xref: /freebsd/sys/cam/ctl/ctl.c (revision bbb29a3c0f2c4565eff6fda70426807b6ed97f8b)
1 /*-
2  * Copyright (c) 2003-2009 Silicon Graphics International Corp.
3  * Copyright (c) 2012 The FreeBSD Foundation
4  * All rights reserved.
5  *
6  * Portions of this software were developed by Edward Tomasz Napierala
7  * under sponsorship from the FreeBSD Foundation.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions, and the following disclaimer,
14  *    without modification.
15  * 2. Redistributions in binary form must reproduce at minimum a disclaimer
16  *    substantially similar to the "NO WARRANTY" disclaimer below
17  *    ("Disclaimer") and any redistribution must be conditioned upon
18  *    including a substantially similar Disclaimer requirement for further
19  *    binary redistribution.
20  *
21  * NO WARRANTY
22  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
23  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
24  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR
25  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
26  * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
30  * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
31  * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
32  * POSSIBILITY OF SUCH DAMAGES.
33  *
34  * $Id: //depot/users/kenm/FreeBSD-test2/sys/cam/ctl/ctl.c#8 $
35  */
36 /*
37  * CAM Target Layer, a SCSI device emulation subsystem.
38  *
39  * Author: Ken Merry <ken@FreeBSD.org>
40  */
41 
42 #define _CTL_C
43 
44 #include <sys/cdefs.h>
45 __FBSDID("$FreeBSD$");
46 
47 #include <sys/param.h>
48 #include <sys/systm.h>
49 #include <sys/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_frontend_internal.h>
76 #include <cam/ctl/ctl_util.h>
77 #include <cam/ctl/ctl_backend.h>
78 #include <cam/ctl/ctl_ioctl.h>
79 #include <cam/ctl/ctl_ha.h>
80 #include <cam/ctl/ctl_private.h>
81 #include <cam/ctl/ctl_debug.h>
82 #include <cam/ctl/ctl_scsi_all.h>
83 #include <cam/ctl/ctl_error.h>
84 
85 struct ctl_softc *control_softc = NULL;
86 
87 /*
88  * Size and alignment macros needed for Copan-specific HA hardware.  These
89  * can go away when the HA code is re-written, and uses busdma for any
90  * hardware.
91  */
92 #define	CTL_ALIGN_8B(target, source, type)				\
93 	if (((uint32_t)source & 0x7) != 0)				\
94 		target = (type)(source + (0x8 - ((uint32_t)source & 0x7)));\
95 	else								\
96 		target = (type)source;
97 
98 #define	CTL_SIZE_8B(target, size)					\
99 	if ((size & 0x7) != 0)						\
100 		target = size + (0x8 - (size & 0x7));			\
101 	else								\
102 		target = size;
103 
104 #define CTL_ALIGN_8B_MARGIN	16
105 
106 /*
107  * Template mode pages.
108  */
109 
110 /*
111  * Note that these are default values only.  The actual values will be
112  * filled in when the user does a mode sense.
113  */
114 static struct copan_debugconf_subpage debugconf_page_default = {
115 	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
116 	DBGCNF_SUBPAGE_CODE,		/* subpage */
117 	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
118 	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
119 	DBGCNF_VERSION,			/* page_version */
120 	{CTL_TIME_IO_DEFAULT_SECS>>8,
121 	 CTL_TIME_IO_DEFAULT_SECS>>0},	/* ctl_time_io_secs */
122 };
123 
124 static struct copan_debugconf_subpage debugconf_page_changeable = {
125 	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
126 	DBGCNF_SUBPAGE_CODE,		/* subpage */
127 	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
128 	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
129 	0,				/* page_version */
130 	{0xff,0xff},			/* ctl_time_io_secs */
131 };
132 
133 static struct scsi_da_rw_recovery_page rw_er_page_default = {
134 	/*page_code*/SMS_RW_ERROR_RECOVERY_PAGE,
135 	/*page_length*/sizeof(struct scsi_da_rw_recovery_page) - 2,
136 	/*byte3*/SMS_RWER_AWRE|SMS_RWER_ARRE,
137 	/*read_retry_count*/0,
138 	/*correction_span*/0,
139 	/*head_offset_count*/0,
140 	/*data_strobe_offset_cnt*/0,
141 	/*byte8*/SMS_RWER_LBPERE,
142 	/*write_retry_count*/0,
143 	/*reserved2*/0,
144 	/*recovery_time_limit*/{0, 0},
145 };
146 
147 static struct scsi_da_rw_recovery_page rw_er_page_changeable = {
148 	/*page_code*/SMS_RW_ERROR_RECOVERY_PAGE,
149 	/*page_length*/sizeof(struct scsi_da_rw_recovery_page) - 2,
150 	/*byte3*/0,
151 	/*read_retry_count*/0,
152 	/*correction_span*/0,
153 	/*head_offset_count*/0,
154 	/*data_strobe_offset_cnt*/0,
155 	/*byte8*/0,
156 	/*write_retry_count*/0,
157 	/*reserved2*/0,
158 	/*recovery_time_limit*/{0, 0},
159 };
160 
161 static struct scsi_format_page format_page_default = {
162 	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
163 	/*page_length*/sizeof(struct scsi_format_page) - 2,
164 	/*tracks_per_zone*/ {0, 0},
165 	/*alt_sectors_per_zone*/ {0, 0},
166 	/*alt_tracks_per_zone*/ {0, 0},
167 	/*alt_tracks_per_lun*/ {0, 0},
168 	/*sectors_per_track*/ {(CTL_DEFAULT_SECTORS_PER_TRACK >> 8) & 0xff,
169 			        CTL_DEFAULT_SECTORS_PER_TRACK & 0xff},
170 	/*bytes_per_sector*/ {0, 0},
171 	/*interleave*/ {0, 0},
172 	/*track_skew*/ {0, 0},
173 	/*cylinder_skew*/ {0, 0},
174 	/*flags*/ SFP_HSEC,
175 	/*reserved*/ {0, 0, 0}
176 };
177 
178 static struct scsi_format_page format_page_changeable = {
179 	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
180 	/*page_length*/sizeof(struct scsi_format_page) - 2,
181 	/*tracks_per_zone*/ {0, 0},
182 	/*alt_sectors_per_zone*/ {0, 0},
183 	/*alt_tracks_per_zone*/ {0, 0},
184 	/*alt_tracks_per_lun*/ {0, 0},
185 	/*sectors_per_track*/ {0, 0},
186 	/*bytes_per_sector*/ {0, 0},
187 	/*interleave*/ {0, 0},
188 	/*track_skew*/ {0, 0},
189 	/*cylinder_skew*/ {0, 0},
190 	/*flags*/ 0,
191 	/*reserved*/ {0, 0, 0}
192 };
193 
194 static struct scsi_rigid_disk_page rigid_disk_page_default = {
195 	/*page_code*/SMS_RIGID_DISK_PAGE,
196 	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
197 	/*cylinders*/ {0, 0, 0},
198 	/*heads*/ CTL_DEFAULT_HEADS,
199 	/*start_write_precomp*/ {0, 0, 0},
200 	/*start_reduced_current*/ {0, 0, 0},
201 	/*step_rate*/ {0, 0},
202 	/*landing_zone_cylinder*/ {0, 0, 0},
203 	/*rpl*/ SRDP_RPL_DISABLED,
204 	/*rotational_offset*/ 0,
205 	/*reserved1*/ 0,
206 	/*rotation_rate*/ {(CTL_DEFAULT_ROTATION_RATE >> 8) & 0xff,
207 			   CTL_DEFAULT_ROTATION_RATE & 0xff},
208 	/*reserved2*/ {0, 0}
209 };
210 
211 static struct scsi_rigid_disk_page rigid_disk_page_changeable = {
212 	/*page_code*/SMS_RIGID_DISK_PAGE,
213 	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
214 	/*cylinders*/ {0, 0, 0},
215 	/*heads*/ 0,
216 	/*start_write_precomp*/ {0, 0, 0},
217 	/*start_reduced_current*/ {0, 0, 0},
218 	/*step_rate*/ {0, 0},
219 	/*landing_zone_cylinder*/ {0, 0, 0},
220 	/*rpl*/ 0,
221 	/*rotational_offset*/ 0,
222 	/*reserved1*/ 0,
223 	/*rotation_rate*/ {0, 0},
224 	/*reserved2*/ {0, 0}
225 };
226 
227 static struct scsi_caching_page caching_page_default = {
228 	/*page_code*/SMS_CACHING_PAGE,
229 	/*page_length*/sizeof(struct scsi_caching_page) - 2,
230 	/*flags1*/ SCP_DISC | SCP_WCE,
231 	/*ret_priority*/ 0,
232 	/*disable_pf_transfer_len*/ {0xff, 0xff},
233 	/*min_prefetch*/ {0, 0},
234 	/*max_prefetch*/ {0xff, 0xff},
235 	/*max_pf_ceiling*/ {0xff, 0xff},
236 	/*flags2*/ 0,
237 	/*cache_segments*/ 0,
238 	/*cache_seg_size*/ {0, 0},
239 	/*reserved*/ 0,
240 	/*non_cache_seg_size*/ {0, 0, 0}
241 };
242 
243 static struct scsi_caching_page caching_page_changeable = {
244 	/*page_code*/SMS_CACHING_PAGE,
245 	/*page_length*/sizeof(struct scsi_caching_page) - 2,
246 	/*flags1*/ SCP_WCE | SCP_RCD,
247 	/*ret_priority*/ 0,
248 	/*disable_pf_transfer_len*/ {0, 0},
249 	/*min_prefetch*/ {0, 0},
250 	/*max_prefetch*/ {0, 0},
251 	/*max_pf_ceiling*/ {0, 0},
252 	/*flags2*/ 0,
253 	/*cache_segments*/ 0,
254 	/*cache_seg_size*/ {0, 0},
255 	/*reserved*/ 0,
256 	/*non_cache_seg_size*/ {0, 0, 0}
257 };
258 
259 static struct scsi_control_page control_page_default = {
260 	/*page_code*/SMS_CONTROL_MODE_PAGE,
261 	/*page_length*/sizeof(struct scsi_control_page) - 2,
262 	/*rlec*/0,
263 	/*queue_flags*/SCP_QUEUE_ALG_RESTRICTED,
264 	/*eca_and_aen*/0,
265 	/*flags4*/SCP_TAS,
266 	/*aen_holdoff_period*/{0, 0},
267 	/*busy_timeout_period*/{0, 0},
268 	/*extended_selftest_completion_time*/{0, 0}
269 };
270 
271 static struct scsi_control_page control_page_changeable = {
272 	/*page_code*/SMS_CONTROL_MODE_PAGE,
273 	/*page_length*/sizeof(struct scsi_control_page) - 2,
274 	/*rlec*/SCP_DSENSE,
275 	/*queue_flags*/SCP_QUEUE_ALG_MASK,
276 	/*eca_and_aen*/SCP_SWP,
277 	/*flags4*/0,
278 	/*aen_holdoff_period*/{0, 0},
279 	/*busy_timeout_period*/{0, 0},
280 	/*extended_selftest_completion_time*/{0, 0}
281 };
282 
283 static struct scsi_info_exceptions_page ie_page_default = {
284 	/*page_code*/SMS_INFO_EXCEPTIONS_PAGE,
285 	/*page_length*/sizeof(struct scsi_info_exceptions_page) - 2,
286 	/*info_flags*/SIEP_FLAGS_DEXCPT,
287 	/*mrie*/0,
288 	/*interval_timer*/{0, 0, 0, 0},
289 	/*report_count*/{0, 0, 0, 0}
290 };
291 
292 static struct scsi_info_exceptions_page ie_page_changeable = {
293 	/*page_code*/SMS_INFO_EXCEPTIONS_PAGE,
294 	/*page_length*/sizeof(struct scsi_info_exceptions_page) - 2,
295 	/*info_flags*/0,
296 	/*mrie*/0,
297 	/*interval_timer*/{0, 0, 0, 0},
298 	/*report_count*/{0, 0, 0, 0}
299 };
300 
301 #define CTL_LBPM_LEN	(sizeof(struct ctl_logical_block_provisioning_page) - 4)
302 
303 static struct ctl_logical_block_provisioning_page lbp_page_default = {{
304 	/*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF,
305 	/*subpage_code*/0x02,
306 	/*page_length*/{CTL_LBPM_LEN >> 8, CTL_LBPM_LEN},
307 	/*flags*/0,
308 	/*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
309 	/*descr*/{}},
310 	{{/*flags*/0,
311 	  /*resource*/0x01,
312 	  /*reserved*/{0, 0},
313 	  /*count*/{0, 0, 0, 0}},
314 	 {/*flags*/0,
315 	  /*resource*/0x02,
316 	  /*reserved*/{0, 0},
317 	  /*count*/{0, 0, 0, 0}},
318 	 {/*flags*/0,
319 	  /*resource*/0xf1,
320 	  /*reserved*/{0, 0},
321 	  /*count*/{0, 0, 0, 0}},
322 	 {/*flags*/0,
323 	  /*resource*/0xf2,
324 	  /*reserved*/{0, 0},
325 	  /*count*/{0, 0, 0, 0}}
326 	}
327 };
328 
329 static struct ctl_logical_block_provisioning_page lbp_page_changeable = {{
330 	/*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF,
331 	/*subpage_code*/0x02,
332 	/*page_length*/{CTL_LBPM_LEN >> 8, CTL_LBPM_LEN},
333 	/*flags*/0,
334 	/*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
335 	/*descr*/{}},
336 	{{/*flags*/0,
337 	  /*resource*/0,
338 	  /*reserved*/{0, 0},
339 	  /*count*/{0, 0, 0, 0}},
340 	 {/*flags*/0,
341 	  /*resource*/0,
342 	  /*reserved*/{0, 0},
343 	  /*count*/{0, 0, 0, 0}},
344 	 {/*flags*/0,
345 	  /*resource*/0,
346 	  /*reserved*/{0, 0},
347 	  /*count*/{0, 0, 0, 0}},
348 	 {/*flags*/0,
349 	  /*resource*/0,
350 	  /*reserved*/{0, 0},
351 	  /*count*/{0, 0, 0, 0}}
352 	}
353 };
354 
355 /*
356  * XXX KDM move these into the softc.
357  */
358 static int rcv_sync_msg;
359 static uint8_t ctl_pause_rtr;
360 
361 SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD, 0, "CAM Target Layer");
362 static int worker_threads = -1;
363 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, worker_threads, CTLFLAG_RDTUN,
364     &worker_threads, 1, "Number of worker threads");
365 static int ctl_debug = CTL_DEBUG_NONE;
366 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, debug, CTLFLAG_RWTUN,
367     &ctl_debug, 0, "Enabled debug flags");
368 
369 /*
370  * Supported pages (0x00), Serial number (0x80), Device ID (0x83),
371  * Extended INQUIRY Data (0x86), Mode Page Policy (0x87),
372  * SCSI Ports (0x88), Third-party Copy (0x8F), Block limits (0xB0),
373  * Block Device Characteristics (0xB1) and Logical Block Provisioning (0xB2)
374  */
375 #define SCSI_EVPD_NUM_SUPPORTED_PAGES	10
376 
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 static int ctl_init(void);
381 void ctl_shutdown(void);
382 static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
383 static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
384 static void ctl_ioctl_online(void *arg);
385 static void ctl_ioctl_offline(void *arg);
386 static int ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id);
387 static int ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id);
388 static int ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio);
389 static int ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio);
390 static int ctl_ioctl_submit_wait(union ctl_io *io);
391 static void ctl_ioctl_datamove(union ctl_io *io);
392 static void ctl_ioctl_done(union ctl_io *io);
393 static void ctl_ioctl_hard_startstop_callback(void *arg,
394 					      struct cfi_metatask *metatask);
395 static void ctl_ioctl_bbrread_callback(void *arg,struct cfi_metatask *metatask);
396 static int ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
397 			      struct ctl_ooa *ooa_hdr,
398 			      struct ctl_ooa_entry *kern_entries);
399 static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
400 		     struct thread *td);
401 static uint32_t ctl_map_lun(int port_num, uint32_t lun);
402 static uint32_t ctl_map_lun_back(int port_num, uint32_t lun);
403 static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
404 			 struct ctl_be_lun *be_lun, struct ctl_id target_id);
405 static int ctl_free_lun(struct ctl_lun *lun);
406 static void ctl_create_lun(struct ctl_be_lun *be_lun);
407 /**
408 static void ctl_failover_change_pages(struct ctl_softc *softc,
409 				      struct ctl_scsiio *ctsio, int master);
410 **/
411 
412 static int ctl_do_mode_select(union ctl_io *io);
413 static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
414 			   uint64_t res_key, uint64_t sa_res_key,
415 			   uint8_t type, uint32_t residx,
416 			   struct ctl_scsiio *ctsio,
417 			   struct scsi_per_res_out *cdb,
418 			   struct scsi_per_res_out_parms* param);
419 static void ctl_pro_preempt_other(struct ctl_lun *lun,
420 				  union ctl_ha_msg *msg);
421 static void ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg);
422 static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
423 static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
424 static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
425 static int ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len);
426 static int ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len);
427 static int ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio,
428 					 int alloc_len);
429 static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
430 					 int alloc_len);
431 static int ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len);
432 static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
433 static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
434 static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
435 static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len);
436 static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2);
437 static ctl_action ctl_check_for_blockage(struct ctl_lun *lun,
438     union ctl_io *pending_io, union ctl_io *ooa_io);
439 static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
440 				union ctl_io *starting_io);
441 static int ctl_check_blocked(struct ctl_lun *lun);
442 static int ctl_scsiio_lun_check(struct ctl_softc *ctl_softc,
443 				struct ctl_lun *lun,
444 				const struct ctl_cmd_entry *entry,
445 				struct ctl_scsiio *ctsio);
446 //static int ctl_check_rtr(union ctl_io *pending_io, struct ctl_softc *softc);
447 static void ctl_failover(void);
448 static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
449 			       struct ctl_scsiio *ctsio);
450 static int ctl_scsiio(struct ctl_scsiio *ctsio);
451 
452 static int ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
453 static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
454 			    ctl_ua_type ua_type);
455 static int ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io,
456 			 ctl_ua_type ua_type);
457 static int ctl_abort_task(union ctl_io *io);
458 static int ctl_abort_task_set(union ctl_io *io);
459 static int ctl_i_t_nexus_reset(union ctl_io *io);
460 static void ctl_run_task(union ctl_io *io);
461 #ifdef CTL_IO_DELAY
462 static void ctl_datamove_timer_wakeup(void *arg);
463 static void ctl_done_timer_wakeup(void *arg);
464 #endif /* CTL_IO_DELAY */
465 
466 static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
467 static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
468 static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
469 static void ctl_datamove_remote_write(union ctl_io *io);
470 static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
471 static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
472 static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
473 static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
474 				    ctl_ha_dt_cb callback);
475 static void ctl_datamove_remote_read(union ctl_io *io);
476 static void ctl_datamove_remote(union ctl_io *io);
477 static int ctl_process_done(union ctl_io *io);
478 static void ctl_lun_thread(void *arg);
479 static void ctl_thresh_thread(void *arg);
480 static void ctl_work_thread(void *arg);
481 static void ctl_enqueue_incoming(union ctl_io *io);
482 static void ctl_enqueue_rtr(union ctl_io *io);
483 static void ctl_enqueue_done(union ctl_io *io);
484 static void ctl_enqueue_isc(union ctl_io *io);
485 static const struct ctl_cmd_entry *
486     ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa);
487 static const struct ctl_cmd_entry *
488     ctl_validate_command(struct ctl_scsiio *ctsio);
489 static int ctl_cmd_applicable(uint8_t lun_type,
490     const struct ctl_cmd_entry *entry);
491 
492 /*
493  * Load the serialization table.  This isn't very pretty, but is probably
494  * the easiest way to do it.
495  */
496 #include "ctl_ser_table.c"
497 
498 /*
499  * We only need to define open, close and ioctl routines for this driver.
500  */
501 static struct cdevsw ctl_cdevsw = {
502 	.d_version =	D_VERSION,
503 	.d_flags =	0,
504 	.d_open =	ctl_open,
505 	.d_close =	ctl_close,
506 	.d_ioctl =	ctl_ioctl,
507 	.d_name =	"ctl",
508 };
509 
510 
511 MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
512 MALLOC_DEFINE(M_CTLIO, "ctlio", "Memory used for CTL requests");
513 
514 static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
515 
516 static moduledata_t ctl_moduledata = {
517 	"ctl",
518 	ctl_module_event_handler,
519 	NULL
520 };
521 
522 DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
523 MODULE_VERSION(ctl, 1);
524 
525 static struct ctl_frontend ioctl_frontend =
526 {
527 	.name = "ioctl",
528 };
529 
530 static void
531 ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
532 			    union ctl_ha_msg *msg_info)
533 {
534 	struct ctl_scsiio *ctsio;
535 
536 	if (msg_info->hdr.original_sc == NULL) {
537 		printf("%s: original_sc == NULL!\n", __func__);
538 		/* XXX KDM now what? */
539 		return;
540 	}
541 
542 	ctsio = &msg_info->hdr.original_sc->scsiio;
543 	ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
544 	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
545 	ctsio->io_hdr.status = msg_info->hdr.status;
546 	ctsio->scsi_status = msg_info->scsi.scsi_status;
547 	ctsio->sense_len = msg_info->scsi.sense_len;
548 	ctsio->sense_residual = msg_info->scsi.sense_residual;
549 	ctsio->residual = msg_info->scsi.residual;
550 	memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
551 	       sizeof(ctsio->sense_data));
552 	memcpy(&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
553 	       &msg_info->scsi.lbalen, sizeof(msg_info->scsi.lbalen));
554 	ctl_enqueue_isc((union ctl_io *)ctsio);
555 }
556 
557 static void
558 ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
559 				union ctl_ha_msg *msg_info)
560 {
561 	struct ctl_scsiio *ctsio;
562 
563 	if (msg_info->hdr.serializing_sc == NULL) {
564 		printf("%s: serializing_sc == NULL!\n", __func__);
565 		/* XXX KDM now what? */
566 		return;
567 	}
568 
569 	ctsio = &msg_info->hdr.serializing_sc->scsiio;
570 #if 0
571 	/*
572 	 * Attempt to catch the situation where an I/O has
573 	 * been freed, and we're using it again.
574 	 */
575 	if (ctsio->io_hdr.io_type == 0xff) {
576 		union ctl_io *tmp_io;
577 		tmp_io = (union ctl_io *)ctsio;
578 		printf("%s: %p use after free!\n", __func__,
579 		       ctsio);
580 		printf("%s: type %d msg %d cdb %x iptl: "
581 		       "%d:%d:%d:%d tag 0x%04x "
582 		       "flag %#x status %x\n",
583 			__func__,
584 			tmp_io->io_hdr.io_type,
585 			tmp_io->io_hdr.msg_type,
586 			tmp_io->scsiio.cdb[0],
587 			tmp_io->io_hdr.nexus.initid.id,
588 			tmp_io->io_hdr.nexus.targ_port,
589 			tmp_io->io_hdr.nexus.targ_target.id,
590 			tmp_io->io_hdr.nexus.targ_lun,
591 			(tmp_io->io_hdr.io_type ==
592 			CTL_IO_TASK) ?
593 			tmp_io->taskio.tag_num :
594 			tmp_io->scsiio.tag_num,
595 		        tmp_io->io_hdr.flags,
596 			tmp_io->io_hdr.status);
597 	}
598 #endif
599 	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
600 	ctl_enqueue_isc((union ctl_io *)ctsio);
601 }
602 
603 /*
604  * ISC (Inter Shelf Communication) event handler.  Events from the HA
605  * subsystem come in here.
606  */
607 static void
608 ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
609 {
610 	struct ctl_softc *ctl_softc;
611 	union ctl_io *io;
612 	struct ctl_prio *presio;
613 	ctl_ha_status isc_status;
614 
615 	ctl_softc = control_softc;
616 	io = NULL;
617 
618 
619 #if 0
620 	printf("CTL: Isc Msg event %d\n", event);
621 #endif
622 	if (event == CTL_HA_EVT_MSG_RECV) {
623 		union ctl_ha_msg msg_info;
624 
625 		isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
626 					     sizeof(msg_info), /*wait*/ 0);
627 #if 0
628 		printf("CTL: msg_type %d\n", msg_info.msg_type);
629 #endif
630 		if (isc_status != 0) {
631 			printf("Error receiving message, status = %d\n",
632 			       isc_status);
633 			return;
634 		}
635 
636 		switch (msg_info.hdr.msg_type) {
637 		case CTL_MSG_SERIALIZE:
638 #if 0
639 			printf("Serialize\n");
640 #endif
641 			io = ctl_alloc_io_nowait(ctl_softc->othersc_pool);
642 			if (io == NULL) {
643 				printf("ctl_isc_event_handler: can't allocate "
644 				       "ctl_io!\n");
645 				/* Bad Juju */
646 				/* Need to set busy and send msg back */
647 				msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
648 				msg_info.hdr.status = CTL_SCSI_ERROR;
649 				msg_info.scsi.scsi_status = SCSI_STATUS_BUSY;
650 				msg_info.scsi.sense_len = 0;
651 			        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
652 				    sizeof(msg_info), 0) > CTL_HA_STATUS_SUCCESS){
653 				}
654 				goto bailout;
655 			}
656 			ctl_zero_io(io);
657 			// populate ctsio from msg_info
658 			io->io_hdr.io_type = CTL_IO_SCSI;
659 			io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
660 			io->io_hdr.original_sc = msg_info.hdr.original_sc;
661 #if 0
662 			printf("pOrig %x\n", (int)msg_info.original_sc);
663 #endif
664 			io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
665 					    CTL_FLAG_IO_ACTIVE;
666 			/*
667 			 * If we're in serialization-only mode, we don't
668 			 * want to go through full done processing.  Thus
669 			 * the COPY flag.
670 			 *
671 			 * XXX KDM add another flag that is more specific.
672 			 */
673 			if (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)
674 				io->io_hdr.flags |= CTL_FLAG_INT_COPY;
675 			io->io_hdr.nexus = msg_info.hdr.nexus;
676 #if 0
677 			printf("targ %d, port %d, iid %d, lun %d\n",
678 			       io->io_hdr.nexus.targ_target.id,
679 			       io->io_hdr.nexus.targ_port,
680 			       io->io_hdr.nexus.initid.id,
681 			       io->io_hdr.nexus.targ_lun);
682 #endif
683 			io->scsiio.tag_num = msg_info.scsi.tag_num;
684 			io->scsiio.tag_type = msg_info.scsi.tag_type;
685 			memcpy(io->scsiio.cdb, msg_info.scsi.cdb,
686 			       CTL_MAX_CDBLEN);
687 			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
688 				const struct ctl_cmd_entry *entry;
689 
690 				entry = ctl_get_cmd_entry(&io->scsiio, NULL);
691 				io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
692 				io->io_hdr.flags |=
693 					entry->flags & CTL_FLAG_DATA_MASK;
694 			}
695 			ctl_enqueue_isc(io);
696 			break;
697 
698 		/* Performed on the Originating SC, XFER mode only */
699 		case CTL_MSG_DATAMOVE: {
700 			struct ctl_sg_entry *sgl;
701 			int i, j;
702 
703 			io = msg_info.hdr.original_sc;
704 			if (io == NULL) {
705 				printf("%s: original_sc == NULL!\n", __func__);
706 				/* XXX KDM do something here */
707 				break;
708 			}
709 			io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
710 			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
711 			/*
712 			 * Keep track of this, we need to send it back over
713 			 * when the datamove is complete.
714 			 */
715 			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
716 
717 			if (msg_info.dt.sg_sequence == 0) {
718 				/*
719 				 * XXX KDM we use the preallocated S/G list
720 				 * here, but we'll need to change this to
721 				 * dynamic allocation if we need larger S/G
722 				 * lists.
723 				 */
724 				if (msg_info.dt.kern_sg_entries >
725 				    sizeof(io->io_hdr.remote_sglist) /
726 				    sizeof(io->io_hdr.remote_sglist[0])) {
727 					printf("%s: number of S/G entries "
728 					    "needed %u > allocated num %zd\n",
729 					    __func__,
730 					    msg_info.dt.kern_sg_entries,
731 					    sizeof(io->io_hdr.remote_sglist)/
732 					    sizeof(io->io_hdr.remote_sglist[0]));
733 
734 					/*
735 					 * XXX KDM send a message back to
736 					 * the other side to shut down the
737 					 * DMA.  The error will come back
738 					 * through via the normal channel.
739 					 */
740 					break;
741 				}
742 				sgl = io->io_hdr.remote_sglist;
743 				memset(sgl, 0,
744 				       sizeof(io->io_hdr.remote_sglist));
745 
746 				io->scsiio.kern_data_ptr = (uint8_t *)sgl;
747 
748 				io->scsiio.kern_sg_entries =
749 					msg_info.dt.kern_sg_entries;
750 				io->scsiio.rem_sg_entries =
751 					msg_info.dt.kern_sg_entries;
752 				io->scsiio.kern_data_len =
753 					msg_info.dt.kern_data_len;
754 				io->scsiio.kern_total_len =
755 					msg_info.dt.kern_total_len;
756 				io->scsiio.kern_data_resid =
757 					msg_info.dt.kern_data_resid;
758 				io->scsiio.kern_rel_offset =
759 					msg_info.dt.kern_rel_offset;
760 				/*
761 				 * Clear out per-DMA flags.
762 				 */
763 				io->io_hdr.flags &= ~CTL_FLAG_RDMA_MASK;
764 				/*
765 				 * Add per-DMA flags that are set for this
766 				 * particular DMA request.
767 				 */
768 				io->io_hdr.flags |= msg_info.dt.flags &
769 						    CTL_FLAG_RDMA_MASK;
770 			} else
771 				sgl = (struct ctl_sg_entry *)
772 					io->scsiio.kern_data_ptr;
773 
774 			for (i = msg_info.dt.sent_sg_entries, j = 0;
775 			     i < (msg_info.dt.sent_sg_entries +
776 			     msg_info.dt.cur_sg_entries); i++, j++) {
777 				sgl[i].addr = msg_info.dt.sg_list[j].addr;
778 				sgl[i].len = msg_info.dt.sg_list[j].len;
779 
780 #if 0
781 				printf("%s: L: %p,%d -> %p,%d j=%d, i=%d\n",
782 				       __func__,
783 				       msg_info.dt.sg_list[j].addr,
784 				       msg_info.dt.sg_list[j].len,
785 				       sgl[i].addr, sgl[i].len, j, i);
786 #endif
787 			}
788 #if 0
789 			memcpy(&sgl[msg_info.dt.sent_sg_entries],
790 			       msg_info.dt.sg_list,
791 			       sizeof(*sgl) * msg_info.dt.cur_sg_entries);
792 #endif
793 
794 			/*
795 			 * If this is the last piece of the I/O, we've got
796 			 * the full S/G list.  Queue processing in the thread.
797 			 * Otherwise wait for the next piece.
798 			 */
799 			if (msg_info.dt.sg_last != 0)
800 				ctl_enqueue_isc(io);
801 			break;
802 		}
803 		/* Performed on the Serializing (primary) SC, XFER mode only */
804 		case CTL_MSG_DATAMOVE_DONE: {
805 			if (msg_info.hdr.serializing_sc == NULL) {
806 				printf("%s: serializing_sc == NULL!\n",
807 				       __func__);
808 				/* XXX KDM now what? */
809 				break;
810 			}
811 			/*
812 			 * We grab the sense information here in case
813 			 * there was a failure, so we can return status
814 			 * back to the initiator.
815 			 */
816 			io = msg_info.hdr.serializing_sc;
817 			io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
818 			io->io_hdr.status = msg_info.hdr.status;
819 			io->scsiio.scsi_status = msg_info.scsi.scsi_status;
820 			io->scsiio.sense_len = msg_info.scsi.sense_len;
821 			io->scsiio.sense_residual =msg_info.scsi.sense_residual;
822 			io->io_hdr.port_status = msg_info.scsi.fetd_status;
823 			io->scsiio.residual = msg_info.scsi.residual;
824 			memcpy(&io->scsiio.sense_data,&msg_info.scsi.sense_data,
825 			       sizeof(io->scsiio.sense_data));
826 			ctl_enqueue_isc(io);
827 			break;
828 		}
829 
830 		/* Preformed on Originating SC, SER_ONLY mode */
831 		case CTL_MSG_R2R:
832 			io = msg_info.hdr.original_sc;
833 			if (io == NULL) {
834 				printf("%s: Major Bummer\n", __func__);
835 				return;
836 			} else {
837 #if 0
838 				printf("pOrig %x\n",(int) ctsio);
839 #endif
840 			}
841 			io->io_hdr.msg_type = CTL_MSG_R2R;
842 			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
843 			ctl_enqueue_isc(io);
844 			break;
845 
846 		/*
847 		 * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
848 		 * mode.
849 		 * Performed on the Originating (i.e. secondary) SC in XFER
850 		 * mode
851 		 */
852 		case CTL_MSG_FINISH_IO:
853 			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER)
854 				ctl_isc_handler_finish_xfer(ctl_softc,
855 							    &msg_info);
856 			else
857 				ctl_isc_handler_finish_ser_only(ctl_softc,
858 								&msg_info);
859 			break;
860 
861 		/* Preformed on Originating SC */
862 		case CTL_MSG_BAD_JUJU:
863 			io = msg_info.hdr.original_sc;
864 			if (io == NULL) {
865 				printf("%s: Bad JUJU!, original_sc is NULL!\n",
866 				       __func__);
867 				break;
868 			}
869 			ctl_copy_sense_data(&msg_info, io);
870 			/*
871 			 * IO should have already been cleaned up on other
872 			 * SC so clear this flag so we won't send a message
873 			 * back to finish the IO there.
874 			 */
875 			io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
876 			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
877 
878 			/* io = msg_info.hdr.serializing_sc; */
879 			io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
880 			ctl_enqueue_isc(io);
881 			break;
882 
883 		/* Handle resets sent from the other side */
884 		case CTL_MSG_MANAGE_TASKS: {
885 			struct ctl_taskio *taskio;
886 			taskio = (struct ctl_taskio *)ctl_alloc_io_nowait(
887 			    ctl_softc->othersc_pool);
888 			if (taskio == NULL) {
889 				printf("ctl_isc_event_handler: can't allocate "
890 				       "ctl_io!\n");
891 				/* Bad Juju */
892 				/* should I just call the proper reset func
893 				   here??? */
894 				goto bailout;
895 			}
896 			ctl_zero_io((union ctl_io *)taskio);
897 			taskio->io_hdr.io_type = CTL_IO_TASK;
898 			taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
899 			taskio->io_hdr.nexus = msg_info.hdr.nexus;
900 			taskio->task_action = msg_info.task.task_action;
901 			taskio->tag_num = msg_info.task.tag_num;
902 			taskio->tag_type = msg_info.task.tag_type;
903 #ifdef CTL_TIME_IO
904 			taskio->io_hdr.start_time = time_uptime;
905 			getbintime(&taskio->io_hdr.start_bt);
906 #if 0
907 			cs_prof_gettime(&taskio->io_hdr.start_ticks);
908 #endif
909 #endif /* CTL_TIME_IO */
910 			ctl_run_task((union ctl_io *)taskio);
911 			break;
912 		}
913 		/* Persistent Reserve action which needs attention */
914 		case CTL_MSG_PERS_ACTION:
915 			presio = (struct ctl_prio *)ctl_alloc_io_nowait(
916 			    ctl_softc->othersc_pool);
917 			if (presio == NULL) {
918 				printf("ctl_isc_event_handler: can't allocate "
919 				       "ctl_io!\n");
920 				/* Bad Juju */
921 				/* Need to set busy and send msg back */
922 				goto bailout;
923 			}
924 			ctl_zero_io((union ctl_io *)presio);
925 			presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
926 			presio->pr_msg = msg_info.pr;
927 			ctl_enqueue_isc((union ctl_io *)presio);
928 			break;
929 		case CTL_MSG_SYNC_FE:
930 			rcv_sync_msg = 1;
931 			break;
932 		default:
933 		        printf("How did I get here?\n");
934 		}
935 	} else if (event == CTL_HA_EVT_MSG_SENT) {
936 		if (param != CTL_HA_STATUS_SUCCESS) {
937 			printf("Bad status from ctl_ha_msg_send status %d\n",
938 			       param);
939 		}
940 		return;
941 	} else if (event == CTL_HA_EVT_DISCONNECT) {
942 		printf("CTL: Got a disconnect from Isc\n");
943 		return;
944 	} else {
945 		printf("ctl_isc_event_handler: Unknown event %d\n", event);
946 		return;
947 	}
948 
949 bailout:
950 	return;
951 }
952 
953 static void
954 ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
955 {
956 	struct scsi_sense_data *sense;
957 
958 	sense = &dest->scsiio.sense_data;
959 	bcopy(&src->scsi.sense_data, sense, sizeof(*sense));
960 	dest->scsiio.scsi_status = src->scsi.scsi_status;
961 	dest->scsiio.sense_len = src->scsi.sense_len;
962 	dest->io_hdr.status = src->hdr.status;
963 }
964 
965 static void
966 ctl_est_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
967 {
968 	ctl_ua_type *pu;
969 
970 	mtx_assert(&lun->lun_lock, MA_OWNED);
971 	pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
972 	if (pu == NULL)
973 		return;
974 	pu[initidx % CTL_MAX_INIT_PER_PORT] |= ua;
975 }
976 
977 static void
978 ctl_est_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
979 {
980 	int i, j;
981 
982 	mtx_assert(&lun->lun_lock, MA_OWNED);
983 	for (i = 0; i < CTL_MAX_PORTS; i++) {
984 		if (lun->pending_ua[i] == NULL)
985 			continue;
986 		for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
987 			if (i * CTL_MAX_INIT_PER_PORT + j == except)
988 				continue;
989 			lun->pending_ua[i][j] |= ua;
990 		}
991 	}
992 }
993 
994 static void
995 ctl_clr_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
996 {
997 	ctl_ua_type *pu;
998 
999 	mtx_assert(&lun->lun_lock, MA_OWNED);
1000 	pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
1001 	if (pu == NULL)
1002 		return;
1003 	pu[initidx % CTL_MAX_INIT_PER_PORT] &= ~ua;
1004 }
1005 
1006 static void
1007 ctl_clr_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1008 {
1009 	int i, j;
1010 
1011 	mtx_assert(&lun->lun_lock, MA_OWNED);
1012 	for (i = 0; i < CTL_MAX_PORTS; i++) {
1013 		if (lun->pending_ua[i] == NULL)
1014 			continue;
1015 		for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
1016 			if (i * CTL_MAX_INIT_PER_PORT + j == except)
1017 				continue;
1018 			lun->pending_ua[i][j] &= ~ua;
1019 		}
1020 	}
1021 }
1022 
1023 static int
1024 ctl_ha_state_sysctl(SYSCTL_HANDLER_ARGS)
1025 {
1026 	struct ctl_softc *softc = (struct ctl_softc *)arg1;
1027 	struct ctl_lun *lun;
1028 	int error, value;
1029 
1030 	if (softc->flags & CTL_FLAG_ACTIVE_SHELF)
1031 		value = 0;
1032 	else
1033 		value = 1;
1034 
1035 	error = sysctl_handle_int(oidp, &value, 0, req);
1036 	if ((error != 0) || (req->newptr == NULL))
1037 		return (error);
1038 
1039 	mtx_lock(&softc->ctl_lock);
1040 	if (value == 0)
1041 		softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1042 	else
1043 		softc->flags &= ~CTL_FLAG_ACTIVE_SHELF;
1044 	STAILQ_FOREACH(lun, &softc->lun_list, links) {
1045 		mtx_lock(&lun->lun_lock);
1046 		ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
1047 		mtx_unlock(&lun->lun_lock);
1048 	}
1049 	mtx_unlock(&softc->ctl_lock);
1050 	return (0);
1051 }
1052 
1053 static int
1054 ctl_init(void)
1055 {
1056 	struct ctl_softc *softc;
1057 	void *other_pool;
1058 	struct ctl_port *port;
1059 	int i, error, retval;
1060 	//int isc_retval;
1061 
1062 	retval = 0;
1063 	ctl_pause_rtr = 0;
1064         rcv_sync_msg = 0;
1065 
1066 	control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
1067 			       M_WAITOK | M_ZERO);
1068 	softc = control_softc;
1069 
1070 	softc->dev = make_dev(&ctl_cdevsw, 0, UID_ROOT, GID_OPERATOR, 0600,
1071 			      "cam/ctl");
1072 
1073 	softc->dev->si_drv1 = softc;
1074 
1075 	/*
1076 	 * By default, return a "bad LUN" peripheral qualifier for unknown
1077 	 * LUNs.  The user can override this default using the tunable or
1078 	 * sysctl.  See the comment in ctl_inquiry_std() for more details.
1079 	 */
1080 	softc->inquiry_pq_no_lun = 1;
1081 	TUNABLE_INT_FETCH("kern.cam.ctl.inquiry_pq_no_lun",
1082 			  &softc->inquiry_pq_no_lun);
1083 	sysctl_ctx_init(&softc->sysctl_ctx);
1084 	softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
1085 		SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
1086 		CTLFLAG_RD, 0, "CAM Target Layer");
1087 
1088 	if (softc->sysctl_tree == NULL) {
1089 		printf("%s: unable to allocate sysctl tree\n", __func__);
1090 		destroy_dev(softc->dev);
1091 		free(control_softc, M_DEVBUF);
1092 		control_softc = NULL;
1093 		return (ENOMEM);
1094 	}
1095 
1096 	SYSCTL_ADD_INT(&softc->sysctl_ctx,
1097 		       SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO,
1098 		       "inquiry_pq_no_lun", CTLFLAG_RW,
1099 		       &softc->inquiry_pq_no_lun, 0,
1100 		       "Report no lun possible for invalid LUNs");
1101 
1102 	mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
1103 	softc->io_zone = uma_zcreate("CTL IO", sizeof(union ctl_io),
1104 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1105 	softc->open_count = 0;
1106 
1107 	/*
1108 	 * Default to actually sending a SYNCHRONIZE CACHE command down to
1109 	 * the drive.
1110 	 */
1111 	softc->flags = CTL_FLAG_REAL_SYNC;
1112 
1113 	/*
1114 	 * In Copan's HA scheme, the "master" and "slave" roles are
1115 	 * figured out through the slot the controller is in.  Although it
1116 	 * is an active/active system, someone has to be in charge.
1117 	 */
1118 	SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1119 	    OID_AUTO, "ha_id", CTLFLAG_RDTUN, &softc->ha_id, 0,
1120 	    "HA head ID (0 - no HA)");
1121 	if (softc->ha_id == 0) {
1122 		softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1123 		softc->is_single = 1;
1124 		softc->port_offset = 0;
1125 	} else
1126 		softc->port_offset = (softc->ha_id - 1) * CTL_MAX_PORTS;
1127 	softc->persis_offset = softc->port_offset * CTL_MAX_INIT_PER_PORT;
1128 
1129 	/*
1130 	 * XXX KDM need to figure out where we want to get our target ID
1131 	 * and WWID.  Is it different on each port?
1132 	 */
1133 	softc->target.id = 0;
1134 	softc->target.wwid[0] = 0x12345678;
1135 	softc->target.wwid[1] = 0x87654321;
1136 	STAILQ_INIT(&softc->lun_list);
1137 	STAILQ_INIT(&softc->pending_lun_queue);
1138 	STAILQ_INIT(&softc->fe_list);
1139 	STAILQ_INIT(&softc->port_list);
1140 	STAILQ_INIT(&softc->be_list);
1141 	ctl_tpc_init(softc);
1142 
1143 	if (ctl_pool_create(softc, "othersc", CTL_POOL_ENTRIES_OTHER_SC,
1144 	                    &other_pool) != 0)
1145 	{
1146 		printf("ctl: can't allocate %d entry other SC pool, "
1147 		       "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1148 		return (ENOMEM);
1149 	}
1150 	softc->othersc_pool = other_pool;
1151 
1152 	if (worker_threads <= 0)
1153 		worker_threads = max(1, mp_ncpus / 4);
1154 	if (worker_threads > CTL_MAX_THREADS)
1155 		worker_threads = CTL_MAX_THREADS;
1156 
1157 	for (i = 0; i < worker_threads; i++) {
1158 		struct ctl_thread *thr = &softc->threads[i];
1159 
1160 		mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1161 		thr->ctl_softc = softc;
1162 		STAILQ_INIT(&thr->incoming_queue);
1163 		STAILQ_INIT(&thr->rtr_queue);
1164 		STAILQ_INIT(&thr->done_queue);
1165 		STAILQ_INIT(&thr->isc_queue);
1166 
1167 		error = kproc_kthread_add(ctl_work_thread, thr,
1168 		    &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1169 		if (error != 0) {
1170 			printf("error creating CTL work thread!\n");
1171 			ctl_pool_free(other_pool);
1172 			return (error);
1173 		}
1174 	}
1175 	error = kproc_kthread_add(ctl_lun_thread, softc,
1176 	    &softc->ctl_proc, NULL, 0, 0, "ctl", "lun");
1177 	if (error != 0) {
1178 		printf("error creating CTL lun thread!\n");
1179 		ctl_pool_free(other_pool);
1180 		return (error);
1181 	}
1182 	error = kproc_kthread_add(ctl_thresh_thread, softc,
1183 	    &softc->ctl_proc, NULL, 0, 0, "ctl", "thresh");
1184 	if (error != 0) {
1185 		printf("error creating CTL threshold thread!\n");
1186 		ctl_pool_free(other_pool);
1187 		return (error);
1188 	}
1189 	if (bootverbose)
1190 		printf("ctl: CAM Target Layer loaded\n");
1191 
1192 	/*
1193 	 * Initialize the ioctl front end.
1194 	 */
1195 	ctl_frontend_register(&ioctl_frontend);
1196 	port = &softc->ioctl_info.port;
1197 	port->frontend = &ioctl_frontend;
1198 	sprintf(softc->ioctl_info.port_name, "ioctl");
1199 	port->port_type = CTL_PORT_IOCTL;
1200 	port->num_requested_ctl_io = 100;
1201 	port->port_name = softc->ioctl_info.port_name;
1202 	port->port_online = ctl_ioctl_online;
1203 	port->port_offline = ctl_ioctl_offline;
1204 	port->onoff_arg = &softc->ioctl_info;
1205 	port->lun_enable = ctl_ioctl_lun_enable;
1206 	port->lun_disable = ctl_ioctl_lun_disable;
1207 	port->targ_lun_arg = &softc->ioctl_info;
1208 	port->fe_datamove = ctl_ioctl_datamove;
1209 	port->fe_done = ctl_ioctl_done;
1210 	port->max_targets = 15;
1211 	port->max_target_id = 15;
1212 
1213 	if (ctl_port_register(&softc->ioctl_info.port) != 0) {
1214 		printf("ctl: ioctl front end registration failed, will "
1215 		       "continue anyway\n");
1216 	}
1217 
1218 	SYSCTL_ADD_PROC(&softc->sysctl_ctx,SYSCTL_CHILDREN(softc->sysctl_tree),
1219 	    OID_AUTO, "ha_state", CTLTYPE_INT | CTLFLAG_RWTUN,
1220 	    softc, 0, ctl_ha_state_sysctl, "I", "HA state for this head");
1221 
1222 #ifdef CTL_IO_DELAY
1223 	if (sizeof(struct callout) > CTL_TIMER_BYTES) {
1224 		printf("sizeof(struct callout) %zd > CTL_TIMER_BYTES %zd\n",
1225 		       sizeof(struct callout), CTL_TIMER_BYTES);
1226 		return (EINVAL);
1227 	}
1228 #endif /* CTL_IO_DELAY */
1229 
1230 	return (0);
1231 }
1232 
1233 void
1234 ctl_shutdown(void)
1235 {
1236 	struct ctl_softc *softc;
1237 	struct ctl_lun *lun, *next_lun;
1238 
1239 	softc = (struct ctl_softc *)control_softc;
1240 
1241 	if (ctl_port_deregister(&softc->ioctl_info.port) != 0)
1242 		printf("ctl: ioctl front end deregistration failed\n");
1243 
1244 	mtx_lock(&softc->ctl_lock);
1245 
1246 	/*
1247 	 * Free up each LUN.
1248 	 */
1249 	for (lun = STAILQ_FIRST(&softc->lun_list); lun != NULL; lun = next_lun){
1250 		next_lun = STAILQ_NEXT(lun, links);
1251 		ctl_free_lun(lun);
1252 	}
1253 
1254 	mtx_unlock(&softc->ctl_lock);
1255 
1256 	ctl_frontend_deregister(&ioctl_frontend);
1257 
1258 #if 0
1259 	ctl_shutdown_thread(softc->work_thread);
1260 	mtx_destroy(&softc->queue_lock);
1261 #endif
1262 
1263 	ctl_tpc_shutdown(softc);
1264 	uma_zdestroy(softc->io_zone);
1265 	mtx_destroy(&softc->ctl_lock);
1266 
1267 	destroy_dev(softc->dev);
1268 
1269 	sysctl_ctx_free(&softc->sysctl_ctx);
1270 
1271 	free(control_softc, M_DEVBUF);
1272 	control_softc = NULL;
1273 
1274 	if (bootverbose)
1275 		printf("ctl: CAM Target Layer unloaded\n");
1276 }
1277 
1278 static int
1279 ctl_module_event_handler(module_t mod, int what, void *arg)
1280 {
1281 
1282 	switch (what) {
1283 	case MOD_LOAD:
1284 		return (ctl_init());
1285 	case MOD_UNLOAD:
1286 		return (EBUSY);
1287 	default:
1288 		return (EOPNOTSUPP);
1289 	}
1290 }
1291 
1292 /*
1293  * XXX KDM should we do some access checks here?  Bump a reference count to
1294  * prevent a CTL module from being unloaded while someone has it open?
1295  */
1296 static int
1297 ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1298 {
1299 	return (0);
1300 }
1301 
1302 static int
1303 ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1304 {
1305 	return (0);
1306 }
1307 
1308 int
1309 ctl_port_enable(ctl_port_type port_type)
1310 {
1311 	struct ctl_softc *softc = control_softc;
1312 	struct ctl_port *port;
1313 
1314 	if (softc->is_single == 0) {
1315 		union ctl_ha_msg msg_info;
1316 		int isc_retval;
1317 
1318 #if 0
1319 		printf("%s: HA mode, synchronizing frontend enable\n",
1320 		        __func__);
1321 #endif
1322 		msg_info.hdr.msg_type = CTL_MSG_SYNC_FE;
1323 	        if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1324 		        sizeof(msg_info), 1 )) > CTL_HA_STATUS_SUCCESS) {
1325 			printf("Sync msg send error retval %d\n", isc_retval);
1326 		}
1327 		if (!rcv_sync_msg) {
1328 			isc_retval=ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
1329 			        sizeof(msg_info), 1);
1330 		}
1331 #if 0
1332         	printf("CTL:Frontend Enable\n");
1333 	} else {
1334 		printf("%s: single mode, skipping frontend synchronization\n",
1335 		        __func__);
1336 #endif
1337 	}
1338 
1339 	STAILQ_FOREACH(port, &softc->port_list, links) {
1340 		if (port_type & port->port_type)
1341 		{
1342 #if 0
1343 			printf("port %d\n", port->targ_port);
1344 #endif
1345 			ctl_port_online(port);
1346 		}
1347 	}
1348 
1349 	return (0);
1350 }
1351 
1352 int
1353 ctl_port_disable(ctl_port_type port_type)
1354 {
1355 	struct ctl_softc *softc;
1356 	struct ctl_port *port;
1357 
1358 	softc = control_softc;
1359 
1360 	STAILQ_FOREACH(port, &softc->port_list, links) {
1361 		if (port_type & port->port_type)
1362 			ctl_port_offline(port);
1363 	}
1364 
1365 	return (0);
1366 }
1367 
1368 /*
1369  * Returns 0 for success, 1 for failure.
1370  * Currently the only failure mode is if there aren't enough entries
1371  * allocated.  So, in case of a failure, look at num_entries_dropped,
1372  * reallocate and try again.
1373  */
1374 int
1375 ctl_port_list(struct ctl_port_entry *entries, int num_entries_alloced,
1376 	      int *num_entries_filled, int *num_entries_dropped,
1377 	      ctl_port_type port_type, int no_virtual)
1378 {
1379 	struct ctl_softc *softc;
1380 	struct ctl_port *port;
1381 	int entries_dropped, entries_filled;
1382 	int retval;
1383 	int i;
1384 
1385 	softc = control_softc;
1386 
1387 	retval = 0;
1388 	entries_filled = 0;
1389 	entries_dropped = 0;
1390 
1391 	i = 0;
1392 	mtx_lock(&softc->ctl_lock);
1393 	STAILQ_FOREACH(port, &softc->port_list, links) {
1394 		struct ctl_port_entry *entry;
1395 
1396 		if ((port->port_type & port_type) == 0)
1397 			continue;
1398 
1399 		if ((no_virtual != 0)
1400 		 && (port->virtual_port != 0))
1401 			continue;
1402 
1403 		if (entries_filled >= num_entries_alloced) {
1404 			entries_dropped++;
1405 			continue;
1406 		}
1407 		entry = &entries[i];
1408 
1409 		entry->port_type = port->port_type;
1410 		strlcpy(entry->port_name, port->port_name,
1411 			sizeof(entry->port_name));
1412 		entry->physical_port = port->physical_port;
1413 		entry->virtual_port = port->virtual_port;
1414 		entry->wwnn = port->wwnn;
1415 		entry->wwpn = port->wwpn;
1416 
1417 		i++;
1418 		entries_filled++;
1419 	}
1420 
1421 	mtx_unlock(&softc->ctl_lock);
1422 
1423 	if (entries_dropped > 0)
1424 		retval = 1;
1425 
1426 	*num_entries_dropped = entries_dropped;
1427 	*num_entries_filled = entries_filled;
1428 
1429 	return (retval);
1430 }
1431 
1432 static void
1433 ctl_ioctl_online(void *arg)
1434 {
1435 	struct ctl_ioctl_info *ioctl_info;
1436 
1437 	ioctl_info = (struct ctl_ioctl_info *)arg;
1438 
1439 	ioctl_info->flags |= CTL_IOCTL_FLAG_ENABLED;
1440 }
1441 
1442 static void
1443 ctl_ioctl_offline(void *arg)
1444 {
1445 	struct ctl_ioctl_info *ioctl_info;
1446 
1447 	ioctl_info = (struct ctl_ioctl_info *)arg;
1448 
1449 	ioctl_info->flags &= ~CTL_IOCTL_FLAG_ENABLED;
1450 }
1451 
1452 /*
1453  * Remove an initiator by port number and initiator ID.
1454  * Returns 0 for success, -1 for failure.
1455  */
1456 int
1457 ctl_remove_initiator(struct ctl_port *port, int iid)
1458 {
1459 	struct ctl_softc *softc = control_softc;
1460 
1461 	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1462 
1463 	if (iid > CTL_MAX_INIT_PER_PORT) {
1464 		printf("%s: initiator ID %u > maximun %u!\n",
1465 		       __func__, iid, CTL_MAX_INIT_PER_PORT);
1466 		return (-1);
1467 	}
1468 
1469 	mtx_lock(&softc->ctl_lock);
1470 	port->wwpn_iid[iid].in_use--;
1471 	port->wwpn_iid[iid].last_use = time_uptime;
1472 	mtx_unlock(&softc->ctl_lock);
1473 
1474 	return (0);
1475 }
1476 
1477 /*
1478  * Add an initiator to the initiator map.
1479  * Returns iid for success, < 0 for failure.
1480  */
1481 int
1482 ctl_add_initiator(struct ctl_port *port, int iid, uint64_t wwpn, char *name)
1483 {
1484 	struct ctl_softc *softc = control_softc;
1485 	time_t best_time;
1486 	int i, best;
1487 
1488 	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1489 
1490 	if (iid >= CTL_MAX_INIT_PER_PORT) {
1491 		printf("%s: WWPN %#jx initiator ID %u > maximum %u!\n",
1492 		       __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
1493 		free(name, M_CTL);
1494 		return (-1);
1495 	}
1496 
1497 	mtx_lock(&softc->ctl_lock);
1498 
1499 	if (iid < 0 && (wwpn != 0 || name != NULL)) {
1500 		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1501 			if (wwpn != 0 && wwpn == port->wwpn_iid[i].wwpn) {
1502 				iid = i;
1503 				break;
1504 			}
1505 			if (name != NULL && port->wwpn_iid[i].name != NULL &&
1506 			    strcmp(name, port->wwpn_iid[i].name) == 0) {
1507 				iid = i;
1508 				break;
1509 			}
1510 		}
1511 	}
1512 
1513 	if (iid < 0) {
1514 		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1515 			if (port->wwpn_iid[i].in_use == 0 &&
1516 			    port->wwpn_iid[i].wwpn == 0 &&
1517 			    port->wwpn_iid[i].name == NULL) {
1518 				iid = i;
1519 				break;
1520 			}
1521 		}
1522 	}
1523 
1524 	if (iid < 0) {
1525 		best = -1;
1526 		best_time = INT32_MAX;
1527 		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1528 			if (port->wwpn_iid[i].in_use == 0) {
1529 				if (port->wwpn_iid[i].last_use < best_time) {
1530 					best = i;
1531 					best_time = port->wwpn_iid[i].last_use;
1532 				}
1533 			}
1534 		}
1535 		iid = best;
1536 	}
1537 
1538 	if (iid < 0) {
1539 		mtx_unlock(&softc->ctl_lock);
1540 		free(name, M_CTL);
1541 		return (-2);
1542 	}
1543 
1544 	if (port->wwpn_iid[iid].in_use > 0 && (wwpn != 0 || name != NULL)) {
1545 		/*
1546 		 * This is not an error yet.
1547 		 */
1548 		if (wwpn != 0 && wwpn == port->wwpn_iid[iid].wwpn) {
1549 #if 0
1550 			printf("%s: port %d iid %u WWPN %#jx arrived"
1551 			    " again\n", __func__, port->targ_port,
1552 			    iid, (uintmax_t)wwpn);
1553 #endif
1554 			goto take;
1555 		}
1556 		if (name != NULL && port->wwpn_iid[iid].name != NULL &&
1557 		    strcmp(name, port->wwpn_iid[iid].name) == 0) {
1558 #if 0
1559 			printf("%s: port %d iid %u name '%s' arrived"
1560 			    " again\n", __func__, port->targ_port,
1561 			    iid, name);
1562 #endif
1563 			goto take;
1564 		}
1565 
1566 		/*
1567 		 * This is an error, but what do we do about it?  The
1568 		 * driver is telling us we have a new WWPN for this
1569 		 * initiator ID, so we pretty much need to use it.
1570 		 */
1571 		printf("%s: port %d iid %u WWPN %#jx '%s' arrived,"
1572 		    " but WWPN %#jx '%s' is still at that address\n",
1573 		    __func__, port->targ_port, iid, wwpn, name,
1574 		    (uintmax_t)port->wwpn_iid[iid].wwpn,
1575 		    port->wwpn_iid[iid].name);
1576 
1577 		/*
1578 		 * XXX KDM clear have_ca and ua_pending on each LUN for
1579 		 * this initiator.
1580 		 */
1581 	}
1582 take:
1583 	free(port->wwpn_iid[iid].name, M_CTL);
1584 	port->wwpn_iid[iid].name = name;
1585 	port->wwpn_iid[iid].wwpn = wwpn;
1586 	port->wwpn_iid[iid].in_use++;
1587 	mtx_unlock(&softc->ctl_lock);
1588 
1589 	return (iid);
1590 }
1591 
1592 static int
1593 ctl_create_iid(struct ctl_port *port, int iid, uint8_t *buf)
1594 {
1595 	int len;
1596 
1597 	switch (port->port_type) {
1598 	case CTL_PORT_FC:
1599 	{
1600 		struct scsi_transportid_fcp *id =
1601 		    (struct scsi_transportid_fcp *)buf;
1602 		if (port->wwpn_iid[iid].wwpn == 0)
1603 			return (0);
1604 		memset(id, 0, sizeof(*id));
1605 		id->format_protocol = SCSI_PROTO_FC;
1606 		scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->n_port_name);
1607 		return (sizeof(*id));
1608 	}
1609 	case CTL_PORT_ISCSI:
1610 	{
1611 		struct scsi_transportid_iscsi_port *id =
1612 		    (struct scsi_transportid_iscsi_port *)buf;
1613 		if (port->wwpn_iid[iid].name == NULL)
1614 			return (0);
1615 		memset(id, 0, 256);
1616 		id->format_protocol = SCSI_TRN_ISCSI_FORMAT_PORT |
1617 		    SCSI_PROTO_ISCSI;
1618 		len = strlcpy(id->iscsi_name, port->wwpn_iid[iid].name, 252) + 1;
1619 		len = roundup2(min(len, 252), 4);
1620 		scsi_ulto2b(len, id->additional_length);
1621 		return (sizeof(*id) + len);
1622 	}
1623 	case CTL_PORT_SAS:
1624 	{
1625 		struct scsi_transportid_sas *id =
1626 		    (struct scsi_transportid_sas *)buf;
1627 		if (port->wwpn_iid[iid].wwpn == 0)
1628 			return (0);
1629 		memset(id, 0, sizeof(*id));
1630 		id->format_protocol = SCSI_PROTO_SAS;
1631 		scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->sas_address);
1632 		return (sizeof(*id));
1633 	}
1634 	default:
1635 	{
1636 		struct scsi_transportid_spi *id =
1637 		    (struct scsi_transportid_spi *)buf;
1638 		memset(id, 0, sizeof(*id));
1639 		id->format_protocol = SCSI_PROTO_SPI;
1640 		scsi_ulto2b(iid, id->scsi_addr);
1641 		scsi_ulto2b(port->targ_port, id->rel_trgt_port_id);
1642 		return (sizeof(*id));
1643 	}
1644 	}
1645 }
1646 
1647 static int
1648 ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id)
1649 {
1650 	return (0);
1651 }
1652 
1653 static int
1654 ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id)
1655 {
1656 	return (0);
1657 }
1658 
1659 /*
1660  * Data movement routine for the CTL ioctl frontend port.
1661  */
1662 static int
1663 ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio)
1664 {
1665 	struct ctl_sg_entry *ext_sglist, *kern_sglist;
1666 	struct ctl_sg_entry ext_entry, kern_entry;
1667 	int ext_sglen, ext_sg_entries, kern_sg_entries;
1668 	int ext_sg_start, ext_offset;
1669 	int len_to_copy, len_copied;
1670 	int kern_watermark, ext_watermark;
1671 	int ext_sglist_malloced;
1672 	int i, j;
1673 
1674 	ext_sglist_malloced = 0;
1675 	ext_sg_start = 0;
1676 	ext_offset = 0;
1677 
1678 	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove\n"));
1679 
1680 	/*
1681 	 * If this flag is set, fake the data transfer.
1682 	 */
1683 	if (ctsio->io_hdr.flags & CTL_FLAG_NO_DATAMOVE) {
1684 		ctsio->ext_data_filled = ctsio->ext_data_len;
1685 		goto bailout;
1686 	}
1687 
1688 	/*
1689 	 * To simplify things here, if we have a single buffer, stick it in
1690 	 * a S/G entry and just make it a single entry S/G list.
1691 	 */
1692 	if (ctsio->io_hdr.flags & CTL_FLAG_EDPTR_SGLIST) {
1693 		int len_seen;
1694 
1695 		ext_sglen = ctsio->ext_sg_entries * sizeof(*ext_sglist);
1696 
1697 		ext_sglist = (struct ctl_sg_entry *)malloc(ext_sglen, M_CTL,
1698 							   M_WAITOK);
1699 		ext_sglist_malloced = 1;
1700 		if (copyin(ctsio->ext_data_ptr, ext_sglist,
1701 				   ext_sglen) != 0) {
1702 			ctl_set_internal_failure(ctsio,
1703 						 /*sks_valid*/ 0,
1704 						 /*retry_count*/ 0);
1705 			goto bailout;
1706 		}
1707 		ext_sg_entries = ctsio->ext_sg_entries;
1708 		len_seen = 0;
1709 		for (i = 0; i < ext_sg_entries; i++) {
1710 			if ((len_seen + ext_sglist[i].len) >=
1711 			     ctsio->ext_data_filled) {
1712 				ext_sg_start = i;
1713 				ext_offset = ctsio->ext_data_filled - len_seen;
1714 				break;
1715 			}
1716 			len_seen += ext_sglist[i].len;
1717 		}
1718 	} else {
1719 		ext_sglist = &ext_entry;
1720 		ext_sglist->addr = ctsio->ext_data_ptr;
1721 		ext_sglist->len = ctsio->ext_data_len;
1722 		ext_sg_entries = 1;
1723 		ext_sg_start = 0;
1724 		ext_offset = ctsio->ext_data_filled;
1725 	}
1726 
1727 	if (ctsio->kern_sg_entries > 0) {
1728 		kern_sglist = (struct ctl_sg_entry *)ctsio->kern_data_ptr;
1729 		kern_sg_entries = ctsio->kern_sg_entries;
1730 	} else {
1731 		kern_sglist = &kern_entry;
1732 		kern_sglist->addr = ctsio->kern_data_ptr;
1733 		kern_sglist->len = ctsio->kern_data_len;
1734 		kern_sg_entries = 1;
1735 	}
1736 
1737 
1738 	kern_watermark = 0;
1739 	ext_watermark = ext_offset;
1740 	len_copied = 0;
1741 	for (i = ext_sg_start, j = 0;
1742 	     i < ext_sg_entries && j < kern_sg_entries;) {
1743 		uint8_t *ext_ptr, *kern_ptr;
1744 
1745 		len_to_copy = ctl_min(ext_sglist[i].len - ext_watermark,
1746 				      kern_sglist[j].len - kern_watermark);
1747 
1748 		ext_ptr = (uint8_t *)ext_sglist[i].addr;
1749 		ext_ptr = ext_ptr + ext_watermark;
1750 		if (ctsio->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
1751 			/*
1752 			 * XXX KDM fix this!
