xref: /freebsd/sys/cam/ctl/ctl.c (revision 9f31240773ae369c8285b700840a35e6bfa9f762)
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/kernel.h>
50 #include <sys/types.h>
51 #include <sys/kthread.h>
52 #include <sys/bio.h>
53 #include <sys/fcntl.h>
54 #include <sys/lock.h>
55 #include <sys/module.h>
56 #include <sys/mutex.h>
57 #include <sys/condvar.h>
58 #include <sys/malloc.h>
59 #include <sys/conf.h>
60 #include <sys/ioccom.h>
61 #include <sys/queue.h>
62 #include <sys/sbuf.h>
63 #include <sys/smp.h>
64 #include <sys/endian.h>
65 #include <sys/sysctl.h>
66 
67 #include <cam/cam.h>
68 #include <cam/scsi/scsi_all.h>
69 #include <cam/scsi/scsi_da.h>
70 #include <cam/ctl/ctl_io.h>
71 #include <cam/ctl/ctl.h>
72 #include <cam/ctl/ctl_frontend.h>
73 #include <cam/ctl/ctl_frontend_internal.h>
74 #include <cam/ctl/ctl_util.h>
75 #include <cam/ctl/ctl_backend.h>
76 #include <cam/ctl/ctl_ioctl.h>
77 #include <cam/ctl/ctl_ha.h>
78 #include <cam/ctl/ctl_private.h>
79 #include <cam/ctl/ctl_debug.h>
80 #include <cam/ctl/ctl_scsi_all.h>
81 #include <cam/ctl/ctl_error.h>
82 
83 struct ctl_softc *control_softc = NULL;
84 
85 /*
86  * Size and alignment macros needed for Copan-specific HA hardware.  These
87  * can go away when the HA code is re-written, and uses busdma for any
88  * hardware.
89  */
90 #define	CTL_ALIGN_8B(target, source, type)				\
91 	if (((uint32_t)source & 0x7) != 0)				\
92 		target = (type)(source + (0x8 - ((uint32_t)source & 0x7)));\
93 	else								\
94 		target = (type)source;
95 
96 #define	CTL_SIZE_8B(target, size)					\
97 	if ((size & 0x7) != 0)						\
98 		target = size + (0x8 - (size & 0x7));			\
99 	else								\
100 		target = size;
101 
102 #define CTL_ALIGN_8B_MARGIN	16
103 
104 /*
105  * Template mode pages.
106  */
107 
108 /*
109  * Note that these are default values only.  The actual values will be
110  * filled in when the user does a mode sense.
111  */
112 static struct copan_power_subpage power_page_default = {
113 	/*page_code*/ PWR_PAGE_CODE | SMPH_SPF,
114 	/*subpage*/ PWR_SUBPAGE_CODE,
115 	/*page_length*/ {(sizeof(struct copan_power_subpage) - 4) & 0xff00,
116 			 (sizeof(struct copan_power_subpage) - 4) & 0x00ff},
117 	/*page_version*/ PWR_VERSION,
118 	/* total_luns */ 26,
119 	/* max_active_luns*/ PWR_DFLT_MAX_LUNS,
120 	/*reserved*/ {0, 0, 0, 0, 0, 0, 0, 0, 0,
121 		      0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
122 		      0, 0, 0, 0, 0, 0}
123 };
124 
125 static struct copan_power_subpage power_page_changeable = {
126 	/*page_code*/ PWR_PAGE_CODE | SMPH_SPF,
127 	/*subpage*/ PWR_SUBPAGE_CODE,
128 	/*page_length*/ {(sizeof(struct copan_power_subpage) - 4) & 0xff00,
129 			 (sizeof(struct copan_power_subpage) - 4) & 0x00ff},
130 	/*page_version*/ 0,
131 	/* total_luns */ 0,
132 	/* max_active_luns*/ 0,
133 	/*reserved*/ {0, 0, 0, 0, 0, 0, 0, 0, 0,
134 		      0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
135 		      0, 0, 0, 0, 0, 0}
136 };
137 
138 static struct copan_aps_subpage aps_page_default = {
139 	APS_PAGE_CODE | SMPH_SPF, //page_code
140 	APS_SUBPAGE_CODE, //subpage
141 	{(sizeof(struct copan_aps_subpage) - 4) & 0xff00,
142 	 (sizeof(struct copan_aps_subpage) - 4) & 0x00ff}, //page_length
143 	APS_VERSION, //page_version
144 	0, //lock_active
145 	{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
146 	0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
147 	0, 0, 0, 0, 0} //reserved
148 };
149 
150 static struct copan_aps_subpage aps_page_changeable = {
151 	APS_PAGE_CODE | SMPH_SPF, //page_code
152 	APS_SUBPAGE_CODE, //subpage
153 	{(sizeof(struct copan_aps_subpage) - 4) & 0xff00,
154 	 (sizeof(struct copan_aps_subpage) - 4) & 0x00ff}, //page_length
155 	0, //page_version
156 	0, //lock_active
157 	{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
158 	0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
159 	0, 0, 0, 0, 0} //reserved
160 };
161 
162 static struct copan_debugconf_subpage debugconf_page_default = {
163 	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
164 	DBGCNF_SUBPAGE_CODE,		/* subpage */
165 	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
166 	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
167 	DBGCNF_VERSION,			/* page_version */
168 	{CTL_TIME_IO_DEFAULT_SECS>>8,
169 	 CTL_TIME_IO_DEFAULT_SECS>>0},	/* ctl_time_io_secs */
170 };
171 
172 static struct copan_debugconf_subpage debugconf_page_changeable = {
173 	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
174 	DBGCNF_SUBPAGE_CODE,		/* subpage */
175 	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
176 	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
177 	0,				/* page_version */
178 	{0xff,0xff},			/* ctl_time_io_secs */
179 };
180 
181 static struct scsi_format_page format_page_default = {
182 	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
183 	/*page_length*/sizeof(struct scsi_format_page) - 2,
184 	/*tracks_per_zone*/ {0, 0},
185 	/*alt_sectors_per_zone*/ {0, 0},
186 	/*alt_tracks_per_zone*/ {0, 0},
187 	/*alt_tracks_per_lun*/ {0, 0},
188 	/*sectors_per_track*/ {(CTL_DEFAULT_SECTORS_PER_TRACK >> 8) & 0xff,
189 			        CTL_DEFAULT_SECTORS_PER_TRACK & 0xff},
190 	/*bytes_per_sector*/ {0, 0},
191 	/*interleave*/ {0, 0},
192 	/*track_skew*/ {0, 0},
193 	/*cylinder_skew*/ {0, 0},
194 	/*flags*/ SFP_HSEC,
195 	/*reserved*/ {0, 0, 0}
196 };
197 
198 static struct scsi_format_page format_page_changeable = {
199 	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
200 	/*page_length*/sizeof(struct scsi_format_page) - 2,
201 	/*tracks_per_zone*/ {0, 0},
202 	/*alt_sectors_per_zone*/ {0, 0},
203 	/*alt_tracks_per_zone*/ {0, 0},
204 	/*alt_tracks_per_lun*/ {0, 0},
205 	/*sectors_per_track*/ {0, 0},
206 	/*bytes_per_sector*/ {0, 0},
207 	/*interleave*/ {0, 0},
208 	/*track_skew*/ {0, 0},
209 	/*cylinder_skew*/ {0, 0},
210 	/*flags*/ 0,
211 	/*reserved*/ {0, 0, 0}
212 };
213 
214 static struct scsi_rigid_disk_page rigid_disk_page_default = {
215 	/*page_code*/SMS_RIGID_DISK_PAGE,
216 	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
217 	/*cylinders*/ {0, 0, 0},
218 	/*heads*/ CTL_DEFAULT_HEADS,
219 	/*start_write_precomp*/ {0, 0, 0},
220 	/*start_reduced_current*/ {0, 0, 0},
221 	/*step_rate*/ {0, 0},
222 	/*landing_zone_cylinder*/ {0, 0, 0},
223 	/*rpl*/ SRDP_RPL_DISABLED,
224 	/*rotational_offset*/ 0,
225 	/*reserved1*/ 0,
226 	/*rotation_rate*/ {(CTL_DEFAULT_ROTATION_RATE >> 8) & 0xff,
227 			   CTL_DEFAULT_ROTATION_RATE & 0xff},
228 	/*reserved2*/ {0, 0}
229 };
230 
231 static struct scsi_rigid_disk_page rigid_disk_page_changeable = {
232 	/*page_code*/SMS_RIGID_DISK_PAGE,
233 	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
234 	/*cylinders*/ {0, 0, 0},
235 	/*heads*/ 0,
236 	/*start_write_precomp*/ {0, 0, 0},
237 	/*start_reduced_current*/ {0, 0, 0},
238 	/*step_rate*/ {0, 0},
239 	/*landing_zone_cylinder*/ {0, 0, 0},
240 	/*rpl*/ 0,
241 	/*rotational_offset*/ 0,
242 	/*reserved1*/ 0,
243 	/*rotation_rate*/ {0, 0},
244 	/*reserved2*/ {0, 0}
245 };
246 
247 static struct scsi_caching_page caching_page_default = {
248 	/*page_code*/SMS_CACHING_PAGE,
249 	/*page_length*/sizeof(struct scsi_caching_page) - 2,
250 	/*flags1*/ SCP_DISC | SCP_WCE,
251 	/*ret_priority*/ 0,
252 	/*disable_pf_transfer_len*/ {0xff, 0xff},
253 	/*min_prefetch*/ {0, 0},
254 	/*max_prefetch*/ {0xff, 0xff},
255 	/*max_pf_ceiling*/ {0xff, 0xff},
256 	/*flags2*/ 0,
257 	/*cache_segments*/ 0,
258 	/*cache_seg_size*/ {0, 0},
259 	/*reserved*/ 0,
260 	/*non_cache_seg_size*/ {0, 0, 0}
261 };
262 
263 static struct scsi_caching_page caching_page_changeable = {
264 	/*page_code*/SMS_CACHING_PAGE,
265 	/*page_length*/sizeof(struct scsi_caching_page) - 2,
266 	/*flags1*/ 0,
267 	/*ret_priority*/ 0,
268 	/*disable_pf_transfer_len*/ {0, 0},
269 	/*min_prefetch*/ {0, 0},
270 	/*max_prefetch*/ {0, 0},
271 	/*max_pf_ceiling*/ {0, 0},
272 	/*flags2*/ 0,
273 	/*cache_segments*/ 0,
274 	/*cache_seg_size*/ {0, 0},
275 	/*reserved*/ 0,
276 	/*non_cache_seg_size*/ {0, 0, 0}
277 };
278 
279 static struct scsi_control_page control_page_default = {
280 	/*page_code*/SMS_CONTROL_MODE_PAGE,
281 	/*page_length*/sizeof(struct scsi_control_page) - 2,
282 	/*rlec*/0,
283 	/*queue_flags*/0,
284 	/*eca_and_aen*/0,
285 	/*flags4*/SCP_TAS,
286 	/*aen_holdoff_period*/{0, 0},
287 	/*busy_timeout_period*/{0, 0},
288 	/*extended_selftest_completion_time*/{0, 0}
289 };
290 
291 static struct scsi_control_page control_page_changeable = {
292 	/*page_code*/SMS_CONTROL_MODE_PAGE,
293 	/*page_length*/sizeof(struct scsi_control_page) - 2,
294 	/*rlec*/SCP_DSENSE,
295 	/*queue_flags*/0,
296 	/*eca_and_aen*/0,
297 	/*flags4*/0,
298 	/*aen_holdoff_period*/{0, 0},
299 	/*busy_timeout_period*/{0, 0},
300 	/*extended_selftest_completion_time*/{0, 0}
301 };
302 
303 
304 /*
305  * XXX KDM move these into the softc.
306  */
307 static int rcv_sync_msg;
308 static int persis_offset;
309 static uint8_t ctl_pause_rtr;
310 static int     ctl_is_single = 1;
311 static int     index_to_aps_page;
312 
313 SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD, 0, "CAM Target Layer");
314 static int worker_threads = -1;
315 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, worker_threads, CTLFLAG_RDTUN,
316     &worker_threads, 1, "Number of worker threads");
317 static int verbose = 0;
318 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, verbose, CTLFLAG_RWTUN,
319     &verbose, 0, "Show SCSI errors returned to initiator");
320 
321 /*
322  * Supported pages (0x00), Serial number (0x80), Device ID (0x83),
323  * SCSI Ports (0x88), Third-party Copy (0x8F), Block limits (0xB0),
324  * Block Device Characteristics (0xB1) and Logical Block Provisioning (0xB2)
325  */
326 #define SCSI_EVPD_NUM_SUPPORTED_PAGES	8
327 
328 static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
329 				  int param);
330 static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
331 static int ctl_init(void);
332 void ctl_shutdown(void);
333 static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
334 static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
335 static void ctl_ioctl_online(void *arg);
336 static void ctl_ioctl_offline(void *arg);
337 static int ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id);
338 static int ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id);
339 static int ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio);
340 static int ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio);
341 static int ctl_ioctl_submit_wait(union ctl_io *io);
342 static void ctl_ioctl_datamove(union ctl_io *io);
343 static void ctl_ioctl_done(union ctl_io *io);
344 static void ctl_ioctl_hard_startstop_callback(void *arg,
345 					      struct cfi_metatask *metatask);
346 static void ctl_ioctl_bbrread_callback(void *arg,struct cfi_metatask *metatask);
347 static int ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
348 			      struct ctl_ooa *ooa_hdr,
349 			      struct ctl_ooa_entry *kern_entries);
350 static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
351 		     struct thread *td);
352 static uint32_t ctl_map_lun(int port_num, uint32_t lun);
353 static uint32_t ctl_map_lun_back(int port_num, uint32_t lun);
354 #ifdef unused
355 static union ctl_io *ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port,
356 				   uint32_t targ_target, uint32_t targ_lun,
357 				   int can_wait);
358 static void ctl_kfree_io(union ctl_io *io);
359 #endif /* unused */
360 static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
361 			 struct ctl_be_lun *be_lun, struct ctl_id target_id);
362 static int ctl_free_lun(struct ctl_lun *lun);
363 static void ctl_create_lun(struct ctl_be_lun *be_lun);
364 /**
365 static void ctl_failover_change_pages(struct ctl_softc *softc,
366 				      struct ctl_scsiio *ctsio, int master);
367 **/
368 
369 static int ctl_do_mode_select(union ctl_io *io);
370 static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
371 			   uint64_t res_key, uint64_t sa_res_key,
372 			   uint8_t type, uint32_t residx,
373 			   struct ctl_scsiio *ctsio,
374 			   struct scsi_per_res_out *cdb,
375 			   struct scsi_per_res_out_parms* param);
376 static void ctl_pro_preempt_other(struct ctl_lun *lun,
377 				  union ctl_ha_msg *msg);
378 static void ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg);
379 static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
380 static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
381 static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
382 static int ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio,
383 					 int alloc_len);
384 static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
385 					 int alloc_len);
386 static int ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len);
387 static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
388 static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
389 static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
390 static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint32_t *len);
391 static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2);
392 static ctl_action ctl_check_for_blockage(union ctl_io *pending_io,
393 					 union ctl_io *ooa_io);
394 static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
395 				union ctl_io *starting_io);
396 static int ctl_check_blocked(struct ctl_lun *lun);
397 static int ctl_scsiio_lun_check(struct ctl_softc *ctl_softc,
398 				struct ctl_lun *lun,
399 				const struct ctl_cmd_entry *entry,
400 				struct ctl_scsiio *ctsio);
401 //static int ctl_check_rtr(union ctl_io *pending_io, struct ctl_softc *softc);
402 static void ctl_failover(void);
403 static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
404 			       struct ctl_scsiio *ctsio);
405 static int ctl_scsiio(struct ctl_scsiio *ctsio);
406 
407 static int ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
408 static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
409 			    ctl_ua_type ua_type);
410 static int ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io,
411 			 ctl_ua_type ua_type);
412 static int ctl_abort_task(union ctl_io *io);
413 static int ctl_abort_task_set(union ctl_io *io);
414 static int ctl_i_t_nexus_reset(union ctl_io *io);
415 static void ctl_run_task(union ctl_io *io);
416 #ifdef CTL_IO_DELAY
417 static void ctl_datamove_timer_wakeup(void *arg);
418 static void ctl_done_timer_wakeup(void *arg);
419 #endif /* CTL_IO_DELAY */
420 
421 static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
422 static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
423 static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
424 static void ctl_datamove_remote_write(union ctl_io *io);
425 static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
426 static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
427 static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
428 static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
429 				    ctl_ha_dt_cb callback);
430 static void ctl_datamove_remote_read(union ctl_io *io);
431 static void ctl_datamove_remote(union ctl_io *io);
432 static int ctl_process_done(union ctl_io *io);
433 static void ctl_lun_thread(void *arg);
434 static void ctl_work_thread(void *arg);
435 static void ctl_enqueue_incoming(union ctl_io *io);
436 static void ctl_enqueue_rtr(union ctl_io *io);
437 static void ctl_enqueue_done(union ctl_io *io);
438 static void ctl_enqueue_isc(union ctl_io *io);
439 static const struct ctl_cmd_entry *
440     ctl_get_cmd_entry(struct ctl_scsiio *ctsio);
441 static const struct ctl_cmd_entry *
442     ctl_validate_command(struct ctl_scsiio *ctsio);
443 static int ctl_cmd_applicable(uint8_t lun_type,
444     const struct ctl_cmd_entry *entry);
445 
446 /*
447  * Load the serialization table.  This isn't very pretty, but is probably
448  * the easiest way to do it.
449  */
450 #include "ctl_ser_table.c"
451 
452 /*
453  * We only need to define open, close and ioctl routines for this driver.
454  */
455 static struct cdevsw ctl_cdevsw = {
456 	.d_version =	D_VERSION,
457 	.d_flags =	0,
458 	.d_open =	ctl_open,
459 	.d_close =	ctl_close,
460 	.d_ioctl =	ctl_ioctl,
461 	.d_name =	"ctl",
462 };
463 
464 
465 MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
466 MALLOC_DEFINE(M_CTLIO, "ctlio", "Memory used for CTL requests");
467 
468 static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
469 
470 static moduledata_t ctl_moduledata = {
471 	"ctl",
472 	ctl_module_event_handler,
473 	NULL
474 };
475 
476 DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
477 MODULE_VERSION(ctl, 1);
478 
479 static struct ctl_frontend ioctl_frontend =
480 {
481 	.name = "ioctl",
482 };
483 
484 static void
485 ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
486 			    union ctl_ha_msg *msg_info)
487 {
488 	struct ctl_scsiio *ctsio;
489 
490 	if (msg_info->hdr.original_sc == NULL) {
491 		printf("%s: original_sc == NULL!\n", __func__);
492 		/* XXX KDM now what? */
493 		return;
494 	}
495 
496 	ctsio = &msg_info->hdr.original_sc->scsiio;
497 	ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
498 	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
499 	ctsio->io_hdr.status = msg_info->hdr.status;
500 	ctsio->scsi_status = msg_info->scsi.scsi_status;
501 	ctsio->sense_len = msg_info->scsi.sense_len;
502 	ctsio->sense_residual = msg_info->scsi.sense_residual;
503 	ctsio->residual = msg_info->scsi.residual;
504 	memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
505 	       sizeof(ctsio->sense_data));
506 	memcpy(&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
507 	       &msg_info->scsi.lbalen, sizeof(msg_info->scsi.lbalen));
508 	ctl_enqueue_isc((union ctl_io *)ctsio);
509 }
510 
511 static void
512 ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
513 				union ctl_ha_msg *msg_info)
514 {
515 	struct ctl_scsiio *ctsio;
516 
517 	if (msg_info->hdr.serializing_sc == NULL) {
518 		printf("%s: serializing_sc == NULL!\n", __func__);
519 		/* XXX KDM now what? */
520 		return;
521 	}
522 
523 	ctsio = &msg_info->hdr.serializing_sc->scsiio;
524 #if 0
525 	/*
526 	 * Attempt to catch the situation where an I/O has
527 	 * been freed, and we're using it again.
528 	 */
529 	if (ctsio->io_hdr.io_type == 0xff) {
530 		union ctl_io *tmp_io;
531 		tmp_io = (union ctl_io *)ctsio;
532 		printf("%s: %p use after free!\n", __func__,
533 		       ctsio);
534 		printf("%s: type %d msg %d cdb %x iptl: "
535 		       "%d:%d:%d:%d tag 0x%04x "
536 		       "flag %#x status %x\n",
537 			__func__,
538 			tmp_io->io_hdr.io_type,
539 			tmp_io->io_hdr.msg_type,
540 			tmp_io->scsiio.cdb[0],
541 			tmp_io->io_hdr.nexus.initid.id,
542 			tmp_io->io_hdr.nexus.targ_port,
543 			tmp_io->io_hdr.nexus.targ_target.id,
544 			tmp_io->io_hdr.nexus.targ_lun,
545 			(tmp_io->io_hdr.io_type ==
546 			CTL_IO_TASK) ?
547 			tmp_io->taskio.tag_num :
548 			tmp_io->scsiio.tag_num,
549 		        tmp_io->io_hdr.flags,
550 			tmp_io->io_hdr.status);
551 	}
552 #endif
553 	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
554 	ctl_enqueue_isc((union ctl_io *)ctsio);
555 }
556 
557 /*
558  * ISC (Inter Shelf Communication) event handler.  Events from the HA
559  * subsystem come in here.
560  */
561 static void
562 ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
563 {
564 	struct ctl_softc *ctl_softc;
565 	union ctl_io *io;
566 	struct ctl_prio *presio;
567 	ctl_ha_status isc_status;
568 
569 	ctl_softc = control_softc;
570 	io = NULL;
571 
572 
573 #if 0
574 	printf("CTL: Isc Msg event %d\n", event);
575 #endif
576 	if (event == CTL_HA_EVT_MSG_RECV) {
577 		union ctl_ha_msg msg_info;
578 
579 		isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
580 					     sizeof(msg_info), /*wait*/ 0);
581 #if 0
582 		printf("CTL: msg_type %d\n", msg_info.msg_type);
583 #endif
584 		if (isc_status != 0) {
585 			printf("Error receiving message, status = %d\n",
586 			       isc_status);
587 			return;
588 		}
589 
590 		switch (msg_info.hdr.msg_type) {
591 		case CTL_MSG_SERIALIZE:
592 #if 0
593 			printf("Serialize\n");
594 #endif
595 			io = ctl_alloc_io((void *)ctl_softc->othersc_pool);
596 			if (io == NULL) {
597 				printf("ctl_isc_event_handler: can't allocate "
598 				       "ctl_io!\n");
599 				/* Bad Juju */
600 				/* Need to set busy and send msg back */
601 				msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
602 				msg_info.hdr.status = CTL_SCSI_ERROR;
603 				msg_info.scsi.scsi_status = SCSI_STATUS_BUSY;
604 				msg_info.scsi.sense_len = 0;
605 			        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
606 				    sizeof(msg_info), 0) > CTL_HA_STATUS_SUCCESS){
607 				}
608 				goto bailout;
609 			}
610 			ctl_zero_io(io);
611 			// populate ctsio from msg_info
612 			io->io_hdr.io_type = CTL_IO_SCSI;
613 			io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
614 			io->io_hdr.original_sc = msg_info.hdr.original_sc;
615 #if 0
616 			printf("pOrig %x\n", (int)msg_info.original_sc);
617 #endif
618 			io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
619 					    CTL_FLAG_IO_ACTIVE;
620 			/*
621 			 * If we're in serialization-only mode, we don't
622 			 * want to go through full done processing.  Thus
623 			 * the COPY flag.
624 			 *
625 			 * XXX KDM add another flag that is more specific.
626 			 */
627 			if (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)
628 				io->io_hdr.flags |= CTL_FLAG_INT_COPY;
629 			io->io_hdr.nexus = msg_info.hdr.nexus;
630 #if 0
631 			printf("targ %d, port %d, iid %d, lun %d\n",
632 			       io->io_hdr.nexus.targ_target.id,
633 			       io->io_hdr.nexus.targ_port,
634 			       io->io_hdr.nexus.initid.id,
635 			       io->io_hdr.nexus.targ_lun);
636 #endif
637 			io->scsiio.tag_num = msg_info.scsi.tag_num;
638 			io->scsiio.tag_type = msg_info.scsi.tag_type;
639 			memcpy(io->scsiio.cdb, msg_info.scsi.cdb,
640 			       CTL_MAX_CDBLEN);
641 			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
642 				const struct ctl_cmd_entry *entry;
643 
644 				entry = ctl_get_cmd_entry(&io->scsiio);
645 				io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
646 				io->io_hdr.flags |=
647 					entry->flags & CTL_FLAG_DATA_MASK;
648 			}
649 			ctl_enqueue_isc(io);
650 			break;
651 
652 		/* Performed on the Originating SC, XFER mode only */
653 		case CTL_MSG_DATAMOVE: {
654 			struct ctl_sg_entry *sgl;
655 			int i, j;
656 
657 			io = msg_info.hdr.original_sc;
658 			if (io == NULL) {
659 				printf("%s: original_sc == NULL!\n", __func__);
660 				/* XXX KDM do something here */
661 				break;
662 			}
663 			io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
664 			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
665 			/*
666 			 * Keep track of this, we need to send it back over
667 			 * when the datamove is complete.
668 			 */
669 			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
670 
671 			if (msg_info.dt.sg_sequence == 0) {
672 				/*
673 				 * XXX KDM we use the preallocated S/G list
674 				 * here, but we'll need to change this to
675 				 * dynamic allocation if we need larger S/G
676 				 * lists.
677 				 */
678 				if (msg_info.dt.kern_sg_entries >
679 				    sizeof(io->io_hdr.remote_sglist) /
680 				    sizeof(io->io_hdr.remote_sglist[0])) {
681 					printf("%s: number of S/G entries "
682 					    "needed %u > allocated num %zd\n",
683 					    __func__,
684 					    msg_info.dt.kern_sg_entries,
685 					    sizeof(io->io_hdr.remote_sglist)/
686 					    sizeof(io->io_hdr.remote_sglist[0]));
687 
688 					/*
689 					 * XXX KDM send a message back to
690 					 * the other side to shut down the
691 					 * DMA.  The error will come back
692 					 * through via the normal channel.
693 					 */
694 					break;
695 				}
696 				sgl = io->io_hdr.remote_sglist;
697 				memset(sgl, 0,
698 				       sizeof(io->io_hdr.remote_sglist));
699 
700 				io->scsiio.kern_data_ptr = (uint8_t *)sgl;
701 
702 				io->scsiio.kern_sg_entries =
703 					msg_info.dt.kern_sg_entries;
704 				io->scsiio.rem_sg_entries =
705 					msg_info.dt.kern_sg_entries;
706 				io->scsiio.kern_data_len =
707 					msg_info.dt.kern_data_len;
708 				io->scsiio.kern_total_len =
709 					msg_info.dt.kern_total_len;
710 				io->scsiio.kern_data_resid =
711 					msg_info.dt.kern_data_resid;
712 				io->scsiio.kern_rel_offset =
713 					msg_info.dt.kern_rel_offset;
714 				/*
715 				 * Clear out per-DMA flags.
716 				 */
717 				io->io_hdr.flags &= ~CTL_FLAG_RDMA_MASK;
718 				/*
719 				 * Add per-DMA flags that are set for this
720 				 * particular DMA request.
721 				 */
722 				io->io_hdr.flags |= msg_info.dt.flags &
723 						    CTL_FLAG_RDMA_MASK;
724 			} else
725 				sgl = (struct ctl_sg_entry *)
726 					io->scsiio.kern_data_ptr;
727 
728 			for (i = msg_info.dt.sent_sg_entries, j = 0;
729 			     i < (msg_info.dt.sent_sg_entries +
730 			     msg_info.dt.cur_sg_entries); i++, j++) {
731 				sgl[i].addr = msg_info.dt.sg_list[j].addr;
732 				sgl[i].len = msg_info.dt.sg_list[j].len;
733 
734 #if 0
735 				printf("%s: L: %p,%d -> %p,%d j=%d, i=%d\n",
736 				       __func__,
737 				       msg_info.dt.sg_list[j].addr,
738 				       msg_info.dt.sg_list[j].len,
739 				       sgl[i].addr, sgl[i].len, j, i);
740 #endif
741 			}
742 #if 0
743 			memcpy(&sgl[msg_info.dt.sent_sg_entries],
744 			       msg_info.dt.sg_list,
745 			       sizeof(*sgl) * msg_info.dt.cur_sg_entries);
746 #endif
747 
748 			/*
749 			 * If this is the last piece of the I/O, we've got
750 			 * the full S/G list.  Queue processing in the thread.
