targ_port ||
11969                             init_id != xio->io_hdr.nexus.initid.id)
11970                                 xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS;
11971                         xio->io_hdr.flags |= CTL_FLAG_ABORT;
11972                         found = 1;
11973                         if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11974                                 union ctl_ha_msg msg_info;
11975
11976                                 msg_info.hdr.nexus = xio->io_hdr.nexus;
11977                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
11978                                 msg_info.task.tag_num = xio->scsiio.tag_num;
11979                                 msg_info.task.tag_type = xio->scsiio.tag_type;
11980                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11981                                 msg_info.hdr.original_sc = NULL;
11982                                 msg_info.hdr.serializing_sc = NULL;
11983                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11984                                     (void *)&msg_info, sizeof(msg_info), 0);
11985                         }
11986                 }
11987         }
11988         return (found);
11989 }
11990
11991 static int
11992 ctl_abort_task_set(union ctl_io *io)
11993 {
11994         struct ctl_softc *softc = control_softc;
11995         struct ctl_lun *lun;
11996         uint32_t targ_lun;
11997
11998         /*
11999          * Look up the LUN.
12000          */
12001         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12002         mtx_lock(&softc->ctl_lock);
12003         if ((targ_lun < CTL_MAX_LUNS) && (softc->ctl_luns[targ_lun] != NULL))
12004                 lun = softc->ctl_luns[targ_lun];
12005         else {
12006                 mtx_unlock(&softc->ctl_lock);
12007                 return (1);
12008         }
12009
12010         mtx_lock(&lun->lun_lock);
12011         mtx_unlock(&softc->ctl_lock);
12012         if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
12013                 ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12014                     io->io_hdr.nexus.initid.id,
12015                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12016         } else { /* CTL_TASK_CLEAR_TASK_SET */
12017                 ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
12018                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12019         }
12020         mtx_unlock(&lun->lun_lock);
12021         return (0);
12022 }
12023
12024 static int
12025 ctl_i_t_nexus_reset(union ctl_io *io)
12026 {
12027         struct ctl_softc *softc = control_softc;
12028         struct ctl_lun *lun;
12029         uint32_t initindex;
12030
12031         initindex = ctl_get_initindex(&io->io_hdr.nexus);
12032         mtx_lock(&softc->ctl_lock);
12033         STAILQ_FOREACH(lun, &softc->lun_list, links) {
12034                 mtx_lock(&lun->lun_lock);
12035                 ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12036                     io->io_hdr.nexus.initid.id,
12037                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12038 #ifdef CTL_WITH_CA
12039                 ctl_clear_mask(lun->have_ca, initindex);
12040 #endif
12041                 lun->pending_ua[initindex] |= CTL_UA_I_T_NEXUS_LOSS;
12042                 mtx_unlock(&lun->lun_lock);
12043         }
12044         mtx_unlock(&softc->ctl_lock);
12045         return (0);
12046 }
12047
12048 static int
12049 ctl_abort_task(union ctl_io *io)
12050 {
12051         union ctl_io *xio;
12052         struct ctl_lun *lun;
12053         struct ctl_softc *ctl_softc;
12054 #if 0
12055         struct sbuf sb;
12056         char printbuf[128];
12057 #endif
12058         int found;
12059         uint32_t targ_lun;
12060
12061         ctl_softc = control_softc;
12062         found = 0;
12063
12064         /*
12065          * Look up the LUN.
12066          */
12067         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12068         mtx_lock(&ctl_softc->ctl_lock);
12069         if ((targ_lun < CTL_MAX_LUNS)
12070          && (ctl_softc->ctl_luns[targ_lun] != NULL))
12071                 lun = ctl_softc->ctl_luns[targ_lun];
12072         else {
12073                 mtx_unlock(&ctl_softc->ctl_lock);
12074                 return (1);
12075         }
12076
12077 #if 0
12078         printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
12079                lun->lun, io->taskio.tag_num, io->taskio.tag_type);
12080 #endif
12081
12082         mtx_lock(&lun->lun_lock);
12083         mtx_unlock(&ctl_softc->ctl_lock);
12084         /*
12085          * Run through the OOA queue and attempt to find the given I/O.
12086          * The target port, initiator ID, tag type and tag number have to
12087          * match the values that we got from the initiator.  If we have an
12088          * untagged command to abort, simply abort the first untagged command
12089          * we come to.  We only allow one untagged command at a time of course.
12090          */
12091 #if 0
12092         TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
12093 #endif
12094         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12095              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12096 #if 0
12097                 sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
12098
12099                 sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
12100                             lun->lun, xio->scsiio.tag_num,
12101                             xio->scsiio.tag_type,
12102                             (xio->io_hdr.blocked_links.tqe_prev
12103                             == NULL) ? "" : " BLOCKED",
12104                             (xio->io_hdr.flags &
12105                             CTL_FLAG_DMA_INPROG) ? " DMA" : "",
12106                             (xio->io_hdr.flags &
12107                             CTL_FLAG_ABORT) ? " ABORT" : "",
12108                             (xio->io_hdr.flags &
12109                             CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
12110                 ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
12111                 sbuf_finish(&sb);
12112                 printf("%s\n", sbuf_data(&sb));
12113 #endif
12114
12115                 if ((xio->io_hdr.nexus.targ_port == io->io_hdr.nexus.targ_port)
12116                  && (xio->io_hdr.nexus.initid.id ==
12117                      io->io_hdr.nexus.initid.id)) {
12118                         /*
12119                          * If the abort says that the task is untagged, the
12120                          * task in the queue must be untagged.  Otherwise,
12121                          * we just check to see whether the tag numbers
12122                          * match.  This is because the QLogic firmware
12123                          * doesn't pass back the tag type in an abort
12124                          * request.
12125                          */
12126 #if 0
12127                         if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
12128                           && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
12129                          || (xio->scsiio.tag_num == io->taskio.tag_num)) {
12130 #endif
12131                         /*
12132                          * XXX KDM we've got problems with FC, because it
12133                          * doesn't send down a tag type with aborts.  So we
12134                          * can only really go by the tag number...
12135                          * This may cause problems with parallel SCSI.
12136                          * Need to figure that out!!
12137                          */
12138                         if (xio->scsiio.tag_num == io->taskio.tag_num) {
12139                                 xio->io_hdr.flags |= CTL_FLAG_ABORT;
12140                                 found = 1;
12141                                 if ((io->io_hdr.flags &
12142                                      CTL_FLAG_FROM_OTHER_SC) == 0 &&
12143                                     !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12144                                         union ctl_ha_msg msg_info;
12145
12146                                         io->io_hdr.flags |=
12147                                                         CTL_FLAG_SENT_2OTHER_SC;
12148                                         msg_info.hdr.nexus = io->io_hdr.nexus;
12149                                         msg_info.task.task_action =
12150                                                 CTL_TASK_ABORT_TASK;
12151                                         msg_info.task.tag_num =
12152                                                 io->taskio.tag_num;
12153                                         msg_info.task.tag_type =
12154                                                 io->taskio.tag_type;
12155                                         msg_info.hdr.msg_type =
12156                                                 CTL_MSG_MANAGE_TASKS;
12157                                         msg_info.hdr.original_sc = NULL;
12158                                         msg_info.hdr.serializing_sc = NULL;
12159 #if 0
12160                                         printf("Sent Abort to other side\n");
12161 #endif
12162                                         if (CTL_HA_STATUS_SUCCESS !=
12163                                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12164                                                 (void *)&msg_info,
12165                                                 sizeof(msg_info), 0)) {
12166                                         }
12167                                 }
12168 #if 0
12169                                 printf("ctl_abort_task: found I/O to abort\n");
12170 #endif
12171                                 break;
12172                         }
12173                 }
12174         }
12175         mtx_unlock(&lun->lun_lock);
12176
12177         if (found == 0) {
12178                 /*
12179                  * This isn't really an error.  It's entirely possible for
12180                  * the abort and command completion to cross on the wire.
12181                  * This is more of an informative/diagnostic error.
