xref: /freebsd/sys/cam/ctl/ctl.c (revision 2e3507c25e42292b45a5482e116d278f5515d04d)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2003-2009 Silicon Graphics International Corp.
5  * Copyright (c) 2012 The FreeBSD Foundation
6  * Copyright (c) 2014-2017 Alexander Motin <mav@FreeBSD.org>
7  * Copyright (c) 2017 Jakub Wojciech Klama <jceel@FreeBSD.org>
8  * Copyright (c) 2018 Marcelo Araujo <araujo@FreeBSD.org>
9  * All rights reserved.
10  *
11  * Portions of this software were developed by Edward Tomasz Napierala
12  * under sponsorship from the FreeBSD Foundation.
13  *
14  * Redistribution and use in source and binary forms, with or without
15  * modification, are permitted provided that the following conditions
16  * are met:
17  * 1. Redistributions of source code must retain the above copyright
18  *    notice, this list of conditions, and the following disclaimer,
19  *    without modification.
20  * 2. Redistributions in binary form must reproduce at minimum a disclaimer
21  *    substantially similar to the "NO WARRANTY" disclaimer below
22  *    ("Disclaimer") and any redistribution must be conditioned upon
23  *    including a substantially similar Disclaimer requirement for further
24  *    binary redistribution.
25  *
26  * NO WARRANTY
27  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
28  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
29  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR
30  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
31  * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
32  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
33  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
34  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
35  * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
36  * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
37  * POSSIBILITY OF SUCH DAMAGES.
38  *
39  * $Id$
40  */
41 /*
42  * CAM Target Layer, a SCSI device emulation subsystem.
43  *
44  * Author: Ken Merry <ken@FreeBSD.org>
45  */
46 
47 #include <sys/param.h>
48 #include <sys/systm.h>
49 #include <sys/ctype.h>
50 #include <sys/kernel.h>
51 #include <sys/types.h>
52 #include <sys/kthread.h>
53 #include <sys/bio.h>
54 #include <sys/fcntl.h>
55 #include <sys/lock.h>
56 #include <sys/module.h>
57 #include <sys/mutex.h>
58 #include <sys/condvar.h>
59 #include <sys/malloc.h>
60 #include <sys/conf.h>
61 #include <sys/ioccom.h>
62 #include <sys/queue.h>
63 #include <sys/sbuf.h>
64 #include <sys/smp.h>
65 #include <sys/endian.h>
66 #include <sys/proc.h>
67 #include <sys/sched.h>
68 #include <sys/sysctl.h>
69 #include <sys/nv.h>
70 #include <sys/dnv.h>
71 #include <vm/uma.h>
72 
73 #include <cam/cam.h>
74 #include <cam/scsi/scsi_all.h>
75 #include <cam/scsi/scsi_cd.h>
76 #include <cam/scsi/scsi_da.h>
77 #include <cam/ctl/ctl_io.h>
78 #include <cam/ctl/ctl.h>
79 #include <cam/ctl/ctl_frontend.h>
80 #include <cam/ctl/ctl_util.h>
81 #include <cam/ctl/ctl_backend.h>
82 #include <cam/ctl/ctl_ioctl.h>
83 #include <cam/ctl/ctl_ha.h>
84 #include <cam/ctl/ctl_private.h>
85 #include <cam/ctl/ctl_debug.h>
86 #include <cam/ctl/ctl_scsi_all.h>
87 #include <cam/ctl/ctl_error.h>
88 
89 struct ctl_softc *control_softc = NULL;
90 
91 /*
92  * Template mode pages.
93  */
94 
95 /*
96  * Note that these are default values only.  The actual values will be
97  * filled in when the user does a mode sense.
98  */
99 const static struct scsi_da_rw_recovery_page rw_er_page_default = {
100 	/*page_code*/SMS_RW_ERROR_RECOVERY_PAGE,
101 	/*page_length*/sizeof(struct scsi_da_rw_recovery_page) - 2,
102 	/*byte3*/SMS_RWER_AWRE|SMS_RWER_ARRE,
103 	/*read_retry_count*/0,
104 	/*correction_span*/0,
105 	/*head_offset_count*/0,
106 	/*data_strobe_offset_cnt*/0,
107 	/*byte8*/SMS_RWER_LBPERE,
108 	/*write_retry_count*/0,
109 	/*reserved2*/0,
110 	/*recovery_time_limit*/{0, 0},
111 };
112 
113 const static struct scsi_da_rw_recovery_page rw_er_page_changeable = {
114 	/*page_code*/SMS_RW_ERROR_RECOVERY_PAGE,
115 	/*page_length*/sizeof(struct scsi_da_rw_recovery_page) - 2,
116 	/*byte3*/SMS_RWER_PER,
117 	/*read_retry_count*/0,
118 	/*correction_span*/0,
119 	/*head_offset_count*/0,
120 	/*data_strobe_offset_cnt*/0,
121 	/*byte8*/SMS_RWER_LBPERE,
122 	/*write_retry_count*/0,
123 	/*reserved2*/0,
124 	/*recovery_time_limit*/{0, 0},
125 };
126 
127 const static struct scsi_format_page format_page_default = {
128 	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
129 	/*page_length*/sizeof(struct scsi_format_page) - 2,
130 	/*tracks_per_zone*/ {0, 0},
131 	/*alt_sectors_per_zone*/ {0, 0},
132 	/*alt_tracks_per_zone*/ {0, 0},
133 	/*alt_tracks_per_lun*/ {0, 0},
134 	/*sectors_per_track*/ {(CTL_DEFAULT_SECTORS_PER_TRACK >> 8) & 0xff,
135 			        CTL_DEFAULT_SECTORS_PER_TRACK & 0xff},
136 	/*bytes_per_sector*/ {0, 0},
137 	/*interleave*/ {0, 0},
138 	/*track_skew*/ {0, 0},
139 	/*cylinder_skew*/ {0, 0},
140 	/*flags*/ SFP_HSEC,
141 	/*reserved*/ {0, 0, 0}
142 };
143 
144 const static struct scsi_format_page format_page_changeable = {
145 	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
146 	/*page_length*/sizeof(struct scsi_format_page) - 2,
147 	/*tracks_per_zone*/ {0, 0},
148 	/*alt_sectors_per_zone*/ {0, 0},
149 	/*alt_tracks_per_zone*/ {0, 0},
150 	/*alt_tracks_per_lun*/ {0, 0},
151 	/*sectors_per_track*/ {0, 0},
152 	/*bytes_per_sector*/ {0, 0},
153 	/*interleave*/ {0, 0},
154 	/*track_skew*/ {0, 0},
155 	/*cylinder_skew*/ {0, 0},
156 	/*flags*/ 0,
157 	/*reserved*/ {0, 0, 0}
158 };
159 
160 const static struct scsi_rigid_disk_page rigid_disk_page_default = {
161 	/*page_code*/SMS_RIGID_DISK_PAGE,
162 	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
163 	/*cylinders*/ {0, 0, 0},
164 	/*heads*/ CTL_DEFAULT_HEADS,
165 	/*start_write_precomp*/ {0, 0, 0},
166 	/*start_reduced_current*/ {0, 0, 0},
167 	/*step_rate*/ {0, 0},
168 	/*landing_zone_cylinder*/ {0, 0, 0},
169 	/*rpl*/ SRDP_RPL_DISABLED,
170 	/*rotational_offset*/ 0,
171 	/*reserved1*/ 0,
172 	/*rotation_rate*/ {(CTL_DEFAULT_ROTATION_RATE >> 8) & 0xff,
173 			   CTL_DEFAULT_ROTATION_RATE & 0xff},
174 	/*reserved2*/ {0, 0}
175 };
176 
177 const static struct scsi_rigid_disk_page rigid_disk_page_changeable = {
178 	/*page_code*/SMS_RIGID_DISK_PAGE,
179 	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
180 	/*cylinders*/ {0, 0, 0},
181 	/*heads*/ 0,
182 	/*start_write_precomp*/ {0, 0, 0},
183 	/*start_reduced_current*/ {0, 0, 0},
184 	/*step_rate*/ {0, 0},
185 	/*landing_zone_cylinder*/ {0, 0, 0},
186 	/*rpl*/ 0,
187 	/*rotational_offset*/ 0,
188 	/*reserved1*/ 0,
189 	/*rotation_rate*/ {0, 0},
190 	/*reserved2*/ {0, 0}
191 };
192 
193 const static struct scsi_da_verify_recovery_page verify_er_page_default = {
194 	/*page_code*/SMS_VERIFY_ERROR_RECOVERY_PAGE,
195 	/*page_length*/sizeof(struct scsi_da_verify_recovery_page) - 2,
196 	/*byte3*/0,
197 	/*read_retry_count*/0,
198 	/*reserved*/{ 0, 0, 0, 0, 0, 0 },
199 	/*recovery_time_limit*/{0, 0},
200 };
201 
202 const static struct scsi_da_verify_recovery_page verify_er_page_changeable = {
203 	/*page_code*/SMS_VERIFY_ERROR_RECOVERY_PAGE,
204 	/*page_length*/sizeof(struct scsi_da_verify_recovery_page) - 2,
205 	/*byte3*/SMS_VER_PER,
206 	/*read_retry_count*/0,
207 	/*reserved*/{ 0, 0, 0, 0, 0, 0 },
208 	/*recovery_time_limit*/{0, 0},
209 };
210 
211 const static struct scsi_caching_page caching_page_default = {
212 	/*page_code*/SMS_CACHING_PAGE,
213 	/*page_length*/sizeof(struct scsi_caching_page) - 2,
214 	/*flags1*/ SCP_DISC | SCP_WCE,
215 	/*ret_priority*/ 0,
216 	/*disable_pf_transfer_len*/ {0xff, 0xff},
217 	/*min_prefetch*/ {0, 0},
218 	/*max_prefetch*/ {0xff, 0xff},
219 	/*max_pf_ceiling*/ {0xff, 0xff},
220 	/*flags2*/ 0,
221 	/*cache_segments*/ 0,
222 	/*cache_seg_size*/ {0, 0},
223 	/*reserved*/ 0,
224 	/*non_cache_seg_size*/ {0, 0, 0}
225 };
226 
227 const static struct scsi_caching_page caching_page_changeable = {
228 	/*page_code*/SMS_CACHING_PAGE,
229 	/*page_length*/sizeof(struct scsi_caching_page) - 2,
230 	/*flags1*/ SCP_WCE | SCP_RCD,
231 	/*ret_priority*/ 0,
232 	/*disable_pf_transfer_len*/ {0, 0},
233 	/*min_prefetch*/ {0, 0},
234 	/*max_prefetch*/ {0, 0},
235 	/*max_pf_ceiling*/ {0, 0},
236 	/*flags2*/ 0,
237 	/*cache_segments*/ 0,
238 	/*cache_seg_size*/ {0, 0},
239 	/*reserved*/ 0,
240 	/*non_cache_seg_size*/ {0, 0, 0}
241 };
242 
243 const static struct scsi_control_page control_page_default = {
244 	/*page_code*/SMS_CONTROL_MODE_PAGE,
245 	/*page_length*/sizeof(struct scsi_control_page) - 2,
246 	/*rlec*/0,
247 	/*queue_flags*/SCP_QUEUE_ALG_RESTRICTED,
248 	/*eca_and_aen*/0,
249 	/*flags4*/SCP_TAS,
250 	/*aen_holdoff_period*/{0, 0},
251 	/*busy_timeout_period*/{0, 0},
252 	/*extended_selftest_completion_time*/{0, 0}
253 };
254 
255 const static struct scsi_control_page control_page_changeable = {
256 	/*page_code*/SMS_CONTROL_MODE_PAGE,
257 	/*page_length*/sizeof(struct scsi_control_page) - 2,
258 	/*rlec*/SCP_DSENSE,
259 	/*queue_flags*/SCP_QUEUE_ALG_MASK | SCP_NUAR,
260 	/*eca_and_aen*/SCP_SWP,
261 	/*flags4*/0,
262 	/*aen_holdoff_period*/{0, 0},
263 	/*busy_timeout_period*/{0, 0},
264 	/*extended_selftest_completion_time*/{0, 0}
265 };
266 
267 #define CTL_CEM_LEN	(sizeof(struct scsi_control_ext_page) - 4)
268 
269 const static struct scsi_control_ext_page control_ext_page_default = {
270 	/*page_code*/SMS_CONTROL_MODE_PAGE | SMPH_SPF,
271 	/*subpage_code*/0x01,
272 	/*page_length*/{CTL_CEM_LEN >> 8, CTL_CEM_LEN},
273 	/*flags*/0,
274 	/*prio*/0,
275 	/*max_sense*/0
276 };
277 
278 const static struct scsi_control_ext_page control_ext_page_changeable = {
279 	/*page_code*/SMS_CONTROL_MODE_PAGE | SMPH_SPF,
280 	/*subpage_code*/0x01,
281 	/*page_length*/{CTL_CEM_LEN >> 8, CTL_CEM_LEN},
282 	/*flags*/0,
283 	/*prio*/0,
284 	/*max_sense*/0xff
285 };
286 
287 const static struct scsi_info_exceptions_page ie_page_default = {
288 	/*page_code*/SMS_INFO_EXCEPTIONS_PAGE,
289 	/*page_length*/sizeof(struct scsi_info_exceptions_page) - 2,
290 	/*info_flags*/SIEP_FLAGS_EWASC,
291 	/*mrie*/SIEP_MRIE_NO,
292 	/*interval_timer*/{0, 0, 0, 0},
293 	/*report_count*/{0, 0, 0, 1}
294 };
295 
296 const static struct scsi_info_exceptions_page ie_page_changeable = {
297 	/*page_code*/SMS_INFO_EXCEPTIONS_PAGE,
298 	/*page_length*/sizeof(struct scsi_info_exceptions_page) - 2,
299 	/*info_flags*/SIEP_FLAGS_EWASC | SIEP_FLAGS_DEXCPT | SIEP_FLAGS_TEST |
300 	    SIEP_FLAGS_LOGERR,
301 	/*mrie*/0x0f,
302 	/*interval_timer*/{0xff, 0xff, 0xff, 0xff},
303 	/*report_count*/{0xff, 0xff, 0xff, 0xff}
304 };
305 
306 #define CTL_LBPM_LEN	(sizeof(struct ctl_logical_block_provisioning_page) - 4)
307 
308 const static struct ctl_logical_block_provisioning_page lbp_page_default = {{
309 	/*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF,
310 	/*subpage_code*/0x02,
311 	/*page_length*/{CTL_LBPM_LEN >> 8, CTL_LBPM_LEN},
312 	/*flags*/0,
313 	/*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
314 	/*descr*/{}},
315 	{{/*flags*/0,
316 	  /*resource*/0x01,
317 	  /*reserved*/{0, 0},
318 	  /*count*/{0, 0, 0, 0}},
319 	 {/*flags*/0,
320 	  /*resource*/0x02,
321 	  /*reserved*/{0, 0},
322 	  /*count*/{0, 0, 0, 0}},
323 	 {/*flags*/0,
324 	  /*resource*/0xf1,
325 	  /*reserved*/{0, 0},
326 	  /*count*/{0, 0, 0, 0}},
327 	 {/*flags*/0,
328 	  /*resource*/0xf2,
329 	  /*reserved*/{0, 0},
330 	  /*count*/{0, 0, 0, 0}}
331 	}
332 };
333 
334 const static struct ctl_logical_block_provisioning_page lbp_page_changeable = {{
335 	/*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF,
336 	/*subpage_code*/0x02,
337 	/*page_length*/{CTL_LBPM_LEN >> 8, CTL_LBPM_LEN},
338 	/*flags*/SLBPP_SITUA,
339 	/*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
340 	/*descr*/{}},
341 	{{/*flags*/0,
342 	  /*resource*/0,
343 	  /*reserved*/{0, 0},
344 	  /*count*/{0, 0, 0, 0}},
345 	 {/*flags*/0,
346 	  /*resource*/0,
347 	  /*reserved*/{0, 0},
348 	  /*count*/{0, 0, 0, 0}},
349 	 {/*flags*/0,
350 	  /*resource*/0,
351 	  /*reserved*/{0, 0},
352 	  /*count*/{0, 0, 0, 0}},
353 	 {/*flags*/0,
354 	  /*resource*/0,
355 	  /*reserved*/{0, 0},
356 	  /*count*/{0, 0, 0, 0}}
357 	}
358 };
359 
360 const static struct scsi_cddvd_capabilities_page cddvd_page_default = {
361 	/*page_code*/SMS_CDDVD_CAPS_PAGE,
362 	/*page_length*/sizeof(struct scsi_cddvd_capabilities_page) - 2,
363 	/*caps1*/0x3f,
364 	/*caps2*/0x00,
365 	/*caps3*/0xf0,
366 	/*caps4*/0x00,
367 	/*caps5*/0x29,
368 	/*caps6*/0x00,
369 	/*obsolete*/{0, 0},
370 	/*nvol_levels*/{0, 0},
371 	/*buffer_size*/{8, 0},
372 	/*obsolete2*/{0, 0},
373 	/*reserved*/0,
374 	/*digital*/0,
375 	/*obsolete3*/0,
376 	/*copy_management*/0,
377 	/*reserved2*/0,
378 	/*rotation_control*/0,
379 	/*cur_write_speed*/0,
380 	/*num_speed_descr*/0,
381 };
382 
383 const static struct scsi_cddvd_capabilities_page cddvd_page_changeable = {
384 	/*page_code*/SMS_CDDVD_CAPS_PAGE,
385 	/*page_length*/sizeof(struct scsi_cddvd_capabilities_page) - 2,
386 	/*caps1*/0,
387 	/*caps2*/0,
388 	/*caps3*/0,
389 	/*caps4*/0,
390 	/*caps5*/0,
391 	/*caps6*/0,
392 	/*obsolete*/{0, 0},
393 	/*nvol_levels*/{0, 0},
394 	/*buffer_size*/{0, 0},
395 	/*obsolete2*/{0, 0},
396 	/*reserved*/0,
397 	/*digital*/0,
398 	/*obsolete3*/0,
399 	/*copy_management*/0,
400 	/*reserved2*/0,
401 	/*rotation_control*/0,
402 	/*cur_write_speed*/0,
403 	/*num_speed_descr*/0,
404 };
405 
406 SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
407     "CAM Target Layer");
408 static int worker_threads = -1;
409 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, worker_threads, CTLFLAG_RDTUN,
410     &worker_threads, 1, "Number of worker threads");
411 static int ctl_debug = CTL_DEBUG_NONE;
412 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, debug, CTLFLAG_RWTUN,
413     &ctl_debug, 0, "Enabled debug flags");
414 static int ctl_lun_map_size = 1024;
415 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, lun_map_size, CTLFLAG_RWTUN,
416     &ctl_lun_map_size, 0, "Size of per-port LUN map (max LUN + 1)");
417 #ifdef  CTL_TIME_IO
418 static int ctl_time_io_secs = CTL_TIME_IO_DEFAULT_SECS;
419 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, time_io_secs, CTLFLAG_RWTUN,
420     &ctl_time_io_secs, 0, "Log requests taking more seconds");
421 #endif
422 
423 /*
424  * Maximum number of LUNs we support.  MUST be a power of 2.
425  */
426 #define	CTL_DEFAULT_MAX_LUNS	1024
427 static int ctl_max_luns = CTL_DEFAULT_MAX_LUNS;
428 TUNABLE_INT("kern.cam.ctl.max_luns", &ctl_max_luns);
429 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, max_luns, CTLFLAG_RDTUN,
430     &ctl_max_luns, CTL_DEFAULT_MAX_LUNS, "Maximum number of LUNs");
431 
432 /*
433  * Maximum number of ports registered at one time.
434  */
435 #define	CTL_DEFAULT_MAX_PORTS		1024
436 static int ctl_max_ports = CTL_DEFAULT_MAX_PORTS;
437 TUNABLE_INT("kern.cam.ctl.max_ports", &ctl_max_ports);
438 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, max_ports, CTLFLAG_RDTUN,
439     &ctl_max_ports, CTL_DEFAULT_MAX_LUNS, "Maximum number of ports");
440 
441 /*
442  * Maximum number of initiators we support.
443  */
444 #define	CTL_MAX_INITIATORS	(CTL_MAX_INIT_PER_PORT * ctl_max_ports)
445 
446 /*
447  * Supported pages (0x00), Serial number (0x80), Device ID (0x83),
448  * Extended INQUIRY Data (0x86), Mode Page Policy (0x87),
449  * SCSI Ports (0x88), Third-party Copy (0x8F), SCSI Feature Sets (0x92),
450  * Block limits (0xB0), Block Device Characteristics (0xB1) and
451  * Logical Block Provisioning (0xB2)
452  */
453 #define SCSI_EVPD_NUM_SUPPORTED_PAGES	11
454 
455 static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
456 				  int param);
457 static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
458 static void ctl_copy_sense_data_back(union ctl_io *src, union ctl_ha_msg *dest);
459 static int ctl_init(void);
460 static int ctl_shutdown(void);
461 static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
462 static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
463 static void ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio);
464 static void ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
465 			      struct ctl_ooa *ooa_hdr,
466 			      struct ctl_ooa_entry *kern_entries);
467 static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
468 		     struct thread *td);
469 static int ctl_enable_lun(struct ctl_lun *lun);
470 static int ctl_disable_lun(struct ctl_lun *lun);
471 static int ctl_free_lun(struct ctl_lun *lun);
472 
473 static int ctl_do_mode_select(union ctl_io *io);
474 static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
475 			   uint64_t res_key, uint64_t sa_res_key,
476 			   uint8_t type, uint32_t residx,
477 			   struct ctl_scsiio *ctsio,
478 			   struct scsi_per_res_out *cdb,
479 			   struct scsi_per_res_out_parms* param);
480 static void ctl_pro_preempt_other(struct ctl_lun *lun,
481 				  union ctl_ha_msg *msg);
482 static void ctl_hndl_per_res_out_on_other_sc(union ctl_io *io);
483 static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
484 static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
485 static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
486 static int ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len);
487 static int ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len);
488 static int ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio,
489 					 int alloc_len);
490 static int ctl_inquiry_evpd_sfs(struct ctl_scsiio *ctsio, int alloc_len);
491 static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
492 					 int alloc_len);
493 static int ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len);
494 static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
495 static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
496 static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
497 static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len);
498 static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2,
499     bool seq);
500 static ctl_action ctl_seq_check(union ctl_io *io1, union ctl_io *io2);
501 static ctl_action ctl_check_for_blockage(struct ctl_lun *lun,
502     union ctl_io *pending_io, const uint8_t *serialize_row,
503     union ctl_io *ooa_io);
504 static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
505 				union ctl_io **starting_io);
506 static void ctl_try_unblock_io(struct ctl_lun *lun, union ctl_io *io,
507     bool skip);
508 static void ctl_try_unblock_others(struct ctl_lun *lun, union ctl_io *io,
509     bool skip);
510 static int ctl_scsiio_lun_check(struct ctl_lun *lun,
511 				const struct ctl_cmd_entry *entry,
512 				struct ctl_scsiio *ctsio);
513 static void ctl_failover_lun(union ctl_io *io);
514 static void ctl_scsiio_precheck(struct ctl_scsiio *ctsio);
515 static int ctl_scsiio(struct ctl_scsiio *ctsio);
516 
517 static int ctl_target_reset(union ctl_io *io);
518 static void ctl_do_lun_reset(struct ctl_lun *lun, uint32_t initidx,
519 			 ctl_ua_type ua_type);
520 static int ctl_lun_reset(union ctl_io *io);
521 static int ctl_abort_task(union ctl_io *io);
522 static int ctl_abort_task_set(union ctl_io *io);
523 static int ctl_query_task(union ctl_io *io, int task_set);
524 static void ctl_i_t_nexus_loss(struct ctl_softc *softc, uint32_t initidx,
525 			      ctl_ua_type ua_type);
526 static int ctl_i_t_nexus_reset(union ctl_io *io);
527 static int ctl_query_async_event(union ctl_io *io);
528 static void ctl_run_task(union ctl_io *io);
529 #ifdef CTL_IO_DELAY
530 static void ctl_datamove_timer_wakeup(void *arg);
531 static void ctl_done_timer_wakeup(void *arg);
532 #endif /* CTL_IO_DELAY */
533 
534 static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
535 static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
536 static int ctl_datamove_remote_dm_write_cb(union ctl_io *io, bool samethr);
537 static void ctl_datamove_remote_write(union ctl_io *io);
538 static int ctl_datamove_remote_dm_read_cb(union ctl_io *io, bool samethr);
539 static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
540 static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
541 static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
542 				    ctl_ha_dt_cb callback);
543 static void ctl_datamove_remote_read(union ctl_io *io);
544 static void ctl_datamove_remote(union ctl_io *io);
545 static void ctl_process_done(union ctl_io *io);
546 static void ctl_thresh_thread(void *arg);
547 static void ctl_work_thread(void *arg);
548 static void ctl_enqueue_incoming(union ctl_io *io);
549 static void ctl_enqueue_rtr(union ctl_io *io);
550 static void ctl_enqueue_done(union ctl_io *io);
551 static void ctl_enqueue_isc(union ctl_io *io);
552 static const struct ctl_cmd_entry *
553     ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa);
554 static const struct ctl_cmd_entry *
555     ctl_validate_command(struct ctl_scsiio *ctsio);
556 static int ctl_cmd_applicable(uint8_t lun_type,
557     const struct ctl_cmd_entry *entry);
558 static int ctl_ha_init(void);
559 static int ctl_ha_shutdown(void);
560 
561 static uint64_t ctl_get_prkey(struct ctl_lun *lun, uint32_t residx);
562 static void ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx);
563 static void ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx);
564 static void ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key);
565 
566 /*
567  * Load the serialization table.  This isn't very pretty, but is probably
568  * the easiest way to do it.
569  */
570 #include "ctl_ser_table.c"
571 
572 /*
573  * We only need to define open, close and ioctl routines for this driver.
574  */
575 static struct cdevsw ctl_cdevsw = {
576 	.d_version =	D_VERSION,
577 	.d_flags =	0,
578 	.d_open =	ctl_open,
579 	.d_close =	ctl_close,
580 	.d_ioctl =	ctl_ioctl,
581 	.d_name =	"ctl",
582 };
583 
584 MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
585 
586 static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
587 
588 static moduledata_t ctl_moduledata = {
589 	"ctl",
590 	ctl_module_event_handler,
591 	NULL
592 };
593 
594 DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
595 MODULE_VERSION(ctl, 1);
596 
597 static struct ctl_frontend ha_frontend =
598 {
599 	.name = "ha",
600 	.init = ctl_ha_init,
601 	.shutdown = ctl_ha_shutdown,
602 };
603 
604 static int
605 ctl_ha_init(void)
606 {
607 	struct ctl_softc *softc = control_softc;
608 
609 	if (ctl_pool_create(softc, "othersc", CTL_POOL_ENTRIES_OTHER_SC,
610 	                    &softc->othersc_pool) != 0)
611 		return (ENOMEM);
612 	if (ctl_ha_msg_init(softc) != CTL_HA_STATUS_SUCCESS) {
613 		ctl_pool_free(softc->othersc_pool);
614 		return (EIO);
615 	}
616 	if (ctl_ha_msg_register(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
617 	    != CTL_HA_STATUS_SUCCESS) {
618 		ctl_ha_msg_destroy(softc);
619 		ctl_pool_free(softc->othersc_pool);
620 		return (EIO);
621 	}
622 	return (0);
623 };
624 
625 static int
626 ctl_ha_shutdown(void)
627 {
628 	struct ctl_softc *softc = control_softc;
629 	struct ctl_port *port;
630 
631 	ctl_ha_msg_shutdown(softc);
632 	if (ctl_ha_msg_deregister(CTL_HA_CHAN_CTL) != CTL_HA_STATUS_SUCCESS)
633 		return (EIO);
634 	if (ctl_ha_msg_destroy(softc) != CTL_HA_STATUS_SUCCESS)
635 		return (EIO);
636 	ctl_pool_free(softc->othersc_pool);
637 	while ((port = STAILQ_FIRST(&ha_frontend.port_list)) != NULL) {
638 		ctl_port_deregister(port);
639 		free(port->port_name, M_CTL);
640 		free(port, M_CTL);
641 	}
642 	return (0);
643 };
644 
645 static void
646 ctl_ha_datamove(union ctl_io *io)
647 {
648 	struct ctl_lun *lun = CTL_LUN(io);
649 	struct ctl_sg_entry *sgl;
650 	union ctl_ha_msg msg;
651 	uint32_t sg_entries_sent;
652 	int do_sg_copy, i, j;
653 
654 	memset(&msg.dt, 0, sizeof(msg.dt));
655 	msg.hdr.msg_type = CTL_MSG_DATAMOVE;
656 	msg.hdr.original_sc = io->io_hdr.remote_io;
657 	msg.hdr.serializing_sc = io;
658 	msg.hdr.nexus = io->io_hdr.nexus;
659 	msg.hdr.status = io->io_hdr.status;
660 	msg.dt.flags = io->io_hdr.flags;
661 
662 	/*
663 	 * We convert everything into a S/G list here.  We can't
664 	 * pass by reference, only by value between controllers.
665 	 * So we can't pass a pointer to the S/G list, only as many
666 	 * S/G entries as we can fit in here.  If it's possible for
667 	 * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
668 	 * then we need to break this up into multiple transfers.
669 	 */
670 	if (io->scsiio.kern_sg_entries == 0) {
671 		msg.dt.kern_sg_entries = 1;
672 #if 0
673 		if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
674 			msg.dt.sg_list[0].addr = io->scsiio.kern_data_ptr;
675 		} else {
676 			/* XXX KDM use busdma here! */
677 			msg.dt.sg_list[0].addr =
678 			    (void *)vtophys(io->scsiio.kern_data_ptr);
679 		}
680 #else
681 		KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0,
682 		    ("HA does not support BUS_ADDR"));
683 		msg.dt.sg_list[0].addr = io->scsiio.kern_data_ptr;
684 #endif
685 		msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
686 		do_sg_copy = 0;
687 	} else {
688 		msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
689 		do_sg_copy = 1;
690 	}
691 
692 	msg.dt.kern_data_len = io->scsiio.kern_data_len;
693 	msg.dt.kern_total_len = io->scsiio.kern_total_len;
694 	msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
695 	msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
696 	msg.dt.sg_sequence = 0;
697 
698 	/*
699 	 * Loop until we've sent all of the S/G entries.  On the
700 	 * other end, we'll recompose these S/G entries into one
701 	 * contiguous list before processing.
702 	 */
703 	for (sg_entries_sent = 0; sg_entries_sent < msg.dt.kern_sg_entries;
704 	    msg.dt.sg_sequence++) {
705 		msg.dt.cur_sg_entries = MIN((sizeof(msg.dt.sg_list) /
706 		    sizeof(msg.dt.sg_list[0])),
707 		    msg.dt.kern_sg_entries - sg_entries_sent);
708 		if (do_sg_copy != 0) {
709 			sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
710 			for (i = sg_entries_sent, j = 0;
711 			     i < msg.dt.cur_sg_entries; i++, j++) {
712 #if 0
713 				if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
714 					msg.dt.sg_list[j].addr = sgl[i].addr;
715 				} else {
716 					/* XXX KDM use busdma here! */
717 					msg.dt.sg_list[j].addr =
718 					    (void *)vtophys(sgl[i].addr);
719 				}
720 #else
721 				KASSERT((io->io_hdr.flags &
722 				    CTL_FLAG_BUS_ADDR) == 0,
723 				    ("HA does not support BUS_ADDR"));
724 				msg.dt.sg_list[j].addr = sgl[i].addr;
725 #endif
726 				msg.dt.sg_list[j].len = sgl[i].len;
727 			}
728 		}
729 
730 		sg_entries_sent += msg.dt.cur_sg_entries;
731 		msg.dt.sg_last = (sg_entries_sent >= msg.dt.kern_sg_entries);
732 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
733 		    sizeof(msg.dt) - sizeof(msg.dt.sg_list) +
734 		    sizeof(struct ctl_sg_entry) * msg.dt.cur_sg_entries,
735 		    M_WAITOK) > CTL_HA_STATUS_SUCCESS) {
736 			io->io_hdr.port_status = 31341;
737 			ctl_datamove_done(io, true);
738 			return;
739 		}
740 		msg.dt.sent_sg_entries = sg_entries_sent;
741 	}
742 
743 	/*
744 	 * Officially handover the request from us to peer.
745 	 * If failover has just happened, then we must return error.
746 	 * If failover happen just after, then it is not our problem.
747 	 */
748 	if (lun)
749 		mtx_lock(&lun->lun_lock);
750 	if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
751 		if (lun)
752 			mtx_unlock(&lun->lun_lock);
753 		io->io_hdr.port_status = 31342;
754 		ctl_datamove_done(io, true);
755 		return;
756 	}
757 	io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
758 	io->io_hdr.flags |= CTL_FLAG_DMA_INPROG;
759 	if (lun)
760 		mtx_unlock(&lun->lun_lock);
761 }
762 
763 static void
764 ctl_ha_done(union ctl_io *io)
765 {
766 	union ctl_ha_msg msg;
767 
768 	if (io->io_hdr.io_type == CTL_IO_SCSI) {
769 		memset(&msg, 0, sizeof(msg));
770 		msg.hdr.msg_type = CTL_MSG_FINISH_IO;
771 		msg.hdr.original_sc = io->io_hdr.remote_io;
772 		msg.hdr.nexus = io->io_hdr.nexus;
773 		msg.hdr.status = io->io_hdr.status;
774 		msg.scsi.scsi_status = io->scsiio.scsi_status;
775 		msg.scsi.tag_num = io->scsiio.tag_num;
776 		msg.scsi.tag_type = io->scsiio.tag_type;
777 		msg.scsi.sense_len = io->scsiio.sense_len;
778 		memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
779 		    io->scsiio.sense_len);
780 		ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
781 		    sizeof(msg.scsi) - sizeof(msg.scsi.sense_data) +
782 		    msg.scsi.sense_len, M_WAITOK);
783 	}
784 	ctl_free_io(io);
785 }
786 
787 static void
788 ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
789 			    union ctl_ha_msg *msg_info)
790 {
791 	struct ctl_scsiio *ctsio;
792 
793 	if (msg_info->hdr.original_sc == NULL) {
794 		printf("%s: original_sc == NULL!\n", __func__);
795 		/* XXX KDM now what? */
796 		return;
797 	}
798 
799 	ctsio = &msg_info->hdr.original_sc->scsiio;
800 	ctsio->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
801 	ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
802 	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
803 	ctsio->io_hdr.status = msg_info->hdr.status;
804 	ctsio->scsi_status = msg_info->scsi.scsi_status;
805 	ctsio->sense_len = msg_info->scsi.sense_len;
806 	memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
807 	       msg_info->scsi.sense_len);
808 	ctl_enqueue_isc((union ctl_io *)ctsio);
809 }
810 
811 static void
812 ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
813 				union ctl_ha_msg *msg_info)
814 {
815 	struct ctl_scsiio *ctsio;
816 
817 	if (msg_info->hdr.serializing_sc == NULL) {
818 		printf("%s: serializing_sc == NULL!\n", __func__);
819 		/* XXX KDM now what? */
820 		return;
821 	}
822 
823 	ctsio = &msg_info->hdr.serializing_sc->scsiio;
824 	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
825 	ctl_enqueue_isc((union ctl_io *)ctsio);
826 }
827 
828 void
829 ctl_isc_announce_lun(struct ctl_lun *lun)
830 {
831 	struct ctl_softc *softc = lun->ctl_softc;
832 	union ctl_ha_msg *msg;
833 	struct ctl_ha_msg_lun_pr_key pr_key;
834 	int i, k;
835 
836 	if (softc->ha_link != CTL_HA_LINK_ONLINE)
837 		return;
838 	mtx_lock(&lun->lun_lock);
839 	i = sizeof(msg->lun);
840 	if (lun->lun_devid)
841 		i += lun->lun_devid->len;
842 	i += sizeof(pr_key) * lun->pr_key_count;
843 alloc:
844 	mtx_unlock(&lun->lun_lock);
845 	msg = malloc(i, M_CTL, M_WAITOK);
846 	mtx_lock(&lun->lun_lock);
847 	k = sizeof(msg->lun);
848 	if (lun->lun_devid)
849 		k += lun->lun_devid->len;
850 	k += sizeof(pr_key) * lun->pr_key_count;
851 	if (i < k) {
852 		free(msg, M_CTL);
853 		i = k;
854 		goto alloc;
855 	}
856 	bzero(&msg->lun, sizeof(msg->lun));
857 	msg->hdr.msg_type = CTL_MSG_LUN_SYNC;
858 	msg->hdr.nexus.targ_lun = lun->lun;
859 	msg->hdr.nexus.targ_mapped_lun = lun->lun;
860 	msg->lun.flags = lun->flags;
861 	msg->lun.pr_generation = lun->pr_generation;
862 	msg->lun.pr_res_idx = lun->pr_res_idx;
863 	msg->lun.pr_res_type = lun->pr_res_type;
864 	msg->lun.pr_key_count = lun->pr_key_count;
865 	i = 0;
866 	if (lun->lun_devid) {
867 		msg->lun.lun_devid_len = lun->lun_devid->len;
868 		memcpy(&msg->lun.data[i], lun->lun_devid->data,
869 		    msg->lun.lun_devid_len);
870 		i += msg->lun.lun_devid_len;
871 	}
872 	for (k = 0; k < CTL_MAX_INITIATORS; k++) {
873 		if ((pr_key.pr_key = ctl_get_prkey(lun, k)) == 0)
874 			continue;
875 		pr_key.pr_iid = k;
876 		memcpy(&msg->lun.data[i], &pr_key, sizeof(pr_key));
877 		i += sizeof(pr_key);
878 	}
879 	mtx_unlock(&lun->lun_lock);
880 	ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->lun, sizeof(msg->lun) + i,
881 	    M_WAITOK);
882 	free(msg, M_CTL);
883 
884 	if (lun->flags & CTL_LUN_PRIMARY_SC) {
885 		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
886 			ctl_isc_announce_mode(lun, -1,
887 			    lun->mode_pages.index[i].page_code & SMPH_PC_MASK,
888 			    lun->mode_pages.index[i].subpage);
889 		}
890 	}
891 }
892 
893 void
894 ctl_isc_announce_port(struct ctl_port *port)
895 {
896 	struct ctl_softc *softc = port->ctl_softc;
897 	union ctl_ha_msg *msg;
898 	int i;
899 
900 	if (port->targ_port < softc->port_min ||
901 	    port->targ_port >= softc->port_max ||
902 	    softc->ha_link != CTL_HA_LINK_ONLINE)
903 		return;
904 	i = sizeof(msg->port) + strlen(port->port_name) + 1;
905 	if (port->lun_map)
906 		i += port->lun_map_size * sizeof(uint32_t);
907 	if (port->port_devid)
908 		i += port->port_devid->len;
909 	if (port->target_devid)
910 		i += port->target_devid->len;
911 	if (port->init_devid)
912 		i += port->init_devid->len;
913 	msg = malloc(i, M_CTL, M_WAITOK);
914 	bzero(&msg->port, sizeof(msg->port));
915 	msg->hdr.msg_type = CTL_MSG_PORT_SYNC;
916 	msg->hdr.nexus.targ_port = port->targ_port;
917 	msg->port.port_type = port->port_type;
918 	msg->port.physical_port = port->physical_port;
919 	msg->port.virtual_port = port->virtual_port;
920 	msg->port.status = port->status;
921 	i = 0;
922 	msg->port.name_len = sprintf(&msg->port.data[i],
923 	    "%d:%s", softc->ha_id, port->port_name) + 1;
924 	i += msg->port.name_len;
925 	if (port->lun_map) {
926 		msg->port.lun_map_len = port->lun_map_size * sizeof(uint32_t);
927 		memcpy(&msg->port.data[i], port->lun_map,
928 		    msg->port.lun_map_len);
929 		i += msg->port.lun_map_len;
930 	}
931 	if (port->port_devid) {
932 		msg->port.port_devid_len = port->port_devid->len;
933 		memcpy(&msg->port.data[i], port->port_devid->data,
934 		    msg->port.port_devid_len);
935 		i += msg->port.port_devid_len;
936 	}
937 	if (port->target_devid) {
938 		msg->port.target_devid_len = port->target_devid->len;
939 		memcpy(&msg->port.data[i], port->target_devid->data,
940 		    msg->port.target_devid_len);
941 		i += msg->port.target_devid_len;
942 	}
943 	if (port->init_devid) {
944 		msg->port.init_devid_len = port->init_devid->len;
945 		memcpy(&msg->port.data[i], port->init_devid->data,
946 		    msg->port.init_devid_len);
947 		i += msg->port.init_devid_len;
948 	}
949 	ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->port, sizeof(msg->port) + i,
950 	    M_WAITOK);
951 	free(msg, M_CTL);
952 }
953 
954 void
955 ctl_isc_announce_iid(struct ctl_port *port, int iid)
956 {
957 	struct ctl_softc *softc = port->ctl_softc;
958 	union ctl_ha_msg *msg;
959 	int i, l;
960 
961 	if (port->targ_port < softc->port_min ||
962 	    port->targ_port >= softc->port_max ||
963 	    softc->ha_link != CTL_HA_LINK_ONLINE)
964 		return;
965 	mtx_lock(&softc->ctl_lock);
966 	i = sizeof(msg->iid);
967 	l = 0;
968 	if (port->wwpn_iid[iid].name)
969 		l = strlen(port->wwpn_iid[iid].name) + 1;
970 	i += l;
971 	msg = malloc(i, M_CTL, M_NOWAIT);
972 	if (msg == NULL) {
973 		mtx_unlock(&softc->ctl_lock);
974 		return;
975 	}
976 	bzero(&msg->iid, sizeof(msg->iid));
977 	msg->hdr.msg_type = CTL_MSG_IID_SYNC;
978 	msg->hdr.nexus.targ_port = port->targ_port;
979 	msg->hdr.nexus.initid = iid;
980 	msg->iid.in_use = port->wwpn_iid[iid].in_use;
981 	msg->iid.name_len = l;
982 	msg->iid.wwpn = port->wwpn_iid[iid].wwpn;
983 	if (port->wwpn_iid[iid].name)
984 		strlcpy(msg->iid.data, port->wwpn_iid[iid].name, l);
985 	mtx_unlock(&softc->ctl_lock);
986 	ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->iid, i, M_NOWAIT);
987 	free(msg, M_CTL);
988 }
989 
990 void
991 ctl_isc_announce_mode(struct ctl_lun *lun, uint32_t initidx,
992     uint8_t page, uint8_t subpage)
993 {
994 	struct ctl_softc *softc = lun->ctl_softc;
995 	union ctl_ha_msg *msg;
996 	u_int i, l;
997 
998 	if (softc->ha_link != CTL_HA_LINK_ONLINE)
999 		return;
1000 	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
1001 		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) ==
1002 		    page && lun->mode_pages.index[i].subpage == subpage)
1003 			break;
1004 	}
1005 	if (i == CTL_NUM_MODE_PAGES)
1006 		return;
1007 
1008 	/* Don't try to replicate pages not present on this device. */
1009 	if (lun->mode_pages.index[i].page_data == NULL)
1010 		return;
1011 
1012 	l = sizeof(msg->mode) + lun->mode_pages.index[i].page_len;
1013 	msg = malloc(l, M_CTL, M_WAITOK | M_ZERO);
1014 	msg->hdr.msg_type = CTL_MSG_MODE_SYNC;
1015 	msg->hdr.nexus.targ_port = initidx / CTL_MAX_INIT_PER_PORT;
1016 	msg->hdr.nexus.initid = initidx % CTL_MAX_INIT_PER_PORT;
1017 	msg->hdr.nexus.targ_lun = lun->lun;
1018 	msg->hdr.nexus.targ_mapped_lun = lun->lun;
1019 	msg->mode.page_code = page;
1020 	msg->mode.subpage = subpage;
1021 	msg->mode.page_len = lun->mode_pages.index[i].page_len;
1022 	memcpy(msg->mode.data, lun->mode_pages.index[i].page_data,
1023 	    msg->mode.page_len);
1024 	ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->mode, l, M_WAITOK);
1025 	free(msg, M_CTL);
1026 }
1027 
1028 static void
1029 ctl_isc_ha_link_up(struct ctl_softc *softc)
1030 {
1031 	struct ctl_port *port;
1032 	struct ctl_lun *lun;
1033 	union ctl_ha_msg msg;
1034 	int i;
1035 
1036 	/* Announce this node parameters to peer for validation. */
1037 	msg.login.msg_type = CTL_MSG_LOGIN;
1038 	msg.login.version = CTL_HA_VERSION;
1039 	msg.login.ha_mode = softc->ha_mode;
1040 	msg.login.ha_id = softc->ha_id;
1041 	msg.login.max_luns = ctl_max_luns;
1042 	msg.login.max_ports = ctl_max_ports;
1043 	msg.login.max_init_per_port = CTL_MAX_INIT_PER_PORT;
1044 	ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg.login, sizeof(msg.login),
1045 	    M_WAITOK);
1046 
1047 	STAILQ_FOREACH(port, &softc->port_list, links) {
1048 		ctl_isc_announce_port(port);
1049 		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1050 			if (port->wwpn_iid[i].in_use)
1051 				ctl_isc_announce_iid(port, i);
1052 		}
1053 	}
1054 	STAILQ_FOREACH(lun, &softc->lun_list, links)
1055 		ctl_isc_announce_lun(lun);
1056 }
1057 
1058 static void
1059 ctl_isc_ha_link_down(struct ctl_softc *softc)
1060 {
1061 	struct ctl_port *port;
1062 	struct ctl_lun *lun;
1063 	union ctl_io *io;
1064 	int i;
1065 
1066 	mtx_lock(&softc->ctl_lock);
1067 	STAILQ_FOREACH(lun, &softc->lun_list, links) {
1068 		mtx_lock(&lun->lun_lock);
1069 		if (lun->flags & CTL_LUN_PEER_SC_PRIMARY) {
1070 			lun->flags &= ~CTL_LUN_PEER_SC_PRIMARY;
1071 			ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
1072 		}
1073 		mtx_unlock(&lun->lun_lock);
1074 
1075 		mtx_unlock(&softc->ctl_lock);
1076 		io = ctl_alloc_io(softc->othersc_pool);
1077 		mtx_lock(&softc->ctl_lock);
1078 		ctl_zero_io(io);
1079 		io->io_hdr.msg_type = CTL_MSG_FAILOVER;
1080 		io->io_hdr.nexus.targ_mapped_lun = lun->lun;
1081 		ctl_enqueue_isc(io);
1082 	}
1083 
1084 	STAILQ_FOREACH(port, &softc->port_list, links) {
1085 		if (port->targ_port >= softc->port_min &&
1086 		    port->targ_port < softc->port_max)
1087 			continue;
1088 		port->status &= ~CTL_PORT_STATUS_ONLINE;
1089 		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1090 			port->wwpn_iid[i].in_use = 0;
1091 			free(port->wwpn_iid[i].name, M_CTL);
1092 			port->wwpn_iid[i].name = NULL;
1093 		}
1094 	}
1095 	mtx_unlock(&softc->ctl_lock);
1096 }
1097 
1098 static void
1099 ctl_isc_ua(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1100 {
1101 	struct ctl_lun *lun;
1102 	uint32_t iid;
1103 
1104 	if (len < sizeof(msg->ua)) {
1105 		printf("%s: Received truncated message %d < %zu\n",
1106 		    __func__, len, sizeof(msg->ua));
1107 		ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1108 		return;
1109 	}
1110 
1111 	mtx_lock(&softc->ctl_lock);
1112 	if (msg->hdr.nexus.targ_mapped_lun >= ctl_max_luns ||
1113 	    (lun = softc->ctl_luns[msg->hdr.nexus.targ_mapped_lun]) == NULL) {
1114 		mtx_unlock(&softc->ctl_lock);
1115 		return;
1116 	}
1117 	mtx_lock(&lun->lun_lock);
1118 	mtx_unlock(&softc->ctl_lock);
1119 	if (msg->ua.ua_type == CTL_UA_THIN_PROV_THRES && msg->ua.ua_set)
1120 		memcpy(lun->ua_tpt_info, msg->ua.ua_info, 8);
1121 	iid = ctl_get_initindex(&msg->hdr.nexus);
1122 	if (msg->ua.ua_all) {
1123 		if (msg->ua.ua_set)
1124 			ctl_est_ua_all(lun, iid, msg->ua.ua_type);
1125 		else
1126 			ctl_clr_ua_all(lun, iid, msg->ua.ua_type);
1127 	} else {
1128 		if (msg->ua.ua_set)
1129 			ctl_est_ua(lun, iid, msg->ua.ua_type);
1130 		else
1131 			ctl_clr_ua(lun, iid, msg->ua.ua_type);
1132 	}
1133 	mtx_unlock(&lun->lun_lock);
1134 }
1135 
1136 static void
1137 ctl_isc_lun_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1138 {
1139 	struct ctl_lun *lun;
1140 	struct ctl_ha_msg_lun_pr_key pr_key;
1141 	int i, k;
1142 	ctl_lun_flags oflags;
1143 	uint32_t targ_lun;
1144 
1145 	if (len < offsetof(struct ctl_ha_msg_lun, data[0])) {
1146 		printf("%s: Received truncated message %d < %zu\n",
1147 		    __func__, len, offsetof(struct ctl_ha_msg_lun, data[0]));
1148 		ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1149 		return;
1150 	}
1151 	i = msg->lun.lun_devid_len + msg->lun.pr_key_count * sizeof(pr_key);
1152 	if (len < offsetof(struct ctl_ha_msg_lun, data[i])) {
1153 		printf("%s: Received truncated message data %d < %zu\n",
1154 		    __func__, len, offsetof(struct ctl_ha_msg_lun, data[i]));
1155 		ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1156 		return;
1157 	}
1158 
1159 	targ_lun = msg->hdr.nexus.targ_mapped_lun;
1160 	mtx_lock(&softc->ctl_lock);
1161 	if (targ_lun >= ctl_max_luns ||
1162 	    (lun = softc->ctl_luns[targ_lun]) == NULL) {
1163 		mtx_unlock(&softc->ctl_lock);
1164 		return;
1165 	}
1166 	mtx_lock(&lun->lun_lock);
1167 	mtx_unlock(&softc->ctl_lock);
1168 	if (lun->flags & CTL_LUN_DISABLED) {
1169 		mtx_unlock(&lun->lun_lock);
1170 		return;
1171 	}
1172 	i = (lun->lun_devid != NULL) ? lun->lun_devid->len : 0;
1173 	if (msg->lun.lun_devid_len != i || (i > 0 &&
1174 	    memcmp(&msg->lun.data[0], lun->lun_devid->data, i) != 0)) {
1175 		mtx_unlock(&lun->lun_lock);
1176 		printf("%s: Received conflicting HA LUN %d\n",
1177 		    __func__, targ_lun);
1178 		return;
1179 	} else {
1180 		/* Record whether peer is primary. */
1181 		oflags = lun->flags;
1182 		if ((msg->lun.flags & CTL_LUN_PRIMARY_SC) &&
1183 		    (msg->lun.flags & CTL_LUN_DISABLED) == 0)
1184 			lun->flags |= CTL_LUN_PEER_SC_PRIMARY;
1185 		else
1186 			lun->flags &= ~CTL_LUN_PEER_SC_PRIMARY;
1187 		if (oflags != lun->flags)
1188 			ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
1189 
1190 		/* If peer is primary and we are not -- use data */
1191 		if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
1192 		    (lun->flags & CTL_LUN_PEER_SC_PRIMARY)) {
1193 			lun->pr_generation = msg->lun.pr_generation;
1194 			lun->pr_res_idx = msg->lun.pr_res_idx;
1195 			lun->pr_res_type = msg->lun.pr_res_type;
1196 			lun->pr_key_count = msg->lun.pr_key_count;
1197 			for (k = 0; k < CTL_MAX_INITIATORS; k++)
1198 				ctl_clr_prkey(lun, k);
1199 			for (k = 0; k < msg->lun.pr_key_count; k++) {
1200 				memcpy(&pr_key, &msg->lun.data[i],
1201 				    sizeof(pr_key));
1202 				ctl_alloc_prkey(lun, pr_key.pr_iid);
1203 				ctl_set_prkey(lun, pr_key.pr_iid,
1204 				    pr_key.pr_key);
1205 				i += sizeof(pr_key);
1206 			}
1207 		}
1208 
1209 		mtx_unlock(&lun->lun_lock);
1210 		CTL_DEBUG_PRINT(("%s: Known LUN %d, peer is %s\n",
1211 		    __func__, targ_lun,
1212 		    (msg->lun.flags & CTL_LUN_PRIMARY_SC) ?
1213 		    "primary" : "secondary"));
1214 
1215 		/* If we are primary but peer doesn't know -- notify */
1216 		if ((lun->flags & CTL_LUN_PRIMARY_SC) &&
1217 		    (msg->lun.flags & CTL_LUN_PEER_SC_PRIMARY) == 0)
1218 			ctl_isc_announce_lun(lun);
1219 	}
1220 }
1221 
1222 static void
1223 ctl_isc_port_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1224 {
1225 	struct ctl_port *port;
1226 	struct ctl_lun *lun;
1227 	int i, new;
1228 
1229 	if (len < offsetof(struct ctl_ha_msg_port, data[0])) {
1230 		printf("%s: Received truncated message %d < %zu\n",
1231 		    __func__, len, offsetof(struct ctl_ha_msg_port, data[0]));
1232 		ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1233 		return;
1234 	}
1235 	i = msg->port.name_len + msg->port.lun_map_len +
1236 	    msg->port.port_devid_len + msg->port.target_devid_len +
1237 	    msg->port.init_devid_len;
1238 	if (len < offsetof(struct ctl_ha_msg_port, data[i])) {
1239 		printf("%s: Received truncated message data %d < %zu\n",
1240 		    __func__, len, offsetof(struct ctl_ha_msg_port, data[i]));
1241 		ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1242 		return;
1243 	}
1244 
1245 	port = softc->ctl_ports[msg->hdr.nexus.targ_port];
1246 	if (port == NULL) {
1247 		CTL_DEBUG_PRINT(("%s: New port %d\n", __func__,
1248 		    msg->hdr.nexus.targ_port));
1249 		new = 1;
1250 		port = malloc(sizeof(*port), M_CTL, M_WAITOK | M_ZERO);
1251 		port->frontend = &ha_frontend;
1252 		port->targ_port = msg->hdr.nexus.targ_port;
1253 		port->fe_datamove = ctl_ha_datamove;
1254 		port->fe_done = ctl_ha_done;
1255 	} else if (port->frontend == &ha_frontend) {
1256 		CTL_DEBUG_PRINT(("%s: Updated port %d\n", __func__,
1257 		    msg->hdr.nexus.targ_port));
1258 		new = 0;
1259 	} else {
1260 		printf("%s: Received conflicting HA port %d\n",
1261 		    __func__, msg->hdr.nexus.targ_port);
1262 		return;
1263 	}
1264 	port->port_type = msg->port.port_type;
1265 	port->physical_port = msg->port.physical_port;
1266 	port->virtual_port = msg->port.virtual_port;
1267 	port->status = msg->port.status;
1268 	i = 0;
1269 	free(port->port_name, M_CTL);
1270 	port->port_name = strndup(&msg->port.data[i], msg->port.name_len,
1271 	    M_CTL);
1272 	i += msg->port.name_len;
1273 	if (msg->port.lun_map_len != 0) {
1274 		if (port->lun_map == NULL ||
1275 		    port->lun_map_size * sizeof(uint32_t) <
1276 		    msg->port.lun_map_len) {
1277 			port->lun_map_size = 0;
1278 			free(port->lun_map, M_CTL);
1279 			port->lun_map = malloc(msg->port.lun_map_len,
1280 			    M_CTL, M_WAITOK);
1281 		}
1282 		memcpy(port->lun_map, &msg->port.data[i], msg->port.lun_map_len);
1283 		port->lun_map_size = msg->port.lun_map_len / sizeof(uint32_t);
1284 		i += msg->port.lun_map_len;
1285 	} else {
1286 		port->lun_map_size = 0;
1287 		free(port->lun_map, M_CTL);
1288 		port->lun_map = NULL;
1289 	}
1290 	if (msg->port.port_devid_len != 0) {
1291 		if (port->port_devid == NULL ||
1292 		    port->port_devid->len < msg->port.port_devid_len) {
1293 			free(port->port_devid, M_CTL);
1294 			port->port_devid = malloc(sizeof(struct ctl_devid) +
1295 			    msg->port.port_devid_len, M_CTL, M_WAITOK);
1296 		}
1297 		memcpy(port->port_devid->data, &msg->port.data[i],
1298 		    msg->port.port_devid_len);
1299 		port->port_devid->len = msg->port.port_devid_len;
1300 		i += msg->port.port_devid_len;
1301 	} else {
1302 		free(port->port_devid, M_CTL);
1303 		port->port_devid = NULL;
1304 	}
1305 	if (msg->port.target_devid_len != 0) {
1306 		if (port->target_devid == NULL ||
1307 		    port->target_devid->len < msg->port.target_devid_len) {
1308 			free(port->target_devid, M_CTL);
1309 			port->target_devid = malloc(sizeof(struct ctl_devid) +
1310 			    msg->port.target_devid_len, M_CTL, M_WAITOK);
1311 		}
1312 		memcpy(port->target_devid->data, &msg->port.data[i],
1313 		    msg->port.target_devid_len);
1314 		port->target_devid->len = msg->port.target_devid_len;
1315 		i += msg->port.target_devid_len;
1316 	} else {
1317 		free(port->target_devid, M_CTL);
1318 		port->target_devid = NULL;
1319 	}
1320 	if (msg->port.init_devid_len != 0) {
1321 		if (port->init_devid == NULL ||
1322 		    port->init_devid->len < msg->port.init_devid_len) {
1323 			free(port->init_devid, M_CTL);
1324 			port->init_devid = malloc(sizeof(struct ctl_devid) +
1325 			    msg->port.init_devid_len, M_CTL, M_WAITOK);
1326 		}
1327 		memcpy(port->init_devid->data, &msg->port.data[i],
1328 		    msg->port.init_devid_len);
1329 		port->init_devid->len = msg->port.init_devid_len;
1330 		i += msg->port.init_devid_len;
1331 	} else {
1332 		free(port->init_devid, M_CTL);
1333 		port->init_devid = NULL;
1334 	}
1335 	if (new) {
1336 		if (ctl_port_register(port) != 0) {
1337 			printf("%s: ctl_port_register() failed with error\n",
1338 			    __func__);
1339 		}
1340 	}
1341 	mtx_lock(&softc->ctl_lock);
1342 	STAILQ_FOREACH(lun, &softc->lun_list, links) {
1343 		if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
1344 			continue;
1345 		mtx_lock(&lun->lun_lock);
1346 		ctl_est_ua_all(lun, -1, CTL_UA_INQ_CHANGE);
1347 		mtx_unlock(&lun->lun_lock);
1348 	}
1349 	mtx_unlock(&softc->ctl_lock);
1350 }
1351 
1352 static void
1353 ctl_isc_iid_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1354 {
1355 	struct ctl_port *port;
1356 	int i, iid;
1357 
1358 	if (len < offsetof(struct ctl_ha_msg_iid, data[0])) {
1359 		printf("%s: Received truncated message %d < %zu\n",
1360 		    __func__, len, offsetof(struct ctl_ha_msg_iid, data[0]));
1361 		ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1362 		return;
1363 	}
1364 	i = msg->iid.name_len;
1365 	if (len < offsetof(struct ctl_ha_msg_iid, data[i])) {
1366 		printf("%s: Received truncated message data %d < %zu\n",
1367 		    __func__, len, offsetof(struct ctl_ha_msg_iid, data[i]));
1368 		ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1369 		return;
1370 	}
1371 
1372 	port = softc->ctl_ports[msg->hdr.nexus.targ_port];
1373 	if (port == NULL) {
1374 		printf("%s: Received IID for unknown port %d\n",
1375 		    __func__, msg->hdr.nexus.targ_port);
1376 		return;
1377 	}
1378 	iid = msg->hdr.nexus.initid;
1379 	if (port->wwpn_iid[iid].in_use != 0 &&
1380 	    msg->iid.in_use == 0)
1381 		ctl_i_t_nexus_loss(softc, iid, CTL_UA_POWERON);
1382 	port->wwpn_iid[iid].in_use = msg->iid.in_use;
1383 	port->wwpn_iid[iid].wwpn = msg->iid.wwpn;
1384 	free(port->wwpn_iid[iid].name, M_CTL);
1385 	if (msg->iid.name_len) {
1386 		port->wwpn_iid[iid].name = strndup(&msg->iid.data[0],
1387 		    msg->iid.name_len, M_CTL);
1388 	} else
1389 		port->wwpn_iid[iid].name = NULL;
1390 }
1391 
1392 static void
1393 ctl_isc_login(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1394 {
1395 
1396 	if (len < sizeof(msg->login)) {
1397 		printf("%s: Received truncated message %d < %zu\n",
1398 		    __func__, len, sizeof(msg->login));
1399 		ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1400 		return;
1401 	}
1402 
1403 	if (msg->login.version != CTL_HA_VERSION) {
1404 		printf("CTL HA peers have different versions %d != %d\n",
1405 		    msg->login.version, CTL_HA_VERSION);
1406 		ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1407 		return;
1408 	}
1409 	if (msg->login.ha_mode != softc->ha_mode) {
1410 		printf("CTL HA peers have different ha_mode %d != %d\n",
1411 		    msg->login.ha_mode, softc->ha_mode);
1412 		ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1413 		return;
1414 	}
1415 	if (msg->login.ha_id == softc->ha_id) {
1416 		printf("CTL HA peers have same ha_id %d\n", msg->login.ha_id);
1417 		ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1418 		return;
1419 	}
1420 	if (msg->login.max_luns != ctl_max_luns ||
1421 	    msg->login.max_ports != ctl_max_ports ||
1422 	    msg->login.max_init_per_port != CTL_MAX_INIT_PER_PORT) {
1423 		printf("CTL HA peers have different limits\n");
1424 		ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1425 		return;
1426 	}
1427 }
1428 
1429 static void
1430 ctl_isc_mode_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1431 {
1432 	struct ctl_lun *lun;
1433 	u_int i;
1434 	uint32_t initidx, targ_lun;
1435 
1436 	if (len < offsetof(struct ctl_ha_msg_mode, data[0])) {
1437 		printf("%s: Received truncated message %d < %zu\n",
1438 		    __func__, len, offsetof(struct ctl_ha_msg_mode, data[0]));
1439 		ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1440 		return;
1441 	}
1442 	i = msg->mode.page_len;
1443 	if (len < offsetof(struct ctl_ha_msg_mode, data[i])) {
1444 		printf("%s: Received truncated message data %d < %zu\n",
1445 		    __func__, len, offsetof(struct ctl_ha_msg_mode, data[i]));
1446 		ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1447 		return;
1448 	}
1449 
1450 	targ_lun = msg->hdr.nexus.targ_mapped_lun;
1451 	mtx_lock(&softc->ctl_lock);
1452 	if (targ_lun >= ctl_max_luns ||
1453 	    (lun = softc->ctl_luns[targ_lun]) == NULL) {
1454 		mtx_unlock(&softc->ctl_lock);
1455 		return;
1456 	}
1457 	mtx_lock(&lun->lun_lock);
1458 	mtx_unlock(&softc->ctl_lock);
1459 	if (lun->flags & CTL_LUN_DISABLED) {
1460 		mtx_unlock(&lun->lun_lock);
1461 		return;
1462 	}
1463 	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
1464 		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) ==
1465 		    msg->mode.page_code &&
1466 		    lun->mode_pages.index[i].subpage == msg->mode.subpage)
1467 			break;
1468 	}
1469 	if (i == CTL_NUM_MODE_PAGES) {
1470 		mtx_unlock(&lun->lun_lock);
1471 		return;
1472 	}
1473 	memcpy(lun->mode_pages.index[i].page_data, msg->mode.data,
1474 	    min(lun->mode_pages.index[i].page_len, msg->mode.page_len));
1475 	initidx = ctl_get_initindex(&msg->hdr.nexus);
1476 	if (initidx != -1)
1477 		ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
1478 	mtx_unlock(&lun->lun_lock);
1479 }
1480 
1481 /*
1482  * ISC (Inter Shelf Communication) event handler.  Events from the HA
1483  * subsystem come in here.
1484  */
1485 static void
1486 ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
1487 {
1488 	struct ctl_softc *softc = control_softc;
1489 	union ctl_io *io;
1490 	struct ctl_prio *presio;
1491 	ctl_ha_status isc_status;
1492 
1493 	CTL_DEBUG_PRINT(("CTL: Isc Msg event %d\n", event));
1494 	if (event == CTL_HA_EVT_MSG_RECV) {
1495 		union ctl_ha_msg *msg, msgbuf;
1496 
1497 		if (param > sizeof(msgbuf))
1498 			msg = malloc(param, M_CTL, M_WAITOK);
1499 		else
1500 			msg = &msgbuf;
1501 		isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, msg, param,
1502 		    M_WAITOK);
1503 		if (isc_status != CTL_HA_STATUS_SUCCESS) {
1504 			printf("%s: Error receiving message: %d\n",
1505 			    __func__, isc_status);
1506 			if (msg != &msgbuf)
1507 				free(msg, M_CTL);
1508 			return;
1509 		}
1510 
1511 		CTL_DEBUG_PRINT(("CTL: msg_type %d len %d\n",
1512 		    msg->hdr.msg_type, param));
1513 		switch (msg->hdr.msg_type) {
1514 		case CTL_MSG_SERIALIZE:
1515 			io = ctl_alloc_io(softc->othersc_pool);
1516 			ctl_zero_io(io);
1517 			// populate ctsio from msg
1518 			io->io_hdr.io_type = CTL_IO_SCSI;
1519 			io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
1520 			io->io_hdr.remote_io = msg->hdr.original_sc;
1521 			io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
1522 					    CTL_FLAG_IO_ACTIVE;
1523 			/*
1524 			 * If we're in serialization-only mode, we don't
1525 			 * want to go through full done processing.  Thus
1526 			 * the COPY flag.
1527 			 *
1528 			 * XXX KDM add another flag that is more specific.
1529 			 */
1530 			if (softc->ha_mode != CTL_HA_MODE_XFER)
1531 				io->io_hdr.flags |= CTL_FLAG_INT_COPY;
1532 			io->io_hdr.nexus = msg->hdr.nexus;
1533 			io->scsiio.priority = msg->scsi.priority;
1534 			io->scsiio.tag_num = msg->scsi.tag_num;
1535 			io->scsiio.tag_type = msg->scsi.tag_type;
1536 #ifdef CTL_TIME_IO
1537 			io->io_hdr.start_time = time_uptime;
1538 			getbinuptime(&io->io_hdr.start_bt);
1539 #endif /* CTL_TIME_IO */
1540 			io->scsiio.cdb_len = msg->scsi.cdb_len;
1541 			memcpy(io->scsiio.cdb, msg->scsi.cdb,
1542 			       CTL_MAX_CDBLEN);
1543 			if (softc->ha_mode == CTL_HA_MODE_XFER) {
1544 				const struct ctl_cmd_entry *entry;
1545 
1546 				entry = ctl_get_cmd_entry(&io->scsiio, NULL);
1547 				io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
1548 				io->io_hdr.flags |=
1549 					entry->flags & CTL_FLAG_DATA_MASK;
1550 			}
1551 			ctl_enqueue_isc(io);
1552 			break;
1553 
1554 		/* Performed on the Originating SC, XFER mode only */
1555 		case CTL_MSG_DATAMOVE: {
1556 			struct ctl_sg_entry *sgl;
1557 			int i, j;
1558 
1559 			io = msg->hdr.original_sc;
1560 			if (io == NULL) {
1561 				printf("%s: original_sc == NULL!\n", __func__);
1562 				/* XXX KDM do something here */
1563 				break;
1564 			}
1565 			io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
1566 			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1567 			/*
1568 			 * Keep track of this, we need to send it back over
1569 			 * when the datamove is complete.
