]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/blob - sys/cam/ctl/ctl.c
Add support for NUAR bit in Control mode page.
[FreeBSD/FreeBSD.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_cd.h>
73 #include <cam/scsi/scsi_da.h>
74 #include <cam/ctl/ctl_io.h>
75 #include <cam/ctl/ctl.h>
76 #include <cam/ctl/ctl_frontend.h>
77 #include <cam/ctl/ctl_util.h>
78 #include <cam/ctl/ctl_backend.h>
79 #include <cam/ctl/ctl_ioctl.h>
80 #include <cam/ctl/ctl_ha.h>
81 #include <cam/ctl/ctl_private.h>
82 #include <cam/ctl/ctl_debug.h>
83 #include <cam/ctl/ctl_scsi_all.h>
84 #include <cam/ctl/ctl_error.h>
85
86 struct ctl_softc *control_softc = NULL;
87
88 /*
89  * Template mode pages.
90  */
91
92 /*
93  * Note that these are default values only.  The actual values will be
94  * filled in when the user does a mode sense.
95  */
96 const static struct scsi_da_rw_recovery_page rw_er_page_default = {
97         /*page_code*/SMS_RW_ERROR_RECOVERY_PAGE,
98         /*page_length*/sizeof(struct scsi_da_rw_recovery_page) - 2,
99         /*byte3*/SMS_RWER_AWRE|SMS_RWER_ARRE,
100         /*read_retry_count*/0,
101         /*correction_span*/0,
102         /*head_offset_count*/0,
103         /*data_strobe_offset_cnt*/0,
104         /*byte8*/SMS_RWER_LBPERE,
105         /*write_retry_count*/0,
106         /*reserved2*/0,
107         /*recovery_time_limit*/{0, 0},
108 };
109
110 const static struct scsi_da_rw_recovery_page rw_er_page_changeable = {
111         /*page_code*/SMS_RW_ERROR_RECOVERY_PAGE,
112         /*page_length*/sizeof(struct scsi_da_rw_recovery_page) - 2,
113         /*byte3*/SMS_RWER_PER,
114         /*read_retry_count*/0,
115         /*correction_span*/0,
116         /*head_offset_count*/0,
117         /*data_strobe_offset_cnt*/0,
118         /*byte8*/SMS_RWER_LBPERE,
119         /*write_retry_count*/0,
120         /*reserved2*/0,
121         /*recovery_time_limit*/{0, 0},
122 };
123
124 const static struct scsi_format_page format_page_default = {
125         /*page_code*/SMS_FORMAT_DEVICE_PAGE,
126         /*page_length*/sizeof(struct scsi_format_page) - 2,
127         /*tracks_per_zone*/ {0, 0},
128         /*alt_sectors_per_zone*/ {0, 0},
129         /*alt_tracks_per_zone*/ {0, 0},
130         /*alt_tracks_per_lun*/ {0, 0},
131         /*sectors_per_track*/ {(CTL_DEFAULT_SECTORS_PER_TRACK >> 8) & 0xff,
132                                 CTL_DEFAULT_SECTORS_PER_TRACK & 0xff},
133         /*bytes_per_sector*/ {0, 0},
134         /*interleave*/ {0, 0},
135         /*track_skew*/ {0, 0},
136         /*cylinder_skew*/ {0, 0},
137         /*flags*/ SFP_HSEC,
138         /*reserved*/ {0, 0, 0}
139 };
140
141 const static struct scsi_format_page format_page_changeable = {
142         /*page_code*/SMS_FORMAT_DEVICE_PAGE,
143         /*page_length*/sizeof(struct scsi_format_page) - 2,
144         /*tracks_per_zone*/ {0, 0},
145         /*alt_sectors_per_zone*/ {0, 0},
146         /*alt_tracks_per_zone*/ {0, 0},
147         /*alt_tracks_per_lun*/ {0, 0},
148         /*sectors_per_track*/ {0, 0},
149         /*bytes_per_sector*/ {0, 0},
150         /*interleave*/ {0, 0},
151         /*track_skew*/ {0, 0},
152         /*cylinder_skew*/ {0, 0},
153         /*flags*/ 0,
154         /*reserved*/ {0, 0, 0}
155 };
156
157 const static struct scsi_rigid_disk_page rigid_disk_page_default = {
158         /*page_code*/SMS_RIGID_DISK_PAGE,
159         /*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
160         /*cylinders*/ {0, 0, 0},
161         /*heads*/ CTL_DEFAULT_HEADS,
162         /*start_write_precomp*/ {0, 0, 0},
163         /*start_reduced_current*/ {0, 0, 0},
164         /*step_rate*/ {0, 0},
165         /*landing_zone_cylinder*/ {0, 0, 0},
166         /*rpl*/ SRDP_RPL_DISABLED,
167         /*rotational_offset*/ 0,
168         /*reserved1*/ 0,
169         /*rotation_rate*/ {(CTL_DEFAULT_ROTATION_RATE >> 8) & 0xff,
170                            CTL_DEFAULT_ROTATION_RATE & 0xff},
171         /*reserved2*/ {0, 0}
172 };
173
174 const static struct scsi_rigid_disk_page rigid_disk_page_changeable = {
175         /*page_code*/SMS_RIGID_DISK_PAGE,
176         /*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
177         /*cylinders*/ {0, 0, 0},
178         /*heads*/ 0,
179         /*start_write_precomp*/ {0, 0, 0},
180         /*start_reduced_current*/ {0, 0, 0},
181         /*step_rate*/ {0, 0},
182         /*landing_zone_cylinder*/ {0, 0, 0},
183         /*rpl*/ 0,
184         /*rotational_offset*/ 0,
185         /*reserved1*/ 0,
186         /*rotation_rate*/ {0, 0},
187         /*reserved2*/ {0, 0}
188 };
189
190 const static struct scsi_da_verify_recovery_page verify_er_page_default = {
191         /*page_code*/SMS_VERIFY_ERROR_RECOVERY_PAGE,
192         /*page_length*/sizeof(struct scsi_da_verify_recovery_page) - 2,
193         /*byte3*/0,
194         /*read_retry_count*/0,
195         /*reserved*/{ 0, 0, 0, 0, 0, 0 },
196         /*recovery_time_limit*/{0, 0},
197 };
198
199 const static struct scsi_da_verify_recovery_page verify_er_page_changeable = {
200         /*page_code*/SMS_VERIFY_ERROR_RECOVERY_PAGE,
201         /*page_length*/sizeof(struct scsi_da_verify_recovery_page) - 2,
202         /*byte3*/SMS_VER_PER,
203         /*read_retry_count*/0,
204         /*reserved*/{ 0, 0, 0, 0, 0, 0 },
205         /*recovery_time_limit*/{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 | SCP_NUAR,
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_EWASC,
288         /*mrie*/SIEP_MRIE_NO,
289         /*interval_timer*/{0, 0, 0, 0},
290         /*report_count*/{0, 0, 0, 1}
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*/SIEP_FLAGS_EWASC | SIEP_FLAGS_DEXCPT | SIEP_FLAGS_TEST |
297             SIEP_FLAGS_LOGERR,
298         /*mrie*/0x0f,
299         /*interval_timer*/{0xff, 0xff, 0xff, 0xff},
300         /*report_count*/{0xff, 0xff, 0xff, 0xff}
301 };
302
303 #define CTL_LBPM_LEN    (sizeof(struct ctl_logical_block_provisioning_page) - 4)
304
305 const static struct ctl_logical_block_provisioning_page lbp_page_default = {{
306         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF,
307         /*subpage_code*/0x02,
308         /*page_length*/{CTL_LBPM_LEN >> 8, CTL_LBPM_LEN},
309         /*flags*/0,
310         /*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
311         /*descr*/{}},
312         {{/*flags*/0,
313           /*resource*/0x01,
314           /*reserved*/{0, 0},
315           /*count*/{0, 0, 0, 0}},
316          {/*flags*/0,
317           /*resource*/0x02,
318           /*reserved*/{0, 0},
319           /*count*/{0, 0, 0, 0}},
320          {/*flags*/0,
321           /*resource*/0xf1,
322           /*reserved*/{0, 0},
323           /*count*/{0, 0, 0, 0}},
324          {/*flags*/0,
325           /*resource*/0xf2,
326           /*reserved*/{0, 0},
327           /*count*/{0, 0, 0, 0}}
328         }
329 };
330
331 const static struct ctl_logical_block_provisioning_page lbp_page_changeable = {{
332         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF,
333         /*subpage_code*/0x02,
334         /*page_length*/{CTL_LBPM_LEN >> 8, CTL_LBPM_LEN},
335         /*flags*/0,
336         /*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
337         /*descr*/{}},
338         {{/*flags*/0,
339           /*resource*/0,
340           /*reserved*/{0, 0},
341           /*count*/{0, 0, 0, 0}},
342          {/*flags*/0,
343           /*resource*/0,
344           /*reserved*/{0, 0},
345           /*count*/{0, 0, 0, 0}},
346          {/*flags*/0,
347           /*resource*/0,
348           /*reserved*/{0, 0},
349           /*count*/{0, 0, 0, 0}},
350          {/*flags*/0,
351           /*resource*/0,
352           /*reserved*/{0, 0},
353           /*count*/{0, 0, 0, 0}}
354         }
355 };
356
357 const static struct scsi_cddvd_capabilities_page cddvd_page_default = {
358         /*page_code*/SMS_CDDVD_CAPS_PAGE,
359         /*page_length*/sizeof(struct scsi_cddvd_capabilities_page) - 2,
360         /*caps1*/0x3f,
361         /*caps2*/0x00,
362         /*caps3*/0xf0,
363         /*caps4*/0x00,
364         /*caps5*/0x29,
365         /*caps6*/0x00,
366         /*obsolete*/{0, 0},
367         /*nvol_levels*/{0, 0},
368         /*buffer_size*/{8, 0},
369         /*obsolete2*/{0, 0},
370         /*reserved*/0,
371         /*digital*/0,
372         /*obsolete3*/0,
373         /*copy_management*/0,
374         /*reserved2*/0,
375         /*rotation_control*/0,
376         /*cur_write_speed*/0,
377         /*num_speed_descr*/0,
378 };
379
380 const static struct scsi_cddvd_capabilities_page cddvd_page_changeable = {
381         /*page_code*/SMS_CDDVD_CAPS_PAGE,
382         /*page_length*/sizeof(struct scsi_cddvd_capabilities_page) - 2,
383         /*caps1*/0,
384         /*caps2*/0,
385         /*caps3*/0,
386         /*caps4*/0,
387         /*caps5*/0,
388         /*caps6*/0,
389         /*obsolete*/{0, 0},
390         /*nvol_levels*/{0, 0},
391         /*buffer_size*/{0, 0},
392         /*obsolete2*/{0, 0},
393         /*reserved*/0,
394         /*digital*/0,
395         /*obsolete3*/0,
396         /*copy_management*/0,
397         /*reserved2*/0,
398         /*rotation_control*/0,
399         /*cur_write_speed*/0,
400         /*num_speed_descr*/0,
401 };
402
403 SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD, 0, "CAM Target Layer");
404 static int worker_threads = -1;
405 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, worker_threads, CTLFLAG_RDTUN,
406     &worker_threads, 1, "Number of worker threads");
407 static int ctl_debug = CTL_DEBUG_NONE;
408 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, debug, CTLFLAG_RWTUN,
409     &ctl_debug, 0, "Enabled debug flags");
410
411 /*
412  * Supported pages (0x00), Serial number (0x80), Device ID (0x83),
413  * Extended INQUIRY Data (0x86), Mode Page Policy (0x87),
414  * SCSI Ports (0x88), Third-party Copy (0x8F), Block limits (0xB0),
415  * Block Device Characteristics (0xB1) and Logical Block Provisioning (0xB2)
416  */
417 #define SCSI_EVPD_NUM_SUPPORTED_PAGES   10
418
419 static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
420                                   int param);
421 static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
422 static void ctl_copy_sense_data_back(union ctl_io *src, union ctl_ha_msg *dest);
423 static int ctl_init(void);
424 void ctl_shutdown(void);
425 static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
426 static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
427 static int ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio);
428 static void ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
429                               struct ctl_ooa *ooa_hdr,
430                               struct ctl_ooa_entry *kern_entries);
431 static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
432                      struct thread *td);
433 static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
434                          struct ctl_be_lun *be_lun);
435 static int ctl_free_lun(struct ctl_lun *lun);
436 static void ctl_create_lun(struct ctl_be_lun *be_lun);
437 static struct ctl_port * ctl_io_port(struct ctl_io_hdr *io_hdr);
438
439 static int ctl_do_mode_select(union ctl_io *io);
440 static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
441                            uint64_t res_key, uint64_t sa_res_key,
442                            uint8_t type, uint32_t residx,
443                            struct ctl_scsiio *ctsio,
444                            struct scsi_per_res_out *cdb,
445                            struct scsi_per_res_out_parms* param);
446 static void ctl_pro_preempt_other(struct ctl_lun *lun,
447                                   union ctl_ha_msg *msg);
448 static void ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg);
449 static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
450 static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
451 static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
452 static int ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len);
453 static int ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len);
454 static int ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio,
455                                          int alloc_len);
456 static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
457                                          int alloc_len);
458 static int ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len);
459 static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
460 static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
461 static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
462 static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len);
463 static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2,
464     bool seq);
465 static ctl_action ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2);
466 static ctl_action ctl_check_for_blockage(struct ctl_lun *lun,
467     union ctl_io *pending_io, union ctl_io *ooa_io);
468 static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
469                                 union ctl_io *starting_io);
470 static int ctl_check_blocked(struct ctl_lun *lun);
471 static int ctl_scsiio_lun_check(struct ctl_lun *lun,
472                                 const struct ctl_cmd_entry *entry,
473                                 struct ctl_scsiio *ctsio);
474 static void ctl_failover_lun(union ctl_io *io);
475 static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
476                                struct ctl_scsiio *ctsio);
477 static int ctl_scsiio(struct ctl_scsiio *ctsio);
478
479 static int ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
480 static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
481                             ctl_ua_type ua_type);
482 static int ctl_do_lun_reset(struct ctl_lun *lun, union ctl_io *io,
483                          ctl_ua_type ua_type);
484 static int ctl_lun_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
485 static int ctl_abort_task(union ctl_io *io);
486 static int ctl_abort_task_set(union ctl_io *io);
487 static int ctl_query_task(union ctl_io *io, int task_set);
488 static int ctl_i_t_nexus_reset(union ctl_io *io);
489 static int ctl_query_async_event(union ctl_io *io);
490 static void ctl_run_task(union ctl_io *io);
491 #ifdef CTL_IO_DELAY
492 static void ctl_datamove_timer_wakeup(void *arg);
493 static void ctl_done_timer_wakeup(void *arg);
494 #endif /* CTL_IO_DELAY */
495
496 static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
497 static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
498 static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
499 static void ctl_datamove_remote_write(union ctl_io *io);
500 static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
501 static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
502 static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
503 static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
504                                     ctl_ha_dt_cb callback);
505 static void ctl_datamove_remote_read(union ctl_io *io);
506 static void ctl_datamove_remote(union ctl_io *io);
507 static void ctl_process_done(union ctl_io *io);
508 static void ctl_lun_thread(void *arg);
509 static void ctl_thresh_thread(void *arg);
510 static void ctl_work_thread(void *arg);
511 static void ctl_enqueue_incoming(union ctl_io *io);
512 static void ctl_enqueue_rtr(union ctl_io *io);
513 static void ctl_enqueue_done(union ctl_io *io);
514 static void ctl_enqueue_isc(union ctl_io *io);
515 static const struct ctl_cmd_entry *
516     ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa);
517 static const struct ctl_cmd_entry *
518     ctl_validate_command(struct ctl_scsiio *ctsio);
519 static int ctl_cmd_applicable(uint8_t lun_type,
520     const struct ctl_cmd_entry *entry);
521
522 static uint64_t ctl_get_prkey(struct ctl_lun *lun, uint32_t residx);
523 static void ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx);
524 static void ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx);
525 static void ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key);
526
527 /*
528  * Load the serialization table.  This isn't very pretty, but is probably
529  * the easiest way to do it.
530  */
531 #include "ctl_ser_table.c"
532
533 /*
534  * We only need to define open, close and ioctl routines for this driver.
535  */
536 static struct cdevsw ctl_cdevsw = {
537         .d_version =    D_VERSION,
538         .d_flags =      0,
539         .d_open =       ctl_open,
540         .d_close =      ctl_close,
541         .d_ioctl =      ctl_ioctl,
542         .d_name =       "ctl",
543 };
544
545
546 MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
547
548 static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
549
550 static moduledata_t ctl_moduledata = {
551         "ctl",
552         ctl_module_event_handler,
553         NULL
554 };
555
556 DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
557 MODULE_VERSION(ctl, 1);
558
559 static struct ctl_frontend ha_frontend =
560 {
561         .name = "ha",
562 };
563
564 static void
565 ctl_ha_datamove(union ctl_io *io)
566 {
567         struct ctl_lun *lun;
568         struct ctl_sg_entry *sgl;
569         union ctl_ha_msg msg;
570         uint32_t sg_entries_sent;
571         int do_sg_copy, i, j;
572
573         lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
574         memset(&msg.dt, 0, sizeof(msg.dt));
575         msg.hdr.msg_type = CTL_MSG_DATAMOVE;
576         msg.hdr.original_sc = io->io_hdr.original_sc;
577         msg.hdr.serializing_sc = io;
578         msg.hdr.nexus = io->io_hdr.nexus;
579         msg.hdr.status = io->io_hdr.status;
580         msg.dt.flags = io->io_hdr.flags;
581
582         /*
583          * We convert everything into a S/G list here.  We can't
584          * pass by reference, only by value between controllers.
585          * So we can't pass a pointer to the S/G list, only as many
586          * S/G entries as we can fit in here.  If it's possible for
587          * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
588          * then we need to break this up into multiple transfers.
589          */
590         if (io->scsiio.kern_sg_entries == 0) {
591                 msg.dt.kern_sg_entries = 1;
592 #if 0
593                 if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
594                         msg.dt.sg_list[0].addr = io->scsiio.kern_data_ptr;
595                 } else {
596                         /* XXX KDM use busdma here! */
597                         msg.dt.sg_list[0].addr =
598                             (void *)vtophys(io->scsiio.kern_data_ptr);
599                 }
600 #else
601                 KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0,
602                     ("HA does not support BUS_ADDR"));
603                 msg.dt.sg_list[0].addr = io->scsiio.kern_data_ptr;
604 #endif
605                 msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
606                 do_sg_copy = 0;
607         } else {
608                 msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
609                 do_sg_copy = 1;
610         }
611
612         msg.dt.kern_data_len = io->scsiio.kern_data_len;
613         msg.dt.kern_total_len = io->scsiio.kern_total_len;
614         msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
615         msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
616         msg.dt.sg_sequence = 0;
617
618         /*
619          * Loop until we've sent all of the S/G entries.  On the
620          * other end, we'll recompose these S/G entries into one
621          * contiguous list before processing.
622          */
623         for (sg_entries_sent = 0; sg_entries_sent < msg.dt.kern_sg_entries;
624             msg.dt.sg_sequence++) {
625                 msg.dt.cur_sg_entries = MIN((sizeof(msg.dt.sg_list) /
626                     sizeof(msg.dt.sg_list[0])),
627                     msg.dt.kern_sg_entries - sg_entries_sent);
628                 if (do_sg_copy != 0) {
629                         sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
630                         for (i = sg_entries_sent, j = 0;
631                              i < msg.dt.cur_sg_entries; i++, j++) {
632 #if 0
633                                 if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
634                                         msg.dt.sg_list[j].addr = sgl[i].addr;
635                                 } else {
636                                         /* XXX KDM use busdma here! */
637                                         msg.dt.sg_list[j].addr =
638                                             (void *)vtophys(sgl[i].addr);
639                                 }
640 #else
641                                 KASSERT((io->io_hdr.flags &
642                                     CTL_FLAG_BUS_ADDR) == 0,
643                                     ("HA does not support BUS_ADDR"));
644                                 msg.dt.sg_list[j].addr = sgl[i].addr;
645 #endif
646                                 msg.dt.sg_list[j].len = sgl[i].len;
647                         }
648                 }
649
650                 sg_entries_sent += msg.dt.cur_sg_entries;
651                 msg.dt.sg_last = (sg_entries_sent >= msg.dt.kern_sg_entries);
652                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
653                     sizeof(msg.dt) - sizeof(msg.dt.sg_list) +
654                     sizeof(struct ctl_sg_entry) * msg.dt.cur_sg_entries,
655                     M_WAITOK) > CTL_HA_STATUS_SUCCESS) {
656                         io->io_hdr.port_status = 31341;
657                         io->scsiio.be_move_done(io);
658                         return;
659                 }
660                 msg.dt.sent_sg_entries = sg_entries_sent;
661         }
662
663         /*
664          * Officially handover the request from us to peer.
665          * If failover has just happened, then we must return error.
666          * If failover happen just after, then it is not our problem.
667          */
668         if (lun)
669                 mtx_lock(&lun->lun_lock);
670         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
671                 if (lun)
672                         mtx_unlock(&lun->lun_lock);
673                 io->io_hdr.port_status = 31342;
674                 io->scsiio.be_move_done(io);
675                 return;
676         }
677         io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
678         io->io_hdr.flags |= CTL_FLAG_DMA_INPROG;
679         if (lun)
680                 mtx_unlock(&lun->lun_lock);
681 }
682
683 static void
684 ctl_ha_done(union ctl_io *io)
685 {
686         union ctl_ha_msg msg;
687
688         if (io->io_hdr.io_type == CTL_IO_SCSI) {
689                 memset(&msg, 0, sizeof(msg));
690                 msg.hdr.msg_type = CTL_MSG_FINISH_IO;
691                 msg.hdr.original_sc = io->io_hdr.original_sc;
692                 msg.hdr.nexus = io->io_hdr.nexus;
693                 msg.hdr.status = io->io_hdr.status;
694                 msg.scsi.scsi_status = io->scsiio.scsi_status;
695                 msg.scsi.tag_num = io->scsiio.tag_num;
696                 msg.scsi.tag_type = io->scsiio.tag_type;
697                 msg.scsi.sense_len = io->scsiio.sense_len;
698                 msg.scsi.sense_residual = io->scsiio.sense_residual;
699                 msg.scsi.residual = io->scsiio.residual;
700                 memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
701                     io->scsiio.sense_len);
702                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
703                     sizeof(msg.scsi) - sizeof(msg.scsi.sense_data) +
704                     msg.scsi.sense_len, M_WAITOK);
705         }
706         ctl_free_io(io);
707 }
708
709 static void
710 ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
711                             union ctl_ha_msg *msg_info)
712 {
713         struct ctl_scsiio *ctsio;
714
715         if (msg_info->hdr.original_sc == NULL) {
716                 printf("%s: original_sc == NULL!\n", __func__);
717                 /* XXX KDM now what? */
718                 return;
719         }
720
721         ctsio = &msg_info->hdr.original_sc->scsiio;
722         ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
723         ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
724         ctsio->io_hdr.status = msg_info->hdr.status;
725         ctsio->scsi_status = msg_info->scsi.scsi_status;
726         ctsio->sense_len = msg_info->scsi.sense_len;
727         ctsio->sense_residual = msg_info->scsi.sense_residual;
728         ctsio->residual = msg_info->scsi.residual;
729         memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
730                msg_info->scsi.sense_len);
731         ctl_enqueue_isc((union ctl_io *)ctsio);
732 }
733
734 static void
735 ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
736                                 union ctl_ha_msg *msg_info)
737 {
738         struct ctl_scsiio *ctsio;
739
740         if (msg_info->hdr.serializing_sc == NULL) {
741                 printf("%s: serializing_sc == NULL!\n", __func__);
742                 /* XXX KDM now what? */
743                 return;
744         }
745
746         ctsio = &msg_info->hdr.serializing_sc->scsiio;
747         ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
748         ctl_enqueue_isc((union ctl_io *)ctsio);
749 }
750
751 void
752 ctl_isc_announce_lun(struct ctl_lun *lun)
753 {
754         struct ctl_softc *softc = lun->ctl_softc;
755         union ctl_ha_msg *msg;
756         struct ctl_ha_msg_lun_pr_key pr_key;
757         int i, k;
758
759         if (softc->ha_link != CTL_HA_LINK_ONLINE)
760                 return;
761         mtx_lock(&lun->lun_lock);
762         i = sizeof(msg->lun);
763         if (lun->lun_devid)
764                 i += lun->lun_devid->len;
765         i += sizeof(pr_key) * lun->pr_key_count;
766 alloc:
767         mtx_unlock(&lun->lun_lock);
768         msg = malloc(i, M_CTL, M_WAITOK);
769         mtx_lock(&lun->lun_lock);
770         k = sizeof(msg->lun);
771         if (lun->lun_devid)
772                 k += lun->lun_devid->len;
773         k += sizeof(pr_key) * lun->pr_key_count;
774         if (i < k) {
775                 free(msg, M_CTL);
776                 i = k;
777                 goto alloc;
778         }
779         bzero(&msg->lun, sizeof(msg->lun));
780         msg->hdr.msg_type = CTL_MSG_LUN_SYNC;
781         msg->hdr.nexus.targ_lun = lun->lun;
782         msg->hdr.nexus.targ_mapped_lun = lun->lun;
783         msg->lun.flags = lun->flags;
784         msg->lun.pr_generation = lun->pr_generation;
785         msg->lun.pr_res_idx = lun->pr_res_idx;
786         msg->lun.pr_res_type = lun->pr_res_type;
787         msg->lun.pr_key_count = lun->pr_key_count;
788         i = 0;
789         if (lun->lun_devid) {
790                 msg->lun.lun_devid_len = lun->lun_devid->len;
791                 memcpy(&msg->lun.data[i], lun->lun_devid->data,
792                     msg->lun.lun_devid_len);
793                 i += msg->lun.lun_devid_len;
794         }
795         for (k = 0; k < CTL_MAX_INITIATORS; k++) {
796                 if ((pr_key.pr_key = ctl_get_prkey(lun, k)) == 0)
797                         continue;
798                 pr_key.pr_iid = k;
799                 memcpy(&msg->lun.data[i], &pr_key, sizeof(pr_key));
800                 i += sizeof(pr_key);
801         }
802         mtx_unlock(&lun->lun_lock);
803         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->port, sizeof(msg->port) + i,
804             M_WAITOK);
805         free(msg, M_CTL);
806
807         if (lun->flags & CTL_LUN_PRIMARY_SC) {
808                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
809                         ctl_isc_announce_mode(lun, -1,
810                             lun->mode_pages.index[i].page_code & SMPH_PC_MASK,
811                             lun->mode_pages.index[i].subpage);
812                 }
813         }
814 }
815
816 void
817 ctl_isc_announce_port(struct ctl_port *port)
818 {
819         struct ctl_softc *softc = port->ctl_softc;
820         union ctl_ha_msg *msg;
821         int i;
822
823         if (port->targ_port < softc->port_min ||
824             port->targ_port >= softc->port_max ||
825             softc->ha_link != CTL_HA_LINK_ONLINE)
826                 return;
827         i = sizeof(msg->port) + strlen(port->port_name) + 1;
828         if (port->lun_map)
829                 i += sizeof(uint32_t) * CTL_MAX_LUNS;
830         if (port->port_devid)
831                 i += port->port_devid->len;
832         if (port->target_devid)
833                 i += port->target_devid->len;
834         if (port->init_devid)
835                 i += port->init_devid->len;
836         msg = malloc(i, M_CTL, M_WAITOK);
837         bzero(&msg->port, sizeof(msg->port));
838         msg->hdr.msg_type = CTL_MSG_PORT_SYNC;
839         msg->hdr.nexus.targ_port = port->targ_port;
840         msg->port.port_type = port->port_type;
841         msg->port.physical_port = port->physical_port;
842         msg->port.virtual_port = port->virtual_port;
843         msg->port.status = port->status;
844         i = 0;
845         msg->port.name_len = sprintf(&msg->port.data[i],
846             "%d:%s", softc->ha_id, port->port_name) + 1;
847         i += msg->port.name_len;
848         if (port->lun_map) {
849                 msg->port.lun_map_len = sizeof(uint32_t) * CTL_MAX_LUNS;
850                 memcpy(&msg->port.data[i], port->lun_map,
851                     msg->port.lun_map_len);
852                 i += msg->port.lun_map_len;
853         }
854         if (port->port_devid) {
855                 msg->port.port_devid_len = port->port_devid->len;
856                 memcpy(&msg->port.data[i], port->port_devid->data,
857                     msg->port.port_devid_len);
858                 i += msg->port.port_devid_len;
859         }
860         if (port->target_devid) {
861                 msg->port.target_devid_len = port->target_devid->len;
862                 memcpy(&msg->port.data[i], port->target_devid->data,
863                     msg->port.target_devid_len);
864                 i += msg->port.target_devid_len;
865         }
866         if (port->init_devid) {
867                 msg->port.init_devid_len = port->init_devid->len;
868                 memcpy(&msg->port.data[i], port->init_devid->data,
869                     msg->port.init_devid_len);
870                 i += msg->port.init_devid_len;
871         }
872         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->port, sizeof(msg->port) + i,
873             M_WAITOK);
874         free(msg, M_CTL);
875 }
876
877 void
878 ctl_isc_announce_iid(struct ctl_port *port, int iid)
879 {
880         struct ctl_softc *softc = port->ctl_softc;
881         union ctl_ha_msg *msg;
882         int i, l;
883
884         if (port->targ_port < softc->port_min ||
885             port->targ_port >= softc->port_max ||
886             softc->ha_link != CTL_HA_LINK_ONLINE)
887                 return;
888         mtx_lock(&softc->ctl_lock);
889         i = sizeof(msg->iid);
890         l = 0;
891         if (port->wwpn_iid[iid].name)
892                 l = strlen(port->wwpn_iid[iid].name) + 1;
893         i += l;
894         msg = malloc(i, M_CTL, M_NOWAIT);
895         if (msg == NULL) {
896                 mtx_unlock(&softc->ctl_lock);
897                 return;
898         }
899         bzero(&msg->iid, sizeof(msg->iid));
900         msg->hdr.msg_type = CTL_MSG_IID_SYNC;
901         msg->hdr.nexus.targ_port = port->targ_port;
902         msg->hdr.nexus.initid = iid;
903         msg->iid.in_use = port->wwpn_iid[iid].in_use;
904         msg->iid.name_len = l;
905         msg->iid.wwpn = port->wwpn_iid[iid].wwpn;
906         if (port->wwpn_iid[iid].name)
907                 strlcpy(msg->iid.data, port->wwpn_iid[iid].name, l);
908         mtx_unlock(&softc->ctl_lock);
909         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->iid, i, M_NOWAIT);
910         free(msg, M_CTL);
911 }
912
913 void
914 ctl_isc_announce_mode(struct ctl_lun *lun, uint32_t initidx,
915     uint8_t page, uint8_t subpage)
916 {
917         struct ctl_softc *softc = lun->ctl_softc;
918         union ctl_ha_msg msg;
919         u_int i;
920
921         if (softc->ha_link != CTL_HA_LINK_ONLINE)
922                 return;
923         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
924                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) ==
925                     page && lun->mode_pages.index[i].subpage == subpage)
926                         break;
927         }
928         if (i == CTL_NUM_MODE_PAGES)
929                 return;
930
931         /* Don't try to replicate pages not present on this device. */
932         if (lun->mode_pages.index[i].page_data == NULL)
933                 return;
934
935         bzero(&msg.mode, sizeof(msg.mode));
936         msg.hdr.msg_type = CTL_MSG_MODE_SYNC;
937         msg.hdr.nexus.targ_port = initidx / CTL_MAX_INIT_PER_PORT;
938         msg.hdr.nexus.initid = initidx % CTL_MAX_INIT_PER_PORT;
939         msg.hdr.nexus.targ_lun = lun->lun;
940         msg.hdr.nexus.targ_mapped_lun = lun->lun;
941         msg.mode.page_code = page;
942         msg.mode.subpage = subpage;
943         msg.mode.page_len = lun->mode_pages.index[i].page_len;
944         memcpy(msg.mode.data, lun->mode_pages.index[i].page_data,
945             msg.mode.page_len);
946         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg.mode, sizeof(msg.mode),
947             M_WAITOK);
948 }
949
950 static void
951 ctl_isc_ha_link_up(struct ctl_softc *softc)
952 {
953         struct ctl_port *port;
954         struct ctl_lun *lun;
955         union ctl_ha_msg msg;
956         int i;
957
958         /* Announce this node parameters to peer for validation. */
959         msg.login.msg_type = CTL_MSG_LOGIN;
960         msg.login.version = CTL_HA_VERSION;
961         msg.login.ha_mode = softc->ha_mode;
962         msg.login.ha_id = softc->ha_id;
963         msg.login.max_luns = CTL_MAX_LUNS;
964         msg.login.max_ports = CTL_MAX_PORTS;
965         msg.login.max_init_per_port = CTL_MAX_INIT_PER_PORT;
966         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg.login, sizeof(msg.login),
967             M_WAITOK);
968
969         STAILQ_FOREACH(port, &softc->port_list, links) {
970                 ctl_isc_announce_port(port);
971                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
972                         if (port->wwpn_iid[i].in_use)
973                                 ctl_isc_announce_iid(port, i);
974                 }
975         }
976         STAILQ_FOREACH(lun, &softc->lun_list, links)
977                 ctl_isc_announce_lun(lun);
978 }
979
980 static void
981 ctl_isc_ha_link_down(struct ctl_softc *softc)
982 {
983         struct ctl_port *port;
984         struct ctl_lun *lun;
985         union ctl_io *io;
986         int i;
987
988         mtx_lock(&softc->ctl_lock);
989         STAILQ_FOREACH(lun, &softc->lun_list, links) {
990                 mtx_lock(&lun->lun_lock);
991                 if (lun->flags & CTL_LUN_PEER_SC_PRIMARY) {
992                         lun->flags &= ~CTL_LUN_PEER_SC_PRIMARY;
993                         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
994                 }
995                 mtx_unlock(&lun->lun_lock);
996
997                 mtx_unlock(&softc->ctl_lock);
998                 io = ctl_alloc_io(softc->othersc_pool);
999                 mtx_lock(&softc->ctl_lock);
1000                 ctl_zero_io(io);
1001                 io->io_hdr.msg_type = CTL_MSG_FAILOVER;
1002                 io->io_hdr.nexus.targ_mapped_lun = lun->lun;
1003                 ctl_enqueue_isc(io);
1004         }
1005
1006         STAILQ_FOREACH(port, &softc->port_list, links) {
1007                 if (port->targ_port >= softc->port_min &&
1008                     port->targ_port < softc->port_max)
1009                         continue;
1010                 port->status &= ~CTL_PORT_STATUS_ONLINE;
1011                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1012                         port->wwpn_iid[i].in_use = 0;
1013                         free(port->wwpn_iid[i].name, M_CTL);
1014                         port->wwpn_iid[i].name = NULL;
1015                 }
1016         }
1017         mtx_unlock(&softc->ctl_lock);
1018 }
1019
1020 static void
1021 ctl_isc_ua(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1022 {
1023         struct ctl_lun *lun;
1024         uint32_t iid = ctl_get_initindex(&msg->hdr.nexus);
1025
1026         mtx_lock(&softc->ctl_lock);
1027         if (msg->hdr.nexus.targ_lun < CTL_MAX_LUNS &&
1028             (lun = softc->ctl_luns[msg->hdr.nexus.targ_mapped_lun]) != NULL) {
1029                 mtx_lock(&lun->lun_lock);
1030                 mtx_unlock(&softc->ctl_lock);
1031                 if (msg->ua.ua_type == CTL_UA_THIN_PROV_THRES &&
1032                     msg->ua.ua_set)
1033                         memcpy(lun->ua_tpt_info, msg->ua.ua_info, 8);
1034                 if (msg->ua.ua_all) {
1035                         if (msg->ua.ua_set)
1036                                 ctl_est_ua_all(lun, iid, msg->ua.ua_type);
1037                         else
1038                                 ctl_clr_ua_all(lun, iid, msg->ua.ua_type);
1039                 } else {
1040                         if (msg->ua.ua_set)
1041                                 ctl_est_ua(lun, iid, msg->ua.ua_type);
1042                         else
1043                                 ctl_clr_ua(lun, iid, msg->ua.ua_type);
1044                 }
1045                 mtx_unlock(&lun->lun_lock);
1046         } else
1047                 mtx_unlock(&softc->ctl_lock);
1048 }
1049
1050 static void
1051 ctl_isc_lun_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1052 {
1053         struct ctl_lun *lun;
1054         struct ctl_ha_msg_lun_pr_key pr_key;
1055         int i, k;
1056         ctl_lun_flags oflags;
1057         uint32_t targ_lun;
1058
1059         targ_lun = msg->hdr.nexus.targ_mapped_lun;
1060         mtx_lock(&softc->ctl_lock);
1061         if ((targ_lun >= CTL_MAX_LUNS) ||
1062             ((lun = softc->ctl_luns[targ_lun]) == NULL)) {
1063                 mtx_unlock(&softc->ctl_lock);
1064                 return;
1065         }
1066         mtx_lock(&lun->lun_lock);
1067         mtx_unlock(&softc->ctl_lock);
1068         if (lun->flags & CTL_LUN_DISABLED) {
1069                 mtx_unlock(&lun->lun_lock);
1070                 return;
1071         }
1072         i = (lun->lun_devid != NULL) ? lun->lun_devid->len : 0;
1073         if (msg->lun.lun_devid_len != i || (i > 0 &&
1074             memcmp(&msg->lun.data[0], lun->lun_devid->data, i) != 0)) {
1075                 mtx_unlock(&lun->lun_lock);
1076                 printf("%s: Received conflicting HA LUN %d\n",
1077                     __func__, msg->hdr.nexus.targ_lun);
1078                 return;
1079         } else {
1080                 /* Record whether peer is primary. */
1081                 oflags = lun->flags;
1082                 if ((msg->lun.flags & CTL_LUN_PRIMARY_SC) &&
1083                     (msg->lun.flags & CTL_LUN_DISABLED) == 0)
1084                         lun->flags |= CTL_LUN_PEER_SC_PRIMARY;
1085                 else
1086                         lun->flags &= ~CTL_LUN_PEER_SC_PRIMARY;
1087                 if (oflags != lun->flags)
1088                         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
1089
1090                 /* If peer is primary and we are not -- use data */
1091                 if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
1092                     (lun->flags & CTL_LUN_PEER_SC_PRIMARY)) {
1093                         lun->pr_generation = msg->lun.pr_generation;
1094                         lun->pr_res_idx = msg->lun.pr_res_idx;
1095                         lun->pr_res_type = msg->lun.pr_res_type;
1096                         lun->pr_key_count = msg->lun.pr_key_count;
1097                         for (k = 0; k < CTL_MAX_INITIATORS; k++)
1098                                 ctl_clr_prkey(lun, k);
1099                         for (k = 0; k < msg->lun.pr_key_count; k++) {
1100                                 memcpy(&pr_key, &msg->lun.data[i],
1101                                     sizeof(pr_key));
1102                                 ctl_alloc_prkey(lun, pr_key.pr_iid);
1103                                 ctl_set_prkey(lun, pr_key.pr_iid,
1104                                     pr_key.pr_key);
1105                                 i += sizeof(pr_key);
1106                         }
1107                 }
1108
1109                 mtx_unlock(&lun->lun_lock);
1110                 CTL_DEBUG_PRINT(("%s: Known LUN %d, peer is %s\n",
1111                     __func__, msg->hdr.nexus.targ_lun,
1112                     (msg->lun.flags & CTL_LUN_PRIMARY_SC) ?
1113                     "primary" : "secondary"));
1114
1115                 /* If we are primary but peer doesn't know -- notify */
1116                 if ((lun->flags & CTL_LUN_PRIMARY_SC) &&
1117                     (msg->lun.flags & CTL_LUN_PEER_SC_PRIMARY) == 0)
1118                         ctl_isc_announce_lun(lun);
1119         }
1120 }
1121
1122 static void
1123 ctl_isc_port_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1124 {
1125         struct ctl_port *port;
1126         struct ctl_lun *lun;
1127         int i, new;
1128
1129         port = softc->ctl_ports[msg->hdr.nexus.targ_port];
1130         if (port == NULL) {
1131                 CTL_DEBUG_PRINT(("%s: New port %d\n", __func__,
1132                     msg->hdr.nexus.targ_port));
1133                 new = 1;
1134                 port = malloc(sizeof(*port), M_CTL, M_WAITOK | M_ZERO);
1135                 port->frontend = &ha_frontend;
1136                 port->targ_port = msg->hdr.nexus.targ_port;
1137                 port->fe_datamove = ctl_ha_datamove;
1138                 port->fe_done = ctl_ha_done;
1139         } else if (port->frontend == &ha_frontend) {
1140                 CTL_DEBUG_PRINT(("%s: Updated port %d\n", __func__,
1141                     msg->hdr.nexus.targ_port));
1142                 new = 0;
1143         } else {
1144                 printf("%s: Received conflicting HA port %d\n",
1145                     __func__, msg->hdr.nexus.targ_port);
1146                 return;
1147         }
1148         port->port_type = msg->port.port_type;
1149         port->physical_port = msg->port.physical_port;
1150         port->virtual_port = msg->port.virtual_port;
1151         port->status = msg->port.status;
1152         i = 0;
1153         free(port->port_name, M_CTL);
1154         port->port_name = strndup(&msg->port.data[i], msg->port.name_len,
1155             M_CTL);
1156         i += msg->port.name_len;
1157         if (msg->port.lun_map_len != 0) {
1158                 if (port->lun_map == NULL)
1159                         port->lun_map = malloc(sizeof(uint32_t) * CTL_MAX_LUNS,
1160                             M_CTL, M_WAITOK);
1161                 memcpy(port->lun_map, &msg->port.data[i],
1162                     sizeof(uint32_t) * CTL_MAX_LUNS);
1163                 i += msg->port.lun_map_len;
1164         } else {
1165                 free(port->lun_map, M_CTL);
1166                 port->lun_map = NULL;
1167         }
1168         if (msg->port.port_devid_len != 0) {
1169                 if (port->port_devid == NULL ||
1170                     port->port_devid->len != msg->port.port_devid_len) {
1171                         free(port->port_devid, M_CTL);
1172                         port->port_devid = malloc(sizeof(struct ctl_devid) +
1173                             msg->port.port_devid_len, M_CTL, M_WAITOK);
1174                 }
1175                 memcpy(port->port_devid->data, &msg->port.data[i],
1176                     msg->port.port_devid_len);
1177                 port->port_devid->len = msg->port.port_devid_len;
1178                 i += msg->port.port_devid_len;
1179         } else {
1180                 free(port->port_devid, M_CTL);
1181                 port->port_devid = NULL;
1182         }
1183         if (msg->port.target_devid_len != 0) {
1184                 if (port->target_devid == NULL ||
1185                     port->target_devid->len != msg->port.target_devid_len) {
1186                         free(port->target_devid, M_CTL);
1187                         port->target_devid = malloc(sizeof(struct ctl_devid) +
1188                             msg->port.target_devid_len, M_CTL, M_WAITOK);
1189                 }
1190                 memcpy(port->target_devid->data, &msg->port.data[i],
1191                     msg->port.target_devid_len);
1192                 port->target_devid->len = msg->port.target_devid_len;
1193                 i += msg->port.target_devid_len;
1194         } else {
1195                 free(port->target_devid, M_CTL);
1196                 port->target_devid = NULL;
1197         }
1198         if (msg->port.init_devid_len != 0) {
1199                 if (port->init_devid == NULL ||
1200                     port->init_devid->len != msg->port.init_devid_len) {
1201                         free(port->init_devid, M_CTL);
1202                         port->init_devid = malloc(sizeof(struct ctl_devid) +
1203                             msg->port.init_devid_len, M_CTL, M_WAITOK);
1204                 }
1205                 memcpy(port->init_devid->data, &msg->port.data[i],
1206                     msg->port.init_devid_len);
1207                 port->init_devid->len = msg->port.init_devid_len;
1208                 i += msg->port.init_devid_len;
1209         } else {
1210                 free(port->init_devid, M_CTL);
1211                 port->init_devid = NULL;
1212         }
1213         if (new) {
1214                 if (ctl_port_register(port) != 0) {
1215                         printf("%s: ctl_port_register() failed with error\n",
1216                             __func__);
1217                 }
1218         }
1219         mtx_lock(&softc->ctl_lock);
1220         STAILQ_FOREACH(lun, &softc->lun_list, links) {
1221                 if (ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
1222                         continue;
1223                 mtx_lock(&lun->lun_lock);
1224                 ctl_est_ua_all(lun, -1, CTL_UA_INQ_CHANGE);
1225                 mtx_unlock(&lun->lun_lock);
1226         }
1227         mtx_unlock(&softc->ctl_lock);
1228 }
1229
1230 static void
1231 ctl_isc_iid_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1232 {
1233         struct ctl_port *port;
1234         int iid;
1235
1236         port = softc->ctl_ports[msg->hdr.nexus.targ_port];
1237         if (port == NULL) {
1238                 printf("%s: Received IID for unknown port %d\n",
1239                     __func__, msg->hdr.nexus.targ_port);
1240                 return;
1241         }
1242         iid = msg->hdr.nexus.initid;
1243         port->wwpn_iid[iid].in_use = msg->iid.in_use;
1244         port->wwpn_iid[iid].wwpn = msg->iid.wwpn;
1245         free(port->wwpn_iid[iid].name, M_CTL);
1246         if (msg->iid.name_len) {
1247                 port->wwpn_iid[iid].name = strndup(&msg->iid.data[0],
1248                     msg->iid.name_len, M_CTL);
1249         } else
1250                 port->wwpn_iid[iid].name = NULL;
1251 }
1252
1253 static void
1254 ctl_isc_login(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1255 {
1256
1257         if (msg->login.version != CTL_HA_VERSION) {
1258                 printf("CTL HA peers have different versions %d != %d\n",
1259                     msg->login.version, CTL_HA_VERSION);
1260                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1261                 return;
1262         }
1263         if (msg->login.ha_mode != softc->ha_mode) {
1264                 printf("CTL HA peers have different ha_mode %d != %d\n",
1265                     msg->login.ha_mode, softc->ha_mode);
1266                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1267                 return;
1268         }
1269         if (msg->login.ha_id == softc->ha_id) {
1270                 printf("CTL HA peers have same ha_id %d\n", msg->login.ha_id);
1271                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1272                 return;
1273         }
1274         if (msg->login.max_luns != CTL_MAX_LUNS ||
1275             msg->login.max_ports != CTL_MAX_PORTS ||
1276             msg->login.max_init_per_port != CTL_MAX_INIT_PER_PORT) {
1277                 printf("CTL HA peers have different limits\n");
1278                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1279                 return;
1280         }
1281 }
1282
1283 static void
1284 ctl_isc_mode_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1285 {
1286         struct ctl_lun *lun;
1287         u_int i;
1288         uint32_t initidx, targ_lun;
1289
1290         targ_lun = msg->hdr.nexus.targ_mapped_lun;
1291         mtx_lock(&softc->ctl_lock);
1292         if ((targ_lun >= CTL_MAX_LUNS) ||
1293             ((lun = softc->ctl_luns[targ_lun]) == NULL)) {
1294                 mtx_unlock(&softc->ctl_lock);
1295                 return;
1296         }
1297         mtx_lock(&lun->lun_lock);
1298         mtx_unlock(&softc->ctl_lock);
1299         if (lun->flags & CTL_LUN_DISABLED) {
1300                 mtx_unlock(&lun->lun_lock);
1301                 return;
1302         }
1303         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
1304                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) ==
1305                     msg->mode.page_code &&
1306                     lun->mode_pages.index[i].subpage == msg->mode.subpage)
1307                         break;
1308         }
1309         if (i == CTL_NUM_MODE_PAGES) {
1310                 mtx_unlock(&lun->lun_lock);
1311                 return;
1312         }
1313         memcpy(lun->mode_pages.index[i].page_data, msg->mode.data,
1314             lun->mode_pages.index[i].page_len);
1315         initidx = ctl_get_initindex(&msg->hdr.nexus);
1316         if (initidx != -1)
1317                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
1318         mtx_unlock(&lun->lun_lock);
1319 }
1320
1321 /*
1322  * ISC (Inter Shelf Communication) event handler.  Events from the HA
1323  * subsystem come in here.
1324  */
1325 static void
1326 ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
1327 {
1328         struct ctl_softc *softc = control_softc;
1329         union ctl_io *io;
1330         struct ctl_prio *presio;
1331         ctl_ha_status isc_status;
1332
1333         CTL_DEBUG_PRINT(("CTL: Isc Msg event %d\n", event));
1334         if (event == CTL_HA_EVT_MSG_RECV) {
1335                 union ctl_ha_msg *msg, msgbuf;
1336
1337                 if (param > sizeof(msgbuf))
1338                         msg = malloc(param, M_CTL, M_WAITOK);
1339                 else
1340                         msg = &msgbuf;
1341                 isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, msg, param,
1342                     M_WAITOK);
1343                 if (isc_status != CTL_HA_STATUS_SUCCESS) {
1344                         printf("%s: Error receiving message: %d\n",
1345                             __func__, isc_status);
1346                         if (msg != &msgbuf)
1347                                 free(msg, M_CTL);
1348                         return;
1349                 }
1350
1351                 CTL_DEBUG_PRINT(("CTL: msg_type %d\n", msg->msg_type));
1352                 switch (msg->hdr.msg_type) {
1353                 case CTL_MSG_SERIALIZE:
1354                         io = ctl_alloc_io(softc->othersc_pool);
1355                         ctl_zero_io(io);
1356                         // populate ctsio from msg
1357                         io->io_hdr.io_type = CTL_IO_SCSI;
1358                         io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
1359                         io->io_hdr.original_sc = msg->hdr.original_sc;
1360                         io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
1361                                             CTL_FLAG_IO_ACTIVE;
1362                         /*
1363                          * If we're in serialization-only mode, we don't
1364                          * want to go through full done processing.  Thus
1365                          * the COPY flag.
1366                          *
1367                          * XXX KDM add another flag that is more specific.
1368                          */
1369                         if (softc->ha_mode != CTL_HA_MODE_XFER)
1370                                 io->io_hdr.flags |= CTL_FLAG_INT_COPY;
1371                         io->io_hdr.nexus = msg->hdr.nexus;
1372 #if 0
1373                         printf("port %u, iid %u, lun %u\n",
1374                                io->io_hdr.nexus.targ_port,
1375                                io->io_hdr.nexus.initid,
1376                                io->io_hdr.nexus.targ_lun);
1377 #endif
1378                         io->scsiio.tag_num = msg->scsi.tag_num;
1379                         io->scsiio.tag_type = msg->scsi.tag_type;
1380 #ifdef CTL_TIME_IO
1381                         io->io_hdr.start_time = time_uptime;
1382                         getbinuptime(&io->io_hdr.start_bt);
1383 #endif /* CTL_TIME_IO */
1384                         io->scsiio.cdb_len = msg->scsi.cdb_len;
1385                         memcpy(io->scsiio.cdb, msg->scsi.cdb,
1386                                CTL_MAX_CDBLEN);
1387                         if (softc->ha_mode == CTL_HA_MODE_XFER) {
1388                                 const struct ctl_cmd_entry *entry;
1389
1390                                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
1391                                 io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
1392                                 io->io_hdr.flags |=
1393                                         entry->flags & CTL_FLAG_DATA_MASK;
1394                         }
1395                         ctl_enqueue_isc(io);
1396                         break;
1397
1398                 /* Performed on the Originating SC, XFER mode only */
1399                 case CTL_MSG_DATAMOVE: {
1400                         struct ctl_sg_entry *sgl;
1401                         int i, j;
1402
1403                         io = msg->hdr.original_sc;
1404                         if (io == NULL) {
1405                                 printf("%s: original_sc == NULL!\n", __func__);
1406                                 /* XXX KDM do something here */
1407                                 break;
1408                         }
1409                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
1410                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1411                         /*
1412                          * Keep track of this, we need to send it back over
1413                          * when the datamove is complete.
1414                          */
1415                         io->io_hdr.serializing_sc = msg->hdr.serializing_sc;
1416                         if (msg->hdr.status == CTL_SUCCESS)
1417                                 io->io_hdr.status = msg->hdr.status;
1418
1419                         if (msg->dt.sg_sequence == 0) {
1420 #ifdef CTL_TIME_IO
1421                                 getbinuptime(&io->io_hdr.dma_start_bt);
1422 #endif
1423                                 i = msg->dt.kern_sg_entries +
1424                                     msg->dt.kern_data_len /
1425                                     CTL_HA_DATAMOVE_SEGMENT + 1;
1426                                 sgl = malloc(sizeof(*sgl) * i, M_CTL,
1427                                     M_WAITOK | M_ZERO);
1428                                 io->io_hdr.remote_sglist = sgl;
1429                                 io->io_hdr.local_sglist =
1430                                     &sgl[msg->dt.kern_sg_entries];
1431
1432                                 io->scsiio.kern_data_ptr = (uint8_t *)sgl;
1433
1434                                 io->scsiio.kern_sg_entries =
1435                                         msg->dt.kern_sg_entries;
1436                                 io->scsiio.rem_sg_entries =
1437                                         msg->dt.kern_sg_entries;
1438                                 io->scsiio.kern_data_len =
1439                                         msg->dt.kern_data_len;
1440                                 io->scsiio.kern_total_len =
1441                                         msg->dt.kern_total_len;
1442                                 io->scsiio.kern_data_resid =
1443                                         msg->dt.kern_data_resid;
1444                                 io->scsiio.kern_rel_offset =
1445                                         msg->dt.kern_rel_offset;
1446                                 io->io_hdr.flags &= ~CTL_FLAG_BUS_ADDR;
1447                                 io->io_hdr.flags |= msg->dt.flags &
1448                                     CTL_FLAG_BUS_ADDR;
1449                         } else
1450                                 sgl = (struct ctl_sg_entry *)
1451                                         io->scsiio.kern_data_ptr;
1452
1453                         for (i = msg->dt.sent_sg_entries, j = 0;
1454                              i < (msg->dt.sent_sg_entries +
1455                              msg->dt.cur_sg_entries); i++, j++) {
1456                                 sgl[i].addr = msg->dt.sg_list[j].addr;
1457                                 sgl[i].len = msg->dt.sg_list[j].len;
1458
1459 #if 0
1460                                 printf("%s: DATAMOVE: %p,%lu j=%d, i=%d\n",
1461                                     __func__, sgl[i].addr, sgl[i].len, j, i);
1462 #endif
1463                         }
1464
1465                         /*
1466                          * If this is the last piece of the I/O, we've got
1467                          * the full S/G list.  Queue processing in the thread.
1468                          * Otherwise wait for the next piece.
1469                          */
1470                         if (msg->dt.sg_last != 0)
1471                                 ctl_enqueue_isc(io);
1472                         break;
1473                 }
1474                 /* Performed on the Serializing (primary) SC, XFER mode only */
1475                 case CTL_MSG_DATAMOVE_DONE: {
1476                         if (msg->hdr.serializing_sc == NULL) {
1477                                 printf("%s: serializing_sc == NULL!\n",
1478                                        __func__);
1479                                 /* XXX KDM now what? */
1480                                 break;
1481                         }
1482                         /*
1483                          * We grab the sense information here in case
1484                          * there was a failure, so we can return status
1485                          * back to the initiator.
1486                          */
1487                         io = msg->hdr.serializing_sc;
1488                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
1489                         io->io_hdr.flags &= ~CTL_FLAG_DMA_INPROG;
1490                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1491                         io->io_hdr.port_status = msg->scsi.fetd_status;
1492                         io->scsiio.residual = msg->scsi.residual;
1493                         if (msg->hdr.status != CTL_STATUS_NONE) {
1494                                 io->io_hdr.status = msg->hdr.status;
1495                                 io->scsiio.scsi_status = msg->scsi.scsi_status;
1496                                 io->scsiio.sense_len = msg->scsi.sense_len;
1497                                 io->scsiio.sense_residual =msg->scsi.sense_residual;
1498                                 memcpy(&io->scsiio.sense_data,
1499                                     &msg->scsi.sense_data,
1500                                     msg->scsi.sense_len);
1501                                 if (msg->hdr.status == CTL_SUCCESS)
1502                                         io->io_hdr.flags |= CTL_FLAG_STATUS_SENT;
1503                         }
1504                         ctl_enqueue_isc(io);
1505                         break;
1506                 }
1507
1508                 /* Preformed on Originating SC, SER_ONLY mode */
1509                 case CTL_MSG_R2R:
1510                         io = msg->hdr.original_sc;
1511                         if (io == NULL) {
1512                                 printf("%s: original_sc == NULL!\n",
1513                                     __func__);
1514                                 break;
1515                         }
1516                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1517                         io->io_hdr.msg_type = CTL_MSG_R2R;
1518                         io->io_hdr.serializing_sc = msg->hdr.serializing_sc;
1519                         ctl_enqueue_isc(io);
1520                         break;
1521
1522                 /*
1523                  * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
1524                  * mode.
1525                  * Performed on the Originating (i.e. secondary) SC in XFER
1526                  * mode
1527                  */
1528                 case CTL_MSG_FINISH_IO:
1529                         if (softc->ha_mode == CTL_HA_MODE_XFER)
1530                                 ctl_isc_handler_finish_xfer(softc, msg);
1531                         else
1532                                 ctl_isc_handler_finish_ser_only(softc, msg);
1533                         break;
1534
1535                 /* Preformed on Originating SC */
1536                 case CTL_MSG_BAD_JUJU:
1537                         io = msg->hdr.original_sc;
1538                         if (io == NULL) {
1539                                 printf("%s: Bad JUJU!, original_sc is NULL!\n",
1540                                        __func__);
1541                                 break;
1542                         }
1543                         ctl_copy_sense_data(msg, io);
1544                         /*
1545                          * IO should have already been cleaned up on other
1546                          * SC so clear this flag so we won't send a message
1547                          * back to finish the IO there.