1753 			 */
1754 			panic("need to implement bus address support");
1755 #if 0
1756 			kern_ptr = bus_to_virt(kern_sglist[j].addr);
1757 #endif
1758 		} else
1759 			kern_ptr = (uint8_t *)kern_sglist[j].addr;
1760 		kern_ptr = kern_ptr + kern_watermark;
1761 
1762 		kern_watermark += len_to_copy;
1763 		ext_watermark += len_to_copy;
1764 
1765 		if ((ctsio->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
1766 		     CTL_FLAG_DATA_IN) {
1767 			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1768 					 "bytes to user\n", len_to_copy));
1769 			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1770 					 "to %p\n", kern_ptr, ext_ptr));
1771 			if (copyout(kern_ptr, ext_ptr, len_to_copy) != 0) {
1772 				ctl_set_internal_failure(ctsio,
1773 							 /*sks_valid*/ 0,
1774 							 /*retry_count*/ 0);
1775 				goto bailout;
1776 			}
1777 		} else {
1778 			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1779 					 "bytes from user\n", len_to_copy));
1780 			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1781 					 "to %p\n", ext_ptr, kern_ptr));
1782 			if (copyin(ext_ptr, kern_ptr, len_to_copy)!= 0){
1783 				ctl_set_internal_failure(ctsio,
1784 							 /*sks_valid*/ 0,
1785 							 /*retry_count*/0);
1786 				goto bailout;
1787 			}
1788 		}
1789 
1790 		len_copied += len_to_copy;
1791 
1792 		if (ext_sglist[i].len == ext_watermark) {
1793 			i++;
1794 			ext_watermark = 0;
1795 		}
1796 
1797 		if (kern_sglist[j].len == kern_watermark) {
1798 			j++;
1799 			kern_watermark = 0;
1800 		}
1801 	}
1802 
1803 	ctsio->ext_data_filled += len_copied;
1804 
1805 	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_sg_entries: %d, "
1806 			 "kern_sg_entries: %d\n", ext_sg_entries,
1807 			 kern_sg_entries));
1808 	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_data_len = %d, "
1809 			 "kern_data_len = %d\n", ctsio->ext_data_len,
1810 			 ctsio->kern_data_len));
1811 
1812 
1813 	/* XXX KDM set residual?? */
1814 bailout:
1815 
1816 	if (ext_sglist_malloced != 0)
1817 		free(ext_sglist, M_CTL);
1818 
1819 	return (CTL_RETVAL_COMPLETE);
1820 }
1821 
1822 /*
1823  * Serialize a command that went down the "wrong" side, and so was sent to
1824  * this controller for execution.  The logic is a little different than the
1825  * standard case in ctl_scsiio_precheck().  Errors in this case need to get
1826  * sent back to the other side, but in the success case, we execute the
1827  * command on this side (XFER mode) or tell the other side to execute it
1828  * (SER_ONLY mode).
1829  */
1830 static int
1831 ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
1832 {
1833 	struct ctl_softc *ctl_softc;
1834 	union ctl_ha_msg msg_info;
1835 	struct ctl_lun *lun;
1836 	int retval = 0;
1837 	uint32_t targ_lun;
1838 
1839 	ctl_softc = control_softc;
1840 
1841 	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
1842 	lun = ctl_softc->ctl_luns[targ_lun];
1843 	if (lun==NULL)
1844 	{
1845 		/*
1846 		 * Why isn't LUN defined? The other side wouldn't
1847 		 * send a cmd if the LUN is undefined.
1848 		 */
1849 		printf("%s: Bad JUJU!, LUN is NULL!\n", __func__);
1850 
1851 		/* "Logical unit not supported" */
1852 		ctl_set_sense_data(&msg_info.scsi.sense_data,
1853 				   lun,
1854 				   /*sense_format*/SSD_TYPE_NONE,
1855 				   /*current_error*/ 1,
1856 				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1857 				   /*asc*/ 0x25,
1858 				   /*ascq*/ 0x00,
1859 				   SSD_ELEM_NONE);
1860 
1861 		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1862 		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1863 		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1864 		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1865 		msg_info.hdr.serializing_sc = NULL;
1866 		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1867 	        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1868 				sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1869 		}
1870 		return(1);
1871 
1872 	}
1873 
1874 	mtx_lock(&lun->lun_lock);
1875     	TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1876 
1877 	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
1878 		(union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
1879 		 ooa_links))) {
1880 	case CTL_ACTION_BLOCK:
1881 		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
1882 		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
1883 				  blocked_links);
1884 		break;
1885 	case CTL_ACTION_PASS:
1886 	case CTL_ACTION_SKIP:
1887 		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
1888 			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
1889 			ctl_enqueue_rtr((union ctl_io *)ctsio);
1890 		} else {
1891 
1892 			/* send msg back to other side */
1893 			msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1894 			msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
1895 			msg_info.hdr.msg_type = CTL_MSG_R2R;
1896 #if 0
1897 			printf("2. pOrig %x\n", (int)msg_info.hdr.original_sc);
1898 #endif
1899 		        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1900 			    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1901 			}
1902 		}
1903 		break;
1904 	case CTL_ACTION_OVERLAP:
1905 		/* OVERLAPPED COMMANDS ATTEMPTED */
1906 		ctl_set_sense_data(&msg_info.scsi.sense_data,
1907 				   lun,
1908 				   /*sense_format*/SSD_TYPE_NONE,
1909 				   /*current_error*/ 1,
1910 				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1911 				   /*asc*/ 0x4E,
1912 				   /*ascq*/ 0x00,
1913 				   SSD_ELEM_NONE);
1914 
1915 		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1916 		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1917 		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1918 		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1919 		msg_info.hdr.serializing_sc = NULL;
1920 		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1921 #if 0
1922 		printf("BAD JUJU:Major Bummer Overlap\n");
1923 #endif
1924 		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1925 		retval = 1;
1926 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1927 		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1928 		}
1929 		break;
1930 	case CTL_ACTION_OVERLAP_TAG:
1931 		/* TAGGED OVERLAPPED COMMANDS (NN = QUEUE TAG) */
1932 		ctl_set_sense_data(&msg_info.scsi.sense_data,
1933 				   lun,
1934 				   /*sense_format*/SSD_TYPE_NONE,
1935 				   /*current_error*/ 1,
1936 				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1937 				   /*asc*/ 0x4D,
1938 				   /*ascq*/ ctsio->tag_num & 0xff,
1939 				   SSD_ELEM_NONE);
1940 
1941 		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1942 		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1943 		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1944 		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1945 		msg_info.hdr.serializing_sc = NULL;
1946 		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1947 #if 0
1948 		printf("BAD JUJU:Major Bummer Overlap Tag\n");
1949 #endif
1950 		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1951 		retval = 1;
1952 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1953 		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1954 		}
1955 		break;
1956 	case CTL_ACTION_ERROR:
1957 	default:
1958 		/* "Internal target failure" */
1959 		ctl_set_sense_data(&msg_info.scsi.sense_data,
1960 				   lun,
1961 				   /*sense_format*/SSD_TYPE_NONE,
1962 				   /*current_error*/ 1,
1963 				   /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
1964 				   /*asc*/ 0x44,
1965 				   /*ascq*/ 0x00,
1966 				   SSD_ELEM_NONE);
1967 
1968 		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1969 		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1970 		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1971 		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1972 		msg_info.hdr.serializing_sc = NULL;
1973 		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1974 #if 0
1975 		printf("BAD JUJU:Major Bummer HW Error\n");
1976 #endif
1977 		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1978 		retval = 1;
1979 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1980 		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1981 		}
1982 		break;
1983 	}
1984 	mtx_unlock(&lun->lun_lock);
1985 	return (retval);
1986 }
1987 
1988 static int
1989 ctl_ioctl_submit_wait(union ctl_io *io)
1990 {
1991 	struct ctl_fe_ioctl_params params;
1992 	ctl_fe_ioctl_state last_state;
1993 	int done, retval;
1994 
1995 	retval = 0;
1996 
1997 	bzero(&params, sizeof(params));
1998 
1999 	mtx_init(&params.ioctl_mtx, "ctliocmtx", NULL, MTX_DEF);
2000 	cv_init(&params.sem, "ctlioccv");
2001 	params.state = CTL_IOCTL_INPROG;
2002 	last_state = params.state;
2003 
2004 	io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = &params;
2005 
2006 	CTL_DEBUG_PRINT(("ctl_ioctl_submit_wait\n"));
2007 
2008 	/* This shouldn't happen */
2009 	if ((retval = ctl_queue(io)) != CTL_RETVAL_COMPLETE)
2010 		return (retval);
2011 
2012 	done = 0;
2013 
2014 	do {
2015 		mtx_lock(&params.ioctl_mtx);
2016 		/*
2017 		 * Check the state here, and don't sleep if the state has
2018 		 * already changed (i.e. wakeup has already occured, but we
2019 		 * weren't waiting yet).
2020 		 */
2021 		if (params.state == last_state) {
2022 			/* XXX KDM cv_wait_sig instead? */
2023 			cv_wait(&params.sem, &params.ioctl_mtx);
2024 		}
2025 		last_state = params.state;
2026 
2027 		switch (params.state) {
2028 		case CTL_IOCTL_INPROG:
2029 			/* Why did we wake up? */
2030 			/* XXX KDM error here? */
2031 			mtx_unlock(&params.ioctl_mtx);
2032 			break;
2033 		case CTL_IOCTL_DATAMOVE:
2034 			CTL_DEBUG_PRINT(("got CTL_IOCTL_DATAMOVE\n"));
2035 
2036 			/*
2037 			 * change last_state back to INPROG to avoid
2038 			 * deadlock on subsequent data moves.
2039 			 */
2040 			params.state = last_state = CTL_IOCTL_INPROG;
2041 
2042 			mtx_unlock(&params.ioctl_mtx);
2043 			ctl_ioctl_do_datamove(&io->scsiio);
2044 			/*
2045 			 * Note that in some cases, most notably writes,
2046 			 * this will queue the I/O and call us back later.
2047 			 * In other cases, generally reads, this routine
2048 			 * will immediately call back and wake us up,
2049 			 * probably using our own context.
2050 			 */
2051 			io->scsiio.be_move_done(io);
2052 			break;
2053 		case CTL_IOCTL_DONE:
2054 			mtx_unlock(&params.ioctl_mtx);
2055 			CTL_DEBUG_PRINT(("got CTL_IOCTL_DONE\n"));
2056 			done = 1;
2057 			break;
2058 		default:
2059 			mtx_unlock(&params.ioctl_mtx);
2060 			/* XXX KDM error here? */
2061 			break;
2062 		}
2063 	} while (done == 0);
2064 
2065 	mtx_destroy(&params.ioctl_mtx);
2066 	cv_destroy(&params.sem);
2067 
2068 	return (CTL_RETVAL_COMPLETE);
2069 }
2070 
2071 static void
2072 ctl_ioctl_datamove(union ctl_io *io)
2073 {
2074 	struct ctl_fe_ioctl_params *params;
2075 
2076 	params = (struct ctl_fe_ioctl_params *)
2077 		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
2078 
2079 	mtx_lock(&params->ioctl_mtx);
2080 	params->state = CTL_IOCTL_DATAMOVE;
2081 	cv_broadcast(&params->sem);
2082 	mtx_unlock(&params->ioctl_mtx);
2083 }
2084 
2085 static void
2086 ctl_ioctl_done(union ctl_io *io)
2087 {
2088 	struct ctl_fe_ioctl_params *params;
2089 
2090 	params = (struct ctl_fe_ioctl_params *)
2091 		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
2092 
2093 	mtx_lock(&params->ioctl_mtx);
2094 	params->state = CTL_IOCTL_DONE;
2095 	cv_broadcast(&params->sem);
2096 	mtx_unlock(&params->ioctl_mtx);
2097 }
2098 
2099 static void
2100 ctl_ioctl_hard_startstop_callback(void *arg, struct cfi_metatask *metatask)
2101 {
2102 	struct ctl_fe_ioctl_startstop_info *sd_info;
2103 
2104 	sd_info = (struct ctl_fe_ioctl_startstop_info *)arg;
2105 
2106 	sd_info->hs_info.status = metatask->status;
2107 	sd_info->hs_info.total_luns = metatask->taskinfo.startstop.total_luns;
2108 	sd_info->hs_info.luns_complete =
2109 		metatask->taskinfo.startstop.luns_complete;
2110 	sd_info->hs_info.luns_failed = metatask->taskinfo.startstop.luns_failed;
2111 
2112 	cv_broadcast(&sd_info->sem);
2113 }
2114 
2115 static void
2116 ctl_ioctl_bbrread_callback(void *arg, struct cfi_metatask *metatask)
2117 {
2118 	struct ctl_fe_ioctl_bbrread_info *fe_bbr_info;
2119 
2120 	fe_bbr_info = (struct ctl_fe_ioctl_bbrread_info *)arg;
2121 
2122 	mtx_lock(fe_bbr_info->lock);
2123 	fe_bbr_info->bbr_info->status = metatask->status;
2124 	fe_bbr_info->bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2125 	fe_bbr_info->wakeup_done = 1;
2126 	mtx_unlock(fe_bbr_info->lock);
2127 
2128 	cv_broadcast(&fe_bbr_info->sem);
2129 }
2130 
2131 /*
2132  * Returns 0 for success, errno for failure.
2133  */
2134 static int
2135 ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
2136 		   struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
2137 {
2138 	union ctl_io *io;
2139 	int retval;
2140 
2141 	retval = 0;
2142 
2143 	mtx_lock(&lun->lun_lock);
2144 	for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
2145 	     (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2146 	     ooa_links)) {
2147 		struct ctl_ooa_entry *entry;
2148 
2149 		/*
2150 		 * If we've got more than we can fit, just count the
2151 		 * remaining entries.
2152 		 */
2153 		if (*cur_fill_num >= ooa_hdr->alloc_num)
2154 			continue;
2155 
2156 		entry = &kern_entries[*cur_fill_num];
2157 
2158 		entry->tag_num = io->scsiio.tag_num;
2159 		entry->lun_num = lun->lun;
2160 #ifdef CTL_TIME_IO
2161 		entry->start_bt = io->io_hdr.start_bt;
2162 #endif
2163 		bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2164 		entry->cdb_len = io->scsiio.cdb_len;
2165 		if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
2166 			entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2167 
2168 		if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2169 			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2170 
2171 		if (io->io_hdr.flags & CTL_FLAG_ABORT)
2172 			entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2173 
2174 		if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2175 			entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2176 
2177 		if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2178 			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2179 	}
2180 	mtx_unlock(&lun->lun_lock);
2181 
2182 	return (retval);
2183 }
2184 
2185 static void *
2186 ctl_copyin_alloc(void *user_addr, int len, char *error_str,
2187 		 size_t error_str_len)
2188 {
2189 	void *kptr;
2190 
2191 	kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2192 
2193 	if (copyin(user_addr, kptr, len) != 0) {
2194 		snprintf(error_str, error_str_len, "Error copying %d bytes "
2195 			 "from user address %p to kernel address %p", len,
2196 			 user_addr, kptr);
2197 		free(kptr, M_CTL);
2198 		return (NULL);
2199 	}
2200 
2201 	return (kptr);
2202 }
2203 
2204 static void
2205 ctl_free_args(int num_args, struct ctl_be_arg *args)
2206 {
2207 	int i;
2208 
2209 	if (args == NULL)
2210 		return;
2211 
2212 	for (i = 0; i < num_args; i++) {
2213 		free(args[i].kname, M_CTL);
2214 		free(args[i].kvalue, M_CTL);
2215 	}
2216 
2217 	free(args, M_CTL);
2218 }
2219 
2220 static struct ctl_be_arg *
2221 ctl_copyin_args(int num_args, struct ctl_be_arg *uargs,
2222 		char *error_str, size_t error_str_len)
2223 {
2224 	struct ctl_be_arg *args;
2225 	int i;
2226 
2227 	args = ctl_copyin_alloc(uargs, num_args * sizeof(*args),
2228 				error_str, error_str_len);
2229 
2230 	if (args == NULL)
2231 		goto bailout;
2232 
2233 	for (i = 0; i < num_args; i++) {
2234 		args[i].kname = NULL;
2235 		args[i].kvalue = NULL;
2236 	}
2237 
2238 	for (i = 0; i < num_args; i++) {
2239 		uint8_t *tmpptr;
2240 
2241 		args[i].kname = ctl_copyin_alloc(args[i].name,
2242 			args[i].namelen, error_str, error_str_len);
2243 		if (args[i].kname == NULL)
2244 			goto bailout;
2245 
2246 		if (args[i].kname[args[i].namelen - 1] != '\0') {
2247 			snprintf(error_str, error_str_len, "Argument %d "
2248 				 "name is not NUL-terminated", i);
2249 			goto bailout;
2250 		}
2251 
2252 		if (args[i].flags & CTL_BEARG_RD) {
2253 			tmpptr = ctl_copyin_alloc(args[i].value,
2254 				args[i].vallen, error_str, error_str_len);
2255 			if (tmpptr == NULL)
2256 				goto bailout;
2257 			if ((args[i].flags & CTL_BEARG_ASCII)
2258 			 && (tmpptr[args[i].vallen - 1] != '\0')) {
2259 				snprintf(error_str, error_str_len, "Argument "
2260 				    "%d value is not NUL-terminated", i);
2261 				goto bailout;
2262 			}
2263 			args[i].kvalue = tmpptr;
2264 		} else {
2265 			args[i].kvalue = malloc(args[i].vallen,
2266 			    M_CTL, M_WAITOK | M_ZERO);
2267 		}
2268 	}
2269 
2270 	return (args);
2271 bailout:
2272 
2273 	ctl_free_args(num_args, args);
2274 
2275 	return (NULL);
2276 }
2277 
2278 static void
2279 ctl_copyout_args(int num_args, struct ctl_be_arg *args)
2280 {
2281 	int i;
2282 
2283 	for (i = 0; i < num_args; i++) {
2284 		if (args[i].flags & CTL_BEARG_WR)
2285 			copyout(args[i].kvalue, args[i].value, args[i].vallen);
2286 	}
2287 }
2288 
2289 /*
2290  * Escape characters that are illegal or not recommended in XML.
2291  */
2292 int
2293 ctl_sbuf_printf_esc(struct sbuf *sb, char *str, int size)
2294 {
2295 	char *end = str + size;
2296 	int retval;
2297 
2298 	retval = 0;
2299 
2300 	for (; *str && str < end; str++) {
2301 		switch (*str) {
2302 		case '&':
2303 			retval = sbuf_printf(sb, "&amp;");
2304 			break;
2305 		case '>':
2306 			retval = sbuf_printf(sb, "&gt;");
2307 			break;
2308 		case '<':
2309 			retval = sbuf_printf(sb, "&lt;");
2310 			break;
2311 		default:
2312 			retval = sbuf_putc(sb, *str);
2313 			break;
2314 		}
2315 
2316 		if (retval != 0)
2317 			break;
2318 
2319 	}
2320 
2321 	return (retval);
2322 }
2323 
2324 static void
2325 ctl_id_sbuf(struct ctl_devid *id, struct sbuf *sb)
2326 {
2327 	struct scsi_vpd_id_descriptor *desc;
2328 	int i;
2329 
2330 	if (id == NULL || id->len < 4)
2331 		return;
2332 	desc = (struct scsi_vpd_id_descriptor *)id->data;
2333 	switch (desc->id_type & SVPD_ID_TYPE_MASK) {
2334 	case SVPD_ID_TYPE_T10:
2335 		sbuf_printf(sb, "t10.");
2336 		break;
2337 	case SVPD_ID_TYPE_EUI64:
2338 		sbuf_printf(sb, "eui.");
2339 		break;
2340 	case SVPD_ID_TYPE_NAA:
2341 		sbuf_printf(sb, "naa.");
2342 		break;
2343 	case SVPD_ID_TYPE_SCSI_NAME:
2344 		break;
2345 	}
2346 	switch (desc->proto_codeset & SVPD_ID_CODESET_MASK) {
2347 	case SVPD_ID_CODESET_BINARY:
2348 		for (i = 0; i < desc->length; i++)
2349 			sbuf_printf(sb, "%02x", desc->identifier[i]);
2350 		break;
2351 	case SVPD_ID_CODESET_ASCII:
2352 		sbuf_printf(sb, "%.*s", (int)desc->length,
2353 		    (char *)desc->identifier);
2354 		break;
2355 	case SVPD_ID_CODESET_UTF8:
2356 		sbuf_printf(sb, "%s", (char *)desc->identifier);
2357 		break;
2358 	}
2359 }
2360 
2361 static int
2362 ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2363 	  struct thread *td)
2364 {
2365 	struct ctl_softc *softc;
2366 	int retval;
2367 
2368 	softc = control_softc;
2369 
2370 	retval = 0;
2371 
2372 	switch (cmd) {
2373 	case CTL_IO: {
2374 		union ctl_io *io;
2375 		void *pool_tmp;
2376 
2377 		/*
2378 		 * If we haven't been "enabled", don't allow any SCSI I/O
2379 		 * to this FETD.
2380 		 */
2381 		if ((softc->ioctl_info.flags & CTL_IOCTL_FLAG_ENABLED) == 0) {
2382 			retval = EPERM;
2383 			break;
2384 		}
2385 
2386 		io = ctl_alloc_io(softc->ioctl_info.port.ctl_pool_ref);
2387 
2388 		/*
2389 		 * Need to save the pool reference so it doesn't get
2390 		 * spammed by the user's ctl_io.
2391 		 */
2392 		pool_tmp = io->io_hdr.pool;
2393 		memcpy(io, (void *)addr, sizeof(*io));
2394 		io->io_hdr.pool = pool_tmp;
2395 
2396 		/*
2397 		 * No status yet, so make sure the status is set properly.
2398 		 */
2399 		io->io_hdr.status = CTL_STATUS_NONE;
2400 
2401 		/*
2402 		 * The user sets the initiator ID, target and LUN IDs.
2403 		 */
2404 		io->io_hdr.nexus.targ_port = softc->ioctl_info.port.targ_port;
2405 		io->io_hdr.flags |= CTL_FLAG_USER_REQ;
2406 		if ((io->io_hdr.io_type == CTL_IO_SCSI)
2407 		 && (io->scsiio.tag_type != CTL_TAG_UNTAGGED))
2408 			io->scsiio.tag_num = softc->ioctl_info.cur_tag_num++;
2409 
2410 		retval = ctl_ioctl_submit_wait(io);
2411 
2412 		if (retval != 0) {
2413 			ctl_free_io(io);
2414 			break;
2415 		}
2416 
2417 		memcpy((void *)addr, io, sizeof(*io));
2418 
2419 		/* return this to our pool */
2420 		ctl_free_io(io);
2421 
2422 		break;
2423 	}
2424 	case CTL_ENABLE_PORT:
2425 	case CTL_DISABLE_PORT:
2426 	case CTL_SET_PORT_WWNS: {
2427 		struct ctl_port *port;
2428 		struct ctl_port_entry *entry;
2429 
2430 		entry = (struct ctl_port_entry *)addr;
2431 
2432 		mtx_lock(&softc->ctl_lock);
2433 		STAILQ_FOREACH(port, &softc->port_list, links) {
2434 			int action, done;
2435 
2436 			action = 0;
2437 			done = 0;
2438 
2439 			if ((entry->port_type == CTL_PORT_NONE)
2440 			 && (entry->targ_port == port->targ_port)) {
2441 				/*
2442 				 * If the user only wants to enable or
2443 				 * disable or set WWNs on a specific port,
2444 				 * do the operation and we're done.
2445 				 */
2446 				action = 1;
2447 				done = 1;
2448 			} else if (entry->port_type & port->port_type) {
2449 				/*
2450 				 * Compare the user's type mask with the
2451 				 * particular frontend type to see if we
2452 				 * have a match.
2453 				 */
2454 				action = 1;
2455 				done = 0;
2456 
2457 				/*
2458 				 * Make sure the user isn't trying to set
2459 				 * WWNs on multiple ports at the same time.
2460 				 */
2461 				if (cmd == CTL_SET_PORT_WWNS) {
2462 					printf("%s: Can't set WWNs on "
2463 					       "multiple ports\n", __func__);
2464 					retval = EINVAL;
2465 					break;
2466 				}
2467 			}
2468 			if (action != 0) {
2469 				/*
2470 				 * XXX KDM we have to drop the lock here,
2471 				 * because the online/offline operations
2472 				 * can potentially block.  We need to
2473 				 * reference count the frontends so they
2474 				 * can't go away,
2475 				 */
2476 				mtx_unlock(&softc->ctl_lock);
2477 
2478 				if (cmd == CTL_ENABLE_PORT) {
2479 					struct ctl_lun *lun;
2480 
2481 					STAILQ_FOREACH(lun, &softc->lun_list,
2482 						       links) {
2483 						port->lun_enable(port->targ_lun_arg,
2484 						    lun->target,
2485 						    lun->lun);
2486 					}
2487 
2488 					ctl_port_online(port);
2489 				} else if (cmd == CTL_DISABLE_PORT) {
2490 					struct ctl_lun *lun;
2491 
2492 					ctl_port_offline(port);
2493 
2494 					STAILQ_FOREACH(lun, &softc->lun_list,
2495 						       links) {
2496 						port->lun_disable(
2497 						    port->targ_lun_arg,
2498 						    lun->target,
2499 						    lun->lun);
2500 					}
2501 				}
2502 
2503 				mtx_lock(&softc->ctl_lock);
2504 
2505 				if (cmd == CTL_SET_PORT_WWNS)
2506 					ctl_port_set_wwns(port,
2507 					    (entry->flags & CTL_PORT_WWNN_VALID) ?
2508 					    1 : 0, entry->wwnn,
2509 					    (entry->flags & CTL_PORT_WWPN_VALID) ?
2510 					    1 : 0, entry->wwpn);
2511 			}
2512 			if (done != 0)
2513 				break;
2514 		}
2515 		mtx_unlock(&softc->ctl_lock);
2516 		break;
2517 	}
2518 	case CTL_GET_PORT_LIST: {
2519 		struct ctl_port *port;
2520 		struct ctl_port_list *list;
2521 		int i;
2522 
2523 		list = (struct ctl_port_list *)addr;
2524 
2525 		if (list->alloc_len != (list->alloc_num *
2526 		    sizeof(struct ctl_port_entry))) {
2527 			printf("%s: CTL_GET_PORT_LIST: alloc_len %u != "
2528 			       "alloc_num %u * sizeof(struct ctl_port_entry) "
2529 			       "%zu\n", __func__, list->alloc_len,
2530 			       list->alloc_num, sizeof(struct ctl_port_entry));
2531 			retval = EINVAL;
2532 			break;
2533 		}
2534 		list->fill_len = 0;
2535 		list->fill_num = 0;
2536 		list->dropped_num = 0;
2537 		i = 0;
2538 		mtx_lock(&softc->ctl_lock);
2539 		STAILQ_FOREACH(port, &softc->port_list, links) {
2540 			struct ctl_port_entry entry, *list_entry;
2541 
2542 			if (list->fill_num >= list->alloc_num) {
2543 				list->dropped_num++;
2544 				continue;
2545 			}
2546 
2547 			entry.port_type = port->port_type;
2548 			strlcpy(entry.port_name, port->port_name,
2549 				sizeof(entry.port_name));
2550 			entry.targ_port = port->targ_port;
2551 			entry.physical_port = port->physical_port;
2552 			entry.virtual_port = port->virtual_port;
2553 			entry.wwnn = port->wwnn;
2554 			entry.wwpn = port->wwpn;
2555 			if (port->status & CTL_PORT_STATUS_ONLINE)
2556 				entry.online = 1;
2557 			else
2558 				entry.online = 0;
2559 
2560 			list_entry = &list->entries[i];
2561 
2562 			retval = copyout(&entry, list_entry, sizeof(entry));
2563 			if (retval != 0) {
2564 				printf("%s: CTL_GET_PORT_LIST: copyout "
2565 				       "returned %d\n", __func__, retval);
2566 				break;
2567 			}
2568 			i++;
2569 			list->fill_num++;
2570 			list->fill_len += sizeof(entry);
2571 		}
2572 		mtx_unlock(&softc->ctl_lock);
2573 
2574 		/*
2575 		 * If this is non-zero, we had a copyout fault, so there's
2576 		 * probably no point in attempting to set the status inside
2577 		 * the structure.
2578 		 */
2579 		if (retval != 0)
2580 			break;
2581 
2582 		if (list->dropped_num > 0)
2583 			list->status = CTL_PORT_LIST_NEED_MORE_SPACE;
2584 		else
2585 			list->status = CTL_PORT_LIST_OK;
2586 		break;
2587 	}
2588 	case CTL_DUMP_OOA: {
2589 		struct ctl_lun *lun;
2590 		union ctl_io *io;
2591 		char printbuf[128];
2592 		struct sbuf sb;
2593 
2594 		mtx_lock(&softc->ctl_lock);
2595 		printf("Dumping OOA queues:\n");
2596 		STAILQ_FOREACH(lun, &softc->lun_list, links) {
2597 			mtx_lock(&lun->lun_lock);
2598 			for (io = (union ctl_io *)TAILQ_FIRST(
2599 			     &lun->ooa_queue); io != NULL;
2600 			     io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2601 			     ooa_links)) {
2602 				sbuf_new(&sb, printbuf, sizeof(printbuf),
2603 					 SBUF_FIXEDLEN);
2604 				sbuf_printf(&sb, "LUN %jd tag 0x%04x%s%s%s%s: ",
2605 					    (intmax_t)lun->lun,
2606 					    io->scsiio.tag_num,
2607 					    (io->io_hdr.flags &
2608 					    CTL_FLAG_BLOCKED) ? "" : " BLOCKED",
2609 					    (io->io_hdr.flags &
2610 					    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
2611 					    (io->io_hdr.flags &
2612 					    CTL_FLAG_ABORT) ? " ABORT" : "",
2613 			                    (io->io_hdr.flags &
2614 		                        CTL_FLAG_IS_WAS_ON_RTR) ? " RTR" : "");
2615 				ctl_scsi_command_string(&io->scsiio, NULL, &sb);
2616 				sbuf_finish(&sb);
2617 				printf("%s\n", sbuf_data(&sb));
2618 			}
2619 			mtx_unlock(&lun->lun_lock);
2620 		}
2621 		printf("OOA queues dump done\n");
2622 		mtx_unlock(&softc->ctl_lock);
2623 		break;
2624 	}
2625 	case CTL_GET_OOA: {
2626 		struct ctl_lun *lun;
2627 		struct ctl_ooa *ooa_hdr;
2628 		struct ctl_ooa_entry *entries;
2629 		uint32_t cur_fill_num;
2630 
2631 		ooa_hdr = (struct ctl_ooa *)addr;
2632 
2633 		if ((ooa_hdr->alloc_len == 0)
2634 		 || (ooa_hdr->alloc_num == 0)) {
2635 			printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2636 			       "must be non-zero\n", __func__,
2637 			       ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2638 			retval = EINVAL;
2639 			break;
2640 		}
2641 
2642 		if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2643 		    sizeof(struct ctl_ooa_entry))) {
2644 			printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2645 			       "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2646 			       __func__, ooa_hdr->alloc_len,
2647 			       ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2648 			retval = EINVAL;
2649 			break;
2650 		}
2651 
2652 		entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2653 		if (entries == NULL) {
2654 			printf("%s: could not allocate %d bytes for OOA "
2655 			       "dump\n", __func__, ooa_hdr->alloc_len);
2656 			retval = ENOMEM;
2657 			break;
2658 		}
2659 
2660 		mtx_lock(&softc->ctl_lock);
2661 		if (((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0)
2662 		 && ((ooa_hdr->lun_num >= CTL_MAX_LUNS)
2663 		  || (softc->ctl_luns[ooa_hdr->lun_num] == NULL))) {
2664 			mtx_unlock(&softc->ctl_lock);
2665 			free(entries, M_CTL);
2666 			printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2667 			       __func__, (uintmax_t)ooa_hdr->lun_num);
2668 			retval = EINVAL;
2669 			break;
2670 		}
2671 
2672 		cur_fill_num = 0;
2673 
2674 		if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2675 			STAILQ_FOREACH(lun, &softc->lun_list, links) {
2676 				retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2677 					ooa_hdr, entries);
2678 				if (retval != 0)
2679 					break;
2680 			}
2681 			if (retval != 0) {
2682 				mtx_unlock(&softc->ctl_lock);
2683 				free(entries, M_CTL);
2684 				break;
2685 			}
2686 		} else {
2687 			lun = softc->ctl_luns[ooa_hdr->lun_num];
2688 
2689 			retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,ooa_hdr,
2690 						    entries);
2691 		}
2692 		mtx_unlock(&softc->ctl_lock);
2693 
2694 		ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2695 		ooa_hdr->fill_len = ooa_hdr->fill_num *
2696 			sizeof(struct ctl_ooa_entry);
2697 		retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2698 		if (retval != 0) {
2699 			printf("%s: error copying out %d bytes for OOA dump\n",
2700 			       __func__, ooa_hdr->fill_len);
2701 		}
2702 
2703 		getbintime(&ooa_hdr->cur_bt);
2704 
2705 		if (cur_fill_num > ooa_hdr->alloc_num) {
2706 			ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2707 			ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2708 		} else {
2709 			ooa_hdr->dropped_num = 0;
2710 			ooa_hdr->status = CTL_OOA_OK;
2711 		}
2712 
2713 		free(entries, M_CTL);
2714 		break;
2715 	}
2716 	case CTL_CHECK_OOA: {
2717 		union ctl_io *io;
2718 		struct ctl_lun *lun;
2719 		struct ctl_ooa_info *ooa_info;
2720 
2721 
2722 		ooa_info = (struct ctl_ooa_info *)addr;
2723 
2724 		if (ooa_info->lun_id >= CTL_MAX_LUNS) {
2725 			ooa_info->status = CTL_OOA_INVALID_LUN;
2726 			break;
2727 		}
2728 		mtx_lock(&softc->ctl_lock);
2729 		lun = softc->ctl_luns[ooa_info->lun_id];
2730 		if (lun == NULL) {
2731 			mtx_unlock(&softc->ctl_lock);
2732 			ooa_info->status = CTL_OOA_INVALID_LUN;
2733 			break;
2734 		}
2735 		mtx_lock(&lun->lun_lock);
2736 		mtx_unlock(&softc->ctl_lock);
2737 		ooa_info->num_entries = 0;
2738 		for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
2739 		     io != NULL; io = (union ctl_io *)TAILQ_NEXT(
2740 		     &io->io_hdr, ooa_links)) {
2741 			ooa_info->num_entries++;
2742 		}
2743 		mtx_unlock(&lun->lun_lock);
2744 
2745 		ooa_info->status = CTL_OOA_SUCCESS;
2746 
2747 		break;
2748 	}
2749 	case CTL_HARD_START:
2750 	case CTL_HARD_STOP: {
2751 		struct ctl_fe_ioctl_startstop_info ss_info;
2752 		struct cfi_metatask *metatask;
2753 		struct mtx hs_mtx;
2754 
2755 		mtx_init(&hs_mtx, "HS Mutex", NULL, MTX_DEF);
2756 
2757 		cv_init(&ss_info.sem, "hard start/stop cv" );
2758 
2759 		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2760 		if (metatask == NULL) {
2761 			retval = ENOMEM;
2762 			mtx_destroy(&hs_mtx);
2763 			break;
2764 		}
2765 
2766 		if (cmd == CTL_HARD_START)
2767 			metatask->tasktype = CFI_TASK_STARTUP;
2768 		else
2769 			metatask->tasktype = CFI_TASK_SHUTDOWN;
2770 
2771 		metatask->callback = ctl_ioctl_hard_startstop_callback;
2772 		metatask->callback_arg = &ss_info;
2773 
2774 		cfi_action(metatask);
2775 
2776 		/* Wait for the callback */
2777 		mtx_lock(&hs_mtx);
2778 		cv_wait_sig(&ss_info.sem, &hs_mtx);
2779 		mtx_unlock(&hs_mtx);
2780 
2781 		/*
2782 		 * All information has been copied from the metatask by the
2783 		 * time cv_broadcast() is called, so we free the metatask here.
2784 		 */
2785 		cfi_free_metatask(metatask);
2786 
2787 		memcpy((void *)addr, &ss_info.hs_info, sizeof(ss_info.hs_info));
2788 
2789 		mtx_destroy(&hs_mtx);
2790 		break;
2791 	}
2792 	case CTL_BBRREAD: {
2793 		struct ctl_bbrread_info *bbr_info;
2794 		struct ctl_fe_ioctl_bbrread_info fe_bbr_info;
2795 		struct mtx bbr_mtx;
2796 		struct cfi_metatask *metatask;
2797 
2798 		bbr_info = (struct ctl_bbrread_info *)addr;
2799 
2800 		bzero(&fe_bbr_info, sizeof(fe_bbr_info));
2801 
2802 		bzero(&bbr_mtx, sizeof(bbr_mtx));
2803 		mtx_init(&bbr_mtx, "BBR Mutex", NULL, MTX_DEF);
2804 
2805 		fe_bbr_info.bbr_info = bbr_info;
2806 		fe_bbr_info.lock = &bbr_mtx;
2807 
2808 		cv_init(&fe_bbr_info.sem, "BBR read cv");
2809 		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2810 
2811 		if (metatask == NULL) {
2812 			mtx_destroy(&bbr_mtx);
2813 			cv_destroy(&fe_bbr_info.sem);
2814 			retval = ENOMEM;
2815 			break;
2816 		}
2817 		metatask->tasktype = CFI_TASK_BBRREAD;
2818 		metatask->callback = ctl_ioctl_bbrread_callback;
2819 		metatask->callback_arg = &fe_bbr_info;
2820 		metatask->taskinfo.bbrread.lun_num = bbr_info->lun_num;
2821 		metatask->taskinfo.bbrread.lba = bbr_info->lba;
2822 		metatask->taskinfo.bbrread.len = bbr_info->len;
2823 
2824 		cfi_action(metatask);
2825 
2826 		mtx_lock(&bbr_mtx);
2827 		while (fe_bbr_info.wakeup_done == 0)
2828 			cv_wait_sig(&fe_bbr_info.sem, &bbr_mtx);
2829 		mtx_unlock(&bbr_mtx);
2830 
2831 		bbr_info->status = metatask->status;
2832 		bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2833 		bbr_info->scsi_status = metatask->taskinfo.bbrread.scsi_status;
2834 		memcpy(&bbr_info->sense_data,
2835 		       &metatask->taskinfo.bbrread.sense_data,
2836 		       ctl_min(sizeof(bbr_info->sense_data),
2837 			       sizeof(metatask->taskinfo.bbrread.sense_data)));
2838 
2839 		cfi_free_metatask(metatask);
2840 
2841 		mtx_destroy(&bbr_mtx);
2842 		cv_destroy(&fe_bbr_info.sem);
2843 
2844 		break;
2845 	}
2846 	case CTL_DELAY_IO: {
2847 		struct ctl_io_delay_info *delay_info;
2848 #ifdef CTL_IO_DELAY
2849 		struct ctl_lun *lun;
2850 #endif /* CTL_IO_DELAY */
2851 
2852 		delay_info = (struct ctl_io_delay_info *)addr;
2853 
2854 #ifdef CTL_IO_DELAY
2855 		mtx_lock(&softc->ctl_lock);
2856 
2857 		if ((delay_info->lun_id >= CTL_MAX_LUNS)
2858 		 || (softc->ctl_luns[delay_info->lun_id] == NULL)) {
2859 			delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2860 		} else {
2861 			lun = softc->ctl_luns[delay_info->lun_id];
2862 			mtx_lock(&lun->lun_lock);
2863 
2864 			delay_info->status = CTL_DELAY_STATUS_OK;
2865 
2866 			switch (delay_info->delay_type) {
2867 			case CTL_DELAY_TYPE_CONT:
2868 				break;
2869 			case CTL_DELAY_TYPE_ONESHOT:
2870 				break;
2871 			default:
2872 				delay_info->status =
2873 					CTL_DELAY_STATUS_INVALID_TYPE;
2874 				break;
2875 			}
2876 
2877 			switch (delay_info->delay_loc) {
2878 			case CTL_DELAY_LOC_DATAMOVE:
2879 				lun->delay_info.datamove_type =
2880 					delay_info->delay_type;
2881 				lun->delay_info.datamove_delay =
2882 					delay_info->delay_secs;
2883 				break;
2884 			case CTL_DELAY_LOC_DONE:
2885 				lun->delay_info.done_type =
2886 					delay_info->delay_type;
2887 				lun->delay_info.done_delay =
2888 					delay_info->delay_secs;
2889 				break;
2890 			default:
2891 				delay_info->status =
2892 					CTL_DELAY_STATUS_INVALID_LOC;
2893 				break;
2894 			}
2895 			mtx_unlock(&lun->lun_lock);
2896 		}
2897 
2898 		mtx_unlock(&softc->ctl_lock);
2899 #else
2900 		delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2901 #endif /* CTL_IO_DELAY */
2902 		break;
2903 	}
2904 	case CTL_REALSYNC_SET: {
2905 		int *syncstate;
2906 
2907 		syncstate = (int *)addr;
2908 
2909 		mtx_lock(&softc->ctl_lock);
2910 		switch (*syncstate) {
2911 		case 0:
2912 			softc->flags &= ~CTL_FLAG_REAL_SYNC;
2913 			break;
2914 		case 1:
2915 			softc->flags |= CTL_FLAG_REAL_SYNC;
2916 			break;
2917 		default:
2918 			retval = EINVAL;
2919 			break;
2920 		}
2921 		mtx_unlock(&softc->ctl_lock);
2922 		break;
2923 	}
2924 	case CTL_REALSYNC_GET: {
2925 		int *syncstate;
2926 
2927 		syncstate = (int*)addr;
2928 
2929 		mtx_lock(&softc->ctl_lock);
2930 		if (softc->flags & CTL_FLAG_REAL_SYNC)
2931 			*syncstate = 1;
2932 		else
2933 			*syncstate = 0;
2934 		mtx_unlock(&softc->ctl_lock);
2935 
2936 		break;
2937 	}
2938 	case CTL_SETSYNC:
2939 	case CTL_GETSYNC: {
2940 		struct ctl_sync_info *sync_info;
2941 		struct ctl_lun *lun;
2942 
2943 		sync_info = (struct ctl_sync_info *)addr;
2944 
2945 		mtx_lock(&softc->ctl_lock);
2946 		lun = softc->ctl_luns[sync_info->lun_id];
2947 		if (lun == NULL) {
2948 			mtx_unlock(&softc->ctl_lock);
2949 			sync_info->status = CTL_GS_SYNC_NO_LUN;
2950 		}
2951 		/*
2952 		 * Get or set the sync interval.  We're not bounds checking
2953 		 * in the set case, hopefully the user won't do something
2954 		 * silly.
2955 		 */
2956 		mtx_lock(&lun->lun_lock);
2957 		mtx_unlock(&softc->ctl_lock);
2958 		if (cmd == CTL_GETSYNC)
2959 			sync_info->sync_interval = lun->sync_interval;
2960 		else
2961 			lun->sync_interval = sync_info->sync_interval;
2962 		mtx_unlock(&lun->lun_lock);
2963 
2964 		sync_info->status = CTL_GS_SYNC_OK;
2965 
2966 		break;
2967 	}
2968 	case CTL_GETSTATS: {
2969 		struct ctl_stats *stats;
2970 		struct ctl_lun *lun;
2971 		int i;
2972 
2973 		stats = (struct ctl_stats *)addr;
2974 
2975 		if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) >
2976 		     stats->alloc_len) {
2977 			stats->status = CTL_SS_NEED_MORE_SPACE;
2978 			stats->num_luns = softc->num_luns;
2979 			break;
2980 		}
2981 		/*
2982 		 * XXX KDM no locking here.  If the LUN list changes,
2983 		 * things can blow up.
2984 		 */
2985 		for (i = 0, lun = STAILQ_FIRST(&softc->lun_list); lun != NULL;
2986 		     i++, lun = STAILQ_NEXT(lun, links)) {
2987 			retval = copyout(&lun->stats, &stats->lun_stats[i],
2988 					 sizeof(lun->stats));
2989 			if (retval != 0)
2990 				break;
2991 		}
2992 		stats->num_luns = softc->num_luns;
2993 		stats->fill_len = sizeof(struct ctl_lun_io_stats) *
2994 				 softc->num_luns;
2995 		stats->status = CTL_SS_OK;
2996 #ifdef CTL_TIME_IO
2997 		stats->flags = CTL_STATS_FLAG_TIME_VALID;
2998 #else
2999 		stats->flags = CTL_STATS_FLAG_NONE;
3000 #endif
3001 		getnanouptime(&stats->timestamp);
3002 		break;
3003 	}
3004 	case CTL_ERROR_INJECT: {
3005 		struct ctl_error_desc *err_desc, *new_err_desc;
3006 		struct ctl_lun *lun;
3007 
3008 		err_desc = (struct ctl_error_desc *)addr;
3009 
3010 		new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
3011 				      M_WAITOK | M_ZERO);
3012 		bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
3013 
3014 		mtx_lock(&softc->ctl_lock);
3015 		lun = softc->ctl_luns[err_desc->lun_id];
3016 		if (lun == NULL) {
3017 			mtx_unlock(&softc->ctl_lock);
3018 			free(new_err_desc, M_CTL);
3019 			printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
3020 			       __func__, (uintmax_t)err_desc->lun_id);
3021 			retval = EINVAL;
3022 			break;
3023 		}
3024 		mtx_lock(&lun->lun_lock);
3025 		mtx_unlock(&softc->ctl_lock);
3026 
3027 		/*
3028 		 * We could do some checking here to verify the validity
3029 		 * of the request, but given the complexity of error
3030 		 * injection requests, the checking logic would be fairly
3031 		 * complex.
3032 		 *
3033 		 * For now, if the request is invalid, it just won't get
3034 		 * executed and might get deleted.
3035 		 */
3036 		STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
3037 
3038 		/*
3039 		 * XXX KDM check to make sure the serial number is unique,
3040 		 * in case we somehow manage to wrap.  That shouldn't
3041 		 * happen for a very long time, but it's the right thing to
3042 		 * do.
3043 		 */
3044 		new_err_desc->serial = lun->error_serial;
3045 		err_desc->serial = lun->error_serial;
3046 		lun->error_serial++;
3047 
3048 		mtx_unlock(&lun->lun_lock);
3049 		break;
3050 	}
3051 	case CTL_ERROR_INJECT_DELETE: {
3052 		struct ctl_error_desc *delete_desc, *desc, *desc2;
3053 		struct ctl_lun *lun;
3054 		int delete_done;
3055 
3056 		delete_desc = (struct ctl_error_desc *)addr;
3057 		delete_done = 0;
3058 
3059 		mtx_lock(&softc->ctl_lock);
3060 		lun = softc->ctl_luns[delete_desc->lun_id];
3061 		if (lun == NULL) {
3062 			mtx_unlock(&softc->ctl_lock);
3063 			printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
3064 			       __func__, (uintmax_t)delete_desc->lun_id);
3065 			retval = EINVAL;
3066 			break;
3067 		}
3068 		mtx_lock(&lun->lun_lock);
3069 		mtx_unlock(&softc->ctl_lock);
3070 		STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
3071 			if (desc->serial != delete_desc->serial)
3072 				continue;
3073 
3074 			STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
3075 				      links);
3076 			free(desc, M_CTL);
3077 			delete_done = 1;
3078 		}
3079 		mtx_unlock(&lun->lun_lock);
3080 		if (delete_done == 0) {
3081 			printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
3082 			       "error serial %ju on LUN %u\n", __func__,
3083 			       delete_desc->serial, delete_desc->lun_id);
3084 			retval = EINVAL;
3085 			break;
3086 		}
3087 		break;
3088 	}
3089 	case CTL_DUMP_STRUCTS: {
3090 		int i, j, k;
3091 		struct ctl_port *port;
3092 		struct ctl_frontend *fe;
3093 
3094 		mtx_lock(&softc->ctl_lock);
3095 		printf("CTL Persistent Reservation information start:\n");
3096 		for (i = 0; i < CTL_MAX_LUNS; i++) {
3097 			struct ctl_lun *lun;
3098 
3099 			lun = softc->ctl_luns[i];
3100 
3101 			if ((lun == NULL)
3102 			 || ((lun->flags & CTL_LUN_DISABLED) != 0))
3103 				continue;
3104 
3105 			for (j = 0; j < (CTL_MAX_PORTS * 2); j++) {
3106 				if (lun->pr_keys[j] == NULL)
3107 					continue;
3108 				for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
3109 					if (lun->pr_keys[j][k] == 0)
3110 						continue;
3111 					printf("  LUN %d port %d iid %d key "
3112 					       "%#jx\n", i, j, k,
3113 					       (uintmax_t)lun->pr_keys[j][k]);
3114 				}
3115 			}
3116 		}
3117 		printf("CTL Persistent Reservation information end\n");
3118 		printf("CTL Ports:\n");
3119 		STAILQ_FOREACH(port, &softc->port_list, links) {
3120 			printf("  Port %d '%s' Frontend '%s' Type %u pp %d vp %d WWNN "
3121 			       "%#jx WWPN %#jx\n", port->targ_port, port->port_name,
3122 			       port->frontend->name, port->port_type,
3123 			       port->physical_port, port->virtual_port,
3124 			       (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
3125 			for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3126 				if (port->wwpn_iid[j].in_use == 0 &&
3127 				    port->wwpn_iid[j].wwpn == 0 &&
3128 				    port->wwpn_iid[j].name == NULL)
3129 					continue;
3130 
3131 				printf("    iid %u use %d WWPN %#jx '%s'\n",
3132 				    j, port->wwpn_iid[j].in_use,
3133 				    (uintmax_t)port->wwpn_iid[j].wwpn,
3134 				    port->wwpn_iid[j].name);
3135 			}
3136 		}
3137 		printf("CTL Port information end\n");
3138 		mtx_unlock(&softc->ctl_lock);
3139 		/*
3140 		 * XXX KDM calling this without a lock.  We'd likely want
3141 		 * to drop the lock before calling the frontend's dump
3142 		 * routine anyway.
3143 		 */
3144 		printf("CTL Frontends:\n");
3145 		STAILQ_FOREACH(fe, &softc->fe_list, links) {
3146 			printf("  Frontend '%s'\n", fe->name);
3147 			if (fe->fe_dump != NULL)
3148 				fe->fe_dump();
3149 		}
3150 		printf("CTL Frontend information end\n");
3151 		break;
3152 	}
3153 	case CTL_LUN_REQ: {
3154 		struct ctl_lun_req *lun_req;
3155 		struct ctl_backend_driver *backend;
3156 
3157 		lun_req = (struct ctl_lun_req *)addr;
3158 
3159 		backend = ctl_backend_find(lun_req->backend);
3160 		if (backend == NULL) {
3161 			lun_req->status = CTL_LUN_ERROR;
3162 			snprintf(lun_req->error_str,
3163 				 sizeof(lun_req->error_str),
3164 				 "Backend \"%s\" not found.",
3165 				 lun_req->backend);
3166 			break;
3167 		}
3168 		if (lun_req->num_be_args > 0) {
3169 			lun_req->kern_be_args = ctl_copyin_args(
3170 				lun_req->num_be_args,
3171 				lun_req->be_args,
3172 				lun_req->error_str,
3173 				sizeof(lun_req->error_str));
3174 			if (lun_req->kern_be_args == NULL) {
3175 				lun_req->status = CTL_LUN_ERROR;
3176 				break;
3177 			}
3178 		}
3179 
3180 		retval = backend->ioctl(dev, cmd, addr, flag, td);
3181 
3182 		if (lun_req->num_be_args > 0) {
3183 			ctl_copyout_args(lun_req->num_be_args,
3184 				      lun_req->kern_be_args);
3185 			ctl_free_args(lun_req->num_be_args,
3186 				      lun_req->kern_be_args);
3187 		}
3188 		break;
3189 	}
3190 	case CTL_LUN_LIST: {
3191 		struct sbuf *sb;
3192 		struct ctl_lun *lun;
3193 		struct ctl_lun_list *list;
3194 		struct ctl_option *opt;
3195 
3196 		list = (struct ctl_lun_list *)addr;
3197 
3198 		/*
3199 		 * Allocate a fixed length sbuf here, based on the length
3200 		 * of the user's buffer.  We could allocate an auto-extending
3201 		 * buffer, and then tell the user how much larger our
3202 		 * amount of data is than his buffer, but that presents
3203 		 * some problems:
3204 		 *
3205 		 * 1.  The sbuf(9) routines use a blocking malloc, and so
3206 		 *     we can't hold a lock while calling them with an
3207 		 *     auto-extending buffer.
3208  		 *
3209 		 * 2.  There is not currently a LUN reference counting
3210 		 *     mechanism, outside of outstanding transactions on
3211 		 *     the LUN's OOA queue.  So a LUN could go away on us
3212 		 *     while we're getting the LUN number, backend-specific
3213 		 *     information, etc.  Thus, given the way things
3214 		 *     currently work, we need to hold the CTL lock while
3215 		 *     grabbing LUN information.
3216 		 *
3217 		 * So, from the user's standpoint, the best thing to do is
3218 		 * allocate what he thinks is a reasonable buffer length,
3219 		 * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
3220 		 * double the buffer length and try again.  (And repeat
3221 		 * that until he succeeds.)
3222 		 */
3223 		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3224 		if (sb == NULL) {
3225 			list->status = CTL_LUN_LIST_ERROR;
3226 			snprintf(list->error_str, sizeof(list->error_str),
3227 				 "Unable to allocate %d bytes for LUN list",
3228 				 list->alloc_len);
3229 			break;
3230 		}
3231 
3232 		sbuf_printf(sb, "<ctllunlist>\n");
3233 
3234 		mtx_lock(&softc->ctl_lock);
3235 		STAILQ_FOREACH(lun, &softc->lun_list, links) {
3236 			mtx_lock(&lun->lun_lock);
3237 			retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3238 					     (uintmax_t)lun->lun);
3239 
3240 			/*
3241 			 * Bail out as soon as we see that we've overfilled
3242 			 * the buffer.
3243 			 */
3244 			if (retval != 0)
3245 				break;
3246 
3247 			retval = sbuf_printf(sb, "\t<backend_type>%s"
3248 					     "</backend_type>\n",
3249 					     (lun->backend == NULL) ?  "none" :
3250 					     lun->backend->name);
3251 
3252 			if (retval != 0)
3253 				break;
3254 
3255 			retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
3256 					     lun->be_lun->lun_type);
3257 
3258 			if (retval != 0)
3259 				break;
3260 
3261 			if (lun->backend == NULL) {
3262 				retval = sbuf_printf(sb, "</lun>\n");
3263 				if (retval != 0)
3264 					break;
3265 				continue;
3266 			}
3267 
3268 			retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
3269 					     (lun->be_lun->maxlba > 0) ?
3270 					     lun->be_lun->maxlba + 1 : 0);
3271 
3272 			if (retval != 0)
3273 				break;
3274 
3275 			retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
3276 					     lun->be_lun->blocksize);
3277 
3278 			if (retval != 0)
3279 				break;
3280 
3281 			retval = sbuf_printf(sb, "\t<serial_number>");
3282 
3283 			if (retval != 0)
3284 				break;
3285 
3286 			retval = ctl_sbuf_printf_esc(sb,
3287 			    lun->be_lun->serial_num,
3288 			    sizeof(lun->be_lun->serial_num));
3289 
3290 			if (retval != 0)
3291 				break;
3292 
3293 			retval = sbuf_printf(sb, "</serial_number>\n");
3294 
3295 			if (retval != 0)
3296 				break;
3297 
3298 			retval = sbuf_printf(sb, "\t<device_id>");
3299 
3300 			if (retval != 0)
3301 				break;
3302 
3303 			retval = ctl_sbuf_printf_esc(sb,
3304 			    lun->be_lun->device_id,
3305 			    sizeof(lun->be_lun->device_id));
3306 
3307 			if (retval != 0)
3308 				break;
3309 
3310 			retval = sbuf_printf(sb, "</device_id>\n");
3311 
3312 			if (retval != 0)
3313 				break;
3314 
3315 			if (lun->backend->lun_info != NULL) {
3316 				retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3317 				if (retval != 0)
3318 					break;
3319 			}
3320 			STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
3321 				retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3322 				    opt->name, opt->value, opt->name);
3323 				if (retval != 0)
3324 					break;
3325 			}
3326 
3327 			retval = sbuf_printf(sb, "</lun>\n");
3328 
3329 			if (retval != 0)
3330 				break;
3331 			mtx_unlock(&lun->lun_lock);
3332 		}
3333 		if (lun != NULL)
3334 			mtx_unlock(&lun->lun_lock);
3335 		mtx_unlock(&softc->ctl_lock);
3336 
3337 		if ((retval != 0)
3338 		 || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3339 			retval = 0;
3340 			sbuf_delete(sb);
3341 			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3342 			snprintf(list->error_str, sizeof(list->error_str),
3343 				 "Out of space, %d bytes is too small",
3344 				 list->alloc_len);
3345 			break;
3346 		}
3347 
3348 		sbuf_finish(sb);
3349 
3350 		retval = copyout(sbuf_data(sb), list->lun_xml,
3351 				 sbuf_len(sb) + 1);
3352 
3353 		list->fill_len = sbuf_len(sb) + 1;
3354 		list->status = CTL_LUN_LIST_OK;
3355 		sbuf_delete(sb);
3356 		break;
3357 	}
3358 	case CTL_ISCSI: {
3359 		struct ctl_iscsi *ci;
3360 		struct ctl_frontend *fe;
3361 
3362 		ci = (struct ctl_iscsi *)addr;
3363 
3364 		fe = ctl_frontend_find("iscsi");
3365 		if (fe == NULL) {
3366 			ci->status = CTL_ISCSI_ERROR;
3367 			snprintf(ci->error_str, sizeof(ci->error_str),
3368 			    "Frontend \"iscsi\" not found.");
3369 			break;
3370 		}
3371 
3372 		retval = fe->ioctl(dev, cmd, addr, flag, td);
3373 		break;
3374 	}
3375 	case CTL_PORT_REQ: {
3376 		struct ctl_req *req;
3377 		struct ctl_frontend *fe;
3378 
3379 		req = (struct ctl_req *)addr;
3380 
3381 		fe = ctl_frontend_find(req->driver);
3382 		if (fe == NULL) {
3383 			req->status = CTL_LUN_ERROR;
3384 			snprintf(req->error_str, sizeof(req->error_str),
3385 			    "Frontend \"%s\" not found.", req->driver);
3386 			break;
3387 		}
3388 		if (req->num_args > 0) {
3389 			req->kern_args = ctl_copyin_args(req->num_args,
3390 			    req->args, req->error_str, sizeof(req->error_str));
3391 			if (req->kern_args == NULL) {
3392 				req->status = CTL_LUN_ERROR;
3393 				break;
3394 			}
3395 		}
3396 
3397 		retval = fe->ioctl(dev, cmd, addr, flag, td);
3398 
3399 		if (req->num_args > 0) {
3400 			ctl_copyout_args(req->num_args, req->kern_args);
3401 			ctl_free_args(req->num_args, req->kern_args);
3402 		}
3403 		break;
3404 	}
3405 	case CTL_PORT_LIST: {
3406 		struct sbuf *sb;
3407 		struct ctl_port *port;
3408 		struct ctl_lun_list *list;
3409 		struct ctl_option *opt;
3410 		int j;
3411 
3412 		list = (struct ctl_lun_list *)addr;
3413 
3414 		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3415 		if (sb == NULL) {
3416 			list->status = CTL_LUN_LIST_ERROR;
3417 			snprintf(list->error_str, sizeof(list->error_str),
3418 				 "Unable to allocate %d bytes for LUN list",
3419 				 list->alloc_len);
3420 			break;
3421 		}
3422 
3423 		sbuf_printf(sb, "<ctlportlist>\n");
3424 
3425 		mtx_lock(&softc->ctl_lock);
3426 		STAILQ_FOREACH(port, &softc->port_list, links) {
3427 			retval = sbuf_printf(sb, "<targ_port id=\"%ju\">\n",
3428 					     (uintmax_t)port->targ_port);
3429 
3430 			/*
3431 			 * Bail out as soon as we see that we've overfilled
3432 			 * the buffer.
3433 			 */
3434 			if (retval != 0)
3435 				break;
3436 
3437 			retval = sbuf_printf(sb, "\t<frontend_type>%s"
3438 			    "</frontend_type>\n", port->frontend->name);
3439 			if (retval != 0)
3440 				break;
3441 
3442 			retval = sbuf_printf(sb, "\t<port_type>%d</port_type>\n",
3443 					     port->port_type);
3444 			if (retval != 0)
3445 				break;
3446 
3447 			retval = sbuf_printf(sb, "\t<online>%s</online>\n",
3448 			    (port->status & CTL_PORT_STATUS_ONLINE) ? "YES" : "NO");
3449 			if (retval != 0)
3450 				break;
3451 
3452 			retval = sbuf_printf(sb, "\t<port_name>%s</port_name>\n",
3453 			    port->port_name);
3454 			if (retval != 0)
3455 				break;
3456 
3457 			retval = sbuf_printf(sb, "\t<physical_port>%d</physical_port>\n",
3458 			    port->physical_port);
3459 			if (retval != 0)
3460 				break;
3461 
3462 			retval = sbuf_printf(sb, "\t<virtual_port>%d</virtual_port>\n",
3463 			    port->virtual_port);
3464 			if (retval != 0)
3465 				break;
3466 
3467 			if (port->target_devid != NULL) {
3468 				sbuf_printf(sb, "\t<target>");
3469 				ctl_id_sbuf(port->target_devid, sb);
3470 				sbuf_printf(sb, "</target>\n");
3471 			}
3472 
3473 			if (port->port_devid != NULL) {
3474 				sbuf_printf(sb, "\t<port>");
3475 				ctl_id_sbuf(port->port_devid, sb);
3476 				sbuf_printf(sb, "</port>\n");
3477 			}
3478 
3479 			if (port->port_info != NULL) {
3480 				retval = port->port_info(port->onoff_arg, sb);
3481 				if (retval != 0)
3482 					break;
3483 			}
3484 			STAILQ_FOREACH(opt, &port->options, links) {
3485 				retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3486 				    opt->name, opt->value, opt->name);
3487 				if (retval != 0)
3488 					break;
3489 			}
3490 
3491 			for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3492 				if (port->wwpn_iid[j].in_use == 0 ||
3493 				    (port->wwpn_iid[j].wwpn == 0 &&
3494 				     port->wwpn_iid[j].name == NULL))
3495 					continue;
3496 
3497 				if (port->wwpn_iid[j].name != NULL)
3498 					retval = sbuf_printf(sb,
3499 					    "\t<initiator>%u %s</initiator>\n",
3500 					    j, port->wwpn_iid[j].name);
3501 				else
3502 					retval = sbuf_printf(sb,
3503 					    "\t<initiator>%u naa.%08jx</initiator>\n",
3504 					    j, port->wwpn_iid[j].wwpn);
3505 				if (retval != 0)
3506 					break;
3507 			}
3508 			if (retval != 0)
3509 				break;
3510 
3511 			retval = sbuf_printf(sb, "</targ_port>\n");
3512 			if (retval != 0)
3513 				break;
3514 		}
3515 		mtx_unlock(&softc->ctl_lock);
3516 
3517 		if ((retval != 0)
3518 		 || ((retval = sbuf_printf(sb, "</ctlportlist>\n")) != 0)) {
3519 			retval = 0;
3520 			sbuf_delete(sb);
3521 			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3522 			snprintf(list->error_str, sizeof(list->error_str),
3523 				 "Out of space, %d bytes is too small",
3524 				 list->alloc_len);
3525 			break;
3526 		}
3527 
3528 		sbuf_finish(sb);
3529 
3530 		retval = copyout(sbuf_data(sb), list->lun_xml,
3531 				 sbuf_len(sb) + 1);
3532 
3533 		list->fill_len = sbuf_len(sb) + 1;
3534 		list->status = CTL_LUN_LIST_OK;
3535 		sbuf_delete(sb);
3536 		break;
3537 	}
3538 	default: {
3539 		/* XXX KDM should we fix this? */
3540 #if 0
3541 		struct ctl_backend_driver *backend;
3542 		unsigned int type;
3543 		int found;
3544 
3545 		found = 0;
3546 
3547 		/*
3548 		 * We encode the backend type as the ioctl type for backend
3549 		 * ioctls.  So parse it out here, and then search for a
3550 		 * backend of this type.
