xref: /freebsd/sys/cam/ctl/ctl.c (revision bd81e07d2761cf1c13063eb49a5c0cb4a6951318)
1 /*-
2  * Copyright (c) 2003-2009 Silicon Graphics International Corp.
3  * Copyright (c) 2012 The FreeBSD Foundation
4  * All rights reserved.
5  *
6  * Portions of this software were developed by Edward Tomasz Napierala
7  * under sponsorship from the FreeBSD Foundation.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions, and the following disclaimer,
14  *    without modification.
15  * 2. Redistributions in binary form must reproduce at minimum a disclaimer
16  *    substantially similar to the "NO WARRANTY" disclaimer below
17  *    ("Disclaimer") and any redistribution must be conditioned upon
18  *    including a substantially similar Disclaimer requirement for further
19  *    binary redistribution.
20  *
21  * NO WARRANTY
22  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
23  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
24  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR
25  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
26  * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
30  * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
31  * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
32  * POSSIBILITY OF SUCH DAMAGES.
33  *
34  * $Id$
35  */
36 /*
37  * CAM Target Layer, a SCSI device emulation subsystem.
38  *
39  * Author: Ken Merry <ken@FreeBSD.org>
40  */
41 
42 #define _CTL_C
43 
44 #include <sys/cdefs.h>
45 __FBSDID("$FreeBSD$");
46 
47 #include <sys/param.h>
48 #include <sys/systm.h>
49 #include <sys/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/sysctl.h>
67 #include <vm/uma.h>
68 
69 #include <cam/cam.h>
70 #include <cam/scsi/scsi_all.h>
71 #include <cam/scsi/scsi_da.h>
72 #include <cam/ctl/ctl_io.h>
73 #include <cam/ctl/ctl.h>
74 #include <cam/ctl/ctl_frontend.h>
75 #include <cam/ctl/ctl_util.h>
76 #include <cam/ctl/ctl_backend.h>
77 #include <cam/ctl/ctl_ioctl.h>
78 #include <cam/ctl/ctl_ha.h>
79 #include <cam/ctl/ctl_private.h>
80 #include <cam/ctl/ctl_debug.h>
81 #include <cam/ctl/ctl_scsi_all.h>
82 #include <cam/ctl/ctl_error.h>
83 
84 struct ctl_softc *control_softc = NULL;
85 
86 /*
87  * Size and alignment macros needed for Copan-specific HA hardware.  These
88  * can go away when the HA code is re-written, and uses busdma for any
89  * hardware.
90  */
91 #define	CTL_ALIGN_8B(target, source, type)				\
92 	if (((uint32_t)source & 0x7) != 0)				\
93 		target = (type)(source + (0x8 - ((uint32_t)source & 0x7)));\
94 	else								\
95 		target = (type)source;
96 
97 #define	CTL_SIZE_8B(target, size)					\
98 	if ((size & 0x7) != 0)						\
99 		target = size + (0x8 - (size & 0x7));			\
100 	else								\
101 		target = size;
102 
103 #define CTL_ALIGN_8B_MARGIN	16
104 
105 /*
106  * Template mode pages.
107  */
108 
109 /*
110  * Note that these are default values only.  The actual values will be
111  * filled in when the user does a mode sense.
112  */
113 const static struct copan_debugconf_subpage debugconf_page_default = {
114 	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
115 	DBGCNF_SUBPAGE_CODE,		/* subpage */
116 	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
117 	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
118 	DBGCNF_VERSION,			/* page_version */
119 	{CTL_TIME_IO_DEFAULT_SECS>>8,
120 	 CTL_TIME_IO_DEFAULT_SECS>>0},	/* ctl_time_io_secs */
121 };
122 
123 const static struct copan_debugconf_subpage debugconf_page_changeable = {
124 	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
125 	DBGCNF_SUBPAGE_CODE,		/* subpage */
126 	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
127 	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
128 	0,				/* page_version */
129 	{0xff,0xff},			/* ctl_time_io_secs */
130 };
131 
132 const static struct scsi_da_rw_recovery_page rw_er_page_default = {
133 	/*page_code*/SMS_RW_ERROR_RECOVERY_PAGE,
134 	/*page_length*/sizeof(struct scsi_da_rw_recovery_page) - 2,
135 	/*byte3*/SMS_RWER_AWRE|SMS_RWER_ARRE,
136 	/*read_retry_count*/0,
137 	/*correction_span*/0,
138 	/*head_offset_count*/0,
139 	/*data_strobe_offset_cnt*/0,
140 	/*byte8*/SMS_RWER_LBPERE,
141 	/*write_retry_count*/0,
142 	/*reserved2*/0,
143 	/*recovery_time_limit*/{0, 0},
144 };
145 
146 const static struct scsi_da_rw_recovery_page rw_er_page_changeable = {
147 	/*page_code*/SMS_RW_ERROR_RECOVERY_PAGE,
148 	/*page_length*/sizeof(struct scsi_da_rw_recovery_page) - 2,
149 	/*byte3*/0,
150 	/*read_retry_count*/0,
151 	/*correction_span*/0,
152 	/*head_offset_count*/0,
153 	/*data_strobe_offset_cnt*/0,
154 	/*byte8*/0,
155 	/*write_retry_count*/0,
156 	/*reserved2*/0,
157 	/*recovery_time_limit*/{0, 0},
158 };
159 
160 const static struct scsi_format_page format_page_default = {
161 	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
162 	/*page_length*/sizeof(struct scsi_format_page) - 2,
163 	/*tracks_per_zone*/ {0, 0},
164 	/*alt_sectors_per_zone*/ {0, 0},
165 	/*alt_tracks_per_zone*/ {0, 0},
166 	/*alt_tracks_per_lun*/ {0, 0},
167 	/*sectors_per_track*/ {(CTL_DEFAULT_SECTORS_PER_TRACK >> 8) & 0xff,
168 			        CTL_DEFAULT_SECTORS_PER_TRACK & 0xff},
169 	/*bytes_per_sector*/ {0, 0},
170 	/*interleave*/ {0, 0},
171 	/*track_skew*/ {0, 0},
172 	/*cylinder_skew*/ {0, 0},
173 	/*flags*/ SFP_HSEC,
174 	/*reserved*/ {0, 0, 0}
175 };
176 
177 const static struct scsi_format_page format_page_changeable = {
178 	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
179 	/*page_length*/sizeof(struct scsi_format_page) - 2,
180 	/*tracks_per_zone*/ {0, 0},
181 	/*alt_sectors_per_zone*/ {0, 0},
182 	/*alt_tracks_per_zone*/ {0, 0},
183 	/*alt_tracks_per_lun*/ {0, 0},
184 	/*sectors_per_track*/ {0, 0},
185 	/*bytes_per_sector*/ {0, 0},
186 	/*interleave*/ {0, 0},
187 	/*track_skew*/ {0, 0},
188 	/*cylinder_skew*/ {0, 0},
189 	/*flags*/ 0,
190 	/*reserved*/ {0, 0, 0}
191 };
192 
193 const static struct scsi_rigid_disk_page rigid_disk_page_default = {
194 	/*page_code*/SMS_RIGID_DISK_PAGE,
195 	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
196 	/*cylinders*/ {0, 0, 0},
197 	/*heads*/ CTL_DEFAULT_HEADS,
198 	/*start_write_precomp*/ {0, 0, 0},
199 	/*start_reduced_current*/ {0, 0, 0},
200 	/*step_rate*/ {0, 0},
201 	/*landing_zone_cylinder*/ {0, 0, 0},
202 	/*rpl*/ SRDP_RPL_DISABLED,
203 	/*rotational_offset*/ 0,
204 	/*reserved1*/ 0,
205 	/*rotation_rate*/ {(CTL_DEFAULT_ROTATION_RATE >> 8) & 0xff,
206 			   CTL_DEFAULT_ROTATION_RATE & 0xff},
207 	/*reserved2*/ {0, 0}
208 };
209 
210 const static struct scsi_rigid_disk_page rigid_disk_page_changeable = {
211 	/*page_code*/SMS_RIGID_DISK_PAGE,
212 	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
213 	/*cylinders*/ {0, 0, 0},
214 	/*heads*/ 0,
215 	/*start_write_precomp*/ {0, 0, 0},
216 	/*start_reduced_current*/ {0, 0, 0},
217 	/*step_rate*/ {0, 0},
218 	/*landing_zone_cylinder*/ {0, 0, 0},
219 	/*rpl*/ 0,
220 	/*rotational_offset*/ 0,
221 	/*reserved1*/ 0,
222 	/*rotation_rate*/ {0, 0},
223 	/*reserved2*/ {0, 0}
224 };
225 
226 const static struct scsi_caching_page caching_page_default = {
227 	/*page_code*/SMS_CACHING_PAGE,
228 	/*page_length*/sizeof(struct scsi_caching_page) - 2,
229 	/*flags1*/ SCP_DISC | SCP_WCE,
230 	/*ret_priority*/ 0,
231 	/*disable_pf_transfer_len*/ {0xff, 0xff},
232 	/*min_prefetch*/ {0, 0},
233 	/*max_prefetch*/ {0xff, 0xff},
234 	/*max_pf_ceiling*/ {0xff, 0xff},
235 	/*flags2*/ 0,
236 	/*cache_segments*/ 0,
237 	/*cache_seg_size*/ {0, 0},
238 	/*reserved*/ 0,
239 	/*non_cache_seg_size*/ {0, 0, 0}
240 };
241 
242 const static struct scsi_caching_page caching_page_changeable = {
243 	/*page_code*/SMS_CACHING_PAGE,
244 	/*page_length*/sizeof(struct scsi_caching_page) - 2,
245 	/*flags1*/ SCP_WCE | SCP_RCD,
246 	/*ret_priority*/ 0,
247 	/*disable_pf_transfer_len*/ {0, 0},
248 	/*min_prefetch*/ {0, 0},
249 	/*max_prefetch*/ {0, 0},
250 	/*max_pf_ceiling*/ {0, 0},
251 	/*flags2*/ 0,
252 	/*cache_segments*/ 0,
253 	/*cache_seg_size*/ {0, 0},
254 	/*reserved*/ 0,
255 	/*non_cache_seg_size*/ {0, 0, 0}
256 };
257 
258 const static struct scsi_control_page control_page_default = {
259 	/*page_code*/SMS_CONTROL_MODE_PAGE,
260 	/*page_length*/sizeof(struct scsi_control_page) - 2,
261 	/*rlec*/0,
262 	/*queue_flags*/SCP_QUEUE_ALG_RESTRICTED,
263 	/*eca_and_aen*/0,
264 	/*flags4*/SCP_TAS,
265 	/*aen_holdoff_period*/{0, 0},
266 	/*busy_timeout_period*/{0, 0},
267 	/*extended_selftest_completion_time*/{0, 0}
268 };
269 
270 const static struct scsi_control_page control_page_changeable = {
271 	/*page_code*/SMS_CONTROL_MODE_PAGE,
272 	/*page_length*/sizeof(struct scsi_control_page) - 2,
273 	/*rlec*/SCP_DSENSE,
274 	/*queue_flags*/SCP_QUEUE_ALG_MASK,
275 	/*eca_and_aen*/SCP_SWP,
276 	/*flags4*/0,
277 	/*aen_holdoff_period*/{0, 0},
278 	/*busy_timeout_period*/{0, 0},
279 	/*extended_selftest_completion_time*/{0, 0}
280 };
281 
282 const static struct scsi_info_exceptions_page ie_page_default = {
283 	/*page_code*/SMS_INFO_EXCEPTIONS_PAGE,
284 	/*page_length*/sizeof(struct scsi_info_exceptions_page) - 2,
285 	/*info_flags*/SIEP_FLAGS_DEXCPT,
286 	/*mrie*/0,
287 	/*interval_timer*/{0, 0, 0, 0},
288 	/*report_count*/{0, 0, 0, 0}
289 };
290 
291 const static struct scsi_info_exceptions_page ie_page_changeable = {
292 	/*page_code*/SMS_INFO_EXCEPTIONS_PAGE,
293 	/*page_length*/sizeof(struct scsi_info_exceptions_page) - 2,
294 	/*info_flags*/0,
295 	/*mrie*/0,
296 	/*interval_timer*/{0, 0, 0, 0},
297 	/*report_count*/{0, 0, 0, 0}
298 };
299 
300 #define CTL_LBPM_LEN	(sizeof(struct ctl_logical_block_provisioning_page) - 4)
301 
302 const static struct ctl_logical_block_provisioning_page lbp_page_default = {{
303 	/*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF,
304 	/*subpage_code*/0x02,
305 	/*page_length*/{CTL_LBPM_LEN >> 8, CTL_LBPM_LEN},
306 	/*flags*/0,
307 	/*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
308 	/*descr*/{}},
309 	{{/*flags*/0,
310 	  /*resource*/0x01,
311 	  /*reserved*/{0, 0},
312 	  /*count*/{0, 0, 0, 0}},
313 	 {/*flags*/0,
314 	  /*resource*/0x02,
315 	  /*reserved*/{0, 0},
316 	  /*count*/{0, 0, 0, 0}},
317 	 {/*flags*/0,
318 	  /*resource*/0xf1,
319 	  /*reserved*/{0, 0},
320 	  /*count*/{0, 0, 0, 0}},
321 	 {/*flags*/0,
322 	  /*resource*/0xf2,
323 	  /*reserved*/{0, 0},
324 	  /*count*/{0, 0, 0, 0}}
325 	}
326 };
327 
328 const static struct ctl_logical_block_provisioning_page lbp_page_changeable = {{
329 	/*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF,
330 	/*subpage_code*/0x02,
331 	/*page_length*/{CTL_LBPM_LEN >> 8, CTL_LBPM_LEN},
332 	/*flags*/0,
333 	/*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
334 	/*descr*/{}},
335 	{{/*flags*/0,
336 	  /*resource*/0,
337 	  /*reserved*/{0, 0},
338 	  /*count*/{0, 0, 0, 0}},
339 	 {/*flags*/0,
340 	  /*resource*/0,
341 	  /*reserved*/{0, 0},
342 	  /*count*/{0, 0, 0, 0}},
343 	 {/*flags*/0,
344 	  /*resource*/0,
345 	  /*reserved*/{0, 0},
346 	  /*count*/{0, 0, 0, 0}},
347 	 {/*flags*/0,
348 	  /*resource*/0,
349 	  /*reserved*/{0, 0},
350 	  /*count*/{0, 0, 0, 0}}
351 	}
352 };
353 
354 /*
355  * XXX KDM move these into the softc.
356  */
357 static int rcv_sync_msg;
358 static uint8_t ctl_pause_rtr;
359 
360 SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD, 0, "CAM Target Layer");
361 static int worker_threads = -1;
362 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, worker_threads, CTLFLAG_RDTUN,
363     &worker_threads, 1, "Number of worker threads");
364 static int ctl_debug = CTL_DEBUG_NONE;
365 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, debug, CTLFLAG_RWTUN,
366     &ctl_debug, 0, "Enabled debug flags");
367 
368 /*
369  * Supported pages (0x00), Serial number (0x80), Device ID (0x83),
370  * Extended INQUIRY Data (0x86), Mode Page Policy (0x87),
371  * SCSI Ports (0x88), Third-party Copy (0x8F), Block limits (0xB0),
372  * Block Device Characteristics (0xB1) and Logical Block Provisioning (0xB2)
373  */
374 #define SCSI_EVPD_NUM_SUPPORTED_PAGES	10
375 
376 #ifdef notyet
377 static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
378 				  int param);
379 static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
380 #endif
381 static int ctl_init(void);
382 void ctl_shutdown(void);
383 static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
384 static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
385 static int ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio);
386 static int ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
387 			      struct ctl_ooa *ooa_hdr,
388 			      struct ctl_ooa_entry *kern_entries);
389 static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
390 		     struct thread *td);
391 static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
392 			 struct ctl_be_lun *be_lun);
393 static int ctl_free_lun(struct ctl_lun *lun);
394 static void ctl_create_lun(struct ctl_be_lun *be_lun);
395 static struct ctl_port * ctl_io_port(struct ctl_io_hdr *io_hdr);
396 /**
397 static void ctl_failover_change_pages(struct ctl_softc *softc,
398 				      struct ctl_scsiio *ctsio, int master);
399 **/
400 
401 static int ctl_do_mode_select(union ctl_io *io);
402 static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
403 			   uint64_t res_key, uint64_t sa_res_key,
404 			   uint8_t type, uint32_t residx,
405 			   struct ctl_scsiio *ctsio,
406 			   struct scsi_per_res_out *cdb,
407 			   struct scsi_per_res_out_parms* param);
408 static void ctl_pro_preempt_other(struct ctl_lun *lun,
409 				  union ctl_ha_msg *msg);
410 static void ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg);
411 static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
412 static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
413 static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
414 static int ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len);
415 static int ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len);
416 static int ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio,
417 					 int alloc_len);
418 static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
419 					 int alloc_len);
420 static int ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len);
421 static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
422 static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
423 static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
424 static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len);
425 static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2,
426     bool seq);
427 static ctl_action ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2);
428 static ctl_action ctl_check_for_blockage(struct ctl_lun *lun,
429     union ctl_io *pending_io, union ctl_io *ooa_io);
430 static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
431 				union ctl_io *starting_io);
432 static int ctl_check_blocked(struct ctl_lun *lun);
433 static int ctl_scsiio_lun_check(struct ctl_lun *lun,
434 				const struct ctl_cmd_entry *entry,
435 				struct ctl_scsiio *ctsio);
436 //static int ctl_check_rtr(union ctl_io *pending_io, struct ctl_softc *softc);
437 #ifdef notyet
438 static void ctl_failover(void);
439 #endif
440 static void ctl_clear_ua(struct ctl_softc *ctl_softc, uint32_t initidx,
441 			 ctl_ua_type ua_type);
442 static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
443 			       struct ctl_scsiio *ctsio);
444 static int ctl_scsiio(struct ctl_scsiio *ctsio);
445 
446 static int ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
447 static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
448 			    ctl_ua_type ua_type);
449 static int ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io,
450 			 ctl_ua_type ua_type);
451 static int ctl_abort_task(union ctl_io *io);
452 static int ctl_abort_task_set(union ctl_io *io);
453 static int ctl_i_t_nexus_reset(union ctl_io *io);
454 static void ctl_run_task(union ctl_io *io);
455 #ifdef CTL_IO_DELAY
456 static void ctl_datamove_timer_wakeup(void *arg);
457 static void ctl_done_timer_wakeup(void *arg);
458 #endif /* CTL_IO_DELAY */
459 
460 static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
461 static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
462 static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
463 static void ctl_datamove_remote_write(union ctl_io *io);
464 static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
465 static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
466 static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
467 static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
468 				    ctl_ha_dt_cb callback);
469 static void ctl_datamove_remote_read(union ctl_io *io);
470 static void ctl_datamove_remote(union ctl_io *io);
471 static int ctl_process_done(union ctl_io *io);
472 static void ctl_lun_thread(void *arg);
473 static void ctl_thresh_thread(void *arg);
474 static void ctl_work_thread(void *arg);
475 static void ctl_enqueue_incoming(union ctl_io *io);
476 static void ctl_enqueue_rtr(union ctl_io *io);
477 static void ctl_enqueue_done(union ctl_io *io);
478 #ifdef notyet
479 static void ctl_enqueue_isc(union ctl_io *io);
480 #endif
481 static const struct ctl_cmd_entry *
482     ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa);
483 static const struct ctl_cmd_entry *
484     ctl_validate_command(struct ctl_scsiio *ctsio);
485 static int ctl_cmd_applicable(uint8_t lun_type,
486     const struct ctl_cmd_entry *entry);
487 
488 /*
489  * Load the serialization table.  This isn't very pretty, but is probably
490  * the easiest way to do it.
491  */
492 #include "ctl_ser_table.c"
493 
494 /*
495  * We only need to define open, close and ioctl routines for this driver.
496  */
497 static struct cdevsw ctl_cdevsw = {
498 	.d_version =	D_VERSION,
499 	.d_flags =	0,
500 	.d_open =	ctl_open,
501 	.d_close =	ctl_close,
502 	.d_ioctl =	ctl_ioctl,
503 	.d_name =	"ctl",
504 };
505 
506 
507 MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
508 
509 static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
510 
511 static moduledata_t ctl_moduledata = {
512 	"ctl",
513 	ctl_module_event_handler,
514 	NULL
515 };
516 
517 DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
518 MODULE_VERSION(ctl, 1);
519 
520 #ifdef notyet
521 static void
522 ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
523 			    union ctl_ha_msg *msg_info)
524 {
525 	struct ctl_scsiio *ctsio;
526 
527 	if (msg_info->hdr.original_sc == NULL) {
528 		printf("%s: original_sc == NULL!\n", __func__);
529 		/* XXX KDM now what? */
530 		return;
531 	}
532 
533 	ctsio = &msg_info->hdr.original_sc->scsiio;
534 	ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
535 	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
536 	ctsio->io_hdr.status = msg_info->hdr.status;
537 	ctsio->scsi_status = msg_info->scsi.scsi_status;
538 	ctsio->sense_len = msg_info->scsi.sense_len;
539 	ctsio->sense_residual = msg_info->scsi.sense_residual;
540 	ctsio->residual = msg_info->scsi.residual;
541 	memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
542 	       sizeof(ctsio->sense_data));
543 	memcpy(&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
544 	       &msg_info->scsi.lbalen, sizeof(msg_info->scsi.lbalen));
545 	ctl_enqueue_isc((union ctl_io *)ctsio);
546 }
547 
548 static void
549 ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
550 				union ctl_ha_msg *msg_info)
551 {
552 	struct ctl_scsiio *ctsio;
553 
554 	if (msg_info->hdr.serializing_sc == NULL) {
555 		printf("%s: serializing_sc == NULL!\n", __func__);
556 		/* XXX KDM now what? */
557 		return;
558 	}
559 
560 	ctsio = &msg_info->hdr.serializing_sc->scsiio;
561 #if 0
562 	/*
563 	 * Attempt to catch the situation where an I/O has
564 	 * been freed, and we're using it again.
565 	 */
566 	if (ctsio->io_hdr.io_type == 0xff) {
567 		union ctl_io *tmp_io;
568 		tmp_io = (union ctl_io *)ctsio;
569 		printf("%s: %p use after free!\n", __func__,
570 		       ctsio);
571 		printf("%s: type %d msg %d cdb %x iptl: "
572 		       "%d:%d:%d:%d tag 0x%04x "
573 		       "flag %#x status %x\n",
574 			__func__,
575 			tmp_io->io_hdr.io_type,
576 			tmp_io->io_hdr.msg_type,
577 			tmp_io->scsiio.cdb[0],
578 			tmp_io->io_hdr.nexus.initid.id,
579 			tmp_io->io_hdr.nexus.targ_port,
580 			tmp_io->io_hdr.nexus.targ_target.id,
581 			tmp_io->io_hdr.nexus.targ_lun,
582 			(tmp_io->io_hdr.io_type ==
583 			CTL_IO_TASK) ?
584 			tmp_io->taskio.tag_num :
585 			tmp_io->scsiio.tag_num,
586 		        tmp_io->io_hdr.flags,
587 			tmp_io->io_hdr.status);
588 	}
589 #endif
590 	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
591 	ctl_enqueue_isc((union ctl_io *)ctsio);
592 }
593 
594 /*
595  * ISC (Inter Shelf Communication) event handler.  Events from the HA
596  * subsystem come in here.
597  */
598 static void
599 ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
600 {
601 	struct ctl_softc *softc;
602 	union ctl_io *io;
603 	struct ctl_prio *presio;
604 	ctl_ha_status isc_status;
605 
606 	softc = control_softc;
607 	io = NULL;
608 
609 
610 #if 0
611 	printf("CTL: Isc Msg event %d\n", event);
612 #endif
613 	if (event == CTL_HA_EVT_MSG_RECV) {
614 		union ctl_ha_msg msg_info;
615 
616 		isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
617 					     sizeof(msg_info), /*wait*/ 0);
618 #if 0
619 		printf("CTL: msg_type %d\n", msg_info.msg_type);
620 #endif
621 		if (isc_status != 0) {
622 			printf("Error receiving message, status = %d\n",
623 			       isc_status);
624 			return;
625 		}
626 
627 		switch (msg_info.hdr.msg_type) {
628 		case CTL_MSG_SERIALIZE:
629 #if 0
630 			printf("Serialize\n");
631 #endif
632 			io = ctl_alloc_io_nowait(softc->othersc_pool);
633 			if (io == NULL) {
634 				printf("ctl_isc_event_handler: can't allocate "
635 				       "ctl_io!\n");
636 				/* Bad Juju */
637 				/* Need to set busy and send msg back */
638 				msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
639 				msg_info.hdr.status = CTL_SCSI_ERROR;
640 				msg_info.scsi.scsi_status = SCSI_STATUS_BUSY;
641 				msg_info.scsi.sense_len = 0;
642 			        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
643 				    sizeof(msg_info), 0) > CTL_HA_STATUS_SUCCESS){
644 				}
645 				goto bailout;
646 			}
647 			ctl_zero_io(io);
648 			// populate ctsio from msg_info
649 			io->io_hdr.io_type = CTL_IO_SCSI;
650 			io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
651 			io->io_hdr.original_sc = msg_info.hdr.original_sc;
652 #if 0
653 			printf("pOrig %x\n", (int)msg_info.original_sc);
654 #endif
655 			io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
656 					    CTL_FLAG_IO_ACTIVE;
657 			/*
658 			 * If we're in serialization-only mode, we don't
659 			 * want to go through full done processing.  Thus
660 			 * the COPY flag.
661 			 *
662 			 * XXX KDM add another flag that is more specific.
663 			 */
664 			if (softc->ha_mode == CTL_HA_MODE_SER_ONLY)
665 				io->io_hdr.flags |= CTL_FLAG_INT_COPY;
666 			io->io_hdr.nexus = msg_info.hdr.nexus;
667 #if 0
668 			printf("targ %d, port %d, iid %d, lun %d\n",
669 			       io->io_hdr.nexus.targ_target.id,
670 			       io->io_hdr.nexus.targ_port,
671 			       io->io_hdr.nexus.initid.id,
672 			       io->io_hdr.nexus.targ_lun);
673 #endif
674 			io->scsiio.tag_num = msg_info.scsi.tag_num;
675 			io->scsiio.tag_type = msg_info.scsi.tag_type;
676 			memcpy(io->scsiio.cdb, msg_info.scsi.cdb,
677 			       CTL_MAX_CDBLEN);
678 			if (softc->ha_mode == CTL_HA_MODE_XFER) {
679 				const struct ctl_cmd_entry *entry;
680 
681 				entry = ctl_get_cmd_entry(&io->scsiio, NULL);
682 				io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
683 				io->io_hdr.flags |=
684 					entry->flags & CTL_FLAG_DATA_MASK;
685 			}
686 			ctl_enqueue_isc(io);
687 			break;
688 
689 		/* Performed on the Originating SC, XFER mode only */
690 		case CTL_MSG_DATAMOVE: {
691 			struct ctl_sg_entry *sgl;
692 			int i, j;
693 
694 			io = msg_info.hdr.original_sc;
695 			if (io == NULL) {
696 				printf("%s: original_sc == NULL!\n", __func__);
697 				/* XXX KDM do something here */
698 				break;
699 			}
700 			io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
701 			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
702 			/*
703 			 * Keep track of this, we need to send it back over
704 			 * when the datamove is complete.
705 			 */
706 			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
707 
708 			if (msg_info.dt.sg_sequence == 0) {
709 				/*
710 				 * XXX KDM we use the preallocated S/G list
711 				 * here, but we'll need to change this to
712 				 * dynamic allocation if we need larger S/G
713 				 * lists.
714 				 */
715 				if (msg_info.dt.kern_sg_entries >
716 				    sizeof(io->io_hdr.remote_sglist) /
717 				    sizeof(io->io_hdr.remote_sglist[0])) {
718 					printf("%s: number of S/G entries "
719 					    "needed %u > allocated num %zd\n",
720 					    __func__,
721 					    msg_info.dt.kern_sg_entries,
722 					    sizeof(io->io_hdr.remote_sglist)/
723 					    sizeof(io->io_hdr.remote_sglist[0]));
724 
725 					/*
726 					 * XXX KDM send a message back to
727 					 * the other side to shut down the
728 					 * DMA.  The error will come back
729 					 * through via the normal channel.
730 					 */
731 					break;
732 				}
733 				sgl = io->io_hdr.remote_sglist;
734 				memset(sgl, 0,
735 				       sizeof(io->io_hdr.remote_sglist));
736 
737 				io->scsiio.kern_data_ptr = (uint8_t *)sgl;
738 
739 				io->scsiio.kern_sg_entries =
740 					msg_info.dt.kern_sg_entries;
741 				io->scsiio.rem_sg_entries =
742 					msg_info.dt.kern_sg_entries;
743 				io->scsiio.kern_data_len =
744 					msg_info.dt.kern_data_len;
745 				io->scsiio.kern_total_len =
746 					msg_info.dt.kern_total_len;
747 				io->scsiio.kern_data_resid =
748 					msg_info.dt.kern_data_resid;
749 				io->scsiio.kern_rel_offset =
750 					msg_info.dt.kern_rel_offset;
751 				/*
752 				 * Clear out per-DMA flags.
753 				 */
754 				io->io_hdr.flags &= ~CTL_FLAG_RDMA_MASK;
755 				/*
756 				 * Add per-DMA flags that are set for this
757 				 * particular DMA request.
758 				 */
759 				io->io_hdr.flags |= msg_info.dt.flags &
760 						    CTL_FLAG_RDMA_MASK;
761 			} else
762 				sgl = (struct ctl_sg_entry *)
763 					io->scsiio.kern_data_ptr;
764 
765 			for (i = msg_info.dt.sent_sg_entries, j = 0;
766 			     i < (msg_info.dt.sent_sg_entries +
767 			     msg_info.dt.cur_sg_entries); i++, j++) {
768 				sgl[i].addr = msg_info.dt.sg_list[j].addr;
769 				sgl[i].len = msg_info.dt.sg_list[j].len;
770 
771 #if 0
772 				printf("%s: L: %p,%d -> %p,%d j=%d, i=%d\n",
773 				       __func__,
774 				       msg_info.dt.sg_list[j].addr,
775 				       msg_info.dt.sg_list[j].len,
776 				       sgl[i].addr, sgl[i].len, j, i);
777 #endif
778 			}
779 #if 0
780 			memcpy(&sgl[msg_info.dt.sent_sg_entries],
781 			       msg_info.dt.sg_list,
782 			       sizeof(*sgl) * msg_info.dt.cur_sg_entries);
783 #endif
784 
785 			/*
786 			 * If this is the last piece of the I/O, we've got
787 			 * the full S/G list.  Queue processing in the thread.
788 			 * Otherwise wait for the next piece.
789 			 */
790 			if (msg_info.dt.sg_last != 0)
791 				ctl_enqueue_isc(io);
792 			break;
793 		}
794 		/* Performed on the Serializing (primary) SC, XFER mode only */
795 		case CTL_MSG_DATAMOVE_DONE: {
796 			if (msg_info.hdr.serializing_sc == NULL) {
797 				printf("%s: serializing_sc == NULL!\n",
798 				       __func__);
799 				/* XXX KDM now what? */
800 				break;
801 			}
802 			/*
803 			 * We grab the sense information here in case
804 			 * there was a failure, so we can return status
805 			 * back to the initiator.
806 			 */
807 			io = msg_info.hdr.serializing_sc;
808 			io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
809 			io->io_hdr.status = msg_info.hdr.status;
810 			io->scsiio.scsi_status = msg_info.scsi.scsi_status;
811 			io->scsiio.sense_len = msg_info.scsi.sense_len;
812 			io->scsiio.sense_residual =msg_info.scsi.sense_residual;
813 			io->io_hdr.port_status = msg_info.scsi.fetd_status;
814 			io->scsiio.residual = msg_info.scsi.residual;
815 			memcpy(&io->scsiio.sense_data,&msg_info.scsi.sense_data,
816 			       sizeof(io->scsiio.sense_data));
817 			ctl_enqueue_isc(io);
818 			break;
819 		}
820 
821 		/* Preformed on Originating SC, SER_ONLY mode */
822 		case CTL_MSG_R2R:
823 			io = msg_info.hdr.original_sc;
824 			if (io == NULL) {
825 				printf("%s: Major Bummer\n", __func__);
826 				return;
827 			} else {
828 #if 0
829 				printf("pOrig %x\n",(int) ctsio);
830 #endif
831 			}
832 			io->io_hdr.msg_type = CTL_MSG_R2R;
833 			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
834 			ctl_enqueue_isc(io);
835 			break;
836 
837 		/*
838 		 * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
839 		 * mode.
840 		 * Performed on the Originating (i.e. secondary) SC in XFER
841 		 * mode
842 		 */
843 		case CTL_MSG_FINISH_IO:
844 			if (softc->ha_mode == CTL_HA_MODE_XFER)
845 				ctl_isc_handler_finish_xfer(softc,
846 							    &msg_info);
847 			else
848 				ctl_isc_handler_finish_ser_only(softc,
849 								&msg_info);
850 			break;
851 
852 		/* Preformed on Originating SC */
853 		case CTL_MSG_BAD_JUJU:
854 			io = msg_info.hdr.original_sc;
855 			if (io == NULL) {
856 				printf("%s: Bad JUJU!, original_sc is NULL!\n",
857 				       __func__);
858 				break;
859 			}
860 			ctl_copy_sense_data(&msg_info, io);
861 			/*
862 			 * IO should have already been cleaned up on other
863 			 * SC so clear this flag so we won't send a message
864 			 * back to finish the IO there.
865 			 */
866 			io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
867 			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
868 
869 			/* io = msg_info.hdr.serializing_sc; */
870 			io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
871 			ctl_enqueue_isc(io);
872 			break;
873 
874 		/* Handle resets sent from the other side */
875 		case CTL_MSG_MANAGE_TASKS: {
876 			struct ctl_taskio *taskio;
877 			taskio = (struct ctl_taskio *)ctl_alloc_io_nowait(
878 			    softc->othersc_pool);
879 			if (taskio == NULL) {
880 				printf("ctl_isc_event_handler: can't allocate "
881 				       "ctl_io!\n");
882 				/* Bad Juju */
883 				/* should I just call the proper reset func
884 				   here??? */
885 				goto bailout;
886 			}
887 			ctl_zero_io((union ctl_io *)taskio);
888 			taskio->io_hdr.io_type = CTL_IO_TASK;
889 			taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
890 			taskio->io_hdr.nexus = msg_info.hdr.nexus;
891 			taskio->task_action = msg_info.task.task_action;
892 			taskio->tag_num = msg_info.task.tag_num;
893 			taskio->tag_type = msg_info.task.tag_type;
894 #ifdef CTL_TIME_IO
895 			taskio->io_hdr.start_time = time_uptime;
896 			getbintime(&taskio->io_hdr.start_bt);
897 #if 0
898 			cs_prof_gettime(&taskio->io_hdr.start_ticks);
899 #endif
900 #endif /* CTL_TIME_IO */
901 			ctl_run_task((union ctl_io *)taskio);
902 			break;
903 		}
904 		/* Persistent Reserve action which needs attention */
905 		case CTL_MSG_PERS_ACTION:
906 			presio = (struct ctl_prio *)ctl_alloc_io_nowait(
907 			    softc->othersc_pool);
908 			if (presio == NULL) {
909 				printf("ctl_isc_event_handler: can't allocate "
910 				       "ctl_io!\n");
911 				/* Bad Juju */
912 				/* Need to set busy and send msg back */
913 				goto bailout;
914 			}
915 			ctl_zero_io((union ctl_io *)presio);
916 			presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
917 			presio->pr_msg = msg_info.pr;
918 			ctl_enqueue_isc((union ctl_io *)presio);
919 			break;
920 		case CTL_MSG_SYNC_FE:
921 			rcv_sync_msg = 1;
922 			break;
923 		default:
924 		        printf("How did I get here?\n");
925 		}
926 	} else if (event == CTL_HA_EVT_MSG_SENT) {
927 		if (param != CTL_HA_STATUS_SUCCESS) {
928 			printf("Bad status from ctl_ha_msg_send status %d\n",
929 			       param);
930 		}
931 		return;
932 	} else if (event == CTL_HA_EVT_DISCONNECT) {
933 		printf("CTL: Got a disconnect from Isc\n");
934 		return;
935 	} else {
936 		printf("ctl_isc_event_handler: Unknown event %d\n", event);
937 		return;
938 	}
939 
940 bailout:
941 	return;
942 }
943 
944 static void
945 ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
946 {
947 	struct scsi_sense_data *sense;
948 
949 	sense = &dest->scsiio.sense_data;
950 	bcopy(&src->scsi.sense_data, sense, sizeof(*sense));
951 	dest->scsiio.scsi_status = src->scsi.scsi_status;
952 	dest->scsiio.sense_len = src->scsi.sense_len;
953 	dest->io_hdr.status = src->hdr.status;
954 }
955 #endif
956 
957 static void
958 ctl_est_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
959 {
960 	ctl_ua_type *pu;
961 
962 	mtx_assert(&lun->lun_lock, MA_OWNED);
963 	pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
964 	if (pu == NULL)
965 		return;
966 	pu[initidx % CTL_MAX_INIT_PER_PORT] |= ua;
967 }
968 
969 static void
970 ctl_est_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
971 {
972 	int i, j;
973 
974 	mtx_assert(&lun->lun_lock, MA_OWNED);
975 	for (i = 0; i < CTL_MAX_PORTS; i++) {
976 		if (lun->pending_ua[i] == NULL)
977 			continue;
978 		for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
979 			if (i * CTL_MAX_INIT_PER_PORT + j == except)
980 				continue;
981 			lun->pending_ua[i][j] |= ua;
982 		}
983 	}
984 }
985 
986 static void
987 ctl_clr_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
988 {
989 	ctl_ua_type *pu;
990 
991 	mtx_assert(&lun->lun_lock, MA_OWNED);
992 	pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
993 	if (pu == NULL)
994 		return;
995 	pu[initidx % CTL_MAX_INIT_PER_PORT] &= ~ua;
996 }
997 
998 static void
999 ctl_clr_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1000 {
1001 	int i, j;
1002 
1003 	mtx_assert(&lun->lun_lock, MA_OWNED);
1004 	for (i = 0; i < CTL_MAX_PORTS; i++) {
1005 		if (lun->pending_ua[i] == NULL)
1006 			continue;
1007 		for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
1008 			if (i * CTL_MAX_INIT_PER_PORT + j == except)
1009 				continue;
1010 			lun->pending_ua[i][j] &= ~ua;
1011 		}
1012 	}
1013 }
1014 
1015 static int
1016 ctl_ha_state_sysctl(SYSCTL_HANDLER_ARGS)
1017 {
1018 	struct ctl_softc *softc = (struct ctl_softc *)arg1;
1019 	struct ctl_lun *lun;
1020 	int error, value;
1021 
1022 	if (softc->flags & CTL_FLAG_ACTIVE_SHELF)
1023 		value = 0;
1024 	else
1025 		value = 1;
1026 
1027 	error = sysctl_handle_int(oidp, &value, 0, req);
1028 	if ((error != 0) || (req->newptr == NULL))
1029 		return (error);
1030 
1031 	mtx_lock(&softc->ctl_lock);
1032 	if (value == 0)
1033 		softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1034 	else
1035 		softc->flags &= ~CTL_FLAG_ACTIVE_SHELF;
1036 	STAILQ_FOREACH(lun, &softc->lun_list, links) {
1037 		mtx_lock(&lun->lun_lock);
1038 		ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
1039 		mtx_unlock(&lun->lun_lock);
1040 	}
1041 	mtx_unlock(&softc->ctl_lock);
1042 	return (0);
1043 }
1044 
1045 static int
1046 ctl_init(void)
1047 {
1048 	struct ctl_softc *softc;
1049 	void *other_pool;
1050 	int i, error, retval;
1051 	//int isc_retval;
1052 
1053 	retval = 0;
1054 	ctl_pause_rtr = 0;
1055         rcv_sync_msg = 0;
1056 
1057 	control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
1058 			       M_WAITOK | M_ZERO);
1059 	softc = control_softc;
1060 
1061 	softc->dev = make_dev(&ctl_cdevsw, 0, UID_ROOT, GID_OPERATOR, 0600,
1062 			      "cam/ctl");
1063 
1064 	softc->dev->si_drv1 = softc;
1065 
1066 	/*
1067 	 * By default, return a "bad LUN" peripheral qualifier for unknown
1068 	 * LUNs.  The user can override this default using the tunable or
1069 	 * sysctl.  See the comment in ctl_inquiry_std() for more details.
1070 	 */
1071 	softc->inquiry_pq_no_lun = 1;
1072 	TUNABLE_INT_FETCH("kern.cam.ctl.inquiry_pq_no_lun",
1073 			  &softc->inquiry_pq_no_lun);
1074 	sysctl_ctx_init(&softc->sysctl_ctx);
1075 	softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
1076 		SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
1077 		CTLFLAG_RD, 0, "CAM Target Layer");
1078 
1079 	if (softc->sysctl_tree == NULL) {
1080 		printf("%s: unable to allocate sysctl tree\n", __func__);
1081 		destroy_dev(softc->dev);
1082 		free(control_softc, M_DEVBUF);
1083 		control_softc = NULL;
1084 		return (ENOMEM);
1085 	}
1086 
1087 	SYSCTL_ADD_INT(&softc->sysctl_ctx,
1088 		       SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO,
1089 		       "inquiry_pq_no_lun", CTLFLAG_RW,
1090 		       &softc->inquiry_pq_no_lun, 0,
1091 		       "Report no lun possible for invalid LUNs");
1092 
1093 	mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
1094 	softc->io_zone = uma_zcreate("CTL IO", sizeof(union ctl_io),
1095 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1096 	softc->open_count = 0;
1097 
1098 	/*
1099 	 * Default to actually sending a SYNCHRONIZE CACHE command down to
1100 	 * the drive.
1101 	 */
1102 	softc->flags = CTL_FLAG_REAL_SYNC;
1103 
1104 	/*
1105 	 * In Copan's HA scheme, the "master" and "slave" roles are
1106 	 * figured out through the slot the controller is in.  Although it
1107 	 * is an active/active system, someone has to be in charge.
1108 	 */
1109 	SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1110 	    OID_AUTO, "ha_id", CTLFLAG_RDTUN, &softc->ha_id, 0,
1111 	    "HA head ID (0 - no HA)");
1112 	if (softc->ha_id == 0) {
1113 		softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1114 		softc->is_single = 1;
1115 		softc->port_offset = 0;
1116 	} else
1117 		softc->port_offset = (softc->ha_id - 1) * CTL_MAX_PORTS;
1118 	softc->persis_offset = softc->port_offset * CTL_MAX_INIT_PER_PORT;
1119 
1120 	STAILQ_INIT(&softc->lun_list);
1121 	STAILQ_INIT(&softc->pending_lun_queue);
1122 	STAILQ_INIT(&softc->fe_list);
1123 	STAILQ_INIT(&softc->port_list);
1124 	STAILQ_INIT(&softc->be_list);
1125 	ctl_tpc_init(softc);
1126 
1127 	if (ctl_pool_create(softc, "othersc", CTL_POOL_ENTRIES_OTHER_SC,
1128 	                    &other_pool) != 0)
1129 	{
1130 		printf("ctl: can't allocate %d entry other SC pool, "
1131 		       "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1132 		return (ENOMEM);
1133 	}
1134 	softc->othersc_pool = other_pool;
1135 
1136 	if (worker_threads <= 0)
1137 		worker_threads = max(1, mp_ncpus / 4);
1138 	if (worker_threads > CTL_MAX_THREADS)
1139 		worker_threads = CTL_MAX_THREADS;
1140 
1141 	for (i = 0; i < worker_threads; i++) {
1142 		struct ctl_thread *thr = &softc->threads[i];
1143 
1144 		mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1145 		thr->ctl_softc = softc;
1146 		STAILQ_INIT(&thr->incoming_queue);
1147 		STAILQ_INIT(&thr->rtr_queue);
1148 		STAILQ_INIT(&thr->done_queue);
1149 		STAILQ_INIT(&thr->isc_queue);
1150 
1151 		error = kproc_kthread_add(ctl_work_thread, thr,
1152 		    &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1153 		if (error != 0) {
1154 			printf("error creating CTL work thread!\n");
1155 			ctl_pool_free(other_pool);
1156 			return (error);
1157 		}
1158 	}
1159 	error = kproc_kthread_add(ctl_lun_thread, softc,
1160 	    &softc->ctl_proc, NULL, 0, 0, "ctl", "lun");
1161 	if (error != 0) {
1162 		printf("error creating CTL lun thread!\n");
1163 		ctl_pool_free(other_pool);
1164 		return (error);
1165 	}
1166 	error = kproc_kthread_add(ctl_thresh_thread, softc,
1167 	    &softc->ctl_proc, NULL, 0, 0, "ctl", "thresh");
1168 	if (error != 0) {
1169 		printf("error creating CTL threshold thread!\n");
1170 		ctl_pool_free(other_pool);
1171 		return (error);
1172 	}
1173 
1174 	SYSCTL_ADD_PROC(&softc->sysctl_ctx,SYSCTL_CHILDREN(softc->sysctl_tree),
1175 	    OID_AUTO, "ha_state", CTLTYPE_INT | CTLFLAG_RWTUN,
1176 	    softc, 0, ctl_ha_state_sysctl, "I", "HA state for this head");
1177 
1178 #ifdef CTL_IO_DELAY
1179 	if (sizeof(struct callout) > CTL_TIMER_BYTES) {
1180 		printf("sizeof(struct callout) %zd > CTL_TIMER_BYTES %zd\n",
1181 		       sizeof(struct callout), CTL_TIMER_BYTES);
1182 		return (EINVAL);
1183 	}
1184 #endif /* CTL_IO_DELAY */
1185 
1186 	return (0);
1187 }
1188 
1189 void
1190 ctl_shutdown(void)
1191 {
1192 	struct ctl_softc *softc;
1193 	struct ctl_lun *lun, *next_lun;
1194 
1195 	softc = (struct ctl_softc *)control_softc;
1196 
1197 	mtx_lock(&softc->ctl_lock);
1198 
1199 	/*
1200 	 * Free up each LUN.
1201 	 */
1202 	for (lun = STAILQ_FIRST(&softc->lun_list); lun != NULL; lun = next_lun){
1203 		next_lun = STAILQ_NEXT(lun, links);
1204 		ctl_free_lun(lun);
1205 	}
1206 
1207 	mtx_unlock(&softc->ctl_lock);
1208 
1209 #if 0
1210 	ctl_shutdown_thread(softc->work_thread);
1211 	mtx_destroy(&softc->queue_lock);
1212 #endif
1213 
1214 	ctl_tpc_shutdown(softc);
1215 	uma_zdestroy(softc->io_zone);
1216 	mtx_destroy(&softc->ctl_lock);
1217 
1218 	destroy_dev(softc->dev);
1219 
1220 	sysctl_ctx_free(&softc->sysctl_ctx);
1221 
1222 	free(control_softc, M_DEVBUF);
1223 	control_softc = NULL;
1224 }
1225 
1226 static int
1227 ctl_module_event_handler(module_t mod, int what, void *arg)
1228 {
1229 
1230 	switch (what) {
1231 	case MOD_LOAD:
1232 		return (ctl_init());
1233 	case MOD_UNLOAD:
1234 		return (EBUSY);
1235 	default:
1236 		return (EOPNOTSUPP);
1237 	}
1238 }
1239 
1240 /*
1241  * XXX KDM should we do some access checks here?  Bump a reference count to
1242  * prevent a CTL module from being unloaded while someone has it open?
1243  */
1244 static int
1245 ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1246 {
1247 	return (0);
1248 }
1249 
1250 static int
1251 ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1252 {
1253 	return (0);
1254 }
1255 
1256 int
1257 ctl_port_enable(ctl_port_type port_type)
1258 {
1259 	struct ctl_softc *softc = control_softc;
1260 	struct ctl_port *port;
1261 
1262 	if (softc->is_single == 0) {
1263 		union ctl_ha_msg msg_info;
1264 		int isc_retval;
1265 
1266 #if 0
1267 		printf("%s: HA mode, synchronizing frontend enable\n",
1268 		        __func__);
1269 #endif
1270 		msg_info.hdr.msg_type = CTL_MSG_SYNC_FE;
1271 	        if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1272 		        sizeof(msg_info), 1 )) > CTL_HA_STATUS_SUCCESS) {
1273 			printf("Sync msg send error retval %d\n", isc_retval);
1274 		}
1275 		if (!rcv_sync_msg) {
1276 			isc_retval=ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
1277 			        sizeof(msg_info), 1);
1278 		}
1279 #if 0
1280         	printf("CTL:Frontend Enable\n");
1281 	} else {
1282 		printf("%s: single mode, skipping frontend synchronization\n",
1283 		        __func__);
1284 #endif
1285 	}
1286 
1287 	STAILQ_FOREACH(port, &softc->port_list, links) {
1288 		if (port_type & port->port_type)
1289 		{
1290 #if 0
1291 			printf("port %d\n", port->targ_port);
1292 #endif
1293 			ctl_port_online(port);
1294 		}
1295 	}
1296 
1297 	return (0);
1298 }
1299 
1300 int
1301 ctl_port_disable(ctl_port_type port_type)
1302 {
1303 	struct ctl_softc *softc;
1304 	struct ctl_port *port;
1305 
1306 	softc = control_softc;
1307 
1308 	STAILQ_FOREACH(port, &softc->port_list, links) {
1309 		if (port_type & port->port_type)
1310 			ctl_port_offline(port);
1311 	}
1312 
1313 	return (0);
1314 }
1315 
1316 /*
1317  * Returns 0 for success, 1 for failure.
1318  * Currently the only failure mode is if there aren't enough entries
1319  * allocated.  So, in case of a failure, look at num_entries_dropped,
1320  * reallocate and try again.
1321  */
1322 int
1323 ctl_port_list(struct ctl_port_entry *entries, int num_entries_alloced,
1324 	      int *num_entries_filled, int *num_entries_dropped,
1325 	      ctl_port_type port_type, int no_virtual)
1326 {
1327 	struct ctl_softc *softc;
1328 	struct ctl_port *port;
1329 	int entries_dropped, entries_filled;
1330 	int retval;
1331 	int i;
1332 
1333 	softc = control_softc;
1334 
1335 	retval = 0;
1336 	entries_filled = 0;
1337 	entries_dropped = 0;
1338 
1339 	i = 0;
1340 	mtx_lock(&softc->ctl_lock);
1341 	STAILQ_FOREACH(port, &softc->port_list, links) {
1342 		struct ctl_port_entry *entry;
1343 
1344 		if ((port->port_type & port_type) == 0)
1345 			continue;
1346 
1347 		if ((no_virtual != 0)
1348 		 && (port->virtual_port != 0))
1349 			continue;
1350 
1351 		if (entries_filled >= num_entries_alloced) {
1352 			entries_dropped++;
1353 			continue;
1354 		}
1355 		entry = &entries[i];
1356 
1357 		entry->port_type = port->port_type;
1358 		strlcpy(entry->port_name, port->port_name,
1359 			sizeof(entry->port_name));
1360 		entry->physical_port = port->physical_port;
1361 		entry->virtual_port = port->virtual_port;
1362 		entry->wwnn = port->wwnn;
1363 		entry->wwpn = port->wwpn;
1364 
1365 		i++;
1366 		entries_filled++;
1367 	}
1368 
1369 	mtx_unlock(&softc->ctl_lock);
1370 
1371 	if (entries_dropped > 0)
1372 		retval = 1;
1373 
1374 	*num_entries_dropped = entries_dropped;
1375 	*num_entries_filled = entries_filled;
1376 
1377 	return (retval);
1378 }
1379 
1380 /*
1381  * Remove an initiator by port number and initiator ID.
1382  * Returns 0 for success, -1 for failure.
1383  */
1384 int
1385 ctl_remove_initiator(struct ctl_port *port, int iid)
1386 {
1387 	struct ctl_softc *softc = control_softc;
1388 
1389 	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1390 
1391 	if (iid > CTL_MAX_INIT_PER_PORT) {
1392 		printf("%s: initiator ID %u > maximun %u!\n",
1393 		       __func__, iid, CTL_MAX_INIT_PER_PORT);
1394 		return (-1);
1395 	}
1396 
1397 	mtx_lock(&softc->ctl_lock);
1398 	port->wwpn_iid[iid].in_use--;
1399 	port->wwpn_iid[iid].last_use = time_uptime;
1400 	mtx_unlock(&softc->ctl_lock);
1401 
1402 	return (0);
1403 }
1404 
1405 /*
1406  * Add an initiator to the initiator map.
1407  * Returns iid for success, < 0 for failure.
1408  */
1409 int
1410 ctl_add_initiator(struct ctl_port *port, int iid, uint64_t wwpn, char *name)
1411 {
1412 	struct ctl_softc *softc = control_softc;
1413 	time_t best_time;
1414 	int i, best;
1415 
1416 	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1417 
1418 	if (iid >= CTL_MAX_INIT_PER_PORT) {
1419 		printf("%s: WWPN %#jx initiator ID %u > maximum %u!\n",
1420 		       __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
1421 		free(name, M_CTL);
1422 		return (-1);
1423 	}
1424 
1425 	mtx_lock(&softc->ctl_lock);
1426 
1427 	if (iid < 0 && (wwpn != 0 || name != NULL)) {
1428 		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1429 			if (wwpn != 0 && wwpn == port->wwpn_iid[i].wwpn) {
1430 				iid = i;
1431 				break;
1432 			}
1433 			if (name != NULL && port->wwpn_iid[i].name != NULL &&
1434 			    strcmp(name, port->wwpn_iid[i].name) == 0) {
1435 				iid = i;
1436 				break;
1437 			}
1438 		}
1439 	}
1440 
1441 	if (iid < 0) {
1442 		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1443 			if (port->wwpn_iid[i].in_use == 0 &&
1444 			    port->wwpn_iid[i].wwpn == 0 &&
1445 			    port->wwpn_iid[i].name == NULL) {
1446 				iid = i;
1447 				break;
1448 			}
1449 		}
1450 	}
1451 
1452 	if (iid < 0) {
1453 		best = -1;
1454 		best_time = INT32_MAX;
1455 		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1456 			if (port->wwpn_iid[i].in_use == 0) {
1457 				if (port->wwpn_iid[i].last_use < best_time) {
1458 					best = i;
1459 					best_time = port->wwpn_iid[i].last_use;
1460 				}
1461 			}
1462 		}
1463 		iid = best;
1464 	}
1465 
1466 	if (iid < 0) {
1467 		mtx_unlock(&softc->ctl_lock);
1468 		free(name, M_CTL);
1469 		return (-2);
1470 	}
1471 
1472 	if (port->wwpn_iid[iid].in_use > 0 && (wwpn != 0 || name != NULL)) {
1473 		/*
1474 		 * This is not an error yet.
1475 		 */
1476 		if (wwpn != 0 && wwpn == port->wwpn_iid[iid].wwpn) {
1477 #if 0
1478 			printf("%s: port %d iid %u WWPN %#jx arrived"
1479 			    " again\n", __func__, port->targ_port,
1480 			    iid, (uintmax_t)wwpn);
1481 #endif
1482 			goto take;
1483 		}
1484 		if (name != NULL && port->wwpn_iid[iid].name != NULL &&
1485 		    strcmp(name, port->wwpn_iid[iid].name) == 0) {
1486 #if 0
1487 			printf("%s: port %d iid %u name '%s' arrived"
1488 			    " again\n", __func__, port->targ_port,
1489 			    iid, name);
1490 #endif
1491 			goto take;
1492 		}
1493 
1494 		/*
1495 		 * This is an error, but what do we do about it?  The
1496 		 * driver is telling us we have a new WWPN for this
1497 		 * initiator ID, so we pretty much need to use it.
1498 		 */
1499 		printf("%s: port %d iid %u WWPN %#jx '%s' arrived,"
1500 		    " but WWPN %#jx '%s' is still at that address\n",
1501 		    __func__, port->targ_port, iid, wwpn, name,
1502 		    (uintmax_t)port->wwpn_iid[iid].wwpn,
1503 		    port->wwpn_iid[iid].name);
1504 
1505 		/*
1506 		 * XXX KDM clear have_ca and ua_pending on each LUN for
1507 		 * this initiator.
1508 		 */
1509 	}
1510 take:
1511 	free(port->wwpn_iid[iid].name, M_CTL);
1512 	port->wwpn_iid[iid].name = name;
1513 	port->wwpn_iid[iid].wwpn = wwpn;
1514 	port->wwpn_iid[iid].in_use++;
1515 	mtx_unlock(&softc->ctl_lock);
1516 
1517 	return (iid);
1518 }
1519 
1520 static int
1521 ctl_create_iid(struct ctl_port *port, int iid, uint8_t *buf)
1522 {
1523 	int len;
1524 
1525 	switch (port->port_type) {
1526 	case CTL_PORT_FC:
1527 	{
1528 		struct scsi_transportid_fcp *id =
1529 		    (struct scsi_transportid_fcp *)buf;
1530 		if (port->wwpn_iid[iid].wwpn == 0)
1531 			return (0);
1532 		memset(id, 0, sizeof(*id));
1533 		id->format_protocol = SCSI_PROTO_FC;
1534 		scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->n_port_name);
1535 		return (sizeof(*id));
1536 	}
1537 	case CTL_PORT_ISCSI:
1538 	{
1539 		struct scsi_transportid_iscsi_port *id =
1540 		    (struct scsi_transportid_iscsi_port *)buf;
1541 		if (port->wwpn_iid[iid].name == NULL)
1542 			return (0);
1543 		memset(id, 0, 256);
1544 		id->format_protocol = SCSI_TRN_ISCSI_FORMAT_PORT |
1545 		    SCSI_PROTO_ISCSI;
1546 		len = strlcpy(id->iscsi_name, port->wwpn_iid[iid].name, 252) + 1;
1547 		len = roundup2(min(len, 252), 4);
1548 		scsi_ulto2b(len, id->additional_length);
1549 		return (sizeof(*id) + len);
1550 	}
1551 	case CTL_PORT_SAS:
1552 	{
1553 		struct scsi_transportid_sas *id =
1554 		    (struct scsi_transportid_sas *)buf;
1555 		if (port->wwpn_iid[iid].wwpn == 0)
1556 			return (0);
1557 		memset(id, 0, sizeof(*id));
1558 		id->format_protocol = SCSI_PROTO_SAS;
1559 		scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->sas_address);
1560 		return (sizeof(*id));
1561 	}
1562 	default:
1563 	{
1564 		struct scsi_transportid_spi *id =
1565 		    (struct scsi_transportid_spi *)buf;
1566 		memset(id, 0, sizeof(*id));
1567 		id->format_protocol = SCSI_PROTO_SPI;
1568 		scsi_ulto2b(iid, id->scsi_addr);
1569 		scsi_ulto2b(port->targ_port, id->rel_trgt_port_id);
1570 		return (sizeof(*id));
1571 	}
1572 	}
1573 }
1574 
1575 /*
1576  * Serialize a command that went down the "wrong" side, and so was sent to
1577  * this controller for execution.  The logic is a little different than the
1578  * standard case in ctl_scsiio_precheck().  Errors in this case need to get
1579  * sent back to the other side, but in the success case, we execute the
1580  * command on this side (XFER mode) or tell the other side to execute it
1581  * (SER_ONLY mode).
1582  */
1583 static int
1584 ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
1585 {
1586 	struct ctl_softc *softc;
1587 	union ctl_ha_msg msg_info;
1588 	struct ctl_lun *lun;
1589 	int retval = 0;
1590 	uint32_t targ_lun;
1591 
1592 	softc = control_softc;
1593 
1594 	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
1595 	lun = softc->ctl_luns[targ_lun];
1596 	if (lun==NULL)
1597 	{
1598 		/*
1599 		 * Why isn't LUN defined? The other side wouldn't
1600 		 * send a cmd if the LUN is undefined.
1601 		 */
1602 		printf("%s: Bad JUJU!, LUN is NULL!\n", __func__);
1603 
1604 		/* "Logical unit not supported" */
1605 		ctl_set_sense_data(&msg_info.scsi.sense_data,
1606 				   lun,
1607 				   /*sense_format*/SSD_TYPE_NONE,
1608 				   /*current_error*/ 1,
1609 				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1610 				   /*asc*/ 0x25,
1611 				   /*ascq*/ 0x00,
1612 				   SSD_ELEM_NONE);
1613 
1614 		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1615 		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1616 		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1617 		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1618 		msg_info.hdr.serializing_sc = NULL;
1619 		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1620 	        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1621 				sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1622 		}
1623 		return(1);
1624 
1625 	}
1626 
1627 	mtx_lock(&lun->lun_lock);
1628     	TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1629 
1630 	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
1631 		(union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
1632 		 ooa_links))) {
1633 	case CTL_ACTION_BLOCK:
1634 		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
1635 		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
1636 				  blocked_links);
1637 		break;
1638 	case CTL_ACTION_PASS:
1639 	case CTL_ACTION_SKIP:
1640 		if (softc->ha_mode == CTL_HA_MODE_XFER) {
1641 			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
1642 			ctl_enqueue_rtr((union ctl_io *)ctsio);
1643 		} else {
1644 
1645 			/* send msg back to other side */
1646 			msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1647 			msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
1648 			msg_info.hdr.msg_type = CTL_MSG_R2R;
1649 #if 0
1650 			printf("2. pOrig %x\n", (int)msg_info.hdr.original_sc);
1651 #endif
1652 		        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1653 			    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1654 			}
1655 		}
1656 		break;
1657 	case CTL_ACTION_OVERLAP:
1658 		/* OVERLAPPED COMMANDS ATTEMPTED */
1659 		ctl_set_sense_data(&msg_info.scsi.sense_data,
1660 				   lun,
1661 				   /*sense_format*/SSD_TYPE_NONE,
1662 				   /*current_error*/ 1,
1663 				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1664 				   /*asc*/ 0x4E,
1665 				   /*ascq*/ 0x00,
1666 				   SSD_ELEM_NONE);
1667 
1668 		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1669 		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1670 		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1671 		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1672 		msg_info.hdr.serializing_sc = NULL;
1673 		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1674 #if 0
1675 		printf("BAD JUJU:Major Bummer Overlap\n");
1676 #endif
1677 		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1678 		retval = 1;
1679 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1680 		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1681 		}
1682 		break;
1683 	case CTL_ACTION_OVERLAP_TAG:
1684 		/* TAGGED OVERLAPPED COMMANDS (NN = QUEUE TAG) */
1685 		ctl_set_sense_data(&msg_info.scsi.sense_data,
1686 				   lun,
1687 				   /*sense_format*/SSD_TYPE_NONE,
1688 				   /*current_error*/ 1,
1689 				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1690 				   /*asc*/ 0x4D,
1691 				   /*ascq*/ ctsio->tag_num & 0xff,
1692 				   SSD_ELEM_NONE);
1693 
1694 		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1695 		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1696 		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1697 		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1698 		msg_info.hdr.serializing_sc = NULL;
1699 		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1700 #if 0
1701 		printf("BAD JUJU:Major Bummer Overlap Tag\n");
1702 #endif
1703 		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1704 		retval = 1;
1705 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1706 		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1707 		}
1708 		break;
1709 	case CTL_ACTION_ERROR:
1710 	default:
1711 		/* "Internal target failure" */
1712 		ctl_set_sense_data(&msg_info.scsi.sense_data,
1713 				   lun,
1714 				   /*sense_format*/SSD_TYPE_NONE,
1715 				   /*current_error*/ 1,
1716 				   /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
1717 				   /*asc*/ 0x44,
1718 				   /*ascq*/ 0x00,
1719 				   SSD_ELEM_NONE);
1720 
1721 		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1722 		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1723 		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1724 		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1725 		msg_info.hdr.serializing_sc = NULL;
1726 		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1727 #if 0
1728 		printf("BAD JUJU:Major Bummer HW Error\n");
1729 #endif
1730 		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1731 		retval = 1;
1732 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1733 		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1734 		}
1735 		break;
1736 	}
1737 	mtx_unlock(&lun->lun_lock);
1738 	return (retval);
1739 }
1740 
1741 /*
1742  * Returns 0 for success, errno for failure.
