]> CyberLeo.Net >> Repos - FreeBSD/stable/10.git/blob - sys/cam/ctl/ctl.c
MFC r288170: Add new report types to REPORT LUNS command.
[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  * Copyright (c) 2015 Alexander Motin <mav@FreeBSD.org>
5  * All rights reserved.
6  *
7  * Portions of this software were developed by Edward Tomasz Napierala
8  * under sponsorship from the FreeBSD Foundation.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions, and the following disclaimer,
15  *    without modification.
16  * 2. Redistributions in binary form must reproduce at minimum a disclaimer
17  *    substantially similar to the "NO WARRANTY" disclaimer below
18  *    ("Disclaimer") and any redistribution must be conditioned upon
19  *    including a substantially similar Disclaimer requirement for further
20  *    binary redistribution.
21  *
22  * NO WARRANTY
23  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
24  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
25  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR
26  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
27  * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
31  * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
32  * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
33  * POSSIBILITY OF SUCH DAMAGES.
34  *
35  * $Id$
36  */
37 /*
38  * CAM Target Layer, a SCSI device emulation subsystem.
39  *
40  * Author: Ken Merry <ken@FreeBSD.org>
41  */
42
43 #define _CTL_C
44
45 #include <sys/cdefs.h>
46 __FBSDID("$FreeBSD$");
47
48 #include <sys/param.h>
49 #include <sys/systm.h>
50 #include <sys/ctype.h>
51 #include <sys/kernel.h>
52 #include <sys/types.h>
53 #include <sys/kthread.h>
54 #include <sys/bio.h>
55 #include <sys/fcntl.h>
56 #include <sys/lock.h>
57 #include <sys/module.h>
58 #include <sys/mutex.h>
59 #include <sys/condvar.h>
60 #include <sys/malloc.h>
61 #include <sys/conf.h>
62 #include <sys/ioccom.h>
63 #include <sys/queue.h>
64 #include <sys/sbuf.h>
65 #include <sys/smp.h>
66 #include <sys/endian.h>
67 #include <sys/sysctl.h>
68 #include <vm/uma.h>
69
70 #include <cam/cam.h>
71 #include <cam/scsi/scsi_all.h>
72 #include <cam/scsi/scsi_da.h>
73 #include <cam/ctl/ctl_io.h>
74 #include <cam/ctl/ctl.h>
75 #include <cam/ctl/ctl_frontend.h>
76 #include <cam/ctl/ctl_util.h>
77 #include <cam/ctl/ctl_backend.h>
78 #include <cam/ctl/ctl_ioctl.h>
79 #include <cam/ctl/ctl_ha.h>
80 #include <cam/ctl/ctl_private.h>
81 #include <cam/ctl/ctl_debug.h>
82 #include <cam/ctl/ctl_scsi_all.h>
83 #include <cam/ctl/ctl_error.h>
84
85 struct ctl_softc *control_softc = NULL;
86
87 /*
88  * Template mode pages.
89  */
90
91 /*
92  * Note that these are default values only.  The actual values will be
93  * filled in when the user does a mode sense.
94  */
95 const static struct copan_debugconf_subpage debugconf_page_default = {
96         DBGCNF_PAGE_CODE | SMPH_SPF,    /* page_code */
97         DBGCNF_SUBPAGE_CODE,            /* subpage */
98         {(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
99          (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
100         DBGCNF_VERSION,                 /* page_version */
101         {CTL_TIME_IO_DEFAULT_SECS>>8,
102          CTL_TIME_IO_DEFAULT_SECS>>0},  /* ctl_time_io_secs */
103 };
104
105 const static struct copan_debugconf_subpage debugconf_page_changeable = {
106         DBGCNF_PAGE_CODE | SMPH_SPF,    /* page_code */
107         DBGCNF_SUBPAGE_CODE,            /* subpage */
108         {(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
109          (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
110         0,                              /* page_version */
111         {0xff,0xff},                    /* ctl_time_io_secs */
112 };
113
114 const static struct scsi_da_rw_recovery_page rw_er_page_default = {
115         /*page_code*/SMS_RW_ERROR_RECOVERY_PAGE,
116         /*page_length*/sizeof(struct scsi_da_rw_recovery_page) - 2,
117         /*byte3*/SMS_RWER_AWRE|SMS_RWER_ARRE,
118         /*read_retry_count*/0,
119         /*correction_span*/0,
120         /*head_offset_count*/0,
121         /*data_strobe_offset_cnt*/0,
122         /*byte8*/SMS_RWER_LBPERE,
123         /*write_retry_count*/0,
124         /*reserved2*/0,
125         /*recovery_time_limit*/{0, 0},
126 };
127
128 const static struct scsi_da_rw_recovery_page rw_er_page_changeable = {
129         /*page_code*/SMS_RW_ERROR_RECOVERY_PAGE,
130         /*page_length*/sizeof(struct scsi_da_rw_recovery_page) - 2,
131         /*byte3*/0,
132         /*read_retry_count*/0,
133         /*correction_span*/0,
134         /*head_offset_count*/0,
135         /*data_strobe_offset_cnt*/0,
136         /*byte8*/0,
137         /*write_retry_count*/0,
138         /*reserved2*/0,
139         /*recovery_time_limit*/{0, 0},
140 };
141
142 const static struct scsi_format_page format_page_default = {
143         /*page_code*/SMS_FORMAT_DEVICE_PAGE,
144         /*page_length*/sizeof(struct scsi_format_page) - 2,
145         /*tracks_per_zone*/ {0, 0},
146         /*alt_sectors_per_zone*/ {0, 0},
147         /*alt_tracks_per_zone*/ {0, 0},
148         /*alt_tracks_per_lun*/ {0, 0},
149         /*sectors_per_track*/ {(CTL_DEFAULT_SECTORS_PER_TRACK >> 8) & 0xff,
150                                 CTL_DEFAULT_SECTORS_PER_TRACK & 0xff},
151         /*bytes_per_sector*/ {0, 0},
152         /*interleave*/ {0, 0},
153         /*track_skew*/ {0, 0},
154         /*cylinder_skew*/ {0, 0},
155         /*flags*/ SFP_HSEC,
156         /*reserved*/ {0, 0, 0}
157 };
158
159 const static struct scsi_format_page format_page_changeable = {
160         /*page_code*/SMS_FORMAT_DEVICE_PAGE,
161         /*page_length*/sizeof(struct scsi_format_page) - 2,
162         /*tracks_per_zone*/ {0, 0},
163         /*alt_sectors_per_zone*/ {0, 0},
164         /*alt_tracks_per_zone*/ {0, 0},
165         /*alt_tracks_per_lun*/ {0, 0},
166         /*sectors_per_track*/ {0, 0},
167         /*bytes_per_sector*/ {0, 0},
168         /*interleave*/ {0, 0},
169         /*track_skew*/ {0, 0},
170         /*cylinder_skew*/ {0, 0},
171         /*flags*/ 0,
172         /*reserved*/ {0, 0, 0}
173 };
174
175 const static struct scsi_rigid_disk_page rigid_disk_page_default = {
176         /*page_code*/SMS_RIGID_DISK_PAGE,
177         /*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
178         /*cylinders*/ {0, 0, 0},
179         /*heads*/ CTL_DEFAULT_HEADS,
180         /*start_write_precomp*/ {0, 0, 0},
181         /*start_reduced_current*/ {0, 0, 0},
182         /*step_rate*/ {0, 0},
183         /*landing_zone_cylinder*/ {0, 0, 0},
184         /*rpl*/ SRDP_RPL_DISABLED,
185         /*rotational_offset*/ 0,
186         /*reserved1*/ 0,
187         /*rotation_rate*/ {(CTL_DEFAULT_ROTATION_RATE >> 8) & 0xff,
188                            CTL_DEFAULT_ROTATION_RATE & 0xff},
189         /*reserved2*/ {0, 0}
190 };
191
192 const static struct scsi_rigid_disk_page rigid_disk_page_changeable = {
193         /*page_code*/SMS_RIGID_DISK_PAGE,
194         /*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
195         /*cylinders*/ {0, 0, 0},
196         /*heads*/ 0,
197         /*start_write_precomp*/ {0, 0, 0},
198         /*start_reduced_current*/ {0, 0, 0},
199         /*step_rate*/ {0, 0},
200         /*landing_zone_cylinder*/ {0, 0, 0},
201         /*rpl*/ 0,
202         /*rotational_offset*/ 0,
203         /*reserved1*/ 0,
204         /*rotation_rate*/ {0, 0},
205         /*reserved2*/ {0, 0}
206 };
207
208 const static struct scsi_caching_page caching_page_default = {
209         /*page_code*/SMS_CACHING_PAGE,
210         /*page_length*/sizeof(struct scsi_caching_page) - 2,
211         /*flags1*/ SCP_DISC | SCP_WCE,
212         /*ret_priority*/ 0,
213         /*disable_pf_transfer_len*/ {0xff, 0xff},
214         /*min_prefetch*/ {0, 0},
215         /*max_prefetch*/ {0xff, 0xff},
216         /*max_pf_ceiling*/ {0xff, 0xff},
217         /*flags2*/ 0,
218         /*cache_segments*/ 0,
219         /*cache_seg_size*/ {0, 0},
220         /*reserved*/ 0,
221         /*non_cache_seg_size*/ {0, 0, 0}
222 };
223
224 const static struct scsi_caching_page caching_page_changeable = {
225         /*page_code*/SMS_CACHING_PAGE,
226         /*page_length*/sizeof(struct scsi_caching_page) - 2,
227         /*flags1*/ SCP_WCE | SCP_RCD,
228         /*ret_priority*/ 0,
229         /*disable_pf_transfer_len*/ {0, 0},
230         /*min_prefetch*/ {0, 0},
231         /*max_prefetch*/ {0, 0},
232         /*max_pf_ceiling*/ {0, 0},
233         /*flags2*/ 0,
234         /*cache_segments*/ 0,
235         /*cache_seg_size*/ {0, 0},
236         /*reserved*/ 0,
237         /*non_cache_seg_size*/ {0, 0, 0}
238 };
239
240 const static struct scsi_control_page control_page_default = {
241         /*page_code*/SMS_CONTROL_MODE_PAGE,
242         /*page_length*/sizeof(struct scsi_control_page) - 2,
243         /*rlec*/0,
244         /*queue_flags*/SCP_QUEUE_ALG_RESTRICTED,
245         /*eca_and_aen*/0,
246         /*flags4*/SCP_TAS,
247         /*aen_holdoff_period*/{0, 0},
248         /*busy_timeout_period*/{0, 0},
249         /*extended_selftest_completion_time*/{0, 0}
250 };
251
252 const static struct scsi_control_page control_page_changeable = {
253         /*page_code*/SMS_CONTROL_MODE_PAGE,
254         /*page_length*/sizeof(struct scsi_control_page) - 2,
255         /*rlec*/SCP_DSENSE,
256         /*queue_flags*/SCP_QUEUE_ALG_MASK,
257         /*eca_and_aen*/SCP_SWP,
258         /*flags4*/0,
259         /*aen_holdoff_period*/{0, 0},
260         /*busy_timeout_period*/{0, 0},
261         /*extended_selftest_completion_time*/{0, 0}
262 };
263
264 #define CTL_CEM_LEN     (sizeof(struct scsi_control_ext_page) - 4)
265
266 const static struct scsi_control_ext_page control_ext_page_default = {
267         /*page_code*/SMS_CONTROL_MODE_PAGE | SMPH_SPF,
268         /*subpage_code*/0x01,
269         /*page_length*/{CTL_CEM_LEN >> 8, CTL_CEM_LEN},
270         /*flags*/0,
271         /*prio*/0,
272         /*max_sense*/0
273 };
274
275 const static struct scsi_control_ext_page control_ext_page_changeable = {
276         /*page_code*/SMS_CONTROL_MODE_PAGE | SMPH_SPF,
277         /*subpage_code*/0x01,
278         /*page_length*/{CTL_CEM_LEN >> 8, CTL_CEM_LEN},
279         /*flags*/0,
280         /*prio*/0,
281         /*max_sense*/0
282 };
283
284 const static struct scsi_info_exceptions_page ie_page_default = {
285         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE,
286         /*page_length*/sizeof(struct scsi_info_exceptions_page) - 2,
287         /*info_flags*/SIEP_FLAGS_DEXCPT,
288         /*mrie*/0,
289         /*interval_timer*/{0, 0, 0, 0},
290         /*report_count*/{0, 0, 0, 0}
291 };
292
293 const static struct scsi_info_exceptions_page ie_page_changeable = {
294         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE,
295         /*page_length*/sizeof(struct scsi_info_exceptions_page) - 2,
296         /*info_flags*/0,
297         /*mrie*/0,
298         /*interval_timer*/{0, 0, 0, 0},
299         /*report_count*/{0, 0, 0, 0}
300 };
301
302 #define CTL_LBPM_LEN    (sizeof(struct ctl_logical_block_provisioning_page) - 4)
303
304 const static struct ctl_logical_block_provisioning_page lbp_page_default = {{
305         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF,
306         /*subpage_code*/0x02,
307         /*page_length*/{CTL_LBPM_LEN >> 8, CTL_LBPM_LEN},
308         /*flags*/0,
309         /*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
310         /*descr*/{}},
311         {{/*flags*/0,
312           /*resource*/0x01,
313           /*reserved*/{0, 0},
314           /*count*/{0, 0, 0, 0}},
315          {/*flags*/0,
316           /*resource*/0x02,
317           /*reserved*/{0, 0},
318           /*count*/{0, 0, 0, 0}},
319          {/*flags*/0,
320           /*resource*/0xf1,
321           /*reserved*/{0, 0},
322           /*count*/{0, 0, 0, 0}},
323          {/*flags*/0,
324           /*resource*/0xf2,
325           /*reserved*/{0, 0},
326           /*count*/{0, 0, 0, 0}}
327         }
328 };
329
330 const static struct ctl_logical_block_provisioning_page lbp_page_changeable = {{
331         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF,
332         /*subpage_code*/0x02,
333         /*page_length*/{CTL_LBPM_LEN >> 8, CTL_LBPM_LEN},
334         /*flags*/0,
335         /*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
336         /*descr*/{}},
337         {{/*flags*/0,
338           /*resource*/0,
339           /*reserved*/{0, 0},
340           /*count*/{0, 0, 0, 0}},
341          {/*flags*/0,
342           /*resource*/0,
343           /*reserved*/{0, 0},
344           /*count*/{0, 0, 0, 0}},
345          {/*flags*/0,
346           /*resource*/0,
347           /*reserved*/{0, 0},
348           /*count*/{0, 0, 0, 0}},
349          {/*flags*/0,
350           /*resource*/0,
351           /*reserved*/{0, 0},
352           /*count*/{0, 0, 0, 0}}
353         }
354 };
355
356 SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD, 0, "CAM Target Layer");
357 static int worker_threads = -1;
358 TUNABLE_INT("kern.cam.ctl.worker_threads", &worker_threads);
359 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, worker_threads, CTLFLAG_RDTUN,
360     &worker_threads, 1, "Number of worker threads");
361 static int ctl_debug = CTL_DEBUG_NONE;
362 TUNABLE_INT("kern.cam.ctl.debug", &ctl_debug);
363 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, debug, CTLFLAG_RWTUN,
364     &ctl_debug, 0, "Enabled debug flags");
365
366 /*
367  * Supported pages (0x00), Serial number (0x80), Device ID (0x83),
368  * Extended INQUIRY Data (0x86), Mode Page Policy (0x87),
369  * SCSI Ports (0x88), Third-party Copy (0x8F), Block limits (0xB0),
370  * Block Device Characteristics (0xB1) and Logical Block Provisioning (0xB2)
371  */
372 #define SCSI_EVPD_NUM_SUPPORTED_PAGES   10
373
374 static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
375                                   int param);
376 static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
377 static void ctl_copy_sense_data_back(union ctl_io *src, union ctl_ha_msg *dest);
378 static int ctl_init(void);
379 void ctl_shutdown(void);
380 static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
381 static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
382 static int ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio);
383 static int ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
384                               struct ctl_ooa *ooa_hdr,
385                               struct ctl_ooa_entry *kern_entries);
386 static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
387                      struct thread *td);
388 static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
389                          struct ctl_be_lun *be_lun);
390 static int ctl_free_lun(struct ctl_lun *lun);
391 static void ctl_create_lun(struct ctl_be_lun *be_lun);
392 static struct ctl_port * ctl_io_port(struct ctl_io_hdr *io_hdr);
393
394 static int ctl_do_mode_select(union ctl_io *io);
395 static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
396                            uint64_t res_key, uint64_t sa_res_key,
397                            uint8_t type, uint32_t residx,
398                            struct ctl_scsiio *ctsio,
399                            struct scsi_per_res_out *cdb,
400                            struct scsi_per_res_out_parms* param);
401 static void ctl_pro_preempt_other(struct ctl_lun *lun,
402                                   union ctl_ha_msg *msg);
403 static void ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg);
404 static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
405 static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
406 static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
407 static int ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len);
408 static int ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len);
409 static int ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio,
410                                          int alloc_len);
411 static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
412                                          int alloc_len);
413 static int ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len);
414 static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
415 static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
416 static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
417 static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len);
418 static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2,
419     bool seq);
420 static ctl_action ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2);
421 static ctl_action ctl_check_for_blockage(struct ctl_lun *lun,
422     union ctl_io *pending_io, union ctl_io *ooa_io);
423 static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
424                                 union ctl_io *starting_io);
425 static int ctl_check_blocked(struct ctl_lun *lun);
426 static int ctl_scsiio_lun_check(struct ctl_lun *lun,
427                                 const struct ctl_cmd_entry *entry,
428                                 struct ctl_scsiio *ctsio);
429 static void ctl_failover_lun(struct ctl_lun *lun);
430 static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
431                                struct ctl_scsiio *ctsio);
432 static int ctl_scsiio(struct ctl_scsiio *ctsio);
433
434 static int ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
435 static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
436                             ctl_ua_type ua_type);
437 static int ctl_do_lun_reset(struct ctl_lun *lun, union ctl_io *io,
438                          ctl_ua_type ua_type);
439 static int ctl_lun_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
440 static int ctl_abort_task(union ctl_io *io);
441 static int ctl_abort_task_set(union ctl_io *io);
442 static int ctl_query_task(union ctl_io *io, int task_set);
443 static int ctl_i_t_nexus_reset(union ctl_io *io);
444 static int ctl_query_async_event(union ctl_io *io);
445 static void ctl_run_task(union ctl_io *io);
446 #ifdef CTL_IO_DELAY
447 static void ctl_datamove_timer_wakeup(void *arg);
448 static void ctl_done_timer_wakeup(void *arg);
449 #endif /* CTL_IO_DELAY */
450
451 static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
452 static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
453 static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
454 static void ctl_datamove_remote_write(union ctl_io *io);
455 static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
456 static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
457 static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
458 static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
459                                     ctl_ha_dt_cb callback);
460 static void ctl_datamove_remote_read(union ctl_io *io);
461 static void ctl_datamove_remote(union ctl_io *io);
462 static int ctl_process_done(union ctl_io *io);
463 static void ctl_lun_thread(void *arg);
464 static void ctl_thresh_thread(void *arg);
465 static void ctl_work_thread(void *arg);
466 static void ctl_enqueue_incoming(union ctl_io *io);
467 static void ctl_enqueue_rtr(union ctl_io *io);
468 static void ctl_enqueue_done(union ctl_io *io);
469 static void ctl_enqueue_isc(union ctl_io *io);
470 static const struct ctl_cmd_entry *
471     ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa);
472 static const struct ctl_cmd_entry *
473     ctl_validate_command(struct ctl_scsiio *ctsio);
474 static int ctl_cmd_applicable(uint8_t lun_type,
475     const struct ctl_cmd_entry *entry);
476
477 static uint64_t ctl_get_prkey(struct ctl_lun *lun, uint32_t residx);
478 static void ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx);
479 static void ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx);
480 static void ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key);
481
482 /*
483  * Load the serialization table.  This isn't very pretty, but is probably
484  * the easiest way to do it.
485  */
486 #include "ctl_ser_table.c"
487
488 /*
489  * We only need to define open, close and ioctl routines for this driver.
490  */
491 static struct cdevsw ctl_cdevsw = {
492         .d_version =    D_VERSION,
493         .d_flags =      0,
494         .d_open =       ctl_open,
495         .d_close =      ctl_close,
496         .d_ioctl =      ctl_ioctl,
497         .d_name =       "ctl",
498 };
499
500
501 MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
502
503 static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
504
505 static moduledata_t ctl_moduledata = {
506         "ctl",
507         ctl_module_event_handler,
508         NULL
509 };
510
511 DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
512 MODULE_VERSION(ctl, 1);
513
514 static struct ctl_frontend ha_frontend =
515 {
516         .name = "ha",
517 };
518
519 static void
520 ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
521                             union ctl_ha_msg *msg_info)
522 {
523         struct ctl_scsiio *ctsio;
524
525         if (msg_info->hdr.original_sc == NULL) {
526                 printf("%s: original_sc == NULL!\n", __func__);
527                 /* XXX KDM now what? */
528                 return;
529         }
530
531         ctsio = &msg_info->hdr.original_sc->scsiio;
532         ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
533         ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
534         ctsio->io_hdr.status = msg_info->hdr.status;
535         ctsio->scsi_status = msg_info->scsi.scsi_status;
536         ctsio->sense_len = msg_info->scsi.sense_len;
537         ctsio->sense_residual = msg_info->scsi.sense_residual;
538         ctsio->residual = msg_info->scsi.residual;
539         memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
540                msg_info->scsi.sense_len);
541         ctl_enqueue_isc((union ctl_io *)ctsio);
542 }
543
544 static void
545 ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
546                                 union ctl_ha_msg *msg_info)
547 {
548         struct ctl_scsiio *ctsio;
549
550         if (msg_info->hdr.serializing_sc == NULL) {
551                 printf("%s: serializing_sc == NULL!\n", __func__);
552                 /* XXX KDM now what? */
553                 return;
554         }
555
556         ctsio = &msg_info->hdr.serializing_sc->scsiio;
557         ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
558         ctl_enqueue_isc((union ctl_io *)ctsio);
559 }
560
561 void
562 ctl_isc_announce_lun(struct ctl_lun *lun)
563 {
564         struct ctl_softc *softc = lun->ctl_softc;
565         union ctl_ha_msg *msg;
566         struct ctl_ha_msg_lun_pr_key pr_key;
567         int i, k;
568
569         if (softc->ha_link != CTL_HA_LINK_ONLINE)
570                 return;
571         mtx_lock(&lun->lun_lock);
572         i = sizeof(msg->lun);
573         if (lun->lun_devid)
574                 i += lun->lun_devid->len;
575         i += sizeof(pr_key) * lun->pr_key_count;
576 alloc:
577         mtx_unlock(&lun->lun_lock);
578         msg = malloc(i, M_CTL, M_WAITOK);
579         mtx_lock(&lun->lun_lock);
580         k = sizeof(msg->lun);
581         if (lun->lun_devid)
582                 k += lun->lun_devid->len;
583         k += sizeof(pr_key) * lun->pr_key_count;
584         if (i < k) {
585                 free(msg, M_CTL);
586                 i = k;
587                 goto alloc;
588         }
589         bzero(&msg->lun, sizeof(msg->lun));
590         msg->hdr.msg_type = CTL_MSG_LUN_SYNC;
591         msg->hdr.nexus.targ_lun = lun->lun;
592         msg->hdr.nexus.targ_mapped_lun = lun->lun;
593         msg->lun.flags = lun->flags;
594         msg->lun.pr_generation = lun->PRGeneration;
595         msg->lun.pr_res_idx = lun->pr_res_idx;
596         msg->lun.pr_res_type = lun->res_type;
597         msg->lun.pr_key_count = lun->pr_key_count;
598         i = 0;
599         if (lun->lun_devid) {
600                 msg->lun.lun_devid_len = lun->lun_devid->len;
601                 memcpy(&msg->lun.data[i], lun->lun_devid->data,
602                     msg->lun.lun_devid_len);
603                 i += msg->lun.lun_devid_len;
604         }
605         for (k = 0; k < CTL_MAX_INITIATORS; k++) {
606                 if ((pr_key.pr_key = ctl_get_prkey(lun, k)) == 0)
607                         continue;
608                 pr_key.pr_iid = k;
609                 memcpy(&msg->lun.data[i], &pr_key, sizeof(pr_key));
610                 i += sizeof(pr_key);
611         }
612         mtx_unlock(&lun->lun_lock);
613         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->port, sizeof(msg->port) + i,
614             M_WAITOK);
615         free(msg, M_CTL);
616
617         if (lun->flags & CTL_LUN_PRIMARY_SC) {
618                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
619                         ctl_isc_announce_mode(lun, -1,
620                             lun->mode_pages.index[i].page_code & SMPH_PC_MASK,
621                             lun->mode_pages.index[i].subpage);
622                 }
623         }
624 }
625
626 void
627 ctl_isc_announce_port(struct ctl_port *port)
628 {
629         struct ctl_softc *softc = control_softc;
630         union ctl_ha_msg *msg;
631         int i;
632
633         if (port->targ_port < softc->port_min ||
634             port->targ_port >= softc->port_max ||
635             softc->ha_link != CTL_HA_LINK_ONLINE)
636                 return;
637         i = sizeof(msg->port) + strlen(port->port_name) + 1;
638         if (port->lun_map)
639                 i += sizeof(uint32_t) * CTL_MAX_LUNS;
640         if (port->port_devid)
641                 i += port->port_devid->len;
642         if (port->target_devid)
643                 i += port->target_devid->len;
644         if (port->init_devid)
645                 i += port->init_devid->len;
646         msg = malloc(i, M_CTL, M_WAITOK);
647         bzero(&msg->port, sizeof(msg->port));
648         msg->hdr.msg_type = CTL_MSG_PORT_SYNC;
649         msg->hdr.nexus.targ_port = port->targ_port;
650         msg->port.port_type = port->port_type;
651         msg->port.physical_port = port->physical_port;
652         msg->port.virtual_port = port->virtual_port;
653         msg->port.status = port->status;
654         i = 0;
655         msg->port.name_len = sprintf(&msg->port.data[i],
656             "%d:%s", softc->ha_id, port->port_name) + 1;
657         i += msg->port.name_len;
658         if (port->lun_map) {
659                 msg->port.lun_map_len = sizeof(uint32_t) * CTL_MAX_LUNS;
660                 memcpy(&msg->port.data[i], port->lun_map,
661                     msg->port.lun_map_len);
662                 i += msg->port.lun_map_len;
663         }
664         if (port->port_devid) {
665                 msg->port.port_devid_len = port->port_devid->len;
666                 memcpy(&msg->port.data[i], port->port_devid->data,
667                     msg->port.port_devid_len);
668                 i += msg->port.port_devid_len;
669         }
670         if (port->target_devid) {
671                 msg->port.target_devid_len = port->target_devid->len;
672                 memcpy(&msg->port.data[i], port->target_devid->data,
673                     msg->port.target_devid_len);
674                 i += msg->port.target_devid_len;
675         }
676         if (port->init_devid) {
677                 msg->port.init_devid_len = port->init_devid->len;
678                 memcpy(&msg->port.data[i], port->init_devid->data,
679                     msg->port.init_devid_len);
680                 i += msg->port.init_devid_len;
681         }
682         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->port, sizeof(msg->port) + i,
683             M_WAITOK);
684         free(msg, M_CTL);
685 }
686
687 void
688 ctl_isc_announce_iid(struct ctl_port *port, int iid)
689 {
690         struct ctl_softc *softc = control_softc;
691         union ctl_ha_msg *msg;
692         int i, l;
693
694         if (port->targ_port < softc->port_min ||
695             port->targ_port >= softc->port_max ||
696             softc->ha_link != CTL_HA_LINK_ONLINE)
697                 return;
698         mtx_lock(&softc->ctl_lock);
699         i = sizeof(msg->iid);
700         l = 0;
701         if (port->wwpn_iid[iid].name)
702                 l = strlen(port->wwpn_iid[iid].name) + 1;
703         i += l;
704         msg = malloc(i, M_CTL, M_NOWAIT);
705         if (msg == NULL) {
706                 mtx_unlock(&softc->ctl_lock);
707                 return;
708         }
709         bzero(&msg->iid, sizeof(msg->iid));
710         msg->hdr.msg_type = CTL_MSG_IID_SYNC;
711         msg->hdr.nexus.targ_port = port->targ_port;
712         msg->hdr.nexus.initid = iid;
713         msg->iid.in_use = port->wwpn_iid[iid].in_use;
714         msg->iid.name_len = l;
715         msg->iid.wwpn = port->wwpn_iid[iid].wwpn;
716         if (port->wwpn_iid[iid].name)
717                 strlcpy(msg->iid.data, port->wwpn_iid[iid].name, l);
718         mtx_unlock(&softc->ctl_lock);
719         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->iid, i, M_NOWAIT);
720         free(msg, M_CTL);
721 }
722
723 void
724 ctl_isc_announce_mode(struct ctl_lun *lun, uint32_t initidx,
725     uint8_t page, uint8_t subpage)
726 {
727         struct ctl_softc *softc = lun->ctl_softc;
728         union ctl_ha_msg msg;
729         int i;
730
731         if (softc->ha_link != CTL_HA_LINK_ONLINE)
732                 return;
733         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
734                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) ==
735                     page && lun->mode_pages.index[i].subpage == subpage)
736                         break;
737         }
738         if (i == CTL_NUM_MODE_PAGES)
739                 return;
740         bzero(&msg.mode, sizeof(msg.mode));
741         msg.hdr.msg_type = CTL_MSG_MODE_SYNC;
742         msg.hdr.nexus.targ_port = initidx / CTL_MAX_INIT_PER_PORT;
743         msg.hdr.nexus.initid = initidx % CTL_MAX_INIT_PER_PORT;
744         msg.hdr.nexus.targ_lun = lun->lun;
745         msg.hdr.nexus.targ_mapped_lun = lun->lun;
746         msg.mode.page_code = page;
747         msg.mode.subpage = subpage;
748         msg.mode.page_len = lun->mode_pages.index[i].page_len;
749         memcpy(msg.mode.data, lun->mode_pages.index[i].page_data,
750             msg.mode.page_len);
751         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg.mode, sizeof(msg.mode),
752             M_WAITOK);
753 }
754
755 static void
756 ctl_isc_ha_link_up(struct ctl_softc *softc)
757 {
758         struct ctl_port *port;
759         struct ctl_lun *lun;
760         union ctl_ha_msg msg;
761         int i;
762
763         /* Announce this node parameters to peer for validation. */
764         msg.login.msg_type = CTL_MSG_LOGIN;
765         msg.login.version = CTL_HA_VERSION;
766         msg.login.ha_mode = softc->ha_mode;
767         msg.login.ha_id = softc->ha_id;
768         msg.login.max_luns = CTL_MAX_LUNS;
769         msg.login.max_ports = CTL_MAX_PORTS;
770         msg.login.max_init_per_port = CTL_MAX_INIT_PER_PORT;
771         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg.login, sizeof(msg.login),
772             M_WAITOK);
773
774         STAILQ_FOREACH(port, &softc->port_list, links) {
775                 ctl_isc_announce_port(port);
776                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
777                         if (port->wwpn_iid[i].in_use)
778                                 ctl_isc_announce_iid(port, i);
779                 }
780         }
781         STAILQ_FOREACH(lun, &softc->lun_list, links)
782                 ctl_isc_announce_lun(lun);
783 }
784
785 static void
786 ctl_isc_ha_link_down(struct ctl_softc *softc)
787 {
788         struct ctl_port *port;
789         struct ctl_lun *lun;
790         union ctl_io *io;
791         int i;
792
793         mtx_lock(&softc->ctl_lock);
794         STAILQ_FOREACH(lun, &softc->lun_list, links) {
795                 mtx_lock(&lun->lun_lock);
796                 if (lun->flags & CTL_LUN_PEER_SC_PRIMARY) {
797                         lun->flags &= ~CTL_LUN_PEER_SC_PRIMARY;
798                         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
799                 }
800                 mtx_unlock(&lun->lun_lock);
801
802                 mtx_unlock(&softc->ctl_lock);
803                 io = ctl_alloc_io(softc->othersc_pool);
804                 mtx_lock(&softc->ctl_lock);
805                 ctl_zero_io(io);
806                 io->io_hdr.msg_type = CTL_MSG_FAILOVER;
807                 io->io_hdr.nexus.targ_mapped_lun = lun->lun;
808                 ctl_enqueue_isc(io);
809         }
810
811         STAILQ_FOREACH(port, &softc->port_list, links) {
812                 if (port->targ_port >= softc->port_min &&
813                     port->targ_port < softc->port_max)
814                         continue;
815                 port->status &= ~CTL_PORT_STATUS_ONLINE;
816                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
817                         port->wwpn_iid[i].in_use = 0;
818                         free(port->wwpn_iid[i].name, M_CTL);
819                         port->wwpn_iid[i].name = NULL;
820                 }
821         }
822         mtx_unlock(&softc->ctl_lock);
823 }
824
825 static void
826 ctl_isc_ua(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
827 {
828         struct ctl_lun *lun;
829         uint32_t iid = ctl_get_initindex(&msg->hdr.nexus);
830
831         mtx_lock(&softc->ctl_lock);
832         if (msg->hdr.nexus.targ_lun < CTL_MAX_LUNS &&
833             (lun = softc->ctl_luns[msg->hdr.nexus.targ_mapped_lun]) != NULL) {
834                 mtx_lock(&lun->lun_lock);
835                 mtx_unlock(&softc->ctl_lock);
836                 if (msg->ua.ua_type == CTL_UA_THIN_PROV_THRES &&
837                     msg->ua.ua_set)
838                         memcpy(lun->ua_tpt_info, msg->ua.ua_info, 8);
839                 if (msg->ua.ua_all) {
840                         if (msg->ua.ua_set)
841                                 ctl_est_ua_all(lun, iid, msg->ua.ua_type);
842                         else
843                                 ctl_clr_ua_all(lun, iid, msg->ua.ua_type);
844                 } else {
845                         if (msg->ua.ua_set)
846                                 ctl_est_ua(lun, iid, msg->ua.ua_type);
847                         else
848                                 ctl_clr_ua(lun, iid, msg->ua.ua_type);
849                 }
850                 mtx_unlock(&lun->lun_lock);
851         } else
852                 mtx_unlock(&softc->ctl_lock);
853 }
854
855 static void
856 ctl_isc_lun_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
857 {
858         struct ctl_lun *lun;
859         struct ctl_ha_msg_lun_pr_key pr_key;
860         int i, k;
861         ctl_lun_flags oflags;
862         uint32_t targ_lun;
863
864         targ_lun = msg->hdr.nexus.targ_mapped_lun;
865         mtx_lock(&softc->ctl_lock);
866         if ((targ_lun >= CTL_MAX_LUNS) ||
867             ((lun = softc->ctl_luns[targ_lun]) == NULL)) {
868                 mtx_unlock(&softc->ctl_lock);
869                 return;
870         }
871         mtx_lock(&lun->lun_lock);
872         mtx_unlock(&softc->ctl_lock);
873         if (lun->flags & CTL_LUN_DISABLED) {
874                 mtx_unlock(&lun->lun_lock);
875                 return;
876         }
877         i = (lun->lun_devid != NULL) ? lun->lun_devid->len : 0;
878         if (msg->lun.lun_devid_len != i || (i > 0 &&
879             memcmp(&msg->lun.data[0], lun->lun_devid->data, i) != 0)) {
880                 mtx_unlock(&lun->lun_lock);
881                 printf("%s: Received conflicting HA LUN %d\n",
882                     __func__, msg->hdr.nexus.targ_lun);
883                 return;
884         } else {
885                 /* Record whether peer is primary. */
886                 oflags = lun->flags;
887                 if ((msg->lun.flags & CTL_LUN_PRIMARY_SC) &&
888                     (msg->lun.flags & CTL_LUN_DISABLED) == 0)
889                         lun->flags |= CTL_LUN_PEER_SC_PRIMARY;
890                 else
891                         lun->flags &= ~CTL_LUN_PEER_SC_PRIMARY;
892                 if (oflags != lun->flags)
893                         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
894
895                 /* If peer is primary and we are not -- use data */
896                 if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
897                     (lun->flags & CTL_LUN_PEER_SC_PRIMARY)) {
898                         lun->PRGeneration = msg->lun.pr_generation;
899                         lun->pr_res_idx = msg->lun.pr_res_idx;
900                         lun->res_type = msg->lun.pr_res_type;
901                         lun->pr_key_count = msg->lun.pr_key_count;
902                         for (k = 0; k < CTL_MAX_INITIATORS; k++)
903                                 ctl_clr_prkey(lun, k);
904                         for (k = 0; k < msg->lun.pr_key_count; k++) {
905                                 memcpy(&pr_key, &msg->lun.data[i],
906                                     sizeof(pr_key));
907                                 ctl_alloc_prkey(lun, pr_key.pr_iid);
908                                 ctl_set_prkey(lun, pr_key.pr_iid,
909                                     pr_key.pr_key);
910                                 i += sizeof(pr_key);
911                         }
912                 }
913
914                 mtx_unlock(&lun->lun_lock);
915                 CTL_DEBUG_PRINT(("%s: Known LUN %d, peer is %s\n",
916                     __func__, msg->hdr.nexus.targ_lun,
917                     (msg->lun.flags & CTL_LUN_PRIMARY_SC) ?
918                     "primary" : "secondary"));
919
920                 /* If we are primary but peer doesn't know -- notify */
921                 if ((lun->flags & CTL_LUN_PRIMARY_SC) &&
922                     (msg->lun.flags & CTL_LUN_PEER_SC_PRIMARY) == 0)
923                         ctl_isc_announce_lun(lun);
924         }
925 }
926
927 static void
928 ctl_isc_port_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
929 {
930         struct ctl_port *port;
931         struct ctl_lun *lun;
932         int i, new;
933
934         port = softc->ctl_ports[msg->hdr.nexus.targ_port];
935         if (port == NULL) {
936                 CTL_DEBUG_PRINT(("%s: New port %d\n", __func__,
937                     msg->hdr.nexus.targ_port));
938                 new = 1;
939                 port = malloc(sizeof(*port), M_CTL, M_WAITOK | M_ZERO);
940                 port->frontend = &ha_frontend;
941                 port->targ_port = msg->hdr.nexus.targ_port;
942         } else if (port->frontend == &ha_frontend) {
943                 CTL_DEBUG_PRINT(("%s: Updated port %d\n", __func__,
944                     msg->hdr.nexus.targ_port));
945                 new = 0;
946         } else {
947                 printf("%s: Received conflicting HA port %d\n",
948                     __func__, msg->hdr.nexus.targ_port);
949                 return;
950         }
951         port->port_type = msg->port.port_type;
952         port->physical_port = msg->port.physical_port;
953         port->virtual_port = msg->port.virtual_port;
954         port->status = msg->port.status;
955         i = 0;
956         free(port->port_name, M_CTL);
957         port->port_name = strndup(&msg->port.data[i], msg->port.name_len,
958             M_CTL);
959         i += msg->port.name_len;
960         if (msg->port.lun_map_len != 0) {
961                 if (port->lun_map == NULL)
962                         port->lun_map = malloc(sizeof(uint32_t) * CTL_MAX_LUNS,
963                             M_CTL, M_WAITOK);
964                 memcpy(port->lun_map, &msg->port.data[i],
965                     sizeof(uint32_t) * CTL_MAX_LUNS);
966                 i += msg->port.lun_map_len;
967         } else {
968                 free(port->lun_map, M_CTL);
969                 port->lun_map = NULL;
970         }
971         if (msg->port.port_devid_len != 0) {
972                 if (port->port_devid == NULL ||
973                     port->port_devid->len != msg->port.port_devid_len) {
974                         free(port->port_devid, M_CTL);
975                         port->port_devid = malloc(sizeof(struct ctl_devid) +
976                             msg->port.port_devid_len, M_CTL, M_WAITOK);
977                 }
978                 memcpy(port->port_devid->data, &msg->port.data[i],
979                     msg->port.port_devid_len);
980                 port->port_devid->len = msg->port.port_devid_len;
981                 i += msg->port.port_devid_len;
982         } else {
983                 free(port->port_devid, M_CTL);
984                 port->port_devid = NULL;
985         }
986         if (msg->port.target_devid_len != 0) {
987                 if (port->target_devid == NULL ||
988                     port->target_devid->len != msg->port.target_devid_len) {
989                         free(port->target_devid, M_CTL);
990                         port->target_devid = malloc(sizeof(struct ctl_devid) +
991                             msg->port.target_devid_len, M_CTL, M_WAITOK);
992                 }
993                 memcpy(port->target_devid->data, &msg->port.data[i],
994                     msg->port.target_devid_len);
995                 port->target_devid->len = msg->port.target_devid_len;
996                 i += msg->port.target_devid_len;
997         } else {
998                 free(port->target_devid, M_CTL);
999                 port->target_devid = NULL;
1000         }
1001         if (msg->port.init_devid_len != 0) {
1002                 if (port->init_devid == NULL ||
1003                     port->init_devid->len != msg->port.init_devid_len) {
1004                         free(port->init_devid, M_CTL);
1005                         port->init_devid = malloc(sizeof(struct ctl_devid) +
1006                             msg->port.init_devid_len, M_CTL, M_WAITOK);
1007                 }
1008                 memcpy(port->init_devid->data, &msg->port.data[i],
1009                     msg->port.init_devid_len);
1010                 port->init_devid->len = msg->port.init_devid_len;
1011                 i += msg->port.init_devid_len;
1012         } else {
1013                 free(port->init_devid, M_CTL);
1014                 port->init_devid = NULL;
1015         }
1016         if (new) {
1017                 if (ctl_port_register(port) != 0) {
1018                         printf("%s: ctl_port_register() failed with error\n",
1019                             __func__);
1020                 }
1021         }
1022         mtx_lock(&softc->ctl_lock);
1023         STAILQ_FOREACH(lun, &softc->lun_list, links) {
1024                 if (ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
1025                         continue;
1026                 mtx_lock(&lun->lun_lock);
1027                 ctl_est_ua_all(lun, -1, CTL_UA_INQ_CHANGE);
1028                 mtx_unlock(&lun->lun_lock);
1029         }
1030         mtx_unlock(&softc->ctl_lock);
1031 }
1032
1033 static void
1034 ctl_isc_iid_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1035 {
1036         struct ctl_port *port;
1037         int iid;
1038
1039         port = softc->ctl_ports[msg->hdr.nexus.targ_port];
1040         if (port == NULL) {
1041                 printf("%s: Received IID for unknown port %d\n",
1042                     __func__, msg->hdr.nexus.targ_port);
1043                 return;
1044         }
1045         iid = msg->hdr.nexus.initid;
1046         port->wwpn_iid[iid].in_use = msg->iid.in_use;
1047         port->wwpn_iid[iid].wwpn = msg->iid.wwpn;
1048         free(port->wwpn_iid[iid].name, M_CTL);
1049         if (msg->iid.name_len) {
1050                 port->wwpn_iid[iid].name = strndup(&msg->iid.data[0],
1051                     msg->iid.name_len, M_CTL);
1052         } else
1053                 port->wwpn_iid[iid].name = NULL;
1054 }
1055
1056 static void
1057 ctl_isc_login(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1058 {
1059
1060         if (msg->login.version != CTL_HA_VERSION) {
1061                 printf("CTL HA peers have different versions %d != %d\n",
1062                     msg->login.version, CTL_HA_VERSION);
1063                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1064                 return;
1065         }
1066         if (msg->login.ha_mode != softc->ha_mode) {
1067                 printf("CTL HA peers have different ha_mode %d != %d\n",
1068                     msg->login.ha_mode, softc->ha_mode);
1069                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1070                 return;
1071         }
1072         if (msg->login.ha_id == softc->ha_id) {
1073                 printf("CTL HA peers have same ha_id %d\n", msg->login.ha_id);
1074                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1075                 return;
1076         }
1077         if (msg->login.max_luns != CTL_MAX_LUNS ||
1078             msg->login.max_ports != CTL_MAX_PORTS ||
1079             msg->login.max_init_per_port != CTL_MAX_INIT_PER_PORT) {
1080                 printf("CTL HA peers have different limits\n");
1081                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1082                 return;
1083         }
1084 }
1085
1086 static void
1087 ctl_isc_mode_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1088 {
1089         struct ctl_lun *lun;
1090         int i;
1091         uint32_t initidx, targ_lun;
1092
1093         targ_lun = msg->hdr.nexus.targ_mapped_lun;
1094         mtx_lock(&softc->ctl_lock);
1095         if ((targ_lun >= CTL_MAX_LUNS) ||
1096             ((lun = softc->ctl_luns[targ_lun]) == NULL)) {
1097                 mtx_unlock(&softc->ctl_lock);
1098                 return;
1099         }
1100         mtx_lock(&lun->lun_lock);
1101         mtx_unlock(&softc->ctl_lock);
1102         if (lun->flags & CTL_LUN_DISABLED) {
1103                 mtx_unlock(&lun->lun_lock);
1104                 return;
1105         }
1106         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
1107                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) ==
1108                     msg->mode.page_code &&
1109                     lun->mode_pages.index[i].subpage == msg->mode.subpage)
1110                         break;
1111         }
1112         if (i == CTL_NUM_MODE_PAGES) {
1113                 mtx_unlock(&lun->lun_lock);
1114                 return;
1115         }
1116         memcpy(lun->mode_pages.index[i].page_data, msg->mode.data,
1117             lun->mode_pages.index[i].page_len);
1118         initidx = ctl_get_initindex(&msg->hdr.nexus);
1119         if (initidx != -1)
1120                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
1121         mtx_unlock(&lun->lun_lock);
1122 }
1123
1124 /*
1125  * ISC (Inter Shelf Communication) event handler.  Events from the HA
1126  * subsystem come in here.
1127  */
1128 static void
1129 ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
1130 {
1131         struct ctl_softc *softc;
1132         union ctl_io *io;
1133         struct ctl_prio *presio;
1134         ctl_ha_status isc_status;
1135
1136         softc = control_softc;
1137         CTL_DEBUG_PRINT(("CTL: Isc Msg event %d\n", event));
1138         if (event == CTL_HA_EVT_MSG_RECV) {
1139                 union ctl_ha_msg *msg, msgbuf;
1140
1141                 if (param > sizeof(msgbuf))
1142                         msg = malloc(param, M_CTL, M_WAITOK);
1143                 else
1144                         msg = &msgbuf;
1145                 isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, msg, param,
1146                     M_WAITOK);
1147                 if (isc_status != CTL_HA_STATUS_SUCCESS) {
1148                         printf("%s: Error receiving message: %d\n",
1149                             __func__, isc_status);
1150                         if (msg != &msgbuf)
1151                                 free(msg, M_CTL);
1152                         return;
1153                 }
1154
1155                 CTL_DEBUG_PRINT(("CTL: msg_type %d\n", msg->msg_type));
1156                 switch (msg->hdr.msg_type) {
1157                 case CTL_MSG_SERIALIZE:
1158                         io = ctl_alloc_io(softc->othersc_pool);
1159                         ctl_zero_io(io);
1160                         // populate ctsio from msg
1161                         io->io_hdr.io_type = CTL_IO_SCSI;
1162                         io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
1163                         io->io_hdr.original_sc = msg->hdr.original_sc;
1164                         io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
1165                                             CTL_FLAG_IO_ACTIVE;
1166                         /*
1167                          * If we're in serialization-only mode, we don't
1168                          * want to go through full done processing.  Thus
1169                          * the COPY flag.
1170                          *
1171                          * XXX KDM add another flag that is more specific.
1172                          */
1173                         if (softc->ha_mode != CTL_HA_MODE_XFER)
1174                                 io->io_hdr.flags |= CTL_FLAG_INT_COPY;
1175                         io->io_hdr.nexus = msg->hdr.nexus;
1176 #if 0
1177                         printf("port %u, iid %u, lun %u\n",
1178                                io->io_hdr.nexus.targ_port,
1179                                io->io_hdr.nexus.initid,
1180                                io->io_hdr.nexus.targ_lun);
1181 #endif
1182                         io->scsiio.tag_num = msg->scsi.tag_num;
1183                         io->scsiio.tag_type = msg->scsi.tag_type;
1184 #ifdef CTL_TIME_IO
1185                         io->io_hdr.start_time = time_uptime;
1186                         getbintime(&io->io_hdr.start_bt);
1187 #endif /* CTL_TIME_IO */
1188                         io->scsiio.cdb_len = msg->scsi.cdb_len;
1189                         memcpy(io->scsiio.cdb, msg->scsi.cdb,
1190                                CTL_MAX_CDBLEN);
1191                         if (softc->ha_mode == CTL_HA_MODE_XFER) {
1192                                 const struct ctl_cmd_entry *entry;
1193
1194                                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
1195                                 io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
1196                                 io->io_hdr.flags |=
1197                                         entry->flags & CTL_FLAG_DATA_MASK;
1198                         }
1199                         ctl_enqueue_isc(io);
1200                         break;
1201
1202                 /* Performed on the Originating SC, XFER mode only */
1203                 case CTL_MSG_DATAMOVE: {
1204                         struct ctl_sg_entry *sgl;
1205                         int i, j;
1206
1207                         io = msg->hdr.original_sc;
1208                         if (io == NULL) {
1209                                 printf("%s: original_sc == NULL!\n", __func__);
1210                                 /* XXX KDM do something here */
1211                                 break;
1212                         }
1213                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
1214                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1215                         /*
1216                          * Keep track of this, we need to send it back over
1217                          * when the datamove is complete.
1218                          */
1219                         io->io_hdr.serializing_sc = msg->hdr.serializing_sc;
1220                         if (msg->hdr.status == CTL_SUCCESS)
1221                                 io->io_hdr.status = msg->hdr.status;
1222
1223                         if (msg->dt.sg_sequence == 0) {
1224                                 i = msg->dt.kern_sg_entries +
1225                                     msg->dt.kern_data_len /
1226                                     CTL_HA_DATAMOVE_SEGMENT + 1;
1227                                 sgl = malloc(sizeof(*sgl) * i, M_CTL,
1228                                     M_WAITOK | M_ZERO);
1229                                 io->io_hdr.remote_sglist = sgl;
1230                                 io->io_hdr.local_sglist =
1231                                     &sgl[msg->dt.kern_sg_entries];
1232
1233                                 io->scsiio.kern_data_ptr = (uint8_t *)sgl;
1234
1235                                 io->scsiio.kern_sg_entries =
1236                                         msg->dt.kern_sg_entries;
1237                                 io->scsiio.rem_sg_entries =
1238                                         msg->dt.kern_sg_entries;
1239                                 io->scsiio.kern_data_len =
1240                                         msg->dt.kern_data_len;
1241                                 io->scsiio.kern_total_len =
1242                                         msg->dt.kern_total_len;
1243                                 io->scsiio.kern_data_resid =
1244                                         msg->dt.kern_data_resid;
1245                                 io->scsiio.kern_rel_offset =
1246                                         msg->dt.kern_rel_offset;
1247                                 io->io_hdr.flags &= ~CTL_FLAG_BUS_ADDR;
1248                                 io->io_hdr.flags |= msg->dt.flags &
1249                                     CTL_FLAG_BUS_ADDR;
1250                         } else
1251                                 sgl = (struct ctl_sg_entry *)
1252                                         io->scsiio.kern_data_ptr;
1253
1254                         for (i = msg->dt.sent_sg_entries, j = 0;
1255                              i < (msg->dt.sent_sg_entries +
1256                              msg->dt.cur_sg_entries); i++, j++) {
1257                                 sgl[i].addr = msg->dt.sg_list[j].addr;
1258                                 sgl[i].len = msg->dt.sg_list[j].len;
1259
1260 #if 0
1261                                 printf("%s: DATAMOVE: %p,%lu j=%d, i=%d\n",
1262                                     __func__, sgl[i].addr, sgl[i].len, j, i);
1263 #endif
1264                         }
1265
1266                         /*
1267                          * If this is the last piece of the I/O, we've got
1268                          * the full S/G list.  Queue processing in the thread.
1269                          * Otherwise wait for the next piece.
1270                          */
1271                         if (msg->dt.sg_last != 0)
1272                                 ctl_enqueue_isc(io);
1273                         break;
1274                 }
1275                 /* Performed on the Serializing (primary) SC, XFER mode only */
1276                 case CTL_MSG_DATAMOVE_DONE: {
1277                         if (msg->hdr.serializing_sc == NULL) {
1278                                 printf("%s: serializing_sc == NULL!\n",
1279                                        __func__);
1280                                 /* XXX KDM now what? */
1281                                 break;
1282                         }
1283                         /*
1284                          * We grab the sense information here in case
1285                          * there was a failure, so we can return status
1286                          * back to the initiator.
1287                          */
1288                         io = msg->hdr.serializing_sc;
1289                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
1290                         io->io_hdr.flags &= ~CTL_FLAG_DMA_INPROG;
1291                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1292                         io->io_hdr.port_status = msg->scsi.fetd_status;
1293                         io->scsiio.residual = msg->scsi.residual;
1294                         if (msg->hdr.status != CTL_STATUS_NONE) {
1295                                 io->io_hdr.status = msg->hdr.status;
1296                                 io->scsiio.scsi_status = msg->scsi.scsi_status;
1297                                 io->scsiio.sense_len = msg->scsi.sense_len;
1298                                 io->scsiio.sense_residual =msg->scsi.sense_residual;
1299                                 memcpy(&io->scsiio.sense_data,
1300                                     &msg->scsi.sense_data,
1301                                     msg->scsi.sense_len);
1302                                 if (msg->hdr.status == CTL_SUCCESS)
1303                                         io->io_hdr.flags |= CTL_FLAG_STATUS_SENT;
1304                         }
1305                         ctl_enqueue_isc(io);
1306                         break;
1307                 }
1308
1309                 /* Preformed on Originating SC, SER_ONLY mode */
1310                 case CTL_MSG_R2R:
1311                         io = msg->hdr.original_sc;
1312                         if (io == NULL) {
1313                                 printf("%s: original_sc == NULL!\n",
1314                                     __func__);
1315                                 break;
1316                         }
1317                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1318                         io->io_hdr.msg_type = CTL_MSG_R2R;
1319                         io->io_hdr.serializing_sc = msg->hdr.serializing_sc;
1320                         ctl_enqueue_isc(io);
1321                         break;
1322
1323                 /*
1324                  * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
1325                  * mode.
1326                  * Performed on the Originating (i.e. secondary) SC in XFER
1327                  * mode
1328                  */
1329                 case CTL_MSG_FINISH_IO:
1330                         if (softc->ha_mode == CTL_HA_MODE_XFER)
1331                                 ctl_isc_handler_finish_xfer(softc, msg);
1332                         else
1333                                 ctl_isc_handler_finish_ser_only(softc, msg);
1334                         break;
1335
1336                 /* Preformed on Originating SC */
1337                 case CTL_MSG_BAD_JUJU:
1338                         io = msg->hdr.original_sc;
1339                         if (io == NULL) {
1340                                 printf("%s: Bad JUJU!, original_sc is NULL!\n",
1341                                        __func__);
1342                                 break;
1343                         }
1344                         ctl_copy_sense_data(msg, io);
1345                         /*
1346                          * IO should have already been cleaned up on other
1347                          * SC so clear this flag so we won't send a message
1348                          * back to finish the IO there.
1349                          */
1350                         io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
1351                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1352
1353                         /* io = msg->hdr.serializing_sc; */
1354                         io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
1355                         ctl_enqueue_isc(io);
1356                         break;
1357
1358                 /* Handle resets sent from the other side */
1359                 case CTL_MSG_MANAGE_TASKS: {
1360                         struct ctl_taskio *taskio;
1361                         taskio = (struct ctl_taskio *)ctl_alloc_io(
1362                             softc->othersc_pool);
1363                         ctl_zero_io((union ctl_io *)taskio);
1364                         taskio->io_hdr.io_type = CTL_IO_TASK;
1365                         taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
1366                         taskio->io_hdr.nexus = msg->hdr.nexus;
1367                         taskio->task_action = msg->task.task_action;
1368                         taskio->tag_num = msg->task.tag_num;
1369                         taskio->tag_type = msg->task.tag_type;
1370 #ifdef CTL_TIME_IO
1371                         taskio->io_hdr.start_time = time_uptime;
1372                         getbintime(&taskio->io_hdr.start_bt);
1373 #endif /* CTL_TIME_IO */
1374                         ctl_run_task((union ctl_io *)taskio);
1375                         break;
1376                 }
1377                 /* Persistent Reserve action which needs attention */
1378                 case CTL_MSG_PERS_ACTION:
1379                         presio = (struct ctl_prio *)ctl_alloc_io(
1380                             softc->othersc_pool);
1381                         ctl_zero_io((union ctl_io *)presio);
1382                         presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
1383                         presio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
1384                         presio->io_hdr.nexus = msg->hdr.nexus;
1385                         presio->pr_msg = msg->pr;
1386                         ctl_enqueue_isc((union ctl_io *)presio);
1387                         break;
1388                 case CTL_MSG_UA:
1389                         ctl_isc_ua(softc, msg, param);
1390                         break;
1391                 case CTL_MSG_PORT_SYNC:
1392                         ctl_isc_port_sync(softc, msg, param);
1393                         break;
1394                 case CTL_MSG_LUN_SYNC:
1395                         ctl_isc_lun_sync(softc, msg, param);
1396                         break;
1397                 case CTL_MSG_IID_SYNC:
1398                         ctl_isc_iid_sync(softc, msg, param);
1399                         break;
1400                 case CTL_MSG_LOGIN:
1401                         ctl_isc_login(softc, msg, param);
1402                         break;
1403                 case CTL_MSG_MODE_SYNC:
1404                         ctl_isc_mode_sync(softc, msg, param);
1405                         break;
1406                 default:
1407                         printf("Received HA message of unknown type %d\n",
1408                             msg->hdr.msg_type);
1409                         ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1410                         break;
1411                 }
1412                 if (msg != &msgbuf)
1413                         free(msg, M_CTL);
1414         } else if (event == CTL_HA_EVT_LINK_CHANGE) {
1415                 printf("CTL: HA link status changed from %d to %d\n",
1416                     softc->ha_link, param);
1417                 if (param == softc->ha_link)
1418                         return;
1419                 if (softc->ha_link == CTL_HA_LINK_ONLINE) {
1420                         softc->ha_link = param;
1421                         ctl_isc_ha_link_down(softc);
1422                 } else {
1423                         softc->ha_link = param;
1424                         if (softc->ha_link == CTL_HA_LINK_ONLINE)
1425                                 ctl_isc_ha_link_up(softc);
1426                 }
1427                 return;
1428         } else {
1429                 printf("ctl_isc_event_handler: Unknown event %d\n", event);
1430                 return;
1431         }
1432 }
1433
1434 static void
1435 ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
1436 {
1437
1438         memcpy(&dest->scsiio.sense_data, &src->scsi.sense_data,
1439             src->scsi.sense_len);
1440         dest->scsiio.scsi_status = src->scsi.scsi_status;
1441         dest->scsiio.sense_len = src->scsi.sense_len;
1442         dest->io_hdr.status = src->hdr.status;
1443 }
1444
1445 static void
1446 ctl_copy_sense_data_back(union ctl_io *src, union ctl_ha_msg *dest)
1447 {
1448
1449         memcpy(&dest->scsi.sense_data, &src->scsiio.sense_data,
1450             src->scsiio.sense_len);
1451         dest->scsi.scsi_status = src->scsiio.scsi_status;
1452         dest->scsi.sense_len = src->scsiio.sense_len;
1453         dest->hdr.status = src->io_hdr.status;
1454 }
1455
1456 void
1457 ctl_est_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
1458 {
1459         struct ctl_softc *softc = lun->ctl_softc;
1460         ctl_ua_type *pu;
1461
1462         if (initidx < softc->init_min || initidx >= softc->init_max)
1463                 return;
1464         mtx_assert(&lun->lun_lock, MA_OWNED);
1465         pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
1466         if (pu == NULL)
1467                 return;
1468         pu[initidx % CTL_MAX_INIT_PER_PORT] |= ua;
1469 }
1470
1471 void
1472 ctl_est_ua_port(struct ctl_lun *lun, int port, uint32_t except, ctl_ua_type ua)
1473 {
1474         int i;
1475
1476         mtx_assert(&lun->lun_lock, MA_OWNED);
1477         if (lun->pending_ua[port] == NULL)
1478                 return;
1479         for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1480                 if (port * CTL_MAX_INIT_PER_PORT + i == except)
1481                         continue;
1482                 lun->pending_ua[port][i] |= ua;
1483         }
1484 }
1485
1486 void
1487 ctl_est_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1488 {
1489         struct ctl_softc *softc = lun->ctl_softc;
1490         int i;
1491
1492         mtx_assert(&lun->lun_lock, MA_OWNED);
1493         for (i = softc->port_min; i < softc->port_max; i++)
1494                 ctl_est_ua_port(lun, i, except, ua);
1495 }
1496
1497 void
1498 ctl_clr_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
1499 {
1500         struct ctl_softc *softc = lun->ctl_softc;
1501         ctl_ua_type *pu;
1502
1503         if (initidx < softc->init_min || initidx >= softc->init_max)
1504                 return;
1505         mtx_assert(&lun->lun_lock, MA_OWNED);
1506         pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
1507         if (pu == NULL)
1508                 return;
1509         pu[initidx % CTL_MAX_INIT_PER_PORT] &= ~ua;
1510 }
1511
1512 void
1513 ctl_clr_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1514 {
1515         struct ctl_softc *softc = lun->ctl_softc;
1516         int i, j;
1517
1518         mtx_assert(&lun->lun_lock, MA_OWNED);
1519         for (i = softc->port_min; i < softc->port_max; i++) {
1520                 if (lun->pending_ua[i] == NULL)
1521                         continue;
1522                 for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
1523                         if (i * CTL_MAX_INIT_PER_PORT + j == except)
1524                                 continue;
1525                         lun->pending_ua[i][j] &= ~ua;
1526                 }
1527         }
1528 }
1529
1530 void
1531 ctl_clr_ua_allluns(struct ctl_softc *ctl_softc, uint32_t initidx,
1532     ctl_ua_type ua_type)
1533 {
1534         struct ctl_lun *lun;
1535
1536         mtx_assert(&ctl_softc->ctl_lock, MA_OWNED);
1537         STAILQ_FOREACH(lun, &ctl_softc->lun_list, links) {
1538                 mtx_lock(&lun->lun_lock);
1539                 ctl_clr_ua(lun, initidx, ua_type);
1540                 mtx_unlock(&lun->lun_lock);
1541         }
1542 }
1543
1544 static int
1545 ctl_ha_role_sysctl(SYSCTL_HANDLER_ARGS)
1546 {
1547         struct ctl_softc *softc = (struct ctl_softc *)arg1;
1548         struct ctl_lun *lun;
1549         struct ctl_lun_req ireq;
1550         int error, value;
1551
1552         value = (softc->flags & CTL_FLAG_ACTIVE_SHELF) ? 0 : 1;
1553         error = sysctl_handle_int(oidp, &value, 0, req);
1554         if ((error != 0) || (req->newptr == NULL))
1555                 return (error);
1556
1557         mtx_lock(&softc->ctl_lock);
1558         if (value == 0)
1559                 softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1560         else
1561                 softc->flags &= ~CTL_FLAG_ACTIVE_SHELF;
1562         STAILQ_FOREACH(lun, &softc->lun_list, links) {
1563                 mtx_unlock(&softc->ctl_lock);
1564                 bzero(&ireq, sizeof(ireq));
1565                 ireq.reqtype = CTL_LUNREQ_MODIFY;
1566                 ireq.reqdata.modify.lun_id = lun->lun;
1567                 lun->backend->ioctl(NULL, CTL_LUN_REQ, (caddr_t)&ireq, 0,
1568                     curthread);
1569                 if (ireq.status != CTL_LUN_OK) {
1570                         printf("%s: CTL_LUNREQ_MODIFY returned %d '%s'\n",
1571                             __func__, ireq.status, ireq.error_str);
1572                 }
1573                 mtx_lock(&softc->ctl_lock);
1574         }
1575         mtx_unlock(&softc->ctl_lock);
1576         return (0);
1577 }
1578
1579 static int
1580 ctl_init(void)
1581 {
1582         struct ctl_softc *softc;
1583         void *other_pool;
1584         int i, error, retval;
1585
1586         retval = 0;
1587         control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
1588                                M_WAITOK | M_ZERO);
1589         softc = control_softc;
1590
1591         softc->dev = make_dev(&ctl_cdevsw, 0, UID_ROOT, GID_OPERATOR, 0600,
1592                               "cam/ctl");
1593
1594         softc->dev->si_drv1 = softc;
1595
1596         sysctl_ctx_init(&softc->sysctl_ctx);
1597         softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
1598                 SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
1599                 CTLFLAG_RD, 0, "CAM Target Layer");
1600
1601         if (softc->sysctl_tree == NULL) {
1602                 printf("%s: unable to allocate sysctl tree\n", __func__);
1603                 destroy_dev(softc->dev);
1604                 free(control_softc, M_DEVBUF);
1605                 control_softc = NULL;
1606                 return (ENOMEM);
1607         }
1608
1609         mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
1610         softc->io_zone = uma_zcreate("CTL IO", sizeof(union ctl_io),
1611             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1612         softc->open_count = 0;
1613
1614         /*
1615          * Default to actually sending a SYNCHRONIZE CACHE command down to
1616          * the drive.