1548                          */
1549                         io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
1550                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1551
1552                         /* io = msg->hdr.serializing_sc; */
1553                         io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
1554                         ctl_enqueue_isc(io);
1555                         break;
1556
1557                 /* Handle resets sent from the other side */
1558                 case CTL_MSG_MANAGE_TASKS: {
1559                         struct ctl_taskio *taskio;
1560                         taskio = (struct ctl_taskio *)ctl_alloc_io(
1561                             softc->othersc_pool);
1562                         ctl_zero_io((union ctl_io *)taskio);
1563                         taskio->io_hdr.io_type = CTL_IO_TASK;
1564                         taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
1565                         taskio->io_hdr.nexus = msg->hdr.nexus;
1566                         taskio->task_action = msg->task.task_action;
1567                         taskio->tag_num = msg->task.tag_num;
1568                         taskio->tag_type = msg->task.tag_type;
1569 #ifdef CTL_TIME_IO
1570                         taskio->io_hdr.start_time = time_uptime;
1571                         getbinuptime(&taskio->io_hdr.start_bt);
1572 #endif /* CTL_TIME_IO */
1573                         ctl_run_task((union ctl_io *)taskio);
1574                         break;
1575                 }
1576                 /* Persistent Reserve action which needs attention */
1577                 case CTL_MSG_PERS_ACTION:
1578                         presio = (struct ctl_prio *)ctl_alloc_io(
1579                             softc->othersc_pool);
1580                         ctl_zero_io((union ctl_io *)presio);
1581                         presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
1582                         presio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
1583                         presio->io_hdr.nexus = msg->hdr.nexus;
1584                         presio->pr_msg = msg->pr;
1585                         ctl_enqueue_isc((union ctl_io *)presio);
1586                         break;
1587                 case CTL_MSG_UA:
1588                         ctl_isc_ua(softc, msg, param);
1589                         break;
1590                 case CTL_MSG_PORT_SYNC:
1591                         ctl_isc_port_sync(softc, msg, param);
1592                         break;
1593                 case CTL_MSG_LUN_SYNC:
1594                         ctl_isc_lun_sync(softc, msg, param);
1595                         break;
1596                 case CTL_MSG_IID_SYNC:
1597                         ctl_isc_iid_sync(softc, msg, param);
1598                         break;
1599                 case CTL_MSG_LOGIN:
1600                         ctl_isc_login(softc, msg, param);
1601                         break;
1602                 case CTL_MSG_MODE_SYNC:
1603                         ctl_isc_mode_sync(softc, msg, param);
1604                         break;
1605                 default:
1606                         printf("Received HA message of unknown type %d\n",
1607                             msg->hdr.msg_type);
1608                         ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1609                         break;
1610                 }
1611                 if (msg != &msgbuf)
1612                         free(msg, M_CTL);
1613         } else if (event == CTL_HA_EVT_LINK_CHANGE) {
1614                 printf("CTL: HA link status changed from %d to %d\n",
1615                     softc->ha_link, param);
1616                 if (param == softc->ha_link)
1617                         return;
1618                 if (softc->ha_link == CTL_HA_LINK_ONLINE) {
1619                         softc->ha_link = param;
1620                         ctl_isc_ha_link_down(softc);
1621                 } else {
1622                         softc->ha_link = param;
1623                         if (softc->ha_link == CTL_HA_LINK_ONLINE)
1624                                 ctl_isc_ha_link_up(softc);
1625                 }
1626                 return;
1627         } else {
1628                 printf("ctl_isc_event_handler: Unknown event %d\n", event);
1629                 return;
1630         }
1631 }
1632
1633 static void
1634 ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
1635 {
1636
1637         memcpy(&dest->scsiio.sense_data, &src->scsi.sense_data,
1638             src->scsi.sense_len);
1639         dest->scsiio.scsi_status = src->scsi.scsi_status;
1640         dest->scsiio.sense_len = src->scsi.sense_len;
1641         dest->io_hdr.status = src->hdr.status;
1642 }
1643
1644 static void
1645 ctl_copy_sense_data_back(union ctl_io *src, union ctl_ha_msg *dest)
1646 {
1647
1648         memcpy(&dest->scsi.sense_data, &src->scsiio.sense_data,
1649             src->scsiio.sense_len);
1650         dest->scsi.scsi_status = src->scsiio.scsi_status;
1651         dest->scsi.sense_len = src->scsiio.sense_len;
1652         dest->hdr.status = src->io_hdr.status;
1653 }
1654
1655 void
1656 ctl_est_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
1657 {
1658         struct ctl_softc *softc = lun->ctl_softc;
1659         ctl_ua_type *pu;
1660
1661         if (initidx < softc->init_min || initidx >= softc->init_max)
1662                 return;
1663         mtx_assert(&lun->lun_lock, MA_OWNED);
1664         pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
1665         if (pu == NULL)
1666                 return;
1667         pu[initidx % CTL_MAX_INIT_PER_PORT] |= ua;
1668 }
1669
1670 void
1671 ctl_est_ua_port(struct ctl_lun *lun, int port, uint32_t except, ctl_ua_type ua)
1672 {
1673         int i;
1674
1675         mtx_assert(&lun->lun_lock, MA_OWNED);
1676         if (lun->pending_ua[port] == NULL)
1677                 return;
1678         for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1679                 if (port * CTL_MAX_INIT_PER_PORT + i == except)
1680                         continue;
1681                 lun->pending_ua[port][i] |= ua;
1682         }
1683 }
1684
1685 void
1686 ctl_est_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1687 {
1688         struct ctl_softc *softc = lun->ctl_softc;
1689         int i;
1690
1691         mtx_assert(&lun->lun_lock, MA_OWNED);
1692         for (i = softc->port_min; i < softc->port_max; i++)
1693                 ctl_est_ua_port(lun, i, except, ua);
1694 }
1695
1696 void
1697 ctl_clr_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
1698 {
1699         struct ctl_softc *softc = lun->ctl_softc;
1700         ctl_ua_type *pu;
1701
1702         if (initidx < softc->init_min || initidx >= softc->init_max)
1703                 return;
1704         mtx_assert(&lun->lun_lock, MA_OWNED);
1705         pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
1706         if (pu == NULL)
1707                 return;
1708         pu[initidx % CTL_MAX_INIT_PER_PORT] &= ~ua;
1709 }
1710
1711 void
1712 ctl_clr_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1713 {
1714         struct ctl_softc *softc = lun->ctl_softc;
1715         int i, j;
1716
1717         mtx_assert(&lun->lun_lock, MA_OWNED);
1718         for (i = softc->port_min; i < softc->port_max; i++) {
1719                 if (lun->pending_ua[i] == NULL)
1720                         continue;
1721                 for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
1722                         if (i * CTL_MAX_INIT_PER_PORT + j == except)
1723                                 continue;
1724                         lun->pending_ua[i][j] &= ~ua;
1725                 }
1726         }
1727 }
1728
1729 void
1730 ctl_clr_ua_allluns(struct ctl_softc *ctl_softc, uint32_t initidx,
1731     ctl_ua_type ua_type)
1732 {
1733         struct ctl_lun *lun;
1734
1735         mtx_assert(&ctl_softc->ctl_lock, MA_OWNED);
1736         STAILQ_FOREACH(lun, &ctl_softc->lun_list, links) {
1737                 mtx_lock(&lun->lun_lock);
1738                 ctl_clr_ua(lun, initidx, ua_type);
1739                 mtx_unlock(&lun->lun_lock);
1740         }
1741 }
1742
1743 static int
1744 ctl_ha_role_sysctl(SYSCTL_HANDLER_ARGS)
1745 {
1746         struct ctl_softc *softc = (struct ctl_softc *)arg1;
1747         struct ctl_lun *lun;
1748         struct ctl_lun_req ireq;
1749         int error, value;
1750
1751         value = (softc->flags & CTL_FLAG_ACTIVE_SHELF) ? 0 : 1;
1752         error = sysctl_handle_int(oidp, &value, 0, req);
1753         if ((error != 0) || (req->newptr == NULL))
1754                 return (error);
1755
1756         mtx_lock(&softc->ctl_lock);
1757         if (value == 0)
1758                 softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1759         else
1760                 softc->flags &= ~CTL_FLAG_ACTIVE_SHELF;
1761         STAILQ_FOREACH(lun, &softc->lun_list, links) {
1762                 mtx_unlock(&softc->ctl_lock);
1763                 bzero(&ireq, sizeof(ireq));
1764                 ireq.reqtype = CTL_LUNREQ_MODIFY;
1765                 ireq.reqdata.modify.lun_id = lun->lun;
1766                 lun->backend->ioctl(NULL, CTL_LUN_REQ, (caddr_t)&ireq, 0,
1767                     curthread);
1768                 if (ireq.status != CTL_LUN_OK) {
1769                         printf("%s: CTL_LUNREQ_MODIFY returned %d '%s'\n",
1770                             __func__, ireq.status, ireq.error_str);
1771                 }
1772                 mtx_lock(&softc->ctl_lock);
1773         }
1774         mtx_unlock(&softc->ctl_lock);
1775         return (0);
1776 }
1777
1778 static int
1779 ctl_init(void)
1780 {
1781         struct make_dev_args args;
1782         struct ctl_softc *softc;
1783         void *other_pool;
1784         int i, error;
1785
1786         softc = control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
1787                                M_WAITOK | M_ZERO);
1788
1789         make_dev_args_init(&args);
1790         args.mda_devsw = &ctl_cdevsw;
1791         args.mda_uid = UID_ROOT;
1792         args.mda_gid = GID_OPERATOR;
1793         args.mda_mode = 0600;
1794         args.mda_si_drv1 = softc;
1795         error = make_dev_s(&args, &softc->dev, "cam/ctl");
1796         if (error != 0) {
1797                 free(control_softc, M_DEVBUF);
1798                 return (error);
1799         }
1800
1801         sysctl_ctx_init(&softc->sysctl_ctx);
1802         softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
1803                 SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
1804                 CTLFLAG_RD, 0, "CAM Target Layer");
1805
1806         if (softc->sysctl_tree == NULL) {
1807                 printf("%s: unable to allocate sysctl tree\n", __func__);
1808                 destroy_dev(softc->dev);
1809                 free(control_softc, M_DEVBUF);
1810                 control_softc = NULL;
1811                 return (ENOMEM);
1812         }
1813
1814         mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
1815         softc->io_zone = uma_zcreate("CTL IO", sizeof(union ctl_io),
1816             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1817         softc->flags = 0;
1818
1819         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1820             OID_AUTO, "ha_mode", CTLFLAG_RDTUN, (int *)&softc->ha_mode, 0,
1821             "HA mode (0 - act/stby, 1 - serialize only, 2 - xfer)");
1822
1823         /*
1824          * In Copan's HA scheme, the "master" and "slave" roles are
1825          * figured out through the slot the controller is in.  Although it
1826          * is an active/active system, someone has to be in charge.
1827          */
1828         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1829             OID_AUTO, "ha_id", CTLFLAG_RDTUN, &softc->ha_id, 0,
1830             "HA head ID (0 - no HA)");
1831         if (softc->ha_id == 0 || softc->ha_id > NUM_HA_SHELVES) {
1832                 softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1833                 softc->is_single = 1;
1834                 softc->port_cnt = CTL_MAX_PORTS;
1835                 softc->port_min = 0;
1836         } else {
1837                 softc->port_cnt = CTL_MAX_PORTS / NUM_HA_SHELVES;
1838                 softc->port_min = (softc->ha_id - 1) * softc->port_cnt;
1839         }
1840         softc->port_max = softc->port_min + softc->port_cnt;
1841         softc->init_min = softc->port_min * CTL_MAX_INIT_PER_PORT;
1842         softc->init_max = softc->port_max * CTL_MAX_INIT_PER_PORT;
1843
1844         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1845             OID_AUTO, "ha_link", CTLFLAG_RD, (int *)&softc->ha_link, 0,
1846             "HA link state (0 - offline, 1 - unknown, 2 - online)");
1847
1848         STAILQ_INIT(&softc->lun_list);
1849         STAILQ_INIT(&softc->pending_lun_queue);
1850         STAILQ_INIT(&softc->fe_list);
1851         STAILQ_INIT(&softc->port_list);
1852         STAILQ_INIT(&softc->be_list);
1853         ctl_tpc_init(softc);
1854
1855         if (ctl_pool_create(softc, "othersc", CTL_POOL_ENTRIES_OTHER_SC,
1856                             &other_pool) != 0)
1857         {
1858                 printf("ctl: can't allocate %d entry other SC pool, "
1859                        "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1860                 return (ENOMEM);
1861         }
1862         softc->othersc_pool = other_pool;
1863
1864         if (worker_threads <= 0)
1865                 worker_threads = max(1, mp_ncpus / 4);
1866         if (worker_threads > CTL_MAX_THREADS)
1867                 worker_threads = CTL_MAX_THREADS;
1868
1869         for (i = 0; i < worker_threads; i++) {
1870                 struct ctl_thread *thr = &softc->threads[i];
1871
1872                 mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1873                 thr->ctl_softc = softc;
1874                 STAILQ_INIT(&thr->incoming_queue);
1875                 STAILQ_INIT(&thr->rtr_queue);
1876                 STAILQ_INIT(&thr->done_queue);
1877                 STAILQ_INIT(&thr->isc_queue);
1878
1879                 error = kproc_kthread_add(ctl_work_thread, thr,
1880                     &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1881                 if (error != 0) {
1882                         printf("error creating CTL work thread!\n");
1883                         ctl_pool_free(other_pool);
1884                         return (error);
1885                 }
1886         }
1887         error = kproc_kthread_add(ctl_lun_thread, softc,
1888             &softc->ctl_proc, NULL, 0, 0, "ctl", "lun");
1889         if (error != 0) {
1890                 printf("error creating CTL lun thread!\n");
1891                 ctl_pool_free(other_pool);
1892                 return (error);
1893         }
1894         error = kproc_kthread_add(ctl_thresh_thread, softc,
1895             &softc->ctl_proc, NULL, 0, 0, "ctl", "thresh");
1896         if (error != 0) {
1897                 printf("error creating CTL threshold thread!\n");
1898                 ctl_pool_free(other_pool);
1899                 return (error);
1900         }
1901
1902         SYSCTL_ADD_PROC(&softc->sysctl_ctx,SYSCTL_CHILDREN(softc->sysctl_tree),
1903             OID_AUTO, "ha_role", CTLTYPE_INT | CTLFLAG_RWTUN,
1904             softc, 0, ctl_ha_role_sysctl, "I", "HA role for this head");
1905
1906         if (softc->is_single == 0) {
1907                 ctl_frontend_register(&ha_frontend);
1908                 if (ctl_ha_msg_init(softc) != CTL_HA_STATUS_SUCCESS) {
1909                         printf("ctl_init: ctl_ha_msg_init failed.\n");
1910                         softc->is_single = 1;
1911                 } else
1912                 if (ctl_ha_msg_register(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
1913                     != CTL_HA_STATUS_SUCCESS) {
1914                         printf("ctl_init: ctl_ha_msg_register failed.\n");
1915                         softc->is_single = 1;
1916                 }
1917         }
1918         return (0);
1919 }
1920
1921 void
1922 ctl_shutdown(void)
1923 {
1924         struct ctl_softc *softc = control_softc;
1925         struct ctl_lun *lun, *next_lun;
1926
1927         if (softc->is_single == 0) {
1928                 ctl_ha_msg_shutdown(softc);
1929                 if (ctl_ha_msg_deregister(CTL_HA_CHAN_CTL)
1930                     != CTL_HA_STATUS_SUCCESS)
1931                         printf("%s: ctl_ha_msg_deregister failed.\n", __func__);
1932                 if (ctl_ha_msg_destroy(softc) != CTL_HA_STATUS_SUCCESS)
1933                         printf("%s: ctl_ha_msg_destroy failed.\n", __func__);
1934                 ctl_frontend_deregister(&ha_frontend);
1935         }
1936
1937         mtx_lock(&softc->ctl_lock);
1938
1939         STAILQ_FOREACH_SAFE(lun, &softc->lun_list, links, next_lun)
1940                 ctl_free_lun(lun);
1941
1942         mtx_unlock(&softc->ctl_lock);
1943
1944 #if 0
1945         ctl_shutdown_thread(softc->work_thread);
1946         mtx_destroy(&softc->queue_lock);
1947 #endif
1948
1949         ctl_tpc_shutdown(softc);
1950         uma_zdestroy(softc->io_zone);
1951         mtx_destroy(&softc->ctl_lock);
1952
1953         destroy_dev(softc->dev);
1954
1955         sysctl_ctx_free(&softc->sysctl_ctx);
1956
1957         free(control_softc, M_DEVBUF);
1958         control_softc = NULL;
1959 }
1960
1961 static int
1962 ctl_module_event_handler(module_t mod, int what, void *arg)
1963 {
1964
1965         switch (what) {
1966         case MOD_LOAD:
1967                 return (ctl_init());
1968         case MOD_UNLOAD:
1969                 return (EBUSY);
1970         default:
1971                 return (EOPNOTSUPP);
1972         }
1973 }
1974
1975 /*
1976  * XXX KDM should we do some access checks here?  Bump a reference count to
1977  * prevent a CTL module from being unloaded while someone has it open?
1978  */
1979 static int
1980 ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1981 {
1982         return (0);
1983 }
1984
1985 static int
1986 ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1987 {
1988         return (0);
1989 }
1990
1991 /*
1992  * Remove an initiator by port number and initiator ID.
1993  * Returns 0 for success, -1 for failure.
1994  */
1995 int
1996 ctl_remove_initiator(struct ctl_port *port, int iid)
1997 {
1998         struct ctl_softc *softc = port->ctl_softc;
1999
2000         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
2001
2002         if (iid > CTL_MAX_INIT_PER_PORT) {
2003                 printf("%s: initiator ID %u > maximun %u!\n",
2004                        __func__, iid, CTL_MAX_INIT_PER_PORT);
2005                 return (-1);
2006         }
2007
2008         mtx_lock(&softc->ctl_lock);
2009         port->wwpn_iid[iid].in_use--;
2010         port->wwpn_iid[iid].last_use = time_uptime;
2011         mtx_unlock(&softc->ctl_lock);
2012         ctl_isc_announce_iid(port, iid);
2013
2014         return (0);
2015 }
2016
2017 /*
2018  * Add an initiator to the initiator map.
2019  * Returns iid for success, < 0 for failure.
2020  */
2021 int
2022 ctl_add_initiator(struct ctl_port *port, int iid, uint64_t wwpn, char *name)
2023 {
2024         struct ctl_softc *softc = port->ctl_softc;
2025         time_t best_time;
2026         int i, best;
2027
2028         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
2029
2030         if (iid >= CTL_MAX_INIT_PER_PORT) {
2031                 printf("%s: WWPN %#jx initiator ID %u > maximum %u!\n",
2032                        __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
2033                 free(name, M_CTL);
2034                 return (-1);
2035         }
2036
2037         mtx_lock(&softc->ctl_lock);
2038
2039         if (iid < 0 && (wwpn != 0 || name != NULL)) {
2040                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2041                         if (wwpn != 0 && wwpn == port->wwpn_iid[i].wwpn) {
2042                                 iid = i;
2043                                 break;
2044                         }
2045                         if (name != NULL && port->wwpn_iid[i].name != NULL &&
2046                             strcmp(name, port->wwpn_iid[i].name) == 0) {
2047                                 iid = i;
2048                                 break;
2049                         }
2050                 }
2051         }
2052
2053         if (iid < 0) {
2054                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2055                         if (port->wwpn_iid[i].in_use == 0 &&
2056                             port->wwpn_iid[i].wwpn == 0 &&
2057                             port->wwpn_iid[i].name == NULL) {
2058                                 iid = i;
2059                                 break;
2060                         }
2061                 }
2062         }
2063
2064         if (iid < 0) {
2065                 best = -1;
2066                 best_time = INT32_MAX;
2067                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2068                         if (port->wwpn_iid[i].in_use == 0) {
2069                                 if (port->wwpn_iid[i].last_use < best_time) {
2070                                         best = i;
2071                                         best_time = port->wwpn_iid[i].last_use;
2072                                 }
2073                         }
2074                 }
2075                 iid = best;
2076         }
2077
2078         if (iid < 0) {
2079                 mtx_unlock(&softc->ctl_lock);
2080                 free(name, M_CTL);
2081                 return (-2);
2082         }
2083
2084         if (port->wwpn_iid[iid].in_use > 0 && (wwpn != 0 || name != NULL)) {
2085                 /*
2086                  * This is not an error yet.
2087                  */
2088                 if (wwpn != 0 && wwpn == port->wwpn_iid[iid].wwpn) {
2089 #if 0
2090                         printf("%s: port %d iid %u WWPN %#jx arrived"
2091                             " again\n", __func__, port->targ_port,
2092                             iid, (uintmax_t)wwpn);
2093 #endif
2094                         goto take;
2095                 }
2096                 if (name != NULL && port->wwpn_iid[iid].name != NULL &&
2097                     strcmp(name, port->wwpn_iid[iid].name) == 0) {
2098 #if 0
2099                         printf("%s: port %d iid %u name '%s' arrived"
2100                             " again\n", __func__, port->targ_port,
2101                             iid, name);
2102 #endif
2103                         goto take;
2104                 }
2105
2106                 /*
2107                  * This is an error, but what do we do about it?  The
2108                  * driver is telling us we have a new WWPN for this
2109                  * initiator ID, so we pretty much need to use it.
2110                  */
2111                 printf("%s: port %d iid %u WWPN %#jx '%s' arrived,"
2112                     " but WWPN %#jx '%s' is still at that address\n",
2113                     __func__, port->targ_port, iid, wwpn, name,
2114                     (uintmax_t)port->wwpn_iid[iid].wwpn,
2115                     port->wwpn_iid[iid].name);
2116
2117                 /*
2118                  * XXX KDM clear have_ca and ua_pending on each LUN for
2119                  * this initiator.
2120                  */
2121         }
2122 take:
2123         free(port->wwpn_iid[iid].name, M_CTL);
2124         port->wwpn_iid[iid].name = name;
2125         port->wwpn_iid[iid].wwpn = wwpn;
2126         port->wwpn_iid[iid].in_use++;
2127         mtx_unlock(&softc->ctl_lock);
2128         ctl_isc_announce_iid(port, iid);
2129
2130         return (iid);
2131 }
2132
2133 static int
2134 ctl_create_iid(struct ctl_port *port, int iid, uint8_t *buf)
2135 {
2136         int len;
2137
2138         switch (port->port_type) {
2139         case CTL_PORT_FC:
2140         {
2141                 struct scsi_transportid_fcp *id =
2142                     (struct scsi_transportid_fcp *)buf;
2143                 if (port->wwpn_iid[iid].wwpn == 0)
2144                         return (0);
2145                 memset(id, 0, sizeof(*id));
2146                 id->format_protocol = SCSI_PROTO_FC;
2147                 scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->n_port_name);
2148                 return (sizeof(*id));
2149         }
2150         case CTL_PORT_ISCSI:
2151         {
2152                 struct scsi_transportid_iscsi_port *id =
2153                     (struct scsi_transportid_iscsi_port *)buf;
2154                 if (port->wwpn_iid[iid].name == NULL)
2155                         return (0);
2156                 memset(id, 0, 256);
2157                 id->format_protocol = SCSI_TRN_ISCSI_FORMAT_PORT |
2158                     SCSI_PROTO_ISCSI;
2159                 len = strlcpy(id->iscsi_name, port->wwpn_iid[iid].name, 252) + 1;
2160                 len = roundup2(min(len, 252), 4);
2161                 scsi_ulto2b(len, id->additional_length);
2162                 return (sizeof(*id) + len);
2163         }
2164         case CTL_PORT_SAS:
2165         {
2166                 struct scsi_transportid_sas *id =
2167                     (struct scsi_transportid_sas *)buf;
2168                 if (port->wwpn_iid[iid].wwpn == 0)
2169                         return (0);
2170                 memset(id, 0, sizeof(*id));
2171                 id->format_protocol = SCSI_PROTO_SAS;
2172                 scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->sas_address);
2173                 return (sizeof(*id));
2174         }
2175         default:
2176         {
2177                 struct scsi_transportid_spi *id =
2178                     (struct scsi_transportid_spi *)buf;
2179                 memset(id, 0, sizeof(*id));
2180                 id->format_protocol = SCSI_PROTO_SPI;
2181                 scsi_ulto2b(iid, id->scsi_addr);
2182                 scsi_ulto2b(port->targ_port, id->rel_trgt_port_id);
2183                 return (sizeof(*id));
2184         }
2185         }
2186 }
2187
2188 /*
2189  * Serialize a command that went down the "wrong" side, and so was sent to
2190  * this controller for execution.  The logic is a little different than the
2191  * standard case in ctl_scsiio_precheck().  Errors in this case need to get
2192  * sent back to the other side, but in the success case, we execute the
2193  * command on this side (XFER mode) or tell the other side to execute it
2194  * (SER_ONLY mode).
2195  */
2196 static int
2197 ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
2198 {
2199         struct ctl_softc *softc = control_softc;
2200         union ctl_ha_msg msg_info;
2201         struct ctl_port *port;
2202         struct ctl_lun *lun;
2203         const struct ctl_cmd_entry *entry;
2204         int retval = 0;
2205         uint32_t targ_lun;
2206
2207         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
2208         mtx_lock(&softc->ctl_lock);
2209
2210         /* Make sure that we know about this port. */
2211         port = ctl_io_port(&ctsio->io_hdr);
2212         if (port == NULL || (port->status & CTL_PORT_STATUS_ONLINE) == 0) {
2213                 ctl_set_internal_failure(ctsio, /*sks_valid*/ 0,
2214                                          /*retry_count*/ 1);
2215                 goto badjuju;
2216         }
2217
2218         /* Make sure that we know about this LUN. */
2219         if ((targ_lun < CTL_MAX_LUNS) &&
2220             ((lun = softc->ctl_luns[targ_lun]) != NULL)) {
2221                 mtx_lock(&lun->lun_lock);
2222                 mtx_unlock(&softc->ctl_lock);
2223                 /*
2224                  * If the LUN is invalid, pretend that it doesn't exist.
2225                  * It will go away as soon as all pending I/O has been
2226                  * completed.
2227                  */
2228                 if (lun->flags & CTL_LUN_DISABLED) {
2229                         mtx_unlock(&lun->lun_lock);
2230                         lun = NULL;
2231                 }
2232         } else {
2233                 mtx_unlock(&softc->ctl_lock);
2234                 lun = NULL;
2235         }
2236         if (lun == NULL) {
2237                 /*
2238                  * The other node would not send this request to us unless
2239                  * received announce that we are primary node for this LUN.
2240                  * If this LUN does not exist now, it is probably result of
2241                  * a race, so respond to initiator in the most opaque way.
2242                  */
2243                 ctl_set_busy(ctsio);
2244                 goto badjuju;
2245         }
2246
2247         entry = ctl_get_cmd_entry(ctsio, NULL);
2248         if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
2249                 mtx_unlock(&lun->lun_lock);
2250                 goto badjuju;
2251         }
2252
2253         ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
2254         ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = lun->be_lun;
2255
2256         /*
2257          * Every I/O goes into the OOA queue for a
2258          * particular LUN, and stays there until completion.
2259          */
2260 #ifdef CTL_TIME_IO
2261         if (TAILQ_EMPTY(&lun->ooa_queue))
2262                 lun->idle_time += getsbinuptime() - lun->last_busy;
2263 #endif
2264         TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2265
2266         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
2267                 (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
2268                  ooa_links))) {
2269         case CTL_ACTION_BLOCK:
2270                 ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
2271                 TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
2272                                   blocked_links);
2273                 mtx_unlock(&lun->lun_lock);
2274                 break;
2275         case CTL_ACTION_PASS:
2276         case CTL_ACTION_SKIP:
2277                 if (softc->ha_mode == CTL_HA_MODE_XFER) {
2278                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
2279                         ctl_enqueue_rtr((union ctl_io *)ctsio);
2280                         mtx_unlock(&lun->lun_lock);
2281                 } else {
2282                         ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
2283                         mtx_unlock(&lun->lun_lock);
2284
2285                         /* send msg back to other side */
2286                         msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
2287                         msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
2288                         msg_info.hdr.msg_type = CTL_MSG_R2R;
2289                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
2290                             sizeof(msg_info.hdr), M_WAITOK);
2291                 }
2292                 break;
2293         case CTL_ACTION_OVERLAP:
2294                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2295                 mtx_unlock(&lun->lun_lock);
2296                 ctl_set_overlapped_cmd(ctsio);
2297                 goto badjuju;
2298         case CTL_ACTION_OVERLAP_TAG:
2299                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2300                 mtx_unlock(&lun->lun_lock);
2301                 ctl_set_overlapped_tag(ctsio, ctsio->tag_num);
2302                 goto badjuju;
2303         case CTL_ACTION_ERROR:
2304         default:
2305                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2306                 mtx_unlock(&lun->lun_lock);
2307
2308                 ctl_set_internal_failure(ctsio, /*sks_valid*/ 0,
2309                                          /*retry_count*/ 0);
2310 badjuju:
2311                 ctl_copy_sense_data_back((union ctl_io *)ctsio, &msg_info);
2312                 msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
2313                 msg_info.hdr.serializing_sc = NULL;
2314                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
2315                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
2316                     sizeof(msg_info.scsi), M_WAITOK);
2317                 retval = 1;
2318                 break;
2319         }
2320         return (retval);
2321 }
2322
2323 /*
2324  * Returns 0 for success, errno for failure.
2325  */
2326 static void
2327 ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
2328                    struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
2329 {
2330         union ctl_io *io;
2331
2332         mtx_lock(&lun->lun_lock);
2333         for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
2334              (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2335              ooa_links)) {
2336                 struct ctl_ooa_entry *entry;
2337
2338                 /*
2339                  * If we've got more than we can fit, just count the
2340                  * remaining entries.
2341                  */
2342                 if (*cur_fill_num >= ooa_hdr->alloc_num)
2343                         continue;
2344
2345                 entry = &kern_entries[*cur_fill_num];
2346
2347                 entry->tag_num = io->scsiio.tag_num;
2348                 entry->lun_num = lun->lun;
2349 #ifdef CTL_TIME_IO
2350                 entry->start_bt = io->io_hdr.start_bt;
2351 #endif
2352                 bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2353                 entry->cdb_len = io->scsiio.cdb_len;
2354                 if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
2355                         entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2356
2357                 if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2358                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2359
2360                 if (io->io_hdr.flags & CTL_FLAG_ABORT)
2361                         entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2362
2363                 if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2364                         entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2365
2366                 if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2367                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2368         }
2369         mtx_unlock(&lun->lun_lock);
2370 }
2371
2372 static void *
2373 ctl_copyin_alloc(void *user_addr, unsigned int len, char *error_str,
2374                  size_t error_str_len)
2375 {
2376         void *kptr;
2377
2378         kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2379
2380         if (copyin(user_addr, kptr, len) != 0) {
2381                 snprintf(error_str, error_str_len, "Error copying %d bytes "
2382                          "from user address %p to kernel address %p", len,
2383                          user_addr, kptr);
2384                 free(kptr, M_CTL);
2385                 return (NULL);
2386         }
2387
2388         return (kptr);
2389 }
2390
2391 static void
2392 ctl_free_args(int num_args, struct ctl_be_arg *args)
2393 {
2394         int i;
2395
2396         if (args == NULL)
2397                 return;
2398
2399         for (i = 0; i < num_args; i++) {
2400                 free(args[i].kname, M_CTL);
2401                 free(args[i].kvalue, M_CTL);
2402         }
2403
2404         free(args, M_CTL);
2405 }
2406
2407 static struct ctl_be_arg *
2408 ctl_copyin_args(int num_args, struct ctl_be_arg *uargs,
2409                 char *error_str, size_t error_str_len)
2410 {
2411         struct ctl_be_arg *args;
2412         int i;
2413
2414         args = ctl_copyin_alloc(uargs, num_args * sizeof(*args),
2415                                 error_str, error_str_len);
2416
2417         if (args == NULL)
2418                 goto bailout;
2419
2420         for (i = 0; i < num_args; i++) {
2421                 args[i].kname = NULL;
2422                 args[i].kvalue = NULL;
2423         }
2424
2425         for (i = 0; i < num_args; i++) {
2426                 uint8_t *tmpptr;
2427
2428                 if (args[i].namelen == 0) {
2429                         snprintf(error_str, error_str_len, "Argument %d "
2430                                  "name length is zero", i);
2431                         goto bailout;
2432                 }
2433
2434                 args[i].kname = ctl_copyin_alloc(args[i].name,
2435                         args[i].namelen, error_str, error_str_len);
2436                 if (args[i].kname == NULL)
2437                         goto bailout;
2438
2439                 if (args[i].kname[args[i].namelen - 1] != '\0') {
2440                         snprintf(error_str, error_str_len, "Argument %d "
2441                                  "name is not NUL-terminated", i);
2442                         goto bailout;
2443                 }
2444
2445                 if (args[i].flags & CTL_BEARG_RD) {
2446                         if (args[i].vallen == 0) {
2447                                 snprintf(error_str, error_str_len, "Argument %d "
2448                                          "value length is zero", i);
2449                                 goto bailout;
2450                         }
2451
2452                         tmpptr = ctl_copyin_alloc(args[i].value,
2453                                 args[i].vallen, error_str, error_str_len);
2454                         if (tmpptr == NULL)
2455                                 goto bailout;
2456
2457                         if ((args[i].flags & CTL_BEARG_ASCII)
2458                          && (tmpptr[args[i].vallen - 1] != '\0')) {
2459                                 snprintf(error_str, error_str_len, "Argument "
2460                                     "%d value is not NUL-terminated", i);
2461                                 free(tmpptr, M_CTL);
2462                                 goto bailout;
2463                         }
2464                         args[i].kvalue = tmpptr;
2465                 } else {
2466                         args[i].kvalue = malloc(args[i].vallen,
2467                             M_CTL, M_WAITOK | M_ZERO);
2468                 }
2469         }
2470
2471         return (args);
2472 bailout:
2473
2474         ctl_free_args(num_args, args);
2475
2476         return (NULL);
2477 }
2478
2479 static void
2480 ctl_copyout_args(int num_args, struct ctl_be_arg *args)
2481 {
2482         int i;
2483
2484         for (i = 0; i < num_args; i++) {
2485                 if (args[i].flags & CTL_BEARG_WR)
2486                         copyout(args[i].kvalue, args[i].value, args[i].vallen);
2487         }
2488 }
2489
2490 /*
2491  * Escape characters that are illegal or not recommended in XML.
2492  */
2493 int
2494 ctl_sbuf_printf_esc(struct sbuf *sb, char *str, int size)
2495 {
2496         char *end = str + size;
2497         int retval;
2498
2499         retval = 0;
2500
2501         for (; *str && str < end; str++) {
2502                 switch (*str) {
2503                 case '&':
2504                         retval = sbuf_printf(sb, "&amp;");
2505                         break;
2506                 case '>':
2507                         retval = sbuf_printf(sb, "&gt;");
2508                         break;
2509                 case '<':
2510                         retval = sbuf_printf(sb, "&lt;");
2511                         break;
2512                 default:
2513                         retval = sbuf_putc(sb, *str);
2514                         break;
2515                 }
2516
2517                 if (retval != 0)
2518                         break;
2519
2520         }
2521
2522         return (retval);
2523 }
2524
2525 static void
2526 ctl_id_sbuf(struct ctl_devid *id, struct sbuf *sb)
2527 {
2528         struct scsi_vpd_id_descriptor *desc;
2529         int i;
2530
2531         if (id == NULL || id->len < 4)
2532                 return;
2533         desc = (struct scsi_vpd_id_descriptor *)id->data;
2534         switch (desc->id_type & SVPD_ID_TYPE_MASK) {
2535         case SVPD_ID_TYPE_T10:
2536                 sbuf_printf(sb, "t10.");
2537                 break;
2538         case SVPD_ID_TYPE_EUI64:
2539                 sbuf_printf(sb, "eui.");
2540                 break;
2541         case SVPD_ID_TYPE_NAA:
2542                 sbuf_printf(sb, "naa.");
2543                 break;
2544         case SVPD_ID_TYPE_SCSI_NAME:
2545                 break;
2546         }
2547         switch (desc->proto_codeset & SVPD_ID_CODESET_MASK) {
2548         case SVPD_ID_CODESET_BINARY:
2549                 for (i = 0; i < desc->length; i++)
2550                         sbuf_printf(sb, "%02x", desc->identifier[i]);
2551                 break;
2552         case SVPD_ID_CODESET_ASCII:
2553                 sbuf_printf(sb, "%.*s", (int)desc->length,
2554                     (char *)desc->identifier);
2555                 break;
2556         case SVPD_ID_CODESET_UTF8:
2557                 sbuf_printf(sb, "%s", (char *)desc->identifier);
2558                 break;
2559         }
2560 }
2561
2562 static int
2563 ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2564           struct thread *td)
2565 {
2566         struct ctl_softc *softc = dev->si_drv1;
2567         struct ctl_lun *lun;
2568         int retval;
2569
2570         retval = 0;
2571
2572         switch (cmd) {
2573         case CTL_IO:
2574                 retval = ctl_ioctl_io(dev, cmd, addr, flag, td);
2575                 break;
2576         case CTL_ENABLE_PORT:
2577         case CTL_DISABLE_PORT:
2578         case CTL_SET_PORT_WWNS: {
2579                 struct ctl_port *port;
2580                 struct ctl_port_entry *entry;
2581
2582                 entry = (struct ctl_port_entry *)addr;
2583                 
2584                 mtx_lock(&softc->ctl_lock);
2585                 STAILQ_FOREACH(port, &softc->port_list, links) {
2586                         int action, done;
2587
2588                         if (port->targ_port < softc->port_min ||
2589                             port->targ_port >= softc->port_max)
2590                                 continue;
2591
2592                         action = 0;
2593                         done = 0;
2594                         if ((entry->port_type == CTL_PORT_NONE)
2595                          && (entry->targ_port == port->targ_port)) {
2596                                 /*
2597                                  * If the user only wants to enable or
2598                                  * disable or set WWNs on a specific port,
2599                                  * do the operation and we're done.
2600                                  */
2601                                 action = 1;
2602                                 done = 1;
2603                         } else if (entry->port_type & port->port_type) {
2604                                 /*
2605                                  * Compare the user's type mask with the
2606                                  * particular frontend type to see if we
2607                                  * have a match.
2608                                  */
2609                                 action = 1;
2610                                 done = 0;
2611
2612                                 /*
2613                                  * Make sure the user isn't trying to set
2614                                  * WWNs on multiple ports at the same time.
2615                                  */
2616                                 if (cmd == CTL_SET_PORT_WWNS) {
2617                                         printf("%s: Can't set WWNs on "
2618                                                "multiple ports\n", __func__);
2619                                         retval = EINVAL;
2620                                         break;
2621                                 }
2622                         }
2623                         if (action == 0)
2624                                 continue;
2625
2626                         /*
2627                          * XXX KDM we have to drop the lock here, because
2628                          * the online/offline operations can potentially
2629                          * block.  We need to reference count the frontends
2630                          * so they can't go away,
2631                          */
2632                         if (cmd == CTL_ENABLE_PORT) {
2633                                 mtx_unlock(&softc->ctl_lock);
2634                                 ctl_port_online(port);
2635                                 mtx_lock(&softc->ctl_lock);
2636                         } else if (cmd == CTL_DISABLE_PORT) {
2637                                 mtx_unlock(&softc->ctl_lock);
2638                                 ctl_port_offline(port);
2639                                 mtx_lock(&softc->ctl_lock);
2640                         } else if (cmd == CTL_SET_PORT_WWNS) {
2641                                 ctl_port_set_wwns(port,
2642                                     (entry->flags & CTL_PORT_WWNN_VALID) ?
2643                                     1 : 0, entry->wwnn,
2644                                     (entry->flags & CTL_PORT_WWPN_VALID) ?
2645                                     1 : 0, entry->wwpn);
2646                         }
2647                         if (done != 0)
2648                                 break;
2649                 }
2650                 mtx_unlock(&softc->ctl_lock);
2651                 break;
2652         }
2653         case CTL_GET_OOA: {
2654                 struct ctl_ooa *ooa_hdr;
2655                 struct ctl_ooa_entry *entries;
2656                 uint32_t cur_fill_num;
2657
2658                 ooa_hdr = (struct ctl_ooa *)addr;
2659
2660                 if ((ooa_hdr->alloc_len == 0)
2661                  || (ooa_hdr->alloc_num == 0)) {
2662                         printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2663                                "must be non-zero\n", __func__,
2664                                ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2665                         retval = EINVAL;
2666                         break;
2667                 }
2668
2669                 if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2670                     sizeof(struct ctl_ooa_entry))) {
2671                         printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2672                                "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2673                                __func__, ooa_hdr->alloc_len,
2674                                ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2675                         retval = EINVAL;
2676                         break;
2677                 }
2678
2679                 entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2680                 if (entries == NULL) {
2681                         printf("%s: could not allocate %d bytes for OOA "
2682                                "dump\n", __func__, ooa_hdr->alloc_len);
2683                         retval = ENOMEM;
2684                         break;
2685                 }
2686
2687                 mtx_lock(&softc->ctl_lock);
2688                 if (((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0)
2689                  && ((ooa_hdr->lun_num >= CTL_MAX_LUNS)
2690                   || (softc->ctl_luns[ooa_hdr->lun_num] == NULL))) {
2691                         mtx_unlock(&softc->ctl_lock);
2692                         free(entries, M_CTL);
2693                         printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2694                                __func__, (uintmax_t)ooa_hdr->lun_num);
2695                         retval = EINVAL;
2696                         break;
2697                 }
2698
2699                 cur_fill_num = 0;
2700
2701                 if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2702                         STAILQ_FOREACH(lun, &softc->lun_list, links) {
2703                                 ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2704                                     ooa_hdr, entries);
2705                         }
2706                 } else {
2707                         lun = softc->ctl_luns[ooa_hdr->lun_num];
2708                         ctl_ioctl_fill_ooa(lun, &cur_fill_num, ooa_hdr,
2709                             entries);
2710                 }
2711                 mtx_unlock(&softc->ctl_lock);
2712
2713                 ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2714                 ooa_hdr->fill_len = ooa_hdr->fill_num *
2715                         sizeof(struct ctl_ooa_entry);
2716                 retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2717                 if (retval != 0) {
2718                         printf("%s: error copying out %d bytes for OOA dump\n", 
2719                                __func__, ooa_hdr->fill_len);
2720                 }
2721
2722                 getbinuptime(&ooa_hdr->cur_bt);
2723
2724                 if (cur_fill_num > ooa_hdr->alloc_num) {
2725                         ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2726                         ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2727                 } else {
2728                         ooa_hdr->dropped_num = 0;
2729                         ooa_hdr->status = CTL_OOA_OK;
2730                 }
2731
2732                 free(entries, M_CTL);
2733                 break;
2734         }
2735         case CTL_DELAY_IO: {
2736                 struct ctl_io_delay_info *delay_info;
2737
2738                 delay_info = (struct ctl_io_delay_info *)addr;
2739
2740 #ifdef CTL_IO_DELAY
2741                 mtx_lock(&softc->ctl_lock);
2742
2743                 if ((delay_info->lun_id >= CTL_MAX_LUNS)
2744                  || (softc->ctl_luns[delay_info->lun_id] == NULL)) {
2745                         delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2746                 } else {
2747                         lun = softc->ctl_luns[delay_info->lun_id];
2748                         mtx_lock(&lun->lun_lock);
2749
2750                         delay_info->status = CTL_DELAY_STATUS_OK;
2751
2752                         switch (delay_info->delay_type) {
2753                         case CTL_DELAY_TYPE_CONT:
2754                                 break;
2755                         case CTL_DELAY_TYPE_ONESHOT:
2756                                 break;
2757                         default:
2758                                 delay_info->status =
2759                                         CTL_DELAY_STATUS_INVALID_TYPE;
2760                                 break;
2761                         }
2762
2763                         switch (delay_info->delay_loc) {
2764                         case CTL_DELAY_LOC_DATAMOVE:
2765                                 lun->delay_info.datamove_type =
2766                                         delay_info->delay_type;
2767                                 lun->delay_info.datamove_delay =
2768                                         delay_info->delay_secs;
2769                                 break;
2770                         case CTL_DELAY_LOC_DONE:
2771                                 lun->delay_info.done_type =
2772                                         delay_info->delay_type;
2773                                 lun->delay_info.done_delay =
2774                                         delay_info->delay_secs;
2775                                 break;
2776                         default:
2777                                 delay_info->status =
2778                                         CTL_DELAY_STATUS_INVALID_LOC;
2779                                 break;
2780                         }
2781                         mtx_unlock(&lun->lun_lock);
2782                 }
2783
2784                 mtx_unlock(&softc->ctl_lock);
2785 #else
2786                 delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2787 #endif /* CTL_IO_DELAY */
2788                 break;
2789         }
2790         case CTL_GETSTATS: {
2791                 struct ctl_stats *stats;
2792                 int i;
2793
2794                 stats = (struct ctl_stats *)addr;
2795
2796                 if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) >
2797                      stats->alloc_len) {
2798                         stats->status = CTL_SS_NEED_MORE_SPACE;
2799                         stats->num_luns = softc->num_luns;
2800                         break;
2801                 }
2802                 /*
2803                  * XXX KDM no locking here.  If the LUN list changes,
2804                  * things can blow up.
2805                  */
2806                 i = 0;
2807                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
2808                         retval = copyout(&lun->stats, &stats->lun_stats[i++],
2809                                          sizeof(lun->stats));
2810                         if (retval != 0)
2811                                 break;
2812                 }
2813                 stats->num_luns = softc->num_luns;
2814                 stats->fill_len = sizeof(struct ctl_lun_io_stats) *
2815                                  softc->num_luns;
2816                 stats->status = CTL_SS_OK;
2817 #ifdef CTL_TIME_IO
2818                 stats->flags = CTL_STATS_FLAG_TIME_VALID;
2819 #else
2820                 stats->flags = CTL_STATS_FLAG_NONE;
2821 #endif
2822                 getnanouptime(&stats->timestamp);
2823                 break;
2824         }
2825         case CTL_ERROR_INJECT: {
2826                 struct ctl_error_desc *err_desc, *new_err_desc;
2827
2828                 err_desc = (struct ctl_error_desc *)addr;
2829
2830                 new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2831                                       M_WAITOK | M_ZERO);
2832                 bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2833
2834                 mtx_lock(&softc->ctl_lock);
2835                 lun = softc->ctl_luns[err_desc->lun_id];
2836                 if (lun == NULL) {
2837                         mtx_unlock(&softc->ctl_lock);
2838                         free(new_err_desc, M_CTL);
2839                         printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2840                                __func__, (uintmax_t)err_desc->lun_id);
2841                         retval = EINVAL;
2842                         break;
2843                 }
2844                 mtx_lock(&lun->lun_lock);
2845                 mtx_unlock(&softc->ctl_lock);
2846
2847                 /*
2848                  * We could do some checking here to verify the validity
2849                  * of the request, but given the complexity of error
2850                  * injection requests, the checking logic would be fairly
2851                  * complex.
2852                  *
2853                  * For now, if the request is invalid, it just won't get
2854                  * executed and might get deleted.
2855                  */
2856                 STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2857
2858                 /*
2859                  * XXX KDM check to make sure the serial number is unique,
2860                  * in case we somehow manage to wrap.  That shouldn't
2861                  * happen for a very long time, but it's the right thing to
2862                  * do.
2863                  */
2864                 new_err_desc->serial = lun->error_serial;
2865                 err_desc->serial = lun->error_serial;
2866                 lun->error_serial++;
2867
2868                 mtx_unlock(&lun->lun_lock);
2869                 break;
2870         }
2871         case CTL_ERROR_INJECT_DELETE: {
2872                 struct ctl_error_desc *delete_desc, *desc, *desc2;
2873                 int delete_done;
2874
2875                 delete_desc = (struct ctl_error_desc *)addr;
2876                 delete_done = 0;
2877
2878                 mtx_lock(&softc->ctl_lock);
2879                 lun = softc->ctl_luns[delete_desc->lun_id];
2880                 if (lun == NULL) {
2881                         mtx_unlock(&softc->ctl_lock);
2882                         printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2883                                __func__, (uintmax_t)delete_desc->lun_id);
2884                         retval = EINVAL;
2885                         break;
2886                 }
2887                 mtx_lock(&lun->lun_lock);
2888                 mtx_unlock(&softc->ctl_lock);
2889                 STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2890                         if (desc->serial != delete_desc->serial)
2891                                 continue;
2892
2893                         STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2894                                       links);
2895                         free(desc, M_CTL);
2896                         delete_done = 1;
2897                 }
2898                 mtx_unlock(&lun->lun_lock);
2899                 if (delete_done == 0) {
2900                         printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2901                                "error serial %ju on LUN %u\n", __func__, 
2902                                delete_desc->serial, delete_desc->lun_id);
2903                         retval = EINVAL;
2904                         break;
2905                 }
2906                 break;
2907         }
2908         case CTL_DUMP_STRUCTS: {
2909                 int i, j, k;
2910                 struct ctl_port *port;
2911                 struct ctl_frontend *fe;
2912
2913                 mtx_lock(&softc->ctl_lock);
2914                 printf("CTL Persistent Reservation information start:\n");
2915                 for (i = 0; i < CTL_MAX_LUNS; i++) {
2916                         lun = softc->ctl_luns[i];
2917
2918                         if ((lun == NULL)
2919                          || ((lun->flags & CTL_LUN_DISABLED) != 0))
2920                                 continue;
2921
2922                         for (j = 0; j < CTL_MAX_PORTS; j++) {
2923                                 if (lun->pr_keys[j] == NULL)
2924                                         continue;
2925                                 for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2926                                         if (lun->pr_keys[j][k] == 0)
2927                                                 continue;
2928                                         printf("  LUN %d port %d iid %d key "
2929                                                "%#jx\n", i, j, k,
2930                                                (uintmax_t)lun->pr_keys[j][k]);
2931                                 }
2932                         }
2933                 }
2934                 printf("CTL Persistent Reservation information end\n");
2935                 printf("CTL Ports:\n");
2936                 STAILQ_FOREACH(port, &softc->port_list, links) {
2937                         printf("  Port %d '%s' Frontend '%s' Type %u pp %d vp %d WWNN "
2938                                "%#jx WWPN %#jx\n", port->targ_port, port->port_name,
2939                                port->frontend->name, port->port_type,
2940                                port->physical_port, port->virtual_port,
2941                                (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
2942                         for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
2943                                 if (port->wwpn_iid[j].in_use == 0 &&
2944                                     port->wwpn_iid[j].wwpn == 0 &&
2945                                     port->wwpn_iid[j].name == NULL)
2946                                         continue;
2947
2948                                 printf("    iid %u use %d WWPN %#jx '%s'\n",
2949                                     j, port->wwpn_iid[j].in_use,
2950                                     (uintmax_t)port->wwpn_iid[j].wwpn,
2951                                     port->wwpn_iid[j].name);
2952                         }
2953                 }
2954                 printf("CTL Port information end\n");
2955                 mtx_unlock(&softc->ctl_lock);
2956                 /*
2957                  * XXX KDM calling this without a lock.  We'd likely want
2958                  * to drop the lock before calling the frontend's dump
2959                  * routine anyway.
2960                  */
2961                 printf("CTL Frontends:\n");
2962                 STAILQ_FOREACH(fe, &softc->fe_list, links) {
2963                         printf("  Frontend '%s'\n", fe->name);
2964                         if (fe->fe_dump != NULL)
2965                                 fe->fe_dump();
2966                 }
2967                 printf("CTL Frontend information end\n");
2968                 break;
2969         }
2970         case CTL_LUN_REQ: {
2971                 struct ctl_lun_req *lun_req;
2972                 struct ctl_backend_driver *backend;
2973
2974                 lun_req = (struct ctl_lun_req *)addr;
2975
2976                 backend = ctl_backend_find(lun_req->backend);
2977                 if (backend == NULL) {
2978                         lun_req->status = CTL_LUN_ERROR;
2979                         snprintf(lun_req->error_str,
2980                                  sizeof(lun_req->error_str),
2981                                  "Backend \"%s\" not found.",
2982                                  lun_req->backend);
2983                         break;
2984                 }
2985                 if (lun_req->num_be_args > 0) {
2986                         lun_req->kern_be_args = ctl_copyin_args(
2987                                 lun_req->num_be_args,
2988                                 lun_req->be_args,
2989                                 lun_req->error_str,
2990                                 sizeof(lun_req->error_str));
2991                         if (lun_req->kern_be_args == NULL) {
2992                                 lun_req->status = CTL_LUN_ERROR;
2993                                 break;
2994                         }
2995                 }
2996
2997                 retval = backend->ioctl(dev, cmd, addr, flag, td);
2998
2999                 if (lun_req->num_be_args > 0) {
3000                         ctl_copyout_args(lun_req->num_be_args,
3001                                       lun_req->kern_be_args);
3002                         ctl_free_args(lun_req->num_be_args,
3003                                       lun_req->kern_be_args);
3004                 }
3005                 break;
3006         }
3007         case CTL_LUN_LIST: {
3008                 struct sbuf *sb;
3009                 struct ctl_lun_list *list;
3010                 struct ctl_option *opt;
3011
3012                 list = (struct ctl_lun_list *)addr;
3013
3014                 /*
3015                  * Allocate a fixed length sbuf here, based on the length
3016                  * of the user's buffer.  We could allocate an auto-extending
3017                  * buffer, and then tell the user how much larger our
3018                  * amount of data is than his buffer, but that presents
3019                  * some problems:
3020                  *
3021                  * 1.  The sbuf(9) routines use a blocking malloc, and so
3022                  *     we can't hold a lock while calling them with an
3023                  *     auto-extending buffer.
3024                  *
3025                  * 2.  There is not currently a LUN reference counting
3026                  *     mechanism, outside of outstanding transactions on
3027                  *     the LUN's OOA queue.  So a LUN could go away on us
3028                  *     while we're getting the LUN number, backend-specific
3029                  *     information, etc.  Thus, given the way things
3030                  *     currently work, we need to hold the CTL lock while
3031                  *     grabbing LUN information.
3032                  *
3033                  * So, from the user's standpoint, the best thing to do is
3034                  * allocate what he thinks is a reasonable buffer length,
3035                  * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
3036                  * double the buffer length and try again.  (And repeat
3037                  * that until he succeeds.)
3038                  */
3039                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3040                 if (sb == NULL) {
3041                         list->status = CTL_LUN_LIST_ERROR;
3042                         snprintf(list->error_str, sizeof(list->error_str),
3043                                  "Unable to allocate %d bytes for LUN list",
3044                                  list->alloc_len);
3045                         break;
3046                 }
3047
3048                 sbuf_printf(sb, "<ctllunlist>\n");
3049
3050                 mtx_lock(&softc->ctl_lock);
3051                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
3052                         mtx_lock(&lun->lun_lock);
3053                         retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3054                                              (uintmax_t)lun->lun);
3055
3056                         /*
3057                          * Bail out as soon as we see that we've overfilled
3058                          * the buffer.
3059                          */
3060                         if (retval != 0)
3061                                 break;
3062
3063                         retval = sbuf_printf(sb, "\t<backend_type>%s"
3064                                              "</backend_type>\n",
3065                                              (lun->backend == NULL) ?  "none" :
3066                                              lun->backend->name);
3067
3068                         if (retval != 0)
3069                                 break;
3070
3071                         retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
3072                                              lun->be_lun->lun_type);
3073
3074                         if (retval != 0)
3075                                 break;
3076
3077                         if (lun->backend == NULL) {
3078                                 retval = sbuf_printf(sb, "</lun>\n");
3079                                 if (retval != 0)
3080                                         break;
3081                                 continue;
3082                         }
3083
3084                         retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
3085                                              (lun->be_lun->maxlba > 0) ?
3086                                              lun->be_lun->maxlba + 1 : 0);
3087
3088                         if (retval != 0)
3089                                 break;
3090
3091                         retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
3092                                              lun->be_lun->blocksize);
3093
3094                         if (retval != 0)
3095                                 break;
3096
3097                         retval = sbuf_printf(sb, "\t<serial_number>");
3098
3099                         if (retval != 0)
3100                                 break;
3101
3102                         retval = ctl_sbuf_printf_esc(sb,
3103                             lun->be_lun->serial_num,
3104                             sizeof(lun->be_lun->serial_num));
3105
3106                         if (retval != 0)
3107                                 break;
3108
3109                         retval = sbuf_printf(sb, "</serial_number>\n");
3110                 
3111                         if (retval != 0)
3112                                 break;
3113
3114                         retval = sbuf_printf(sb, "\t<device_id>");
3115
3116                         if (retval != 0)
3117                                 break;
3118
3119                         retval = ctl_sbuf_printf_esc(sb,
3120                             lun->be_lun->device_id,
3121                             sizeof(lun->be_lun->device_id));
3122
3123                         if (retval != 0)
3124                                 break;
3125
3126                         retval = sbuf_printf(sb, "</device_id>\n");
3127
3128                         if (retval != 0)
3129                                 break;
3130
3131                         if (lun->backend->lun_info != NULL) {
3132                                 retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3133                                 if (retval != 0)
3134                                         break;
3135                         }
3136                         STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
3137                                 retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3138                                     opt->name, opt->value, opt->name);
3139                                 if (retval != 0)
3140                                         break;
3141                         }
3142
3143                         retval = sbuf_printf(sb, "</lun>\n");
3144
3145                         if (retval != 0)
3146                                 break;
3147                         mtx_unlock(&lun->lun_lock);
3148                 }
3149                 if (lun != NULL)
3150                         mtx_unlock(&lun->lun_lock);
3151                 mtx_unlock(&softc->ctl_lock);
3152
3153                 if ((retval != 0)
3154                  || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3155                         retval = 0;
3156                         sbuf_delete(sb);
3157                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3158                         snprintf(list->error_str, sizeof(list->error_str),
3159                                  "Out of space, %d bytes is too small",
3160                                  list->alloc_len);
3161                         break;
3162                 }
3163
3164                 sbuf_finish(sb);
3165
3166                 retval = copyout(sbuf_data(sb), list->lun_xml,
3167                                  sbuf_len(sb) + 1);
3168
3169                 list->fill_len = sbuf_len(sb) + 1;
3170                 list->status = CTL_LUN_LIST_OK;
3171                 sbuf_delete(sb);
3172                 break;
3173         }
3174         case CTL_ISCSI: {
3175                 struct ctl_iscsi *ci;
3176                 struct ctl_frontend *fe;
3177
3178                 ci = (struct ctl_iscsi *)addr;
3179
3180                 fe = ctl_frontend_find("iscsi");
3181                 if (fe == NULL) {
3182                         ci->status = CTL_ISCSI_ERROR;
3183                         snprintf(ci->error_str, sizeof(ci->error_str),
3184                             "Frontend \"iscsi\" not found.");
3185                         break;
3186                 }
3187
3188                 retval = fe->ioctl(dev, cmd, addr, flag, td);
3189                 break;
3190         }
3191         case CTL_PORT_REQ: {
3192                 struct ctl_req *req;
3193                 struct ctl_frontend *fe;
3194
3195                 req = (struct ctl_req *)addr;
3196
3197                 fe = ctl_frontend_find(req->driver);
3198                 if (fe == NULL) {
3199                         req->status = CTL_LUN_ERROR;
3200                         snprintf(req->error_str, sizeof(req->error_str),
3201                             "Frontend \"%s\" not found.", req->driver);
3202                         break;
3203                 }
3204                 if (req->num_args > 0) {
3205                         req->kern_args = ctl_copyin_args(req->num_args,
3206                             req->args, req->error_str, sizeof(req->error_str));
3207                         if (req->kern_args == NULL) {
3208                                 req->status = CTL_LUN_ERROR;
3209                                 break;
3210                         }
3211                 }
3212
3213                 if (fe->ioctl)
3214                         retval = fe->ioctl(dev, cmd, addr, flag, td);
3215                 else
3216                         retval = ENODEV;
3217
3218                 if (req->num_args > 0) {
3219                         ctl_copyout_args(req->num_args, req->kern_args);
3220                         ctl_free_args(req->num_args, req->kern_args);
3221                 }
3222                 break;
3223         }
3224         case CTL_PORT_LIST: {
3225                 struct sbuf *sb;
3226                 struct ctl_port *port;
3227                 struct ctl_lun_list *list;
3228                 struct ctl_option *opt;
3229                 int j;
3230                 uint32_t plun;
3231
3232                 list = (struct ctl_lun_list *)addr;
3233
3234                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3235                 if (sb == NULL) {
3236                         list->status = CTL_LUN_LIST_ERROR;
3237                         snprintf(list->error_str, sizeof(list->error_str),
3238                                  "Unable to allocate %d bytes for LUN list",
3239                                  list->alloc_len);
3240                         break;
3241                 }
3242
3243                 sbuf_printf(sb, "<ctlportlist>\n");
3244
3245                 mtx_lock(&softc->ctl_lock);
3246                 STAILQ_FOREACH(port, &softc->port_list, links) {
3247                         retval = sbuf_printf(sb, "<targ_port id=\"%ju\">\n",
3248                                              (uintmax_t)port->targ_port);
3249
3250                         /*
3251                          * Bail out as soon as we see that we've overfilled
3252                          * the buffer.
3253                          */
3254                         if (retval != 0)
3255                                 break;
3256
3257                         retval = sbuf_printf(sb, "\t<frontend_type>%s"
3258                             "</frontend_type>\n", port->frontend->name);
3259                         if (retval != 0)
3260                                 break;
3261
3262                         retval = sbuf_printf(sb, "\t<port_type>%d</port_type>\n",
3263                                              port->port_type);
3264                         if (retval != 0)
3265                                 break;
3266
3267                         retval = sbuf_printf(sb, "\t<online>%s</online>\n",
3268                             (port->status & CTL_PORT_STATUS_ONLINE) ? "YES" : "NO");
3269                         if (retval != 0)
3270                                 break;
3271
3272                         retval = sbuf_printf(sb, "\t<port_name>%s</port_name>\n",
3273                             port->port_name);
3274                         if (retval != 0)
3275                                 break;
3276
3277                         retval = sbuf_printf(sb, "\t<physical_port>%d</physical_port>\n",
3278                             port->physical_port);
3279                         if (retval != 0)
3280                                 break;
3281
3282                         retval = sbuf_printf(sb, "\t<virtual_port>%d</virtual_port>\n",
3283                             port->virtual_port);
3284                         if (retval != 0)
3285                                 break;
3286
3287                         if (port->target_devid != NULL) {
3288                                 sbuf_printf(sb, "\t<target>");
3289                                 ctl_id_sbuf(port->target_devid, sb);
3290                                 sbuf_printf(sb, "</target>\n");
3291                         }
3292
3293                         if (port->port_devid != NULL) {
3294                                 sbuf_printf(sb, "\t<port>");
3295                                 ctl_id_sbuf(port->port_devid, sb);
3296                                 sbuf_printf(sb, "</port>\n");
3297                         }
3298
3299                         if (port->port_info != NULL) {
3300                                 retval = port->port_info(port->onoff_arg, sb);
3301                                 if (retval != 0)
3302                                         break;
3303                         }
3304                         STAILQ_FOREACH(opt, &port->options, links) {
3305                                 retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3306                                     opt->name, opt->value, opt->name);
3307                                 if (retval != 0)
3308                                         break;
3309                         }
3310
3311                         if (port->lun_map != NULL) {
3312                                 sbuf_printf(sb, "\t<lun_map>on</lun_map>\n");
3313                                 for (j = 0; j < CTL_MAX_LUNS; j++) {
3314                                         plun = ctl_lun_map_from_port(port, j);
3315                                         if (plun >= CTL_MAX_LUNS)
3316                                                 continue;
3317                                         sbuf_printf(sb,
3318                                             "\t<lun id=\"%u\">%u</lun>\n",
3319                                             j, plun);
3320                                 }
3321                         }
3322
3323                         for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3324                                 if (port->wwpn_iid[j].in_use == 0 ||
3325                                     (port->wwpn_iid[j].wwpn == 0 &&
3326                                      port->wwpn_iid[j].name == NULL))
3327                                         continue;
3328
3329                                 if (port->wwpn_iid[j].name != NULL)
3330                                         retval = sbuf_printf(sb,
3331                                             "\t<initiator id=\"%u\">%s</initiator>\n",
3332                                             j, port->wwpn_iid[j].name);
3333                                 else
3334                                         retval = sbuf_printf(sb,
3335                                             "\t<initiator id=\"%u\">naa.%08jx</initiator>\n",
3336                                             j, port->wwpn_iid[j].wwpn);
3337                                 if (retval != 0)
3338                                         break;
3339                         }
3340                         if (retval != 0)
3341                                 break;
3342
3343                         retval = sbuf_printf(sb, "</targ_port>\n");
3344                         if (retval != 0)
3345                                 break;
3346                 }
3347                 mtx_unlock(&softc->ctl_lock);
3348
3349                 if ((retval != 0)
3350                  || ((retval = sbuf_printf(sb, "</ctlportlist>\n")) != 0)) {
3351                         retval = 0;
3352                         sbuf_delete(sb);
3353                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3354                         snprintf(list->error_str, sizeof(list->error_str),
3355                                  "Out of space, %d bytes is too small",
3356                                  list->alloc_len);
3357                         break;
3358                 }
3359
3360                 sbuf_finish(sb);
3361
3362                 retval = copyout(sbuf_data(sb), list->lun_xml,
3363                                  sbuf_len(sb) + 1);
3364
3365                 list->fill_len = sbuf_len(sb) + 1;
3366                 list->status = CTL_LUN_LIST_OK;
3367                 sbuf_delete(sb);
3368                 break;
3369         }
3370         case CTL_LUN_MAP: {
3371                 struct ctl_lun_map *lm  = (struct ctl_lun_map *)addr;
3372                 struct ctl_port *port;
3373
3374                 mtx_lock(&softc->ctl_lock);
3375                 if (lm->port < softc->port_min ||
3376                     lm->port >= softc->port_max ||
3377                     (port = softc->ctl_ports[lm->port]) == NULL) {
3378                         mtx_unlock(&softc->ctl_lock);
3379                         return (ENXIO);
3380                 }
3381                 if (port->status & CTL_PORT_STATUS_ONLINE) {
3382                         STAILQ_FOREACH(lun, &softc->lun_list, links) {
3383                                 if (ctl_lun_map_to_port(port, lun->lun) >=
3384                                     CTL_MAX_LUNS)
3385                                         continue;
3386                                 mtx_lock(&lun->lun_lock);
3387                                 ctl_est_ua_port(lun, lm->port, -1,
3388                                     CTL_UA_LUN_CHANGE);
3389                                 mtx_unlock(&lun->lun_lock);
3390                         }
3391                 }
3392                 mtx_unlock(&softc->ctl_lock); // XXX: port_enable sleeps
3393                 if (lm->plun < CTL_MAX_LUNS) {
3394                         if (lm->lun == UINT32_MAX)
3395                                 retval = ctl_lun_map_unset(port, lm->plun);
3396                         else if (lm->lun < CTL_MAX_LUNS &&
3397                             softc->ctl_luns[lm->lun] != NULL)
3398                                 retval = ctl_lun_map_set(port, lm->plun, lm->lun);
3399                         else
3400                                 return (ENXIO);
3401                 } else if (lm->plun == UINT32_MAX) {
3402                         if (lm->lun == UINT32_MAX)
3403                                 retval = ctl_lun_map_deinit(port);
3404                         else
3405                                 retval = ctl_lun_map_init(port);
3406                 } else
3407                         return (ENXIO);
3408                 if (port->status & CTL_PORT_STATUS_ONLINE)
3409                         ctl_isc_announce_port(port);
3410                 break;
3411         }
3412         default: {
3413                 /* XXX KDM should we fix this? */
3414 #if 0
3415                 struct ctl_backend_driver *backend;
3416                 unsigned int type;
3417                 int found;
3418
3419                 found = 0;
3420
3421                 /*
3422                  * We encode the backend type as the ioctl type for backend
3423                  * ioctls.  So parse it out here, and then search for a
3424                  * backend of this type.