1570 			 */
1571 			io->io_hdr.remote_io = msg->hdr.serializing_sc;
1572 			if (msg->hdr.status == CTL_SUCCESS)
1573 				io->io_hdr.status = msg->hdr.status;
1574 
1575 			if (msg->dt.sg_sequence == 0) {
1576 #ifdef CTL_TIME_IO
1577 				getbinuptime(&io->io_hdr.dma_start_bt);
1578 #endif
1579 				i = msg->dt.kern_sg_entries +
1580 				    msg->dt.kern_data_len /
1581 				    CTL_HA_DATAMOVE_SEGMENT + 1;
1582 				sgl = malloc(sizeof(*sgl) * i, M_CTL,
1583 				    M_WAITOK | M_ZERO);
1584 				CTL_RSGL(io) = sgl;
1585 				CTL_LSGL(io) = &sgl[msg->dt.kern_sg_entries];
1586 
1587 				io->scsiio.kern_data_ptr = (uint8_t *)sgl;
1588 
1589 				io->scsiio.kern_sg_entries =
1590 					msg->dt.kern_sg_entries;
1591 				io->scsiio.rem_sg_entries =
1592 					msg->dt.kern_sg_entries;
1593 				io->scsiio.kern_data_len =
1594 					msg->dt.kern_data_len;
1595 				io->scsiio.kern_total_len =
1596 					msg->dt.kern_total_len;
1597 				io->scsiio.kern_data_resid =
1598 					msg->dt.kern_data_resid;
1599 				io->scsiio.kern_rel_offset =
1600 					msg->dt.kern_rel_offset;
1601 				io->io_hdr.flags &= ~CTL_FLAG_BUS_ADDR;
1602 				io->io_hdr.flags |= msg->dt.flags &
1603 				    CTL_FLAG_BUS_ADDR;
1604 			} else
1605 				sgl = (struct ctl_sg_entry *)
1606 					io->scsiio.kern_data_ptr;
1607 
1608 			for (i = msg->dt.sent_sg_entries, j = 0;
1609 			     i < (msg->dt.sent_sg_entries +
1610 			     msg->dt.cur_sg_entries); i++, j++) {
1611 				sgl[i].addr = msg->dt.sg_list[j].addr;
1612 				sgl[i].len = msg->dt.sg_list[j].len;
1613 			}
1614 
1615 			/*
1616 			 * If this is the last piece of the I/O, we've got
1617 			 * the full S/G list.  Queue processing in the thread.
1618 			 * Otherwise wait for the next piece.
1619 			 */
1620 			if (msg->dt.sg_last != 0)
1621 				ctl_enqueue_isc(io);
1622 			break;
1623 		}
1624 		/* Performed on the Serializing (primary) SC, XFER mode only */
1625 		case CTL_MSG_DATAMOVE_DONE: {
1626 			if (msg->hdr.serializing_sc == NULL) {
1627 				printf("%s: serializing_sc == NULL!\n",
1628 				       __func__);
1629 				/* XXX KDM now what? */
1630 				break;
1631 			}
1632 			/*
1633 			 * We grab the sense information here in case
1634 			 * there was a failure, so we can return status
1635 			 * back to the initiator.
1636 			 */
1637 			io = msg->hdr.serializing_sc;
1638 			io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
1639 			io->io_hdr.flags &= ~CTL_FLAG_DMA_INPROG;
1640 			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1641 			io->io_hdr.port_status = msg->scsi.port_status;
1642 			io->scsiio.kern_data_resid = msg->scsi.kern_data_resid;
1643 			if (msg->hdr.status != CTL_STATUS_NONE) {
1644 				io->io_hdr.status = msg->hdr.status;
1645 				io->scsiio.scsi_status = msg->scsi.scsi_status;
1646 				io->scsiio.sense_len = msg->scsi.sense_len;
1647 				memcpy(&io->scsiio.sense_data,
1648 				    &msg->scsi.sense_data,
1649 				    msg->scsi.sense_len);
1650 				if (msg->hdr.status == CTL_SUCCESS)
1651 					io->io_hdr.flags |= CTL_FLAG_STATUS_SENT;
1652 			}
1653 			ctl_enqueue_isc(io);
1654 			break;
1655 		}
1656 
1657 		/* Preformed on Originating SC, SER_ONLY mode */
1658 		case CTL_MSG_R2R:
1659 			io = msg->hdr.original_sc;
1660 			if (io == NULL) {
1661 				printf("%s: original_sc == NULL!\n",
1662 				    __func__);
1663 				break;
1664 			}
1665 			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1666 			io->io_hdr.msg_type = CTL_MSG_R2R;
1667 			io->io_hdr.remote_io = msg->hdr.serializing_sc;
1668 			ctl_enqueue_isc(io);
1669 			break;
1670 
1671 		/*
1672 		 * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
1673 		 * mode.
1674 		 * Performed on the Originating (i.e. secondary) SC in XFER
1675 		 * mode
1676 		 */
1677 		case CTL_MSG_FINISH_IO:
1678 			if (softc->ha_mode == CTL_HA_MODE_XFER)
1679 				ctl_isc_handler_finish_xfer(softc, msg);
1680 			else
1681 				ctl_isc_handler_finish_ser_only(softc, msg);
1682 			break;
1683 
1684 		/* Preformed on Originating SC */
1685 		case CTL_MSG_BAD_JUJU:
1686 			io = msg->hdr.original_sc;
1687 			if (io == NULL) {
1688 				printf("%s: Bad JUJU!, original_sc is NULL!\n",
1689 				       __func__);
1690 				break;
1691 			}
1692 			ctl_copy_sense_data(msg, io);
1693 			/*
1694 			 * IO should have already been cleaned up on other
1695 			 * SC so clear this flag so we won't send a message
1696 			 * back to finish the IO there.
1697 			 */
1698 			io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
1699 			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1700 
1701 			/* io = msg->hdr.serializing_sc; */
1702 			io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
1703 			ctl_enqueue_isc(io);
1704 			break;
1705 
1706 		/* Handle resets sent from the other side */
1707 		case CTL_MSG_MANAGE_TASKS: {
1708 			struct ctl_taskio *taskio;
1709 			taskio = (struct ctl_taskio *)ctl_alloc_io(
1710 			    softc->othersc_pool);
1711 			ctl_zero_io((union ctl_io *)taskio);
1712 			taskio->io_hdr.io_type = CTL_IO_TASK;
1713 			taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
1714 			taskio->io_hdr.nexus = msg->hdr.nexus;
1715 			taskio->task_action = msg->task.task_action;
1716 			taskio->tag_num = msg->task.tag_num;
1717 			taskio->tag_type = msg->task.tag_type;
1718 #ifdef CTL_TIME_IO
1719 			taskio->io_hdr.start_time = time_uptime;
1720 			getbinuptime(&taskio->io_hdr.start_bt);
1721 #endif /* CTL_TIME_IO */
1722 			ctl_run_task((union ctl_io *)taskio);
1723 			break;
1724 		}
1725 		/* Persistent Reserve action which needs attention */
1726 		case CTL_MSG_PERS_ACTION:
1727 			presio = (struct ctl_prio *)ctl_alloc_io(
1728 			    softc->othersc_pool);
1729 			ctl_zero_io((union ctl_io *)presio);
1730 			presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
1731 			presio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
1732 			presio->io_hdr.nexus = msg->hdr.nexus;
1733 			presio->pr_msg = msg->pr;
1734 			ctl_enqueue_isc((union ctl_io *)presio);
1735 			break;
1736 		case CTL_MSG_UA:
1737 			ctl_isc_ua(softc, msg, param);
1738 			break;
1739 		case CTL_MSG_PORT_SYNC:
1740 			ctl_isc_port_sync(softc, msg, param);
1741 			break;
1742 		case CTL_MSG_LUN_SYNC:
1743 			ctl_isc_lun_sync(softc, msg, param);
1744 			break;
1745 		case CTL_MSG_IID_SYNC:
1746 			ctl_isc_iid_sync(softc, msg, param);
1747 			break;
1748 		case CTL_MSG_LOGIN:
1749 			ctl_isc_login(softc, msg, param);
1750 			break;
1751 		case CTL_MSG_MODE_SYNC:
1752 			ctl_isc_mode_sync(softc, msg, param);
1753 			break;
1754 		default:
1755 			printf("Received HA message of unknown type %d\n",
1756 			    msg->hdr.msg_type);
1757 			ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1758 			break;
1759 		}
1760 		if (msg != &msgbuf)
1761 			free(msg, M_CTL);
1762 	} else if (event == CTL_HA_EVT_LINK_CHANGE) {
1763 		printf("CTL: HA link status changed from %d to %d\n",
1764 		    softc->ha_link, param);
1765 		if (param == softc->ha_link)
1766 			return;
1767 		if (softc->ha_link == CTL_HA_LINK_ONLINE) {
1768 			softc->ha_link = param;
1769 			ctl_isc_ha_link_down(softc);
1770 		} else {
1771 			softc->ha_link = param;
1772 			if (softc->ha_link == CTL_HA_LINK_ONLINE)
1773 				ctl_isc_ha_link_up(softc);
1774 		}
1775 		return;
1776 	} else {
1777 		printf("ctl_isc_event_handler: Unknown event %d\n", event);
1778 		return;
1779 	}
1780 }
1781 
1782 static void
1783 ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
1784 {
1785 
1786 	memcpy(&dest->scsiio.sense_data, &src->scsi.sense_data,
1787 	    src->scsi.sense_len);
1788 	dest->scsiio.scsi_status = src->scsi.scsi_status;
1789 	dest->scsiio.sense_len = src->scsi.sense_len;
1790 	dest->io_hdr.status = src->hdr.status;
1791 }
1792 
1793 static void
1794 ctl_copy_sense_data_back(union ctl_io *src, union ctl_ha_msg *dest)
1795 {
1796 
1797 	memcpy(&dest->scsi.sense_data, &src->scsiio.sense_data,
1798 	    src->scsiio.sense_len);
1799 	dest->scsi.scsi_status = src->scsiio.scsi_status;
1800 	dest->scsi.sense_len = src->scsiio.sense_len;
1801 	dest->hdr.status = src->io_hdr.status;
1802 }
1803 
1804 void
1805 ctl_est_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
1806 {
1807 	struct ctl_softc *softc = lun->ctl_softc;
1808 	ctl_ua_type *pu;
1809 
1810 	if (initidx < softc->init_min || initidx >= softc->init_max)
1811 		return;
1812 	mtx_assert(&lun->lun_lock, MA_OWNED);
1813 	pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
1814 	if (pu == NULL)
1815 		return;
1816 	pu[initidx % CTL_MAX_INIT_PER_PORT] |= ua;
1817 }
1818 
1819 void
1820 ctl_est_ua_port(struct ctl_lun *lun, int port, uint32_t except, ctl_ua_type ua)
1821 {
1822 	int i;
1823 
1824 	mtx_assert(&lun->lun_lock, MA_OWNED);
1825 	if (lun->pending_ua[port] == NULL)
1826 		return;
1827 	for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1828 		if (port * CTL_MAX_INIT_PER_PORT + i == except)
1829 			continue;
1830 		lun->pending_ua[port][i] |= ua;
1831 	}
1832 }
1833 
1834 void
1835 ctl_est_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1836 {
1837 	struct ctl_softc *softc = lun->ctl_softc;
1838 	int i;
1839 
1840 	mtx_assert(&lun->lun_lock, MA_OWNED);
1841 	for (i = softc->port_min; i < softc->port_max; i++)
1842 		ctl_est_ua_port(lun, i, except, ua);
1843 }
1844 
1845 void
1846 ctl_clr_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
1847 {
1848 	struct ctl_softc *softc = lun->ctl_softc;
1849 	ctl_ua_type *pu;
1850 
1851 	if (initidx < softc->init_min || initidx >= softc->init_max)
1852 		return;
1853 	mtx_assert(&lun->lun_lock, MA_OWNED);
1854 	pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
1855 	if (pu == NULL)
1856 		return;
1857 	pu[initidx % CTL_MAX_INIT_PER_PORT] &= ~ua;
1858 }
1859 
1860 void
1861 ctl_clr_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1862 {
1863 	struct ctl_softc *softc = lun->ctl_softc;
1864 	int i, j;
1865 
1866 	mtx_assert(&lun->lun_lock, MA_OWNED);
1867 	for (i = softc->port_min; i < softc->port_max; i++) {
1868 		if (lun->pending_ua[i] == NULL)
1869 			continue;
1870 		for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
1871 			if (i * CTL_MAX_INIT_PER_PORT + j == except)
1872 				continue;
1873 			lun->pending_ua[i][j] &= ~ua;
1874 		}
1875 	}
1876 }
1877 
1878 void
1879 ctl_clr_ua_allluns(struct ctl_softc *ctl_softc, uint32_t initidx,
1880     ctl_ua_type ua_type)
1881 {
1882 	struct ctl_lun *lun;
1883 
1884 	mtx_assert(&ctl_softc->ctl_lock, MA_OWNED);
1885 	STAILQ_FOREACH(lun, &ctl_softc->lun_list, links) {
1886 		mtx_lock(&lun->lun_lock);
1887 		ctl_clr_ua(lun, initidx, ua_type);
1888 		mtx_unlock(&lun->lun_lock);
1889 	}
1890 }
1891 
1892 static int
1893 ctl_ha_role_sysctl(SYSCTL_HANDLER_ARGS)
1894 {
1895 	struct ctl_softc *softc = (struct ctl_softc *)arg1;
1896 	struct ctl_lun *lun;
1897 	struct ctl_lun_req ireq;
1898 	int error, value;
1899 
1900 	value = (softc->flags & CTL_FLAG_ACTIVE_SHELF) ? 0 : 1;
1901 	error = sysctl_handle_int(oidp, &value, 0, req);
1902 	if ((error != 0) || (req->newptr == NULL))
1903 		return (error);
1904 
1905 	mtx_lock(&softc->ctl_lock);
1906 	if (value == 0)
1907 		softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1908 	else
1909 		softc->flags &= ~CTL_FLAG_ACTIVE_SHELF;
1910 	STAILQ_FOREACH(lun, &softc->lun_list, links) {
1911 		mtx_unlock(&softc->ctl_lock);
1912 		bzero(&ireq, sizeof(ireq));
1913 		ireq.reqtype = CTL_LUNREQ_MODIFY;
1914 		ireq.reqdata.modify.lun_id = lun->lun;
1915 		lun->backend->ioctl(NULL, CTL_LUN_REQ, (caddr_t)&ireq, 0,
1916 		    curthread);
1917 		if (ireq.status != CTL_LUN_OK) {
1918 			printf("%s: CTL_LUNREQ_MODIFY returned %d '%s'\n",
1919 			    __func__, ireq.status, ireq.error_str);
1920 		}
1921 		mtx_lock(&softc->ctl_lock);
1922 	}
1923 	mtx_unlock(&softc->ctl_lock);
1924 	return (0);
1925 }
1926 
1927 static int
1928 ctl_init(void)
1929 {
1930 	struct make_dev_args args;
1931 	struct ctl_softc *softc;
1932 	int i, error;
1933 
1934 	softc = control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
1935 			       M_WAITOK | M_ZERO);
1936 
1937 	make_dev_args_init(&args);
1938 	args.mda_devsw = &ctl_cdevsw;
1939 	args.mda_uid = UID_ROOT;
1940 	args.mda_gid = GID_OPERATOR;
1941 	args.mda_mode = 0600;
1942 	args.mda_si_drv1 = softc;
1943 	args.mda_si_drv2 = NULL;
1944 	error = make_dev_s(&args, &softc->dev, "cam/ctl");
1945 	if (error != 0) {
1946 		free(softc, M_DEVBUF);
1947 		control_softc = NULL;
1948 		return (error);
1949 	}
1950 
1951 	sysctl_ctx_init(&softc->sysctl_ctx);
1952 	softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
1953 		SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
1954 		CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "CAM Target Layer");
1955 
1956 	if (softc->sysctl_tree == NULL) {
1957 		printf("%s: unable to allocate sysctl tree\n", __func__);
1958 		destroy_dev(softc->dev);
1959 		free(softc, M_DEVBUF);
1960 		control_softc = NULL;
1961 		return (ENOMEM);
1962 	}
1963 
1964 	mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
1965 	softc->io_zone = uma_zcreate("CTL IO", sizeof(union ctl_io),
1966 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1967 	softc->flags = 0;
1968 
1969 	SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1970 	    OID_AUTO, "ha_mode", CTLFLAG_RDTUN, (int *)&softc->ha_mode, 0,
1971 	    "HA mode (0 - act/stby, 1 - serialize only, 2 - xfer)");
1972 
1973 	if (ctl_max_luns <= 0 || powerof2(ctl_max_luns) == 0) {
1974 		printf("Bad value %d for kern.cam.ctl.max_luns, must be a power of two, using %d\n",
1975 		    ctl_max_luns, CTL_DEFAULT_MAX_LUNS);
1976 		ctl_max_luns = CTL_DEFAULT_MAX_LUNS;
1977 	}
1978 	softc->ctl_luns = malloc(sizeof(struct ctl_lun *) * ctl_max_luns,
1979 	    M_DEVBUF, M_WAITOK | M_ZERO);
1980 	softc->ctl_lun_mask = malloc(sizeof(uint32_t) *
1981 	    ((ctl_max_luns + 31) / 32), M_DEVBUF, M_WAITOK | M_ZERO);
1982 	if (ctl_max_ports <= 0 || powerof2(ctl_max_ports) == 0) {
1983 		printf("Bad value %d for kern.cam.ctl.max_ports, must be a power of two, using %d\n",
1984 		    ctl_max_ports, CTL_DEFAULT_MAX_PORTS);
1985 		ctl_max_ports = CTL_DEFAULT_MAX_PORTS;
1986 	}
1987 	softc->ctl_port_mask = malloc(sizeof(uint32_t) *
1988 	  ((ctl_max_ports + 31) / 32), M_DEVBUF, M_WAITOK | M_ZERO);
1989 	softc->ctl_ports = malloc(sizeof(struct ctl_port *) * ctl_max_ports,
1990 	     M_DEVBUF, M_WAITOK | M_ZERO);
1991 
1992 	/*
1993 	 * In Copan's HA scheme, the "master" and "slave" roles are
1994 	 * figured out through the slot the controller is in.  Although it
1995 	 * is an active/active system, someone has to be in charge.
1996 	 */
1997 	SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1998 	    OID_AUTO, "ha_id", CTLFLAG_RDTUN, &softc->ha_id, 0,
1999 	    "HA head ID (0 - no HA)");
2000 	if (softc->ha_id == 0 || softc->ha_id > NUM_HA_SHELVES) {
2001 		softc->flags |= CTL_FLAG_ACTIVE_SHELF;
2002 		softc->is_single = 1;
2003 		softc->port_cnt = ctl_max_ports;
2004 		softc->port_min = 0;
2005 	} else {
2006 		softc->port_cnt = ctl_max_ports / NUM_HA_SHELVES;
2007 		softc->port_min = (softc->ha_id - 1) * softc->port_cnt;
2008 	}
2009 	softc->port_max = softc->port_min + softc->port_cnt;
2010 	softc->init_min = softc->port_min * CTL_MAX_INIT_PER_PORT;
2011 	softc->init_max = softc->port_max * CTL_MAX_INIT_PER_PORT;
2012 
2013 	SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
2014 	    OID_AUTO, "ha_link", CTLFLAG_RD, (int *)&softc->ha_link, 0,
2015 	    "HA link state (0 - offline, 1 - unknown, 2 - online)");
2016 
2017 	STAILQ_INIT(&softc->lun_list);
2018 	STAILQ_INIT(&softc->fe_list);
2019 	STAILQ_INIT(&softc->port_list);
2020 	STAILQ_INIT(&softc->be_list);
2021 	ctl_tpc_init(softc);
2022 
2023 	if (worker_threads <= 0)
2024 		worker_threads = max(1, mp_ncpus / 4);
2025 	if (worker_threads > CTL_MAX_THREADS)
2026 		worker_threads = CTL_MAX_THREADS;
2027 
2028 	for (i = 0; i < worker_threads; i++) {
2029 		struct ctl_thread *thr = &softc->threads[i];
2030 
2031 		mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
2032 		thr->ctl_softc = softc;
2033 		STAILQ_INIT(&thr->incoming_queue);
2034 		STAILQ_INIT(&thr->rtr_queue);
2035 		STAILQ_INIT(&thr->done_queue);
2036 		STAILQ_INIT(&thr->isc_queue);
2037 
2038 		error = kproc_kthread_add(ctl_work_thread, thr,
2039 		    &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
2040 		if (error != 0) {
2041 			printf("error creating CTL work thread!\n");
2042 			return (error);
2043 		}
2044 	}
2045 	error = kproc_kthread_add(ctl_thresh_thread, softc,
2046 	    &softc->ctl_proc, &softc->thresh_thread, 0, 0, "ctl", "thresh");
2047 	if (error != 0) {
2048 		printf("error creating CTL threshold thread!\n");
2049 		return (error);
2050 	}
2051 
2052 	SYSCTL_ADD_PROC(&softc->sysctl_ctx,SYSCTL_CHILDREN(softc->sysctl_tree),
2053 	    OID_AUTO, "ha_role",
2054 	    CTLTYPE_INT | CTLFLAG_RWTUN | CTLFLAG_MPSAFE,
2055 	    softc, 0, ctl_ha_role_sysctl, "I", "HA role for this head");
2056 
2057 	if (softc->is_single == 0) {
2058 		if (ctl_frontend_register(&ha_frontend) != 0)
2059 			softc->is_single = 1;
2060 	}
2061 	return (0);
2062 }
2063 
2064 static int
2065 ctl_shutdown(void)
2066 {
2067 	struct ctl_softc *softc = control_softc;
2068 	int i;
2069 
2070 	if (softc->is_single == 0)
2071 		ctl_frontend_deregister(&ha_frontend);
2072 
2073 	destroy_dev(softc->dev);
2074 
2075 	/* Shutdown CTL threads. */
2076 	softc->shutdown = 1;
2077 	for (i = 0; i < worker_threads; i++) {
2078 		struct ctl_thread *thr = &softc->threads[i];
2079 		while (thr->thread != NULL) {
2080 			wakeup(thr);
2081 			if (thr->thread != NULL)
2082 				pause("CTL thr shutdown", 1);
2083 		}
2084 		mtx_destroy(&thr->queue_lock);
2085 	}
2086 	while (softc->thresh_thread != NULL) {
2087 		wakeup(softc->thresh_thread);
2088 		if (softc->thresh_thread != NULL)
2089 			pause("CTL thr shutdown", 1);
2090 	}
2091 
2092 	ctl_tpc_shutdown(softc);
2093 	uma_zdestroy(softc->io_zone);
2094 	mtx_destroy(&softc->ctl_lock);
2095 
2096 	free(softc->ctl_luns, M_DEVBUF);
2097 	free(softc->ctl_lun_mask, M_DEVBUF);
2098 	free(softc->ctl_port_mask, M_DEVBUF);
2099 	free(softc->ctl_ports, M_DEVBUF);
2100 
2101 	sysctl_ctx_free(&softc->sysctl_ctx);
2102 
2103 	free(softc, M_DEVBUF);
2104 	control_softc = NULL;
2105 	return (0);
2106 }
2107 
2108 static int
2109 ctl_module_event_handler(module_t mod, int what, void *arg)
2110 {
2111 
2112 	switch (what) {
2113 	case MOD_LOAD:
2114 		return (ctl_init());
2115 	case MOD_UNLOAD:
2116 		return (ctl_shutdown());
2117 	default:
2118 		return (EOPNOTSUPP);
2119 	}
2120 }
2121 
2122 /*
2123  * XXX KDM should we do some access checks here?  Bump a reference count to
2124  * prevent a CTL module from being unloaded while someone has it open?
2125  */
2126 static int
2127 ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
2128 {
2129 	return (0);
2130 }
2131 
2132 static int
2133 ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
2134 {
2135 	return (0);
2136 }
2137 
2138 /*
2139  * Remove an initiator by port number and initiator ID.
2140  * Returns 0 for success, -1 for failure.
2141  */
2142 int
2143 ctl_remove_initiator(struct ctl_port *port, int iid)
2144 {
2145 	struct ctl_softc *softc = port->ctl_softc;
2146 	int last;
2147 
2148 	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
2149 
2150 	if (iid > CTL_MAX_INIT_PER_PORT) {
2151 		printf("%s: initiator ID %u > maximun %u!\n",
2152 		       __func__, iid, CTL_MAX_INIT_PER_PORT);
2153 		return (-1);
2154 	}
2155 
2156 	mtx_lock(&softc->ctl_lock);
2157 	last = (--port->wwpn_iid[iid].in_use == 0);
2158 	port->wwpn_iid[iid].last_use = time_uptime;
2159 	mtx_unlock(&softc->ctl_lock);
2160 	if (last)
2161 		ctl_i_t_nexus_loss(softc, iid, CTL_UA_POWERON);
2162 	ctl_isc_announce_iid(port, iid);
2163 
2164 	return (0);
2165 }
2166 
2167 /*
2168  * Add an initiator to the initiator map.
2169  * Returns iid for success, < 0 for failure.
2170  */
2171 int
2172 ctl_add_initiator(struct ctl_port *port, int iid, uint64_t wwpn, char *name)
2173 {
2174 	struct ctl_softc *softc = port->ctl_softc;
2175 	time_t best_time;
2176 	int i, best;
2177 
2178 	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
2179 
2180 	if (iid >= CTL_MAX_INIT_PER_PORT) {
2181 		printf("%s: WWPN %#jx initiator ID %u > maximum %u!\n",
2182 		       __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
2183 		free(name, M_CTL);
2184 		return (-1);
2185 	}
2186 
2187 	mtx_lock(&softc->ctl_lock);
2188 
2189 	if (iid < 0 && (wwpn != 0 || name != NULL)) {
2190 		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2191 			if (wwpn != 0 && wwpn == port->wwpn_iid[i].wwpn) {
2192 				iid = i;
2193 				break;
2194 			}
2195 			if (name != NULL && port->wwpn_iid[i].name != NULL &&
2196 			    strcmp(name, port->wwpn_iid[i].name) == 0) {
2197 				iid = i;
2198 				break;
2199 			}
2200 		}
2201 	}
2202 
2203 	if (iid < 0) {
2204 		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2205 			if (port->wwpn_iid[i].in_use == 0 &&
2206 			    port->wwpn_iid[i].wwpn == 0 &&
2207 			    port->wwpn_iid[i].name == NULL) {
2208 				iid = i;
2209 				break;
2210 			}
2211 		}
2212 	}
2213 
2214 	if (iid < 0) {
2215 		best = -1;
2216 		best_time = INT32_MAX;
2217 		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2218 			if (port->wwpn_iid[i].in_use == 0) {
2219 				if (port->wwpn_iid[i].last_use < best_time) {
2220 					best = i;
2221 					best_time = port->wwpn_iid[i].last_use;
2222 				}
2223 			}
2224 		}
2225 		iid = best;
2226 	}
2227 
2228 	if (iid < 0) {
2229 		mtx_unlock(&softc->ctl_lock);
2230 		free(name, M_CTL);
2231 		return (-2);
2232 	}
2233 
2234 	if (port->wwpn_iid[iid].in_use > 0 && (wwpn != 0 || name != NULL)) {
2235 		/*
2236 		 * This is not an error yet.
2237 		 */
2238 		if (wwpn != 0 && wwpn == port->wwpn_iid[iid].wwpn) {
2239 #if 0
2240 			printf("%s: port %d iid %u WWPN %#jx arrived"
2241 			    " again\n", __func__, port->targ_port,
2242 			    iid, (uintmax_t)wwpn);
2243 #endif
2244 			goto take;
2245 		}
2246 		if (name != NULL && port->wwpn_iid[iid].name != NULL &&
2247 		    strcmp(name, port->wwpn_iid[iid].name) == 0) {
2248 #if 0
2249 			printf("%s: port %d iid %u name '%s' arrived"
2250 			    " again\n", __func__, port->targ_port,
2251 			    iid, name);
2252 #endif
2253 			goto take;
2254 		}
2255 
2256 		/*
2257 		 * This is an error, but what do we do about it?  The
2258 		 * driver is telling us we have a new WWPN for this
2259 		 * initiator ID, so we pretty much need to use it.
2260 		 */
2261 		printf("%s: port %d iid %u WWPN %#jx '%s' arrived,"
2262 		    " but WWPN %#jx '%s' is still at that address\n",
2263 		    __func__, port->targ_port, iid, wwpn, name,
2264 		    (uintmax_t)port->wwpn_iid[iid].wwpn,
2265 		    port->wwpn_iid[iid].name);
2266 	}
2267 take:
2268 	free(port->wwpn_iid[iid].name, M_CTL);
2269 	port->wwpn_iid[iid].name = name;
2270 	port->wwpn_iid[iid].wwpn = wwpn;
2271 	port->wwpn_iid[iid].in_use++;
2272 	mtx_unlock(&softc->ctl_lock);
2273 	ctl_isc_announce_iid(port, iid);
2274 
2275 	return (iid);
2276 }
2277 
2278 static int
2279 ctl_create_iid(struct ctl_port *port, int iid, uint8_t *buf)
2280 {
2281 	int len;
2282 
2283 	switch (port->port_type) {
2284 	case CTL_PORT_FC:
2285 	{
2286 		struct scsi_transportid_fcp *id =
2287 		    (struct scsi_transportid_fcp *)buf;
2288 		if (port->wwpn_iid[iid].wwpn == 0)
2289 			return (0);
2290 		memset(id, 0, sizeof(*id));
2291 		id->format_protocol = SCSI_PROTO_FC;
2292 		scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->n_port_name);
2293 		return (sizeof(*id));
2294 	}
2295 	case CTL_PORT_ISCSI:
2296 	{
2297 		struct scsi_transportid_iscsi_port *id =
2298 		    (struct scsi_transportid_iscsi_port *)buf;
2299 		if (port->wwpn_iid[iid].name == NULL)
2300 			return (0);
2301 		memset(id, 0, 256);
2302 		id->format_protocol = SCSI_TRN_ISCSI_FORMAT_PORT |
2303 		    SCSI_PROTO_ISCSI;
2304 		len = strlcpy(id->iscsi_name, port->wwpn_iid[iid].name, 252) + 1;
2305 		len = roundup2(min(len, 252), 4);
2306 		scsi_ulto2b(len, id->additional_length);
2307 		return (sizeof(*id) + len);
2308 	}
2309 	case CTL_PORT_SAS:
2310 	{
2311 		struct scsi_transportid_sas *id =
2312 		    (struct scsi_transportid_sas *)buf;
2313 		if (port->wwpn_iid[iid].wwpn == 0)
2314 			return (0);
2315 		memset(id, 0, sizeof(*id));
2316 		id->format_protocol = SCSI_PROTO_SAS;
2317 		scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->sas_address);
2318 		return (sizeof(*id));
2319 	}
2320 	default:
2321 	{
2322 		struct scsi_transportid_spi *id =
2323 		    (struct scsi_transportid_spi *)buf;
2324 		memset(id, 0, sizeof(*id));
2325 		id->format_protocol = SCSI_PROTO_SPI;
2326 		scsi_ulto2b(iid, id->scsi_addr);
2327 		scsi_ulto2b(port->targ_port, id->rel_trgt_port_id);
2328 		return (sizeof(*id));
2329 	}
2330 	}
2331 }
2332 
2333 /*
2334  * Serialize a command that went down the "wrong" side, and so was sent to
2335  * this controller for execution.  The logic is a little different than the
2336  * standard case in ctl_scsiio_precheck().  Errors in this case need to get
2337  * sent back to the other side, but in the success case, we execute the
2338  * command on this side (XFER mode) or tell the other side to execute it
2339  * (SER_ONLY mode).
2340  */
2341 static void
2342 ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
2343 {
2344 	struct ctl_softc *softc = CTL_SOFTC(ctsio);
2345 	struct ctl_port *port = CTL_PORT(ctsio);
2346 	union ctl_ha_msg msg_info;
2347 	struct ctl_lun *lun;
2348 	const struct ctl_cmd_entry *entry;
2349 	union ctl_io *bio;
2350 	uint32_t targ_lun;
2351 
2352 	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
2353 
2354 	/* Make sure that we know about this port. */
2355 	if (port == NULL || (port->status & CTL_PORT_STATUS_ONLINE) == 0) {
2356 		ctl_set_internal_failure(ctsio, /*sks_valid*/ 0,
2357 					 /*retry_count*/ 1);
2358 		goto badjuju;
2359 	}
2360 
2361 	/* Make sure that we know about this LUN. */
2362 	mtx_lock(&softc->ctl_lock);
2363 	if (targ_lun >= ctl_max_luns ||
2364 	    (lun = softc->ctl_luns[targ_lun]) == NULL) {
2365 		mtx_unlock(&softc->ctl_lock);
2366 
2367 		/*
2368 		 * The other node would not send this request to us unless
2369 		 * received announce that we are primary node for this LUN.
2370 		 * If this LUN does not exist now, it is probably result of
2371 		 * a race, so respond to initiator in the most opaque way.
2372 		 */
2373 		ctl_set_busy(ctsio);
2374 		goto badjuju;
2375 	}
2376 	mtx_lock(&lun->lun_lock);
2377 	mtx_unlock(&softc->ctl_lock);
2378 
2379 	/*
2380 	 * If the LUN is invalid, pretend that it doesn't exist.
2381 	 * It will go away as soon as all pending I/Os completed.
2382 	 */
2383 	if (lun->flags & CTL_LUN_DISABLED) {
2384 		mtx_unlock(&lun->lun_lock);
2385 		ctl_set_busy(ctsio);
2386 		goto badjuju;
2387 	}
2388 
2389 	entry = ctl_get_cmd_entry(ctsio, NULL);
2390 	ctsio->seridx = entry->seridx;
2391 	if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
2392 		mtx_unlock(&lun->lun_lock);
2393 		goto badjuju;
2394 	}
2395 
2396 	CTL_LUN(ctsio) = lun;
2397 	CTL_BACKEND_LUN(ctsio) = lun->be_lun;
2398 
2399 	/*
2400 	 * Every I/O goes into the OOA queue for a
2401 	 * particular LUN, and stays there until completion.
2402 	 */
2403 #ifdef CTL_TIME_IO
2404 	if (LIST_EMPTY(&lun->ooa_queue))
2405 		lun->idle_time += getsbinuptime() - lun->last_busy;
2406 #endif
2407 	LIST_INSERT_HEAD(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2408 
2409 	bio = (union ctl_io *)LIST_NEXT(&ctsio->io_hdr, ooa_links);
2410 	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio, &bio)) {
2411 	case CTL_ACTION_PASS:
2412 	case CTL_ACTION_SKIP:
2413 		if (softc->ha_mode == CTL_HA_MODE_XFER) {
2414 			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
2415 			ctl_enqueue_rtr((union ctl_io *)ctsio);
2416 			mtx_unlock(&lun->lun_lock);
2417 		} else {
2418 			ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
2419 			mtx_unlock(&lun->lun_lock);
2420 
2421 			/* send msg back to other side */
2422 			msg_info.hdr.original_sc = ctsio->io_hdr.remote_io;
2423 			msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
2424 			msg_info.hdr.msg_type = CTL_MSG_R2R;
2425 			ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
2426 			    sizeof(msg_info.hdr), M_WAITOK);
2427 		}
2428 		break;
2429 	case CTL_ACTION_BLOCK:
2430 		ctsio->io_hdr.blocker = bio;
2431 		TAILQ_INSERT_TAIL(&bio->io_hdr.blocked_queue, &ctsio->io_hdr,
2432 				  blocked_links);
2433 		mtx_unlock(&lun->lun_lock);
2434 		break;
2435 	case CTL_ACTION_OVERLAP:
2436 		LIST_REMOVE(&ctsio->io_hdr, ooa_links);
2437 		mtx_unlock(&lun->lun_lock);
2438 		ctl_set_overlapped_cmd(ctsio);
2439 		goto badjuju;
2440 	case CTL_ACTION_OVERLAP_TAG:
2441 		LIST_REMOVE(&ctsio->io_hdr, ooa_links);
2442 		mtx_unlock(&lun->lun_lock);
2443 		ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
2444 badjuju:
2445 		ctl_copy_sense_data_back((union ctl_io *)ctsio, &msg_info);
2446 		msg_info.hdr.original_sc = ctsio->io_hdr.remote_io;
2447 		msg_info.hdr.serializing_sc = NULL;
2448 		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
2449 		ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
2450 		    sizeof(msg_info.scsi), M_WAITOK);
2451 		ctl_free_io((union ctl_io *)ctsio);
2452 		break;
2453 	default:
2454 		__assert_unreachable();
2455 	}
2456 }
2457 
2458 /*
2459  * Returns 0 for success, errno for failure.
2460  */
2461 static void
2462 ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
2463 		   struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
2464 {
2465 	struct ctl_io_hdr *ioh;
2466 
2467 	mtx_lock(&lun->lun_lock);
2468 	ioh = LIST_FIRST(&lun->ooa_queue);
2469 	if (ioh == NULL) {
2470 		mtx_unlock(&lun->lun_lock);
2471 		return;
2472 	}
2473 	while (LIST_NEXT(ioh, ooa_links) != NULL)
2474 		ioh = LIST_NEXT(ioh, ooa_links);
2475 	for ( ; ioh; ioh = LIST_PREV(ioh, &lun->ooa_queue, ctl_io_hdr, ooa_links)) {
2476 		union ctl_io *io = (union ctl_io *)ioh;
2477 		struct ctl_ooa_entry *entry;
2478 
2479 		/*
2480 		 * If we've got more than we can fit, just count the
2481 		 * remaining entries.
2482 		 */
2483 		if (*cur_fill_num >= ooa_hdr->alloc_num) {
2484 			(*cur_fill_num)++;
2485 			continue;
2486 		}
2487 
2488 		entry = &kern_entries[*cur_fill_num];
2489 
2490 		entry->tag_num = io->scsiio.tag_num;
2491 		entry->tag_type = io->scsiio.tag_type;
2492 		entry->lun_num = lun->lun;
2493 #ifdef CTL_TIME_IO
2494 		entry->start_bt = io->io_hdr.start_bt;
2495 #endif
2496 		bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2497 		entry->cdb_len = io->scsiio.cdb_len;
2498 		if (io->io_hdr.blocker != NULL)
2499 			entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2500 
2501 		if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2502 			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2503 
2504 		if (io->io_hdr.flags & CTL_FLAG_ABORT)
2505 			entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2506 
2507 		if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2508 			entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2509 
2510 		if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2511 			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2512 
2513 		if (io->io_hdr.flags & CTL_FLAG_STATUS_QUEUED)
2514 			entry->cmd_flags |= CTL_OOACMD_FLAG_STATUS_QUEUED;
2515 
2516 		if (io->io_hdr.flags & CTL_FLAG_STATUS_SENT)
2517 			entry->cmd_flags |= CTL_OOACMD_FLAG_STATUS_SENT;
2518 		(*cur_fill_num)++;
2519 	}
2520 	mtx_unlock(&lun->lun_lock);
2521 }
2522 
2523 /*
2524  * Escape characters that are illegal or not recommended in XML.
2525  */
2526 int
2527 ctl_sbuf_printf_esc(struct sbuf *sb, char *str, int size)
2528 {
2529 	char *end = str + size;
2530 	int retval;
2531 
2532 	retval = 0;
2533 
2534 	for (; *str && str < end; str++) {
2535 		switch (*str) {
2536 		case '&':
2537 			retval = sbuf_cat(sb, "&amp;");
2538 			break;
2539 		case '>':
2540 			retval = sbuf_cat(sb, "&gt;");
2541 			break;
2542 		case '<':
2543 			retval = sbuf_cat(sb, "&lt;");
2544 			break;
2545 		default:
2546 			retval = sbuf_putc(sb, *str);
2547 			break;
2548 		}
2549 
2550 		if (retval != 0)
2551 			break;
2552 	}
2553 
2554 	return (retval);
2555 }
2556 
2557 static void
2558 ctl_id_sbuf(struct ctl_devid *id, struct sbuf *sb)
2559 {
2560 	struct scsi_vpd_id_descriptor *desc;
2561 	int i;
2562 
2563 	if (id == NULL || id->len < 4)
2564 		return;
2565 	desc = (struct scsi_vpd_id_descriptor *)id->data;
2566 	switch (desc->id_type & SVPD_ID_TYPE_MASK) {
2567 	case SVPD_ID_TYPE_T10:
2568 		sbuf_cat(sb, "t10.");
2569 		break;
2570 	case SVPD_ID_TYPE_EUI64:
2571 		sbuf_cat(sb, "eui.");
2572 		break;
2573 	case SVPD_ID_TYPE_NAA:
2574 		sbuf_cat(sb, "naa.");
2575 		break;
2576 	case SVPD_ID_TYPE_SCSI_NAME:
2577 		break;
2578 	}
2579 	switch (desc->proto_codeset & SVPD_ID_CODESET_MASK) {
2580 	case SVPD_ID_CODESET_BINARY:
2581 		for (i = 0; i < desc->length; i++)
2582 			sbuf_printf(sb, "%02x", desc->identifier[i]);
2583 		break;
2584 	case SVPD_ID_CODESET_ASCII:
2585 		sbuf_printf(sb, "%.*s", (int)desc->length,
2586 		    (char *)desc->identifier);
2587 		break;
2588 	case SVPD_ID_CODESET_UTF8:
2589 		sbuf_cat(sb, (char *)desc->identifier);
2590 		break;
2591 	}
2592 }
2593 
2594 static int
2595 ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2596 	  struct thread *td)
2597 {
2598 	struct ctl_softc *softc = dev->si_drv1;
2599 	struct ctl_port *port;
2600 	struct ctl_lun *lun;
2601 	int retval;
2602 
2603 	retval = 0;
2604 
2605 	switch (cmd) {
2606 	case CTL_IO:
2607 		retval = ctl_ioctl_io(dev, cmd, addr, flag, td);
2608 		break;
2609 	case CTL_ENABLE_PORT:
2610 	case CTL_DISABLE_PORT:
2611 	case CTL_SET_PORT_WWNS: {
2612 		struct ctl_port *port;
2613 		struct ctl_port_entry *entry;
2614 
2615 		entry = (struct ctl_port_entry *)addr;
2616 
2617 		mtx_lock(&softc->ctl_lock);
2618 		STAILQ_FOREACH(port, &softc->port_list, links) {
2619 			int action, done;
2620 
2621 			if (port->targ_port < softc->port_min ||
2622 			    port->targ_port >= softc->port_max)
2623 				continue;
2624 
2625 			action = 0;
2626 			done = 0;
2627 			if ((entry->port_type == CTL_PORT_NONE)
2628 			 && (entry->targ_port == port->targ_port)) {
2629 				/*
2630 				 * If the user only wants to enable or
2631 				 * disable or set WWNs on a specific port,
2632 				 * do the operation and we're done.
2633 				 */
2634 				action = 1;
2635 				done = 1;
2636 			} else if (entry->port_type & port->port_type) {
2637 				/*
2638 				 * Compare the user's type mask with the
2639 				 * particular frontend type to see if we
2640 				 * have a match.
2641 				 */
2642 				action = 1;
2643 				done = 0;
2644 
2645 				/*
2646 				 * Make sure the user isn't trying to set
2647 				 * WWNs on multiple ports at the same time.
2648 				 */
2649 				if (cmd == CTL_SET_PORT_WWNS) {
2650 					printf("%s: Can't set WWNs on "
2651 					       "multiple ports\n", __func__);
2652 					retval = EINVAL;
2653 					break;
2654 				}
2655 			}
2656 			if (action == 0)
2657 				continue;
2658 
2659 			/*
2660 			 * XXX KDM we have to drop the lock here, because
2661 			 * the online/offline operations can potentially
2662 			 * block.  We need to reference count the frontends
2663 			 * so they can't go away,
2664 			 */
2665 			if (cmd == CTL_ENABLE_PORT) {
2666 				mtx_unlock(&softc->ctl_lock);
2667 				ctl_port_online(port);
2668 				mtx_lock(&softc->ctl_lock);
2669 			} else if (cmd == CTL_DISABLE_PORT) {
2670 				mtx_unlock(&softc->ctl_lock);
2671 				ctl_port_offline(port);
2672 				mtx_lock(&softc->ctl_lock);
2673 			} else if (cmd == CTL_SET_PORT_WWNS) {
2674 				ctl_port_set_wwns(port,
2675 				    (entry->flags & CTL_PORT_WWNN_VALID) ?
2676 				    1 : 0, entry->wwnn,
2677 				    (entry->flags & CTL_PORT_WWPN_VALID) ?
2678 				    1 : 0, entry->wwpn);
2679 			}
2680 			if (done != 0)
2681 				break;
2682 		}
2683 		mtx_unlock(&softc->ctl_lock);
2684 		break;
2685 	}
2686 	case CTL_GET_OOA: {
2687 		struct ctl_ooa *ooa_hdr;
2688 		struct ctl_ooa_entry *entries;
2689 		uint32_t cur_fill_num;
2690 
2691 		ooa_hdr = (struct ctl_ooa *)addr;
2692 
2693 		if ((ooa_hdr->alloc_len == 0)
2694 		 || (ooa_hdr->alloc_num == 0)) {
2695 			printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2696 			       "must be non-zero\n", __func__,
2697 			       ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2698 			retval = EINVAL;
2699 			break;
2700 		}
2701 
2702 		if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2703 		    sizeof(struct ctl_ooa_entry))) {
2704 			printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2705 			       "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2706 			       __func__, ooa_hdr->alloc_len,
2707 			       ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2708 			retval = EINVAL;
2709 			break;
2710 		}
2711 
2712 		entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2713 		if (entries == NULL) {
2714 			printf("%s: could not allocate %d bytes for OOA "
2715 			       "dump\n", __func__, ooa_hdr->alloc_len);
2716 			retval = ENOMEM;
2717 			break;
2718 		}
2719 
2720 		mtx_lock(&softc->ctl_lock);
2721 		if ((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0 &&
2722 		    (ooa_hdr->lun_num >= ctl_max_luns ||
2723 		     softc->ctl_luns[ooa_hdr->lun_num] == NULL)) {
2724 			mtx_unlock(&softc->ctl_lock);
2725 			free(entries, M_CTL);
2726 			printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2727 			       __func__, (uintmax_t)ooa_hdr->lun_num);
2728 			retval = EINVAL;
2729 			break;
2730 		}
2731 
2732 		cur_fill_num = 0;
2733 
2734 		if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2735 			STAILQ_FOREACH(lun, &softc->lun_list, links) {
2736 				ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2737 				    ooa_hdr, entries);
2738 			}
2739 		} else {
2740 			lun = softc->ctl_luns[ooa_hdr->lun_num];
2741 			ctl_ioctl_fill_ooa(lun, &cur_fill_num, ooa_hdr,
2742 			    entries);
2743 		}
2744 		mtx_unlock(&softc->ctl_lock);
2745 
2746 		ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2747 		ooa_hdr->fill_len = ooa_hdr->fill_num *
2748 			sizeof(struct ctl_ooa_entry);
2749 		retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2750 		if (retval != 0) {
2751 			printf("%s: error copying out %d bytes for OOA dump\n",
2752 			       __func__, ooa_hdr->fill_len);
2753 		}
2754 
2755 		getbinuptime(&ooa_hdr->cur_bt);
2756 
2757 		if (cur_fill_num > ooa_hdr->alloc_num) {
2758 			ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2759 			ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2760 		} else {
2761 			ooa_hdr->dropped_num = 0;
2762 			ooa_hdr->status = CTL_OOA_OK;
2763 		}
2764 
2765 		free(entries, M_CTL);
2766 		break;
2767 	}
2768 	case CTL_DELAY_IO: {
2769 		struct ctl_io_delay_info *delay_info;
2770 
2771 		delay_info = (struct ctl_io_delay_info *)addr;
2772 
2773 #ifdef CTL_IO_DELAY
2774 		mtx_lock(&softc->ctl_lock);
2775 		if (delay_info->lun_id >= ctl_max_luns ||
2776 		    (lun = softc->ctl_luns[delay_info->lun_id]) == NULL) {
2777 			mtx_unlock(&softc->ctl_lock);
2778 			delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2779 			break;
2780 		}
2781 		mtx_lock(&lun->lun_lock);
2782 		mtx_unlock(&softc->ctl_lock);
2783 		delay_info->status = CTL_DELAY_STATUS_OK;
2784 		switch (delay_info->delay_type) {
2785 		case CTL_DELAY_TYPE_CONT:
2786 		case CTL_DELAY_TYPE_ONESHOT:
2787 			break;
2788 		default:
2789 			delay_info->status = CTL_DELAY_STATUS_INVALID_TYPE;
2790 			break;
2791 		}
2792 		switch (delay_info->delay_loc) {
2793 		case CTL_DELAY_LOC_DATAMOVE:
2794 			lun->delay_info.datamove_type = delay_info->delay_type;
2795 			lun->delay_info.datamove_delay = delay_info->delay_secs;
2796 			break;
2797 		case CTL_DELAY_LOC_DONE:
2798 			lun->delay_info.done_type = delay_info->delay_type;
2799 			lun->delay_info.done_delay = delay_info->delay_secs;
2800 			break;
2801 		default:
2802 			delay_info->status = CTL_DELAY_STATUS_INVALID_LOC;
2803 			break;
2804 		}
2805 		mtx_unlock(&lun->lun_lock);
2806 #else
2807 		delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2808 #endif /* CTL_IO_DELAY */
2809 		break;
2810 	}
2811 	case CTL_ERROR_INJECT: {
2812 		struct ctl_error_desc *err_desc, *new_err_desc;
2813 
2814 		err_desc = (struct ctl_error_desc *)addr;
2815 
2816 		new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2817 				      M_WAITOK | M_ZERO);
2818 		bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2819 
2820 		mtx_lock(&softc->ctl_lock);
2821 		if (err_desc->lun_id >= ctl_max_luns ||
2822 		    (lun = softc->ctl_luns[err_desc->lun_id]) == NULL) {
2823 			mtx_unlock(&softc->ctl_lock);
2824 			free(new_err_desc, M_CTL);
2825 			printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2826 			       __func__, (uintmax_t)err_desc->lun_id);
2827 			retval = EINVAL;
2828 			break;
2829 		}
2830 		mtx_lock(&lun->lun_lock);
2831 		mtx_unlock(&softc->ctl_lock);
2832 
2833 		/*
2834 		 * We could do some checking here to verify the validity
2835 		 * of the request, but given the complexity of error
2836 		 * injection requests, the checking logic would be fairly
2837 		 * complex.
2838 		 *
2839 		 * For now, if the request is invalid, it just won't get
2840 		 * executed and might get deleted.
2841 		 */
2842 		STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2843 
2844 		/*
2845 		 * XXX KDM check to make sure the serial number is unique,
2846 		 * in case we somehow manage to wrap.  That shouldn't
2847 		 * happen for a very long time, but it's the right thing to
2848 		 * do.
2849 		 */
2850 		new_err_desc->serial = lun->error_serial;
2851 		err_desc->serial = lun->error_serial;
2852 		lun->error_serial++;
2853 
2854 		mtx_unlock(&lun->lun_lock);
2855 		break;
2856 	}
2857 	case CTL_ERROR_INJECT_DELETE: {
2858 		struct ctl_error_desc *delete_desc, *desc, *desc2;
2859 		int delete_done;
2860 
2861 		delete_desc = (struct ctl_error_desc *)addr;
2862 		delete_done = 0;
2863 
2864 		mtx_lock(&softc->ctl_lock);
2865 		if (delete_desc->lun_id >= ctl_max_luns ||
2866 		    (lun = softc->ctl_luns[delete_desc->lun_id]) == NULL) {
2867 			mtx_unlock(&softc->ctl_lock);
2868 			printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2869 			       __func__, (uintmax_t)delete_desc->lun_id);
2870 			retval = EINVAL;
2871 			break;
2872 		}
2873 		mtx_lock(&lun->lun_lock);
2874 		mtx_unlock(&softc->ctl_lock);
2875 		STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2876 			if (desc->serial != delete_desc->serial)
2877 				continue;
2878 
2879 			STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2880 				      links);
2881 			free(desc, M_CTL);
2882 			delete_done = 1;
2883 		}
2884 		mtx_unlock(&lun->lun_lock);
2885 		if (delete_done == 0) {
2886 			printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2887 			       "error serial %ju on LUN %u\n", __func__,
2888 			       delete_desc->serial, delete_desc->lun_id);
2889 			retval = EINVAL;
2890 			break;
2891 		}
2892 		break;
2893 	}
2894 	case CTL_DUMP_STRUCTS: {
2895 		int j, k;
2896 		struct ctl_port *port;
2897 		struct ctl_frontend *fe;
2898 
2899 		mtx_lock(&softc->ctl_lock);
2900 		printf("CTL Persistent Reservation information start:\n");
2901 		STAILQ_FOREACH(lun, &softc->lun_list, links) {
2902 			mtx_lock(&lun->lun_lock);
2903 			if ((lun->flags & CTL_LUN_DISABLED) != 0) {
2904 				mtx_unlock(&lun->lun_lock);
2905 				continue;
2906 			}
2907 
2908 			for (j = 0; j < ctl_max_ports; j++) {
2909 				if (lun->pr_keys[j] == NULL)
2910 					continue;
2911 				for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2912 					if (lun->pr_keys[j][k] == 0)
2913 						continue;
2914 					printf("  LUN %ju port %d iid %d key "
2915 					       "%#jx\n", lun->lun, j, k,
2916 					       (uintmax_t)lun->pr_keys[j][k]);
2917 				}
2918 			}
2919 			mtx_unlock(&lun->lun_lock);
2920 		}
2921 		printf("CTL Persistent Reservation information end\n");
2922 		printf("CTL Ports:\n");
2923 		STAILQ_FOREACH(port, &softc->port_list, links) {
2924 			printf("  Port %d '%s' Frontend '%s' Type %u pp %d vp %d WWNN "
2925 			       "%#jx WWPN %#jx\n", port->targ_port, port->port_name,
2926 			       port->frontend->name, port->port_type,
2927 			       port->physical_port, port->virtual_port,
2928 			       (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
2929 			for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
2930 				if (port->wwpn_iid[j].in_use == 0 &&
2931 				    port->wwpn_iid[j].wwpn == 0 &&
2932 				    port->wwpn_iid[j].name == NULL)
2933 					continue;
2934 
2935 				printf("    iid %u use %d WWPN %#jx '%s'\n",
2936 				    j, port->wwpn_iid[j].in_use,
2937 				    (uintmax_t)port->wwpn_iid[j].wwpn,
2938 				    port->wwpn_iid[j].name);
2939 			}
2940 		}
2941 		printf("CTL Port information end\n");
2942 		mtx_unlock(&softc->ctl_lock);
2943 		/*
2944 		 * XXX KDM calling this without a lock.  We'd likely want
2945 		 * to drop the lock before calling the frontend's dump
2946 		 * routine anyway.
2947 		 */
2948 		printf("CTL Frontends:\n");
2949 		STAILQ_FOREACH(fe, &softc->fe_list, links) {
2950 			printf("  Frontend '%s'\n", fe->name);
2951 			if (fe->fe_dump != NULL)
2952 				fe->fe_dump();
2953 		}
2954 		printf("CTL Frontend information end\n");
2955 		break;
2956 	}
2957 	case CTL_LUN_REQ: {
2958 		struct ctl_lun_req *lun_req;
2959 		struct ctl_backend_driver *backend;
2960 		void *packed;
2961 		nvlist_t *tmp_args_nvl;
2962 		size_t packed_len;
2963 
2964 		lun_req = (struct ctl_lun_req *)addr;
2965 		tmp_args_nvl = lun_req->args_nvl;
2966 
2967 		backend = ctl_backend_find(lun_req->backend);
2968 		if (backend == NULL) {
2969 			lun_req->status = CTL_LUN_ERROR;
2970 			snprintf(lun_req->error_str,
2971 				 sizeof(lun_req->error_str),
2972 				 "Backend \"%s\" not found.",
2973 				 lun_req->backend);
2974 			break;
2975 		}
2976 
2977 		if (lun_req->args != NULL) {
2978 			if (lun_req->args_len > CTL_MAX_ARGS_LEN) {
2979 				lun_req->status = CTL_LUN_ERROR;
2980 				snprintf(lun_req->error_str, sizeof(lun_req->error_str),
2981 				    "Too big args.");
2982 				break;
2983 			}
2984 			packed = malloc(lun_req->args_len, M_CTL, M_WAITOK);
2985 			if (copyin(lun_req->args, packed, lun_req->args_len) != 0) {
2986 				free(packed, M_CTL);
2987 				lun_req->status = CTL_LUN_ERROR;
2988 				snprintf(lun_req->error_str, sizeof(lun_req->error_str),
2989 				    "Cannot copyin args.");
2990 				break;
2991 			}
2992 			lun_req->args_nvl = nvlist_unpack(packed,
2993 			    lun_req->args_len, 0);
2994 			free(packed, M_CTL);
2995 
2996 			if (lun_req->args_nvl == NULL) {
2997 				lun_req->status = CTL_LUN_ERROR;
2998 				snprintf(lun_req->error_str, sizeof(lun_req->error_str),
2999 				    "Cannot unpack args nvlist.");
3000 				break;
3001 			}
3002 		} else
3003 			lun_req->args_nvl = nvlist_create(0);
3004 
3005 		lun_req->result_nvl = NULL;
3006 		retval = backend->ioctl(dev, cmd, addr, flag, td);
3007 		nvlist_destroy(lun_req->args_nvl);
3008 		lun_req->args_nvl = tmp_args_nvl;
3009 
3010 		if (lun_req->result_nvl != NULL) {
3011 			if (lun_req->result != NULL) {
3012 				packed = nvlist_pack(lun_req->result_nvl,
3013 				    &packed_len);
3014 				if (packed == NULL) {
3015 					lun_req->status = CTL_LUN_ERROR;
3016 					snprintf(lun_req->error_str,
3017 					    sizeof(lun_req->error_str),
3018 					    "Cannot pack result nvlist.");
3019 					break;
3020 				}
3021 
3022 				if (packed_len > lun_req->result_len) {
3023 					lun_req->status = CTL_LUN_ERROR;
3024 					snprintf(lun_req->error_str,
3025 					    sizeof(lun_req->error_str),
3026 					    "Result nvlist too large.");
3027 					free(packed, M_NVLIST);
3028 					break;
3029 				}
3030 
3031 				if (copyout(packed, lun_req->result, packed_len)) {
3032 					lun_req->status = CTL_LUN_ERROR;
3033 					snprintf(lun_req->error_str,
3034 					    sizeof(lun_req->error_str),
3035 					    "Cannot copyout() the result.");
3036 					free(packed, M_NVLIST);
3037 					break;
3038 				}
3039 
3040 				lun_req->result_len = packed_len;
3041 				free(packed, M_NVLIST);
3042 			}
3043 
3044 			nvlist_destroy(lun_req->result_nvl);
3045 		}
3046 		break;
3047 	}
3048 	case CTL_LUN_LIST: {
3049 		struct sbuf *sb;
3050 		struct ctl_lun_list *list;
3051 		const char *name, *value;
3052 		void *cookie;
3053 		int type;
3054 
3055 		list = (struct ctl_lun_list *)addr;
3056 
3057 		/*
3058 		 * Allocate a fixed length sbuf here, based on the length
3059 		 * of the user's buffer.  We could allocate an auto-extending
3060 		 * buffer, and then tell the user how much larger our
3061 		 * amount of data is than his buffer, but that presents
3062 		 * some problems:
3063 		 *
3064 		 * 1.  The sbuf(9) routines use a blocking malloc, and so
3065 		 *     we can't hold a lock while calling them with an
3066 		 *     auto-extending buffer.
3067  		 *
3068 		 * 2.  There is not currently a LUN reference counting
3069 		 *     mechanism, outside of outstanding transactions on
3070 		 *     the LUN's OOA queue.  So a LUN could go away on us
3071 		 *     while we're getting the LUN number, backend-specific
3072 		 *     information, etc.  Thus, given the way things
3073 		 *     currently work, we need to hold the CTL lock while
3074 		 *     grabbing LUN information.
3075 		 *
3076 		 * So, from the user's standpoint, the best thing to do is
3077 		 * allocate what he thinks is a reasonable buffer length,
3078 		 * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
3079 		 * double the buffer length and try again.  (And repeat
3080 		 * that until he succeeds.)
3081 		 */
3082 		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3083 		if (sb == NULL) {
3084 			list->status = CTL_LUN_LIST_ERROR;
3085 			snprintf(list->error_str, sizeof(list->error_str),
3086 				 "Unable to allocate %d bytes for LUN list",
3087 				 list->alloc_len);
3088 			break;
3089 		}
3090 
3091 		sbuf_cat(sb, "<ctllunlist>\n");
3092 
3093 		mtx_lock(&softc->ctl_lock);
3094 		STAILQ_FOREACH(lun, &softc->lun_list, links) {
3095 			mtx_lock(&lun->lun_lock);
3096 			retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3097 					     (uintmax_t)lun->lun);
3098 
3099 			/*
3100 			 * Bail out as soon as we see that we've overfilled
3101 			 * the buffer.
3102 			 */
3103 			if (retval != 0)
3104 				break;
3105 
3106 			retval = sbuf_printf(sb, "\t<backend_type>%s"
3107 					     "</backend_type>\n",
3108 					     (lun->backend == NULL) ?  "none" :
3109 					     lun->backend->name);
3110 
3111 			if (retval != 0)
3112 				break;
3113 
3114 			retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
3115 					     lun->be_lun->lun_type);
3116 
3117 			if (retval != 0)
3118 				break;
3119 
3120 			if (lun->backend == NULL) {
3121 				retval = sbuf_cat(sb, "</lun>\n");
3122 				if (retval != 0)
3123 					break;
3124 				continue;
3125 			}
3126 
3127 			retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
3128 					     (lun->be_lun->maxlba > 0) ?
3129 					     lun->be_lun->maxlba + 1 : 0);
3130 
3131 			if (retval != 0)
3132 				break;
3133 
3134 			retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
3135 					     lun->be_lun->blocksize);
3136 
3137 			if (retval != 0)
3138 				break;
3139 
3140 			retval = sbuf_cat(sb, "\t<serial_number>");
3141 
3142 			if (retval != 0)
3143 				break;
3144 
3145 			retval = ctl_sbuf_printf_esc(sb,
3146 			    lun->be_lun->serial_num,
3147 			    sizeof(lun->be_lun->serial_num));
3148 
3149 			if (retval != 0)
3150 				break;
3151 
3152 			retval = sbuf_cat(sb, "</serial_number>\n");
3153 
3154 			if (retval != 0)
3155 				break;
3156 
3157 			retval = sbuf_cat(sb, "\t<device_id>");
3158 
3159 			if (retval != 0)
3160 				break;
3161 
3162 			retval = ctl_sbuf_printf_esc(sb,
3163 			    lun->be_lun->device_id,
3164 			    sizeof(lun->be_lun->device_id));
3165 
3166 			if (retval != 0)
3167 				break;
3168 
3169 			retval = sbuf_cat(sb, "</device_id>\n");
3170 
3171 			if (retval != 0)
3172 				break;
3173 
3174 			if (lun->backend->lun_info != NULL) {
3175 				retval = lun->backend->lun_info(lun->be_lun, sb);
3176 				if (retval != 0)
3177 					break;
3178 			}
3179 
3180 			cookie = NULL;
3181 			while ((name = nvlist_next(lun->be_lun->options, &type,
3182 			    &cookie)) != NULL) {
3183 				sbuf_printf(sb, "\t<%s>", name);
3184 
3185 				if (type == NV_TYPE_STRING) {
3186 					value = dnvlist_get_string(
3187 					    lun->be_lun->options, name, NULL);
3188 					if (value != NULL)
3189 						sbuf_cat(sb, value);
3190 				}
3191 
3192 				sbuf_printf(sb, "</%s>\n", name);
3193 			}
3194 
3195 			retval = sbuf_cat(sb, "</lun>\n");
3196 
3197 			if (retval != 0)
3198 				break;
3199 			mtx_unlock(&lun->lun_lock);
3200 		}
3201 		if (lun != NULL)
3202 			mtx_unlock(&lun->lun_lock);
3203 		mtx_unlock(&softc->ctl_lock);
3204 
3205 		if ((retval != 0)
3206 		 || ((retval = sbuf_cat(sb, "</ctllunlist>\n")) != 0)) {
3207 			retval = 0;
3208 			sbuf_delete(sb);
3209 			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3210 			snprintf(list->error_str, sizeof(list->error_str),
3211 				 "Out of space, %d bytes is too small",
3212 				 list->alloc_len);
3213 			break;
3214 		}
3215 
3216 		sbuf_finish(sb);
3217 
3218 		retval = copyout(sbuf_data(sb), list->lun_xml,
3219 				 sbuf_len(sb) + 1);
3220 
3221 		list->fill_len = sbuf_len(sb) + 1;
3222 		list->status = CTL_LUN_LIST_OK;
3223 		sbuf_delete(sb);
3224 		break;
3225 	}
3226 	case CTL_ISCSI: {
3227 		struct ctl_iscsi *ci;
3228 		struct ctl_frontend *fe;
3229 
3230 		ci = (struct ctl_iscsi *)addr;
3231 
3232 		fe = ctl_frontend_find("iscsi");
3233 		if (fe == NULL) {
3234 			ci->status = CTL_ISCSI_ERROR;
3235 			snprintf(ci->error_str, sizeof(ci->error_str),
3236 			    "Frontend \"iscsi\" not found.");
3237 			break;
3238 		}
3239 
3240 		retval = fe->ioctl(dev, cmd, addr, flag, td);
3241 		break;
3242 	}
3243 	case CTL_PORT_REQ: {
3244 		struct ctl_req *req;
3245 		struct ctl_frontend *fe;
3246 		void *packed;
3247 		nvlist_t *tmp_args_nvl;
3248 		size_t packed_len;
3249 
3250 		req = (struct ctl_req *)addr;
3251 		tmp_args_nvl = req->args_nvl;
3252 
3253 		fe = ctl_frontend_find(req->driver);
3254 		if (fe == NULL) {
3255 			req->status = CTL_LUN_ERROR;
3256 			snprintf(req->error_str, sizeof(req->error_str),
3257 			    "Frontend \"%s\" not found.", req->driver);
3258 			break;
3259 		}
3260 
3261 		if (req->args != NULL) {
3262 			if (req->args_len > CTL_MAX_ARGS_LEN) {
3263 				req->status = CTL_LUN_ERROR;
3264 				snprintf(req->error_str, sizeof(req->error_str),
3265 				    "Too big args.");
3266 				break;
3267 			}
3268 			packed = malloc(req->args_len, M_CTL, M_WAITOK);
3269 			if (copyin(req->args, packed, req->args_len) != 0) {
3270 				free(packed, M_CTL);
3271 				req->status = CTL_LUN_ERROR;
3272 				snprintf(req->error_str, sizeof(req->error_str),
3273 				    "Cannot copyin args.");
3274 				break;
3275 			}
3276 			req->args_nvl = nvlist_unpack(packed,
3277 			    req->args_len, 0);
3278 			free(packed, M_CTL);
3279 
3280 			if (req->args_nvl == NULL) {
3281 				req->status = CTL_LUN_ERROR;
3282 				snprintf(req->error_str, sizeof(req->error_str),
3283 				    "Cannot unpack args nvlist.");
3284 				break;
3285 			}
3286 		} else
3287 			req->args_nvl = nvlist_create(0);
3288 
3289 		req->result_nvl = NULL;
3290 		if (fe->ioctl)
3291 			retval = fe->ioctl(dev, cmd, addr, flag, td);
3292 		else
3293 			retval = ENODEV;
3294 
3295 		nvlist_destroy(req->args_nvl);
3296 		req->args_nvl = tmp_args_nvl;
3297 
3298 		if (req->result_nvl != NULL) {
3299 			if (req->result != NULL) {
3300 				packed = nvlist_pack(req->result_nvl,
3301 				    &packed_len);
3302 				if (packed == NULL) {
3303 					req->status = CTL_LUN_ERROR;
3304 					snprintf(req->error_str,
3305 					    sizeof(req->error_str),
3306 					    "Cannot pack result nvlist.");
3307 					break;
3308 				}
3309 
3310 				if (packed_len > req->result_len) {
3311 					req->status = CTL_LUN_ERROR;
3312 					snprintf(req->error_str,
3313 					    sizeof(req->error_str),
3314 					    "Result nvlist too large.");
3315 					free(packed, M_NVLIST);
3316 					break;
3317 				}
3318 
3319 				if (copyout(packed, req->result, packed_len)) {
3320 					req->status = CTL_LUN_ERROR;
3321 					snprintf(req->error_str,
3322 					    sizeof(req->error_str),
3323 					    "Cannot copyout() the result.");
3324 					free(packed, M_NVLIST);
3325 					break;
3326 				}
3327 
3328 				req->result_len = packed_len;
3329 				free(packed, M_NVLIST);
3330 			}
3331 
3332 			nvlist_destroy(req->result_nvl);
3333 		}
3334 		break;
3335 	}
3336 	case CTL_PORT_LIST: {
3337 		struct sbuf *sb;
3338 		struct ctl_port *port;
3339 		struct ctl_lun_list *list;
3340 		const char *name, *value;
3341 		void *cookie;
3342 		int j, type;
3343 		uint32_t plun;
3344 
3345 		list = (struct ctl_lun_list *)addr;
3346 
3347 		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3348 		if (sb == NULL) {
3349 			list->status = CTL_LUN_LIST_ERROR;
3350 			snprintf(list->error_str, sizeof(list->error_str),
3351 				 "Unable to allocate %d bytes for LUN list",
3352 				 list->alloc_len);
3353 			break;
3354 		}
3355 
3356 		sbuf_cat(sb, "<ctlportlist>\n");
3357 
3358 		mtx_lock(&softc->ctl_lock);
3359 		STAILQ_FOREACH(port, &softc->port_list, links) {
3360 			retval = sbuf_printf(sb, "<targ_port id=\"%ju\">\n",
3361 					     (uintmax_t)port->targ_port);
3362 
3363 			/*
3364 			 * Bail out as soon as we see that we've overfilled
3365 			 * the buffer.