1617          */
1618         softc->flags = CTL_FLAG_REAL_SYNC;
1619
1620         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1621             OID_AUTO, "ha_mode", CTLFLAG_RDTUN, (int *)&softc->ha_mode, 0,
1622             "HA mode (0 - act/stby, 1 - serialize only, 2 - xfer)");
1623
1624         /*
1625          * In Copan's HA scheme, the "master" and "slave" roles are
1626          * figured out through the slot the controller is in.  Although it
1627          * is an active/active system, someone has to be in charge.
1628          */
1629         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1630             OID_AUTO, "ha_id", CTLFLAG_RDTUN, &softc->ha_id, 0,
1631             "HA head ID (0 - no HA)");
1632         if (softc->ha_id == 0 || softc->ha_id > NUM_TARGET_PORT_GROUPS) {
1633                 softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1634                 softc->is_single = 1;
1635                 softc->port_cnt = CTL_MAX_PORTS;
1636                 softc->port_min = 0;
1637         } else {
1638                 softc->port_cnt = CTL_MAX_PORTS / NUM_TARGET_PORT_GROUPS;
1639                 softc->port_min = (softc->ha_id - 1) * softc->port_cnt;
1640         }
1641         softc->port_max = softc->port_min + softc->port_cnt;
1642         softc->init_min = softc->port_min * CTL_MAX_INIT_PER_PORT;
1643         softc->init_max = softc->port_max * CTL_MAX_INIT_PER_PORT;
1644
1645         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1646             OID_AUTO, "ha_link", CTLFLAG_RD, (int *)&softc->ha_link, 0,
1647             "HA link state (0 - offline, 1 - unknown, 2 - online)");
1648
1649         STAILQ_INIT(&softc->lun_list);
1650         STAILQ_INIT(&softc->pending_lun_queue);
1651         STAILQ_INIT(&softc->fe_list);
1652         STAILQ_INIT(&softc->port_list);
1653         STAILQ_INIT(&softc->be_list);
1654         ctl_tpc_init(softc);
1655
1656         if (ctl_pool_create(softc, "othersc", CTL_POOL_ENTRIES_OTHER_SC,
1657                             &other_pool) != 0)
1658         {
1659                 printf("ctl: can't allocate %d entry other SC pool, "
1660                        "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1661                 return (ENOMEM);
1662         }
1663         softc->othersc_pool = other_pool;
1664
1665         if (worker_threads <= 0)
1666                 worker_threads = max(1, mp_ncpus / 4);
1667         if (worker_threads > CTL_MAX_THREADS)
1668                 worker_threads = CTL_MAX_THREADS;
1669
1670         for (i = 0; i < worker_threads; i++) {
1671                 struct ctl_thread *thr = &softc->threads[i];
1672
1673                 mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1674                 thr->ctl_softc = softc;
1675                 STAILQ_INIT(&thr->incoming_queue);
1676                 STAILQ_INIT(&thr->rtr_queue);
1677                 STAILQ_INIT(&thr->done_queue);
1678                 STAILQ_INIT(&thr->isc_queue);
1679
1680                 error = kproc_kthread_add(ctl_work_thread, thr,
1681                     &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1682                 if (error != 0) {
1683                         printf("error creating CTL work thread!\n");
1684                         ctl_pool_free(other_pool);
1685                         return (error);
1686                 }
1687         }
1688         error = kproc_kthread_add(ctl_lun_thread, softc,
1689             &softc->ctl_proc, NULL, 0, 0, "ctl", "lun");
1690         if (error != 0) {
1691                 printf("error creating CTL lun thread!\n");
1692                 ctl_pool_free(other_pool);
1693                 return (error);
1694         }
1695         error = kproc_kthread_add(ctl_thresh_thread, softc,
1696             &softc->ctl_proc, NULL, 0, 0, "ctl", "thresh");
1697         if (error != 0) {
1698                 printf("error creating CTL threshold thread!\n");
1699                 ctl_pool_free(other_pool);
1700                 return (error);
1701         }
1702
1703         SYSCTL_ADD_PROC(&softc->sysctl_ctx,SYSCTL_CHILDREN(softc->sysctl_tree),
1704             OID_AUTO, "ha_role", CTLTYPE_INT | CTLFLAG_RWTUN,
1705             softc, 0, ctl_ha_role_sysctl, "I", "HA role for this head");
1706
1707         if (softc->is_single == 0) {
1708                 ctl_frontend_register(&ha_frontend);
1709                 if (ctl_ha_msg_init(softc) != CTL_HA_STATUS_SUCCESS) {
1710                         printf("ctl_init: ctl_ha_msg_init failed.\n");
1711                         softc->is_single = 1;
1712                 } else
1713                 if (ctl_ha_msg_register(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
1714                     != CTL_HA_STATUS_SUCCESS) {
1715                         printf("ctl_init: ctl_ha_msg_register failed.\n");
1716                         softc->is_single = 1;
1717                 }
1718         }
1719         return (0);
1720 }
1721
1722 void
1723 ctl_shutdown(void)
1724 {
1725         struct ctl_softc *softc;
1726         struct ctl_lun *lun, *next_lun;
1727
1728         softc = (struct ctl_softc *)control_softc;
1729
1730         if (softc->is_single == 0) {
1731                 ctl_ha_msg_shutdown(softc);
1732                 if (ctl_ha_msg_deregister(CTL_HA_CHAN_CTL)
1733                     != CTL_HA_STATUS_SUCCESS)
1734                         printf("%s: ctl_ha_msg_deregister failed.\n", __func__);
1735                 if (ctl_ha_msg_destroy(softc) != CTL_HA_STATUS_SUCCESS)
1736                         printf("%s: ctl_ha_msg_destroy failed.\n", __func__);
1737                 ctl_frontend_deregister(&ha_frontend);
1738         }
1739
1740         mtx_lock(&softc->ctl_lock);
1741
1742         /*
1743          * Free up each LUN.
1744          */
1745         for (lun = STAILQ_FIRST(&softc->lun_list); lun != NULL; lun = next_lun){
1746                 next_lun = STAILQ_NEXT(lun, links);
1747                 ctl_free_lun(lun);
1748         }
1749
1750         mtx_unlock(&softc->ctl_lock);
1751
1752 #if 0
1753         ctl_shutdown_thread(softc->work_thread);
1754         mtx_destroy(&softc->queue_lock);
1755 #endif
1756
1757         ctl_tpc_shutdown(softc);
1758         uma_zdestroy(softc->io_zone);
1759         mtx_destroy(&softc->ctl_lock);
1760
1761         destroy_dev(softc->dev);
1762
1763         sysctl_ctx_free(&softc->sysctl_ctx);
1764
1765         free(control_softc, M_DEVBUF);
1766         control_softc = NULL;
1767 }
1768
1769 static int
1770 ctl_module_event_handler(module_t mod, int what, void *arg)
1771 {
1772
1773         switch (what) {
1774         case MOD_LOAD:
1775                 return (ctl_init());
1776         case MOD_UNLOAD:
1777                 return (EBUSY);
1778         default:
1779                 return (EOPNOTSUPP);
1780         }
1781 }
1782
1783 /*
1784  * XXX KDM should we do some access checks here?  Bump a reference count to
1785  * prevent a CTL module from being unloaded while someone has it open?
1786  */
1787 static int
1788 ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1789 {
1790         return (0);
1791 }
1792
1793 static int
1794 ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1795 {
1796         return (0);
1797 }
1798
1799 /*
1800  * Remove an initiator by port number and initiator ID.
1801  * Returns 0 for success, -1 for failure.
1802  */
1803 int
1804 ctl_remove_initiator(struct ctl_port *port, int iid)
1805 {
1806         struct ctl_softc *softc = control_softc;
1807
1808         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1809
1810         if (iid > CTL_MAX_INIT_PER_PORT) {
1811                 printf("%s: initiator ID %u > maximun %u!\n",
1812                        __func__, iid, CTL_MAX_INIT_PER_PORT);
1813                 return (-1);
1814         }
1815
1816         mtx_lock(&softc->ctl_lock);
1817         port->wwpn_iid[iid].in_use--;
1818         port->wwpn_iid[iid].last_use = time_uptime;
1819         mtx_unlock(&softc->ctl_lock);
1820         ctl_isc_announce_iid(port, iid);
1821
1822         return (0);
1823 }
1824
1825 /*
1826  * Add an initiator to the initiator map.
1827  * Returns iid for success, < 0 for failure.
1828  */
1829 int
1830 ctl_add_initiator(struct ctl_port *port, int iid, uint64_t wwpn, char *name)
1831 {
1832         struct ctl_softc *softc = control_softc;
1833         time_t best_time;
1834         int i, best;
1835
1836         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1837
1838         if (iid >= CTL_MAX_INIT_PER_PORT) {
1839                 printf("%s: WWPN %#jx initiator ID %u > maximum %u!\n",
1840                        __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
1841                 free(name, M_CTL);
1842                 return (-1);
1843         }
1844
1845         mtx_lock(&softc->ctl_lock);
1846
1847         if (iid < 0 && (wwpn != 0 || name != NULL)) {
1848                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1849                         if (wwpn != 0 && wwpn == port->wwpn_iid[i].wwpn) {
1850                                 iid = i;
1851                                 break;
1852                         }
1853                         if (name != NULL && port->wwpn_iid[i].name != NULL &&
1854                             strcmp(name, port->wwpn_iid[i].name) == 0) {
1855                                 iid = i;
1856                                 break;
1857                         }
1858                 }
1859         }
1860
1861         if (iid < 0) {
1862                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1863                         if (port->wwpn_iid[i].in_use == 0 &&
1864                             port->wwpn_iid[i].wwpn == 0 &&
1865                             port->wwpn_iid[i].name == NULL) {
1866                                 iid = i;
1867                                 break;
1868                         }
1869                 }
1870         }
1871
1872         if (iid < 0) {
1873                 best = -1;
1874                 best_time = INT32_MAX;
1875                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1876                         if (port->wwpn_iid[i].in_use == 0) {
1877                                 if (port->wwpn_iid[i].last_use < best_time) {
1878                                         best = i;
1879                                         best_time = port->wwpn_iid[i].last_use;
1880                                 }
1881                         }
1882                 }
1883                 iid = best;
1884         }
1885
1886         if (iid < 0) {
1887                 mtx_unlock(&softc->ctl_lock);
1888                 free(name, M_CTL);
1889                 return (-2);
1890         }
1891
1892         if (port->wwpn_iid[iid].in_use > 0 && (wwpn != 0 || name != NULL)) {
1893                 /*
1894                  * This is not an error yet.
1895                  */
1896                 if (wwpn != 0 && wwpn == port->wwpn_iid[iid].wwpn) {
1897 #if 0
1898                         printf("%s: port %d iid %u WWPN %#jx arrived"
1899                             " again\n", __func__, port->targ_port,
1900                             iid, (uintmax_t)wwpn);
1901 #endif
1902                         goto take;
1903                 }
1904                 if (name != NULL && port->wwpn_iid[iid].name != NULL &&
1905                     strcmp(name, port->wwpn_iid[iid].name) == 0) {
1906 #if 0
1907                         printf("%s: port %d iid %u name '%s' arrived"
1908                             " again\n", __func__, port->targ_port,
1909                             iid, name);
1910 #endif
1911                         goto take;
1912                 }
1913
1914                 /*
1915                  * This is an error, but what do we do about it?  The
1916                  * driver is telling us we have a new WWPN for this
1917                  * initiator ID, so we pretty much need to use it.
1918                  */
1919                 printf("%s: port %d iid %u WWPN %#jx '%s' arrived,"
1920                     " but WWPN %#jx '%s' is still at that address\n",
1921                     __func__, port->targ_port, iid, wwpn, name,
1922                     (uintmax_t)port->wwpn_iid[iid].wwpn,
1923                     port->wwpn_iid[iid].name);
1924
1925                 /*
1926                  * XXX KDM clear have_ca and ua_pending on each LUN for
1927                  * this initiator.
1928                  */
1929         }
1930 take:
1931         free(port->wwpn_iid[iid].name, M_CTL);
1932         port->wwpn_iid[iid].name = name;
1933         port->wwpn_iid[iid].wwpn = wwpn;
1934         port->wwpn_iid[iid].in_use++;
1935         mtx_unlock(&softc->ctl_lock);
1936         ctl_isc_announce_iid(port, iid);
1937
1938         return (iid);
1939 }
1940
1941 static int
1942 ctl_create_iid(struct ctl_port *port, int iid, uint8_t *buf)
1943 {
1944         int len;
1945
1946         switch (port->port_type) {
1947         case CTL_PORT_FC:
1948         {
1949                 struct scsi_transportid_fcp *id =
1950                     (struct scsi_transportid_fcp *)buf;
1951                 if (port->wwpn_iid[iid].wwpn == 0)
1952                         return (0);
1953                 memset(id, 0, sizeof(*id));
1954                 id->format_protocol = SCSI_PROTO_FC;
1955                 scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->n_port_name);
1956                 return (sizeof(*id));
1957         }
1958         case CTL_PORT_ISCSI:
1959         {
1960                 struct scsi_transportid_iscsi_port *id =
1961                     (struct scsi_transportid_iscsi_port *)buf;
1962                 if (port->wwpn_iid[iid].name == NULL)
1963                         return (0);
1964                 memset(id, 0, 256);
1965                 id->format_protocol = SCSI_TRN_ISCSI_FORMAT_PORT |
1966                     SCSI_PROTO_ISCSI;
1967                 len = strlcpy(id->iscsi_name, port->wwpn_iid[iid].name, 252) + 1;
1968                 len = roundup2(min(len, 252), 4);
1969                 scsi_ulto2b(len, id->additional_length);
1970                 return (sizeof(*id) + len);
1971         }
1972         case CTL_PORT_SAS:
1973         {
1974                 struct scsi_transportid_sas *id =
1975                     (struct scsi_transportid_sas *)buf;
1976                 if (port->wwpn_iid[iid].wwpn == 0)
1977                         return (0);
1978                 memset(id, 0, sizeof(*id));
1979                 id->format_protocol = SCSI_PROTO_SAS;
1980                 scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->sas_address);
1981                 return (sizeof(*id));
1982         }
1983         default:
1984         {
1985                 struct scsi_transportid_spi *id =
1986                     (struct scsi_transportid_spi *)buf;
1987                 memset(id, 0, sizeof(*id));
1988                 id->format_protocol = SCSI_PROTO_SPI;
1989                 scsi_ulto2b(iid, id->scsi_addr);
1990                 scsi_ulto2b(port->targ_port, id->rel_trgt_port_id);
1991                 return (sizeof(*id));
1992         }
1993         }
1994 }
1995
1996 /*
1997  * Serialize a command that went down the "wrong" side, and so was sent to
1998  * this controller for execution.  The logic is a little different than the
1999  * standard case in ctl_scsiio_precheck().  Errors in this case need to get
2000  * sent back to the other side, but in the success case, we execute the
2001  * command on this side (XFER mode) or tell the other side to execute it
2002  * (SER_ONLY mode).
2003  */
2004 static int
2005 ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
2006 {
2007         struct ctl_softc *softc;
2008         union ctl_ha_msg msg_info;
2009         struct ctl_lun *lun;
2010         const struct ctl_cmd_entry *entry;
2011         int retval = 0;
2012         uint32_t targ_lun;
2013
2014         softc = control_softc;
2015
2016         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
2017         mtx_lock(&softc->ctl_lock);
2018         if ((targ_lun < CTL_MAX_LUNS) &&
2019             ((lun = softc->ctl_luns[targ_lun]) != NULL)) {
2020                 mtx_lock(&lun->lun_lock);
2021                 mtx_unlock(&softc->ctl_lock);
2022                 /*
2023                  * If the LUN is invalid, pretend that it doesn't exist.
2024                  * It will go away as soon as all pending I/O has been
2025                  * completed.
2026                  */
2027                 if (lun->flags & CTL_LUN_DISABLED) {
2028                         mtx_unlock(&lun->lun_lock);
2029                         lun = NULL;
2030                 }
2031         } else {
2032                 mtx_unlock(&softc->ctl_lock);
2033                 lun = NULL;
2034         }
2035         if (lun == NULL) {
2036                 /*
2037                  * The other node would not send this request to us unless
2038                  * received announce that we are primary node for this LUN.
2039                  * If this LUN does not exist now, it is probably result of
2040                  * a race, so respond to initiator in the most opaque way.
2041                  */
2042                 ctl_set_busy(ctsio);
2043                 ctl_copy_sense_data_back((union ctl_io *)ctsio, &msg_info);
2044                 msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
2045                 msg_info.hdr.serializing_sc = NULL;
2046                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
2047                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
2048                     sizeof(msg_info.scsi), M_WAITOK);
2049                 return(1);
2050         }
2051
2052         entry = ctl_get_cmd_entry(ctsio, NULL);
2053         if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
2054                 mtx_unlock(&lun->lun_lock);
2055                 ctl_copy_sense_data_back((union ctl_io *)ctsio, &msg_info);
2056                 msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
2057                 msg_info.hdr.serializing_sc = NULL;
2058                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
2059                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
2060                     sizeof(msg_info.scsi), M_WAITOK);
2061                 return(1);
2062         }
2063
2064         ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
2065         ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = lun->be_lun;
2066
2067         /*
2068          * Every I/O goes into the OOA queue for a
2069          * particular LUN, and stays there until completion.
2070          */
2071 #ifdef CTL_TIME_IO
2072         if (TAILQ_EMPTY(&lun->ooa_queue))
2073                 lun->idle_time += getsbinuptime() - lun->last_busy;
2074 #endif
2075         TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2076
2077         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
2078                 (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
2079                  ooa_links))) {
2080         case CTL_ACTION_BLOCK:
2081                 ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
2082                 TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
2083                                   blocked_links);
2084                 mtx_unlock(&lun->lun_lock);
2085                 break;
2086         case CTL_ACTION_PASS:
2087         case CTL_ACTION_SKIP:
2088                 if (softc->ha_mode == CTL_HA_MODE_XFER) {
2089                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
2090                         ctl_enqueue_rtr((union ctl_io *)ctsio);
2091                         mtx_unlock(&lun->lun_lock);
2092                 } else {
2093                         ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
2094                         mtx_unlock(&lun->lun_lock);
2095
2096                         /* send msg back to other side */
2097                         msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
2098                         msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
2099                         msg_info.hdr.msg_type = CTL_MSG_R2R;
2100                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
2101                             sizeof(msg_info.hdr), M_WAITOK);
2102                 }
2103                 break;
2104         case CTL_ACTION_OVERLAP:
2105                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2106                 mtx_unlock(&lun->lun_lock);
2107                 retval = 1;
2108
2109                 ctl_set_overlapped_cmd(ctsio);
2110                 ctl_copy_sense_data_back((union ctl_io *)ctsio, &msg_info);
2111                 msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
2112                 msg_info.hdr.serializing_sc = NULL;
2113                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
2114                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
2115                     sizeof(msg_info.scsi), M_WAITOK);
2116                 break;
2117         case CTL_ACTION_OVERLAP_TAG:
2118                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2119                 mtx_unlock(&lun->lun_lock);
2120                 retval = 1;
2121                 ctl_set_overlapped_tag(ctsio, ctsio->tag_num);
2122                 ctl_copy_sense_data_back((union ctl_io *)ctsio, &msg_info);
2123                 msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
2124                 msg_info.hdr.serializing_sc = NULL;
2125                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
2126                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
2127                     sizeof(msg_info.scsi), M_WAITOK);
2128                 break;
2129         case CTL_ACTION_ERROR:
2130         default:
2131                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2132                 mtx_unlock(&lun->lun_lock);
2133                 retval = 1;
2134
2135                 ctl_set_internal_failure(ctsio, /*sks_valid*/ 0,
2136                                          /*retry_count*/ 0);
2137                 ctl_copy_sense_data_back((union ctl_io *)ctsio, &msg_info);
2138                 msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
2139                 msg_info.hdr.serializing_sc = NULL;
2140                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
2141                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
2142                     sizeof(msg_info.scsi), M_WAITOK);
2143                 break;
2144         }
2145         return (retval);
2146 }
2147
2148 /*
2149  * Returns 0 for success, errno for failure.
2150  */
2151 static int
2152 ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
2153                    struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
2154 {
2155         union ctl_io *io;
2156         int retval;
2157
2158         retval = 0;
2159
2160         mtx_lock(&lun->lun_lock);
2161         for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
2162              (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2163              ooa_links)) {
2164                 struct ctl_ooa_entry *entry;
2165
2166                 /*
2167                  * If we've got more than we can fit, just count the
2168                  * remaining entries.
2169                  */
2170                 if (*cur_fill_num >= ooa_hdr->alloc_num)
2171                         continue;
2172
2173                 entry = &kern_entries[*cur_fill_num];
2174
2175                 entry->tag_num = io->scsiio.tag_num;
2176                 entry->lun_num = lun->lun;
2177 #ifdef CTL_TIME_IO
2178                 entry->start_bt = io->io_hdr.start_bt;
2179 #endif
2180                 bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2181                 entry->cdb_len = io->scsiio.cdb_len;
2182                 if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
2183                         entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2184
2185                 if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2186                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2187
2188                 if (io->io_hdr.flags & CTL_FLAG_ABORT)
2189                         entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2190
2191                 if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2192                         entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2193
2194                 if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2195                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2196         }
2197         mtx_unlock(&lun->lun_lock);
2198
2199         return (retval);
2200 }
2201
2202 static void *
2203 ctl_copyin_alloc(void *user_addr, int len, char *error_str,
2204                  size_t error_str_len)
2205 {
2206         void *kptr;
2207
2208         kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2209
2210         if (copyin(user_addr, kptr, len) != 0) {
2211                 snprintf(error_str, error_str_len, "Error copying %d bytes "
2212                          "from user address %p to kernel address %p", len,
2213                          user_addr, kptr);
2214                 free(kptr, M_CTL);
2215                 return (NULL);
2216         }
2217
2218         return (kptr);
2219 }
2220
2221 static void
2222 ctl_free_args(int num_args, struct ctl_be_arg *args)
2223 {
2224         int i;
2225
2226         if (args == NULL)
2227                 return;
2228
2229         for (i = 0; i < num_args; i++) {
2230                 free(args[i].kname, M_CTL);
2231                 free(args[i].kvalue, M_CTL);
2232         }
2233
2234         free(args, M_CTL);
2235 }
2236
2237 static struct ctl_be_arg *
2238 ctl_copyin_args(int num_args, struct ctl_be_arg *uargs,
2239                 char *error_str, size_t error_str_len)
2240 {
2241         struct ctl_be_arg *args;
2242         int i;
2243
2244         args = ctl_copyin_alloc(uargs, num_args * sizeof(*args),
2245                                 error_str, error_str_len);
2246
2247         if (args == NULL)
2248                 goto bailout;
2249
2250         for (i = 0; i < num_args; i++) {
2251                 args[i].kname = NULL;
2252                 args[i].kvalue = NULL;
2253         }
2254
2255         for (i = 0; i < num_args; i++) {
2256                 uint8_t *tmpptr;
2257
2258                 args[i].kname = ctl_copyin_alloc(args[i].name,
2259                         args[i].namelen, error_str, error_str_len);
2260                 if (args[i].kname == NULL)
2261                         goto bailout;
2262
2263                 if (args[i].kname[args[i].namelen - 1] != '\0') {
2264                         snprintf(error_str, error_str_len, "Argument %d "
2265                                  "name is not NUL-terminated", i);
2266                         goto bailout;
2267                 }
2268
2269                 if (args[i].flags & CTL_BEARG_RD) {
2270                         tmpptr = ctl_copyin_alloc(args[i].value,
2271                                 args[i].vallen, error_str, error_str_len);
2272                         if (tmpptr == NULL)
2273                                 goto bailout;
2274                         if ((args[i].flags & CTL_BEARG_ASCII)
2275                          && (tmpptr[args[i].vallen - 1] != '\0')) {
2276                                 snprintf(error_str, error_str_len, "Argument "
2277                                     "%d value is not NUL-terminated", i);
2278                                 goto bailout;
2279                         }
2280                         args[i].kvalue = tmpptr;
2281                 } else {
2282                         args[i].kvalue = malloc(args[i].vallen,
2283                             M_CTL, M_WAITOK | M_ZERO);
2284                 }
2285         }
2286
2287         return (args);
2288 bailout:
2289
2290         ctl_free_args(num_args, args);
2291
2292         return (NULL);
2293 }
2294
2295 static void
2296 ctl_copyout_args(int num_args, struct ctl_be_arg *args)
2297 {
2298         int i;
2299
2300         for (i = 0; i < num_args; i++) {
2301                 if (args[i].flags & CTL_BEARG_WR)
2302                         copyout(args[i].kvalue, args[i].value, args[i].vallen);
2303         }
2304 }
2305
2306 /*
2307  * Escape characters that are illegal or not recommended in XML.
2308  */
2309 int
2310 ctl_sbuf_printf_esc(struct sbuf *sb, char *str, int size)
2311 {
2312         char *end = str + size;
2313         int retval;
2314
2315         retval = 0;
2316
2317         for (; *str && str < end; str++) {
2318                 switch (*str) {
2319                 case '&':
2320                         retval = sbuf_printf(sb, "&amp;");
2321                         break;
2322                 case '>':
2323                         retval = sbuf_printf(sb, "&gt;");
2324                         break;
2325                 case '<':
2326                         retval = sbuf_printf(sb, "&lt;");
2327                         break;
2328                 default:
2329                         retval = sbuf_putc(sb, *str);
2330                         break;
2331                 }
2332
2333                 if (retval != 0)
2334                         break;
2335
2336         }
2337
2338         return (retval);
2339 }
2340
2341 static void
2342 ctl_id_sbuf(struct ctl_devid *id, struct sbuf *sb)
2343 {
2344         struct scsi_vpd_id_descriptor *desc;
2345         int i;
2346
2347         if (id == NULL || id->len < 4)
2348                 return;
2349         desc = (struct scsi_vpd_id_descriptor *)id->data;
2350         switch (desc->id_type & SVPD_ID_TYPE_MASK) {
2351         case SVPD_ID_TYPE_T10:
2352                 sbuf_printf(sb, "t10.");
2353                 break;
2354         case SVPD_ID_TYPE_EUI64:
2355                 sbuf_printf(sb, "eui.");
2356                 break;
2357         case SVPD_ID_TYPE_NAA:
2358                 sbuf_printf(sb, "naa.");
2359                 break;
2360         case SVPD_ID_TYPE_SCSI_NAME:
2361                 break;
2362         }
2363         switch (desc->proto_codeset & SVPD_ID_CODESET_MASK) {
2364         case SVPD_ID_CODESET_BINARY:
2365                 for (i = 0; i < desc->length; i++)
2366                         sbuf_printf(sb, "%02x", desc->identifier[i]);
2367                 break;
2368         case SVPD_ID_CODESET_ASCII:
2369                 sbuf_printf(sb, "%.*s", (int)desc->length,
2370                     (char *)desc->identifier);
2371                 break;
2372         case SVPD_ID_CODESET_UTF8:
2373                 sbuf_printf(sb, "%s", (char *)desc->identifier);
2374                 break;
2375         }
2376 }
2377
2378 static int
2379 ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2380           struct thread *td)
2381 {
2382         struct ctl_softc *softc;
2383         struct ctl_lun *lun;
2384         int retval;
2385
2386         softc = control_softc;
2387
2388         retval = 0;
2389
2390         switch (cmd) {
2391         case CTL_IO:
2392                 retval = ctl_ioctl_io(dev, cmd, addr, flag, td);
2393                 break;
2394         case CTL_ENABLE_PORT:
2395         case CTL_DISABLE_PORT:
2396         case CTL_SET_PORT_WWNS: {
2397                 struct ctl_port *port;
2398                 struct ctl_port_entry *entry;
2399
2400                 entry = (struct ctl_port_entry *)addr;
2401                 
2402                 mtx_lock(&softc->ctl_lock);
2403                 STAILQ_FOREACH(port, &softc->port_list, links) {
2404                         int action, done;
2405
2406                         if (port->targ_port < softc->port_min ||
2407                             port->targ_port >= softc->port_max)
2408                                 continue;
2409
2410                         action = 0;
2411                         done = 0;
2412                         if ((entry->port_type == CTL_PORT_NONE)
2413                          && (entry->targ_port == port->targ_port)) {
2414                                 /*
2415                                  * If the user only wants to enable or
2416                                  * disable or set WWNs on a specific port,
2417                                  * do the operation and we're done.
2418                                  */
2419                                 action = 1;
2420                                 done = 1;
2421                         } else if (entry->port_type & port->port_type) {
2422                                 /*
2423                                  * Compare the user's type mask with the
2424                                  * particular frontend type to see if we
2425                                  * have a match.
2426                                  */
2427                                 action = 1;
2428                                 done = 0;
2429
2430                                 /*
2431                                  * Make sure the user isn't trying to set
2432                                  * WWNs on multiple ports at the same time.
2433                                  */
2434                                 if (cmd == CTL_SET_PORT_WWNS) {
2435                                         printf("%s: Can't set WWNs on "
2436                                                "multiple ports\n", __func__);
2437                                         retval = EINVAL;
2438                                         break;
2439                                 }
2440                         }
2441                         if (action == 0)
2442                                 continue;
2443
2444                         /*
2445                          * XXX KDM we have to drop the lock here, because
2446                          * the online/offline operations can potentially
2447                          * block.  We need to reference count the frontends
2448                          * so they can't go away,
2449                          */
2450                         if (cmd == CTL_ENABLE_PORT) {
2451                                 mtx_unlock(&softc->ctl_lock);
2452                                 ctl_port_online(port);
2453                                 mtx_lock(&softc->ctl_lock);
2454                         } else if (cmd == CTL_DISABLE_PORT) {
2455                                 mtx_unlock(&softc->ctl_lock);
2456                                 ctl_port_offline(port);
2457                                 mtx_lock(&softc->ctl_lock);
2458                         } else if (cmd == CTL_SET_PORT_WWNS) {
2459                                 ctl_port_set_wwns(port,
2460                                     (entry->flags & CTL_PORT_WWNN_VALID) ?
2461                                     1 : 0, entry->wwnn,
2462                                     (entry->flags & CTL_PORT_WWPN_VALID) ?
2463                                     1 : 0, entry->wwpn);
2464                         }
2465                         if (done != 0)
2466                                 break;
2467                 }
2468                 mtx_unlock(&softc->ctl_lock);
2469                 break;
2470         }
2471         case CTL_GET_PORT_LIST: {
2472                 struct ctl_port *port;
2473                 struct ctl_port_list *list;
2474                 int i;
2475
2476                 list = (struct ctl_port_list *)addr;
2477
2478                 if (list->alloc_len != (list->alloc_num *
2479                     sizeof(struct ctl_port_entry))) {
2480                         printf("%s: CTL_GET_PORT_LIST: alloc_len %u != "
2481                                "alloc_num %u * sizeof(struct ctl_port_entry) "
2482                                "%zu\n", __func__, list->alloc_len,
2483                                list->alloc_num, sizeof(struct ctl_port_entry));
2484                         retval = EINVAL;
2485                         break;
2486                 }
2487                 list->fill_len = 0;
2488                 list->fill_num = 0;
2489                 list->dropped_num = 0;
2490                 i = 0;
2491                 mtx_lock(&softc->ctl_lock);
2492                 STAILQ_FOREACH(port, &softc->port_list, links) {
2493                         struct ctl_port_entry entry, *list_entry;
2494
2495                         if (list->fill_num >= list->alloc_num) {
2496                                 list->dropped_num++;
2497                                 continue;
2498                         }
2499
2500                         entry.port_type = port->port_type;
2501                         strlcpy(entry.port_name, port->port_name,
2502                                 sizeof(entry.port_name));
2503                         entry.targ_port = port->targ_port;
2504                         entry.physical_port = port->physical_port;
2505                         entry.virtual_port = port->virtual_port;
2506                         entry.wwnn = port->wwnn;
2507                         entry.wwpn = port->wwpn;
2508                         if (port->status & CTL_PORT_STATUS_ONLINE)
2509                                 entry.online = 1;
2510                         else
2511                                 entry.online = 0;
2512
2513                         list_entry = &list->entries[i];
2514
2515                         retval = copyout(&entry, list_entry, sizeof(entry));
2516                         if (retval != 0) {
2517                                 printf("%s: CTL_GET_PORT_LIST: copyout "
2518                                        "returned %d\n", __func__, retval);
2519                                 break;
2520                         }
2521                         i++;
2522                         list->fill_num++;
2523                         list->fill_len += sizeof(entry);
2524                 }
2525                 mtx_unlock(&softc->ctl_lock);
2526
2527                 /*
2528                  * If this is non-zero, we had a copyout fault, so there's
2529                  * probably no point in attempting to set the status inside
2530                  * the structure.
2531                  */
2532                 if (retval != 0)
2533                         break;
2534
2535                 if (list->dropped_num > 0)
2536                         list->status = CTL_PORT_LIST_NEED_MORE_SPACE;
2537                 else
2538                         list->status = CTL_PORT_LIST_OK;
2539                 break;
2540         }
2541         case CTL_DUMP_OOA: {
2542                 union ctl_io *io;
2543                 char printbuf[128];
2544                 struct sbuf sb;
2545
2546                 mtx_lock(&softc->ctl_lock);
2547                 printf("Dumping OOA queues:\n");
2548                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
2549                         mtx_lock(&lun->lun_lock);
2550                         for (io = (union ctl_io *)TAILQ_FIRST(
2551                              &lun->ooa_queue); io != NULL;
2552                              io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2553                              ooa_links)) {
2554                                 sbuf_new(&sb, printbuf, sizeof(printbuf),
2555                                          SBUF_FIXEDLEN);
2556                                 sbuf_printf(&sb, "LUN %jd tag 0x%04x%s%s%s%s: ",
2557                                             (intmax_t)lun->lun,
2558                                             io->scsiio.tag_num,
2559                                             (io->io_hdr.flags &
2560                                             CTL_FLAG_BLOCKED) ? "" : " BLOCKED",
2561                                             (io->io_hdr.flags &
2562                                             CTL_FLAG_DMA_INPROG) ? " DMA" : "",
2563                                             (io->io_hdr.flags &
2564                                             CTL_FLAG_ABORT) ? " ABORT" : "",
2565                                             (io->io_hdr.flags &
2566                                         CTL_FLAG_IS_WAS_ON_RTR) ? " RTR" : "");
2567                                 ctl_scsi_command_string(&io->scsiio, NULL, &sb);
2568                                 sbuf_finish(&sb);
2569                                 printf("%s\n", sbuf_data(&sb));
2570                         }
2571                         mtx_unlock(&lun->lun_lock);
2572                 }
2573                 printf("OOA queues dump done\n");
2574                 mtx_unlock(&softc->ctl_lock);
2575                 break;
2576         }
2577         case CTL_GET_OOA: {
2578                 struct ctl_ooa *ooa_hdr;
2579                 struct ctl_ooa_entry *entries;
2580                 uint32_t cur_fill_num;
2581
2582                 ooa_hdr = (struct ctl_ooa *)addr;
2583
2584                 if ((ooa_hdr->alloc_len == 0)
2585                  || (ooa_hdr->alloc_num == 0)) {
2586                         printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2587                                "must be non-zero\n", __func__,
2588                                ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2589                         retval = EINVAL;
2590                         break;
2591                 }
2592
2593                 if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2594                     sizeof(struct ctl_ooa_entry))) {
2595                         printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2596                                "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2597                                __func__, ooa_hdr->alloc_len,
2598                                ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2599                         retval = EINVAL;
2600                         break;
2601                 }
2602
2603                 entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2604                 if (entries == NULL) {
2605                         printf("%s: could not allocate %d bytes for OOA "
2606                                "dump\n", __func__, ooa_hdr->alloc_len);
2607                         retval = ENOMEM;
2608                         break;
2609                 }
2610
2611                 mtx_lock(&softc->ctl_lock);
2612                 if (((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0)
2613                  && ((ooa_hdr->lun_num >= CTL_MAX_LUNS)
2614                   || (softc->ctl_luns[ooa_hdr->lun_num] == NULL))) {
2615                         mtx_unlock(&softc->ctl_lock);
2616                         free(entries, M_CTL);
2617                         printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2618                                __func__, (uintmax_t)ooa_hdr->lun_num);
2619                         retval = EINVAL;
2620                         break;
2621                 }
2622
2623                 cur_fill_num = 0;
2624
2625                 if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2626                         STAILQ_FOREACH(lun, &softc->lun_list, links) {
2627                                 retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2628                                         ooa_hdr, entries);
2629                                 if (retval != 0)
2630                                         break;
2631                         }
2632                         if (retval != 0) {
2633                                 mtx_unlock(&softc->ctl_lock);
2634                                 free(entries, M_CTL);
2635                                 break;
2636                         }
2637                 } else {
2638                         lun = softc->ctl_luns[ooa_hdr->lun_num];
2639
2640                         retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,ooa_hdr,
2641                                                     entries);
2642                 }
2643                 mtx_unlock(&softc->ctl_lock);
2644
2645                 ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2646                 ooa_hdr->fill_len = ooa_hdr->fill_num *
2647                         sizeof(struct ctl_ooa_entry);
2648                 retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2649                 if (retval != 0) {
2650                         printf("%s: error copying out %d bytes for OOA dump\n", 
2651                                __func__, ooa_hdr->fill_len);
2652                 }
2653
2654                 getbintime(&ooa_hdr->cur_bt);
2655
2656                 if (cur_fill_num > ooa_hdr->alloc_num) {
2657                         ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2658                         ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2659                 } else {
2660                         ooa_hdr->dropped_num = 0;
2661                         ooa_hdr->status = CTL_OOA_OK;
2662                 }
2663
2664                 free(entries, M_CTL);
2665                 break;
2666         }
2667         case CTL_CHECK_OOA: {
2668                 union ctl_io *io;
2669                 struct ctl_ooa_info *ooa_info;
2670
2671
2672                 ooa_info = (struct ctl_ooa_info *)addr;
2673
2674                 if (ooa_info->lun_id >= CTL_MAX_LUNS) {
2675                         ooa_info->status = CTL_OOA_INVALID_LUN;
2676                         break;
2677                 }
2678                 mtx_lock(&softc->ctl_lock);
2679                 lun = softc->ctl_luns[ooa_info->lun_id];
2680                 if (lun == NULL) {
2681                         mtx_unlock(&softc->ctl_lock);
2682                         ooa_info->status = CTL_OOA_INVALID_LUN;
2683                         break;
2684                 }
2685                 mtx_lock(&lun->lun_lock);
2686                 mtx_unlock(&softc->ctl_lock);
2687                 ooa_info->num_entries = 0;
2688                 for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
2689                      io != NULL; io = (union ctl_io *)TAILQ_NEXT(
2690                      &io->io_hdr, ooa_links)) {
2691                         ooa_info->num_entries++;
2692                 }
2693                 mtx_unlock(&lun->lun_lock);
2694
2695                 ooa_info->status = CTL_OOA_SUCCESS;
2696
2697                 break;
2698         }
2699         case CTL_DELAY_IO: {
2700                 struct ctl_io_delay_info *delay_info;
2701
2702                 delay_info = (struct ctl_io_delay_info *)addr;
2703
2704 #ifdef CTL_IO_DELAY
2705                 mtx_lock(&softc->ctl_lock);
2706
2707                 if ((delay_info->lun_id >= CTL_MAX_LUNS)
2708                  || (softc->ctl_luns[delay_info->lun_id] == NULL)) {
2709                         delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2710                 } else {
2711                         lun = softc->ctl_luns[delay_info->lun_id];
2712                         mtx_lock(&lun->lun_lock);
2713
2714                         delay_info->status = CTL_DELAY_STATUS_OK;
2715
2716                         switch (delay_info->delay_type) {
2717                         case CTL_DELAY_TYPE_CONT:
2718                                 break;
2719                         case CTL_DELAY_TYPE_ONESHOT:
2720                                 break;
2721                         default:
2722                                 delay_info->status =
2723                                         CTL_DELAY_STATUS_INVALID_TYPE;
2724                                 break;
2725                         }
2726
2727                         switch (delay_info->delay_loc) {
2728                         case CTL_DELAY_LOC_DATAMOVE:
2729                                 lun->delay_info.datamove_type =
2730                                         delay_info->delay_type;
2731                                 lun->delay_info.datamove_delay =
2732                                         delay_info->delay_secs;
2733                                 break;
2734                         case CTL_DELAY_LOC_DONE:
2735                                 lun->delay_info.done_type =
2736                                         delay_info->delay_type;
2737                                 lun->delay_info.done_delay =
2738                                         delay_info->delay_secs;
2739                                 break;
2740                         default:
2741                                 delay_info->status =
2742                                         CTL_DELAY_STATUS_INVALID_LOC;
2743                                 break;
2744                         }
2745                         mtx_unlock(&lun->lun_lock);
2746                 }
2747
2748                 mtx_unlock(&softc->ctl_lock);
2749 #else
2750                 delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2751 #endif /* CTL_IO_DELAY */
2752                 break;
2753         }
2754         case CTL_REALSYNC_SET: {
2755                 int *syncstate;
2756
2757                 syncstate = (int *)addr;
2758
2759                 mtx_lock(&softc->ctl_lock);
2760                 switch (*syncstate) {
2761                 case 0:
2762                         softc->flags &= ~CTL_FLAG_REAL_SYNC;
2763                         break;
2764                 case 1:
2765                         softc->flags |= CTL_FLAG_REAL_SYNC;
2766                         break;
2767                 default:
2768                         retval = EINVAL;
2769                         break;
2770                 }
2771                 mtx_unlock(&softc->ctl_lock);
2772                 break;
2773         }
2774         case CTL_REALSYNC_GET: {
2775                 int *syncstate;
2776
2777                 syncstate = (int*)addr;
2778
2779                 mtx_lock(&softc->ctl_lock);
2780                 if (softc->flags & CTL_FLAG_REAL_SYNC)
2781                         *syncstate = 1;
2782                 else
2783                         *syncstate = 0;
2784                 mtx_unlock(&softc->ctl_lock);
2785
2786                 break;
2787         }
2788         case CTL_SETSYNC:
2789         case CTL_GETSYNC: {
2790                 struct ctl_sync_info *sync_info;
2791
2792                 sync_info = (struct ctl_sync_info *)addr;
2793
2794                 mtx_lock(&softc->ctl_lock);
2795                 lun = softc->ctl_luns[sync_info->lun_id];
2796                 if (lun == NULL) {
2797                         mtx_unlock(&softc->ctl_lock);
2798                         sync_info->status = CTL_GS_SYNC_NO_LUN;
2799                         break;
2800                 }
2801                 /*
2802                  * Get or set the sync interval.  We're not bounds checking
2803                  * in the set case, hopefully the user won't do something
2804                  * silly.
2805                  */
2806                 mtx_lock(&lun->lun_lock);
2807                 mtx_unlock(&softc->ctl_lock);
2808                 if (cmd == CTL_GETSYNC)
2809                         sync_info->sync_interval = lun->sync_interval;
2810                 else
2811                         lun->sync_interval = sync_info->sync_interval;
2812                 mtx_unlock(&lun->lun_lock);
2813
2814                 sync_info->status = CTL_GS_SYNC_OK;
2815
2816                 break;
2817         }
2818         case CTL_GETSTATS: {
2819                 struct ctl_stats *stats;
2820                 int i;
2821
2822                 stats = (struct ctl_stats *)addr;
2823
2824                 if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) >
2825                      stats->alloc_len) {
2826                         stats->status = CTL_SS_NEED_MORE_SPACE;
2827                         stats->num_luns = softc->num_luns;
2828                         break;
2829                 }
2830                 /*
2831                  * XXX KDM no locking here.  If the LUN list changes,
2832                  * things can blow up.
2833                  */
2834                 for (i = 0, lun = STAILQ_FIRST(&softc->lun_list); lun != NULL;
2835                      i++, lun = STAILQ_NEXT(lun, links)) {
2836                         retval = copyout(&lun->stats, &stats->lun_stats[i],
2837                                          sizeof(lun->stats));
2838                         if (retval != 0)
2839                                 break;
2840                 }
2841                 stats->num_luns = softc->num_luns;
2842                 stats->fill_len = sizeof(struct ctl_lun_io_stats) *
2843                                  softc->num_luns;
2844                 stats->status = CTL_SS_OK;
2845 #ifdef CTL_TIME_IO
2846                 stats->flags = CTL_STATS_FLAG_TIME_VALID;
2847 #else
2848                 stats->flags = CTL_STATS_FLAG_NONE;
2849 #endif
2850                 getnanouptime(&stats->timestamp);
2851                 break;
2852         }
2853         case CTL_ERROR_INJECT: {
2854                 struct ctl_error_desc *err_desc, *new_err_desc;
2855
2856                 err_desc = (struct ctl_error_desc *)addr;
2857
2858                 new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2859                                       M_WAITOK | M_ZERO);
2860                 bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2861
2862                 mtx_lock(&softc->ctl_lock);
2863                 lun = softc->ctl_luns[err_desc->lun_id];
2864                 if (lun == NULL) {
2865                         mtx_unlock(&softc->ctl_lock);
2866                         free(new_err_desc, M_CTL);
2867                         printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2868                                __func__, (uintmax_t)err_desc->lun_id);
2869                         retval = EINVAL;
2870                         break;
2871                 }
2872                 mtx_lock(&lun->lun_lock);
2873                 mtx_unlock(&softc->ctl_lock);
2874
2875                 /*
2876                  * We could do some checking here to verify the validity
2877                  * of the request, but given the complexity of error
2878                  * injection requests, the checking logic would be fairly
2879                  * complex.
2880                  *
2881                  * For now, if the request is invalid, it just won't get
2882                  * executed and might get deleted.
2883                  */
2884                 STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2885
2886                 /*
2887                  * XXX KDM check to make sure the serial number is unique,
2888                  * in case we somehow manage to wrap.  That shouldn't
2889                  * happen for a very long time, but it's the right thing to
2890                  * do.
2891                  */
2892                 new_err_desc->serial = lun->error_serial;
2893                 err_desc->serial = lun->error_serial;
2894                 lun->error_serial++;
2895
2896                 mtx_unlock(&lun->lun_lock);
2897                 break;
2898         }
2899         case CTL_ERROR_INJECT_DELETE: {
2900                 struct ctl_error_desc *delete_desc, *desc, *desc2;
2901                 int delete_done;
2902
2903                 delete_desc = (struct ctl_error_desc *)addr;
2904                 delete_done = 0;
2905
2906                 mtx_lock(&softc->ctl_lock);
2907                 lun = softc->ctl_luns[delete_desc->lun_id];
2908                 if (lun == NULL) {
2909                         mtx_unlock(&softc->ctl_lock);
2910                         printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2911                                __func__, (uintmax_t)delete_desc->lun_id);
2912                         retval = EINVAL;
2913                         break;
2914                 }
2915                 mtx_lock(&lun->lun_lock);
2916                 mtx_unlock(&softc->ctl_lock);
2917                 STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2918                         if (desc->serial != delete_desc->serial)
2919                                 continue;
2920
2921                         STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2922                                       links);
2923                         free(desc, M_CTL);
2924                         delete_done = 1;
2925                 }
2926                 mtx_unlock(&lun->lun_lock);
2927                 if (delete_done == 0) {
2928                         printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2929                                "error serial %ju on LUN %u\n", __func__, 
2930                                delete_desc->serial, delete_desc->lun_id);
2931                         retval = EINVAL;
2932                         break;
2933                 }
2934                 break;
2935         }
2936         case CTL_DUMP_STRUCTS: {
2937                 int i, j, k;
2938                 struct ctl_port *port;
2939                 struct ctl_frontend *fe;
2940
2941                 mtx_lock(&softc->ctl_lock);
2942                 printf("CTL Persistent Reservation information start:\n");
2943                 for (i = 0; i < CTL_MAX_LUNS; i++) {
2944                         lun = softc->ctl_luns[i];
2945
2946                         if ((lun == NULL)
2947                          || ((lun->flags & CTL_LUN_DISABLED) != 0))
2948                                 continue;
2949
2950                         for (j = 0; j < CTL_MAX_PORTS; j++) {
2951                                 if (lun->pr_keys[j] == NULL)
2952                                         continue;
2953                                 for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2954                                         if (lun->pr_keys[j][k] == 0)
2955                                                 continue;
2956                                         printf("  LUN %d port %d iid %d key "
2957                                                "%#jx\n", i, j, k,
2958                                                (uintmax_t)lun->pr_keys[j][k]);
2959                                 }
2960                         }
2961                 }
2962                 printf("CTL Persistent Reservation information end\n");
2963                 printf("CTL Ports:\n");
2964                 STAILQ_FOREACH(port, &softc->port_list, links) {
2965                         printf("  Port %d '%s' Frontend '%s' Type %u pp %d vp %d WWNN "
2966                                "%#jx WWPN %#jx\n", port->targ_port, port->port_name,
2967                                port->frontend->name, port->port_type,
2968                                port->physical_port, port->virtual_port,
2969                                (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
2970                         for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
2971                                 if (port->wwpn_iid[j].in_use == 0 &&
2972                                     port->wwpn_iid[j].wwpn == 0 &&
2973                                     port->wwpn_iid[j].name == NULL)
2974                                         continue;
2975
2976                                 printf("    iid %u use %d WWPN %#jx '%s'\n",
2977                                     j, port->wwpn_iid[j].in_use,
2978                                     (uintmax_t)port->wwpn_iid[j].wwpn,
2979                                     port->wwpn_iid[j].name);
2980                         }
2981                 }
2982                 printf("CTL Port information end\n");
2983                 mtx_unlock(&softc->ctl_lock);
2984                 /*
2985                  * XXX KDM calling this without a lock.  We'd likely want
2986                  * to drop the lock before calling the frontend's dump
2987                  * routine anyway.
2988                  */
2989                 printf("CTL Frontends:\n");
2990                 STAILQ_FOREACH(fe, &softc->fe_list, links) {
2991                         printf("  Frontend '%s'\n", fe->name);
2992                         if (fe->fe_dump != NULL)
2993                                 fe->fe_dump();
2994                 }
2995                 printf("CTL Frontend information end\n");
2996                 break;
2997         }
2998         case CTL_LUN_REQ: {
2999                 struct ctl_lun_req *lun_req;
3000                 struct ctl_backend_driver *backend;
3001
3002                 lun_req = (struct ctl_lun_req *)addr;
3003
3004                 backend = ctl_backend_find(lun_req->backend);
3005                 if (backend == NULL) {
3006                         lun_req->status = CTL_LUN_ERROR;
3007                         snprintf(lun_req->error_str,
3008                                  sizeof(lun_req->error_str),
3009                                  "Backend \"%s\" not found.",
3010                                  lun_req->backend);
3011                         break;
3012                 }
3013                 if (lun_req->num_be_args > 0) {
3014                         lun_req->kern_be_args = ctl_copyin_args(
3015                                 lun_req->num_be_args,
3016                                 lun_req->be_args,
3017                                 lun_req->error_str,
3018                                 sizeof(lun_req->error_str));
3019                         if (lun_req->kern_be_args == NULL) {
3020                                 lun_req->status = CTL_LUN_ERROR;
3021                                 break;
3022                         }
3023                 }
3024
3025                 retval = backend->ioctl(dev, cmd, addr, flag, td);
3026
3027                 if (lun_req->num_be_args > 0) {
3028                         ctl_copyout_args(lun_req->num_be_args,
3029                                       lun_req->kern_be_args);
3030                         ctl_free_args(lun_req->num_be_args,
3031                                       lun_req->kern_be_args);
3032                 }
3033                 break;
3034         }
3035         case CTL_LUN_LIST: {
3036                 struct sbuf *sb;
3037                 struct ctl_lun_list *list;
3038                 struct ctl_option *opt;
3039
3040                 list = (struct ctl_lun_list *)addr;
3041
3042                 /*
3043                  * Allocate a fixed length sbuf here, based on the length
3044                  * of the user's buffer.  We could allocate an auto-extending
3045                  * buffer, and then tell the user how much larger our
3046                  * amount of data is than his buffer, but that presents
3047                  * some problems:
3048                  *
3049                  * 1.  The sbuf(9) routines use a blocking malloc, and so
3050                  *     we can't hold a lock while calling them with an
3051                  *     auto-extending buffer.
3052                  *
3053                  * 2.  There is not currently a LUN reference counting
3054                  *     mechanism, outside of outstanding transactions on
3055                  *     the LUN's OOA queue.  So a LUN could go away on us
3056                  *     while we're getting the LUN number, backend-specific
3057                  *     information, etc.  Thus, given the way things
3058                  *     currently work, we need to hold the CTL lock while
3059                  *     grabbing LUN information.
3060                  *
3061                  * So, from the user's standpoint, the best thing to do is
3062                  * allocate what he thinks is a reasonable buffer length,
3063                  * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
3064                  * double the buffer length and try again.  (And repeat
3065                  * that until he succeeds.)
3066                  */
3067                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3068                 if (sb == NULL) {
3069                         list->status = CTL_LUN_LIST_ERROR;
3070                         snprintf(list->error_str, sizeof(list->error_str),
3071                                  "Unable to allocate %d bytes for LUN list",
3072                                  list->alloc_len);
3073                         break;
3074                 }
3075
3076                 sbuf_printf(sb, "<ctllunlist>\n");
3077
3078                 mtx_lock(&softc->ctl_lock);
3079                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
3080                         mtx_lock(&lun->lun_lock);
3081                         retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3082                                              (uintmax_t)lun->lun);
3083
3084                         /*
3085                          * Bail out as soon as we see that we've overfilled
3086                          * the buffer.
3087                          */
3088                         if (retval != 0)
3089                                 break;
3090
3091                         retval = sbuf_printf(sb, "\t<backend_type>%s"
3092                                              "</backend_type>\n",
3093                                              (lun->backend == NULL) ?  "none" :
3094                                              lun->backend->name);
3095
3096                         if (retval != 0)
3097                                 break;
3098
3099                         retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
3100                                              lun->be_lun->lun_type);
3101
3102                         if (retval != 0)
3103                                 break;
3104
3105                         if (lun->backend == NULL) {
3106                                 retval = sbuf_printf(sb, "</lun>\n");
3107                                 if (retval != 0)
3108                                         break;
3109                                 continue;
3110                         }
3111
3112                         retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
3113                                              (lun->be_lun->maxlba > 0) ?
3114                                              lun->be_lun->maxlba + 1 : 0);
3115
3116                         if (retval != 0)
3117                                 break;
3118
3119                         retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
3120                                              lun->be_lun->blocksize);
3121
3122                         if (retval != 0)
3123                                 break;
3124
3125                         retval = sbuf_printf(sb, "\t<serial_number>");
3126
3127                         if (retval != 0)
3128                                 break;
3129
3130                         retval = ctl_sbuf_printf_esc(sb,
3131                             lun->be_lun->serial_num,
3132                             sizeof(lun->be_lun->serial_num));
3133
3134                         if (retval != 0)
3135                                 break;
3136
3137                         retval = sbuf_printf(sb, "</serial_number>\n");
3138                 
3139                         if (retval != 0)
3140                                 break;
3141
3142                         retval = sbuf_printf(sb, "\t<device_id>");
3143
3144                         if (retval != 0)
3145                                 break;
3146
3147                         retval = ctl_sbuf_printf_esc(sb,
3148                             lun->be_lun->device_id,
3149                             sizeof(lun->be_lun->device_id));
3150
3151                         if (retval != 0)
3152                                 break;
3153
3154                         retval = sbuf_printf(sb, "</device_id>\n");
3155
3156                         if (retval != 0)
3157                                 break;
3158
3159                         if (lun->backend->lun_info != NULL) {
3160                                 retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3161                                 if (retval != 0)
3162                                         break;
3163                         }
3164                         STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
3165                                 retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3166                                     opt->name, opt->value, opt->name);
3167                                 if (retval != 0)
3168                                         break;
3169                         }
3170
3171                         retval = sbuf_printf(sb, "</lun>\n");
3172
3173                         if (retval != 0)
3174                                 break;
3175                         mtx_unlock(&lun->lun_lock);
3176                 }
3177                 if (lun != NULL)
3178                         mtx_unlock(&lun->lun_lock);
3179                 mtx_unlock(&softc->ctl_lock);
3180
3181                 if ((retval != 0)
3182                  || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3183                         retval = 0;
3184                         sbuf_delete(sb);
3185                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3186                         snprintf(list->error_str, sizeof(list->error_str),
3187                                  "Out of space, %d bytes is too small",
3188                                  list->alloc_len);
3189                         break;
3190                 }
3191
3192                 sbuf_finish(sb);
3193
3194                 retval = copyout(sbuf_data(sb), list->lun_xml,
3195                                  sbuf_len(sb) + 1);
3196
3197                 list->fill_len = sbuf_len(sb) + 1;
3198                 list->status = CTL_LUN_LIST_OK;
3199                 sbuf_delete(sb);
3200                 break;
3201         }
3202         case CTL_ISCSI: {
3203                 struct ctl_iscsi *ci;
3204                 struct ctl_frontend *fe;
3205
3206                 ci = (struct ctl_iscsi *)addr;
3207
3208                 fe = ctl_frontend_find("iscsi");
3209                 if (fe == NULL) {
3210                         ci->status = CTL_ISCSI_ERROR;
3211                         snprintf(ci->error_str, sizeof(ci->error_str),
3212                             "Frontend \"iscsi\" not found.");
3213                         break;
3214                 }
3215
3216                 retval = fe->ioctl(dev, cmd, addr, flag, td);
3217                 break;
3218         }
3219         case CTL_PORT_REQ: {
3220                 struct ctl_req *req;
3221                 struct ctl_frontend *fe;
3222
3223                 req = (struct ctl_req *)addr;
3224
3225                 fe = ctl_frontend_find(req->driver);
3226                 if (fe == NULL) {
3227                         req->status = CTL_LUN_ERROR;
3228                         snprintf(req->error_str, sizeof(req->error_str),
3229                             "Frontend \"%s\" not found.", req->driver);
3230                         break;
3231                 }
3232                 if (req->num_args > 0) {
3233                         req->kern_args = ctl_copyin_args(req->num_args,
3234                             req->args, req->error_str, sizeof(req->error_str));
3235                         if (req->kern_args == NULL) {
3236                                 req->status = CTL_LUN_ERROR;
3237                                 break;
3238                         }
3239                 }
3240
3241                 if (fe->ioctl)
3242                         retval = fe->ioctl(dev, cmd, addr, flag, td);
3243                 else
3244                         retval = ENODEV;
3245
3246                 if (req->num_args > 0) {
3247                         ctl_copyout_args(req->num_args, req->kern_args);
3248                         ctl_free_args(req->num_args, req->kern_args);
3249                 }
3250                 break;
3251         }
3252         case CTL_PORT_LIST: {
3253                 struct sbuf *sb;
3254                 struct ctl_port *port;
3255                 struct ctl_lun_list *list;
3256                 struct ctl_option *opt;
3257                 int j;
3258                 uint32_t plun;
3259
3260                 list = (struct ctl_lun_list *)addr;
3261
3262                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3263                 if (sb == NULL) {
3264                         list->status = CTL_LUN_LIST_ERROR;
3265                         snprintf(list->error_str, sizeof(list->error_str),
3266                                  "Unable to allocate %d bytes for LUN list",
3267                                  list->alloc_len);
3268                         break;
3269                 }
3270
3271                 sbuf_printf(sb, "<ctlportlist>\n");
3272
3273                 mtx_lock(&softc->ctl_lock);
3274                 STAILQ_FOREACH(port, &softc->port_list, links) {
3275                         retval = sbuf_printf(sb, "<targ_port id=\"%ju\">\n",
3276                                              (uintmax_t)port->targ_port);
3277
3278                         /*
3279                          * Bail out as soon as we see that we've overfilled
3280                          * the buffer.