12182                  */
12183 #if 0
12184                 printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
12185                        "%d:%d:%d:%d tag %d type %d\n",
12186                        io->io_hdr.nexus.initid.id,
12187                        io->io_hdr.nexus.targ_port,
12188                        io->io_hdr.nexus.targ_target.id,
12189                        io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
12190                        io->taskio.tag_type);
12191 #endif
12192         }
12193         return (0);
12194 }
12195
12196 static void
12197 ctl_run_task(union ctl_io *io)
12198 {
12199         struct ctl_softc *ctl_softc = control_softc;
12200         int retval = 1;
12201         const char *task_desc;
12202
12203         CTL_DEBUG_PRINT(("ctl_run_task\n"));
12204
12205         KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
12206             ("ctl_run_task: Unextected io_type %d\n",
12207              io->io_hdr.io_type));
12208
12209         task_desc = ctl_scsi_task_string(&io->taskio);
12210         if (task_desc != NULL) {
12211 #ifdef NEEDTOPORT
12212                 csevent_log(CSC_CTL | CSC_SHELF_SW |
12213                             CTL_TASK_REPORT,
12214                             csevent_LogType_Trace,
12215                             csevent_Severity_Information,
12216                             csevent_AlertLevel_Green,
12217                             csevent_FRU_Firmware,
12218                             csevent_FRU_Unknown,
12219                             "CTL: received task: %s",task_desc);
12220 #endif
12221         } else {
12222 #ifdef NEEDTOPORT
12223                 csevent_log(CSC_CTL | CSC_SHELF_SW |
12224                             CTL_TASK_REPORT,
12225                             csevent_LogType_Trace,
12226                             csevent_Severity_Information,
12227                             csevent_AlertLevel_Green,
12228                             csevent_FRU_Firmware,
12229                             csevent_FRU_Unknown,
12230                             "CTL: received unknown task "
12231                             "type: %d (%#x)",
12232                             io->taskio.task_action,
12233                             io->taskio.task_action);
12234 #endif
12235         }
12236         switch (io->taskio.task_action) {
12237         case CTL_TASK_ABORT_TASK:
12238                 retval = ctl_abort_task(io);
12239                 break;
12240         case CTL_TASK_ABORT_TASK_SET:
12241         case CTL_TASK_CLEAR_TASK_SET:
12242                 retval = ctl_abort_task_set(io);
12243                 break;
12244         case CTL_TASK_CLEAR_ACA:
12245                 break;
12246         case CTL_TASK_I_T_NEXUS_RESET:
12247                 retval = ctl_i_t_nexus_reset(io);
12248                 break;
12249         case CTL_TASK_LUN_RESET: {
12250                 struct ctl_lun *lun;
12251                 uint32_t targ_lun;
12252
12253                 targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12254                 mtx_lock(&ctl_softc->ctl_lock);
12255                 if ((targ_lun < CTL_MAX_LUNS)
12256                  && (ctl_softc->ctl_luns[targ_lun] != NULL))
12257                         lun = ctl_softc->ctl_luns[targ_lun];
12258                 else {
12259                         mtx_unlock(&ctl_softc->ctl_lock);
12260                         retval = 1;
12261                         break;
12262                 }
12263
12264                 if (!(io->io_hdr.flags &
12265                     CTL_FLAG_FROM_OTHER_SC)) {
12266                         union ctl_ha_msg msg_info;
12267
12268                         io->io_hdr.flags |=
12269                                 CTL_FLAG_SENT_2OTHER_SC;
12270                         msg_info.hdr.msg_type =
12271                                 CTL_MSG_MANAGE_TASKS;
12272                         msg_info.hdr.nexus = io->io_hdr.nexus;
12273                         msg_info.task.task_action =
12274                                 CTL_TASK_LUN_RESET;
12275                         msg_info.hdr.original_sc = NULL;
12276                         msg_info.hdr.serializing_sc = NULL;
12277                         if (CTL_HA_STATUS_SUCCESS !=
12278                             ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12279                             (void *)&msg_info,
12280                             sizeof(msg_info), 0)) {
12281                         }
12282                 }
12283
12284                 retval = ctl_lun_reset(lun, io,
12285                                        CTL_UA_LUN_RESET);
12286                 mtx_unlock(&ctl_softc->ctl_lock);
12287                 break;
12288         }
12289         case CTL_TASK_TARGET_RESET:
12290                 retval = ctl_target_reset(ctl_softc, io, CTL_UA_TARG_RESET);
12291                 break;
12292         case CTL_TASK_BUS_RESET:
12293                 retval = ctl_bus_reset(ctl_softc, io);
12294                 break;
12295         case CTL_TASK_PORT_LOGIN:
12296                 break;
12297         case CTL_TASK_PORT_LOGOUT:
12298                 break;
12299         default:
12300                 printf("ctl_run_task: got unknown task management event %d\n",
12301                        io->taskio.task_action);
12302                 break;
12303         }
12304         if (retval == 0)
12305                 io->io_hdr.status = CTL_SUCCESS;
12306         else
12307                 io->io_hdr.status = CTL_ERROR;
12308         ctl_done(io);
12309 }
12310
12311 /*
12312  * For HA operation.  Handle commands that come in from the other
12313  * controller.
12314  */
12315 static void
12316 ctl_handle_isc(union ctl_io *io)
12317 {
12318         int free_io;
12319         struct ctl_lun *lun;
12320         struct ctl_softc *ctl_softc;
12321         uint32_t targ_lun;
12322
12323         ctl_softc = control_softc;
12324
12325         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12326         lun = ctl_softc->ctl_luns[targ_lun];
12327
12328         switch (io->io_hdr.msg_type) {
12329         case CTL_MSG_SERIALIZE:
12330                 free_io = ctl_serialize_other_sc_cmd(&io->scsiio);
12331                 break;
12332         case CTL_MSG_R2R: {
12333                 const struct ctl_cmd_entry *entry;
12334
12335                 /*
12336                  * This is only used in SER_ONLY mode.
12337                  */
12338                 free_io = 0;
12339                 entry = ctl_get_cmd_entry(&io->scsiio);
12340                 mtx_lock(&lun->lun_lock);
12341                 if (ctl_scsiio_lun_check(ctl_softc, lun,
12342                     entry, (struct ctl_scsiio *)io) != 0) {
12343                         mtx_unlock(&lun->lun_lock);
12344                         ctl_done(io);
12345                         break;
12346                 }
12347                 io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
12348                 mtx_unlock(&lun->lun_lock);
12349                 ctl_enqueue_rtr(io);
12350                 break;
12351         }
12352         case CTL_MSG_FINISH_IO:
12353                 if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
12354                         free_io = 0;
12355                         ctl_done(io);
12356                 } else {
12357                         free_io = 1;
12358                         mtx_lock(&lun->lun_lock);
12359                         TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
12360                                      ooa_links);
12361                         ctl_check_blocked(lun);
12362                         mtx_unlock(&lun->lun_lock);
12363                 }
12364                 break;
12365         case CTL_MSG_PERS_ACTION:
12366                 ctl_hndl_per_res_out_on_other_sc(
12367                         (union ctl_ha_msg *)&io->presio.pr_msg);
12368                 free_io = 1;
12369                 break;
12370         case CTL_MSG_BAD_JUJU:
12371                 free_io = 0;
12372                 ctl_done(io);
12373                 break;
12374         case CTL_MSG_DATAMOVE:
12375                 /* Only used in XFER mode */
12376                 free_io = 0;
12377                 ctl_datamove_remote(io);
12378                 break;
12379         case CTL_MSG_DATAMOVE_DONE:
12380                 /* Only used in XFER mode */
12381                 free_io = 0;
12382                 io->scsiio.be_move_done(io);
12383                 break;
12384         default:
12385                 free_io = 1;
12386                 printf("%s: Invalid message type %d\n",
12387                        __func__, io->io_hdr.msg_type);
12388                 break;
12389         }
12390         if (free_io)
12391                 ctl_free_io(io);
12392
12393 }
12394
12395
12396 /*
12397  * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12398  * there is no match.
12399  */
12400 static ctl_lun_error_pattern
12401 ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12402 {
12403         const struct ctl_cmd_entry *entry;
12404         ctl_lun_error_pattern filtered_pattern, pattern;
12405
12406         pattern = desc->error_pattern;
12407
12408         /*
12409          * XXX KDM we need more data passed into this function to match a
12410          * custom pattern, and we actually need to implement custom pattern
12411          * matching.
12412          */
12413         if (pattern & CTL_LUN_PAT_CMD)
12414                 return (CTL_LUN_PAT_CMD);
12415
12416         if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12417                 return (CTL_LUN_PAT_ANY);
12418
12419         entry = ctl_get_cmd_entry(ctsio);
12420
12421         filtered_pattern = entry->pattern & pattern;
12422
12423         /*
12424          * If the user requested specific flags in the pattern (e.g.
12425          * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12426          * flags.
12427          *
12428          * If the user did not specify any flags, it doesn't matter whether
12429          * or not the command supports the flags.
12430          */
12431         if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12432              (pattern & ~CTL_LUN_PAT_MASK))
12433                 return (CTL_LUN_PAT_NONE);
12434
12435         /*
12436          * If the user asked for a range check, see if the requested LBA
12437          * range overlaps with this command's LBA range.
12438          */
12439         if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12440                 uint64_t lba1;
12441                 uint32_t len1;
12442                 ctl_action action;
12443                 int retval;
12444
12445                 retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12446                 if (retval != 0)
12447                         return (CTL_LUN_PAT_NONE);
12448
12449                 action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12450                                               desc->lba_range.len);
12451                 /*
12452                  * A "pass" means that the LBA ranges don't overlap, so
12453                  * this doesn't match the user's range criteria.
12454                  */
12455                 if (action == CTL_ACTION_PASS)
12456                         return (CTL_LUN_PAT_NONE);
12457         }
12458
12459         return (filtered_pattern);
12460 }
12461
12462 static void
12463 ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12464 {
12465         struct ctl_error_desc *desc, *desc2;
12466
12467         mtx_assert(&lun->lun_lock, MA_OWNED);
12468
12469         STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12470                 ctl_lun_error_pattern pattern;
12471                 /*
12472                  * Check to see whether this particular command matches
12473                  * the pattern in the descriptor.
12474                  */
12475                 pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12476                 if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12477                         continue;
12478
12479                 switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12480                 case CTL_LUN_INJ_ABORTED:
12481                         ctl_set_aborted(&io->scsiio);
12482                         break;
12483                 case CTL_LUN_INJ_MEDIUM_ERR:
12484                         ctl_set_medium_error(&io->scsiio);
12485                         break;
12486                 case CTL_LUN_INJ_UA:
12487                         /* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12488                          * OCCURRED */
12489                         ctl_set_ua(&io->scsiio, 0x29, 0x00);
12490                         break;
12491                 case CTL_LUN_INJ_CUSTOM:
12492                         /*
12493                          * We're assuming the user knows what he is doing.
12494                          * Just copy the sense information without doing
12495                          * checks.
12496                          */
12497                         bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12498                               ctl_min(sizeof(desc->custom_sense),
12499                                       sizeof(io->scsiio.sense_data)));
12500                         io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12501                         io->scsiio.sense_len = SSD_FULL_SIZE;
12502                         io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12503                         break;
12504                 case CTL_LUN_INJ_NONE:
12505                 default:
12506                         /*
12507                          * If this is an error injection type we don't know
12508                          * about, clear the continuous flag (if it is set)
12509                          * so it will get deleted below.