3425                  */
3426                 type = _IOC_TYPE(cmd);
3427
3428                 STAILQ_FOREACH(backend, &softc->be_list, links) {
3429                         if (backend->type == type) {
3430                                 found = 1;
3431                                 break;
3432                         }
3433                 }
3434                 if (found == 0) {
3435                         printf("ctl: unknown ioctl command %#lx or backend "
3436                                "%d\n", cmd, type);
3437                         retval = EINVAL;
3438                         break;
3439                 }
3440                 retval = backend->ioctl(dev, cmd, addr, flag, td);
3441 #endif
3442                 retval = ENOTTY;
3443                 break;
3444         }
3445         }
3446         return (retval);
3447 }
3448
3449 uint32_t
3450 ctl_get_initindex(struct ctl_nexus *nexus)
3451 {
3452         return (nexus->initid + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3453 }
3454
3455 int
3456 ctl_lun_map_init(struct ctl_port *port)
3457 {
3458         struct ctl_softc *softc = port->ctl_softc;
3459         struct ctl_lun *lun;
3460         uint32_t i;
3461
3462         if (port->lun_map == NULL)
3463                 port->lun_map = malloc(sizeof(uint32_t) * CTL_MAX_LUNS,
3464                     M_CTL, M_NOWAIT);
3465         if (port->lun_map == NULL)
3466                 return (ENOMEM);
3467         for (i = 0; i < CTL_MAX_LUNS; i++)
3468                 port->lun_map[i] = UINT32_MAX;
3469         if (port->status & CTL_PORT_STATUS_ONLINE) {
3470                 if (port->lun_disable != NULL) {
3471                         STAILQ_FOREACH(lun, &softc->lun_list, links)
3472                                 port->lun_disable(port->targ_lun_arg, lun->lun);
3473                 }
3474                 ctl_isc_announce_port(port);
3475         }
3476         return (0);
3477 }
3478
3479 int
3480 ctl_lun_map_deinit(struct ctl_port *port)
3481 {
3482         struct ctl_softc *softc = port->ctl_softc;
3483         struct ctl_lun *lun;
3484
3485         if (port->lun_map == NULL)
3486                 return (0);
3487         free(port->lun_map, M_CTL);
3488         port->lun_map = NULL;
3489         if (port->status & CTL_PORT_STATUS_ONLINE) {
3490                 if (port->lun_enable != NULL) {
3491                         STAILQ_FOREACH(lun, &softc->lun_list, links)
3492                                 port->lun_enable(port->targ_lun_arg, lun->lun);
3493                 }
3494                 ctl_isc_announce_port(port);
3495         }
3496         return (0);
3497 }
3498
3499 int
3500 ctl_lun_map_set(struct ctl_port *port, uint32_t plun, uint32_t glun)
3501 {
3502         int status;
3503         uint32_t old;
3504
3505         if (port->lun_map == NULL) {
3506                 status = ctl_lun_map_init(port);
3507                 if (status != 0)
3508                         return (status);
3509         }
3510         old = port->lun_map[plun];
3511         port->lun_map[plun] = glun;
3512         if ((port->status & CTL_PORT_STATUS_ONLINE) && old >= CTL_MAX_LUNS) {
3513                 if (port->lun_enable != NULL)
3514                         port->lun_enable(port->targ_lun_arg, plun);
3515                 ctl_isc_announce_port(port);
3516         }
3517         return (0);
3518 }
3519
3520 int
3521 ctl_lun_map_unset(struct ctl_port *port, uint32_t plun)
3522 {
3523         uint32_t old;
3524
3525         if (port->lun_map == NULL)
3526                 return (0);
3527         old = port->lun_map[plun];
3528         port->lun_map[plun] = UINT32_MAX;
3529         if ((port->status & CTL_PORT_STATUS_ONLINE) && old < CTL_MAX_LUNS) {
3530                 if (port->lun_disable != NULL)
3531                         port->lun_disable(port->targ_lun_arg, plun);
3532                 ctl_isc_announce_port(port);
3533         }
3534         return (0);
3535 }
3536
3537 uint32_t
3538 ctl_lun_map_from_port(struct ctl_port *port, uint32_t lun_id)
3539 {
3540
3541         if (port == NULL)
3542                 return (UINT32_MAX);
3543         if (port->lun_map == NULL || lun_id >= CTL_MAX_LUNS)
3544                 return (lun_id);
3545         return (port->lun_map[lun_id]);
3546 }
3547
3548 uint32_t
3549 ctl_lun_map_to_port(struct ctl_port *port, uint32_t lun_id)
3550 {
3551         uint32_t i;
3552
3553         if (port == NULL)
3554                 return (UINT32_MAX);
3555         if (port->lun_map == NULL)
3556                 return (lun_id);
3557         for (i = 0; i < CTL_MAX_LUNS; i++) {
3558                 if (port->lun_map[i] == lun_id)
3559                         return (i);
3560         }
3561         return (UINT32_MAX);
3562 }
3563
3564 uint32_t
3565 ctl_decode_lun(uint64_t encoded)
3566 {
3567         uint8_t lun[8];
3568         uint32_t result = 0xffffffff;
3569
3570         be64enc(lun, encoded);
3571         switch (lun[0] & RPL_LUNDATA_ATYP_MASK) {
3572         case RPL_LUNDATA_ATYP_PERIPH:
3573                 if ((lun[0] & 0x3f) == 0 && lun[2] == 0 && lun[3] == 0 &&
3574                     lun[4] == 0 && lun[5] == 0 && lun[6] == 0 && lun[7] == 0)
3575                         result = lun[1];
3576                 break;
3577         case RPL_LUNDATA_ATYP_FLAT:
3578                 if (lun[2] == 0 && lun[3] == 0 && lun[4] == 0 && lun[5] == 0 &&
3579                     lun[6] == 0 && lun[7] == 0)
3580                         result = ((lun[0] & 0x3f) << 8) + lun[1];
3581                 break;
3582         case RPL_LUNDATA_ATYP_EXTLUN:
3583                 switch (lun[0] & RPL_LUNDATA_EXT_EAM_MASK) {
3584                 case 0x02:
3585                         switch (lun[0] & RPL_LUNDATA_EXT_LEN_MASK) {
3586                         case 0x00:
3587                                 result = lun[1];
3588                                 break;
3589                         case 0x10:
3590                                 result = (lun[1] << 16) + (lun[2] << 8) +
3591                                     lun[3];
3592                                 break;
3593                         case 0x20:
3594                                 if (lun[1] == 0 && lun[6] == 0 && lun[7] == 0)
3595                                         result = (lun[2] << 24) +
3596                                             (lun[3] << 16) + (lun[4] << 8) +
3597                                             lun[5];
3598                                 break;
3599                         }
3600                         break;
3601                 case RPL_LUNDATA_EXT_EAM_NOT_SPEC:
3602                         result = 0xffffffff;
3603                         break;
3604                 }
3605                 break;
3606         }
3607         return (result);
3608 }
3609
3610 uint64_t
3611 ctl_encode_lun(uint32_t decoded)
3612 {
3613         uint64_t l = decoded;
3614
3615         if (l <= 0xff)
3616                 return (((uint64_t)RPL_LUNDATA_ATYP_PERIPH << 56) | (l << 48));
3617         if (l <= 0x3fff)
3618                 return (((uint64_t)RPL_LUNDATA_ATYP_FLAT << 56) | (l << 48));
3619         if (l <= 0xffffff)
3620                 return (((uint64_t)(RPL_LUNDATA_ATYP_EXTLUN | 0x12) << 56) |
3621                     (l << 32));
3622         return ((((uint64_t)RPL_LUNDATA_ATYP_EXTLUN | 0x22) << 56) | (l << 16));
3623 }
3624
3625 static struct ctl_port *
3626 ctl_io_port(struct ctl_io_hdr *io_hdr)
3627 {
3628
3629         return (control_softc->ctl_ports[io_hdr->nexus.targ_port]);
3630 }
3631
3632 int
3633 ctl_ffz(uint32_t *mask, uint32_t first, uint32_t last)
3634 {
3635         int i;
3636
3637         for (i = first; i < last; i++) {
3638                 if ((mask[i / 32] & (1 << (i % 32))) == 0)
3639                         return (i);
3640         }
3641         return (-1);
3642 }
3643
3644 int
3645 ctl_set_mask(uint32_t *mask, uint32_t bit)
3646 {
3647         uint32_t chunk, piece;
3648
3649         chunk = bit >> 5;
3650         piece = bit % (sizeof(uint32_t) * 8);
3651
3652         if ((mask[chunk] & (1 << piece)) != 0)
3653                 return (-1);
3654         else
3655                 mask[chunk] |= (1 << piece);
3656
3657         return (0);
3658 }
3659
3660 int
3661 ctl_clear_mask(uint32_t *mask, uint32_t bit)
3662 {
3663         uint32_t chunk, piece;
3664
3665         chunk = bit >> 5;
3666         piece = bit % (sizeof(uint32_t) * 8);
3667
3668         if ((mask[chunk] & (1 << piece)) == 0)
3669                 return (-1);
3670         else
3671                 mask[chunk] &= ~(1 << piece);
3672
3673         return (0);
3674 }
3675
3676 int
3677 ctl_is_set(uint32_t *mask, uint32_t bit)
3678 {
3679         uint32_t chunk, piece;
3680
3681         chunk = bit >> 5;
3682         piece = bit % (sizeof(uint32_t) * 8);
3683
3684         if ((mask[chunk] & (1 << piece)) == 0)
3685                 return (0);
3686         else
3687                 return (1);
3688 }
3689
3690 static uint64_t
3691 ctl_get_prkey(struct ctl_lun *lun, uint32_t residx)
3692 {
3693         uint64_t *t;
3694
3695         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3696         if (t == NULL)
3697                 return (0);
3698         return (t[residx % CTL_MAX_INIT_PER_PORT]);
3699 }
3700
3701 static void
3702 ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx)
3703 {
3704         uint64_t *t;
3705
3706         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3707         if (t == NULL)
3708                 return;
3709         t[residx % CTL_MAX_INIT_PER_PORT] = 0;
3710 }
3711
3712 static void
3713 ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx)
3714 {
3715         uint64_t *p;
3716         u_int i;
3717
3718         i = residx/CTL_MAX_INIT_PER_PORT;
3719         if (lun->pr_keys[i] != NULL)
3720                 return;
3721         mtx_unlock(&lun->lun_lock);
3722         p = malloc(sizeof(uint64_t) * CTL_MAX_INIT_PER_PORT, M_CTL,
3723             M_WAITOK | M_ZERO);
3724         mtx_lock(&lun->lun_lock);
3725         if (lun->pr_keys[i] == NULL)
3726                 lun->pr_keys[i] = p;
3727         else
3728                 free(p, M_CTL);
3729 }
3730
3731 static void
3732 ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key)
3733 {
3734         uint64_t *t;
3735
3736         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3737         KASSERT(t != NULL, ("prkey %d is not allocated", residx));
3738         t[residx % CTL_MAX_INIT_PER_PORT] = key;
3739 }
3740
3741 /*
3742  * ctl_softc, pool_name, total_ctl_io are passed in.
3743  * npool is passed out.
3744  */
3745 int
3746 ctl_pool_create(struct ctl_softc *ctl_softc, const char *pool_name,
3747                 uint32_t total_ctl_io, void **npool)
3748 {
3749 #ifdef IO_POOLS
3750         struct ctl_io_pool *pool;
3751
3752         pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3753                                             M_NOWAIT | M_ZERO);
3754         if (pool == NULL)
3755                 return (ENOMEM);
3756
3757         snprintf(pool->name, sizeof(pool->name), "CTL IO %s", pool_name);
3758         pool->ctl_softc = ctl_softc;
3759         pool->zone = uma_zsecond_create(pool->name, NULL,
3760             NULL, NULL, NULL, ctl_softc->io_zone);
3761         /* uma_prealloc(pool->zone, total_ctl_io); */
3762
3763         *npool = pool;
3764 #else
3765         *npool = ctl_softc->io_zone;
3766 #endif
3767         return (0);
3768 }
3769
3770 void
3771 ctl_pool_free(struct ctl_io_pool *pool)
3772 {
3773
3774         if (pool == NULL)
3775                 return;
3776
3777 #ifdef IO_POOLS
3778         uma_zdestroy(pool->zone);
3779         free(pool, M_CTL);
3780 #endif
3781 }
3782
3783 union ctl_io *
3784 ctl_alloc_io(void *pool_ref)
3785 {
3786         union ctl_io *io;
3787 #ifdef IO_POOLS
3788         struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3789
3790         io = uma_zalloc(pool->zone, M_WAITOK);
3791 #else
3792         io = uma_zalloc((uma_zone_t)pool_ref, M_WAITOK);
3793 #endif
3794         if (io != NULL)
3795                 io->io_hdr.pool = pool_ref;
3796         return (io);
3797 }
3798
3799 union ctl_io *
3800 ctl_alloc_io_nowait(void *pool_ref)
3801 {
3802         union ctl_io *io;
3803 #ifdef IO_POOLS
3804         struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3805
3806         io = uma_zalloc(pool->zone, M_NOWAIT);
3807 #else
3808         io = uma_zalloc((uma_zone_t)pool_ref, M_NOWAIT);
3809 #endif
3810         if (io != NULL)
3811                 io->io_hdr.pool = pool_ref;
3812         return (io);
3813 }
3814
3815 void
3816 ctl_free_io(union ctl_io *io)
3817 {
3818 #ifdef IO_POOLS
3819         struct ctl_io_pool *pool;
3820 #endif
3821
3822         if (io == NULL)
3823                 return;
3824
3825 #ifdef IO_POOLS
3826         pool = (struct ctl_io_pool *)io->io_hdr.pool;
3827         uma_zfree(pool->zone, io);
3828 #else
3829         uma_zfree((uma_zone_t)io->io_hdr.pool, io);
3830 #endif
3831 }
3832
3833 void
3834 ctl_zero_io(union ctl_io *io)
3835 {
3836         void *pool_ref;
3837
3838         if (io == NULL)
3839                 return;
3840
3841         /*
3842          * May need to preserve linked list pointers at some point too.
3843          */
3844         pool_ref = io->io_hdr.pool;
3845         memset(io, 0, sizeof(*io));
3846         io->io_hdr.pool = pool_ref;
3847 }
3848
3849 int
3850 ctl_expand_number(const char *buf, uint64_t *num)
3851 {
3852         char *endptr;
3853         uint64_t number;
3854         unsigned shift;
3855
3856         number = strtoq(buf, &endptr, 0);
3857
3858         switch (tolower((unsigned char)*endptr)) {
3859         case 'e':
3860                 shift = 60;
3861                 break;
3862         case 'p':
3863                 shift = 50;
3864                 break;
3865         case 't':
3866                 shift = 40;
3867                 break;
3868         case 'g':
3869                 shift = 30;
3870                 break;
3871         case 'm':
3872                 shift = 20;
3873                 break;
3874         case 'k':
3875                 shift = 10;
3876                 break;
3877         case 'b':
3878         case '\0': /* No unit. */
3879                 *num = number;
3880                 return (0);
3881         default:
3882                 /* Unrecognized unit. */
3883                 return (-1);
3884         }
3885
3886         if ((number << shift) >> shift != number) {
3887                 /* Overflow */
3888                 return (-1);
3889         }
3890         *num = number << shift;
3891         return (0);
3892 }
3893
3894
3895 /*
3896  * This routine could be used in the future to load default and/or saved
3897  * mode page parameters for a particuar lun.
3898  */
3899 static int
3900 ctl_init_page_index(struct ctl_lun *lun)
3901 {
3902         int i, page_code;
3903         struct ctl_page_index *page_index;
3904         const char *value;
3905         uint64_t ival;
3906
3907         memcpy(&lun->mode_pages.index, page_index_template,
3908                sizeof(page_index_template));
3909
3910         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
3911
3912                 page_index = &lun->mode_pages.index[i];
3913                 if (lun->be_lun->lun_type == T_DIRECT &&
3914                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
3915                         continue;
3916                 if (lun->be_lun->lun_type == T_PROCESSOR &&
3917                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
3918                         continue;
3919                 if (lun->be_lun->lun_type == T_CDROM &&
3920                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
3921                         continue;
3922
3923                 page_code = page_index->page_code & SMPH_PC_MASK;
3924                 switch (page_code) {
3925                 case SMS_RW_ERROR_RECOVERY_PAGE: {
3926                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
3927                             ("subpage %#x for page %#x is incorrect!",
3928                             page_index->subpage, page_code));
3929                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT],
3930                                &rw_er_page_default,
3931                                sizeof(rw_er_page_default));
3932                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CHANGEABLE],
3933                                &rw_er_page_changeable,
3934                                sizeof(rw_er_page_changeable));
3935                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_DEFAULT],
3936                                &rw_er_page_default,
3937                                sizeof(rw_er_page_default));
3938                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_SAVED],
3939                                &rw_er_page_default,
3940                                sizeof(rw_er_page_default));
3941                         page_index->page_data =
3942                                 (uint8_t *)lun->mode_pages.rw_er_page;
3943                         break;
3944                 }
3945                 case SMS_FORMAT_DEVICE_PAGE: {
3946                         struct scsi_format_page *format_page;
3947
3948                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
3949                             ("subpage %#x for page %#x is incorrect!",
3950                             page_index->subpage, page_code));
3951
3952                         /*
3953                          * Sectors per track are set above.  Bytes per
3954                          * sector need to be set here on a per-LUN basis.
3955                          */
3956                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
3957                                &format_page_default,
3958                                sizeof(format_page_default));
3959                         memcpy(&lun->mode_pages.format_page[
3960                                CTL_PAGE_CHANGEABLE], &format_page_changeable,
3961                                sizeof(format_page_changeable));
3962                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
3963                                &format_page_default,
3964                                sizeof(format_page_default));
3965                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
3966                                &format_page_default,
3967                                sizeof(format_page_default));
3968
3969                         format_page = &lun->mode_pages.format_page[
3970                                 CTL_PAGE_CURRENT];
3971                         scsi_ulto2b(lun->be_lun->blocksize,
3972                                     format_page->bytes_per_sector);
3973
3974                         format_page = &lun->mode_pages.format_page[
3975                                 CTL_PAGE_DEFAULT];
3976                         scsi_ulto2b(lun->be_lun->blocksize,
3977                                     format_page->bytes_per_sector);
3978
3979                         format_page = &lun->mode_pages.format_page[
3980                                 CTL_PAGE_SAVED];
3981                         scsi_ulto2b(lun->be_lun->blocksize,
3982                                     format_page->bytes_per_sector);
3983
3984                         page_index->page_data =
3985                                 (uint8_t *)lun->mode_pages.format_page;
3986                         break;
3987                 }
3988                 case SMS_RIGID_DISK_PAGE: {
3989                         struct scsi_rigid_disk_page *rigid_disk_page;
3990                         uint32_t sectors_per_cylinder;
3991                         uint64_t cylinders;
3992 #ifndef __XSCALE__
3993                         int shift;
3994 #endif /* !__XSCALE__ */
3995
3996                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
3997                             ("subpage %#x for page %#x is incorrect!",
3998                             page_index->subpage, page_code));
3999
4000                         /*
4001                          * Rotation rate and sectors per track are set
4002                          * above.  We calculate the cylinders here based on
4003                          * capacity.  Due to the number of heads and
4004                          * sectors per track we're using, smaller arrays
4005                          * may turn out to have 0 cylinders.  Linux and
4006                          * FreeBSD don't pay attention to these mode pages
4007                          * to figure out capacity, but Solaris does.  It
4008                          * seems to deal with 0 cylinders just fine, and
4009                          * works out a fake geometry based on the capacity.
4010                          */
4011                         memcpy(&lun->mode_pages.rigid_disk_page[
4012                                CTL_PAGE_DEFAULT], &rigid_disk_page_default,
4013                                sizeof(rigid_disk_page_default));
4014                         memcpy(&lun->mode_pages.rigid_disk_page[
4015                                CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
4016                                sizeof(rigid_disk_page_changeable));
4017
4018                         sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
4019                                 CTL_DEFAULT_HEADS;
4020
4021                         /*
4022                          * The divide method here will be more accurate,
4023                          * probably, but results in floating point being
4024                          * used in the kernel on i386 (__udivdi3()).  On the
4025                          * XScale, though, __udivdi3() is implemented in
4026                          * software.
4027                          *
4028                          * The shift method for cylinder calculation is
4029                          * accurate if sectors_per_cylinder is a power of
4030                          * 2.  Otherwise it might be slightly off -- you
4031                          * might have a bit of a truncation problem.
4032                          */
4033 #ifdef  __XSCALE__
4034                         cylinders = (lun->be_lun->maxlba + 1) /
4035                                 sectors_per_cylinder;
4036 #else
4037                         for (shift = 31; shift > 0; shift--) {
4038                                 if (sectors_per_cylinder & (1 << shift))
4039                                         break;
4040                         }
4041                         cylinders = (lun->be_lun->maxlba + 1) >> shift;
4042 #endif
4043
4044                         /*
4045                          * We've basically got 3 bytes, or 24 bits for the
4046                          * cylinder size in the mode page.  If we're over,
4047                          * just round down to 2^24.
4048                          */
4049                         if (cylinders > 0xffffff)
4050                                 cylinders = 0xffffff;
4051
4052                         rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4053                                 CTL_PAGE_DEFAULT];
4054                         scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4055
4056                         if ((value = ctl_get_opt(&lun->be_lun->options,
4057                             "rpm")) != NULL) {
4058                                 scsi_ulto2b(strtol(value, NULL, 0),
4059                                      rigid_disk_page->rotation_rate);
4060                         }
4061
4062                         memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_CURRENT],
4063                                &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
4064                                sizeof(rigid_disk_page_default));
4065                         memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_SAVED],
4066                                &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
4067                                sizeof(rigid_disk_page_default));
4068
4069                         page_index->page_data =
4070                                 (uint8_t *)lun->mode_pages.rigid_disk_page;
4071                         break;
4072                 }
4073                 case SMS_VERIFY_ERROR_RECOVERY_PAGE: {
4074                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4075                             ("subpage %#x for page %#x is incorrect!",
4076                             page_index->subpage, page_code));
4077                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_CURRENT],
4078                                &verify_er_page_default,
4079                                sizeof(verify_er_page_default));
4080                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_CHANGEABLE],
4081                                &verify_er_page_changeable,
4082                                sizeof(verify_er_page_changeable));
4083                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_DEFAULT],
4084                                &verify_er_page_default,
4085                                sizeof(verify_er_page_default));
4086                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_SAVED],
4087                                &verify_er_page_default,
4088                                sizeof(verify_er_page_default));
4089                         page_index->page_data =
4090                                 (uint8_t *)lun->mode_pages.verify_er_page;
4091                         break;
4092                 }
4093                 case SMS_CACHING_PAGE: {
4094                         struct scsi_caching_page *caching_page;
4095
4096                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4097                             ("subpage %#x for page %#x is incorrect!",
4098                             page_index->subpage, page_code));
4099                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4100                                &caching_page_default,
4101                                sizeof(caching_page_default));
4102                         memcpy(&lun->mode_pages.caching_page[
4103                                CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4104                                sizeof(caching_page_changeable));
4105                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4106                                &caching_page_default,
4107                                sizeof(caching_page_default));
4108                         caching_page = &lun->mode_pages.caching_page[
4109                             CTL_PAGE_SAVED];
4110                         value = ctl_get_opt(&lun->be_lun->options, "writecache");
4111                         if (value != NULL && strcmp(value, "off") == 0)
4112                                 caching_page->flags1 &= ~SCP_WCE;
4113                         value = ctl_get_opt(&lun->be_lun->options, "readcache");
4114                         if (value != NULL && strcmp(value, "off") == 0)
4115                                 caching_page->flags1 |= SCP_RCD;
4116                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4117                                &lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4118                                sizeof(caching_page_default));
4119                         page_index->page_data =
4120                                 (uint8_t *)lun->mode_pages.caching_page;
4121                         break;
4122                 }
4123                 case SMS_CONTROL_MODE_PAGE: {
4124                         switch (page_index->subpage) {
4125                         case SMS_SUBPAGE_PAGE_0: {
4126                                 struct scsi_control_page *control_page;
4127
4128                                 memcpy(&lun->mode_pages.control_page[
4129                                     CTL_PAGE_DEFAULT],
4130                                        &control_page_default,
4131                                        sizeof(control_page_default));
4132                                 memcpy(&lun->mode_pages.control_page[
4133                                     CTL_PAGE_CHANGEABLE],
4134                                        &control_page_changeable,
4135                                        sizeof(control_page_changeable));
4136                                 memcpy(&lun->mode_pages.control_page[
4137                                     CTL_PAGE_SAVED],
4138                                        &control_page_default,
4139                                        sizeof(control_page_default));
4140                                 control_page = &lun->mode_pages.control_page[
4141                                     CTL_PAGE_SAVED];
4142                                 value = ctl_get_opt(&lun->be_lun->options,
4143                                     "reordering");
4144                                 if (value != NULL &&
4145                                     strcmp(value, "unrestricted") == 0) {
4146                                         control_page->queue_flags &=
4147                                             ~SCP_QUEUE_ALG_MASK;
4148                                         control_page->queue_flags |=
4149                                             SCP_QUEUE_ALG_UNRESTRICTED;
4150                                 }
4151                                 memcpy(&lun->mode_pages.control_page[
4152                                     CTL_PAGE_CURRENT],
4153                                        &lun->mode_pages.control_page[
4154                                     CTL_PAGE_SAVED],
4155                                        sizeof(control_page_default));
4156                                 page_index->page_data =
4157                                     (uint8_t *)lun->mode_pages.control_page;
4158                                 break;
4159                         }
4160                         case 0x01:
4161                                 memcpy(&lun->mode_pages.control_ext_page[
4162                                     CTL_PAGE_DEFAULT],
4163                                        &control_ext_page_default,
4164                                        sizeof(control_ext_page_default));
4165                                 memcpy(&lun->mode_pages.control_ext_page[
4166                                     CTL_PAGE_CHANGEABLE],
4167                                        &control_ext_page_changeable,
4168                                        sizeof(control_ext_page_changeable));
4169                                 memcpy(&lun->mode_pages.control_ext_page[
4170                                     CTL_PAGE_SAVED],
4171                                        &control_ext_page_default,
4172                                        sizeof(control_ext_page_default));
4173                                 memcpy(&lun->mode_pages.control_ext_page[
4174                                     CTL_PAGE_CURRENT],
4175                                        &lun->mode_pages.control_ext_page[
4176                                     CTL_PAGE_SAVED],
4177                                        sizeof(control_ext_page_default));
4178                                 page_index->page_data =
4179                                     (uint8_t *)lun->mode_pages.control_ext_page;
4180                                 break;
4181                         default:
4182                                 panic("subpage %#x for page %#x is incorrect!",
4183                                       page_index->subpage, page_code);
4184                         }
4185                         break;
4186                 }
4187                 case SMS_INFO_EXCEPTIONS_PAGE: {
4188                         switch (page_index->subpage) {
4189                         case SMS_SUBPAGE_PAGE_0:
4190                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_CURRENT],
4191                                        &ie_page_default,
4192                                        sizeof(ie_page_default));
4193                                 memcpy(&lun->mode_pages.ie_page[
4194                                        CTL_PAGE_CHANGEABLE], &ie_page_changeable,
4195                                        sizeof(ie_page_changeable));
4196                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_DEFAULT],
4197                                        &ie_page_default,
4198                                        sizeof(ie_page_default));
4199                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_SAVED],
4200                                        &ie_page_default,
4201                                        sizeof(ie_page_default));
4202                                 page_index->page_data =
4203                                         (uint8_t *)lun->mode_pages.ie_page;
4204                                 break;
4205                         case 0x02: {
4206                                 struct ctl_logical_block_provisioning_page *page;
4207
4208                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_DEFAULT],
4209                                        &lbp_page_default,
4210                                        sizeof(lbp_page_default));
4211                                 memcpy(&lun->mode_pages.lbp_page[
4212                                        CTL_PAGE_CHANGEABLE], &lbp_page_changeable,
4213                                        sizeof(lbp_page_changeable));
4214                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4215                                        &lbp_page_default,
4216                                        sizeof(lbp_page_default));
4217                                 page = &lun->mode_pages.lbp_page[CTL_PAGE_SAVED];
4218                                 value = ctl_get_opt(&lun->be_lun->options,
4219                                     "avail-threshold");
4220                                 if (value != NULL &&
4221                                     ctl_expand_number(value, &ival) == 0) {
4222                                         page->descr[0].flags |= SLBPPD_ENABLED |
4223                                             SLBPPD_ARMING_DEC;
4224                                         if (lun->be_lun->blocksize)
4225                                                 ival /= lun->be_lun->blocksize;
4226                                         else
4227                                                 ival /= 512;
4228                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4229                                             page->descr[0].count);
4230                                 }
4231                                 value = ctl_get_opt(&lun->be_lun->options,
4232                                     "used-threshold");
4233                                 if (value != NULL &&
4234                                     ctl_expand_number(value, &ival) == 0) {
4235                                         page->descr[1].flags |= SLBPPD_ENABLED |
4236                                             SLBPPD_ARMING_INC;
4237                                         if (lun->be_lun->blocksize)
4238                                                 ival /= lun->be_lun->blocksize;
4239                                         else
4240                                                 ival /= 512;
4241                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4242                                             page->descr[1].count);
4243                                 }
4244                                 value = ctl_get_opt(&lun->be_lun->options,
4245                                     "pool-avail-threshold");
4246                                 if (value != NULL &&
4247                                     ctl_expand_number(value, &ival) == 0) {
4248                                         page->descr[2].flags |= SLBPPD_ENABLED |
4249                                             SLBPPD_ARMING_DEC;
4250                                         if (lun->be_lun->blocksize)
4251                                                 ival /= lun->be_lun->blocksize;
4252                                         else
4253                                                 ival /= 512;
4254                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4255                                             page->descr[2].count);
4256                                 }
4257                                 value = ctl_get_opt(&lun->be_lun->options,
4258                                     "pool-used-threshold");
4259                                 if (value != NULL &&
4260                                     ctl_expand_number(value, &ival) == 0) {
4261                                         page->descr[3].flags |= SLBPPD_ENABLED |
4262                                             SLBPPD_ARMING_INC;
4263                                         if (lun->be_lun->blocksize)
4264                                                 ival /= lun->be_lun->blocksize;
4265                                         else
4266                                                 ival /= 512;
4267                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4268                                             page->descr[3].count);
4269                                 }
4270                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_CURRENT],
4271                                        &lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4272                                        sizeof(lbp_page_default));
4273                                 page_index->page_data =
4274                                         (uint8_t *)lun->mode_pages.lbp_page;
4275                                 break;
4276                         }
4277                         default:
4278                                 panic("subpage %#x for page %#x is incorrect!",
4279                                       page_index->subpage, page_code);
4280                         }
4281                         break;
4282                 }
4283                 case SMS_CDDVD_CAPS_PAGE:{
4284                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4285                             ("subpage %#x for page %#x is incorrect!",
4286                             page_index->subpage, page_code));
4287                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_DEFAULT],
4288                                &cddvd_page_default,
4289                                sizeof(cddvd_page_default));
4290                         memcpy(&lun->mode_pages.cddvd_page[
4291                                CTL_PAGE_CHANGEABLE], &cddvd_page_changeable,
4292                                sizeof(cddvd_page_changeable));
4293                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_SAVED],
4294                                &cddvd_page_default,
4295                                sizeof(cddvd_page_default));
4296                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_CURRENT],
4297                                &lun->mode_pages.cddvd_page[CTL_PAGE_SAVED],
4298                                sizeof(cddvd_page_default));
4299                         page_index->page_data =
4300                                 (uint8_t *)lun->mode_pages.cddvd_page;
4301                         break;
4302                 }
4303                 default:
4304                         panic("invalid page code value %#x", page_code);
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                 if (lun->be_lun->lun_type == T_DIRECT &&
4325                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
4326                         continue;
4327                 if (lun->be_lun->lun_type == T_PROCESSOR &&
4328                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
4329                         continue;
4330                 if (lun->be_lun->lun_type == T_CDROM &&
4331                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
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         lun->log_pages.index[4].page_data = (uint8_t *)&lun->log_pages.ie_page;
4356         lun->log_pages.index[4].page_len = sizeof(lun->log_pages.ie_page);
4357
4358         return (CTL_RETVAL_COMPLETE);
4359 }
4360
4361 static int
4362 hex2bin(const char *str, uint8_t *buf, int buf_size)
4363 {
4364         int i;
4365         u_char c;
4366
4367         memset(buf, 0, buf_size);
4368         while (isspace(str[0]))
4369                 str++;
4370         if (str[0] == '0' && (str[1] == 'x' || str[1] == 'X'))
4371                 str += 2;
4372         buf_size *= 2;
4373         for (i = 0; str[i] != 0 && i < buf_size; i++) {
4374                 c = str[i];
4375                 if (isdigit(c))
4376                         c -= '0';
4377                 else if (isalpha(c))
4378                         c -= isupper(c) ? 'A' - 10 : 'a' - 10;
4379                 else
4380                         break;
4381                 if (c >= 16)
4382                         break;
4383                 if ((i & 1) == 0)
4384                         buf[i / 2] |= (c << 4);
4385                 else
4386                         buf[i / 2] |= c;
4387         }
4388         return ((i + 1) / 2);
4389 }
4390
4391 /*
4392  * LUN allocation.
4393  *
4394  * Requirements:
4395  * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4396  *   wants us to allocate the LUN and he can block.
4397  * - ctl_softc is always set
4398  * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4399  *
4400  * Returns 0 for success, non-zero (errno) for failure.
4401  */
4402 static int
4403 ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4404               struct ctl_be_lun *const be_lun)
4405 {
4406         struct ctl_lun *nlun, *lun;
4407         struct scsi_vpd_id_descriptor *desc;
4408         struct scsi_vpd_id_t10 *t10id;
4409         const char *eui, *naa, *scsiname, *vendor, *value;
4410         int lun_number, i, lun_malloced;
4411         int devidlen, idlen1, idlen2 = 0, len;
4412
4413         if (be_lun == NULL)
4414                 return (EINVAL);
4415
4416         /*
4417          * We currently only support Direct Access or Processor LUN types.
4418          */
4419         switch (be_lun->lun_type) {
4420         case T_DIRECT:
4421         case T_PROCESSOR:
4422         case T_CDROM:
4423                 break;
4424         case T_SEQUENTIAL:
4425         case T_CHANGER:
4426         default:
4427                 be_lun->lun_config_status(be_lun->be_lun,
4428                                           CTL_LUN_CONFIG_FAILURE);
4429                 break;
4430         }
4431         if (ctl_lun == NULL) {
4432                 lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4433                 lun_malloced = 1;
4434         } else {
4435                 lun_malloced = 0;
4436                 lun = ctl_lun;
4437         }
4438
4439         memset(lun, 0, sizeof(*lun));
4440         if (lun_malloced)
4441                 lun->flags = CTL_LUN_MALLOCED;
4442
4443         /* Generate LUN ID. */
4444         devidlen = max(CTL_DEVID_MIN_LEN,
4445             strnlen(be_lun->device_id, CTL_DEVID_LEN));
4446         idlen1 = sizeof(*t10id) + devidlen;
4447         len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4448         scsiname = ctl_get_opt(&be_lun->options, "scsiname");
4449         if (scsiname != NULL) {
4450                 idlen2 = roundup2(strlen(scsiname) + 1, 4);
4451                 len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4452         }
4453         eui = ctl_get_opt(&be_lun->options, "eui");
4454         if (eui != NULL) {
4455                 len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4456         }
4457         naa = ctl_get_opt(&be_lun->options, "naa");
4458         if (naa != NULL) {
4459                 len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4460         }
4461         lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4462             M_CTL, M_WAITOK | M_ZERO);
4463         desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4464         desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4465         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4466         desc->length = idlen1;
4467         t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4468         memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4469         if ((vendor = ctl_get_opt(&be_lun->options, "vendor")) == NULL) {
4470                 strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4471         } else {
4472                 strncpy(t10id->vendor, vendor,
4473                     min(sizeof(t10id->vendor), strlen(vendor)));
4474         }
4475         strncpy((char *)t10id->vendor_spec_id,
4476             (char *)be_lun->device_id, devidlen);
4477         if (scsiname != NULL) {
4478                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4479                     desc->length);
4480                 desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4481                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4482                     SVPD_ID_TYPE_SCSI_NAME;
4483                 desc->length = idlen2;
4484                 strlcpy(desc->identifier, scsiname, idlen2);
4485         }
4486         if (eui != NULL) {
4487                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4488                     desc->length);
4489                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4490                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4491                     SVPD_ID_TYPE_EUI64;
4492                 desc->length = hex2bin(eui, desc->identifier, 16);
4493                 desc->length = desc->length > 12 ? 16 :
4494                     (desc->length > 8 ? 12 : 8);
4495                 len -= 16 - desc->length;
4496         }
4497         if (naa != NULL) {
4498                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4499                     desc->length);
4500                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4501                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4502                     SVPD_ID_TYPE_NAA;
4503                 desc->length = hex2bin(naa, desc->identifier, 16);
4504                 desc->length = desc->length > 8 ? 16 : 8;
4505                 len -= 16 - desc->length;
4506         }
4507         lun->lun_devid->len = len;
4508
4509         mtx_lock(&ctl_softc->ctl_lock);
4510         /*
4511          * See if the caller requested a particular LUN number.  If so, see
4512          * if it is available.  Otherwise, allocate the first available LUN.
4513          */
4514         if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4515                 if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4516                  || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4517                         mtx_unlock(&ctl_softc->ctl_lock);
4518                         if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4519                                 printf("ctl: requested LUN ID %d is higher "
4520                                        "than CTL_MAX_LUNS - 1 (%d)\n",
4521                                        be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4522                         } else {
4523                                 /*
4524                                  * XXX KDM return an error, or just assign
4525                                  * another LUN ID in this case??
4526                                  */
4527                                 printf("ctl: requested LUN ID %d is already "
4528                                        "in use\n", be_lun->req_lun_id);
4529                         }
4530                         if (lun->flags & CTL_LUN_MALLOCED)
4531                                 free(lun, M_CTL);
4532                         be_lun->lun_config_status(be_lun->be_lun,
4533                                                   CTL_LUN_CONFIG_FAILURE);
4534                         return (ENOSPC);
4535                 }
4536                 lun_number = be_lun->req_lun_id;
4537         } else {
4538                 lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, 0, CTL_MAX_LUNS);
4539                 if (lun_number == -1) {
4540                         mtx_unlock(&ctl_softc->ctl_lock);
4541                         printf("ctl: can't allocate LUN, out of LUNs\n");
4542                         if (lun->flags & CTL_LUN_MALLOCED)
4543                                 free(lun, M_CTL);
4544                         be_lun->lun_config_status(be_lun->be_lun,
4545                                                   CTL_LUN_CONFIG_FAILURE);
4546                         return (ENOSPC);
4547                 }
4548         }
4549         ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4550
4551         mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4552         lun->lun = lun_number;
4553         lun->be_lun = be_lun;
4554         /*
4555          * The processor LUN is always enabled.  Disk LUNs come on line
4556          * disabled, and must be enabled by the backend.
4557          */
4558         lun->flags |= CTL_LUN_DISABLED;
4559         lun->backend = be_lun->be;
4560         be_lun->ctl_lun = lun;
4561         be_lun->lun_id = lun_number;
4562         atomic_add_int(&be_lun->be->num_luns, 1);
4563         if (be_lun->flags & CTL_LUN_FLAG_EJECTED)
4564                 lun->flags |= CTL_LUN_EJECTED;
4565         if (be_lun->flags & CTL_LUN_FLAG_NO_MEDIA)
4566                 lun->flags |= CTL_LUN_NO_MEDIA;
4567         if (be_lun->flags & CTL_LUN_FLAG_STOPPED)
4568                 lun->flags |= CTL_LUN_STOPPED;
4569
4570         if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4571                 lun->flags |= CTL_LUN_PRIMARY_SC;
4572
4573         value = ctl_get_opt(&be_lun->options, "removable");
4574         if (value != NULL) {
4575                 if (strcmp(value, "on") == 0)
4576                         lun->flags |= CTL_LUN_REMOVABLE;
4577         } else if (be_lun->lun_type == T_CDROM)
4578                 lun->flags |= CTL_LUN_REMOVABLE;
4579
4580         lun->ctl_softc = ctl_softc;
4581 #ifdef CTL_TIME_IO
4582         lun->last_busy = getsbinuptime();
4583 #endif
4584         TAILQ_INIT(&lun->ooa_queue);
4585         TAILQ_INIT(&lun->blocked_queue);
4586         STAILQ_INIT(&lun->error_list);
4587         lun->ie_reported = 1;
4588         callout_init_mtx(&lun->ie_callout, &lun->lun_lock, 0);
4589         ctl_tpc_lun_init(lun);
4590
4591         /*
4592          * Initialize the mode and log page index.
4593          */
4594         ctl_init_page_index(lun);
4595         ctl_init_log_page_index(lun);
4596
4597         /*
4598          * Now, before we insert this lun on the lun list, set the lun
4599          * inventory changed UA for all other luns.
4600          */
4601         STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4602                 mtx_lock(&nlun->lun_lock);
4603                 ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4604                 mtx_unlock(&nlun->lun_lock);
4605         }
4606
4607         STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4608
4609         ctl_softc->ctl_luns[lun_number] = lun;
4610
4611         ctl_softc->num_luns++;
4612
4613         /* Setup statistics gathering */
4614         lun->stats.device_type = be_lun->lun_type;
4615         lun->stats.lun_number = lun_number;
4616         lun->stats.blocksize = be_lun->blocksize;
4617         if (be_lun->blocksize == 0)
4618                 lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4619         for (i = 0;i < CTL_MAX_PORTS;i++)
4620                 lun->stats.ports[i].targ_port = i;
4621
4622         mtx_unlock(&ctl_softc->ctl_lock);
4623
4624         lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4625         return (0);
4626 }
4627
4628 /*
4629  * Delete a LUN.
4630  * Assumptions:
4631  * - LUN has already been marked invalid and any pending I/O has been taken
4632  *   care of.
4633  */
4634 static int
4635 ctl_free_lun(struct ctl_lun *lun)
4636 {
4637         struct ctl_softc *softc;
4638         struct ctl_lun *nlun;
4639         int i;
4640
4641         softc = lun->ctl_softc;
4642
4643         mtx_assert(&softc->ctl_lock, MA_OWNED);
4644
4645         STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4646
4647         ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4648
4649         softc->ctl_luns[lun->lun] = NULL;
4650
4651         if (!TAILQ_EMPTY(&lun->ooa_queue))
4652                 panic("Freeing a LUN %p with outstanding I/O!!\n", lun);
4653
4654         softc->num_luns--;
4655
4656         /*
4657          * Tell the backend to free resources, if this LUN has a backend.
4658          */
4659         atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4660         lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4661
4662         lun->ie_reportcnt = UINT32_MAX;
4663         callout_drain(&lun->ie_callout);
4664
4665         ctl_tpc_lun_shutdown(lun);
4666         mtx_destroy(&lun->lun_lock);
4667         free(lun->lun_devid, M_CTL);
4668         for (i = 0; i < CTL_MAX_PORTS; i++)
4669                 free(lun->pending_ua[i], M_CTL);
4670         for (i = 0; i < CTL_MAX_PORTS; i++)
4671                 free(lun->pr_keys[i], M_CTL);
4672         free(lun->write_buffer, M_CTL);
4673         if (lun->flags & CTL_LUN_MALLOCED)
4674                 free(lun, M_CTL);
4675
4676         STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4677                 mtx_lock(&nlun->lun_lock);
4678                 ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4679                 mtx_unlock(&nlun->lun_lock);
4680         }
4681
4682         return (0);
4683 }
4684
4685 static void
4686 ctl_create_lun(struct ctl_be_lun *be_lun)
4687 {
4688
4689         /*
4690          * ctl_alloc_lun() should handle all potential failure cases.
4691          */
4692         ctl_alloc_lun(control_softc, NULL, be_lun);
4693 }
4694
4695 int
4696 ctl_add_lun(struct ctl_be_lun *be_lun)
4697 {
4698         struct ctl_softc *softc = control_softc;
4699
4700         mtx_lock(&softc->ctl_lock);
4701         STAILQ_INSERT_TAIL(&softc->pending_lun_queue, be_lun, links);
4702         mtx_unlock(&softc->ctl_lock);
4703         wakeup(&softc->pending_lun_queue);
4704
4705         return (0);
4706 }
4707
4708 int
4709 ctl_enable_lun(struct ctl_be_lun *be_lun)
4710 {
4711         struct ctl_softc *softc;
4712         struct ctl_port *port, *nport;
4713         struct ctl_lun *lun;
4714         int retval;
4715
4716         lun = (struct ctl_lun *)be_lun->ctl_lun;
4717         softc = lun->ctl_softc;
4718
4719         mtx_lock(&softc->ctl_lock);
4720         mtx_lock(&lun->lun_lock);
4721         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4722                 /*
4723                  * eh?  Why did we get called if the LUN is already
4724                  * enabled?
4725                  */
4726                 mtx_unlock(&lun->lun_lock);
4727                 mtx_unlock(&softc->ctl_lock);
4728                 return (0);
4729         }
4730         lun->flags &= ~CTL_LUN_DISABLED;
4731         mtx_unlock(&lun->lun_lock);
4732
4733         STAILQ_FOREACH_SAFE(port, &softc->port_list, links, nport) {
4734                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4735                     port->lun_map != NULL || port->lun_enable == NULL)
4736                         continue;
4737
4738                 /*
4739                  * Drop the lock while we call the FETD's enable routine.
4740                  * This can lead to a callback into CTL (at least in the
4741                  * case of the internal initiator frontend.
4742                  */
4743                 mtx_unlock(&softc->ctl_lock);
4744                 retval = port->lun_enable(port->targ_lun_arg, lun->lun);
4745                 mtx_lock(&softc->ctl_lock);
4746                 if (retval != 0) {
4747                         printf("%s: FETD %s port %d returned error "
4748                                "%d for lun_enable on lun %jd\n",
4749                                __func__, port->port_name, port->targ_port,
4750                                retval, (intmax_t)lun->lun);
4751                 }
4752         }
4753
4754         mtx_unlock(&softc->ctl_lock);
4755         ctl_isc_announce_lun(lun);
4756
4757         return (0);
4758 }
4759
4760 int
4761 ctl_disable_lun(struct ctl_be_lun *be_lun)
4762 {
4763         struct ctl_softc *softc;
4764         struct ctl_port *port;
4765         struct ctl_lun *lun;
4766         int retval;
4767
4768         lun = (struct ctl_lun *)be_lun->ctl_lun;
4769         softc = lun->ctl_softc;
4770
4771         mtx_lock(&softc->ctl_lock);
4772         mtx_lock(&lun->lun_lock);
4773         if (lun->flags & CTL_LUN_DISABLED) {
4774                 mtx_unlock(&lun->lun_lock);
4775                 mtx_unlock(&softc->ctl_lock);
4776                 return (0);
4777         }
4778         lun->flags |= CTL_LUN_DISABLED;
4779         mtx_unlock(&lun->lun_lock);
4780
4781         STAILQ_FOREACH(port, &softc->port_list, links) {
4782                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4783                     port->lun_map != NULL || port->lun_disable == NULL)
4784                         continue;
4785
4786                 /*
4787                  * Drop the lock before we call the frontend's disable
4788                  * routine, to avoid lock order reversals.
4789                  *
4790                  * XXX KDM what happens if the frontend list changes while
4791                  * we're traversing it?  It's unlikely, but should be handled.
4792                  */
4793                 mtx_unlock(&softc->ctl_lock);
4794                 retval = port->lun_disable(port->targ_lun_arg, lun->lun);
4795                 mtx_lock(&softc->ctl_lock);
4796                 if (retval != 0) {
4797                         printf("%s: FETD %s port %d returned error "
4798                                "%d for lun_disable on lun %jd\n",
4799                                __func__, port->port_name, port->targ_port,
4800                                retval, (intmax_t)lun->lun);
4801                 }
4802         }
4803
4804         mtx_unlock(&softc->ctl_lock);
4805         ctl_isc_announce_lun(lun);
4806
4807         return (0);
4808 }
4809
4810 int
4811 ctl_start_lun(struct ctl_be_lun *be_lun)
4812 {
4813         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4814
4815         mtx_lock(&lun->lun_lock);
4816         lun->flags &= ~CTL_LUN_STOPPED;
4817         mtx_unlock(&lun->lun_lock);
4818         return (0);
4819 }
4820
4821 int
4822 ctl_stop_lun(struct ctl_be_lun *be_lun)
4823 {
4824         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4825
4826         mtx_lock(&lun->lun_lock);
4827         lun->flags |= CTL_LUN_STOPPED;
4828         mtx_unlock(&lun->lun_lock);
4829         return (0);
4830 }
4831
4832 int
4833 ctl_lun_no_media(struct ctl_be_lun *be_lun)
4834 {
4835         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4836
4837         mtx_lock(&lun->lun_lock);
4838         lun->flags |= CTL_LUN_NO_MEDIA;
4839         mtx_unlock(&lun->lun_lock);
4840         return (0);
4841 }
4842
4843 int
4844 ctl_lun_has_media(struct ctl_be_lun *be_lun)
4845 {
4846         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4847         union ctl_ha_msg msg;
4848
4849         mtx_lock(&lun->lun_lock);
4850         lun->flags &= ~(CTL_LUN_NO_MEDIA | CTL_LUN_EJECTED);
4851         if (lun->flags & CTL_LUN_REMOVABLE)
4852                 ctl_est_ua_all(lun, -1, CTL_UA_MEDIUM_CHANGE);
4853         mtx_unlock(&lun->lun_lock);
4854         if ((lun->flags & CTL_LUN_REMOVABLE) &&
4855             lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
4856                 bzero(&msg.ua, sizeof(msg.ua));
4857                 msg.hdr.msg_type = CTL_MSG_UA;
4858                 msg.hdr.nexus.initid = -1;
4859                 msg.hdr.nexus.targ_port = -1;
4860                 msg.hdr.nexus.targ_lun = lun->lun;
4861                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
4862                 msg.ua.ua_all = 1;
4863                 msg.ua.ua_set = 1;
4864                 msg.ua.ua_type = CTL_UA_MEDIUM_CHANGE;
4865                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.ua),
4866                     M_WAITOK);
4867         }
4868         return (0);
4869 }
4870
4871 int
4872 ctl_lun_ejected(struct ctl_be_lun *be_lun)
4873 {
4874         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4875
4876         mtx_lock(&lun->lun_lock);
4877         lun->flags |= CTL_LUN_EJECTED;
4878         mtx_unlock(&lun->lun_lock);
4879         return (0);
4880 }
4881
4882 int
4883 ctl_lun_primary(struct ctl_be_lun *be_lun)
4884 {
4885         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4886
4887         mtx_lock(&lun->lun_lock);
4888         lun->flags |= CTL_LUN_PRIMARY_SC;
4889         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
4890         mtx_unlock(&lun->lun_lock);
4891         ctl_isc_announce_lun(lun);
4892         return (0);
4893 }
4894
4895 int
4896 ctl_lun_secondary(struct ctl_be_lun *be_lun)
4897 {
4898         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4899
4900         mtx_lock(&lun->lun_lock);
4901         lun->flags &= ~CTL_LUN_PRIMARY_SC;
4902         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
4903         mtx_unlock(&lun->lun_lock);
4904         ctl_isc_announce_lun(lun);
4905         return (0);
4906 }
4907
4908 int
4909 ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4910 {
4911         struct ctl_softc *softc;
4912         struct ctl_lun *lun;
4913
4914         lun = (struct ctl_lun *)be_lun->ctl_lun;
4915         softc = lun->ctl_softc;
4916
4917         mtx_lock(&lun->lun_lock);
4918
4919         /*
4920          * The LUN needs to be disabled before it can be marked invalid.
4921          */
4922         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4923                 mtx_unlock(&lun->lun_lock);
4924                 return (-1);
4925         }
4926         /*
4927          * Mark the LUN invalid.
4928          */
4929         lun->flags |= CTL_LUN_INVALID;
4930
4931         /*
4932          * If there is nothing in the OOA queue, go ahead and free the LUN.
4933          * If we have something in the OOA queue, we'll free it when the
4934          * last I/O completes.
4935          */
4936         if (TAILQ_EMPTY(&lun->ooa_queue)) {
4937                 mtx_unlock(&lun->lun_lock);
4938                 mtx_lock(&softc->ctl_lock);
4939                 ctl_free_lun(lun);
4940                 mtx_unlock(&softc->ctl_lock);
4941         } else
4942                 mtx_unlock(&lun->lun_lock);
4943
4944         return (0);
4945 }
4946
4947 void
4948 ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
4949 {
4950         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4951         union ctl_ha_msg msg;
4952
4953         mtx_lock(&lun->lun_lock);
4954         ctl_est_ua_all(lun, -1, CTL_UA_CAPACITY_CHANGE);
4955         mtx_unlock(&lun->lun_lock);
4956         if (lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
4957                 /* Send msg to other side. */
4958                 bzero(&msg.ua, sizeof(msg.ua));
4959                 msg.hdr.msg_type = CTL_MSG_UA;
4960                 msg.hdr.nexus.initid = -1;
4961                 msg.hdr.nexus.targ_port = -1;
4962                 msg.hdr.nexus.targ_lun = lun->lun;
4963                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
4964                 msg.ua.ua_all = 1;
4965                 msg.ua.ua_set = 1;
4966                 msg.ua.ua_type = CTL_UA_CAPACITY_CHANGE;
4967                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.ua),
4968                     M_WAITOK);
4969         }
4970 }
4971
4972 /*
4973  * Backend "memory move is complete" callback for requests that never
4974  * make it down to say RAIDCore's configuration code.
4975  */
4976 int
4977 ctl_config_move_done(union ctl_io *io)
4978 {
4979         int retval;
4980
4981         CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
4982         KASSERT(io->io_hdr.io_type == CTL_IO_SCSI,
4983             ("Config I/O type isn't CTL_IO_SCSI (%d)!", io->io_hdr.io_type));
4984
4985         if ((io->io_hdr.port_status != 0) &&
4986             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
4987              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
4988                 /*
4989                  * For hardware error sense keys, the sense key
4990                  * specific value is defined to be a retry count,
4991                  * but we use it to pass back an internal FETD
4992                  * error code.  XXX KDM  Hopefully the FETD is only
4993                  * using 16 bits for an error code, since that's
4994                  * all the space we have in the sks field.
4995                  */
4996                 ctl_set_internal_failure(&io->scsiio,
4997                                          /*sks_valid*/ 1,
4998                                          /*retry_count*/
4999                                          io->io_hdr.port_status);
5000         }
5001
5002         if (ctl_debug & CTL_DEBUG_CDB_DATA)
5003                 ctl_data_print(io);
5004         if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) ||
5005             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
5006              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) ||
5007             ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
5008                 /*
5009                  * XXX KDM just assuming a single pointer here, and not a
5010                  * S/G list.  If we start using S/G lists for config data,
5011                  * we'll need to know how to clean them up here as well.
5012                  */
5013                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5014                         free(io->scsiio.kern_data_ptr, M_CTL);
5015                 ctl_done(io);
5016                 retval = CTL_RETVAL_COMPLETE;
5017         } else {
5018                 /*
5019                  * XXX KDM now we need to continue data movement.  Some
5020                  * options:
5021                  * - call ctl_scsiio() again?  We don't do this for data
5022                  *   writes, because for those at least we know ahead of
5023                  *   time where the write will go and how long it is.  For
5024                  *   config writes, though, that information is largely
5025                  *   contained within the write itself, thus we need to
5026                  *   parse out the data again.
5027                  *
5028                  * - Call some other function once the data is in?
5029                  */
5030
5031                 /*
5032                  * XXX KDM call ctl_scsiio() again for now, and check flag
5033                  * bits to see whether we're allocated or not.
5034                  */
5035                 retval = ctl_scsiio(&io->scsiio);
5036         }
5037         return (retval);
5038 }
5039
5040 /*
5041  * This gets called by a backend driver when it is done with a
5042  * data_submit method.
5043  */
5044 void
5045 ctl_data_submit_done(union ctl_io *io)
5046 {
5047         /*
5048          * If the IO_CONT flag is set, we need to call the supplied
5049          * function to continue processing the I/O, instead of completing
5050          * the I/O just yet.
5051          *
5052          * If there is an error, though, we don't want to keep processing.
5053          * Instead, just send status back to the initiator.
5054          */
5055         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5056             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5057             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5058              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5059                 io->scsiio.io_cont(io);
5060                 return;
5061         }
5062         ctl_done(io);
5063 }
5064
5065 /*
5066  * This gets called by a backend driver when it is done with a
5067  * configuration write.
5068  */
5069 void
5070 ctl_config_write_done(union ctl_io *io)
5071 {
5072         uint8_t *buf;
5073
5074         /*
5075          * If the IO_CONT flag is set, we need to call the supplied
5076          * function to continue processing the I/O, instead of completing
5077          * the I/O just yet.
5078          *
5079          * If there is an error, though, we don't want to keep processing.
5080          * Instead, just send status back to the initiator.
5081          */
5082         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5083             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5084             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5085              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5086                 io->scsiio.io_cont(io);
5087                 return;
5088         }
5089         /*
5090          * Since a configuration write can be done for commands that actually
5091          * have data allocated, like write buffer, and commands that have
5092          * no data, like start/stop unit, we need to check here.
5093          */
5094         if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5095                 buf = io->scsiio.kern_data_ptr;
5096         else
5097                 buf = NULL;
5098         ctl_done(io);
5099         if (buf)
5100                 free(buf, M_CTL);
5101 }
5102
5103 void
5104 ctl_config_read_done(union ctl_io *io)
5105 {
5106         uint8_t *buf;
5107
5108         /*
5109          * If there is some error -- we are done, skip data transfer.
5110          */
5111         if ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0 ||
5112             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
5113              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
5114                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5115                         buf = io->scsiio.kern_data_ptr;
5116                 else
5117                         buf = NULL;
5118                 ctl_done(io);
5119                 if (buf)
5120                         free(buf, M_CTL);
5121                 return;
5122         }
5123
5124         /*
5125          * If the IO_CONT flag is set, we need to call the supplied
5126          * function to continue processing the I/O, instead of completing
5127          * the I/O just yet.
5128          */
5129         if (io->io_hdr.flags & CTL_FLAG_IO_CONT) {
5130                 io->scsiio.io_cont(io);
5131                 return;
5132         }
5133
5134         ctl_datamove(io);
5135 }
5136
5137 /*
5138  * SCSI release command.