3281                          */
3282                         if (retval != 0)
3283                                 break;
3284
3285                         retval = sbuf_printf(sb, "\t<frontend_type>%s"
3286                             "</frontend_type>\n", port->frontend->name);
3287                         if (retval != 0)
3288                                 break;
3289
3290                         retval = sbuf_printf(sb, "\t<port_type>%d</port_type>\n",
3291                                              port->port_type);
3292                         if (retval != 0)
3293                                 break;
3294
3295                         retval = sbuf_printf(sb, "\t<online>%s</online>\n",
3296                             (port->status & CTL_PORT_STATUS_ONLINE) ? "YES" : "NO");
3297                         if (retval != 0)
3298                                 break;
3299
3300                         retval = sbuf_printf(sb, "\t<port_name>%s</port_name>\n",
3301                             port->port_name);
3302                         if (retval != 0)
3303                                 break;
3304
3305                         retval = sbuf_printf(sb, "\t<physical_port>%d</physical_port>\n",
3306                             port->physical_port);
3307                         if (retval != 0)
3308                                 break;
3309
3310                         retval = sbuf_printf(sb, "\t<virtual_port>%d</virtual_port>\n",
3311                             port->virtual_port);
3312                         if (retval != 0)
3313                                 break;
3314
3315                         if (port->target_devid != NULL) {
3316                                 sbuf_printf(sb, "\t<target>");
3317                                 ctl_id_sbuf(port->target_devid, sb);
3318                                 sbuf_printf(sb, "</target>\n");
3319                         }
3320
3321                         if (port->port_devid != NULL) {
3322                                 sbuf_printf(sb, "\t<port>");
3323                                 ctl_id_sbuf(port->port_devid, sb);
3324                                 sbuf_printf(sb, "</port>\n");
3325                         }
3326
3327                         if (port->port_info != NULL) {
3328                                 retval = port->port_info(port->onoff_arg, sb);
3329                                 if (retval != 0)
3330                                         break;
3331                         }
3332                         STAILQ_FOREACH(opt, &port->options, links) {
3333                                 retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3334                                     opt->name, opt->value, opt->name);
3335                                 if (retval != 0)
3336                                         break;
3337                         }
3338
3339                         if (port->lun_map != NULL) {
3340                                 sbuf_printf(sb, "\t<lun_map>on</lun_map>\n");
3341                                 for (j = 0; j < CTL_MAX_LUNS; j++) {
3342                                         plun = ctl_lun_map_from_port(port, j);
3343                                         if (plun >= CTL_MAX_LUNS)
3344                                                 continue;
3345                                         sbuf_printf(sb,
3346                                             "\t<lun id=\"%u\">%u</lun>\n",
3347                                             j, plun);
3348                                 }
3349                         }
3350
3351                         for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3352                                 if (port->wwpn_iid[j].in_use == 0 ||
3353                                     (port->wwpn_iid[j].wwpn == 0 &&
3354                                      port->wwpn_iid[j].name == NULL))
3355                                         continue;
3356
3357                                 if (port->wwpn_iid[j].name != NULL)
3358                                         retval = sbuf_printf(sb,
3359                                             "\t<initiator id=\"%u\">%s</initiator>\n",
3360                                             j, port->wwpn_iid[j].name);
3361                                 else
3362                                         retval = sbuf_printf(sb,
3363                                             "\t<initiator id=\"%u\">naa.%08jx</initiator>\n",
3364                                             j, port->wwpn_iid[j].wwpn);
3365                                 if (retval != 0)
3366                                         break;
3367                         }
3368                         if (retval != 0)
3369                                 break;
3370
3371                         retval = sbuf_printf(sb, "</targ_port>\n");
3372                         if (retval != 0)
3373                                 break;
3374                 }
3375                 mtx_unlock(&softc->ctl_lock);
3376
3377                 if ((retval != 0)
3378                  || ((retval = sbuf_printf(sb, "</ctlportlist>\n")) != 0)) {
3379                         retval = 0;
3380                         sbuf_delete(sb);
3381                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3382                         snprintf(list->error_str, sizeof(list->error_str),
3383                                  "Out of space, %d bytes is too small",
3384                                  list->alloc_len);
3385                         break;
3386                 }
3387
3388                 sbuf_finish(sb);
3389
3390                 retval = copyout(sbuf_data(sb), list->lun_xml,
3391                                  sbuf_len(sb) + 1);
3392
3393                 list->fill_len = sbuf_len(sb) + 1;
3394                 list->status = CTL_LUN_LIST_OK;
3395                 sbuf_delete(sb);
3396                 break;
3397         }
3398         case CTL_LUN_MAP: {
3399                 struct ctl_lun_map *lm  = (struct ctl_lun_map *)addr;
3400                 struct ctl_port *port;
3401
3402                 mtx_lock(&softc->ctl_lock);
3403                 if (lm->port < softc->port_min ||
3404                     lm->port >= softc->port_max ||
3405                     (port = softc->ctl_ports[lm->port]) == NULL) {
3406                         mtx_unlock(&softc->ctl_lock);
3407                         return (ENXIO);
3408                 }
3409                 if (port->status & CTL_PORT_STATUS_ONLINE) {
3410                         STAILQ_FOREACH(lun, &softc->lun_list, links) {
3411                                 if (ctl_lun_map_to_port(port, lun->lun) >=
3412                                     CTL_MAX_LUNS)
3413                                         continue;
3414                                 mtx_lock(&lun->lun_lock);
3415                                 ctl_est_ua_port(lun, lm->port, -1,
3416                                     CTL_UA_LUN_CHANGE);
3417                                 mtx_unlock(&lun->lun_lock);
3418                         }
3419                 }
3420                 mtx_unlock(&softc->ctl_lock); // XXX: port_enable sleeps
3421                 if (lm->plun < CTL_MAX_LUNS) {
3422                         if (lm->lun == UINT32_MAX)
3423                                 retval = ctl_lun_map_unset(port, lm->plun);
3424                         else if (lm->lun < CTL_MAX_LUNS &&
3425                             softc->ctl_luns[lm->lun] != NULL)
3426                                 retval = ctl_lun_map_set(port, lm->plun, lm->lun);
3427                         else
3428                                 return (ENXIO);
3429                 } else if (lm->plun == UINT32_MAX) {
3430                         if (lm->lun == UINT32_MAX)
3431                                 retval = ctl_lun_map_deinit(port);
3432                         else
3433                                 retval = ctl_lun_map_init(port);
3434                 } else
3435                         return (ENXIO);
3436                 if (port->status & CTL_PORT_STATUS_ONLINE)
3437                         ctl_isc_announce_port(port);
3438                 break;
3439         }
3440         default: {
3441                 /* XXX KDM should we fix this? */
3442 #if 0
3443                 struct ctl_backend_driver *backend;
3444                 unsigned int type;
3445                 int found;
3446
3447                 found = 0;
3448
3449                 /*
3450                  * We encode the backend type as the ioctl type for backend
3451                  * ioctls.  So parse it out here, and then search for a
3452                  * backend of this type.
3453                  */
3454                 type = _IOC_TYPE(cmd);
3455
3456                 STAILQ_FOREACH(backend, &softc->be_list, links) {
3457                         if (backend->type == type) {
3458                                 found = 1;
3459                                 break;
3460                         }
3461                 }
3462                 if (found == 0) {
3463                         printf("ctl: unknown ioctl command %#lx or backend "
3464                                "%d\n", cmd, type);
3465                         retval = EINVAL;
3466                         break;
3467                 }
3468                 retval = backend->ioctl(dev, cmd, addr, flag, td);
3469 #endif
3470                 retval = ENOTTY;
3471                 break;
3472         }
3473         }
3474         return (retval);
3475 }
3476
3477 uint32_t
3478 ctl_get_initindex(struct ctl_nexus *nexus)
3479 {
3480         return (nexus->initid + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3481 }
3482
3483 int
3484 ctl_lun_map_init(struct ctl_port *port)
3485 {
3486         struct ctl_softc *softc = control_softc;
3487         struct ctl_lun *lun;
3488         uint32_t i;
3489
3490         if (port->lun_map == NULL)
3491                 port->lun_map = malloc(sizeof(uint32_t) * CTL_MAX_LUNS,
3492                     M_CTL, M_NOWAIT);
3493         if (port->lun_map == NULL)
3494                 return (ENOMEM);
3495         for (i = 0; i < CTL_MAX_LUNS; i++)
3496                 port->lun_map[i] = UINT32_MAX;
3497         if (port->status & CTL_PORT_STATUS_ONLINE) {
3498                 if (port->lun_disable != NULL) {
3499                         STAILQ_FOREACH(lun, &softc->lun_list, links)
3500                                 port->lun_disable(port->targ_lun_arg, lun->lun);
3501                 }
3502                 ctl_isc_announce_port(port);
3503         }
3504         return (0);
3505 }
3506
3507 int
3508 ctl_lun_map_deinit(struct ctl_port *port)
3509 {
3510         struct ctl_softc *softc = control_softc;
3511         struct ctl_lun *lun;
3512
3513         if (port->lun_map == NULL)
3514                 return (0);
3515         free(port->lun_map, M_CTL);
3516         port->lun_map = NULL;
3517         if (port->status & CTL_PORT_STATUS_ONLINE) {
3518                 if (port->lun_enable != NULL) {
3519                         STAILQ_FOREACH(lun, &softc->lun_list, links)
3520                                 port->lun_enable(port->targ_lun_arg, lun->lun);
3521                 }
3522                 ctl_isc_announce_port(port);
3523         }
3524         return (0);
3525 }
3526
3527 int
3528 ctl_lun_map_set(struct ctl_port *port, uint32_t plun, uint32_t glun)
3529 {
3530         int status;
3531         uint32_t old;
3532
3533         if (port->lun_map == NULL) {
3534                 status = ctl_lun_map_init(port);
3535                 if (status != 0)
3536                         return (status);
3537         }
3538         old = port->lun_map[plun];
3539         port->lun_map[plun] = glun;
3540         if ((port->status & CTL_PORT_STATUS_ONLINE) && old >= CTL_MAX_LUNS) {
3541                 if (port->lun_enable != NULL)
3542                         port->lun_enable(port->targ_lun_arg, plun);
3543                 ctl_isc_announce_port(port);
3544         }
3545         return (0);
3546 }
3547
3548 int
3549 ctl_lun_map_unset(struct ctl_port *port, uint32_t plun)
3550 {
3551         uint32_t old;
3552
3553         if (port->lun_map == NULL)
3554                 return (0);
3555         old = port->lun_map[plun];
3556         port->lun_map[plun] = UINT32_MAX;
3557         if ((port->status & CTL_PORT_STATUS_ONLINE) && old < CTL_MAX_LUNS) {
3558                 if (port->lun_disable != NULL)
3559                         port->lun_disable(port->targ_lun_arg, plun);
3560                 ctl_isc_announce_port(port);
3561         }
3562         return (0);
3563 }
3564
3565 uint32_t
3566 ctl_lun_map_from_port(struct ctl_port *port, uint32_t lun_id)
3567 {
3568
3569         if (port == NULL)
3570                 return (UINT32_MAX);
3571         if (port->lun_map == NULL || lun_id >= CTL_MAX_LUNS)
3572                 return (lun_id);
3573         return (port->lun_map[lun_id]);
3574 }
3575
3576 uint32_t
3577 ctl_lun_map_to_port(struct ctl_port *port, uint32_t lun_id)
3578 {
3579         uint32_t i;
3580
3581         if (port == NULL)
3582                 return (UINT32_MAX);
3583         if (port->lun_map == NULL)
3584                 return (lun_id);
3585         for (i = 0; i < CTL_MAX_LUNS; i++) {
3586                 if (port->lun_map[i] == lun_id)
3587                         return (i);
3588         }
3589         return (UINT32_MAX);
3590 }
3591
3592 static struct ctl_port *
3593 ctl_io_port(struct ctl_io_hdr *io_hdr)
3594 {
3595
3596         return (control_softc->ctl_ports[io_hdr->nexus.targ_port]);
3597 }
3598
3599 int
3600 ctl_ffz(uint32_t *mask, uint32_t first, uint32_t last)
3601 {
3602         int i;
3603
3604         for (i = first; i < last; i++) {
3605                 if ((mask[i / 32] & (1 << (i % 32))) == 0)
3606                         return (i);
3607         }
3608         return (-1);
3609 }
3610
3611 int
3612 ctl_set_mask(uint32_t *mask, uint32_t bit)
3613 {
3614         uint32_t chunk, piece;
3615
3616         chunk = bit >> 5;
3617         piece = bit % (sizeof(uint32_t) * 8);
3618
3619         if ((mask[chunk] & (1 << piece)) != 0)
3620                 return (-1);
3621         else
3622                 mask[chunk] |= (1 << piece);
3623
3624         return (0);
3625 }
3626
3627 int
3628 ctl_clear_mask(uint32_t *mask, uint32_t bit)
3629 {
3630         uint32_t chunk, piece;
3631
3632         chunk = bit >> 5;
3633         piece = bit % (sizeof(uint32_t) * 8);
3634
3635         if ((mask[chunk] & (1 << piece)) == 0)
3636                 return (-1);
3637         else
3638                 mask[chunk] &= ~(1 << piece);
3639
3640         return (0);
3641 }
3642
3643 int
3644 ctl_is_set(uint32_t *mask, uint32_t bit)
3645 {
3646         uint32_t chunk, piece;
3647
3648         chunk = bit >> 5;
3649         piece = bit % (sizeof(uint32_t) * 8);
3650
3651         if ((mask[chunk] & (1 << piece)) == 0)
3652                 return (0);
3653         else
3654                 return (1);
3655 }
3656
3657 static uint64_t
3658 ctl_get_prkey(struct ctl_lun *lun, uint32_t residx)
3659 {
3660         uint64_t *t;
3661
3662         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3663         if (t == NULL)
3664                 return (0);
3665         return (t[residx % CTL_MAX_INIT_PER_PORT]);
3666 }
3667
3668 static void
3669 ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx)
3670 {
3671         uint64_t *t;
3672
3673         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3674         if (t == NULL)
3675                 return;
3676         t[residx % CTL_MAX_INIT_PER_PORT] = 0;
3677 }
3678
3679 static void
3680 ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx)
3681 {
3682         uint64_t *p;
3683         u_int i;
3684
3685         i = residx/CTL_MAX_INIT_PER_PORT;
3686         if (lun->pr_keys[i] != NULL)
3687                 return;
3688         mtx_unlock(&lun->lun_lock);
3689         p = malloc(sizeof(uint64_t) * CTL_MAX_INIT_PER_PORT, M_CTL,
3690             M_WAITOK | M_ZERO);
3691         mtx_lock(&lun->lun_lock);
3692         if (lun->pr_keys[i] == NULL)
3693                 lun->pr_keys[i] = p;
3694         else
3695                 free(p, M_CTL);
3696 }
3697
3698 static void
3699 ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key)
3700 {
3701         uint64_t *t;
3702
3703         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3704         KASSERT(t != NULL, ("prkey %d is not allocated", residx));
3705         t[residx % CTL_MAX_INIT_PER_PORT] = key;
3706 }
3707
3708 /*
3709  * ctl_softc, pool_name, total_ctl_io are passed in.
3710  * npool is passed out.
3711  */
3712 int
3713 ctl_pool_create(struct ctl_softc *ctl_softc, const char *pool_name,
3714                 uint32_t total_ctl_io, void **npool)
3715 {
3716 #ifdef IO_POOLS
3717         struct ctl_io_pool *pool;
3718
3719         pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3720                                             M_NOWAIT | M_ZERO);
3721         if (pool == NULL)
3722                 return (ENOMEM);
3723
3724         snprintf(pool->name, sizeof(pool->name), "CTL IO %s", pool_name);
3725         pool->ctl_softc = ctl_softc;
3726         pool->zone = uma_zsecond_create(pool->name, NULL,
3727             NULL, NULL, NULL, ctl_softc->io_zone);
3728         /* uma_prealloc(pool->zone, total_ctl_io); */
3729
3730         *npool = pool;
3731 #else
3732         *npool = ctl_softc->io_zone;
3733 #endif
3734         return (0);
3735 }
3736
3737 void
3738 ctl_pool_free(struct ctl_io_pool *pool)
3739 {
3740
3741         if (pool == NULL)
3742                 return;
3743
3744 #ifdef IO_POOLS
3745         uma_zdestroy(pool->zone);
3746         free(pool, M_CTL);
3747 #endif
3748 }
3749
3750 union ctl_io *
3751 ctl_alloc_io(void *pool_ref)
3752 {
3753         union ctl_io *io;
3754 #ifdef IO_POOLS
3755         struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3756
3757         io = uma_zalloc(pool->zone, M_WAITOK);
3758 #else
3759         io = uma_zalloc((uma_zone_t)pool_ref, M_WAITOK);
3760 #endif
3761         if (io != NULL)
3762                 io->io_hdr.pool = pool_ref;
3763         return (io);
3764 }
3765
3766 union ctl_io *
3767 ctl_alloc_io_nowait(void *pool_ref)
3768 {
3769         union ctl_io *io;
3770 #ifdef IO_POOLS
3771         struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3772
3773         io = uma_zalloc(pool->zone, M_NOWAIT);
3774 #else
3775         io = uma_zalloc((uma_zone_t)pool_ref, M_NOWAIT);
3776 #endif
3777         if (io != NULL)
3778                 io->io_hdr.pool = pool_ref;
3779         return (io);
3780 }
3781
3782 void
3783 ctl_free_io(union ctl_io *io)
3784 {
3785 #ifdef IO_POOLS
3786         struct ctl_io_pool *pool;
3787 #endif
3788
3789         if (io == NULL)
3790                 return;
3791
3792 #ifdef IO_POOLS
3793         pool = (struct ctl_io_pool *)io->io_hdr.pool;
3794         uma_zfree(pool->zone, io);
3795 #else
3796         uma_zfree((uma_zone_t)io->io_hdr.pool, io);
3797 #endif
3798 }
3799
3800 void
3801 ctl_zero_io(union ctl_io *io)
3802 {
3803         void *pool_ref;
3804
3805         if (io == NULL)
3806                 return;
3807
3808         /*
3809          * May need to preserve linked list pointers at some point too.
3810          */
3811         pool_ref = io->io_hdr.pool;
3812         memset(io, 0, sizeof(*io));
3813         io->io_hdr.pool = pool_ref;
3814 }
3815
3816 /*
3817  * This routine is currently used for internal copies of ctl_ios that need
3818  * to persist for some reason after we've already returned status to the
3819  * FETD.  (Thus the flag set.)
3820  *
3821  * XXX XXX
3822  * Note that this makes a blind copy of all fields in the ctl_io, except
3823  * for the pool reference.  This includes any memory that has been
3824  * allocated!  That memory will no longer be valid after done has been
3825  * called, so this would be VERY DANGEROUS for command that actually does
3826  * any reads or writes.  Right now (11/7/2005), this is only used for immediate
3827  * start and stop commands, which don't transfer any data, so this is not a
3828  * problem.  If it is used for anything else, the caller would also need to
3829  * allocate data buffer space and this routine would need to be modified to
3830  * copy the data buffer(s) as well.
3831  */
3832 void
3833 ctl_copy_io(union ctl_io *src, union ctl_io *dest)
3834 {
3835         void *pool_ref;
3836
3837         if ((src == NULL)
3838          || (dest == NULL))
3839                 return;
3840
3841         /*
3842          * May need to preserve linked list pointers at some point too.
3843          */
3844         pool_ref = dest->io_hdr.pool;
3845
3846         memcpy(dest, src, MIN(sizeof(*src), sizeof(*dest)));
3847
3848         dest->io_hdr.pool = pool_ref;
3849         /*
3850          * We need to know that this is an internal copy, and doesn't need
3851          * to get passed back to the FETD that allocated it.
3852          */
3853         dest->io_hdr.flags |= CTL_FLAG_INT_COPY;
3854 }
3855
3856 int
3857 ctl_expand_number(const char *buf, uint64_t *num)
3858 {
3859         char *endptr;
3860         uint64_t number;
3861         unsigned shift;
3862
3863         number = strtoq(buf, &endptr, 0);
3864
3865         switch (tolower((unsigned char)*endptr)) {
3866         case 'e':
3867                 shift = 60;
3868                 break;
3869         case 'p':
3870                 shift = 50;
3871                 break;
3872         case 't':
3873                 shift = 40;
3874                 break;
3875         case 'g':
3876                 shift = 30;
3877                 break;
3878         case 'm':
3879                 shift = 20;
3880                 break;
3881         case 'k':
3882                 shift = 10;
3883                 break;
3884         case 'b':
3885         case '\0': /* No unit. */
3886                 *num = number;
3887                 return (0);
3888         default:
3889                 /* Unrecognized unit. */
3890                 return (-1);
3891         }
3892
3893         if ((number << shift) >> shift != number) {
3894                 /* Overflow */
3895                 return (-1);
3896         }
3897         *num = number << shift;
3898         return (0);
3899 }
3900
3901
3902 /*
3903  * This routine could be used in the future to load default and/or saved
3904  * mode page parameters for a particuar lun.
3905  */
3906 static int
3907 ctl_init_page_index(struct ctl_lun *lun)
3908 {
3909         int i;
3910         struct ctl_page_index *page_index;
3911         const char *value;
3912         uint64_t ival;
3913
3914         memcpy(&lun->mode_pages.index, page_index_template,
3915                sizeof(page_index_template));
3916
3917         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
3918
3919                 page_index = &lun->mode_pages.index[i];
3920                 /*
3921                  * If this is a disk-only mode page, there's no point in
3922                  * setting it up.  For some pages, we have to have some
3923                  * basic information about the disk in order to calculate the
3924                  * mode page data.
3925                  */
3926                 if ((lun->be_lun->lun_type != T_DIRECT)
3927                  && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
3928                         continue;
3929
3930                 switch (page_index->page_code & SMPH_PC_MASK) {
3931                 case SMS_RW_ERROR_RECOVERY_PAGE: {
3932                         if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3933                                 panic("subpage is incorrect!");
3934                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT],
3935                                &rw_er_page_default,
3936                                sizeof(rw_er_page_default));
3937                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CHANGEABLE],
3938                                &rw_er_page_changeable,
3939                                sizeof(rw_er_page_changeable));
3940                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_DEFAULT],
3941                                &rw_er_page_default,
3942                                sizeof(rw_er_page_default));
3943                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_SAVED],
3944                                &rw_er_page_default,
3945                                sizeof(rw_er_page_default));
3946                         page_index->page_data =
3947                                 (uint8_t *)lun->mode_pages.rw_er_page;
3948                         break;
3949                 }
3950                 case SMS_FORMAT_DEVICE_PAGE: {
3951                         struct scsi_format_page *format_page;
3952
3953                         if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3954                                 panic("subpage is incorrect!");
3955
3956                         /*
3957                          * Sectors per track are set above.  Bytes per
3958                          * sector need to be set here on a per-LUN basis.
3959                          */
3960                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
3961                                &format_page_default,
3962                                sizeof(format_page_default));
3963                         memcpy(&lun->mode_pages.format_page[
3964                                CTL_PAGE_CHANGEABLE], &format_page_changeable,
3965                                sizeof(format_page_changeable));
3966                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
3967                                &format_page_default,
3968                                sizeof(format_page_default));
3969                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
3970                                &format_page_default,
3971                                sizeof(format_page_default));
3972
3973                         format_page = &lun->mode_pages.format_page[
3974                                 CTL_PAGE_CURRENT];
3975                         scsi_ulto2b(lun->be_lun->blocksize,
3976                                     format_page->bytes_per_sector);
3977
3978                         format_page = &lun->mode_pages.format_page[
3979                                 CTL_PAGE_DEFAULT];
3980                         scsi_ulto2b(lun->be_lun->blocksize,
3981                                     format_page->bytes_per_sector);
3982
3983                         format_page = &lun->mode_pages.format_page[
3984                                 CTL_PAGE_SAVED];
3985                         scsi_ulto2b(lun->be_lun->blocksize,
3986                                     format_page->bytes_per_sector);
3987
3988                         page_index->page_data =
3989                                 (uint8_t *)lun->mode_pages.format_page;
3990                         break;
3991                 }
3992                 case SMS_RIGID_DISK_PAGE: {
3993                         struct scsi_rigid_disk_page *rigid_disk_page;
3994                         uint32_t sectors_per_cylinder;
3995                         uint64_t cylinders;
3996 #ifndef __XSCALE__
3997                         int shift;
3998 #endif /* !__XSCALE__ */
3999
4000                         if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4001                                 panic("invalid subpage value %d",
4002                                       page_index->subpage);
4003
4004                         /*
4005                          * Rotation rate and sectors per track are set
4006                          * above.  We calculate the cylinders here based on
4007                          * capacity.  Due to the number of heads and
4008                          * sectors per track we're using, smaller arrays
4009                          * may turn out to have 0 cylinders.  Linux and
4010                          * FreeBSD don't pay attention to these mode pages
4011                          * to figure out capacity, but Solaris does.  It
4012                          * seems to deal with 0 cylinders just fine, and
4013                          * works out a fake geometry based on the capacity.
4014                          */
4015                         memcpy(&lun->mode_pages.rigid_disk_page[
4016                                CTL_PAGE_DEFAULT], &rigid_disk_page_default,
4017                                sizeof(rigid_disk_page_default));
4018                         memcpy(&lun->mode_pages.rigid_disk_page[
4019                                CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
4020                                sizeof(rigid_disk_page_changeable));
4021
4022                         sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
4023                                 CTL_DEFAULT_HEADS;
4024
4025                         /*
4026                          * The divide method here will be more accurate,
4027                          * probably, but results in floating point being
4028                          * used in the kernel on i386 (__udivdi3()).  On the
4029                          * XScale, though, __udivdi3() is implemented in
4030                          * software.
4031                          *
4032                          * The shift method for cylinder calculation is
4033                          * accurate if sectors_per_cylinder is a power of
4034                          * 2.  Otherwise it might be slightly off -- you
4035                          * might have a bit of a truncation problem.
4036                          */
4037 #ifdef  __XSCALE__
4038                         cylinders = (lun->be_lun->maxlba + 1) /
4039                                 sectors_per_cylinder;
4040 #else
4041                         for (shift = 31; shift > 0; shift--) {
4042                                 if (sectors_per_cylinder & (1 << shift))
4043                                         break;
4044                         }
4045                         cylinders = (lun->be_lun->maxlba + 1) >> shift;
4046 #endif
4047
4048                         /*
4049                          * We've basically got 3 bytes, or 24 bits for the
4050                          * cylinder size in the mode page.  If we're over,
4051                          * just round down to 2^24.
4052                          */
4053                         if (cylinders > 0xffffff)
4054                                 cylinders = 0xffffff;
4055
4056                         rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4057                                 CTL_PAGE_DEFAULT];
4058                         scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4059
4060                         if ((value = ctl_get_opt(&lun->be_lun->options,
4061                             "rpm")) != NULL) {
4062                                 scsi_ulto2b(strtol(value, NULL, 0),
4063                                      rigid_disk_page->rotation_rate);
4064                         }
4065
4066                         memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_CURRENT],
4067                                &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
4068                                sizeof(rigid_disk_page_default));
4069                         memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_SAVED],
4070                                &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
4071                                sizeof(rigid_disk_page_default));
4072
4073                         page_index->page_data =
4074                                 (uint8_t *)lun->mode_pages.rigid_disk_page;
4075                         break;
4076                 }
4077                 case SMS_CACHING_PAGE: {
4078                         struct scsi_caching_page *caching_page;
4079
4080                         if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4081                                 panic("invalid subpage value %d",
4082                                       page_index->subpage);
4083                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4084                                &caching_page_default,
4085                                sizeof(caching_page_default));
4086                         memcpy(&lun->mode_pages.caching_page[
4087                                CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4088                                sizeof(caching_page_changeable));
4089                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4090                                &caching_page_default,
4091                                sizeof(caching_page_default));
4092                         caching_page = &lun->mode_pages.caching_page[
4093                             CTL_PAGE_SAVED];
4094                         value = ctl_get_opt(&lun->be_lun->options, "writecache");
4095                         if (value != NULL && strcmp(value, "off") == 0)
4096                                 caching_page->flags1 &= ~SCP_WCE;
4097                         value = ctl_get_opt(&lun->be_lun->options, "readcache");
4098                         if (value != NULL && strcmp(value, "off") == 0)
4099                                 caching_page->flags1 |= SCP_RCD;
4100                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4101                                &lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4102                                sizeof(caching_page_default));
4103                         page_index->page_data =
4104                                 (uint8_t *)lun->mode_pages.caching_page;
4105                         break;
4106                 }
4107                 case SMS_CONTROL_MODE_PAGE: {
4108                         switch (page_index->subpage) {
4109                         case SMS_SUBPAGE_PAGE_0: {
4110                                 struct scsi_control_page *control_page;
4111
4112                                 memcpy(&lun->mode_pages.control_page[
4113                                     CTL_PAGE_DEFAULT],
4114                                        &control_page_default,
4115                                        sizeof(control_page_default));
4116                                 memcpy(&lun->mode_pages.control_page[
4117                                     CTL_PAGE_CHANGEABLE],
4118                                        &control_page_changeable,
4119                                        sizeof(control_page_changeable));
4120                                 memcpy(&lun->mode_pages.control_page[
4121                                     CTL_PAGE_SAVED],
4122                                        &control_page_default,
4123                                        sizeof(control_page_default));
4124                                 control_page = &lun->mode_pages.control_page[
4125                                     CTL_PAGE_SAVED];
4126                                 value = ctl_get_opt(&lun->be_lun->options,
4127                                     "reordering");
4128                                 if (value != NULL &&
4129                                     strcmp(value, "unrestricted") == 0) {
4130                                         control_page->queue_flags &=
4131                                             ~SCP_QUEUE_ALG_MASK;
4132                                         control_page->queue_flags |=
4133                                             SCP_QUEUE_ALG_UNRESTRICTED;
4134                                 }
4135                                 memcpy(&lun->mode_pages.control_page[
4136                                     CTL_PAGE_CURRENT],
4137                                        &lun->mode_pages.control_page[
4138                                     CTL_PAGE_SAVED],
4139                                        sizeof(control_page_default));
4140                                 page_index->page_data =
4141                                     (uint8_t *)lun->mode_pages.control_page;
4142                                 break;
4143                         }
4144                         case 0x01:
4145                                 memcpy(&lun->mode_pages.control_ext_page[
4146                                     CTL_PAGE_DEFAULT],
4147                                        &control_ext_page_default,
4148                                        sizeof(control_ext_page_default));
4149                                 memcpy(&lun->mode_pages.control_ext_page[
4150                                     CTL_PAGE_CHANGEABLE],
4151                                        &control_ext_page_changeable,
4152                                        sizeof(control_ext_page_changeable));
4153                                 memcpy(&lun->mode_pages.control_ext_page[
4154                                     CTL_PAGE_SAVED],
4155                                        &control_ext_page_default,
4156                                        sizeof(control_ext_page_default));
4157                                 memcpy(&lun->mode_pages.control_ext_page[
4158                                     CTL_PAGE_CURRENT],
4159                                        &lun->mode_pages.control_ext_page[
4160                                     CTL_PAGE_SAVED],
4161                                        sizeof(control_ext_page_default));
4162                                 page_index->page_data =
4163                                     (uint8_t *)lun->mode_pages.control_ext_page;
4164                                 break;
4165                         }
4166                         break;
4167                 }
4168                 case SMS_INFO_EXCEPTIONS_PAGE: {
4169                         switch (page_index->subpage) {
4170                         case SMS_SUBPAGE_PAGE_0:
4171                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_CURRENT],
4172                                        &ie_page_default,
4173                                        sizeof(ie_page_default));
4174                                 memcpy(&lun->mode_pages.ie_page[
4175                                        CTL_PAGE_CHANGEABLE], &ie_page_changeable,
4176                                        sizeof(ie_page_changeable));
4177                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_DEFAULT],
4178                                        &ie_page_default,
4179                                        sizeof(ie_page_default));
4180                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_SAVED],
4181                                        &ie_page_default,
4182                                        sizeof(ie_page_default));
4183                                 page_index->page_data =
4184                                         (uint8_t *)lun->mode_pages.ie_page;
4185                                 break;
4186                         case 0x02: {
4187                                 struct ctl_logical_block_provisioning_page *page;
4188
4189                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_DEFAULT],
4190                                        &lbp_page_default,
4191                                        sizeof(lbp_page_default));
4192                                 memcpy(&lun->mode_pages.lbp_page[
4193                                        CTL_PAGE_CHANGEABLE], &lbp_page_changeable,
4194                                        sizeof(lbp_page_changeable));
4195                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4196                                        &lbp_page_default,
4197                                        sizeof(lbp_page_default));
4198                                 page = &lun->mode_pages.lbp_page[CTL_PAGE_SAVED];
4199                                 value = ctl_get_opt(&lun->be_lun->options,
4200                                     "avail-threshold");
4201                                 if (value != NULL &&
4202                                     ctl_expand_number(value, &ival) == 0) {
4203                                         page->descr[0].flags |= SLBPPD_ENABLED |
4204                                             SLBPPD_ARMING_DEC;
4205                                         if (lun->be_lun->blocksize)
4206                                                 ival /= lun->be_lun->blocksize;
4207                                         else
4208                                                 ival /= 512;
4209                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4210                                             page->descr[0].count);
4211                                 }
4212                                 value = ctl_get_opt(&lun->be_lun->options,
4213                                     "used-threshold");
4214                                 if (value != NULL &&
4215                                     ctl_expand_number(value, &ival) == 0) {
4216                                         page->descr[1].flags |= SLBPPD_ENABLED |
4217                                             SLBPPD_ARMING_INC;
4218                                         if (lun->be_lun->blocksize)
4219                                                 ival /= lun->be_lun->blocksize;
4220                                         else
4221                                                 ival /= 512;
4222                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4223                                             page->descr[1].count);
4224                                 }
4225                                 value = ctl_get_opt(&lun->be_lun->options,
4226                                     "pool-avail-threshold");
4227                                 if (value != NULL &&
4228                                     ctl_expand_number(value, &ival) == 0) {
4229                                         page->descr[2].flags |= SLBPPD_ENABLED |
4230                                             SLBPPD_ARMING_DEC;
4231                                         if (lun->be_lun->blocksize)
4232                                                 ival /= lun->be_lun->blocksize;
4233                                         else
4234                                                 ival /= 512;
4235                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4236                                             page->descr[2].count);
4237                                 }
4238                                 value = ctl_get_opt(&lun->be_lun->options,
4239                                     "pool-used-threshold");
4240                                 if (value != NULL &&
4241                                     ctl_expand_number(value, &ival) == 0) {
4242                                         page->descr[3].flags |= SLBPPD_ENABLED |
4243                                             SLBPPD_ARMING_INC;
4244                                         if (lun->be_lun->blocksize)
4245                                                 ival /= lun->be_lun->blocksize;
4246                                         else
4247                                                 ival /= 512;
4248                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4249                                             page->descr[3].count);
4250                                 }
4251                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_CURRENT],
4252                                        &lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4253                                        sizeof(lbp_page_default));
4254                                 page_index->page_data =
4255                                         (uint8_t *)lun->mode_pages.lbp_page;
4256                         }}
4257                         break;
4258                 }
4259                 case SMS_VENDOR_SPECIFIC_PAGE:{
4260                         switch (page_index->subpage) {
4261                         case DBGCNF_SUBPAGE_CODE: {
4262                                 struct copan_debugconf_subpage *current_page,
4263                                                                *saved_page;
4264
4265                                 memcpy(&lun->mode_pages.debugconf_subpage[
4266                                        CTL_PAGE_CURRENT],
4267                                        &debugconf_page_default,
4268                                        sizeof(debugconf_page_default));
4269                                 memcpy(&lun->mode_pages.debugconf_subpage[
4270                                        CTL_PAGE_CHANGEABLE],
4271                                        &debugconf_page_changeable,
4272                                        sizeof(debugconf_page_changeable));
4273                                 memcpy(&lun->mode_pages.debugconf_subpage[
4274                                        CTL_PAGE_DEFAULT],
4275                                        &debugconf_page_default,
4276                                        sizeof(debugconf_page_default));
4277                                 memcpy(&lun->mode_pages.debugconf_subpage[
4278                                        CTL_PAGE_SAVED],
4279                                        &debugconf_page_default,
4280                                        sizeof(debugconf_page_default));
4281                                 page_index->page_data =
4282                                         (uint8_t *)lun->mode_pages.debugconf_subpage;
4283
4284                                 current_page = (struct copan_debugconf_subpage *)
4285                                         (page_index->page_data +
4286                                          (page_index->page_len *
4287                                           CTL_PAGE_CURRENT));
4288                                 saved_page = (struct copan_debugconf_subpage *)
4289                                         (page_index->page_data +
4290                                          (page_index->page_len *
4291                                           CTL_PAGE_SAVED));
4292                                 break;
4293                         }
4294                         default:
4295                                 panic("invalid subpage value %d",
4296                                       page_index->subpage);
4297                                 break;
4298                         }
4299                         break;
4300                 }
4301                 default:
4302                         panic("invalid page value %d",
4303                               page_index->page_code & SMPH_PC_MASK);
4304                         break;
4305         }
4306         }
4307
4308         return (CTL_RETVAL_COMPLETE);
4309 }
4310
4311 static int
4312 ctl_init_log_page_index(struct ctl_lun *lun)
4313 {
4314         struct ctl_page_index *page_index;
4315         int i, j, k, prev;
4316
4317         memcpy(&lun->log_pages.index, log_page_index_template,
4318                sizeof(log_page_index_template));
4319
4320         prev = -1;
4321         for (i = 0, j = 0, k = 0; i < CTL_NUM_LOG_PAGES; i++) {
4322
4323                 page_index = &lun->log_pages.index[i];
4324                 /*
4325                  * If this is a disk-only mode page, there's no point in
4326                  * setting it up.  For some pages, we have to have some
4327                  * basic information about the disk in order to calculate the
4328                  * mode page data.
4329                  */
4330                 if ((lun->be_lun->lun_type != T_DIRECT)
4331                  && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
4332                         continue;
4333
4334                 if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING &&
4335                      lun->backend->lun_attr == NULL)
4336                         continue;
4337
4338                 if (page_index->page_code != prev) {
4339                         lun->log_pages.pages_page[j] = page_index->page_code;
4340                         prev = page_index->page_code;
4341                         j++;
4342                 }
4343                 lun->log_pages.subpages_page[k*2] = page_index->page_code;
4344                 lun->log_pages.subpages_page[k*2+1] = page_index->subpage;
4345                 k++;
4346         }
4347         lun->log_pages.index[0].page_data = &lun->log_pages.pages_page[0];
4348         lun->log_pages.index[0].page_len = j;
4349         lun->log_pages.index[1].page_data = &lun->log_pages.subpages_page[0];
4350         lun->log_pages.index[1].page_len = k * 2;
4351         lun->log_pages.index[2].page_data = &lun->log_pages.lbp_page[0];
4352         lun->log_pages.index[2].page_len = 12*CTL_NUM_LBP_PARAMS;
4353         lun->log_pages.index[3].page_data = (uint8_t *)&lun->log_pages.stat_page;
4354         lun->log_pages.index[3].page_len = sizeof(lun->log_pages.stat_page);
4355
4356         return (CTL_RETVAL_COMPLETE);
4357 }
4358
4359 static int
4360 hex2bin(const char *str, uint8_t *buf, int buf_size)
4361 {
4362         int i;
4363         u_char c;
4364
4365         memset(buf, 0, buf_size);
4366         while (isspace(str[0]))
4367                 str++;
4368         if (str[0] == '0' && (str[1] == 'x' || str[1] == 'X'))
4369                 str += 2;
4370         buf_size *= 2;
4371         for (i = 0; str[i] != 0 && i < buf_size; i++) {
4372                 c = str[i];
4373                 if (isdigit(c))
4374                         c -= '0';
4375                 else if (isalpha(c))
4376                         c -= isupper(c) ? 'A' - 10 : 'a' - 10;
4377                 else
4378                         break;
4379                 if (c >= 16)
4380                         break;
4381                 if ((i & 1) == 0)
4382                         buf[i / 2] |= (c << 4);
4383                 else
4384                         buf[i / 2] |= c;
4385         }
4386         return ((i + 1) / 2);
4387 }
4388
4389 /*
4390  * LUN allocation.
4391  *
4392  * Requirements:
4393  * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4394  *   wants us to allocate the LUN and he can block.
4395  * - ctl_softc is always set
4396  * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4397  *
4398  * Returns 0 for success, non-zero (errno) for failure.
4399  */
4400 static int
4401 ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4402               struct ctl_be_lun *const be_lun)
4403 {
4404         struct ctl_lun *nlun, *lun;
4405         struct scsi_vpd_id_descriptor *desc;
4406         struct scsi_vpd_id_t10 *t10id;
4407         const char *eui, *naa, *scsiname, *vendor;
4408         int lun_number, i, lun_malloced;
4409         int devidlen, idlen1, idlen2 = 0, len;
4410
4411         if (be_lun == NULL)
4412                 return (EINVAL);
4413
4414         /*
4415          * We currently only support Direct Access or Processor LUN types.
4416          */
4417         switch (be_lun->lun_type) {
4418         case T_DIRECT:
4419                 break;
4420         case T_PROCESSOR:
4421                 break;
4422         case T_SEQUENTIAL:
4423         case T_CHANGER:
4424         default:
4425                 be_lun->lun_config_status(be_lun->be_lun,
4426                                           CTL_LUN_CONFIG_FAILURE);
4427                 break;
4428         }
4429         if (ctl_lun == NULL) {
4430                 lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4431                 lun_malloced = 1;
4432         } else {
4433                 lun_malloced = 0;
4434                 lun = ctl_lun;
4435         }
4436
4437         memset(lun, 0, sizeof(*lun));
4438         if (lun_malloced)
4439                 lun->flags = CTL_LUN_MALLOCED;
4440
4441         /* Generate LUN ID. */
4442         devidlen = max(CTL_DEVID_MIN_LEN,
4443             strnlen(be_lun->device_id, CTL_DEVID_LEN));
4444         idlen1 = sizeof(*t10id) + devidlen;
4445         len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4446         scsiname = ctl_get_opt(&be_lun->options, "scsiname");
4447         if (scsiname != NULL) {
4448                 idlen2 = roundup2(strlen(scsiname) + 1, 4);
4449                 len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4450         }
4451         eui = ctl_get_opt(&be_lun->options, "eui");
4452         if (eui != NULL) {
4453                 len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4454         }
4455         naa = ctl_get_opt(&be_lun->options, "naa");
4456         if (naa != NULL) {
4457                 len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4458         }
4459         lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4460             M_CTL, M_WAITOK | M_ZERO);
4461         desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4462         desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4463         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4464         desc->length = idlen1;
4465         t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4466         memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4467         if ((vendor = ctl_get_opt(&be_lun->options, "vendor")) == NULL) {
4468                 strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4469         } else {
4470                 strncpy(t10id->vendor, vendor,
4471                     min(sizeof(t10id->vendor), strlen(vendor)));
4472         }
4473         strncpy((char *)t10id->vendor_spec_id,
4474             (char *)be_lun->device_id, devidlen);
4475         if (scsiname != NULL) {
4476                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4477                     desc->length);
4478                 desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4479                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4480                     SVPD_ID_TYPE_SCSI_NAME;
4481                 desc->length = idlen2;
4482                 strlcpy(desc->identifier, scsiname, idlen2);
4483         }
4484         if (eui != NULL) {
4485                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4486                     desc->length);
4487                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4488                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4489                     SVPD_ID_TYPE_EUI64;
4490                 desc->length = hex2bin(eui, desc->identifier, 16);
4491                 desc->length = desc->length > 12 ? 16 :
4492                     (desc->length > 8 ? 12 : 8);
4493                 len -= 16 - desc->length;
4494         }
4495         if (naa != NULL) {
4496                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4497                     desc->length);
4498                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4499                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4500                     SVPD_ID_TYPE_NAA;
4501                 desc->length = hex2bin(naa, desc->identifier, 16);
4502                 desc->length = desc->length > 8 ? 16 : 8;
4503                 len -= 16 - desc->length;
4504         }
4505         lun->lun_devid->len = len;
4506
4507         mtx_lock(&ctl_softc->ctl_lock);
4508         /*
4509          * See if the caller requested a particular LUN number.  If so, see
4510          * if it is available.  Otherwise, allocate the first available LUN.
4511          */
4512         if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4513                 if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4514                  || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4515                         mtx_unlock(&ctl_softc->ctl_lock);
4516                         if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4517                                 printf("ctl: requested LUN ID %d is higher "
4518                                        "than CTL_MAX_LUNS - 1 (%d)\n",
4519                                        be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4520                         } else {
4521                                 /*
4522                                  * XXX KDM return an error, or just assign
4523                                  * another LUN ID in this case??
4524                                  */
4525                                 printf("ctl: requested LUN ID %d is already "
4526                                        "in use\n", be_lun->req_lun_id);
4527                         }
4528                         if (lun->flags & CTL_LUN_MALLOCED)
4529                                 free(lun, M_CTL);
4530                         be_lun->lun_config_status(be_lun->be_lun,
4531                                                   CTL_LUN_CONFIG_FAILURE);
4532                         return (ENOSPC);
4533                 }
4534                 lun_number = be_lun->req_lun_id;
4535         } else {
4536                 lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, 0, CTL_MAX_LUNS);
4537                 if (lun_number == -1) {
4538                         mtx_unlock(&ctl_softc->ctl_lock);
4539                         printf("ctl: can't allocate LUN, out of LUNs\n");
4540                         if (lun->flags & CTL_LUN_MALLOCED)
4541                                 free(lun, M_CTL);
4542                         be_lun->lun_config_status(be_lun->be_lun,
4543                                                   CTL_LUN_CONFIG_FAILURE);
4544                         return (ENOSPC);
4545                 }
4546         }
4547         ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4548
4549         mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4550         lun->lun = lun_number;
4551         lun->be_lun = be_lun;
4552         /*
4553          * The processor LUN is always enabled.  Disk LUNs come on line
4554          * disabled, and must be enabled by the backend.
4555          */
4556         lun->flags |= CTL_LUN_DISABLED;
4557         lun->backend = be_lun->be;
4558         be_lun->ctl_lun = lun;
4559         be_lun->lun_id = lun_number;
4560         atomic_add_int(&be_lun->be->num_luns, 1);
4561         if (be_lun->flags & CTL_LUN_FLAG_OFFLINE)
4562                 lun->flags |= CTL_LUN_OFFLINE;
4563
4564         if (be_lun->flags & CTL_LUN_FLAG_POWERED_OFF)
4565                 lun->flags |= CTL_LUN_STOPPED;
4566
4567         if (be_lun->flags & CTL_LUN_FLAG_INOPERABLE)
4568                 lun->flags |= CTL_LUN_INOPERABLE;
4569
4570         if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4571                 lun->flags |= CTL_LUN_PRIMARY_SC;
4572
4573         lun->ctl_softc = ctl_softc;
4574 #ifdef CTL_TIME_IO
4575         lun->last_busy = getsbinuptime();
4576 #endif
4577         TAILQ_INIT(&lun->ooa_queue);
4578         TAILQ_INIT(&lun->blocked_queue);
4579         STAILQ_INIT(&lun->error_list);
4580         ctl_tpc_lun_init(lun);
4581
4582         /*
4583          * Initialize the mode and log page index.
4584          */
4585         ctl_init_page_index(lun);
4586         ctl_init_log_page_index(lun);
4587
4588         /*
4589          * Now, before we insert this lun on the lun list, set the lun
4590          * inventory changed UA for all other luns.
4591          */
4592         STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4593                 mtx_lock(&nlun->lun_lock);
4594                 ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4595                 mtx_unlock(&nlun->lun_lock);
4596         }
4597
4598         STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4599
4600         ctl_softc->ctl_luns[lun_number] = lun;
4601
4602         ctl_softc->num_luns++;
4603
4604         /* Setup statistics gathering */
4605         lun->stats.device_type = be_lun->lun_type;
4606         lun->stats.lun_number = lun_number;
4607         if (lun->stats.device_type == T_DIRECT)
4608                 lun->stats.blocksize = be_lun->blocksize;
4609         else
4610                 lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4611         for (i = 0;i < CTL_MAX_PORTS;i++)
4612                 lun->stats.ports[i].targ_port = i;
4613
4614         mtx_unlock(&ctl_softc->ctl_lock);
4615
4616         lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4617         return (0);
4618 }
4619
4620 /*
4621  * Delete a LUN.
4622  * Assumptions:
4623  * - LUN has already been marked invalid and any pending I/O has been taken
4624  *   care of.
4625  */
4626 static int
4627 ctl_free_lun(struct ctl_lun *lun)
4628 {
4629         struct ctl_softc *softc;
4630         struct ctl_lun *nlun;
4631         int i;
4632
4633         softc = lun->ctl_softc;
4634
4635         mtx_assert(&softc->ctl_lock, MA_OWNED);
4636
4637         STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4638
4639         ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4640
4641         softc->ctl_luns[lun->lun] = NULL;
4642
4643         if (!TAILQ_EMPTY(&lun->ooa_queue))
4644                 panic("Freeing a LUN %p with outstanding I/O!!\n", lun);
4645
4646         softc->num_luns--;
4647
4648         /*
4649          * Tell the backend to free resources, if this LUN has a backend.
4650          */
4651         atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4652         lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4653
4654         ctl_tpc_lun_shutdown(lun);
4655         mtx_destroy(&lun->lun_lock);
4656         free(lun->lun_devid, M_CTL);
4657         for (i = 0; i < CTL_MAX_PORTS; i++)
4658                 free(lun->pending_ua[i], M_CTL);
4659         for (i = 0; i < CTL_MAX_PORTS; i++)
4660                 free(lun->pr_keys[i], M_CTL);
4661         free(lun->write_buffer, M_CTL);
4662         if (lun->flags & CTL_LUN_MALLOCED)
4663                 free(lun, M_CTL);
4664
4665         STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4666                 mtx_lock(&nlun->lun_lock);
4667                 ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4668                 mtx_unlock(&nlun->lun_lock);
4669         }
4670
4671         return (0);
4672 }
4673
4674 static void
4675 ctl_create_lun(struct ctl_be_lun *be_lun)
4676 {
4677         struct ctl_softc *softc;
4678
4679         softc = control_softc;
4680
4681         /*
4682          * ctl_alloc_lun() should handle all potential failure cases.
4683          */
4684         ctl_alloc_lun(softc, NULL, be_lun);
4685 }
4686
4687 int
4688 ctl_add_lun(struct ctl_be_lun *be_lun)
4689 {
4690         struct ctl_softc *softc = control_softc;
4691
4692         mtx_lock(&softc->ctl_lock);
4693         STAILQ_INSERT_TAIL(&softc->pending_lun_queue, be_lun, links);
4694         mtx_unlock(&softc->ctl_lock);
4695         wakeup(&softc->pending_lun_queue);
4696
4697         return (0);
4698 }
4699
4700 int
4701 ctl_enable_lun(struct ctl_be_lun *be_lun)
4702 {
4703         struct ctl_softc *softc;
4704         struct ctl_port *port, *nport;
4705         struct ctl_lun *lun;
4706         int retval;
4707
4708         lun = (struct ctl_lun *)be_lun->ctl_lun;
4709         softc = lun->ctl_softc;
4710
4711         mtx_lock(&softc->ctl_lock);
4712         mtx_lock(&lun->lun_lock);
4713         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4714                 /*
4715                  * eh?  Why did we get called if the LUN is already
4716                  * enabled?
4717                  */
4718                 mtx_unlock(&lun->lun_lock);
4719                 mtx_unlock(&softc->ctl_lock);
4720                 return (0);
4721         }
4722         lun->flags &= ~CTL_LUN_DISABLED;
4723         mtx_unlock(&lun->lun_lock);
4724
4725         for (port = STAILQ_FIRST(&softc->port_list); port != NULL; port = nport) {
4726                 nport = STAILQ_NEXT(port, links);
4727                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4728                     port->lun_map != NULL || port->lun_enable == NULL)
4729                         continue;
4730
4731                 /*
4732                  * Drop the lock while we call the FETD's enable routine.
4733                  * This can lead to a callback into CTL (at least in the
4734                  * case of the internal initiator frontend.
4735                  */
4736                 mtx_unlock(&softc->ctl_lock);
4737                 retval = port->lun_enable(port->targ_lun_arg, lun->lun);
4738                 mtx_lock(&softc->ctl_lock);
4739                 if (retval != 0) {
4740                         printf("%s: FETD %s port %d returned error "
4741                                "%d for lun_enable on lun %jd\n",
4742                                __func__, port->port_name, port->targ_port,
4743                                retval, (intmax_t)lun->lun);
4744                 }
4745         }
4746
4747         mtx_unlock(&softc->ctl_lock);
4748         ctl_isc_announce_lun(lun);
4749
4750         return (0);
4751 }
4752
4753 int
4754 ctl_disable_lun(struct ctl_be_lun *be_lun)
4755 {
4756         struct ctl_softc *softc;
4757         struct ctl_port *port;
4758         struct ctl_lun *lun;
4759         int retval;
4760
4761         lun = (struct ctl_lun *)be_lun->ctl_lun;
4762         softc = lun->ctl_softc;
4763
4764         mtx_lock(&softc->ctl_lock);
4765         mtx_lock(&lun->lun_lock);
4766         if (lun->flags & CTL_LUN_DISABLED) {
4767                 mtx_unlock(&lun->lun_lock);
4768                 mtx_unlock(&softc->ctl_lock);
4769                 return (0);
4770         }
4771         lun->flags |= CTL_LUN_DISABLED;
4772         mtx_unlock(&lun->lun_lock);
4773
4774         STAILQ_FOREACH(port, &softc->port_list, links) {
4775                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4776                     port->lun_map != NULL || port->lun_disable == NULL)
4777                         continue;
4778
4779                 /*
4780                  * Drop the lock before we call the frontend's disable
4781                  * routine, to avoid lock order reversals.
4782                  *
4783                  * XXX KDM what happens if the frontend list changes while
4784                  * we're traversing it?  It's unlikely, but should be handled.
4785                  */
4786                 mtx_unlock(&softc->ctl_lock);
4787                 retval = port->lun_disable(port->targ_lun_arg, lun->lun);
4788                 mtx_lock(&softc->ctl_lock);
4789                 if (retval != 0) {
4790                         printf("%s: FETD %s port %d returned error "
4791                                "%d for lun_disable on lun %jd\n",
4792                                __func__, port->port_name, port->targ_port,
4793                                retval, (intmax_t)lun->lun);
4794                 }
4795         }
4796
4797         mtx_unlock(&softc->ctl_lock);
4798         ctl_isc_announce_lun(lun);
4799
4800         return (0);
4801 }
4802
4803 int
4804 ctl_start_lun(struct ctl_be_lun *be_lun)
4805 {
4806         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4807
4808         mtx_lock(&lun->lun_lock);
4809         lun->flags &= ~CTL_LUN_STOPPED;
4810         mtx_unlock(&lun->lun_lock);
4811         return (0);
4812 }
4813
4814 int
4815 ctl_stop_lun(struct ctl_be_lun *be_lun)
4816 {
4817         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4818
4819         mtx_lock(&lun->lun_lock);
4820         lun->flags |= CTL_LUN_STOPPED;
4821         mtx_unlock(&lun->lun_lock);
4822         return (0);
4823 }
4824
4825 int
4826 ctl_lun_offline(struct ctl_be_lun *be_lun)
4827 {
4828         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4829
4830         mtx_lock(&lun->lun_lock);
4831         lun->flags |= CTL_LUN_OFFLINE;
4832         mtx_unlock(&lun->lun_lock);
4833         return (0);
4834 }
4835
4836 int
4837 ctl_lun_online(struct ctl_be_lun *be_lun)
4838 {
4839         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4840
4841         mtx_lock(&lun->lun_lock);
4842         lun->flags &= ~CTL_LUN_OFFLINE;
4843         mtx_unlock(&lun->lun_lock);
4844         return (0);
4845 }
4846
4847 int
4848 ctl_lun_primary(struct ctl_be_lun *be_lun)
4849 {
4850         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4851
4852         mtx_lock(&lun->lun_lock);
4853         lun->flags |= CTL_LUN_PRIMARY_SC;
4854         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
4855         mtx_unlock(&lun->lun_lock);
4856         ctl_isc_announce_lun(lun);
4857         return (0);
4858 }
4859
4860 int
4861 ctl_lun_secondary(struct ctl_be_lun *be_lun)
4862 {
4863         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4864
4865         mtx_lock(&lun->lun_lock);
4866         lun->flags &= ~CTL_LUN_PRIMARY_SC;
4867         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
4868         mtx_unlock(&lun->lun_lock);
4869         ctl_isc_announce_lun(lun);
4870         return (0);
4871 }
4872
4873 int
4874 ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4875 {
4876         struct ctl_softc *softc;
4877         struct ctl_lun *lun;
4878
4879         lun = (struct ctl_lun *)be_lun->ctl_lun;
4880         softc = lun->ctl_softc;
4881
4882         mtx_lock(&lun->lun_lock);
4883
4884         /*
4885          * The LUN needs to be disabled before it can be marked invalid.
4886          */
4887         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4888                 mtx_unlock(&lun->lun_lock);
4889                 return (-1);
4890         }
4891         /*
4892          * Mark the LUN invalid.
4893          */
4894         lun->flags |= CTL_LUN_INVALID;
4895
4896         /*
4897          * If there is nothing in the OOA queue, go ahead and free the LUN.
4898          * If we have something in the OOA queue, we'll free it when the
4899          * last I/O completes.
4900          */
4901         if (TAILQ_EMPTY(&lun->ooa_queue)) {
4902                 mtx_unlock(&lun->lun_lock);
4903                 mtx_lock(&softc->ctl_lock);
4904                 ctl_free_lun(lun);
4905                 mtx_unlock(&softc->ctl_lock);
4906         } else
4907                 mtx_unlock(&lun->lun_lock);
4908
4909         return (0);
4910 }
4911
4912 int
4913 ctl_lun_inoperable(struct ctl_be_lun *be_lun)
4914 {
4915         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4916
4917         mtx_lock(&lun->lun_lock);
4918         lun->flags |= CTL_LUN_INOPERABLE;
4919         mtx_unlock(&lun->lun_lock);
4920         return (0);
4921 }
4922
4923 int
4924 ctl_lun_operable(struct ctl_be_lun *be_lun)
4925 {
4926         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4927
4928         mtx_lock(&lun->lun_lock);
4929         lun->flags &= ~CTL_LUN_INOPERABLE;
4930         mtx_unlock(&lun->lun_lock);
4931         return (0);
4932 }
4933
4934 void
4935 ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
4936 {
4937         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4938         union ctl_ha_msg msg;
4939
4940         mtx_lock(&lun->lun_lock);
4941         ctl_est_ua_all(lun, -1, CTL_UA_CAPACITY_CHANGED);
4942         mtx_unlock(&lun->lun_lock);
4943         if (lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
4944                 /* Send msg to other side. */
4945                 bzero(&msg.ua, sizeof(msg.ua));
4946                 msg.hdr.msg_type = CTL_MSG_UA;
4947                 msg.hdr.nexus.initid = -1;
4948                 msg.hdr.nexus.targ_port = -1;
4949                 msg.hdr.nexus.targ_lun = lun->lun;
4950                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
4951                 msg.ua.ua_all = 1;
4952                 msg.ua.ua_set = 1;
4953                 msg.ua.ua_type = CTL_UA_CAPACITY_CHANGED;
4954                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.ua),
4955                     M_WAITOK);
4956         }
4957 }
4958
4959 /*
4960  * Backend "memory move is complete" callback for requests that never
4961  * make it down to say RAIDCore's configuration code.
4962  */
4963 int
4964 ctl_config_move_done(union ctl_io *io)
4965 {
4966         int retval;
4967
4968         CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
4969         KASSERT(io->io_hdr.io_type == CTL_IO_SCSI,
4970             ("Config I/O type isn't CTL_IO_SCSI (%d)!", io->io_hdr.io_type));
4971
4972         if ((io->io_hdr.port_status != 0) &&
4973             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
4974              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
4975                 /*
4976                  * For hardware error sense keys, the sense key
4977                  * specific value is defined to be a retry count,
4978                  * but we use it to pass back an internal FETD
4979                  * error code.  XXX KDM  Hopefully the FETD is only
4980                  * using 16 bits for an error code, since that's
4981                  * all the space we have in the sks field.