3551 		 */
3552 		type = _IOC_TYPE(cmd);
3553 
3554 		STAILQ_FOREACH(backend, &softc->be_list, links) {
3555 			if (backend->type == type) {
3556 				found = 1;
3557 				break;
3558 			}
3559 		}
3560 		if (found == 0) {
3561 			printf("ctl: unknown ioctl command %#lx or backend "
3562 			       "%d\n", cmd, type);
3563 			retval = EINVAL;
3564 			break;
3565 		}
3566 		retval = backend->ioctl(dev, cmd, addr, flag, td);
3567 #endif
3568 		retval = ENOTTY;
3569 		break;
3570 	}
3571 	}
3572 	return (retval);
3573 }
3574 
3575 uint32_t
3576 ctl_get_initindex(struct ctl_nexus *nexus)
3577 {
3578 	if (nexus->targ_port < CTL_MAX_PORTS)
3579 		return (nexus->initid.id +
3580 			(nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3581 	else
3582 		return (nexus->initid.id +
3583 		       ((nexus->targ_port - CTL_MAX_PORTS) *
3584 			CTL_MAX_INIT_PER_PORT));
3585 }
3586 
3587 uint32_t
3588 ctl_get_resindex(struct ctl_nexus *nexus)
3589 {
3590 	return (nexus->initid.id + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3591 }
3592 
3593 uint32_t
3594 ctl_port_idx(int port_num)
3595 {
3596 	if (port_num < CTL_MAX_PORTS)
3597 		return(port_num);
3598 	else
3599 		return(port_num - CTL_MAX_PORTS);
3600 }
3601 
3602 static uint32_t
3603 ctl_map_lun(int port_num, uint32_t lun_id)
3604 {
3605 	struct ctl_port *port;
3606 
3607 	port = control_softc->ctl_ports[ctl_port_idx(port_num)];
3608 	if (port == NULL)
3609 		return (UINT32_MAX);
3610 	if (port->lun_map == NULL)
3611 		return (lun_id);
3612 	return (port->lun_map(port->targ_lun_arg, lun_id));
3613 }
3614 
3615 static uint32_t
3616 ctl_map_lun_back(int port_num, uint32_t lun_id)
3617 {
3618 	struct ctl_port *port;
3619 	uint32_t i;
3620 
3621 	port = control_softc->ctl_ports[ctl_port_idx(port_num)];
3622 	if (port->lun_map == NULL)
3623 		return (lun_id);
3624 	for (i = 0; i < CTL_MAX_LUNS; i++) {
3625 		if (port->lun_map(port->targ_lun_arg, i) == lun_id)
3626 			return (i);
3627 	}
3628 	return (UINT32_MAX);
3629 }
3630 
3631 /*
3632  * Note:  This only works for bitmask sizes that are at least 32 bits, and
3633  * that are a power of 2.
3634  */
3635 int
3636 ctl_ffz(uint32_t *mask, uint32_t size)
3637 {
3638 	uint32_t num_chunks, num_pieces;
3639 	int i, j;
3640 
3641 	num_chunks = (size >> 5);
3642 	if (num_chunks == 0)
3643 		num_chunks++;
3644 	num_pieces = ctl_min((sizeof(uint32_t) * 8), size);
3645 
3646 	for (i = 0; i < num_chunks; i++) {
3647 		for (j = 0; j < num_pieces; j++) {
3648 			if ((mask[i] & (1 << j)) == 0)
3649 				return ((i << 5) + j);
3650 		}
3651 	}
3652 
3653 	return (-1);
3654 }
3655 
3656 int
3657 ctl_set_mask(uint32_t *mask, uint32_t bit)
3658 {
3659 	uint32_t chunk, piece;
3660 
3661 	chunk = bit >> 5;
3662 	piece = bit % (sizeof(uint32_t) * 8);
3663 
3664 	if ((mask[chunk] & (1 << piece)) != 0)
3665 		return (-1);
3666 	else
3667 		mask[chunk] |= (1 << piece);
3668 
3669 	return (0);
3670 }
3671 
3672 int
3673 ctl_clear_mask(uint32_t *mask, uint32_t bit)
3674 {
3675 	uint32_t chunk, piece;
3676 
3677 	chunk = bit >> 5;
3678 	piece = bit % (sizeof(uint32_t) * 8);
3679 
3680 	if ((mask[chunk] & (1 << piece)) == 0)
3681 		return (-1);
3682 	else
3683 		mask[chunk] &= ~(1 << piece);
3684 
3685 	return (0);
3686 }
3687 
3688 int
3689 ctl_is_set(uint32_t *mask, uint32_t bit)
3690 {
3691 	uint32_t chunk, piece;
3692 
3693 	chunk = bit >> 5;
3694 	piece = bit % (sizeof(uint32_t) * 8);
3695 
3696 	if ((mask[chunk] & (1 << piece)) == 0)
3697 		return (0);
3698 	else
3699 		return (1);
3700 }
3701 
3702 static uint64_t
3703 ctl_get_prkey(struct ctl_lun *lun, uint32_t residx)
3704 {
3705 	uint64_t *t;
3706 
3707 	t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3708 	if (t == NULL)
3709 		return (0);
3710 	return (t[residx % CTL_MAX_INIT_PER_PORT]);
3711 }
3712 
3713 static void
3714 ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx)
3715 {
3716 	uint64_t *t;
3717 
3718 	t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3719 	if (t == NULL)
3720 		return;
3721 	t[residx % CTL_MAX_INIT_PER_PORT] = 0;
3722 }
3723 
3724 static void
3725 ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx)
3726 {
3727 	uint64_t *p;
3728 	u_int i;
3729 
3730 	i = residx/CTL_MAX_INIT_PER_PORT;
3731 	if (lun->pr_keys[i] != NULL)
3732 		return;
3733 	mtx_unlock(&lun->lun_lock);
3734 	p = malloc(sizeof(uint64_t) * CTL_MAX_INIT_PER_PORT, M_CTL,
3735 	    M_WAITOK | M_ZERO);
3736 	mtx_lock(&lun->lun_lock);
3737 	if (lun->pr_keys[i] == NULL)
3738 		lun->pr_keys[i] = p;
3739 	else
3740 		free(p, M_CTL);
3741 }
3742 
3743 static void
3744 ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key)
3745 {
3746 	uint64_t *t;
3747 
3748 	t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3749 	KASSERT(t != NULL, ("prkey %d is not allocated", residx));
3750 	t[residx % CTL_MAX_INIT_PER_PORT] = key;
3751 }
3752 
3753 /*
3754  * ctl_softc, pool_name, total_ctl_io are passed in.
3755  * npool is passed out.
3756  */
3757 int
3758 ctl_pool_create(struct ctl_softc *ctl_softc, const char *pool_name,
3759 		uint32_t total_ctl_io, void **npool)
3760 {
3761 #ifdef IO_POOLS
3762 	struct ctl_io_pool *pool;
3763 
3764 	pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3765 					    M_NOWAIT | M_ZERO);
3766 	if (pool == NULL)
3767 		return (ENOMEM);
3768 
3769 	snprintf(pool->name, sizeof(pool->name), "CTL IO %s", pool_name);
3770 	pool->ctl_softc = ctl_softc;
3771 	pool->zone = uma_zsecond_create(pool->name, NULL,
3772 	    NULL, NULL, NULL, ctl_softc->io_zone);
3773 	/* uma_prealloc(pool->zone, total_ctl_io); */
3774 
3775 	*npool = pool;
3776 #else
3777 	*npool = ctl_softc->io_zone;
3778 #endif
3779 	return (0);
3780 }
3781 
3782 void
3783 ctl_pool_free(struct ctl_io_pool *pool)
3784 {
3785 
3786 	if (pool == NULL)
3787 		return;
3788 
3789 #ifdef IO_POOLS
3790 	uma_zdestroy(pool->zone);
3791 	free(pool, M_CTL);
3792 #endif
3793 }
3794 
3795 union ctl_io *
3796 ctl_alloc_io(void *pool_ref)
3797 {
3798 	union ctl_io *io;
3799 #ifdef IO_POOLS
3800 	struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3801 
3802 	io = uma_zalloc(pool->zone, M_WAITOK);
3803 #else
3804 	io = uma_zalloc((uma_zone_t)pool_ref, M_WAITOK);
3805 #endif
3806 	if (io != NULL)
3807 		io->io_hdr.pool = pool_ref;
3808 	return (io);
3809 }
3810 
3811 union ctl_io *
3812 ctl_alloc_io_nowait(void *pool_ref)
3813 {
3814 	union ctl_io *io;
3815 #ifdef IO_POOLS
3816 	struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3817 
3818 	io = uma_zalloc(pool->zone, M_NOWAIT);
3819 #else
3820 	io = uma_zalloc((uma_zone_t)pool_ref, M_NOWAIT);
3821 #endif
3822 	if (io != NULL)
3823 		io->io_hdr.pool = pool_ref;
3824 	return (io);
3825 }
3826 
3827 void
3828 ctl_free_io(union ctl_io *io)
3829 {
3830 #ifdef IO_POOLS
3831 	struct ctl_io_pool *pool;
3832 #endif
3833 
3834 	if (io == NULL)
3835 		return;
3836 
3837 #ifdef IO_POOLS
3838 	pool = (struct ctl_io_pool *)io->io_hdr.pool;
3839 	uma_zfree(pool->zone, io);
3840 #else
3841 	uma_zfree((uma_zone_t)io->io_hdr.pool, io);
3842 #endif
3843 }
3844 
3845 void
3846 ctl_zero_io(union ctl_io *io)
3847 {
3848 	void *pool_ref;
3849 
3850 	if (io == NULL)
3851 		return;
3852 
3853 	/*
3854 	 * May need to preserve linked list pointers at some point too.
3855 	 */
3856 	pool_ref = io->io_hdr.pool;
3857 	memset(io, 0, sizeof(*io));
3858 	io->io_hdr.pool = pool_ref;
3859 }
3860 
3861 /*
3862  * This routine is currently used for internal copies of ctl_ios that need
3863  * to persist for some reason after we've already returned status to the
3864  * FETD.  (Thus the flag set.)
3865  *
3866  * XXX XXX
3867  * Note that this makes a blind copy of all fields in the ctl_io, except
3868  * for the pool reference.  This includes any memory that has been
3869  * allocated!  That memory will no longer be valid after done has been
3870  * called, so this would be VERY DANGEROUS for command that actually does
3871  * any reads or writes.  Right now (11/7/2005), this is only used for immediate
3872  * start and stop commands, which don't transfer any data, so this is not a
3873  * problem.  If it is used for anything else, the caller would also need to
3874  * allocate data buffer space and this routine would need to be modified to
3875  * copy the data buffer(s) as well.
3876  */
3877 void
3878 ctl_copy_io(union ctl_io *src, union ctl_io *dest)
3879 {
3880 	void *pool_ref;
3881 
3882 	if ((src == NULL)
3883 	 || (dest == NULL))
3884 		return;
3885 
3886 	/*
3887 	 * May need to preserve linked list pointers at some point too.
3888 	 */
3889 	pool_ref = dest->io_hdr.pool;
3890 
3891 	memcpy(dest, src, ctl_min(sizeof(*src), sizeof(*dest)));
3892 
3893 	dest->io_hdr.pool = pool_ref;
3894 	/*
3895 	 * We need to know that this is an internal copy, and doesn't need
3896 	 * to get passed back to the FETD that allocated it.
3897 	 */
3898 	dest->io_hdr.flags |= CTL_FLAG_INT_COPY;
3899 }
3900 
3901 static int
3902 ctl_expand_number(const char *buf, uint64_t *num)
3903 {
3904 	char *endptr;
3905 	uint64_t number;
3906 	unsigned shift;
3907 
3908 	number = strtoq(buf, &endptr, 0);
3909 
3910 	switch (tolower((unsigned char)*endptr)) {
3911 	case 'e':
3912 		shift = 60;
3913 		break;
3914 	case 'p':
3915 		shift = 50;
3916 		break;
3917 	case 't':
3918 		shift = 40;
3919 		break;
3920 	case 'g':
3921 		shift = 30;
3922 		break;
3923 	case 'm':
3924 		shift = 20;
3925 		break;
3926 	case 'k':
3927 		shift = 10;
3928 		break;
3929 	case 'b':
3930 	case '\0': /* No unit. */
3931 		*num = number;
3932 		return (0);
3933 	default:
3934 		/* Unrecognized unit. */
3935 		return (-1);
3936 	}
3937 
3938 	if ((number << shift) >> shift != number) {
3939 		/* Overflow */
3940 		return (-1);
3941 	}
3942 	*num = number << shift;
3943 	return (0);
3944 }
3945 
3946 
3947 /*
3948  * This routine could be used in the future to load default and/or saved
3949  * mode page parameters for a particuar lun.
3950  */
3951 static int
3952 ctl_init_page_index(struct ctl_lun *lun)
3953 {
3954 	int i;
3955 	struct ctl_page_index *page_index;
3956 	const char *value;
3957 	uint64_t ival;
3958 
3959 	memcpy(&lun->mode_pages.index, page_index_template,
3960 	       sizeof(page_index_template));
3961 
3962 	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
3963 
3964 		page_index = &lun->mode_pages.index[i];
3965 		/*
3966 		 * If this is a disk-only mode page, there's no point in
3967 		 * setting it up.  For some pages, we have to have some
3968 		 * basic information about the disk in order to calculate the
3969 		 * mode page data.
3970 		 */
3971 		if ((lun->be_lun->lun_type != T_DIRECT)
3972 		 && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
3973 			continue;
3974 
3975 		switch (page_index->page_code & SMPH_PC_MASK) {
3976 		case SMS_RW_ERROR_RECOVERY_PAGE: {
3977 			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3978 				panic("subpage is incorrect!");
3979 			memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT],
3980 			       &rw_er_page_default,
3981 			       sizeof(rw_er_page_default));
3982 			memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CHANGEABLE],
3983 			       &rw_er_page_changeable,
3984 			       sizeof(rw_er_page_changeable));
3985 			memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_DEFAULT],
3986 			       &rw_er_page_default,
3987 			       sizeof(rw_er_page_default));
3988 			memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_SAVED],
3989 			       &rw_er_page_default,
3990 			       sizeof(rw_er_page_default));
3991 			page_index->page_data =
3992 				(uint8_t *)lun->mode_pages.rw_er_page;
3993 			break;
3994 		}
3995 		case SMS_FORMAT_DEVICE_PAGE: {
3996 			struct scsi_format_page *format_page;
3997 
3998 			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3999 				panic("subpage is incorrect!");
4000 
4001 			/*
4002 			 * Sectors per track are set above.  Bytes per
4003 			 * sector need to be set here on a per-LUN basis.
4004 			 */
4005 			memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
4006 			       &format_page_default,
4007 			       sizeof(format_page_default));
4008 			memcpy(&lun->mode_pages.format_page[
4009 			       CTL_PAGE_CHANGEABLE], &format_page_changeable,
4010 			       sizeof(format_page_changeable));
4011 			memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
4012 			       &format_page_default,
4013 			       sizeof(format_page_default));
4014 			memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
4015 			       &format_page_default,
4016 			       sizeof(format_page_default));
4017 
4018 			format_page = &lun->mode_pages.format_page[
4019 				CTL_PAGE_CURRENT];
4020 			scsi_ulto2b(lun->be_lun->blocksize,
4021 				    format_page->bytes_per_sector);
4022 
4023 			format_page = &lun->mode_pages.format_page[
4024 				CTL_PAGE_DEFAULT];
4025 			scsi_ulto2b(lun->be_lun->blocksize,
4026 				    format_page->bytes_per_sector);
4027 
4028 			format_page = &lun->mode_pages.format_page[
4029 				CTL_PAGE_SAVED];
4030 			scsi_ulto2b(lun->be_lun->blocksize,
4031 				    format_page->bytes_per_sector);
4032 
4033 			page_index->page_data =
4034 				(uint8_t *)lun->mode_pages.format_page;
4035 			break;
4036 		}
4037 		case SMS_RIGID_DISK_PAGE: {
4038 			struct scsi_rigid_disk_page *rigid_disk_page;
4039 			uint32_t sectors_per_cylinder;
4040 			uint64_t cylinders;
4041 #ifndef	__XSCALE__
4042 			int shift;
4043 #endif /* !__XSCALE__ */
4044 
4045 			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4046 				panic("invalid subpage value %d",
4047 				      page_index->subpage);
4048 
4049 			/*
4050 			 * Rotation rate and sectors per track are set
4051 			 * above.  We calculate the cylinders here based on
4052 			 * capacity.  Due to the number of heads and
4053 			 * sectors per track we're using, smaller arrays
4054 			 * may turn out to have 0 cylinders.  Linux and
4055 			 * FreeBSD don't pay attention to these mode pages
4056 			 * to figure out capacity, but Solaris does.  It
4057 			 * seems to deal with 0 cylinders just fine, and
4058 			 * works out a fake geometry based on the capacity.
4059 			 */
4060 			memcpy(&lun->mode_pages.rigid_disk_page[
4061 			       CTL_PAGE_DEFAULT], &rigid_disk_page_default,
4062 			       sizeof(rigid_disk_page_default));
4063 			memcpy(&lun->mode_pages.rigid_disk_page[
4064 			       CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
4065 			       sizeof(rigid_disk_page_changeable));
4066 
4067 			sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
4068 				CTL_DEFAULT_HEADS;
4069 
4070 			/*
4071 			 * The divide method here will be more accurate,
4072 			 * probably, but results in floating point being
4073 			 * used in the kernel on i386 (__udivdi3()).  On the
4074 			 * XScale, though, __udivdi3() is implemented in
4075 			 * software.
4076 			 *
4077 			 * The shift method for cylinder calculation is
4078 			 * accurate if sectors_per_cylinder is a power of
4079 			 * 2.  Otherwise it might be slightly off -- you
4080 			 * might have a bit of a truncation problem.
4081 			 */
4082 #ifdef	__XSCALE__
4083 			cylinders = (lun->be_lun->maxlba + 1) /
4084 				sectors_per_cylinder;
4085 #else
4086 			for (shift = 31; shift > 0; shift--) {
4087 				if (sectors_per_cylinder & (1 << shift))
4088 					break;
4089 			}
4090 			cylinders = (lun->be_lun->maxlba + 1) >> shift;
4091 #endif
4092 
4093 			/*
4094 			 * We've basically got 3 bytes, or 24 bits for the
4095 			 * cylinder size in the mode page.  If we're over,
4096 			 * just round down to 2^24.
4097 			 */
4098 			if (cylinders > 0xffffff)
4099 				cylinders = 0xffffff;
4100 
4101 			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4102 				CTL_PAGE_DEFAULT];
4103 			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4104 
4105 			if ((value = ctl_get_opt(&lun->be_lun->options,
4106 			    "rpm")) != NULL) {
4107 				scsi_ulto2b(strtol(value, NULL, 0),
4108 				     rigid_disk_page->rotation_rate);
4109 			}
4110 
4111 			memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_CURRENT],
4112 			       &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
4113 			       sizeof(rigid_disk_page_default));
4114 			memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_SAVED],
4115 			       &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
4116 			       sizeof(rigid_disk_page_default));
4117 
4118 			page_index->page_data =
4119 				(uint8_t *)lun->mode_pages.rigid_disk_page;
4120 			break;
4121 		}
4122 		case SMS_CACHING_PAGE: {
4123 			struct scsi_caching_page *caching_page;
4124 
4125 			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4126 				panic("invalid subpage value %d",
4127 				      page_index->subpage);
4128 			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4129 			       &caching_page_default,
4130 			       sizeof(caching_page_default));
4131 			memcpy(&lun->mode_pages.caching_page[
4132 			       CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4133 			       sizeof(caching_page_changeable));
4134 			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4135 			       &caching_page_default,
4136 			       sizeof(caching_page_default));
4137 			caching_page = &lun->mode_pages.caching_page[
4138 			    CTL_PAGE_SAVED];
4139 			value = ctl_get_opt(&lun->be_lun->options, "writecache");
4140 			if (value != NULL && strcmp(value, "off") == 0)
4141 				caching_page->flags1 &= ~SCP_WCE;
4142 			value = ctl_get_opt(&lun->be_lun->options, "readcache");
4143 			if (value != NULL && strcmp(value, "off") == 0)
4144 				caching_page->flags1 |= SCP_RCD;
4145 			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4146 			       &lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4147 			       sizeof(caching_page_default));
4148 			page_index->page_data =
4149 				(uint8_t *)lun->mode_pages.caching_page;
4150 			break;
4151 		}
4152 		case SMS_CONTROL_MODE_PAGE: {
4153 			struct scsi_control_page *control_page;
4154 
4155 			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4156 				panic("invalid subpage value %d",
4157 				      page_index->subpage);
4158 
4159 			memcpy(&lun->mode_pages.control_page[CTL_PAGE_DEFAULT],
4160 			       &control_page_default,
4161 			       sizeof(control_page_default));
4162 			memcpy(&lun->mode_pages.control_page[
4163 			       CTL_PAGE_CHANGEABLE], &control_page_changeable,
4164 			       sizeof(control_page_changeable));
4165 			memcpy(&lun->mode_pages.control_page[CTL_PAGE_SAVED],
4166 			       &control_page_default,
4167 			       sizeof(control_page_default));
4168 			control_page = &lun->mode_pages.control_page[
4169 			    CTL_PAGE_SAVED];
4170 			value = ctl_get_opt(&lun->be_lun->options, "reordering");
4171 			if (value != NULL && strcmp(value, "unrestricted") == 0) {
4172 				control_page->queue_flags &= ~SCP_QUEUE_ALG_MASK;
4173 				control_page->queue_flags |= SCP_QUEUE_ALG_UNRESTRICTED;
4174 			}
4175 			memcpy(&lun->mode_pages.control_page[CTL_PAGE_CURRENT],
4176 			       &lun->mode_pages.control_page[CTL_PAGE_SAVED],
4177 			       sizeof(control_page_default));
4178 			page_index->page_data =
4179 				(uint8_t *)lun->mode_pages.control_page;
4180 			break;
4181 
4182 		}
4183 		case SMS_INFO_EXCEPTIONS_PAGE: {
4184 			switch (page_index->subpage) {
4185 			case SMS_SUBPAGE_PAGE_0:
4186 				memcpy(&lun->mode_pages.ie_page[CTL_PAGE_CURRENT],
4187 				       &ie_page_default,
4188 				       sizeof(ie_page_default));
4189 				memcpy(&lun->mode_pages.ie_page[
4190 				       CTL_PAGE_CHANGEABLE], &ie_page_changeable,
4191 				       sizeof(ie_page_changeable));
4192 				memcpy(&lun->mode_pages.ie_page[CTL_PAGE_DEFAULT],
4193 				       &ie_page_default,
4194 				       sizeof(ie_page_default));
4195 				memcpy(&lun->mode_pages.ie_page[CTL_PAGE_SAVED],
4196 				       &ie_page_default,
4197 				       sizeof(ie_page_default));
4198 				page_index->page_data =
4199 					(uint8_t *)lun->mode_pages.ie_page;
4200 				break;
4201 			case 0x02: {
4202 				struct ctl_logical_block_provisioning_page *page;
4203 
4204 				memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_DEFAULT],
4205 				       &lbp_page_default,
4206 				       sizeof(lbp_page_default));
4207 				memcpy(&lun->mode_pages.lbp_page[
4208 				       CTL_PAGE_CHANGEABLE], &lbp_page_changeable,
4209 				       sizeof(lbp_page_changeable));
4210 				memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4211 				       &lbp_page_default,
4212 				       sizeof(lbp_page_default));
4213 				page = &lun->mode_pages.lbp_page[CTL_PAGE_SAVED];
4214 				value = ctl_get_opt(&lun->be_lun->options,
4215 				    "avail-threshold");
4216 				if (value != NULL &&
4217 				    ctl_expand_number(value, &ival) == 0) {
4218 					page->descr[0].flags |= SLBPPD_ENABLED |
4219 					    SLBPPD_ARMING_DEC;
4220 					if (lun->be_lun->blocksize)
4221 						ival /= lun->be_lun->blocksize;
4222 					else
4223 						ival /= 512;
4224 					scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4225 					    page->descr[0].count);
4226 				}
4227 				value = ctl_get_opt(&lun->be_lun->options,
4228 				    "used-threshold");
4229 				if (value != NULL &&
4230 				    ctl_expand_number(value, &ival) == 0) {
4231 					page->descr[1].flags |= SLBPPD_ENABLED |
4232 					    SLBPPD_ARMING_INC;
4233 					if (lun->be_lun->blocksize)
4234 						ival /= lun->be_lun->blocksize;
4235 					else
4236 						ival /= 512;
4237 					scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4238 					    page->descr[1].count);
4239 				}
4240 				value = ctl_get_opt(&lun->be_lun->options,
4241 				    "pool-avail-threshold");
4242 				if (value != NULL &&
4243 				    ctl_expand_number(value, &ival) == 0) {
4244 					page->descr[2].flags |= SLBPPD_ENABLED |
4245 					    SLBPPD_ARMING_DEC;
4246 					if (lun->be_lun->blocksize)
4247 						ival /= lun->be_lun->blocksize;
4248 					else
4249 						ival /= 512;
4250 					scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4251 					    page->descr[2].count);
4252 				}
4253 				value = ctl_get_opt(&lun->be_lun->options,
4254 				    "pool-used-threshold");
4255 				if (value != NULL &&
4256 				    ctl_expand_number(value, &ival) == 0) {
4257 					page->descr[3].flags |= SLBPPD_ENABLED |
4258 					    SLBPPD_ARMING_INC;
4259 					if (lun->be_lun->blocksize)
4260 						ival /= lun->be_lun->blocksize;
4261 					else
4262 						ival /= 512;
4263 					scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4264 					    page->descr[3].count);
4265 				}
4266 				memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_CURRENT],
4267 				       &lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4268 				       sizeof(lbp_page_default));
4269 				page_index->page_data =
4270 					(uint8_t *)lun->mode_pages.lbp_page;
4271 			}}
4272 			break;
4273 		}
4274 		case SMS_VENDOR_SPECIFIC_PAGE:{
4275 			switch (page_index->subpage) {
4276 			case DBGCNF_SUBPAGE_CODE: {
4277 				struct copan_debugconf_subpage *current_page,
4278 							       *saved_page;
4279 
4280 				memcpy(&lun->mode_pages.debugconf_subpage[
4281 				       CTL_PAGE_CURRENT],
4282 				       &debugconf_page_default,
4283 				       sizeof(debugconf_page_default));
4284 				memcpy(&lun->mode_pages.debugconf_subpage[
4285 				       CTL_PAGE_CHANGEABLE],
4286 				       &debugconf_page_changeable,
4287 				       sizeof(debugconf_page_changeable));
4288 				memcpy(&lun->mode_pages.debugconf_subpage[
4289 				       CTL_PAGE_DEFAULT],
4290 				       &debugconf_page_default,
4291 				       sizeof(debugconf_page_default));
4292 				memcpy(&lun->mode_pages.debugconf_subpage[
4293 				       CTL_PAGE_SAVED],
4294 				       &debugconf_page_default,
4295 				       sizeof(debugconf_page_default));
4296 				page_index->page_data =
4297 					(uint8_t *)lun->mode_pages.debugconf_subpage;
4298 
4299 				current_page = (struct copan_debugconf_subpage *)
4300 					(page_index->page_data +
4301 					 (page_index->page_len *
4302 					  CTL_PAGE_CURRENT));
4303 				saved_page = (struct copan_debugconf_subpage *)
4304 					(page_index->page_data +
4305 					 (page_index->page_len *
4306 					  CTL_PAGE_SAVED));
4307 				break;
4308 			}
4309 			default:
4310 				panic("invalid subpage value %d",
4311 				      page_index->subpage);
4312 				break;
4313 			}
4314    			break;
4315 		}
4316 		default:
4317 			panic("invalid page value %d",
4318 			      page_index->page_code & SMPH_PC_MASK);
4319 			break;
4320     	}
4321 	}
4322 
4323 	return (CTL_RETVAL_COMPLETE);
4324 }
4325 
4326 static int
4327 ctl_init_log_page_index(struct ctl_lun *lun)
4328 {
4329 	struct ctl_page_index *page_index;
4330 	int i, j, k, prev;
4331 
4332 	memcpy(&lun->log_pages.index, log_page_index_template,
4333 	       sizeof(log_page_index_template));
4334 
4335 	prev = -1;
4336 	for (i = 0, j = 0, k = 0; i < CTL_NUM_LOG_PAGES; i++) {
4337 
4338 		page_index = &lun->log_pages.index[i];
4339 		/*
4340 		 * If this is a disk-only mode page, there's no point in
4341 		 * setting it up.  For some pages, we have to have some
4342 		 * basic information about the disk in order to calculate the
4343 		 * mode page data.
4344 		 */
4345 		if ((lun->be_lun->lun_type != T_DIRECT)
4346 		 && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
4347 			continue;
4348 
4349 		if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING &&
4350 		     lun->backend->lun_attr == NULL)
4351 			continue;
4352 
4353 		if (page_index->page_code != prev) {
4354 			lun->log_pages.pages_page[j] = page_index->page_code;
4355 			prev = page_index->page_code;
4356 			j++;
4357 		}
4358 		lun->log_pages.subpages_page[k*2] = page_index->page_code;
4359 		lun->log_pages.subpages_page[k*2+1] = page_index->subpage;
4360 		k++;
4361 	}
4362 	lun->log_pages.index[0].page_data = &lun->log_pages.pages_page[0];
4363 	lun->log_pages.index[0].page_len = j;
4364 	lun->log_pages.index[1].page_data = &lun->log_pages.subpages_page[0];
4365 	lun->log_pages.index[1].page_len = k * 2;
4366 	lun->log_pages.index[2].page_data = &lun->log_pages.lbp_page[0];
4367 	lun->log_pages.index[2].page_len = 12*CTL_NUM_LBP_PARAMS;
4368 
4369 	return (CTL_RETVAL_COMPLETE);
4370 }
4371 
4372 static int
4373 hex2bin(const char *str, uint8_t *buf, int buf_size)
4374 {
4375 	int i;
4376 	u_char c;
4377 
4378 	memset(buf, 0, buf_size);
4379 	while (isspace(str[0]))
4380 		str++;
4381 	if (str[0] == '0' && (str[1] == 'x' || str[1] == 'X'))
4382 		str += 2;
4383 	buf_size *= 2;
4384 	for (i = 0; str[i] != 0 && i < buf_size; i++) {
4385 		c = str[i];
4386 		if (isdigit(c))
4387 			c -= '0';
4388 		else if (isalpha(c))
4389 			c -= isupper(c) ? 'A' - 10 : 'a' - 10;
4390 		else
4391 			break;
4392 		if (c >= 16)
4393 			break;
4394 		if ((i & 1) == 0)
4395 			buf[i / 2] |= (c << 4);
4396 		else
4397 			buf[i / 2] |= c;
4398 	}
4399 	return ((i + 1) / 2);
4400 }
4401 
4402 /*
4403  * LUN allocation.
4404  *
4405  * Requirements:
4406  * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4407  *   wants us to allocate the LUN and he can block.
4408  * - ctl_softc is always set
4409  * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4410  *
4411  * Returns 0 for success, non-zero (errno) for failure.
4412  */
4413 static int
4414 ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4415 	      struct ctl_be_lun *const be_lun, struct ctl_id target_id)
4416 {
4417 	struct ctl_lun *nlun, *lun;
4418 	struct ctl_port *port;
4419 	struct scsi_vpd_id_descriptor *desc;
4420 	struct scsi_vpd_id_t10 *t10id;
4421 	const char *eui, *naa, *scsiname, *vendor, *value;
4422 	int lun_number, i, lun_malloced;
4423 	int devidlen, idlen1, idlen2 = 0, len;
4424 
4425 	if (be_lun == NULL)
4426 		return (EINVAL);
4427 
4428 	/*
4429 	 * We currently only support Direct Access or Processor LUN types.
4430 	 */
4431 	switch (be_lun->lun_type) {
4432 	case T_DIRECT:
4433 		break;
4434 	case T_PROCESSOR:
4435 		break;
4436 	case T_SEQUENTIAL:
4437 	case T_CHANGER:
4438 	default:
4439 		be_lun->lun_config_status(be_lun->be_lun,
4440 					  CTL_LUN_CONFIG_FAILURE);
4441 		break;
4442 	}
4443 	if (ctl_lun == NULL) {
4444 		lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4445 		lun_malloced = 1;
4446 	} else {
4447 		lun_malloced = 0;
4448 		lun = ctl_lun;
4449 	}
4450 
4451 	memset(lun, 0, sizeof(*lun));
4452 	if (lun_malloced)
4453 		lun->flags = CTL_LUN_MALLOCED;
4454 
4455 	/* Generate LUN ID. */
4456 	devidlen = max(CTL_DEVID_MIN_LEN,
4457 	    strnlen(be_lun->device_id, CTL_DEVID_LEN));
4458 	idlen1 = sizeof(*t10id) + devidlen;
4459 	len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4460 	scsiname = ctl_get_opt(&be_lun->options, "scsiname");
4461 	if (scsiname != NULL) {
4462 		idlen2 = roundup2(strlen(scsiname) + 1, 4);
4463 		len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4464 	}
4465 	eui = ctl_get_opt(&be_lun->options, "eui");
4466 	if (eui != NULL) {
4467 		len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4468 	}
4469 	naa = ctl_get_opt(&be_lun->options, "naa");
4470 	if (naa != NULL) {
4471 		len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4472 	}
4473 	lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4474 	    M_CTL, M_WAITOK | M_ZERO);
4475 	desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4476 	desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4477 	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4478 	desc->length = idlen1;
4479 	t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4480 	memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4481 	if ((vendor = ctl_get_opt(&be_lun->options, "vendor")) == NULL) {
4482 		strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4483 	} else {
4484 		strncpy(t10id->vendor, vendor,
4485 		    min(sizeof(t10id->vendor), strlen(vendor)));
4486 	}
4487 	strncpy((char *)t10id->vendor_spec_id,
4488 	    (char *)be_lun->device_id, devidlen);
4489 	if (scsiname != NULL) {
4490 		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4491 		    desc->length);
4492 		desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4493 		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4494 		    SVPD_ID_TYPE_SCSI_NAME;
4495 		desc->length = idlen2;
4496 		strlcpy(desc->identifier, scsiname, idlen2);
4497 	}
4498 	if (eui != NULL) {
4499 		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4500 		    desc->length);
4501 		desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4502 		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4503 		    SVPD_ID_TYPE_EUI64;
4504 		desc->length = hex2bin(eui, desc->identifier, 16);
4505 		desc->length = desc->length > 12 ? 16 :
4506 		    (desc->length > 8 ? 12 : 8);
4507 		len -= 16 - desc->length;
4508 	}
4509 	if (naa != NULL) {
4510 		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4511 		    desc->length);
4512 		desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4513 		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4514 		    SVPD_ID_TYPE_NAA;
4515 		desc->length = hex2bin(naa, desc->identifier, 16);
4516 		desc->length = desc->length > 8 ? 16 : 8;
4517 		len -= 16 - desc->length;
4518 	}
4519 	lun->lun_devid->len = len;
4520 
4521 	mtx_lock(&ctl_softc->ctl_lock);
4522 	/*
4523 	 * See if the caller requested a particular LUN number.  If so, see
4524 	 * if it is available.  Otherwise, allocate the first available LUN.
4525 	 */
4526 	if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4527 		if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4528 		 || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4529 			mtx_unlock(&ctl_softc->ctl_lock);
4530 			if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4531 				printf("ctl: requested LUN ID %d is higher "
4532 				       "than CTL_MAX_LUNS - 1 (%d)\n",
4533 				       be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4534 			} else {
4535 				/*
4536 				 * XXX KDM return an error, or just assign
4537 				 * another LUN ID in this case??
4538 				 */
4539 				printf("ctl: requested LUN ID %d is already "
4540 				       "in use\n", be_lun->req_lun_id);
4541 			}
4542 			if (lun->flags & CTL_LUN_MALLOCED)
4543 				free(lun, M_CTL);
4544 			be_lun->lun_config_status(be_lun->be_lun,
4545 						  CTL_LUN_CONFIG_FAILURE);
4546 			return (ENOSPC);
4547 		}
4548 		lun_number = be_lun->req_lun_id;
4549 	} else {
4550 		lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, CTL_MAX_LUNS);
4551 		if (lun_number == -1) {
4552 			mtx_unlock(&ctl_softc->ctl_lock);
4553 			printf("ctl: can't allocate LUN on target %ju, out of "
4554 			       "LUNs\n", (uintmax_t)target_id.id);
4555 			if (lun->flags & CTL_LUN_MALLOCED)
4556 				free(lun, M_CTL);
4557 			be_lun->lun_config_status(be_lun->be_lun,
4558 						  CTL_LUN_CONFIG_FAILURE);
4559 			return (ENOSPC);
4560 		}
4561 	}
4562 	ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4563 
4564 	mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4565 	lun->target = target_id;
4566 	lun->lun = lun_number;
4567 	lun->be_lun = be_lun;
4568 	/*
4569 	 * The processor LUN is always enabled.  Disk LUNs come on line
4570 	 * disabled, and must be enabled by the backend.
4571 	 */
4572 	lun->flags |= CTL_LUN_DISABLED;
4573 	lun->backend = be_lun->be;
4574 	be_lun->ctl_lun = lun;
4575 	be_lun->lun_id = lun_number;
4576 	atomic_add_int(&be_lun->be->num_luns, 1);
4577 	if (be_lun->flags & CTL_LUN_FLAG_OFFLINE)
4578 		lun->flags |= CTL_LUN_OFFLINE;
4579 
4580 	if (be_lun->flags & CTL_LUN_FLAG_POWERED_OFF)
4581 		lun->flags |= CTL_LUN_STOPPED;
4582 
4583 	if (be_lun->flags & CTL_LUN_FLAG_INOPERABLE)
4584 		lun->flags |= CTL_LUN_INOPERABLE;
4585 
4586 	if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4587 		lun->flags |= CTL_LUN_PRIMARY_SC;
4588 
4589 	value = ctl_get_opt(&be_lun->options, "readonly");
4590 	if (value != NULL && strcmp(value, "on") == 0)
4591 		lun->flags |= CTL_LUN_READONLY;
4592 
4593 	lun->ctl_softc = ctl_softc;
4594 	TAILQ_INIT(&lun->ooa_queue);
4595 	TAILQ_INIT(&lun->blocked_queue);
4596 	STAILQ_INIT(&lun->error_list);
4597 	ctl_tpc_lun_init(lun);
4598 
4599 	/*
4600 	 * Initialize the mode and log page index.
4601 	 */
4602 	ctl_init_page_index(lun);
4603 	ctl_init_log_page_index(lun);
4604 
4605 	/*
4606 	 * Now, before we insert this lun on the lun list, set the lun
4607 	 * inventory changed UA for all other luns.
4608 	 */
4609 	STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4610 		mtx_lock(&nlun->lun_lock);
4611 		ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4612 		mtx_unlock(&nlun->lun_lock);
4613 	}
4614 
4615 	STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4616 
4617 	ctl_softc->ctl_luns[lun_number] = lun;
4618 
4619 	ctl_softc->num_luns++;
4620 
4621 	/* Setup statistics gathering */
4622 	lun->stats.device_type = be_lun->lun_type;
4623 	lun->stats.lun_number = lun_number;
4624 	if (lun->stats.device_type == T_DIRECT)
4625 		lun->stats.blocksize = be_lun->blocksize;
4626 	else
4627 		lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4628 	for (i = 0;i < CTL_MAX_PORTS;i++)
4629 		lun->stats.ports[i].targ_port = i;
4630 
4631 	mtx_unlock(&ctl_softc->ctl_lock);
4632 
4633 	lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4634 
4635 	/*
4636 	 * Run through each registered FETD and bring it online if it isn't
4637 	 * already.  Enable the target ID if it hasn't been enabled, and
4638 	 * enable this particular LUN.
4639 	 */
4640 	STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4641 		int retval;
4642 
4643 		retval = port->lun_enable(port->targ_lun_arg, target_id,lun_number);
4644 		if (retval != 0) {
4645 			printf("ctl_alloc_lun: FETD %s port %d returned error "
4646 			       "%d for lun_enable on target %ju lun %d\n",
4647 			       port->port_name, port->targ_port, retval,
4648 			       (uintmax_t)target_id.id, lun_number);
4649 		} else
4650 			port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4651 	}
4652 	return (0);
4653 }
4654 
4655 /*
4656  * Delete a LUN.
4657  * Assumptions:
4658  * - LUN has already been marked invalid and any pending I/O has been taken
4659  *   care of.
4660  */
4661 static int
4662 ctl_free_lun(struct ctl_lun *lun)
4663 {
4664 	struct ctl_softc *softc;
4665 #if 0
4666 	struct ctl_port *port;
4667 #endif
4668 	struct ctl_lun *nlun;
4669 	int i;
4670 
4671 	softc = lun->ctl_softc;
4672 
4673 	mtx_assert(&softc->ctl_lock, MA_OWNED);
4674 
4675 	STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4676 
4677 	ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4678 
4679 	softc->ctl_luns[lun->lun] = NULL;
4680 
4681 	if (!TAILQ_EMPTY(&lun->ooa_queue))
4682 		panic("Freeing a LUN %p with outstanding I/O!!\n", lun);
4683 
4684 	softc->num_luns--;
4685 
4686 	/*
4687 	 * XXX KDM this scheme only works for a single target/multiple LUN
4688 	 * setup.  It needs to be revamped for a multiple target scheme.
4689 	 *
4690 	 * XXX KDM this results in port->lun_disable() getting called twice,
4691 	 * once when ctl_disable_lun() is called, and a second time here.
4692 	 * We really need to re-think the LUN disable semantics.  There
4693 	 * should probably be several steps/levels to LUN removal:
4694 	 *  - disable
4695 	 *  - invalidate
4696 	 *  - free
4697  	 *
4698 	 * Right now we only have a disable method when communicating to
4699 	 * the front end ports, at least for individual LUNs.
4700 	 */
4701 #if 0
4702 	STAILQ_FOREACH(port, &softc->port_list, links) {
4703 		int retval;
4704 
4705 		retval = port->lun_disable(port->targ_lun_arg, lun->target,
4706 					 lun->lun);
4707 		if (retval != 0) {
4708 			printf("ctl_free_lun: FETD %s port %d returned error "
4709 			       "%d for lun_disable on target %ju lun %jd\n",
4710 			       port->port_name, port->targ_port, retval,
4711 			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4712 		}
4713 
4714 		if (STAILQ_FIRST(&softc->lun_list) == NULL) {
4715 			port->status &= ~CTL_PORT_STATUS_LUN_ONLINE;
4716 
4717 			retval = port->targ_disable(port->targ_lun_arg,lun->target);
4718 			if (retval != 0) {
4719 				printf("ctl_free_lun: FETD %s port %d "
4720 				       "returned error %d for targ_disable on "
4721 				       "target %ju\n", port->port_name,
4722 				       port->targ_port, retval,
4723 				       (uintmax_t)lun->target.id);
4724 			} else
4725 				port->status &= ~CTL_PORT_STATUS_TARG_ONLINE;
4726 
4727 			if ((port->status & CTL_PORT_STATUS_TARG_ONLINE) != 0)
4728 				continue;
4729 
4730 #if 0
4731 			port->port_offline(port->onoff_arg);
4732 			port->status &= ~CTL_PORT_STATUS_ONLINE;
4733 #endif
4734 		}
4735 	}
4736 #endif
4737 
4738 	/*
4739 	 * Tell the backend to free resources, if this LUN has a backend.
4740 	 */
4741 	atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4742 	lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4743 
4744 	ctl_tpc_lun_shutdown(lun);
4745 	mtx_destroy(&lun->lun_lock);
4746 	free(lun->lun_devid, M_CTL);
4747 	for (i = 0; i < CTL_MAX_PORTS; i++)
4748 		free(lun->pending_ua[i], M_CTL);
4749 	for (i = 0; i < 2 * CTL_MAX_PORTS; i++)
4750 		free(lun->pr_keys[i], M_CTL);
4751 	free(lun->write_buffer, M_CTL);
4752 	if (lun->flags & CTL_LUN_MALLOCED)
4753 		free(lun, M_CTL);
4754 
4755 	STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4756 		mtx_lock(&nlun->lun_lock);
4757 		ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4758 		mtx_unlock(&nlun->lun_lock);
4759 	}
4760 
4761 	return (0);
4762 }
4763 
4764 static void
4765 ctl_create_lun(struct ctl_be_lun *be_lun)
4766 {
4767 	struct ctl_softc *ctl_softc;
4768 
4769 	ctl_softc = control_softc;
4770 
4771 	/*
4772 	 * ctl_alloc_lun() should handle all potential failure cases.
4773 	 */
4774 	ctl_alloc_lun(ctl_softc, NULL, be_lun, ctl_softc->target);
4775 }
4776 
4777 int
4778 ctl_add_lun(struct ctl_be_lun *be_lun)
4779 {
4780 	struct ctl_softc *ctl_softc = control_softc;
4781 
4782 	mtx_lock(&ctl_softc->ctl_lock);
4783 	STAILQ_INSERT_TAIL(&ctl_softc->pending_lun_queue, be_lun, links);
4784 	mtx_unlock(&ctl_softc->ctl_lock);
4785 	wakeup(&ctl_softc->pending_lun_queue);
4786 
4787 	return (0);
4788 }
4789 
4790 int
4791 ctl_enable_lun(struct ctl_be_lun *be_lun)
4792 {
4793 	struct ctl_softc *ctl_softc;
4794 	struct ctl_port *port, *nport;
4795 	struct ctl_lun *lun;
4796 	int retval;
4797 
4798 	ctl_softc = control_softc;
4799 
4800 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4801 
4802 	mtx_lock(&ctl_softc->ctl_lock);
4803 	mtx_lock(&lun->lun_lock);
4804 	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4805 		/*
4806 		 * eh?  Why did we get called if the LUN is already
4807 		 * enabled?
4808 		 */
4809 		mtx_unlock(&lun->lun_lock);
4810 		mtx_unlock(&ctl_softc->ctl_lock);
4811 		return (0);
4812 	}
4813 	lun->flags &= ~CTL_LUN_DISABLED;
4814 	mtx_unlock(&lun->lun_lock);
4815 
4816 	for (port = STAILQ_FIRST(&ctl_softc->port_list); port != NULL; port = nport) {
4817 		nport = STAILQ_NEXT(port, links);
4818 
4819 		/*
4820 		 * Drop the lock while we call the FETD's enable routine.
4821 		 * This can lead to a callback into CTL (at least in the
4822 		 * case of the internal initiator frontend.
4823 		 */
4824 		mtx_unlock(&ctl_softc->ctl_lock);
4825 		retval = port->lun_enable(port->targ_lun_arg, lun->target,lun->lun);
4826 		mtx_lock(&ctl_softc->ctl_lock);
4827 		if (retval != 0) {
4828 			printf("%s: FETD %s port %d returned error "
4829 			       "%d for lun_enable on target %ju lun %jd\n",
4830 			       __func__, port->port_name, port->targ_port, retval,
4831 			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4832 		}
4833 #if 0
4834 		 else {
4835             /* NOTE:  TODO:  why does lun enable affect port status? */
4836 			port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4837 		}
4838 #endif
4839 	}
4840 
4841 	mtx_unlock(&ctl_softc->ctl_lock);
4842 
4843 	return (0);
4844 }
4845 
4846 int
4847 ctl_disable_lun(struct ctl_be_lun *be_lun)
4848 {
4849 	struct ctl_softc *ctl_softc;
4850 	struct ctl_port *port;
4851 	struct ctl_lun *lun;
4852 	int retval;
4853 
4854 	ctl_softc = control_softc;
4855 
4856 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4857 
4858 	mtx_lock(&ctl_softc->ctl_lock);
4859 	mtx_lock(&lun->lun_lock);
4860 	if (lun->flags & CTL_LUN_DISABLED) {
4861 		mtx_unlock(&lun->lun_lock);
4862 		mtx_unlock(&ctl_softc->ctl_lock);
4863 		return (0);
4864 	}
4865 	lun->flags |= CTL_LUN_DISABLED;
4866 	mtx_unlock(&lun->lun_lock);
4867 
4868 	STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4869 		mtx_unlock(&ctl_softc->ctl_lock);
4870 		/*
4871 		 * Drop the lock before we call the frontend's disable
4872 		 * routine, to avoid lock order reversals.
4873 		 *
4874 		 * XXX KDM what happens if the frontend list changes while
4875 		 * we're traversing it?  It's unlikely, but should be handled.
4876 		 */
4877 		retval = port->lun_disable(port->targ_lun_arg, lun->target,
4878 					 lun->lun);
4879 		mtx_lock(&ctl_softc->ctl_lock);
4880 		if (retval != 0) {
4881 			printf("ctl_alloc_lun: FETD %s port %d returned error "
4882 			       "%d for lun_disable on target %ju lun %jd\n",
4883 			       port->port_name, port->targ_port, retval,
4884 			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4885 		}
4886 	}
4887 
4888 	mtx_unlock(&ctl_softc->ctl_lock);
4889 
4890 	return (0);
4891 }
4892 
4893 int
4894 ctl_start_lun(struct ctl_be_lun *be_lun)
4895 {
4896 	struct ctl_softc *ctl_softc;
4897 	struct ctl_lun *lun;
4898 
4899 	ctl_softc = control_softc;
4900 
4901 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4902 
4903 	mtx_lock(&lun->lun_lock);
4904 	lun->flags &= ~CTL_LUN_STOPPED;
4905 	mtx_unlock(&lun->lun_lock);
4906 
4907 	return (0);
4908 }
4909 
4910 int
4911 ctl_stop_lun(struct ctl_be_lun *be_lun)
4912 {
4913 	struct ctl_softc *ctl_softc;
4914 	struct ctl_lun *lun;
4915 
4916 	ctl_softc = control_softc;
4917 
4918 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4919 
4920 	mtx_lock(&lun->lun_lock);
4921 	lun->flags |= CTL_LUN_STOPPED;
4922 	mtx_unlock(&lun->lun_lock);
4923 
4924 	return (0);
4925 }
4926 
4927 int
4928 ctl_lun_offline(struct ctl_be_lun *be_lun)
4929 {
4930 	struct ctl_softc *ctl_softc;
4931 	struct ctl_lun *lun;
4932 
4933 	ctl_softc = control_softc;
4934 
4935 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4936 
4937 	mtx_lock(&lun->lun_lock);
4938 	lun->flags |= CTL_LUN_OFFLINE;
4939 	mtx_unlock(&lun->lun_lock);
4940 
4941 	return (0);
4942 }
4943 
4944 int
4945 ctl_lun_online(struct ctl_be_lun *be_lun)
4946 {
4947 	struct ctl_softc *ctl_softc;
4948 	struct ctl_lun *lun;
4949 
4950 	ctl_softc = control_softc;
4951 
4952 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4953 
4954 	mtx_lock(&lun->lun_lock);
4955 	lun->flags &= ~CTL_LUN_OFFLINE;
4956 	mtx_unlock(&lun->lun_lock);
4957 
4958 	return (0);
4959 }
4960 
4961 int
4962 ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4963 {
4964 	struct ctl_softc *ctl_softc;
4965 	struct ctl_lun *lun;
4966 
4967 	ctl_softc = control_softc;
4968 
4969 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4970 
4971 	mtx_lock(&lun->lun_lock);
4972 
4973 	/*
4974 	 * The LUN needs to be disabled before it can be marked invalid.
4975 	 */
4976 	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4977 		mtx_unlock(&lun->lun_lock);
4978 		return (-1);
4979 	}
4980 	/*
4981 	 * Mark the LUN invalid.
4982 	 */
4983 	lun->flags |= CTL_LUN_INVALID;
4984 
4985 	/*
4986 	 * If there is nothing in the OOA queue, go ahead and free the LUN.
4987 	 * If we have something in the OOA queue, we'll free it when the
4988 	 * last I/O completes.
4989 	 */
4990 	if (TAILQ_EMPTY(&lun->ooa_queue)) {
4991 		mtx_unlock(&lun->lun_lock);
4992 		mtx_lock(&ctl_softc->ctl_lock);
4993 		ctl_free_lun(lun);
4994 		mtx_unlock(&ctl_softc->ctl_lock);
4995 	} else
4996 		mtx_unlock(&lun->lun_lock);
4997 
4998 	return (0);
4999 }
5000 
5001 int
5002 ctl_lun_inoperable(struct ctl_be_lun *be_lun)
5003 {
5004 	struct ctl_softc *ctl_softc;
5005 	struct ctl_lun *lun;
5006 
5007 	ctl_softc = control_softc;
5008 	lun = (struct ctl_lun *)be_lun->ctl_lun;
5009 
5010 	mtx_lock(&lun->lun_lock);
5011 	lun->flags |= CTL_LUN_INOPERABLE;
5012 	mtx_unlock(&lun->lun_lock);
5013 
5014 	return (0);
5015 }
5016 
5017 int
5018 ctl_lun_operable(struct ctl_be_lun *be_lun)
5019 {
5020 	struct ctl_softc *ctl_softc;
5021 	struct ctl_lun *lun;
5022 
5023 	ctl_softc = control_softc;
5024 	lun = (struct ctl_lun *)be_lun->ctl_lun;
5025 
5026 	mtx_lock(&lun->lun_lock);
5027 	lun->flags &= ~CTL_LUN_INOPERABLE;
5028 	mtx_unlock(&lun->lun_lock);
5029 
5030 	return (0);
5031 }
5032 
5033 void
5034 ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
5035 {
5036 	struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
5037 
5038 	mtx_lock(&lun->lun_lock);
5039 	ctl_est_ua_all(lun, -1, CTL_UA_CAPACITY_CHANGED);
5040 	mtx_unlock(&lun->lun_lock);
5041 }
5042 
5043 /*
5044  * Backend "memory move is complete" callback for requests that never
5045  * make it down to say RAIDCore's configuration code.
5046  */
5047 int
5048 ctl_config_move_done(union ctl_io *io)
5049 {
5050 	int retval;
5051 
5052 	CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
5053 	KASSERT(io->io_hdr.io_type == CTL_IO_SCSI,
5054 	    ("Config I/O type isn't CTL_IO_SCSI (%d)!", io->io_hdr.io_type));
5055 
5056 	if ((io->io_hdr.port_status != 0) &&
5057 	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5058 	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5059 		/*
5060 		 * For hardware error sense keys, the sense key
5061 		 * specific value is defined to be a retry count,
5062 		 * but we use it to pass back an internal FETD
5063 		 * error code.  XXX KDM  Hopefully the FETD is only
5064 		 * using 16 bits for an error code, since that's
5065 		 * all the space we have in the sks field.
5066 		 */
5067 		ctl_set_internal_failure(&io->scsiio,
5068 					 /*sks_valid*/ 1,
5069 					 /*retry_count*/
5070 					 io->io_hdr.port_status);
5071 	}
5072 
5073 	if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) ||
5074 	    ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
5075 	     (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) ||
5076 	    ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
5077 		/*
5078 		 * XXX KDM just assuming a single pointer here, and not a
5079 		 * S/G list.  If we start using S/G lists for config data,
5080 		 * we'll need to know how to clean them up here as well.
5081 		 */
5082 		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5083 			free(io->scsiio.kern_data_ptr, M_CTL);
5084 		ctl_done(io);
5085 		retval = CTL_RETVAL_COMPLETE;
5086 	} else {
5087 		/*
5088 		 * XXX KDM now we need to continue data movement.  Some
5089 		 * options:
5090 		 * - call ctl_scsiio() again?  We don't do this for data
5091 		 *   writes, because for those at least we know ahead of
5092 		 *   time where the write will go and how long it is.  For
5093 		 *   config writes, though, that information is largely
5094 		 *   contained within the write itself, thus we need to
5095 		 *   parse out the data again.
5096 		 *
5097 		 * - Call some other function once the data is in?
5098 		 */
5099 		if (ctl_debug & CTL_DEBUG_CDB_DATA)
5100 			ctl_data_print(io);
5101 
5102 		/*
5103 		 * XXX KDM call ctl_scsiio() again for now, and check flag
5104 		 * bits to see whether we're allocated or not.
5105 		 */
5106 		retval = ctl_scsiio(&io->scsiio);
5107 	}
5108 	return (retval);
5109 }
5110 
5111 /*
5112  * This gets called by a backend driver when it is done with a
5113  * data_submit method.
5114  */
5115 void
5116 ctl_data_submit_done(union ctl_io *io)
5117 {
5118 	/*
5119 	 * If the IO_CONT flag is set, we need to call the supplied
5120 	 * function to continue processing the I/O, instead of completing
5121 	 * the I/O just yet.
5122 	 *
5123 	 * If there is an error, though, we don't want to keep processing.
5124 	 * Instead, just send status back to the initiator.
5125 	 */
5126 	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5127 	    (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5128 	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5129 	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5130 		io->scsiio.io_cont(io);
5131 		return;
5132 	}
5133 	ctl_done(io);
5134 }
5135 
5136 /*
5137  * This gets called by a backend driver when it is done with a
5138  * configuration write.
5139  */
5140 void
5141 ctl_config_write_done(union ctl_io *io)
5142 {
5143 	uint8_t *buf;
5144 
5145 	/*
5146 	 * If the IO_CONT flag is set, we need to call the supplied
5147 	 * function to continue processing the I/O, instead of completing
5148 	 * the I/O just yet.
5149 	 *
5150 	 * If there is an error, though, we don't want to keep processing.
5151 	 * Instead, just send status back to the initiator.
5152 	 */
5153 	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5154 	    (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5155 	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5156 	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5157 		io->scsiio.io_cont(io);
5158 		return;
5159 	}
5160 	/*
5161 	 * Since a configuration write can be done for commands that actually
5162 	 * have data allocated, like write buffer, and commands that have
5163 	 * no data, like start/stop unit, we need to check here.
5164 	 */
5165 	if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5166 		buf = io->scsiio.kern_data_ptr;
5167 	else
5168 		buf = NULL;
5169 	ctl_done(io);
5170 	if (buf)
5171 		free(buf, M_CTL);
5172 }
5173 
5174 void
5175 ctl_config_read_done(union ctl_io *io)
5176 {
5177 	uint8_t *buf;
5178 
5179 	/*
5180 	 * If there is some error -- we are done, skip data transfer.
5181 	 */
5182 	if ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0 ||
5183 	    ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
5184 	     (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
5185 		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5186 			buf = io->scsiio.kern_data_ptr;
5187 		else
5188 			buf = NULL;
5189 		ctl_done(io);
5190 		if (buf)
5191 			free(buf, M_CTL);
5192 		return;
5193 	}
5194 
5195 	/*
5196 	 * If the IO_CONT flag is set, we need to call the supplied
5197 	 * function to continue processing the I/O, instead of completing
5198 	 * the I/O just yet.
5199 	 */
5200 	if (io->io_hdr.flags & CTL_FLAG_IO_CONT) {
5201 		io->scsiio.io_cont(io);
5202 		return;
5203 	}
5204 
5205 	ctl_datamove(io);
5206 }
5207 
5208 /*
5209  * SCSI release command.
5210  */
5211 int
5212 ctl_scsi_release(struct ctl_scsiio *ctsio)
5213 {
5214 	int length, longid, thirdparty_id, resv_id;
5215 	struct ctl_softc *ctl_softc;
5216 	struct ctl_lun *lun;
5217 	uint32_t residx;
5218 
5219 	length = 0;
5220 	resv_id = 0;
5221 
5222 	CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5223 
5224 	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5225 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5226 	ctl_softc = control_softc;
5227 
5228 	switch (ctsio->cdb[0]) {
5229 	case RELEASE_10: {
5230 		struct scsi_release_10 *cdb;
5231 
5232 		cdb = (struct scsi_release_10 *)ctsio->cdb;
5233 
5234 		if (cdb->byte2 & SR10_LONGID)
5235 			longid = 1;
5236 		else
5237 			thirdparty_id = cdb->thirdparty_id;
5238 
5239 		resv_id = cdb->resv_id;
5240 		length = scsi_2btoul(cdb->length);
5241 		break;
5242 	}
5243 	}
5244 
5245 
5246 	/*
5247 	 * XXX KDM right now, we only support LUN reservation.  We don't
5248 	 * support 3rd party reservations, or extent reservations, which
5249 	 * might actually need the parameter list.  If we've gotten this
5250 	 * far, we've got a LUN reservation.  Anything else got kicked out
5251 	 * above.  So, according to SPC, ignore the length.
5252 	 */
5253 	length = 0;
5254 
5255 	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5256 	 && (length > 0)) {
5257 		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5258 		ctsio->kern_data_len = length;
5259 		ctsio->kern_total_len = length;
5260 		ctsio->kern_data_resid = 0;
5261 		ctsio->kern_rel_offset = 0;
5262 		ctsio->kern_sg_entries = 0;
5263 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5264 		ctsio->be_move_done = ctl_config_move_done;
5265 		ctl_datamove((union ctl_io *)ctsio);
5266 
5267 		return (CTL_RETVAL_COMPLETE);
5268 	}
5269 
5270 	if (length > 0)
5271 		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5272 
5273 	mtx_lock(&lun->lun_lock);
5274 
5275 	/*
5276 	 * According to SPC, it is not an error for an intiator to attempt
5277 	 * to release a reservation on a LUN that isn't reserved, or that
5278 	 * is reserved by another initiator.  The reservation can only be
5279 	 * released, though, by the initiator who made it or by one of
5280 	 * several reset type events.
5281 	 */
5282 	if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
5283 			lun->flags &= ~CTL_LUN_RESERVED;
5284 
5285 	mtx_unlock(&lun->lun_lock);
5286 
5287 	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5288 		free(ctsio->kern_data_ptr, M_CTL);
5289 		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5290 	}
5291 
5292 	ctl_set_success(ctsio);
5293 	ctl_done((union ctl_io *)ctsio);
5294 	return (CTL_RETVAL_COMPLETE);
5295 }
5296 
5297 int
5298 ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5299 {
5300 	int extent, thirdparty, longid;
5301 	int resv_id, length;
5302 	uint64_t thirdparty_id;
5303 	struct ctl_softc *ctl_softc;
5304 	struct ctl_lun *lun;
5305 	uint32_t residx;
5306 
5307 	extent = 0;
5308 	thirdparty = 0;
5309 	longid = 0;
5310 	resv_id = 0;
5311 	length = 0;
5312 	thirdparty_id = 0;
5313 
5314 	CTL_DEBUG_PRINT(("ctl_reserve\n"));
5315 
5316 	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5317 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5318 	ctl_softc = control_softc;
5319 
5320 	switch (ctsio->cdb[0]) {
5321 	case RESERVE_10: {
5322 		struct scsi_reserve_10 *cdb;
5323 
5324 		cdb = (struct scsi_reserve_10 *)ctsio->cdb;
5325 
5326 		if (cdb->byte2 & SR10_LONGID)
5327 			longid = 1;
5328 		else
5329 			thirdparty_id = cdb->thirdparty_id;
5330 
5331 		resv_id = cdb->resv_id;
5332 		length = scsi_2btoul(cdb->length);
5333 		break;
5334 	}
5335 	}
5336 
5337 	/*
5338 	 * XXX KDM right now, we only support LUN reservation.  We don't
5339 	 * support 3rd party reservations, or extent reservations, which
5340 	 * might actually need the parameter list.  If we've gotten this
5341 	 * far, we've got a LUN reservation.  Anything else got kicked out
5342 	 * above.  So, according to SPC, ignore the length.