5139  */
5140 int
5141 ctl_scsi_release(struct ctl_scsiio *ctsio)
5142 {
5143         struct ctl_lun *lun;
5144         uint32_t residx;
5145
5146         CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5147
5148         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5149         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5150
5151         /*
5152          * XXX KDM right now, we only support LUN reservation.  We don't
5153          * support 3rd party reservations, or extent reservations, which
5154          * might actually need the parameter list.  If we've gotten this
5155          * far, we've got a LUN reservation.  Anything else got kicked out
5156          * above.  So, according to SPC, ignore the length.
5157          */
5158
5159         mtx_lock(&lun->lun_lock);
5160
5161         /*
5162          * According to SPC, it is not an error for an intiator to attempt
5163          * to release a reservation on a LUN that isn't reserved, or that
5164          * is reserved by another initiator.  The reservation can only be
5165          * released, though, by the initiator who made it or by one of
5166          * several reset type events.
5167          */
5168         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
5169                         lun->flags &= ~CTL_LUN_RESERVED;
5170
5171         mtx_unlock(&lun->lun_lock);
5172
5173         ctl_set_success(ctsio);
5174         ctl_done((union ctl_io *)ctsio);
5175         return (CTL_RETVAL_COMPLETE);
5176 }
5177
5178 int
5179 ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5180 {
5181         struct ctl_lun *lun;
5182         uint32_t residx;
5183
5184         CTL_DEBUG_PRINT(("ctl_reserve\n"));
5185
5186         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5187         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5188
5189         /*
5190          * XXX KDM right now, we only support LUN reservation.  We don't
5191          * support 3rd party reservations, or extent reservations, which
5192          * might actually need the parameter list.  If we've gotten this
5193          * far, we've got a LUN reservation.  Anything else got kicked out
5194          * above.  So, according to SPC, ignore the length.
5195          */
5196
5197         mtx_lock(&lun->lun_lock);
5198         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx != residx)) {
5199                 ctl_set_reservation_conflict(ctsio);
5200                 goto bailout;
5201         }
5202
5203         /* SPC-3 exceptions to SPC-2 RESERVE and RELEASE behavior. */
5204         if (lun->flags & CTL_LUN_PR_RESERVED) {
5205                 ctl_set_success(ctsio);
5206                 goto bailout;
5207         }
5208
5209         lun->flags |= CTL_LUN_RESERVED;
5210         lun->res_idx = residx;
5211         ctl_set_success(ctsio);
5212
5213 bailout:
5214         mtx_unlock(&lun->lun_lock);
5215         ctl_done((union ctl_io *)ctsio);
5216         return (CTL_RETVAL_COMPLETE);
5217 }
5218
5219 int
5220 ctl_start_stop(struct ctl_scsiio *ctsio)
5221 {
5222         struct scsi_start_stop_unit *cdb;
5223         struct ctl_lun *lun;
5224         int retval;
5225
5226         CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5227
5228         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5229         cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5230
5231         if ((cdb->how & SSS_PC_MASK) == 0) {
5232                 if ((lun->flags & CTL_LUN_PR_RESERVED) &&
5233                     (cdb->how & SSS_START) == 0) {
5234                         uint32_t residx;
5235
5236                         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5237                         if (ctl_get_prkey(lun, residx) == 0 ||
5238                             (lun->pr_res_idx != residx && lun->pr_res_type < 4)) {
5239
5240                                 ctl_set_reservation_conflict(ctsio);
5241                                 ctl_done((union ctl_io *)ctsio);
5242                                 return (CTL_RETVAL_COMPLETE);
5243                         }
5244                 }
5245
5246                 if ((cdb->how & SSS_LOEJ) &&
5247                     (lun->flags & CTL_LUN_REMOVABLE) == 0) {
5248                         ctl_set_invalid_field(ctsio,
5249                                               /*sks_valid*/ 1,
5250                                               /*command*/ 1,
5251                                               /*field*/ 4,
5252                                               /*bit_valid*/ 1,
5253                                               /*bit*/ 1);
5254                         ctl_done((union ctl_io *)ctsio);
5255                         return (CTL_RETVAL_COMPLETE);
5256                 }
5257
5258                 if ((cdb->how & SSS_START) == 0 && (cdb->how & SSS_LOEJ) &&
5259                     lun->prevent_count > 0) {
5260                         /* "Medium removal prevented" */
5261                         ctl_set_sense(ctsio, /*current_error*/ 1,
5262                             /*sense_key*/(lun->flags & CTL_LUN_NO_MEDIA) ?
5263                              SSD_KEY_NOT_READY : SSD_KEY_ILLEGAL_REQUEST,
5264                             /*asc*/ 0x53, /*ascq*/ 0x02, SSD_ELEM_NONE);
5265                         ctl_done((union ctl_io *)ctsio);
5266                         return (CTL_RETVAL_COMPLETE);
5267                 }
5268         }
5269
5270         retval = lun->backend->config_write((union ctl_io *)ctsio);
5271         return (retval);
5272 }
5273
5274 int
5275 ctl_prevent_allow(struct ctl_scsiio *ctsio)
5276 {
5277         struct ctl_lun *lun;
5278         struct scsi_prevent *cdb;
5279         int retval;
5280         uint32_t initidx;
5281
5282         CTL_DEBUG_PRINT(("ctl_prevent_allow\n"));
5283
5284         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5285         cdb = (struct scsi_prevent *)ctsio->cdb;
5286
5287         if ((lun->flags & CTL_LUN_REMOVABLE) == 0) {
5288                 ctl_set_invalid_opcode(ctsio);
5289                 ctl_done((union ctl_io *)ctsio);
5290                 return (CTL_RETVAL_COMPLETE);
5291         }
5292
5293         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5294         mtx_lock(&lun->lun_lock);
5295         if ((cdb->how & PR_PREVENT) &&
5296             ctl_is_set(lun->prevent, initidx) == 0) {
5297                 ctl_set_mask(lun->prevent, initidx);
5298                 lun->prevent_count++;
5299         } else if ((cdb->how & PR_PREVENT) == 0 &&
5300             ctl_is_set(lun->prevent, initidx)) {
5301                 ctl_clear_mask(lun->prevent, initidx);
5302                 lun->prevent_count--;
5303         }
5304         mtx_unlock(&lun->lun_lock);
5305         retval = lun->backend->config_write((union ctl_io *)ctsio);
5306         return (retval);
5307 }
5308
5309 /*
5310  * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5311  * we don't really do anything with the LBA and length fields if the user
5312  * passes them in.  Instead we'll just flush out the cache for the entire
5313  * LUN.
5314  */
5315 int
5316 ctl_sync_cache(struct ctl_scsiio *ctsio)
5317 {
5318         struct ctl_lun *lun;
5319         struct ctl_softc *softc;
5320         struct ctl_lba_len_flags *lbalen;
5321         uint64_t starting_lba;
5322         uint32_t block_count;
5323         int retval;
5324         uint8_t byte2;
5325
5326         CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5327
5328         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5329         softc = lun->ctl_softc;
5330         retval = 0;
5331
5332         switch (ctsio->cdb[0]) {
5333         case SYNCHRONIZE_CACHE: {
5334                 struct scsi_sync_cache *cdb;
5335                 cdb = (struct scsi_sync_cache *)ctsio->cdb;
5336
5337                 starting_lba = scsi_4btoul(cdb->begin_lba);
5338                 block_count = scsi_2btoul(cdb->lb_count);
5339                 byte2 = cdb->byte2;
5340                 break;
5341         }
5342         case SYNCHRONIZE_CACHE_16: {
5343                 struct scsi_sync_cache_16 *cdb;
5344                 cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5345
5346                 starting_lba = scsi_8btou64(cdb->begin_lba);
5347                 block_count = scsi_4btoul(cdb->lb_count);
5348                 byte2 = cdb->byte2;
5349                 break;
5350         }
5351         default:
5352                 ctl_set_invalid_opcode(ctsio);
5353                 ctl_done((union ctl_io *)ctsio);
5354                 goto bailout;
5355                 break; /* NOTREACHED */
5356         }
5357
5358         /*
5359          * We check the LBA and length, but don't do anything with them.
5360          * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5361          * get flushed.  This check will just help satisfy anyone who wants
5362          * to see an error for an out of range LBA.
5363          */
5364         if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5365                 ctl_set_lba_out_of_range(ctsio);
5366                 ctl_done((union ctl_io *)ctsio);
5367                 goto bailout;
5368         }
5369
5370         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5371         lbalen->lba = starting_lba;
5372         lbalen->len = block_count;
5373         lbalen->flags = byte2;
5374         retval = lun->backend->config_write((union ctl_io *)ctsio);
5375
5376 bailout:
5377         return (retval);
5378 }
5379
5380 int
5381 ctl_format(struct ctl_scsiio *ctsio)
5382 {
5383         struct scsi_format *cdb;
5384         struct ctl_lun *lun;
5385         int length, defect_list_len;
5386
5387         CTL_DEBUG_PRINT(("ctl_format\n"));
5388
5389         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5390
5391         cdb = (struct scsi_format *)ctsio->cdb;
5392
5393         length = 0;
5394         if (cdb->byte2 & SF_FMTDATA) {
5395                 if (cdb->byte2 & SF_LONGLIST)
5396                         length = sizeof(struct scsi_format_header_long);
5397                 else
5398                         length = sizeof(struct scsi_format_header_short);
5399         }
5400
5401         if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5402          && (length > 0)) {
5403                 ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5404                 ctsio->kern_data_len = length;
5405                 ctsio->kern_total_len = length;
5406                 ctsio->kern_data_resid = 0;
5407                 ctsio->kern_rel_offset = 0;
5408                 ctsio->kern_sg_entries = 0;
5409                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5410                 ctsio->be_move_done = ctl_config_move_done;
5411                 ctl_datamove((union ctl_io *)ctsio);
5412
5413                 return (CTL_RETVAL_COMPLETE);
5414         }
5415
5416         defect_list_len = 0;
5417
5418         if (cdb->byte2 & SF_FMTDATA) {
5419                 if (cdb->byte2 & SF_LONGLIST) {
5420                         struct scsi_format_header_long *header;
5421
5422                         header = (struct scsi_format_header_long *)
5423                                 ctsio->kern_data_ptr;
5424
5425                         defect_list_len = scsi_4btoul(header->defect_list_len);
5426                         if (defect_list_len != 0) {
5427                                 ctl_set_invalid_field(ctsio,
5428                                                       /*sks_valid*/ 1,
5429                                                       /*command*/ 0,
5430                                                       /*field*/ 2,
5431                                                       /*bit_valid*/ 0,
5432                                                       /*bit*/ 0);
5433                                 goto bailout;
5434                         }
5435                 } else {
5436                         struct scsi_format_header_short *header;
5437
5438                         header = (struct scsi_format_header_short *)
5439                                 ctsio->kern_data_ptr;
5440
5441                         defect_list_len = scsi_2btoul(header->defect_list_len);
5442                         if (defect_list_len != 0) {
5443                                 ctl_set_invalid_field(ctsio,
5444                                                       /*sks_valid*/ 1,
5445                                                       /*command*/ 0,
5446                                                       /*field*/ 2,
5447                                                       /*bit_valid*/ 0,
5448                                                       /*bit*/ 0);
5449                                 goto bailout;
5450                         }
5451                 }
5452         }
5453
5454         ctl_set_success(ctsio);
5455 bailout:
5456
5457         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5458                 free(ctsio->kern_data_ptr, M_CTL);
5459                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5460         }
5461
5462         ctl_done((union ctl_io *)ctsio);
5463         return (CTL_RETVAL_COMPLETE);
5464 }
5465
5466 int
5467 ctl_read_buffer(struct ctl_scsiio *ctsio)
5468 {
5469         struct ctl_lun *lun;
5470         uint64_t buffer_offset;
5471         uint32_t len;
5472         uint8_t byte2;
5473         static uint8_t descr[4];
5474         static uint8_t echo_descr[4] = { 0 };
5475
5476         CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5477         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5478         switch (ctsio->cdb[0]) {
5479         case READ_BUFFER: {
5480                 struct scsi_read_buffer *cdb;
5481
5482                 cdb = (struct scsi_read_buffer *)ctsio->cdb;
5483                 buffer_offset = scsi_3btoul(cdb->offset);
5484                 len = scsi_3btoul(cdb->length);
5485                 byte2 = cdb->byte2;
5486                 break;
5487         }
5488         case READ_BUFFER_16: {
5489                 struct scsi_read_buffer_16 *cdb;
5490
5491                 cdb = (struct scsi_read_buffer_16 *)ctsio->cdb;
5492                 buffer_offset = scsi_8btou64(cdb->offset);
5493                 len = scsi_4btoul(cdb->length);
5494                 byte2 = cdb->byte2;
5495                 break;
5496         }
5497         default: /* This shouldn't happen. */
5498                 ctl_set_invalid_opcode(ctsio);
5499                 ctl_done((union ctl_io *)ctsio);
5500                 return (CTL_RETVAL_COMPLETE);
5501         }
5502
5503         if ((byte2 & RWB_MODE) != RWB_MODE_DATA &&
5504             (byte2 & RWB_MODE) != RWB_MODE_ECHO_DESCR &&
5505             (byte2 & RWB_MODE) != RWB_MODE_DESCR) {
5506                 ctl_set_invalid_field(ctsio,
5507                                       /*sks_valid*/ 1,
5508                                       /*command*/ 1,
5509                                       /*field*/ 1,
5510                                       /*bit_valid*/ 1,
5511                                       /*bit*/ 4);
5512                 ctl_done((union ctl_io *)ctsio);
5513                 return (CTL_RETVAL_COMPLETE);
5514         }
5515
5516         if (buffer_offset > CTL_WRITE_BUFFER_SIZE ||
5517             buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5518                 ctl_set_invalid_field(ctsio,
5519                                       /*sks_valid*/ 1,
5520                                       /*command*/ 1,
5521                                       /*field*/ 6,
5522                                       /*bit_valid*/ 0,
5523                                       /*bit*/ 0);
5524                 ctl_done((union ctl_io *)ctsio);
5525                 return (CTL_RETVAL_COMPLETE);
5526         }
5527
5528         if ((byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5529                 descr[0] = 0;
5530                 scsi_ulto3b(CTL_WRITE_BUFFER_SIZE, &descr[1]);
5531                 ctsio->kern_data_ptr = descr;
5532                 len = min(len, sizeof(descr));
5533         } else if ((byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5534                 ctsio->kern_data_ptr = echo_descr;
5535                 len = min(len, sizeof(echo_descr));
5536         } else {
5537                 if (lun->write_buffer == NULL) {
5538                         lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5539                             M_CTL, M_WAITOK);
5540                 }
5541                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5542         }
5543         ctsio->kern_data_len = len;
5544         ctsio->kern_total_len = len;
5545         ctsio->kern_data_resid = 0;
5546         ctsio->kern_rel_offset = 0;
5547         ctsio->kern_sg_entries = 0;
5548         ctl_set_success(ctsio);
5549         ctsio->be_move_done = ctl_config_move_done;
5550         ctl_datamove((union ctl_io *)ctsio);
5551         return (CTL_RETVAL_COMPLETE);
5552 }
5553
5554 int
5555 ctl_write_buffer(struct ctl_scsiio *ctsio)
5556 {
5557         struct scsi_write_buffer *cdb;
5558         struct ctl_lun *lun;
5559         int buffer_offset, len;
5560
5561         CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5562
5563         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5564         cdb = (struct scsi_write_buffer *)ctsio->cdb;
5565
5566         if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5567                 ctl_set_invalid_field(ctsio,
5568                                       /*sks_valid*/ 1,
5569                                       /*command*/ 1,
5570                                       /*field*/ 1,
5571                                       /*bit_valid*/ 1,
5572                                       /*bit*/ 4);
5573                 ctl_done((union ctl_io *)ctsio);
5574                 return (CTL_RETVAL_COMPLETE);
5575         }
5576
5577         len = scsi_3btoul(cdb->length);
5578         buffer_offset = scsi_3btoul(cdb->offset);
5579
5580         if (buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5581                 ctl_set_invalid_field(ctsio,
5582                                       /*sks_valid*/ 1,
5583                                       /*command*/ 1,
5584                                       /*field*/ 6,
5585                                       /*bit_valid*/ 0,
5586                                       /*bit*/ 0);
5587                 ctl_done((union ctl_io *)ctsio);
5588                 return (CTL_RETVAL_COMPLETE);
5589         }
5590
5591         /*
5592          * If we've got a kernel request that hasn't been malloced yet,
5593          * malloc it and tell the caller the data buffer is here.
5594          */
5595         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5596                 if (lun->write_buffer == NULL) {
5597                         lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5598                             M_CTL, M_WAITOK);
5599                 }
5600                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5601                 ctsio->kern_data_len = len;
5602                 ctsio->kern_total_len = len;
5603                 ctsio->kern_data_resid = 0;
5604                 ctsio->kern_rel_offset = 0;
5605                 ctsio->kern_sg_entries = 0;
5606                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5607                 ctsio->be_move_done = ctl_config_move_done;
5608                 ctl_datamove((union ctl_io *)ctsio);
5609
5610                 return (CTL_RETVAL_COMPLETE);
5611         }
5612
5613         ctl_set_success(ctsio);
5614         ctl_done((union ctl_io *)ctsio);
5615         return (CTL_RETVAL_COMPLETE);
5616 }
5617
5618 int
5619 ctl_write_same(struct ctl_scsiio *ctsio)
5620 {
5621         struct ctl_lun *lun;
5622         struct ctl_lba_len_flags *lbalen;
5623         uint64_t lba;
5624         uint32_t num_blocks;
5625         int len, retval;
5626         uint8_t byte2;
5627
5628         CTL_DEBUG_PRINT(("ctl_write_same\n"));
5629
5630         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5631
5632         switch (ctsio->cdb[0]) {
5633         case WRITE_SAME_10: {
5634                 struct scsi_write_same_10 *cdb;
5635
5636                 cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5637
5638                 lba = scsi_4btoul(cdb->addr);
5639                 num_blocks = scsi_2btoul(cdb->length);
5640                 byte2 = cdb->byte2;
5641                 break;
5642         }
5643         case WRITE_SAME_16: {
5644                 struct scsi_write_same_16 *cdb;
5645
5646                 cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5647
5648                 lba = scsi_8btou64(cdb->addr);
5649                 num_blocks = scsi_4btoul(cdb->length);
5650                 byte2 = cdb->byte2;
5651                 break;
5652         }
5653         default:
5654                 /*
5655                  * We got a command we don't support.  This shouldn't
5656                  * happen, commands should be filtered out above us.
5657                  */
5658                 ctl_set_invalid_opcode(ctsio);
5659                 ctl_done((union ctl_io *)ctsio);
5660
5661                 return (CTL_RETVAL_COMPLETE);
5662                 break; /* NOTREACHED */
5663         }
5664
5665         /* ANCHOR flag can be used only together with UNMAP */
5666         if ((byte2 & SWS_UNMAP) == 0 && (byte2 & SWS_ANCHOR) != 0) {
5667                 ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
5668                     /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
5669                 ctl_done((union ctl_io *)ctsio);
5670                 return (CTL_RETVAL_COMPLETE);
5671         }
5672
5673         /*
5674          * The first check is to make sure we're in bounds, the second
5675          * check is to catch wrap-around problems.  If the lba + num blocks
5676          * is less than the lba, then we've wrapped around and the block
5677          * range is invalid anyway.
5678          */
5679         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5680          || ((lba + num_blocks) < lba)) {
5681                 ctl_set_lba_out_of_range(ctsio);
5682                 ctl_done((union ctl_io *)ctsio);
5683                 return (CTL_RETVAL_COMPLETE);
5684         }
5685
5686         /* Zero number of blocks means "to the last logical block" */
5687         if (num_blocks == 0) {
5688                 if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5689                         ctl_set_invalid_field(ctsio,
5690                                               /*sks_valid*/ 0,
5691                                               /*command*/ 1,
5692                                               /*field*/ 0,
5693                                               /*bit_valid*/ 0,
5694                                               /*bit*/ 0);
5695                         ctl_done((union ctl_io *)ctsio);
5696                         return (CTL_RETVAL_COMPLETE);
5697                 }
5698                 num_blocks = (lun->be_lun->maxlba + 1) - lba;
5699         }
5700
5701         len = lun->be_lun->blocksize;
5702
5703         /*
5704          * If we've got a kernel request that hasn't been malloced yet,
5705          * malloc it and tell the caller the data buffer is here.
5706          */
5707         if ((byte2 & SWS_NDOB) == 0 &&
5708             (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5709                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);
5710                 ctsio->kern_data_len = len;
5711                 ctsio->kern_total_len = len;
5712                 ctsio->kern_data_resid = 0;
5713                 ctsio->kern_rel_offset = 0;
5714                 ctsio->kern_sg_entries = 0;
5715                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5716                 ctsio->be_move_done = ctl_config_move_done;
5717                 ctl_datamove((union ctl_io *)ctsio);
5718
5719                 return (CTL_RETVAL_COMPLETE);
5720         }
5721
5722         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5723         lbalen->lba = lba;
5724         lbalen->len = num_blocks;
5725         lbalen->flags = byte2;
5726         retval = lun->backend->config_write((union ctl_io *)ctsio);
5727
5728         return (retval);
5729 }
5730
5731 int
5732 ctl_unmap(struct ctl_scsiio *ctsio)
5733 {
5734         struct ctl_lun *lun;
5735         struct scsi_unmap *cdb;
5736         struct ctl_ptr_len_flags *ptrlen;
5737         struct scsi_unmap_header *hdr;
5738         struct scsi_unmap_desc *buf, *end, *endnz, *range;
5739         uint64_t lba;
5740         uint32_t num_blocks;
5741         int len, retval;
5742         uint8_t byte2;
5743
5744         CTL_DEBUG_PRINT(("ctl_unmap\n"));
5745
5746         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5747         cdb = (struct scsi_unmap *)ctsio->cdb;
5748
5749         len = scsi_2btoul(cdb->length);
5750         byte2 = cdb->byte2;
5751
5752         /*
5753          * If we've got a kernel request that hasn't been malloced yet,
5754          * malloc it and tell the caller the data buffer is here.
5755          */
5756         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5757                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);
5758                 ctsio->kern_data_len = len;
5759                 ctsio->kern_total_len = len;
5760                 ctsio->kern_data_resid = 0;
5761                 ctsio->kern_rel_offset = 0;
5762                 ctsio->kern_sg_entries = 0;
5763                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5764                 ctsio->be_move_done = ctl_config_move_done;
5765                 ctl_datamove((union ctl_io *)ctsio);
5766
5767                 return (CTL_RETVAL_COMPLETE);
5768         }
5769
5770         len = ctsio->kern_total_len - ctsio->kern_data_resid;
5771         hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
5772         if (len < sizeof (*hdr) ||
5773             len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
5774             len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
5775             scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
5776                 ctl_set_invalid_field(ctsio,
5777                                       /*sks_valid*/ 0,
5778                                       /*command*/ 0,
5779                                       /*field*/ 0,
5780                                       /*bit_valid*/ 0,
5781                                       /*bit*/ 0);
5782                 goto done;
5783         }
5784         len = scsi_2btoul(hdr->desc_length);
5785         buf = (struct scsi_unmap_desc *)(hdr + 1);
5786         end = buf + len / sizeof(*buf);
5787
5788         endnz = buf;
5789         for (range = buf; range < end; range++) {
5790                 lba = scsi_8btou64(range->lba);
5791                 num_blocks = scsi_4btoul(range->length);
5792                 if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5793                  || ((lba + num_blocks) < lba)) {
5794                         ctl_set_lba_out_of_range(ctsio);
5795                         ctl_done((union ctl_io *)ctsio);
5796                         return (CTL_RETVAL_COMPLETE);
5797                 }
5798                 if (num_blocks != 0)
5799                         endnz = range + 1;
5800         }
5801
5802         /*
5803          * Block backend can not handle zero last range.
5804          * Filter it out and return if there is nothing left.
5805          */
5806         len = (uint8_t *)endnz - (uint8_t *)buf;
5807         if (len == 0) {
5808                 ctl_set_success(ctsio);
5809                 goto done;
5810         }
5811
5812         mtx_lock(&lun->lun_lock);
5813         ptrlen = (struct ctl_ptr_len_flags *)
5814             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5815         ptrlen->ptr = (void *)buf;
5816         ptrlen->len = len;
5817         ptrlen->flags = byte2;
5818         ctl_check_blocked(lun);
5819         mtx_unlock(&lun->lun_lock);
5820
5821         retval = lun->backend->config_write((union ctl_io *)ctsio);
5822         return (retval);
5823
5824 done:
5825         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5826                 free(ctsio->kern_data_ptr, M_CTL);
5827                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5828         }
5829         ctl_done((union ctl_io *)ctsio);
5830         return (CTL_RETVAL_COMPLETE);
5831 }
5832
5833 int
5834 ctl_default_page_handler(struct ctl_scsiio *ctsio,
5835                          struct ctl_page_index *page_index, uint8_t *page_ptr)
5836 {
5837         struct ctl_lun *lun;
5838         uint8_t *current_cp, *saved_cp;
5839         int set_ua;
5840         uint32_t initidx;
5841
5842         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5843         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5844         set_ua = 0;
5845
5846         current_cp = (page_index->page_data + (page_index->page_len *
5847             CTL_PAGE_CURRENT));
5848         saved_cp = (page_index->page_data + (page_index->page_len *
5849             CTL_PAGE_SAVED));
5850
5851         mtx_lock(&lun->lun_lock);
5852         if (memcmp(current_cp, page_ptr, page_index->page_len)) {
5853                 memcpy(current_cp, page_ptr, page_index->page_len);
5854                 memcpy(saved_cp, page_ptr, page_index->page_len);
5855                 set_ua = 1;
5856         }
5857         if (set_ua != 0)
5858                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
5859         mtx_unlock(&lun->lun_lock);
5860         if (set_ua) {
5861                 ctl_isc_announce_mode(lun,
5862                     ctl_get_initindex(&ctsio->io_hdr.nexus),
5863                     page_index->page_code, page_index->subpage);
5864         }
5865         return (CTL_RETVAL_COMPLETE);
5866 }
5867
5868 static void
5869 ctl_ie_timer(void *arg)
5870 {
5871         struct ctl_lun *lun = arg;
5872         uint64_t t;
5873
5874         if (lun->ie_asc == 0)
5875                 return;
5876
5877         if (lun->MODE_IE.mrie == SIEP_MRIE_UA)
5878                 ctl_est_ua_all(lun, -1, CTL_UA_IE);
5879         else
5880                 lun->ie_reported = 0;
5881
5882         if (lun->ie_reportcnt < scsi_4btoul(lun->MODE_IE.report_count)) {
5883                 lun->ie_reportcnt++;
5884                 t = scsi_4btoul(lun->MODE_IE.interval_timer);
5885                 if (t == 0 || t == UINT32_MAX)
5886                         t = 3000;  /* 5 min */
5887                 callout_schedule(&lun->ie_callout, t * hz / 10);
5888         }
5889 }
5890
5891 int
5892 ctl_ie_page_handler(struct ctl_scsiio *ctsio,
5893                          struct ctl_page_index *page_index, uint8_t *page_ptr)
5894 {
5895         struct scsi_info_exceptions_page *pg;
5896         struct ctl_lun *lun;
5897         uint64_t t;
5898
5899         (void)ctl_default_page_handler(ctsio, page_index, page_ptr);
5900
5901         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5902         pg = (struct scsi_info_exceptions_page *)page_ptr;
5903         mtx_lock(&lun->lun_lock);
5904         if (pg->info_flags & SIEP_FLAGS_TEST) {
5905                 lun->ie_asc = 0x5d;
5906                 lun->ie_ascq = 0xff;
5907                 if (pg->mrie == SIEP_MRIE_UA) {
5908                         ctl_est_ua_all(lun, -1, CTL_UA_IE);
5909                         lun->ie_reported = 1;
5910                 } else {
5911                         ctl_clr_ua_all(lun, -1, CTL_UA_IE);
5912                         lun->ie_reported = -1;
5913                 }
5914                 lun->ie_reportcnt = 1;
5915                 if (lun->ie_reportcnt < scsi_4btoul(pg->report_count)) {
5916                         lun->ie_reportcnt++;
5917                         t = scsi_4btoul(pg->interval_timer);
5918                         if (t == 0 || t == UINT32_MAX)
5919                                 t = 3000;  /* 5 min */
5920                         callout_reset(&lun->ie_callout, t * hz / 10,
5921                             ctl_ie_timer, lun);
5922                 }
5923         } else {
5924                 lun->ie_asc = 0;
5925                 lun->ie_ascq = 0;
5926                 lun->ie_reported = 1;
5927                 ctl_clr_ua_all(lun, -1, CTL_UA_IE);
5928                 lun->ie_reportcnt = UINT32_MAX;
5929                 callout_stop(&lun->ie_callout);
5930         }
5931         mtx_unlock(&lun->lun_lock);
5932         return (CTL_RETVAL_COMPLETE);
5933 }
5934
5935 static int
5936 ctl_do_mode_select(union ctl_io *io)
5937 {
5938         struct scsi_mode_page_header *page_header;
5939         struct ctl_page_index *page_index;
5940         struct ctl_scsiio *ctsio;
5941         int page_len, page_len_offset, page_len_size;
5942         union ctl_modepage_info *modepage_info;
5943         struct ctl_lun *lun;
5944         int *len_left, *len_used;
5945         int retval, i;
5946
5947         ctsio = &io->scsiio;
5948         page_index = NULL;
5949         page_len = 0;
5950         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5951
5952         modepage_info = (union ctl_modepage_info *)
5953                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
5954         len_left = &modepage_info->header.len_left;
5955         len_used = &modepage_info->header.len_used;
5956
5957 do_next_page:
5958
5959         page_header = (struct scsi_mode_page_header *)
5960                 (ctsio->kern_data_ptr + *len_used);
5961
5962         if (*len_left == 0) {
5963                 free(ctsio->kern_data_ptr, M_CTL);
5964                 ctl_set_success(ctsio);
5965                 ctl_done((union ctl_io *)ctsio);
5966                 return (CTL_RETVAL_COMPLETE);
5967         } else if (*len_left < sizeof(struct scsi_mode_page_header)) {
5968
5969                 free(ctsio->kern_data_ptr, M_CTL);
5970                 ctl_set_param_len_error(ctsio);
5971                 ctl_done((union ctl_io *)ctsio);
5972                 return (CTL_RETVAL_COMPLETE);
5973
5974         } else if ((page_header->page_code & SMPH_SPF)
5975                 && (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
5976
5977                 free(ctsio->kern_data_ptr, M_CTL);
5978                 ctl_set_param_len_error(ctsio);
5979                 ctl_done((union ctl_io *)ctsio);
5980                 return (CTL_RETVAL_COMPLETE);
5981         }
5982
5983
5984         /*
5985          * XXX KDM should we do something with the block descriptor?
5986          */
5987         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
5988                 page_index = &lun->mode_pages.index[i];
5989                 if (lun->be_lun->lun_type == T_DIRECT &&
5990                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
5991                         continue;
5992                 if (lun->be_lun->lun_type == T_PROCESSOR &&
5993                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
5994                         continue;
5995                 if (lun->be_lun->lun_type == T_CDROM &&
5996                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
5997                         continue;
5998
5999                 if ((page_index->page_code & SMPH_PC_MASK) !=
6000                     (page_header->page_code & SMPH_PC_MASK))
6001                         continue;
6002
6003                 /*
6004                  * If neither page has a subpage code, then we've got a
6005                  * match.
6006                  */
6007                 if (((page_index->page_code & SMPH_SPF) == 0)
6008                  && ((page_header->page_code & SMPH_SPF) == 0)) {
6009                         page_len = page_header->page_length;
6010                         break;
6011                 }
6012
6013                 /*
6014                  * If both pages have subpages, then the subpage numbers
6015                  * have to match.
6016                  */
6017                 if ((page_index->page_code & SMPH_SPF)
6018                   && (page_header->page_code & SMPH_SPF)) {
6019                         struct scsi_mode_page_header_sp *sph;
6020
6021                         sph = (struct scsi_mode_page_header_sp *)page_header;
6022                         if (page_index->subpage == sph->subpage) {
6023                                 page_len = scsi_2btoul(sph->page_length);
6024                                 break;
6025                         }
6026                 }
6027         }
6028
6029         /*
6030          * If we couldn't find the page, or if we don't have a mode select
6031          * handler for it, send back an error to the user.
6032          */
6033         if ((i >= CTL_NUM_MODE_PAGES)
6034          || (page_index->select_handler == NULL)) {
6035                 ctl_set_invalid_field(ctsio,
6036                                       /*sks_valid*/ 1,
6037                                       /*command*/ 0,
6038                                       /*field*/ *len_used,
6039                                       /*bit_valid*/ 0,
6040                                       /*bit*/ 0);
6041                 free(ctsio->kern_data_ptr, M_CTL);
6042                 ctl_done((union ctl_io *)ctsio);
6043                 return (CTL_RETVAL_COMPLETE);
6044         }
6045
6046         if (page_index->page_code & SMPH_SPF) {
6047                 page_len_offset = 2;
6048                 page_len_size = 2;
6049         } else {
6050                 page_len_size = 1;
6051                 page_len_offset = 1;
6052         }
6053
6054         /*
6055          * If the length the initiator gives us isn't the one we specify in
6056          * the mode page header, or if they didn't specify enough data in
6057          * the CDB to avoid truncating this page, kick out the request.
6058          */
6059         if ((page_len != (page_index->page_len - page_len_offset -
6060                           page_len_size))
6061          || (*len_left < page_index->page_len)) {
6062
6063
6064                 ctl_set_invalid_field(ctsio,
6065                                       /*sks_valid*/ 1,
6066                                       /*command*/ 0,
6067                                       /*field*/ *len_used + page_len_offset,
6068                                       /*bit_valid*/ 0,
6069                                       /*bit*/ 0);
6070                 free(ctsio->kern_data_ptr, M_CTL);
6071                 ctl_done((union ctl_io *)ctsio);
6072                 return (CTL_RETVAL_COMPLETE);
6073         }
6074
6075         /*
6076          * Run through the mode page, checking to make sure that the bits
6077          * the user changed are actually legal for him to change.
6078          */
6079         for (i = 0; i < page_index->page_len; i++) {
6080                 uint8_t *user_byte, *change_mask, *current_byte;
6081                 int bad_bit;
6082                 int j;
6083
6084                 user_byte = (uint8_t *)page_header + i;
6085                 change_mask = page_index->page_data +
6086                               (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6087                 current_byte = page_index->page_data +
6088                                (page_index->page_len * CTL_PAGE_CURRENT) + i;
6089
6090                 /*
6091                  * Check to see whether the user set any bits in this byte
6092                  * that he is not allowed to set.
6093                  */
6094                 if ((*user_byte & ~(*change_mask)) ==
6095                     (*current_byte & ~(*change_mask)))
6096                         continue;
6097
6098                 /*
6099                  * Go through bit by bit to determine which one is illegal.
6100                  */
6101                 bad_bit = 0;
6102                 for (j = 7; j >= 0; j--) {
6103                         if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6104                             (((1 << i) & ~(*change_mask)) & *current_byte)) {
6105                                 bad_bit = i;
6106                                 break;
6107                         }
6108                 }
6109                 ctl_set_invalid_field(ctsio,
6110                                       /*sks_valid*/ 1,
6111                                       /*command*/ 0,
6112                                       /*field*/ *len_used + i,
6113                                       /*bit_valid*/ 1,
6114                                       /*bit*/ bad_bit);
6115                 free(ctsio->kern_data_ptr, M_CTL);
6116                 ctl_done((union ctl_io *)ctsio);
6117                 return (CTL_RETVAL_COMPLETE);
6118         }
6119
6120         /*
6121          * Decrement these before we call the page handler, since we may
6122          * end up getting called back one way or another before the handler
6123          * returns to this context.
6124          */
6125         *len_left -= page_index->page_len;
6126         *len_used += page_index->page_len;
6127
6128         retval = page_index->select_handler(ctsio, page_index,
6129                                             (uint8_t *)page_header);
6130
6131         /*
6132          * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6133          * wait until this queued command completes to finish processing
6134          * the mode page.  If it returns anything other than
6135          * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6136          * already set the sense information, freed the data pointer, and
6137          * completed the io for us.
6138          */
6139         if (retval != CTL_RETVAL_COMPLETE)
6140                 goto bailout_no_done;
6141
6142         /*
6143          * If the initiator sent us more than one page, parse the next one.
6144          */
6145         if (*len_left > 0)
6146                 goto do_next_page;
6147
6148         ctl_set_success(ctsio);
6149         free(ctsio->kern_data_ptr, M_CTL);
6150         ctl_done((union ctl_io *)ctsio);
6151
6152 bailout_no_done:
6153
6154         return (CTL_RETVAL_COMPLETE);
6155
6156 }
6157
6158 int
6159 ctl_mode_select(struct ctl_scsiio *ctsio)
6160 {
6161         int param_len, pf, sp;
6162         int header_size, bd_len;
6163         union ctl_modepage_info *modepage_info;
6164
6165         switch (ctsio->cdb[0]) {
6166         case MODE_SELECT_6: {
6167                 struct scsi_mode_select_6 *cdb;
6168
6169                 cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6170
6171                 pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6172                 sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6173                 param_len = cdb->length;
6174                 header_size = sizeof(struct scsi_mode_header_6);
6175                 break;
6176         }
6177         case MODE_SELECT_10: {
6178                 struct scsi_mode_select_10 *cdb;
6179
6180                 cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6181
6182                 pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6183                 sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6184                 param_len = scsi_2btoul(cdb->length);
6185                 header_size = sizeof(struct scsi_mode_header_10);
6186                 break;
6187         }
6188         default:
6189                 ctl_set_invalid_opcode(ctsio);
6190                 ctl_done((union ctl_io *)ctsio);
6191                 return (CTL_RETVAL_COMPLETE);
6192         }
6193
6194         /*
6195          * From SPC-3:
6196          * "A parameter list length of zero indicates that the Data-Out Buffer
6197          * shall be empty. This condition shall not be considered as an error."
6198          */
6199         if (param_len == 0) {
6200                 ctl_set_success(ctsio);
6201                 ctl_done((union ctl_io *)ctsio);
6202                 return (CTL_RETVAL_COMPLETE);
6203         }
6204
6205         /*
6206          * Since we'll hit this the first time through, prior to
6207          * allocation, we don't need to free a data buffer here.
6208          */
6209         if (param_len < header_size) {
6210                 ctl_set_param_len_error(ctsio);
6211                 ctl_done((union ctl_io *)ctsio);
6212                 return (CTL_RETVAL_COMPLETE);
6213         }
6214
6215         /*
6216          * Allocate the data buffer and grab the user's data.  In theory,
6217          * we shouldn't have to sanity check the parameter list length here
6218          * because the maximum size is 64K.  We should be able to malloc
6219          * that much without too many problems.
6220          */
6221         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6222                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6223                 ctsio->kern_data_len = param_len;
6224                 ctsio->kern_total_len = param_len;
6225                 ctsio->kern_data_resid = 0;
6226                 ctsio->kern_rel_offset = 0;
6227                 ctsio->kern_sg_entries = 0;
6228                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6229                 ctsio->be_move_done = ctl_config_move_done;
6230                 ctl_datamove((union ctl_io *)ctsio);
6231
6232                 return (CTL_RETVAL_COMPLETE);
6233         }
6234
6235         switch (ctsio->cdb[0]) {
6236         case MODE_SELECT_6: {
6237                 struct scsi_mode_header_6 *mh6;
6238
6239                 mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6240                 bd_len = mh6->blk_desc_len;
6241                 break;
6242         }
6243         case MODE_SELECT_10: {
6244                 struct scsi_mode_header_10 *mh10;
6245
6246                 mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6247                 bd_len = scsi_2btoul(mh10->blk_desc_len);
6248                 break;
6249         }
6250         default:
6251                 panic("%s: Invalid CDB type %#x", __func__, ctsio->cdb[0]);
6252         }
6253
6254         if (param_len < (header_size + bd_len)) {
6255                 free(ctsio->kern_data_ptr, M_CTL);
6256                 ctl_set_param_len_error(ctsio);
6257                 ctl_done((union ctl_io *)ctsio);
6258                 return (CTL_RETVAL_COMPLETE);
6259         }
6260
6261         /*
6262          * Set the IO_CONT flag, so that if this I/O gets passed to
6263          * ctl_config_write_done(), it'll get passed back to
6264          * ctl_do_mode_select() for further processing, or completion if
6265          * we're all done.
6266          */
6267         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6268         ctsio->io_cont = ctl_do_mode_select;
6269
6270         modepage_info = (union ctl_modepage_info *)
6271                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6272         memset(modepage_info, 0, sizeof(*modepage_info));
6273         modepage_info->header.len_left = param_len - header_size - bd_len;
6274         modepage_info->header.len_used = header_size + bd_len;
6275
6276         return (ctl_do_mode_select((union ctl_io *)ctsio));
6277 }
6278
6279 int
6280 ctl_mode_sense(struct ctl_scsiio *ctsio)
6281 {
6282         struct ctl_lun *lun;
6283         int pc, page_code, dbd, llba, subpage;
6284         int alloc_len, page_len, header_len, total_len;
6285         struct scsi_mode_block_descr *block_desc;
6286         struct ctl_page_index *page_index;
6287
6288         dbd = 0;
6289         llba = 0;
6290         block_desc = NULL;
6291
6292         CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6293
6294         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6295         switch (ctsio->cdb[0]) {
6296         case MODE_SENSE_6: {
6297                 struct scsi_mode_sense_6 *cdb;
6298
6299                 cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6300
6301                 header_len = sizeof(struct scsi_mode_hdr_6);
6302                 if (cdb->byte2 & SMS_DBD)
6303                         dbd = 1;
6304                 else
6305                         header_len += sizeof(struct scsi_mode_block_descr);
6306
6307                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6308                 page_code = cdb->page & SMS_PAGE_CODE;
6309                 subpage = cdb->subpage;
6310                 alloc_len = cdb->length;
6311                 break;
6312         }
6313         case MODE_SENSE_10: {
6314                 struct scsi_mode_sense_10 *cdb;
6315
6316                 cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6317
6318                 header_len = sizeof(struct scsi_mode_hdr_10);
6319
6320                 if (cdb->byte2 & SMS_DBD)
6321                         dbd = 1;
6322                 else
6323                         header_len += sizeof(struct scsi_mode_block_descr);
6324                 if (cdb->byte2 & SMS10_LLBAA)
6325                         llba = 1;
6326                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6327                 page_code = cdb->page & SMS_PAGE_CODE;
6328                 subpage = cdb->subpage;
6329                 alloc_len = scsi_2btoul(cdb->length);
6330                 break;
6331         }
6332         default:
6333                 ctl_set_invalid_opcode(ctsio);
6334                 ctl_done((union ctl_io *)ctsio);
6335                 return (CTL_RETVAL_COMPLETE);
6336                 break; /* NOTREACHED */
6337         }
6338
6339         /*
6340          * We have to make a first pass through to calculate the size of
6341          * the pages that match the user's query.  Then we allocate enough
6342          * memory to hold it, and actually copy the data into the buffer.
6343          */
6344         switch (page_code) {
6345         case SMS_ALL_PAGES_PAGE: {
6346                 u_int i;
6347
6348                 page_len = 0;
6349
6350                 /*
6351                  * At the moment, values other than 0 and 0xff here are
6352                  * reserved according to SPC-3.
6353                  */
6354                 if ((subpage != SMS_SUBPAGE_PAGE_0)
6355                  && (subpage != SMS_SUBPAGE_ALL)) {
6356                         ctl_set_invalid_field(ctsio,
6357                                               /*sks_valid*/ 1,
6358                                               /*command*/ 1,
6359                                               /*field*/ 3,
6360                                               /*bit_valid*/ 0,
6361                                               /*bit*/ 0);
6362                         ctl_done((union ctl_io *)ctsio);
6363                         return (CTL_RETVAL_COMPLETE);
6364                 }
6365
6366                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6367                         page_index = &lun->mode_pages.index[i];
6368
6369                         /* Make sure the page is supported for this dev type */
6370                         if (lun->be_lun->lun_type == T_DIRECT &&
6371                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6372                                 continue;
6373                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6374                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6375                                 continue;
6376                         if (lun->be_lun->lun_type == T_CDROM &&
6377                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6378                                 continue;
6379
6380                         /*
6381                          * We don't use this subpage if the user didn't
6382                          * request all subpages.
6383                          */
6384                         if ((page_index->subpage != 0)
6385                          && (subpage == SMS_SUBPAGE_PAGE_0))
6386                                 continue;
6387
6388 #if 0
6389                         printf("found page %#x len %d\n",
6390                                page_index->page_code & SMPH_PC_MASK,
6391                                page_index->page_len);
6392 #endif
6393                         page_len += page_index->page_len;
6394                 }
6395                 break;
6396         }
6397         default: {
6398                 u_int i;
6399
6400                 page_len = 0;
6401
6402                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6403                         page_index = &lun->mode_pages.index[i];
6404
6405                         /* Make sure the page is supported for this dev type */
6406                         if (lun->be_lun->lun_type == T_DIRECT &&
6407                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6408                                 continue;
6409                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6410                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6411                                 continue;
6412                         if (lun->be_lun->lun_type == T_CDROM &&
6413                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6414                                 continue;
6415
6416                         /* Look for the right page code */
6417                         if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6418                                 continue;
6419
6420                         /* Look for the right subpage or the subpage wildcard*/
6421                         if ((page_index->subpage != subpage)
6422                          && (subpage != SMS_SUBPAGE_ALL))
6423                                 continue;
6424
6425 #if 0
6426                         printf("found page %#x len %d\n",
6427                                page_index->page_code & SMPH_PC_MASK,
6428                                page_index->page_len);
6429 #endif
6430
6431                         page_len += page_index->page_len;
6432                 }
6433
6434                 if (page_len == 0) {
6435                         ctl_set_invalid_field(ctsio,
6436                                               /*sks_valid*/ 1,
6437                                               /*command*/ 1,
6438                                               /*field*/ 2,
6439                                               /*bit_valid*/ 1,
6440                                               /*bit*/ 5);
6441                         ctl_done((union ctl_io *)ctsio);
6442                         return (CTL_RETVAL_COMPLETE);
6443                 }
6444                 break;
6445         }
6446         }
6447
6448         total_len = header_len + page_len;
6449 #if 0
6450         printf("header_len = %d, page_len = %d, total_len = %d\n",
6451                header_len, page_len, total_len);
6452 #endif
6453
6454         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6455         ctsio->kern_sg_entries = 0;
6456         ctsio->kern_data_resid = 0;
6457         ctsio->kern_rel_offset = 0;
6458         if (total_len < alloc_len) {
6459                 ctsio->residual = alloc_len - total_len;
6460                 ctsio->kern_data_len = total_len;
6461                 ctsio->kern_total_len = total_len;
6462         } else {
6463                 ctsio->residual = 0;
6464                 ctsio->kern_data_len = alloc_len;
6465                 ctsio->kern_total_len = alloc_len;
6466         }
6467
6468         switch (ctsio->cdb[0]) {
6469         case MODE_SENSE_6: {
6470                 struct scsi_mode_hdr_6 *header;
6471
6472                 header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6473
6474                 header->datalen = MIN(total_len - 1, 254);
6475                 if (lun->be_lun->lun_type == T_DIRECT) {
6476                         header->dev_specific = 0x10; /* DPOFUA */
6477                         if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6478                             (lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0)
6479                                 header->dev_specific |= 0x80; /* WP */
6480                 }
6481                 if (dbd)
6482                         header->block_descr_len = 0;
6483                 else
6484                         header->block_descr_len =
6485                                 sizeof(struct scsi_mode_block_descr);
6486                 block_desc = (struct scsi_mode_block_descr *)&header[1];
6487                 break;
6488         }
6489         case MODE_SENSE_10: {
6490                 struct scsi_mode_hdr_10 *header;
6491                 int datalen;
6492
6493                 header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6494
6495                 datalen = MIN(total_len - 2, 65533);
6496                 scsi_ulto2b(datalen, header->datalen);
6497                 if (lun->be_lun->lun_type == T_DIRECT) {
6498                         header->dev_specific = 0x10; /* DPOFUA */
6499                         if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6500                             (lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0)
6501                                 header->dev_specific |= 0x80; /* WP */
6502                 }
6503                 if (dbd)
6504                         scsi_ulto2b(0, header->block_descr_len);
6505                 else
6506                         scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
6507                                     header->block_descr_len);
6508                 block_desc = (struct scsi_mode_block_descr *)&header[1];
6509                 break;
6510         }
6511         default:
6512                 panic("%s: Invalid CDB type %#x", __func__, ctsio->cdb[0]);
6513         }
6514
6515         /*
6516          * If we've got a disk, use its blocksize in the block
6517          * descriptor.  Otherwise, just set it to 0.
6518          */
6519         if (dbd == 0) {
6520                 if (lun->be_lun->lun_type == T_DIRECT)
6521                         scsi_ulto3b(lun->be_lun->blocksize,
6522                                     block_desc->block_len);
6523                 else
6524                         scsi_ulto3b(0, block_desc->block_len);
6525         }
6526
6527         switch (page_code) {
6528         case SMS_ALL_PAGES_PAGE: {
6529                 int i, data_used;
6530
6531                 data_used = header_len;
6532                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6533                         struct ctl_page_index *page_index;
6534
6535                         page_index = &lun->mode_pages.index[i];
6536                         if (lun->be_lun->lun_type == T_DIRECT &&
6537                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6538                                 continue;
6539                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6540                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6541                                 continue;
6542                         if (lun->be_lun->lun_type == T_CDROM &&
6543                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6544                                 continue;
6545
6546                         /*
6547                          * We don't use this subpage if the user didn't
6548                          * request all subpages.  We already checked (above)
6549                          * to make sure the user only specified a subpage
6550                          * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
6551                          */
6552                         if ((page_index->subpage != 0)
6553                          && (subpage == SMS_SUBPAGE_PAGE_0))
6554                                 continue;
6555
6556                         /*
6557                          * Call the handler, if it exists, to update the
6558                          * page to the latest values.
6559                          */
6560                         if (page_index->sense_handler != NULL)
6561                                 page_index->sense_handler(ctsio, page_index,pc);
6562
6563                         memcpy(ctsio->kern_data_ptr + data_used,
6564                                page_index->page_data +
6565                                (page_index->page_len * pc),
6566                                page_index->page_len);
6567                         data_used += page_index->page_len;
6568                 }
6569                 break;
6570         }
6571         default: {
6572                 int i, data_used;
6573
6574                 data_used = header_len;
6575
6576                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6577                         struct ctl_page_index *page_index;
6578
6579                         page_index = &lun->mode_pages.index[i];
6580
6581                         /* Look for the right page code */
6582                         if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6583                                 continue;
6584
6585                         /* Look for the right subpage or the subpage wildcard*/
6586                         if ((page_index->subpage != subpage)
6587                          && (subpage != SMS_SUBPAGE_ALL))
6588                                 continue;
6589
6590                         /* Make sure the page is supported for this dev type */
6591                         if (lun->be_lun->lun_type == T_DIRECT &&
6592                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6593                                 continue;
6594                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6595                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6596                                 continue;
6597                         if (lun->be_lun->lun_type == T_CDROM &&
6598                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6599                                 continue;
6600
6601                         /*
6602                          * Call the handler, if it exists, to update the
6603                          * page to the latest values.
6604                          */
6605                         if (page_index->sense_handler != NULL)
6606                                 page_index->sense_handler(ctsio, page_index,pc);
6607
6608                         memcpy(ctsio->kern_data_ptr + data_used,
6609                                page_index->page_data +
6610                                (page_index->page_len * pc),
6611                                page_index->page_len);
6612                         data_used += page_index->page_len;
6613                 }
6614                 break;
6615         }
6616         }
6617
6618         ctl_set_success(ctsio);
6619         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6620         ctsio->be_move_done = ctl_config_move_done;
6621         ctl_datamove((union ctl_io *)ctsio);
6622         return (CTL_RETVAL_COMPLETE);
6623 }
6624
6625 int
6626 ctl_lbp_log_sense_handler(struct ctl_scsiio *ctsio,
6627                                struct ctl_page_index *page_index,
6628                                int pc)
6629 {
6630         struct ctl_lun *lun;
6631         struct scsi_log_param_header *phdr;
6632         uint8_t *data;
6633         uint64_t val;
6634
6635         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6636         data = page_index->page_data;
6637
6638         if (lun->backend->lun_attr != NULL &&
6639             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksavail"))
6640              != UINT64_MAX) {
6641                 phdr = (struct scsi_log_param_header *)data;
6642                 scsi_ulto2b(0x0001, phdr->param_code);
6643                 phdr->param_control = SLP_LBIN | SLP_LP;
6644                 phdr->param_len = 8;
6645                 data = (uint8_t *)(phdr + 1);
6646                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6647                 data[4] = 0x02; /* per-pool */
6648                 data += phdr->param_len;
6649         }
6650
6651         if (lun->backend->lun_attr != NULL &&
6652             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksused"))
6653              != UINT64_MAX) {
6654                 phdr = (struct scsi_log_param_header *)data;
6655                 scsi_ulto2b(0x0002, phdr->param_code);
6656                 phdr->param_control = SLP_LBIN | SLP_LP;
6657                 phdr->param_len = 8;
6658                 data = (uint8_t *)(phdr + 1);
6659                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6660                 data[4] = 0x01; /* per-LUN */
6661                 data += phdr->param_len;
6662         }
6663
6664         if (lun->backend->lun_attr != NULL &&
6665             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksavail"))
6666              != UINT64_MAX) {
6667                 phdr = (struct scsi_log_param_header *)data;
6668                 scsi_ulto2b(0x00f1, phdr->param_code);
6669                 phdr->param_control = SLP_LBIN | SLP_LP;
6670                 phdr->param_len = 8;
6671                 data = (uint8_t *)(phdr + 1);
6672                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6673                 data[4] = 0x02; /* per-pool */
6674                 data += phdr->param_len;
6675         }
6676
6677         if (lun->backend->lun_attr != NULL &&
6678             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksused"))
6679              != UINT64_MAX) {
6680                 phdr = (struct scsi_log_param_header *)data;
6681                 scsi_ulto2b(0x00f2, phdr->param_code);
6682                 phdr->param_control = SLP_LBIN | SLP_LP;
6683                 phdr->param_len = 8;
6684                 data = (uint8_t *)(phdr + 1);
6685                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6686                 data[4] = 0x02; /* per-pool */
6687                 data += phdr->param_len;
6688         }
6689
6690         page_index->page_len = data - page_index->page_data;
6691         return (0);
6692 }
6693
6694 int
6695 ctl_sap_log_sense_handler(struct ctl_scsiio *ctsio,
6696                                struct ctl_page_index *page_index,
6697                                int pc)
6698 {
6699         struct ctl_lun *lun;
6700         struct stat_page *data;
6701         uint64_t rn, wn, rb, wb;
6702         struct bintime rt, wt;
6703         int i;
6704
6705         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6706         data = (struct stat_page *)page_index->page_data;
6707
6708         scsi_ulto2b(SLP_SAP, data->sap.hdr.param_code);
6709         data->sap.hdr.param_control = SLP_LBIN;
6710         data->sap.hdr.param_len = sizeof(struct scsi_log_stat_and_perf) -
6711             sizeof(struct scsi_log_param_header);
6712         rn = wn = rb = wb = 0;
6713         bintime_clear(&rt);
6714         bintime_clear(&wt);
6715         for (i = 0; i < CTL_MAX_PORTS; i++) {
6716                 rn += lun->stats.ports[i].operations[CTL_STATS_READ];
6717                 wn += lun->stats.ports[i].operations[CTL_STATS_WRITE];
6718                 rb += lun->stats.ports[i].bytes[CTL_STATS_READ];
6719                 wb += lun->stats.ports[i].bytes[CTL_STATS_WRITE];
6720                 bintime_add(&rt, &lun->stats.ports[i].time[CTL_STATS_READ]);
6721                 bintime_add(&wt, &lun->stats.ports[i].time[CTL_STATS_WRITE]);
6722         }
6723         scsi_u64to8b(rn, data->sap.read_num);
6724         scsi_u64to8b(wn, data->sap.write_num);
6725         if (lun->stats.blocksize > 0) {
6726                 scsi_u64to8b(wb / lun->stats.blocksize,
6727                     data->sap.recvieved_lba);
6728                 scsi_u64to8b(rb / lun->stats.blocksize,
6729                     data->sap.transmitted_lba);
6730         }
6731         scsi_u64to8b((uint64_t)rt.sec * 1000 + rt.frac / (UINT64_MAX / 1000),
6732             data->sap.read_int);
6733         scsi_u64to8b((uint64_t)wt.sec * 1000 + wt.frac / (UINT64_MAX / 1000),
6734             data->sap.write_int);
6735         scsi_u64to8b(0, data->sap.weighted_num);
6736         scsi_u64to8b(0, data->sap.weighted_int);
6737         scsi_ulto2b(SLP_IT, data->it.hdr.param_code);
6738         data->it.hdr.param_control = SLP_LBIN;
6739         data->it.hdr.param_len = sizeof(struct scsi_log_idle_time) -
6740             sizeof(struct scsi_log_param_header);
6741 #ifdef CTL_TIME_IO
6742         scsi_u64to8b(lun->idle_time / SBT_1MS, data->it.idle_int);
6743 #endif
6744         scsi_ulto2b(SLP_TI, data->ti.hdr.param_code);
6745         data->it.hdr.param_control = SLP_LBIN;
6746         data->ti.hdr.param_len = sizeof(struct scsi_log_time_interval) -
6747             sizeof(struct scsi_log_param_header);
6748         scsi_ulto4b(3, data->ti.exponent);
6749         scsi_ulto4b(1, data->ti.integer);
6750         return (0);
6751 }
6752
6753 int
6754 ctl_ie_log_sense_handler(struct ctl_scsiio *ctsio,
6755                                struct ctl_page_index *page_index,
6756                                int pc)
6757 {
6758         struct ctl_lun *lun;
6759         struct scsi_log_informational_exceptions *data;
6760
6761         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6762         data = (struct scsi_log_informational_exceptions *)page_index->page_data;
6763
6764         scsi_ulto2b(SLP_IE_GEN, data->hdr.param_code);
6765         data->hdr.param_control = SLP_LBIN;
6766         data->hdr.param_len = sizeof(struct scsi_log_informational_exceptions) -
6767             sizeof(struct scsi_log_param_header);
6768         data->ie_asc = lun->ie_asc;
6769         data->ie_ascq = lun->ie_ascq;
6770         data->temperature = 0xff;
6771         return (0);
6772 }
6773
6774 int
6775 ctl_log_sense(struct ctl_scsiio *ctsio)
6776 {
6777         struct ctl_lun *lun;
6778         int i, pc, page_code, subpage;
6779         int alloc_len, total_len;
6780         struct ctl_page_index *page_index;
6781         struct scsi_log_sense *cdb;
6782         struct scsi_log_header *header;
6783
6784         CTL_DEBUG_PRINT(("ctl_log_sense\n"));
6785
6786         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6787         cdb = (struct scsi_log_sense *)ctsio->cdb;
6788         pc = (cdb->page & SLS_PAGE_CTRL_MASK) >> 6;
6789         page_code = cdb->page & SLS_PAGE_CODE;
6790         subpage = cdb->subpage;
6791         alloc_len = scsi_2btoul(cdb->length);
6792
6793         page_index = NULL;
6794         for (i = 0; i < CTL_NUM_LOG_PAGES; i++) {
6795                 page_index = &lun->log_pages.index[i];
6796
6797                 /* Look for the right page code */
6798                 if ((page_index->page_code & SL_PAGE_CODE) != page_code)
6799                         continue;
6800
6801                 /* Look for the right subpage or the subpage wildcard*/
6802                 if (page_index->subpage != subpage)
6803                         continue;
6804
6805                 break;
6806         }
6807         if (i >= CTL_NUM_LOG_PAGES) {
6808                 ctl_set_invalid_field(ctsio,
6809                                       /*sks_valid*/ 1,
6810                                       /*command*/ 1,
6811                                       /*field*/ 2,
6812                                       /*bit_valid*/ 0,
6813                                       /*bit*/ 0);
6814                 ctl_done((union ctl_io *)ctsio);
6815                 return (CTL_RETVAL_COMPLETE);
6816         }
6817
6818         total_len = sizeof(struct scsi_log_header) + page_index->page_len;
6819
6820         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6821         ctsio->kern_sg_entries = 0;
6822         ctsio->kern_data_resid = 0;
6823         ctsio->kern_rel_offset = 0;
6824         if (total_len < alloc_len) {
6825                 ctsio->residual = alloc_len - total_len;
6826                 ctsio->kern_data_len = total_len;
6827                 ctsio->kern_total_len = total_len;
6828         } else {
6829                 ctsio->residual = 0;
6830                 ctsio->kern_data_len = alloc_len;
6831                 ctsio->kern_total_len = alloc_len;
6832         }
6833
6834         header = (struct scsi_log_header *)ctsio->kern_data_ptr;
6835         header->page = page_index->page_code;
6836         if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING)
6837                 header->page |= SL_DS;
6838         if (page_index->subpage) {
6839                 header->page |= SL_SPF;
6840                 header->subpage = page_index->subpage;
6841         }
6842         scsi_ulto2b(page_index->page_len, header->datalen);
6843
6844         /*
6845          * Call the handler, if it exists, to update the
6846          * page to the latest values.