4982                  */
4983                 ctl_set_internal_failure(&io->scsiio,
4984                                          /*sks_valid*/ 1,
4985                                          /*retry_count*/
4986                                          io->io_hdr.port_status);
4987         }
4988
4989         if (ctl_debug & CTL_DEBUG_CDB_DATA)
4990                 ctl_data_print(io);
4991         if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) ||
4992             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
4993              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) ||
4994             ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
4995                 /*
4996                  * XXX KDM just assuming a single pointer here, and not a
4997                  * S/G list.  If we start using S/G lists for config data,
4998                  * we'll need to know how to clean them up here as well.
4999                  */
5000                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5001                         free(io->scsiio.kern_data_ptr, M_CTL);
5002                 ctl_done(io);
5003                 retval = CTL_RETVAL_COMPLETE;
5004         } else {
5005                 /*
5006                  * XXX KDM now we need to continue data movement.  Some
5007                  * options:
5008                  * - call ctl_scsiio() again?  We don't do this for data
5009                  *   writes, because for those at least we know ahead of
5010                  *   time where the write will go and how long it is.  For
5011                  *   config writes, though, that information is largely
5012                  *   contained within the write itself, thus we need to
5013                  *   parse out the data again.
5014                  *
5015                  * - Call some other function once the data is in?
5016                  */
5017
5018                 /*
5019                  * XXX KDM call ctl_scsiio() again for now, and check flag
5020                  * bits to see whether we're allocated or not.
5021                  */
5022                 retval = ctl_scsiio(&io->scsiio);
5023         }
5024         return (retval);
5025 }
5026
5027 /*
5028  * This gets called by a backend driver when it is done with a
5029  * data_submit method.
5030  */
5031 void
5032 ctl_data_submit_done(union ctl_io *io)
5033 {
5034         /*
5035          * If the IO_CONT flag is set, we need to call the supplied
5036          * function to continue processing the I/O, instead of completing
5037          * the I/O just yet.
5038          *
5039          * If there is an error, though, we don't want to keep processing.
5040          * Instead, just send status back to the initiator.
5041          */
5042         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5043             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5044             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5045              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5046                 io->scsiio.io_cont(io);
5047                 return;
5048         }
5049         ctl_done(io);
5050 }
5051
5052 /*
5053  * This gets called by a backend driver when it is done with a
5054  * configuration write.
5055  */
5056 void
5057 ctl_config_write_done(union ctl_io *io)
5058 {
5059         uint8_t *buf;
5060
5061         /*
5062          * If the IO_CONT flag is set, we need to call the supplied
5063          * function to continue processing the I/O, instead of completing
5064          * the I/O just yet.
5065          *
5066          * If there is an error, though, we don't want to keep processing.
5067          * Instead, just send status back to the initiator.
5068          */
5069         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5070             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5071             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5072              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5073                 io->scsiio.io_cont(io);
5074                 return;
5075         }
5076         /*
5077          * Since a configuration write can be done for commands that actually
5078          * have data allocated, like write buffer, and commands that have
5079          * no data, like start/stop unit, we need to check here.
5080          */
5081         if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5082                 buf = io->scsiio.kern_data_ptr;
5083         else
5084                 buf = NULL;
5085         ctl_done(io);
5086         if (buf)
5087                 free(buf, M_CTL);
5088 }
5089
5090 void
5091 ctl_config_read_done(union ctl_io *io)
5092 {
5093         uint8_t *buf;
5094
5095         /*
5096          * If there is some error -- we are done, skip data transfer.
5097          */
5098         if ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0 ||
5099             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
5100              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
5101                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5102                         buf = io->scsiio.kern_data_ptr;
5103                 else
5104                         buf = NULL;
5105                 ctl_done(io);
5106                 if (buf)
5107                         free(buf, M_CTL);
5108                 return;
5109         }
5110
5111         /*
5112          * If the IO_CONT flag is set, we need to call the supplied
5113          * function to continue processing the I/O, instead of completing
5114          * the I/O just yet.
5115          */
5116         if (io->io_hdr.flags & CTL_FLAG_IO_CONT) {
5117                 io->scsiio.io_cont(io);
5118                 return;
5119         }
5120
5121         ctl_datamove(io);
5122 }
5123
5124 /*
5125  * SCSI release command.
5126  */
5127 int
5128 ctl_scsi_release(struct ctl_scsiio *ctsio)
5129 {
5130         int length, longid, thirdparty_id, resv_id;
5131         struct ctl_lun *lun;
5132         uint32_t residx;
5133
5134         length = 0;
5135         resv_id = 0;
5136
5137         CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5138
5139         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5140         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5141
5142         switch (ctsio->cdb[0]) {
5143         case RELEASE_10: {
5144                 struct scsi_release_10 *cdb;
5145
5146                 cdb = (struct scsi_release_10 *)ctsio->cdb;
5147
5148                 if (cdb->byte2 & SR10_LONGID)
5149                         longid = 1;
5150                 else
5151                         thirdparty_id = cdb->thirdparty_id;
5152
5153                 resv_id = cdb->resv_id;
5154                 length = scsi_2btoul(cdb->length);
5155                 break;
5156         }
5157         }
5158
5159
5160         /*
5161          * XXX KDM right now, we only support LUN reservation.  We don't
5162          * support 3rd party reservations, or extent reservations, which
5163          * might actually need the parameter list.  If we've gotten this
5164          * far, we've got a LUN reservation.  Anything else got kicked out
5165          * above.  So, according to SPC, ignore the length.
5166          */
5167         length = 0;
5168
5169         if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5170          && (length > 0)) {
5171                 ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5172                 ctsio->kern_data_len = length;
5173                 ctsio->kern_total_len = length;
5174                 ctsio->kern_data_resid = 0;
5175                 ctsio->kern_rel_offset = 0;
5176                 ctsio->kern_sg_entries = 0;
5177                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5178                 ctsio->be_move_done = ctl_config_move_done;
5179                 ctl_datamove((union ctl_io *)ctsio);
5180
5181                 return (CTL_RETVAL_COMPLETE);
5182         }
5183
5184         if (length > 0)
5185                 thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5186
5187         mtx_lock(&lun->lun_lock);
5188
5189         /*
5190          * According to SPC, it is not an error for an intiator to attempt
5191          * to release a reservation on a LUN that isn't reserved, or that
5192          * is reserved by another initiator.  The reservation can only be
5193          * released, though, by the initiator who made it or by one of
5194          * several reset type events.
5195          */
5196         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
5197                         lun->flags &= ~CTL_LUN_RESERVED;
5198
5199         mtx_unlock(&lun->lun_lock);
5200
5201         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5202                 free(ctsio->kern_data_ptr, M_CTL);
5203                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5204         }
5205
5206         ctl_set_success(ctsio);
5207         ctl_done((union ctl_io *)ctsio);
5208         return (CTL_RETVAL_COMPLETE);
5209 }
5210
5211 int
5212 ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5213 {
5214         int extent, thirdparty, longid;
5215         int resv_id, length;
5216         uint64_t thirdparty_id;
5217         struct ctl_lun *lun;
5218         uint32_t residx;
5219
5220         extent = 0;
5221         thirdparty = 0;
5222         longid = 0;
5223         resv_id = 0;
5224         length = 0;
5225         thirdparty_id = 0;
5226
5227         CTL_DEBUG_PRINT(("ctl_reserve\n"));
5228
5229         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5230         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5231
5232         switch (ctsio->cdb[0]) {
5233         case RESERVE_10: {
5234                 struct scsi_reserve_10 *cdb;
5235
5236                 cdb = (struct scsi_reserve_10 *)ctsio->cdb;
5237
5238                 if (cdb->byte2 & SR10_LONGID)
5239                         longid = 1;
5240                 else
5241                         thirdparty_id = cdb->thirdparty_id;
5242
5243                 resv_id = cdb->resv_id;
5244                 length = scsi_2btoul(cdb->length);
5245                 break;
5246         }
5247         }
5248
5249         /*
5250          * XXX KDM right now, we only support LUN reservation.  We don't
5251          * support 3rd party reservations, or extent reservations, which
5252          * might actually need the parameter list.  If we've gotten this
5253          * far, we've got a LUN reservation.  Anything else got kicked out
5254          * above.  So, according to SPC, ignore the length.
5255          */
5256         length = 0;
5257
5258         if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5259          && (length > 0)) {
5260                 ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5261                 ctsio->kern_data_len = length;
5262                 ctsio->kern_total_len = length;
5263                 ctsio->kern_data_resid = 0;
5264                 ctsio->kern_rel_offset = 0;
5265                 ctsio->kern_sg_entries = 0;
5266                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5267                 ctsio->be_move_done = ctl_config_move_done;
5268                 ctl_datamove((union ctl_io *)ctsio);
5269
5270                 return (CTL_RETVAL_COMPLETE);
5271         }
5272
5273         if (length > 0)
5274                 thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5275
5276         mtx_lock(&lun->lun_lock);
5277         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx != residx)) {
5278                 ctl_set_reservation_conflict(ctsio);
5279                 goto bailout;
5280         }
5281
5282         lun->flags |= CTL_LUN_RESERVED;
5283         lun->res_idx = residx;
5284
5285         ctl_set_success(ctsio);
5286
5287 bailout:
5288         mtx_unlock(&lun->lun_lock);
5289
5290         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5291                 free(ctsio->kern_data_ptr, M_CTL);
5292                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5293         }
5294
5295         ctl_done((union ctl_io *)ctsio);
5296         return (CTL_RETVAL_COMPLETE);
5297 }
5298
5299 int
5300 ctl_start_stop(struct ctl_scsiio *ctsio)
5301 {
5302         struct scsi_start_stop_unit *cdb;
5303         struct ctl_lun *lun;
5304         int retval;
5305
5306         CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5307
5308         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5309         retval = 0;
5310
5311         cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5312
5313         /*
5314          * XXX KDM
5315          * We don't support the immediate bit on a stop unit.  In order to
5316          * do that, we would need to code up a way to know that a stop is
5317          * pending, and hold off any new commands until it completes, one
5318          * way or another.  Then we could accept or reject those commands
5319          * depending on its status.  We would almost need to do the reverse
5320          * of what we do below for an immediate start -- return the copy of
5321          * the ctl_io to the FETD with status to send to the host (and to
5322          * free the copy!) and then free the original I/O once the stop
5323          * actually completes.  That way, the OOA queue mechanism can work
5324          * to block commands that shouldn't proceed.  Another alternative
5325          * would be to put the copy in the queue in place of the original,
5326          * and return the original back to the caller.  That could be
5327          * slightly safer..
5328          */
5329         if ((cdb->byte2 & SSS_IMMED)
5330          && ((cdb->how & SSS_START) == 0)) {
5331                 ctl_set_invalid_field(ctsio,
5332                                       /*sks_valid*/ 1,
5333                                       /*command*/ 1,
5334                                       /*field*/ 1,
5335                                       /*bit_valid*/ 1,
5336                                       /*bit*/ 0);
5337                 ctl_done((union ctl_io *)ctsio);
5338                 return (CTL_RETVAL_COMPLETE);
5339         }
5340
5341         if ((lun->flags & CTL_LUN_PR_RESERVED)
5342          && ((cdb->how & SSS_START)==0)) {
5343                 uint32_t residx;
5344
5345                 residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5346                 if (ctl_get_prkey(lun, residx) == 0
5347                  || (lun->pr_res_idx!=residx && lun->res_type < 4)) {
5348
5349                         ctl_set_reservation_conflict(ctsio);
5350                         ctl_done((union ctl_io *)ctsio);
5351                         return (CTL_RETVAL_COMPLETE);
5352                 }
5353         }
5354
5355         /*
5356          * If there is no backend on this device, we can't start or stop
5357          * it.  In theory we shouldn't get any start/stop commands in the
5358          * first place at this level if the LUN doesn't have a backend.
5359          * That should get stopped by the command decode code.
5360          */
5361         if (lun->backend == NULL) {
5362                 ctl_set_invalid_opcode(ctsio);
5363                 ctl_done((union ctl_io *)ctsio);
5364                 return (CTL_RETVAL_COMPLETE);
5365         }
5366
5367         /*
5368          * XXX KDM Copan-specific offline behavior.
5369          * Figure out a reasonable way to port this?
5370          */
5371 #ifdef NEEDTOPORT
5372         mtx_lock(&lun->lun_lock);
5373
5374         if (((cdb->byte2 & SSS_ONOFFLINE) == 0)
5375          && (lun->flags & CTL_LUN_OFFLINE)) {
5376                 /*
5377                  * If the LUN is offline, and the on/offline bit isn't set,
5378                  * reject the start or stop.  Otherwise, let it through.
5379                  */
5380                 mtx_unlock(&lun->lun_lock);
5381                 ctl_set_lun_not_ready(ctsio);
5382                 ctl_done((union ctl_io *)ctsio);
5383         } else {
5384                 mtx_unlock(&lun->lun_lock);
5385 #endif /* NEEDTOPORT */
5386                 /*
5387                  * This could be a start or a stop when we're online,
5388                  * or a stop/offline or start/online.  A start or stop when
5389                  * we're offline is covered in the case above.
5390                  */
5391                 /*
5392                  * In the non-immediate case, we send the request to
5393                  * the backend and return status to the user when
5394                  * it is done.
5395                  *
5396                  * In the immediate case, we allocate a new ctl_io
5397                  * to hold a copy of the request, and send that to
5398                  * the backend.  We then set good status on the
5399                  * user's request and return it immediately.
5400                  */
5401                 if (cdb->byte2 & SSS_IMMED) {
5402                         union ctl_io *new_io;
5403
5404                         new_io = ctl_alloc_io(ctsio->io_hdr.pool);
5405                         ctl_copy_io((union ctl_io *)ctsio, new_io);
5406                         retval = lun->backend->config_write(new_io);
5407                         ctl_set_success(ctsio);
5408                         ctl_done((union ctl_io *)ctsio);
5409                 } else {
5410                         retval = lun->backend->config_write(
5411                                 (union ctl_io *)ctsio);
5412                 }
5413 #ifdef NEEDTOPORT
5414         }
5415 #endif
5416         return (retval);
5417 }
5418
5419 /*
5420  * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5421  * we don't really do anything with the LBA and length fields if the user
5422  * passes them in.  Instead we'll just flush out the cache for the entire
5423  * LUN.
5424  */
5425 int
5426 ctl_sync_cache(struct ctl_scsiio *ctsio)
5427 {
5428         struct ctl_lun *lun;
5429         struct ctl_softc *softc;
5430         struct ctl_lba_len_flags *lbalen;
5431         uint64_t starting_lba;
5432         uint32_t block_count;
5433         int retval;
5434         uint8_t byte2;
5435
5436         CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5437
5438         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5439         softc = lun->ctl_softc;
5440         retval = 0;
5441
5442         switch (ctsio->cdb[0]) {
5443         case SYNCHRONIZE_CACHE: {
5444                 struct scsi_sync_cache *cdb;
5445                 cdb = (struct scsi_sync_cache *)ctsio->cdb;
5446
5447                 starting_lba = scsi_4btoul(cdb->begin_lba);
5448                 block_count = scsi_2btoul(cdb->lb_count);
5449                 byte2 = cdb->byte2;
5450                 break;
5451         }
5452         case SYNCHRONIZE_CACHE_16: {
5453                 struct scsi_sync_cache_16 *cdb;
5454                 cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5455
5456                 starting_lba = scsi_8btou64(cdb->begin_lba);
5457                 block_count = scsi_4btoul(cdb->lb_count);
5458                 byte2 = cdb->byte2;
5459                 break;
5460         }
5461         default:
5462                 ctl_set_invalid_opcode(ctsio);
5463                 ctl_done((union ctl_io *)ctsio);
5464                 goto bailout;
5465                 break; /* NOTREACHED */
5466         }
5467
5468         /*
5469          * We check the LBA and length, but don't do anything with them.
5470          * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5471          * get flushed.  This check will just help satisfy anyone who wants
5472          * to see an error for an out of range LBA.
5473          */
5474         if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5475                 ctl_set_lba_out_of_range(ctsio);
5476                 ctl_done((union ctl_io *)ctsio);
5477                 goto bailout;
5478         }
5479
5480         /*
5481          * If this LUN has no backend, we can't flush the cache anyway.
5482          */
5483         if (lun->backend == NULL) {
5484                 ctl_set_invalid_opcode(ctsio);
5485                 ctl_done((union ctl_io *)ctsio);
5486                 goto bailout;
5487         }
5488
5489         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5490         lbalen->lba = starting_lba;
5491         lbalen->len = block_count;
5492         lbalen->flags = byte2;
5493
5494         /*
5495          * Check to see whether we're configured to send the SYNCHRONIZE
5496          * CACHE command directly to the back end.
5497          */
5498         mtx_lock(&lun->lun_lock);
5499         if ((softc->flags & CTL_FLAG_REAL_SYNC)
5500          && (++(lun->sync_count) >= lun->sync_interval)) {
5501                 lun->sync_count = 0;
5502                 mtx_unlock(&lun->lun_lock);
5503                 retval = lun->backend->config_write((union ctl_io *)ctsio);
5504         } else {
5505                 mtx_unlock(&lun->lun_lock);
5506                 ctl_set_success(ctsio);
5507                 ctl_done((union ctl_io *)ctsio);
5508         }
5509
5510 bailout:
5511
5512         return (retval);
5513 }
5514
5515 int
5516 ctl_format(struct ctl_scsiio *ctsio)
5517 {
5518         struct scsi_format *cdb;
5519         struct ctl_lun *lun;
5520         int length, defect_list_len;
5521
5522         CTL_DEBUG_PRINT(("ctl_format\n"));
5523
5524         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5525
5526         cdb = (struct scsi_format *)ctsio->cdb;
5527
5528         length = 0;
5529         if (cdb->byte2 & SF_FMTDATA) {
5530                 if (cdb->byte2 & SF_LONGLIST)
5531                         length = sizeof(struct scsi_format_header_long);
5532                 else
5533                         length = sizeof(struct scsi_format_header_short);
5534         }
5535
5536         if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5537          && (length > 0)) {
5538                 ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5539                 ctsio->kern_data_len = length;
5540                 ctsio->kern_total_len = length;
5541                 ctsio->kern_data_resid = 0;
5542                 ctsio->kern_rel_offset = 0;
5543                 ctsio->kern_sg_entries = 0;
5544                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5545                 ctsio->be_move_done = ctl_config_move_done;
5546                 ctl_datamove((union ctl_io *)ctsio);
5547
5548                 return (CTL_RETVAL_COMPLETE);
5549         }
5550
5551         defect_list_len = 0;
5552
5553         if (cdb->byte2 & SF_FMTDATA) {
5554                 if (cdb->byte2 & SF_LONGLIST) {
5555                         struct scsi_format_header_long *header;
5556
5557                         header = (struct scsi_format_header_long *)
5558                                 ctsio->kern_data_ptr;
5559
5560                         defect_list_len = scsi_4btoul(header->defect_list_len);
5561                         if (defect_list_len != 0) {
5562                                 ctl_set_invalid_field(ctsio,
5563                                                       /*sks_valid*/ 1,
5564                                                       /*command*/ 0,
5565                                                       /*field*/ 2,
5566                                                       /*bit_valid*/ 0,
5567                                                       /*bit*/ 0);
5568                                 goto bailout;
5569                         }
5570                 } else {
5571                         struct scsi_format_header_short *header;
5572
5573                         header = (struct scsi_format_header_short *)
5574                                 ctsio->kern_data_ptr;
5575
5576                         defect_list_len = scsi_2btoul(header->defect_list_len);
5577                         if (defect_list_len != 0) {
5578                                 ctl_set_invalid_field(ctsio,
5579                                                       /*sks_valid*/ 1,
5580                                                       /*command*/ 0,
5581                                                       /*field*/ 2,
5582                                                       /*bit_valid*/ 0,
5583                                                       /*bit*/ 0);
5584                                 goto bailout;
5585                         }
5586                 }
5587         }
5588
5589         /*
5590          * The format command will clear out the "Medium format corrupted"
5591          * status if set by the configuration code.  That status is really
5592          * just a way to notify the host that we have lost the media, and
5593          * get them to issue a command that will basically make them think
5594          * they're blowing away the media.
5595          */
5596         mtx_lock(&lun->lun_lock);
5597         lun->flags &= ~CTL_LUN_INOPERABLE;
5598         mtx_unlock(&lun->lun_lock);
5599
5600         ctl_set_success(ctsio);
5601 bailout:
5602
5603         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5604                 free(ctsio->kern_data_ptr, M_CTL);
5605                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5606         }
5607
5608         ctl_done((union ctl_io *)ctsio);
5609         return (CTL_RETVAL_COMPLETE);
5610 }
5611
5612 int
5613 ctl_read_buffer(struct ctl_scsiio *ctsio)
5614 {
5615         struct ctl_lun *lun;
5616         uint64_t buffer_offset;
5617         uint32_t len;
5618         uint8_t byte2;
5619         static uint8_t descr[4];
5620         static uint8_t echo_descr[4] = { 0 };
5621
5622         CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5623         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5624         switch (ctsio->cdb[0]) {
5625         case READ_BUFFER: {
5626                 struct scsi_read_buffer *cdb;
5627
5628                 cdb = (struct scsi_read_buffer *)ctsio->cdb;
5629                 buffer_offset = scsi_3btoul(cdb->offset);
5630                 len = scsi_3btoul(cdb->length);
5631                 byte2 = cdb->byte2;
5632                 break;
5633         }
5634         case READ_BUFFER_16: {
5635                 struct scsi_read_buffer_16 *cdb;
5636
5637                 cdb = (struct scsi_read_buffer_16 *)ctsio->cdb;
5638                 buffer_offset = scsi_8btou64(cdb->offset);
5639                 len = scsi_4btoul(cdb->length);
5640                 byte2 = cdb->byte2;
5641                 break;
5642         }
5643         default: /* This shouldn't happen. */
5644                 ctl_set_invalid_opcode(ctsio);
5645                 ctl_done((union ctl_io *)ctsio);
5646                 return (CTL_RETVAL_COMPLETE);
5647         }
5648
5649         if ((byte2 & RWB_MODE) != RWB_MODE_DATA &&
5650             (byte2 & RWB_MODE) != RWB_MODE_ECHO_DESCR &&
5651             (byte2 & RWB_MODE) != RWB_MODE_DESCR) {
5652                 ctl_set_invalid_field(ctsio,
5653                                       /*sks_valid*/ 1,
5654                                       /*command*/ 1,
5655                                       /*field*/ 1,
5656                                       /*bit_valid*/ 1,
5657                                       /*bit*/ 4);
5658                 ctl_done((union ctl_io *)ctsio);
5659                 return (CTL_RETVAL_COMPLETE);
5660         }
5661
5662         if (buffer_offset > CTL_WRITE_BUFFER_SIZE ||
5663             buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5664                 ctl_set_invalid_field(ctsio,
5665                                       /*sks_valid*/ 1,
5666                                       /*command*/ 1,
5667                                       /*field*/ 6,
5668                                       /*bit_valid*/ 0,
5669                                       /*bit*/ 0);
5670                 ctl_done((union ctl_io *)ctsio);
5671                 return (CTL_RETVAL_COMPLETE);
5672         }
5673
5674         if ((byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5675                 descr[0] = 0;
5676                 scsi_ulto3b(CTL_WRITE_BUFFER_SIZE, &descr[1]);
5677                 ctsio->kern_data_ptr = descr;
5678                 len = min(len, sizeof(descr));
5679         } else if ((byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5680                 ctsio->kern_data_ptr = echo_descr;
5681                 len = min(len, sizeof(echo_descr));
5682         } else {
5683                 if (lun->write_buffer == NULL) {
5684                         lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5685                             M_CTL, M_WAITOK);
5686                 }
5687                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5688         }
5689         ctsio->kern_data_len = len;
5690         ctsio->kern_total_len = len;
5691         ctsio->kern_data_resid = 0;
5692         ctsio->kern_rel_offset = 0;
5693         ctsio->kern_sg_entries = 0;
5694         ctl_set_success(ctsio);
5695         ctsio->be_move_done = ctl_config_move_done;
5696         ctl_datamove((union ctl_io *)ctsio);
5697         return (CTL_RETVAL_COMPLETE);
5698 }
5699
5700 int
5701 ctl_write_buffer(struct ctl_scsiio *ctsio)
5702 {
5703         struct scsi_write_buffer *cdb;
5704         struct ctl_lun *lun;
5705         int buffer_offset, len;
5706
5707         CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5708
5709         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5710         cdb = (struct scsi_write_buffer *)ctsio->cdb;
5711
5712         if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5713                 ctl_set_invalid_field(ctsio,
5714                                       /*sks_valid*/ 1,
5715                                       /*command*/ 1,
5716                                       /*field*/ 1,
5717                                       /*bit_valid*/ 1,
5718                                       /*bit*/ 4);
5719                 ctl_done((union ctl_io *)ctsio);
5720                 return (CTL_RETVAL_COMPLETE);
5721         }
5722
5723         len = scsi_3btoul(cdb->length);
5724         buffer_offset = scsi_3btoul(cdb->offset);
5725
5726         if (buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5727                 ctl_set_invalid_field(ctsio,
5728                                       /*sks_valid*/ 1,
5729                                       /*command*/ 1,
5730                                       /*field*/ 6,
5731                                       /*bit_valid*/ 0,
5732                                       /*bit*/ 0);
5733                 ctl_done((union ctl_io *)ctsio);
5734                 return (CTL_RETVAL_COMPLETE);
5735         }
5736
5737         /*
5738          * If we've got a kernel request that hasn't been malloced yet,
5739          * malloc it and tell the caller the data buffer is here.
5740          */
5741         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5742                 if (lun->write_buffer == NULL) {
5743                         lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5744                             M_CTL, M_WAITOK);
5745                 }
5746                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5747                 ctsio->kern_data_len = len;
5748                 ctsio->kern_total_len = len;
5749                 ctsio->kern_data_resid = 0;
5750                 ctsio->kern_rel_offset = 0;
5751                 ctsio->kern_sg_entries = 0;
5752                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5753                 ctsio->be_move_done = ctl_config_move_done;
5754                 ctl_datamove((union ctl_io *)ctsio);
5755
5756                 return (CTL_RETVAL_COMPLETE);
5757         }
5758
5759         ctl_set_success(ctsio);
5760         ctl_done((union ctl_io *)ctsio);
5761         return (CTL_RETVAL_COMPLETE);
5762 }
5763
5764 int
5765 ctl_write_same(struct ctl_scsiio *ctsio)
5766 {
5767         struct ctl_lun *lun;
5768         struct ctl_lba_len_flags *lbalen;
5769         uint64_t lba;
5770         uint32_t num_blocks;
5771         int len, retval;
5772         uint8_t byte2;
5773
5774         retval = CTL_RETVAL_COMPLETE;
5775
5776         CTL_DEBUG_PRINT(("ctl_write_same\n"));
5777
5778         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5779
5780         switch (ctsio->cdb[0]) {
5781         case WRITE_SAME_10: {
5782                 struct scsi_write_same_10 *cdb;
5783
5784                 cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5785
5786                 lba = scsi_4btoul(cdb->addr);
5787                 num_blocks = scsi_2btoul(cdb->length);
5788                 byte2 = cdb->byte2;
5789                 break;
5790         }
5791         case WRITE_SAME_16: {
5792                 struct scsi_write_same_16 *cdb;
5793
5794                 cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5795
5796                 lba = scsi_8btou64(cdb->addr);
5797                 num_blocks = scsi_4btoul(cdb->length);
5798                 byte2 = cdb->byte2;
5799                 break;
5800         }
5801         default:
5802                 /*
5803                  * We got a command we don't support.  This shouldn't
5804                  * happen, commands should be filtered out above us.
5805                  */
5806                 ctl_set_invalid_opcode(ctsio);
5807                 ctl_done((union ctl_io *)ctsio);
5808
5809                 return (CTL_RETVAL_COMPLETE);
5810                 break; /* NOTREACHED */
5811         }
5812
5813         /* NDOB and ANCHOR flags can be used only together with UNMAP */
5814         if ((byte2 & SWS_UNMAP) == 0 &&
5815             (byte2 & (SWS_NDOB | SWS_ANCHOR)) != 0) {
5816                 ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
5817                     /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
5818                 ctl_done((union ctl_io *)ctsio);
5819                 return (CTL_RETVAL_COMPLETE);
5820         }
5821
5822         /*
5823          * The first check is to make sure we're in bounds, the second
5824          * check is to catch wrap-around problems.  If the lba + num blocks
5825          * is less than the lba, then we've wrapped around and the block
5826          * range is invalid anyway.
5827          */
5828         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5829          || ((lba + num_blocks) < lba)) {
5830                 ctl_set_lba_out_of_range(ctsio);
5831                 ctl_done((union ctl_io *)ctsio);
5832                 return (CTL_RETVAL_COMPLETE);
5833         }
5834
5835         /* Zero number of blocks means "to the last logical block" */
5836         if (num_blocks == 0) {
5837                 if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5838                         ctl_set_invalid_field(ctsio,
5839                                               /*sks_valid*/ 0,
5840                                               /*command*/ 1,
5841                                               /*field*/ 0,
5842                                               /*bit_valid*/ 0,
5843                                               /*bit*/ 0);
5844                         ctl_done((union ctl_io *)ctsio);
5845                         return (CTL_RETVAL_COMPLETE);
5846                 }
5847                 num_blocks = (lun->be_lun->maxlba + 1) - lba;
5848         }
5849
5850         len = lun->be_lun->blocksize;
5851
5852         /*
5853          * If we've got a kernel request that hasn't been malloced yet,
5854          * malloc it and tell the caller the data buffer is here.
5855          */
5856         if ((byte2 & SWS_NDOB) == 0 &&
5857             (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5858                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
5859                 ctsio->kern_data_len = len;
5860                 ctsio->kern_total_len = len;
5861                 ctsio->kern_data_resid = 0;
5862                 ctsio->kern_rel_offset = 0;
5863                 ctsio->kern_sg_entries = 0;
5864                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5865                 ctsio->be_move_done = ctl_config_move_done;
5866                 ctl_datamove((union ctl_io *)ctsio);
5867
5868                 return (CTL_RETVAL_COMPLETE);
5869         }
5870
5871         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5872         lbalen->lba = lba;
5873         lbalen->len = num_blocks;
5874         lbalen->flags = byte2;
5875         retval = lun->backend->config_write((union ctl_io *)ctsio);
5876
5877         return (retval);
5878 }
5879
5880 int
5881 ctl_unmap(struct ctl_scsiio *ctsio)
5882 {
5883         struct ctl_lun *lun;
5884         struct scsi_unmap *cdb;
5885         struct ctl_ptr_len_flags *ptrlen;
5886         struct scsi_unmap_header *hdr;
5887         struct scsi_unmap_desc *buf, *end, *endnz, *range;
5888         uint64_t lba;
5889         uint32_t num_blocks;
5890         int len, retval;
5891         uint8_t byte2;
5892
5893         retval = CTL_RETVAL_COMPLETE;
5894
5895         CTL_DEBUG_PRINT(("ctl_unmap\n"));
5896
5897         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5898         cdb = (struct scsi_unmap *)ctsio->cdb;
5899
5900         len = scsi_2btoul(cdb->length);
5901         byte2 = cdb->byte2;
5902
5903         /*
5904          * If we've got a kernel request that hasn't been malloced yet,
5905          * malloc it and tell the caller the data buffer is here.
5906          */
5907         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5908                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
5909                 ctsio->kern_data_len = len;
5910                 ctsio->kern_total_len = len;
5911                 ctsio->kern_data_resid = 0;
5912                 ctsio->kern_rel_offset = 0;
5913                 ctsio->kern_sg_entries = 0;
5914                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5915                 ctsio->be_move_done = ctl_config_move_done;
5916                 ctl_datamove((union ctl_io *)ctsio);
5917
5918                 return (CTL_RETVAL_COMPLETE);
5919         }
5920
5921         len = ctsio->kern_total_len - ctsio->kern_data_resid;
5922         hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
5923         if (len < sizeof (*hdr) ||
5924             len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
5925             len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
5926             scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
5927                 ctl_set_invalid_field(ctsio,
5928                                       /*sks_valid*/ 0,
5929                                       /*command*/ 0,
5930                                       /*field*/ 0,
5931                                       /*bit_valid*/ 0,
5932                                       /*bit*/ 0);
5933                 goto done;
5934         }
5935         len = scsi_2btoul(hdr->desc_length);
5936         buf = (struct scsi_unmap_desc *)(hdr + 1);
5937         end = buf + len / sizeof(*buf);
5938
5939         endnz = buf;
5940         for (range = buf; range < end; range++) {
5941                 lba = scsi_8btou64(range->lba);
5942                 num_blocks = scsi_4btoul(range->length);
5943                 if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5944                  || ((lba + num_blocks) < lba)) {
5945                         ctl_set_lba_out_of_range(ctsio);
5946                         ctl_done((union ctl_io *)ctsio);
5947                         return (CTL_RETVAL_COMPLETE);
5948                 }
5949                 if (num_blocks != 0)
5950                         endnz = range + 1;
5951         }
5952
5953         /*
5954          * Block backend can not handle zero last range.
5955          * Filter it out and return if there is nothing left.
5956          */
5957         len = (uint8_t *)endnz - (uint8_t *)buf;
5958         if (len == 0) {
5959                 ctl_set_success(ctsio);
5960                 goto done;
5961         }
5962
5963         mtx_lock(&lun->lun_lock);
5964         ptrlen = (struct ctl_ptr_len_flags *)
5965             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5966         ptrlen->ptr = (void *)buf;
5967         ptrlen->len = len;
5968         ptrlen->flags = byte2;
5969         ctl_check_blocked(lun);
5970         mtx_unlock(&lun->lun_lock);
5971
5972         retval = lun->backend->config_write((union ctl_io *)ctsio);
5973         return (retval);
5974
5975 done:
5976         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5977                 free(ctsio->kern_data_ptr, M_CTL);
5978                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5979         }
5980         ctl_done((union ctl_io *)ctsio);
5981         return (CTL_RETVAL_COMPLETE);
5982 }
5983
5984 /*
5985  * Note that this function currently doesn't actually do anything inside
5986  * CTL to enforce things if the DQue bit is turned on.
5987  *
5988  * Also note that this function can't be used in the default case, because
5989  * the DQue bit isn't set in the changeable mask for the control mode page
5990  * anyway.  This is just here as an example for how to implement a page
5991  * handler, and a placeholder in case we want to allow the user to turn
5992  * tagged queueing on and off.
5993  *
5994  * The D_SENSE bit handling is functional, however, and will turn
5995  * descriptor sense on and off for a given LUN.
5996  */
5997 int
5998 ctl_control_page_handler(struct ctl_scsiio *ctsio,
5999                          struct ctl_page_index *page_index, uint8_t *page_ptr)
6000 {
6001         struct scsi_control_page *current_cp, *saved_cp, *user_cp;
6002         struct ctl_lun *lun;
6003         int set_ua;
6004         uint32_t initidx;
6005
6006         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6007         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6008         set_ua = 0;
6009
6010         user_cp = (struct scsi_control_page *)page_ptr;
6011         current_cp = (struct scsi_control_page *)
6012                 (page_index->page_data + (page_index->page_len *
6013                 CTL_PAGE_CURRENT));
6014         saved_cp = (struct scsi_control_page *)
6015                 (page_index->page_data + (page_index->page_len *
6016                 CTL_PAGE_SAVED));
6017
6018         mtx_lock(&lun->lun_lock);
6019         if (((current_cp->rlec & SCP_DSENSE) == 0)
6020          && ((user_cp->rlec & SCP_DSENSE) != 0)) {
6021                 /*
6022                  * Descriptor sense is currently turned off and the user
6023                  * wants to turn it on.
6024                  */
6025                 current_cp->rlec |= SCP_DSENSE;
6026                 saved_cp->rlec |= SCP_DSENSE;
6027                 lun->flags |= CTL_LUN_SENSE_DESC;
6028                 set_ua = 1;
6029         } else if (((current_cp->rlec & SCP_DSENSE) != 0)
6030                 && ((user_cp->rlec & SCP_DSENSE) == 0)) {
6031                 /*
6032                  * Descriptor sense is currently turned on, and the user
6033                  * wants to turn it off.
6034                  */
6035                 current_cp->rlec &= ~SCP_DSENSE;
6036                 saved_cp->rlec &= ~SCP_DSENSE;
6037                 lun->flags &= ~CTL_LUN_SENSE_DESC;
6038                 set_ua = 1;
6039         }
6040         if ((current_cp->queue_flags & SCP_QUEUE_ALG_MASK) !=
6041             (user_cp->queue_flags & SCP_QUEUE_ALG_MASK)) {
6042                 current_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
6043                 current_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
6044                 saved_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
6045                 saved_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
6046                 set_ua = 1;
6047         }
6048         if ((current_cp->eca_and_aen & SCP_SWP) !=
6049             (user_cp->eca_and_aen & SCP_SWP)) {
6050                 current_cp->eca_and_aen &= ~SCP_SWP;
6051                 current_cp->eca_and_aen |= user_cp->eca_and_aen & SCP_SWP;
6052                 saved_cp->eca_and_aen &= ~SCP_SWP;
6053                 saved_cp->eca_and_aen |= user_cp->eca_and_aen & SCP_SWP;
6054                 set_ua = 1;
6055         }
6056         if (set_ua != 0)
6057                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
6058         mtx_unlock(&lun->lun_lock);
6059         if (set_ua) {
6060                 ctl_isc_announce_mode(lun,
6061                     ctl_get_initindex(&ctsio->io_hdr.nexus),
6062                     page_index->page_code, page_index->subpage);
6063         }
6064         return (0);
6065 }
6066
6067 int
6068 ctl_caching_sp_handler(struct ctl_scsiio *ctsio,
6069                      struct ctl_page_index *page_index, uint8_t *page_ptr)
6070 {
6071         struct scsi_caching_page *current_cp, *saved_cp, *user_cp;
6072         struct ctl_lun *lun;
6073         int set_ua;
6074         uint32_t initidx;
6075
6076         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6077         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6078         set_ua = 0;
6079
6080         user_cp = (struct scsi_caching_page *)page_ptr;
6081         current_cp = (struct scsi_caching_page *)
6082                 (page_index->page_data + (page_index->page_len *
6083                 CTL_PAGE_CURRENT));
6084         saved_cp = (struct scsi_caching_page *)
6085                 (page_index->page_data + (page_index->page_len *
6086                 CTL_PAGE_SAVED));
6087
6088         mtx_lock(&lun->lun_lock);
6089         if ((current_cp->flags1 & (SCP_WCE | SCP_RCD)) !=
6090             (user_cp->flags1 & (SCP_WCE | SCP_RCD))) {
6091                 current_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
6092                 current_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
6093                 saved_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
6094                 saved_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
6095                 set_ua = 1;
6096         }
6097         if (set_ua != 0)
6098                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
6099         mtx_unlock(&lun->lun_lock);
6100         if (set_ua) {
6101                 ctl_isc_announce_mode(lun,
6102                     ctl_get_initindex(&ctsio->io_hdr.nexus),
6103                     page_index->page_code, page_index->subpage);
6104         }
6105         return (0);
6106 }
6107
6108 int
6109 ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio,
6110                                 struct ctl_page_index *page_index,
6111                                 uint8_t *page_ptr)
6112 {
6113         uint8_t *c;
6114         int i;
6115
6116         c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs;
6117         ctl_time_io_secs =
6118                 (c[0] << 8) |
6119                 (c[1] << 0) |
6120                 0;
6121         CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs));
6122         printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs);
6123         printf("page data:");
6124         for (i=0; i<8; i++)
6125                 printf(" %.2x",page_ptr[i]);
6126         printf("\n");
6127         return (0);
6128 }
6129
6130 int
6131 ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio,
6132                                struct ctl_page_index *page_index,
6133                                int pc)
6134 {
6135         struct copan_debugconf_subpage *page;
6136
6137         page = (struct copan_debugconf_subpage *)page_index->page_data +
6138                 (page_index->page_len * pc);
6139
6140         switch (pc) {
6141         case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6142         case SMS_PAGE_CTRL_DEFAULT >> 6:
6143         case SMS_PAGE_CTRL_SAVED >> 6:
6144                 /*
6145                  * We don't update the changable or default bits for this page.
6146                  */
6147                 break;
6148         case SMS_PAGE_CTRL_CURRENT >> 6:
6149                 page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8;
6150                 page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0;
6151                 break;
6152         default:
6153 #ifdef NEEDTOPORT
6154                 EPRINT(0, "Invalid PC %d!!", pc);
6155 #endif /* NEEDTOPORT */
6156                 break;
6157         }
6158         return (0);
6159 }
6160
6161
6162 static int
6163 ctl_do_mode_select(union ctl_io *io)
6164 {
6165         struct scsi_mode_page_header *page_header;
6166         struct ctl_page_index *page_index;
6167         struct ctl_scsiio *ctsio;
6168         int control_dev, page_len;
6169         int page_len_offset, page_len_size;
6170         union ctl_modepage_info *modepage_info;
6171         struct ctl_lun *lun;
6172         int *len_left, *len_used;
6173         int retval, i;
6174
6175         ctsio = &io->scsiio;
6176         page_index = NULL;
6177         page_len = 0;
6178         retval = CTL_RETVAL_COMPLETE;
6179
6180         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6181
6182         if (lun->be_lun->lun_type != T_DIRECT)
6183                 control_dev = 1;
6184         else
6185                 control_dev = 0;
6186
6187         modepage_info = (union ctl_modepage_info *)
6188                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6189         len_left = &modepage_info->header.len_left;
6190         len_used = &modepage_info->header.len_used;
6191
6192 do_next_page:
6193
6194         page_header = (struct scsi_mode_page_header *)
6195                 (ctsio->kern_data_ptr + *len_used);
6196
6197         if (*len_left == 0) {
6198                 free(ctsio->kern_data_ptr, M_CTL);
6199                 ctl_set_success(ctsio);
6200                 ctl_done((union ctl_io *)ctsio);
6201                 return (CTL_RETVAL_COMPLETE);
6202         } else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6203
6204                 free(ctsio->kern_data_ptr, M_CTL);
6205                 ctl_set_param_len_error(ctsio);
6206                 ctl_done((union ctl_io *)ctsio);
6207                 return (CTL_RETVAL_COMPLETE);
6208
6209         } else if ((page_header->page_code & SMPH_SPF)
6210                 && (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6211
6212                 free(ctsio->kern_data_ptr, M_CTL);
6213                 ctl_set_param_len_error(ctsio);
6214                 ctl_done((union ctl_io *)ctsio);
6215                 return (CTL_RETVAL_COMPLETE);
6216         }
6217
6218
6219         /*
6220          * XXX KDM should we do something with the block descriptor?
6221          */
6222         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6223
6224                 if ((control_dev != 0)
6225                  && (lun->mode_pages.index[i].page_flags &
6226                      CTL_PAGE_FLAG_DISK_ONLY))
6227                         continue;
6228
6229                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
6230                     (page_header->page_code & SMPH_PC_MASK))
6231                         continue;
6232
6233                 /*
6234                  * If neither page has a subpage code, then we've got a
6235                  * match.
6236                  */
6237                 if (((lun->mode_pages.index[i].page_code & SMPH_SPF) == 0)
6238                  && ((page_header->page_code & SMPH_SPF) == 0)) {
6239                         page_index = &lun->mode_pages.index[i];
6240                         page_len = page_header->page_length;
6241                         break;
6242                 }
6243
6244                 /*
6245                  * If both pages have subpages, then the subpage numbers
6246                  * have to match.
6247                  */
6248                 if ((lun->mode_pages.index[i].page_code & SMPH_SPF)
6249                   && (page_header->page_code & SMPH_SPF)) {
6250                         struct scsi_mode_page_header_sp *sph;
6251
6252                         sph = (struct scsi_mode_page_header_sp *)page_header;
6253
6254                         if (lun->mode_pages.index[i].subpage ==
6255                             sph->subpage) {
6256                                 page_index = &lun->mode_pages.index[i];
6257                                 page_len = scsi_2btoul(sph->page_length);
6258                                 break;
6259                         }
6260                 }
6261         }
6262
6263         /*
6264          * If we couldn't find the page, or if we don't have a mode select
6265          * handler for it, send back an error to the user.
6266          */
6267         if ((page_index == NULL)
6268          || (page_index->select_handler == NULL)) {
6269                 ctl_set_invalid_field(ctsio,
6270                                       /*sks_valid*/ 1,
6271                                       /*command*/ 0,
6272                                       /*field*/ *len_used,
6273                                       /*bit_valid*/ 0,
6274                                       /*bit*/ 0);
6275                 free(ctsio->kern_data_ptr, M_CTL);
6276                 ctl_done((union ctl_io *)ctsio);
6277                 return (CTL_RETVAL_COMPLETE);
6278         }
6279
6280         if (page_index->page_code & SMPH_SPF) {
6281                 page_len_offset = 2;
6282                 page_len_size = 2;
6283         } else {
6284                 page_len_size = 1;
6285                 page_len_offset = 1;
6286         }
6287
6288         /*
6289          * If the length the initiator gives us isn't the one we specify in
6290          * the mode page header, or if they didn't specify enough data in
6291          * the CDB to avoid truncating this page, kick out the request.
6292          */
6293         if ((page_len != (page_index->page_len - page_len_offset -
6294                           page_len_size))
6295          || (*len_left < page_index->page_len)) {
6296
6297
6298                 ctl_set_invalid_field(ctsio,
6299                                       /*sks_valid*/ 1,
6300                                       /*command*/ 0,
6301                                       /*field*/ *len_used + page_len_offset,
6302                                       /*bit_valid*/ 0,
6303                                       /*bit*/ 0);
6304                 free(ctsio->kern_data_ptr, M_CTL);
6305                 ctl_done((union ctl_io *)ctsio);
6306                 return (CTL_RETVAL_COMPLETE);
6307         }
6308
6309         /*
6310          * Run through the mode page, checking to make sure that the bits
6311          * the user changed are actually legal for him to change.
6312          */
6313         for (i = 0; i < page_index->page_len; i++) {
6314                 uint8_t *user_byte, *change_mask, *current_byte;
6315                 int bad_bit;
6316                 int j;
6317
6318                 user_byte = (uint8_t *)page_header + i;
6319                 change_mask = page_index->page_data +
6320                               (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6321                 current_byte = page_index->page_data +
6322                                (page_index->page_len * CTL_PAGE_CURRENT) + i;
6323
6324                 /*
6325                  * Check to see whether the user set any bits in this byte
6326                  * that he is not allowed to set.
6327                  */
6328                 if ((*user_byte & ~(*change_mask)) ==
6329                     (*current_byte & ~(*change_mask)))
6330                         continue;
6331
6332                 /*
6333                  * Go through bit by bit to determine which one is illegal.
6334                  */
6335                 bad_bit = 0;
6336                 for (j = 7; j >= 0; j--) {
6337                         if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6338                             (((1 << i) & ~(*change_mask)) & *current_byte)) {
6339                                 bad_bit = i;
6340                                 break;
6341                         }
6342                 }
6343                 ctl_set_invalid_field(ctsio,
6344                                       /*sks_valid*/ 1,
6345                                       /*command*/ 0,
6346                                       /*field*/ *len_used + i,
6347                                       /*bit_valid*/ 1,
6348                                       /*bit*/ bad_bit);
6349                 free(ctsio->kern_data_ptr, M_CTL);
6350                 ctl_done((union ctl_io *)ctsio);
6351                 return (CTL_RETVAL_COMPLETE);
6352         }
6353
6354         /*
6355          * Decrement these before we call the page handler, since we may
6356          * end up getting called back one way or another before the handler
6357          * returns to this context.
6358          */
6359         *len_left -= page_index->page_len;
6360         *len_used += page_index->page_len;
6361
6362         retval = page_index->select_handler(ctsio, page_index,
6363                                             (uint8_t *)page_header);
6364
6365         /*
6366          * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6367          * wait until this queued command completes to finish processing
6368          * the mode page.  If it returns anything other than
6369          * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6370          * already set the sense information, freed the data pointer, and
6371          * completed the io for us.
6372          */
6373         if (retval != CTL_RETVAL_COMPLETE)
6374                 goto bailout_no_done;
6375
6376         /*
6377          * If the initiator sent us more than one page, parse the next one.
6378          */
6379         if (*len_left > 0)
6380                 goto do_next_page;
6381
6382         ctl_set_success(ctsio);
6383         free(ctsio->kern_data_ptr, M_CTL);
6384         ctl_done((union ctl_io *)ctsio);
6385
6386 bailout_no_done:
6387
6388         return (CTL_RETVAL_COMPLETE);
6389
6390 }
6391
6392 int
6393 ctl_mode_select(struct ctl_scsiio *ctsio)
6394 {
6395         int param_len, pf, sp;
6396         int header_size, bd_len;
6397         int len_left, len_used;
6398         struct ctl_page_index *page_index;
6399         struct ctl_lun *lun;
6400         int control_dev, page_len;
6401         union ctl_modepage_info *modepage_info;
6402         int retval;
6403
6404         pf = 0;
6405         sp = 0;
6406         page_len = 0;
6407         len_used = 0;
6408         len_left = 0;
6409         retval = 0;
6410         bd_len = 0;
6411         page_index = NULL;
6412
6413         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6414
6415         if (lun->be_lun->lun_type != T_DIRECT)
6416                 control_dev = 1;
6417         else
6418                 control_dev = 0;
6419
6420         switch (ctsio->cdb[0]) {
6421         case MODE_SELECT_6: {
6422                 struct scsi_mode_select_6 *cdb;
6423
6424                 cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6425
6426                 pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6427                 sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6428
6429                 param_len = cdb->length;
6430                 header_size = sizeof(struct scsi_mode_header_6);
6431                 break;
6432         }
6433         case MODE_SELECT_10: {
6434                 struct scsi_mode_select_10 *cdb;
6435
6436                 cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6437
6438                 pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6439                 sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6440
6441                 param_len = scsi_2btoul(cdb->length);
6442                 header_size = sizeof(struct scsi_mode_header_10);
6443                 break;
6444         }
6445         default:
6446                 ctl_set_invalid_opcode(ctsio);
6447                 ctl_done((union ctl_io *)ctsio);
6448                 return (CTL_RETVAL_COMPLETE);
6449                 break; /* NOTREACHED */
6450         }
6451
6452         /*
6453          * From SPC-3:
6454          * "A parameter list length of zero indicates that the Data-Out Buffer
6455          * shall be empty. This condition shall not be considered as an error."
6456          */
6457         if (param_len == 0) {
6458                 ctl_set_success(ctsio);
6459                 ctl_done((union ctl_io *)ctsio);
6460                 return (CTL_RETVAL_COMPLETE);
6461         }
6462
6463         /*
6464          * Since we'll hit this the first time through, prior to
6465          * allocation, we don't need to free a data buffer here.
6466          */
6467         if (param_len < header_size) {
6468                 ctl_set_param_len_error(ctsio);
6469                 ctl_done((union ctl_io *)ctsio);
6470                 return (CTL_RETVAL_COMPLETE);
6471         }
6472
6473         /*
6474          * Allocate the data buffer and grab the user's data.  In theory,
6475          * we shouldn't have to sanity check the parameter list length here
6476          * because the maximum size is 64K.  We should be able to malloc
6477          * that much without too many problems.
6478          */
6479         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6480                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6481                 ctsio->kern_data_len = param_len;
6482                 ctsio->kern_total_len = param_len;
6483                 ctsio->kern_data_resid = 0;
6484                 ctsio->kern_rel_offset = 0;
6485                 ctsio->kern_sg_entries = 0;
6486                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6487                 ctsio->be_move_done = ctl_config_move_done;
6488                 ctl_datamove((union ctl_io *)ctsio);
6489
6490                 return (CTL_RETVAL_COMPLETE);
6491         }
6492
6493         switch (ctsio->cdb[0]) {
6494         case MODE_SELECT_6: {
6495                 struct scsi_mode_header_6 *mh6;
6496
6497                 mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6498                 bd_len = mh6->blk_desc_len;
6499                 break;
6500         }
6501         case MODE_SELECT_10: {
6502                 struct scsi_mode_header_10 *mh10;
6503
6504                 mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6505                 bd_len = scsi_2btoul(mh10->blk_desc_len);
6506                 break;
6507         }
6508         default:
6509                 panic("Invalid CDB type %#x", ctsio->cdb[0]);
6510                 break;
6511         }
6512
6513         if (param_len < (header_size + bd_len)) {
6514                 free(ctsio->kern_data_ptr, M_CTL);
6515                 ctl_set_param_len_error(ctsio);
6516                 ctl_done((union ctl_io *)ctsio);
6517                 return (CTL_RETVAL_COMPLETE);
6518         }
6519
6520         /*
6521          * Set the IO_CONT flag, so that if this I/O gets passed to
6522          * ctl_config_write_done(), it'll get passed back to
6523          * ctl_do_mode_select() for further processing, or completion if
6524          * we're all done.
6525          */
6526         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6527         ctsio->io_cont = ctl_do_mode_select;
6528
6529         modepage_info = (union ctl_modepage_info *)
6530                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6531
6532         memset(modepage_info, 0, sizeof(*modepage_info));
6533
6534         len_left = param_len - header_size - bd_len;
6535         len_used = header_size + bd_len;
6536
6537         modepage_info->header.len_left = len_left;
6538         modepage_info->header.len_used = len_used;
6539
6540         return (ctl_do_mode_select((union ctl_io *)ctsio));
6541 }
6542
6543 int
6544 ctl_mode_sense(struct ctl_scsiio *ctsio)
6545 {
6546         struct ctl_lun *lun;
6547         int pc, page_code, dbd, llba, subpage;
6548         int alloc_len, page_len, header_len, total_len;
6549         struct scsi_mode_block_descr *block_desc;
6550         struct ctl_page_index *page_index;
6551         int control_dev;
6552
6553         dbd = 0;
6554         llba = 0;
6555         block_desc = NULL;
6556         page_index = NULL;
6557
6558         CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6559
6560         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6561
6562         if (lun->be_lun->lun_type != T_DIRECT)
6563                 control_dev = 1;
6564         else
6565                 control_dev = 0;
6566
6567         switch (ctsio->cdb[0]) {
6568         case MODE_SENSE_6: {
6569                 struct scsi_mode_sense_6 *cdb;
6570
6571                 cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6572
6573                 header_len = sizeof(struct scsi_mode_hdr_6);
6574                 if (cdb->byte2 & SMS_DBD)
6575                         dbd = 1;
6576                 else
6577                         header_len += sizeof(struct scsi_mode_block_descr);
6578
6579                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6580                 page_code = cdb->page & SMS_PAGE_CODE;
6581                 subpage = cdb->subpage;
6582                 alloc_len = cdb->length;
6583                 break;
6584         }
6585         case MODE_SENSE_10: {
6586                 struct scsi_mode_sense_10 *cdb;
6587
6588                 cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6589
6590                 header_len = sizeof(struct scsi_mode_hdr_10);
6591
6592                 if (cdb->byte2 & SMS_DBD)
6593                         dbd = 1;
6594                 else
6595                         header_len += sizeof(struct scsi_mode_block_descr);
6596                 if (cdb->byte2 & SMS10_LLBAA)
6597                         llba = 1;
6598                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6599                 page_code = cdb->page & SMS_PAGE_CODE;
6600                 subpage = cdb->subpage;
6601                 alloc_len = scsi_2btoul(cdb->length);
6602                 break;
6603         }
6604         default:
6605                 ctl_set_invalid_opcode(ctsio);
6606                 ctl_done((union ctl_io *)ctsio);
6607                 return (CTL_RETVAL_COMPLETE);
6608                 break; /* NOTREACHED */
6609         }
6610
6611         /*
6612          * We have to make a first pass through to calculate the size of
6613          * the pages that match the user's query.  Then we allocate enough
6614          * memory to hold it, and actually copy the data into the buffer.
6615          */
6616         switch (page_code) {
6617         case SMS_ALL_PAGES_PAGE: {
6618                 int i;
6619
6620                 page_len = 0;
6621
6622                 /*
6623                  * At the moment, values other than 0 and 0xff here are
6624                  * reserved according to SPC-3.
6625                  */
6626                 if ((subpage != SMS_SUBPAGE_PAGE_0)
6627                  && (subpage != SMS_SUBPAGE_ALL)) {
6628                         ctl_set_invalid_field(ctsio,
6629                                               /*sks_valid*/ 1,
6630                                               /*command*/ 1,
6631                                               /*field*/ 3,
6632                                               /*bit_valid*/ 0,
6633                                               /*bit*/ 0);
6634                         ctl_done((union ctl_io *)ctsio);
6635                         return (CTL_RETVAL_COMPLETE);
6636                 }
6637
6638                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6639                         if ((control_dev != 0)
6640                          && (lun->mode_pages.index[i].page_flags &
6641                              CTL_PAGE_FLAG_DISK_ONLY))
6642                                 continue;
6643
6644                         /*
6645                          * We don't use this subpage if the user didn't
6646                          * request all subpages.
6647                          */
6648                         if ((lun->mode_pages.index[i].subpage != 0)
6649                          && (subpage == SMS_SUBPAGE_PAGE_0))
6650                                 continue;
6651
6652 #if 0
6653                         printf("found page %#x len %d\n",
6654                                lun->mode_pages.index[i].page_code &
6655                                SMPH_PC_MASK,
6656                                lun->mode_pages.index[i].page_len);
6657 #endif
6658                         page_len += lun->mode_pages.index[i].page_len;
6659                 }
6660                 break;
6661         }
6662         default: {
6663                 int i;
6664
6665                 page_len = 0;
6666
6667                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6668                         /* Look for the right page code */
6669                         if ((lun->mode_pages.index[i].page_code &
6670                              SMPH_PC_MASK) != page_code)
6671                                 continue;
6672
6673                         /* Look for the right subpage or the subpage wildcard*/
6674                         if ((lun->mode_pages.index[i].subpage != subpage)
6675                          && (subpage != SMS_SUBPAGE_ALL))
6676                                 continue;
6677
6678                         /* Make sure the page is supported for this dev type */
6679                         if ((control_dev != 0)
6680                          && (lun->mode_pages.index[i].page_flags &
6681                              CTL_PAGE_FLAG_DISK_ONLY))
6682                                 continue;
6683
6684 #if 0
6685                         printf("found page %#x len %d\n",
6686                                lun->mode_pages.index[i].page_code &
6687                                SMPH_PC_MASK,
6688                                lun->mode_pages.index[i].page_len);
6689 #endif
6690
6691                         page_len += lun->mode_pages.index[i].page_len;
6692                 }
6693
6694                 if (page_len == 0) {
6695                         ctl_set_invalid_field(ctsio,
6696                                               /*sks_valid*/ 1,
6697                                               /*command*/ 1,
6698                                               /*field*/ 2,
6699                                               /*bit_valid*/ 1,
6700                                               /*bit*/ 5);
6701                         ctl_done((union ctl_io *)ctsio);
6702                         return (CTL_RETVAL_COMPLETE);
6703                 }
6704                 break;
6705         }
6706         }
6707
6708         total_len = header_len + page_len;
6709 #if 0
6710         printf("header_len = %d, page_len = %d, total_len = %d\n",
6711                header_len, page_len, total_len);
6712 #endif
6713
6714         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6715         ctsio->kern_sg_entries = 0;
6716         ctsio->kern_data_resid = 0;
6717         ctsio->kern_rel_offset = 0;
6718         if (total_len < alloc_len) {
6719                 ctsio->residual = alloc_len - total_len;
6720                 ctsio->kern_data_len = total_len;
6721                 ctsio->kern_total_len = total_len;
6722         } else {
6723                 ctsio->residual = 0;
6724                 ctsio->kern_data_len = alloc_len;
6725                 ctsio->kern_total_len = alloc_len;
6726         }
6727
6728         switch (ctsio->cdb[0]) {
6729         case MODE_SENSE_6: {
6730                 struct scsi_mode_hdr_6 *header;
6731
6732                 header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6733
6734                 header->datalen = MIN(total_len - 1, 254);
6735                 if (control_dev == 0) {
6736                         header->dev_specific = 0x10; /* DPOFUA */
6737                         if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6738                             (lun->mode_pages.control_page[CTL_PAGE_CURRENT]
6739                             .eca_and_aen & SCP_SWP) != 0)
6740                                     header->dev_specific |= 0x80; /* WP */
6741                 }
6742                 if (dbd)
6743                         header->block_descr_len = 0;
6744                 else
6745                         header->block_descr_len =
6746                                 sizeof(struct scsi_mode_block_descr);
6747                 block_desc = (struct scsi_mode_block_descr *)&header[1];
6748                 break;
6749         }
6750         case MODE_SENSE_10: {
6751                 struct scsi_mode_hdr_10 *header;
6752                 int datalen;
6753
6754                 header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6755
6756                 datalen = MIN(total_len - 2, 65533);
6757                 scsi_ulto2b(datalen, header->datalen);
6758                 if (control_dev == 0) {
6759                         header->dev_specific = 0x10; /* DPOFUA */
6760                         if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6761                             (lun->mode_pages.control_page[CTL_PAGE_CURRENT]
6762                             .eca_and_aen & SCP_SWP) != 0)
6763                                     header->dev_specific |= 0x80; /* WP */
6764                 }
6765                 if (dbd)
6766                         scsi_ulto2b(0, header->block_descr_len);
6767                 else
6768                         scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
6769                                     header->block_descr_len);
6770                 block_desc = (struct scsi_mode_block_descr *)&header[1];
6771                 break;
6772         }
6773         default:
6774                 panic("invalid CDB type %#x", ctsio->cdb[0]);
6775                 break; /* NOTREACHED */
6776         }
6777
6778         /*
6779          * If we've got a disk, use its blocksize in the block
6780          * descriptor.  Otherwise, just set it to 0.