5343 	 */
5344 	length = 0;
5345 
5346 	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5347 	 && (length > 0)) {
5348 		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5349 		ctsio->kern_data_len = length;
5350 		ctsio->kern_total_len = length;
5351 		ctsio->kern_data_resid = 0;
5352 		ctsio->kern_rel_offset = 0;
5353 		ctsio->kern_sg_entries = 0;
5354 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5355 		ctsio->be_move_done = ctl_config_move_done;
5356 		ctl_datamove((union ctl_io *)ctsio);
5357 
5358 		return (CTL_RETVAL_COMPLETE);
5359 	}
5360 
5361 	if (length > 0)
5362 		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5363 
5364 	mtx_lock(&lun->lun_lock);
5365 	if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx != residx)) {
5366 		ctl_set_reservation_conflict(ctsio);
5367 		goto bailout;
5368 	}
5369 
5370 	lun->flags |= CTL_LUN_RESERVED;
5371 	lun->res_idx = residx;
5372 
5373 	ctl_set_success(ctsio);
5374 
5375 bailout:
5376 	mtx_unlock(&lun->lun_lock);
5377 
5378 	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5379 		free(ctsio->kern_data_ptr, M_CTL);
5380 		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5381 	}
5382 
5383 	ctl_done((union ctl_io *)ctsio);
5384 	return (CTL_RETVAL_COMPLETE);
5385 }
5386 
5387 int
5388 ctl_start_stop(struct ctl_scsiio *ctsio)
5389 {
5390 	struct scsi_start_stop_unit *cdb;
5391 	struct ctl_lun *lun;
5392 	struct ctl_softc *ctl_softc;
5393 	int retval;
5394 
5395 	CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5396 
5397 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5398 	ctl_softc = control_softc;
5399 	retval = 0;
5400 
5401 	cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5402 
5403 	/*
5404 	 * XXX KDM
5405 	 * We don't support the immediate bit on a stop unit.  In order to
5406 	 * do that, we would need to code up a way to know that a stop is
5407 	 * pending, and hold off any new commands until it completes, one
5408 	 * way or another.  Then we could accept or reject those commands
5409 	 * depending on its status.  We would almost need to do the reverse
5410 	 * of what we do below for an immediate start -- return the copy of
5411 	 * the ctl_io to the FETD with status to send to the host (and to
5412 	 * free the copy!) and then free the original I/O once the stop
5413 	 * actually completes.  That way, the OOA queue mechanism can work
5414 	 * to block commands that shouldn't proceed.  Another alternative
5415 	 * would be to put the copy in the queue in place of the original,
5416 	 * and return the original back to the caller.  That could be
5417 	 * slightly safer..
5418 	 */
5419 	if ((cdb->byte2 & SSS_IMMED)
5420 	 && ((cdb->how & SSS_START) == 0)) {
5421 		ctl_set_invalid_field(ctsio,
5422 				      /*sks_valid*/ 1,
5423 				      /*command*/ 1,
5424 				      /*field*/ 1,
5425 				      /*bit_valid*/ 1,
5426 				      /*bit*/ 0);
5427 		ctl_done((union ctl_io *)ctsio);
5428 		return (CTL_RETVAL_COMPLETE);
5429 	}
5430 
5431 	if ((lun->flags & CTL_LUN_PR_RESERVED)
5432 	 && ((cdb->how & SSS_START)==0)) {
5433 		uint32_t residx;
5434 
5435 		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5436 		if (ctl_get_prkey(lun, residx) == 0
5437 		 || (lun->pr_res_idx!=residx && lun->res_type < 4)) {
5438 
5439 			ctl_set_reservation_conflict(ctsio);
5440 			ctl_done((union ctl_io *)ctsio);
5441 			return (CTL_RETVAL_COMPLETE);
5442 		}
5443 	}
5444 
5445 	/*
5446 	 * If there is no backend on this device, we can't start or stop
5447 	 * it.  In theory we shouldn't get any start/stop commands in the
5448 	 * first place at this level if the LUN doesn't have a backend.
5449 	 * That should get stopped by the command decode code.
5450 	 */
5451 	if (lun->backend == NULL) {
5452 		ctl_set_invalid_opcode(ctsio);
5453 		ctl_done((union ctl_io *)ctsio);
5454 		return (CTL_RETVAL_COMPLETE);
5455 	}
5456 
5457 	/*
5458 	 * XXX KDM Copan-specific offline behavior.
5459 	 * Figure out a reasonable way to port this?
5460 	 */
5461 #ifdef NEEDTOPORT
5462 	mtx_lock(&lun->lun_lock);
5463 
5464 	if (((cdb->byte2 & SSS_ONOFFLINE) == 0)
5465 	 && (lun->flags & CTL_LUN_OFFLINE)) {
5466 		/*
5467 		 * If the LUN is offline, and the on/offline bit isn't set,
5468 		 * reject the start or stop.  Otherwise, let it through.
5469 		 */
5470 		mtx_unlock(&lun->lun_lock);
5471 		ctl_set_lun_not_ready(ctsio);
5472 		ctl_done((union ctl_io *)ctsio);
5473 	} else {
5474 		mtx_unlock(&lun->lun_lock);
5475 #endif /* NEEDTOPORT */
5476 		/*
5477 		 * This could be a start or a stop when we're online,
5478 		 * or a stop/offline or start/online.  A start or stop when
5479 		 * we're offline is covered in the case above.
5480 		 */
5481 		/*
5482 		 * In the non-immediate case, we send the request to
5483 		 * the backend and return status to the user when
5484 		 * it is done.
5485 		 *
5486 		 * In the immediate case, we allocate a new ctl_io
5487 		 * to hold a copy of the request, and send that to
5488 		 * the backend.  We then set good status on the
5489 		 * user's request and return it immediately.
5490 		 */
5491 		if (cdb->byte2 & SSS_IMMED) {
5492 			union ctl_io *new_io;
5493 
5494 			new_io = ctl_alloc_io(ctsio->io_hdr.pool);
5495 			ctl_copy_io((union ctl_io *)ctsio, new_io);
5496 			retval = lun->backend->config_write(new_io);
5497 			ctl_set_success(ctsio);
5498 			ctl_done((union ctl_io *)ctsio);
5499 		} else {
5500 			retval = lun->backend->config_write(
5501 				(union ctl_io *)ctsio);
5502 		}
5503 #ifdef NEEDTOPORT
5504 	}
5505 #endif
5506 	return (retval);
5507 }
5508 
5509 /*
5510  * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5511  * we don't really do anything with the LBA and length fields if the user
5512  * passes them in.  Instead we'll just flush out the cache for the entire
5513  * LUN.
5514  */
5515 int
5516 ctl_sync_cache(struct ctl_scsiio *ctsio)
5517 {
5518 	struct ctl_lun *lun;
5519 	struct ctl_softc *ctl_softc;
5520 	uint64_t starting_lba;
5521 	uint32_t block_count;
5522 	int retval;
5523 
5524 	CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5525 
5526 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5527 	ctl_softc = control_softc;
5528 	retval = 0;
5529 
5530 	switch (ctsio->cdb[0]) {
5531 	case SYNCHRONIZE_CACHE: {
5532 		struct scsi_sync_cache *cdb;
5533 		cdb = (struct scsi_sync_cache *)ctsio->cdb;
5534 
5535 		starting_lba = scsi_4btoul(cdb->begin_lba);
5536 		block_count = scsi_2btoul(cdb->lb_count);
5537 		break;
5538 	}
5539 	case SYNCHRONIZE_CACHE_16: {
5540 		struct scsi_sync_cache_16 *cdb;
5541 		cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5542 
5543 		starting_lba = scsi_8btou64(cdb->begin_lba);
5544 		block_count = scsi_4btoul(cdb->lb_count);
5545 		break;
5546 	}
5547 	default:
5548 		ctl_set_invalid_opcode(ctsio);
5549 		ctl_done((union ctl_io *)ctsio);
5550 		goto bailout;
5551 		break; /* NOTREACHED */
5552 	}
5553 
5554 	/*
5555 	 * We check the LBA and length, but don't do anything with them.
5556 	 * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5557 	 * get flushed.  This check will just help satisfy anyone who wants
5558 	 * to see an error for an out of range LBA.
5559 	 */
5560 	if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5561 		ctl_set_lba_out_of_range(ctsio);
5562 		ctl_done((union ctl_io *)ctsio);
5563 		goto bailout;
5564 	}
5565 
5566 	/*
5567 	 * If this LUN has no backend, we can't flush the cache anyway.
5568 	 */
5569 	if (lun->backend == NULL) {
5570 		ctl_set_invalid_opcode(ctsio);
5571 		ctl_done((union ctl_io *)ctsio);
5572 		goto bailout;
5573 	}
5574 
5575 	/*
5576 	 * Check to see whether we're configured to send the SYNCHRONIZE
5577 	 * CACHE command directly to the back end.
5578 	 */
5579 	mtx_lock(&lun->lun_lock);
5580 	if ((ctl_softc->flags & CTL_FLAG_REAL_SYNC)
5581 	 && (++(lun->sync_count) >= lun->sync_interval)) {
5582 		lun->sync_count = 0;
5583 		mtx_unlock(&lun->lun_lock);
5584 		retval = lun->backend->config_write((union ctl_io *)ctsio);
5585 	} else {
5586 		mtx_unlock(&lun->lun_lock);
5587 		ctl_set_success(ctsio);
5588 		ctl_done((union ctl_io *)ctsio);
5589 	}
5590 
5591 bailout:
5592 
5593 	return (retval);
5594 }
5595 
5596 int
5597 ctl_format(struct ctl_scsiio *ctsio)
5598 {
5599 	struct scsi_format *cdb;
5600 	struct ctl_lun *lun;
5601 	struct ctl_softc *ctl_softc;
5602 	int length, defect_list_len;
5603 
5604 	CTL_DEBUG_PRINT(("ctl_format\n"));
5605 
5606 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5607 	ctl_softc = control_softc;
5608 
5609 	cdb = (struct scsi_format *)ctsio->cdb;
5610 
5611 	length = 0;
5612 	if (cdb->byte2 & SF_FMTDATA) {
5613 		if (cdb->byte2 & SF_LONGLIST)
5614 			length = sizeof(struct scsi_format_header_long);
5615 		else
5616 			length = sizeof(struct scsi_format_header_short);
5617 	}
5618 
5619 	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5620 	 && (length > 0)) {
5621 		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5622 		ctsio->kern_data_len = length;
5623 		ctsio->kern_total_len = length;
5624 		ctsio->kern_data_resid = 0;
5625 		ctsio->kern_rel_offset = 0;
5626 		ctsio->kern_sg_entries = 0;
5627 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5628 		ctsio->be_move_done = ctl_config_move_done;
5629 		ctl_datamove((union ctl_io *)ctsio);
5630 
5631 		return (CTL_RETVAL_COMPLETE);
5632 	}
5633 
5634 	defect_list_len = 0;
5635 
5636 	if (cdb->byte2 & SF_FMTDATA) {
5637 		if (cdb->byte2 & SF_LONGLIST) {
5638 			struct scsi_format_header_long *header;
5639 
5640 			header = (struct scsi_format_header_long *)
5641 				ctsio->kern_data_ptr;
5642 
5643 			defect_list_len = scsi_4btoul(header->defect_list_len);
5644 			if (defect_list_len != 0) {
5645 				ctl_set_invalid_field(ctsio,
5646 						      /*sks_valid*/ 1,
5647 						      /*command*/ 0,
5648 						      /*field*/ 2,
5649 						      /*bit_valid*/ 0,
5650 						      /*bit*/ 0);
5651 				goto bailout;
5652 			}
5653 		} else {
5654 			struct scsi_format_header_short *header;
5655 
5656 			header = (struct scsi_format_header_short *)
5657 				ctsio->kern_data_ptr;
5658 
5659 			defect_list_len = scsi_2btoul(header->defect_list_len);
5660 			if (defect_list_len != 0) {
5661 				ctl_set_invalid_field(ctsio,
5662 						      /*sks_valid*/ 1,
5663 						      /*command*/ 0,
5664 						      /*field*/ 2,
5665 						      /*bit_valid*/ 0,
5666 						      /*bit*/ 0);
5667 				goto bailout;
5668 			}
5669 		}
5670 	}
5671 
5672 	/*
5673 	 * The format command will clear out the "Medium format corrupted"
5674 	 * status if set by the configuration code.  That status is really
5675 	 * just a way to notify the host that we have lost the media, and
5676 	 * get them to issue a command that will basically make them think
5677 	 * they're blowing away the media.
5678 	 */
5679 	mtx_lock(&lun->lun_lock);
5680 	lun->flags &= ~CTL_LUN_INOPERABLE;
5681 	mtx_unlock(&lun->lun_lock);
5682 
5683 	ctl_set_success(ctsio);
5684 bailout:
5685 
5686 	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5687 		free(ctsio->kern_data_ptr, M_CTL);
5688 		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5689 	}
5690 
5691 	ctl_done((union ctl_io *)ctsio);
5692 	return (CTL_RETVAL_COMPLETE);
5693 }
5694 
5695 int
5696 ctl_read_buffer(struct ctl_scsiio *ctsio)
5697 {
5698 	struct scsi_read_buffer *cdb;
5699 	struct ctl_lun *lun;
5700 	int buffer_offset, len;
5701 	static uint8_t descr[4];
5702 	static uint8_t echo_descr[4] = { 0 };
5703 
5704 	CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5705 
5706 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5707 	cdb = (struct scsi_read_buffer *)ctsio->cdb;
5708 
5709 	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA &&
5710 	    (cdb->byte2 & RWB_MODE) != RWB_MODE_ECHO_DESCR &&
5711 	    (cdb->byte2 & RWB_MODE) != RWB_MODE_DESCR) {
5712 		ctl_set_invalid_field(ctsio,
5713 				      /*sks_valid*/ 1,
5714 				      /*command*/ 1,
5715 				      /*field*/ 1,
5716 				      /*bit_valid*/ 1,
5717 				      /*bit*/ 4);
5718 		ctl_done((union ctl_io *)ctsio);
5719 		return (CTL_RETVAL_COMPLETE);
5720 	}
5721 
5722 	len = scsi_3btoul(cdb->length);
5723 	buffer_offset = scsi_3btoul(cdb->offset);
5724 
5725 	if (buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5726 		ctl_set_invalid_field(ctsio,
5727 				      /*sks_valid*/ 1,
5728 				      /*command*/ 1,
5729 				      /*field*/ 6,
5730 				      /*bit_valid*/ 0,
5731 				      /*bit*/ 0);
5732 		ctl_done((union ctl_io *)ctsio);
5733 		return (CTL_RETVAL_COMPLETE);
5734 	}
5735 
5736 	if ((cdb->byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5737 		descr[0] = 0;
5738 		scsi_ulto3b(CTL_WRITE_BUFFER_SIZE, &descr[1]);
5739 		ctsio->kern_data_ptr = descr;
5740 		len = min(len, sizeof(descr));
5741 	} else if ((cdb->byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5742 		ctsio->kern_data_ptr = echo_descr;
5743 		len = min(len, sizeof(echo_descr));
5744 	} else {
5745 		if (lun->write_buffer == NULL) {
5746 			lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5747 			    M_CTL, M_WAITOK);
5748 		}
5749 		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5750 	}
5751 	ctsio->kern_data_len = len;
5752 	ctsio->kern_total_len = len;
5753 	ctsio->kern_data_resid = 0;
5754 	ctsio->kern_rel_offset = 0;
5755 	ctsio->kern_sg_entries = 0;
5756 	ctl_set_success(ctsio);
5757 	ctsio->be_move_done = ctl_config_move_done;
5758 	ctl_datamove((union ctl_io *)ctsio);
5759 	return (CTL_RETVAL_COMPLETE);
5760 }
5761 
5762 int
5763 ctl_write_buffer(struct ctl_scsiio *ctsio)
5764 {
5765 	struct scsi_write_buffer *cdb;
5766 	struct ctl_lun *lun;
5767 	int buffer_offset, len;
5768 
5769 	CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5770 
5771 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5772 	cdb = (struct scsi_write_buffer *)ctsio->cdb;
5773 
5774 	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5775 		ctl_set_invalid_field(ctsio,
5776 				      /*sks_valid*/ 1,
5777 				      /*command*/ 1,
5778 				      /*field*/ 1,
5779 				      /*bit_valid*/ 1,
5780 				      /*bit*/ 4);
5781 		ctl_done((union ctl_io *)ctsio);
5782 		return (CTL_RETVAL_COMPLETE);
5783 	}
5784 
5785 	len = scsi_3btoul(cdb->length);
5786 	buffer_offset = scsi_3btoul(cdb->offset);
5787 
5788 	if (buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5789 		ctl_set_invalid_field(ctsio,
5790 				      /*sks_valid*/ 1,
5791 				      /*command*/ 1,
5792 				      /*field*/ 6,
5793 				      /*bit_valid*/ 0,
5794 				      /*bit*/ 0);
5795 		ctl_done((union ctl_io *)ctsio);
5796 		return (CTL_RETVAL_COMPLETE);
5797 	}
5798 
5799 	/*
5800 	 * If we've got a kernel request that hasn't been malloced yet,
5801 	 * malloc it and tell the caller the data buffer is here.
5802 	 */
5803 	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5804 		if (lun->write_buffer == NULL) {
5805 			lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5806 			    M_CTL, M_WAITOK);
5807 		}
5808 		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5809 		ctsio->kern_data_len = len;
5810 		ctsio->kern_total_len = len;
5811 		ctsio->kern_data_resid = 0;
5812 		ctsio->kern_rel_offset = 0;
5813 		ctsio->kern_sg_entries = 0;
5814 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5815 		ctsio->be_move_done = ctl_config_move_done;
5816 		ctl_datamove((union ctl_io *)ctsio);
5817 
5818 		return (CTL_RETVAL_COMPLETE);
5819 	}
5820 
5821 	ctl_set_success(ctsio);
5822 	ctl_done((union ctl_io *)ctsio);
5823 	return (CTL_RETVAL_COMPLETE);
5824 }
5825 
5826 int
5827 ctl_write_same(struct ctl_scsiio *ctsio)
5828 {
5829 	struct ctl_lun *lun;
5830 	struct ctl_lba_len_flags *lbalen;
5831 	uint64_t lba;
5832 	uint32_t num_blocks;
5833 	int len, retval;
5834 	uint8_t byte2;
5835 
5836 	retval = CTL_RETVAL_COMPLETE;
5837 
5838 	CTL_DEBUG_PRINT(("ctl_write_same\n"));
5839 
5840 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5841 
5842 	switch (ctsio->cdb[0]) {
5843 	case WRITE_SAME_10: {
5844 		struct scsi_write_same_10 *cdb;
5845 
5846 		cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5847 
5848 		lba = scsi_4btoul(cdb->addr);
5849 		num_blocks = scsi_2btoul(cdb->length);
5850 		byte2 = cdb->byte2;
5851 		break;
5852 	}
5853 	case WRITE_SAME_16: {
5854 		struct scsi_write_same_16 *cdb;
5855 
5856 		cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5857 
5858 		lba = scsi_8btou64(cdb->addr);
5859 		num_blocks = scsi_4btoul(cdb->length);
5860 		byte2 = cdb->byte2;
5861 		break;
5862 	}
5863 	default:
5864 		/*
5865 		 * We got a command we don't support.  This shouldn't
5866 		 * happen, commands should be filtered out above us.
5867 		 */
5868 		ctl_set_invalid_opcode(ctsio);
5869 		ctl_done((union ctl_io *)ctsio);
5870 
5871 		return (CTL_RETVAL_COMPLETE);
5872 		break; /* NOTREACHED */
5873 	}
5874 
5875 	/* NDOB and ANCHOR flags can be used only together with UNMAP */
5876 	if ((byte2 & SWS_UNMAP) == 0 &&
5877 	    (byte2 & (SWS_NDOB | SWS_ANCHOR)) != 0) {
5878 		ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
5879 		    /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
5880 		ctl_done((union ctl_io *)ctsio);
5881 		return (CTL_RETVAL_COMPLETE);
5882 	}
5883 
5884 	/*
5885 	 * The first check is to make sure we're in bounds, the second
5886 	 * check is to catch wrap-around problems.  If the lba + num blocks
5887 	 * is less than the lba, then we've wrapped around and the block
5888 	 * range is invalid anyway.
5889 	 */
5890 	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5891 	 || ((lba + num_blocks) < lba)) {
5892 		ctl_set_lba_out_of_range(ctsio);
5893 		ctl_done((union ctl_io *)ctsio);
5894 		return (CTL_RETVAL_COMPLETE);
5895 	}
5896 
5897 	/* Zero number of blocks means "to the last logical block" */
5898 	if (num_blocks == 0) {
5899 		if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5900 			ctl_set_invalid_field(ctsio,
5901 					      /*sks_valid*/ 0,
5902 					      /*command*/ 1,
5903 					      /*field*/ 0,
5904 					      /*bit_valid*/ 0,
5905 					      /*bit*/ 0);
5906 			ctl_done((union ctl_io *)ctsio);
5907 			return (CTL_RETVAL_COMPLETE);
5908 		}
5909 		num_blocks = (lun->be_lun->maxlba + 1) - lba;
5910 	}
5911 
5912 	len = lun->be_lun->blocksize;
5913 
5914 	/*
5915 	 * If we've got a kernel request that hasn't been malloced yet,
5916 	 * malloc it and tell the caller the data buffer is here.
5917 	 */
5918 	if ((byte2 & SWS_NDOB) == 0 &&
5919 	    (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5920 		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
5921 		ctsio->kern_data_len = len;
5922 		ctsio->kern_total_len = len;
5923 		ctsio->kern_data_resid = 0;
5924 		ctsio->kern_rel_offset = 0;
5925 		ctsio->kern_sg_entries = 0;
5926 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5927 		ctsio->be_move_done = ctl_config_move_done;
5928 		ctl_datamove((union ctl_io *)ctsio);
5929 
5930 		return (CTL_RETVAL_COMPLETE);
5931 	}
5932 
5933 	lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5934 	lbalen->lba = lba;
5935 	lbalen->len = num_blocks;
5936 	lbalen->flags = byte2;
5937 	retval = lun->backend->config_write((union ctl_io *)ctsio);
5938 
5939 	return (retval);
5940 }
5941 
5942 int
5943 ctl_unmap(struct ctl_scsiio *ctsio)
5944 {
5945 	struct ctl_lun *lun;
5946 	struct scsi_unmap *cdb;
5947 	struct ctl_ptr_len_flags *ptrlen;
5948 	struct scsi_unmap_header *hdr;
5949 	struct scsi_unmap_desc *buf, *end, *endnz, *range;
5950 	uint64_t lba;
5951 	uint32_t num_blocks;
5952 	int len, retval;
5953 	uint8_t byte2;
5954 
5955 	retval = CTL_RETVAL_COMPLETE;
5956 
5957 	CTL_DEBUG_PRINT(("ctl_unmap\n"));
5958 
5959 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5960 	cdb = (struct scsi_unmap *)ctsio->cdb;
5961 
5962 	len = scsi_2btoul(cdb->length);
5963 	byte2 = cdb->byte2;
5964 
5965 	/*
5966 	 * If we've got a kernel request that hasn't been malloced yet,
5967 	 * malloc it and tell the caller the data buffer is here.
5968 	 */
5969 	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5970 		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
5971 		ctsio->kern_data_len = len;
5972 		ctsio->kern_total_len = len;
5973 		ctsio->kern_data_resid = 0;
5974 		ctsio->kern_rel_offset = 0;
5975 		ctsio->kern_sg_entries = 0;
5976 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5977 		ctsio->be_move_done = ctl_config_move_done;
5978 		ctl_datamove((union ctl_io *)ctsio);
5979 
5980 		return (CTL_RETVAL_COMPLETE);
5981 	}
5982 
5983 	len = ctsio->kern_total_len - ctsio->kern_data_resid;
5984 	hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
5985 	if (len < sizeof (*hdr) ||
5986 	    len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
5987 	    len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
5988 	    scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
5989 		ctl_set_invalid_field(ctsio,
5990 				      /*sks_valid*/ 0,
5991 				      /*command*/ 0,
5992 				      /*field*/ 0,
5993 				      /*bit_valid*/ 0,
5994 				      /*bit*/ 0);
5995 		goto done;
5996 	}
5997 	len = scsi_2btoul(hdr->desc_length);
5998 	buf = (struct scsi_unmap_desc *)(hdr + 1);
5999 	end = buf + len / sizeof(*buf);
6000 
6001 	endnz = buf;
6002 	for (range = buf; range < end; range++) {
6003 		lba = scsi_8btou64(range->lba);
6004 		num_blocks = scsi_4btoul(range->length);
6005 		if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
6006 		 || ((lba + num_blocks) < lba)) {
6007 			ctl_set_lba_out_of_range(ctsio);
6008 			ctl_done((union ctl_io *)ctsio);
6009 			return (CTL_RETVAL_COMPLETE);
6010 		}
6011 		if (num_blocks != 0)
6012 			endnz = range + 1;
6013 	}
6014 
6015 	/*
6016 	 * Block backend can not handle zero last range.
6017 	 * Filter it out and return if there is nothing left.
6018 	 */
6019 	len = (uint8_t *)endnz - (uint8_t *)buf;
6020 	if (len == 0) {
6021 		ctl_set_success(ctsio);
6022 		goto done;
6023 	}
6024 
6025 	mtx_lock(&lun->lun_lock);
6026 	ptrlen = (struct ctl_ptr_len_flags *)
6027 	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
6028 	ptrlen->ptr = (void *)buf;
6029 	ptrlen->len = len;
6030 	ptrlen->flags = byte2;
6031 	ctl_check_blocked(lun);
6032 	mtx_unlock(&lun->lun_lock);
6033 
6034 	retval = lun->backend->config_write((union ctl_io *)ctsio);
6035 	return (retval);
6036 
6037 done:
6038 	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
6039 		free(ctsio->kern_data_ptr, M_CTL);
6040 		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
6041 	}
6042 	ctl_done((union ctl_io *)ctsio);
6043 	return (CTL_RETVAL_COMPLETE);
6044 }
6045 
6046 /*
6047  * Note that this function currently doesn't actually do anything inside
6048  * CTL to enforce things if the DQue bit is turned on.
6049  *
6050  * Also note that this function can't be used in the default case, because
6051  * the DQue bit isn't set in the changeable mask for the control mode page
6052  * anyway.  This is just here as an example for how to implement a page
6053  * handler, and a placeholder in case we want to allow the user to turn
6054  * tagged queueing on and off.
6055  *
6056  * The D_SENSE bit handling is functional, however, and will turn
6057  * descriptor sense on and off for a given LUN.
6058  */
6059 int
6060 ctl_control_page_handler(struct ctl_scsiio *ctsio,
6061 			 struct ctl_page_index *page_index, uint8_t *page_ptr)
6062 {
6063 	struct scsi_control_page *current_cp, *saved_cp, *user_cp;
6064 	struct ctl_lun *lun;
6065 	struct ctl_softc *softc;
6066 	int set_ua;
6067 	uint32_t initidx;
6068 
6069 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6070 	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6071 	set_ua = 0;
6072 
6073 	user_cp = (struct scsi_control_page *)page_ptr;
6074 	current_cp = (struct scsi_control_page *)
6075 		(page_index->page_data + (page_index->page_len *
6076 		CTL_PAGE_CURRENT));
6077 	saved_cp = (struct scsi_control_page *)
6078 		(page_index->page_data + (page_index->page_len *
6079 		CTL_PAGE_SAVED));
6080 
6081 	softc = control_softc;
6082 
6083 	mtx_lock(&lun->lun_lock);
6084 	if (((current_cp->rlec & SCP_DSENSE) == 0)
6085 	 && ((user_cp->rlec & SCP_DSENSE) != 0)) {
6086 		/*
6087 		 * Descriptor sense is currently turned off and the user
6088 		 * wants to turn it on.
6089 		 */
6090 		current_cp->rlec |= SCP_DSENSE;
6091 		saved_cp->rlec |= SCP_DSENSE;
6092 		lun->flags |= CTL_LUN_SENSE_DESC;
6093 		set_ua = 1;
6094 	} else if (((current_cp->rlec & SCP_DSENSE) != 0)
6095 		&& ((user_cp->rlec & SCP_DSENSE) == 0)) {
6096 		/*
6097 		 * Descriptor sense is currently turned on, and the user
6098 		 * wants to turn it off.
6099 		 */
6100 		current_cp->rlec &= ~SCP_DSENSE;
6101 		saved_cp->rlec &= ~SCP_DSENSE;
6102 		lun->flags &= ~CTL_LUN_SENSE_DESC;
6103 		set_ua = 1;
6104 	}
6105 	if ((current_cp->queue_flags & SCP_QUEUE_ALG_MASK) !=
6106 	    (user_cp->queue_flags & SCP_QUEUE_ALG_MASK)) {
6107 		current_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
6108 		current_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
6109 		saved_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
6110 		saved_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
6111 		set_ua = 1;
6112 	}
6113 	if ((current_cp->eca_and_aen & SCP_SWP) !=
6114 	    (user_cp->eca_and_aen & SCP_SWP)) {
6115 		current_cp->eca_and_aen &= ~SCP_SWP;
6116 		current_cp->eca_and_aen |= user_cp->eca_and_aen & SCP_SWP;
6117 		saved_cp->eca_and_aen &= ~SCP_SWP;
6118 		saved_cp->eca_and_aen |= user_cp->eca_and_aen & SCP_SWP;
6119 		set_ua = 1;
6120 	}
6121 	if (set_ua != 0)
6122 		ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
6123 	mtx_unlock(&lun->lun_lock);
6124 
6125 	return (0);
6126 }
6127 
6128 int
6129 ctl_caching_sp_handler(struct ctl_scsiio *ctsio,
6130 		     struct ctl_page_index *page_index, uint8_t *page_ptr)
6131 {
6132 	struct scsi_caching_page *current_cp, *saved_cp, *user_cp;
6133 	struct ctl_lun *lun;
6134 	int set_ua;
6135 	uint32_t initidx;
6136 
6137 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6138 	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6139 	set_ua = 0;
6140 
6141 	user_cp = (struct scsi_caching_page *)page_ptr;
6142 	current_cp = (struct scsi_caching_page *)
6143 		(page_index->page_data + (page_index->page_len *
6144 		CTL_PAGE_CURRENT));
6145 	saved_cp = (struct scsi_caching_page *)
6146 		(page_index->page_data + (page_index->page_len *
6147 		CTL_PAGE_SAVED));
6148 
6149 	mtx_lock(&lun->lun_lock);
6150 	if ((current_cp->flags1 & (SCP_WCE | SCP_RCD)) !=
6151 	    (user_cp->flags1 & (SCP_WCE | SCP_RCD))) {
6152 		current_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
6153 		current_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
6154 		saved_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
6155 		saved_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
6156 		set_ua = 1;
6157 	}
6158 	if (set_ua != 0)
6159 		ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
6160 	mtx_unlock(&lun->lun_lock);
6161 
6162 	return (0);
6163 }
6164 
6165 int
6166 ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio,
6167 				struct ctl_page_index *page_index,
6168 				uint8_t *page_ptr)
6169 {
6170 	uint8_t *c;
6171 	int i;
6172 
6173 	c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs;
6174 	ctl_time_io_secs =
6175 		(c[0] << 8) |
6176 		(c[1] << 0) |
6177 		0;
6178 	CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs));
6179 	printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs);
6180 	printf("page data:");
6181 	for (i=0; i<8; i++)
6182 		printf(" %.2x",page_ptr[i]);
6183 	printf("\n");
6184 	return (0);
6185 }
6186 
6187 int
6188 ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio,
6189 			       struct ctl_page_index *page_index,
6190 			       int pc)
6191 {
6192 	struct copan_debugconf_subpage *page;
6193 
6194 	page = (struct copan_debugconf_subpage *)page_index->page_data +
6195 		(page_index->page_len * pc);
6196 
6197 	switch (pc) {
6198 	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6199 	case SMS_PAGE_CTRL_DEFAULT >> 6:
6200 	case SMS_PAGE_CTRL_SAVED >> 6:
6201 		/*
6202 		 * We don't update the changable or default bits for this page.
6203 		 */
6204 		break;
6205 	case SMS_PAGE_CTRL_CURRENT >> 6:
6206 		page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8;
6207 		page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0;
6208 		break;
6209 	default:
6210 #ifdef NEEDTOPORT
6211 		EPRINT(0, "Invalid PC %d!!", pc);
6212 #endif /* NEEDTOPORT */
6213 		break;
6214 	}
6215 	return (0);
6216 }
6217 
6218 
6219 static int
6220 ctl_do_mode_select(union ctl_io *io)
6221 {
6222 	struct scsi_mode_page_header *page_header;
6223 	struct ctl_page_index *page_index;
6224 	struct ctl_scsiio *ctsio;
6225 	int control_dev, page_len;
6226 	int page_len_offset, page_len_size;
6227 	union ctl_modepage_info *modepage_info;
6228 	struct ctl_lun *lun;
6229 	int *len_left, *len_used;
6230 	int retval, i;
6231 
6232 	ctsio = &io->scsiio;
6233 	page_index = NULL;
6234 	page_len = 0;
6235 	retval = CTL_RETVAL_COMPLETE;
6236 
6237 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6238 
6239 	if (lun->be_lun->lun_type != T_DIRECT)
6240 		control_dev = 1;
6241 	else
6242 		control_dev = 0;
6243 
6244 	modepage_info = (union ctl_modepage_info *)
6245 		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6246 	len_left = &modepage_info->header.len_left;
6247 	len_used = &modepage_info->header.len_used;
6248 
6249 do_next_page:
6250 
6251 	page_header = (struct scsi_mode_page_header *)
6252 		(ctsio->kern_data_ptr + *len_used);
6253 
6254 	if (*len_left == 0) {
6255 		free(ctsio->kern_data_ptr, M_CTL);
6256 		ctl_set_success(ctsio);
6257 		ctl_done((union ctl_io *)ctsio);
6258 		return (CTL_RETVAL_COMPLETE);
6259 	} else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6260 
6261 		free(ctsio->kern_data_ptr, M_CTL);
6262 		ctl_set_param_len_error(ctsio);
6263 		ctl_done((union ctl_io *)ctsio);
6264 		return (CTL_RETVAL_COMPLETE);
6265 
6266 	} else if ((page_header->page_code & SMPH_SPF)
6267 		&& (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6268 
6269 		free(ctsio->kern_data_ptr, M_CTL);
6270 		ctl_set_param_len_error(ctsio);
6271 		ctl_done((union ctl_io *)ctsio);
6272 		return (CTL_RETVAL_COMPLETE);
6273 	}
6274 
6275 
6276 	/*
6277 	 * XXX KDM should we do something with the block descriptor?
6278 	 */
6279 	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6280 
6281 		if ((control_dev != 0)
6282 		 && (lun->mode_pages.index[i].page_flags &
6283 		     CTL_PAGE_FLAG_DISK_ONLY))
6284 			continue;
6285 
6286 		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
6287 		    (page_header->page_code & SMPH_PC_MASK))
6288 			continue;
6289 
6290 		/*
6291 		 * If neither page has a subpage code, then we've got a
6292 		 * match.
6293 		 */
6294 		if (((lun->mode_pages.index[i].page_code & SMPH_SPF) == 0)
6295 		 && ((page_header->page_code & SMPH_SPF) == 0)) {
6296 			page_index = &lun->mode_pages.index[i];
6297 			page_len = page_header->page_length;
6298 			break;
6299 		}
6300 
6301 		/*
6302 		 * If both pages have subpages, then the subpage numbers
6303 		 * have to match.
6304 		 */
6305 		if ((lun->mode_pages.index[i].page_code & SMPH_SPF)
6306 		  && (page_header->page_code & SMPH_SPF)) {
6307 			struct scsi_mode_page_header_sp *sph;
6308 
6309 			sph = (struct scsi_mode_page_header_sp *)page_header;
6310 
6311 			if (lun->mode_pages.index[i].subpage ==
6312 			    sph->subpage) {
6313 				page_index = &lun->mode_pages.index[i];
6314 				page_len = scsi_2btoul(sph->page_length);
6315 				break;
6316 			}
6317 		}
6318 	}
6319 
6320 	/*
6321 	 * If we couldn't find the page, or if we don't have a mode select
6322 	 * handler for it, send back an error to the user.
6323 	 */
6324 	if ((page_index == NULL)
6325 	 || (page_index->select_handler == NULL)) {
6326 		ctl_set_invalid_field(ctsio,
6327 				      /*sks_valid*/ 1,
6328 				      /*command*/ 0,
6329 				      /*field*/ *len_used,
6330 				      /*bit_valid*/ 0,
6331 				      /*bit*/ 0);
6332 		free(ctsio->kern_data_ptr, M_CTL);
6333 		ctl_done((union ctl_io *)ctsio);
6334 		return (CTL_RETVAL_COMPLETE);
6335 	}
6336 
6337 	if (page_index->page_code & SMPH_SPF) {
6338 		page_len_offset = 2;
6339 		page_len_size = 2;
6340 	} else {
6341 		page_len_size = 1;
6342 		page_len_offset = 1;
6343 	}
6344 
6345 	/*
6346 	 * If the length the initiator gives us isn't the one we specify in
6347 	 * the mode page header, or if they didn't specify enough data in
6348 	 * the CDB to avoid truncating this page, kick out the request.
6349 	 */
6350 	if ((page_len != (page_index->page_len - page_len_offset -
6351 			  page_len_size))
6352 	 || (*len_left < page_index->page_len)) {
6353 
6354 
6355 		ctl_set_invalid_field(ctsio,
6356 				      /*sks_valid*/ 1,
6357 				      /*command*/ 0,
6358 				      /*field*/ *len_used + page_len_offset,
6359 				      /*bit_valid*/ 0,
6360 				      /*bit*/ 0);
6361 		free(ctsio->kern_data_ptr, M_CTL);
6362 		ctl_done((union ctl_io *)ctsio);
6363 		return (CTL_RETVAL_COMPLETE);
6364 	}
6365 
6366 	/*
6367 	 * Run through the mode page, checking to make sure that the bits
6368 	 * the user changed are actually legal for him to change.
6369 	 */
6370 	for (i = 0; i < page_index->page_len; i++) {
6371 		uint8_t *user_byte, *change_mask, *current_byte;
6372 		int bad_bit;
6373 		int j;
6374 
6375 		user_byte = (uint8_t *)page_header + i;
6376 		change_mask = page_index->page_data +
6377 			      (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6378 		current_byte = page_index->page_data +
6379 			       (page_index->page_len * CTL_PAGE_CURRENT) + i;
6380 
6381 		/*
6382 		 * Check to see whether the user set any bits in this byte
6383 		 * that he is not allowed to set.
6384 		 */
6385 		if ((*user_byte & ~(*change_mask)) ==
6386 		    (*current_byte & ~(*change_mask)))
6387 			continue;
6388 
6389 		/*
6390 		 * Go through bit by bit to determine which one is illegal.
6391 		 */
6392 		bad_bit = 0;
6393 		for (j = 7; j >= 0; j--) {
6394 			if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6395 			    (((1 << i) & ~(*change_mask)) & *current_byte)) {
6396 				bad_bit = i;
6397 				break;
6398 			}
6399 		}
6400 		ctl_set_invalid_field(ctsio,
6401 				      /*sks_valid*/ 1,
6402 				      /*command*/ 0,
6403 				      /*field*/ *len_used + i,
6404 				      /*bit_valid*/ 1,
6405 				      /*bit*/ bad_bit);
6406 		free(ctsio->kern_data_ptr, M_CTL);
6407 		ctl_done((union ctl_io *)ctsio);
6408 		return (CTL_RETVAL_COMPLETE);
6409 	}
6410 
6411 	/*
6412 	 * Decrement these before we call the page handler, since we may
6413 	 * end up getting called back one way or another before the handler
6414 	 * returns to this context.
6415 	 */
6416 	*len_left -= page_index->page_len;
6417 	*len_used += page_index->page_len;
6418 
6419 	retval = page_index->select_handler(ctsio, page_index,
6420 					    (uint8_t *)page_header);
6421 
6422 	/*
6423 	 * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6424 	 * wait until this queued command completes to finish processing
6425 	 * the mode page.  If it returns anything other than
6426 	 * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6427 	 * already set the sense information, freed the data pointer, and
6428 	 * completed the io for us.
6429 	 */
6430 	if (retval != CTL_RETVAL_COMPLETE)
6431 		goto bailout_no_done;
6432 
6433 	/*
6434 	 * If the initiator sent us more than one page, parse the next one.
6435 	 */
6436 	if (*len_left > 0)
6437 		goto do_next_page;
6438 
6439 	ctl_set_success(ctsio);
6440 	free(ctsio->kern_data_ptr, M_CTL);
6441 	ctl_done((union ctl_io *)ctsio);
6442 
6443 bailout_no_done:
6444 
6445 	return (CTL_RETVAL_COMPLETE);
6446 
6447 }
6448 
6449 int
6450 ctl_mode_select(struct ctl_scsiio *ctsio)
6451 {
6452 	int param_len, pf, sp;
6453 	int header_size, bd_len;
6454 	int len_left, len_used;
6455 	struct ctl_page_index *page_index;
6456 	struct ctl_lun *lun;
6457 	int control_dev, page_len;
6458 	union ctl_modepage_info *modepage_info;
6459 	int retval;
6460 
6461 	pf = 0;
6462 	sp = 0;
6463 	page_len = 0;
6464 	len_used = 0;
6465 	len_left = 0;
6466 	retval = 0;
6467 	bd_len = 0;
6468 	page_index = NULL;
6469 
6470 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6471 
6472 	if (lun->be_lun->lun_type != T_DIRECT)
6473 		control_dev = 1;
6474 	else
6475 		control_dev = 0;
6476 
6477 	switch (ctsio->cdb[0]) {
6478 	case MODE_SELECT_6: {
6479 		struct scsi_mode_select_6 *cdb;
6480 
6481 		cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6482 
6483 		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6484 		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6485 
6486 		param_len = cdb->length;
6487 		header_size = sizeof(struct scsi_mode_header_6);
6488 		break;
6489 	}
6490 	case MODE_SELECT_10: {
6491 		struct scsi_mode_select_10 *cdb;
6492 
6493 		cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6494 
6495 		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6496 		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6497 
6498 		param_len = scsi_2btoul(cdb->length);
6499 		header_size = sizeof(struct scsi_mode_header_10);
6500 		break;
6501 	}
6502 	default:
6503 		ctl_set_invalid_opcode(ctsio);
6504 		ctl_done((union ctl_io *)ctsio);
6505 		return (CTL_RETVAL_COMPLETE);
6506 		break; /* NOTREACHED */
6507 	}
6508 
6509 	/*
6510 	 * From SPC-3:
6511 	 * "A parameter list length of zero indicates that the Data-Out Buffer
6512 	 * shall be empty. This condition shall not be considered as an error."
6513 	 */
6514 	if (param_len == 0) {
6515 		ctl_set_success(ctsio);
6516 		ctl_done((union ctl_io *)ctsio);
6517 		return (CTL_RETVAL_COMPLETE);
6518 	}
6519 
6520 	/*
6521 	 * Since we'll hit this the first time through, prior to
6522 	 * allocation, we don't need to free a data buffer here.
6523 	 */
6524 	if (param_len < header_size) {
6525 		ctl_set_param_len_error(ctsio);
6526 		ctl_done((union ctl_io *)ctsio);
6527 		return (CTL_RETVAL_COMPLETE);
6528 	}
6529 
6530 	/*
6531 	 * Allocate the data buffer and grab the user's data.  In theory,
6532 	 * we shouldn't have to sanity check the parameter list length here
6533 	 * because the maximum size is 64K.  We should be able to malloc
6534 	 * that much without too many problems.
6535 	 */
6536 	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6537 		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6538 		ctsio->kern_data_len = param_len;
6539 		ctsio->kern_total_len = param_len;
6540 		ctsio->kern_data_resid = 0;
6541 		ctsio->kern_rel_offset = 0;
6542 		ctsio->kern_sg_entries = 0;
6543 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6544 		ctsio->be_move_done = ctl_config_move_done;
6545 		ctl_datamove((union ctl_io *)ctsio);
6546 
6547 		return (CTL_RETVAL_COMPLETE);
6548 	}
6549 
6550 	switch (ctsio->cdb[0]) {
6551 	case MODE_SELECT_6: {
6552 		struct scsi_mode_header_6 *mh6;
6553 
6554 		mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6555 		bd_len = mh6->blk_desc_len;
6556 		break;
6557 	}
6558 	case MODE_SELECT_10: {
6559 		struct scsi_mode_header_10 *mh10;
6560 
6561 		mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6562 		bd_len = scsi_2btoul(mh10->blk_desc_len);
6563 		break;
6564 	}
6565 	default:
6566 		panic("Invalid CDB type %#x", ctsio->cdb[0]);
6567 		break;
6568 	}
6569 
6570 	if (param_len < (header_size + bd_len)) {
6571 		free(ctsio->kern_data_ptr, M_CTL);
6572 		ctl_set_param_len_error(ctsio);
6573 		ctl_done((union ctl_io *)ctsio);
6574 		return (CTL_RETVAL_COMPLETE);
6575 	}
6576 
6577 	/*
6578 	 * Set the IO_CONT flag, so that if this I/O gets passed to
6579 	 * ctl_config_write_done(), it'll get passed back to
6580 	 * ctl_do_mode_select() for further processing, or completion if
6581 	 * we're all done.
6582 	 */
6583 	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6584 	ctsio->io_cont = ctl_do_mode_select;
6585 
6586 	modepage_info = (union ctl_modepage_info *)
6587 		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6588 
6589 	memset(modepage_info, 0, sizeof(*modepage_info));
6590 
6591 	len_left = param_len - header_size - bd_len;
6592 	len_used = header_size + bd_len;
6593 
6594 	modepage_info->header.len_left = len_left;
6595 	modepage_info->header.len_used = len_used;
6596 
6597 	return (ctl_do_mode_select((union ctl_io *)ctsio));
6598 }
6599 
6600 int
6601 ctl_mode_sense(struct ctl_scsiio *ctsio)
6602 {
6603 	struct ctl_lun *lun;
6604 	int pc, page_code, dbd, llba, subpage;
6605 	int alloc_len, page_len, header_len, total_len;
6606 	struct scsi_mode_block_descr *block_desc;
6607 	struct ctl_page_index *page_index;
6608 	int control_dev;
6609 
6610 	dbd = 0;
6611 	llba = 0;
6612 	block_desc = NULL;
6613 	page_index = NULL;
6614 
6615 	CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6616 
6617 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6618 
6619 	if (lun->be_lun->lun_type != T_DIRECT)
6620 		control_dev = 1;
6621 	else
6622 		control_dev = 0;
6623 
6624 	switch (ctsio->cdb[0]) {
6625 	case MODE_SENSE_6: {
6626 		struct scsi_mode_sense_6 *cdb;
6627 
6628 		cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6629 
6630 		header_len = sizeof(struct scsi_mode_hdr_6);
6631 		if (cdb->byte2 & SMS_DBD)
6632 			dbd = 1;
6633 		else
6634 			header_len += sizeof(struct scsi_mode_block_descr);
6635 
6636 		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6637 		page_code = cdb->page & SMS_PAGE_CODE;
6638 		subpage = cdb->subpage;
6639 		alloc_len = cdb->length;
6640 		break;
6641 	}
6642 	case MODE_SENSE_10: {
6643 		struct scsi_mode_sense_10 *cdb;
6644 
6645 		cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6646 
6647 		header_len = sizeof(struct scsi_mode_hdr_10);
6648 
6649 		if (cdb->byte2 & SMS_DBD)
6650 			dbd = 1;
6651 		else
6652 			header_len += sizeof(struct scsi_mode_block_descr);
6653 		if (cdb->byte2 & SMS10_LLBAA)
6654 			llba = 1;
6655 		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6656 		page_code = cdb->page & SMS_PAGE_CODE;
6657 		subpage = cdb->subpage;
6658 		alloc_len = scsi_2btoul(cdb->length);
6659 		break;
6660 	}
6661 	default:
6662 		ctl_set_invalid_opcode(ctsio);
6663 		ctl_done((union ctl_io *)ctsio);
6664 		return (CTL_RETVAL_COMPLETE);
6665 		break; /* NOTREACHED */
6666 	}
6667 
6668 	/*
6669 	 * We have to make a first pass through to calculate the size of
6670 	 * the pages that match the user's query.  Then we allocate enough
6671 	 * memory to hold it, and actually copy the data into the buffer.
6672 	 */
6673 	switch (page_code) {
6674 	case SMS_ALL_PAGES_PAGE: {
6675 		int i;
6676 
6677 		page_len = 0;
6678 
6679 		/*
6680 		 * At the moment, values other than 0 and 0xff here are
6681 		 * reserved according to SPC-3.
6682 		 */
6683 		if ((subpage != SMS_SUBPAGE_PAGE_0)
6684 		 && (subpage != SMS_SUBPAGE_ALL)) {
6685 			ctl_set_invalid_field(ctsio,
6686 					      /*sks_valid*/ 1,
6687 					      /*command*/ 1,
6688 					      /*field*/ 3,
6689 					      /*bit_valid*/ 0,
6690 					      /*bit*/ 0);
6691 			ctl_done((union ctl_io *)ctsio);
6692 			return (CTL_RETVAL_COMPLETE);
6693 		}
6694 
6695 		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6696 			if ((control_dev != 0)
6697 			 && (lun->mode_pages.index[i].page_flags &
6698 			     CTL_PAGE_FLAG_DISK_ONLY))
6699 				continue;
6700 
6701 			/*
6702 			 * We don't use this subpage if the user didn't
6703 			 * request all subpages.
6704 			 */
6705 			if ((lun->mode_pages.index[i].subpage != 0)
6706 			 && (subpage == SMS_SUBPAGE_PAGE_0))
6707 				continue;
6708 
6709 #if 0
6710 			printf("found page %#x len %d\n",
6711 			       lun->mode_pages.index[i].page_code &
6712 			       SMPH_PC_MASK,
6713 			       lun->mode_pages.index[i].page_len);
6714 #endif
6715 			page_len += lun->mode_pages.index[i].page_len;
6716 		}
6717 		break;
6718 	}
6719 	default: {
6720 		int i;
6721 
6722 		page_len = 0;
6723 
6724 		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6725 			/* Look for the right page code */
6726 			if ((lun->mode_pages.index[i].page_code &
6727 			     SMPH_PC_MASK) != page_code)
6728 				continue;
6729 
6730 			/* Look for the right subpage or the subpage wildcard*/
6731 			if ((lun->mode_pages.index[i].subpage != subpage)
6732 			 && (subpage != SMS_SUBPAGE_ALL))
6733 				continue;
6734 
6735 			/* Make sure the page is supported for this dev type */
6736 			if ((control_dev != 0)
6737 			 && (lun->mode_pages.index[i].page_flags &
6738 			     CTL_PAGE_FLAG_DISK_ONLY))
6739 				continue;
6740 
6741 #if 0
6742 			printf("found page %#x len %d\n",
6743 			       lun->mode_pages.index[i].page_code &
6744 			       SMPH_PC_MASK,
6745 			       lun->mode_pages.index[i].page_len);
6746 #endif
6747 
6748 			page_len += lun->mode_pages.index[i].page_len;
6749 		}
6750 
6751 		if (page_len == 0) {
6752 			ctl_set_invalid_field(ctsio,
6753 					      /*sks_valid*/ 1,
6754 					      /*command*/ 1,
6755 					      /*field*/ 2,
6756 					      /*bit_valid*/ 1,
6757 					      /*bit*/ 5);
6758 			ctl_done((union ctl_io *)ctsio);
6759 			return (CTL_RETVAL_COMPLETE);
6760 		}
6761 		break;
6762 	}
6763 	}
6764 
6765 	total_len = header_len + page_len;
6766 #if 0
6767 	printf("header_len = %d, page_len = %d, total_len = %d\n",
6768 	       header_len, page_len, total_len);
6769 #endif
6770 
6771 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6772 	ctsio->kern_sg_entries = 0;
6773 	ctsio->kern_data_resid = 0;
6774 	ctsio->kern_rel_offset = 0;
6775 	if (total_len < alloc_len) {
6776 		ctsio->residual = alloc_len - total_len;
6777 		ctsio->kern_data_len = total_len;
6778 		ctsio->kern_total_len = total_len;
6779 	} else {
6780 		ctsio->residual = 0;
6781 		ctsio->kern_data_len = alloc_len;
6782 		ctsio->kern_total_len = alloc_len;
6783 	}
6784 
6785 	switch (ctsio->cdb[0]) {
6786 	case MODE_SENSE_6: {
6787 		struct scsi_mode_hdr_6 *header;
6788 
6789 		header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6790 
6791 		header->datalen = ctl_min(total_len - 1, 254);
6792 		if (control_dev == 0) {
6793 			header->dev_specific = 0x10; /* DPOFUA */
6794 			if ((lun->flags & CTL_LUN_READONLY) ||
6795 			    (lun->mode_pages.control_page[CTL_PAGE_CURRENT]
6796 			    .eca_and_aen & SCP_SWP) != 0)
6797 				    header->dev_specific |= 0x80; /* WP */
6798 		}
6799 		if (dbd)
6800 			header->block_descr_len = 0;
6801 		else
6802 			header->block_descr_len =
6803 				sizeof(struct scsi_mode_block_descr);
6804 		block_desc = (struct scsi_mode_block_descr *)&header[1];
6805 		break;
6806 	}
6807 	case MODE_SENSE_10: {
6808 		struct scsi_mode_hdr_10 *header;
6809 		int datalen;
6810 
6811 		header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6812 
6813 		datalen = ctl_min(total_len - 2, 65533);
6814 		scsi_ulto2b(datalen, header->datalen);
6815 		if (control_dev == 0) {
6816 			header->dev_specific = 0x10; /* DPOFUA */
6817 			if ((lun->flags & CTL_LUN_READONLY) ||
6818 			    (lun->mode_pages.control_page[CTL_PAGE_CURRENT]
6819 			    .eca_and_aen & SCP_SWP) != 0)
6820 				    header->dev_specific |= 0x80; /* WP */
6821 		}
6822 		if (dbd)
6823 			scsi_ulto2b(0, header->block_descr_len);
6824 		else
6825 			scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
6826 				    header->block_descr_len);
6827 		block_desc = (struct scsi_mode_block_descr *)&header[1];
6828 		break;
6829 	}
6830 	default:
6831 		panic("invalid CDB type %#x", ctsio->cdb[0]);
6832 		break; /* NOTREACHED */
6833 	}
6834 
6835 	/*
6836 	 * If we've got a disk, use its blocksize in the block
6837 	 * descriptor.  Otherwise, just set it to 0.
6838 	 */
6839 	if (dbd == 0) {
6840 		if (control_dev == 0)
6841 			scsi_ulto3b(lun->be_lun->blocksize,
6842 				    block_desc->block_len);
6843 		else
6844 			scsi_ulto3b(0, block_desc->block_len);
6845 	}
6846 
6847 	switch (page_code) {
6848 	case SMS_ALL_PAGES_PAGE: {
6849 		int i, data_used;
6850 
6851 		data_used = header_len;
6852 		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6853 			struct ctl_page_index *page_index;
6854 
6855 			page_index = &lun->mode_pages.index[i];
6856 
6857 			if ((control_dev != 0)
6858 			 && (page_index->page_flags &
6859 			    CTL_PAGE_FLAG_DISK_ONLY))
6860 				continue;
6861 
6862 			/*
6863 			 * We don't use this subpage if the user didn't
6864 			 * request all subpages.  We already checked (above)
6865 			 * to make sure the user only specified a subpage
6866 			 * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
6867 			 */
6868 			if ((page_index->subpage != 0)
6869 			 && (subpage == SMS_SUBPAGE_PAGE_0))
6870 				continue;
6871 
6872 			/*
6873 			 * Call the handler, if it exists, to update the
6874 			 * page to the latest values.
6875 			 */
6876 			if (page_index->sense_handler != NULL)
6877 				page_index->sense_handler(ctsio, page_index,pc);
6878 
6879 			memcpy(ctsio->kern_data_ptr + data_used,
6880 			       page_index->page_data +
6881 			       (page_index->page_len * pc),
6882 			       page_index->page_len);
6883 			data_used += page_index->page_len;
6884 		}
6885 		break;
6886 	}
6887 	default: {
6888 		int i, data_used;
6889 
6890 		data_used = header_len;
6891 
6892 		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6893 			struct ctl_page_index *page_index;
6894 
6895 			page_index = &lun->mode_pages.index[i];
6896 
6897 			/* Look for the right page code */
6898 			if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6899 				continue;
6900 
6901 			/* Look for the right subpage or the subpage wildcard*/
6902 			if ((page_index->subpage != subpage)
6903 			 && (subpage != SMS_SUBPAGE_ALL))
6904 				continue;
6905 
6906 			/* Make sure the page is supported for this dev type */
6907 			if ((control_dev != 0)
6908 			 && (page_index->page_flags &
6909 			     CTL_PAGE_FLAG_DISK_ONLY))
6910 				continue;
6911 
6912 			/*
6913 			 * Call the handler, if it exists, to update the
6914 			 * page to the latest values.
6915 			 */
6916 			if (page_index->sense_handler != NULL)
6917 				page_index->sense_handler(ctsio, page_index,pc);
6918 
6919 			memcpy(ctsio->kern_data_ptr + data_used,
6920 			       page_index->page_data +
6921 			       (page_index->page_len * pc),
6922 			       page_index->page_len);
6923 			data_used += page_index->page_len;
6924 		}
6925 		break;
6926 	}
6927 	}
6928 
6929 	ctl_set_success(ctsio);
6930 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6931 	ctsio->be_move_done = ctl_config_move_done;
6932 	ctl_datamove((union ctl_io *)ctsio);
6933 	return (CTL_RETVAL_COMPLETE);
6934 }
6935 
6936 int
6937 ctl_lbp_log_sense_handler(struct ctl_scsiio *ctsio,
6938 			       struct ctl_page_index *page_index,
6939 			       int pc)
6940 {
6941 	struct ctl_lun *lun;
6942 	struct scsi_log_param_header *phdr;
6943 	uint8_t *data;
6944 	uint64_t val;
6945 
6946 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6947 	data = page_index->page_data;
6948 
6949 	if (lun->backend->lun_attr != NULL &&
6950 	    (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksavail"))
6951 	     != UINT64_MAX) {
6952 		phdr = (struct scsi_log_param_header *)data;
6953 		scsi_ulto2b(0x0001, phdr->param_code);
6954 		phdr->param_control = SLP_LBIN | SLP_LP;
6955 		phdr->param_len = 8;
6956 		data = (uint8_t *)(phdr + 1);
6957 		scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6958 		data[4] = 0x02; /* per-pool */
6959 		data += phdr->param_len;
6960 	}
6961 
6962 	if (lun->backend->lun_attr != NULL &&
6963 	    (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksused"))
6964 	     != UINT64_MAX) {
6965 		phdr = (struct scsi_log_param_header *)data;
6966 		scsi_ulto2b(0x0002, phdr->param_code);
6967 		phdr->param_control = SLP_LBIN | SLP_LP;
6968 		phdr->param_len = 8;
6969 		data = (uint8_t *)(phdr + 1);
6970 		scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6971 		data[4] = 0x01; /* per-LUN */
6972 		data += phdr->param_len;
6973 	}
6974 
6975 	if (lun->backend->lun_attr != NULL &&
6976 	    (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksavail"))
6977 	     != UINT64_MAX) {
6978 		phdr = (struct scsi_log_param_header *)data;
6979 		scsi_ulto2b(0x00f1, phdr->param_code);
6980 		phdr->param_control = SLP_LBIN | SLP_LP;
6981 		phdr->param_len = 8;
6982 		data = (uint8_t *)(phdr + 1);
6983 		scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6984 		data[4] = 0x02; /* per-pool */
6985 		data += phdr->param_len;
6986 	}
6987 
6988 	if (lun->backend->lun_attr != NULL &&
6989 	    (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksused"))
6990 	     != UINT64_MAX) {
6991 		phdr = (struct scsi_log_param_header *)data;
6992 		scsi_ulto2b(0x00f2, phdr->param_code);
6993 		phdr->param_control = SLP_LBIN | SLP_LP;
6994 		phdr->param_len = 8;
6995 		data = (uint8_t *)(phdr + 1);
6996 		scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6997 		data[4] = 0x02; /* per-pool */
6998 		data += phdr->param_len;
6999 	}
7000 
7001 	page_index->page_len = data - page_index->page_data;
7002 	return (0);
7003 }
7004 
7005 int
7006 ctl_log_sense(struct ctl_scsiio *ctsio)
7007 {
7008 	struct ctl_lun *lun;
7009 	int i, pc, page_code, subpage;
7010 	int alloc_len, total_len;
7011 	struct ctl_page_index *page_index;
7012 	struct scsi_log_sense *cdb;
7013 	struct scsi_log_header *header;
7014 
7015 	CTL_DEBUG_PRINT(("ctl_log_sense\n"));
7016 
7017 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7018 	cdb = (struct scsi_log_sense *)ctsio->cdb;
7019 	pc = (cdb->page & SLS_PAGE_CTRL_MASK) >> 6;
7020 	page_code = cdb->page & SLS_PAGE_CODE;
7021 	subpage = cdb->subpage;
7022 	alloc_len = scsi_2btoul(cdb->length);
7023 
7024 	page_index = NULL;
7025 	for (i = 0; i < CTL_NUM_LOG_PAGES; i++) {
7026 		page_index = &lun->log_pages.index[i];
7027 
7028 		/* Look for the right page code */
7029 		if ((page_index->page_code & SL_PAGE_CODE) != page_code)
7030 			continue;
7031 
7032 		/* Look for the right subpage or the subpage wildcard*/
7033 		if (page_index->subpage != subpage)
7034 			continue;
7035 
7036 		break;
7037 	}
7038 	if (i >= CTL_NUM_LOG_PAGES) {
7039 		ctl_set_invalid_field(ctsio,
7040 				      /*sks_valid*/ 1,
7041 				      /*command*/ 1,
7042 				      /*field*/ 2,
7043 				      /*bit_valid*/ 0,
7044 				      /*bit*/ 0);
7045 		ctl_done((union ctl_io *)ctsio);
7046 		return (CTL_RETVAL_COMPLETE);
7047 	}
7048 
7049 	total_len = sizeof(struct scsi_log_header) + page_index->page_len;
7050 
7051 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7052 	ctsio->kern_sg_entries = 0;
7053 	ctsio->kern_data_resid = 0;
7054 	ctsio->kern_rel_offset = 0;
7055 	if (total_len < alloc_len) {
7056 		ctsio->residual = alloc_len - total_len;
7057 		ctsio->kern_data_len = total_len;
7058 		ctsio->kern_total_len = total_len;
7059 	} else {
7060 		ctsio->residual = 0;
7061 		ctsio->kern_data_len = alloc_len;
7062 		ctsio->kern_total_len = alloc_len;
7063 	}
7064 
7065 	header = (struct scsi_log_header *)ctsio->kern_data_ptr;
7066 	header->page = page_index->page_code;
7067 	if (page_index->subpage) {
7068 		header->page |= SL_SPF;
7069 		header->subpage = page_index->subpage;
7070 	}
7071 	scsi_ulto2b(page_index->page_len, header->datalen);
7072 
7073 	/*
7074 	 * Call the handler, if it exists, to update the
7075 	 * page to the latest values.
7076 	 */
7077 	if (page_index->sense_handler != NULL)
7078 		page_index->sense_handler(ctsio, page_index, pc);
7079 
7080 	memcpy(header + 1, page_index->page_data, page_index->page_len);
7081 
7082 	ctl_set_success(ctsio);
7083 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7084 	ctsio->be_move_done = ctl_config_move_done;
7085 	ctl_datamove((union ctl_io *)ctsio);
7086 	return (CTL_RETVAL_COMPLETE);
7087 }
7088 
7089 int
7090 ctl_read_capacity(struct ctl_scsiio *ctsio)
7091 {
7092 	struct scsi_read_capacity *cdb;
7093 	struct scsi_read_capacity_data *data;
7094 	struct ctl_lun *lun;
7095 	uint32_t lba;
7096 
7097 	CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
7098 
7099 	cdb = (struct scsi_read_capacity *)ctsio->cdb;
7100 
7101 	lba = scsi_4btoul(cdb->addr);
7102 	if (((cdb->pmi & SRC_PMI) == 0)
7103 	 && (lba != 0)) {
7104 		ctl_set_invalid_field(/*ctsio*/ ctsio,
7105 				      /*sks_valid*/ 1,
7106 				      /*command*/ 1,
7107 				      /*field*/ 2,
7108 				      /*bit_valid*/ 0,
7109 				      /*bit*/ 0);
7110 		ctl_done((union ctl_io *)ctsio);
7111 		return (CTL_RETVAL_COMPLETE);
7112 	}
7113 
7114 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7115 
7116 	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7117 	data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
7118 	ctsio->residual = 0;
7119 	ctsio->kern_data_len = sizeof(*data);
7120 	ctsio->kern_total_len = sizeof(*data);
7121 	ctsio->kern_data_resid = 0;
7122 	ctsio->kern_rel_offset = 0;
7123 	ctsio->kern_sg_entries = 0;
7124 
7125 	/*
7126 	 * If the maximum LBA is greater than 0xfffffffe, the user must
7127 	 * issue a SERVICE ACTION IN (16) command, with the read capacity
7128 	 * serivce action set.