751 			 * Otherwise wait for the next piece.
752 			 */
753 			if (msg_info.dt.sg_last != 0)
754 				ctl_enqueue_isc(io);
755 			break;
756 		}
757 		/* Performed on the Serializing (primary) SC, XFER mode only */
758 		case CTL_MSG_DATAMOVE_DONE: {
759 			if (msg_info.hdr.serializing_sc == NULL) {
760 				printf("%s: serializing_sc == NULL!\n",
761 				       __func__);
762 				/* XXX KDM now what? */
763 				break;
764 			}
765 			/*
766 			 * We grab the sense information here in case
767 			 * there was a failure, so we can return status
768 			 * back to the initiator.
769 			 */
770 			io = msg_info.hdr.serializing_sc;
771 			io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
772 			io->io_hdr.status = msg_info.hdr.status;
773 			io->scsiio.scsi_status = msg_info.scsi.scsi_status;
774 			io->scsiio.sense_len = msg_info.scsi.sense_len;
775 			io->scsiio.sense_residual =msg_info.scsi.sense_residual;
776 			io->io_hdr.port_status = msg_info.scsi.fetd_status;
777 			io->scsiio.residual = msg_info.scsi.residual;
778 			memcpy(&io->scsiio.sense_data,&msg_info.scsi.sense_data,
779 			       sizeof(io->scsiio.sense_data));
780 			ctl_enqueue_isc(io);
781 			break;
782 		}
783 
784 		/* Preformed on Originating SC, SER_ONLY mode */
785 		case CTL_MSG_R2R:
786 			io = msg_info.hdr.original_sc;
787 			if (io == NULL) {
788 				printf("%s: Major Bummer\n", __func__);
789 				return;
790 			} else {
791 #if 0
792 				printf("pOrig %x\n",(int) ctsio);
793 #endif
794 			}
795 			io->io_hdr.msg_type = CTL_MSG_R2R;
796 			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
797 			ctl_enqueue_isc(io);
798 			break;
799 
800 		/*
801 		 * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
802 		 * mode.
803 		 * Performed on the Originating (i.e. secondary) SC in XFER
804 		 * mode
805 		 */
806 		case CTL_MSG_FINISH_IO:
807 			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER)
808 				ctl_isc_handler_finish_xfer(ctl_softc,
809 							    &msg_info);
810 			else
811 				ctl_isc_handler_finish_ser_only(ctl_softc,
812 								&msg_info);
813 			break;
814 
815 		/* Preformed on Originating SC */
816 		case CTL_MSG_BAD_JUJU:
817 			io = msg_info.hdr.original_sc;
818 			if (io == NULL) {
819 				printf("%s: Bad JUJU!, original_sc is NULL!\n",
820 				       __func__);
821 				break;
822 			}
823 			ctl_copy_sense_data(&msg_info, io);
824 			/*
825 			 * IO should have already been cleaned up on other
826 			 * SC so clear this flag so we won't send a message
827 			 * back to finish the IO there.
828 			 */
829 			io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
830 			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
831 
832 			/* io = msg_info.hdr.serializing_sc; */
833 			io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
834 			ctl_enqueue_isc(io);
835 			break;
836 
837 		/* Handle resets sent from the other side */
838 		case CTL_MSG_MANAGE_TASKS: {
839 			struct ctl_taskio *taskio;
840 			taskio = (struct ctl_taskio *)ctl_alloc_io(
841 				(void *)ctl_softc->othersc_pool);
842 			if (taskio == NULL) {
843 				printf("ctl_isc_event_handler: can't allocate "
844 				       "ctl_io!\n");
845 				/* Bad Juju */
846 				/* should I just call the proper reset func
847 				   here??? */
848 				goto bailout;
849 			}
850 			ctl_zero_io((union ctl_io *)taskio);
851 			taskio->io_hdr.io_type = CTL_IO_TASK;
852 			taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
853 			taskio->io_hdr.nexus = msg_info.hdr.nexus;
854 			taskio->task_action = msg_info.task.task_action;
855 			taskio->tag_num = msg_info.task.tag_num;
856 			taskio->tag_type = msg_info.task.tag_type;
857 #ifdef CTL_TIME_IO
858 			taskio->io_hdr.start_time = time_uptime;
859 			getbintime(&taskio->io_hdr.start_bt);
860 #if 0
861 			cs_prof_gettime(&taskio->io_hdr.start_ticks);
862 #endif
863 #endif /* CTL_TIME_IO */
864 			ctl_run_task((union ctl_io *)taskio);
865 			break;
866 		}
867 		/* Persistent Reserve action which needs attention */
868 		case CTL_MSG_PERS_ACTION:
869 			presio = (struct ctl_prio *)ctl_alloc_io(
870 				(void *)ctl_softc->othersc_pool);
871 			if (presio == NULL) {
872 				printf("ctl_isc_event_handler: can't allocate "
873 				       "ctl_io!\n");
874 				/* Bad Juju */
875 				/* Need to set busy and send msg back */
876 				goto bailout;
877 			}
878 			ctl_zero_io((union ctl_io *)presio);
879 			presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
880 			presio->pr_msg = msg_info.pr;
881 			ctl_enqueue_isc((union ctl_io *)presio);
882 			break;
883 		case CTL_MSG_SYNC_FE:
884 			rcv_sync_msg = 1;
885 			break;
886 		case CTL_MSG_APS_LOCK: {
887 			// It's quicker to execute this then to
888 			// queue it.
889 			struct ctl_lun *lun;
890 			struct ctl_page_index *page_index;
891 			struct copan_aps_subpage *current_sp;
892 			uint32_t targ_lun;
893 
894 			targ_lun = msg_info.hdr.nexus.targ_mapped_lun;
895 			lun = ctl_softc->ctl_luns[targ_lun];
896 			mtx_lock(&lun->lun_lock);
897 			page_index = &lun->mode_pages.index[index_to_aps_page];
898 			current_sp = (struct copan_aps_subpage *)
899 				     (page_index->page_data +
900 				     (page_index->page_len * CTL_PAGE_CURRENT));
901 
902 			current_sp->lock_active = msg_info.aps.lock_flag;
903 			mtx_unlock(&lun->lun_lock);
904 		        break;
905 		}
906 		default:
907 		        printf("How did I get here?\n");
908 		}
909 	} else if (event == CTL_HA_EVT_MSG_SENT) {
910 		if (param != CTL_HA_STATUS_SUCCESS) {
911 			printf("Bad status from ctl_ha_msg_send status %d\n",
912 			       param);
913 		}
914 		return;
915 	} else if (event == CTL_HA_EVT_DISCONNECT) {
916 		printf("CTL: Got a disconnect from Isc\n");
917 		return;
918 	} else {
919 		printf("ctl_isc_event_handler: Unknown event %d\n", event);
920 		return;
921 	}
922 
923 bailout:
924 	return;
925 }
926 
927 static void
928 ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
929 {
930 	struct scsi_sense_data *sense;
931 
932 	sense = &dest->scsiio.sense_data;
933 	bcopy(&src->scsi.sense_data, sense, sizeof(*sense));
934 	dest->scsiio.scsi_status = src->scsi.scsi_status;
935 	dest->scsiio.sense_len = src->scsi.sense_len;
936 	dest->io_hdr.status = src->hdr.status;
937 }
938 
939 static int
940 ctl_init(void)
941 {
942 	struct ctl_softc *softc;
943 	struct ctl_io_pool *internal_pool, *emergency_pool, *other_pool;
944 	struct ctl_port *port;
945         uint8_t sc_id =0;
946 	int i, error, retval;
947 	//int isc_retval;
948 
949 	retval = 0;
950 	ctl_pause_rtr = 0;
951         rcv_sync_msg = 0;
952 
953 	control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
954 			       M_WAITOK | M_ZERO);
955 	softc = control_softc;
956 
957 	softc->dev = make_dev(&ctl_cdevsw, 0, UID_ROOT, GID_OPERATOR, 0600,
958 			      "cam/ctl");
959 
960 	softc->dev->si_drv1 = softc;
961 
962 	/*
963 	 * By default, return a "bad LUN" peripheral qualifier for unknown
964 	 * LUNs.  The user can override this default using the tunable or
965 	 * sysctl.  See the comment in ctl_inquiry_std() for more details.
966 	 */
967 	softc->inquiry_pq_no_lun = 1;
968 	TUNABLE_INT_FETCH("kern.cam.ctl.inquiry_pq_no_lun",
969 			  &softc->inquiry_pq_no_lun);
970 	sysctl_ctx_init(&softc->sysctl_ctx);
971 	softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
972 		SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
973 		CTLFLAG_RD, 0, "CAM Target Layer");
974 
975 	if (softc->sysctl_tree == NULL) {
976 		printf("%s: unable to allocate sysctl tree\n", __func__);
977 		destroy_dev(softc->dev);
978 		free(control_softc, M_DEVBUF);
979 		control_softc = NULL;
980 		return (ENOMEM);
981 	}
982 
983 	SYSCTL_ADD_INT(&softc->sysctl_ctx,
984 		       SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO,
985 		       "inquiry_pq_no_lun", CTLFLAG_RW,
986 		       &softc->inquiry_pq_no_lun, 0,
987 		       "Report no lun possible for invalid LUNs");
988 
989 	mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
990 	mtx_init(&softc->pool_lock, "CTL pool mutex", NULL, MTX_DEF);
991 	softc->open_count = 0;
992 
993 	/*
994 	 * Default to actually sending a SYNCHRONIZE CACHE command down to
995 	 * the drive.
996 	 */
997 	softc->flags = CTL_FLAG_REAL_SYNC;
998 
999 	/*
1000 	 * In Copan's HA scheme, the "master" and "slave" roles are
1001 	 * figured out through the slot the controller is in.  Although it
1002 	 * is an active/active system, someone has to be in charge.
1003  	 */
1004 #ifdef NEEDTOPORT
1005         scmicro_rw(SCMICRO_GET_SHELF_ID, &sc_id);
1006 #endif
1007 
1008         if (sc_id == 0) {
1009 		softc->flags |= CTL_FLAG_MASTER_SHELF;
1010 		persis_offset = 0;
1011 	} else
1012 		persis_offset = CTL_MAX_INITIATORS;
1013 
1014 	/*
1015 	 * XXX KDM need to figure out where we want to get our target ID
1016 	 * and WWID.  Is it different on each port?
1017 	 */
1018 	softc->target.id = 0;
1019 	softc->target.wwid[0] = 0x12345678;
1020 	softc->target.wwid[1] = 0x87654321;
1021 	STAILQ_INIT(&softc->lun_list);
1022 	STAILQ_INIT(&softc->pending_lun_queue);
1023 	STAILQ_INIT(&softc->fe_list);
1024 	STAILQ_INIT(&softc->port_list);
1025 	STAILQ_INIT(&softc->be_list);
1026 	STAILQ_INIT(&softc->io_pools);
1027 	ctl_tpc_init(softc);
1028 
1029 	if (ctl_pool_create(softc, CTL_POOL_INTERNAL, CTL_POOL_ENTRIES_INTERNAL,
1030 			    &internal_pool)!= 0){
1031 		printf("ctl: can't allocate %d entry internal pool, "
1032 		       "exiting\n", CTL_POOL_ENTRIES_INTERNAL);
1033 		return (ENOMEM);
1034 	}
1035 
1036 	if (ctl_pool_create(softc, CTL_POOL_EMERGENCY,
1037 			    CTL_POOL_ENTRIES_EMERGENCY, &emergency_pool) != 0) {
1038 		printf("ctl: can't allocate %d entry emergency pool, "
1039 		       "exiting\n", CTL_POOL_ENTRIES_EMERGENCY);
1040 		ctl_pool_free(internal_pool);
1041 		return (ENOMEM);
1042 	}
1043 
1044 	if (ctl_pool_create(softc, CTL_POOL_4OTHERSC, CTL_POOL_ENTRIES_OTHER_SC,
1045 	                    &other_pool) != 0)
1046 	{
1047 		printf("ctl: can't allocate %d entry other SC pool, "
1048 		       "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1049 		ctl_pool_free(internal_pool);
1050 		ctl_pool_free(emergency_pool);
1051 		return (ENOMEM);
1052 	}
1053 
1054 	softc->internal_pool = internal_pool;
1055 	softc->emergency_pool = emergency_pool;
1056 	softc->othersc_pool = other_pool;
1057 
1058 	if (worker_threads <= 0)
1059 		worker_threads = max(1, mp_ncpus / 4);
1060 	if (worker_threads > CTL_MAX_THREADS)
1061 		worker_threads = CTL_MAX_THREADS;
1062 
1063 	for (i = 0; i < worker_threads; i++) {
1064 		struct ctl_thread *thr = &softc->threads[i];
1065 
1066 		mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1067 		thr->ctl_softc = softc;
1068 		STAILQ_INIT(&thr->incoming_queue);
1069 		STAILQ_INIT(&thr->rtr_queue);
1070 		STAILQ_INIT(&thr->done_queue);
1071 		STAILQ_INIT(&thr->isc_queue);
1072 
1073 		error = kproc_kthread_add(ctl_work_thread, thr,
1074 		    &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1075 		if (error != 0) {
1076 			printf("error creating CTL work thread!\n");
1077 			ctl_pool_free(internal_pool);
1078 			ctl_pool_free(emergency_pool);
1079 			ctl_pool_free(other_pool);
1080 			return (error);
1081 		}
1082 	}
1083 	error = kproc_kthread_add(ctl_lun_thread, softc,
1084 	    &softc->ctl_proc, NULL, 0, 0, "ctl", "lun");
1085 	if (error != 0) {
1086 		printf("error creating CTL lun thread!\n");
1087 		ctl_pool_free(internal_pool);
1088 		ctl_pool_free(emergency_pool);
1089 		ctl_pool_free(other_pool);
1090 		return (error);
1091 	}
1092 	if (bootverbose)
1093 		printf("ctl: CAM Target Layer loaded\n");
1094 
1095 	/*
1096 	 * Initialize the ioctl front end.
1097 	 */
1098 	ctl_frontend_register(&ioctl_frontend);
1099 	port = &softc->ioctl_info.port;
1100 	port->frontend = &ioctl_frontend;
1101 	sprintf(softc->ioctl_info.port_name, "ioctl");
1102 	port->port_type = CTL_PORT_IOCTL;
1103 	port->num_requested_ctl_io = 100;
1104 	port->port_name = softc->ioctl_info.port_name;
1105 	port->port_online = ctl_ioctl_online;
1106 	port->port_offline = ctl_ioctl_offline;
1107 	port->onoff_arg = &softc->ioctl_info;
1108 	port->lun_enable = ctl_ioctl_lun_enable;
1109 	port->lun_disable = ctl_ioctl_lun_disable;
1110 	port->targ_lun_arg = &softc->ioctl_info;
1111 	port->fe_datamove = ctl_ioctl_datamove;
1112 	port->fe_done = ctl_ioctl_done;
1113 	port->max_targets = 15;
1114 	port->max_target_id = 15;
1115 
1116 	if (ctl_port_register(&softc->ioctl_info.port,
1117 	                  (softc->flags & CTL_FLAG_MASTER_SHELF)) != 0) {
1118 		printf("ctl: ioctl front end registration failed, will "
1119 		       "continue anyway\n");
1120 	}
1121 
1122 #ifdef CTL_IO_DELAY
1123 	if (sizeof(struct callout) > CTL_TIMER_BYTES) {
1124 		printf("sizeof(struct callout) %zd > CTL_TIMER_BYTES %zd\n",
1125 		       sizeof(struct callout), CTL_TIMER_BYTES);
1126 		return (EINVAL);
1127 	}
1128 #endif /* CTL_IO_DELAY */
1129 
1130 	return (0);
1131 }
1132 
1133 void
1134 ctl_shutdown(void)
1135 {
1136 	struct ctl_softc *softc;
1137 	struct ctl_lun *lun, *next_lun;
1138 	struct ctl_io_pool *pool;
1139 
1140 	softc = (struct ctl_softc *)control_softc;
1141 
1142 	if (ctl_port_deregister(&softc->ioctl_info.port) != 0)
1143 		printf("ctl: ioctl front end deregistration failed\n");
1144 
1145 	mtx_lock(&softc->ctl_lock);
1146 
1147 	/*
1148 	 * Free up each LUN.
1149 	 */
1150 	for (lun = STAILQ_FIRST(&softc->lun_list); lun != NULL; lun = next_lun){
1151 		next_lun = STAILQ_NEXT(lun, links);
1152 		ctl_free_lun(lun);
1153 	}
1154 
1155 	mtx_unlock(&softc->ctl_lock);
1156 
1157 	ctl_frontend_deregister(&ioctl_frontend);
1158 
1159 	/*
1160 	 * This will rip the rug out from under any FETDs or anyone else
1161 	 * that has a pool allocated.  Since we increment our module
1162 	 * refcount any time someone outside the main CTL module allocates
1163 	 * a pool, we shouldn't have any problems here.  The user won't be
1164 	 * able to unload the CTL module until client modules have
1165 	 * successfully unloaded.
1166 	 */
1167 	while ((pool = STAILQ_FIRST(&softc->io_pools)) != NULL)
1168 		ctl_pool_free(pool);
1169 
1170 #if 0
1171 	ctl_shutdown_thread(softc->work_thread);
1172 	mtx_destroy(&softc->queue_lock);
1173 #endif
1174 
1175 	ctl_tpc_shutdown(softc);
1176 	mtx_destroy(&softc->pool_lock);
1177 	mtx_destroy(&softc->ctl_lock);
1178 
1179 	destroy_dev(softc->dev);
1180 
1181 	sysctl_ctx_free(&softc->sysctl_ctx);
1182 
1183 	free(control_softc, M_DEVBUF);
1184 	control_softc = NULL;
1185 
1186 	if (bootverbose)
1187 		printf("ctl: CAM Target Layer unloaded\n");
1188 }
1189 
1190 static int
1191 ctl_module_event_handler(module_t mod, int what, void *arg)
1192 {
1193 
1194 	switch (what) {
1195 	case MOD_LOAD:
1196 		return (ctl_init());
1197 	case MOD_UNLOAD:
1198 		return (EBUSY);
1199 	default:
1200 		return (EOPNOTSUPP);
1201 	}
1202 }
1203 
1204 /*
1205  * XXX KDM should we do some access checks here?  Bump a reference count to
1206  * prevent a CTL module from being unloaded while someone has it open?
1207  */
1208 static int
1209 ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1210 {
1211 	return (0);
1212 }
1213 
1214 static int
1215 ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1216 {
1217 	return (0);
1218 }
1219 
1220 int
1221 ctl_port_enable(ctl_port_type port_type)
1222 {
1223 	struct ctl_softc *softc;
1224 	struct ctl_port *port;
1225 
1226 	if (ctl_is_single == 0) {
1227 		union ctl_ha_msg msg_info;
1228 		int isc_retval;
1229 
1230 #if 0
1231 		printf("%s: HA mode, synchronizing frontend enable\n",
1232 		        __func__);
1233 #endif
1234 		msg_info.hdr.msg_type = CTL_MSG_SYNC_FE;
1235 	        if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1236 		        sizeof(msg_info), 1 )) > CTL_HA_STATUS_SUCCESS) {
1237 			printf("Sync msg send error retval %d\n", isc_retval);
1238 		}
1239 		if (!rcv_sync_msg) {
1240 			isc_retval=ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
1241 			        sizeof(msg_info), 1);
1242 		}
1243 #if 0
1244         	printf("CTL:Frontend Enable\n");
1245 	} else {
1246 		printf("%s: single mode, skipping frontend synchronization\n",
1247 		        __func__);
1248 #endif
1249 	}
1250 
1251 	softc = control_softc;
1252 
1253 	STAILQ_FOREACH(port, &softc->port_list, links) {
1254 		if (port_type & port->port_type)
1255 		{
1256 #if 0
1257 			printf("port %d\n", port->targ_port);
1258 #endif
1259 			ctl_port_online(port);
1260 		}
1261 	}
1262 
1263 	return (0);
1264 }
1265 
1266 int
1267 ctl_port_disable(ctl_port_type port_type)
1268 {
1269 	struct ctl_softc *softc;
1270 	struct ctl_port *port;
1271 
1272 	softc = control_softc;
1273 
1274 	STAILQ_FOREACH(port, &softc->port_list, links) {
1275 		if (port_type & port->port_type)
1276 			ctl_port_offline(port);
1277 	}
1278 
1279 	return (0);
1280 }
1281 
1282 /*
1283  * Returns 0 for success, 1 for failure.
1284  * Currently the only failure mode is if there aren't enough entries
1285  * allocated.  So, in case of a failure, look at num_entries_dropped,
1286  * reallocate and try again.
1287  */
1288 int
1289 ctl_port_list(struct ctl_port_entry *entries, int num_entries_alloced,
1290 	      int *num_entries_filled, int *num_entries_dropped,
1291 	      ctl_port_type port_type, int no_virtual)
1292 {
1293 	struct ctl_softc *softc;
1294 	struct ctl_port *port;
1295 	int entries_dropped, entries_filled;
1296 	int retval;
1297 	int i;
1298 
1299 	softc = control_softc;
1300 
1301 	retval = 0;
1302 	entries_filled = 0;
1303 	entries_dropped = 0;
1304 
1305 	i = 0;
1306 	mtx_lock(&softc->ctl_lock);
1307 	STAILQ_FOREACH(port, &softc->port_list, links) {
1308 		struct ctl_port_entry *entry;
1309 
1310 		if ((port->port_type & port_type) == 0)
1311 			continue;
1312 
1313 		if ((no_virtual != 0)
1314 		 && (port->virtual_port != 0))
1315 			continue;
1316 
1317 		if (entries_filled >= num_entries_alloced) {
1318 			entries_dropped++;
1319 			continue;
1320 		}
1321 		entry = &entries[i];
1322 
1323 		entry->port_type = port->port_type;
1324 		strlcpy(entry->port_name, port->port_name,
1325 			sizeof(entry->port_name));
1326 		entry->physical_port = port->physical_port;
1327 		entry->virtual_port = port->virtual_port;
1328 		entry->wwnn = port->wwnn;
1329 		entry->wwpn = port->wwpn;
1330 
1331 		i++;
1332 		entries_filled++;
1333 	}
1334 
1335 	mtx_unlock(&softc->ctl_lock);
1336 
1337 	if (entries_dropped > 0)
1338 		retval = 1;
1339 
1340 	*num_entries_dropped = entries_dropped;
1341 	*num_entries_filled = entries_filled;
1342 
1343 	return (retval);
1344 }
1345 
1346 static void
1347 ctl_ioctl_online(void *arg)
1348 {
1349 	struct ctl_ioctl_info *ioctl_info;
1350 
1351 	ioctl_info = (struct ctl_ioctl_info *)arg;
1352 
1353 	ioctl_info->flags |= CTL_IOCTL_FLAG_ENABLED;
1354 }
1355 
1356 static void
1357 ctl_ioctl_offline(void *arg)
1358 {
1359 	struct ctl_ioctl_info *ioctl_info;
1360 
1361 	ioctl_info = (struct ctl_ioctl_info *)arg;
1362 
1363 	ioctl_info->flags &= ~CTL_IOCTL_FLAG_ENABLED;
1364 }
1365 
1366 /*
1367  * Remove an initiator by port number and initiator ID.
1368  * Returns 0 for success, -1 for failure.
1369  */
1370 int
1371 ctl_remove_initiator(struct ctl_port *port, int iid)
1372 {
1373 	struct ctl_softc *softc = control_softc;
1374 
1375 	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1376 
1377 	if (iid > CTL_MAX_INIT_PER_PORT) {
1378 		printf("%s: initiator ID %u > maximun %u!\n",
1379 		       __func__, iid, CTL_MAX_INIT_PER_PORT);
1380 		return (-1);
1381 	}
1382 
1383 	mtx_lock(&softc->ctl_lock);
1384 	port->wwpn_iid[iid].in_use--;
1385 	port->wwpn_iid[iid].last_use = time_uptime;
1386 	mtx_unlock(&softc->ctl_lock);
1387 
1388 	return (0);
1389 }
1390 
1391 /*
1392  * Add an initiator to the initiator map.
1393  * Returns iid for success, < 0 for failure.
1394  */
1395 int
1396 ctl_add_initiator(struct ctl_port *port, int iid, uint64_t wwpn, char *name)
1397 {
1398 	struct ctl_softc *softc = control_softc;
1399 	time_t best_time;
1400 	int i, best;
1401 
1402 	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1403 
1404 	if (iid >= CTL_MAX_INIT_PER_PORT) {
1405 		printf("%s: WWPN %#jx initiator ID %u > maximum %u!\n",
1406 		       __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
1407 		free(name, M_CTL);
1408 		return (-1);
1409 	}
1410 
1411 	mtx_lock(&softc->ctl_lock);
1412 
1413 	if (iid < 0 && (wwpn != 0 || name != NULL)) {
1414 		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1415 			if (wwpn != 0 && wwpn == port->wwpn_iid[i].wwpn) {
1416 				iid = i;
1417 				break;
1418 			}
1419 			if (name != NULL && port->wwpn_iid[i].name != NULL &&
1420 			    strcmp(name, port->wwpn_iid[i].name) == 0) {
1421 				iid = i;
1422 				break;
1423 			}
1424 		}
1425 	}
1426 
1427 	if (iid < 0) {
1428 		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1429 			if (port->wwpn_iid[i].in_use == 0 &&
1430 			    port->wwpn_iid[i].wwpn == 0 &&
1431 			    port->wwpn_iid[i].name == NULL) {
1432 				iid = i;
1433 				break;
1434 			}
1435 		}
1436 	}
1437 
1438 	if (iid < 0) {
1439 		best = -1;
1440 		best_time = INT32_MAX;
1441 		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1442 			if (port->wwpn_iid[i].in_use == 0) {
1443 				if (port->wwpn_iid[i].last_use < best_time) {
1444 					best = i;
1445 					best_time = port->wwpn_iid[i].last_use;
1446 				}
1447 			}
1448 		}
1449 		iid = best;
1450 	}
1451 
1452 	if (iid < 0) {
1453 		mtx_unlock(&softc->ctl_lock);
1454 		free(name, M_CTL);
1455 		return (-2);
1456 	}
1457 
1458 	if (port->wwpn_iid[iid].in_use > 0 && (wwpn != 0 || name != NULL)) {
1459 		/*
1460 		 * This is not an error yet.
1461 		 */
1462 		if (wwpn != 0 && wwpn == port->wwpn_iid[iid].wwpn) {
1463 #if 0
1464 			printf("%s: port %d iid %u WWPN %#jx arrived"
1465 			    " again\n", __func__, port->targ_port,
1466 			    iid, (uintmax_t)wwpn);
1467 #endif
1468 			goto take;
1469 		}
1470 		if (name != NULL && port->wwpn_iid[iid].name != NULL &&
1471 		    strcmp(name, port->wwpn_iid[iid].name) == 0) {
1472 #if 0
1473 			printf("%s: port %d iid %u name '%s' arrived"
1474 			    " again\n", __func__, port->targ_port,
1475 			    iid, name);
1476 #endif
1477 			goto take;
1478 		}
1479 
1480 		/*
1481 		 * This is an error, but what do we do about it?  The
1482 		 * driver is telling us we have a new WWPN for this
1483 		 * initiator ID, so we pretty much need to use it.
1484 		 */
1485 		printf("%s: port %d iid %u WWPN %#jx '%s' arrived,"
1486 		    " but WWPN %#jx '%s' is still at that address\n",
1487 		    __func__, port->targ_port, iid, wwpn, name,
1488 		    (uintmax_t)port->wwpn_iid[iid].wwpn,
1489 		    port->wwpn_iid[iid].name);
1490 
1491 		/*
1492 		 * XXX KDM clear have_ca and ua_pending on each LUN for
1493 		 * this initiator.