12510                          */
12511                         desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12512                         break;
12513                 }
12514                 /*
12515                  * By default, each error injection action is a one-shot
12516                  */
12517                 if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12518                         continue;
12519
12520                 STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12521
12522                 free(desc, M_CTL);
12523         }
12524 }
12525
12526 #ifdef CTL_IO_DELAY
12527 static void
12528 ctl_datamove_timer_wakeup(void *arg)
12529 {
12530         union ctl_io *io;
12531
12532         io = (union ctl_io *)arg;
12533
12534         ctl_datamove(io);
12535 }
12536 #endif /* CTL_IO_DELAY */
12537
12538 void
12539 ctl_datamove(union ctl_io *io)
12540 {
12541         void (*fe_datamove)(union ctl_io *io);
12542
12543         mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
12544
12545         CTL_DEBUG_PRINT(("ctl_datamove\n"));
12546
12547 #ifdef CTL_TIME_IO
12548         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12549                 char str[256];
12550                 char path_str[64];
12551                 struct sbuf sb;
12552
12553                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
12554                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12555
12556                 sbuf_cat(&sb, path_str);
12557                 switch (io->io_hdr.io_type) {
12558                 case CTL_IO_SCSI:
12559                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12560                         sbuf_printf(&sb, "\n");
12561                         sbuf_cat(&sb, path_str);
12562                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12563                                     io->scsiio.tag_num, io->scsiio.tag_type);
12564                         break;
12565                 case CTL_IO_TASK:
12566                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12567                                     "Tag Type: %d\n", io->taskio.task_action,
12568                                     io->taskio.tag_num, io->taskio.tag_type);
12569                         break;
12570                 default:
12571                         printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12572                         panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12573                         break;
12574                 }
12575                 sbuf_cat(&sb, path_str);
12576                 sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12577                             (intmax_t)time_uptime - io->io_hdr.start_time);
12578                 sbuf_finish(&sb);
12579                 printf("%s", sbuf_data(&sb));
12580         }
12581 #endif /* CTL_TIME_IO */
12582
12583 #ifdef CTL_IO_DELAY
12584         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12585                 struct ctl_lun *lun;
12586
12587                 lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12588
12589                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12590         } else {
12591                 struct ctl_lun *lun;
12592
12593                 lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12594                 if ((lun != NULL)
12595                  && (lun->delay_info.datamove_delay > 0)) {
12596                         struct callout *callout;
12597
12598                         callout = (struct callout *)&io->io_hdr.timer_bytes;
12599                         callout_init(callout, /*mpsafe*/ 1);
12600                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12601                         callout_reset(callout,
12602                                       lun->delay_info.datamove_delay * hz,
12603                                       ctl_datamove_timer_wakeup, io);
12604                         if (lun->delay_info.datamove_type ==
12605                             CTL_DELAY_TYPE_ONESHOT)
12606                                 lun->delay_info.datamove_delay = 0;
12607                         return;
12608                 }
12609         }
12610 #endif
12611
12612         /*
12613          * This command has been aborted.  Set the port status, so we fail
12614          * the data move.
12615          */
12616         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12617                 printf("ctl_datamove: tag 0x%04x on (%ju:%d:%ju:%d) aborted\n",
12618                        io->scsiio.tag_num,(uintmax_t)io->io_hdr.nexus.initid.id,
12619                        io->io_hdr.nexus.targ_port,
12620                        (uintmax_t)io->io_hdr.nexus.targ_target.id,
12621                        io->io_hdr.nexus.targ_lun);
12622                 io->io_hdr.port_status = 31337;
12623                 /*
12624                  * Note that the backend, in this case, will get the
12625                  * callback in its context.  In other cases it may get
12626                  * called in the frontend's interrupt thread context.
12627                  */
12628                 io->scsiio.be_move_done(io);
12629                 return;
12630         }
12631
12632         /*
12633          * If we're in XFER mode and this I/O is from the other shelf
12634          * controller, we need to send the DMA to the other side to
12635          * actually transfer the data to/from the host.  In serialize only
12636          * mode the transfer happens below CTL and ctl_datamove() is only
12637          * called on the machine that originally received the I/O.
12638          */
12639         if ((control_softc->ha_mode == CTL_HA_MODE_XFER)
12640          && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12641                 union ctl_ha_msg msg;
12642                 uint32_t sg_entries_sent;
12643                 int do_sg_copy;
12644                 int i;
12645
12646                 memset(&msg, 0, sizeof(msg));
12647                 msg.hdr.msg_type = CTL_MSG_DATAMOVE;
12648                 msg.hdr.original_sc = io->io_hdr.original_sc;
12649                 msg.hdr.serializing_sc = io;
12650                 msg.hdr.nexus = io->io_hdr.nexus;
12651                 msg.dt.flags = io->io_hdr.flags;
12652                 /*
12653                  * We convert everything into a S/G list here.  We can't
12654                  * pass by reference, only by value between controllers.
12655                  * So we can't pass a pointer to the S/G list, only as many
12656                  * S/G entries as we can fit in here.  If it's possible for
12657                  * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
12658                  * then we need to break this up into multiple transfers.
12659                  */
12660                 if (io->scsiio.kern_sg_entries == 0) {
12661                         msg.dt.kern_sg_entries = 1;
12662                         /*
12663                          * If this is in cached memory, flush the cache
12664                          * before we send the DMA request to the other
12665                          * controller.  We want to do this in either the
12666                          * read or the write case.  The read case is
12667                          * straightforward.  In the write case, we want to
12668                          * make sure nothing is in the local cache that
12669                          * could overwrite the DMAed data.
12670                          */
12671                         if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12672                                 /*
12673                                  * XXX KDM use bus_dmamap_sync() here.
12674                                  */
12675                         }
12676
12677                         /*
12678                          * Convert to a physical address if this is a
12679                          * virtual address.
12680                          */
12681                         if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
12682                                 msg.dt.sg_list[0].addr =
12683                                         io->scsiio.kern_data_ptr;
12684                         } else {
12685                                 /*
12686                                  * XXX KDM use busdma here!
12687                                  */
12688 #if 0
12689                                 msg.dt.sg_list[0].addr = (void *)
12690                                         vtophys(io->scsiio.kern_data_ptr);
12691 #endif
12692                         }
12693
12694                         msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
12695                         do_sg_copy = 0;
12696                 } else {
12697                         struct ctl_sg_entry *sgl;
12698
12699                         do_sg_copy = 1;
12700                         msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
12701                         sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
12702                         if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12703                                 /*
12704                                  * XXX KDM use bus_dmamap_sync() here.
12705                                  */
12706                         }
12707                 }
12708
12709                 msg.dt.kern_data_len = io->scsiio.kern_data_len;
12710                 msg.dt.kern_total_len = io->scsiio.kern_total_len;
12711                 msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
12712                 msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
12713                 msg.dt.sg_sequence = 0;
12714
12715                 /*
12716                  * Loop until we've sent all of the S/G entries.  On the
12717                  * other end, we'll recompose these S/G entries into one
12718                  * contiguous list before passing it to the
12719                  */
12720                 for (sg_entries_sent = 0; sg_entries_sent <
12721                      msg.dt.kern_sg_entries; msg.dt.sg_sequence++) {
12722                         msg.dt.cur_sg_entries = ctl_min((sizeof(msg.dt.sg_list)/
12723                                 sizeof(msg.dt.sg_list[0])),
12724                                 msg.dt.kern_sg_entries - sg_entries_sent);
12725
12726                         if (do_sg_copy != 0) {
12727                                 struct ctl_sg_entry *sgl;
12728                                 int j;
12729
12730                                 sgl = (struct ctl_sg_entry *)
12731                                         io->scsiio.kern_data_ptr;
12732                                 /*
12733                                  * If this is in cached memory, flush the cache
12734                                  * before we send the DMA request to the other
12735                                  * controller.  We want to do this in either
12736                                  * the * read or the write case.  The read
12737                                  * case is straightforward.  In the write
12738                                  * case, we want to make sure nothing is
12739                                  * in the local cache that could overwrite
12740                                  * the DMAed data.
12741                                  */
12742
12743                                 for (i = sg_entries_sent, j = 0;
12744                                      i < msg.dt.cur_sg_entries; i++, j++) {
12745                                         if ((io->io_hdr.flags &
12746                                              CTL_FLAG_NO_DATASYNC) == 0) {
12747                                                 /*
12748                                                  * XXX KDM use bus_dmamap_sync()
12749                                                  */
12750                                         }
12751                                         if ((io->io_hdr.flags &
12752                                              CTL_FLAG_BUS_ADDR) == 0) {
12753                                                 /*
12754                                                  * XXX KDM use busdma.
12755                                                  */
12756 #if 0
12757                                                 msg.dt.sg_list[j].addr =(void *)
12758                                                        vtophys(sgl[i].addr);
12759 #endif
12760                                         } else {
12761                                                 msg.dt.sg_list[j].addr =
12762                                                         sgl[i].addr;
12763                                         }
12764                                         msg.dt.sg_list[j].len = sgl[i].len;
12765                                 }
12766                         }
12767
12768                         sg_entries_sent += msg.dt.cur_sg_entries;
12769                         if (sg_entries_sent >= msg.dt.kern_sg_entries)
12770                                 msg.dt.sg_last = 1;
12771                         else
12772                                 msg.dt.sg_last = 0;
12773
12774                         /*
12775                          * XXX KDM drop and reacquire the lock here?