7129 	 */
7130 	if (lun->be_lun->maxlba > 0xfffffffe)
7131 		scsi_ulto4b(0xffffffff, data->addr);
7132 	else
7133 		scsi_ulto4b(lun->be_lun->maxlba, data->addr);
7134 
7135 	/*
7136 	 * XXX KDM this may not be 512 bytes...
7137 	 */
7138 	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7139 
7140 	ctl_set_success(ctsio);
7141 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7142 	ctsio->be_move_done = ctl_config_move_done;
7143 	ctl_datamove((union ctl_io *)ctsio);
7144 	return (CTL_RETVAL_COMPLETE);
7145 }
7146 
7147 int
7148 ctl_read_capacity_16(struct ctl_scsiio *ctsio)
7149 {
7150 	struct scsi_read_capacity_16 *cdb;
7151 	struct scsi_read_capacity_data_long *data;
7152 	struct ctl_lun *lun;
7153 	uint64_t lba;
7154 	uint32_t alloc_len;
7155 
7156 	CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
7157 
7158 	cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
7159 
7160 	alloc_len = scsi_4btoul(cdb->alloc_len);
7161 	lba = scsi_8btou64(cdb->addr);
7162 
7163 	if ((cdb->reladr & SRC16_PMI)
7164 	 && (lba != 0)) {
7165 		ctl_set_invalid_field(/*ctsio*/ ctsio,
7166 				      /*sks_valid*/ 1,
7167 				      /*command*/ 1,
7168 				      /*field*/ 2,
7169 				      /*bit_valid*/ 0,
7170 				      /*bit*/ 0);
7171 		ctl_done((union ctl_io *)ctsio);
7172 		return (CTL_RETVAL_COMPLETE);
7173 	}
7174 
7175 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7176 
7177 	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7178 	data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
7179 
7180 	if (sizeof(*data) < alloc_len) {
7181 		ctsio->residual = alloc_len - sizeof(*data);
7182 		ctsio->kern_data_len = sizeof(*data);
7183 		ctsio->kern_total_len = sizeof(*data);
7184 	} else {
7185 		ctsio->residual = 0;
7186 		ctsio->kern_data_len = alloc_len;
7187 		ctsio->kern_total_len = alloc_len;
7188 	}
7189 	ctsio->kern_data_resid = 0;
7190 	ctsio->kern_rel_offset = 0;
7191 	ctsio->kern_sg_entries = 0;
7192 
7193 	scsi_u64to8b(lun->be_lun->maxlba, data->addr);
7194 	/* XXX KDM this may not be 512 bytes... */
7195 	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7196 	data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
7197 	scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
7198 	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
7199 		data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ;
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 (CTL_RETVAL_COMPLETE);
7206 }
7207 
7208 int
7209 ctl_get_lba_status(struct ctl_scsiio *ctsio)
7210 {
7211 	struct scsi_get_lba_status *cdb;
7212 	struct scsi_get_lba_status_data *data;
7213 	struct ctl_lun *lun;
7214 	struct ctl_lba_len_flags *lbalen;
7215 	uint64_t lba;
7216 	uint32_t alloc_len, total_len;
7217 	int retval;
7218 
7219 	CTL_DEBUG_PRINT(("ctl_get_lba_status\n"));
7220 
7221 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7222 	cdb = (struct scsi_get_lba_status *)ctsio->cdb;
7223 	lba = scsi_8btou64(cdb->addr);
7224 	alloc_len = scsi_4btoul(cdb->alloc_len);
7225 
7226 	if (lba > lun->be_lun->maxlba) {
7227 		ctl_set_lba_out_of_range(ctsio);
7228 		ctl_done((union ctl_io *)ctsio);
7229 		return (CTL_RETVAL_COMPLETE);
7230 	}
7231 
7232 	total_len = sizeof(*data) + sizeof(data->descr[0]);
7233 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7234 	data = (struct scsi_get_lba_status_data *)ctsio->kern_data_ptr;
7235 
7236 	if (total_len < alloc_len) {
7237 		ctsio->residual = alloc_len - total_len;
7238 		ctsio->kern_data_len = total_len;
7239 		ctsio->kern_total_len = total_len;
7240 	} else {
7241 		ctsio->residual = 0;
7242 		ctsio->kern_data_len = alloc_len;
7243 		ctsio->kern_total_len = alloc_len;
7244 	}
7245 	ctsio->kern_data_resid = 0;
7246 	ctsio->kern_rel_offset = 0;
7247 	ctsio->kern_sg_entries = 0;
7248 
7249 	/* Fill dummy data in case backend can't tell anything. */
7250 	scsi_ulto4b(4 + sizeof(data->descr[0]), data->length);
7251 	scsi_u64to8b(lba, data->descr[0].addr);
7252 	scsi_ulto4b(MIN(UINT32_MAX, lun->be_lun->maxlba + 1 - lba),
7253 	    data->descr[0].length);
7254 	data->descr[0].status = 0; /* Mapped or unknown. */
7255 
7256 	ctl_set_success(ctsio);
7257 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7258 	ctsio->be_move_done = ctl_config_move_done;
7259 
7260 	lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
7261 	lbalen->lba = lba;
7262 	lbalen->len = total_len;
7263 	lbalen->flags = 0;
7264 	retval = lun->backend->config_read((union ctl_io *)ctsio);
7265 	return (CTL_RETVAL_COMPLETE);
7266 }
7267 
7268 int
7269 ctl_read_defect(struct ctl_scsiio *ctsio)
7270 {
7271 	struct scsi_read_defect_data_10 *ccb10;
7272 	struct scsi_read_defect_data_12 *ccb12;
7273 	struct scsi_read_defect_data_hdr_10 *data10;
7274 	struct scsi_read_defect_data_hdr_12 *data12;
7275 	uint32_t alloc_len, data_len;
7276 	uint8_t format;
7277 
7278 	CTL_DEBUG_PRINT(("ctl_read_defect\n"));
7279 
7280 	if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7281 		ccb10 = (struct scsi_read_defect_data_10 *)&ctsio->cdb;
7282 		format = ccb10->format;
7283 		alloc_len = scsi_2btoul(ccb10->alloc_length);
7284 		data_len = sizeof(*data10);
7285 	} else {
7286 		ccb12 = (struct scsi_read_defect_data_12 *)&ctsio->cdb;
7287 		format = ccb12->format;
7288 		alloc_len = scsi_4btoul(ccb12->alloc_length);
7289 		data_len = sizeof(*data12);
7290 	}
7291 	if (alloc_len == 0) {
7292 		ctl_set_success(ctsio);
7293 		ctl_done((union ctl_io *)ctsio);
7294 		return (CTL_RETVAL_COMPLETE);
7295 	}
7296 
7297 	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
7298 	if (data_len < alloc_len) {
7299 		ctsio->residual = alloc_len - data_len;
7300 		ctsio->kern_data_len = data_len;
7301 		ctsio->kern_total_len = data_len;
7302 	} else {
7303 		ctsio->residual = 0;
7304 		ctsio->kern_data_len = alloc_len;
7305 		ctsio->kern_total_len = alloc_len;
7306 	}
7307 	ctsio->kern_data_resid = 0;
7308 	ctsio->kern_rel_offset = 0;
7309 	ctsio->kern_sg_entries = 0;
7310 
7311 	if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7312 		data10 = (struct scsi_read_defect_data_hdr_10 *)
7313 		    ctsio->kern_data_ptr;
7314 		data10->format = format;
7315 		scsi_ulto2b(0, data10->length);
7316 	} else {
7317 		data12 = (struct scsi_read_defect_data_hdr_12 *)
7318 		    ctsio->kern_data_ptr;
7319 		data12->format = format;
7320 		scsi_ulto2b(0, data12->generation);
7321 		scsi_ulto4b(0, data12->length);
7322 	}
7323 
7324 	ctl_set_success(ctsio);
7325 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7326 	ctsio->be_move_done = ctl_config_move_done;
7327 	ctl_datamove((union ctl_io *)ctsio);
7328 	return (CTL_RETVAL_COMPLETE);
7329 }
7330 
7331 int
7332 ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7333 {
7334 	struct scsi_maintenance_in *cdb;
7335 	int retval;
7336 	int alloc_len, ext, total_len = 0, g, p, pc, pg, gs, os;
7337 	int num_target_port_groups, num_target_ports;
7338 	struct ctl_lun *lun;
7339 	struct ctl_softc *softc;
7340 	struct ctl_port *port;
7341 	struct scsi_target_group_data *rtg_ptr;
7342 	struct scsi_target_group_data_extended *rtg_ext_ptr;
7343 	struct scsi_target_port_group_descriptor *tpg_desc;
7344 
7345 	CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7346 
7347 	cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7348 	softc = control_softc;
7349 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7350 
7351 	retval = CTL_RETVAL_COMPLETE;
7352 
7353 	switch (cdb->byte2 & STG_PDF_MASK) {
7354 	case STG_PDF_LENGTH:
7355 		ext = 0;
7356 		break;
7357 	case STG_PDF_EXTENDED:
7358 		ext = 1;
7359 		break;
7360 	default:
7361 		ctl_set_invalid_field(/*ctsio*/ ctsio,
7362 				      /*sks_valid*/ 1,
7363 				      /*command*/ 1,
7364 				      /*field*/ 2,
7365 				      /*bit_valid*/ 1,
7366 				      /*bit*/ 5);
7367 		ctl_done((union ctl_io *)ctsio);
7368 		return(retval);
7369 	}
7370 
7371 	if (softc->is_single)
7372 		num_target_port_groups = 1;
7373 	else
7374 		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
7375 	num_target_ports = 0;
7376 	mtx_lock(&softc->ctl_lock);
7377 	STAILQ_FOREACH(port, &softc->port_list, links) {
7378 		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7379 			continue;
7380 		if (ctl_map_lun_back(port->targ_port, lun->lun) >= CTL_MAX_LUNS)
7381 			continue;
7382 		num_target_ports++;
7383 	}
7384 	mtx_unlock(&softc->ctl_lock);
7385 
7386 	if (ext)
7387 		total_len = sizeof(struct scsi_target_group_data_extended);
7388 	else
7389 		total_len = sizeof(struct scsi_target_group_data);
7390 	total_len += sizeof(struct scsi_target_port_group_descriptor) *
7391 		num_target_port_groups +
7392 	    sizeof(struct scsi_target_port_descriptor) *
7393 		num_target_ports * num_target_port_groups;
7394 
7395 	alloc_len = scsi_4btoul(cdb->length);
7396 
7397 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7398 
7399 	ctsio->kern_sg_entries = 0;
7400 
7401 	if (total_len < alloc_len) {
7402 		ctsio->residual = alloc_len - total_len;
7403 		ctsio->kern_data_len = total_len;
7404 		ctsio->kern_total_len = total_len;
7405 	} else {
7406 		ctsio->residual = 0;
7407 		ctsio->kern_data_len = alloc_len;
7408 		ctsio->kern_total_len = alloc_len;
7409 	}
7410 	ctsio->kern_data_resid = 0;
7411 	ctsio->kern_rel_offset = 0;
7412 
7413 	if (ext) {
7414 		rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7415 		    ctsio->kern_data_ptr;
7416 		scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7417 		rtg_ext_ptr->format_type = 0x10;
7418 		rtg_ext_ptr->implicit_transition_time = 0;
7419 		tpg_desc = &rtg_ext_ptr->groups[0];
7420 	} else {
7421 		rtg_ptr = (struct scsi_target_group_data *)
7422 		    ctsio->kern_data_ptr;
7423 		scsi_ulto4b(total_len - 4, rtg_ptr->length);
7424 		tpg_desc = &rtg_ptr->groups[0];
7425 	}
7426 
7427 	mtx_lock(&softc->ctl_lock);
7428 	pg = softc->port_offset / CTL_MAX_PORTS;
7429 	if (softc->flags & CTL_FLAG_ACTIVE_SHELF) {
7430 		if (softc->ha_mode == CTL_HA_MODE_ACT_STBY) {
7431 			gs = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7432 			os = TPG_ASYMMETRIC_ACCESS_STANDBY;
7433 		} else if (lun->flags & CTL_LUN_PRIMARY_SC) {
7434 			gs = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7435 			os = TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7436 		} else {
7437 			gs = TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7438 			os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7439 		}
7440 	} else {
7441 		gs = TPG_ASYMMETRIC_ACCESS_STANDBY;
7442 		os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7443 	}
7444 	for (g = 0; g < num_target_port_groups; g++) {
7445 		tpg_desc->pref_state = (g == pg) ? gs : os;
7446 		tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP;
7447 		scsi_ulto2b(g + 1, tpg_desc->target_port_group);
7448 		tpg_desc->status = TPG_IMPLICIT;
7449 		pc = 0;
7450 		STAILQ_FOREACH(port, &softc->port_list, links) {
7451 			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7452 				continue;
7453 			if (ctl_map_lun_back(port->targ_port, lun->lun) >=
7454 			    CTL_MAX_LUNS)
7455 				continue;
7456 			p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
7457 			scsi_ulto2b(p, tpg_desc->descriptors[pc].
7458 			    relative_target_port_identifier);
7459 			pc++;
7460 		}
7461 		tpg_desc->target_port_count = pc;
7462 		tpg_desc = (struct scsi_target_port_group_descriptor *)
7463 		    &tpg_desc->descriptors[pc];
7464 	}
7465 	mtx_unlock(&softc->ctl_lock);
7466 
7467 	ctl_set_success(ctsio);
7468 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7469 	ctsio->be_move_done = ctl_config_move_done;
7470 	ctl_datamove((union ctl_io *)ctsio);
7471 	return(retval);
7472 }
7473 
7474 int
7475 ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7476 {
7477 	struct ctl_lun *lun;
7478 	struct scsi_report_supported_opcodes *cdb;
7479 	const struct ctl_cmd_entry *entry, *sentry;
7480 	struct scsi_report_supported_opcodes_all *all;
7481 	struct scsi_report_supported_opcodes_descr *descr;
7482 	struct scsi_report_supported_opcodes_one *one;
7483 	int retval;
7484 	int alloc_len, total_len;
7485 	int opcode, service_action, i, j, num;
7486 
7487 	CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7488 
7489 	cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7490 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7491 
7492 	retval = CTL_RETVAL_COMPLETE;
7493 
7494 	opcode = cdb->requested_opcode;
7495 	service_action = scsi_2btoul(cdb->requested_service_action);
7496 	switch (cdb->options & RSO_OPTIONS_MASK) {
7497 	case RSO_OPTIONS_ALL:
7498 		num = 0;
7499 		for (i = 0; i < 256; i++) {
7500 			entry = &ctl_cmd_table[i];
7501 			if (entry->flags & CTL_CMD_FLAG_SA5) {
7502 				for (j = 0; j < 32; j++) {
7503 					sentry = &((const struct ctl_cmd_entry *)
7504 					    entry->execute)[j];
7505 					if (ctl_cmd_applicable(
7506 					    lun->be_lun->lun_type, sentry))
7507 						num++;
7508 				}
7509 			} else {
7510 				if (ctl_cmd_applicable(lun->be_lun->lun_type,
7511 				    entry))
7512 					num++;
7513 			}
7514 		}
7515 		total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7516 		    num * sizeof(struct scsi_report_supported_opcodes_descr);
7517 		break;
7518 	case RSO_OPTIONS_OC:
7519 		if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7520 			ctl_set_invalid_field(/*ctsio*/ ctsio,
7521 					      /*sks_valid*/ 1,
7522 					      /*command*/ 1,
7523 					      /*field*/ 2,
7524 					      /*bit_valid*/ 1,
7525 					      /*bit*/ 2);
7526 			ctl_done((union ctl_io *)ctsio);
7527 			return (CTL_RETVAL_COMPLETE);
7528 		}
7529 		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7530 		break;
7531 	case RSO_OPTIONS_OC_SA:
7532 		if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7533 		    service_action >= 32) {
7534 			ctl_set_invalid_field(/*ctsio*/ ctsio,
7535 					      /*sks_valid*/ 1,
7536 					      /*command*/ 1,
7537 					      /*field*/ 2,
7538 					      /*bit_valid*/ 1,
7539 					      /*bit*/ 2);
7540 			ctl_done((union ctl_io *)ctsio);
7541 			return (CTL_RETVAL_COMPLETE);
7542 		}
7543 		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7544 		break;
7545 	default:
7546 		ctl_set_invalid_field(/*ctsio*/ ctsio,
7547 				      /*sks_valid*/ 1,
7548 				      /*command*/ 1,
7549 				      /*field*/ 2,
7550 				      /*bit_valid*/ 1,
7551 				      /*bit*/ 2);
7552 		ctl_done((union ctl_io *)ctsio);
7553 		return (CTL_RETVAL_COMPLETE);
7554 	}
7555 
7556 	alloc_len = scsi_4btoul(cdb->length);
7557 
7558 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7559 
7560 	ctsio->kern_sg_entries = 0;
7561 
7562 	if (total_len < alloc_len) {
7563 		ctsio->residual = alloc_len - total_len;
7564 		ctsio->kern_data_len = total_len;
7565 		ctsio->kern_total_len = total_len;
7566 	} else {
7567 		ctsio->residual = 0;
7568 		ctsio->kern_data_len = alloc_len;
7569 		ctsio->kern_total_len = alloc_len;
7570 	}
7571 	ctsio->kern_data_resid = 0;
7572 	ctsio->kern_rel_offset = 0;
7573 
7574 	switch (cdb->options & RSO_OPTIONS_MASK) {
7575 	case RSO_OPTIONS_ALL:
7576 		all = (struct scsi_report_supported_opcodes_all *)
7577 		    ctsio->kern_data_ptr;
7578 		num = 0;
7579 		for (i = 0; i < 256; i++) {
7580 			entry = &ctl_cmd_table[i];
7581 			if (entry->flags & CTL_CMD_FLAG_SA5) {
7582 				for (j = 0; j < 32; j++) {
7583 					sentry = &((const struct ctl_cmd_entry *)
7584 					    entry->execute)[j];
7585 					if (!ctl_cmd_applicable(
7586 					    lun->be_lun->lun_type, sentry))
7587 						continue;
7588 					descr = &all->descr[num++];
7589 					descr->opcode = i;
7590 					scsi_ulto2b(j, descr->service_action);
7591 					descr->flags = RSO_SERVACTV;
7592 					scsi_ulto2b(sentry->length,
7593 					    descr->cdb_length);
7594 				}
7595 			} else {
7596 				if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7597 				    entry))
7598 					continue;
7599 				descr = &all->descr[num++];
7600 				descr->opcode = i;
7601 				scsi_ulto2b(0, descr->service_action);
7602 				descr->flags = 0;
7603 				scsi_ulto2b(entry->length, descr->cdb_length);
7604 			}
7605 		}
7606 		scsi_ulto4b(
7607 		    num * sizeof(struct scsi_report_supported_opcodes_descr),
7608 		    all->length);
7609 		break;
7610 	case RSO_OPTIONS_OC:
7611 		one = (struct scsi_report_supported_opcodes_one *)
7612 		    ctsio->kern_data_ptr;
7613 		entry = &ctl_cmd_table[opcode];
7614 		goto fill_one;
7615 	case RSO_OPTIONS_OC_SA:
7616 		one = (struct scsi_report_supported_opcodes_one *)
7617 		    ctsio->kern_data_ptr;
7618 		entry = &ctl_cmd_table[opcode];
7619 		entry = &((const struct ctl_cmd_entry *)
7620 		    entry->execute)[service_action];
7621 fill_one:
7622 		if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7623 			one->support = 3;
7624 			scsi_ulto2b(entry->length, one->cdb_length);
7625 			one->cdb_usage[0] = opcode;
7626 			memcpy(&one->cdb_usage[1], entry->usage,
7627 			    entry->length - 1);
7628 		} else
7629 			one->support = 1;
7630 		break;
7631 	}
7632 
7633 	ctl_set_success(ctsio);
7634 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7635 	ctsio->be_move_done = ctl_config_move_done;
7636 	ctl_datamove((union ctl_io *)ctsio);
7637 	return(retval);
7638 }
7639 
7640 int
7641 ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7642 {
7643 	struct scsi_report_supported_tmf *cdb;
7644 	struct scsi_report_supported_tmf_data *data;
7645 	int retval;
7646 	int alloc_len, total_len;
7647 
7648 	CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7649 
7650 	cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7651 
7652 	retval = CTL_RETVAL_COMPLETE;
7653 
7654 	total_len = sizeof(struct scsi_report_supported_tmf_data);
7655 	alloc_len = scsi_4btoul(cdb->length);
7656 
7657 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7658 
7659 	ctsio->kern_sg_entries = 0;
7660 
7661 	if (total_len < alloc_len) {
7662 		ctsio->residual = alloc_len - total_len;
7663 		ctsio->kern_data_len = total_len;
7664 		ctsio->kern_total_len = total_len;
7665 	} else {
7666 		ctsio->residual = 0;
7667 		ctsio->kern_data_len = alloc_len;
7668 		ctsio->kern_total_len = alloc_len;
7669 	}
7670 	ctsio->kern_data_resid = 0;
7671 	ctsio->kern_rel_offset = 0;
7672 
7673 	data = (struct scsi_report_supported_tmf_data *)ctsio->kern_data_ptr;
7674 	data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_TRS;
7675 	data->byte2 |= RST_ITNRS;
7676 
7677 	ctl_set_success(ctsio);
7678 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7679 	ctsio->be_move_done = ctl_config_move_done;
7680 	ctl_datamove((union ctl_io *)ctsio);
7681 	return (retval);
7682 }
7683 
7684 int
7685 ctl_report_timestamp(struct ctl_scsiio *ctsio)
7686 {
7687 	struct scsi_report_timestamp *cdb;
7688 	struct scsi_report_timestamp_data *data;
7689 	struct timeval tv;
7690 	int64_t timestamp;
7691 	int retval;
7692 	int alloc_len, total_len;
7693 
7694 	CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7695 
7696 	cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7697 
7698 	retval = CTL_RETVAL_COMPLETE;
7699 
7700 	total_len = sizeof(struct scsi_report_timestamp_data);
7701 	alloc_len = scsi_4btoul(cdb->length);
7702 
7703 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7704 
7705 	ctsio->kern_sg_entries = 0;
7706 
7707 	if (total_len < alloc_len) {
7708 		ctsio->residual = alloc_len - total_len;
7709 		ctsio->kern_data_len = total_len;
7710 		ctsio->kern_total_len = total_len;
7711 	} else {
7712 		ctsio->residual = 0;
7713 		ctsio->kern_data_len = alloc_len;
7714 		ctsio->kern_total_len = alloc_len;
7715 	}
7716 	ctsio->kern_data_resid = 0;
7717 	ctsio->kern_rel_offset = 0;
7718 
7719 	data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7720 	scsi_ulto2b(sizeof(*data) - 2, data->length);
7721 	data->origin = RTS_ORIG_OUTSIDE;
7722 	getmicrotime(&tv);
7723 	timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7724 	scsi_ulto4b(timestamp >> 16, data->timestamp);
7725 	scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7726 
7727 	ctl_set_success(ctsio);
7728 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7729 	ctsio->be_move_done = ctl_config_move_done;
7730 	ctl_datamove((union ctl_io *)ctsio);
7731 	return (retval);
7732 }
7733 
7734 int
7735 ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7736 {
7737 	struct scsi_per_res_in *cdb;
7738 	int alloc_len, total_len = 0;
7739 	/* struct scsi_per_res_in_rsrv in_data; */
7740 	struct ctl_lun *lun;
7741 	struct ctl_softc *softc;
7742 	uint64_t key;
7743 
7744 	CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7745 
7746 	softc = control_softc;
7747 
7748 	cdb = (struct scsi_per_res_in *)ctsio->cdb;
7749 
7750 	alloc_len = scsi_2btoul(cdb->length);
7751 
7752 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7753 
7754 retry:
7755 	mtx_lock(&lun->lun_lock);
7756 	switch (cdb->action) {
7757 	case SPRI_RK: /* read keys */
7758 		total_len = sizeof(struct scsi_per_res_in_keys) +
7759 			lun->pr_key_count *
7760 			sizeof(struct scsi_per_res_key);
7761 		break;
7762 	case SPRI_RR: /* read reservation */
7763 		if (lun->flags & CTL_LUN_PR_RESERVED)
7764 			total_len = sizeof(struct scsi_per_res_in_rsrv);
7765 		else
7766 			total_len = sizeof(struct scsi_per_res_in_header);
7767 		break;
7768 	case SPRI_RC: /* report capabilities */
7769 		total_len = sizeof(struct scsi_per_res_cap);
7770 		break;
7771 	case SPRI_RS: /* read full status */
7772 		total_len = sizeof(struct scsi_per_res_in_header) +
7773 		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7774 		    lun->pr_key_count;
7775 		break;
7776 	default:
7777 		panic("Invalid PR type %x", cdb->action);
7778 	}
7779 	mtx_unlock(&lun->lun_lock);
7780 
7781 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7782 
7783 	if (total_len < alloc_len) {
7784 		ctsio->residual = alloc_len - total_len;
7785 		ctsio->kern_data_len = total_len;
7786 		ctsio->kern_total_len = total_len;
7787 	} else {
7788 		ctsio->residual = 0;
7789 		ctsio->kern_data_len = alloc_len;
7790 		ctsio->kern_total_len = alloc_len;
7791 	}
7792 
7793 	ctsio->kern_data_resid = 0;
7794 	ctsio->kern_rel_offset = 0;
7795 	ctsio->kern_sg_entries = 0;
7796 
7797 	mtx_lock(&lun->lun_lock);
7798 	switch (cdb->action) {
7799 	case SPRI_RK: { // read keys
7800         struct scsi_per_res_in_keys *res_keys;
7801 		int i, key_count;
7802 
7803 		res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7804 
7805 		/*
7806 		 * We had to drop the lock to allocate our buffer, which
7807 		 * leaves time for someone to come in with another
7808 		 * persistent reservation.  (That is unlikely, though,
7809 		 * since this should be the only persistent reservation
7810 		 * command active right now.)
7811 		 */
7812 		if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7813 		    (lun->pr_key_count *
7814 		     sizeof(struct scsi_per_res_key)))){
7815 			mtx_unlock(&lun->lun_lock);
7816 			free(ctsio->kern_data_ptr, M_CTL);
7817 			printf("%s: reservation length changed, retrying\n",
7818 			       __func__);
7819 			goto retry;
7820 		}
7821 
7822 		scsi_ulto4b(lun->PRGeneration, res_keys->header.generation);
7823 
7824 		scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7825 			     lun->pr_key_count, res_keys->header.length);
7826 
7827 		for (i = 0, key_count = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7828 			if ((key = ctl_get_prkey(lun, i)) == 0)
7829 				continue;
7830 
7831 			/*
7832 			 * We used lun->pr_key_count to calculate the
7833 			 * size to allocate.  If it turns out the number of
7834 			 * initiators with the registered flag set is
7835 			 * larger than that (i.e. they haven't been kept in
7836 			 * sync), we've got a problem.
7837 			 */
7838 			if (key_count >= lun->pr_key_count) {
7839 #ifdef NEEDTOPORT
7840 				csevent_log(CSC_CTL | CSC_SHELF_SW |
7841 					    CTL_PR_ERROR,
7842 					    csevent_LogType_Fault,
7843 					    csevent_AlertLevel_Yellow,
7844 					    csevent_FRU_ShelfController,
7845 					    csevent_FRU_Firmware,
7846 				        csevent_FRU_Unknown,
7847 					    "registered keys %d >= key "
7848 					    "count %d", key_count,
7849 					    lun->pr_key_count);
7850 #endif
7851 				key_count++;
7852 				continue;
7853 			}
7854 			scsi_u64to8b(key, res_keys->keys[key_count].key);
7855 			key_count++;
7856 		}
7857 		break;
7858 	}
7859 	case SPRI_RR: { // read reservation
7860 		struct scsi_per_res_in_rsrv *res;
7861 		int tmp_len, header_only;
7862 
7863 		res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7864 
7865 		scsi_ulto4b(lun->PRGeneration, res->header.generation);
7866 
7867 		if (lun->flags & CTL_LUN_PR_RESERVED)
7868 		{
7869 			tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7870 			scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7871 				    res->header.length);
7872 			header_only = 0;
7873 		} else {
7874 			tmp_len = sizeof(struct scsi_per_res_in_header);
7875 			scsi_ulto4b(0, res->header.length);
7876 			header_only = 1;
7877 		}
7878 
7879 		/*
7880 		 * We had to drop the lock to allocate our buffer, which
7881 		 * leaves time for someone to come in with another
7882 		 * persistent reservation.  (That is unlikely, though,
7883 		 * since this should be the only persistent reservation
7884 		 * command active right now.)
7885 		 */
7886 		if (tmp_len != total_len) {
7887 			mtx_unlock(&lun->lun_lock);
7888 			free(ctsio->kern_data_ptr, M_CTL);
7889 			printf("%s: reservation status changed, retrying\n",
7890 			       __func__);
7891 			goto retry;
7892 		}
7893 
7894 		/*
7895 		 * No reservation held, so we're done.
7896 		 */
7897 		if (header_only != 0)
7898 			break;
7899 
7900 		/*
7901 		 * If the registration is an All Registrants type, the key
7902 		 * is 0, since it doesn't really matter.
7903 		 */
7904 		if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7905 			scsi_u64to8b(ctl_get_prkey(lun, lun->pr_res_idx),
7906 			    res->data.reservation);
7907 		}
7908 		res->data.scopetype = lun->res_type;
7909 		break;
7910 	}
7911 	case SPRI_RC:     //report capabilities
7912 	{
7913 		struct scsi_per_res_cap *res_cap;
7914 		uint16_t type_mask;
7915 
7916 		res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7917 		scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7918 		res_cap->flags2 |= SPRI_TMV | SPRI_ALLOW_5;
7919 		type_mask = SPRI_TM_WR_EX_AR |
7920 			    SPRI_TM_EX_AC_RO |
7921 			    SPRI_TM_WR_EX_RO |
7922 			    SPRI_TM_EX_AC |
7923 			    SPRI_TM_WR_EX |
7924 			    SPRI_TM_EX_AC_AR;
7925 		scsi_ulto2b(type_mask, res_cap->type_mask);
7926 		break;
7927 	}
7928 	case SPRI_RS: { // read full status
7929 		struct scsi_per_res_in_full *res_status;
7930 		struct scsi_per_res_in_full_desc *res_desc;
7931 		struct ctl_port *port;
7932 		int i, len;
7933 
7934 		res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7935 
7936 		/*
7937 		 * We had to drop the lock to allocate our buffer, which
7938 		 * leaves time for someone to come in with another
7939 		 * persistent reservation.  (That is unlikely, though,
7940 		 * since this should be the only persistent reservation
7941 		 * command active right now.)
7942 		 */
7943 		if (total_len < (sizeof(struct scsi_per_res_in_header) +
7944 		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7945 		     lun->pr_key_count)){
7946 			mtx_unlock(&lun->lun_lock);
7947 			free(ctsio->kern_data_ptr, M_CTL);
7948 			printf("%s: reservation length changed, retrying\n",
7949 			       __func__);
7950 			goto retry;
7951 		}
7952 
7953 		scsi_ulto4b(lun->PRGeneration, res_status->header.generation);
7954 
7955 		res_desc = &res_status->desc[0];
7956 		for (i = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7957 			if ((key = ctl_get_prkey(lun, i)) == 0)
7958 				continue;
7959 
7960 			scsi_u64to8b(key, res_desc->res_key.key);
7961 			if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7962 			    (lun->pr_res_idx == i ||
7963 			     lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7964 				res_desc->flags = SPRI_FULL_R_HOLDER;
7965 				res_desc->scopetype = lun->res_type;
7966 			}
7967 			scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7968 			    res_desc->rel_trgt_port_id);
7969 			len = 0;
7970 			port = softc->ctl_ports[
7971 			    ctl_port_idx(i / CTL_MAX_INIT_PER_PORT)];
7972 			if (port != NULL)
7973 				len = ctl_create_iid(port,
7974 				    i % CTL_MAX_INIT_PER_PORT,
7975 				    res_desc->transport_id);
7976 			scsi_ulto4b(len, res_desc->additional_length);
7977 			res_desc = (struct scsi_per_res_in_full_desc *)
7978 			    &res_desc->transport_id[len];
7979 		}
7980 		scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7981 		    res_status->header.length);
7982 		break;
7983 	}
7984 	default:
7985 		/*
7986 		 * This is a bug, because we just checked for this above,
7987 		 * and should have returned an error.
7988 		 */
7989 		panic("Invalid PR type %x", cdb->action);
7990 		break; /* NOTREACHED */
7991 	}
7992 	mtx_unlock(&lun->lun_lock);
7993 
7994 	ctl_set_success(ctsio);
7995 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7996 	ctsio->be_move_done = ctl_config_move_done;
7997 	ctl_datamove((union ctl_io *)ctsio);
7998 	return (CTL_RETVAL_COMPLETE);
7999 }
8000 
8001 static void
8002 ctl_est_res_ua(struct ctl_lun *lun, uint32_t residx, ctl_ua_type ua)
8003 {
8004 	int off = lun->ctl_softc->persis_offset;
8005 
8006 	if (residx >= off && residx < off + CTL_MAX_INITIATORS)
8007 		ctl_est_ua(lun, residx - off, ua);
8008 }
8009 
8010 /*
8011  * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
8012  * it should return.
8013  */
8014 static int
8015 ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
8016 		uint64_t sa_res_key, uint8_t type, uint32_t residx,
8017 		struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
8018 		struct scsi_per_res_out_parms* param)
8019 {
8020 	union ctl_ha_msg persis_io;
8021 	int retval, i;
8022 	int isc_retval;
8023 
8024 	retval = 0;
8025 
8026 	mtx_lock(&lun->lun_lock);
8027 	if (sa_res_key == 0) {
8028 		if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8029 			/* validate scope and type */
8030 			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
8031 			     SPR_LU_SCOPE) {
8032 				mtx_unlock(&lun->lun_lock);
8033 				ctl_set_invalid_field(/*ctsio*/ ctsio,
8034 						      /*sks_valid*/ 1,
8035 						      /*command*/ 1,
8036 						      /*field*/ 2,
8037 						      /*bit_valid*/ 1,
8038 						      /*bit*/ 4);
8039 				ctl_done((union ctl_io *)ctsio);
8040 				return (1);
8041 			}
8042 
8043 		        if (type>8 || type==2 || type==4 || type==0) {
8044 				mtx_unlock(&lun->lun_lock);
8045 				ctl_set_invalid_field(/*ctsio*/ ctsio,
8046        	           				      /*sks_valid*/ 1,
8047 						      /*command*/ 1,
8048 						      /*field*/ 2,
8049 						      /*bit_valid*/ 1,
8050 						      /*bit*/ 0);
8051 				ctl_done((union ctl_io *)ctsio);
8052 				return (1);
8053 		        }
8054 
8055 			/*
8056 			 * Unregister everybody else and build UA for
8057 			 * them
8058 			 */
8059 			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8060 				if (i == residx || ctl_get_prkey(lun, i) == 0)
8061 					continue;
8062 
8063 				ctl_clr_prkey(lun, i);
8064 				ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
8065 			}
8066 			lun->pr_key_count = 1;
8067 			lun->res_type = type;
8068 			if (lun->res_type != SPR_TYPE_WR_EX_AR
8069 			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8070 				lun->pr_res_idx = residx;
8071 
8072 			/* send msg to other side */
8073 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8074 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8075 			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8076 			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8077 			persis_io.pr.pr_info.res_type = type;
8078 			memcpy(persis_io.pr.pr_info.sa_res_key,
8079 			       param->serv_act_res_key,
8080 			       sizeof(param->serv_act_res_key));
8081 			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8082 			     &persis_io, sizeof(persis_io), 0)) >
8083 			     CTL_HA_STATUS_SUCCESS) {
8084 				printf("CTL:Persis Out error returned "
8085 				       "from ctl_ha_msg_send %d\n",
8086 				       isc_retval);
8087 			}
8088 		} else {
8089 			/* not all registrants */
8090 			mtx_unlock(&lun->lun_lock);
8091 			free(ctsio->kern_data_ptr, M_CTL);
8092 			ctl_set_invalid_field(ctsio,
8093 					      /*sks_valid*/ 1,
8094 					      /*command*/ 0,
8095 					      /*field*/ 8,
8096 					      /*bit_valid*/ 0,
8097 					      /*bit*/ 0);
8098 			ctl_done((union ctl_io *)ctsio);
8099 			return (1);
8100 		}
8101 	} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8102 		|| !(lun->flags & CTL_LUN_PR_RESERVED)) {
8103 		int found = 0;
8104 
8105 		if (res_key == sa_res_key) {
8106 			/* special case */
8107 			/*
8108 			 * The spec implies this is not good but doesn't
8109 			 * say what to do. There are two choices either
8110 			 * generate a res conflict or check condition
8111 			 * with illegal field in parameter data. Since
8112 			 * that is what is done when the sa_res_key is
8113 			 * zero I'll take that approach since this has
8114 			 * to do with the sa_res_key.
8115 			 */
8116 			mtx_unlock(&lun->lun_lock);
8117 			free(ctsio->kern_data_ptr, M_CTL);
8118 			ctl_set_invalid_field(ctsio,
8119 					      /*sks_valid*/ 1,
8120 					      /*command*/ 0,
8121 					      /*field*/ 8,
8122 					      /*bit_valid*/ 0,
8123 					      /*bit*/ 0);
8124 			ctl_done((union ctl_io *)ctsio);
8125 			return (1);
8126 		}
8127 
8128 		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8129 			if (ctl_get_prkey(lun, i) != sa_res_key)
8130 				continue;
8131 
8132 			found = 1;
8133 			ctl_clr_prkey(lun, i);
8134 			lun->pr_key_count--;
8135 			ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
8136 		}
8137 		if (!found) {
8138 			mtx_unlock(&lun->lun_lock);
8139 			free(ctsio->kern_data_ptr, M_CTL);
8140 			ctl_set_reservation_conflict(ctsio);
8141 			ctl_done((union ctl_io *)ctsio);
8142 			return (CTL_RETVAL_COMPLETE);
8143 		}
8144 		/* send msg to other side */
8145 		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8146 		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8147 		persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8148 		persis_io.pr.pr_info.residx = lun->pr_res_idx;
8149 		persis_io.pr.pr_info.res_type = type;
8150 		memcpy(persis_io.pr.pr_info.sa_res_key,
8151 		       param->serv_act_res_key,
8152 		       sizeof(param->serv_act_res_key));
8153 		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8154 		     &persis_io, sizeof(persis_io), 0)) >
8155 		     CTL_HA_STATUS_SUCCESS) {
8156 			printf("CTL:Persis Out error returned from "
8157 			       "ctl_ha_msg_send %d\n", isc_retval);
8158 		}
8159 	} else {
8160 		/* Reserved but not all registrants */
8161 		/* sa_res_key is res holder */
8162 		if (sa_res_key == ctl_get_prkey(lun, lun->pr_res_idx)) {
8163 			/* validate scope and type */
8164 			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
8165 			     SPR_LU_SCOPE) {
8166 				mtx_unlock(&lun->lun_lock);
8167 				ctl_set_invalid_field(/*ctsio*/ ctsio,
8168 						      /*sks_valid*/ 1,
8169 						      /*command*/ 1,
8170 						      /*field*/ 2,
8171 						      /*bit_valid*/ 1,
8172 						      /*bit*/ 4);
8173 				ctl_done((union ctl_io *)ctsio);
8174 				return (1);
8175 			}
8176 
8177 			if (type>8 || type==2 || type==4 || type==0) {
8178 				mtx_unlock(&lun->lun_lock);
8179 				ctl_set_invalid_field(/*ctsio*/ ctsio,
8180 						      /*sks_valid*/ 1,
8181 						      /*command*/ 1,
8182 						      /*field*/ 2,
8183 						      /*bit_valid*/ 1,
8184 						      /*bit*/ 0);
8185 				ctl_done((union ctl_io *)ctsio);
8186 				return (1);
8187 			}
8188 
8189 			/*
8190 			 * Do the following:
8191 			 * if sa_res_key != res_key remove all
8192 			 * registrants w/sa_res_key and generate UA
8193 			 * for these registrants(Registrations
8194 			 * Preempted) if it wasn't an exclusive
8195 			 * reservation generate UA(Reservations
8196 			 * Preempted) for all other registered nexuses
8197 			 * if the type has changed. Establish the new
8198 			 * reservation and holder. If res_key and
8199 			 * sa_res_key are the same do the above
8200 			 * except don't unregister the res holder.
8201 			 */
8202 
8203 			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8204 				if (i == residx || ctl_get_prkey(lun, i) == 0)
8205 					continue;
8206 
8207 				if (sa_res_key == ctl_get_prkey(lun, i)) {
8208 					ctl_clr_prkey(lun, i);
8209 					lun->pr_key_count--;
8210 					ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
8211 				} else if (type != lun->res_type
8212 					&& (lun->res_type == SPR_TYPE_WR_EX_RO
8213 					 || lun->res_type ==SPR_TYPE_EX_AC_RO)){
8214 					ctl_est_res_ua(lun, i, CTL_UA_RES_RELEASE);
8215 				}
8216 			}
8217 			lun->res_type = type;
8218 			if (lun->res_type != SPR_TYPE_WR_EX_AR
8219 			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8220 				lun->pr_res_idx = residx;
8221 			else
8222 				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8223 
8224 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8225 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8226 			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8227 			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8228 			persis_io.pr.pr_info.res_type = type;
8229 			memcpy(persis_io.pr.pr_info.sa_res_key,
8230 			       param->serv_act_res_key,
8231 			       sizeof(param->serv_act_res_key));
8232 			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8233 			     &persis_io, sizeof(persis_io), 0)) >
8234 			     CTL_HA_STATUS_SUCCESS) {
8235 				printf("CTL:Persis Out error returned "
8236 				       "from ctl_ha_msg_send %d\n",
8237 				       isc_retval);
8238 			}
8239 		} else {
8240 			/*
8241 			 * sa_res_key is not the res holder just
8242 			 * remove registrants
8243 			 */
8244 			int found=0;
8245 
8246 			for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8247 				if (sa_res_key != ctl_get_prkey(lun, i))
8248 					continue;
8249 
8250 				found = 1;
8251 				ctl_clr_prkey(lun, i);
8252 				lun->pr_key_count--;
8253 				ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
8254 			}
8255 
8256 			if (!found) {
8257 				mtx_unlock(&lun->lun_lock);
8258 				free(ctsio->kern_data_ptr, M_CTL);
8259 				ctl_set_reservation_conflict(ctsio);
8260 				ctl_done((union ctl_io *)ctsio);
8261 		        	return (1);
8262 			}
8263 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8264 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8265 			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8266 			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8267 			persis_io.pr.pr_info.res_type = type;
8268 			memcpy(persis_io.pr.pr_info.sa_res_key,
8269 			       param->serv_act_res_key,
8270 			       sizeof(param->serv_act_res_key));
8271 			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8272 			     &persis_io, sizeof(persis_io), 0)) >
8273 			     CTL_HA_STATUS_SUCCESS) {
8274 				printf("CTL:Persis Out error returned "
8275 				       "from ctl_ha_msg_send %d\n",
8276 				isc_retval);
8277 			}
8278 		}
8279 	}
8280 
8281 	lun->PRGeneration++;
8282 	mtx_unlock(&lun->lun_lock);
8283 
8284 	return (retval);
8285 }
8286 
8287 static void
8288 ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
8289 {
8290 	uint64_t sa_res_key;
8291 	int i;
8292 
8293 	sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
8294 
8295 	if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8296 	 || lun->pr_res_idx == CTL_PR_NO_RESERVATION
8297 	 || sa_res_key != ctl_get_prkey(lun, lun->pr_res_idx)) {
8298 		if (sa_res_key == 0) {
8299 			/*
8300 			 * Unregister everybody else and build UA for
8301 			 * them
8302 			 */
8303 			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8304 				if (i == msg->pr.pr_info.residx ||
8305 				    ctl_get_prkey(lun, i) == 0)
8306 					continue;
8307 
8308 				ctl_clr_prkey(lun, i);
8309 				ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
8310 			}
8311 
8312 			lun->pr_key_count = 1;
8313 			lun->res_type = msg->pr.pr_info.res_type;
8314 			if (lun->res_type != SPR_TYPE_WR_EX_AR
8315 			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8316 				lun->pr_res_idx = msg->pr.pr_info.residx;
8317 		} else {
8318 		        for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8319 				if (sa_res_key == ctl_get_prkey(lun, i))
8320 					continue;
8321 
8322 				ctl_clr_prkey(lun, i);
8323 				lun->pr_key_count--;
8324 				ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
8325 			}
8326 		}
8327 	} else {
8328 		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8329 			if (i == msg->pr.pr_info.residx ||
8330 			    ctl_get_prkey(lun, i) == 0)
8331 				continue;
8332 
8333 			if (sa_res_key == ctl_get_prkey(lun, i)) {
8334 				ctl_clr_prkey(lun, i);
8335 				lun->pr_key_count--;
8336 				ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
8337 			} else if (msg->pr.pr_info.res_type != lun->res_type
8338 				&& (lun->res_type == SPR_TYPE_WR_EX_RO
8339 				 || lun->res_type == SPR_TYPE_EX_AC_RO)) {
8340 				ctl_est_res_ua(lun, i, CTL_UA_RES_RELEASE);
8341 			}
8342 		}
8343 		lun->res_type = msg->pr.pr_info.res_type;
8344 		if (lun->res_type != SPR_TYPE_WR_EX_AR
8345 		 && lun->res_type != SPR_TYPE_EX_AC_AR)
8346 			lun->pr_res_idx = msg->pr.pr_info.residx;
8347 		else
8348 			lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8349 	}
8350 	lun->PRGeneration++;
8351 
8352 }
8353 
8354 
8355 int
8356 ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8357 {
8358 	int retval;
8359 	int isc_retval;
8360 	u_int32_t param_len;
8361 	struct scsi_per_res_out *cdb;
8362 	struct ctl_lun *lun;
8363 	struct scsi_per_res_out_parms* param;
8364 	struct ctl_softc *softc;
8365 	uint32_t residx;
8366 	uint64_t res_key, sa_res_key, key;
8367 	uint8_t type;
8368 	union ctl_ha_msg persis_io;
8369 	int    i;
8370 
8371 	CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8372 
8373 	retval = CTL_RETVAL_COMPLETE;
8374 
8375 	softc = control_softc;
8376 
8377 	cdb = (struct scsi_per_res_out *)ctsio->cdb;
8378 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8379 
8380 	/*
8381 	 * We only support whole-LUN scope.  The scope & type are ignored for
8382 	 * register, register and ignore existing key and clear.
8383 	 * We sometimes ignore scope and type on preempts too!!
8384 	 * Verify reservation type here as well.
8385 	 */
8386 	type = cdb->scope_type & SPR_TYPE_MASK;
8387 	if ((cdb->action == SPRO_RESERVE)
8388 	 || (cdb->action == SPRO_RELEASE)) {
8389 		if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8390 			ctl_set_invalid_field(/*ctsio*/ ctsio,
8391 					      /*sks_valid*/ 1,
8392 					      /*command*/ 1,
8393 					      /*field*/ 2,
8394 					      /*bit_valid*/ 1,
8395 					      /*bit*/ 4);
8396 			ctl_done((union ctl_io *)ctsio);
8397 			return (CTL_RETVAL_COMPLETE);
8398 		}
8399 
8400 		if (type>8 || type==2 || type==4 || type==0) {
8401 			ctl_set_invalid_field(/*ctsio*/ ctsio,
8402 					      /*sks_valid*/ 1,
8403 					      /*command*/ 1,
8404 					      /*field*/ 2,
8405 					      /*bit_valid*/ 1,
8406 					      /*bit*/ 0);
8407 			ctl_done((union ctl_io *)ctsio);
8408 			return (CTL_RETVAL_COMPLETE);
8409 		}
8410 	}
8411 
8412 	param_len = scsi_4btoul(cdb->length);
8413 
8414 	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8415 		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8416 		ctsio->kern_data_len = param_len;
8417 		ctsio->kern_total_len = param_len;
8418 		ctsio->kern_data_resid = 0;
8419 		ctsio->kern_rel_offset = 0;
8420 		ctsio->kern_sg_entries = 0;
8421 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8422 		ctsio->be_move_done = ctl_config_move_done;
8423 		ctl_datamove((union ctl_io *)ctsio);
8424 
8425 		return (CTL_RETVAL_COMPLETE);
8426 	}
8427 
8428 	param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8429 
8430 	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8431 	res_key = scsi_8btou64(param->res_key.key);
8432 	sa_res_key = scsi_8btou64(param->serv_act_res_key);
8433 
8434 	/*
8435 	 * Validate the reservation key here except for SPRO_REG_IGNO
8436 	 * This must be done for all other service actions
8437 	 */
8438 	if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8439 		mtx_lock(&lun->lun_lock);
8440 		if ((key = ctl_get_prkey(lun, residx)) != 0) {
8441 			if (res_key != key) {
8442 				/*
8443 				 * The current key passed in doesn't match
8444 				 * the one the initiator previously
8445 				 * registered.
8446 				 */
8447 				mtx_unlock(&lun->lun_lock);
8448 				free(ctsio->kern_data_ptr, M_CTL);
8449 				ctl_set_reservation_conflict(ctsio);
8450 				ctl_done((union ctl_io *)ctsio);
8451 				return (CTL_RETVAL_COMPLETE);
8452 			}
8453 		} else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8454 			/*
8455 			 * We are not registered
8456 			 */
8457 			mtx_unlock(&lun->lun_lock);
8458 			free(ctsio->kern_data_ptr, M_CTL);
8459 			ctl_set_reservation_conflict(ctsio);
8460 			ctl_done((union ctl_io *)ctsio);
8461 			return (CTL_RETVAL_COMPLETE);
8462 		} else if (res_key != 0) {
8463 			/*
8464 			 * We are not registered and trying to register but
8465 			 * the register key isn't zero.
8466 			 */
8467 			mtx_unlock(&lun->lun_lock);
8468 			free(ctsio->kern_data_ptr, M_CTL);
8469 			ctl_set_reservation_conflict(ctsio);
8470 			ctl_done((union ctl_io *)ctsio);
8471 			return (CTL_RETVAL_COMPLETE);
8472 		}
8473 		mtx_unlock(&lun->lun_lock);
8474 	}
8475 
8476 	switch (cdb->action & SPRO_ACTION_MASK) {
8477 	case SPRO_REGISTER:
8478 	case SPRO_REG_IGNO: {
8479 
8480 #if 0
8481 		printf("Registration received\n");
8482 #endif
8483 
8484 		/*
8485 		 * We don't support any of these options, as we report in
8486 		 * the read capabilities request (see
8487 		 * ctl_persistent_reserve_in(), above).
8488 		 */
8489 		if ((param->flags & SPR_SPEC_I_PT)
8490 		 || (param->flags & SPR_ALL_TG_PT)
8491 		 || (param->flags & SPR_APTPL)) {
8492 			int bit_ptr;
8493 
8494 			if (param->flags & SPR_APTPL)
8495 				bit_ptr = 0;
8496 			else if (param->flags & SPR_ALL_TG_PT)
8497 				bit_ptr = 2;
8498 			else /* SPR_SPEC_I_PT */
8499 				bit_ptr = 3;
8500 
8501 			free(ctsio->kern_data_ptr, M_CTL);
8502 			ctl_set_invalid_field(ctsio,
8503 					      /*sks_valid*/ 1,
8504 					      /*command*/ 0,
8505 					      /*field*/ 20,
8506 					      /*bit_valid*/ 1,
8507 					      /*bit*/ bit_ptr);
8508 			ctl_done((union ctl_io *)ctsio);
8509 			return (CTL_RETVAL_COMPLETE);
8510 		}
8511 
8512 		mtx_lock(&lun->lun_lock);
8513 
8514 		/*
8515 		 * The initiator wants to clear the
8516 		 * key/unregister.
8517 		 */
8518 		if (sa_res_key == 0) {
8519 			if ((res_key == 0
8520 			  && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8521 			 || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8522 			  && ctl_get_prkey(lun, residx) == 0)) {
8523 				mtx_unlock(&lun->lun_lock);
8524 				goto done;
8525 			}
8526 
8527 			ctl_clr_prkey(lun, residx);
8528 			lun->pr_key_count--;
8529 
8530 			if (residx == lun->pr_res_idx) {
8531 				lun->flags &= ~CTL_LUN_PR_RESERVED;
8532 				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8533 
8534 				if ((lun->res_type == SPR_TYPE_WR_EX_RO
8535 				  || lun->res_type == SPR_TYPE_EX_AC_RO)
8536 				 && lun->pr_key_count) {
8537 					/*
8538 					 * If the reservation is a registrants
8539 					 * only type we need to generate a UA
8540 					 * for other registered inits.  The
8541 					 * sense code should be RESERVATIONS
8542 					 * RELEASED
8543 					 */
8544 
8545 					for (i = 0; i < CTL_MAX_INITIATORS;i++){
8546 						if (ctl_get_prkey(lun, i +
8547 						    softc->persis_offset) == 0)
8548 							continue;
8549 						ctl_est_ua(lun, i,
8550 						    CTL_UA_RES_RELEASE);
8551 					}
8552 				}
8553 				lun->res_type = 0;
8554 			} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8555 				if (lun->pr_key_count==0) {
8556 					lun->flags &= ~CTL_LUN_PR_RESERVED;
8557 					lun->res_type = 0;
8558 					lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8559 				}
8560 			}
8561 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8562 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8563 			persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8564 			persis_io.pr.pr_info.residx = residx;
8565 			if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8566 			     &persis_io, sizeof(persis_io), 0 )) >
8567 			     CTL_HA_STATUS_SUCCESS) {
8568 				printf("CTL:Persis Out error returned from "
8569 				       "ctl_ha_msg_send %d\n", isc_retval);
8570 			}
8571 		} else /* sa_res_key != 0 */ {
8572 
8573 			/*
8574 			 * If we aren't registered currently then increment
8575 			 * the key count and set the registered flag.
8576 			 */
8577 			ctl_alloc_prkey(lun, residx);
8578 			if (ctl_get_prkey(lun, residx) == 0)
8579 				lun->pr_key_count++;
8580 			ctl_set_prkey(lun, residx, sa_res_key);
8581 
8582 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8583 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8584 			persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8585 			persis_io.pr.pr_info.residx = residx;
8586 			memcpy(persis_io.pr.pr_info.sa_res_key,
8587 			       param->serv_act_res_key,
8588 			       sizeof(param->serv_act_res_key));
8589 			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8590 			     &persis_io, sizeof(persis_io), 0)) >
8591 			     CTL_HA_STATUS_SUCCESS) {
8592 				printf("CTL:Persis Out error returned from "
8593 				       "ctl_ha_msg_send %d\n", isc_retval);
8594 			}
8595 		}
8596 		lun->PRGeneration++;
8597 		mtx_unlock(&lun->lun_lock);
8598 
8599 		break;
8600 	}
8601 	case SPRO_RESERVE:
8602 #if 0
8603                 printf("Reserve executed type %d\n", type);
8604 #endif
8605 		mtx_lock(&lun->lun_lock);
8606 		if (lun->flags & CTL_LUN_PR_RESERVED) {
8607 			/*
8608 			 * if this isn't the reservation holder and it's
8609 			 * not a "all registrants" type or if the type is
8610 			 * different then we have a conflict
8611 			 */
8612 			if ((lun->pr_res_idx != residx
8613 			  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8614 			 || lun->res_type != type) {
8615 				mtx_unlock(&lun->lun_lock);
8616 				free(ctsio->kern_data_ptr, M_CTL);
8617 				ctl_set_reservation_conflict(ctsio);
8618 				ctl_done((union ctl_io *)ctsio);
8619 				return (CTL_RETVAL_COMPLETE);
8620 			}
8621 			mtx_unlock(&lun->lun_lock);
8622 		} else /* create a reservation */ {
8623 			/*
8624 			 * If it's not an "all registrants" type record
8625 			 * reservation holder
8626 			 */
8627 			if (type != SPR_TYPE_WR_EX_AR
8628 			 && type != SPR_TYPE_EX_AC_AR)
8629 				lun->pr_res_idx = residx; /* Res holder */
8630 			else
8631 				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8632 
8633 			lun->flags |= CTL_LUN_PR_RESERVED;
8634 			lun->res_type = type;
8635 
8636 			mtx_unlock(&lun->lun_lock);
8637 
8638 			/* send msg to other side */
8639 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8640 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8641 			persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8642 			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8643 			persis_io.pr.pr_info.res_type = type;
8644 			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8645 			     &persis_io, sizeof(persis_io), 0)) >
8646 			     CTL_HA_STATUS_SUCCESS) {
8647 				printf("CTL:Persis Out error returned from "
8648 				       "ctl_ha_msg_send %d\n", isc_retval);
8649 			}
8650 		}
8651 		break;
8652 
8653 	case SPRO_RELEASE:
8654 		mtx_lock(&lun->lun_lock);
8655 		if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8656 			/* No reservation exists return good status */
8657 			mtx_unlock(&lun->lun_lock);
8658 			goto done;
8659 		}
8660 		/*
8661 		 * Is this nexus a reservation holder?
8662 		 */
8663 		if (lun->pr_res_idx != residx
8664 		 && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8665 			/*
8666 			 * not a res holder return good status but
8667 			 * do nothing
8668 			 */
8669 			mtx_unlock(&lun->lun_lock);
8670 			goto done;
8671 		}
8672 
8673 		if (lun->res_type != type) {
8674 			mtx_unlock(&lun->lun_lock);
8675 			free(ctsio->kern_data_ptr, M_CTL);
8676 			ctl_set_illegal_pr_release(ctsio);
8677 			ctl_done((union ctl_io *)ctsio);
8678 			return (CTL_RETVAL_COMPLETE);
8679 		}
8680 
8681 		/* okay to release */
8682 		lun->flags &= ~CTL_LUN_PR_RESERVED;
8683 		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8684 		lun->res_type = 0;
8685 
8686 		/*
8687 		 * if this isn't an exclusive access
8688 		 * res generate UA for all other
8689 		 * registrants.
8690 		 */
8691 		if (type != SPR_TYPE_EX_AC
8692 		 && type != SPR_TYPE_WR_EX) {
8693 			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8694 				if (i == residx ||
8695 				    ctl_get_prkey(lun,
8696 				     i + softc->persis_offset) == 0)
8697 					continue;
8698 				ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8699 			}
8700 		}
8701 		mtx_unlock(&lun->lun_lock);
8702 		/* Send msg to other side */
8703 		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8704 		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8705 		persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8706 		if ((isc_retval=ctl_ha_msg_send( CTL_HA_CHAN_CTL, &persis_io,
8707 		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8708 			printf("CTL:Persis Out error returned from "
8709 			       "ctl_ha_msg_send %d\n", isc_retval);
8710 		}
8711 		break;
8712 
8713 	case SPRO_CLEAR:
8714 		/* send msg to other side */
8715 
8716 		mtx_lock(&lun->lun_lock);
8717 		lun->flags &= ~CTL_LUN_PR_RESERVED;
8718 		lun->res_type = 0;
8719 		lun->pr_key_count = 0;
8720 		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8721 
8722 		ctl_clr_prkey(lun, residx);
8723 		for (i=0; i < 2*CTL_MAX_INITIATORS; i++)
8724 			if (ctl_get_prkey(lun, i) != 0) {
8725 				ctl_clr_prkey(lun, i);
8726 				ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
8727 			}
8728 		lun->PRGeneration++;
8729 		mtx_unlock(&lun->lun_lock);
8730 		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8731 		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8732 		persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8733 		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8734 		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8735 			printf("CTL:Persis Out error returned from "
8736 			       "ctl_ha_msg_send %d\n", isc_retval);
8737 		}
8738 		break;
8739 
8740 	case SPRO_PREEMPT:
8741 	case SPRO_PRE_ABO: {
8742 		int nretval;
8743 
8744 		nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8745 					  residx, ctsio, cdb, param);
8746 		if (nretval != 0)
8747 			return (CTL_RETVAL_COMPLETE);
8748 		break;
8749 	}
8750 	default:
8751 		panic("Invalid PR type %x", cdb->action);
8752 	}
8753 
8754 done:
8755 	free(ctsio->kern_data_ptr, M_CTL);
8756 	ctl_set_success(ctsio);
8757 	ctl_done((union ctl_io *)ctsio);
8758 
8759 	return (retval);
8760 }
8761 
8762 /*
8763  * This routine is for handling a message from the other SC pertaining to
8764  * persistent reserve out. All the error checking will have been done
8765  * so only perorming the action need be done here to keep the two
8766  * in sync.
8767  */
8768 static void
8769 ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8770 {
8771 	struct ctl_lun *lun;
8772 	struct ctl_softc *softc;
8773 	int i;
8774 	uint32_t targ_lun;
8775 
8776 	softc = control_softc;
8777 
8778 	targ_lun = msg->hdr.nexus.targ_mapped_lun;
8779 	lun = softc->ctl_luns[targ_lun];
8780 	mtx_lock(&lun->lun_lock);
8781 	switch(msg->pr.pr_info.action) {
8782 	case CTL_PR_REG_KEY:
8783 		ctl_alloc_prkey(lun, msg->pr.pr_info.residx);
8784 		if (ctl_get_prkey(lun, msg->pr.pr_info.residx) == 0)
8785 			lun->pr_key_count++;
8786 		ctl_set_prkey(lun, msg->pr.pr_info.residx,
8787 		    scsi_8btou64(msg->pr.pr_info.sa_res_key));
8788 		lun->PRGeneration++;
8789 		break;
8790 
8791 	case CTL_PR_UNREG_KEY:
8792 		ctl_clr_prkey(lun, msg->pr.pr_info.residx);
8793 		lun->pr_key_count--;
8794 
8795 		/* XXX Need to see if the reservation has been released */
8796 		/* if so do we need to generate UA? */
8797 		if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8798 			lun->flags &= ~CTL_LUN_PR_RESERVED;
8799 			lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8800 
8801 			if ((lun->res_type == SPR_TYPE_WR_EX_RO
8802 			  || lun->res_type == SPR_TYPE_EX_AC_RO)
8803 			 && lun->pr_key_count) {
8804 				/*
8805 				 * If the reservation is a registrants
8806 				 * only type we need to generate a UA
8807 				 * for other registered inits.  The
8808 				 * sense code should be RESERVATIONS
8809 				 * RELEASED
8810 				 */
8811 
8812 				for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8813 					if (ctl_get_prkey(lun, i +
8814 					    softc->persis_offset) == 0)
8815 						continue;
8816 
8817 					ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8818 				}
8819 			}
8820 			lun->res_type = 0;
8821 		} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8822 			if (lun->pr_key_count==0) {
8823 				lun->flags &= ~CTL_LUN_PR_RESERVED;
8824 				lun->res_type = 0;
8825 				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8826 			}
8827 		}
8828 		lun->PRGeneration++;
8829 		break;
8830 
8831 	case CTL_PR_RESERVE:
8832 		lun->flags |= CTL_LUN_PR_RESERVED;
8833 		lun->res_type = msg->pr.pr_info.res_type;
8834 		lun->pr_res_idx = msg->pr.pr_info.residx;
8835 
8836 		break;
8837 
8838 	case CTL_PR_RELEASE:
8839 		/*
8840 		 * if this isn't an exclusive access res generate UA for all
8841 		 * other registrants.
8842 		 */
8843 		if (lun->res_type != SPR_TYPE_EX_AC
8844 		 && lun->res_type != SPR_TYPE_WR_EX) {
8845 			for (i = 0; i < CTL_MAX_INITIATORS; i++)
8846 				if (ctl_get_prkey(lun, i + softc->persis_offset) != 0)
8847 					ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8848 		}
8849 
8850 		lun->flags &= ~CTL_LUN_PR_RESERVED;
8851 		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8852 		lun->res_type = 0;
8853 		break;
8854 
8855 	case CTL_PR_PREEMPT:
8856 		ctl_pro_preempt_other(lun, msg);
8857 		break;
8858 	case CTL_PR_CLEAR:
8859 		lun->flags &= ~CTL_LUN_PR_RESERVED;
8860 		lun->res_type = 0;
8861 		lun->pr_key_count = 0;
8862 		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8863 
8864 		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8865 			if (ctl_get_prkey(lun, i) == 0)
8866 				continue;
8867 			ctl_clr_prkey(lun, i);
8868 			ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
8869 		}
8870 		lun->PRGeneration++;
8871 		break;
8872 	}
8873 
8874 	mtx_unlock(&lun->lun_lock);
8875 }
8876 
8877 int
8878 ctl_read_write(struct ctl_scsiio *ctsio)
8879 {
8880 	struct ctl_lun *lun;
8881 	struct ctl_lba_len_flags *lbalen;
8882 	uint64_t lba;
8883 	uint32_t num_blocks;
8884 	int flags, retval;
8885 	int isread;
8886 
8887 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8888 
8889 	CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8890 
8891 	flags = 0;
8892 	retval = CTL_RETVAL_COMPLETE;
8893 
8894 	isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8895 	      || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8896 	switch (ctsio->cdb[0]) {
8897 	case READ_6:
8898 	case WRITE_6: {
8899 		struct scsi_rw_6 *cdb;
8900 
8901 		cdb = (struct scsi_rw_6 *)ctsio->cdb;
8902 
8903 		lba = scsi_3btoul(cdb->addr);
8904 		/* only 5 bits are valid in the most significant address byte */
8905 		lba &= 0x1fffff;
8906 		num_blocks = cdb->length;
8907 		/*
8908 		 * This is correct according to SBC-2.