1494 		 */
1495 	}
1496 take:
1497 	free(port->wwpn_iid[iid].name, M_CTL);
1498 	port->wwpn_iid[iid].name = name;
1499 	port->wwpn_iid[iid].wwpn = wwpn;
1500 	port->wwpn_iid[iid].in_use++;
1501 	mtx_unlock(&softc->ctl_lock);
1502 
1503 	return (iid);
1504 }
1505 
1506 static int
1507 ctl_create_iid(struct ctl_port *port, int iid, uint8_t *buf)
1508 {
1509 	int len;
1510 
1511 	switch (port->port_type) {
1512 	case CTL_PORT_FC:
1513 	{
1514 		struct scsi_transportid_fcp *id =
1515 		    (struct scsi_transportid_fcp *)buf;
1516 		if (port->wwpn_iid[iid].wwpn == 0)
1517 			return (0);
1518 		memset(id, 0, sizeof(*id));
1519 		id->format_protocol = SCSI_PROTO_FC;
1520 		scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->n_port_name);
1521 		return (sizeof(*id));
1522 	}
1523 	case CTL_PORT_ISCSI:
1524 	{
1525 		struct scsi_transportid_iscsi_port *id =
1526 		    (struct scsi_transportid_iscsi_port *)buf;
1527 		if (port->wwpn_iid[iid].name == NULL)
1528 			return (0);
1529 		memset(id, 0, 256);
1530 		id->format_protocol = SCSI_TRN_ISCSI_FORMAT_PORT |
1531 		    SCSI_PROTO_ISCSI;
1532 		len = strlcpy(id->iscsi_name, port->wwpn_iid[iid].name, 252) + 1;
1533 		len = roundup2(min(len, 252), 4);
1534 		scsi_ulto2b(len, id->additional_length);
1535 		return (sizeof(*id) + len);
1536 	}
1537 	case CTL_PORT_SAS:
1538 	{
1539 		struct scsi_transportid_sas *id =
1540 		    (struct scsi_transportid_sas *)buf;
1541 		if (port->wwpn_iid[iid].wwpn == 0)
1542 			return (0);
1543 		memset(id, 0, sizeof(*id));
1544 		id->format_protocol = SCSI_PROTO_SAS;
1545 		scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->sas_address);
1546 		return (sizeof(*id));
1547 	}
1548 	default:
1549 	{
1550 		struct scsi_transportid_spi *id =
1551 		    (struct scsi_transportid_spi *)buf;
1552 		memset(id, 0, sizeof(*id));
1553 		id->format_protocol = SCSI_PROTO_SPI;
1554 		scsi_ulto2b(iid, id->scsi_addr);
1555 		scsi_ulto2b(port->targ_port, id->rel_trgt_port_id);
1556 		return (sizeof(*id));
1557 	}
1558 	}
1559 }
1560 
1561 static int
1562 ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id)
1563 {
1564 	return (0);
1565 }
1566 
1567 static int
1568 ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id)
1569 {
1570 	return (0);
1571 }
1572 
1573 /*
1574  * Data movement routine for the CTL ioctl frontend port.
1575  */
1576 static int
1577 ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio)
1578 {
1579 	struct ctl_sg_entry *ext_sglist, *kern_sglist;
1580 	struct ctl_sg_entry ext_entry, kern_entry;
1581 	int ext_sglen, ext_sg_entries, kern_sg_entries;
1582 	int ext_sg_start, ext_offset;
1583 	int len_to_copy, len_copied;
1584 	int kern_watermark, ext_watermark;
1585 	int ext_sglist_malloced;
1586 	int i, j;
1587 
1588 	ext_sglist_malloced = 0;
1589 	ext_sg_start = 0;
1590 	ext_offset = 0;
1591 
1592 	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove\n"));
1593 
1594 	/*
1595 	 * If this flag is set, fake the data transfer.
1596 	 */
1597 	if (ctsio->io_hdr.flags & CTL_FLAG_NO_DATAMOVE) {
1598 		ctsio->ext_data_filled = ctsio->ext_data_len;
1599 		goto bailout;
1600 	}
1601 
1602 	/*
1603 	 * To simplify things here, if we have a single buffer, stick it in
1604 	 * a S/G entry and just make it a single entry S/G list.
1605 	 */
1606 	if (ctsio->io_hdr.flags & CTL_FLAG_EDPTR_SGLIST) {
1607 		int len_seen;
1608 
1609 		ext_sglen = ctsio->ext_sg_entries * sizeof(*ext_sglist);
1610 
1611 		ext_sglist = (struct ctl_sg_entry *)malloc(ext_sglen, M_CTL,
1612 							   M_WAITOK);
1613 		ext_sglist_malloced = 1;
1614 		if (copyin(ctsio->ext_data_ptr, ext_sglist,
1615 				   ext_sglen) != 0) {
1616 			ctl_set_internal_failure(ctsio,
1617 						 /*sks_valid*/ 0,
1618 						 /*retry_count*/ 0);
1619 			goto bailout;
1620 		}
1621 		ext_sg_entries = ctsio->ext_sg_entries;
1622 		len_seen = 0;
1623 		for (i = 0; i < ext_sg_entries; i++) {
1624 			if ((len_seen + ext_sglist[i].len) >=
1625 			     ctsio->ext_data_filled) {
1626 				ext_sg_start = i;
1627 				ext_offset = ctsio->ext_data_filled - len_seen;
1628 				break;
1629 			}
1630 			len_seen += ext_sglist[i].len;
1631 		}
1632 	} else {
1633 		ext_sglist = &ext_entry;
1634 		ext_sglist->addr = ctsio->ext_data_ptr;
1635 		ext_sglist->len = ctsio->ext_data_len;
1636 		ext_sg_entries = 1;
1637 		ext_sg_start = 0;
1638 		ext_offset = ctsio->ext_data_filled;
1639 	}
1640 
1641 	if (ctsio->kern_sg_entries > 0) {
1642 		kern_sglist = (struct ctl_sg_entry *)ctsio->kern_data_ptr;
1643 		kern_sg_entries = ctsio->kern_sg_entries;
1644 	} else {
1645 		kern_sglist = &kern_entry;
1646 		kern_sglist->addr = ctsio->kern_data_ptr;
1647 		kern_sglist->len = ctsio->kern_data_len;
1648 		kern_sg_entries = 1;
1649 	}
1650 
1651 
1652 	kern_watermark = 0;
1653 	ext_watermark = ext_offset;
1654 	len_copied = 0;
1655 	for (i = ext_sg_start, j = 0;
1656 	     i < ext_sg_entries && j < kern_sg_entries;) {
1657 		uint8_t *ext_ptr, *kern_ptr;
1658 
1659 		len_to_copy = ctl_min(ext_sglist[i].len - ext_watermark,
1660 				      kern_sglist[j].len - kern_watermark);
1661 
1662 		ext_ptr = (uint8_t *)ext_sglist[i].addr;
1663 		ext_ptr = ext_ptr + ext_watermark;
1664 		if (ctsio->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
1665 			/*
1666 			 * XXX KDM fix this!
1667 			 */
1668 			panic("need to implement bus address support");
1669 #if 0
1670 			kern_ptr = bus_to_virt(kern_sglist[j].addr);
1671 #endif
1672 		} else
1673 			kern_ptr = (uint8_t *)kern_sglist[j].addr;
1674 		kern_ptr = kern_ptr + kern_watermark;
1675 
1676 		kern_watermark += len_to_copy;
1677 		ext_watermark += len_to_copy;
1678 
1679 		if ((ctsio->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
1680 		     CTL_FLAG_DATA_IN) {
1681 			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1682 					 "bytes to user\n", len_to_copy));
1683 			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1684 					 "to %p\n", kern_ptr, ext_ptr));
1685 			if (copyout(kern_ptr, ext_ptr, len_to_copy) != 0) {
1686 				ctl_set_internal_failure(ctsio,
1687 							 /*sks_valid*/ 0,
1688 							 /*retry_count*/ 0);
1689 				goto bailout;
1690 			}
1691 		} else {
1692 			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1693 					 "bytes from user\n", len_to_copy));
1694 			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1695 					 "to %p\n", ext_ptr, kern_ptr));
1696 			if (copyin(ext_ptr, kern_ptr, len_to_copy)!= 0){
1697 				ctl_set_internal_failure(ctsio,
1698 							 /*sks_valid*/ 0,
1699 							 /*retry_count*/0);
1700 				goto bailout;
1701 			}
1702 		}
1703 
1704 		len_copied += len_to_copy;
1705 
1706 		if (ext_sglist[i].len == ext_watermark) {
1707 			i++;
1708 			ext_watermark = 0;
1709 		}
1710 
1711 		if (kern_sglist[j].len == kern_watermark) {
1712 			j++;
1713 			kern_watermark = 0;
1714 		}
1715 	}
1716 
1717 	ctsio->ext_data_filled += len_copied;
1718 
1719 	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_sg_entries: %d, "
1720 			 "kern_sg_entries: %d\n", ext_sg_entries,
1721 			 kern_sg_entries));
1722 	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_data_len = %d, "
1723 			 "kern_data_len = %d\n", ctsio->ext_data_len,
1724 			 ctsio->kern_data_len));
1725 
1726 
1727 	/* XXX KDM set residual?? */
1728 bailout:
1729 
1730 	if (ext_sglist_malloced != 0)
1731 		free(ext_sglist, M_CTL);
1732 
1733 	return (CTL_RETVAL_COMPLETE);
1734 }
1735 
1736 /*
1737  * Serialize a command that went down the "wrong" side, and so was sent to
1738  * this controller for execution.  The logic is a little different than the
1739  * standard case in ctl_scsiio_precheck().  Errors in this case need to get
1740  * sent back to the other side, but in the success case, we execute the
1741  * command on this side (XFER mode) or tell the other side to execute it
1742  * (SER_ONLY mode).
1743  */
1744 static int
1745 ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
1746 {
1747 	struct ctl_softc *ctl_softc;
1748 	union ctl_ha_msg msg_info;
1749 	struct ctl_lun *lun;
1750 	int retval = 0;
1751 	uint32_t targ_lun;
1752 
1753 	ctl_softc = control_softc;
1754 
1755 	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
1756 	lun = ctl_softc->ctl_luns[targ_lun];
1757 	if (lun==NULL)
1758 	{
1759 		/*
1760 		 * Why isn't LUN defined? The other side wouldn't
1761 		 * send a cmd if the LUN is undefined.
1762 		 */
1763 		printf("%s: Bad JUJU!, LUN is NULL!\n", __func__);
1764 
1765 		/* "Logical unit not supported" */
1766 		ctl_set_sense_data(&msg_info.scsi.sense_data,
1767 				   lun,
1768 				   /*sense_format*/SSD_TYPE_NONE,
1769 				   /*current_error*/ 1,
1770 				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1771 				   /*asc*/ 0x25,
1772 				   /*ascq*/ 0x00,
1773 				   SSD_ELEM_NONE);
1774 
1775 		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1776 		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1777 		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1778 		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1779 		msg_info.hdr.serializing_sc = NULL;
1780 		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1781 	        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1782 				sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1783 		}
1784 		return(1);
1785 
1786 	}
1787 
1788 	mtx_lock(&lun->lun_lock);
1789     	TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1790 
1791 	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
1792 		(union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
1793 		 ooa_links))) {
1794 	case CTL_ACTION_BLOCK:
1795 		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
1796 		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
1797 				  blocked_links);
1798 		break;
1799 	case CTL_ACTION_PASS:
1800 	case CTL_ACTION_SKIP:
1801 		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
1802 			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
1803 			ctl_enqueue_rtr((union ctl_io *)ctsio);
1804 		} else {
1805 
1806 			/* send msg back to other side */
1807 			msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1808 			msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
1809 			msg_info.hdr.msg_type = CTL_MSG_R2R;
1810 #if 0
1811 			printf("2. pOrig %x\n", (int)msg_info.hdr.original_sc);
1812 #endif
1813 		        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1814 			    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1815 			}
1816 		}
1817 		break;
1818 	case CTL_ACTION_OVERLAP:
1819 		/* OVERLAPPED COMMANDS ATTEMPTED */
1820 		ctl_set_sense_data(&msg_info.scsi.sense_data,
1821 				   lun,
1822 				   /*sense_format*/SSD_TYPE_NONE,
1823 				   /*current_error*/ 1,
1824 				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1825 				   /*asc*/ 0x4E,
1826 				   /*ascq*/ 0x00,
1827 				   SSD_ELEM_NONE);
1828 
1829 		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1830 		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1831 		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1832 		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1833 		msg_info.hdr.serializing_sc = NULL;
1834 		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1835 #if 0
1836 		printf("BAD JUJU:Major Bummer Overlap\n");
1837 #endif
1838 		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1839 		retval = 1;
1840 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1841 		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1842 		}
1843 		break;
1844 	case CTL_ACTION_OVERLAP_TAG:
1845 		/* TAGGED OVERLAPPED COMMANDS (NN = QUEUE TAG) */
1846 		ctl_set_sense_data(&msg_info.scsi.sense_data,
1847 				   lun,
1848 				   /*sense_format*/SSD_TYPE_NONE,
1849 				   /*current_error*/ 1,
1850 				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1851 				   /*asc*/ 0x4D,
1852 				   /*ascq*/ ctsio->tag_num & 0xff,
1853 				   SSD_ELEM_NONE);
1854 
1855 		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1856 		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1857 		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1858 		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1859 		msg_info.hdr.serializing_sc = NULL;
1860 		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1861 #if 0
1862 		printf("BAD JUJU:Major Bummer Overlap Tag\n");
1863 #endif
1864 		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1865 		retval = 1;
1866 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1867 		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1868 		}
1869 		break;
1870 	case CTL_ACTION_ERROR:
1871 	default:
1872 		/* "Internal target failure" */
1873 		ctl_set_sense_data(&msg_info.scsi.sense_data,
1874 				   lun,
1875 				   /*sense_format*/SSD_TYPE_NONE,
1876 				   /*current_error*/ 1,
1877 				   /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
1878 				   /*asc*/ 0x44,
1879 				   /*ascq*/ 0x00,
1880 				   SSD_ELEM_NONE);
1881 
1882 		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1883 		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1884 		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1885 		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1886 		msg_info.hdr.serializing_sc = NULL;
1887 		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1888 #if 0
1889 		printf("BAD JUJU:Major Bummer HW Error\n");
1890 #endif
1891 		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1892 		retval = 1;
1893 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1894 		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1895 		}
1896 		break;
1897 	}
1898 	mtx_unlock(&lun->lun_lock);
1899 	return (retval);
1900 }
1901 
1902 static int
1903 ctl_ioctl_submit_wait(union ctl_io *io)
1904 {
1905 	struct ctl_fe_ioctl_params params;
1906 	ctl_fe_ioctl_state last_state;
1907 	int done, retval;
1908 
1909 	retval = 0;
1910 
1911 	bzero(&params, sizeof(params));
1912 
1913 	mtx_init(&params.ioctl_mtx, "ctliocmtx", NULL, MTX_DEF);
1914 	cv_init(&params.sem, "ctlioccv");
1915 	params.state = CTL_IOCTL_INPROG;
1916 	last_state = params.state;
1917 
1918 	io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = &params;
1919 
1920 	CTL_DEBUG_PRINT(("ctl_ioctl_submit_wait\n"));
1921 
1922 	/* This shouldn't happen */
1923 	if ((retval = ctl_queue(io)) != CTL_RETVAL_COMPLETE)
1924 		return (retval);
1925 
1926 	done = 0;
1927 
1928 	do {
1929 		mtx_lock(&params.ioctl_mtx);
1930 		/*
1931 		 * Check the state here, and don't sleep if the state has
1932 		 * already changed (i.e. wakeup has already occured, but we
1933 		 * weren't waiting yet).
1934 		 */
1935 		if (params.state == last_state) {
1936 			/* XXX KDM cv_wait_sig instead? */
1937 			cv_wait(&params.sem, &params.ioctl_mtx);
1938 		}
1939 		last_state = params.state;
1940 
1941 		switch (params.state) {
1942 		case CTL_IOCTL_INPROG:
1943 			/* Why did we wake up? */
1944 			/* XXX KDM error here? */
1945 			mtx_unlock(&params.ioctl_mtx);
1946 			break;
1947 		case CTL_IOCTL_DATAMOVE:
1948 			CTL_DEBUG_PRINT(("got CTL_IOCTL_DATAMOVE\n"));
1949 
1950 			/*
1951 			 * change last_state back to INPROG to avoid
1952 			 * deadlock on subsequent data moves.
1953 			 */
1954 			params.state = last_state = CTL_IOCTL_INPROG;
1955 
1956 			mtx_unlock(&params.ioctl_mtx);
1957 			ctl_ioctl_do_datamove(&io->scsiio);
1958 			/*
1959 			 * Note that in some cases, most notably writes,
1960 			 * this will queue the I/O and call us back later.
1961 			 * In other cases, generally reads, this routine
1962 			 * will immediately call back and wake us up,
1963 			 * probably using our own context.
1964 			 */
1965 			io->scsiio.be_move_done(io);
1966 			break;
1967 		case CTL_IOCTL_DONE:
1968 			mtx_unlock(&params.ioctl_mtx);
1969 			CTL_DEBUG_PRINT(("got CTL_IOCTL_DONE\n"));
1970 			done = 1;
1971 			break;
1972 		default:
1973 			mtx_unlock(&params.ioctl_mtx);
1974 			/* XXX KDM error here? */
1975 			break;
1976 		}
1977 	} while (done == 0);
1978 
1979 	mtx_destroy(&params.ioctl_mtx);
1980 	cv_destroy(&params.sem);
1981 
1982 	return (CTL_RETVAL_COMPLETE);
1983 }
1984 
1985 static void
1986 ctl_ioctl_datamove(union ctl_io *io)
1987 {
1988 	struct ctl_fe_ioctl_params *params;
1989 
1990 	params = (struct ctl_fe_ioctl_params *)
1991 		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
1992 
1993 	mtx_lock(&params->ioctl_mtx);
1994 	params->state = CTL_IOCTL_DATAMOVE;
1995 	cv_broadcast(&params->sem);
1996 	mtx_unlock(&params->ioctl_mtx);
1997 }
1998 
1999 static void
2000 ctl_ioctl_done(union ctl_io *io)
2001 {
2002 	struct ctl_fe_ioctl_params *params;
2003 
2004 	params = (struct ctl_fe_ioctl_params *)
2005 		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
2006 
2007 	mtx_lock(&params->ioctl_mtx);
2008 	params->state = CTL_IOCTL_DONE;
2009 	cv_broadcast(&params->sem);
2010 	mtx_unlock(&params->ioctl_mtx);
2011 }
2012 
2013 static void
2014 ctl_ioctl_hard_startstop_callback(void *arg, struct cfi_metatask *metatask)
2015 {
2016 	struct ctl_fe_ioctl_startstop_info *sd_info;
2017 
2018 	sd_info = (struct ctl_fe_ioctl_startstop_info *)arg;
2019 
2020 	sd_info->hs_info.status = metatask->status;
2021 	sd_info->hs_info.total_luns = metatask->taskinfo.startstop.total_luns;
2022 	sd_info->hs_info.luns_complete =
2023 		metatask->taskinfo.startstop.luns_complete;
2024 	sd_info->hs_info.luns_failed = metatask->taskinfo.startstop.luns_failed;
2025 
2026 	cv_broadcast(&sd_info->sem);
2027 }
2028 
2029 static void
2030 ctl_ioctl_bbrread_callback(void *arg, struct cfi_metatask *metatask)
2031 {
2032 	struct ctl_fe_ioctl_bbrread_info *fe_bbr_info;
2033 
2034 	fe_bbr_info = (struct ctl_fe_ioctl_bbrread_info *)arg;
2035 
2036 	mtx_lock(fe_bbr_info->lock);
2037 	fe_bbr_info->bbr_info->status = metatask->status;
2038 	fe_bbr_info->bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2039 	fe_bbr_info->wakeup_done = 1;
2040 	mtx_unlock(fe_bbr_info->lock);
2041 
2042 	cv_broadcast(&fe_bbr_info->sem);
2043 }
2044 
2045 /*
2046  * Returns 0 for success, errno for failure.
2047  */
2048 static int
2049 ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
2050 		   struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
2051 {
2052 	union ctl_io *io;
2053 	int retval;
2054 
2055 	retval = 0;
2056 
2057 	mtx_lock(&lun->lun_lock);
2058 	for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
2059 	     (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2060 	     ooa_links)) {
2061 		struct ctl_ooa_entry *entry;
2062 
2063 		/*
2064 		 * If we've got more than we can fit, just count the
2065 		 * remaining entries.
2066 		 */
2067 		if (*cur_fill_num >= ooa_hdr->alloc_num)
2068 			continue;
2069 
2070 		entry = &kern_entries[*cur_fill_num];
2071 
2072 		entry->tag_num = io->scsiio.tag_num;
2073 		entry->lun_num = lun->lun;
2074 #ifdef CTL_TIME_IO
2075 		entry->start_bt = io->io_hdr.start_bt;
2076 #endif
2077 		bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2078 		entry->cdb_len = io->scsiio.cdb_len;
2079 		if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
2080 			entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2081 
2082 		if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2083 			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2084 
2085 		if (io->io_hdr.flags & CTL_FLAG_ABORT)
2086 			entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2087 
2088 		if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2089 			entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2090 
2091 		if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2092 			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2093 	}
2094 	mtx_unlock(&lun->lun_lock);
2095 
2096 	return (retval);
2097 }
2098 
2099 static void *
2100 ctl_copyin_alloc(void *user_addr, int len, char *error_str,
2101 		 size_t error_str_len)
2102 {
2103 	void *kptr;
2104 
2105 	kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2106 
2107 	if (copyin(user_addr, kptr, len) != 0) {
2108 		snprintf(error_str, error_str_len, "Error copying %d bytes "
2109 			 "from user address %p to kernel address %p", len,
2110 			 user_addr, kptr);
2111 		free(kptr, M_CTL);
2112 		return (NULL);
2113 	}
2114 
2115 	return (kptr);
2116 }
2117 
2118 static void
2119 ctl_free_args(int num_args, struct ctl_be_arg *args)
2120 {
2121 	int i;
2122 
2123 	if (args == NULL)
2124 		return;
2125 
2126 	for (i = 0; i < num_args; i++) {
2127 		free(args[i].kname, M_CTL);
2128 		free(args[i].kvalue, M_CTL);
2129 	}
2130 
2131 	free(args, M_CTL);
2132 }
2133 
2134 static struct ctl_be_arg *
2135 ctl_copyin_args(int num_args, struct ctl_be_arg *uargs,
2136 		char *error_str, size_t error_str_len)
2137 {
2138 	struct ctl_be_arg *args;
2139 	int i;
2140 
2141 	args = ctl_copyin_alloc(uargs, num_args * sizeof(*args),
2142 				error_str, error_str_len);
2143 
2144 	if (args == NULL)
2145 		goto bailout;
2146 
2147 	for (i = 0; i < num_args; i++) {
2148 		args[i].kname = NULL;
2149 		args[i].kvalue = NULL;
2150 	}
2151 
2152 	for (i = 0; i < num_args; i++) {
2153 		uint8_t *tmpptr;
2154 
2155 		args[i].kname = ctl_copyin_alloc(args[i].name,
2156 			args[i].namelen, error_str, error_str_len);
2157 		if (args[i].kname == NULL)
2158 			goto bailout;
2159 
2160 		if (args[i].kname[args[i].namelen - 1] != '\0') {
2161 			snprintf(error_str, error_str_len, "Argument %d "
2162 				 "name is not NUL-terminated", i);
2163 			goto bailout;
2164 		}
2165 
2166 		if (args[i].flags & CTL_BEARG_RD) {
2167 			tmpptr = ctl_copyin_alloc(args[i].value,
2168 				args[i].vallen, error_str, error_str_len);
2169 			if (tmpptr == NULL)
2170 				goto bailout;
2171 			if ((args[i].flags & CTL_BEARG_ASCII)
2172 			 && (tmpptr[args[i].vallen - 1] != '\0')) {
2173 				snprintf(error_str, error_str_len, "Argument "
2174 				    "%d value is not NUL-terminated", i);
2175 				goto bailout;
2176 			}
2177 			args[i].kvalue = tmpptr;
2178 		} else {
2179 			args[i].kvalue = malloc(args[i].vallen,
2180 			    M_CTL, M_WAITOK | M_ZERO);
2181 		}
2182 	}
2183 
2184 	return (args);
2185 bailout:
2186 
2187 	ctl_free_args(num_args, args);
2188 
2189 	return (NULL);
2190 }
2191 
2192 static void
2193 ctl_copyout_args(int num_args, struct ctl_be_arg *args)
2194 {
2195 	int i;
2196 
2197 	for (i = 0; i < num_args; i++) {
2198 		if (args[i].flags & CTL_BEARG_WR)
2199 			copyout(args[i].kvalue, args[i].value, args[i].vallen);
2200 	}
2201 }
2202 
2203 /*
2204  * Escape characters that are illegal or not recommended in XML.
2205  */
2206 int
2207 ctl_sbuf_printf_esc(struct sbuf *sb, char *str)
2208 {
2209 	int retval;
2210 
2211 	retval = 0;
2212 
2213 	for (; *str; str++) {
2214 		switch (*str) {
2215 		case '&':
2216 			retval = sbuf_printf(sb, "&amp;");
2217 			break;
2218 		case '>':
2219 			retval = sbuf_printf(sb, "&gt;");
2220 			break;
2221 		case '<':
2222 			retval = sbuf_printf(sb, "&lt;");
2223 			break;
2224 		default:
2225 			retval = sbuf_putc(sb, *str);
2226 			break;
2227 		}
2228 
2229 		if (retval != 0)
2230 			break;
2231 
2232 	}
2233 
2234 	return (retval);
2235 }
2236 
2237 static int
2238 ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2239 	  struct thread *td)
2240 {
2241 	struct ctl_softc *softc;
2242 	int retval;
2243 
2244 	softc = control_softc;
2245 
2246 	retval = 0;
2247 
2248 	switch (cmd) {
2249 	case CTL_IO: {
2250 		union ctl_io *io;
2251 		void *pool_tmp;
2252 
2253 		/*
2254 		 * If we haven't been "enabled", don't allow any SCSI I/O
2255 		 * to this FETD.
2256 		 */
2257 		if ((softc->ioctl_info.flags & CTL_IOCTL_FLAG_ENABLED) == 0) {
2258 			retval = EPERM;
2259 			break;
2260 		}
2261 
2262 		io = ctl_alloc_io(softc->ioctl_info.port.ctl_pool_ref);
2263 		if (io == NULL) {
2264 			printf("ctl_ioctl: can't allocate ctl_io!\n");
2265 			retval = ENOSPC;
2266 			break;
2267 		}
2268 
2269 		/*
2270 		 * Need to save the pool reference so it doesn't get
2271 		 * spammed by the user's ctl_io.
2272 		 */
2273 		pool_tmp = io->io_hdr.pool;
2274 
2275 		memcpy(io, (void *)addr, sizeof(*io));
2276 
2277 		io->io_hdr.pool = pool_tmp;
2278 		/*
2279 		 * No status yet, so make sure the status is set properly.
2280 		 */
2281 		io->io_hdr.status = CTL_STATUS_NONE;
2282 
2283 		/*
2284 		 * The user sets the initiator ID, target and LUN IDs.
2285 		 */
2286 		io->io_hdr.nexus.targ_port = softc->ioctl_info.port.targ_port;
2287 		io->io_hdr.flags |= CTL_FLAG_USER_REQ;
2288 		if ((io->io_hdr.io_type == CTL_IO_SCSI)
2289 		 && (io->scsiio.tag_type != CTL_TAG_UNTAGGED))
2290 			io->scsiio.tag_num = softc->ioctl_info.cur_tag_num++;
2291 
2292 		retval = ctl_ioctl_submit_wait(io);
2293 
2294 		if (retval != 0) {
2295 			ctl_free_io(io);
2296 			break;
2297 		}
2298 
2299 		memcpy((void *)addr, io, sizeof(*io));
2300 
2301 		/* return this to our pool */
2302 		ctl_free_io(io);
2303 
2304 		break;
2305 	}
2306 	case CTL_ENABLE_PORT:
2307 	case CTL_DISABLE_PORT:
2308 	case CTL_SET_PORT_WWNS: {
2309 		struct ctl_port *port;
2310 		struct ctl_port_entry *entry;
2311 
2312 		entry = (struct ctl_port_entry *)addr;
2313 
2314 		mtx_lock(&softc->ctl_lock);
2315 		STAILQ_FOREACH(port, &softc->port_list, links) {
2316 			int action, done;
2317 
2318 			action = 0;
2319 			done = 0;
2320 
2321 			if ((entry->port_type == CTL_PORT_NONE)
2322 			 && (entry->targ_port == port->targ_port)) {
2323 				/*
2324 				 * If the user only wants to enable or
2325 				 * disable or set WWNs on a specific port,
2326 				 * do the operation and we're done.
2327 				 */
2328 				action = 1;
2329 				done = 1;
2330 			} else if (entry->port_type & port->port_type) {
2331 				/*
2332 				 * Compare the user's type mask with the
2333 				 * particular frontend type to see if we
2334 				 * have a match.
2335 				 */
2336 				action = 1;
2337 				done = 0;
2338 
2339 				/*
2340 				 * Make sure the user isn't trying to set
2341 				 * WWNs on multiple ports at the same time.