12776                          */
12777                         if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12778                             sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
12779                                 /*
12780                                  * XXX do something here.
12781                                  */
12782                         }
12783
12784                         msg.dt.sent_sg_entries = sg_entries_sent;
12785                 }
12786                 io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12787                 if (io->io_hdr.flags & CTL_FLAG_FAILOVER)
12788                         ctl_failover_io(io, /*have_lock*/ 0);
12789
12790         } else {
12791
12792                 /*
12793                  * Lookup the fe_datamove() function for this particular
12794                  * front end.
12795                  */
12796                 fe_datamove =
12797                     control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12798
12799                 fe_datamove(io);
12800         }
12801 }
12802
12803 static void
12804 ctl_send_datamove_done(union ctl_io *io, int have_lock)
12805 {
12806         union ctl_ha_msg msg;
12807         int isc_status;
12808
12809         memset(&msg, 0, sizeof(msg));
12810
12811         msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12812         msg.hdr.original_sc = io;
12813         msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12814         msg.hdr.nexus = io->io_hdr.nexus;
12815         msg.hdr.status = io->io_hdr.status;
12816         msg.scsi.tag_num = io->scsiio.tag_num;
12817         msg.scsi.tag_type = io->scsiio.tag_type;
12818         msg.scsi.scsi_status = io->scsiio.scsi_status;
12819         memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12820                sizeof(io->scsiio.sense_data));
12821         msg.scsi.sense_len = io->scsiio.sense_len;
12822         msg.scsi.sense_residual = io->scsiio.sense_residual;
12823         msg.scsi.fetd_status = io->io_hdr.port_status;
12824         msg.scsi.residual = io->scsiio.residual;
12825         io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12826
12827         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12828                 ctl_failover_io(io, /*have_lock*/ have_lock);
12829                 return;
12830         }
12831
12832         isc_status = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0);
12833         if (isc_status > CTL_HA_STATUS_SUCCESS) {
12834                 /* XXX do something if this fails */
12835         }
12836
12837 }
12838
12839 /*
12840  * The DMA to the remote side is done, now we need to tell the other side
12841  * we're done so it can continue with its data movement.
12842  */
12843 static void
12844 ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12845 {
12846         union ctl_io *io;
12847
12848         io = rq->context;
12849
12850         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12851                 printf("%s: ISC DMA write failed with error %d", __func__,
12852                        rq->ret);
12853                 ctl_set_internal_failure(&io->scsiio,
12854                                          /*sks_valid*/ 1,
12855                                          /*retry_count*/ rq->ret);
12856         }
12857
12858         ctl_dt_req_free(rq);
12859
12860         /*
12861          * In this case, we had to malloc the memory locally.  Free it.
12862          */
12863         if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12864                 int i;
12865                 for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12866                         free(io->io_hdr.local_sglist[i].addr, M_CTL);
12867         }
12868         /*
12869          * The data is in local and remote memory, so now we need to send
12870          * status (good or back) back to the other side.
12871          */
12872         ctl_send_datamove_done(io, /*have_lock*/ 0);
12873 }
12874
12875 /*
12876  * We've moved the data from the host/controller into local memory.  Now we
12877  * need to push it over to the remote controller's memory.
12878  */
12879 static int
12880 ctl_datamove_remote_dm_write_cb(union ctl_io *io)
12881 {
12882         int retval;
12883
12884         retval = 0;
12885
12886         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12887                                           ctl_datamove_remote_write_cb);
12888
12889         return (retval);
12890 }
12891
12892 static void
12893 ctl_datamove_remote_write(union ctl_io *io)
12894 {
12895         int retval;
12896         void (*fe_datamove)(union ctl_io *io);
12897
12898         /*
12899          * - Get the data from the host/HBA into local memory.
12900          * - DMA memory from the local controller to the remote controller.
12901          * - Send status back to the remote controller.
12902          */
12903
12904         retval = ctl_datamove_remote_sgl_setup(io);
12905         if (retval != 0)
12906                 return;
12907
12908         /* Switch the pointer over so the FETD knows what to do */
12909         io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12910
12911         /*
12912          * Use a custom move done callback, since we need to send completion
12913          * back to the other controller, not to the backend on this side.
12914          */
12915         io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12916
12917         fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12918
12919         fe_datamove(io);
12920
12921         return;
12922
12923 }
12924
12925 static int
12926 ctl_datamove_remote_dm_read_cb(union ctl_io *io)
12927 {
12928 #if 0
12929         char str[256];
12930         char path_str[64];
12931         struct sbuf sb;
12932 #endif
12933
12934         /*
12935          * In this case, we had to malloc the memory locally.  Free it.
12936          */
12937         if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12938                 int i;
12939                 for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12940                         free(io->io_hdr.local_sglist[i].addr, M_CTL);
12941         }
12942
12943 #if 0
12944         scsi_path_string(io, path_str, sizeof(path_str));
12945         sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12946         sbuf_cat(&sb, path_str);
12947         scsi_command_string(&io->scsiio, NULL, &sb);
12948         sbuf_printf(&sb, "\n");
12949         sbuf_cat(&sb, path_str);
12950         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12951                     io->scsiio.tag_num, io->scsiio.tag_type);
12952         sbuf_cat(&sb, path_str);
12953         sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
12954                     io->io_hdr.flags, io->io_hdr.status);
12955         sbuf_finish(&sb);
12956         printk("%s", sbuf_data(&sb));
12957 #endif
12958
12959
12960         /*
12961          * The read is done, now we need to send status (good or bad) back
12962          * to the other side.
12963          */
12964         ctl_send_datamove_done(io, /*have_lock*/ 0);
12965
12966         return (0);
12967 }
12968
12969 static void
12970 ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
12971 {
12972         union ctl_io *io;
12973         void (*fe_datamove)(union ctl_io *io);
12974
12975         io = rq->context;
12976
12977         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12978                 printf("%s: ISC DMA read failed with error %d", __func__,
12979                        rq->ret);
12980                 ctl_set_internal_failure(&io->scsiio,
12981                                          /*sks_valid*/ 1,
12982                                          /*retry_count*/ rq->ret);
12983         }
12984
12985         ctl_dt_req_free(rq);
12986
12987         /* Switch the pointer over so the FETD knows what to do */
12988         io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12989
12990         /*
12991          * Use a custom move done callback, since we need to send completion
12992          * back to the other controller, not to the backend on this side.
12993          */
12994         io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
12995
12996         /* XXX KDM add checks like the ones in ctl_datamove? */
12997
12998         fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12999
13000         fe_datamove(io);
13001 }
13002
13003 static int
13004 ctl_datamove_remote_sgl_setup(union ctl_io *io)
13005 {
13006         struct ctl_sg_entry *local_sglist, *remote_sglist;
13007         struct ctl_sg_entry *local_dma_sglist, *remote_dma_sglist;
13008         struct ctl_softc *softc;
13009         int retval;
13010         int i;
13011
13012         retval = 0;
13013         softc = control_softc;
13014
13015         local_sglist = io->io_hdr.local_sglist;
13016         local_dma_sglist = io->io_hdr.local_dma_sglist;
13017         remote_sglist = io->io_hdr.remote_sglist;
13018         remote_dma_sglist = io->io_hdr.remote_dma_sglist;
13019
13020         if (io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) {
13021                 for (i = 0; i < io->scsiio.kern_sg_entries; i++) {
13022                         local_sglist[i].len = remote_sglist[i].len;
13023
13024                         /*
13025                          * XXX Detect the situation where the RS-level I/O
13026                          * redirector on the other side has already read the
13027                          * data off of the AOR RS on this side, and
13028                          * transferred it to remote (mirror) memory on the
13029                          * other side.  Since we already have the data in
13030                          * memory here, we just need to use it.
13031                          *
13032                          * XXX KDM this can probably be removed once we
13033                          * get the cache device code in and take the
13034                          * current AOR implementation out.
13035                          */
13036 #ifdef NEEDTOPORT
13037                         if ((remote_sglist[i].addr >=
13038                              (void *)vtophys(softc->mirr->addr))
13039                          && (remote_sglist[i].addr <
13040                              ((void *)vtophys(softc->mirr->addr) +
13041                              CacheMirrorOffset))) {
13042                                 local_sglist[i].addr = remote_sglist[i].addr -
13043                                         CacheMirrorOffset;
13044                                 if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13045                                      CTL_FLAG_DATA_IN)
13046                                         io->io_hdr.flags |= CTL_FLAG_REDIR_DONE;
13047                         } else {
13048                                 local_sglist[i].addr = remote_sglist[i].addr +
13049                                         CacheMirrorOffset;
13050                         }
13051 #endif
13052 #if 0
13053                         printf("%s: local %p, remote %p, len %d\n",
13054                                __func__, local_sglist[i].addr,
13055                                remote_sglist[i].addr, local_sglist[i].len);
13056 #endif
13057                 }
13058         } else {
13059                 uint32_t len_to_go;
13060
13061                 /*
13062                  * In this case, we don't have automatically allocated
13063                  * memory for this I/O on this controller.  This typically
13064                  * happens with internal CTL I/O -- e.g. inquiry, mode
13065                  * sense, etc.  Anything coming from RAIDCore will have
13066                  * a mirror area available.
13067                  */
13068                 len_to_go = io->scsiio.kern_data_len;
13069
13070                 /*
13071                  * Clear the no datasync flag, we have to use malloced
13072                  * buffers.
13073                  */
13074                 io->io_hdr.flags &= ~CTL_FLAG_NO_DATASYNC;
13075
13076                 /*
13077                  * The difficult thing here is that the size of the various
13078                  * S/G segments may be different than the size from the
13079                  * remote controller.  That'll make it harder when DMAing
13080                  * the data back to the other side.