6781          */
6782         if (dbd == 0) {
6783                 if (control_dev == 0)
6784                         scsi_ulto3b(lun->be_lun->blocksize,
6785                                     block_desc->block_len);
6786                 else
6787                         scsi_ulto3b(0, block_desc->block_len);
6788         }
6789
6790         switch (page_code) {
6791         case SMS_ALL_PAGES_PAGE: {
6792                 int i, data_used;
6793
6794                 data_used = header_len;
6795                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6796                         struct ctl_page_index *page_index;
6797
6798                         page_index = &lun->mode_pages.index[i];
6799
6800                         if ((control_dev != 0)
6801                          && (page_index->page_flags &
6802                             CTL_PAGE_FLAG_DISK_ONLY))
6803                                 continue;
6804
6805                         /*
6806                          * We don't use this subpage if the user didn't
6807                          * request all subpages.  We already checked (above)
6808                          * to make sure the user only specified a subpage
6809                          * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
6810                          */
6811                         if ((page_index->subpage != 0)
6812                          && (subpage == SMS_SUBPAGE_PAGE_0))
6813                                 continue;
6814
6815                         /*
6816                          * Call the handler, if it exists, to update the
6817                          * page to the latest values.
6818                          */
6819                         if (page_index->sense_handler != NULL)
6820                                 page_index->sense_handler(ctsio, page_index,pc);
6821
6822                         memcpy(ctsio->kern_data_ptr + data_used,
6823                                page_index->page_data +
6824                                (page_index->page_len * pc),
6825                                page_index->page_len);
6826                         data_used += page_index->page_len;
6827                 }
6828                 break;
6829         }
6830         default: {
6831                 int i, data_used;
6832
6833                 data_used = header_len;
6834
6835                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6836                         struct ctl_page_index *page_index;
6837
6838                         page_index = &lun->mode_pages.index[i];
6839
6840                         /* Look for the right page code */
6841                         if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6842                                 continue;
6843
6844                         /* Look for the right subpage or the subpage wildcard*/
6845                         if ((page_index->subpage != subpage)
6846                          && (subpage != SMS_SUBPAGE_ALL))
6847                                 continue;
6848
6849                         /* Make sure the page is supported for this dev type */
6850                         if ((control_dev != 0)
6851                          && (page_index->page_flags &
6852                              CTL_PAGE_FLAG_DISK_ONLY))
6853                                 continue;
6854
6855                         /*
6856                          * Call the handler, if it exists, to update the
6857                          * page to the latest values.
6858                          */
6859                         if (page_index->sense_handler != NULL)
6860                                 page_index->sense_handler(ctsio, page_index,pc);
6861
6862                         memcpy(ctsio->kern_data_ptr + data_used,
6863                                page_index->page_data +
6864                                (page_index->page_len * pc),
6865                                page_index->page_len);
6866                         data_used += page_index->page_len;
6867                 }
6868                 break;
6869         }
6870         }
6871
6872         ctl_set_success(ctsio);
6873         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6874         ctsio->be_move_done = ctl_config_move_done;
6875         ctl_datamove((union ctl_io *)ctsio);
6876         return (CTL_RETVAL_COMPLETE);
6877 }
6878
6879 int
6880 ctl_lbp_log_sense_handler(struct ctl_scsiio *ctsio,
6881                                struct ctl_page_index *page_index,
6882                                int pc)
6883 {
6884         struct ctl_lun *lun;
6885         struct scsi_log_param_header *phdr;
6886         uint8_t *data;
6887         uint64_t val;
6888
6889         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6890         data = page_index->page_data;
6891
6892         if (lun->backend->lun_attr != NULL &&
6893             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksavail"))
6894              != UINT64_MAX) {
6895                 phdr = (struct scsi_log_param_header *)data;
6896                 scsi_ulto2b(0x0001, phdr->param_code);
6897                 phdr->param_control = SLP_LBIN | SLP_LP;
6898                 phdr->param_len = 8;
6899                 data = (uint8_t *)(phdr + 1);
6900                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6901                 data[4] = 0x02; /* per-pool */
6902                 data += phdr->param_len;
6903         }
6904
6905         if (lun->backend->lun_attr != NULL &&
6906             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksused"))
6907              != UINT64_MAX) {
6908                 phdr = (struct scsi_log_param_header *)data;
6909                 scsi_ulto2b(0x0002, phdr->param_code);
6910                 phdr->param_control = SLP_LBIN | SLP_LP;
6911                 phdr->param_len = 8;
6912                 data = (uint8_t *)(phdr + 1);
6913                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6914                 data[4] = 0x01; /* per-LUN */
6915                 data += phdr->param_len;
6916         }
6917
6918         if (lun->backend->lun_attr != NULL &&
6919             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksavail"))
6920              != UINT64_MAX) {
6921                 phdr = (struct scsi_log_param_header *)data;
6922                 scsi_ulto2b(0x00f1, phdr->param_code);
6923                 phdr->param_control = SLP_LBIN | SLP_LP;
6924                 phdr->param_len = 8;
6925                 data = (uint8_t *)(phdr + 1);
6926                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6927                 data[4] = 0x02; /* per-pool */
6928                 data += phdr->param_len;
6929         }
6930
6931         if (lun->backend->lun_attr != NULL &&
6932             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksused"))
6933              != UINT64_MAX) {
6934                 phdr = (struct scsi_log_param_header *)data;
6935                 scsi_ulto2b(0x00f2, phdr->param_code);
6936                 phdr->param_control = SLP_LBIN | SLP_LP;
6937                 phdr->param_len = 8;
6938                 data = (uint8_t *)(phdr + 1);
6939                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6940                 data[4] = 0x02; /* per-pool */
6941                 data += phdr->param_len;
6942         }
6943
6944         page_index->page_len = data - page_index->page_data;
6945         return (0);
6946 }
6947
6948 int
6949 ctl_sap_log_sense_handler(struct ctl_scsiio *ctsio,
6950                                struct ctl_page_index *page_index,
6951                                int pc)
6952 {
6953         struct ctl_lun *lun;
6954         struct stat_page *data;
6955         uint64_t rn, wn, rb, wb;
6956         struct bintime rt, wt;
6957         int i;
6958
6959         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6960         data = (struct stat_page *)page_index->page_data;
6961
6962         scsi_ulto2b(SLP_SAP, data->sap.hdr.param_code);
6963         data->sap.hdr.param_control = SLP_LBIN;
6964         data->sap.hdr.param_len = sizeof(struct scsi_log_stat_and_perf) -
6965             sizeof(struct scsi_log_param_header);
6966         rn = wn = rb = wb = 0;
6967         bintime_clear(&rt);
6968         bintime_clear(&wt);
6969         for (i = 0; i < CTL_MAX_PORTS; i++) {
6970                 rn += lun->stats.ports[i].operations[CTL_STATS_READ];
6971                 wn += lun->stats.ports[i].operations[CTL_STATS_WRITE];
6972                 rb += lun->stats.ports[i].bytes[CTL_STATS_READ];
6973                 wb += lun->stats.ports[i].bytes[CTL_STATS_WRITE];
6974                 bintime_add(&rt, &lun->stats.ports[i].time[CTL_STATS_READ]);
6975                 bintime_add(&wt, &lun->stats.ports[i].time[CTL_STATS_WRITE]);
6976         }
6977         scsi_u64to8b(rn, data->sap.read_num);
6978         scsi_u64to8b(wn, data->sap.write_num);
6979         if (lun->stats.blocksize > 0) {
6980                 scsi_u64to8b(wb / lun->stats.blocksize,
6981                     data->sap.recvieved_lba);
6982                 scsi_u64to8b(rb / lun->stats.blocksize,
6983                     data->sap.transmitted_lba);
6984         }
6985         scsi_u64to8b((uint64_t)rt.sec * 1000 + rt.frac / (UINT64_MAX / 1000),
6986             data->sap.read_int);
6987         scsi_u64to8b((uint64_t)wt.sec * 1000 + wt.frac / (UINT64_MAX / 1000),
6988             data->sap.write_int);
6989         scsi_u64to8b(0, data->sap.weighted_num);
6990         scsi_u64to8b(0, data->sap.weighted_int);
6991         scsi_ulto2b(SLP_IT, data->it.hdr.param_code);
6992         data->it.hdr.param_control = SLP_LBIN;
6993         data->it.hdr.param_len = sizeof(struct scsi_log_idle_time) -
6994             sizeof(struct scsi_log_param_header);
6995 #ifdef CTL_TIME_IO
6996         scsi_u64to8b(lun->idle_time / SBT_1MS, data->it.idle_int);
6997 #endif
6998         scsi_ulto2b(SLP_TI, data->ti.hdr.param_code);
6999         data->it.hdr.param_control = SLP_LBIN;
7000         data->ti.hdr.param_len = sizeof(struct scsi_log_time_interval) -
7001             sizeof(struct scsi_log_param_header);
7002         scsi_ulto4b(3, data->ti.exponent);
7003         scsi_ulto4b(1, data->ti.integer);
7004
7005         page_index->page_len = sizeof(*data);
7006         return (0);
7007 }
7008
7009 int
7010 ctl_log_sense(struct ctl_scsiio *ctsio)
7011 {
7012         struct ctl_lun *lun;
7013         int i, pc, page_code, subpage;
7014         int alloc_len, total_len;
7015         struct ctl_page_index *page_index;
7016         struct scsi_log_sense *cdb;
7017         struct scsi_log_header *header;
7018
7019         CTL_DEBUG_PRINT(("ctl_log_sense\n"));
7020
7021         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7022         cdb = (struct scsi_log_sense *)ctsio->cdb;
7023         pc = (cdb->page & SLS_PAGE_CTRL_MASK) >> 6;
7024         page_code = cdb->page & SLS_PAGE_CODE;
7025         subpage = cdb->subpage;
7026         alloc_len = scsi_2btoul(cdb->length);
7027
7028         page_index = NULL;
7029         for (i = 0; i < CTL_NUM_LOG_PAGES; i++) {
7030                 page_index = &lun->log_pages.index[i];
7031
7032                 /* Look for the right page code */
7033                 if ((page_index->page_code & SL_PAGE_CODE) != page_code)
7034                         continue;
7035
7036                 /* Look for the right subpage or the subpage wildcard*/
7037                 if (page_index->subpage != subpage)
7038                         continue;
7039
7040                 break;
7041         }
7042         if (i >= CTL_NUM_LOG_PAGES) {
7043                 ctl_set_invalid_field(ctsio,
7044                                       /*sks_valid*/ 1,
7045                                       /*command*/ 1,
7046                                       /*field*/ 2,
7047                                       /*bit_valid*/ 0,
7048                                       /*bit*/ 0);
7049                 ctl_done((union ctl_io *)ctsio);
7050                 return (CTL_RETVAL_COMPLETE);
7051         }
7052
7053         total_len = sizeof(struct scsi_log_header) + page_index->page_len;
7054
7055         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7056         ctsio->kern_sg_entries = 0;
7057         ctsio->kern_data_resid = 0;
7058         ctsio->kern_rel_offset = 0;
7059         if (total_len < alloc_len) {
7060                 ctsio->residual = alloc_len - total_len;
7061                 ctsio->kern_data_len = total_len;
7062                 ctsio->kern_total_len = total_len;
7063         } else {
7064                 ctsio->residual = 0;
7065                 ctsio->kern_data_len = alloc_len;
7066                 ctsio->kern_total_len = alloc_len;
7067         }
7068
7069         header = (struct scsi_log_header *)ctsio->kern_data_ptr;
7070         header->page = page_index->page_code;
7071         if (page_index->subpage) {
7072                 header->page |= SL_SPF;
7073                 header->subpage = page_index->subpage;
7074         }
7075         scsi_ulto2b(page_index->page_len, header->datalen);
7076
7077         /*
7078          * Call the handler, if it exists, to update the
7079          * page to the latest values.
7080          */
7081         if (page_index->sense_handler != NULL)
7082                 page_index->sense_handler(ctsio, page_index, pc);
7083
7084         memcpy(header + 1, page_index->page_data, page_index->page_len);
7085
7086         ctl_set_success(ctsio);
7087         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7088         ctsio->be_move_done = ctl_config_move_done;
7089         ctl_datamove((union ctl_io *)ctsio);
7090         return (CTL_RETVAL_COMPLETE);
7091 }
7092
7093 int
7094 ctl_read_capacity(struct ctl_scsiio *ctsio)
7095 {
7096         struct scsi_read_capacity *cdb;
7097         struct scsi_read_capacity_data *data;
7098         struct ctl_lun *lun;
7099         uint32_t lba;
7100
7101         CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
7102
7103         cdb = (struct scsi_read_capacity *)ctsio->cdb;
7104
7105         lba = scsi_4btoul(cdb->addr);
7106         if (((cdb->pmi & SRC_PMI) == 0)
7107          && (lba != 0)) {
7108                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7109                                       /*sks_valid*/ 1,
7110                                       /*command*/ 1,
7111                                       /*field*/ 2,
7112                                       /*bit_valid*/ 0,
7113                                       /*bit*/ 0);
7114                 ctl_done((union ctl_io *)ctsio);
7115                 return (CTL_RETVAL_COMPLETE);
7116         }
7117
7118         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7119
7120         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7121         data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
7122         ctsio->residual = 0;
7123         ctsio->kern_data_len = sizeof(*data);
7124         ctsio->kern_total_len = sizeof(*data);
7125         ctsio->kern_data_resid = 0;
7126         ctsio->kern_rel_offset = 0;
7127         ctsio->kern_sg_entries = 0;
7128
7129         /*
7130          * If the maximum LBA is greater than 0xfffffffe, the user must
7131          * issue a SERVICE ACTION IN (16) command, with the read capacity
7132          * serivce action set.
7133          */
7134         if (lun->be_lun->maxlba > 0xfffffffe)
7135                 scsi_ulto4b(0xffffffff, data->addr);
7136         else
7137                 scsi_ulto4b(lun->be_lun->maxlba, data->addr);
7138
7139         /*
7140          * XXX KDM this may not be 512 bytes...
7141          */
7142         scsi_ulto4b(lun->be_lun->blocksize, data->length);
7143
7144         ctl_set_success(ctsio);
7145         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7146         ctsio->be_move_done = ctl_config_move_done;
7147         ctl_datamove((union ctl_io *)ctsio);
7148         return (CTL_RETVAL_COMPLETE);
7149 }
7150
7151 int
7152 ctl_read_capacity_16(struct ctl_scsiio *ctsio)
7153 {
7154         struct scsi_read_capacity_16 *cdb;
7155         struct scsi_read_capacity_data_long *data;
7156         struct ctl_lun *lun;
7157         uint64_t lba;
7158         uint32_t alloc_len;
7159
7160         CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
7161
7162         cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
7163
7164         alloc_len = scsi_4btoul(cdb->alloc_len);
7165         lba = scsi_8btou64(cdb->addr);
7166
7167         if ((cdb->reladr & SRC16_PMI)
7168          && (lba != 0)) {
7169                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7170                                       /*sks_valid*/ 1,
7171                                       /*command*/ 1,
7172                                       /*field*/ 2,
7173                                       /*bit_valid*/ 0,
7174                                       /*bit*/ 0);
7175                 ctl_done((union ctl_io *)ctsio);
7176                 return (CTL_RETVAL_COMPLETE);
7177         }
7178
7179         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7180
7181         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7182         data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
7183
7184         if (sizeof(*data) < alloc_len) {
7185                 ctsio->residual = alloc_len - sizeof(*data);
7186                 ctsio->kern_data_len = sizeof(*data);
7187                 ctsio->kern_total_len = sizeof(*data);
7188         } else {
7189                 ctsio->residual = 0;
7190                 ctsio->kern_data_len = alloc_len;
7191                 ctsio->kern_total_len = alloc_len;
7192         }
7193         ctsio->kern_data_resid = 0;
7194         ctsio->kern_rel_offset = 0;
7195         ctsio->kern_sg_entries = 0;
7196
7197         scsi_u64to8b(lun->be_lun->maxlba, data->addr);
7198         /* XXX KDM this may not be 512 bytes... */
7199         scsi_ulto4b(lun->be_lun->blocksize, data->length);
7200         data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
7201         scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
7202         if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
7203                 data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ;
7204
7205         ctl_set_success(ctsio);
7206         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7207         ctsio->be_move_done = ctl_config_move_done;
7208         ctl_datamove((union ctl_io *)ctsio);
7209         return (CTL_RETVAL_COMPLETE);
7210 }
7211
7212 int
7213 ctl_get_lba_status(struct ctl_scsiio *ctsio)
7214 {
7215         struct scsi_get_lba_status *cdb;
7216         struct scsi_get_lba_status_data *data;
7217         struct ctl_lun *lun;
7218         struct ctl_lba_len_flags *lbalen;
7219         uint64_t lba;
7220         uint32_t alloc_len, total_len;
7221         int retval;
7222
7223         CTL_DEBUG_PRINT(("ctl_get_lba_status\n"));
7224
7225         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7226         cdb = (struct scsi_get_lba_status *)ctsio->cdb;
7227         lba = scsi_8btou64(cdb->addr);
7228         alloc_len = scsi_4btoul(cdb->alloc_len);
7229
7230         if (lba > lun->be_lun->maxlba) {
7231                 ctl_set_lba_out_of_range(ctsio);
7232                 ctl_done((union ctl_io *)ctsio);
7233                 return (CTL_RETVAL_COMPLETE);
7234         }
7235
7236         total_len = sizeof(*data) + sizeof(data->descr[0]);
7237         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7238         data = (struct scsi_get_lba_status_data *)ctsio->kern_data_ptr;
7239
7240         if (total_len < alloc_len) {
7241                 ctsio->residual = alloc_len - total_len;
7242                 ctsio->kern_data_len = total_len;
7243                 ctsio->kern_total_len = total_len;
7244         } else {
7245                 ctsio->residual = 0;
7246                 ctsio->kern_data_len = alloc_len;
7247                 ctsio->kern_total_len = alloc_len;
7248         }
7249         ctsio->kern_data_resid = 0;
7250         ctsio->kern_rel_offset = 0;
7251         ctsio->kern_sg_entries = 0;
7252
7253         /* Fill dummy data in case backend can't tell anything. */
7254         scsi_ulto4b(4 + sizeof(data->descr[0]), data->length);
7255         scsi_u64to8b(lba, data->descr[0].addr);
7256         scsi_ulto4b(MIN(UINT32_MAX, lun->be_lun->maxlba + 1 - lba),
7257             data->descr[0].length);
7258         data->descr[0].status = 0; /* Mapped or unknown. */
7259
7260         ctl_set_success(ctsio);
7261         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7262         ctsio->be_move_done = ctl_config_move_done;
7263
7264         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
7265         lbalen->lba = lba;
7266         lbalen->len = total_len;
7267         lbalen->flags = 0;
7268         retval = lun->backend->config_read((union ctl_io *)ctsio);
7269         return (CTL_RETVAL_COMPLETE);
7270 }
7271
7272 int
7273 ctl_read_defect(struct ctl_scsiio *ctsio)
7274 {
7275         struct scsi_read_defect_data_10 *ccb10;
7276         struct scsi_read_defect_data_12 *ccb12;
7277         struct scsi_read_defect_data_hdr_10 *data10;
7278         struct scsi_read_defect_data_hdr_12 *data12;
7279         uint32_t alloc_len, data_len;
7280         uint8_t format;
7281
7282         CTL_DEBUG_PRINT(("ctl_read_defect\n"));
7283
7284         if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7285                 ccb10 = (struct scsi_read_defect_data_10 *)&ctsio->cdb;
7286                 format = ccb10->format;
7287                 alloc_len = scsi_2btoul(ccb10->alloc_length);
7288                 data_len = sizeof(*data10);
7289         } else {
7290                 ccb12 = (struct scsi_read_defect_data_12 *)&ctsio->cdb;
7291                 format = ccb12->format;
7292                 alloc_len = scsi_4btoul(ccb12->alloc_length);
7293                 data_len = sizeof(*data12);
7294         }
7295         if (alloc_len == 0) {
7296                 ctl_set_success(ctsio);
7297                 ctl_done((union ctl_io *)ctsio);
7298                 return (CTL_RETVAL_COMPLETE);
7299         }
7300
7301         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
7302         if (data_len < alloc_len) {
7303                 ctsio->residual = alloc_len - data_len;
7304                 ctsio->kern_data_len = data_len;
7305                 ctsio->kern_total_len = data_len;
7306         } else {
7307                 ctsio->residual = 0;
7308                 ctsio->kern_data_len = alloc_len;
7309                 ctsio->kern_total_len = alloc_len;
7310         }
7311         ctsio->kern_data_resid = 0;
7312         ctsio->kern_rel_offset = 0;
7313         ctsio->kern_sg_entries = 0;
7314
7315         if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7316                 data10 = (struct scsi_read_defect_data_hdr_10 *)
7317                     ctsio->kern_data_ptr;
7318                 data10->format = format;
7319                 scsi_ulto2b(0, data10->length);
7320         } else {
7321                 data12 = (struct scsi_read_defect_data_hdr_12 *)
7322                     ctsio->kern_data_ptr;
7323                 data12->format = format;
7324                 scsi_ulto2b(0, data12->generation);
7325                 scsi_ulto4b(0, data12->length);
7326         }
7327
7328         ctl_set_success(ctsio);
7329         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7330         ctsio->be_move_done = ctl_config_move_done;
7331         ctl_datamove((union ctl_io *)ctsio);
7332         return (CTL_RETVAL_COMPLETE);
7333 }
7334
7335 int
7336 ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7337 {
7338         struct scsi_maintenance_in *cdb;
7339         int retval;
7340         int alloc_len, ext, total_len = 0, g, pc, pg, gs, os;
7341         int num_target_port_groups, num_target_ports;
7342         struct ctl_lun *lun;
7343         struct ctl_softc *softc;
7344         struct ctl_port *port;
7345         struct scsi_target_group_data *rtg_ptr;
7346         struct scsi_target_group_data_extended *rtg_ext_ptr;
7347         struct scsi_target_port_group_descriptor *tpg_desc;
7348
7349         CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7350
7351         cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7352         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7353         softc = lun->ctl_softc;
7354
7355         retval = CTL_RETVAL_COMPLETE;
7356
7357         switch (cdb->byte2 & STG_PDF_MASK) {
7358         case STG_PDF_LENGTH:
7359                 ext = 0;
7360                 break;
7361         case STG_PDF_EXTENDED:
7362                 ext = 1;
7363                 break;
7364         default:
7365                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7366                                       /*sks_valid*/ 1,
7367                                       /*command*/ 1,
7368                                       /*field*/ 2,
7369                                       /*bit_valid*/ 1,
7370                                       /*bit*/ 5);
7371                 ctl_done((union ctl_io *)ctsio);
7372                 return(retval);
7373         }
7374
7375         if (softc->is_single)
7376                 num_target_port_groups = 1;
7377         else
7378                 num_target_port_groups = NUM_TARGET_PORT_GROUPS;
7379         num_target_ports = 0;
7380         mtx_lock(&softc->ctl_lock);
7381         STAILQ_FOREACH(port, &softc->port_list, links) {
7382                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7383                         continue;
7384                 if (ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
7385                         continue;
7386                 num_target_ports++;
7387         }
7388         mtx_unlock(&softc->ctl_lock);
7389
7390         if (ext)
7391                 total_len = sizeof(struct scsi_target_group_data_extended);
7392         else
7393                 total_len = sizeof(struct scsi_target_group_data);
7394         total_len += sizeof(struct scsi_target_port_group_descriptor) *
7395                 num_target_port_groups +
7396             sizeof(struct scsi_target_port_descriptor) * num_target_ports;
7397
7398         alloc_len = scsi_4btoul(cdb->length);
7399
7400         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7401
7402         ctsio->kern_sg_entries = 0;
7403
7404         if (total_len < alloc_len) {
7405                 ctsio->residual = alloc_len - total_len;
7406                 ctsio->kern_data_len = total_len;
7407                 ctsio->kern_total_len = total_len;
7408         } else {
7409                 ctsio->residual = 0;
7410                 ctsio->kern_data_len = alloc_len;
7411                 ctsio->kern_total_len = alloc_len;
7412         }
7413         ctsio->kern_data_resid = 0;
7414         ctsio->kern_rel_offset = 0;
7415
7416         if (ext) {
7417                 rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7418                     ctsio->kern_data_ptr;
7419                 scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7420                 rtg_ext_ptr->format_type = 0x10;
7421                 rtg_ext_ptr->implicit_transition_time = 0;
7422                 tpg_desc = &rtg_ext_ptr->groups[0];
7423         } else {
7424                 rtg_ptr = (struct scsi_target_group_data *)
7425                     ctsio->kern_data_ptr;
7426                 scsi_ulto4b(total_len - 4, rtg_ptr->length);
7427                 tpg_desc = &rtg_ptr->groups[0];
7428         }
7429
7430         mtx_lock(&softc->ctl_lock);
7431         pg = softc->port_min / softc->port_cnt;
7432         if (softc->ha_link == CTL_HA_LINK_OFFLINE)
7433                 gs = TPG_ASYMMETRIC_ACCESS_UNAVAILABLE;
7434         else if (softc->ha_link == CTL_HA_LINK_UNKNOWN)
7435                 gs = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7436         else if (softc->ha_mode == CTL_HA_MODE_ACT_STBY)
7437                 gs = TPG_ASYMMETRIC_ACCESS_STANDBY;
7438         else
7439                 gs = TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7440         if (lun->flags & CTL_LUN_PRIMARY_SC) {
7441                 os = gs;
7442                 gs = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7443         } else
7444                 os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7445         for (g = 0; g < num_target_port_groups; g++) {
7446                 tpg_desc->pref_state = (g == pg) ? gs : os;
7447                 tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP |
7448                     TPG_U_SUP | TPG_T_SUP;
7449                 scsi_ulto2b(g + 1, tpg_desc->target_port_group);
7450                 tpg_desc->status = TPG_IMPLICIT;
7451                 pc = 0;
7452                 STAILQ_FOREACH(port, &softc->port_list, links) {
7453                         if (port->targ_port < g * softc->port_cnt ||
7454                             port->targ_port >= (g + 1) * softc->port_cnt)
7455                                 continue;
7456                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7457                                 continue;
7458                         if (ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
7459                                 continue;
7460                         scsi_ulto2b(port->targ_port, tpg_desc->descriptors[pc].
7461                             relative_target_port_identifier);
7462                         pc++;
7463                 }
7464                 tpg_desc->target_port_count = pc;
7465                 tpg_desc = (struct scsi_target_port_group_descriptor *)
7466                     &tpg_desc->descriptors[pc];
7467         }
7468         mtx_unlock(&softc->ctl_lock);
7469
7470         ctl_set_success(ctsio);
7471         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7472         ctsio->be_move_done = ctl_config_move_done;
7473         ctl_datamove((union ctl_io *)ctsio);
7474         return(retval);
7475 }
7476
7477 int
7478 ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7479 {
7480         struct ctl_lun *lun;
7481         struct scsi_report_supported_opcodes *cdb;
7482         const struct ctl_cmd_entry *entry, *sentry;
7483         struct scsi_report_supported_opcodes_all *all;
7484         struct scsi_report_supported_opcodes_descr *descr;
7485         struct scsi_report_supported_opcodes_one *one;
7486         int retval;
7487         int alloc_len, total_len;
7488         int opcode, service_action, i, j, num;
7489
7490         CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7491
7492         cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7493         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7494
7495         retval = CTL_RETVAL_COMPLETE;
7496
7497         opcode = cdb->requested_opcode;
7498         service_action = scsi_2btoul(cdb->requested_service_action);
7499         switch (cdb->options & RSO_OPTIONS_MASK) {
7500         case RSO_OPTIONS_ALL:
7501                 num = 0;
7502                 for (i = 0; i < 256; i++) {
7503                         entry = &ctl_cmd_table[i];
7504                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7505                                 for (j = 0; j < 32; j++) {
7506                                         sentry = &((const struct ctl_cmd_entry *)
7507                                             entry->execute)[j];
7508                                         if (ctl_cmd_applicable(
7509                                             lun->be_lun->lun_type, sentry))
7510                                                 num++;
7511                                 }
7512                         } else {
7513                                 if (ctl_cmd_applicable(lun->be_lun->lun_type,
7514                                     entry))
7515                                         num++;
7516                         }
7517                 }
7518                 total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7519                     num * sizeof(struct scsi_report_supported_opcodes_descr);
7520                 break;
7521         case RSO_OPTIONS_OC:
7522                 if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7523                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7524                                               /*sks_valid*/ 1,
7525                                               /*command*/ 1,
7526                                               /*field*/ 2,
7527                                               /*bit_valid*/ 1,
7528                                               /*bit*/ 2);
7529                         ctl_done((union ctl_io *)ctsio);
7530                         return (CTL_RETVAL_COMPLETE);
7531                 }
7532                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7533                 break;
7534         case RSO_OPTIONS_OC_SA:
7535                 if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7536                     service_action >= 32) {
7537                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7538                                               /*sks_valid*/ 1,
7539                                               /*command*/ 1,
7540                                               /*field*/ 2,
7541                                               /*bit_valid*/ 1,
7542                                               /*bit*/ 2);
7543                         ctl_done((union ctl_io *)ctsio);
7544                         return (CTL_RETVAL_COMPLETE);
7545                 }
7546                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7547                 break;
7548         default:
7549                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7550                                       /*sks_valid*/ 1,
7551                                       /*command*/ 1,
7552                                       /*field*/ 2,
7553                                       /*bit_valid*/ 1,
7554                                       /*bit*/ 2);
7555                 ctl_done((union ctl_io *)ctsio);
7556                 return (CTL_RETVAL_COMPLETE);
7557         }
7558
7559         alloc_len = scsi_4btoul(cdb->length);
7560
7561         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7562
7563         ctsio->kern_sg_entries = 0;
7564
7565         if (total_len < alloc_len) {
7566                 ctsio->residual = alloc_len - total_len;
7567                 ctsio->kern_data_len = total_len;
7568                 ctsio->kern_total_len = total_len;
7569         } else {
7570                 ctsio->residual = 0;
7571                 ctsio->kern_data_len = alloc_len;
7572                 ctsio->kern_total_len = alloc_len;
7573         }
7574         ctsio->kern_data_resid = 0;
7575         ctsio->kern_rel_offset = 0;
7576
7577         switch (cdb->options & RSO_OPTIONS_MASK) {
7578         case RSO_OPTIONS_ALL:
7579                 all = (struct scsi_report_supported_opcodes_all *)
7580                     ctsio->kern_data_ptr;
7581                 num = 0;
7582                 for (i = 0; i < 256; i++) {
7583                         entry = &ctl_cmd_table[i];
7584                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7585                                 for (j = 0; j < 32; j++) {
7586                                         sentry = &((const struct ctl_cmd_entry *)
7587                                             entry->execute)[j];
7588                                         if (!ctl_cmd_applicable(
7589                                             lun->be_lun->lun_type, sentry))
7590                                                 continue;
7591                                         descr = &all->descr[num++];
7592                                         descr->opcode = i;
7593                                         scsi_ulto2b(j, descr->service_action);
7594                                         descr->flags = RSO_SERVACTV;
7595                                         scsi_ulto2b(sentry->length,
7596                                             descr->cdb_length);
7597                                 }
7598                         } else {
7599                                 if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7600                                     entry))
7601                                         continue;
7602                                 descr = &all->descr[num++];
7603                                 descr->opcode = i;
7604                                 scsi_ulto2b(0, descr->service_action);
7605                                 descr->flags = 0;
7606                                 scsi_ulto2b(entry->length, descr->cdb_length);
7607                         }
7608                 }
7609                 scsi_ulto4b(
7610                     num * sizeof(struct scsi_report_supported_opcodes_descr),
7611                     all->length);
7612                 break;
7613         case RSO_OPTIONS_OC:
7614                 one = (struct scsi_report_supported_opcodes_one *)
7615                     ctsio->kern_data_ptr;
7616                 entry = &ctl_cmd_table[opcode];
7617                 goto fill_one;
7618         case RSO_OPTIONS_OC_SA:
7619                 one = (struct scsi_report_supported_opcodes_one *)
7620                     ctsio->kern_data_ptr;
7621                 entry = &ctl_cmd_table[opcode];
7622                 entry = &((const struct ctl_cmd_entry *)
7623                     entry->execute)[service_action];
7624 fill_one:
7625                 if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7626                         one->support = 3;
7627                         scsi_ulto2b(entry->length, one->cdb_length);
7628                         one->cdb_usage[0] = opcode;
7629                         memcpy(&one->cdb_usage[1], entry->usage,
7630                             entry->length - 1);
7631                 } else
7632                         one->support = 1;
7633                 break;
7634         }
7635
7636         ctl_set_success(ctsio);
7637         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7638         ctsio->be_move_done = ctl_config_move_done;
7639         ctl_datamove((union ctl_io *)ctsio);
7640         return(retval);
7641 }
7642
7643 int
7644 ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7645 {
7646         struct scsi_report_supported_tmf *cdb;
7647         struct scsi_report_supported_tmf_data *data;
7648         int retval;
7649         int alloc_len, total_len;
7650
7651         CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7652
7653         cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7654
7655         retval = CTL_RETVAL_COMPLETE;
7656
7657         total_len = sizeof(struct scsi_report_supported_tmf_data);
7658         alloc_len = scsi_4btoul(cdb->length);
7659
7660         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7661
7662         ctsio->kern_sg_entries = 0;
7663
7664         if (total_len < alloc_len) {
7665                 ctsio->residual = alloc_len - total_len;
7666                 ctsio->kern_data_len = total_len;
7667                 ctsio->kern_total_len = total_len;
7668         } else {
7669                 ctsio->residual = 0;
7670                 ctsio->kern_data_len = alloc_len;
7671                 ctsio->kern_total_len = alloc_len;
7672         }
7673         ctsio->kern_data_resid = 0;
7674         ctsio->kern_rel_offset = 0;
7675
7676         data = (struct scsi_report_supported_tmf_data *)ctsio->kern_data_ptr;
7677         data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_QTS |
7678             RST_TRS;
7679         data->byte2 |= RST_QAES | RST_QTSS | RST_ITNRS;
7680
7681         ctl_set_success(ctsio);
7682         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7683         ctsio->be_move_done = ctl_config_move_done;
7684         ctl_datamove((union ctl_io *)ctsio);
7685         return (retval);
7686 }
7687
7688 int
7689 ctl_report_timestamp(struct ctl_scsiio *ctsio)
7690 {
7691         struct scsi_report_timestamp *cdb;
7692         struct scsi_report_timestamp_data *data;
7693         struct timeval tv;
7694         int64_t timestamp;
7695         int retval;
7696         int alloc_len, total_len;
7697
7698         CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7699
7700         cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7701
7702         retval = CTL_RETVAL_COMPLETE;
7703
7704         total_len = sizeof(struct scsi_report_timestamp_data);
7705         alloc_len = scsi_4btoul(cdb->length);
7706
7707         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7708
7709         ctsio->kern_sg_entries = 0;
7710
7711         if (total_len < alloc_len) {
7712                 ctsio->residual = alloc_len - total_len;
7713                 ctsio->kern_data_len = total_len;
7714                 ctsio->kern_total_len = total_len;
7715         } else {
7716                 ctsio->residual = 0;
7717                 ctsio->kern_data_len = alloc_len;
7718                 ctsio->kern_total_len = alloc_len;
7719         }
7720         ctsio->kern_data_resid = 0;
7721         ctsio->kern_rel_offset = 0;
7722
7723         data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7724         scsi_ulto2b(sizeof(*data) - 2, data->length);
7725         data->origin = RTS_ORIG_OUTSIDE;
7726         getmicrotime(&tv);
7727         timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7728         scsi_ulto4b(timestamp >> 16, data->timestamp);
7729         scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7730
7731         ctl_set_success(ctsio);
7732         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7733         ctsio->be_move_done = ctl_config_move_done;
7734         ctl_datamove((union ctl_io *)ctsio);
7735         return (retval);
7736 }
7737
7738 int
7739 ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7740 {
7741         struct scsi_per_res_in *cdb;
7742         int alloc_len, total_len = 0;
7743         /* struct scsi_per_res_in_rsrv in_data; */
7744         struct ctl_lun *lun;
7745         struct ctl_softc *softc;
7746         uint64_t key;
7747
7748         CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7749
7750         cdb = (struct scsi_per_res_in *)ctsio->cdb;
7751
7752         alloc_len = scsi_2btoul(cdb->length);
7753
7754         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7755         softc = lun->ctl_softc;
7756
7757 retry:
7758         mtx_lock(&lun->lun_lock);
7759         switch (cdb->action) {
7760         case SPRI_RK: /* read keys */
7761                 total_len = sizeof(struct scsi_per_res_in_keys) +
7762                         lun->pr_key_count *
7763                         sizeof(struct scsi_per_res_key);
7764                 break;
7765         case SPRI_RR: /* read reservation */
7766                 if (lun->flags & CTL_LUN_PR_RESERVED)
7767                         total_len = sizeof(struct scsi_per_res_in_rsrv);
7768                 else
7769                         total_len = sizeof(struct scsi_per_res_in_header);
7770                 break;
7771         case SPRI_RC: /* report capabilities */
7772                 total_len = sizeof(struct scsi_per_res_cap);
7773                 break;
7774         case SPRI_RS: /* read full status */
7775                 total_len = sizeof(struct scsi_per_res_in_header) +
7776                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7777                     lun->pr_key_count;
7778                 break;
7779         default:
7780                 panic("Invalid PR type %x", cdb->action);
7781         }
7782         mtx_unlock(&lun->lun_lock);
7783
7784         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7785
7786         if (total_len < alloc_len) {
7787                 ctsio->residual = alloc_len - total_len;
7788                 ctsio->kern_data_len = total_len;
7789                 ctsio->kern_total_len = total_len;
7790         } else {
7791                 ctsio->residual = 0;
7792                 ctsio->kern_data_len = alloc_len;
7793                 ctsio->kern_total_len = alloc_len;
7794         }
7795
7796         ctsio->kern_data_resid = 0;
7797         ctsio->kern_rel_offset = 0;
7798         ctsio->kern_sg_entries = 0;
7799
7800         mtx_lock(&lun->lun_lock);
7801         switch (cdb->action) {
7802         case SPRI_RK: { // read keys
7803         struct scsi_per_res_in_keys *res_keys;
7804                 int i, key_count;
7805
7806                 res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7807
7808                 /*
7809                  * We had to drop the lock to allocate our buffer, which
7810                  * leaves time for someone to come in with another
7811                  * persistent reservation.  (That is unlikely, though,
7812                  * since this should be the only persistent reservation
7813                  * command active right now.)
7814                  */
7815                 if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7816                     (lun->pr_key_count *
7817                      sizeof(struct scsi_per_res_key)))){
7818                         mtx_unlock(&lun->lun_lock);
7819                         free(ctsio->kern_data_ptr, M_CTL);
7820                         printf("%s: reservation length changed, retrying\n",
7821                                __func__);
7822                         goto retry;
7823                 }
7824
7825                 scsi_ulto4b(lun->PRGeneration, res_keys->header.generation);
7826
7827                 scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7828                              lun->pr_key_count, res_keys->header.length);
7829
7830                 for (i = 0, key_count = 0; i < CTL_MAX_INITIATORS; i++) {
7831                         if ((key = ctl_get_prkey(lun, i)) == 0)
7832                                 continue;
7833
7834                         /*
7835                          * We used lun->pr_key_count to calculate the
7836                          * size to allocate.  If it turns out the number of
7837                          * initiators with the registered flag set is
7838                          * larger than that (i.e. they haven't been kept in
7839                          * sync), we've got a problem.
7840                          */
7841                         if (key_count >= lun->pr_key_count) {
7842 #ifdef NEEDTOPORT
7843                                 csevent_log(CSC_CTL | CSC_SHELF_SW |
7844                                             CTL_PR_ERROR,
7845                                             csevent_LogType_Fault,
7846                                             csevent_AlertLevel_Yellow,
7847                                             csevent_FRU_ShelfController,
7848                                             csevent_FRU_Firmware,
7849                                         csevent_FRU_Unknown,
7850                                             "registered keys %d >= key "
7851                                             "count %d", key_count,
7852                                             lun->pr_key_count);
7853 #endif
7854                                 key_count++;
7855                                 continue;
7856                         }
7857                         scsi_u64to8b(key, res_keys->keys[key_count].key);
7858                         key_count++;
7859                 }
7860                 break;
7861         }
7862         case SPRI_RR: { // read reservation
7863                 struct scsi_per_res_in_rsrv *res;
7864                 int tmp_len, header_only;
7865
7866                 res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7867
7868                 scsi_ulto4b(lun->PRGeneration, res->header.generation);
7869
7870                 if (lun->flags & CTL_LUN_PR_RESERVED)
7871                 {
7872                         tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7873                         scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7874                                     res->header.length);
7875                         header_only = 0;
7876                 } else {
7877                         tmp_len = sizeof(struct scsi_per_res_in_header);
7878                         scsi_ulto4b(0, res->header.length);
7879                         header_only = 1;
7880                 }
7881
7882                 /*
7883                  * We had to drop the lock to allocate our buffer, which
7884                  * leaves time for someone to come in with another
7885                  * persistent reservation.  (That is unlikely, though,
7886                  * since this should be the only persistent reservation
7887                  * command active right now.)
7888                  */
7889                 if (tmp_len != total_len) {
7890                         mtx_unlock(&lun->lun_lock);
7891                         free(ctsio->kern_data_ptr, M_CTL);
7892                         printf("%s: reservation status changed, retrying\n",
7893                                __func__);
7894                         goto retry;
7895                 }
7896
7897                 /*
7898                  * No reservation held, so we're done.
7899                  */
7900                 if (header_only != 0)
7901                         break;
7902
7903                 /*
7904                  * If the registration is an All Registrants type, the key
7905                  * is 0, since it doesn't really matter.
7906                  */
7907                 if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7908                         scsi_u64to8b(ctl_get_prkey(lun, lun->pr_res_idx),
7909                             res->data.reservation);
7910                 }
7911                 res->data.scopetype = lun->res_type;
7912                 break;
7913         }
7914         case SPRI_RC:     //report capabilities
7915         {
7916                 struct scsi_per_res_cap *res_cap;
7917                 uint16_t type_mask;
7918
7919                 res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7920                 scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7921                 res_cap->flags2 |= SPRI_TMV | SPRI_ALLOW_5;
7922                 type_mask = SPRI_TM_WR_EX_AR |
7923                             SPRI_TM_EX_AC_RO |
7924                             SPRI_TM_WR_EX_RO |
7925                             SPRI_TM_EX_AC |
7926                             SPRI_TM_WR_EX |
7927                             SPRI_TM_EX_AC_AR;
7928                 scsi_ulto2b(type_mask, res_cap->type_mask);
7929                 break;
7930         }
7931         case SPRI_RS: { // read full status
7932                 struct scsi_per_res_in_full *res_status;
7933                 struct scsi_per_res_in_full_desc *res_desc;
7934                 struct ctl_port *port;
7935                 int i, len;
7936
7937                 res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7938
7939                 /*
7940                  * We had to drop the lock to allocate our buffer, which
7941                  * leaves time for someone to come in with another
7942                  * persistent reservation.  (That is unlikely, though,
7943                  * since this should be the only persistent reservation
7944                  * command active right now.)
7945                  */
7946                 if (total_len < (sizeof(struct scsi_per_res_in_header) +
7947                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7948                      lun->pr_key_count)){
7949                         mtx_unlock(&lun->lun_lock);
7950                         free(ctsio->kern_data_ptr, M_CTL);
7951                         printf("%s: reservation length changed, retrying\n",
7952                                __func__);
7953                         goto retry;
7954                 }
7955
7956                 scsi_ulto4b(lun->PRGeneration, res_status->header.generation);
7957
7958                 res_desc = &res_status->desc[0];
7959                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7960                         if ((key = ctl_get_prkey(lun, i)) == 0)
7961                                 continue;
7962
7963                         scsi_u64to8b(key, res_desc->res_key.key);
7964                         if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7965                             (lun->pr_res_idx == i ||
7966                              lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7967                                 res_desc->flags = SPRI_FULL_R_HOLDER;
7968                                 res_desc->scopetype = lun->res_type;
7969                         }
7970                         scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7971                             res_desc->rel_trgt_port_id);
7972                         len = 0;
7973                         port = softc->ctl_ports[i / CTL_MAX_INIT_PER_PORT];
7974                         if (port != NULL)
7975                                 len = ctl_create_iid(port,
7976                                     i % CTL_MAX_INIT_PER_PORT,
7977                                     res_desc->transport_id);
7978                         scsi_ulto4b(len, res_desc->additional_length);
7979                         res_desc = (struct scsi_per_res_in_full_desc *)
7980                             &res_desc->transport_id[len];
7981                 }
7982                 scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7983                     res_status->header.length);
7984                 break;
7985         }
7986         default:
7987                 /*
7988                  * This is a bug, because we just checked for this above,
7989                  * and should have returned an error.
7990                  */
7991                 panic("Invalid PR type %x", cdb->action);
7992                 break; /* NOTREACHED */
7993         }
7994         mtx_unlock(&lun->lun_lock);
7995
7996         ctl_set_success(ctsio);
7997         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7998         ctsio->be_move_done = ctl_config_move_done;
7999         ctl_datamove((union ctl_io *)ctsio);
8000         return (CTL_RETVAL_COMPLETE);
8001 }
8002
8003 /*
8004  * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
8005  * it should return.
8006  */
8007 static int
8008 ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
8009                 uint64_t sa_res_key, uint8_t type, uint32_t residx,
8010                 struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
8011                 struct scsi_per_res_out_parms* param)
8012 {
8013         union ctl_ha_msg persis_io;
8014         int i;
8015
8016         mtx_lock(&lun->lun_lock);
8017         if (sa_res_key == 0) {
8018                 if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8019                         /* validate scope and type */
8020                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
8021                              SPR_LU_SCOPE) {
8022                                 mtx_unlock(&lun->lun_lock);
8023                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
8024                                                       /*sks_valid*/ 1,
8025                                                       /*command*/ 1,
8026                                                       /*field*/ 2,
8027                                                       /*bit_valid*/ 1,
8028                                                       /*bit*/ 4);
8029                                 ctl_done((union ctl_io *)ctsio);
8030                                 return (1);
8031                         }
8032
8033                         if (type>8 || type==2 || type==4 || type==0) {
8034                                 mtx_unlock(&lun->lun_lock);
8035                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
8036                                                       /*sks_valid*/ 1,
8037                                                       /*command*/ 1,
8038                                                       /*field*/ 2,
8039                                                       /*bit_valid*/ 1,
8040                                                       /*bit*/ 0);
8041                                 ctl_done((union ctl_io *)ctsio);
8042                                 return (1);
8043                         }
8044
8045                         /*
8046                          * Unregister everybody else and build UA for
8047                          * them
8048                          */
8049                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
8050                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8051                                         continue;
8052
8053                                 ctl_clr_prkey(lun, i);
8054                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8055                         }
8056                         lun->pr_key_count = 1;
8057                         lun->res_type = type;
8058                         if (lun->res_type != SPR_TYPE_WR_EX_AR
8059                          && lun->res_type != SPR_TYPE_EX_AC_AR)
8060                                 lun->pr_res_idx = residx;
8061                         lun->PRGeneration++;
8062                         mtx_unlock(&lun->lun_lock);
8063
8064                         /* send msg to other side */
8065                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8066                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8067                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8068                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8069                         persis_io.pr.pr_info.res_type = type;
8070                         memcpy(persis_io.pr.pr_info.sa_res_key,
8071                                param->serv_act_res_key,
8072                                sizeof(param->serv_act_res_key));
8073                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8074                             sizeof(persis_io.pr), M_WAITOK);
8075                 } else {
8076                         /* not all registrants */
8077                         mtx_unlock(&lun->lun_lock);
8078                         free(ctsio->kern_data_ptr, M_CTL);
8079                         ctl_set_invalid_field(ctsio,
8080                                               /*sks_valid*/ 1,
8081                                               /*command*/ 0,
8082                                               /*field*/ 8,
8083                                               /*bit_valid*/ 0,
8084                                               /*bit*/ 0);
8085                         ctl_done((union ctl_io *)ctsio);
8086                         return (1);
8087                 }
8088         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8089                 || !(lun->flags & CTL_LUN_PR_RESERVED)) {
8090                 int found = 0;
8091
8092                 if (res_key == sa_res_key) {
8093                         /* special case */
8094                         /*
8095                          * The spec implies this is not good but doesn't
8096                          * say what to do. There are two choices either
8097                          * generate a res conflict or check condition
8098                          * with illegal field in parameter data. Since
8099                          * that is what is done when the sa_res_key is
8100                          * zero I'll take that approach since this has
8101                          * to do with the sa_res_key.
8102                          */
8103                         mtx_unlock(&lun->lun_lock);
8104                         free(ctsio->kern_data_ptr, M_CTL);
8105                         ctl_set_invalid_field(ctsio,
8106                                               /*sks_valid*/ 1,
8107                                               /*command*/ 0,
8108                                               /*field*/ 8,
8109                                               /*bit_valid*/ 0,
8110                                               /*bit*/ 0);
8111                         ctl_done((union ctl_io *)ctsio);
8112                         return (1);
8113                 }
8114
8115                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8116                         if (ctl_get_prkey(lun, i) != sa_res_key)
8117                                 continue;
8118
8119                         found = 1;
8120                         ctl_clr_prkey(lun, i);
8121                         lun->pr_key_count--;
8122                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8123                 }
8124                 if (!found) {
8125                         mtx_unlock(&lun->lun_lock);
8126                         free(ctsio->kern_data_ptr, M_CTL);
8127                         ctl_set_reservation_conflict(ctsio);
8128                         ctl_done((union ctl_io *)ctsio);
8129                         return (CTL_RETVAL_COMPLETE);
8130                 }
8131                 lun->PRGeneration++;
8132                 mtx_unlock(&lun->lun_lock);
8133
8134                 /* send msg to other side */
8135                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8136                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8137                 persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8138                 persis_io.pr.pr_info.residx = lun->pr_res_idx;
8139                 persis_io.pr.pr_info.res_type = type;
8140                 memcpy(persis_io.pr.pr_info.sa_res_key,
8141                        param->serv_act_res_key,
8142                        sizeof(param->serv_act_res_key));
8143                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8144                     sizeof(persis_io.pr), M_WAITOK);
8145         } else {
8146                 /* Reserved but not all registrants */
8147                 /* sa_res_key is res holder */
8148                 if (sa_res_key == ctl_get_prkey(lun, lun->pr_res_idx)) {
8149                         /* validate scope and type */
8150                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
8151                              SPR_LU_SCOPE) {
8152                                 mtx_unlock(&lun->lun_lock);
8153                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
8154                                                       /*sks_valid*/ 1,
8155                                                       /*command*/ 1,
8156                                                       /*field*/ 2,
8157                                                       /*bit_valid*/ 1,
8158                                                       /*bit*/ 4);
8159                                 ctl_done((union ctl_io *)ctsio);
8160                                 return (1);
8161                         }
8162
8163                         if (type>8 || type==2 || type==4 || type==0) {
8164                                 mtx_unlock(&lun->lun_lock);
8165                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
8166                                                       /*sks_valid*/ 1,
8167                                                       /*command*/ 1,
8168                                                       /*field*/ 2,
8169                                                       /*bit_valid*/ 1,
8170                                                       /*bit*/ 0);
8171                                 ctl_done((union ctl_io *)ctsio);
8172                                 return (1);
8173                         }
8174
8175                         /*
8176                          * Do the following:
8177                          * if sa_res_key != res_key remove all
8178                          * registrants w/sa_res_key and generate UA
8179                          * for these registrants(Registrations
8180                          * Preempted) if it wasn't an exclusive
8181                          * reservation generate UA(Reservations
8182                          * Preempted) for all other registered nexuses
8183                          * if the type has changed. Establish the new
8184                          * reservation and holder. If res_key and
8185                          * sa_res_key are the same do the above
8186                          * except don't unregister the res holder.
8187                          */
8188
8189                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
8190                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8191                                         continue;
8192
8193                                 if (sa_res_key == ctl_get_prkey(lun, i)) {
8194                                         ctl_clr_prkey(lun, i);
8195                                         lun->pr_key_count--;
8196                                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8197                                 } else if (type != lun->res_type
8198                                         && (lun->res_type == SPR_TYPE_WR_EX_RO
8199                                          || lun->res_type ==SPR_TYPE_EX_AC_RO)){
8200                                         ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8201                                 }
8202                         }
8203                         lun->res_type = type;
8204                         if (lun->res_type != SPR_TYPE_WR_EX_AR
8205                          && lun->res_type != SPR_TYPE_EX_AC_AR)
8206                                 lun->pr_res_idx = residx;
8207                         else
8208                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8209                         lun->PRGeneration++;
8210                         mtx_unlock(&lun->lun_lock);
8211
8212                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8213                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8214                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8215                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8216                         persis_io.pr.pr_info.res_type = type;
8217                         memcpy(persis_io.pr.pr_info.sa_res_key,
8218                                param->serv_act_res_key,
8219                                sizeof(param->serv_act_res_key));
8220                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8221                             sizeof(persis_io.pr), M_WAITOK);
8222                 } else {
8223                         /*
8224                          * sa_res_key is not the res holder just
8225                          * remove registrants
8226                          */
8227                         int found=0;
8228
8229                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8230                                 if (sa_res_key != ctl_get_prkey(lun, i))
8231                                         continue;
8232
8233                                 found = 1;
8234                                 ctl_clr_prkey(lun, i);
8235                                 lun->pr_key_count--;
8236                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8237                         }
8238
8239                         if (!found) {
8240                                 mtx_unlock(&lun->lun_lock);
8241                                 free(ctsio->kern_data_ptr, M_CTL);
8242                                 ctl_set_reservation_conflict(ctsio);
8243                                 ctl_done((union ctl_io *)ctsio);
8244                                 return (1);
8245                         }
8246                         lun->PRGeneration++;
8247                         mtx_unlock(&lun->lun_lock);
8248
8249                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8250                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8251                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8252                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8253                         persis_io.pr.pr_info.res_type = type;
8254                         memcpy(persis_io.pr.pr_info.sa_res_key,
8255                                param->serv_act_res_key,
8256                                sizeof(param->serv_act_res_key));
8257                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8258                             sizeof(persis_io.pr), M_WAITOK);
8259                 }
8260         }
8261         return (0);
8262 }
8263
8264 static void
8265 ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
8266 {
8267         uint64_t sa_res_key;
8268         int i;
8269
8270         sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
8271
8272         if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8273          || lun->pr_res_idx == CTL_PR_NO_RESERVATION
8274          || sa_res_key != ctl_get_prkey(lun, lun->pr_res_idx)) {
8275                 if (sa_res_key == 0) {
8276                         /*
8277                          * Unregister everybody else and build UA for
8278                          * them
8279                          */
8280                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
8281                                 if (i == msg->pr.pr_info.residx ||
8282                                     ctl_get_prkey(lun, i) == 0)
8283                                         continue;
8284
8285                                 ctl_clr_prkey(lun, i);
8286                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8287                         }
8288
8289                         lun->pr_key_count = 1;
8290                         lun->res_type = msg->pr.pr_info.res_type;
8291                         if (lun->res_type != SPR_TYPE_WR_EX_AR
8292                          && lun->res_type != SPR_TYPE_EX_AC_AR)
8293                                 lun->pr_res_idx = msg->pr.pr_info.residx;
8294                 } else {
8295                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8296                                 if (sa_res_key == ctl_get_prkey(lun, i))
8297                                         continue;
8298
8299                                 ctl_clr_prkey(lun, i);
8300                                 lun->pr_key_count--;
8301                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8302                         }
8303                 }
8304         } else {
8305                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8306                         if (i == msg->pr.pr_info.residx ||
8307                             ctl_get_prkey(lun, i) == 0)
8308                                 continue;
8309
8310                         if (sa_res_key == ctl_get_prkey(lun, i)) {
8311                                 ctl_clr_prkey(lun, i);
8312                                 lun->pr_key_count--;
8313                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8314                         } else if (msg->pr.pr_info.res_type != lun->res_type
8315                                 && (lun->res_type == SPR_TYPE_WR_EX_RO
8316                                  || lun->res_type == SPR_TYPE_EX_AC_RO)) {
8317                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8318                         }
8319                 }
8320                 lun->res_type = msg->pr.pr_info.res_type;
8321                 if (lun->res_type != SPR_TYPE_WR_EX_AR
8322                  && lun->res_type != SPR_TYPE_EX_AC_AR)
8323                         lun->pr_res_idx = msg->pr.pr_info.residx;
8324                 else
8325                         lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8326         }
8327         lun->PRGeneration++;
8328
8329 }
8330
8331
8332 int
8333 ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8334 {
8335         int retval;
8336         u_int32_t param_len;
8337         struct scsi_per_res_out *cdb;
8338         struct ctl_lun *lun;
8339         struct scsi_per_res_out_parms* param;
8340         struct ctl_softc *softc;
8341         uint32_t residx;
8342         uint64_t res_key, sa_res_key, key;
8343         uint8_t type;
8344         union ctl_ha_msg persis_io;
8345         int    i;
8346
8347         CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8348
8349         retval = CTL_RETVAL_COMPLETE;
8350
8351         cdb = (struct scsi_per_res_out *)ctsio->cdb;
8352         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8353         softc = lun->ctl_softc;
8354
8355         /*
8356          * We only support whole-LUN scope.  The scope & type are ignored for
8357          * register, register and ignore existing key and clear.
8358          * We sometimes ignore scope and type on preempts too!!
8359          * Verify reservation type here as well.
8360          */
8361         type = cdb->scope_type & SPR_TYPE_MASK;
8362         if ((cdb->action == SPRO_RESERVE)
8363          || (cdb->action == SPRO_RELEASE)) {
8364                 if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8365                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8366                                               /*sks_valid*/ 1,
8367                                               /*command*/ 1,
8368                                               /*field*/ 2,
8369                                               /*bit_valid*/ 1,
8370                                               /*bit*/ 4);
8371                         ctl_done((union ctl_io *)ctsio);
8372                         return (CTL_RETVAL_COMPLETE);
8373                 }
8374
8375                 if (type>8 || type==2 || type==4 || type==0) {
8376                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8377                                               /*sks_valid*/ 1,
8378                                               /*command*/ 1,
8379                                               /*field*/ 2,
8380                                               /*bit_valid*/ 1,
8381                                               /*bit*/ 0);
8382                         ctl_done((union ctl_io *)ctsio);
8383                         return (CTL_RETVAL_COMPLETE);
8384                 }
8385         }
8386
8387         param_len = scsi_4btoul(cdb->length);
8388
8389         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8390                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8391                 ctsio->kern_data_len = param_len;
8392                 ctsio->kern_total_len = param_len;
8393                 ctsio->kern_data_resid = 0;
8394                 ctsio->kern_rel_offset = 0;
8395                 ctsio->kern_sg_entries = 0;
8396                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8397                 ctsio->be_move_done = ctl_config_move_done;
8398                 ctl_datamove((union ctl_io *)ctsio);
8399
8400                 return (CTL_RETVAL_COMPLETE);
8401         }
8402
8403         param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8404
8405         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
8406         res_key = scsi_8btou64(param->res_key.key);
8407         sa_res_key = scsi_8btou64(param->serv_act_res_key);
8408
8409         /*
8410          * Validate the reservation key here except for SPRO_REG_IGNO
8411          * This must be done for all other service actions
8412          */
8413         if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8414                 mtx_lock(&lun->lun_lock);
8415                 if ((key = ctl_get_prkey(lun, residx)) != 0) {
8416                         if (res_key != key) {
8417                                 /*
8418                                  * The current key passed in doesn't match
8419                                  * the one the initiator previously
8420                                  * registered.
8421                                  */
8422                                 mtx_unlock(&lun->lun_lock);
8423                                 free(ctsio->kern_data_ptr, M_CTL);
8424                                 ctl_set_reservation_conflict(ctsio);
8425                                 ctl_done((union ctl_io *)ctsio);
8426                                 return (CTL_RETVAL_COMPLETE);
8427                         }
8428                 } else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8429                         /*
8430                          * We are not registered
8431                          */
8432                         mtx_unlock(&lun->lun_lock);
8433                         free(ctsio->kern_data_ptr, M_CTL);
8434                         ctl_set_reservation_conflict(ctsio);
8435                         ctl_done((union ctl_io *)ctsio);
8436                         return (CTL_RETVAL_COMPLETE);
8437                 } else if (res_key != 0) {
8438                         /*
8439                          * We are not registered and trying to register but
8440                          * the register key isn't zero.
8441                          */
8442                         mtx_unlock(&lun->lun_lock);
8443                         free(ctsio->kern_data_ptr, M_CTL);
8444                         ctl_set_reservation_conflict(ctsio);
8445                         ctl_done((union ctl_io *)ctsio);
8446                         return (CTL_RETVAL_COMPLETE);
8447                 }
8448                 mtx_unlock(&lun->lun_lock);
8449         }
8450
8451         switch (cdb->action & SPRO_ACTION_MASK) {
8452         case SPRO_REGISTER:
8453         case SPRO_REG_IGNO: {
8454
8455 #if 0
8456                 printf("Registration received\n");
8457 #endif
8458
8459                 /*
8460                  * We don't support any of these options, as we report in
8461                  * the read capabilities request (see
8462                  * ctl_persistent_reserve_in(), above).