1743  */
1744 static int
1745 ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
1746 		   struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
1747 {
1748 	union ctl_io *io;
1749 	int retval;
1750 
1751 	retval = 0;
1752 
1753 	mtx_lock(&lun->lun_lock);
1754 	for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
1755 	     (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
1756 	     ooa_links)) {
1757 		struct ctl_ooa_entry *entry;
1758 
1759 		/*
1760 		 * If we've got more than we can fit, just count the
1761 		 * remaining entries.
1762 		 */
1763 		if (*cur_fill_num >= ooa_hdr->alloc_num)
1764 			continue;
1765 
1766 		entry = &kern_entries[*cur_fill_num];
1767 
1768 		entry->tag_num = io->scsiio.tag_num;
1769 		entry->lun_num = lun->lun;
1770 #ifdef CTL_TIME_IO
1771 		entry->start_bt = io->io_hdr.start_bt;
1772 #endif
1773 		bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
1774 		entry->cdb_len = io->scsiio.cdb_len;
1775 		if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
1776 			entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
1777 
1778 		if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
1779 			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
1780 
1781 		if (io->io_hdr.flags & CTL_FLAG_ABORT)
1782 			entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
1783 
1784 		if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
1785 			entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
1786 
1787 		if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
1788 			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
1789 	}
1790 	mtx_unlock(&lun->lun_lock);
1791 
1792 	return (retval);
1793 }
1794 
1795 static void *
1796 ctl_copyin_alloc(void *user_addr, int len, char *error_str,
1797 		 size_t error_str_len)
1798 {
1799 	void *kptr;
1800 
1801 	kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
1802 
1803 	if (copyin(user_addr, kptr, len) != 0) {
1804 		snprintf(error_str, error_str_len, "Error copying %d bytes "
1805 			 "from user address %p to kernel address %p", len,
1806 			 user_addr, kptr);
1807 		free(kptr, M_CTL);
1808 		return (NULL);
1809 	}
1810 
1811 	return (kptr);
1812 }
1813 
1814 static void
1815 ctl_free_args(int num_args, struct ctl_be_arg *args)
1816 {
1817 	int i;
1818 
1819 	if (args == NULL)
1820 		return;
1821 
1822 	for (i = 0; i < num_args; i++) {
1823 		free(args[i].kname, M_CTL);
1824 		free(args[i].kvalue, M_CTL);
1825 	}
1826 
1827 	free(args, M_CTL);
1828 }
1829 
1830 static struct ctl_be_arg *
1831 ctl_copyin_args(int num_args, struct ctl_be_arg *uargs,
1832 		char *error_str, size_t error_str_len)
1833 {
1834 	struct ctl_be_arg *args;
1835 	int i;
1836 
1837 	args = ctl_copyin_alloc(uargs, num_args * sizeof(*args),
1838 				error_str, error_str_len);
1839 
1840 	if (args == NULL)
1841 		goto bailout;
1842 
1843 	for (i = 0; i < num_args; i++) {
1844 		args[i].kname = NULL;
1845 		args[i].kvalue = NULL;
1846 	}
1847 
1848 	for (i = 0; i < num_args; i++) {
1849 		uint8_t *tmpptr;
1850 
1851 		args[i].kname = ctl_copyin_alloc(args[i].name,
1852 			args[i].namelen, error_str, error_str_len);
1853 		if (args[i].kname == NULL)
1854 			goto bailout;
1855 
1856 		if (args[i].kname[args[i].namelen - 1] != '\0') {
1857 			snprintf(error_str, error_str_len, "Argument %d "
1858 				 "name is not NUL-terminated", i);
1859 			goto bailout;
1860 		}
1861 
1862 		if (args[i].flags & CTL_BEARG_RD) {
1863 			tmpptr = ctl_copyin_alloc(args[i].value,
1864 				args[i].vallen, error_str, error_str_len);
1865 			if (tmpptr == NULL)
1866 				goto bailout;
1867 			if ((args[i].flags & CTL_BEARG_ASCII)
1868 			 && (tmpptr[args[i].vallen - 1] != '\0')) {
1869 				snprintf(error_str, error_str_len, "Argument "
1870 				    "%d value is not NUL-terminated", i);
1871 				goto bailout;
1872 			}
1873 			args[i].kvalue = tmpptr;
1874 		} else {
1875 			args[i].kvalue = malloc(args[i].vallen,
1876 			    M_CTL, M_WAITOK | M_ZERO);
1877 		}
1878 	}
1879 
1880 	return (args);
1881 bailout:
1882 
1883 	ctl_free_args(num_args, args);
1884 
1885 	return (NULL);
1886 }
1887 
1888 static void
1889 ctl_copyout_args(int num_args, struct ctl_be_arg *args)
1890 {
1891 	int i;
1892 
1893 	for (i = 0; i < num_args; i++) {
1894 		if (args[i].flags & CTL_BEARG_WR)
1895 			copyout(args[i].kvalue, args[i].value, args[i].vallen);
1896 	}
1897 }
1898 
1899 /*
1900  * Escape characters that are illegal or not recommended in XML.
1901  */
1902 int
1903 ctl_sbuf_printf_esc(struct sbuf *sb, char *str, int size)
1904 {
1905 	char *end = str + size;
1906 	int retval;
1907 
1908 	retval = 0;
1909 
1910 	for (; *str && str < end; str++) {
1911 		switch (*str) {
1912 		case '&':
1913 			retval = sbuf_printf(sb, "&amp;");
1914 			break;
1915 		case '>':
1916 			retval = sbuf_printf(sb, "&gt;");
1917 			break;
1918 		case '<':
1919 			retval = sbuf_printf(sb, "&lt;");
1920 			break;
1921 		default:
1922 			retval = sbuf_putc(sb, *str);
1923 			break;
1924 		}
1925 
1926 		if (retval != 0)
1927 			break;
1928 
1929 	}
1930 
1931 	return (retval);
1932 }
1933 
1934 static void
1935 ctl_id_sbuf(struct ctl_devid *id, struct sbuf *sb)
1936 {
1937 	struct scsi_vpd_id_descriptor *desc;
1938 	int i;
1939 
1940 	if (id == NULL || id->len < 4)
1941 		return;
1942 	desc = (struct scsi_vpd_id_descriptor *)id->data;
1943 	switch (desc->id_type & SVPD_ID_TYPE_MASK) {
1944 	case SVPD_ID_TYPE_T10:
1945 		sbuf_printf(sb, "t10.");
1946 		break;
1947 	case SVPD_ID_TYPE_EUI64:
1948 		sbuf_printf(sb, "eui.");
1949 		break;
1950 	case SVPD_ID_TYPE_NAA:
1951 		sbuf_printf(sb, "naa.");
1952 		break;
1953 	case SVPD_ID_TYPE_SCSI_NAME:
1954 		break;
1955 	}
1956 	switch (desc->proto_codeset & SVPD_ID_CODESET_MASK) {
1957 	case SVPD_ID_CODESET_BINARY:
1958 		for (i = 0; i < desc->length; i++)
1959 			sbuf_printf(sb, "%02x", desc->identifier[i]);
1960 		break;
1961 	case SVPD_ID_CODESET_ASCII:
1962 		sbuf_printf(sb, "%.*s", (int)desc->length,
1963 		    (char *)desc->identifier);
1964 		break;
1965 	case SVPD_ID_CODESET_UTF8:
1966 		sbuf_printf(sb, "%s", (char *)desc->identifier);
1967 		break;
1968 	}
1969 }
1970 
1971 static int
1972 ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
1973 	  struct thread *td)
1974 {
1975 	struct ctl_softc *softc;
1976 	int retval;
1977 
1978 	softc = control_softc;
1979 
1980 	retval = 0;
1981 
1982 	switch (cmd) {
1983 	case CTL_IO:
1984 		retval = ctl_ioctl_io(dev, cmd, addr, flag, td);
1985 		break;
1986 	case CTL_ENABLE_PORT:
1987 	case CTL_DISABLE_PORT:
1988 	case CTL_SET_PORT_WWNS: {
1989 		struct ctl_port *port;
1990 		struct ctl_port_entry *entry;
1991 
1992 		entry = (struct ctl_port_entry *)addr;
1993 
1994 		mtx_lock(&softc->ctl_lock);
1995 		STAILQ_FOREACH(port, &softc->port_list, links) {
1996 			int action, done;
1997 
1998 			action = 0;
1999 			done = 0;
2000 
2001 			if ((entry->port_type == CTL_PORT_NONE)
2002 			 && (entry->targ_port == port->targ_port)) {
2003 				/*
2004 				 * If the user only wants to enable or
2005 				 * disable or set WWNs on a specific port,
2006 				 * do the operation and we're done.
2007 				 */
2008 				action = 1;
2009 				done = 1;
2010 			} else if (entry->port_type & port->port_type) {
2011 				/*
2012 				 * Compare the user's type mask with the
2013 				 * particular frontend type to see if we
2014 				 * have a match.
2015 				 */
2016 				action = 1;
2017 				done = 0;
2018 
2019 				/*
2020 				 * Make sure the user isn't trying to set
2021 				 * WWNs on multiple ports at the same time.
2022 				 */
2023 				if (cmd == CTL_SET_PORT_WWNS) {
2024 					printf("%s: Can't set WWNs on "
2025 					       "multiple ports\n", __func__);
2026 					retval = EINVAL;
2027 					break;
2028 				}
2029 			}
2030 			if (action != 0) {
2031 				/*
2032 				 * XXX KDM we have to drop the lock here,
2033 				 * because the online/offline operations
2034 				 * can potentially block.  We need to
2035 				 * reference count the frontends so they
2036 				 * can't go away,
2037 				 */
2038 				mtx_unlock(&softc->ctl_lock);
2039 
2040 				if (cmd == CTL_ENABLE_PORT) {
2041 					ctl_port_online(port);
2042 				} else if (cmd == CTL_DISABLE_PORT) {
2043 					ctl_port_offline(port);
2044 				}
2045 
2046 				mtx_lock(&softc->ctl_lock);
2047 
2048 				if (cmd == CTL_SET_PORT_WWNS)
2049 					ctl_port_set_wwns(port,
2050 					    (entry->flags & CTL_PORT_WWNN_VALID) ?
2051 					    1 : 0, entry->wwnn,
2052 					    (entry->flags & CTL_PORT_WWPN_VALID) ?
2053 					    1 : 0, entry->wwpn);
2054 			}
2055 			if (done != 0)
2056 				break;
2057 		}
2058 		mtx_unlock(&softc->ctl_lock);
2059 		break;
2060 	}
2061 	case CTL_GET_PORT_LIST: {
2062 		struct ctl_port *port;
2063 		struct ctl_port_list *list;
2064 		int i;
2065 
2066 		list = (struct ctl_port_list *)addr;
2067 
2068 		if (list->alloc_len != (list->alloc_num *
2069 		    sizeof(struct ctl_port_entry))) {
2070 			printf("%s: CTL_GET_PORT_LIST: alloc_len %u != "
2071 			       "alloc_num %u * sizeof(struct ctl_port_entry) "
2072 			       "%zu\n", __func__, list->alloc_len,
2073 			       list->alloc_num, sizeof(struct ctl_port_entry));
2074 			retval = EINVAL;
2075 			break;
2076 		}
2077 		list->fill_len = 0;
2078 		list->fill_num = 0;
2079 		list->dropped_num = 0;
2080 		i = 0;
2081 		mtx_lock(&softc->ctl_lock);
2082 		STAILQ_FOREACH(port, &softc->port_list, links) {
2083 			struct ctl_port_entry entry, *list_entry;
2084 
2085 			if (list->fill_num >= list->alloc_num) {
2086 				list->dropped_num++;
2087 				continue;
2088 			}
2089 
2090 			entry.port_type = port->port_type;
2091 			strlcpy(entry.port_name, port->port_name,
2092 				sizeof(entry.port_name));
2093 			entry.targ_port = port->targ_port;
2094 			entry.physical_port = port->physical_port;
2095 			entry.virtual_port = port->virtual_port;
2096 			entry.wwnn = port->wwnn;
2097 			entry.wwpn = port->wwpn;
2098 			if (port->status & CTL_PORT_STATUS_ONLINE)
2099 				entry.online = 1;
2100 			else
2101 				entry.online = 0;
2102 
2103 			list_entry = &list->entries[i];
2104 
2105 			retval = copyout(&entry, list_entry, sizeof(entry));
2106 			if (retval != 0) {
2107 				printf("%s: CTL_GET_PORT_LIST: copyout "
2108 				       "returned %d\n", __func__, retval);
2109 				break;
2110 			}
2111 			i++;
2112 			list->fill_num++;
2113 			list->fill_len += sizeof(entry);
2114 		}
2115 		mtx_unlock(&softc->ctl_lock);
2116 
2117 		/*
2118 		 * If this is non-zero, we had a copyout fault, so there's
2119 		 * probably no point in attempting to set the status inside
2120 		 * the structure.
2121 		 */
2122 		if (retval != 0)
2123 			break;
2124 
2125 		if (list->dropped_num > 0)
2126 			list->status = CTL_PORT_LIST_NEED_MORE_SPACE;
2127 		else
2128 			list->status = CTL_PORT_LIST_OK;
2129 		break;
2130 	}
2131 	case CTL_DUMP_OOA: {
2132 		struct ctl_lun *lun;
2133 		union ctl_io *io;
2134 		char printbuf[128];
2135 		struct sbuf sb;
2136 
2137 		mtx_lock(&softc->ctl_lock);
2138 		printf("Dumping OOA queues:\n");
2139 		STAILQ_FOREACH(lun, &softc->lun_list, links) {
2140 			mtx_lock(&lun->lun_lock);
2141 			for (io = (union ctl_io *)TAILQ_FIRST(
2142 			     &lun->ooa_queue); io != NULL;
2143 			     io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2144 			     ooa_links)) {
2145 				sbuf_new(&sb, printbuf, sizeof(printbuf),
2146 					 SBUF_FIXEDLEN);
2147 				sbuf_printf(&sb, "LUN %jd tag 0x%04x%s%s%s%s: ",
2148 					    (intmax_t)lun->lun,
2149 					    io->scsiio.tag_num,
2150 					    (io->io_hdr.flags &
2151 					    CTL_FLAG_BLOCKED) ? "" : " BLOCKED",
2152 					    (io->io_hdr.flags &
2153 					    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
2154 					    (io->io_hdr.flags &
2155 					    CTL_FLAG_ABORT) ? " ABORT" : "",
2156 			                    (io->io_hdr.flags &
2157 		                        CTL_FLAG_IS_WAS_ON_RTR) ? " RTR" : "");
2158 				ctl_scsi_command_string(&io->scsiio, NULL, &sb);
2159 				sbuf_finish(&sb);
2160 				printf("%s\n", sbuf_data(&sb));
2161 			}
2162 			mtx_unlock(&lun->lun_lock);
2163 		}
2164 		printf("OOA queues dump done\n");
2165 		mtx_unlock(&softc->ctl_lock);
2166 		break;
2167 	}
2168 	case CTL_GET_OOA: {
2169 		struct ctl_lun *lun;
2170 		struct ctl_ooa *ooa_hdr;
2171 		struct ctl_ooa_entry *entries;
2172 		uint32_t cur_fill_num;
2173 
2174 		ooa_hdr = (struct ctl_ooa *)addr;
2175 
2176 		if ((ooa_hdr->alloc_len == 0)
2177 		 || (ooa_hdr->alloc_num == 0)) {
2178 			printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2179 			       "must be non-zero\n", __func__,
2180 			       ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2181 			retval = EINVAL;
2182 			break;
2183 		}
2184 
2185 		if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2186 		    sizeof(struct ctl_ooa_entry))) {
2187 			printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2188 			       "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2189 			       __func__, ooa_hdr->alloc_len,
2190 			       ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2191 			retval = EINVAL;
2192 			break;
2193 		}
2194 
2195 		entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2196 		if (entries == NULL) {
2197 			printf("%s: could not allocate %d bytes for OOA "
2198 			       "dump\n", __func__, ooa_hdr->alloc_len);
2199 			retval = ENOMEM;
2200 			break;
2201 		}
2202 
2203 		mtx_lock(&softc->ctl_lock);
2204 		if (((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0)
2205 		 && ((ooa_hdr->lun_num >= CTL_MAX_LUNS)
2206 		  || (softc->ctl_luns[ooa_hdr->lun_num] == NULL))) {
2207 			mtx_unlock(&softc->ctl_lock);
2208 			free(entries, M_CTL);
2209 			printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2210 			       __func__, (uintmax_t)ooa_hdr->lun_num);
2211 			retval = EINVAL;
2212 			break;
2213 		}
2214 
2215 		cur_fill_num = 0;
2216 
2217 		if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2218 			STAILQ_FOREACH(lun, &softc->lun_list, links) {
2219 				retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2220 					ooa_hdr, entries);
2221 				if (retval != 0)
2222 					break;
2223 			}
2224 			if (retval != 0) {
2225 				mtx_unlock(&softc->ctl_lock);
2226 				free(entries, M_CTL);
2227 				break;
2228 			}
2229 		} else {
2230 			lun = softc->ctl_luns[ooa_hdr->lun_num];
2231 
2232 			retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,ooa_hdr,
2233 						    entries);
2234 		}
2235 		mtx_unlock(&softc->ctl_lock);
2236 
2237 		ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2238 		ooa_hdr->fill_len = ooa_hdr->fill_num *
2239 			sizeof(struct ctl_ooa_entry);
2240 		retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2241 		if (retval != 0) {
2242 			printf("%s: error copying out %d bytes for OOA dump\n",
2243 			       __func__, ooa_hdr->fill_len);
2244 		}
2245 
2246 		getbintime(&ooa_hdr->cur_bt);
2247 
2248 		if (cur_fill_num > ooa_hdr->alloc_num) {
2249 			ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2250 			ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2251 		} else {
2252 			ooa_hdr->dropped_num = 0;
2253 			ooa_hdr->status = CTL_OOA_OK;
2254 		}
2255 
2256 		free(entries, M_CTL);
2257 		break;
2258 	}
2259 	case CTL_CHECK_OOA: {
2260 		union ctl_io *io;
2261 		struct ctl_lun *lun;
2262 		struct ctl_ooa_info *ooa_info;
2263 
2264 
2265 		ooa_info = (struct ctl_ooa_info *)addr;
2266 
2267 		if (ooa_info->lun_id >= CTL_MAX_LUNS) {
2268 			ooa_info->status = CTL_OOA_INVALID_LUN;
2269 			break;
2270 		}
2271 		mtx_lock(&softc->ctl_lock);
2272 		lun = softc->ctl_luns[ooa_info->lun_id];
2273 		if (lun == NULL) {
2274 			mtx_unlock(&softc->ctl_lock);
2275 			ooa_info->status = CTL_OOA_INVALID_LUN;
2276 			break;
2277 		}
2278 		mtx_lock(&lun->lun_lock);
2279 		mtx_unlock(&softc->ctl_lock);
2280 		ooa_info->num_entries = 0;
2281 		for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
2282 		     io != NULL; io = (union ctl_io *)TAILQ_NEXT(
2283 		     &io->io_hdr, ooa_links)) {
2284 			ooa_info->num_entries++;
2285 		}
2286 		mtx_unlock(&lun->lun_lock);
2287 
2288 		ooa_info->status = CTL_OOA_SUCCESS;
2289 
2290 		break;
2291 	}
2292 	case CTL_DELAY_IO: {
2293 		struct ctl_io_delay_info *delay_info;
2294 #ifdef CTL_IO_DELAY
2295 		struct ctl_lun *lun;
2296 #endif /* CTL_IO_DELAY */
2297 
2298 		delay_info = (struct ctl_io_delay_info *)addr;
2299 
2300 #ifdef CTL_IO_DELAY
2301 		mtx_lock(&softc->ctl_lock);
2302 
2303 		if ((delay_info->lun_id >= CTL_MAX_LUNS)
2304 		 || (softc->ctl_luns[delay_info->lun_id] == NULL)) {
2305 			delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2306 		} else {
2307 			lun = softc->ctl_luns[delay_info->lun_id];
2308 			mtx_lock(&lun->lun_lock);
2309 
2310 			delay_info->status = CTL_DELAY_STATUS_OK;
2311 
2312 			switch (delay_info->delay_type) {
2313 			case CTL_DELAY_TYPE_CONT:
2314 				break;
2315 			case CTL_DELAY_TYPE_ONESHOT:
2316 				break;
2317 			default:
2318 				delay_info->status =
2319 					CTL_DELAY_STATUS_INVALID_TYPE;
2320 				break;
2321 			}
2322 
2323 			switch (delay_info->delay_loc) {
2324 			case CTL_DELAY_LOC_DATAMOVE:
2325 				lun->delay_info.datamove_type =
2326 					delay_info->delay_type;
2327 				lun->delay_info.datamove_delay =
2328 					delay_info->delay_secs;
2329 				break;
2330 			case CTL_DELAY_LOC_DONE:
2331 				lun->delay_info.done_type =
2332 					delay_info->delay_type;
2333 				lun->delay_info.done_delay =
2334 					delay_info->delay_secs;
2335 				break;
2336 			default:
2337 				delay_info->status =
2338 					CTL_DELAY_STATUS_INVALID_LOC;
2339 				break;
2340 			}
2341 			mtx_unlock(&lun->lun_lock);
2342 		}
2343 
2344 		mtx_unlock(&softc->ctl_lock);
2345 #else
2346 		delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2347 #endif /* CTL_IO_DELAY */
2348 		break;
2349 	}
2350 	case CTL_REALSYNC_SET: {
2351 		int *syncstate;
2352 
2353 		syncstate = (int *)addr;
2354 
2355 		mtx_lock(&softc->ctl_lock);
2356 		switch (*syncstate) {
2357 		case 0:
2358 			softc->flags &= ~CTL_FLAG_REAL_SYNC;
2359 			break;
2360 		case 1:
2361 			softc->flags |= CTL_FLAG_REAL_SYNC;
2362 			break;
2363 		default:
2364 			retval = EINVAL;
2365 			break;
2366 		}
2367 		mtx_unlock(&softc->ctl_lock);
2368 		break;
2369 	}
2370 	case CTL_REALSYNC_GET: {
2371 		int *syncstate;
2372 
2373 		syncstate = (int*)addr;
2374 
2375 		mtx_lock(&softc->ctl_lock);
2376 		if (softc->flags & CTL_FLAG_REAL_SYNC)
2377 			*syncstate = 1;
2378 		else
2379 			*syncstate = 0;
2380 		mtx_unlock(&softc->ctl_lock);
2381 
2382 		break;
2383 	}
2384 	case CTL_SETSYNC:
2385 	case CTL_GETSYNC: {
2386 		struct ctl_sync_info *sync_info;
2387 		struct ctl_lun *lun;
2388 
2389 		sync_info = (struct ctl_sync_info *)addr;
2390 
2391 		mtx_lock(&softc->ctl_lock);
2392 		lun = softc->ctl_luns[sync_info->lun_id];
2393 		if (lun == NULL) {
2394 			mtx_unlock(&softc->ctl_lock);
2395 			sync_info->status = CTL_GS_SYNC_NO_LUN;
2396 		}
2397 		/*
2398 		 * Get or set the sync interval.  We're not bounds checking
2399 		 * in the set case, hopefully the user won't do something
2400 		 * silly.
2401 		 */
2402 		mtx_lock(&lun->lun_lock);
2403 		mtx_unlock(&softc->ctl_lock);
2404 		if (cmd == CTL_GETSYNC)
2405 			sync_info->sync_interval = lun->sync_interval;
2406 		else
2407 			lun->sync_interval = sync_info->sync_interval;
2408 		mtx_unlock(&lun->lun_lock);
2409 
2410 		sync_info->status = CTL_GS_SYNC_OK;
2411 
2412 		break;
2413 	}
2414 	case CTL_GETSTATS: {
2415 		struct ctl_stats *stats;
2416 		struct ctl_lun *lun;
2417 		int i;
2418 
2419 		stats = (struct ctl_stats *)addr;
2420 
2421 		if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) >
2422 		     stats->alloc_len) {
2423 			stats->status = CTL_SS_NEED_MORE_SPACE;
2424 			stats->num_luns = softc->num_luns;
2425 			break;
2426 		}
2427 		/*
2428 		 * XXX KDM no locking here.  If the LUN list changes,
2429 		 * things can blow up.
2430 		 */
2431 		for (i = 0, lun = STAILQ_FIRST(&softc->lun_list); lun != NULL;
2432 		     i++, lun = STAILQ_NEXT(lun, links)) {
2433 			retval = copyout(&lun->stats, &stats->lun_stats[i],
2434 					 sizeof(lun->stats));
2435 			if (retval != 0)
2436 				break;
2437 		}
2438 		stats->num_luns = softc->num_luns;
2439 		stats->fill_len = sizeof(struct ctl_lun_io_stats) *
2440 				 softc->num_luns;
2441 		stats->status = CTL_SS_OK;
2442 #ifdef CTL_TIME_IO
2443 		stats->flags = CTL_STATS_FLAG_TIME_VALID;
2444 #else
2445 		stats->flags = CTL_STATS_FLAG_NONE;
2446 #endif
2447 		getnanouptime(&stats->timestamp);
2448 		break;
2449 	}
2450 	case CTL_ERROR_INJECT: {
2451 		struct ctl_error_desc *err_desc, *new_err_desc;
2452 		struct ctl_lun *lun;
2453 
2454 		err_desc = (struct ctl_error_desc *)addr;
2455 
2456 		new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2457 				      M_WAITOK | M_ZERO);
2458 		bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2459 
2460 		mtx_lock(&softc->ctl_lock);
2461 		lun = softc->ctl_luns[err_desc->lun_id];
2462 		if (lun == NULL) {
2463 			mtx_unlock(&softc->ctl_lock);
2464 			free(new_err_desc, M_CTL);
2465 			printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2466 			       __func__, (uintmax_t)err_desc->lun_id);
2467 			retval = EINVAL;
2468 			break;
2469 		}
2470 		mtx_lock(&lun->lun_lock);
2471 		mtx_unlock(&softc->ctl_lock);
2472 
2473 		/*
2474 		 * We could do some checking here to verify the validity
2475 		 * of the request, but given the complexity of error
2476 		 * injection requests, the checking logic would be fairly
2477 		 * complex.
2478 		 *
2479 		 * For now, if the request is invalid, it just won't get
2480 		 * executed and might get deleted.
2481 		 */
2482 		STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2483 
2484 		/*
2485 		 * XXX KDM check to make sure the serial number is unique,
2486 		 * in case we somehow manage to wrap.  That shouldn't
2487 		 * happen for a very long time, but it's the right thing to
2488 		 * do.
2489 		 */
2490 		new_err_desc->serial = lun->error_serial;
2491 		err_desc->serial = lun->error_serial;
2492 		lun->error_serial++;
2493 
2494 		mtx_unlock(&lun->lun_lock);
2495 		break;
2496 	}
2497 	case CTL_ERROR_INJECT_DELETE: {
2498 		struct ctl_error_desc *delete_desc, *desc, *desc2;
2499 		struct ctl_lun *lun;
2500 		int delete_done;
2501 
2502 		delete_desc = (struct ctl_error_desc *)addr;
2503 		delete_done = 0;
2504 
2505 		mtx_lock(&softc->ctl_lock);
2506 		lun = softc->ctl_luns[delete_desc->lun_id];
2507 		if (lun == NULL) {
2508 			mtx_unlock(&softc->ctl_lock);
2509 			printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2510 			       __func__, (uintmax_t)delete_desc->lun_id);
2511 			retval = EINVAL;
2512 			break;
2513 		}
2514 		mtx_lock(&lun->lun_lock);
2515 		mtx_unlock(&softc->ctl_lock);
2516 		STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2517 			if (desc->serial != delete_desc->serial)
2518 				continue;
2519 
2520 			STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2521 				      links);
2522 			free(desc, M_CTL);
2523 			delete_done = 1;
2524 		}
2525 		mtx_unlock(&lun->lun_lock);
2526 		if (delete_done == 0) {
2527 			printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2528 			       "error serial %ju on LUN %u\n", __func__,
2529 			       delete_desc->serial, delete_desc->lun_id);
2530 			retval = EINVAL;
2531 			break;
2532 		}
2533 		break;
2534 	}
2535 	case CTL_DUMP_STRUCTS: {
2536 		int i, j, k;
2537 		struct ctl_port *port;
2538 		struct ctl_frontend *fe;
2539 
2540 		mtx_lock(&softc->ctl_lock);
2541 		printf("CTL Persistent Reservation information start:\n");
2542 		for (i = 0; i < CTL_MAX_LUNS; i++) {
2543 			struct ctl_lun *lun;
2544 
2545 			lun = softc->ctl_luns[i];
2546 
2547 			if ((lun == NULL)
2548 			 || ((lun->flags & CTL_LUN_DISABLED) != 0))
2549 				continue;
2550 
2551 			for (j = 0; j < (CTL_MAX_PORTS * 2); j++) {
2552 				if (lun->pr_keys[j] == NULL)
2553 					continue;
2554 				for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2555 					if (lun->pr_keys[j][k] == 0)
2556 						continue;
2557 					printf("  LUN %d port %d iid %d key "
2558 					       "%#jx\n", i, j, k,
2559 					       (uintmax_t)lun->pr_keys[j][k]);
2560 				}
2561 			}
2562 		}
2563 		printf("CTL Persistent Reservation information end\n");
2564 		printf("CTL Ports:\n");
2565 		STAILQ_FOREACH(port, &softc->port_list, links) {
2566 			printf("  Port %d '%s' Frontend '%s' Type %u pp %d vp %d WWNN "
2567 			       "%#jx WWPN %#jx\n", port->targ_port, port->port_name,
2568 			       port->frontend->name, port->port_type,
2569 			       port->physical_port, port->virtual_port,
2570 			       (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
2571 			for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
2572 				if (port->wwpn_iid[j].in_use == 0 &&
2573 				    port->wwpn_iid[j].wwpn == 0 &&
2574 				    port->wwpn_iid[j].name == NULL)
2575 					continue;
2576 
2577 				printf("    iid %u use %d WWPN %#jx '%s'\n",
2578 				    j, port->wwpn_iid[j].in_use,
2579 				    (uintmax_t)port->wwpn_iid[j].wwpn,
2580 				    port->wwpn_iid[j].name);
2581 			}
2582 		}
2583 		printf("CTL Port information end\n");
2584 		mtx_unlock(&softc->ctl_lock);
2585 		/*
2586 		 * XXX KDM calling this without a lock.  We'd likely want
2587 		 * to drop the lock before calling the frontend's dump
2588 		 * routine anyway.
2589 		 */
2590 		printf("CTL Frontends:\n");
2591 		STAILQ_FOREACH(fe, &softc->fe_list, links) {
2592 			printf("  Frontend '%s'\n", fe->name);
2593 			if (fe->fe_dump != NULL)
2594 				fe->fe_dump();
2595 		}
2596 		printf("CTL Frontend information end\n");
2597 		break;
2598 	}
2599 	case CTL_LUN_REQ: {
2600 		struct ctl_lun_req *lun_req;
2601 		struct ctl_backend_driver *backend;
2602 
2603 		lun_req = (struct ctl_lun_req *)addr;
2604 
2605 		backend = ctl_backend_find(lun_req->backend);
2606 		if (backend == NULL) {
2607 			lun_req->status = CTL_LUN_ERROR;
2608 			snprintf(lun_req->error_str,
2609 				 sizeof(lun_req->error_str),
2610 				 "Backend \"%s\" not found.",
2611 				 lun_req->backend);
2612 			break;
2613 		}
2614 		if (lun_req->num_be_args > 0) {
2615 			lun_req->kern_be_args = ctl_copyin_args(
2616 				lun_req->num_be_args,
2617 				lun_req->be_args,
2618 				lun_req->error_str,
2619 				sizeof(lun_req->error_str));
2620 			if (lun_req->kern_be_args == NULL) {
2621 				lun_req->status = CTL_LUN_ERROR;
2622 				break;
2623 			}
2624 		}
2625 
2626 		retval = backend->ioctl(dev, cmd, addr, flag, td);
2627 
2628 		if (lun_req->num_be_args > 0) {
2629 			ctl_copyout_args(lun_req->num_be_args,
2630 				      lun_req->kern_be_args);
2631 			ctl_free_args(lun_req->num_be_args,
2632 				      lun_req->kern_be_args);
2633 		}
2634 		break;
2635 	}
2636 	case CTL_LUN_LIST: {
2637 		struct sbuf *sb;
2638 		struct ctl_lun *lun;
2639 		struct ctl_lun_list *list;
2640 		struct ctl_option *opt;
2641 
2642 		list = (struct ctl_lun_list *)addr;
2643 
2644 		/*
2645 		 * Allocate a fixed length sbuf here, based on the length
2646 		 * of the user's buffer.  We could allocate an auto-extending
2647 		 * buffer, and then tell the user how much larger our
2648 		 * amount of data is than his buffer, but that presents
2649 		 * some problems:
2650 		 *
2651 		 * 1.  The sbuf(9) routines use a blocking malloc, and so
2652 		 *     we can't hold a lock while calling them with an
2653 		 *     auto-extending buffer.
2654  		 *
2655 		 * 2.  There is not currently a LUN reference counting
2656 		 *     mechanism, outside of outstanding transactions on
2657 		 *     the LUN's OOA queue.  So a LUN could go away on us
2658 		 *     while we're getting the LUN number, backend-specific
2659 		 *     information, etc.  Thus, given the way things
2660 		 *     currently work, we need to hold the CTL lock while
2661 		 *     grabbing LUN information.
2662 		 *
2663 		 * So, from the user's standpoint, the best thing to do is
2664 		 * allocate what he thinks is a reasonable buffer length,
2665 		 * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
2666 		 * double the buffer length and try again.  (And repeat
2667 		 * that until he succeeds.)
2668 		 */
2669 		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
2670 		if (sb == NULL) {
2671 			list->status = CTL_LUN_LIST_ERROR;
2672 			snprintf(list->error_str, sizeof(list->error_str),
2673 				 "Unable to allocate %d bytes for LUN list",
2674 				 list->alloc_len);
2675 			break;
2676 		}
2677 
2678 		sbuf_printf(sb, "<ctllunlist>\n");
2679 
2680 		mtx_lock(&softc->ctl_lock);
2681 		STAILQ_FOREACH(lun, &softc->lun_list, links) {
2682 			mtx_lock(&lun->lun_lock);
2683 			retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
2684 					     (uintmax_t)lun->lun);
2685 
2686 			/*
2687 			 * Bail out as soon as we see that we've overfilled
2688 			 * the buffer.
2689 			 */
2690 			if (retval != 0)
2691 				break;
2692 
2693 			retval = sbuf_printf(sb, "\t<backend_type>%s"
2694 					     "</backend_type>\n",
2695 					     (lun->backend == NULL) ?  "none" :
2696 					     lun->backend->name);
2697 
2698 			if (retval != 0)
2699 				break;
2700 
2701 			retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
2702 					     lun->be_lun->lun_type);
2703 
2704 			if (retval != 0)
2705 				break;
2706 
2707 			if (lun->backend == NULL) {
2708 				retval = sbuf_printf(sb, "</lun>\n");
2709 				if (retval != 0)
2710 					break;
2711 				continue;
2712 			}
2713 
2714 			retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
2715 					     (lun->be_lun->maxlba > 0) ?
2716 					     lun->be_lun->maxlba + 1 : 0);
2717 
2718 			if (retval != 0)
2719 				break;
2720 
2721 			retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
2722 					     lun->be_lun->blocksize);
2723 
2724 			if (retval != 0)
2725 				break;
2726 
2727 			retval = sbuf_printf(sb, "\t<serial_number>");
2728 
2729 			if (retval != 0)
2730 				break;
2731 
2732 			retval = ctl_sbuf_printf_esc(sb,
2733 			    lun->be_lun->serial_num,
2734 			    sizeof(lun->be_lun->serial_num));
2735 
2736 			if (retval != 0)
2737 				break;
2738 
2739 			retval = sbuf_printf(sb, "</serial_number>\n");
2740 
2741 			if (retval != 0)
2742 				break;
2743 
2744 			retval = sbuf_printf(sb, "\t<device_id>");
2745 
2746 			if (retval != 0)
2747 				break;
2748 
2749 			retval = ctl_sbuf_printf_esc(sb,
2750 			    lun->be_lun->device_id,
2751 			    sizeof(lun->be_lun->device_id));
2752 
2753 			if (retval != 0)
2754 				break;
2755 
2756 			retval = sbuf_printf(sb, "</device_id>\n");
2757 
2758 			if (retval != 0)
2759 				break;
2760 
2761 			if (lun->backend->lun_info != NULL) {
2762 				retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
2763 				if (retval != 0)
2764 					break;
2765 			}
2766 			STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
2767 				retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
2768 				    opt->name, opt->value, opt->name);
2769 				if (retval != 0)
2770 					break;
2771 			}
2772 
2773 			retval = sbuf_printf(sb, "</lun>\n");
2774 
2775 			if (retval != 0)
2776 				break;
2777 			mtx_unlock(&lun->lun_lock);
2778 		}
2779 		if (lun != NULL)
2780 			mtx_unlock(&lun->lun_lock);
2781 		mtx_unlock(&softc->ctl_lock);
2782 
2783 		if ((retval != 0)
2784 		 || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
2785 			retval = 0;
2786 			sbuf_delete(sb);
2787 			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
2788 			snprintf(list->error_str, sizeof(list->error_str),
2789 				 "Out of space, %d bytes is too small",
2790 				 list->alloc_len);
2791 			break;
2792 		}
2793 
2794 		sbuf_finish(sb);
2795 
2796 		retval = copyout(sbuf_data(sb), list->lun_xml,
2797 				 sbuf_len(sb) + 1);
2798 
2799 		list->fill_len = sbuf_len(sb) + 1;
2800 		list->status = CTL_LUN_LIST_OK;
2801 		sbuf_delete(sb);
2802 		break;
2803 	}
2804 	case CTL_ISCSI: {
2805 		struct ctl_iscsi *ci;
2806 		struct ctl_frontend *fe;
2807 
2808 		ci = (struct ctl_iscsi *)addr;
2809 
2810 		fe = ctl_frontend_find("iscsi");
2811 		if (fe == NULL) {
2812 			ci->status = CTL_ISCSI_ERROR;
2813 			snprintf(ci->error_str, sizeof(ci->error_str),
2814 			    "Frontend \"iscsi\" not found.");
2815 			break;
2816 		}
2817 
2818 		retval = fe->ioctl(dev, cmd, addr, flag, td);
2819 		break;
2820 	}
2821 	case CTL_PORT_REQ: {
2822 		struct ctl_req *req;
2823 		struct ctl_frontend *fe;
2824 
2825 		req = (struct ctl_req *)addr;
2826 
2827 		fe = ctl_frontend_find(req->driver);
2828 		if (fe == NULL) {
2829 			req->status = CTL_LUN_ERROR;
2830 			snprintf(req->error_str, sizeof(req->error_str),
2831 			    "Frontend \"%s\" not found.", req->driver);
2832 			break;
2833 		}
2834 		if (req->num_args > 0) {
2835 			req->kern_args = ctl_copyin_args(req->num_args,
2836 			    req->args, req->error_str, sizeof(req->error_str));
2837 			if (req->kern_args == NULL) {
2838 				req->status = CTL_LUN_ERROR;
2839 				break;
2840 			}
2841 		}
2842 
2843 		retval = fe->ioctl(dev, cmd, addr, flag, td);
2844 
2845 		if (req->num_args > 0) {
2846 			ctl_copyout_args(req->num_args, req->kern_args);
2847 			ctl_free_args(req->num_args, req->kern_args);
2848 		}
2849 		break;
2850 	}
2851 	case CTL_PORT_LIST: {
2852 		struct sbuf *sb;
2853 		struct ctl_port *port;
2854 		struct ctl_lun_list *list;
2855 		struct ctl_option *opt;
2856 		int j;
2857 		uint32_t plun;
2858 
2859 		list = (struct ctl_lun_list *)addr;
2860 
2861 		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
2862 		if (sb == NULL) {
2863 			list->status = CTL_LUN_LIST_ERROR;
2864 			snprintf(list->error_str, sizeof(list->error_str),
2865 				 "Unable to allocate %d bytes for LUN list",
2866 				 list->alloc_len);
2867 			break;
2868 		}
2869 
2870 		sbuf_printf(sb, "<ctlportlist>\n");
2871 
2872 		mtx_lock(&softc->ctl_lock);
2873 		STAILQ_FOREACH(port, &softc->port_list, links) {
2874 			retval = sbuf_printf(sb, "<targ_port id=\"%ju\">\n",
2875 					     (uintmax_t)port->targ_port);
2876 
2877 			/*
2878 			 * Bail out as soon as we see that we've overfilled
2879 			 * the buffer.
2880 			 */
2881 			if (retval != 0)
2882 				break;
2883 
2884 			retval = sbuf_printf(sb, "\t<frontend_type>%s"
2885 			    "</frontend_type>\n", port->frontend->name);
2886 			if (retval != 0)
2887 				break;
2888 
2889 			retval = sbuf_printf(sb, "\t<port_type>%d</port_type>\n",
2890 					     port->port_type);
2891 			if (retval != 0)
2892 				break;
2893 
2894 			retval = sbuf_printf(sb, "\t<online>%s</online>\n",
2895 			    (port->status & CTL_PORT_STATUS_ONLINE) ? "YES" : "NO");
2896 			if (retval != 0)
2897 				break;
2898 
2899 			retval = sbuf_printf(sb, "\t<port_name>%s</port_name>\n",
2900 			    port->port_name);
2901 			if (retval != 0)
2902 				break;
2903 
2904 			retval = sbuf_printf(sb, "\t<physical_port>%d</physical_port>\n",
2905 			    port->physical_port);
2906 			if (retval != 0)
2907 				break;
2908 
2909 			retval = sbuf_printf(sb, "\t<virtual_port>%d</virtual_port>\n",
2910 			    port->virtual_port);
2911 			if (retval != 0)
2912 				break;
2913 
2914 			if (port->target_devid != NULL) {
2915 				sbuf_printf(sb, "\t<target>");
2916 				ctl_id_sbuf(port->target_devid, sb);
2917 				sbuf_printf(sb, "</target>\n");
2918 			}
2919 
2920 			if (port->port_devid != NULL) {
2921 				sbuf_printf(sb, "\t<port>");
2922 				ctl_id_sbuf(port->port_devid, sb);
2923 				sbuf_printf(sb, "</port>\n");
2924 			}
2925 
2926 			if (port->port_info != NULL) {
2927 				retval = port->port_info(port->onoff_arg, sb);
2928 				if (retval != 0)
2929 					break;
2930 			}
2931 			STAILQ_FOREACH(opt, &port->options, links) {
2932 				retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
2933 				    opt->name, opt->value, opt->name);
2934 				if (retval != 0)
2935 					break;
2936 			}
2937 
2938 			if (port->lun_map != NULL) {
2939 				sbuf_printf(sb, "\t<lun_map>on</lun_map>\n");
2940 				for (j = 0; j < CTL_MAX_LUNS; j++) {
2941 					plun = ctl_lun_map_from_port(port, j);
2942 					if (plun >= CTL_MAX_LUNS)
2943 						continue;
2944 					sbuf_printf(sb,
2945 					    "\t<lun id=\"%u\">%u</lun>\n",
2946 					    j, plun);
2947 				}
2948 			}
2949 
2950 			for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
2951 				if (port->wwpn_iid[j].in_use == 0 ||
2952 				    (port->wwpn_iid[j].wwpn == 0 &&
2953 				     port->wwpn_iid[j].name == NULL))
2954 					continue;
2955 
2956 				if (port->wwpn_iid[j].name != NULL)
2957 					retval = sbuf_printf(sb,
2958 					    "\t<initiator id=\"%u\">%s</initiator>\n",
2959 					    j, port->wwpn_iid[j].name);
2960 				else
2961 					retval = sbuf_printf(sb,
2962 					    "\t<initiator id=\"%u\">naa.%08jx</initiator>\n",
2963 					    j, port->wwpn_iid[j].wwpn);
2964 				if (retval != 0)
2965 					break;
2966 			}
2967 			if (retval != 0)
2968 				break;
2969 
2970 			retval = sbuf_printf(sb, "</targ_port>\n");
2971 			if (retval != 0)
2972 				break;
2973 		}
2974 		mtx_unlock(&softc->ctl_lock);
2975 
2976 		if ((retval != 0)
2977 		 || ((retval = sbuf_printf(sb, "</ctlportlist>\n")) != 0)) {
2978 			retval = 0;
2979 			sbuf_delete(sb);
2980 			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
2981 			snprintf(list->error_str, sizeof(list->error_str),
2982 				 "Out of space, %d bytes is too small",
2983 				 list->alloc_len);
2984 			break;
2985 		}
2986 
2987 		sbuf_finish(sb);
2988 
2989 		retval = copyout(sbuf_data(sb), list->lun_xml,
2990 				 sbuf_len(sb) + 1);
2991 
2992 		list->fill_len = sbuf_len(sb) + 1;
2993 		list->status = CTL_LUN_LIST_OK;
2994 		sbuf_delete(sb);
2995 		break;
2996 	}
2997 	case CTL_LUN_MAP: {
2998 		struct ctl_lun_map *lm  = (struct ctl_lun_map *)addr;
2999 		struct ctl_port *port;
3000 
3001 		mtx_lock(&softc->ctl_lock);
3002 		if (lm->port >= CTL_MAX_PORTS ||
3003 		    (port = softc->ctl_ports[lm->port]) == NULL) {
3004 			mtx_unlock(&softc->ctl_lock);
3005 			return (ENXIO);
3006 		}
3007 		mtx_unlock(&softc->ctl_lock); // XXX: port_enable sleeps
3008 		if (lm->plun < CTL_MAX_LUNS) {
3009 			if (lm->lun == UINT32_MAX)
3010 				retval = ctl_lun_map_unset(port, lm->plun);
3011 			else if (lm->lun < CTL_MAX_LUNS &&
3012 			    softc->ctl_luns[lm->lun] != NULL)
3013 				retval = ctl_lun_map_set(port, lm->plun, lm->lun);
3014 			else
3015 				return (ENXIO);
3016 		} else if (lm->plun == UINT32_MAX) {
3017 			if (lm->lun == UINT32_MAX)
3018 				retval = ctl_lun_map_deinit(port);
3019 			else
3020 				retval = ctl_lun_map_init(port);
3021 		} else
3022 			return (ENXIO);
3023 		break;
3024 	}
3025 	default: {
3026 		/* XXX KDM should we fix this? */
3027 #if 0
3028 		struct ctl_backend_driver *backend;
3029 		unsigned int type;
3030 		int found;
3031 
3032 		found = 0;
3033 
3034 		/*
3035 		 * We encode the backend type as the ioctl type for backend
3036 		 * ioctls.  So parse it out here, and then search for a
3037 		 * backend of this type.
3038 		 */
3039 		type = _IOC_TYPE(cmd);
3040 
3041 		STAILQ_FOREACH(backend, &softc->be_list, links) {
3042 			if (backend->type == type) {
3043 				found = 1;
3044 				break;
3045 			}
3046 		}
3047 		if (found == 0) {
3048 			printf("ctl: unknown ioctl command %#lx or backend "
3049 			       "%d\n", cmd, type);
3050 			retval = EINVAL;
3051 			break;
3052 		}
3053 		retval = backend->ioctl(dev, cmd, addr, flag, td);
3054 #endif
3055 		retval = ENOTTY;
3056 		break;
3057 	}
3058 	}
3059 	return (retval);
3060 }
3061 
3062 uint32_t
3063 ctl_get_initindex(struct ctl_nexus *nexus)
3064 {
3065 	if (nexus->targ_port < CTL_MAX_PORTS)
3066 		return (nexus->initid.id +
3067 			(nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3068 	else
3069 		return (nexus->initid.id +
3070 		       ((nexus->targ_port - CTL_MAX_PORTS) *
3071 			CTL_MAX_INIT_PER_PORT));
3072 }
3073 
3074 uint32_t
3075 ctl_get_resindex(struct ctl_nexus *nexus)
3076 {
3077 	return (nexus->initid.id + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3078 }
3079 
3080 uint32_t
3081 ctl_port_idx(int port_num)
3082 {
3083 	if (port_num < CTL_MAX_PORTS)
3084 		return(port_num);
3085 	else
3086 		return(port_num - CTL_MAX_PORTS);
3087 }
3088 
3089 int
3090 ctl_lun_map_init(struct ctl_port *port)
3091 {
3092 	struct ctl_softc *softc = control_softc;
3093 	struct ctl_lun *lun;
3094 	uint32_t i;
3095 
3096 	if (port->lun_map == NULL)
3097 		port->lun_map = malloc(sizeof(uint32_t) * CTL_MAX_LUNS,
3098 		    M_CTL, M_NOWAIT);
3099 	if (port->lun_map == NULL)
3100 		return (ENOMEM);
3101 	for (i = 0; i < CTL_MAX_LUNS; i++)
3102 		port->lun_map[i] = UINT32_MAX;
3103 	if (port->status & CTL_PORT_STATUS_ONLINE) {
3104 		STAILQ_FOREACH(lun, &softc->lun_list, links)
3105 			port->lun_disable(port->targ_lun_arg, lun->lun);
3106 	}
3107 	return (0);
3108 }
3109 
3110 int
3111 ctl_lun_map_deinit(struct ctl_port *port)
3112 {
3113 	struct ctl_softc *softc = control_softc;
3114 	struct ctl_lun *lun;
3115 
3116 	if (port->lun_map == NULL)
3117 		return (0);
3118 	free(port->lun_map, M_CTL);
3119 	port->lun_map = NULL;
3120 	if (port->status & CTL_PORT_STATUS_ONLINE) {
3121 		STAILQ_FOREACH(lun, &softc->lun_list, links)
3122 			port->lun_enable(port->targ_lun_arg, lun->lun);
3123 	}
3124 	return (0);
3125 }
3126 
3127 int
3128 ctl_lun_map_set(struct ctl_port *port, uint32_t plun, uint32_t glun)
3129 {
3130 	int status;
3131 	uint32_t old;
3132 
3133 	if (port->lun_map == NULL) {
3134 		status = ctl_lun_map_init(port);
3135 		if (status != 0)
3136 			return (status);
3137 	}
3138 	old = port->lun_map[plun];
3139 	port->lun_map[plun] = glun;
3140 	if ((port->status & CTL_PORT_STATUS_ONLINE) && old >= CTL_MAX_LUNS)
3141 		port->lun_enable(port->targ_lun_arg, plun);
3142 	return (0);
3143 }
3144 
3145 int
3146 ctl_lun_map_unset(struct ctl_port *port, uint32_t plun)
3147 {
3148 	uint32_t old;
3149 
3150 	if (port->lun_map == NULL)
3151 		return (0);
3152 	old = port->lun_map[plun];
3153 	port->lun_map[plun] = UINT32_MAX;
3154 	if ((port->status & CTL_PORT_STATUS_ONLINE) && old < CTL_MAX_LUNS)
3155 		port->lun_disable(port->targ_lun_arg, plun);
3156 	return (0);
3157 }
3158 
3159 uint32_t
3160 ctl_lun_map_from_port(struct ctl_port *port, uint32_t lun_id)
3161 {
3162 
3163 	if (port == NULL)
3164 		return (UINT32_MAX);
3165 	if (port->lun_map == NULL || lun_id >= CTL_MAX_LUNS)
3166 		return (lun_id);
3167 	return (port->lun_map[lun_id]);
3168 }
3169 
3170 uint32_t
3171 ctl_lun_map_to_port(struct ctl_port *port, uint32_t lun_id)
3172 {
3173 	uint32_t i;
3174 
3175 	if (port == NULL)
3176 		return (UINT32_MAX);
3177 	if (port->lun_map == NULL)
3178 		return (lun_id);
3179 	for (i = 0; i < CTL_MAX_LUNS; i++) {
3180 		if (port->lun_map[i] == lun_id)
3181 			return (i);
3182 	}
3183 	return (UINT32_MAX);
3184 }
3185 
3186 static struct ctl_port *
3187 ctl_io_port(struct ctl_io_hdr *io_hdr)
3188 {
3189 	int port_num;
3190 
3191 	port_num = io_hdr->nexus.targ_port;
3192 	return (control_softc->ctl_ports[ctl_port_idx(port_num)]);
3193 }
3194 
3195 /*
3196  * Note:  This only works for bitmask sizes that are at least 32 bits, and
3197  * that are a power of 2.
3198  */
3199 int
3200 ctl_ffz(uint32_t *mask, uint32_t size)
3201 {
3202 	uint32_t num_chunks, num_pieces;
3203 	int i, j;
3204 
3205 	num_chunks = (size >> 5);
3206 	if (num_chunks == 0)
3207 		num_chunks++;
3208 	num_pieces = MIN((sizeof(uint32_t) * 8), size);
3209 
3210 	for (i = 0; i < num_chunks; i++) {
3211 		for (j = 0; j < num_pieces; j++) {
3212 			if ((mask[i] & (1 << j)) == 0)
3213 				return ((i << 5) + j);
3214 		}
3215 	}
3216 
3217 	return (-1);
3218 }
3219 
3220 int
3221 ctl_set_mask(uint32_t *mask, uint32_t bit)
3222 {
3223 	uint32_t chunk, piece;
3224 
3225 	chunk = bit >> 5;
3226 	piece = bit % (sizeof(uint32_t) * 8);
3227 
3228 	if ((mask[chunk] & (1 << piece)) != 0)
3229 		return (-1);
3230 	else
3231 		mask[chunk] |= (1 << piece);
3232 
3233 	return (0);
3234 }
3235 
3236 int
3237 ctl_clear_mask(uint32_t *mask, uint32_t bit)
3238 {
3239 	uint32_t chunk, piece;
3240 
3241 	chunk = bit >> 5;
3242 	piece = bit % (sizeof(uint32_t) * 8);
3243 
3244 	if ((mask[chunk] & (1 << piece)) == 0)
3245 		return (-1);
3246 	else
3247 		mask[chunk] &= ~(1 << piece);
3248 
3249 	return (0);
3250 }
3251 
3252 int
3253 ctl_is_set(uint32_t *mask, uint32_t bit)
3254 {
3255 	uint32_t chunk, piece;
3256 
3257 	chunk = bit >> 5;
3258 	piece = bit % (sizeof(uint32_t) * 8);
3259 
3260 	if ((mask[chunk] & (1 << piece)) == 0)
3261 		return (0);
3262 	else
3263 		return (1);
3264 }
3265 
3266 static uint64_t
3267 ctl_get_prkey(struct ctl_lun *lun, uint32_t residx)
3268 {
3269 	uint64_t *t;
3270 
3271 	t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3272 	if (t == NULL)
3273 		return (0);
3274 	return (t[residx % CTL_MAX_INIT_PER_PORT]);
3275 }
3276 
3277 static void
3278 ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx)
3279 {
3280 	uint64_t *t;
3281 
3282 	t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3283 	if (t == NULL)
3284 		return;
3285 	t[residx % CTL_MAX_INIT_PER_PORT] = 0;
3286 }
3287 
3288 static void
3289 ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx)
3290 {
3291 	uint64_t *p;
3292 	u_int i;
3293 
3294 	i = residx/CTL_MAX_INIT_PER_PORT;
3295 	if (lun->pr_keys[i] != NULL)
3296 		return;
3297 	mtx_unlock(&lun->lun_lock);
3298 	p = malloc(sizeof(uint64_t) * CTL_MAX_INIT_PER_PORT, M_CTL,
3299 	    M_WAITOK | M_ZERO);
3300 	mtx_lock(&lun->lun_lock);
3301 	if (lun->pr_keys[i] == NULL)
3302 		lun->pr_keys[i] = p;
3303 	else
3304 		free(p, M_CTL);
3305 }
3306 
3307 static void
3308 ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key)
3309 {
3310 	uint64_t *t;
3311 
3312 	t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3313 	KASSERT(t != NULL, ("prkey %d is not allocated", residx));
3314 	t[residx % CTL_MAX_INIT_PER_PORT] = key;
3315 }
3316 
3317 /*
3318  * ctl_softc, pool_name, total_ctl_io are passed in.
3319  * npool is passed out.
3320  */
3321 int
3322 ctl_pool_create(struct ctl_softc *ctl_softc, const char *pool_name,
3323 		uint32_t total_ctl_io, void **npool)
3324 {
3325 #ifdef IO_POOLS
3326 	struct ctl_io_pool *pool;
3327 
3328 	pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3329 					    M_NOWAIT | M_ZERO);
3330 	if (pool == NULL)
3331 		return (ENOMEM);
3332 
3333 	snprintf(pool->name, sizeof(pool->name), "CTL IO %s", pool_name);
3334 	pool->ctl_softc = ctl_softc;
3335 	pool->zone = uma_zsecond_create(pool->name, NULL,
3336 	    NULL, NULL, NULL, ctl_softc->io_zone);
3337 	/* uma_prealloc(pool->zone, total_ctl_io); */
3338 
3339 	*npool = pool;
3340 #else
3341 	*npool = ctl_softc->io_zone;
3342 #endif
3343 	return (0);
3344 }
3345 
3346 void
3347 ctl_pool_free(struct ctl_io_pool *pool)
3348 {
3349 
3350 	if (pool == NULL)
3351 		return;
3352 
3353 #ifdef IO_POOLS
3354 	uma_zdestroy(pool->zone);
3355 	free(pool, M_CTL);
3356 #endif
3357 }
3358 
3359 union ctl_io *
3360 ctl_alloc_io(void *pool_ref)
3361 {
3362 	union ctl_io *io;
3363 #ifdef IO_POOLS
3364 	struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3365 
3366 	io = uma_zalloc(pool->zone, M_WAITOK);
3367 #else
3368 	io = uma_zalloc((uma_zone_t)pool_ref, M_WAITOK);
3369 #endif
3370 	if (io != NULL)
3371 		io->io_hdr.pool = pool_ref;
3372 	return (io);
3373 }
3374 
3375 union ctl_io *
3376 ctl_alloc_io_nowait(void *pool_ref)
3377 {
3378 	union ctl_io *io;
3379 #ifdef IO_POOLS
3380 	struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3381 
3382 	io = uma_zalloc(pool->zone, M_NOWAIT);
3383 #else
3384 	io = uma_zalloc((uma_zone_t)pool_ref, M_NOWAIT);
3385 #endif
3386 	if (io != NULL)
3387 		io->io_hdr.pool = pool_ref;
3388 	return (io);
3389 }
3390 
3391 void
3392 ctl_free_io(union ctl_io *io)
3393 {
3394 #ifdef IO_POOLS
3395 	struct ctl_io_pool *pool;
3396 #endif
3397 
3398 	if (io == NULL)
3399 		return;
3400 
3401 #ifdef IO_POOLS
3402 	pool = (struct ctl_io_pool *)io->io_hdr.pool;
3403 	uma_zfree(pool->zone, io);
3404 #else
3405 	uma_zfree((uma_zone_t)io->io_hdr.pool, io);
3406 #endif
3407 }
3408 
3409 void
3410 ctl_zero_io(union ctl_io *io)
3411 {
3412 	void *pool_ref;
3413 
3414 	if (io == NULL)
3415 		return;
3416 
3417 	/*
3418 	 * May need to preserve linked list pointers at some point too.
3419 	 */
3420 	pool_ref = io->io_hdr.pool;
3421 	memset(io, 0, sizeof(*io));
3422 	io->io_hdr.pool = pool_ref;
3423 }
3424 
3425 /*
3426  * This routine is currently used for internal copies of ctl_ios that need
3427  * to persist for some reason after we've already returned status to the
3428  * FETD.  (Thus the flag set.)
3429  *
3430  * XXX XXX
3431  * Note that this makes a blind copy of all fields in the ctl_io, except
3432  * for the pool reference.  This includes any memory that has been
3433  * allocated!  That memory will no longer be valid after done has been
3434  * called, so this would be VERY DANGEROUS for command that actually does
3435  * any reads or writes.  Right now (11/7/2005), this is only used for immediate
3436  * start and stop commands, which don't transfer any data, so this is not a
3437  * problem.  If it is used for anything else, the caller would also need to
3438  * allocate data buffer space and this routine would need to be modified to
3439  * copy the data buffer(s) as well.
3440  */
3441 void
3442 ctl_copy_io(union ctl_io *src, union ctl_io *dest)
3443 {
3444 	void *pool_ref;
3445 
3446 	if ((src == NULL)
3447 	 || (dest == NULL))
3448 		return;
3449 
3450 	/*
3451 	 * May need to preserve linked list pointers at some point too.
3452 	 */
3453 	pool_ref = dest->io_hdr.pool;
3454 
3455 	memcpy(dest, src, MIN(sizeof(*src), sizeof(*dest)));
3456 
3457 	dest->io_hdr.pool = pool_ref;
3458 	/*
3459 	 * We need to know that this is an internal copy, and doesn't need
3460 	 * to get passed back to the FETD that allocated it.
3461 	 */
3462 	dest->io_hdr.flags |= CTL_FLAG_INT_COPY;
3463 }
3464 
3465 int
3466 ctl_expand_number(const char *buf, uint64_t *num)
3467 {
3468 	char *endptr;
3469 	uint64_t number;
3470 	unsigned shift;
3471 
3472 	number = strtoq(buf, &endptr, 0);
3473 
3474 	switch (tolower((unsigned char)*endptr)) {
3475 	case 'e':
3476 		shift = 60;
3477 		break;
3478 	case 'p':
3479 		shift = 50;
3480 		break;
3481 	case 't':
3482 		shift = 40;
3483 		break;
3484 	case 'g':
3485 		shift = 30;
3486 		break;
3487 	case 'm':
3488 		shift = 20;
3489 		break;
3490 	case 'k':
3491 		shift = 10;
3492 		break;
3493 	case 'b':
3494 	case '\0': /* No unit. */
3495 		*num = number;
3496 		return (0);
3497 	default:
3498 		/* Unrecognized unit. */
3499 		return (-1);
3500 	}
3501 
3502 	if ((number << shift) >> shift != number) {
3503 		/* Overflow */
3504 		return (-1);
3505 	}
3506 	*num = number << shift;
3507 	return (0);
3508 }
3509 
3510 
3511 /*
3512  * This routine could be used in the future to load default and/or saved
3513  * mode page parameters for a particuar lun.