8909 		 */
8910 		if (num_blocks == 0)
8911 			num_blocks = 256;
8912 		break;
8913 	}
8914 	case READ_10:
8915 	case WRITE_10: {
8916 		struct scsi_rw_10 *cdb;
8917 
8918 		cdb = (struct scsi_rw_10 *)ctsio->cdb;
8919 		if (cdb->byte2 & SRW10_FUA)
8920 			flags |= CTL_LLF_FUA;
8921 		if (cdb->byte2 & SRW10_DPO)
8922 			flags |= CTL_LLF_DPO;
8923 		lba = scsi_4btoul(cdb->addr);
8924 		num_blocks = scsi_2btoul(cdb->length);
8925 		break;
8926 	}
8927 	case WRITE_VERIFY_10: {
8928 		struct scsi_write_verify_10 *cdb;
8929 
8930 		cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8931 		flags |= CTL_LLF_FUA;
8932 		if (cdb->byte2 & SWV_DPO)
8933 			flags |= CTL_LLF_DPO;
8934 		lba = scsi_4btoul(cdb->addr);
8935 		num_blocks = scsi_2btoul(cdb->length);
8936 		break;
8937 	}
8938 	case READ_12:
8939 	case WRITE_12: {
8940 		struct scsi_rw_12 *cdb;
8941 
8942 		cdb = (struct scsi_rw_12 *)ctsio->cdb;
8943 		if (cdb->byte2 & SRW12_FUA)
8944 			flags |= CTL_LLF_FUA;
8945 		if (cdb->byte2 & SRW12_DPO)
8946 			flags |= CTL_LLF_DPO;
8947 		lba = scsi_4btoul(cdb->addr);
8948 		num_blocks = scsi_4btoul(cdb->length);
8949 		break;
8950 	}
8951 	case WRITE_VERIFY_12: {
8952 		struct scsi_write_verify_12 *cdb;
8953 
8954 		cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8955 		flags |= CTL_LLF_FUA;
8956 		if (cdb->byte2 & SWV_DPO)
8957 			flags |= CTL_LLF_DPO;
8958 		lba = scsi_4btoul(cdb->addr);
8959 		num_blocks = scsi_4btoul(cdb->length);
8960 		break;
8961 	}
8962 	case READ_16:
8963 	case WRITE_16: {
8964 		struct scsi_rw_16 *cdb;
8965 
8966 		cdb = (struct scsi_rw_16 *)ctsio->cdb;
8967 		if (cdb->byte2 & SRW12_FUA)
8968 			flags |= CTL_LLF_FUA;
8969 		if (cdb->byte2 & SRW12_DPO)
8970 			flags |= CTL_LLF_DPO;
8971 		lba = scsi_8btou64(cdb->addr);
8972 		num_blocks = scsi_4btoul(cdb->length);
8973 		break;
8974 	}
8975 	case WRITE_ATOMIC_16: {
8976 		struct scsi_rw_16 *cdb;
8977 
8978 		if (lun->be_lun->atomicblock == 0) {
8979 			ctl_set_invalid_opcode(ctsio);
8980 			ctl_done((union ctl_io *)ctsio);
8981 			return (CTL_RETVAL_COMPLETE);
8982 		}
8983 
8984 		cdb = (struct scsi_rw_16 *)ctsio->cdb;
8985 		if (cdb->byte2 & SRW12_FUA)
8986 			flags |= CTL_LLF_FUA;
8987 		if (cdb->byte2 & SRW12_DPO)
8988 			flags |= CTL_LLF_DPO;
8989 		lba = scsi_8btou64(cdb->addr);
8990 		num_blocks = scsi_4btoul(cdb->length);
8991 		if (num_blocks > lun->be_lun->atomicblock) {
8992 			ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
8993 			    /*command*/ 1, /*field*/ 12, /*bit_valid*/ 0,
8994 			    /*bit*/ 0);
8995 			ctl_done((union ctl_io *)ctsio);
8996 			return (CTL_RETVAL_COMPLETE);
8997 		}
8998 		break;
8999 	}
9000 	case WRITE_VERIFY_16: {
9001 		struct scsi_write_verify_16 *cdb;
9002 
9003 		cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
9004 		flags |= CTL_LLF_FUA;
9005 		if (cdb->byte2 & SWV_DPO)
9006 			flags |= CTL_LLF_DPO;
9007 		lba = scsi_8btou64(cdb->addr);
9008 		num_blocks = scsi_4btoul(cdb->length);
9009 		break;
9010 	}
9011 	default:
9012 		/*
9013 		 * We got a command we don't support.  This shouldn't
9014 		 * happen, commands should be filtered out above us.
9015 		 */
9016 		ctl_set_invalid_opcode(ctsio);
9017 		ctl_done((union ctl_io *)ctsio);
9018 
9019 		return (CTL_RETVAL_COMPLETE);
9020 		break; /* NOTREACHED */
9021 	}
9022 
9023 	/*
9024 	 * The first check is to make sure we're in bounds, the second
9025 	 * check is to catch wrap-around problems.  If the lba + num blocks
9026 	 * is less than the lba, then we've wrapped around and the block
9027 	 * range is invalid anyway.
9028 	 */
9029 	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9030 	 || ((lba + num_blocks) < lba)) {
9031 		ctl_set_lba_out_of_range(ctsio);
9032 		ctl_done((union ctl_io *)ctsio);
9033 		return (CTL_RETVAL_COMPLETE);
9034 	}
9035 
9036 	/*
9037 	 * According to SBC-3, a transfer length of 0 is not an error.
9038 	 * Note that this cannot happen with WRITE(6) or READ(6), since 0
9039 	 * translates to 256 blocks for those commands.
9040 	 */
9041 	if (num_blocks == 0) {
9042 		ctl_set_success(ctsio);
9043 		ctl_done((union ctl_io *)ctsio);
9044 		return (CTL_RETVAL_COMPLETE);
9045 	}
9046 
9047 	/* Set FUA and/or DPO if caches are disabled. */
9048 	if (isread) {
9049 		if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9050 		    SCP_RCD) != 0)
9051 			flags |= CTL_LLF_FUA | CTL_LLF_DPO;
9052 	} else {
9053 		if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9054 		    SCP_WCE) == 0)
9055 			flags |= CTL_LLF_FUA;
9056 	}
9057 
9058 	lbalen = (struct ctl_lba_len_flags *)
9059 	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9060 	lbalen->lba = lba;
9061 	lbalen->len = num_blocks;
9062 	lbalen->flags = (isread ? CTL_LLF_READ : CTL_LLF_WRITE) | flags;
9063 
9064 	ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9065 	ctsio->kern_rel_offset = 0;
9066 
9067 	CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
9068 
9069 	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9070 
9071 	return (retval);
9072 }
9073 
9074 static int
9075 ctl_cnw_cont(union ctl_io *io)
9076 {
9077 	struct ctl_scsiio *ctsio;
9078 	struct ctl_lun *lun;
9079 	struct ctl_lba_len_flags *lbalen;
9080 	int retval;
9081 
9082 	ctsio = &io->scsiio;
9083 	ctsio->io_hdr.status = CTL_STATUS_NONE;
9084 	ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
9085 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9086 	lbalen = (struct ctl_lba_len_flags *)
9087 	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9088 	lbalen->flags &= ~CTL_LLF_COMPARE;
9089 	lbalen->flags |= CTL_LLF_WRITE;
9090 
9091 	CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
9092 	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9093 	return (retval);
9094 }
9095 
9096 int
9097 ctl_cnw(struct ctl_scsiio *ctsio)
9098 {
9099 	struct ctl_lun *lun;
9100 	struct ctl_lba_len_flags *lbalen;
9101 	uint64_t lba;
9102 	uint32_t num_blocks;
9103 	int flags, retval;
9104 
9105 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9106 
9107 	CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
9108 
9109 	flags = 0;
9110 	retval = CTL_RETVAL_COMPLETE;
9111 
9112 	switch (ctsio->cdb[0]) {
9113 	case COMPARE_AND_WRITE: {
9114 		struct scsi_compare_and_write *cdb;
9115 
9116 		cdb = (struct scsi_compare_and_write *)ctsio->cdb;
9117 		if (cdb->byte2 & SRW10_FUA)
9118 			flags |= CTL_LLF_FUA;
9119 		if (cdb->byte2 & SRW10_DPO)
9120 			flags |= CTL_LLF_DPO;
9121 		lba = scsi_8btou64(cdb->addr);
9122 		num_blocks = cdb->length;
9123 		break;
9124 	}
9125 	default:
9126 		/*
9127 		 * We got a command we don't support.  This shouldn't
9128 		 * happen, commands should be filtered out above us.
9129 		 */
9130 		ctl_set_invalid_opcode(ctsio);
9131 		ctl_done((union ctl_io *)ctsio);
9132 
9133 		return (CTL_RETVAL_COMPLETE);
9134 		break; /* NOTREACHED */
9135 	}
9136 
9137 	/*
9138 	 * The first check is to make sure we're in bounds, the second
9139 	 * check is to catch wrap-around problems.  If the lba + num blocks
9140 	 * is less than the lba, then we've wrapped around and the block
9141 	 * range is invalid anyway.
9142 	 */
9143 	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9144 	 || ((lba + num_blocks) < lba)) {
9145 		ctl_set_lba_out_of_range(ctsio);
9146 		ctl_done((union ctl_io *)ctsio);
9147 		return (CTL_RETVAL_COMPLETE);
9148 	}
9149 
9150 	/*
9151 	 * According to SBC-3, a transfer length of 0 is not an error.
9152 	 */
9153 	if (num_blocks == 0) {
9154 		ctl_set_success(ctsio);
9155 		ctl_done((union ctl_io *)ctsio);
9156 		return (CTL_RETVAL_COMPLETE);
9157 	}
9158 
9159 	/* Set FUA if write cache is disabled. */
9160 	if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9161 	    SCP_WCE) == 0)
9162 		flags |= CTL_LLF_FUA;
9163 
9164 	ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
9165 	ctsio->kern_rel_offset = 0;
9166 
9167 	/*
9168 	 * Set the IO_CONT flag, so that if this I/O gets passed to
9169 	 * ctl_data_submit_done(), it'll get passed back to
9170 	 * ctl_ctl_cnw_cont() for further processing.
9171 	 */
9172 	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
9173 	ctsio->io_cont = ctl_cnw_cont;
9174 
9175 	lbalen = (struct ctl_lba_len_flags *)
9176 	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9177 	lbalen->lba = lba;
9178 	lbalen->len = num_blocks;
9179 	lbalen->flags = CTL_LLF_COMPARE | flags;
9180 
9181 	CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
9182 	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9183 	return (retval);
9184 }
9185 
9186 int
9187 ctl_verify(struct ctl_scsiio *ctsio)
9188 {
9189 	struct ctl_lun *lun;
9190 	struct ctl_lba_len_flags *lbalen;
9191 	uint64_t lba;
9192 	uint32_t num_blocks;
9193 	int bytchk, flags;
9194 	int retval;
9195 
9196 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9197 
9198 	CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
9199 
9200 	bytchk = 0;
9201 	flags = CTL_LLF_FUA;
9202 	retval = CTL_RETVAL_COMPLETE;
9203 
9204 	switch (ctsio->cdb[0]) {
9205 	case VERIFY_10: {
9206 		struct scsi_verify_10 *cdb;
9207 
9208 		cdb = (struct scsi_verify_10 *)ctsio->cdb;
9209 		if (cdb->byte2 & SVFY_BYTCHK)
9210 			bytchk = 1;
9211 		if (cdb->byte2 & SVFY_DPO)
9212 			flags |= CTL_LLF_DPO;
9213 		lba = scsi_4btoul(cdb->addr);
9214 		num_blocks = scsi_2btoul(cdb->length);
9215 		break;
9216 	}
9217 	case VERIFY_12: {
9218 		struct scsi_verify_12 *cdb;
9219 
9220 		cdb = (struct scsi_verify_12 *)ctsio->cdb;
9221 		if (cdb->byte2 & SVFY_BYTCHK)
9222 			bytchk = 1;
9223 		if (cdb->byte2 & SVFY_DPO)
9224 			flags |= CTL_LLF_DPO;
9225 		lba = scsi_4btoul(cdb->addr);
9226 		num_blocks = scsi_4btoul(cdb->length);
9227 		break;
9228 	}
9229 	case VERIFY_16: {
9230 		struct scsi_rw_16 *cdb;
9231 
9232 		cdb = (struct scsi_rw_16 *)ctsio->cdb;
9233 		if (cdb->byte2 & SVFY_BYTCHK)
9234 			bytchk = 1;
9235 		if (cdb->byte2 & SVFY_DPO)
9236 			flags |= CTL_LLF_DPO;
9237 		lba = scsi_8btou64(cdb->addr);
9238 		num_blocks = scsi_4btoul(cdb->length);
9239 		break;
9240 	}
9241 	default:
9242 		/*
9243 		 * We got a command we don't support.  This shouldn't
9244 		 * happen, commands should be filtered out above us.
9245 		 */
9246 		ctl_set_invalid_opcode(ctsio);
9247 		ctl_done((union ctl_io *)ctsio);
9248 		return (CTL_RETVAL_COMPLETE);
9249 	}
9250 
9251 	/*
9252 	 * The first check is to make sure we're in bounds, the second
9253 	 * check is to catch wrap-around problems.  If the lba + num blocks
9254 	 * is less than the lba, then we've wrapped around and the block
9255 	 * range is invalid anyway.
9256 	 */
9257 	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9258 	 || ((lba + num_blocks) < lba)) {
9259 		ctl_set_lba_out_of_range(ctsio);
9260 		ctl_done((union ctl_io *)ctsio);
9261 		return (CTL_RETVAL_COMPLETE);
9262 	}
9263 
9264 	/*
9265 	 * According to SBC-3, a transfer length of 0 is not an error.
9266 	 */
9267 	if (num_blocks == 0) {
9268 		ctl_set_success(ctsio);
9269 		ctl_done((union ctl_io *)ctsio);
9270 		return (CTL_RETVAL_COMPLETE);
9271 	}
9272 
9273 	lbalen = (struct ctl_lba_len_flags *)
9274 	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9275 	lbalen->lba = lba;
9276 	lbalen->len = num_blocks;
9277 	if (bytchk) {
9278 		lbalen->flags = CTL_LLF_COMPARE | flags;
9279 		ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9280 	} else {
9281 		lbalen->flags = CTL_LLF_VERIFY | flags;
9282 		ctsio->kern_total_len = 0;
9283 	}
9284 	ctsio->kern_rel_offset = 0;
9285 
9286 	CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9287 	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9288 	return (retval);
9289 }
9290 
9291 int
9292 ctl_report_luns(struct ctl_scsiio *ctsio)
9293 {
9294 	struct scsi_report_luns *cdb;
9295 	struct scsi_report_luns_data *lun_data;
9296 	struct ctl_lun *lun, *request_lun;
9297 	int num_luns, retval;
9298 	uint32_t alloc_len, lun_datalen;
9299 	int num_filled, well_known;
9300 	uint32_t initidx, targ_lun_id, lun_id;
9301 
9302 	retval = CTL_RETVAL_COMPLETE;
9303 	well_known = 0;
9304 
9305 	cdb = (struct scsi_report_luns *)ctsio->cdb;
9306 
9307 	CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9308 
9309 	mtx_lock(&control_softc->ctl_lock);
9310 	num_luns = control_softc->num_luns;
9311 	mtx_unlock(&control_softc->ctl_lock);
9312 
9313 	switch (cdb->select_report) {
9314 	case RPL_REPORT_DEFAULT:
9315 	case RPL_REPORT_ALL:
9316 		break;
9317 	case RPL_REPORT_WELLKNOWN:
9318 		well_known = 1;
9319 		num_luns = 0;
9320 		break;
9321 	default:
9322 		ctl_set_invalid_field(ctsio,
9323 				      /*sks_valid*/ 1,
9324 				      /*command*/ 1,
9325 				      /*field*/ 2,
9326 				      /*bit_valid*/ 0,
9327 				      /*bit*/ 0);
9328 		ctl_done((union ctl_io *)ctsio);
9329 		return (retval);
9330 		break; /* NOTREACHED */
9331 	}
9332 
9333 	alloc_len = scsi_4btoul(cdb->length);
9334 	/*
9335 	 * The initiator has to allocate at least 16 bytes for this request,
9336 	 * so he can at least get the header and the first LUN.  Otherwise
9337 	 * we reject the request (per SPC-3 rev 14, section 6.21).
9338 	 */
9339 	if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9340 	    sizeof(struct scsi_report_luns_lundata))) {
9341 		ctl_set_invalid_field(ctsio,
9342 				      /*sks_valid*/ 1,
9343 				      /*command*/ 1,
9344 				      /*field*/ 6,
9345 				      /*bit_valid*/ 0,
9346 				      /*bit*/ 0);
9347 		ctl_done((union ctl_io *)ctsio);
9348 		return (retval);
9349 	}
9350 
9351 	request_lun = (struct ctl_lun *)
9352 		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9353 
9354 	lun_datalen = sizeof(*lun_data) +
9355 		(num_luns * sizeof(struct scsi_report_luns_lundata));
9356 
9357 	ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9358 	lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9359 	ctsio->kern_sg_entries = 0;
9360 
9361 	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9362 
9363 	mtx_lock(&control_softc->ctl_lock);
9364 	for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
9365 		lun_id = ctl_map_lun(ctsio->io_hdr.nexus.targ_port, targ_lun_id);
9366 		if (lun_id >= CTL_MAX_LUNS)
9367 			continue;
9368 		lun = control_softc->ctl_luns[lun_id];
9369 		if (lun == NULL)
9370 			continue;
9371 
9372 		if (targ_lun_id <= 0xff) {
9373 			/*
9374 			 * Peripheral addressing method, bus number 0.
9375 			 */
9376 			lun_data->luns[num_filled].lundata[0] =
9377 				RPL_LUNDATA_ATYP_PERIPH;
9378 			lun_data->luns[num_filled].lundata[1] = targ_lun_id;
9379 			num_filled++;
9380 		} else if (targ_lun_id <= 0x3fff) {
9381 			/*
9382 			 * Flat addressing method.
9383 			 */
9384 			lun_data->luns[num_filled].lundata[0] =
9385 				RPL_LUNDATA_ATYP_FLAT | (targ_lun_id >> 8);
9386 			lun_data->luns[num_filled].lundata[1] =
9387 				(targ_lun_id & 0xff);
9388 			num_filled++;
9389 		} else if (targ_lun_id <= 0xffffff) {
9390 			/*
9391 			 * Extended flat addressing method.
9392 			 */
9393 			lun_data->luns[num_filled].lundata[0] =
9394 			    RPL_LUNDATA_ATYP_EXTLUN | 0x12;
9395 			scsi_ulto3b(targ_lun_id,
9396 			    &lun_data->luns[num_filled].lundata[1]);
9397 			num_filled++;
9398 		} else {
9399 			printf("ctl_report_luns: bogus LUN number %jd, "
9400 			       "skipping\n", (intmax_t)targ_lun_id);
9401 		}
9402 		/*
9403 		 * According to SPC-3, rev 14 section 6.21:
9404 		 *
9405 		 * "The execution of a REPORT LUNS command to any valid and
9406 		 * installed logical unit shall clear the REPORTED LUNS DATA
9407 		 * HAS CHANGED unit attention condition for all logical
9408 		 * units of that target with respect to the requesting
9409 		 * initiator. A valid and installed logical unit is one
9410 		 * having a PERIPHERAL QUALIFIER of 000b in the standard
9411 		 * INQUIRY data (see 6.4.2)."
9412 		 *
9413 		 * If request_lun is NULL, the LUN this report luns command
9414 		 * was issued to is either disabled or doesn't exist. In that
9415 		 * case, we shouldn't clear any pending lun change unit
9416 		 * attention.
9417 		 */
9418 		if (request_lun != NULL) {
9419 			mtx_lock(&lun->lun_lock);
9420 			ctl_clr_ua(lun, initidx, CTL_UA_RES_RELEASE);
9421 			mtx_unlock(&lun->lun_lock);
9422 		}
9423 	}
9424 	mtx_unlock(&control_softc->ctl_lock);
9425 
9426 	/*
9427 	 * It's quite possible that we've returned fewer LUNs than we allocated
9428 	 * space for.  Trim it.
9429 	 */
9430 	lun_datalen = sizeof(*lun_data) +
9431 		(num_filled * sizeof(struct scsi_report_luns_lundata));
9432 
9433 	if (lun_datalen < alloc_len) {
9434 		ctsio->residual = alloc_len - lun_datalen;
9435 		ctsio->kern_data_len = lun_datalen;
9436 		ctsio->kern_total_len = lun_datalen;
9437 	} else {
9438 		ctsio->residual = 0;
9439 		ctsio->kern_data_len = alloc_len;
9440 		ctsio->kern_total_len = alloc_len;
9441 	}
9442 	ctsio->kern_data_resid = 0;
9443 	ctsio->kern_rel_offset = 0;
9444 	ctsio->kern_sg_entries = 0;
9445 
9446 	/*
9447 	 * We set this to the actual data length, regardless of how much
9448 	 * space we actually have to return results.  If the user looks at
9449 	 * this value, he'll know whether or not he allocated enough space
9450 	 * and reissue the command if necessary.  We don't support well
9451 	 * known logical units, so if the user asks for that, return none.
9452 	 */
9453 	scsi_ulto4b(lun_datalen - 8, lun_data->length);
9454 
9455 	/*
9456 	 * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9457 	 * this request.
9458 	 */
9459 	ctl_set_success(ctsio);
9460 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9461 	ctsio->be_move_done = ctl_config_move_done;
9462 	ctl_datamove((union ctl_io *)ctsio);
9463 	return (retval);
9464 }
9465 
9466 int
9467 ctl_request_sense(struct ctl_scsiio *ctsio)
9468 {
9469 	struct scsi_request_sense *cdb;
9470 	struct scsi_sense_data *sense_ptr;
9471 	struct ctl_lun *lun;
9472 	uint32_t initidx;
9473 	int have_error;
9474 	scsi_sense_data_type sense_format;
9475 	ctl_ua_type ua_type;
9476 
9477 	cdb = (struct scsi_request_sense *)ctsio->cdb;
9478 
9479 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9480 
9481 	CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9482 
9483 	/*
9484 	 * Determine which sense format the user wants.
9485 	 */
9486 	if (cdb->byte2 & SRS_DESC)
9487 		sense_format = SSD_TYPE_DESC;
9488 	else
9489 		sense_format = SSD_TYPE_FIXED;
9490 
9491 	ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9492 	sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9493 	ctsio->kern_sg_entries = 0;
9494 
9495 	/*
9496 	 * struct scsi_sense_data, which is currently set to 256 bytes, is
9497 	 * larger than the largest allowed value for the length field in the
9498 	 * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9499 	 */
9500 	ctsio->residual = 0;
9501 	ctsio->kern_data_len = cdb->length;
9502 	ctsio->kern_total_len = cdb->length;
9503 
9504 	ctsio->kern_data_resid = 0;
9505 	ctsio->kern_rel_offset = 0;
9506 	ctsio->kern_sg_entries = 0;
9507 
9508 	/*
9509 	 * If we don't have a LUN, we don't have any pending sense.
9510 	 */
9511 	if (lun == NULL)
9512 		goto no_sense;
9513 
9514 	have_error = 0;
9515 	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9516 	/*
9517 	 * Check for pending sense, and then for pending unit attentions.
9518 	 * Pending sense gets returned first, then pending unit attentions.
9519 	 */
9520 	mtx_lock(&lun->lun_lock);
9521 #ifdef CTL_WITH_CA
9522 	if (ctl_is_set(lun->have_ca, initidx)) {
9523 		scsi_sense_data_type stored_format;
9524 
9525 		/*
9526 		 * Check to see which sense format was used for the stored
9527 		 * sense data.
9528 		 */
9529 		stored_format = scsi_sense_type(&lun->pending_sense[initidx]);
9530 
9531 		/*
9532 		 * If the user requested a different sense format than the
9533 		 * one we stored, then we need to convert it to the other
9534 		 * format.  If we're going from descriptor to fixed format
9535 		 * sense data, we may lose things in translation, depending
9536 		 * on what options were used.
9537 		 *
9538 		 * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9539 		 * for some reason we'll just copy it out as-is.
9540 		 */
9541 		if ((stored_format == SSD_TYPE_FIXED)
9542 		 && (sense_format == SSD_TYPE_DESC))
9543 			ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9544 			    &lun->pending_sense[initidx],
9545 			    (struct scsi_sense_data_desc *)sense_ptr);
9546 		else if ((stored_format == SSD_TYPE_DESC)
9547 		      && (sense_format == SSD_TYPE_FIXED))
9548 			ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9549 			    &lun->pending_sense[initidx],
9550 			    (struct scsi_sense_data_fixed *)sense_ptr);
9551 		else
9552 			memcpy(sense_ptr, &lun->pending_sense[initidx],
9553 			       ctl_min(sizeof(*sense_ptr),
9554 			       sizeof(lun->pending_sense[initidx])));
9555 
9556 		ctl_clear_mask(lun->have_ca, initidx);
9557 		have_error = 1;
9558 	} else
9559 #endif
9560 	{
9561 		ua_type = ctl_build_ua(lun, initidx, sense_ptr, sense_format);
9562 		if (ua_type != CTL_UA_NONE)
9563 			have_error = 1;
9564 	}
9565 	mtx_unlock(&lun->lun_lock);
9566 
9567 	/*
9568 	 * We already have a pending error, return it.
9569 	 */
9570 	if (have_error != 0) {
9571 		/*
9572 		 * We report the SCSI status as OK, since the status of the
9573 		 * request sense command itself is OK.
9574 		 * We report 0 for the sense length, because we aren't doing
9575 		 * autosense in this case.  We're reporting sense as
9576 		 * parameter data.
9577 		 */
9578 		ctl_set_success(ctsio);
9579 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9580 		ctsio->be_move_done = ctl_config_move_done;
9581 		ctl_datamove((union ctl_io *)ctsio);
9582 		return (CTL_RETVAL_COMPLETE);
9583 	}
9584 
9585 no_sense:
9586 
9587 	/*
9588 	 * No sense information to report, so we report that everything is
9589 	 * okay.
9590 	 */
9591 	ctl_set_sense_data(sense_ptr,
9592 			   lun,
9593 			   sense_format,
9594 			   /*current_error*/ 1,
9595 			   /*sense_key*/ SSD_KEY_NO_SENSE,
9596 			   /*asc*/ 0x00,
9597 			   /*ascq*/ 0x00,
9598 			   SSD_ELEM_NONE);
9599 
9600 	/*
9601 	 * We report 0 for the sense length, because we aren't doing
9602 	 * autosense in this case.  We're reporting sense as parameter data.
9603 	 */
9604 	ctl_set_success(ctsio);
9605 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9606 	ctsio->be_move_done = ctl_config_move_done;
9607 	ctl_datamove((union ctl_io *)ctsio);
9608 	return (CTL_RETVAL_COMPLETE);
9609 }
9610 
9611 int
9612 ctl_tur(struct ctl_scsiio *ctsio)
9613 {
9614 
9615 	CTL_DEBUG_PRINT(("ctl_tur\n"));
9616 
9617 	ctl_set_success(ctsio);
9618 	ctl_done((union ctl_io *)ctsio);
9619 
9620 	return (CTL_RETVAL_COMPLETE);
9621 }
9622 
9623 #ifdef notyet
9624 static int
9625 ctl_cmddt_inquiry(struct ctl_scsiio *ctsio)
9626 {
9627 
9628 }
9629 #endif
9630 
9631 static int
9632 ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9633 {
9634 	struct scsi_vpd_supported_pages *pages;
9635 	int sup_page_size;
9636 	struct ctl_lun *lun;
9637 
9638 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9639 
9640 	sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9641 	    SCSI_EVPD_NUM_SUPPORTED_PAGES;
9642 	ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9643 	pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9644 	ctsio->kern_sg_entries = 0;
9645 
9646 	if (sup_page_size < alloc_len) {
9647 		ctsio->residual = alloc_len - sup_page_size;
9648 		ctsio->kern_data_len = sup_page_size;
9649 		ctsio->kern_total_len = sup_page_size;
9650 	} else {
9651 		ctsio->residual = 0;
9652 		ctsio->kern_data_len = alloc_len;
9653 		ctsio->kern_total_len = alloc_len;
9654 	}
9655 	ctsio->kern_data_resid = 0;
9656 	ctsio->kern_rel_offset = 0;
9657 	ctsio->kern_sg_entries = 0;
9658 
9659 	/*
9660 	 * The control device is always connected.  The disk device, on the
9661 	 * other hand, may not be online all the time.  Need to change this
9662 	 * to figure out whether the disk device is actually online or not.
9663 	 */
9664 	if (lun != NULL)
9665 		pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9666 				lun->be_lun->lun_type;
9667 	else
9668 		pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9669 
9670 	pages->length = SCSI_EVPD_NUM_SUPPORTED_PAGES;
9671 	/* Supported VPD pages */
9672 	pages->page_list[0] = SVPD_SUPPORTED_PAGES;
9673 	/* Serial Number */
9674 	pages->page_list[1] = SVPD_UNIT_SERIAL_NUMBER;
9675 	/* Device Identification */
9676 	pages->page_list[2] = SVPD_DEVICE_ID;
9677 	/* Extended INQUIRY Data */
9678 	pages->page_list[3] = SVPD_EXTENDED_INQUIRY_DATA;
9679 	/* Mode Page Policy */
9680 	pages->page_list[4] = SVPD_MODE_PAGE_POLICY;
9681 	/* SCSI Ports */
9682 	pages->page_list[5] = SVPD_SCSI_PORTS;
9683 	/* Third-party Copy */
9684 	pages->page_list[6] = SVPD_SCSI_TPC;
9685 	/* Block limits */
9686 	pages->page_list[7] = SVPD_BLOCK_LIMITS;
9687 	/* Block Device Characteristics */
9688 	pages->page_list[8] = SVPD_BDC;
9689 	/* Logical Block Provisioning */
9690 	pages->page_list[9] = SVPD_LBP;
9691 
9692 	ctl_set_success(ctsio);
9693 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9694 	ctsio->be_move_done = ctl_config_move_done;
9695 	ctl_datamove((union ctl_io *)ctsio);
9696 	return (CTL_RETVAL_COMPLETE);
9697 }
9698 
9699 static int
9700 ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9701 {
9702 	struct scsi_vpd_unit_serial_number *sn_ptr;
9703 	struct ctl_lun *lun;
9704 	int data_len;
9705 
9706 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9707 
9708 	data_len = 4 + CTL_SN_LEN;
9709 	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9710 	sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9711 	if (data_len < alloc_len) {
9712 		ctsio->residual = alloc_len - data_len;
9713 		ctsio->kern_data_len = data_len;
9714 		ctsio->kern_total_len = data_len;
9715 	} else {
9716 		ctsio->residual = 0;
9717 		ctsio->kern_data_len = alloc_len;
9718 		ctsio->kern_total_len = alloc_len;
9719 	}
9720 	ctsio->kern_data_resid = 0;
9721 	ctsio->kern_rel_offset = 0;
9722 	ctsio->kern_sg_entries = 0;
9723 
9724 	/*
9725 	 * The control device is always connected.  The disk device, on the
9726 	 * other hand, may not be online all the time.  Need to change this
9727 	 * to figure out whether the disk device is actually online or not.
9728 	 */
9729 	if (lun != NULL)
9730 		sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9731 				  lun->be_lun->lun_type;
9732 	else
9733 		sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9734 
9735 	sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9736 	sn_ptr->length = CTL_SN_LEN;
9737 	/*
9738 	 * If we don't have a LUN, we just leave the serial number as
9739 	 * all spaces.
9740 	 */
9741 	if (lun != NULL) {
9742 		strncpy((char *)sn_ptr->serial_num,
9743 			(char *)lun->be_lun->serial_num, CTL_SN_LEN);
9744 	} else
9745 		memset(sn_ptr->serial_num, 0x20, CTL_SN_LEN);
9746 
9747 	ctl_set_success(ctsio);
9748 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9749 	ctsio->be_move_done = ctl_config_move_done;
9750 	ctl_datamove((union ctl_io *)ctsio);
9751 	return (CTL_RETVAL_COMPLETE);
9752 }
9753 
9754 
9755 static int
9756 ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len)
9757 {
9758 	struct scsi_vpd_extended_inquiry_data *eid_ptr;
9759 	struct ctl_lun *lun;
9760 	int data_len;
9761 
9762 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9763 
9764 	data_len = sizeof(struct scsi_vpd_extended_inquiry_data);
9765 	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9766 	eid_ptr = (struct scsi_vpd_extended_inquiry_data *)ctsio->kern_data_ptr;
9767 	ctsio->kern_sg_entries = 0;
9768 
9769 	if (data_len < alloc_len) {
9770 		ctsio->residual = alloc_len - data_len;
9771 		ctsio->kern_data_len = data_len;
9772 		ctsio->kern_total_len = data_len;
9773 	} else {
9774 		ctsio->residual = 0;
9775 		ctsio->kern_data_len = alloc_len;
9776 		ctsio->kern_total_len = alloc_len;
9777 	}
9778 	ctsio->kern_data_resid = 0;
9779 	ctsio->kern_rel_offset = 0;
9780 	ctsio->kern_sg_entries = 0;
9781 
9782 	/*
9783 	 * The control device is always connected.  The disk device, on the
9784 	 * other hand, may not be online all the time.
9785 	 */
9786 	if (lun != NULL)
9787 		eid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9788 				     lun->be_lun->lun_type;
9789 	else
9790 		eid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9791 	eid_ptr->page_code = SVPD_EXTENDED_INQUIRY_DATA;
9792 	eid_ptr->page_length = data_len - 4;
9793 	eid_ptr->flags2 = SVPD_EID_HEADSUP | SVPD_EID_ORDSUP | SVPD_EID_SIMPSUP;
9794 	eid_ptr->flags3 = SVPD_EID_V_SUP;
9795 
9796 	ctl_set_success(ctsio);
9797 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9798 	ctsio->be_move_done = ctl_config_move_done;
9799 	ctl_datamove((union ctl_io *)ctsio);
9800 	return (CTL_RETVAL_COMPLETE);
9801 }
9802 
9803 static int
9804 ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len)
9805 {
9806 	struct scsi_vpd_mode_page_policy *mpp_ptr;
9807 	struct ctl_lun *lun;
9808 	int data_len;
9809 
9810 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9811 
9812 	data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9813 	    sizeof(struct scsi_vpd_mode_page_policy_descr);
9814 
9815 	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9816 	mpp_ptr = (struct scsi_vpd_mode_page_policy *)ctsio->kern_data_ptr;
9817 	ctsio->kern_sg_entries = 0;
9818 
9819 	if (data_len < alloc_len) {
9820 		ctsio->residual = alloc_len - data_len;
9821 		ctsio->kern_data_len = data_len;
9822 		ctsio->kern_total_len = data_len;
9823 	} else {
9824 		ctsio->residual = 0;
9825 		ctsio->kern_data_len = alloc_len;
9826 		ctsio->kern_total_len = alloc_len;
9827 	}
9828 	ctsio->kern_data_resid = 0;
9829 	ctsio->kern_rel_offset = 0;
9830 	ctsio->kern_sg_entries = 0;
9831 
9832 	/*
9833 	 * The control device is always connected.  The disk device, on the
9834 	 * other hand, may not be online all the time.
9835 	 */
9836 	if (lun != NULL)
9837 		mpp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9838 				     lun->be_lun->lun_type;
9839 	else
9840 		mpp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9841 	mpp_ptr->page_code = SVPD_MODE_PAGE_POLICY;
9842 	scsi_ulto2b(data_len - 4, mpp_ptr->page_length);
9843 	mpp_ptr->descr[0].page_code = 0x3f;
9844 	mpp_ptr->descr[0].subpage_code = 0xff;
9845 	mpp_ptr->descr[0].policy = SVPD_MPP_SHARED;
9846 
9847 	ctl_set_success(ctsio);
9848 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9849 	ctsio->be_move_done = ctl_config_move_done;
9850 	ctl_datamove((union ctl_io *)ctsio);
9851 	return (CTL_RETVAL_COMPLETE);
9852 }
9853 
9854 static int
9855 ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9856 {
9857 	struct scsi_vpd_device_id *devid_ptr;
9858 	struct scsi_vpd_id_descriptor *desc;
9859 	struct ctl_softc *ctl_softc;
9860 	struct ctl_lun *lun;
9861 	struct ctl_port *port;
9862 	int data_len;
9863 	uint8_t proto;
9864 
9865 	ctl_softc = control_softc;
9866 
9867 	port = ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)];
9868 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9869 
9870 	data_len = sizeof(struct scsi_vpd_device_id) +
9871 	    sizeof(struct scsi_vpd_id_descriptor) +
9872 		sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9873 	    sizeof(struct scsi_vpd_id_descriptor) +
9874 		sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9875 	if (lun && lun->lun_devid)
9876 		data_len += lun->lun_devid->len;
9877 	if (port->port_devid)
9878 		data_len += port->port_devid->len;
9879 	if (port->target_devid)
9880 		data_len += port->target_devid->len;
9881 
9882 	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9883 	devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9884 	ctsio->kern_sg_entries = 0;
9885 
9886 	if (data_len < alloc_len) {
9887 		ctsio->residual = alloc_len - data_len;
9888 		ctsio->kern_data_len = data_len;
9889 		ctsio->kern_total_len = data_len;
9890 	} else {
9891 		ctsio->residual = 0;
9892 		ctsio->kern_data_len = alloc_len;
9893 		ctsio->kern_total_len = alloc_len;
9894 	}
9895 	ctsio->kern_data_resid = 0;
9896 	ctsio->kern_rel_offset = 0;
9897 	ctsio->kern_sg_entries = 0;
9898 
9899 	/*
9900 	 * The control device is always connected.  The disk device, on the
9901 	 * other hand, may not be online all the time.
9902 	 */
9903 	if (lun != NULL)
9904 		devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9905 				     lun->be_lun->lun_type;
9906 	else
9907 		devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9908 	devid_ptr->page_code = SVPD_DEVICE_ID;
9909 	scsi_ulto2b(data_len - 4, devid_ptr->length);
9910 
9911 	if (port->port_type == CTL_PORT_FC)
9912 		proto = SCSI_PROTO_FC << 4;
9913 	else if (port->port_type == CTL_PORT_ISCSI)
9914 		proto = SCSI_PROTO_ISCSI << 4;
9915 	else
9916 		proto = SCSI_PROTO_SPI << 4;
9917 	desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9918 
9919 	/*
9920 	 * We're using a LUN association here.  i.e., this device ID is a
9921 	 * per-LUN identifier.
9922 	 */
9923 	if (lun && lun->lun_devid) {
9924 		memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
9925 		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9926 		    lun->lun_devid->len);
9927 	}
9928 
9929 	/*
9930 	 * This is for the WWPN which is a port association.
9931 	 */
9932 	if (port->port_devid) {
9933 		memcpy(desc, port->port_devid->data, port->port_devid->len);
9934 		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9935 		    port->port_devid->len);
9936 	}
9937 
9938 	/*
9939 	 * This is for the Relative Target Port(type 4h) identifier
9940 	 */
9941 	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9942 	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9943 	    SVPD_ID_TYPE_RELTARG;
9944 	desc->length = 4;
9945 	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
9946 	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9947 	    sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9948 
9949 	/*
9950 	 * This is for the Target Port Group(type 5h) identifier
9951 	 */
9952 	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9953 	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9954 	    SVPD_ID_TYPE_TPORTGRP;
9955 	desc->length = 4;
9956 	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS + 1,
9957 	    &desc->identifier[2]);
9958 	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9959 	    sizeof(struct scsi_vpd_id_trgt_port_grp_id));
9960 
9961 	/*
9962 	 * This is for the Target identifier
9963 	 */
9964 	if (port->target_devid) {
9965 		memcpy(desc, port->target_devid->data, port->target_devid->len);
9966 	}
9967 
9968 	ctl_set_success(ctsio);
9969 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9970 	ctsio->be_move_done = ctl_config_move_done;
9971 	ctl_datamove((union ctl_io *)ctsio);
9972 	return (CTL_RETVAL_COMPLETE);
9973 }
9974 
9975 static int
9976 ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
9977 {
9978 	struct ctl_softc *softc = control_softc;
9979 	struct scsi_vpd_scsi_ports *sp;
9980 	struct scsi_vpd_port_designation *pd;
9981 	struct scsi_vpd_port_designation_cont *pdc;
9982 	struct ctl_lun *lun;
9983 	struct ctl_port *port;
9984 	int data_len, num_target_ports, iid_len, id_len, g, pg, p;
9985 	int num_target_port_groups;
9986 
9987 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9988 
9989 	if (softc->is_single)
9990 		num_target_port_groups = 1;
9991 	else
9992 		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
9993 	num_target_ports = 0;
9994 	iid_len = 0;
9995 	id_len = 0;
9996 	mtx_lock(&softc->ctl_lock);
9997 	STAILQ_FOREACH(port, &softc->port_list, links) {
9998 		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9999 			continue;
10000 		if (lun != NULL &&
10001 		    ctl_map_lun_back(port->targ_port, lun->lun) >=
10002 		    CTL_MAX_LUNS)
10003 			continue;
10004 		num_target_ports++;
10005 		if (port->init_devid)
10006 			iid_len += port->init_devid->len;
10007 		if (port->port_devid)
10008 			id_len += port->port_devid->len;
10009 	}
10010 	mtx_unlock(&softc->ctl_lock);
10011 
10012 	data_len = sizeof(struct scsi_vpd_scsi_ports) + num_target_port_groups *
10013 	    num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
10014 	     sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
10015 	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10016 	sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
10017 	ctsio->kern_sg_entries = 0;
10018 
10019 	if (data_len < alloc_len) {
10020 		ctsio->residual = alloc_len - data_len;
10021 		ctsio->kern_data_len = data_len;
10022 		ctsio->kern_total_len = data_len;
10023 	} else {
10024 		ctsio->residual = 0;
10025 		ctsio->kern_data_len = alloc_len;
10026 		ctsio->kern_total_len = alloc_len;
10027 	}
10028 	ctsio->kern_data_resid = 0;
10029 	ctsio->kern_rel_offset = 0;
10030 	ctsio->kern_sg_entries = 0;
10031 
10032 	/*
10033 	 * The control device is always connected.  The disk device, on the
10034 	 * other hand, may not be online all the time.  Need to change this
10035 	 * to figure out whether the disk device is actually online or not.
10036 	 */
10037 	if (lun != NULL)
10038 		sp->device = (SID_QUAL_LU_CONNECTED << 5) |
10039 				  lun->be_lun->lun_type;
10040 	else
10041 		sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10042 
10043 	sp->page_code = SVPD_SCSI_PORTS;
10044 	scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
10045 	    sp->page_length);
10046 	pd = &sp->design[0];
10047 
10048 	mtx_lock(&softc->ctl_lock);
10049 	pg = softc->port_offset / CTL_MAX_PORTS;
10050 	for (g = 0; g < num_target_port_groups; g++) {
10051 		STAILQ_FOREACH(port, &softc->port_list, links) {
10052 			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
10053 				continue;
10054 			if (lun != NULL &&
10055 			    ctl_map_lun_back(port->targ_port, lun->lun) >=
10056 			    CTL_MAX_LUNS)
10057 				continue;
10058 			p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
10059 			scsi_ulto2b(p, pd->relative_port_id);
10060 			if (port->init_devid && g == pg) {
10061 				iid_len = port->init_devid->len;
10062 				memcpy(pd->initiator_transportid,
10063 				    port->init_devid->data, port->init_devid->len);
10064 			} else
10065 				iid_len = 0;
10066 			scsi_ulto2b(iid_len, pd->initiator_transportid_length);
10067 			pdc = (struct scsi_vpd_port_designation_cont *)
10068 			    (&pd->initiator_transportid[iid_len]);
10069 			if (port->port_devid && g == pg) {
10070 				id_len = port->port_devid->len;
10071 				memcpy(pdc->target_port_descriptors,
10072 				    port->port_devid->data, port->port_devid->len);
10073 			} else
10074 				id_len = 0;
10075 			scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
10076 			pd = (struct scsi_vpd_port_designation *)
10077 			    ((uint8_t *)pdc->target_port_descriptors + id_len);
10078 		}
10079 	}
10080 	mtx_unlock(&softc->ctl_lock);
10081 
10082 	ctl_set_success(ctsio);
10083 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10084 	ctsio->be_move_done = ctl_config_move_done;
10085 	ctl_datamove((union ctl_io *)ctsio);
10086 	return (CTL_RETVAL_COMPLETE);
10087 }
10088 
10089 static int
10090 ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
10091 {
10092 	struct scsi_vpd_block_limits *bl_ptr;
10093 	struct ctl_lun *lun;
10094 	int bs;
10095 
10096 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10097 
10098 	ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
10099 	bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
10100 	ctsio->kern_sg_entries = 0;
10101 
10102 	if (sizeof(*bl_ptr) < alloc_len) {
10103 		ctsio->residual = alloc_len - sizeof(*bl_ptr);
10104 		ctsio->kern_data_len = sizeof(*bl_ptr);
10105 		ctsio->kern_total_len = sizeof(*bl_ptr);
10106 	} else {
10107 		ctsio->residual = 0;
10108 		ctsio->kern_data_len = alloc_len;
10109 		ctsio->kern_total_len = alloc_len;
10110 	}
10111 	ctsio->kern_data_resid = 0;
10112 	ctsio->kern_rel_offset = 0;
10113 	ctsio->kern_sg_entries = 0;
10114 
10115 	/*
10116 	 * The control device is always connected.  The disk device, on the
10117 	 * other hand, may not be online all the time.  Need to change this
10118 	 * to figure out whether the disk device is actually online or not.
10119 	 */
10120 	if (lun != NULL)
10121 		bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10122 				  lun->be_lun->lun_type;
10123 	else
10124 		bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10125 
10126 	bl_ptr->page_code = SVPD_BLOCK_LIMITS;
10127 	scsi_ulto2b(sizeof(*bl_ptr) - 4, bl_ptr->page_length);
10128 	bl_ptr->max_cmp_write_len = 0xff;
10129 	scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
10130 	if (lun != NULL) {
10131 		bs = lun->be_lun->blocksize;
10132 		scsi_ulto4b(MAXPHYS / bs, bl_ptr->opt_txfer_len);
10133 		if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10134 			scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_lba_cnt);
10135 			scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_blk_cnt);
10136 			if (lun->be_lun->pblockexp != 0) {
10137 				scsi_ulto4b((1 << lun->be_lun->pblockexp),
10138 				    bl_ptr->opt_unmap_grain);
10139 				scsi_ulto4b(0x80000000 | lun->be_lun->pblockoff,
10140 				    bl_ptr->unmap_grain_align);
10141 			}
10142 		}
10143 		scsi_ulto4b(lun->be_lun->atomicblock,
10144 		    bl_ptr->max_atomic_transfer_length);
10145 		scsi_ulto4b(0, bl_ptr->atomic_alignment);
10146 		scsi_ulto4b(0, bl_ptr->atomic_transfer_length_granularity);
10147 	}
10148 	scsi_u64to8b(UINT64_MAX, bl_ptr->max_write_same_length);
10149 
10150 	ctl_set_success(ctsio);
10151 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10152 	ctsio->be_move_done = ctl_config_move_done;
10153 	ctl_datamove((union ctl_io *)ctsio);
10154 	return (CTL_RETVAL_COMPLETE);
10155 }
10156 
10157 static int
10158 ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len)
10159 {
10160 	struct scsi_vpd_block_device_characteristics *bdc_ptr;
10161 	struct ctl_lun *lun;
10162 	const char *value;
10163 	u_int i;
10164 
10165 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10166 
10167 	ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO);
10168 	bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr;
10169 	ctsio->kern_sg_entries = 0;
10170 
10171 	if (sizeof(*bdc_ptr) < alloc_len) {
10172 		ctsio->residual = alloc_len - sizeof(*bdc_ptr);
10173 		ctsio->kern_data_len = sizeof(*bdc_ptr);
10174 		ctsio->kern_total_len = sizeof(*bdc_ptr);
10175 	} else {
10176 		ctsio->residual = 0;
10177 		ctsio->kern_data_len = alloc_len;
10178 		ctsio->kern_total_len = alloc_len;
10179 	}
10180 	ctsio->kern_data_resid = 0;
10181 	ctsio->kern_rel_offset = 0;
10182 	ctsio->kern_sg_entries = 0;
10183 
10184 	/*
10185 	 * The control device is always connected.  The disk device, on the
10186 	 * other hand, may not be online all the time.  Need to change this
10187 	 * to figure out whether the disk device is actually online or not.
10188 	 */
10189 	if (lun != NULL)
10190 		bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10191 				  lun->be_lun->lun_type;
10192 	else
10193 		bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10194 	bdc_ptr->page_code = SVPD_BDC;
10195 	scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length);
10196 	if (lun != NULL &&
10197 	    (value = ctl_get_opt(&lun->be_lun->options, "rpm")) != NULL)
10198 		i = strtol(value, NULL, 0);
10199 	else
10200 		i = CTL_DEFAULT_ROTATION_RATE;
10201 	scsi_ulto2b(i, bdc_ptr->medium_rotation_rate);
10202 	if (lun != NULL &&
10203 	    (value = ctl_get_opt(&lun->be_lun->options, "formfactor")) != NULL)
10204 		i = strtol(value, NULL, 0);
10205 	else
10206 		i = 0;
10207 	bdc_ptr->wab_wac_ff = (i & 0x0f);
10208 	bdc_ptr->flags = SVPD_FUAB | SVPD_VBULS;
10209 
10210 	ctl_set_success(ctsio);
10211 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10212 	ctsio->be_move_done = ctl_config_move_done;
10213 	ctl_datamove((union ctl_io *)ctsio);
10214 	return (CTL_RETVAL_COMPLETE);
10215 }
10216 
10217 static int
10218 ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
10219 {
10220 	struct scsi_vpd_logical_block_prov *lbp_ptr;
10221 	struct ctl_lun *lun;
10222 
10223 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10224 
10225 	ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
10226 	lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
10227 	ctsio->kern_sg_entries = 0;
10228 
10229 	if (sizeof(*lbp_ptr) < alloc_len) {
10230 		ctsio->residual = alloc_len - sizeof(*lbp_ptr);
10231 		ctsio->kern_data_len = sizeof(*lbp_ptr);
10232 		ctsio->kern_total_len = sizeof(*lbp_ptr);
10233 	} else {
10234 		ctsio->residual = 0;
10235 		ctsio->kern_data_len = alloc_len;
10236 		ctsio->kern_total_len = alloc_len;
10237 	}
10238 	ctsio->kern_data_resid = 0;
10239 	ctsio->kern_rel_offset = 0;
10240 	ctsio->kern_sg_entries = 0;
10241 
10242 	/*
10243 	 * The control device is always connected.  The disk device, on the
10244 	 * other hand, may not be online all the time.  Need to change this
10245 	 * to figure out whether the disk device is actually online or not.
10246 	 */
10247 	if (lun != NULL)
10248 		lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10249 				  lun->be_lun->lun_type;
10250 	else
10251 		lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10252 
10253 	lbp_ptr->page_code = SVPD_LBP;
10254 	scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length);
10255 	lbp_ptr->threshold_exponent = CTL_LBP_EXPONENT;
10256 	if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10257 		lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 |
10258 		    SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP;
10259 		lbp_ptr->prov_type = SVPD_LBP_THIN;
10260 	}
10261 
10262 	ctl_set_success(ctsio);
10263 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10264 	ctsio->be_move_done = ctl_config_move_done;
10265 	ctl_datamove((union ctl_io *)ctsio);
10266 	return (CTL_RETVAL_COMPLETE);
10267 }
10268 
10269 static int
10270 ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
10271 {
10272 	struct scsi_inquiry *cdb;
10273 	int alloc_len, retval;
10274 
10275 	cdb = (struct scsi_inquiry *)ctsio->cdb;
10276 
10277 	retval = CTL_RETVAL_COMPLETE;
10278 
10279 	alloc_len = scsi_2btoul(cdb->length);
10280 
10281 	switch (cdb->page_code) {
10282 	case SVPD_SUPPORTED_PAGES:
10283 		retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
10284 		break;
10285 	case SVPD_UNIT_SERIAL_NUMBER:
10286 		retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
10287 		break;
10288 	case SVPD_DEVICE_ID:
10289 		retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
10290 		break;
10291 	case SVPD_EXTENDED_INQUIRY_DATA:
10292 		retval = ctl_inquiry_evpd_eid(ctsio, alloc_len);
10293 		break;
10294 	case SVPD_MODE_PAGE_POLICY:
10295 		retval = ctl_inquiry_evpd_mpp(ctsio, alloc_len);
10296 		break;
10297 	case SVPD_SCSI_PORTS:
10298 		retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
10299 		break;
10300 	case SVPD_SCSI_TPC:
10301 		retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
10302 		break;
10303 	case SVPD_BLOCK_LIMITS:
10304 		retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
10305 		break;
10306 	case SVPD_BDC:
10307 		retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len);
10308 		break;
10309 	case SVPD_LBP:
10310 		retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
10311 		break;
10312 	default:
10313 		ctl_set_invalid_field(ctsio,
10314 				      /*sks_valid*/ 1,
10315 				      /*command*/ 1,
10316 				      /*field*/ 2,
10317 				      /*bit_valid*/ 0,
10318 				      /*bit*/ 0);
10319 		ctl_done((union ctl_io *)ctsio);
10320 		retval = CTL_RETVAL_COMPLETE;
10321 		break;
10322 	}
10323 
10324 	return (retval);
10325 }
10326 
10327 static int
10328 ctl_inquiry_std(struct ctl_scsiio *ctsio)
10329 {
10330 	struct scsi_inquiry_data *inq_ptr;
10331 	struct scsi_inquiry *cdb;
10332 	struct ctl_softc *ctl_softc;
10333 	struct ctl_lun *lun;
10334 	char *val;
10335 	uint32_t alloc_len, data_len;
10336 	ctl_port_type port_type;
10337 
10338 	ctl_softc = control_softc;
10339 
10340 	/*
10341 	 * Figure out whether we're talking to a Fibre Channel port or not.
10342 	 * We treat the ioctl front end, and any SCSI adapters, as packetized
10343 	 * SCSI front ends.
10344 	 */
10345 	port_type = ctl_softc->ctl_ports[
10346 	    ctl_port_idx(ctsio->io_hdr.nexus.targ_port)]->port_type;
10347 	if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
10348 		port_type = CTL_PORT_SCSI;
10349 
10350 	lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10351 	cdb = (struct scsi_inquiry *)ctsio->cdb;
10352 	alloc_len = scsi_2btoul(cdb->length);
10353 
10354 	/*
10355 	 * We malloc the full inquiry data size here and fill it
10356 	 * in.  If the user only asks for less, we'll give him
10357 	 * that much.
10358 	 */
10359 	data_len = offsetof(struct scsi_inquiry_data, vendor_specific1);
10360 	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10361 	inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
10362 	ctsio->kern_sg_entries = 0;
10363 	ctsio->kern_data_resid = 0;
10364 	ctsio->kern_rel_offset = 0;
10365 
10366 	if (data_len < alloc_len) {
10367 		ctsio->residual = alloc_len - data_len;
10368 		ctsio->kern_data_len = data_len;
10369 		ctsio->kern_total_len = data_len;
10370 	} else {
10371 		ctsio->residual = 0;
10372 		ctsio->kern_data_len = alloc_len;
10373 		ctsio->kern_total_len = alloc_len;
10374 	}
10375 
10376 	/*
10377 	 * If we have a LUN configured, report it as connected.  Otherwise,
10378 	 * report that it is offline or no device is supported, depending
10379 	 * on the value of inquiry_pq_no_lun.
10380 	 *
10381 	 * According to the spec (SPC-4 r34), the peripheral qualifier
10382 	 * SID_QUAL_LU_OFFLINE (001b) is used in the following scenario:
10383 	 *
10384 	 * "A peripheral device having the specified peripheral device type
10385 	 * is not connected to this logical unit. However, the device
10386 	 * server is capable of supporting the specified peripheral device
10387 	 * type on this logical unit."
10388 	 *
10389 	 * According to the same spec, the peripheral qualifier
10390 	 * SID_QUAL_BAD_LU (011b) is used in this scenario:
10391 	 *
10392 	 * "The device server is not capable of supporting a peripheral
10393 	 * device on this logical unit. For this peripheral qualifier the
10394 	 * peripheral device type shall be set to 1Fh. All other peripheral
10395 	 * device type values are reserved for this peripheral qualifier."
10396 	 *
10397 	 * Given the text, it would seem that we probably want to report that
10398 	 * the LUN is offline here.  There is no LUN connected, but we can
10399 	 * support a LUN at the given LUN number.
10400 	 *
10401 	 * In the real world, though, it sounds like things are a little
10402 	 * different:
10403 	 *
10404 	 * - Linux, when presented with a LUN with the offline peripheral
10405 	 *   qualifier, will create an sg driver instance for it.  So when
10406 	 *   you attach it to CTL, you wind up with a ton of sg driver
10407 	 *   instances.  (One for every LUN that Linux bothered to probe.)
10408 	 *   Linux does this despite the fact that it issues a REPORT LUNs
10409 	 *   to LUN 0 to get the inventory of supported LUNs.
10410 	 *
10411 	 * - There is other anecdotal evidence (from Emulex folks) about
10412 	 *   arrays that use the offline peripheral qualifier for LUNs that
10413 	 *   are on the "passive" path in an active/passive array.
10414 	 *
10415 	 * So the solution is provide a hopefully reasonable default
10416 	 * (return bad/no LUN) and allow the user to change the behavior
10417 	 * with a tunable/sysctl variable.
10418 	 */
10419 	if (lun != NULL)
10420 		inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10421 				  lun->be_lun->lun_type;
10422 	else if (ctl_softc->inquiry_pq_no_lun == 0)
10423 		inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10424 	else
10425 		inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10426 
10427 	/* RMB in byte 2 is 0 */
10428 	inq_ptr->version = SCSI_REV_SPC4;
10429 
10430 	/*
10431 	 * According to SAM-3, even if a device only supports a single
10432 	 * level of LUN addressing, it should still set the HISUP bit:
10433 	 *
10434 	 * 4.9.1 Logical unit numbers overview
10435 	 *
10436 	 * All logical unit number formats described in this standard are
10437 	 * hierarchical in structure even when only a single level in that
10438 	 * hierarchy is used. The HISUP bit shall be set to one in the
10439 	 * standard INQUIRY data (see SPC-2) when any logical unit number
10440 	 * format described in this standard is used.  Non-hierarchical
10441 	 * formats are outside the scope of this standard.
10442 	 *
10443 	 * Therefore we set the HiSup bit here.
10444 	 *
10445 	 * The reponse format is 2, per SPC-3.
10446 	 */
10447 	inq_ptr->response_format = SID_HiSup | 2;
10448 
10449 	inq_ptr->additional_length = data_len -
10450 	    (offsetof(struct scsi_inquiry_data, additional_length) + 1);
10451 	CTL_DEBUG_PRINT(("additional_length = %d\n",
10452 			 inq_ptr->additional_length));
10453 
10454 	inq_ptr->spc3_flags = SPC3_SID_3PC | SPC3_SID_TPGS_IMPLICIT;
10455 	/* 16 bit addressing */
10456 	if (port_type == CTL_PORT_SCSI)
10457 		inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10458 	/* XXX set the SID_MultiP bit here if we're actually going to
10459 	   respond on multiple ports */
10460 	inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10461 
10462 	/* 16 bit data bus, synchronous transfers */
10463 	if (port_type == CTL_PORT_SCSI)
10464 		inq_ptr->flags = SID_WBus16 | SID_Sync;
10465 	/*
10466 	 * XXX KDM do we want to support tagged queueing on the control
10467 	 * device at all?
10468 	 */
10469 	if ((lun == NULL)
10470 	 || (lun->be_lun->lun_type != T_PROCESSOR))
10471 		inq_ptr->flags |= SID_CmdQue;
10472 	/*
10473 	 * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10474 	 * We have 8 bytes for the vendor name, and 16 bytes for the device
10475 	 * name and 4 bytes for the revision.
10476 	 */
10477 	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10478 	    "vendor")) == NULL) {
10479 		strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
10480 	} else {
10481 		memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10482 		strncpy(inq_ptr->vendor, val,
10483 		    min(sizeof(inq_ptr->vendor), strlen(val)));
10484 	}
10485 	if (lun == NULL) {
10486 		strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10487 		    sizeof(inq_ptr->product));
10488 	} else if ((val = ctl_get_opt(&lun->be_lun->options, "product")) == NULL) {
10489 		switch (lun->be_lun->lun_type) {
10490 		case T_DIRECT:
10491 			strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10492 			    sizeof(inq_ptr->product));
10493 			break;
10494 		case T_PROCESSOR:
10495 			strncpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT,
10496 			    sizeof(inq_ptr->product));
10497 			break;
10498 		default:
10499 			strncpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT,
10500 			    sizeof(inq_ptr->product));
10501 			break;
10502 		}
10503 	} else {
10504 		memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10505 		strncpy(inq_ptr->product, val,
10506 		    min(sizeof(inq_ptr->product), strlen(val)));
10507 	}
10508 
10509 	/*
10510 	 * XXX make this a macro somewhere so it automatically gets
10511 	 * incremented when we make changes.
10512 	 */
10513 	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10514 	    "revision")) == NULL) {
10515 		strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10516 	} else {
10517 		memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10518 		strncpy(inq_ptr->revision, val,
10519 		    min(sizeof(inq_ptr->revision), strlen(val)));
10520 	}
10521 
10522 	/*
10523 	 * For parallel SCSI, we support double transition and single
10524 	 * transition clocking.  We also support QAS (Quick Arbitration
10525 	 * and Selection) and Information Unit transfers on both the
10526 	 * control and array devices.
10527 	 */
10528 	if (port_type == CTL_PORT_SCSI)
10529 		inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10530 				    SID_SPI_IUS;
10531 
10532 	/* SAM-5 (no version claimed) */
10533 	scsi_ulto2b(0x00A0, inq_ptr->version1);
10534 	/* SPC-4 (no version claimed) */
10535 	scsi_ulto2b(0x0460, inq_ptr->version2);
10536 	if (port_type == CTL_PORT_FC) {
10537 		/* FCP-2 ANSI INCITS.350:2003 */
10538 		scsi_ulto2b(0x0917, inq_ptr->version3);
10539 	} else if (port_type == CTL_PORT_SCSI) {
10540 		/* SPI-4 ANSI INCITS.362:200x */
10541 		scsi_ulto2b(0x0B56, inq_ptr->version3);
10542 	} else if (port_type == CTL_PORT_ISCSI) {
10543 		/* iSCSI (no version claimed) */
10544 		scsi_ulto2b(0x0960, inq_ptr->version3);
10545 	} else if (port_type == CTL_PORT_SAS) {
10546 		/* SAS (no version claimed) */
10547 		scsi_ulto2b(0x0BE0, inq_ptr->version3);
10548 	}
10549 
10550 	if (lun == NULL) {
10551 		/* SBC-4 (no version claimed) */
10552 		scsi_ulto2b(0x0600, inq_ptr->version4);
10553 	} else {
10554 		switch (lun->be_lun->lun_type) {
10555 		case T_DIRECT:
10556 			/* SBC-4 (no version claimed) */
10557 			scsi_ulto2b(0x0600, inq_ptr->version4);
10558 			break;
10559 		case T_PROCESSOR:
10560 		default:
10561 			break;
10562 		}
10563 	}
10564 
10565 	ctl_set_success(ctsio);
10566 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10567 	ctsio->be_move_done = ctl_config_move_done;
10568 	ctl_datamove((union ctl_io *)ctsio);
10569 	return (CTL_RETVAL_COMPLETE);
10570 }
10571 
10572 int
10573 ctl_inquiry(struct ctl_scsiio *ctsio)
10574 {
10575 	struct scsi_inquiry *cdb;
10576 	int retval;
10577 
10578 	CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10579 
10580 	cdb = (struct scsi_inquiry *)ctsio->cdb;
10581 	if (cdb->byte2 & SI_EVPD)
10582 		retval = ctl_inquiry_evpd(ctsio);
10583 	else if (cdb->page_code == 0)
10584 		retval = ctl_inquiry_std(ctsio);
10585 	else {
10586 		ctl_set_invalid_field(ctsio,
10587 				      /*sks_valid*/ 1,
10588 				      /*command*/ 1,
10589 				      /*field*/ 2,
10590 				      /*bit_valid*/ 0,
10591 				      /*bit*/ 0);
10592 		ctl_done((union ctl_io *)ctsio);
10593 		return (CTL_RETVAL_COMPLETE);
10594 	}
10595 
10596 	return (retval);
10597 }
10598 
10599 /*
10600  * For known CDB types, parse the LBA and length.