2342 				 */
2343 				if (cmd == CTL_SET_PORT_WWNS) {
2344 					printf("%s: Can't set WWNs on "
2345 					       "multiple ports\n", __func__);
2346 					retval = EINVAL;
2347 					break;
2348 				}
2349 			}
2350 			if (action != 0) {
2351 				/*
2352 				 * XXX KDM we have to drop the lock here,
2353 				 * because the online/offline operations
2354 				 * can potentially block.  We need to
2355 				 * reference count the frontends so they
2356 				 * can't go away,
2357 				 */
2358 				mtx_unlock(&softc->ctl_lock);
2359 
2360 				if (cmd == CTL_ENABLE_PORT) {
2361 					struct ctl_lun *lun;
2362 
2363 					STAILQ_FOREACH(lun, &softc->lun_list,
2364 						       links) {
2365 						port->lun_enable(port->targ_lun_arg,
2366 						    lun->target,
2367 						    lun->lun);
2368 					}
2369 
2370 					ctl_port_online(port);
2371 				} else if (cmd == CTL_DISABLE_PORT) {
2372 					struct ctl_lun *lun;
2373 
2374 					ctl_port_offline(port);
2375 
2376 					STAILQ_FOREACH(lun, &softc->lun_list,
2377 						       links) {
2378 						port->lun_disable(
2379 						    port->targ_lun_arg,
2380 						    lun->target,
2381 						    lun->lun);
2382 					}
2383 				}
2384 
2385 				mtx_lock(&softc->ctl_lock);
2386 
2387 				if (cmd == CTL_SET_PORT_WWNS)
2388 					ctl_port_set_wwns(port,
2389 					    (entry->flags & CTL_PORT_WWNN_VALID) ?
2390 					    1 : 0, entry->wwnn,
2391 					    (entry->flags & CTL_PORT_WWPN_VALID) ?
2392 					    1 : 0, entry->wwpn);
2393 			}
2394 			if (done != 0)
2395 				break;
2396 		}
2397 		mtx_unlock(&softc->ctl_lock);
2398 		break;
2399 	}
2400 	case CTL_GET_PORT_LIST: {
2401 		struct ctl_port *port;
2402 		struct ctl_port_list *list;
2403 		int i;
2404 
2405 		list = (struct ctl_port_list *)addr;
2406 
2407 		if (list->alloc_len != (list->alloc_num *
2408 		    sizeof(struct ctl_port_entry))) {
2409 			printf("%s: CTL_GET_PORT_LIST: alloc_len %u != "
2410 			       "alloc_num %u * sizeof(struct ctl_port_entry) "
2411 			       "%zu\n", __func__, list->alloc_len,
2412 			       list->alloc_num, sizeof(struct ctl_port_entry));
2413 			retval = EINVAL;
2414 			break;
2415 		}
2416 		list->fill_len = 0;
2417 		list->fill_num = 0;
2418 		list->dropped_num = 0;
2419 		i = 0;
2420 		mtx_lock(&softc->ctl_lock);
2421 		STAILQ_FOREACH(port, &softc->port_list, links) {
2422 			struct ctl_port_entry entry, *list_entry;
2423 
2424 			if (list->fill_num >= list->alloc_num) {
2425 				list->dropped_num++;
2426 				continue;
2427 			}
2428 
2429 			entry.port_type = port->port_type;
2430 			strlcpy(entry.port_name, port->port_name,
2431 				sizeof(entry.port_name));
2432 			entry.targ_port = port->targ_port;
2433 			entry.physical_port = port->physical_port;
2434 			entry.virtual_port = port->virtual_port;
2435 			entry.wwnn = port->wwnn;
2436 			entry.wwpn = port->wwpn;
2437 			if (port->status & CTL_PORT_STATUS_ONLINE)
2438 				entry.online = 1;
2439 			else
2440 				entry.online = 0;
2441 
2442 			list_entry = &list->entries[i];
2443 
2444 			retval = copyout(&entry, list_entry, sizeof(entry));
2445 			if (retval != 0) {
2446 				printf("%s: CTL_GET_PORT_LIST: copyout "
2447 				       "returned %d\n", __func__, retval);
2448 				break;
2449 			}
2450 			i++;
2451 			list->fill_num++;
2452 			list->fill_len += sizeof(entry);
2453 		}
2454 		mtx_unlock(&softc->ctl_lock);
2455 
2456 		/*
2457 		 * If this is non-zero, we had a copyout fault, so there's
2458 		 * probably no point in attempting to set the status inside
2459 		 * the structure.
2460 		 */
2461 		if (retval != 0)
2462 			break;
2463 
2464 		if (list->dropped_num > 0)
2465 			list->status = CTL_PORT_LIST_NEED_MORE_SPACE;
2466 		else
2467 			list->status = CTL_PORT_LIST_OK;
2468 		break;
2469 	}
2470 	case CTL_DUMP_OOA: {
2471 		struct ctl_lun *lun;
2472 		union ctl_io *io;
2473 		char printbuf[128];
2474 		struct sbuf sb;
2475 
2476 		mtx_lock(&softc->ctl_lock);
2477 		printf("Dumping OOA queues:\n");
2478 		STAILQ_FOREACH(lun, &softc->lun_list, links) {
2479 			mtx_lock(&lun->lun_lock);
2480 			for (io = (union ctl_io *)TAILQ_FIRST(
2481 			     &lun->ooa_queue); io != NULL;
2482 			     io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2483 			     ooa_links)) {
2484 				sbuf_new(&sb, printbuf, sizeof(printbuf),
2485 					 SBUF_FIXEDLEN);
2486 				sbuf_printf(&sb, "LUN %jd tag 0x%04x%s%s%s%s: ",
2487 					    (intmax_t)lun->lun,
2488 					    io->scsiio.tag_num,
2489 					    (io->io_hdr.flags &
2490 					    CTL_FLAG_BLOCKED) ? "" : " BLOCKED",
2491 					    (io->io_hdr.flags &
2492 					    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
2493 					    (io->io_hdr.flags &
2494 					    CTL_FLAG_ABORT) ? " ABORT" : "",
2495 			                    (io->io_hdr.flags &
2496 		                        CTL_FLAG_IS_WAS_ON_RTR) ? " RTR" : "");
2497 				ctl_scsi_command_string(&io->scsiio, NULL, &sb);
2498 				sbuf_finish(&sb);
2499 				printf("%s\n", sbuf_data(&sb));
2500 			}
2501 			mtx_unlock(&lun->lun_lock);
2502 		}
2503 		printf("OOA queues dump done\n");
2504 		mtx_unlock(&softc->ctl_lock);
2505 		break;
2506 	}
2507 	case CTL_GET_OOA: {
2508 		struct ctl_lun *lun;
2509 		struct ctl_ooa *ooa_hdr;
2510 		struct ctl_ooa_entry *entries;
2511 		uint32_t cur_fill_num;
2512 
2513 		ooa_hdr = (struct ctl_ooa *)addr;
2514 
2515 		if ((ooa_hdr->alloc_len == 0)
2516 		 || (ooa_hdr->alloc_num == 0)) {
2517 			printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2518 			       "must be non-zero\n", __func__,
2519 			       ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2520 			retval = EINVAL;
2521 			break;
2522 		}
2523 
2524 		if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2525 		    sizeof(struct ctl_ooa_entry))) {
2526 			printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2527 			       "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2528 			       __func__, ooa_hdr->alloc_len,
2529 			       ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2530 			retval = EINVAL;
2531 			break;
2532 		}
2533 
2534 		entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2535 		if (entries == NULL) {
2536 			printf("%s: could not allocate %d bytes for OOA "
2537 			       "dump\n", __func__, ooa_hdr->alloc_len);
2538 			retval = ENOMEM;
2539 			break;
2540 		}
2541 
2542 		mtx_lock(&softc->ctl_lock);
2543 		if (((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0)
2544 		 && ((ooa_hdr->lun_num > CTL_MAX_LUNS)
2545 		  || (softc->ctl_luns[ooa_hdr->lun_num] == NULL))) {
2546 			mtx_unlock(&softc->ctl_lock);
2547 			free(entries, M_CTL);
2548 			printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2549 			       __func__, (uintmax_t)ooa_hdr->lun_num);
2550 			retval = EINVAL;
2551 			break;
2552 		}
2553 
2554 		cur_fill_num = 0;
2555 
2556 		if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2557 			STAILQ_FOREACH(lun, &softc->lun_list, links) {
2558 				retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2559 					ooa_hdr, entries);
2560 				if (retval != 0)
2561 					break;
2562 			}
2563 			if (retval != 0) {
2564 				mtx_unlock(&softc->ctl_lock);
2565 				free(entries, M_CTL);
2566 				break;
2567 			}
2568 		} else {
2569 			lun = softc->ctl_luns[ooa_hdr->lun_num];
2570 
2571 			retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,ooa_hdr,
2572 						    entries);
2573 		}
2574 		mtx_unlock(&softc->ctl_lock);
2575 
2576 		ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2577 		ooa_hdr->fill_len = ooa_hdr->fill_num *
2578 			sizeof(struct ctl_ooa_entry);
2579 		retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2580 		if (retval != 0) {
2581 			printf("%s: error copying out %d bytes for OOA dump\n",
2582 			       __func__, ooa_hdr->fill_len);
2583 		}
2584 
2585 		getbintime(&ooa_hdr->cur_bt);
2586 
2587 		if (cur_fill_num > ooa_hdr->alloc_num) {
2588 			ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2589 			ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2590 		} else {
2591 			ooa_hdr->dropped_num = 0;
2592 			ooa_hdr->status = CTL_OOA_OK;
2593 		}
2594 
2595 		free(entries, M_CTL);
2596 		break;
2597 	}
2598 	case CTL_CHECK_OOA: {
2599 		union ctl_io *io;
2600 		struct ctl_lun *lun;
2601 		struct ctl_ooa_info *ooa_info;
2602 
2603 
2604 		ooa_info = (struct ctl_ooa_info *)addr;
2605 
2606 		if (ooa_info->lun_id >= CTL_MAX_LUNS) {
2607 			ooa_info->status = CTL_OOA_INVALID_LUN;
2608 			break;
2609 		}
2610 		mtx_lock(&softc->ctl_lock);
2611 		lun = softc->ctl_luns[ooa_info->lun_id];
2612 		if (lun == NULL) {
2613 			mtx_unlock(&softc->ctl_lock);
2614 			ooa_info->status = CTL_OOA_INVALID_LUN;
2615 			break;
2616 		}
2617 		mtx_lock(&lun->lun_lock);
2618 		mtx_unlock(&softc->ctl_lock);
2619 		ooa_info->num_entries = 0;
2620 		for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
2621 		     io != NULL; io = (union ctl_io *)TAILQ_NEXT(
2622 		     &io->io_hdr, ooa_links)) {
2623 			ooa_info->num_entries++;
2624 		}
2625 		mtx_unlock(&lun->lun_lock);
2626 
2627 		ooa_info->status = CTL_OOA_SUCCESS;
2628 
2629 		break;
2630 	}
2631 	case CTL_HARD_START:
2632 	case CTL_HARD_STOP: {
2633 		struct ctl_fe_ioctl_startstop_info ss_info;
2634 		struct cfi_metatask *metatask;
2635 		struct mtx hs_mtx;
2636 
2637 		mtx_init(&hs_mtx, "HS Mutex", NULL, MTX_DEF);
2638 
2639 		cv_init(&ss_info.sem, "hard start/stop cv" );
2640 
2641 		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2642 		if (metatask == NULL) {
2643 			retval = ENOMEM;
2644 			mtx_destroy(&hs_mtx);
2645 			break;
2646 		}
2647 
2648 		if (cmd == CTL_HARD_START)
2649 			metatask->tasktype = CFI_TASK_STARTUP;
2650 		else
2651 			metatask->tasktype = CFI_TASK_SHUTDOWN;
2652 
2653 		metatask->callback = ctl_ioctl_hard_startstop_callback;
2654 		metatask->callback_arg = &ss_info;
2655 
2656 		cfi_action(metatask);
2657 
2658 		/* Wait for the callback */
2659 		mtx_lock(&hs_mtx);
2660 		cv_wait_sig(&ss_info.sem, &hs_mtx);
2661 		mtx_unlock(&hs_mtx);
2662 
2663 		/*
2664 		 * All information has been copied from the metatask by the
2665 		 * time cv_broadcast() is called, so we free the metatask here.
2666 		 */
2667 		cfi_free_metatask(metatask);
2668 
2669 		memcpy((void *)addr, &ss_info.hs_info, sizeof(ss_info.hs_info));
2670 
2671 		mtx_destroy(&hs_mtx);
2672 		break;
2673 	}
2674 	case CTL_BBRREAD: {
2675 		struct ctl_bbrread_info *bbr_info;
2676 		struct ctl_fe_ioctl_bbrread_info fe_bbr_info;
2677 		struct mtx bbr_mtx;
2678 		struct cfi_metatask *metatask;
2679 
2680 		bbr_info = (struct ctl_bbrread_info *)addr;
2681 
2682 		bzero(&fe_bbr_info, sizeof(fe_bbr_info));
2683 
2684 		bzero(&bbr_mtx, sizeof(bbr_mtx));
2685 		mtx_init(&bbr_mtx, "BBR Mutex", NULL, MTX_DEF);
2686 
2687 		fe_bbr_info.bbr_info = bbr_info;
2688 		fe_bbr_info.lock = &bbr_mtx;
2689 
2690 		cv_init(&fe_bbr_info.sem, "BBR read cv");
2691 		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2692 
2693 		if (metatask == NULL) {
2694 			mtx_destroy(&bbr_mtx);
2695 			cv_destroy(&fe_bbr_info.sem);
2696 			retval = ENOMEM;
2697 			break;
2698 		}
2699 		metatask->tasktype = CFI_TASK_BBRREAD;
2700 		metatask->callback = ctl_ioctl_bbrread_callback;
2701 		metatask->callback_arg = &fe_bbr_info;
2702 		metatask->taskinfo.bbrread.lun_num = bbr_info->lun_num;
2703 		metatask->taskinfo.bbrread.lba = bbr_info->lba;
2704 		metatask->taskinfo.bbrread.len = bbr_info->len;
2705 
2706 		cfi_action(metatask);
2707 
2708 		mtx_lock(&bbr_mtx);
2709 		while (fe_bbr_info.wakeup_done == 0)
2710 			cv_wait_sig(&fe_bbr_info.sem, &bbr_mtx);
2711 		mtx_unlock(&bbr_mtx);
2712 
2713 		bbr_info->status = metatask->status;
2714 		bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2715 		bbr_info->scsi_status = metatask->taskinfo.bbrread.scsi_status;
2716 		memcpy(&bbr_info->sense_data,
2717 		       &metatask->taskinfo.bbrread.sense_data,
2718 		       ctl_min(sizeof(bbr_info->sense_data),
2719 			       sizeof(metatask->taskinfo.bbrread.sense_data)));
2720 
2721 		cfi_free_metatask(metatask);
2722 
2723 		mtx_destroy(&bbr_mtx);
2724 		cv_destroy(&fe_bbr_info.sem);
2725 
2726 		break;
2727 	}
2728 	case CTL_DELAY_IO: {
2729 		struct ctl_io_delay_info *delay_info;
2730 #ifdef CTL_IO_DELAY
2731 		struct ctl_lun *lun;
2732 #endif /* CTL_IO_DELAY */
2733 
2734 		delay_info = (struct ctl_io_delay_info *)addr;
2735 
2736 #ifdef CTL_IO_DELAY
2737 		mtx_lock(&softc->ctl_lock);
2738 
2739 		if ((delay_info->lun_id > CTL_MAX_LUNS)
2740 		 || (softc->ctl_luns[delay_info->lun_id] == NULL)) {
2741 			delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2742 		} else {
2743 			lun = softc->ctl_luns[delay_info->lun_id];
2744 			mtx_lock(&lun->lun_lock);
2745 
2746 			delay_info->status = CTL_DELAY_STATUS_OK;
2747 
2748 			switch (delay_info->delay_type) {
2749 			case CTL_DELAY_TYPE_CONT:
2750 				break;
2751 			case CTL_DELAY_TYPE_ONESHOT:
2752 				break;
2753 			default:
2754 				delay_info->status =
2755 					CTL_DELAY_STATUS_INVALID_TYPE;
2756 				break;
2757 			}
2758 
2759 			switch (delay_info->delay_loc) {
2760 			case CTL_DELAY_LOC_DATAMOVE:
2761 				lun->delay_info.datamove_type =
2762 					delay_info->delay_type;
2763 				lun->delay_info.datamove_delay =
2764 					delay_info->delay_secs;
2765 				break;
2766 			case CTL_DELAY_LOC_DONE:
2767 				lun->delay_info.done_type =
2768 					delay_info->delay_type;
2769 				lun->delay_info.done_delay =
2770 					delay_info->delay_secs;
2771 				break;
2772 			default:
2773 				delay_info->status =
2774 					CTL_DELAY_STATUS_INVALID_LOC;
2775 				break;
2776 			}
2777 			mtx_unlock(&lun->lun_lock);
2778 		}
2779 
2780 		mtx_unlock(&softc->ctl_lock);
2781 #else
2782 		delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2783 #endif /* CTL_IO_DELAY */
2784 		break;
2785 	}
2786 	case CTL_REALSYNC_SET: {
2787 		int *syncstate;
2788 
2789 		syncstate = (int *)addr;
2790 
2791 		mtx_lock(&softc->ctl_lock);
2792 		switch (*syncstate) {
2793 		case 0:
2794 			softc->flags &= ~CTL_FLAG_REAL_SYNC;
2795 			break;
2796 		case 1:
2797 			softc->flags |= CTL_FLAG_REAL_SYNC;
2798 			break;
2799 		default:
2800 			retval = EINVAL;
2801 			break;
2802 		}
2803 		mtx_unlock(&softc->ctl_lock);
2804 		break;
2805 	}
2806 	case CTL_REALSYNC_GET: {
2807 		int *syncstate;
2808 
2809 		syncstate = (int*)addr;
2810 
2811 		mtx_lock(&softc->ctl_lock);
2812 		if (softc->flags & CTL_FLAG_REAL_SYNC)
2813 			*syncstate = 1;
2814 		else
2815 			*syncstate = 0;
2816 		mtx_unlock(&softc->ctl_lock);
2817 
2818 		break;
2819 	}
2820 	case CTL_SETSYNC:
2821 	case CTL_GETSYNC: {
2822 		struct ctl_sync_info *sync_info;
2823 		struct ctl_lun *lun;
2824 
2825 		sync_info = (struct ctl_sync_info *)addr;
2826 
2827 		mtx_lock(&softc->ctl_lock);
2828 		lun = softc->ctl_luns[sync_info->lun_id];
2829 		if (lun == NULL) {
2830 			mtx_unlock(&softc->ctl_lock);
2831 			sync_info->status = CTL_GS_SYNC_NO_LUN;
2832 		}
2833 		/*
2834 		 * Get or set the sync interval.  We're not bounds checking
2835 		 * in the set case, hopefully the user won't do something
2836 		 * silly.
2837 		 */
2838 		mtx_lock(&lun->lun_lock);
2839 		mtx_unlock(&softc->ctl_lock);
2840 		if (cmd == CTL_GETSYNC)
2841 			sync_info->sync_interval = lun->sync_interval;
2842 		else
2843 			lun->sync_interval = sync_info->sync_interval;
2844 		mtx_unlock(&lun->lun_lock);
2845 
2846 		sync_info->status = CTL_GS_SYNC_OK;
2847 
2848 		break;
2849 	}
2850 	case CTL_GETSTATS: {
2851 		struct ctl_stats *stats;
2852 		struct ctl_lun *lun;
2853 		int i;
2854 
2855 		stats = (struct ctl_stats *)addr;
2856 
2857 		if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) >
2858 		     stats->alloc_len) {
2859 			stats->status = CTL_SS_NEED_MORE_SPACE;
2860 			stats->num_luns = softc->num_luns;
2861 			break;
2862 		}
2863 		/*
2864 		 * XXX KDM no locking here.  If the LUN list changes,
2865 		 * things can blow up.
2866 		 */
2867 		for (i = 0, lun = STAILQ_FIRST(&softc->lun_list); lun != NULL;
2868 		     i++, lun = STAILQ_NEXT(lun, links)) {
2869 			retval = copyout(&lun->stats, &stats->lun_stats[i],
2870 					 sizeof(lun->stats));
2871 			if (retval != 0)
2872 				break;
2873 		}
2874 		stats->num_luns = softc->num_luns;
2875 		stats->fill_len = sizeof(struct ctl_lun_io_stats) *
2876 				 softc->num_luns;
2877 		stats->status = CTL_SS_OK;
2878 #ifdef CTL_TIME_IO
2879 		stats->flags = CTL_STATS_FLAG_TIME_VALID;
2880 #else
2881 		stats->flags = CTL_STATS_FLAG_NONE;
2882 #endif
2883 		getnanouptime(&stats->timestamp);
2884 		break;
2885 	}
2886 	case CTL_ERROR_INJECT: {
2887 		struct ctl_error_desc *err_desc, *new_err_desc;
2888 		struct ctl_lun *lun;
2889 
2890 		err_desc = (struct ctl_error_desc *)addr;
2891 
2892 		new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2893 				      M_WAITOK | M_ZERO);
2894 		bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2895 
2896 		mtx_lock(&softc->ctl_lock);
2897 		lun = softc->ctl_luns[err_desc->lun_id];
2898 		if (lun == NULL) {
2899 			mtx_unlock(&softc->ctl_lock);
2900 			printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2901 			       __func__, (uintmax_t)err_desc->lun_id);
2902 			retval = EINVAL;
2903 			break;
2904 		}
2905 		mtx_lock(&lun->lun_lock);
2906 		mtx_unlock(&softc->ctl_lock);
2907 
2908 		/*
2909 		 * We could do some checking here to verify the validity
2910 		 * of the request, but given the complexity of error
2911 		 * injection requests, the checking logic would be fairly
2912 		 * complex.
2913 		 *
2914 		 * For now, if the request is invalid, it just won't get
2915 		 * executed and might get deleted.
2916 		 */
2917 		STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2918 
2919 		/*
2920 		 * XXX KDM check to make sure the serial number is unique,
2921 		 * in case we somehow manage to wrap.  That shouldn't
2922 		 * happen for a very long time, but it's the right thing to
2923 		 * do.
2924 		 */
2925 		new_err_desc->serial = lun->error_serial;
2926 		err_desc->serial = lun->error_serial;
2927 		lun->error_serial++;
2928 
2929 		mtx_unlock(&lun->lun_lock);
2930 		break;
2931 	}
2932 	case CTL_ERROR_INJECT_DELETE: {
2933 		struct ctl_error_desc *delete_desc, *desc, *desc2;
2934 		struct ctl_lun *lun;
2935 		int delete_done;
2936 
2937 		delete_desc = (struct ctl_error_desc *)addr;
2938 		delete_done = 0;
2939 
2940 		mtx_lock(&softc->ctl_lock);
2941 		lun = softc->ctl_luns[delete_desc->lun_id];
2942 		if (lun == NULL) {
2943 			mtx_unlock(&softc->ctl_lock);
2944 			printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2945 			       __func__, (uintmax_t)delete_desc->lun_id);
2946 			retval = EINVAL;
2947 			break;
2948 		}
2949 		mtx_lock(&lun->lun_lock);
2950 		mtx_unlock(&softc->ctl_lock);
2951 		STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2952 			if (desc->serial != delete_desc->serial)
2953 				continue;
2954 
2955 			STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2956 				      links);
2957 			free(desc, M_CTL);
2958 			delete_done = 1;
2959 		}
2960 		mtx_unlock(&lun->lun_lock);
2961 		if (delete_done == 0) {
2962 			printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2963 			       "error serial %ju on LUN %u\n", __func__,
2964 			       delete_desc->serial, delete_desc->lun_id);
2965 			retval = EINVAL;
2966 			break;
2967 		}
2968 		break;
2969 	}
2970 	case CTL_DUMP_STRUCTS: {
2971 		int i, j, k, idx;
2972 		struct ctl_port *port;
2973 		struct ctl_frontend *fe;
2974 
2975 		mtx_lock(&softc->ctl_lock);
2976 		printf("CTL Persistent Reservation information start:\n");
2977 		for (i = 0; i < CTL_MAX_LUNS; i++) {
2978 			struct ctl_lun *lun;
2979 
2980 			lun = softc->ctl_luns[i];
2981 
2982 			if ((lun == NULL)
2983 			 || ((lun->flags & CTL_LUN_DISABLED) != 0))
2984 				continue;
2985 
2986 			for (j = 0; j < (CTL_MAX_PORTS * 2); j++) {
2987 				for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2988 					idx = j * CTL_MAX_INIT_PER_PORT + k;
2989 					if (lun->per_res[idx].registered == 0)
2990 						continue;
2991 					printf("  LUN %d port %d iid %d key "
2992 					       "%#jx\n", i, j, k,
2993 					       (uintmax_t)scsi_8btou64(
2994 					       lun->per_res[idx].res_key.key));
2995 				}
2996 			}
2997 		}
2998 		printf("CTL Persistent Reservation information end\n");
2999 		printf("CTL Ports:\n");
3000 		STAILQ_FOREACH(port, &softc->port_list, links) {
3001 			printf("  Port %d '%s' Frontend '%s' Type %u pp %d vp %d WWNN "
3002 			       "%#jx WWPN %#jx\n", port->targ_port, port->port_name,
3003 			       port->frontend->name, port->port_type,
3004 			       port->physical_port, port->virtual_port,
3005 			       (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
3006 			for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3007 				if (port->wwpn_iid[j].in_use == 0 &&
3008 				    port->wwpn_iid[j].wwpn == 0 &&
3009 				    port->wwpn_iid[j].name == NULL)
3010 					continue;
3011 
3012 				printf("    iid %u use %d WWPN %#jx '%s'\n",
3013 				    j, port->wwpn_iid[j].in_use,
3014 				    (uintmax_t)port->wwpn_iid[j].wwpn,
3015 				    port->wwpn_iid[j].name);
3016 			}
3017 		}
3018 		printf("CTL Port information end\n");
3019 		mtx_unlock(&softc->ctl_lock);
3020 		/*
3021 		 * XXX KDM calling this without a lock.  We'd likely want
3022 		 * to drop the lock before calling the frontend's dump
3023 		 * routine anyway.
3024 		 */
3025 		printf("CTL Frontends:\n");
3026 		STAILQ_FOREACH(fe, &softc->fe_list, links) {
3027 			printf("  Frontend '%s'\n", fe->name);
3028 			if (fe->fe_dump != NULL)
3029 				fe->fe_dump();
3030 		}
3031 		printf("CTL Frontend information end\n");
3032 		break;
3033 	}
3034 	case CTL_LUN_REQ: {
3035 		struct ctl_lun_req *lun_req;
3036 		struct ctl_backend_driver *backend;
3037 
3038 		lun_req = (struct ctl_lun_req *)addr;
3039 
3040 		backend = ctl_backend_find(lun_req->backend);
3041 		if (backend == NULL) {
3042 			lun_req->status = CTL_LUN_ERROR;
3043 			snprintf(lun_req->error_str,
3044 				 sizeof(lun_req->error_str),
3045 				 "Backend \"%s\" not found.",
3046 				 lun_req->backend);
3047 			break;
3048 		}
3049 		if (lun_req->num_be_args > 0) {
3050 			lun_req->kern_be_args = ctl_copyin_args(
3051 				lun_req->num_be_args,
3052 				lun_req->be_args,
3053 				lun_req->error_str,
3054 				sizeof(lun_req->error_str));
3055 			if (lun_req->kern_be_args == NULL) {
3056 				lun_req->status = CTL_LUN_ERROR;
3057 				break;
3058 			}
3059 		}
3060 
3061 		retval = backend->ioctl(dev, cmd, addr, flag, td);
3062 
3063 		if (lun_req->num_be_args > 0) {
3064 			ctl_copyout_args(lun_req->num_be_args,
3065 				      lun_req->kern_be_args);
3066 			ctl_free_args(lun_req->num_be_args,
3067 				      lun_req->kern_be_args);
3068 		}
3069 		break;
3070 	}
3071 	case CTL_LUN_LIST: {
3072 		struct sbuf *sb;
3073 		struct ctl_lun *lun;
3074 		struct ctl_lun_list *list;
3075 		struct ctl_option *opt;
3076 
3077 		list = (struct ctl_lun_list *)addr;
3078 
3079 		/*
3080 		 * Allocate a fixed length sbuf here, based on the length
3081 		 * of the user's buffer.  We could allocate an auto-extending
3082 		 * buffer, and then tell the user how much larger our
3083 		 * amount of data is than his buffer, but that presents
3084 		 * some problems:
3085 		 *
3086 		 * 1.  The sbuf(9) routines use a blocking malloc, and so
3087 		 *     we can't hold a lock while calling them with an
3088 		 *     auto-extending buffer.