3514  */
3515 static int
3516 ctl_init_page_index(struct ctl_lun *lun)
3517 {
3518 	int i;
3519 	struct ctl_page_index *page_index;
3520 	const char *value;
3521 	uint64_t ival;
3522 
3523 	memcpy(&lun->mode_pages.index, page_index_template,
3524 	       sizeof(page_index_template));
3525 
3526 	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
3527 
3528 		page_index = &lun->mode_pages.index[i];
3529 		/*
3530 		 * If this is a disk-only mode page, there's no point in
3531 		 * setting it up.  For some pages, we have to have some
3532 		 * basic information about the disk in order to calculate the
3533 		 * mode page data.
3534 		 */
3535 		if ((lun->be_lun->lun_type != T_DIRECT)
3536 		 && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
3537 			continue;
3538 
3539 		switch (page_index->page_code & SMPH_PC_MASK) {
3540 		case SMS_RW_ERROR_RECOVERY_PAGE: {
3541 			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3542 				panic("subpage is incorrect!");
3543 			memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT],
3544 			       &rw_er_page_default,
3545 			       sizeof(rw_er_page_default));
3546 			memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CHANGEABLE],
3547 			       &rw_er_page_changeable,
3548 			       sizeof(rw_er_page_changeable));
3549 			memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_DEFAULT],
3550 			       &rw_er_page_default,
3551 			       sizeof(rw_er_page_default));
3552 			memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_SAVED],
3553 			       &rw_er_page_default,
3554 			       sizeof(rw_er_page_default));
3555 			page_index->page_data =
3556 				(uint8_t *)lun->mode_pages.rw_er_page;
3557 			break;
3558 		}
3559 		case SMS_FORMAT_DEVICE_PAGE: {
3560 			struct scsi_format_page *format_page;
3561 
3562 			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3563 				panic("subpage is incorrect!");
3564 
3565 			/*
3566 			 * Sectors per track are set above.  Bytes per
3567 			 * sector need to be set here on a per-LUN basis.
3568 			 */
3569 			memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
3570 			       &format_page_default,
3571 			       sizeof(format_page_default));
3572 			memcpy(&lun->mode_pages.format_page[
3573 			       CTL_PAGE_CHANGEABLE], &format_page_changeable,
3574 			       sizeof(format_page_changeable));
3575 			memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
3576 			       &format_page_default,
3577 			       sizeof(format_page_default));
3578 			memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
3579 			       &format_page_default,
3580 			       sizeof(format_page_default));
3581 
3582 			format_page = &lun->mode_pages.format_page[
3583 				CTL_PAGE_CURRENT];
3584 			scsi_ulto2b(lun->be_lun->blocksize,
3585 				    format_page->bytes_per_sector);
3586 
3587 			format_page = &lun->mode_pages.format_page[
3588 				CTL_PAGE_DEFAULT];
3589 			scsi_ulto2b(lun->be_lun->blocksize,
3590 				    format_page->bytes_per_sector);
3591 
3592 			format_page = &lun->mode_pages.format_page[
3593 				CTL_PAGE_SAVED];
3594 			scsi_ulto2b(lun->be_lun->blocksize,
3595 				    format_page->bytes_per_sector);
3596 
3597 			page_index->page_data =
3598 				(uint8_t *)lun->mode_pages.format_page;
3599 			break;
3600 		}
3601 		case SMS_RIGID_DISK_PAGE: {
3602 			struct scsi_rigid_disk_page *rigid_disk_page;
3603 			uint32_t sectors_per_cylinder;
3604 			uint64_t cylinders;
3605 #ifndef	__XSCALE__
3606 			int shift;
3607 #endif /* !__XSCALE__ */
3608 
3609 			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3610 				panic("invalid subpage value %d",
3611 				      page_index->subpage);
3612 
3613 			/*
3614 			 * Rotation rate and sectors per track are set
3615 			 * above.  We calculate the cylinders here based on
3616 			 * capacity.  Due to the number of heads and
3617 			 * sectors per track we're using, smaller arrays
3618 			 * may turn out to have 0 cylinders.  Linux and
3619 			 * FreeBSD don't pay attention to these mode pages
3620 			 * to figure out capacity, but Solaris does.  It
3621 			 * seems to deal with 0 cylinders just fine, and
3622 			 * works out a fake geometry based on the capacity.
3623 			 */
3624 			memcpy(&lun->mode_pages.rigid_disk_page[
3625 			       CTL_PAGE_DEFAULT], &rigid_disk_page_default,
3626 			       sizeof(rigid_disk_page_default));
3627 			memcpy(&lun->mode_pages.rigid_disk_page[
3628 			       CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
3629 			       sizeof(rigid_disk_page_changeable));
3630 
3631 			sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
3632 				CTL_DEFAULT_HEADS;
3633 
3634 			/*
3635 			 * The divide method here will be more accurate,
3636 			 * probably, but results in floating point being
3637 			 * used in the kernel on i386 (__udivdi3()).  On the
3638 			 * XScale, though, __udivdi3() is implemented in
3639 			 * software.
3640 			 *
3641 			 * The shift method for cylinder calculation is
3642 			 * accurate if sectors_per_cylinder is a power of
3643 			 * 2.  Otherwise it might be slightly off -- you
3644 			 * might have a bit of a truncation problem.
3645 			 */
3646 #ifdef	__XSCALE__
3647 			cylinders = (lun->be_lun->maxlba + 1) /
3648 				sectors_per_cylinder;
3649 #else
3650 			for (shift = 31; shift > 0; shift--) {
3651 				if (sectors_per_cylinder & (1 << shift))
3652 					break;
3653 			}
3654 			cylinders = (lun->be_lun->maxlba + 1) >> shift;
3655 #endif
3656 
3657 			/*
3658 			 * We've basically got 3 bytes, or 24 bits for the
3659 			 * cylinder size in the mode page.  If we're over,
3660 			 * just round down to 2^24.
3661 			 */
3662 			if (cylinders > 0xffffff)
3663 				cylinders = 0xffffff;
3664 
3665 			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
3666 				CTL_PAGE_DEFAULT];
3667 			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
3668 
3669 			if ((value = ctl_get_opt(&lun->be_lun->options,
3670 			    "rpm")) != NULL) {
3671 				scsi_ulto2b(strtol(value, NULL, 0),
3672 				     rigid_disk_page->rotation_rate);
3673 			}
3674 
3675 			memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_CURRENT],
3676 			       &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
3677 			       sizeof(rigid_disk_page_default));
3678 			memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_SAVED],
3679 			       &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
3680 			       sizeof(rigid_disk_page_default));
3681 
3682 			page_index->page_data =
3683 				(uint8_t *)lun->mode_pages.rigid_disk_page;
3684 			break;
3685 		}
3686 		case SMS_CACHING_PAGE: {
3687 			struct scsi_caching_page *caching_page;
3688 
3689 			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3690 				panic("invalid subpage value %d",
3691 				      page_index->subpage);
3692 			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
3693 			       &caching_page_default,
3694 			       sizeof(caching_page_default));
3695 			memcpy(&lun->mode_pages.caching_page[
3696 			       CTL_PAGE_CHANGEABLE], &caching_page_changeable,
3697 			       sizeof(caching_page_changeable));
3698 			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
3699 			       &caching_page_default,
3700 			       sizeof(caching_page_default));
3701 			caching_page = &lun->mode_pages.caching_page[
3702 			    CTL_PAGE_SAVED];
3703 			value = ctl_get_opt(&lun->be_lun->options, "writecache");
3704 			if (value != NULL && strcmp(value, "off") == 0)
3705 				caching_page->flags1 &= ~SCP_WCE;
3706 			value = ctl_get_opt(&lun->be_lun->options, "readcache");
3707 			if (value != NULL && strcmp(value, "off") == 0)
3708 				caching_page->flags1 |= SCP_RCD;
3709 			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
3710 			       &lun->mode_pages.caching_page[CTL_PAGE_SAVED],
3711 			       sizeof(caching_page_default));
3712 			page_index->page_data =
3713 				(uint8_t *)lun->mode_pages.caching_page;
3714 			break;
3715 		}
3716 		case SMS_CONTROL_MODE_PAGE: {
3717 			struct scsi_control_page *control_page;
3718 
3719 			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3720 				panic("invalid subpage value %d",
3721 				      page_index->subpage);
3722 
3723 			memcpy(&lun->mode_pages.control_page[CTL_PAGE_DEFAULT],
3724 			       &control_page_default,
3725 			       sizeof(control_page_default));
3726 			memcpy(&lun->mode_pages.control_page[
3727 			       CTL_PAGE_CHANGEABLE], &control_page_changeable,
3728 			       sizeof(control_page_changeable));
3729 			memcpy(&lun->mode_pages.control_page[CTL_PAGE_SAVED],
3730 			       &control_page_default,
3731 			       sizeof(control_page_default));
3732 			control_page = &lun->mode_pages.control_page[
3733 			    CTL_PAGE_SAVED];
3734 			value = ctl_get_opt(&lun->be_lun->options, "reordering");
3735 			if (value != NULL && strcmp(value, "unrestricted") == 0) {
3736 				control_page->queue_flags &= ~SCP_QUEUE_ALG_MASK;
3737 				control_page->queue_flags |= SCP_QUEUE_ALG_UNRESTRICTED;
3738 			}
3739 			memcpy(&lun->mode_pages.control_page[CTL_PAGE_CURRENT],
3740 			       &lun->mode_pages.control_page[CTL_PAGE_SAVED],
3741 			       sizeof(control_page_default));
3742 			page_index->page_data =
3743 				(uint8_t *)lun->mode_pages.control_page;
3744 			break;
3745 
3746 		}
3747 		case SMS_INFO_EXCEPTIONS_PAGE: {
3748 			switch (page_index->subpage) {
3749 			case SMS_SUBPAGE_PAGE_0:
3750 				memcpy(&lun->mode_pages.ie_page[CTL_PAGE_CURRENT],
3751 				       &ie_page_default,
3752 				       sizeof(ie_page_default));
3753 				memcpy(&lun->mode_pages.ie_page[
3754 				       CTL_PAGE_CHANGEABLE], &ie_page_changeable,
3755 				       sizeof(ie_page_changeable));
3756 				memcpy(&lun->mode_pages.ie_page[CTL_PAGE_DEFAULT],
3757 				       &ie_page_default,
3758 				       sizeof(ie_page_default));
3759 				memcpy(&lun->mode_pages.ie_page[CTL_PAGE_SAVED],
3760 				       &ie_page_default,
3761 				       sizeof(ie_page_default));
3762 				page_index->page_data =
3763 					(uint8_t *)lun->mode_pages.ie_page;
3764 				break;
3765 			case 0x02: {
3766 				struct ctl_logical_block_provisioning_page *page;
3767 
3768 				memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_DEFAULT],
3769 				       &lbp_page_default,
3770 				       sizeof(lbp_page_default));
3771 				memcpy(&lun->mode_pages.lbp_page[
3772 				       CTL_PAGE_CHANGEABLE], &lbp_page_changeable,
3773 				       sizeof(lbp_page_changeable));
3774 				memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
3775 				       &lbp_page_default,
3776 				       sizeof(lbp_page_default));
3777 				page = &lun->mode_pages.lbp_page[CTL_PAGE_SAVED];
3778 				value = ctl_get_opt(&lun->be_lun->options,
3779 				    "avail-threshold");
3780 				if (value != NULL &&
3781 				    ctl_expand_number(value, &ival) == 0) {
3782 					page->descr[0].flags |= SLBPPD_ENABLED |
3783 					    SLBPPD_ARMING_DEC;
3784 					if (lun->be_lun->blocksize)
3785 						ival /= lun->be_lun->blocksize;
3786 					else
3787 						ival /= 512;
3788 					scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
3789 					    page->descr[0].count);
3790 				}
3791 				value = ctl_get_opt(&lun->be_lun->options,
3792 				    "used-threshold");
3793 				if (value != NULL &&
3794 				    ctl_expand_number(value, &ival) == 0) {
3795 					page->descr[1].flags |= SLBPPD_ENABLED |
3796 					    SLBPPD_ARMING_INC;
3797 					if (lun->be_lun->blocksize)
3798 						ival /= lun->be_lun->blocksize;
3799 					else
3800 						ival /= 512;
3801 					scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
3802 					    page->descr[1].count);
3803 				}
3804 				value = ctl_get_opt(&lun->be_lun->options,
3805 				    "pool-avail-threshold");
3806 				if (value != NULL &&
3807 				    ctl_expand_number(value, &ival) == 0) {
3808 					page->descr[2].flags |= SLBPPD_ENABLED |
3809 					    SLBPPD_ARMING_DEC;
3810 					if (lun->be_lun->blocksize)
3811 						ival /= lun->be_lun->blocksize;
3812 					else
3813 						ival /= 512;
3814 					scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
3815 					    page->descr[2].count);
3816 				}
3817 				value = ctl_get_opt(&lun->be_lun->options,
3818 				    "pool-used-threshold");
3819 				if (value != NULL &&
3820 				    ctl_expand_number(value, &ival) == 0) {
3821 					page->descr[3].flags |= SLBPPD_ENABLED |
3822 					    SLBPPD_ARMING_INC;
3823 					if (lun->be_lun->blocksize)
3824 						ival /= lun->be_lun->blocksize;
3825 					else
3826 						ival /= 512;
3827 					scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
3828 					    page->descr[3].count);
3829 				}
3830 				memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_CURRENT],
3831 				       &lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
3832 				       sizeof(lbp_page_default));
3833 				page_index->page_data =
3834 					(uint8_t *)lun->mode_pages.lbp_page;
3835 			}}
3836 			break;
3837 		}
3838 		case SMS_VENDOR_SPECIFIC_PAGE:{
3839 			switch (page_index->subpage) {
3840 			case DBGCNF_SUBPAGE_CODE: {
3841 				struct copan_debugconf_subpage *current_page,
3842 							       *saved_page;
3843 
3844 				memcpy(&lun->mode_pages.debugconf_subpage[
3845 				       CTL_PAGE_CURRENT],
3846 				       &debugconf_page_default,
3847 				       sizeof(debugconf_page_default));
3848 				memcpy(&lun->mode_pages.debugconf_subpage[
3849 				       CTL_PAGE_CHANGEABLE],
3850 				       &debugconf_page_changeable,
3851 				       sizeof(debugconf_page_changeable));
3852 				memcpy(&lun->mode_pages.debugconf_subpage[
3853 				       CTL_PAGE_DEFAULT],
3854 				       &debugconf_page_default,
3855 				       sizeof(debugconf_page_default));
3856 				memcpy(&lun->mode_pages.debugconf_subpage[
3857 				       CTL_PAGE_SAVED],
3858 				       &debugconf_page_default,
3859 				       sizeof(debugconf_page_default));
3860 				page_index->page_data =
3861 					(uint8_t *)lun->mode_pages.debugconf_subpage;
3862 
3863 				current_page = (struct copan_debugconf_subpage *)
3864 					(page_index->page_data +
3865 					 (page_index->page_len *
3866 					  CTL_PAGE_CURRENT));
3867 				saved_page = (struct copan_debugconf_subpage *)
3868 					(page_index->page_data +
3869 					 (page_index->page_len *
3870 					  CTL_PAGE_SAVED));
3871 				break;
3872 			}
3873 			default:
3874 				panic("invalid subpage value %d",
3875 				      page_index->subpage);
3876 				break;
3877 			}
3878    			break;
3879 		}
3880 		default:
3881 			panic("invalid page value %d",
3882 			      page_index->page_code & SMPH_PC_MASK);
3883 			break;
3884     	}
3885 	}
3886 
3887 	return (CTL_RETVAL_COMPLETE);
3888 }
3889 
3890 static int
3891 ctl_init_log_page_index(struct ctl_lun *lun)
3892 {
3893 	struct ctl_page_index *page_index;
3894 	int i, j, k, prev;
3895 
3896 	memcpy(&lun->log_pages.index, log_page_index_template,
3897 	       sizeof(log_page_index_template));
3898 
3899 	prev = -1;
3900 	for (i = 0, j = 0, k = 0; i < CTL_NUM_LOG_PAGES; i++) {
3901 
3902 		page_index = &lun->log_pages.index[i];
3903 		/*
3904 		 * If this is a disk-only mode page, there's no point in
3905 		 * setting it up.  For some pages, we have to have some
3906 		 * basic information about the disk in order to calculate the
3907 		 * mode page data.
3908 		 */
3909 		if ((lun->be_lun->lun_type != T_DIRECT)
3910 		 && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
3911 			continue;
3912 
3913 		if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING &&
3914 		     lun->backend->lun_attr == NULL)
3915 			continue;
3916 
3917 		if (page_index->page_code != prev) {
3918 			lun->log_pages.pages_page[j] = page_index->page_code;
3919 			prev = page_index->page_code;
3920 			j++;
3921 		}
3922 		lun->log_pages.subpages_page[k*2] = page_index->page_code;
3923 		lun->log_pages.subpages_page[k*2+1] = page_index->subpage;
3924 		k++;
3925 	}
3926 	lun->log_pages.index[0].page_data = &lun->log_pages.pages_page[0];
3927 	lun->log_pages.index[0].page_len = j;
3928 	lun->log_pages.index[1].page_data = &lun->log_pages.subpages_page[0];
3929 	lun->log_pages.index[1].page_len = k * 2;
3930 	lun->log_pages.index[2].page_data = &lun->log_pages.lbp_page[0];
3931 	lun->log_pages.index[2].page_len = 12*CTL_NUM_LBP_PARAMS;
3932 	lun->log_pages.index[3].page_data = (uint8_t *)&lun->log_pages.stat_page;
3933 	lun->log_pages.index[3].page_len = sizeof(lun->log_pages.stat_page);
3934 
3935 	return (CTL_RETVAL_COMPLETE);
3936 }
3937 
3938 static int
3939 hex2bin(const char *str, uint8_t *buf, int buf_size)
3940 {
3941 	int i;
3942 	u_char c;
3943 
3944 	memset(buf, 0, buf_size);
3945 	while (isspace(str[0]))
3946 		str++;
3947 	if (str[0] == '0' && (str[1] == 'x' || str[1] == 'X'))
3948 		str += 2;
3949 	buf_size *= 2;
3950 	for (i = 0; str[i] != 0 && i < buf_size; i++) {
3951 		c = str[i];
3952 		if (isdigit(c))
3953 			c -= '0';
3954 		else if (isalpha(c))
3955 			c -= isupper(c) ? 'A' - 10 : 'a' - 10;
3956 		else
3957 			break;
3958 		if (c >= 16)
3959 			break;
3960 		if ((i & 1) == 0)
3961 			buf[i / 2] |= (c << 4);
3962 		else
3963 			buf[i / 2] |= c;
3964 	}
3965 	return ((i + 1) / 2);
3966 }
3967 
3968 /*
3969  * LUN allocation.
3970  *
3971  * Requirements:
3972  * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
3973  *   wants us to allocate the LUN and he can block.
3974  * - ctl_softc is always set
3975  * - be_lun is set if the LUN has a backend (needed for disk LUNs)
3976  *
3977  * Returns 0 for success, non-zero (errno) for failure.
3978  */
3979 static int
3980 ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
3981 	      struct ctl_be_lun *const be_lun)
3982 {
3983 	struct ctl_lun *nlun, *lun;
3984 	struct scsi_vpd_id_descriptor *desc;
3985 	struct scsi_vpd_id_t10 *t10id;
3986 	const char *eui, *naa, *scsiname, *vendor, *value;
3987 	int lun_number, i, lun_malloced;
3988 	int devidlen, idlen1, idlen2 = 0, len;
3989 
3990 	if (be_lun == NULL)
3991 		return (EINVAL);
3992 
3993 	/*
3994 	 * We currently only support Direct Access or Processor LUN types.
3995 	 */
3996 	switch (be_lun->lun_type) {
3997 	case T_DIRECT:
3998 		break;
3999 	case T_PROCESSOR:
4000 		break;
4001 	case T_SEQUENTIAL:
4002 	case T_CHANGER:
4003 	default:
4004 		be_lun->lun_config_status(be_lun->be_lun,
4005 					  CTL_LUN_CONFIG_FAILURE);
4006 		break;
4007 	}
4008 	if (ctl_lun == NULL) {
4009 		lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4010 		lun_malloced = 1;
4011 	} else {
4012 		lun_malloced = 0;
4013 		lun = ctl_lun;
4014 	}
4015 
4016 	memset(lun, 0, sizeof(*lun));
4017 	if (lun_malloced)
4018 		lun->flags = CTL_LUN_MALLOCED;
4019 
4020 	/* Generate LUN ID. */
4021 	devidlen = max(CTL_DEVID_MIN_LEN,
4022 	    strnlen(be_lun->device_id, CTL_DEVID_LEN));
4023 	idlen1 = sizeof(*t10id) + devidlen;
4024 	len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4025 	scsiname = ctl_get_opt(&be_lun->options, "scsiname");
4026 	if (scsiname != NULL) {
4027 		idlen2 = roundup2(strlen(scsiname) + 1, 4);
4028 		len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4029 	}
4030 	eui = ctl_get_opt(&be_lun->options, "eui");
4031 	if (eui != NULL) {
4032 		len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4033 	}
4034 	naa = ctl_get_opt(&be_lun->options, "naa");
4035 	if (naa != NULL) {
4036 		len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4037 	}
4038 	lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4039 	    M_CTL, M_WAITOK | M_ZERO);
4040 	desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4041 	desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4042 	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4043 	desc->length = idlen1;
4044 	t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4045 	memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4046 	if ((vendor = ctl_get_opt(&be_lun->options, "vendor")) == NULL) {
4047 		strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4048 	} else {
4049 		strncpy(t10id->vendor, vendor,
4050 		    min(sizeof(t10id->vendor), strlen(vendor)));
4051 	}
4052 	strncpy((char *)t10id->vendor_spec_id,
4053 	    (char *)be_lun->device_id, devidlen);
4054 	if (scsiname != NULL) {
4055 		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4056 		    desc->length);
4057 		desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4058 		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4059 		    SVPD_ID_TYPE_SCSI_NAME;
4060 		desc->length = idlen2;
4061 		strlcpy(desc->identifier, scsiname, idlen2);
4062 	}
4063 	if (eui != NULL) {
4064 		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4065 		    desc->length);
4066 		desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4067 		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4068 		    SVPD_ID_TYPE_EUI64;
4069 		desc->length = hex2bin(eui, desc->identifier, 16);
4070 		desc->length = desc->length > 12 ? 16 :
4071 		    (desc->length > 8 ? 12 : 8);
4072 		len -= 16 - desc->length;
4073 	}
4074 	if (naa != NULL) {
4075 		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4076 		    desc->length);
4077 		desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4078 		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4079 		    SVPD_ID_TYPE_NAA;
4080 		desc->length = hex2bin(naa, desc->identifier, 16);
4081 		desc->length = desc->length > 8 ? 16 : 8;
4082 		len -= 16 - desc->length;
4083 	}
4084 	lun->lun_devid->len = len;
4085 
4086 	mtx_lock(&ctl_softc->ctl_lock);
4087 	/*
4088 	 * See if the caller requested a particular LUN number.  If so, see
4089 	 * if it is available.  Otherwise, allocate the first available LUN.
4090 	 */
4091 	if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4092 		if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4093 		 || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4094 			mtx_unlock(&ctl_softc->ctl_lock);
4095 			if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4096 				printf("ctl: requested LUN ID %d is higher "
4097 				       "than CTL_MAX_LUNS - 1 (%d)\n",
4098 				       be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4099 			} else {
4100 				/*
4101 				 * XXX KDM return an error, or just assign
4102 				 * another LUN ID in this case??
4103 				 */
4104 				printf("ctl: requested LUN ID %d is already "
4105 				       "in use\n", be_lun->req_lun_id);
4106 			}
4107 			if (lun->flags & CTL_LUN_MALLOCED)
4108 				free(lun, M_CTL);
4109 			be_lun->lun_config_status(be_lun->be_lun,
4110 						  CTL_LUN_CONFIG_FAILURE);
4111 			return (ENOSPC);
4112 		}
4113 		lun_number = be_lun->req_lun_id;
4114 	} else {
4115 		lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, CTL_MAX_LUNS);
4116 		if (lun_number == -1) {
4117 			mtx_unlock(&ctl_softc->ctl_lock);
4118 			printf("ctl: can't allocate LUN, out of LUNs\n");
4119 			if (lun->flags & CTL_LUN_MALLOCED)
4120 				free(lun, M_CTL);
4121 			be_lun->lun_config_status(be_lun->be_lun,
4122 						  CTL_LUN_CONFIG_FAILURE);
4123 			return (ENOSPC);
4124 		}
4125 	}
4126 	ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4127 
4128 	mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4129 	lun->lun = lun_number;
4130 	lun->be_lun = be_lun;
4131 	/*
4132 	 * The processor LUN is always enabled.  Disk LUNs come on line
4133 	 * disabled, and must be enabled by the backend.
4134 	 */
4135 	lun->flags |= CTL_LUN_DISABLED;
4136 	lun->backend = be_lun->be;
4137 	be_lun->ctl_lun = lun;
4138 	be_lun->lun_id = lun_number;
4139 	atomic_add_int(&be_lun->be->num_luns, 1);
4140 	if (be_lun->flags & CTL_LUN_FLAG_OFFLINE)
4141 		lun->flags |= CTL_LUN_OFFLINE;
4142 
4143 	if (be_lun->flags & CTL_LUN_FLAG_POWERED_OFF)
4144 		lun->flags |= CTL_LUN_STOPPED;
4145 
4146 	if (be_lun->flags & CTL_LUN_FLAG_INOPERABLE)
4147 		lun->flags |= CTL_LUN_INOPERABLE;
4148 
4149 	if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4150 		lun->flags |= CTL_LUN_PRIMARY_SC;
4151 
4152 	value = ctl_get_opt(&be_lun->options, "readonly");
4153 	if (value != NULL && strcmp(value, "on") == 0)
4154 		lun->flags |= CTL_LUN_READONLY;
4155 
4156 	lun->serseq = CTL_LUN_SERSEQ_OFF;
4157 	if (be_lun->flags & CTL_LUN_FLAG_SERSEQ_READ)
4158 		lun->serseq = CTL_LUN_SERSEQ_READ;
4159 	value = ctl_get_opt(&be_lun->options, "serseq");
4160 	if (value != NULL && strcmp(value, "on") == 0)
4161 		lun->serseq = CTL_LUN_SERSEQ_ON;
4162 	else if (value != NULL && strcmp(value, "read") == 0)
4163 		lun->serseq = CTL_LUN_SERSEQ_READ;
4164 	else if (value != NULL && strcmp(value, "off") == 0)
4165 		lun->serseq = CTL_LUN_SERSEQ_OFF;
4166 
4167 	lun->ctl_softc = ctl_softc;
4168 #ifdef CTL_TIME_IO
4169 	lun->last_busy = getsbinuptime();
4170 #endif
4171 	TAILQ_INIT(&lun->ooa_queue);
4172 	TAILQ_INIT(&lun->blocked_queue);
4173 	STAILQ_INIT(&lun->error_list);
4174 	ctl_tpc_lun_init(lun);
4175 
4176 	/*
4177 	 * Initialize the mode and log page index.
4178 	 */
4179 	ctl_init_page_index(lun);
4180 	ctl_init_log_page_index(lun);
4181 
4182 	/*
4183 	 * Now, before we insert this lun on the lun list, set the lun
4184 	 * inventory changed UA for all other luns.
4185 	 */
4186 	STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4187 		mtx_lock(&nlun->lun_lock);
4188 		ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4189 		mtx_unlock(&nlun->lun_lock);
4190 	}
4191 
4192 	STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4193 
4194 	ctl_softc->ctl_luns[lun_number] = lun;
4195 
4196 	ctl_softc->num_luns++;
4197 
4198 	/* Setup statistics gathering */
4199 	lun->stats.device_type = be_lun->lun_type;
4200 	lun->stats.lun_number = lun_number;
4201 	if (lun->stats.device_type == T_DIRECT)
4202 		lun->stats.blocksize = be_lun->blocksize;
4203 	else
4204 		lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4205 	for (i = 0;i < CTL_MAX_PORTS;i++)
4206 		lun->stats.ports[i].targ_port = i;
4207 
4208 	mtx_unlock(&ctl_softc->ctl_lock);
4209 
4210 	lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4211 	return (0);
4212 }
4213 
4214 /*
4215  * Delete a LUN.
4216  * Assumptions:
4217  * - LUN has already been marked invalid and any pending I/O has been taken
4218  *   care of.
4219  */
4220 static int
4221 ctl_free_lun(struct ctl_lun *lun)
4222 {
4223 	struct ctl_softc *softc;
4224 	struct ctl_lun *nlun;
4225 	int i;
4226 
4227 	softc = lun->ctl_softc;
4228 
4229 	mtx_assert(&softc->ctl_lock, MA_OWNED);
4230 
4231 	STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4232 
4233 	ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4234 
4235 	softc->ctl_luns[lun->lun] = NULL;
4236 
4237 	if (!TAILQ_EMPTY(&lun->ooa_queue))
4238 		panic("Freeing a LUN %p with outstanding I/O!!\n", lun);
4239 
4240 	softc->num_luns--;
4241 
4242 	/*
4243 	 * Tell the backend to free resources, if this LUN has a backend.
4244 	 */
4245 	atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4246 	lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4247 
4248 	ctl_tpc_lun_shutdown(lun);
4249 	mtx_destroy(&lun->lun_lock);
4250 	free(lun->lun_devid, M_CTL);
4251 	for (i = 0; i < CTL_MAX_PORTS; i++)
4252 		free(lun->pending_ua[i], M_CTL);
4253 	for (i = 0; i < 2 * CTL_MAX_PORTS; i++)
4254 		free(lun->pr_keys[i], M_CTL);
4255 	free(lun->write_buffer, M_CTL);
4256 	if (lun->flags & CTL_LUN_MALLOCED)
4257 		free(lun, M_CTL);
4258 
4259 	STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4260 		mtx_lock(&nlun->lun_lock);
4261 		ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4262 		mtx_unlock(&nlun->lun_lock);
4263 	}
4264 
4265 	return (0);
4266 }
4267 
4268 static void
4269 ctl_create_lun(struct ctl_be_lun *be_lun)
4270 {
4271 	struct ctl_softc *softc;
4272 
4273 	softc = control_softc;
4274 
4275 	/*
4276 	 * ctl_alloc_lun() should handle all potential failure cases.
4277 	 */
4278 	ctl_alloc_lun(softc, NULL, be_lun);
4279 }
4280 
4281 int
4282 ctl_add_lun(struct ctl_be_lun *be_lun)
4283 {
4284 	struct ctl_softc *softc = control_softc;
4285 
4286 	mtx_lock(&softc->ctl_lock);
4287 	STAILQ_INSERT_TAIL(&softc->pending_lun_queue, be_lun, links);
4288 	mtx_unlock(&softc->ctl_lock);
4289 	wakeup(&softc->pending_lun_queue);
4290 
4291 	return (0);
4292 }
4293 
4294 int
4295 ctl_enable_lun(struct ctl_be_lun *be_lun)
4296 {
4297 	struct ctl_softc *softc;
4298 	struct ctl_port *port, *nport;
4299 	struct ctl_lun *lun;
4300 	int retval;
4301 
4302 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4303 	softc = lun->ctl_softc;
4304 
4305 	mtx_lock(&softc->ctl_lock);
4306 	mtx_lock(&lun->lun_lock);
4307 	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4308 		/*
4309 		 * eh?  Why did we get called if the LUN is already
4310 		 * enabled?
4311 		 */
4312 		mtx_unlock(&lun->lun_lock);
4313 		mtx_unlock(&softc->ctl_lock);
4314 		return (0);
4315 	}
4316 	lun->flags &= ~CTL_LUN_DISABLED;
4317 	mtx_unlock(&lun->lun_lock);
4318 
4319 	for (port = STAILQ_FIRST(&softc->port_list); port != NULL; port = nport) {
4320 		nport = STAILQ_NEXT(port, links);
4321 		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4322 		    port->lun_map != NULL)
4323 			continue;
4324 
4325 		/*
4326 		 * Drop the lock while we call the FETD's enable routine.
4327 		 * This can lead to a callback into CTL (at least in the
4328 		 * case of the internal initiator frontend.
4329 		 */
4330 		mtx_unlock(&softc->ctl_lock);
4331 		retval = port->lun_enable(port->targ_lun_arg, lun->lun);
4332 		mtx_lock(&softc->ctl_lock);
4333 		if (retval != 0) {
4334 			printf("%s: FETD %s port %d returned error "
4335 			       "%d for lun_enable on lun %jd\n",
4336 			       __func__, port->port_name, port->targ_port,
4337 			       retval, (intmax_t)lun->lun);
4338 		}
4339 	}
4340 
4341 	mtx_unlock(&softc->ctl_lock);
4342 
4343 	return (0);
4344 }
4345 
4346 int
4347 ctl_disable_lun(struct ctl_be_lun *be_lun)
4348 {
4349 	struct ctl_softc *softc;
4350 	struct ctl_port *port;
4351 	struct ctl_lun *lun;
4352 	int retval;
4353 
4354 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4355 	softc = lun->ctl_softc;
4356 
4357 	mtx_lock(&softc->ctl_lock);
4358 	mtx_lock(&lun->lun_lock);
4359 	if (lun->flags & CTL_LUN_DISABLED) {
4360 		mtx_unlock(&lun->lun_lock);
4361 		mtx_unlock(&softc->ctl_lock);
4362 		return (0);
4363 	}
4364 	lun->flags |= CTL_LUN_DISABLED;
4365 	mtx_unlock(&lun->lun_lock);
4366 
4367 	STAILQ_FOREACH(port, &softc->port_list, links) {
4368 		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4369 		    port->lun_map != NULL)
4370 			continue;
4371 		mtx_unlock(&softc->ctl_lock);
4372 		/*
4373 		 * Drop the lock before we call the frontend's disable
4374 		 * routine, to avoid lock order reversals.
4375 		 *
4376 		 * XXX KDM what happens if the frontend list changes while
4377 		 * we're traversing it?  It's unlikely, but should be handled.
4378 		 */
4379 		retval = port->lun_disable(port->targ_lun_arg, lun->lun);
4380 		mtx_lock(&softc->ctl_lock);
4381 		if (retval != 0) {
4382 			printf("%s: FETD %s port %d returned error "
4383 			       "%d for lun_disable on lun %jd\n",
4384 			       __func__, port->port_name, port->targ_port,
4385 			       retval, (intmax_t)lun->lun);
4386 		}
4387 	}
4388 
4389 	mtx_unlock(&softc->ctl_lock);
4390 
4391 	return (0);
4392 }
4393 
4394 int
4395 ctl_start_lun(struct ctl_be_lun *be_lun)
4396 {
4397 	struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4398 
4399 	mtx_lock(&lun->lun_lock);
4400 	lun->flags &= ~CTL_LUN_STOPPED;
4401 	mtx_unlock(&lun->lun_lock);
4402 	return (0);
4403 }
4404 
4405 int
4406 ctl_stop_lun(struct ctl_be_lun *be_lun)
4407 {
4408 	struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4409 
4410 	mtx_lock(&lun->lun_lock);
4411 	lun->flags |= CTL_LUN_STOPPED;
4412 	mtx_unlock(&lun->lun_lock);
4413 	return (0);
4414 }
4415 
4416 int
4417 ctl_lun_offline(struct ctl_be_lun *be_lun)
4418 {
4419 	struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4420 
4421 	mtx_lock(&lun->lun_lock);
4422 	lun->flags |= CTL_LUN_OFFLINE;
4423 	mtx_unlock(&lun->lun_lock);
4424 	return (0);
4425 }
4426 
4427 int
4428 ctl_lun_online(struct ctl_be_lun *be_lun)
4429 {
4430 	struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4431 
4432 	mtx_lock(&lun->lun_lock);
4433 	lun->flags &= ~CTL_LUN_OFFLINE;
4434 	mtx_unlock(&lun->lun_lock);
4435 	return (0);
4436 }
4437 
4438 int
4439 ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4440 {
4441 	struct ctl_softc *softc;
4442 	struct ctl_lun *lun;
4443 
4444 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4445 	softc = lun->ctl_softc;
4446 
4447 	mtx_lock(&lun->lun_lock);
4448 
4449 	/*
4450 	 * The LUN needs to be disabled before it can be marked invalid.
4451 	 */
4452 	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4453 		mtx_unlock(&lun->lun_lock);
4454 		return (-1);
4455 	}
4456 	/*
4457 	 * Mark the LUN invalid.
4458 	 */
4459 	lun->flags |= CTL_LUN_INVALID;
4460 
4461 	/*
4462 	 * If there is nothing in the OOA queue, go ahead and free the LUN.
4463 	 * If we have something in the OOA queue, we'll free it when the
4464 	 * last I/O completes.
4465 	 */
4466 	if (TAILQ_EMPTY(&lun->ooa_queue)) {
4467 		mtx_unlock(&lun->lun_lock);
4468 		mtx_lock(&softc->ctl_lock);
4469 		ctl_free_lun(lun);
4470 		mtx_unlock(&softc->ctl_lock);
4471 	} else
4472 		mtx_unlock(&lun->lun_lock);
4473 
4474 	return (0);
4475 }
4476 
4477 int
4478 ctl_lun_inoperable(struct ctl_be_lun *be_lun)
4479 {
4480 	struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4481 
4482 	mtx_lock(&lun->lun_lock);
4483 	lun->flags |= CTL_LUN_INOPERABLE;
4484 	mtx_unlock(&lun->lun_lock);
4485 	return (0);
4486 }
4487 
4488 int
4489 ctl_lun_operable(struct ctl_be_lun *be_lun)
4490 {
4491 	struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4492 
4493 	mtx_lock(&lun->lun_lock);
4494 	lun->flags &= ~CTL_LUN_INOPERABLE;
4495 	mtx_unlock(&lun->lun_lock);
4496 	return (0);
4497 }
4498 
4499 void
4500 ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
4501 {
4502 	struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4503 
4504 	mtx_lock(&lun->lun_lock);
4505 	ctl_est_ua_all(lun, -1, CTL_UA_CAPACITY_CHANGED);
4506 	mtx_unlock(&lun->lun_lock);
4507 }
4508 
4509 /*
4510  * Backend "memory move is complete" callback for requests that never
4511  * make it down to say RAIDCore's configuration code.
4512  */
4513 int
4514 ctl_config_move_done(union ctl_io *io)
4515 {
4516 	int retval;
4517 
4518 	CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
4519 	KASSERT(io->io_hdr.io_type == CTL_IO_SCSI,
4520 	    ("Config I/O type isn't CTL_IO_SCSI (%d)!", io->io_hdr.io_type));
4521 
4522 	if ((io->io_hdr.port_status != 0) &&
4523 	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
4524 	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
4525 		/*
4526 		 * For hardware error sense keys, the sense key
4527 		 * specific value is defined to be a retry count,
4528 		 * but we use it to pass back an internal FETD
4529 		 * error code.  XXX KDM  Hopefully the FETD is only
4530 		 * using 16 bits for an error code, since that's
4531 		 * all the space we have in the sks field.
4532 		 */
4533 		ctl_set_internal_failure(&io->scsiio,
4534 					 /*sks_valid*/ 1,
4535 					 /*retry_count*/
4536 					 io->io_hdr.port_status);
4537 	}
4538 
4539 	if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) ||
4540 	    ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
4541 	     (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) ||
4542 	    ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
4543 		/*
4544 		 * XXX KDM just assuming a single pointer here, and not a
4545 		 * S/G list.  If we start using S/G lists for config data,
4546 		 * we'll need to know how to clean them up here as well.
4547 		 */
4548 		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
4549 			free(io->scsiio.kern_data_ptr, M_CTL);
4550 		ctl_done(io);
4551 		retval = CTL_RETVAL_COMPLETE;
4552 	} else {
4553 		/*
4554 		 * XXX KDM now we need to continue data movement.  Some
4555 		 * options:
4556 		 * - call ctl_scsiio() again?  We don't do this for data
4557 		 *   writes, because for those at least we know ahead of
4558 		 *   time where the write will go and how long it is.  For
4559 		 *   config writes, though, that information is largely
4560 		 *   contained within the write itself, thus we need to
4561 		 *   parse out the data again.
4562 		 *
4563 		 * - Call some other function once the data is in?
4564 		 */
4565 		if (ctl_debug & CTL_DEBUG_CDB_DATA)
4566 			ctl_data_print(io);
4567 
4568 		/*
4569 		 * XXX KDM call ctl_scsiio() again for now, and check flag
4570 		 * bits to see whether we're allocated or not.
4571 		 */
4572 		retval = ctl_scsiio(&io->scsiio);
4573 	}
4574 	return (retval);
4575 }
4576 
4577 /*
4578  * This gets called by a backend driver when it is done with a
4579  * data_submit method.
4580  */
4581 void
4582 ctl_data_submit_done(union ctl_io *io)
4583 {
4584 	/*
4585 	 * If the IO_CONT flag is set, we need to call the supplied
4586 	 * function to continue processing the I/O, instead of completing
4587 	 * the I/O just yet.
4588 	 *
4589 	 * If there is an error, though, we don't want to keep processing.
4590 	 * Instead, just send status back to the initiator.
4591 	 */
4592 	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
4593 	    (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
4594 	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
4595 	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
4596 		io->scsiio.io_cont(io);
4597 		return;
4598 	}
4599 	ctl_done(io);
4600 }
4601 
4602 /*
4603  * This gets called by a backend driver when it is done with a
4604  * configuration write.
4605  */
4606 void
4607 ctl_config_write_done(union ctl_io *io)
4608 {
4609 	uint8_t *buf;
4610 
4611 	/*
4612 	 * If the IO_CONT flag is set, we need to call the supplied
4613 	 * function to continue processing the I/O, instead of completing
4614 	 * the I/O just yet.
4615 	 *
4616 	 * If there is an error, though, we don't want to keep processing.
4617 	 * Instead, just send status back to the initiator.
4618 	 */
4619 	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
4620 	    (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
4621 	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
4622 	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
4623 		io->scsiio.io_cont(io);
4624 		return;
4625 	}
4626 	/*
4627 	 * Since a configuration write can be done for commands that actually
4628 	 * have data allocated, like write buffer, and commands that have
4629 	 * no data, like start/stop unit, we need to check here.
4630 	 */
4631 	if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
4632 		buf = io->scsiio.kern_data_ptr;
4633 	else
4634 		buf = NULL;
4635 	ctl_done(io);
4636 	if (buf)
4637 		free(buf, M_CTL);
4638 }
4639 
4640 void
4641 ctl_config_read_done(union ctl_io *io)
4642 {
4643 	uint8_t *buf;
4644 
4645 	/*
4646 	 * If there is some error -- we are done, skip data transfer.
4647 	 */
4648 	if ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0 ||
4649 	    ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
4650 	     (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
4651 		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
4652 			buf = io->scsiio.kern_data_ptr;
4653 		else
4654 			buf = NULL;
4655 		ctl_done(io);
4656 		if (buf)
4657 			free(buf, M_CTL);
4658 		return;
4659 	}
4660 
4661 	/*
4662 	 * If the IO_CONT flag is set, we need to call the supplied
4663 	 * function to continue processing the I/O, instead of completing
4664 	 * the I/O just yet.
4665 	 */
4666 	if (io->io_hdr.flags & CTL_FLAG_IO_CONT) {
4667 		io->scsiio.io_cont(io);
4668 		return;
4669 	}
4670 
4671 	ctl_datamove(io);
4672 }
4673 
4674 /*
4675  * SCSI release command.
4676  */
4677 int
4678 ctl_scsi_release(struct ctl_scsiio *ctsio)
4679 {
4680 	int length, longid, thirdparty_id, resv_id;
4681 	struct ctl_lun *lun;
4682 	uint32_t residx;
4683 
4684 	length = 0;
4685 	resv_id = 0;
4686 
4687 	CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
4688 
4689 	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
4690 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
4691 
4692 	switch (ctsio->cdb[0]) {
4693 	case RELEASE_10: {
4694 		struct scsi_release_10 *cdb;
4695 
4696 		cdb = (struct scsi_release_10 *)ctsio->cdb;
4697 
4698 		if (cdb->byte2 & SR10_LONGID)
4699 			longid = 1;
4700 		else
4701 			thirdparty_id = cdb->thirdparty_id;
4702 
4703 		resv_id = cdb->resv_id;
4704 		length = scsi_2btoul(cdb->length);
4705 		break;
4706 	}
4707 	}
4708 
4709 
4710 	/*
4711 	 * XXX KDM right now, we only support LUN reservation.  We don't
4712 	 * support 3rd party reservations, or extent reservations, which
4713 	 * might actually need the parameter list.  If we've gotten this
4714 	 * far, we've got a LUN reservation.  Anything else got kicked out
4715 	 * above.  So, according to SPC, ignore the length.
4716 	 */
4717 	length = 0;
4718 
4719 	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
4720 	 && (length > 0)) {
4721 		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
4722 		ctsio->kern_data_len = length;
4723 		ctsio->kern_total_len = length;
4724 		ctsio->kern_data_resid = 0;
4725 		ctsio->kern_rel_offset = 0;
4726 		ctsio->kern_sg_entries = 0;
4727 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
4728 		ctsio->be_move_done = ctl_config_move_done;
4729 		ctl_datamove((union ctl_io *)ctsio);
4730 
4731 		return (CTL_RETVAL_COMPLETE);
4732 	}
4733 
4734 	if (length > 0)
4735 		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
4736 
4737 	mtx_lock(&lun->lun_lock);
4738 
4739 	/*
4740 	 * According to SPC, it is not an error for an intiator to attempt
4741 	 * to release a reservation on a LUN that isn't reserved, or that
4742 	 * is reserved by another initiator.  The reservation can only be
4743 	 * released, though, by the initiator who made it or by one of
4744 	 * several reset type events.
4745 	 */
4746 	if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
4747 			lun->flags &= ~CTL_LUN_RESERVED;
4748 
4749 	mtx_unlock(&lun->lun_lock);
4750 
4751 	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
4752 		free(ctsio->kern_data_ptr, M_CTL);
4753 		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
4754 	}
4755 
4756 	ctl_set_success(ctsio);
4757 	ctl_done((union ctl_io *)ctsio);
4758 	return (CTL_RETVAL_COMPLETE);
4759 }
4760 
4761 int
4762 ctl_scsi_reserve(struct ctl_scsiio *ctsio)
4763 {
4764 	int extent, thirdparty, longid;
4765 	int resv_id, length;
4766 	uint64_t thirdparty_id;
4767 	struct ctl_lun *lun;
4768 	uint32_t residx;
4769 
4770 	extent = 0;
4771 	thirdparty = 0;
4772 	longid = 0;
4773 	resv_id = 0;
4774 	length = 0;
4775 	thirdparty_id = 0;
4776 
4777 	CTL_DEBUG_PRINT(("ctl_reserve\n"));
4778 
4779 	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
4780 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
4781 
4782 	switch (ctsio->cdb[0]) {
4783 	case RESERVE_10: {
4784 		struct scsi_reserve_10 *cdb;
4785 
4786 		cdb = (struct scsi_reserve_10 *)ctsio->cdb;
4787 
4788 		if (cdb->byte2 & SR10_LONGID)
4789 			longid = 1;
4790 		else
4791 			thirdparty_id = cdb->thirdparty_id;
4792 
4793 		resv_id = cdb->resv_id;
4794 		length = scsi_2btoul(cdb->length);
4795 		break;
4796 	}
4797 	}
4798 
4799 	/*
4800 	 * XXX KDM right now, we only support LUN reservation.  We don't
4801 	 * support 3rd party reservations, or extent reservations, which
4802 	 * might actually need the parameter list.  If we've gotten this
4803 	 * far, we've got a LUN reservation.  Anything else got kicked out
4804 	 * above.  So, according to SPC, ignore the length.
4805 	 */
4806 	length = 0;
4807 
4808 	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
4809 	 && (length > 0)) {
4810 		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
4811 		ctsio->kern_data_len = length;
4812 		ctsio->kern_total_len = length;
4813 		ctsio->kern_data_resid = 0;
4814 		ctsio->kern_rel_offset = 0;
4815 		ctsio->kern_sg_entries = 0;
4816 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
4817 		ctsio->be_move_done = ctl_config_move_done;
4818 		ctl_datamove((union ctl_io *)ctsio);
4819 
4820 		return (CTL_RETVAL_COMPLETE);
4821 	}
4822 
4823 	if (length > 0)
4824 		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
4825 
4826 	mtx_lock(&lun->lun_lock);
4827 	if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx != residx)) {
4828 		ctl_set_reservation_conflict(ctsio);
4829 		goto bailout;
4830 	}
4831 
4832 	lun->flags |= CTL_LUN_RESERVED;
4833 	lun->res_idx = residx;
4834 
4835 	ctl_set_success(ctsio);
4836 
4837 bailout:
4838 	mtx_unlock(&lun->lun_lock);
4839 
4840 	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
4841 		free(ctsio->kern_data_ptr, M_CTL);
4842 		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
4843 	}
4844 
4845 	ctl_done((union ctl_io *)ctsio);
4846 	return (CTL_RETVAL_COMPLETE);
4847 }
4848 
4849 int
4850 ctl_start_stop(struct ctl_scsiio *ctsio)
4851 {
4852 	struct scsi_start_stop_unit *cdb;
4853 	struct ctl_lun *lun;
4854 	int retval;
4855 
4856 	CTL_DEBUG_PRINT(("ctl_start_stop\n"));
4857 
4858 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
4859 	retval = 0;
4860 
4861 	cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
4862 
4863 	/*
4864 	 * XXX KDM
4865 	 * We don't support the immediate bit on a stop unit.  In order to
4866 	 * do that, we would need to code up a way to know that a stop is
4867 	 * pending, and hold off any new commands until it completes, one
4868 	 * way or another.  Then we could accept or reject those commands
4869 	 * depending on its status.  We would almost need to do the reverse
4870 	 * of what we do below for an immediate start -- return the copy of
4871 	 * the ctl_io to the FETD with status to send to the host (and to
4872 	 * free the copy!) and then free the original I/O once the stop
4873 	 * actually completes.  That way, the OOA queue mechanism can work
4874 	 * to block commands that shouldn't proceed.  Another alternative
4875 	 * would be to put the copy in the queue in place of the original,
4876 	 * and return the original back to the caller.  That could be
4877 	 * slightly safer..
4878 	 */
4879 	if ((cdb->byte2 & SSS_IMMED)
4880 	 && ((cdb->how & SSS_START) == 0)) {
4881 		ctl_set_invalid_field(ctsio,
4882 				      /*sks_valid*/ 1,
4883 				      /*command*/ 1,
4884 				      /*field*/ 1,
4885 				      /*bit_valid*/ 1,
4886 				      /*bit*/ 0);
4887 		ctl_done((union ctl_io *)ctsio);
4888 		return (CTL_RETVAL_COMPLETE);
4889 	}
4890 
4891 	if ((lun->flags & CTL_LUN_PR_RESERVED)
4892 	 && ((cdb->how & SSS_START)==0)) {
4893 		uint32_t residx;
4894 
4895 		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
4896 		if (ctl_get_prkey(lun, residx) == 0
4897 		 || (lun->pr_res_idx!=residx && lun->res_type < 4)) {
4898 
4899 			ctl_set_reservation_conflict(ctsio);
4900 			ctl_done((union ctl_io *)ctsio);
4901 			return (CTL_RETVAL_COMPLETE);
4902 		}
4903 	}
4904 
4905 	/*
4906 	 * If there is no backend on this device, we can't start or stop
4907 	 * it.  In theory we shouldn't get any start/stop commands in the
4908 	 * first place at this level if the LUN doesn't have a backend.
4909 	 * That should get stopped by the command decode code.
4910 	 */
4911 	if (lun->backend == NULL) {
4912 		ctl_set_invalid_opcode(ctsio);
4913 		ctl_done((union ctl_io *)ctsio);
4914 		return (CTL_RETVAL_COMPLETE);
4915 	}
4916 
4917 	/*
4918 	 * XXX KDM Copan-specific offline behavior.
4919 	 * Figure out a reasonable way to port this?
4920 	 */
4921 #ifdef NEEDTOPORT
4922 	mtx_lock(&lun->lun_lock);
4923 
4924 	if (((cdb->byte2 & SSS_ONOFFLINE) == 0)
4925 	 && (lun->flags & CTL_LUN_OFFLINE)) {
4926 		/*
4927 		 * If the LUN is offline, and the on/offline bit isn't set,
4928 		 * reject the start or stop.  Otherwise, let it through.
4929 		 */
4930 		mtx_unlock(&lun->lun_lock);
4931 		ctl_set_lun_not_ready(ctsio);
4932 		ctl_done((union ctl_io *)ctsio);
4933 	} else {
4934 		mtx_unlock(&lun->lun_lock);
4935 #endif /* NEEDTOPORT */
4936 		/*
4937 		 * This could be a start or a stop when we're online,
4938 		 * or a stop/offline or start/online.  A start or stop when
4939 		 * we're offline is covered in the case above.
4940 		 */
4941 		/*
4942 		 * In the non-immediate case, we send the request to
4943 		 * the backend and return status to the user when
4944 		 * it is done.
4945 		 *
4946 		 * In the immediate case, we allocate a new ctl_io
4947 		 * to hold a copy of the request, and send that to
4948 		 * the backend.  We then set good status on the
4949 		 * user's request and return it immediately.
4950 		 */
4951 		if (cdb->byte2 & SSS_IMMED) {
4952 			union ctl_io *new_io;
4953 
4954 			new_io = ctl_alloc_io(ctsio->io_hdr.pool);
4955 			ctl_copy_io((union ctl_io *)ctsio, new_io);
4956 			retval = lun->backend->config_write(new_io);
4957 			ctl_set_success(ctsio);
4958 			ctl_done((union ctl_io *)ctsio);
4959 		} else {
4960 			retval = lun->backend->config_write(
4961 				(union ctl_io *)ctsio);
4962 		}
4963 #ifdef NEEDTOPORT
4964 	}
4965 #endif
4966 	return (retval);
4967 }
4968 
4969 /*
4970  * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
4971  * we don't really do anything with the LBA and length fields if the user
4972  * passes them in.  Instead we'll just flush out the cache for the entire
4973  * LUN.
4974  */
4975 int
4976 ctl_sync_cache(struct ctl_scsiio *ctsio)
4977 {
4978 	struct ctl_lun *lun;
4979 	struct ctl_softc *softc;
4980 	struct ctl_lba_len_flags *lbalen;
4981 	uint64_t starting_lba;
4982 	uint32_t block_count;
4983 	int retval;
4984 	uint8_t byte2;
4985 
4986 	CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
4987 
4988 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
4989 	softc = lun->ctl_softc;
4990 	retval = 0;
4991 
4992 	switch (ctsio->cdb[0]) {
4993 	case SYNCHRONIZE_CACHE: {
4994 		struct scsi_sync_cache *cdb;
4995 		cdb = (struct scsi_sync_cache *)ctsio->cdb;
4996 
4997 		starting_lba = scsi_4btoul(cdb->begin_lba);
4998 		block_count = scsi_2btoul(cdb->lb_count);
4999 		byte2 = cdb->byte2;
5000 		break;
5001 	}
5002 	case SYNCHRONIZE_CACHE_16: {
5003 		struct scsi_sync_cache_16 *cdb;
5004 		cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5005 
5006 		starting_lba = scsi_8btou64(cdb->begin_lba);
5007 		block_count = scsi_4btoul(cdb->lb_count);
5008 		byte2 = cdb->byte2;
5009 		break;
5010 	}
5011 	default:
5012 		ctl_set_invalid_opcode(ctsio);
5013 		ctl_done((union ctl_io *)ctsio);
5014 		goto bailout;
5015 		break; /* NOTREACHED */
5016 	}
5017 
5018 	/*
5019 	 * We check the LBA and length, but don't do anything with them.
5020 	 * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5021 	 * get flushed.  This check will just help satisfy anyone who wants
5022 	 * to see an error for an out of range LBA.
5023 	 */
5024 	if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5025 		ctl_set_lba_out_of_range(ctsio);
5026 		ctl_done((union ctl_io *)ctsio);
5027 		goto bailout;
5028 	}
5029 
5030 	/*
5031 	 * If this LUN has no backend, we can't flush the cache anyway.
5032 	 */
5033 	if (lun->backend == NULL) {
5034 		ctl_set_invalid_opcode(ctsio);
5035 		ctl_done((union ctl_io *)ctsio);
5036 		goto bailout;
5037 	}
5038 
5039 	lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5040 	lbalen->lba = starting_lba;
5041 	lbalen->len = block_count;
5042 	lbalen->flags = byte2;
5043 
5044 	/*
5045 	 * Check to see whether we're configured to send the SYNCHRONIZE
5046 	 * CACHE command directly to the back end.
5047 	 */
5048 	mtx_lock(&lun->lun_lock);
5049 	if ((softc->flags & CTL_FLAG_REAL_SYNC)
5050 	 && (++(lun->sync_count) >= lun->sync_interval)) {
5051 		lun->sync_count = 0;
5052 		mtx_unlock(&lun->lun_lock);
5053 		retval = lun->backend->config_write((union ctl_io *)ctsio);
5054 	} else {
5055 		mtx_unlock(&lun->lun_lock);
5056 		ctl_set_success(ctsio);
5057 		ctl_done((union ctl_io *)ctsio);
5058 	}
5059 
5060 bailout:
5061 
5062 	return (retval);
5063 }
5064 
5065 int
5066 ctl_format(struct ctl_scsiio *ctsio)
5067 {
5068 	struct scsi_format *cdb;
5069 	struct ctl_lun *lun;
5070 	int length, defect_list_len;
5071 
5072 	CTL_DEBUG_PRINT(("ctl_format\n"));
5073 
5074 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5075 
5076 	cdb = (struct scsi_format *)ctsio->cdb;
5077 
5078 	length = 0;
5079 	if (cdb->byte2 & SF_FMTDATA) {
5080 		if (cdb->byte2 & SF_LONGLIST)
5081 			length = sizeof(struct scsi_format_header_long);
5082 		else
5083 			length = sizeof(struct scsi_format_header_short);
5084 	}
5085 
5086 	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5087 	 && (length > 0)) {
5088 		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5089 		ctsio->kern_data_len = length;
5090 		ctsio->kern_total_len = length;
5091 		ctsio->kern_data_resid = 0;
5092 		ctsio->kern_rel_offset = 0;
5093 		ctsio->kern_sg_entries = 0;
5094 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5095 		ctsio->be_move_done = ctl_config_move_done;
5096 		ctl_datamove((union ctl_io *)ctsio);
5097 
5098 		return (CTL_RETVAL_COMPLETE);
5099 	}
5100 
5101 	defect_list_len = 0;
5102 
5103 	if (cdb->byte2 & SF_FMTDATA) {
5104 		if (cdb->byte2 & SF_LONGLIST) {
5105 			struct scsi_format_header_long *header;
5106 
5107 			header = (struct scsi_format_header_long *)
5108 				ctsio->kern_data_ptr;
5109 
5110 			defect_list_len = scsi_4btoul(header->defect_list_len);
5111 			if (defect_list_len != 0) {
5112 				ctl_set_invalid_field(ctsio,
5113 						      /*sks_valid*/ 1,
5114 						      /*command*/ 0,
5115 						      /*field*/ 2,
5116 						      /*bit_valid*/ 0,
5117 						      /*bit*/ 0);
5118 				goto bailout;
5119 			}
5120 		} else {
5121 			struct scsi_format_header_short *header;
5122 
5123 			header = (struct scsi_format_header_short *)
5124 				ctsio->kern_data_ptr;
5125 
5126 			defect_list_len = scsi_2btoul(header->defect_list_len);
5127 			if (defect_list_len != 0) {
5128 				ctl_set_invalid_field(ctsio,
5129 						      /*sks_valid*/ 1,
5130 						      /*command*/ 0,
5131 						      /*field*/ 2,
5132 						      /*bit_valid*/ 0,
5133 						      /*bit*/ 0);
5134 				goto bailout;
5135 			}
5136 		}
5137 	}
5138 
5139 	/*
5140 	 * The format command will clear out the "Medium format corrupted"
5141 	 * status if set by the configuration code.  That status is really
5142 	 * just a way to notify the host that we have lost the media, and
5143 	 * get them to issue a command that will basically make them think
5144 	 * they're blowing away the media.
5145 	 */
5146 	mtx_lock(&lun->lun_lock);
5147 	lun->flags &= ~CTL_LUN_INOPERABLE;
5148 	mtx_unlock(&lun->lun_lock);
5149 
5150 	ctl_set_success(ctsio);
5151 bailout:
5152 
5153 	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5154 		free(ctsio->kern_data_ptr, M_CTL);
5155 		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5156 	}
5157 
5158 	ctl_done((union ctl_io *)ctsio);
5159 	return (CTL_RETVAL_COMPLETE);
5160 }
5161 
5162 int
5163 ctl_read_buffer(struct ctl_scsiio *ctsio)
5164 {
5165 	struct scsi_read_buffer *cdb;
5166 	struct ctl_lun *lun;
5167 	int buffer_offset, len;
5168 	static uint8_t descr[4];
5169 	static uint8_t echo_descr[4] = { 0 };
5170 
5171 	CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5172 
5173 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5174 	cdb = (struct scsi_read_buffer *)ctsio->cdb;
5175 
5176 	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA &&
5177 	    (cdb->byte2 & RWB_MODE) != RWB_MODE_ECHO_DESCR &&
5178 	    (cdb->byte2 & RWB_MODE) != RWB_MODE_DESCR) {
5179 		ctl_set_invalid_field(ctsio,
5180 				      /*sks_valid*/ 1,
5181 				      /*command*/ 1,
5182 				      /*field*/ 1,
5183 				      /*bit_valid*/ 1,
5184 				      /*bit*/ 4);
5185 		ctl_done((union ctl_io *)ctsio);
5186 		return (CTL_RETVAL_COMPLETE);
5187 	}
5188 
5189 	len = scsi_3btoul(cdb->length);
5190 	buffer_offset = scsi_3btoul(cdb->offset);
5191 
5192 	if (buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5193 		ctl_set_invalid_field(ctsio,
5194 				      /*sks_valid*/ 1,
5195 				      /*command*/ 1,
5196 				      /*field*/ 6,
5197 				      /*bit_valid*/ 0,
5198 				      /*bit*/ 0);
5199 		ctl_done((union ctl_io *)ctsio);
5200 		return (CTL_RETVAL_COMPLETE);
5201 	}
5202 
5203 	if ((cdb->byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5204 		descr[0] = 0;
5205 		scsi_ulto3b(CTL_WRITE_BUFFER_SIZE, &descr[1]);
5206 		ctsio->kern_data_ptr = descr;
5207 		len = min(len, sizeof(descr));
5208 	} else if ((cdb->byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5209 		ctsio->kern_data_ptr = echo_descr;
5210 		len = min(len, sizeof(echo_descr));
5211 	} else {
5212 		if (lun->write_buffer == NULL) {
5213 			lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5214 			    M_CTL, M_WAITOK);
5215 		}
5216 		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5217 	}
5218 	ctsio->kern_data_len = len;
5219 	ctsio->kern_total_len = len;
5220 	ctsio->kern_data_resid = 0;
5221 	ctsio->kern_rel_offset = 0;
5222 	ctsio->kern_sg_entries = 0;
5223 	ctl_set_success(ctsio);
5224 	ctsio->be_move_done = ctl_config_move_done;
5225 	ctl_datamove((union ctl_io *)ctsio);
5226 	return (CTL_RETVAL_COMPLETE);
5227 }
5228 
5229 int
5230 ctl_write_buffer(struct ctl_scsiio *ctsio)
5231 {
5232 	struct scsi_write_buffer *cdb;
5233 	struct ctl_lun *lun;
5234 	int buffer_offset, len;
5235 
5236 	CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5237 
5238 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5239 	cdb = (struct scsi_write_buffer *)ctsio->cdb;
5240 
5241 	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5242 		ctl_set_invalid_field(ctsio,
5243 				      /*sks_valid*/ 1,
5244 				      /*command*/ 1,
5245 				      /*field*/ 1,
5246 				      /*bit_valid*/ 1,
5247 				      /*bit*/ 4);
5248 		ctl_done((union ctl_io *)ctsio);
5249 		return (CTL_RETVAL_COMPLETE);
5250 	}
5251 
5252 	len = scsi_3btoul(cdb->length);
5253 	buffer_offset = scsi_3btoul(cdb->offset);
5254 
5255 	if (buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5256 		ctl_set_invalid_field(ctsio,
5257 				      /*sks_valid*/ 1,
5258 				      /*command*/ 1,
5259 				      /*field*/ 6,
5260 				      /*bit_valid*/ 0,
5261 				      /*bit*/ 0);
5262 		ctl_done((union ctl_io *)ctsio);
5263 		return (CTL_RETVAL_COMPLETE);
5264 	}
5265 
5266 	/*
5267 	 * If we've got a kernel request that hasn't been malloced yet,
5268 	 * malloc it and tell the caller the data buffer is here.