10601  */
10602 static int
10603 ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len)
10604 {
10605 	if (io->io_hdr.io_type != CTL_IO_SCSI)
10606 		return (1);
10607 
10608 	switch (io->scsiio.cdb[0]) {
10609 	case COMPARE_AND_WRITE: {
10610 		struct scsi_compare_and_write *cdb;
10611 
10612 		cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10613 
10614 		*lba = scsi_8btou64(cdb->addr);
10615 		*len = cdb->length;
10616 		break;
10617 	}
10618 	case READ_6:
10619 	case WRITE_6: {
10620 		struct scsi_rw_6 *cdb;
10621 
10622 		cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10623 
10624 		*lba = scsi_3btoul(cdb->addr);
10625 		/* only 5 bits are valid in the most significant address byte */
10626 		*lba &= 0x1fffff;
10627 		*len = cdb->length;
10628 		break;
10629 	}
10630 	case READ_10:
10631 	case WRITE_10: {
10632 		struct scsi_rw_10 *cdb;
10633 
10634 		cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10635 
10636 		*lba = scsi_4btoul(cdb->addr);
10637 		*len = scsi_2btoul(cdb->length);
10638 		break;
10639 	}
10640 	case WRITE_VERIFY_10: {
10641 		struct scsi_write_verify_10 *cdb;
10642 
10643 		cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10644 
10645 		*lba = scsi_4btoul(cdb->addr);
10646 		*len = scsi_2btoul(cdb->length);
10647 		break;
10648 	}
10649 	case READ_12:
10650 	case WRITE_12: {
10651 		struct scsi_rw_12 *cdb;
10652 
10653 		cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10654 
10655 		*lba = scsi_4btoul(cdb->addr);
10656 		*len = scsi_4btoul(cdb->length);
10657 		break;
10658 	}
10659 	case WRITE_VERIFY_12: {
10660 		struct scsi_write_verify_12 *cdb;
10661 
10662 		cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10663 
10664 		*lba = scsi_4btoul(cdb->addr);
10665 		*len = scsi_4btoul(cdb->length);
10666 		break;
10667 	}
10668 	case READ_16:
10669 	case WRITE_16:
10670 	case WRITE_ATOMIC_16: {
10671 		struct scsi_rw_16 *cdb;
10672 
10673 		cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10674 
10675 		*lba = scsi_8btou64(cdb->addr);
10676 		*len = scsi_4btoul(cdb->length);
10677 		break;
10678 	}
10679 	case WRITE_VERIFY_16: {
10680 		struct scsi_write_verify_16 *cdb;
10681 
10682 		cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10683 
10684 		*lba = scsi_8btou64(cdb->addr);
10685 		*len = scsi_4btoul(cdb->length);
10686 		break;
10687 	}
10688 	case WRITE_SAME_10: {
10689 		struct scsi_write_same_10 *cdb;
10690 
10691 		cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10692 
10693 		*lba = scsi_4btoul(cdb->addr);
10694 		*len = scsi_2btoul(cdb->length);
10695 		break;
10696 	}
10697 	case WRITE_SAME_16: {
10698 		struct scsi_write_same_16 *cdb;
10699 
10700 		cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10701 
10702 		*lba = scsi_8btou64(cdb->addr);
10703 		*len = scsi_4btoul(cdb->length);
10704 		break;
10705 	}
10706 	case VERIFY_10: {
10707 		struct scsi_verify_10 *cdb;
10708 
10709 		cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10710 
10711 		*lba = scsi_4btoul(cdb->addr);
10712 		*len = scsi_2btoul(cdb->length);
10713 		break;
10714 	}
10715 	case VERIFY_12: {
10716 		struct scsi_verify_12 *cdb;
10717 
10718 		cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10719 
10720 		*lba = scsi_4btoul(cdb->addr);
10721 		*len = scsi_4btoul(cdb->length);
10722 		break;
10723 	}
10724 	case VERIFY_16: {
10725 		struct scsi_verify_16 *cdb;
10726 
10727 		cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10728 
10729 		*lba = scsi_8btou64(cdb->addr);
10730 		*len = scsi_4btoul(cdb->length);
10731 		break;
10732 	}
10733 	case UNMAP: {
10734 		*lba = 0;
10735 		*len = UINT64_MAX;
10736 		break;
10737 	}
10738 	case SERVICE_ACTION_IN: {	/* GET LBA STATUS */
10739 		struct scsi_get_lba_status *cdb;
10740 
10741 		cdb = (struct scsi_get_lba_status *)io->scsiio.cdb;
10742 		*lba = scsi_8btou64(cdb->addr);
10743 		*len = UINT32_MAX;
10744 		break;
10745 	}
10746 	default:
10747 		return (1);
10748 		break; /* NOTREACHED */
10749 	}
10750 
10751 	return (0);
10752 }
10753 
10754 static ctl_action
10755 ctl_extent_check_lba(uint64_t lba1, uint64_t len1, uint64_t lba2, uint64_t len2)
10756 {
10757 	uint64_t endlba1, endlba2;
10758 
10759 	endlba1 = lba1 + len1 - 1;
10760 	endlba2 = lba2 + len2 - 1;
10761 
10762 	if ((endlba1 < lba2)
10763 	 || (endlba2 < lba1))
10764 		return (CTL_ACTION_PASS);
10765 	else
10766 		return (CTL_ACTION_BLOCK);
10767 }
10768 
10769 static int
10770 ctl_extent_check_unmap(union ctl_io *io, uint64_t lba2, uint64_t len2)
10771 {
10772 	struct ctl_ptr_len_flags *ptrlen;
10773 	struct scsi_unmap_desc *buf, *end, *range;
10774 	uint64_t lba;
10775 	uint32_t len;
10776 
10777 	/* If not UNMAP -- go other way. */
10778 	if (io->io_hdr.io_type != CTL_IO_SCSI ||
10779 	    io->scsiio.cdb[0] != UNMAP)
10780 		return (CTL_ACTION_ERROR);
10781 
10782 	/* If UNMAP without data -- block and wait for data. */
10783 	ptrlen = (struct ctl_ptr_len_flags *)
10784 	    &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
10785 	if ((io->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0 ||
10786 	    ptrlen->ptr == NULL)
10787 		return (CTL_ACTION_BLOCK);
10788 
10789 	/* UNMAP with data -- check for collision. */
10790 	buf = (struct scsi_unmap_desc *)ptrlen->ptr;
10791 	end = buf + ptrlen->len / sizeof(*buf);
10792 	for (range = buf; range < end; range++) {
10793 		lba = scsi_8btou64(range->lba);
10794 		len = scsi_4btoul(range->length);
10795 		if ((lba < lba2 + len2) && (lba + len > lba2))
10796 			return (CTL_ACTION_BLOCK);
10797 	}
10798 	return (CTL_ACTION_PASS);
10799 }
10800 
10801 static ctl_action
10802 ctl_extent_check(union ctl_io *io1, union ctl_io *io2)
10803 {
10804 	uint64_t lba1, lba2;
10805 	uint64_t len1, len2;
10806 	int retval;
10807 
10808 	if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10809 		return (CTL_ACTION_ERROR);
10810 
10811 	retval = ctl_extent_check_unmap(io2, lba1, len1);
10812 	if (retval != CTL_ACTION_ERROR)
10813 		return (retval);
10814 
10815 	if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10816 		return (CTL_ACTION_ERROR);
10817 
10818 	return (ctl_extent_check_lba(lba1, len1, lba2, len2));
10819 }
10820 
10821 static ctl_action
10822 ctl_check_for_blockage(struct ctl_lun *lun, union ctl_io *pending_io,
10823     union ctl_io *ooa_io)
10824 {
10825 	const struct ctl_cmd_entry *pending_entry, *ooa_entry;
10826 	ctl_serialize_action *serialize_row;
10827 
10828 	/*
10829 	 * The initiator attempted multiple untagged commands at the same
10830 	 * time.  Can't do that.
10831 	 */
10832 	if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10833 	 && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10834 	 && ((pending_io->io_hdr.nexus.targ_port ==
10835 	      ooa_io->io_hdr.nexus.targ_port)
10836 	  && (pending_io->io_hdr.nexus.initid.id ==
10837 	      ooa_io->io_hdr.nexus.initid.id))
10838 	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10839 		return (CTL_ACTION_OVERLAP);
10840 
10841 	/*
10842 	 * The initiator attempted to send multiple tagged commands with
10843 	 * the same ID.  (It's fine if different initiators have the same
10844 	 * tag ID.)
10845 	 *
10846 	 * Even if all of those conditions are true, we don't kill the I/O
10847 	 * if the command ahead of us has been aborted.  We won't end up
10848 	 * sending it to the FETD, and it's perfectly legal to resend a
10849 	 * command with the same tag number as long as the previous
10850 	 * instance of this tag number has been aborted somehow.
10851 	 */
10852 	if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10853 	 && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10854 	 && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
10855 	 && ((pending_io->io_hdr.nexus.targ_port ==
10856 	      ooa_io->io_hdr.nexus.targ_port)
10857 	  && (pending_io->io_hdr.nexus.initid.id ==
10858 	      ooa_io->io_hdr.nexus.initid.id))
10859 	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10860 		return (CTL_ACTION_OVERLAP_TAG);
10861 
10862 	/*
10863 	 * If we get a head of queue tag, SAM-3 says that we should
10864 	 * immediately execute it.
10865 	 *
10866 	 * What happens if this command would normally block for some other
10867 	 * reason?  e.g. a request sense with a head of queue tag
10868 	 * immediately after a write.  Normally that would block, but this
10869 	 * will result in its getting executed immediately...
10870 	 *
10871 	 * We currently return "pass" instead of "skip", so we'll end up
10872 	 * going through the rest of the queue to check for overlapped tags.
10873 	 *
10874 	 * XXX KDM check for other types of blockage first??
10875 	 */
10876 	if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10877 		return (CTL_ACTION_PASS);
10878 
10879 	/*
10880 	 * Ordered tags have to block until all items ahead of them
10881 	 * have completed.  If we get called with an ordered tag, we always
10882 	 * block, if something else is ahead of us in the queue.
10883 	 */
10884 	if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
10885 		return (CTL_ACTION_BLOCK);
10886 
10887 	/*
10888 	 * Simple tags get blocked until all head of queue and ordered tags
10889 	 * ahead of them have completed.  I'm lumping untagged commands in
10890 	 * with simple tags here.  XXX KDM is that the right thing to do?
10891 	 */
10892 	if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10893 	  || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
10894 	 && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10895 	  || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
10896 		return (CTL_ACTION_BLOCK);
10897 
10898 	pending_entry = ctl_get_cmd_entry(&pending_io->scsiio, NULL);
10899 	ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio, NULL);
10900 
10901 	serialize_row = ctl_serialize_table[ooa_entry->seridx];
10902 
10903 	switch (serialize_row[pending_entry->seridx]) {
10904 	case CTL_SER_BLOCK:
10905 		return (CTL_ACTION_BLOCK);
10906 	case CTL_SER_EXTENT:
10907 		return (ctl_extent_check(pending_io, ooa_io));
10908 	case CTL_SER_EXTENTOPT:
10909 		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
10910 		    & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
10911 			return (ctl_extent_check(pending_io, ooa_io));
10912 		/* FALLTHROUGH */
10913 	case CTL_SER_PASS:
10914 		return (CTL_ACTION_PASS);
10915 	case CTL_SER_BLOCKOPT:
10916 		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
10917 		    & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
10918 			return (CTL_ACTION_BLOCK);
10919 		return (CTL_ACTION_PASS);
10920 	case CTL_SER_SKIP:
10921 		return (CTL_ACTION_SKIP);
10922 	default:
10923 		panic("invalid serialization value %d",
10924 		      serialize_row[pending_entry->seridx]);
10925 	}
10926 
10927 	return (CTL_ACTION_ERROR);
10928 }
10929 
10930 /*
10931  * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
10932  * Assumptions:
10933  * - pending_io is generally either incoming, or on the blocked queue
10934  * - starting I/O is the I/O we want to start the check with.
10935  */
10936 static ctl_action
10937 ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
10938 	      union ctl_io *starting_io)
10939 {
10940 	union ctl_io *ooa_io;
10941 	ctl_action action;
10942 
10943 	mtx_assert(&lun->lun_lock, MA_OWNED);
10944 
10945 	/*
10946 	 * Run back along the OOA queue, starting with the current
10947 	 * blocked I/O and going through every I/O before it on the
10948 	 * queue.  If starting_io is NULL, we'll just end up returning
10949 	 * CTL_ACTION_PASS.
10950 	 */
10951 	for (ooa_io = starting_io; ooa_io != NULL;
10952 	     ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
10953 	     ooa_links)){
10954 
10955 		/*
10956 		 * This routine just checks to see whether
10957 		 * cur_blocked is blocked by ooa_io, which is ahead
10958 		 * of it in the queue.  It doesn't queue/dequeue
10959 		 * cur_blocked.
10960 		 */
10961 		action = ctl_check_for_blockage(lun, pending_io, ooa_io);
10962 		switch (action) {
10963 		case CTL_ACTION_BLOCK:
10964 		case CTL_ACTION_OVERLAP:
10965 		case CTL_ACTION_OVERLAP_TAG:
10966 		case CTL_ACTION_SKIP:
10967 		case CTL_ACTION_ERROR:
10968 			return (action);
10969 			break; /* NOTREACHED */
10970 		case CTL_ACTION_PASS:
10971 			break;
10972 		default:
10973 			panic("invalid action %d", action);
10974 			break;  /* NOTREACHED */
10975 		}
10976 	}
10977 
10978 	return (CTL_ACTION_PASS);
10979 }
10980 
10981 /*
10982  * Assumptions:
10983  * - An I/O has just completed, and has been removed from the per-LUN OOA
10984  *   queue, so some items on the blocked queue may now be unblocked.
10985  */
10986 static int
10987 ctl_check_blocked(struct ctl_lun *lun)
10988 {
10989 	union ctl_io *cur_blocked, *next_blocked;
10990 
10991 	mtx_assert(&lun->lun_lock, MA_OWNED);
10992 
10993 	/*
10994 	 * Run forward from the head of the blocked queue, checking each
10995 	 * entry against the I/Os prior to it on the OOA queue to see if
10996 	 * there is still any blockage.
10997 	 *
10998 	 * We cannot use the TAILQ_FOREACH() macro, because it can't deal
10999 	 * with our removing a variable on it while it is traversing the
11000 	 * list.
11001 	 */
11002 	for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
11003 	     cur_blocked != NULL; cur_blocked = next_blocked) {
11004 		union ctl_io *prev_ooa;
11005 		ctl_action action;
11006 
11007 		next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
11008 							  blocked_links);
11009 
11010 		prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
11011 						      ctl_ooaq, ooa_links);
11012 
11013 		/*
11014 		 * If cur_blocked happens to be the first item in the OOA
11015 		 * queue now, prev_ooa will be NULL, and the action
11016 		 * returned will just be CTL_ACTION_PASS.
11017 		 */
11018 		action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
11019 
11020 		switch (action) {
11021 		case CTL_ACTION_BLOCK:
11022 			/* Nothing to do here, still blocked */
11023 			break;
11024 		case CTL_ACTION_OVERLAP:
11025 		case CTL_ACTION_OVERLAP_TAG:
11026 			/*
11027 			 * This shouldn't happen!  In theory we've already
11028 			 * checked this command for overlap...
11029 			 */
11030 			break;
11031 		case CTL_ACTION_PASS:
11032 		case CTL_ACTION_SKIP: {
11033 			struct ctl_softc *softc;
11034 			const struct ctl_cmd_entry *entry;
11035 			int isc_retval;
11036 
11037 			/*
11038 			 * The skip case shouldn't happen, this transaction
11039 			 * should have never made it onto the blocked queue.
11040 			 */
11041 			/*
11042 			 * This I/O is no longer blocked, we can remove it
11043 			 * from the blocked queue.  Since this is a TAILQ
11044 			 * (doubly linked list), we can do O(1) removals
11045 			 * from any place on the list.
11046 			 */
11047 			TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
11048 				     blocked_links);
11049 			cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11050 
11051 			if (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC){
11052 				/*
11053 				 * Need to send IO back to original side to
11054 				 * run
11055 				 */
11056 				union ctl_ha_msg msg_info;
11057 
11058 				msg_info.hdr.original_sc =
11059 					cur_blocked->io_hdr.original_sc;
11060 				msg_info.hdr.serializing_sc = cur_blocked;
11061 				msg_info.hdr.msg_type = CTL_MSG_R2R;
11062 				if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11063 				     &msg_info, sizeof(msg_info), 0)) >
11064 				     CTL_HA_STATUS_SUCCESS) {
11065 					printf("CTL:Check Blocked error from "
11066 					       "ctl_ha_msg_send %d\n",
11067 					       isc_retval);
11068 				}
11069 				break;
11070 			}
11071 			entry = ctl_get_cmd_entry(&cur_blocked->scsiio, NULL);
11072 			softc = control_softc;
11073 
11074 			/*
11075 			 * Check this I/O for LUN state changes that may
11076 			 * have happened while this command was blocked.
11077 			 * The LUN state may have been changed by a command
11078 			 * ahead of us in the queue, so we need to re-check
11079 			 * for any states that can be caused by SCSI
11080 			 * commands.
11081 			 */
11082 			if (ctl_scsiio_lun_check(softc, lun, entry,
11083 						 &cur_blocked->scsiio) == 0) {
11084 				cur_blocked->io_hdr.flags |=
11085 				                      CTL_FLAG_IS_WAS_ON_RTR;
11086 				ctl_enqueue_rtr(cur_blocked);
11087 			} else
11088 				ctl_done(cur_blocked);
11089 			break;
11090 		}
11091 		default:
11092 			/*
11093 			 * This probably shouldn't happen -- we shouldn't
11094 			 * get CTL_ACTION_ERROR, or anything else.
11095 			 */
11096 			break;
11097 		}
11098 	}
11099 
11100 	return (CTL_RETVAL_COMPLETE);
11101 }
11102 
11103 /*
11104  * This routine (with one exception) checks LUN flags that can be set by
11105  * commands ahead of us in the OOA queue.  These flags have to be checked
11106  * when a command initially comes in, and when we pull a command off the
11107  * blocked queue and are preparing to execute it.  The reason we have to
11108  * check these flags for commands on the blocked queue is that the LUN
11109  * state may have been changed by a command ahead of us while we're on the
11110  * blocked queue.
11111  *
11112  * Ordering is somewhat important with these checks, so please pay
11113  * careful attention to the placement of any new checks.
11114  */
11115 static int
11116 ctl_scsiio_lun_check(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
11117     const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
11118 {
11119 	int retval;
11120 	uint32_t residx;
11121 
11122 	retval = 0;
11123 
11124 	mtx_assert(&lun->lun_lock, MA_OWNED);
11125 
11126 	/*
11127 	 * If this shelf is a secondary shelf controller, we have to reject
11128 	 * any media access commands.
11129 	 */
11130 	if ((ctl_softc->flags & CTL_FLAG_ACTIVE_SHELF) == 0 &&
11131 	    (entry->flags & CTL_CMD_FLAG_OK_ON_SECONDARY) == 0) {
11132 		ctl_set_lun_standby(ctsio);
11133 		retval = 1;
11134 		goto bailout;
11135 	}
11136 
11137 	if (entry->pattern & CTL_LUN_PAT_WRITE) {
11138 		if (lun->flags & CTL_LUN_READONLY) {
11139 			ctl_set_sense(ctsio, /*current_error*/ 1,
11140 			    /*sense_key*/ SSD_KEY_DATA_PROTECT,
11141 			    /*asc*/ 0x27, /*ascq*/ 0x01, SSD_ELEM_NONE);
11142 			retval = 1;
11143 			goto bailout;
11144 		}
11145 		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT]
11146 		    .eca_and_aen & SCP_SWP) != 0) {
11147 			ctl_set_sense(ctsio, /*current_error*/ 1,
11148 			    /*sense_key*/ SSD_KEY_DATA_PROTECT,
11149 			    /*asc*/ 0x27, /*ascq*/ 0x02, SSD_ELEM_NONE);
11150 			retval = 1;
11151 			goto bailout;
11152 		}
11153 	}
11154 
11155 	/*
11156 	 * Check for a reservation conflict.  If this command isn't allowed
11157 	 * even on reserved LUNs, and if this initiator isn't the one who
11158 	 * reserved us, reject the command with a reservation conflict.
11159 	 */
11160 	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
11161 	if ((lun->flags & CTL_LUN_RESERVED)
11162 	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
11163 		if (lun->res_idx != residx) {
11164 			ctl_set_reservation_conflict(ctsio);
11165 			retval = 1;
11166 			goto bailout;
11167 		}
11168 	}
11169 
11170 	if ((lun->flags & CTL_LUN_PR_RESERVED) == 0 ||
11171 	    (entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV)) {
11172 		/* No reservation or command is allowed. */;
11173 	} else if ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_WRESV) &&
11174 	    (lun->res_type == SPR_TYPE_WR_EX ||
11175 	     lun->res_type == SPR_TYPE_WR_EX_RO ||
11176 	     lun->res_type == SPR_TYPE_WR_EX_AR)) {
11177 		/* The command is allowed for Write Exclusive resv. */;
11178 	} else {
11179 		/*
11180 		 * if we aren't registered or it's a res holder type
11181 		 * reservation and this isn't the res holder then set a
11182 		 * conflict.
11183 		 */
11184 		if (ctl_get_prkey(lun, residx) == 0
11185 		 || (residx != lun->pr_res_idx && lun->res_type < 4)) {
11186 			ctl_set_reservation_conflict(ctsio);
11187 			retval = 1;
11188 			goto bailout;
11189 		}
11190 
11191 	}
11192 
11193 	if ((lun->flags & CTL_LUN_OFFLINE)
11194 	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_OFFLINE) == 0)) {
11195 		ctl_set_lun_not_ready(ctsio);
11196 		retval = 1;
11197 		goto bailout;
11198 	}
11199 
11200 	/*
11201 	 * If the LUN is stopped, see if this particular command is allowed
11202 	 * for a stopped lun.  Otherwise, reject it with 0x04,0x02.
11203 	 */
11204 	if ((lun->flags & CTL_LUN_STOPPED)
11205 	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_STOPPED) == 0)) {
11206 		/* "Logical unit not ready, initializing cmd. required" */
11207 		ctl_set_lun_stopped(ctsio);
11208 		retval = 1;
11209 		goto bailout;
11210 	}
11211 
11212 	if ((lun->flags & CTL_LUN_INOPERABLE)
11213 	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_INOPERABLE) == 0)) {
11214 		/* "Medium format corrupted" */
11215 		ctl_set_medium_format_corrupted(ctsio);
11216 		retval = 1;
11217 		goto bailout;
11218 	}
11219 
11220 bailout:
11221 	return (retval);
11222 
11223 }
11224 
11225 static void
11226 ctl_failover_io(union ctl_io *io, int have_lock)
11227 {
11228 	ctl_set_busy(&io->scsiio);
11229 	ctl_done(io);
11230 }
11231 
11232 static void
11233 ctl_failover(void)
11234 {
11235 	struct ctl_lun *lun;
11236 	struct ctl_softc *ctl_softc;
11237 	union ctl_io *next_io, *pending_io;
11238 	union ctl_io *io;
11239 	int lun_idx;
11240 
11241 	ctl_softc = control_softc;
11242 
11243 	mtx_lock(&ctl_softc->ctl_lock);
11244 	/*
11245 	 * Remove any cmds from the other SC from the rtr queue.  These
11246 	 * will obviously only be for LUNs for which we're the primary.
11247 	 * We can't send status or get/send data for these commands.
11248 	 * Since they haven't been executed yet, we can just remove them.
11249 	 * We'll either abort them or delete them below, depending on
11250 	 * which HA mode we're in.
11251 	 */
11252 #ifdef notyet
11253 	mtx_lock(&ctl_softc->queue_lock);
11254 	for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->rtr_queue);
11255 	     io != NULL; io = next_io) {
11256 		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
11257 		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
11258 			STAILQ_REMOVE(&ctl_softc->rtr_queue, &io->io_hdr,
11259 				      ctl_io_hdr, links);
11260 	}
11261 	mtx_unlock(&ctl_softc->queue_lock);
11262 #endif
11263 
11264 	for (lun_idx=0; lun_idx < ctl_softc->num_luns; lun_idx++) {
11265 		lun = ctl_softc->ctl_luns[lun_idx];
11266 		if (lun==NULL)
11267 			continue;
11268 
11269 		/*
11270 		 * Processor LUNs are primary on both sides.
11271 		 * XXX will this always be true?
11272 		 */
11273 		if (lun->be_lun->lun_type == T_PROCESSOR)
11274 			continue;
11275 
11276 		if ((lun->flags & CTL_LUN_PRIMARY_SC)
11277 		 && (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11278 			printf("FAILOVER: primary lun %d\n", lun_idx);
11279 		        /*
11280 			 * Remove all commands from the other SC. First from the
11281 			 * blocked queue then from the ooa queue. Once we have
11282 			 * removed them. Call ctl_check_blocked to see if there
11283 			 * is anything that can run.
11284 			 */
11285 			for (io = (union ctl_io *)TAILQ_FIRST(
11286 			     &lun->blocked_queue); io != NULL; io = next_io) {
11287 
11288 		        	next_io = (union ctl_io *)TAILQ_NEXT(
11289 				    &io->io_hdr, blocked_links);
11290 
11291 				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11292 					TAILQ_REMOVE(&lun->blocked_queue,
11293 						     &io->io_hdr,blocked_links);
11294 					io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11295 					TAILQ_REMOVE(&lun->ooa_queue,
11296 						     &io->io_hdr, ooa_links);
11297 
11298 					ctl_free_io(io);
11299 				}
11300 			}
11301 
11302 			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11303 	     		     io != NULL; io = next_io) {
11304 
11305 		        	next_io = (union ctl_io *)TAILQ_NEXT(
11306 				    &io->io_hdr, ooa_links);
11307 
11308 				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11309 
11310 					TAILQ_REMOVE(&lun->ooa_queue,
11311 						&io->io_hdr,
11312 					     	ooa_links);
11313 
11314 					ctl_free_io(io);
11315 				}
11316 			}
11317 			ctl_check_blocked(lun);
11318 		} else if ((lun->flags & CTL_LUN_PRIMARY_SC)
11319 			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
11320 
11321 			printf("FAILOVER: primary lun %d\n", lun_idx);
11322 			/*
11323 			 * Abort all commands from the other SC.  We can't
11324 			 * send status back for them now.  These should get
11325 			 * cleaned up when they are completed or come out
11326 			 * for a datamove operation.
11327 			 */
11328 			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11329 	     		     io != NULL; io = next_io) {
11330 		        	next_io = (union ctl_io *)TAILQ_NEXT(
11331 					&io->io_hdr, ooa_links);
11332 
11333 				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
11334 					io->io_hdr.flags |= CTL_FLAG_ABORT;
11335 			}
11336 		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11337 			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
11338 
11339 			printf("FAILOVER: secondary lun %d\n", lun_idx);
11340 
11341 			lun->flags |= CTL_LUN_PRIMARY_SC;
11342 
11343 			/*
11344 			 * We send all I/O that was sent to this controller
11345 			 * and redirected to the other side back with
11346 			 * busy status, and have the initiator retry it.
11347 			 * Figuring out how much data has been transferred,
11348 			 * etc. and picking up where we left off would be
11349 			 * very tricky.
11350 			 *
11351 			 * XXX KDM need to remove I/O from the blocked
11352 			 * queue as well!
11353 			 */
11354 			for (pending_io = (union ctl_io *)TAILQ_FIRST(
11355 			     &lun->ooa_queue); pending_io != NULL;
11356 			     pending_io = next_io) {
11357 
11358 				next_io =  (union ctl_io *)TAILQ_NEXT(
11359 					&pending_io->io_hdr, ooa_links);
11360 
11361 				pending_io->io_hdr.flags &=
11362 					~CTL_FLAG_SENT_2OTHER_SC;
11363 
11364 				if (pending_io->io_hdr.flags &
11365 				    CTL_FLAG_IO_ACTIVE) {
11366 					pending_io->io_hdr.flags |=
11367 						CTL_FLAG_FAILOVER;
11368 				} else {
11369 					ctl_set_busy(&pending_io->scsiio);
11370 					ctl_done(pending_io);
11371 				}
11372 			}
11373 
11374 			ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
11375 		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11376 			&& (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11377 			printf("FAILOVER: secondary lun %d\n", lun_idx);
11378 			/*
11379 			 * if the first io on the OOA is not on the RtR queue
11380 			 * add it.
11381 			 */
11382 			lun->flags |= CTL_LUN_PRIMARY_SC;
11383 
11384 			pending_io = (union ctl_io *)TAILQ_FIRST(
11385 			    &lun->ooa_queue);
11386 			if (pending_io==NULL) {
11387 				printf("Nothing on OOA queue\n");
11388 				continue;
11389 			}
11390 
11391 			pending_io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11392 			if ((pending_io->io_hdr.flags &
11393 			     CTL_FLAG_IS_WAS_ON_RTR) == 0) {
11394 				pending_io->io_hdr.flags |=
11395 				    CTL_FLAG_IS_WAS_ON_RTR;
11396 				ctl_enqueue_rtr(pending_io);
11397 			}
11398 #if 0
11399 			else
11400 			{
11401 				printf("Tag 0x%04x is running\n",
11402 				      pending_io->scsiio.tag_num);
11403 			}
11404 #endif
11405 
11406 			next_io = (union ctl_io *)TAILQ_NEXT(
11407 			    &pending_io->io_hdr, ooa_links);
11408 			for (pending_io=next_io; pending_io != NULL;
11409 			     pending_io = next_io) {
11410 				pending_io->io_hdr.flags &=
11411 				    ~CTL_FLAG_SENT_2OTHER_SC;
11412 				next_io = (union ctl_io *)TAILQ_NEXT(
11413 					&pending_io->io_hdr, ooa_links);
11414 				if (pending_io->io_hdr.flags &
11415 				    CTL_FLAG_IS_WAS_ON_RTR) {
11416 #if 0
11417 				        printf("Tag 0x%04x is running\n",
11418 				      		pending_io->scsiio.tag_num);
11419 #endif
11420 					continue;
11421 				}
11422 
11423 				switch (ctl_check_ooa(lun, pending_io,
11424 			            (union ctl_io *)TAILQ_PREV(
11425 				    &pending_io->io_hdr, ctl_ooaq,
11426 				    ooa_links))) {
11427 
11428 				case CTL_ACTION_BLOCK:
11429 					TAILQ_INSERT_TAIL(&lun->blocked_queue,
11430 							  &pending_io->io_hdr,
11431 							  blocked_links);
11432 					pending_io->io_hdr.flags |=
11433 					    CTL_FLAG_BLOCKED;
11434 					break;
11435 				case CTL_ACTION_PASS:
11436 				case CTL_ACTION_SKIP:
11437 					pending_io->io_hdr.flags |=
11438 					    CTL_FLAG_IS_WAS_ON_RTR;
11439 					ctl_enqueue_rtr(pending_io);
11440 					break;
11441 				case CTL_ACTION_OVERLAP:
11442 					ctl_set_overlapped_cmd(
11443 					    (struct ctl_scsiio *)pending_io);
11444 					ctl_done(pending_io);
11445 					break;
11446 				case CTL_ACTION_OVERLAP_TAG:
11447 					ctl_set_overlapped_tag(
11448 					    (struct ctl_scsiio *)pending_io,
11449 					    pending_io->scsiio.tag_num & 0xff);
11450 					ctl_done(pending_io);
11451 					break;
11452 				case CTL_ACTION_ERROR:
11453 				default:
11454 					ctl_set_internal_failure(
11455 						(struct ctl_scsiio *)pending_io,
11456 						0,  // sks_valid
11457 						0); //retry count
11458 					ctl_done(pending_io);
11459 					break;
11460 				}
11461 			}
11462 
11463 			ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
11464 		} else {
11465 			panic("Unhandled HA mode failover, LUN flags = %#x, "
11466 			      "ha_mode = #%x", lun->flags, ctl_softc->ha_mode);
11467 		}
11468 	}
11469 	ctl_pause_rtr = 0;
11470 	mtx_unlock(&ctl_softc->ctl_lock);
11471 }
11472 
11473 static int
11474 ctl_scsiio_precheck(struct ctl_softc *ctl_softc, struct ctl_scsiio *ctsio)
11475 {
11476 	struct ctl_lun *lun;
11477 	const struct ctl_cmd_entry *entry;
11478 	uint32_t initidx, targ_lun;
11479 	int retval;
11480 
11481 	retval = 0;
11482 
11483 	lun = NULL;
11484 
11485 	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11486 	if ((targ_lun < CTL_MAX_LUNS)
11487 	 && ((lun = ctl_softc->ctl_luns[targ_lun]) != NULL)) {
11488 		/*
11489 		 * If the LUN is invalid, pretend that it doesn't exist.
11490 		 * It will go away as soon as all pending I/O has been
11491 		 * completed.
11492 		 */
11493 		mtx_lock(&lun->lun_lock);
11494 		if (lun->flags & CTL_LUN_DISABLED) {
11495 			mtx_unlock(&lun->lun_lock);
11496 			lun = NULL;
11497 			ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11498 			ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11499 		} else {
11500 			ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
11501 			ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
11502 				lun->be_lun;
11503 			if (lun->be_lun->lun_type == T_PROCESSOR) {
11504 				ctsio->io_hdr.flags |= CTL_FLAG_CONTROL_DEV;
11505 			}
11506 
11507 			/*
11508 			 * Every I/O goes into the OOA queue for a
11509 			 * particular LUN, and stays there until completion.
11510 			 */
11511 			TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr,
11512 			    ooa_links);
11513 		}
11514 	} else {
11515 		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11516 		ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11517 	}
11518 
11519 	/* Get command entry and return error if it is unsuppotyed. */
11520 	entry = ctl_validate_command(ctsio);
11521 	if (entry == NULL) {
11522 		if (lun)
11523 			mtx_unlock(&lun->lun_lock);
11524 		return (retval);
11525 	}
11526 
11527 	ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11528 	ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11529 
11530 	/*
11531 	 * Check to see whether we can send this command to LUNs that don't
11532 	 * exist.  This should pretty much only be the case for inquiry
11533 	 * and request sense.  Further checks, below, really require having
11534 	 * a LUN, so we can't really check the command anymore.  Just put
11535 	 * it on the rtr queue.
11536 	 */
11537 	if (lun == NULL) {
11538 		if (entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) {
11539 			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11540 			ctl_enqueue_rtr((union ctl_io *)ctsio);
11541 			return (retval);
11542 		}
11543 
11544 		ctl_set_unsupported_lun(ctsio);
11545 		ctl_done((union ctl_io *)ctsio);
11546 		CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11547 		return (retval);
11548 	} else {
11549 		/*
11550 		 * Make sure we support this particular command on this LUN.
11551 		 * e.g., we don't support writes to the control LUN.
11552 		 */
11553 		if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11554 			mtx_unlock(&lun->lun_lock);
11555 			ctl_set_invalid_opcode(ctsio);
11556 			ctl_done((union ctl_io *)ctsio);
11557 			return (retval);
11558 		}
11559 	}
11560 
11561 	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11562 
11563 #ifdef CTL_WITH_CA
11564 	/*
11565 	 * If we've got a request sense, it'll clear the contingent
11566 	 * allegiance condition.  Otherwise, if we have a CA condition for
11567 	 * this initiator, clear it, because it sent down a command other
11568 	 * than request sense.
11569 	 */
11570 	if ((ctsio->cdb[0] != REQUEST_SENSE)
11571 	 && (ctl_is_set(lun->have_ca, initidx)))
11572 		ctl_clear_mask(lun->have_ca, initidx);
11573 #endif
11574 
11575 	/*
11576 	 * If the command has this flag set, it handles its own unit
11577 	 * attention reporting, we shouldn't do anything.  Otherwise we
11578 	 * check for any pending unit attentions, and send them back to the
11579 	 * initiator.  We only do this when a command initially comes in,
11580 	 * not when we pull it off the blocked queue.
11581 	 *
11582 	 * According to SAM-3, section 5.3.2, the order that things get
11583 	 * presented back to the host is basically unit attentions caused
11584 	 * by some sort of reset event, busy status, reservation conflicts
11585 	 * or task set full, and finally any other status.
11586 	 *
11587 	 * One issue here is that some of the unit attentions we report
11588 	 * don't fall into the "reset" category (e.g. "reported luns data
11589 	 * has changed").  So reporting it here, before the reservation
11590 	 * check, may be technically wrong.  I guess the only thing to do
11591 	 * would be to check for and report the reset events here, and then
11592 	 * check for the other unit attention types after we check for a
11593 	 * reservation conflict.
11594 	 *
11595 	 * XXX KDM need to fix this
11596 	 */
11597 	if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11598 		ctl_ua_type ua_type;
11599 		scsi_sense_data_type sense_format;
11600 
11601 		if (lun->flags & CTL_LUN_SENSE_DESC)
11602 			sense_format = SSD_TYPE_DESC;
11603 		else
11604 			sense_format = SSD_TYPE_FIXED;
11605 
11606 		ua_type = ctl_build_ua(lun, initidx, &ctsio->sense_data,
11607 		    sense_format);
11608 		if (ua_type != CTL_UA_NONE) {
11609 			mtx_unlock(&lun->lun_lock);
11610 			ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11611 			ctsio->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
11612 			ctsio->sense_len = SSD_FULL_SIZE;
11613 			ctl_done((union ctl_io *)ctsio);
11614 			return (retval);
11615 		}
11616 	}
11617 
11618 
11619 	if (ctl_scsiio_lun_check(ctl_softc, lun, entry, ctsio) != 0) {
11620 		mtx_unlock(&lun->lun_lock);
11621 		ctl_done((union ctl_io *)ctsio);
11622 		return (retval);
11623 	}
11624 
11625 	/*
11626 	 * XXX CHD this is where we want to send IO to other side if
11627 	 * this LUN is secondary on this SC. We will need to make a copy
11628 	 * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11629 	 * the copy we send as FROM_OTHER.
11630 	 * We also need to stuff the address of the original IO so we can
11631 	 * find it easily. Something similar will need be done on the other
11632 	 * side so when we are done we can find the copy.
11633 	 */
11634 	if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11635 		union ctl_ha_msg msg_info;
11636 		int isc_retval;
11637 
11638 		ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11639 
11640 		msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11641 		msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11642 #if 0
11643 		printf("1. ctsio %p\n", ctsio);
11644 #endif
11645 		msg_info.hdr.serializing_sc = NULL;
11646 		msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11647 		msg_info.scsi.tag_num = ctsio->tag_num;
11648 		msg_info.scsi.tag_type = ctsio->tag_type;
11649 		memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11650 
11651 		ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11652 
11653 		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11654 		    (void *)&msg_info, sizeof(msg_info), 0)) >
11655 		    CTL_HA_STATUS_SUCCESS) {
11656 			printf("CTL:precheck, ctl_ha_msg_send returned %d\n",
11657 			       isc_retval);
11658 			printf("CTL:opcode is %x\n", ctsio->cdb[0]);
11659 		} else {
11660 #if 0
11661 			printf("CTL:Precheck sent msg, opcode is %x\n",opcode);
11662 #endif
11663 		}
11664 
11665 		/*
11666 		 * XXX KDM this I/O is off the incoming queue, but hasn't
11667 		 * been inserted on any other queue.  We may need to come
11668 		 * up with a holding queue while we wait for serialization
11669 		 * so that we have an idea of what we're waiting for from
11670 		 * the other side.
11671 		 */
11672 		mtx_unlock(&lun->lun_lock);
11673 		return (retval);
11674 	}
11675 
11676 	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11677 			      (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11678 			      ctl_ooaq, ooa_links))) {
11679 	case CTL_ACTION_BLOCK:
11680 		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11681 		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11682 				  blocked_links);
11683 		mtx_unlock(&lun->lun_lock);
11684 		return (retval);
11685 	case CTL_ACTION_PASS:
11686 	case CTL_ACTION_SKIP:
11687 		ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11688 		mtx_unlock(&lun->lun_lock);
11689 		ctl_enqueue_rtr((union ctl_io *)ctsio);
11690 		break;
11691 	case CTL_ACTION_OVERLAP:
11692 		mtx_unlock(&lun->lun_lock);
11693 		ctl_set_overlapped_cmd(ctsio);
11694 		ctl_done((union ctl_io *)ctsio);
11695 		break;
11696 	case CTL_ACTION_OVERLAP_TAG:
11697 		mtx_unlock(&lun->lun_lock);
11698 		ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11699 		ctl_done((union ctl_io *)ctsio);
11700 		break;
11701 	case CTL_ACTION_ERROR:
11702 	default:
11703 		mtx_unlock(&lun->lun_lock);
11704 		ctl_set_internal_failure(ctsio,
11705 					 /*sks_valid*/ 0,
11706 					 /*retry_count*/ 0);
11707 		ctl_done((union ctl_io *)ctsio);
11708 		break;
11709 	}
11710 	return (retval);
11711 }
11712 
11713 const struct ctl_cmd_entry *
11714 ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa)
11715 {
11716 	const struct ctl_cmd_entry *entry;
11717 	int service_action;
11718 
11719 	entry = &ctl_cmd_table[ctsio->cdb[0]];
11720 	if (sa)
11721 		*sa = ((entry->flags & CTL_CMD_FLAG_SA5) != 0);
11722 	if (entry->flags & CTL_CMD_FLAG_SA5) {
11723 		service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11724 		entry = &((const struct ctl_cmd_entry *)
11725 		    entry->execute)[service_action];
11726 	}
11727 	return (entry);
11728 }
11729 
11730 const struct ctl_cmd_entry *
11731 ctl_validate_command(struct ctl_scsiio *ctsio)
11732 {
11733 	const struct ctl_cmd_entry *entry;
11734 	int i, sa;
11735 	uint8_t diff;
11736 
11737 	entry = ctl_get_cmd_entry(ctsio, &sa);
11738 	if (entry->execute == NULL) {
11739 		if (sa)
11740 			ctl_set_invalid_field(ctsio,
11741 					      /*sks_valid*/ 1,
11742 					      /*command*/ 1,
11743 					      /*field*/ 1,
11744 					      /*bit_valid*/ 1,
11745 					      /*bit*/ 4);
11746 		else
11747 			ctl_set_invalid_opcode(ctsio);
11748 		ctl_done((union ctl_io *)ctsio);
11749 		return (NULL);
11750 	}
11751 	KASSERT(entry->length > 0,
11752 	    ("Not defined length for command 0x%02x/0x%02x",
11753 	     ctsio->cdb[0], ctsio->cdb[1]));
11754 	for (i = 1; i < entry->length; i++) {
11755 		diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11756 		if (diff == 0)
11757 			continue;
11758 		ctl_set_invalid_field(ctsio,
11759 				      /*sks_valid*/ 1,
11760 				      /*command*/ 1,
11761 				      /*field*/ i,
11762 				      /*bit_valid*/ 1,
11763 				      /*bit*/ fls(diff) - 1);
11764 		ctl_done((union ctl_io *)ctsio);
11765 		return (NULL);
11766 	}
11767 	return (entry);
11768 }
11769 
11770 static int
11771 ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11772 {
11773 
11774 	switch (lun_type) {
11775 	case T_PROCESSOR:
11776 		if (((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0) &&
11777 		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11778 			return (0);
11779 		break;
11780 	case T_DIRECT:
11781 		if (((entry->flags & CTL_CMD_FLAG_OK_ON_SLUN) == 0) &&
11782 		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11783 			return (0);
11784 		break;
11785 	default:
11786 		return (0);
11787 	}
11788 	return (1);
11789 }
11790 
11791 static int
11792 ctl_scsiio(struct ctl_scsiio *ctsio)
11793 {
11794 	int retval;
11795 	const struct ctl_cmd_entry *entry;
11796 
11797 	retval = CTL_RETVAL_COMPLETE;
11798 
11799 	CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11800 
11801 	entry = ctl_get_cmd_entry(ctsio, NULL);
11802 
11803 	/*
11804 	 * If this I/O has been aborted, just send it straight to
11805 	 * ctl_done() without executing it.
11806 	 */
11807 	if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11808 		ctl_done((union ctl_io *)ctsio);
11809 		goto bailout;
11810 	}
11811 
11812 	/*
11813 	 * All the checks should have been handled by ctl_scsiio_precheck().
11814 	 * We should be clear now to just execute the I/O.
11815 	 */
11816 	retval = entry->execute(ctsio);
11817 
11818 bailout:
11819 	return (retval);
11820 }
11821 
11822 /*
11823  * Since we only implement one target right now, a bus reset simply resets
11824  * our single target.
11825  */
11826 static int
11827 ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io)
11828 {
11829 	return(ctl_target_reset(ctl_softc, io, CTL_UA_BUS_RESET));
11830 }
11831 
11832 static int
11833 ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
11834 		 ctl_ua_type ua_type)
11835 {
11836 	struct ctl_lun *lun;
11837 	int retval;
11838 
11839 	if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11840 		union ctl_ha_msg msg_info;
11841 
11842 		io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11843 		msg_info.hdr.nexus = io->io_hdr.nexus;
11844 		if (ua_type==CTL_UA_TARG_RESET)
11845 			msg_info.task.task_action = CTL_TASK_TARGET_RESET;
11846 		else
11847 			msg_info.task.task_action = CTL_TASK_BUS_RESET;
11848 		msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11849 		msg_info.hdr.original_sc = NULL;
11850 		msg_info.hdr.serializing_sc = NULL;
11851 		if (CTL_HA_STATUS_SUCCESS != ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11852 		    (void *)&msg_info, sizeof(msg_info), 0)) {
11853 		}
11854 	}
11855 	retval = 0;
11856 
11857 	mtx_lock(&ctl_softc->ctl_lock);
11858 	STAILQ_FOREACH(lun, &ctl_softc->lun_list, links)
11859 		retval += ctl_lun_reset(lun, io, ua_type);
11860 	mtx_unlock(&ctl_softc->ctl_lock);
11861 
11862 	return (retval);
11863 }
11864 
11865 /*
11866  * The LUN should always be set.  The I/O is optional, and is used to
11867  * distinguish between I/Os sent by this initiator, and by other
11868  * initiators.  We set unit attention for initiators other than this one.
11869  * SAM-3 is vague on this point.  It does say that a unit attention should
11870  * be established for other initiators when a LUN is reset (see section
11871  * 5.7.3), but it doesn't specifically say that the unit attention should
11872  * be established for this particular initiator when a LUN is reset.  Here
11873  * is the relevant text, from SAM-3 rev 8:
11874  *
11875  * 5.7.2 When a SCSI initiator port aborts its own tasks
11876  *
11877  * When a SCSI initiator port causes its own task(s) to be aborted, no
11878  * notification that the task(s) have been aborted shall be returned to
11879  * the SCSI initiator port other than the completion response for the
11880  * command or task management function action that caused the task(s) to
11881  * be aborted and notification(s) associated with related effects of the
11882  * action (e.g., a reset unit attention condition).
11883  *
11884  * XXX KDM for now, we're setting unit attention for all initiators.
11885  */
11886 static int
11887 ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
11888 {
11889 	union ctl_io *xio;
11890 #if 0
11891 	uint32_t initidx;
11892 #endif
11893 #ifdef CTL_WITH_CA
11894 	int i;
11895 #endif
11896 
11897 	mtx_lock(&lun->lun_lock);
11898 	/*
11899 	 * Run through the OOA queue and abort each I/O.
11900 	 */
11901 #if 0
11902 	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
11903 #endif
11904 	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11905 	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11906 		xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
11907 	}
11908 
11909 	/*
11910 	 * This version sets unit attention for every
11911 	 */
11912 #if 0
11913 	initidx = ctl_get_initindex(&io->io_hdr.nexus);
11914 	ctl_est_ua_all(lun, initidx, ua_type);
11915 #else
11916 	ctl_est_ua_all(lun, -1, ua_type);
11917 #endif
11918 
11919 	/*
11920 	 * A reset (any kind, really) clears reservations established with
11921 	 * RESERVE/RELEASE.  It does not clear reservations established
11922 	 * with PERSISTENT RESERVE OUT, but we don't support that at the
11923 	 * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
11924 	 * reservations made with the RESERVE/RELEASE commands, because
11925 	 * those commands are obsolete in SPC-3.
11926 	 */
11927 	lun->flags &= ~CTL_LUN_RESERVED;
11928 
11929 #ifdef CTL_WITH_CA
11930 	for (i = 0; i < CTL_MAX_INITIATORS; i++)
11931 		ctl_clear_mask(lun->have_ca, i);
11932 #endif
11933 	mtx_unlock(&lun->lun_lock);
11934 
11935 	return (0);
11936 }
11937 
11938 static void
11939 ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
11940     int other_sc)
11941 {
11942 	union ctl_io *xio;
11943 
11944 	mtx_assert(&lun->lun_lock, MA_OWNED);
11945 
11946 	/*
11947 	 * Run through the OOA queue and attempt to find the given I/O.
11948 	 * The target port, initiator ID, tag type and tag number have to
11949 	 * match the values that we got from the initiator.  If we have an
11950 	 * untagged command to abort, simply abort the first untagged command
11951 	 * we come to.  We only allow one untagged command at a time of course.
11952 	 */
11953 	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11954 	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11955 
11956 		if ((targ_port == UINT32_MAX ||
11957 		     targ_port == xio->io_hdr.nexus.targ_port) &&
11958 		    (init_id == UINT32_MAX ||
11959 		     init_id == xio->io_hdr.nexus.initid.id)) {
11960 			if (targ_port != xio->io_hdr.nexus.targ_port ||
11961 			    init_id != xio->io_hdr.nexus.initid.id)
11962 				xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS;
11963 			xio->io_hdr.flags |= CTL_FLAG_ABORT;
11964 			if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11965 				union ctl_ha_msg msg_info;
11966 
11967 				msg_info.hdr.nexus = xio->io_hdr.nexus;
11968 				msg_info.task.task_action = CTL_TASK_ABORT_TASK;
11969 				msg_info.task.tag_num = xio->scsiio.tag_num;
11970 				msg_info.task.tag_type = xio->scsiio.tag_type;
11971 				msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11972 				msg_info.hdr.original_sc = NULL;
11973 				msg_info.hdr.serializing_sc = NULL;
11974 				ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11975 				    (void *)&msg_info, sizeof(msg_info), 0);
11976 			}
11977 		}
11978 	}
11979 }
11980 
11981 static int
11982 ctl_abort_task_set(union ctl_io *io)
11983 {
11984 	struct ctl_softc *softc = control_softc;
11985 	struct ctl_lun *lun;
11986 	uint32_t targ_lun;
11987 
11988 	/*
11989 	 * Look up the LUN.
11990 	 */
11991 	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11992 	mtx_lock(&softc->ctl_lock);
11993 	if ((targ_lun < CTL_MAX_LUNS) && (softc->ctl_luns[targ_lun] != NULL))
11994 		lun = softc->ctl_luns[targ_lun];
11995 	else {
11996 		mtx_unlock(&softc->ctl_lock);
11997 		return (1);
11998 	}
11999 
12000 	mtx_lock(&lun->lun_lock);
12001 	mtx_unlock(&softc->ctl_lock);
12002 	if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
12003 		ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12004 		    io->io_hdr.nexus.initid.id,
12005 		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12006 	} else { /* CTL_TASK_CLEAR_TASK_SET */
12007 		ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
12008 		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12009 	}
12010 	mtx_unlock(&lun->lun_lock);
12011 	return (0);
12012 }
12013 
12014 static int
12015 ctl_i_t_nexus_reset(union ctl_io *io)
12016 {
12017 	struct ctl_softc *softc = control_softc;
12018 	struct ctl_lun *lun;
12019 	uint32_t initidx, residx;
12020 
12021 	initidx = ctl_get_initindex(&io->io_hdr.nexus);
12022 	residx = ctl_get_resindex(&io->io_hdr.nexus);
12023 	mtx_lock(&softc->ctl_lock);
12024 	STAILQ_FOREACH(lun, &softc->lun_list, links) {
12025 		mtx_lock(&lun->lun_lock);
12026 		ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12027 		    io->io_hdr.nexus.initid.id,
12028 		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12029 #ifdef CTL_WITH_CA
12030 		ctl_clear_mask(lun->have_ca, initidx);
12031 #endif
12032 		if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
12033 			lun->flags &= ~CTL_LUN_RESERVED;
12034 		ctl_est_ua(lun, initidx, CTL_UA_I_T_NEXUS_LOSS);
12035 		mtx_unlock(&lun->lun_lock);
12036 	}
12037 	mtx_unlock(&softc->ctl_lock);
12038 	return (0);
12039 }
12040 
12041 static int
12042 ctl_abort_task(union ctl_io *io)
12043 {
12044 	union ctl_io *xio;
12045 	struct ctl_lun *lun;
12046 	struct ctl_softc *ctl_softc;
12047 #if 0
12048 	struct sbuf sb;
12049 	char printbuf[128];
12050 #endif
12051 	int found;
12052 	uint32_t targ_lun;
12053 
12054 	ctl_softc = control_softc;
12055 	found = 0;
12056 
12057 	/*
12058 	 * Look up the LUN.
12059 	 */
12060 	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12061 	mtx_lock(&ctl_softc->ctl_lock);
12062 	if ((targ_lun < CTL_MAX_LUNS)
12063 	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
12064 		lun = ctl_softc->ctl_luns[targ_lun];
12065 	else {
12066 		mtx_unlock(&ctl_softc->ctl_lock);
12067 		return (1);
12068 	}
12069 
12070 #if 0
12071 	printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
12072 	       lun->lun, io->taskio.tag_num, io->taskio.tag_type);
12073 #endif
12074 
12075 	mtx_lock(&lun->lun_lock);
12076 	mtx_unlock(&ctl_softc->ctl_lock);
12077 	/*
12078 	 * Run through the OOA queue and attempt to find the given I/O.
12079 	 * The target port, initiator ID, tag type and tag number have to
12080 	 * match the values that we got from the initiator.  If we have an
12081 	 * untagged command to abort, simply abort the first untagged command
12082 	 * we come to.  We only allow one untagged command at a time of course.
12083 	 */
12084 #if 0
12085 	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
12086 #endif
12087 	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12088 	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12089 #if 0
12090 		sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
12091 
12092 		sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
12093 			    lun->lun, xio->scsiio.tag_num,
12094 			    xio->scsiio.tag_type,
12095 			    (xio->io_hdr.blocked_links.tqe_prev
12096 			    == NULL) ? "" : " BLOCKED",
12097 			    (xio->io_hdr.flags &
12098 			    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
12099 			    (xio->io_hdr.flags &
12100 			    CTL_FLAG_ABORT) ? " ABORT" : "",
12101 			    (xio->io_hdr.flags &
12102 			    CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
12103 		ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
12104 		sbuf_finish(&sb);
12105 		printf("%s\n", sbuf_data(&sb));
12106 #endif
12107 
12108 		if ((xio->io_hdr.nexus.targ_port == io->io_hdr.nexus.targ_port)
12109 		 && (xio->io_hdr.nexus.initid.id ==
12110 		     io->io_hdr.nexus.initid.id)) {
12111 			/*
12112 			 * If the abort says that the task is untagged, the
12113 			 * task in the queue must be untagged.  Otherwise,
12114 			 * we just check to see whether the tag numbers
12115 			 * match.  This is because the QLogic firmware
12116 			 * doesn't pass back the tag type in an abort
12117 			 * request.
12118 			 */
12119 #if 0
12120 			if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
12121 			  && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
12122 			 || (xio->scsiio.tag_num == io->taskio.tag_num)) {
12123 #endif
12124 			/*
12125 			 * XXX KDM we've got problems with FC, because it
12126 			 * doesn't send down a tag type with aborts.  So we
12127 			 * can only really go by the tag number...
12128 			 * This may cause problems with parallel SCSI.
12129 			 * Need to figure that out!!
12130 			 */
12131 			if (xio->scsiio.tag_num == io->taskio.tag_num) {
12132 				xio->io_hdr.flags |= CTL_FLAG_ABORT;
12133 				found = 1;
12134 				if ((io->io_hdr.flags &
12135 				     CTL_FLAG_FROM_OTHER_SC) == 0 &&
12136 				    !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12137 					union ctl_ha_msg msg_info;
12138 
12139 					io->io_hdr.flags |=
12140 					                CTL_FLAG_SENT_2OTHER_SC;
12141 					msg_info.hdr.nexus = io->io_hdr.nexus;
12142 					msg_info.task.task_action =
12143 						CTL_TASK_ABORT_TASK;
12144 					msg_info.task.tag_num =
12145 						io->taskio.tag_num;
12146 					msg_info.task.tag_type =
12147 						io->taskio.tag_type;
12148 					msg_info.hdr.msg_type =
12149 						CTL_MSG_MANAGE_TASKS;
12150 					msg_info.hdr.original_sc = NULL;
12151 					msg_info.hdr.serializing_sc = NULL;
12152 #if 0
12153 					printf("Sent Abort to other side\n");
12154 #endif
12155 					if (CTL_HA_STATUS_SUCCESS !=
12156 					        ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12157 		    				(void *)&msg_info,
12158 						sizeof(msg_info), 0)) {
12159 					}
12160 				}
12161 #if 0
12162 				printf("ctl_abort_task: found I/O to abort\n");
12163 #endif
12164 				break;
12165 			}
12166 		}
12167 	}
12168 	mtx_unlock(&lun->lun_lock);
12169 
12170 	if (found == 0) {
12171 		/*
12172 		 * This isn't really an error.  It's entirely possible for
12173 		 * the abort and command completion to cross on the wire.
12174 		 * This is more of an informative/diagnostic error.
12175 		 */
12176 #if 0
12177 		printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
12178 		       "%d:%d:%d:%d tag %d type %d\n",
12179 		       io->io_hdr.nexus.initid.id,
12180 		       io->io_hdr.nexus.targ_port,
12181 		       io->io_hdr.nexus.targ_target.id,
12182 		       io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
12183 		       io->taskio.tag_type);
12184 #endif
12185 	}
12186 	return (0);
12187 }
12188 
12189 static void
12190 ctl_run_task(union ctl_io *io)
12191 {
12192 	struct ctl_softc *ctl_softc = control_softc;
12193 	int retval = 1;
12194 	const char *task_desc;
12195 
12196 	CTL_DEBUG_PRINT(("ctl_run_task\n"));
12197 
12198 	KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
12199 	    ("ctl_run_task: Unextected io_type %d\n",
12200 	     io->io_hdr.io_type));
12201 
12202 	task_desc = ctl_scsi_task_string(&io->taskio);
12203 	if (task_desc != NULL) {
12204 #ifdef NEEDTOPORT
12205 		csevent_log(CSC_CTL | CSC_SHELF_SW |
12206 			    CTL_TASK_REPORT,
12207 			    csevent_LogType_Trace,
12208 			    csevent_Severity_Information,
12209 			    csevent_AlertLevel_Green,
12210 			    csevent_FRU_Firmware,
12211 			    csevent_FRU_Unknown,
12212 			    "CTL: received task: %s",task_desc);
12213 #endif
12214 	} else {
12215 #ifdef NEEDTOPORT
12216 		csevent_log(CSC_CTL | CSC_SHELF_SW |
12217 			    CTL_TASK_REPORT,
12218 			    csevent_LogType_Trace,
12219 			    csevent_Severity_Information,
12220 			    csevent_AlertLevel_Green,
12221 			    csevent_FRU_Firmware,
12222 			    csevent_FRU_Unknown,
12223 			    "CTL: received unknown task "
12224 			    "type: %d (%#x)",
12225 			    io->taskio.task_action,
12226 			    io->taskio.task_action);
12227 #endif
12228 	}
12229 	switch (io->taskio.task_action) {
12230 	case CTL_TASK_ABORT_TASK:
12231 		retval = ctl_abort_task(io);
12232 		break;
12233 	case CTL_TASK_ABORT_TASK_SET:
12234 	case CTL_TASK_CLEAR_TASK_SET:
12235 		retval = ctl_abort_task_set(io);
12236 		break;
12237 	case CTL_TASK_CLEAR_ACA:
12238 		break;
12239 	case CTL_TASK_I_T_NEXUS_RESET:
12240 		retval = ctl_i_t_nexus_reset(io);
12241 		break;
12242 	case CTL_TASK_LUN_RESET: {
12243 		struct ctl_lun *lun;
12244 		uint32_t targ_lun;
12245 
12246 		targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12247 		mtx_lock(&ctl_softc->ctl_lock);
12248 		if ((targ_lun < CTL_MAX_LUNS)
12249 		 && (ctl_softc->ctl_luns[targ_lun] != NULL))
12250 			lun = ctl_softc->ctl_luns[targ_lun];
12251 		else {
12252 			mtx_unlock(&ctl_softc->ctl_lock);
12253 			retval = 1;
12254 			break;
12255 		}
12256 
12257 		if (!(io->io_hdr.flags &
12258 		    CTL_FLAG_FROM_OTHER_SC)) {
12259 			union ctl_ha_msg msg_info;
12260 
12261 			io->io_hdr.flags |=
12262 				CTL_FLAG_SENT_2OTHER_SC;
12263 			msg_info.hdr.msg_type =
12264 				CTL_MSG_MANAGE_TASKS;
12265 			msg_info.hdr.nexus = io->io_hdr.nexus;
12266 			msg_info.task.task_action =
12267 				CTL_TASK_LUN_RESET;
12268 			msg_info.hdr.original_sc = NULL;
12269 			msg_info.hdr.serializing_sc = NULL;
12270 			if (CTL_HA_STATUS_SUCCESS !=
12271 			    ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12272 			    (void *)&msg_info,
12273 			    sizeof(msg_info), 0)) {
12274 			}
12275 		}
12276 
12277 		retval = ctl_lun_reset(lun, io,
12278 				       CTL_UA_LUN_RESET);
12279 		mtx_unlock(&ctl_softc->ctl_lock);
12280 		break;
12281 	}
12282 	case CTL_TASK_TARGET_RESET:
12283 		retval = ctl_target_reset(ctl_softc, io, CTL_UA_TARG_RESET);
12284 		break;
12285 	case CTL_TASK_BUS_RESET:
12286 		retval = ctl_bus_reset(ctl_softc, io);
12287 		break;
12288 	case CTL_TASK_PORT_LOGIN:
12289 		break;
12290 	case CTL_TASK_PORT_LOGOUT:
12291 		break;
12292 	default:
12293 		printf("ctl_run_task: got unknown task management event %d\n",
12294 		       io->taskio.task_action);
12295 		break;
12296 	}
12297 	if (retval == 0)
12298 		io->io_hdr.status = CTL_SUCCESS;
12299 	else
12300 		io->io_hdr.status = CTL_ERROR;
12301 	ctl_done(io);
12302 }
12303 
12304 /*
12305  * For HA operation.  Handle commands that come in from the other
12306  * controller.
12307  */
12308 static void
12309 ctl_handle_isc(union ctl_io *io)
12310 {
12311 	int free_io;
12312 	struct ctl_lun *lun;
12313 	struct ctl_softc *ctl_softc;
12314 	uint32_t targ_lun;
12315 
12316 	ctl_softc = control_softc;
12317 
12318 	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12319 	lun = ctl_softc->ctl_luns[targ_lun];
12320 
12321 	switch (io->io_hdr.msg_type) {
12322 	case CTL_MSG_SERIALIZE:
12323 		free_io = ctl_serialize_other_sc_cmd(&io->scsiio);
12324 		break;
12325 	case CTL_MSG_R2R: {
12326 		const struct ctl_cmd_entry *entry;
12327 
12328 		/*
12329 		 * This is only used in SER_ONLY mode.