8463                  */
8464                 if ((param->flags & SPR_SPEC_I_PT)
8465                  || (param->flags & SPR_ALL_TG_PT)
8466                  || (param->flags & SPR_APTPL)) {
8467                         int bit_ptr;
8468
8469                         if (param->flags & SPR_APTPL)
8470                                 bit_ptr = 0;
8471                         else if (param->flags & SPR_ALL_TG_PT)
8472                                 bit_ptr = 2;
8473                         else /* SPR_SPEC_I_PT */
8474                                 bit_ptr = 3;
8475
8476                         free(ctsio->kern_data_ptr, M_CTL);
8477                         ctl_set_invalid_field(ctsio,
8478                                               /*sks_valid*/ 1,
8479                                               /*command*/ 0,
8480                                               /*field*/ 20,
8481                                               /*bit_valid*/ 1,
8482                                               /*bit*/ bit_ptr);
8483                         ctl_done((union ctl_io *)ctsio);
8484                         return (CTL_RETVAL_COMPLETE);
8485                 }
8486
8487                 mtx_lock(&lun->lun_lock);
8488
8489                 /*
8490                  * The initiator wants to clear the
8491                  * key/unregister.
8492                  */
8493                 if (sa_res_key == 0) {
8494                         if ((res_key == 0
8495                           && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8496                          || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8497                           && ctl_get_prkey(lun, residx) == 0)) {
8498                                 mtx_unlock(&lun->lun_lock);
8499                                 goto done;
8500                         }
8501
8502                         ctl_clr_prkey(lun, residx);
8503                         lun->pr_key_count--;
8504
8505                         if (residx == lun->pr_res_idx) {
8506                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8507                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8508
8509                                 if ((lun->res_type == SPR_TYPE_WR_EX_RO
8510                                   || lun->res_type == SPR_TYPE_EX_AC_RO)
8511                                  && lun->pr_key_count) {
8512                                         /*
8513                                          * If the reservation is a registrants
8514                                          * only type we need to generate a UA
8515                                          * for other registered inits.  The
8516                                          * sense code should be RESERVATIONS
8517                                          * RELEASED
8518                                          */
8519
8520                                         for (i = softc->init_min; i < softc->init_max; i++){
8521                                                 if (ctl_get_prkey(lun, i) == 0)
8522                                                         continue;
8523                                                 ctl_est_ua(lun, i,
8524                                                     CTL_UA_RES_RELEASE);
8525                                         }
8526                                 }
8527                                 lun->res_type = 0;
8528                         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8529                                 if (lun->pr_key_count==0) {
8530                                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8531                                         lun->res_type = 0;
8532                                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8533                                 }
8534                         }
8535                         lun->PRGeneration++;
8536                         mtx_unlock(&lun->lun_lock);
8537
8538                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8539                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8540                         persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8541                         persis_io.pr.pr_info.residx = residx;
8542                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8543                             sizeof(persis_io.pr), M_WAITOK);
8544                 } else /* sa_res_key != 0 */ {
8545
8546                         /*
8547                          * If we aren't registered currently then increment
8548                          * the key count and set the registered flag.
8549                          */
8550                         ctl_alloc_prkey(lun, residx);
8551                         if (ctl_get_prkey(lun, residx) == 0)
8552                                 lun->pr_key_count++;
8553                         ctl_set_prkey(lun, residx, sa_res_key);
8554                         lun->PRGeneration++;
8555                         mtx_unlock(&lun->lun_lock);
8556
8557                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8558                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8559                         persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8560                         persis_io.pr.pr_info.residx = residx;
8561                         memcpy(persis_io.pr.pr_info.sa_res_key,
8562                                param->serv_act_res_key,
8563                                sizeof(param->serv_act_res_key));
8564                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8565                             sizeof(persis_io.pr), M_WAITOK);
8566                 }
8567
8568                 break;
8569         }
8570         case SPRO_RESERVE:
8571 #if 0
8572                 printf("Reserve executed type %d\n", type);
8573 #endif
8574                 mtx_lock(&lun->lun_lock);
8575                 if (lun->flags & CTL_LUN_PR_RESERVED) {
8576                         /*
8577                          * if this isn't the reservation holder and it's
8578                          * not a "all registrants" type or if the type is
8579                          * different then we have a conflict
8580                          */
8581                         if ((lun->pr_res_idx != residx
8582                           && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8583                          || lun->res_type != type) {
8584                                 mtx_unlock(&lun->lun_lock);
8585                                 free(ctsio->kern_data_ptr, M_CTL);
8586                                 ctl_set_reservation_conflict(ctsio);
8587                                 ctl_done((union ctl_io *)ctsio);
8588                                 return (CTL_RETVAL_COMPLETE);
8589                         }
8590                         mtx_unlock(&lun->lun_lock);
8591                 } else /* create a reservation */ {
8592                         /*
8593                          * If it's not an "all registrants" type record
8594                          * reservation holder
8595                          */
8596                         if (type != SPR_TYPE_WR_EX_AR
8597                          && type != SPR_TYPE_EX_AC_AR)
8598                                 lun->pr_res_idx = residx; /* Res holder */
8599                         else
8600                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8601
8602                         lun->flags |= CTL_LUN_PR_RESERVED;
8603                         lun->res_type = type;
8604
8605                         mtx_unlock(&lun->lun_lock);
8606
8607                         /* send msg to other side */
8608                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8609                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8610                         persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8611                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8612                         persis_io.pr.pr_info.res_type = type;
8613                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8614                             sizeof(persis_io.pr), M_WAITOK);
8615                 }
8616                 break;
8617
8618         case SPRO_RELEASE:
8619                 mtx_lock(&lun->lun_lock);
8620                 if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8621                         /* No reservation exists return good status */
8622                         mtx_unlock(&lun->lun_lock);
8623                         goto done;
8624                 }
8625                 /*
8626                  * Is this nexus a reservation holder?
8627                  */
8628                 if (lun->pr_res_idx != residx
8629                  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8630                         /*
8631                          * not a res holder return good status but
8632                          * do nothing
8633                          */
8634                         mtx_unlock(&lun->lun_lock);
8635                         goto done;
8636                 }
8637
8638                 if (lun->res_type != type) {
8639                         mtx_unlock(&lun->lun_lock);
8640                         free(ctsio->kern_data_ptr, M_CTL);
8641                         ctl_set_illegal_pr_release(ctsio);
8642                         ctl_done((union ctl_io *)ctsio);
8643                         return (CTL_RETVAL_COMPLETE);
8644                 }
8645
8646                 /* okay to release */
8647                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8648                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8649                 lun->res_type = 0;
8650
8651                 /*
8652                  * if this isn't an exclusive access
8653                  * res generate UA for all other
8654                  * registrants.
8655                  */
8656                 if (type != SPR_TYPE_EX_AC
8657                  && type != SPR_TYPE_WR_EX) {
8658                         for (i = softc->init_min; i < softc->init_max; i++) {
8659                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8660                                         continue;
8661                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8662                         }
8663                 }
8664                 mtx_unlock(&lun->lun_lock);
8665
8666                 /* Send msg to other side */
8667                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8668                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8669                 persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8670                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8671                      sizeof(persis_io.pr), M_WAITOK);
8672                 break;
8673
8674         case SPRO_CLEAR:
8675                 /* send msg to other side */
8676
8677                 mtx_lock(&lun->lun_lock);
8678                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8679                 lun->res_type = 0;
8680                 lun->pr_key_count = 0;
8681                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8682
8683                 ctl_clr_prkey(lun, residx);
8684                 for (i = 0; i < CTL_MAX_INITIATORS; i++)
8685                         if (ctl_get_prkey(lun, i) != 0) {
8686                                 ctl_clr_prkey(lun, i);
8687                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8688                         }
8689                 lun->PRGeneration++;
8690                 mtx_unlock(&lun->lun_lock);
8691
8692                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8693                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8694                 persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8695                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8696                      sizeof(persis_io.pr), M_WAITOK);
8697                 break;
8698
8699         case SPRO_PREEMPT:
8700         case SPRO_PRE_ABO: {
8701                 int nretval;
8702
8703                 nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8704                                           residx, ctsio, cdb, param);
8705                 if (nretval != 0)
8706                         return (CTL_RETVAL_COMPLETE);
8707                 break;
8708         }
8709         default:
8710                 panic("Invalid PR type %x", cdb->action);
8711         }
8712
8713 done:
8714         free(ctsio->kern_data_ptr, M_CTL);
8715         ctl_set_success(ctsio);
8716         ctl_done((union ctl_io *)ctsio);
8717
8718         return (retval);
8719 }
8720
8721 /*
8722  * This routine is for handling a message from the other SC pertaining to
8723  * persistent reserve out. All the error checking will have been done
8724  * so only perorming the action need be done here to keep the two
8725  * in sync.
8726  */
8727 static void
8728 ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8729 {
8730         struct ctl_lun *lun;
8731         struct ctl_softc *softc;
8732         int i;
8733         uint32_t residx, targ_lun;
8734
8735         softc = control_softc;
8736         targ_lun = msg->hdr.nexus.targ_mapped_lun;
8737         mtx_lock(&softc->ctl_lock);
8738         if ((targ_lun >= CTL_MAX_LUNS) ||
8739             ((lun = softc->ctl_luns[targ_lun]) == NULL)) {
8740                 mtx_unlock(&softc->ctl_lock);
8741                 return;
8742         }
8743         mtx_lock(&lun->lun_lock);
8744         mtx_unlock(&softc->ctl_lock);
8745         if (lun->flags & CTL_LUN_DISABLED) {
8746                 mtx_unlock(&lun->lun_lock);
8747                 return;
8748         }
8749         residx = ctl_get_initindex(&msg->hdr.nexus);
8750         switch(msg->pr.pr_info.action) {
8751         case CTL_PR_REG_KEY:
8752                 ctl_alloc_prkey(lun, msg->pr.pr_info.residx);
8753                 if (ctl_get_prkey(lun, msg->pr.pr_info.residx) == 0)
8754                         lun->pr_key_count++;
8755                 ctl_set_prkey(lun, msg->pr.pr_info.residx,
8756                     scsi_8btou64(msg->pr.pr_info.sa_res_key));
8757                 lun->PRGeneration++;
8758                 break;
8759
8760         case CTL_PR_UNREG_KEY:
8761                 ctl_clr_prkey(lun, msg->pr.pr_info.residx);
8762                 lun->pr_key_count--;
8763
8764                 /* XXX Need to see if the reservation has been released */
8765                 /* if so do we need to generate UA? */
8766                 if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8767                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8768                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8769
8770                         if ((lun->res_type == SPR_TYPE_WR_EX_RO
8771                           || lun->res_type == SPR_TYPE_EX_AC_RO)
8772                          && lun->pr_key_count) {
8773                                 /*
8774                                  * If the reservation is a registrants
8775                                  * only type we need to generate a UA
8776                                  * for other registered inits.  The
8777                                  * sense code should be RESERVATIONS
8778                                  * RELEASED
8779                                  */
8780
8781                                 for (i = softc->init_min; i < softc->init_max; i++) {
8782                                         if (ctl_get_prkey(lun, i) == 0)
8783                                                 continue;
8784
8785                                         ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8786                                 }
8787                         }
8788                         lun->res_type = 0;
8789                 } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8790                         if (lun->pr_key_count==0) {
8791                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8792                                 lun->res_type = 0;
8793                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8794                         }
8795                 }
8796                 lun->PRGeneration++;
8797                 break;
8798
8799         case CTL_PR_RESERVE:
8800                 lun->flags |= CTL_LUN_PR_RESERVED;
8801                 lun->res_type = msg->pr.pr_info.res_type;
8802                 lun->pr_res_idx = msg->pr.pr_info.residx;
8803
8804                 break;
8805
8806         case CTL_PR_RELEASE:
8807                 /*
8808                  * if this isn't an exclusive access res generate UA for all
8809                  * other registrants.
8810                  */
8811                 if (lun->res_type != SPR_TYPE_EX_AC
8812                  && lun->res_type != SPR_TYPE_WR_EX) {
8813                         for (i = softc->init_min; i < softc->init_max; i++)
8814                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8815                                         continue;
8816                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8817                 }
8818
8819                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8820                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8821                 lun->res_type = 0;
8822                 break;
8823
8824         case CTL_PR_PREEMPT:
8825                 ctl_pro_preempt_other(lun, msg);
8826                 break;
8827         case CTL_PR_CLEAR:
8828                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8829                 lun->res_type = 0;
8830                 lun->pr_key_count = 0;
8831                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8832
8833                 for (i=0; i < CTL_MAX_INITIATORS; i++) {
8834                         if (ctl_get_prkey(lun, i) == 0)
8835                                 continue;
8836                         ctl_clr_prkey(lun, i);
8837                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8838                 }
8839                 lun->PRGeneration++;
8840                 break;
8841         }
8842
8843         mtx_unlock(&lun->lun_lock);
8844 }
8845
8846 int
8847 ctl_read_write(struct ctl_scsiio *ctsio)
8848 {
8849         struct ctl_lun *lun;
8850         struct ctl_lba_len_flags *lbalen;
8851         uint64_t lba;
8852         uint32_t num_blocks;
8853         int flags, retval;
8854         int isread;
8855
8856         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8857
8858         CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8859
8860         flags = 0;
8861         retval = CTL_RETVAL_COMPLETE;
8862
8863         isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8864               || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8865         switch (ctsio->cdb[0]) {
8866         case READ_6:
8867         case WRITE_6: {
8868                 struct scsi_rw_6 *cdb;
8869
8870                 cdb = (struct scsi_rw_6 *)ctsio->cdb;
8871
8872                 lba = scsi_3btoul(cdb->addr);
8873                 /* only 5 bits are valid in the most significant address byte */
8874                 lba &= 0x1fffff;
8875                 num_blocks = cdb->length;
8876                 /*
8877                  * This is correct according to SBC-2.
8878                  */
8879                 if (num_blocks == 0)
8880                         num_blocks = 256;
8881                 break;
8882         }
8883         case READ_10:
8884         case WRITE_10: {
8885                 struct scsi_rw_10 *cdb;
8886
8887                 cdb = (struct scsi_rw_10 *)ctsio->cdb;
8888                 if (cdb->byte2 & SRW10_FUA)
8889                         flags |= CTL_LLF_FUA;
8890                 if (cdb->byte2 & SRW10_DPO)
8891                         flags |= CTL_LLF_DPO;
8892                 lba = scsi_4btoul(cdb->addr);
8893                 num_blocks = scsi_2btoul(cdb->length);
8894                 break;
8895         }
8896         case WRITE_VERIFY_10: {
8897                 struct scsi_write_verify_10 *cdb;
8898
8899                 cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8900                 flags |= CTL_LLF_FUA;
8901                 if (cdb->byte2 & SWV_DPO)
8902                         flags |= CTL_LLF_DPO;
8903                 lba = scsi_4btoul(cdb->addr);
8904                 num_blocks = scsi_2btoul(cdb->length);
8905                 break;
8906         }
8907         case READ_12:
8908         case WRITE_12: {
8909                 struct scsi_rw_12 *cdb;
8910
8911                 cdb = (struct scsi_rw_12 *)ctsio->cdb;
8912                 if (cdb->byte2 & SRW12_FUA)
8913                         flags |= CTL_LLF_FUA;
8914                 if (cdb->byte2 & SRW12_DPO)
8915                         flags |= CTL_LLF_DPO;
8916                 lba = scsi_4btoul(cdb->addr);
8917                 num_blocks = scsi_4btoul(cdb->length);
8918                 break;
8919         }
8920         case WRITE_VERIFY_12: {
8921                 struct scsi_write_verify_12 *cdb;
8922
8923                 cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8924                 flags |= CTL_LLF_FUA;
8925                 if (cdb->byte2 & SWV_DPO)
8926                         flags |= CTL_LLF_DPO;
8927                 lba = scsi_4btoul(cdb->addr);
8928                 num_blocks = scsi_4btoul(cdb->length);
8929                 break;
8930         }
8931         case READ_16:
8932         case WRITE_16: {
8933                 struct scsi_rw_16 *cdb;
8934
8935                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
8936                 if (cdb->byte2 & SRW12_FUA)
8937                         flags |= CTL_LLF_FUA;
8938                 if (cdb->byte2 & SRW12_DPO)
8939                         flags |= CTL_LLF_DPO;
8940                 lba = scsi_8btou64(cdb->addr);
8941                 num_blocks = scsi_4btoul(cdb->length);
8942                 break;
8943         }
8944         case WRITE_ATOMIC_16: {
8945                 struct scsi_write_atomic_16 *cdb;
8946
8947                 if (lun->be_lun->atomicblock == 0) {
8948                         ctl_set_invalid_opcode(ctsio);
8949                         ctl_done((union ctl_io *)ctsio);
8950                         return (CTL_RETVAL_COMPLETE);
8951                 }
8952
8953                 cdb = (struct scsi_write_atomic_16 *)ctsio->cdb;
8954                 if (cdb->byte2 & SRW12_FUA)
8955                         flags |= CTL_LLF_FUA;
8956                 if (cdb->byte2 & SRW12_DPO)
8957                         flags |= CTL_LLF_DPO;
8958                 lba = scsi_8btou64(cdb->addr);
8959                 num_blocks = scsi_2btoul(cdb->length);
8960                 if (num_blocks > lun->be_lun->atomicblock) {
8961                         ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
8962                             /*command*/ 1, /*field*/ 12, /*bit_valid*/ 0,
8963                             /*bit*/ 0);
8964                         ctl_done((union ctl_io *)ctsio);
8965                         return (CTL_RETVAL_COMPLETE);
8966                 }
8967                 break;
8968         }
8969         case WRITE_VERIFY_16: {
8970                 struct scsi_write_verify_16 *cdb;
8971
8972                 cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
8973                 flags |= CTL_LLF_FUA;
8974                 if (cdb->byte2 & SWV_DPO)
8975                         flags |= CTL_LLF_DPO;
8976                 lba = scsi_8btou64(cdb->addr);
8977                 num_blocks = scsi_4btoul(cdb->length);
8978                 break;
8979         }
8980         default:
8981                 /*
8982                  * We got a command we don't support.  This shouldn't
8983                  * happen, commands should be filtered out above us.
8984                  */
8985                 ctl_set_invalid_opcode(ctsio);
8986                 ctl_done((union ctl_io *)ctsio);
8987
8988                 return (CTL_RETVAL_COMPLETE);
8989                 break; /* NOTREACHED */
8990         }
8991
8992         /*
8993          * The first check is to make sure we're in bounds, the second
8994          * check is to catch wrap-around problems.  If the lba + num blocks
8995          * is less than the lba, then we've wrapped around and the block
8996          * range is invalid anyway.
8997          */
8998         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8999          || ((lba + num_blocks) < lba)) {
9000                 ctl_set_lba_out_of_range(ctsio);
9001                 ctl_done((union ctl_io *)ctsio);
9002                 return (CTL_RETVAL_COMPLETE);
9003         }
9004
9005         /*
9006          * According to SBC-3, a transfer length of 0 is not an error.
9007          * Note that this cannot happen with WRITE(6) or READ(6), since 0
9008          * translates to 256 blocks for those commands.
9009          */
9010         if (num_blocks == 0) {
9011                 ctl_set_success(ctsio);
9012                 ctl_done((union ctl_io *)ctsio);
9013                 return (CTL_RETVAL_COMPLETE);
9014         }
9015
9016         /* Set FUA and/or DPO if caches are disabled. */
9017         if (isread) {
9018                 if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9019                     SCP_RCD) != 0)
9020                         flags |= CTL_LLF_FUA | CTL_LLF_DPO;
9021         } else {
9022                 if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9023                     SCP_WCE) == 0)
9024                         flags |= CTL_LLF_FUA;
9025         }
9026
9027         lbalen = (struct ctl_lba_len_flags *)
9028             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9029         lbalen->lba = lba;
9030         lbalen->len = num_blocks;
9031         lbalen->flags = (isread ? CTL_LLF_READ : CTL_LLF_WRITE) | flags;
9032
9033         ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9034         ctsio->kern_rel_offset = 0;
9035
9036         CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
9037
9038         retval = lun->backend->data_submit((union ctl_io *)ctsio);
9039
9040         return (retval);
9041 }
9042
9043 static int
9044 ctl_cnw_cont(union ctl_io *io)
9045 {
9046         struct ctl_scsiio *ctsio;
9047         struct ctl_lun *lun;
9048         struct ctl_lba_len_flags *lbalen;
9049         int retval;
9050
9051         ctsio = &io->scsiio;
9052         ctsio->io_hdr.status = CTL_STATUS_NONE;
9053         ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
9054         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9055         lbalen = (struct ctl_lba_len_flags *)
9056             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9057         lbalen->flags &= ~CTL_LLF_COMPARE;
9058         lbalen->flags |= CTL_LLF_WRITE;
9059
9060         CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
9061         retval = lun->backend->data_submit((union ctl_io *)ctsio);
9062         return (retval);
9063 }
9064
9065 int
9066 ctl_cnw(struct ctl_scsiio *ctsio)
9067 {
9068         struct ctl_lun *lun;
9069         struct ctl_lba_len_flags *lbalen;
9070         uint64_t lba;
9071         uint32_t num_blocks;
9072         int flags, retval;
9073
9074         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9075
9076         CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
9077
9078         flags = 0;
9079         retval = CTL_RETVAL_COMPLETE;
9080
9081         switch (ctsio->cdb[0]) {
9082         case COMPARE_AND_WRITE: {
9083                 struct scsi_compare_and_write *cdb;
9084
9085                 cdb = (struct scsi_compare_and_write *)ctsio->cdb;
9086                 if (cdb->byte2 & SRW10_FUA)
9087                         flags |= CTL_LLF_FUA;
9088                 if (cdb->byte2 & SRW10_DPO)
9089                         flags |= CTL_LLF_DPO;
9090                 lba = scsi_8btou64(cdb->addr);
9091                 num_blocks = cdb->length;
9092                 break;
9093         }
9094         default:
9095                 /*
9096                  * We got a command we don't support.  This shouldn't
9097                  * happen, commands should be filtered out above us.
9098                  */
9099                 ctl_set_invalid_opcode(ctsio);
9100                 ctl_done((union ctl_io *)ctsio);
9101
9102                 return (CTL_RETVAL_COMPLETE);
9103                 break; /* NOTREACHED */
9104         }
9105
9106         /*
9107          * The first check is to make sure we're in bounds, the second
9108          * check is to catch wrap-around problems.  If the lba + num blocks
9109          * is less than the lba, then we've wrapped around and the block
9110          * range is invalid anyway.
9111          */
9112         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9113          || ((lba + num_blocks) < lba)) {
9114                 ctl_set_lba_out_of_range(ctsio);
9115                 ctl_done((union ctl_io *)ctsio);
9116                 return (CTL_RETVAL_COMPLETE);
9117         }
9118
9119         /*
9120          * According to SBC-3, a transfer length of 0 is not an error.
9121          */
9122         if (num_blocks == 0) {
9123                 ctl_set_success(ctsio);
9124                 ctl_done((union ctl_io *)ctsio);
9125                 return (CTL_RETVAL_COMPLETE);
9126         }
9127
9128         /* Set FUA if write cache is disabled. */
9129         if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9130             SCP_WCE) == 0)
9131                 flags |= CTL_LLF_FUA;
9132
9133         ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
9134         ctsio->kern_rel_offset = 0;
9135
9136         /*
9137          * Set the IO_CONT flag, so that if this I/O gets passed to
9138          * ctl_data_submit_done(), it'll get passed back to
9139          * ctl_ctl_cnw_cont() for further processing.
9140          */
9141         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
9142         ctsio->io_cont = ctl_cnw_cont;
9143
9144         lbalen = (struct ctl_lba_len_flags *)
9145             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9146         lbalen->lba = lba;
9147         lbalen->len = num_blocks;
9148         lbalen->flags = CTL_LLF_COMPARE | flags;
9149
9150         CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
9151         retval = lun->backend->data_submit((union ctl_io *)ctsio);
9152         return (retval);
9153 }
9154
9155 int
9156 ctl_verify(struct ctl_scsiio *ctsio)
9157 {
9158         struct ctl_lun *lun;
9159         struct ctl_lba_len_flags *lbalen;
9160         uint64_t lba;
9161         uint32_t num_blocks;
9162         int bytchk, flags;
9163         int retval;
9164
9165         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9166
9167         CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
9168
9169         bytchk = 0;
9170         flags = CTL_LLF_FUA;
9171         retval = CTL_RETVAL_COMPLETE;
9172
9173         switch (ctsio->cdb[0]) {
9174         case VERIFY_10: {
9175                 struct scsi_verify_10 *cdb;
9176
9177                 cdb = (struct scsi_verify_10 *)ctsio->cdb;
9178                 if (cdb->byte2 & SVFY_BYTCHK)
9179                         bytchk = 1;
9180                 if (cdb->byte2 & SVFY_DPO)
9181                         flags |= CTL_LLF_DPO;
9182                 lba = scsi_4btoul(cdb->addr);
9183                 num_blocks = scsi_2btoul(cdb->length);
9184                 break;
9185         }
9186         case VERIFY_12: {
9187                 struct scsi_verify_12 *cdb;
9188
9189                 cdb = (struct scsi_verify_12 *)ctsio->cdb;
9190                 if (cdb->byte2 & SVFY_BYTCHK)
9191                         bytchk = 1;
9192                 if (cdb->byte2 & SVFY_DPO)
9193                         flags |= CTL_LLF_DPO;
9194                 lba = scsi_4btoul(cdb->addr);
9195                 num_blocks = scsi_4btoul(cdb->length);
9196                 break;
9197         }
9198         case VERIFY_16: {
9199                 struct scsi_rw_16 *cdb;
9200
9201                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
9202                 if (cdb->byte2 & SVFY_BYTCHK)
9203                         bytchk = 1;
9204                 if (cdb->byte2 & SVFY_DPO)
9205                         flags |= CTL_LLF_DPO;
9206                 lba = scsi_8btou64(cdb->addr);
9207                 num_blocks = scsi_4btoul(cdb->length);
9208                 break;
9209         }
9210         default:
9211                 /*
9212                  * We got a command we don't support.  This shouldn't
9213                  * happen, commands should be filtered out above us.
9214                  */
9215                 ctl_set_invalid_opcode(ctsio);
9216                 ctl_done((union ctl_io *)ctsio);
9217                 return (CTL_RETVAL_COMPLETE);
9218         }
9219
9220         /*
9221          * The first check is to make sure we're in bounds, the second
9222          * check is to catch wrap-around problems.  If the lba + num blocks
9223          * is less than the lba, then we've wrapped around and the block
9224          * range is invalid anyway.
9225          */
9226         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9227          || ((lba + num_blocks) < lba)) {
9228                 ctl_set_lba_out_of_range(ctsio);
9229                 ctl_done((union ctl_io *)ctsio);
9230                 return (CTL_RETVAL_COMPLETE);
9231         }
9232
9233         /*
9234          * According to SBC-3, a transfer length of 0 is not an error.
9235          */
9236         if (num_blocks == 0) {
9237                 ctl_set_success(ctsio);
9238                 ctl_done((union ctl_io *)ctsio);
9239                 return (CTL_RETVAL_COMPLETE);
9240         }
9241
9242         lbalen = (struct ctl_lba_len_flags *)
9243             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9244         lbalen->lba = lba;
9245         lbalen->len = num_blocks;
9246         if (bytchk) {
9247                 lbalen->flags = CTL_LLF_COMPARE | flags;
9248                 ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9249         } else {
9250                 lbalen->flags = CTL_LLF_VERIFY | flags;
9251                 ctsio->kern_total_len = 0;
9252         }
9253         ctsio->kern_rel_offset = 0;
9254
9255         CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9256         retval = lun->backend->data_submit((union ctl_io *)ctsio);
9257         return (retval);
9258 }
9259
9260 int
9261 ctl_report_luns(struct ctl_scsiio *ctsio)
9262 {
9263         struct ctl_softc *softc = control_softc;
9264         struct scsi_report_luns *cdb;
9265         struct scsi_report_luns_data *lun_data;
9266         struct ctl_lun *lun, *request_lun;
9267         struct ctl_port *port;
9268         int num_luns, retval;
9269         uint32_t alloc_len, lun_datalen;
9270         int num_filled;
9271         uint32_t initidx, targ_lun_id, lun_id;
9272
9273         retval = CTL_RETVAL_COMPLETE;
9274         cdb = (struct scsi_report_luns *)ctsio->cdb;
9275         port = ctl_io_port(&ctsio->io_hdr);
9276
9277         CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9278
9279         mtx_lock(&softc->ctl_lock);
9280         num_luns = 0;
9281         for (targ_lun_id = 0; targ_lun_id < CTL_MAX_LUNS; targ_lun_id++) {
9282                 if (ctl_lun_map_from_port(port, targ_lun_id) < CTL_MAX_LUNS)
9283                         num_luns++;
9284         }
9285         mtx_unlock(&softc->ctl_lock);
9286
9287         switch (cdb->select_report) {
9288         case RPL_REPORT_DEFAULT:
9289         case RPL_REPORT_ALL:
9290         case RPL_REPORT_NONSUBSID:
9291                 break;
9292         case RPL_REPORT_WELLKNOWN:
9293         case RPL_REPORT_ADMIN:
9294         case RPL_REPORT_CONGLOM:
9295                 num_luns = 0;
9296                 break;
9297         default:
9298                 ctl_set_invalid_field(ctsio,
9299                                       /*sks_valid*/ 1,
9300                                       /*command*/ 1,
9301                                       /*field*/ 2,
9302                                       /*bit_valid*/ 0,
9303                                       /*bit*/ 0);
9304                 ctl_done((union ctl_io *)ctsio);
9305                 return (retval);
9306                 break; /* NOTREACHED */
9307         }
9308
9309         alloc_len = scsi_4btoul(cdb->length);
9310         /*
9311          * The initiator has to allocate at least 16 bytes for this request,
9312          * so he can at least get the header and the first LUN.  Otherwise
9313          * we reject the request (per SPC-3 rev 14, section 6.21).
9314          */
9315         if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9316             sizeof(struct scsi_report_luns_lundata))) {
9317                 ctl_set_invalid_field(ctsio,
9318                                       /*sks_valid*/ 1,
9319                                       /*command*/ 1,
9320                                       /*field*/ 6,
9321                                       /*bit_valid*/ 0,
9322                                       /*bit*/ 0);
9323                 ctl_done((union ctl_io *)ctsio);
9324                 return (retval);
9325         }
9326
9327         request_lun = (struct ctl_lun *)
9328                 ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9329
9330         lun_datalen = sizeof(*lun_data) +
9331                 (num_luns * sizeof(struct scsi_report_luns_lundata));
9332
9333         ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9334         lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9335         ctsio->kern_sg_entries = 0;
9336
9337         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9338
9339         mtx_lock(&softc->ctl_lock);
9340         for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
9341                 lun_id = ctl_lun_map_from_port(port, targ_lun_id);
9342                 if (lun_id >= CTL_MAX_LUNS)
9343                         continue;
9344                 lun = softc->ctl_luns[lun_id];
9345                 if (lun == NULL)
9346                         continue;
9347
9348                 if (targ_lun_id <= 0xff) {
9349                         /*
9350                          * Peripheral addressing method, bus number 0.
9351                          */
9352                         lun_data->luns[num_filled].lundata[0] =
9353                                 RPL_LUNDATA_ATYP_PERIPH;
9354                         lun_data->luns[num_filled].lundata[1] = targ_lun_id;
9355                         num_filled++;
9356                 } else if (targ_lun_id <= 0x3fff) {
9357                         /*
9358                          * Flat addressing method.
9359                          */
9360                         lun_data->luns[num_filled].lundata[0] =
9361                                 RPL_LUNDATA_ATYP_FLAT | (targ_lun_id >> 8);
9362                         lun_data->luns[num_filled].lundata[1] =
9363                                 (targ_lun_id & 0xff);
9364                         num_filled++;
9365                 } else if (targ_lun_id <= 0xffffff) {
9366                         /*
9367                          * Extended flat addressing method.
9368                          */
9369                         lun_data->luns[num_filled].lundata[0] =
9370                             RPL_LUNDATA_ATYP_EXTLUN | 0x12;
9371                         scsi_ulto3b(targ_lun_id,
9372                             &lun_data->luns[num_filled].lundata[1]);
9373                         num_filled++;
9374                 } else {
9375                         printf("ctl_report_luns: bogus LUN number %jd, "
9376                                "skipping\n", (intmax_t)targ_lun_id);
9377                 }
9378                 /*
9379                  * According to SPC-3, rev 14 section 6.21:
9380                  *
9381                  * "The execution of a REPORT LUNS command to any valid and
9382                  * installed logical unit shall clear the REPORTED LUNS DATA
9383                  * HAS CHANGED unit attention condition for all logical
9384                  * units of that target with respect to the requesting
9385                  * initiator. A valid and installed logical unit is one
9386                  * having a PERIPHERAL QUALIFIER of 000b in the standard
9387                  * INQUIRY data (see 6.4.2)."
9388                  *
9389                  * If request_lun is NULL, the LUN this report luns command
9390                  * was issued to is either disabled or doesn't exist. In that
9391                  * case, we shouldn't clear any pending lun change unit
9392                  * attention.
9393                  */
9394                 if (request_lun != NULL) {
9395                         mtx_lock(&lun->lun_lock);
9396                         ctl_clr_ua(lun, initidx, CTL_UA_LUN_CHANGE);
9397                         mtx_unlock(&lun->lun_lock);
9398                 }
9399         }
9400         mtx_unlock(&softc->ctl_lock);
9401
9402         /*
9403          * It's quite possible that we've returned fewer LUNs than we allocated
9404          * space for.  Trim it.
9405          */
9406         lun_datalen = sizeof(*lun_data) +
9407                 (num_filled * sizeof(struct scsi_report_luns_lundata));
9408
9409         if (lun_datalen < alloc_len) {
9410                 ctsio->residual = alloc_len - lun_datalen;
9411                 ctsio->kern_data_len = lun_datalen;
9412                 ctsio->kern_total_len = lun_datalen;
9413         } else {
9414                 ctsio->residual = 0;
9415                 ctsio->kern_data_len = alloc_len;
9416                 ctsio->kern_total_len = alloc_len;
9417         }
9418         ctsio->kern_data_resid = 0;
9419         ctsio->kern_rel_offset = 0;
9420         ctsio->kern_sg_entries = 0;
9421
9422         /*
9423          * We set this to the actual data length, regardless of how much
9424          * space we actually have to return results.  If the user looks at
9425          * this value, he'll know whether or not he allocated enough space
9426          * and reissue the command if necessary.  We don't support well
9427          * known logical units, so if the user asks for that, return none.
9428          */
9429         scsi_ulto4b(lun_datalen - 8, lun_data->length);
9430
9431         /*
9432          * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9433          * this request.
9434          */
9435         ctl_set_success(ctsio);
9436         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9437         ctsio->be_move_done = ctl_config_move_done;
9438         ctl_datamove((union ctl_io *)ctsio);
9439         return (retval);
9440 }
9441
9442 int
9443 ctl_request_sense(struct ctl_scsiio *ctsio)
9444 {
9445         struct scsi_request_sense *cdb;
9446         struct scsi_sense_data *sense_ptr;
9447         struct ctl_softc *ctl_softc;
9448         struct ctl_lun *lun;
9449         uint32_t initidx;
9450         int have_error;
9451         scsi_sense_data_type sense_format;
9452         ctl_ua_type ua_type;
9453
9454         cdb = (struct scsi_request_sense *)ctsio->cdb;
9455
9456         ctl_softc = control_softc;
9457         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9458
9459         CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9460
9461         /*
9462          * Determine which sense format the user wants.
9463          */
9464         if (cdb->byte2 & SRS_DESC)
9465                 sense_format = SSD_TYPE_DESC;
9466         else
9467                 sense_format = SSD_TYPE_FIXED;
9468
9469         ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9470         sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9471         ctsio->kern_sg_entries = 0;
9472
9473         /*
9474          * struct scsi_sense_data, which is currently set to 256 bytes, is
9475          * larger than the largest allowed value for the length field in the
9476          * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9477          */
9478         ctsio->residual = 0;
9479         ctsio->kern_data_len = cdb->length;
9480         ctsio->kern_total_len = cdb->length;
9481
9482         ctsio->kern_data_resid = 0;
9483         ctsio->kern_rel_offset = 0;
9484         ctsio->kern_sg_entries = 0;
9485
9486         /*
9487          * If we don't have a LUN, we don't have any pending sense.
9488          */
9489         if (lun == NULL)
9490                 goto no_sense;
9491
9492         have_error = 0;
9493         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9494         /*
9495          * Check for pending sense, and then for pending unit attentions.
9496          * Pending sense gets returned first, then pending unit attentions.
9497          */
9498         mtx_lock(&lun->lun_lock);
9499 #ifdef CTL_WITH_CA
9500         if (ctl_is_set(lun->have_ca, initidx)) {
9501                 scsi_sense_data_type stored_format;
9502
9503                 /*
9504                  * Check to see which sense format was used for the stored
9505                  * sense data.
9506                  */
9507                 stored_format = scsi_sense_type(&lun->pending_sense[initidx]);
9508
9509                 /*
9510                  * If the user requested a different sense format than the
9511                  * one we stored, then we need to convert it to the other
9512                  * format.  If we're going from descriptor to fixed format
9513                  * sense data, we may lose things in translation, depending
9514                  * on what options were used.
9515                  *
9516                  * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9517                  * for some reason we'll just copy it out as-is.
9518                  */
9519                 if ((stored_format == SSD_TYPE_FIXED)
9520                  && (sense_format == SSD_TYPE_DESC))
9521                         ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9522                             &lun->pending_sense[initidx],
9523                             (struct scsi_sense_data_desc *)sense_ptr);
9524                 else if ((stored_format == SSD_TYPE_DESC)
9525                       && (sense_format == SSD_TYPE_FIXED))
9526                         ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9527                             &lun->pending_sense[initidx],
9528                             (struct scsi_sense_data_fixed *)sense_ptr);
9529                 else
9530                         memcpy(sense_ptr, &lun->pending_sense[initidx],
9531                                MIN(sizeof(*sense_ptr),
9532                                sizeof(lun->pending_sense[initidx])));
9533
9534                 ctl_clear_mask(lun->have_ca, initidx);
9535                 have_error = 1;
9536         } else
9537 #endif
9538         {
9539                 ua_type = ctl_build_ua(lun, initidx, sense_ptr, sense_format);
9540                 if (ua_type != CTL_UA_NONE)
9541                         have_error = 1;
9542                 if (ua_type == CTL_UA_LUN_CHANGE) {
9543                         mtx_unlock(&lun->lun_lock);
9544                         mtx_lock(&ctl_softc->ctl_lock);
9545                         ctl_clr_ua_allluns(ctl_softc, initidx, ua_type);
9546                         mtx_unlock(&ctl_softc->ctl_lock);
9547                         mtx_lock(&lun->lun_lock);
9548                 }
9549
9550         }
9551         mtx_unlock(&lun->lun_lock);
9552
9553         /*
9554          * We already have a pending error, return it.
9555          */
9556         if (have_error != 0) {
9557                 /*
9558                  * We report the SCSI status as OK, since the status of the
9559                  * request sense command itself is OK.
9560                  * We report 0 for the sense length, because we aren't doing
9561                  * autosense in this case.  We're reporting sense as
9562                  * parameter data.
9563                  */
9564                 ctl_set_success(ctsio);
9565                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9566                 ctsio->be_move_done = ctl_config_move_done;
9567                 ctl_datamove((union ctl_io *)ctsio);
9568                 return (CTL_RETVAL_COMPLETE);
9569         }
9570
9571 no_sense:
9572
9573         /*
9574          * No sense information to report, so we report that everything is
9575          * okay.
9576          */
9577         ctl_set_sense_data(sense_ptr,
9578                            lun,
9579                            sense_format,
9580                            /*current_error*/ 1,
9581                            /*sense_key*/ SSD_KEY_NO_SENSE,
9582                            /*asc*/ 0x00,
9583                            /*ascq*/ 0x00,
9584                            SSD_ELEM_NONE);
9585
9586         /*
9587          * We report 0 for the sense length, because we aren't doing
9588          * autosense in this case.  We're reporting sense as parameter data.
9589          */
9590         ctl_set_success(ctsio);
9591         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9592         ctsio->be_move_done = ctl_config_move_done;
9593         ctl_datamove((union ctl_io *)ctsio);
9594         return (CTL_RETVAL_COMPLETE);
9595 }
9596
9597 int
9598 ctl_tur(struct ctl_scsiio *ctsio)
9599 {
9600
9601         CTL_DEBUG_PRINT(("ctl_tur\n"));
9602
9603         ctl_set_success(ctsio);
9604         ctl_done((union ctl_io *)ctsio);
9605
9606         return (CTL_RETVAL_COMPLETE);
9607 }
9608
9609 /*
9610  * SCSI VPD page 0x00, the Supported VPD Pages page.
9611  */
9612 static int
9613 ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9614 {
9615         struct scsi_vpd_supported_pages *pages;
9616         int sup_page_size;
9617         struct ctl_lun *lun;
9618         int p;
9619
9620         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9621
9622         sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9623             SCSI_EVPD_NUM_SUPPORTED_PAGES;
9624         ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9625         pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9626         ctsio->kern_sg_entries = 0;
9627
9628         if (sup_page_size < alloc_len) {
9629                 ctsio->residual = alloc_len - sup_page_size;
9630                 ctsio->kern_data_len = sup_page_size;
9631                 ctsio->kern_total_len = sup_page_size;
9632         } else {
9633                 ctsio->residual = 0;
9634                 ctsio->kern_data_len = alloc_len;
9635                 ctsio->kern_total_len = alloc_len;
9636         }
9637         ctsio->kern_data_resid = 0;
9638         ctsio->kern_rel_offset = 0;
9639         ctsio->kern_sg_entries = 0;
9640
9641         /*
9642          * The control device is always connected.  The disk device, on the
9643          * other hand, may not be online all the time.  Need to change this
9644          * to figure out whether the disk device is actually online or not.
9645          */
9646         if (lun != NULL)
9647                 pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9648                                 lun->be_lun->lun_type;
9649         else
9650                 pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9651
9652         p = 0;
9653         /* Supported VPD pages */
9654         pages->page_list[p++] = SVPD_SUPPORTED_PAGES;
9655         /* Serial Number */
9656         pages->page_list[p++] = SVPD_UNIT_SERIAL_NUMBER;
9657         /* Device Identification */
9658         pages->page_list[p++] = SVPD_DEVICE_ID;
9659         /* Extended INQUIRY Data */
9660         pages->page_list[p++] = SVPD_EXTENDED_INQUIRY_DATA;
9661         /* Mode Page Policy */
9662         pages->page_list[p++] = SVPD_MODE_PAGE_POLICY;
9663         /* SCSI Ports */
9664         pages->page_list[p++] = SVPD_SCSI_PORTS;
9665         /* Third-party Copy */
9666         pages->page_list[p++] = SVPD_SCSI_TPC;
9667         if (lun != NULL && lun->be_lun->lun_type == T_DIRECT) {
9668                 /* Block limits */
9669                 pages->page_list[p++] = SVPD_BLOCK_LIMITS;
9670                 /* Block Device Characteristics */
9671                 pages->page_list[p++] = SVPD_BDC;
9672                 /* Logical Block Provisioning */
9673                 pages->page_list[p++] = SVPD_LBP;
9674         }
9675         pages->length = p;
9676
9677         ctl_set_success(ctsio);
9678         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9679         ctsio->be_move_done = ctl_config_move_done;
9680         ctl_datamove((union ctl_io *)ctsio);
9681         return (CTL_RETVAL_COMPLETE);
9682 }
9683
9684 /*
9685  * SCSI VPD page 0x80, the Unit Serial Number page.
9686  */
9687 static int
9688 ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9689 {
9690         struct scsi_vpd_unit_serial_number *sn_ptr;
9691         struct ctl_lun *lun;
9692         int data_len;
9693
9694         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9695
9696         data_len = 4 + CTL_SN_LEN;
9697         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9698         sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9699         if (data_len < alloc_len) {
9700                 ctsio->residual = alloc_len - data_len;
9701                 ctsio->kern_data_len = data_len;
9702                 ctsio->kern_total_len = data_len;
9703         } else {
9704                 ctsio->residual = 0;
9705                 ctsio->kern_data_len = alloc_len;
9706                 ctsio->kern_total_len = alloc_len;
9707         }
9708         ctsio->kern_data_resid = 0;
9709         ctsio->kern_rel_offset = 0;
9710         ctsio->kern_sg_entries = 0;
9711
9712         /*
9713          * The control device is always connected.  The disk device, on the
9714          * other hand, may not be online all the time.  Need to change this
9715          * to figure out whether the disk device is actually online or not.
9716          */
9717         if (lun != NULL)
9718                 sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9719                                   lun->be_lun->lun_type;
9720         else
9721                 sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9722
9723         sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9724         sn_ptr->length = CTL_SN_LEN;
9725         /*
9726          * If we don't have a LUN, we just leave the serial number as
9727          * all spaces.
9728          */
9729         if (lun != NULL) {
9730                 strncpy((char *)sn_ptr->serial_num,
9731                         (char *)lun->be_lun->serial_num, CTL_SN_LEN);
9732         } else
9733                 memset(sn_ptr->serial_num, 0x20, CTL_SN_LEN);
9734
9735         ctl_set_success(ctsio);
9736         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9737         ctsio->be_move_done = ctl_config_move_done;
9738         ctl_datamove((union ctl_io *)ctsio);
9739         return (CTL_RETVAL_COMPLETE);
9740 }
9741
9742
9743 /*
9744  * SCSI VPD page 0x86, the Extended INQUIRY Data page.
9745  */
9746 static int
9747 ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len)
9748 {
9749         struct scsi_vpd_extended_inquiry_data *eid_ptr;
9750         struct ctl_lun *lun;
9751         int data_len;
9752
9753         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9754
9755         data_len = sizeof(struct scsi_vpd_extended_inquiry_data);
9756         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9757         eid_ptr = (struct scsi_vpd_extended_inquiry_data *)ctsio->kern_data_ptr;
9758         ctsio->kern_sg_entries = 0;
9759
9760         if (data_len < alloc_len) {
9761                 ctsio->residual = alloc_len - data_len;
9762                 ctsio->kern_data_len = data_len;
9763                 ctsio->kern_total_len = data_len;
9764         } else {
9765                 ctsio->residual = 0;
9766                 ctsio->kern_data_len = alloc_len;
9767                 ctsio->kern_total_len = alloc_len;
9768         }
9769         ctsio->kern_data_resid = 0;
9770         ctsio->kern_rel_offset = 0;
9771         ctsio->kern_sg_entries = 0;
9772
9773         /*
9774          * The control device is always connected.  The disk device, on the
9775          * other hand, may not be online all the time.
9776          */
9777         if (lun != NULL)
9778                 eid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9779                                      lun->be_lun->lun_type;
9780         else
9781                 eid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9782         eid_ptr->page_code = SVPD_EXTENDED_INQUIRY_DATA;
9783         scsi_ulto2b(data_len - 4, eid_ptr->page_length);
9784         /*
9785          * We support head of queue, ordered and simple tags.
9786          */
9787         eid_ptr->flags2 = SVPD_EID_HEADSUP | SVPD_EID_ORDSUP | SVPD_EID_SIMPSUP;
9788         /*
9789          * Volatile cache supported.
9790          */
9791         eid_ptr->flags3 = SVPD_EID_V_SUP;
9792
9793         /*
9794          * This means that we clear the REPORTED LUNS DATA HAS CHANGED unit
9795          * attention for a particular IT nexus on all LUNs once we report
9796          * it to that nexus once.  This bit is required as of SPC-4.
9797          */
9798         eid_ptr->flags4 = SVPD_EID_LUICLT;
9799
9800         /*
9801          * XXX KDM in order to correctly answer this, we would need
9802          * information from the SIM to determine how much sense data it
9803          * can send.  So this would really be a path inquiry field, most
9804          * likely.  This can be set to a maximum of 252 according to SPC-4,
9805          * but the hardware may or may not be able to support that much.
9806          * 0 just means that the maximum sense data length is not reported.
9807          */
9808         eid_ptr->max_sense_length = 0;
9809
9810         ctl_set_success(ctsio);
9811         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9812         ctsio->be_move_done = ctl_config_move_done;
9813         ctl_datamove((union ctl_io *)ctsio);
9814         return (CTL_RETVAL_COMPLETE);
9815 }
9816
9817 static int
9818 ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len)
9819 {
9820         struct scsi_vpd_mode_page_policy *mpp_ptr;
9821         struct ctl_lun *lun;
9822         int data_len;
9823
9824         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9825
9826         data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9827             sizeof(struct scsi_vpd_mode_page_policy_descr);
9828
9829         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9830         mpp_ptr = (struct scsi_vpd_mode_page_policy *)ctsio->kern_data_ptr;
9831         ctsio->kern_sg_entries = 0;
9832
9833         if (data_len < alloc_len) {
9834                 ctsio->residual = alloc_len - data_len;
9835                 ctsio->kern_data_len = data_len;
9836                 ctsio->kern_total_len = data_len;
9837         } else {
9838                 ctsio->residual = 0;
9839                 ctsio->kern_data_len = alloc_len;
9840                 ctsio->kern_total_len = alloc_len;
9841         }
9842         ctsio->kern_data_resid = 0;
9843         ctsio->kern_rel_offset = 0;
9844         ctsio->kern_sg_entries = 0;
9845
9846         /*
9847          * The control device is always connected.  The disk device, on the
9848          * other hand, may not be online all the time.
9849          */
9850         if (lun != NULL)
9851                 mpp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9852                                      lun->be_lun->lun_type;
9853         else
9854                 mpp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9855         mpp_ptr->page_code = SVPD_MODE_PAGE_POLICY;
9856         scsi_ulto2b(data_len - 4, mpp_ptr->page_length);
9857         mpp_ptr->descr[0].page_code = 0x3f;
9858         mpp_ptr->descr[0].subpage_code = 0xff;
9859         mpp_ptr->descr[0].policy = SVPD_MPP_SHARED;
9860
9861         ctl_set_success(ctsio);
9862         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9863         ctsio->be_move_done = ctl_config_move_done;
9864         ctl_datamove((union ctl_io *)ctsio);
9865         return (CTL_RETVAL_COMPLETE);
9866 }
9867
9868 /*
9869  * SCSI VPD page 0x83, the Device Identification page.
9870  */
9871 static int
9872 ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9873 {
9874         struct scsi_vpd_device_id *devid_ptr;
9875         struct scsi_vpd_id_descriptor *desc;
9876         struct ctl_softc *softc;
9877         struct ctl_lun *lun;
9878         struct ctl_port *port;
9879         int data_len;
9880         uint8_t proto;
9881
9882         softc = control_softc;
9883
9884         port = ctl_io_port(&ctsio->io_hdr);
9885         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9886
9887         data_len = sizeof(struct scsi_vpd_device_id) +
9888             sizeof(struct scsi_vpd_id_descriptor) +
9889                 sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9890             sizeof(struct scsi_vpd_id_descriptor) +
9891                 sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9892         if (lun && lun->lun_devid)
9893                 data_len += lun->lun_devid->len;
9894         if (port && port->port_devid)
9895                 data_len += port->port_devid->len;
9896         if (port && port->target_devid)
9897                 data_len += port->target_devid->len;
9898
9899         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9900         devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9901         ctsio->kern_sg_entries = 0;
9902
9903         if (data_len < alloc_len) {
9904                 ctsio->residual = alloc_len - data_len;
9905                 ctsio->kern_data_len = data_len;
9906                 ctsio->kern_total_len = data_len;
9907         } else {
9908                 ctsio->residual = 0;
9909                 ctsio->kern_data_len = alloc_len;
9910                 ctsio->kern_total_len = alloc_len;
9911         }
9912         ctsio->kern_data_resid = 0;
9913         ctsio->kern_rel_offset = 0;
9914         ctsio->kern_sg_entries = 0;
9915
9916         /*
9917          * The control device is always connected.  The disk device, on the
9918          * other hand, may not be online all the time.
9919          */
9920         if (lun != NULL)
9921                 devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9922                                      lun->be_lun->lun_type;
9923         else
9924                 devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9925         devid_ptr->page_code = SVPD_DEVICE_ID;
9926         scsi_ulto2b(data_len - 4, devid_ptr->length);
9927
9928         if (port && port->port_type == CTL_PORT_FC)
9929                 proto = SCSI_PROTO_FC << 4;
9930         else if (port && port->port_type == CTL_PORT_ISCSI)
9931                 proto = SCSI_PROTO_ISCSI << 4;
9932         else
9933                 proto = SCSI_PROTO_SPI << 4;
9934         desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9935
9936         /*
9937          * We're using a LUN association here.  i.e., this device ID is a
9938          * per-LUN identifier.
9939          */
9940         if (lun && lun->lun_devid) {
9941                 memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
9942                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9943                     lun->lun_devid->len);
9944         }
9945
9946         /*
9947          * This is for the WWPN which is a port association.
9948          */
9949         if (port && port->port_devid) {
9950                 memcpy(desc, port->port_devid->data, port->port_devid->len);
9951                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9952                     port->port_devid->len);
9953         }
9954
9955         /*
9956          * This is for the Relative Target Port(type 4h) identifier
9957          */
9958         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9959         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9960             SVPD_ID_TYPE_RELTARG;
9961         desc->length = 4;
9962         scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
9963         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9964             sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9965
9966         /*
9967          * This is for the Target Port Group(type 5h) identifier
9968          */
9969         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9970         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9971             SVPD_ID_TYPE_TPORTGRP;
9972         desc->length = 4;
9973         scsi_ulto2b(ctsio->io_hdr.nexus.targ_port / softc->port_cnt + 1,
9974             &desc->identifier[2]);
9975         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9976             sizeof(struct scsi_vpd_id_trgt_port_grp_id));
9977
9978         /*
9979          * This is for the Target identifier
9980          */
9981         if (port && port->target_devid) {
9982                 memcpy(desc, port->target_devid->data, port->target_devid->len);
9983         }
9984
9985         ctl_set_success(ctsio);
9986         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9987         ctsio->be_move_done = ctl_config_move_done;
9988         ctl_datamove((union ctl_io *)ctsio);
9989         return (CTL_RETVAL_COMPLETE);
9990 }
9991
9992 static int
9993 ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
9994 {
9995         struct ctl_softc *softc = control_softc;
9996         struct scsi_vpd_scsi_ports *sp;
9997         struct scsi_vpd_port_designation *pd;
9998         struct scsi_vpd_port_designation_cont *pdc;
9999         struct ctl_lun *lun;
10000         struct ctl_port *port;
10001         int data_len, num_target_ports, iid_len, id_len;
10002
10003         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10004
10005         num_target_ports = 0;
10006         iid_len = 0;
10007         id_len = 0;
10008         mtx_lock(&softc->ctl_lock);
10009         STAILQ_FOREACH(port, &softc->port_list, links) {
10010                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
10011                         continue;
10012                 if (lun != NULL &&
10013                     ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
10014                         continue;
10015                 num_target_ports++;
10016                 if (port->init_devid)
10017                         iid_len += port->init_devid->len;
10018                 if (port->port_devid)
10019                         id_len += port->port_devid->len;
10020         }
10021         mtx_unlock(&softc->ctl_lock);
10022
10023         data_len = sizeof(struct scsi_vpd_scsi_ports) +
10024             num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
10025              sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
10026         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10027         sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
10028         ctsio->kern_sg_entries = 0;
10029
10030         if (data_len < alloc_len) {
10031                 ctsio->residual = alloc_len - data_len;
10032                 ctsio->kern_data_len = data_len;
10033                 ctsio->kern_total_len = data_len;
10034         } else {
10035                 ctsio->residual = 0;
10036                 ctsio->kern_data_len = alloc_len;
10037                 ctsio->kern_total_len = alloc_len;
10038         }
10039         ctsio->kern_data_resid = 0;
10040         ctsio->kern_rel_offset = 0;
10041         ctsio->kern_sg_entries = 0;
10042
10043         /*
10044          * The control device is always connected.  The disk device, on the
10045          * other hand, may not be online all the time.  Need to change this
10046          * to figure out whether the disk device is actually online or not.
10047          */
10048         if (lun != NULL)
10049                 sp->device = (SID_QUAL_LU_CONNECTED << 5) |
10050                                   lun->be_lun->lun_type;
10051         else
10052                 sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10053
10054         sp->page_code = SVPD_SCSI_PORTS;
10055         scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
10056             sp->page_length);
10057         pd = &sp->design[0];
10058
10059         mtx_lock(&softc->ctl_lock);
10060         STAILQ_FOREACH(port, &softc->port_list, links) {
10061                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
10062                         continue;
10063                 if (lun != NULL &&
10064                     ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
10065                         continue;
10066                 scsi_ulto2b(port->targ_port, pd->relative_port_id);
10067                 if (port->init_devid) {
10068                         iid_len = port->init_devid->len;
10069                         memcpy(pd->initiator_transportid,
10070                             port->init_devid->data, port->init_devid->len);
10071                 } else
10072                         iid_len = 0;
10073                 scsi_ulto2b(iid_len, pd->initiator_transportid_length);
10074                 pdc = (struct scsi_vpd_port_designation_cont *)
10075                     (&pd->initiator_transportid[iid_len]);
10076                 if (port->port_devid) {
10077                         id_len = port->port_devid->len;
10078                         memcpy(pdc->target_port_descriptors,
10079                             port->port_devid->data, port->port_devid->len);
10080                 } else
10081                         id_len = 0;
10082                 scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
10083                 pd = (struct scsi_vpd_port_designation *)
10084                     ((uint8_t *)pdc->target_port_descriptors + id_len);
10085         }
10086         mtx_unlock(&softc->ctl_lock);
10087
10088         ctl_set_success(ctsio);
10089         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10090         ctsio->be_move_done = ctl_config_move_done;
10091         ctl_datamove((union ctl_io *)ctsio);
10092         return (CTL_RETVAL_COMPLETE);
10093 }
10094
10095 static int
10096 ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
10097 {
10098         struct scsi_vpd_block_limits *bl_ptr;
10099         struct ctl_lun *lun;
10100         int bs;
10101
10102         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10103
10104         ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
10105         bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
10106         ctsio->kern_sg_entries = 0;
10107
10108         if (sizeof(*bl_ptr) < alloc_len) {
10109                 ctsio->residual = alloc_len - sizeof(*bl_ptr);
10110                 ctsio->kern_data_len = sizeof(*bl_ptr);
10111                 ctsio->kern_total_len = sizeof(*bl_ptr);
10112         } else {
10113                 ctsio->residual = 0;
10114                 ctsio->kern_data_len = alloc_len;
10115                 ctsio->kern_total_len = alloc_len;
10116         }
10117         ctsio->kern_data_resid = 0;
10118         ctsio->kern_rel_offset = 0;
10119         ctsio->kern_sg_entries = 0;
10120
10121         /*
10122          * The control device is always connected.  The disk device, on the
10123          * other hand, may not be online all the time.  Need to change this
10124          * to figure out whether the disk device is actually online or not.
10125          */
10126         if (lun != NULL)
10127                 bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10128                                   lun->be_lun->lun_type;
10129         else
10130                 bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10131
10132         bl_ptr->page_code = SVPD_BLOCK_LIMITS;
10133         scsi_ulto2b(sizeof(*bl_ptr) - 4, bl_ptr->page_length);
10134         bl_ptr->max_cmp_write_len = 0xff;
10135         scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
10136         if (lun != NULL) {
10137                 bs = lun->be_lun->blocksize;
10138                 scsi_ulto4b(lun->be_lun->opttxferlen, bl_ptr->opt_txfer_len);
10139                 if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10140                         scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_lba_cnt);
10141                         scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_blk_cnt);
10142                         if (lun->be_lun->ublockexp != 0) {
10143                                 scsi_ulto4b((1 << lun->be_lun->ublockexp),
10144                                     bl_ptr->opt_unmap_grain);
10145                                 scsi_ulto4b(0x80000000 | lun->be_lun->ublockoff,
10146                                     bl_ptr->unmap_grain_align);
10147                         }
10148                 }
10149                 scsi_ulto4b(lun->be_lun->atomicblock,
10150                     bl_ptr->max_atomic_transfer_length);
10151                 scsi_ulto4b(0, bl_ptr->atomic_alignment);
10152                 scsi_ulto4b(0, bl_ptr->atomic_transfer_length_granularity);
10153                 scsi_ulto4b(0, bl_ptr->max_atomic_transfer_length_with_atomic_boundary);
10154                 scsi_ulto4b(0, bl_ptr->max_atomic_boundary_size);
10155         }
10156         scsi_u64to8b(UINT64_MAX, bl_ptr->max_write_same_length);
10157
10158         ctl_set_success(ctsio);
10159         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10160         ctsio->be_move_done = ctl_config_move_done;
10161         ctl_datamove((union ctl_io *)ctsio);
10162         return (CTL_RETVAL_COMPLETE);
10163 }
10164
10165 static int
10166 ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len)
10167 {
10168         struct scsi_vpd_block_device_characteristics *bdc_ptr;
10169         struct ctl_lun *lun;
10170         const char *value;
10171         u_int i;
10172
10173         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10174
10175         ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO);
10176         bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr;
10177         ctsio->kern_sg_entries = 0;
10178
10179         if (sizeof(*bdc_ptr) < alloc_len) {
10180                 ctsio->residual = alloc_len - sizeof(*bdc_ptr);
10181                 ctsio->kern_data_len = sizeof(*bdc_ptr);
10182                 ctsio->kern_total_len = sizeof(*bdc_ptr);
10183         } else {
10184                 ctsio->residual = 0;
10185                 ctsio->kern_data_len = alloc_len;
10186                 ctsio->kern_total_len = alloc_len;
10187         }
10188         ctsio->kern_data_resid = 0;
10189         ctsio->kern_rel_offset = 0;
10190         ctsio->kern_sg_entries = 0;
10191
10192         /*
10193          * The control device is always connected.  The disk device, on the
10194          * other hand, may not be online all the time.  Need to change this
10195          * to figure out whether the disk device is actually online or not.