5269 	 */
5270 	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5271 		if (lun->write_buffer == NULL) {
5272 			lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5273 			    M_CTL, M_WAITOK);
5274 		}
5275 		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5276 		ctsio->kern_data_len = len;
5277 		ctsio->kern_total_len = len;
5278 		ctsio->kern_data_resid = 0;
5279 		ctsio->kern_rel_offset = 0;
5280 		ctsio->kern_sg_entries = 0;
5281 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5282 		ctsio->be_move_done = ctl_config_move_done;
5283 		ctl_datamove((union ctl_io *)ctsio);
5284 
5285 		return (CTL_RETVAL_COMPLETE);
5286 	}
5287 
5288 	ctl_set_success(ctsio);
5289 	ctl_done((union ctl_io *)ctsio);
5290 	return (CTL_RETVAL_COMPLETE);
5291 }
5292 
5293 int
5294 ctl_write_same(struct ctl_scsiio *ctsio)
5295 {
5296 	struct ctl_lun *lun;
5297 	struct ctl_lba_len_flags *lbalen;
5298 	uint64_t lba;
5299 	uint32_t num_blocks;
5300 	int len, retval;
5301 	uint8_t byte2;
5302 
5303 	retval = CTL_RETVAL_COMPLETE;
5304 
5305 	CTL_DEBUG_PRINT(("ctl_write_same\n"));
5306 
5307 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5308 
5309 	switch (ctsio->cdb[0]) {
5310 	case WRITE_SAME_10: {
5311 		struct scsi_write_same_10 *cdb;
5312 
5313 		cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5314 
5315 		lba = scsi_4btoul(cdb->addr);
5316 		num_blocks = scsi_2btoul(cdb->length);
5317 		byte2 = cdb->byte2;
5318 		break;
5319 	}
5320 	case WRITE_SAME_16: {
5321 		struct scsi_write_same_16 *cdb;
5322 
5323 		cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5324 
5325 		lba = scsi_8btou64(cdb->addr);
5326 		num_blocks = scsi_4btoul(cdb->length);
5327 		byte2 = cdb->byte2;
5328 		break;
5329 	}
5330 	default:
5331 		/*
5332 		 * We got a command we don't support.  This shouldn't
5333 		 * happen, commands should be filtered out above us.
5334 		 */
5335 		ctl_set_invalid_opcode(ctsio);
5336 		ctl_done((union ctl_io *)ctsio);
5337 
5338 		return (CTL_RETVAL_COMPLETE);
5339 		break; /* NOTREACHED */
5340 	}
5341 
5342 	/* NDOB and ANCHOR flags can be used only together with UNMAP */
5343 	if ((byte2 & SWS_UNMAP) == 0 &&
5344 	    (byte2 & (SWS_NDOB | SWS_ANCHOR)) != 0) {
5345 		ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
5346 		    /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
5347 		ctl_done((union ctl_io *)ctsio);
5348 		return (CTL_RETVAL_COMPLETE);
5349 	}
5350 
5351 	/*
5352 	 * The first check is to make sure we're in bounds, the second
5353 	 * check is to catch wrap-around problems.  If the lba + num blocks
5354 	 * is less than the lba, then we've wrapped around and the block
5355 	 * range is invalid anyway.
5356 	 */
5357 	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5358 	 || ((lba + num_blocks) < lba)) {
5359 		ctl_set_lba_out_of_range(ctsio);
5360 		ctl_done((union ctl_io *)ctsio);
5361 		return (CTL_RETVAL_COMPLETE);
5362 	}
5363 
5364 	/* Zero number of blocks means "to the last logical block" */
5365 	if (num_blocks == 0) {
5366 		if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5367 			ctl_set_invalid_field(ctsio,
5368 					      /*sks_valid*/ 0,
5369 					      /*command*/ 1,
5370 					      /*field*/ 0,
5371 					      /*bit_valid*/ 0,
5372 					      /*bit*/ 0);
5373 			ctl_done((union ctl_io *)ctsio);
5374 			return (CTL_RETVAL_COMPLETE);
5375 		}
5376 		num_blocks = (lun->be_lun->maxlba + 1) - lba;
5377 	}
5378 
5379 	len = lun->be_lun->blocksize;
5380 
5381 	/*
5382 	 * If we've got a kernel request that hasn't been malloced yet,
5383 	 * malloc it and tell the caller the data buffer is here.
5384 	 */
5385 	if ((byte2 & SWS_NDOB) == 0 &&
5386 	    (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5387 		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
5388 		ctsio->kern_data_len = len;
5389 		ctsio->kern_total_len = len;
5390 		ctsio->kern_data_resid = 0;
5391 		ctsio->kern_rel_offset = 0;
5392 		ctsio->kern_sg_entries = 0;
5393 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5394 		ctsio->be_move_done = ctl_config_move_done;
5395 		ctl_datamove((union ctl_io *)ctsio);
5396 
5397 		return (CTL_RETVAL_COMPLETE);
5398 	}
5399 
5400 	lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5401 	lbalen->lba = lba;
5402 	lbalen->len = num_blocks;
5403 	lbalen->flags = byte2;
5404 	retval = lun->backend->config_write((union ctl_io *)ctsio);
5405 
5406 	return (retval);
5407 }
5408 
5409 int
5410 ctl_unmap(struct ctl_scsiio *ctsio)
5411 {
5412 	struct ctl_lun *lun;
5413 	struct scsi_unmap *cdb;
5414 	struct ctl_ptr_len_flags *ptrlen;
5415 	struct scsi_unmap_header *hdr;
5416 	struct scsi_unmap_desc *buf, *end, *endnz, *range;
5417 	uint64_t lba;
5418 	uint32_t num_blocks;
5419 	int len, retval;
5420 	uint8_t byte2;
5421 
5422 	retval = CTL_RETVAL_COMPLETE;
5423 
5424 	CTL_DEBUG_PRINT(("ctl_unmap\n"));
5425 
5426 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5427 	cdb = (struct scsi_unmap *)ctsio->cdb;
5428 
5429 	len = scsi_2btoul(cdb->length);
5430 	byte2 = cdb->byte2;
5431 
5432 	/*
5433 	 * If we've got a kernel request that hasn't been malloced yet,
5434 	 * malloc it and tell the caller the data buffer is here.
5435 	 */
5436 	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5437 		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
5438 		ctsio->kern_data_len = len;
5439 		ctsio->kern_total_len = len;
5440 		ctsio->kern_data_resid = 0;
5441 		ctsio->kern_rel_offset = 0;
5442 		ctsio->kern_sg_entries = 0;
5443 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5444 		ctsio->be_move_done = ctl_config_move_done;
5445 		ctl_datamove((union ctl_io *)ctsio);
5446 
5447 		return (CTL_RETVAL_COMPLETE);
5448 	}
5449 
5450 	len = ctsio->kern_total_len - ctsio->kern_data_resid;
5451 	hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
5452 	if (len < sizeof (*hdr) ||
5453 	    len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
5454 	    len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
5455 	    scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
5456 		ctl_set_invalid_field(ctsio,
5457 				      /*sks_valid*/ 0,
5458 				      /*command*/ 0,
5459 				      /*field*/ 0,
5460 				      /*bit_valid*/ 0,
5461 				      /*bit*/ 0);
5462 		goto done;
5463 	}
5464 	len = scsi_2btoul(hdr->desc_length);
5465 	buf = (struct scsi_unmap_desc *)(hdr + 1);
5466 	end = buf + len / sizeof(*buf);
5467 
5468 	endnz = buf;
5469 	for (range = buf; range < end; range++) {
5470 		lba = scsi_8btou64(range->lba);
5471 		num_blocks = scsi_4btoul(range->length);
5472 		if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5473 		 || ((lba + num_blocks) < lba)) {
5474 			ctl_set_lba_out_of_range(ctsio);
5475 			ctl_done((union ctl_io *)ctsio);
5476 			return (CTL_RETVAL_COMPLETE);
5477 		}
5478 		if (num_blocks != 0)
5479 			endnz = range + 1;
5480 	}
5481 
5482 	/*
5483 	 * Block backend can not handle zero last range.
5484 	 * Filter it out and return if there is nothing left.
5485 	 */
5486 	len = (uint8_t *)endnz - (uint8_t *)buf;
5487 	if (len == 0) {
5488 		ctl_set_success(ctsio);
5489 		goto done;
5490 	}
5491 
5492 	mtx_lock(&lun->lun_lock);
5493 	ptrlen = (struct ctl_ptr_len_flags *)
5494 	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5495 	ptrlen->ptr = (void *)buf;
5496 	ptrlen->len = len;
5497 	ptrlen->flags = byte2;
5498 	ctl_check_blocked(lun);
5499 	mtx_unlock(&lun->lun_lock);
5500 
5501 	retval = lun->backend->config_write((union ctl_io *)ctsio);
5502 	return (retval);
5503 
5504 done:
5505 	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5506 		free(ctsio->kern_data_ptr, M_CTL);
5507 		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5508 	}
5509 	ctl_done((union ctl_io *)ctsio);
5510 	return (CTL_RETVAL_COMPLETE);
5511 }
5512 
5513 /*
5514  * Note that this function currently doesn't actually do anything inside
5515  * CTL to enforce things if the DQue bit is turned on.
5516  *
5517  * Also note that this function can't be used in the default case, because
5518  * the DQue bit isn't set in the changeable mask for the control mode page
5519  * anyway.  This is just here as an example for how to implement a page
5520  * handler, and a placeholder in case we want to allow the user to turn
5521  * tagged queueing on and off.
5522  *
5523  * The D_SENSE bit handling is functional, however, and will turn
5524  * descriptor sense on and off for a given LUN.
5525  */
5526 int
5527 ctl_control_page_handler(struct ctl_scsiio *ctsio,
5528 			 struct ctl_page_index *page_index, uint8_t *page_ptr)
5529 {
5530 	struct scsi_control_page *current_cp, *saved_cp, *user_cp;
5531 	struct ctl_lun *lun;
5532 	int set_ua;
5533 	uint32_t initidx;
5534 
5535 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5536 	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5537 	set_ua = 0;
5538 
5539 	user_cp = (struct scsi_control_page *)page_ptr;
5540 	current_cp = (struct scsi_control_page *)
5541 		(page_index->page_data + (page_index->page_len *
5542 		CTL_PAGE_CURRENT));
5543 	saved_cp = (struct scsi_control_page *)
5544 		(page_index->page_data + (page_index->page_len *
5545 		CTL_PAGE_SAVED));
5546 
5547 	mtx_lock(&lun->lun_lock);
5548 	if (((current_cp->rlec & SCP_DSENSE) == 0)
5549 	 && ((user_cp->rlec & SCP_DSENSE) != 0)) {
5550 		/*
5551 		 * Descriptor sense is currently turned off and the user
5552 		 * wants to turn it on.
5553 		 */
5554 		current_cp->rlec |= SCP_DSENSE;
5555 		saved_cp->rlec |= SCP_DSENSE;
5556 		lun->flags |= CTL_LUN_SENSE_DESC;
5557 		set_ua = 1;
5558 	} else if (((current_cp->rlec & SCP_DSENSE) != 0)
5559 		&& ((user_cp->rlec & SCP_DSENSE) == 0)) {
5560 		/*
5561 		 * Descriptor sense is currently turned on, and the user
5562 		 * wants to turn it off.
5563 		 */
5564 		current_cp->rlec &= ~SCP_DSENSE;
5565 		saved_cp->rlec &= ~SCP_DSENSE;
5566 		lun->flags &= ~CTL_LUN_SENSE_DESC;
5567 		set_ua = 1;
5568 	}
5569 	if ((current_cp->queue_flags & SCP_QUEUE_ALG_MASK) !=
5570 	    (user_cp->queue_flags & SCP_QUEUE_ALG_MASK)) {
5571 		current_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
5572 		current_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
5573 		saved_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
5574 		saved_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
5575 		set_ua = 1;
5576 	}
5577 	if ((current_cp->eca_and_aen & SCP_SWP) !=
5578 	    (user_cp->eca_and_aen & SCP_SWP)) {
5579 		current_cp->eca_and_aen &= ~SCP_SWP;
5580 		current_cp->eca_and_aen |= user_cp->eca_and_aen & SCP_SWP;
5581 		saved_cp->eca_and_aen &= ~SCP_SWP;
5582 		saved_cp->eca_and_aen |= user_cp->eca_and_aen & SCP_SWP;
5583 		set_ua = 1;
5584 	}
5585 	if (set_ua != 0)
5586 		ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
5587 	mtx_unlock(&lun->lun_lock);
5588 
5589 	return (0);
5590 }
5591 
5592 int
5593 ctl_caching_sp_handler(struct ctl_scsiio *ctsio,
5594 		     struct ctl_page_index *page_index, uint8_t *page_ptr)
5595 {
5596 	struct scsi_caching_page *current_cp, *saved_cp, *user_cp;
5597 	struct ctl_lun *lun;
5598 	int set_ua;
5599 	uint32_t initidx;
5600 
5601 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5602 	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5603 	set_ua = 0;
5604 
5605 	user_cp = (struct scsi_caching_page *)page_ptr;
5606 	current_cp = (struct scsi_caching_page *)
5607 		(page_index->page_data + (page_index->page_len *
5608 		CTL_PAGE_CURRENT));
5609 	saved_cp = (struct scsi_caching_page *)
5610 		(page_index->page_data + (page_index->page_len *
5611 		CTL_PAGE_SAVED));
5612 
5613 	mtx_lock(&lun->lun_lock);
5614 	if ((current_cp->flags1 & (SCP_WCE | SCP_RCD)) !=
5615 	    (user_cp->flags1 & (SCP_WCE | SCP_RCD))) {
5616 		current_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
5617 		current_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
5618 		saved_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
5619 		saved_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
5620 		set_ua = 1;
5621 	}
5622 	if (set_ua != 0)
5623 		ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
5624 	mtx_unlock(&lun->lun_lock);
5625 
5626 	return (0);
5627 }
5628 
5629 int
5630 ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio,
5631 				struct ctl_page_index *page_index,
5632 				uint8_t *page_ptr)
5633 {
5634 	uint8_t *c;
5635 	int i;
5636 
5637 	c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs;
5638 	ctl_time_io_secs =
5639 		(c[0] << 8) |
5640 		(c[1] << 0) |
5641 		0;
5642 	CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs));
5643 	printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs);
5644 	printf("page data:");
5645 	for (i=0; i<8; i++)
5646 		printf(" %.2x",page_ptr[i]);
5647 	printf("\n");
5648 	return (0);
5649 }
5650 
5651 int
5652 ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio,
5653 			       struct ctl_page_index *page_index,
5654 			       int pc)
5655 {
5656 	struct copan_debugconf_subpage *page;
5657 
5658 	page = (struct copan_debugconf_subpage *)page_index->page_data +
5659 		(page_index->page_len * pc);
5660 
5661 	switch (pc) {
5662 	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
5663 	case SMS_PAGE_CTRL_DEFAULT >> 6:
5664 	case SMS_PAGE_CTRL_SAVED >> 6:
5665 		/*
5666 		 * We don't update the changable or default bits for this page.
5667 		 */
5668 		break;
5669 	case SMS_PAGE_CTRL_CURRENT >> 6:
5670 		page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8;
5671 		page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0;
5672 		break;
5673 	default:
5674 #ifdef NEEDTOPORT
5675 		EPRINT(0, "Invalid PC %d!!", pc);
5676 #endif /* NEEDTOPORT */
5677 		break;
5678 	}
5679 	return (0);
5680 }
5681 
5682 
5683 static int
5684 ctl_do_mode_select(union ctl_io *io)
5685 {
5686 	struct scsi_mode_page_header *page_header;
5687 	struct ctl_page_index *page_index;
5688 	struct ctl_scsiio *ctsio;
5689 	int control_dev, page_len;
5690 	int page_len_offset, page_len_size;
5691 	union ctl_modepage_info *modepage_info;
5692 	struct ctl_lun *lun;
5693 	int *len_left, *len_used;
5694 	int retval, i;
5695 
5696 	ctsio = &io->scsiio;
5697 	page_index = NULL;
5698 	page_len = 0;
5699 	retval = CTL_RETVAL_COMPLETE;
5700 
5701 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5702 
5703 	if (lun->be_lun->lun_type != T_DIRECT)
5704 		control_dev = 1;
5705 	else
5706 		control_dev = 0;
5707 
5708 	modepage_info = (union ctl_modepage_info *)
5709 		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
5710 	len_left = &modepage_info->header.len_left;
5711 	len_used = &modepage_info->header.len_used;
5712 
5713 do_next_page:
5714 
5715 	page_header = (struct scsi_mode_page_header *)
5716 		(ctsio->kern_data_ptr + *len_used);
5717 
5718 	if (*len_left == 0) {
5719 		free(ctsio->kern_data_ptr, M_CTL);
5720 		ctl_set_success(ctsio);
5721 		ctl_done((union ctl_io *)ctsio);
5722 		return (CTL_RETVAL_COMPLETE);
5723 	} else if (*len_left < sizeof(struct scsi_mode_page_header)) {
5724 
5725 		free(ctsio->kern_data_ptr, M_CTL);
5726 		ctl_set_param_len_error(ctsio);
5727 		ctl_done((union ctl_io *)ctsio);
5728 		return (CTL_RETVAL_COMPLETE);
5729 
5730 	} else if ((page_header->page_code & SMPH_SPF)
5731 		&& (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
5732 
5733 		free(ctsio->kern_data_ptr, M_CTL);
5734 		ctl_set_param_len_error(ctsio);
5735 		ctl_done((union ctl_io *)ctsio);
5736 		return (CTL_RETVAL_COMPLETE);
5737 	}
5738 
5739 
5740 	/*
5741 	 * XXX KDM should we do something with the block descriptor?
5742 	 */
5743 	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
5744 
5745 		if ((control_dev != 0)
5746 		 && (lun->mode_pages.index[i].page_flags &
5747 		     CTL_PAGE_FLAG_DISK_ONLY))
5748 			continue;
5749 
5750 		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
5751 		    (page_header->page_code & SMPH_PC_MASK))
5752 			continue;
5753 
5754 		/*
5755 		 * If neither page has a subpage code, then we've got a
5756 		 * match.
5757 		 */
5758 		if (((lun->mode_pages.index[i].page_code & SMPH_SPF) == 0)
5759 		 && ((page_header->page_code & SMPH_SPF) == 0)) {
5760 			page_index = &lun->mode_pages.index[i];
5761 			page_len = page_header->page_length;
5762 			break;
5763 		}
5764 
5765 		/*
5766 		 * If both pages have subpages, then the subpage numbers
5767 		 * have to match.
5768 		 */
5769 		if ((lun->mode_pages.index[i].page_code & SMPH_SPF)
5770 		  && (page_header->page_code & SMPH_SPF)) {
5771 			struct scsi_mode_page_header_sp *sph;
5772 
5773 			sph = (struct scsi_mode_page_header_sp *)page_header;
5774 
5775 			if (lun->mode_pages.index[i].subpage ==
5776 			    sph->subpage) {
5777 				page_index = &lun->mode_pages.index[i];
5778 				page_len = scsi_2btoul(sph->page_length);
5779 				break;
5780 			}
5781 		}
5782 	}
5783 
5784 	/*
5785 	 * If we couldn't find the page, or if we don't have a mode select
5786 	 * handler for it, send back an error to the user.
5787 	 */
5788 	if ((page_index == NULL)
5789 	 || (page_index->select_handler == NULL)) {
5790 		ctl_set_invalid_field(ctsio,
5791 				      /*sks_valid*/ 1,
5792 				      /*command*/ 0,
5793 				      /*field*/ *len_used,
5794 				      /*bit_valid*/ 0,
5795 				      /*bit*/ 0);
5796 		free(ctsio->kern_data_ptr, M_CTL);
5797 		ctl_done((union ctl_io *)ctsio);
5798 		return (CTL_RETVAL_COMPLETE);
5799 	}
5800 
5801 	if (page_index->page_code & SMPH_SPF) {
5802 		page_len_offset = 2;
5803 		page_len_size = 2;
5804 	} else {
5805 		page_len_size = 1;
5806 		page_len_offset = 1;
5807 	}
5808 
5809 	/*
5810 	 * If the length the initiator gives us isn't the one we specify in
5811 	 * the mode page header, or if they didn't specify enough data in
5812 	 * the CDB to avoid truncating this page, kick out the request.
5813 	 */
5814 	if ((page_len != (page_index->page_len - page_len_offset -
5815 			  page_len_size))
5816 	 || (*len_left < page_index->page_len)) {
5817 
5818 
5819 		ctl_set_invalid_field(ctsio,
5820 				      /*sks_valid*/ 1,
5821 				      /*command*/ 0,
5822 				      /*field*/ *len_used + page_len_offset,
5823 				      /*bit_valid*/ 0,
5824 				      /*bit*/ 0);
5825 		free(ctsio->kern_data_ptr, M_CTL);
5826 		ctl_done((union ctl_io *)ctsio);
5827 		return (CTL_RETVAL_COMPLETE);
5828 	}
5829 
5830 	/*
5831 	 * Run through the mode page, checking to make sure that the bits
5832 	 * the user changed are actually legal for him to change.
5833 	 */
5834 	for (i = 0; i < page_index->page_len; i++) {
5835 		uint8_t *user_byte, *change_mask, *current_byte;
5836 		int bad_bit;
5837 		int j;
5838 
5839 		user_byte = (uint8_t *)page_header + i;
5840 		change_mask = page_index->page_data +
5841 			      (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
5842 		current_byte = page_index->page_data +
5843 			       (page_index->page_len * CTL_PAGE_CURRENT) + i;
5844 
5845 		/*
5846 		 * Check to see whether the user set any bits in this byte
5847 		 * that he is not allowed to set.
5848 		 */
5849 		if ((*user_byte & ~(*change_mask)) ==
5850 		    (*current_byte & ~(*change_mask)))
5851 			continue;
5852 
5853 		/*
5854 		 * Go through bit by bit to determine which one is illegal.
5855 		 */
5856 		bad_bit = 0;
5857 		for (j = 7; j >= 0; j--) {
5858 			if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
5859 			    (((1 << i) & ~(*change_mask)) & *current_byte)) {
5860 				bad_bit = i;
5861 				break;
5862 			}
5863 		}
5864 		ctl_set_invalid_field(ctsio,
5865 				      /*sks_valid*/ 1,
5866 				      /*command*/ 0,
5867 				      /*field*/ *len_used + i,
5868 				      /*bit_valid*/ 1,
5869 				      /*bit*/ bad_bit);
5870 		free(ctsio->kern_data_ptr, M_CTL);
5871 		ctl_done((union ctl_io *)ctsio);
5872 		return (CTL_RETVAL_COMPLETE);
5873 	}
5874 
5875 	/*
5876 	 * Decrement these before we call the page handler, since we may
5877 	 * end up getting called back one way or another before the handler
5878 	 * returns to this context.
5879 	 */
5880 	*len_left -= page_index->page_len;
5881 	*len_used += page_index->page_len;
5882 
5883 	retval = page_index->select_handler(ctsio, page_index,
5884 					    (uint8_t *)page_header);
5885 
5886 	/*
5887 	 * If the page handler returns CTL_RETVAL_QUEUED, then we need to
5888 	 * wait until this queued command completes to finish processing
5889 	 * the mode page.  If it returns anything other than
5890 	 * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
5891 	 * already set the sense information, freed the data pointer, and
5892 	 * completed the io for us.
5893 	 */
5894 	if (retval != CTL_RETVAL_COMPLETE)
5895 		goto bailout_no_done;
5896 
5897 	/*
5898 	 * If the initiator sent us more than one page, parse the next one.
5899 	 */
5900 	if (*len_left > 0)
5901 		goto do_next_page;
5902 
5903 	ctl_set_success(ctsio);
5904 	free(ctsio->kern_data_ptr, M_CTL);
5905 	ctl_done((union ctl_io *)ctsio);
5906 
5907 bailout_no_done:
5908 
5909 	return (CTL_RETVAL_COMPLETE);
5910 
5911 }
5912 
5913 int
5914 ctl_mode_select(struct ctl_scsiio *ctsio)
5915 {
5916 	int param_len, pf, sp;
5917 	int header_size, bd_len;
5918 	int len_left, len_used;
5919 	struct ctl_page_index *page_index;
5920 	struct ctl_lun *lun;
5921 	int control_dev, page_len;
5922 	union ctl_modepage_info *modepage_info;
5923 	int retval;
5924 
5925 	pf = 0;
5926 	sp = 0;
5927 	page_len = 0;
5928 	len_used = 0;
5929 	len_left = 0;
5930 	retval = 0;
5931 	bd_len = 0;
5932 	page_index = NULL;
5933 
5934 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5935 
5936 	if (lun->be_lun->lun_type != T_DIRECT)
5937 		control_dev = 1;
5938 	else
5939 		control_dev = 0;
5940 
5941 	switch (ctsio->cdb[0]) {
5942 	case MODE_SELECT_6: {
5943 		struct scsi_mode_select_6 *cdb;
5944 
5945 		cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
5946 
5947 		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
5948 		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
5949 
5950 		param_len = cdb->length;
5951 		header_size = sizeof(struct scsi_mode_header_6);
5952 		break;
5953 	}
5954 	case MODE_SELECT_10: {
5955 		struct scsi_mode_select_10 *cdb;
5956 
5957 		cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
5958 
5959 		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
5960 		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
5961 
5962 		param_len = scsi_2btoul(cdb->length);
5963 		header_size = sizeof(struct scsi_mode_header_10);
5964 		break;
5965 	}
5966 	default:
5967 		ctl_set_invalid_opcode(ctsio);
5968 		ctl_done((union ctl_io *)ctsio);
5969 		return (CTL_RETVAL_COMPLETE);
5970 		break; /* NOTREACHED */
5971 	}
5972 
5973 	/*
5974 	 * From SPC-3:
5975 	 * "A parameter list length of zero indicates that the Data-Out Buffer
5976 	 * shall be empty. This condition shall not be considered as an error."
5977 	 */
5978 	if (param_len == 0) {
5979 		ctl_set_success(ctsio);
5980 		ctl_done((union ctl_io *)ctsio);
5981 		return (CTL_RETVAL_COMPLETE);
5982 	}
5983 
5984 	/*
5985 	 * Since we'll hit this the first time through, prior to
5986 	 * allocation, we don't need to free a data buffer here.
5987 	 */
5988 	if (param_len < header_size) {
5989 		ctl_set_param_len_error(ctsio);
5990 		ctl_done((union ctl_io *)ctsio);
5991 		return (CTL_RETVAL_COMPLETE);
5992 	}
5993 
5994 	/*
5995 	 * Allocate the data buffer and grab the user's data.  In theory,
5996 	 * we shouldn't have to sanity check the parameter list length here
5997 	 * because the maximum size is 64K.  We should be able to malloc
5998 	 * that much without too many problems.
5999 	 */
6000 	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6001 		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6002 		ctsio->kern_data_len = param_len;
6003 		ctsio->kern_total_len = param_len;
6004 		ctsio->kern_data_resid = 0;
6005 		ctsio->kern_rel_offset = 0;
6006 		ctsio->kern_sg_entries = 0;
6007 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6008 		ctsio->be_move_done = ctl_config_move_done;
6009 		ctl_datamove((union ctl_io *)ctsio);
6010 
6011 		return (CTL_RETVAL_COMPLETE);
6012 	}
6013 
6014 	switch (ctsio->cdb[0]) {
6015 	case MODE_SELECT_6: {
6016 		struct scsi_mode_header_6 *mh6;
6017 
6018 		mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6019 		bd_len = mh6->blk_desc_len;
6020 		break;
6021 	}
6022 	case MODE_SELECT_10: {
6023 		struct scsi_mode_header_10 *mh10;
6024 
6025 		mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6026 		bd_len = scsi_2btoul(mh10->blk_desc_len);
6027 		break;
6028 	}
6029 	default:
6030 		panic("Invalid CDB type %#x", ctsio->cdb[0]);
6031 		break;
6032 	}
6033 
6034 	if (param_len < (header_size + bd_len)) {
6035 		free(ctsio->kern_data_ptr, M_CTL);
6036 		ctl_set_param_len_error(ctsio);
6037 		ctl_done((union ctl_io *)ctsio);
6038 		return (CTL_RETVAL_COMPLETE);
6039 	}
6040 
6041 	/*
6042 	 * Set the IO_CONT flag, so that if this I/O gets passed to
6043 	 * ctl_config_write_done(), it'll get passed back to
6044 	 * ctl_do_mode_select() for further processing, or completion if
6045 	 * we're all done.
6046 	 */
6047 	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6048 	ctsio->io_cont = ctl_do_mode_select;
6049 
6050 	modepage_info = (union ctl_modepage_info *)
6051 		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6052 
6053 	memset(modepage_info, 0, sizeof(*modepage_info));
6054 
6055 	len_left = param_len - header_size - bd_len;
6056 	len_used = header_size + bd_len;
6057 
6058 	modepage_info->header.len_left = len_left;
6059 	modepage_info->header.len_used = len_used;
6060 
6061 	return (ctl_do_mode_select((union ctl_io *)ctsio));
6062 }
6063 
6064 int
6065 ctl_mode_sense(struct ctl_scsiio *ctsio)
6066 {
6067 	struct ctl_lun *lun;
6068 	int pc, page_code, dbd, llba, subpage;
6069 	int alloc_len, page_len, header_len, total_len;
6070 	struct scsi_mode_block_descr *block_desc;
6071 	struct ctl_page_index *page_index;
6072 	int control_dev;
6073 
6074 	dbd = 0;
6075 	llba = 0;
6076 	block_desc = NULL;
6077 	page_index = NULL;
6078 
6079 	CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6080 
6081 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6082 
6083 	if (lun->be_lun->lun_type != T_DIRECT)
6084 		control_dev = 1;
6085 	else
6086 		control_dev = 0;
6087 
6088 	switch (ctsio->cdb[0]) {
6089 	case MODE_SENSE_6: {
6090 		struct scsi_mode_sense_6 *cdb;
6091 
6092 		cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6093 
6094 		header_len = sizeof(struct scsi_mode_hdr_6);
6095 		if (cdb->byte2 & SMS_DBD)
6096 			dbd = 1;
6097 		else
6098 			header_len += sizeof(struct scsi_mode_block_descr);
6099 
6100 		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6101 		page_code = cdb->page & SMS_PAGE_CODE;
6102 		subpage = cdb->subpage;
6103 		alloc_len = cdb->length;
6104 		break;
6105 	}
6106 	case MODE_SENSE_10: {
6107 		struct scsi_mode_sense_10 *cdb;
6108 
6109 		cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6110 
6111 		header_len = sizeof(struct scsi_mode_hdr_10);
6112 
6113 		if (cdb->byte2 & SMS_DBD)
6114 			dbd = 1;
6115 		else
6116 			header_len += sizeof(struct scsi_mode_block_descr);
6117 		if (cdb->byte2 & SMS10_LLBAA)
6118 			llba = 1;
6119 		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6120 		page_code = cdb->page & SMS_PAGE_CODE;
6121 		subpage = cdb->subpage;
6122 		alloc_len = scsi_2btoul(cdb->length);
6123 		break;
6124 	}
6125 	default:
6126 		ctl_set_invalid_opcode(ctsio);
6127 		ctl_done((union ctl_io *)ctsio);
6128 		return (CTL_RETVAL_COMPLETE);
6129 		break; /* NOTREACHED */
6130 	}
6131 
6132 	/*
6133 	 * We have to make a first pass through to calculate the size of
6134 	 * the pages that match the user's query.  Then we allocate enough
6135 	 * memory to hold it, and actually copy the data into the buffer.
6136 	 */
6137 	switch (page_code) {
6138 	case SMS_ALL_PAGES_PAGE: {
6139 		int i;
6140 
6141 		page_len = 0;
6142 
6143 		/*
6144 		 * At the moment, values other than 0 and 0xff here are
6145 		 * reserved according to SPC-3.
6146 		 */
6147 		if ((subpage != SMS_SUBPAGE_PAGE_0)
6148 		 && (subpage != SMS_SUBPAGE_ALL)) {
6149 			ctl_set_invalid_field(ctsio,
6150 					      /*sks_valid*/ 1,
6151 					      /*command*/ 1,
6152 					      /*field*/ 3,
6153 					      /*bit_valid*/ 0,
6154 					      /*bit*/ 0);
6155 			ctl_done((union ctl_io *)ctsio);
6156 			return (CTL_RETVAL_COMPLETE);
6157 		}
6158 
6159 		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6160 			if ((control_dev != 0)
6161 			 && (lun->mode_pages.index[i].page_flags &
6162 			     CTL_PAGE_FLAG_DISK_ONLY))
6163 				continue;
6164 
6165 			/*
6166 			 * We don't use this subpage if the user didn't
6167 			 * request all subpages.
6168 			 */
6169 			if ((lun->mode_pages.index[i].subpage != 0)
6170 			 && (subpage == SMS_SUBPAGE_PAGE_0))
6171 				continue;
6172 
6173 #if 0
6174 			printf("found page %#x len %d\n",
6175 			       lun->mode_pages.index[i].page_code &
6176 			       SMPH_PC_MASK,
6177 			       lun->mode_pages.index[i].page_len);
6178 #endif
6179 			page_len += lun->mode_pages.index[i].page_len;
6180 		}
6181 		break;
6182 	}
6183 	default: {
6184 		int i;
6185 
6186 		page_len = 0;
6187 
6188 		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6189 			/* Look for the right page code */
6190 			if ((lun->mode_pages.index[i].page_code &
6191 			     SMPH_PC_MASK) != page_code)
6192 				continue;
6193 
6194 			/* Look for the right subpage or the subpage wildcard*/
6195 			if ((lun->mode_pages.index[i].subpage != subpage)
6196 			 && (subpage != SMS_SUBPAGE_ALL))
6197 				continue;
6198 
6199 			/* Make sure the page is supported for this dev type */
6200 			if ((control_dev != 0)
6201 			 && (lun->mode_pages.index[i].page_flags &
6202 			     CTL_PAGE_FLAG_DISK_ONLY))
6203 				continue;
6204 
6205 #if 0
6206 			printf("found page %#x len %d\n",
6207 			       lun->mode_pages.index[i].page_code &
6208 			       SMPH_PC_MASK,
6209 			       lun->mode_pages.index[i].page_len);
6210 #endif
6211 
6212 			page_len += lun->mode_pages.index[i].page_len;
6213 		}
6214 
6215 		if (page_len == 0) {
6216 			ctl_set_invalid_field(ctsio,
6217 					      /*sks_valid*/ 1,
6218 					      /*command*/ 1,
6219 					      /*field*/ 2,
6220 					      /*bit_valid*/ 1,
6221 					      /*bit*/ 5);
6222 			ctl_done((union ctl_io *)ctsio);
6223 			return (CTL_RETVAL_COMPLETE);
6224 		}
6225 		break;
6226 	}
6227 	}
6228 
6229 	total_len = header_len + page_len;
6230 #if 0
6231 	printf("header_len = %d, page_len = %d, total_len = %d\n",
6232 	       header_len, page_len, total_len);
6233 #endif
6234 
6235 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6236 	ctsio->kern_sg_entries = 0;
6237 	ctsio->kern_data_resid = 0;
6238 	ctsio->kern_rel_offset = 0;
6239 	if (total_len < alloc_len) {
6240 		ctsio->residual = alloc_len - total_len;
6241 		ctsio->kern_data_len = total_len;
6242 		ctsio->kern_total_len = total_len;
6243 	} else {
6244 		ctsio->residual = 0;
6245 		ctsio->kern_data_len = alloc_len;
6246 		ctsio->kern_total_len = alloc_len;
6247 	}
6248 
6249 	switch (ctsio->cdb[0]) {
6250 	case MODE_SENSE_6: {
6251 		struct scsi_mode_hdr_6 *header;
6252 
6253 		header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6254 
6255 		header->datalen = MIN(total_len - 1, 254);
6256 		if (control_dev == 0) {
6257 			header->dev_specific = 0x10; /* DPOFUA */
6258 			if ((lun->flags & CTL_LUN_READONLY) ||
6259 			    (lun->mode_pages.control_page[CTL_PAGE_CURRENT]
6260 			    .eca_and_aen & SCP_SWP) != 0)
6261 				    header->dev_specific |= 0x80; /* WP */
6262 		}
6263 		if (dbd)
6264 			header->block_descr_len = 0;
6265 		else
6266 			header->block_descr_len =
6267 				sizeof(struct scsi_mode_block_descr);
6268 		block_desc = (struct scsi_mode_block_descr *)&header[1];
6269 		break;
6270 	}
6271 	case MODE_SENSE_10: {
6272 		struct scsi_mode_hdr_10 *header;
6273 		int datalen;
6274 
6275 		header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6276 
6277 		datalen = MIN(total_len - 2, 65533);
6278 		scsi_ulto2b(datalen, header->datalen);
6279 		if (control_dev == 0) {
6280 			header->dev_specific = 0x10; /* DPOFUA */
6281 			if ((lun->flags & CTL_LUN_READONLY) ||
6282 			    (lun->mode_pages.control_page[CTL_PAGE_CURRENT]
6283 			    .eca_and_aen & SCP_SWP) != 0)
6284 				    header->dev_specific |= 0x80; /* WP */
6285 		}
6286 		if (dbd)
6287 			scsi_ulto2b(0, header->block_descr_len);
6288 		else
6289 			scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
6290 				    header->block_descr_len);
6291 		block_desc = (struct scsi_mode_block_descr *)&header[1];
6292 		break;
6293 	}
6294 	default:
6295 		panic("invalid CDB type %#x", ctsio->cdb[0]);
6296 		break; /* NOTREACHED */
6297 	}
6298 
6299 	/*
6300 	 * If we've got a disk, use its blocksize in the block
6301 	 * descriptor.  Otherwise, just set it to 0.
6302 	 */
6303 	if (dbd == 0) {
6304 		if (control_dev == 0)
6305 			scsi_ulto3b(lun->be_lun->blocksize,
6306 				    block_desc->block_len);
6307 		else
6308 			scsi_ulto3b(0, block_desc->block_len);
6309 	}
6310 
6311 	switch (page_code) {
6312 	case SMS_ALL_PAGES_PAGE: {
6313 		int i, data_used;
6314 
6315 		data_used = header_len;
6316 		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6317 			struct ctl_page_index *page_index;
6318 
6319 			page_index = &lun->mode_pages.index[i];
6320 
6321 			if ((control_dev != 0)
6322 			 && (page_index->page_flags &
6323 			    CTL_PAGE_FLAG_DISK_ONLY))
6324 				continue;
6325 
6326 			/*
6327 			 * We don't use this subpage if the user didn't
6328 			 * request all subpages.  We already checked (above)
6329 			 * to make sure the user only specified a subpage
6330 			 * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
6331 			 */
6332 			if ((page_index->subpage != 0)
6333 			 && (subpage == SMS_SUBPAGE_PAGE_0))
6334 				continue;
6335 
6336 			/*
6337 			 * Call the handler, if it exists, to update the
6338 			 * page to the latest values.
6339 			 */
6340 			if (page_index->sense_handler != NULL)
6341 				page_index->sense_handler(ctsio, page_index,pc);
6342 
6343 			memcpy(ctsio->kern_data_ptr + data_used,
6344 			       page_index->page_data +
6345 			       (page_index->page_len * pc),
6346 			       page_index->page_len);
6347 			data_used += page_index->page_len;
6348 		}
6349 		break;
6350 	}
6351 	default: {
6352 		int i, data_used;
6353 
6354 		data_used = header_len;
6355 
6356 		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6357 			struct ctl_page_index *page_index;
6358 
6359 			page_index = &lun->mode_pages.index[i];
6360 
6361 			/* Look for the right page code */
6362 			if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6363 				continue;
6364 
6365 			/* Look for the right subpage or the subpage wildcard*/
6366 			if ((page_index->subpage != subpage)
6367 			 && (subpage != SMS_SUBPAGE_ALL))
6368 				continue;
6369 
6370 			/* Make sure the page is supported for this dev type */
6371 			if ((control_dev != 0)
6372 			 && (page_index->page_flags &
6373 			     CTL_PAGE_FLAG_DISK_ONLY))
6374 				continue;
6375 
6376 			/*
6377 			 * Call the handler, if it exists, to update the
6378 			 * page to the latest values.
6379 			 */
6380 			if (page_index->sense_handler != NULL)
6381 				page_index->sense_handler(ctsio, page_index,pc);
6382 
6383 			memcpy(ctsio->kern_data_ptr + data_used,
6384 			       page_index->page_data +
6385 			       (page_index->page_len * pc),
6386 			       page_index->page_len);
6387 			data_used += page_index->page_len;
6388 		}
6389 		break;
6390 	}
6391 	}
6392 
6393 	ctl_set_success(ctsio);
6394 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6395 	ctsio->be_move_done = ctl_config_move_done;
6396 	ctl_datamove((union ctl_io *)ctsio);
6397 	return (CTL_RETVAL_COMPLETE);
6398 }
6399 
6400 int
6401 ctl_lbp_log_sense_handler(struct ctl_scsiio *ctsio,
6402 			       struct ctl_page_index *page_index,
6403 			       int pc)
6404 {
6405 	struct ctl_lun *lun;
6406 	struct scsi_log_param_header *phdr;
6407 	uint8_t *data;
6408 	uint64_t val;
6409 
6410 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6411 	data = page_index->page_data;
6412 
6413 	if (lun->backend->lun_attr != NULL &&
6414 	    (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksavail"))
6415 	     != UINT64_MAX) {
6416 		phdr = (struct scsi_log_param_header *)data;
6417 		scsi_ulto2b(0x0001, phdr->param_code);
6418 		phdr->param_control = SLP_LBIN | SLP_LP;
6419 		phdr->param_len = 8;
6420 		data = (uint8_t *)(phdr + 1);
6421 		scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6422 		data[4] = 0x02; /* per-pool */
6423 		data += phdr->param_len;
6424 	}
6425 
6426 	if (lun->backend->lun_attr != NULL &&
6427 	    (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksused"))
6428 	     != UINT64_MAX) {
6429 		phdr = (struct scsi_log_param_header *)data;
6430 		scsi_ulto2b(0x0002, phdr->param_code);
6431 		phdr->param_control = SLP_LBIN | SLP_LP;
6432 		phdr->param_len = 8;
6433 		data = (uint8_t *)(phdr + 1);
6434 		scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6435 		data[4] = 0x01; /* per-LUN */
6436 		data += phdr->param_len;
6437 	}
6438 
6439 	if (lun->backend->lun_attr != NULL &&
6440 	    (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksavail"))
6441 	     != UINT64_MAX) {
6442 		phdr = (struct scsi_log_param_header *)data;
6443 		scsi_ulto2b(0x00f1, phdr->param_code);
6444 		phdr->param_control = SLP_LBIN | SLP_LP;
6445 		phdr->param_len = 8;
6446 		data = (uint8_t *)(phdr + 1);
6447 		scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6448 		data[4] = 0x02; /* per-pool */
6449 		data += phdr->param_len;
6450 	}
6451 
6452 	if (lun->backend->lun_attr != NULL &&
6453 	    (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksused"))
6454 	     != UINT64_MAX) {
6455 		phdr = (struct scsi_log_param_header *)data;
6456 		scsi_ulto2b(0x00f2, phdr->param_code);
6457 		phdr->param_control = SLP_LBIN | SLP_LP;
6458 		phdr->param_len = 8;
6459 		data = (uint8_t *)(phdr + 1);
6460 		scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6461 		data[4] = 0x02; /* per-pool */
6462 		data += phdr->param_len;
6463 	}
6464 
6465 	page_index->page_len = data - page_index->page_data;
6466 	return (0);
6467 }
6468 
6469 int
6470 ctl_sap_log_sense_handler(struct ctl_scsiio *ctsio,
6471 			       struct ctl_page_index *page_index,
6472 			       int pc)
6473 {
6474 	struct ctl_lun *lun;
6475 	struct stat_page *data;
6476 	uint64_t rn, wn, rb, wb;
6477 	struct bintime rt, wt;
6478 	int i;
6479 
6480 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6481 	data = (struct stat_page *)page_index->page_data;
6482 
6483 	scsi_ulto2b(SLP_SAP, data->sap.hdr.param_code);
6484 	data->sap.hdr.param_control = SLP_LBIN;
6485 	data->sap.hdr.param_len = sizeof(struct scsi_log_stat_and_perf) -
6486 	    sizeof(struct scsi_log_param_header);
6487 	rn = wn = rb = wb = 0;
6488 	bintime_clear(&rt);
6489 	bintime_clear(&wt);
6490 	for (i = 0; i < CTL_MAX_PORTS; i++) {
6491 		rn += lun->stats.ports[i].operations[CTL_STATS_READ];
6492 		wn += lun->stats.ports[i].operations[CTL_STATS_WRITE];
6493 		rb += lun->stats.ports[i].bytes[CTL_STATS_READ];
6494 		wb += lun->stats.ports[i].bytes[CTL_STATS_WRITE];
6495 		bintime_add(&rt, &lun->stats.ports[i].time[CTL_STATS_READ]);
6496 		bintime_add(&wt, &lun->stats.ports[i].time[CTL_STATS_WRITE]);
6497 	}
6498 	scsi_u64to8b(rn, data->sap.read_num);
6499 	scsi_u64to8b(wn, data->sap.write_num);
6500 	if (lun->stats.blocksize > 0) {
6501 		scsi_u64to8b(wb / lun->stats.blocksize,
6502 		    data->sap.recvieved_lba);
6503 		scsi_u64to8b(rb / lun->stats.blocksize,
6504 		    data->sap.transmitted_lba);
6505 	}
6506 	scsi_u64to8b((uint64_t)rt.sec * 1000 + rt.frac / (UINT64_MAX / 1000),
6507 	    data->sap.read_int);
6508 	scsi_u64to8b((uint64_t)wt.sec * 1000 + wt.frac / (UINT64_MAX / 1000),
6509 	    data->sap.write_int);
6510 	scsi_u64to8b(0, data->sap.weighted_num);
6511 	scsi_u64to8b(0, data->sap.weighted_int);
6512 	scsi_ulto2b(SLP_IT, data->it.hdr.param_code);
6513 	data->it.hdr.param_control = SLP_LBIN;
6514 	data->it.hdr.param_len = sizeof(struct scsi_log_idle_time) -
6515 	    sizeof(struct scsi_log_param_header);
6516 #ifdef CTL_TIME_IO
6517 	scsi_u64to8b(lun->idle_time / SBT_1MS, data->it.idle_int);
6518 #endif
6519 	scsi_ulto2b(SLP_TI, data->ti.hdr.param_code);
6520 	data->it.hdr.param_control = SLP_LBIN;
6521 	data->ti.hdr.param_len = sizeof(struct scsi_log_time_interval) -
6522 	    sizeof(struct scsi_log_param_header);
6523 	scsi_ulto4b(3, data->ti.exponent);
6524 	scsi_ulto4b(1, data->ti.integer);
6525 
6526 	page_index->page_len = sizeof(*data);
6527 	return (0);
6528 }
6529 
6530 int
6531 ctl_log_sense(struct ctl_scsiio *ctsio)
6532 {
6533 	struct ctl_lun *lun;
6534 	int i, pc, page_code, subpage;
6535 	int alloc_len, total_len;
6536 	struct ctl_page_index *page_index;
6537 	struct scsi_log_sense *cdb;
6538 	struct scsi_log_header *header;
6539 
6540 	CTL_DEBUG_PRINT(("ctl_log_sense\n"));
6541 
6542 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6543 	cdb = (struct scsi_log_sense *)ctsio->cdb;
6544 	pc = (cdb->page & SLS_PAGE_CTRL_MASK) >> 6;
6545 	page_code = cdb->page & SLS_PAGE_CODE;
6546 	subpage = cdb->subpage;
6547 	alloc_len = scsi_2btoul(cdb->length);
6548 
6549 	page_index = NULL;
6550 	for (i = 0; i < CTL_NUM_LOG_PAGES; i++) {
6551 		page_index = &lun->log_pages.index[i];
6552 
6553 		/* Look for the right page code */
6554 		if ((page_index->page_code & SL_PAGE_CODE) != page_code)
6555 			continue;
6556 
6557 		/* Look for the right subpage or the subpage wildcard*/
6558 		if (page_index->subpage != subpage)
6559 			continue;
6560 
6561 		break;
6562 	}
6563 	if (i >= CTL_NUM_LOG_PAGES) {
6564 		ctl_set_invalid_field(ctsio,
6565 				      /*sks_valid*/ 1,
6566 				      /*command*/ 1,
6567 				      /*field*/ 2,
6568 				      /*bit_valid*/ 0,
6569 				      /*bit*/ 0);
6570 		ctl_done((union ctl_io *)ctsio);
6571 		return (CTL_RETVAL_COMPLETE);
6572 	}
6573 
6574 	total_len = sizeof(struct scsi_log_header) + page_index->page_len;
6575 
6576 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6577 	ctsio->kern_sg_entries = 0;
6578 	ctsio->kern_data_resid = 0;
6579 	ctsio->kern_rel_offset = 0;
6580 	if (total_len < alloc_len) {
6581 		ctsio->residual = alloc_len - total_len;
6582 		ctsio->kern_data_len = total_len;
6583 		ctsio->kern_total_len = total_len;
6584 	} else {
6585 		ctsio->residual = 0;
6586 		ctsio->kern_data_len = alloc_len;
6587 		ctsio->kern_total_len = alloc_len;
6588 	}
6589 
6590 	header = (struct scsi_log_header *)ctsio->kern_data_ptr;
6591 	header->page = page_index->page_code;
6592 	if (page_index->subpage) {
6593 		header->page |= SL_SPF;
6594 		header->subpage = page_index->subpage;
6595 	}
6596 	scsi_ulto2b(page_index->page_len, header->datalen);
6597 
6598 	/*
6599 	 * Call the handler, if it exists, to update the
6600 	 * page to the latest values.
6601 	 */
6602 	if (page_index->sense_handler != NULL)
6603 		page_index->sense_handler(ctsio, page_index, pc);
6604 
6605 	memcpy(header + 1, page_index->page_data, page_index->page_len);
6606 
6607 	ctl_set_success(ctsio);
6608 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6609 	ctsio->be_move_done = ctl_config_move_done;
6610 	ctl_datamove((union ctl_io *)ctsio);
6611 	return (CTL_RETVAL_COMPLETE);
6612 }
6613 
6614 int
6615 ctl_read_capacity(struct ctl_scsiio *ctsio)
6616 {
6617 	struct scsi_read_capacity *cdb;
6618 	struct scsi_read_capacity_data *data;
6619 	struct ctl_lun *lun;
6620 	uint32_t lba;
6621 
6622 	CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
6623 
6624 	cdb = (struct scsi_read_capacity *)ctsio->cdb;
6625 
6626 	lba = scsi_4btoul(cdb->addr);
6627 	if (((cdb->pmi & SRC_PMI) == 0)
6628 	 && (lba != 0)) {
6629 		ctl_set_invalid_field(/*ctsio*/ ctsio,
6630 				      /*sks_valid*/ 1,
6631 				      /*command*/ 1,
6632 				      /*field*/ 2,
6633 				      /*bit_valid*/ 0,
6634 				      /*bit*/ 0);
6635 		ctl_done((union ctl_io *)ctsio);
6636 		return (CTL_RETVAL_COMPLETE);
6637 	}
6638 
6639 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6640 
6641 	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6642 	data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
6643 	ctsio->residual = 0;
6644 	ctsio->kern_data_len = sizeof(*data);
6645 	ctsio->kern_total_len = sizeof(*data);
6646 	ctsio->kern_data_resid = 0;
6647 	ctsio->kern_rel_offset = 0;
6648 	ctsio->kern_sg_entries = 0;
6649 
6650 	/*
6651 	 * If the maximum LBA is greater than 0xfffffffe, the user must
6652 	 * issue a SERVICE ACTION IN (16) command, with the read capacity
6653 	 * serivce action set.
6654 	 */
6655 	if (lun->be_lun->maxlba > 0xfffffffe)
6656 		scsi_ulto4b(0xffffffff, data->addr);
6657 	else
6658 		scsi_ulto4b(lun->be_lun->maxlba, data->addr);
6659 
6660 	/*
6661 	 * XXX KDM this may not be 512 bytes...
6662 	 */
6663 	scsi_ulto4b(lun->be_lun->blocksize, data->length);
6664 
6665 	ctl_set_success(ctsio);
6666 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6667 	ctsio->be_move_done = ctl_config_move_done;
6668 	ctl_datamove((union ctl_io *)ctsio);
6669 	return (CTL_RETVAL_COMPLETE);
6670 }
6671 
6672 int
6673 ctl_read_capacity_16(struct ctl_scsiio *ctsio)
6674 {
6675 	struct scsi_read_capacity_16 *cdb;
6676 	struct scsi_read_capacity_data_long *data;
6677 	struct ctl_lun *lun;
6678 	uint64_t lba;
6679 	uint32_t alloc_len;
6680 
6681 	CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
6682 
6683 	cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
6684 
6685 	alloc_len = scsi_4btoul(cdb->alloc_len);
6686 	lba = scsi_8btou64(cdb->addr);
6687 
6688 	if ((cdb->reladr & SRC16_PMI)
6689 	 && (lba != 0)) {
6690 		ctl_set_invalid_field(/*ctsio*/ ctsio,
6691 				      /*sks_valid*/ 1,
6692 				      /*command*/ 1,
6693 				      /*field*/ 2,
6694 				      /*bit_valid*/ 0,
6695 				      /*bit*/ 0);
6696 		ctl_done((union ctl_io *)ctsio);
6697 		return (CTL_RETVAL_COMPLETE);
6698 	}
6699 
6700 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6701 
6702 	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6703 	data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
6704 
6705 	if (sizeof(*data) < alloc_len) {
6706 		ctsio->residual = alloc_len - sizeof(*data);
6707 		ctsio->kern_data_len = sizeof(*data);
6708 		ctsio->kern_total_len = sizeof(*data);
6709 	} else {
6710 		ctsio->residual = 0;
6711 		ctsio->kern_data_len = alloc_len;
6712 		ctsio->kern_total_len = alloc_len;
6713 	}
6714 	ctsio->kern_data_resid = 0;
6715 	ctsio->kern_rel_offset = 0;
6716 	ctsio->kern_sg_entries = 0;
6717 
6718 	scsi_u64to8b(lun->be_lun->maxlba, data->addr);
6719 	/* XXX KDM this may not be 512 bytes... */
6720 	scsi_ulto4b(lun->be_lun->blocksize, data->length);
6721 	data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
6722 	scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
6723 	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
6724 		data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ;
6725 
6726 	ctl_set_success(ctsio);
6727 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6728 	ctsio->be_move_done = ctl_config_move_done;
6729 	ctl_datamove((union ctl_io *)ctsio);
6730 	return (CTL_RETVAL_COMPLETE);
6731 }
6732 
6733 int
6734 ctl_get_lba_status(struct ctl_scsiio *ctsio)
6735 {
6736 	struct scsi_get_lba_status *cdb;
6737 	struct scsi_get_lba_status_data *data;
6738 	struct ctl_lun *lun;
6739 	struct ctl_lba_len_flags *lbalen;
6740 	uint64_t lba;
6741 	uint32_t alloc_len, total_len;
6742 	int retval;
6743 
6744 	CTL_DEBUG_PRINT(("ctl_get_lba_status\n"));
6745 
6746 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6747 	cdb = (struct scsi_get_lba_status *)ctsio->cdb;
6748 	lba = scsi_8btou64(cdb->addr);
6749 	alloc_len = scsi_4btoul(cdb->alloc_len);
6750 
6751 	if (lba > lun->be_lun->maxlba) {
6752 		ctl_set_lba_out_of_range(ctsio);
6753 		ctl_done((union ctl_io *)ctsio);
6754 		return (CTL_RETVAL_COMPLETE);
6755 	}
6756 
6757 	total_len = sizeof(*data) + sizeof(data->descr[0]);
6758 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6759 	data = (struct scsi_get_lba_status_data *)ctsio->kern_data_ptr;
6760 
6761 	if (total_len < alloc_len) {
6762 		ctsio->residual = alloc_len - total_len;
6763 		ctsio->kern_data_len = total_len;
6764 		ctsio->kern_total_len = total_len;
6765 	} else {
6766 		ctsio->residual = 0;
6767 		ctsio->kern_data_len = alloc_len;
6768 		ctsio->kern_total_len = alloc_len;
6769 	}
6770 	ctsio->kern_data_resid = 0;
6771 	ctsio->kern_rel_offset = 0;
6772 	ctsio->kern_sg_entries = 0;
6773 
6774 	/* Fill dummy data in case backend can't tell anything. */
6775 	scsi_ulto4b(4 + sizeof(data->descr[0]), data->length);
6776 	scsi_u64to8b(lba, data->descr[0].addr);
6777 	scsi_ulto4b(MIN(UINT32_MAX, lun->be_lun->maxlba + 1 - lba),
6778 	    data->descr[0].length);
6779 	data->descr[0].status = 0; /* Mapped or unknown. */
6780 
6781 	ctl_set_success(ctsio);
6782 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6783 	ctsio->be_move_done = ctl_config_move_done;
6784 
6785 	lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
6786 	lbalen->lba = lba;
6787 	lbalen->len = total_len;
6788 	lbalen->flags = 0;
6789 	retval = lun->backend->config_read((union ctl_io *)ctsio);
6790 	return (CTL_RETVAL_COMPLETE);
6791 }
6792 
6793 int
6794 ctl_read_defect(struct ctl_scsiio *ctsio)
6795 {
6796 	struct scsi_read_defect_data_10 *ccb10;
6797 	struct scsi_read_defect_data_12 *ccb12;
6798 	struct scsi_read_defect_data_hdr_10 *data10;
6799 	struct scsi_read_defect_data_hdr_12 *data12;
6800 	uint32_t alloc_len, data_len;
6801 	uint8_t format;
6802 
6803 	CTL_DEBUG_PRINT(("ctl_read_defect\n"));
6804 
6805 	if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
6806 		ccb10 = (struct scsi_read_defect_data_10 *)&ctsio->cdb;
6807 		format = ccb10->format;
6808 		alloc_len = scsi_2btoul(ccb10->alloc_length);
6809 		data_len = sizeof(*data10);
6810 	} else {
6811 		ccb12 = (struct scsi_read_defect_data_12 *)&ctsio->cdb;
6812 		format = ccb12->format;
6813 		alloc_len = scsi_4btoul(ccb12->alloc_length);
6814 		data_len = sizeof(*data12);
6815 	}
6816 	if (alloc_len == 0) {
6817 		ctl_set_success(ctsio);
6818 		ctl_done((union ctl_io *)ctsio);
6819 		return (CTL_RETVAL_COMPLETE);
6820 	}
6821 
6822 	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
6823 	if (data_len < alloc_len) {
6824 		ctsio->residual = alloc_len - data_len;
6825 		ctsio->kern_data_len = data_len;
6826 		ctsio->kern_total_len = data_len;
6827 	} else {
6828 		ctsio->residual = 0;
6829 		ctsio->kern_data_len = alloc_len;
6830 		ctsio->kern_total_len = alloc_len;
6831 	}
6832 	ctsio->kern_data_resid = 0;
6833 	ctsio->kern_rel_offset = 0;
6834 	ctsio->kern_sg_entries = 0;
6835 
6836 	if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
6837 		data10 = (struct scsi_read_defect_data_hdr_10 *)
6838 		    ctsio->kern_data_ptr;
6839 		data10->format = format;
6840 		scsi_ulto2b(0, data10->length);
6841 	} else {
6842 		data12 = (struct scsi_read_defect_data_hdr_12 *)
6843 		    ctsio->kern_data_ptr;
6844 		data12->format = format;
6845 		scsi_ulto2b(0, data12->generation);
6846 		scsi_ulto4b(0, data12->length);
6847 	}
6848 
6849 	ctl_set_success(ctsio);
6850 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6851 	ctsio->be_move_done = ctl_config_move_done;
6852 	ctl_datamove((union ctl_io *)ctsio);
6853 	return (CTL_RETVAL_COMPLETE);
6854 }
6855 
6856 int
6857 ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
6858 {
6859 	struct scsi_maintenance_in *cdb;
6860 	int retval;
6861 	int alloc_len, ext, total_len = 0, g, p, pc, pg, gs, os;
6862 	int num_target_port_groups, num_target_ports;
6863 	struct ctl_lun *lun;
6864 	struct ctl_softc *softc;
6865 	struct ctl_port *port;
6866 	struct scsi_target_group_data *rtg_ptr;
6867 	struct scsi_target_group_data_extended *rtg_ext_ptr;
6868 	struct scsi_target_port_group_descriptor *tpg_desc;
6869 
6870 	CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
6871 
6872 	cdb = (struct scsi_maintenance_in *)ctsio->cdb;
6873 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6874 	softc = lun->ctl_softc;
6875 
6876 	retval = CTL_RETVAL_COMPLETE;
6877 
6878 	switch (cdb->byte2 & STG_PDF_MASK) {
6879 	case STG_PDF_LENGTH:
6880 		ext = 0;
6881 		break;
6882 	case STG_PDF_EXTENDED:
6883 		ext = 1;
6884 		break;
6885 	default:
6886 		ctl_set_invalid_field(/*ctsio*/ ctsio,
6887 				      /*sks_valid*/ 1,
6888 				      /*command*/ 1,
6889 				      /*field*/ 2,
6890 				      /*bit_valid*/ 1,
6891 				      /*bit*/ 5);
6892 		ctl_done((union ctl_io *)ctsio);
6893 		return(retval);
6894 	}
6895 
6896 	if (softc->is_single)
6897 		num_target_port_groups = 1;
6898 	else
6899 		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
6900 	num_target_ports = 0;
6901 	mtx_lock(&softc->ctl_lock);
6902 	STAILQ_FOREACH(port, &softc->port_list, links) {
6903 		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
6904 			continue;
6905 		if (ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
6906 			continue;
6907 		num_target_ports++;
6908 	}
6909 	mtx_unlock(&softc->ctl_lock);
6910 
6911 	if (ext)
6912 		total_len = sizeof(struct scsi_target_group_data_extended);
6913 	else
6914 		total_len = sizeof(struct scsi_target_group_data);
6915 	total_len += sizeof(struct scsi_target_port_group_descriptor) *
6916 		num_target_port_groups +
6917 	    sizeof(struct scsi_target_port_descriptor) *
6918 		num_target_ports * num_target_port_groups;
6919 
6920 	alloc_len = scsi_4btoul(cdb->length);
6921 
6922 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6923 
6924 	ctsio->kern_sg_entries = 0;
6925 
6926 	if (total_len < alloc_len) {
6927 		ctsio->residual = alloc_len - total_len;
6928 		ctsio->kern_data_len = total_len;
6929 		ctsio->kern_total_len = total_len;
6930 	} else {
6931 		ctsio->residual = 0;
6932 		ctsio->kern_data_len = alloc_len;
6933 		ctsio->kern_total_len = alloc_len;
6934 	}
6935 	ctsio->kern_data_resid = 0;
6936 	ctsio->kern_rel_offset = 0;
6937 
6938 	if (ext) {
6939 		rtg_ext_ptr = (struct scsi_target_group_data_extended *)
6940 		    ctsio->kern_data_ptr;
6941 		scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
6942 		rtg_ext_ptr->format_type = 0x10;
6943 		rtg_ext_ptr->implicit_transition_time = 0;
6944 		tpg_desc = &rtg_ext_ptr->groups[0];
6945 	} else {
6946 		rtg_ptr = (struct scsi_target_group_data *)
6947 		    ctsio->kern_data_ptr;
6948 		scsi_ulto4b(total_len - 4, rtg_ptr->length);
6949 		tpg_desc = &rtg_ptr->groups[0];
6950 	}
6951 
6952 	mtx_lock(&softc->ctl_lock);
6953 	pg = softc->port_offset / CTL_MAX_PORTS;
6954 	if (softc->flags & CTL_FLAG_ACTIVE_SHELF) {
6955 		if (softc->ha_mode == CTL_HA_MODE_ACT_STBY) {
6956 			gs = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
6957 			os = TPG_ASYMMETRIC_ACCESS_STANDBY;
6958 		} else if (lun->flags & CTL_LUN_PRIMARY_SC) {
6959 			gs = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
6960 			os = TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
6961 		} else {
6962 			gs = TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
6963 			os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
6964 		}
6965 	} else {
6966 		gs = TPG_ASYMMETRIC_ACCESS_STANDBY;
6967 		os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
6968 	}
6969 	for (g = 0; g < num_target_port_groups; g++) {
6970 		tpg_desc->pref_state = (g == pg) ? gs : os;
6971 		tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP;
6972 		scsi_ulto2b(g + 1, tpg_desc->target_port_group);
6973 		tpg_desc->status = TPG_IMPLICIT;
6974 		pc = 0;
6975 		STAILQ_FOREACH(port, &softc->port_list, links) {
6976 			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
6977 				continue;
6978 			if (ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
6979 				continue;
6980 			p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
6981 			scsi_ulto2b(p, tpg_desc->descriptors[pc].