3366 			 */
3367 			if (retval != 0)
3368 				break;
3369 
3370 			retval = sbuf_printf(sb, "\t<frontend_type>%s"
3371 			    "</frontend_type>\n", port->frontend->name);
3372 			if (retval != 0)
3373 				break;
3374 
3375 			retval = sbuf_printf(sb, "\t<port_type>%d</port_type>\n",
3376 					     port->port_type);
3377 			if (retval != 0)
3378 				break;
3379 
3380 			retval = sbuf_printf(sb, "\t<online>%s</online>\n",
3381 			    (port->status & CTL_PORT_STATUS_ONLINE) ? "YES" : "NO");
3382 			if (retval != 0)
3383 				break;
3384 
3385 			retval = sbuf_printf(sb, "\t<port_name>%s</port_name>\n",
3386 			    port->port_name);
3387 			if (retval != 0)
3388 				break;
3389 
3390 			retval = sbuf_printf(sb, "\t<physical_port>%d</physical_port>\n",
3391 			    port->physical_port);
3392 			if (retval != 0)
3393 				break;
3394 
3395 			retval = sbuf_printf(sb, "\t<virtual_port>%d</virtual_port>\n",
3396 			    port->virtual_port);
3397 			if (retval != 0)
3398 				break;
3399 
3400 			if (port->target_devid != NULL) {
3401 				sbuf_cat(sb, "\t<target>");
3402 				ctl_id_sbuf(port->target_devid, sb);
3403 				sbuf_cat(sb, "</target>\n");
3404 			}
3405 
3406 			if (port->port_devid != NULL) {
3407 				sbuf_cat(sb, "\t<port>");
3408 				ctl_id_sbuf(port->port_devid, sb);
3409 				sbuf_cat(sb, "</port>\n");
3410 			}
3411 
3412 			if (port->port_info != NULL) {
3413 				retval = port->port_info(port->onoff_arg, sb);
3414 				if (retval != 0)
3415 					break;
3416 			}
3417 
3418 			cookie = NULL;
3419 			while ((name = nvlist_next(port->options, &type,
3420 			    &cookie)) != NULL) {
3421 				sbuf_printf(sb, "\t<%s>", name);
3422 
3423 				if (type == NV_TYPE_STRING) {
3424 					value = dnvlist_get_string(port->options,
3425 					    name, NULL);
3426 					if (value != NULL)
3427 						sbuf_printf(sb, "%s", value);
3428 				}
3429 
3430 				sbuf_printf(sb, "</%s>\n", name);
3431 			}
3432 
3433 			if (port->lun_map != NULL) {
3434 				sbuf_cat(sb, "\t<lun_map>on</lun_map>\n");
3435 				for (j = 0; j < port->lun_map_size; j++) {
3436 					plun = ctl_lun_map_from_port(port, j);
3437 					if (plun == UINT32_MAX)
3438 						continue;
3439 					sbuf_printf(sb,
3440 					    "\t<lun id=\"%u\">%u</lun>\n",
3441 					    j, plun);
3442 				}
3443 			}
3444 
3445 			for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3446 				if (port->wwpn_iid[j].in_use == 0 ||
3447 				    (port->wwpn_iid[j].wwpn == 0 &&
3448 				     port->wwpn_iid[j].name == NULL))
3449 					continue;
3450 
3451 				if (port->wwpn_iid[j].name != NULL)
3452 					retval = sbuf_printf(sb,
3453 					    "\t<initiator id=\"%u\">%s</initiator>\n",
3454 					    j, port->wwpn_iid[j].name);
3455 				else
3456 					retval = sbuf_printf(sb,
3457 					    "\t<initiator id=\"%u\">naa.%08jx</initiator>\n",
3458 					    j, port->wwpn_iid[j].wwpn);
3459 				if (retval != 0)
3460 					break;
3461 			}
3462 			if (retval != 0)
3463 				break;
3464 
3465 			retval = sbuf_cat(sb, "</targ_port>\n");
3466 			if (retval != 0)
3467 				break;
3468 		}
3469 		mtx_unlock(&softc->ctl_lock);
3470 
3471 		if ((retval != 0)
3472 		 || ((retval = sbuf_cat(sb, "</ctlportlist>\n")) != 0)) {
3473 			retval = 0;
3474 			sbuf_delete(sb);
3475 			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3476 			snprintf(list->error_str, sizeof(list->error_str),
3477 				 "Out of space, %d bytes is too small",
3478 				 list->alloc_len);
3479 			break;
3480 		}
3481 
3482 		sbuf_finish(sb);
3483 
3484 		retval = copyout(sbuf_data(sb), list->lun_xml,
3485 				 sbuf_len(sb) + 1);
3486 
3487 		list->fill_len = sbuf_len(sb) + 1;
3488 		list->status = CTL_LUN_LIST_OK;
3489 		sbuf_delete(sb);
3490 		break;
3491 	}
3492 	case CTL_LUN_MAP: {
3493 		struct ctl_lun_map *lm  = (struct ctl_lun_map *)addr;
3494 		struct ctl_port *port;
3495 
3496 		mtx_lock(&softc->ctl_lock);
3497 		if (lm->port < softc->port_min ||
3498 		    lm->port >= softc->port_max ||
3499 		    (port = softc->ctl_ports[lm->port]) == NULL) {
3500 			mtx_unlock(&softc->ctl_lock);
3501 			return (ENXIO);
3502 		}
3503 		if (port->status & CTL_PORT_STATUS_ONLINE) {
3504 			STAILQ_FOREACH(lun, &softc->lun_list, links) {
3505 				if (ctl_lun_map_to_port(port, lun->lun) ==
3506 				    UINT32_MAX)
3507 					continue;
3508 				mtx_lock(&lun->lun_lock);
3509 				ctl_est_ua_port(lun, lm->port, -1,
3510 				    CTL_UA_LUN_CHANGE);
3511 				mtx_unlock(&lun->lun_lock);
3512 			}
3513 		}
3514 		mtx_unlock(&softc->ctl_lock); // XXX: port_enable sleeps
3515 		if (lm->plun != UINT32_MAX) {
3516 			if (lm->lun == UINT32_MAX)
3517 				retval = ctl_lun_map_unset(port, lm->plun);
3518 			else if (lm->lun < ctl_max_luns &&
3519 			    softc->ctl_luns[lm->lun] != NULL)
3520 				retval = ctl_lun_map_set(port, lm->plun, lm->lun);
3521 			else
3522 				return (ENXIO);
3523 		} else {
3524 			if (lm->lun == UINT32_MAX)
3525 				retval = ctl_lun_map_deinit(port);
3526 			else
3527 				retval = ctl_lun_map_init(port);
3528 		}
3529 		if (port->status & CTL_PORT_STATUS_ONLINE)
3530 			ctl_isc_announce_port(port);
3531 		break;
3532 	}
3533 	case CTL_GET_LUN_STATS: {
3534 		struct ctl_get_io_stats *stats = (struct ctl_get_io_stats *)addr;
3535 		int i;
3536 
3537 		/*
3538 		 * XXX KDM no locking here.  If the LUN list changes,
3539 		 * things can blow up.
3540 		 */
3541 		i = 0;
3542 		stats->status = CTL_SS_OK;
3543 		stats->fill_len = 0;
3544 		STAILQ_FOREACH(lun, &softc->lun_list, links) {
3545 			if (lun->lun < stats->first_item)
3546 				continue;
3547 			if (stats->fill_len + sizeof(lun->stats) >
3548 			    stats->alloc_len) {
3549 				stats->status = CTL_SS_NEED_MORE_SPACE;
3550 				break;
3551 			}
3552 			retval = copyout(&lun->stats, &stats->stats[i++],
3553 					 sizeof(lun->stats));
3554 			if (retval != 0)
3555 				break;
3556 			stats->fill_len += sizeof(lun->stats);
3557 		}
3558 		stats->num_items = softc->num_luns;
3559 		stats->flags = CTL_STATS_FLAG_NONE;
3560 #ifdef CTL_TIME_IO
3561 		stats->flags |= CTL_STATS_FLAG_TIME_VALID;
3562 #endif
3563 		getnanouptime(&stats->timestamp);
3564 		break;
3565 	}
3566 	case CTL_GET_PORT_STATS: {
3567 		struct ctl_get_io_stats *stats = (struct ctl_get_io_stats *)addr;
3568 		int i;
3569 
3570 		/*
3571 		 * XXX KDM no locking here.  If the LUN list changes,
3572 		 * things can blow up.
3573 		 */
3574 		i = 0;
3575 		stats->status = CTL_SS_OK;
3576 		stats->fill_len = 0;
3577 		STAILQ_FOREACH(port, &softc->port_list, links) {
3578 			if (port->targ_port < stats->first_item)
3579 				continue;
3580 			if (stats->fill_len + sizeof(port->stats) >
3581 			    stats->alloc_len) {
3582 				stats->status = CTL_SS_NEED_MORE_SPACE;
3583 				break;
3584 			}
3585 			retval = copyout(&port->stats, &stats->stats[i++],
3586 					 sizeof(port->stats));
3587 			if (retval != 0)
3588 				break;
3589 			stats->fill_len += sizeof(port->stats);
3590 		}
3591 		stats->num_items = softc->num_ports;
3592 		stats->flags = CTL_STATS_FLAG_NONE;
3593 #ifdef CTL_TIME_IO
3594 		stats->flags |= CTL_STATS_FLAG_TIME_VALID;
3595 #endif
3596 		getnanouptime(&stats->timestamp);
3597 		break;
3598 	}
3599 	default: {
3600 		/* XXX KDM should we fix this? */
3601 #if 0
3602 		struct ctl_backend_driver *backend;
3603 		unsigned int type;
3604 		int found;
3605 
3606 		found = 0;
3607 
3608 		/*
3609 		 * We encode the backend type as the ioctl type for backend
3610 		 * ioctls.  So parse it out here, and then search for a
3611 		 * backend of this type.
3612 		 */
3613 		type = _IOC_TYPE(cmd);
3614 
3615 		STAILQ_FOREACH(backend, &softc->be_list, links) {
3616 			if (backend->type == type) {
3617 				found = 1;
3618 				break;
3619 			}
3620 		}
3621 		if (found == 0) {
3622 			printf("ctl: unknown ioctl command %#lx or backend "
3623 			       "%d\n", cmd, type);
3624 			retval = EINVAL;
3625 			break;
3626 		}
3627 		retval = backend->ioctl(dev, cmd, addr, flag, td);
3628 #endif
3629 		retval = ENOTTY;
3630 		break;
3631 	}
3632 	}
3633 	return (retval);
3634 }
3635 
3636 uint32_t
3637 ctl_get_initindex(struct ctl_nexus *nexus)
3638 {
3639 	return (nexus->initid + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3640 }
3641 
3642 int
3643 ctl_lun_map_init(struct ctl_port *port)
3644 {
3645 	struct ctl_softc *softc = port->ctl_softc;
3646 	struct ctl_lun *lun;
3647 	int size = ctl_lun_map_size;
3648 	uint32_t i;
3649 
3650 	if (port->lun_map == NULL || port->lun_map_size < size) {
3651 		port->lun_map_size = 0;
3652 		free(port->lun_map, M_CTL);
3653 		port->lun_map = malloc(size * sizeof(uint32_t),
3654 		    M_CTL, M_NOWAIT);
3655 	}
3656 	if (port->lun_map == NULL)
3657 		return (ENOMEM);
3658 	for (i = 0; i < size; i++)
3659 		port->lun_map[i] = UINT32_MAX;
3660 	port->lun_map_size = size;
3661 	if (port->status & CTL_PORT_STATUS_ONLINE) {
3662 		if (port->lun_disable != NULL) {
3663 			STAILQ_FOREACH(lun, &softc->lun_list, links)
3664 				port->lun_disable(port->targ_lun_arg, lun->lun);
3665 		}
3666 		ctl_isc_announce_port(port);
3667 	}
3668 	return (0);
3669 }
3670 
3671 int
3672 ctl_lun_map_deinit(struct ctl_port *port)
3673 {
3674 	struct ctl_softc *softc = port->ctl_softc;
3675 	struct ctl_lun *lun;
3676 
3677 	if (port->lun_map == NULL)
3678 		return (0);
3679 	port->lun_map_size = 0;
3680 	free(port->lun_map, M_CTL);
3681 	port->lun_map = NULL;
3682 	if (port->status & CTL_PORT_STATUS_ONLINE) {
3683 		if (port->lun_enable != NULL) {
3684 			STAILQ_FOREACH(lun, &softc->lun_list, links)
3685 				port->lun_enable(port->targ_lun_arg, lun->lun);
3686 		}
3687 		ctl_isc_announce_port(port);
3688 	}
3689 	return (0);
3690 }
3691 
3692 int
3693 ctl_lun_map_set(struct ctl_port *port, uint32_t plun, uint32_t glun)
3694 {
3695 	int status;
3696 	uint32_t old;
3697 
3698 	if (port->lun_map == NULL) {
3699 		status = ctl_lun_map_init(port);
3700 		if (status != 0)
3701 			return (status);
3702 	}
3703 	if (plun >= port->lun_map_size)
3704 		return (EINVAL);
3705 	old = port->lun_map[plun];
3706 	port->lun_map[plun] = glun;
3707 	if ((port->status & CTL_PORT_STATUS_ONLINE) && old == UINT32_MAX) {
3708 		if (port->lun_enable != NULL)
3709 			port->lun_enable(port->targ_lun_arg, plun);
3710 		ctl_isc_announce_port(port);
3711 	}
3712 	return (0);
3713 }
3714 
3715 int
3716 ctl_lun_map_unset(struct ctl_port *port, uint32_t plun)
3717 {
3718 	uint32_t old;
3719 
3720 	if (port->lun_map == NULL || plun >= port->lun_map_size)
3721 		return (0);
3722 	old = port->lun_map[plun];
3723 	port->lun_map[plun] = UINT32_MAX;
3724 	if ((port->status & CTL_PORT_STATUS_ONLINE) && old != UINT32_MAX) {
3725 		if (port->lun_disable != NULL)
3726 			port->lun_disable(port->targ_lun_arg, plun);
3727 		ctl_isc_announce_port(port);
3728 	}
3729 	return (0);
3730 }
3731 
3732 uint32_t
3733 ctl_lun_map_from_port(struct ctl_port *port, uint32_t lun_id)
3734 {
3735 
3736 	if (port == NULL)
3737 		return (UINT32_MAX);
3738 	if (port->lun_map == NULL)
3739 		return (lun_id);
3740 	if (lun_id > port->lun_map_size)
3741 		return (UINT32_MAX);
3742 	return (port->lun_map[lun_id]);
3743 }
3744 
3745 uint32_t
3746 ctl_lun_map_to_port(struct ctl_port *port, uint32_t lun_id)
3747 {
3748 	uint32_t i;
3749 
3750 	if (port == NULL)
3751 		return (UINT32_MAX);
3752 	if (port->lun_map == NULL)
3753 		return (lun_id);
3754 	for (i = 0; i < port->lun_map_size; i++) {
3755 		if (port->lun_map[i] == lun_id)
3756 			return (i);
3757 	}
3758 	return (UINT32_MAX);
3759 }
3760 
3761 uint32_t
3762 ctl_decode_lun(uint64_t encoded)
3763 {
3764 	uint8_t lun[8];
3765 	uint32_t result = 0xffffffff;
3766 
3767 	be64enc(lun, encoded);
3768 	switch (lun[0] & RPL_LUNDATA_ATYP_MASK) {
3769 	case RPL_LUNDATA_ATYP_PERIPH:
3770 		if ((lun[0] & 0x3f) == 0 && lun[2] == 0 && lun[3] == 0 &&
3771 		    lun[4] == 0 && lun[5] == 0 && lun[6] == 0 && lun[7] == 0)
3772 			result = lun[1];
3773 		break;
3774 	case RPL_LUNDATA_ATYP_FLAT:
3775 		if (lun[2] == 0 && lun[3] == 0 && lun[4] == 0 && lun[5] == 0 &&
3776 		    lun[6] == 0 && lun[7] == 0)
3777 			result = ((lun[0] & 0x3f) << 8) + lun[1];
3778 		break;
3779 	case RPL_LUNDATA_ATYP_EXTLUN:
3780 		switch (lun[0] & RPL_LUNDATA_EXT_EAM_MASK) {
3781 		case 0x02:
3782 			switch (lun[0] & RPL_LUNDATA_EXT_LEN_MASK) {
3783 			case 0x00:
3784 				result = lun[1];
3785 				break;
3786 			case 0x10:
3787 				result = (lun[1] << 16) + (lun[2] << 8) +
3788 				    lun[3];
3789 				break;
3790 			case 0x20:
3791 				if (lun[1] == 0 && lun[6] == 0 && lun[7] == 0)
3792 					result = (lun[2] << 24) +
3793 					    (lun[3] << 16) + (lun[4] << 8) +
3794 					    lun[5];
3795 				break;
3796 			}
3797 			break;
3798 		case RPL_LUNDATA_EXT_EAM_NOT_SPEC:
3799 			result = 0xffffffff;
3800 			break;
3801 		}
3802 		break;
3803 	}
3804 	return (result);
3805 }
3806 
3807 uint64_t
3808 ctl_encode_lun(uint32_t decoded)
3809 {
3810 	uint64_t l = decoded;
3811 
3812 	if (l <= 0xff)
3813 		return (((uint64_t)RPL_LUNDATA_ATYP_PERIPH << 56) | (l << 48));
3814 	if (l <= 0x3fff)
3815 		return (((uint64_t)RPL_LUNDATA_ATYP_FLAT << 56) | (l << 48));
3816 	if (l <= 0xffffff)
3817 		return (((uint64_t)(RPL_LUNDATA_ATYP_EXTLUN | 0x12) << 56) |
3818 		    (l << 32));
3819 	return ((((uint64_t)RPL_LUNDATA_ATYP_EXTLUN | 0x22) << 56) | (l << 16));
3820 }
3821 
3822 int
3823 ctl_ffz(uint32_t *mask, uint32_t first, uint32_t last)
3824 {
3825 	int i;
3826 
3827 	for (i = first; i < last; i++) {
3828 		if ((mask[i / 32] & (1 << (i % 32))) == 0)
3829 			return (i);
3830 	}
3831 	return (-1);
3832 }
3833 
3834 int
3835 ctl_set_mask(uint32_t *mask, uint32_t bit)
3836 {
3837 	uint32_t chunk, piece;
3838 
3839 	chunk = bit >> 5;
3840 	piece = bit % (sizeof(uint32_t) * 8);
3841 
3842 	if ((mask[chunk] & (1 << piece)) != 0)
3843 		return (-1);
3844 	else
3845 		mask[chunk] |= (1 << piece);
3846 
3847 	return (0);
3848 }
3849 
3850 int
3851 ctl_clear_mask(uint32_t *mask, uint32_t bit)
3852 {
3853 	uint32_t chunk, piece;
3854 
3855 	chunk = bit >> 5;
3856 	piece = bit % (sizeof(uint32_t) * 8);
3857 
3858 	if ((mask[chunk] & (1 << piece)) == 0)
3859 		return (-1);
3860 	else
3861 		mask[chunk] &= ~(1 << piece);
3862 
3863 	return (0);
3864 }
3865 
3866 int
3867 ctl_is_set(uint32_t *mask, uint32_t bit)
3868 {
3869 	uint32_t chunk, piece;
3870 
3871 	chunk = bit >> 5;
3872 	piece = bit % (sizeof(uint32_t) * 8);
3873 
3874 	if ((mask[chunk] & (1 << piece)) == 0)
3875 		return (0);
3876 	else
3877 		return (1);
3878 }
3879 
3880 static uint64_t
3881 ctl_get_prkey(struct ctl_lun *lun, uint32_t residx)
3882 {
3883 	uint64_t *t;
3884 
3885 	t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3886 	if (t == NULL)
3887 		return (0);
3888 	return (t[residx % CTL_MAX_INIT_PER_PORT]);
3889 }
3890 
3891 static void
3892 ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx)
3893 {
3894 	uint64_t *t;
3895 
3896 	t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3897 	if (t == NULL)
3898 		return;
3899 	t[residx % CTL_MAX_INIT_PER_PORT] = 0;
3900 }
3901 
3902 static void
3903 ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx)
3904 {
3905 	uint64_t *p;
3906 	u_int i;
3907 
3908 	i = residx/CTL_MAX_INIT_PER_PORT;
3909 	if (lun->pr_keys[i] != NULL)
3910 		return;
3911 	mtx_unlock(&lun->lun_lock);
3912 	p = malloc(sizeof(uint64_t) * CTL_MAX_INIT_PER_PORT, M_CTL,
3913 	    M_WAITOK | M_ZERO);
3914 	mtx_lock(&lun->lun_lock);
3915 	if (lun->pr_keys[i] == NULL)
3916 		lun->pr_keys[i] = p;
3917 	else
3918 		free(p, M_CTL);
3919 }
3920 
3921 static void
3922 ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key)
3923 {
3924 	uint64_t *t;
3925 
3926 	t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3927 	KASSERT(t != NULL, ("prkey %d is not allocated", residx));
3928 	t[residx % CTL_MAX_INIT_PER_PORT] = key;
3929 }
3930 
3931 /*
3932  * ctl_softc, pool_name, total_ctl_io are passed in.
3933  * npool is passed out.
3934  */
3935 int
3936 ctl_pool_create(struct ctl_softc *ctl_softc, const char *pool_name,
3937 		uint32_t total_ctl_io, void **npool)
3938 {
3939 	struct ctl_io_pool *pool;
3940 
3941 	pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3942 					    M_NOWAIT | M_ZERO);
3943 	if (pool == NULL)
3944 		return (ENOMEM);
3945 
3946 	snprintf(pool->name, sizeof(pool->name), "CTL IO %s", pool_name);
3947 	pool->ctl_softc = ctl_softc;
3948 #ifdef IO_POOLS
3949 	pool->zone = uma_zsecond_create(pool->name, NULL,
3950 	    NULL, NULL, NULL, ctl_softc->io_zone);
3951 	/* uma_prealloc(pool->zone, total_ctl_io); */
3952 #else
3953 	pool->zone = ctl_softc->io_zone;
3954 #endif
3955 
3956 	*npool = pool;
3957 	return (0);
3958 }
3959 
3960 void
3961 ctl_pool_free(struct ctl_io_pool *pool)
3962 {
3963 
3964 	if (pool == NULL)
3965 		return;
3966 
3967 #ifdef IO_POOLS
3968 	uma_zdestroy(pool->zone);
3969 #endif
3970 	free(pool, M_CTL);
3971 }
3972 
3973 union ctl_io *
3974 ctl_alloc_io(void *pool_ref)
3975 {
3976 	struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3977 	union ctl_io *io;
3978 
3979 	io = uma_zalloc(pool->zone, M_WAITOK);
3980 	if (io != NULL) {
3981 		io->io_hdr.pool = pool_ref;
3982 		CTL_SOFTC(io) = pool->ctl_softc;
3983 		TAILQ_INIT(&io->io_hdr.blocked_queue);
3984 	}
3985 	return (io);
3986 }
3987 
3988 union ctl_io *
3989 ctl_alloc_io_nowait(void *pool_ref)
3990 {
3991 	struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3992 	union ctl_io *io;
3993 
3994 	io = uma_zalloc(pool->zone, M_NOWAIT);
3995 	if (io != NULL) {
3996 		io->io_hdr.pool = pool_ref;
3997 		CTL_SOFTC(io) = pool->ctl_softc;
3998 		TAILQ_INIT(&io->io_hdr.blocked_queue);
3999 	}
4000 	return (io);
4001 }
4002 
4003 void
4004 ctl_free_io(union ctl_io *io)
4005 {
4006 	struct ctl_io_pool *pool;
4007 
4008 	if (io == NULL)
4009 		return;
4010 
4011 	pool = (struct ctl_io_pool *)io->io_hdr.pool;
4012 	uma_zfree(pool->zone, io);
4013 }
4014 
4015 void
4016 ctl_zero_io(union ctl_io *io)
4017 {
4018 	struct ctl_io_pool *pool;
4019 
4020 	if (io == NULL)
4021 		return;
4022 
4023 	/*
4024 	 * May need to preserve linked list pointers at some point too.
4025 	 */
4026 	pool = io->io_hdr.pool;
4027 	memset(io, 0, sizeof(*io));
4028 	io->io_hdr.pool = pool;
4029 	CTL_SOFTC(io) = pool->ctl_softc;
4030 	TAILQ_INIT(&io->io_hdr.blocked_queue);
4031 }
4032 
4033 int
4034 ctl_expand_number(const char *buf, uint64_t *num)
4035 {
4036 	char *endptr;
4037 	uint64_t number;
4038 	unsigned shift;
4039 
4040 	number = strtoq(buf, &endptr, 0);
4041 
4042 	switch (tolower((unsigned char)*endptr)) {
4043 	case 'e':
4044 		shift = 60;
4045 		break;
4046 	case 'p':
4047 		shift = 50;
4048 		break;
4049 	case 't':
4050 		shift = 40;
4051 		break;
4052 	case 'g':
4053 		shift = 30;
4054 		break;
4055 	case 'm':
4056 		shift = 20;
4057 		break;
4058 	case 'k':
4059 		shift = 10;
4060 		break;
4061 	case 'b':
4062 	case '\0': /* No unit. */
4063 		*num = number;
4064 		return (0);
4065 	default:
4066 		/* Unrecognized unit. */
4067 		return (-1);
4068 	}
4069 
4070 	if ((number << shift) >> shift != number) {
4071 		/* Overflow */
4072 		return (-1);
4073 	}
4074 	*num = number << shift;
4075 	return (0);
4076 }
4077 
4078 /*
4079  * This routine could be used in the future to load default and/or saved
4080  * mode page parameters for a particuar lun.
4081  */
4082 static int
4083 ctl_init_page_index(struct ctl_lun *lun)
4084 {
4085 	int i, page_code;
4086 	struct ctl_page_index *page_index;
4087 	const char *value;
4088 	uint64_t ival;
4089 
4090 	memcpy(&lun->mode_pages.index, page_index_template,
4091 	       sizeof(page_index_template));
4092 
4093 	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
4094 		page_index = &lun->mode_pages.index[i];
4095 		if (lun->be_lun->lun_type == T_DIRECT &&
4096 		    (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
4097 			continue;
4098 		if (lun->be_lun->lun_type == T_PROCESSOR &&
4099 		    (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
4100 			continue;
4101 		if (lun->be_lun->lun_type == T_CDROM &&
4102 		    (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
4103 			continue;
4104 
4105 		page_code = page_index->page_code & SMPH_PC_MASK;
4106 		switch (page_code) {
4107 		case SMS_RW_ERROR_RECOVERY_PAGE: {
4108 			KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4109 			    ("subpage %#x for page %#x is incorrect!",
4110 			    page_index->subpage, page_code));
4111 			memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT],
4112 			       &rw_er_page_default,
4113 			       sizeof(rw_er_page_default));
4114 			memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CHANGEABLE],
4115 			       &rw_er_page_changeable,
4116 			       sizeof(rw_er_page_changeable));
4117 			memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_DEFAULT],
4118 			       &rw_er_page_default,
4119 			       sizeof(rw_er_page_default));
4120 			memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_SAVED],
4121 			       &rw_er_page_default,
4122 			       sizeof(rw_er_page_default));
4123 			page_index->page_data =
4124 				(uint8_t *)lun->mode_pages.rw_er_page;
4125 			break;
4126 		}
4127 		case SMS_FORMAT_DEVICE_PAGE: {
4128 			struct scsi_format_page *format_page;
4129 
4130 			KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4131 			    ("subpage %#x for page %#x is incorrect!",
4132 			    page_index->subpage, page_code));
4133 
4134 			/*
4135 			 * Sectors per track are set above.  Bytes per
4136 			 * sector need to be set here on a per-LUN basis.
4137 			 */
4138 			memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
4139 			       &format_page_default,
4140 			       sizeof(format_page_default));
4141 			memcpy(&lun->mode_pages.format_page[
4142 			       CTL_PAGE_CHANGEABLE], &format_page_changeable,
4143 			       sizeof(format_page_changeable));
4144 			memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
4145 			       &format_page_default,
4146 			       sizeof(format_page_default));
4147 			memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
4148 			       &format_page_default,
4149 			       sizeof(format_page_default));
4150 
4151 			format_page = &lun->mode_pages.format_page[
4152 				CTL_PAGE_CURRENT];
4153 			scsi_ulto2b(lun->be_lun->blocksize,
4154 				    format_page->bytes_per_sector);
4155 
4156 			format_page = &lun->mode_pages.format_page[
4157 				CTL_PAGE_DEFAULT];
4158 			scsi_ulto2b(lun->be_lun->blocksize,
4159 				    format_page->bytes_per_sector);
4160 
4161 			format_page = &lun->mode_pages.format_page[
4162 				CTL_PAGE_SAVED];
4163 			scsi_ulto2b(lun->be_lun->blocksize,
4164 				    format_page->bytes_per_sector);
4165 
4166 			page_index->page_data =
4167 				(uint8_t *)lun->mode_pages.format_page;
4168 			break;
4169 		}
4170 		case SMS_RIGID_DISK_PAGE: {
4171 			struct scsi_rigid_disk_page *rigid_disk_page;
4172 			uint32_t sectors_per_cylinder;
4173 			uint64_t cylinders;
4174 #ifndef	__XSCALE__
4175 			int shift;
4176 #endif /* !__XSCALE__ */
4177 
4178 			KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4179 			    ("subpage %#x for page %#x is incorrect!",
4180 			    page_index->subpage, page_code));
4181 
4182 			/*
4183 			 * Rotation rate and sectors per track are set
4184 			 * above.  We calculate the cylinders here based on
4185 			 * capacity.  Due to the number of heads and
4186 			 * sectors per track we're using, smaller arrays
4187 			 * may turn out to have 0 cylinders.  Linux and
4188 			 * FreeBSD don't pay attention to these mode pages
4189 			 * to figure out capacity, but Solaris does.  It
4190 			 * seems to deal with 0 cylinders just fine, and
4191 			 * works out a fake geometry based on the capacity.
4192 			 */
4193 			memcpy(&lun->mode_pages.rigid_disk_page[
4194 			       CTL_PAGE_DEFAULT], &rigid_disk_page_default,
4195 			       sizeof(rigid_disk_page_default));
4196 			memcpy(&lun->mode_pages.rigid_disk_page[
4197 			       CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
4198 			       sizeof(rigid_disk_page_changeable));
4199 
4200 			sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
4201 				CTL_DEFAULT_HEADS;
4202 
4203 			/*
4204 			 * The divide method here will be more accurate,
4205 			 * probably, but results in floating point being
4206 			 * used in the kernel on i386 (__udivdi3()).  On the
4207 			 * XScale, though, __udivdi3() is implemented in
4208 			 * software.
4209 			 *
4210 			 * The shift method for cylinder calculation is
4211 			 * accurate if sectors_per_cylinder is a power of
4212 			 * 2.  Otherwise it might be slightly off -- you
4213 			 * might have a bit of a truncation problem.
4214 			 */
4215 #ifdef	__XSCALE__
4216 			cylinders = (lun->be_lun->maxlba + 1) /
4217 				sectors_per_cylinder;
4218 #else
4219 			for (shift = 31; shift > 0; shift--) {
4220 				if (sectors_per_cylinder & (1 << shift))
4221 					break;
4222 			}
4223 			cylinders = (lun->be_lun->maxlba + 1) >> shift;
4224 #endif
4225 
4226 			/*
4227 			 * We've basically got 3 bytes, or 24 bits for the
4228 			 * cylinder size in the mode page.  If we're over,
4229 			 * just round down to 2^24.
4230 			 */
4231 			if (cylinders > 0xffffff)
4232 				cylinders = 0xffffff;
4233 
4234 			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4235 				CTL_PAGE_DEFAULT];
4236 			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4237 
4238 			if ((value = dnvlist_get_string(lun->be_lun->options,
4239 			    "rpm", NULL)) != NULL) {
4240 				scsi_ulto2b(strtol(value, NULL, 0),
4241 				     rigid_disk_page->rotation_rate);
4242 			}
4243 
4244 			memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_CURRENT],
4245 			       &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
4246 			       sizeof(rigid_disk_page_default));
4247 			memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_SAVED],
4248 			       &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
4249 			       sizeof(rigid_disk_page_default));
4250 
4251 			page_index->page_data =
4252 				(uint8_t *)lun->mode_pages.rigid_disk_page;
4253 			break;
4254 		}
4255 		case SMS_VERIFY_ERROR_RECOVERY_PAGE: {
4256 			KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4257 			    ("subpage %#x for page %#x is incorrect!",
4258 			    page_index->subpage, page_code));
4259 			memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_CURRENT],
4260 			       &verify_er_page_default,
4261 			       sizeof(verify_er_page_default));
4262 			memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_CHANGEABLE],
4263 			       &verify_er_page_changeable,
4264 			       sizeof(verify_er_page_changeable));
4265 			memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_DEFAULT],
4266 			       &verify_er_page_default,
4267 			       sizeof(verify_er_page_default));
4268 			memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_SAVED],
4269 			       &verify_er_page_default,
4270 			       sizeof(verify_er_page_default));
4271 			page_index->page_data =
4272 				(uint8_t *)lun->mode_pages.verify_er_page;
4273 			break;
4274 		}
4275 		case SMS_CACHING_PAGE: {
4276 			struct scsi_caching_page *caching_page;
4277 
4278 			KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4279 			    ("subpage %#x for page %#x is incorrect!",
4280 			    page_index->subpage, page_code));
4281 			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4282 			       &caching_page_default,
4283 			       sizeof(caching_page_default));
4284 			memcpy(&lun->mode_pages.caching_page[
4285 			       CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4286 			       sizeof(caching_page_changeable));
4287 			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4288 			       &caching_page_default,
4289 			       sizeof(caching_page_default));
4290 			caching_page = &lun->mode_pages.caching_page[
4291 			    CTL_PAGE_SAVED];
4292 			value = dnvlist_get_string(lun->be_lun->options,
4293 			    "writecache", NULL);
4294 			if (value != NULL && strcmp(value, "off") == 0)
4295 				caching_page->flags1 &= ~SCP_WCE;
4296 			value = dnvlist_get_string(lun->be_lun->options,
4297 			    "readcache", NULL);
4298 			if (value != NULL && strcmp(value, "off") == 0)
4299 				caching_page->flags1 |= SCP_RCD;
4300 			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4301 			       &lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4302 			       sizeof(caching_page_default));
4303 			page_index->page_data =
4304 				(uint8_t *)lun->mode_pages.caching_page;
4305 			break;
4306 		}
4307 		case SMS_CONTROL_MODE_PAGE: {
4308 			switch (page_index->subpage) {
4309 			case SMS_SUBPAGE_PAGE_0: {
4310 				struct scsi_control_page *control_page;
4311 
4312 				memcpy(&lun->mode_pages.control_page[
4313 				    CTL_PAGE_DEFAULT],
4314 				       &control_page_default,
4315 				       sizeof(control_page_default));
4316 				memcpy(&lun->mode_pages.control_page[
4317 				    CTL_PAGE_CHANGEABLE],
4318 				       &control_page_changeable,
4319 				       sizeof(control_page_changeable));
4320 				memcpy(&lun->mode_pages.control_page[
4321 				    CTL_PAGE_SAVED],
4322 				       &control_page_default,
4323 				       sizeof(control_page_default));
4324 				control_page = &lun->mode_pages.control_page[
4325 				    CTL_PAGE_SAVED];
4326 				value = dnvlist_get_string(lun->be_lun->options,
4327 				    "reordering", NULL);
4328 				if (value != NULL &&
4329 				    strcmp(value, "unrestricted") == 0) {
4330 					control_page->queue_flags &=
4331 					    ~SCP_QUEUE_ALG_MASK;
4332 					control_page->queue_flags |=
4333 					    SCP_QUEUE_ALG_UNRESTRICTED;
4334 				}
4335 				memcpy(&lun->mode_pages.control_page[
4336 				    CTL_PAGE_CURRENT],
4337 				       &lun->mode_pages.control_page[
4338 				    CTL_PAGE_SAVED],
4339 				       sizeof(control_page_default));
4340 				page_index->page_data =
4341 				    (uint8_t *)lun->mode_pages.control_page;
4342 				break;
4343 			}
4344 			case 0x01:
4345 				memcpy(&lun->mode_pages.control_ext_page[
4346 				    CTL_PAGE_DEFAULT],
4347 				       &control_ext_page_default,
4348 				       sizeof(control_ext_page_default));
4349 				memcpy(&lun->mode_pages.control_ext_page[
4350 				    CTL_PAGE_CHANGEABLE],
4351 				       &control_ext_page_changeable,
4352 				       sizeof(control_ext_page_changeable));
4353 				memcpy(&lun->mode_pages.control_ext_page[
4354 				    CTL_PAGE_SAVED],
4355 				       &control_ext_page_default,
4356 				       sizeof(control_ext_page_default));
4357 				memcpy(&lun->mode_pages.control_ext_page[
4358 				    CTL_PAGE_CURRENT],
4359 				       &lun->mode_pages.control_ext_page[
4360 				    CTL_PAGE_SAVED],
4361 				       sizeof(control_ext_page_default));
4362 				page_index->page_data =
4363 				    (uint8_t *)lun->mode_pages.control_ext_page;
4364 				break;
4365 			default:
4366 				panic("subpage %#x for page %#x is incorrect!",
4367 				      page_index->subpage, page_code);
4368 			}
4369 			break;
4370 		}
4371 		case SMS_INFO_EXCEPTIONS_PAGE: {
4372 			switch (page_index->subpage) {
4373 			case SMS_SUBPAGE_PAGE_0:
4374 				memcpy(&lun->mode_pages.ie_page[CTL_PAGE_CURRENT],
4375 				       &ie_page_default,
4376 				       sizeof(ie_page_default));
4377 				memcpy(&lun->mode_pages.ie_page[
4378 				       CTL_PAGE_CHANGEABLE], &ie_page_changeable,
4379 				       sizeof(ie_page_changeable));
4380 				memcpy(&lun->mode_pages.ie_page[CTL_PAGE_DEFAULT],
4381 				       &ie_page_default,
4382 				       sizeof(ie_page_default));
4383 				memcpy(&lun->mode_pages.ie_page[CTL_PAGE_SAVED],
4384 				       &ie_page_default,
4385 				       sizeof(ie_page_default));
4386 				page_index->page_data =
4387 					(uint8_t *)lun->mode_pages.ie_page;
4388 				break;
4389 			case 0x02: {
4390 				struct ctl_logical_block_provisioning_page *page;
4391 
4392 				memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_DEFAULT],
4393 				       &lbp_page_default,
4394 				       sizeof(lbp_page_default));
4395 				memcpy(&lun->mode_pages.lbp_page[
4396 				       CTL_PAGE_CHANGEABLE], &lbp_page_changeable,
4397 				       sizeof(lbp_page_changeable));
4398 				memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4399 				       &lbp_page_default,
4400 				       sizeof(lbp_page_default));
4401 				page = &lun->mode_pages.lbp_page[CTL_PAGE_SAVED];
4402 				value = dnvlist_get_string(lun->be_lun->options,
4403 				    "avail-threshold", NULL);
4404 				if (value != NULL &&
4405 				    ctl_expand_number(value, &ival) == 0) {
4406 					page->descr[0].flags |= SLBPPD_ENABLED |
4407 					    SLBPPD_ARMING_DEC;
4408 					if (lun->be_lun->blocksize)
4409 						ival /= lun->be_lun->blocksize;
4410 					else
4411 						ival /= 512;
4412 					scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4413 					    page->descr[0].count);
4414 				}
4415 				value = dnvlist_get_string(lun->be_lun->options,
4416 				    "used-threshold", NULL);
4417 				if (value != NULL &&
4418 				    ctl_expand_number(value, &ival) == 0) {
4419 					page->descr[1].flags |= SLBPPD_ENABLED |
4420 					    SLBPPD_ARMING_INC;
4421 					if (lun->be_lun->blocksize)
4422 						ival /= lun->be_lun->blocksize;
4423 					else
4424 						ival /= 512;
4425 					scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4426 					    page->descr[1].count);
4427 				}
4428 				value = dnvlist_get_string(lun->be_lun->options,
4429 				    "pool-avail-threshold", NULL);
4430 				if (value != NULL &&
4431 				    ctl_expand_number(value, &ival) == 0) {
4432 					page->descr[2].flags |= SLBPPD_ENABLED |
4433 					    SLBPPD_ARMING_DEC;
4434 					if (lun->be_lun->blocksize)
4435 						ival /= lun->be_lun->blocksize;
4436 					else
4437 						ival /= 512;
4438 					scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4439 					    page->descr[2].count);
4440 				}
4441 				value = dnvlist_get_string(lun->be_lun->options,
4442 				    "pool-used-threshold", NULL);
4443 				if (value != NULL &&
4444 				    ctl_expand_number(value, &ival) == 0) {
4445 					page->descr[3].flags |= SLBPPD_ENABLED |
4446 					    SLBPPD_ARMING_INC;
4447 					if (lun->be_lun->blocksize)
4448 						ival /= lun->be_lun->blocksize;
4449 					else
4450 						ival /= 512;
4451 					scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4452 					    page->descr[3].count);
4453 				}
4454 				memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_CURRENT],
4455 				       &lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4456 				       sizeof(lbp_page_default));
4457 				page_index->page_data =
4458 					(uint8_t *)lun->mode_pages.lbp_page;
4459 				break;
4460 			}
4461 			default:
4462 				panic("subpage %#x for page %#x is incorrect!",
4463 				      page_index->subpage, page_code);
4464 			}
4465 			break;
4466 		}
4467 		case SMS_CDDVD_CAPS_PAGE:{
4468 			KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4469 			    ("subpage %#x for page %#x is incorrect!",
4470 			    page_index->subpage, page_code));
4471 			memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_DEFAULT],
4472 			       &cddvd_page_default,
4473 			       sizeof(cddvd_page_default));
4474 			memcpy(&lun->mode_pages.cddvd_page[
4475 			       CTL_PAGE_CHANGEABLE], &cddvd_page_changeable,
4476 			       sizeof(cddvd_page_changeable));
4477 			memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_SAVED],
4478 			       &cddvd_page_default,
4479 			       sizeof(cddvd_page_default));
4480 			memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_CURRENT],
4481 			       &lun->mode_pages.cddvd_page[CTL_PAGE_SAVED],
4482 			       sizeof(cddvd_page_default));
4483 			page_index->page_data =
4484 				(uint8_t *)lun->mode_pages.cddvd_page;
4485 			break;
4486 		}
4487 		default:
4488 			panic("invalid page code value %#x", page_code);
4489 		}
4490 	}
4491 
4492 	return (CTL_RETVAL_COMPLETE);
4493 }
4494 
4495 static int
4496 ctl_init_log_page_index(struct ctl_lun *lun)
4497 {
4498 	struct ctl_page_index *page_index;
4499 	int i, j, k, prev;
4500 
4501 	memcpy(&lun->log_pages.index, log_page_index_template,
4502 	       sizeof(log_page_index_template));
4503 
4504 	prev = -1;
4505 	for (i = 0, j = 0, k = 0; i < CTL_NUM_LOG_PAGES; i++) {
4506 		page_index = &lun->log_pages.index[i];
4507 		if (lun->be_lun->lun_type == T_DIRECT &&
4508 		    (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
4509 			continue;
4510 		if (lun->be_lun->lun_type == T_PROCESSOR &&
4511 		    (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
4512 			continue;
4513 		if (lun->be_lun->lun_type == T_CDROM &&
4514 		    (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
4515 			continue;
4516 
4517 		if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING &&
4518 		    lun->backend->lun_attr == NULL)
4519 			continue;
4520 
4521 		if (page_index->page_code != prev) {
4522 			lun->log_pages.pages_page[j] = page_index->page_code;
4523 			prev = page_index->page_code;
4524 			j++;
4525 		}
4526 		lun->log_pages.subpages_page[k*2] = page_index->page_code;
4527 		lun->log_pages.subpages_page[k*2+1] = page_index->subpage;
4528 		k++;
4529 	}
4530 	lun->log_pages.index[0].page_data = &lun->log_pages.pages_page[0];
4531 	lun->log_pages.index[0].page_len = j;
4532 	lun->log_pages.index[1].page_data = &lun->log_pages.subpages_page[0];
4533 	lun->log_pages.index[1].page_len = k * 2;
4534 	lun->log_pages.index[2].page_data = (uint8_t *)&lun->log_pages.temp_page;
4535 	lun->log_pages.index[2].page_len = sizeof(lun->log_pages.temp_page);
4536 	lun->log_pages.index[3].page_data = &lun->log_pages.lbp_page[0];
4537 	lun->log_pages.index[3].page_len = 12*CTL_NUM_LBP_PARAMS;
4538 	lun->log_pages.index[4].page_data = (uint8_t *)&lun->log_pages.stat_page;
4539 	lun->log_pages.index[4].page_len = sizeof(lun->log_pages.stat_page);
4540 	lun->log_pages.index[5].page_data = (uint8_t *)&lun->log_pages.ie_page;
4541 	lun->log_pages.index[5].page_len = sizeof(lun->log_pages.ie_page);
4542 
4543 	return (CTL_RETVAL_COMPLETE);
4544 }
4545 
4546 static int
4547 hex2bin(const char *str, uint8_t *buf, int buf_size)
4548 {
4549 	int i;
4550 	u_char c;
4551 
4552 	memset(buf, 0, buf_size);
4553 	while (isspace(str[0]))
4554 		str++;
4555 	if (str[0] == '0' && (str[1] == 'x' || str[1] == 'X'))
4556 		str += 2;
4557 	buf_size *= 2;
4558 	for (i = 0; str[i] != 0 && i < buf_size; i++) {
4559 		while (str[i] == '-')	/* Skip dashes in UUIDs. */
4560 			str++;
4561 		c = str[i];
4562 		if (isdigit(c))
4563 			c -= '0';
4564 		else if (isalpha(c))
4565 			c -= isupper(c) ? 'A' - 10 : 'a' - 10;
4566 		else
4567 			break;
4568 		if (c >= 16)
4569 			break;
4570 		if ((i & 1) == 0)
4571 			buf[i / 2] |= (c << 4);
4572 		else
4573 			buf[i / 2] |= c;
4574 	}
4575 	return ((i + 1) / 2);
4576 }
4577 
4578 /*
4579  * Add LUN.
4580  *
4581  * Returns 0 for success, non-zero (errno) for failure.
4582  */
4583 int
4584 ctl_add_lun(struct ctl_be_lun *be_lun)
4585 {
4586 	struct ctl_softc *ctl_softc = control_softc;
4587 	struct ctl_lun *nlun, *lun;
4588 	struct scsi_vpd_id_descriptor *desc;
4589 	struct scsi_vpd_id_t10 *t10id;
4590 	const char *eui, *naa, *scsiname, *uuid, *vendor, *value;
4591 	int lun_number;
4592 	int devidlen, idlen1, idlen2 = 0, len;
4593 
4594 	/*
4595 	 * We support only Direct Access, CD-ROM or Processor LUN types.
4596 	 */
4597 	switch (be_lun->lun_type) {
4598 	case T_DIRECT:
4599 	case T_PROCESSOR:
4600 	case T_CDROM:
4601 		break;
4602 	case T_SEQUENTIAL:
4603 	case T_CHANGER:
4604 	default:
4605 		return (EINVAL);
4606 	}
4607 	lun = malloc(sizeof(*lun), M_CTL, M_WAITOK | M_ZERO);
4608 
4609 	lun->pending_sense = malloc(sizeof(struct scsi_sense_data *) *
4610 	    ctl_max_ports, M_DEVBUF, M_WAITOK | M_ZERO);
4611 	lun->pending_ua = malloc(sizeof(ctl_ua_type *) * ctl_max_ports,
4612 	    M_DEVBUF, M_WAITOK | M_ZERO);
4613 	lun->pr_keys = malloc(sizeof(uint64_t *) * ctl_max_ports,
4614 	    M_DEVBUF, M_WAITOK | M_ZERO);
4615 
4616 	/* Generate LUN ID. */
4617 	devidlen = max(CTL_DEVID_MIN_LEN,
4618 	    strnlen(be_lun->device_id, CTL_DEVID_LEN));
4619 	idlen1 = sizeof(*t10id) + devidlen;
4620 	len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4621 	scsiname = dnvlist_get_string(be_lun->options, "scsiname", NULL);
4622 	if (scsiname != NULL) {
4623 		idlen2 = roundup2(strlen(scsiname) + 1, 4);
4624 		len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4625 	}
4626 	eui = dnvlist_get_string(be_lun->options, "eui", NULL);
4627 	if (eui != NULL) {
4628 		len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4629 	}
4630 	naa = dnvlist_get_string(be_lun->options, "naa", NULL);
4631 	if (naa != NULL) {
4632 		len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4633 	}
4634 	uuid = dnvlist_get_string(be_lun->options, "uuid", NULL);
4635 	if (uuid != NULL) {
4636 		len += sizeof(struct scsi_vpd_id_descriptor) + 18;
4637 	}
4638 	lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4639 	    M_CTL, M_WAITOK | M_ZERO);
4640 	desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4641 	desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4642 	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4643 	desc->length = idlen1;
4644 	t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4645 	memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4646 	if ((vendor = dnvlist_get_string(be_lun->options, "vendor", NULL)) == NULL) {
4647 		strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4648 	} else {
4649 		strncpy(t10id->vendor, vendor,
4650 		    min(sizeof(t10id->vendor), strlen(vendor)));
4651 	}
4652 	strncpy((char *)t10id->vendor_spec_id,
4653 	    (char *)be_lun->device_id, devidlen);
4654 	if (scsiname != NULL) {
4655 		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4656 		    desc->length);
4657 		desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4658 		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4659 		    SVPD_ID_TYPE_SCSI_NAME;
4660 		desc->length = idlen2;
4661 		strlcpy(desc->identifier, scsiname, idlen2);
4662 	}
4663 	if (eui != NULL) {
4664 		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4665 		    desc->length);
4666 		desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4667 		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4668 		    SVPD_ID_TYPE_EUI64;
4669 		desc->length = hex2bin(eui, desc->identifier, 16);
4670 		desc->length = desc->length > 12 ? 16 :
4671 		    (desc->length > 8 ? 12 : 8);
4672 		len -= 16 - desc->length;
4673 	}
4674 	if (naa != NULL) {
4675 		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4676 		    desc->length);
4677 		desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4678 		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4679 		    SVPD_ID_TYPE_NAA;
4680 		desc->length = hex2bin(naa, desc->identifier, 16);
4681 		desc->length = desc->length > 8 ? 16 : 8;
4682 		len -= 16 - desc->length;
4683 	}
4684 	if (uuid != NULL) {
4685 		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4686 		    desc->length);
4687 		desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4688 		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4689 		    SVPD_ID_TYPE_UUID;
4690 		desc->identifier[0] = 0x10;
4691 		hex2bin(uuid, &desc->identifier[2], 16);
4692 		desc->length = 18;
4693 	}
4694 	lun->lun_devid->len = len;
4695 
4696 	mtx_lock(&ctl_softc->ctl_lock);
4697 	/*
4698 	 * See if the caller requested a particular LUN number.  If so, see
4699 	 * if it is available.  Otherwise, allocate the first available LUN.
4700 	 */
4701 	if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4702 		if ((be_lun->req_lun_id > (ctl_max_luns - 1))
4703 		 || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4704 			mtx_unlock(&ctl_softc->ctl_lock);
4705 			if (be_lun->req_lun_id > (ctl_max_luns - 1)) {
4706 				printf("ctl: requested LUN ID %d is higher "
4707 				       "than ctl_max_luns - 1 (%d)\n",
4708 				       be_lun->req_lun_id, ctl_max_luns - 1);
4709 			} else {
4710 				/*
4711 				 * XXX KDM return an error, or just assign
4712 				 * another LUN ID in this case??
4713 				 */
4714 				printf("ctl: requested LUN ID %d is already "
4715 				       "in use\n", be_lun->req_lun_id);
4716 			}
4717 fail:
4718 			free(lun->lun_devid, M_CTL);
4719 			free(lun, M_CTL);
4720 			return (ENOSPC);
4721 		}
4722 		lun_number = be_lun->req_lun_id;
4723 	} else {
4724 		lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, 0, ctl_max_luns);
4725 		if (lun_number == -1) {
4726 			mtx_unlock(&ctl_softc->ctl_lock);
4727 			printf("ctl: can't allocate LUN, out of LUNs\n");
4728 			goto fail;
4729 		}
4730 	}
4731 	ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4732 	mtx_unlock(&ctl_softc->ctl_lock);
4733 
4734 	mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4735 	lun->lun = lun_number;
4736 	lun->be_lun = be_lun;
4737 	/*
4738 	 * The processor LUN is always enabled.  Disk LUNs come on line
4739 	 * disabled, and must be enabled by the backend.
4740 	 */
4741 	lun->flags |= CTL_LUN_DISABLED;
4742 	lun->backend = be_lun->be;
4743 	be_lun->ctl_lun = lun;
4744 	be_lun->lun_id = lun_number;
4745 	if (be_lun->flags & CTL_LUN_FLAG_EJECTED)
4746 		lun->flags |= CTL_LUN_EJECTED;
4747 	if (be_lun->flags & CTL_LUN_FLAG_NO_MEDIA)
4748 		lun->flags |= CTL_LUN_NO_MEDIA;
4749 	if (be_lun->flags & CTL_LUN_FLAG_STOPPED)
4750 		lun->flags |= CTL_LUN_STOPPED;
4751 
4752 	if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4753 		lun->flags |= CTL_LUN_PRIMARY_SC;
4754 
4755 	value = dnvlist_get_string(be_lun->options, "removable", NULL);
4756 	if (value != NULL) {
4757 		if (strcmp(value, "on") == 0)
4758 			lun->flags |= CTL_LUN_REMOVABLE;
4759 	} else if (be_lun->lun_type == T_CDROM)
4760 		lun->flags |= CTL_LUN_REMOVABLE;
4761 
4762 	lun->ctl_softc = ctl_softc;
4763 #ifdef CTL_TIME_IO
4764 	lun->last_busy = getsbinuptime();
4765 #endif
4766 	LIST_INIT(&lun->ooa_queue);
4767 	STAILQ_INIT(&lun->error_list);
4768 	lun->ie_reported = 1;
4769 	callout_init_mtx(&lun->ie_callout, &lun->lun_lock, 0);
4770 	ctl_tpc_lun_init(lun);
4771 	if (lun->flags & CTL_LUN_REMOVABLE) {
4772 		lun->prevent = malloc((CTL_MAX_INITIATORS + 31) / 32 * 4,
4773 		    M_CTL, M_WAITOK);
4774 	}
4775 
4776 	/*
4777 	 * Initialize the mode and log page index.
4778 	 */
4779 	ctl_init_page_index(lun);
4780 	ctl_init_log_page_index(lun);
4781 
4782 	/* Setup statistics gathering */
4783 	lun->stats.item = lun_number;
4784 
4785 	/*
4786 	 * Now, before we insert this lun on the lun list, set the lun
4787 	 * inventory changed UA for all other luns.
4788 	 */
4789 	mtx_lock(&ctl_softc->ctl_lock);
4790 	STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4791 		mtx_lock(&nlun->lun_lock);
4792 		ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4793 		mtx_unlock(&nlun->lun_lock);
4794 	}
4795 	STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4796 	ctl_softc->ctl_luns[lun_number] = lun;
4797 	ctl_softc->num_luns++;
4798 	mtx_unlock(&ctl_softc->ctl_lock);
4799 
4800 	/*
4801 	 * We successfully added the LUN, attempt to enable it.
4802 	 */
4803 	if (ctl_enable_lun(lun) != 0) {
4804 		printf("%s: ctl_enable_lun() failed!\n", __func__);
4805 		mtx_lock(&ctl_softc->ctl_lock);
4806 		STAILQ_REMOVE(&ctl_softc->lun_list, lun, ctl_lun, links);
4807 		ctl_clear_mask(ctl_softc->ctl_lun_mask, lun_number);
4808 		ctl_softc->ctl_luns[lun_number] = NULL;
4809 		ctl_softc->num_luns--;
4810 		mtx_unlock(&ctl_softc->ctl_lock);
4811 		free(lun->lun_devid, M_CTL);
4812 		free(lun, M_CTL);
4813 		return (EIO);
4814 	}
4815 
4816 	return (0);
4817 }
4818 
4819 /*
4820  * Free LUN that has no active requests.
4821  */
4822 static int
4823 ctl_free_lun(struct ctl_lun *lun)
4824 {
4825 	struct ctl_softc *softc = lun->ctl_softc;
4826 	struct ctl_lun *nlun;
4827 	int i;
4828 
4829 	KASSERT(LIST_EMPTY(&lun->ooa_queue),
4830 	    ("Freeing a LUN %p with outstanding I/O!\n", lun));
4831 
4832 	mtx_lock(&softc->ctl_lock);
4833 	STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4834 	ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4835 	softc->ctl_luns[lun->lun] = NULL;
4836 	softc->num_luns--;
4837 	STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4838 		mtx_lock(&nlun->lun_lock);
4839 		ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4840 		mtx_unlock(&nlun->lun_lock);
4841 	}
4842 	mtx_unlock(&softc->ctl_lock);
4843 
4844 	/*
4845 	 * Tell the backend to free resources, if this LUN has a backend.
4846 	 */
4847 	lun->be_lun->lun_shutdown(lun->be_lun);
4848 
4849 	lun->ie_reportcnt = UINT32_MAX;
4850 	callout_drain(&lun->ie_callout);
4851 	ctl_tpc_lun_shutdown(lun);
4852 	mtx_destroy(&lun->lun_lock);
4853 	free(lun->lun_devid, M_CTL);
4854 	for (i = 0; i < ctl_max_ports; i++)
4855 		free(lun->pending_ua[i], M_CTL);
4856 	free(lun->pending_ua, M_DEVBUF);
4857 	for (i = 0; i < ctl_max_ports; i++)
4858 		free(lun->pr_keys[i], M_CTL);
4859 	free(lun->pr_keys, M_DEVBUF);
4860 	free(lun->write_buffer, M_CTL);
4861 	free(lun->prevent, M_CTL);
4862 	free(lun, M_CTL);
4863 
4864 	return (0);
4865 }
4866 
4867 static int
4868 ctl_enable_lun(struct ctl_lun *lun)
4869 {
4870 	struct ctl_softc *softc;
4871 	struct ctl_port *port, *nport;
4872 	int retval;
4873 
4874 	softc = lun->ctl_softc;
4875 
4876 	mtx_lock(&softc->ctl_lock);
4877 	mtx_lock(&lun->lun_lock);
4878 	KASSERT((lun->flags & CTL_LUN_DISABLED) != 0,
4879 	    ("%s: LUN not disabled", __func__));
4880 	lun->flags &= ~CTL_LUN_DISABLED;
4881 	mtx_unlock(&lun->lun_lock);
4882 
4883 	STAILQ_FOREACH_SAFE(port, &softc->port_list, links, nport) {
4884 		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4885 		    port->lun_map != NULL || port->lun_enable == NULL)
4886 			continue;
4887 
4888 		/*
4889 		 * Drop the lock while we call the FETD's enable routine.
4890 		 * This can lead to a callback into CTL (at least in the
4891 		 * case of the internal initiator frontend.
4892 		 */
4893 		mtx_unlock(&softc->ctl_lock);
4894 		retval = port->lun_enable(port->targ_lun_arg, lun->lun);
4895 		mtx_lock(&softc->ctl_lock);
4896 		if (retval != 0) {
4897 			printf("%s: FETD %s port %d returned error "
4898 			       "%d for lun_enable on lun %jd\n",
4899 			       __func__, port->port_name, port->targ_port,
4900 			       retval, (intmax_t)lun->lun);
4901 		}
4902 	}
4903 
4904 	mtx_unlock(&softc->ctl_lock);
4905 	ctl_isc_announce_lun(lun);
4906 
4907 	return (0);
4908 }
4909 
4910 static int
4911 ctl_disable_lun(struct ctl_lun *lun)
4912 {
4913 	struct ctl_softc *softc;
4914 	struct ctl_port *port;
4915 	int retval;
4916 
4917 	softc = lun->ctl_softc;
4918 
4919 	mtx_lock(&softc->ctl_lock);
4920 	mtx_lock(&lun->lun_lock);
4921 	KASSERT((lun->flags & CTL_LUN_DISABLED) == 0,
4922 	    ("%s: LUN not enabled", __func__));
4923 	lun->flags |= CTL_LUN_DISABLED;
4924 	mtx_unlock(&lun->lun_lock);
4925 
4926 	STAILQ_FOREACH(port, &softc->port_list, links) {
4927 		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4928 		    port->lun_map != NULL || port->lun_disable == NULL)
4929 			continue;
4930 
4931 		/*
4932 		 * Drop the lock before we call the frontend's disable
4933 		 * routine, to avoid lock order reversals.
4934 		 *
4935 		 * XXX KDM what happens if the frontend list changes while
4936 		 * we're traversing it?  It's unlikely, but should be handled.
4937 		 */
4938 		mtx_unlock(&softc->ctl_lock);
4939 		retval = port->lun_disable(port->targ_lun_arg, lun->lun);
4940 		mtx_lock(&softc->ctl_lock);
4941 		if (retval != 0) {
4942 			printf("%s: FETD %s port %d returned error "
4943 			       "%d for lun_disable on lun %jd\n",
4944 			       __func__, port->port_name, port->targ_port,
4945 			       retval, (intmax_t)lun->lun);
4946 		}
4947 	}
4948 
4949 	mtx_unlock(&softc->ctl_lock);
4950 	ctl_isc_announce_lun(lun);
4951 
4952 	return (0);
4953 }
4954 
4955 int
4956 ctl_start_lun(struct ctl_be_lun *be_lun)
4957 {
4958 	struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4959 
4960 	mtx_lock(&lun->lun_lock);
4961 	lun->flags &= ~CTL_LUN_STOPPED;
4962 	mtx_unlock(&lun->lun_lock);
4963 	return (0);
4964 }
4965 
4966 int
4967 ctl_stop_lun(struct ctl_be_lun *be_lun)
4968 {
4969 	struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4970 
4971 	mtx_lock(&lun->lun_lock);
4972 	lun->flags |= CTL_LUN_STOPPED;
4973 	mtx_unlock(&lun->lun_lock);
4974 	return (0);
4975 }
4976 
4977 int
4978 ctl_lun_no_media(struct ctl_be_lun *be_lun)
4979 {
4980 	struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4981 
4982 	mtx_lock(&lun->lun_lock);
4983 	lun->flags |= CTL_LUN_NO_MEDIA;
4984 	mtx_unlock(&lun->lun_lock);
4985 	return (0);
4986 }
4987 
4988 int
4989 ctl_lun_has_media(struct ctl_be_lun *be_lun)
4990 {
4991 	struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4992 	union ctl_ha_msg msg;
4993 
4994 	mtx_lock(&lun->lun_lock);
4995 	lun->flags &= ~(CTL_LUN_NO_MEDIA | CTL_LUN_EJECTED);
4996 	if (lun->flags & CTL_LUN_REMOVABLE)
4997 		ctl_est_ua_all(lun, -1, CTL_UA_MEDIUM_CHANGE);
4998 	mtx_unlock(&lun->lun_lock);
4999 	if ((lun->flags & CTL_LUN_REMOVABLE) &&
5000 	    lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
5001 		bzero(&msg.ua, sizeof(msg.ua));
5002 		msg.hdr.msg_type = CTL_MSG_UA;
5003 		msg.hdr.nexus.initid = -1;
5004 		msg.hdr.nexus.targ_port = -1;
5005 		msg.hdr.nexus.targ_lun = lun->lun;
5006 		msg.hdr.nexus.targ_mapped_lun = lun->lun;
5007 		msg.ua.ua_all = 1;
5008 		msg.ua.ua_set = 1;
5009 		msg.ua.ua_type = CTL_UA_MEDIUM_CHANGE;
5010 		ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.ua),
5011 		    M_WAITOK);
5012 	}
5013 	return (0);
5014 }
5015 
5016 int
5017 ctl_lun_ejected(struct ctl_be_lun *be_lun)
5018 {
5019 	struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
5020 
5021 	mtx_lock(&lun->lun_lock);
5022 	lun->flags |= CTL_LUN_EJECTED;
5023 	mtx_unlock(&lun->lun_lock);
5024 	return (0);
5025 }
5026 
5027 int
5028 ctl_lun_primary(struct ctl_be_lun *be_lun)
5029 {
5030 	struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
5031 
5032 	mtx_lock(&lun->lun_lock);
5033 	lun->flags |= CTL_LUN_PRIMARY_SC;
5034 	ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
5035 	mtx_unlock(&lun->lun_lock);
5036 	ctl_isc_announce_lun(lun);
5037 	return (0);
5038 }
5039 
5040 int
5041 ctl_lun_secondary(struct ctl_be_lun *be_lun)
5042 {
5043 	struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
5044 
5045 	mtx_lock(&lun->lun_lock);
5046 	lun->flags &= ~CTL_LUN_PRIMARY_SC;
5047 	ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
5048 	mtx_unlock(&lun->lun_lock);
5049 	ctl_isc_announce_lun(lun);
5050 	return (0);
5051 }
5052 
5053 /*
5054  * Remove LUN.  If there are active requests, wait for completion.