6847          */
6848         if (page_index->sense_handler != NULL)
6849                 page_index->sense_handler(ctsio, page_index, pc);
6850
6851         memcpy(header + 1, page_index->page_data, page_index->page_len);
6852
6853         ctl_set_success(ctsio);
6854         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6855         ctsio->be_move_done = ctl_config_move_done;
6856         ctl_datamove((union ctl_io *)ctsio);
6857         return (CTL_RETVAL_COMPLETE);
6858 }
6859
6860 int
6861 ctl_read_capacity(struct ctl_scsiio *ctsio)
6862 {
6863         struct scsi_read_capacity *cdb;
6864         struct scsi_read_capacity_data *data;
6865         struct ctl_lun *lun;
6866         uint32_t lba;
6867
6868         CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
6869
6870         cdb = (struct scsi_read_capacity *)ctsio->cdb;
6871
6872         lba = scsi_4btoul(cdb->addr);
6873         if (((cdb->pmi & SRC_PMI) == 0)
6874          && (lba != 0)) {
6875                 ctl_set_invalid_field(/*ctsio*/ ctsio,
6876                                       /*sks_valid*/ 1,
6877                                       /*command*/ 1,
6878                                       /*field*/ 2,
6879                                       /*bit_valid*/ 0,
6880                                       /*bit*/ 0);
6881                 ctl_done((union ctl_io *)ctsio);
6882                 return (CTL_RETVAL_COMPLETE);
6883         }
6884
6885         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6886
6887         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6888         data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
6889         ctsio->residual = 0;
6890         ctsio->kern_data_len = sizeof(*data);
6891         ctsio->kern_total_len = sizeof(*data);
6892         ctsio->kern_data_resid = 0;
6893         ctsio->kern_rel_offset = 0;
6894         ctsio->kern_sg_entries = 0;
6895
6896         /*
6897          * If the maximum LBA is greater than 0xfffffffe, the user must
6898          * issue a SERVICE ACTION IN (16) command, with the read capacity
6899          * serivce action set.
6900          */
6901         if (lun->be_lun->maxlba > 0xfffffffe)
6902                 scsi_ulto4b(0xffffffff, data->addr);
6903         else
6904                 scsi_ulto4b(lun->be_lun->maxlba, data->addr);
6905
6906         /*
6907          * XXX KDM this may not be 512 bytes...
6908          */
6909         scsi_ulto4b(lun->be_lun->blocksize, data->length);
6910
6911         ctl_set_success(ctsio);
6912         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6913         ctsio->be_move_done = ctl_config_move_done;
6914         ctl_datamove((union ctl_io *)ctsio);
6915         return (CTL_RETVAL_COMPLETE);
6916 }
6917
6918 int
6919 ctl_read_capacity_16(struct ctl_scsiio *ctsio)
6920 {
6921         struct scsi_read_capacity_16 *cdb;
6922         struct scsi_read_capacity_data_long *data;
6923         struct ctl_lun *lun;
6924         uint64_t lba;
6925         uint32_t alloc_len;
6926
6927         CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
6928
6929         cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
6930
6931         alloc_len = scsi_4btoul(cdb->alloc_len);
6932         lba = scsi_8btou64(cdb->addr);
6933
6934         if ((cdb->reladr & SRC16_PMI)
6935          && (lba != 0)) {
6936                 ctl_set_invalid_field(/*ctsio*/ ctsio,
6937                                       /*sks_valid*/ 1,
6938                                       /*command*/ 1,
6939                                       /*field*/ 2,
6940                                       /*bit_valid*/ 0,
6941                                       /*bit*/ 0);
6942                 ctl_done((union ctl_io *)ctsio);
6943                 return (CTL_RETVAL_COMPLETE);
6944         }
6945
6946         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6947
6948         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6949         data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
6950
6951         if (sizeof(*data) < alloc_len) {
6952                 ctsio->residual = alloc_len - sizeof(*data);
6953                 ctsio->kern_data_len = sizeof(*data);
6954                 ctsio->kern_total_len = sizeof(*data);
6955         } else {
6956                 ctsio->residual = 0;
6957                 ctsio->kern_data_len = alloc_len;
6958                 ctsio->kern_total_len = alloc_len;
6959         }
6960         ctsio->kern_data_resid = 0;
6961         ctsio->kern_rel_offset = 0;
6962         ctsio->kern_sg_entries = 0;
6963
6964         scsi_u64to8b(lun->be_lun->maxlba, data->addr);
6965         /* XXX KDM this may not be 512 bytes... */
6966         scsi_ulto4b(lun->be_lun->blocksize, data->length);
6967         data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
6968         scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
6969         if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
6970                 data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ;
6971
6972         ctl_set_success(ctsio);
6973         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6974         ctsio->be_move_done = ctl_config_move_done;
6975         ctl_datamove((union ctl_io *)ctsio);
6976         return (CTL_RETVAL_COMPLETE);
6977 }
6978
6979 int
6980 ctl_get_lba_status(struct ctl_scsiio *ctsio)
6981 {
6982         struct scsi_get_lba_status *cdb;
6983         struct scsi_get_lba_status_data *data;
6984         struct ctl_lun *lun;
6985         struct ctl_lba_len_flags *lbalen;
6986         uint64_t lba;
6987         uint32_t alloc_len, total_len;
6988         int retval;
6989
6990         CTL_DEBUG_PRINT(("ctl_get_lba_status\n"));
6991
6992         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6993         cdb = (struct scsi_get_lba_status *)ctsio->cdb;
6994         lba = scsi_8btou64(cdb->addr);
6995         alloc_len = scsi_4btoul(cdb->alloc_len);
6996
6997         if (lba > lun->be_lun->maxlba) {
6998                 ctl_set_lba_out_of_range(ctsio);
6999                 ctl_done((union ctl_io *)ctsio);
7000                 return (CTL_RETVAL_COMPLETE);
7001         }
7002
7003         total_len = sizeof(*data) + sizeof(data->descr[0]);
7004         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7005         data = (struct scsi_get_lba_status_data *)ctsio->kern_data_ptr;
7006
7007         if (total_len < alloc_len) {
7008                 ctsio->residual = alloc_len - total_len;
7009                 ctsio->kern_data_len = total_len;
7010                 ctsio->kern_total_len = total_len;
7011         } else {
7012                 ctsio->residual = 0;
7013                 ctsio->kern_data_len = alloc_len;
7014                 ctsio->kern_total_len = alloc_len;
7015         }
7016         ctsio->kern_data_resid = 0;
7017         ctsio->kern_rel_offset = 0;
7018         ctsio->kern_sg_entries = 0;
7019
7020         /* Fill dummy data in case backend can't tell anything. */
7021         scsi_ulto4b(4 + sizeof(data->descr[0]), data->length);
7022         scsi_u64to8b(lba, data->descr[0].addr);
7023         scsi_ulto4b(MIN(UINT32_MAX, lun->be_lun->maxlba + 1 - lba),
7024             data->descr[0].length);
7025         data->descr[0].status = 0; /* Mapped or unknown. */
7026
7027         ctl_set_success(ctsio);
7028         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7029         ctsio->be_move_done = ctl_config_move_done;
7030
7031         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
7032         lbalen->lba = lba;
7033         lbalen->len = total_len;
7034         lbalen->flags = 0;
7035         retval = lun->backend->config_read((union ctl_io *)ctsio);
7036         return (CTL_RETVAL_COMPLETE);
7037 }
7038
7039 int
7040 ctl_read_defect(struct ctl_scsiio *ctsio)
7041 {
7042         struct scsi_read_defect_data_10 *ccb10;
7043         struct scsi_read_defect_data_12 *ccb12;
7044         struct scsi_read_defect_data_hdr_10 *data10;
7045         struct scsi_read_defect_data_hdr_12 *data12;
7046         uint32_t alloc_len, data_len;
7047         uint8_t format;
7048
7049         CTL_DEBUG_PRINT(("ctl_read_defect\n"));
7050
7051         if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7052                 ccb10 = (struct scsi_read_defect_data_10 *)&ctsio->cdb;
7053                 format = ccb10->format;
7054                 alloc_len = scsi_2btoul(ccb10->alloc_length);
7055                 data_len = sizeof(*data10);
7056         } else {
7057                 ccb12 = (struct scsi_read_defect_data_12 *)&ctsio->cdb;
7058                 format = ccb12->format;
7059                 alloc_len = scsi_4btoul(ccb12->alloc_length);
7060                 data_len = sizeof(*data12);
7061         }
7062         if (alloc_len == 0) {
7063                 ctl_set_success(ctsio);
7064                 ctl_done((union ctl_io *)ctsio);
7065                 return (CTL_RETVAL_COMPLETE);
7066         }
7067
7068         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
7069         if (data_len < alloc_len) {
7070                 ctsio->residual = alloc_len - data_len;
7071                 ctsio->kern_data_len = data_len;
7072                 ctsio->kern_total_len = data_len;
7073         } else {
7074                 ctsio->residual = 0;
7075                 ctsio->kern_data_len = alloc_len;
7076                 ctsio->kern_total_len = alloc_len;
7077         }
7078         ctsio->kern_data_resid = 0;
7079         ctsio->kern_rel_offset = 0;
7080         ctsio->kern_sg_entries = 0;
7081
7082         if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7083                 data10 = (struct scsi_read_defect_data_hdr_10 *)
7084                     ctsio->kern_data_ptr;
7085                 data10->format = format;
7086                 scsi_ulto2b(0, data10->length);
7087         } else {
7088                 data12 = (struct scsi_read_defect_data_hdr_12 *)
7089                     ctsio->kern_data_ptr;
7090                 data12->format = format;
7091                 scsi_ulto2b(0, data12->generation);
7092                 scsi_ulto4b(0, data12->length);
7093         }
7094
7095         ctl_set_success(ctsio);
7096         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7097         ctsio->be_move_done = ctl_config_move_done;
7098         ctl_datamove((union ctl_io *)ctsio);
7099         return (CTL_RETVAL_COMPLETE);
7100 }
7101
7102 int
7103 ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7104 {
7105         struct scsi_maintenance_in *cdb;
7106         int retval;
7107         int alloc_len, ext, total_len = 0, g, pc, pg, ts, os;
7108         int num_ha_groups, num_target_ports, shared_group;
7109         struct ctl_lun *lun;
7110         struct ctl_softc *softc;
7111         struct ctl_port *port;
7112         struct scsi_target_group_data *rtg_ptr;
7113         struct scsi_target_group_data_extended *rtg_ext_ptr;
7114         struct scsi_target_port_group_descriptor *tpg_desc;
7115
7116         CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7117
7118         cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7119         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7120         softc = lun->ctl_softc;
7121
7122         retval = CTL_RETVAL_COMPLETE;
7123
7124         switch (cdb->byte2 & STG_PDF_MASK) {
7125         case STG_PDF_LENGTH:
7126                 ext = 0;
7127                 break;
7128         case STG_PDF_EXTENDED:
7129                 ext = 1;
7130                 break;
7131         default:
7132                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7133                                       /*sks_valid*/ 1,
7134                                       /*command*/ 1,
7135                                       /*field*/ 2,
7136                                       /*bit_valid*/ 1,
7137                                       /*bit*/ 5);
7138                 ctl_done((union ctl_io *)ctsio);
7139                 return(retval);
7140         }
7141
7142         num_target_ports = 0;
7143         shared_group = (softc->is_single != 0);
7144         mtx_lock(&softc->ctl_lock);
7145         STAILQ_FOREACH(port, &softc->port_list, links) {
7146                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7147                         continue;
7148                 if (ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
7149                         continue;
7150                 num_target_ports++;
7151                 if (port->status & CTL_PORT_STATUS_HA_SHARED)
7152                         shared_group = 1;
7153         }
7154         mtx_unlock(&softc->ctl_lock);
7155         num_ha_groups = (softc->is_single) ? 0 : NUM_HA_SHELVES;
7156
7157         if (ext)
7158                 total_len = sizeof(struct scsi_target_group_data_extended);
7159         else
7160                 total_len = sizeof(struct scsi_target_group_data);
7161         total_len += sizeof(struct scsi_target_port_group_descriptor) *
7162                 (shared_group + num_ha_groups) +
7163             sizeof(struct scsi_target_port_descriptor) * num_target_ports;
7164
7165         alloc_len = scsi_4btoul(cdb->length);
7166
7167         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7168
7169         ctsio->kern_sg_entries = 0;
7170
7171         if (total_len < alloc_len) {
7172                 ctsio->residual = alloc_len - total_len;
7173                 ctsio->kern_data_len = total_len;
7174                 ctsio->kern_total_len = total_len;
7175         } else {
7176                 ctsio->residual = 0;
7177                 ctsio->kern_data_len = alloc_len;
7178                 ctsio->kern_total_len = alloc_len;
7179         }
7180         ctsio->kern_data_resid = 0;
7181         ctsio->kern_rel_offset = 0;
7182
7183         if (ext) {
7184                 rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7185                     ctsio->kern_data_ptr;
7186                 scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7187                 rtg_ext_ptr->format_type = 0x10;
7188                 rtg_ext_ptr->implicit_transition_time = 0;
7189                 tpg_desc = &rtg_ext_ptr->groups[0];
7190         } else {
7191                 rtg_ptr = (struct scsi_target_group_data *)
7192                     ctsio->kern_data_ptr;
7193                 scsi_ulto4b(total_len - 4, rtg_ptr->length);
7194                 tpg_desc = &rtg_ptr->groups[0];
7195         }
7196
7197         mtx_lock(&softc->ctl_lock);
7198         pg = softc->port_min / softc->port_cnt;
7199         if (lun->flags & (CTL_LUN_PRIMARY_SC | CTL_LUN_PEER_SC_PRIMARY)) {
7200                 /* Some shelf is known to be primary. */
7201                 if (softc->ha_link == CTL_HA_LINK_OFFLINE)
7202                         os = TPG_ASYMMETRIC_ACCESS_UNAVAILABLE;
7203                 else if (softc->ha_link == CTL_HA_LINK_UNKNOWN)
7204                         os = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7205                 else if (softc->ha_mode == CTL_HA_MODE_ACT_STBY)
7206                         os = TPG_ASYMMETRIC_ACCESS_STANDBY;
7207                 else
7208                         os = TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7209                 if (lun->flags & CTL_LUN_PRIMARY_SC) {
7210                         ts = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7211                 } else {
7212                         ts = os;
7213                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7214                 }
7215         } else {
7216                 /* No known primary shelf. */
7217                 if (softc->ha_link == CTL_HA_LINK_OFFLINE) {
7218                         ts = TPG_ASYMMETRIC_ACCESS_UNAVAILABLE;
7219                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7220                 } else if (softc->ha_link == CTL_HA_LINK_UNKNOWN) {
7221                         ts = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7222                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7223                 } else {
7224                         ts = os = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7225                 }
7226         }
7227         if (shared_group) {
7228                 tpg_desc->pref_state = ts;
7229                 tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP |
7230                     TPG_U_SUP | TPG_T_SUP;
7231                 scsi_ulto2b(1, tpg_desc->target_port_group);
7232                 tpg_desc->status = TPG_IMPLICIT;
7233                 pc = 0;
7234                 STAILQ_FOREACH(port, &softc->port_list, links) {
7235                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7236                                 continue;
7237                         if (!softc->is_single &&
7238                             (port->status & CTL_PORT_STATUS_HA_SHARED) == 0)
7239                                 continue;
7240                         if (ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
7241                                 continue;
7242                         scsi_ulto2b(port->targ_port, tpg_desc->descriptors[pc].
7243                             relative_target_port_identifier);
7244                         pc++;
7245                 }
7246                 tpg_desc->target_port_count = pc;
7247                 tpg_desc = (struct scsi_target_port_group_descriptor *)
7248                     &tpg_desc->descriptors[pc];
7249         }
7250         for (g = 0; g < num_ha_groups; g++) {
7251                 tpg_desc->pref_state = (g == pg) ? ts : os;
7252                 tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP |
7253                     TPG_U_SUP | TPG_T_SUP;
7254                 scsi_ulto2b(2 + g, tpg_desc->target_port_group);
7255                 tpg_desc->status = TPG_IMPLICIT;
7256                 pc = 0;
7257                 STAILQ_FOREACH(port, &softc->port_list, links) {
7258                         if (port->targ_port < g * softc->port_cnt ||
7259                             port->targ_port >= (g + 1) * softc->port_cnt)
7260                                 continue;
7261                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7262                                 continue;
7263                         if (port->status & CTL_PORT_STATUS_HA_SHARED)
7264                                 continue;
7265                         if (ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
7266                                 continue;
7267                         scsi_ulto2b(port->targ_port, tpg_desc->descriptors[pc].
7268                             relative_target_port_identifier);
7269                         pc++;
7270                 }
7271                 tpg_desc->target_port_count = pc;
7272                 tpg_desc = (struct scsi_target_port_group_descriptor *)
7273                     &tpg_desc->descriptors[pc];
7274         }
7275         mtx_unlock(&softc->ctl_lock);
7276
7277         ctl_set_success(ctsio);
7278         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7279         ctsio->be_move_done = ctl_config_move_done;
7280         ctl_datamove((union ctl_io *)ctsio);
7281         return(retval);
7282 }
7283
7284 int
7285 ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7286 {
7287         struct ctl_lun *lun;
7288         struct scsi_report_supported_opcodes *cdb;
7289         const struct ctl_cmd_entry *entry, *sentry;
7290         struct scsi_report_supported_opcodes_all *all;
7291         struct scsi_report_supported_opcodes_descr *descr;
7292         struct scsi_report_supported_opcodes_one *one;
7293         int retval;
7294         int alloc_len, total_len;
7295         int opcode, service_action, i, j, num;
7296
7297         CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7298
7299         cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7300         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7301
7302         retval = CTL_RETVAL_COMPLETE;
7303
7304         opcode = cdb->requested_opcode;
7305         service_action = scsi_2btoul(cdb->requested_service_action);
7306         switch (cdb->options & RSO_OPTIONS_MASK) {
7307         case RSO_OPTIONS_ALL:
7308                 num = 0;
7309                 for (i = 0; i < 256; i++) {
7310                         entry = &ctl_cmd_table[i];
7311                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7312                                 for (j = 0; j < 32; j++) {
7313                                         sentry = &((const struct ctl_cmd_entry *)
7314                                             entry->execute)[j];
7315                                         if (ctl_cmd_applicable(
7316                                             lun->be_lun->lun_type, sentry))
7317                                                 num++;
7318                                 }
7319                         } else {
7320                                 if (ctl_cmd_applicable(lun->be_lun->lun_type,
7321                                     entry))
7322                                         num++;
7323                         }
7324                 }
7325                 total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7326                     num * sizeof(struct scsi_report_supported_opcodes_descr);
7327                 break;
7328         case RSO_OPTIONS_OC:
7329                 if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7330                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7331                                               /*sks_valid*/ 1,
7332                                               /*command*/ 1,
7333                                               /*field*/ 2,
7334                                               /*bit_valid*/ 1,
7335                                               /*bit*/ 2);
7336                         ctl_done((union ctl_io *)ctsio);
7337                         return (CTL_RETVAL_COMPLETE);
7338                 }
7339                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7340                 break;
7341         case RSO_OPTIONS_OC_SA:
7342                 if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7343                     service_action >= 32) {
7344                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7345                                               /*sks_valid*/ 1,
7346                                               /*command*/ 1,
7347                                               /*field*/ 2,
7348                                               /*bit_valid*/ 1,
7349                                               /*bit*/ 2);
7350                         ctl_done((union ctl_io *)ctsio);
7351                         return (CTL_RETVAL_COMPLETE);
7352                 }
7353                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7354                 break;
7355         default:
7356                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7357                                       /*sks_valid*/ 1,
7358                                       /*command*/ 1,
7359                                       /*field*/ 2,
7360                                       /*bit_valid*/ 1,
7361                                       /*bit*/ 2);
7362                 ctl_done((union ctl_io *)ctsio);
7363                 return (CTL_RETVAL_COMPLETE);
7364         }
7365
7366         alloc_len = scsi_4btoul(cdb->length);
7367
7368         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7369
7370         ctsio->kern_sg_entries = 0;
7371
7372         if (total_len < alloc_len) {
7373                 ctsio->residual = alloc_len - total_len;
7374                 ctsio->kern_data_len = total_len;
7375                 ctsio->kern_total_len = total_len;
7376         } else {
7377                 ctsio->residual = 0;
7378                 ctsio->kern_data_len = alloc_len;
7379                 ctsio->kern_total_len = alloc_len;
7380         }
7381         ctsio->kern_data_resid = 0;
7382         ctsio->kern_rel_offset = 0;
7383
7384         switch (cdb->options & RSO_OPTIONS_MASK) {
7385         case RSO_OPTIONS_ALL:
7386                 all = (struct scsi_report_supported_opcodes_all *)
7387                     ctsio->kern_data_ptr;
7388                 num = 0;
7389                 for (i = 0; i < 256; i++) {
7390                         entry = &ctl_cmd_table[i];
7391                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7392                                 for (j = 0; j < 32; j++) {
7393                                         sentry = &((const struct ctl_cmd_entry *)
7394                                             entry->execute)[j];
7395                                         if (!ctl_cmd_applicable(
7396                                             lun->be_lun->lun_type, sentry))
7397                                                 continue;
7398                                         descr = &all->descr[num++];
7399                                         descr->opcode = i;
7400                                         scsi_ulto2b(j, descr->service_action);
7401                                         descr->flags = RSO_SERVACTV;
7402                                         scsi_ulto2b(sentry->length,
7403                                             descr->cdb_length);
7404                                 }
7405                         } else {
7406                                 if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7407                                     entry))
7408                                         continue;
7409                                 descr = &all->descr[num++];
7410                                 descr->opcode = i;
7411                                 scsi_ulto2b(0, descr->service_action);
7412                                 descr->flags = 0;
7413                                 scsi_ulto2b(entry->length, descr->cdb_length);
7414                         }
7415                 }
7416                 scsi_ulto4b(
7417                     num * sizeof(struct scsi_report_supported_opcodes_descr),
7418                     all->length);
7419                 break;
7420         case RSO_OPTIONS_OC:
7421                 one = (struct scsi_report_supported_opcodes_one *)
7422                     ctsio->kern_data_ptr;
7423                 entry = &ctl_cmd_table[opcode];
7424                 goto fill_one;
7425         case RSO_OPTIONS_OC_SA:
7426                 one = (struct scsi_report_supported_opcodes_one *)
7427                     ctsio->kern_data_ptr;
7428                 entry = &ctl_cmd_table[opcode];
7429                 entry = &((const struct ctl_cmd_entry *)
7430                     entry->execute)[service_action];
7431 fill_one:
7432                 if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7433                         one->support = 3;
7434                         scsi_ulto2b(entry->length, one->cdb_length);
7435                         one->cdb_usage[0] = opcode;
7436                         memcpy(&one->cdb_usage[1], entry->usage,
7437                             entry->length - 1);
7438                 } else
7439                         one->support = 1;
7440                 break;
7441         }
7442
7443         ctl_set_success(ctsio);
7444         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7445         ctsio->be_move_done = ctl_config_move_done;
7446         ctl_datamove((union ctl_io *)ctsio);
7447         return(retval);
7448 }
7449
7450 int
7451 ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7452 {
7453         struct scsi_report_supported_tmf *cdb;
7454         struct scsi_report_supported_tmf_data *data;
7455         int retval;
7456         int alloc_len, total_len;
7457
7458         CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7459
7460         cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7461
7462         retval = CTL_RETVAL_COMPLETE;
7463
7464         total_len = sizeof(struct scsi_report_supported_tmf_data);
7465         alloc_len = scsi_4btoul(cdb->length);
7466
7467         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7468
7469         ctsio->kern_sg_entries = 0;
7470
7471         if (total_len < alloc_len) {
7472                 ctsio->residual = alloc_len - total_len;
7473                 ctsio->kern_data_len = total_len;
7474                 ctsio->kern_total_len = total_len;
7475         } else {
7476                 ctsio->residual = 0;
7477                 ctsio->kern_data_len = alloc_len;
7478                 ctsio->kern_total_len = alloc_len;
7479         }
7480         ctsio->kern_data_resid = 0;
7481         ctsio->kern_rel_offset = 0;
7482
7483         data = (struct scsi_report_supported_tmf_data *)ctsio->kern_data_ptr;
7484         data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_QTS |
7485             RST_TRS;
7486         data->byte2 |= RST_QAES | RST_QTSS | RST_ITNRS;
7487
7488         ctl_set_success(ctsio);
7489         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7490         ctsio->be_move_done = ctl_config_move_done;
7491         ctl_datamove((union ctl_io *)ctsio);
7492         return (retval);
7493 }
7494
7495 int
7496 ctl_report_timestamp(struct ctl_scsiio *ctsio)
7497 {
7498         struct scsi_report_timestamp *cdb;
7499         struct scsi_report_timestamp_data *data;
7500         struct timeval tv;
7501         int64_t timestamp;
7502         int retval;
7503         int alloc_len, total_len;
7504
7505         CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7506
7507         cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7508
7509         retval = CTL_RETVAL_COMPLETE;
7510
7511         total_len = sizeof(struct scsi_report_timestamp_data);
7512         alloc_len = scsi_4btoul(cdb->length);
7513
7514         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7515
7516         ctsio->kern_sg_entries = 0;
7517
7518         if (total_len < alloc_len) {
7519                 ctsio->residual = alloc_len - total_len;
7520                 ctsio->kern_data_len = total_len;
7521                 ctsio->kern_total_len = total_len;
7522         } else {
7523                 ctsio->residual = 0;
7524                 ctsio->kern_data_len = alloc_len;
7525                 ctsio->kern_total_len = alloc_len;
7526         }
7527         ctsio->kern_data_resid = 0;
7528         ctsio->kern_rel_offset = 0;
7529
7530         data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7531         scsi_ulto2b(sizeof(*data) - 2, data->length);
7532         data->origin = RTS_ORIG_OUTSIDE;
7533         getmicrotime(&tv);
7534         timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7535         scsi_ulto4b(timestamp >> 16, data->timestamp);
7536         scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7537
7538         ctl_set_success(ctsio);
7539         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7540         ctsio->be_move_done = ctl_config_move_done;
7541         ctl_datamove((union ctl_io *)ctsio);
7542         return (retval);
7543 }
7544
7545 int
7546 ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7547 {
7548         struct scsi_per_res_in *cdb;
7549         int alloc_len, total_len = 0;
7550         /* struct scsi_per_res_in_rsrv in_data; */
7551         struct ctl_lun *lun;
7552         struct ctl_softc *softc;
7553         uint64_t key;
7554
7555         CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7556
7557         cdb = (struct scsi_per_res_in *)ctsio->cdb;
7558
7559         alloc_len = scsi_2btoul(cdb->length);
7560
7561         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7562         softc = lun->ctl_softc;
7563
7564 retry:
7565         mtx_lock(&lun->lun_lock);
7566         switch (cdb->action) {
7567         case SPRI_RK: /* read keys */
7568                 total_len = sizeof(struct scsi_per_res_in_keys) +
7569                         lun->pr_key_count *
7570                         sizeof(struct scsi_per_res_key);
7571                 break;
7572         case SPRI_RR: /* read reservation */
7573                 if (lun->flags & CTL_LUN_PR_RESERVED)
7574                         total_len = sizeof(struct scsi_per_res_in_rsrv);
7575                 else
7576                         total_len = sizeof(struct scsi_per_res_in_header);
7577                 break;
7578         case SPRI_RC: /* report capabilities */
7579                 total_len = sizeof(struct scsi_per_res_cap);
7580                 break;
7581         case SPRI_RS: /* read full status */
7582                 total_len = sizeof(struct scsi_per_res_in_header) +
7583                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7584                     lun->pr_key_count;
7585                 break;
7586         default:
7587                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
7588         }
7589         mtx_unlock(&lun->lun_lock);
7590
7591         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7592
7593         if (total_len < alloc_len) {
7594                 ctsio->residual = alloc_len - total_len;
7595                 ctsio->kern_data_len = total_len;
7596                 ctsio->kern_total_len = total_len;
7597         } else {
7598                 ctsio->residual = 0;
7599                 ctsio->kern_data_len = alloc_len;
7600                 ctsio->kern_total_len = alloc_len;
7601         }
7602
7603         ctsio->kern_data_resid = 0;
7604         ctsio->kern_rel_offset = 0;
7605         ctsio->kern_sg_entries = 0;
7606
7607         mtx_lock(&lun->lun_lock);
7608         switch (cdb->action) {
7609         case SPRI_RK: { // read keys
7610         struct scsi_per_res_in_keys *res_keys;
7611                 int i, key_count;
7612
7613                 res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7614
7615                 /*
7616                  * We had to drop the lock to allocate our buffer, which
7617                  * leaves time for someone to come in with another
7618                  * persistent reservation.  (That is unlikely, though,
7619                  * since this should be the only persistent reservation
7620                  * command active right now.)
7621                  */
7622                 if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7623                     (lun->pr_key_count *
7624                      sizeof(struct scsi_per_res_key)))){
7625                         mtx_unlock(&lun->lun_lock);
7626                         free(ctsio->kern_data_ptr, M_CTL);
7627                         printf("%s: reservation length changed, retrying\n",
7628                                __func__);
7629                         goto retry;
7630                 }
7631
7632                 scsi_ulto4b(lun->pr_generation, res_keys->header.generation);
7633
7634                 scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7635                              lun->pr_key_count, res_keys->header.length);
7636
7637                 for (i = 0, key_count = 0; i < CTL_MAX_INITIATORS; i++) {
7638                         if ((key = ctl_get_prkey(lun, i)) == 0)
7639                                 continue;
7640
7641                         /*
7642                          * We used lun->pr_key_count to calculate the
7643                          * size to allocate.  If it turns out the number of
7644                          * initiators with the registered flag set is
7645                          * larger than that (i.e. they haven't been kept in
7646                          * sync), we've got a problem.
7647                          */
7648                         if (key_count >= lun->pr_key_count) {
7649                                 key_count++;
7650                                 continue;
7651                         }
7652                         scsi_u64to8b(key, res_keys->keys[key_count].key);
7653                         key_count++;
7654                 }
7655                 break;
7656         }
7657         case SPRI_RR: { // read reservation
7658                 struct scsi_per_res_in_rsrv *res;
7659                 int tmp_len, header_only;
7660
7661                 res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7662
7663                 scsi_ulto4b(lun->pr_generation, res->header.generation);
7664
7665                 if (lun->flags & CTL_LUN_PR_RESERVED)
7666                 {
7667                         tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7668                         scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7669                                     res->header.length);
7670                         header_only = 0;
7671                 } else {
7672                         tmp_len = sizeof(struct scsi_per_res_in_header);
7673                         scsi_ulto4b(0, res->header.length);
7674                         header_only = 1;
7675                 }
7676
7677                 /*
7678                  * We had to drop the lock to allocate our buffer, which
7679                  * leaves time for someone to come in with another
7680                  * persistent reservation.  (That is unlikely, though,
7681                  * since this should be the only persistent reservation
7682                  * command active right now.)
7683                  */
7684                 if (tmp_len != total_len) {
7685                         mtx_unlock(&lun->lun_lock);
7686                         free(ctsio->kern_data_ptr, M_CTL);
7687                         printf("%s: reservation status changed, retrying\n",
7688                                __func__);
7689                         goto retry;
7690                 }
7691
7692                 /*
7693                  * No reservation held, so we're done.
7694                  */
7695                 if (header_only != 0)
7696                         break;
7697
7698                 /*
7699                  * If the registration is an All Registrants type, the key
7700                  * is 0, since it doesn't really matter.
7701                  */
7702                 if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7703                         scsi_u64to8b(ctl_get_prkey(lun, lun->pr_res_idx),
7704                             res->data.reservation);
7705                 }
7706                 res->data.scopetype = lun->pr_res_type;
7707                 break;
7708         }
7709         case SPRI_RC:     //report capabilities
7710         {
7711                 struct scsi_per_res_cap *res_cap;
7712                 uint16_t type_mask;
7713
7714                 res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7715                 scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7716                 res_cap->flags1 = SPRI_CRH;
7717                 res_cap->flags2 = SPRI_TMV | SPRI_ALLOW_5;
7718                 type_mask = SPRI_TM_WR_EX_AR |
7719                             SPRI_TM_EX_AC_RO |
7720                             SPRI_TM_WR_EX_RO |
7721                             SPRI_TM_EX_AC |
7722                             SPRI_TM_WR_EX |
7723                             SPRI_TM_EX_AC_AR;
7724                 scsi_ulto2b(type_mask, res_cap->type_mask);
7725                 break;
7726         }
7727         case SPRI_RS: { // read full status
7728                 struct scsi_per_res_in_full *res_status;
7729                 struct scsi_per_res_in_full_desc *res_desc;
7730                 struct ctl_port *port;
7731                 int i, len;
7732
7733                 res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7734
7735                 /*
7736                  * We had to drop the lock to allocate our buffer, which
7737                  * leaves time for someone to come in with another
7738                  * persistent reservation.  (That is unlikely, though,
7739                  * since this should be the only persistent reservation
7740                  * command active right now.)
7741                  */
7742                 if (total_len < (sizeof(struct scsi_per_res_in_header) +
7743                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7744                      lun->pr_key_count)){
7745                         mtx_unlock(&lun->lun_lock);
7746                         free(ctsio->kern_data_ptr, M_CTL);
7747                         printf("%s: reservation length changed, retrying\n",
7748                                __func__);
7749                         goto retry;
7750                 }
7751
7752                 scsi_ulto4b(lun->pr_generation, res_status->header.generation);
7753
7754                 res_desc = &res_status->desc[0];
7755                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7756                         if ((key = ctl_get_prkey(lun, i)) == 0)
7757                                 continue;
7758
7759                         scsi_u64to8b(key, res_desc->res_key.key);
7760                         if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7761                             (lun->pr_res_idx == i ||
7762                              lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7763                                 res_desc->flags = SPRI_FULL_R_HOLDER;
7764                                 res_desc->scopetype = lun->pr_res_type;
7765                         }
7766                         scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7767                             res_desc->rel_trgt_port_id);
7768                         len = 0;
7769                         port = softc->ctl_ports[i / CTL_MAX_INIT_PER_PORT];
7770                         if (port != NULL)
7771                                 len = ctl_create_iid(port,
7772                                     i % CTL_MAX_INIT_PER_PORT,
7773                                     res_desc->transport_id);
7774                         scsi_ulto4b(len, res_desc->additional_length);
7775                         res_desc = (struct scsi_per_res_in_full_desc *)
7776                             &res_desc->transport_id[len];
7777                 }
7778                 scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7779                     res_status->header.length);
7780                 break;
7781         }
7782         default:
7783                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
7784         }
7785         mtx_unlock(&lun->lun_lock);
7786
7787         ctl_set_success(ctsio);
7788         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7789         ctsio->be_move_done = ctl_config_move_done;
7790         ctl_datamove((union ctl_io *)ctsio);
7791         return (CTL_RETVAL_COMPLETE);
7792 }
7793
7794 /*
7795  * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7796  * it should return.
7797  */
7798 static int
7799 ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7800                 uint64_t sa_res_key, uint8_t type, uint32_t residx,
7801                 struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7802                 struct scsi_per_res_out_parms* param)
7803 {
7804         union ctl_ha_msg persis_io;
7805         int i;
7806
7807         mtx_lock(&lun->lun_lock);
7808         if (sa_res_key == 0) {
7809                 if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7810                         /* validate scope and type */
7811                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7812                              SPR_LU_SCOPE) {
7813                                 mtx_unlock(&lun->lun_lock);
7814                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7815                                                       /*sks_valid*/ 1,
7816                                                       /*command*/ 1,
7817                                                       /*field*/ 2,
7818                                                       /*bit_valid*/ 1,
7819                                                       /*bit*/ 4);
7820                                 ctl_done((union ctl_io *)ctsio);
7821                                 return (1);
7822                         }
7823
7824                         if (type>8 || type==2 || type==4 || type==0) {
7825                                 mtx_unlock(&lun->lun_lock);
7826                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7827                                                       /*sks_valid*/ 1,
7828                                                       /*command*/ 1,
7829                                                       /*field*/ 2,
7830                                                       /*bit_valid*/ 1,
7831                                                       /*bit*/ 0);
7832                                 ctl_done((union ctl_io *)ctsio);
7833                                 return (1);
7834                         }
7835
7836                         /*
7837                          * Unregister everybody else and build UA for
7838                          * them
7839                          */
7840                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
7841                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
7842                                         continue;
7843
7844                                 ctl_clr_prkey(lun, i);
7845                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7846                         }
7847                         lun->pr_key_count = 1;
7848                         lun->pr_res_type = type;
7849                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
7850                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
7851                                 lun->pr_res_idx = residx;
7852                         lun->pr_generation++;
7853                         mtx_unlock(&lun->lun_lock);
7854
7855                         /* send msg to other side */
7856                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7857                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7858                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7859                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
7860                         persis_io.pr.pr_info.res_type = type;
7861                         memcpy(persis_io.pr.pr_info.sa_res_key,
7862                                param->serv_act_res_key,
7863                                sizeof(param->serv_act_res_key));
7864                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7865                             sizeof(persis_io.pr), M_WAITOK);
7866                 } else {
7867                         /* not all registrants */
7868                         mtx_unlock(&lun->lun_lock);
7869                         free(ctsio->kern_data_ptr, M_CTL);
7870                         ctl_set_invalid_field(ctsio,
7871                                               /*sks_valid*/ 1,
7872                                               /*command*/ 0,
7873                                               /*field*/ 8,
7874                                               /*bit_valid*/ 0,
7875                                               /*bit*/ 0);
7876                         ctl_done((union ctl_io *)ctsio);
7877                         return (1);
7878                 }
7879         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7880                 || !(lun->flags & CTL_LUN_PR_RESERVED)) {
7881                 int found = 0;
7882
7883                 if (res_key == sa_res_key) {
7884                         /* special case */
7885                         /*
7886                          * The spec implies this is not good but doesn't
7887                          * say what to do. There are two choices either
7888                          * generate a res conflict or check condition
7889                          * with illegal field in parameter data. Since
7890                          * that is what is done when the sa_res_key is
7891                          * zero I'll take that approach since this has
7892                          * to do with the sa_res_key.
7893                          */
7894                         mtx_unlock(&lun->lun_lock);
7895                         free(ctsio->kern_data_ptr, M_CTL);
7896                         ctl_set_invalid_field(ctsio,
7897                                               /*sks_valid*/ 1,
7898                                               /*command*/ 0,
7899                                               /*field*/ 8,
7900                                               /*bit_valid*/ 0,
7901                                               /*bit*/ 0);
7902                         ctl_done((union ctl_io *)ctsio);
7903                         return (1);
7904                 }
7905
7906                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7907                         if (ctl_get_prkey(lun, i) != sa_res_key)
7908                                 continue;
7909
7910                         found = 1;
7911                         ctl_clr_prkey(lun, i);
7912                         lun->pr_key_count--;
7913                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7914                 }
7915                 if (!found) {
7916                         mtx_unlock(&lun->lun_lock);
7917                         free(ctsio->kern_data_ptr, M_CTL);
7918                         ctl_set_reservation_conflict(ctsio);
7919                         ctl_done((union ctl_io *)ctsio);
7920                         return (CTL_RETVAL_COMPLETE);
7921                 }
7922                 lun->pr_generation++;
7923                 mtx_unlock(&lun->lun_lock);
7924
7925                 /* send msg to other side */
7926                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7927                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7928                 persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7929                 persis_io.pr.pr_info.residx = lun->pr_res_idx;
7930                 persis_io.pr.pr_info.res_type = type;
7931                 memcpy(persis_io.pr.pr_info.sa_res_key,
7932                        param->serv_act_res_key,
7933                        sizeof(param->serv_act_res_key));
7934                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7935                     sizeof(persis_io.pr), M_WAITOK);
7936         } else {
7937                 /* Reserved but not all registrants */
7938                 /* sa_res_key is res holder */
7939                 if (sa_res_key == ctl_get_prkey(lun, lun->pr_res_idx)) {
7940                         /* validate scope and type */
7941                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7942                              SPR_LU_SCOPE) {
7943                                 mtx_unlock(&lun->lun_lock);
7944                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7945                                                       /*sks_valid*/ 1,
7946                                                       /*command*/ 1,
7947                                                       /*field*/ 2,
7948                                                       /*bit_valid*/ 1,
7949                                                       /*bit*/ 4);
7950                                 ctl_done((union ctl_io *)ctsio);
7951                                 return (1);
7952                         }
7953
7954                         if (type>8 || type==2 || type==4 || type==0) {
7955                                 mtx_unlock(&lun->lun_lock);
7956                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7957                                                       /*sks_valid*/ 1,
7958                                                       /*command*/ 1,
7959                                                       /*field*/ 2,
7960                                                       /*bit_valid*/ 1,
7961                                                       /*bit*/ 0);
7962                                 ctl_done((union ctl_io *)ctsio);
7963                                 return (1);
7964                         }
7965
7966                         /*
7967                          * Do the following:
7968                          * if sa_res_key != res_key remove all
7969                          * registrants w/sa_res_key and generate UA
7970                          * for these registrants(Registrations
7971                          * Preempted) if it wasn't an exclusive
7972                          * reservation generate UA(Reservations
7973                          * Preempted) for all other registered nexuses
7974                          * if the type has changed. Establish the new
7975                          * reservation and holder. If res_key and
7976                          * sa_res_key are the same do the above
7977                          * except don't unregister the res holder.
7978                          */
7979
7980                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
7981                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
7982                                         continue;
7983
7984                                 if (sa_res_key == ctl_get_prkey(lun, i)) {
7985                                         ctl_clr_prkey(lun, i);
7986                                         lun->pr_key_count--;
7987                                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7988                                 } else if (type != lun->pr_res_type &&
7989                                     (lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
7990                                      lun->pr_res_type == SPR_TYPE_EX_AC_RO)) {
7991                                         ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
7992                                 }
7993                         }
7994                         lun->pr_res_type = type;
7995                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
7996                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
7997                                 lun->pr_res_idx = residx;
7998                         else
7999                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8000                         lun->pr_generation++;
8001                         mtx_unlock(&lun->lun_lock);
8002
8003                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8004                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8005                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8006                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8007                         persis_io.pr.pr_info.res_type = type;
8008                         memcpy(persis_io.pr.pr_info.sa_res_key,
8009                                param->serv_act_res_key,
8010                                sizeof(param->serv_act_res_key));
8011                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8012                             sizeof(persis_io.pr), M_WAITOK);
8013                 } else {
8014                         /*
8015                          * sa_res_key is not the res holder just
8016                          * remove registrants
8017                          */
8018                         int found=0;
8019
8020                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8021                                 if (sa_res_key != ctl_get_prkey(lun, i))
8022                                         continue;
8023
8024                                 found = 1;
8025                                 ctl_clr_prkey(lun, i);
8026                                 lun->pr_key_count--;
8027                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8028                         }
8029
8030                         if (!found) {
8031                                 mtx_unlock(&lun->lun_lock);
8032                                 free(ctsio->kern_data_ptr, M_CTL);
8033                                 ctl_set_reservation_conflict(ctsio);
8034                                 ctl_done((union ctl_io *)ctsio);
8035                                 return (1);
8036                         }
8037                         lun->pr_generation++;
8038                         mtx_unlock(&lun->lun_lock);
8039
8040                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8041                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8042                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8043                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8044                         persis_io.pr.pr_info.res_type = type;
8045                         memcpy(persis_io.pr.pr_info.sa_res_key,
8046                                param->serv_act_res_key,
8047                                sizeof(param->serv_act_res_key));
8048                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8049                             sizeof(persis_io.pr), M_WAITOK);
8050                 }
8051         }
8052         return (0);
8053 }
8054
8055 static void
8056 ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
8057 {
8058         uint64_t sa_res_key;
8059         int i;
8060
8061         sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
8062
8063         if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8064          || lun->pr_res_idx == CTL_PR_NO_RESERVATION
8065          || sa_res_key != ctl_get_prkey(lun, lun->pr_res_idx)) {
8066                 if (sa_res_key == 0) {
8067                         /*
8068                          * Unregister everybody else and build UA for
8069                          * them
8070                          */
8071                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
8072                                 if (i == msg->pr.pr_info.residx ||
8073                                     ctl_get_prkey(lun, i) == 0)
8074                                         continue;
8075
8076                                 ctl_clr_prkey(lun, i);
8077                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8078                         }
8079
8080                         lun->pr_key_count = 1;
8081                         lun->pr_res_type = msg->pr.pr_info.res_type;
8082                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8083                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8084                                 lun->pr_res_idx = msg->pr.pr_info.residx;
8085                 } else {
8086                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8087                                 if (sa_res_key == ctl_get_prkey(lun, i))
8088                                         continue;
8089
8090                                 ctl_clr_prkey(lun, i);
8091                                 lun->pr_key_count--;
8092                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8093                         }
8094                 }
8095         } else {
8096                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8097                         if (i == msg->pr.pr_info.residx ||
8098                             ctl_get_prkey(lun, i) == 0)
8099                                 continue;
8100
8101                         if (sa_res_key == ctl_get_prkey(lun, i)) {
8102                                 ctl_clr_prkey(lun, i);
8103                                 lun->pr_key_count--;
8104                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8105                         } else if (msg->pr.pr_info.res_type != lun->pr_res_type
8106                             && (lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8107                              lun->pr_res_type == SPR_TYPE_EX_AC_RO)) {
8108                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8109                         }
8110                 }
8111                 lun->pr_res_type = msg->pr.pr_info.res_type;
8112                 if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8113                     lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8114                         lun->pr_res_idx = msg->pr.pr_info.residx;
8115                 else
8116                         lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8117         }
8118         lun->pr_generation++;
8119
8120 }
8121
8122
8123 int
8124 ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8125 {
8126         int retval;
8127         u_int32_t param_len;
8128         struct scsi_per_res_out *cdb;
8129         struct ctl_lun *lun;
8130         struct scsi_per_res_out_parms* param;
8131         struct ctl_softc *softc;
8132         uint32_t residx;
8133         uint64_t res_key, sa_res_key, key;
8134         uint8_t type;
8135         union ctl_ha_msg persis_io;
8136         int    i;
8137
8138         CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8139
8140         retval = CTL_RETVAL_COMPLETE;
8141
8142         cdb = (struct scsi_per_res_out *)ctsio->cdb;
8143         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8144         softc = lun->ctl_softc;
8145
8146         /*
8147          * We only support whole-LUN scope.  The scope & type are ignored for
8148          * register, register and ignore existing key and clear.
8149          * We sometimes ignore scope and type on preempts too!!
8150          * Verify reservation type here as well.
8151          */
8152         type = cdb->scope_type & SPR_TYPE_MASK;
8153         if ((cdb->action == SPRO_RESERVE)
8154          || (cdb->action == SPRO_RELEASE)) {
8155                 if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8156                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8157                                               /*sks_valid*/ 1,
8158                                               /*command*/ 1,
8159                                               /*field*/ 2,
8160                                               /*bit_valid*/ 1,
8161                                               /*bit*/ 4);
8162                         ctl_done((union ctl_io *)ctsio);
8163                         return (CTL_RETVAL_COMPLETE);
8164                 }
8165
8166                 if (type>8 || type==2 || type==4 || type==0) {
8167                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8168                                               /*sks_valid*/ 1,
8169                                               /*command*/ 1,
8170                                               /*field*/ 2,
8171                                               /*bit_valid*/ 1,
8172                                               /*bit*/ 0);
8173                         ctl_done((union ctl_io *)ctsio);
8174                         return (CTL_RETVAL_COMPLETE);
8175                 }
8176         }
8177
8178         param_len = scsi_4btoul(cdb->length);
8179
8180         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8181                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8182                 ctsio->kern_data_len = param_len;
8183                 ctsio->kern_total_len = param_len;
8184                 ctsio->kern_data_resid = 0;
8185                 ctsio->kern_rel_offset = 0;
8186                 ctsio->kern_sg_entries = 0;
8187                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8188                 ctsio->be_move_done = ctl_config_move_done;
8189                 ctl_datamove((union ctl_io *)ctsio);
8190
8191                 return (CTL_RETVAL_COMPLETE);
8192         }
8193
8194         param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8195
8196         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
8197         res_key = scsi_8btou64(param->res_key.key);
8198         sa_res_key = scsi_8btou64(param->serv_act_res_key);
8199
8200         /*
8201          * Validate the reservation key here except for SPRO_REG_IGNO
8202          * This must be done for all other service actions
8203          */
8204         if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8205                 mtx_lock(&lun->lun_lock);
8206                 if ((key = ctl_get_prkey(lun, residx)) != 0) {
8207                         if (res_key != key) {
8208                                 /*
8209                                  * The current key passed in doesn't match
8210                                  * the one the initiator previously
8211                                  * registered.
8212                                  */
8213                                 mtx_unlock(&lun->lun_lock);
8214                                 free(ctsio->kern_data_ptr, M_CTL);
8215                                 ctl_set_reservation_conflict(ctsio);
8216                                 ctl_done((union ctl_io *)ctsio);
8217                                 return (CTL_RETVAL_COMPLETE);
8218                         }
8219                 } else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8220                         /*
8221                          * We are not registered
8222                          */
8223                         mtx_unlock(&lun->lun_lock);
8224                         free(ctsio->kern_data_ptr, M_CTL);
8225                         ctl_set_reservation_conflict(ctsio);
8226                         ctl_done((union ctl_io *)ctsio);
8227                         return (CTL_RETVAL_COMPLETE);
8228                 } else if (res_key != 0) {
8229                         /*
8230                          * We are not registered and trying to register but
8231                          * the register key isn't zero.
8232                          */
8233                         mtx_unlock(&lun->lun_lock);
8234                         free(ctsio->kern_data_ptr, M_CTL);
8235                         ctl_set_reservation_conflict(ctsio);
8236                         ctl_done((union ctl_io *)ctsio);
8237                         return (CTL_RETVAL_COMPLETE);
8238                 }
8239                 mtx_unlock(&lun->lun_lock);
8240         }
8241
8242         switch (cdb->action & SPRO_ACTION_MASK) {
8243         case SPRO_REGISTER:
8244         case SPRO_REG_IGNO: {
8245
8246 #if 0
8247                 printf("Registration received\n");
8248 #endif
8249
8250                 /*
8251                  * We don't support any of these options, as we report in
8252                  * the read capabilities request (see
8253                  * ctl_persistent_reserve_in(), above).
8254                  */
8255                 if ((param->flags & SPR_SPEC_I_PT)
8256                  || (param->flags & SPR_ALL_TG_PT)
8257                  || (param->flags & SPR_APTPL)) {
8258                         int bit_ptr;
8259
8260                         if (param->flags & SPR_APTPL)
8261                                 bit_ptr = 0;
8262                         else if (param->flags & SPR_ALL_TG_PT)
8263                                 bit_ptr = 2;
8264                         else /* SPR_SPEC_I_PT */
8265                                 bit_ptr = 3;
8266
8267                         free(ctsio->kern_data_ptr, M_CTL);
8268                         ctl_set_invalid_field(ctsio,
8269                                               /*sks_valid*/ 1,
8270                                               /*command*/ 0,
8271                                               /*field*/ 20,
8272                                               /*bit_valid*/ 1,
8273                                               /*bit*/ bit_ptr);
8274                         ctl_done((union ctl_io *)ctsio);
8275                         return (CTL_RETVAL_COMPLETE);
8276                 }
8277
8278                 mtx_lock(&lun->lun_lock);
8279
8280                 /*
8281                  * The initiator wants to clear the
8282                  * key/unregister.
8283                  */
8284                 if (sa_res_key == 0) {
8285                         if ((res_key == 0
8286                           && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8287                          || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8288                           && ctl_get_prkey(lun, residx) == 0)) {
8289                                 mtx_unlock(&lun->lun_lock);
8290                                 goto done;
8291                         }
8292
8293                         ctl_clr_prkey(lun, residx);
8294                         lun->pr_key_count--;
8295
8296                         if (residx == lun->pr_res_idx) {
8297                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8298                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8299
8300                                 if ((lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8301                                      lun->pr_res_type == SPR_TYPE_EX_AC_RO) &&
8302                                     lun->pr_key_count) {
8303                                         /*
8304                                          * If the reservation is a registrants
8305                                          * only type we need to generate a UA
8306                                          * for other registered inits.  The
8307                                          * sense code should be RESERVATIONS
8308                                          * RELEASED
8309                                          */
8310
8311                                         for (i = softc->init_min; i < softc->init_max; i++){
8312                                                 if (ctl_get_prkey(lun, i) == 0)
8313                                                         continue;
8314                                                 ctl_est_ua(lun, i,
8315                                                     CTL_UA_RES_RELEASE);
8316                                         }
8317                                 }
8318                                 lun->pr_res_type = 0;
8319                         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8320                                 if (lun->pr_key_count==0) {
8321                                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8322                                         lun->pr_res_type = 0;
8323                                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8324                                 }
8325                         }
8326                         lun->pr_generation++;
8327                         mtx_unlock(&lun->lun_lock);
8328
8329                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8330                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8331                         persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8332                         persis_io.pr.pr_info.residx = residx;
8333                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8334                             sizeof(persis_io.pr), M_WAITOK);
8335                 } else /* sa_res_key != 0 */ {
8336
8337                         /*
8338                          * If we aren't registered currently then increment
8339                          * the key count and set the registered flag.
8340                          */
8341                         ctl_alloc_prkey(lun, residx);
8342                         if (ctl_get_prkey(lun, residx) == 0)
8343                                 lun->pr_key_count++;
8344                         ctl_set_prkey(lun, residx, sa_res_key);
8345                         lun->pr_generation++;
8346                         mtx_unlock(&lun->lun_lock);
8347
8348                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8349                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8350                         persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8351                         persis_io.pr.pr_info.residx = residx;
8352                         memcpy(persis_io.pr.pr_info.sa_res_key,
8353                                param->serv_act_res_key,
8354                                sizeof(param->serv_act_res_key));
8355                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8356                             sizeof(persis_io.pr), M_WAITOK);
8357                 }
8358
8359                 break;
8360         }
8361         case SPRO_RESERVE:
8362 #if 0
8363                 printf("Reserve executed type %d\n", type);
8364 #endif
8365                 mtx_lock(&lun->lun_lock);
8366                 if (lun->flags & CTL_LUN_PR_RESERVED) {
8367                         /*
8368                          * if this isn't the reservation holder and it's
8369                          * not a "all registrants" type or if the type is
8370                          * different then we have a conflict
8371                          */
8372                         if ((lun->pr_res_idx != residx
8373                           && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8374                          || lun->pr_res_type != type) {
8375                                 mtx_unlock(&lun->lun_lock);
8376                                 free(ctsio->kern_data_ptr, M_CTL);
8377                                 ctl_set_reservation_conflict(ctsio);
8378                                 ctl_done((union ctl_io *)ctsio);
8379                                 return (CTL_RETVAL_COMPLETE);
8380                         }
8381                         mtx_unlock(&lun->lun_lock);
8382                 } else /* create a reservation */ {
8383                         /*
8384                          * If it's not an "all registrants" type record
8385                          * reservation holder
8386                          */
8387                         if (type != SPR_TYPE_WR_EX_AR
8388                          && type != SPR_TYPE_EX_AC_AR)
8389                                 lun->pr_res_idx = residx; /* Res holder */
8390                         else
8391                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8392
8393                         lun->flags |= CTL_LUN_PR_RESERVED;
8394                         lun->pr_res_type = type;
8395
8396                         mtx_unlock(&lun->lun_lock);
8397
8398                         /* send msg to other side */
8399                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8400                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8401                         persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8402                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8403                         persis_io.pr.pr_info.res_type = type;
8404                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8405                             sizeof(persis_io.pr), M_WAITOK);
8406                 }
8407                 break;
8408
8409         case SPRO_RELEASE:
8410                 mtx_lock(&lun->lun_lock);
8411                 if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8412                         /* No reservation exists return good status */
8413                         mtx_unlock(&lun->lun_lock);
8414                         goto done;
8415                 }
8416                 /*
8417                  * Is this nexus a reservation holder?
8418                  */
8419                 if (lun->pr_res_idx != residx
8420                  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8421                         /*
8422                          * not a res holder return good status but
8423                          * do nothing
8424                          */
8425                         mtx_unlock(&lun->lun_lock);
8426                         goto done;
8427                 }
8428
8429                 if (lun->pr_res_type != type) {
8430                         mtx_unlock(&lun->lun_lock);
8431                         free(ctsio->kern_data_ptr, M_CTL);
8432                         ctl_set_illegal_pr_release(ctsio);
8433                         ctl_done((union ctl_io *)ctsio);
8434                         return (CTL_RETVAL_COMPLETE);
8435                 }
8436
8437                 /* okay to release */
8438                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8439                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8440                 lun->pr_res_type = 0;
8441
8442                 /*
8443                  * If this isn't an exclusive access reservation and NUAR
8444                  * is not set, generate UA for all other registrants.
8445                  */
8446                 if (type != SPR_TYPE_EX_AC && type != SPR_TYPE_WR_EX &&
8447                     (lun->MODE_CTRL.queue_flags & SCP_NUAR) == 0) {
8448                         for (i = softc->init_min; i < softc->init_max; i++) {
8449                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8450                                         continue;
8451                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8452                         }
8453                 }
8454                 mtx_unlock(&lun->lun_lock);
8455
8456                 /* Send msg to other side */
8457                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8458                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8459                 persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8460                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8461                      sizeof(persis_io.pr), M_WAITOK);
8462                 break;
8463
8464         case SPRO_CLEAR:
8465                 /* send msg to other side */
8466
8467                 mtx_lock(&lun->lun_lock);
8468                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8469                 lun->pr_res_type = 0;
8470                 lun->pr_key_count = 0;
8471                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8472
8473                 ctl_clr_prkey(lun, residx);
8474                 for (i = 0; i < CTL_MAX_INITIATORS; i++)
8475                         if (ctl_get_prkey(lun, i) != 0) {
8476                                 ctl_clr_prkey(lun, i);
8477                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8478                         }
8479                 lun->pr_generation++;
8480                 mtx_unlock(&lun->lun_lock);
8481
8482                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8483                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8484                 persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8485                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8486                      sizeof(persis_io.pr), M_WAITOK);
8487                 break;
8488
8489         case SPRO_PREEMPT:
8490         case SPRO_PRE_ABO: {
8491                 int nretval;
8492
8493                 nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8494                                           residx, ctsio, cdb, param);
8495                 if (nretval != 0)
8496                         return (CTL_RETVAL_COMPLETE);
8497                 break;
8498         }
8499         default:
8500                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
8501         }
8502
8503 done:
8504         free(ctsio->kern_data_ptr, M_CTL);
8505         ctl_set_success(ctsio);
8506         ctl_done((union ctl_io *)ctsio);
8507
8508         return (retval);
8509 }
8510
8511 /*
8512  * This routine is for handling a message from the other SC pertaining to
8513  * persistent reserve out. All the error checking will have been done
8514  * so only perorming the action need be done here to keep the two
8515  * in sync.
8516  */
8517 static void
8518 ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8519 {
8520         struct ctl_softc *softc = control_softc;
8521         struct ctl_lun *lun;
8522         int i;
8523         uint32_t residx, targ_lun;
8524
8525         targ_lun = msg->hdr.nexus.targ_mapped_lun;
8526         mtx_lock(&softc->ctl_lock);
8527         if ((targ_lun >= CTL_MAX_LUNS) ||
8528             ((lun = softc->ctl_luns[targ_lun]) == NULL)) {
8529                 mtx_unlock(&softc->ctl_lock);
8530                 return;
8531         }
8532         mtx_lock(&lun->lun_lock);
8533         mtx_unlock(&softc->ctl_lock);
8534         if (lun->flags & CTL_LUN_DISABLED) {
8535                 mtx_unlock(&lun->lun_lock);
8536                 return;
8537         }
8538         residx = ctl_get_initindex(&msg->hdr.nexus);
8539         switch(msg->pr.pr_info.action) {
8540         case CTL_PR_REG_KEY:
8541                 ctl_alloc_prkey(lun, msg->pr.pr_info.residx);
8542                 if (ctl_get_prkey(lun, msg->pr.pr_info.residx) == 0)
8543                         lun->pr_key_count++;
8544                 ctl_set_prkey(lun, msg->pr.pr_info.residx,
8545                     scsi_8btou64(msg->pr.pr_info.sa_res_key));
8546                 lun->pr_generation++;
8547                 break;
8548
8549         case CTL_PR_UNREG_KEY:
8550                 ctl_clr_prkey(lun, msg->pr.pr_info.residx);
8551                 lun->pr_key_count--;
8552
8553                 /* XXX Need to see if the reservation has been released */
8554                 /* if so do we need to generate UA? */
8555                 if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8556                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8557                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8558
8559                         if ((lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8560                              lun->pr_res_type == SPR_TYPE_EX_AC_RO) &&
8561                             lun->pr_key_count) {
8562                                 /*
8563                                  * If the reservation is a registrants
8564                                  * only type we need to generate a UA
8565                                  * for other registered inits.  The
8566                                  * sense code should be RESERVATIONS
8567                                  * RELEASED
8568                                  */
8569
8570                                 for (i = softc->init_min; i < softc->init_max; i++) {
8571                                         if (ctl_get_prkey(lun, i) == 0)
8572                                                 continue;
8573
8574                                         ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8575                                 }
8576                         }
8577                         lun->pr_res_type = 0;
8578                 } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8579                         if (lun->pr_key_count==0) {
8580                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8581                                 lun->pr_res_type = 0;
8582                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8583                         }
8584                 }
8585                 lun->pr_generation++;
8586                 break;
8587
8588         case CTL_PR_RESERVE:
8589                 lun->flags |= CTL_LUN_PR_RESERVED;
8590                 lun->pr_res_type = msg->pr.pr_info.res_type;
8591                 lun->pr_res_idx = msg->pr.pr_info.residx;
8592
8593                 break;
8594
8595         case CTL_PR_RELEASE:
8596                 /*
8597                  * If this isn't an exclusive access reservation and NUAR
8598                  * is not set, generate UA for all other registrants.