6982 			    relative_target_port_identifier);
6983 			pc++;
6984 		}
6985 		tpg_desc->target_port_count = pc;
6986 		tpg_desc = (struct scsi_target_port_group_descriptor *)
6987 		    &tpg_desc->descriptors[pc];
6988 	}
6989 	mtx_unlock(&softc->ctl_lock);
6990 
6991 	ctl_set_success(ctsio);
6992 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6993 	ctsio->be_move_done = ctl_config_move_done;
6994 	ctl_datamove((union ctl_io *)ctsio);
6995 	return(retval);
6996 }
6997 
6998 int
6999 ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7000 {
7001 	struct ctl_lun *lun;
7002 	struct scsi_report_supported_opcodes *cdb;
7003 	const struct ctl_cmd_entry *entry, *sentry;
7004 	struct scsi_report_supported_opcodes_all *all;
7005 	struct scsi_report_supported_opcodes_descr *descr;
7006 	struct scsi_report_supported_opcodes_one *one;
7007 	int retval;
7008 	int alloc_len, total_len;
7009 	int opcode, service_action, i, j, num;
7010 
7011 	CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7012 
7013 	cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7014 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7015 
7016 	retval = CTL_RETVAL_COMPLETE;
7017 
7018 	opcode = cdb->requested_opcode;
7019 	service_action = scsi_2btoul(cdb->requested_service_action);
7020 	switch (cdb->options & RSO_OPTIONS_MASK) {
7021 	case RSO_OPTIONS_ALL:
7022 		num = 0;
7023 		for (i = 0; i < 256; i++) {
7024 			entry = &ctl_cmd_table[i];
7025 			if (entry->flags & CTL_CMD_FLAG_SA5) {
7026 				for (j = 0; j < 32; j++) {
7027 					sentry = &((const struct ctl_cmd_entry *)
7028 					    entry->execute)[j];
7029 					if (ctl_cmd_applicable(
7030 					    lun->be_lun->lun_type, sentry))
7031 						num++;
7032 				}
7033 			} else {
7034 				if (ctl_cmd_applicable(lun->be_lun->lun_type,
7035 				    entry))
7036 					num++;
7037 			}
7038 		}
7039 		total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7040 		    num * sizeof(struct scsi_report_supported_opcodes_descr);
7041 		break;
7042 	case RSO_OPTIONS_OC:
7043 		if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7044 			ctl_set_invalid_field(/*ctsio*/ ctsio,
7045 					      /*sks_valid*/ 1,
7046 					      /*command*/ 1,
7047 					      /*field*/ 2,
7048 					      /*bit_valid*/ 1,
7049 					      /*bit*/ 2);
7050 			ctl_done((union ctl_io *)ctsio);
7051 			return (CTL_RETVAL_COMPLETE);
7052 		}
7053 		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7054 		break;
7055 	case RSO_OPTIONS_OC_SA:
7056 		if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7057 		    service_action >= 32) {
7058 			ctl_set_invalid_field(/*ctsio*/ ctsio,
7059 					      /*sks_valid*/ 1,
7060 					      /*command*/ 1,
7061 					      /*field*/ 2,
7062 					      /*bit_valid*/ 1,
7063 					      /*bit*/ 2);
7064 			ctl_done((union ctl_io *)ctsio);
7065 			return (CTL_RETVAL_COMPLETE);
7066 		}
7067 		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7068 		break;
7069 	default:
7070 		ctl_set_invalid_field(/*ctsio*/ ctsio,
7071 				      /*sks_valid*/ 1,
7072 				      /*command*/ 1,
7073 				      /*field*/ 2,
7074 				      /*bit_valid*/ 1,
7075 				      /*bit*/ 2);
7076 		ctl_done((union ctl_io *)ctsio);
7077 		return (CTL_RETVAL_COMPLETE);
7078 	}
7079 
7080 	alloc_len = scsi_4btoul(cdb->length);
7081 
7082 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7083 
7084 	ctsio->kern_sg_entries = 0;
7085 
7086 	if (total_len < alloc_len) {
7087 		ctsio->residual = alloc_len - total_len;
7088 		ctsio->kern_data_len = total_len;
7089 		ctsio->kern_total_len = total_len;
7090 	} else {
7091 		ctsio->residual = 0;
7092 		ctsio->kern_data_len = alloc_len;
7093 		ctsio->kern_total_len = alloc_len;
7094 	}
7095 	ctsio->kern_data_resid = 0;
7096 	ctsio->kern_rel_offset = 0;
7097 
7098 	switch (cdb->options & RSO_OPTIONS_MASK) {
7099 	case RSO_OPTIONS_ALL:
7100 		all = (struct scsi_report_supported_opcodes_all *)
7101 		    ctsio->kern_data_ptr;
7102 		num = 0;
7103 		for (i = 0; i < 256; i++) {
7104 			entry = &ctl_cmd_table[i];
7105 			if (entry->flags & CTL_CMD_FLAG_SA5) {
7106 				for (j = 0; j < 32; j++) {
7107 					sentry = &((const struct ctl_cmd_entry *)
7108 					    entry->execute)[j];
7109 					if (!ctl_cmd_applicable(
7110 					    lun->be_lun->lun_type, sentry))
7111 						continue;
7112 					descr = &all->descr[num++];
7113 					descr->opcode = i;
7114 					scsi_ulto2b(j, descr->service_action);
7115 					descr->flags = RSO_SERVACTV;
7116 					scsi_ulto2b(sentry->length,
7117 					    descr->cdb_length);
7118 				}
7119 			} else {
7120 				if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7121 				    entry))
7122 					continue;
7123 				descr = &all->descr[num++];
7124 				descr->opcode = i;
7125 				scsi_ulto2b(0, descr->service_action);
7126 				descr->flags = 0;
7127 				scsi_ulto2b(entry->length, descr->cdb_length);
7128 			}
7129 		}
7130 		scsi_ulto4b(
7131 		    num * sizeof(struct scsi_report_supported_opcodes_descr),
7132 		    all->length);
7133 		break;
7134 	case RSO_OPTIONS_OC:
7135 		one = (struct scsi_report_supported_opcodes_one *)
7136 		    ctsio->kern_data_ptr;
7137 		entry = &ctl_cmd_table[opcode];
7138 		goto fill_one;
7139 	case RSO_OPTIONS_OC_SA:
7140 		one = (struct scsi_report_supported_opcodes_one *)
7141 		    ctsio->kern_data_ptr;
7142 		entry = &ctl_cmd_table[opcode];
7143 		entry = &((const struct ctl_cmd_entry *)
7144 		    entry->execute)[service_action];
7145 fill_one:
7146 		if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7147 			one->support = 3;
7148 			scsi_ulto2b(entry->length, one->cdb_length);
7149 			one->cdb_usage[0] = opcode;
7150 			memcpy(&one->cdb_usage[1], entry->usage,
7151 			    entry->length - 1);
7152 		} else
7153 			one->support = 1;
7154 		break;
7155 	}
7156 
7157 	ctl_set_success(ctsio);
7158 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7159 	ctsio->be_move_done = ctl_config_move_done;
7160 	ctl_datamove((union ctl_io *)ctsio);
7161 	return(retval);
7162 }
7163 
7164 int
7165 ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7166 {
7167 	struct scsi_report_supported_tmf *cdb;
7168 	struct scsi_report_supported_tmf_data *data;
7169 	int retval;
7170 	int alloc_len, total_len;
7171 
7172 	CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7173 
7174 	cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7175 
7176 	retval = CTL_RETVAL_COMPLETE;
7177 
7178 	total_len = sizeof(struct scsi_report_supported_tmf_data);
7179 	alloc_len = scsi_4btoul(cdb->length);
7180 
7181 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7182 
7183 	ctsio->kern_sg_entries = 0;
7184 
7185 	if (total_len < alloc_len) {
7186 		ctsio->residual = alloc_len - total_len;
7187 		ctsio->kern_data_len = total_len;
7188 		ctsio->kern_total_len = total_len;
7189 	} else {
7190 		ctsio->residual = 0;
7191 		ctsio->kern_data_len = alloc_len;
7192 		ctsio->kern_total_len = alloc_len;
7193 	}
7194 	ctsio->kern_data_resid = 0;
7195 	ctsio->kern_rel_offset = 0;
7196 
7197 	data = (struct scsi_report_supported_tmf_data *)ctsio->kern_data_ptr;
7198 	data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_TRS;
7199 	data->byte2 |= RST_ITNRS;
7200 
7201 	ctl_set_success(ctsio);
7202 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7203 	ctsio->be_move_done = ctl_config_move_done;
7204 	ctl_datamove((union ctl_io *)ctsio);
7205 	return (retval);
7206 }
7207 
7208 int
7209 ctl_report_timestamp(struct ctl_scsiio *ctsio)
7210 {
7211 	struct scsi_report_timestamp *cdb;
7212 	struct scsi_report_timestamp_data *data;
7213 	struct timeval tv;
7214 	int64_t timestamp;
7215 	int retval;
7216 	int alloc_len, total_len;
7217 
7218 	CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7219 
7220 	cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7221 
7222 	retval = CTL_RETVAL_COMPLETE;
7223 
7224 	total_len = sizeof(struct scsi_report_timestamp_data);
7225 	alloc_len = scsi_4btoul(cdb->length);
7226 
7227 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7228 
7229 	ctsio->kern_sg_entries = 0;
7230 
7231 	if (total_len < alloc_len) {
7232 		ctsio->residual = alloc_len - total_len;
7233 		ctsio->kern_data_len = total_len;
7234 		ctsio->kern_total_len = total_len;
7235 	} else {
7236 		ctsio->residual = 0;
7237 		ctsio->kern_data_len = alloc_len;
7238 		ctsio->kern_total_len = alloc_len;
7239 	}
7240 	ctsio->kern_data_resid = 0;
7241 	ctsio->kern_rel_offset = 0;
7242 
7243 	data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7244 	scsi_ulto2b(sizeof(*data) - 2, data->length);
7245 	data->origin = RTS_ORIG_OUTSIDE;
7246 	getmicrotime(&tv);
7247 	timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7248 	scsi_ulto4b(timestamp >> 16, data->timestamp);
7249 	scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7250 
7251 	ctl_set_success(ctsio);
7252 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7253 	ctsio->be_move_done = ctl_config_move_done;
7254 	ctl_datamove((union ctl_io *)ctsio);
7255 	return (retval);
7256 }
7257 
7258 int
7259 ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7260 {
7261 	struct scsi_per_res_in *cdb;
7262 	int alloc_len, total_len = 0;
7263 	/* struct scsi_per_res_in_rsrv in_data; */
7264 	struct ctl_lun *lun;
7265 	struct ctl_softc *softc;
7266 	uint64_t key;
7267 
7268 	CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7269 
7270 	cdb = (struct scsi_per_res_in *)ctsio->cdb;
7271 
7272 	alloc_len = scsi_2btoul(cdb->length);
7273 
7274 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7275 	softc = lun->ctl_softc;
7276 
7277 retry:
7278 	mtx_lock(&lun->lun_lock);
7279 	switch (cdb->action) {
7280 	case SPRI_RK: /* read keys */
7281 		total_len = sizeof(struct scsi_per_res_in_keys) +
7282 			lun->pr_key_count *
7283 			sizeof(struct scsi_per_res_key);
7284 		break;
7285 	case SPRI_RR: /* read reservation */
7286 		if (lun->flags & CTL_LUN_PR_RESERVED)
7287 			total_len = sizeof(struct scsi_per_res_in_rsrv);
7288 		else
7289 			total_len = sizeof(struct scsi_per_res_in_header);
7290 		break;
7291 	case SPRI_RC: /* report capabilities */
7292 		total_len = sizeof(struct scsi_per_res_cap);
7293 		break;
7294 	case SPRI_RS: /* read full status */
7295 		total_len = sizeof(struct scsi_per_res_in_header) +
7296 		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7297 		    lun->pr_key_count;
7298 		break;
7299 	default:
7300 		panic("Invalid PR type %x", cdb->action);
7301 	}
7302 	mtx_unlock(&lun->lun_lock);
7303 
7304 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7305 
7306 	if (total_len < alloc_len) {
7307 		ctsio->residual = alloc_len - total_len;
7308 		ctsio->kern_data_len = total_len;
7309 		ctsio->kern_total_len = total_len;
7310 	} else {
7311 		ctsio->residual = 0;
7312 		ctsio->kern_data_len = alloc_len;
7313 		ctsio->kern_total_len = alloc_len;
7314 	}
7315 
7316 	ctsio->kern_data_resid = 0;
7317 	ctsio->kern_rel_offset = 0;
7318 	ctsio->kern_sg_entries = 0;
7319 
7320 	mtx_lock(&lun->lun_lock);
7321 	switch (cdb->action) {
7322 	case SPRI_RK: { // read keys
7323         struct scsi_per_res_in_keys *res_keys;
7324 		int i, key_count;
7325 
7326 		res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7327 
7328 		/*
7329 		 * We had to drop the lock to allocate our buffer, which
7330 		 * leaves time for someone to come in with another
7331 		 * persistent reservation.  (That is unlikely, though,
7332 		 * since this should be the only persistent reservation
7333 		 * command active right now.)
7334 		 */
7335 		if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7336 		    (lun->pr_key_count *
7337 		     sizeof(struct scsi_per_res_key)))){
7338 			mtx_unlock(&lun->lun_lock);
7339 			free(ctsio->kern_data_ptr, M_CTL);
7340 			printf("%s: reservation length changed, retrying\n",
7341 			       __func__);
7342 			goto retry;
7343 		}
7344 
7345 		scsi_ulto4b(lun->PRGeneration, res_keys->header.generation);
7346 
7347 		scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7348 			     lun->pr_key_count, res_keys->header.length);
7349 
7350 		for (i = 0, key_count = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7351 			if ((key = ctl_get_prkey(lun, i)) == 0)
7352 				continue;
7353 
7354 			/*
7355 			 * We used lun->pr_key_count to calculate the
7356 			 * size to allocate.  If it turns out the number of
7357 			 * initiators with the registered flag set is
7358 			 * larger than that (i.e. they haven't been kept in
7359 			 * sync), we've got a problem.
7360 			 */
7361 			if (key_count >= lun->pr_key_count) {
7362 #ifdef NEEDTOPORT
7363 				csevent_log(CSC_CTL | CSC_SHELF_SW |
7364 					    CTL_PR_ERROR,
7365 					    csevent_LogType_Fault,
7366 					    csevent_AlertLevel_Yellow,
7367 					    csevent_FRU_ShelfController,
7368 					    csevent_FRU_Firmware,
7369 				        csevent_FRU_Unknown,
7370 					    "registered keys %d >= key "
7371 					    "count %d", key_count,
7372 					    lun->pr_key_count);
7373 #endif
7374 				key_count++;
7375 				continue;
7376 			}
7377 			scsi_u64to8b(key, res_keys->keys[key_count].key);
7378 			key_count++;
7379 		}
7380 		break;
7381 	}
7382 	case SPRI_RR: { // read reservation
7383 		struct scsi_per_res_in_rsrv *res;
7384 		int tmp_len, header_only;
7385 
7386 		res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7387 
7388 		scsi_ulto4b(lun->PRGeneration, res->header.generation);
7389 
7390 		if (lun->flags & CTL_LUN_PR_RESERVED)
7391 		{
7392 			tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7393 			scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7394 				    res->header.length);
7395 			header_only = 0;
7396 		} else {
7397 			tmp_len = sizeof(struct scsi_per_res_in_header);
7398 			scsi_ulto4b(0, res->header.length);
7399 			header_only = 1;
7400 		}
7401 
7402 		/*
7403 		 * We had to drop the lock to allocate our buffer, which
7404 		 * leaves time for someone to come in with another
7405 		 * persistent reservation.  (That is unlikely, though,
7406 		 * since this should be the only persistent reservation
7407 		 * command active right now.)
7408 		 */
7409 		if (tmp_len != total_len) {
7410 			mtx_unlock(&lun->lun_lock);
7411 			free(ctsio->kern_data_ptr, M_CTL);
7412 			printf("%s: reservation status changed, retrying\n",
7413 			       __func__);
7414 			goto retry;
7415 		}
7416 
7417 		/*
7418 		 * No reservation held, so we're done.
7419 		 */
7420 		if (header_only != 0)
7421 			break;
7422 
7423 		/*
7424 		 * If the registration is an All Registrants type, the key
7425 		 * is 0, since it doesn't really matter.
7426 		 */
7427 		if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7428 			scsi_u64to8b(ctl_get_prkey(lun, lun->pr_res_idx),
7429 			    res->data.reservation);
7430 		}
7431 		res->data.scopetype = lun->res_type;
7432 		break;
7433 	}
7434 	case SPRI_RC:     //report capabilities
7435 	{
7436 		struct scsi_per_res_cap *res_cap;
7437 		uint16_t type_mask;
7438 
7439 		res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7440 		scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7441 		res_cap->flags2 |= SPRI_TMV | SPRI_ALLOW_5;
7442 		type_mask = SPRI_TM_WR_EX_AR |
7443 			    SPRI_TM_EX_AC_RO |
7444 			    SPRI_TM_WR_EX_RO |
7445 			    SPRI_TM_EX_AC |
7446 			    SPRI_TM_WR_EX |
7447 			    SPRI_TM_EX_AC_AR;
7448 		scsi_ulto2b(type_mask, res_cap->type_mask);
7449 		break;
7450 	}
7451 	case SPRI_RS: { // read full status
7452 		struct scsi_per_res_in_full *res_status;
7453 		struct scsi_per_res_in_full_desc *res_desc;
7454 		struct ctl_port *port;
7455 		int i, len;
7456 
7457 		res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7458 
7459 		/*
7460 		 * We had to drop the lock to allocate our buffer, which
7461 		 * leaves time for someone to come in with another
7462 		 * persistent reservation.  (That is unlikely, though,
7463 		 * since this should be the only persistent reservation
7464 		 * command active right now.)
7465 		 */
7466 		if (total_len < (sizeof(struct scsi_per_res_in_header) +
7467 		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7468 		     lun->pr_key_count)){
7469 			mtx_unlock(&lun->lun_lock);
7470 			free(ctsio->kern_data_ptr, M_CTL);
7471 			printf("%s: reservation length changed, retrying\n",
7472 			       __func__);
7473 			goto retry;
7474 		}
7475 
7476 		scsi_ulto4b(lun->PRGeneration, res_status->header.generation);
7477 
7478 		res_desc = &res_status->desc[0];
7479 		for (i = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7480 			if ((key = ctl_get_prkey(lun, i)) == 0)
7481 				continue;
7482 
7483 			scsi_u64to8b(key, res_desc->res_key.key);
7484 			if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7485 			    (lun->pr_res_idx == i ||
7486 			     lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7487 				res_desc->flags = SPRI_FULL_R_HOLDER;
7488 				res_desc->scopetype = lun->res_type;
7489 			}
7490 			scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7491 			    res_desc->rel_trgt_port_id);
7492 			len = 0;
7493 			port = softc->ctl_ports[
7494 			    ctl_port_idx(i / CTL_MAX_INIT_PER_PORT)];
7495 			if (port != NULL)
7496 				len = ctl_create_iid(port,
7497 				    i % CTL_MAX_INIT_PER_PORT,
7498 				    res_desc->transport_id);
7499 			scsi_ulto4b(len, res_desc->additional_length);
7500 			res_desc = (struct scsi_per_res_in_full_desc *)
7501 			    &res_desc->transport_id[len];
7502 		}
7503 		scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7504 		    res_status->header.length);
7505 		break;
7506 	}
7507 	default:
7508 		/*
7509 		 * This is a bug, because we just checked for this above,
7510 		 * and should have returned an error.
7511 		 */
7512 		panic("Invalid PR type %x", cdb->action);
7513 		break; /* NOTREACHED */
7514 	}
7515 	mtx_unlock(&lun->lun_lock);
7516 
7517 	ctl_set_success(ctsio);
7518 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7519 	ctsio->be_move_done = ctl_config_move_done;
7520 	ctl_datamove((union ctl_io *)ctsio);
7521 	return (CTL_RETVAL_COMPLETE);
7522 }
7523 
7524 static void
7525 ctl_est_res_ua(struct ctl_lun *lun, uint32_t residx, ctl_ua_type ua)
7526 {
7527 	int off = lun->ctl_softc->persis_offset;
7528 
7529 	if (residx >= off && residx < off + CTL_MAX_INITIATORS)
7530 		ctl_est_ua(lun, residx - off, ua);
7531 }
7532 
7533 /*
7534  * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7535  * it should return.
7536  */
7537 static int
7538 ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7539 		uint64_t sa_res_key, uint8_t type, uint32_t residx,
7540 		struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7541 		struct scsi_per_res_out_parms* param)
7542 {
7543 	union ctl_ha_msg persis_io;
7544 	int retval, i;
7545 	int isc_retval;
7546 
7547 	retval = 0;
7548 
7549 	mtx_lock(&lun->lun_lock);
7550 	if (sa_res_key == 0) {
7551 		if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7552 			/* validate scope and type */
7553 			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7554 			     SPR_LU_SCOPE) {
7555 				mtx_unlock(&lun->lun_lock);
7556 				ctl_set_invalid_field(/*ctsio*/ ctsio,
7557 						      /*sks_valid*/ 1,
7558 						      /*command*/ 1,
7559 						      /*field*/ 2,
7560 						      /*bit_valid*/ 1,
7561 						      /*bit*/ 4);
7562 				ctl_done((union ctl_io *)ctsio);
7563 				return (1);
7564 			}
7565 
7566 		        if (type>8 || type==2 || type==4 || type==0) {
7567 				mtx_unlock(&lun->lun_lock);
7568 				ctl_set_invalid_field(/*ctsio*/ ctsio,
7569        	           				      /*sks_valid*/ 1,
7570 						      /*command*/ 1,
7571 						      /*field*/ 2,
7572 						      /*bit_valid*/ 1,
7573 						      /*bit*/ 0);
7574 				ctl_done((union ctl_io *)ctsio);
7575 				return (1);
7576 		        }
7577 
7578 			/*
7579 			 * Unregister everybody else and build UA for
7580 			 * them
7581 			 */
7582 			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7583 				if (i == residx || ctl_get_prkey(lun, i) == 0)
7584 					continue;
7585 
7586 				ctl_clr_prkey(lun, i);
7587 				ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
7588 			}
7589 			lun->pr_key_count = 1;
7590 			lun->res_type = type;
7591 			if (lun->res_type != SPR_TYPE_WR_EX_AR
7592 			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7593 				lun->pr_res_idx = residx;
7594 
7595 			/* send msg to other side */
7596 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7597 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7598 			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7599 			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7600 			persis_io.pr.pr_info.res_type = type;
7601 			memcpy(persis_io.pr.pr_info.sa_res_key,
7602 			       param->serv_act_res_key,
7603 			       sizeof(param->serv_act_res_key));
7604 			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7605 			     &persis_io, sizeof(persis_io), 0)) >
7606 			     CTL_HA_STATUS_SUCCESS) {
7607 				printf("CTL:Persis Out error returned "
7608 				       "from ctl_ha_msg_send %d\n",
7609 				       isc_retval);
7610 			}
7611 		} else {
7612 			/* not all registrants */
7613 			mtx_unlock(&lun->lun_lock);
7614 			free(ctsio->kern_data_ptr, M_CTL);
7615 			ctl_set_invalid_field(ctsio,
7616 					      /*sks_valid*/ 1,
7617 					      /*command*/ 0,
7618 					      /*field*/ 8,
7619 					      /*bit_valid*/ 0,
7620 					      /*bit*/ 0);
7621 			ctl_done((union ctl_io *)ctsio);
7622 			return (1);
7623 		}
7624 	} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7625 		|| !(lun->flags & CTL_LUN_PR_RESERVED)) {
7626 		int found = 0;
7627 
7628 		if (res_key == sa_res_key) {
7629 			/* special case */
7630 			/*
7631 			 * The spec implies this is not good but doesn't
7632 			 * say what to do. There are two choices either
7633 			 * generate a res conflict or check condition
7634 			 * with illegal field in parameter data. Since
7635 			 * that is what is done when the sa_res_key is
7636 			 * zero I'll take that approach since this has
7637 			 * to do with the sa_res_key.
7638 			 */
7639 			mtx_unlock(&lun->lun_lock);
7640 			free(ctsio->kern_data_ptr, M_CTL);
7641 			ctl_set_invalid_field(ctsio,
7642 					      /*sks_valid*/ 1,
7643 					      /*command*/ 0,
7644 					      /*field*/ 8,
7645 					      /*bit_valid*/ 0,
7646 					      /*bit*/ 0);
7647 			ctl_done((union ctl_io *)ctsio);
7648 			return (1);
7649 		}
7650 
7651 		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7652 			if (ctl_get_prkey(lun, i) != sa_res_key)
7653 				continue;
7654 
7655 			found = 1;
7656 			ctl_clr_prkey(lun, i);
7657 			lun->pr_key_count--;
7658 			ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
7659 		}
7660 		if (!found) {
7661 			mtx_unlock(&lun->lun_lock);
7662 			free(ctsio->kern_data_ptr, M_CTL);
7663 			ctl_set_reservation_conflict(ctsio);
7664 			ctl_done((union ctl_io *)ctsio);
7665 			return (CTL_RETVAL_COMPLETE);
7666 		}
7667 		/* send msg to other side */
7668 		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7669 		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7670 		persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7671 		persis_io.pr.pr_info.residx = lun->pr_res_idx;
7672 		persis_io.pr.pr_info.res_type = type;
7673 		memcpy(persis_io.pr.pr_info.sa_res_key,
7674 		       param->serv_act_res_key,
7675 		       sizeof(param->serv_act_res_key));
7676 		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7677 		     &persis_io, sizeof(persis_io), 0)) >
7678 		     CTL_HA_STATUS_SUCCESS) {
7679 			printf("CTL:Persis Out error returned from "
7680 			       "ctl_ha_msg_send %d\n", isc_retval);
7681 		}
7682 	} else {
7683 		/* Reserved but not all registrants */
7684 		/* sa_res_key is res holder */
7685 		if (sa_res_key == ctl_get_prkey(lun, lun->pr_res_idx)) {
7686 			/* validate scope and type */
7687 			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7688 			     SPR_LU_SCOPE) {
7689 				mtx_unlock(&lun->lun_lock);
7690 				ctl_set_invalid_field(/*ctsio*/ ctsio,
7691 						      /*sks_valid*/ 1,
7692 						      /*command*/ 1,
7693 						      /*field*/ 2,
7694 						      /*bit_valid*/ 1,
7695 						      /*bit*/ 4);
7696 				ctl_done((union ctl_io *)ctsio);
7697 				return (1);
7698 			}
7699 
7700 			if (type>8 || type==2 || type==4 || type==0) {
7701 				mtx_unlock(&lun->lun_lock);
7702 				ctl_set_invalid_field(/*ctsio*/ ctsio,
7703 						      /*sks_valid*/ 1,
7704 						      /*command*/ 1,
7705 						      /*field*/ 2,
7706 						      /*bit_valid*/ 1,
7707 						      /*bit*/ 0);
7708 				ctl_done((union ctl_io *)ctsio);
7709 				return (1);
7710 			}
7711 
7712 			/*
7713 			 * Do the following:
7714 			 * if sa_res_key != res_key remove all
7715 			 * registrants w/sa_res_key and generate UA
7716 			 * for these registrants(Registrations
7717 			 * Preempted) if it wasn't an exclusive
7718 			 * reservation generate UA(Reservations
7719 			 * Preempted) for all other registered nexuses
7720 			 * if the type has changed. Establish the new
7721 			 * reservation and holder. If res_key and
7722 			 * sa_res_key are the same do the above
7723 			 * except don't unregister the res holder.
7724 			 */
7725 
7726 			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7727 				if (i == residx || ctl_get_prkey(lun, i) == 0)
7728 					continue;
7729 
7730 				if (sa_res_key == ctl_get_prkey(lun, i)) {
7731 					ctl_clr_prkey(lun, i);
7732 					lun->pr_key_count--;
7733 					ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
7734 				} else if (type != lun->res_type
7735 					&& (lun->res_type == SPR_TYPE_WR_EX_RO
7736 					 || lun->res_type ==SPR_TYPE_EX_AC_RO)){
7737 					ctl_est_res_ua(lun, i, CTL_UA_RES_RELEASE);
7738 				}
7739 			}
7740 			lun->res_type = type;
7741 			if (lun->res_type != SPR_TYPE_WR_EX_AR
7742 			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7743 				lun->pr_res_idx = residx;
7744 			else
7745 				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
7746 
7747 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7748 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7749 			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7750 			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7751 			persis_io.pr.pr_info.res_type = type;
7752 			memcpy(persis_io.pr.pr_info.sa_res_key,
7753 			       param->serv_act_res_key,
7754 			       sizeof(param->serv_act_res_key));
7755 			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7756 			     &persis_io, sizeof(persis_io), 0)) >
7757 			     CTL_HA_STATUS_SUCCESS) {
7758 				printf("CTL:Persis Out error returned "
7759 				       "from ctl_ha_msg_send %d\n",
7760 				       isc_retval);
7761 			}
7762 		} else {
7763 			/*
7764 			 * sa_res_key is not the res holder just
7765 			 * remove registrants
7766 			 */
7767 			int found=0;
7768 
7769 			for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7770 				if (sa_res_key != ctl_get_prkey(lun, i))
7771 					continue;
7772 
7773 				found = 1;
7774 				ctl_clr_prkey(lun, i);
7775 				lun->pr_key_count--;
7776 				ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
7777 			}
7778 
7779 			if (!found) {
7780 				mtx_unlock(&lun->lun_lock);
7781 				free(ctsio->kern_data_ptr, M_CTL);
7782 				ctl_set_reservation_conflict(ctsio);
7783 				ctl_done((union ctl_io *)ctsio);
7784 		        	return (1);
7785 			}
7786 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7787 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7788 			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7789 			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7790 			persis_io.pr.pr_info.res_type = type;
7791 			memcpy(persis_io.pr.pr_info.sa_res_key,
7792 			       param->serv_act_res_key,
7793 			       sizeof(param->serv_act_res_key));
7794 			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7795 			     &persis_io, sizeof(persis_io), 0)) >
7796 			     CTL_HA_STATUS_SUCCESS) {
7797 				printf("CTL:Persis Out error returned "
7798 				       "from ctl_ha_msg_send %d\n",
7799 				isc_retval);
7800 			}
7801 		}
7802 	}
7803 
7804 	lun->PRGeneration++;
7805 	mtx_unlock(&lun->lun_lock);
7806 
7807 	return (retval);
7808 }
7809 
7810 static void
7811 ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
7812 {
7813 	uint64_t sa_res_key;
7814 	int i;
7815 
7816 	sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
7817 
7818 	if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7819 	 || lun->pr_res_idx == CTL_PR_NO_RESERVATION
7820 	 || sa_res_key != ctl_get_prkey(lun, lun->pr_res_idx)) {
7821 		if (sa_res_key == 0) {
7822 			/*
7823 			 * Unregister everybody else and build UA for
7824 			 * them
7825 			 */
7826 			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7827 				if (i == msg->pr.pr_info.residx ||
7828 				    ctl_get_prkey(lun, i) == 0)
7829 					continue;
7830 
7831 				ctl_clr_prkey(lun, i);
7832 				ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
7833 			}
7834 
7835 			lun->pr_key_count = 1;
7836 			lun->res_type = msg->pr.pr_info.res_type;
7837 			if (lun->res_type != SPR_TYPE_WR_EX_AR
7838 			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7839 				lun->pr_res_idx = msg->pr.pr_info.residx;
7840 		} else {
7841 		        for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7842 				if (sa_res_key == ctl_get_prkey(lun, i))
7843 					continue;
7844 
7845 				ctl_clr_prkey(lun, i);
7846 				lun->pr_key_count--;
7847 				ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
7848 			}
7849 		}
7850 	} else {
7851 		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7852 			if (i == msg->pr.pr_info.residx ||
7853 			    ctl_get_prkey(lun, i) == 0)
7854 				continue;
7855 
7856 			if (sa_res_key == ctl_get_prkey(lun, i)) {
7857 				ctl_clr_prkey(lun, i);
7858 				lun->pr_key_count--;
7859 				ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
7860 			} else if (msg->pr.pr_info.res_type != lun->res_type
7861 				&& (lun->res_type == SPR_TYPE_WR_EX_RO
7862 				 || lun->res_type == SPR_TYPE_EX_AC_RO)) {
7863 				ctl_est_res_ua(lun, i, CTL_UA_RES_RELEASE);
7864 			}
7865 		}
7866 		lun->res_type = msg->pr.pr_info.res_type;
7867 		if (lun->res_type != SPR_TYPE_WR_EX_AR
7868 		 && lun->res_type != SPR_TYPE_EX_AC_AR)
7869 			lun->pr_res_idx = msg->pr.pr_info.residx;
7870 		else
7871 			lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
7872 	}
7873 	lun->PRGeneration++;
7874 
7875 }
7876 
7877 
7878 int
7879 ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
7880 {
7881 	int retval;
7882 	int isc_retval;
7883 	u_int32_t param_len;
7884 	struct scsi_per_res_out *cdb;
7885 	struct ctl_lun *lun;
7886 	struct scsi_per_res_out_parms* param;
7887 	struct ctl_softc *softc;
7888 	uint32_t residx;
7889 	uint64_t res_key, sa_res_key, key;
7890 	uint8_t type;
7891 	union ctl_ha_msg persis_io;
7892 	int    i;
7893 
7894 	CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
7895 
7896 	retval = CTL_RETVAL_COMPLETE;
7897 
7898 	cdb = (struct scsi_per_res_out *)ctsio->cdb;
7899 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7900 	softc = lun->ctl_softc;
7901 
7902 	/*
7903 	 * We only support whole-LUN scope.  The scope & type are ignored for
7904 	 * register, register and ignore existing key and clear.
7905 	 * We sometimes ignore scope and type on preempts too!!
7906 	 * Verify reservation type here as well.
7907 	 */
7908 	type = cdb->scope_type & SPR_TYPE_MASK;
7909 	if ((cdb->action == SPRO_RESERVE)
7910 	 || (cdb->action == SPRO_RELEASE)) {
7911 		if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
7912 			ctl_set_invalid_field(/*ctsio*/ ctsio,
7913 					      /*sks_valid*/ 1,
7914 					      /*command*/ 1,
7915 					      /*field*/ 2,
7916 					      /*bit_valid*/ 1,
7917 					      /*bit*/ 4);
7918 			ctl_done((union ctl_io *)ctsio);
7919 			return (CTL_RETVAL_COMPLETE);
7920 		}
7921 
7922 		if (type>8 || type==2 || type==4 || type==0) {
7923 			ctl_set_invalid_field(/*ctsio*/ ctsio,
7924 					      /*sks_valid*/ 1,
7925 					      /*command*/ 1,
7926 					      /*field*/ 2,
7927 					      /*bit_valid*/ 1,
7928 					      /*bit*/ 0);
7929 			ctl_done((union ctl_io *)ctsio);
7930 			return (CTL_RETVAL_COMPLETE);
7931 		}
7932 	}
7933 
7934 	param_len = scsi_4btoul(cdb->length);
7935 
7936 	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
7937 		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
7938 		ctsio->kern_data_len = param_len;
7939 		ctsio->kern_total_len = param_len;
7940 		ctsio->kern_data_resid = 0;
7941 		ctsio->kern_rel_offset = 0;
7942 		ctsio->kern_sg_entries = 0;
7943 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7944 		ctsio->be_move_done = ctl_config_move_done;
7945 		ctl_datamove((union ctl_io *)ctsio);
7946 
7947 		return (CTL_RETVAL_COMPLETE);
7948 	}
7949 
7950 	param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
7951 
7952 	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
7953 	res_key = scsi_8btou64(param->res_key.key);
7954 	sa_res_key = scsi_8btou64(param->serv_act_res_key);
7955 
7956 	/*
7957 	 * Validate the reservation key here except for SPRO_REG_IGNO
7958 	 * This must be done for all other service actions
7959 	 */
7960 	if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
7961 		mtx_lock(&lun->lun_lock);
7962 		if ((key = ctl_get_prkey(lun, residx)) != 0) {
7963 			if (res_key != key) {
7964 				/*
7965 				 * The current key passed in doesn't match
7966 				 * the one the initiator previously
7967 				 * registered.
7968 				 */
7969 				mtx_unlock(&lun->lun_lock);
7970 				free(ctsio->kern_data_ptr, M_CTL);
7971 				ctl_set_reservation_conflict(ctsio);
7972 				ctl_done((union ctl_io *)ctsio);
7973 				return (CTL_RETVAL_COMPLETE);
7974 			}
7975 		} else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
7976 			/*
7977 			 * We are not registered
7978 			 */
7979 			mtx_unlock(&lun->lun_lock);
7980 			free(ctsio->kern_data_ptr, M_CTL);
7981 			ctl_set_reservation_conflict(ctsio);
7982 			ctl_done((union ctl_io *)ctsio);
7983 			return (CTL_RETVAL_COMPLETE);
7984 		} else if (res_key != 0) {
7985 			/*
7986 			 * We are not registered and trying to register but
7987 			 * the register key isn't zero.
7988 			 */
7989 			mtx_unlock(&lun->lun_lock);
7990 			free(ctsio->kern_data_ptr, M_CTL);
7991 			ctl_set_reservation_conflict(ctsio);
7992 			ctl_done((union ctl_io *)ctsio);
7993 			return (CTL_RETVAL_COMPLETE);
7994 		}
7995 		mtx_unlock(&lun->lun_lock);
7996 	}
7997 
7998 	switch (cdb->action & SPRO_ACTION_MASK) {
7999 	case SPRO_REGISTER:
8000 	case SPRO_REG_IGNO: {
8001 
8002 #if 0
8003 		printf("Registration received\n");
8004 #endif
8005 
8006 		/*
8007 		 * We don't support any of these options, as we report in
8008 		 * the read capabilities request (see
8009 		 * ctl_persistent_reserve_in(), above).
8010 		 */
8011 		if ((param->flags & SPR_SPEC_I_PT)
8012 		 || (param->flags & SPR_ALL_TG_PT)
8013 		 || (param->flags & SPR_APTPL)) {
8014 			int bit_ptr;
8015 
8016 			if (param->flags & SPR_APTPL)
8017 				bit_ptr = 0;
8018 			else if (param->flags & SPR_ALL_TG_PT)
8019 				bit_ptr = 2;
8020 			else /* SPR_SPEC_I_PT */
8021 				bit_ptr = 3;
8022 
8023 			free(ctsio->kern_data_ptr, M_CTL);
8024 			ctl_set_invalid_field(ctsio,
8025 					      /*sks_valid*/ 1,
8026 					      /*command*/ 0,
8027 					      /*field*/ 20,
8028 					      /*bit_valid*/ 1,
8029 					      /*bit*/ bit_ptr);
8030 			ctl_done((union ctl_io *)ctsio);
8031 			return (CTL_RETVAL_COMPLETE);
8032 		}
8033 
8034 		mtx_lock(&lun->lun_lock);
8035 
8036 		/*
8037 		 * The initiator wants to clear the
8038 		 * key/unregister.
8039 		 */
8040 		if (sa_res_key == 0) {
8041 			if ((res_key == 0
8042 			  && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8043 			 || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8044 			  && ctl_get_prkey(lun, residx) == 0)) {
8045 				mtx_unlock(&lun->lun_lock);
8046 				goto done;
8047 			}
8048 
8049 			ctl_clr_prkey(lun, residx);
8050 			lun->pr_key_count--;
8051 
8052 			if (residx == lun->pr_res_idx) {
8053 				lun->flags &= ~CTL_LUN_PR_RESERVED;
8054 				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8055 
8056 				if ((lun->res_type == SPR_TYPE_WR_EX_RO
8057 				  || lun->res_type == SPR_TYPE_EX_AC_RO)
8058 				 && lun->pr_key_count) {
8059 					/*
8060 					 * If the reservation is a registrants
8061 					 * only type we need to generate a UA
8062 					 * for other registered inits.  The
8063 					 * sense code should be RESERVATIONS
8064 					 * RELEASED
8065 					 */
8066 
8067 					for (i = 0; i < CTL_MAX_INITIATORS;i++){
8068 						if (ctl_get_prkey(lun, i +
8069 						    softc->persis_offset) == 0)
8070 							continue;
8071 						ctl_est_ua(lun, i,
8072 						    CTL_UA_RES_RELEASE);
8073 					}
8074 				}
8075 				lun->res_type = 0;
8076 			} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8077 				if (lun->pr_key_count==0) {
8078 					lun->flags &= ~CTL_LUN_PR_RESERVED;
8079 					lun->res_type = 0;
8080 					lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8081 				}
8082 			}
8083 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8084 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8085 			persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8086 			persis_io.pr.pr_info.residx = residx;
8087 			if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8088 			     &persis_io, sizeof(persis_io), 0 )) >
8089 			     CTL_HA_STATUS_SUCCESS) {
8090 				printf("CTL:Persis Out error returned from "
8091 				       "ctl_ha_msg_send %d\n", isc_retval);
8092 			}
8093 		} else /* sa_res_key != 0 */ {
8094 
8095 			/*
8096 			 * If we aren't registered currently then increment
8097 			 * the key count and set the registered flag.
8098 			 */
8099 			ctl_alloc_prkey(lun, residx);
8100 			if (ctl_get_prkey(lun, residx) == 0)
8101 				lun->pr_key_count++;
8102 			ctl_set_prkey(lun, residx, sa_res_key);
8103 
8104 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8105 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8106 			persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8107 			persis_io.pr.pr_info.residx = residx;
8108 			memcpy(persis_io.pr.pr_info.sa_res_key,
8109 			       param->serv_act_res_key,
8110 			       sizeof(param->serv_act_res_key));
8111 			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8112 			     &persis_io, sizeof(persis_io), 0)) >
8113 			     CTL_HA_STATUS_SUCCESS) {
8114 				printf("CTL:Persis Out error returned from "
8115 				       "ctl_ha_msg_send %d\n", isc_retval);
8116 			}
8117 		}
8118 		lun->PRGeneration++;
8119 		mtx_unlock(&lun->lun_lock);
8120 
8121 		break;
8122 	}
8123 	case SPRO_RESERVE:
8124 #if 0
8125                 printf("Reserve executed type %d\n", type);
8126 #endif
8127 		mtx_lock(&lun->lun_lock);
8128 		if (lun->flags & CTL_LUN_PR_RESERVED) {
8129 			/*
8130 			 * if this isn't the reservation holder and it's
8131 			 * not a "all registrants" type or if the type is
8132 			 * different then we have a conflict
8133 			 */
8134 			if ((lun->pr_res_idx != residx
8135 			  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8136 			 || lun->res_type != type) {
8137 				mtx_unlock(&lun->lun_lock);
8138 				free(ctsio->kern_data_ptr, M_CTL);
8139 				ctl_set_reservation_conflict(ctsio);
8140 				ctl_done((union ctl_io *)ctsio);
8141 				return (CTL_RETVAL_COMPLETE);
8142 			}
8143 			mtx_unlock(&lun->lun_lock);
8144 		} else /* create a reservation */ {
8145 			/*
8146 			 * If it's not an "all registrants" type record
8147 			 * reservation holder
8148 			 */
8149 			if (type != SPR_TYPE_WR_EX_AR
8150 			 && type != SPR_TYPE_EX_AC_AR)
8151 				lun->pr_res_idx = residx; /* Res holder */
8152 			else
8153 				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8154 
8155 			lun->flags |= CTL_LUN_PR_RESERVED;
8156 			lun->res_type = type;
8157 
8158 			mtx_unlock(&lun->lun_lock);
8159 
8160 			/* send msg to other side */
8161 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8162 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8163 			persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8164 			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8165 			persis_io.pr.pr_info.res_type = type;
8166 			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8167 			     &persis_io, sizeof(persis_io), 0)) >
8168 			     CTL_HA_STATUS_SUCCESS) {
8169 				printf("CTL:Persis Out error returned from "
8170 				       "ctl_ha_msg_send %d\n", isc_retval);
8171 			}
8172 		}
8173 		break;
8174 
8175 	case SPRO_RELEASE:
8176 		mtx_lock(&lun->lun_lock);
8177 		if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8178 			/* No reservation exists return good status */
8179 			mtx_unlock(&lun->lun_lock);
8180 			goto done;
8181 		}
8182 		/*
8183 		 * Is this nexus a reservation holder?
8184 		 */
8185 		if (lun->pr_res_idx != residx
8186 		 && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8187 			/*
8188 			 * not a res holder return good status but
8189 			 * do nothing
8190 			 */
8191 			mtx_unlock(&lun->lun_lock);
8192 			goto done;
8193 		}
8194 
8195 		if (lun->res_type != type) {
8196 			mtx_unlock(&lun->lun_lock);
8197 			free(ctsio->kern_data_ptr, M_CTL);
8198 			ctl_set_illegal_pr_release(ctsio);
8199 			ctl_done((union ctl_io *)ctsio);
8200 			return (CTL_RETVAL_COMPLETE);
8201 		}
8202 
8203 		/* okay to release */
8204 		lun->flags &= ~CTL_LUN_PR_RESERVED;
8205 		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8206 		lun->res_type = 0;
8207 
8208 		/*
8209 		 * if this isn't an exclusive access
8210 		 * res generate UA for all other
8211 		 * registrants.
8212 		 */
8213 		if (type != SPR_TYPE_EX_AC
8214 		 && type != SPR_TYPE_WR_EX) {
8215 			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8216 				if (i == residx ||
8217 				    ctl_get_prkey(lun,
8218 				     i + softc->persis_offset) == 0)
8219 					continue;
8220 				ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8221 			}
8222 		}
8223 		mtx_unlock(&lun->lun_lock);
8224 		/* Send msg to other side */
8225 		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8226 		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8227 		persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8228 		if ((isc_retval=ctl_ha_msg_send( CTL_HA_CHAN_CTL, &persis_io,
8229 		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8230 			printf("CTL:Persis Out error returned from "
8231 			       "ctl_ha_msg_send %d\n", isc_retval);
8232 		}
8233 		break;
8234 
8235 	case SPRO_CLEAR:
8236 		/* send msg to other side */
8237 
8238 		mtx_lock(&lun->lun_lock);
8239 		lun->flags &= ~CTL_LUN_PR_RESERVED;
8240 		lun->res_type = 0;
8241 		lun->pr_key_count = 0;
8242 		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8243 
8244 		ctl_clr_prkey(lun, residx);
8245 		for (i=0; i < 2*CTL_MAX_INITIATORS; i++)
8246 			if (ctl_get_prkey(lun, i) != 0) {
8247 				ctl_clr_prkey(lun, i);
8248 				ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
8249 			}
8250 		lun->PRGeneration++;
8251 		mtx_unlock(&lun->lun_lock);
8252 		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8253 		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8254 		persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8255 		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8256 		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8257 			printf("CTL:Persis Out error returned from "
8258 			       "ctl_ha_msg_send %d\n", isc_retval);
8259 		}
8260 		break;
8261 
8262 	case SPRO_PREEMPT:
8263 	case SPRO_PRE_ABO: {
8264 		int nretval;
8265 
8266 		nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8267 					  residx, ctsio, cdb, param);
8268 		if (nretval != 0)
8269 			return (CTL_RETVAL_COMPLETE);
8270 		break;
8271 	}
8272 	default:
8273 		panic("Invalid PR type %x", cdb->action);
8274 	}
8275 
8276 done:
8277 	free(ctsio->kern_data_ptr, M_CTL);
8278 	ctl_set_success(ctsio);
8279 	ctl_done((union ctl_io *)ctsio);
8280 
8281 	return (retval);
8282 }
8283 
8284 /*
8285  * This routine is for handling a message from the other SC pertaining to
8286  * persistent reserve out. All the error checking will have been done
8287  * so only perorming the action need be done here to keep the two
8288  * in sync.
8289  */
8290 static void
8291 ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8292 {
8293 	struct ctl_lun *lun;
8294 	struct ctl_softc *softc;
8295 	int i;
8296 	uint32_t targ_lun;
8297 
8298 	softc = control_softc;
8299 
8300 	targ_lun = msg->hdr.nexus.targ_mapped_lun;
8301 	lun = softc->ctl_luns[targ_lun];
8302 	mtx_lock(&lun->lun_lock);
8303 	switch(msg->pr.pr_info.action) {
8304 	case CTL_PR_REG_KEY:
8305 		ctl_alloc_prkey(lun, msg->pr.pr_info.residx);
8306 		if (ctl_get_prkey(lun, msg->pr.pr_info.residx) == 0)
8307 			lun->pr_key_count++;
8308 		ctl_set_prkey(lun, msg->pr.pr_info.residx,
8309 		    scsi_8btou64(msg->pr.pr_info.sa_res_key));
8310 		lun->PRGeneration++;
8311 		break;
8312 
8313 	case CTL_PR_UNREG_KEY:
8314 		ctl_clr_prkey(lun, msg->pr.pr_info.residx);
8315 		lun->pr_key_count--;
8316 
8317 		/* XXX Need to see if the reservation has been released */
8318 		/* if so do we need to generate UA? */
8319 		if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8320 			lun->flags &= ~CTL_LUN_PR_RESERVED;
8321 			lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8322 
8323 			if ((lun->res_type == SPR_TYPE_WR_EX_RO
8324 			  || lun->res_type == SPR_TYPE_EX_AC_RO)
8325 			 && lun->pr_key_count) {
8326 				/*
8327 				 * If the reservation is a registrants
8328 				 * only type we need to generate a UA
8329 				 * for other registered inits.  The
8330 				 * sense code should be RESERVATIONS
8331 				 * RELEASED
8332 				 */
8333 
8334 				for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8335 					if (ctl_get_prkey(lun, i +
8336 					    softc->persis_offset) == 0)
8337 						continue;
8338 
8339 					ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8340 				}
8341 			}
8342 			lun->res_type = 0;
8343 		} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8344 			if (lun->pr_key_count==0) {
8345 				lun->flags &= ~CTL_LUN_PR_RESERVED;
8346 				lun->res_type = 0;
8347 				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8348 			}
8349 		}
8350 		lun->PRGeneration++;
8351 		break;
8352 
8353 	case CTL_PR_RESERVE:
8354 		lun->flags |= CTL_LUN_PR_RESERVED;
8355 		lun->res_type = msg->pr.pr_info.res_type;
8356 		lun->pr_res_idx = msg->pr.pr_info.residx;
8357 
8358 		break;
8359 
8360 	case CTL_PR_RELEASE:
8361 		/*
8362 		 * if this isn't an exclusive access res generate UA for all
8363 		 * other registrants.
8364 		 */
8365 		if (lun->res_type != SPR_TYPE_EX_AC
8366 		 && lun->res_type != SPR_TYPE_WR_EX) {
8367 			for (i = 0; i < CTL_MAX_INITIATORS; i++)
8368 				if (ctl_get_prkey(lun, i + softc->persis_offset) != 0)
8369 					ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8370 		}
8371 
8372 		lun->flags &= ~CTL_LUN_PR_RESERVED;
8373 		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8374 		lun->res_type = 0;
8375 		break;
8376 
8377 	case CTL_PR_PREEMPT:
8378 		ctl_pro_preempt_other(lun, msg);
8379 		break;
8380 	case CTL_PR_CLEAR:
8381 		lun->flags &= ~CTL_LUN_PR_RESERVED;
8382 		lun->res_type = 0;
8383 		lun->pr_key_count = 0;
8384 		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8385 
8386 		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8387 			if (ctl_get_prkey(lun, i) == 0)
8388 				continue;
8389 			ctl_clr_prkey(lun, i);
8390 			ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
8391 		}
8392 		lun->PRGeneration++;
8393 		break;
8394 	}
8395 
8396 	mtx_unlock(&lun->lun_lock);
8397 }
8398 
8399 int
8400 ctl_read_write(struct ctl_scsiio *ctsio)
8401 {
8402 	struct ctl_lun *lun;
8403 	struct ctl_lba_len_flags *lbalen;
8404 	uint64_t lba;
8405 	uint32_t num_blocks;
8406 	int flags, retval;
8407 	int isread;
8408 
8409 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8410 
8411 	CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8412 
8413 	flags = 0;
8414 	retval = CTL_RETVAL_COMPLETE;
8415 
8416 	isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8417 	      || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8418 	switch (ctsio->cdb[0]) {
8419 	case READ_6:
8420 	case WRITE_6: {
8421 		struct scsi_rw_6 *cdb;
8422 
8423 		cdb = (struct scsi_rw_6 *)ctsio->cdb;
8424 
8425 		lba = scsi_3btoul(cdb->addr);
8426 		/* only 5 bits are valid in the most significant address byte */
8427 		lba &= 0x1fffff;
8428 		num_blocks = cdb->length;
8429 		/*
8430 		 * This is correct according to SBC-2.
8431 		 */
8432 		if (num_blocks == 0)
8433 			num_blocks = 256;
8434 		break;
8435 	}
8436 	case READ_10:
8437 	case WRITE_10: {
8438 		struct scsi_rw_10 *cdb;
8439 
8440 		cdb = (struct scsi_rw_10 *)ctsio->cdb;
8441 		if (cdb->byte2 & SRW10_FUA)
8442 			flags |= CTL_LLF_FUA;
8443 		if (cdb->byte2 & SRW10_DPO)
8444 			flags |= CTL_LLF_DPO;
8445 		lba = scsi_4btoul(cdb->addr);
8446 		num_blocks = scsi_2btoul(cdb->length);
8447 		break;
8448 	}
8449 	case WRITE_VERIFY_10: {
8450 		struct scsi_write_verify_10 *cdb;
8451 
8452 		cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8453 		flags |= CTL_LLF_FUA;
8454 		if (cdb->byte2 & SWV_DPO)
8455 			flags |= CTL_LLF_DPO;
8456 		lba = scsi_4btoul(cdb->addr);
8457 		num_blocks = scsi_2btoul(cdb->length);
8458 		break;
8459 	}
8460 	case READ_12:
8461 	case WRITE_12: {
8462 		struct scsi_rw_12 *cdb;
8463 
8464 		cdb = (struct scsi_rw_12 *)ctsio->cdb;
8465 		if (cdb->byte2 & SRW12_FUA)
8466 			flags |= CTL_LLF_FUA;
8467 		if (cdb->byte2 & SRW12_DPO)
8468 			flags |= CTL_LLF_DPO;
8469 		lba = scsi_4btoul(cdb->addr);
8470 		num_blocks = scsi_4btoul(cdb->length);
8471 		break;
8472 	}
8473 	case WRITE_VERIFY_12: {
8474 		struct scsi_write_verify_12 *cdb;
8475 
8476 		cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8477 		flags |= CTL_LLF_FUA;
8478 		if (cdb->byte2 & SWV_DPO)
8479 			flags |= CTL_LLF_DPO;
8480 		lba = scsi_4btoul(cdb->addr);
8481 		num_blocks = scsi_4btoul(cdb->length);
8482 		break;
8483 	}
8484 	case READ_16:
8485 	case WRITE_16: {
8486 		struct scsi_rw_16 *cdb;
8487 
8488 		cdb = (struct scsi_rw_16 *)ctsio->cdb;
8489 		if (cdb->byte2 & SRW12_FUA)
8490 			flags |= CTL_LLF_FUA;
8491 		if (cdb->byte2 & SRW12_DPO)
8492 			flags |= CTL_LLF_DPO;
8493 		lba = scsi_8btou64(cdb->addr);
8494 		num_blocks = scsi_4btoul(cdb->length);
8495 		break;
8496 	}
8497 	case WRITE_ATOMIC_16: {
8498 		struct scsi_rw_16 *cdb;
8499 
8500 		if (lun->be_lun->atomicblock == 0) {
8501 			ctl_set_invalid_opcode(ctsio);
8502 			ctl_done((union ctl_io *)ctsio);
8503 			return (CTL_RETVAL_COMPLETE);
8504 		}
8505 
8506 		cdb = (struct scsi_rw_16 *)ctsio->cdb;
8507 		if (cdb->byte2 & SRW12_FUA)
8508 			flags |= CTL_LLF_FUA;
8509 		if (cdb->byte2 & SRW12_DPO)
8510 			flags |= CTL_LLF_DPO;
8511 		lba = scsi_8btou64(cdb->addr);
8512 		num_blocks = scsi_4btoul(cdb->length);
8513 		if (num_blocks > lun->be_lun->atomicblock) {
8514 			ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
8515 			    /*command*/ 1, /*field*/ 12, /*bit_valid*/ 0,
8516 			    /*bit*/ 0);
8517 			ctl_done((union ctl_io *)ctsio);
8518 			return (CTL_RETVAL_COMPLETE);
8519 		}
8520 		break;
8521 	}
8522 	case WRITE_VERIFY_16: {
8523 		struct scsi_write_verify_16 *cdb;
8524 
8525 		cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
8526 		flags |= CTL_LLF_FUA;
8527 		if (cdb->byte2 & SWV_DPO)
8528 			flags |= CTL_LLF_DPO;
8529 		lba = scsi_8btou64(cdb->addr);
8530 		num_blocks = scsi_4btoul(cdb->length);
8531 		break;
8532 	}
8533 	default:
8534 		/*
8535 		 * We got a command we don't support.  This shouldn't
8536 		 * happen, commands should be filtered out above us.
8537 		 */
8538 		ctl_set_invalid_opcode(ctsio);
8539 		ctl_done((union ctl_io *)ctsio);
8540 
8541 		return (CTL_RETVAL_COMPLETE);
8542 		break; /* NOTREACHED */
8543 	}
8544 
8545 	/*
8546 	 * The first check is to make sure we're in bounds, the second
8547 	 * check is to catch wrap-around problems.  If the lba + num blocks
8548 	 * is less than the lba, then we've wrapped around and the block
8549 	 * range is invalid anyway.
8550 	 */
8551 	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8552 	 || ((lba + num_blocks) < lba)) {
8553 		ctl_set_lba_out_of_range(ctsio);
8554 		ctl_done((union ctl_io *)ctsio);
8555 		return (CTL_RETVAL_COMPLETE);
8556 	}
8557 
8558 	/*
8559 	 * According to SBC-3, a transfer length of 0 is not an error.
8560 	 * Note that this cannot happen with WRITE(6) or READ(6), since 0
8561 	 * translates to 256 blocks for those commands.
8562 	 */
8563 	if (num_blocks == 0) {
8564 		ctl_set_success(ctsio);
8565 		ctl_done((union ctl_io *)ctsio);
8566 		return (CTL_RETVAL_COMPLETE);
8567 	}
8568 
8569 	/* Set FUA and/or DPO if caches are disabled. */
8570 	if (isread) {
8571 		if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
8572 		    SCP_RCD) != 0)
8573 			flags |= CTL_LLF_FUA | CTL_LLF_DPO;
8574 	} else {
8575 		if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
8576 		    SCP_WCE) == 0)
8577 			flags |= CTL_LLF_FUA;
8578 	}
8579 
8580 	lbalen = (struct ctl_lba_len_flags *)
8581 	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8582 	lbalen->lba = lba;
8583 	lbalen->len = num_blocks;
8584 	lbalen->flags = (isread ? CTL_LLF_READ : CTL_LLF_WRITE) | flags;
8585 
8586 	ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8587 	ctsio->kern_rel_offset = 0;
8588 
8589 	CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
8590 
8591 	retval = lun->backend->data_submit((union ctl_io *)ctsio);
8592 
8593 	return (retval);
8594 }
8595 
8596 static int
8597 ctl_cnw_cont(union ctl_io *io)
8598 {
8599 	struct ctl_scsiio *ctsio;
8600 	struct ctl_lun *lun;
8601 	struct ctl_lba_len_flags *lbalen;
8602 	int retval;
8603 
8604 	ctsio = &io->scsiio;
8605 	ctsio->io_hdr.status = CTL_STATUS_NONE;
8606 	ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
8607 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8608 	lbalen = (struct ctl_lba_len_flags *)
8609 	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8610 	lbalen->flags &= ~CTL_LLF_COMPARE;
8611 	lbalen->flags |= CTL_LLF_WRITE;
8612 
8613 	CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
8614 	retval = lun->backend->data_submit((union ctl_io *)ctsio);
8615 	return (retval);
8616 }
8617 
8618 int
8619 ctl_cnw(struct ctl_scsiio *ctsio)
8620 {
8621 	struct ctl_lun *lun;
8622 	struct ctl_lba_len_flags *lbalen;
8623 	uint64_t lba;
8624 	uint32_t num_blocks;
8625 	int flags, retval;
8626 
8627 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8628 
8629 	CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
8630 
8631 	flags = 0;
8632 	retval = CTL_RETVAL_COMPLETE;
8633 
8634 	switch (ctsio->cdb[0]) {
8635 	case COMPARE_AND_WRITE: {
8636 		struct scsi_compare_and_write *cdb;
8637 
8638 		cdb = (struct scsi_compare_and_write *)ctsio->cdb;
8639 		if (cdb->byte2 & SRW10_FUA)
8640 			flags |= CTL_LLF_FUA;
8641 		if (cdb->byte2 & SRW10_DPO)
8642 			flags |= CTL_LLF_DPO;
8643 		lba = scsi_8btou64(cdb->addr);
8644 		num_blocks = cdb->length;
8645 		break;
8646 	}
8647 	default:
8648 		/*
8649 		 * We got a command we don't support.  This shouldn't
8650 		 * happen, commands should be filtered out above us.