5055  *
5056  * Returns 0 for success, non-zero (errno) for failure.
5057  * Completion is reported to backed via the lun_shutdown() method.
5058  */
5059 int
5060 ctl_remove_lun(struct ctl_be_lun *be_lun)
5061 {
5062 	struct ctl_lun *lun;
5063 
5064 	lun = (struct ctl_lun *)be_lun->ctl_lun;
5065 
5066 	ctl_disable_lun(lun);
5067 
5068 	mtx_lock(&lun->lun_lock);
5069 	lun->flags |= CTL_LUN_INVALID;
5070 
5071 	/*
5072 	 * If there is nothing in the OOA queue, go ahead and free the LUN.
5073 	 * If we have something in the OOA queue, we'll free it when the
5074 	 * last I/O completes.
5075 	 */
5076 	if (LIST_EMPTY(&lun->ooa_queue)) {
5077 		mtx_unlock(&lun->lun_lock);
5078 		ctl_free_lun(lun);
5079 	} else
5080 		mtx_unlock(&lun->lun_lock);
5081 
5082 	return (0);
5083 }
5084 
5085 void
5086 ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
5087 {
5088 	struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
5089 	union ctl_ha_msg msg;
5090 
5091 	mtx_lock(&lun->lun_lock);
5092 	ctl_est_ua_all(lun, -1, CTL_UA_CAPACITY_CHANGE);
5093 	mtx_unlock(&lun->lun_lock);
5094 	if (lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
5095 		/* Send msg to other side. */
5096 		bzero(&msg.ua, sizeof(msg.ua));
5097 		msg.hdr.msg_type = CTL_MSG_UA;
5098 		msg.hdr.nexus.initid = -1;
5099 		msg.hdr.nexus.targ_port = -1;
5100 		msg.hdr.nexus.targ_lun = lun->lun;
5101 		msg.hdr.nexus.targ_mapped_lun = lun->lun;
5102 		msg.ua.ua_all = 1;
5103 		msg.ua.ua_set = 1;
5104 		msg.ua.ua_type = CTL_UA_CAPACITY_CHANGE;
5105 		ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.ua),
5106 		    M_WAITOK);
5107 	}
5108 }
5109 
5110 /*
5111  * Backend "memory move is complete" callback for requests that never
5112  * make it down to say RAIDCore's configuration code.
5113  */
5114 int
5115 ctl_config_move_done(union ctl_io *io, bool samethr)
5116 {
5117 	int retval;
5118 
5119 	CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
5120 	KASSERT(io->io_hdr.io_type == CTL_IO_SCSI,
5121 	    ("%s: unexpected I/O type %x", __func__, io->io_hdr.io_type));
5122 
5123 	if (ctl_debug & CTL_DEBUG_CDB_DATA)
5124 		ctl_data_print(io);
5125 	if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) ||
5126 	    ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
5127 	     (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) ||
5128 	    ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
5129 		/*
5130 		 * XXX KDM just assuming a single pointer here, and not a
5131 		 * S/G list.  If we start using S/G lists for config data,
5132 		 * we'll need to know how to clean them up here as well.
5133 		 */
5134 		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5135 			free(io->scsiio.kern_data_ptr, M_CTL);
5136 		ctl_done(io);
5137 		retval = CTL_RETVAL_COMPLETE;
5138 	} else {
5139 		/*
5140 		 * XXX KDM now we need to continue data movement.  Some
5141 		 * options:
5142 		 * - call ctl_scsiio() again?  We don't do this for data
5143 		 *   writes, because for those at least we know ahead of
5144 		 *   time where the write will go and how long it is.  For
5145 		 *   config writes, though, that information is largely
5146 		 *   contained within the write itself, thus we need to
5147 		 *   parse out the data again.
5148 		 *
5149 		 * - Call some other function once the data is in?
5150 		 */
5151 
5152 		/*
5153 		 * XXX KDM call ctl_scsiio() again for now, and check flag
5154 		 * bits to see whether we're allocated or not.
5155 		 */
5156 		retval = ctl_scsiio(&io->scsiio);
5157 	}
5158 	return (retval);
5159 }
5160 
5161 /*
5162  * This gets called by a backend driver when it is done with a
5163  * data_submit method.
5164  */
5165 void
5166 ctl_data_submit_done(union ctl_io *io)
5167 {
5168 	/*
5169 	 * If the IO_CONT flag is set, we need to call the supplied
5170 	 * function to continue processing the I/O, instead of completing
5171 	 * the I/O just yet.
5172 	 *
5173 	 * If there is an error, though, we don't want to keep processing.
5174 	 * Instead, just send status back to the initiator.
5175 	 */
5176 	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5177 	    (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5178 	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5179 	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5180 		io->scsiio.io_cont(io);
5181 		return;
5182 	}
5183 	ctl_done(io);
5184 }
5185 
5186 /*
5187  * This gets called by a backend driver when it is done with a
5188  * configuration write.
5189  */
5190 void
5191 ctl_config_write_done(union ctl_io *io)
5192 {
5193 	uint8_t *buf;
5194 
5195 	/*
5196 	 * If the IO_CONT flag is set, we need to call the supplied
5197 	 * function to continue processing the I/O, instead of completing
5198 	 * the I/O just yet.
5199 	 *
5200 	 * If there is an error, though, we don't want to keep processing.
5201 	 * Instead, just send status back to the initiator.
5202 	 */
5203 	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5204 	    (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5205 	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5206 	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5207 		io->scsiio.io_cont(io);
5208 		return;
5209 	}
5210 	/*
5211 	 * Since a configuration write can be done for commands that actually
5212 	 * have data allocated, like write buffer, and commands that have
5213 	 * no data, like start/stop unit, we need to check here.
5214 	 */
5215 	if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5216 		buf = io->scsiio.kern_data_ptr;
5217 	else
5218 		buf = NULL;
5219 	ctl_done(io);
5220 	if (buf)
5221 		free(buf, M_CTL);
5222 }
5223 
5224 void
5225 ctl_config_read_done(union ctl_io *io)
5226 {
5227 	uint8_t *buf;
5228 
5229 	/*
5230 	 * If there is some error -- we are done, skip data transfer.
5231 	 */
5232 	if ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0 ||
5233 	    ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
5234 	     (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
5235 		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5236 			buf = io->scsiio.kern_data_ptr;
5237 		else
5238 			buf = NULL;
5239 		ctl_done(io);
5240 		if (buf)
5241 			free(buf, M_CTL);
5242 		return;
5243 	}
5244 
5245 	/*
5246 	 * If the IO_CONT flag is set, we need to call the supplied
5247 	 * function to continue processing the I/O, instead of completing
5248 	 * the I/O just yet.
5249 	 */
5250 	if (io->io_hdr.flags & CTL_FLAG_IO_CONT) {
5251 		io->scsiio.io_cont(io);
5252 		return;
5253 	}
5254 
5255 	ctl_datamove(io);
5256 }
5257 
5258 /*
5259  * SCSI release command.
5260  */
5261 int
5262 ctl_scsi_release(struct ctl_scsiio *ctsio)
5263 {
5264 	struct ctl_lun *lun = CTL_LUN(ctsio);
5265 	uint32_t residx;
5266 
5267 	CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5268 
5269 	residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5270 
5271 	/*
5272 	 * XXX KDM right now, we only support LUN reservation.  We don't
5273 	 * support 3rd party reservations, or extent reservations, which
5274 	 * might actually need the parameter list.  If we've gotten this
5275 	 * far, we've got a LUN reservation.  Anything else got kicked out
5276 	 * above.  So, according to SPC, ignore the length.
5277 	 */
5278 
5279 	mtx_lock(&lun->lun_lock);
5280 
5281 	/*
5282 	 * According to SPC, it is not an error for an intiator to attempt
5283 	 * to release a reservation on a LUN that isn't reserved, or that
5284 	 * is reserved by another initiator.  The reservation can only be
5285 	 * released, though, by the initiator who made it or by one of
5286 	 * several reset type events.
5287 	 */
5288 	if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
5289 			lun->flags &= ~CTL_LUN_RESERVED;
5290 
5291 	mtx_unlock(&lun->lun_lock);
5292 
5293 	ctl_set_success(ctsio);
5294 	ctl_done((union ctl_io *)ctsio);
5295 	return (CTL_RETVAL_COMPLETE);
5296 }
5297 
5298 int
5299 ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5300 {
5301 	struct ctl_lun *lun = CTL_LUN(ctsio);
5302 	uint32_t residx;
5303 
5304 	CTL_DEBUG_PRINT(("ctl_reserve\n"));
5305 
5306 	residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5307 
5308 	/*
5309 	 * XXX KDM right now, we only support LUN reservation.  We don't
5310 	 * support 3rd party reservations, or extent reservations, which
5311 	 * might actually need the parameter list.  If we've gotten this
5312 	 * far, we've got a LUN reservation.  Anything else got kicked out
5313 	 * above.  So, according to SPC, ignore the length.
5314 	 */
5315 
5316 	mtx_lock(&lun->lun_lock);
5317 	if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx != residx)) {
5318 		ctl_set_reservation_conflict(ctsio);
5319 		goto bailout;
5320 	}
5321 
5322 	/* SPC-3 exceptions to SPC-2 RESERVE and RELEASE behavior. */
5323 	if (lun->flags & CTL_LUN_PR_RESERVED) {
5324 		ctl_set_success(ctsio);
5325 		goto bailout;
5326 	}
5327 
5328 	lun->flags |= CTL_LUN_RESERVED;
5329 	lun->res_idx = residx;
5330 	ctl_set_success(ctsio);
5331 
5332 bailout:
5333 	mtx_unlock(&lun->lun_lock);
5334 	ctl_done((union ctl_io *)ctsio);
5335 	return (CTL_RETVAL_COMPLETE);
5336 }
5337 
5338 int
5339 ctl_start_stop(struct ctl_scsiio *ctsio)
5340 {
5341 	struct ctl_lun *lun = CTL_LUN(ctsio);
5342 	struct scsi_start_stop_unit *cdb;
5343 	int retval;
5344 
5345 	CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5346 
5347 	cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5348 
5349 	if ((cdb->how & SSS_PC_MASK) == 0) {
5350 		if ((lun->flags & CTL_LUN_PR_RESERVED) &&
5351 		    (cdb->how & SSS_START) == 0) {
5352 			uint32_t residx;
5353 
5354 			residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5355 			if (ctl_get_prkey(lun, residx) == 0 ||
5356 			    (lun->pr_res_idx != residx && lun->pr_res_type < 4)) {
5357 				ctl_set_reservation_conflict(ctsio);
5358 				ctl_done((union ctl_io *)ctsio);
5359 				return (CTL_RETVAL_COMPLETE);
5360 			}
5361 		}
5362 
5363 		if ((cdb->how & SSS_LOEJ) &&
5364 		    (lun->flags & CTL_LUN_REMOVABLE) == 0) {
5365 			ctl_set_invalid_field(ctsio,
5366 					      /*sks_valid*/ 1,
5367 					      /*command*/ 1,
5368 					      /*field*/ 4,
5369 					      /*bit_valid*/ 1,
5370 					      /*bit*/ 1);
5371 			ctl_done((union ctl_io *)ctsio);
5372 			return (CTL_RETVAL_COMPLETE);
5373 		}
5374 
5375 		if ((cdb->how & SSS_START) == 0 && (cdb->how & SSS_LOEJ) &&
5376 		    lun->prevent_count > 0) {
5377 			/* "Medium removal prevented" */
5378 			ctl_set_sense(ctsio, /*current_error*/ 1,
5379 			    /*sense_key*/(lun->flags & CTL_LUN_NO_MEDIA) ?
5380 			     SSD_KEY_NOT_READY : SSD_KEY_ILLEGAL_REQUEST,
5381 			    /*asc*/ 0x53, /*ascq*/ 0x02, SSD_ELEM_NONE);
5382 			ctl_done((union ctl_io *)ctsio);
5383 			return (CTL_RETVAL_COMPLETE);
5384 		}
5385 	}
5386 
5387 	retval = lun->backend->config_write((union ctl_io *)ctsio);
5388 	return (retval);
5389 }
5390 
5391 int
5392 ctl_prevent_allow(struct ctl_scsiio *ctsio)
5393 {
5394 	struct ctl_lun *lun = CTL_LUN(ctsio);
5395 	struct scsi_prevent *cdb;
5396 	int retval;
5397 	uint32_t initidx;
5398 
5399 	CTL_DEBUG_PRINT(("ctl_prevent_allow\n"));
5400 
5401 	cdb = (struct scsi_prevent *)ctsio->cdb;
5402 
5403 	if ((lun->flags & CTL_LUN_REMOVABLE) == 0 || lun->prevent == NULL) {
5404 		ctl_set_invalid_opcode(ctsio);
5405 		ctl_done((union ctl_io *)ctsio);
5406 		return (CTL_RETVAL_COMPLETE);
5407 	}
5408 
5409 	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5410 	mtx_lock(&lun->lun_lock);
5411 	if ((cdb->how & PR_PREVENT) &&
5412 	    ctl_is_set(lun->prevent, initidx) == 0) {
5413 		ctl_set_mask(lun->prevent, initidx);
5414 		lun->prevent_count++;
5415 	} else if ((cdb->how & PR_PREVENT) == 0 &&
5416 	    ctl_is_set(lun->prevent, initidx)) {
5417 		ctl_clear_mask(lun->prevent, initidx);
5418 		lun->prevent_count--;
5419 	}
5420 	mtx_unlock(&lun->lun_lock);
5421 	retval = lun->backend->config_write((union ctl_io *)ctsio);
5422 	return (retval);
5423 }
5424 
5425 /*
5426  * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5427  * we don't really do anything with the LBA and length fields if the user
5428  * passes them in.  Instead we'll just flush out the cache for the entire
5429  * LUN.
5430  */
5431 int
5432 ctl_sync_cache(struct ctl_scsiio *ctsio)
5433 {
5434 	struct ctl_lun *lun = CTL_LUN(ctsio);
5435 	struct ctl_lba_len_flags *lbalen;
5436 	uint64_t starting_lba;
5437 	uint32_t block_count;
5438 	int retval;
5439 	uint8_t byte2;
5440 
5441 	CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5442 
5443 	retval = 0;
5444 
5445 	switch (ctsio->cdb[0]) {
5446 	case SYNCHRONIZE_CACHE: {
5447 		struct scsi_sync_cache *cdb;
5448 		cdb = (struct scsi_sync_cache *)ctsio->cdb;
5449 
5450 		starting_lba = scsi_4btoul(cdb->begin_lba);
5451 		block_count = scsi_2btoul(cdb->lb_count);
5452 		byte2 = cdb->byte2;
5453 		break;
5454 	}
5455 	case SYNCHRONIZE_CACHE_16: {
5456 		struct scsi_sync_cache_16 *cdb;
5457 		cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5458 
5459 		starting_lba = scsi_8btou64(cdb->begin_lba);
5460 		block_count = scsi_4btoul(cdb->lb_count);
5461 		byte2 = cdb->byte2;
5462 		break;
5463 	}
5464 	default:
5465 		ctl_set_invalid_opcode(ctsio);
5466 		ctl_done((union ctl_io *)ctsio);
5467 		goto bailout;
5468 		break; /* NOTREACHED */
5469 	}
5470 
5471 	/*
5472 	 * We check the LBA and length, but don't do anything with them.
5473 	 * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5474 	 * get flushed.  This check will just help satisfy anyone who wants
5475 	 * to see an error for an out of range LBA.
5476 	 */
5477 	if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5478 		ctl_set_lba_out_of_range(ctsio,
5479 		    MAX(starting_lba, lun->be_lun->maxlba + 1));
5480 		ctl_done((union ctl_io *)ctsio);
5481 		goto bailout;
5482 	}
5483 
5484 	lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5485 	lbalen->lba = starting_lba;
5486 	lbalen->len = block_count;
5487 	lbalen->flags = byte2;
5488 	retval = lun->backend->config_write((union ctl_io *)ctsio);
5489 
5490 bailout:
5491 	return (retval);
5492 }
5493 
5494 int
5495 ctl_format(struct ctl_scsiio *ctsio)
5496 {
5497 	struct scsi_format *cdb;
5498 	int length, defect_list_len;
5499 
5500 	CTL_DEBUG_PRINT(("ctl_format\n"));
5501 
5502 	cdb = (struct scsi_format *)ctsio->cdb;
5503 
5504 	length = 0;
5505 	if (cdb->byte2 & SF_FMTDATA) {
5506 		if (cdb->byte2 & SF_LONGLIST)
5507 			length = sizeof(struct scsi_format_header_long);
5508 		else
5509 			length = sizeof(struct scsi_format_header_short);
5510 	}
5511 
5512 	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5513 	 && (length > 0)) {
5514 		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5515 		ctsio->kern_data_len = length;
5516 		ctsio->kern_total_len = length;
5517 		ctsio->kern_rel_offset = 0;
5518 		ctsio->kern_sg_entries = 0;
5519 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5520 		ctsio->be_move_done = ctl_config_move_done;
5521 		ctl_datamove((union ctl_io *)ctsio);
5522 
5523 		return (CTL_RETVAL_COMPLETE);
5524 	}
5525 
5526 	defect_list_len = 0;
5527 
5528 	if (cdb->byte2 & SF_FMTDATA) {
5529 		if (cdb->byte2 & SF_LONGLIST) {
5530 			struct scsi_format_header_long *header;
5531 
5532 			header = (struct scsi_format_header_long *)
5533 				ctsio->kern_data_ptr;
5534 
5535 			defect_list_len = scsi_4btoul(header->defect_list_len);
5536 			if (defect_list_len != 0) {
5537 				ctl_set_invalid_field(ctsio,
5538 						      /*sks_valid*/ 1,
5539 						      /*command*/ 0,
5540 						      /*field*/ 2,
5541 						      /*bit_valid*/ 0,
5542 						      /*bit*/ 0);
5543 				goto bailout;
5544 			}
5545 		} else {
5546 			struct scsi_format_header_short *header;
5547 
5548 			header = (struct scsi_format_header_short *)
5549 				ctsio->kern_data_ptr;
5550 
5551 			defect_list_len = scsi_2btoul(header->defect_list_len);
5552 			if (defect_list_len != 0) {
5553 				ctl_set_invalid_field(ctsio,
5554 						      /*sks_valid*/ 1,
5555 						      /*command*/ 0,
5556 						      /*field*/ 2,
5557 						      /*bit_valid*/ 0,
5558 						      /*bit*/ 0);
5559 				goto bailout;
5560 			}
5561 		}
5562 	}
5563 
5564 	ctl_set_success(ctsio);
5565 bailout:
5566 
5567 	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5568 		free(ctsio->kern_data_ptr, M_CTL);
5569 		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5570 	}
5571 
5572 	ctl_done((union ctl_io *)ctsio);
5573 	return (CTL_RETVAL_COMPLETE);
5574 }
5575 
5576 int
5577 ctl_read_buffer(struct ctl_scsiio *ctsio)
5578 {
5579 	struct ctl_lun *lun = CTL_LUN(ctsio);
5580 	uint64_t buffer_offset;
5581 	uint32_t len;
5582 	uint8_t byte2;
5583 	static uint8_t descr[4];
5584 	static uint8_t echo_descr[4] = { 0 };
5585 
5586 	CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5587 
5588 	switch (ctsio->cdb[0]) {
5589 	case READ_BUFFER: {
5590 		struct scsi_read_buffer *cdb;
5591 
5592 		cdb = (struct scsi_read_buffer *)ctsio->cdb;
5593 		buffer_offset = scsi_3btoul(cdb->offset);
5594 		len = scsi_3btoul(cdb->length);
5595 		byte2 = cdb->byte2;
5596 		break;
5597 	}
5598 	case READ_BUFFER_16: {
5599 		struct scsi_read_buffer_16 *cdb;
5600 
5601 		cdb = (struct scsi_read_buffer_16 *)ctsio->cdb;
5602 		buffer_offset = scsi_8btou64(cdb->offset);
5603 		len = scsi_4btoul(cdb->length);
5604 		byte2 = cdb->byte2;
5605 		break;
5606 	}
5607 	default: /* This shouldn't happen. */
5608 		ctl_set_invalid_opcode(ctsio);
5609 		ctl_done((union ctl_io *)ctsio);
5610 		return (CTL_RETVAL_COMPLETE);
5611 	}
5612 
5613 	if (buffer_offset > CTL_WRITE_BUFFER_SIZE ||
5614 	    buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5615 		ctl_set_invalid_field(ctsio,
5616 				      /*sks_valid*/ 1,
5617 				      /*command*/ 1,
5618 				      /*field*/ 6,
5619 				      /*bit_valid*/ 0,
5620 				      /*bit*/ 0);
5621 		ctl_done((union ctl_io *)ctsio);
5622 		return (CTL_RETVAL_COMPLETE);
5623 	}
5624 
5625 	if ((byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5626 		descr[0] = 0;
5627 		scsi_ulto3b(CTL_WRITE_BUFFER_SIZE, &descr[1]);
5628 		ctsio->kern_data_ptr = descr;
5629 		len = min(len, sizeof(descr));
5630 	} else if ((byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5631 		ctsio->kern_data_ptr = echo_descr;
5632 		len = min(len, sizeof(echo_descr));
5633 	} else {
5634 		if (lun->write_buffer == NULL) {
5635 			lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5636 			    M_CTL, M_WAITOK);
5637 		}
5638 		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5639 	}
5640 	ctsio->kern_data_len = len;
5641 	ctsio->kern_total_len = len;
5642 	ctsio->kern_rel_offset = 0;
5643 	ctsio->kern_sg_entries = 0;
5644 	ctl_set_success(ctsio);
5645 	ctsio->be_move_done = ctl_config_move_done;
5646 	ctl_datamove((union ctl_io *)ctsio);
5647 	return (CTL_RETVAL_COMPLETE);
5648 }
5649 
5650 int
5651 ctl_write_buffer(struct ctl_scsiio *ctsio)
5652 {
5653 	struct ctl_lun *lun = CTL_LUN(ctsio);
5654 	struct scsi_write_buffer *cdb;
5655 	int buffer_offset, len;
5656 
5657 	CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5658 
5659 	cdb = (struct scsi_write_buffer *)ctsio->cdb;
5660 
5661 	len = scsi_3btoul(cdb->length);
5662 	buffer_offset = scsi_3btoul(cdb->offset);
5663 
5664 	if (buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5665 		ctl_set_invalid_field(ctsio,
5666 				      /*sks_valid*/ 1,
5667 				      /*command*/ 1,
5668 				      /*field*/ 6,
5669 				      /*bit_valid*/ 0,
5670 				      /*bit*/ 0);
5671 		ctl_done((union ctl_io *)ctsio);
5672 		return (CTL_RETVAL_COMPLETE);
5673 	}
5674 
5675 	/*
5676 	 * If we've got a kernel request that hasn't been malloced yet,
5677 	 * malloc it and tell the caller the data buffer is here.
5678 	 */
5679 	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5680 		if (lun->write_buffer == NULL) {
5681 			lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5682 			    M_CTL, M_WAITOK);
5683 		}
5684 		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5685 		ctsio->kern_data_len = len;
5686 		ctsio->kern_total_len = len;
5687 		ctsio->kern_rel_offset = 0;
5688 		ctsio->kern_sg_entries = 0;
5689 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5690 		ctsio->be_move_done = ctl_config_move_done;
5691 		ctl_datamove((union ctl_io *)ctsio);
5692 
5693 		return (CTL_RETVAL_COMPLETE);
5694 	}
5695 
5696 	ctl_set_success(ctsio);
5697 	ctl_done((union ctl_io *)ctsio);
5698 	return (CTL_RETVAL_COMPLETE);
5699 }
5700 
5701 static int
5702 ctl_write_same_cont(union ctl_io *io)
5703 {
5704 	struct ctl_lun *lun = CTL_LUN(io);
5705 	struct ctl_scsiio *ctsio;
5706 	struct ctl_lba_len_flags *lbalen;
5707 	int retval;
5708 
5709 	ctsio = &io->scsiio;
5710 	ctsio->io_hdr.status = CTL_STATUS_NONE;
5711 	lbalen = (struct ctl_lba_len_flags *)
5712 	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5713 	lbalen->lba += lbalen->len;
5714 	if ((lun->be_lun->maxlba + 1) - lbalen->lba <= UINT32_MAX) {
5715 		ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
5716 		lbalen->len = (lun->be_lun->maxlba + 1) - lbalen->lba;
5717 	}
5718 
5719 	CTL_DEBUG_PRINT(("ctl_write_same_cont: calling config_write()\n"));
5720 	retval = lun->backend->config_write((union ctl_io *)ctsio);
5721 	return (retval);
5722 }
5723 
5724 int
5725 ctl_write_same(struct ctl_scsiio *ctsio)
5726 {
5727 	struct ctl_lun *lun = CTL_LUN(ctsio);
5728 	struct ctl_lba_len_flags *lbalen;
5729 	const char *val;
5730 	uint64_t lba, ival;
5731 	uint32_t num_blocks;
5732 	int len, retval;
5733 	uint8_t byte2;
5734 
5735 	CTL_DEBUG_PRINT(("ctl_write_same\n"));
5736 
5737 	switch (ctsio->cdb[0]) {
5738 	case WRITE_SAME_10: {
5739 		struct scsi_write_same_10 *cdb;
5740 
5741 		cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5742 
5743 		lba = scsi_4btoul(cdb->addr);
5744 		num_blocks = scsi_2btoul(cdb->length);
5745 		byte2 = cdb->byte2;
5746 		break;
5747 	}
5748 	case WRITE_SAME_16: {
5749 		struct scsi_write_same_16 *cdb;
5750 
5751 		cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5752 
5753 		lba = scsi_8btou64(cdb->addr);
5754 		num_blocks = scsi_4btoul(cdb->length);
5755 		byte2 = cdb->byte2;
5756 		break;
5757 	}
5758 	default:
5759 		/*
5760 		 * We got a command we don't support.  This shouldn't
5761 		 * happen, commands should be filtered out above us.
5762 		 */
5763 		ctl_set_invalid_opcode(ctsio);
5764 		ctl_done((union ctl_io *)ctsio);
5765 
5766 		return (CTL_RETVAL_COMPLETE);
5767 		break; /* NOTREACHED */
5768 	}
5769 
5770 	/* ANCHOR flag can be used only together with UNMAP */
5771 	if ((byte2 & SWS_UNMAP) == 0 && (byte2 & SWS_ANCHOR) != 0) {
5772 		ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
5773 		    /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
5774 		ctl_done((union ctl_io *)ctsio);
5775 		return (CTL_RETVAL_COMPLETE);
5776 	}
5777 
5778 	/*
5779 	 * The first check is to make sure we're in bounds, the second
5780 	 * check is to catch wrap-around problems.  If the lba + num blocks
5781 	 * is less than the lba, then we've wrapped around and the block
5782 	 * range is invalid anyway.
5783 	 */
5784 	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5785 	 || ((lba + num_blocks) < lba)) {
5786 		ctl_set_lba_out_of_range(ctsio,
5787 		    MAX(lba, lun->be_lun->maxlba + 1));
5788 		ctl_done((union ctl_io *)ctsio);
5789 		return (CTL_RETVAL_COMPLETE);
5790 	}
5791 
5792 	/* Zero number of blocks means "to the last logical block" */
5793 	if (num_blocks == 0) {
5794 		ival = UINT64_MAX;
5795 		val = dnvlist_get_string(lun->be_lun->options,
5796 		    "write_same_max_lba", NULL);
5797 		if (val != NULL)
5798 			ctl_expand_number(val, &ival);
5799 		if ((lun->be_lun->maxlba + 1) - lba > ival) {
5800 			ctl_set_invalid_field(ctsio,
5801 			    /*sks_valid*/ 1, /*command*/ 1,
5802 			    /*field*/ ctsio->cdb[0] == WRITE_SAME_10 ? 7 : 10,
5803 			    /*bit_valid*/ 0, /*bit*/ 0);
5804 			ctl_done((union ctl_io *)ctsio);
5805 			return (CTL_RETVAL_COMPLETE);
5806 		}
5807 		if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5808 			ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
5809 			ctsio->io_cont = ctl_write_same_cont;
5810 			num_blocks = 1 << 31;
5811 		} else
5812 			num_blocks = (lun->be_lun->maxlba + 1) - lba;
5813 	}
5814 
5815 	len = lun->be_lun->blocksize;
5816 
5817 	/*
5818 	 * If we've got a kernel request that hasn't been malloced yet,
5819 	 * malloc it and tell the caller the data buffer is here.
5820 	 */
5821 	if ((byte2 & SWS_NDOB) == 0 &&
5822 	    (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5823 		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);
5824 		ctsio->kern_data_len = len;
5825 		ctsio->kern_total_len = len;
5826 		ctsio->kern_rel_offset = 0;
5827 		ctsio->kern_sg_entries = 0;
5828 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5829 		ctsio->be_move_done = ctl_config_move_done;
5830 		ctl_datamove((union ctl_io *)ctsio);
5831 
5832 		return (CTL_RETVAL_COMPLETE);
5833 	}
5834 
5835 	lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5836 	lbalen->lba = lba;
5837 	lbalen->len = num_blocks;
5838 	lbalen->flags = byte2;
5839 	retval = lun->backend->config_write((union ctl_io *)ctsio);
5840 
5841 	return (retval);
5842 }
5843 
5844 int
5845 ctl_unmap(struct ctl_scsiio *ctsio)
5846 {
5847 	struct ctl_lun *lun = CTL_LUN(ctsio);
5848 	struct scsi_unmap *cdb;
5849 	struct ctl_ptr_len_flags *ptrlen;
5850 	struct scsi_unmap_header *hdr;
5851 	struct scsi_unmap_desc *buf, *end, *endnz, *range;
5852 	uint64_t lba;
5853 	uint32_t num_blocks;
5854 	int len, retval;
5855 	uint8_t byte2;
5856 
5857 	CTL_DEBUG_PRINT(("ctl_unmap\n"));
5858 
5859 	cdb = (struct scsi_unmap *)ctsio->cdb;
5860 	len = scsi_2btoul(cdb->length);
5861 	byte2 = cdb->byte2;
5862 
5863 	/*
5864 	 * If we've got a kernel request that hasn't been malloced yet,
5865 	 * malloc it and tell the caller the data buffer is here.
5866 	 */
5867 	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5868 		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);
5869 		ctsio->kern_data_len = len;
5870 		ctsio->kern_total_len = len;
5871 		ctsio->kern_rel_offset = 0;
5872 		ctsio->kern_sg_entries = 0;
5873 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5874 		ctsio->be_move_done = ctl_config_move_done;
5875 		ctl_datamove((union ctl_io *)ctsio);
5876 
5877 		return (CTL_RETVAL_COMPLETE);
5878 	}
5879 
5880 	len = ctsio->kern_total_len - ctsio->kern_data_resid;
5881 	hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
5882 	if (len < sizeof (*hdr) ||
5883 	    len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
5884 	    len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
5885 	    scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
5886 		ctl_set_invalid_field(ctsio,
5887 				      /*sks_valid*/ 0,
5888 				      /*command*/ 0,
5889 				      /*field*/ 0,
5890 				      /*bit_valid*/ 0,
5891 				      /*bit*/ 0);
5892 		goto done;
5893 	}
5894 	len = scsi_2btoul(hdr->desc_length);
5895 	buf = (struct scsi_unmap_desc *)(hdr + 1);
5896 	end = buf + len / sizeof(*buf);
5897 
5898 	endnz = buf;
5899 	for (range = buf; range < end; range++) {
5900 		lba = scsi_8btou64(range->lba);
5901 		num_blocks = scsi_4btoul(range->length);
5902 		if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5903 		 || ((lba + num_blocks) < lba)) {
5904 			ctl_set_lba_out_of_range(ctsio,
5905 			    MAX(lba, lun->be_lun->maxlba + 1));
5906 			ctl_done((union ctl_io *)ctsio);
5907 			return (CTL_RETVAL_COMPLETE);
5908 		}
5909 		if (num_blocks != 0)
5910 			endnz = range + 1;
5911 	}
5912 
5913 	/*
5914 	 * Block backend can not handle zero last range.
5915 	 * Filter it out and return if there is nothing left.
5916 	 */
5917 	len = (uint8_t *)endnz - (uint8_t *)buf;
5918 	if (len == 0) {
5919 		ctl_set_success(ctsio);
5920 		goto done;
5921 	}
5922 
5923 	mtx_lock(&lun->lun_lock);
5924 	ptrlen = (struct ctl_ptr_len_flags *)
5925 	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5926 	ptrlen->ptr = (void *)buf;
5927 	ptrlen->len = len;
5928 	ptrlen->flags = byte2;
5929 	ctl_try_unblock_others(lun, (union ctl_io *)ctsio, FALSE);
5930 	mtx_unlock(&lun->lun_lock);
5931 
5932 	retval = lun->backend->config_write((union ctl_io *)ctsio);
5933 	return (retval);
5934 
5935 done:
5936 	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5937 		free(ctsio->kern_data_ptr, M_CTL);
5938 		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5939 	}
5940 	ctl_done((union ctl_io *)ctsio);
5941 	return (CTL_RETVAL_COMPLETE);
5942 }
5943 
5944 int
5945 ctl_default_page_handler(struct ctl_scsiio *ctsio,
5946 			 struct ctl_page_index *page_index, uint8_t *page_ptr)
5947 {
5948 	struct ctl_lun *lun = CTL_LUN(ctsio);
5949 	uint8_t *current_cp;
5950 	int set_ua;
5951 	uint32_t initidx;
5952 
5953 	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5954 	set_ua = 0;
5955 
5956 	current_cp = (page_index->page_data + (page_index->page_len *
5957 	    CTL_PAGE_CURRENT));
5958 
5959 	mtx_lock(&lun->lun_lock);
5960 	if (memcmp(current_cp, page_ptr, page_index->page_len)) {
5961 		memcpy(current_cp, page_ptr, page_index->page_len);
5962 		set_ua = 1;
5963 	}
5964 	if (set_ua != 0)
5965 		ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
5966 	mtx_unlock(&lun->lun_lock);
5967 	if (set_ua) {
5968 		ctl_isc_announce_mode(lun,
5969 		    ctl_get_initindex(&ctsio->io_hdr.nexus),
5970 		    page_index->page_code, page_index->subpage);
5971 	}
5972 	return (CTL_RETVAL_COMPLETE);
5973 }
5974 
5975 static void
5976 ctl_ie_timer(void *arg)
5977 {
5978 	struct ctl_lun *lun = arg;
5979 	uint64_t t;
5980 
5981 	if (lun->ie_asc == 0)
5982 		return;
5983 
5984 	if (lun->MODE_IE.mrie == SIEP_MRIE_UA)
5985 		ctl_est_ua_all(lun, -1, CTL_UA_IE);
5986 	else
5987 		lun->ie_reported = 0;
5988 
5989 	if (lun->ie_reportcnt < scsi_4btoul(lun->MODE_IE.report_count)) {
5990 		lun->ie_reportcnt++;
5991 		t = scsi_4btoul(lun->MODE_IE.interval_timer);
5992 		if (t == 0 || t == UINT32_MAX)
5993 			t = 3000;  /* 5 min */
5994 		callout_schedule_sbt(&lun->ie_callout, SBT_1S / 10 * t,
5995 		    SBT_1S / 10, 0);
5996 	}
5997 }
5998 
5999 int
6000 ctl_ie_page_handler(struct ctl_scsiio *ctsio,
6001 			 struct ctl_page_index *page_index, uint8_t *page_ptr)
6002 {
6003 	struct ctl_lun *lun = CTL_LUN(ctsio);
6004 	struct scsi_info_exceptions_page *pg;
6005 	uint64_t t;
6006 
6007 	(void)ctl_default_page_handler(ctsio, page_index, page_ptr);
6008 
6009 	pg = (struct scsi_info_exceptions_page *)page_ptr;
6010 	mtx_lock(&lun->lun_lock);
6011 	if (pg->info_flags & SIEP_FLAGS_TEST) {
6012 		lun->ie_asc = 0x5d;
6013 		lun->ie_ascq = 0xff;
6014 		if (pg->mrie == SIEP_MRIE_UA) {
6015 			ctl_est_ua_all(lun, -1, CTL_UA_IE);
6016 			lun->ie_reported = 1;
6017 		} else {
6018 			ctl_clr_ua_all(lun, -1, CTL_UA_IE);
6019 			lun->ie_reported = -1;
6020 		}
6021 		lun->ie_reportcnt = 1;
6022 		if (lun->ie_reportcnt < scsi_4btoul(pg->report_count)) {
6023 			lun->ie_reportcnt++;
6024 			t = scsi_4btoul(pg->interval_timer);
6025 			if (t == 0 || t == UINT32_MAX)
6026 				t = 3000;  /* 5 min */
6027 			callout_reset_sbt(&lun->ie_callout, SBT_1S / 10 * t,
6028 			    SBT_1S / 10, ctl_ie_timer, lun, 0);
6029 		}
6030 	} else {
6031 		lun->ie_asc = 0;
6032 		lun->ie_ascq = 0;
6033 		lun->ie_reported = 1;
6034 		ctl_clr_ua_all(lun, -1, CTL_UA_IE);
6035 		lun->ie_reportcnt = UINT32_MAX;
6036 		callout_stop(&lun->ie_callout);
6037 	}
6038 	mtx_unlock(&lun->lun_lock);
6039 	return (CTL_RETVAL_COMPLETE);
6040 }
6041 
6042 static int
6043 ctl_do_mode_select(union ctl_io *io)
6044 {
6045 	struct ctl_lun *lun = CTL_LUN(io);
6046 	struct scsi_mode_page_header *page_header;
6047 	struct ctl_page_index *page_index;
6048 	struct ctl_scsiio *ctsio;
6049 	int page_len, page_len_offset, page_len_size;
6050 	union ctl_modepage_info *modepage_info;
6051 	uint16_t *len_left, *len_used;
6052 	int retval, i;
6053 
6054 	ctsio = &io->scsiio;
6055 	page_index = NULL;
6056 	page_len = 0;
6057 
6058 	modepage_info = (union ctl_modepage_info *)
6059 		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6060 	len_left = &modepage_info->header.len_left;
6061 	len_used = &modepage_info->header.len_used;
6062 
6063 do_next_page:
6064 
6065 	page_header = (struct scsi_mode_page_header *)
6066 		(ctsio->kern_data_ptr + *len_used);
6067 
6068 	if (*len_left == 0) {
6069 		free(ctsio->kern_data_ptr, M_CTL);
6070 		ctl_set_success(ctsio);
6071 		ctl_done((union ctl_io *)ctsio);
6072 		return (CTL_RETVAL_COMPLETE);
6073 	} else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6074 		free(ctsio->kern_data_ptr, M_CTL);
6075 		ctl_set_param_len_error(ctsio);
6076 		ctl_done((union ctl_io *)ctsio);
6077 		return (CTL_RETVAL_COMPLETE);
6078 
6079 	} else if ((page_header->page_code & SMPH_SPF)
6080 		&& (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6081 		free(ctsio->kern_data_ptr, M_CTL);
6082 		ctl_set_param_len_error(ctsio);
6083 		ctl_done((union ctl_io *)ctsio);
6084 		return (CTL_RETVAL_COMPLETE);
6085 	}
6086 
6087 	/*
6088 	 * XXX KDM should we do something with the block descriptor?
6089 	 */
6090 	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6091 		page_index = &lun->mode_pages.index[i];
6092 		if (lun->be_lun->lun_type == T_DIRECT &&
6093 		    (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6094 			continue;
6095 		if (lun->be_lun->lun_type == T_PROCESSOR &&
6096 		    (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6097 			continue;
6098 		if (lun->be_lun->lun_type == T_CDROM &&
6099 		    (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6100 			continue;
6101 
6102 		if ((page_index->page_code & SMPH_PC_MASK) !=
6103 		    (page_header->page_code & SMPH_PC_MASK))
6104 			continue;
6105 
6106 		/*
6107 		 * If neither page has a subpage code, then we've got a
6108 		 * match.
6109 		 */
6110 		if (((page_index->page_code & SMPH_SPF) == 0)
6111 		 && ((page_header->page_code & SMPH_SPF) == 0)) {
6112 			page_len = page_header->page_length;
6113 			break;
6114 		}
6115 
6116 		/*
6117 		 * If both pages have subpages, then the subpage numbers
6118 		 * have to match.
6119 		 */
6120 		if ((page_index->page_code & SMPH_SPF)
6121 		  && (page_header->page_code & SMPH_SPF)) {
6122 			struct scsi_mode_page_header_sp *sph;
6123 
6124 			sph = (struct scsi_mode_page_header_sp *)page_header;
6125 			if (page_index->subpage == sph->subpage) {
6126 				page_len = scsi_2btoul(sph->page_length);
6127 				break;
6128 			}
6129 		}
6130 	}
6131 
6132 	/*
6133 	 * If we couldn't find the page, or if we don't have a mode select
6134 	 * handler for it, send back an error to the user.
6135 	 */
6136 	if ((i >= CTL_NUM_MODE_PAGES)
6137 	 || (page_index->select_handler == NULL)) {
6138 		ctl_set_invalid_field(ctsio,
6139 				      /*sks_valid*/ 1,
6140 				      /*command*/ 0,
6141 				      /*field*/ *len_used,
6142 				      /*bit_valid*/ 0,
6143 				      /*bit*/ 0);
6144 		free(ctsio->kern_data_ptr, M_CTL);
6145 		ctl_done((union ctl_io *)ctsio);
6146 		return (CTL_RETVAL_COMPLETE);
6147 	}
6148 
6149 	if (page_index->page_code & SMPH_SPF) {
6150 		page_len_offset = 2;
6151 		page_len_size = 2;
6152 	} else {
6153 		page_len_size = 1;
6154 		page_len_offset = 1;
6155 	}
6156 
6157 	/*
6158 	 * If the length the initiator gives us isn't the one we specify in
6159 	 * the mode page header, or if they didn't specify enough data in
6160 	 * the CDB to avoid truncating this page, kick out the request.
6161 	 */
6162 	if (page_len != page_index->page_len - page_len_offset - page_len_size) {
6163 		ctl_set_invalid_field(ctsio,
6164 				      /*sks_valid*/ 1,
6165 				      /*command*/ 0,
6166 				      /*field*/ *len_used + page_len_offset,
6167 				      /*bit_valid*/ 0,
6168 				      /*bit*/ 0);
6169 		free(ctsio->kern_data_ptr, M_CTL);
6170 		ctl_done((union ctl_io *)ctsio);
6171 		return (CTL_RETVAL_COMPLETE);
6172 	}
6173 	if (*len_left < page_index->page_len) {
6174 		free(ctsio->kern_data_ptr, M_CTL);
6175 		ctl_set_param_len_error(ctsio);
6176 		ctl_done((union ctl_io *)ctsio);
6177 		return (CTL_RETVAL_COMPLETE);
6178 	}
6179 
6180 	/*
6181 	 * Run through the mode page, checking to make sure that the bits
6182 	 * the user changed are actually legal for him to change.
6183 	 */
6184 	for (i = 0; i < page_index->page_len; i++) {
6185 		uint8_t *user_byte, *change_mask, *current_byte;
6186 		int bad_bit;
6187 		int j;
6188 
6189 		user_byte = (uint8_t *)page_header + i;
6190 		change_mask = page_index->page_data +
6191 			      (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6192 		current_byte = page_index->page_data +
6193 			       (page_index->page_len * CTL_PAGE_CURRENT) + i;
6194 
6195 		/*
6196 		 * Check to see whether the user set any bits in this byte
6197 		 * that he is not allowed to set.
6198 		 */
6199 		if ((*user_byte & ~(*change_mask)) ==
6200 		    (*current_byte & ~(*change_mask)))
6201 			continue;
6202 
6203 		/*
6204 		 * Go through bit by bit to determine which one is illegal.
6205 		 */
6206 		bad_bit = 0;
6207 		for (j = 7; j >= 0; j--) {
6208 			if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6209 			    (((1 << i) & ~(*change_mask)) & *current_byte)) {
6210 				bad_bit = i;
6211 				break;
6212 			}
6213 		}
6214 		ctl_set_invalid_field(ctsio,
6215 				      /*sks_valid*/ 1,
6216 				      /*command*/ 0,
6217 				      /*field*/ *len_used + i,
6218 				      /*bit_valid*/ 1,
6219 				      /*bit*/ bad_bit);
6220 		free(ctsio->kern_data_ptr, M_CTL);
6221 		ctl_done((union ctl_io *)ctsio);
6222 		return (CTL_RETVAL_COMPLETE);
6223 	}
6224 
6225 	/*
6226 	 * Decrement these before we call the page handler, since we may
6227 	 * end up getting called back one way or another before the handler
6228 	 * returns to this context.
6229 	 */
6230 	*len_left -= page_index->page_len;
6231 	*len_used += page_index->page_len;
6232 
6233 	retval = page_index->select_handler(ctsio, page_index,
6234 					    (uint8_t *)page_header);
6235 
6236 	/*
6237 	 * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6238 	 * wait until this queued command completes to finish processing
6239 	 * the mode page.  If it returns anything other than
6240 	 * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6241 	 * already set the sense information, freed the data pointer, and
6242 	 * completed the io for us.
6243 	 */
6244 	if (retval != CTL_RETVAL_COMPLETE)
6245 		goto bailout_no_done;
6246 
6247 	/*
6248 	 * If the initiator sent us more than one page, parse the next one.
6249 	 */
6250 	if (*len_left > 0)
6251 		goto do_next_page;
6252 
6253 	ctl_set_success(ctsio);
6254 	free(ctsio->kern_data_ptr, M_CTL);
6255 	ctl_done((union ctl_io *)ctsio);
6256 
6257 bailout_no_done:
6258 
6259 	return (CTL_RETVAL_COMPLETE);
6260 
6261 }
6262 
6263 int
6264 ctl_mode_select(struct ctl_scsiio *ctsio)
6265 {
6266 	struct ctl_lun *lun = CTL_LUN(ctsio);
6267 	union ctl_modepage_info *modepage_info;
6268 	int bd_len, i, header_size, param_len, rtd;
6269 	uint32_t initidx;
6270 
6271 	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6272 	switch (ctsio->cdb[0]) {
6273 	case MODE_SELECT_6: {
6274 		struct scsi_mode_select_6 *cdb;
6275 
6276 		cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6277 
6278 		rtd = (cdb->byte2 & SMS_RTD) ? 1 : 0;
6279 		param_len = cdb->length;
6280 		header_size = sizeof(struct scsi_mode_header_6);
6281 		break;
6282 	}
6283 	case MODE_SELECT_10: {
6284 		struct scsi_mode_select_10 *cdb;
6285 
6286 		cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6287 
6288 		rtd = (cdb->byte2 & SMS_RTD) ? 1 : 0;
6289 		param_len = scsi_2btoul(cdb->length);
6290 		header_size = sizeof(struct scsi_mode_header_10);
6291 		break;
6292 	}
6293 	default:
6294 		ctl_set_invalid_opcode(ctsio);
6295 		ctl_done((union ctl_io *)ctsio);
6296 		return (CTL_RETVAL_COMPLETE);
6297 	}
6298 
6299 	if (rtd) {
6300 		if (param_len != 0) {
6301 			ctl_set_invalid_field(ctsio, /*sks_valid*/ 0,
6302 			    /*command*/ 1, /*field*/ 0,
6303 			    /*bit_valid*/ 0, /*bit*/ 0);
6304 			ctl_done((union ctl_io *)ctsio);
6305 			return (CTL_RETVAL_COMPLETE);
6306 		}
6307 
6308 		/* Revert to defaults. */
6309 		ctl_init_page_index(lun);
6310 		mtx_lock(&lun->lun_lock);
6311 		ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
6312 		mtx_unlock(&lun->lun_lock);
6313 		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6314 			ctl_isc_announce_mode(lun, -1,
6315 			    lun->mode_pages.index[i].page_code & SMPH_PC_MASK,
6316 			    lun->mode_pages.index[i].subpage);
6317 		}
6318 		ctl_set_success(ctsio);
6319 		ctl_done((union ctl_io *)ctsio);
6320 		return (CTL_RETVAL_COMPLETE);
6321 	}
6322 
6323 	/*
6324 	 * From SPC-3:
6325 	 * "A parameter list length of zero indicates that the Data-Out Buffer
6326 	 * shall be empty. This condition shall not be considered as an error."
6327 	 */
6328 	if (param_len == 0) {
6329 		ctl_set_success(ctsio);
6330 		ctl_done((union ctl_io *)ctsio);
6331 		return (CTL_RETVAL_COMPLETE);
6332 	}
6333 
6334 	/*
6335 	 * Since we'll hit this the first time through, prior to
6336 	 * allocation, we don't need to free a data buffer here.
6337 	 */
6338 	if (param_len < header_size) {
6339 		ctl_set_param_len_error(ctsio);
6340 		ctl_done((union ctl_io *)ctsio);
6341 		return (CTL_RETVAL_COMPLETE);
6342 	}
6343 
6344 	/*
6345 	 * Allocate the data buffer and grab the user's data.  In theory,
6346 	 * we shouldn't have to sanity check the parameter list length here
6347 	 * because the maximum size is 64K.  We should be able to malloc
6348 	 * that much without too many problems.
6349 	 */
6350 	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6351 		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6352 		ctsio->kern_data_len = param_len;
6353 		ctsio->kern_total_len = param_len;
6354 		ctsio->kern_rel_offset = 0;
6355 		ctsio->kern_sg_entries = 0;
6356 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6357 		ctsio->be_move_done = ctl_config_move_done;
6358 		ctl_datamove((union ctl_io *)ctsio);
6359 
6360 		return (CTL_RETVAL_COMPLETE);
6361 	}
6362 
6363 	switch (ctsio->cdb[0]) {
6364 	case MODE_SELECT_6: {
6365 		struct scsi_mode_header_6 *mh6;
6366 
6367 		mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6368 		bd_len = mh6->blk_desc_len;
6369 		break;
6370 	}
6371 	case MODE_SELECT_10: {
6372 		struct scsi_mode_header_10 *mh10;
6373 
6374 		mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6375 		bd_len = scsi_2btoul(mh10->blk_desc_len);
6376 		break;
6377 	}
6378 	default:
6379 		panic("%s: Invalid CDB type %#x", __func__, ctsio->cdb[0]);
6380 	}
6381 
6382 	if (param_len < (header_size + bd_len)) {
6383 		free(ctsio->kern_data_ptr, M_CTL);
6384 		ctl_set_param_len_error(ctsio);
6385 		ctl_done((union ctl_io *)ctsio);
6386 		return (CTL_RETVAL_COMPLETE);
6387 	}
6388 
6389 	/*
6390 	 * Set the IO_CONT flag, so that if this I/O gets passed to
6391 	 * ctl_config_write_done(), it'll get passed back to
6392 	 * ctl_do_mode_select() for further processing, or completion if
6393 	 * we're all done.
6394 	 */
6395 	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6396 	ctsio->io_cont = ctl_do_mode_select;
6397 
6398 	modepage_info = (union ctl_modepage_info *)
6399 		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6400 	memset(modepage_info, 0, sizeof(*modepage_info));
6401 	modepage_info->header.len_left = param_len - header_size - bd_len;
6402 	modepage_info->header.len_used = header_size + bd_len;
6403 
6404 	return (ctl_do_mode_select((union ctl_io *)ctsio));
6405 }
6406 
6407 int
6408 ctl_mode_sense(struct ctl_scsiio *ctsio)
6409 {
6410 	struct ctl_lun *lun = CTL_LUN(ctsio);
6411 	int pc, page_code, llba, subpage;
6412 	int alloc_len, page_len, header_len, bd_len, total_len;
6413 	void *block_desc;
6414 	struct ctl_page_index *page_index;
6415 
6416 	llba = 0;
6417 
6418 	CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6419 
6420 	switch (ctsio->cdb[0]) {
6421 	case MODE_SENSE_6: {
6422 		struct scsi_mode_sense_6 *cdb;
6423 
6424 		cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6425 
6426 		header_len = sizeof(struct scsi_mode_hdr_6);
6427 		if (cdb->byte2 & SMS_DBD)
6428 			bd_len = 0;
6429 		else
6430 			bd_len = sizeof(struct scsi_mode_block_descr);
6431 		header_len += bd_len;
6432 
6433 		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6434 		page_code = cdb->page & SMS_PAGE_CODE;
6435 		subpage = cdb->subpage;
6436 		alloc_len = cdb->length;
6437 		break;
6438 	}
6439 	case MODE_SENSE_10: {
6440 		struct scsi_mode_sense_10 *cdb;
6441 
6442 		cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6443 
6444 		header_len = sizeof(struct scsi_mode_hdr_10);
6445 		if (cdb->byte2 & SMS_DBD) {
6446 			bd_len = 0;
6447 		} else if (lun->be_lun->lun_type == T_DIRECT) {
6448 			if (cdb->byte2 & SMS10_LLBAA) {
6449 				llba = 1;
6450 				bd_len = sizeof(struct scsi_mode_block_descr_dlong);
6451 			} else
6452 				bd_len = sizeof(struct scsi_mode_block_descr_dshort);
6453 		} else
6454 			bd_len = sizeof(struct scsi_mode_block_descr);
6455 		header_len += bd_len;
6456 
6457 		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6458 		page_code = cdb->page & SMS_PAGE_CODE;
6459 		subpage = cdb->subpage;
6460 		alloc_len = scsi_2btoul(cdb->length);
6461 		break;
6462 	}
6463 	default:
6464 		ctl_set_invalid_opcode(ctsio);
6465 		ctl_done((union ctl_io *)ctsio);
6466 		return (CTL_RETVAL_COMPLETE);
6467 		break; /* NOTREACHED */
6468 	}
6469 
6470 	/*
6471 	 * We have to make a first pass through to calculate the size of
6472 	 * the pages that match the user's query.  Then we allocate enough
6473 	 * memory to hold it, and actually copy the data into the buffer.
6474 	 */
6475 	switch (page_code) {
6476 	case SMS_ALL_PAGES_PAGE: {
6477 		u_int i;
6478 
6479 		page_len = 0;
6480 
6481 		/*
6482 		 * At the moment, values other than 0 and 0xff here are
6483 		 * reserved according to SPC-3.
6484 		 */
6485 		if ((subpage != SMS_SUBPAGE_PAGE_0)
6486 		 && (subpage != SMS_SUBPAGE_ALL)) {
6487 			ctl_set_invalid_field(ctsio,
6488 					      /*sks_valid*/ 1,
6489 					      /*command*/ 1,
6490 					      /*field*/ 3,
6491 					      /*bit_valid*/ 0,
6492 					      /*bit*/ 0);
6493 			ctl_done((union ctl_io *)ctsio);
6494 			return (CTL_RETVAL_COMPLETE);
6495 		}
6496 
6497 		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6498 			page_index = &lun->mode_pages.index[i];
6499 
6500 			/* Make sure the page is supported for this dev type */
6501 			if (lun->be_lun->lun_type == T_DIRECT &&
6502 			    (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6503 				continue;
6504 			if (lun->be_lun->lun_type == T_PROCESSOR &&
6505 			    (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6506 				continue;
6507 			if (lun->be_lun->lun_type == T_CDROM &&
6508 			    (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6509 				continue;
6510 
6511 			/*
6512 			 * We don't use this subpage if the user didn't
6513 			 * request all subpages.
6514 			 */
6515 			if ((page_index->subpage != 0)
6516 			 && (subpage == SMS_SUBPAGE_PAGE_0))
6517 				continue;
6518 
6519 			page_len += page_index->page_len;
6520 		}
6521 		break;
6522 	}
6523 	default: {
6524 		u_int i;
6525 
6526 		page_len = 0;
6527 
6528 		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6529 			page_index = &lun->mode_pages.index[i];
6530 
6531 			/* Make sure the page is supported for this dev type */
6532 			if (lun->be_lun->lun_type == T_DIRECT &&
6533 			    (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6534 				continue;
6535 			if (lun->be_lun->lun_type == T_PROCESSOR &&
6536 			    (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6537 				continue;
6538 			if (lun->be_lun->lun_type == T_CDROM &&
6539 			    (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6540 				continue;
6541 
6542 			/* Look for the right page code */
6543 			if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6544 				continue;
6545 
6546 			/* Look for the right subpage or the subpage wildcard*/
6547 			if ((page_index->subpage != subpage)
6548 			 && (subpage != SMS_SUBPAGE_ALL))
6549 				continue;
6550 
6551 			page_len += page_index->page_len;
6552 		}
6553 
6554 		if (page_len == 0) {
6555 			ctl_set_invalid_field(ctsio,
6556 					      /*sks_valid*/ 1,
6557 					      /*command*/ 1,
6558 					      /*field*/ 2,
6559 					      /*bit_valid*/ 1,
6560 					      /*bit*/ 5);
6561 			ctl_done((union ctl_io *)ctsio);
6562 			return (CTL_RETVAL_COMPLETE);
6563 		}
6564 		break;
6565 	}
6566 	}
6567 
6568 	total_len = header_len + page_len;
6569 
6570 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6571 	ctsio->kern_sg_entries = 0;
6572 	ctsio->kern_rel_offset = 0;
6573 	ctsio->kern_data_len = min(total_len, alloc_len);
6574 	ctsio->kern_total_len = ctsio->kern_data_len;
6575 
6576 	switch (ctsio->cdb[0]) {
6577 	case MODE_SENSE_6: {
6578 		struct scsi_mode_hdr_6 *header;
6579 
6580 		header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6581 
6582 		header->datalen = MIN(total_len - 1, 254);
6583 		if (lun->be_lun->lun_type == T_DIRECT) {
6584 			header->dev_specific = 0x10; /* DPOFUA */
6585 			if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6586 			    (lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0)
6587 				header->dev_specific |= 0x80; /* WP */
6588 		}
6589 		header->block_descr_len = bd_len;
6590 		block_desc = &header[1];
6591 		break;
6592 	}
6593 	case MODE_SENSE_10: {
6594 		struct scsi_mode_hdr_10 *header;
6595 		int datalen;
6596 
6597 		header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6598 
6599 		datalen = MIN(total_len - 2, 65533);
6600 		scsi_ulto2b(datalen, header->datalen);
6601 		if (lun->be_lun->lun_type == T_DIRECT) {
6602 			header->dev_specific = 0x10; /* DPOFUA */
6603 			if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6604 			    (lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0)
6605 				header->dev_specific |= 0x80; /* WP */
6606 		}
6607 		if (llba)
6608 			header->flags |= SMH_LONGLBA;
6609 		scsi_ulto2b(bd_len, header->block_descr_len);
6610 		block_desc = &header[1];
6611 		break;
6612 	}
6613 	default:
6614 		panic("%s: Invalid CDB type %#x", __func__, ctsio->cdb[0]);
6615 	}
6616 
6617 	/*
6618 	 * If we've got a disk, use its blocksize in the block
6619 	 * descriptor.  Otherwise, just set it to 0.
6620 	 */
6621 	if (bd_len > 0) {
6622 		if (lun->be_lun->lun_type == T_DIRECT) {
6623 			if (llba) {
6624 				struct scsi_mode_block_descr_dlong *bd = block_desc;
6625 				if (lun->be_lun->maxlba != 0)
6626 					scsi_u64to8b(lun->be_lun->maxlba + 1,
6627 					    bd->num_blocks);
6628 				scsi_ulto4b(lun->be_lun->blocksize,
6629 				    bd->block_len);
6630 			} else {
6631 				struct scsi_mode_block_descr_dshort *bd = block_desc;
6632 				if (lun->be_lun->maxlba != 0)
6633 					scsi_ulto4b(MIN(lun->be_lun->maxlba+1,
6634 					    UINT32_MAX), bd->num_blocks);
6635 				scsi_ulto3b(lun->be_lun->blocksize,
6636 				    bd->block_len);
6637 			}
6638 		} else {
6639 			struct scsi_mode_block_descr *bd = block_desc;
6640 			scsi_ulto3b(0, bd->block_len);
6641 		}
6642 	}
6643 
6644 	switch (page_code) {
6645 	case SMS_ALL_PAGES_PAGE: {
6646 		int i, data_used;
6647 
6648 		data_used = header_len;
6649 		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6650 			struct ctl_page_index *page_index;
6651 
6652 			page_index = &lun->mode_pages.index[i];
6653 			if (lun->be_lun->lun_type == T_DIRECT &&
6654 			    (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6655 				continue;
6656 			if (lun->be_lun->lun_type == T_PROCESSOR &&
6657 			    (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6658 				continue;
6659 			if (lun->be_lun->lun_type == T_CDROM &&
6660 			    (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6661 				continue;
6662 
6663 			/*
6664 			 * We don't use this subpage if the user didn't
6665 			 * request all subpages.  We already checked (above)
6666 			 * to make sure the user only specified a subpage
6667 			 * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
6668 			 */
6669 			if ((page_index->subpage != 0)
6670 			 && (subpage == SMS_SUBPAGE_PAGE_0))
6671 				continue;
6672 
6673 			/*
6674 			 * Call the handler, if it exists, to update the
6675 			 * page to the latest values.
6676 			 */
6677 			if (page_index->sense_handler != NULL)
6678 				page_index->sense_handler(ctsio, page_index,pc);
6679 
6680 			memcpy(ctsio->kern_data_ptr + data_used,
6681 			       page_index->page_data +
6682 			       (page_index->page_len * pc),
6683 			       page_index->page_len);
6684 			data_used += page_index->page_len;
6685 		}
6686 		break;
6687 	}
6688 	default: {
6689 		int i, data_used;
6690 
6691 		data_used = header_len;
6692 
6693 		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6694 			struct ctl_page_index *page_index;
6695 
6696 			page_index = &lun->mode_pages.index[i];
6697 
6698 			/* Look for the right page code */
6699 			if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6700 				continue;
6701 
6702 			/* Look for the right subpage or the subpage wildcard*/
6703 			if ((page_index->subpage != subpage)
6704 			 && (subpage != SMS_SUBPAGE_ALL))
6705 				continue;
6706 
6707 			/* Make sure the page is supported for this dev type */
6708 			if (lun->be_lun->lun_type == T_DIRECT &&
6709 			    (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6710 				continue;
6711 			if (lun->be_lun->lun_type == T_PROCESSOR &&
6712 			    (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6713 				continue;
6714 			if (lun->be_lun->lun_type == T_CDROM &&
6715 			    (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6716 				continue;
6717 
6718 			/*
6719 			 * Call the handler, if it exists, to update the
6720 			 * page to the latest values.