8599                  */
8600                 if (lun->pr_res_type != SPR_TYPE_EX_AC &&
8601                     lun->pr_res_type != SPR_TYPE_WR_EX &&
8602                     (lun->MODE_CTRL.queue_flags & SCP_NUAR) == 0) {
8603                         for (i = softc->init_min; i < softc->init_max; i++)
8604                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8605                                         continue;
8606                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8607                 }
8608
8609                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8610                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8611                 lun->pr_res_type = 0;
8612                 break;
8613
8614         case CTL_PR_PREEMPT:
8615                 ctl_pro_preempt_other(lun, msg);
8616                 break;
8617         case CTL_PR_CLEAR:
8618                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8619                 lun->pr_res_type = 0;
8620                 lun->pr_key_count = 0;
8621                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8622
8623                 for (i=0; i < CTL_MAX_INITIATORS; i++) {
8624                         if (ctl_get_prkey(lun, i) == 0)
8625                                 continue;
8626                         ctl_clr_prkey(lun, i);
8627                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8628                 }
8629                 lun->pr_generation++;
8630                 break;
8631         }
8632
8633         mtx_unlock(&lun->lun_lock);
8634 }
8635
8636 int
8637 ctl_read_write(struct ctl_scsiio *ctsio)
8638 {
8639         struct ctl_lun *lun;
8640         struct ctl_lba_len_flags *lbalen;
8641         uint64_t lba;
8642         uint32_t num_blocks;
8643         int flags, retval;
8644         int isread;
8645
8646         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8647
8648         CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8649
8650         flags = 0;
8651         isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8652               || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8653         switch (ctsio->cdb[0]) {
8654         case READ_6:
8655         case WRITE_6: {
8656                 struct scsi_rw_6 *cdb;
8657
8658                 cdb = (struct scsi_rw_6 *)ctsio->cdb;
8659
8660                 lba = scsi_3btoul(cdb->addr);
8661                 /* only 5 bits are valid in the most significant address byte */
8662                 lba &= 0x1fffff;
8663                 num_blocks = cdb->length;
8664                 /*
8665                  * This is correct according to SBC-2.
8666                  */
8667                 if (num_blocks == 0)
8668                         num_blocks = 256;
8669                 break;
8670         }
8671         case READ_10:
8672         case WRITE_10: {
8673                 struct scsi_rw_10 *cdb;
8674
8675                 cdb = (struct scsi_rw_10 *)ctsio->cdb;
8676                 if (cdb->byte2 & SRW10_FUA)
8677                         flags |= CTL_LLF_FUA;
8678                 if (cdb->byte2 & SRW10_DPO)
8679                         flags |= CTL_LLF_DPO;
8680                 lba = scsi_4btoul(cdb->addr);
8681                 num_blocks = scsi_2btoul(cdb->length);
8682                 break;
8683         }
8684         case WRITE_VERIFY_10: {
8685                 struct scsi_write_verify_10 *cdb;
8686
8687                 cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8688                 flags |= CTL_LLF_FUA;
8689                 if (cdb->byte2 & SWV_DPO)
8690                         flags |= CTL_LLF_DPO;
8691                 lba = scsi_4btoul(cdb->addr);
8692                 num_blocks = scsi_2btoul(cdb->length);
8693                 break;
8694         }
8695         case READ_12:
8696         case WRITE_12: {
8697                 struct scsi_rw_12 *cdb;
8698
8699                 cdb = (struct scsi_rw_12 *)ctsio->cdb;
8700                 if (cdb->byte2 & SRW12_FUA)
8701                         flags |= CTL_LLF_FUA;
8702                 if (cdb->byte2 & SRW12_DPO)
8703                         flags |= CTL_LLF_DPO;
8704                 lba = scsi_4btoul(cdb->addr);
8705                 num_blocks = scsi_4btoul(cdb->length);
8706                 break;
8707         }
8708         case WRITE_VERIFY_12: {
8709                 struct scsi_write_verify_12 *cdb;
8710
8711                 cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8712                 flags |= CTL_LLF_FUA;
8713                 if (cdb->byte2 & SWV_DPO)
8714                         flags |= CTL_LLF_DPO;
8715                 lba = scsi_4btoul(cdb->addr);
8716                 num_blocks = scsi_4btoul(cdb->length);
8717                 break;
8718         }
8719         case READ_16:
8720         case WRITE_16: {
8721                 struct scsi_rw_16 *cdb;
8722
8723                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
8724                 if (cdb->byte2 & SRW12_FUA)
8725                         flags |= CTL_LLF_FUA;
8726                 if (cdb->byte2 & SRW12_DPO)
8727                         flags |= CTL_LLF_DPO;
8728                 lba = scsi_8btou64(cdb->addr);
8729                 num_blocks = scsi_4btoul(cdb->length);
8730                 break;
8731         }
8732         case WRITE_ATOMIC_16: {
8733                 struct scsi_write_atomic_16 *cdb;
8734
8735                 if (lun->be_lun->atomicblock == 0) {
8736                         ctl_set_invalid_opcode(ctsio);
8737                         ctl_done((union ctl_io *)ctsio);
8738                         return (CTL_RETVAL_COMPLETE);
8739                 }
8740
8741                 cdb = (struct scsi_write_atomic_16 *)ctsio->cdb;
8742                 if (cdb->byte2 & SRW12_FUA)
8743                         flags |= CTL_LLF_FUA;
8744                 if (cdb->byte2 & SRW12_DPO)
8745                         flags |= CTL_LLF_DPO;
8746                 lba = scsi_8btou64(cdb->addr);
8747                 num_blocks = scsi_2btoul(cdb->length);
8748                 if (num_blocks > lun->be_lun->atomicblock) {
8749                         ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
8750                             /*command*/ 1, /*field*/ 12, /*bit_valid*/ 0,
8751                             /*bit*/ 0);
8752                         ctl_done((union ctl_io *)ctsio);
8753                         return (CTL_RETVAL_COMPLETE);
8754                 }
8755                 break;
8756         }
8757         case WRITE_VERIFY_16: {
8758                 struct scsi_write_verify_16 *cdb;
8759
8760                 cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
8761                 flags |= CTL_LLF_FUA;
8762                 if (cdb->byte2 & SWV_DPO)
8763                         flags |= CTL_LLF_DPO;
8764                 lba = scsi_8btou64(cdb->addr);
8765                 num_blocks = scsi_4btoul(cdb->length);
8766                 break;
8767         }
8768         default:
8769                 /*
8770                  * We got a command we don't support.  This shouldn't
8771                  * happen, commands should be filtered out above us.
8772                  */
8773                 ctl_set_invalid_opcode(ctsio);
8774                 ctl_done((union ctl_io *)ctsio);
8775
8776                 return (CTL_RETVAL_COMPLETE);
8777                 break; /* NOTREACHED */
8778         }
8779
8780         /*
8781          * The first check is to make sure we're in bounds, the second
8782          * check is to catch wrap-around problems.  If the lba + num blocks
8783          * is less than the lba, then we've wrapped around and the block
8784          * range is invalid anyway.
8785          */
8786         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8787          || ((lba + num_blocks) < lba)) {
8788                 ctl_set_lba_out_of_range(ctsio);
8789                 ctl_done((union ctl_io *)ctsio);
8790                 return (CTL_RETVAL_COMPLETE);
8791         }
8792
8793         /*
8794          * According to SBC-3, a transfer length of 0 is not an error.
8795          * Note that this cannot happen with WRITE(6) or READ(6), since 0
8796          * translates to 256 blocks for those commands.
8797          */
8798         if (num_blocks == 0) {
8799                 ctl_set_success(ctsio);
8800                 ctl_done((union ctl_io *)ctsio);
8801                 return (CTL_RETVAL_COMPLETE);
8802         }
8803
8804         /* Set FUA and/or DPO if caches are disabled. */
8805         if (isread) {
8806                 if ((lun->MODE_CACHING.flags1 & SCP_RCD) != 0)
8807                         flags |= CTL_LLF_FUA | CTL_LLF_DPO;
8808         } else {
8809                 if ((lun->MODE_CACHING.flags1 & SCP_WCE) == 0)
8810                         flags |= CTL_LLF_FUA;
8811         }
8812
8813         lbalen = (struct ctl_lba_len_flags *)
8814             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8815         lbalen->lba = lba;
8816         lbalen->len = num_blocks;
8817         lbalen->flags = (isread ? CTL_LLF_READ : CTL_LLF_WRITE) | flags;
8818
8819         ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8820         ctsio->kern_rel_offset = 0;
8821
8822         CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
8823
8824         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8825         return (retval);
8826 }
8827
8828 static int
8829 ctl_cnw_cont(union ctl_io *io)
8830 {
8831         struct ctl_scsiio *ctsio;
8832         struct ctl_lun *lun;
8833         struct ctl_lba_len_flags *lbalen;
8834         int retval;
8835
8836         ctsio = &io->scsiio;
8837         ctsio->io_hdr.status = CTL_STATUS_NONE;
8838         ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
8839         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8840         lbalen = (struct ctl_lba_len_flags *)
8841             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8842         lbalen->flags &= ~CTL_LLF_COMPARE;
8843         lbalen->flags |= CTL_LLF_WRITE;
8844
8845         CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
8846         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8847         return (retval);
8848 }
8849
8850 int
8851 ctl_cnw(struct ctl_scsiio *ctsio)
8852 {
8853         struct ctl_lun *lun;
8854         struct ctl_lba_len_flags *lbalen;
8855         uint64_t lba;
8856         uint32_t num_blocks;
8857         int flags, retval;
8858
8859         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8860
8861         CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
8862
8863         flags = 0;
8864         switch (ctsio->cdb[0]) {
8865         case COMPARE_AND_WRITE: {
8866                 struct scsi_compare_and_write *cdb;
8867
8868                 cdb = (struct scsi_compare_and_write *)ctsio->cdb;
8869                 if (cdb->byte2 & SRW10_FUA)
8870                         flags |= CTL_LLF_FUA;
8871                 if (cdb->byte2 & SRW10_DPO)
8872                         flags |= CTL_LLF_DPO;
8873                 lba = scsi_8btou64(cdb->addr);
8874                 num_blocks = cdb->length;
8875                 break;
8876         }
8877         default:
8878                 /*
8879                  * We got a command we don't support.  This shouldn't
8880                  * happen, commands should be filtered out above us.
8881                  */
8882                 ctl_set_invalid_opcode(ctsio);
8883                 ctl_done((union ctl_io *)ctsio);
8884
8885                 return (CTL_RETVAL_COMPLETE);
8886                 break; /* NOTREACHED */
8887         }
8888
8889         /*
8890          * The first check is to make sure we're in bounds, the second
8891          * check is to catch wrap-around problems.  If the lba + num blocks
8892          * is less than the lba, then we've wrapped around and the block
8893          * range is invalid anyway.
8894          */
8895         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8896          || ((lba + num_blocks) < lba)) {
8897                 ctl_set_lba_out_of_range(ctsio);
8898                 ctl_done((union ctl_io *)ctsio);
8899                 return (CTL_RETVAL_COMPLETE);
8900         }
8901
8902         /*
8903          * According to SBC-3, a transfer length of 0 is not an error.
8904          */
8905         if (num_blocks == 0) {
8906                 ctl_set_success(ctsio);
8907                 ctl_done((union ctl_io *)ctsio);
8908                 return (CTL_RETVAL_COMPLETE);
8909         }
8910
8911         /* Set FUA if write cache is disabled. */
8912         if ((lun->MODE_CACHING.flags1 & SCP_WCE) == 0)
8913                 flags |= CTL_LLF_FUA;
8914
8915         ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
8916         ctsio->kern_rel_offset = 0;
8917
8918         /*
8919          * Set the IO_CONT flag, so that if this I/O gets passed to
8920          * ctl_data_submit_done(), it'll get passed back to
8921          * ctl_ctl_cnw_cont() for further processing.
8922          */
8923         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
8924         ctsio->io_cont = ctl_cnw_cont;
8925
8926         lbalen = (struct ctl_lba_len_flags *)
8927             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8928         lbalen->lba = lba;
8929         lbalen->len = num_blocks;
8930         lbalen->flags = CTL_LLF_COMPARE | flags;
8931
8932         CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
8933         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8934         return (retval);
8935 }
8936
8937 int
8938 ctl_verify(struct ctl_scsiio *ctsio)
8939 {
8940         struct ctl_lun *lun;
8941         struct ctl_lba_len_flags *lbalen;
8942         uint64_t lba;
8943         uint32_t num_blocks;
8944         int bytchk, flags;
8945         int retval;
8946
8947         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8948
8949         CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
8950
8951         bytchk = 0;
8952         flags = CTL_LLF_FUA;
8953         switch (ctsio->cdb[0]) {
8954         case VERIFY_10: {
8955                 struct scsi_verify_10 *cdb;
8956
8957                 cdb = (struct scsi_verify_10 *)ctsio->cdb;
8958                 if (cdb->byte2 & SVFY_BYTCHK)
8959                         bytchk = 1;
8960                 if (cdb->byte2 & SVFY_DPO)
8961                         flags |= CTL_LLF_DPO;
8962                 lba = scsi_4btoul(cdb->addr);
8963                 num_blocks = scsi_2btoul(cdb->length);
8964                 break;
8965         }
8966         case VERIFY_12: {
8967                 struct scsi_verify_12 *cdb;
8968
8969                 cdb = (struct scsi_verify_12 *)ctsio->cdb;
8970                 if (cdb->byte2 & SVFY_BYTCHK)
8971                         bytchk = 1;
8972                 if (cdb->byte2 & SVFY_DPO)
8973                         flags |= CTL_LLF_DPO;
8974                 lba = scsi_4btoul(cdb->addr);
8975                 num_blocks = scsi_4btoul(cdb->length);
8976                 break;
8977         }
8978         case VERIFY_16: {
8979                 struct scsi_rw_16 *cdb;
8980
8981                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
8982                 if (cdb->byte2 & SVFY_BYTCHK)
8983                         bytchk = 1;
8984                 if (cdb->byte2 & SVFY_DPO)
8985                         flags |= CTL_LLF_DPO;
8986                 lba = scsi_8btou64(cdb->addr);
8987                 num_blocks = scsi_4btoul(cdb->length);
8988                 break;
8989         }
8990         default:
8991                 /*
8992                  * We got a command we don't support.  This shouldn't
8993                  * happen, commands should be filtered out above us.
8994                  */
8995                 ctl_set_invalid_opcode(ctsio);
8996                 ctl_done((union ctl_io *)ctsio);
8997                 return (CTL_RETVAL_COMPLETE);
8998         }
8999
9000         /*
9001          * The first check is to make sure we're in bounds, the second
9002          * check is to catch wrap-around problems.  If the lba + num blocks
9003          * is less than the lba, then we've wrapped around and the block
9004          * range is invalid anyway.
9005          */
9006         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9007          || ((lba + num_blocks) < lba)) {
9008                 ctl_set_lba_out_of_range(ctsio);
9009                 ctl_done((union ctl_io *)ctsio);
9010                 return (CTL_RETVAL_COMPLETE);
9011         }
9012
9013         /*
9014          * According to SBC-3, a transfer length of 0 is not an error.
9015          */
9016         if (num_blocks == 0) {
9017                 ctl_set_success(ctsio);
9018                 ctl_done((union ctl_io *)ctsio);
9019                 return (CTL_RETVAL_COMPLETE);
9020         }
9021
9022         lbalen = (struct ctl_lba_len_flags *)
9023             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9024         lbalen->lba = lba;
9025         lbalen->len = num_blocks;
9026         if (bytchk) {
9027                 lbalen->flags = CTL_LLF_COMPARE | flags;
9028                 ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9029         } else {
9030                 lbalen->flags = CTL_LLF_VERIFY | flags;
9031                 ctsio->kern_total_len = 0;
9032         }
9033         ctsio->kern_rel_offset = 0;
9034
9035         CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9036         retval = lun->backend->data_submit((union ctl_io *)ctsio);
9037         return (retval);
9038 }
9039
9040 int
9041 ctl_report_luns(struct ctl_scsiio *ctsio)
9042 {
9043         struct ctl_softc *softc;
9044         struct scsi_report_luns *cdb;
9045         struct scsi_report_luns_data *lun_data;
9046         struct ctl_lun *lun, *request_lun;
9047         struct ctl_port *port;
9048         int num_luns, retval;
9049         uint32_t alloc_len, lun_datalen;
9050         int num_filled;
9051         uint32_t initidx, targ_lun_id, lun_id;
9052
9053         retval = CTL_RETVAL_COMPLETE;
9054         cdb = (struct scsi_report_luns *)ctsio->cdb;
9055         port = ctl_io_port(&ctsio->io_hdr);
9056         softc = port->ctl_softc;
9057
9058         CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9059
9060         mtx_lock(&softc->ctl_lock);
9061         num_luns = 0;
9062         for (targ_lun_id = 0; targ_lun_id < CTL_MAX_LUNS; targ_lun_id++) {
9063                 if (ctl_lun_map_from_port(port, targ_lun_id) < CTL_MAX_LUNS)
9064                         num_luns++;
9065         }
9066         mtx_unlock(&softc->ctl_lock);
9067
9068         switch (cdb->select_report) {
9069         case RPL_REPORT_DEFAULT:
9070         case RPL_REPORT_ALL:
9071         case RPL_REPORT_NONSUBSID:
9072                 break;
9073         case RPL_REPORT_WELLKNOWN:
9074         case RPL_REPORT_ADMIN:
9075         case RPL_REPORT_CONGLOM:
9076                 num_luns = 0;
9077                 break;
9078         default:
9079                 ctl_set_invalid_field(ctsio,
9080                                       /*sks_valid*/ 1,
9081                                       /*command*/ 1,
9082                                       /*field*/ 2,
9083                                       /*bit_valid*/ 0,
9084                                       /*bit*/ 0);
9085                 ctl_done((union ctl_io *)ctsio);
9086                 return (retval);
9087                 break; /* NOTREACHED */
9088         }
9089
9090         alloc_len = scsi_4btoul(cdb->length);
9091         /*
9092          * The initiator has to allocate at least 16 bytes for this request,
9093          * so he can at least get the header and the first LUN.  Otherwise
9094          * we reject the request (per SPC-3 rev 14, section 6.21).
9095          */
9096         if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9097             sizeof(struct scsi_report_luns_lundata))) {
9098                 ctl_set_invalid_field(ctsio,
9099                                       /*sks_valid*/ 1,
9100                                       /*command*/ 1,
9101                                       /*field*/ 6,
9102                                       /*bit_valid*/ 0,
9103                                       /*bit*/ 0);
9104                 ctl_done((union ctl_io *)ctsio);
9105                 return (retval);
9106         }
9107
9108         request_lun = (struct ctl_lun *)
9109                 ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9110
9111         lun_datalen = sizeof(*lun_data) +
9112                 (num_luns * sizeof(struct scsi_report_luns_lundata));
9113
9114         ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9115         lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9116         ctsio->kern_sg_entries = 0;
9117
9118         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9119
9120         mtx_lock(&softc->ctl_lock);
9121         for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
9122                 lun_id = ctl_lun_map_from_port(port, targ_lun_id);
9123                 if (lun_id >= CTL_MAX_LUNS)
9124                         continue;
9125                 lun = softc->ctl_luns[lun_id];
9126                 if (lun == NULL)
9127                         continue;
9128
9129                 be64enc(lun_data->luns[num_filled++].lundata,
9130                     ctl_encode_lun(targ_lun_id));
9131
9132                 /*
9133                  * According to SPC-3, rev 14 section 6.21:
9134                  *
9135                  * "The execution of a REPORT LUNS command to any valid and
9136                  * installed logical unit shall clear the REPORTED LUNS DATA
9137                  * HAS CHANGED unit attention condition for all logical
9138                  * units of that target with respect to the requesting
9139                  * initiator. A valid and installed logical unit is one
9140                  * having a PERIPHERAL QUALIFIER of 000b in the standard
9141                  * INQUIRY data (see 6.4.2)."
9142                  *
9143                  * If request_lun is NULL, the LUN this report luns command
9144                  * was issued to is either disabled or doesn't exist. In that
9145                  * case, we shouldn't clear any pending lun change unit
9146                  * attention.
9147                  */
9148                 if (request_lun != NULL) {
9149                         mtx_lock(&lun->lun_lock);
9150                         ctl_clr_ua(lun, initidx, CTL_UA_LUN_CHANGE);
9151                         mtx_unlock(&lun->lun_lock);
9152                 }
9153         }
9154         mtx_unlock(&softc->ctl_lock);
9155
9156         /*
9157          * It's quite possible that we've returned fewer LUNs than we allocated
9158          * space for.  Trim it.
9159          */
9160         lun_datalen = sizeof(*lun_data) +
9161                 (num_filled * sizeof(struct scsi_report_luns_lundata));
9162
9163         if (lun_datalen < alloc_len) {
9164                 ctsio->residual = alloc_len - lun_datalen;
9165                 ctsio->kern_data_len = lun_datalen;
9166                 ctsio->kern_total_len = lun_datalen;
9167         } else {
9168                 ctsio->residual = 0;
9169                 ctsio->kern_data_len = alloc_len;
9170                 ctsio->kern_total_len = alloc_len;
9171         }
9172         ctsio->kern_data_resid = 0;
9173         ctsio->kern_rel_offset = 0;
9174         ctsio->kern_sg_entries = 0;
9175
9176         /*
9177          * We set this to the actual data length, regardless of how much
9178          * space we actually have to return results.  If the user looks at
9179          * this value, he'll know whether or not he allocated enough space
9180          * and reissue the command if necessary.  We don't support well
9181          * known logical units, so if the user asks for that, return none.
9182          */
9183         scsi_ulto4b(lun_datalen - 8, lun_data->length);
9184
9185         /*
9186          * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9187          * this request.
9188          */
9189         ctl_set_success(ctsio);
9190         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9191         ctsio->be_move_done = ctl_config_move_done;
9192         ctl_datamove((union ctl_io *)ctsio);
9193         return (retval);
9194 }
9195
9196 int
9197 ctl_request_sense(struct ctl_scsiio *ctsio)
9198 {
9199         struct scsi_request_sense *cdb;
9200         struct scsi_sense_data *sense_ptr;
9201         struct ctl_softc *softc;
9202         struct ctl_lun *lun;
9203         uint32_t initidx;
9204         int have_error;
9205         scsi_sense_data_type sense_format;
9206         ctl_ua_type ua_type;
9207         uint8_t asc = 0, ascq = 0;
9208
9209         cdb = (struct scsi_request_sense *)ctsio->cdb;
9210
9211         softc = control_softc;
9212         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9213
9214         CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9215
9216         /*
9217          * Determine which sense format the user wants.
9218          */
9219         if (cdb->byte2 & SRS_DESC)
9220                 sense_format = SSD_TYPE_DESC;
9221         else
9222                 sense_format = SSD_TYPE_FIXED;
9223
9224         ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9225         sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9226         ctsio->kern_sg_entries = 0;
9227
9228         /*
9229          * struct scsi_sense_data, which is currently set to 256 bytes, is
9230          * larger than the largest allowed value for the length field in the
9231          * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9232          */
9233         ctsio->residual = 0;
9234         ctsio->kern_data_len = cdb->length;
9235         ctsio->kern_total_len = cdb->length;
9236
9237         ctsio->kern_data_resid = 0;
9238         ctsio->kern_rel_offset = 0;
9239         ctsio->kern_sg_entries = 0;
9240
9241         /*
9242          * If we don't have a LUN, we don't have any pending sense.
9243          */
9244         if (lun == NULL ||
9245             ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
9246              softc->ha_link < CTL_HA_LINK_UNKNOWN)) {
9247                 /* "Logical unit not supported" */
9248                 ctl_set_sense_data(sense_ptr, NULL, sense_format,
9249                     /*current_error*/ 1,
9250                     /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
9251                     /*asc*/ 0x25,
9252                     /*ascq*/ 0x00,
9253                     SSD_ELEM_NONE);
9254                 goto send;
9255         }
9256
9257         have_error = 0;
9258         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9259         /*
9260          * Check for pending sense, and then for pending unit attentions.
9261          * Pending sense gets returned first, then pending unit attentions.
9262          */
9263         mtx_lock(&lun->lun_lock);
9264 #ifdef CTL_WITH_CA
9265         if (ctl_is_set(lun->have_ca, initidx)) {
9266                 scsi_sense_data_type stored_format;
9267
9268                 /*
9269                  * Check to see which sense format was used for the stored
9270                  * sense data.
9271                  */
9272                 stored_format = scsi_sense_type(&lun->pending_sense[initidx]);
9273
9274                 /*
9275                  * If the user requested a different sense format than the
9276                  * one we stored, then we need to convert it to the other
9277                  * format.  If we're going from descriptor to fixed format
9278                  * sense data, we may lose things in translation, depending
9279                  * on what options were used.
9280                  *
9281                  * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9282                  * for some reason we'll just copy it out as-is.
9283                  */
9284                 if ((stored_format == SSD_TYPE_FIXED)
9285                  && (sense_format == SSD_TYPE_DESC))
9286                         ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9287                             &lun->pending_sense[initidx],
9288                             (struct scsi_sense_data_desc *)sense_ptr);
9289                 else if ((stored_format == SSD_TYPE_DESC)
9290                       && (sense_format == SSD_TYPE_FIXED))
9291                         ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9292                             &lun->pending_sense[initidx],
9293                             (struct scsi_sense_data_fixed *)sense_ptr);
9294                 else
9295                         memcpy(sense_ptr, &lun->pending_sense[initidx],
9296                                MIN(sizeof(*sense_ptr),
9297                                sizeof(lun->pending_sense[initidx])));
9298
9299                 ctl_clear_mask(lun->have_ca, initidx);
9300                 have_error = 1;
9301         } else
9302 #endif
9303         if (have_error == 0) {
9304                 ua_type = ctl_build_ua(lun, initidx, sense_ptr, sense_format);
9305                 if (ua_type != CTL_UA_NONE)
9306                         have_error = 1;
9307                 if (ua_type == CTL_UA_LUN_CHANGE) {
9308                         mtx_unlock(&lun->lun_lock);
9309                         mtx_lock(&softc->ctl_lock);
9310                         ctl_clr_ua_allluns(softc, initidx, ua_type);
9311                         mtx_unlock(&softc->ctl_lock);
9312                         mtx_lock(&lun->lun_lock);
9313                 }
9314         }
9315         if (have_error == 0) {
9316                 /*
9317                  * Report informational exception if have one and allowed.
9318                  */
9319                 if (lun->MODE_IE.mrie != SIEP_MRIE_NO) {
9320                         asc = lun->ie_asc;
9321                         ascq = lun->ie_ascq;
9322                 }
9323                 ctl_set_sense_data(sense_ptr, lun, sense_format,
9324                     /*current_error*/ 1,
9325                     /*sense_key*/ SSD_KEY_NO_SENSE,
9326                     /*asc*/ asc,
9327                     /*ascq*/ ascq,
9328                     SSD_ELEM_NONE);
9329         }
9330         mtx_unlock(&lun->lun_lock);
9331
9332 send:
9333         /*
9334          * We report the SCSI status as OK, since the status of the command
9335          * itself is OK.  We're reporting sense as parameter data.
9336          */
9337         ctl_set_success(ctsio);
9338         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9339         ctsio->be_move_done = ctl_config_move_done;
9340         ctl_datamove((union ctl_io *)ctsio);
9341         return (CTL_RETVAL_COMPLETE);
9342 }
9343
9344 int
9345 ctl_tur(struct ctl_scsiio *ctsio)
9346 {
9347
9348         CTL_DEBUG_PRINT(("ctl_tur\n"));
9349
9350         ctl_set_success(ctsio);
9351         ctl_done((union ctl_io *)ctsio);
9352
9353         return (CTL_RETVAL_COMPLETE);
9354 }
9355
9356 /*
9357  * SCSI VPD page 0x00, the Supported VPD Pages page.
9358  */
9359 static int
9360 ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9361 {
9362         struct scsi_vpd_supported_pages *pages;
9363         int sup_page_size;
9364         struct ctl_lun *lun;
9365         int p;
9366
9367         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9368
9369         sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9370             SCSI_EVPD_NUM_SUPPORTED_PAGES;
9371         ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9372         pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9373         ctsio->kern_sg_entries = 0;
9374
9375         if (sup_page_size < alloc_len) {
9376                 ctsio->residual = alloc_len - sup_page_size;
9377                 ctsio->kern_data_len = sup_page_size;
9378                 ctsio->kern_total_len = sup_page_size;
9379         } else {
9380                 ctsio->residual = 0;
9381                 ctsio->kern_data_len = alloc_len;
9382                 ctsio->kern_total_len = alloc_len;
9383         }
9384         ctsio->kern_data_resid = 0;
9385         ctsio->kern_rel_offset = 0;
9386         ctsio->kern_sg_entries = 0;
9387
9388         /*
9389          * The control device is always connected.  The disk device, on the
9390          * other hand, may not be online all the time.  Need to change this
9391          * to figure out whether the disk device is actually online or not.
9392          */
9393         if (lun != NULL)
9394                 pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9395                                 lun->be_lun->lun_type;
9396         else
9397                 pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9398
9399         p = 0;
9400         /* Supported VPD pages */
9401         pages->page_list[p++] = SVPD_SUPPORTED_PAGES;
9402         /* Serial Number */
9403         pages->page_list[p++] = SVPD_UNIT_SERIAL_NUMBER;
9404         /* Device Identification */
9405         pages->page_list[p++] = SVPD_DEVICE_ID;
9406         /* Extended INQUIRY Data */
9407         pages->page_list[p++] = SVPD_EXTENDED_INQUIRY_DATA;
9408         /* Mode Page Policy */
9409         pages->page_list[p++] = SVPD_MODE_PAGE_POLICY;
9410         /* SCSI Ports */
9411         pages->page_list[p++] = SVPD_SCSI_PORTS;
9412         /* Third-party Copy */
9413         pages->page_list[p++] = SVPD_SCSI_TPC;
9414         if (lun != NULL && lun->be_lun->lun_type == T_DIRECT) {
9415                 /* Block limits */
9416                 pages->page_list[p++] = SVPD_BLOCK_LIMITS;
9417                 /* Block Device Characteristics */
9418                 pages->page_list[p++] = SVPD_BDC;
9419                 /* Logical Block Provisioning */
9420                 pages->page_list[p++] = SVPD_LBP;
9421         }
9422         pages->length = p;
9423
9424         ctl_set_success(ctsio);
9425         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9426         ctsio->be_move_done = ctl_config_move_done;
9427         ctl_datamove((union ctl_io *)ctsio);
9428         return (CTL_RETVAL_COMPLETE);
9429 }
9430
9431 /*
9432  * SCSI VPD page 0x80, the Unit Serial Number page.
9433  */
9434 static int
9435 ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9436 {
9437         struct scsi_vpd_unit_serial_number *sn_ptr;
9438         struct ctl_lun *lun;
9439         int data_len;
9440
9441         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9442
9443         data_len = 4 + CTL_SN_LEN;
9444         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9445         sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9446         if (data_len < alloc_len) {
9447                 ctsio->residual = alloc_len - data_len;
9448                 ctsio->kern_data_len = data_len;
9449                 ctsio->kern_total_len = data_len;
9450         } else {
9451                 ctsio->residual = 0;
9452                 ctsio->kern_data_len = alloc_len;
9453                 ctsio->kern_total_len = alloc_len;
9454         }
9455         ctsio->kern_data_resid = 0;
9456         ctsio->kern_rel_offset = 0;
9457         ctsio->kern_sg_entries = 0;
9458
9459         /*
9460          * The control device is always connected.  The disk device, on the
9461          * other hand, may not be online all the time.  Need to change this
9462          * to figure out whether the disk device is actually online or not.
9463          */
9464         if (lun != NULL)
9465                 sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9466                                   lun->be_lun->lun_type;
9467         else
9468                 sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9469
9470         sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9471         sn_ptr->length = CTL_SN_LEN;
9472         /*
9473          * If we don't have a LUN, we just leave the serial number as
9474          * all spaces.
9475          */
9476         if (lun != NULL) {
9477                 strncpy((char *)sn_ptr->serial_num,
9478                         (char *)lun->be_lun->serial_num, CTL_SN_LEN);
9479         } else
9480                 memset(sn_ptr->serial_num, 0x20, CTL_SN_LEN);
9481
9482         ctl_set_success(ctsio);
9483         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9484         ctsio->be_move_done = ctl_config_move_done;
9485         ctl_datamove((union ctl_io *)ctsio);
9486         return (CTL_RETVAL_COMPLETE);
9487 }
9488
9489
9490 /*
9491  * SCSI VPD page 0x86, the Extended INQUIRY Data page.
9492  */
9493 static int
9494 ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len)
9495 {
9496         struct scsi_vpd_extended_inquiry_data *eid_ptr;
9497         struct ctl_lun *lun;
9498         int data_len;
9499
9500         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9501
9502         data_len = sizeof(struct scsi_vpd_extended_inquiry_data);
9503         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9504         eid_ptr = (struct scsi_vpd_extended_inquiry_data *)ctsio->kern_data_ptr;
9505         ctsio->kern_sg_entries = 0;
9506
9507         if (data_len < alloc_len) {
9508                 ctsio->residual = alloc_len - data_len;
9509                 ctsio->kern_data_len = data_len;
9510                 ctsio->kern_total_len = data_len;
9511         } else {
9512                 ctsio->residual = 0;
9513                 ctsio->kern_data_len = alloc_len;
9514                 ctsio->kern_total_len = alloc_len;
9515         }
9516         ctsio->kern_data_resid = 0;
9517         ctsio->kern_rel_offset = 0;
9518         ctsio->kern_sg_entries = 0;
9519
9520         /*
9521          * The control device is always connected.  The disk device, on the
9522          * other hand, may not be online all the time.
9523          */
9524         if (lun != NULL)
9525                 eid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9526                                      lun->be_lun->lun_type;
9527         else
9528                 eid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9529         eid_ptr->page_code = SVPD_EXTENDED_INQUIRY_DATA;
9530         scsi_ulto2b(data_len - 4, eid_ptr->page_length);
9531         /*
9532          * We support head of queue, ordered and simple tags.
9533          */
9534         eid_ptr->flags2 = SVPD_EID_HEADSUP | SVPD_EID_ORDSUP | SVPD_EID_SIMPSUP;
9535         /*
9536          * Volatile cache supported.
9537          */
9538         eid_ptr->flags3 = SVPD_EID_V_SUP;
9539
9540         /*
9541          * This means that we clear the REPORTED LUNS DATA HAS CHANGED unit
9542          * attention for a particular IT nexus on all LUNs once we report
9543          * it to that nexus once.  This bit is required as of SPC-4.
9544          */
9545         eid_ptr->flags4 = SVPD_EID_LUICLT;
9546
9547         /*
9548          * XXX KDM in order to correctly answer this, we would need
9549          * information from the SIM to determine how much sense data it
9550          * can send.  So this would really be a path inquiry field, most
9551          * likely.  This can be set to a maximum of 252 according to SPC-4,
9552          * but the hardware may or may not be able to support that much.
9553          * 0 just means that the maximum sense data length is not reported.
9554          */
9555         eid_ptr->max_sense_length = 0;
9556
9557         ctl_set_success(ctsio);
9558         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9559         ctsio->be_move_done = ctl_config_move_done;
9560         ctl_datamove((union ctl_io *)ctsio);
9561         return (CTL_RETVAL_COMPLETE);
9562 }
9563
9564 static int
9565 ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len)
9566 {
9567         struct scsi_vpd_mode_page_policy *mpp_ptr;
9568         struct ctl_lun *lun;
9569         int data_len;
9570
9571         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9572
9573         data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9574             sizeof(struct scsi_vpd_mode_page_policy_descr);
9575
9576         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9577         mpp_ptr = (struct scsi_vpd_mode_page_policy *)ctsio->kern_data_ptr;
9578         ctsio->kern_sg_entries = 0;
9579
9580         if (data_len < alloc_len) {
9581                 ctsio->residual = alloc_len - data_len;
9582                 ctsio->kern_data_len = data_len;
9583                 ctsio->kern_total_len = data_len;
9584         } else {
9585                 ctsio->residual = 0;
9586                 ctsio->kern_data_len = alloc_len;
9587                 ctsio->kern_total_len = alloc_len;
9588         }
9589         ctsio->kern_data_resid = 0;
9590         ctsio->kern_rel_offset = 0;
9591         ctsio->kern_sg_entries = 0;
9592
9593         /*
9594          * The control device is always connected.  The disk device, on the
9595          * other hand, may not be online all the time.
9596          */
9597         if (lun != NULL)
9598                 mpp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9599                                      lun->be_lun->lun_type;
9600         else
9601                 mpp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9602         mpp_ptr->page_code = SVPD_MODE_PAGE_POLICY;
9603         scsi_ulto2b(data_len - 4, mpp_ptr->page_length);
9604         mpp_ptr->descr[0].page_code = 0x3f;
9605         mpp_ptr->descr[0].subpage_code = 0xff;
9606         mpp_ptr->descr[0].policy = SVPD_MPP_SHARED;
9607
9608         ctl_set_success(ctsio);
9609         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9610         ctsio->be_move_done = ctl_config_move_done;
9611         ctl_datamove((union ctl_io *)ctsio);
9612         return (CTL_RETVAL_COMPLETE);
9613 }
9614
9615 /*
9616  * SCSI VPD page 0x83, the Device Identification page.
9617  */
9618 static int
9619 ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9620 {
9621         struct scsi_vpd_device_id *devid_ptr;
9622         struct scsi_vpd_id_descriptor *desc;
9623         struct ctl_softc *softc;
9624         struct ctl_lun *lun;
9625         struct ctl_port *port;
9626         int data_len, g;
9627         uint8_t proto;
9628
9629         softc = control_softc;
9630
9631         port = ctl_io_port(&ctsio->io_hdr);
9632         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9633
9634         data_len = sizeof(struct scsi_vpd_device_id) +
9635             sizeof(struct scsi_vpd_id_descriptor) +
9636                 sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9637             sizeof(struct scsi_vpd_id_descriptor) +
9638                 sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9639         if (lun && lun->lun_devid)
9640                 data_len += lun->lun_devid->len;
9641         if (port && port->port_devid)
9642                 data_len += port->port_devid->len;
9643         if (port && port->target_devid)
9644                 data_len += port->target_devid->len;
9645
9646         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9647         devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9648         ctsio->kern_sg_entries = 0;
9649
9650         if (data_len < alloc_len) {
9651                 ctsio->residual = alloc_len - data_len;
9652                 ctsio->kern_data_len = data_len;
9653                 ctsio->kern_total_len = data_len;
9654         } else {
9655                 ctsio->residual = 0;
9656                 ctsio->kern_data_len = alloc_len;
9657                 ctsio->kern_total_len = alloc_len;
9658         }
9659         ctsio->kern_data_resid = 0;
9660         ctsio->kern_rel_offset = 0;
9661         ctsio->kern_sg_entries = 0;
9662
9663         /*
9664          * The control device is always connected.  The disk device, on the
9665          * other hand, may not be online all the time.
9666          */
9667         if (lun != NULL)
9668                 devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9669                                      lun->be_lun->lun_type;
9670         else
9671                 devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9672         devid_ptr->page_code = SVPD_DEVICE_ID;
9673         scsi_ulto2b(data_len - 4, devid_ptr->length);
9674
9675         if (port && port->port_type == CTL_PORT_FC)
9676                 proto = SCSI_PROTO_FC << 4;
9677         else if (port && port->port_type == CTL_PORT_ISCSI)
9678                 proto = SCSI_PROTO_ISCSI << 4;
9679         else
9680                 proto = SCSI_PROTO_SPI << 4;
9681         desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9682
9683         /*
9684          * We're using a LUN association here.  i.e., this device ID is a
9685          * per-LUN identifier.
9686          */
9687         if (lun && lun->lun_devid) {
9688                 memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
9689                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9690                     lun->lun_devid->len);
9691         }
9692
9693         /*
9694          * This is for the WWPN which is a port association.
9695          */
9696         if (port && port->port_devid) {
9697                 memcpy(desc, port->port_devid->data, port->port_devid->len);
9698                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9699                     port->port_devid->len);
9700         }
9701
9702         /*
9703          * This is for the Relative Target Port(type 4h) identifier
9704          */
9705         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9706         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9707             SVPD_ID_TYPE_RELTARG;
9708         desc->length = 4;
9709         scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
9710         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9711             sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9712
9713         /*
9714          * This is for the Target Port Group(type 5h) identifier
9715          */
9716         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9717         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9718             SVPD_ID_TYPE_TPORTGRP;
9719         desc->length = 4;
9720         if (softc->is_single ||
9721             (port && port->status & CTL_PORT_STATUS_HA_SHARED))
9722                 g = 1;
9723         else
9724                 g = 2 + ctsio->io_hdr.nexus.targ_port / softc->port_cnt;
9725         scsi_ulto2b(g, &desc->identifier[2]);
9726         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9727             sizeof(struct scsi_vpd_id_trgt_port_grp_id));
9728
9729         /*
9730          * This is for the Target identifier
9731          */
9732         if (port && port->target_devid) {
9733                 memcpy(desc, port->target_devid->data, port->target_devid->len);
9734         }
9735
9736         ctl_set_success(ctsio);
9737         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9738         ctsio->be_move_done = ctl_config_move_done;
9739         ctl_datamove((union ctl_io *)ctsio);
9740         return (CTL_RETVAL_COMPLETE);
9741 }
9742
9743 static int
9744 ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
9745 {
9746         struct ctl_softc *softc = control_softc;
9747         struct scsi_vpd_scsi_ports *sp;
9748         struct scsi_vpd_port_designation *pd;
9749         struct scsi_vpd_port_designation_cont *pdc;
9750         struct ctl_lun *lun;
9751         struct ctl_port *port;
9752         int data_len, num_target_ports, iid_len, id_len;
9753
9754         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9755
9756         num_target_ports = 0;
9757         iid_len = 0;
9758         id_len = 0;
9759         mtx_lock(&softc->ctl_lock);
9760         STAILQ_FOREACH(port, &softc->port_list, links) {
9761                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9762                         continue;
9763                 if (lun != NULL &&
9764                     ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
9765                         continue;
9766                 num_target_ports++;
9767                 if (port->init_devid)
9768                         iid_len += port->init_devid->len;
9769                 if (port->port_devid)
9770                         id_len += port->port_devid->len;
9771         }
9772         mtx_unlock(&softc->ctl_lock);
9773
9774         data_len = sizeof(struct scsi_vpd_scsi_ports) +
9775             num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
9776              sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
9777         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9778         sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
9779         ctsio->kern_sg_entries = 0;
9780
9781         if (data_len < alloc_len) {
9782                 ctsio->residual = alloc_len - data_len;
9783                 ctsio->kern_data_len = data_len;
9784                 ctsio->kern_total_len = data_len;
9785         } else {
9786                 ctsio->residual = 0;
9787                 ctsio->kern_data_len = alloc_len;
9788                 ctsio->kern_total_len = alloc_len;
9789         }
9790         ctsio->kern_data_resid = 0;
9791         ctsio->kern_rel_offset = 0;
9792         ctsio->kern_sg_entries = 0;
9793
9794         /*
9795          * The control device is always connected.  The disk device, on the
9796          * other hand, may not be online all the time.  Need to change this
9797          * to figure out whether the disk device is actually online or not.
9798          */
9799         if (lun != NULL)
9800                 sp->device = (SID_QUAL_LU_CONNECTED << 5) |
9801                                   lun->be_lun->lun_type;
9802         else
9803                 sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9804
9805         sp->page_code = SVPD_SCSI_PORTS;
9806         scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
9807             sp->page_length);
9808         pd = &sp->design[0];
9809
9810         mtx_lock(&softc->ctl_lock);
9811         STAILQ_FOREACH(port, &softc->port_list, links) {
9812                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9813                         continue;
9814                 if (lun != NULL &&
9815                     ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
9816                         continue;
9817                 scsi_ulto2b(port->targ_port, pd->relative_port_id);
9818                 if (port->init_devid) {
9819                         iid_len = port->init_devid->len;
9820                         memcpy(pd->initiator_transportid,
9821                             port->init_devid->data, port->init_devid->len);
9822                 } else
9823                         iid_len = 0;
9824                 scsi_ulto2b(iid_len, pd->initiator_transportid_length);
9825                 pdc = (struct scsi_vpd_port_designation_cont *)
9826                     (&pd->initiator_transportid[iid_len]);
9827                 if (port->port_devid) {
9828                         id_len = port->port_devid->len;
9829                         memcpy(pdc->target_port_descriptors,
9830                             port->port_devid->data, port->port_devid->len);
9831                 } else
9832                         id_len = 0;
9833                 scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
9834                 pd = (struct scsi_vpd_port_designation *)
9835                     ((uint8_t *)pdc->target_port_descriptors + id_len);
9836         }
9837         mtx_unlock(&softc->ctl_lock);
9838
9839         ctl_set_success(ctsio);
9840         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9841         ctsio->be_move_done = ctl_config_move_done;
9842         ctl_datamove((union ctl_io *)ctsio);
9843         return (CTL_RETVAL_COMPLETE);
9844 }
9845
9846 static int
9847 ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
9848 {
9849         struct scsi_vpd_block_limits *bl_ptr;
9850         struct ctl_lun *lun;
9851         uint64_t ival;
9852
9853         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9854
9855         ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
9856         bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
9857         ctsio->kern_sg_entries = 0;
9858
9859         if (sizeof(*bl_ptr) < alloc_len) {
9860                 ctsio->residual = alloc_len - sizeof(*bl_ptr);
9861                 ctsio->kern_data_len = sizeof(*bl_ptr);
9862                 ctsio->kern_total_len = sizeof(*bl_ptr);
9863         } else {
9864                 ctsio->residual = 0;
9865                 ctsio->kern_data_len = alloc_len;
9866                 ctsio->kern_total_len = alloc_len;
9867         }
9868         ctsio->kern_data_resid = 0;
9869         ctsio->kern_rel_offset = 0;
9870         ctsio->kern_sg_entries = 0;
9871
9872         /*
9873          * The control device is always connected.  The disk device, on the
9874          * other hand, may not be online all the time.  Need to change this
9875          * to figure out whether the disk device is actually online or not.
9876          */
9877         if (lun != NULL)
9878                 bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9879                                   lun->be_lun->lun_type;
9880         else
9881                 bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9882
9883         bl_ptr->page_code = SVPD_BLOCK_LIMITS;
9884         scsi_ulto2b(sizeof(*bl_ptr) - 4, bl_ptr->page_length);
9885         bl_ptr->max_cmp_write_len = 0xff;
9886         scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
9887         if (lun != NULL) {
9888                 scsi_ulto4b(lun->be_lun->opttxferlen, bl_ptr->opt_txfer_len);
9889                 if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9890                         ival = 0xffffffff;
9891                         ctl_get_opt_number(&lun->be_lun->options,
9892                             "unmap_max_lba", &ival);
9893                         scsi_ulto4b(ival, bl_ptr->max_unmap_lba_cnt);
9894                         ival = 0xffffffff;
9895                         ctl_get_opt_number(&lun->be_lun->options,
9896                             "unmap_max_descr", &ival);
9897                         scsi_ulto4b(ival, bl_ptr->max_unmap_blk_cnt);
9898                         if (lun->be_lun->ublockexp != 0) {
9899                                 scsi_ulto4b((1 << lun->be_lun->ublockexp),
9900                                     bl_ptr->opt_unmap_grain);
9901                                 scsi_ulto4b(0x80000000 | lun->be_lun->ublockoff,
9902                                     bl_ptr->unmap_grain_align);
9903                         }
9904                 }
9905                 scsi_ulto4b(lun->be_lun->atomicblock,
9906                     bl_ptr->max_atomic_transfer_length);
9907                 scsi_ulto4b(0, bl_ptr->atomic_alignment);
9908                 scsi_ulto4b(0, bl_ptr->atomic_transfer_length_granularity);
9909                 scsi_ulto4b(0, bl_ptr->max_atomic_transfer_length_with_atomic_boundary);
9910                 scsi_ulto4b(0, bl_ptr->max_atomic_boundary_size);
9911                 ival = UINT64_MAX;
9912                 ctl_get_opt_number(&lun->be_lun->options, "write_same_max_lba", &ival);
9913                 scsi_u64to8b(ival, bl_ptr->max_write_same_length);
9914         }
9915
9916         ctl_set_success(ctsio);
9917         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9918         ctsio->be_move_done = ctl_config_move_done;
9919         ctl_datamove((union ctl_io *)ctsio);
9920         return (CTL_RETVAL_COMPLETE);
9921 }
9922
9923 static int
9924 ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len)
9925 {
9926         struct scsi_vpd_block_device_characteristics *bdc_ptr;
9927         struct ctl_lun *lun;
9928         const char *value;
9929         u_int i;
9930
9931         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9932
9933         ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO);
9934         bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr;
9935         ctsio->kern_sg_entries = 0;
9936
9937         if (sizeof(*bdc_ptr) < alloc_len) {
9938                 ctsio->residual = alloc_len - sizeof(*bdc_ptr);
9939                 ctsio->kern_data_len = sizeof(*bdc_ptr);
9940                 ctsio->kern_total_len = sizeof(*bdc_ptr);
9941         } else {
9942                 ctsio->residual = 0;
9943                 ctsio->kern_data_len = alloc_len;
9944                 ctsio->kern_total_len = alloc_len;
9945         }
9946         ctsio->kern_data_resid = 0;
9947         ctsio->kern_rel_offset = 0;
9948         ctsio->kern_sg_entries = 0;
9949
9950         /*
9951          * The control device is always connected.  The disk device, on the
9952          * other hand, may not be online all the time.  Need to change this
9953          * to figure out whether the disk device is actually online or not.
9954          */
9955         if (lun != NULL)
9956                 bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9957                                   lun->be_lun->lun_type;
9958         else
9959                 bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9960         bdc_ptr->page_code = SVPD_BDC;
9961         scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length);
9962         if (lun != NULL &&
9963             (value = ctl_get_opt(&lun->be_lun->options, "rpm")) != NULL)
9964                 i = strtol(value, NULL, 0);
9965         else
9966                 i = CTL_DEFAULT_ROTATION_RATE;
9967         scsi_ulto2b(i, bdc_ptr->medium_rotation_rate);
9968         if (lun != NULL &&
9969             (value = ctl_get_opt(&lun->be_lun->options, "formfactor")) != NULL)
9970                 i = strtol(value, NULL, 0);
9971         else
9972                 i = 0;
9973         bdc_ptr->wab_wac_ff = (i & 0x0f);
9974         bdc_ptr->flags = SVPD_FUAB | SVPD_VBULS;
9975
9976         ctl_set_success(ctsio);
9977         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9978         ctsio->be_move_done = ctl_config_move_done;
9979         ctl_datamove((union ctl_io *)ctsio);
9980         return (CTL_RETVAL_COMPLETE);
9981 }
9982
9983 static int
9984 ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
9985 {
9986         struct scsi_vpd_logical_block_prov *lbp_ptr;
9987         struct ctl_lun *lun;
9988         const char *value;
9989
9990         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9991
9992         ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
9993         lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
9994         ctsio->kern_sg_entries = 0;
9995
9996         if (sizeof(*lbp_ptr) < alloc_len) {
9997                 ctsio->residual = alloc_len - sizeof(*lbp_ptr);
9998                 ctsio->kern_data_len = sizeof(*lbp_ptr);
9999                 ctsio->kern_total_len = sizeof(*lbp_ptr);
10000         } else {
10001                 ctsio->residual = 0;
10002                 ctsio->kern_data_len = alloc_len;
10003                 ctsio->kern_total_len = alloc_len;
10004         }
10005         ctsio->kern_data_resid = 0;
10006         ctsio->kern_rel_offset = 0;
10007         ctsio->kern_sg_entries = 0;
10008
10009         /*
10010          * The control device is always connected.  The disk device, on the
10011          * other hand, may not be online all the time.  Need to change this
10012          * to figure out whether the disk device is actually online or not.
10013          */
10014         if (lun != NULL)
10015                 lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10016                                   lun->be_lun->lun_type;
10017         else
10018                 lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10019
10020         lbp_ptr->page_code = SVPD_LBP;
10021         scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length);
10022         lbp_ptr->threshold_exponent = CTL_LBP_EXPONENT;
10023         if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10024                 lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 |
10025                     SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP;
10026                 value = ctl_get_opt(&lun->be_lun->options, "provisioning_type");
10027                 if (value != NULL) {
10028                         if (strcmp(value, "resource") == 0)
10029                                 lbp_ptr->prov_type = SVPD_LBP_RESOURCE;
10030                         else if (strcmp(value, "thin") == 0)
10031                                 lbp_ptr->prov_type = SVPD_LBP_THIN;
10032                 } else
10033                         lbp_ptr->prov_type = SVPD_LBP_THIN;
10034         }
10035
10036         ctl_set_success(ctsio);
10037         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10038         ctsio->be_move_done = ctl_config_move_done;
10039         ctl_datamove((union ctl_io *)ctsio);
10040         return (CTL_RETVAL_COMPLETE);
10041 }
10042
10043 /*
10044  * INQUIRY with the EVPD bit set.
10045  */
10046 static int
10047 ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
10048 {
10049         struct ctl_lun *lun;
10050         struct scsi_inquiry *cdb;
10051         int alloc_len, retval;
10052
10053         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10054         cdb = (struct scsi_inquiry *)ctsio->cdb;
10055         alloc_len = scsi_2btoul(cdb->length);
10056
10057         switch (cdb->page_code) {
10058         case SVPD_SUPPORTED_PAGES:
10059                 retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
10060                 break;
10061         case SVPD_UNIT_SERIAL_NUMBER:
10062                 retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
10063                 break;
10064         case SVPD_DEVICE_ID:
10065                 retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
10066                 break;
10067         case SVPD_EXTENDED_INQUIRY_DATA:
10068                 retval = ctl_inquiry_evpd_eid(ctsio, alloc_len);
10069                 break;
10070         case SVPD_MODE_PAGE_POLICY:
10071                 retval = ctl_inquiry_evpd_mpp(ctsio, alloc_len);
10072                 break;
10073         case SVPD_SCSI_PORTS:
10074                 retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
10075                 break;
10076         case SVPD_SCSI_TPC:
10077                 retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
10078                 break;
10079         case SVPD_BLOCK_LIMITS:
10080                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10081                         goto err;
10082                 retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
10083                 break;
10084         case SVPD_BDC:
10085                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10086                         goto err;
10087                 retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len);
10088                 break;
10089         case SVPD_LBP:
10090                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10091                         goto err;
10092                 retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
10093                 break;
10094         default:
10095 err:
10096                 ctl_set_invalid_field(ctsio,
10097                                       /*sks_valid*/ 1,
10098                                       /*command*/ 1,
10099                                       /*field*/ 2,
10100                                       /*bit_valid*/ 0,
10101                                       /*bit*/ 0);
10102                 ctl_done((union ctl_io *)ctsio);
10103                 retval = CTL_RETVAL_COMPLETE;
10104                 break;
10105         }
10106
10107         return (retval);
10108 }
10109
10110 /*
10111  * Standard INQUIRY data.
10112  */
10113 static int
10114 ctl_inquiry_std(struct ctl_scsiio *ctsio)
10115 {
10116         struct scsi_inquiry_data *inq_ptr;
10117         struct scsi_inquiry *cdb;
10118         struct ctl_softc *softc = control_softc;
10119         struct ctl_port *port;
10120         struct ctl_lun *lun;
10121         char *val;
10122         uint32_t alloc_len, data_len;
10123         ctl_port_type port_type;
10124
10125         port = ctl_io_port(&ctsio->io_hdr);
10126         port_type = port->port_type;
10127         if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
10128                 port_type = CTL_PORT_SCSI;
10129
10130         lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10131         cdb = (struct scsi_inquiry *)ctsio->cdb;
10132         alloc_len = scsi_2btoul(cdb->length);
10133
10134         /*
10135          * We malloc the full inquiry data size here and fill it
10136          * in.  If the user only asks for less, we'll give him
10137          * that much.
10138          */
10139         data_len = offsetof(struct scsi_inquiry_data, vendor_specific1);
10140         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10141         inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
10142         ctsio->kern_sg_entries = 0;
10143         ctsio->kern_data_resid = 0;
10144         ctsio->kern_rel_offset = 0;
10145
10146         if (data_len < alloc_len) {
10147                 ctsio->residual = alloc_len - data_len;
10148                 ctsio->kern_data_len = data_len;
10149                 ctsio->kern_total_len = data_len;
10150         } else {
10151                 ctsio->residual = 0;
10152                 ctsio->kern_data_len = alloc_len;
10153                 ctsio->kern_total_len = alloc_len;
10154         }
10155
10156         if (lun != NULL) {
10157                 if ((lun->flags & CTL_LUN_PRIMARY_SC) ||
10158                     softc->ha_link >= CTL_HA_LINK_UNKNOWN) {
10159                         inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10160                             lun->be_lun->lun_type;
10161                 } else {
10162                         inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) |
10163                             lun->be_lun->lun_type;
10164                 }
10165                 if (lun->flags & CTL_LUN_REMOVABLE)
10166                         inq_ptr->dev_qual2 |= SID_RMB;
10167         } else
10168                 inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10169
10170         /* RMB in byte 2 is 0 */
10171         inq_ptr->version = SCSI_REV_SPC4;
10172
10173         /*
10174          * According to SAM-3, even if a device only supports a single
10175          * level of LUN addressing, it should still set the HISUP bit:
10176          *
10177          * 4.9.1 Logical unit numbers overview
10178          *
10179          * All logical unit number formats described in this standard are
10180          * hierarchical in structure even when only a single level in that
10181          * hierarchy is used. The HISUP bit shall be set to one in the
10182          * standard INQUIRY data (see SPC-2) when any logical unit number
10183          * format described in this standard is used.  Non-hierarchical
10184          * formats are outside the scope of this standard.
10185          *
10186          * Therefore we set the HiSup bit here.
10187          *
10188          * The response format is 2, per SPC-3.
10189          */
10190         inq_ptr->response_format = SID_HiSup | 2;
10191
10192         inq_ptr->additional_length = data_len -
10193             (offsetof(struct scsi_inquiry_data, additional_length) + 1);
10194         CTL_DEBUG_PRINT(("additional_length = %d\n",
10195                          inq_ptr->additional_length));
10196
10197         inq_ptr->spc3_flags = SPC3_SID_3PC | SPC3_SID_TPGS_IMPLICIT;
10198         if (port_type == CTL_PORT_SCSI)
10199                 inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10200         inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10201         inq_ptr->flags = SID_CmdQue;
10202         if (port_type == CTL_PORT_SCSI)
10203                 inq_ptr->flags |= SID_WBus16 | SID_Sync;
10204
10205         /*
10206          * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10207          * We have 8 bytes for the vendor name, and 16 bytes for the device
10208          * name and 4 bytes for the revision.
10209          */
10210         if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10211             "vendor")) == NULL) {
10212                 strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
10213         } else {
10214                 memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10215                 strncpy(inq_ptr->vendor, val,
10216                     min(sizeof(inq_ptr->vendor), strlen(val)));
10217         }
10218         if (lun == NULL) {
10219                 strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10220                     sizeof(inq_ptr->product));
10221         } else if ((val = ctl_get_opt(&lun->be_lun->options, "product")) == NULL) {
10222                 switch (lun->be_lun->lun_type) {
10223                 case T_DIRECT:
10224                         strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10225                             sizeof(inq_ptr->product));
10226                         break;
10227                 case T_PROCESSOR:
10228                         strncpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT,
10229                             sizeof(inq_ptr->product));
10230                         break;
10231                 case T_CDROM:
10232                         strncpy(inq_ptr->product, CTL_CDROM_PRODUCT,
10233                             sizeof(inq_ptr->product));
10234                         break;
10235                 default:
10236                         strncpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT,
10237                             sizeof(inq_ptr->product));
10238                         break;
10239                 }
10240         } else {
10241                 memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10242                 strncpy(inq_ptr->product, val,
10243                     min(sizeof(inq_ptr->product), strlen(val)));
10244         }
10245
10246         /*
10247          * XXX make this a macro somewhere so it automatically gets
10248          * incremented when we make changes.
10249          */
10250         if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10251             "revision")) == NULL) {
10252                 strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10253         } else {
10254                 memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10255                 strncpy(inq_ptr->revision, val,
10256                     min(sizeof(inq_ptr->revision), strlen(val)));
10257         }
10258
10259         /*
10260          * For parallel SCSI, we support double transition and single
10261          * transition clocking.  We also support QAS (Quick Arbitration
10262          * and Selection) and Information Unit transfers on both the
10263          * control and array devices.