8651 		 */
8652 		ctl_set_invalid_opcode(ctsio);
8653 		ctl_done((union ctl_io *)ctsio);
8654 
8655 		return (CTL_RETVAL_COMPLETE);
8656 		break; /* NOTREACHED */
8657 	}
8658 
8659 	/*
8660 	 * The first check is to make sure we're in bounds, the second
8661 	 * check is to catch wrap-around problems.  If the lba + num blocks
8662 	 * is less than the lba, then we've wrapped around and the block
8663 	 * range is invalid anyway.
8664 	 */
8665 	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8666 	 || ((lba + num_blocks) < lba)) {
8667 		ctl_set_lba_out_of_range(ctsio);
8668 		ctl_done((union ctl_io *)ctsio);
8669 		return (CTL_RETVAL_COMPLETE);
8670 	}
8671 
8672 	/*
8673 	 * According to SBC-3, a transfer length of 0 is not an error.
8674 	 */
8675 	if (num_blocks == 0) {
8676 		ctl_set_success(ctsio);
8677 		ctl_done((union ctl_io *)ctsio);
8678 		return (CTL_RETVAL_COMPLETE);
8679 	}
8680 
8681 	/* Set FUA if write cache is disabled. */
8682 	if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
8683 	    SCP_WCE) == 0)
8684 		flags |= CTL_LLF_FUA;
8685 
8686 	ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
8687 	ctsio->kern_rel_offset = 0;
8688 
8689 	/*
8690 	 * Set the IO_CONT flag, so that if this I/O gets passed to
8691 	 * ctl_data_submit_done(), it'll get passed back to
8692 	 * ctl_ctl_cnw_cont() for further processing.
8693 	 */
8694 	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
8695 	ctsio->io_cont = ctl_cnw_cont;
8696 
8697 	lbalen = (struct ctl_lba_len_flags *)
8698 	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8699 	lbalen->lba = lba;
8700 	lbalen->len = num_blocks;
8701 	lbalen->flags = CTL_LLF_COMPARE | flags;
8702 
8703 	CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
8704 	retval = lun->backend->data_submit((union ctl_io *)ctsio);
8705 	return (retval);
8706 }
8707 
8708 int
8709 ctl_verify(struct ctl_scsiio *ctsio)
8710 {
8711 	struct ctl_lun *lun;
8712 	struct ctl_lba_len_flags *lbalen;
8713 	uint64_t lba;
8714 	uint32_t num_blocks;
8715 	int bytchk, flags;
8716 	int retval;
8717 
8718 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8719 
8720 	CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
8721 
8722 	bytchk = 0;
8723 	flags = CTL_LLF_FUA;
8724 	retval = CTL_RETVAL_COMPLETE;
8725 
8726 	switch (ctsio->cdb[0]) {
8727 	case VERIFY_10: {
8728 		struct scsi_verify_10 *cdb;
8729 
8730 		cdb = (struct scsi_verify_10 *)ctsio->cdb;
8731 		if (cdb->byte2 & SVFY_BYTCHK)
8732 			bytchk = 1;
8733 		if (cdb->byte2 & SVFY_DPO)
8734 			flags |= CTL_LLF_DPO;
8735 		lba = scsi_4btoul(cdb->addr);
8736 		num_blocks = scsi_2btoul(cdb->length);
8737 		break;
8738 	}
8739 	case VERIFY_12: {
8740 		struct scsi_verify_12 *cdb;
8741 
8742 		cdb = (struct scsi_verify_12 *)ctsio->cdb;
8743 		if (cdb->byte2 & SVFY_BYTCHK)
8744 			bytchk = 1;
8745 		if (cdb->byte2 & SVFY_DPO)
8746 			flags |= CTL_LLF_DPO;
8747 		lba = scsi_4btoul(cdb->addr);
8748 		num_blocks = scsi_4btoul(cdb->length);
8749 		break;
8750 	}
8751 	case VERIFY_16: {
8752 		struct scsi_rw_16 *cdb;
8753 
8754 		cdb = (struct scsi_rw_16 *)ctsio->cdb;
8755 		if (cdb->byte2 & SVFY_BYTCHK)
8756 			bytchk = 1;
8757 		if (cdb->byte2 & SVFY_DPO)
8758 			flags |= CTL_LLF_DPO;
8759 		lba = scsi_8btou64(cdb->addr);
8760 		num_blocks = scsi_4btoul(cdb->length);
8761 		break;
8762 	}
8763 	default:
8764 		/*
8765 		 * We got a command we don't support.  This shouldn't
8766 		 * happen, commands should be filtered out above us.
8767 		 */
8768 		ctl_set_invalid_opcode(ctsio);
8769 		ctl_done((union ctl_io *)ctsio);
8770 		return (CTL_RETVAL_COMPLETE);
8771 	}
8772 
8773 	/*
8774 	 * The first check is to make sure we're in bounds, the second
8775 	 * check is to catch wrap-around problems.  If the lba + num blocks
8776 	 * is less than the lba, then we've wrapped around and the block
8777 	 * range is invalid anyway.
8778 	 */
8779 	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8780 	 || ((lba + num_blocks) < lba)) {
8781 		ctl_set_lba_out_of_range(ctsio);
8782 		ctl_done((union ctl_io *)ctsio);
8783 		return (CTL_RETVAL_COMPLETE);
8784 	}
8785 
8786 	/*
8787 	 * According to SBC-3, a transfer length of 0 is not an error.
8788 	 */
8789 	if (num_blocks == 0) {
8790 		ctl_set_success(ctsio);
8791 		ctl_done((union ctl_io *)ctsio);
8792 		return (CTL_RETVAL_COMPLETE);
8793 	}
8794 
8795 	lbalen = (struct ctl_lba_len_flags *)
8796 	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8797 	lbalen->lba = lba;
8798 	lbalen->len = num_blocks;
8799 	if (bytchk) {
8800 		lbalen->flags = CTL_LLF_COMPARE | flags;
8801 		ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8802 	} else {
8803 		lbalen->flags = CTL_LLF_VERIFY | flags;
8804 		ctsio->kern_total_len = 0;
8805 	}
8806 	ctsio->kern_rel_offset = 0;
8807 
8808 	CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
8809 	retval = lun->backend->data_submit((union ctl_io *)ctsio);
8810 	return (retval);
8811 }
8812 
8813 int
8814 ctl_report_luns(struct ctl_scsiio *ctsio)
8815 {
8816 	struct ctl_softc *softc = control_softc;
8817 	struct scsi_report_luns *cdb;
8818 	struct scsi_report_luns_data *lun_data;
8819 	struct ctl_lun *lun, *request_lun;
8820 	struct ctl_port *port;
8821 	int num_luns, retval;
8822 	uint32_t alloc_len, lun_datalen;
8823 	int num_filled, well_known;
8824 	uint32_t initidx, targ_lun_id, lun_id;
8825 
8826 	retval = CTL_RETVAL_COMPLETE;
8827 	well_known = 0;
8828 
8829 	cdb = (struct scsi_report_luns *)ctsio->cdb;
8830 	port = ctl_io_port(&ctsio->io_hdr);
8831 
8832 	CTL_DEBUG_PRINT(("ctl_report_luns\n"));
8833 
8834 	mtx_lock(&softc->ctl_lock);
8835 	num_luns = 0;
8836 	for (targ_lun_id = 0; targ_lun_id < CTL_MAX_LUNS; targ_lun_id++) {
8837 		if (ctl_lun_map_from_port(port, targ_lun_id) < CTL_MAX_LUNS)
8838 			num_luns++;
8839 	}
8840 	mtx_unlock(&softc->ctl_lock);
8841 
8842 	switch (cdb->select_report) {
8843 	case RPL_REPORT_DEFAULT:
8844 	case RPL_REPORT_ALL:
8845 		break;
8846 	case RPL_REPORT_WELLKNOWN:
8847 		well_known = 1;
8848 		num_luns = 0;
8849 		break;
8850 	default:
8851 		ctl_set_invalid_field(ctsio,
8852 				      /*sks_valid*/ 1,
8853 				      /*command*/ 1,
8854 				      /*field*/ 2,
8855 				      /*bit_valid*/ 0,
8856 				      /*bit*/ 0);
8857 		ctl_done((union ctl_io *)ctsio);
8858 		return (retval);
8859 		break; /* NOTREACHED */
8860 	}
8861 
8862 	alloc_len = scsi_4btoul(cdb->length);
8863 	/*
8864 	 * The initiator has to allocate at least 16 bytes for this request,
8865 	 * so he can at least get the header and the first LUN.  Otherwise
8866 	 * we reject the request (per SPC-3 rev 14, section 6.21).
8867 	 */
8868 	if (alloc_len < (sizeof(struct scsi_report_luns_data) +
8869 	    sizeof(struct scsi_report_luns_lundata))) {
8870 		ctl_set_invalid_field(ctsio,
8871 				      /*sks_valid*/ 1,
8872 				      /*command*/ 1,
8873 				      /*field*/ 6,
8874 				      /*bit_valid*/ 0,
8875 				      /*bit*/ 0);
8876 		ctl_done((union ctl_io *)ctsio);
8877 		return (retval);
8878 	}
8879 
8880 	request_lun = (struct ctl_lun *)
8881 		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8882 
8883 	lun_datalen = sizeof(*lun_data) +
8884 		(num_luns * sizeof(struct scsi_report_luns_lundata));
8885 
8886 	ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
8887 	lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
8888 	ctsio->kern_sg_entries = 0;
8889 
8890 	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
8891 
8892 	mtx_lock(&softc->ctl_lock);
8893 	for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
8894 		lun_id = ctl_lun_map_from_port(port, targ_lun_id);
8895 		if (lun_id >= CTL_MAX_LUNS)
8896 			continue;
8897 		lun = softc->ctl_luns[lun_id];
8898 		if (lun == NULL)
8899 			continue;
8900 
8901 		if (targ_lun_id <= 0xff) {
8902 			/*
8903 			 * Peripheral addressing method, bus number 0.
8904 			 */
8905 			lun_data->luns[num_filled].lundata[0] =
8906 				RPL_LUNDATA_ATYP_PERIPH;
8907 			lun_data->luns[num_filled].lundata[1] = targ_lun_id;
8908 			num_filled++;
8909 		} else if (targ_lun_id <= 0x3fff) {
8910 			/*
8911 			 * Flat addressing method.
8912 			 */
8913 			lun_data->luns[num_filled].lundata[0] =
8914 				RPL_LUNDATA_ATYP_FLAT | (targ_lun_id >> 8);
8915 			lun_data->luns[num_filled].lundata[1] =
8916 				(targ_lun_id & 0xff);
8917 			num_filled++;
8918 		} else if (targ_lun_id <= 0xffffff) {
8919 			/*
8920 			 * Extended flat addressing method.
8921 			 */
8922 			lun_data->luns[num_filled].lundata[0] =
8923 			    RPL_LUNDATA_ATYP_EXTLUN | 0x12;
8924 			scsi_ulto3b(targ_lun_id,
8925 			    &lun_data->luns[num_filled].lundata[1]);
8926 			num_filled++;
8927 		} else {
8928 			printf("ctl_report_luns: bogus LUN number %jd, "
8929 			       "skipping\n", (intmax_t)targ_lun_id);
8930 		}
8931 		/*
8932 		 * According to SPC-3, rev 14 section 6.21:
8933 		 *
8934 		 * "The execution of a REPORT LUNS command to any valid and
8935 		 * installed logical unit shall clear the REPORTED LUNS DATA
8936 		 * HAS CHANGED unit attention condition for all logical
8937 		 * units of that target with respect to the requesting
8938 		 * initiator. A valid and installed logical unit is one
8939 		 * having a PERIPHERAL QUALIFIER of 000b in the standard
8940 		 * INQUIRY data (see 6.4.2)."
8941 		 *
8942 		 * If request_lun is NULL, the LUN this report luns command
8943 		 * was issued to is either disabled or doesn't exist. In that
8944 		 * case, we shouldn't clear any pending lun change unit
8945 		 * attention.
8946 		 */
8947 		if (request_lun != NULL) {
8948 			mtx_lock(&lun->lun_lock);
8949 			ctl_clr_ua(lun, initidx, CTL_UA_RES_RELEASE);
8950 			mtx_unlock(&lun->lun_lock);
8951 		}
8952 	}
8953 	mtx_unlock(&softc->ctl_lock);
8954 
8955 	/*
8956 	 * It's quite possible that we've returned fewer LUNs than we allocated
8957 	 * space for.  Trim it.
8958 	 */
8959 	lun_datalen = sizeof(*lun_data) +
8960 		(num_filled * sizeof(struct scsi_report_luns_lundata));
8961 
8962 	if (lun_datalen < alloc_len) {
8963 		ctsio->residual = alloc_len - lun_datalen;
8964 		ctsio->kern_data_len = lun_datalen;
8965 		ctsio->kern_total_len = lun_datalen;
8966 	} else {
8967 		ctsio->residual = 0;
8968 		ctsio->kern_data_len = alloc_len;
8969 		ctsio->kern_total_len = alloc_len;
8970 	}
8971 	ctsio->kern_data_resid = 0;
8972 	ctsio->kern_rel_offset = 0;
8973 	ctsio->kern_sg_entries = 0;
8974 
8975 	/*
8976 	 * We set this to the actual data length, regardless of how much
8977 	 * space we actually have to return results.  If the user looks at
8978 	 * this value, he'll know whether or not he allocated enough space
8979 	 * and reissue the command if necessary.  We don't support well
8980 	 * known logical units, so if the user asks for that, return none.
8981 	 */
8982 	scsi_ulto4b(lun_datalen - 8, lun_data->length);
8983 
8984 	/*
8985 	 * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
8986 	 * this request.
8987 	 */
8988 	ctl_set_success(ctsio);
8989 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8990 	ctsio->be_move_done = ctl_config_move_done;
8991 	ctl_datamove((union ctl_io *)ctsio);
8992 	return (retval);
8993 }
8994 
8995 int
8996 ctl_request_sense(struct ctl_scsiio *ctsio)
8997 {
8998 	struct scsi_request_sense *cdb;
8999 	struct scsi_sense_data *sense_ptr;
9000 	struct ctl_softc *ctl_softc;
9001 	struct ctl_lun *lun;
9002 	uint32_t initidx;
9003 	int have_error;
9004 	scsi_sense_data_type sense_format;
9005 	ctl_ua_type ua_type;
9006 
9007 	cdb = (struct scsi_request_sense *)ctsio->cdb;
9008 
9009 	ctl_softc = control_softc;
9010 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9011 
9012 	CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9013 
9014 	/*
9015 	 * Determine which sense format the user wants.
9016 	 */
9017 	if (cdb->byte2 & SRS_DESC)
9018 		sense_format = SSD_TYPE_DESC;
9019 	else
9020 		sense_format = SSD_TYPE_FIXED;
9021 
9022 	ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9023 	sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9024 	ctsio->kern_sg_entries = 0;
9025 
9026 	/*
9027 	 * struct scsi_sense_data, which is currently set to 256 bytes, is
9028 	 * larger than the largest allowed value for the length field in the
9029 	 * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9030 	 */
9031 	ctsio->residual = 0;
9032 	ctsio->kern_data_len = cdb->length;
9033 	ctsio->kern_total_len = cdb->length;
9034 
9035 	ctsio->kern_data_resid = 0;
9036 	ctsio->kern_rel_offset = 0;
9037 	ctsio->kern_sg_entries = 0;
9038 
9039 	/*
9040 	 * If we don't have a LUN, we don't have any pending sense.
9041 	 */
9042 	if (lun == NULL)
9043 		goto no_sense;
9044 
9045 	have_error = 0;
9046 	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9047 	/*
9048 	 * Check for pending sense, and then for pending unit attentions.
9049 	 * Pending sense gets returned first, then pending unit attentions.
9050 	 */
9051 	mtx_lock(&lun->lun_lock);
9052 #ifdef CTL_WITH_CA
9053 	if (ctl_is_set(lun->have_ca, initidx)) {
9054 		scsi_sense_data_type stored_format;
9055 
9056 		/*
9057 		 * Check to see which sense format was used for the stored
9058 		 * sense data.
9059 		 */
9060 		stored_format = scsi_sense_type(&lun->pending_sense[initidx]);
9061 
9062 		/*
9063 		 * If the user requested a different sense format than the
9064 		 * one we stored, then we need to convert it to the other
9065 		 * format.  If we're going from descriptor to fixed format
9066 		 * sense data, we may lose things in translation, depending
9067 		 * on what options were used.
9068 		 *
9069 		 * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9070 		 * for some reason we'll just copy it out as-is.
9071 		 */
9072 		if ((stored_format == SSD_TYPE_FIXED)
9073 		 && (sense_format == SSD_TYPE_DESC))
9074 			ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9075 			    &lun->pending_sense[initidx],
9076 			    (struct scsi_sense_data_desc *)sense_ptr);
9077 		else if ((stored_format == SSD_TYPE_DESC)
9078 		      && (sense_format == SSD_TYPE_FIXED))
9079 			ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9080 			    &lun->pending_sense[initidx],
9081 			    (struct scsi_sense_data_fixed *)sense_ptr);
9082 		else
9083 			memcpy(sense_ptr, &lun->pending_sense[initidx],
9084 			       MIN(sizeof(*sense_ptr),
9085 			       sizeof(lun->pending_sense[initidx])));
9086 
9087 		ctl_clear_mask(lun->have_ca, initidx);
9088 		have_error = 1;
9089 	} else
9090 #endif
9091 	{
9092 		ua_type = ctl_build_ua(lun, initidx, sense_ptr, sense_format);
9093 		if (ua_type != CTL_UA_NONE)
9094 			have_error = 1;
9095 		if (ua_type == CTL_UA_LUN_CHANGE) {
9096 			mtx_unlock(&lun->lun_lock);
9097 			mtx_lock(&ctl_softc->ctl_lock);
9098 			ctl_clear_ua(ctl_softc, initidx, ua_type);
9099 			mtx_unlock(&ctl_softc->ctl_lock);
9100 			mtx_lock(&lun->lun_lock);
9101 		}
9102 
9103 	}
9104 	mtx_unlock(&lun->lun_lock);
9105 
9106 	/*
9107 	 * We already have a pending error, return it.
9108 	 */
9109 	if (have_error != 0) {
9110 		/*
9111 		 * We report the SCSI status as OK, since the status of the
9112 		 * request sense command itself is OK.
9113 		 * We report 0 for the sense length, because we aren't doing
9114 		 * autosense in this case.  We're reporting sense as
9115 		 * parameter data.
9116 		 */
9117 		ctl_set_success(ctsio);
9118 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9119 		ctsio->be_move_done = ctl_config_move_done;
9120 		ctl_datamove((union ctl_io *)ctsio);
9121 		return (CTL_RETVAL_COMPLETE);
9122 	}
9123 
9124 no_sense:
9125 
9126 	/*
9127 	 * No sense information to report, so we report that everything is
9128 	 * okay.
9129 	 */
9130 	ctl_set_sense_data(sense_ptr,
9131 			   lun,
9132 			   sense_format,
9133 			   /*current_error*/ 1,
9134 			   /*sense_key*/ SSD_KEY_NO_SENSE,
9135 			   /*asc*/ 0x00,
9136 			   /*ascq*/ 0x00,
9137 			   SSD_ELEM_NONE);
9138 
9139 	/*
9140 	 * We report 0 for the sense length, because we aren't doing
9141 	 * autosense in this case.  We're reporting sense as parameter data.
9142 	 */
9143 	ctl_set_success(ctsio);
9144 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9145 	ctsio->be_move_done = ctl_config_move_done;
9146 	ctl_datamove((union ctl_io *)ctsio);
9147 	return (CTL_RETVAL_COMPLETE);
9148 }
9149 
9150 int
9151 ctl_tur(struct ctl_scsiio *ctsio)
9152 {
9153 
9154 	CTL_DEBUG_PRINT(("ctl_tur\n"));
9155 
9156 	ctl_set_success(ctsio);
9157 	ctl_done((union ctl_io *)ctsio);
9158 
9159 	return (CTL_RETVAL_COMPLETE);
9160 }
9161 
9162 #ifdef notyet
9163 static int
9164 ctl_cmddt_inquiry(struct ctl_scsiio *ctsio)
9165 {
9166 
9167 }
9168 #endif
9169 
9170 /*
9171  * SCSI VPD page 0x00, the Supported VPD Pages page.
9172  */
9173 static int
9174 ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9175 {
9176 	struct scsi_vpd_supported_pages *pages;
9177 	int sup_page_size;
9178 	struct ctl_lun *lun;
9179 	int p;
9180 
9181 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9182 
9183 	sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9184 	    SCSI_EVPD_NUM_SUPPORTED_PAGES;
9185 	ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9186 	pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9187 	ctsio->kern_sg_entries = 0;
9188 
9189 	if (sup_page_size < alloc_len) {
9190 		ctsio->residual = alloc_len - sup_page_size;
9191 		ctsio->kern_data_len = sup_page_size;
9192 		ctsio->kern_total_len = sup_page_size;
9193 	} else {
9194 		ctsio->residual = 0;
9195 		ctsio->kern_data_len = alloc_len;
9196 		ctsio->kern_total_len = alloc_len;
9197 	}
9198 	ctsio->kern_data_resid = 0;
9199 	ctsio->kern_rel_offset = 0;
9200 	ctsio->kern_sg_entries = 0;
9201 
9202 	/*
9203 	 * The control device is always connected.  The disk device, on the
9204 	 * other hand, may not be online all the time.  Need to change this
9205 	 * to figure out whether the disk device is actually online or not.
9206 	 */
9207 	if (lun != NULL)
9208 		pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9209 				lun->be_lun->lun_type;
9210 	else
9211 		pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9212 
9213 	p = 0;
9214 	/* Supported VPD pages */
9215 	pages->page_list[p++] = SVPD_SUPPORTED_PAGES;
9216 	/* Serial Number */
9217 	pages->page_list[p++] = SVPD_UNIT_SERIAL_NUMBER;
9218 	/* Device Identification */
9219 	pages->page_list[p++] = SVPD_DEVICE_ID;
9220 	/* Extended INQUIRY Data */
9221 	pages->page_list[p++] = SVPD_EXTENDED_INQUIRY_DATA;
9222 	/* Mode Page Policy */
9223 	pages->page_list[p++] = SVPD_MODE_PAGE_POLICY;
9224 	/* SCSI Ports */
9225 	pages->page_list[p++] = SVPD_SCSI_PORTS;
9226 	/* Third-party Copy */
9227 	pages->page_list[p++] = SVPD_SCSI_TPC;
9228 	if (lun != NULL && lun->be_lun->lun_type == T_DIRECT) {
9229 		/* Block limits */
9230 		pages->page_list[p++] = SVPD_BLOCK_LIMITS;
9231 		/* Block Device Characteristics */
9232 		pages->page_list[p++] = SVPD_BDC;
9233 		/* Logical Block Provisioning */
9234 		pages->page_list[p++] = SVPD_LBP;
9235 	}
9236 	pages->length = p;
9237 
9238 	ctl_set_success(ctsio);
9239 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9240 	ctsio->be_move_done = ctl_config_move_done;
9241 	ctl_datamove((union ctl_io *)ctsio);
9242 	return (CTL_RETVAL_COMPLETE);
9243 }
9244 
9245 /*
9246  * SCSI VPD page 0x80, the Unit Serial Number page.
9247  */
9248 static int
9249 ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9250 {
9251 	struct scsi_vpd_unit_serial_number *sn_ptr;
9252 	struct ctl_lun *lun;
9253 	int data_len;
9254 
9255 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9256 
9257 	data_len = 4 + CTL_SN_LEN;
9258 	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9259 	sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9260 	if (data_len < alloc_len) {
9261 		ctsio->residual = alloc_len - data_len;
9262 		ctsio->kern_data_len = data_len;
9263 		ctsio->kern_total_len = data_len;
9264 	} else {
9265 		ctsio->residual = 0;
9266 		ctsio->kern_data_len = alloc_len;
9267 		ctsio->kern_total_len = alloc_len;
9268 	}
9269 	ctsio->kern_data_resid = 0;
9270 	ctsio->kern_rel_offset = 0;
9271 	ctsio->kern_sg_entries = 0;
9272 
9273 	/*
9274 	 * The control device is always connected.  The disk device, on the
9275 	 * other hand, may not be online all the time.  Need to change this
9276 	 * to figure out whether the disk device is actually online or not.
9277 	 */
9278 	if (lun != NULL)
9279 		sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9280 				  lun->be_lun->lun_type;
9281 	else
9282 		sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9283 
9284 	sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9285 	sn_ptr->length = CTL_SN_LEN;
9286 	/*
9287 	 * If we don't have a LUN, we just leave the serial number as
9288 	 * all spaces.
9289 	 */
9290 	if (lun != NULL) {
9291 		strncpy((char *)sn_ptr->serial_num,
9292 			(char *)lun->be_lun->serial_num, CTL_SN_LEN);
9293 	} else
9294 		memset(sn_ptr->serial_num, 0x20, CTL_SN_LEN);
9295 
9296 	ctl_set_success(ctsio);
9297 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9298 	ctsio->be_move_done = ctl_config_move_done;
9299 	ctl_datamove((union ctl_io *)ctsio);
9300 	return (CTL_RETVAL_COMPLETE);
9301 }
9302 
9303 
9304 /*
9305  * SCSI VPD page 0x86, the Extended INQUIRY Data page.
9306  */
9307 static int
9308 ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len)
9309 {
9310 	struct scsi_vpd_extended_inquiry_data *eid_ptr;
9311 	struct ctl_lun *lun;
9312 	int data_len;
9313 
9314 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9315 
9316 	data_len = sizeof(struct scsi_vpd_extended_inquiry_data);
9317 	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9318 	eid_ptr = (struct scsi_vpd_extended_inquiry_data *)ctsio->kern_data_ptr;
9319 	ctsio->kern_sg_entries = 0;
9320 
9321 	if (data_len < alloc_len) {
9322 		ctsio->residual = alloc_len - data_len;
9323 		ctsio->kern_data_len = data_len;
9324 		ctsio->kern_total_len = data_len;
9325 	} else {
9326 		ctsio->residual = 0;
9327 		ctsio->kern_data_len = alloc_len;
9328 		ctsio->kern_total_len = alloc_len;
9329 	}
9330 	ctsio->kern_data_resid = 0;
9331 	ctsio->kern_rel_offset = 0;
9332 	ctsio->kern_sg_entries = 0;
9333 
9334 	/*
9335 	 * The control device is always connected.  The disk device, on the
9336 	 * other hand, may not be online all the time.
9337 	 */
9338 	if (lun != NULL)
9339 		eid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9340 				     lun->be_lun->lun_type;
9341 	else
9342 		eid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9343 	eid_ptr->page_code = SVPD_EXTENDED_INQUIRY_DATA;
9344 	scsi_ulto2b(data_len - 4, eid_ptr->page_length);
9345 	/*
9346 	 * We support head of queue, ordered and simple tags.
9347 	 */
9348 	eid_ptr->flags2 = SVPD_EID_HEADSUP | SVPD_EID_ORDSUP | SVPD_EID_SIMPSUP;
9349 	/*
9350 	 * Volatile cache supported.
9351 	 */
9352 	eid_ptr->flags3 = SVPD_EID_V_SUP;
9353 
9354 	/*
9355 	 * This means that we clear the REPORTED LUNS DATA HAS CHANGED unit
9356 	 * attention for a particular IT nexus on all LUNs once we report
9357 	 * it to that nexus once.  This bit is required as of SPC-4.
9358 	 */
9359 	eid_ptr->flags4 = SVPD_EID_LUICLT;
9360 
9361 	/*
9362 	 * XXX KDM in order to correctly answer this, we would need
9363 	 * information from the SIM to determine how much sense data it
9364 	 * can send.  So this would really be a path inquiry field, most
9365 	 * likely.  This can be set to a maximum of 252 according to SPC-4,
9366 	 * but the hardware may or may not be able to support that much.
9367 	 * 0 just means that the maximum sense data length is not reported.
9368 	 */
9369 	eid_ptr->max_sense_length = 0;
9370 
9371 	ctl_set_success(ctsio);
9372 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9373 	ctsio->be_move_done = ctl_config_move_done;
9374 	ctl_datamove((union ctl_io *)ctsio);
9375 	return (CTL_RETVAL_COMPLETE);
9376 }
9377 
9378 static int
9379 ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len)
9380 {
9381 	struct scsi_vpd_mode_page_policy *mpp_ptr;
9382 	struct ctl_lun *lun;
9383 	int data_len;
9384 
9385 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9386 
9387 	data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9388 	    sizeof(struct scsi_vpd_mode_page_policy_descr);
9389 
9390 	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9391 	mpp_ptr = (struct scsi_vpd_mode_page_policy *)ctsio->kern_data_ptr;
9392 	ctsio->kern_sg_entries = 0;
9393 
9394 	if (data_len < alloc_len) {
9395 		ctsio->residual = alloc_len - data_len;
9396 		ctsio->kern_data_len = data_len;
9397 		ctsio->kern_total_len = data_len;
9398 	} else {
9399 		ctsio->residual = 0;
9400 		ctsio->kern_data_len = alloc_len;
9401 		ctsio->kern_total_len = alloc_len;
9402 	}
9403 	ctsio->kern_data_resid = 0;
9404 	ctsio->kern_rel_offset = 0;
9405 	ctsio->kern_sg_entries = 0;
9406 
9407 	/*
9408 	 * The control device is always connected.  The disk device, on the
9409 	 * other hand, may not be online all the time.
9410 	 */
9411 	if (lun != NULL)
9412 		mpp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9413 				     lun->be_lun->lun_type;
9414 	else
9415 		mpp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9416 	mpp_ptr->page_code = SVPD_MODE_PAGE_POLICY;
9417 	scsi_ulto2b(data_len - 4, mpp_ptr->page_length);
9418 	mpp_ptr->descr[0].page_code = 0x3f;
9419 	mpp_ptr->descr[0].subpage_code = 0xff;
9420 	mpp_ptr->descr[0].policy = SVPD_MPP_SHARED;
9421 
9422 	ctl_set_success(ctsio);
9423 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9424 	ctsio->be_move_done = ctl_config_move_done;
9425 	ctl_datamove((union ctl_io *)ctsio);
9426 	return (CTL_RETVAL_COMPLETE);
9427 }
9428 
9429 /*
9430  * SCSI VPD page 0x83, the Device Identification page.
9431  */
9432 static int
9433 ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9434 {
9435 	struct scsi_vpd_device_id *devid_ptr;
9436 	struct scsi_vpd_id_descriptor *desc;
9437 	struct ctl_softc *softc;
9438 	struct ctl_lun *lun;
9439 	struct ctl_port *port;
9440 	int data_len;
9441 	uint8_t proto;
9442 
9443 	softc = control_softc;
9444 
9445 	port = softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)];
9446 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9447 
9448 	data_len = sizeof(struct scsi_vpd_device_id) +
9449 	    sizeof(struct scsi_vpd_id_descriptor) +
9450 		sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9451 	    sizeof(struct scsi_vpd_id_descriptor) +
9452 		sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9453 	if (lun && lun->lun_devid)
9454 		data_len += lun->lun_devid->len;
9455 	if (port->port_devid)
9456 		data_len += port->port_devid->len;
9457 	if (port->target_devid)
9458 		data_len += port->target_devid->len;
9459 
9460 	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9461 	devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9462 	ctsio->kern_sg_entries = 0;
9463 
9464 	if (data_len < alloc_len) {
9465 		ctsio->residual = alloc_len - data_len;
9466 		ctsio->kern_data_len = data_len;
9467 		ctsio->kern_total_len = data_len;
9468 	} else {
9469 		ctsio->residual = 0;
9470 		ctsio->kern_data_len = alloc_len;
9471 		ctsio->kern_total_len = alloc_len;
9472 	}
9473 	ctsio->kern_data_resid = 0;
9474 	ctsio->kern_rel_offset = 0;
9475 	ctsio->kern_sg_entries = 0;
9476 
9477 	/*
9478 	 * The control device is always connected.  The disk device, on the
9479 	 * other hand, may not be online all the time.
9480 	 */
9481 	if (lun != NULL)
9482 		devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9483 				     lun->be_lun->lun_type;
9484 	else
9485 		devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9486 	devid_ptr->page_code = SVPD_DEVICE_ID;
9487 	scsi_ulto2b(data_len - 4, devid_ptr->length);
9488 
9489 	if (port->port_type == CTL_PORT_FC)
9490 		proto = SCSI_PROTO_FC << 4;
9491 	else if (port->port_type == CTL_PORT_ISCSI)
9492 		proto = SCSI_PROTO_ISCSI << 4;
9493 	else
9494 		proto = SCSI_PROTO_SPI << 4;
9495 	desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9496 
9497 	/*
9498 	 * We're using a LUN association here.  i.e., this device ID is a
9499 	 * per-LUN identifier.
9500 	 */
9501 	if (lun && lun->lun_devid) {
9502 		memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
9503 		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9504 		    lun->lun_devid->len);
9505 	}
9506 
9507 	/*
9508 	 * This is for the WWPN which is a port association.
9509 	 */
9510 	if (port->port_devid) {
9511 		memcpy(desc, port->port_devid->data, port->port_devid->len);
9512 		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9513 		    port->port_devid->len);
9514 	}
9515 
9516 	/*
9517 	 * This is for the Relative Target Port(type 4h) identifier
9518 	 */
9519 	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9520 	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9521 	    SVPD_ID_TYPE_RELTARG;
9522 	desc->length = 4;
9523 	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
9524 	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9525 	    sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9526 
9527 	/*
9528 	 * This is for the Target Port Group(type 5h) identifier
9529 	 */
9530 	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9531 	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9532 	    SVPD_ID_TYPE_TPORTGRP;
9533 	desc->length = 4;
9534 	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS + 1,
9535 	    &desc->identifier[2]);
9536 	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9537 	    sizeof(struct scsi_vpd_id_trgt_port_grp_id));
9538 
9539 	/*
9540 	 * This is for the Target identifier
9541 	 */
9542 	if (port->target_devid) {
9543 		memcpy(desc, port->target_devid->data, port->target_devid->len);
9544 	}
9545 
9546 	ctl_set_success(ctsio);
9547 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9548 	ctsio->be_move_done = ctl_config_move_done;
9549 	ctl_datamove((union ctl_io *)ctsio);
9550 	return (CTL_RETVAL_COMPLETE);
9551 }
9552 
9553 static int
9554 ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
9555 {
9556 	struct ctl_softc *softc = control_softc;
9557 	struct scsi_vpd_scsi_ports *sp;
9558 	struct scsi_vpd_port_designation *pd;
9559 	struct scsi_vpd_port_designation_cont *pdc;
9560 	struct ctl_lun *lun;
9561 	struct ctl_port *port;
9562 	int data_len, num_target_ports, iid_len, id_len, g, pg, p;
9563 	int num_target_port_groups;
9564 
9565 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9566 
9567 	if (softc->is_single)
9568 		num_target_port_groups = 1;
9569 	else
9570 		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
9571 	num_target_ports = 0;
9572 	iid_len = 0;
9573 	id_len = 0;
9574 	mtx_lock(&softc->ctl_lock);
9575 	STAILQ_FOREACH(port, &softc->port_list, links) {
9576 		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9577 			continue;
9578 		if (lun != NULL &&
9579 		    ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
9580 			continue;
9581 		num_target_ports++;
9582 		if (port->init_devid)
9583 			iid_len += port->init_devid->len;
9584 		if (port->port_devid)
9585 			id_len += port->port_devid->len;
9586 	}
9587 	mtx_unlock(&softc->ctl_lock);
9588 
9589 	data_len = sizeof(struct scsi_vpd_scsi_ports) + num_target_port_groups *
9590 	    num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
9591 	     sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
9592 	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9593 	sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
9594 	ctsio->kern_sg_entries = 0;
9595 
9596 	if (data_len < alloc_len) {
9597 		ctsio->residual = alloc_len - data_len;
9598 		ctsio->kern_data_len = data_len;
9599 		ctsio->kern_total_len = data_len;
9600 	} else {
9601 		ctsio->residual = 0;
9602 		ctsio->kern_data_len = alloc_len;
9603 		ctsio->kern_total_len = alloc_len;
9604 	}
9605 	ctsio->kern_data_resid = 0;
9606 	ctsio->kern_rel_offset = 0;
9607 	ctsio->kern_sg_entries = 0;
9608 
9609 	/*
9610 	 * The control device is always connected.  The disk device, on the
9611 	 * other hand, may not be online all the time.  Need to change this
9612 	 * to figure out whether the disk device is actually online or not.
9613 	 */
9614 	if (lun != NULL)
9615 		sp->device = (SID_QUAL_LU_CONNECTED << 5) |
9616 				  lun->be_lun->lun_type;
9617 	else
9618 		sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9619 
9620 	sp->page_code = SVPD_SCSI_PORTS;
9621 	scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
9622 	    sp->page_length);
9623 	pd = &sp->design[0];
9624 
9625 	mtx_lock(&softc->ctl_lock);
9626 	pg = softc->port_offset / CTL_MAX_PORTS;
9627 	for (g = 0; g < num_target_port_groups; g++) {
9628 		STAILQ_FOREACH(port, &softc->port_list, links) {
9629 			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9630 				continue;
9631 			if (lun != NULL &&
9632 			    ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
9633 				continue;
9634 			p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
9635 			scsi_ulto2b(p, pd->relative_port_id);
9636 			if (port->init_devid && g == pg) {
9637 				iid_len = port->init_devid->len;
9638 				memcpy(pd->initiator_transportid,
9639 				    port->init_devid->data, port->init_devid->len);
9640 			} else
9641 				iid_len = 0;
9642 			scsi_ulto2b(iid_len, pd->initiator_transportid_length);
9643 			pdc = (struct scsi_vpd_port_designation_cont *)
9644 			    (&pd->initiator_transportid[iid_len]);
9645 			if (port->port_devid && g == pg) {
9646 				id_len = port->port_devid->len;
9647 				memcpy(pdc->target_port_descriptors,
9648 				    port->port_devid->data, port->port_devid->len);
9649 			} else
9650 				id_len = 0;
9651 			scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
9652 			pd = (struct scsi_vpd_port_designation *)
9653 			    ((uint8_t *)pdc->target_port_descriptors + id_len);
9654 		}
9655 	}
9656 	mtx_unlock(&softc->ctl_lock);
9657 
9658 	ctl_set_success(ctsio);
9659 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9660 	ctsio->be_move_done = ctl_config_move_done;
9661 	ctl_datamove((union ctl_io *)ctsio);
9662 	return (CTL_RETVAL_COMPLETE);
9663 }
9664 
9665 static int
9666 ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
9667 {
9668 	struct scsi_vpd_block_limits *bl_ptr;
9669 	struct ctl_lun *lun;
9670 	int bs;
9671 
9672 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9673 
9674 	ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
9675 	bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
9676 	ctsio->kern_sg_entries = 0;
9677 
9678 	if (sizeof(*bl_ptr) < alloc_len) {
9679 		ctsio->residual = alloc_len - sizeof(*bl_ptr);
9680 		ctsio->kern_data_len = sizeof(*bl_ptr);
9681 		ctsio->kern_total_len = sizeof(*bl_ptr);
9682 	} else {
9683 		ctsio->residual = 0;
9684 		ctsio->kern_data_len = alloc_len;
9685 		ctsio->kern_total_len = alloc_len;
9686 	}
9687 	ctsio->kern_data_resid = 0;
9688 	ctsio->kern_rel_offset = 0;
9689 	ctsio->kern_sg_entries = 0;
9690 
9691 	/*
9692 	 * The control device is always connected.  The disk device, on the
9693 	 * other hand, may not be online all the time.  Need to change this
9694 	 * to figure out whether the disk device is actually online or not.
9695 	 */
9696 	if (lun != NULL)
9697 		bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9698 				  lun->be_lun->lun_type;
9699 	else
9700 		bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9701 
9702 	bl_ptr->page_code = SVPD_BLOCK_LIMITS;
9703 	scsi_ulto2b(sizeof(*bl_ptr) - 4, bl_ptr->page_length);
9704 	bl_ptr->max_cmp_write_len = 0xff;
9705 	scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
9706 	if (lun != NULL) {
9707 		bs = lun->be_lun->blocksize;
9708 		scsi_ulto4b(lun->be_lun->opttxferlen, bl_ptr->opt_txfer_len);
9709 		if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9710 			scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_lba_cnt);
9711 			scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_blk_cnt);
9712 			if (lun->be_lun->ublockexp != 0) {
9713 				scsi_ulto4b((1 << lun->be_lun->ublockexp),
9714 				    bl_ptr->opt_unmap_grain);
9715 				scsi_ulto4b(0x80000000 | lun->be_lun->ublockoff,
9716 				    bl_ptr->unmap_grain_align);
9717 			}
9718 		}
9719 		scsi_ulto4b(lun->be_lun->atomicblock,
9720 		    bl_ptr->max_atomic_transfer_length);
9721 		scsi_ulto4b(0, bl_ptr->atomic_alignment);
9722 		scsi_ulto4b(0, bl_ptr->atomic_transfer_length_granularity);
9723 	}
9724 	scsi_u64to8b(UINT64_MAX, bl_ptr->max_write_same_length);
9725 
9726 	ctl_set_success(ctsio);
9727 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9728 	ctsio->be_move_done = ctl_config_move_done;
9729 	ctl_datamove((union ctl_io *)ctsio);
9730 	return (CTL_RETVAL_COMPLETE);
9731 }
9732 
9733 static int
9734 ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len)
9735 {
9736 	struct scsi_vpd_block_device_characteristics *bdc_ptr;
9737 	struct ctl_lun *lun;
9738 	const char *value;
9739 	u_int i;
9740 
9741 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9742 
9743 	ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO);
9744 	bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr;
9745 	ctsio->kern_sg_entries = 0;
9746 
9747 	if (sizeof(*bdc_ptr) < alloc_len) {
9748 		ctsio->residual = alloc_len - sizeof(*bdc_ptr);
9749 		ctsio->kern_data_len = sizeof(*bdc_ptr);
9750 		ctsio->kern_total_len = sizeof(*bdc_ptr);
9751 	} else {
9752 		ctsio->residual = 0;
9753 		ctsio->kern_data_len = alloc_len;
9754 		ctsio->kern_total_len = alloc_len;
9755 	}
9756 	ctsio->kern_data_resid = 0;
9757 	ctsio->kern_rel_offset = 0;
9758 	ctsio->kern_sg_entries = 0;
9759 
9760 	/*
9761 	 * The control device is always connected.  The disk device, on the
9762 	 * other hand, may not be online all the time.  Need to change this
9763 	 * to figure out whether the disk device is actually online or not.
9764 	 */
9765 	if (lun != NULL)
9766 		bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9767 				  lun->be_lun->lun_type;
9768 	else
9769 		bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9770 	bdc_ptr->page_code = SVPD_BDC;
9771 	scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length);
9772 	if (lun != NULL &&
9773 	    (value = ctl_get_opt(&lun->be_lun->options, "rpm")) != NULL)
9774 		i = strtol(value, NULL, 0);
9775 	else
9776 		i = CTL_DEFAULT_ROTATION_RATE;
9777 	scsi_ulto2b(i, bdc_ptr->medium_rotation_rate);
9778 	if (lun != NULL &&
9779 	    (value = ctl_get_opt(&lun->be_lun->options, "formfactor")) != NULL)
9780 		i = strtol(value, NULL, 0);
9781 	else
9782 		i = 0;
9783 	bdc_ptr->wab_wac_ff = (i & 0x0f);
9784 	bdc_ptr->flags = SVPD_FUAB | SVPD_VBULS;
9785 
9786 	ctl_set_success(ctsio);
9787 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9788 	ctsio->be_move_done = ctl_config_move_done;
9789 	ctl_datamove((union ctl_io *)ctsio);
9790 	return (CTL_RETVAL_COMPLETE);
9791 }
9792 
9793 static int
9794 ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
9795 {
9796 	struct scsi_vpd_logical_block_prov *lbp_ptr;
9797 	struct ctl_lun *lun;
9798 
9799 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9800 
9801 	ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
9802 	lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
9803 	ctsio->kern_sg_entries = 0;
9804 
9805 	if (sizeof(*lbp_ptr) < alloc_len) {
9806 		ctsio->residual = alloc_len - sizeof(*lbp_ptr);
9807 		ctsio->kern_data_len = sizeof(*lbp_ptr);
9808 		ctsio->kern_total_len = sizeof(*lbp_ptr);
9809 	} else {
9810 		ctsio->residual = 0;
9811 		ctsio->kern_data_len = alloc_len;
9812 		ctsio->kern_total_len = alloc_len;
9813 	}
9814 	ctsio->kern_data_resid = 0;
9815 	ctsio->kern_rel_offset = 0;
9816 	ctsio->kern_sg_entries = 0;
9817 
9818 	/*
9819 	 * The control device is always connected.  The disk device, on the
9820 	 * other hand, may not be online all the time.  Need to change this
9821 	 * to figure out whether the disk device is actually online or not.
9822 	 */
9823 	if (lun != NULL)
9824 		lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9825 				  lun->be_lun->lun_type;
9826 	else
9827 		lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9828 
9829 	lbp_ptr->page_code = SVPD_LBP;
9830 	scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length);
9831 	lbp_ptr->threshold_exponent = CTL_LBP_EXPONENT;
9832 	if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9833 		lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 |
9834 		    SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP;
9835 		lbp_ptr->prov_type = SVPD_LBP_THIN;
9836 	}
9837 
9838 	ctl_set_success(ctsio);
9839 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9840 	ctsio->be_move_done = ctl_config_move_done;
9841 	ctl_datamove((union ctl_io *)ctsio);
9842 	return (CTL_RETVAL_COMPLETE);
9843 }
9844 
9845 /*
9846  * INQUIRY with the EVPD bit set.
9847  */
9848 static int
9849 ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
9850 {
9851 	struct ctl_lun *lun;
9852 	struct scsi_inquiry *cdb;
9853 	int alloc_len, retval;
9854 
9855 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9856 	cdb = (struct scsi_inquiry *)ctsio->cdb;
9857 	alloc_len = scsi_2btoul(cdb->length);
9858 
9859 	switch (cdb->page_code) {
9860 	case SVPD_SUPPORTED_PAGES:
9861 		retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
9862 		break;
9863 	case SVPD_UNIT_SERIAL_NUMBER:
9864 		retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
9865 		break;
9866 	case SVPD_DEVICE_ID:
9867 		retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
9868 		break;
9869 	case SVPD_EXTENDED_INQUIRY_DATA:
9870 		retval = ctl_inquiry_evpd_eid(ctsio, alloc_len);
9871 		break;
9872 	case SVPD_MODE_PAGE_POLICY:
9873 		retval = ctl_inquiry_evpd_mpp(ctsio, alloc_len);
9874 		break;
9875 	case SVPD_SCSI_PORTS:
9876 		retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
9877 		break;
9878 	case SVPD_SCSI_TPC:
9879 		retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
9880 		break;
9881 	case SVPD_BLOCK_LIMITS:
9882 		if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
9883 			goto err;
9884 		retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
9885 		break;
9886 	case SVPD_BDC:
9887 		if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
9888 			goto err;
9889 		retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len);
9890 		break;
9891 	case SVPD_LBP:
9892 		if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
9893 			goto err;
9894 		retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
9895 		break;
9896 	default:
9897 err:
9898 		ctl_set_invalid_field(ctsio,
9899 				      /*sks_valid*/ 1,
9900 				      /*command*/ 1,
9901 				      /*field*/ 2,
9902 				      /*bit_valid*/ 0,
9903 				      /*bit*/ 0);
9904 		ctl_done((union ctl_io *)ctsio);
9905 		retval = CTL_RETVAL_COMPLETE;
9906 		break;
9907 	}
9908 
9909 	return (retval);
9910 }
9911 
9912 /*
9913  * Standard INQUIRY data.
9914  */
9915 static int
9916 ctl_inquiry_std(struct ctl_scsiio *ctsio)
9917 {
9918 	struct scsi_inquiry_data *inq_ptr;
9919 	struct scsi_inquiry *cdb;
9920 	struct ctl_softc *softc;
9921 	struct ctl_lun *lun;
9922 	char *val;
9923 	uint32_t alloc_len, data_len;
9924 	ctl_port_type port_type;
9925 
9926 	softc = control_softc;
9927 
9928 	/*
9929 	 * Figure out whether we're talking to a Fibre Channel port or not.
9930 	 * We treat the ioctl front end, and any SCSI adapters, as packetized
9931 	 * SCSI front ends.
9932 	 */
9933 	port_type = softc->ctl_ports[
9934 	    ctl_port_idx(ctsio->io_hdr.nexus.targ_port)]->port_type;
9935 	if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
9936 		port_type = CTL_PORT_SCSI;
9937 
9938 	lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9939 	cdb = (struct scsi_inquiry *)ctsio->cdb;
9940 	alloc_len = scsi_2btoul(cdb->length);
9941 
9942 	/*
9943 	 * We malloc the full inquiry data size here and fill it
9944 	 * in.  If the user only asks for less, we'll give him
9945 	 * that much.
9946 	 */
9947 	data_len = offsetof(struct scsi_inquiry_data, vendor_specific1);
9948 	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9949 	inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
9950 	ctsio->kern_sg_entries = 0;
9951 	ctsio->kern_data_resid = 0;
9952 	ctsio->kern_rel_offset = 0;
9953 
9954 	if (data_len < alloc_len) {
9955 		ctsio->residual = alloc_len - data_len;
9956 		ctsio->kern_data_len = data_len;
9957 		ctsio->kern_total_len = data_len;
9958 	} else {
9959 		ctsio->residual = 0;
9960 		ctsio->kern_data_len = alloc_len;
9961 		ctsio->kern_total_len = alloc_len;
9962 	}
9963 
9964 	/*
9965 	 * If we have a LUN configured, report it as connected.  Otherwise,
9966 	 * report that it is offline or no device is supported, depending
9967 	 * on the value of inquiry_pq_no_lun.
9968 	 *
9969 	 * According to the spec (SPC-4 r34), the peripheral qualifier
9970 	 * SID_QUAL_LU_OFFLINE (001b) is used in the following scenario:
9971 	 *
9972 	 * "A peripheral device having the specified peripheral device type
9973 	 * is not connected to this logical unit. However, the device
9974 	 * server is capable of supporting the specified peripheral device
9975 	 * type on this logical unit."
9976 	 *
9977 	 * According to the same spec, the peripheral qualifier
9978 	 * SID_QUAL_BAD_LU (011b) is used in this scenario:
9979 	 *
9980 	 * "The device server is not capable of supporting a peripheral
9981 	 * device on this logical unit. For this peripheral qualifier the
9982 	 * peripheral device type shall be set to 1Fh. All other peripheral
9983 	 * device type values are reserved for this peripheral qualifier."
9984 	 *
9985 	 * Given the text, it would seem that we probably want to report that
9986 	 * the LUN is offline here.  There is no LUN connected, but we can
9987 	 * support a LUN at the given LUN number.
9988 	 *
9989 	 * In the real world, though, it sounds like things are a little
9990 	 * different:
9991 	 *
9992 	 * - Linux, when presented with a LUN with the offline peripheral
9993 	 *   qualifier, will create an sg driver instance for it.  So when
9994 	 *   you attach it to CTL, you wind up with a ton of sg driver
9995 	 *   instances.  (One for every LUN that Linux bothered to probe.)
9996 	 *   Linux does this despite the fact that it issues a REPORT LUNs
9997 	 *   to LUN 0 to get the inventory of supported LUNs.
9998 	 *
9999 	 * - There is other anecdotal evidence (from Emulex folks) about
10000 	 *   arrays that use the offline peripheral qualifier for LUNs that
10001 	 *   are on the "passive" path in an active/passive array.
10002 	 *
10003 	 * So the solution is provide a hopefully reasonable default
10004 	 * (return bad/no LUN) and allow the user to change the behavior
10005 	 * with a tunable/sysctl variable.
10006 	 */
10007 	if (lun != NULL)
10008 		inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10009 				  lun->be_lun->lun_type;
10010 	else if (softc->inquiry_pq_no_lun == 0)
10011 		inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10012 	else
10013 		inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10014 
10015 	/* RMB in byte 2 is 0 */
10016 	inq_ptr->version = SCSI_REV_SPC4;
10017 
10018 	/*
10019 	 * According to SAM-3, even if a device only supports a single
10020 	 * level of LUN addressing, it should still set the HISUP bit:
10021 	 *
10022 	 * 4.9.1 Logical unit numbers overview
10023 	 *
10024 	 * All logical unit number formats described in this standard are
10025 	 * hierarchical in structure even when only a single level in that
10026 	 * hierarchy is used. The HISUP bit shall be set to one in the
10027 	 * standard INQUIRY data (see SPC-2) when any logical unit number
10028 	 * format described in this standard is used.  Non-hierarchical
10029 	 * formats are outside the scope of this standard.
10030 	 *
10031 	 * Therefore we set the HiSup bit here.
10032 	 *
10033 	 * The reponse format is 2, per SPC-3.
10034 	 */
10035 	inq_ptr->response_format = SID_HiSup | 2;
10036 
10037 	inq_ptr->additional_length = data_len -
10038 	    (offsetof(struct scsi_inquiry_data, additional_length) + 1);
10039 	CTL_DEBUG_PRINT(("additional_length = %d\n",
10040 			 inq_ptr->additional_length));
10041 
10042 	inq_ptr->spc3_flags = SPC3_SID_3PC | SPC3_SID_TPGS_IMPLICIT;
10043 	/* 16 bit addressing */
10044 	if (port_type == CTL_PORT_SCSI)
10045 		inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10046 	/* XXX set the SID_MultiP bit here if we're actually going to
10047 	   respond on multiple ports */
10048 	inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10049 
10050 	/* 16 bit data bus, synchronous transfers */
10051 	if (port_type == CTL_PORT_SCSI)
10052 		inq_ptr->flags = SID_WBus16 | SID_Sync;
10053 	/*
10054 	 * XXX KDM do we want to support tagged queueing on the control
10055 	 * device at all?
10056 	 */
10057 	if ((lun == NULL)
10058 	 || (lun->be_lun->lun_type != T_PROCESSOR))
10059 		inq_ptr->flags |= SID_CmdQue;
10060 	/*
10061 	 * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10062 	 * We have 8 bytes for the vendor name, and 16 bytes for the device
10063 	 * name and 4 bytes for the revision.
10064 	 */
10065 	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10066 	    "vendor")) == NULL) {
10067 		strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
10068 	} else {
10069 		memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10070 		strncpy(inq_ptr->vendor, val,
10071 		    min(sizeof(inq_ptr->vendor), strlen(val)));
10072 	}
10073 	if (lun == NULL) {
10074 		strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10075 		    sizeof(inq_ptr->product));
10076 	} else if ((val = ctl_get_opt(&lun->be_lun->options, "product")) == NULL) {
10077 		switch (lun->be_lun->lun_type) {
10078 		case T_DIRECT:
10079 			strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10080 			    sizeof(inq_ptr->product));
10081 			break;
10082 		case T_PROCESSOR:
10083 			strncpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT,
10084 			    sizeof(inq_ptr->product));
10085 			break;
10086 		default:
10087 			strncpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT,
10088 			    sizeof(inq_ptr->product));
10089 			break;
10090 		}
10091 	} else {
10092 		memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10093 		strncpy(inq_ptr->product, val,
10094 		    min(sizeof(inq_ptr->product), strlen(val)));
10095 	}
10096 
10097 	/*
10098 	 * XXX make this a macro somewhere so it automatically gets
10099 	 * incremented when we make changes.
10100 	 */
10101 	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10102 	    "revision")) == NULL) {
10103 		strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10104 	} else {
10105 		memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10106 		strncpy(inq_ptr->revision, val,
10107 		    min(sizeof(inq_ptr->revision), strlen(val)));
10108 	}
10109 
10110 	/*
10111 	 * For parallel SCSI, we support double transition and single
10112 	 * transition clocking.  We also support QAS (Quick Arbitration
10113 	 * and Selection) and Information Unit transfers on both the
10114 	 * control and array devices.
10115 	 */
10116 	if (port_type == CTL_PORT_SCSI)
10117 		inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10118 				    SID_SPI_IUS;
10119 
10120 	/* SAM-5 (no version claimed) */
10121 	scsi_ulto2b(0x00A0, inq_ptr->version1);
10122 	/* SPC-4 (no version claimed) */
10123 	scsi_ulto2b(0x0460, inq_ptr->version2);
10124 	if (port_type == CTL_PORT_FC) {
10125 		/* FCP-2 ANSI INCITS.350:2003 */
10126 		scsi_ulto2b(0x0917, inq_ptr->version3);
10127 	} else if (port_type == CTL_PORT_SCSI) {
10128 		/* SPI-4 ANSI INCITS.362:200x */
10129 		scsi_ulto2b(0x0B56, inq_ptr->version3);
10130 	} else if (port_type == CTL_PORT_ISCSI) {
10131 		/* iSCSI (no version claimed) */
10132 		scsi_ulto2b(0x0960, inq_ptr->version3);
10133 	} else if (port_type == CTL_PORT_SAS) {
10134 		/* SAS (no version claimed) */
10135 		scsi_ulto2b(0x0BE0, inq_ptr->version3);
10136 	}
10137 
10138 	if (lun == NULL) {
10139 		/* SBC-4 (no version claimed) */
10140 		scsi_ulto2b(0x0600, inq_ptr->version4);
10141 	} else {
10142 		switch (lun->be_lun->lun_type) {
10143 		case T_DIRECT:
10144 			/* SBC-4 (no version claimed) */
10145 			scsi_ulto2b(0x0600, inq_ptr->version4);
10146 			break;
10147 		case T_PROCESSOR:
10148 		default:
10149 			break;
10150 		}
10151 	}
10152 
10153 	ctl_set_success(ctsio);
10154 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10155 	ctsio->be_move_done = ctl_config_move_done;
10156 	ctl_datamove((union ctl_io *)ctsio);
10157 	return (CTL_RETVAL_COMPLETE);
10158 }
10159 
10160 int
10161 ctl_inquiry(struct ctl_scsiio *ctsio)
10162 {
10163 	struct scsi_inquiry *cdb;
10164 	int retval;
10165 
10166 	CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10167 
10168 	cdb = (struct scsi_inquiry *)ctsio->cdb;
10169 	if (cdb->byte2 & SI_EVPD)
10170 		retval = ctl_inquiry_evpd(ctsio);
10171 	else if (cdb->page_code == 0)
10172 		retval = ctl_inquiry_std(ctsio);
10173 	else {
10174 		ctl_set_invalid_field(ctsio,
10175 				      /*sks_valid*/ 1,
10176 				      /*command*/ 1,
10177 				      /*field*/ 2,
10178 				      /*bit_valid*/ 0,
10179 				      /*bit*/ 0);
10180 		ctl_done((union ctl_io *)ctsio);
10181 		return (CTL_RETVAL_COMPLETE);
10182 	}
10183 
10184 	return (retval);
10185 }
10186 
10187 /*
10188  * For known CDB types, parse the LBA and length.