6721 			 */
6722 			if (page_index->sense_handler != NULL)
6723 				page_index->sense_handler(ctsio, page_index,pc);
6724 
6725 			memcpy(ctsio->kern_data_ptr + data_used,
6726 			       page_index->page_data +
6727 			       (page_index->page_len * pc),
6728 			       page_index->page_len);
6729 			data_used += page_index->page_len;
6730 		}
6731 		break;
6732 	}
6733 	}
6734 
6735 	ctl_set_success(ctsio);
6736 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6737 	ctsio->be_move_done = ctl_config_move_done;
6738 	ctl_datamove((union ctl_io *)ctsio);
6739 	return (CTL_RETVAL_COMPLETE);
6740 }
6741 
6742 int
6743 ctl_temp_log_sense_handler(struct ctl_scsiio *ctsio,
6744 			       struct ctl_page_index *page_index,
6745 			       int pc)
6746 {
6747 	struct ctl_lun *lun = CTL_LUN(ctsio);
6748 	struct scsi_log_temperature *data;
6749 	const char *value;
6750 
6751 	data = (struct scsi_log_temperature *)page_index->page_data;
6752 
6753 	scsi_ulto2b(SLP_TEMPERATURE, data->hdr.param_code);
6754 	data->hdr.param_control = SLP_LBIN;
6755 	data->hdr.param_len = sizeof(struct scsi_log_temperature) -
6756 	    sizeof(struct scsi_log_param_header);
6757 	if ((value = dnvlist_get_string(lun->be_lun->options, "temperature",
6758 	    NULL)) != NULL)
6759 		data->temperature = strtol(value, NULL, 0);
6760 	else
6761 		data->temperature = 0xff;
6762 	data++;
6763 
6764 	scsi_ulto2b(SLP_REFTEMPERATURE, data->hdr.param_code);
6765 	data->hdr.param_control = SLP_LBIN;
6766 	data->hdr.param_len = sizeof(struct scsi_log_temperature) -
6767 	    sizeof(struct scsi_log_param_header);
6768 	if ((value = dnvlist_get_string(lun->be_lun->options, "reftemperature",
6769 	    NULL)) != NULL)
6770 		data->temperature = strtol(value, NULL, 0);
6771 	else
6772 		data->temperature = 0xff;
6773 	return (0);
6774 }
6775 
6776 int
6777 ctl_lbp_log_sense_handler(struct ctl_scsiio *ctsio,
6778 			       struct ctl_page_index *page_index,
6779 			       int pc)
6780 {
6781 	struct ctl_lun *lun = CTL_LUN(ctsio);
6782 	struct scsi_log_param_header *phdr;
6783 	uint8_t *data;
6784 	uint64_t val;
6785 
6786 	data = page_index->page_data;
6787 
6788 	if (lun->backend->lun_attr != NULL &&
6789 	    (val = lun->backend->lun_attr(lun->be_lun, "blocksavail"))
6790 	     != UINT64_MAX) {
6791 		phdr = (struct scsi_log_param_header *)data;
6792 		scsi_ulto2b(0x0001, phdr->param_code);
6793 		phdr->param_control = SLP_LBIN | SLP_LP;
6794 		phdr->param_len = 8;
6795 		data = (uint8_t *)(phdr + 1);
6796 		scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6797 		data[4] = 0x02; /* per-pool */
6798 		data += phdr->param_len;
6799 	}
6800 
6801 	if (lun->backend->lun_attr != NULL &&
6802 	    (val = lun->backend->lun_attr(lun->be_lun, "blocksused"))
6803 	     != UINT64_MAX) {
6804 		phdr = (struct scsi_log_param_header *)data;
6805 		scsi_ulto2b(0x0002, phdr->param_code);
6806 		phdr->param_control = SLP_LBIN | SLP_LP;
6807 		phdr->param_len = 8;
6808 		data = (uint8_t *)(phdr + 1);
6809 		scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6810 		data[4] = 0x01; /* per-LUN */
6811 		data += phdr->param_len;
6812 	}
6813 
6814 	if (lun->backend->lun_attr != NULL &&
6815 	    (val = lun->backend->lun_attr(lun->be_lun, "poolblocksavail"))
6816 	     != UINT64_MAX) {
6817 		phdr = (struct scsi_log_param_header *)data;
6818 		scsi_ulto2b(0x00f1, phdr->param_code);
6819 		phdr->param_control = SLP_LBIN | SLP_LP;
6820 		phdr->param_len = 8;
6821 		data = (uint8_t *)(phdr + 1);
6822 		scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6823 		data[4] = 0x02; /* per-pool */
6824 		data += phdr->param_len;
6825 	}
6826 
6827 	if (lun->backend->lun_attr != NULL &&
6828 	    (val = lun->backend->lun_attr(lun->be_lun, "poolblocksused"))
6829 	     != UINT64_MAX) {
6830 		phdr = (struct scsi_log_param_header *)data;
6831 		scsi_ulto2b(0x00f2, phdr->param_code);
6832 		phdr->param_control = SLP_LBIN | SLP_LP;
6833 		phdr->param_len = 8;
6834 		data = (uint8_t *)(phdr + 1);
6835 		scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6836 		data[4] = 0x02; /* per-pool */
6837 		data += phdr->param_len;
6838 	}
6839 
6840 	page_index->page_len = data - page_index->page_data;
6841 	return (0);
6842 }
6843 
6844 int
6845 ctl_sap_log_sense_handler(struct ctl_scsiio *ctsio,
6846 			       struct ctl_page_index *page_index,
6847 			       int pc)
6848 {
6849 	struct ctl_lun *lun = CTL_LUN(ctsio);
6850 	struct stat_page *data;
6851 	struct bintime *t;
6852 
6853 	data = (struct stat_page *)page_index->page_data;
6854 
6855 	scsi_ulto2b(SLP_SAP, data->sap.hdr.param_code);
6856 	data->sap.hdr.param_control = SLP_LBIN;
6857 	data->sap.hdr.param_len = sizeof(struct scsi_log_stat_and_perf) -
6858 	    sizeof(struct scsi_log_param_header);
6859 	scsi_u64to8b(lun->stats.operations[CTL_STATS_READ],
6860 	    data->sap.read_num);
6861 	scsi_u64to8b(lun->stats.operations[CTL_STATS_WRITE],
6862 	    data->sap.write_num);
6863 	if (lun->be_lun->blocksize > 0) {
6864 		scsi_u64to8b(lun->stats.bytes[CTL_STATS_WRITE] /
6865 		    lun->be_lun->blocksize, data->sap.recvieved_lba);
6866 		scsi_u64to8b(lun->stats.bytes[CTL_STATS_READ] /
6867 		    lun->be_lun->blocksize, data->sap.transmitted_lba);
6868 	}
6869 	t = &lun->stats.time[CTL_STATS_READ];
6870 	scsi_u64to8b((uint64_t)t->sec * 1000 + t->frac / (UINT64_MAX / 1000),
6871 	    data->sap.read_int);
6872 	t = &lun->stats.time[CTL_STATS_WRITE];
6873 	scsi_u64to8b((uint64_t)t->sec * 1000 + t->frac / (UINT64_MAX / 1000),
6874 	    data->sap.write_int);
6875 	scsi_u64to8b(0, data->sap.weighted_num);
6876 	scsi_u64to8b(0, data->sap.weighted_int);
6877 	scsi_ulto2b(SLP_IT, data->it.hdr.param_code);
6878 	data->it.hdr.param_control = SLP_LBIN;
6879 	data->it.hdr.param_len = sizeof(struct scsi_log_idle_time) -
6880 	    sizeof(struct scsi_log_param_header);
6881 #ifdef CTL_TIME_IO
6882 	scsi_u64to8b(lun->idle_time / SBT_1MS, data->it.idle_int);
6883 #endif
6884 	scsi_ulto2b(SLP_TI, data->ti.hdr.param_code);
6885 	data->it.hdr.param_control = SLP_LBIN;
6886 	data->ti.hdr.param_len = sizeof(struct scsi_log_time_interval) -
6887 	    sizeof(struct scsi_log_param_header);
6888 	scsi_ulto4b(3, data->ti.exponent);
6889 	scsi_ulto4b(1, data->ti.integer);
6890 	return (0);
6891 }
6892 
6893 int
6894 ctl_ie_log_sense_handler(struct ctl_scsiio *ctsio,
6895 			       struct ctl_page_index *page_index,
6896 			       int pc)
6897 {
6898 	struct ctl_lun *lun = CTL_LUN(ctsio);
6899 	struct scsi_log_informational_exceptions *data;
6900 	const char *value;
6901 
6902 	data = (struct scsi_log_informational_exceptions *)page_index->page_data;
6903 
6904 	scsi_ulto2b(SLP_IE_GEN, data->hdr.param_code);
6905 	data->hdr.param_control = SLP_LBIN;
6906 	data->hdr.param_len = sizeof(struct scsi_log_informational_exceptions) -
6907 	    sizeof(struct scsi_log_param_header);
6908 	data->ie_asc = lun->ie_asc;
6909 	data->ie_ascq = lun->ie_ascq;
6910 	if ((value = dnvlist_get_string(lun->be_lun->options, "temperature",
6911 	    NULL)) != NULL)
6912 		data->temperature = strtol(value, NULL, 0);
6913 	else
6914 		data->temperature = 0xff;
6915 	return (0);
6916 }
6917 
6918 int
6919 ctl_log_sense(struct ctl_scsiio *ctsio)
6920 {
6921 	struct ctl_lun *lun = CTL_LUN(ctsio);
6922 	int i, pc, page_code, subpage;
6923 	int alloc_len, total_len;
6924 	struct ctl_page_index *page_index;
6925 	struct scsi_log_sense *cdb;
6926 	struct scsi_log_header *header;
6927 
6928 	CTL_DEBUG_PRINT(("ctl_log_sense\n"));
6929 
6930 	cdb = (struct scsi_log_sense *)ctsio->cdb;
6931 	pc = (cdb->page & SLS_PAGE_CTRL_MASK) >> 6;
6932 	page_code = cdb->page & SLS_PAGE_CODE;
6933 	subpage = cdb->subpage;
6934 	alloc_len = scsi_2btoul(cdb->length);
6935 
6936 	page_index = NULL;
6937 	for (i = 0; i < CTL_NUM_LOG_PAGES; i++) {
6938 		page_index = &lun->log_pages.index[i];
6939 
6940 		/* Look for the right page code */
6941 		if ((page_index->page_code & SL_PAGE_CODE) != page_code)
6942 			continue;
6943 
6944 		/* Look for the right subpage or the subpage wildcard*/
6945 		if (page_index->subpage != subpage)
6946 			continue;
6947 
6948 		break;
6949 	}
6950 	if (i >= CTL_NUM_LOG_PAGES) {
6951 		ctl_set_invalid_field(ctsio,
6952 				      /*sks_valid*/ 1,
6953 				      /*command*/ 1,
6954 				      /*field*/ 2,
6955 				      /*bit_valid*/ 0,
6956 				      /*bit*/ 0);
6957 		ctl_done((union ctl_io *)ctsio);
6958 		return (CTL_RETVAL_COMPLETE);
6959 	}
6960 
6961 	total_len = sizeof(struct scsi_log_header) + page_index->page_len;
6962 
6963 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6964 	ctsio->kern_sg_entries = 0;
6965 	ctsio->kern_rel_offset = 0;
6966 	ctsio->kern_data_len = min(total_len, alloc_len);
6967 	ctsio->kern_total_len = ctsio->kern_data_len;
6968 
6969 	header = (struct scsi_log_header *)ctsio->kern_data_ptr;
6970 	header->page = page_index->page_code;
6971 	if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING)
6972 		header->page |= SL_DS;
6973 	if (page_index->subpage) {
6974 		header->page |= SL_SPF;
6975 		header->subpage = page_index->subpage;
6976 	}
6977 	scsi_ulto2b(page_index->page_len, header->datalen);
6978 
6979 	/*
6980 	 * Call the handler, if it exists, to update the
6981 	 * page to the latest values.
6982 	 */
6983 	if (page_index->sense_handler != NULL)
6984 		page_index->sense_handler(ctsio, page_index, pc);
6985 
6986 	memcpy(header + 1, page_index->page_data, page_index->page_len);
6987 
6988 	ctl_set_success(ctsio);
6989 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6990 	ctsio->be_move_done = ctl_config_move_done;
6991 	ctl_datamove((union ctl_io *)ctsio);
6992 	return (CTL_RETVAL_COMPLETE);
6993 }
6994 
6995 int
6996 ctl_read_capacity(struct ctl_scsiio *ctsio)
6997 {
6998 	struct ctl_lun *lun = CTL_LUN(ctsio);
6999 	struct scsi_read_capacity *cdb;
7000 	struct scsi_read_capacity_data *data;
7001 	uint32_t lba;
7002 
7003 	CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
7004 
7005 	cdb = (struct scsi_read_capacity *)ctsio->cdb;
7006 
7007 	lba = scsi_4btoul(cdb->addr);
7008 	if (((cdb->pmi & SRC_PMI) == 0)
7009 	 && (lba != 0)) {
7010 		ctl_set_invalid_field(/*ctsio*/ ctsio,
7011 				      /*sks_valid*/ 1,
7012 				      /*command*/ 1,
7013 				      /*field*/ 2,
7014 				      /*bit_valid*/ 0,
7015 				      /*bit*/ 0);
7016 		ctl_done((union ctl_io *)ctsio);
7017 		return (CTL_RETVAL_COMPLETE);
7018 	}
7019 
7020 	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7021 	data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
7022 	ctsio->kern_data_len = sizeof(*data);
7023 	ctsio->kern_total_len = sizeof(*data);
7024 	ctsio->kern_rel_offset = 0;
7025 	ctsio->kern_sg_entries = 0;
7026 
7027 	/*
7028 	 * If the maximum LBA is greater than 0xfffffffe, the user must
7029 	 * issue a SERVICE ACTION IN (16) command, with the read capacity
7030 	 * serivce action set.
7031 	 */
7032 	if (lun->be_lun->maxlba > 0xfffffffe)
7033 		scsi_ulto4b(0xffffffff, data->addr);
7034 	else
7035 		scsi_ulto4b(lun->be_lun->maxlba, data->addr);
7036 
7037 	/*
7038 	 * XXX KDM this may not be 512 bytes...
7039 	 */
7040 	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7041 
7042 	ctl_set_success(ctsio);
7043 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7044 	ctsio->be_move_done = ctl_config_move_done;
7045 	ctl_datamove((union ctl_io *)ctsio);
7046 	return (CTL_RETVAL_COMPLETE);
7047 }
7048 
7049 int
7050 ctl_read_capacity_16(struct ctl_scsiio *ctsio)
7051 {
7052 	struct ctl_lun *lun = CTL_LUN(ctsio);
7053 	struct scsi_read_capacity_16 *cdb;
7054 	struct scsi_read_capacity_data_long *data;
7055 	uint64_t lba;
7056 	uint32_t alloc_len;
7057 
7058 	CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
7059 
7060 	cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
7061 
7062 	alloc_len = scsi_4btoul(cdb->alloc_len);
7063 	lba = scsi_8btou64(cdb->addr);
7064 
7065 	if ((cdb->reladr & SRC16_PMI)
7066 	 && (lba != 0)) {
7067 		ctl_set_invalid_field(/*ctsio*/ ctsio,
7068 				      /*sks_valid*/ 1,
7069 				      /*command*/ 1,
7070 				      /*field*/ 2,
7071 				      /*bit_valid*/ 0,
7072 				      /*bit*/ 0);
7073 		ctl_done((union ctl_io *)ctsio);
7074 		return (CTL_RETVAL_COMPLETE);
7075 	}
7076 
7077 	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7078 	data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
7079 	ctsio->kern_rel_offset = 0;
7080 	ctsio->kern_sg_entries = 0;
7081 	ctsio->kern_data_len = min(sizeof(*data), alloc_len);
7082 	ctsio->kern_total_len = ctsio->kern_data_len;
7083 
7084 	scsi_u64to8b(lun->be_lun->maxlba, data->addr);
7085 	/* XXX KDM this may not be 512 bytes... */
7086 	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7087 	data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
7088 	scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
7089 	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
7090 		data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ;
7091 
7092 	ctl_set_success(ctsio);
7093 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7094 	ctsio->be_move_done = ctl_config_move_done;
7095 	ctl_datamove((union ctl_io *)ctsio);
7096 	return (CTL_RETVAL_COMPLETE);
7097 }
7098 
7099 int
7100 ctl_get_lba_status(struct ctl_scsiio *ctsio)
7101 {
7102 	struct ctl_lun *lun = CTL_LUN(ctsio);
7103 	struct scsi_get_lba_status *cdb;
7104 	struct scsi_get_lba_status_data *data;
7105 	struct ctl_lba_len_flags *lbalen;
7106 	uint64_t lba;
7107 	uint32_t alloc_len, total_len;
7108 	int retval;
7109 
7110 	CTL_DEBUG_PRINT(("ctl_get_lba_status\n"));
7111 
7112 	cdb = (struct scsi_get_lba_status *)ctsio->cdb;
7113 	lba = scsi_8btou64(cdb->addr);
7114 	alloc_len = scsi_4btoul(cdb->alloc_len);
7115 
7116 	if (lba > lun->be_lun->maxlba) {
7117 		ctl_set_lba_out_of_range(ctsio, lba);
7118 		ctl_done((union ctl_io *)ctsio);
7119 		return (CTL_RETVAL_COMPLETE);
7120 	}
7121 
7122 	total_len = sizeof(*data) + sizeof(data->descr[0]);
7123 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7124 	data = (struct scsi_get_lba_status_data *)ctsio->kern_data_ptr;
7125 	ctsio->kern_rel_offset = 0;
7126 	ctsio->kern_sg_entries = 0;
7127 	ctsio->kern_data_len = min(total_len, alloc_len);
7128 	ctsio->kern_total_len = ctsio->kern_data_len;
7129 
7130 	/* Fill dummy data in case backend can't tell anything. */
7131 	scsi_ulto4b(4 + sizeof(data->descr[0]), data->length);
7132 	scsi_u64to8b(lba, data->descr[0].addr);
7133 	scsi_ulto4b(MIN(UINT32_MAX, lun->be_lun->maxlba + 1 - lba),
7134 	    data->descr[0].length);
7135 	data->descr[0].status = 0; /* Mapped or unknown. */
7136 
7137 	ctl_set_success(ctsio);
7138 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7139 	ctsio->be_move_done = ctl_config_move_done;
7140 
7141 	lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
7142 	lbalen->lba = lba;
7143 	lbalen->len = total_len;
7144 	lbalen->flags = 0;
7145 	retval = lun->backend->config_read((union ctl_io *)ctsio);
7146 	return (retval);
7147 }
7148 
7149 int
7150 ctl_read_defect(struct ctl_scsiio *ctsio)
7151 {
7152 	struct scsi_read_defect_data_10 *ccb10;
7153 	struct scsi_read_defect_data_12 *ccb12;
7154 	struct scsi_read_defect_data_hdr_10 *data10;
7155 	struct scsi_read_defect_data_hdr_12 *data12;
7156 	uint32_t alloc_len, data_len;
7157 	uint8_t format;
7158 
7159 	CTL_DEBUG_PRINT(("ctl_read_defect\n"));
7160 
7161 	if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7162 		ccb10 = (struct scsi_read_defect_data_10 *)&ctsio->cdb;
7163 		format = ccb10->format;
7164 		alloc_len = scsi_2btoul(ccb10->alloc_length);
7165 		data_len = sizeof(*data10);
7166 	} else {
7167 		ccb12 = (struct scsi_read_defect_data_12 *)&ctsio->cdb;
7168 		format = ccb12->format;
7169 		alloc_len = scsi_4btoul(ccb12->alloc_length);
7170 		data_len = sizeof(*data12);
7171 	}
7172 	if (alloc_len == 0) {
7173 		ctl_set_success(ctsio);
7174 		ctl_done((union ctl_io *)ctsio);
7175 		return (CTL_RETVAL_COMPLETE);
7176 	}
7177 
7178 	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
7179 	ctsio->kern_rel_offset = 0;
7180 	ctsio->kern_sg_entries = 0;
7181 	ctsio->kern_data_len = min(data_len, alloc_len);
7182 	ctsio->kern_total_len = ctsio->kern_data_len;
7183 
7184 	if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7185 		data10 = (struct scsi_read_defect_data_hdr_10 *)
7186 		    ctsio->kern_data_ptr;
7187 		data10->format = format;
7188 		scsi_ulto2b(0, data10->length);
7189 	} else {
7190 		data12 = (struct scsi_read_defect_data_hdr_12 *)
7191 		    ctsio->kern_data_ptr;
7192 		data12->format = format;
7193 		scsi_ulto2b(0, data12->generation);
7194 		scsi_ulto4b(0, data12->length);
7195 	}
7196 
7197 	ctl_set_success(ctsio);
7198 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7199 	ctsio->be_move_done = ctl_config_move_done;
7200 	ctl_datamove((union ctl_io *)ctsio);
7201 	return (CTL_RETVAL_COMPLETE);
7202 }
7203 
7204 int
7205 ctl_report_ident_info(struct ctl_scsiio *ctsio)
7206 {
7207 	struct ctl_lun *lun = CTL_LUN(ctsio);
7208 	struct scsi_report_ident_info *cdb;
7209 	struct scsi_report_ident_info_data *rii_ptr;
7210 	struct scsi_report_ident_info_descr *riid_ptr;
7211 	const char *oii, *otii;
7212 	int retval, alloc_len, total_len = 0, len = 0;
7213 
7214 	CTL_DEBUG_PRINT(("ctl_report_ident_info\n"));
7215 
7216 	cdb = (struct scsi_report_ident_info *)ctsio->cdb;
7217 	retval = CTL_RETVAL_COMPLETE;
7218 
7219 	total_len = sizeof(struct scsi_report_ident_info_data);
7220 	switch (cdb->type) {
7221 	case RII_LUII:
7222 		oii = dnvlist_get_string(lun->be_lun->options,
7223 		    "ident_info", NULL);
7224 		if (oii)
7225 			len = strlen(oii);	/* Approximately */
7226 		break;
7227 	case RII_LUTII:
7228 		otii = dnvlist_get_string(lun->be_lun->options,
7229 		    "text_ident_info", NULL);
7230 		if (otii)
7231 			len = strlen(otii) + 1;	/* NULL-terminated */
7232 		break;
7233 	case RII_IIS:
7234 		len = 2 * sizeof(struct scsi_report_ident_info_descr);
7235 		break;
7236 	default:
7237 		ctl_set_invalid_field(/*ctsio*/ ctsio,
7238 				      /*sks_valid*/ 1,
7239 				      /*command*/ 1,
7240 				      /*field*/ 11,
7241 				      /*bit_valid*/ 1,
7242 				      /*bit*/ 2);
7243 		ctl_done((union ctl_io *)ctsio);
7244 		return(retval);
7245 	}
7246 	total_len += len;
7247 	alloc_len = scsi_4btoul(cdb->length);
7248 
7249 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7250 	ctsio->kern_sg_entries = 0;
7251 	ctsio->kern_rel_offset = 0;
7252 	ctsio->kern_data_len = min(total_len, alloc_len);
7253 	ctsio->kern_total_len = ctsio->kern_data_len;
7254 
7255 	rii_ptr = (struct scsi_report_ident_info_data *)ctsio->kern_data_ptr;
7256 	switch (cdb->type) {
7257 	case RII_LUII:
7258 		if (oii) {
7259 			if (oii[0] == '0' && oii[1] == 'x')
7260 				len = hex2bin(oii, (uint8_t *)(rii_ptr + 1), len);
7261 			else
7262 				strncpy((uint8_t *)(rii_ptr + 1), oii, len);
7263 		}
7264 		break;
7265 	case RII_LUTII:
7266 		if (otii)
7267 			strlcpy((uint8_t *)(rii_ptr + 1), otii, len);
7268 		break;
7269 	case RII_IIS:
7270 		riid_ptr = (struct scsi_report_ident_info_descr *)(rii_ptr + 1);
7271 		riid_ptr->type = RII_LUII;
7272 		scsi_ulto2b(0xffff, riid_ptr->length);
7273 		riid_ptr++;
7274 		riid_ptr->type = RII_LUTII;
7275 		scsi_ulto2b(0xffff, riid_ptr->length);
7276 	}
7277 	scsi_ulto2b(len, rii_ptr->length);
7278 
7279 	ctl_set_success(ctsio);
7280 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7281 	ctsio->be_move_done = ctl_config_move_done;
7282 	ctl_datamove((union ctl_io *)ctsio);
7283 	return(retval);
7284 }
7285 
7286 int
7287 ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7288 {
7289 	struct ctl_softc *softc = CTL_SOFTC(ctsio);
7290 	struct ctl_lun *lun = CTL_LUN(ctsio);
7291 	struct scsi_maintenance_in *cdb;
7292 	int retval;
7293 	int alloc_len, ext, total_len = 0, g, pc, pg, ts, os;
7294 	int num_ha_groups, num_target_ports, shared_group;
7295 	struct ctl_port *port;
7296 	struct scsi_target_group_data *rtg_ptr;
7297 	struct scsi_target_group_data_extended *rtg_ext_ptr;
7298 	struct scsi_target_port_group_descriptor *tpg_desc;
7299 
7300 	CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7301 
7302 	cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7303 	retval = CTL_RETVAL_COMPLETE;
7304 
7305 	switch (cdb->byte2 & STG_PDF_MASK) {
7306 	case STG_PDF_LENGTH:
7307 		ext = 0;
7308 		break;
7309 	case STG_PDF_EXTENDED:
7310 		ext = 1;
7311 		break;
7312 	default:
7313 		ctl_set_invalid_field(/*ctsio*/ ctsio,
7314 				      /*sks_valid*/ 1,
7315 				      /*command*/ 1,
7316 				      /*field*/ 2,
7317 				      /*bit_valid*/ 1,
7318 				      /*bit*/ 5);
7319 		ctl_done((union ctl_io *)ctsio);
7320 		return(retval);
7321 	}
7322 
7323 	num_target_ports = 0;
7324 	shared_group = (softc->is_single != 0);
7325 	mtx_lock(&softc->ctl_lock);
7326 	STAILQ_FOREACH(port, &softc->port_list, links) {
7327 		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7328 			continue;
7329 		if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
7330 			continue;
7331 		num_target_ports++;
7332 		if (port->status & CTL_PORT_STATUS_HA_SHARED)
7333 			shared_group = 1;
7334 	}
7335 	mtx_unlock(&softc->ctl_lock);
7336 	num_ha_groups = (softc->is_single) ? 0 : NUM_HA_SHELVES;
7337 
7338 	if (ext)
7339 		total_len = sizeof(struct scsi_target_group_data_extended);
7340 	else
7341 		total_len = sizeof(struct scsi_target_group_data);
7342 	total_len += sizeof(struct scsi_target_port_group_descriptor) *
7343 		(shared_group + num_ha_groups) +
7344 	    sizeof(struct scsi_target_port_descriptor) * num_target_ports;
7345 
7346 	alloc_len = scsi_4btoul(cdb->length);
7347 
7348 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7349 	ctsio->kern_sg_entries = 0;
7350 	ctsio->kern_rel_offset = 0;
7351 	ctsio->kern_data_len = min(total_len, alloc_len);
7352 	ctsio->kern_total_len = ctsio->kern_data_len;
7353 
7354 	if (ext) {
7355 		rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7356 		    ctsio->kern_data_ptr;
7357 		scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7358 		rtg_ext_ptr->format_type = 0x10;
7359 		rtg_ext_ptr->implicit_transition_time = 0;
7360 		tpg_desc = &rtg_ext_ptr->groups[0];
7361 	} else {
7362 		rtg_ptr = (struct scsi_target_group_data *)
7363 		    ctsio->kern_data_ptr;
7364 		scsi_ulto4b(total_len - 4, rtg_ptr->length);
7365 		tpg_desc = &rtg_ptr->groups[0];
7366 	}
7367 
7368 	mtx_lock(&softc->ctl_lock);
7369 	pg = softc->port_min / softc->port_cnt;
7370 	if (lun->flags & (CTL_LUN_PRIMARY_SC | CTL_LUN_PEER_SC_PRIMARY)) {
7371 		/* Some shelf is known to be primary. */
7372 		if (softc->ha_link == CTL_HA_LINK_OFFLINE)
7373 			os = TPG_ASYMMETRIC_ACCESS_UNAVAILABLE;
7374 		else if (softc->ha_link == CTL_HA_LINK_UNKNOWN)
7375 			os = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7376 		else if (softc->ha_mode == CTL_HA_MODE_ACT_STBY)
7377 			os = TPG_ASYMMETRIC_ACCESS_STANDBY;
7378 		else
7379 			os = TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7380 		if (lun->flags & CTL_LUN_PRIMARY_SC) {
7381 			ts = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7382 		} else {
7383 			ts = os;
7384 			os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7385 		}
7386 	} else {
7387 		/* No known primary shelf. */
7388 		if (softc->ha_link == CTL_HA_LINK_OFFLINE) {
7389 			ts = TPG_ASYMMETRIC_ACCESS_UNAVAILABLE;
7390 			os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7391 		} else if (softc->ha_link == CTL_HA_LINK_UNKNOWN) {
7392 			ts = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7393 			os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7394 		} else {
7395 			ts = os = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7396 		}
7397 	}
7398 	if (shared_group) {
7399 		tpg_desc->pref_state = ts;
7400 		tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP |
7401 		    TPG_U_SUP | TPG_T_SUP;
7402 		scsi_ulto2b(1, tpg_desc->target_port_group);
7403 		tpg_desc->status = TPG_IMPLICIT;
7404 		pc = 0;
7405 		STAILQ_FOREACH(port, &softc->port_list, links) {
7406 			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7407 				continue;
7408 			if (!softc->is_single &&
7409 			    (port->status & CTL_PORT_STATUS_HA_SHARED) == 0)
7410 				continue;
7411 			if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
7412 				continue;
7413 			scsi_ulto2b(port->targ_port, tpg_desc->descriptors[pc].
7414 			    relative_target_port_identifier);
7415 			pc++;
7416 		}
7417 		tpg_desc->target_port_count = pc;
7418 		tpg_desc = (struct scsi_target_port_group_descriptor *)
7419 		    &tpg_desc->descriptors[pc];
7420 	}
7421 	for (g = 0; g < num_ha_groups; g++) {
7422 		tpg_desc->pref_state = (g == pg) ? ts : os;
7423 		tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP |
7424 		    TPG_U_SUP | TPG_T_SUP;
7425 		scsi_ulto2b(2 + g, tpg_desc->target_port_group);
7426 		tpg_desc->status = TPG_IMPLICIT;
7427 		pc = 0;
7428 		STAILQ_FOREACH(port, &softc->port_list, links) {
7429 			if (port->targ_port < g * softc->port_cnt ||
7430 			    port->targ_port >= (g + 1) * softc->port_cnt)
7431 				continue;
7432 			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7433 				continue;
7434 			if (port->status & CTL_PORT_STATUS_HA_SHARED)
7435 				continue;
7436 			if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
7437 				continue;
7438 			scsi_ulto2b(port->targ_port, tpg_desc->descriptors[pc].
7439 			    relative_target_port_identifier);
7440 			pc++;
7441 		}
7442 		tpg_desc->target_port_count = pc;
7443 		tpg_desc = (struct scsi_target_port_group_descriptor *)
7444 		    &tpg_desc->descriptors[pc];
7445 	}
7446 	mtx_unlock(&softc->ctl_lock);
7447 
7448 	ctl_set_success(ctsio);
7449 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7450 	ctsio->be_move_done = ctl_config_move_done;
7451 	ctl_datamove((union ctl_io *)ctsio);
7452 	return(retval);
7453 }
7454 
7455 int
7456 ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7457 {
7458 	struct ctl_lun *lun = CTL_LUN(ctsio);
7459 	struct scsi_report_supported_opcodes *cdb;
7460 	const struct ctl_cmd_entry *entry, *sentry;
7461 	struct scsi_report_supported_opcodes_all *all;
7462 	struct scsi_report_supported_opcodes_descr *descr;
7463 	struct scsi_report_supported_opcodes_one *one;
7464 	int retval;
7465 	int alloc_len, total_len;
7466 	int opcode, service_action, i, j, num;
7467 
7468 	CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7469 
7470 	cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7471 	retval = CTL_RETVAL_COMPLETE;
7472 
7473 	opcode = cdb->requested_opcode;
7474 	service_action = scsi_2btoul(cdb->requested_service_action);
7475 	switch (cdb->options & RSO_OPTIONS_MASK) {
7476 	case RSO_OPTIONS_ALL:
7477 		num = 0;
7478 		for (i = 0; i < 256; i++) {
7479 			entry = &ctl_cmd_table[i];
7480 			if (entry->flags & CTL_CMD_FLAG_SA5) {
7481 				for (j = 0; j < 32; j++) {
7482 					sentry = &((const struct ctl_cmd_entry *)
7483 					    entry->execute)[j];
7484 					if (ctl_cmd_applicable(
7485 					    lun->be_lun->lun_type, sentry))
7486 						num++;
7487 				}
7488 			} else {
7489 				if (ctl_cmd_applicable(lun->be_lun->lun_type,
7490 				    entry))
7491 					num++;
7492 			}
7493 		}
7494 		total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7495 		    num * sizeof(struct scsi_report_supported_opcodes_descr);
7496 		break;
7497 	case RSO_OPTIONS_OC:
7498 		if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7499 			ctl_set_invalid_field(/*ctsio*/ ctsio,
7500 					      /*sks_valid*/ 1,
7501 					      /*command*/ 1,
7502 					      /*field*/ 2,
7503 					      /*bit_valid*/ 1,
7504 					      /*bit*/ 2);
7505 			ctl_done((union ctl_io *)ctsio);
7506 			return (CTL_RETVAL_COMPLETE);
7507 		}
7508 		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7509 		break;
7510 	case RSO_OPTIONS_OC_SA:
7511 		if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7512 		    service_action >= 32) {
7513 			ctl_set_invalid_field(/*ctsio*/ ctsio,
7514 					      /*sks_valid*/ 1,
7515 					      /*command*/ 1,
7516 					      /*field*/ 2,
7517 					      /*bit_valid*/ 1,
7518 					      /*bit*/ 2);
7519 			ctl_done((union ctl_io *)ctsio);
7520 			return (CTL_RETVAL_COMPLETE);
7521 		}
7522 		/* FALLTHROUGH */
7523 	case RSO_OPTIONS_OC_ASA:
7524 		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7525 		break;
7526 	default:
7527 		ctl_set_invalid_field(/*ctsio*/ ctsio,
7528 				      /*sks_valid*/ 1,
7529 				      /*command*/ 1,
7530 				      /*field*/ 2,
7531 				      /*bit_valid*/ 1,
7532 				      /*bit*/ 2);
7533 		ctl_done((union ctl_io *)ctsio);
7534 		return (CTL_RETVAL_COMPLETE);
7535 	}
7536 
7537 	alloc_len = scsi_4btoul(cdb->length);
7538 
7539 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7540 	ctsio->kern_sg_entries = 0;
7541 	ctsio->kern_rel_offset = 0;
7542 	ctsio->kern_data_len = min(total_len, alloc_len);
7543 	ctsio->kern_total_len = ctsio->kern_data_len;
7544 
7545 	switch (cdb->options & RSO_OPTIONS_MASK) {
7546 	case RSO_OPTIONS_ALL:
7547 		all = (struct scsi_report_supported_opcodes_all *)
7548 		    ctsio->kern_data_ptr;
7549 		num = 0;
7550 		for (i = 0; i < 256; i++) {
7551 			entry = &ctl_cmd_table[i];
7552 			if (entry->flags & CTL_CMD_FLAG_SA5) {
7553 				for (j = 0; j < 32; j++) {
7554 					sentry = &((const struct ctl_cmd_entry *)
7555 					    entry->execute)[j];
7556 					if (!ctl_cmd_applicable(
7557 					    lun->be_lun->lun_type, sentry))
7558 						continue;
7559 					descr = &all->descr[num++];
7560 					descr->opcode = i;
7561 					scsi_ulto2b(j, descr->service_action);
7562 					descr->flags = RSO_SERVACTV;
7563 					scsi_ulto2b(sentry->length,
7564 					    descr->cdb_length);
7565 				}
7566 			} else {
7567 				if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7568 				    entry))
7569 					continue;
7570 				descr = &all->descr[num++];
7571 				descr->opcode = i;
7572 				scsi_ulto2b(0, descr->service_action);
7573 				descr->flags = 0;
7574 				scsi_ulto2b(entry->length, descr->cdb_length);
7575 			}
7576 		}
7577 		scsi_ulto4b(
7578 		    num * sizeof(struct scsi_report_supported_opcodes_descr),
7579 		    all->length);
7580 		break;
7581 	case RSO_OPTIONS_OC:
7582 		one = (struct scsi_report_supported_opcodes_one *)
7583 		    ctsio->kern_data_ptr;
7584 		entry = &ctl_cmd_table[opcode];
7585 		goto fill_one;
7586 	case RSO_OPTIONS_OC_SA:
7587 		one = (struct scsi_report_supported_opcodes_one *)
7588 		    ctsio->kern_data_ptr;
7589 		entry = &ctl_cmd_table[opcode];
7590 		entry = &((const struct ctl_cmd_entry *)
7591 		    entry->execute)[service_action];
7592 fill_one:
7593 		if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7594 			one->support = 3;
7595 			scsi_ulto2b(entry->length, one->cdb_length);
7596 			one->cdb_usage[0] = opcode;
7597 			memcpy(&one->cdb_usage[1], entry->usage,
7598 			    entry->length - 1);
7599 		} else
7600 			one->support = 1;
7601 		break;
7602 	case RSO_OPTIONS_OC_ASA:
7603 		one = (struct scsi_report_supported_opcodes_one *)
7604 		    ctsio->kern_data_ptr;
7605 		entry = &ctl_cmd_table[opcode];
7606 		if (entry->flags & CTL_CMD_FLAG_SA5) {
7607 			entry = &((const struct ctl_cmd_entry *)
7608 			    entry->execute)[service_action];
7609 		} else if (service_action != 0) {
7610 			one->support = 1;
7611 			break;
7612 		}
7613 		goto fill_one;
7614 	}
7615 
7616 	ctl_set_success(ctsio);
7617 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7618 	ctsio->be_move_done = ctl_config_move_done;
7619 	ctl_datamove((union ctl_io *)ctsio);
7620 	return(retval);
7621 }
7622 
7623 int
7624 ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7625 {
7626 	struct scsi_report_supported_tmf *cdb;
7627 	struct scsi_report_supported_tmf_ext_data *data;
7628 	int retval;
7629 	int alloc_len, total_len;
7630 
7631 	CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7632 
7633 	cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7634 
7635 	retval = CTL_RETVAL_COMPLETE;
7636 
7637 	if (cdb->options & RST_REPD)
7638 		total_len = sizeof(struct scsi_report_supported_tmf_ext_data);
7639 	else
7640 		total_len = sizeof(struct scsi_report_supported_tmf_data);
7641 	alloc_len = scsi_4btoul(cdb->length);
7642 
7643 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7644 	ctsio->kern_sg_entries = 0;
7645 	ctsio->kern_rel_offset = 0;
7646 	ctsio->kern_data_len = min(total_len, alloc_len);
7647 	ctsio->kern_total_len = ctsio->kern_data_len;
7648 
7649 	data = (struct scsi_report_supported_tmf_ext_data *)ctsio->kern_data_ptr;
7650 	data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_QTS |
7651 	    RST_TRS;
7652 	data->byte2 |= RST_QAES | RST_QTSS | RST_ITNRS;
7653 	data->length = total_len - 4;
7654 
7655 	ctl_set_success(ctsio);
7656 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7657 	ctsio->be_move_done = ctl_config_move_done;
7658 	ctl_datamove((union ctl_io *)ctsio);
7659 	return (retval);
7660 }
7661 
7662 int
7663 ctl_report_timestamp(struct ctl_scsiio *ctsio)
7664 {
7665 	struct scsi_report_timestamp *cdb;
7666 	struct scsi_report_timestamp_data *data;
7667 	struct timeval tv;
7668 	int64_t timestamp;
7669 	int retval;
7670 	int alloc_len, total_len;
7671 
7672 	CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7673 
7674 	cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7675 
7676 	retval = CTL_RETVAL_COMPLETE;
7677 
7678 	total_len = sizeof(struct scsi_report_timestamp_data);
7679 	alloc_len = scsi_4btoul(cdb->length);
7680 
7681 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7682 	ctsio->kern_sg_entries = 0;
7683 	ctsio->kern_rel_offset = 0;
7684 	ctsio->kern_data_len = min(total_len, alloc_len);
7685 	ctsio->kern_total_len = ctsio->kern_data_len;
7686 
7687 	data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7688 	scsi_ulto2b(sizeof(*data) - 2, data->length);
7689 	data->origin = RTS_ORIG_OUTSIDE;
7690 	getmicrotime(&tv);
7691 	timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7692 	scsi_ulto4b(timestamp >> 16, data->timestamp);
7693 	scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7694 
7695 	ctl_set_success(ctsio);
7696 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7697 	ctsio->be_move_done = ctl_config_move_done;
7698 	ctl_datamove((union ctl_io *)ctsio);
7699 	return (retval);
7700 }
7701 
7702 int
7703 ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7704 {
7705 	struct ctl_softc *softc = CTL_SOFTC(ctsio);
7706 	struct ctl_lun *lun = CTL_LUN(ctsio);
7707 	struct scsi_per_res_in *cdb;
7708 	int alloc_len, total_len = 0;
7709 	/* struct scsi_per_res_in_rsrv in_data; */
7710 	uint64_t key;
7711 
7712 	CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7713 
7714 	cdb = (struct scsi_per_res_in *)ctsio->cdb;
7715 
7716 	alloc_len = scsi_2btoul(cdb->length);
7717 
7718 retry:
7719 	mtx_lock(&lun->lun_lock);
7720 	switch (cdb->action) {
7721 	case SPRI_RK: /* read keys */
7722 		total_len = sizeof(struct scsi_per_res_in_keys) +
7723 			lun->pr_key_count *
7724 			sizeof(struct scsi_per_res_key);
7725 		break;
7726 	case SPRI_RR: /* read reservation */
7727 		if (lun->flags & CTL_LUN_PR_RESERVED)
7728 			total_len = sizeof(struct scsi_per_res_in_rsrv);
7729 		else
7730 			total_len = sizeof(struct scsi_per_res_in_header);
7731 		break;
7732 	case SPRI_RC: /* report capabilities */
7733 		total_len = sizeof(struct scsi_per_res_cap);
7734 		break;
7735 	case SPRI_RS: /* read full status */
7736 		total_len = sizeof(struct scsi_per_res_in_header) +
7737 		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7738 		    lun->pr_key_count;
7739 		break;
7740 	default:
7741 		panic("%s: Invalid PR type %#x", __func__, cdb->action);
7742 	}
7743 	mtx_unlock(&lun->lun_lock);
7744 
7745 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7746 	ctsio->kern_rel_offset = 0;
7747 	ctsio->kern_sg_entries = 0;
7748 	ctsio->kern_data_len = min(total_len, alloc_len);
7749 	ctsio->kern_total_len = ctsio->kern_data_len;
7750 
7751 	mtx_lock(&lun->lun_lock);
7752 	switch (cdb->action) {
7753 	case SPRI_RK: { // read keys
7754         struct scsi_per_res_in_keys *res_keys;
7755 		int i, key_count;
7756 
7757 		res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7758 
7759 		/*
7760 		 * We had to drop the lock to allocate our buffer, which
7761 		 * leaves time for someone to come in with another
7762 		 * persistent reservation.  (That is unlikely, though,
7763 		 * since this should be the only persistent reservation
7764 		 * command active right now.)
7765 		 */
7766 		if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7767 		    (lun->pr_key_count *
7768 		     sizeof(struct scsi_per_res_key)))){
7769 			mtx_unlock(&lun->lun_lock);
7770 			free(ctsio->kern_data_ptr, M_CTL);
7771 			printf("%s: reservation length changed, retrying\n",
7772 			       __func__);
7773 			goto retry;
7774 		}
7775 
7776 		scsi_ulto4b(lun->pr_generation, res_keys->header.generation);
7777 
7778 		scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7779 			     lun->pr_key_count, res_keys->header.length);
7780 
7781 		for (i = 0, key_count = 0; i < CTL_MAX_INITIATORS; i++) {
7782 			if ((key = ctl_get_prkey(lun, i)) == 0)
7783 				continue;
7784 
7785 			/*
7786 			 * We used lun->pr_key_count to calculate the
7787 			 * size to allocate.  If it turns out the number of
7788 			 * initiators with the registered flag set is
7789 			 * larger than that (i.e. they haven't been kept in
7790 			 * sync), we've got a problem.
7791 			 */
7792 			if (key_count >= lun->pr_key_count) {
7793 				key_count++;
7794 				continue;
7795 			}
7796 			scsi_u64to8b(key, res_keys->keys[key_count].key);
7797 			key_count++;
7798 		}
7799 		break;
7800 	}
7801 	case SPRI_RR: { // read reservation
7802 		struct scsi_per_res_in_rsrv *res;
7803 		int tmp_len, header_only;
7804 
7805 		res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7806 
7807 		scsi_ulto4b(lun->pr_generation, res->header.generation);
7808 
7809 		if (lun->flags & CTL_LUN_PR_RESERVED)
7810 		{
7811 			tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7812 			scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7813 				    res->header.length);
7814 			header_only = 0;
7815 		} else {
7816 			tmp_len = sizeof(struct scsi_per_res_in_header);
7817 			scsi_ulto4b(0, res->header.length);
7818 			header_only = 1;
7819 		}
7820 
7821 		/*
7822 		 * We had to drop the lock to allocate our buffer, which
7823 		 * leaves time for someone to come in with another
7824 		 * persistent reservation.  (That is unlikely, though,
7825 		 * since this should be the only persistent reservation
7826 		 * command active right now.)
7827 		 */
7828 		if (tmp_len != total_len) {
7829 			mtx_unlock(&lun->lun_lock);
7830 			free(ctsio->kern_data_ptr, M_CTL);
7831 			printf("%s: reservation status changed, retrying\n",
7832 			       __func__);
7833 			goto retry;
7834 		}
7835 
7836 		/*
7837 		 * No reservation held, so we're done.
7838 		 */
7839 		if (header_only != 0)
7840 			break;
7841 
7842 		/*
7843 		 * If the registration is an All Registrants type, the key
7844 		 * is 0, since it doesn't really matter.
7845 		 */
7846 		if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7847 			scsi_u64to8b(ctl_get_prkey(lun, lun->pr_res_idx),
7848 			    res->data.reservation);
7849 		}
7850 		res->data.scopetype = lun->pr_res_type;
7851 		break;
7852 	}
7853 	case SPRI_RC:     //report capabilities
7854 	{
7855 		struct scsi_per_res_cap *res_cap;
7856 		uint16_t type_mask;
7857 
7858 		res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7859 		scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7860 		res_cap->flags1 = SPRI_CRH;
7861 		res_cap->flags2 = SPRI_TMV | SPRI_ALLOW_5;
7862 		type_mask = SPRI_TM_WR_EX_AR |
7863 			    SPRI_TM_EX_AC_RO |
7864 			    SPRI_TM_WR_EX_RO |
7865 			    SPRI_TM_EX_AC |
7866 			    SPRI_TM_WR_EX |
7867 			    SPRI_TM_EX_AC_AR;
7868 		scsi_ulto2b(type_mask, res_cap->type_mask);
7869 		break;
7870 	}
7871 	case SPRI_RS: { // read full status
7872 		struct scsi_per_res_in_full *res_status;
7873 		struct scsi_per_res_in_full_desc *res_desc;
7874 		struct ctl_port *port;
7875 		int i, len;
7876 
7877 		res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7878 
7879 		/*
7880 		 * We had to drop the lock to allocate our buffer, which
7881 		 * leaves time for someone to come in with another
7882 		 * persistent reservation.  (That is unlikely, though,
7883 		 * since this should be the only persistent reservation
7884 		 * command active right now.)
7885 		 */
7886 		if (total_len < (sizeof(struct scsi_per_res_in_header) +
7887 		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7888 		     lun->pr_key_count)){
7889 			mtx_unlock(&lun->lun_lock);
7890 			free(ctsio->kern_data_ptr, M_CTL);
7891 			printf("%s: reservation length changed, retrying\n",
7892 			       __func__);
7893 			goto retry;
7894 		}
7895 
7896 		scsi_ulto4b(lun->pr_generation, res_status->header.generation);
7897 
7898 		res_desc = &res_status->desc[0];
7899 		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7900 			if ((key = ctl_get_prkey(lun, i)) == 0)
7901 				continue;
7902 
7903 			scsi_u64to8b(key, res_desc->res_key.key);
7904 			if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7905 			    (lun->pr_res_idx == i ||
7906 			     lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7907 				res_desc->flags = SPRI_FULL_R_HOLDER;
7908 				res_desc->scopetype = lun->pr_res_type;
7909 			}
7910 			scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7911 			    res_desc->rel_trgt_port_id);
7912 			len = 0;
7913 			port = softc->ctl_ports[i / CTL_MAX_INIT_PER_PORT];
7914 			if (port != NULL)
7915 				len = ctl_create_iid(port,
7916 				    i % CTL_MAX_INIT_PER_PORT,
7917 				    res_desc->transport_id);
7918 			scsi_ulto4b(len, res_desc->additional_length);
7919 			res_desc = (struct scsi_per_res_in_full_desc *)
7920 			    &res_desc->transport_id[len];
7921 		}
7922 		scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7923 		    res_status->header.length);
7924 		break;
7925 	}
7926 	default:
7927 		panic("%s: Invalid PR type %#x", __func__, cdb->action);
7928 	}
7929 	mtx_unlock(&lun->lun_lock);
7930 
7931 	ctl_set_success(ctsio);
7932 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7933 	ctsio->be_move_done = ctl_config_move_done;
7934 	ctl_datamove((union ctl_io *)ctsio);
7935 	return (CTL_RETVAL_COMPLETE);
7936 }
7937 
7938 /*
7939  * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7940  * it should return.
7941  */
7942 static int
7943 ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7944 		uint64_t sa_res_key, uint8_t type, uint32_t residx,
7945 		struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7946 		struct scsi_per_res_out_parms* param)
7947 {
7948 	union ctl_ha_msg persis_io;
7949 	int i;
7950 
7951 	mtx_lock(&lun->lun_lock);
7952 	if (sa_res_key == 0) {
7953 		if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7954 			/* validate scope and type */
7955 			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7956 			     SPR_LU_SCOPE) {
7957 				mtx_unlock(&lun->lun_lock);
7958 				ctl_set_invalid_field(/*ctsio*/ ctsio,
7959 						      /*sks_valid*/ 1,
7960 						      /*command*/ 1,
7961 						      /*field*/ 2,
7962 						      /*bit_valid*/ 1,
7963 						      /*bit*/ 4);
7964 				ctl_done((union ctl_io *)ctsio);
7965 				return (1);
7966 			}
7967 
7968 		        if (type>8 || type==2 || type==4 || type==0) {
7969 				mtx_unlock(&lun->lun_lock);
7970 				ctl_set_invalid_field(/*ctsio*/ ctsio,
7971        	           				      /*sks_valid*/ 1,
7972 						      /*command*/ 1,
7973 						      /*field*/ 2,
7974 						      /*bit_valid*/ 1,
7975 						      /*bit*/ 0);
7976 				ctl_done((union ctl_io *)ctsio);
7977 				return (1);
7978 		        }
7979 
7980 			/*
7981 			 * Unregister everybody else and build UA for
7982 			 * them
7983 			 */
7984 			for(i = 0; i < CTL_MAX_INITIATORS; i++) {
7985 				if (i == residx || ctl_get_prkey(lun, i) == 0)
7986 					continue;
7987 
7988 				ctl_clr_prkey(lun, i);
7989 				ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7990 			}
7991 			lun->pr_key_count = 1;
7992 			lun->pr_res_type = type;
7993 			if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
7994 			    lun->pr_res_type != SPR_TYPE_EX_AC_AR)
7995 				lun->pr_res_idx = residx;
7996 			lun->pr_generation++;
7997 			mtx_unlock(&lun->lun_lock);
7998 
7999 			/* send msg to other side */
8000 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8001 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8002 			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8003 			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8004 			persis_io.pr.pr_info.res_type = type;
8005 			memcpy(persis_io.pr.pr_info.sa_res_key,
8006 			       param->serv_act_res_key,
8007 			       sizeof(param->serv_act_res_key));
8008 			ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8009 			    sizeof(persis_io.pr), M_WAITOK);
8010 		} else {
8011 			/* not all registrants */
8012 			mtx_unlock(&lun->lun_lock);
8013 			free(ctsio->kern_data_ptr, M_CTL);
8014 			ctl_set_invalid_field(ctsio,
8015 					      /*sks_valid*/ 1,
8016 					      /*command*/ 0,
8017 					      /*field*/ 8,
8018 					      /*bit_valid*/ 0,
8019 					      /*bit*/ 0);
8020 			ctl_done((union ctl_io *)ctsio);
8021 			return (1);
8022 		}
8023 	} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8024 		|| !(lun->flags & CTL_LUN_PR_RESERVED)) {
8025 		int found = 0;
8026 
8027 		if (res_key == sa_res_key) {
8028 			/* special case */
8029 			/*
8030 			 * The spec implies this is not good but doesn't
8031 			 * say what to do. There are two choices either
8032 			 * generate a res conflict or check condition
8033 			 * with illegal field in parameter data. Since
8034 			 * that is what is done when the sa_res_key is
8035 			 * zero I'll take that approach since this has
8036 			 * to do with the sa_res_key.
8037 			 */
8038 			mtx_unlock(&lun->lun_lock);
8039 			free(ctsio->kern_data_ptr, M_CTL);
8040 			ctl_set_invalid_field(ctsio,
8041 					      /*sks_valid*/ 1,
8042 					      /*command*/ 0,
8043 					      /*field*/ 8,
8044 					      /*bit_valid*/ 0,
8045 					      /*bit*/ 0);
8046 			ctl_done((union ctl_io *)ctsio);
8047 			return (1);
8048 		}
8049 
8050 		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8051 			if (ctl_get_prkey(lun, i) != sa_res_key)
8052 				continue;
8053 
8054 			found = 1;
8055 			ctl_clr_prkey(lun, i);
8056 			lun->pr_key_count--;
8057 			ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8058 		}
8059 		if (!found) {
8060 			mtx_unlock(&lun->lun_lock);
8061 			free(ctsio->kern_data_ptr, M_CTL);
8062 			ctl_set_reservation_conflict(ctsio);
8063 			ctl_done((union ctl_io *)ctsio);
8064 			return (CTL_RETVAL_COMPLETE);
8065 		}
8066 		lun->pr_generation++;
8067 		mtx_unlock(&lun->lun_lock);
8068 
8069 		/* send msg to other side */
8070 		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8071 		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8072 		persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8073 		persis_io.pr.pr_info.residx = lun->pr_res_idx;
8074 		persis_io.pr.pr_info.res_type = type;
8075 		memcpy(persis_io.pr.pr_info.sa_res_key,
8076 		       param->serv_act_res_key,
8077 		       sizeof(param->serv_act_res_key));
8078 		ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8079 		    sizeof(persis_io.pr), M_WAITOK);
8080 	} else {
8081 		/* Reserved but not all registrants */
8082 		/* sa_res_key is res holder */
8083 		if (sa_res_key == ctl_get_prkey(lun, lun->pr_res_idx)) {
8084 			/* validate scope and type */
8085 			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
8086 			     SPR_LU_SCOPE) {
8087 				mtx_unlock(&lun->lun_lock);
8088 				ctl_set_invalid_field(/*ctsio*/ ctsio,
8089 						      /*sks_valid*/ 1,
8090 						      /*command*/ 1,
8091 						      /*field*/ 2,
8092 						      /*bit_valid*/ 1,
8093 						      /*bit*/ 4);
8094 				ctl_done((union ctl_io *)ctsio);
8095 				return (1);
8096 			}
8097 
8098 			if (type>8 || type==2 || type==4 || type==0) {
8099 				mtx_unlock(&lun->lun_lock);
8100 				ctl_set_invalid_field(/*ctsio*/ ctsio,
8101 						      /*sks_valid*/ 1,
8102 						      /*command*/ 1,
8103 						      /*field*/ 2,
8104 						      /*bit_valid*/ 1,
8105 						      /*bit*/ 0);
8106 				ctl_done((union ctl_io *)ctsio);
8107 				return (1);
8108 			}
8109 
8110 			/*
8111 			 * Do the following:
8112 			 * if sa_res_key != res_key remove all
8113 			 * registrants w/sa_res_key and generate UA
8114 			 * for these registrants(Registrations
8115 			 * Preempted) if it wasn't an exclusive
8116 			 * reservation generate UA(Reservations
8117 			 * Preempted) for all other registered nexuses
8118 			 * if the type has changed. Establish the new
8119 			 * reservation and holder. If res_key and
8120 			 * sa_res_key are the same do the above
8121 			 * except don't unregister the res holder.
8122 			 */
8123 
8124 			for(i = 0; i < CTL_MAX_INITIATORS; i++) {
8125 				if (i == residx || ctl_get_prkey(lun, i) == 0)
8126 					continue;
8127 
8128 				if (sa_res_key == ctl_get_prkey(lun, i)) {
8129 					ctl_clr_prkey(lun, i);
8130 					lun->pr_key_count--;
8131 					ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8132 				} else if (type != lun->pr_res_type &&
8133 				    (lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8134 				     lun->pr_res_type == SPR_TYPE_EX_AC_RO)) {
8135 					ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8136 				}
8137 			}
8138 			lun->pr_res_type = type;
8139 			if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8140 			    lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8141 				lun->pr_res_idx = residx;
8142 			else
8143 				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8144 			lun->pr_generation++;
8145 			mtx_unlock(&lun->lun_lock);
8146 
8147 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8148 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8149 			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8150 			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8151 			persis_io.pr.pr_info.res_type = type;
8152 			memcpy(persis_io.pr.pr_info.sa_res_key,
8153 			       param->serv_act_res_key,
8154 			       sizeof(param->serv_act_res_key));
8155 			ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8156 			    sizeof(persis_io.pr), M_WAITOK);
8157 		} else {
8158 			/*
8159 			 * sa_res_key is not the res holder just
8160 			 * remove registrants
8161 			 */
8162 			int found=0;
8163 
8164 			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8165 				if (sa_res_key != ctl_get_prkey(lun, i))
8166 					continue;
8167 
8168 				found = 1;
8169 				ctl_clr_prkey(lun, i);
8170 				lun->pr_key_count--;
8171 				ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8172 			}
8173 
8174 			if (!found) {
8175 				mtx_unlock(&lun->lun_lock);
8176 				free(ctsio->kern_data_ptr, M_CTL);
8177 				ctl_set_reservation_conflict(ctsio);
8178 				ctl_done((union ctl_io *)ctsio);
8179 		        	return (1);
8180 			}
8181 			lun->pr_generation++;
8182 			mtx_unlock(&lun->lun_lock);
8183 
8184 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8185 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8186 			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8187 			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8188 			persis_io.pr.pr_info.res_type = type;
8189 			memcpy(persis_io.pr.pr_info.sa_res_key,
8190 			       param->serv_act_res_key,
8191 			       sizeof(param->serv_act_res_key));
8192 			ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8193 			    sizeof(persis_io.pr), M_WAITOK);
8194 		}
8195 	}
8196 	return (0);
8197 }
8198 
8199 static void
8200 ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
8201 {
8202 	uint64_t sa_res_key;
8203 	int i;
8204 
8205 	sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
8206 
8207 	if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8208 	 || lun->pr_res_idx == CTL_PR_NO_RESERVATION
8209 	 || sa_res_key != ctl_get_prkey(lun, lun->pr_res_idx)) {
8210 		if (sa_res_key == 0) {
8211 			/*
8212 			 * Unregister everybody else and build UA for
8213 			 * them
8214 			 */
8215 			for(i = 0; i < CTL_MAX_INITIATORS; i++) {
8216 				if (i == msg->pr.pr_info.residx ||
8217 				    ctl_get_prkey(lun, i) == 0)
8218 					continue;
8219 
8220 				ctl_clr_prkey(lun, i);
8221 				ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8222 			}
8223 
8224 			lun->pr_key_count = 1;
8225 			lun->pr_res_type = msg->pr.pr_info.res_type;
8226 			if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8227 			    lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8228 				lun->pr_res_idx = msg->pr.pr_info.residx;
8229 		} else {
8230 		        for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8231 				if (sa_res_key == ctl_get_prkey(lun, i))
8232 					continue;
8233 
8234 				ctl_clr_prkey(lun, i);
8235 				lun->pr_key_count--;
8236 				ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8237 			}
8238 		}
8239 	} else {
8240 		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8241 			if (i == msg->pr.pr_info.residx ||
8242 			    ctl_get_prkey(lun, i) == 0)
8243 				continue;
8244 
8245 			if (sa_res_key == ctl_get_prkey(lun, i)) {
8246 				ctl_clr_prkey(lun, i);
8247 				lun->pr_key_count--;
8248 				ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8249 			} else if (msg->pr.pr_info.res_type != lun->pr_res_type
8250 			    && (lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8251 			     lun->pr_res_type == SPR_TYPE_EX_AC_RO)) {
8252 				ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8253 			}
8254 		}
8255 		lun->pr_res_type = msg->pr.pr_info.res_type;
8256 		if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8257 		    lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8258 			lun->pr_res_idx = msg->pr.pr_info.residx;
8259 		else
8260 			lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8261 	}
8262 	lun->pr_generation++;
8263 
8264 }
8265 
8266 int
8267 ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8268 {
8269 	struct ctl_softc *softc = CTL_SOFTC(ctsio);
8270 	struct ctl_lun *lun = CTL_LUN(ctsio);
8271 	int retval;
8272 	uint32_t param_len;
8273 	struct scsi_per_res_out *cdb;
8274 	struct scsi_per_res_out_parms* param;
8275 	uint32_t residx;
8276 	uint64_t res_key, sa_res_key, key;
8277 	uint8_t type;
8278 	union ctl_ha_msg persis_io;
8279 	int    i;
8280 
8281 	CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8282 
8283 	cdb = (struct scsi_per_res_out *)ctsio->cdb;
8284 	retval = CTL_RETVAL_COMPLETE;
8285 
8286 	/*
8287 	 * We only support whole-LUN scope.  The scope & type are ignored for
8288 	 * register, register and ignore existing key and clear.
8289 	 * We sometimes ignore scope and type on preempts too!!
8290 	 * Verify reservation type here as well.
8291 	 */
8292 	type = cdb->scope_type & SPR_TYPE_MASK;
8293 	if ((cdb->action == SPRO_RESERVE)
8294 	 || (cdb->action == SPRO_RELEASE)) {
8295 		if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8296 			ctl_set_invalid_field(/*ctsio*/ ctsio,
8297 					      /*sks_valid*/ 1,
8298 					      /*command*/ 1,
8299 					      /*field*/ 2,
8300 					      /*bit_valid*/ 1,
8301 					      /*bit*/ 4);
8302 			ctl_done((union ctl_io *)ctsio);
8303 			return (CTL_RETVAL_COMPLETE);
8304 		}
8305 
8306 		if (type>8 || type==2 || type==4 || type==0) {
8307 			ctl_set_invalid_field(/*ctsio*/ ctsio,
8308 					      /*sks_valid*/ 1,
8309 					      /*command*/ 1,
8310 					      /*field*/ 2,
8311 					      /*bit_valid*/ 1,
8312 					      /*bit*/ 0);
8313 			ctl_done((union ctl_io *)ctsio);
8314 			return (CTL_RETVAL_COMPLETE);
8315 		}
8316 	}
8317 
8318 	param_len = scsi_4btoul(cdb->length);
8319 
8320 	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8321 		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8322 		ctsio->kern_data_len = param_len;
8323 		ctsio->kern_total_len = param_len;
8324 		ctsio->kern_rel_offset = 0;
8325 		ctsio->kern_sg_entries = 0;
8326 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8327 		ctsio->be_move_done = ctl_config_move_done;
8328 		ctl_datamove((union ctl_io *)ctsio);
8329 
8330 		return (CTL_RETVAL_COMPLETE);
8331 	}
8332 
8333 	param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8334 
8335 	residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
8336 	res_key = scsi_8btou64(param->res_key.key);
8337 	sa_res_key = scsi_8btou64(param->serv_act_res_key);
8338 
8339 	/*
8340 	 * Validate the reservation key here except for SPRO_REG_IGNO
8341 	 * This must be done for all other service actions
8342 	 */
8343 	if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8344 		mtx_lock(&lun->lun_lock);
8345 		if ((key = ctl_get_prkey(lun, residx)) != 0) {
8346 			if (res_key != key) {
8347 				/*
8348 				 * The current key passed in doesn't match
8349 				 * the one the initiator previously
8350 				 * registered.
8351 				 */
8352 				mtx_unlock(&lun->lun_lock);
8353 				free(ctsio->kern_data_ptr, M_CTL);
8354 				ctl_set_reservation_conflict(ctsio);
8355 				ctl_done((union ctl_io *)ctsio);
8356 				return (CTL_RETVAL_COMPLETE);
8357 			}
8358 		} else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8359 			/*
8360 			 * We are not registered
8361 			 */
8362 			mtx_unlock(&lun->lun_lock);
8363 			free(ctsio->kern_data_ptr, M_CTL);
8364 			ctl_set_reservation_conflict(ctsio);
8365 			ctl_done((union ctl_io *)ctsio);
8366 			return (CTL_RETVAL_COMPLETE);
8367 		} else if (res_key != 0) {
8368 			/*
8369 			 * We are not registered and trying to register but
8370 			 * the register key isn't zero.
8371 			 */
8372 			mtx_unlock(&lun->lun_lock);
8373 			free(ctsio->kern_data_ptr, M_CTL);
8374 			ctl_set_reservation_conflict(ctsio);
8375 			ctl_done((union ctl_io *)ctsio);
8376 			return (CTL_RETVAL_COMPLETE);
8377 		}
8378 		mtx_unlock(&lun->lun_lock);
8379 	}
8380 
8381 	switch (cdb->action & SPRO_ACTION_MASK) {
8382 	case SPRO_REGISTER:
8383 	case SPRO_REG_IGNO: {
8384 		/*
8385 		 * We don't support any of these options, as we report in
8386 		 * the read capabilities request (see
8387 		 * ctl_persistent_reserve_in(), above).
8388 		 */
8389 		if ((param->flags & SPR_SPEC_I_PT)
8390 		 || (param->flags & SPR_ALL_TG_PT)
8391 		 || (param->flags & SPR_APTPL)) {
8392 			int bit_ptr;
8393 
8394 			if (param->flags & SPR_APTPL)
8395 				bit_ptr = 0;
8396 			else if (param->flags & SPR_ALL_TG_PT)
8397 				bit_ptr = 2;
8398 			else /* SPR_SPEC_I_PT */
8399 				bit_ptr = 3;
8400 
8401 			free(ctsio->kern_data_ptr, M_CTL);
8402 			ctl_set_invalid_field(ctsio,
8403 					      /*sks_valid*/ 1,
8404 					      /*command*/ 0,
8405 					      /*field*/ 20,
8406 					      /*bit_valid*/ 1,
8407 					      /*bit*/ bit_ptr);
8408 			ctl_done((union ctl_io *)ctsio);
8409 			return (CTL_RETVAL_COMPLETE);
8410 		}
8411 
8412 		mtx_lock(&lun->lun_lock);
8413 
8414 		/*
8415 		 * The initiator wants to clear the
8416 		 * key/unregister.
8417 		 */
8418 		if (sa_res_key == 0) {
8419 			if ((res_key == 0
8420 			  && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8421 			 || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8422 			  && ctl_get_prkey(lun, residx) == 0)) {
8423 				mtx_unlock(&lun->lun_lock);
8424 				goto done;
8425 			}
8426 
8427 			ctl_clr_prkey(lun, residx);
8428 			lun->pr_key_count--;
8429 
8430 			if (residx == lun->pr_res_idx) {
8431 				lun->flags &= ~CTL_LUN_PR_RESERVED;
8432 				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8433 
8434 				if ((lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8435 				     lun->pr_res_type == SPR_TYPE_EX_AC_RO) &&
8436 				    lun->pr_key_count) {
8437 					/*
8438 					 * If the reservation is a registrants
8439 					 * only type we need to generate a UA
8440 					 * for other registered inits.  The
8441 					 * sense code should be RESERVATIONS
8442 					 * RELEASED
8443 					 */
8444 
8445 					for (i = softc->init_min; i < softc->init_max; i++){
8446 						if (ctl_get_prkey(lun, i) == 0)
8447 							continue;
8448 						ctl_est_ua(lun, i,
8449 						    CTL_UA_RES_RELEASE);
8450 					}
8451 				}
8452 				lun->pr_res_type = 0;
8453 			} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8454 				if (lun->pr_key_count==0) {
8455 					lun->flags &= ~CTL_LUN_PR_RESERVED;
8456 					lun->pr_res_type = 0;
8457 					lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8458 				}
8459 			}
8460 			lun->pr_generation++;
8461 			mtx_unlock(&lun->lun_lock);
8462 
8463 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8464 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8465 			persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8466 			persis_io.pr.pr_info.residx = residx;
8467 			ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8468 			    sizeof(persis_io.pr), M_WAITOK);
8469 		} else /* sa_res_key != 0 */ {
8470 			/*
8471 			 * If we aren't registered currently then increment
8472 			 * the key count and set the registered flag.