10264          */
10265         if (port_type == CTL_PORT_SCSI)
10266                 inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10267                                     SID_SPI_IUS;
10268
10269         /* SAM-5 (no version claimed) */
10270         scsi_ulto2b(0x00A0, inq_ptr->version1);
10271         /* SPC-4 (no version claimed) */
10272         scsi_ulto2b(0x0460, inq_ptr->version2);
10273         if (port_type == CTL_PORT_FC) {
10274                 /* FCP-2 ANSI INCITS.350:2003 */
10275                 scsi_ulto2b(0x0917, inq_ptr->version3);
10276         } else if (port_type == CTL_PORT_SCSI) {
10277                 /* SPI-4 ANSI INCITS.362:200x */
10278                 scsi_ulto2b(0x0B56, inq_ptr->version3);
10279         } else if (port_type == CTL_PORT_ISCSI) {
10280                 /* iSCSI (no version claimed) */
10281                 scsi_ulto2b(0x0960, inq_ptr->version3);
10282         } else if (port_type == CTL_PORT_SAS) {
10283                 /* SAS (no version claimed) */
10284                 scsi_ulto2b(0x0BE0, inq_ptr->version3);
10285         }
10286
10287         if (lun == NULL) {
10288                 /* SBC-4 (no version claimed) */
10289                 scsi_ulto2b(0x0600, inq_ptr->version4);
10290         } else {
10291                 switch (lun->be_lun->lun_type) {
10292                 case T_DIRECT:
10293                         /* SBC-4 (no version claimed) */
10294                         scsi_ulto2b(0x0600, inq_ptr->version4);
10295                         break;
10296                 case T_PROCESSOR:
10297                         break;
10298                 case T_CDROM:
10299                         /* MMC-6 (no version claimed) */
10300                         scsi_ulto2b(0x04E0, inq_ptr->version4);
10301                         break;
10302                 default:
10303                         break;
10304                 }
10305         }
10306
10307         ctl_set_success(ctsio);
10308         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10309         ctsio->be_move_done = ctl_config_move_done;
10310         ctl_datamove((union ctl_io *)ctsio);
10311         return (CTL_RETVAL_COMPLETE);
10312 }
10313
10314 int
10315 ctl_inquiry(struct ctl_scsiio *ctsio)
10316 {
10317         struct scsi_inquiry *cdb;
10318         int retval;
10319
10320         CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10321
10322         cdb = (struct scsi_inquiry *)ctsio->cdb;
10323         if (cdb->byte2 & SI_EVPD)
10324                 retval = ctl_inquiry_evpd(ctsio);
10325         else if (cdb->page_code == 0)
10326                 retval = ctl_inquiry_std(ctsio);
10327         else {
10328                 ctl_set_invalid_field(ctsio,
10329                                       /*sks_valid*/ 1,
10330                                       /*command*/ 1,
10331                                       /*field*/ 2,
10332                                       /*bit_valid*/ 0,
10333                                       /*bit*/ 0);
10334                 ctl_done((union ctl_io *)ctsio);
10335                 return (CTL_RETVAL_COMPLETE);
10336         }
10337
10338         return (retval);
10339 }
10340
10341 int
10342 ctl_get_config(struct ctl_scsiio *ctsio)
10343 {
10344         struct scsi_get_config_header *hdr;
10345         struct scsi_get_config_feature *feature;
10346         struct scsi_get_config *cdb;
10347         struct ctl_lun *lun;
10348         uint32_t alloc_len, data_len;
10349         int rt, starting;
10350
10351         lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10352         cdb = (struct scsi_get_config *)ctsio->cdb;
10353         rt = (cdb->rt & SGC_RT_MASK);
10354         starting = scsi_2btoul(cdb->starting_feature);
10355         alloc_len = scsi_2btoul(cdb->length);
10356
10357         data_len = sizeof(struct scsi_get_config_header) +
10358             sizeof(struct scsi_get_config_feature) + 8 +
10359             sizeof(struct scsi_get_config_feature) + 8 +
10360             sizeof(struct scsi_get_config_feature) + 4 +
10361             sizeof(struct scsi_get_config_feature) + 4 +
10362             sizeof(struct scsi_get_config_feature) + 8 +
10363             sizeof(struct scsi_get_config_feature) +
10364             sizeof(struct scsi_get_config_feature) + 4 +
10365             sizeof(struct scsi_get_config_feature) + 4 +
10366             sizeof(struct scsi_get_config_feature) + 4 +
10367             sizeof(struct scsi_get_config_feature) + 4 +
10368             sizeof(struct scsi_get_config_feature) + 4 +
10369             sizeof(struct scsi_get_config_feature) + 4;
10370         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10371         ctsio->kern_sg_entries = 0;
10372         ctsio->kern_data_resid = 0;
10373         ctsio->kern_rel_offset = 0;
10374
10375         hdr = (struct scsi_get_config_header *)ctsio->kern_data_ptr;
10376         if (lun->flags & CTL_LUN_NO_MEDIA)
10377                 scsi_ulto2b(0x0000, hdr->current_profile);
10378         else
10379                 scsi_ulto2b(0x0010, hdr->current_profile);
10380         feature = (struct scsi_get_config_feature *)(hdr + 1);
10381
10382         if (starting > 0x003b)
10383                 goto done;
10384         if (starting > 0x003a)
10385                 goto f3b;
10386         if (starting > 0x002b)
10387                 goto f3a;
10388         if (starting > 0x002a)
10389                 goto f2b;
10390         if (starting > 0x001f)
10391                 goto f2a;
10392         if (starting > 0x001e)
10393                 goto f1f;
10394         if (starting > 0x001d)
10395                 goto f1e;
10396         if (starting > 0x0010)
10397                 goto f1d;
10398         if (starting > 0x0003)
10399                 goto f10;
10400         if (starting > 0x0002)
10401                 goto f3;
10402         if (starting > 0x0001)
10403                 goto f2;
10404         if (starting > 0x0000)
10405                 goto f1;
10406
10407         /* Profile List */
10408         scsi_ulto2b(0x0000, feature->feature_code);
10409         feature->flags = SGC_F_PERSISTENT | SGC_F_CURRENT;
10410         feature->add_length = 8;
10411         scsi_ulto2b(0x0008, &feature->feature_data[0]); /* CD-ROM */
10412         feature->feature_data[2] = 0x00;
10413         scsi_ulto2b(0x0010, &feature->feature_data[4]); /* DVD-ROM */
10414         feature->feature_data[6] = 0x01;
10415         feature = (struct scsi_get_config_feature *)
10416             &feature->feature_data[feature->add_length];
10417
10418 f1:     /* Core */
10419         scsi_ulto2b(0x0001, feature->feature_code);
10420         feature->flags = 0x08 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10421         feature->add_length = 8;
10422         scsi_ulto4b(0x00000000, &feature->feature_data[0]);
10423         feature->feature_data[4] = 0x03;
10424         feature = (struct scsi_get_config_feature *)
10425             &feature->feature_data[feature->add_length];
10426
10427 f2:     /* Morphing */
10428         scsi_ulto2b(0x0002, feature->feature_code);
10429         feature->flags = 0x04 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10430         feature->add_length = 4;
10431         feature->feature_data[0] = 0x02;
10432         feature = (struct scsi_get_config_feature *)
10433             &feature->feature_data[feature->add_length];
10434
10435 f3:     /* Removable Medium */
10436         scsi_ulto2b(0x0003, feature->feature_code);
10437         feature->flags = 0x04 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10438         feature->add_length = 4;
10439         feature->feature_data[0] = 0x39;
10440         feature = (struct scsi_get_config_feature *)
10441             &feature->feature_data[feature->add_length];
10442
10443         if (rt == SGC_RT_CURRENT && (lun->flags & CTL_LUN_NO_MEDIA))
10444                 goto done;
10445
10446 f10:    /* Random Read */
10447         scsi_ulto2b(0x0010, feature->feature_code);
10448         feature->flags = 0x00;
10449         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10450                 feature->flags |= SGC_F_CURRENT;
10451         feature->add_length = 8;
10452         scsi_ulto4b(lun->be_lun->blocksize, &feature->feature_data[0]);
10453         scsi_ulto2b(1, &feature->feature_data[4]);
10454         feature->feature_data[6] = 0x00;
10455         feature = (struct scsi_get_config_feature *)
10456             &feature->feature_data[feature->add_length];
10457
10458 f1d:    /* Multi-Read */
10459         scsi_ulto2b(0x001D, feature->feature_code);
10460         feature->flags = 0x00;
10461         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10462                 feature->flags |= SGC_F_CURRENT;
10463         feature->add_length = 0;
10464         feature = (struct scsi_get_config_feature *)
10465             &feature->feature_data[feature->add_length];
10466
10467 f1e:    /* CD Read */
10468         scsi_ulto2b(0x001E, feature->feature_code);
10469         feature->flags = 0x00;
10470         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10471                 feature->flags |= SGC_F_CURRENT;
10472         feature->add_length = 4;
10473         feature->feature_data[0] = 0x00;
10474         feature = (struct scsi_get_config_feature *)
10475             &feature->feature_data[feature->add_length];
10476
10477 f1f:    /* DVD Read */
10478         scsi_ulto2b(0x001F, feature->feature_code);
10479         feature->flags = 0x08;
10480         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10481                 feature->flags |= SGC_F_CURRENT;
10482         feature->add_length = 4;
10483         feature->feature_data[0] = 0x01;
10484         feature->feature_data[2] = 0x03;
10485         feature = (struct scsi_get_config_feature *)
10486             &feature->feature_data[feature->add_length];
10487
10488 f2a:    /* DVD+RW */
10489         scsi_ulto2b(0x002A, feature->feature_code);
10490         feature->flags = 0x04;
10491         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10492                 feature->flags |= SGC_F_CURRENT;
10493         feature->add_length = 4;
10494         feature->feature_data[0] = 0x00;
10495         feature->feature_data[1] = 0x00;
10496         feature = (struct scsi_get_config_feature *)
10497             &feature->feature_data[feature->add_length];
10498
10499 f2b:    /* DVD+R */
10500         scsi_ulto2b(0x002B, feature->feature_code);
10501         feature->flags = 0x00;
10502         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10503                 feature->flags |= SGC_F_CURRENT;
10504         feature->add_length = 4;
10505         feature->feature_data[0] = 0x00;
10506         feature = (struct scsi_get_config_feature *)
10507             &feature->feature_data[feature->add_length];
10508
10509 f3a:    /* DVD+RW Dual Layer */
10510         scsi_ulto2b(0x003A, feature->feature_code);
10511         feature->flags = 0x00;
10512         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10513                 feature->flags |= SGC_F_CURRENT;
10514         feature->add_length = 4;
10515         feature->feature_data[0] = 0x00;
10516         feature->feature_data[1] = 0x00;
10517         feature = (struct scsi_get_config_feature *)
10518             &feature->feature_data[feature->add_length];
10519
10520 f3b:    /* DVD+R Dual Layer */
10521         scsi_ulto2b(0x003B, feature->feature_code);
10522         feature->flags = 0x00;
10523         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10524                 feature->flags |= SGC_F_CURRENT;
10525         feature->add_length = 4;
10526         feature->feature_data[0] = 0x00;
10527         feature = (struct scsi_get_config_feature *)
10528             &feature->feature_data[feature->add_length];
10529
10530 done:
10531         data_len = (uint8_t *)feature - (uint8_t *)hdr;
10532         if (rt == SGC_RT_SPECIFIC && data_len > 4) {
10533                 feature = (struct scsi_get_config_feature *)(hdr + 1);
10534                 if (scsi_2btoul(feature->feature_code) == starting)
10535                         feature = (struct scsi_get_config_feature *)
10536                             &feature->feature_data[feature->add_length];
10537                 data_len = (uint8_t *)feature - (uint8_t *)hdr;
10538         }
10539         scsi_ulto4b(data_len - 4, hdr->data_length);
10540         if (data_len < alloc_len) {
10541                 ctsio->residual = alloc_len - data_len;
10542                 ctsio->kern_data_len = data_len;
10543                 ctsio->kern_total_len = data_len;
10544         } else {
10545                 ctsio->residual = 0;
10546                 ctsio->kern_data_len = alloc_len;
10547                 ctsio->kern_total_len = alloc_len;
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_get_event_status(struct ctl_scsiio *ctsio)
10559 {
10560         struct scsi_get_event_status_header *hdr;
10561         struct scsi_get_event_status *cdb;
10562         struct ctl_lun *lun;
10563         uint32_t alloc_len, data_len;
10564         int notif_class;
10565
10566         lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10567         cdb = (struct scsi_get_event_status *)ctsio->cdb;
10568         if ((cdb->byte2 & SGESN_POLLED) == 0) {
10569                 ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 1,
10570                     /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
10571                 ctl_done((union ctl_io *)ctsio);
10572                 return (CTL_RETVAL_COMPLETE);
10573         }
10574         notif_class = cdb->notif_class;
10575         alloc_len = scsi_2btoul(cdb->length);
10576
10577         data_len = sizeof(struct scsi_get_event_status_header);
10578         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10579         ctsio->kern_sg_entries = 0;
10580         ctsio->kern_data_resid = 0;
10581         ctsio->kern_rel_offset = 0;
10582
10583         if (data_len < alloc_len) {
10584                 ctsio->residual = alloc_len - data_len;
10585                 ctsio->kern_data_len = data_len;
10586                 ctsio->kern_total_len = data_len;
10587         } else {
10588                 ctsio->residual = 0;
10589                 ctsio->kern_data_len = alloc_len;
10590                 ctsio->kern_total_len = alloc_len;
10591         }
10592
10593         hdr = (struct scsi_get_event_status_header *)ctsio->kern_data_ptr;
10594         scsi_ulto2b(0, hdr->descr_length);
10595         hdr->nea_class = SGESN_NEA;
10596         hdr->supported_class = 0;
10597
10598         ctl_set_success(ctsio);
10599         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10600         ctsio->be_move_done = ctl_config_move_done;
10601         ctl_datamove((union ctl_io *)ctsio);
10602         return (CTL_RETVAL_COMPLETE);
10603 }
10604
10605 int
10606 ctl_mechanism_status(struct ctl_scsiio *ctsio)
10607 {
10608         struct scsi_mechanism_status_header *hdr;
10609         struct scsi_mechanism_status *cdb;
10610         struct ctl_lun *lun;
10611         uint32_t alloc_len, data_len;
10612
10613         lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10614         cdb = (struct scsi_mechanism_status *)ctsio->cdb;
10615         alloc_len = scsi_2btoul(cdb->length);
10616
10617         data_len = sizeof(struct scsi_mechanism_status_header);
10618         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10619         ctsio->kern_sg_entries = 0;
10620         ctsio->kern_data_resid = 0;
10621         ctsio->kern_rel_offset = 0;
10622
10623         if (data_len < alloc_len) {
10624                 ctsio->residual = alloc_len - data_len;
10625                 ctsio->kern_data_len = data_len;
10626                 ctsio->kern_total_len = data_len;
10627         } else {
10628                 ctsio->residual = 0;
10629                 ctsio->kern_data_len = alloc_len;
10630                 ctsio->kern_total_len = alloc_len;
10631         }
10632
10633         hdr = (struct scsi_mechanism_status_header *)ctsio->kern_data_ptr;
10634         hdr->state1 = 0x00;
10635         hdr->state2 = 0xe0;
10636         scsi_ulto3b(0, hdr->lba);
10637         hdr->slots_num = 0;
10638         scsi_ulto2b(0, hdr->slots_length);
10639
10640         ctl_set_success(ctsio);
10641         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10642         ctsio->be_move_done = ctl_config_move_done;
10643         ctl_datamove((union ctl_io *)ctsio);
10644         return (CTL_RETVAL_COMPLETE);
10645 }
10646
10647 static void
10648 ctl_ultomsf(uint32_t lba, uint8_t *buf)
10649 {
10650
10651         lba += 150;
10652         buf[0] = 0;
10653         buf[1] = bin2bcd((lba / 75) / 60);
10654         buf[2] = bin2bcd((lba / 75) % 60);
10655         buf[3] = bin2bcd(lba % 75);
10656 }
10657
10658 int
10659 ctl_read_toc(struct ctl_scsiio *ctsio)
10660 {
10661         struct scsi_read_toc_hdr *hdr;
10662         struct scsi_read_toc_type01_descr *descr;
10663         struct scsi_read_toc *cdb;
10664         struct ctl_lun *lun;
10665         uint32_t alloc_len, data_len;
10666         int format, msf;
10667
10668         lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10669         cdb = (struct scsi_read_toc *)ctsio->cdb;
10670         msf = (cdb->byte2 & CD_MSF) != 0;
10671         format = cdb->format;
10672         alloc_len = scsi_2btoul(cdb->data_len);
10673
10674         data_len = sizeof(struct scsi_read_toc_hdr);
10675         if (format == 0)
10676                 data_len += 2 * sizeof(struct scsi_read_toc_type01_descr);
10677         else
10678                 data_len += sizeof(struct scsi_read_toc_type01_descr);
10679         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10680         ctsio->kern_sg_entries = 0;
10681         ctsio->kern_data_resid = 0;
10682         ctsio->kern_rel_offset = 0;
10683
10684         if (data_len < alloc_len) {
10685                 ctsio->residual = alloc_len - data_len;
10686                 ctsio->kern_data_len = data_len;
10687                 ctsio->kern_total_len = data_len;
10688         } else {
10689                 ctsio->residual = 0;
10690                 ctsio->kern_data_len = alloc_len;
10691                 ctsio->kern_total_len = alloc_len;
10692         }
10693
10694         hdr = (struct scsi_read_toc_hdr *)ctsio->kern_data_ptr;
10695         if (format == 0) {
10696                 scsi_ulto2b(0x12, hdr->data_length);
10697                 hdr->first = 1;
10698                 hdr->last = 1;
10699                 descr = (struct scsi_read_toc_type01_descr *)(hdr + 1);
10700                 descr->addr_ctl = 0x14;
10701                 descr->track_number = 1;
10702                 if (msf)
10703                         ctl_ultomsf(0, descr->track_start);
10704                 else
10705                         scsi_ulto4b(0, descr->track_start);
10706                 descr++;
10707                 descr->addr_ctl = 0x14;
10708                 descr->track_number = 0xaa;
10709                 if (msf)
10710                         ctl_ultomsf(lun->be_lun->maxlba+1, descr->track_start);
10711                 else
10712                         scsi_ulto4b(lun->be_lun->maxlba+1, descr->track_start);
10713         } else {
10714                 scsi_ulto2b(0x0a, hdr->data_length);
10715                 hdr->first = 1;
10716                 hdr->last = 1;
10717                 descr = (struct scsi_read_toc_type01_descr *)(hdr + 1);
10718                 descr->addr_ctl = 0x14;
10719                 descr->track_number = 1;
10720                 if (msf)
10721                         ctl_ultomsf(0, descr->track_start);
10722                 else
10723                         scsi_ulto4b(0, descr->track_start);
10724         }
10725
10726         ctl_set_success(ctsio);
10727         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10728         ctsio->be_move_done = ctl_config_move_done;
10729         ctl_datamove((union ctl_io *)ctsio);
10730         return (CTL_RETVAL_COMPLETE);
10731 }
10732
10733 /*
10734  * For known CDB types, parse the LBA and length.
10735  */
10736 static int
10737 ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len)
10738 {
10739         if (io->io_hdr.io_type != CTL_IO_SCSI)
10740                 return (1);
10741
10742         switch (io->scsiio.cdb[0]) {
10743         case COMPARE_AND_WRITE: {
10744                 struct scsi_compare_and_write *cdb;
10745
10746                 cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10747
10748                 *lba = scsi_8btou64(cdb->addr);
10749                 *len = cdb->length;
10750                 break;
10751         }
10752         case READ_6:
10753         case WRITE_6: {
10754                 struct scsi_rw_6 *cdb;
10755
10756                 cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10757
10758                 *lba = scsi_3btoul(cdb->addr);
10759                 /* only 5 bits are valid in the most significant address byte */
10760                 *lba &= 0x1fffff;
10761                 *len = cdb->length;
10762                 break;
10763         }
10764         case READ_10:
10765         case WRITE_10: {
10766                 struct scsi_rw_10 *cdb;
10767
10768                 cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10769
10770                 *lba = scsi_4btoul(cdb->addr);
10771                 *len = scsi_2btoul(cdb->length);
10772                 break;
10773         }
10774         case WRITE_VERIFY_10: {
10775                 struct scsi_write_verify_10 *cdb;
10776
10777                 cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10778
10779                 *lba = scsi_4btoul(cdb->addr);
10780                 *len = scsi_2btoul(cdb->length);
10781                 break;
10782         }
10783         case READ_12:
10784         case WRITE_12: {
10785                 struct scsi_rw_12 *cdb;
10786
10787                 cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10788
10789                 *lba = scsi_4btoul(cdb->addr);
10790                 *len = scsi_4btoul(cdb->length);
10791                 break;
10792         }
10793         case WRITE_VERIFY_12: {
10794                 struct scsi_write_verify_12 *cdb;
10795
10796                 cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10797
10798                 *lba = scsi_4btoul(cdb->addr);
10799                 *len = scsi_4btoul(cdb->length);
10800                 break;
10801         }
10802         case READ_16:
10803         case WRITE_16: {
10804                 struct scsi_rw_16 *cdb;
10805
10806                 cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10807
10808                 *lba = scsi_8btou64(cdb->addr);
10809                 *len = scsi_4btoul(cdb->length);
10810                 break;
10811         }
10812         case WRITE_ATOMIC_16: {
10813                 struct scsi_write_atomic_16 *cdb;
10814
10815                 cdb = (struct scsi_write_atomic_16 *)io->scsiio.cdb;
10816
10817                 *lba = scsi_8btou64(cdb->addr);
10818                 *len = scsi_2btoul(cdb->length);
10819                 break;
10820         }
10821         case WRITE_VERIFY_16: {
10822                 struct scsi_write_verify_16 *cdb;
10823
10824                 cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10825
10826                 *lba = scsi_8btou64(cdb->addr);
10827                 *len = scsi_4btoul(cdb->length);
10828                 break;
10829         }
10830         case WRITE_SAME_10: {
10831                 struct scsi_write_same_10 *cdb;
10832
10833                 cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10834
10835                 *lba = scsi_4btoul(cdb->addr);
10836                 *len = scsi_2btoul(cdb->length);
10837                 break;
10838         }
10839         case WRITE_SAME_16: {
10840                 struct scsi_write_same_16 *cdb;
10841
10842                 cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10843
10844                 *lba = scsi_8btou64(cdb->addr);
10845                 *len = scsi_4btoul(cdb->length);
10846                 break;
10847         }
10848         case VERIFY_10: {
10849                 struct scsi_verify_10 *cdb;
10850
10851                 cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10852
10853                 *lba = scsi_4btoul(cdb->addr);
10854                 *len = scsi_2btoul(cdb->length);
10855                 break;
10856         }
10857         case VERIFY_12: {
10858                 struct scsi_verify_12 *cdb;
10859
10860                 cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10861
10862                 *lba = scsi_4btoul(cdb->addr);
10863                 *len = scsi_4btoul(cdb->length);
10864                 break;
10865         }
10866         case VERIFY_16: {
10867                 struct scsi_verify_16 *cdb;
10868
10869                 cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10870
10871                 *lba = scsi_8btou64(cdb->addr);
10872                 *len = scsi_4btoul(cdb->length);
10873                 break;
10874         }
10875         case UNMAP: {
10876                 *lba = 0;
10877                 *len = UINT64_MAX;
10878                 break;
10879         }
10880         case SERVICE_ACTION_IN: {       /* GET LBA STATUS */
10881                 struct scsi_get_lba_status *cdb;
10882
10883                 cdb = (struct scsi_get_lba_status *)io->scsiio.cdb;
10884                 *lba = scsi_8btou64(cdb->addr);
10885                 *len = UINT32_MAX;
10886                 break;
10887         }
10888         default:
10889                 return (1);
10890                 break; /* NOTREACHED */
10891         }
10892
10893         return (0);
10894 }
10895
10896 static ctl_action
10897 ctl_extent_check_lba(uint64_t lba1, uint64_t len1, uint64_t lba2, uint64_t len2,
10898     bool seq)
10899 {
10900         uint64_t endlba1, endlba2;
10901
10902         endlba1 = lba1 + len1 - (seq ? 0 : 1);
10903         endlba2 = lba2 + len2 - 1;
10904
10905         if ((endlba1 < lba2) || (endlba2 < lba1))
10906                 return (CTL_ACTION_PASS);
10907         else
10908                 return (CTL_ACTION_BLOCK);
10909 }
10910
10911 static int
10912 ctl_extent_check_unmap(union ctl_io *io, uint64_t lba2, uint64_t len2)
10913 {
10914         struct ctl_ptr_len_flags *ptrlen;
10915         struct scsi_unmap_desc *buf, *end, *range;
10916         uint64_t lba;
10917         uint32_t len;
10918
10919         /* If not UNMAP -- go other way. */
10920         if (io->io_hdr.io_type != CTL_IO_SCSI ||
10921             io->scsiio.cdb[0] != UNMAP)
10922                 return (CTL_ACTION_ERROR);
10923
10924         /* If UNMAP without data -- block and wait for data. */
10925         ptrlen = (struct ctl_ptr_len_flags *)
10926             &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
10927         if ((io->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0 ||
10928             ptrlen->ptr == NULL)
10929                 return (CTL_ACTION_BLOCK);
10930
10931         /* UNMAP with data -- check for collision. */
10932         buf = (struct scsi_unmap_desc *)ptrlen->ptr;
10933         end = buf + ptrlen->len / sizeof(*buf);
10934         for (range = buf; range < end; range++) {
10935                 lba = scsi_8btou64(range->lba);
10936                 len = scsi_4btoul(range->length);
10937                 if ((lba < lba2 + len2) && (lba + len > lba2))
10938                         return (CTL_ACTION_BLOCK);
10939         }
10940         return (CTL_ACTION_PASS);
10941 }
10942
10943 static ctl_action
10944 ctl_extent_check(union ctl_io *io1, union ctl_io *io2, bool seq)
10945 {
10946         uint64_t lba1, lba2;
10947         uint64_t len1, len2;
10948         int retval;
10949
10950         if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10951                 return (CTL_ACTION_ERROR);
10952
10953         retval = ctl_extent_check_unmap(io1, lba2, len2);
10954         if (retval != CTL_ACTION_ERROR)
10955                 return (retval);
10956
10957         if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10958                 return (CTL_ACTION_ERROR);
10959
10960         if (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE)
10961                 seq = FALSE;
10962         return (ctl_extent_check_lba(lba1, len1, lba2, len2, seq));
10963 }
10964
10965 static ctl_action
10966 ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2)
10967 {
10968         uint64_t lba1, lba2;
10969         uint64_t len1, len2;
10970
10971         if (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE)
10972                 return (CTL_ACTION_PASS);
10973         if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10974                 return (CTL_ACTION_ERROR);
10975         if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10976                 return (CTL_ACTION_ERROR);
10977
10978         if (lba1 + len1 == lba2)
10979                 return (CTL_ACTION_BLOCK);
10980         return (CTL_ACTION_PASS);
10981 }
10982
10983 static ctl_action
10984 ctl_check_for_blockage(struct ctl_lun *lun, union ctl_io *pending_io,
10985     union ctl_io *ooa_io)
10986 {
10987         const struct ctl_cmd_entry *pending_entry, *ooa_entry;
10988         const ctl_serialize_action *serialize_row;
10989
10990         /*
10991          * The initiator attempted multiple untagged commands at the same
10992          * time.  Can't do that.
10993          */
10994         if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10995          && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10996          && ((pending_io->io_hdr.nexus.targ_port ==
10997               ooa_io->io_hdr.nexus.targ_port)
10998           && (pending_io->io_hdr.nexus.initid ==
10999               ooa_io->io_hdr.nexus.initid))
11000          && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
11001               CTL_FLAG_STATUS_SENT)) == 0))
11002                 return (CTL_ACTION_OVERLAP);
11003
11004         /*
11005          * The initiator attempted to send multiple tagged commands with
11006          * the same ID.  (It's fine if different initiators have the same
11007          * tag ID.)
11008          *
11009          * Even if all of those conditions are true, we don't kill the I/O
11010          * if the command ahead of us has been aborted.  We won't end up
11011          * sending it to the FETD, and it's perfectly legal to resend a
11012          * command with the same tag number as long as the previous
11013          * instance of this tag number has been aborted somehow.
11014          */
11015         if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
11016          && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
11017          && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
11018          && ((pending_io->io_hdr.nexus.targ_port ==
11019               ooa_io->io_hdr.nexus.targ_port)
11020           && (pending_io->io_hdr.nexus.initid ==
11021               ooa_io->io_hdr.nexus.initid))
11022          && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
11023               CTL_FLAG_STATUS_SENT)) == 0))
11024                 return (CTL_ACTION_OVERLAP_TAG);
11025
11026         /*
11027          * If we get a head of queue tag, SAM-3 says that we should
11028          * immediately execute it.
11029          *
11030          * What happens if this command would normally block for some other
11031          * reason?  e.g. a request sense with a head of queue tag
11032          * immediately after a write.  Normally that would block, but this
11033          * will result in its getting executed immediately...
11034          *
11035          * We currently return "pass" instead of "skip", so we'll end up
11036          * going through the rest of the queue to check for overlapped tags.
11037          *
11038          * XXX KDM check for other types of blockage first??
11039          */
11040         if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
11041                 return (CTL_ACTION_PASS);
11042
11043         /*
11044          * Ordered tags have to block until all items ahead of them
11045          * have completed.  If we get called with an ordered tag, we always
11046          * block, if something else is ahead of us in the queue.
11047          */
11048         if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
11049                 return (CTL_ACTION_BLOCK);
11050
11051         /*
11052          * Simple tags get blocked until all head of queue and ordered tags
11053          * ahead of them have completed.  I'm lumping untagged commands in
11054          * with simple tags here.  XXX KDM is that the right thing to do?
11055          */
11056         if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
11057           || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
11058          && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
11059           || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
11060                 return (CTL_ACTION_BLOCK);
11061
11062         pending_entry = ctl_get_cmd_entry(&pending_io->scsiio, NULL);
11063         KASSERT(pending_entry->seridx < CTL_SERIDX_COUNT,
11064             ("%s: Invalid seridx %d for pending CDB %02x %02x @ %p",
11065              __func__, pending_entry->seridx, pending_io->scsiio.cdb[0],
11066              pending_io->scsiio.cdb[1], pending_io));
11067         ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio, NULL);
11068         if (ooa_entry->seridx == CTL_SERIDX_INVLD)
11069                 return (CTL_ACTION_PASS); /* Unsupported command in OOA queue */
11070         KASSERT(ooa_entry->seridx < CTL_SERIDX_COUNT,
11071             ("%s: Invalid seridx %d for ooa CDB %02x %02x @ %p",
11072              __func__, ooa_entry->seridx, ooa_io->scsiio.cdb[0],
11073              ooa_io->scsiio.cdb[1], ooa_io));
11074
11075         serialize_row = ctl_serialize_table[ooa_entry->seridx];
11076
11077         switch (serialize_row[pending_entry->seridx]) {
11078         case CTL_SER_BLOCK:
11079                 return (CTL_ACTION_BLOCK);
11080         case CTL_SER_EXTENT:
11081                 return (ctl_extent_check(ooa_io, pending_io,
11082                     (lun->be_lun && lun->be_lun->serseq == CTL_LUN_SERSEQ_ON)));
11083         case CTL_SER_EXTENTOPT:
11084                 if ((lun->MODE_CTRL.queue_flags & SCP_QUEUE_ALG_MASK) !=
11085                     SCP_QUEUE_ALG_UNRESTRICTED)
11086                         return (ctl_extent_check(ooa_io, pending_io,
11087                             (lun->be_lun &&
11088                              lun->be_lun->serseq == CTL_LUN_SERSEQ_ON)));
11089                 return (CTL_ACTION_PASS);
11090         case CTL_SER_EXTENTSEQ:
11091                 if (lun->be_lun && lun->be_lun->serseq != CTL_LUN_SERSEQ_OFF)
11092                         return (ctl_extent_check_seq(ooa_io, pending_io));
11093                 return (CTL_ACTION_PASS);
11094         case CTL_SER_PASS:
11095                 return (CTL_ACTION_PASS);
11096         case CTL_SER_BLOCKOPT:
11097                 if ((lun->MODE_CTRL.queue_flags & SCP_QUEUE_ALG_MASK) !=
11098                     SCP_QUEUE_ALG_UNRESTRICTED)
11099                         return (CTL_ACTION_BLOCK);
11100                 return (CTL_ACTION_PASS);
11101         case CTL_SER_SKIP:
11102                 return (CTL_ACTION_SKIP);
11103         default:
11104                 panic("%s: Invalid serialization value %d for %d => %d",
11105                     __func__, serialize_row[pending_entry->seridx],
11106                     pending_entry->seridx, ooa_entry->seridx);
11107         }
11108
11109         return (CTL_ACTION_ERROR);
11110 }
11111
11112 /*
11113  * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
11114  * Assumptions:
11115  * - pending_io is generally either incoming, or on the blocked queue
11116  * - starting I/O is the I/O we want to start the check with.
11117  */
11118 static ctl_action
11119 ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
11120               union ctl_io *starting_io)
11121 {
11122         union ctl_io *ooa_io;
11123         ctl_action action;
11124
11125         mtx_assert(&lun->lun_lock, MA_OWNED);
11126
11127         /*
11128          * Run back along the OOA queue, starting with the current
11129          * blocked I/O and going through every I/O before it on the
11130          * queue.  If starting_io is NULL, we'll just end up returning
11131          * CTL_ACTION_PASS.
11132          */
11133         for (ooa_io = starting_io; ooa_io != NULL;
11134              ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
11135              ooa_links)){
11136
11137                 /*
11138                  * This routine just checks to see whether
11139                  * cur_blocked is blocked by ooa_io, which is ahead
11140                  * of it in the queue.  It doesn't queue/dequeue
11141                  * cur_blocked.
11142                  */
11143                 action = ctl_check_for_blockage(lun, pending_io, ooa_io);
11144                 switch (action) {
11145                 case CTL_ACTION_BLOCK:
11146                 case CTL_ACTION_OVERLAP:
11147                 case CTL_ACTION_OVERLAP_TAG:
11148                 case CTL_ACTION_SKIP:
11149                 case CTL_ACTION_ERROR:
11150                         return (action);
11151                         break; /* NOTREACHED */
11152                 case CTL_ACTION_PASS:
11153                         break;
11154                 default:
11155                         panic("%s: Invalid action %d\n", __func__, action);
11156                 }
11157         }
11158
11159         return (CTL_ACTION_PASS);
11160 }
11161
11162 /*
11163  * Assumptions:
11164  * - An I/O has just completed, and has been removed from the per-LUN OOA
11165  *   queue, so some items on the blocked queue may now be unblocked.
11166  */
11167 static int
11168 ctl_check_blocked(struct ctl_lun *lun)
11169 {
11170         struct ctl_softc *softc = lun->ctl_softc;
11171         union ctl_io *cur_blocked, *next_blocked;
11172
11173         mtx_assert(&lun->lun_lock, MA_OWNED);
11174
11175         /*
11176          * Run forward from the head of the blocked queue, checking each
11177          * entry against the I/Os prior to it on the OOA queue to see if
11178          * there is still any blockage.
11179          *
11180          * We cannot use the TAILQ_FOREACH() macro, because it can't deal
11181          * with our removing a variable on it while it is traversing the
11182          * list.
11183          */
11184         for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
11185              cur_blocked != NULL; cur_blocked = next_blocked) {
11186                 union ctl_io *prev_ooa;
11187                 ctl_action action;
11188
11189                 next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
11190                                                           blocked_links);
11191
11192                 prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
11193                                                       ctl_ooaq, ooa_links);
11194
11195                 /*
11196                  * If cur_blocked happens to be the first item in the OOA
11197                  * queue now, prev_ooa will be NULL, and the action
11198                  * returned will just be CTL_ACTION_PASS.
11199                  */
11200                 action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
11201
11202                 switch (action) {
11203                 case CTL_ACTION_BLOCK:
11204                         /* Nothing to do here, still blocked */
11205                         break;
11206                 case CTL_ACTION_OVERLAP:
11207                 case CTL_ACTION_OVERLAP_TAG:
11208                         /*
11209                          * This shouldn't happen!  In theory we've already
11210                          * checked this command for overlap...
11211                          */
11212                         break;
11213                 case CTL_ACTION_PASS:
11214                 case CTL_ACTION_SKIP: {
11215                         const struct ctl_cmd_entry *entry;
11216
11217                         /*
11218                          * The skip case shouldn't happen, this transaction
11219                          * should have never made it onto the blocked queue.
11220                          */
11221                         /*
11222                          * This I/O is no longer blocked, we can remove it
11223                          * from the blocked queue.  Since this is a TAILQ
11224                          * (doubly linked list), we can do O(1) removals
11225                          * from any place on the list.
11226                          */
11227                         TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
11228                                      blocked_links);
11229                         cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11230
11231                         if ((softc->ha_mode != CTL_HA_MODE_XFER) &&
11232                             (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)){
11233                                 /*
11234                                  * Need to send IO back to original side to
11235                                  * run
11236                                  */
11237                                 union ctl_ha_msg msg_info;
11238
11239                                 cur_blocked->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11240                                 msg_info.hdr.original_sc =
11241                                         cur_blocked->io_hdr.original_sc;
11242                                 msg_info.hdr.serializing_sc = cur_blocked;
11243                                 msg_info.hdr.msg_type = CTL_MSG_R2R;
11244                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11245                                     sizeof(msg_info.hdr), M_NOWAIT);
11246                                 break;
11247                         }
11248                         entry = ctl_get_cmd_entry(&cur_blocked->scsiio, NULL);
11249
11250                         /*
11251                          * Check this I/O for LUN state changes that may
11252                          * have happened while this command was blocked.
11253                          * The LUN state may have been changed by a command
11254                          * ahead of us in the queue, so we need to re-check
11255                          * for any states that can be caused by SCSI
11256                          * commands.
11257                          */
11258                         if (ctl_scsiio_lun_check(lun, entry,
11259                                                  &cur_blocked->scsiio) == 0) {
11260                                 cur_blocked->io_hdr.flags |=
11261                                                       CTL_FLAG_IS_WAS_ON_RTR;
11262                                 ctl_enqueue_rtr(cur_blocked);
11263                         } else
11264                                 ctl_done(cur_blocked);
11265                         break;
11266                 }
11267                 default:
11268                         /*
11269                          * This probably shouldn't happen -- we shouldn't
11270                          * get CTL_ACTION_ERROR, or anything else.
11271                          */
11272                         break;
11273                 }
11274         }
11275
11276         return (CTL_RETVAL_COMPLETE);
11277 }
11278
11279 /*
11280  * This routine (with one exception) checks LUN flags that can be set by
11281  * commands ahead of us in the OOA queue.  These flags have to be checked
11282  * when a command initially comes in, and when we pull a command off the
11283  * blocked queue and are preparing to execute it.  The reason we have to
11284  * check these flags for commands on the blocked queue is that the LUN
11285  * state may have been changed by a command ahead of us while we're on the
11286  * blocked queue.
11287  *
11288  * Ordering is somewhat important with these checks, so please pay
11289  * careful attention to the placement of any new checks.
11290  */
11291 static int
11292 ctl_scsiio_lun_check(struct ctl_lun *lun,
11293     const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
11294 {
11295         struct ctl_softc *softc = lun->ctl_softc;
11296         int retval;
11297         uint32_t residx;
11298
11299         retval = 0;
11300
11301         mtx_assert(&lun->lun_lock, MA_OWNED);
11302
11303         /*
11304          * If this shelf is a secondary shelf controller, we may have to
11305          * reject some commands disallowed by HA mode and link state.
11306          */
11307         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11308                 if (softc->ha_link == CTL_HA_LINK_OFFLINE &&
11309                     (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
11310                         ctl_set_lun_unavail(ctsio);
11311                         retval = 1;
11312                         goto bailout;
11313                 }
11314                 if ((lun->flags & CTL_LUN_PEER_SC_PRIMARY) == 0 &&
11315                     (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
11316                         ctl_set_lun_transit(ctsio);
11317                         retval = 1;
11318                         goto bailout;
11319                 }
11320                 if (softc->ha_mode == CTL_HA_MODE_ACT_STBY &&
11321                     (entry->flags & CTL_CMD_FLAG_OK_ON_STANDBY) == 0) {
11322                         ctl_set_lun_standby(ctsio);
11323                         retval = 1;
11324                         goto bailout;
11325                 }
11326
11327                 /* The rest of checks are only done on executing side */
11328                 if (softc->ha_mode == CTL_HA_MODE_XFER)
11329                         goto bailout;
11330         }
11331
11332         if (entry->pattern & CTL_LUN_PAT_WRITE) {
11333                 if (lun->be_lun &&
11334                     lun->be_lun->flags & CTL_LUN_FLAG_READONLY) {
11335                         ctl_set_hw_write_protected(ctsio);
11336                         retval = 1;
11337                         goto bailout;
11338                 }
11339                 if ((lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0) {
11340                         ctl_set_sense(ctsio, /*current_error*/ 1,
11341                             /*sense_key*/ SSD_KEY_DATA_PROTECT,
11342                             /*asc*/ 0x27, /*ascq*/ 0x02, SSD_ELEM_NONE);
11343                         retval = 1;
11344                         goto bailout;
11345                 }
11346         }
11347
11348         /*
11349          * Check for a reservation conflict.  If this command isn't allowed
11350          * even on reserved LUNs, and if this initiator isn't the one who
11351          * reserved us, reject the command with a reservation conflict.
11352          */
11353         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11354         if ((lun->flags & CTL_LUN_RESERVED)
11355          && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
11356                 if (lun->res_idx != residx) {
11357                         ctl_set_reservation_conflict(ctsio);
11358                         retval = 1;
11359                         goto bailout;
11360                 }
11361         }
11362
11363         if ((lun->flags & CTL_LUN_PR_RESERVED) == 0 ||
11364             (entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV)) {
11365                 /* No reservation or command is allowed. */;
11366         } else if ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_WRESV) &&
11367             (lun->pr_res_type == SPR_TYPE_WR_EX ||
11368              lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
11369              lun->pr_res_type == SPR_TYPE_WR_EX_AR)) {
11370                 /* The command is allowed for Write Exclusive resv. */;
11371         } else {
11372                 /*
11373                  * if we aren't registered or it's a res holder type
11374                  * reservation and this isn't the res holder then set a
11375                  * conflict.
11376                  */
11377                 if (ctl_get_prkey(lun, residx) == 0 ||
11378                     (residx != lun->pr_res_idx && lun->pr_res_type < 4)) {
11379                         ctl_set_reservation_conflict(ctsio);
11380                         retval = 1;
11381                         goto bailout;
11382                 }
11383         }
11384
11385         if ((entry->flags & CTL_CMD_FLAG_OK_ON_NO_MEDIA) == 0) {
11386                 if (lun->flags & CTL_LUN_EJECTED)
11387                         ctl_set_lun_ejected(ctsio);
11388                 else if (lun->flags & CTL_LUN_NO_MEDIA) {
11389                         if (lun->flags & CTL_LUN_REMOVABLE)
11390                                 ctl_set_lun_no_media(ctsio);
11391                         else
11392                                 ctl_set_lun_int_reqd(ctsio);
11393                 } else if (lun->flags & CTL_LUN_STOPPED)
11394                         ctl_set_lun_stopped(ctsio);
11395                 else
11396                         goto bailout;
11397                 retval = 1;
11398                 goto bailout;
11399         }
11400
11401 bailout:
11402         return (retval);
11403 }
11404
11405 static void
11406 ctl_failover_io(union ctl_io *io, int have_lock)
11407 {
11408         ctl_set_busy(&io->scsiio);
11409         ctl_done(io);
11410 }
11411
11412 static void
11413 ctl_failover_lun(union ctl_io *rio)
11414 {
11415         struct ctl_softc *softc = control_softc;
11416         struct ctl_lun *lun;
11417         struct ctl_io_hdr *io, *next_io;
11418         uint32_t targ_lun;
11419
11420         targ_lun = rio->io_hdr.nexus.targ_mapped_lun;
11421         CTL_DEBUG_PRINT(("FAILOVER for lun %ju\n", targ_lun));
11422
11423         /* Find and lock the LUN. */
11424         mtx_lock(&softc->ctl_lock);
11425         if ((targ_lun < CTL_MAX_LUNS) &&
11426             ((lun = softc->ctl_luns[targ_lun]) != NULL)) {
11427                 mtx_lock(&lun->lun_lock);
11428                 mtx_unlock(&softc->ctl_lock);
11429                 if (lun->flags & CTL_LUN_DISABLED) {
11430                         mtx_unlock(&lun->lun_lock);
11431                         return;
11432                 }
11433         } else {
11434                 mtx_unlock(&softc->ctl_lock);
11435                 return;
11436         }
11437
11438         if (softc->ha_mode == CTL_HA_MODE_XFER) {
11439                 TAILQ_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
11440                         /* We are master */
11441                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11442                                 if (io->flags & CTL_FLAG_IO_ACTIVE) {
11443                                         io->flags |= CTL_FLAG_ABORT;
11444                                         io->flags |= CTL_FLAG_FAILOVER;
11445                                 } else { /* This can be only due to DATAMOVE */
11446                                         io->msg_type = CTL_MSG_DATAMOVE_DONE;
11447                                         io->flags &= ~CTL_FLAG_DMA_INPROG;
11448                                         io->flags |= CTL_FLAG_IO_ACTIVE;
11449                                         io->port_status = 31340;
11450                                         ctl_enqueue_isc((union ctl_io *)io);
11451                                 }
11452                         }
11453                         /* We are slave */
11454                         if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
11455                                 io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11456                                 if (io->flags & CTL_FLAG_IO_ACTIVE) {
11457                                         io->flags |= CTL_FLAG_FAILOVER;
11458                                 } else {
11459                                         ctl_set_busy(&((union ctl_io *)io)->
11460                                             scsiio);
11461                                         ctl_done((union ctl_io *)io);
11462                                 }
11463                         }
11464                 }
11465         } else { /* SERIALIZE modes */
11466                 TAILQ_FOREACH_SAFE(io, &lun->blocked_queue, blocked_links,
11467                     next_io) {
11468                         /* We are master */
11469                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11470                                 TAILQ_REMOVE(&lun->blocked_queue, io,
11471                                     blocked_links);
11472                                 io->flags &= ~CTL_FLAG_BLOCKED;
11473                                 TAILQ_REMOVE(&lun->ooa_queue, io, ooa_links);
11474                                 ctl_free_io((union ctl_io *)io);
11475                         }
11476                 }
11477                 TAILQ_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
11478                         /* We are master */
11479                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11480                                 TAILQ_REMOVE(&lun->ooa_queue, io, ooa_links);
11481                                 ctl_free_io((union ctl_io *)io);
11482                         }
11483                         /* We are slave */
11484                         if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
11485                                 io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11486                                 if (!(io->flags & CTL_FLAG_IO_ACTIVE)) {
11487                                         ctl_set_busy(&((union ctl_io *)io)->
11488                                             scsiio);
11489                                         ctl_done((union ctl_io *)io);
11490                                 }
11491                         }
11492                 }
11493                 ctl_check_blocked(lun);
11494         }
11495         mtx_unlock(&lun->lun_lock);
11496 }
11497
11498 static int
11499 ctl_scsiio_precheck(struct ctl_softc *softc, struct ctl_scsiio *ctsio)
11500 {
11501         struct ctl_lun *lun;
11502         const struct ctl_cmd_entry *entry;
11503         uint32_t initidx, targ_lun;
11504         int retval;
11505
11506         retval = 0;
11507
11508         lun = NULL;
11509
11510         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11511         if ((targ_lun < CTL_MAX_LUNS)
11512          && ((lun = softc->ctl_luns[targ_lun]) != NULL)) {
11513                 /*
11514                  * If the LUN is invalid, pretend that it doesn't exist.
11515                  * It will go away as soon as all pending I/O has been
11516                  * completed.
11517                  */
11518                 mtx_lock(&lun->lun_lock);
11519                 if (lun->flags & CTL_LUN_DISABLED) {
11520                         mtx_unlock(&lun->lun_lock);
11521                         lun = NULL;
11522                         ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11523                         ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11524                 } else {
11525                         ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
11526                         ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
11527                                 lun->be_lun;
11528
11529                         /*
11530                          * Every I/O goes into the OOA queue for a
11531                          * particular LUN, and stays there until completion.
11532                          */
11533 #ifdef CTL_TIME_IO
11534                         if (TAILQ_EMPTY(&lun->ooa_queue)) {
11535                                 lun->idle_time += getsbinuptime() -
11536                                     lun->last_busy;
11537                         }
11538 #endif
11539                         TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr,
11540                             ooa_links);
11541                 }
11542         } else {
11543                 ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11544                 ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11545         }
11546
11547         /* Get command entry and return error if it is unsuppotyed. */
11548         entry = ctl_validate_command(ctsio);
11549         if (entry == NULL) {
11550                 if (lun)
11551                         mtx_unlock(&lun->lun_lock);
11552                 return (retval);
11553         }
11554
11555         ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11556         ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11557
11558         /*
11559          * Check to see whether we can send this command to LUNs that don't
11560          * exist.  This should pretty much only be the case for inquiry
11561          * and request sense.  Further checks, below, really require having
11562          * a LUN, so we can't really check the command anymore.  Just put
11563          * it on the rtr queue.
11564          */
11565         if (lun == NULL) {
11566                 if (entry->flags & CTL_CMD_FLAG_OK_ON_NO_LUN) {
11567                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11568                         ctl_enqueue_rtr((union ctl_io *)ctsio);
11569                         return (retval);
11570                 }
11571
11572                 ctl_set_unsupported_lun(ctsio);
11573                 ctl_done((union ctl_io *)ctsio);
11574                 CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11575                 return (retval);
11576         } else {
11577                 /*
11578                  * Make sure we support this particular command on this LUN.
11579                  * e.g., we don't support writes to the control LUN.
11580                  */
11581                 if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11582                         mtx_unlock(&lun->lun_lock);
11583                         ctl_set_invalid_opcode(ctsio);
11584                         ctl_done((union ctl_io *)ctsio);
11585                         return (retval);
11586                 }
11587         }
11588
11589         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11590
11591 #ifdef CTL_WITH_CA
11592         /*
11593          * If we've got a request sense, it'll clear the contingent
11594          * allegiance condition.  Otherwise, if we have a CA condition for
11595          * this initiator, clear it, because it sent down a command other
11596          * than request sense.
11597          */
11598         if ((ctsio->cdb[0] != REQUEST_SENSE)
11599          && (ctl_is_set(lun->have_ca, initidx)))
11600                 ctl_clear_mask(lun->have_ca, initidx);
11601 #endif
11602
11603         /*
11604          * If the command has this flag set, it handles its own unit
11605          * attention reporting, we shouldn't do anything.  Otherwise we
11606          * check for any pending unit attentions, and send them back to the
11607          * initiator.  We only do this when a command initially comes in,
11608          * not when we pull it off the blocked queue.
11609          *
11610          * According to SAM-3, section 5.3.2, the order that things get
11611          * presented back to the host is basically unit attentions caused
11612          * by some sort of reset event, busy status, reservation conflicts
11613          * or task set full, and finally any other status.
11614          *
11615          * One issue here is that some of the unit attentions we report
11616          * don't fall into the "reset" category (e.g. "reported luns data
11617          * has changed").  So reporting it here, before the reservation
11618          * check, may be technically wrong.  I guess the only thing to do
11619          * would be to check for and report the reset events here, and then
11620          * check for the other unit attention types after we check for a
11621          * reservation conflict.
11622          *
11623          * XXX KDM need to fix this
11624          */
11625         if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11626                 ctl_ua_type ua_type;
11627
11628                 ua_type = ctl_build_ua(lun, initidx, &ctsio->sense_data,
11629                     SSD_TYPE_NONE);
11630                 if (ua_type != CTL_UA_NONE) {
11631                         mtx_unlock(&lun->lun_lock);
11632                         ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11633                         ctsio->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
11634                         ctsio->sense_len = SSD_FULL_SIZE;
11635                         ctl_done((union ctl_io *)ctsio);
11636                         return (retval);
11637                 }
11638         }
11639
11640
11641         if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
11642                 mtx_unlock(&lun->lun_lock);
11643                 ctl_done((union ctl_io *)ctsio);
11644                 return (retval);
11645         }
11646
11647         /*
11648          * XXX CHD this is where we want to send IO to other side if
11649          * this LUN is secondary on this SC. We will need to make a copy
11650          * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11651          * the copy we send as FROM_OTHER.
11652          * We also need to stuff the address of the original IO so we can
11653          * find it easily. Something similar will need be done on the other
11654          * side so when we are done we can find the copy.
11655          */
11656         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
11657             (lun->flags & CTL_LUN_PEER_SC_PRIMARY) != 0 &&
11658             (entry->flags & CTL_CMD_FLAG_RUN_HERE) == 0) {
11659                 union ctl_ha_msg msg_info;
11660                 int isc_retval;
11661
11662                 ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11663                 ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11664                 mtx_unlock(&lun->lun_lock);
11665
11666                 msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11667                 msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11668                 msg_info.hdr.serializing_sc = NULL;
11669                 msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11670                 msg_info.scsi.tag_num = ctsio->tag_num;
11671                 msg_info.scsi.tag_type = ctsio->tag_type;
11672                 msg_info.scsi.cdb_len = ctsio->cdb_len;
11673                 memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11674
11675                 if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11676                     sizeof(msg_info.scsi) - sizeof(msg_info.scsi.sense_data),
11677                     M_WAITOK)) > CTL_HA_STATUS_SUCCESS) {
11678                         ctl_set_busy(ctsio);
11679                         ctl_done((union ctl_io *)ctsio);
11680                         return (retval);
11681                 }
11682                 return (retval);
11683         }
11684
11685         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11686                               (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11687                               ctl_ooaq, ooa_links))) {
11688         case CTL_ACTION_BLOCK:
11689                 ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11690                 TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11691                                   blocked_links);
11692                 mtx_unlock(&lun->lun_lock);
11693                 return (retval);
11694         case CTL_ACTION_PASS:
11695         case CTL_ACTION_SKIP:
11696                 ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11697                 mtx_unlock(&lun->lun_lock);
11698                 ctl_enqueue_rtr((union ctl_io *)ctsio);
11699                 break;
11700         case CTL_ACTION_OVERLAP:
11701                 mtx_unlock(&lun->lun_lock);
11702                 ctl_set_overlapped_cmd(ctsio);
11703                 ctl_done((union ctl_io *)ctsio);
11704                 break;
11705         case CTL_ACTION_OVERLAP_TAG:
11706                 mtx_unlock(&lun->lun_lock);
11707                 ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11708                 ctl_done((union ctl_io *)ctsio);
11709                 break;
11710         case CTL_ACTION_ERROR:
11711         default:
11712                 mtx_unlock(&lun->lun_lock);
11713                 ctl_set_internal_failure(ctsio,
11714                                          /*sks_valid*/ 0,
11715                                          /*retry_count*/ 0);
11716                 ctl_done((union ctl_io *)ctsio);
11717                 break;
11718         }
11719         return (retval);
11720 }
11721
11722 const struct ctl_cmd_entry *
11723 ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa)
11724 {
11725         const struct ctl_cmd_entry *entry;
11726         int service_action;
11727
11728         entry = &ctl_cmd_table[ctsio->cdb[0]];
11729         if (sa)
11730                 *sa = ((entry->flags & CTL_CMD_FLAG_SA5) != 0);
11731         if (entry->flags & CTL_CMD_FLAG_SA5) {
11732                 service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11733                 entry = &((const struct ctl_cmd_entry *)
11734                     entry->execute)[service_action];
11735         }
11736         return (entry);
11737 }
11738
11739 const struct ctl_cmd_entry *
11740 ctl_validate_command(struct ctl_scsiio *ctsio)
11741 {
11742         const struct ctl_cmd_entry *entry;
11743         int i, sa;
11744         uint8_t diff;
11745
11746         entry = ctl_get_cmd_entry(ctsio, &sa);
11747         if (entry->execute == NULL) {
11748                 if (sa)
11749                         ctl_set_invalid_field(ctsio,
11750                                               /*sks_valid*/ 1,
11751                                               /*command*/ 1,
11752                                               /*field*/ 1,
11753                                               /*bit_valid*/ 1,
11754                                               /*bit*/ 4);
11755                 else
11756                         ctl_set_invalid_opcode(ctsio);
11757                 ctl_done((union ctl_io *)ctsio);
11758                 return (NULL);
11759         }
11760         KASSERT(entry->length > 0,
11761             ("Not defined length for command 0x%02x/0x%02x",
11762              ctsio->cdb[0], ctsio->cdb[1]));
11763         for (i = 1; i < entry->length; i++) {
11764                 diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11765                 if (diff == 0)
11766                         continue;
11767                 ctl_set_invalid_field(ctsio,
11768                                       /*sks_valid*/ 1,
11769                                       /*command*/ 1,
11770                                       /*field*/ i,
11771                                       /*bit_valid*/ 1,
11772                                       /*bit*/ fls(diff) - 1);
11773                 ctl_done((union ctl_io *)ctsio);
11774                 return (NULL);
11775         }
11776         return (entry);
11777 }
11778
11779 static int
11780 ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11781 {
11782
11783         switch (lun_type) {
11784         case T_DIRECT:
11785                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_DIRECT) == 0)
11786                         return (0);
11787                 break;
11788         case T_PROCESSOR:
11789                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0)
11790                         return (0);
11791                 break;
11792         case T_CDROM:
11793                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_CDROM) == 0)
11794                         return (0);
11795                 break;
11796         default:
11797                 return (0);
11798         }
11799         return (1);
11800 }
11801
11802 static int
11803 ctl_scsiio(struct ctl_scsiio *ctsio)
11804 {
11805         int retval;
11806         const struct ctl_cmd_entry *entry;
11807
11808         retval = CTL_RETVAL_COMPLETE;
11809
11810         CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11811
11812         entry = ctl_get_cmd_entry(ctsio, NULL);
11813
11814         /*
11815          * If this I/O has been aborted, just send it straight to
11816          * ctl_done() without executing it.
11817          */
11818         if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11819                 ctl_done((union ctl_io *)ctsio);
11820                 goto bailout;
11821         }
11822
11823         /*
11824          * All the checks should have been handled by ctl_scsiio_precheck().
11825          * We should be clear now to just execute the I/O.
11826          */
11827         retval = entry->execute(ctsio);
11828
11829 bailout:
11830         return (retval);
11831 }
11832
11833 /*
11834  * Since we only implement one target right now, a bus reset simply resets
11835  * our single target.
11836  */
11837 static int
11838 ctl_bus_reset(struct ctl_softc *softc, union ctl_io *io)
11839 {
11840         return(ctl_target_reset(softc, io, CTL_UA_BUS_RESET));
11841 }
11842
11843 static int
11844 ctl_target_reset(struct ctl_softc *softc, union ctl_io *io,
11845                  ctl_ua_type ua_type)
11846 {
11847         struct ctl_port *port;
11848         struct ctl_lun *lun;
11849         int retval;
11850
11851         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11852                 union ctl_ha_msg msg_info;
11853
11854                 msg_info.hdr.nexus = io->io_hdr.nexus;
11855                 if (ua_type==CTL_UA_TARG_RESET)
11856                         msg_info.task.task_action = CTL_TASK_TARGET_RESET;
11857                 else
11858                         msg_info.task.task_action = CTL_TASK_BUS_RESET;
11859                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11860                 msg_info.hdr.original_sc = NULL;
11861                 msg_info.hdr.serializing_sc = NULL;
11862                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11863                     sizeof(msg_info.task), M_WAITOK);
11864         }
11865         retval = 0;
11866
11867         mtx_lock(&softc->ctl_lock);
11868         port = ctl_io_port(&io->io_hdr);
11869         STAILQ_FOREACH(lun, &softc->lun_list, links) {
11870                 if (port != NULL &&
11871                     ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
11872                         continue;
11873                 retval += ctl_do_lun_reset(lun, io, ua_type);
11874         }
11875         mtx_unlock(&softc->ctl_lock);
11876         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11877         return (retval);
11878 }
11879
11880 /*
11881  * The LUN should always be set.  The I/O is optional, and is used to
11882  * distinguish between I/Os sent by this initiator, and by other
11883  * initiators.  We set unit attention for initiators other than this one.
11884  * SAM-3 is vague on this point.  It does say that a unit attention should
11885  * be established for other initiators when a LUN is reset (see section
11886  * 5.7.3), but it doesn't specifically say that the unit attention should
11887  * be established for this particular initiator when a LUN is reset.  Here
11888  * is the relevant text, from SAM-3 rev 8:
11889  *
11890  * 5.7.2 When a SCSI initiator port aborts its own tasks
11891  *
11892  * When a SCSI initiator port causes its own task(s) to be aborted, no
11893  * notification that the task(s) have been aborted shall be returned to
11894  * the SCSI initiator port other than the completion response for the
11895  * command or task management function action that caused the task(s) to
11896  * be aborted and notification(s) associated with related effects of the
11897  * action (e.g., a reset unit attention condition).
11898  *
11899  * XXX KDM for now, we're setting unit attention for all initiators.
11900  */
11901 static int
11902 ctl_do_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
11903 {
11904         union ctl_io *xio;
11905 #if 0
11906         uint32_t initidx;
11907 #endif
11908         int i;
11909
11910         mtx_lock(&lun->lun_lock);
11911         /*
11912          * Run through the OOA queue and abort each I/O.
11913          */
11914         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11915              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11916                 xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
11917         }
11918
11919         /*
11920          * This version sets unit attention for every
11921          */
11922 #if 0
11923         initidx = ctl_get_initindex(&io->io_hdr.nexus);
11924         ctl_est_ua_all(lun, initidx, ua_type);
11925 #else
11926         ctl_est_ua_all(lun, -1, ua_type);
11927 #endif
11928
11929         /*
11930          * A reset (any kind, really) clears reservations established with
11931          * RESERVE/RELEASE.  It does not clear reservations established
11932          * with PERSISTENT RESERVE OUT, but we don't support that at the
11933          * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
11934          * reservations made with the RESERVE/RELEASE commands, because
11935          * those commands are obsolete in SPC-3.