13081                  */
13082                 for (i = 0; (i < sizeof(io->io_hdr.remote_sglist) /
13083                      sizeof(io->io_hdr.remote_sglist[0])) &&
13084                      (len_to_go > 0); i++) {
13085                         local_sglist[i].len = ctl_min(len_to_go, 131072);
13086                         CTL_SIZE_8B(local_dma_sglist[i].len,
13087                                     local_sglist[i].len);
13088                         local_sglist[i].addr =
13089                                 malloc(local_dma_sglist[i].len, M_CTL,M_WAITOK);
13090
13091                         local_dma_sglist[i].addr = local_sglist[i].addr;
13092
13093                         if (local_sglist[i].addr == NULL) {
13094                                 int j;
13095
13096                                 printf("malloc failed for %zd bytes!",
13097                                        local_dma_sglist[i].len);
13098                                 for (j = 0; j < i; j++) {
13099                                         free(local_sglist[j].addr, M_CTL);
13100                                 }
13101                                 ctl_set_internal_failure(&io->scsiio,
13102                                                          /*sks_valid*/ 1,
13103                                                          /*retry_count*/ 4857);
13104                                 retval = 1;
13105                                 goto bailout_error;
13106                                 
13107                         }
13108                         /* XXX KDM do we need a sync here? */
13109
13110                         len_to_go -= local_sglist[i].len;
13111                 }
13112                 /*
13113                  * Reset the number of S/G entries accordingly.  The
13114                  * original number of S/G entries is available in
13115                  * rem_sg_entries.
13116                  */
13117                 io->scsiio.kern_sg_entries = i;
13118
13119 #if 0
13120                 printf("%s: kern_sg_entries = %d\n", __func__,
13121                        io->scsiio.kern_sg_entries);
13122                 for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13123                         printf("%s: sg[%d] = %p, %d (DMA: %d)\n", __func__, i,
13124                                local_sglist[i].addr, local_sglist[i].len,
13125                                local_dma_sglist[i].len);
13126 #endif
13127         }
13128
13129
13130         return (retval);
13131
13132 bailout_error:
13133
13134         ctl_send_datamove_done(io, /*have_lock*/ 0);
13135
13136         return (retval);
13137 }
13138
13139 static int
13140 ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
13141                          ctl_ha_dt_cb callback)
13142 {
13143         struct ctl_ha_dt_req *rq;
13144         struct ctl_sg_entry *remote_sglist, *local_sglist;
13145         struct ctl_sg_entry *remote_dma_sglist, *local_dma_sglist;
13146         uint32_t local_used, remote_used, total_used;
13147         int retval;
13148         int i, j;
13149
13150         retval = 0;
13151
13152         rq = ctl_dt_req_alloc();
13153
13154         /*
13155          * If we failed to allocate the request, and if the DMA didn't fail
13156          * anyway, set busy status.  This is just a resource allocation
13157          * failure.
13158          */
13159         if ((rq == NULL)
13160          && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE))
13161                 ctl_set_busy(&io->scsiio);
13162
13163         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) {
13164
13165                 if (rq != NULL)
13166                         ctl_dt_req_free(rq);
13167
13168                 /*
13169                  * The data move failed.  We need to return status back
13170                  * to the other controller.  No point in trying to DMA
13171                  * data to the remote controller.
13172                  */
13173
13174                 ctl_send_datamove_done(io, /*have_lock*/ 0);
13175
13176                 retval = 1;
13177
13178                 goto bailout;
13179         }
13180
13181         local_sglist = io->io_hdr.local_sglist;
13182         local_dma_sglist = io->io_hdr.local_dma_sglist;
13183         remote_sglist = io->io_hdr.remote_sglist;
13184         remote_dma_sglist = io->io_hdr.remote_dma_sglist;
13185         local_used = 0;
13186         remote_used = 0;
13187         total_used = 0;
13188
13189         if (io->io_hdr.flags & CTL_FLAG_REDIR_DONE) {
13190                 rq->ret = CTL_HA_STATUS_SUCCESS;
13191                 rq->context = io;
13192                 callback(rq);
13193                 goto bailout;
13194         }
13195
13196         /*
13197          * Pull/push the data over the wire from/to the other controller.
13198          * This takes into account the possibility that the local and
13199          * remote sglists may not be identical in terms of the size of
13200          * the elements and the number of elements.
13201          *
13202          * One fundamental assumption here is that the length allocated for
13203          * both the local and remote sglists is identical.  Otherwise, we've
13204          * essentially got a coding error of some sort.
13205          */
13206         for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
13207                 int isc_ret;
13208                 uint32_t cur_len, dma_length;
13209                 uint8_t *tmp_ptr;
13210
13211                 rq->id = CTL_HA_DATA_CTL;
13212                 rq->command = command;
13213                 rq->context = io;
13214
13215                 /*
13216                  * Both pointers should be aligned.  But it is possible
13217                  * that the allocation length is not.  They should both
13218                  * also have enough slack left over at the end, though,
13219                  * to round up to the next 8 byte boundary.
13220                  */
13221                 cur_len = ctl_min(local_sglist[i].len - local_used,
13222                                   remote_sglist[j].len - remote_used);
13223
13224                 /*
13225                  * In this case, we have a size issue and need to decrease
13226                  * the size, except in the case where we actually have less
13227                  * than 8 bytes left.  In that case, we need to increase
13228                  * the DMA length to get the last bit.
13229                  */
13230                 if ((cur_len & 0x7) != 0) {
13231                         if (cur_len > 0x7) {
13232                                 cur_len = cur_len - (cur_len & 0x7);
13233                                 dma_length = cur_len;
13234                         } else {
13235                                 CTL_SIZE_8B(dma_length, cur_len);
13236                         }
13237
13238                 } else
13239                         dma_length = cur_len;
13240
13241                 /*
13242                  * If we had to allocate memory for this I/O, instead of using
13243                  * the non-cached mirror memory, we'll need to flush the cache
13244                  * before trying to DMA to the other controller.
13245                  *
13246                  * We could end up doing this multiple times for the same
13247                  * segment if we have a larger local segment than remote
13248                  * segment.  That shouldn't be an issue.
13249                  */
13250                 if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
13251                         /*
13252                          * XXX KDM use bus_dmamap_sync() here.
13253                          */
13254                 }
13255
13256                 rq->size = dma_length;
13257
13258                 tmp_ptr = (uint8_t *)local_sglist[i].addr;
13259                 tmp_ptr += local_used;
13260
13261                 /* Use physical addresses when talking to ISC hardware */
13262                 if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
13263                         /* XXX KDM use busdma */
13264 #if 0
13265                         rq->local = vtophys(tmp_ptr);
13266 #endif
13267                 } else
13268                         rq->local = tmp_ptr;
13269
13270                 tmp_ptr = (uint8_t *)remote_sglist[j].addr;
13271                 tmp_ptr += remote_used;
13272                 rq->remote = tmp_ptr;
13273
13274                 rq->callback = NULL;
13275
13276                 local_used += cur_len;
13277                 if (local_used >= local_sglist[i].len) {
13278                         i++;
13279                         local_used = 0;
13280                 }
13281
13282                 remote_used += cur_len;
13283                 if (remote_used >= remote_sglist[j].len) {
13284                         j++;
13285                         remote_used = 0;
13286                 }
13287                 total_used += cur_len;
13288
13289                 if (total_used >= io->scsiio.kern_data_len)
13290                         rq->callback = callback;
13291
13292                 if ((rq->size & 0x7) != 0) {
13293                         printf("%s: warning: size %d is not on 8b boundary\n",
13294                                __func__, rq->size);
13295                 }
13296                 if (((uintptr_t)rq->local & 0x7) != 0) {
13297                         printf("%s: warning: local %p not on 8b boundary\n",
13298                                __func__, rq->local);
13299                 }
13300                 if (((uintptr_t)rq->remote & 0x7) != 0) {
13301                         printf("%s: warning: remote %p not on 8b boundary\n",
13302                                __func__, rq->local);
13303                 }
13304 #if 0
13305                 printf("%s: %s: local %#x remote %#x size %d\n", __func__,
13306                        (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
13307                        rq->local, rq->remote, rq->size);
13308 #endif
13309
13310                 isc_ret = ctl_dt_single(rq);
13311                 if (isc_ret == CTL_HA_STATUS_WAIT)
13312                         continue;
13313
13314                 if (isc_ret == CTL_HA_STATUS_DISCONNECT) {
13315                         rq->ret = CTL_HA_STATUS_SUCCESS;
13316                 } else {
13317                         rq->ret = isc_ret;
13318                 }
13319                 callback(rq);
13320                 goto bailout;
13321         }
13322
13323 bailout:
13324         return (retval);
13325
13326 }
13327
13328 static void
13329 ctl_datamove_remote_read(union ctl_io *io)
13330 {
13331         int retval;
13332         int i;
13333
13334         /*
13335          * This will send an error to the other controller in the case of a
13336          * failure.
13337          */
13338         retval = ctl_datamove_remote_sgl_setup(io);
13339         if (retval != 0)
13340                 return;
13341
13342         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
13343                                           ctl_datamove_remote_read_cb);
13344         if ((retval != 0)
13345          && ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0)) {
13346                 /*
13347                  * Make sure we free memory if there was an error..  The
13348                  * ctl_datamove_remote_xfer() function will send the
13349                  * datamove done message, or call the callback with an
13350                  * error if there is a problem.