8473 			 */
8474 			ctl_alloc_prkey(lun, residx);
8475 			if (ctl_get_prkey(lun, residx) == 0)
8476 				lun->pr_key_count++;
8477 			ctl_set_prkey(lun, residx, sa_res_key);
8478 			lun->pr_generation++;
8479 			mtx_unlock(&lun->lun_lock);
8480 
8481 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8482 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8483 			persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8484 			persis_io.pr.pr_info.residx = residx;
8485 			memcpy(persis_io.pr.pr_info.sa_res_key,
8486 			       param->serv_act_res_key,
8487 			       sizeof(param->serv_act_res_key));
8488 			ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8489 			    sizeof(persis_io.pr), M_WAITOK);
8490 		}
8491 
8492 		break;
8493 	}
8494 	case SPRO_RESERVE:
8495 		mtx_lock(&lun->lun_lock);
8496 		if (lun->flags & CTL_LUN_PR_RESERVED) {
8497 			/*
8498 			 * if this isn't the reservation holder and it's
8499 			 * not a "all registrants" type or if the type is
8500 			 * different then we have a conflict
8501 			 */
8502 			if ((lun->pr_res_idx != residx
8503 			  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8504 			 || lun->pr_res_type != type) {
8505 				mtx_unlock(&lun->lun_lock);
8506 				free(ctsio->kern_data_ptr, M_CTL);
8507 				ctl_set_reservation_conflict(ctsio);
8508 				ctl_done((union ctl_io *)ctsio);
8509 				return (CTL_RETVAL_COMPLETE);
8510 			}
8511 			mtx_unlock(&lun->lun_lock);
8512 		} else /* create a reservation */ {
8513 			/*
8514 			 * If it's not an "all registrants" type record
8515 			 * reservation holder
8516 			 */
8517 			if (type != SPR_TYPE_WR_EX_AR
8518 			 && type != SPR_TYPE_EX_AC_AR)
8519 				lun->pr_res_idx = residx; /* Res holder */
8520 			else
8521 				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8522 
8523 			lun->flags |= CTL_LUN_PR_RESERVED;
8524 			lun->pr_res_type = type;
8525 
8526 			mtx_unlock(&lun->lun_lock);
8527 
8528 			/* send msg to other side */
8529 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8530 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8531 			persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8532 			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8533 			persis_io.pr.pr_info.res_type = type;
8534 			ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8535 			    sizeof(persis_io.pr), M_WAITOK);
8536 		}
8537 		break;
8538 
8539 	case SPRO_RELEASE:
8540 		mtx_lock(&lun->lun_lock);
8541 		if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8542 			/* No reservation exists return good status */
8543 			mtx_unlock(&lun->lun_lock);
8544 			goto done;
8545 		}
8546 		/*
8547 		 * Is this nexus a reservation holder?
8548 		 */
8549 		if (lun->pr_res_idx != residx
8550 		 && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8551 			/*
8552 			 * not a res holder return good status but
8553 			 * do nothing
8554 			 */
8555 			mtx_unlock(&lun->lun_lock);
8556 			goto done;
8557 		}
8558 
8559 		if (lun->pr_res_type != type) {
8560 			mtx_unlock(&lun->lun_lock);
8561 			free(ctsio->kern_data_ptr, M_CTL);
8562 			ctl_set_illegal_pr_release(ctsio);
8563 			ctl_done((union ctl_io *)ctsio);
8564 			return (CTL_RETVAL_COMPLETE);
8565 		}
8566 
8567 		/* okay to release */
8568 		lun->flags &= ~CTL_LUN_PR_RESERVED;
8569 		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8570 		lun->pr_res_type = 0;
8571 
8572 		/*
8573 		 * If this isn't an exclusive access reservation and NUAR
8574 		 * is not set, generate UA for all other registrants.
8575 		 */
8576 		if (type != SPR_TYPE_EX_AC && type != SPR_TYPE_WR_EX &&
8577 		    (lun->MODE_CTRL.queue_flags & SCP_NUAR) == 0) {
8578 			for (i = softc->init_min; i < softc->init_max; i++) {
8579 				if (i == residx || ctl_get_prkey(lun, i) == 0)
8580 					continue;
8581 				ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8582 			}
8583 		}
8584 		mtx_unlock(&lun->lun_lock);
8585 
8586 		/* Send msg to other side */
8587 		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8588 		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8589 		persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8590 		ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8591 		     sizeof(persis_io.pr), M_WAITOK);
8592 		break;
8593 
8594 	case SPRO_CLEAR:
8595 		/* send msg to other side */
8596 
8597 		mtx_lock(&lun->lun_lock);
8598 		lun->flags &= ~CTL_LUN_PR_RESERVED;
8599 		lun->pr_res_type = 0;
8600 		lun->pr_key_count = 0;
8601 		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8602 
8603 		ctl_clr_prkey(lun, residx);
8604 		for (i = 0; i < CTL_MAX_INITIATORS; i++)
8605 			if (ctl_get_prkey(lun, i) != 0) {
8606 				ctl_clr_prkey(lun, i);
8607 				ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8608 			}
8609 		lun->pr_generation++;
8610 		mtx_unlock(&lun->lun_lock);
8611 
8612 		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8613 		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8614 		persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8615 		ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8616 		     sizeof(persis_io.pr), M_WAITOK);
8617 		break;
8618 
8619 	case SPRO_PREEMPT:
8620 	case SPRO_PRE_ABO: {
8621 		int nretval;
8622 
8623 		nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8624 					  residx, ctsio, cdb, param);
8625 		if (nretval != 0)
8626 			return (CTL_RETVAL_COMPLETE);
8627 		break;
8628 	}
8629 	default:
8630 		panic("%s: Invalid PR type %#x", __func__, cdb->action);
8631 	}
8632 
8633 done:
8634 	free(ctsio->kern_data_ptr, M_CTL);
8635 	ctl_set_success(ctsio);
8636 	ctl_done((union ctl_io *)ctsio);
8637 
8638 	return (retval);
8639 }
8640 
8641 /*
8642  * This routine is for handling a message from the other SC pertaining to
8643  * persistent reserve out. All the error checking will have been done
8644  * so only performing the action need be done here to keep the two
8645  * in sync.
8646  */
8647 static void
8648 ctl_hndl_per_res_out_on_other_sc(union ctl_io *io)
8649 {
8650 	struct ctl_softc *softc = CTL_SOFTC(io);
8651 	union ctl_ha_msg *msg = (union ctl_ha_msg *)&io->presio.pr_msg;
8652 	struct ctl_lun *lun;
8653 	int i;
8654 	uint32_t residx, targ_lun;
8655 
8656 	targ_lun = msg->hdr.nexus.targ_mapped_lun;
8657 	mtx_lock(&softc->ctl_lock);
8658 	if (targ_lun >= ctl_max_luns ||
8659 	    (lun = softc->ctl_luns[targ_lun]) == NULL) {
8660 		mtx_unlock(&softc->ctl_lock);
8661 		return;
8662 	}
8663 	mtx_lock(&lun->lun_lock);
8664 	mtx_unlock(&softc->ctl_lock);
8665 	if (lun->flags & CTL_LUN_DISABLED) {
8666 		mtx_unlock(&lun->lun_lock);
8667 		return;
8668 	}
8669 	residx = ctl_get_initindex(&msg->hdr.nexus);
8670 	switch(msg->pr.pr_info.action) {
8671 	case CTL_PR_REG_KEY:
8672 		ctl_alloc_prkey(lun, msg->pr.pr_info.residx);
8673 		if (ctl_get_prkey(lun, msg->pr.pr_info.residx) == 0)
8674 			lun->pr_key_count++;
8675 		ctl_set_prkey(lun, msg->pr.pr_info.residx,
8676 		    scsi_8btou64(msg->pr.pr_info.sa_res_key));
8677 		lun->pr_generation++;
8678 		break;
8679 
8680 	case CTL_PR_UNREG_KEY:
8681 		ctl_clr_prkey(lun, msg->pr.pr_info.residx);
8682 		lun->pr_key_count--;
8683 
8684 		/* XXX Need to see if the reservation has been released */
8685 		/* if so do we need to generate UA? */
8686 		if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8687 			lun->flags &= ~CTL_LUN_PR_RESERVED;
8688 			lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8689 
8690 			if ((lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8691 			     lun->pr_res_type == SPR_TYPE_EX_AC_RO) &&
8692 			    lun->pr_key_count) {
8693 				/*
8694 				 * If the reservation is a registrants
8695 				 * only type we need to generate a UA
8696 				 * for other registered inits.  The
8697 				 * sense code should be RESERVATIONS
8698 				 * RELEASED
8699 				 */
8700 
8701 				for (i = softc->init_min; i < softc->init_max; i++) {
8702 					if (ctl_get_prkey(lun, i) == 0)
8703 						continue;
8704 
8705 					ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8706 				}
8707 			}
8708 			lun->pr_res_type = 0;
8709 		} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8710 			if (lun->pr_key_count==0) {
8711 				lun->flags &= ~CTL_LUN_PR_RESERVED;
8712 				lun->pr_res_type = 0;
8713 				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8714 			}
8715 		}
8716 		lun->pr_generation++;
8717 		break;
8718 
8719 	case CTL_PR_RESERVE:
8720 		lun->flags |= CTL_LUN_PR_RESERVED;
8721 		lun->pr_res_type = msg->pr.pr_info.res_type;
8722 		lun->pr_res_idx = msg->pr.pr_info.residx;
8723 
8724 		break;
8725 
8726 	case CTL_PR_RELEASE:
8727 		/*
8728 		 * If this isn't an exclusive access reservation and NUAR
8729 		 * is not set, generate UA for all other registrants.
8730 		 */
8731 		if (lun->pr_res_type != SPR_TYPE_EX_AC &&
8732 		    lun->pr_res_type != SPR_TYPE_WR_EX &&
8733 		    (lun->MODE_CTRL.queue_flags & SCP_NUAR) == 0) {
8734 			for (i = softc->init_min; i < softc->init_max; i++) {
8735 				if (i == residx || ctl_get_prkey(lun, i) == 0)
8736 					continue;
8737 				ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8738 			}
8739 		}
8740 
8741 		lun->flags &= ~CTL_LUN_PR_RESERVED;
8742 		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8743 		lun->pr_res_type = 0;
8744 		break;
8745 
8746 	case CTL_PR_PREEMPT:
8747 		ctl_pro_preempt_other(lun, msg);
8748 		break;
8749 	case CTL_PR_CLEAR:
8750 		lun->flags &= ~CTL_LUN_PR_RESERVED;
8751 		lun->pr_res_type = 0;
8752 		lun->pr_key_count = 0;
8753 		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8754 
8755 		for (i=0; i < CTL_MAX_INITIATORS; i++) {
8756 			if (ctl_get_prkey(lun, i) == 0)
8757 				continue;
8758 			ctl_clr_prkey(lun, i);
8759 			ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8760 		}
8761 		lun->pr_generation++;
8762 		break;
8763 	}
8764 
8765 	mtx_unlock(&lun->lun_lock);
8766 }
8767 
8768 int
8769 ctl_read_write(struct ctl_scsiio *ctsio)
8770 {
8771 	struct ctl_lun *lun = CTL_LUN(ctsio);
8772 	struct ctl_lba_len_flags *lbalen;
8773 	uint64_t lba;
8774 	uint32_t num_blocks;
8775 	int flags, retval;
8776 	int isread;
8777 
8778 	CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8779 
8780 	flags = 0;
8781 	isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8782 	      || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8783 	switch (ctsio->cdb[0]) {
8784 	case READ_6:
8785 	case WRITE_6: {
8786 		struct scsi_rw_6 *cdb;
8787 
8788 		cdb = (struct scsi_rw_6 *)ctsio->cdb;
8789 
8790 		lba = scsi_3btoul(cdb->addr);
8791 		/* only 5 bits are valid in the most significant address byte */
8792 		lba &= 0x1fffff;
8793 		num_blocks = cdb->length;
8794 		/*
8795 		 * This is correct according to SBC-2.
8796 		 */
8797 		if (num_blocks == 0)
8798 			num_blocks = 256;
8799 		break;
8800 	}
8801 	case READ_10:
8802 	case WRITE_10: {
8803 		struct scsi_rw_10 *cdb;
8804 
8805 		cdb = (struct scsi_rw_10 *)ctsio->cdb;
8806 		if (cdb->byte2 & SRW10_FUA)
8807 			flags |= CTL_LLF_FUA;
8808 		if (cdb->byte2 & SRW10_DPO)
8809 			flags |= CTL_LLF_DPO;
8810 		lba = scsi_4btoul(cdb->addr);
8811 		num_blocks = scsi_2btoul(cdb->length);
8812 		break;
8813 	}
8814 	case WRITE_VERIFY_10: {
8815 		struct scsi_write_verify_10 *cdb;
8816 
8817 		cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8818 		flags |= CTL_LLF_FUA;
8819 		if (cdb->byte2 & SWV_DPO)
8820 			flags |= CTL_LLF_DPO;
8821 		lba = scsi_4btoul(cdb->addr);
8822 		num_blocks = scsi_2btoul(cdb->length);
8823 		break;
8824 	}
8825 	case READ_12:
8826 	case WRITE_12: {
8827 		struct scsi_rw_12 *cdb;
8828 
8829 		cdb = (struct scsi_rw_12 *)ctsio->cdb;
8830 		if (cdb->byte2 & SRW12_FUA)
8831 			flags |= CTL_LLF_FUA;
8832 		if (cdb->byte2 & SRW12_DPO)
8833 			flags |= CTL_LLF_DPO;
8834 		lba = scsi_4btoul(cdb->addr);
8835 		num_blocks = scsi_4btoul(cdb->length);
8836 		break;
8837 	}
8838 	case WRITE_VERIFY_12: {
8839 		struct scsi_write_verify_12 *cdb;
8840 
8841 		cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8842 		flags |= CTL_LLF_FUA;
8843 		if (cdb->byte2 & SWV_DPO)
8844 			flags |= CTL_LLF_DPO;
8845 		lba = scsi_4btoul(cdb->addr);
8846 		num_blocks = scsi_4btoul(cdb->length);
8847 		break;
8848 	}
8849 	case READ_16:
8850 	case WRITE_16: {
8851 		struct scsi_rw_16 *cdb;
8852 
8853 		cdb = (struct scsi_rw_16 *)ctsio->cdb;
8854 		if (cdb->byte2 & SRW12_FUA)
8855 			flags |= CTL_LLF_FUA;
8856 		if (cdb->byte2 & SRW12_DPO)
8857 			flags |= CTL_LLF_DPO;
8858 		lba = scsi_8btou64(cdb->addr);
8859 		num_blocks = scsi_4btoul(cdb->length);
8860 		break;
8861 	}
8862 	case WRITE_ATOMIC_16: {
8863 		struct scsi_write_atomic_16 *cdb;
8864 
8865 		if (lun->be_lun->atomicblock == 0) {
8866 			ctl_set_invalid_opcode(ctsio);
8867 			ctl_done((union ctl_io *)ctsio);
8868 			return (CTL_RETVAL_COMPLETE);
8869 		}
8870 
8871 		cdb = (struct scsi_write_atomic_16 *)ctsio->cdb;
8872 		if (cdb->byte2 & SRW12_FUA)
8873 			flags |= CTL_LLF_FUA;
8874 		if (cdb->byte2 & SRW12_DPO)
8875 			flags |= CTL_LLF_DPO;
8876 		lba = scsi_8btou64(cdb->addr);
8877 		num_blocks = scsi_2btoul(cdb->length);
8878 		if (num_blocks > lun->be_lun->atomicblock) {
8879 			ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
8880 			    /*command*/ 1, /*field*/ 12, /*bit_valid*/ 0,
8881 			    /*bit*/ 0);
8882 			ctl_done((union ctl_io *)ctsio);
8883 			return (CTL_RETVAL_COMPLETE);
8884 		}
8885 		break;
8886 	}
8887 	case WRITE_VERIFY_16: {
8888 		struct scsi_write_verify_16 *cdb;
8889 
8890 		cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
8891 		flags |= CTL_LLF_FUA;
8892 		if (cdb->byte2 & SWV_DPO)
8893 			flags |= CTL_LLF_DPO;
8894 		lba = scsi_8btou64(cdb->addr);
8895 		num_blocks = scsi_4btoul(cdb->length);
8896 		break;
8897 	}
8898 	default:
8899 		/*
8900 		 * We got a command we don't support.  This shouldn't
8901 		 * happen, commands should be filtered out above us.
8902 		 */
8903 		ctl_set_invalid_opcode(ctsio);
8904 		ctl_done((union ctl_io *)ctsio);
8905 
8906 		return (CTL_RETVAL_COMPLETE);
8907 		break; /* NOTREACHED */
8908 	}
8909 
8910 	/*
8911 	 * The first check is to make sure we're in bounds, the second
8912 	 * check is to catch wrap-around problems.  If the lba + num blocks
8913 	 * is less than the lba, then we've wrapped around and the block
8914 	 * range is invalid anyway.
8915 	 */
8916 	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8917 	 || ((lba + num_blocks) < lba)) {
8918 		ctl_set_lba_out_of_range(ctsio,
8919 		    MAX(lba, lun->be_lun->maxlba + 1));
8920 		ctl_done((union ctl_io *)ctsio);
8921 		return (CTL_RETVAL_COMPLETE);
8922 	}
8923 
8924 	/*
8925 	 * According to SBC-3, a transfer length of 0 is not an error.
8926 	 * Note that this cannot happen with WRITE(6) or READ(6), since 0
8927 	 * translates to 256 blocks for those commands.
8928 	 */
8929 	if (num_blocks == 0) {
8930 		ctl_set_success(ctsio);
8931 		ctl_done((union ctl_io *)ctsio);
8932 		return (CTL_RETVAL_COMPLETE);
8933 	}
8934 
8935 	/* Set FUA and/or DPO if caches are disabled. */
8936 	if (isread) {
8937 		if ((lun->MODE_CACHING.flags1 & SCP_RCD) != 0)
8938 			flags |= CTL_LLF_FUA | CTL_LLF_DPO;
8939 	} else {
8940 		if ((lun->MODE_CACHING.flags1 & SCP_WCE) == 0)
8941 			flags |= CTL_LLF_FUA;
8942 	}
8943 
8944 	lbalen = (struct ctl_lba_len_flags *)
8945 	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8946 	lbalen->lba = lba;
8947 	lbalen->len = num_blocks;
8948 	lbalen->flags = (isread ? CTL_LLF_READ : CTL_LLF_WRITE) | flags;
8949 
8950 	ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8951 	ctsio->kern_rel_offset = 0;
8952 
8953 	CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
8954 
8955 	retval = lun->backend->data_submit((union ctl_io *)ctsio);
8956 	return (retval);
8957 }
8958 
8959 static int
8960 ctl_cnw_cont(union ctl_io *io)
8961 {
8962 	struct ctl_lun *lun = CTL_LUN(io);
8963 	struct ctl_scsiio *ctsio;
8964 	struct ctl_lba_len_flags *lbalen;
8965 	int retval;
8966 
8967 	ctsio = &io->scsiio;
8968 	ctsio->io_hdr.status = CTL_STATUS_NONE;
8969 	ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
8970 	lbalen = (struct ctl_lba_len_flags *)
8971 	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8972 	lbalen->flags &= ~CTL_LLF_COMPARE;
8973 	lbalen->flags |= CTL_LLF_WRITE;
8974 
8975 	CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
8976 	retval = lun->backend->data_submit((union ctl_io *)ctsio);
8977 	return (retval);
8978 }
8979 
8980 int
8981 ctl_cnw(struct ctl_scsiio *ctsio)
8982 {
8983 	struct ctl_lun *lun = CTL_LUN(ctsio);
8984 	struct ctl_lba_len_flags *lbalen;
8985 	uint64_t lba;
8986 	uint32_t num_blocks;
8987 	int flags, retval;
8988 
8989 	CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
8990 
8991 	flags = 0;
8992 	switch (ctsio->cdb[0]) {
8993 	case COMPARE_AND_WRITE: {
8994 		struct scsi_compare_and_write *cdb;
8995 
8996 		cdb = (struct scsi_compare_and_write *)ctsio->cdb;
8997 		if (cdb->byte2 & SRW10_FUA)
8998 			flags |= CTL_LLF_FUA;
8999 		if (cdb->byte2 & SRW10_DPO)
9000 			flags |= CTL_LLF_DPO;
9001 		lba = scsi_8btou64(cdb->addr);
9002 		num_blocks = cdb->length;
9003 		break;
9004 	}
9005 	default:
9006 		/*
9007 		 * We got a command we don't support.  This shouldn't
9008 		 * happen, commands should be filtered out above us.
9009 		 */
9010 		ctl_set_invalid_opcode(ctsio);
9011 		ctl_done((union ctl_io *)ctsio);
9012 
9013 		return (CTL_RETVAL_COMPLETE);
9014 		break; /* NOTREACHED */
9015 	}
9016 
9017 	/*
9018 	 * The first check is to make sure we're in bounds, the second
9019 	 * check is to catch wrap-around problems.  If the lba + num blocks
9020 	 * is less than the lba, then we've wrapped around and the block
9021 	 * range is invalid anyway.
9022 	 */
9023 	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9024 	 || ((lba + num_blocks) < lba)) {
9025 		ctl_set_lba_out_of_range(ctsio,
9026 		    MAX(lba, lun->be_lun->maxlba + 1));
9027 		ctl_done((union ctl_io *)ctsio);
9028 		return (CTL_RETVAL_COMPLETE);
9029 	}
9030 
9031 	/*
9032 	 * According to SBC-3, a transfer length of 0 is not an error.
9033 	 */
9034 	if (num_blocks == 0) {
9035 		ctl_set_success(ctsio);
9036 		ctl_done((union ctl_io *)ctsio);
9037 		return (CTL_RETVAL_COMPLETE);
9038 	}
9039 
9040 	/* Set FUA if write cache is disabled. */
9041 	if ((lun->MODE_CACHING.flags1 & SCP_WCE) == 0)
9042 		flags |= CTL_LLF_FUA;
9043 
9044 	ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
9045 	ctsio->kern_rel_offset = 0;
9046 
9047 	/*
9048 	 * Set the IO_CONT flag, so that if this I/O gets passed to
9049 	 * ctl_data_submit_done(), it'll get passed back to
9050 	 * ctl_ctl_cnw_cont() for further processing.
9051 	 */
9052 	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
9053 	ctsio->io_cont = ctl_cnw_cont;
9054 
9055 	lbalen = (struct ctl_lba_len_flags *)
9056 	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9057 	lbalen->lba = lba;
9058 	lbalen->len = num_blocks;
9059 	lbalen->flags = CTL_LLF_COMPARE | flags;
9060 
9061 	CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
9062 	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9063 	return (retval);
9064 }
9065 
9066 int
9067 ctl_verify(struct ctl_scsiio *ctsio)
9068 {
9069 	struct ctl_lun *lun = CTL_LUN(ctsio);
9070 	struct ctl_lba_len_flags *lbalen;
9071 	uint64_t lba;
9072 	uint32_t num_blocks;
9073 	int bytchk, flags;
9074 	int retval;
9075 
9076 	CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
9077 
9078 	bytchk = 0;
9079 	flags = CTL_LLF_FUA;
9080 	switch (ctsio->cdb[0]) {
9081 	case VERIFY_10: {
9082 		struct scsi_verify_10 *cdb;
9083 
9084 		cdb = (struct scsi_verify_10 *)ctsio->cdb;
9085 		if (cdb->byte2 & SVFY_BYTCHK)
9086 			bytchk = 1;
9087 		if (cdb->byte2 & SVFY_DPO)
9088 			flags |= CTL_LLF_DPO;
9089 		lba = scsi_4btoul(cdb->addr);
9090 		num_blocks = scsi_2btoul(cdb->length);
9091 		break;
9092 	}
9093 	case VERIFY_12: {
9094 		struct scsi_verify_12 *cdb;
9095 
9096 		cdb = (struct scsi_verify_12 *)ctsio->cdb;
9097 		if (cdb->byte2 & SVFY_BYTCHK)
9098 			bytchk = 1;
9099 		if (cdb->byte2 & SVFY_DPO)
9100 			flags |= CTL_LLF_DPO;
9101 		lba = scsi_4btoul(cdb->addr);
9102 		num_blocks = scsi_4btoul(cdb->length);
9103 		break;
9104 	}
9105 	case VERIFY_16: {
9106 		struct scsi_rw_16 *cdb;
9107 
9108 		cdb = (struct scsi_rw_16 *)ctsio->cdb;
9109 		if (cdb->byte2 & SVFY_BYTCHK)
9110 			bytchk = 1;
9111 		if (cdb->byte2 & SVFY_DPO)
9112 			flags |= CTL_LLF_DPO;
9113 		lba = scsi_8btou64(cdb->addr);
9114 		num_blocks = scsi_4btoul(cdb->length);
9115 		break;
9116 	}
9117 	default:
9118 		/*
9119 		 * We got a command we don't support.  This shouldn't
9120 		 * happen, commands should be filtered out above us.
9121 		 */
9122 		ctl_set_invalid_opcode(ctsio);
9123 		ctl_done((union ctl_io *)ctsio);
9124 		return (CTL_RETVAL_COMPLETE);
9125 	}
9126 
9127 	/*
9128 	 * The first check is to make sure we're in bounds, the second
9129 	 * check is to catch wrap-around problems.  If the lba + num blocks
9130 	 * is less than the lba, then we've wrapped around and the block
9131 	 * range is invalid anyway.
9132 	 */
9133 	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9134 	 || ((lba + num_blocks) < lba)) {
9135 		ctl_set_lba_out_of_range(ctsio,
9136 		    MAX(lba, lun->be_lun->maxlba + 1));
9137 		ctl_done((union ctl_io *)ctsio);
9138 		return (CTL_RETVAL_COMPLETE);
9139 	}
9140 
9141 	/*
9142 	 * According to SBC-3, a transfer length of 0 is not an error.
9143 	 */
9144 	if (num_blocks == 0) {
9145 		ctl_set_success(ctsio);
9146 		ctl_done((union ctl_io *)ctsio);
9147 		return (CTL_RETVAL_COMPLETE);
9148 	}
9149 
9150 	lbalen = (struct ctl_lba_len_flags *)
9151 	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9152 	lbalen->lba = lba;
9153 	lbalen->len = num_blocks;
9154 	if (bytchk) {
9155 		lbalen->flags = CTL_LLF_COMPARE | flags;
9156 		ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9157 	} else {
9158 		lbalen->flags = CTL_LLF_VERIFY | flags;
9159 		ctsio->kern_total_len = 0;
9160 	}
9161 	ctsio->kern_rel_offset = 0;
9162 
9163 	CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9164 	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9165 	return (retval);
9166 }
9167 
9168 int
9169 ctl_report_luns(struct ctl_scsiio *ctsio)
9170 {
9171 	struct ctl_softc *softc = CTL_SOFTC(ctsio);
9172 	struct ctl_port *port = CTL_PORT(ctsio);
9173 	struct ctl_lun *lun, *request_lun = CTL_LUN(ctsio);
9174 	struct scsi_report_luns *cdb;
9175 	struct scsi_report_luns_data *lun_data;
9176 	int num_filled, num_luns, num_port_luns, retval;
9177 	uint32_t alloc_len, lun_datalen;
9178 	uint32_t initidx, targ_lun_id, lun_id;
9179 
9180 	retval = CTL_RETVAL_COMPLETE;
9181 	cdb = (struct scsi_report_luns *)ctsio->cdb;
9182 
9183 	CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9184 
9185 	num_luns = 0;
9186 	num_port_luns = port->lun_map ? port->lun_map_size : ctl_max_luns;
9187 	mtx_lock(&softc->ctl_lock);
9188 	for (targ_lun_id = 0; targ_lun_id < num_port_luns; targ_lun_id++) {
9189 		if (ctl_lun_map_from_port(port, targ_lun_id) != UINT32_MAX)
9190 			num_luns++;
9191 	}
9192 	mtx_unlock(&softc->ctl_lock);
9193 
9194 	switch (cdb->select_report) {
9195 	case RPL_REPORT_DEFAULT:
9196 	case RPL_REPORT_ALL:
9197 	case RPL_REPORT_NONSUBSID:
9198 		break;
9199 	case RPL_REPORT_WELLKNOWN:
9200 	case RPL_REPORT_ADMIN:
9201 	case RPL_REPORT_CONGLOM:
9202 		num_luns = 0;
9203 		break;
9204 	default:
9205 		ctl_set_invalid_field(ctsio,
9206 				      /*sks_valid*/ 1,
9207 				      /*command*/ 1,
9208 				      /*field*/ 2,
9209 				      /*bit_valid*/ 0,
9210 				      /*bit*/ 0);
9211 		ctl_done((union ctl_io *)ctsio);
9212 		return (retval);
9213 		break; /* NOTREACHED */
9214 	}
9215 
9216 	alloc_len = scsi_4btoul(cdb->length);
9217 	/*
9218 	 * The initiator has to allocate at least 16 bytes for this request,
9219 	 * so he can at least get the header and the first LUN.  Otherwise
9220 	 * we reject the request (per SPC-3 rev 14, section 6.21).
9221 	 */
9222 	if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9223 	    sizeof(struct scsi_report_luns_lundata))) {
9224 		ctl_set_invalid_field(ctsio,
9225 				      /*sks_valid*/ 1,
9226 				      /*command*/ 1,
9227 				      /*field*/ 6,
9228 				      /*bit_valid*/ 0,
9229 				      /*bit*/ 0);
9230 		ctl_done((union ctl_io *)ctsio);
9231 		return (retval);
9232 	}
9233 
9234 	lun_datalen = sizeof(*lun_data) +
9235 		(num_luns * sizeof(struct scsi_report_luns_lundata));
9236 
9237 	ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9238 	lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9239 	ctsio->kern_sg_entries = 0;
9240 
9241 	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9242 
9243 	mtx_lock(&softc->ctl_lock);
9244 	for (targ_lun_id = 0, num_filled = 0;
9245 	    targ_lun_id < num_port_luns && num_filled < num_luns;
9246 	    targ_lun_id++) {
9247 		lun_id = ctl_lun_map_from_port(port, targ_lun_id);
9248 		if (lun_id == UINT32_MAX)
9249 			continue;
9250 		lun = softc->ctl_luns[lun_id];
9251 		if (lun == NULL)
9252 			continue;
9253 
9254 		be64enc(lun_data->luns[num_filled++].lundata,
9255 		    ctl_encode_lun(targ_lun_id));
9256 
9257 		/*
9258 		 * According to SPC-3, rev 14 section 6.21:
9259 		 *
9260 		 * "The execution of a REPORT LUNS command to any valid and
9261 		 * installed logical unit shall clear the REPORTED LUNS DATA
9262 		 * HAS CHANGED unit attention condition for all logical
9263 		 * units of that target with respect to the requesting
9264 		 * initiator. A valid and installed logical unit is one
9265 		 * having a PERIPHERAL QUALIFIER of 000b in the standard
9266 		 * INQUIRY data (see 6.4.2)."
9267 		 *
9268 		 * If request_lun is NULL, the LUN this report luns command
9269 		 * was issued to is either disabled or doesn't exist. In that
9270 		 * case, we shouldn't clear any pending lun change unit
9271 		 * attention.
9272 		 */
9273 		if (request_lun != NULL) {
9274 			mtx_lock(&lun->lun_lock);
9275 			ctl_clr_ua(lun, initidx, CTL_UA_LUN_CHANGE);
9276 			mtx_unlock(&lun->lun_lock);
9277 		}
9278 	}
9279 	mtx_unlock(&softc->ctl_lock);
9280 
9281 	/*
9282 	 * It's quite possible that we've returned fewer LUNs than we allocated
9283 	 * space for.  Trim it.
9284 	 */
9285 	lun_datalen = sizeof(*lun_data) +
9286 		(num_filled * sizeof(struct scsi_report_luns_lundata));
9287 	ctsio->kern_rel_offset = 0;
9288 	ctsio->kern_sg_entries = 0;
9289 	ctsio->kern_data_len = min(lun_datalen, alloc_len);
9290 	ctsio->kern_total_len = ctsio->kern_data_len;
9291 
9292 	/*
9293 	 * We set this to the actual data length, regardless of how much
9294 	 * space we actually have to return results.  If the user looks at
9295 	 * this value, he'll know whether or not he allocated enough space
9296 	 * and reissue the command if necessary.  We don't support well
9297 	 * known logical units, so if the user asks for that, return none.
9298 	 */
9299 	scsi_ulto4b(lun_datalen - 8, lun_data->length);
9300 
9301 	/*
9302 	 * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9303 	 * this request.
9304 	 */
9305 	ctl_set_success(ctsio);
9306 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9307 	ctsio->be_move_done = ctl_config_move_done;
9308 	ctl_datamove((union ctl_io *)ctsio);
9309 	return (retval);
9310 }
9311 
9312 int
9313 ctl_request_sense(struct ctl_scsiio *ctsio)
9314 {
9315 	struct ctl_softc *softc = CTL_SOFTC(ctsio);
9316 	struct ctl_lun *lun = CTL_LUN(ctsio);
9317 	struct scsi_request_sense *cdb;
9318 	struct scsi_sense_data *sense_ptr, *ps;
9319 	uint32_t initidx;
9320 	int have_error;
9321 	u_int sense_len = SSD_FULL_SIZE;
9322 	scsi_sense_data_type sense_format;
9323 	ctl_ua_type ua_type;
9324 	uint8_t asc = 0, ascq = 0;
9325 
9326 	cdb = (struct scsi_request_sense *)ctsio->cdb;
9327 
9328 	CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9329 
9330 	/*
9331 	 * Determine which sense format the user wants.
9332 	 */
9333 	if (cdb->byte2 & SRS_DESC)
9334 		sense_format = SSD_TYPE_DESC;
9335 	else
9336 		sense_format = SSD_TYPE_FIXED;
9337 
9338 	ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9339 	sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9340 	ctsio->kern_sg_entries = 0;
9341 	ctsio->kern_rel_offset = 0;
9342 
9343 	/*
9344 	 * struct scsi_sense_data, which is currently set to 256 bytes, is
9345 	 * larger than the largest allowed value for the length field in the
9346 	 * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9347 	 */
9348 	ctsio->kern_data_len = cdb->length;
9349 	ctsio->kern_total_len = cdb->length;
9350 
9351 	/*
9352 	 * If we don't have a LUN, we don't have any pending sense.
9353 	 */
9354 	if (lun == NULL ||
9355 	    ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
9356 	     softc->ha_link < CTL_HA_LINK_UNKNOWN)) {
9357 		/* "Logical unit not supported" */
9358 		ctl_set_sense_data(sense_ptr, &sense_len, NULL, sense_format,
9359 		    /*current_error*/ 1,
9360 		    /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
9361 		    /*asc*/ 0x25,
9362 		    /*ascq*/ 0x00,
9363 		    SSD_ELEM_NONE);
9364 		goto send;
9365 	}
9366 
9367 	have_error = 0;
9368 	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9369 	/*
9370 	 * Check for pending sense, and then for pending unit attentions.
9371 	 * Pending sense gets returned first, then pending unit attentions.
9372 	 */
9373 	mtx_lock(&lun->lun_lock);
9374 	ps = lun->pending_sense[initidx / CTL_MAX_INIT_PER_PORT];
9375 	if (ps != NULL)
9376 		ps += initidx % CTL_MAX_INIT_PER_PORT;
9377 	if (ps != NULL && ps->error_code != 0) {
9378 		scsi_sense_data_type stored_format;
9379 
9380 		/*
9381 		 * Check to see which sense format was used for the stored
9382 		 * sense data.
9383 		 */
9384 		stored_format = scsi_sense_type(ps);
9385 
9386 		/*
9387 		 * If the user requested a different sense format than the
9388 		 * one we stored, then we need to convert it to the other
9389 		 * format.  If we're going from descriptor to fixed format
9390 		 * sense data, we may lose things in translation, depending
9391 		 * on what options were used.
9392 		 *
9393 		 * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9394 		 * for some reason we'll just copy it out as-is.
9395 		 */
9396 		if ((stored_format == SSD_TYPE_FIXED)
9397 		 && (sense_format == SSD_TYPE_DESC))
9398 			ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9399 			    ps, (struct scsi_sense_data_desc *)sense_ptr);
9400 		else if ((stored_format == SSD_TYPE_DESC)
9401 		      && (sense_format == SSD_TYPE_FIXED))
9402 			ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9403 			    ps, (struct scsi_sense_data_fixed *)sense_ptr);
9404 		else
9405 			memcpy(sense_ptr, ps, sizeof(*sense_ptr));
9406 
9407 		ps->error_code = 0;
9408 		have_error = 1;
9409 	} else {
9410 		ua_type = ctl_build_ua(lun, initidx, sense_ptr, &sense_len,
9411 		    sense_format);
9412 		if (ua_type != CTL_UA_NONE)
9413 			have_error = 1;
9414 	}
9415 	if (have_error == 0) {
9416 		/*
9417 		 * Report informational exception if have one and allowed.
9418 		 */
9419 		if (lun->MODE_IE.mrie != SIEP_MRIE_NO) {
9420 			asc = lun->ie_asc;
9421 			ascq = lun->ie_ascq;
9422 		}
9423 		ctl_set_sense_data(sense_ptr, &sense_len, lun, sense_format,
9424 		    /*current_error*/ 1,
9425 		    /*sense_key*/ SSD_KEY_NO_SENSE,
9426 		    /*asc*/ asc,
9427 		    /*ascq*/ ascq,
9428 		    SSD_ELEM_NONE);
9429 	}
9430 	mtx_unlock(&lun->lun_lock);
9431 
9432 send:
9433 	/*
9434 	 * We report the SCSI status as OK, since the status of the command
9435 	 * itself is OK.  We're reporting sense as parameter data.
9436 	 */
9437 	ctl_set_success(ctsio);
9438 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9439 	ctsio->be_move_done = ctl_config_move_done;
9440 	ctl_datamove((union ctl_io *)ctsio);
9441 	return (CTL_RETVAL_COMPLETE);
9442 }
9443 
9444 int
9445 ctl_tur(struct ctl_scsiio *ctsio)
9446 {
9447 
9448 	CTL_DEBUG_PRINT(("ctl_tur\n"));
9449 
9450 	ctl_set_success(ctsio);
9451 	ctl_done((union ctl_io *)ctsio);
9452 
9453 	return (CTL_RETVAL_COMPLETE);
9454 }
9455 
9456 /*
9457  * SCSI VPD page 0x00, the Supported VPD Pages page.
9458  */
9459 static int
9460 ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9461 {
9462 	struct ctl_lun *lun = CTL_LUN(ctsio);
9463 	struct scsi_vpd_supported_pages *pages;
9464 	int sup_page_size;
9465 	int p;
9466 
9467 	sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9468 	    SCSI_EVPD_NUM_SUPPORTED_PAGES;
9469 	ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9470 	pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9471 	ctsio->kern_rel_offset = 0;
9472 	ctsio->kern_sg_entries = 0;
9473 	ctsio->kern_data_len = min(sup_page_size, alloc_len);
9474 	ctsio->kern_total_len = ctsio->kern_data_len;
9475 
9476 	/*
9477 	 * The control device is always connected.  The disk device, on the
9478 	 * other hand, may not be online all the time.  Need to change this
9479 	 * to figure out whether the disk device is actually online or not.
9480 	 */
9481 	if (lun != NULL)
9482 		pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9483 				lun->be_lun->lun_type;
9484 	else
9485 		pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9486 
9487 	p = 0;
9488 	/* Supported VPD pages */
9489 	pages->page_list[p++] = SVPD_SUPPORTED_PAGES;
9490 	/* Serial Number */
9491 	pages->page_list[p++] = SVPD_UNIT_SERIAL_NUMBER;
9492 	/* Device Identification */
9493 	pages->page_list[p++] = SVPD_DEVICE_ID;
9494 	/* Extended INQUIRY Data */
9495 	pages->page_list[p++] = SVPD_EXTENDED_INQUIRY_DATA;
9496 	/* Mode Page Policy */
9497 	pages->page_list[p++] = SVPD_MODE_PAGE_POLICY;
9498 	/* SCSI Ports */
9499 	pages->page_list[p++] = SVPD_SCSI_PORTS;
9500 	/* Third-party Copy */
9501 	pages->page_list[p++] = SVPD_SCSI_TPC;
9502 	/* SCSI Feature Sets */
9503 	pages->page_list[p++] = SVPD_SCSI_SFS;
9504 	if (lun != NULL && lun->be_lun->lun_type == T_DIRECT) {
9505 		/* Block limits */
9506 		pages->page_list[p++] = SVPD_BLOCK_LIMITS;
9507 		/* Block Device Characteristics */
9508 		pages->page_list[p++] = SVPD_BDC;
9509 		/* Logical Block Provisioning */
9510 		pages->page_list[p++] = SVPD_LBP;
9511 	}
9512 	pages->length = p;
9513 
9514 	ctl_set_success(ctsio);
9515 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9516 	ctsio->be_move_done = ctl_config_move_done;
9517 	ctl_datamove((union ctl_io *)ctsio);
9518 	return (CTL_RETVAL_COMPLETE);
9519 }
9520 
9521 /*
9522  * SCSI VPD page 0x80, the Unit Serial Number page.
9523  */
9524 static int
9525 ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9526 {
9527 	struct ctl_lun *lun = CTL_LUN(ctsio);
9528 	struct scsi_vpd_unit_serial_number *sn_ptr;
9529 	int data_len;
9530 
9531 	data_len = 4 + CTL_SN_LEN;
9532 	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9533 	sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9534 	ctsio->kern_rel_offset = 0;
9535 	ctsio->kern_sg_entries = 0;
9536 	ctsio->kern_data_len = min(data_len, alloc_len);
9537 	ctsio->kern_total_len = ctsio->kern_data_len;
9538 
9539 	/*
9540 	 * The control device is always connected.  The disk device, on the
9541 	 * other hand, may not be online all the time.  Need to change this
9542 	 * to figure out whether the disk device is actually online or not.
9543 	 */
9544 	if (lun != NULL)
9545 		sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9546 				  lun->be_lun->lun_type;
9547 	else
9548 		sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9549 
9550 	sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9551 	sn_ptr->length = CTL_SN_LEN;
9552 	/*
9553 	 * If we don't have a LUN, we just leave the serial number as
9554 	 * all spaces.
9555 	 */
9556 	if (lun != NULL) {
9557 		strncpy((char *)sn_ptr->serial_num,
9558 			(char *)lun->be_lun->serial_num, CTL_SN_LEN);
9559 	} else
9560 		memset(sn_ptr->serial_num, 0x20, CTL_SN_LEN);
9561 
9562 	ctl_set_success(ctsio);
9563 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9564 	ctsio->be_move_done = ctl_config_move_done;
9565 	ctl_datamove((union ctl_io *)ctsio);
9566 	return (CTL_RETVAL_COMPLETE);
9567 }
9568 
9569 /*
9570  * SCSI VPD page 0x86, the Extended INQUIRY Data page.
9571  */
9572 static int
9573 ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len)
9574 {
9575 	struct ctl_lun *lun = CTL_LUN(ctsio);
9576 	struct scsi_vpd_extended_inquiry_data *eid_ptr;
9577 	int data_len;
9578 
9579 	data_len = sizeof(struct scsi_vpd_extended_inquiry_data);
9580 	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9581 	eid_ptr = (struct scsi_vpd_extended_inquiry_data *)ctsio->kern_data_ptr;
9582 	ctsio->kern_sg_entries = 0;
9583 	ctsio->kern_rel_offset = 0;
9584 	ctsio->kern_data_len = min(data_len, alloc_len);
9585 	ctsio->kern_total_len = ctsio->kern_data_len;
9586 
9587 	/*
9588 	 * The control device is always connected.  The disk device, on the
9589 	 * other hand, may not be online all the time.
9590 	 */
9591 	if (lun != NULL)
9592 		eid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9593 				     lun->be_lun->lun_type;
9594 	else
9595 		eid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9596 	eid_ptr->page_code = SVPD_EXTENDED_INQUIRY_DATA;
9597 	scsi_ulto2b(data_len - 4, eid_ptr->page_length);
9598 	/*
9599 	 * We support head of queue, ordered and simple tags.
9600 	 */
9601 	eid_ptr->flags2 = SVPD_EID_HEADSUP | SVPD_EID_ORDSUP | SVPD_EID_SIMPSUP;
9602 	/*
9603 	 * Volatile cache supported.
9604 	 */
9605 	eid_ptr->flags3 = SVPD_EID_V_SUP;
9606 
9607 	/*
9608 	 * This means that we clear the REPORTED LUNS DATA HAS CHANGED unit
9609 	 * attention for a particular IT nexus on all LUNs once we report
9610 	 * it to that nexus once.  This bit is required as of SPC-4.
9611 	 */
9612 	eid_ptr->flags4 = SVPD_EID_LUICLR;
9613 
9614 	/*
9615 	 * We support revert to defaults (RTD) bit in MODE SELECT.
9616 	 */
9617 	eid_ptr->flags5 = SVPD_EID_RTD_SUP;
9618 
9619 	/*
9620 	 * XXX KDM in order to correctly answer this, we would need
9621 	 * information from the SIM to determine how much sense data it
9622 	 * can send.  So this would really be a path inquiry field, most
9623 	 * likely.  This can be set to a maximum of 252 according to SPC-4,
9624 	 * but the hardware may or may not be able to support that much.
9625 	 * 0 just means that the maximum sense data length is not reported.
9626 	 */
9627 	eid_ptr->max_sense_length = 0;
9628 
9629 	ctl_set_success(ctsio);
9630 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9631 	ctsio->be_move_done = ctl_config_move_done;
9632 	ctl_datamove((union ctl_io *)ctsio);
9633 	return (CTL_RETVAL_COMPLETE);
9634 }
9635 
9636 static int
9637 ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len)
9638 {
9639 	struct ctl_lun *lun = CTL_LUN(ctsio);
9640 	struct scsi_vpd_mode_page_policy *mpp_ptr;
9641 	int data_len;
9642 
9643 	data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9644 	    sizeof(struct scsi_vpd_mode_page_policy_descr);
9645 
9646 	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9647 	mpp_ptr = (struct scsi_vpd_mode_page_policy *)ctsio->kern_data_ptr;
9648 	ctsio->kern_rel_offset = 0;
9649 	ctsio->kern_sg_entries = 0;
9650 	ctsio->kern_data_len = min(data_len, alloc_len);
9651 	ctsio->kern_total_len = ctsio->kern_data_len;
9652 
9653 	/*
9654 	 * The control device is always connected.  The disk device, on the
9655 	 * other hand, may not be online all the time.
9656 	 */
9657 	if (lun != NULL)
9658 		mpp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9659 				     lun->be_lun->lun_type;
9660 	else
9661 		mpp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9662 	mpp_ptr->page_code = SVPD_MODE_PAGE_POLICY;
9663 	scsi_ulto2b(data_len - 4, mpp_ptr->page_length);
9664 	mpp_ptr->descr[0].page_code = 0x3f;
9665 	mpp_ptr->descr[0].subpage_code = 0xff;
9666 	mpp_ptr->descr[0].policy = SVPD_MPP_SHARED;
9667 
9668 	ctl_set_success(ctsio);
9669 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9670 	ctsio->be_move_done = ctl_config_move_done;
9671 	ctl_datamove((union ctl_io *)ctsio);
9672 	return (CTL_RETVAL_COMPLETE);
9673 }
9674 
9675 /*
9676  * SCSI VPD page 0x83, the Device Identification page.
9677  */
9678 static int
9679 ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9680 {
9681 	struct ctl_softc *softc = CTL_SOFTC(ctsio);
9682 	struct ctl_port *port = CTL_PORT(ctsio);
9683 	struct ctl_lun *lun = CTL_LUN(ctsio);
9684 	struct scsi_vpd_device_id *devid_ptr;
9685 	struct scsi_vpd_id_descriptor *desc;
9686 	int data_len, g;
9687 	uint8_t proto;
9688 
9689 	data_len = sizeof(struct scsi_vpd_device_id) +
9690 	    sizeof(struct scsi_vpd_id_descriptor) +
9691 		sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9692 	    sizeof(struct scsi_vpd_id_descriptor) +
9693 		sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9694 	if (lun && lun->lun_devid)
9695 		data_len += lun->lun_devid->len;
9696 	if (port && port->port_devid)
9697 		data_len += port->port_devid->len;
9698 	if (port && port->target_devid)
9699 		data_len += port->target_devid->len;
9700 
9701 	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9702 	devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9703 	ctsio->kern_sg_entries = 0;
9704 	ctsio->kern_rel_offset = 0;
9705 	ctsio->kern_sg_entries = 0;
9706 	ctsio->kern_data_len = min(data_len, alloc_len);
9707 	ctsio->kern_total_len = ctsio->kern_data_len;
9708 
9709 	/*
9710 	 * The control device is always connected.  The disk device, on the
9711 	 * other hand, may not be online all the time.
9712 	 */
9713 	if (lun != NULL)
9714 		devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9715 				     lun->be_lun->lun_type;
9716 	else
9717 		devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9718 	devid_ptr->page_code = SVPD_DEVICE_ID;
9719 	scsi_ulto2b(data_len - 4, devid_ptr->length);
9720 
9721 	if (port && port->port_type == CTL_PORT_FC)
9722 		proto = SCSI_PROTO_FC << 4;
9723 	else if (port && port->port_type == CTL_PORT_SAS)
9724 		proto = SCSI_PROTO_SAS << 4;
9725 	else if (port && port->port_type == CTL_PORT_ISCSI)
9726 		proto = SCSI_PROTO_ISCSI << 4;
9727 	else
9728 		proto = SCSI_PROTO_SPI << 4;
9729 	desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9730 
9731 	/*
9732 	 * We're using a LUN association here.  i.e., this device ID is a
9733 	 * per-LUN identifier.
9734 	 */
9735 	if (lun && lun->lun_devid) {
9736 		memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
9737 		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9738 		    lun->lun_devid->len);
9739 	}
9740 
9741 	/*
9742 	 * This is for the WWPN which is a port association.
9743 	 */
9744 	if (port && port->port_devid) {
9745 		memcpy(desc, port->port_devid->data, port->port_devid->len);
9746 		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9747 		    port->port_devid->len);
9748 	}
9749 
9750 	/*
9751 	 * This is for the Relative Target Port(type 4h) identifier
9752 	 */
9753 	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9754 	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9755 	    SVPD_ID_TYPE_RELTARG;
9756 	desc->length = 4;
9757 	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
9758 	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9759 	    sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9760 
9761 	/*
9762 	 * This is for the Target Port Group(type 5h) identifier
9763 	 */
9764 	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9765 	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9766 	    SVPD_ID_TYPE_TPORTGRP;
9767 	desc->length = 4;
9768 	if (softc->is_single ||
9769 	    (port && port->status & CTL_PORT_STATUS_HA_SHARED))
9770 		g = 1;
9771 	else
9772 		g = 2 + ctsio->io_hdr.nexus.targ_port / softc->port_cnt;
9773 	scsi_ulto2b(g, &desc->identifier[2]);
9774 	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9775 	    sizeof(struct scsi_vpd_id_trgt_port_grp_id));
9776 
9777 	/*
9778 	 * This is for the Target identifier
9779 	 */
9780 	if (port && port->target_devid) {
9781 		memcpy(desc, port->target_devid->data, port->target_devid->len);
9782 	}
9783 
9784 	ctl_set_success(ctsio);
9785 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9786 	ctsio->be_move_done = ctl_config_move_done;
9787 	ctl_datamove((union ctl_io *)ctsio);
9788 	return (CTL_RETVAL_COMPLETE);
9789 }
9790 
9791 static int
9792 ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
9793 {
9794 	struct ctl_softc *softc = CTL_SOFTC(ctsio);
9795 	struct ctl_lun *lun = CTL_LUN(ctsio);
9796 	struct scsi_vpd_scsi_ports *sp;
9797 	struct scsi_vpd_port_designation *pd;
9798 	struct scsi_vpd_port_designation_cont *pdc;
9799 	struct ctl_port *port;
9800 	int data_len, num_target_ports, iid_len, id_len;
9801 
9802 	num_target_ports = 0;
9803 	iid_len = 0;
9804 	id_len = 0;
9805 	mtx_lock(&softc->ctl_lock);
9806 	STAILQ_FOREACH(port, &softc->port_list, links) {
9807 		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9808 			continue;
9809 		if (lun != NULL &&
9810 		    ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
9811 			continue;
9812 		num_target_ports++;
9813 		if (port->init_devid)
9814 			iid_len += port->init_devid->len;
9815 		if (port->port_devid)
9816 			id_len += port->port_devid->len;
9817 	}
9818 	mtx_unlock(&softc->ctl_lock);
9819 
9820 	data_len = sizeof(struct scsi_vpd_scsi_ports) +
9821 	    num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
9822 	     sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
9823 	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9824 	sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
9825 	ctsio->kern_sg_entries = 0;
9826 	ctsio->kern_rel_offset = 0;
9827 	ctsio->kern_sg_entries = 0;
9828 	ctsio->kern_data_len = min(data_len, alloc_len);
9829 	ctsio->kern_total_len = ctsio->kern_data_len;
9830 
9831 	/*
9832 	 * The control device is always connected.  The disk device, on the
9833 	 * other hand, may not be online all the time.  Need to change this
9834 	 * to figure out whether the disk device is actually online or not.
9835 	 */
9836 	if (lun != NULL)
9837 		sp->device = (SID_QUAL_LU_CONNECTED << 5) |
9838 				  lun->be_lun->lun_type;
9839 	else
9840 		sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9841 
9842 	sp->page_code = SVPD_SCSI_PORTS;
9843 	scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
9844 	    sp->page_length);
9845 	pd = &sp->design[0];
9846 
9847 	mtx_lock(&softc->ctl_lock);
9848 	STAILQ_FOREACH(port, &softc->port_list, links) {
9849 		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9850 			continue;
9851 		if (lun != NULL &&
9852 		    ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
9853 			continue;
9854 		scsi_ulto2b(port->targ_port, pd->relative_port_id);
9855 		if (port->init_devid) {
9856 			iid_len = port->init_devid->len;
9857 			memcpy(pd->initiator_transportid,
9858 			    port->init_devid->data, port->init_devid->len);
9859 		} else
9860 			iid_len = 0;
9861 		scsi_ulto2b(iid_len, pd->initiator_transportid_length);
9862 		pdc = (struct scsi_vpd_port_designation_cont *)
9863 		    (&pd->initiator_transportid[iid_len]);
9864 		if (port->port_devid) {
9865 			id_len = port->port_devid->len;
9866 			memcpy(pdc->target_port_descriptors,
9867 			    port->port_devid->data, port->port_devid->len);
9868 		} else
9869 			id_len = 0;
9870 		scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
9871 		pd = (struct scsi_vpd_port_designation *)
9872 		    ((uint8_t *)pdc->target_port_descriptors + id_len);
9873 	}
9874 	mtx_unlock(&softc->ctl_lock);
9875 
9876 	ctl_set_success(ctsio);
9877 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9878 	ctsio->be_move_done = ctl_config_move_done;
9879 	ctl_datamove((union ctl_io *)ctsio);
9880 	return (CTL_RETVAL_COMPLETE);
9881 }
9882 
9883 static int
9884 ctl_inquiry_evpd_sfs(struct ctl_scsiio *ctsio, int alloc_len)
9885 {
9886 	struct ctl_lun *lun = CTL_LUN(ctsio);
9887 	struct scsi_vpd_sfs *sfs_ptr;
9888 	int sfs_page_size, n;
9889 
9890 	sfs_page_size = sizeof(*sfs_ptr) + 5 * 2;
9891 	ctsio->kern_data_ptr = malloc(sfs_page_size, M_CTL, M_WAITOK | M_ZERO);
9892 	sfs_ptr = (struct scsi_vpd_sfs *)ctsio->kern_data_ptr;
9893 	ctsio->kern_sg_entries = 0;
9894 	ctsio->kern_rel_offset = 0;
9895 	ctsio->kern_sg_entries = 0;
9896 	ctsio->kern_data_len = min(sfs_page_size, alloc_len);
9897 	ctsio->kern_total_len = ctsio->kern_data_len;
9898 
9899 	/*
9900 	 * The control device is always connected.  The disk device, on the
9901 	 * other hand, may not be online all the time.  Need to change this
9902 	 * to figure out whether the disk device is actually online or not.
9903 	 */
9904 	if (lun != NULL)
9905 		sfs_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9906 				  lun->be_lun->lun_type;
9907 	else
9908 		sfs_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9909 
9910 	sfs_ptr->page_code = SVPD_SCSI_SFS;
9911 	n = 0;
9912 	/* Discovery 2016 */
9913 	scsi_ulto2b(0x0001, &sfs_ptr->codes[2 * n++]);
9914 	if (lun != NULL && lun->be_lun->lun_type == T_DIRECT) {
9915 		 /* SBC Base 2016 */
9916 		scsi_ulto2b(0x0101, &sfs_ptr->codes[2 * n++]);
9917 		 /* SBC Base 2010 */
9918 		scsi_ulto2b(0x0102, &sfs_ptr->codes[2 * n++]);
9919 		if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9920 			/* Basic Provisioning 2016 */
9921 			scsi_ulto2b(0x0103, &sfs_ptr->codes[2 * n++]);
9922 		}
9923 		/* Drive Maintenance 2016 */
9924 		//scsi_ulto2b(0x0104, &sfs_ptr->codes[2 * n++]);
9925 	}
9926 	scsi_ulto2b(4 + 2 * n, sfs_ptr->page_length);
9927 
9928 	ctl_set_success(ctsio);
9929 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9930 	ctsio->be_move_done = ctl_config_move_done;
9931 	ctl_datamove((union ctl_io *)ctsio);
9932 	return (CTL_RETVAL_COMPLETE);
9933 }
9934 
9935 static int
9936 ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
9937 {
9938 	struct ctl_lun *lun = CTL_LUN(ctsio);
9939 	struct scsi_vpd_block_limits *bl_ptr;
9940 	const char *val;
9941 	uint64_t ival;
9942 
9943 	ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
9944 	bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
9945 	ctsio->kern_sg_entries = 0;
9946 	ctsio->kern_rel_offset = 0;
9947 	ctsio->kern_sg_entries = 0;
9948 	ctsio->kern_data_len = min(sizeof(*bl_ptr), alloc_len);
9949 	ctsio->kern_total_len = ctsio->kern_data_len;
9950 
9951 	/*
9952 	 * The control device is always connected.  The disk device, on the
9953 	 * other hand, may not be online all the time.  Need to change this
9954 	 * to figure out whether the disk device is actually online or not.
9955 	 */
9956 	if (lun != NULL)
9957 		bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9958 				  lun->be_lun->lun_type;
9959 	else
9960 		bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9961 
9962 	bl_ptr->page_code = SVPD_BLOCK_LIMITS;
9963 	scsi_ulto2b(sizeof(*bl_ptr) - 4, bl_ptr->page_length);
9964 	bl_ptr->max_cmp_write_len = 0xff;
9965 	scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
9966 	if (lun != NULL) {
9967 		scsi_ulto4b(lun->be_lun->opttxferlen, bl_ptr->opt_txfer_len);
9968 		if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9969 			ival = 0xffffffff;
9970 			val = dnvlist_get_string(lun->be_lun->options,
9971 			    "unmap_max_lba", NULL);
9972 			if (val != NULL)
9973 				ctl_expand_number(val, &ival);
9974 			scsi_ulto4b(ival, bl_ptr->max_unmap_lba_cnt);
9975 			ival = 0xffffffff;
9976 			val = dnvlist_get_string(lun->be_lun->options,
9977 			    "unmap_max_descr", NULL);
9978 			if (val != NULL)
9979 				ctl_expand_number(val, &ival);
9980 			scsi_ulto4b(ival, bl_ptr->max_unmap_blk_cnt);
9981 			if (lun->be_lun->ublockexp != 0) {
9982 				scsi_ulto4b((1 << lun->be_lun->ublockexp),
9983 				    bl_ptr->opt_unmap_grain);
9984 				scsi_ulto4b(0x80000000 | lun->be_lun->ublockoff,
9985 				    bl_ptr->unmap_grain_align);
9986 			}
9987 		}
9988 		scsi_ulto4b(lun->be_lun->atomicblock,
9989 		    bl_ptr->max_atomic_transfer_length);
9990 		scsi_ulto4b(0, bl_ptr->atomic_alignment);
9991 		scsi_ulto4b(0, bl_ptr->atomic_transfer_length_granularity);
9992 		scsi_ulto4b(0, bl_ptr->max_atomic_transfer_length_with_atomic_boundary);
9993 		scsi_ulto4b(0, bl_ptr->max_atomic_boundary_size);
9994 		ival = UINT64_MAX;
9995 		val = dnvlist_get_string(lun->be_lun->options,
9996 		    "write_same_max_lba", NULL);
9997 		if (val != NULL)
9998 			ctl_expand_number(val, &ival);
9999 		scsi_u64to8b(ival, bl_ptr->max_write_same_length);
10000 		if (lun->be_lun->maxlba + 1 > ival)
10001 			bl_ptr->flags |= SVPD_BL_WSNZ;
10002 	}
10003 
10004 	ctl_set_success(ctsio);
10005 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10006 	ctsio->be_move_done = ctl_config_move_done;
10007 	ctl_datamove((union ctl_io *)ctsio);
10008 	return (CTL_RETVAL_COMPLETE);
10009 }
10010 
10011 static int
10012 ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len)
10013 {
10014 	struct ctl_lun *lun = CTL_LUN(ctsio);
10015 	struct scsi_vpd_block_device_characteristics *bdc_ptr;
10016 	const char *value;
10017 	u_int i;
10018 
10019 	ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO);
10020 	bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr;
10021 	ctsio->kern_sg_entries = 0;
10022 	ctsio->kern_rel_offset = 0;
10023 	ctsio->kern_data_len = min(sizeof(*bdc_ptr), alloc_len);
10024 	ctsio->kern_total_len = ctsio->kern_data_len;
10025 
10026 	/*
10027 	 * The control device is always connected.  The disk device, on the
10028 	 * other hand, may not be online all the time.  Need to change this
10029 	 * to figure out whether the disk device is actually online or not.
10030 	 */
10031 	if (lun != NULL)
10032 		bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10033 				  lun->be_lun->lun_type;
10034 	else
10035 		bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10036 	bdc_ptr->page_code = SVPD_BDC;
10037 	scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length);
10038 	if (lun != NULL &&
10039 	    (value = dnvlist_get_string(lun->be_lun->options, "rpm", NULL)) != NULL)
10040 		i = strtol(value, NULL, 0);
10041 	else
10042 		i = CTL_DEFAULT_ROTATION_RATE;
10043 	scsi_ulto2b(i, bdc_ptr->medium_rotation_rate);
10044 	if (lun != NULL &&
10045 	    (value = dnvlist_get_string(lun->be_lun->options, "formfactor", NULL)) != NULL)
10046 		i = strtol(value, NULL, 0);
10047 	else
10048 		i = 0;
10049 	bdc_ptr->wab_wac_ff = (i & 0x0f);
10050 	bdc_ptr->flags = SVPD_RBWZ | SVPD_FUAB | SVPD_VBULS;
10051 
10052 	ctl_set_success(ctsio);
10053 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10054 	ctsio->be_move_done = ctl_config_move_done;
10055 	ctl_datamove((union ctl_io *)ctsio);
10056 	return (CTL_RETVAL_COMPLETE);
10057 }
10058 
10059 static int
10060 ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
10061 {
10062 	struct ctl_lun *lun = CTL_LUN(ctsio);
10063 	struct scsi_vpd_logical_block_prov *lbp_ptr;
10064 	const char *value;
10065 
10066 	ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
10067 	lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
10068 	ctsio->kern_sg_entries = 0;
10069 	ctsio->kern_rel_offset = 0;
10070 	ctsio->kern_data_len = min(sizeof(*lbp_ptr), alloc_len);
10071 	ctsio->kern_total_len = ctsio->kern_data_len;
10072 
10073 	/*
10074 	 * The control device is always connected.  The disk device, on the
10075 	 * other hand, may not be online all the time.  Need to change this
10076 	 * to figure out whether the disk device is actually online or not.
10077 	 */
10078 	if (lun != NULL)
10079 		lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10080 				  lun->be_lun->lun_type;
10081 	else
10082 		lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10083 
10084 	lbp_ptr->page_code = SVPD_LBP;
10085 	scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length);
10086 	lbp_ptr->threshold_exponent = CTL_LBP_EXPONENT;
10087 	if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10088 		lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 |
10089 		    SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP;
10090 		value = dnvlist_get_string(lun->be_lun->options,
10091 		    "provisioning_type", NULL);
10092 		if (value != NULL) {
10093 			if (strcmp(value, "resource") == 0)
10094 				lbp_ptr->prov_type = SVPD_LBP_RESOURCE;
10095 			else if (strcmp(value, "thin") == 0)
10096 				lbp_ptr->prov_type = SVPD_LBP_THIN;
10097 		} else
10098 			lbp_ptr->prov_type = SVPD_LBP_THIN;
10099 	}
10100 
10101 	ctl_set_success(ctsio);
10102 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10103 	ctsio->be_move_done = ctl_config_move_done;
10104 	ctl_datamove((union ctl_io *)ctsio);
10105 	return (CTL_RETVAL_COMPLETE);
10106 }
10107 
10108 /*
10109  * INQUIRY with the EVPD bit set.
10110  */
10111 static int
10112 ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
10113 {
10114 	struct ctl_lun *lun = CTL_LUN(ctsio);
10115 	struct scsi_inquiry *cdb;
10116 	int alloc_len, retval;
10117 
10118 	cdb = (struct scsi_inquiry *)ctsio->cdb;
10119 	alloc_len = scsi_2btoul(cdb->length);
10120 
10121 	switch (cdb->page_code) {
10122 	case SVPD_SUPPORTED_PAGES:
10123 		retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
10124 		break;
10125 	case SVPD_UNIT_SERIAL_NUMBER:
10126 		retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
10127 		break;
10128 	case SVPD_DEVICE_ID:
10129 		retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
10130 		break;
10131 	case SVPD_EXTENDED_INQUIRY_DATA:
10132 		retval = ctl_inquiry_evpd_eid(ctsio, alloc_len);
10133 		break;
10134 	case SVPD_MODE_PAGE_POLICY:
10135 		retval = ctl_inquiry_evpd_mpp(ctsio, alloc_len);
10136 		break;
10137 	case SVPD_SCSI_PORTS:
10138 		retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
10139 		break;
10140 	case SVPD_SCSI_TPC:
10141 		retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
10142 		break;
10143 	case SVPD_SCSI_SFS:
10144 		retval = ctl_inquiry_evpd_sfs(ctsio, alloc_len);
10145 		break;
10146 	case SVPD_BLOCK_LIMITS:
10147 		if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10148 			goto err;
10149 		retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
10150 		break;
10151 	case SVPD_BDC:
10152 		if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10153 			goto err;
10154 		retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len);
10155 		break;
10156 	case SVPD_LBP:
10157 		if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10158 			goto err;
10159 		retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
10160 		break;
10161 	default:
10162 err:
10163 		ctl_set_invalid_field(ctsio,
10164 				      /*sks_valid*/ 1,
10165 				      /*command*/ 1,
10166 				      /*field*/ 2,
10167 				      /*bit_valid*/ 0,
10168 				      /*bit*/ 0);
10169 		ctl_done((union ctl_io *)ctsio);
10170 		retval = CTL_RETVAL_COMPLETE;
10171 		break;
10172 	}
10173 
10174 	return (retval);
10175 }
10176 
10177 /*
10178  * Standard INQUIRY data.
10179  */
10180 static int
10181 ctl_inquiry_std(struct ctl_scsiio *ctsio)
10182 {
10183 	struct ctl_softc *softc = CTL_SOFTC(ctsio);
10184 	struct ctl_port *port = CTL_PORT(ctsio);
10185 	struct ctl_lun *lun = CTL_LUN(ctsio);
10186 	struct scsi_inquiry_data *inq_ptr;
10187 	struct scsi_inquiry *cdb;
10188 	const char *val;
10189 	uint32_t alloc_len, data_len;
10190 	ctl_port_type port_type;
10191 
10192 	port_type = port->port_type;
10193 	if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
10194 		port_type = CTL_PORT_SCSI;
10195 
10196 	cdb = (struct scsi_inquiry *)ctsio->cdb;
10197 	alloc_len = scsi_2btoul(cdb->length);
10198 
10199 	/*
10200 	 * We malloc the full inquiry data size here and fill it
10201 	 * in.  If the user only asks for less, we'll give him
10202 	 * that much.