3089  		 *
3090 		 * 2.  There is not currently a LUN reference counting
3091 		 *     mechanism, outside of outstanding transactions on
3092 		 *     the LUN's OOA queue.  So a LUN could go away on us
3093 		 *     while we're getting the LUN number, backend-specific
3094 		 *     information, etc.  Thus, given the way things
3095 		 *     currently work, we need to hold the CTL lock while
3096 		 *     grabbing LUN information.
3097 		 *
3098 		 * So, from the user's standpoint, the best thing to do is
3099 		 * allocate what he thinks is a reasonable buffer length,
3100 		 * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
3101 		 * double the buffer length and try again.  (And repeat
3102 		 * that until he succeeds.)
3103 		 */
3104 		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3105 		if (sb == NULL) {
3106 			list->status = CTL_LUN_LIST_ERROR;
3107 			snprintf(list->error_str, sizeof(list->error_str),
3108 				 "Unable to allocate %d bytes for LUN list",
3109 				 list->alloc_len);
3110 			break;
3111 		}
3112 
3113 		sbuf_printf(sb, "<ctllunlist>\n");
3114 
3115 		mtx_lock(&softc->ctl_lock);
3116 		STAILQ_FOREACH(lun, &softc->lun_list, links) {
3117 			mtx_lock(&lun->lun_lock);
3118 			retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3119 					     (uintmax_t)lun->lun);
3120 
3121 			/*
3122 			 * Bail out as soon as we see that we've overfilled
3123 			 * the buffer.
3124 			 */
3125 			if (retval != 0)
3126 				break;
3127 
3128 			retval = sbuf_printf(sb, "\t<backend_type>%s"
3129 					     "</backend_type>\n",
3130 					     (lun->backend == NULL) ?  "none" :
3131 					     lun->backend->name);
3132 
3133 			if (retval != 0)
3134 				break;
3135 
3136 			retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
3137 					     lun->be_lun->lun_type);
3138 
3139 			if (retval != 0)
3140 				break;
3141 
3142 			if (lun->backend == NULL) {
3143 				retval = sbuf_printf(sb, "</lun>\n");
3144 				if (retval != 0)
3145 					break;
3146 				continue;
3147 			}
3148 
3149 			retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
3150 					     (lun->be_lun->maxlba > 0) ?
3151 					     lun->be_lun->maxlba + 1 : 0);
3152 
3153 			if (retval != 0)
3154 				break;
3155 
3156 			retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
3157 					     lun->be_lun->blocksize);
3158 
3159 			if (retval != 0)
3160 				break;
3161 
3162 			retval = sbuf_printf(sb, "\t<serial_number>");
3163 
3164 			if (retval != 0)
3165 				break;
3166 
3167 			retval = ctl_sbuf_printf_esc(sb,
3168 						     lun->be_lun->serial_num);
3169 
3170 			if (retval != 0)
3171 				break;
3172 
3173 			retval = sbuf_printf(sb, "</serial_number>\n");
3174 
3175 			if (retval != 0)
3176 				break;
3177 
3178 			retval = sbuf_printf(sb, "\t<device_id>");
3179 
3180 			if (retval != 0)
3181 				break;
3182 
3183 			retval = ctl_sbuf_printf_esc(sb,lun->be_lun->device_id);
3184 
3185 			if (retval != 0)
3186 				break;
3187 
3188 			retval = sbuf_printf(sb, "</device_id>\n");
3189 
3190 			if (retval != 0)
3191 				break;
3192 
3193 			if (lun->backend->lun_info != NULL) {
3194 				retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3195 				if (retval != 0)
3196 					break;
3197 			}
3198 			STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
3199 				retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3200 				    opt->name, opt->value, opt->name);
3201 				if (retval != 0)
3202 					break;
3203 			}
3204 
3205 			retval = sbuf_printf(sb, "</lun>\n");
3206 
3207 			if (retval != 0)
3208 				break;
3209 			mtx_unlock(&lun->lun_lock);
3210 		}
3211 		if (lun != NULL)
3212 			mtx_unlock(&lun->lun_lock);
3213 		mtx_unlock(&softc->ctl_lock);
3214 
3215 		if ((retval != 0)
3216 		 || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3217 			retval = 0;
3218 			sbuf_delete(sb);
3219 			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3220 			snprintf(list->error_str, sizeof(list->error_str),
3221 				 "Out of space, %d bytes is too small",
3222 				 list->alloc_len);
3223 			break;
3224 		}
3225 
3226 		sbuf_finish(sb);
3227 
3228 		retval = copyout(sbuf_data(sb), list->lun_xml,
3229 				 sbuf_len(sb) + 1);
3230 
3231 		list->fill_len = sbuf_len(sb) + 1;
3232 		list->status = CTL_LUN_LIST_OK;
3233 		sbuf_delete(sb);
3234 		break;
3235 	}
3236 	case CTL_ISCSI: {
3237 		struct ctl_iscsi *ci;
3238 		struct ctl_frontend *fe;
3239 
3240 		ci = (struct ctl_iscsi *)addr;
3241 
3242 		fe = ctl_frontend_find("iscsi");
3243 		if (fe == NULL) {
3244 			ci->status = CTL_ISCSI_ERROR;
3245 			snprintf(ci->error_str, sizeof(ci->error_str),
3246 			    "Frontend \"iscsi\" not found.");
3247 			break;
3248 		}
3249 
3250 		retval = fe->ioctl(dev, cmd, addr, flag, td);
3251 		break;
3252 	}
3253 	case CTL_PORT_REQ: {
3254 		struct ctl_req *req;
3255 		struct ctl_frontend *fe;
3256 
3257 		req = (struct ctl_req *)addr;
3258 
3259 		fe = ctl_frontend_find(req->driver);
3260 		if (fe == NULL) {
3261 			req->status = CTL_LUN_ERROR;
3262 			snprintf(req->error_str, sizeof(req->error_str),
3263 			    "Frontend \"%s\" not found.", req->driver);
3264 			break;
3265 		}
3266 		if (req->num_args > 0) {
3267 			req->kern_args = ctl_copyin_args(req->num_args,
3268 			    req->args, req->error_str, sizeof(req->error_str));
3269 			if (req->kern_args == NULL) {
3270 				req->status = CTL_LUN_ERROR;
3271 				break;
3272 			}
3273 		}
3274 
3275 		retval = fe->ioctl(dev, cmd, addr, flag, td);
3276 
3277 		if (req->num_args > 0) {
3278 			ctl_copyout_args(req->num_args, req->kern_args);
3279 			ctl_free_args(req->num_args, req->kern_args);
3280 		}
3281 		break;
3282 	}
3283 	case CTL_PORT_LIST: {
3284 		struct sbuf *sb;
3285 		struct ctl_port *port;
3286 		struct ctl_lun_list *list;
3287 		struct ctl_option *opt;
3288 
3289 		list = (struct ctl_lun_list *)addr;
3290 
3291 		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3292 		if (sb == NULL) {
3293 			list->status = CTL_LUN_LIST_ERROR;
3294 			snprintf(list->error_str, sizeof(list->error_str),
3295 				 "Unable to allocate %d bytes for LUN list",
3296 				 list->alloc_len);
3297 			break;
3298 		}
3299 
3300 		sbuf_printf(sb, "<ctlportlist>\n");
3301 
3302 		mtx_lock(&softc->ctl_lock);
3303 		STAILQ_FOREACH(port, &softc->port_list, links) {
3304 			retval = sbuf_printf(sb, "<targ_port id=\"%ju\">\n",
3305 					     (uintmax_t)port->targ_port);
3306 
3307 			/*
3308 			 * Bail out as soon as we see that we've overfilled
3309 			 * the buffer.
3310 			 */
3311 			if (retval != 0)
3312 				break;
3313 
3314 			retval = sbuf_printf(sb, "\t<frontend_type>%s"
3315 			    "</frontend_type>\n", port->frontend->name);
3316 			if (retval != 0)
3317 				break;
3318 
3319 			retval = sbuf_printf(sb, "\t<port_type>%d</port_type>\n",
3320 					     port->port_type);
3321 			if (retval != 0)
3322 				break;
3323 
3324 			retval = sbuf_printf(sb, "\t<online>%s</online>\n",
3325 			    (port->status & CTL_PORT_STATUS_ONLINE) ? "YES" : "NO");
3326 			if (retval != 0)
3327 				break;
3328 
3329 			retval = sbuf_printf(sb, "\t<port_name>%s</port_name>\n",
3330 			    port->port_name);
3331 			if (retval != 0)
3332 				break;
3333 
3334 			retval = sbuf_printf(sb, "\t<physical_port>%d</physical_port>\n",
3335 			    port->physical_port);
3336 			if (retval != 0)
3337 				break;
3338 
3339 			retval = sbuf_printf(sb, "\t<virtual_port>%d</virtual_port>\n",
3340 			    port->virtual_port);
3341 			if (retval != 0)
3342 				break;
3343 
3344 			retval = sbuf_printf(sb, "\t<wwnn>%#jx</wwnn>\n",
3345 			    (uintmax_t)port->wwnn);
3346 			if (retval != 0)
3347 				break;
3348 
3349 			retval = sbuf_printf(sb, "\t<wwpn>%#jx</wwpn>\n",
3350 			    (uintmax_t)port->wwpn);
3351 			if (retval != 0)
3352 				break;
3353 
3354 			if (port->port_info != NULL) {
3355 				retval = port->port_info(port->onoff_arg, sb);
3356 				if (retval != 0)
3357 					break;
3358 			}
3359 			STAILQ_FOREACH(opt, &port->options, links) {
3360 				retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3361 				    opt->name, opt->value, opt->name);
3362 				if (retval != 0)
3363 					break;
3364 			}
3365 
3366 			retval = sbuf_printf(sb, "</targ_port>\n");
3367 			if (retval != 0)
3368 				break;
3369 		}
3370 		mtx_unlock(&softc->ctl_lock);
3371 
3372 		if ((retval != 0)
3373 		 || ((retval = sbuf_printf(sb, "</ctlportlist>\n")) != 0)) {
3374 			retval = 0;
3375 			sbuf_delete(sb);
3376 			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3377 			snprintf(list->error_str, sizeof(list->error_str),
3378 				 "Out of space, %d bytes is too small",
3379 				 list->alloc_len);
3380 			break;
3381 		}
3382 
3383 		sbuf_finish(sb);
3384 
3385 		retval = copyout(sbuf_data(sb), list->lun_xml,
3386 				 sbuf_len(sb) + 1);
3387 
3388 		list->fill_len = sbuf_len(sb) + 1;
3389 		list->status = CTL_LUN_LIST_OK;
3390 		sbuf_delete(sb);
3391 		break;
3392 	}
3393 	default: {
3394 		/* XXX KDM should we fix this? */
3395 #if 0
3396 		struct ctl_backend_driver *backend;
3397 		unsigned int type;
3398 		int found;
3399 
3400 		found = 0;
3401 
3402 		/*
3403 		 * We encode the backend type as the ioctl type for backend
3404 		 * ioctls.  So parse it out here, and then search for a
3405 		 * backend of this type.
3406 		 */
3407 		type = _IOC_TYPE(cmd);
3408 
3409 		STAILQ_FOREACH(backend, &softc->be_list, links) {
3410 			if (backend->type == type) {
3411 				found = 1;
3412 				break;
3413 			}
3414 		}
3415 		if (found == 0) {
3416 			printf("ctl: unknown ioctl command %#lx or backend "
3417 			       "%d\n", cmd, type);
3418 			retval = EINVAL;
3419 			break;
3420 		}
3421 		retval = backend->ioctl(dev, cmd, addr, flag, td);
3422 #endif
3423 		retval = ENOTTY;
3424 		break;
3425 	}
3426 	}
3427 	return (retval);
3428 }
3429 
3430 uint32_t
3431 ctl_get_initindex(struct ctl_nexus *nexus)
3432 {
3433 	if (nexus->targ_port < CTL_MAX_PORTS)
3434 		return (nexus->initid.id +
3435 			(nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3436 	else
3437 		return (nexus->initid.id +
3438 		       ((nexus->targ_port - CTL_MAX_PORTS) *
3439 			CTL_MAX_INIT_PER_PORT));
3440 }
3441 
3442 uint32_t
3443 ctl_get_resindex(struct ctl_nexus *nexus)
3444 {
3445 	return (nexus->initid.id + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3446 }
3447 
3448 uint32_t
3449 ctl_port_idx(int port_num)
3450 {
3451 	if (port_num < CTL_MAX_PORTS)
3452 		return(port_num);
3453 	else
3454 		return(port_num - CTL_MAX_PORTS);
3455 }
3456 
3457 static uint32_t
3458 ctl_map_lun(int port_num, uint32_t lun_id)
3459 {
3460 	struct ctl_port *port;
3461 
3462 	port = control_softc->ctl_ports[ctl_port_idx(port_num)];
3463 	if (port == NULL)
3464 		return (UINT32_MAX);
3465 	if (port->lun_map == NULL)
3466 		return (lun_id);
3467 	return (port->lun_map(port->targ_lun_arg, lun_id));
3468 }
3469 
3470 static uint32_t
3471 ctl_map_lun_back(int port_num, uint32_t lun_id)
3472 {
3473 	struct ctl_port *port;
3474 	uint32_t i;
3475 
3476 	port = control_softc->ctl_ports[ctl_port_idx(port_num)];
3477 	if (port->lun_map == NULL)
3478 		return (lun_id);
3479 	for (i = 0; i < CTL_MAX_LUNS; i++) {
3480 		if (port->lun_map(port->targ_lun_arg, i) == lun_id)
3481 			return (i);
3482 	}
3483 	return (UINT32_MAX);
3484 }
3485 
3486 /*
3487  * Note:  This only works for bitmask sizes that are at least 32 bits, and
3488  * that are a power of 2.
3489  */
3490 int
3491 ctl_ffz(uint32_t *mask, uint32_t size)
3492 {
3493 	uint32_t num_chunks, num_pieces;
3494 	int i, j;
3495 
3496 	num_chunks = (size >> 5);
3497 	if (num_chunks == 0)
3498 		num_chunks++;
3499 	num_pieces = ctl_min((sizeof(uint32_t) * 8), size);
3500 
3501 	for (i = 0; i < num_chunks; i++) {
3502 		for (j = 0; j < num_pieces; j++) {
3503 			if ((mask[i] & (1 << j)) == 0)
3504 				return ((i << 5) + j);
3505 		}
3506 	}
3507 
3508 	return (-1);
3509 }
3510 
3511 int
3512 ctl_set_mask(uint32_t *mask, uint32_t bit)
3513 {
3514 	uint32_t chunk, piece;
3515 
3516 	chunk = bit >> 5;
3517 	piece = bit % (sizeof(uint32_t) * 8);
3518 
3519 	if ((mask[chunk] & (1 << piece)) != 0)
3520 		return (-1);
3521 	else
3522 		mask[chunk] |= (1 << piece);
3523 
3524 	return (0);
3525 }
3526 
3527 int
3528 ctl_clear_mask(uint32_t *mask, uint32_t bit)
3529 {
3530 	uint32_t chunk, piece;
3531 
3532 	chunk = bit >> 5;
3533 	piece = bit % (sizeof(uint32_t) * 8);
3534 
3535 	if ((mask[chunk] & (1 << piece)) == 0)
3536 		return (-1);
3537 	else
3538 		mask[chunk] &= ~(1 << piece);
3539 
3540 	return (0);
3541 }
3542 
3543 int
3544 ctl_is_set(uint32_t *mask, uint32_t bit)
3545 {
3546 	uint32_t chunk, piece;
3547 
3548 	chunk = bit >> 5;
3549 	piece = bit % (sizeof(uint32_t) * 8);
3550 
3551 	if ((mask[chunk] & (1 << piece)) == 0)
3552 		return (0);
3553 	else
3554 		return (1);
3555 }
3556 
3557 #ifdef unused
3558 /*
3559  * The bus, target and lun are optional, they can be filled in later.
3560  * can_wait is used to determine whether we can wait on the malloc or not.
3561  */
3562 union ctl_io*
3563 ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port, uint32_t targ_target,
3564 	      uint32_t targ_lun, int can_wait)
3565 {
3566 	union ctl_io *io;
3567 
3568 	if (can_wait)
3569 		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_WAITOK);
3570 	else
3571 		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_NOWAIT);
3572 
3573 	if (io != NULL) {
3574 		io->io_hdr.io_type = io_type;
3575 		io->io_hdr.targ_port = targ_port;
3576 		/*
3577 		 * XXX KDM this needs to change/go away.  We need to move
3578 		 * to a preallocated pool of ctl_scsiio structures.
3579 		 */
3580 		io->io_hdr.nexus.targ_target.id = targ_target;
3581 		io->io_hdr.nexus.targ_lun = targ_lun;
3582 	}
3583 
3584 	return (io);
3585 }
3586 
3587 void
3588 ctl_kfree_io(union ctl_io *io)
3589 {
3590 	free(io, M_CTL);
3591 }
3592 #endif /* unused */
3593 
3594 /*
3595  * ctl_softc, pool_type, total_ctl_io are passed in.
3596  * npool is passed out.
3597  */
3598 int
3599 ctl_pool_create(struct ctl_softc *ctl_softc, ctl_pool_type pool_type,
3600 		uint32_t total_ctl_io, struct ctl_io_pool **npool)
3601 {
3602 	uint32_t i;
3603 	union ctl_io *cur_io, *next_io;
3604 	struct ctl_io_pool *pool;
3605 	int retval;
3606 
3607 	retval = 0;
3608 
3609 	pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3610 					    M_NOWAIT | M_ZERO);
3611 	if (pool == NULL) {
3612 		retval = ENOMEM;
3613 		goto bailout;
3614 	}
3615 
3616 	pool->type = pool_type;
3617 	pool->ctl_softc = ctl_softc;
3618 
3619 	mtx_lock(&ctl_softc->pool_lock);
3620 	pool->id = ctl_softc->cur_pool_id++;
3621 	mtx_unlock(&ctl_softc->pool_lock);
3622 
3623 	pool->flags = CTL_POOL_FLAG_NONE;
3624 	pool->refcount = 1;		/* Reference for validity. */
3625 	STAILQ_INIT(&pool->free_queue);
3626 
3627 	/*
3628 	 * XXX KDM other options here:
3629 	 * - allocate a page at a time
3630 	 * - allocate one big chunk of memory.
3631 	 * Page allocation might work well, but would take a little more
3632 	 * tracking.
3633 	 */
3634 	for (i = 0; i < total_ctl_io; i++) {
3635 		cur_io = (union ctl_io *)malloc(sizeof(*cur_io), M_CTLIO,
3636 						M_NOWAIT);
3637 		if (cur_io == NULL) {
3638 			retval = ENOMEM;
3639 			break;
3640 		}
3641 		cur_io->io_hdr.pool = pool;
3642 		STAILQ_INSERT_TAIL(&pool->free_queue, &cur_io->io_hdr, links);
3643 		pool->total_ctl_io++;
3644 		pool->free_ctl_io++;
3645 	}
3646 
3647 	if (retval != 0) {
3648 		for (cur_io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3649 		     cur_io != NULL; cur_io = next_io) {
3650 			next_io = (union ctl_io *)STAILQ_NEXT(&cur_io->io_hdr,
3651 							      links);
3652 			STAILQ_REMOVE(&pool->free_queue, &cur_io->io_hdr,
3653 				      ctl_io_hdr, links);
3654 			free(cur_io, M_CTLIO);
3655 		}
3656 
3657 		free(pool, M_CTL);
3658 		goto bailout;
3659 	}
3660 	mtx_lock(&ctl_softc->pool_lock);
3661 	ctl_softc->num_pools++;
3662 	STAILQ_INSERT_TAIL(&ctl_softc->io_pools, pool, links);
3663 	/*
3664 	 * Increment our usage count if this is an external consumer, so we
3665 	 * can't get unloaded until the external consumer (most likely a
3666 	 * FETD) unloads and frees his pool.
3667 	 *
3668 	 * XXX KDM will this increment the caller's module use count, or
3669 	 * mine?
3670 	 */
3671 #if 0
3672 	if ((pool_type != CTL_POOL_EMERGENCY)
3673 	 && (pool_type != CTL_POOL_INTERNAL)
3674 	 && (pool_type != CTL_POOL_4OTHERSC))
3675 		MOD_INC_USE_COUNT;
3676 #endif
3677 
3678 	mtx_unlock(&ctl_softc->pool_lock);
3679 
3680 	*npool = pool;
3681 
3682 bailout:
3683 
3684 	return (retval);
3685 }
3686 
3687 static int
3688 ctl_pool_acquire(struct ctl_io_pool *pool)
3689 {
3690 
3691 	mtx_assert(&pool->ctl_softc->pool_lock, MA_OWNED);
3692 
3693 	if (pool->flags & CTL_POOL_FLAG_INVALID)
3694 		return (EINVAL);
3695 
3696 	pool->refcount++;
3697 
3698 	return (0);
3699 }
3700 
3701 static void
3702 ctl_pool_release(struct ctl_io_pool *pool)
3703 {
3704 	struct ctl_softc *ctl_softc = pool->ctl_softc;
3705 	union ctl_io *io;
3706 
3707 	mtx_assert(&ctl_softc->pool_lock, MA_OWNED);
3708 
3709 	if (--pool->refcount != 0)
3710 		return;
3711 
3712 	while ((io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue)) != NULL) {
3713 		STAILQ_REMOVE(&pool->free_queue, &io->io_hdr, ctl_io_hdr,
3714 			      links);
3715 		free(io, M_CTLIO);
3716 	}
3717 
3718 	STAILQ_REMOVE(&ctl_softc->io_pools, pool, ctl_io_pool, links);
3719 	ctl_softc->num_pools--;
3720 
3721 	/*
3722 	 * XXX KDM will this decrement the caller's usage count or mine?
3723 	 */
3724 #if 0
3725 	if ((pool->type != CTL_POOL_EMERGENCY)
3726 	 && (pool->type != CTL_POOL_INTERNAL)
3727 	 && (pool->type != CTL_POOL_4OTHERSC))
3728 		MOD_DEC_USE_COUNT;
3729 #endif
3730 
3731 	free(pool, M_CTL);
3732 }
3733 
3734 void
3735 ctl_pool_free(struct ctl_io_pool *pool)
3736 {
3737 	struct ctl_softc *ctl_softc;
3738 
3739 	if (pool == NULL)
3740 		return;
3741 
3742 	ctl_softc = pool->ctl_softc;
3743 	mtx_lock(&ctl_softc->pool_lock);
3744 	pool->flags |= CTL_POOL_FLAG_INVALID;
3745 	ctl_pool_release(pool);
3746 	mtx_unlock(&ctl_softc->pool_lock);
3747 }
3748 
3749 /*
3750  * This routine does not block (except for spinlocks of course).
3751  * It tries to allocate a ctl_io union from the caller's pool as quickly as
3752  * possible.
3753  */
3754 union ctl_io *
3755 ctl_alloc_io(void *pool_ref)
3756 {
3757 	union ctl_io *io;
3758 	struct ctl_softc *ctl_softc;
3759 	struct ctl_io_pool *pool, *npool;
3760 	struct ctl_io_pool *emergency_pool;
3761 
3762 	pool = (struct ctl_io_pool *)pool_ref;
3763 
3764 	if (pool == NULL) {
3765 		printf("%s: pool is NULL\n", __func__);
3766 		return (NULL);
3767 	}
3768 
3769 	emergency_pool = NULL;
3770 
3771 	ctl_softc = pool->ctl_softc;
3772 
3773 	mtx_lock(&ctl_softc->pool_lock);
3774 	/*
3775 	 * First, try to get the io structure from the user's pool.
3776 	 */
3777 	if (ctl_pool_acquire(pool) == 0) {
3778 		io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3779 		if (io != NULL) {
3780 			STAILQ_REMOVE_HEAD(&pool->free_queue, links);
3781 			pool->total_allocated++;
3782 			pool->free_ctl_io--;
3783 			mtx_unlock(&ctl_softc->pool_lock);
3784 			return (io);
3785 		} else
3786 			ctl_pool_release(pool);
3787 	}
3788 	/*
3789 	 * If he doesn't have any io structures left, search for an
3790 	 * emergency pool and grab one from there.
3791 	 */
3792 	STAILQ_FOREACH(npool, &ctl_softc->io_pools, links) {
3793 		if (npool->type != CTL_POOL_EMERGENCY)
3794 			continue;
3795 
3796 		if (ctl_pool_acquire(npool) != 0)
3797 			continue;
3798 
3799 		emergency_pool = npool;
3800 
3801 		io = (union ctl_io *)STAILQ_FIRST(&npool->free_queue);
3802 		if (io != NULL) {
3803 			STAILQ_REMOVE_HEAD(&npool->free_queue, links);
3804 			npool->total_allocated++;
3805 			npool->free_ctl_io--;
3806 			mtx_unlock(&ctl_softc->pool_lock);
3807 			return (io);
3808 		} else
3809 			ctl_pool_release(npool);
3810 	}
3811 
3812 	/* Drop the spinlock before we malloc */
3813 	mtx_unlock(&ctl_softc->pool_lock);
3814 
3815 	/*
3816 	 * The emergency pool (if it exists) didn't have one, so try an
3817 	 * atomic (i.e. nonblocking) malloc and see if we get lucky.
3818 	 */
3819 	io = (union ctl_io *)malloc(sizeof(*io), M_CTLIO, M_NOWAIT);
3820 	if (io != NULL) {
3821 		/*
3822 		 * If the emergency pool exists but is empty, add this
3823 		 * ctl_io to its list when it gets freed.
3824 		 */
3825 		if (emergency_pool != NULL) {
3826 			mtx_lock(&ctl_softc->pool_lock);
3827 			if (ctl_pool_acquire(emergency_pool) == 0) {
3828 				io->io_hdr.pool = emergency_pool;
3829 				emergency_pool->total_ctl_io++;
3830 				/*
3831 				 * Need to bump this, otherwise
3832 				 * total_allocated and total_freed won't
3833 				 * match when we no longer have anything
3834 				 * outstanding.
3835 				 */
3836 				emergency_pool->total_allocated++;
3837 			}
3838 			mtx_unlock(&ctl_softc->pool_lock);
3839 		} else
3840 			io->io_hdr.pool = NULL;
3841 	}
3842 
3843 	return (io);
3844 }
3845 
3846 void
3847 ctl_free_io(union ctl_io *io)
3848 {
3849 	if (io == NULL)
3850 		return;
3851 
3852 	/*
3853 	 * If this ctl_io has a pool, return it to that pool.
3854 	 */
3855 	if (io->io_hdr.pool != NULL) {
3856 		struct ctl_io_pool *pool;
3857 
3858 		pool = (struct ctl_io_pool *)io->io_hdr.pool;
3859 		mtx_lock(&pool->ctl_softc->pool_lock);
3860 		io->io_hdr.io_type = 0xff;
3861 		STAILQ_INSERT_TAIL(&pool->free_queue, &io->io_hdr, links);
3862 		pool->total_freed++;
3863 		pool->free_ctl_io++;
3864 		ctl_pool_release(pool);
3865 		mtx_unlock(&pool->ctl_softc->pool_lock);
3866 	} else {
3867 		/*
3868 		 * Otherwise, just free it.  We probably malloced it and
3869 		 * the emergency pool wasn't available.
3870 		 */
3871 		free(io, M_CTLIO);
3872 	}
3873 
3874 }
3875 
3876 void
3877 ctl_zero_io(union ctl_io *io)
3878 {
3879 	void *pool_ref;
3880 
3881 	if (io == NULL)
3882 		return;
3883 
3884 	/*
3885 	 * May need to preserve linked list pointers at some point too.
3886 	 */
3887 	pool_ref = io->io_hdr.pool;
3888 
3889 	memset(io, 0, sizeof(*io));
3890 
3891 	io->io_hdr.pool = pool_ref;
3892 }
3893 
3894 /*
3895  * This routine is currently used for internal copies of ctl_ios that need
3896  * to persist for some reason after we've already returned status to the
3897  * FETD.  (Thus the flag set.)
3898  *
3899  * XXX XXX
3900  * Note that this makes a blind copy of all fields in the ctl_io, except
3901  * for the pool reference.  This includes any memory that has been
3902  * allocated!  That memory will no longer be valid after done has been
3903  * called, so this would be VERY DANGEROUS for command that actually does
3904  * any reads or writes.  Right now (11/7/2005), this is only used for immediate
3905  * start and stop commands, which don't transfer any data, so this is not a
3906  * problem.  If it is used for anything else, the caller would also need to
3907  * allocate data buffer space and this routine would need to be modified to
3908  * copy the data buffer(s) as well.