13351                  */
13352                 for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13353                         free(io->io_hdr.local_sglist[i].addr, M_CTL);
13354         }
13355
13356         return;
13357 }
13358
13359 /*
13360  * Process a datamove request from the other controller.  This is used for
13361  * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
13362  * first.  Once that is complete, the data gets DMAed into the remote
13363  * controller's memory.  For reads, we DMA from the remote controller's
13364  * memory into our memory first, and then move it out to the FETD.
13365  */
13366 static void
13367 ctl_datamove_remote(union ctl_io *io)
13368 {
13369         struct ctl_softc *softc;
13370
13371         softc = control_softc;
13372
13373         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
13374
13375         /*
13376          * Note that we look for an aborted I/O here, but don't do some of
13377          * the other checks that ctl_datamove() normally does.
13378          * We don't need to run the datamove delay code, since that should
13379          * have been done if need be on the other controller.
13380          */
13381         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
13382                 printf("%s: tag 0x%04x on (%d:%d:%d:%d) aborted\n", __func__,
13383                        io->scsiio.tag_num, io->io_hdr.nexus.initid.id,
13384                        io->io_hdr.nexus.targ_port,
13385                        io->io_hdr.nexus.targ_target.id,
13386                        io->io_hdr.nexus.targ_lun);
13387                 io->io_hdr.port_status = 31338;
13388                 ctl_send_datamove_done(io, /*have_lock*/ 0);
13389                 return;
13390         }
13391
13392         if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT) {
13393                 ctl_datamove_remote_write(io);
13394         } else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN){
13395                 ctl_datamove_remote_read(io);
13396         } else {
13397                 union ctl_ha_msg msg;
13398                 struct scsi_sense_data *sense;
13399                 uint8_t sks[3];
13400                 int retry_count;
13401
13402                 memset(&msg, 0, sizeof(msg));
13403
13404                 msg.hdr.msg_type = CTL_MSG_BAD_JUJU;
13405                 msg.hdr.status = CTL_SCSI_ERROR;
13406                 msg.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
13407
13408                 retry_count = 4243;
13409
13410                 sense = &msg.scsi.sense_data;
13411                 sks[0] = SSD_SCS_VALID;
13412                 sks[1] = (retry_count >> 8) & 0xff;
13413                 sks[2] = retry_count & 0xff;
13414
13415                 /* "Internal target failure" */
13416                 scsi_set_sense_data(sense,
13417                                     /*sense_format*/ SSD_TYPE_NONE,
13418                                     /*current_error*/ 1,
13419                                     /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
13420                                     /*asc*/ 0x44,
13421                                     /*ascq*/ 0x00,
13422                                     /*type*/ SSD_ELEM_SKS,
13423                                     /*size*/ sizeof(sks),
13424                                     /*data*/ sks,
13425                                     SSD_ELEM_NONE);
13426
13427                 io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
13428                 if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
13429                         ctl_failover_io(io, /*have_lock*/ 1);
13430                         return;
13431                 }
13432
13433                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0) >
13434                     CTL_HA_STATUS_SUCCESS) {
13435                         /* XXX KDM what to do if this fails? */
13436                 }
13437                 return;
13438         }
13439         
13440 }
13441
13442 static int
13443 ctl_process_done(union ctl_io *io)
13444 {
13445         struct ctl_lun *lun;
13446         struct ctl_softc *ctl_softc;
13447         void (*fe_done)(union ctl_io *io);
13448         uint32_t targ_port = ctl_port_idx(io->io_hdr.nexus.targ_port);
13449
13450         CTL_DEBUG_PRINT(("ctl_process_done\n"));
13451
13452         fe_done =
13453             control_softc->ctl_ports[targ_port]->fe_done;
13454
13455 #ifdef CTL_TIME_IO
13456         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
13457                 char str[256];
13458                 char path_str[64];
13459                 struct sbuf sb;
13460
13461                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
13462                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13463
13464                 sbuf_cat(&sb, path_str);
13465                 switch (io->io_hdr.io_type) {
13466                 case CTL_IO_SCSI:
13467                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
13468                         sbuf_printf(&sb, "\n");
13469                         sbuf_cat(&sb, path_str);
13470                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13471                                     io->scsiio.tag_num, io->scsiio.tag_type);
13472                         break;
13473                 case CTL_IO_TASK:
13474                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
13475                                     "Tag Type: %d\n", io->taskio.task_action,
13476                                     io->taskio.tag_num, io->taskio.tag_type);
13477                         break;
13478                 default:
13479                         printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13480                         panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13481                         break;
13482                 }
13483                 sbuf_cat(&sb, path_str);
13484                 sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
13485                             (intmax_t)time_uptime - io->io_hdr.start_time);
13486                 sbuf_finish(&sb);
13487                 printf("%s", sbuf_data(&sb));
13488         }
13489 #endif /* CTL_TIME_IO */
13490
13491         switch (io->io_hdr.io_type) {
13492         case CTL_IO_SCSI:
13493                 break;
13494         case CTL_IO_TASK:
13495                 if (bootverbose || verbose > 0)
13496                         ctl_io_error_print(io, NULL);
13497                 if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
13498                         ctl_free_io(io);
13499                 else
13500                         fe_done(io);
13501                 return (CTL_RETVAL_COMPLETE);
13502                 break;
13503         default:
13504                 printf("ctl_process_done: invalid io type %d\n",
13505                        io->io_hdr.io_type);
13506                 panic("ctl_process_done: invalid io type %d\n",
13507                       io->io_hdr.io_type);
13508                 break; /* NOTREACHED */
13509         }
13510
13511         lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13512         if (lun == NULL) {
13513                 CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
13514                                  io->io_hdr.nexus.targ_mapped_lun));
13515                 fe_done(io);
13516                 goto bailout;
13517         }
13518         ctl_softc = lun->ctl_softc;
13519
13520         mtx_lock(&lun->lun_lock);
13521
13522         /*
13523          * Check to see if we have any errors to inject here.  We only
13524          * inject errors for commands that don't already have errors set.
13525          */
13526         if ((STAILQ_FIRST(&lun->error_list) != NULL)
13527          && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))
13528                 ctl_inject_error(lun, io);
13529
13530         /*
13531          * XXX KDM how do we treat commands that aren't completed
13532          * successfully?
13533          *
13534          * XXX KDM should we also track I/O latency?
13535          */
13536         if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
13537             io->io_hdr.io_type == CTL_IO_SCSI) {
13538 #ifdef CTL_TIME_IO
13539                 struct bintime cur_bt;
13540 #endif
13541                 int type;
13542
13543                 if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13544                     CTL_FLAG_DATA_IN)
13545                         type = CTL_STATS_READ;
13546                 else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13547                     CTL_FLAG_DATA_OUT)
13548                         type = CTL_STATS_WRITE;
13549                 else
13550                         type = CTL_STATS_NO_IO;
13551
13552                 lun->stats.ports[targ_port].bytes[type] +=
13553                     io->scsiio.kern_total_len;
13554                 lun->stats.ports[targ_port].operations[type]++;
13555 #ifdef CTL_TIME_IO
13556                 bintime_add(&lun->stats.ports[targ_port].dma_time[type],
13557                    &io->io_hdr.dma_bt);
13558                 lun->stats.ports[targ_port].num_dmas[type] +=
13559                     io->io_hdr.num_dmas;
13560                 getbintime(&cur_bt);
13561                 bintime_sub(&cur_bt, &io->io_hdr.start_bt);
13562                 bintime_add(&lun->stats.ports[targ_port].time[type], &cur_bt);
13563 #endif
13564         }
13565
13566         /*
13567          * Remove this from the OOA queue.
13568          */
13569         TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13570
13571         /*
13572          * Run through the blocked queue on this LUN and see if anything
13573          * has become unblocked, now that this transaction is done.
13574          */
13575         ctl_check_blocked(lun);
13576
13577         /*
13578          * If the LUN has been invalidated, free it if there is nothing
13579          * left on its OOA queue.
13580          */
13581         if ((lun->flags & CTL_LUN_INVALID)
13582          && TAILQ_EMPTY(&lun->ooa_queue)) {
13583                 mtx_unlock(&lun->lun_lock);
13584                 mtx_lock(&ctl_softc->ctl_lock);
13585                 ctl_free_lun(lun);
13586                 mtx_unlock(&ctl_softc->ctl_lock);
13587         } else
13588                 mtx_unlock(&lun->lun_lock);
13589
13590         /*
13591          * If this command has been aborted, make sure we set the status
13592          * properly.  The FETD is responsible for freeing the I/O and doing
13593          * whatever it needs to do to clean up its state.
13594          */
13595         if (io->io_hdr.flags & CTL_FLAG_ABORT)
13596                 ctl_set_task_aborted(&io->scsiio);
13597
13598         /*
13599          * We print out status for every task management command.  For SCSI
13600          * commands, we filter out any unit attention errors; they happen
13601          * on every boot, and would clutter up the log.  Note:  task
13602          * management commands aren't printed here, they are printed above,
13603          * since they should never even make it down here.
13604          */
13605         switch (io->io_hdr.io_type) {
13606         case CTL_IO_SCSI: {
13607                 int error_code, sense_key, asc, ascq;
13608
13609                 sense_key = 0;
13610
13611                 if (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SCSI_ERROR)
13612                  && (io->scsiio.scsi_status == SCSI_STATUS_CHECK_COND)) {
13613                         /*
13614                          * Since this is just for printing, no need to
13615                          * show errors here.