10203 	 */
10204 	data_len = offsetof(struct scsi_inquiry_data, vendor_specific1);
10205 	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10206 	inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
10207 	ctsio->kern_sg_entries = 0;
10208 	ctsio->kern_rel_offset = 0;
10209 	ctsio->kern_data_len = min(data_len, alloc_len);
10210 	ctsio->kern_total_len = ctsio->kern_data_len;
10211 
10212 	if (lun != NULL) {
10213 		if ((lun->flags & CTL_LUN_PRIMARY_SC) ||
10214 		    softc->ha_link >= CTL_HA_LINK_UNKNOWN) {
10215 			inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10216 			    lun->be_lun->lun_type;
10217 		} else {
10218 			inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) |
10219 			    lun->be_lun->lun_type;
10220 		}
10221 		if (lun->flags & CTL_LUN_REMOVABLE)
10222 			inq_ptr->dev_qual2 |= SID_RMB;
10223 	} else
10224 		inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10225 
10226 	/* RMB in byte 2 is 0 */
10227 	inq_ptr->version = SCSI_REV_SPC5;
10228 
10229 	/*
10230 	 * According to SAM-3, even if a device only supports a single
10231 	 * level of LUN addressing, it should still set the HISUP bit:
10232 	 *
10233 	 * 4.9.1 Logical unit numbers overview
10234 	 *
10235 	 * All logical unit number formats described in this standard are
10236 	 * hierarchical in structure even when only a single level in that
10237 	 * hierarchy is used. The HISUP bit shall be set to one in the
10238 	 * standard INQUIRY data (see SPC-2) when any logical unit number
10239 	 * format described in this standard is used.  Non-hierarchical
10240 	 * formats are outside the scope of this standard.
10241 	 *
10242 	 * Therefore we set the HiSup bit here.
10243 	 *
10244 	 * The response format is 2, per SPC-3.
10245 	 */
10246 	inq_ptr->response_format = SID_HiSup | 2;
10247 
10248 	inq_ptr->additional_length = data_len -
10249 	    (offsetof(struct scsi_inquiry_data, additional_length) + 1);
10250 	CTL_DEBUG_PRINT(("additional_length = %d\n",
10251 			 inq_ptr->additional_length));
10252 
10253 	inq_ptr->spc3_flags = SPC3_SID_3PC | SPC3_SID_TPGS_IMPLICIT;
10254 	if (port_type == CTL_PORT_SCSI)
10255 		inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10256 	inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10257 	inq_ptr->flags = SID_CmdQue;
10258 	if (port_type == CTL_PORT_SCSI)
10259 		inq_ptr->flags |= SID_WBus16 | SID_Sync;
10260 
10261 	/*
10262 	 * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10263 	 * We have 8 bytes for the vendor name, and 16 bytes for the device
10264 	 * name and 4 bytes for the revision.
10265 	 */
10266 	if (lun == NULL || (val = dnvlist_get_string(lun->be_lun->options,
10267 	    "vendor", NULL)) == NULL) {
10268 		strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
10269 	} else {
10270 		memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10271 		strncpy(inq_ptr->vendor, val,
10272 		    min(sizeof(inq_ptr->vendor), strlen(val)));
10273 	}
10274 	if (lun == NULL) {
10275 		strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10276 		    sizeof(inq_ptr->product));
10277 	} else if ((val = dnvlist_get_string(lun->be_lun->options, "product",
10278 	    NULL)) == NULL) {
10279 		switch (lun->be_lun->lun_type) {
10280 		case T_DIRECT:
10281 			strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10282 			    sizeof(inq_ptr->product));
10283 			break;
10284 		case T_PROCESSOR:
10285 			strncpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT,
10286 			    sizeof(inq_ptr->product));
10287 			break;
10288 		case T_CDROM:
10289 			strncpy(inq_ptr->product, CTL_CDROM_PRODUCT,
10290 			    sizeof(inq_ptr->product));
10291 			break;
10292 		default:
10293 			strncpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT,
10294 			    sizeof(inq_ptr->product));
10295 			break;
10296 		}
10297 	} else {
10298 		memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10299 		strncpy(inq_ptr->product, val,
10300 		    min(sizeof(inq_ptr->product), strlen(val)));
10301 	}
10302 
10303 	/*
10304 	 * XXX make this a macro somewhere so it automatically gets
10305 	 * incremented when we make changes.
10306 	 */
10307 	if (lun == NULL || (val = dnvlist_get_string(lun->be_lun->options,
10308 	    "revision", NULL)) == NULL) {
10309 		strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10310 	} else {
10311 		memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10312 		strncpy(inq_ptr->revision, val,
10313 		    min(sizeof(inq_ptr->revision), strlen(val)));
10314 	}
10315 
10316 	/*
10317 	 * For parallel SCSI, we support double transition and single
10318 	 * transition clocking.  We also support QAS (Quick Arbitration
10319 	 * and Selection) and Information Unit transfers on both the
10320 	 * control and array devices.
10321 	 */
10322 	if (port_type == CTL_PORT_SCSI)
10323 		inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10324 				    SID_SPI_IUS;
10325 
10326 	/* SAM-6 (no version claimed) */
10327 	scsi_ulto2b(0x00C0, inq_ptr->version1);
10328 	/* SPC-5 (no version claimed) */
10329 	scsi_ulto2b(0x05C0, inq_ptr->version2);
10330 	if (port_type == CTL_PORT_FC) {
10331 		/* FCP-2 ANSI INCITS.350:2003 */
10332 		scsi_ulto2b(0x0917, inq_ptr->version3);
10333 	} else if (port_type == CTL_PORT_SCSI) {
10334 		/* SPI-4 ANSI INCITS.362:200x */
10335 		scsi_ulto2b(0x0B56, inq_ptr->version3);
10336 	} else if (port_type == CTL_PORT_ISCSI) {
10337 		/* iSCSI (no version claimed) */
10338 		scsi_ulto2b(0x0960, inq_ptr->version3);
10339 	} else if (port_type == CTL_PORT_SAS) {
10340 		/* SAS (no version claimed) */
10341 		scsi_ulto2b(0x0BE0, inq_ptr->version3);
10342 	} else if (port_type == CTL_PORT_UMASS) {
10343 		/* USB Mass Storage Class Bulk-Only Transport, Revision 1.0 */
10344 		scsi_ulto2b(0x1730, inq_ptr->version3);
10345 	}
10346 
10347 	if (lun == NULL) {
10348 		/* SBC-4 (no version claimed) */
10349 		scsi_ulto2b(0x0600, inq_ptr->version4);
10350 	} else {
10351 		switch (lun->be_lun->lun_type) {
10352 		case T_DIRECT:
10353 			/* SBC-4 (no version claimed) */
10354 			scsi_ulto2b(0x0600, inq_ptr->version4);
10355 			break;
10356 		case T_PROCESSOR:
10357 			break;
10358 		case T_CDROM:
10359 			/* MMC-6 (no version claimed) */
10360 			scsi_ulto2b(0x04E0, inq_ptr->version4);
10361 			break;
10362 		default:
10363 			break;
10364 		}
10365 	}
10366 
10367 	ctl_set_success(ctsio);
10368 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10369 	ctsio->be_move_done = ctl_config_move_done;
10370 	ctl_datamove((union ctl_io *)ctsio);
10371 	return (CTL_RETVAL_COMPLETE);
10372 }
10373 
10374 int
10375 ctl_inquiry(struct ctl_scsiio *ctsio)
10376 {
10377 	struct scsi_inquiry *cdb;
10378 	int retval;
10379 
10380 	CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10381 
10382 	cdb = (struct scsi_inquiry *)ctsio->cdb;
10383 	if (cdb->byte2 & SI_EVPD)
10384 		retval = ctl_inquiry_evpd(ctsio);
10385 	else if (cdb->page_code == 0)
10386 		retval = ctl_inquiry_std(ctsio);
10387 	else {
10388 		ctl_set_invalid_field(ctsio,
10389 				      /*sks_valid*/ 1,
10390 				      /*command*/ 1,
10391 				      /*field*/ 2,
10392 				      /*bit_valid*/ 0,
10393 				      /*bit*/ 0);
10394 		ctl_done((union ctl_io *)ctsio);
10395 		return (CTL_RETVAL_COMPLETE);
10396 	}
10397 
10398 	return (retval);
10399 }
10400 
10401 int
10402 ctl_get_config(struct ctl_scsiio *ctsio)
10403 {
10404 	struct ctl_lun *lun = CTL_LUN(ctsio);
10405 	struct scsi_get_config_header *hdr;
10406 	struct scsi_get_config_feature *feature;
10407 	struct scsi_get_config *cdb;
10408 	uint32_t alloc_len, data_len;
10409 	int rt, starting;
10410 
10411 	cdb = (struct scsi_get_config *)ctsio->cdb;
10412 	rt = (cdb->rt & SGC_RT_MASK);
10413 	starting = scsi_2btoul(cdb->starting_feature);
10414 	alloc_len = scsi_2btoul(cdb->length);
10415 
10416 	data_len = sizeof(struct scsi_get_config_header) +
10417 	    sizeof(struct scsi_get_config_feature) + 8 +
10418 	    sizeof(struct scsi_get_config_feature) + 8 +
10419 	    sizeof(struct scsi_get_config_feature) + 4 +
10420 	    sizeof(struct scsi_get_config_feature) + 4 +
10421 	    sizeof(struct scsi_get_config_feature) + 8 +
10422 	    sizeof(struct scsi_get_config_feature) +
10423 	    sizeof(struct scsi_get_config_feature) + 4 +
10424 	    sizeof(struct scsi_get_config_feature) + 4 +
10425 	    sizeof(struct scsi_get_config_feature) + 4 +
10426 	    sizeof(struct scsi_get_config_feature) + 4 +
10427 	    sizeof(struct scsi_get_config_feature) + 4 +
10428 	    sizeof(struct scsi_get_config_feature) + 4;
10429 	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10430 	ctsio->kern_sg_entries = 0;
10431 	ctsio->kern_rel_offset = 0;
10432 
10433 	hdr = (struct scsi_get_config_header *)ctsio->kern_data_ptr;
10434 	if (lun->flags & CTL_LUN_NO_MEDIA)
10435 		scsi_ulto2b(0x0000, hdr->current_profile);
10436 	else
10437 		scsi_ulto2b(0x0010, hdr->current_profile);
10438 	feature = (struct scsi_get_config_feature *)(hdr + 1);
10439 
10440 	if (starting > 0x003b)
10441 		goto done;
10442 	if (starting > 0x003a)
10443 		goto f3b;
10444 	if (starting > 0x002b)
10445 		goto f3a;
10446 	if (starting > 0x002a)
10447 		goto f2b;
10448 	if (starting > 0x001f)
10449 		goto f2a;
10450 	if (starting > 0x001e)
10451 		goto f1f;
10452 	if (starting > 0x001d)
10453 		goto f1e;
10454 	if (starting > 0x0010)
10455 		goto f1d;
10456 	if (starting > 0x0003)
10457 		goto f10;
10458 	if (starting > 0x0002)
10459 		goto f3;
10460 	if (starting > 0x0001)
10461 		goto f2;
10462 	if (starting > 0x0000)
10463 		goto f1;
10464 
10465 	/* Profile List */
10466 	scsi_ulto2b(0x0000, feature->feature_code);
10467 	feature->flags = SGC_F_PERSISTENT | SGC_F_CURRENT;
10468 	feature->add_length = 8;
10469 	scsi_ulto2b(0x0008, &feature->feature_data[0]);	/* CD-ROM */
10470 	feature->feature_data[2] = 0x00;
10471 	scsi_ulto2b(0x0010, &feature->feature_data[4]);	/* DVD-ROM */
10472 	feature->feature_data[6] = 0x01;
10473 	feature = (struct scsi_get_config_feature *)
10474 	    &feature->feature_data[feature->add_length];
10475 
10476 f1:	/* Core */
10477 	scsi_ulto2b(0x0001, feature->feature_code);
10478 	feature->flags = 0x08 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10479 	feature->add_length = 8;
10480 	scsi_ulto4b(0x00000000, &feature->feature_data[0]);
10481 	feature->feature_data[4] = 0x03;
10482 	feature = (struct scsi_get_config_feature *)
10483 	    &feature->feature_data[feature->add_length];
10484 
10485 f2:	/* Morphing */
10486 	scsi_ulto2b(0x0002, feature->feature_code);
10487 	feature->flags = 0x04 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10488 	feature->add_length = 4;
10489 	feature->feature_data[0] = 0x02;
10490 	feature = (struct scsi_get_config_feature *)
10491 	    &feature->feature_data[feature->add_length];
10492 
10493 f3:	/* Removable Medium */
10494 	scsi_ulto2b(0x0003, feature->feature_code);
10495 	feature->flags = 0x04 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10496 	feature->add_length = 4;
10497 	feature->feature_data[0] = 0x39;
10498 	feature = (struct scsi_get_config_feature *)
10499 	    &feature->feature_data[feature->add_length];
10500 
10501 	if (rt == SGC_RT_CURRENT && (lun->flags & CTL_LUN_NO_MEDIA))
10502 		goto done;
10503 
10504 f10:	/* Random Read */
10505 	scsi_ulto2b(0x0010, feature->feature_code);
10506 	feature->flags = 0x00;
10507 	if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10508 		feature->flags |= SGC_F_CURRENT;
10509 	feature->add_length = 8;
10510 	scsi_ulto4b(lun->be_lun->blocksize, &feature->feature_data[0]);
10511 	scsi_ulto2b(1, &feature->feature_data[4]);
10512 	feature->feature_data[6] = 0x00;
10513 	feature = (struct scsi_get_config_feature *)
10514 	    &feature->feature_data[feature->add_length];
10515 
10516 f1d:	/* Multi-Read */
10517 	scsi_ulto2b(0x001D, feature->feature_code);
10518 	feature->flags = 0x00;
10519 	if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10520 		feature->flags |= SGC_F_CURRENT;
10521 	feature->add_length = 0;
10522 	feature = (struct scsi_get_config_feature *)
10523 	    &feature->feature_data[feature->add_length];
10524 
10525 f1e:	/* CD Read */
10526 	scsi_ulto2b(0x001E, feature->feature_code);
10527 	feature->flags = 0x00;
10528 	if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10529 		feature->flags |= SGC_F_CURRENT;
10530 	feature->add_length = 4;
10531 	feature->feature_data[0] = 0x00;
10532 	feature = (struct scsi_get_config_feature *)
10533 	    &feature->feature_data[feature->add_length];
10534 
10535 f1f:	/* DVD Read */
10536 	scsi_ulto2b(0x001F, feature->feature_code);
10537 	feature->flags = 0x08;
10538 	if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10539 		feature->flags |= SGC_F_CURRENT;
10540 	feature->add_length = 4;
10541 	feature->feature_data[0] = 0x01;
10542 	feature->feature_data[2] = 0x03;
10543 	feature = (struct scsi_get_config_feature *)
10544 	    &feature->feature_data[feature->add_length];
10545 
10546 f2a:	/* DVD+RW */
10547 	scsi_ulto2b(0x002A, feature->feature_code);
10548 	feature->flags = 0x04;
10549 	if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10550 		feature->flags |= SGC_F_CURRENT;
10551 	feature->add_length = 4;
10552 	feature->feature_data[0] = 0x00;
10553 	feature->feature_data[1] = 0x00;
10554 	feature = (struct scsi_get_config_feature *)
10555 	    &feature->feature_data[feature->add_length];
10556 
10557 f2b:	/* DVD+R */
10558 	scsi_ulto2b(0x002B, feature->feature_code);
10559 	feature->flags = 0x00;
10560 	if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10561 		feature->flags |= SGC_F_CURRENT;
10562 	feature->add_length = 4;
10563 	feature->feature_data[0] = 0x00;
10564 	feature = (struct scsi_get_config_feature *)
10565 	    &feature->feature_data[feature->add_length];
10566 
10567 f3a:	/* DVD+RW Dual Layer */
10568 	scsi_ulto2b(0x003A, feature->feature_code);
10569 	feature->flags = 0x00;
10570 	if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10571 		feature->flags |= SGC_F_CURRENT;
10572 	feature->add_length = 4;
10573 	feature->feature_data[0] = 0x00;
10574 	feature->feature_data[1] = 0x00;
10575 	feature = (struct scsi_get_config_feature *)
10576 	    &feature->feature_data[feature->add_length];
10577 
10578 f3b:	/* DVD+R Dual Layer */
10579 	scsi_ulto2b(0x003B, feature->feature_code);
10580 	feature->flags = 0x00;
10581 	if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10582 		feature->flags |= SGC_F_CURRENT;
10583 	feature->add_length = 4;
10584 	feature->feature_data[0] = 0x00;
10585 	feature = (struct scsi_get_config_feature *)
10586 	    &feature->feature_data[feature->add_length];
10587 
10588 done:
10589 	data_len = (uint8_t *)feature - (uint8_t *)hdr;
10590 	if (rt == SGC_RT_SPECIFIC && data_len > 4) {
10591 		feature = (struct scsi_get_config_feature *)(hdr + 1);
10592 		if (scsi_2btoul(feature->feature_code) == starting)
10593 			feature = (struct scsi_get_config_feature *)
10594 			    &feature->feature_data[feature->add_length];
10595 		data_len = (uint8_t *)feature - (uint8_t *)hdr;
10596 	}
10597 	scsi_ulto4b(data_len - 4, hdr->data_length);
10598 	ctsio->kern_data_len = min(data_len, alloc_len);
10599 	ctsio->kern_total_len = ctsio->kern_data_len;
10600 
10601 	ctl_set_success(ctsio);
10602 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10603 	ctsio->be_move_done = ctl_config_move_done;
10604 	ctl_datamove((union ctl_io *)ctsio);
10605 	return (CTL_RETVAL_COMPLETE);
10606 }
10607 
10608 int
10609 ctl_get_event_status(struct ctl_scsiio *ctsio)
10610 {
10611 	struct scsi_get_event_status_header *hdr;
10612 	struct scsi_get_event_status *cdb;
10613 	uint32_t alloc_len, data_len;
10614 
10615 	cdb = (struct scsi_get_event_status *)ctsio->cdb;
10616 	if ((cdb->byte2 & SGESN_POLLED) == 0) {
10617 		ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 1,
10618 		    /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
10619 		ctl_done((union ctl_io *)ctsio);
10620 		return (CTL_RETVAL_COMPLETE);
10621 	}
10622 	alloc_len = scsi_2btoul(cdb->length);
10623 
10624 	data_len = sizeof(struct scsi_get_event_status_header);
10625 	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10626 	ctsio->kern_sg_entries = 0;
10627 	ctsio->kern_rel_offset = 0;
10628 	ctsio->kern_data_len = min(data_len, alloc_len);
10629 	ctsio->kern_total_len = ctsio->kern_data_len;
10630 
10631 	hdr = (struct scsi_get_event_status_header *)ctsio->kern_data_ptr;
10632 	scsi_ulto2b(0, hdr->descr_length);
10633 	hdr->nea_class = SGESN_NEA;
10634 	hdr->supported_class = 0;
10635 
10636 	ctl_set_success(ctsio);
10637 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10638 	ctsio->be_move_done = ctl_config_move_done;
10639 	ctl_datamove((union ctl_io *)ctsio);
10640 	return (CTL_RETVAL_COMPLETE);
10641 }
10642 
10643 int
10644 ctl_mechanism_status(struct ctl_scsiio *ctsio)
10645 {
10646 	struct scsi_mechanism_status_header *hdr;
10647 	struct scsi_mechanism_status *cdb;
10648 	uint32_t alloc_len, data_len;
10649 
10650 	cdb = (struct scsi_mechanism_status *)ctsio->cdb;
10651 	alloc_len = scsi_2btoul(cdb->length);
10652 
10653 	data_len = sizeof(struct scsi_mechanism_status_header);
10654 	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10655 	ctsio->kern_sg_entries = 0;
10656 	ctsio->kern_rel_offset = 0;
10657 	ctsio->kern_data_len = min(data_len, alloc_len);
10658 	ctsio->kern_total_len = ctsio->kern_data_len;
10659 
10660 	hdr = (struct scsi_mechanism_status_header *)ctsio->kern_data_ptr;
10661 	hdr->state1 = 0x00;
10662 	hdr->state2 = 0xe0;
10663 	scsi_ulto3b(0, hdr->lba);
10664 	hdr->slots_num = 0;
10665 	scsi_ulto2b(0, hdr->slots_length);
10666 
10667 	ctl_set_success(ctsio);
10668 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10669 	ctsio->be_move_done = ctl_config_move_done;
10670 	ctl_datamove((union ctl_io *)ctsio);
10671 	return (CTL_RETVAL_COMPLETE);
10672 }
10673 
10674 static void
10675 ctl_ultomsf(uint32_t lba, uint8_t *buf)
10676 {
10677 
10678 	lba += 150;
10679 	buf[0] = 0;
10680 	buf[1] = bin2bcd((lba / 75) / 60);
10681 	buf[2] = bin2bcd((lba / 75) % 60);
10682 	buf[3] = bin2bcd(lba % 75);
10683 }
10684 
10685 int
10686 ctl_read_toc(struct ctl_scsiio *ctsio)
10687 {
10688 	struct ctl_lun *lun = CTL_LUN(ctsio);
10689 	struct scsi_read_toc_hdr *hdr;
10690 	struct scsi_read_toc_type01_descr *descr;
10691 	struct scsi_read_toc *cdb;
10692 	uint32_t alloc_len, data_len;
10693 	int format, msf;
10694 
10695 	cdb = (struct scsi_read_toc *)ctsio->cdb;
10696 	msf = (cdb->byte2 & CD_MSF) != 0;
10697 	format = cdb->format;
10698 	alloc_len = scsi_2btoul(cdb->data_len);
10699 
10700 	data_len = sizeof(struct scsi_read_toc_hdr);
10701 	if (format == 0)
10702 		data_len += 2 * sizeof(struct scsi_read_toc_type01_descr);
10703 	else
10704 		data_len += sizeof(struct scsi_read_toc_type01_descr);
10705 	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10706 	ctsio->kern_sg_entries = 0;
10707 	ctsio->kern_rel_offset = 0;
10708 	ctsio->kern_data_len = min(data_len, alloc_len);
10709 	ctsio->kern_total_len = ctsio->kern_data_len;
10710 
10711 	hdr = (struct scsi_read_toc_hdr *)ctsio->kern_data_ptr;
10712 	if (format == 0) {
10713 		scsi_ulto2b(0x12, hdr->data_length);
10714 		hdr->first = 1;
10715 		hdr->last = 1;
10716 		descr = (struct scsi_read_toc_type01_descr *)(hdr + 1);
10717 		descr->addr_ctl = 0x14;
10718 		descr->track_number = 1;
10719 		if (msf)
10720 			ctl_ultomsf(0, descr->track_start);
10721 		else
10722 			scsi_ulto4b(0, descr->track_start);
10723 		descr++;
10724 		descr->addr_ctl = 0x14;
10725 		descr->track_number = 0xaa;
10726 		if (msf)
10727 			ctl_ultomsf(lun->be_lun->maxlba+1, descr->track_start);
10728 		else
10729 			scsi_ulto4b(lun->be_lun->maxlba+1, descr->track_start);
10730 	} else {
10731 		scsi_ulto2b(0x0a, hdr->data_length);
10732 		hdr->first = 1;
10733 		hdr->last = 1;
10734 		descr = (struct scsi_read_toc_type01_descr *)(hdr + 1);
10735 		descr->addr_ctl = 0x14;
10736 		descr->track_number = 1;
10737 		if (msf)
10738 			ctl_ultomsf(0, descr->track_start);
10739 		else
10740 			scsi_ulto4b(0, descr->track_start);
10741 	}
10742 
10743 	ctl_set_success(ctsio);
10744 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10745 	ctsio->be_move_done = ctl_config_move_done;
10746 	ctl_datamove((union ctl_io *)ctsio);
10747 	return (CTL_RETVAL_COMPLETE);
10748 }
10749 
10750 /*
10751  * For known CDB types, parse the LBA and length.
10752  */
10753 static int
10754 ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len)
10755 {
10756 
10757 	KASSERT(io->io_hdr.io_type == CTL_IO_SCSI,
10758 	    ("%s: unexpected I/O type %x", __func__, io->io_hdr.io_type));
10759 
10760 	switch (io->scsiio.cdb[0]) {
10761 	case COMPARE_AND_WRITE: {
10762 		struct scsi_compare_and_write *cdb;
10763 
10764 		cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10765 
10766 		*lba = scsi_8btou64(cdb->addr);
10767 		*len = cdb->length;
10768 		break;
10769 	}
10770 	case READ_6:
10771 	case WRITE_6: {
10772 		struct scsi_rw_6 *cdb;
10773 
10774 		cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10775 
10776 		*lba = scsi_3btoul(cdb->addr);
10777 		/* only 5 bits are valid in the most significant address byte */
10778 		*lba &= 0x1fffff;
10779 		*len = cdb->length;
10780 		break;
10781 	}
10782 	case READ_10:
10783 	case WRITE_10: {
10784 		struct scsi_rw_10 *cdb;
10785 
10786 		cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10787 
10788 		*lba = scsi_4btoul(cdb->addr);
10789 		*len = scsi_2btoul(cdb->length);
10790 		break;
10791 	}
10792 	case WRITE_VERIFY_10: {
10793 		struct scsi_write_verify_10 *cdb;
10794 
10795 		cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10796 
10797 		*lba = scsi_4btoul(cdb->addr);
10798 		*len = scsi_2btoul(cdb->length);
10799 		break;
10800 	}
10801 	case READ_12:
10802 	case WRITE_12: {
10803 		struct scsi_rw_12 *cdb;
10804 
10805 		cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10806 
10807 		*lba = scsi_4btoul(cdb->addr);
10808 		*len = scsi_4btoul(cdb->length);
10809 		break;
10810 	}
10811 	case WRITE_VERIFY_12: {
10812 		struct scsi_write_verify_12 *cdb;
10813 
10814 		cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10815 
10816 		*lba = scsi_4btoul(cdb->addr);
10817 		*len = scsi_4btoul(cdb->length);
10818 		break;
10819 	}
10820 	case READ_16:
10821 	case WRITE_16: {
10822 		struct scsi_rw_16 *cdb;
10823 
10824 		cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10825 
10826 		*lba = scsi_8btou64(cdb->addr);
10827 		*len = scsi_4btoul(cdb->length);
10828 		break;
10829 	}
10830 	case WRITE_ATOMIC_16: {
10831 		struct scsi_write_atomic_16 *cdb;
10832 
10833 		cdb = (struct scsi_write_atomic_16 *)io->scsiio.cdb;
10834 
10835 		*lba = scsi_8btou64(cdb->addr);
10836 		*len = scsi_2btoul(cdb->length);
10837 		break;
10838 	}
10839 	case WRITE_VERIFY_16: {
10840 		struct scsi_write_verify_16 *cdb;
10841 
10842 		cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10843 
10844 		*lba = scsi_8btou64(cdb->addr);
10845 		*len = scsi_4btoul(cdb->length);
10846 		break;
10847 	}
10848 	case WRITE_SAME_10: {
10849 		struct scsi_write_same_10 *cdb;
10850 
10851 		cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10852 
10853 		*lba = scsi_4btoul(cdb->addr);
10854 		*len = scsi_2btoul(cdb->length);
10855 		break;
10856 	}
10857 	case WRITE_SAME_16: {
10858 		struct scsi_write_same_16 *cdb;
10859 
10860 		cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10861 
10862 		*lba = scsi_8btou64(cdb->addr);
10863 		*len = scsi_4btoul(cdb->length);
10864 		break;
10865 	}
10866 	case VERIFY_10: {
10867 		struct scsi_verify_10 *cdb;
10868 
10869 		cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10870 
10871 		*lba = scsi_4btoul(cdb->addr);
10872 		*len = scsi_2btoul(cdb->length);
10873 		break;
10874 	}
10875 	case VERIFY_12: {
10876 		struct scsi_verify_12 *cdb;
10877 
10878 		cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10879 
10880 		*lba = scsi_4btoul(cdb->addr);
10881 		*len = scsi_4btoul(cdb->length);
10882 		break;
10883 	}
10884 	case VERIFY_16: {
10885 		struct scsi_verify_16 *cdb;
10886 
10887 		cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10888 
10889 		*lba = scsi_8btou64(cdb->addr);
10890 		*len = scsi_4btoul(cdb->length);
10891 		break;
10892 	}
10893 	case UNMAP: {
10894 		*lba = 0;
10895 		*len = UINT64_MAX;
10896 		break;
10897 	}
10898 	case SERVICE_ACTION_IN: {	/* GET LBA STATUS */
10899 		struct scsi_get_lba_status *cdb;
10900 
10901 		cdb = (struct scsi_get_lba_status *)io->scsiio.cdb;
10902 		*lba = scsi_8btou64(cdb->addr);
10903 		*len = UINT32_MAX;
10904 		break;
10905 	}
10906 	default:
10907 		*lba = 0;
10908 		*len = UINT64_MAX;
10909 		return (1);
10910 	}
10911 
10912 	return (0);
10913 }
10914 
10915 static ctl_action
10916 ctl_extent_check_lba(uint64_t lba1, uint64_t len1, uint64_t lba2, uint64_t len2,
10917     bool seq)
10918 {
10919 	uint64_t endlba1, endlba2;
10920 
10921 	endlba1 = lba1 + len1 - (seq ? 0 : 1);
10922 	endlba2 = lba2 + len2 - 1;
10923 
10924 	if ((endlba1 < lba2) || (endlba2 < lba1))
10925 		return (CTL_ACTION_PASS);
10926 	else
10927 		return (CTL_ACTION_BLOCK);
10928 }
10929 
10930 static int
10931 ctl_extent_check_unmap(union ctl_io *io, uint64_t lba2, uint64_t len2)
10932 {
10933 	struct ctl_ptr_len_flags *ptrlen;
10934 	struct scsi_unmap_desc *buf, *end, *range;
10935 	uint64_t lba;
10936 	uint32_t len;
10937 
10938 	KASSERT(io->io_hdr.io_type == CTL_IO_SCSI,
10939 	    ("%s: unexpected I/O type %x", __func__, io->io_hdr.io_type));
10940 
10941 	/* If not UNMAP -- go other way. */
10942 	if (io->scsiio.cdb[0] != UNMAP)
10943 		return (CTL_ACTION_SKIP);
10944 
10945 	/* If UNMAP without data -- block and wait for data. */
10946 	ptrlen = (struct ctl_ptr_len_flags *)
10947 	    &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
10948 	if ((io->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0 ||
10949 	    ptrlen->ptr == NULL)
10950 		return (CTL_ACTION_BLOCK);
10951 
10952 	/* UNMAP with data -- check for collision. */
10953 	buf = (struct scsi_unmap_desc *)ptrlen->ptr;
10954 	end = buf + ptrlen->len / sizeof(*buf);
10955 	for (range = buf; range < end; range++) {
10956 		lba = scsi_8btou64(range->lba);
10957 		len = scsi_4btoul(range->length);
10958 		if ((lba < lba2 + len2) && (lba + len > lba2))
10959 			return (CTL_ACTION_BLOCK);
10960 	}
10961 	return (CTL_ACTION_PASS);
10962 }
10963 
10964 static ctl_action
10965 ctl_extent_check(union ctl_io *io1, union ctl_io *io2, bool seq)
10966 {
10967 	uint64_t lba1, lba2;
10968 	uint64_t len1, len2;
10969 	int retval;
10970 
10971 	retval = ctl_get_lba_len(io2, &lba2, &len2);
10972 	KASSERT(retval == 0, ("ctl_get_lba_len() error"));
10973 
10974 	retval = ctl_extent_check_unmap(io1, lba2, len2);
10975 	if (retval != CTL_ACTION_SKIP)
10976 		return (retval);
10977 
10978 	retval = ctl_get_lba_len(io1, &lba1, &len1);
10979 	KASSERT(retval == 0, ("ctl_get_lba_len() error"));
10980 
10981 	if (seq && (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE))
10982 		seq = FALSE;
10983 	return (ctl_extent_check_lba(lba1, len1, lba2, len2, seq));
10984 }
10985 
10986 static ctl_action
10987 ctl_seq_check(union ctl_io *io1, union ctl_io *io2)
10988 {
10989 	uint64_t lba1, lba2;
10990 	uint64_t len1, len2;
10991 	int retval __diagused;
10992 
10993 	if (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE)
10994 		return (CTL_ACTION_PASS);
10995 	retval = ctl_get_lba_len(io1, &lba1, &len1);
10996 	KASSERT(retval == 0, ("ctl_get_lba_len() error"));
10997 	retval = ctl_get_lba_len(io2, &lba2, &len2);
10998 	KASSERT(retval == 0, ("ctl_get_lba_len() error"));
10999 
11000 	if (lba1 + len1 == lba2)
11001 		return (CTL_ACTION_BLOCK);
11002 	return (CTL_ACTION_PASS);
11003 }
11004 
11005 static ctl_action
11006 ctl_check_for_blockage(struct ctl_lun *lun, union ctl_io *pending_io,
11007     const uint8_t *serialize_row, union ctl_io *ooa_io)
11008 {
11009 
11010 	/*
11011 	 * The initiator attempted multiple untagged commands at the same
11012 	 * time.  Can't do that.
11013 	 */
11014 	if (__predict_false(pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
11015 	 && __predict_false(ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
11016 	 && ((pending_io->io_hdr.nexus.targ_port ==
11017 	      ooa_io->io_hdr.nexus.targ_port)
11018 	  && (pending_io->io_hdr.nexus.initid ==
11019 	      ooa_io->io_hdr.nexus.initid))
11020 	 && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
11021 	      CTL_FLAG_STATUS_SENT)) == 0))
11022 		return (CTL_ACTION_OVERLAP);
11023 
11024 	/*
11025 	 * The initiator attempted to send multiple tagged commands with
11026 	 * the same ID.  (It's fine if different initiators have the same
11027 	 * tag ID.)
11028 	 *
11029 	 * Even if all of those conditions are true, we don't kill the I/O
11030 	 * if the command ahead of us has been aborted.  We won't end up
11031 	 * sending it to the FETD, and it's perfectly legal to resend a
11032 	 * command with the same tag number as long as the previous
11033 	 * instance of this tag number has been aborted somehow.
11034 	 */
11035 	if (__predict_true(pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
11036 	 && __predict_true(ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
11037 	 && __predict_false(pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
11038 	 && ((pending_io->io_hdr.nexus.targ_port ==
11039 	      ooa_io->io_hdr.nexus.targ_port)
11040 	  && (pending_io->io_hdr.nexus.initid ==
11041 	      ooa_io->io_hdr.nexus.initid))
11042 	 && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
11043 	      CTL_FLAG_STATUS_SENT)) == 0))
11044 		return (CTL_ACTION_OVERLAP_TAG);
11045 
11046 	/*
11047 	 * If we get a head of queue tag, SAM-3 says that we should
11048 	 * immediately execute it.
11049 	 *
11050 	 * What happens if this command would normally block for some other
11051 	 * reason?  e.g. a request sense with a head of queue tag
11052 	 * immediately after a write.  Normally that would block, but this
11053 	 * will result in its getting executed immediately...
11054 	 *
11055 	 * We currently return "pass" instead of "skip", so we'll end up
11056 	 * going through the rest of the queue to check for overlapped tags.
11057 	 *
11058 	 * XXX KDM check for other types of blockage first??
11059 	 */
11060 	if (__predict_false(pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE))
11061 		return (CTL_ACTION_PASS);
11062 
11063 	/*
11064 	 * Simple tags get blocked until all head of queue and ordered tags
11065 	 * ahead of them have completed.  I'm lumping untagged commands in
11066 	 * with simple tags here.  XXX KDM is that the right thing to do?
11067 	 */
11068 	if (__predict_false(ooa_io->scsiio.tag_type == CTL_TAG_ORDERED) ||
11069 	    __predict_false(ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE))
11070 		return (CTL_ACTION_BLOCK);
11071 
11072 	/* Unsupported command in OOA queue. */
11073 	if (__predict_false(ooa_io->scsiio.seridx == CTL_SERIDX_INVLD))
11074 		return (CTL_ACTION_PASS);
11075 
11076 	switch (serialize_row[ooa_io->scsiio.seridx]) {
11077 	case CTL_SER_SEQ:
11078 		if (lun->be_lun->serseq != CTL_LUN_SERSEQ_OFF)
11079 			return (ctl_seq_check(ooa_io, pending_io));
11080 		/* FALLTHROUGH */
11081 	case CTL_SER_PASS:
11082 		return (CTL_ACTION_PASS);
11083 	case CTL_SER_EXTENTOPT:
11084 		if ((lun->MODE_CTRL.queue_flags & SCP_QUEUE_ALG_MASK) ==
11085 		    SCP_QUEUE_ALG_UNRESTRICTED)
11086 			return (CTL_ACTION_PASS);
11087 		/* FALLTHROUGH */
11088 	case CTL_SER_EXTENT:
11089 		return (ctl_extent_check(ooa_io, pending_io,
11090 		    (lun->be_lun->serseq == CTL_LUN_SERSEQ_ON)));
11091 	case CTL_SER_BLOCKOPT:
11092 		if ((lun->MODE_CTRL.queue_flags & SCP_QUEUE_ALG_MASK) ==
11093 		    SCP_QUEUE_ALG_UNRESTRICTED)
11094 			return (CTL_ACTION_PASS);
11095 		/* FALLTHROUGH */
11096 	case CTL_SER_BLOCK:
11097 		return (CTL_ACTION_BLOCK);
11098 	default:
11099 		__assert_unreachable();
11100 	}
11101 }
11102 
11103 /*
11104  * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
11105  * Assumptions:
11106  * - pending_io is generally either incoming, or on the blocked queue
11107  * - starting I/O is the I/O we want to start the check with.
11108  */
11109 static ctl_action
11110 ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
11111 	      union ctl_io **starting_io)
11112 {
11113 	union ctl_io *ooa_io = *starting_io;
11114 	const uint8_t *serialize_row;
11115 	ctl_action action;
11116 
11117 	mtx_assert(&lun->lun_lock, MA_OWNED);
11118 
11119 	/*
11120 	 * Aborted commands are not going to be executed and may even
11121 	 * not report completion, so we don't care about their order.
11122 	 * Let them complete ASAP to clean the OOA queue.
11123 	 */
11124 	if (__predict_false(pending_io->io_hdr.flags & CTL_FLAG_ABORT))
11125 		return (CTL_ACTION_SKIP);
11126 
11127 	/*
11128 	 * Ordered tags have to block until all items ahead of them have
11129 	 * completed.  If we get called with an ordered tag, we always
11130 	 * block, if something else is ahead of us in the queue.
11131 	 */
11132 	if ((pending_io->scsiio.tag_type == CTL_TAG_ORDERED) &&
11133 	    (ooa_io != NULL))
11134 		return (CTL_ACTION_BLOCK);
11135 
11136 	serialize_row = ctl_serialize_table[pending_io->scsiio.seridx];
11137 
11138 	/*
11139 	 * Run back along the OOA queue, starting with the current
11140 	 * blocked I/O and going through every I/O before it on the
11141 	 * queue.  If starting_io is NULL, we'll just end up returning
11142 	 * CTL_ACTION_PASS.
11143 	 */
11144 	for (; ooa_io != NULL;
11145 	     ooa_io = (union ctl_io *)LIST_NEXT(&ooa_io->io_hdr, ooa_links)) {
11146 		action = ctl_check_for_blockage(lun, pending_io, serialize_row,
11147 		    ooa_io);
11148 		if (action != CTL_ACTION_PASS) {
11149 			*starting_io = ooa_io;
11150 			return (action);
11151 		}
11152 	}
11153 
11154 	*starting_io = NULL;
11155 	return (CTL_ACTION_PASS);
11156 }
11157 
11158 /*
11159  * Try to unblock the specified I/O.
11160  *
11161  * skip parameter allows explicitly skip present blocker of the I/O,
11162  * starting from the previous one on OOA queue.  It can be used when
11163  * we know for sure that the blocker I/O does no longer count.
11164  */
11165 static void
11166 ctl_try_unblock_io(struct ctl_lun *lun, union ctl_io *io, bool skip)
11167 {
11168 	struct ctl_softc *softc = lun->ctl_softc;
11169 	union ctl_io *bio, *obio;
11170 	const struct ctl_cmd_entry *entry;
11171 	union ctl_ha_msg msg_info;
11172 	ctl_action action;
11173 
11174 	mtx_assert(&lun->lun_lock, MA_OWNED);
11175 
11176 	if (io->io_hdr.blocker == NULL)
11177 		return;
11178 
11179 	obio = bio = io->io_hdr.blocker;
11180 	if (skip)
11181 		bio = (union ctl_io *)LIST_NEXT(&bio->io_hdr, ooa_links);
11182 	action = ctl_check_ooa(lun, io, &bio);
11183 	if (action == CTL_ACTION_BLOCK) {
11184 		/* Still blocked, but may be by different I/O now. */
11185 		if (bio != obio) {
11186 			TAILQ_REMOVE(&obio->io_hdr.blocked_queue,
11187 			    &io->io_hdr, blocked_links);
11188 			TAILQ_INSERT_TAIL(&bio->io_hdr.blocked_queue,
11189 			    &io->io_hdr, blocked_links);
11190 			io->io_hdr.blocker = bio;
11191 		}
11192 		return;
11193 	}
11194 
11195 	/* No longer blocked, one way or another. */
11196 	TAILQ_REMOVE(&obio->io_hdr.blocked_queue, &io->io_hdr, blocked_links);
11197 	io->io_hdr.blocker = NULL;
11198 
11199 	switch (action) {
11200 	case CTL_ACTION_PASS:
11201 	case CTL_ACTION_SKIP:
11202 
11203 		/* Serializing commands from the other SC retire there. */
11204 		if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) &&
11205 		    (softc->ha_mode != CTL_HA_MODE_XFER)) {
11206 			io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11207 			msg_info.hdr.original_sc = io->io_hdr.remote_io;
11208 			msg_info.hdr.serializing_sc = io;
11209 			msg_info.hdr.msg_type = CTL_MSG_R2R;
11210 			ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11211 			    sizeof(msg_info.hdr), M_NOWAIT);
11212 			break;
11213 		}
11214 
11215 		/*
11216 		 * Check this I/O for LUN state changes that may have happened
11217 		 * while this command was blocked. The LUN state may have been
11218 		 * changed by a command ahead of us in the queue.
11219 		 */
11220 		entry = ctl_get_cmd_entry(&io->scsiio, NULL);
11221 		if (ctl_scsiio_lun_check(lun, entry, &io->scsiio) != 0) {
11222 			ctl_done(io);
11223 			break;
11224 		}
11225 
11226 		io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11227 		ctl_enqueue_rtr(io);
11228 		break;
11229 	default:
11230 		__assert_unreachable();
11231 	case CTL_ACTION_OVERLAP:
11232 		ctl_set_overlapped_cmd(&io->scsiio);
11233 		goto error;
11234 	case CTL_ACTION_OVERLAP_TAG:
11235 		ctl_set_overlapped_tag(&io->scsiio,
11236 		    io->scsiio.tag_num & 0xff);
11237 error:
11238 		/* Serializing commands from the other SC are done here. */
11239 		if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) &&
11240 		    (softc->ha_mode != CTL_HA_MODE_XFER)) {
11241 			ctl_try_unblock_others(lun, io, TRUE);
11242 			LIST_REMOVE(&io->io_hdr, ooa_links);
11243 
11244 			ctl_copy_sense_data_back(io, &msg_info);
11245 			msg_info.hdr.original_sc = io->io_hdr.remote_io;
11246 			msg_info.hdr.serializing_sc = NULL;
11247 			msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
11248 			ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11249 			    sizeof(msg_info.scsi), M_WAITOK);
11250 			ctl_free_io(io);
11251 			break;
11252 		}
11253 
11254 		ctl_done(io);
11255 		break;
11256 	}
11257 }
11258 
11259 /*
11260  * Try to unblock I/Os blocked by the specified I/O.
11261  *
11262  * skip parameter allows explicitly skip the specified I/O as blocker,
11263  * starting from the previous one on the OOA queue.  It can be used when
11264  * we know for sure that the specified I/O does no longer count (done).
11265  * It has to be still on OOA queue though so that we know where to start.
11266  */
11267 static void
11268 ctl_try_unblock_others(struct ctl_lun *lun, union ctl_io *bio, bool skip)
11269 {
11270 	union ctl_io *io, *next_io;
11271 
11272 	mtx_assert(&lun->lun_lock, MA_OWNED);
11273 
11274 	for (io = (union ctl_io *)TAILQ_FIRST(&bio->io_hdr.blocked_queue);
11275 	     io != NULL; io = next_io) {
11276 		next_io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr, blocked_links);
11277 
11278 		KASSERT(io->io_hdr.blocker != NULL,
11279 		    ("I/O %p on blocked list without blocker", io));
11280 		ctl_try_unblock_io(lun, io, skip);
11281 	}
11282 	KASSERT(!skip || TAILQ_EMPTY(&bio->io_hdr.blocked_queue),
11283 	    ("blocked_queue is not empty after skipping %p", bio));
11284 }
11285 
11286 /*
11287  * This routine (with one exception) checks LUN flags that can be set by
11288  * commands ahead of us in the OOA queue.  These flags have to be checked
11289  * when a command initially comes in, and when we pull a command off the
11290  * blocked queue and are preparing to execute it.  The reason we have to
11291  * check these flags for commands on the blocked queue is that the LUN
11292  * state may have been changed by a command ahead of us while we're on the
11293  * blocked queue.
11294  *
11295  * Ordering is somewhat important with these checks, so please pay
11296  * careful attention to the placement of any new checks.
11297  */
11298 static int
11299 ctl_scsiio_lun_check(struct ctl_lun *lun,
11300     const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
11301 {
11302 	struct ctl_softc *softc = lun->ctl_softc;
11303 	int retval;
11304 	uint32_t residx;
11305 
11306 	retval = 0;
11307 
11308 	mtx_assert(&lun->lun_lock, MA_OWNED);
11309 
11310 	/*
11311 	 * If this shelf is a secondary shelf controller, we may have to
11312 	 * reject some commands disallowed by HA mode and link state.
11313 	 */
11314 	if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11315 		if (softc->ha_link == CTL_HA_LINK_OFFLINE &&
11316 		    (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
11317 			ctl_set_lun_unavail(ctsio);
11318 			retval = 1;
11319 			goto bailout;
11320 		}
11321 		if ((lun->flags & CTL_LUN_PEER_SC_PRIMARY) == 0 &&
11322 		    (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
11323 			ctl_set_lun_transit(ctsio);
11324 			retval = 1;
11325 			goto bailout;
11326 		}
11327 		if (softc->ha_mode == CTL_HA_MODE_ACT_STBY &&
11328 		    (entry->flags & CTL_CMD_FLAG_OK_ON_STANDBY) == 0) {
11329 			ctl_set_lun_standby(ctsio);
11330 			retval = 1;
11331 			goto bailout;
11332 		}
11333 
11334 		/* The rest of checks are only done on executing side */
11335 		if (softc->ha_mode == CTL_HA_MODE_XFER)
11336 			goto bailout;
11337 	}
11338 
11339 	if (entry->pattern & CTL_LUN_PAT_WRITE) {
11340 		if (lun->be_lun->flags & CTL_LUN_FLAG_READONLY) {
11341 			ctl_set_hw_write_protected(ctsio);
11342 			retval = 1;
11343 			goto bailout;
11344 		}
11345 		if ((lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0) {
11346 			ctl_set_sense(ctsio, /*current_error*/ 1,
11347 			    /*sense_key*/ SSD_KEY_DATA_PROTECT,
11348 			    /*asc*/ 0x27, /*ascq*/ 0x02, SSD_ELEM_NONE);
11349 			retval = 1;
11350 			goto bailout;
11351 		}
11352 	}
11353 
11354 	/*
11355 	 * Check for a reservation conflict.  If this command isn't allowed
11356 	 * even on reserved LUNs, and if this initiator isn't the one who
11357 	 * reserved us, reject the command with a reservation conflict.
11358 	 */
11359 	residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11360 	if ((lun->flags & CTL_LUN_RESERVED)
11361 	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
11362 		if (lun->res_idx != residx) {
11363 			ctl_set_reservation_conflict(ctsio);
11364 			retval = 1;
11365 			goto bailout;
11366 		}
11367 	}
11368 
11369 	if ((lun->flags & CTL_LUN_PR_RESERVED) == 0 ||
11370 	    (entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV)) {
11371 		/* No reservation or command is allowed. */;
11372 	} else if ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_WRESV) &&
11373 	    (lun->pr_res_type == SPR_TYPE_WR_EX ||
11374 	     lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
11375 	     lun->pr_res_type == SPR_TYPE_WR_EX_AR)) {
11376 		/* The command is allowed for Write Exclusive resv. */;
11377 	} else {
11378 		/*
11379 		 * if we aren't registered or it's a res holder type
11380 		 * reservation and this isn't the res holder then set a
11381 		 * conflict.
11382 		 */
11383 		if (ctl_get_prkey(lun, residx) == 0 ||
11384 		    (residx != lun->pr_res_idx && lun->pr_res_type < 4)) {
11385 			ctl_set_reservation_conflict(ctsio);
11386 			retval = 1;
11387 			goto bailout;
11388 		}
11389 	}
11390 
11391 	if ((entry->flags & CTL_CMD_FLAG_OK_ON_NO_MEDIA) == 0) {
11392 		if (lun->flags & CTL_LUN_EJECTED)
11393 			ctl_set_lun_ejected(ctsio);
11394 		else if (lun->flags & CTL_LUN_NO_MEDIA) {
11395 			if (lun->flags & CTL_LUN_REMOVABLE)
11396 				ctl_set_lun_no_media(ctsio);
11397 			else
11398 				ctl_set_lun_int_reqd(ctsio);
11399 		} else if (lun->flags & CTL_LUN_STOPPED)
11400 			ctl_set_lun_stopped(ctsio);
11401 		else
11402 			goto bailout;
11403 		retval = 1;
11404 		goto bailout;
11405 	}
11406 
11407 bailout:
11408 	return (retval);
11409 }
11410 
11411 static void
11412 ctl_failover_io(union ctl_io *io, int have_lock)
11413 {
11414 	ctl_set_busy(&io->scsiio);
11415 	ctl_done(io);
11416 }
11417 
11418 static void
11419 ctl_failover_lun(union ctl_io *rio)
11420 {
11421 	struct ctl_softc *softc = CTL_SOFTC(rio);
11422 	struct ctl_lun *lun;
11423 	struct ctl_io_hdr *io, *next_io;
11424 	uint32_t targ_lun;
11425 
11426 	targ_lun = rio->io_hdr.nexus.targ_mapped_lun;
11427 	CTL_DEBUG_PRINT(("FAILOVER for lun %u\n", targ_lun));
11428 
11429 	/* Find and lock the LUN. */
11430 	mtx_lock(&softc->ctl_lock);
11431 	if (targ_lun > ctl_max_luns ||
11432 	    (lun = softc->ctl_luns[targ_lun]) == NULL) {
11433 		mtx_unlock(&softc->ctl_lock);
11434 		return;
11435 	}
11436 	mtx_lock(&lun->lun_lock);
11437 	mtx_unlock(&softc->ctl_lock);
11438 	if (lun->flags & CTL_LUN_DISABLED) {
11439 		mtx_unlock(&lun->lun_lock);
11440 		return;
11441 	}
11442 
11443 	if (softc->ha_mode == CTL_HA_MODE_XFER) {
11444 		LIST_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
11445 			/* We are master */
11446 			if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11447 				if (io->flags & CTL_FLAG_IO_ACTIVE) {
11448 					io->flags |= CTL_FLAG_ABORT |
11449 					    CTL_FLAG_FAILOVER;
11450 					ctl_try_unblock_io(lun,
11451 					    (union ctl_io *)io, FALSE);
11452 				} else { /* This can be only due to DATAMOVE */
11453 					io->msg_type = CTL_MSG_DATAMOVE_DONE;
11454 					io->flags &= ~CTL_FLAG_DMA_INPROG;
11455 					io->flags |= CTL_FLAG_IO_ACTIVE;
11456 					io->port_status = 31340;
11457 					ctl_enqueue_isc((union ctl_io *)io);
11458 				}
11459 			} else
11460 			/* We are slave */
11461 			if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
11462 				io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11463 				if (io->flags & CTL_FLAG_IO_ACTIVE) {
11464 					io->flags |= CTL_FLAG_FAILOVER;
11465 				} else {
11466 					ctl_set_busy(&((union ctl_io *)io)->
11467 					    scsiio);
11468 					ctl_done((union ctl_io *)io);
11469 				}
11470 			}
11471 		}
11472 	} else { /* SERIALIZE modes */
11473 		LIST_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
11474 			/* We are master */
11475 			if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11476 				if (io->blocker != NULL) {
11477 					TAILQ_REMOVE(&io->blocker->io_hdr.blocked_queue,
11478 					    io, blocked_links);
11479 					io->blocker = NULL;
11480 				}
11481 				ctl_try_unblock_others(lun, (union ctl_io *)io,
11482 				    TRUE);
11483 				LIST_REMOVE(io, ooa_links);
11484 				ctl_free_io((union ctl_io *)io);
11485 			} else
11486 			/* We are slave */
11487 			if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
11488 				io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11489 				if (!(io->flags & CTL_FLAG_IO_ACTIVE)) {
11490 					ctl_set_busy(&((union ctl_io *)io)->
11491 					    scsiio);
11492 					ctl_done((union ctl_io *)io);
11493 				}
11494 			}
11495 		}
11496 	}
11497 	mtx_unlock(&lun->lun_lock);
11498 }
11499 
11500 static void
11501 ctl_scsiio_precheck(struct ctl_scsiio *ctsio)
11502 {
11503 	struct ctl_softc *softc = CTL_SOFTC(ctsio);
11504 	struct ctl_lun *lun;
11505 	const struct ctl_cmd_entry *entry;
11506 	union ctl_io *bio;
11507 	uint32_t initidx, targ_lun;
11508 
11509 	lun = NULL;
11510 	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11511 	if (targ_lun < ctl_max_luns)
11512 		lun = softc->ctl_luns[targ_lun];
11513 	if (lun) {
11514 		/*
11515 		 * If the LUN is invalid, pretend that it doesn't exist.
11516 		 * It will go away as soon as all pending I/O has been
11517 		 * completed.
11518 		 */
11519 		mtx_lock(&lun->lun_lock);
11520 		if (lun->flags & CTL_LUN_DISABLED) {
11521 			mtx_unlock(&lun->lun_lock);
11522 			lun = NULL;
11523 		}
11524 	}
11525 	CTL_LUN(ctsio) = lun;
11526 	if (lun) {
11527 		CTL_BACKEND_LUN(ctsio) = lun->be_lun;
11528 
11529 		/*
11530 		 * Every I/O goes into the OOA queue for a particular LUN,
11531 		 * and stays there until completion.
11532 		 */
11533 #ifdef CTL_TIME_IO
11534 		if (LIST_EMPTY(&lun->ooa_queue))
11535 			lun->idle_time += getsbinuptime() - lun->last_busy;
11536 #endif
11537 		LIST_INSERT_HEAD(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
11538 	}
11539 
11540 	/* Get command entry and return error if it is unsuppotyed. */
11541 	entry = ctl_validate_command(ctsio);
11542 	if (entry == NULL) {
11543 		if (lun)
11544 			mtx_unlock(&lun->lun_lock);
11545 		return;
11546 	}
11547 
11548 	ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11549 	ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11550 
11551 	/*
11552 	 * Check to see whether we can send this command to LUNs that don't
11553 	 * exist.  This should pretty much only be the case for inquiry
11554 	 * and request sense.  Further checks, below, really require having
11555 	 * a LUN, so we can't really check the command anymore.  Just put
11556 	 * it on the rtr queue.
11557 	 */
11558 	if (lun == NULL) {
11559 		if (entry->flags & CTL_CMD_FLAG_OK_ON_NO_LUN) {
11560 			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11561 			ctl_enqueue_rtr((union ctl_io *)ctsio);
11562 			return;
11563 		}
11564 
11565 		ctl_set_unsupported_lun(ctsio);
11566 		ctl_done((union ctl_io *)ctsio);
11567 		CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11568 		return;
11569 	} else {
11570 		/*
11571 		 * Make sure we support this particular command on this LUN.
11572 		 * e.g., we don't support writes to the control LUN.
11573 		 */
11574 		if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11575 			mtx_unlock(&lun->lun_lock);
11576 			ctl_set_invalid_opcode(ctsio);
11577 			ctl_done((union ctl_io *)ctsio);
11578 			return;
11579 		}
11580 	}
11581 
11582 	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11583 
11584 	/*
11585 	 * If we've got a request sense, it'll clear the contingent
11586 	 * allegiance condition.  Otherwise, if we have a CA condition for
11587 	 * this initiator, clear it, because it sent down a command other
11588 	 * than request sense.
11589 	 */
11590 	if (ctsio->cdb[0] != REQUEST_SENSE) {
11591 		struct scsi_sense_data *ps;
11592 
11593 		ps = lun->pending_sense[initidx / CTL_MAX_INIT_PER_PORT];
11594 		if (ps != NULL)
11595 			ps[initidx % CTL_MAX_INIT_PER_PORT].error_code = 0;
11596 	}
11597 
11598 	/*
11599 	 * If the command has this flag set, it handles its own unit
11600 	 * attention reporting, we shouldn't do anything.  Otherwise we
11601 	 * check for any pending unit attentions, and send them back to the
11602 	 * initiator.  We only do this when a command initially comes in,
11603 	 * not when we pull it off the blocked queue.
11604 	 *
11605 	 * According to SAM-3, section 5.3.2, the order that things get
11606 	 * presented back to the host is basically unit attentions caused
11607 	 * by some sort of reset event, busy status, reservation conflicts
11608 	 * or task set full, and finally any other status.
11609 	 *
11610 	 * One issue here is that some of the unit attentions we report
11611 	 * don't fall into the "reset" category (e.g. "reported luns data
11612 	 * has changed").  So reporting it here, before the reservation
11613 	 * check, may be technically wrong.  I guess the only thing to do
11614 	 * would be to check for and report the reset events here, and then
11615 	 * check for the other unit attention types after we check for a
11616 	 * reservation conflict.
11617 	 *
11618 	 * XXX KDM need to fix this
11619 	 */
11620 	if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11621 		ctl_ua_type ua_type;
11622 		u_int sense_len = 0;
11623 
11624 		ua_type = ctl_build_ua(lun, initidx, &ctsio->sense_data,
11625 		    &sense_len, SSD_TYPE_NONE);
11626 		if (ua_type != CTL_UA_NONE) {
11627 			mtx_unlock(&lun->lun_lock);
11628 			ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11629 			ctsio->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
11630 			ctsio->sense_len = sense_len;
11631 			ctl_done((union ctl_io *)ctsio);
11632 			return;
11633 		}
11634 	}
11635 
11636 	if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
11637 		mtx_unlock(&lun->lun_lock);
11638 		ctl_done((union ctl_io *)ctsio);
11639 		return;
11640 	}
11641 
11642 	/*
11643 	 * XXX CHD this is where we want to send IO to other side if
11644 	 * this LUN is secondary on this SC. We will need to make a copy
11645 	 * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11646 	 * the copy we send as FROM_OTHER.
11647 	 * We also need to stuff the address of the original IO so we can
11648 	 * find it easily. Something similar will need be done on the other
11649 	 * side so when we are done we can find the copy.
11650 	 */
11651 	if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
11652 	    (lun->flags & CTL_LUN_PEER_SC_PRIMARY) != 0 &&
11653 	    (entry->flags & CTL_CMD_FLAG_RUN_HERE) == 0) {
11654 		union ctl_ha_msg msg_info;
11655 		int isc_retval;
11656 
11657 		ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11658 		ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11659 		mtx_unlock(&lun->lun_lock);
11660 
11661 		msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11662 		msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11663 		msg_info.hdr.serializing_sc = NULL;
11664 		msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11665 		msg_info.scsi.tag_num = ctsio->tag_num;
11666 		msg_info.scsi.tag_type = ctsio->tag_type;
11667 		memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11668 		msg_info.scsi.cdb_len = ctsio->cdb_len;
11669 		msg_info.scsi.priority = ctsio->priority;
11670 
11671 		if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11672 		    sizeof(msg_info.scsi) - sizeof(msg_info.scsi.sense_data),
11673 		    M_WAITOK)) > CTL_HA_STATUS_SUCCESS) {
11674 			ctsio->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11675 			ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
11676 			ctl_set_busy(ctsio);
11677 			ctl_done((union ctl_io *)ctsio);
11678 			return;
11679 		}
11680 		return;
11681 	}
11682 
11683 	bio = (union ctl_io *)LIST_NEXT(&ctsio->io_hdr, ooa_links);
11684 	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio, &bio)) {
11685 	case CTL_ACTION_PASS:
11686 	case CTL_ACTION_SKIP:
11687 		ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11688 		mtx_unlock(&lun->lun_lock);
11689 		ctl_enqueue_rtr((union ctl_io *)ctsio);
11690 		break;
11691 	case CTL_ACTION_BLOCK:
11692 		ctsio->io_hdr.blocker = bio;
11693 		TAILQ_INSERT_TAIL(&bio->io_hdr.blocked_queue, &ctsio->io_hdr,
11694 				  blocked_links);
11695 		mtx_unlock(&lun->lun_lock);
11696 		break;
11697 	case CTL_ACTION_OVERLAP:
11698 		mtx_unlock(&lun->lun_lock);
11699 		ctl_set_overlapped_cmd(ctsio);
11700 		ctl_done((union ctl_io *)ctsio);
11701 		break;
11702 	case CTL_ACTION_OVERLAP_TAG:
11703 		mtx_unlock(&lun->lun_lock);
11704 		ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11705 		ctl_done((union ctl_io *)ctsio);
11706 		break;
11707 	default:
11708 		__assert_unreachable();
11709 	}
11710 }
11711 
11712 const struct ctl_cmd_entry *
11713 ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa)
11714 {
11715 	const struct ctl_cmd_entry *entry;
11716 	int service_action;
11717 
11718 	entry = &ctl_cmd_table[ctsio->cdb[0]];
11719 	if (sa)
11720 		*sa = ((entry->flags & CTL_CMD_FLAG_SA5) != 0);
11721 	if (entry->flags & CTL_CMD_FLAG_SA5) {
11722 		service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11723 		entry = &((const struct ctl_cmd_entry *)
11724 		    entry->execute)[service_action];
11725 	}
11726 	return (entry);
11727 }
11728 
11729 const struct ctl_cmd_entry *
11730 ctl_validate_command(struct ctl_scsiio *ctsio)
11731 {
11732 	const struct ctl_cmd_entry *entry;
11733 	int i, sa;
11734 	uint8_t diff;
11735 
11736 	entry = ctl_get_cmd_entry(ctsio, &sa);
11737 	ctsio->seridx = entry->seridx;
11738 	if (entry->execute == NULL) {
11739 		if (sa)
11740 			ctl_set_invalid_field(ctsio,
11741 					      /*sks_valid*/ 1,
11742 					      /*command*/ 1,
11743 					      /*field*/ 1,
11744 					      /*bit_valid*/ 1,
11745 					      /*bit*/ 4);
11746 		else
11747 			ctl_set_invalid_opcode(ctsio);
11748 		ctl_done((union ctl_io *)ctsio);
11749 		return (NULL);
11750 	}
11751 	KASSERT(entry->length > 0,
11752 	    ("Not defined length for command 0x%02x/0x%02x",
11753 	     ctsio->cdb[0], ctsio->cdb[1]));
11754 	for (i = 1; i < entry->length; i++) {
11755 		diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11756 		if (diff == 0)
11757 			continue;
11758 		ctl_set_invalid_field(ctsio,
11759 				      /*sks_valid*/ 1,
11760 				      /*command*/ 1,
11761 				      /*field*/ i,
11762 				      /*bit_valid*/ 1,
11763 				      /*bit*/ fls(diff) - 1);
11764 		ctl_done((union ctl_io *)ctsio);
11765 		return (NULL);
11766 	}
11767 	return (entry);
11768 }
11769 
11770 static int
11771 ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11772 {
11773 
11774 	switch (lun_type) {
11775 	case T_DIRECT:
11776 		if ((entry->flags & CTL_CMD_FLAG_OK_ON_DIRECT) == 0)
11777 			return (0);
11778 		break;
11779 	case T_PROCESSOR:
11780 		if ((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0)
11781 			return (0);
11782 		break;
11783 	case T_CDROM:
11784 		if ((entry->flags & CTL_CMD_FLAG_OK_ON_CDROM) == 0)
11785 			return (0);
11786 		break;
11787 	default:
11788 		return (0);
11789 	}
11790 	return (1);
11791 }
11792 
11793 static int
11794 ctl_scsiio(struct ctl_scsiio *ctsio)
11795 {
11796 	int retval;
11797 	const struct ctl_cmd_entry *entry;
11798 
11799 	retval = CTL_RETVAL_COMPLETE;
11800 
11801 	CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11802 
11803 	entry = ctl_get_cmd_entry(ctsio, NULL);
11804 
11805 	/*
11806 	 * If this I/O has been aborted, just send it straight to
11807 	 * ctl_done() without executing it.
11808 	 */
11809 	if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11810 		ctl_done((union ctl_io *)ctsio);
11811 		goto bailout;
11812 	}
11813 
11814 	/*
11815 	 * All the checks should have been handled by ctl_scsiio_precheck().
11816 	 * We should be clear now to just execute the I/O.
11817 	 */
11818 	retval = entry->execute(ctsio);
11819 
11820 bailout:
11821 	return (retval);
11822 }
11823 
11824 static int
11825 ctl_target_reset(union ctl_io *io)
11826 {
11827 	struct ctl_softc *softc = CTL_SOFTC(io);
11828 	struct ctl_port *port = CTL_PORT(io);
11829 	struct ctl_lun *lun;
11830 	uint32_t initidx;
11831 	ctl_ua_type ua_type;
11832 
11833 	if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11834 		union ctl_ha_msg msg_info;
11835 
11836 		msg_info.hdr.nexus = io->io_hdr.nexus;
11837 		msg_info.task.task_action = io->taskio.task_action;
11838 		msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11839 		msg_info.hdr.original_sc = NULL;
11840 		msg_info.hdr.serializing_sc = NULL;
11841 		ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11842 		    sizeof(msg_info.task), M_WAITOK);
11843 	}
11844 
11845 	initidx = ctl_get_initindex(&io->io_hdr.nexus);
11846 	if (io->taskio.task_action == CTL_TASK_TARGET_RESET)
11847 		ua_type = CTL_UA_TARG_RESET;
11848 	else
11849 		ua_type = CTL_UA_BUS_RESET;
11850 	mtx_lock(&softc->ctl_lock);
11851 	STAILQ_FOREACH(lun, &softc->lun_list, links) {
11852 		if (port != NULL &&
11853 		    ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
11854 			continue;
11855 		ctl_do_lun_reset(lun, initidx, ua_type);
11856 	}
11857 	mtx_unlock(&softc->ctl_lock);
11858 	io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11859 	return (0);
11860 }
11861 
11862 /*
11863  * The LUN should always be set.  The I/O is optional, and is used to
11864  * distinguish between I/Os sent by this initiator, and by other
11865  * initiators.  We set unit attention for initiators other than this one.