10196          */
10197         if (lun != NULL)
10198                 bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10199                                   lun->be_lun->lun_type;
10200         else
10201                 bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10202         bdc_ptr->page_code = SVPD_BDC;
10203         scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length);
10204         if (lun != NULL &&
10205             (value = ctl_get_opt(&lun->be_lun->options, "rpm")) != NULL)
10206                 i = strtol(value, NULL, 0);
10207         else
10208                 i = CTL_DEFAULT_ROTATION_RATE;
10209         scsi_ulto2b(i, bdc_ptr->medium_rotation_rate);
10210         if (lun != NULL &&
10211             (value = ctl_get_opt(&lun->be_lun->options, "formfactor")) != NULL)
10212                 i = strtol(value, NULL, 0);
10213         else
10214                 i = 0;
10215         bdc_ptr->wab_wac_ff = (i & 0x0f);
10216         bdc_ptr->flags = SVPD_FUAB | SVPD_VBULS;
10217
10218         ctl_set_success(ctsio);
10219         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10220         ctsio->be_move_done = ctl_config_move_done;
10221         ctl_datamove((union ctl_io *)ctsio);
10222         return (CTL_RETVAL_COMPLETE);
10223 }
10224
10225 static int
10226 ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
10227 {
10228         struct scsi_vpd_logical_block_prov *lbp_ptr;
10229         struct ctl_lun *lun;
10230
10231         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10232
10233         ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
10234         lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
10235         ctsio->kern_sg_entries = 0;
10236
10237         if (sizeof(*lbp_ptr) < alloc_len) {
10238                 ctsio->residual = alloc_len - sizeof(*lbp_ptr);
10239                 ctsio->kern_data_len = sizeof(*lbp_ptr);
10240                 ctsio->kern_total_len = sizeof(*lbp_ptr);
10241         } else {
10242                 ctsio->residual = 0;
10243                 ctsio->kern_data_len = alloc_len;
10244                 ctsio->kern_total_len = alloc_len;
10245         }
10246         ctsio->kern_data_resid = 0;
10247         ctsio->kern_rel_offset = 0;
10248         ctsio->kern_sg_entries = 0;
10249
10250         /*
10251          * The control device is always connected.  The disk device, on the
10252          * other hand, may not be online all the time.  Need to change this
10253          * to figure out whether the disk device is actually online or not.
10254          */
10255         if (lun != NULL)
10256                 lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10257                                   lun->be_lun->lun_type;
10258         else
10259                 lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10260
10261         lbp_ptr->page_code = SVPD_LBP;
10262         scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length);
10263         lbp_ptr->threshold_exponent = CTL_LBP_EXPONENT;
10264         if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10265                 lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 |
10266                     SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP;
10267                 lbp_ptr->prov_type = SVPD_LBP_THIN;
10268         }
10269
10270         ctl_set_success(ctsio);
10271         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10272         ctsio->be_move_done = ctl_config_move_done;
10273         ctl_datamove((union ctl_io *)ctsio);
10274         return (CTL_RETVAL_COMPLETE);
10275 }
10276
10277 /*
10278  * INQUIRY with the EVPD bit set.
10279  */
10280 static int
10281 ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
10282 {
10283         struct ctl_lun *lun;
10284         struct scsi_inquiry *cdb;
10285         int alloc_len, retval;
10286
10287         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10288         cdb = (struct scsi_inquiry *)ctsio->cdb;
10289         alloc_len = scsi_2btoul(cdb->length);
10290
10291         switch (cdb->page_code) {
10292         case SVPD_SUPPORTED_PAGES:
10293                 retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
10294                 break;
10295         case SVPD_UNIT_SERIAL_NUMBER:
10296                 retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
10297                 break;
10298         case SVPD_DEVICE_ID:
10299                 retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
10300                 break;
10301         case SVPD_EXTENDED_INQUIRY_DATA:
10302                 retval = ctl_inquiry_evpd_eid(ctsio, alloc_len);
10303                 break;
10304         case SVPD_MODE_PAGE_POLICY:
10305                 retval = ctl_inquiry_evpd_mpp(ctsio, alloc_len);
10306                 break;
10307         case SVPD_SCSI_PORTS:
10308                 retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
10309                 break;
10310         case SVPD_SCSI_TPC:
10311                 retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
10312                 break;
10313         case SVPD_BLOCK_LIMITS:
10314                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10315                         goto err;
10316                 retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
10317                 break;
10318         case SVPD_BDC:
10319                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10320                         goto err;
10321                 retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len);
10322                 break;
10323         case SVPD_LBP:
10324                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10325                         goto err;
10326                 retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
10327                 break;
10328         default:
10329 err:
10330                 ctl_set_invalid_field(ctsio,
10331                                       /*sks_valid*/ 1,
10332                                       /*command*/ 1,
10333                                       /*field*/ 2,
10334                                       /*bit_valid*/ 0,
10335                                       /*bit*/ 0);
10336                 ctl_done((union ctl_io *)ctsio);
10337                 retval = CTL_RETVAL_COMPLETE;
10338                 break;
10339         }
10340
10341         return (retval);
10342 }
10343
10344 /*
10345  * Standard INQUIRY data.
10346  */
10347 static int
10348 ctl_inquiry_std(struct ctl_scsiio *ctsio)
10349 {
10350         struct scsi_inquiry_data *inq_ptr;
10351         struct scsi_inquiry *cdb;
10352         struct ctl_softc *softc;
10353         struct ctl_port *port;
10354         struct ctl_lun *lun;
10355         char *val;
10356         uint32_t alloc_len, data_len;
10357         ctl_port_type port_type;
10358
10359         softc = control_softc;
10360
10361         /*
10362          * Figure out whether we're talking to a Fibre Channel port or not.
10363          * We treat the ioctl front end, and any SCSI adapters, as packetized
10364          * SCSI front ends.
10365          */
10366         port = ctl_io_port(&ctsio->io_hdr);
10367         if (port != NULL)
10368                 port_type = port->port_type;
10369         else
10370                 port_type = CTL_PORT_SCSI;
10371         if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
10372                 port_type = CTL_PORT_SCSI;
10373
10374         lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10375         cdb = (struct scsi_inquiry *)ctsio->cdb;
10376         alloc_len = scsi_2btoul(cdb->length);
10377
10378         /*
10379          * We malloc the full inquiry data size here and fill it
10380          * in.  If the user only asks for less, we'll give him
10381          * that much.
10382          */
10383         data_len = offsetof(struct scsi_inquiry_data, vendor_specific1);
10384         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10385         inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
10386         ctsio->kern_sg_entries = 0;
10387         ctsio->kern_data_resid = 0;
10388         ctsio->kern_rel_offset = 0;
10389
10390         if (data_len < alloc_len) {
10391                 ctsio->residual = alloc_len - data_len;
10392                 ctsio->kern_data_len = data_len;
10393                 ctsio->kern_total_len = data_len;
10394         } else {
10395                 ctsio->residual = 0;
10396                 ctsio->kern_data_len = alloc_len;
10397                 ctsio->kern_total_len = alloc_len;
10398         }
10399
10400         if (lun != NULL) {
10401                 if ((lun->flags & CTL_LUN_PRIMARY_SC) ||
10402                     softc->ha_link >= CTL_HA_LINK_UNKNOWN) {
10403                         inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10404                             lun->be_lun->lun_type;
10405                 } else {
10406                         inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) |
10407                             lun->be_lun->lun_type;
10408                 }
10409         } else
10410                 inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10411
10412         /* RMB in byte 2 is 0 */
10413         inq_ptr->version = SCSI_REV_SPC4;
10414
10415         /*
10416          * According to SAM-3, even if a device only supports a single
10417          * level of LUN addressing, it should still set the HISUP bit:
10418          *
10419          * 4.9.1 Logical unit numbers overview
10420          *
10421          * All logical unit number formats described in this standard are
10422          * hierarchical in structure even when only a single level in that
10423          * hierarchy is used. The HISUP bit shall be set to one in the
10424          * standard INQUIRY data (see SPC-2) when any logical unit number
10425          * format described in this standard is used.  Non-hierarchical
10426          * formats are outside the scope of this standard.
10427          *
10428          * Therefore we set the HiSup bit here.
10429          *
10430          * The reponse format is 2, per SPC-3.
10431          */
10432         inq_ptr->response_format = SID_HiSup | 2;
10433
10434         inq_ptr->additional_length = data_len -
10435             (offsetof(struct scsi_inquiry_data, additional_length) + 1);
10436         CTL_DEBUG_PRINT(("additional_length = %d\n",
10437                          inq_ptr->additional_length));
10438
10439         inq_ptr->spc3_flags = SPC3_SID_3PC | SPC3_SID_TPGS_IMPLICIT;
10440         /* 16 bit addressing */
10441         if (port_type == CTL_PORT_SCSI)
10442                 inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10443         /* XXX set the SID_MultiP bit here if we're actually going to
10444            respond on multiple ports */
10445         inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10446
10447         /* 16 bit data bus, synchronous transfers */
10448         if (port_type == CTL_PORT_SCSI)
10449                 inq_ptr->flags = SID_WBus16 | SID_Sync;
10450         /*
10451          * XXX KDM do we want to support tagged queueing on the control
10452          * device at all?
10453          */
10454         if ((lun == NULL)
10455          || (lun->be_lun->lun_type != T_PROCESSOR))
10456                 inq_ptr->flags |= SID_CmdQue;
10457         /*
10458          * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10459          * We have 8 bytes for the vendor name, and 16 bytes for the device
10460          * name and 4 bytes for the revision.
10461          */
10462         if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10463             "vendor")) == NULL) {
10464                 strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
10465         } else {
10466                 memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10467                 strncpy(inq_ptr->vendor, val,
10468                     min(sizeof(inq_ptr->vendor), strlen(val)));
10469         }
10470         if (lun == NULL) {
10471                 strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10472                     sizeof(inq_ptr->product));
10473         } else if ((val = ctl_get_opt(&lun->be_lun->options, "product")) == NULL) {
10474                 switch (lun->be_lun->lun_type) {
10475                 case T_DIRECT:
10476                         strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10477                             sizeof(inq_ptr->product));
10478                         break;
10479                 case T_PROCESSOR:
10480                         strncpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT,
10481                             sizeof(inq_ptr->product));
10482                         break;
10483                 default:
10484                         strncpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT,
10485                             sizeof(inq_ptr->product));
10486                         break;
10487                 }
10488         } else {
10489                 memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10490                 strncpy(inq_ptr->product, val,
10491                     min(sizeof(inq_ptr->product), strlen(val)));
10492         }
10493
10494         /*
10495          * XXX make this a macro somewhere so it automatically gets
10496          * incremented when we make changes.
10497          */
10498         if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10499             "revision")) == NULL) {
10500                 strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10501         } else {
10502                 memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10503                 strncpy(inq_ptr->revision, val,
10504                     min(sizeof(inq_ptr->revision), strlen(val)));
10505         }
10506
10507         /*
10508          * For parallel SCSI, we support double transition and single
10509          * transition clocking.  We also support QAS (Quick Arbitration
10510          * and Selection) and Information Unit transfers on both the
10511          * control and array devices.
10512          */
10513         if (port_type == CTL_PORT_SCSI)
10514                 inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10515                                     SID_SPI_IUS;
10516
10517         /* SAM-5 (no version claimed) */
10518         scsi_ulto2b(0x00A0, inq_ptr->version1);
10519         /* SPC-4 (no version claimed) */
10520         scsi_ulto2b(0x0460, inq_ptr->version2);
10521         if (port_type == CTL_PORT_FC) {
10522                 /* FCP-2 ANSI INCITS.350:2003 */
10523                 scsi_ulto2b(0x0917, inq_ptr->version3);
10524         } else if (port_type == CTL_PORT_SCSI) {
10525                 /* SPI-4 ANSI INCITS.362:200x */
10526                 scsi_ulto2b(0x0B56, inq_ptr->version3);
10527         } else if (port_type == CTL_PORT_ISCSI) {
10528                 /* iSCSI (no version claimed) */
10529                 scsi_ulto2b(0x0960, inq_ptr->version3);
10530         } else if (port_type == CTL_PORT_SAS) {
10531                 /* SAS (no version claimed) */
10532                 scsi_ulto2b(0x0BE0, inq_ptr->version3);
10533         }
10534
10535         if (lun == NULL) {
10536                 /* SBC-4 (no version claimed) */
10537                 scsi_ulto2b(0x0600, inq_ptr->version4);
10538         } else {
10539                 switch (lun->be_lun->lun_type) {
10540                 case T_DIRECT:
10541                         /* SBC-4 (no version claimed) */
10542                         scsi_ulto2b(0x0600, inq_ptr->version4);
10543                         break;
10544                 case T_PROCESSOR:
10545                 default:
10546                         break;
10547                 }
10548         }
10549
10550         ctl_set_success(ctsio);
10551         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10552         ctsio->be_move_done = ctl_config_move_done;
10553         ctl_datamove((union ctl_io *)ctsio);
10554         return (CTL_RETVAL_COMPLETE);
10555 }
10556
10557 int
10558 ctl_inquiry(struct ctl_scsiio *ctsio)
10559 {
10560         struct scsi_inquiry *cdb;
10561         int retval;
10562
10563         CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10564
10565         cdb = (struct scsi_inquiry *)ctsio->cdb;
10566         if (cdb->byte2 & SI_EVPD)
10567                 retval = ctl_inquiry_evpd(ctsio);
10568         else if (cdb->page_code == 0)
10569                 retval = ctl_inquiry_std(ctsio);
10570         else {
10571                 ctl_set_invalid_field(ctsio,
10572                                       /*sks_valid*/ 1,
10573                                       /*command*/ 1,
10574                                       /*field*/ 2,
10575                                       /*bit_valid*/ 0,
10576                                       /*bit*/ 0);
10577                 ctl_done((union ctl_io *)ctsio);
10578                 return (CTL_RETVAL_COMPLETE);
10579         }
10580
10581         return (retval);
10582 }
10583
10584 /*
10585  * For known CDB types, parse the LBA and length.
10586  */
10587 static int
10588 ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len)
10589 {
10590         if (io->io_hdr.io_type != CTL_IO_SCSI)
10591                 return (1);
10592
10593         switch (io->scsiio.cdb[0]) {
10594         case COMPARE_AND_WRITE: {
10595                 struct scsi_compare_and_write *cdb;
10596
10597                 cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10598
10599                 *lba = scsi_8btou64(cdb->addr);
10600                 *len = cdb->length;
10601                 break;
10602         }
10603         case READ_6:
10604         case WRITE_6: {
10605                 struct scsi_rw_6 *cdb;
10606
10607                 cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10608
10609                 *lba = scsi_3btoul(cdb->addr);
10610                 /* only 5 bits are valid in the most significant address byte */
10611                 *lba &= 0x1fffff;
10612                 *len = cdb->length;
10613                 break;
10614         }
10615         case READ_10:
10616         case WRITE_10: {
10617                 struct scsi_rw_10 *cdb;
10618
10619                 cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10620
10621                 *lba = scsi_4btoul(cdb->addr);
10622                 *len = scsi_2btoul(cdb->length);
10623                 break;
10624         }
10625         case WRITE_VERIFY_10: {
10626                 struct scsi_write_verify_10 *cdb;
10627
10628                 cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10629
10630                 *lba = scsi_4btoul(cdb->addr);
10631                 *len = scsi_2btoul(cdb->length);
10632                 break;
10633         }
10634         case READ_12:
10635         case WRITE_12: {
10636                 struct scsi_rw_12 *cdb;
10637
10638                 cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10639
10640                 *lba = scsi_4btoul(cdb->addr);
10641                 *len = scsi_4btoul(cdb->length);
10642                 break;
10643         }
10644         case WRITE_VERIFY_12: {
10645                 struct scsi_write_verify_12 *cdb;
10646
10647                 cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10648
10649                 *lba = scsi_4btoul(cdb->addr);
10650                 *len = scsi_4btoul(cdb->length);
10651                 break;
10652         }
10653         case READ_16:
10654         case WRITE_16: {
10655                 struct scsi_rw_16 *cdb;
10656
10657                 cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10658
10659                 *lba = scsi_8btou64(cdb->addr);
10660                 *len = scsi_4btoul(cdb->length);
10661                 break;
10662         }
10663         case WRITE_ATOMIC_16: {
10664                 struct scsi_write_atomic_16 *cdb;
10665
10666                 cdb = (struct scsi_write_atomic_16 *)io->scsiio.cdb;
10667
10668                 *lba = scsi_8btou64(cdb->addr);
10669                 *len = scsi_2btoul(cdb->length);
10670                 break;
10671         }
10672         case WRITE_VERIFY_16: {
10673                 struct scsi_write_verify_16 *cdb;
10674
10675                 cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10676
10677                 *lba = scsi_8btou64(cdb->addr);
10678                 *len = scsi_4btoul(cdb->length);
10679                 break;
10680         }
10681         case WRITE_SAME_10: {
10682                 struct scsi_write_same_10 *cdb;
10683
10684                 cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10685
10686                 *lba = scsi_4btoul(cdb->addr);
10687                 *len = scsi_2btoul(cdb->length);
10688                 break;
10689         }
10690         case WRITE_SAME_16: {
10691                 struct scsi_write_same_16 *cdb;
10692
10693                 cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10694
10695                 *lba = scsi_8btou64(cdb->addr);
10696                 *len = scsi_4btoul(cdb->length);
10697                 break;
10698         }
10699         case VERIFY_10: {
10700                 struct scsi_verify_10 *cdb;
10701
10702                 cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10703
10704                 *lba = scsi_4btoul(cdb->addr);
10705                 *len = scsi_2btoul(cdb->length);
10706                 break;
10707         }
10708         case VERIFY_12: {
10709                 struct scsi_verify_12 *cdb;
10710
10711                 cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10712
10713                 *lba = scsi_4btoul(cdb->addr);
10714                 *len = scsi_4btoul(cdb->length);
10715                 break;
10716         }
10717         case VERIFY_16: {
10718                 struct scsi_verify_16 *cdb;
10719
10720                 cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10721
10722                 *lba = scsi_8btou64(cdb->addr);
10723                 *len = scsi_4btoul(cdb->length);
10724                 break;
10725         }
10726         case UNMAP: {
10727                 *lba = 0;
10728                 *len = UINT64_MAX;
10729                 break;
10730         }
10731         case SERVICE_ACTION_IN: {       /* GET LBA STATUS */
10732                 struct scsi_get_lba_status *cdb;
10733
10734                 cdb = (struct scsi_get_lba_status *)io->scsiio.cdb;
10735                 *lba = scsi_8btou64(cdb->addr);
10736                 *len = UINT32_MAX;
10737                 break;
10738         }
10739         default:
10740                 return (1);
10741                 break; /* NOTREACHED */
10742         }
10743
10744         return (0);
10745 }
10746
10747 static ctl_action
10748 ctl_extent_check_lba(uint64_t lba1, uint64_t len1, uint64_t lba2, uint64_t len2,
10749     bool seq)
10750 {
10751         uint64_t endlba1, endlba2;
10752
10753         endlba1 = lba1 + len1 - (seq ? 0 : 1);
10754         endlba2 = lba2 + len2 - 1;
10755
10756         if ((endlba1 < lba2) || (endlba2 < lba1))
10757                 return (CTL_ACTION_PASS);
10758         else
10759                 return (CTL_ACTION_BLOCK);
10760 }
10761
10762 static int
10763 ctl_extent_check_unmap(union ctl_io *io, uint64_t lba2, uint64_t len2)
10764 {
10765         struct ctl_ptr_len_flags *ptrlen;
10766         struct scsi_unmap_desc *buf, *end, *range;
10767         uint64_t lba;
10768         uint32_t len;
10769
10770         /* If not UNMAP -- go other way. */
10771         if (io->io_hdr.io_type != CTL_IO_SCSI ||
10772             io->scsiio.cdb[0] != UNMAP)
10773                 return (CTL_ACTION_ERROR);
10774
10775         /* If UNMAP without data -- block and wait for data. */
10776         ptrlen = (struct ctl_ptr_len_flags *)
10777             &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
10778         if ((io->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0 ||
10779             ptrlen->ptr == NULL)
10780                 return (CTL_ACTION_BLOCK);
10781
10782         /* UNMAP with data -- check for collision. */
10783         buf = (struct scsi_unmap_desc *)ptrlen->ptr;
10784         end = buf + ptrlen->len / sizeof(*buf);
10785         for (range = buf; range < end; range++) {
10786                 lba = scsi_8btou64(range->lba);
10787                 len = scsi_4btoul(range->length);
10788                 if ((lba < lba2 + len2) && (lba + len > lba2))
10789                         return (CTL_ACTION_BLOCK);
10790         }
10791         return (CTL_ACTION_PASS);
10792 }
10793
10794 static ctl_action
10795 ctl_extent_check(union ctl_io *io1, union ctl_io *io2, bool seq)
10796 {
10797         uint64_t lba1, lba2;
10798         uint64_t len1, len2;
10799         int retval;
10800
10801         if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10802                 return (CTL_ACTION_ERROR);
10803
10804         retval = ctl_extent_check_unmap(io1, lba2, len2);
10805         if (retval != CTL_ACTION_ERROR)
10806                 return (retval);
10807
10808         if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10809                 return (CTL_ACTION_ERROR);
10810
10811         if (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE)
10812                 seq = FALSE;
10813         return (ctl_extent_check_lba(lba1, len1, lba2, len2, seq));
10814 }
10815
10816 static ctl_action
10817 ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2)
10818 {
10819         uint64_t lba1, lba2;
10820         uint64_t len1, len2;
10821
10822         if (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE)
10823                 return (CTL_ACTION_PASS);
10824         if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10825                 return (CTL_ACTION_ERROR);
10826         if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10827                 return (CTL_ACTION_ERROR);
10828
10829         if (lba1 + len1 == lba2)
10830                 return (CTL_ACTION_BLOCK);
10831         return (CTL_ACTION_PASS);
10832 }
10833
10834 static ctl_action
10835 ctl_check_for_blockage(struct ctl_lun *lun, union ctl_io *pending_io,
10836     union ctl_io *ooa_io)
10837 {
10838         const struct ctl_cmd_entry *pending_entry, *ooa_entry;
10839         ctl_serialize_action *serialize_row;
10840
10841         /*
10842          * The initiator attempted multiple untagged commands at the same
10843          * time.  Can't do that.
10844          */
10845         if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10846          && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10847          && ((pending_io->io_hdr.nexus.targ_port ==
10848               ooa_io->io_hdr.nexus.targ_port)
10849           && (pending_io->io_hdr.nexus.initid ==
10850               ooa_io->io_hdr.nexus.initid))
10851          && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
10852               CTL_FLAG_STATUS_SENT)) == 0))
10853                 return (CTL_ACTION_OVERLAP);
10854
10855         /*
10856          * The initiator attempted to send multiple tagged commands with
10857          * the same ID.  (It's fine if different initiators have the same
10858          * tag ID.)
10859          *
10860          * Even if all of those conditions are true, we don't kill the I/O
10861          * if the command ahead of us has been aborted.  We won't end up
10862          * sending it to the FETD, and it's perfectly legal to resend a
10863          * command with the same tag number as long as the previous
10864          * instance of this tag number has been aborted somehow.
10865          */
10866         if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10867          && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10868          && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
10869          && ((pending_io->io_hdr.nexus.targ_port ==
10870               ooa_io->io_hdr.nexus.targ_port)
10871           && (pending_io->io_hdr.nexus.initid ==
10872               ooa_io->io_hdr.nexus.initid))
10873          && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
10874               CTL_FLAG_STATUS_SENT)) == 0))
10875                 return (CTL_ACTION_OVERLAP_TAG);
10876
10877         /*
10878          * If we get a head of queue tag, SAM-3 says that we should
10879          * immediately execute it.
10880          *
10881          * What happens if this command would normally block for some other
10882          * reason?  e.g. a request sense with a head of queue tag
10883          * immediately after a write.  Normally that would block, but this
10884          * will result in its getting executed immediately...
10885          *
10886          * We currently return "pass" instead of "skip", so we'll end up
10887          * going through the rest of the queue to check for overlapped tags.
10888          *
10889          * XXX KDM check for other types of blockage first??
10890          */
10891         if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10892                 return (CTL_ACTION_PASS);
10893
10894         /*
10895          * Ordered tags have to block until all items ahead of them
10896          * have completed.  If we get called with an ordered tag, we always
10897          * block, if something else is ahead of us in the queue.
10898          */
10899         if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
10900                 return (CTL_ACTION_BLOCK);
10901
10902         /*
10903          * Simple tags get blocked until all head of queue and ordered tags
10904          * ahead of them have completed.  I'm lumping untagged commands in
10905          * with simple tags here.  XXX KDM is that the right thing to do?
10906          */
10907         if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10908           || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
10909          && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10910           || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
10911                 return (CTL_ACTION_BLOCK);
10912
10913         pending_entry = ctl_get_cmd_entry(&pending_io->scsiio, NULL);
10914         ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio, NULL);
10915
10916         serialize_row = ctl_serialize_table[ooa_entry->seridx];
10917
10918         switch (serialize_row[pending_entry->seridx]) {
10919         case CTL_SER_BLOCK:
10920                 return (CTL_ACTION_BLOCK);
10921         case CTL_SER_EXTENT:
10922                 return (ctl_extent_check(ooa_io, pending_io,
10923                     (lun->be_lun && lun->be_lun->serseq == CTL_LUN_SERSEQ_ON)));
10924         case CTL_SER_EXTENTOPT:
10925                 if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
10926                     & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
10927                         return (ctl_extent_check(ooa_io, pending_io,
10928                             (lun->be_lun &&
10929                              lun->be_lun->serseq == CTL_LUN_SERSEQ_ON)));
10930                 return (CTL_ACTION_PASS);
10931         case CTL_SER_EXTENTSEQ:
10932                 if (lun->be_lun && lun->be_lun->serseq != CTL_LUN_SERSEQ_OFF)
10933                         return (ctl_extent_check_seq(ooa_io, pending_io));
10934                 return (CTL_ACTION_PASS);
10935         case CTL_SER_PASS:
10936                 return (CTL_ACTION_PASS);
10937         case CTL_SER_BLOCKOPT:
10938                 if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
10939                     & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
10940                         return (CTL_ACTION_BLOCK);
10941                 return (CTL_ACTION_PASS);
10942         case CTL_SER_SKIP:
10943                 return (CTL_ACTION_SKIP);
10944         default:
10945                 panic("invalid serialization value %d",
10946                       serialize_row[pending_entry->seridx]);
10947         }
10948
10949         return (CTL_ACTION_ERROR);
10950 }
10951
10952 /*
10953  * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
10954  * Assumptions:
10955  * - pending_io is generally either incoming, or on the blocked queue
10956  * - starting I/O is the I/O we want to start the check with.
10957  */
10958 static ctl_action
10959 ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
10960               union ctl_io *starting_io)
10961 {
10962         union ctl_io *ooa_io;
10963         ctl_action action;
10964
10965         mtx_assert(&lun->lun_lock, MA_OWNED);
10966
10967         /*
10968          * Run back along the OOA queue, starting with the current
10969          * blocked I/O and going through every I/O before it on the
10970          * queue.  If starting_io is NULL, we'll just end up returning
10971          * CTL_ACTION_PASS.
10972          */
10973         for (ooa_io = starting_io; ooa_io != NULL;
10974              ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
10975              ooa_links)){
10976
10977                 /*
10978                  * This routine just checks to see whether
10979                  * cur_blocked is blocked by ooa_io, which is ahead
10980                  * of it in the queue.  It doesn't queue/dequeue
10981                  * cur_blocked.
10982                  */
10983                 action = ctl_check_for_blockage(lun, pending_io, ooa_io);
10984                 switch (action) {
10985                 case CTL_ACTION_BLOCK:
10986                 case CTL_ACTION_OVERLAP:
10987                 case CTL_ACTION_OVERLAP_TAG:
10988                 case CTL_ACTION_SKIP:
10989                 case CTL_ACTION_ERROR:
10990                         return (action);
10991                         break; /* NOTREACHED */
10992                 case CTL_ACTION_PASS:
10993                         break;
10994                 default:
10995                         panic("invalid action %d", action);
10996                         break;  /* NOTREACHED */
10997                 }
10998         }
10999
11000         return (CTL_ACTION_PASS);
11001 }
11002
11003 /*
11004  * Assumptions:
11005  * - An I/O has just completed, and has been removed from the per-LUN OOA
11006  *   queue, so some items on the blocked queue may now be unblocked.
11007  */
11008 static int
11009 ctl_check_blocked(struct ctl_lun *lun)
11010 {
11011         struct ctl_softc *softc = lun->ctl_softc;
11012         union ctl_io *cur_blocked, *next_blocked;
11013
11014         mtx_assert(&lun->lun_lock, MA_OWNED);
11015
11016         /*
11017          * Run forward from the head of the blocked queue, checking each
11018          * entry against the I/Os prior to it on the OOA queue to see if
11019          * there is still any blockage.
11020          *
11021          * We cannot use the TAILQ_FOREACH() macro, because it can't deal
11022          * with our removing a variable on it while it is traversing the
11023          * list.
11024          */
11025         for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
11026              cur_blocked != NULL; cur_blocked = next_blocked) {
11027                 union ctl_io *prev_ooa;
11028                 ctl_action action;
11029
11030                 next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
11031                                                           blocked_links);
11032
11033                 prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
11034                                                       ctl_ooaq, ooa_links);
11035
11036                 /*
11037                  * If cur_blocked happens to be the first item in the OOA
11038                  * queue now, prev_ooa will be NULL, and the action
11039                  * returned will just be CTL_ACTION_PASS.
11040                  */
11041                 action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
11042
11043                 switch (action) {
11044                 case CTL_ACTION_BLOCK:
11045                         /* Nothing to do here, still blocked */
11046                         break;
11047                 case CTL_ACTION_OVERLAP:
11048                 case CTL_ACTION_OVERLAP_TAG:
11049                         /*
11050                          * This shouldn't happen!  In theory we've already
11051                          * checked this command for overlap...
11052                          */
11053                         break;
11054                 case CTL_ACTION_PASS:
11055                 case CTL_ACTION_SKIP: {
11056                         const struct ctl_cmd_entry *entry;
11057
11058                         /*
11059                          * The skip case shouldn't happen, this transaction
11060                          * should have never made it onto the blocked queue.
11061                          */
11062                         /*
11063                          * This I/O is no longer blocked, we can remove it
11064                          * from the blocked queue.  Since this is a TAILQ
11065                          * (doubly linked list), we can do O(1) removals
11066                          * from any place on the list.
11067                          */
11068                         TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
11069                                      blocked_links);
11070                         cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11071
11072                         if ((softc->ha_mode != CTL_HA_MODE_XFER) &&
11073                             (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)){
11074                                 /*
11075                                  * Need to send IO back to original side to
11076                                  * run
11077                                  */
11078                                 union ctl_ha_msg msg_info;
11079
11080                                 cur_blocked->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11081                                 msg_info.hdr.original_sc =
11082                                         cur_blocked->io_hdr.original_sc;
11083                                 msg_info.hdr.serializing_sc = cur_blocked;
11084                                 msg_info.hdr.msg_type = CTL_MSG_R2R;
11085                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11086                                     sizeof(msg_info.hdr), M_NOWAIT);
11087                                 break;
11088                         }
11089                         entry = ctl_get_cmd_entry(&cur_blocked->scsiio, NULL);
11090
11091                         /*
11092                          * Check this I/O for LUN state changes that may
11093                          * have happened while this command was blocked.
11094                          * The LUN state may have been changed by a command
11095                          * ahead of us in the queue, so we need to re-check
11096                          * for any states that can be caused by SCSI
11097                          * commands.
11098                          */
11099                         if (ctl_scsiio_lun_check(lun, entry,
11100                                                  &cur_blocked->scsiio) == 0) {
11101                                 cur_blocked->io_hdr.flags |=
11102                                                       CTL_FLAG_IS_WAS_ON_RTR;
11103                                 ctl_enqueue_rtr(cur_blocked);
11104                         } else
11105                                 ctl_done(cur_blocked);
11106                         break;
11107                 }
11108                 default:
11109                         /*
11110                          * This probably shouldn't happen -- we shouldn't
11111                          * get CTL_ACTION_ERROR, or anything else.
11112                          */
11113                         break;
11114                 }
11115         }
11116
11117         return (CTL_RETVAL_COMPLETE);
11118 }
11119
11120 /*
11121  * This routine (with one exception) checks LUN flags that can be set by
11122  * commands ahead of us in the OOA queue.  These flags have to be checked
11123  * when a command initially comes in, and when we pull a command off the
11124  * blocked queue and are preparing to execute it.  The reason we have to
11125  * check these flags for commands on the blocked queue is that the LUN
11126  * state may have been changed by a command ahead of us while we're on the
11127  * blocked queue.
11128  *
11129  * Ordering is somewhat important with these checks, so please pay
11130  * careful attention to the placement of any new checks.
11131  */
11132 static int
11133 ctl_scsiio_lun_check(struct ctl_lun *lun,
11134     const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
11135 {
11136         struct ctl_softc *softc = lun->ctl_softc;
11137         int retval;
11138         uint32_t residx;
11139
11140         retval = 0;
11141
11142         mtx_assert(&lun->lun_lock, MA_OWNED);
11143
11144         /*
11145          * If this shelf is a secondary shelf controller, we may have to
11146          * reject some commands disallowed by HA mode and link state.
11147          */
11148         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11149                 if (softc->ha_link == CTL_HA_LINK_OFFLINE &&
11150                     (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
11151                         ctl_set_lun_unavail(ctsio);
11152                         retval = 1;
11153                         goto bailout;
11154                 }
11155                 if ((lun->flags & CTL_LUN_PEER_SC_PRIMARY) == 0 &&
11156                     (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
11157                         ctl_set_lun_transit(ctsio);
11158                         retval = 1;
11159                         goto bailout;
11160                 }
11161                 if (softc->ha_mode == CTL_HA_MODE_ACT_STBY &&
11162                     (entry->flags & CTL_CMD_FLAG_OK_ON_STANDBY) == 0) {
11163                         ctl_set_lun_standby(ctsio);
11164                         retval = 1;
11165                         goto bailout;
11166                 }
11167
11168                 /* The rest of checks are only done on executing side */
11169                 if (softc->ha_mode == CTL_HA_MODE_XFER)
11170                         goto bailout;
11171         }
11172
11173         if (entry->pattern & CTL_LUN_PAT_WRITE) {
11174                 if (lun->be_lun &&
11175                     lun->be_lun->flags & CTL_LUN_FLAG_READONLY) {
11176                         ctl_set_hw_write_protected(ctsio);
11177                         retval = 1;
11178                         goto bailout;
11179                 }
11180                 if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT]
11181                     .eca_and_aen & SCP_SWP) != 0) {
11182                         ctl_set_sense(ctsio, /*current_error*/ 1,
11183                             /*sense_key*/ SSD_KEY_DATA_PROTECT,
11184                             /*asc*/ 0x27, /*ascq*/ 0x02, SSD_ELEM_NONE);
11185                         retval = 1;
11186                         goto bailout;
11187                 }
11188         }
11189
11190         /*
11191          * Check for a reservation conflict.  If this command isn't allowed
11192          * even on reserved LUNs, and if this initiator isn't the one who
11193          * reserved us, reject the command with a reservation conflict.
11194          */
11195         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11196         if ((lun->flags & CTL_LUN_RESERVED)
11197          && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
11198                 if (lun->res_idx != residx) {
11199                         ctl_set_reservation_conflict(ctsio);
11200                         retval = 1;
11201                         goto bailout;
11202                 }
11203         }
11204
11205         if ((lun->flags & CTL_LUN_PR_RESERVED) == 0 ||
11206             (entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV)) {
11207                 /* No reservation or command is allowed. */;
11208         } else if ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_WRESV) &&
11209             (lun->res_type == SPR_TYPE_WR_EX ||
11210              lun->res_type == SPR_TYPE_WR_EX_RO ||
11211              lun->res_type == SPR_TYPE_WR_EX_AR)) {
11212                 /* The command is allowed for Write Exclusive resv. */;
11213         } else {
11214                 /*
11215                  * if we aren't registered or it's a res holder type
11216                  * reservation and this isn't the res holder then set a
11217                  * conflict.
11218                  */
11219                 if (ctl_get_prkey(lun, residx) == 0
11220                  || (residx != lun->pr_res_idx && lun->res_type < 4)) {
11221                         ctl_set_reservation_conflict(ctsio);
11222                         retval = 1;
11223                         goto bailout;
11224                 }
11225         }
11226
11227         if ((lun->flags & CTL_LUN_OFFLINE)
11228          && ((entry->flags & CTL_CMD_FLAG_OK_ON_STANDBY) == 0)) {
11229                 ctl_set_lun_not_ready(ctsio);
11230                 retval = 1;
11231                 goto bailout;
11232         }
11233
11234         if ((lun->flags & CTL_LUN_STOPPED)
11235          && ((entry->flags & CTL_CMD_FLAG_OK_ON_STOPPED) == 0)) {
11236                 /* "Logical unit not ready, initializing cmd. required" */
11237                 ctl_set_lun_stopped(ctsio);
11238                 retval = 1;
11239                 goto bailout;
11240         }
11241
11242         if ((lun->flags & CTL_LUN_INOPERABLE)
11243          && ((entry->flags & CTL_CMD_FLAG_OK_ON_INOPERABLE) == 0)) {
11244                 /* "Medium format corrupted" */
11245                 ctl_set_medium_format_corrupted(ctsio);
11246                 retval = 1;
11247                 goto bailout;
11248         }
11249
11250 bailout:
11251         return (retval);
11252 }
11253
11254 static void
11255 ctl_failover_io(union ctl_io *io, int have_lock)
11256 {
11257         ctl_set_busy(&io->scsiio);
11258         ctl_done(io);
11259 }
11260
11261 static void
11262 ctl_failover_lun(struct ctl_lun *lun)
11263 {
11264         struct ctl_softc *softc = lun->ctl_softc;
11265         struct ctl_io_hdr *io, *next_io;
11266
11267         CTL_DEBUG_PRINT(("FAILOVER for lun %ju\n", lun->lun));
11268         if (softc->ha_mode == CTL_HA_MODE_XFER) {
11269                 TAILQ_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
11270                         /* We are master */
11271                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11272                                 if (io->flags & CTL_FLAG_IO_ACTIVE) {
11273                                         io->flags |= CTL_FLAG_ABORT;
11274                                         io->flags |= CTL_FLAG_FAILOVER;
11275                                 } else { /* This can be only due to DATAMOVE */
11276                                         io->msg_type = CTL_MSG_DATAMOVE_DONE;
11277                                         io->flags &= ~CTL_FLAG_DMA_INPROG;
11278                                         io->flags |= CTL_FLAG_IO_ACTIVE;
11279                                         io->port_status = 31340;
11280                                         ctl_enqueue_isc((union ctl_io *)io);
11281                                 }
11282                         }
11283                         /* We are slave */
11284                         if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
11285                                 io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11286                                 if (io->flags & CTL_FLAG_IO_ACTIVE) {
11287                                         io->flags |= CTL_FLAG_FAILOVER;
11288                                 } else {
11289                                         ctl_set_busy(&((union ctl_io *)io)->
11290                                             scsiio);
11291                                         ctl_done((union ctl_io *)io);
11292                                 }
11293                         }
11294                 }
11295         } else { /* SERIALIZE modes */
11296                 TAILQ_FOREACH_SAFE(io, &lun->blocked_queue, blocked_links,
11297                     next_io) {
11298                         /* We are master */
11299                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11300                                 TAILQ_REMOVE(&lun->blocked_queue, io,
11301                                     blocked_links);
11302                                 io->flags &= ~CTL_FLAG_BLOCKED;
11303                                 TAILQ_REMOVE(&lun->ooa_queue, io, ooa_links);
11304                                 ctl_free_io((union ctl_io *)io);
11305                         }
11306                 }
11307                 TAILQ_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
11308                         /* We are master */
11309                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11310                                 TAILQ_REMOVE(&lun->ooa_queue, io, ooa_links);
11311                                 ctl_free_io((union ctl_io *)io);
11312                         }
11313                         /* We are slave */
11314                         if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
11315                                 io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11316                                 if (!(io->flags & CTL_FLAG_IO_ACTIVE)) {
11317                                         ctl_set_busy(&((union ctl_io *)io)->
11318                                             scsiio);
11319                                         ctl_done((union ctl_io *)io);
11320                                 }
11321                         }
11322                 }
11323                 ctl_check_blocked(lun);
11324         }
11325 }
11326
11327 static int
11328 ctl_scsiio_precheck(struct ctl_softc *softc, struct ctl_scsiio *ctsio)
11329 {
11330         struct ctl_lun *lun;
11331         const struct ctl_cmd_entry *entry;
11332         uint32_t initidx, targ_lun;
11333         int retval;
11334
11335         retval = 0;
11336
11337         lun = NULL;
11338
11339         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11340         if ((targ_lun < CTL_MAX_LUNS)
11341          && ((lun = softc->ctl_luns[targ_lun]) != NULL)) {
11342                 /*
11343                  * If the LUN is invalid, pretend that it doesn't exist.
11344                  * It will go away as soon as all pending I/O has been
11345                  * completed.
11346                  */
11347                 mtx_lock(&lun->lun_lock);
11348                 if (lun->flags & CTL_LUN_DISABLED) {
11349                         mtx_unlock(&lun->lun_lock);
11350                         lun = NULL;
11351                         ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11352                         ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11353                 } else {
11354                         ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
11355                         ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
11356                                 lun->be_lun;
11357
11358                         /*
11359                          * Every I/O goes into the OOA queue for a
11360                          * particular LUN, and stays there until completion.
11361                          */
11362 #ifdef CTL_TIME_IO
11363                         if (TAILQ_EMPTY(&lun->ooa_queue)) {
11364                                 lun->idle_time += getsbinuptime() -
11365                                     lun->last_busy;
11366                         }
11367 #endif
11368                         TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr,
11369                             ooa_links);
11370                 }
11371         } else {
11372                 ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11373                 ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11374         }
11375
11376         /* Get command entry and return error if it is unsuppotyed. */
11377         entry = ctl_validate_command(ctsio);
11378         if (entry == NULL) {
11379                 if (lun)
11380                         mtx_unlock(&lun->lun_lock);
11381                 return (retval);
11382         }
11383
11384         ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11385         ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11386
11387         /*
11388          * Check to see whether we can send this command to LUNs that don't
11389          * exist.  This should pretty much only be the case for inquiry
11390          * and request sense.  Further checks, below, really require having
11391          * a LUN, so we can't really check the command anymore.  Just put
11392          * it on the rtr queue.
11393          */
11394         if (lun == NULL) {
11395                 if (entry->flags & CTL_CMD_FLAG_OK_ON_NO_LUN) {
11396                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11397                         ctl_enqueue_rtr((union ctl_io *)ctsio);
11398                         return (retval);
11399                 }
11400
11401                 ctl_set_unsupported_lun(ctsio);
11402                 ctl_done((union ctl_io *)ctsio);
11403                 CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11404                 return (retval);
11405         } else {
11406                 /*
11407                  * Make sure we support this particular command on this LUN.
11408                  * e.g., we don't support writes to the control LUN.
11409                  */
11410                 if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11411                         mtx_unlock(&lun->lun_lock);
11412                         ctl_set_invalid_opcode(ctsio);
11413                         ctl_done((union ctl_io *)ctsio);
11414                         return (retval);
11415                 }
11416         }
11417
11418         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11419
11420 #ifdef CTL_WITH_CA
11421         /*
11422          * If we've got a request sense, it'll clear the contingent
11423          * allegiance condition.  Otherwise, if we have a CA condition for
11424          * this initiator, clear it, because it sent down a command other
11425          * than request sense.
11426          */
11427         if ((ctsio->cdb[0] != REQUEST_SENSE)
11428          && (ctl_is_set(lun->have_ca, initidx)))
11429                 ctl_clear_mask(lun->have_ca, initidx);
11430 #endif
11431
11432         /*
11433          * If the command has this flag set, it handles its own unit
11434          * attention reporting, we shouldn't do anything.  Otherwise we
11435          * check for any pending unit attentions, and send them back to the
11436          * initiator.  We only do this when a command initially comes in,
11437          * not when we pull it off the blocked queue.
11438          *
11439          * According to SAM-3, section 5.3.2, the order that things get
11440          * presented back to the host is basically unit attentions caused
11441          * by some sort of reset event, busy status, reservation conflicts
11442          * or task set full, and finally any other status.
11443          *
11444          * One issue here is that some of the unit attentions we report
11445          * don't fall into the "reset" category (e.g. "reported luns data
11446          * has changed").  So reporting it here, before the reservation
11447          * check, may be technically wrong.  I guess the only thing to do
11448          * would be to check for and report the reset events here, and then
11449          * check for the other unit attention types after we check for a
11450          * reservation conflict.
11451          *
11452          * XXX KDM need to fix this
11453          */
11454         if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11455                 ctl_ua_type ua_type;
11456
11457                 ua_type = ctl_build_ua(lun, initidx, &ctsio->sense_data,
11458                     SSD_TYPE_NONE);
11459                 if (ua_type != CTL_UA_NONE) {
11460                         mtx_unlock(&lun->lun_lock);
11461                         ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11462                         ctsio->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
11463                         ctsio->sense_len = SSD_FULL_SIZE;
11464                         ctl_done((union ctl_io *)ctsio);
11465                         return (retval);
11466                 }
11467         }
11468
11469
11470         if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
11471                 mtx_unlock(&lun->lun_lock);
11472                 ctl_done((union ctl_io *)ctsio);
11473                 return (retval);
11474         }
11475
11476         /*
11477          * XXX CHD this is where we want to send IO to other side if
11478          * this LUN is secondary on this SC. We will need to make a copy
11479          * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11480          * the copy we send as FROM_OTHER.
11481          * We also need to stuff the address of the original IO so we can
11482          * find it easily. Something similar will need be done on the other
11483          * side so when we are done we can find the copy.
11484          */
11485         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
11486             (lun->flags & CTL_LUN_PEER_SC_PRIMARY) != 0 &&
11487             (entry->flags & CTL_CMD_FLAG_RUN_HERE) == 0) {
11488                 union ctl_ha_msg msg_info;
11489                 int isc_retval;
11490
11491                 ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11492                 ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11493                 mtx_unlock(&lun->lun_lock);
11494
11495                 msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11496                 msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11497                 msg_info.hdr.serializing_sc = NULL;
11498                 msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11499                 msg_info.scsi.tag_num = ctsio->tag_num;
11500                 msg_info.scsi.tag_type = ctsio->tag_type;
11501                 msg_info.scsi.cdb_len = ctsio->cdb_len;
11502                 memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11503
11504                 if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11505                     sizeof(msg_info.scsi) - sizeof(msg_info.scsi.sense_data),
11506                     M_WAITOK)) > CTL_HA_STATUS_SUCCESS) {
11507                         ctl_set_busy(ctsio);
11508                         ctl_done((union ctl_io *)ctsio);
11509                         return (retval);
11510                 }
11511                 return (retval);
11512         }
11513
11514         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11515                               (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11516                               ctl_ooaq, ooa_links))) {
11517         case CTL_ACTION_BLOCK:
11518                 ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11519                 TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11520                                   blocked_links);
11521                 mtx_unlock(&lun->lun_lock);
11522                 return (retval);
11523         case CTL_ACTION_PASS:
11524         case CTL_ACTION_SKIP:
11525                 ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11526                 mtx_unlock(&lun->lun_lock);
11527                 ctl_enqueue_rtr((union ctl_io *)ctsio);
11528                 break;
11529         case CTL_ACTION_OVERLAP:
11530                 mtx_unlock(&lun->lun_lock);
11531                 ctl_set_overlapped_cmd(ctsio);
11532                 ctl_done((union ctl_io *)ctsio);
11533                 break;
11534         case CTL_ACTION_OVERLAP_TAG:
11535                 mtx_unlock(&lun->lun_lock);
11536                 ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11537                 ctl_done((union ctl_io *)ctsio);
11538                 break;
11539         case CTL_ACTION_ERROR:
11540         default:
11541                 mtx_unlock(&lun->lun_lock);
11542                 ctl_set_internal_failure(ctsio,
11543                                          /*sks_valid*/ 0,
11544                                          /*retry_count*/ 0);
11545                 ctl_done((union ctl_io *)ctsio);
11546                 break;
11547         }
11548         return (retval);
11549 }
11550
11551 const struct ctl_cmd_entry *
11552 ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa)
11553 {
11554         const struct ctl_cmd_entry *entry;
11555         int service_action;
11556
11557         entry = &ctl_cmd_table[ctsio->cdb[0]];
11558         if (sa)
11559                 *sa = ((entry->flags & CTL_CMD_FLAG_SA5) != 0);
11560         if (entry->flags & CTL_CMD_FLAG_SA5) {
11561                 service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11562                 entry = &((const struct ctl_cmd_entry *)
11563                     entry->execute)[service_action];
11564         }
11565         return (entry);
11566 }
11567
11568 const struct ctl_cmd_entry *
11569 ctl_validate_command(struct ctl_scsiio *ctsio)
11570 {
11571         const struct ctl_cmd_entry *entry;
11572         int i, sa;
11573         uint8_t diff;
11574
11575         entry = ctl_get_cmd_entry(ctsio, &sa);
11576         if (entry->execute == NULL) {
11577                 if (sa)
11578                         ctl_set_invalid_field(ctsio,
11579                                               /*sks_valid*/ 1,
11580                                               /*command*/ 1,
11581                                               /*field*/ 1,
11582                                               /*bit_valid*/ 1,
11583                                               /*bit*/ 4);
11584                 else
11585                         ctl_set_invalid_opcode(ctsio);
11586                 ctl_done((union ctl_io *)ctsio);
11587                 return (NULL);
11588         }
11589         KASSERT(entry->length > 0,
11590             ("Not defined length for command 0x%02x/0x%02x",
11591              ctsio->cdb[0], ctsio->cdb[1]));
11592         for (i = 1; i < entry->length; i++) {
11593                 diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11594                 if (diff == 0)
11595                         continue;
11596                 ctl_set_invalid_field(ctsio,
11597                                       /*sks_valid*/ 1,
11598                                       /*command*/ 1,
11599                                       /*field*/ i,
11600                                       /*bit_valid*/ 1,
11601                                       /*bit*/ fls(diff) - 1);
11602                 ctl_done((union ctl_io *)ctsio);
11603                 return (NULL);
11604         }
11605         return (entry);
11606 }
11607
11608 static int
11609 ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11610 {
11611
11612         switch (lun_type) {
11613         case T_PROCESSOR:
11614                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0)
11615                         return (0);
11616                 break;
11617         case T_DIRECT:
11618                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_SLUN) == 0)
11619                         return (0);
11620                 break;
11621         default:
11622                 return (0);
11623         }
11624         return (1);
11625 }
11626
11627 static int
11628 ctl_scsiio(struct ctl_scsiio *ctsio)
11629 {
11630         int retval;
11631         const struct ctl_cmd_entry *entry;
11632
11633         retval = CTL_RETVAL_COMPLETE;
11634
11635         CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11636
11637         entry = ctl_get_cmd_entry(ctsio, NULL);
11638
11639         /*
11640          * If this I/O has been aborted, just send it straight to
11641          * ctl_done() without executing it.
11642          */
11643         if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11644                 ctl_done((union ctl_io *)ctsio);
11645                 goto bailout;
11646         }
11647
11648         /*
11649          * All the checks should have been handled by ctl_scsiio_precheck().
11650          * We should be clear now to just execute the I/O.
11651          */
11652         retval = entry->execute(ctsio);
11653
11654 bailout:
11655         return (retval);
11656 }
11657
11658 /*
11659  * Since we only implement one target right now, a bus reset simply resets
11660  * our single target.
11661  */
11662 static int
11663 ctl_bus_reset(struct ctl_softc *softc, union ctl_io *io)
11664 {
11665         return(ctl_target_reset(softc, io, CTL_UA_BUS_RESET));
11666 }
11667
11668 static int
11669 ctl_target_reset(struct ctl_softc *softc, union ctl_io *io,
11670                  ctl_ua_type ua_type)
11671 {
11672         struct ctl_port *port;
11673         struct ctl_lun *lun;
11674         int retval;
11675
11676         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11677                 union ctl_ha_msg msg_info;
11678
11679                 msg_info.hdr.nexus = io->io_hdr.nexus;
11680                 if (ua_type==CTL_UA_TARG_RESET)
11681                         msg_info.task.task_action = CTL_TASK_TARGET_RESET;
11682                 else
11683                         msg_info.task.task_action = CTL_TASK_BUS_RESET;
11684                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11685                 msg_info.hdr.original_sc = NULL;
11686                 msg_info.hdr.serializing_sc = NULL;
11687                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11688                     sizeof(msg_info.task), M_WAITOK);
11689         }
11690         retval = 0;
11691
11692         mtx_lock(&softc->ctl_lock);
11693         port = softc->ctl_ports[io->io_hdr.nexus.targ_port];
11694         STAILQ_FOREACH(lun, &softc->lun_list, links) {
11695                 if (port != NULL &&
11696                     ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
11697                         continue;
11698                 retval += ctl_do_lun_reset(lun, io, ua_type);
11699         }
11700         mtx_unlock(&softc->ctl_lock);
11701         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11702         return (retval);
11703 }
11704
11705 /*
11706  * The LUN should always be set.  The I/O is optional, and is used to
11707  * distinguish between I/Os sent by this initiator, and by other
11708  * initiators.  We set unit attention for initiators other than this one.
11709  * SAM-3 is vague on this point.  It does say that a unit attention should
11710  * be established for other initiators when a LUN is reset (see section
11711  * 5.7.3), but it doesn't specifically say that the unit attention should
11712  * be established for this particular initiator when a LUN is reset.  Here
11713  * is the relevant text, from SAM-3 rev 8:
11714  *
11715  * 5.7.2 When a SCSI initiator port aborts its own tasks
11716  *
11717  * When a SCSI initiator port causes its own task(s) to be aborted, no
11718  * notification that the task(s) have been aborted shall be returned to
11719  * the SCSI initiator port other than the completion response for the
11720  * command or task management function action that caused the task(s) to
11721  * be aborted and notification(s) associated with related effects of the
11722  * action (e.g., a reset unit attention condition).
11723  *
11724  * XXX KDM for now, we're setting unit attention for all initiators.
11725  */
11726 static int
11727 ctl_do_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
11728 {
11729         union ctl_io *xio;
11730 #if 0
11731         uint32_t initidx;
11732 #endif
11733 #ifdef CTL_WITH_CA
11734         int i;
11735 #endif
11736
11737         mtx_lock(&lun->lun_lock);
11738         /*
11739          * Run through the OOA queue and abort each I/O.
11740          */
11741         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11742              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11743                 xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
11744         }
11745
11746         /*
11747          * This version sets unit attention for every
11748          */
11749 #if 0
11750         initidx = ctl_get_initindex(&io->io_hdr.nexus);
11751         ctl_est_ua_all(lun, initidx, ua_type);
11752 #else
11753         ctl_est_ua_all(lun, -1, ua_type);
11754 #endif
11755
11756         /*
11757          * A reset (any kind, really) clears reservations established with
11758          * RESERVE/RELEASE.  It does not clear reservations established
11759          * with PERSISTENT RESERVE OUT, but we don't support that at the
11760          * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
11761          * reservations made with the RESERVE/RELEASE commands, because
11762          * those commands are obsolete in SPC-3.
11763          */
11764         lun->flags &= ~CTL_LUN_RESERVED;
11765
11766 #ifdef CTL_WITH_CA
11767         for (i = 0; i < CTL_MAX_INITIATORS; i++)
11768                 ctl_clear_mask(lun->have_ca, i);
11769 #endif
11770         mtx_unlock(&lun->lun_lock);
11771
11772         return (0);
11773 }
11774
11775 static int
11776 ctl_lun_reset(struct ctl_softc *softc, union ctl_io *io)
11777 {
11778         struct ctl_lun *lun;
11779         uint32_t targ_lun;
11780         int retval;
11781
11782         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11783         mtx_lock(&softc->ctl_lock);
11784         if ((targ_lun >= CTL_MAX_LUNS) ||
11785             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11786                 mtx_unlock(&softc->ctl_lock);
11787                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11788                 return (1);
11789         }
11790         retval = ctl_do_lun_reset(lun, io, CTL_UA_LUN_RESET);
11791         mtx_unlock(&softc->ctl_lock);
11792         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11793
11794         if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0) {
11795                 union ctl_ha_msg msg_info;
11796
11797                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11798                 msg_info.hdr.nexus = io->io_hdr.nexus;
11799                 msg_info.task.task_action = CTL_TASK_LUN_RESET;
11800                 msg_info.hdr.original_sc = NULL;
11801                 msg_info.hdr.serializing_sc = NULL;
11802                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11803                     sizeof(msg_info.task), M_WAITOK);
11804         }
11805         return (retval);
11806 }
11807
11808 static void
11809 ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
11810     int other_sc)
11811 {
11812         union ctl_io *xio;
11813
11814         mtx_assert(&lun->lun_lock, MA_OWNED);
11815
11816         /*
11817          * Run through the OOA queue and attempt to find the given I/O.
11818          * The target port, initiator ID, tag type and tag number have to
11819          * match the values that we got from the initiator.  If we have an
11820          * untagged command to abort, simply abort the first untagged command
11821          * we come to.  We only allow one untagged command at a time of course.
11822          */
11823         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11824              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11825
11826                 if ((targ_port == UINT32_MAX ||
11827                      targ_port == xio->io_hdr.nexus.targ_port) &&
11828                     (init_id == UINT32_MAX ||
11829                      init_id == xio->io_hdr.nexus.initid)) {
11830                         if (targ_port != xio->io_hdr.nexus.targ_port ||
11831                             init_id != xio->io_hdr.nexus.initid)
11832                                 xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS;
11833                         xio->io_hdr.flags |= CTL_FLAG_ABORT;
11834                         if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11835                                 union ctl_ha_msg msg_info;
11836
11837                                 msg_info.hdr.nexus = xio->io_hdr.nexus;
11838                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
11839                                 msg_info.task.tag_num = xio->scsiio.tag_num;
11840                                 msg_info.task.tag_type = xio->scsiio.tag_type;
11841                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11842                                 msg_info.hdr.original_sc = NULL;
11843                                 msg_info.hdr.serializing_sc = NULL;
11844                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11845                                     sizeof(msg_info.task), M_NOWAIT);
11846                         }
11847                 }
11848         }
11849 }
11850
11851 static int
11852 ctl_abort_task_set(union ctl_io *io)
11853 {
11854         struct ctl_softc *softc = control_softc;
11855         struct ctl_lun *lun;
11856         uint32_t targ_lun;
11857
11858         /*
11859          * Look up the LUN.
11860          */
11861         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11862         mtx_lock(&softc->ctl_lock);
11863         if ((targ_lun >= CTL_MAX_LUNS) ||
11864             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11865                 mtx_unlock(&softc->ctl_lock);
11866                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11867                 return (1);
11868         }
11869
11870         mtx_lock(&lun->lun_lock);
11871         mtx_unlock(&softc->ctl_lock);
11872         if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
11873                 ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
11874                     io->io_hdr.nexus.initid,
11875                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11876         } else { /* CTL_TASK_CLEAR_TASK_SET */
11877                 ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
11878                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11879         }
11880         mtx_unlock(&lun->lun_lock);
11881         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11882         return (0);
11883 }
11884
11885 static int
11886 ctl_i_t_nexus_reset(union ctl_io *io)
11887 {
11888         struct ctl_softc *softc = control_softc;
11889         struct ctl_lun *lun;
11890         uint32_t initidx;
11891
11892         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11893                 union ctl_ha_msg msg_info;
11894
11895                 msg_info.hdr.nexus = io->io_hdr.nexus;
11896                 msg_info.task.task_action = CTL_TASK_I_T_NEXUS_RESET;
11897                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11898                 msg_info.hdr.original_sc = NULL;
11899                 msg_info.hdr.serializing_sc = NULL;
11900                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11901                     sizeof(msg_info.task), M_WAITOK);
11902         }
11903
11904         initidx = ctl_get_initindex(&io->io_hdr.nexus);
11905         mtx_lock(&softc->ctl_lock);
11906         STAILQ_FOREACH(lun, &softc->lun_list, links) {
11907                 mtx_lock(&lun->lun_lock);
11908                 ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
11909                     io->io_hdr.nexus.initid, 1);
11910 #ifdef CTL_WITH_CA
11911                 ctl_clear_mask(lun->have_ca, initidx);
11912 #endif
11913                 if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == initidx))
11914                         lun->flags &= ~CTL_LUN_RESERVED;
11915                 ctl_est_ua(lun, initidx, CTL_UA_I_T_NEXUS_LOSS);
11916                 mtx_unlock(&lun->lun_lock);
11917         }
11918         mtx_unlock(&softc->ctl_lock);
11919         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11920         return (0);
11921 }
11922
11923 static int
11924 ctl_abort_task(union ctl_io *io)
11925 {
11926         union ctl_io *xio;
11927         struct ctl_lun *lun;
11928         struct ctl_softc *softc;
11929 #if 0
11930         struct sbuf sb;
11931         char printbuf[128];
11932 #endif
11933         int found;
11934         uint32_t targ_lun;
11935
11936         softc = control_softc;
11937         found = 0;
11938
11939         /*
11940          * Look up the LUN.