10189  */
10190 static int
10191 ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len)
10192 {
10193 	if (io->io_hdr.io_type != CTL_IO_SCSI)
10194 		return (1);
10195 
10196 	switch (io->scsiio.cdb[0]) {
10197 	case COMPARE_AND_WRITE: {
10198 		struct scsi_compare_and_write *cdb;
10199 
10200 		cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10201 
10202 		*lba = scsi_8btou64(cdb->addr);
10203 		*len = cdb->length;
10204 		break;
10205 	}
10206 	case READ_6:
10207 	case WRITE_6: {
10208 		struct scsi_rw_6 *cdb;
10209 
10210 		cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10211 
10212 		*lba = scsi_3btoul(cdb->addr);
10213 		/* only 5 bits are valid in the most significant address byte */
10214 		*lba &= 0x1fffff;
10215 		*len = cdb->length;
10216 		break;
10217 	}
10218 	case READ_10:
10219 	case WRITE_10: {
10220 		struct scsi_rw_10 *cdb;
10221 
10222 		cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10223 
10224 		*lba = scsi_4btoul(cdb->addr);
10225 		*len = scsi_2btoul(cdb->length);
10226 		break;
10227 	}
10228 	case WRITE_VERIFY_10: {
10229 		struct scsi_write_verify_10 *cdb;
10230 
10231 		cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10232 
10233 		*lba = scsi_4btoul(cdb->addr);
10234 		*len = scsi_2btoul(cdb->length);
10235 		break;
10236 	}
10237 	case READ_12:
10238 	case WRITE_12: {
10239 		struct scsi_rw_12 *cdb;
10240 
10241 		cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10242 
10243 		*lba = scsi_4btoul(cdb->addr);
10244 		*len = scsi_4btoul(cdb->length);
10245 		break;
10246 	}
10247 	case WRITE_VERIFY_12: {
10248 		struct scsi_write_verify_12 *cdb;
10249 
10250 		cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10251 
10252 		*lba = scsi_4btoul(cdb->addr);
10253 		*len = scsi_4btoul(cdb->length);
10254 		break;
10255 	}
10256 	case READ_16:
10257 	case WRITE_16:
10258 	case WRITE_ATOMIC_16: {
10259 		struct scsi_rw_16 *cdb;
10260 
10261 		cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10262 
10263 		*lba = scsi_8btou64(cdb->addr);
10264 		*len = scsi_4btoul(cdb->length);
10265 		break;
10266 	}
10267 	case WRITE_VERIFY_16: {
10268 		struct scsi_write_verify_16 *cdb;
10269 
10270 		cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10271 
10272 		*lba = scsi_8btou64(cdb->addr);
10273 		*len = scsi_4btoul(cdb->length);
10274 		break;
10275 	}
10276 	case WRITE_SAME_10: {
10277 		struct scsi_write_same_10 *cdb;
10278 
10279 		cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10280 
10281 		*lba = scsi_4btoul(cdb->addr);
10282 		*len = scsi_2btoul(cdb->length);
10283 		break;
10284 	}
10285 	case WRITE_SAME_16: {
10286 		struct scsi_write_same_16 *cdb;
10287 
10288 		cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10289 
10290 		*lba = scsi_8btou64(cdb->addr);
10291 		*len = scsi_4btoul(cdb->length);
10292 		break;
10293 	}
10294 	case VERIFY_10: {
10295 		struct scsi_verify_10 *cdb;
10296 
10297 		cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10298 
10299 		*lba = scsi_4btoul(cdb->addr);
10300 		*len = scsi_2btoul(cdb->length);
10301 		break;
10302 	}
10303 	case VERIFY_12: {
10304 		struct scsi_verify_12 *cdb;
10305 
10306 		cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10307 
10308 		*lba = scsi_4btoul(cdb->addr);
10309 		*len = scsi_4btoul(cdb->length);
10310 		break;
10311 	}
10312 	case VERIFY_16: {
10313 		struct scsi_verify_16 *cdb;
10314 
10315 		cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10316 
10317 		*lba = scsi_8btou64(cdb->addr);
10318 		*len = scsi_4btoul(cdb->length);
10319 		break;
10320 	}
10321 	case UNMAP: {
10322 		*lba = 0;
10323 		*len = UINT64_MAX;
10324 		break;
10325 	}
10326 	case SERVICE_ACTION_IN: {	/* GET LBA STATUS */
10327 		struct scsi_get_lba_status *cdb;
10328 
10329 		cdb = (struct scsi_get_lba_status *)io->scsiio.cdb;
10330 		*lba = scsi_8btou64(cdb->addr);
10331 		*len = UINT32_MAX;
10332 		break;
10333 	}
10334 	default:
10335 		return (1);
10336 		break; /* NOTREACHED */
10337 	}
10338 
10339 	return (0);
10340 }
10341 
10342 static ctl_action
10343 ctl_extent_check_lba(uint64_t lba1, uint64_t len1, uint64_t lba2, uint64_t len2,
10344     bool seq)
10345 {
10346 	uint64_t endlba1, endlba2;
10347 
10348 	endlba1 = lba1 + len1 - (seq ? 0 : 1);
10349 	endlba2 = lba2 + len2 - 1;
10350 
10351 	if ((endlba1 < lba2) || (endlba2 < lba1))
10352 		return (CTL_ACTION_PASS);
10353 	else
10354 		return (CTL_ACTION_BLOCK);
10355 }
10356 
10357 static int
10358 ctl_extent_check_unmap(union ctl_io *io, uint64_t lba2, uint64_t len2)
10359 {
10360 	struct ctl_ptr_len_flags *ptrlen;
10361 	struct scsi_unmap_desc *buf, *end, *range;
10362 	uint64_t lba;
10363 	uint32_t len;
10364 
10365 	/* If not UNMAP -- go other way. */
10366 	if (io->io_hdr.io_type != CTL_IO_SCSI ||
10367 	    io->scsiio.cdb[0] != UNMAP)
10368 		return (CTL_ACTION_ERROR);
10369 
10370 	/* If UNMAP without data -- block and wait for data. */
10371 	ptrlen = (struct ctl_ptr_len_flags *)
10372 	    &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
10373 	if ((io->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0 ||
10374 	    ptrlen->ptr == NULL)
10375 		return (CTL_ACTION_BLOCK);
10376 
10377 	/* UNMAP with data -- check for collision. */
10378 	buf = (struct scsi_unmap_desc *)ptrlen->ptr;
10379 	end = buf + ptrlen->len / sizeof(*buf);
10380 	for (range = buf; range < end; range++) {
10381 		lba = scsi_8btou64(range->lba);
10382 		len = scsi_4btoul(range->length);
10383 		if ((lba < lba2 + len2) && (lba + len > lba2))
10384 			return (CTL_ACTION_BLOCK);
10385 	}
10386 	return (CTL_ACTION_PASS);
10387 }
10388 
10389 static ctl_action
10390 ctl_extent_check(union ctl_io *io1, union ctl_io *io2, bool seq)
10391 {
10392 	uint64_t lba1, lba2;
10393 	uint64_t len1, len2;
10394 	int retval;
10395 
10396 	if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10397 		return (CTL_ACTION_ERROR);
10398 
10399 	retval = ctl_extent_check_unmap(io1, lba2, len2);
10400 	if (retval != CTL_ACTION_ERROR)
10401 		return (retval);
10402 
10403 	if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10404 		return (CTL_ACTION_ERROR);
10405 
10406 	return (ctl_extent_check_lba(lba1, len1, lba2, len2, seq));
10407 }
10408 
10409 static ctl_action
10410 ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2)
10411 {
10412 	uint64_t lba1, lba2;
10413 	uint64_t len1, len2;
10414 
10415 	if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10416 		return (CTL_ACTION_ERROR);
10417 	if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10418 		return (CTL_ACTION_ERROR);
10419 
10420 	if (lba1 + len1 == lba2)
10421 		return (CTL_ACTION_BLOCK);
10422 	return (CTL_ACTION_PASS);
10423 }
10424 
10425 static ctl_action
10426 ctl_check_for_blockage(struct ctl_lun *lun, union ctl_io *pending_io,
10427     union ctl_io *ooa_io)
10428 {
10429 	const struct ctl_cmd_entry *pending_entry, *ooa_entry;
10430 	ctl_serialize_action *serialize_row;
10431 
10432 	/*
10433 	 * The initiator attempted multiple untagged commands at the same
10434 	 * time.  Can't do that.
10435 	 */
10436 	if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10437 	 && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10438 	 && ((pending_io->io_hdr.nexus.targ_port ==
10439 	      ooa_io->io_hdr.nexus.targ_port)
10440 	  && (pending_io->io_hdr.nexus.initid.id ==
10441 	      ooa_io->io_hdr.nexus.initid.id))
10442 	 && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
10443 	      CTL_FLAG_STATUS_SENT)) == 0))
10444 		return (CTL_ACTION_OVERLAP);
10445 
10446 	/*
10447 	 * The initiator attempted to send multiple tagged commands with
10448 	 * the same ID.  (It's fine if different initiators have the same
10449 	 * tag ID.)
10450 	 *
10451 	 * Even if all of those conditions are true, we don't kill the I/O
10452 	 * if the command ahead of us has been aborted.  We won't end up
10453 	 * sending it to the FETD, and it's perfectly legal to resend a
10454 	 * command with the same tag number as long as the previous
10455 	 * instance of this tag number has been aborted somehow.
10456 	 */
10457 	if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10458 	 && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10459 	 && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
10460 	 && ((pending_io->io_hdr.nexus.targ_port ==
10461 	      ooa_io->io_hdr.nexus.targ_port)
10462 	  && (pending_io->io_hdr.nexus.initid.id ==
10463 	      ooa_io->io_hdr.nexus.initid.id))
10464 	 && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
10465 	      CTL_FLAG_STATUS_SENT)) == 0))
10466 		return (CTL_ACTION_OVERLAP_TAG);
10467 
10468 	/*
10469 	 * If we get a head of queue tag, SAM-3 says that we should
10470 	 * immediately execute it.
10471 	 *
10472 	 * What happens if this command would normally block for some other
10473 	 * reason?  e.g. a request sense with a head of queue tag
10474 	 * immediately after a write.  Normally that would block, but this
10475 	 * will result in its getting executed immediately...
10476 	 *
10477 	 * We currently return "pass" instead of "skip", so we'll end up
10478 	 * going through the rest of the queue to check for overlapped tags.
10479 	 *
10480 	 * XXX KDM check for other types of blockage first??
10481 	 */
10482 	if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10483 		return (CTL_ACTION_PASS);
10484 
10485 	/*
10486 	 * Ordered tags have to block until all items ahead of them
10487 	 * have completed.  If we get called with an ordered tag, we always
10488 	 * block, if something else is ahead of us in the queue.
10489 	 */
10490 	if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
10491 		return (CTL_ACTION_BLOCK);
10492 
10493 	/*
10494 	 * Simple tags get blocked until all head of queue and ordered tags
10495 	 * ahead of them have completed.  I'm lumping untagged commands in
10496 	 * with simple tags here.  XXX KDM is that the right thing to do?
10497 	 */
10498 	if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10499 	  || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
10500 	 && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10501 	  || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
10502 		return (CTL_ACTION_BLOCK);
10503 
10504 	pending_entry = ctl_get_cmd_entry(&pending_io->scsiio, NULL);
10505 	ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio, NULL);
10506 
10507 	serialize_row = ctl_serialize_table[ooa_entry->seridx];
10508 
10509 	switch (serialize_row[pending_entry->seridx]) {
10510 	case CTL_SER_BLOCK:
10511 		return (CTL_ACTION_BLOCK);
10512 	case CTL_SER_EXTENT:
10513 		return (ctl_extent_check(ooa_io, pending_io,
10514 		    (lun->serseq == CTL_LUN_SERSEQ_ON)));
10515 	case CTL_SER_EXTENTOPT:
10516 		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
10517 		    & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
10518 			return (ctl_extent_check(ooa_io, pending_io,
10519 			    (lun->serseq == CTL_LUN_SERSEQ_ON)));
10520 		return (CTL_ACTION_PASS);
10521 	case CTL_SER_EXTENTSEQ:
10522 		if (lun->serseq != CTL_LUN_SERSEQ_OFF)
10523 			return (ctl_extent_check_seq(ooa_io, pending_io));
10524 		return (CTL_ACTION_PASS);
10525 	case CTL_SER_PASS:
10526 		return (CTL_ACTION_PASS);
10527 	case CTL_SER_BLOCKOPT:
10528 		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
10529 		    & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
10530 			return (CTL_ACTION_BLOCK);
10531 		return (CTL_ACTION_PASS);
10532 	case CTL_SER_SKIP:
10533 		return (CTL_ACTION_SKIP);
10534 	default:
10535 		panic("invalid serialization value %d",
10536 		      serialize_row[pending_entry->seridx]);
10537 	}
10538 
10539 	return (CTL_ACTION_ERROR);
10540 }
10541 
10542 /*
10543  * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
10544  * Assumptions:
10545  * - pending_io is generally either incoming, or on the blocked queue
10546  * - starting I/O is the I/O we want to start the check with.
10547  */
10548 static ctl_action
10549 ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
10550 	      union ctl_io *starting_io)
10551 {
10552 	union ctl_io *ooa_io;
10553 	ctl_action action;
10554 
10555 	mtx_assert(&lun->lun_lock, MA_OWNED);
10556 
10557 	/*
10558 	 * Run back along the OOA queue, starting with the current
10559 	 * blocked I/O and going through every I/O before it on the
10560 	 * queue.  If starting_io is NULL, we'll just end up returning
10561 	 * CTL_ACTION_PASS.
10562 	 */
10563 	for (ooa_io = starting_io; ooa_io != NULL;
10564 	     ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
10565 	     ooa_links)){
10566 
10567 		/*
10568 		 * This routine just checks to see whether
10569 		 * cur_blocked is blocked by ooa_io, which is ahead
10570 		 * of it in the queue.  It doesn't queue/dequeue
10571 		 * cur_blocked.
10572 		 */
10573 		action = ctl_check_for_blockage(lun, pending_io, ooa_io);
10574 		switch (action) {
10575 		case CTL_ACTION_BLOCK:
10576 		case CTL_ACTION_OVERLAP:
10577 		case CTL_ACTION_OVERLAP_TAG:
10578 		case CTL_ACTION_SKIP:
10579 		case CTL_ACTION_ERROR:
10580 			return (action);
10581 			break; /* NOTREACHED */
10582 		case CTL_ACTION_PASS:
10583 			break;
10584 		default:
10585 			panic("invalid action %d", action);
10586 			break;  /* NOTREACHED */
10587 		}
10588 	}
10589 
10590 	return (CTL_ACTION_PASS);
10591 }
10592 
10593 /*
10594  * Assumptions:
10595  * - An I/O has just completed, and has been removed from the per-LUN OOA
10596  *   queue, so some items on the blocked queue may now be unblocked.
10597  */
10598 static int
10599 ctl_check_blocked(struct ctl_lun *lun)
10600 {
10601 	union ctl_io *cur_blocked, *next_blocked;
10602 
10603 	mtx_assert(&lun->lun_lock, MA_OWNED);
10604 
10605 	/*
10606 	 * Run forward from the head of the blocked queue, checking each
10607 	 * entry against the I/Os prior to it on the OOA queue to see if
10608 	 * there is still any blockage.
10609 	 *
10610 	 * We cannot use the TAILQ_FOREACH() macro, because it can't deal
10611 	 * with our removing a variable on it while it is traversing the
10612 	 * list.
10613 	 */
10614 	for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
10615 	     cur_blocked != NULL; cur_blocked = next_blocked) {
10616 		union ctl_io *prev_ooa;
10617 		ctl_action action;
10618 
10619 		next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
10620 							  blocked_links);
10621 
10622 		prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
10623 						      ctl_ooaq, ooa_links);
10624 
10625 		/*
10626 		 * If cur_blocked happens to be the first item in the OOA
10627 		 * queue now, prev_ooa will be NULL, and the action
10628 		 * returned will just be CTL_ACTION_PASS.
10629 		 */
10630 		action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
10631 
10632 		switch (action) {
10633 		case CTL_ACTION_BLOCK:
10634 			/* Nothing to do here, still blocked */
10635 			break;
10636 		case CTL_ACTION_OVERLAP:
10637 		case CTL_ACTION_OVERLAP_TAG:
10638 			/*
10639 			 * This shouldn't happen!  In theory we've already
10640 			 * checked this command for overlap...
10641 			 */
10642 			break;
10643 		case CTL_ACTION_PASS:
10644 		case CTL_ACTION_SKIP: {
10645 			const struct ctl_cmd_entry *entry;
10646 			int isc_retval;
10647 
10648 			/*
10649 			 * The skip case shouldn't happen, this transaction
10650 			 * should have never made it onto the blocked queue.
10651 			 */
10652 			/*
10653 			 * This I/O is no longer blocked, we can remove it
10654 			 * from the blocked queue.  Since this is a TAILQ
10655 			 * (doubly linked list), we can do O(1) removals
10656 			 * from any place on the list.
10657 			 */
10658 			TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
10659 				     blocked_links);
10660 			cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
10661 
10662 			if (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC){
10663 				/*
10664 				 * Need to send IO back to original side to
10665 				 * run
10666 				 */
10667 				union ctl_ha_msg msg_info;
10668 
10669 				msg_info.hdr.original_sc =
10670 					cur_blocked->io_hdr.original_sc;
10671 				msg_info.hdr.serializing_sc = cur_blocked;
10672 				msg_info.hdr.msg_type = CTL_MSG_R2R;
10673 				if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
10674 				     &msg_info, sizeof(msg_info), 0)) >
10675 				     CTL_HA_STATUS_SUCCESS) {
10676 					printf("CTL:Check Blocked error from "
10677 					       "ctl_ha_msg_send %d\n",
10678 					       isc_retval);
10679 				}
10680 				break;
10681 			}
10682 			entry = ctl_get_cmd_entry(&cur_blocked->scsiio, NULL);
10683 
10684 			/*
10685 			 * Check this I/O for LUN state changes that may
10686 			 * have happened while this command was blocked.
10687 			 * The LUN state may have been changed by a command
10688 			 * ahead of us in the queue, so we need to re-check
10689 			 * for any states that can be caused by SCSI
10690 			 * commands.
10691 			 */
10692 			if (ctl_scsiio_lun_check(lun, entry,
10693 						 &cur_blocked->scsiio) == 0) {
10694 				cur_blocked->io_hdr.flags |=
10695 				                      CTL_FLAG_IS_WAS_ON_RTR;
10696 				ctl_enqueue_rtr(cur_blocked);
10697 			} else
10698 				ctl_done(cur_blocked);
10699 			break;
10700 		}
10701 		default:
10702 			/*
10703 			 * This probably shouldn't happen -- we shouldn't
10704 			 * get CTL_ACTION_ERROR, or anything else.
10705 			 */
10706 			break;
10707 		}
10708 	}
10709 
10710 	return (CTL_RETVAL_COMPLETE);
10711 }
10712 
10713 /*
10714  * This routine (with one exception) checks LUN flags that can be set by
10715  * commands ahead of us in the OOA queue.  These flags have to be checked
10716  * when a command initially comes in, and when we pull a command off the
10717  * blocked queue and are preparing to execute it.  The reason we have to
10718  * check these flags for commands on the blocked queue is that the LUN
10719  * state may have been changed by a command ahead of us while we're on the
10720  * blocked queue.
10721  *
10722  * Ordering is somewhat important with these checks, so please pay
10723  * careful attention to the placement of any new checks.
10724  */
10725 static int
10726 ctl_scsiio_lun_check(struct ctl_lun *lun,
10727     const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
10728 {
10729 	struct ctl_softc *softc = lun->ctl_softc;
10730 	int retval;
10731 	uint32_t residx;
10732 
10733 	retval = 0;
10734 
10735 	mtx_assert(&lun->lun_lock, MA_OWNED);
10736 
10737 	/*
10738 	 * If this shelf is a secondary shelf controller, we have to reject
10739 	 * any media access commands.
10740 	 */
10741 	if ((softc->flags & CTL_FLAG_ACTIVE_SHELF) == 0 &&
10742 	    (entry->flags & CTL_CMD_FLAG_OK_ON_SECONDARY) == 0) {
10743 		ctl_set_lun_standby(ctsio);
10744 		retval = 1;
10745 		goto bailout;
10746 	}
10747 
10748 	if (entry->pattern & CTL_LUN_PAT_WRITE) {
10749 		if (lun->flags & CTL_LUN_READONLY) {
10750 			ctl_set_sense(ctsio, /*current_error*/ 1,
10751 			    /*sense_key*/ SSD_KEY_DATA_PROTECT,
10752 			    /*asc*/ 0x27, /*ascq*/ 0x01, SSD_ELEM_NONE);
10753 			retval = 1;
10754 			goto bailout;
10755 		}
10756 		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT]
10757 		    .eca_and_aen & SCP_SWP) != 0) {
10758 			ctl_set_sense(ctsio, /*current_error*/ 1,
10759 			    /*sense_key*/ SSD_KEY_DATA_PROTECT,
10760 			    /*asc*/ 0x27, /*ascq*/ 0x02, SSD_ELEM_NONE);
10761 			retval = 1;
10762 			goto bailout;
10763 		}
10764 	}
10765 
10766 	/*
10767 	 * Check for a reservation conflict.  If this command isn't allowed
10768 	 * even on reserved LUNs, and if this initiator isn't the one who
10769 	 * reserved us, reject the command with a reservation conflict.
10770 	 */
10771 	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
10772 	if ((lun->flags & CTL_LUN_RESERVED)
10773 	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
10774 		if (lun->res_idx != residx) {
10775 			ctl_set_reservation_conflict(ctsio);
10776 			retval = 1;
10777 			goto bailout;
10778 		}
10779 	}
10780 
10781 	if ((lun->flags & CTL_LUN_PR_RESERVED) == 0 ||
10782 	    (entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV)) {
10783 		/* No reservation or command is allowed. */;
10784 	} else if ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_WRESV) &&
10785 	    (lun->res_type == SPR_TYPE_WR_EX ||
10786 	     lun->res_type == SPR_TYPE_WR_EX_RO ||
10787 	     lun->res_type == SPR_TYPE_WR_EX_AR)) {
10788 		/* The command is allowed for Write Exclusive resv. */;
10789 	} else {
10790 		/*
10791 		 * if we aren't registered or it's a res holder type
10792 		 * reservation and this isn't the res holder then set a
10793 		 * conflict.
10794 		 */
10795 		if (ctl_get_prkey(lun, residx) == 0
10796 		 || (residx != lun->pr_res_idx && lun->res_type < 4)) {
10797 			ctl_set_reservation_conflict(ctsio);
10798 			retval = 1;
10799 			goto bailout;
10800 		}
10801 
10802 	}
10803 
10804 	if ((lun->flags & CTL_LUN_OFFLINE)
10805 	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_OFFLINE) == 0)) {
10806 		ctl_set_lun_not_ready(ctsio);
10807 		retval = 1;
10808 		goto bailout;
10809 	}
10810 
10811 	/*
10812 	 * If the LUN is stopped, see if this particular command is allowed
10813 	 * for a stopped lun.  Otherwise, reject it with 0x04,0x02.
10814 	 */
10815 	if ((lun->flags & CTL_LUN_STOPPED)
10816 	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_STOPPED) == 0)) {
10817 		/* "Logical unit not ready, initializing cmd. required" */
10818 		ctl_set_lun_stopped(ctsio);
10819 		retval = 1;
10820 		goto bailout;
10821 	}
10822 
10823 	if ((lun->flags & CTL_LUN_INOPERABLE)
10824 	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_INOPERABLE) == 0)) {
10825 		/* "Medium format corrupted" */
10826 		ctl_set_medium_format_corrupted(ctsio);
10827 		retval = 1;
10828 		goto bailout;
10829 	}
10830 
10831 bailout:
10832 	return (retval);
10833 
10834 }
10835 
10836 static void
10837 ctl_failover_io(union ctl_io *io, int have_lock)
10838 {
10839 	ctl_set_busy(&io->scsiio);
10840 	ctl_done(io);
10841 }
10842 
10843 #ifdef notyet
10844 static void
10845 ctl_failover(void)
10846 {
10847 	struct ctl_lun *lun;
10848 	struct ctl_softc *softc;
10849 	union ctl_io *next_io, *pending_io;
10850 	union ctl_io *io;
10851 	int lun_idx;
10852 
10853 	softc = control_softc;
10854 
10855 	mtx_lock(&softc->ctl_lock);
10856 	/*
10857 	 * Remove any cmds from the other SC from the rtr queue.  These
10858 	 * will obviously only be for LUNs for which we're the primary.
10859 	 * We can't send status or get/send data for these commands.
10860 	 * Since they haven't been executed yet, we can just remove them.
10861 	 * We'll either abort them or delete them below, depending on
10862 	 * which HA mode we're in.
10863 	 */
10864 #ifdef notyet
10865 	mtx_lock(&softc->queue_lock);
10866 	for (io = (union ctl_io *)STAILQ_FIRST(&softc->rtr_queue);
10867 	     io != NULL; io = next_io) {
10868 		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
10869 		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
10870 			STAILQ_REMOVE(&softc->rtr_queue, &io->io_hdr,
10871 				      ctl_io_hdr, links);
10872 	}
10873 	mtx_unlock(&softc->queue_lock);
10874 #endif
10875 
10876 	for (lun_idx=0; lun_idx < softc->num_luns; lun_idx++) {
10877 		lun = softc->ctl_luns[lun_idx];
10878 		if (lun==NULL)
10879 			continue;
10880 
10881 		/*
10882 		 * Processor LUNs are primary on both sides.
10883 		 * XXX will this always be true?
10884 		 */
10885 		if (lun->be_lun->lun_type == T_PROCESSOR)
10886 			continue;
10887 
10888 		if ((lun->flags & CTL_LUN_PRIMARY_SC)
10889 		 && (softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
10890 			printf("FAILOVER: primary lun %d\n", lun_idx);
10891 		        /*
10892 			 * Remove all commands from the other SC. First from the
10893 			 * blocked queue then from the ooa queue. Once we have
10894 			 * removed them. Call ctl_check_blocked to see if there
10895 			 * is anything that can run.
10896 			 */
10897 			for (io = (union ctl_io *)TAILQ_FIRST(
10898 			     &lun->blocked_queue); io != NULL; io = next_io) {
10899 
10900 		        	next_io = (union ctl_io *)TAILQ_NEXT(
10901 				    &io->io_hdr, blocked_links);
10902 
10903 				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
10904 					TAILQ_REMOVE(&lun->blocked_queue,
10905 						     &io->io_hdr,blocked_links);
10906 					io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
10907 					TAILQ_REMOVE(&lun->ooa_queue,
10908 						     &io->io_hdr, ooa_links);
10909 
10910 					ctl_free_io(io);
10911 				}
10912 			}
10913 
10914 			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
10915 	     		     io != NULL; io = next_io) {
10916 
10917 		        	next_io = (union ctl_io *)TAILQ_NEXT(
10918 				    &io->io_hdr, ooa_links);
10919 
10920 				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
10921 
10922 					TAILQ_REMOVE(&lun->ooa_queue,
10923 						&io->io_hdr,
10924 					     	ooa_links);
10925 
10926 					ctl_free_io(io);
10927 				}
10928 			}
10929 			ctl_check_blocked(lun);
10930 		} else if ((lun->flags & CTL_LUN_PRIMARY_SC)
10931 			&& (softc->ha_mode == CTL_HA_MODE_XFER)) {
10932 
10933 			printf("FAILOVER: primary lun %d\n", lun_idx);
10934 			/*
10935 			 * Abort all commands from the other SC.  We can't
10936 			 * send status back for them now.  These should get
10937 			 * cleaned up when they are completed or come out
10938 			 * for a datamove operation.
10939 			 */
10940 			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
10941 	     		     io != NULL; io = next_io) {
10942 		        	next_io = (union ctl_io *)TAILQ_NEXT(
10943 					&io->io_hdr, ooa_links);
10944 
10945 				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
10946 					io->io_hdr.flags |= CTL_FLAG_ABORT;
10947 			}
10948 		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
10949 			&& (softc->ha_mode == CTL_HA_MODE_XFER)) {
10950 
10951 			printf("FAILOVER: secondary lun %d\n", lun_idx);
10952 
10953 			lun->flags |= CTL_LUN_PRIMARY_SC;
10954 
10955 			/*
10956 			 * We send all I/O that was sent to this controller
10957 			 * and redirected to the other side back with
10958 			 * busy status, and have the initiator retry it.
10959 			 * Figuring out how much data has been transferred,
10960 			 * etc. and picking up where we left off would be
10961 			 * very tricky.
10962 			 *
10963 			 * XXX KDM need to remove I/O from the blocked
10964 			 * queue as well!
10965 			 */
10966 			for (pending_io = (union ctl_io *)TAILQ_FIRST(
10967 			     &lun->ooa_queue); pending_io != NULL;
10968 			     pending_io = next_io) {
10969 
10970 				next_io =  (union ctl_io *)TAILQ_NEXT(
10971 					&pending_io->io_hdr, ooa_links);
10972 
10973 				pending_io->io_hdr.flags &=
10974 					~CTL_FLAG_SENT_2OTHER_SC;
10975 
10976 				if (pending_io->io_hdr.flags &
10977 				    CTL_FLAG_IO_ACTIVE) {
10978 					pending_io->io_hdr.flags |=
10979 						CTL_FLAG_FAILOVER;
10980 				} else {
10981 					ctl_set_busy(&pending_io->scsiio);
10982 					ctl_done(pending_io);
10983 				}
10984 			}
10985 
10986 			ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
10987 		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
10988 			&& (softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
10989 			printf("FAILOVER: secondary lun %d\n", lun_idx);
10990 			/*
10991 			 * if the first io on the OOA is not on the RtR queue
10992 			 * add it.
10993 			 */
10994 			lun->flags |= CTL_LUN_PRIMARY_SC;
10995 
10996 			pending_io = (union ctl_io *)TAILQ_FIRST(
10997 			    &lun->ooa_queue);
10998 			if (pending_io==NULL) {
10999 				printf("Nothing on OOA queue\n");
11000 				continue;
11001 			}
11002 
11003 			pending_io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11004 			if ((pending_io->io_hdr.flags &
11005 			     CTL_FLAG_IS_WAS_ON_RTR) == 0) {
11006 				pending_io->io_hdr.flags |=
11007 				    CTL_FLAG_IS_WAS_ON_RTR;
11008 				ctl_enqueue_rtr(pending_io);
11009 			}
11010 #if 0
11011 			else
11012 			{
11013 				printf("Tag 0x%04x is running\n",
11014 				      pending_io->scsiio.tag_num);
11015 			}
11016 #endif
11017 
11018 			next_io = (union ctl_io *)TAILQ_NEXT(
11019 			    &pending_io->io_hdr, ooa_links);
11020 			for (pending_io=next_io; pending_io != NULL;
11021 			     pending_io = next_io) {
11022 				pending_io->io_hdr.flags &=
11023 				    ~CTL_FLAG_SENT_2OTHER_SC;
11024 				next_io = (union ctl_io *)TAILQ_NEXT(
11025 					&pending_io->io_hdr, ooa_links);
11026 				if (pending_io->io_hdr.flags &
11027 				    CTL_FLAG_IS_WAS_ON_RTR) {
11028 #if 0
11029 				        printf("Tag 0x%04x is running\n",
11030 				      		pending_io->scsiio.tag_num);
11031 #endif
11032 					continue;
11033 				}
11034 
11035 				switch (ctl_check_ooa(lun, pending_io,
11036 			            (union ctl_io *)TAILQ_PREV(
11037 				    &pending_io->io_hdr, ctl_ooaq,
11038 				    ooa_links))) {
11039 
11040 				case CTL_ACTION_BLOCK:
11041 					TAILQ_INSERT_TAIL(&lun->blocked_queue,
11042 							  &pending_io->io_hdr,
11043 							  blocked_links);
11044 					pending_io->io_hdr.flags |=
11045 					    CTL_FLAG_BLOCKED;
11046 					break;
11047 				case CTL_ACTION_PASS:
11048 				case CTL_ACTION_SKIP:
11049 					pending_io->io_hdr.flags |=
11050 					    CTL_FLAG_IS_WAS_ON_RTR;
11051 					ctl_enqueue_rtr(pending_io);
11052 					break;
11053 				case CTL_ACTION_OVERLAP:
11054 					ctl_set_overlapped_cmd(
11055 					    (struct ctl_scsiio *)pending_io);
11056 					ctl_done(pending_io);
11057 					break;
11058 				case CTL_ACTION_OVERLAP_TAG:
11059 					ctl_set_overlapped_tag(
11060 					    (struct ctl_scsiio *)pending_io,
11061 					    pending_io->scsiio.tag_num & 0xff);
11062 					ctl_done(pending_io);
11063 					break;
11064 				case CTL_ACTION_ERROR:
11065 				default:
11066 					ctl_set_internal_failure(
11067 						(struct ctl_scsiio *)pending_io,
11068 						0,  // sks_valid
11069 						0); //retry count
11070 					ctl_done(pending_io);
11071 					break;
11072 				}
11073 			}
11074 
11075 			ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
11076 		} else {
11077 			panic("Unhandled HA mode failover, LUN flags = %#x, "
11078 			      "ha_mode = #%x", lun->flags, softc->ha_mode);
11079 		}
11080 	}
11081 	ctl_pause_rtr = 0;
11082 	mtx_unlock(&softc->ctl_lock);
11083 }
11084 #endif
11085 
11086 static void
11087 ctl_clear_ua(struct ctl_softc *ctl_softc, uint32_t initidx,
11088 	     ctl_ua_type ua_type)
11089 {
11090 	struct ctl_lun *lun;
11091 	ctl_ua_type *pu;
11092 
11093 	mtx_assert(&ctl_softc->ctl_lock, MA_OWNED);
11094 
11095 	STAILQ_FOREACH(lun, &ctl_softc->lun_list, links) {
11096 		mtx_lock(&lun->lun_lock);
11097 		pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
11098 		if (pu != NULL)
11099 			pu[initidx % CTL_MAX_INIT_PER_PORT] &= ~ua_type;
11100 		mtx_unlock(&lun->lun_lock);
11101 	}
11102 }
11103 
11104 static int
11105 ctl_scsiio_precheck(struct ctl_softc *softc, struct ctl_scsiio *ctsio)
11106 {
11107 	struct ctl_lun *lun;
11108 	const struct ctl_cmd_entry *entry;
11109 	uint32_t initidx, targ_lun;
11110 	int retval;
11111 
11112 	retval = 0;
11113 
11114 	lun = NULL;
11115 
11116 	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11117 	if ((targ_lun < CTL_MAX_LUNS)
11118 	 && ((lun = softc->ctl_luns[targ_lun]) != NULL)) {
11119 		/*
11120 		 * If the LUN is invalid, pretend that it doesn't exist.
11121 		 * It will go away as soon as all pending I/O has been
11122 		 * completed.
11123 		 */
11124 		mtx_lock(&lun->lun_lock);
11125 		if (lun->flags & CTL_LUN_DISABLED) {
11126 			mtx_unlock(&lun->lun_lock);
11127 			lun = NULL;
11128 			ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11129 			ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11130 		} else {
11131 			ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
11132 			ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
11133 				lun->be_lun;
11134 			if (lun->be_lun->lun_type == T_PROCESSOR) {
11135 				ctsio->io_hdr.flags |= CTL_FLAG_CONTROL_DEV;
11136 			}
11137 
11138 			/*
11139 			 * Every I/O goes into the OOA queue for a
11140 			 * particular LUN, and stays there until completion.
11141 			 */
11142 #ifdef CTL_TIME_IO
11143 			if (TAILQ_EMPTY(&lun->ooa_queue)) {
11144 				lun->idle_time += getsbinuptime() -
11145 				    lun->last_busy;
11146 			}
11147 #endif
11148 			TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr,
11149 			    ooa_links);
11150 		}
11151 	} else {
11152 		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11153 		ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11154 	}
11155 
11156 	/* Get command entry and return error if it is unsuppotyed. */
11157 	entry = ctl_validate_command(ctsio);
11158 	if (entry == NULL) {
11159 		if (lun)
11160 			mtx_unlock(&lun->lun_lock);
11161 		return (retval);
11162 	}
11163 
11164 	ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11165 	ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11166 
11167 	/*
11168 	 * Check to see whether we can send this command to LUNs that don't
11169 	 * exist.  This should pretty much only be the case for inquiry
11170 	 * and request sense.  Further checks, below, really require having
11171 	 * a LUN, so we can't really check the command anymore.  Just put
11172 	 * it on the rtr queue.
11173 	 */
11174 	if (lun == NULL) {
11175 		if (entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) {
11176 			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11177 			ctl_enqueue_rtr((union ctl_io *)ctsio);
11178 			return (retval);
11179 		}
11180 
11181 		ctl_set_unsupported_lun(ctsio);
11182 		ctl_done((union ctl_io *)ctsio);
11183 		CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11184 		return (retval);
11185 	} else {
11186 		/*
11187 		 * Make sure we support this particular command on this LUN.
11188 		 * e.g., we don't support writes to the control LUN.
11189 		 */
11190 		if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11191 			mtx_unlock(&lun->lun_lock);
11192 			ctl_set_invalid_opcode(ctsio);
11193 			ctl_done((union ctl_io *)ctsio);
11194 			return (retval);
11195 		}
11196 	}
11197 
11198 	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11199 
11200 #ifdef CTL_WITH_CA
11201 	/*
11202 	 * If we've got a request sense, it'll clear the contingent
11203 	 * allegiance condition.  Otherwise, if we have a CA condition for
11204 	 * this initiator, clear it, because it sent down a command other
11205 	 * than request sense.
11206 	 */
11207 	if ((ctsio->cdb[0] != REQUEST_SENSE)
11208 	 && (ctl_is_set(lun->have_ca, initidx)))
11209 		ctl_clear_mask(lun->have_ca, initidx);
11210 #endif
11211 
11212 	/*
11213 	 * If the command has this flag set, it handles its own unit
11214 	 * attention reporting, we shouldn't do anything.  Otherwise we
11215 	 * check for any pending unit attentions, and send them back to the
11216 	 * initiator.  We only do this when a command initially comes in,
11217 	 * not when we pull it off the blocked queue.
11218 	 *
11219 	 * According to SAM-3, section 5.3.2, the order that things get
11220 	 * presented back to the host is basically unit attentions caused
11221 	 * by some sort of reset event, busy status, reservation conflicts
11222 	 * or task set full, and finally any other status.
11223 	 *
11224 	 * One issue here is that some of the unit attentions we report
11225 	 * don't fall into the "reset" category (e.g. "reported luns data
11226 	 * has changed").  So reporting it here, before the reservation
11227 	 * check, may be technically wrong.  I guess the only thing to do
11228 	 * would be to check for and report the reset events here, and then
11229 	 * check for the other unit attention types after we check for a
11230 	 * reservation conflict.
11231 	 *
11232 	 * XXX KDM need to fix this
11233 	 */
11234 	if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11235 		ctl_ua_type ua_type;
11236 		scsi_sense_data_type sense_format;
11237 
11238 		if (lun->flags & CTL_LUN_SENSE_DESC)
11239 			sense_format = SSD_TYPE_DESC;
11240 		else
11241 			sense_format = SSD_TYPE_FIXED;
11242 
11243 		ua_type = ctl_build_ua(lun, initidx, &ctsio->sense_data,
11244 		    sense_format);
11245 		if (ua_type != CTL_UA_NONE) {
11246 			mtx_unlock(&lun->lun_lock);
11247 			ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11248 			ctsio->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
11249 			ctsio->sense_len = SSD_FULL_SIZE;
11250 			ctl_done((union ctl_io *)ctsio);
11251 			return (retval);
11252 		}
11253 	}
11254 
11255 
11256 	if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
11257 		mtx_unlock(&lun->lun_lock);
11258 		ctl_done((union ctl_io *)ctsio);
11259 		return (retval);
11260 	}
11261 
11262 	/*
11263 	 * XXX CHD this is where we want to send IO to other side if
11264 	 * this LUN is secondary on this SC. We will need to make a copy
11265 	 * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11266 	 * the copy we send as FROM_OTHER.
11267 	 * We also need to stuff the address of the original IO so we can
11268 	 * find it easily. Something similar will need be done on the other
11269 	 * side so when we are done we can find the copy.
11270 	 */
11271 	if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11272 		union ctl_ha_msg msg_info;
11273 		int isc_retval;
11274 
11275 		ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11276 
11277 		msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11278 		msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11279 #if 0
11280 		printf("1. ctsio %p\n", ctsio);
11281 #endif
11282 		msg_info.hdr.serializing_sc = NULL;
11283 		msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11284 		msg_info.scsi.tag_num = ctsio->tag_num;
11285 		msg_info.scsi.tag_type = ctsio->tag_type;
11286 		memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11287 
11288 		ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11289 
11290 		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11291 		    (void *)&msg_info, sizeof(msg_info), 0)) >
11292 		    CTL_HA_STATUS_SUCCESS) {
11293 			printf("CTL:precheck, ctl_ha_msg_send returned %d\n",
11294 			       isc_retval);
11295 			printf("CTL:opcode is %x\n", ctsio->cdb[0]);
11296 		} else {
11297 #if 0
11298 			printf("CTL:Precheck sent msg, opcode is %x\n",opcode);
11299 #endif
11300 		}
11301 
11302 		/*
11303 		 * XXX KDM this I/O is off the incoming queue, but hasn't
11304 		 * been inserted on any other queue.  We may need to come
11305 		 * up with a holding queue while we wait for serialization
11306 		 * so that we have an idea of what we're waiting for from
11307 		 * the other side.
11308 		 */
11309 		mtx_unlock(&lun->lun_lock);
11310 		return (retval);
11311 	}
11312 
11313 	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11314 			      (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11315 			      ctl_ooaq, ooa_links))) {
11316 	case CTL_ACTION_BLOCK:
11317 		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11318 		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11319 				  blocked_links);
11320 		mtx_unlock(&lun->lun_lock);
11321 		return (retval);
11322 	case CTL_ACTION_PASS:
11323 	case CTL_ACTION_SKIP:
11324 		ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11325 		mtx_unlock(&lun->lun_lock);
11326 		ctl_enqueue_rtr((union ctl_io *)ctsio);
11327 		break;
11328 	case CTL_ACTION_OVERLAP:
11329 		mtx_unlock(&lun->lun_lock);
11330 		ctl_set_overlapped_cmd(ctsio);
11331 		ctl_done((union ctl_io *)ctsio);
11332 		break;
11333 	case CTL_ACTION_OVERLAP_TAG:
11334 		mtx_unlock(&lun->lun_lock);
11335 		ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11336 		ctl_done((union ctl_io *)ctsio);
11337 		break;
11338 	case CTL_ACTION_ERROR:
11339 	default:
11340 		mtx_unlock(&lun->lun_lock);
11341 		ctl_set_internal_failure(ctsio,
11342 					 /*sks_valid*/ 0,
11343 					 /*retry_count*/ 0);
11344 		ctl_done((union ctl_io *)ctsio);
11345 		break;
11346 	}
11347 	return (retval);
11348 }
11349 
11350 const struct ctl_cmd_entry *
11351 ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa)
11352 {
11353 	const struct ctl_cmd_entry *entry;
11354 	int service_action;
11355 
11356 	entry = &ctl_cmd_table[ctsio->cdb[0]];
11357 	if (sa)
11358 		*sa = ((entry->flags & CTL_CMD_FLAG_SA5) != 0);
11359 	if (entry->flags & CTL_CMD_FLAG_SA5) {
11360 		service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11361 		entry = &((const struct ctl_cmd_entry *)
11362 		    entry->execute)[service_action];
11363 	}
11364 	return (entry);
11365 }
11366 
11367 const struct ctl_cmd_entry *
11368 ctl_validate_command(struct ctl_scsiio *ctsio)
11369 {
11370 	const struct ctl_cmd_entry *entry;
11371 	int i, sa;
11372 	uint8_t diff;
11373 
11374 	entry = ctl_get_cmd_entry(ctsio, &sa);
11375 	if (entry->execute == NULL) {
11376 		if (sa)
11377 			ctl_set_invalid_field(ctsio,
11378 					      /*sks_valid*/ 1,
11379 					      /*command*/ 1,
11380 					      /*field*/ 1,
11381 					      /*bit_valid*/ 1,
11382 					      /*bit*/ 4);
11383 		else
11384 			ctl_set_invalid_opcode(ctsio);
11385 		ctl_done((union ctl_io *)ctsio);
11386 		return (NULL);
11387 	}
11388 	KASSERT(entry->length > 0,
11389 	    ("Not defined length for command 0x%02x/0x%02x",
11390 	     ctsio->cdb[0], ctsio->cdb[1]));
11391 	for (i = 1; i < entry->length; i++) {
11392 		diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11393 		if (diff == 0)
11394 			continue;
11395 		ctl_set_invalid_field(ctsio,
11396 				      /*sks_valid*/ 1,
11397 				      /*command*/ 1,
11398 				      /*field*/ i,
11399 				      /*bit_valid*/ 1,
11400 				      /*bit*/ fls(diff) - 1);
11401 		ctl_done((union ctl_io *)ctsio);
11402 		return (NULL);
11403 	}
11404 	return (entry);
11405 }
11406 
11407 static int
11408 ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11409 {
11410 
11411 	switch (lun_type) {
11412 	case T_PROCESSOR:
11413 		if (((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0) &&
11414 		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11415 			return (0);
11416 		break;
11417 	case T_DIRECT:
11418 		if (((entry->flags & CTL_CMD_FLAG_OK_ON_SLUN) == 0) &&
11419 		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11420 			return (0);
11421 		break;
11422 	default:
11423 		return (0);
11424 	}
11425 	return (1);
11426 }
11427 
11428 static int
11429 ctl_scsiio(struct ctl_scsiio *ctsio)
11430 {
11431 	int retval;
11432 	const struct ctl_cmd_entry *entry;
11433 
11434 	retval = CTL_RETVAL_COMPLETE;
11435 
11436 	CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11437 
11438 	entry = ctl_get_cmd_entry(ctsio, NULL);
11439 
11440 	/*
11441 	 * If this I/O has been aborted, just send it straight to
11442 	 * ctl_done() without executing it.
11443 	 */
11444 	if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11445 		ctl_done((union ctl_io *)ctsio);
11446 		goto bailout;
11447 	}
11448 
11449 	/*
11450 	 * All the checks should have been handled by ctl_scsiio_precheck().
11451 	 * We should be clear now to just execute the I/O.
11452 	 */
11453 	retval = entry->execute(ctsio);
11454 
11455 bailout:
11456 	return (retval);
11457 }
11458 
11459 /*
11460  * Since we only implement one target right now, a bus reset simply resets
11461  * our single target.
11462  */
11463 static int
11464 ctl_bus_reset(struct ctl_softc *softc, union ctl_io *io)
11465 {
11466 	return(ctl_target_reset(softc, io, CTL_UA_BUS_RESET));
11467 }
11468 
11469 static int
11470 ctl_target_reset(struct ctl_softc *softc, union ctl_io *io,
11471 		 ctl_ua_type ua_type)
11472 {
11473 	struct ctl_lun *lun;
11474 	int retval;
11475 
11476 	if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11477 		union ctl_ha_msg msg_info;
11478 
11479 		io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11480 		msg_info.hdr.nexus = io->io_hdr.nexus;
11481 		if (ua_type==CTL_UA_TARG_RESET)
11482 			msg_info.task.task_action = CTL_TASK_TARGET_RESET;
11483 		else
11484 			msg_info.task.task_action = CTL_TASK_BUS_RESET;
11485 		msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11486 		msg_info.hdr.original_sc = NULL;
11487 		msg_info.hdr.serializing_sc = NULL;
11488 		if (CTL_HA_STATUS_SUCCESS != ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11489 		    (void *)&msg_info, sizeof(msg_info), 0)) {
11490 		}
11491 	}
11492 	retval = 0;
11493 
11494 	mtx_lock(&softc->ctl_lock);
11495 	STAILQ_FOREACH(lun, &softc->lun_list, links)
11496 		retval += ctl_lun_reset(lun, io, ua_type);
11497 	mtx_unlock(&softc->ctl_lock);
11498 
11499 	return (retval);
11500 }
11501 
11502 /*
11503  * The LUN should always be set.  The I/O is optional, and is used to
11504  * distinguish between I/Os sent by this initiator, and by other
11505  * initiators.  We set unit attention for initiators other than this one.
11506  * SAM-3 is vague on this point.  It does say that a unit attention should
11507  * be established for other initiators when a LUN is reset (see section
11508  * 5.7.3), but it doesn't specifically say that the unit attention should
11509  * be established for this particular initiator when a LUN is reset.  Here
11510  * is the relevant text, from SAM-3 rev 8:
11511  *
11512  * 5.7.2 When a SCSI initiator port aborts its own tasks
11513  *
11514  * When a SCSI initiator port causes its own task(s) to be aborted, no
11515  * notification that the task(s) have been aborted shall be returned to
11516  * the SCSI initiator port other than the completion response for the
11517  * command or task management function action that caused the task(s) to
11518  * be aborted and notification(s) associated with related effects of the
11519  * action (e.g., a reset unit attention condition).
11520  *
11521  * XXX KDM for now, we're setting unit attention for all initiators.
11522  */
11523 static int
11524 ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
11525 {
11526 	union ctl_io *xio;
11527 #if 0
11528 	uint32_t initidx;
11529 #endif
11530 #ifdef CTL_WITH_CA
11531 	int i;
11532 #endif
11533 
11534 	mtx_lock(&lun->lun_lock);
11535 	/*
11536 	 * Run through the OOA queue and abort each I/O.
11537 	 */
11538 	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11539 	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11540 		xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
11541 	}
11542 
11543 	/*
11544 	 * This version sets unit attention for every
11545 	 */
11546 #if 0
11547 	initidx = ctl_get_initindex(&io->io_hdr.nexus);
11548 	ctl_est_ua_all(lun, initidx, ua_type);
11549 #else
11550 	ctl_est_ua_all(lun, -1, ua_type);
11551 #endif
11552 
11553 	/*
11554 	 * A reset (any kind, really) clears reservations established with
11555 	 * RESERVE/RELEASE.  It does not clear reservations established
11556 	 * with PERSISTENT RESERVE OUT, but we don't support that at the
11557 	 * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
11558 	 * reservations made with the RESERVE/RELEASE commands, because
11559 	 * those commands are obsolete in SPC-3.
11560 	 */
11561 	lun->flags &= ~CTL_LUN_RESERVED;
11562 
11563 #ifdef CTL_WITH_CA
11564 	for (i = 0; i < CTL_MAX_INITIATORS; i++)
11565 		ctl_clear_mask(lun->have_ca, i);
11566 #endif
11567 	mtx_unlock(&lun->lun_lock);
11568 
11569 	return (0);
11570 }
11571 
11572 static void
11573 ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
11574     int other_sc)
11575 {
11576 	union ctl_io *xio;
11577 
11578 	mtx_assert(&lun->lun_lock, MA_OWNED);
11579 
11580 	/*
11581 	 * Run through the OOA queue and attempt to find the given I/O.
11582 	 * The target port, initiator ID, tag type and tag number have to
11583 	 * match the values that we got from the initiator.  If we have an
11584 	 * untagged command to abort, simply abort the first untagged command
11585 	 * we come to.  We only allow one untagged command at a time of course.
11586 	 */
11587 	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11588 	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11589 
11590 		if ((targ_port == UINT32_MAX ||
11591 		     targ_port == xio->io_hdr.nexus.targ_port) &&
11592 		    (init_id == UINT32_MAX ||
11593 		     init_id == xio->io_hdr.nexus.initid.id)) {
11594 			if (targ_port != xio->io_hdr.nexus.targ_port ||
11595 			    init_id != xio->io_hdr.nexus.initid.id)
11596 				xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS;
11597 			xio->io_hdr.flags |= CTL_FLAG_ABORT;
11598 			if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11599 				union ctl_ha_msg msg_info;
11600 
11601 				msg_info.hdr.nexus = xio->io_hdr.nexus;
11602 				msg_info.task.task_action = CTL_TASK_ABORT_TASK;
11603 				msg_info.task.tag_num = xio->scsiio.tag_num;
11604 				msg_info.task.tag_type = xio->scsiio.tag_type;
11605 				msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11606 				msg_info.hdr.original_sc = NULL;
11607 				msg_info.hdr.serializing_sc = NULL;
11608 				ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11609 				    (void *)&msg_info, sizeof(msg_info), 0);
11610 			}
11611 		}
11612 	}
11613 }
11614 
11615 static int
11616 ctl_abort_task_set(union ctl_io *io)
11617 {
11618 	struct ctl_softc *softc = control_softc;
11619 	struct ctl_lun *lun;
11620 	uint32_t targ_lun;
11621 
11622 	/*
11623 	 * Look up the LUN.
11624 	 */
11625 	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11626 	mtx_lock(&softc->ctl_lock);
11627 	if ((targ_lun < CTL_MAX_LUNS) && (softc->ctl_luns[targ_lun] != NULL))
11628 		lun = softc->ctl_luns[targ_lun];
11629 	else {
11630 		mtx_unlock(&softc->ctl_lock);
11631 		return (1);
11632 	}
11633 
11634 	mtx_lock(&lun->lun_lock);
11635 	mtx_unlock(&softc->ctl_lock);
11636 	if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
11637 		ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
11638 		    io->io_hdr.nexus.initid.id,
11639 		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11640 	} else { /* CTL_TASK_CLEAR_TASK_SET */
11641 		ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
11642 		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11643 	}
11644 	mtx_unlock(&lun->lun_lock);
11645 	return (0);
11646 }
11647 
11648 static int
11649 ctl_i_t_nexus_reset(union ctl_io *io)
11650 {
11651 	struct ctl_softc *softc = control_softc;
11652 	struct ctl_lun *lun;
11653 	uint32_t initidx, residx;
11654 
11655 	initidx = ctl_get_initindex(&io->io_hdr.nexus);
11656 	residx = ctl_get_resindex(&io->io_hdr.nexus);
11657 	mtx_lock(&softc->ctl_lock);
11658 	STAILQ_FOREACH(lun, &softc->lun_list, links) {
11659 		mtx_lock(&lun->lun_lock);
11660 		ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
11661 		    io->io_hdr.nexus.initid.id,
11662 		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11663 #ifdef CTL_WITH_CA
11664 		ctl_clear_mask(lun->have_ca, initidx);
11665 #endif
11666 		if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
11667 			lun->flags &= ~CTL_LUN_RESERVED;
11668 		ctl_est_ua(lun, initidx, CTL_UA_I_T_NEXUS_LOSS);
11669 		mtx_unlock(&lun->lun_lock);
11670 	}
11671 	mtx_unlock(&softc->ctl_lock);
11672 	return (0);
11673 }
11674 
11675 static int
11676 ctl_abort_task(union ctl_io *io)
11677 {
11678 	union ctl_io *xio;
11679 	struct ctl_lun *lun;
11680 	struct ctl_softc *softc;
11681 #if 0
11682 	struct sbuf sb;
11683 	char printbuf[128];
11684 #endif
11685 	int found;
11686 	uint32_t targ_lun;
11687 
11688 	softc = control_softc;
11689 	found = 0;
11690 
11691 	/*
11692 	 * Look up the LUN.
11693 	 */
11694 	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11695 	mtx_lock(&softc->ctl_lock);
11696 	if ((targ_lun < CTL_MAX_LUNS)
11697 	 && (softc->ctl_luns[targ_lun] != NULL))
11698 		lun = softc->ctl_luns[targ_lun];
11699 	else {
11700 		mtx_unlock(&softc->ctl_lock);
11701 		return (1);
11702 	}
11703 
11704 #if 0
11705 	printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
11706 	       lun->lun, io->taskio.tag_num, io->taskio.tag_type);
11707 #endif
11708 
11709 	mtx_lock(&lun->lun_lock);
11710 	mtx_unlock(&softc->ctl_lock);
11711 	/*
11712 	 * Run through the OOA queue and attempt to find the given I/O.
11713 	 * The target port, initiator ID, tag type and tag number have to
11714 	 * match the values that we got from the initiator.  If we have an
11715 	 * untagged command to abort, simply abort the first untagged command
11716 	 * we come to.  We only allow one untagged command at a time of course.
11717 	 */
11718 	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11719 	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11720 #if 0
11721 		sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
11722 
11723 		sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
11724 			    lun->lun, xio->scsiio.tag_num,
11725 			    xio->scsiio.tag_type,
11726 			    (xio->io_hdr.blocked_links.tqe_prev
11727 			    == NULL) ? "" : " BLOCKED",
11728 			    (xio->io_hdr.flags &
11729 			    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
11730 			    (xio->io_hdr.flags &
11731 			    CTL_FLAG_ABORT) ? " ABORT" : "",
11732 			    (xio->io_hdr.flags &
11733 			    CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
11734 		ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
11735 		sbuf_finish(&sb);
11736 		printf("%s\n", sbuf_data(&sb));
11737 #endif
11738 
11739 		if ((xio->io_hdr.nexus.targ_port != io->io_hdr.nexus.targ_port)
11740 		 || (xio->io_hdr.nexus.initid.id != io->io_hdr.nexus.initid.id)
11741 		 || (xio->io_hdr.flags & CTL_FLAG_ABORT))
11742 			continue;
11743 
11744 		/*
11745 		 * If the abort says that the task is untagged, the
11746 		 * task in the queue must be untagged.  Otherwise,
11747 		 * we just check to see whether the tag numbers
11748 		 * match.  This is because the QLogic firmware
11749 		 * doesn't pass back the tag type in an abort
11750 		 * request.
11751 		 */
11752 #if 0
11753 		if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
11754 		  && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
11755 		 || (xio->scsiio.tag_num == io->taskio.tag_num))
11756 #endif
11757 		/*
11758 		 * XXX KDM we've got problems with FC, because it
11759 		 * doesn't send down a tag type with aborts.  So we
11760 		 * can only really go by the tag number...
11761 		 * This may cause problems with parallel SCSI.
11762 		 * Need to figure that out!!
11763 		 */
11764 		if (xio->scsiio.tag_num == io->taskio.tag_num) {
11765 			xio->io_hdr.flags |= CTL_FLAG_ABORT;
11766 			found = 1;
11767 			if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0 &&
11768 			    !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11769 				union ctl_ha_msg msg_info;
11770 
11771 				io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11772 				msg_info.hdr.nexus = io->io_hdr.nexus;
11773 				msg_info.task.task_action = CTL_TASK_ABORT_TASK;
11774 				msg_info.task.tag_num = io->taskio.tag_num;
11775 				msg_info.task.tag_type = io->taskio.tag_type;
11776 				msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11777 				msg_info.hdr.original_sc = NULL;
11778 				msg_info.hdr.serializing_sc = NULL;
11779 #if 0
11780 				printf("Sent Abort to other side\n");
11781 #endif
11782 				if (ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11783 				    (void *)&msg_info, sizeof(msg_info), 0) !=
11784 				    CTL_HA_STATUS_SUCCESS) {
11785 				}
11786 			}
11787 #if 0
11788 			printf("ctl_abort_task: found I/O to abort\n");
11789 #endif
11790 		}
11791 	}
11792 	mtx_unlock(&lun->lun_lock);
11793 
11794 	if (found == 0) {
11795 		/*
11796 		 * This isn't really an error.  It's entirely possible for
11797 		 * the abort and command completion to cross on the wire.
11798 		 * This is more of an informative/diagnostic error.
11799 		 */
11800 #if 0
11801 		printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
11802 		       "%d:%d:%d:%d tag %d type %d\n",
11803 		       io->io_hdr.nexus.initid.id,
11804 		       io->io_hdr.nexus.targ_port,
11805 		       io->io_hdr.nexus.targ_target.id,
11806 		       io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
11807 		       io->taskio.tag_type);
11808 #endif
11809 	}
11810 	return (0);
11811 }
11812 
11813 static void
11814 ctl_run_task(union ctl_io *io)
11815 {
11816 	struct ctl_softc *softc = control_softc;
11817 	int retval = 1;
11818 	const char *task_desc;
11819 
11820 	CTL_DEBUG_PRINT(("ctl_run_task\n"));
11821 
11822 	KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
11823 	    ("ctl_run_task: Unextected io_type %d\n",
11824 	     io->io_hdr.io_type));
11825 
11826 	task_desc = ctl_scsi_task_string(&io->taskio);
11827 	if (task_desc != NULL) {
11828 #ifdef NEEDTOPORT
11829 		csevent_log(CSC_CTL | CSC_SHELF_SW |
11830 			    CTL_TASK_REPORT,
11831 			    csevent_LogType_Trace,
11832 			    csevent_Severity_Information,
11833 			    csevent_AlertLevel_Green,
11834 			    csevent_FRU_Firmware,
11835 			    csevent_FRU_Unknown,
11836 			    "CTL: received task: %s",task_desc);
11837 #endif
11838 	} else {
11839 #ifdef NEEDTOPORT
11840 		csevent_log(CSC_CTL | CSC_SHELF_SW |
11841 			    CTL_TASK_REPORT,
11842 			    csevent_LogType_Trace,
11843 			    csevent_Severity_Information,
11844 			    csevent_AlertLevel_Green,
11845 			    csevent_FRU_Firmware,
11846 			    csevent_FRU_Unknown,
11847 			    "CTL: received unknown task "
11848 			    "type: %d (%#x)",
11849 			    io->taskio.task_action,
11850 			    io->taskio.task_action);
11851 #endif
11852 	}
11853 	switch (io->taskio.task_action) {
11854 	case CTL_TASK_ABORT_TASK:
11855 		retval = ctl_abort_task(io);
11856 		break;
11857 	case CTL_TASK_ABORT_TASK_SET:
11858 	case CTL_TASK_CLEAR_TASK_SET:
11859 		retval = ctl_abort_task_set(io);
11860 		break;
11861 	case CTL_TASK_CLEAR_ACA:
11862 		break;
11863 	case CTL_TASK_I_T_NEXUS_RESET:
11864 		retval = ctl_i_t_nexus_reset(io);
11865 		break;
11866 	case CTL_TASK_LUN_RESET: {
11867 		struct ctl_lun *lun;
11868 		uint32_t targ_lun;
11869 
11870 		targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11871 		mtx_lock(&softc->ctl_lock);
11872 		if ((targ_lun < CTL_MAX_LUNS)
11873 		 && (softc->ctl_luns[targ_lun] != NULL))
11874 			lun = softc->ctl_luns[targ_lun];
11875 		else {
11876 			mtx_unlock(&softc->ctl_lock);
11877 			retval = 1;
11878 			break;
11879 		}
11880 
11881 		if (!(io->io_hdr.flags &
11882 		    CTL_FLAG_FROM_OTHER_SC)) {
11883 			union ctl_ha_msg msg_info;
11884 
11885 			io->io_hdr.flags |=
11886 				CTL_FLAG_SENT_2OTHER_SC;
11887 			msg_info.hdr.msg_type =
11888 				CTL_MSG_MANAGE_TASKS;
11889 			msg_info.hdr.nexus = io->io_hdr.nexus;
11890 			msg_info.task.task_action =
11891 				CTL_TASK_LUN_RESET;
11892 			msg_info.hdr.original_sc = NULL;
11893 			msg_info.hdr.serializing_sc = NULL;
11894 			if (CTL_HA_STATUS_SUCCESS !=
11895 			    ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11896 			    (void *)&msg_info,
11897 			    sizeof(msg_info), 0)) {
11898 			}
11899 		}
11900 
11901 		retval = ctl_lun_reset(lun, io,
11902 				       CTL_UA_LUN_RESET);
11903 		mtx_unlock(&softc->ctl_lock);
11904 		break;
11905 	}
11906 	case CTL_TASK_TARGET_RESET:
11907 		retval = ctl_target_reset(softc, io, CTL_UA_TARG_RESET);
11908 		break;
11909 	case CTL_TASK_BUS_RESET:
11910 		retval = ctl_bus_reset(softc, io);
11911 		break;
11912 	case CTL_TASK_PORT_LOGIN:
11913 		break;
11914 	case CTL_TASK_PORT_LOGOUT:
11915 		break;
11916 	default:
11917 		printf("ctl_run_task: got unknown task management event %d\n",
11918 		       io->taskio.task_action);
11919 		break;
11920 	}
11921 	if (retval == 0)
11922 		io->io_hdr.status = CTL_SUCCESS;
11923 	else
11924 		io->io_hdr.status = CTL_ERROR;
11925 	ctl_done(io);
11926 }
11927 
11928 /*
11929  * For HA operation.  Handle commands that come in from the other
11930  * controller.
11931  */
11932 static void
11933 ctl_handle_isc(union ctl_io *io)
11934 {
11935 	int free_io;
11936 	struct ctl_lun *lun;
11937 	struct ctl_softc *softc;
11938 	uint32_t targ_lun;
11939 
11940 	softc = control_softc;
11941 
11942 	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11943 	lun = softc->ctl_luns[targ_lun];
11944 
11945 	switch (io->io_hdr.msg_type) {
11946 	case CTL_MSG_SERIALIZE:
11947 		free_io = ctl_serialize_other_sc_cmd(&io->scsiio);
11948 		break;
11949 	case CTL_MSG_R2R: {
11950 		const struct ctl_cmd_entry *entry;
11951 
11952 		/*
11953 		 * This is only used in SER_ONLY mode.
11954 		 */
11955 		free_io = 0;
11956 		entry = ctl_get_cmd_entry(&io->scsiio, NULL);
11957 		mtx_lock(&lun->lun_lock);
11958 		if (ctl_scsiio_lun_check(lun,
11959 		    entry, (struct ctl_scsiio *)io) != 0) {
11960 			mtx_unlock(&lun->lun_lock);
11961 			ctl_done(io);
11962 			break;
11963 		}
11964 		io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11965 		mtx_unlock(&lun->lun_lock);
11966 		ctl_enqueue_rtr(io);
11967 		break;
11968 	}
11969 	case CTL_MSG_FINISH_IO:
11970 		if (softc->ha_mode == CTL_HA_MODE_XFER) {
11971 			free_io = 0;
11972 			ctl_done(io);
11973 		} else {
11974 			free_io = 1;
11975 			mtx_lock(&lun->lun_lock);
11976 			TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
11977 				     ooa_links);
11978 			ctl_check_blocked(lun);
11979 			mtx_unlock(&lun->lun_lock);
11980 		}
11981 		break;
11982 	case CTL_MSG_PERS_ACTION:
11983 		ctl_hndl_per_res_out_on_other_sc(
11984 			(union ctl_ha_msg *)&io->presio.pr_msg);
11985 		free_io = 1;
11986 		break;
11987 	case CTL_MSG_BAD_JUJU:
11988 		free_io = 0;
11989 		ctl_done(io);
11990 		break;
11991 	case CTL_MSG_DATAMOVE:
11992 		/* Only used in XFER mode */
11993 		free_io = 0;
11994 		ctl_datamove_remote(io);
11995 		break;
11996 	case CTL_MSG_DATAMOVE_DONE:
11997 		/* Only used in XFER mode */
11998 		free_io = 0;
11999 		io->scsiio.be_move_done(io);
12000 		break;
12001 	default:
12002 		free_io = 1;
12003 		printf("%s: Invalid message type %d\n",
12004 		       __func__, io->io_hdr.msg_type);
12005 		break;
12006 	}
12007 	if (free_io)
12008 		ctl_free_io(io);
12009 
12010 }
12011 
12012 
12013 /*
12014  * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12015  * there is no match.