11866  * SAM-3 is vague on this point.  It does say that a unit attention should
11867  * be established for other initiators when a LUN is reset (see section
11868  * 5.7.3), but it doesn't specifically say that the unit attention should
11869  * be established for this particular initiator when a LUN is reset.  Here
11870  * is the relevant text, from SAM-3 rev 8:
11871  *
11872  * 5.7.2 When a SCSI initiator port aborts its own tasks
11873  *
11874  * When a SCSI initiator port causes its own task(s) to be aborted, no
11875  * notification that the task(s) have been aborted shall be returned to
11876  * the SCSI initiator port other than the completion response for the
11877  * command or task management function action that caused the task(s) to
11878  * be aborted and notification(s) associated with related effects of the
11879  * action (e.g., a reset unit attention condition).
11880  *
11881  * XXX KDM for now, we're setting unit attention for all initiators.
11882  */
11883 static void
11884 ctl_do_lun_reset(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua_type)
11885 {
11886 	struct ctl_io_hdr *xioh;
11887 	int i;
11888 
11889 	mtx_lock(&lun->lun_lock);
11890 	/* Abort tasks. */
11891 	LIST_FOREACH(xioh, &lun->ooa_queue, ooa_links) {
11892 		xioh->flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
11893 		ctl_try_unblock_io(lun, (union ctl_io *)xioh, FALSE);
11894 	}
11895 	/* Clear CA. */
11896 	for (i = 0; i < ctl_max_ports; i++) {
11897 		free(lun->pending_sense[i], M_CTL);
11898 		lun->pending_sense[i] = NULL;
11899 	}
11900 	/* Clear reservation. */
11901 	lun->flags &= ~CTL_LUN_RESERVED;
11902 	/* Clear prevent media removal. */
11903 	if (lun->prevent) {
11904 		for (i = 0; i < CTL_MAX_INITIATORS; i++)
11905 			ctl_clear_mask(lun->prevent, i);
11906 		lun->prevent_count = 0;
11907 	}
11908 	/* Clear TPC status */
11909 	ctl_tpc_lun_clear(lun, -1);
11910 	/* Establish UA. */
11911 #if 0
11912 	ctl_est_ua_all(lun, initidx, ua_type);
11913 #else
11914 	ctl_est_ua_all(lun, -1, ua_type);
11915 #endif
11916 	mtx_unlock(&lun->lun_lock);
11917 }
11918 
11919 static int
11920 ctl_lun_reset(union ctl_io *io)
11921 {
11922 	struct ctl_softc *softc = CTL_SOFTC(io);
11923 	struct ctl_lun *lun;
11924 	uint32_t targ_lun, initidx;
11925 
11926 	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11927 	initidx = ctl_get_initindex(&io->io_hdr.nexus);
11928 	mtx_lock(&softc->ctl_lock);
11929 	if (targ_lun >= ctl_max_luns ||
11930 	    (lun = softc->ctl_luns[targ_lun]) == NULL) {
11931 		mtx_unlock(&softc->ctl_lock);
11932 		io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11933 		return (1);
11934 	}
11935 	ctl_do_lun_reset(lun, initidx, CTL_UA_LUN_RESET);
11936 	mtx_unlock(&softc->ctl_lock);
11937 	io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11938 
11939 	if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0) {
11940 		union ctl_ha_msg msg_info;
11941 
11942 		msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11943 		msg_info.hdr.nexus = io->io_hdr.nexus;
11944 		msg_info.task.task_action = CTL_TASK_LUN_RESET;
11945 		msg_info.hdr.original_sc = NULL;
11946 		msg_info.hdr.serializing_sc = NULL;
11947 		ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11948 		    sizeof(msg_info.task), M_WAITOK);
11949 	}
11950 	return (0);
11951 }
11952 
11953 static void
11954 ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
11955     int other_sc)
11956 {
11957 	struct ctl_io_hdr *xioh;
11958 
11959 	mtx_assert(&lun->lun_lock, MA_OWNED);
11960 
11961 	/*
11962 	 * Run through the OOA queue and attempt to find the given I/O.
11963 	 * The target port, initiator ID, tag type and tag number have to
11964 	 * match the values that we got from the initiator.  If we have an
11965 	 * untagged command to abort, simply abort the first untagged command
11966 	 * we come to.  We only allow one untagged command at a time of course.
11967 	 */
11968 	LIST_FOREACH(xioh, &lun->ooa_queue, ooa_links) {
11969 		union ctl_io *xio = (union ctl_io *)xioh;
11970 		if ((targ_port == UINT32_MAX ||
11971 		     targ_port == xioh->nexus.targ_port) &&
11972 		    (init_id == UINT32_MAX ||
11973 		     init_id == xioh->nexus.initid)) {
11974 			if (targ_port != xioh->nexus.targ_port ||
11975 			    init_id != xioh->nexus.initid)
11976 				xioh->flags |= CTL_FLAG_ABORT_STATUS;
11977 			xioh->flags |= CTL_FLAG_ABORT;
11978 			if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11979 				union ctl_ha_msg msg_info;
11980 
11981 				msg_info.hdr.nexus = xioh->nexus;
11982 				msg_info.task.task_action = CTL_TASK_ABORT_TASK;
11983 				msg_info.task.tag_num = xio->scsiio.tag_num;
11984 				msg_info.task.tag_type = xio->scsiio.tag_type;
11985 				msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11986 				msg_info.hdr.original_sc = NULL;
11987 				msg_info.hdr.serializing_sc = NULL;
11988 				ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11989 				    sizeof(msg_info.task), M_NOWAIT);
11990 			}
11991 			ctl_try_unblock_io(lun, xio, FALSE);
11992 		}
11993 	}
11994 }
11995 
11996 static int
11997 ctl_abort_task_set(union ctl_io *io)
11998 {
11999 	struct ctl_softc *softc = CTL_SOFTC(io);
12000 	struct ctl_lun *lun;
12001 	uint32_t targ_lun;
12002 
12003 	/*
12004 	 * Look up the LUN.
12005 	 */
12006 	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12007 	mtx_lock(&softc->ctl_lock);
12008 	if (targ_lun >= ctl_max_luns ||
12009 	    (lun = softc->ctl_luns[targ_lun]) == NULL) {
12010 		mtx_unlock(&softc->ctl_lock);
12011 		io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12012 		return (1);
12013 	}
12014 
12015 	mtx_lock(&lun->lun_lock);
12016 	mtx_unlock(&softc->ctl_lock);
12017 	if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
12018 		ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12019 		    io->io_hdr.nexus.initid,
12020 		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12021 	} else { /* CTL_TASK_CLEAR_TASK_SET */
12022 		ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
12023 		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12024 	}
12025 	mtx_unlock(&lun->lun_lock);
12026 	io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12027 	return (0);
12028 }
12029 
12030 static void
12031 ctl_i_t_nexus_loss(struct ctl_softc *softc, uint32_t initidx,
12032     ctl_ua_type ua_type)
12033 {
12034 	struct ctl_lun *lun;
12035 	struct scsi_sense_data *ps;
12036 	uint32_t p, i;
12037 
12038 	p = initidx / CTL_MAX_INIT_PER_PORT;
12039 	i = initidx % CTL_MAX_INIT_PER_PORT;
12040 	mtx_lock(&softc->ctl_lock);
12041 	STAILQ_FOREACH(lun, &softc->lun_list, links) {
12042 		mtx_lock(&lun->lun_lock);
12043 		/* Abort tasks. */
12044 		ctl_abort_tasks_lun(lun, p, i, 1);
12045 		/* Clear CA. */
12046 		ps = lun->pending_sense[p];
12047 		if (ps != NULL)
12048 			ps[i].error_code = 0;
12049 		/* Clear reservation. */
12050 		if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == initidx))
12051 			lun->flags &= ~CTL_LUN_RESERVED;
12052 		/* Clear prevent media removal. */
12053 		if (lun->prevent && ctl_is_set(lun->prevent, initidx)) {
12054 			ctl_clear_mask(lun->prevent, initidx);
12055 			lun->prevent_count--;
12056 		}
12057 		/* Clear TPC status */
12058 		ctl_tpc_lun_clear(lun, initidx);
12059 		/* Establish UA. */
12060 		ctl_est_ua(lun, initidx, ua_type);
12061 		mtx_unlock(&lun->lun_lock);
12062 	}
12063 	mtx_unlock(&softc->ctl_lock);
12064 }
12065 
12066 static int
12067 ctl_i_t_nexus_reset(union ctl_io *io)
12068 {
12069 	struct ctl_softc *softc = CTL_SOFTC(io);
12070 	uint32_t initidx;
12071 
12072 	if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12073 		union ctl_ha_msg msg_info;
12074 
12075 		msg_info.hdr.nexus = io->io_hdr.nexus;
12076 		msg_info.task.task_action = CTL_TASK_I_T_NEXUS_RESET;
12077 		msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12078 		msg_info.hdr.original_sc = NULL;
12079 		msg_info.hdr.serializing_sc = NULL;
12080 		ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
12081 		    sizeof(msg_info.task), M_WAITOK);
12082 	}
12083 
12084 	initidx = ctl_get_initindex(&io->io_hdr.nexus);
12085 	ctl_i_t_nexus_loss(softc, initidx, CTL_UA_I_T_NEXUS_LOSS);
12086 	io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12087 	return (0);
12088 }
12089 
12090 static int
12091 ctl_abort_task(union ctl_io *io)
12092 {
12093 	struct ctl_softc *softc = CTL_SOFTC(io);
12094 	struct ctl_io_hdr *xioh;
12095 	struct ctl_lun *lun;
12096 	uint32_t targ_lun;
12097 
12098 	/*
12099 	 * Look up the LUN.
12100 	 */
12101 	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12102 	mtx_lock(&softc->ctl_lock);
12103 	if (targ_lun >= ctl_max_luns ||
12104 	    (lun = softc->ctl_luns[targ_lun]) == NULL) {
12105 		mtx_unlock(&softc->ctl_lock);
12106 		io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12107 		return (1);
12108 	}
12109 
12110 	mtx_lock(&lun->lun_lock);
12111 	mtx_unlock(&softc->ctl_lock);
12112 	/*
12113 	 * Run through the OOA queue and attempt to find the given I/O.
12114 	 * The target port, initiator ID, tag type and tag number have to
12115 	 * match the values that we got from the initiator.  If we have an
12116 	 * untagged command to abort, simply abort the first untagged command
12117 	 * we come to.  We only allow one untagged command at a time of course.
12118 	 */
12119 	LIST_FOREACH(xioh, &lun->ooa_queue, ooa_links) {
12120 		union ctl_io *xio = (union ctl_io *)xioh;
12121 		if ((xioh->nexus.targ_port != io->io_hdr.nexus.targ_port)
12122 		 || (xioh->nexus.initid != io->io_hdr.nexus.initid)
12123 		 || (xioh->flags & CTL_FLAG_ABORT))
12124 			continue;
12125 
12126 		/*
12127 		 * If the abort says that the task is untagged, the
12128 		 * task in the queue must be untagged.  Otherwise,
12129 		 * we just check to see whether the tag numbers
12130 		 * match.  This is because the QLogic firmware
12131 		 * doesn't pass back the tag type in an abort
12132 		 * request.
12133 		 */
12134 #if 0
12135 		if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
12136 		  && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
12137 		 || (xio->scsiio.tag_num == io->taskio.tag_num)) {
12138 #else
12139 		/*
12140 		 * XXX KDM we've got problems with FC, because it
12141 		 * doesn't send down a tag type with aborts.  So we
12142 		 * can only really go by the tag number...
12143 		 * This may cause problems with parallel SCSI.
12144 		 * Need to figure that out!!
12145 		 */
12146 		if (xio->scsiio.tag_num == io->taskio.tag_num) {
12147 #endif
12148 			xioh->flags |= CTL_FLAG_ABORT;
12149 			if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0 &&
12150 			    !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12151 				union ctl_ha_msg msg_info;
12152 
12153 				msg_info.hdr.nexus = io->io_hdr.nexus;
12154 				msg_info.task.task_action = CTL_TASK_ABORT_TASK;
12155 				msg_info.task.tag_num = io->taskio.tag_num;
12156 				msg_info.task.tag_type = io->taskio.tag_type;
12157 				msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12158 				msg_info.hdr.original_sc = NULL;
12159 				msg_info.hdr.serializing_sc = NULL;
12160 				ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
12161 				    sizeof(msg_info.task), M_NOWAIT);
12162 			}
12163 			ctl_try_unblock_io(lun, xio, FALSE);
12164 		}
12165 	}
12166 	mtx_unlock(&lun->lun_lock);
12167 	io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12168 	return (0);
12169 }
12170 
12171 static int
12172 ctl_query_task(union ctl_io *io, int task_set)
12173 {
12174 	struct ctl_softc *softc = CTL_SOFTC(io);
12175 	struct ctl_io_hdr *xioh;
12176 	struct ctl_lun *lun;
12177 	int found = 0;
12178 	uint32_t targ_lun;
12179 
12180 	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12181 	mtx_lock(&softc->ctl_lock);
12182 	if (targ_lun >= ctl_max_luns ||
12183 	    (lun = softc->ctl_luns[targ_lun]) == NULL) {
12184 		mtx_unlock(&softc->ctl_lock);
12185 		io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12186 		return (1);
12187 	}
12188 	mtx_lock(&lun->lun_lock);
12189 	mtx_unlock(&softc->ctl_lock);
12190 	LIST_FOREACH(xioh, &lun->ooa_queue, ooa_links) {
12191 		union ctl_io *xio = (union ctl_io *)xioh;
12192 		if ((xioh->nexus.targ_port != io->io_hdr.nexus.targ_port)
12193 		 || (xioh->nexus.initid != io->io_hdr.nexus.initid)
12194 		 || (xioh->flags & CTL_FLAG_ABORT))
12195 			continue;
12196 
12197 		if (task_set || xio->scsiio.tag_num == io->taskio.tag_num) {
12198 			found = 1;
12199 			break;
12200 		}
12201 	}
12202 	mtx_unlock(&lun->lun_lock);
12203 	if (found)
12204 		io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED;
12205 	else
12206 		io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12207 	return (0);
12208 }
12209 
12210 static int
12211 ctl_query_async_event(union ctl_io *io)
12212 {
12213 	struct ctl_softc *softc = CTL_SOFTC(io);
12214 	struct ctl_lun *lun;
12215 	ctl_ua_type ua;
12216 	uint32_t targ_lun, initidx;
12217 
12218 	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12219 	mtx_lock(&softc->ctl_lock);
12220 	if (targ_lun >= ctl_max_luns ||
12221 	    (lun = softc->ctl_luns[targ_lun]) == NULL) {
12222 		mtx_unlock(&softc->ctl_lock);
12223 		io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12224 		return (1);
12225 	}
12226 	mtx_lock(&lun->lun_lock);
12227 	mtx_unlock(&softc->ctl_lock);
12228 	initidx = ctl_get_initindex(&io->io_hdr.nexus);
12229 	ua = ctl_build_qae(lun, initidx, io->taskio.task_resp);
12230 	mtx_unlock(&lun->lun_lock);
12231 	if (ua != CTL_UA_NONE)
12232 		io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED;
12233 	else
12234 		io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12235 	return (0);
12236 }
12237 
12238 static void
12239 ctl_run_task(union ctl_io *io)
12240 {
12241 	int retval = 1;
12242 
12243 	CTL_DEBUG_PRINT(("ctl_run_task\n"));
12244 	KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
12245 	    ("ctl_run_task: Unextected io_type %d\n", io->io_hdr.io_type));
12246 	io->taskio.task_status = CTL_TASK_FUNCTION_NOT_SUPPORTED;
12247 	bzero(io->taskio.task_resp, sizeof(io->taskio.task_resp));
12248 	switch (io->taskio.task_action) {
12249 	case CTL_TASK_ABORT_TASK:
12250 		retval = ctl_abort_task(io);
12251 		break;
12252 	case CTL_TASK_ABORT_TASK_SET:
12253 	case CTL_TASK_CLEAR_TASK_SET:
12254 		retval = ctl_abort_task_set(io);
12255 		break;
12256 	case CTL_TASK_CLEAR_ACA:
12257 		break;
12258 	case CTL_TASK_I_T_NEXUS_RESET:
12259 		retval = ctl_i_t_nexus_reset(io);
12260 		break;
12261 	case CTL_TASK_LUN_RESET:
12262 		retval = ctl_lun_reset(io);
12263 		break;
12264 	case CTL_TASK_TARGET_RESET:
12265 	case CTL_TASK_BUS_RESET:
12266 		retval = ctl_target_reset(io);
12267 		break;
12268 	case CTL_TASK_PORT_LOGIN:
12269 		break;
12270 	case CTL_TASK_PORT_LOGOUT:
12271 		break;
12272 	case CTL_TASK_QUERY_TASK:
12273 		retval = ctl_query_task(io, 0);
12274 		break;
12275 	case CTL_TASK_QUERY_TASK_SET:
12276 		retval = ctl_query_task(io, 1);
12277 		break;
12278 	case CTL_TASK_QUERY_ASYNC_EVENT:
12279 		retval = ctl_query_async_event(io);
12280 		break;
12281 	default:
12282 		printf("%s: got unknown task management event %d\n",
12283 		       __func__, io->taskio.task_action);
12284 		break;
12285 	}
12286 	if (retval == 0)
12287 		io->io_hdr.status = CTL_SUCCESS;
12288 	else
12289 		io->io_hdr.status = CTL_ERROR;
12290 	ctl_done(io);
12291 }
12292 
12293 /*
12294  * For HA operation.  Handle commands that come in from the other
12295  * controller.
12296  */
12297 static void
12298 ctl_handle_isc(union ctl_io *io)
12299 {
12300 	struct ctl_softc *softc = CTL_SOFTC(io);
12301 	struct ctl_lun *lun;
12302 	const struct ctl_cmd_entry *entry;
12303 	uint32_t targ_lun;
12304 
12305 	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12306 	switch (io->io_hdr.msg_type) {
12307 	case CTL_MSG_SERIALIZE:
12308 		ctl_serialize_other_sc_cmd(&io->scsiio);
12309 		break;
12310 	case CTL_MSG_R2R:		/* Only used in SER_ONLY mode. */
12311 		entry = ctl_get_cmd_entry(&io->scsiio, NULL);
12312 		if (targ_lun >= ctl_max_luns ||
12313 		    (lun = softc->ctl_luns[targ_lun]) == NULL) {
12314 			ctl_done(io);
12315 			break;
12316 		}
12317 		mtx_lock(&lun->lun_lock);
12318 		if (ctl_scsiio_lun_check(lun, entry, &io->scsiio) != 0) {
12319 			mtx_unlock(&lun->lun_lock);
12320 			ctl_done(io);
12321 			break;
12322 		}
12323 		io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
12324 		mtx_unlock(&lun->lun_lock);
12325 		ctl_enqueue_rtr(io);
12326 		break;
12327 	case CTL_MSG_FINISH_IO:
12328 		if (softc->ha_mode == CTL_HA_MODE_XFER) {
12329 			ctl_done(io);
12330 			break;
12331 		}
12332 		if (targ_lun >= ctl_max_luns ||
12333 		    (lun = softc->ctl_luns[targ_lun]) == NULL) {
12334 			ctl_free_io(io);
12335 			break;
12336 		}
12337 		mtx_lock(&lun->lun_lock);
12338 		ctl_try_unblock_others(lun, io, TRUE);
12339 		LIST_REMOVE(&io->io_hdr, ooa_links);
12340 		mtx_unlock(&lun->lun_lock);
12341 		ctl_free_io(io);
12342 		break;
12343 	case CTL_MSG_PERS_ACTION:
12344 		ctl_hndl_per_res_out_on_other_sc(io);
12345 		ctl_free_io(io);
12346 		break;
12347 	case CTL_MSG_BAD_JUJU:
12348 		ctl_done(io);
12349 		break;
12350 	case CTL_MSG_DATAMOVE:		/* Only used in XFER mode */
12351 		ctl_datamove_remote(io);
12352 		break;
12353 	case CTL_MSG_DATAMOVE_DONE:	/* Only used in XFER mode */
12354 		ctl_datamove_done(io, false);
12355 		break;
12356 	case CTL_MSG_FAILOVER:
12357 		ctl_failover_lun(io);
12358 		ctl_free_io(io);
12359 		break;
12360 	default:
12361 		printf("%s: Invalid message type %d\n",
12362 		       __func__, io->io_hdr.msg_type);
12363 		ctl_free_io(io);
12364 		break;
12365 	}
12366 
12367 }
12368 
12369 /*
12370  * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12371  * there is no match.
12372  */
12373 static ctl_lun_error_pattern
12374 ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12375 {
12376 	const struct ctl_cmd_entry *entry;
12377 	ctl_lun_error_pattern filtered_pattern, pattern;
12378 
12379 	pattern = desc->error_pattern;
12380 
12381 	/*
12382 	 * XXX KDM we need more data passed into this function to match a
12383 	 * custom pattern, and we actually need to implement custom pattern
12384 	 * matching.
12385 	 */
12386 	if (pattern & CTL_LUN_PAT_CMD)
12387 		return (CTL_LUN_PAT_CMD);
12388 
12389 	if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12390 		return (CTL_LUN_PAT_ANY);
12391 
12392 	entry = ctl_get_cmd_entry(ctsio, NULL);
12393 
12394 	filtered_pattern = entry->pattern & pattern;
12395 
12396 	/*
12397 	 * If the user requested specific flags in the pattern (e.g.
12398 	 * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12399 	 * flags.
12400 	 *
12401 	 * If the user did not specify any flags, it doesn't matter whether
12402 	 * or not the command supports the flags.
12403 	 */
12404 	if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12405 	     (pattern & ~CTL_LUN_PAT_MASK))
12406 		return (CTL_LUN_PAT_NONE);
12407 
12408 	/*
12409 	 * If the user asked for a range check, see if the requested LBA
12410 	 * range overlaps with this command's LBA range.
12411 	 */
12412 	if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12413 		uint64_t lba1;
12414 		uint64_t len1;
12415 		ctl_action action;
12416 		int retval;
12417 
12418 		retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12419 		if (retval != 0)
12420 			return (CTL_LUN_PAT_NONE);
12421 
12422 		action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12423 					      desc->lba_range.len, FALSE);
12424 		/*
12425 		 * A "pass" means that the LBA ranges don't overlap, so
12426 		 * this doesn't match the user's range criteria.
12427 		 */
12428 		if (action == CTL_ACTION_PASS)
12429 			return (CTL_LUN_PAT_NONE);
12430 	}
12431 
12432 	return (filtered_pattern);
12433 }
12434 
12435 static void
12436 ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12437 {
12438 	struct ctl_error_desc *desc, *desc2;
12439 
12440 	mtx_assert(&lun->lun_lock, MA_OWNED);
12441 
12442 	STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12443 		ctl_lun_error_pattern pattern;
12444 		/*
12445 		 * Check to see whether this particular command matches
12446 		 * the pattern in the descriptor.
12447 		 */
12448 		pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12449 		if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12450 			continue;
12451 
12452 		switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12453 		case CTL_LUN_INJ_ABORTED:
12454 			ctl_set_aborted(&io->scsiio);
12455 			break;
12456 		case CTL_LUN_INJ_MEDIUM_ERR:
12457 			ctl_set_medium_error(&io->scsiio,
12458 			    (io->io_hdr.flags & CTL_FLAG_DATA_MASK) !=
12459 			     CTL_FLAG_DATA_OUT);
12460 			break;
12461 		case CTL_LUN_INJ_UA:
12462 			/* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12463 			 * OCCURRED */
12464 			ctl_set_ua(&io->scsiio, 0x29, 0x00);
12465 			break;
12466 		case CTL_LUN_INJ_CUSTOM:
12467 			/*
12468 			 * We're assuming the user knows what he is doing.
12469 			 * Just copy the sense information without doing
12470 			 * checks.
12471 			 */
12472 			bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12473 			      MIN(sizeof(desc->custom_sense),
12474 				  sizeof(io->scsiio.sense_data)));
12475 			io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12476 			io->scsiio.sense_len = SSD_FULL_SIZE;
12477 			io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12478 			break;
12479 		case CTL_LUN_INJ_NONE:
12480 		default:
12481 			/*
12482 			 * If this is an error injection type we don't know
12483 			 * about, clear the continuous flag (if it is set)
12484 			 * so it will get deleted below.
12485 			 */
12486 			desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12487 			break;
12488 		}
12489 		/*
12490 		 * By default, each error injection action is a one-shot
12491 		 */
12492 		if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12493 			continue;
12494 
12495 		STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12496 
12497 		free(desc, M_CTL);
12498 	}
12499 }
12500 
12501 #ifdef CTL_IO_DELAY
12502 static void
12503 ctl_datamove_timer_wakeup(void *arg)
12504 {
12505 	union ctl_io *io;
12506 
12507 	io = (union ctl_io *)arg;
12508 
12509 	ctl_datamove(io);
12510 }
12511 #endif /* CTL_IO_DELAY */
12512 
12513 static void
12514 ctl_datamove_done_process(union ctl_io *io)
12515 {
12516 #ifdef CTL_TIME_IO
12517 	struct bintime cur_bt;
12518 #endif
12519 
12520 	KASSERT(io->io_hdr.io_type == CTL_IO_SCSI,
12521 	    ("%s: unexpected I/O type %x", __func__, io->io_hdr.io_type));
12522 
12523 #ifdef CTL_TIME_IO
12524 	getbinuptime(&cur_bt);
12525 	bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt);
12526 	bintime_add(&io->io_hdr.dma_bt, &cur_bt);
12527 #endif
12528 	io->io_hdr.num_dmas++;
12529 
12530 	if ((io->io_hdr.port_status != 0) &&
12531 	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
12532 	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
12533 		ctl_set_internal_failure(&io->scsiio, /*sks_valid*/ 1,
12534 		    /*retry_count*/ io->io_hdr.port_status);
12535 	} else if (io->scsiio.kern_data_resid != 0 &&
12536 	    (io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT &&
12537 	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
12538 	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
12539 		ctl_set_invalid_field_ciu(&io->scsiio);
12540 	} else if (ctl_debug & CTL_DEBUG_CDB_DATA)
12541 		ctl_data_print(io);
12542 }
12543 
12544 void
12545 ctl_datamove_done(union ctl_io *io, bool samethr)
12546 {
12547 
12548 	ctl_datamove_done_process(io);
12549 	io->scsiio.be_move_done(io, samethr);
12550 }
12551 
12552 void
12553 ctl_datamove(union ctl_io *io)
12554 {
12555 	void (*fe_datamove)(union ctl_io *io);
12556 
12557 	mtx_assert(&((struct ctl_softc *)CTL_SOFTC(io))->ctl_lock, MA_NOTOWNED);
12558 
12559 	CTL_DEBUG_PRINT(("ctl_datamove\n"));
12560 
12561 	/* No data transferred yet.  Frontend must update this when done. */
12562 	io->scsiio.kern_data_resid = io->scsiio.kern_data_len;
12563 
12564 #ifdef CTL_TIME_IO
12565 	getbinuptime(&io->io_hdr.dma_start_bt);
12566 #endif /* CTL_TIME_IO */
12567 
12568 #ifdef CTL_IO_DELAY
12569 	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12570 		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12571 	} else {
12572 		struct ctl_lun *lun;
12573 
12574 		lun = CTL_LUN(io);
12575 		if ((lun != NULL)
12576 		 && (lun->delay_info.datamove_delay > 0)) {
12577 			callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
12578 			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12579 			callout_reset(&io->io_hdr.delay_callout,
12580 				      lun->delay_info.datamove_delay * hz,
12581 				      ctl_datamove_timer_wakeup, io);
12582 			if (lun->delay_info.datamove_type ==
12583 			    CTL_DELAY_TYPE_ONESHOT)
12584 				lun->delay_info.datamove_delay = 0;
12585 			return;
12586 		}
12587 	}
12588 #endif
12589 
12590 	/*
12591 	 * This command has been aborted.  Set the port status, so we fail
12592 	 * the data move.
12593 	 */
12594 	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12595 		printf("ctl_datamove: tag 0x%jx on (%u:%u:%u) aborted\n",
12596 		       io->scsiio.tag_num, io->io_hdr.nexus.initid,
12597 		       io->io_hdr.nexus.targ_port,
12598 		       io->io_hdr.nexus.targ_lun);
12599 		io->io_hdr.port_status = 31337;
12600 		ctl_datamove_done_process(io);
12601 		io->scsiio.be_move_done(io, true);
12602 		return;
12603 	}
12604 
12605 	/* Don't confuse frontend with zero length data move. */
12606 	if (io->scsiio.kern_data_len == 0) {
12607 		ctl_datamove_done_process(io);
12608 		io->scsiio.be_move_done(io, true);
12609 		return;
12610 	}
12611 
12612 	fe_datamove = CTL_PORT(io)->fe_datamove;
12613 	fe_datamove(io);
12614 }
12615 
12616 static void
12617 ctl_send_datamove_done(union ctl_io *io, int have_lock)
12618 {
12619 	union ctl_ha_msg msg;
12620 #ifdef CTL_TIME_IO
12621 	struct bintime cur_bt;
12622 #endif
12623 
12624 	memset(&msg, 0, sizeof(msg));
12625 	msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12626 	msg.hdr.original_sc = io;
12627 	msg.hdr.serializing_sc = io->io_hdr.remote_io;
12628 	msg.hdr.nexus = io->io_hdr.nexus;
12629 	msg.hdr.status = io->io_hdr.status;
12630 	msg.scsi.kern_data_resid = io->scsiio.kern_data_resid;
12631 	msg.scsi.tag_num = io->scsiio.tag_num;
12632 	msg.scsi.tag_type = io->scsiio.tag_type;
12633 	msg.scsi.scsi_status = io->scsiio.scsi_status;
12634 	memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12635 	       io->scsiio.sense_len);
12636 	msg.scsi.sense_len = io->scsiio.sense_len;
12637 	msg.scsi.port_status = io->io_hdr.port_status;
12638 	io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12639 	if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12640 		ctl_failover_io(io, /*have_lock*/ have_lock);
12641 		return;
12642 	}
12643 	ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12644 	    sizeof(msg.scsi) - sizeof(msg.scsi.sense_data) +
12645 	    msg.scsi.sense_len, M_WAITOK);
12646 
12647 #ifdef CTL_TIME_IO
12648 	getbinuptime(&cur_bt);
12649 	bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt);
12650 	bintime_add(&io->io_hdr.dma_bt, &cur_bt);
12651 #endif
12652 	io->io_hdr.num_dmas++;
12653 }
12654 
12655 /*
12656  * The DMA to the remote side is done, now we need to tell the other side
12657  * we're done so it can continue with its data movement.
12658  */
12659 static void
12660 ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12661 {
12662 	union ctl_io *io;
12663 	uint32_t i;
12664 
12665 	io = rq->context;
12666 
12667 	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12668 		printf("%s: ISC DMA write failed with error %d", __func__,
12669 		       rq->ret);
12670 		ctl_set_internal_failure(&io->scsiio,
12671 					 /*sks_valid*/ 1,
12672 					 /*retry_count*/ rq->ret);
12673 	}
12674 
12675 	ctl_dt_req_free(rq);
12676 
12677 	for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12678 		free(CTL_LSGLT(io)[i].addr, M_CTL);
12679 	free(CTL_RSGL(io), M_CTL);
12680 	CTL_RSGL(io) = NULL;
12681 	CTL_LSGL(io) = NULL;
12682 
12683 	/*
12684 	 * The data is in local and remote memory, so now we need to send
12685 	 * status (good or back) back to the other side.
12686 	 */
12687 	ctl_send_datamove_done(io, /*have_lock*/ 0);
12688 }
12689 
12690 /*
12691  * We've moved the data from the host/controller into local memory.  Now we
12692  * need to push it over to the remote controller's memory.
12693  */
12694 static int
12695 ctl_datamove_remote_dm_write_cb(union ctl_io *io, bool samethr)
12696 {
12697 	int retval;
12698 
12699 	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12700 					  ctl_datamove_remote_write_cb);
12701 	return (retval);
12702 }
12703 
12704 static void
12705 ctl_datamove_remote_write(union ctl_io *io)
12706 {
12707 	int retval;
12708 	void (*fe_datamove)(union ctl_io *io);
12709 
12710 	/*
12711 	 * - Get the data from the host/HBA into local memory.
12712 	 * - DMA memory from the local controller to the remote controller.
12713 	 * - Send status back to the remote controller.
12714 	 */
12715 
12716 	retval = ctl_datamove_remote_sgl_setup(io);
12717 	if (retval != 0)
12718 		return;
12719 
12720 	/* Switch the pointer over so the FETD knows what to do */
12721 	io->scsiio.kern_data_ptr = (uint8_t *)CTL_LSGL(io);
12722 
12723 	/*
12724 	 * Use a custom move done callback, since we need to send completion
12725 	 * back to the other controller, not to the backend on this side.
12726 	 */
12727 	io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12728 
12729 	fe_datamove = CTL_PORT(io)->fe_datamove;
12730 	fe_datamove(io);
12731 }
12732 
12733 static int
12734 ctl_datamove_remote_dm_read_cb(union ctl_io *io, bool samethr)
12735 {
12736 	uint32_t i;
12737 
12738 	for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12739 		free(CTL_LSGLT(io)[i].addr, M_CTL);
12740 	free(CTL_RSGL(io), M_CTL);
12741 	CTL_RSGL(io) = NULL;
12742 	CTL_LSGL(io) = NULL;
12743 
12744 	/*
12745 	 * The read is done, now we need to send status (good or bad) back
12746 	 * to the other side.
12747 	 */
12748 	ctl_send_datamove_done(io, /*have_lock*/ 0);
12749 
12750 	return (0);
12751 }
12752 
12753 static void
12754 ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
12755 {
12756 	union ctl_io *io;
12757 	void (*fe_datamove)(union ctl_io *io);
12758 
12759 	io = rq->context;
12760 
12761 	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12762 		printf("%s: ISC DMA read failed with error %d\n", __func__,
12763 		       rq->ret);
12764 		ctl_set_internal_failure(&io->scsiio,
12765 					 /*sks_valid*/ 1,
12766 					 /*retry_count*/ rq->ret);
12767 	}
12768 
12769 	ctl_dt_req_free(rq);
12770 
12771 	/* Switch the pointer over so the FETD knows what to do */
12772 	io->scsiio.kern_data_ptr = (uint8_t *)CTL_LSGL(io);
12773 
12774 	/*
12775 	 * Use a custom move done callback, since we need to send completion
12776 	 * back to the other controller, not to the backend on this side.
12777 	 */
12778 	io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
12779 
12780 	/* XXX KDM add checks like the ones in ctl_datamove? */
12781 
12782 	fe_datamove = CTL_PORT(io)->fe_datamove;
12783 	fe_datamove(io);
12784 }
12785 
12786 static int
12787 ctl_datamove_remote_sgl_setup(union ctl_io *io)
12788 {
12789 	struct ctl_sg_entry *local_sglist;
12790 	uint32_t len_to_go;
12791 	int retval;
12792 	int i;
12793 
12794 	retval = 0;
12795 	local_sglist = CTL_LSGL(io);
12796 	len_to_go = io->scsiio.kern_data_len;
12797 
12798 	/*
12799 	 * The difficult thing here is that the size of the various
12800 	 * S/G segments may be different than the size from the
12801 	 * remote controller.  That'll make it harder when DMAing
12802 	 * the data back to the other side.
12803 	 */
12804 	for (i = 0; len_to_go > 0; i++) {
12805 		local_sglist[i].len = MIN(len_to_go, CTL_HA_DATAMOVE_SEGMENT);
12806 		local_sglist[i].addr =
12807 		    malloc(local_sglist[i].len, M_CTL, M_WAITOK);
12808 
12809 		len_to_go -= local_sglist[i].len;
12810 	}
12811 	/*
12812 	 * Reset the number of S/G entries accordingly.  The original
12813 	 * number of S/G entries is available in rem_sg_entries.
12814 	 */
12815 	io->scsiio.kern_sg_entries = i;
12816 
12817 	return (retval);
12818 }
12819 
12820 static int
12821 ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
12822 			 ctl_ha_dt_cb callback)
12823 {
12824 	struct ctl_ha_dt_req *rq;
12825 	struct ctl_sg_entry *remote_sglist, *local_sglist;
12826 	uint32_t local_used, remote_used, total_used;
12827 	int i, j, isc_ret;
12828 
12829 	rq = ctl_dt_req_alloc();
12830 
12831 	/*
12832 	 * If we failed to allocate the request, and if the DMA didn't fail
12833 	 * anyway, set busy status.  This is just a resource allocation
12834 	 * failure.
12835 	 */
12836 	if ((rq == NULL)
12837 	 && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
12838 	     (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS))
12839 		ctl_set_busy(&io->scsiio);
12840 
12841 	if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
12842 	    (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) {
12843 		if (rq != NULL)
12844 			ctl_dt_req_free(rq);
12845 
12846 		/*
12847 		 * The data move failed.  We need to return status back
12848 		 * to the other controller.  No point in trying to DMA
12849 		 * data to the remote controller.
12850 		 */
12851 
12852 		ctl_send_datamove_done(io, /*have_lock*/ 0);
12853 
12854 		return (1);
12855 	}
12856 
12857 	local_sglist = CTL_LSGL(io);
12858 	remote_sglist = CTL_RSGL(io);
12859 	local_used = 0;
12860 	remote_used = 0;
12861 	total_used = 0;
12862 
12863 	/*
12864 	 * Pull/push the data over the wire from/to the other controller.
12865 	 * This takes into account the possibility that the local and
12866 	 * remote sglists may not be identical in terms of the size of
12867 	 * the elements and the number of elements.
12868 	 *
12869 	 * One fundamental assumption here is that the length allocated for
12870 	 * both the local and remote sglists is identical.  Otherwise, we've
12871 	 * essentially got a coding error of some sort.
12872 	 */
12873 	isc_ret = CTL_HA_STATUS_SUCCESS;
12874 	for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
12875 		uint32_t cur_len;
12876 		uint8_t *tmp_ptr;
12877 
12878 		rq->command = command;
12879 		rq->context = io;
12880 
12881 		/*
12882 		 * Both pointers should be aligned.  But it is possible
12883 		 * that the allocation length is not.  They should both
12884 		 * also have enough slack left over at the end, though,
12885 		 * to round up to the next 8 byte boundary.
12886 		 */
12887 		cur_len = MIN(local_sglist[i].len - local_used,
12888 			      remote_sglist[j].len - remote_used);
12889 		rq->size = cur_len;
12890 
12891 		tmp_ptr = (uint8_t *)local_sglist[i].addr;
12892 		tmp_ptr += local_used;
12893 
12894 #if 0
12895 		/* Use physical addresses when talking to ISC hardware */
12896 		if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
12897 			/* XXX KDM use busdma */
12898 			rq->local = vtophys(tmp_ptr);
12899 		} else
12900 			rq->local = tmp_ptr;
12901 #else
12902 		KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0,
12903 		    ("HA does not support BUS_ADDR"));
12904 		rq->local = tmp_ptr;
12905 #endif
12906 
12907 		tmp_ptr = (uint8_t *)remote_sglist[j].addr;
12908 		tmp_ptr += remote_used;
12909 		rq->remote = tmp_ptr;
12910 
12911 		rq->callback = NULL;
12912 
12913 		local_used += cur_len;
12914 		if (local_used >= local_sglist[i].len) {
12915 			i++;
12916 			local_used = 0;
12917 		}
12918 
12919 		remote_used += cur_len;
12920 		if (remote_used >= remote_sglist[j].len) {
12921 			j++;
12922 			remote_used = 0;
12923 		}
12924 		total_used += cur_len;
12925 
12926 		if (total_used >= io->scsiio.kern_data_len)
12927 			rq->callback = callback;
12928 
12929 		isc_ret = ctl_dt_single(rq);
12930 		if (isc_ret > CTL_HA_STATUS_SUCCESS)
12931 			break;
12932 	}
12933 	if (isc_ret != CTL_HA_STATUS_WAIT) {
12934 		rq->ret = isc_ret;
12935 		callback(rq);
12936 	}
12937 
12938 	return (0);
12939 }
12940 
12941 static void
12942 ctl_datamove_remote_read(union ctl_io *io)
12943 {
12944 	int retval;
12945 	uint32_t i;
12946 
12947 	/*
12948 	 * This will send an error to the other controller in the case of a
12949 	 * failure.
12950 	 */
12951 	retval = ctl_datamove_remote_sgl_setup(io);
12952 	if (retval != 0)
12953 		return;
12954 
12955 	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
12956 					  ctl_datamove_remote_read_cb);
12957 	if (retval != 0) {
12958 		/*
12959 		 * Make sure we free memory if there was an error..  The
12960 		 * ctl_datamove_remote_xfer() function will send the
12961 		 * datamove done message, or call the callback with an
12962 		 * error if there is a problem.
12963 		 */
12964 		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12965 			free(CTL_LSGLT(io)[i].addr, M_CTL);
12966 		free(CTL_RSGL(io), M_CTL);
12967 		CTL_RSGL(io) = NULL;
12968 		CTL_LSGL(io) = NULL;
12969 	}
12970 }
12971 
12972 /*
12973  * Process a datamove request from the other controller.  This is used for
12974  * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
12975  * first.  Once that is complete, the data gets DMAed into the remote
12976  * controller's memory.  For reads, we DMA from the remote controller's
12977  * memory into our memory first, and then move it out to the FETD.
12978  */
12979 static void
12980 ctl_datamove_remote(union ctl_io *io)
12981 {
12982 
12983 	mtx_assert(&((struct ctl_softc *)CTL_SOFTC(io))->ctl_lock, MA_NOTOWNED);
12984 
12985 	if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12986 		ctl_failover_io(io, /*have_lock*/ 0);
12987 		return;
12988 	}
12989 
12990 	/*
12991 	 * Note that we look for an aborted I/O here, but don't do some of
12992 	 * the other checks that ctl_datamove() normally does.
12993 	 * We don't need to run the datamove delay code, since that should
12994 	 * have been done if need be on the other controller.
12995 	 */
12996 	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12997 		printf("%s: tag 0x%jx on (%u:%u:%u) aborted\n", __func__,
12998 		       io->scsiio.tag_num, io->io_hdr.nexus.initid,
12999 		       io->io_hdr.nexus.targ_port,
13000 		       io->io_hdr.nexus.targ_lun);
13001 		io->io_hdr.port_status = 31338;
13002 		ctl_send_datamove_done(io, /*have_lock*/ 0);
13003 		return;
13004 	}
13005 
13006 	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT)
13007 		ctl_datamove_remote_write(io);
13008 	else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
13009 		ctl_datamove_remote_read(io);
13010 	else {
13011 		io->io_hdr.port_status = 31339;
13012 		ctl_send_datamove_done(io, /*have_lock*/ 0);
13013 	}
13014 }
13015 
13016 static void
13017 ctl_process_done(union ctl_io *io)
13018 {
13019 	struct ctl_softc *softc = CTL_SOFTC(io);
13020 	struct ctl_port *port = CTL_PORT(io);
13021 	struct ctl_lun *lun = CTL_LUN(io);
13022 	void (*fe_done)(union ctl_io *io);
13023 	union ctl_ha_msg msg;
13024 
13025 	CTL_DEBUG_PRINT(("ctl_process_done\n"));
13026 	fe_done = port->fe_done;
13027 
13028 #ifdef CTL_TIME_IO
13029 	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
13030 		char str[256];
13031 		char path_str[64];
13032 		struct sbuf sb;
13033 
13034 		ctl_scsi_path_string(io, path_str, sizeof(path_str));
13035 		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13036 
13037 		ctl_io_sbuf(io, &sb);
13038 		sbuf_cat(&sb, path_str);
13039 		sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
13040 			    (intmax_t)time_uptime - io->io_hdr.start_time);
13041 		sbuf_finish(&sb);
13042 		printf("%s", sbuf_data(&sb));
13043 	}
13044 #endif /* CTL_TIME_IO */
13045 
13046 	switch (io->io_hdr.io_type) {
13047 	case CTL_IO_SCSI:
13048 		break;
13049 	case CTL_IO_TASK:
13050 		if (ctl_debug & CTL_DEBUG_INFO)
13051 			ctl_io_error_print(io, NULL);
13052 		fe_done(io);
13053 		return;
13054 	default:
13055 		panic("%s: Invalid CTL I/O type %d\n",
13056 		    __func__, io->io_hdr.io_type);
13057 	}
13058 
13059 	if (lun == NULL) {
13060 		CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
13061 				 io->io_hdr.nexus.targ_mapped_lun));
13062 		goto bailout;
13063 	}
13064 
13065 	mtx_lock(&lun->lun_lock);
13066 
13067 	/*
13068 	 * Check to see if we have any informational exception and status
13069 	 * of this command can be modified to report it in form of either
13070 	 * RECOVERED ERROR or NO SENSE, depending on MRIE mode page field.
13071 	 */
13072 	if (lun->ie_reported == 0 && lun->ie_asc != 0 &&
13073 	    io->io_hdr.status == CTL_SUCCESS &&
13074 	    (io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0) {
13075 		uint8_t mrie = lun->MODE_IE.mrie;
13076 		uint8_t per = ((lun->MODE_RWER.byte3 & SMS_RWER_PER) ||
13077 		    (lun->MODE_VER.byte3 & SMS_VER_PER));
13078 		if (((mrie == SIEP_MRIE_REC_COND && per) ||
13079 		     mrie == SIEP_MRIE_REC_UNCOND ||
13080 		     mrie == SIEP_MRIE_NO_SENSE) &&
13081 		    (ctl_get_cmd_entry(&io->scsiio, NULL)->flags &
13082 		     CTL_CMD_FLAG_NO_SENSE) == 0) {
13083 			ctl_set_sense(&io->scsiio,
13084 			      /*current_error*/ 1,
13085 			      /*sense_key*/ (mrie == SIEP_MRIE_NO_SENSE) ?
13086 			        SSD_KEY_NO_SENSE : SSD_KEY_RECOVERED_ERROR,
13087 			      /*asc*/ lun->ie_asc,
13088 			      /*ascq*/ lun->ie_ascq,
13089 			      SSD_ELEM_NONE);
13090 			lun->ie_reported = 1;
13091 		}
13092 	} else if (lun->ie_reported < 0)
13093 		lun->ie_reported = 0;
13094 
13095 	/*
13096 	 * Check to see if we have any errors to inject here.  We only
13097 	 * inject errors for commands that don't already have errors set.
13098 	 */
13099 	if (!STAILQ_EMPTY(&lun->error_list) &&
13100 	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) &&
13101 	    ((io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0))
13102 		ctl_inject_error(lun, io);
13103 
13104 	/*
13105 	 * XXX KDM how do we treat commands that aren't completed
13106 	 * successfully?
13107 	 *
13108 	 * XXX KDM should we also track I/O latency?
13109 	 */
13110 	if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
13111 	    io->io_hdr.io_type == CTL_IO_SCSI) {
13112 		int type;
13113 #ifdef CTL_TIME_IO
13114 		struct bintime bt;
13115 
13116 		getbinuptime(&bt);
13117 		bintime_sub(&bt, &io->io_hdr.start_bt);
13118 #endif
13119 		if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13120 		    CTL_FLAG_DATA_IN)
13121 			type = CTL_STATS_READ;
13122 		else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13123 		    CTL_FLAG_DATA_OUT)
13124 			type = CTL_STATS_WRITE;
13125 		else
13126 			type = CTL_STATS_NO_IO;
13127 
13128 		lun->stats.bytes[type] += io->scsiio.kern_total_len;
13129 		lun->stats.operations[type] ++;
13130 		lun->stats.dmas[type] += io->io_hdr.num_dmas;
13131 #ifdef CTL_TIME_IO
13132 		bintime_add(&lun->stats.dma_time[type], &io->io_hdr.dma_bt);
13133 		bintime_add(&lun->stats.time[type], &bt);
13134 #endif
13135 
13136 		mtx_lock(&port->port_lock);
13137 		port->stats.bytes[type] += io->scsiio.kern_total_len;
13138 		port->stats.operations[type] ++;
13139 		port->stats.dmas[type] += io->io_hdr.num_dmas;
13140 #ifdef CTL_TIME_IO
13141 		bintime_add(&port->stats.dma_time[type], &io->io_hdr.dma_bt);
13142 		bintime_add(&port->stats.time[type], &bt);
13143 #endif
13144 		mtx_unlock(&port->port_lock);
13145 	}
13146 
13147 	/*
13148 	 * Run through the blocked queue of this I/O and see if anything
13149 	 * can be unblocked, now that this I/O is done and will be removed.
13150 	 * We need to do it before removal to have OOA position to start.
13151 	 */
13152 	ctl_try_unblock_others(lun, io, TRUE);
13153 
13154 	/*
13155 	 * Remove this from the OOA queue.
13156 	 */
13157 	LIST_REMOVE(&io->io_hdr, ooa_links);
13158 #ifdef CTL_TIME_IO
13159 	if (LIST_EMPTY(&lun->ooa_queue))
13160 		lun->last_busy = getsbinuptime();
13161 #endif
13162 
13163 	/*
13164 	 * If the LUN has been invalidated, free it if there is nothing
13165 	 * left on its OOA queue.
13166 	 */
13167 	if ((lun->flags & CTL_LUN_INVALID)
13168 	 && LIST_EMPTY(&lun->ooa_queue)) {
13169 		mtx_unlock(&lun->lun_lock);
13170 		ctl_free_lun(lun);
13171 	} else
13172 		mtx_unlock(&lun->lun_lock);
13173 
13174 bailout:
13175 
13176 	/*
13177 	 * If this command has been aborted, make sure we set the status
13178 	 * properly.  The FETD is responsible for freeing the I/O and doing
13179 	 * whatever it needs to do to clean up its state.
13180 	 */
13181 	if (io->io_hdr.flags & CTL_FLAG_ABORT)
13182 		ctl_set_task_aborted(&io->scsiio);
13183 
13184 	/*
13185 	 * If enabled, print command error status.
13186 	 */
13187 	if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS &&
13188 	    (ctl_debug & CTL_DEBUG_INFO) != 0)
13189 		ctl_io_error_print(io, NULL);
13190 
13191 	/*
13192 	 * Tell the FETD or the other shelf controller we're done with this
13193 	 * command.  Note that only SCSI commands get to this point.  Task
13194 	 * management commands are completed above.
13195 	 */
13196 	if ((softc->ha_mode != CTL_HA_MODE_XFER) &&
13197 	    (io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)) {
13198 		memset(&msg, 0, sizeof(msg));
13199 		msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13200 		msg.hdr.serializing_sc = io->io_hdr.remote_io;
13201 		msg.hdr.nexus = io->io_hdr.nexus;
13202 		ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13203 		    sizeof(msg.scsi) - sizeof(msg.scsi.sense_data),
13204 		    M_WAITOK);
13205 	}
13206 
13207 	fe_done(io);
13208 }
13209 
13210 /*
13211  * Front end should call this if it doesn't do autosense.  When the request
13212  * sense comes back in from the initiator, we'll dequeue this and send it.
13213  */
13214 int
13215 ctl_queue_sense(union ctl_io *io)
13216 {
13217 	struct ctl_softc *softc = CTL_SOFTC(io);
13218 	struct ctl_port *port = CTL_PORT(io);
13219 	struct ctl_lun *lun;
13220 	struct scsi_sense_data *ps;
13221 	uint32_t initidx, p, targ_lun;
13222 
13223 	CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13224 
13225 	targ_lun = ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13226 
13227 	/*
13228 	 * LUN lookup will likely move to the ctl_work_thread() once we
13229 	 * have our new queueing infrastructure (that doesn't put things on
13230 	 * a per-LUN queue initially).  That is so that we can handle
13231 	 * things like an INQUIRY to a LUN that we don't have enabled.  We
13232 	 * can't deal with that right now.
13233 	 * If we don't have a LUN for this, just toss the sense information.
13234 	 */
13235 	mtx_lock(&softc->ctl_lock);
13236 	if (targ_lun >= ctl_max_luns ||
13237 	    (lun = softc->ctl_luns[targ_lun]) == NULL) {
13238 		mtx_unlock(&softc->ctl_lock);
13239 		goto bailout;
13240 	}
13241 	mtx_lock(&lun->lun_lock);
13242 	mtx_unlock(&softc->ctl_lock);
13243 
13244 	initidx = ctl_get_initindex(&io->io_hdr.nexus);
13245 	p = initidx / CTL_MAX_INIT_PER_PORT;
13246 	if (lun->pending_sense[p] == NULL) {
13247 		lun->pending_sense[p] = malloc(sizeof(*ps) * CTL_MAX_INIT_PER_PORT,
13248 		    M_CTL, M_NOWAIT | M_ZERO);
13249 	}
13250 	if ((ps = lun->pending_sense[p]) != NULL) {
13251 		ps += initidx % CTL_MAX_INIT_PER_PORT;
13252 		memset(ps, 0, sizeof(*ps));
13253 		memcpy(ps, &io->scsiio.sense_data, io->scsiio.sense_len);
13254 	}
13255 	mtx_unlock(&lun->lun_lock);
13256 
13257 bailout:
13258 	ctl_free_io(io);
13259 	return (CTL_RETVAL_COMPLETE);
13260 }
13261 
13262 /*
13263  * Primary command inlet from frontend ports.  All SCSI and task I/O
13264  * requests must go through this function.
13265  */
13266 int
13267 ctl_queue(union ctl_io *io)
13268 {
13269 	struct ctl_port *port = CTL_PORT(io);
13270 
13271 	CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13272 
13273 #ifdef CTL_TIME_IO
13274 	io->io_hdr.start_time = time_uptime;
13275 	getbinuptime(&io->io_hdr.start_bt);
13276 #endif /* CTL_TIME_IO */
13277 
13278 	/* Map FE-specific LUN ID into global one. */
13279 	io->io_hdr.nexus.targ_mapped_lun =
13280 	    ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13281 
13282 	switch (io->io_hdr.io_type) {
13283 	case CTL_IO_SCSI:
13284 	case CTL_IO_TASK:
13285 		if (ctl_debug & CTL_DEBUG_CDB)
13286 			ctl_io_print(io);
13287 		ctl_enqueue_incoming(io);
13288 		break;
13289 	default:
13290 		printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13291 		return (EINVAL);
13292 	}
13293 
13294 	return (CTL_RETVAL_COMPLETE);
13295 }
13296 
13297 int
13298 ctl_run(union ctl_io *io)
13299 {
13300 	struct ctl_port *port = CTL_PORT(io);
13301 
13302 	CTL_DEBUG_PRINT(("ctl_run cdb[0]=%02X\n", io->scsiio.cdb[0]));
13303 
13304 #ifdef CTL_TIME_IO
13305 	io->io_hdr.start_time = time_uptime;
13306 	getbinuptime(&io->io_hdr.start_bt);
13307 #endif /* CTL_TIME_IO */
13308 
13309 	/* Map FE-specific LUN ID into global one. */
13310 	io->io_hdr.nexus.targ_mapped_lun =
13311 	    ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13312 
13313 	switch (io->io_hdr.io_type) {
13314 	case CTL_IO_SCSI:
13315 		if (ctl_debug & CTL_DEBUG_CDB)
13316 			ctl_io_print(io);
13317 		ctl_scsiio_precheck(&io->scsiio);
13318 		break;
13319 	case CTL_IO_TASK:
13320 		if (ctl_debug & CTL_DEBUG_CDB)
13321 			ctl_io_print(io);
13322 		ctl_run_task(io);
13323 		break;
13324 	default:
13325 		printf("ctl_run: unknown I/O type %d\n", io->io_hdr.io_type);
13326 		return (EINVAL);
13327 	}
13328 
13329 	return (CTL_RETVAL_COMPLETE);
13330 }
13331 
13332 #ifdef CTL_IO_DELAY
13333 static void
13334 ctl_done_timer_wakeup(void *arg)
13335 {
13336 	union ctl_io *io;
13337 
13338 	io = (union ctl_io *)arg;
13339 	ctl_done(io);
13340 }
13341 #endif /* CTL_IO_DELAY */
13342 
13343 void
13344 ctl_serseq_done(union ctl_io *io)
13345 {
13346 	struct ctl_lun *lun = CTL_LUN(io);
13347 
13348 	/* This is racy, but should not be a problem. */
13349 	if (!TAILQ_EMPTY(&io->io_hdr.blocked_queue)) {
13350 		mtx_lock(&lun->lun_lock);
13351 		io->io_hdr.flags |= CTL_FLAG_SERSEQ_DONE;
13352 		ctl_try_unblock_others(lun, io, FALSE);
13353 		mtx_unlock(&lun->lun_lock);
13354 	} else
13355 		io->io_hdr.flags |= CTL_FLAG_SERSEQ_DONE;
13356 }
13357 
13358 void
13359 ctl_done(union ctl_io *io)
13360 {
13361 
13362 	/*
13363 	 * Enable this to catch duplicate completion issues.
13364 	 */
13365 #if 0
13366 	if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13367 		printf("%s: type %d msg %d cdb %x iptl: "
13368 		       "%u:%u:%u tag 0x%04x "
13369 		       "flag %#x status %x\n",
13370 			__func__,
13371 			io->io_hdr.io_type,
13372 			io->io_hdr.msg_type,
13373 			io->scsiio.cdb[0],
13374 			io->io_hdr.nexus.initid,
13375 			io->io_hdr.nexus.targ_port,
13376 			io->io_hdr.nexus.targ_lun,
13377 			(io->io_hdr.io_type ==
13378 			CTL_IO_TASK) ?
13379 			io->taskio.tag_num :
13380 			io->scsiio.tag_num,
13381 		        io->io_hdr.flags,
13382 			io->io_hdr.status);
13383 	} else
13384 		io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13385 #endif
13386 
13387 	/*
13388 	 * This is an internal copy of an I/O, and should not go through
13389 	 * the normal done processing logic.
13390 	 */
13391 	if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
13392 		return;
13393 
13394 #ifdef CTL_IO_DELAY
13395 	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13396 		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13397 	} else {
13398 		struct ctl_lun *lun = CTL_LUN(io);
13399 
13400 		if ((lun != NULL)
13401 		 && (lun->delay_info.done_delay > 0)) {
13402 			callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
13403 			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13404 			callout_reset(&io->io_hdr.delay_callout,
13405 				      lun->delay_info.done_delay * hz,
13406 				      ctl_done_timer_wakeup, io);
13407 			if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13408 				lun->delay_info.done_delay = 0;
13409 			return;
13410 		}
13411 	}
13412 #endif /* CTL_IO_DELAY */
13413 
13414 	ctl_enqueue_done(io);
13415 }
13416 
13417 static void
13418 ctl_work_thread(void *arg)
13419 {
13420 	struct ctl_thread *thr = (struct ctl_thread *)arg;
13421 	struct ctl_softc *softc = thr->ctl_softc;
13422 	union ctl_io *io;
13423 	int retval;
13424 
13425 	CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13426 	thread_lock(curthread);
13427 	sched_prio(curthread, PUSER - 1);
13428 	thread_unlock(curthread);
13429 
13430 	while (!softc->shutdown) {
13431 		/*
13432 		 * We handle the queues in this order:
13433 		 * - ISC
13434 		 * - done queue (to free up resources, unblock other commands)
13435 		 * - incoming queue
13436 		 * - RtR queue
13437 		 *
13438 		 * If those queues are empty, we break out of the loop and
13439 		 * go to sleep.
13440 		 */
13441 		mtx_lock(&thr->queue_lock);
13442 		io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
13443 		if (io != NULL) {
13444 			STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
13445 			mtx_unlock(&thr->queue_lock);
13446 			ctl_handle_isc(io);
13447 			continue;
13448 		}
13449 		io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
13450 		if (io != NULL) {
13451 			STAILQ_REMOVE_HEAD(&thr->done_queue, links);
13452 			/* clear any blocked commands, call fe_done */
13453 			mtx_unlock(&thr->queue_lock);
13454 			ctl_process_done(io);
13455 			continue;
13456 		}
13457 		io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
13458 		if (io != NULL) {
13459 			STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
13460 			mtx_unlock(&thr->queue_lock);
13461 			if (io->io_hdr.io_type == CTL_IO_TASK)
13462 				ctl_run_task(io);
13463 			else
13464 				ctl_scsiio_precheck(&io->scsiio);
13465 			continue;
13466 		}
13467 		io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
13468 		if (io != NULL) {
13469 			STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
13470 			mtx_unlock(&thr->queue_lock);
13471 			retval = ctl_scsiio(&io->scsiio);
13472 			if (retval != CTL_RETVAL_COMPLETE)
13473 				CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13474 			continue;
13475 		}
13476 
13477 		/* Sleep until we have something to do. */
13478 		mtx_sleep(thr, &thr->queue_lock, PDROP, "-", 0);
13479 	}
13480 	thr->thread = NULL;
13481 	kthread_exit();
13482 }
13483 
13484 static void
13485 ctl_thresh_thread(void *arg)
13486 {
13487 	struct ctl_softc *softc = (struct ctl_softc *)arg;
13488 	struct ctl_lun *lun;
13489 	struct ctl_logical_block_provisioning_page *page;
13490 	const char *attr;
13491 	union ctl_ha_msg msg;
13492 	uint64_t thres, val;
13493 	int i, e, set;
13494 
13495 	CTL_DEBUG_PRINT(("ctl_thresh_thread starting\n"));
13496 	thread_lock(curthread);
13497 	sched_prio(curthread, PUSER - 1);
13498 	thread_unlock(curthread);
13499 
13500 	while (!softc->shutdown) {
13501 		mtx_lock(&softc->ctl_lock);
13502 		STAILQ_FOREACH(lun, &softc->lun_list, links) {
13503 			if ((lun->flags & CTL_LUN_DISABLED) ||
13504 			    (lun->flags & CTL_LUN_NO_MEDIA) ||
13505 			    lun->backend->lun_attr == NULL)
13506 				continue;
13507 			if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
13508 			    softc->ha_mode == CTL_HA_MODE_XFER)
13509 				continue;
13510 			if ((lun->MODE_RWER.byte8 & SMS_RWER_LBPERE) == 0)
13511 				continue;
13512 			e = 0;
13513 			page = &lun->MODE_LBP;
13514 			for (i = 0; i < CTL_NUM_LBP_THRESH; i++) {
13515 				if ((page->descr[i].flags & SLBPPD_ENABLED) == 0)
13516 					continue;
13517 				thres = scsi_4btoul(page->descr[i].count);
13518 				thres <<= CTL_LBP_EXPONENT;
13519 				switch (page->descr[i].resource) {
13520 				case 0x01:
13521 					attr = "blocksavail";
13522 					break;
13523 				case 0x02:
13524 					attr = "blocksused";
13525 					break;
13526 				case 0xf1:
13527 					attr = "poolblocksavail";
13528 					break;
13529 				case 0xf2:
13530 					attr = "poolblocksused";
13531 					break;
13532 				default:
13533 					continue;
13534 				}
13535 				mtx_unlock(&softc->ctl_lock); // XXX
13536 				val = lun->backend->lun_attr(lun->be_lun, attr);
13537 				mtx_lock(&softc->ctl_lock);
13538 				if (val == UINT64_MAX)
13539 					continue;
13540 				if ((page->descr[i].flags & SLBPPD_ARMING_MASK)
13541 				    == SLBPPD_ARMING_INC)
13542 					e = (val >= thres);
13543 				else
13544 					e = (val <= thres);
13545 				if (e)
13546 					break;
13547 			}
13548 			mtx_lock(&lun->lun_lock);
13549 			if (e) {
13550 				scsi_u64to8b((uint8_t *)&page->descr[i] -
13551 				    (uint8_t *)page, lun->ua_tpt_info);
13552 				if (lun->lasttpt == 0 ||
13553 				    time_uptime - lun->lasttpt >= CTL_LBP_UA_PERIOD) {
13554 					lun->lasttpt = time_uptime;
13555 					ctl_est_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13556 					set = 1;
13557 				} else
13558 					set = 0;
13559 			} else {
13560 				lun->lasttpt = 0;
13561 				ctl_clr_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13562 				set = -1;
13563 			}
13564 			mtx_unlock(&lun->lun_lock);
13565 			if (set != 0 &&
13566 			    lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
13567 				/* Send msg to other side. */
13568 				bzero(&msg.ua, sizeof(msg.ua));
13569 				msg.hdr.msg_type = CTL_MSG_UA;
13570 				msg.hdr.nexus.initid = -1;
13571 				msg.hdr.nexus.targ_port = -1;
13572 				msg.hdr.nexus.targ_lun = lun->lun;
13573 				msg.hdr.nexus.targ_mapped_lun = lun->lun;
13574 				msg.ua.ua_all = 1;
13575 				msg.ua.ua_set = (set > 0);
13576 				msg.ua.ua_type = CTL_UA_THIN_PROV_THRES;
13577 				memcpy(msg.ua.ua_info, lun->ua_tpt_info, 8);
13578 				mtx_unlock(&softc->ctl_lock); // XXX
13579 				ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13580 				    sizeof(msg.ua), M_WAITOK);
13581 				mtx_lock(&softc->ctl_lock);
13582 			}
13583 		}
13584 		mtx_sleep(&softc->thresh_thread, &softc->ctl_lock,
13585 		    PDROP, "-", CTL_LBP_PERIOD * hz);
13586 	}
13587 	softc->thresh_thread = NULL;
13588 	kthread_exit();
13589 }
13590 
13591 static void
13592 ctl_enqueue_incoming(union ctl_io *io)
13593 {
13594 	struct ctl_softc *softc = CTL_SOFTC(io);
13595 	struct ctl_thread *thr;
13596 	u_int idx;
13597 
13598 	idx = (io->io_hdr.nexus.targ_port * 127 +
13599 	       io->io_hdr.nexus.initid) % worker_threads;
13600 	thr = &softc->threads[idx];
13601 	mtx_lock(&thr->queue_lock);
13602 	STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
13603 	mtx_unlock(&thr->queue_lock);
13604 	wakeup(thr);
13605 }
13606 
13607 static void
13608 ctl_enqueue_rtr(union ctl_io *io)
13609 {
13610 	struct ctl_softc *softc = CTL_SOFTC(io);
13611 	struct ctl_thread *thr;
13612 
13613 	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13614 	mtx_lock(&thr->queue_lock);
13615 	STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
13616 	mtx_unlock(&thr->queue_lock);
13617 	wakeup(thr);
13618 }
13619 
13620 static void
13621 ctl_enqueue_done(union ctl_io *io)
13622 {
13623 	struct ctl_softc *softc = CTL_SOFTC(io);
13624 	struct ctl_thread *thr;
13625 
13626 	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13627 	mtx_lock(&thr->queue_lock);
13628 	STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
13629 	mtx_unlock(&thr->queue_lock);
13630 	wakeup(thr);
13631 }
13632 
13633 static void
13634 ctl_enqueue_isc(union ctl_io *io)
13635 {
13636 	struct ctl_softc *softc = CTL_SOFTC(io);
13637 	struct ctl_thread *thr;
13638 
13639 	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13640 	mtx_lock(&thr->queue_lock);
13641 	STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
13642 	mtx_unlock(&thr->queue_lock);
13643 	wakeup(thr);
13644 }
13645 
13646 /*
13647  *  vim: ts=8
13648  */
13649