12330 		 */
12331 		free_io = 0;
12332 		entry = ctl_get_cmd_entry(&io->scsiio, NULL);
12333 		mtx_lock(&lun->lun_lock);
12334 		if (ctl_scsiio_lun_check(ctl_softc, lun,
12335 		    entry, (struct ctl_scsiio *)io) != 0) {
12336 			mtx_unlock(&lun->lun_lock);
12337 			ctl_done(io);
12338 			break;
12339 		}
12340 		io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
12341 		mtx_unlock(&lun->lun_lock);
12342 		ctl_enqueue_rtr(io);
12343 		break;
12344 	}
12345 	case CTL_MSG_FINISH_IO:
12346 		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
12347 			free_io = 0;
12348 			ctl_done(io);
12349 		} else {
12350 			free_io = 1;
12351 			mtx_lock(&lun->lun_lock);
12352 			TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
12353 				     ooa_links);
12354 			ctl_check_blocked(lun);
12355 			mtx_unlock(&lun->lun_lock);
12356 		}
12357 		break;
12358 	case CTL_MSG_PERS_ACTION:
12359 		ctl_hndl_per_res_out_on_other_sc(
12360 			(union ctl_ha_msg *)&io->presio.pr_msg);
12361 		free_io = 1;
12362 		break;
12363 	case CTL_MSG_BAD_JUJU:
12364 		free_io = 0;
12365 		ctl_done(io);
12366 		break;
12367 	case CTL_MSG_DATAMOVE:
12368 		/* Only used in XFER mode */
12369 		free_io = 0;
12370 		ctl_datamove_remote(io);
12371 		break;
12372 	case CTL_MSG_DATAMOVE_DONE:
12373 		/* Only used in XFER mode */
12374 		free_io = 0;
12375 		io->scsiio.be_move_done(io);
12376 		break;
12377 	default:
12378 		free_io = 1;
12379 		printf("%s: Invalid message type %d\n",
12380 		       __func__, io->io_hdr.msg_type);
12381 		break;
12382 	}
12383 	if (free_io)
12384 		ctl_free_io(io);
12385 
12386 }
12387 
12388 
12389 /*
12390  * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12391  * there is no match.
12392  */
12393 static ctl_lun_error_pattern
12394 ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12395 {
12396 	const struct ctl_cmd_entry *entry;
12397 	ctl_lun_error_pattern filtered_pattern, pattern;
12398 
12399 	pattern = desc->error_pattern;
12400 
12401 	/*
12402 	 * XXX KDM we need more data passed into this function to match a
12403 	 * custom pattern, and we actually need to implement custom pattern
12404 	 * matching.
12405 	 */
12406 	if (pattern & CTL_LUN_PAT_CMD)
12407 		return (CTL_LUN_PAT_CMD);
12408 
12409 	if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12410 		return (CTL_LUN_PAT_ANY);
12411 
12412 	entry = ctl_get_cmd_entry(ctsio, NULL);
12413 
12414 	filtered_pattern = entry->pattern & pattern;
12415 
12416 	/*
12417 	 * If the user requested specific flags in the pattern (e.g.
12418 	 * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12419 	 * flags.
12420 	 *
12421 	 * If the user did not specify any flags, it doesn't matter whether
12422 	 * or not the command supports the flags.
12423 	 */
12424 	if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12425 	     (pattern & ~CTL_LUN_PAT_MASK))
12426 		return (CTL_LUN_PAT_NONE);
12427 
12428 	/*
12429 	 * If the user asked for a range check, see if the requested LBA
12430 	 * range overlaps with this command's LBA range.
12431 	 */
12432 	if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12433 		uint64_t lba1;
12434 		uint64_t len1;
12435 		ctl_action action;
12436 		int retval;
12437 
12438 		retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12439 		if (retval != 0)
12440 			return (CTL_LUN_PAT_NONE);
12441 
12442 		action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12443 					      desc->lba_range.len);
12444 		/*
12445 		 * A "pass" means that the LBA ranges don't overlap, so
12446 		 * this doesn't match the user's range criteria.
12447 		 */
12448 		if (action == CTL_ACTION_PASS)
12449 			return (CTL_LUN_PAT_NONE);
12450 	}
12451 
12452 	return (filtered_pattern);
12453 }
12454 
12455 static void
12456 ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12457 {
12458 	struct ctl_error_desc *desc, *desc2;
12459 
12460 	mtx_assert(&lun->lun_lock, MA_OWNED);
12461 
12462 	STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12463 		ctl_lun_error_pattern pattern;
12464 		/*
12465 		 * Check to see whether this particular command matches
12466 		 * the pattern in the descriptor.
12467 		 */
12468 		pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12469 		if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12470 			continue;
12471 
12472 		switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12473 		case CTL_LUN_INJ_ABORTED:
12474 			ctl_set_aborted(&io->scsiio);
12475 			break;
12476 		case CTL_LUN_INJ_MEDIUM_ERR:
12477 			ctl_set_medium_error(&io->scsiio);
12478 			break;
12479 		case CTL_LUN_INJ_UA:
12480 			/* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12481 			 * OCCURRED */
12482 			ctl_set_ua(&io->scsiio, 0x29, 0x00);
12483 			break;
12484 		case CTL_LUN_INJ_CUSTOM:
12485 			/*
12486 			 * We're assuming the user knows what he is doing.
12487 			 * Just copy the sense information without doing
12488 			 * checks.
12489 			 */
12490 			bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12491 			      ctl_min(sizeof(desc->custom_sense),
12492 				      sizeof(io->scsiio.sense_data)));
12493 			io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12494 			io->scsiio.sense_len = SSD_FULL_SIZE;
12495 			io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12496 			break;
12497 		case CTL_LUN_INJ_NONE:
12498 		default:
12499 			/*
12500 			 * If this is an error injection type we don't know
12501 			 * about, clear the continuous flag (if it is set)
12502 			 * so it will get deleted below.
12503 			 */
12504 			desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12505 			break;
12506 		}
12507 		/*
12508 		 * By default, each error injection action is a one-shot
12509 		 */
12510 		if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12511 			continue;
12512 
12513 		STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12514 
12515 		free(desc, M_CTL);
12516 	}
12517 }
12518 
12519 #ifdef CTL_IO_DELAY
12520 static void
12521 ctl_datamove_timer_wakeup(void *arg)
12522 {
12523 	union ctl_io *io;
12524 
12525 	io = (union ctl_io *)arg;
12526 
12527 	ctl_datamove(io);
12528 }
12529 #endif /* CTL_IO_DELAY */
12530 
12531 void
12532 ctl_datamove(union ctl_io *io)
12533 {
12534 	void (*fe_datamove)(union ctl_io *io);
12535 
12536 	mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
12537 
12538 	CTL_DEBUG_PRINT(("ctl_datamove\n"));
12539 
12540 #ifdef CTL_TIME_IO
12541 	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12542 		char str[256];
12543 		char path_str[64];
12544 		struct sbuf sb;
12545 
12546 		ctl_scsi_path_string(io, path_str, sizeof(path_str));
12547 		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12548 
12549 		sbuf_cat(&sb, path_str);
12550 		switch (io->io_hdr.io_type) {
12551 		case CTL_IO_SCSI:
12552 			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12553 			sbuf_printf(&sb, "\n");
12554 			sbuf_cat(&sb, path_str);
12555 			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12556 				    io->scsiio.tag_num, io->scsiio.tag_type);
12557 			break;
12558 		case CTL_IO_TASK:
12559 			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12560 				    "Tag Type: %d\n", io->taskio.task_action,
12561 				    io->taskio.tag_num, io->taskio.tag_type);
12562 			break;
12563 		default:
12564 			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12565 			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12566 			break;
12567 		}
12568 		sbuf_cat(&sb, path_str);
12569 		sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12570 			    (intmax_t)time_uptime - io->io_hdr.start_time);
12571 		sbuf_finish(&sb);
12572 		printf("%s", sbuf_data(&sb));
12573 	}
12574 #endif /* CTL_TIME_IO */
12575 
12576 #ifdef CTL_IO_DELAY
12577 	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12578 		struct ctl_lun *lun;
12579 
12580 		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12581 
12582 		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12583 	} else {
12584 		struct ctl_lun *lun;
12585 
12586 		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12587 		if ((lun != NULL)
12588 		 && (lun->delay_info.datamove_delay > 0)) {
12589 			struct callout *callout;
12590 
12591 			callout = (struct callout *)&io->io_hdr.timer_bytes;
12592 			callout_init(callout, /*mpsafe*/ 1);
12593 			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12594 			callout_reset(callout,
12595 				      lun->delay_info.datamove_delay * hz,
12596 				      ctl_datamove_timer_wakeup, io);
12597 			if (lun->delay_info.datamove_type ==
12598 			    CTL_DELAY_TYPE_ONESHOT)
12599 				lun->delay_info.datamove_delay = 0;
12600 			return;
12601 		}
12602 	}
12603 #endif
12604 
12605 	/*
12606 	 * This command has been aborted.  Set the port status, so we fail
12607 	 * the data move.
12608 	 */
12609 	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12610 		printf("ctl_datamove: tag 0x%04x on (%ju:%d:%ju:%d) aborted\n",
12611 		       io->scsiio.tag_num,(uintmax_t)io->io_hdr.nexus.initid.id,
12612 		       io->io_hdr.nexus.targ_port,
12613 		       (uintmax_t)io->io_hdr.nexus.targ_target.id,
12614 		       io->io_hdr.nexus.targ_lun);
12615 		io->io_hdr.port_status = 31337;
12616 		/*
12617 		 * Note that the backend, in this case, will get the
12618 		 * callback in its context.  In other cases it may get
12619 		 * called in the frontend's interrupt thread context.
12620 		 */
12621 		io->scsiio.be_move_done(io);
12622 		return;
12623 	}
12624 
12625 	/* Don't confuse frontend with zero length data move. */
12626 	if (io->scsiio.kern_data_len == 0) {
12627 		io->scsiio.be_move_done(io);
12628 		return;
12629 	}
12630 
12631 	/*
12632 	 * If we're in XFER mode and this I/O is from the other shelf
12633 	 * controller, we need to send the DMA to the other side to
12634 	 * actually transfer the data to/from the host.  In serialize only
12635 	 * mode the transfer happens below CTL and ctl_datamove() is only
12636 	 * called on the machine that originally received the I/O.
12637 	 */
12638 	if ((control_softc->ha_mode == CTL_HA_MODE_XFER)
12639 	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12640 		union ctl_ha_msg msg;
12641 		uint32_t sg_entries_sent;
12642 		int do_sg_copy;
12643 		int i;
12644 
12645 		memset(&msg, 0, sizeof(msg));
12646 		msg.hdr.msg_type = CTL_MSG_DATAMOVE;
12647 		msg.hdr.original_sc = io->io_hdr.original_sc;
12648 		msg.hdr.serializing_sc = io;
12649 		msg.hdr.nexus = io->io_hdr.nexus;
12650 		msg.dt.flags = io->io_hdr.flags;
12651 		/*
12652 		 * We convert everything into a S/G list here.  We can't
12653 		 * pass by reference, only by value between controllers.
12654 		 * So we can't pass a pointer to the S/G list, only as many
12655 		 * S/G entries as we can fit in here.  If it's possible for
12656 		 * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
12657 		 * then we need to break this up into multiple transfers.
12658 		 */
12659 		if (io->scsiio.kern_sg_entries == 0) {
12660 			msg.dt.kern_sg_entries = 1;
12661 			/*
12662 			 * If this is in cached memory, flush the cache
12663 			 * before we send the DMA request to the other
12664 			 * controller.  We want to do this in either the
12665 			 * read or the write case.  The read case is
12666 			 * straightforward.  In the write case, we want to
12667 			 * make sure nothing is in the local cache that
12668 			 * could overwrite the DMAed data.
12669 			 */
12670 			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12671 				/*
12672 				 * XXX KDM use bus_dmamap_sync() here.
12673 				 */
12674 			}
12675 
12676 			/*
12677 			 * Convert to a physical address if this is a
12678 			 * virtual address.
12679 			 */
12680 			if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
12681 				msg.dt.sg_list[0].addr =
12682 					io->scsiio.kern_data_ptr;
12683 			} else {
12684 				/*
12685 				 * XXX KDM use busdma here!
12686 				 */
12687 #if 0
12688 				msg.dt.sg_list[0].addr = (void *)
12689 					vtophys(io->scsiio.kern_data_ptr);
12690 #endif
12691 			}
12692 
12693 			msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
12694 			do_sg_copy = 0;
12695 		} else {
12696 			struct ctl_sg_entry *sgl;
12697 
12698 			do_sg_copy = 1;
12699 			msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
12700 			sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
12701 			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12702 				/*
12703 				 * XXX KDM use bus_dmamap_sync() here.
12704 				 */
12705 			}
12706 		}
12707 
12708 		msg.dt.kern_data_len = io->scsiio.kern_data_len;
12709 		msg.dt.kern_total_len = io->scsiio.kern_total_len;
12710 		msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
12711 		msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
12712 		msg.dt.sg_sequence = 0;
12713 
12714 		/*
12715 		 * Loop until we've sent all of the S/G entries.  On the
12716 		 * other end, we'll recompose these S/G entries into one
12717 		 * contiguous list before passing it to the
12718 		 */
12719 		for (sg_entries_sent = 0; sg_entries_sent <
12720 		     msg.dt.kern_sg_entries; msg.dt.sg_sequence++) {
12721 			msg.dt.cur_sg_entries = ctl_min((sizeof(msg.dt.sg_list)/
12722 				sizeof(msg.dt.sg_list[0])),
12723 				msg.dt.kern_sg_entries - sg_entries_sent);
12724 
12725 			if (do_sg_copy != 0) {
12726 				struct ctl_sg_entry *sgl;
12727 				int j;
12728 
12729 				sgl = (struct ctl_sg_entry *)
12730 					io->scsiio.kern_data_ptr;
12731 				/*
12732 				 * If this is in cached memory, flush the cache
12733 				 * before we send the DMA request to the other
12734 				 * controller.  We want to do this in either
12735 				 * the * read or the write case.  The read
12736 				 * case is straightforward.  In the write
12737 				 * case, we want to make sure nothing is
12738 				 * in the local cache that could overwrite
12739 				 * the DMAed data.
12740 				 */
12741 
12742 				for (i = sg_entries_sent, j = 0;
12743 				     i < msg.dt.cur_sg_entries; i++, j++) {
12744 					if ((io->io_hdr.flags &
12745 					     CTL_FLAG_NO_DATASYNC) == 0) {
12746 						/*
12747 						 * XXX KDM use bus_dmamap_sync()
12748 						 */
12749 					}
12750 					if ((io->io_hdr.flags &
12751 					     CTL_FLAG_BUS_ADDR) == 0) {
12752 						/*
12753 						 * XXX KDM use busdma.
12754 						 */
12755 #if 0
12756 						msg.dt.sg_list[j].addr =(void *)
12757 						       vtophys(sgl[i].addr);
12758 #endif
12759 					} else {
12760 						msg.dt.sg_list[j].addr =
12761 							sgl[i].addr;
12762 					}
12763 					msg.dt.sg_list[j].len = sgl[i].len;
12764 				}
12765 			}
12766 
12767 			sg_entries_sent += msg.dt.cur_sg_entries;
12768 			if (sg_entries_sent >= msg.dt.kern_sg_entries)
12769 				msg.dt.sg_last = 1;
12770 			else
12771 				msg.dt.sg_last = 0;
12772 
12773 			/*
12774 			 * XXX KDM drop and reacquire the lock here?
12775 			 */
12776 			if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12777 			    sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
12778 				/*
12779 				 * XXX do something here.
12780 				 */
12781 			}
12782 
12783 			msg.dt.sent_sg_entries = sg_entries_sent;
12784 		}
12785 		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12786 		if (io->io_hdr.flags & CTL_FLAG_FAILOVER)
12787 			ctl_failover_io(io, /*have_lock*/ 0);
12788 
12789 	} else {
12790 
12791 		/*
12792 		 * Lookup the fe_datamove() function for this particular
12793 		 * front end.
12794 		 */
12795 		fe_datamove =
12796 		    control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12797 
12798 		fe_datamove(io);
12799 	}
12800 }
12801 
12802 static void
12803 ctl_send_datamove_done(union ctl_io *io, int have_lock)
12804 {
12805 	union ctl_ha_msg msg;
12806 	int isc_status;
12807 
12808 	memset(&msg, 0, sizeof(msg));
12809 
12810 	msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12811 	msg.hdr.original_sc = io;
12812 	msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12813 	msg.hdr.nexus = io->io_hdr.nexus;
12814 	msg.hdr.status = io->io_hdr.status;
12815 	msg.scsi.tag_num = io->scsiio.tag_num;
12816 	msg.scsi.tag_type = io->scsiio.tag_type;
12817 	msg.scsi.scsi_status = io->scsiio.scsi_status;
12818 	memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12819 	       sizeof(io->scsiio.sense_data));
12820 	msg.scsi.sense_len = io->scsiio.sense_len;
12821 	msg.scsi.sense_residual = io->scsiio.sense_residual;
12822 	msg.scsi.fetd_status = io->io_hdr.port_status;
12823 	msg.scsi.residual = io->scsiio.residual;
12824 	io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12825 
12826 	if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12827 		ctl_failover_io(io, /*have_lock*/ have_lock);
12828 		return;
12829 	}
12830 
12831 	isc_status = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0);
12832 	if (isc_status > CTL_HA_STATUS_SUCCESS) {
12833 		/* XXX do something if this fails */
12834 	}
12835 
12836 }
12837 
12838 /*
12839  * The DMA to the remote side is done, now we need to tell the other side
12840  * we're done so it can continue with its data movement.
12841  */
12842 static void
12843 ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12844 {
12845 	union ctl_io *io;
12846 
12847 	io = rq->context;
12848 
12849 	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12850 		printf("%s: ISC DMA write failed with error %d", __func__,
12851 		       rq->ret);
12852 		ctl_set_internal_failure(&io->scsiio,
12853 					 /*sks_valid*/ 1,
12854 					 /*retry_count*/ rq->ret);
12855 	}
12856 
12857 	ctl_dt_req_free(rq);
12858 
12859 	/*
12860 	 * In this case, we had to malloc the memory locally.  Free it.
12861 	 */
12862 	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12863 		int i;
12864 		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12865 			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12866 	}
12867 	/*
12868 	 * The data is in local and remote memory, so now we need to send
12869 	 * status (good or back) back to the other side.
12870 	 */
12871 	ctl_send_datamove_done(io, /*have_lock*/ 0);
12872 }
12873 
12874 /*
12875  * We've moved the data from the host/controller into local memory.  Now we
12876  * need to push it over to the remote controller's memory.
12877  */
12878 static int
12879 ctl_datamove_remote_dm_write_cb(union ctl_io *io)
12880 {
12881 	int retval;
12882 
12883 	retval = 0;
12884 
12885 	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12886 					  ctl_datamove_remote_write_cb);
12887 
12888 	return (retval);
12889 }
12890 
12891 static void
12892 ctl_datamove_remote_write(union ctl_io *io)
12893 {
12894 	int retval;
12895 	void (*fe_datamove)(union ctl_io *io);
12896 
12897 	/*
12898 	 * - Get the data from the host/HBA into local memory.
12899 	 * - DMA memory from the local controller to the remote controller.
12900 	 * - Send status back to the remote controller.
12901 	 */
12902 
12903 	retval = ctl_datamove_remote_sgl_setup(io);
12904 	if (retval != 0)
12905 		return;
12906 
12907 	/* Switch the pointer over so the FETD knows what to do */
12908 	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12909 
12910 	/*
12911 	 * Use a custom move done callback, since we need to send completion
12912 	 * back to the other controller, not to the backend on this side.
12913 	 */
12914 	io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12915 
12916 	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12917 
12918 	fe_datamove(io);
12919 
12920 	return;
12921 
12922 }
12923 
12924 static int
12925 ctl_datamove_remote_dm_read_cb(union ctl_io *io)
12926 {
12927 #if 0
12928 	char str[256];
12929 	char path_str[64];
12930 	struct sbuf sb;
12931 #endif
12932 
12933 	/*
12934 	 * In this case, we had to malloc the memory locally.  Free it.
12935 	 */
12936 	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12937 		int i;
12938 		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12939 			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12940 	}
12941 
12942 #if 0
12943 	scsi_path_string(io, path_str, sizeof(path_str));
12944 	sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12945 	sbuf_cat(&sb, path_str);
12946 	scsi_command_string(&io->scsiio, NULL, &sb);
12947 	sbuf_printf(&sb, "\n");
12948 	sbuf_cat(&sb, path_str);
12949 	sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12950 		    io->scsiio.tag_num, io->scsiio.tag_type);
12951 	sbuf_cat(&sb, path_str);
12952 	sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
12953 		    io->io_hdr.flags, io->io_hdr.status);
12954 	sbuf_finish(&sb);
12955 	printk("%s", sbuf_data(&sb));
12956 #endif
12957 
12958 
12959 	/*
12960 	 * The read is done, now we need to send status (good or bad) back
12961 	 * to the other side.
12962 	 */
12963 	ctl_send_datamove_done(io, /*have_lock*/ 0);
12964 
12965 	return (0);
12966 }
12967 
12968 static void
12969 ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
12970 {
12971 	union ctl_io *io;
12972 	void (*fe_datamove)(union ctl_io *io);
12973 
12974 	io = rq->context;
12975 
12976 	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12977 		printf("%s: ISC DMA read failed with error %d", __func__,
12978 		       rq->ret);
12979 		ctl_set_internal_failure(&io->scsiio,
12980 					 /*sks_valid*/ 1,
12981 					 /*retry_count*/ rq->ret);
12982 	}
12983 
12984 	ctl_dt_req_free(rq);
12985 
12986 	/* Switch the pointer over so the FETD knows what to do */
12987 	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12988 
12989 	/*
12990 	 * Use a custom move done callback, since we need to send completion
12991 	 * back to the other controller, not to the backend on this side.
12992 	 */
12993 	io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
12994 
12995 	/* XXX KDM add checks like the ones in ctl_datamove? */
12996 
12997 	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12998 
12999 	fe_datamove(io);
13000 }
13001 
13002 static int
13003 ctl_datamove_remote_sgl_setup(union ctl_io *io)
13004 {
13005 	struct ctl_sg_entry *local_sglist, *remote_sglist;
13006 	struct ctl_sg_entry *local_dma_sglist, *remote_dma_sglist;
13007 	struct ctl_softc *softc;
13008 	int retval;
13009 	int i;
13010 
13011 	retval = 0;
13012 	softc = control_softc;
13013 
13014 	local_sglist = io->io_hdr.local_sglist;
13015 	local_dma_sglist = io->io_hdr.local_dma_sglist;
13016 	remote_sglist = io->io_hdr.remote_sglist;
13017 	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
13018 
13019 	if (io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) {
13020 		for (i = 0; i < io->scsiio.kern_sg_entries; i++) {
13021 			local_sglist[i].len = remote_sglist[i].len;
13022 
13023 			/*
13024 			 * XXX Detect the situation where the RS-level I/O
13025 			 * redirector on the other side has already read the
13026 			 * data off of the AOR RS on this side, and
13027 			 * transferred it to remote (mirror) memory on the
13028 			 * other side.  Since we already have the data in
13029 			 * memory here, we just need to use it.
13030 			 *
13031 			 * XXX KDM this can probably be removed once we
13032 			 * get the cache device code in and take the
13033 			 * current AOR implementation out.
13034 			 */
13035 #ifdef NEEDTOPORT
13036 			if ((remote_sglist[i].addr >=
13037 			     (void *)vtophys(softc->mirr->addr))
13038 			 && (remote_sglist[i].addr <
13039 			     ((void *)vtophys(softc->mirr->addr) +
13040 			     CacheMirrorOffset))) {
13041 				local_sglist[i].addr = remote_sglist[i].addr -
13042 					CacheMirrorOffset;
13043 				if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13044 				     CTL_FLAG_DATA_IN)
13045 					io->io_hdr.flags |= CTL_FLAG_REDIR_DONE;
13046 			} else {
13047 				local_sglist[i].addr = remote_sglist[i].addr +
13048 					CacheMirrorOffset;
13049 			}
13050 #endif
13051 #if 0
13052 			printf("%s: local %p, remote %p, len %d\n",
13053 			       __func__, local_sglist[i].addr,
13054 			       remote_sglist[i].addr, local_sglist[i].len);
13055 #endif
13056 		}
13057 	} else {
13058 		uint32_t len_to_go;
13059 
13060 		/*
13061 		 * In this case, we don't have automatically allocated
13062 		 * memory for this I/O on this controller.  This typically
13063 		 * happens with internal CTL I/O -- e.g. inquiry, mode
13064 		 * sense, etc.  Anything coming from RAIDCore will have
13065 		 * a mirror area available.
13066 		 */
13067 		len_to_go = io->scsiio.kern_data_len;
13068 
13069 		/*
13070 		 * Clear the no datasync flag, we have to use malloced
13071 		 * buffers.
13072 		 */
13073 		io->io_hdr.flags &= ~CTL_FLAG_NO_DATASYNC;
13074 
13075 		/*
13076 		 * The difficult thing here is that the size of the various
13077 		 * S/G segments may be different than the size from the
13078 		 * remote controller.  That'll make it harder when DMAing
13079 		 * the data back to the other side.
13080 		 */
13081 		for (i = 0; (i < sizeof(io->io_hdr.remote_sglist) /
13082 		     sizeof(io->io_hdr.remote_sglist[0])) &&
13083 		     (len_to_go > 0); i++) {
13084 			local_sglist[i].len = ctl_min(len_to_go, 131072);
13085 			CTL_SIZE_8B(local_dma_sglist[i].len,
13086 				    local_sglist[i].len);
13087 			local_sglist[i].addr =
13088 				malloc(local_dma_sglist[i].len, M_CTL,M_WAITOK);
13089 
13090 			local_dma_sglist[i].addr = local_sglist[i].addr;
13091 
13092 			if (local_sglist[i].addr == NULL) {
13093 				int j;
13094 
13095 				printf("malloc failed for %zd bytes!",
13096 				       local_dma_sglist[i].len);
13097 				for (j = 0; j < i; j++) {
13098 					free(local_sglist[j].addr, M_CTL);
13099 				}
13100 				ctl_set_internal_failure(&io->scsiio,
13101 							 /*sks_valid*/ 1,
13102 							 /*retry_count*/ 4857);
13103 				retval = 1;
13104 				goto bailout_error;
13105 
13106 			}
13107 			/* XXX KDM do we need a sync here? */
13108 
13109 			len_to_go -= local_sglist[i].len;
13110 		}
13111 		/*
13112 		 * Reset the number of S/G entries accordingly.  The
13113 		 * original number of S/G entries is available in
13114 		 * rem_sg_entries.
13115 		 */
13116 		io->scsiio.kern_sg_entries = i;
13117 
13118 #if 0
13119 		printf("%s: kern_sg_entries = %d\n", __func__,
13120 		       io->scsiio.kern_sg_entries);
13121 		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13122 			printf("%s: sg[%d] = %p, %d (DMA: %d)\n", __func__, i,
13123 			       local_sglist[i].addr, local_sglist[i].len,
13124 			       local_dma_sglist[i].len);
13125 #endif
13126 	}
13127 
13128 
13129 	return (retval);
13130 
13131 bailout_error:
13132 
13133 	ctl_send_datamove_done(io, /*have_lock*/ 0);
13134 
13135 	return (retval);
13136 }
13137 
13138 static int
13139 ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
13140 			 ctl_ha_dt_cb callback)
13141 {
13142 	struct ctl_ha_dt_req *rq;
13143 	struct ctl_sg_entry *remote_sglist, *local_sglist;
13144 	struct ctl_sg_entry *remote_dma_sglist, *local_dma_sglist;
13145 	uint32_t local_used, remote_used, total_used;
13146 	int retval;
13147 	int i, j;
13148 
13149 	retval = 0;
13150 
13151 	rq = ctl_dt_req_alloc();
13152 
13153 	/*
13154 	 * If we failed to allocate the request, and if the DMA didn't fail
13155 	 * anyway, set busy status.  This is just a resource allocation
13156 	 * failure.
13157 	 */
13158 	if ((rq == NULL)
13159 	 && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE))
13160 		ctl_set_busy(&io->scsiio);
13161 
13162 	if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) {
13163 
13164 		if (rq != NULL)
13165 			ctl_dt_req_free(rq);
13166 
13167 		/*
13168 		 * The data move failed.  We need to return status back
13169 		 * to the other controller.  No point in trying to DMA
13170 		 * data to the remote controller.
13171 		 */
13172 
13173 		ctl_send_datamove_done(io, /*have_lock*/ 0);
13174 
13175 		retval = 1;
13176 
13177 		goto bailout;
13178 	}
13179 
13180 	local_sglist = io->io_hdr.local_sglist;
13181 	local_dma_sglist = io->io_hdr.local_dma_sglist;
13182 	remote_sglist = io->io_hdr.remote_sglist;
13183 	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
13184 	local_used = 0;
13185 	remote_used = 0;
13186 	total_used = 0;
13187 
13188 	if (io->io_hdr.flags & CTL_FLAG_REDIR_DONE) {
13189 		rq->ret = CTL_HA_STATUS_SUCCESS;
13190 		rq->context = io;
13191 		callback(rq);
13192 		goto bailout;
13193 	}
13194 
13195 	/*
13196 	 * Pull/push the data over the wire from/to the other controller.
13197 	 * This takes into account the possibility that the local and
13198 	 * remote sglists may not be identical in terms of the size of
13199 	 * the elements and the number of elements.
13200 	 *
13201 	 * One fundamental assumption here is that the length allocated for
13202 	 * both the local and remote sglists is identical.  Otherwise, we've
13203 	 * essentially got a coding error of some sort.
13204 	 */
13205 	for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
13206 		int isc_ret;
13207 		uint32_t cur_len, dma_length;
13208 		uint8_t *tmp_ptr;
13209 
13210 		rq->id = CTL_HA_DATA_CTL;
13211 		rq->command = command;
13212 		rq->context = io;
13213 
13214 		/*
13215 		 * Both pointers should be aligned.  But it is possible
13216 		 * that the allocation length is not.  They should both
13217 		 * also have enough slack left over at the end, though,
13218 		 * to round up to the next 8 byte boundary.
13219 		 */
13220 		cur_len = ctl_min(local_sglist[i].len - local_used,
13221 				  remote_sglist[j].len - remote_used);
13222 
13223 		/*
13224 		 * In this case, we have a size issue and need to decrease
13225 		 * the size, except in the case where we actually have less
13226 		 * than 8 bytes left.  In that case, we need to increase
13227 		 * the DMA length to get the last bit.
13228 		 */
13229 		if ((cur_len & 0x7) != 0) {
13230 			if (cur_len > 0x7) {
13231 				cur_len = cur_len - (cur_len & 0x7);
13232 				dma_length = cur_len;
13233 			} else {
13234 				CTL_SIZE_8B(dma_length, cur_len);
13235 			}
13236 
13237 		} else
13238 			dma_length = cur_len;
13239 
13240 		/*
13241 		 * If we had to allocate memory for this I/O, instead of using
13242 		 * the non-cached mirror memory, we'll need to flush the cache
13243 		 * before trying to DMA to the other controller.
13244 		 *
13245 		 * We could end up doing this multiple times for the same
13246 		 * segment if we have a larger local segment than remote
13247 		 * segment.  That shouldn't be an issue.
13248 		 */
13249 		if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
13250 			/*
13251 			 * XXX KDM use bus_dmamap_sync() here.
13252 			 */
13253 		}
13254 
13255 		rq->size = dma_length;
13256 
13257 		tmp_ptr = (uint8_t *)local_sglist[i].addr;
13258 		tmp_ptr += local_used;
13259 
13260 		/* Use physical addresses when talking to ISC hardware */
13261 		if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
13262 			/* XXX KDM use busdma */
13263 #if 0
13264 			rq->local = vtophys(tmp_ptr);
13265 #endif
13266 		} else
13267 			rq->local = tmp_ptr;
13268 
13269 		tmp_ptr = (uint8_t *)remote_sglist[j].addr;
13270 		tmp_ptr += remote_used;
13271 		rq->remote = tmp_ptr;
13272 
13273 		rq->callback = NULL;
13274 
13275 		local_used += cur_len;
13276 		if (local_used >= local_sglist[i].len) {
13277 			i++;
13278 			local_used = 0;
13279 		}
13280 
13281 		remote_used += cur_len;
13282 		if (remote_used >= remote_sglist[j].len) {
13283 			j++;
13284 			remote_used = 0;
13285 		}
13286 		total_used += cur_len;
13287 
13288 		if (total_used >= io->scsiio.kern_data_len)
13289 			rq->callback = callback;
13290 
13291 		if ((rq->size & 0x7) != 0) {
13292 			printf("%s: warning: size %d is not on 8b boundary\n",
13293 			       __func__, rq->size);
13294 		}
13295 		if (((uintptr_t)rq->local & 0x7) != 0) {
13296 			printf("%s: warning: local %p not on 8b boundary\n",
13297 			       __func__, rq->local);
13298 		}
13299 		if (((uintptr_t)rq->remote & 0x7) != 0) {
13300 			printf("%s: warning: remote %p not on 8b boundary\n",
13301 			       __func__, rq->local);
13302 		}
13303 #if 0
13304 		printf("%s: %s: local %#x remote %#x size %d\n", __func__,
13305 		       (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
13306 		       rq->local, rq->remote, rq->size);
13307 #endif
13308 
13309 		isc_ret = ctl_dt_single(rq);
13310 		if (isc_ret == CTL_HA_STATUS_WAIT)
13311 			continue;
13312 
13313 		if (isc_ret == CTL_HA_STATUS_DISCONNECT) {
13314 			rq->ret = CTL_HA_STATUS_SUCCESS;
13315 		} else {
13316 			rq->ret = isc_ret;
13317 		}
13318 		callback(rq);
13319 		goto bailout;
13320 	}
13321 
13322 bailout:
13323 	return (retval);
13324 
13325 }
13326 
13327 static void
13328 ctl_datamove_remote_read(union ctl_io *io)
13329 {
13330 	int retval;
13331 	int i;
13332 
13333 	/*
13334 	 * This will send an error to the other controller in the case of a
13335 	 * failure.
13336 	 */
13337 	retval = ctl_datamove_remote_sgl_setup(io);
13338 	if (retval != 0)
13339 		return;
13340 
13341 	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
13342 					  ctl_datamove_remote_read_cb);
13343 	if ((retval != 0)
13344 	 && ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0)) {
13345 		/*
13346 		 * Make sure we free memory if there was an error..  The
13347 		 * ctl_datamove_remote_xfer() function will send the
13348 		 * datamove done message, or call the callback with an
13349 		 * error if there is a problem.
13350 		 */
13351 		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13352 			free(io->io_hdr.local_sglist[i].addr, M_CTL);
13353 	}
13354 
13355 	return;
13356 }
13357 
13358 /*
13359  * Process a datamove request from the other controller.  This is used for
13360  * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
13361  * first.  Once that is complete, the data gets DMAed into the remote
13362  * controller's memory.  For reads, we DMA from the remote controller's
13363  * memory into our memory first, and then move it out to the FETD.
13364  */
13365 static void
13366 ctl_datamove_remote(union ctl_io *io)
13367 {
13368 	struct ctl_softc *softc;
13369 
13370 	softc = control_softc;
13371 
13372 	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
13373 
13374 	/*
13375 	 * Note that we look for an aborted I/O here, but don't do some of
13376 	 * the other checks that ctl_datamove() normally does.
13377 	 * We don't need to run the datamove delay code, since that should
13378 	 * have been done if need be on the other controller.
13379 	 */
13380 	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
13381 		printf("%s: tag 0x%04x on (%d:%d:%d:%d) aborted\n", __func__,
13382 		       io->scsiio.tag_num, io->io_hdr.nexus.initid.id,
13383 		       io->io_hdr.nexus.targ_port,
13384 		       io->io_hdr.nexus.targ_target.id,
13385 		       io->io_hdr.nexus.targ_lun);
13386 		io->io_hdr.port_status = 31338;
13387 		ctl_send_datamove_done(io, /*have_lock*/ 0);
13388 		return;
13389 	}
13390 
13391 	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT) {
13392 		ctl_datamove_remote_write(io);
13393 	} else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN){
13394 		ctl_datamove_remote_read(io);
13395 	} else {
13396 		union ctl_ha_msg msg;
13397 		struct scsi_sense_data *sense;
13398 		uint8_t sks[3];
13399 		int retry_count;
13400 
13401 		memset(&msg, 0, sizeof(msg));
13402 
13403 		msg.hdr.msg_type = CTL_MSG_BAD_JUJU;
13404 		msg.hdr.status = CTL_SCSI_ERROR;
13405 		msg.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
13406 
13407 		retry_count = 4243;
13408 
13409 		sense = &msg.scsi.sense_data;
13410 		sks[0] = SSD_SCS_VALID;
13411 		sks[1] = (retry_count >> 8) & 0xff;
13412 		sks[2] = retry_count & 0xff;
13413 
13414 		/* "Internal target failure" */
13415 		scsi_set_sense_data(sense,
13416 				    /*sense_format*/ SSD_TYPE_NONE,
13417 				    /*current_error*/ 1,
13418 				    /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
13419 				    /*asc*/ 0x44,
13420 				    /*ascq*/ 0x00,
13421 				    /*type*/ SSD_ELEM_SKS,
13422 				    /*size*/ sizeof(sks),
13423 				    /*data*/ sks,
13424 				    SSD_ELEM_NONE);
13425 
13426 		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
13427 		if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
13428 			ctl_failover_io(io, /*have_lock*/ 1);
13429 			return;
13430 		}
13431 
13432 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0) >
13433 		    CTL_HA_STATUS_SUCCESS) {
13434 			/* XXX KDM what to do if this fails? */
13435 		}
13436 		return;
13437 	}
13438 
13439 }
13440 
13441 static int
13442 ctl_process_done(union ctl_io *io)
13443 {
13444 	struct ctl_lun *lun;
13445 	struct ctl_softc *ctl_softc = control_softc;
13446 	void (*fe_done)(union ctl_io *io);
13447 	uint32_t targ_port = ctl_port_idx(io->io_hdr.nexus.targ_port);
13448 
13449 	CTL_DEBUG_PRINT(("ctl_process_done\n"));
13450 
13451 	fe_done =
13452 	    control_softc->ctl_ports[targ_port]->fe_done;
13453 
13454 #ifdef CTL_TIME_IO
13455 	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
13456 		char str[256];
13457 		char path_str[64];
13458 		struct sbuf sb;
13459 
13460 		ctl_scsi_path_string(io, path_str, sizeof(path_str));
13461 		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13462 
13463 		sbuf_cat(&sb, path_str);
13464 		switch (io->io_hdr.io_type) {
13465 		case CTL_IO_SCSI:
13466 			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
13467 			sbuf_printf(&sb, "\n");
13468 			sbuf_cat(&sb, path_str);
13469 			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13470 				    io->scsiio.tag_num, io->scsiio.tag_type);
13471 			break;
13472 		case CTL_IO_TASK:
13473 			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
13474 				    "Tag Type: %d\n", io->taskio.task_action,
13475 				    io->taskio.tag_num, io->taskio.tag_type);
13476 			break;
13477 		default:
13478 			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13479 			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13480 			break;
13481 		}
13482 		sbuf_cat(&sb, path_str);
13483 		sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
13484 			    (intmax_t)time_uptime - io->io_hdr.start_time);
13485 		sbuf_finish(&sb);
13486 		printf("%s", sbuf_data(&sb));
13487 	}
13488 #endif /* CTL_TIME_IO */
13489 
13490 	switch (io->io_hdr.io_type) {
13491 	case CTL_IO_SCSI:
13492 		break;
13493 	case CTL_IO_TASK:
13494 		if (bootverbose || (ctl_debug & CTL_DEBUG_INFO))
13495 			ctl_io_error_print(io, NULL);
13496 		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
13497 			ctl_free_io(io);
13498 		else
13499 			fe_done(io);
13500 		return (CTL_RETVAL_COMPLETE);
13501 	default:
13502 		panic("ctl_process_done: invalid io type %d\n",
13503 		      io->io_hdr.io_type);
13504 		break; /* NOTREACHED */
13505 	}
13506 
13507 	lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13508 	if (lun == NULL) {
13509 		CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
13510 				 io->io_hdr.nexus.targ_mapped_lun));
13511 		goto bailout;
13512 	}
13513 
13514 	mtx_lock(&lun->lun_lock);
13515 
13516 	/*
13517 	 * Check to see if we have any errors to inject here.  We only
13518 	 * inject errors for commands that don't already have errors set.
13519 	 */
13520 	if ((STAILQ_FIRST(&lun->error_list) != NULL) &&
13521 	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) &&
13522 	    ((io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0))
13523 		ctl_inject_error(lun, io);
13524 
13525 	/*
13526 	 * XXX KDM how do we treat commands that aren't completed
13527 	 * successfully?
13528 	 *
13529 	 * XXX KDM should we also track I/O latency?
13530 	 */
13531 	if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
13532 	    io->io_hdr.io_type == CTL_IO_SCSI) {
13533 #ifdef CTL_TIME_IO
13534 		struct bintime cur_bt;
13535 #endif
13536 		int type;
13537 
13538 		if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13539 		    CTL_FLAG_DATA_IN)
13540 			type = CTL_STATS_READ;
13541 		else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13542 		    CTL_FLAG_DATA_OUT)
13543 			type = CTL_STATS_WRITE;
13544 		else
13545 			type = CTL_STATS_NO_IO;
13546 
13547 		lun->stats.ports[targ_port].bytes[type] +=
13548 		    io->scsiio.kern_total_len;
13549 		lun->stats.ports[targ_port].operations[type]++;
13550 #ifdef CTL_TIME_IO
13551 		bintime_add(&lun->stats.ports[targ_port].dma_time[type],
13552 		   &io->io_hdr.dma_bt);
13553 		lun->stats.ports[targ_port].num_dmas[type] +=
13554 		    io->io_hdr.num_dmas;
13555 		getbintime(&cur_bt);
13556 		bintime_sub(&cur_bt, &io->io_hdr.start_bt);
13557 		bintime_add(&lun->stats.ports[targ_port].time[type], &cur_bt);
13558 #endif
13559 	}
13560 
13561 	/*
13562 	 * Remove this from the OOA queue.
13563 	 */
13564 	TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13565 
13566 	/*
13567 	 * Run through the blocked queue on this LUN and see if anything
13568 	 * has become unblocked, now that this transaction is done.
13569 	 */
13570 	ctl_check_blocked(lun);
13571 
13572 	/*
13573 	 * If the LUN has been invalidated, free it if there is nothing
13574 	 * left on its OOA queue.
13575 	 */
13576 	if ((lun->flags & CTL_LUN_INVALID)
13577 	 && TAILQ_EMPTY(&lun->ooa_queue)) {
13578 		mtx_unlock(&lun->lun_lock);
13579 		mtx_lock(&ctl_softc->ctl_lock);
13580 		ctl_free_lun(lun);
13581 		mtx_unlock(&ctl_softc->ctl_lock);
13582 	} else
13583 		mtx_unlock(&lun->lun_lock);
13584 
13585 bailout:
13586 
13587 	/*
13588 	 * If this command has been aborted, make sure we set the status
13589 	 * properly.  The FETD is responsible for freeing the I/O and doing
13590 	 * whatever it needs to do to clean up its state.
13591 	 */
13592 	if (io->io_hdr.flags & CTL_FLAG_ABORT)
13593 		ctl_set_task_aborted(&io->scsiio);
13594 
13595 	/*
13596 	 * If enabled, print command error status.
13597 	 * We don't print UAs unless debugging was enabled explicitly.
13598 	 */
13599 	do {
13600 		if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)
13601 			break;
13602 		if (!bootverbose && (ctl_debug & CTL_DEBUG_INFO) == 0)
13603 			break;
13604 		if ((ctl_debug & CTL_DEBUG_INFO) == 0 &&
13605 		    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SCSI_ERROR) &&
13606 		     (io->scsiio.scsi_status == SCSI_STATUS_CHECK_COND)) {
13607 			int error_code, sense_key, asc, ascq;
13608 
13609 			scsi_extract_sense_len(&io->scsiio.sense_data,
13610 			    io->scsiio.sense_len, &error_code, &sense_key,
13611 			    &asc, &ascq, /*show_errors*/ 0);
13612 			if (sense_key == SSD_KEY_UNIT_ATTENTION)
13613 				break;
13614 		}
13615 
13616 		ctl_io_error_print(io, NULL);
13617 	} while (0);
13618 
13619 	/*
13620 	 * Tell the FETD or the other shelf controller we're done with this
13621 	 * command.  Note that only SCSI commands get to this point.  Task
13622 	 * management commands are completed above.
13623 	 *
13624 	 * We only send status to the other controller if we're in XFER
13625 	 * mode.  In SER_ONLY mode, the I/O is done on the controller that
13626 	 * received the I/O (from CTL's perspective), and so the status is
13627 	 * generated there.
13628 	 *
13629 	 * XXX KDM if we hold the lock here, we could cause a deadlock
13630 	 * if the frontend comes back in in this context to queue
13631 	 * something.
13632 	 */
13633 	if ((ctl_softc->ha_mode == CTL_HA_MODE_XFER)
13634 	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
13635 		union ctl_ha_msg msg;
13636 
13637 		memset(&msg, 0, sizeof(msg));
13638 		msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13639 		msg.hdr.original_sc = io->io_hdr.original_sc;
13640 		msg.hdr.nexus = io->io_hdr.nexus;
13641 		msg.hdr.status = io->io_hdr.status;
13642 		msg.scsi.scsi_status = io->scsiio.scsi_status;
13643 		msg.scsi.tag_num = io->scsiio.tag_num;
13644 		msg.scsi.tag_type = io->scsiio.tag_type;
13645 		msg.scsi.sense_len = io->scsiio.sense_len;
13646 		msg.scsi.sense_residual = io->scsiio.sense_residual;
13647 		msg.scsi.residual = io->scsiio.residual;
13648 		memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
13649 		       sizeof(io->scsiio.sense_data));
13650 		/*
13651 		 * We copy this whether or not this is an I/O-related
13652 		 * command.  Otherwise, we'd have to go and check to see
13653 		 * whether it's a read/write command, and it really isn't
13654 		 * worth it.
13655 		 */
13656 		memcpy(&msg.scsi.lbalen,
13657 		       &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
13658 		       sizeof(msg.scsi.lbalen));
13659 
13660 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13661 				sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
13662 			/* XXX do something here */
13663 		}
13664 
13665 		ctl_free_io(io);
13666 	} else
13667 		fe_done(io);
13668 
13669 	return (CTL_RETVAL_COMPLETE);
13670 }
13671 
13672 #ifdef CTL_WITH_CA
13673 /*
13674  * Front end should call this if it doesn't do autosense.  When the request
13675  * sense comes back in from the initiator, we'll dequeue this and send it.
13676  */
13677 int
13678 ctl_queue_sense(union ctl_io *io)
13679 {
13680 	struct ctl_lun *lun;
13681 	struct ctl_softc *ctl_softc;
13682 	uint32_t initidx, targ_lun;
13683 
13684 	ctl_softc = control_softc;
13685 
13686 	CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13687 
13688 	/*
13689 	 * LUN lookup will likely move to the ctl_work_thread() once we
13690 	 * have our new queueing infrastructure (that doesn't put things on
13691 	 * a per-LUN queue initially).  That is so that we can handle
13692 	 * things like an INQUIRY to a LUN that we don't have enabled.  We
13693 	 * can't deal with that right now.
13694 	 */
13695 	mtx_lock(&ctl_softc->ctl_lock);
13696 
13697 	/*
13698 	 * If we don't have a LUN for this, just toss the sense
13699 	 * information.
13700 	 */
13701 	targ_lun = io->io_hdr.nexus.targ_lun;
13702 	targ_lun = ctl_map_lun(io->io_hdr.nexus.targ_port, targ_lun);
13703 	if ((targ_lun < CTL_MAX_LUNS)
13704 	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
13705 		lun = ctl_softc->ctl_luns[targ_lun];
13706 	else
13707 		goto bailout;
13708 
13709 	initidx = ctl_get_initindex(&io->io_hdr.nexus);
13710 
13711 	mtx_lock(&lun->lun_lock);
13712 	/*
13713 	 * Already have CA set for this LUN...toss the sense information.
13714 	 */
13715 	if (ctl_is_set(lun->have_ca, initidx)) {
13716 		mtx_unlock(&lun->lun_lock);
13717 		goto bailout;
13718 	}
13719 
13720 	memcpy(&lun->pending_sense[initidx], &io->scsiio.sense_data,
13721 	       ctl_min(sizeof(lun->pending_sense[initidx]),
13722 	       sizeof(io->scsiio.sense_data)));
13723 	ctl_set_mask(lun->have_ca, initidx);
13724 	mtx_unlock(&lun->lun_lock);
13725 
13726 bailout:
13727 	mtx_unlock(&ctl_softc->ctl_lock);
13728 
13729 	ctl_free_io(io);
13730 
13731 	return (CTL_RETVAL_COMPLETE);
13732 }
13733 #endif
13734 
13735 /*
13736  * Primary command inlet from frontend ports.  All SCSI and task I/O
13737  * requests must go through this function.
13738  */
13739 int
13740 ctl_queue(union ctl_io *io)
13741 {
13742 	struct ctl_softc *ctl_softc;
13743 
13744 	CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13745 
13746 	ctl_softc = control_softc;
13747 
13748 #ifdef CTL_TIME_IO
13749 	io->io_hdr.start_time = time_uptime;
13750 	getbintime(&io->io_hdr.start_bt);
13751 #endif /* CTL_TIME_IO */
13752 
13753 	/* Map FE-specific LUN ID into global one. */
13754 	io->io_hdr.nexus.targ_mapped_lun =
13755 	    ctl_map_lun(io->io_hdr.nexus.targ_port, io->io_hdr.nexus.targ_lun);
13756 
13757 	switch (io->io_hdr.io_type) {
13758 	case CTL_IO_SCSI:
13759 	case CTL_IO_TASK:
13760 		if (ctl_debug & CTL_DEBUG_CDB)
13761 			ctl_io_print(io);
13762 		ctl_enqueue_incoming(io);
13763 		break;
13764 	default:
13765 		printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13766 		return (EINVAL);
13767 	}
13768 
13769 	return (CTL_RETVAL_COMPLETE);
13770 }
13771 
13772 #ifdef CTL_IO_DELAY
13773 static void
13774 ctl_done_timer_wakeup(void *arg)
13775 {
13776 	union ctl_io *io;
13777 
13778 	io = (union ctl_io *)arg;
13779 	ctl_done(io);
13780 }
13781 #endif /* CTL_IO_DELAY */
13782 
13783 void
13784 ctl_done(union ctl_io *io)
13785 {
13786 	struct ctl_softc *ctl_softc;
13787 
13788 	ctl_softc = control_softc;
13789 
13790 	/*
13791 	 * Enable this to catch duplicate completion issues.
13792 	 */
13793 #if 0
13794 	if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13795 		printf("%s: type %d msg %d cdb %x iptl: "
13796 		       "%d:%d:%d:%d tag 0x%04x "
13797 		       "flag %#x status %x\n",
13798 			__func__,
13799 			io->io_hdr.io_type,
13800 			io->io_hdr.msg_type,
13801 			io->scsiio.cdb[0],
13802 			io->io_hdr.nexus.initid.id,
13803 			io->io_hdr.nexus.targ_port,
13804 			io->io_hdr.nexus.targ_target.id,
13805 			io->io_hdr.nexus.targ_lun,
13806 			(io->io_hdr.io_type ==
13807 			CTL_IO_TASK) ?
13808 			io->taskio.tag_num :
13809 			io->scsiio.tag_num,
13810 		        io->io_hdr.flags,
13811 			io->io_hdr.status);
13812 	} else
13813 		io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13814 #endif
13815 
13816 	/*
13817 	 * This is an internal copy of an I/O, and should not go through
13818 	 * the normal done processing logic.
13819 	 */
13820 	if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
13821 		return;
13822 
13823 	/*
13824 	 * We need to send a msg to the serializing shelf to finish the IO
13825 	 * as well.  We don't send a finish message to the other shelf if
13826 	 * this is a task management command.  Task management commands
13827 	 * aren't serialized in the OOA queue, but rather just executed on
13828 	 * both shelf controllers for commands that originated on that
13829 	 * controller.
13830 	 */
13831 	if ((io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)
13832 	 && (io->io_hdr.io_type != CTL_IO_TASK)) {
13833 		union ctl_ha_msg msg_io;
13834 
13835 		msg_io.hdr.msg_type = CTL_MSG_FINISH_IO;
13836 		msg_io.hdr.serializing_sc = io->io_hdr.serializing_sc;
13837 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_io,
13838 		    sizeof(msg_io), 0 ) != CTL_HA_STATUS_SUCCESS) {
13839 		}
13840 		/* continue on to finish IO */
13841 	}
13842 #ifdef CTL_IO_DELAY
13843 	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13844 		struct ctl_lun *lun;
13845 
13846 		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13847 
13848 		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13849 	} else {
13850 		struct ctl_lun *lun;
13851 
13852 		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13853 
13854 		if ((lun != NULL)
13855 		 && (lun->delay_info.done_delay > 0)) {
13856 			struct callout *callout;
13857 
13858 			callout = (struct callout *)&io->io_hdr.timer_bytes;
13859 			callout_init(callout, /*mpsafe*/ 1);
13860 			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13861 			callout_reset(callout,
13862 				      lun->delay_info.done_delay * hz,
13863 				      ctl_done_timer_wakeup, io);
13864 			if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13865 				lun->delay_info.done_delay = 0;
13866 			return;
13867 		}
13868 	}
13869 #endif /* CTL_IO_DELAY */
13870 
13871 	ctl_enqueue_done(io);
13872 }
13873 
13874 int
13875 ctl_isc(struct ctl_scsiio *ctsio)
13876 {
13877 	struct ctl_lun *lun;
13878 	int retval;
13879 
13880 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13881 
13882 	CTL_DEBUG_PRINT(("ctl_isc: command: %02x\n", ctsio->cdb[0]));
13883 
13884 	CTL_DEBUG_PRINT(("ctl_isc: calling data_submit()\n"));
13885 
13886 	retval = lun->backend->data_submit((union ctl_io *)ctsio);
13887 
13888 	return (retval);
13889 }
13890 
13891 
13892 static void
13893 ctl_work_thread(void *arg)
13894 {
13895 	struct ctl_thread *thr = (struct ctl_thread *)arg;
13896 	struct ctl_softc *softc = thr->ctl_softc;
13897 	union ctl_io *io;
13898 	int retval;
13899 
13900 	CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13901 
13902 	for (;;) {
13903 		retval = 0;
13904 
13905 		/*
13906 		 * We handle the queues in this order:
13907 		 * - ISC
13908 		 * - done queue (to free up resources, unblock other commands)
13909 		 * - RtR queue
13910 		 * - incoming queue
13911 		 *
13912 		 * If those queues are empty, we break out of the loop and
13913 		 * go to sleep.
13914 		 */
13915 		mtx_lock(&thr->queue_lock);
13916 		io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
13917 		if (io != NULL) {
13918 			STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
13919 			mtx_unlock(&thr->queue_lock);
13920 			ctl_handle_isc(io);
13921 			continue;
13922 		}
13923 		io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
13924 		if (io != NULL) {
13925 			STAILQ_REMOVE_HEAD(&thr->done_queue, links);
13926 			/* clear any blocked commands, call fe_done */
13927 			mtx_unlock(&thr->queue_lock);
13928 			retval = ctl_process_done(io);
13929 			continue;
13930 		}
13931 		io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
13932 		if (io != NULL) {
13933 			STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
13934 			mtx_unlock(&thr->queue_lock);
13935 			if (io->io_hdr.io_type == CTL_IO_TASK)
13936 				ctl_run_task(io);
13937 			else
13938 				ctl_scsiio_precheck(softc, &io->scsiio);
13939 			continue;
13940 		}
13941 		if (!ctl_pause_rtr) {
13942 			io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
13943 			if (io != NULL) {
13944 				STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
13945 				mtx_unlock(&thr->queue_lock);
13946 				retval = ctl_scsiio(&io->scsiio);
13947 				if (retval != CTL_RETVAL_COMPLETE)
13948 					CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13949 				continue;
13950 			}
13951 		}
13952 
13953 		/* Sleep until we have something to do. */
13954 		mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
13955 	}
13956 }
13957 
13958 static void
13959 ctl_lun_thread(void *arg)
13960 {
13961 	struct ctl_softc *softc = (struct ctl_softc *)arg;
13962 	struct ctl_be_lun *be_lun;
13963 	int retval;
13964 
13965 	CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
13966 
13967 	for (;;) {
13968 		retval = 0;
13969 		mtx_lock(&softc->ctl_lock);
13970 		be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
13971 		if (be_lun != NULL) {
13972 			STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
13973 			mtx_unlock(&softc->ctl_lock);
13974 			ctl_create_lun(be_lun);
13975 			continue;
13976 		}
13977 
13978 		/* Sleep until we have something to do. */
13979 		mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
13980 		    PDROP | PRIBIO, "-", 0);
13981 	}
13982 }
13983 
13984 static void
13985 ctl_thresh_thread(void *arg)
13986 {
13987 	struct ctl_softc *softc = (struct ctl_softc *)arg;
13988 	struct ctl_lun *lun;
13989 	struct ctl_be_lun *be_lun;
13990 	struct scsi_da_rw_recovery_page *rwpage;
13991 	struct ctl_logical_block_provisioning_page *page;
13992 	const char *attr;
13993 	uint64_t thres, val;
13994 	int i, e;
13995 
13996 	CTL_DEBUG_PRINT(("ctl_thresh_thread starting\n"));
13997 
13998 	for (;;) {
13999 		mtx_lock(&softc->ctl_lock);
14000 		STAILQ_FOREACH(lun, &softc->lun_list, links) {
14001 			be_lun = lun->be_lun;
14002 			if ((lun->flags & CTL_LUN_DISABLED) ||
14003 			    (lun->flags & CTL_LUN_OFFLINE) ||
14004 			    lun->backend->lun_attr == NULL)
14005 				continue;
14006 			rwpage = &lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT];
14007 			if ((rwpage->byte8 & SMS_RWER_LBPERE) == 0)
14008 				continue;
14009 			e = 0;
14010 			page = &lun->mode_pages.lbp_page[CTL_PAGE_CURRENT];
14011 			for (i = 0; i < CTL_NUM_LBP_THRESH; i++) {
14012 				if ((page->descr[i].flags & SLBPPD_ENABLED) == 0)
14013 					continue;
14014 				thres = scsi_4btoul(page->descr[i].count);
14015 				thres <<= CTL_LBP_EXPONENT;
14016 				switch (page->descr[i].resource) {
14017 				case 0x01:
14018 					attr = "blocksavail";
14019 					break;
14020 				case 0x02:
14021 					attr = "blocksused";
14022 					break;
14023 				case 0xf1:
14024 					attr = "poolblocksavail";
14025 					break;
14026 				case 0xf2:
14027 					attr = "poolblocksused";
14028 					break;
14029 				default:
14030 					continue;
14031 				}
14032 				mtx_unlock(&softc->ctl_lock); // XXX
14033 				val = lun->backend->lun_attr(
14034 				    lun->be_lun->be_lun, attr);
14035 				mtx_lock(&softc->ctl_lock);
14036 				if (val == UINT64_MAX)
14037 					continue;
14038 				if ((page->descr[i].flags & SLBPPD_ARMING_MASK)
14039 				    == SLBPPD_ARMING_INC)
14040 					e |= (val >= thres);
14041 				else
14042 					e |= (val <= thres);
14043 			}
14044 			mtx_lock(&lun->lun_lock);
14045 			if (e) {
14046 				if (lun->lasttpt == 0 ||
14047 				    time_uptime - lun->lasttpt >= CTL_LBP_UA_PERIOD) {
14048 					lun->lasttpt = time_uptime;
14049 					ctl_est_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
14050 				}
14051 			} else {
14052 				lun->lasttpt = 0;
14053 				ctl_clr_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
14054 			}
14055 			mtx_unlock(&lun->lun_lock);
14056 		}
14057 		mtx_unlock(&softc->ctl_lock);
14058 		pause("-", CTL_LBP_PERIOD * hz);
14059 	}
14060 }
14061 
14062 static void
14063 ctl_enqueue_incoming(union ctl_io *io)
14064 {
14065 	struct ctl_softc *softc = control_softc;
14066 	struct ctl_thread *thr;
14067 	u_int idx;
14068 
14069 	idx = (io->io_hdr.nexus.targ_port * 127 +
14070 	       io->io_hdr.nexus.initid.id) % worker_threads;
14071 	thr = &softc->threads[idx];
14072 	mtx_lock(&thr->queue_lock);
14073 	STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
14074 	mtx_unlock(&thr->queue_lock);
14075 	wakeup(thr);
14076 }
14077 
14078 static void
14079 ctl_enqueue_rtr(union ctl_io *io)
14080 {
14081 	struct ctl_softc *softc = control_softc;
14082 	struct ctl_thread *thr;
14083 
14084 	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14085 	mtx_lock(&thr->queue_lock);
14086 	STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
14087 	mtx_unlock(&thr->queue_lock);
14088 	wakeup(thr);
14089 }
14090 
14091 static void
14092 ctl_enqueue_done(union ctl_io *io)
14093 {
14094 	struct ctl_softc *softc = control_softc;
14095 	struct ctl_thread *thr;
14096 
14097 	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14098 	mtx_lock(&thr->queue_lock);
14099 	STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
14100 	mtx_unlock(&thr->queue_lock);
14101 	wakeup(thr);
14102 }
14103 
14104 static void
14105 ctl_enqueue_isc(union ctl_io *io)
14106 {
14107 	struct ctl_softc *softc = control_softc;
14108 	struct ctl_thread *thr;
14109 
14110 	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14111 	mtx_lock(&thr->queue_lock);
14112 	STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
14113 	mtx_unlock(&thr->queue_lock);
14114 	wakeup(thr);
14115 }
14116 
14117 /* Initialization and failover */
14118 
14119 void
14120 ctl_init_isc_msg(void)
14121 {
14122 	printf("CTL: Still calling this thing\n");
14123 }
14124 
14125 /*
14126  * Init component
14127  * 	Initializes component into configuration defined by bootMode
14128  *	(see hasc-sv.c)
14129  *  	returns hasc_Status:
14130  * 		OK
14131  *		ERROR - fatal error
14132  */
14133 static ctl_ha_comp_status
14134 ctl_isc_init(struct ctl_ha_component *c)
14135 {
14136 	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
14137 
14138 	c->status = ret;
14139 	return ret;
14140 }
14141 
14142 /* Start component
14143  * 	Starts component in state requested. If component starts successfully,
14144  *	it must set its own state to the requestrd state
14145  *	When requested state is HASC_STATE_HA, the component may refine it
14146  * 	by adding _SLAVE or _MASTER flags.
14147  *	Currently allowed state transitions are:
14148  *	UNKNOWN->HA		- initial startup
14149  *	UNKNOWN->SINGLE - initial startup when no parter detected
14150  *	HA->SINGLE		- failover
14151  * returns ctl_ha_comp_status:
14152  * 		OK	- component successfully started in requested state
14153  *		FAILED  - could not start the requested state, failover may
14154  * 			  be possible
14155  *		ERROR	- fatal error detected, no future startup possible
14156  */
14157 static ctl_ha_comp_status
14158 ctl_isc_start(struct ctl_ha_component *c, ctl_ha_state state)
14159 {
14160 	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
14161 
14162 	printf("%s: go\n", __func__);
14163 
14164 	// UNKNOWN->HA or UNKNOWN->SINGLE (bootstrap)
14165 	if (c->state == CTL_HA_STATE_UNKNOWN ) {
14166 		control_softc->is_single = 0;
14167 		if (ctl_ha_msg_create(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
14168 		    != CTL_HA_STATUS_SUCCESS) {
14169 			printf("ctl_isc_start: ctl_ha_msg_create failed.\n");
14170 			ret = CTL_HA_COMP_STATUS_ERROR;
14171 		}
14172 	} else if (CTL_HA_STATE_IS_HA(c->state)
14173 		&& CTL_HA_STATE_IS_SINGLE(state)){
14174 		// HA->SINGLE transition
14175 	        ctl_failover();
14176 		control_softc->is_single = 1;
14177 	} else {
14178 		printf("ctl_isc_start:Invalid state transition %X->%X\n",
14179 		       c->state, state);
14180 		ret = CTL_HA_COMP_STATUS_ERROR;
14181 	}
14182 	if (CTL_HA_STATE_IS_SINGLE(state))
14183 		control_softc->is_single = 1;
14184 
14185 	c->state = state;
14186 	c->status = ret;
14187 	return ret;
14188 }
14189 
14190 /*
14191  * Quiesce component
14192  * The component must clear any error conditions (set status to OK) and
14193  * prepare itself to another Start call
14194  * returns ctl_ha_comp_status:
14195  * 	OK
14196  *	ERROR
14197  */
14198 static ctl_ha_comp_status
14199 ctl_isc_quiesce(struct ctl_ha_component *c)
14200 {
14201 	int ret = CTL_HA_COMP_STATUS_OK;
14202 
14203 	ctl_pause_rtr = 1;
14204 	c->status = ret;
14205 	return ret;
14206 }
14207 
14208 struct ctl_ha_component ctl_ha_component_ctlisc =
14209 {
14210 	.name = "CTL ISC",
14211 	.state = CTL_HA_STATE_UNKNOWN,
14212 	.init = ctl_isc_init,
14213 	.start = ctl_isc_start,
14214 	.quiesce = ctl_isc_quiesce
14215 };
14216 
14217 /*
14218  *  vim: ts=8
14219  */
14220