12016  */
12017 static ctl_lun_error_pattern
12018 ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12019 {
12020 	const struct ctl_cmd_entry *entry;
12021 	ctl_lun_error_pattern filtered_pattern, pattern;
12022 
12023 	pattern = desc->error_pattern;
12024 
12025 	/*
12026 	 * XXX KDM we need more data passed into this function to match a
12027 	 * custom pattern, and we actually need to implement custom pattern
12028 	 * matching.
12029 	 */
12030 	if (pattern & CTL_LUN_PAT_CMD)
12031 		return (CTL_LUN_PAT_CMD);
12032 
12033 	if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12034 		return (CTL_LUN_PAT_ANY);
12035 
12036 	entry = ctl_get_cmd_entry(ctsio, NULL);
12037 
12038 	filtered_pattern = entry->pattern & pattern;
12039 
12040 	/*
12041 	 * If the user requested specific flags in the pattern (e.g.
12042 	 * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12043 	 * flags.
12044 	 *
12045 	 * If the user did not specify any flags, it doesn't matter whether
12046 	 * or not the command supports the flags.
12047 	 */
12048 	if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12049 	     (pattern & ~CTL_LUN_PAT_MASK))
12050 		return (CTL_LUN_PAT_NONE);
12051 
12052 	/*
12053 	 * If the user asked for a range check, see if the requested LBA
12054 	 * range overlaps with this command's LBA range.
12055 	 */
12056 	if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12057 		uint64_t lba1;
12058 		uint64_t len1;
12059 		ctl_action action;
12060 		int retval;
12061 
12062 		retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12063 		if (retval != 0)
12064 			return (CTL_LUN_PAT_NONE);
12065 
12066 		action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12067 					      desc->lba_range.len, FALSE);
12068 		/*
12069 		 * A "pass" means that the LBA ranges don't overlap, so
12070 		 * this doesn't match the user's range criteria.
12071 		 */
12072 		if (action == CTL_ACTION_PASS)
12073 			return (CTL_LUN_PAT_NONE);
12074 	}
12075 
12076 	return (filtered_pattern);
12077 }
12078 
12079 static void
12080 ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12081 {
12082 	struct ctl_error_desc *desc, *desc2;
12083 
12084 	mtx_assert(&lun->lun_lock, MA_OWNED);
12085 
12086 	STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12087 		ctl_lun_error_pattern pattern;
12088 		/*
12089 		 * Check to see whether this particular command matches
12090 		 * the pattern in the descriptor.
12091 		 */
12092 		pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12093 		if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12094 			continue;
12095 
12096 		switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12097 		case CTL_LUN_INJ_ABORTED:
12098 			ctl_set_aborted(&io->scsiio);
12099 			break;
12100 		case CTL_LUN_INJ_MEDIUM_ERR:
12101 			ctl_set_medium_error(&io->scsiio);
12102 			break;
12103 		case CTL_LUN_INJ_UA:
12104 			/* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12105 			 * OCCURRED */
12106 			ctl_set_ua(&io->scsiio, 0x29, 0x00);
12107 			break;
12108 		case CTL_LUN_INJ_CUSTOM:
12109 			/*
12110 			 * We're assuming the user knows what he is doing.
12111 			 * Just copy the sense information without doing
12112 			 * checks.
12113 			 */
12114 			bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12115 			      MIN(sizeof(desc->custom_sense),
12116 				  sizeof(io->scsiio.sense_data)));
12117 			io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12118 			io->scsiio.sense_len = SSD_FULL_SIZE;
12119 			io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12120 			break;
12121 		case CTL_LUN_INJ_NONE:
12122 		default:
12123 			/*
12124 			 * If this is an error injection type we don't know
12125 			 * about, clear the continuous flag (if it is set)
12126 			 * so it will get deleted below.
12127 			 */
12128 			desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12129 			break;
12130 		}
12131 		/*
12132 		 * By default, each error injection action is a one-shot
12133 		 */
12134 		if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12135 			continue;
12136 
12137 		STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12138 
12139 		free(desc, M_CTL);
12140 	}
12141 }
12142 
12143 #ifdef CTL_IO_DELAY
12144 static void
12145 ctl_datamove_timer_wakeup(void *arg)
12146 {
12147 	union ctl_io *io;
12148 
12149 	io = (union ctl_io *)arg;
12150 
12151 	ctl_datamove(io);
12152 }
12153 #endif /* CTL_IO_DELAY */
12154 
12155 void
12156 ctl_datamove(union ctl_io *io)
12157 {
12158 	void (*fe_datamove)(union ctl_io *io);
12159 
12160 	mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
12161 
12162 	CTL_DEBUG_PRINT(("ctl_datamove\n"));
12163 
12164 #ifdef CTL_TIME_IO
12165 	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12166 		char str[256];
12167 		char path_str[64];
12168 		struct sbuf sb;
12169 
12170 		ctl_scsi_path_string(io, path_str, sizeof(path_str));
12171 		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12172 
12173 		sbuf_cat(&sb, path_str);
12174 		switch (io->io_hdr.io_type) {
12175 		case CTL_IO_SCSI:
12176 			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12177 			sbuf_printf(&sb, "\n");
12178 			sbuf_cat(&sb, path_str);
12179 			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12180 				    io->scsiio.tag_num, io->scsiio.tag_type);
12181 			break;
12182 		case CTL_IO_TASK:
12183 			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12184 				    "Tag Type: %d\n", io->taskio.task_action,
12185 				    io->taskio.tag_num, io->taskio.tag_type);
12186 			break;
12187 		default:
12188 			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12189 			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12190 			break;
12191 		}
12192 		sbuf_cat(&sb, path_str);
12193 		sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12194 			    (intmax_t)time_uptime - io->io_hdr.start_time);
12195 		sbuf_finish(&sb);
12196 		printf("%s", sbuf_data(&sb));
12197 	}
12198 #endif /* CTL_TIME_IO */
12199 
12200 #ifdef CTL_IO_DELAY
12201 	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12202 		struct ctl_lun *lun;
12203 
12204 		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12205 
12206 		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12207 	} else {
12208 		struct ctl_lun *lun;
12209 
12210 		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12211 		if ((lun != NULL)
12212 		 && (lun->delay_info.datamove_delay > 0)) {
12213 			struct callout *callout;
12214 
12215 			callout = (struct callout *)&io->io_hdr.timer_bytes;
12216 			callout_init(callout, /*mpsafe*/ 1);
12217 			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12218 			callout_reset(callout,
12219 				      lun->delay_info.datamove_delay * hz,
12220 				      ctl_datamove_timer_wakeup, io);
12221 			if (lun->delay_info.datamove_type ==
12222 			    CTL_DELAY_TYPE_ONESHOT)
12223 				lun->delay_info.datamove_delay = 0;
12224 			return;
12225 		}
12226 	}
12227 #endif
12228 
12229 	/*
12230 	 * This command has been aborted.  Set the port status, so we fail
12231 	 * the data move.
12232 	 */
12233 	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12234 		printf("ctl_datamove: tag 0x%04x on (%ju:%d:%ju:%d) aborted\n",
12235 		       io->scsiio.tag_num,(uintmax_t)io->io_hdr.nexus.initid.id,
12236 		       io->io_hdr.nexus.targ_port,
12237 		       (uintmax_t)io->io_hdr.nexus.targ_target.id,
12238 		       io->io_hdr.nexus.targ_lun);
12239 		io->io_hdr.port_status = 31337;
12240 		/*
12241 		 * Note that the backend, in this case, will get the
12242 		 * callback in its context.  In other cases it may get
12243 		 * called in the frontend's interrupt thread context.
12244 		 */
12245 		io->scsiio.be_move_done(io);
12246 		return;
12247 	}
12248 
12249 	/* Don't confuse frontend with zero length data move. */
12250 	if (io->scsiio.kern_data_len == 0) {
12251 		io->scsiio.be_move_done(io);
12252 		return;
12253 	}
12254 
12255 	/*
12256 	 * If we're in XFER mode and this I/O is from the other shelf
12257 	 * controller, we need to send the DMA to the other side to
12258 	 * actually transfer the data to/from the host.  In serialize only
12259 	 * mode the transfer happens below CTL and ctl_datamove() is only
12260 	 * called on the machine that originally received the I/O.
12261 	 */
12262 	if ((control_softc->ha_mode == CTL_HA_MODE_XFER)
12263 	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12264 		union ctl_ha_msg msg;
12265 		uint32_t sg_entries_sent;
12266 		int do_sg_copy;
12267 		int i;
12268 
12269 		memset(&msg, 0, sizeof(msg));
12270 		msg.hdr.msg_type = CTL_MSG_DATAMOVE;
12271 		msg.hdr.original_sc = io->io_hdr.original_sc;
12272 		msg.hdr.serializing_sc = io;
12273 		msg.hdr.nexus = io->io_hdr.nexus;
12274 		msg.dt.flags = io->io_hdr.flags;
12275 		/*
12276 		 * We convert everything into a S/G list here.  We can't
12277 		 * pass by reference, only by value between controllers.
12278 		 * So we can't pass a pointer to the S/G list, only as many
12279 		 * S/G entries as we can fit in here.  If it's possible for
12280 		 * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
12281 		 * then we need to break this up into multiple transfers.
12282 		 */
12283 		if (io->scsiio.kern_sg_entries == 0) {
12284 			msg.dt.kern_sg_entries = 1;
12285 			/*
12286 			 * If this is in cached memory, flush the cache
12287 			 * before we send the DMA request to the other
12288 			 * controller.  We want to do this in either the
12289 			 * read or the write case.  The read case is
12290 			 * straightforward.  In the write case, we want to
12291 			 * make sure nothing is in the local cache that
12292 			 * could overwrite the DMAed data.
12293 			 */
12294 			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12295 				/*
12296 				 * XXX KDM use bus_dmamap_sync() here.
12297 				 */
12298 			}
12299 
12300 			/*
12301 			 * Convert to a physical address if this is a
12302 			 * virtual address.
12303 			 */
12304 			if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
12305 				msg.dt.sg_list[0].addr =
12306 					io->scsiio.kern_data_ptr;
12307 			} else {
12308 				/*
12309 				 * XXX KDM use busdma here!
12310 				 */
12311 #if 0
12312 				msg.dt.sg_list[0].addr = (void *)
12313 					vtophys(io->scsiio.kern_data_ptr);
12314 #endif
12315 			}
12316 
12317 			msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
12318 			do_sg_copy = 0;
12319 		} else {
12320 			struct ctl_sg_entry *sgl;
12321 
12322 			do_sg_copy = 1;
12323 			msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
12324 			sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
12325 			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12326 				/*
12327 				 * XXX KDM use bus_dmamap_sync() here.
12328 				 */
12329 			}
12330 		}
12331 
12332 		msg.dt.kern_data_len = io->scsiio.kern_data_len;
12333 		msg.dt.kern_total_len = io->scsiio.kern_total_len;
12334 		msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
12335 		msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
12336 		msg.dt.sg_sequence = 0;
12337 
12338 		/*
12339 		 * Loop until we've sent all of the S/G entries.  On the
12340 		 * other end, we'll recompose these S/G entries into one
12341 		 * contiguous list before passing it to the
12342 		 */
12343 		for (sg_entries_sent = 0; sg_entries_sent <
12344 		     msg.dt.kern_sg_entries; msg.dt.sg_sequence++) {
12345 			msg.dt.cur_sg_entries = MIN((sizeof(msg.dt.sg_list)/
12346 				sizeof(msg.dt.sg_list[0])),
12347 				msg.dt.kern_sg_entries - sg_entries_sent);
12348 
12349 			if (do_sg_copy != 0) {
12350 				struct ctl_sg_entry *sgl;
12351 				int j;
12352 
12353 				sgl = (struct ctl_sg_entry *)
12354 					io->scsiio.kern_data_ptr;
12355 				/*
12356 				 * If this is in cached memory, flush the cache
12357 				 * before we send the DMA request to the other
12358 				 * controller.  We want to do this in either
12359 				 * the * read or the write case.  The read
12360 				 * case is straightforward.  In the write
12361 				 * case, we want to make sure nothing is
12362 				 * in the local cache that could overwrite
12363 				 * the DMAed data.
12364 				 */
12365 
12366 				for (i = sg_entries_sent, j = 0;
12367 				     i < msg.dt.cur_sg_entries; i++, j++) {
12368 					if ((io->io_hdr.flags &
12369 					     CTL_FLAG_NO_DATASYNC) == 0) {
12370 						/*
12371 						 * XXX KDM use bus_dmamap_sync()
12372 						 */
12373 					}
12374 					if ((io->io_hdr.flags &
12375 					     CTL_FLAG_BUS_ADDR) == 0) {
12376 						/*
12377 						 * XXX KDM use busdma.
12378 						 */
12379 #if 0
12380 						msg.dt.sg_list[j].addr =(void *)
12381 						       vtophys(sgl[i].addr);
12382 #endif
12383 					} else {
12384 						msg.dt.sg_list[j].addr =
12385 							sgl[i].addr;
12386 					}
12387 					msg.dt.sg_list[j].len = sgl[i].len;
12388 				}
12389 			}
12390 
12391 			sg_entries_sent += msg.dt.cur_sg_entries;
12392 			if (sg_entries_sent >= msg.dt.kern_sg_entries)
12393 				msg.dt.sg_last = 1;
12394 			else
12395 				msg.dt.sg_last = 0;
12396 
12397 			/*
12398 			 * XXX KDM drop and reacquire the lock here?
12399 			 */
12400 			if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12401 			    sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
12402 				/*
12403 				 * XXX do something here.
12404 				 */
12405 			}
12406 
12407 			msg.dt.sent_sg_entries = sg_entries_sent;
12408 		}
12409 		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12410 		if (io->io_hdr.flags & CTL_FLAG_FAILOVER)
12411 			ctl_failover_io(io, /*have_lock*/ 0);
12412 
12413 	} else {
12414 
12415 		/*
12416 		 * Lookup the fe_datamove() function for this particular
12417 		 * front end.
12418 		 */
12419 		fe_datamove =
12420 		    control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12421 
12422 		fe_datamove(io);
12423 	}
12424 }
12425 
12426 static void
12427 ctl_send_datamove_done(union ctl_io *io, int have_lock)
12428 {
12429 	union ctl_ha_msg msg;
12430 	int isc_status;
12431 
12432 	memset(&msg, 0, sizeof(msg));
12433 
12434 	msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12435 	msg.hdr.original_sc = io;
12436 	msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12437 	msg.hdr.nexus = io->io_hdr.nexus;
12438 	msg.hdr.status = io->io_hdr.status;
12439 	msg.scsi.tag_num = io->scsiio.tag_num;
12440 	msg.scsi.tag_type = io->scsiio.tag_type;
12441 	msg.scsi.scsi_status = io->scsiio.scsi_status;
12442 	memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12443 	       sizeof(io->scsiio.sense_data));
12444 	msg.scsi.sense_len = io->scsiio.sense_len;
12445 	msg.scsi.sense_residual = io->scsiio.sense_residual;
12446 	msg.scsi.fetd_status = io->io_hdr.port_status;
12447 	msg.scsi.residual = io->scsiio.residual;
12448 	io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12449 
12450 	if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12451 		ctl_failover_io(io, /*have_lock*/ have_lock);
12452 		return;
12453 	}
12454 
12455 	isc_status = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0);
12456 	if (isc_status > CTL_HA_STATUS_SUCCESS) {
12457 		/* XXX do something if this fails */
12458 	}
12459 
12460 }
12461 
12462 /*
12463  * The DMA to the remote side is done, now we need to tell the other side
12464  * we're done so it can continue with its data movement.
12465  */
12466 static void
12467 ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12468 {
12469 	union ctl_io *io;
12470 
12471 	io = rq->context;
12472 
12473 	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12474 		printf("%s: ISC DMA write failed with error %d", __func__,
12475 		       rq->ret);
12476 		ctl_set_internal_failure(&io->scsiio,
12477 					 /*sks_valid*/ 1,
12478 					 /*retry_count*/ rq->ret);
12479 	}
12480 
12481 	ctl_dt_req_free(rq);
12482 
12483 	/*
12484 	 * In this case, we had to malloc the memory locally.  Free it.
12485 	 */
12486 	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12487 		int i;
12488 		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12489 			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12490 	}
12491 	/*
12492 	 * The data is in local and remote memory, so now we need to send
12493 	 * status (good or back) back to the other side.
12494 	 */
12495 	ctl_send_datamove_done(io, /*have_lock*/ 0);
12496 }
12497 
12498 /*
12499  * We've moved the data from the host/controller into local memory.  Now we
12500  * need to push it over to the remote controller's memory.
12501  */
12502 static int
12503 ctl_datamove_remote_dm_write_cb(union ctl_io *io)
12504 {
12505 	int retval;
12506 
12507 	retval = 0;
12508 
12509 	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12510 					  ctl_datamove_remote_write_cb);
12511 
12512 	return (retval);
12513 }
12514 
12515 static void
12516 ctl_datamove_remote_write(union ctl_io *io)
12517 {
12518 	int retval;
12519 	void (*fe_datamove)(union ctl_io *io);
12520 
12521 	/*
12522 	 * - Get the data from the host/HBA into local memory.
12523 	 * - DMA memory from the local controller to the remote controller.
12524 	 * - Send status back to the remote controller.
12525 	 */
12526 
12527 	retval = ctl_datamove_remote_sgl_setup(io);
12528 	if (retval != 0)
12529 		return;
12530 
12531 	/* Switch the pointer over so the FETD knows what to do */
12532 	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12533 
12534 	/*
12535 	 * Use a custom move done callback, since we need to send completion
12536 	 * back to the other controller, not to the backend on this side.
12537 	 */
12538 	io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12539 
12540 	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12541 
12542 	fe_datamove(io);
12543 
12544 	return;
12545 
12546 }
12547 
12548 static int
12549 ctl_datamove_remote_dm_read_cb(union ctl_io *io)
12550 {
12551 #if 0
12552 	char str[256];
12553 	char path_str[64];
12554 	struct sbuf sb;
12555 #endif
12556 
12557 	/*
12558 	 * In this case, we had to malloc the memory locally.  Free it.
12559 	 */
12560 	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12561 		int i;
12562 		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12563 			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12564 	}
12565 
12566 #if 0
12567 	scsi_path_string(io, path_str, sizeof(path_str));
12568 	sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12569 	sbuf_cat(&sb, path_str);
12570 	scsi_command_string(&io->scsiio, NULL, &sb);
12571 	sbuf_printf(&sb, "\n");
12572 	sbuf_cat(&sb, path_str);
12573 	sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12574 		    io->scsiio.tag_num, io->scsiio.tag_type);
12575 	sbuf_cat(&sb, path_str);
12576 	sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
12577 		    io->io_hdr.flags, io->io_hdr.status);
12578 	sbuf_finish(&sb);
12579 	printk("%s", sbuf_data(&sb));
12580 #endif
12581 
12582 
12583 	/*
12584 	 * The read is done, now we need to send status (good or bad) back
12585 	 * to the other side.
12586 	 */
12587 	ctl_send_datamove_done(io, /*have_lock*/ 0);
12588 
12589 	return (0);
12590 }
12591 
12592 static void
12593 ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
12594 {
12595 	union ctl_io *io;
12596 	void (*fe_datamove)(union ctl_io *io);
12597 
12598 	io = rq->context;
12599 
12600 	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12601 		printf("%s: ISC DMA read failed with error %d", __func__,
12602 		       rq->ret);
12603 		ctl_set_internal_failure(&io->scsiio,
12604 					 /*sks_valid*/ 1,
12605 					 /*retry_count*/ rq->ret);
12606 	}
12607 
12608 	ctl_dt_req_free(rq);
12609 
12610 	/* Switch the pointer over so the FETD knows what to do */
12611 	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12612 
12613 	/*
12614 	 * Use a custom move done callback, since we need to send completion
12615 	 * back to the other controller, not to the backend on this side.
12616 	 */
12617 	io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
12618 
12619 	/* XXX KDM add checks like the ones in ctl_datamove? */
12620 
12621 	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12622 
12623 	fe_datamove(io);
12624 }
12625 
12626 static int
12627 ctl_datamove_remote_sgl_setup(union ctl_io *io)
12628 {
12629 	struct ctl_sg_entry *local_sglist, *remote_sglist;
12630 	struct ctl_sg_entry *local_dma_sglist, *remote_dma_sglist;
12631 	struct ctl_softc *softc;
12632 	int retval;
12633 	int i;
12634 
12635 	retval = 0;
12636 	softc = control_softc;
12637 
12638 	local_sglist = io->io_hdr.local_sglist;
12639 	local_dma_sglist = io->io_hdr.local_dma_sglist;
12640 	remote_sglist = io->io_hdr.remote_sglist;
12641 	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
12642 
12643 	if (io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) {
12644 		for (i = 0; i < io->scsiio.kern_sg_entries; i++) {
12645 			local_sglist[i].len = remote_sglist[i].len;
12646 
12647 			/*
12648 			 * XXX Detect the situation where the RS-level I/O
12649 			 * redirector on the other side has already read the
12650 			 * data off of the AOR RS on this side, and
12651 			 * transferred it to remote (mirror) memory on the
12652 			 * other side.  Since we already have the data in
12653 			 * memory here, we just need to use it.
12654 			 *
12655 			 * XXX KDM this can probably be removed once we
12656 			 * get the cache device code in and take the
12657 			 * current AOR implementation out.
12658 			 */
12659 #ifdef NEEDTOPORT
12660 			if ((remote_sglist[i].addr >=
12661 			     (void *)vtophys(softc->mirr->addr))
12662 			 && (remote_sglist[i].addr <
12663 			     ((void *)vtophys(softc->mirr->addr) +
12664 			     CacheMirrorOffset))) {
12665 				local_sglist[i].addr = remote_sglist[i].addr -
12666 					CacheMirrorOffset;
12667 				if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
12668 				     CTL_FLAG_DATA_IN)
12669 					io->io_hdr.flags |= CTL_FLAG_REDIR_DONE;
12670 			} else {
12671 				local_sglist[i].addr = remote_sglist[i].addr +
12672 					CacheMirrorOffset;
12673 			}
12674 #endif
12675 #if 0
12676 			printf("%s: local %p, remote %p, len %d\n",
12677 			       __func__, local_sglist[i].addr,
12678 			       remote_sglist[i].addr, local_sglist[i].len);
12679 #endif
12680 		}
12681 	} else {
12682 		uint32_t len_to_go;
12683 
12684 		/*
12685 		 * In this case, we don't have automatically allocated
12686 		 * memory for this I/O on this controller.  This typically
12687 		 * happens with internal CTL I/O -- e.g. inquiry, mode
12688 		 * sense, etc.  Anything coming from RAIDCore will have
12689 		 * a mirror area available.
12690 		 */
12691 		len_to_go = io->scsiio.kern_data_len;
12692 
12693 		/*
12694 		 * Clear the no datasync flag, we have to use malloced
12695 		 * buffers.
12696 		 */
12697 		io->io_hdr.flags &= ~CTL_FLAG_NO_DATASYNC;
12698 
12699 		/*
12700 		 * The difficult thing here is that the size of the various
12701 		 * S/G segments may be different than the size from the
12702 		 * remote controller.  That'll make it harder when DMAing
12703 		 * the data back to the other side.
12704 		 */
12705 		for (i = 0; (i < sizeof(io->io_hdr.remote_sglist) /
12706 		     sizeof(io->io_hdr.remote_sglist[0])) &&
12707 		     (len_to_go > 0); i++) {
12708 			local_sglist[i].len = MIN(len_to_go, 131072);
12709 			CTL_SIZE_8B(local_dma_sglist[i].len,
12710 				    local_sglist[i].len);
12711 			local_sglist[i].addr =
12712 				malloc(local_dma_sglist[i].len, M_CTL,M_WAITOK);
12713 
12714 			local_dma_sglist[i].addr = local_sglist[i].addr;
12715 
12716 			if (local_sglist[i].addr == NULL) {
12717 				int j;
12718 
12719 				printf("malloc failed for %zd bytes!",
12720 				       local_dma_sglist[i].len);
12721 				for (j = 0; j < i; j++) {
12722 					free(local_sglist[j].addr, M_CTL);
12723 				}
12724 				ctl_set_internal_failure(&io->scsiio,
12725 							 /*sks_valid*/ 1,
12726 							 /*retry_count*/ 4857);
12727 				retval = 1;
12728 				goto bailout_error;
12729 
12730 			}
12731 			/* XXX KDM do we need a sync here? */
12732 
12733 			len_to_go -= local_sglist[i].len;
12734 		}
12735 		/*
12736 		 * Reset the number of S/G entries accordingly.  The
12737 		 * original number of S/G entries is available in
12738 		 * rem_sg_entries.
12739 		 */
12740 		io->scsiio.kern_sg_entries = i;
12741 
12742 #if 0
12743 		printf("%s: kern_sg_entries = %d\n", __func__,
12744 		       io->scsiio.kern_sg_entries);
12745 		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12746 			printf("%s: sg[%d] = %p, %d (DMA: %d)\n", __func__, i,
12747 			       local_sglist[i].addr, local_sglist[i].len,
12748 			       local_dma_sglist[i].len);
12749 #endif
12750 	}
12751 
12752 
12753 	return (retval);
12754 
12755 bailout_error:
12756 
12757 	ctl_send_datamove_done(io, /*have_lock*/ 0);
12758 
12759 	return (retval);
12760 }
12761 
12762 static int
12763 ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
12764 			 ctl_ha_dt_cb callback)
12765 {
12766 	struct ctl_ha_dt_req *rq;
12767 	struct ctl_sg_entry *remote_sglist, *local_sglist;
12768 	struct ctl_sg_entry *remote_dma_sglist, *local_dma_sglist;
12769 	uint32_t local_used, remote_used, total_used;
12770 	int retval;
12771 	int i, j;
12772 
12773 	retval = 0;
12774 
12775 	rq = ctl_dt_req_alloc();
12776 
12777 	/*
12778 	 * If we failed to allocate the request, and if the DMA didn't fail
12779 	 * anyway, set busy status.  This is just a resource allocation
12780 	 * failure.
12781 	 */
12782 	if ((rq == NULL)
12783 	 && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE))
12784 		ctl_set_busy(&io->scsiio);
12785 
12786 	if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) {
12787 
12788 		if (rq != NULL)
12789 			ctl_dt_req_free(rq);
12790 
12791 		/*
12792 		 * The data move failed.  We need to return status back
12793 		 * to the other controller.  No point in trying to DMA
12794 		 * data to the remote controller.
12795 		 */
12796 
12797 		ctl_send_datamove_done(io, /*have_lock*/ 0);
12798 
12799 		retval = 1;
12800 
12801 		goto bailout;
12802 	}
12803 
12804 	local_sglist = io->io_hdr.local_sglist;
12805 	local_dma_sglist = io->io_hdr.local_dma_sglist;
12806 	remote_sglist = io->io_hdr.remote_sglist;
12807 	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
12808 	local_used = 0;
12809 	remote_used = 0;
12810 	total_used = 0;
12811 
12812 	if (io->io_hdr.flags & CTL_FLAG_REDIR_DONE) {
12813 		rq->ret = CTL_HA_STATUS_SUCCESS;
12814 		rq->context = io;
12815 		callback(rq);
12816 		goto bailout;
12817 	}
12818 
12819 	/*
12820 	 * Pull/push the data over the wire from/to the other controller.
12821 	 * This takes into account the possibility that the local and
12822 	 * remote sglists may not be identical in terms of the size of
12823 	 * the elements and the number of elements.
12824 	 *
12825 	 * One fundamental assumption here is that the length allocated for
12826 	 * both the local and remote sglists is identical.  Otherwise, we've
12827 	 * essentially got a coding error of some sort.
12828 	 */
12829 	for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
12830 		int isc_ret;
12831 		uint32_t cur_len, dma_length;
12832 		uint8_t *tmp_ptr;
12833 
12834 		rq->id = CTL_HA_DATA_CTL;
12835 		rq->command = command;
12836 		rq->context = io;
12837 
12838 		/*
12839 		 * Both pointers should be aligned.  But it is possible
12840 		 * that the allocation length is not.  They should both
12841 		 * also have enough slack left over at the end, though,
12842 		 * to round up to the next 8 byte boundary.
12843 		 */
12844 		cur_len = MIN(local_sglist[i].len - local_used,
12845 			      remote_sglist[j].len - remote_used);
12846 
12847 		/*
12848 		 * In this case, we have a size issue and need to decrease
12849 		 * the size, except in the case where we actually have less
12850 		 * than 8 bytes left.  In that case, we need to increase
12851 		 * the DMA length to get the last bit.
12852 		 */
12853 		if ((cur_len & 0x7) != 0) {
12854 			if (cur_len > 0x7) {
12855 				cur_len = cur_len - (cur_len & 0x7);
12856 				dma_length = cur_len;
12857 			} else {
12858 				CTL_SIZE_8B(dma_length, cur_len);
12859 			}
12860 
12861 		} else
12862 			dma_length = cur_len;
12863 
12864 		/*
12865 		 * If we had to allocate memory for this I/O, instead of using
12866 		 * the non-cached mirror memory, we'll need to flush the cache
12867 		 * before trying to DMA to the other controller.
12868 		 *
12869 		 * We could end up doing this multiple times for the same
12870 		 * segment if we have a larger local segment than remote
12871 		 * segment.  That shouldn't be an issue.
12872 		 */
12873 		if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12874 			/*
12875 			 * XXX KDM use bus_dmamap_sync() here.
12876 			 */
12877 		}
12878 
12879 		rq->size = dma_length;
12880 
12881 		tmp_ptr = (uint8_t *)local_sglist[i].addr;
12882 		tmp_ptr += local_used;
12883 
12884 		/* Use physical addresses when talking to ISC hardware */
12885 		if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
12886 			/* XXX KDM use busdma */
12887 #if 0
12888 			rq->local = vtophys(tmp_ptr);
12889 #endif
12890 		} else
12891 			rq->local = tmp_ptr;
12892 
12893 		tmp_ptr = (uint8_t *)remote_sglist[j].addr;
12894 		tmp_ptr += remote_used;
12895 		rq->remote = tmp_ptr;
12896 
12897 		rq->callback = NULL;
12898 
12899 		local_used += cur_len;
12900 		if (local_used >= local_sglist[i].len) {
12901 			i++;
12902 			local_used = 0;
12903 		}
12904 
12905 		remote_used += cur_len;
12906 		if (remote_used >= remote_sglist[j].len) {
12907 			j++;
12908 			remote_used = 0;
12909 		}
12910 		total_used += cur_len;
12911 
12912 		if (total_used >= io->scsiio.kern_data_len)
12913 			rq->callback = callback;
12914 
12915 		if ((rq->size & 0x7) != 0) {
12916 			printf("%s: warning: size %d is not on 8b boundary\n",
12917 			       __func__, rq->size);
12918 		}
12919 		if (((uintptr_t)rq->local & 0x7) != 0) {
12920 			printf("%s: warning: local %p not on 8b boundary\n",
12921 			       __func__, rq->local);
12922 		}
12923 		if (((uintptr_t)rq->remote & 0x7) != 0) {
12924 			printf("%s: warning: remote %p not on 8b boundary\n",
12925 			       __func__, rq->local);
12926 		}
12927 #if 0
12928 		printf("%s: %s: local %#x remote %#x size %d\n", __func__,
12929 		       (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
12930 		       rq->local, rq->remote, rq->size);
12931 #endif
12932 
12933 		isc_ret = ctl_dt_single(rq);
12934 		if (isc_ret == CTL_HA_STATUS_WAIT)
12935 			continue;
12936 
12937 		if (isc_ret == CTL_HA_STATUS_DISCONNECT) {
12938 			rq->ret = CTL_HA_STATUS_SUCCESS;
12939 		} else {
12940 			rq->ret = isc_ret;
12941 		}
12942 		callback(rq);
12943 		goto bailout;
12944 	}
12945 
12946 bailout:
12947 	return (retval);
12948 
12949 }
12950 
12951 static void
12952 ctl_datamove_remote_read(union ctl_io *io)
12953 {
12954 	int retval;
12955 	int i;
12956 
12957 	/*
12958 	 * This will send an error to the other controller in the case of a
12959 	 * failure.
12960 	 */
12961 	retval = ctl_datamove_remote_sgl_setup(io);
12962 	if (retval != 0)
12963 		return;
12964 
12965 	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
12966 					  ctl_datamove_remote_read_cb);
12967 	if ((retval != 0)
12968 	 && ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0)) {
12969 		/*
12970 		 * Make sure we free memory if there was an error..  The
12971 		 * ctl_datamove_remote_xfer() function will send the
12972 		 * datamove done message, or call the callback with an
12973 		 * error if there is a problem.
12974 		 */
12975 		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12976 			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12977 	}
12978 
12979 	return;
12980 }
12981 
12982 /*
12983  * Process a datamove request from the other controller.  This is used for
12984  * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
12985  * first.  Once that is complete, the data gets DMAed into the remote
12986  * controller's memory.  For reads, we DMA from the remote controller's
12987  * memory into our memory first, and then move it out to the FETD.
12988  */
12989 static void
12990 ctl_datamove_remote(union ctl_io *io)
12991 {
12992 	struct ctl_softc *softc;
12993 
12994 	softc = control_softc;
12995 
12996 	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
12997 
12998 	/*
12999 	 * Note that we look for an aborted I/O here, but don't do some of
13000 	 * the other checks that ctl_datamove() normally does.
13001 	 * We don't need to run the datamove delay code, since that should
13002 	 * have been done if need be on the other controller.
13003 	 */
13004 	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
13005 		printf("%s: tag 0x%04x on (%d:%d:%d:%d) aborted\n", __func__,
13006 		       io->scsiio.tag_num, io->io_hdr.nexus.initid.id,
13007 		       io->io_hdr.nexus.targ_port,
13008 		       io->io_hdr.nexus.targ_target.id,
13009 		       io->io_hdr.nexus.targ_lun);
13010 		io->io_hdr.port_status = 31338;
13011 		ctl_send_datamove_done(io, /*have_lock*/ 0);
13012 		return;
13013 	}
13014 
13015 	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT) {
13016 		ctl_datamove_remote_write(io);
13017 	} else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN){
13018 		ctl_datamove_remote_read(io);
13019 	} else {
13020 		union ctl_ha_msg msg;
13021 		struct scsi_sense_data *sense;
13022 		uint8_t sks[3];
13023 		int retry_count;
13024 
13025 		memset(&msg, 0, sizeof(msg));
13026 
13027 		msg.hdr.msg_type = CTL_MSG_BAD_JUJU;
13028 		msg.hdr.status = CTL_SCSI_ERROR;
13029 		msg.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
13030 
13031 		retry_count = 4243;
13032 
13033 		sense = &msg.scsi.sense_data;
13034 		sks[0] = SSD_SCS_VALID;
13035 		sks[1] = (retry_count >> 8) & 0xff;
13036 		sks[2] = retry_count & 0xff;
13037 
13038 		/* "Internal target failure" */
13039 		scsi_set_sense_data(sense,
13040 				    /*sense_format*/ SSD_TYPE_NONE,
13041 				    /*current_error*/ 1,
13042 				    /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
13043 				    /*asc*/ 0x44,
13044 				    /*ascq*/ 0x00,
13045 				    /*type*/ SSD_ELEM_SKS,
13046 				    /*size*/ sizeof(sks),
13047 				    /*data*/ sks,
13048 				    SSD_ELEM_NONE);
13049 
13050 		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
13051 		if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
13052 			ctl_failover_io(io, /*have_lock*/ 1);
13053 			return;
13054 		}
13055 
13056 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0) >
13057 		    CTL_HA_STATUS_SUCCESS) {
13058 			/* XXX KDM what to do if this fails? */
13059 		}
13060 		return;
13061 	}
13062 
13063 }
13064 
13065 static int
13066 ctl_process_done(union ctl_io *io)
13067 {
13068 	struct ctl_lun *lun;
13069 	struct ctl_softc *softc = control_softc;
13070 	void (*fe_done)(union ctl_io *io);
13071 	uint32_t targ_port = ctl_port_idx(io->io_hdr.nexus.targ_port);
13072 
13073 	CTL_DEBUG_PRINT(("ctl_process_done\n"));
13074 
13075 	fe_done = softc->ctl_ports[targ_port]->fe_done;
13076 
13077 #ifdef CTL_TIME_IO
13078 	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
13079 		char str[256];
13080 		char path_str[64];
13081 		struct sbuf sb;
13082 
13083 		ctl_scsi_path_string(io, path_str, sizeof(path_str));
13084 		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13085 
13086 		sbuf_cat(&sb, path_str);
13087 		switch (io->io_hdr.io_type) {
13088 		case CTL_IO_SCSI:
13089 			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
13090 			sbuf_printf(&sb, "\n");
13091 			sbuf_cat(&sb, path_str);
13092 			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13093 				    io->scsiio.tag_num, io->scsiio.tag_type);
13094 			break;
13095 		case CTL_IO_TASK:
13096 			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
13097 				    "Tag Type: %d\n", io->taskio.task_action,
13098 				    io->taskio.tag_num, io->taskio.tag_type);
13099 			break;
13100 		default:
13101 			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13102 			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13103 			break;
13104 		}
13105 		sbuf_cat(&sb, path_str);
13106 		sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
13107 			    (intmax_t)time_uptime - io->io_hdr.start_time);
13108 		sbuf_finish(&sb);
13109 		printf("%s", sbuf_data(&sb));
13110 	}
13111 #endif /* CTL_TIME_IO */
13112 
13113 	switch (io->io_hdr.io_type) {
13114 	case CTL_IO_SCSI:
13115 		break;
13116 	case CTL_IO_TASK:
13117 		if (ctl_debug & CTL_DEBUG_INFO)
13118 			ctl_io_error_print(io, NULL);
13119 		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
13120 			ctl_free_io(io);
13121 		else
13122 			fe_done(io);
13123 		return (CTL_RETVAL_COMPLETE);
13124 	default:
13125 		panic("ctl_process_done: invalid io type %d\n",
13126 		      io->io_hdr.io_type);
13127 		break; /* NOTREACHED */
13128 	}
13129 
13130 	lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13131 	if (lun == NULL) {
13132 		CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
13133 				 io->io_hdr.nexus.targ_mapped_lun));
13134 		goto bailout;
13135 	}
13136 
13137 	mtx_lock(&lun->lun_lock);
13138 
13139 	/*
13140 	 * Check to see if we have any errors to inject here.  We only
13141 	 * inject errors for commands that don't already have errors set.
13142 	 */
13143 	if ((STAILQ_FIRST(&lun->error_list) != NULL) &&
13144 	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) &&
13145 	    ((io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0))
13146 		ctl_inject_error(lun, io);
13147 
13148 	/*
13149 	 * XXX KDM how do we treat commands that aren't completed
13150 	 * successfully?
13151 	 *
13152 	 * XXX KDM should we also track I/O latency?
13153 	 */
13154 	if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
13155 	    io->io_hdr.io_type == CTL_IO_SCSI) {
13156 #ifdef CTL_TIME_IO
13157 		struct bintime cur_bt;
13158 #endif
13159 		int type;
13160 
13161 		if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13162 		    CTL_FLAG_DATA_IN)
13163 			type = CTL_STATS_READ;
13164 		else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13165 		    CTL_FLAG_DATA_OUT)
13166 			type = CTL_STATS_WRITE;
13167 		else
13168 			type = CTL_STATS_NO_IO;
13169 
13170 		lun->stats.ports[targ_port].bytes[type] +=
13171 		    io->scsiio.kern_total_len;
13172 		lun->stats.ports[targ_port].operations[type]++;
13173 #ifdef CTL_TIME_IO
13174 		bintime_add(&lun->stats.ports[targ_port].dma_time[type],
13175 		   &io->io_hdr.dma_bt);
13176 		lun->stats.ports[targ_port].num_dmas[type] +=
13177 		    io->io_hdr.num_dmas;
13178 		getbintime(&cur_bt);
13179 		bintime_sub(&cur_bt, &io->io_hdr.start_bt);
13180 		bintime_add(&lun->stats.ports[targ_port].time[type], &cur_bt);
13181 #endif
13182 	}
13183 
13184 	/*
13185 	 * Remove this from the OOA queue.
13186 	 */
13187 	TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13188 #ifdef CTL_TIME_IO
13189 	if (TAILQ_EMPTY(&lun->ooa_queue))
13190 		lun->last_busy = getsbinuptime();
13191 #endif
13192 
13193 	/*
13194 	 * Run through the blocked queue on this LUN and see if anything
13195 	 * has become unblocked, now that this transaction is done.
13196 	 */
13197 	ctl_check_blocked(lun);
13198 
13199 	/*
13200 	 * If the LUN has been invalidated, free it if there is nothing
13201 	 * left on its OOA queue.
13202 	 */
13203 	if ((lun->flags & CTL_LUN_INVALID)
13204 	 && TAILQ_EMPTY(&lun->ooa_queue)) {
13205 		mtx_unlock(&lun->lun_lock);
13206 		mtx_lock(&softc->ctl_lock);
13207 		ctl_free_lun(lun);
13208 		mtx_unlock(&softc->ctl_lock);
13209 	} else
13210 		mtx_unlock(&lun->lun_lock);
13211 
13212 bailout:
13213 
13214 	/*
13215 	 * If this command has been aborted, make sure we set the status
13216 	 * properly.  The FETD is responsible for freeing the I/O and doing
13217 	 * whatever it needs to do to clean up its state.
13218 	 */
13219 	if (io->io_hdr.flags & CTL_FLAG_ABORT)
13220 		ctl_set_task_aborted(&io->scsiio);
13221 
13222 	/*
13223 	 * If enabled, print command error status.
13224 	 */
13225 	if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS &&
13226 	    (ctl_debug & CTL_DEBUG_INFO) != 0)
13227 		ctl_io_error_print(io, NULL);
13228 
13229 	/*
13230 	 * Tell the FETD or the other shelf controller we're done with this
13231 	 * command.  Note that only SCSI commands get to this point.  Task
13232 	 * management commands are completed above.
13233 	 *
13234 	 * We only send status to the other controller if we're in XFER
13235 	 * mode.  In SER_ONLY mode, the I/O is done on the controller that
13236 	 * received the I/O (from CTL's perspective), and so the status is
13237 	 * generated there.
13238 	 *
13239 	 * XXX KDM if we hold the lock here, we could cause a deadlock
13240 	 * if the frontend comes back in in this context to queue
13241 	 * something.
13242 	 */
13243 	if ((softc->ha_mode == CTL_HA_MODE_XFER)
13244 	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
13245 		union ctl_ha_msg msg;
13246 
13247 		memset(&msg, 0, sizeof(msg));
13248 		msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13249 		msg.hdr.original_sc = io->io_hdr.original_sc;
13250 		msg.hdr.nexus = io->io_hdr.nexus;
13251 		msg.hdr.status = io->io_hdr.status;
13252 		msg.scsi.scsi_status = io->scsiio.scsi_status;
13253 		msg.scsi.tag_num = io->scsiio.tag_num;
13254 		msg.scsi.tag_type = io->scsiio.tag_type;
13255 		msg.scsi.sense_len = io->scsiio.sense_len;
13256 		msg.scsi.sense_residual = io->scsiio.sense_residual;
13257 		msg.scsi.residual = io->scsiio.residual;
13258 		memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
13259 		       sizeof(io->scsiio.sense_data));
13260 		/*
13261 		 * We copy this whether or not this is an I/O-related
13262 		 * command.  Otherwise, we'd have to go and check to see
13263 		 * whether it's a read/write command, and it really isn't
13264 		 * worth it.
13265 		 */
13266 		memcpy(&msg.scsi.lbalen,
13267 		       &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
13268 		       sizeof(msg.scsi.lbalen));
13269 
13270 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13271 				sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
13272 			/* XXX do something here */
13273 		}
13274 
13275 		ctl_free_io(io);
13276 	} else
13277 		fe_done(io);
13278 
13279 	return (CTL_RETVAL_COMPLETE);
13280 }
13281 
13282 #ifdef CTL_WITH_CA
13283 /*
13284  * Front end should call this if it doesn't do autosense.  When the request
13285  * sense comes back in from the initiator, we'll dequeue this and send it.
13286  */
13287 int
13288 ctl_queue_sense(union ctl_io *io)
13289 {
13290 	struct ctl_lun *lun;
13291 	struct ctl_port *port;
13292 	struct ctl_softc *softc;
13293 	uint32_t initidx, targ_lun;
13294 
13295 	softc = control_softc;
13296 
13297 	CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13298 
13299 	/*
13300 	 * LUN lookup will likely move to the ctl_work_thread() once we
13301 	 * have our new queueing infrastructure (that doesn't put things on
13302 	 * a per-LUN queue initially).  That is so that we can handle
13303 	 * things like an INQUIRY to a LUN that we don't have enabled.  We
13304 	 * can't deal with that right now.
13305 	 */
13306 	mtx_lock(&softc->ctl_lock);
13307 
13308 	/*
13309 	 * If we don't have a LUN for this, just toss the sense
13310 	 * information.
13311 	 */
13312 	port = ctl_io_port(&ctsio->io_hdr);
13313 	targ_lun = ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13314 	if ((targ_lun < CTL_MAX_LUNS)
13315 	 && (softc->ctl_luns[targ_lun] != NULL))
13316 		lun = softc->ctl_luns[targ_lun];
13317 	else
13318 		goto bailout;
13319 
13320 	initidx = ctl_get_initindex(&io->io_hdr.nexus);
13321 
13322 	mtx_lock(&lun->lun_lock);
13323 	/*
13324 	 * Already have CA set for this LUN...toss the sense information.
13325 	 */
13326 	if (ctl_is_set(lun->have_ca, initidx)) {
13327 		mtx_unlock(&lun->lun_lock);
13328 		goto bailout;
13329 	}
13330 
13331 	memcpy(&lun->pending_sense[initidx], &io->scsiio.sense_data,
13332 	       MIN(sizeof(lun->pending_sense[initidx]),
13333 	       sizeof(io->scsiio.sense_data)));
13334 	ctl_set_mask(lun->have_ca, initidx);
13335 	mtx_unlock(&lun->lun_lock);
13336 
13337 bailout:
13338 	mtx_unlock(&softc->ctl_lock);
13339 
13340 	ctl_free_io(io);
13341 
13342 	return (CTL_RETVAL_COMPLETE);
13343 }
13344 #endif
13345 
13346 /*
13347  * Primary command inlet from frontend ports.  All SCSI and task I/O
13348  * requests must go through this function.
13349  */
13350 int
13351 ctl_queue(union ctl_io *io)
13352 {
13353 	struct ctl_port *port;
13354 
13355 	CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13356 
13357 #ifdef CTL_TIME_IO
13358 	io->io_hdr.start_time = time_uptime;
13359 	getbintime(&io->io_hdr.start_bt);
13360 #endif /* CTL_TIME_IO */
13361 
13362 	/* Map FE-specific LUN ID into global one. */
13363 	port = ctl_io_port(&io->io_hdr);
13364 	io->io_hdr.nexus.targ_mapped_lun =
13365 	    ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13366 
13367 	switch (io->io_hdr.io_type) {
13368 	case CTL_IO_SCSI:
13369 	case CTL_IO_TASK:
13370 		if (ctl_debug & CTL_DEBUG_CDB)
13371 			ctl_io_print(io);
13372 		ctl_enqueue_incoming(io);
13373 		break;
13374 	default:
13375 		printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13376 		return (EINVAL);
13377 	}
13378 
13379 	return (CTL_RETVAL_COMPLETE);
13380 }
13381 
13382 #ifdef CTL_IO_DELAY
13383 static void
13384 ctl_done_timer_wakeup(void *arg)
13385 {
13386 	union ctl_io *io;
13387 
13388 	io = (union ctl_io *)arg;
13389 	ctl_done(io);
13390 }
13391 #endif /* CTL_IO_DELAY */
13392 
13393 void
13394 ctl_done(union ctl_io *io)
13395 {
13396 
13397 	/*
13398 	 * Enable this to catch duplicate completion issues.
13399 	 */
13400 #if 0
13401 	if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13402 		printf("%s: type %d msg %d cdb %x iptl: "
13403 		       "%d:%d:%d:%d tag 0x%04x "
13404 		       "flag %#x status %x\n",
13405 			__func__,
13406 			io->io_hdr.io_type,
13407 			io->io_hdr.msg_type,
13408 			io->scsiio.cdb[0],
13409 			io->io_hdr.nexus.initid.id,
13410 			io->io_hdr.nexus.targ_port,
13411 			io->io_hdr.nexus.targ_target.id,
13412 			io->io_hdr.nexus.targ_lun,
13413 			(io->io_hdr.io_type ==
13414 			CTL_IO_TASK) ?
13415 			io->taskio.tag_num :
13416 			io->scsiio.tag_num,
13417 		        io->io_hdr.flags,
13418 			io->io_hdr.status);
13419 	} else
13420 		io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13421 #endif
13422 
13423 	/*
13424 	 * This is an internal copy of an I/O, and should not go through
13425 	 * the normal done processing logic.
13426 	 */
13427 	if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
13428 		return;
13429 
13430 	/*
13431 	 * We need to send a msg to the serializing shelf to finish the IO
13432 	 * as well.  We don't send a finish message to the other shelf if
13433 	 * this is a task management command.  Task management commands
13434 	 * aren't serialized in the OOA queue, but rather just executed on
13435 	 * both shelf controllers for commands that originated on that
13436 	 * controller.
13437 	 */
13438 	if ((io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)
13439 	 && (io->io_hdr.io_type != CTL_IO_TASK)) {
13440 		union ctl_ha_msg msg_io;
13441 
13442 		msg_io.hdr.msg_type = CTL_MSG_FINISH_IO;
13443 		msg_io.hdr.serializing_sc = io->io_hdr.serializing_sc;
13444 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_io,
13445 		    sizeof(msg_io), 0 ) != CTL_HA_STATUS_SUCCESS) {
13446 		}
13447 		/* continue on to finish IO */
13448 	}
13449 #ifdef CTL_IO_DELAY
13450 	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13451 		struct ctl_lun *lun;
13452 
13453 		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13454 
13455 		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13456 	} else {
13457 		struct ctl_lun *lun;
13458 
13459 		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13460 
13461 		if ((lun != NULL)
13462 		 && (lun->delay_info.done_delay > 0)) {
13463 			struct callout *callout;
13464 
13465 			callout = (struct callout *)&io->io_hdr.timer_bytes;
13466 			callout_init(callout, /*mpsafe*/ 1);
13467 			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13468 			callout_reset(callout,
13469 				      lun->delay_info.done_delay * hz,
13470 				      ctl_done_timer_wakeup, io);
13471 			if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13472 				lun->delay_info.done_delay = 0;
13473 			return;
13474 		}
13475 	}
13476 #endif /* CTL_IO_DELAY */
13477 
13478 	ctl_enqueue_done(io);
13479 }
13480 
13481 int
13482 ctl_isc(struct ctl_scsiio *ctsio)
13483 {
13484 	struct ctl_lun *lun;
13485 	int retval;
13486 
13487 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13488 
13489 	CTL_DEBUG_PRINT(("ctl_isc: command: %02x\n", ctsio->cdb[0]));
13490 
13491 	CTL_DEBUG_PRINT(("ctl_isc: calling data_submit()\n"));
13492 
13493 	retval = lun->backend->data_submit((union ctl_io *)ctsio);
13494 
13495 	return (retval);
13496 }
13497 
13498 
13499 static void
13500 ctl_work_thread(void *arg)
13501 {
13502 	struct ctl_thread *thr = (struct ctl_thread *)arg;
13503 	struct ctl_softc *softc = thr->ctl_softc;
13504 	union ctl_io *io;
13505 	int retval;
13506 
13507 	CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13508 
13509 	for (;;) {
13510 		retval = 0;
13511 
13512 		/*
13513 		 * We handle the queues in this order:
13514 		 * - ISC
13515 		 * - done queue (to free up resources, unblock other commands)
13516 		 * - RtR queue
13517 		 * - incoming queue
13518 		 *
13519 		 * If those queues are empty, we break out of the loop and
13520 		 * go to sleep.
13521 		 */
13522 		mtx_lock(&thr->queue_lock);
13523 		io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
13524 		if (io != NULL) {
13525 			STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
13526 			mtx_unlock(&thr->queue_lock);
13527 			ctl_handle_isc(io);
13528 			continue;
13529 		}
13530 		io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
13531 		if (io != NULL) {
13532 			STAILQ_REMOVE_HEAD(&thr->done_queue, links);
13533 			/* clear any blocked commands, call fe_done */
13534 			mtx_unlock(&thr->queue_lock);
13535 			retval = ctl_process_done(io);
13536 			continue;
13537 		}
13538 		io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
13539 		if (io != NULL) {
13540 			STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
13541 			mtx_unlock(&thr->queue_lock);
13542 			if (io->io_hdr.io_type == CTL_IO_TASK)
13543 				ctl_run_task(io);
13544 			else
13545 				ctl_scsiio_precheck(softc, &io->scsiio);
13546 			continue;
13547 		}
13548 		if (!ctl_pause_rtr) {
13549 			io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
13550 			if (io != NULL) {
13551 				STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
13552 				mtx_unlock(&thr->queue_lock);
13553 				retval = ctl_scsiio(&io->scsiio);
13554 				if (retval != CTL_RETVAL_COMPLETE)
13555 					CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13556 				continue;
13557 			}
13558 		}
13559 
13560 		/* Sleep until we have something to do. */
13561 		mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
13562 	}
13563 }
13564 
13565 static void
13566 ctl_lun_thread(void *arg)
13567 {
13568 	struct ctl_softc *softc = (struct ctl_softc *)arg;
13569 	struct ctl_be_lun *be_lun;
13570 	int retval;
13571 
13572 	CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
13573 
13574 	for (;;) {
13575 		retval = 0;
13576 		mtx_lock(&softc->ctl_lock);
13577 		be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
13578 		if (be_lun != NULL) {
13579 			STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
13580 			mtx_unlock(&softc->ctl_lock);
13581 			ctl_create_lun(be_lun);
13582 			continue;
13583 		}
13584 
13585 		/* Sleep until we have something to do. */
13586 		mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
13587 		    PDROP | PRIBIO, "-", 0);
13588 	}
13589 }
13590 
13591 static void
13592 ctl_thresh_thread(void *arg)
13593 {
13594 	struct ctl_softc *softc = (struct ctl_softc *)arg;
13595 	struct ctl_lun *lun;
13596 	struct ctl_be_lun *be_lun;
13597 	struct scsi_da_rw_recovery_page *rwpage;
13598 	struct ctl_logical_block_provisioning_page *page;
13599 	const char *attr;
13600 	uint64_t thres, val;
13601 	int i, e;
13602 
13603 	CTL_DEBUG_PRINT(("ctl_thresh_thread starting\n"));
13604 
13605 	for (;;) {
13606 		mtx_lock(&softc->ctl_lock);
13607 		STAILQ_FOREACH(lun, &softc->lun_list, links) {
13608 			be_lun = lun->be_lun;
13609 			if ((lun->flags & CTL_LUN_DISABLED) ||
13610 			    (lun->flags & CTL_LUN_OFFLINE) ||
13611 			    lun->backend->lun_attr == NULL)
13612 				continue;
13613 			rwpage = &lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT];
13614 			if ((rwpage->byte8 & SMS_RWER_LBPERE) == 0)
13615 				continue;
13616 			e = 0;
13617 			page = &lun->mode_pages.lbp_page[CTL_PAGE_CURRENT];
13618 			for (i = 0; i < CTL_NUM_LBP_THRESH; i++) {
13619 				if ((page->descr[i].flags & SLBPPD_ENABLED) == 0)
13620 					continue;
13621 				thres = scsi_4btoul(page->descr[i].count);
13622 				thres <<= CTL_LBP_EXPONENT;
13623 				switch (page->descr[i].resource) {
13624 				case 0x01:
13625 					attr = "blocksavail";
13626 					break;
13627 				case 0x02:
13628 					attr = "blocksused";
13629 					break;
13630 				case 0xf1:
13631 					attr = "poolblocksavail";
13632 					break;
13633 				case 0xf2:
13634 					attr = "poolblocksused";
13635 					break;
13636 				default:
13637 					continue;
13638 				}
13639 				mtx_unlock(&softc->ctl_lock); // XXX
13640 				val = lun->backend->lun_attr(
13641 				    lun->be_lun->be_lun, attr);
13642 				mtx_lock(&softc->ctl_lock);
13643 				if (val == UINT64_MAX)
13644 					continue;
13645 				if ((page->descr[i].flags & SLBPPD_ARMING_MASK)
13646 				    == SLBPPD_ARMING_INC)
13647 					e |= (val >= thres);
13648 				else
13649 					e |= (val <= thres);
13650 			}
13651 			mtx_lock(&lun->lun_lock);
13652 			if (e) {
13653 				if (lun->lasttpt == 0 ||
13654 				    time_uptime - lun->lasttpt >= CTL_LBP_UA_PERIOD) {
13655 					lun->lasttpt = time_uptime;
13656 					ctl_est_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13657 				}
13658 			} else {
13659 				lun->lasttpt = 0;
13660 				ctl_clr_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13661 			}
13662 			mtx_unlock(&lun->lun_lock);
13663 		}
13664 		mtx_unlock(&softc->ctl_lock);
13665 		pause("-", CTL_LBP_PERIOD * hz);
13666 	}
13667 }
13668 
13669 static void
13670 ctl_enqueue_incoming(union ctl_io *io)
13671 {
13672 	struct ctl_softc *softc = control_softc;
13673 	struct ctl_thread *thr;
13674 	u_int idx;
13675 
13676 	idx = (io->io_hdr.nexus.targ_port * 127 +
13677 	       io->io_hdr.nexus.initid.id) % worker_threads;
13678 	thr = &softc->threads[idx];
13679 	mtx_lock(&thr->queue_lock);
13680 	STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
13681 	mtx_unlock(&thr->queue_lock);
13682 	wakeup(thr);
13683 }
13684 
13685 static void
13686 ctl_enqueue_rtr(union ctl_io *io)
13687 {
13688 	struct ctl_softc *softc = control_softc;
13689 	struct ctl_thread *thr;
13690 
13691 	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13692 	mtx_lock(&thr->queue_lock);
13693 	STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
13694 	mtx_unlock(&thr->queue_lock);
13695 	wakeup(thr);
13696 }
13697 
13698 static void
13699 ctl_enqueue_done(union ctl_io *io)
13700 {
13701 	struct ctl_softc *softc = control_softc;
13702 	struct ctl_thread *thr;
13703 
13704 	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13705 	mtx_lock(&thr->queue_lock);
13706 	STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
13707 	mtx_unlock(&thr->queue_lock);
13708 	wakeup(thr);
13709 }
13710 
13711 #ifdef notyet
13712 static void
13713 ctl_enqueue_isc(union ctl_io *io)
13714 {
13715 	struct ctl_softc *softc = control_softc;
13716 	struct ctl_thread *thr;
13717 
13718 	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13719 	mtx_lock(&thr->queue_lock);
13720 	STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
13721 	mtx_unlock(&thr->queue_lock);
13722 	wakeup(thr);
13723 }
13724 
13725 /* Initialization and failover */
13726 
13727 void
13728 ctl_init_isc_msg(void)
13729 {
13730 	printf("CTL: Still calling this thing\n");
13731 }
13732 
13733 /*
13734  * Init component
13735  * 	Initializes component into configuration defined by bootMode
13736  *	(see hasc-sv.c)
13737  *  	returns hasc_Status:
13738  * 		OK
13739  *		ERROR - fatal error
13740  */
13741 static ctl_ha_comp_status
13742 ctl_isc_init(struct ctl_ha_component *c)
13743 {
13744 	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
13745 
13746 	c->status = ret;
13747 	return ret;
13748 }
13749 
13750 /* Start component
13751  * 	Starts component in state requested. If component starts successfully,
13752  *	it must set its own state to the requestrd state
13753  *	When requested state is HASC_STATE_HA, the component may refine it
13754  * 	by adding _SLAVE or _MASTER flags.
13755  *	Currently allowed state transitions are:
13756  *	UNKNOWN->HA		- initial startup
13757  *	UNKNOWN->SINGLE - initial startup when no parter detected
13758  *	HA->SINGLE		- failover
13759  * returns ctl_ha_comp_status:
13760  * 		OK	- component successfully started in requested state
13761  *		FAILED  - could not start the requested state, failover may
13762  * 			  be possible
13763  *		ERROR	- fatal error detected, no future startup possible
13764  */
13765 static ctl_ha_comp_status
13766 ctl_isc_start(struct ctl_ha_component *c, ctl_ha_state state)
13767 {
13768 	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
13769 
13770 	printf("%s: go\n", __func__);
13771 
13772 	// UNKNOWN->HA or UNKNOWN->SINGLE (bootstrap)
13773 	if (c->state == CTL_HA_STATE_UNKNOWN ) {
13774 		control_softc->is_single = 0;
13775 		if (ctl_ha_msg_create(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
13776 		    != CTL_HA_STATUS_SUCCESS) {
13777 			printf("ctl_isc_start: ctl_ha_msg_create failed.\n");
13778 			ret = CTL_HA_COMP_STATUS_ERROR;
13779 		}
13780 	} else if (CTL_HA_STATE_IS_HA(c->state)
13781 		&& CTL_HA_STATE_IS_SINGLE(state)){
13782 		// HA->SINGLE transition
13783 	        ctl_failover();
13784 		control_softc->is_single = 1;
13785 	} else {
13786 		printf("ctl_isc_start:Invalid state transition %X->%X\n",
13787 		       c->state, state);
13788 		ret = CTL_HA_COMP_STATUS_ERROR;
13789 	}
13790 	if (CTL_HA_STATE_IS_SINGLE(state))
13791 		control_softc->is_single = 1;
13792 
13793 	c->state = state;
13794 	c->status = ret;
13795 	return ret;
13796 }
13797 
13798 /*
13799  * Quiesce component
13800  * The component must clear any error conditions (set status to OK) and
13801  * prepare itself to another Start call
13802  * returns ctl_ha_comp_status:
13803  * 	OK
13804  *	ERROR
13805  */
13806 static ctl_ha_comp_status
13807 ctl_isc_quiesce(struct ctl_ha_component *c)
13808 {
13809 	int ret = CTL_HA_COMP_STATUS_OK;
13810 
13811 	ctl_pause_rtr = 1;
13812 	c->status = ret;
13813 	return ret;
13814 }
13815 
13816 struct ctl_ha_component ctl_ha_component_ctlisc =
13817 {
13818 	.name = "CTL ISC",
13819 	.state = CTL_HA_STATE_UNKNOWN,
13820 	.init = ctl_isc_init,
13821 	.start = ctl_isc_start,
13822 	.quiesce = ctl_isc_quiesce
13823 };
13824 #endif
13825 
13826 /*
13827  *  vim: ts=8
13828  */
13829