3909  */
3910 void
3911 ctl_copy_io(union ctl_io *src, union ctl_io *dest)
3912 {
3913 	void *pool_ref;
3914 
3915 	if ((src == NULL)
3916 	 || (dest == NULL))
3917 		return;
3918 
3919 	/*
3920 	 * May need to preserve linked list pointers at some point too.
3921 	 */
3922 	pool_ref = dest->io_hdr.pool;
3923 
3924 	memcpy(dest, src, ctl_min(sizeof(*src), sizeof(*dest)));
3925 
3926 	dest->io_hdr.pool = pool_ref;
3927 	/*
3928 	 * We need to know that this is an internal copy, and doesn't need
3929 	 * to get passed back to the FETD that allocated it.
3930 	 */
3931 	dest->io_hdr.flags |= CTL_FLAG_INT_COPY;
3932 }
3933 
3934 #ifdef NEEDTOPORT
3935 static void
3936 ctl_update_power_subpage(struct copan_power_subpage *page)
3937 {
3938 	int num_luns, num_partitions, config_type;
3939 	struct ctl_softc *softc;
3940 	cs_BOOL_t aor_present, shelf_50pct_power;
3941 	cs_raidset_personality_t rs_type;
3942 	int max_active_luns;
3943 
3944 	softc = control_softc;
3945 
3946 	/* subtract out the processor LUN */
3947 	num_luns = softc->num_luns - 1;
3948 	/*
3949 	 * Default to 7 LUNs active, which was the only number we allowed
3950 	 * in the past.
3951 	 */
3952 	max_active_luns = 7;
3953 
3954 	num_partitions = config_GetRsPartitionInfo();
3955 	config_type = config_GetConfigType();
3956 	shelf_50pct_power = config_GetShelfPowerMode();
3957 	aor_present = config_IsAorRsPresent();
3958 
3959 	rs_type = ddb_GetRsRaidType(1);
3960 	if ((rs_type != CS_RAIDSET_PERSONALITY_RAID5)
3961 	 && (rs_type != CS_RAIDSET_PERSONALITY_RAID1)) {
3962 		EPRINT(0, "Unsupported RS type %d!", rs_type);
3963 	}
3964 
3965 
3966 	page->total_luns = num_luns;
3967 
3968 	switch (config_type) {
3969 	case 40:
3970 		/*
3971 		 * In a 40 drive configuration, it doesn't matter what DC
3972 		 * cards we have, whether we have AOR enabled or not,
3973 		 * partitioning or not, or what type of RAIDset we have.
3974 		 * In that scenario, we can power up every LUN we present
3975 		 * to the user.
3976 		 */
3977 		max_active_luns = num_luns;
3978 
3979 		break;
3980 	case 64:
3981 		if (shelf_50pct_power == CS_FALSE) {
3982 			/* 25% power */
3983 			if (aor_present == CS_TRUE) {
3984 				if (rs_type ==
3985 				     CS_RAIDSET_PERSONALITY_RAID5) {
3986 					max_active_luns = 7;
3987 				} else if (rs_type ==
3988 					 CS_RAIDSET_PERSONALITY_RAID1){
3989 					max_active_luns = 14;
3990 				} else {
3991 					/* XXX KDM now what?? */
3992 				}
3993 			} else {
3994 				if (rs_type ==
3995 				     CS_RAIDSET_PERSONALITY_RAID5) {
3996 					max_active_luns = 8;
3997 				} else if (rs_type ==
3998 					 CS_RAIDSET_PERSONALITY_RAID1){
3999 					max_active_luns = 16;
4000 				} else {
4001 					/* XXX KDM now what?? */
4002 				}
4003 			}
4004 		} else {
4005 			/* 50% power */
4006 			/*
4007 			 * With 50% power in a 64 drive configuration, we
4008 			 * can power all LUNs we present.
4009 			 */
4010 			max_active_luns = num_luns;
4011 		}
4012 		break;
4013 	case 112:
4014 		if (shelf_50pct_power == CS_FALSE) {
4015 			/* 25% power */
4016 			if (aor_present == CS_TRUE) {
4017 				if (rs_type ==
4018 				     CS_RAIDSET_PERSONALITY_RAID5) {
4019 					max_active_luns = 7;
4020 				} else if (rs_type ==
4021 					 CS_RAIDSET_PERSONALITY_RAID1){
4022 					max_active_luns = 14;
4023 				} else {
4024 					/* XXX KDM now what?? */
4025 				}
4026 			} else {
4027 				if (rs_type ==
4028 				     CS_RAIDSET_PERSONALITY_RAID5) {
4029 					max_active_luns = 8;
4030 				} else if (rs_type ==
4031 					 CS_RAIDSET_PERSONALITY_RAID1){
4032 					max_active_luns = 16;
4033 				} else {
4034 					/* XXX KDM now what?? */
4035 				}
4036 			}
4037 		} else {
4038 			/* 50% power */
4039 			if (aor_present == CS_TRUE) {
4040 				if (rs_type ==
4041 				     CS_RAIDSET_PERSONALITY_RAID5) {
4042 					max_active_luns = 14;
4043 				} else if (rs_type ==
4044 					 CS_RAIDSET_PERSONALITY_RAID1){
4045 					/*
4046 					 * We're assuming here that disk
4047 					 * caching is enabled, and so we're
4048 					 * able to power up half of each
4049 					 * LUN, and cache all writes.
4050 					 */
4051 					max_active_luns = num_luns;
4052 				} else {
4053 					/* XXX KDM now what?? */
4054 				}
4055 			} else {
4056 				if (rs_type ==
4057 				     CS_RAIDSET_PERSONALITY_RAID5) {
4058 					max_active_luns = 15;
4059 				} else if (rs_type ==
4060 					 CS_RAIDSET_PERSONALITY_RAID1){
4061 					max_active_luns = 30;
4062 				} else {
4063 					/* XXX KDM now what?? */
4064 				}
4065 			}
4066 		}
4067 		break;
4068 	default:
4069 		/*
4070 		 * In this case, we have an unknown configuration, so we
4071 		 * just use the default from above.
4072 		 */
4073 		break;
4074 	}
4075 
4076 	page->max_active_luns = max_active_luns;
4077 #if 0
4078 	printk("%s: total_luns = %d, max_active_luns = %d\n", __func__,
4079 	       page->total_luns, page->max_active_luns);
4080 #endif
4081 }
4082 #endif /* NEEDTOPORT */
4083 
4084 /*
4085  * This routine could be used in the future to load default and/or saved
4086  * mode page parameters for a particuar lun.
4087  */
4088 static int
4089 ctl_init_page_index(struct ctl_lun *lun)
4090 {
4091 	int i;
4092 	struct ctl_page_index *page_index;
4093 	struct ctl_softc *softc;
4094 
4095 	memcpy(&lun->mode_pages.index, page_index_template,
4096 	       sizeof(page_index_template));
4097 
4098 	softc = lun->ctl_softc;
4099 
4100 	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
4101 
4102 		page_index = &lun->mode_pages.index[i];
4103 		/*
4104 		 * If this is a disk-only mode page, there's no point in
4105 		 * setting it up.  For some pages, we have to have some
4106 		 * basic information about the disk in order to calculate the
4107 		 * mode page data.
4108 		 */
4109 		if ((lun->be_lun->lun_type != T_DIRECT)
4110 		 && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
4111 			continue;
4112 
4113 		switch (page_index->page_code & SMPH_PC_MASK) {
4114 		case SMS_FORMAT_DEVICE_PAGE: {
4115 			struct scsi_format_page *format_page;
4116 
4117 			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4118 				panic("subpage is incorrect!");
4119 
4120 			/*
4121 			 * Sectors per track are set above.  Bytes per
4122 			 * sector need to be set here on a per-LUN basis.
4123 			 */
4124 			memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
4125 			       &format_page_default,
4126 			       sizeof(format_page_default));
4127 			memcpy(&lun->mode_pages.format_page[
4128 			       CTL_PAGE_CHANGEABLE], &format_page_changeable,
4129 			       sizeof(format_page_changeable));
4130 			memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
4131 			       &format_page_default,
4132 			       sizeof(format_page_default));
4133 			memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
4134 			       &format_page_default,
4135 			       sizeof(format_page_default));
4136 
4137 			format_page = &lun->mode_pages.format_page[
4138 				CTL_PAGE_CURRENT];
4139 			scsi_ulto2b(lun->be_lun->blocksize,
4140 				    format_page->bytes_per_sector);
4141 
4142 			format_page = &lun->mode_pages.format_page[
4143 				CTL_PAGE_DEFAULT];
4144 			scsi_ulto2b(lun->be_lun->blocksize,
4145 				    format_page->bytes_per_sector);
4146 
4147 			format_page = &lun->mode_pages.format_page[
4148 				CTL_PAGE_SAVED];
4149 			scsi_ulto2b(lun->be_lun->blocksize,
4150 				    format_page->bytes_per_sector);
4151 
4152 			page_index->page_data =
4153 				(uint8_t *)lun->mode_pages.format_page;
4154 			break;
4155 		}
4156 		case SMS_RIGID_DISK_PAGE: {
4157 			struct scsi_rigid_disk_page *rigid_disk_page;
4158 			uint32_t sectors_per_cylinder;
4159 			uint64_t cylinders;
4160 #ifndef	__XSCALE__
4161 			int shift;
4162 #endif /* !__XSCALE__ */
4163 
4164 			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4165 				panic("invalid subpage value %d",
4166 				      page_index->subpage);
4167 
4168 			/*
4169 			 * Rotation rate and sectors per track are set
4170 			 * above.  We calculate the cylinders here based on
4171 			 * capacity.  Due to the number of heads and
4172 			 * sectors per track we're using, smaller arrays
4173 			 * may turn out to have 0 cylinders.  Linux and
4174 			 * FreeBSD don't pay attention to these mode pages
4175 			 * to figure out capacity, but Solaris does.  It
4176 			 * seems to deal with 0 cylinders just fine, and
4177 			 * works out a fake geometry based on the capacity.
4178 			 */
4179 			memcpy(&lun->mode_pages.rigid_disk_page[
4180 			       CTL_PAGE_CURRENT], &rigid_disk_page_default,
4181 			       sizeof(rigid_disk_page_default));
4182 			memcpy(&lun->mode_pages.rigid_disk_page[
4183 			       CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
4184 			       sizeof(rigid_disk_page_changeable));
4185 			memcpy(&lun->mode_pages.rigid_disk_page[
4186 			       CTL_PAGE_DEFAULT], &rigid_disk_page_default,
4187 			       sizeof(rigid_disk_page_default));
4188 			memcpy(&lun->mode_pages.rigid_disk_page[
4189 			       CTL_PAGE_SAVED], &rigid_disk_page_default,
4190 			       sizeof(rigid_disk_page_default));
4191 
4192 			sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
4193 				CTL_DEFAULT_HEADS;
4194 
4195 			/*
4196 			 * The divide method here will be more accurate,
4197 			 * probably, but results in floating point being
4198 			 * used in the kernel on i386 (__udivdi3()).  On the
4199 			 * XScale, though, __udivdi3() is implemented in
4200 			 * software.
4201 			 *
4202 			 * The shift method for cylinder calculation is
4203 			 * accurate if sectors_per_cylinder is a power of
4204 			 * 2.  Otherwise it might be slightly off -- you
4205 			 * might have a bit of a truncation problem.
4206 			 */
4207 #ifdef	__XSCALE__
4208 			cylinders = (lun->be_lun->maxlba + 1) /
4209 				sectors_per_cylinder;
4210 #else
4211 			for (shift = 31; shift > 0; shift--) {
4212 				if (sectors_per_cylinder & (1 << shift))
4213 					break;
4214 			}
4215 			cylinders = (lun->be_lun->maxlba + 1) >> shift;
4216 #endif
4217 
4218 			/*
4219 			 * We've basically got 3 bytes, or 24 bits for the
4220 			 * cylinder size in the mode page.  If we're over,
4221 			 * just round down to 2^24.
4222 			 */
4223 			if (cylinders > 0xffffff)
4224 				cylinders = 0xffffff;
4225 
4226 			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4227 				CTL_PAGE_CURRENT];
4228 			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4229 
4230 			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4231 				CTL_PAGE_DEFAULT];
4232 			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4233 
4234 			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4235 				CTL_PAGE_SAVED];
4236 			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4237 
4238 			page_index->page_data =
4239 				(uint8_t *)lun->mode_pages.rigid_disk_page;
4240 			break;
4241 		}
4242 		case SMS_CACHING_PAGE: {
4243 
4244 			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4245 				panic("invalid subpage value %d",
4246 				      page_index->subpage);
4247 			/*
4248 			 * Defaults should be okay here, no calculations
4249 			 * needed.
4250 			 */
4251 			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4252 			       &caching_page_default,
4253 			       sizeof(caching_page_default));
4254 			memcpy(&lun->mode_pages.caching_page[
4255 			       CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4256 			       sizeof(caching_page_changeable));
4257 			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4258 			       &caching_page_default,
4259 			       sizeof(caching_page_default));
4260 			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4261 			       &caching_page_default,
4262 			       sizeof(caching_page_default));
4263 			page_index->page_data =
4264 				(uint8_t *)lun->mode_pages.caching_page;
4265 			break;
4266 		}
4267 		case SMS_CONTROL_MODE_PAGE: {
4268 
4269 			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4270 				panic("invalid subpage value %d",
4271 				      page_index->subpage);
4272 
4273 			/*
4274 			 * Defaults should be okay here, no calculations
4275 			 * needed.
4276 			 */
4277 			memcpy(&lun->mode_pages.control_page[CTL_PAGE_CURRENT],
4278 			       &control_page_default,
4279 			       sizeof(control_page_default));
4280 			memcpy(&lun->mode_pages.control_page[
4281 			       CTL_PAGE_CHANGEABLE], &control_page_changeable,
4282 			       sizeof(control_page_changeable));
4283 			memcpy(&lun->mode_pages.control_page[CTL_PAGE_DEFAULT],
4284 			       &control_page_default,
4285 			       sizeof(control_page_default));
4286 			memcpy(&lun->mode_pages.control_page[CTL_PAGE_SAVED],
4287 			       &control_page_default,
4288 			       sizeof(control_page_default));
4289 			page_index->page_data =
4290 				(uint8_t *)lun->mode_pages.control_page;
4291 			break;
4292 
4293 		}
4294 		case SMS_VENDOR_SPECIFIC_PAGE:{
4295 			switch (page_index->subpage) {
4296 			case PWR_SUBPAGE_CODE: {
4297 				struct copan_power_subpage *current_page,
4298 							   *saved_page;
4299 
4300 				memcpy(&lun->mode_pages.power_subpage[
4301 				       CTL_PAGE_CURRENT],
4302 				       &power_page_default,
4303 				       sizeof(power_page_default));
4304 				memcpy(&lun->mode_pages.power_subpage[
4305 				       CTL_PAGE_CHANGEABLE],
4306 				       &power_page_changeable,
4307 				       sizeof(power_page_changeable));
4308 				memcpy(&lun->mode_pages.power_subpage[
4309 				       CTL_PAGE_DEFAULT],
4310 				       &power_page_default,
4311 				       sizeof(power_page_default));
4312 				memcpy(&lun->mode_pages.power_subpage[
4313 				       CTL_PAGE_SAVED],
4314 				       &power_page_default,
4315 				       sizeof(power_page_default));
4316 				page_index->page_data =
4317 				    (uint8_t *)lun->mode_pages.power_subpage;
4318 
4319 				current_page = (struct copan_power_subpage *)
4320 					(page_index->page_data +
4321 					 (page_index->page_len *
4322 					  CTL_PAGE_CURRENT));
4323 			        saved_page = (struct copan_power_subpage *)
4324 				        (page_index->page_data +
4325 					 (page_index->page_len *
4326 					  CTL_PAGE_SAVED));
4327 				break;
4328 			}
4329 			case APS_SUBPAGE_CODE: {
4330 				struct copan_aps_subpage *current_page,
4331 							 *saved_page;
4332 
4333 				// This gets set multiple times but
4334 				// it should always be the same. It's
4335 				// only done during init so who cares.
4336 				index_to_aps_page = i;
4337 
4338 				memcpy(&lun->mode_pages.aps_subpage[
4339 				       CTL_PAGE_CURRENT],
4340 				       &aps_page_default,
4341 				       sizeof(aps_page_default));
4342 				memcpy(&lun->mode_pages.aps_subpage[
4343 				       CTL_PAGE_CHANGEABLE],
4344 				       &aps_page_changeable,
4345 				       sizeof(aps_page_changeable));
4346 				memcpy(&lun->mode_pages.aps_subpage[
4347 				       CTL_PAGE_DEFAULT],
4348 				       &aps_page_default,
4349 				       sizeof(aps_page_default));
4350 				memcpy(&lun->mode_pages.aps_subpage[
4351 				       CTL_PAGE_SAVED],
4352 				       &aps_page_default,
4353 				       sizeof(aps_page_default));
4354 				page_index->page_data =
4355 					(uint8_t *)lun->mode_pages.aps_subpage;
4356 
4357 				current_page = (struct copan_aps_subpage *)
4358 					(page_index->page_data +
4359 					 (page_index->page_len *
4360 					  CTL_PAGE_CURRENT));
4361 				saved_page = (struct copan_aps_subpage *)
4362 					(page_index->page_data +
4363 					 (page_index->page_len *
4364 					  CTL_PAGE_SAVED));
4365 				break;
4366 			}
4367 			case DBGCNF_SUBPAGE_CODE: {
4368 				struct copan_debugconf_subpage *current_page,
4369 							       *saved_page;
4370 
4371 				memcpy(&lun->mode_pages.debugconf_subpage[
4372 				       CTL_PAGE_CURRENT],
4373 				       &debugconf_page_default,
4374 				       sizeof(debugconf_page_default));
4375 				memcpy(&lun->mode_pages.debugconf_subpage[
4376 				       CTL_PAGE_CHANGEABLE],
4377 				       &debugconf_page_changeable,
4378 				       sizeof(debugconf_page_changeable));
4379 				memcpy(&lun->mode_pages.debugconf_subpage[
4380 				       CTL_PAGE_DEFAULT],
4381 				       &debugconf_page_default,
4382 				       sizeof(debugconf_page_default));
4383 				memcpy(&lun->mode_pages.debugconf_subpage[
4384 				       CTL_PAGE_SAVED],
4385 				       &debugconf_page_default,
4386 				       sizeof(debugconf_page_default));
4387 				page_index->page_data =
4388 					(uint8_t *)lun->mode_pages.debugconf_subpage;
4389 
4390 				current_page = (struct copan_debugconf_subpage *)
4391 					(page_index->page_data +
4392 					 (page_index->page_len *
4393 					  CTL_PAGE_CURRENT));
4394 				saved_page = (struct copan_debugconf_subpage *)
4395 					(page_index->page_data +
4396 					 (page_index->page_len *
4397 					  CTL_PAGE_SAVED));
4398 				break;
4399 			}
4400 			default:
4401 				panic("invalid subpage value %d",
4402 				      page_index->subpage);
4403 				break;
4404 			}
4405    			break;
4406 		}
4407 		default:
4408 			panic("invalid page value %d",
4409 			      page_index->page_code & SMPH_PC_MASK);
4410 			break;
4411     	}
4412 	}
4413 
4414 	return (CTL_RETVAL_COMPLETE);
4415 }
4416 
4417 /*
4418  * LUN allocation.
4419  *
4420  * Requirements:
4421  * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4422  *   wants us to allocate the LUN and he can block.
4423  * - ctl_softc is always set
4424  * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4425  *
4426  * Returns 0 for success, non-zero (errno) for failure.
4427  */
4428 static int
4429 ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4430 	      struct ctl_be_lun *const be_lun, struct ctl_id target_id)
4431 {
4432 	struct ctl_lun *nlun, *lun;
4433 	struct ctl_port *port;
4434 	struct scsi_vpd_id_descriptor *desc;
4435 	struct scsi_vpd_id_t10 *t10id;
4436 	const char *eui, *naa, *scsiname, *vendor;
4437 	int lun_number, i, lun_malloced;
4438 	int devidlen, idlen1, idlen2 = 0, len;
4439 
4440 	if (be_lun == NULL)
4441 		return (EINVAL);
4442 
4443 	/*
4444 	 * We currently only support Direct Access or Processor LUN types.
4445 	 */
4446 	switch (be_lun->lun_type) {
4447 	case T_DIRECT:
4448 		break;
4449 	case T_PROCESSOR:
4450 		break;
4451 	case T_SEQUENTIAL:
4452 	case T_CHANGER:
4453 	default:
4454 		be_lun->lun_config_status(be_lun->be_lun,
4455 					  CTL_LUN_CONFIG_FAILURE);
4456 		break;
4457 	}
4458 	if (ctl_lun == NULL) {
4459 		lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4460 		lun_malloced = 1;
4461 	} else {
4462 		lun_malloced = 0;
4463 		lun = ctl_lun;
4464 	}
4465 
4466 	memset(lun, 0, sizeof(*lun));
4467 	if (lun_malloced)
4468 		lun->flags = CTL_LUN_MALLOCED;
4469 
4470 	/* Generate LUN ID. */
4471 	devidlen = max(CTL_DEVID_MIN_LEN,
4472 	    strnlen(be_lun->device_id, CTL_DEVID_LEN));
4473 	idlen1 = sizeof(*t10id) + devidlen;
4474 	len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4475 	scsiname = ctl_get_opt(&be_lun->options, "scsiname");
4476 	if (scsiname != NULL) {
4477 		idlen2 = roundup2(strlen(scsiname) + 1, 4);
4478 		len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4479 	}
4480 	eui = ctl_get_opt(&be_lun->options, "eui");
4481 	if (eui != NULL) {
4482 		len += sizeof(struct scsi_vpd_id_descriptor) + 8;
4483 	}
4484 	naa = ctl_get_opt(&be_lun->options, "naa");
4485 	if (naa != NULL) {
4486 		len += sizeof(struct scsi_vpd_id_descriptor) + 8;
4487 	}
4488 	lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4489 	    M_CTL, M_WAITOK | M_ZERO);
4490 	lun->lun_devid->len = len;
4491 	desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4492 	desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4493 	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4494 	desc->length = idlen1;
4495 	t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4496 	memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4497 	if ((vendor = ctl_get_opt(&be_lun->options, "vendor")) == NULL) {
4498 		strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4499 	} else {
4500 		strncpy(t10id->vendor, vendor,
4501 		    min(sizeof(t10id->vendor), strlen(vendor)));
4502 	}
4503 	strncpy((char *)t10id->vendor_spec_id,
4504 	    (char *)be_lun->device_id, devidlen);
4505 	if (scsiname != NULL) {
4506 		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4507 		    desc->length);
4508 		desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4509 		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4510 		    SVPD_ID_TYPE_SCSI_NAME;
4511 		desc->length = idlen2;
4512 		strlcpy(desc->identifier, scsiname, idlen2);
4513 	}
4514 	if (eui != NULL) {
4515 		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4516 		    desc->length);
4517 		desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4518 		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4519 		    SVPD_ID_TYPE_EUI64;
4520 		desc->length = 8;
4521 		scsi_u64to8b(strtouq(eui, NULL, 0), desc->identifier);
4522 	}
4523 	if (naa != NULL) {
4524 		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4525 		    desc->length);
4526 		desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4527 		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4528 		    SVPD_ID_TYPE_NAA;
4529 		desc->length = 8;
4530 		scsi_u64to8b(strtouq(naa, NULL, 0), desc->identifier);
4531 	}
4532 
4533 	mtx_lock(&ctl_softc->ctl_lock);
4534 	/*
4535 	 * See if the caller requested a particular LUN number.  If so, see
4536 	 * if it is available.  Otherwise, allocate the first available LUN.
4537 	 */
4538 	if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4539 		if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4540 		 || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4541 			mtx_unlock(&ctl_softc->ctl_lock);
4542 			if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4543 				printf("ctl: requested LUN ID %d is higher "
4544 				       "than CTL_MAX_LUNS - 1 (%d)\n",
4545 				       be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4546 			} else {
4547 				/*
4548 				 * XXX KDM return an error, or just assign
4549 				 * another LUN ID in this case??
4550 				 */
4551 				printf("ctl: requested LUN ID %d is already "
4552 				       "in use\n", be_lun->req_lun_id);
4553 			}
4554 			if (lun->flags & CTL_LUN_MALLOCED)
4555 				free(lun, M_CTL);
4556 			be_lun->lun_config_status(be_lun->be_lun,
4557 						  CTL_LUN_CONFIG_FAILURE);
4558 			return (ENOSPC);
4559 		}
4560 		lun_number = be_lun->req_lun_id;
4561 	} else {
4562 		lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, CTL_MAX_LUNS);
4563 		if (lun_number == -1) {
4564 			mtx_unlock(&ctl_softc->ctl_lock);
4565 			printf("ctl: can't allocate LUN on target %ju, out of "
4566 			       "LUNs\n", (uintmax_t)target_id.id);
4567 			if (lun->flags & CTL_LUN_MALLOCED)
4568 				free(lun, M_CTL);
4569 			be_lun->lun_config_status(be_lun->be_lun,
4570 						  CTL_LUN_CONFIG_FAILURE);
4571 			return (ENOSPC);
4572 		}
4573 	}
4574 	ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4575 
4576 	mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4577 	lun->target = target_id;
4578 	lun->lun = lun_number;
4579 	lun->be_lun = be_lun;
4580 	/*
4581 	 * The processor LUN is always enabled.  Disk LUNs come on line
4582 	 * disabled, and must be enabled by the backend.
4583 	 */
4584 	lun->flags |= CTL_LUN_DISABLED;
4585 	lun->backend = be_lun->be;
4586 	be_lun->ctl_lun = lun;
4587 	be_lun->lun_id = lun_number;
4588 	atomic_add_int(&be_lun->be->num_luns, 1);
4589 	if (be_lun->flags & CTL_LUN_FLAG_POWERED_OFF)
4590 		lun->flags |= CTL_LUN_STOPPED;
4591 
4592 	if (be_lun->flags & CTL_LUN_FLAG_INOPERABLE)
4593 		lun->flags |= CTL_LUN_INOPERABLE;
4594 
4595 	if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4596 		lun->flags |= CTL_LUN_PRIMARY_SC;
4597 
4598 	lun->ctl_softc = ctl_softc;
4599 	TAILQ_INIT(&lun->ooa_queue);
4600 	TAILQ_INIT(&lun->blocked_queue);
4601 	STAILQ_INIT(&lun->error_list);
4602 	ctl_tpc_lun_init(lun);
4603 
4604 	/*
4605 	 * Initialize the mode page index.
4606 	 */
4607 	ctl_init_page_index(lun);
4608 
4609 	/*
4610 	 * Set the poweron UA for all initiators on this LUN only.
4611 	 */
4612 	for (i = 0; i < CTL_MAX_INITIATORS; i++)
4613 		lun->pending_ua[i] = CTL_UA_POWERON;
4614 
4615 	/*
4616 	 * Now, before we insert this lun on the lun list, set the lun
4617 	 * inventory changed UA for all other luns.
4618 	 */
4619 	STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4620 		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4621 			nlun->pending_ua[i] |= CTL_UA_LUN_CHANGE;
4622 		}
4623 	}
4624 
4625 	STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4626 
4627 	ctl_softc->ctl_luns[lun_number] = lun;
4628 
4629 	ctl_softc->num_luns++;
4630 
4631 	/* Setup statistics gathering */
4632 	lun->stats.device_type = be_lun->lun_type;
4633 	lun->stats.lun_number = lun_number;
4634 	if (lun->stats.device_type == T_DIRECT)
4635 		lun->stats.blocksize = be_lun->blocksize;
4636 	else
4637 		lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4638 	for (i = 0;i < CTL_MAX_PORTS;i++)
4639 		lun->stats.ports[i].targ_port = i;
4640 
4641 	mtx_unlock(&ctl_softc->ctl_lock);
4642 
4643 	lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4644 
4645 	/*
4646 	 * Run through each registered FETD and bring it online if it isn't
4647 	 * already.  Enable the target ID if it hasn't been enabled, and
4648 	 * enable this particular LUN.