11941          */
11942         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11943         mtx_lock(&softc->ctl_lock);
11944         if ((targ_lun >= CTL_MAX_LUNS) ||
11945             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11946                 mtx_unlock(&softc->ctl_lock);
11947                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11948                 return (1);
11949         }
11950
11951 #if 0
11952         printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
11953                lun->lun, io->taskio.tag_num, io->taskio.tag_type);
11954 #endif
11955
11956         mtx_lock(&lun->lun_lock);
11957         mtx_unlock(&softc->ctl_lock);
11958         /*
11959          * Run through the OOA queue and attempt to find the given I/O.
11960          * The target port, initiator ID, tag type and tag number have to
11961          * match the values that we got from the initiator.  If we have an
11962          * untagged command to abort, simply abort the first untagged command
11963          * we come to.  We only allow one untagged command at a time of course.
11964          */
11965         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11966              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11967 #if 0
11968                 sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
11969
11970                 sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
11971                             lun->lun, xio->scsiio.tag_num,
11972                             xio->scsiio.tag_type,
11973                             (xio->io_hdr.blocked_links.tqe_prev
11974                             == NULL) ? "" : " BLOCKED",
11975                             (xio->io_hdr.flags &
11976                             CTL_FLAG_DMA_INPROG) ? " DMA" : "",
11977                             (xio->io_hdr.flags &
11978                             CTL_FLAG_ABORT) ? " ABORT" : "",
11979                             (xio->io_hdr.flags &
11980                             CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
11981                 ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
11982                 sbuf_finish(&sb);
11983                 printf("%s\n", sbuf_data(&sb));
11984 #endif
11985
11986                 if ((xio->io_hdr.nexus.targ_port != io->io_hdr.nexus.targ_port)
11987                  || (xio->io_hdr.nexus.initid != io->io_hdr.nexus.initid)
11988                  || (xio->io_hdr.flags & CTL_FLAG_ABORT))
11989                         continue;
11990
11991                 /*
11992                  * If the abort says that the task is untagged, the
11993                  * task in the queue must be untagged.  Otherwise,
11994                  * we just check to see whether the tag numbers
11995                  * match.  This is because the QLogic firmware
11996                  * doesn't pass back the tag type in an abort
11997                  * request.
11998                  */
11999 #if 0
12000                 if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
12001                   && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
12002                  || (xio->scsiio.tag_num == io->taskio.tag_num))
12003 #endif
12004                 /*
12005                  * XXX KDM we've got problems with FC, because it
12006                  * doesn't send down a tag type with aborts.  So we
12007                  * can only really go by the tag number...
12008                  * This may cause problems with parallel SCSI.
12009                  * Need to figure that out!!
12010                  */
12011                 if (xio->scsiio.tag_num == io->taskio.tag_num) {
12012                         xio->io_hdr.flags |= CTL_FLAG_ABORT;
12013                         found = 1;
12014                         if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0 &&
12015                             !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12016                                 union ctl_ha_msg msg_info;
12017
12018                                 msg_info.hdr.nexus = io->io_hdr.nexus;
12019                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
12020                                 msg_info.task.tag_num = io->taskio.tag_num;
12021                                 msg_info.task.tag_type = io->taskio.tag_type;
12022                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12023                                 msg_info.hdr.original_sc = NULL;
12024                                 msg_info.hdr.serializing_sc = NULL;
12025 #if 0
12026                                 printf("Sent Abort to other side\n");
12027 #endif
12028                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
12029                                     sizeof(msg_info.task), M_NOWAIT);
12030                         }
12031 #if 0
12032                         printf("ctl_abort_task: found I/O to abort\n");
12033 #endif
12034                 }
12035         }
12036         mtx_unlock(&lun->lun_lock);
12037
12038         if (found == 0) {
12039                 /*
12040                  * This isn't really an error.  It's entirely possible for
12041                  * the abort and command completion to cross on the wire.
12042                  * This is more of an informative/diagnostic error.
12043                  */
12044 #if 0
12045                 printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
12046                        "%u:%u:%u tag %d type %d\n",
12047                        io->io_hdr.nexus.initid,
12048                        io->io_hdr.nexus.targ_port,
12049                        io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
12050                        io->taskio.tag_type);
12051 #endif
12052         }
12053         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12054         return (0);
12055 }
12056
12057 static int
12058 ctl_query_task(union ctl_io *io, int task_set)
12059 {
12060         union ctl_io *xio;
12061         struct ctl_lun *lun;
12062         struct ctl_softc *softc;
12063         int found = 0;
12064         uint32_t targ_lun;
12065
12066         softc = control_softc;
12067         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12068         mtx_lock(&softc->ctl_lock);
12069         if ((targ_lun >= CTL_MAX_LUNS) ||
12070             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12071                 mtx_unlock(&softc->ctl_lock);
12072                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12073                 return (1);
12074         }
12075         mtx_lock(&lun->lun_lock);
12076         mtx_unlock(&softc->ctl_lock);
12077         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12078              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12079
12080                 if ((xio->io_hdr.nexus.targ_port != io->io_hdr.nexus.targ_port)
12081                  || (xio->io_hdr.nexus.initid != io->io_hdr.nexus.initid)
12082                  || (xio->io_hdr.flags & CTL_FLAG_ABORT))
12083                         continue;
12084
12085                 if (task_set || xio->scsiio.tag_num == io->taskio.tag_num) {
12086                         found = 1;
12087                         break;
12088                 }
12089         }
12090         mtx_unlock(&lun->lun_lock);
12091         if (found)
12092                 io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED;
12093         else
12094                 io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12095         return (0);
12096 }
12097
12098 static int
12099 ctl_query_async_event(union ctl_io *io)
12100 {
12101         struct ctl_lun *lun;
12102         struct ctl_softc *softc;
12103         ctl_ua_type ua;
12104         uint32_t targ_lun, initidx;
12105
12106         softc = control_softc;
12107         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12108         mtx_lock(&softc->ctl_lock);
12109         if ((targ_lun >= CTL_MAX_LUNS) ||
12110             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12111                 mtx_unlock(&softc->ctl_lock);
12112                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12113                 return (1);
12114         }
12115         mtx_lock(&lun->lun_lock);
12116         mtx_unlock(&softc->ctl_lock);
12117         initidx = ctl_get_initindex(&io->io_hdr.nexus);
12118         ua = ctl_build_qae(lun, initidx, io->taskio.task_resp);
12119         mtx_unlock(&lun->lun_lock);
12120         if (ua != CTL_UA_NONE)
12121                 io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED;
12122         else
12123                 io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12124         return (0);
12125 }
12126
12127 static void
12128 ctl_run_task(union ctl_io *io)
12129 {
12130         struct ctl_softc *softc = control_softc;
12131         int retval = 1;
12132
12133         CTL_DEBUG_PRINT(("ctl_run_task\n"));
12134         KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
12135             ("ctl_run_task: Unextected io_type %d\n", io->io_hdr.io_type));
12136         io->taskio.task_status = CTL_TASK_FUNCTION_NOT_SUPPORTED;
12137         bzero(io->taskio.task_resp, sizeof(io->taskio.task_resp));
12138         switch (io->taskio.task_action) {
12139         case CTL_TASK_ABORT_TASK:
12140                 retval = ctl_abort_task(io);
12141                 break;
12142         case CTL_TASK_ABORT_TASK_SET:
12143         case CTL_TASK_CLEAR_TASK_SET:
12144                 retval = ctl_abort_task_set(io);
12145                 break;
12146         case CTL_TASK_CLEAR_ACA:
12147                 break;
12148         case CTL_TASK_I_T_NEXUS_RESET:
12149                 retval = ctl_i_t_nexus_reset(io);
12150                 break;
12151         case CTL_TASK_LUN_RESET:
12152                 retval = ctl_lun_reset(softc, io);
12153                 break;
12154         case CTL_TASK_TARGET_RESET:
12155                 retval = ctl_target_reset(softc, io, CTL_UA_TARG_RESET);
12156                 break;
12157         case CTL_TASK_BUS_RESET:
12158                 retval = ctl_bus_reset(softc, io);
12159                 break;
12160         case CTL_TASK_PORT_LOGIN:
12161                 break;
12162         case CTL_TASK_PORT_LOGOUT:
12163                 break;
12164         case CTL_TASK_QUERY_TASK:
12165                 retval = ctl_query_task(io, 0);
12166                 break;
12167         case CTL_TASK_QUERY_TASK_SET:
12168                 retval = ctl_query_task(io, 1);
12169                 break;
12170         case CTL_TASK_QUERY_ASYNC_EVENT:
12171                 retval = ctl_query_async_event(io);
12172                 break;
12173         default:
12174                 printf("%s: got unknown task management event %d\n",
12175                        __func__, io->taskio.task_action);
12176                 break;
12177         }
12178         if (retval == 0)
12179                 io->io_hdr.status = CTL_SUCCESS;
12180         else
12181                 io->io_hdr.status = CTL_ERROR;
12182         ctl_done(io);
12183 }
12184
12185 /*
12186  * For HA operation.  Handle commands that come in from the other
12187  * controller.
12188  */
12189 static void
12190 ctl_handle_isc(union ctl_io *io)
12191 {
12192         int free_io;
12193         struct ctl_lun *lun;
12194         struct ctl_softc *softc;
12195         uint32_t targ_lun;
12196
12197         softc = control_softc;
12198
12199         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12200         lun = softc->ctl_luns[targ_lun];
12201
12202         switch (io->io_hdr.msg_type) {
12203         case CTL_MSG_SERIALIZE:
12204                 free_io = ctl_serialize_other_sc_cmd(&io->scsiio);
12205                 break;
12206         case CTL_MSG_R2R: {
12207                 const struct ctl_cmd_entry *entry;
12208
12209                 /*
12210                  * This is only used in SER_ONLY mode.
12211                  */
12212                 free_io = 0;
12213                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
12214                 mtx_lock(&lun->lun_lock);
12215                 if (ctl_scsiio_lun_check(lun,
12216                     entry, (struct ctl_scsiio *)io) != 0) {
12217                         mtx_unlock(&lun->lun_lock);
12218                         ctl_done(io);
12219                         break;
12220                 }
12221                 io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
12222                 mtx_unlock(&lun->lun_lock);
12223                 ctl_enqueue_rtr(io);
12224                 break;
12225         }
12226         case CTL_MSG_FINISH_IO:
12227                 if (softc->ha_mode == CTL_HA_MODE_XFER) {
12228                         free_io = 0;
12229                         ctl_done(io);
12230                 } else {
12231                         free_io = 1;
12232                         mtx_lock(&lun->lun_lock);
12233                         TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
12234                                      ooa_links);
12235                         ctl_check_blocked(lun);
12236                         mtx_unlock(&lun->lun_lock);
12237                 }
12238                 break;
12239         case CTL_MSG_PERS_ACTION:
12240                 ctl_hndl_per_res_out_on_other_sc(
12241                         (union ctl_ha_msg *)&io->presio.pr_msg);
12242                 free_io = 1;
12243                 break;
12244         case CTL_MSG_BAD_JUJU:
12245                 free_io = 0;
12246                 ctl_done(io);
12247                 break;
12248         case CTL_MSG_DATAMOVE:
12249                 /* Only used in XFER mode */
12250                 free_io = 0;
12251                 ctl_datamove_remote(io);
12252                 break;
12253         case CTL_MSG_DATAMOVE_DONE:
12254                 /* Only used in XFER mode */
12255                 free_io = 0;
12256                 io->scsiio.be_move_done(io);
12257                 break;
12258         case CTL_MSG_FAILOVER:
12259                 mtx_lock(&lun->lun_lock);
12260                 ctl_failover_lun(lun);
12261                 mtx_unlock(&lun->lun_lock);
12262                 free_io = 1;
12263                 break;
12264         default:
12265                 free_io = 1;
12266                 printf("%s: Invalid message type %d\n",
12267                        __func__, io->io_hdr.msg_type);
12268                 break;
12269         }
12270         if (free_io)
12271                 ctl_free_io(io);
12272
12273 }
12274
12275
12276 /*
12277  * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12278  * there is no match.
12279  */
12280 static ctl_lun_error_pattern
12281 ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12282 {
12283         const struct ctl_cmd_entry *entry;
12284         ctl_lun_error_pattern filtered_pattern, pattern;
12285
12286         pattern = desc->error_pattern;
12287
12288         /*
12289          * XXX KDM we need more data passed into this function to match a
12290          * custom pattern, and we actually need to implement custom pattern
12291          * matching.
12292          */
12293         if (pattern & CTL_LUN_PAT_CMD)
12294                 return (CTL_LUN_PAT_CMD);
12295
12296         if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12297                 return (CTL_LUN_PAT_ANY);
12298
12299         entry = ctl_get_cmd_entry(ctsio, NULL);
12300
12301         filtered_pattern = entry->pattern & pattern;
12302
12303         /*
12304          * If the user requested specific flags in the pattern (e.g.
12305          * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12306          * flags.
12307          *
12308          * If the user did not specify any flags, it doesn't matter whether
12309          * or not the command supports the flags.
12310          */
12311         if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12312              (pattern & ~CTL_LUN_PAT_MASK))
12313                 return (CTL_LUN_PAT_NONE);
12314
12315         /*
12316          * If the user asked for a range check, see if the requested LBA
12317          * range overlaps with this command's LBA range.
12318          */
12319         if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12320                 uint64_t lba1;
12321                 uint64_t len1;
12322                 ctl_action action;
12323                 int retval;
12324
12325                 retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12326                 if (retval != 0)
12327                         return (CTL_LUN_PAT_NONE);
12328
12329                 action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12330                                               desc->lba_range.len, FALSE);
12331                 /*
12332                  * A "pass" means that the LBA ranges don't overlap, so
12333                  * this doesn't match the user's range criteria.
12334                  */
12335                 if (action == CTL_ACTION_PASS)
12336                         return (CTL_LUN_PAT_NONE);
12337         }
12338
12339         return (filtered_pattern);
12340 }
12341
12342 static void
12343 ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12344 {
12345         struct ctl_error_desc *desc, *desc2;
12346
12347         mtx_assert(&lun->lun_lock, MA_OWNED);
12348
12349         STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12350                 ctl_lun_error_pattern pattern;
12351                 /*
12352                  * Check to see whether this particular command matches
12353                  * the pattern in the descriptor.
12354                  */
12355                 pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12356                 if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12357                         continue;
12358
12359                 switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12360                 case CTL_LUN_INJ_ABORTED:
12361                         ctl_set_aborted(&io->scsiio);
12362                         break;
12363                 case CTL_LUN_INJ_MEDIUM_ERR:
12364                         ctl_set_medium_error(&io->scsiio,
12365                             (io->io_hdr.flags & CTL_FLAG_DATA_MASK) !=
12366                              CTL_FLAG_DATA_OUT);
12367                         break;
12368                 case CTL_LUN_INJ_UA:
12369                         /* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12370                          * OCCURRED */
12371                         ctl_set_ua(&io->scsiio, 0x29, 0x00);
12372                         break;
12373                 case CTL_LUN_INJ_CUSTOM:
12374                         /*
12375                          * We're assuming the user knows what he is doing.
12376                          * Just copy the sense information without doing
12377                          * checks.
12378                          */
12379                         bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12380                               MIN(sizeof(desc->custom_sense),
12381                                   sizeof(io->scsiio.sense_data)));
12382                         io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12383                         io->scsiio.sense_len = SSD_FULL_SIZE;
12384                         io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12385                         break;
12386                 case CTL_LUN_INJ_NONE:
12387                 default:
12388                         /*
12389                          * If this is an error injection type we don't know
12390                          * about, clear the continuous flag (if it is set)
12391                          * so it will get deleted below.
12392                          */
12393                         desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12394                         break;
12395                 }
12396                 /*
12397                  * By default, each error injection action is a one-shot
12398                  */
12399                 if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12400                         continue;
12401
12402                 STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12403
12404                 free(desc, M_CTL);
12405         }
12406 }
12407
12408 #ifdef CTL_IO_DELAY
12409 static void
12410 ctl_datamove_timer_wakeup(void *arg)
12411 {
12412         union ctl_io *io;
12413
12414         io = (union ctl_io *)arg;
12415
12416         ctl_datamove(io);
12417 }
12418 #endif /* CTL_IO_DELAY */
12419
12420 void
12421 ctl_datamove(union ctl_io *io)
12422 {
12423         struct ctl_lun *lun;
12424         void (*fe_datamove)(union ctl_io *io);
12425
12426         mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
12427
12428         CTL_DEBUG_PRINT(("ctl_datamove\n"));
12429
12430         lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12431 #ifdef CTL_TIME_IO
12432         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12433                 char str[256];
12434                 char path_str[64];
12435                 struct sbuf sb;
12436
12437                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
12438                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12439
12440                 sbuf_cat(&sb, path_str);
12441                 switch (io->io_hdr.io_type) {
12442                 case CTL_IO_SCSI:
12443                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12444                         sbuf_printf(&sb, "\n");
12445                         sbuf_cat(&sb, path_str);
12446                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12447                                     io->scsiio.tag_num, io->scsiio.tag_type);
12448                         break;
12449                 case CTL_IO_TASK:
12450                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12451                                     "Tag Type: %d\n", io->taskio.task_action,
12452                                     io->taskio.tag_num, io->taskio.tag_type);
12453                         break;
12454                 default:
12455                         printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12456                         panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12457                         break;
12458                 }
12459                 sbuf_cat(&sb, path_str);
12460                 sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12461                             (intmax_t)time_uptime - io->io_hdr.start_time);
12462                 sbuf_finish(&sb);
12463                 printf("%s", sbuf_data(&sb));
12464         }
12465 #endif /* CTL_TIME_IO */
12466
12467 #ifdef CTL_IO_DELAY
12468         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12469                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12470         } else {
12471                 if ((lun != NULL)
12472                  && (lun->delay_info.datamove_delay > 0)) {
12473
12474                         callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
12475                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12476                         callout_reset(&io->io_hdr.delay_callout,
12477                                       lun->delay_info.datamove_delay * hz,
12478                                       ctl_datamove_timer_wakeup, io);
12479                         if (lun->delay_info.datamove_type ==
12480                             CTL_DELAY_TYPE_ONESHOT)
12481                                 lun->delay_info.datamove_delay = 0;
12482                         return;
12483                 }
12484         }
12485 #endif
12486
12487         /*
12488          * This command has been aborted.  Set the port status, so we fail
12489          * the data move.
12490          */
12491         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12492                 printf("ctl_datamove: tag 0x%04x on (%u:%u:%u) aborted\n",
12493                        io->scsiio.tag_num, io->io_hdr.nexus.initid,
12494                        io->io_hdr.nexus.targ_port,
12495                        io->io_hdr.nexus.targ_lun);
12496                 io->io_hdr.port_status = 31337;
12497                 /*
12498                  * Note that the backend, in this case, will get the
12499                  * callback in its context.  In other cases it may get
12500                  * called in the frontend's interrupt thread context.
12501                  */
12502                 io->scsiio.be_move_done(io);
12503                 return;
12504         }
12505
12506         /* Don't confuse frontend with zero length data move. */
12507         if (io->scsiio.kern_data_len == 0) {
12508                 io->scsiio.be_move_done(io);
12509                 return;
12510         }
12511
12512         /*
12513          * If we're in XFER mode and this I/O is from the other shelf
12514          * controller, we need to send the DMA to the other side to
12515          * actually transfer the data to/from the host.  In serialize only
12516          * mode the transfer happens below CTL and ctl_datamove() is only
12517          * called on the machine that originally received the I/O.
12518          */
12519         if ((control_softc->ha_mode == CTL_HA_MODE_XFER)
12520          && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12521                 union ctl_ha_msg msg;
12522                 uint32_t sg_entries_sent;
12523                 int do_sg_copy;
12524                 int i;
12525
12526                 memset(&msg, 0, sizeof(msg));
12527                 msg.hdr.msg_type = CTL_MSG_DATAMOVE;
12528                 msg.hdr.original_sc = io->io_hdr.original_sc;
12529                 msg.hdr.serializing_sc = io;
12530                 msg.hdr.nexus = io->io_hdr.nexus;
12531                 msg.hdr.status = io->io_hdr.status;
12532                 msg.dt.flags = io->io_hdr.flags;
12533                 /*
12534                  * We convert everything into a S/G list here.  We can't
12535                  * pass by reference, only by value between controllers.
12536                  * So we can't pass a pointer to the S/G list, only as many
12537                  * S/G entries as we can fit in here.  If it's possible for
12538                  * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
12539                  * then we need to break this up into multiple transfers.
12540                  */
12541                 if (io->scsiio.kern_sg_entries == 0) {
12542                         msg.dt.kern_sg_entries = 1;
12543 #if 0
12544                         /*
12545                          * Convert to a physical address if this is a
12546                          * virtual address.
12547                          */
12548                         if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
12549                                 msg.dt.sg_list[0].addr =
12550                                         io->scsiio.kern_data_ptr;
12551                         } else {
12552                                 /*
12553                                  * XXX KDM use busdma here!
12554                                  */
12555                                 msg.dt.sg_list[0].addr = (void *)
12556                                         vtophys(io->scsiio.kern_data_ptr);
12557                         }
12558 #else
12559                         KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0,
12560                             ("HA does not support BUS_ADDR"));
12561                         msg.dt.sg_list[0].addr = io->scsiio.kern_data_ptr;
12562 #endif
12563
12564                         msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
12565                         do_sg_copy = 0;
12566                 } else {
12567                         msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
12568                         do_sg_copy = 1;
12569                 }
12570
12571                 msg.dt.kern_data_len = io->scsiio.kern_data_len;
12572                 msg.dt.kern_total_len = io->scsiio.kern_total_len;
12573                 msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
12574                 msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
12575                 msg.dt.sg_sequence = 0;
12576
12577                 /*
12578                  * Loop until we've sent all of the S/G entries.  On the
12579                  * other end, we'll recompose these S/G entries into one
12580                  * contiguous list before passing it to the
12581                  */
12582                 for (sg_entries_sent = 0; sg_entries_sent <
12583                      msg.dt.kern_sg_entries; msg.dt.sg_sequence++) {
12584                         msg.dt.cur_sg_entries = MIN((sizeof(msg.dt.sg_list)/
12585                                 sizeof(msg.dt.sg_list[0])),
12586                                 msg.dt.kern_sg_entries - sg_entries_sent);
12587
12588                         if (do_sg_copy != 0) {
12589                                 struct ctl_sg_entry *sgl;
12590                                 int j;
12591
12592                                 sgl = (struct ctl_sg_entry *)
12593                                         io->scsiio.kern_data_ptr;
12594                                 /*
12595                                  * If this is in cached memory, flush the cache
12596                                  * before we send the DMA request to the other
12597                                  * controller.  We want to do this in either
12598                                  * the * read or the write case.  The read
12599                                  * case is straightforward.  In the write
12600                                  * case, we want to make sure nothing is
12601                                  * in the local cache that could overwrite
12602                                  * the DMAed data.
12603                                  */
12604
12605                                 for (i = sg_entries_sent, j = 0;
12606                                      i < msg.dt.cur_sg_entries; i++, j++) {
12607 #if 0
12608                                         if ((io->io_hdr.flags &
12609                                              CTL_FLAG_BUS_ADDR) == 0) {
12610                                                 /*
12611                                                  * XXX KDM use busdma.
12612                                                  */
12613                                                 msg.dt.sg_list[j].addr =(void *)
12614                                                        vtophys(sgl[i].addr);
12615                                         } else {
12616                                                 msg.dt.sg_list[j].addr =
12617                                                         sgl[i].addr;
12618                                         }
12619 #else
12620                                         KASSERT((io->io_hdr.flags &
12621                                             CTL_FLAG_BUS_ADDR) == 0,
12622                                             ("HA does not support BUS_ADDR"));
12623                                         msg.dt.sg_list[j].addr = sgl[i].addr;
12624 #endif
12625                                         msg.dt.sg_list[j].len = sgl[i].len;
12626                                 }
12627                         }
12628
12629                         sg_entries_sent += msg.dt.cur_sg_entries;
12630                         if (sg_entries_sent >= msg.dt.kern_sg_entries)
12631                                 msg.dt.sg_last = 1;
12632                         else
12633                                 msg.dt.sg_last = 0;
12634
12635                         if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12636                             sizeof(msg.dt) - sizeof(msg.dt.sg_list) +
12637                             sizeof(struct ctl_sg_entry)*msg.dt.cur_sg_entries,
12638                             M_WAITOK) > CTL_HA_STATUS_SUCCESS) {
12639                                 io->io_hdr.port_status = 31341;
12640                                 io->scsiio.be_move_done(io);
12641                                 return;
12642                         }
12643
12644                         msg.dt.sent_sg_entries = sg_entries_sent;
12645                 }
12646
12647                 /*
12648                  * Officially handover the request from us to peer.
12649                  * If failover has just happened, then we must return error.
12650                  * If failover happen just after, then it is not our problem.
12651                  */
12652                 if (lun)
12653                         mtx_lock(&lun->lun_lock);
12654                 if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12655                         if (lun)
12656                                 mtx_unlock(&lun->lun_lock);
12657                         io->io_hdr.port_status = 31342;
12658                         io->scsiio.be_move_done(io);
12659                         return;
12660                 }
12661                 io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12662                 io->io_hdr.flags |= CTL_FLAG_DMA_INPROG;
12663                 if (lun)
12664                         mtx_unlock(&lun->lun_lock);
12665         } else {
12666
12667                 /*
12668                  * Lookup the fe_datamove() function for this particular
12669                  * front end.
12670                  */
12671                 fe_datamove = ctl_io_port(&io->io_hdr)->fe_datamove;
12672
12673                 fe_datamove(io);
12674         }
12675 }
12676
12677 static void
12678 ctl_send_datamove_done(union ctl_io *io, int have_lock)
12679 {
12680         union ctl_ha_msg msg;
12681
12682         memset(&msg, 0, sizeof(msg));
12683
12684         msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12685         msg.hdr.original_sc = io;
12686         msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12687         msg.hdr.nexus = io->io_hdr.nexus;
12688         msg.hdr.status = io->io_hdr.status;
12689         msg.scsi.tag_num = io->scsiio.tag_num;
12690         msg.scsi.tag_type = io->scsiio.tag_type;
12691         msg.scsi.scsi_status = io->scsiio.scsi_status;
12692         memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12693                io->scsiio.sense_len);
12694         msg.scsi.sense_len = io->scsiio.sense_len;
12695         msg.scsi.sense_residual = io->scsiio.sense_residual;
12696         msg.scsi.fetd_status = io->io_hdr.port_status;
12697         msg.scsi.residual = io->scsiio.residual;
12698         io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12699
12700         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12701                 ctl_failover_io(io, /*have_lock*/ have_lock);
12702                 return;
12703         }
12704
12705         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12706             sizeof(msg.scsi) - sizeof(msg.scsi.sense_data) +
12707             msg.scsi.sense_len, M_WAITOK);
12708 }
12709
12710 /*
12711  * The DMA to the remote side is done, now we need to tell the other side
12712  * we're done so it can continue with its data movement.
12713  */
12714 static void
12715 ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12716 {
12717         union ctl_io *io;
12718         int i;
12719
12720         io = rq->context;
12721
12722         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12723                 printf("%s: ISC DMA write failed with error %d", __func__,
12724                        rq->ret);
12725                 ctl_set_internal_failure(&io->scsiio,
12726                                          /*sks_valid*/ 1,
12727                                          /*retry_count*/ rq->ret);
12728         }
12729
12730         ctl_dt_req_free(rq);
12731
12732         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12733                 free(io->io_hdr.local_sglist[i].addr, M_CTL);
12734         free(io->io_hdr.remote_sglist, M_CTL);
12735         io->io_hdr.remote_sglist = NULL;
12736         io->io_hdr.local_sglist = NULL;
12737
12738         /*
12739          * The data is in local and remote memory, so now we need to send
12740          * status (good or back) back to the other side.
12741          */
12742         ctl_send_datamove_done(io, /*have_lock*/ 0);
12743 }
12744
12745 /*
12746  * We've moved the data from the host/controller into local memory.  Now we
12747  * need to push it over to the remote controller's memory.
12748  */
12749 static int
12750 ctl_datamove_remote_dm_write_cb(union ctl_io *io)
12751 {
12752         int retval;
12753
12754         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12755                                           ctl_datamove_remote_write_cb);
12756         return (retval);
12757 }
12758
12759 static void
12760 ctl_datamove_remote_write(union ctl_io *io)
12761 {
12762         int retval;
12763         void (*fe_datamove)(union ctl_io *io);
12764
12765         /*
12766          * - Get the data from the host/HBA into local memory.
12767          * - DMA memory from the local controller to the remote controller.
12768          * - Send status back to the remote controller.
12769          */
12770
12771         retval = ctl_datamove_remote_sgl_setup(io);
12772         if (retval != 0)
12773                 return;
12774
12775         /* Switch the pointer over so the FETD knows what to do */
12776         io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12777
12778         /*
12779          * Use a custom move done callback, since we need to send completion
12780          * back to the other controller, not to the backend on this side.
12781          */
12782         io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12783
12784         fe_datamove = ctl_io_port(&io->io_hdr)->fe_datamove;
12785         fe_datamove(io);
12786 }
12787
12788 static int
12789 ctl_datamove_remote_dm_read_cb(union ctl_io *io)
12790 {
12791 #if 0
12792         char str[256];
12793         char path_str[64];
12794         struct sbuf sb;
12795 #endif
12796         int i;
12797
12798         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12799                 free(io->io_hdr.local_sglist[i].addr, M_CTL);
12800         free(io->io_hdr.remote_sglist, M_CTL);
12801         io->io_hdr.remote_sglist = NULL;
12802         io->io_hdr.local_sglist = NULL;
12803
12804 #if 0
12805         scsi_path_string(io, path_str, sizeof(path_str));
12806         sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12807         sbuf_cat(&sb, path_str);
12808         scsi_command_string(&io->scsiio, NULL, &sb);
12809         sbuf_printf(&sb, "\n");
12810         sbuf_cat(&sb, path_str);
12811         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12812                     io->scsiio.tag_num, io->scsiio.tag_type);
12813         sbuf_cat(&sb, path_str);
12814         sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
12815                     io->io_hdr.flags, io->io_hdr.status);
12816         sbuf_finish(&sb);
12817         printk("%s", sbuf_data(&sb));
12818 #endif
12819
12820
12821         /*
12822          * The read is done, now we need to send status (good or bad) back
12823          * to the other side.
12824          */
12825         ctl_send_datamove_done(io, /*have_lock*/ 0);
12826
12827         return (0);
12828 }
12829
12830 static void
12831 ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
12832 {
12833         union ctl_io *io;
12834         void (*fe_datamove)(union ctl_io *io);
12835
12836         io = rq->context;
12837
12838         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12839                 printf("%s: ISC DMA read failed with error %d\n", __func__,
12840                        rq->ret);
12841                 ctl_set_internal_failure(&io->scsiio,
12842                                          /*sks_valid*/ 1,
12843                                          /*retry_count*/ rq->ret);
12844         }
12845
12846         ctl_dt_req_free(rq);
12847
12848         /* Switch the pointer over so the FETD knows what to do */
12849         io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12850
12851         /*
12852          * Use a custom move done callback, since we need to send completion
12853          * back to the other controller, not to the backend on this side.
12854          */
12855         io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
12856
12857         /* XXX KDM add checks like the ones in ctl_datamove? */
12858
12859         fe_datamove = ctl_io_port(&io->io_hdr)->fe_datamove;
12860         fe_datamove(io);
12861 }
12862
12863 static int
12864 ctl_datamove_remote_sgl_setup(union ctl_io *io)
12865 {
12866         struct ctl_sg_entry *local_sglist;
12867         struct ctl_softc *softc;
12868         uint32_t len_to_go;
12869         int retval;
12870         int i;
12871
12872         retval = 0;
12873         softc = control_softc;
12874         local_sglist = io->io_hdr.local_sglist;
12875         len_to_go = io->scsiio.kern_data_len;
12876
12877         /*
12878          * The difficult thing here is that the size of the various
12879          * S/G segments may be different than the size from the
12880          * remote controller.  That'll make it harder when DMAing
12881          * the data back to the other side.
12882          */
12883         for (i = 0; len_to_go > 0; i++) {
12884                 local_sglist[i].len = MIN(len_to_go, CTL_HA_DATAMOVE_SEGMENT);
12885                 local_sglist[i].addr =
12886                     malloc(local_sglist[i].len, M_CTL, M_WAITOK);
12887
12888                 len_to_go -= local_sglist[i].len;
12889         }
12890         /*
12891          * Reset the number of S/G entries accordingly.  The original
12892          * number of S/G entries is available in rem_sg_entries.
12893          */
12894         io->scsiio.kern_sg_entries = i;
12895
12896 #if 0
12897         printf("%s: kern_sg_entries = %d\n", __func__,
12898                io->scsiio.kern_sg_entries);
12899         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12900                 printf("%s: sg[%d] = %p, %lu\n", __func__, i,
12901                        local_sglist[i].addr, local_sglist[i].len);
12902 #endif
12903
12904         return (retval);
12905 }
12906
12907 static int
12908 ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
12909                          ctl_ha_dt_cb callback)
12910 {
12911         struct ctl_ha_dt_req *rq;
12912         struct ctl_sg_entry *remote_sglist, *local_sglist;
12913         uint32_t local_used, remote_used, total_used;
12914         int i, j, isc_ret;
12915
12916         rq = ctl_dt_req_alloc();
12917
12918         /*
12919          * If we failed to allocate the request, and if the DMA didn't fail
12920          * anyway, set busy status.  This is just a resource allocation
12921          * failure.
12922          */
12923         if ((rq == NULL)
12924          && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
12925              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS))
12926                 ctl_set_busy(&io->scsiio);
12927
12928         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
12929             (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) {
12930
12931                 if (rq != NULL)
12932                         ctl_dt_req_free(rq);
12933
12934                 /*
12935                  * The data move failed.  We need to return status back
12936                  * to the other controller.  No point in trying to DMA
12937                  * data to the remote controller.
12938                  */
12939
12940                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12941
12942                 return (1);
12943         }
12944
12945         local_sglist = io->io_hdr.local_sglist;
12946         remote_sglist = io->io_hdr.remote_sglist;
12947         local_used = 0;
12948         remote_used = 0;
12949         total_used = 0;
12950
12951         /*
12952          * Pull/push the data over the wire from/to the other controller.
12953          * This takes into account the possibility that the local and
12954          * remote sglists may not be identical in terms of the size of
12955          * the elements and the number of elements.
12956          *
12957          * One fundamental assumption here is that the length allocated for
12958          * both the local and remote sglists is identical.  Otherwise, we've
12959          * essentially got a coding error of some sort.
12960          */
12961         isc_ret = CTL_HA_STATUS_SUCCESS;
12962         for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
12963                 uint32_t cur_len;
12964                 uint8_t *tmp_ptr;
12965
12966                 rq->command = command;
12967                 rq->context = io;
12968
12969                 /*
12970                  * Both pointers should be aligned.  But it is possible
12971                  * that the allocation length is not.  They should both
12972                  * also have enough slack left over at the end, though,
12973                  * to round up to the next 8 byte boundary.
12974                  */
12975                 cur_len = MIN(local_sglist[i].len - local_used,
12976                               remote_sglist[j].len - remote_used);
12977                 rq->size = cur_len;
12978
12979                 tmp_ptr = (uint8_t *)local_sglist[i].addr;
12980                 tmp_ptr += local_used;
12981
12982 #if 0
12983                 /* Use physical addresses when talking to ISC hardware */
12984                 if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
12985                         /* XXX KDM use busdma */
12986                         rq->local = vtophys(tmp_ptr);
12987                 } else
12988                         rq->local = tmp_ptr;
12989 #else
12990                 KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0,
12991                     ("HA does not support BUS_ADDR"));
12992                 rq->local = tmp_ptr;
12993 #endif
12994
12995                 tmp_ptr = (uint8_t *)remote_sglist[j].addr;
12996                 tmp_ptr += remote_used;
12997                 rq->remote = tmp_ptr;
12998
12999                 rq->callback = NULL;
13000
13001                 local_used += cur_len;
13002                 if (local_used >= local_sglist[i].len) {
13003                         i++;
13004                         local_used = 0;
13005                 }
13006
13007                 remote_used += cur_len;
13008                 if (remote_used >= remote_sglist[j].len) {
13009                         j++;
13010                         remote_used = 0;
13011                 }
13012                 total_used += cur_len;
13013
13014                 if (total_used >= io->scsiio.kern_data_len)
13015                         rq->callback = callback;
13016
13017 #if 0
13018                 printf("%s: %s: local %p remote %p size %d\n", __func__,
13019                        (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
13020                        rq->local, rq->remote, rq->size);
13021 #endif
13022
13023                 isc_ret = ctl_dt_single(rq);
13024                 if (isc_ret > CTL_HA_STATUS_SUCCESS)
13025                         break;
13026         }
13027         if (isc_ret != CTL_HA_STATUS_WAIT) {
13028                 rq->ret = isc_ret;
13029                 callback(rq);
13030         }
13031
13032         return (0);
13033 }
13034
13035 static void
13036 ctl_datamove_remote_read(union ctl_io *io)
13037 {
13038         int retval;
13039         int i;
13040
13041         /*
13042          * This will send an error to the other controller in the case of a
13043          * failure.
13044          */
13045         retval = ctl_datamove_remote_sgl_setup(io);
13046         if (retval != 0)
13047                 return;
13048
13049         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
13050                                           ctl_datamove_remote_read_cb);
13051         if (retval != 0) {
13052                 /*
13053                  * Make sure we free memory if there was an error..  The
13054                  * ctl_datamove_remote_xfer() function will send the
13055                  * datamove done message, or call the callback with an
13056                  * error if there is a problem.
13057                  */
13058                 for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13059                         free(io->io_hdr.local_sglist[i].addr, M_CTL);
13060                 free(io->io_hdr.remote_sglist, M_CTL);
13061                 io->io_hdr.remote_sglist = NULL;
13062                 io->io_hdr.local_sglist = NULL;
13063         }
13064 }
13065
13066 /*
13067  * Process a datamove request from the other controller.  This is used for
13068  * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
13069  * first.  Once that is complete, the data gets DMAed into the remote
13070  * controller's memory.  For reads, we DMA from the remote controller's
13071  * memory into our memory first, and then move it out to the FETD.
13072  */
13073 static void
13074 ctl_datamove_remote(union ctl_io *io)
13075 {
13076
13077         mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
13078
13079         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
13080                 ctl_failover_io(io, /*have_lock*/ 0);
13081                 return;
13082         }
13083
13084         /*
13085          * Note that we look for an aborted I/O here, but don't do some of
13086          * the other checks that ctl_datamove() normally does.
13087          * We don't need to run the datamove delay code, since that should
13088          * have been done if need be on the other controller.
13089          */
13090         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
13091                 printf("%s: tag 0x%04x on (%u:%u:%u) aborted\n", __func__,
13092                        io->scsiio.tag_num, io->io_hdr.nexus.initid,
13093                        io->io_hdr.nexus.targ_port,
13094                        io->io_hdr.nexus.targ_lun);
13095                 io->io_hdr.port_status = 31338;
13096                 ctl_send_datamove_done(io, /*have_lock*/ 0);
13097                 return;
13098         }
13099
13100         if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT)
13101                 ctl_datamove_remote_write(io);
13102         else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
13103                 ctl_datamove_remote_read(io);
13104         else {
13105                 io->io_hdr.port_status = 31339;
13106                 ctl_send_datamove_done(io, /*have_lock*/ 0);
13107         }
13108 }
13109
13110 static int
13111 ctl_process_done(union ctl_io *io)
13112 {
13113         struct ctl_lun *lun;
13114         struct ctl_softc *softc = control_softc;
13115         void (*fe_done)(union ctl_io *io);
13116         union ctl_ha_msg msg;
13117         uint32_t targ_port = io->io_hdr.nexus.targ_port;
13118
13119         CTL_DEBUG_PRINT(("ctl_process_done\n"));
13120
13121         if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0)
13122                 fe_done = softc->ctl_ports[targ_port]->fe_done;
13123         else
13124                 fe_done = NULL;
13125
13126 #ifdef CTL_TIME_IO
13127         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
13128                 char str[256];
13129                 char path_str[64];
13130                 struct sbuf sb;
13131
13132                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
13133                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13134
13135                 sbuf_cat(&sb, path_str);
13136                 switch (io->io_hdr.io_type) {
13137                 case CTL_IO_SCSI:
13138                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
13139                         sbuf_printf(&sb, "\n");
13140                         sbuf_cat(&sb, path_str);
13141                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13142                                     io->scsiio.tag_num, io->scsiio.tag_type);
13143                         break;
13144                 case CTL_IO_TASK:
13145                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
13146                                     "Tag Type: %d\n", io->taskio.task_action,
13147                                     io->taskio.tag_num, io->taskio.tag_type);
13148                         break;
13149                 default:
13150                         printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13151                         panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13152                         break;
13153                 }
13154                 sbuf_cat(&sb, path_str);
13155                 sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
13156                             (intmax_t)time_uptime - io->io_hdr.start_time);
13157                 sbuf_finish(&sb);
13158                 printf("%s", sbuf_data(&sb));
13159         }
13160 #endif /* CTL_TIME_IO */
13161
13162         switch (io->io_hdr.io_type) {
13163         case CTL_IO_SCSI:
13164                 break;
13165         case CTL_IO_TASK:
13166                 if (ctl_debug & CTL_DEBUG_INFO)
13167                         ctl_io_error_print(io, NULL);
13168                 if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
13169                         ctl_free_io(io);
13170                 else
13171                         fe_done(io);
13172                 return (CTL_RETVAL_COMPLETE);
13173         default:
13174                 panic("ctl_process_done: invalid io type %d\n",
13175                       io->io_hdr.io_type);
13176                 break; /* NOTREACHED */
13177         }
13178
13179         lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13180         if (lun == NULL) {
13181                 CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
13182                                  io->io_hdr.nexus.targ_mapped_lun));
13183                 goto bailout;
13184         }
13185
13186         mtx_lock(&lun->lun_lock);
13187
13188         /*
13189          * Check to see if we have any errors to inject here.  We only
13190          * inject errors for commands that don't already have errors set.
13191          */
13192         if ((STAILQ_FIRST(&lun->error_list) != NULL) &&
13193             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) &&
13194             ((io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0))
13195                 ctl_inject_error(lun, io);
13196
13197         /*
13198          * XXX KDM how do we treat commands that aren't completed
13199          * successfully?
13200          *
13201          * XXX KDM should we also track I/O latency?
13202          */
13203         if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
13204             io->io_hdr.io_type == CTL_IO_SCSI) {
13205 #ifdef CTL_TIME_IO
13206                 struct bintime cur_bt;
13207 #endif
13208                 int type;
13209
13210                 if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13211                     CTL_FLAG_DATA_IN)
13212                         type = CTL_STATS_READ;
13213                 else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13214                     CTL_FLAG_DATA_OUT)
13215                         type = CTL_STATS_WRITE;
13216                 else
13217                         type = CTL_STATS_NO_IO;
13218
13219                 lun->stats.ports[targ_port].bytes[type] +=
13220                     io->scsiio.kern_total_len;
13221                 lun->stats.ports[targ_port].operations[type]++;
13222 #ifdef CTL_TIME_IO
13223                 bintime_add(&lun->stats.ports[targ_port].dma_time[type],
13224                    &io->io_hdr.dma_bt);
13225                 lun->stats.ports[targ_port].num_dmas[type] +=
13226                     io->io_hdr.num_dmas;
13227                 getbintime(&cur_bt);
13228                 bintime_sub(&cur_bt, &io->io_hdr.start_bt);
13229                 bintime_add(&lun->stats.ports[targ_port].time[type], &cur_bt);
13230 #endif
13231         }
13232
13233         /*
13234          * Remove this from the OOA queue.
13235          */
13236         TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13237 #ifdef CTL_TIME_IO
13238         if (TAILQ_EMPTY(&lun->ooa_queue))
13239                 lun->last_busy = getsbinuptime();
13240 #endif
13241
13242         /*
13243          * Run through the blocked queue on this LUN and see if anything
13244          * has become unblocked, now that this transaction is done.
13245          */
13246         ctl_check_blocked(lun);
13247
13248         /*
13249          * If the LUN has been invalidated, free it if there is nothing
13250          * left on its OOA queue.
13251          */
13252         if ((lun->flags & CTL_LUN_INVALID)
13253          && TAILQ_EMPTY(&lun->ooa_queue)) {
13254                 mtx_unlock(&lun->lun_lock);
13255                 mtx_lock(&softc->ctl_lock);
13256                 ctl_free_lun(lun);
13257                 mtx_unlock(&softc->ctl_lock);
13258         } else
13259                 mtx_unlock(&lun->lun_lock);
13260
13261 bailout:
13262
13263         /*
13264          * If this command has been aborted, make sure we set the status
13265          * properly.  The FETD is responsible for freeing the I/O and doing
13266          * whatever it needs to do to clean up its state.
13267          */
13268         if (io->io_hdr.flags & CTL_FLAG_ABORT)
13269                 ctl_set_task_aborted(&io->scsiio);
13270
13271         /*
13272          * If enabled, print command error status.
13273          */
13274         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS &&
13275             (ctl_debug & CTL_DEBUG_INFO) != 0)
13276                 ctl_io_error_print(io, NULL);
13277
13278         /*
13279          * Tell the FETD or the other shelf controller we're done with this
13280          * command.  Note that only SCSI commands get to this point.  Task
13281          * management commands are completed above.
13282          */
13283         if ((softc->ha_mode != CTL_HA_MODE_XFER) &&
13284             (io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)) {
13285                 memset(&msg, 0, sizeof(msg));
13286                 msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13287                 msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
13288                 msg.hdr.nexus = io->io_hdr.nexus;
13289                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13290                     sizeof(msg.scsi) - sizeof(msg.scsi.sense_data),
13291                     M_WAITOK);
13292         }
13293         if ((softc->ha_mode == CTL_HA_MODE_XFER)
13294          && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
13295                 memset(&msg, 0, sizeof(msg));
13296                 msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13297                 msg.hdr.original_sc = io->io_hdr.original_sc;
13298                 msg.hdr.nexus = io->io_hdr.nexus;
13299                 msg.hdr.status = io->io_hdr.status;
13300                 msg.scsi.scsi_status = io->scsiio.scsi_status;
13301                 msg.scsi.tag_num = io->scsiio.tag_num;
13302                 msg.scsi.tag_type = io->scsiio.tag_type;
13303                 msg.scsi.sense_len = io->scsiio.sense_len;
13304                 msg.scsi.sense_residual = io->scsiio.sense_residual;
13305                 msg.scsi.residual = io->scsiio.residual;
13306                 memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
13307                        io->scsiio.sense_len);
13308
13309                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13310                     sizeof(msg.scsi) - sizeof(msg.scsi.sense_data) +
13311                     msg.scsi.sense_len, M_WAITOK);
13312                 ctl_free_io(io);
13313         } else 
13314                 fe_done(io);
13315
13316         return (CTL_RETVAL_COMPLETE);
13317 }
13318
13319 #ifdef CTL_WITH_CA
13320 /*
13321  * Front end should call this if it doesn't do autosense.  When the request
13322  * sense comes back in from the initiator, we'll dequeue this and send it.
13323  */
13324 int
13325 ctl_queue_sense(union ctl_io *io)
13326 {
13327         struct ctl_lun *lun;
13328         struct ctl_port *port;
13329         struct ctl_softc *softc;
13330         uint32_t initidx, targ_lun;
13331
13332         softc = control_softc;
13333
13334         CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13335
13336         /*
13337          * LUN lookup will likely move to the ctl_work_thread() once we
13338          * have our new queueing infrastructure (that doesn't put things on
13339          * a per-LUN queue initially).  That is so that we can handle
13340          * things like an INQUIRY to a LUN that we don't have enabled.  We
13341          * can't deal with that right now.
13342          */
13343         mtx_lock(&softc->ctl_lock);
13344
13345         /*
13346          * If we don't have a LUN for this, just toss the sense
13347          * information.
13348          */
13349         port = ctl_io_port(&ctsio->io_hdr);
13350         targ_lun = ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13351         if ((targ_lun < CTL_MAX_LUNS)
13352          && (softc->ctl_luns[targ_lun] != NULL))
13353                 lun = softc->ctl_luns[targ_lun];
13354         else
13355                 goto bailout;
13356
13357         initidx = ctl_get_initindex(&io->io_hdr.nexus);
13358
13359         mtx_lock(&lun->lun_lock);
13360         /*
13361          * Already have CA set for this LUN...toss the sense information.
13362          */
13363         if (ctl_is_set(lun->have_ca, initidx)) {
13364                 mtx_unlock(&lun->lun_lock);
13365                 goto bailout;
13366         }
13367
13368         memcpy(&lun->pending_sense[initidx], &io->scsiio.sense_data,
13369                MIN(sizeof(lun->pending_sense[initidx]),
13370                sizeof(io->scsiio.sense_data)));
13371         ctl_set_mask(lun->have_ca, initidx);
13372         mtx_unlock(&lun->lun_lock);
13373
13374 bailout:
13375         mtx_unlock(&softc->ctl_lock);
13376
13377         ctl_free_io(io);
13378
13379         return (CTL_RETVAL_COMPLETE);
13380 }
13381 #endif
13382
13383 /*
13384  * Primary command inlet from frontend ports.  All SCSI and task I/O
13385  * requests must go through this function.
13386  */
13387 int
13388 ctl_queue(union ctl_io *io)
13389 {
13390         struct ctl_port *port;
13391
13392         CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13393
13394 #ifdef CTL_TIME_IO
13395         io->io_hdr.start_time = time_uptime;
13396         getbintime(&io->io_hdr.start_bt);
13397 #endif /* CTL_TIME_IO */
13398
13399         /* Map FE-specific LUN ID into global one. */
13400         port = ctl_io_port(&io->io_hdr);
13401         io->io_hdr.nexus.targ_mapped_lun =
13402             ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13403
13404         switch (io->io_hdr.io_type) {
13405         case CTL_IO_SCSI:
13406         case CTL_IO_TASK:
13407                 if (ctl_debug & CTL_DEBUG_CDB)
13408                         ctl_io_print(io);
13409                 ctl_enqueue_incoming(io);
13410                 break;
13411         default:
13412                 printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13413                 return (EINVAL);
13414         }
13415
13416         return (CTL_RETVAL_COMPLETE);
13417 }
13418
13419 #ifdef CTL_IO_DELAY
13420 static void
13421 ctl_done_timer_wakeup(void *arg)
13422 {
13423         union ctl_io *io;
13424
13425         io = (union ctl_io *)arg;
13426         ctl_done(io);
13427 }
13428 #endif /* CTL_IO_DELAY */
13429
13430 void
13431 ctl_serseq_done(union ctl_io *io)
13432 {
13433         struct ctl_lun *lun;
13434
13435         lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13436         if (lun->be_lun == NULL ||
13437             lun->be_lun->serseq == CTL_LUN_SERSEQ_OFF)
13438                 return;
13439         mtx_lock(&lun->lun_lock);
13440         io->io_hdr.flags |= CTL_FLAG_SERSEQ_DONE;
13441         ctl_check_blocked(lun);
13442         mtx_unlock(&lun->lun_lock);
13443 }
13444
13445 void
13446 ctl_done(union ctl_io *io)
13447 {
13448
13449         /*
13450          * Enable this to catch duplicate completion issues.
13451          */
13452 #if 0
13453         if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13454                 printf("%s: type %d msg %d cdb %x iptl: "
13455                        "%u:%u:%u tag 0x%04x "
13456                        "flag %#x status %x\n",
13457                         __func__,
13458                         io->io_hdr.io_type,
13459                         io->io_hdr.msg_type,
13460                         io->scsiio.cdb[0],
13461                         io->io_hdr.nexus.initid,
13462                         io->io_hdr.nexus.targ_port,
13463                         io->io_hdr.nexus.targ_lun,
13464                         (io->io_hdr.io_type ==
13465                         CTL_IO_TASK) ?
13466                         io->taskio.tag_num :
13467                         io->scsiio.tag_num,
13468                         io->io_hdr.flags,
13469                         io->io_hdr.status);
13470         } else
13471                 io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13472 #endif
13473
13474         /*
13475          * This is an internal copy of an I/O, and should not go through
13476          * the normal done processing logic.
13477          */
13478         if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
13479                 return;
13480
13481 #ifdef CTL_IO_DELAY
13482         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13483                 struct ctl_lun *lun;
13484
13485                 lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13486
13487                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13488         } else {
13489                 struct ctl_lun *lun;
13490
13491                 lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13492
13493                 if ((lun != NULL)
13494                  && (lun->delay_info.done_delay > 0)) {
13495
13496                         callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
13497                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13498                         callout_reset(&io->io_hdr.delay_callout,
13499                                       lun->delay_info.done_delay * hz,
13500                                       ctl_done_timer_wakeup, io);
13501                         if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13502                                 lun->delay_info.done_delay = 0;
13503                         return;
13504                 }
13505         }
13506 #endif /* CTL_IO_DELAY */
13507
13508         ctl_enqueue_done(io);
13509 }
13510
13511 static void
13512 ctl_work_thread(void *arg)
13513 {
13514         struct ctl_thread *thr = (struct ctl_thread *)arg;
13515         struct ctl_softc *softc = thr->ctl_softc;
13516         union ctl_io *io;
13517         int retval;
13518
13519         CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13520
13521         for (;;) {
13522                 retval = 0;
13523
13524                 /*
13525                  * We handle the queues in this order:
13526                  * - ISC
13527                  * - done queue (to free up resources, unblock other commands)
13528                  * - RtR queue
13529                  * - incoming queue
13530                  *
13531                  * If those queues are empty, we break out of the loop and
13532                  * go to sleep.
13533                  */
13534                 mtx_lock(&thr->queue_lock);
13535                 io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
13536                 if (io != NULL) {
13537                         STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
13538                         mtx_unlock(&thr->queue_lock);
13539                         ctl_handle_isc(io);
13540                         continue;
13541                 }
13542                 io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
13543                 if (io != NULL) {
13544                         STAILQ_REMOVE_HEAD(&thr->done_queue, links);
13545                         /* clear any blocked commands, call fe_done */
13546                         mtx_unlock(&thr->queue_lock);
13547                         retval = ctl_process_done(io);
13548                         continue;
13549                 }
13550                 io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
13551                 if (io != NULL) {
13552                         STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
13553                         mtx_unlock(&thr->queue_lock);
13554                         if (io->io_hdr.io_type == CTL_IO_TASK)
13555                                 ctl_run_task(io);
13556                         else
13557                                 ctl_scsiio_precheck(softc, &io->scsiio);
13558                         continue;
13559                 }
13560                 io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
13561                 if (io != NULL) {
13562                         STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
13563                         mtx_unlock(&thr->queue_lock);
13564                         retval = ctl_scsiio(&io->scsiio);
13565                         if (retval != CTL_RETVAL_COMPLETE)
13566                                 CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13567                         continue;
13568                 }
13569
13570                 /* Sleep until we have something to do. */
13571                 mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
13572         }
13573 }
13574
13575 static void
13576 ctl_lun_thread(void *arg)
13577 {
13578         struct ctl_softc *softc = (struct ctl_softc *)arg;
13579         struct ctl_be_lun *be_lun;
13580         int retval;
13581
13582         CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
13583
13584         for (;;) {
13585                 retval = 0;
13586                 mtx_lock(&softc->ctl_lock);
13587                 be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
13588                 if (be_lun != NULL) {
13589                         STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
13590                         mtx_unlock(&softc->ctl_lock);
13591                         ctl_create_lun(be_lun);
13592                         continue;
13593                 }
13594
13595                 /* Sleep until we have something to do. */
13596                 mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
13597                     PDROP | PRIBIO, "-", 0);
13598         }
13599 }
13600
13601 static void
13602 ctl_thresh_thread(void *arg)
13603 {
13604         struct ctl_softc *softc = (struct ctl_softc *)arg;
13605         struct ctl_lun *lun;
13606         struct ctl_be_lun *be_lun;
13607         struct scsi_da_rw_recovery_page *rwpage;
13608         struct ctl_logical_block_provisioning_page *page;
13609         const char *attr;
13610         union ctl_ha_msg msg;
13611         uint64_t thres, val;
13612         int i, e, set;
13613
13614         CTL_DEBUG_PRINT(("ctl_thresh_thread starting\n"));
13615
13616         for (;;) {
13617                 mtx_lock(&softc->ctl_lock);
13618                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
13619                         be_lun = lun->be_lun;
13620                         if ((lun->flags & CTL_LUN_DISABLED) ||
13621                             (lun->flags & CTL_LUN_OFFLINE) ||
13622                             lun->backend->lun_attr == NULL)
13623                                 continue;
13624                         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
13625                             softc->ha_mode == CTL_HA_MODE_XFER)
13626                                 continue;
13627                         rwpage = &lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT];
13628                         if ((rwpage->byte8 & SMS_RWER_LBPERE) == 0)
13629                                 continue;
13630                         e = 0;
13631                         page = &lun->mode_pages.lbp_page[CTL_PAGE_CURRENT];
13632                         for (i = 0; i < CTL_NUM_LBP_THRESH; i++) {
13633                                 if ((page->descr[i].flags & SLBPPD_ENABLED) == 0)
13634                                         continue;
13635                                 thres = scsi_4btoul(page->descr[i].count);
13636                                 thres <<= CTL_LBP_EXPONENT;
13637                                 switch (page->descr[i].resource) {
13638                                 case 0x01:
13639                                         attr = "blocksavail";
13640                                         break;
13641                                 case 0x02:
13642                                         attr = "blocksused";
13643                                         break;
13644                                 case 0xf1:
13645                                         attr = "poolblocksavail";
13646                                         break;
13647                                 case 0xf2:
13648                                         attr = "poolblocksused";
13649                                         break;
13650                                 default:
13651                                         continue;
13652                                 }
13653                                 mtx_unlock(&softc->ctl_lock); // XXX
13654                                 val = lun->backend->lun_attr(
13655                                     lun->be_lun->be_lun, attr);
13656                                 mtx_lock(&softc->ctl_lock);
13657                                 if (val == UINT64_MAX)
13658                                         continue;
13659                                 if ((page->descr[i].flags & SLBPPD_ARMING_MASK)
13660                                     == SLBPPD_ARMING_INC)
13661                                         e = (val >= thres);
13662                                 else
13663                                         e = (val <= thres);
13664                                 if (e)
13665                                         break;
13666                         }
13667                         mtx_lock(&lun->lun_lock);
13668                         if (e) {
13669                                 scsi_u64to8b((uint8_t *)&page->descr[i] -
13670                                     (uint8_t *)page, lun->ua_tpt_info);
13671                                 if (lun->lasttpt == 0 ||
13672                                     time_uptime - lun->lasttpt >= CTL_LBP_UA_PERIOD) {
13673                                         lun->lasttpt = time_uptime;
13674                                         ctl_est_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13675                                         set = 1;
13676                                 } else
13677                                         set = 0;
13678                         } else {
13679                                 lun->lasttpt = 0;
13680                                 ctl_clr_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13681                                 set = -1;
13682                         }
13683                         mtx_unlock(&lun->lun_lock);
13684                         if (set != 0 &&
13685                             lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
13686                                 /* Send msg to other side. */
13687                                 bzero(&msg.ua, sizeof(msg.ua));
13688                                 msg.hdr.msg_type = CTL_MSG_UA;
13689                                 msg.hdr.nexus.initid = -1;
13690                                 msg.hdr.nexus.targ_port = -1;
13691                                 msg.hdr.nexus.targ_lun = lun->lun;
13692                                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
13693                                 msg.ua.ua_all = 1;
13694                                 msg.ua.ua_set = (set > 0);
13695                                 msg.ua.ua_type = CTL_UA_THIN_PROV_THRES;
13696                                 memcpy(msg.ua.ua_info, lun->ua_tpt_info, 8);
13697                                 mtx_unlock(&softc->ctl_lock); // XXX
13698                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13699                                     sizeof(msg.ua), M_WAITOK);
13700                                 mtx_lock(&softc->ctl_lock);
13701                         }
13702                 }
13703                 mtx_unlock(&softc->ctl_lock);
13704                 pause("-", CTL_LBP_PERIOD * hz);
13705         }
13706 }
13707
13708 static void
13709 ctl_enqueue_incoming(union ctl_io *io)
13710 {
13711         struct ctl_softc *softc = control_softc;
13712         struct ctl_thread *thr;
13713         u_int idx;
13714
13715         idx = (io->io_hdr.nexus.targ_port * 127 +
13716                io->io_hdr.nexus.initid) % worker_threads;
13717         thr = &softc->threads[idx];
13718         mtx_lock(&thr->queue_lock);
13719         STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
13720         mtx_unlock(&thr->queue_lock);
13721         wakeup(thr);
13722 }
13723
13724 static void
13725 ctl_enqueue_rtr(union ctl_io *io)
13726 {
13727         struct ctl_softc *softc = control_softc;
13728         struct ctl_thread *thr;
13729
13730         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13731         mtx_lock(&thr->queue_lock);
13732         STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
13733         mtx_unlock(&thr->queue_lock);
13734         wakeup(thr);
13735 }
13736
13737 static void
13738 ctl_enqueue_done(union ctl_io *io)
13739 {
13740         struct ctl_softc *softc = control_softc;
13741         struct ctl_thread *thr;
13742
13743         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13744         mtx_lock(&thr->queue_lock);
13745         STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
13746         mtx_unlock(&thr->queue_lock);
13747         wakeup(thr);
13748 }
13749
13750 static void
13751 ctl_enqueue_isc(union ctl_io *io)
13752 {
13753         struct ctl_softc *softc = control_softc;
13754         struct ctl_thread *thr;
13755
13756         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13757         mtx_lock(&thr->queue_lock);
13758         STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
13759         mtx_unlock(&thr->queue_lock);
13760         wakeup(thr);
13761 }
13762
13763 /*
13764  *  vim: ts=8
13765  */