13616                          */
13617                         scsi_extract_sense_len(&io->scsiio.sense_data,
13618                                                io->scsiio.sense_len,
13619                                                &error_code,
13620                                                &sense_key,
13621                                                &asc,
13622                                                &ascq,
13623                                                /*show_errors*/ 0);
13624                 }
13625
13626                 if (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
13627                  && (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SCSI_ERROR)
13628                   || (io->scsiio.scsi_status != SCSI_STATUS_CHECK_COND)
13629                   || (sense_key != SSD_KEY_UNIT_ATTENTION))) {
13630
13631                         if ((time_uptime - ctl_softc->last_print_jiffies) <= 0){
13632                                 ctl_softc->skipped_prints++;
13633                         } else {
13634                                 uint32_t skipped_prints;
13635
13636                                 skipped_prints = ctl_softc->skipped_prints;
13637
13638                                 ctl_softc->skipped_prints = 0;
13639                                 ctl_softc->last_print_jiffies = time_uptime;
13640
13641                                 if (skipped_prints > 0) {
13642 #ifdef NEEDTOPORT
13643                                         csevent_log(CSC_CTL | CSC_SHELF_SW |
13644                                             CTL_ERROR_REPORT,
13645                                             csevent_LogType_Trace,
13646                                             csevent_Severity_Information,
13647                                             csevent_AlertLevel_Green,
13648                                             csevent_FRU_Firmware,
13649                                             csevent_FRU_Unknown,
13650                                             "High CTL error volume, %d prints "
13651                                             "skipped", skipped_prints);
13652 #endif
13653                                 }
13654                                 if (bootverbose || verbose > 0)
13655                                         ctl_io_error_print(io, NULL);
13656                         }
13657                 }
13658                 break;
13659         }
13660         case CTL_IO_TASK:
13661                 if (bootverbose || verbose > 0)
13662                         ctl_io_error_print(io, NULL);
13663                 break;
13664         default:
13665                 break;
13666         }
13667
13668         /*
13669          * Tell the FETD or the other shelf controller we're done with this
13670          * command.  Note that only SCSI commands get to this point.  Task
13671          * management commands are completed above.
13672          *
13673          * We only send status to the other controller if we're in XFER
13674          * mode.  In SER_ONLY mode, the I/O is done on the controller that
13675          * received the I/O (from CTL's perspective), and so the status is
13676          * generated there.
13677          * 
13678          * XXX KDM if we hold the lock here, we could cause a deadlock
13679          * if the frontend comes back in in this context to queue
13680          * something.
13681          */
13682         if ((ctl_softc->ha_mode == CTL_HA_MODE_XFER)
13683          && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
13684                 union ctl_ha_msg msg;
13685
13686                 memset(&msg, 0, sizeof(msg));
13687                 msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13688                 msg.hdr.original_sc = io->io_hdr.original_sc;
13689                 msg.hdr.nexus = io->io_hdr.nexus;
13690                 msg.hdr.status = io->io_hdr.status;
13691                 msg.scsi.scsi_status = io->scsiio.scsi_status;
13692                 msg.scsi.tag_num = io->scsiio.tag_num;
13693                 msg.scsi.tag_type = io->scsiio.tag_type;
13694                 msg.scsi.sense_len = io->scsiio.sense_len;
13695                 msg.scsi.sense_residual = io->scsiio.sense_residual;
13696                 msg.scsi.residual = io->scsiio.residual;
13697                 memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
13698                        sizeof(io->scsiio.sense_data));
13699                 /*
13700                  * We copy this whether or not this is an I/O-related
13701                  * command.  Otherwise, we'd have to go and check to see
13702                  * whether it's a read/write command, and it really isn't
13703                  * worth it.
13704                  */
13705                 memcpy(&msg.scsi.lbalen,
13706                        &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
13707                        sizeof(msg.scsi.lbalen));
13708
13709                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13710                                 sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
13711                         /* XXX do something here */
13712                 }
13713
13714                 ctl_free_io(io);
13715         } else 
13716                 fe_done(io);
13717
13718 bailout:
13719
13720         return (CTL_RETVAL_COMPLETE);
13721 }
13722
13723 #ifdef CTL_WITH_CA
13724 /*
13725  * Front end should call this if it doesn't do autosense.  When the request
13726  * sense comes back in from the initiator, we'll dequeue this and send it.
13727  */
13728 int
13729 ctl_queue_sense(union ctl_io *io)
13730 {
13731         struct ctl_lun *lun;
13732         struct ctl_softc *ctl_softc;
13733         uint32_t initidx, targ_lun;
13734
13735         ctl_softc = control_softc;
13736
13737         CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13738
13739         /*
13740          * LUN lookup will likely move to the ctl_work_thread() once we
13741          * have our new queueing infrastructure (that doesn't put things on
13742          * a per-LUN queue initially).  That is so that we can handle
13743          * things like an INQUIRY to a LUN that we don't have enabled.  We
13744          * can't deal with that right now.
13745          */
13746         mtx_lock(&ctl_softc->ctl_lock);
13747
13748         /*
13749          * If we don't have a LUN for this, just toss the sense
13750          * information.
13751          */
13752         targ_lun = io->io_hdr.nexus.targ_lun;
13753         targ_lun = ctl_map_lun(io->io_hdr.nexus.targ_port, targ_lun);
13754         if ((targ_lun < CTL_MAX_LUNS)
13755          && (ctl_softc->ctl_luns[targ_lun] != NULL))
13756                 lun = ctl_softc->ctl_luns[targ_lun];
13757         else
13758                 goto bailout;
13759
13760         initidx = ctl_get_initindex(&io->io_hdr.nexus);
13761
13762         mtx_lock(&lun->lun_lock);
13763         /*
13764          * Already have CA set for this LUN...toss the sense information.
13765          */
13766         if (ctl_is_set(lun->have_ca, initidx)) {
13767                 mtx_unlock(&lun->lun_lock);
13768                 goto bailout;
13769         }
13770
13771         memcpy(&lun->pending_sense[initidx], &io->scsiio.sense_data,
13772                ctl_min(sizeof(lun->pending_sense[initidx]),
13773                sizeof(io->scsiio.sense_data)));
13774         ctl_set_mask(lun->have_ca, initidx);
13775         mtx_unlock(&lun->lun_lock);
13776
13777 bailout:
13778         mtx_unlock(&ctl_softc->ctl_lock);
13779
13780         ctl_free_io(io);
13781
13782         return (CTL_RETVAL_COMPLETE);
13783 }
13784 #endif
13785
13786 /*
13787  * Primary command inlet from frontend ports.  All SCSI and task I/O
13788  * requests must go through this function.
13789  */
13790 int
13791 ctl_queue(union ctl_io *io)
13792 {
13793         struct ctl_softc *ctl_softc;
13794
13795         CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13796
13797         ctl_softc = control_softc;
13798
13799 #ifdef CTL_TIME_IO
13800         io->io_hdr.start_time = time_uptime;
13801         getbintime(&io->io_hdr.start_bt);
13802 #endif /* CTL_TIME_IO */
13803
13804         /* Map FE-specific LUN ID into global one. */
13805         io->io_hdr.nexus.targ_mapped_lun =
13806             ctl_map_lun(io->io_hdr.nexus.targ_port, io->io_hdr.nexus.targ_lun);
13807
13808         switch (io->io_hdr.io_type) {
13809         case CTL_IO_SCSI:
13810         case CTL_IO_TASK:
13811                 ctl_enqueue_incoming(io);
13812                 break;
13813         default:
13814                 printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13815                 return (EINVAL);
13816         }
13817
13818         return (CTL_RETVAL_COMPLETE);
13819 }
13820
13821 #ifdef CTL_IO_DELAY
13822 static void
13823 ctl_done_timer_wakeup(void *arg)
13824 {
13825         union ctl_io *io;
13826
13827         io = (union ctl_io *)arg;
13828         ctl_done(io);
13829 }
13830 #endif /* CTL_IO_DELAY */
13831
13832 void
13833 ctl_done(union ctl_io *io)
13834 {
13835         struct ctl_softc *ctl_softc;
13836
13837         ctl_softc = control_softc;
13838
13839         /*
13840          * Enable this to catch duplicate completion issues.
13841          */
13842 #if 0
13843         if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13844                 printf("%s: type %d msg %d cdb %x iptl: "
13845                        "%d:%d:%d:%d tag 0x%04x "
13846                        "flag %#x status %x\n",
13847                         __func__,
13848                         io->io_hdr.io_type,
13849                         io->io_hdr.msg_type,
13850                         io->scsiio.cdb[0],
13851                         io->io_hdr.nexus.initid.id,
13852                         io->io_hdr.nexus.targ_port,
13853                         io->io_hdr.nexus.targ_target.id,
13854                         io->io_hdr.nexus.targ_lun,
13855                         (io->io_hdr.io_type ==
13856                         CTL_IO_TASK) ?
13857                         io->taskio.tag_num :
13858                         io->scsiio.tag_num,
13859                         io->io_hdr.flags,
13860                         io->io_hdr.status);
13861         } else
13862                 io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13863 #endif
13864
13865         /*
13866          * This is an internal copy of an I/O, and should not go through
13867          * the normal done processing logic.
13868          */
13869         if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
13870                 return;
13871
13872         /*
13873          * We need to send a msg to the serializing shelf to finish the IO
13874          * as well.  We don't send a finish message to the other shelf if
13875          * this is a task management command.  Task management commands
13876          * aren't serialized in the OOA queue, but rather just executed on
13877          * both shelf controllers for commands that originated on that
13878          * controller.