4649 	 */
4650 	STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4651 		int retval;
4652 
4653 		retval = port->lun_enable(port->targ_lun_arg, target_id,lun_number);
4654 		if (retval != 0) {
4655 			printf("ctl_alloc_lun: FETD %s port %d returned error "
4656 			       "%d for lun_enable on target %ju lun %d\n",
4657 			       port->port_name, port->targ_port, retval,
4658 			       (uintmax_t)target_id.id, lun_number);
4659 		} else
4660 			port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4661 	}
4662 	return (0);
4663 }
4664 
4665 /*
4666  * Delete a LUN.
4667  * Assumptions:
4668  * - LUN has already been marked invalid and any pending I/O has been taken
4669  *   care of.
4670  */
4671 static int
4672 ctl_free_lun(struct ctl_lun *lun)
4673 {
4674 	struct ctl_softc *softc;
4675 #if 0
4676 	struct ctl_port *port;
4677 #endif
4678 	struct ctl_lun *nlun;
4679 	int i;
4680 
4681 	softc = lun->ctl_softc;
4682 
4683 	mtx_assert(&softc->ctl_lock, MA_OWNED);
4684 
4685 	STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4686 
4687 	ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4688 
4689 	softc->ctl_luns[lun->lun] = NULL;
4690 
4691 	if (!TAILQ_EMPTY(&lun->ooa_queue))
4692 		panic("Freeing a LUN %p with outstanding I/O!!\n", lun);
4693 
4694 	softc->num_luns--;
4695 
4696 	/*
4697 	 * XXX KDM this scheme only works for a single target/multiple LUN
4698 	 * setup.  It needs to be revamped for a multiple target scheme.
4699 	 *
4700 	 * XXX KDM this results in port->lun_disable() getting called twice,
4701 	 * once when ctl_disable_lun() is called, and a second time here.
4702 	 * We really need to re-think the LUN disable semantics.  There
4703 	 * should probably be several steps/levels to LUN removal:
4704 	 *  - disable
4705 	 *  - invalidate
4706 	 *  - free
4707  	 *
4708 	 * Right now we only have a disable method when communicating to
4709 	 * the front end ports, at least for individual LUNs.
4710 	 */
4711 #if 0
4712 	STAILQ_FOREACH(port, &softc->port_list, links) {
4713 		int retval;
4714 
4715 		retval = port->lun_disable(port->targ_lun_arg, lun->target,
4716 					 lun->lun);
4717 		if (retval != 0) {
4718 			printf("ctl_free_lun: FETD %s port %d returned error "
4719 			       "%d for lun_disable on target %ju lun %jd\n",
4720 			       port->port_name, port->targ_port, retval,
4721 			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4722 		}
4723 
4724 		if (STAILQ_FIRST(&softc->lun_list) == NULL) {
4725 			port->status &= ~CTL_PORT_STATUS_LUN_ONLINE;
4726 
4727 			retval = port->targ_disable(port->targ_lun_arg,lun->target);
4728 			if (retval != 0) {
4729 				printf("ctl_free_lun: FETD %s port %d "
4730 				       "returned error %d for targ_disable on "
4731 				       "target %ju\n", port->port_name,
4732 				       port->targ_port, retval,
4733 				       (uintmax_t)lun->target.id);
4734 			} else
4735 				port->status &= ~CTL_PORT_STATUS_TARG_ONLINE;
4736 
4737 			if ((port->status & CTL_PORT_STATUS_TARG_ONLINE) != 0)
4738 				continue;
4739 
4740 #if 0
4741 			port->port_offline(port->onoff_arg);
4742 			port->status &= ~CTL_PORT_STATUS_ONLINE;
4743 #endif
4744 		}
4745 	}
4746 #endif
4747 
4748 	/*
4749 	 * Tell the backend to free resources, if this LUN has a backend.
4750 	 */
4751 	atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4752 	lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4753 
4754 	ctl_tpc_lun_shutdown(lun);
4755 	mtx_destroy(&lun->lun_lock);
4756 	free(lun->lun_devid, M_CTL);
4757 	if (lun->flags & CTL_LUN_MALLOCED)
4758 		free(lun, M_CTL);
4759 
4760 	STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4761 		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4762 			nlun->pending_ua[i] |= CTL_UA_LUN_CHANGE;
4763 		}
4764 	}
4765 
4766 	return (0);
4767 }
4768 
4769 static void
4770 ctl_create_lun(struct ctl_be_lun *be_lun)
4771 {
4772 	struct ctl_softc *ctl_softc;
4773 
4774 	ctl_softc = control_softc;
4775 
4776 	/*
4777 	 * ctl_alloc_lun() should handle all potential failure cases.
4778 	 */
4779 	ctl_alloc_lun(ctl_softc, NULL, be_lun, ctl_softc->target);
4780 }
4781 
4782 int
4783 ctl_add_lun(struct ctl_be_lun *be_lun)
4784 {
4785 	struct ctl_softc *ctl_softc = control_softc;
4786 
4787 	mtx_lock(&ctl_softc->ctl_lock);
4788 	STAILQ_INSERT_TAIL(&ctl_softc->pending_lun_queue, be_lun, links);
4789 	mtx_unlock(&ctl_softc->ctl_lock);
4790 	wakeup(&ctl_softc->pending_lun_queue);
4791 
4792 	return (0);
4793 }
4794 
4795 int
4796 ctl_enable_lun(struct ctl_be_lun *be_lun)
4797 {
4798 	struct ctl_softc *ctl_softc;
4799 	struct ctl_port *port, *nport;
4800 	struct ctl_lun *lun;
4801 	int retval;
4802 
4803 	ctl_softc = control_softc;
4804 
4805 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4806 
4807 	mtx_lock(&ctl_softc->ctl_lock);
4808 	mtx_lock(&lun->lun_lock);
4809 	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4810 		/*
4811 		 * eh?  Why did we get called if the LUN is already
4812 		 * enabled?
4813 		 */
4814 		mtx_unlock(&lun->lun_lock);
4815 		mtx_unlock(&ctl_softc->ctl_lock);
4816 		return (0);
4817 	}
4818 	lun->flags &= ~CTL_LUN_DISABLED;
4819 	mtx_unlock(&lun->lun_lock);
4820 
4821 	for (port = STAILQ_FIRST(&ctl_softc->port_list); port != NULL; port = nport) {
4822 		nport = STAILQ_NEXT(port, links);
4823 
4824 		/*
4825 		 * Drop the lock while we call the FETD's enable routine.
4826 		 * This can lead to a callback into CTL (at least in the
4827 		 * case of the internal initiator frontend.
4828 		 */
4829 		mtx_unlock(&ctl_softc->ctl_lock);
4830 		retval = port->lun_enable(port->targ_lun_arg, lun->target,lun->lun);
4831 		mtx_lock(&ctl_softc->ctl_lock);
4832 		if (retval != 0) {
4833 			printf("%s: FETD %s port %d returned error "
4834 			       "%d for lun_enable on target %ju lun %jd\n",
4835 			       __func__, port->port_name, port->targ_port, retval,
4836 			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4837 		}
4838 #if 0
4839 		 else {
4840             /* NOTE:  TODO:  why does lun enable affect port status? */
4841 			port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4842 		}
4843 #endif
4844 	}
4845 
4846 	mtx_unlock(&ctl_softc->ctl_lock);
4847 
4848 	return (0);
4849 }
4850 
4851 int
4852 ctl_disable_lun(struct ctl_be_lun *be_lun)
4853 {
4854 	struct ctl_softc *ctl_softc;
4855 	struct ctl_port *port;
4856 	struct ctl_lun *lun;
4857 	int retval;
4858 
4859 	ctl_softc = control_softc;
4860 
4861 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4862 
4863 	mtx_lock(&ctl_softc->ctl_lock);
4864 	mtx_lock(&lun->lun_lock);
4865 	if (lun->flags & CTL_LUN_DISABLED) {
4866 		mtx_unlock(&lun->lun_lock);
4867 		mtx_unlock(&ctl_softc->ctl_lock);
4868 		return (0);
4869 	}
4870 	lun->flags |= CTL_LUN_DISABLED;
4871 	mtx_unlock(&lun->lun_lock);
4872 
4873 	STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4874 		mtx_unlock(&ctl_softc->ctl_lock);
4875 		/*
4876 		 * Drop the lock before we call the frontend's disable
4877 		 * routine, to avoid lock order reversals.
4878 		 *
4879 		 * XXX KDM what happens if the frontend list changes while
4880 		 * we're traversing it?  It's unlikely, but should be handled.
4881 		 */
4882 		retval = port->lun_disable(port->targ_lun_arg, lun->target,
4883 					 lun->lun);
4884 		mtx_lock(&ctl_softc->ctl_lock);
4885 		if (retval != 0) {
4886 			printf("ctl_alloc_lun: FETD %s port %d returned error "
4887 			       "%d for lun_disable on target %ju lun %jd\n",
4888 			       port->port_name, port->targ_port, retval,
4889 			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4890 		}
4891 	}
4892 
4893 	mtx_unlock(&ctl_softc->ctl_lock);
4894 
4895 	return (0);
4896 }
4897 
4898 int
4899 ctl_start_lun(struct ctl_be_lun *be_lun)
4900 {
4901 	struct ctl_softc *ctl_softc;
4902 	struct ctl_lun *lun;
4903 
4904 	ctl_softc = control_softc;
4905 
4906 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4907 
4908 	mtx_lock(&lun->lun_lock);
4909 	lun->flags &= ~CTL_LUN_STOPPED;
4910 	mtx_unlock(&lun->lun_lock);
4911 
4912 	return (0);
4913 }
4914 
4915 int
4916 ctl_stop_lun(struct ctl_be_lun *be_lun)
4917 {
4918 	struct ctl_softc *ctl_softc;
4919 	struct ctl_lun *lun;
4920 
4921 	ctl_softc = control_softc;
4922 
4923 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4924 
4925 	mtx_lock(&lun->lun_lock);
4926 	lun->flags |= CTL_LUN_STOPPED;
4927 	mtx_unlock(&lun->lun_lock);
4928 
4929 	return (0);
4930 }
4931 
4932 int
4933 ctl_lun_offline(struct ctl_be_lun *be_lun)
4934 {
4935 	struct ctl_softc *ctl_softc;
4936 	struct ctl_lun *lun;
4937 
4938 	ctl_softc = control_softc;
4939 
4940 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4941 
4942 	mtx_lock(&lun->lun_lock);
4943 	lun->flags |= CTL_LUN_OFFLINE;
4944 	mtx_unlock(&lun->lun_lock);
4945 
4946 	return (0);
4947 }
4948 
4949 int
4950 ctl_lun_online(struct ctl_be_lun *be_lun)
4951 {
4952 	struct ctl_softc *ctl_softc;
4953 	struct ctl_lun *lun;
4954 
4955 	ctl_softc = control_softc;
4956 
4957 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4958 
4959 	mtx_lock(&lun->lun_lock);
4960 	lun->flags &= ~CTL_LUN_OFFLINE;
4961 	mtx_unlock(&lun->lun_lock);
4962 
4963 	return (0);
4964 }
4965 
4966 int
4967 ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4968 {
4969 	struct ctl_softc *ctl_softc;
4970 	struct ctl_lun *lun;
4971 
4972 	ctl_softc = control_softc;
4973 
4974 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4975 
4976 	mtx_lock(&lun->lun_lock);
4977 
4978 	/*
4979 	 * The LUN needs to be disabled before it can be marked invalid.
4980 	 */
4981 	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4982 		mtx_unlock(&lun->lun_lock);
4983 		return (-1);
4984 	}
4985 	/*
4986 	 * Mark the LUN invalid.
4987 	 */
4988 	lun->flags |= CTL_LUN_INVALID;
4989 
4990 	/*
4991 	 * If there is nothing in the OOA queue, go ahead and free the LUN.
4992 	 * If we have something in the OOA queue, we'll free it when the
4993 	 * last I/O completes.
4994 	 */
4995 	if (TAILQ_EMPTY(&lun->ooa_queue)) {
4996 		mtx_unlock(&lun->lun_lock);
4997 		mtx_lock(&ctl_softc->ctl_lock);
4998 		ctl_free_lun(lun);
4999 		mtx_unlock(&ctl_softc->ctl_lock);
5000 	} else
5001 		mtx_unlock(&lun->lun_lock);
5002 
5003 	return (0);
5004 }
5005 
5006 int
5007 ctl_lun_inoperable(struct ctl_be_lun *be_lun)
5008 {
5009 	struct ctl_softc *ctl_softc;
5010 	struct ctl_lun *lun;
5011 
5012 	ctl_softc = control_softc;
5013 	lun = (struct ctl_lun *)be_lun->ctl_lun;
5014 
5015 	mtx_lock(&lun->lun_lock);
5016 	lun->flags |= CTL_LUN_INOPERABLE;
5017 	mtx_unlock(&lun->lun_lock);
5018 
5019 	return (0);
5020 }
5021 
5022 int
5023 ctl_lun_operable(struct ctl_be_lun *be_lun)
5024 {
5025 	struct ctl_softc *ctl_softc;
5026 	struct ctl_lun *lun;
5027 
5028 	ctl_softc = control_softc;
5029 	lun = (struct ctl_lun *)be_lun->ctl_lun;
5030 
5031 	mtx_lock(&lun->lun_lock);
5032 	lun->flags &= ~CTL_LUN_INOPERABLE;
5033 	mtx_unlock(&lun->lun_lock);
5034 
5035 	return (0);
5036 }
5037 
5038 int
5039 ctl_lun_power_lock(struct ctl_be_lun *be_lun, struct ctl_nexus *nexus,
5040 		   int lock)
5041 {
5042 	struct ctl_softc *softc;
5043 	struct ctl_lun *lun;
5044 	struct copan_aps_subpage *current_sp;
5045 	struct ctl_page_index *page_index;
5046 	int i;
5047 
5048 	softc = control_softc;
5049 
5050 	mtx_lock(&softc->ctl_lock);
5051 
5052 	lun = (struct ctl_lun *)be_lun->ctl_lun;
5053 	mtx_lock(&lun->lun_lock);
5054 
5055 	page_index = NULL;
5056 	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
5057 		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
5058 		     APS_PAGE_CODE)
5059 			continue;
5060 
5061 		if (lun->mode_pages.index[i].subpage != APS_SUBPAGE_CODE)
5062 			continue;
5063 		page_index = &lun->mode_pages.index[i];
5064 	}
5065 
5066 	if (page_index == NULL) {
5067 		mtx_unlock(&lun->lun_lock);
5068 		mtx_unlock(&softc->ctl_lock);
5069 		printf("%s: APS subpage not found for lun %ju!\n", __func__,
5070 		       (uintmax_t)lun->lun);
5071 		return (1);
5072 	}
5073 #if 0
5074 	if ((softc->aps_locked_lun != 0)
5075 	 && (softc->aps_locked_lun != lun->lun)) {
5076 		printf("%s: attempt to lock LUN %llu when %llu is already "
5077 		       "locked\n");
5078 		mtx_unlock(&lun->lun_lock);
5079 		mtx_unlock(&softc->ctl_lock);
5080 		return (1);
5081 	}
5082 #endif
5083 
5084 	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
5085 		(page_index->page_len * CTL_PAGE_CURRENT));
5086 
5087 	if (lock != 0) {
5088 		current_sp->lock_active = APS_LOCK_ACTIVE;
5089 		softc->aps_locked_lun = lun->lun;
5090 	} else {
5091 		current_sp->lock_active = 0;
5092 		softc->aps_locked_lun = 0;
5093 	}
5094 
5095 
5096 	/*
5097 	 * If we're in HA mode, try to send the lock message to the other
5098 	 * side.
5099 	 */
5100 	if (ctl_is_single == 0) {
5101 		int isc_retval;
5102 		union ctl_ha_msg lock_msg;
5103 
5104 		lock_msg.hdr.nexus = *nexus;
5105 		lock_msg.hdr.msg_type = CTL_MSG_APS_LOCK;
5106 		if (lock != 0)
5107 			lock_msg.aps.lock_flag = 1;
5108 		else
5109 			lock_msg.aps.lock_flag = 0;
5110 		isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &lock_msg,
5111 					 sizeof(lock_msg), 0);
5112 		if (isc_retval > CTL_HA_STATUS_SUCCESS) {
5113 			printf("%s: APS (lock=%d) error returned from "
5114 			       "ctl_ha_msg_send: %d\n", __func__, lock, isc_retval);
5115 			mtx_unlock(&lun->lun_lock);
5116 			mtx_unlock(&softc->ctl_lock);
5117 			return (1);
5118 		}
5119 	}
5120 
5121 	mtx_unlock(&lun->lun_lock);
5122 	mtx_unlock(&softc->ctl_lock);
5123 
5124 	return (0);
5125 }
5126 
5127 void
5128 ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
5129 {
5130 	struct ctl_lun *lun;
5131 	struct ctl_softc *softc;
5132 	int i;
5133 
5134 	softc = control_softc;
5135 
5136 	lun = (struct ctl_lun *)be_lun->ctl_lun;
5137 
5138 	mtx_lock(&lun->lun_lock);
5139 
5140 	for (i = 0; i < CTL_MAX_INITIATORS; i++)
5141 		lun->pending_ua[i] |= CTL_UA_CAPACITY_CHANGED;
5142 
5143 	mtx_unlock(&lun->lun_lock);
5144 }
5145 
5146 /*
5147  * Backend "memory move is complete" callback for requests that never
5148  * make it down to say RAIDCore's configuration code.
5149  */
5150 int
5151 ctl_config_move_done(union ctl_io *io)
5152 {
5153 	int retval;
5154 
5155 	retval = CTL_RETVAL_COMPLETE;
5156 
5157 
5158 	CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
5159 	/*
5160 	 * XXX KDM this shouldn't happen, but what if it does?
5161 	 */
5162 	if (io->io_hdr.io_type != CTL_IO_SCSI)
5163 		panic("I/O type isn't CTL_IO_SCSI!");
5164 
5165 	if ((io->io_hdr.port_status == 0)
5166 	 && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
5167 	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE))
5168 		io->io_hdr.status = CTL_SUCCESS;
5169 	else if ((io->io_hdr.port_status != 0)
5170 	      && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
5171 	      && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)){
5172 		/*
5173 		 * For hardware error sense keys, the sense key
5174 		 * specific value is defined to be a retry count,
5175 		 * but we use it to pass back an internal FETD
5176 		 * error code.  XXX KDM  Hopefully the FETD is only
5177 		 * using 16 bits for an error code, since that's
5178 		 * all the space we have in the sks field.
5179 		 */
5180 		ctl_set_internal_failure(&io->scsiio,
5181 					 /*sks_valid*/ 1,
5182 					 /*retry_count*/
5183 					 io->io_hdr.port_status);
5184 		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5185 			free(io->scsiio.kern_data_ptr, M_CTL);
5186 		ctl_done(io);
5187 		goto bailout;
5188 	}
5189 
5190 	if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
5191 	 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
5192 	 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
5193 		/*
5194 		 * XXX KDM just assuming a single pointer here, and not a
5195 		 * S/G list.  If we start using S/G lists for config data,
5196 		 * we'll need to know how to clean them up here as well.
5197 		 */
5198 		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5199 			free(io->scsiio.kern_data_ptr, M_CTL);
5200 		/* Hopefully the user has already set the status... */
5201 		ctl_done(io);
5202 	} else {
5203 		/*
5204 		 * XXX KDM now we need to continue data movement.  Some
5205 		 * options:
5206 		 * - call ctl_scsiio() again?  We don't do this for data
5207 		 *   writes, because for those at least we know ahead of
5208 		 *   time where the write will go and how long it is.  For
5209 		 *   config writes, though, that information is largely
5210 		 *   contained within the write itself, thus we need to
5211 		 *   parse out the data again.
5212 		 *
5213 		 * - Call some other function once the data is in?
5214 		 */
5215 
5216 		/*
5217 		 * XXX KDM call ctl_scsiio() again for now, and check flag
5218 		 * bits to see whether we're allocated or not.
5219 		 */
5220 		retval = ctl_scsiio(&io->scsiio);
5221 	}
5222 bailout:
5223 	return (retval);
5224 }
5225 
5226 /*
5227  * This gets called by a backend driver when it is done with a
5228  * data_submit method.
5229  */
5230 void
5231 ctl_data_submit_done(union ctl_io *io)
5232 {
5233 	/*
5234 	 * If the IO_CONT flag is set, we need to call the supplied
5235 	 * function to continue processing the I/O, instead of completing
5236 	 * the I/O just yet.
5237 	 *
5238 	 * If there is an error, though, we don't want to keep processing.
5239 	 * Instead, just send status back to the initiator.
5240 	 */
5241 	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5242 	    (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5243 	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5244 	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5245 		io->scsiio.io_cont(io);
5246 		return;
5247 	}
5248 	ctl_done(io);
5249 }
5250 
5251 /*
5252  * This gets called by a backend driver when it is done with a
5253  * configuration write.
5254  */
5255 void
5256 ctl_config_write_done(union ctl_io *io)
5257 {
5258 	/*
5259 	 * If the IO_CONT flag is set, we need to call the supplied
5260 	 * function to continue processing the I/O, instead of completing
5261 	 * the I/O just yet.
5262 	 *
5263 	 * If there is an error, though, we don't want to keep processing.
5264 	 * Instead, just send status back to the initiator.
5265 	 */
5266 	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT)
5267 	 && (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)
5268 	  || ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))) {
5269 		io->scsiio.io_cont(io);
5270 		return;
5271 	}
5272 	/*
5273 	 * Since a configuration write can be done for commands that actually
5274 	 * have data allocated, like write buffer, and commands that have
5275 	 * no data, like start/stop unit, we need to check here.
5276 	 */
5277 	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT)
5278 		free(io->scsiio.kern_data_ptr, M_CTL);
5279 	ctl_done(io);
5280 }
5281 
5282 /*
5283  * SCSI release command.
5284  */
5285 int
5286 ctl_scsi_release(struct ctl_scsiio *ctsio)
5287 {
5288 	int length, longid, thirdparty_id, resv_id;
5289 	struct ctl_softc *ctl_softc;
5290 	struct ctl_lun *lun;
5291 
5292 	length = 0;
5293 	resv_id = 0;
5294 
5295 	CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5296 
5297 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5298 	ctl_softc = control_softc;
5299 
5300 	switch (ctsio->cdb[0]) {
5301 	case RELEASE_10: {
5302 		struct scsi_release_10 *cdb;
5303 
5304 		cdb = (struct scsi_release_10 *)ctsio->cdb;
5305 
5306 		if (cdb->byte2 & SR10_LONGID)
5307 			longid = 1;
5308 		else
5309 			thirdparty_id = cdb->thirdparty_id;
5310 
5311 		resv_id = cdb->resv_id;
5312 		length = scsi_2btoul(cdb->length);
5313 		break;
5314 	}
5315 	}
5316 
5317 
5318 	/*
5319 	 * XXX KDM right now, we only support LUN reservation.  We don't
5320 	 * support 3rd party reservations, or extent reservations, which
5321 	 * might actually need the parameter list.  If we've gotten this
5322 	 * far, we've got a LUN reservation.  Anything else got kicked out
5323 	 * above.  So, according to SPC, ignore the length.
5324 	 */
5325 	length = 0;
5326 
5327 	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5328 	 && (length > 0)) {
5329 		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5330 		ctsio->kern_data_len = length;
5331 		ctsio->kern_total_len = length;
5332 		ctsio->kern_data_resid = 0;
5333 		ctsio->kern_rel_offset = 0;
5334 		ctsio->kern_sg_entries = 0;
5335 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5336 		ctsio->be_move_done = ctl_config_move_done;
5337 		ctl_datamove((union ctl_io *)ctsio);
5338 
5339 		return (CTL_RETVAL_COMPLETE);
5340 	}
5341 
5342 	if (length > 0)
5343 		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5344 
5345 	mtx_lock(&lun->lun_lock);
5346 
5347 	/*
5348 	 * According to SPC, it is not an error for an intiator to attempt
5349 	 * to release a reservation on a LUN that isn't reserved, or that
5350 	 * is reserved by another initiator.  The reservation can only be
5351 	 * released, though, by the initiator who made it or by one of
5352 	 * several reset type events.
5353 	 */
5354 	if (lun->flags & CTL_LUN_RESERVED) {
5355 		if ((ctsio->io_hdr.nexus.initid.id == lun->rsv_nexus.initid.id)
5356 		 && (ctsio->io_hdr.nexus.targ_port == lun->rsv_nexus.targ_port)
5357 		 && (ctsio->io_hdr.nexus.targ_target.id ==
5358 		     lun->rsv_nexus.targ_target.id)) {
5359 			lun->flags &= ~CTL_LUN_RESERVED;
5360 		}
5361 	}
5362 
5363 	mtx_unlock(&lun->lun_lock);
5364 
5365 	ctsio->scsi_status = SCSI_STATUS_OK;
5366 	ctsio->io_hdr.status = CTL_SUCCESS;
5367 
5368 	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5369 		free(ctsio->kern_data_ptr, M_CTL);
5370 		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5371 	}
5372 
5373 	ctl_done((union ctl_io *)ctsio);
5374 	return (CTL_RETVAL_COMPLETE);
5375 }
5376 
5377 int
5378 ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5379 {
5380 	int extent, thirdparty, longid;
5381 	int resv_id, length;
5382 	uint64_t thirdparty_id;
5383 	struct ctl_softc *ctl_softc;
5384 	struct ctl_lun *lun;
5385 
5386 	extent = 0;
5387 	thirdparty = 0;
5388 	longid = 0;
5389 	resv_id = 0;
5390 	length = 0;
5391 	thirdparty_id = 0;
5392 
5393 	CTL_DEBUG_PRINT(("ctl_reserve\n"));
5394 
5395 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5396 	ctl_softc = control_softc;
5397 
5398 	switch (ctsio->cdb[0]) {
5399 	case RESERVE_10: {
5400 		struct scsi_reserve_10 *cdb;
5401 
5402 		cdb = (struct scsi_reserve_10 *)ctsio->cdb;
5403 
5404 		if (cdb->byte2 & SR10_LONGID)
5405 			longid = 1;
5406 		else
5407 			thirdparty_id = cdb->thirdparty_id;
5408 
5409 		resv_id = cdb->resv_id;
5410 		length = scsi_2btoul(cdb->length);
5411 		break;
5412 	}
5413 	}
5414 
5415 	/*
5416 	 * XXX KDM right now, we only support LUN reservation.  We don't
5417 	 * support 3rd party reservations, or extent reservations, which
5418 	 * might actually need the parameter list.  If we've gotten this
5419 	 * far, we've got a LUN reservation.  Anything else got kicked out
5420 	 * above.  So, according to SPC, ignore the length.
5421 	 */
5422 	length = 0;
5423 
5424 	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5425 	 && (length > 0)) {
5426 		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5427 		ctsio->kern_data_len = length;
5428 		ctsio->kern_total_len = length;
5429 		ctsio->kern_data_resid = 0;
5430 		ctsio->kern_rel_offset = 0;
5431 		ctsio->kern_sg_entries = 0;
5432 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5433 		ctsio->be_move_done = ctl_config_move_done;
5434 		ctl_datamove((union ctl_io *)ctsio);
5435 
5436 		return (CTL_RETVAL_COMPLETE);
5437 	}
5438 
5439 	if (length > 0)
5440 		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5441 
5442 	mtx_lock(&lun->lun_lock);
5443 	if (lun->flags & CTL_LUN_RESERVED) {
5444 		if ((ctsio->io_hdr.nexus.initid.id != lun->rsv_nexus.initid.id)
5445 		 || (ctsio->io_hdr.nexus.targ_port != lun->rsv_nexus.targ_port)
5446 		 || (ctsio->io_hdr.nexus.targ_target.id !=
5447 		     lun->rsv_nexus.targ_target.id)) {
5448 			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
5449 			ctsio->io_hdr.status = CTL_SCSI_ERROR;
5450 			goto bailout;
5451 		}
5452 	}
5453 
5454 	lun->flags |= CTL_LUN_RESERVED;
5455 	lun->rsv_nexus = ctsio->io_hdr.nexus;
5456 
5457 	ctsio->scsi_status = SCSI_STATUS_OK;
5458 	ctsio->io_hdr.status = CTL_SUCCESS;
5459 
5460 bailout:
5461 	mtx_unlock(&lun->lun_lock);
5462 
5463 	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5464 		free(ctsio->kern_data_ptr, M_CTL);
5465 		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5466 	}
5467 
5468 	ctl_done((union ctl_io *)ctsio);
5469 	return (CTL_RETVAL_COMPLETE);
5470 }
5471 
5472 int
5473 ctl_start_stop(struct ctl_scsiio *ctsio)
5474 {
5475 	struct scsi_start_stop_unit *cdb;
5476 	struct ctl_lun *lun;
5477 	struct ctl_softc *ctl_softc;
5478 	int retval;
5479 
5480 	CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5481 
5482 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5483 	ctl_softc = control_softc;
5484 	retval = 0;
5485 
5486 	cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5487 
5488 	/*
5489 	 * XXX KDM
5490 	 * We don't support the immediate bit on a stop unit.  In order to
5491 	 * do that, we would need to code up a way to know that a stop is
5492 	 * pending, and hold off any new commands until it completes, one
5493 	 * way or another.  Then we could accept or reject those commands
5494 	 * depending on its status.  We would almost need to do the reverse
5495 	 * of what we do below for an immediate start -- return the copy of
5496 	 * the ctl_io to the FETD with status to send to the host (and to
5497 	 * free the copy!) and then free the original I/O once the stop
5498 	 * actually completes.  That way, the OOA queue mechanism can work
5499 	 * to block commands that shouldn't proceed.  Another alternative
5500 	 * would be to put the copy in the queue in place of the original,
5501 	 * and return the original back to the caller.  That could be
5502 	 * slightly safer..