11936          */
11937         lun->flags &= ~CTL_LUN_RESERVED;
11938
11939 #ifdef CTL_WITH_CA
11940         for (i = 0; i < CTL_MAX_INITIATORS; i++)
11941                 ctl_clear_mask(lun->have_ca, i);
11942 #endif
11943         lun->prevent_count = 0;
11944         for (i = 0; i < CTL_MAX_INITIATORS; i++)
11945                 ctl_clear_mask(lun->prevent, i);
11946         mtx_unlock(&lun->lun_lock);
11947
11948         return (0);
11949 }
11950
11951 static int
11952 ctl_lun_reset(struct ctl_softc *softc, union ctl_io *io)
11953 {
11954         struct ctl_lun *lun;
11955         uint32_t targ_lun;
11956         int retval;
11957
11958         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11959         mtx_lock(&softc->ctl_lock);
11960         if ((targ_lun >= CTL_MAX_LUNS) ||
11961             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11962                 mtx_unlock(&softc->ctl_lock);
11963                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11964                 return (1);
11965         }
11966         retval = ctl_do_lun_reset(lun, io, CTL_UA_LUN_RESET);
11967         mtx_unlock(&softc->ctl_lock);
11968         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11969
11970         if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0) {
11971                 union ctl_ha_msg msg_info;
11972
11973                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11974                 msg_info.hdr.nexus = io->io_hdr.nexus;
11975                 msg_info.task.task_action = CTL_TASK_LUN_RESET;
11976                 msg_info.hdr.original_sc = NULL;
11977                 msg_info.hdr.serializing_sc = NULL;
11978                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11979                     sizeof(msg_info.task), M_WAITOK);
11980         }
11981         return (retval);
11982 }
11983
11984 static void
11985 ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
11986     int other_sc)
11987 {
11988         union ctl_io *xio;
11989
11990         mtx_assert(&lun->lun_lock, MA_OWNED);
11991
11992         /*
11993          * Run through the OOA queue and attempt to find the given I/O.
11994          * The target port, initiator ID, tag type and tag number have to
11995          * match the values that we got from the initiator.  If we have an
11996          * untagged command to abort, simply abort the first untagged command
11997          * we come to.  We only allow one untagged command at a time of course.
11998          */
11999         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12000              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12001
12002                 if ((targ_port == UINT32_MAX ||
12003                      targ_port == xio->io_hdr.nexus.targ_port) &&
12004                     (init_id == UINT32_MAX ||
12005                      init_id == xio->io_hdr.nexus.initid)) {
12006                         if (targ_port != xio->io_hdr.nexus.targ_port ||
12007                             init_id != xio->io_hdr.nexus.initid)
12008                                 xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS;
12009                         xio->io_hdr.flags |= CTL_FLAG_ABORT;
12010                         if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12011                                 union ctl_ha_msg msg_info;
12012
12013                                 msg_info.hdr.nexus = xio->io_hdr.nexus;
12014                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
12015                                 msg_info.task.tag_num = xio->scsiio.tag_num;
12016                                 msg_info.task.tag_type = xio->scsiio.tag_type;
12017                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12018                                 msg_info.hdr.original_sc = NULL;
12019                                 msg_info.hdr.serializing_sc = NULL;
12020                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
12021                                     sizeof(msg_info.task), M_NOWAIT);
12022                         }
12023                 }
12024         }
12025 }
12026
12027 static int
12028 ctl_abort_task_set(union ctl_io *io)
12029 {
12030         struct ctl_softc *softc = control_softc;
12031         struct ctl_lun *lun;
12032         uint32_t targ_lun;
12033
12034         /*
12035          * Look up the LUN.
12036          */
12037         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12038         mtx_lock(&softc->ctl_lock);
12039         if ((targ_lun >= CTL_MAX_LUNS) ||
12040             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12041                 mtx_unlock(&softc->ctl_lock);
12042                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12043                 return (1);
12044         }
12045
12046         mtx_lock(&lun->lun_lock);
12047         mtx_unlock(&softc->ctl_lock);
12048         if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
12049                 ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12050                     io->io_hdr.nexus.initid,
12051                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12052         } else { /* CTL_TASK_CLEAR_TASK_SET */
12053                 ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
12054                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12055         }
12056         mtx_unlock(&lun->lun_lock);
12057         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12058         return (0);
12059 }
12060
12061 static int
12062 ctl_i_t_nexus_reset(union ctl_io *io)
12063 {
12064         struct ctl_softc *softc = control_softc;
12065         struct ctl_lun *lun;
12066         uint32_t initidx;
12067
12068         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12069                 union ctl_ha_msg msg_info;
12070
12071                 msg_info.hdr.nexus = io->io_hdr.nexus;
12072                 msg_info.task.task_action = CTL_TASK_I_T_NEXUS_RESET;
12073                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12074                 msg_info.hdr.original_sc = NULL;
12075                 msg_info.hdr.serializing_sc = NULL;
12076                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
12077                     sizeof(msg_info.task), M_WAITOK);
12078         }
12079
12080         initidx = ctl_get_initindex(&io->io_hdr.nexus);
12081         mtx_lock(&softc->ctl_lock);
12082         STAILQ_FOREACH(lun, &softc->lun_list, links) {
12083                 mtx_lock(&lun->lun_lock);
12084                 ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12085                     io->io_hdr.nexus.initid, 1);
12086 #ifdef CTL_WITH_CA
12087                 ctl_clear_mask(lun->have_ca, initidx);
12088 #endif
12089                 if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == initidx))
12090                         lun->flags &= ~CTL_LUN_RESERVED;
12091                 if (ctl_is_set(lun->prevent, initidx)) {
12092                         ctl_clear_mask(lun->prevent, initidx);
12093                         lun->prevent_count--;
12094                 }
12095                 ctl_est_ua(lun, initidx, CTL_UA_I_T_NEXUS_LOSS);
12096                 mtx_unlock(&lun->lun_lock);
12097         }
12098         mtx_unlock(&softc->ctl_lock);
12099         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12100         return (0);
12101 }
12102
12103 static int
12104 ctl_abort_task(union ctl_io *io)
12105 {
12106         union ctl_io *xio;
12107         struct ctl_lun *lun;
12108         struct ctl_softc *softc;
12109 #if 0
12110         struct sbuf sb;
12111         char printbuf[128];
12112 #endif
12113         int found;
12114         uint32_t targ_lun;
12115
12116         softc = control_softc;
12117         found = 0;
12118
12119         /*
12120          * Look up the LUN.
12121          */
12122         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12123         mtx_lock(&softc->ctl_lock);
12124         if ((targ_lun >= CTL_MAX_LUNS) ||
12125             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12126                 mtx_unlock(&softc->ctl_lock);
12127                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12128                 return (1);
12129         }
12130
12131 #if 0
12132         printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
12133                lun->lun, io->taskio.tag_num, io->taskio.tag_type);
12134 #endif
12135
12136         mtx_lock(&lun->lun_lock);
12137         mtx_unlock(&softc->ctl_lock);
12138         /*
12139          * Run through the OOA queue and attempt to find the given I/O.
12140          * The target port, initiator ID, tag type and tag number have to
12141          * match the values that we got from the initiator.  If we have an
12142          * untagged command to abort, simply abort the first untagged command
12143          * we come to.  We only allow one untagged command at a time of course.
12144          */
12145         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12146              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12147 #if 0
12148                 sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
12149
12150                 sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
12151                             lun->lun, xio->scsiio.tag_num,
12152                             xio->scsiio.tag_type,
12153                             (xio->io_hdr.blocked_links.tqe_prev
12154                             == NULL) ? "" : " BLOCKED",
12155                             (xio->io_hdr.flags &
12156                             CTL_FLAG_DMA_INPROG) ? " DMA" : "",
12157                             (xio->io_hdr.flags &
12158                             CTL_FLAG_ABORT) ? " ABORT" : "",
12159                             (xio->io_hdr.flags &
12160                             CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
12161                 ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
12162                 sbuf_finish(&sb);
12163                 printf("%s\n", sbuf_data(&sb));
12164 #endif
12165
12166                 if ((xio->io_hdr.nexus.targ_port != io->io_hdr.nexus.targ_port)
12167                  || (xio->io_hdr.nexus.initid != io->io_hdr.nexus.initid)
12168                  || (xio->io_hdr.flags & CTL_FLAG_ABORT))
12169                         continue;
12170
12171                 /*
12172                  * If the abort says that the task is untagged, the
12173                  * task in the queue must be untagged.  Otherwise,
12174                  * we just check to see whether the tag numbers
12175                  * match.  This is because the QLogic firmware
12176                  * doesn't pass back the tag type in an abort
12177                  * request.
12178                  */
12179 #if 0
12180                 if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
12181                   && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
12182                  || (xio->scsiio.tag_num == io->taskio.tag_num))
12183 #endif
12184                 /*
12185                  * XXX KDM we've got problems with FC, because it
12186                  * doesn't send down a tag type with aborts.  So we
12187                  * can only really go by the tag number...
12188                  * This may cause problems with parallel SCSI.
12189                  * Need to figure that out!!
12190                  */
12191                 if (xio->scsiio.tag_num == io->taskio.tag_num) {
12192                         xio->io_hdr.flags |= CTL_FLAG_ABORT;
12193                         found = 1;
12194                         if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0 &&
12195                             !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12196                                 union ctl_ha_msg msg_info;
12197
12198                                 msg_info.hdr.nexus = io->io_hdr.nexus;
12199                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
12200                                 msg_info.task.tag_num = io->taskio.tag_num;
12201                                 msg_info.task.tag_type = io->taskio.tag_type;
12202                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12203                                 msg_info.hdr.original_sc = NULL;
12204                                 msg_info.hdr.serializing_sc = NULL;
12205 #if 0
12206                                 printf("Sent Abort to other side\n");
12207 #endif
12208                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
12209                                     sizeof(msg_info.task), M_NOWAIT);
12210                         }
12211 #if 0
12212                         printf("ctl_abort_task: found I/O to abort\n");
12213 #endif
12214                 }
12215         }
12216         mtx_unlock(&lun->lun_lock);
12217
12218         if (found == 0) {
12219                 /*
12220                  * This isn't really an error.  It's entirely possible for
12221                  * the abort and command completion to cross on the wire.
12222                  * This is more of an informative/diagnostic error.
12223                  */
12224 #if 0
12225                 printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
12226                        "%u:%u:%u tag %d type %d\n",
12227                        io->io_hdr.nexus.initid,
12228                        io->io_hdr.nexus.targ_port,
12229                        io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
12230                        io->taskio.tag_type);
12231 #endif
12232         }
12233         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12234         return (0);
12235 }
12236
12237 static int
12238 ctl_query_task(union ctl_io *io, int task_set)
12239 {
12240         union ctl_io *xio;
12241         struct ctl_lun *lun;
12242         struct ctl_softc *softc;
12243         int found = 0;
12244         uint32_t targ_lun;
12245
12246         softc = control_softc;
12247         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12248         mtx_lock(&softc->ctl_lock);
12249         if ((targ_lun >= CTL_MAX_LUNS) ||
12250             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12251                 mtx_unlock(&softc->ctl_lock);
12252                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12253                 return (1);
12254         }
12255         mtx_lock(&lun->lun_lock);
12256         mtx_unlock(&softc->ctl_lock);
12257         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12258              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12259
12260                 if ((xio->io_hdr.nexus.targ_port != io->io_hdr.nexus.targ_port)
12261                  || (xio->io_hdr.nexus.initid != io->io_hdr.nexus.initid)
12262                  || (xio->io_hdr.flags & CTL_FLAG_ABORT))
12263                         continue;
12264
12265                 if (task_set || xio->scsiio.tag_num == io->taskio.tag_num) {
12266                         found = 1;
12267                         break;
12268                 }
12269         }
12270         mtx_unlock(&lun->lun_lock);
12271         if (found)
12272                 io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED;
12273         else
12274                 io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12275         return (0);
12276 }
12277
12278 static int
12279 ctl_query_async_event(union ctl_io *io)
12280 {
12281         struct ctl_lun *lun;
12282         struct ctl_softc *softc;
12283         ctl_ua_type ua;
12284         uint32_t targ_lun, initidx;
12285
12286         softc = control_softc;
12287         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12288         mtx_lock(&softc->ctl_lock);
12289         if ((targ_lun >= CTL_MAX_LUNS) ||
12290             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12291                 mtx_unlock(&softc->ctl_lock);
12292                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12293                 return (1);
12294         }
12295         mtx_lock(&lun->lun_lock);
12296         mtx_unlock(&softc->ctl_lock);
12297         initidx = ctl_get_initindex(&io->io_hdr.nexus);
12298         ua = ctl_build_qae(lun, initidx, io->taskio.task_resp);
12299         mtx_unlock(&lun->lun_lock);
12300         if (ua != CTL_UA_NONE)
12301                 io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED;
12302         else
12303                 io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12304         return (0);
12305 }
12306
12307 static void
12308 ctl_run_task(union ctl_io *io)
12309 {
12310         struct ctl_softc *softc = control_softc;
12311         int retval = 1;
12312
12313         CTL_DEBUG_PRINT(("ctl_run_task\n"));
12314         KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
12315             ("ctl_run_task: Unextected io_type %d\n", io->io_hdr.io_type));
12316         io->taskio.task_status = CTL_TASK_FUNCTION_NOT_SUPPORTED;
12317         bzero(io->taskio.task_resp, sizeof(io->taskio.task_resp));
12318         switch (io->taskio.task_action) {
12319         case CTL_TASK_ABORT_TASK:
12320                 retval = ctl_abort_task(io);
12321                 break;
12322         case CTL_TASK_ABORT_TASK_SET:
12323         case CTL_TASK_CLEAR_TASK_SET:
12324                 retval = ctl_abort_task_set(io);
12325                 break;
12326         case CTL_TASK_CLEAR_ACA:
12327                 break;
12328         case CTL_TASK_I_T_NEXUS_RESET:
12329                 retval = ctl_i_t_nexus_reset(io);
12330                 break;
12331         case CTL_TASK_LUN_RESET:
12332                 retval = ctl_lun_reset(softc, io);
12333                 break;
12334         case CTL_TASK_TARGET_RESET:
12335                 retval = ctl_target_reset(softc, io, CTL_UA_TARG_RESET);
12336                 break;
12337         case CTL_TASK_BUS_RESET:
12338                 retval = ctl_bus_reset(softc, io);
12339                 break;
12340         case CTL_TASK_PORT_LOGIN:
12341                 break;
12342         case CTL_TASK_PORT_LOGOUT:
12343                 break;
12344         case CTL_TASK_QUERY_TASK:
12345                 retval = ctl_query_task(io, 0);
12346                 break;
12347         case CTL_TASK_QUERY_TASK_SET:
12348                 retval = ctl_query_task(io, 1);
12349                 break;
12350         case CTL_TASK_QUERY_ASYNC_EVENT:
12351                 retval = ctl_query_async_event(io);
12352                 break;
12353         default:
12354                 printf("%s: got unknown task management event %d\n",
12355                        __func__, io->taskio.task_action);
12356                 break;
12357         }
12358         if (retval == 0)
12359                 io->io_hdr.status = CTL_SUCCESS;
12360         else
12361                 io->io_hdr.status = CTL_ERROR;
12362         ctl_done(io);
12363 }
12364
12365 /*
12366  * For HA operation.  Handle commands that come in from the other
12367  * controller.
12368  */
12369 static void
12370 ctl_handle_isc(union ctl_io *io)
12371 {
12372         int free_io;
12373         struct ctl_lun *lun;
12374         struct ctl_softc *softc = control_softc;
12375         uint32_t targ_lun;
12376
12377         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12378         lun = softc->ctl_luns[targ_lun];
12379
12380         switch (io->io_hdr.msg_type) {
12381         case CTL_MSG_SERIALIZE:
12382                 free_io = ctl_serialize_other_sc_cmd(&io->scsiio);
12383                 break;
12384         case CTL_MSG_R2R: {
12385                 const struct ctl_cmd_entry *entry;
12386
12387                 /*
12388                  * This is only used in SER_ONLY mode.
12389                  */
12390                 free_io = 0;
12391                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
12392                 mtx_lock(&lun->lun_lock);
12393                 if (ctl_scsiio_lun_check(lun,
12394                     entry, (struct ctl_scsiio *)io) != 0) {
12395                         mtx_unlock(&lun->lun_lock);
12396                         ctl_done(io);
12397                         break;
12398                 }
12399                 io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
12400                 mtx_unlock(&lun->lun_lock);
12401                 ctl_enqueue_rtr(io);
12402                 break;
12403         }
12404         case CTL_MSG_FINISH_IO:
12405                 if (softc->ha_mode == CTL_HA_MODE_XFER) {
12406                         free_io = 0;
12407                         ctl_done(io);
12408                 } else {
12409                         free_io = 1;
12410                         mtx_lock(&lun->lun_lock);
12411                         TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
12412                                      ooa_links);
12413                         ctl_check_blocked(lun);
12414                         mtx_unlock(&lun->lun_lock);
12415                 }
12416                 break;
12417         case CTL_MSG_PERS_ACTION:
12418                 ctl_hndl_per_res_out_on_other_sc(
12419                         (union ctl_ha_msg *)&io->presio.pr_msg);
12420                 free_io = 1;
12421                 break;
12422         case CTL_MSG_BAD_JUJU:
12423                 free_io = 0;
12424                 ctl_done(io);
12425                 break;
12426         case CTL_MSG_DATAMOVE:
12427                 /* Only used in XFER mode */
12428                 free_io = 0;
12429                 ctl_datamove_remote(io);
12430                 break;
12431         case CTL_MSG_DATAMOVE_DONE:
12432                 /* Only used in XFER mode */
12433                 free_io = 0;
12434                 io->scsiio.be_move_done(io);
12435                 break;
12436         case CTL_MSG_FAILOVER:
12437                 ctl_failover_lun(io);
12438                 free_io = 1;
12439                 break;
12440         default:
12441                 free_io = 1;
12442                 printf("%s: Invalid message type %d\n",
12443                        __func__, io->io_hdr.msg_type);
12444                 break;
12445         }
12446         if (free_io)
12447                 ctl_free_io(io);
12448
12449 }
12450
12451
12452 /*
12453  * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12454  * there is no match.
12455  */
12456 static ctl_lun_error_pattern
12457 ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12458 {
12459         const struct ctl_cmd_entry *entry;
12460         ctl_lun_error_pattern filtered_pattern, pattern;
12461
12462         pattern = desc->error_pattern;
12463
12464         /*
12465          * XXX KDM we need more data passed into this function to match a
12466          * custom pattern, and we actually need to implement custom pattern
12467          * matching.
12468          */
12469         if (pattern & CTL_LUN_PAT_CMD)
12470                 return (CTL_LUN_PAT_CMD);
12471
12472         if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12473                 return (CTL_LUN_PAT_ANY);
12474
12475         entry = ctl_get_cmd_entry(ctsio, NULL);
12476
12477         filtered_pattern = entry->pattern & pattern;
12478
12479         /*
12480          * If the user requested specific flags in the pattern (e.g.
12481          * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12482          * flags.
12483          *
12484          * If the user did not specify any flags, it doesn't matter whether
12485          * or not the command supports the flags.
12486          */
12487         if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12488              (pattern & ~CTL_LUN_PAT_MASK))
12489                 return (CTL_LUN_PAT_NONE);
12490
12491         /*
12492          * If the user asked for a range check, see if the requested LBA
12493          * range overlaps with this command's LBA range.
12494          */
12495         if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12496                 uint64_t lba1;
12497                 uint64_t len1;
12498                 ctl_action action;
12499                 int retval;
12500
12501                 retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12502                 if (retval != 0)
12503                         return (CTL_LUN_PAT_NONE);
12504
12505                 action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12506                                               desc->lba_range.len, FALSE);
12507                 /*
12508                  * A "pass" means that the LBA ranges don't overlap, so
12509                  * this doesn't match the user's range criteria.
12510                  */
12511                 if (action == CTL_ACTION_PASS)
12512                         return (CTL_LUN_PAT_NONE);
12513         }
12514
12515         return (filtered_pattern);
12516 }
12517
12518 static void
12519 ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12520 {
12521         struct ctl_error_desc *desc, *desc2;
12522
12523         mtx_assert(&lun->lun_lock, MA_OWNED);
12524
12525         STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12526                 ctl_lun_error_pattern pattern;
12527                 /*
12528                  * Check to see whether this particular command matches
12529                  * the pattern in the descriptor.
12530                  */
12531                 pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12532                 if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12533                         continue;
12534
12535                 switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12536                 case CTL_LUN_INJ_ABORTED:
12537                         ctl_set_aborted(&io->scsiio);
12538                         break;
12539                 case CTL_LUN_INJ_MEDIUM_ERR:
12540                         ctl_set_medium_error(&io->scsiio,
12541                             (io->io_hdr.flags & CTL_FLAG_DATA_MASK) !=
12542                              CTL_FLAG_DATA_OUT);
12543                         break;
12544                 case CTL_LUN_INJ_UA:
12545                         /* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12546                          * OCCURRED */
12547                         ctl_set_ua(&io->scsiio, 0x29, 0x00);
12548                         break;
12549                 case CTL_LUN_INJ_CUSTOM:
12550                         /*
12551                          * We're assuming the user knows what he is doing.
12552                          * Just copy the sense information without doing
12553                          * checks.
12554                          */
12555                         bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12556                               MIN(sizeof(desc->custom_sense),
12557                                   sizeof(io->scsiio.sense_data)));
12558                         io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12559                         io->scsiio.sense_len = SSD_FULL_SIZE;
12560                         io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12561                         break;
12562                 case CTL_LUN_INJ_NONE:
12563                 default:
12564                         /*
12565                          * If this is an error injection type we don't know
12566                          * about, clear the continuous flag (if it is set)
12567                          * so it will get deleted below.
12568                          */
12569                         desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12570                         break;
12571                 }
12572                 /*
12573                  * By default, each error injection action is a one-shot
12574                  */
12575                 if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12576                         continue;
12577
12578                 STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12579
12580                 free(desc, M_CTL);
12581         }
12582 }
12583
12584 #ifdef CTL_IO_DELAY
12585 static void
12586 ctl_datamove_timer_wakeup(void *arg)
12587 {
12588         union ctl_io *io;
12589
12590         io = (union ctl_io *)arg;
12591
12592         ctl_datamove(io);
12593 }
12594 #endif /* CTL_IO_DELAY */
12595
12596 void
12597 ctl_datamove(union ctl_io *io)
12598 {
12599         struct ctl_lun *lun;
12600         void (*fe_datamove)(union ctl_io *io);
12601
12602         mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
12603
12604         CTL_DEBUG_PRINT(("ctl_datamove\n"));
12605
12606         lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12607 #ifdef CTL_TIME_IO
12608         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12609                 char str[256];
12610                 char path_str[64];
12611                 struct sbuf sb;
12612
12613                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
12614                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12615
12616                 sbuf_cat(&sb, path_str);
12617                 switch (io->io_hdr.io_type) {
12618                 case CTL_IO_SCSI:
12619                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12620                         sbuf_printf(&sb, "\n");
12621                         sbuf_cat(&sb, path_str);
12622                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12623                                     io->scsiio.tag_num, io->scsiio.tag_type);
12624                         break;
12625                 case CTL_IO_TASK:
12626                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12627                                     "Tag Type: %d\n", io->taskio.task_action,
12628                                     io->taskio.tag_num, io->taskio.tag_type);
12629                         break;
12630                 default:
12631                         panic("%s: Invalid CTL I/O type %d\n",
12632                             __func__, io->io_hdr.io_type);
12633                 }
12634                 sbuf_cat(&sb, path_str);
12635                 sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12636                             (intmax_t)time_uptime - io->io_hdr.start_time);
12637                 sbuf_finish(&sb);
12638                 printf("%s", sbuf_data(&sb));
12639         }
12640 #endif /* CTL_TIME_IO */
12641
12642 #ifdef CTL_IO_DELAY
12643         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12644                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12645         } else {
12646                 if ((lun != NULL)
12647                  && (lun->delay_info.datamove_delay > 0)) {
12648
12649                         callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
12650                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12651                         callout_reset(&io->io_hdr.delay_callout,
12652                                       lun->delay_info.datamove_delay * hz,
12653                                       ctl_datamove_timer_wakeup, io);
12654                         if (lun->delay_info.datamove_type ==
12655                             CTL_DELAY_TYPE_ONESHOT)
12656                                 lun->delay_info.datamove_delay = 0;
12657                         return;
12658                 }
12659         }
12660 #endif
12661
12662         /*
12663          * This command has been aborted.  Set the port status, so we fail
12664          * the data move.
12665          */
12666         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12667                 printf("ctl_datamove: tag 0x%04x on (%u:%u:%u) aborted\n",
12668                        io->scsiio.tag_num, io->io_hdr.nexus.initid,
12669                        io->io_hdr.nexus.targ_port,
12670                        io->io_hdr.nexus.targ_lun);
12671                 io->io_hdr.port_status = 31337;
12672                 /*
12673                  * Note that the backend, in this case, will get the
12674                  * callback in its context.  In other cases it may get
12675                  * called in the frontend's interrupt thread context.
12676                  */
12677                 io->scsiio.be_move_done(io);
12678                 return;
12679         }
12680
12681         /* Don't confuse frontend with zero length data move. */
12682         if (io->scsiio.kern_data_len == 0) {
12683                 io->scsiio.be_move_done(io);
12684                 return;
12685         }
12686
12687         fe_datamove = ctl_io_port(&io->io_hdr)->fe_datamove;
12688         fe_datamove(io);
12689 }
12690
12691 static void
12692 ctl_send_datamove_done(union ctl_io *io, int have_lock)
12693 {
12694         union ctl_ha_msg msg;
12695 #ifdef CTL_TIME_IO
12696         struct bintime cur_bt;
12697 #endif
12698
12699         memset(&msg, 0, sizeof(msg));
12700         msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12701         msg.hdr.original_sc = io;
12702         msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12703         msg.hdr.nexus = io->io_hdr.nexus;
12704         msg.hdr.status = io->io_hdr.status;
12705         msg.scsi.tag_num = io->scsiio.tag_num;
12706         msg.scsi.tag_type = io->scsiio.tag_type;
12707         msg.scsi.scsi_status = io->scsiio.scsi_status;
12708         memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12709                io->scsiio.sense_len);
12710         msg.scsi.sense_len = io->scsiio.sense_len;
12711         msg.scsi.sense_residual = io->scsiio.sense_residual;
12712         msg.scsi.fetd_status = io->io_hdr.port_status;
12713         msg.scsi.residual = io->scsiio.residual;
12714         io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12715         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12716                 ctl_failover_io(io, /*have_lock*/ have_lock);
12717                 return;
12718         }
12719         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12720             sizeof(msg.scsi) - sizeof(msg.scsi.sense_data) +
12721             msg.scsi.sense_len, M_WAITOK);
12722
12723 #ifdef CTL_TIME_IO
12724         getbinuptime(&cur_bt);
12725         bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt);
12726         bintime_add(&io->io_hdr.dma_bt, &cur_bt);
12727 #endif
12728         io->io_hdr.num_dmas++;
12729 }
12730
12731 /*
12732  * The DMA to the remote side is done, now we need to tell the other side
12733  * we're done so it can continue with its data movement.
12734  */
12735 static void
12736 ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12737 {
12738         union ctl_io *io;
12739         uint32_t i;
12740
12741         io = rq->context;
12742
12743         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12744                 printf("%s: ISC DMA write failed with error %d", __func__,
12745                        rq->ret);
12746                 ctl_set_internal_failure(&io->scsiio,
12747                                          /*sks_valid*/ 1,
12748                                          /*retry_count*/ rq->ret);
12749         }
12750
12751         ctl_dt_req_free(rq);
12752
12753         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12754                 free(io->io_hdr.local_sglist[i].addr, M_CTL);
12755         free(io->io_hdr.remote_sglist, M_CTL);
12756         io->io_hdr.remote_sglist = NULL;
12757         io->io_hdr.local_sglist = NULL;
12758
12759         /*
12760          * The data is in local and remote memory, so now we need to send
12761          * status (good or back) back to the other side.
12762          */
12763         ctl_send_datamove_done(io, /*have_lock*/ 0);
12764 }
12765
12766 /*
12767  * We've moved the data from the host/controller into local memory.  Now we
12768  * need to push it over to the remote controller's memory.
12769  */
12770 static int
12771 ctl_datamove_remote_dm_write_cb(union ctl_io *io)
12772 {
12773         int retval;
12774
12775         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12776                                           ctl_datamove_remote_write_cb);
12777         return (retval);
12778 }
12779
12780 static void
12781 ctl_datamove_remote_write(union ctl_io *io)
12782 {
12783         int retval;
12784         void (*fe_datamove)(union ctl_io *io);
12785
12786         /*
12787          * - Get the data from the host/HBA into local memory.
12788          * - DMA memory from the local controller to the remote controller.
12789          * - Send status back to the remote controller.
12790          */
12791
12792         retval = ctl_datamove_remote_sgl_setup(io);
12793         if (retval != 0)
12794                 return;
12795
12796         /* Switch the pointer over so the FETD knows what to do */
12797         io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12798
12799         /*
12800          * Use a custom move done callback, since we need to send completion
12801          * back to the other controller, not to the backend on this side.
12802          */
12803         io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12804
12805         fe_datamove = ctl_io_port(&io->io_hdr)->fe_datamove;
12806         fe_datamove(io);
12807 }
12808
12809 static int
12810 ctl_datamove_remote_dm_read_cb(union ctl_io *io)
12811 {
12812 #if 0
12813         char str[256];
12814         char path_str[64];
12815         struct sbuf sb;
12816 #endif
12817         uint32_t i;
12818
12819         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12820                 free(io->io_hdr.local_sglist[i].addr, M_CTL);
12821         free(io->io_hdr.remote_sglist, M_CTL);
12822         io->io_hdr.remote_sglist = NULL;
12823         io->io_hdr.local_sglist = NULL;
12824
12825 #if 0
12826         scsi_path_string(io, path_str, sizeof(path_str));
12827         sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12828         sbuf_cat(&sb, path_str);
12829         scsi_command_string(&io->scsiio, NULL, &sb);
12830         sbuf_printf(&sb, "\n");
12831         sbuf_cat(&sb, path_str);
12832         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12833                     io->scsiio.tag_num, io->scsiio.tag_type);
12834         sbuf_cat(&sb, path_str);
12835         sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
12836                     io->io_hdr.flags, io->io_hdr.status);
12837         sbuf_finish(&sb);
12838         printk("%s", sbuf_data(&sb));
12839 #endif
12840
12841
12842         /*
12843          * The read is done, now we need to send status (good or bad) back
12844          * to the other side.
12845          */
12846         ctl_send_datamove_done(io, /*have_lock*/ 0);
12847
12848         return (0);
12849 }
12850
12851 static void
12852 ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
12853 {
12854         union ctl_io *io;
12855         void (*fe_datamove)(union ctl_io *io);
12856
12857         io = rq->context;
12858
12859         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12860                 printf("%s: ISC DMA read failed with error %d\n", __func__,
12861                        rq->ret);
12862                 ctl_set_internal_failure(&io->scsiio,
12863                                          /*sks_valid*/ 1,
12864                                          /*retry_count*/ rq->ret);
12865         }
12866
12867         ctl_dt_req_free(rq);
12868
12869         /* Switch the pointer over so the FETD knows what to do */
12870         io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12871
12872         /*
12873          * Use a custom move done callback, since we need to send completion
12874          * back to the other controller, not to the backend on this side.
12875          */
12876         io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
12877
12878         /* XXX KDM add checks like the ones in ctl_datamove? */
12879
12880         fe_datamove = ctl_io_port(&io->io_hdr)->fe_datamove;
12881         fe_datamove(io);
12882 }
12883
12884 static int
12885 ctl_datamove_remote_sgl_setup(union ctl_io *io)
12886 {
12887         struct ctl_sg_entry *local_sglist;
12888         uint32_t len_to_go;
12889         int retval;
12890         int i;
12891
12892         retval = 0;
12893         local_sglist = io->io_hdr.local_sglist;
12894         len_to_go = io->scsiio.kern_data_len;
12895
12896         /*
12897          * The difficult thing here is that the size of the various
12898          * S/G segments may be different than the size from the
12899          * remote controller.  That'll make it harder when DMAing
12900          * the data back to the other side.
12901          */
12902         for (i = 0; len_to_go > 0; i++) {
12903                 local_sglist[i].len = MIN(len_to_go, CTL_HA_DATAMOVE_SEGMENT);
12904                 local_sglist[i].addr =
12905                     malloc(local_sglist[i].len, M_CTL, M_WAITOK);
12906
12907                 len_to_go -= local_sglist[i].len;
12908         }
12909         /*
12910          * Reset the number of S/G entries accordingly.  The original
12911          * number of S/G entries is available in rem_sg_entries.
12912          */
12913         io->scsiio.kern_sg_entries = i;
12914
12915 #if 0
12916         printf("%s: kern_sg_entries = %d\n", __func__,
12917                io->scsiio.kern_sg_entries);
12918         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12919                 printf("%s: sg[%d] = %p, %lu\n", __func__, i,
12920                        local_sglist[i].addr, local_sglist[i].len);
12921 #endif
12922
12923         return (retval);
12924 }
12925
12926 static int
12927 ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
12928                          ctl_ha_dt_cb callback)
12929 {
12930         struct ctl_ha_dt_req *rq;
12931         struct ctl_sg_entry *remote_sglist, *local_sglist;
12932         uint32_t local_used, remote_used, total_used;
12933         int i, j, isc_ret;
12934
12935         rq = ctl_dt_req_alloc();
12936
12937         /*
12938          * If we failed to allocate the request, and if the DMA didn't fail
12939          * anyway, set busy status.  This is just a resource allocation
12940          * failure.
12941          */
12942         if ((rq == NULL)
12943          && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
12944              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS))
12945                 ctl_set_busy(&io->scsiio);
12946
12947         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
12948             (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) {
12949
12950                 if (rq != NULL)
12951                         ctl_dt_req_free(rq);
12952
12953                 /*
12954                  * The data move failed.  We need to return status back
12955                  * to the other controller.  No point in trying to DMA
12956                  * data to the remote controller.
12957                  */
12958
12959                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12960
12961                 return (1);
12962         }
12963
12964         local_sglist = io->io_hdr.local_sglist;
12965         remote_sglist = io->io_hdr.remote_sglist;
12966         local_used = 0;
12967         remote_used = 0;
12968         total_used = 0;
12969
12970         /*
12971          * Pull/push the data over the wire from/to the other controller.
12972          * This takes into account the possibility that the local and
12973          * remote sglists may not be identical in terms of the size of
12974          * the elements and the number of elements.
12975          *
12976          * One fundamental assumption here is that the length allocated for
12977          * both the local and remote sglists is identical.  Otherwise, we've
12978          * essentially got a coding error of some sort.
12979          */
12980         isc_ret = CTL_HA_STATUS_SUCCESS;
12981         for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
12982                 uint32_t cur_len;
12983                 uint8_t *tmp_ptr;
12984
12985                 rq->command = command;
12986                 rq->context = io;
12987
12988                 /*
12989                  * Both pointers should be aligned.  But it is possible
12990                  * that the allocation length is not.  They should both
12991                  * also have enough slack left over at the end, though,
12992                  * to round up to the next 8 byte boundary.
12993                  */
12994                 cur_len = MIN(local_sglist[i].len - local_used,
12995                               remote_sglist[j].len - remote_used);
12996                 rq->size = cur_len;
12997
12998                 tmp_ptr = (uint8_t *)local_sglist[i].addr;
12999                 tmp_ptr += local_used;
13000
13001 #if 0
13002                 /* Use physical addresses when talking to ISC hardware */
13003                 if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
13004                         /* XXX KDM use busdma */
13005                         rq->local = vtophys(tmp_ptr);
13006                 } else
13007                         rq->local = tmp_ptr;
13008 #else
13009                 KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0,
13010                     ("HA does not support BUS_ADDR"));
13011                 rq->local = tmp_ptr;
13012 #endif
13013
13014                 tmp_ptr = (uint8_t *)remote_sglist[j].addr;
13015                 tmp_ptr += remote_used;
13016                 rq->remote = tmp_ptr;
13017
13018                 rq->callback = NULL;
13019
13020                 local_used += cur_len;
13021                 if (local_used >= local_sglist[i].len) {
13022                         i++;
13023                         local_used = 0;
13024                 }
13025
13026                 remote_used += cur_len;
13027                 if (remote_used >= remote_sglist[j].len) {
13028                         j++;
13029                         remote_used = 0;
13030                 }
13031                 total_used += cur_len;
13032
13033                 if (total_used >= io->scsiio.kern_data_len)
13034                         rq->callback = callback;
13035
13036 #if 0
13037                 printf("%s: %s: local %p remote %p size %d\n", __func__,
13038                        (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
13039                        rq->local, rq->remote, rq->size);
13040 #endif
13041
13042                 isc_ret = ctl_dt_single(rq);
13043                 if (isc_ret > CTL_HA_STATUS_SUCCESS)
13044                         break;
13045         }
13046         if (isc_ret != CTL_HA_STATUS_WAIT) {
13047                 rq->ret = isc_ret;
13048                 callback(rq);
13049         }
13050
13051         return (0);
13052 }
13053
13054 static void
13055 ctl_datamove_remote_read(union ctl_io *io)
13056 {
13057         int retval;
13058         uint32_t i;
13059
13060         /*
13061          * This will send an error to the other controller in the case of a
13062          * failure.
13063          */
13064         retval = ctl_datamove_remote_sgl_setup(io);
13065         if (retval != 0)
13066                 return;
13067
13068         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
13069                                           ctl_datamove_remote_read_cb);
13070         if (retval != 0) {
13071                 /*
13072                  * Make sure we free memory if there was an error..  The
13073                  * ctl_datamove_remote_xfer() function will send the
13074                  * datamove done message, or call the callback with an
13075                  * error if there is a problem.
13076                  */
13077                 for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13078                         free(io->io_hdr.local_sglist[i].addr, M_CTL);
13079                 free(io->io_hdr.remote_sglist, M_CTL);
13080                 io->io_hdr.remote_sglist = NULL;
13081                 io->io_hdr.local_sglist = NULL;
13082         }
13083 }
13084
13085 /*
13086  * Process a datamove request from the other controller.  This is used for
13087  * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
13088  * first.  Once that is complete, the data gets DMAed into the remote
13089  * controller's memory.  For reads, we DMA from the remote controller's
13090  * memory into our memory first, and then move it out to the FETD.
13091  */
13092 static void
13093 ctl_datamove_remote(union ctl_io *io)
13094 {
13095
13096         mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
13097
13098         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
13099                 ctl_failover_io(io, /*have_lock*/ 0);
13100                 return;
13101         }
13102
13103         /*
13104          * Note that we look for an aborted I/O here, but don't do some of
13105          * the other checks that ctl_datamove() normally does.
13106          * We don't need to run the datamove delay code, since that should
13107          * have been done if need be on the other controller.
13108          */
13109         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
13110                 printf("%s: tag 0x%04x on (%u:%u:%u) aborted\n", __func__,
13111                        io->scsiio.tag_num, io->io_hdr.nexus.initid,
13112                        io->io_hdr.nexus.targ_port,
13113                        io->io_hdr.nexus.targ_lun);
13114                 io->io_hdr.port_status = 31338;
13115                 ctl_send_datamove_done(io, /*have_lock*/ 0);
13116                 return;
13117         }
13118
13119         if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT)
13120                 ctl_datamove_remote_write(io);
13121         else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
13122                 ctl_datamove_remote_read(io);
13123         else {
13124                 io->io_hdr.port_status = 31339;
13125                 ctl_send_datamove_done(io, /*have_lock*/ 0);
13126         }
13127 }
13128
13129 static void
13130 ctl_process_done(union ctl_io *io)
13131 {
13132         struct ctl_lun *lun;
13133         struct ctl_softc *softc = control_softc;
13134         void (*fe_done)(union ctl_io *io);
13135         union ctl_ha_msg msg;
13136         uint32_t targ_port = io->io_hdr.nexus.targ_port;
13137
13138         CTL_DEBUG_PRINT(("ctl_process_done\n"));
13139         fe_done = softc->ctl_ports[targ_port]->fe_done;
13140
13141 #ifdef CTL_TIME_IO
13142         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
13143                 char str[256];
13144                 char path_str[64];
13145                 struct sbuf sb;
13146
13147                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
13148                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13149
13150                 sbuf_cat(&sb, path_str);
13151                 switch (io->io_hdr.io_type) {
13152                 case CTL_IO_SCSI:
13153                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
13154                         sbuf_printf(&sb, "\n");
13155                         sbuf_cat(&sb, path_str);
13156                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13157                                     io->scsiio.tag_num, io->scsiio.tag_type);
13158                         break;
13159                 case CTL_IO_TASK:
13160                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
13161                                     "Tag Type: %d\n", io->taskio.task_action,
13162                                     io->taskio.tag_num, io->taskio.tag_type);
13163                         break;
13164                 default:
13165                         panic("%s: Invalid CTL I/O type %d\n",
13166                             __func__, io->io_hdr.io_type);
13167                 }
13168                 sbuf_cat(&sb, path_str);
13169                 sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
13170                             (intmax_t)time_uptime - io->io_hdr.start_time);
13171                 sbuf_finish(&sb);
13172                 printf("%s", sbuf_data(&sb));
13173         }
13174 #endif /* CTL_TIME_IO */
13175
13176         switch (io->io_hdr.io_type) {
13177         case CTL_IO_SCSI:
13178                 break;
13179         case CTL_IO_TASK:
13180                 if (ctl_debug & CTL_DEBUG_INFO)
13181                         ctl_io_error_print(io, NULL);
13182                 fe_done(io);
13183                 return;
13184         default:
13185                 panic("%s: Invalid CTL I/O type %d\n",
13186                     __func__, io->io_hdr.io_type);
13187         }
13188
13189         lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13190         if (lun == NULL) {
13191                 CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
13192                                  io->io_hdr.nexus.targ_mapped_lun));
13193                 goto bailout;
13194         }
13195
13196         mtx_lock(&lun->lun_lock);
13197
13198         /*
13199          * Check to see if we have any informational exception and status
13200          * of this command can be modified to report it in form of either
13201          * RECOVERED ERROR or NO SENSE, depending on MRIE mode page field.
13202          */
13203         if (lun->ie_reported == 0 && lun->ie_asc != 0 &&
13204             io->io_hdr.status == CTL_SUCCESS &&
13205             (io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0) {
13206                 uint8_t mrie = lun->MODE_IE.mrie;
13207                 uint8_t per = ((lun->MODE_RWER.byte3 & SMS_RWER_PER) ||
13208                     (lun->MODE_VER.byte3 & SMS_VER_PER));
13209                 if (((mrie == SIEP_MRIE_REC_COND && per) ||
13210                      mrie == SIEP_MRIE_REC_UNCOND ||
13211                      mrie == SIEP_MRIE_NO_SENSE) &&
13212                     (ctl_get_cmd_entry(&io->scsiio, NULL)->flags &
13213                      CTL_CMD_FLAG_NO_SENSE) == 0) {
13214                         ctl_set_sense(&io->scsiio,
13215                               /*current_error*/ 1,
13216                               /*sense_key*/ (mrie == SIEP_MRIE_NO_SENSE) ?
13217                                 SSD_KEY_NO_SENSE : SSD_KEY_RECOVERED_ERROR,
13218                               /*asc*/ lun->ie_asc,
13219                               /*ascq*/ lun->ie_ascq,
13220                               SSD_ELEM_NONE);
13221                         lun->ie_reported = 1;
13222                 }
13223         } else if (lun->ie_reported < 0)
13224                 lun->ie_reported = 0;
13225
13226         /*
13227          * Check to see if we have any errors to inject here.  We only
13228          * inject errors for commands that don't already have errors set.
13229          */
13230         if (!STAILQ_EMPTY(&lun->error_list) &&
13231             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) &&
13232             ((io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0))
13233                 ctl_inject_error(lun, io);
13234
13235         /*
13236          * XXX KDM how do we treat commands that aren't completed
13237          * successfully?
13238          *
13239          * XXX KDM should we also track I/O latency?
13240          */
13241         if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
13242             io->io_hdr.io_type == CTL_IO_SCSI) {
13243 #ifdef CTL_TIME_IO
13244                 struct bintime cur_bt;
13245 #endif
13246                 int type;
13247
13248                 if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13249                     CTL_FLAG_DATA_IN)
13250                         type = CTL_STATS_READ;
13251                 else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13252                     CTL_FLAG_DATA_OUT)
13253                         type = CTL_STATS_WRITE;
13254                 else
13255                         type = CTL_STATS_NO_IO;
13256
13257                 lun->stats.ports[targ_port].bytes[type] +=
13258                     io->scsiio.kern_total_len;
13259                 lun->stats.ports[targ_port].operations[type]++;
13260 #ifdef CTL_TIME_IO
13261                 bintime_add(&lun->stats.ports[targ_port].dma_time[type],
13262                    &io->io_hdr.dma_bt);
13263                 getbinuptime(&cur_bt);
13264                 bintime_sub(&cur_bt, &io->io_hdr.start_bt);
13265                 bintime_add(&lun->stats.ports[targ_port].time[type], &cur_bt);
13266 #endif
13267                 lun->stats.ports[targ_port].num_dmas[type] +=
13268                     io->io_hdr.num_dmas;
13269         }
13270
13271         /*
13272          * Remove this from the OOA queue.
13273          */
13274         TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13275 #ifdef CTL_TIME_IO
13276         if (TAILQ_EMPTY(&lun->ooa_queue))
13277                 lun->last_busy = getsbinuptime();
13278 #endif
13279
13280         /*
13281          * Run through the blocked queue on this LUN and see if anything
13282          * has become unblocked, now that this transaction is done.
13283          */
13284         ctl_check_blocked(lun);
13285
13286         /*
13287          * If the LUN has been invalidated, free it if there is nothing
13288          * left on its OOA queue.
13289          */
13290         if ((lun->flags & CTL_LUN_INVALID)
13291          && TAILQ_EMPTY(&lun->ooa_queue)) {
13292                 mtx_unlock(&lun->lun_lock);
13293                 mtx_lock(&softc->ctl_lock);
13294                 ctl_free_lun(lun);
13295                 mtx_unlock(&softc->ctl_lock);
13296         } else
13297                 mtx_unlock(&lun->lun_lock);
13298
13299 bailout:
13300
13301         /*
13302          * If this command has been aborted, make sure we set the status
13303          * properly.  The FETD is responsible for freeing the I/O and doing
13304          * whatever it needs to do to clean up its state.
13305          */
13306         if (io->io_hdr.flags & CTL_FLAG_ABORT)
13307                 ctl_set_task_aborted(&io->scsiio);
13308
13309         /*
13310          * If enabled, print command error status.
13311          */
13312         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS &&
13313             (ctl_debug & CTL_DEBUG_INFO) != 0)
13314                 ctl_io_error_print(io, NULL);
13315
13316         /*
13317          * Tell the FETD or the other shelf controller we're done with this
13318          * command.  Note that only SCSI commands get to this point.  Task
13319          * management commands are completed above.
13320          */
13321         if ((softc->ha_mode != CTL_HA_MODE_XFER) &&
13322             (io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)) {
13323                 memset(&msg, 0, sizeof(msg));
13324                 msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13325                 msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
13326                 msg.hdr.nexus = io->io_hdr.nexus;
13327                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13328                     sizeof(msg.scsi) - sizeof(msg.scsi.sense_data),
13329                     M_WAITOK);
13330         }
13331
13332         fe_done(io);
13333 }
13334
13335 #ifdef CTL_WITH_CA
13336 /*
13337  * Front end should call this if it doesn't do autosense.  When the request
13338  * sense comes back in from the initiator, we'll dequeue this and send it.
13339  */
13340 int
13341 ctl_queue_sense(union ctl_io *io)
13342 {
13343         struct ctl_lun *lun;
13344         struct ctl_port *port;
13345         struct ctl_softc *softc;
13346         uint32_t initidx, targ_lun;
13347
13348         softc = control_softc;
13349
13350         CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13351
13352         /*
13353          * LUN lookup will likely move to the ctl_work_thread() once we
13354          * have our new queueing infrastructure (that doesn't put things on
13355          * a per-LUN queue initially).  That is so that we can handle
13356          * things like an INQUIRY to a LUN that we don't have enabled.  We
13357          * can't deal with that right now.
13358          */
13359         mtx_lock(&softc->ctl_lock);
13360
13361         /*
13362          * If we don't have a LUN for this, just toss the sense
13363          * information.
13364          */
13365         port = ctl_io_port(&ctsio->io_hdr);
13366         targ_lun = ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13367         if ((targ_lun < CTL_MAX_LUNS)
13368          && (softc->ctl_luns[targ_lun] != NULL))
13369                 lun = softc->ctl_luns[targ_lun];
13370         else
13371                 goto bailout;
13372
13373         initidx = ctl_get_initindex(&io->io_hdr.nexus);
13374
13375         mtx_lock(&lun->lun_lock);
13376         /*
13377          * Already have CA set for this LUN...toss the sense information.
13378          */
13379         if (ctl_is_set(lun->have_ca, initidx)) {
13380                 mtx_unlock(&lun->lun_lock);
13381                 goto bailout;
13382         }
13383
13384         memcpy(&lun->pending_sense[initidx], &io->scsiio.sense_data,
13385                MIN(sizeof(lun->pending_sense[initidx]),
13386                sizeof(io->scsiio.sense_data)));
13387         ctl_set_mask(lun->have_ca, initidx);
13388         mtx_unlock(&lun->lun_lock);
13389
13390 bailout:
13391         mtx_unlock(&softc->ctl_lock);
13392
13393         ctl_free_io(io);
13394
13395         return (CTL_RETVAL_COMPLETE);
13396 }
13397 #endif
13398
13399 /*
13400  * Primary command inlet from frontend ports.  All SCSI and task I/O
13401  * requests must go through this function.
13402  */
13403 int
13404 ctl_queue(union ctl_io *io)
13405 {
13406         struct ctl_port *port;
13407
13408         CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13409
13410 #ifdef CTL_TIME_IO
13411         io->io_hdr.start_time = time_uptime;
13412         getbinuptime(&io->io_hdr.start_bt);
13413 #endif /* CTL_TIME_IO */
13414
13415         /* Map FE-specific LUN ID into global one. */
13416         port = ctl_io_port(&io->io_hdr);
13417         io->io_hdr.nexus.targ_mapped_lun =
13418             ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13419
13420         switch (io->io_hdr.io_type) {
13421         case CTL_IO_SCSI:
13422         case CTL_IO_TASK:
13423                 if (ctl_debug & CTL_DEBUG_CDB)
13424                         ctl_io_print(io);
13425                 ctl_enqueue_incoming(io);
13426                 break;
13427         default:
13428                 printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13429                 return (EINVAL);
13430         }
13431
13432         return (CTL_RETVAL_COMPLETE);
13433 }
13434
13435 #ifdef CTL_IO_DELAY
13436 static void
13437 ctl_done_timer_wakeup(void *arg)
13438 {
13439         union ctl_io *io;
13440
13441         io = (union ctl_io *)arg;
13442         ctl_done(io);
13443 }
13444 #endif /* CTL_IO_DELAY */
13445
13446 void
13447 ctl_serseq_done(union ctl_io *io)
13448 {
13449         struct ctl_lun *lun;
13450
13451         lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13452         if (lun->be_lun == NULL ||
13453             lun->be_lun->serseq == CTL_LUN_SERSEQ_OFF)
13454                 return;
13455         mtx_lock(&lun->lun_lock);
13456         io->io_hdr.flags |= CTL_FLAG_SERSEQ_DONE;
13457         ctl_check_blocked(lun);
13458         mtx_unlock(&lun->lun_lock);
13459 }
13460
13461 void
13462 ctl_done(union ctl_io *io)
13463 {
13464
13465         /*
13466          * Enable this to catch duplicate completion issues.
13467          */
13468 #if 0
13469         if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13470                 printf("%s: type %d msg %d cdb %x iptl: "
13471                        "%u:%u:%u tag 0x%04x "
13472                        "flag %#x status %x\n",
13473                         __func__,
13474                         io->io_hdr.io_type,
13475                         io->io_hdr.msg_type,
13476                         io->scsiio.cdb[0],
13477                         io->io_hdr.nexus.initid,
13478                         io->io_hdr.nexus.targ_port,
13479                         io->io_hdr.nexus.targ_lun,
13480                         (io->io_hdr.io_type ==
13481                         CTL_IO_TASK) ?
13482                         io->taskio.tag_num :
13483                         io->scsiio.tag_num,
13484                         io->io_hdr.flags,
13485                         io->io_hdr.status);
13486         } else
13487                 io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13488 #endif
13489
13490         /*
13491          * This is an internal copy of an I/O, and should not go through
13492          * the normal done processing logic.
13493          */
13494         if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
13495                 return;
13496
13497 #ifdef CTL_IO_DELAY
13498         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13499                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13500         } else {
13501                 struct ctl_lun *lun;
13502
13503                 lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13504
13505                 if ((lun != NULL)
13506                  && (lun->delay_info.done_delay > 0)) {
13507
13508                         callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
13509                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13510                         callout_reset(&io->io_hdr.delay_callout,
13511                                       lun->delay_info.done_delay * hz,
13512                                       ctl_done_timer_wakeup, io);
13513                         if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13514                                 lun->delay_info.done_delay = 0;
13515                         return;
13516                 }
13517         }
13518 #endif /* CTL_IO_DELAY */
13519
13520         ctl_enqueue_done(io);
13521 }
13522
13523 static void
13524 ctl_work_thread(void *arg)
13525 {
13526         struct ctl_thread *thr = (struct ctl_thread *)arg;
13527         struct ctl_softc *softc = thr->ctl_softc;
13528         union ctl_io *io;
13529         int retval;
13530
13531         CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13532
13533         for (;;) {
13534                 /*
13535                  * We handle the queues in this order:
13536                  * - ISC
13537                  * - done queue (to free up resources, unblock other commands)
13538                  * - RtR queue
13539                  * - incoming queue
13540                  *
13541                  * If those queues are empty, we break out of the loop and
13542                  * go to sleep.
13543                  */
13544                 mtx_lock(&thr->queue_lock);
13545                 io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
13546                 if (io != NULL) {
13547                         STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
13548                         mtx_unlock(&thr->queue_lock);
13549                         ctl_handle_isc(io);
13550                         continue;
13551                 }
13552                 io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
13553                 if (io != NULL) {
13554                         STAILQ_REMOVE_HEAD(&thr->done_queue, links);
13555                         /* clear any blocked commands, call fe_done */
13556                         mtx_unlock(&thr->queue_lock);
13557                         ctl_process_done(io);
13558                         continue;
13559                 }
13560                 io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
13561                 if (io != NULL) {
13562                         STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
13563                         mtx_unlock(&thr->queue_lock);
13564                         if (io->io_hdr.io_type == CTL_IO_TASK)
13565                                 ctl_run_task(io);
13566                         else
13567                                 ctl_scsiio_precheck(softc, &io->scsiio);
13568                         continue;
13569                 }
13570                 io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
13571                 if (io != NULL) {
13572                         STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
13573                         mtx_unlock(&thr->queue_lock);
13574                         retval = ctl_scsiio(&io->scsiio);
13575                         if (retval != CTL_RETVAL_COMPLETE)
13576                                 CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13577                         continue;
13578                 }
13579
13580                 /* Sleep until we have something to do. */
13581                 mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
13582         }
13583 }
13584
13585 static void
13586 ctl_lun_thread(void *arg)
13587 {
13588         struct ctl_softc *softc = (struct ctl_softc *)arg;
13589         struct ctl_be_lun *be_lun;
13590
13591         CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
13592
13593         for (;;) {
13594                 mtx_lock(&softc->ctl_lock);
13595                 be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
13596                 if (be_lun != NULL) {
13597                         STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
13598                         mtx_unlock(&softc->ctl_lock);
13599                         ctl_create_lun(be_lun);
13600                         continue;
13601                 }
13602
13603                 /* Sleep until we have something to do. */
13604                 mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
13605                     PDROP | PRIBIO, "-", 0);
13606         }
13607 }
13608
13609 static void
13610 ctl_thresh_thread(void *arg)
13611 {
13612         struct ctl_softc *softc = (struct ctl_softc *)arg;
13613         struct ctl_lun *lun;
13614         struct ctl_logical_block_provisioning_page *page;
13615         const char *attr;
13616         union ctl_ha_msg msg;
13617         uint64_t thres, val;
13618         int i, e, set;
13619
13620         CTL_DEBUG_PRINT(("ctl_thresh_thread starting\n"));
13621
13622         for (;;) {
13623                 mtx_lock(&softc->ctl_lock);
13624                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
13625                         if ((lun->flags & CTL_LUN_DISABLED) ||
13626                             (lun->flags & CTL_LUN_NO_MEDIA) ||
13627                             lun->backend->lun_attr == NULL)
13628                                 continue;
13629                         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
13630                             softc->ha_mode == CTL_HA_MODE_XFER)
13631                                 continue;
13632                         if ((lun->MODE_RWER.byte8 & SMS_RWER_LBPERE) == 0)
13633                                 continue;
13634                         e = 0;
13635                         page = &lun->MODE_LBP;
13636                         for (i = 0; i < CTL_NUM_LBP_THRESH; i++) {
13637                                 if ((page->descr[i].flags & SLBPPD_ENABLED) == 0)
13638                                         continue;
13639                                 thres = scsi_4btoul(page->descr[i].count);
13640                                 thres <<= CTL_LBP_EXPONENT;
13641                                 switch (page->descr[i].resource) {
13642                                 case 0x01:
13643                                         attr = "blocksavail";
13644                                         break;
13645                                 case 0x02:
13646                                         attr = "blocksused";
13647                                         break;
13648                                 case 0xf1:
13649                                         attr = "poolblocksavail";
13650                                         break;
13651                                 case 0xf2:
13652                                         attr = "poolblocksused";
13653                                         break;
13654                                 default:
13655                                         continue;
13656                                 }
13657                                 mtx_unlock(&softc->ctl_lock); // XXX
13658                                 val = lun->backend->lun_attr(
13659                                     lun->be_lun->be_lun, attr);
13660                                 mtx_lock(&softc->ctl_lock);
13661                                 if (val == UINT64_MAX)
13662                                         continue;
13663                                 if ((page->descr[i].flags & SLBPPD_ARMING_MASK)
13664                                     == SLBPPD_ARMING_INC)
13665                                         e = (val >= thres);
13666                                 else
13667                                         e = (val <= thres);
13668                                 if (e)
13669                                         break;
13670                         }
13671                         mtx_lock(&lun->lun_lock);
13672                         if (e) {
13673                                 scsi_u64to8b((uint8_t *)&page->descr[i] -
13674                                     (uint8_t *)page, lun->ua_tpt_info);
13675                                 if (lun->lasttpt == 0 ||
13676                                     time_uptime - lun->lasttpt >= CTL_LBP_UA_PERIOD) {
13677                                         lun->lasttpt = time_uptime;
13678                                         ctl_est_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13679                                         set = 1;
13680                                 } else
13681                                         set = 0;
13682                         } else {
13683                                 lun->lasttpt = 0;
13684                                 ctl_clr_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13685                                 set = -1;
13686                         }
13687                         mtx_unlock(&lun->lun_lock);
13688                         if (set != 0 &&
13689                             lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
13690                                 /* Send msg to other side. */
13691                                 bzero(&msg.ua, sizeof(msg.ua));
13692                                 msg.hdr.msg_type = CTL_MSG_UA;
13693                                 msg.hdr.nexus.initid = -1;
13694                                 msg.hdr.nexus.targ_port = -1;
13695                                 msg.hdr.nexus.targ_lun = lun->lun;
13696                                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
13697                                 msg.ua.ua_all = 1;
13698                                 msg.ua.ua_set = (set > 0);
13699                                 msg.ua.ua_type = CTL_UA_THIN_PROV_THRES;
13700                                 memcpy(msg.ua.ua_info, lun->ua_tpt_info, 8);
13701                                 mtx_unlock(&softc->ctl_lock); // XXX
13702                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13703                                     sizeof(msg.ua), M_WAITOK);
13704                                 mtx_lock(&softc->ctl_lock);
13705                         }
13706                 }
13707                 mtx_unlock(&softc->ctl_lock);
13708                 pause("-", CTL_LBP_PERIOD * hz);
13709         }
13710 }
13711
13712 static void
13713 ctl_enqueue_incoming(union ctl_io *io)
13714 {
13715         struct ctl_softc *softc = control_softc;
13716         struct ctl_thread *thr;
13717         u_int idx;
13718
13719         idx = (io->io_hdr.nexus.targ_port * 127 +
13720                io->io_hdr.nexus.initid) % worker_threads;
13721         thr = &softc->threads[idx];
13722         mtx_lock(&thr->queue_lock);
13723         STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
13724         mtx_unlock(&thr->queue_lock);
13725         wakeup(thr);
13726 }
13727
13728 static void
13729 ctl_enqueue_rtr(union ctl_io *io)
13730 {
13731         struct ctl_softc *softc = control_softc;
13732         struct ctl_thread *thr;
13733
13734         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13735         mtx_lock(&thr->queue_lock);
13736         STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
13737         mtx_unlock(&thr->queue_lock);
13738         wakeup(thr);
13739 }
13740
13741 static void
13742 ctl_enqueue_done(union ctl_io *io)
13743 {
13744         struct ctl_softc *softc = control_softc;
13745         struct ctl_thread *thr;
13746
13747         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13748         mtx_lock(&thr->queue_lock);
13749         STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
13750         mtx_unlock(&thr->queue_lock);
13751         wakeup(thr);
13752 }
13753
13754 static void
13755 ctl_enqueue_isc(union ctl_io *io)
13756 {
13757         struct ctl_softc *softc = control_softc;
13758         struct ctl_thread *thr;
13759
13760         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13761         mtx_lock(&thr->queue_lock);
13762         STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
13763         mtx_unlock(&thr->queue_lock);
13764         wakeup(thr);
13765 }
13766
13767 /*
13768  *  vim: ts=8
13769  */