13879          */
13880         if ((io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)
13881          && (io->io_hdr.io_type != CTL_IO_TASK)) {
13882                 union ctl_ha_msg msg_io;
13883
13884                 msg_io.hdr.msg_type = CTL_MSG_FINISH_IO;
13885                 msg_io.hdr.serializing_sc = io->io_hdr.serializing_sc;
13886                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_io,
13887                     sizeof(msg_io), 0 ) != CTL_HA_STATUS_SUCCESS) {
13888                 }
13889                 /* continue on to finish IO */
13890         }
13891 #ifdef CTL_IO_DELAY
13892         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13893                 struct ctl_lun *lun;
13894
13895                 lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13896
13897                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13898         } else {
13899                 struct ctl_lun *lun;
13900
13901                 lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13902
13903                 if ((lun != NULL)
13904                  && (lun->delay_info.done_delay > 0)) {
13905                         struct callout *callout;
13906
13907                         callout = (struct callout *)&io->io_hdr.timer_bytes;
13908                         callout_init(callout, /*mpsafe*/ 1);
13909                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13910                         callout_reset(callout,
13911                                       lun->delay_info.done_delay * hz,
13912                                       ctl_done_timer_wakeup, io);
13913                         if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13914                                 lun->delay_info.done_delay = 0;
13915                         return;
13916                 }
13917         }
13918 #endif /* CTL_IO_DELAY */
13919
13920         ctl_enqueue_done(io);
13921 }
13922
13923 int
13924 ctl_isc(struct ctl_scsiio *ctsio)
13925 {
13926         struct ctl_lun *lun;
13927         int retval;
13928
13929         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13930
13931         CTL_DEBUG_PRINT(("ctl_isc: command: %02x\n", ctsio->cdb[0]));
13932
13933         CTL_DEBUG_PRINT(("ctl_isc: calling data_submit()\n"));
13934
13935         retval = lun->backend->data_submit((union ctl_io *)ctsio);
13936
13937         return (retval);
13938 }
13939
13940
13941 static void
13942 ctl_work_thread(void *arg)
13943 {
13944         struct ctl_thread *thr = (struct ctl_thread *)arg;
13945         struct ctl_softc *softc = thr->ctl_softc;
13946         union ctl_io *io;
13947         int retval;
13948
13949         CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13950
13951         for (;;) {
13952                 retval = 0;
13953
13954                 /*
13955                  * We handle the queues in this order:
13956                  * - ISC
13957                  * - done queue (to free up resources, unblock other commands)
13958                  * - RtR queue
13959                  * - incoming queue
13960                  *
13961                  * If those queues are empty, we break out of the loop and
13962                  * go to sleep.
13963                  */
13964                 mtx_lock(&thr->queue_lock);
13965                 io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
13966                 if (io != NULL) {
13967                         STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
13968                         mtx_unlock(&thr->queue_lock);
13969                         ctl_handle_isc(io);
13970                         continue;
13971                 }
13972                 io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
13973                 if (io != NULL) {
13974                         STAILQ_REMOVE_HEAD(&thr->done_queue, links);
13975                         /* clear any blocked commands, call fe_done */
13976                         mtx_unlock(&thr->queue_lock);
13977                         retval = ctl_process_done(io);
13978                         continue;
13979                 }
13980                 io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
13981                 if (io != NULL) {
13982                         STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
13983                         mtx_unlock(&thr->queue_lock);
13984                         if (io->io_hdr.io_type == CTL_IO_TASK)
13985                                 ctl_run_task(io);
13986                         else
13987                                 ctl_scsiio_precheck(softc, &io->scsiio);
13988                         continue;
13989                 }
13990                 if (!ctl_pause_rtr) {
13991                         io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
13992                         if (io != NULL) {
13993                                 STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
13994                                 mtx_unlock(&thr->queue_lock);
13995                                 retval = ctl_scsiio(&io->scsiio);
13996                                 if (retval != CTL_RETVAL_COMPLETE)
13997                                         CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13998                                 continue;
13999                         }
14000                 }
14001
14002                 /* Sleep until we have something to do. */
14003                 mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
14004         }
14005 }
14006
14007 static void
14008 ctl_lun_thread(void *arg)
14009 {
14010         struct ctl_softc *softc = (struct ctl_softc *)arg;
14011         struct ctl_be_lun *be_lun;
14012         int retval;
14013
14014         CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
14015
14016         for (;;) {
14017                 retval = 0;
14018                 mtx_lock(&softc->ctl_lock);
14019                 be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
14020                 if (be_lun != NULL) {
14021                         STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
14022                         mtx_unlock(&softc->ctl_lock);
14023                         ctl_create_lun(be_lun);
14024                         continue;
14025                 }
14026
14027                 /* Sleep until we have something to do. */
14028                 mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
14029                     PDROP | PRIBIO, "-", 0);
14030         }
14031 }
14032
14033 static void
14034 ctl_enqueue_incoming(union ctl_io *io)
14035 {
14036         struct ctl_softc *softc = control_softc;
14037         struct ctl_thread *thr;
14038         u_int idx;
14039
14040         idx = (io->io_hdr.nexus.targ_port * 127 +
14041                io->io_hdr.nexus.initid.id) % worker_threads;
14042         thr = &softc->threads[idx];
14043         mtx_lock(&thr->queue_lock);
14044         STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
14045         mtx_unlock(&thr->queue_lock);
14046         wakeup(thr);
14047 }
14048
14049 static void
14050 ctl_enqueue_rtr(union ctl_io *io)
14051 {
14052         struct ctl_softc *softc = control_softc;
14053         struct ctl_thread *thr;
14054
14055         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14056         mtx_lock(&thr->queue_lock);
14057         STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
14058         mtx_unlock(&thr->queue_lock);
14059         wakeup(thr);
14060 }
14061
14062 static void
14063 ctl_enqueue_done(union ctl_io *io)
14064 {
14065         struct ctl_softc *softc = control_softc;
14066         struct ctl_thread *thr;
14067
14068         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14069         mtx_lock(&thr->queue_lock);
14070         STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
14071         mtx_unlock(&thr->queue_lock);
14072         wakeup(thr);
14073 }
14074
14075 static void
14076 ctl_enqueue_isc(union ctl_io *io)
14077 {
14078         struct ctl_softc *softc = control_softc;
14079         struct ctl_thread *thr;
14080
14081         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14082         mtx_lock(&thr->queue_lock);
14083         STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
14084         mtx_unlock(&thr->queue_lock);
14085         wakeup(thr);
14086 }
14087
14088 /* Initialization and failover */
14089
14090 void
14091 ctl_init_isc_msg(void)
14092 {
14093         printf("CTL: Still calling this thing\n");
14094 }
14095
14096 /*
14097  * Init component
14098  *      Initializes component into configuration defined by bootMode
14099  *      (see hasc-sv.c)
14100  *      returns hasc_Status:
14101  *              OK
14102  *              ERROR - fatal error
14103  */
14104 static ctl_ha_comp_status
14105 ctl_isc_init(struct ctl_ha_component *c)
14106 {
14107         ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
14108
14109         c->status = ret;
14110         return ret;
14111 }
14112
14113 /* Start component
14114  *      Starts component in state requested. If component starts successfully,
14115  *      it must set its own state to the requestrd state
14116  *      When requested state is HASC_STATE_HA, the component may refine it
14117  *      by adding _SLAVE or _MASTER flags.
14118  *      Currently allowed state transitions are:
14119  *      UNKNOWN->HA             - initial startup
14120  *      UNKNOWN->SINGLE - initial startup when no parter detected
14121  *      HA->SINGLE              - failover
14122  * returns ctl_ha_comp_status:
14123  *              OK      - component successfully started in requested state
14124  *              FAILED  - could not start the requested state, failover may
14125  *                        be possible
14126  *              ERROR   - fatal error detected, no future startup possible
14127  */
14128 static ctl_ha_comp_status
14129 ctl_isc_start(struct ctl_ha_component *c, ctl_ha_state state)
14130 {
14131         ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
14132
14133         printf("%s: go\n", __func__);
14134
14135         // UNKNOWN->HA or UNKNOWN->SINGLE (bootstrap)
14136         if (c->state == CTL_HA_STATE_UNKNOWN ) {
14137                 ctl_is_single = 0;
14138                 if (ctl_ha_msg_create(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
14139                     != CTL_HA_STATUS_SUCCESS) {
14140                         printf("ctl_isc_start: ctl_ha_msg_create failed.\n");
14141                         ret = CTL_HA_COMP_STATUS_ERROR;
14142                 }
14143         } else if (CTL_HA_STATE_IS_HA(c->state)
14144                 && CTL_HA_STATE_IS_SINGLE(state)){
14145                 // HA->SINGLE transition
14146                 ctl_failover();
14147                 ctl_is_single = 1;
14148         } else {
14149                 printf("ctl_isc_start:Invalid state transition %X->%X\n",
14150                        c->state, state);
14151                 ret = CTL_HA_COMP_STATUS_ERROR;
14152         }
14153         if (CTL_HA_STATE_IS_SINGLE(state))
14154                 ctl_is_single = 1;
14155
14156         c->state = state;
14157         c->status = ret;
14158         return ret;
14159 }
14160
14161 /*
14162  * Quiesce component
14163  * The component must clear any error conditions (set status to OK) and
14164  * prepare itself to another Start call
14165  * returns ctl_ha_comp_status:
14166  *      OK
14167  *      ERROR
14168  */
14169 static ctl_ha_comp_status
14170 ctl_isc_quiesce(struct ctl_ha_component *c)
14171 {
14172         int ret = CTL_HA_COMP_STATUS_OK;
14173
14174         ctl_pause_rtr = 1;
14175         c->status = ret;
14176         return ret;
14177 }
14178
14179 struct ctl_ha_component ctl_ha_component_ctlisc =
14180 {
14181         .name = "CTL ISC",
14182         .state = CTL_HA_STATE_UNKNOWN,
14183         .init = ctl_isc_init,
14184         .start = ctl_isc_start,
14185         .quiesce = ctl_isc_quiesce
14186 };
14187
14188 /*
14189  *  vim: ts=8
14190  */