5503 	 */
5504 	if ((cdb->byte2 & SSS_IMMED)
5505 	 && ((cdb->how & SSS_START) == 0)) {
5506 		ctl_set_invalid_field(ctsio,
5507 				      /*sks_valid*/ 1,
5508 				      /*command*/ 1,
5509 				      /*field*/ 1,
5510 				      /*bit_valid*/ 1,
5511 				      /*bit*/ 0);
5512 		ctl_done((union ctl_io *)ctsio);
5513 		return (CTL_RETVAL_COMPLETE);
5514 	}
5515 
5516 	if ((lun->flags & CTL_LUN_PR_RESERVED)
5517 	 && ((cdb->how & SSS_START)==0)) {
5518 		uint32_t residx;
5519 
5520 		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5521 		if (!lun->per_res[residx].registered
5522 		 || (lun->pr_res_idx!=residx && lun->res_type < 4)) {
5523 
5524 			ctl_set_reservation_conflict(ctsio);
5525 			ctl_done((union ctl_io *)ctsio);
5526 			return (CTL_RETVAL_COMPLETE);
5527 		}
5528 	}
5529 
5530 	/*
5531 	 * If there is no backend on this device, we can't start or stop
5532 	 * it.  In theory we shouldn't get any start/stop commands in the
5533 	 * first place at this level if the LUN doesn't have a backend.
5534 	 * That should get stopped by the command decode code.
5535 	 */
5536 	if (lun->backend == NULL) {
5537 		ctl_set_invalid_opcode(ctsio);
5538 		ctl_done((union ctl_io *)ctsio);
5539 		return (CTL_RETVAL_COMPLETE);
5540 	}
5541 
5542 	/*
5543 	 * XXX KDM Copan-specific offline behavior.
5544 	 * Figure out a reasonable way to port this?
5545 	 */
5546 #ifdef NEEDTOPORT
5547 	mtx_lock(&lun->lun_lock);
5548 
5549 	if (((cdb->byte2 & SSS_ONOFFLINE) == 0)
5550 	 && (lun->flags & CTL_LUN_OFFLINE)) {
5551 		/*
5552 		 * If the LUN is offline, and the on/offline bit isn't set,
5553 		 * reject the start or stop.  Otherwise, let it through.
5554 		 */
5555 		mtx_unlock(&lun->lun_lock);
5556 		ctl_set_lun_not_ready(ctsio);
5557 		ctl_done((union ctl_io *)ctsio);
5558 	} else {
5559 		mtx_unlock(&lun->lun_lock);
5560 #endif /* NEEDTOPORT */
5561 		/*
5562 		 * This could be a start or a stop when we're online,
5563 		 * or a stop/offline or start/online.  A start or stop when
5564 		 * we're offline is covered in the case above.
5565 		 */
5566 		/*
5567 		 * In the non-immediate case, we send the request to
5568 		 * the backend and return status to the user when
5569 		 * it is done.
5570 		 *
5571 		 * In the immediate case, we allocate a new ctl_io
5572 		 * to hold a copy of the request, and send that to
5573 		 * the backend.  We then set good status on the
5574 		 * user's request and return it immediately.
5575 		 */
5576 		if (cdb->byte2 & SSS_IMMED) {
5577 			union ctl_io *new_io;
5578 
5579 			new_io = ctl_alloc_io(ctsio->io_hdr.pool);
5580 			if (new_io == NULL) {
5581 				ctl_set_busy(ctsio);
5582 				ctl_done((union ctl_io *)ctsio);
5583 			} else {
5584 				ctl_copy_io((union ctl_io *)ctsio,
5585 					    new_io);
5586 				retval = lun->backend->config_write(new_io);
5587 				ctl_set_success(ctsio);
5588 				ctl_done((union ctl_io *)ctsio);
5589 			}
5590 		} else {
5591 			retval = lun->backend->config_write(
5592 				(union ctl_io *)ctsio);
5593 		}
5594 #ifdef NEEDTOPORT
5595 	}
5596 #endif
5597 	return (retval);
5598 }
5599 
5600 /*
5601  * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5602  * we don't really do anything with the LBA and length fields if the user
5603  * passes them in.  Instead we'll just flush out the cache for the entire
5604  * LUN.
5605  */
5606 int
5607 ctl_sync_cache(struct ctl_scsiio *ctsio)
5608 {
5609 	struct ctl_lun *lun;
5610 	struct ctl_softc *ctl_softc;
5611 	uint64_t starting_lba;
5612 	uint32_t block_count;
5613 	int retval;
5614 
5615 	CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5616 
5617 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5618 	ctl_softc = control_softc;
5619 	retval = 0;
5620 
5621 	switch (ctsio->cdb[0]) {
5622 	case SYNCHRONIZE_CACHE: {
5623 		struct scsi_sync_cache *cdb;
5624 		cdb = (struct scsi_sync_cache *)ctsio->cdb;
5625 
5626 		starting_lba = scsi_4btoul(cdb->begin_lba);
5627 		block_count = scsi_2btoul(cdb->lb_count);
5628 		break;
5629 	}
5630 	case SYNCHRONIZE_CACHE_16: {
5631 		struct scsi_sync_cache_16 *cdb;
5632 		cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5633 
5634 		starting_lba = scsi_8btou64(cdb->begin_lba);
5635 		block_count = scsi_4btoul(cdb->lb_count);
5636 		break;
5637 	}
5638 	default:
5639 		ctl_set_invalid_opcode(ctsio);
5640 		ctl_done((union ctl_io *)ctsio);
5641 		goto bailout;
5642 		break; /* NOTREACHED */
5643 	}
5644 
5645 	/*
5646 	 * We check the LBA and length, but don't do anything with them.
5647 	 * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5648 	 * get flushed.  This check will just help satisfy anyone who wants
5649 	 * to see an error for an out of range LBA.
5650 	 */
5651 	if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5652 		ctl_set_lba_out_of_range(ctsio);
5653 		ctl_done((union ctl_io *)ctsio);
5654 		goto bailout;
5655 	}
5656 
5657 	/*
5658 	 * If this LUN has no backend, we can't flush the cache anyway.
5659 	 */
5660 	if (lun->backend == NULL) {
5661 		ctl_set_invalid_opcode(ctsio);
5662 		ctl_done((union ctl_io *)ctsio);
5663 		goto bailout;
5664 	}
5665 
5666 	/*
5667 	 * Check to see whether we're configured to send the SYNCHRONIZE
5668 	 * CACHE command directly to the back end.
5669 	 */
5670 	mtx_lock(&lun->lun_lock);
5671 	if ((ctl_softc->flags & CTL_FLAG_REAL_SYNC)
5672 	 && (++(lun->sync_count) >= lun->sync_interval)) {
5673 		lun->sync_count = 0;
5674 		mtx_unlock(&lun->lun_lock);
5675 		retval = lun->backend->config_write((union ctl_io *)ctsio);
5676 	} else {
5677 		mtx_unlock(&lun->lun_lock);
5678 		ctl_set_success(ctsio);
5679 		ctl_done((union ctl_io *)ctsio);
5680 	}
5681 
5682 bailout:
5683 
5684 	return (retval);
5685 }
5686 
5687 int
5688 ctl_format(struct ctl_scsiio *ctsio)
5689 {
5690 	struct scsi_format *cdb;
5691 	struct ctl_lun *lun;
5692 	struct ctl_softc *ctl_softc;
5693 	int length, defect_list_len;
5694 
5695 	CTL_DEBUG_PRINT(("ctl_format\n"));
5696 
5697 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5698 	ctl_softc = control_softc;
5699 
5700 	cdb = (struct scsi_format *)ctsio->cdb;
5701 
5702 	length = 0;
5703 	if (cdb->byte2 & SF_FMTDATA) {
5704 		if (cdb->byte2 & SF_LONGLIST)
5705 			length = sizeof(struct scsi_format_header_long);
5706 		else
5707 			length = sizeof(struct scsi_format_header_short);
5708 	}
5709 
5710 	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5711 	 && (length > 0)) {
5712 		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5713 		ctsio->kern_data_len = length;
5714 		ctsio->kern_total_len = length;
5715 		ctsio->kern_data_resid = 0;
5716 		ctsio->kern_rel_offset = 0;
5717 		ctsio->kern_sg_entries = 0;
5718 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5719 		ctsio->be_move_done = ctl_config_move_done;
5720 		ctl_datamove((union ctl_io *)ctsio);
5721 
5722 		return (CTL_RETVAL_COMPLETE);
5723 	}
5724 
5725 	defect_list_len = 0;
5726 
5727 	if (cdb->byte2 & SF_FMTDATA) {
5728 		if (cdb->byte2 & SF_LONGLIST) {
5729 			struct scsi_format_header_long *header;
5730 
5731 			header = (struct scsi_format_header_long *)
5732 				ctsio->kern_data_ptr;
5733 
5734 			defect_list_len = scsi_4btoul(header->defect_list_len);
5735 			if (defect_list_len != 0) {
5736 				ctl_set_invalid_field(ctsio,
5737 						      /*sks_valid*/ 1,
5738 						      /*command*/ 0,
5739 						      /*field*/ 2,
5740 						      /*bit_valid*/ 0,
5741 						      /*bit*/ 0);
5742 				goto bailout;
5743 			}
5744 		} else {
5745 			struct scsi_format_header_short *header;
5746 
5747 			header = (struct scsi_format_header_short *)
5748 				ctsio->kern_data_ptr;
5749 
5750 			defect_list_len = scsi_2btoul(header->defect_list_len);
5751 			if (defect_list_len != 0) {
5752 				ctl_set_invalid_field(ctsio,
5753 						      /*sks_valid*/ 1,
5754 						      /*command*/ 0,
5755 						      /*field*/ 2,
5756 						      /*bit_valid*/ 0,
5757 						      /*bit*/ 0);
5758 				goto bailout;
5759 			}
5760 		}
5761 	}
5762 
5763 	/*
5764 	 * The format command will clear out the "Medium format corrupted"
5765 	 * status if set by the configuration code.  That status is really
5766 	 * just a way to notify the host that we have lost the media, and
5767 	 * get them to issue a command that will basically make them think
5768 	 * they're blowing away the media.
5769 	 */
5770 	mtx_lock(&lun->lun_lock);
5771 	lun->flags &= ~CTL_LUN_INOPERABLE;
5772 	mtx_unlock(&lun->lun_lock);
5773 
5774 	ctsio->scsi_status = SCSI_STATUS_OK;
5775 	ctsio->io_hdr.status = CTL_SUCCESS;
5776 bailout:
5777 
5778 	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5779 		free(ctsio->kern_data_ptr, M_CTL);
5780 		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5781 	}
5782 
5783 	ctl_done((union ctl_io *)ctsio);
5784 	return (CTL_RETVAL_COMPLETE);
5785 }
5786 
5787 int
5788 ctl_read_buffer(struct ctl_scsiio *ctsio)
5789 {
5790 	struct scsi_read_buffer *cdb;
5791 	struct ctl_lun *lun;
5792 	int buffer_offset, len;
5793 	static uint8_t descr[4];
5794 	static uint8_t echo_descr[4] = { 0 };
5795 
5796 	CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5797 
5798 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5799 	cdb = (struct scsi_read_buffer *)ctsio->cdb;
5800 
5801 	if (lun->flags & CTL_LUN_PR_RESERVED) {
5802 		uint32_t residx;
5803 
5804 		/*
5805 		 * XXX KDM need a lock here.
5806 		 */
5807 		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5808 		if ((lun->res_type == SPR_TYPE_EX_AC
5809 		  && residx != lun->pr_res_idx)
5810 		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
5811 		   || lun->res_type == SPR_TYPE_EX_AC_AR)
5812 		  && !lun->per_res[residx].registered)) {
5813 			ctl_set_reservation_conflict(ctsio);
5814 			ctl_done((union ctl_io *)ctsio);
5815 			return (CTL_RETVAL_COMPLETE);
5816 	        }
5817 	}
5818 
5819 	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA &&
5820 	    (cdb->byte2 & RWB_MODE) != RWB_MODE_ECHO_DESCR &&
5821 	    (cdb->byte2 & RWB_MODE) != RWB_MODE_DESCR) {
5822 		ctl_set_invalid_field(ctsio,
5823 				      /*sks_valid*/ 1,
5824 				      /*command*/ 1,
5825 				      /*field*/ 1,
5826 				      /*bit_valid*/ 1,
5827 				      /*bit*/ 4);
5828 		ctl_done((union ctl_io *)ctsio);
5829 		return (CTL_RETVAL_COMPLETE);
5830 	}
5831 
5832 	len = scsi_3btoul(cdb->length);
5833 	buffer_offset = scsi_3btoul(cdb->offset);
5834 
5835 	if (buffer_offset + len > sizeof(lun->write_buffer)) {
5836 		ctl_set_invalid_field(ctsio,
5837 				      /*sks_valid*/ 1,
5838 				      /*command*/ 1,
5839 				      /*field*/ 6,
5840 				      /*bit_valid*/ 0,
5841 				      /*bit*/ 0);
5842 		ctl_done((union ctl_io *)ctsio);
5843 		return (CTL_RETVAL_COMPLETE);
5844 	}
5845 
5846 	if ((cdb->byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5847 		descr[0] = 0;
5848 		scsi_ulto3b(sizeof(lun->write_buffer), &descr[1]);
5849 		ctsio->kern_data_ptr = descr;
5850 		len = min(len, sizeof(descr));
5851 	} else if ((cdb->byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5852 		ctsio->kern_data_ptr = echo_descr;
5853 		len = min(len, sizeof(echo_descr));
5854 	} else
5855 		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5856 	ctsio->kern_data_len = len;
5857 	ctsio->kern_total_len = len;
5858 	ctsio->kern_data_resid = 0;
5859 	ctsio->kern_rel_offset = 0;
5860 	ctsio->kern_sg_entries = 0;
5861 	ctsio->be_move_done = ctl_config_move_done;
5862 	ctl_datamove((union ctl_io *)ctsio);
5863 
5864 	return (CTL_RETVAL_COMPLETE);
5865 }
5866 
5867 int
5868 ctl_write_buffer(struct ctl_scsiio *ctsio)
5869 {
5870 	struct scsi_write_buffer *cdb;
5871 	struct ctl_lun *lun;
5872 	int buffer_offset, len;
5873 
5874 	CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5875 
5876 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5877 	cdb = (struct scsi_write_buffer *)ctsio->cdb;
5878 
5879 	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5880 		ctl_set_invalid_field(ctsio,
5881 				      /*sks_valid*/ 1,
5882 				      /*command*/ 1,
5883 				      /*field*/ 1,
5884 				      /*bit_valid*/ 1,
5885 				      /*bit*/ 4);
5886 		ctl_done((union ctl_io *)ctsio);
5887 		return (CTL_RETVAL_COMPLETE);
5888 	}
5889 
5890 	len = scsi_3btoul(cdb->length);
5891 	buffer_offset = scsi_3btoul(cdb->offset);
5892 
5893 	if (buffer_offset + len > sizeof(lun->write_buffer)) {
5894 		ctl_set_invalid_field(ctsio,
5895 				      /*sks_valid*/ 1,
5896 				      /*command*/ 1,
5897 				      /*field*/ 6,
5898 				      /*bit_valid*/ 0,
5899 				      /*bit*/ 0);
5900 		ctl_done((union ctl_io *)ctsio);
5901 		return (CTL_RETVAL_COMPLETE);
5902 	}
5903 
5904 	/*
5905 	 * If we've got a kernel request that hasn't been malloced yet,
5906 	 * malloc it and tell the caller the data buffer is here.
5907 	 */
5908 	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5909 		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5910 		ctsio->kern_data_len = len;
5911 		ctsio->kern_total_len = len;
5912 		ctsio->kern_data_resid = 0;
5913 		ctsio->kern_rel_offset = 0;
5914 		ctsio->kern_sg_entries = 0;
5915 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5916 		ctsio->be_move_done = ctl_config_move_done;
5917 		ctl_datamove((union ctl_io *)ctsio);
5918 
5919 		return (CTL_RETVAL_COMPLETE);
5920 	}
5921 
5922 	ctl_done((union ctl_io *)ctsio);
5923 
5924 	return (CTL_RETVAL_COMPLETE);
5925 }
5926 
5927 int
5928 ctl_write_same(struct ctl_scsiio *ctsio)
5929 {
5930 	struct ctl_lun *lun;
5931 	struct ctl_lba_len_flags *lbalen;
5932 	uint64_t lba;
5933 	uint32_t num_blocks;
5934 	int len, retval;
5935 	uint8_t byte2;
5936 
5937 	retval = CTL_RETVAL_COMPLETE;
5938 
5939 	CTL_DEBUG_PRINT(("ctl_write_same\n"));
5940 
5941 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5942 
5943 	switch (ctsio->cdb[0]) {
5944 	case WRITE_SAME_10: {
5945 		struct scsi_write_same_10 *cdb;
5946 
5947 		cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5948 
5949 		lba = scsi_4btoul(cdb->addr);
5950 		num_blocks = scsi_2btoul(cdb->length);
5951 		byte2 = cdb->byte2;
5952 		break;
5953 	}
5954 	case WRITE_SAME_16: {
5955 		struct scsi_write_same_16 *cdb;
5956 
5957 		cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5958 
5959 		lba = scsi_8btou64(cdb->addr);
5960 		num_blocks = scsi_4btoul(cdb->length);
5961 		byte2 = cdb->byte2;
5962 		break;
5963 	}
5964 	default:
5965 		/*
5966 		 * We got a command we don't support.  This shouldn't
5967 		 * happen, commands should be filtered out above us.
5968 		 */
5969 		ctl_set_invalid_opcode(ctsio);
5970 		ctl_done((union ctl_io *)ctsio);
5971 
5972 		return (CTL_RETVAL_COMPLETE);
5973 		break; /* NOTREACHED */
5974 	}
5975 
5976 	/*
5977 	 * The first check is to make sure we're in bounds, the second
5978 	 * check is to catch wrap-around problems.  If the lba + num blocks
5979 	 * is less than the lba, then we've wrapped around and the block
5980 	 * range is invalid anyway.
5981 	 */
5982 	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5983 	 || ((lba + num_blocks) < lba)) {
5984 		ctl_set_lba_out_of_range(ctsio);
5985 		ctl_done((union ctl_io *)ctsio);
5986 		return (CTL_RETVAL_COMPLETE);
5987 	}
5988 
5989 	/* Zero number of blocks means "to the last logical block" */
5990 	if (num_blocks == 0) {
5991 		if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5992 			ctl_set_invalid_field(ctsio,
5993 					      /*sks_valid*/ 0,
5994 					      /*command*/ 1,
5995 					      /*field*/ 0,
5996 					      /*bit_valid*/ 0,
5997 					      /*bit*/ 0);
5998 			ctl_done((union ctl_io *)ctsio);
5999 			return (CTL_RETVAL_COMPLETE);
6000 		}
6001 		num_blocks = (lun->be_lun->maxlba + 1) - lba;
6002 	}
6003 
6004 	len = lun->be_lun->blocksize;
6005 
6006 	/*
6007 	 * If we've got a kernel request that hasn't been malloced yet,
6008 	 * malloc it and tell the caller the data buffer is here.
6009 	 */
6010 	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6011 		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
6012 		ctsio->kern_data_len = len;
6013 		ctsio->kern_total_len = len;
6014 		ctsio->kern_data_resid = 0;
6015 		ctsio->kern_rel_offset = 0;
6016 		ctsio->kern_sg_entries = 0;
6017 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6018 		ctsio->be_move_done = ctl_config_move_done;
6019 		ctl_datamove((union ctl_io *)ctsio);
6020 
6021 		return (CTL_RETVAL_COMPLETE);
6022 	}
6023 
6024 	lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
6025 	lbalen->lba = lba;
6026 	lbalen->len = num_blocks;
6027 	lbalen->flags = byte2;
6028 	retval = lun->backend->config_write((union ctl_io *)ctsio);
6029 
6030 	return (retval);
6031 }
6032 
6033 int
6034 ctl_unmap(struct ctl_scsiio *ctsio)
6035 {
6036 	struct ctl_lun *lun;
6037 	struct scsi_unmap *cdb;
6038 	struct ctl_ptr_len_flags *ptrlen;
6039 	struct scsi_unmap_header *hdr;
6040 	struct scsi_unmap_desc *buf, *end;
6041 	uint64_t lba;
6042 	uint32_t num_blocks;
6043 	int len, retval;
6044 	uint8_t byte2;
6045 
6046 	retval = CTL_RETVAL_COMPLETE;
6047 
6048 	CTL_DEBUG_PRINT(("ctl_unmap\n"));
6049 
6050 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6051 	cdb = (struct scsi_unmap *)ctsio->cdb;
6052 
6053 	len = scsi_2btoul(cdb->length);
6054 	byte2 = cdb->byte2;
6055 
6056 	/*
6057 	 * If we've got a kernel request that hasn't been malloced yet,
6058 	 * malloc it and tell the caller the data buffer is here.
6059 	 */
6060 	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6061 		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
6062 		ctsio->kern_data_len = len;
6063 		ctsio->kern_total_len = len;
6064 		ctsio->kern_data_resid = 0;
6065 		ctsio->kern_rel_offset = 0;
6066 		ctsio->kern_sg_entries = 0;
6067 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6068 		ctsio->be_move_done = ctl_config_move_done;
6069 		ctl_datamove((union ctl_io *)ctsio);
6070 
6071 		return (CTL_RETVAL_COMPLETE);
6072 	}
6073 
6074 	len = ctsio->kern_total_len - ctsio->kern_data_resid;
6075 	hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
6076 	if (len < sizeof (*hdr) ||
6077 	    len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
6078 	    len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
6079 	    scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
6080 		ctl_set_invalid_field(ctsio,
6081 				      /*sks_valid*/ 0,
6082 				      /*command*/ 0,
6083 				      /*field*/ 0,
6084 				      /*bit_valid*/ 0,
6085 				      /*bit*/ 0);
6086 		ctl_done((union ctl_io *)ctsio);
6087 		return (CTL_RETVAL_COMPLETE);
6088 	}
6089 	len = scsi_2btoul(hdr->desc_length);
6090 	buf = (struct scsi_unmap_desc *)(hdr + 1);
6091 	end = buf + len / sizeof(*buf);
6092 
6093 	ptrlen = (struct ctl_ptr_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
6094 	ptrlen->ptr = (void *)buf;
6095 	ptrlen->len = len;
6096 	ptrlen->flags = byte2;
6097 
6098 	for (; buf < end; buf++) {
6099 		lba = scsi_8btou64(buf->lba);
6100 		num_blocks = scsi_4btoul(buf->length);
6101 		if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
6102 		 || ((lba + num_blocks) < lba)) {
6103 			ctl_set_lba_out_of_range(ctsio);
6104 			ctl_done((union ctl_io *)ctsio);
6105 			return (CTL_RETVAL_COMPLETE);
6106 		}
6107 	}
6108 
6109 	retval = lun->backend->config_write((union ctl_io *)ctsio);
6110 
6111 	return (retval);
6112 }
6113 
6114 /*
6115  * Note that this function currently doesn't actually do anything inside
6116  * CTL to enforce things if the DQue bit is turned on.
6117  *
6118  * Also note that this function can't be used in the default case, because
6119  * the DQue bit isn't set in the changeable mask for the control mode page
6120  * anyway.  This is just here as an example for how to implement a page
6121  * handler, and a placeholder in case we want to allow the user to turn
6122  * tagged queueing on and off.
6123  *
6124  * The D_SENSE bit handling is functional, however, and will turn
6125  * descriptor sense on and off for a given LUN.
6126  */
6127 int
6128 ctl_control_page_handler(struct ctl_scsiio *ctsio,
6129 			 struct ctl_page_index *page_index, uint8_t *page_ptr)
6130 {
6131 	struct scsi_control_page *current_cp, *saved_cp, *user_cp;
6132 	struct ctl_lun *lun;
6133 	struct ctl_softc *softc;
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_control_page *)page_ptr;
6142 	current_cp = (struct scsi_control_page *)
6143 		(page_index->page_data + (page_index->page_len *
6144 		CTL_PAGE_CURRENT));
6145 	saved_cp = (struct scsi_control_page *)
6146 		(page_index->page_data + (page_index->page_len *
6147 		CTL_PAGE_SAVED));
6148 
6149 	softc = control_softc;
6150 
6151 	mtx_lock(&lun->lun_lock);
6152 	if (((current_cp->rlec & SCP_DSENSE) == 0)
6153 	 && ((user_cp->rlec & SCP_DSENSE) != 0)) {
6154 		/*
6155 		 * Descriptor sense is currently turned off and the user
6156 		 * wants to turn it on.
6157 		 */
6158 		current_cp->rlec |= SCP_DSENSE;
6159 		saved_cp->rlec |= SCP_DSENSE;
6160 		lun->flags |= CTL_LUN_SENSE_DESC;
6161 		set_ua = 1;
6162 	} else if (((current_cp->rlec & SCP_DSENSE) != 0)
6163 		&& ((user_cp->rlec & SCP_DSENSE) == 0)) {
6164 		/*
6165 		 * Descriptor sense is currently turned on, and the user
6166 		 * wants to turn it off.
6167 		 */
6168 		current_cp->rlec &= ~SCP_DSENSE;
6169 		saved_cp->rlec &= ~SCP_DSENSE;
6170 		lun->flags &= ~CTL_LUN_SENSE_DESC;
6171 		set_ua = 1;
6172 	}
6173 	if (current_cp->queue_flags & SCP_QUEUE_DQUE) {
6174 		if (user_cp->queue_flags & SCP_QUEUE_DQUE) {
6175 #ifdef NEEDTOPORT
6176 			csevent_log(CSC_CTL | CSC_SHELF_SW |
6177 				    CTL_UNTAG_TO_UNTAG,
6178 				    csevent_LogType_Trace,
6179 				    csevent_Severity_Information,
6180 				    csevent_AlertLevel_Green,
6181 				    csevent_FRU_Firmware,
6182 				    csevent_FRU_Unknown,
6183 				    "Received untagged to untagged transition");
6184 #endif /* NEEDTOPORT */
6185 		} else {
6186 #ifdef NEEDTOPORT
6187 			csevent_log(CSC_CTL | CSC_SHELF_SW |
6188 				    CTL_UNTAG_TO_TAG,
6189 				    csevent_LogType_ConfigChange,
6190 				    csevent_Severity_Information,
6191 				    csevent_AlertLevel_Green,
6192 				    csevent_FRU_Firmware,
6193 				    csevent_FRU_Unknown,
6194 				    "Received untagged to tagged "
6195 				    "queueing transition");
6196 #endif /* NEEDTOPORT */
6197 
6198 			current_cp->queue_flags &= ~SCP_QUEUE_DQUE;
6199 			saved_cp->queue_flags &= ~SCP_QUEUE_DQUE;
6200 			set_ua = 1;
6201 		}
6202 	} else {
6203 		if (user_cp->queue_flags & SCP_QUEUE_DQUE) {
6204 #ifdef NEEDTOPORT
6205 			csevent_log(CSC_CTL | CSC_SHELF_SW |
6206 				    CTL_TAG_TO_UNTAG,
6207 				    csevent_LogType_ConfigChange,
6208 				    csevent_Severity_Warning,
6209 				    csevent_AlertLevel_Yellow,
6210 				    csevent_FRU_Firmware,
6211 				    csevent_FRU_Unknown,
6212 				    "Received tagged queueing to untagged "
6213 				    "transition");
6214 #endif /* NEEDTOPORT */
6215 
6216 			current_cp->queue_flags |= SCP_QUEUE_DQUE;
6217 			saved_cp->queue_flags |= SCP_QUEUE_DQUE;
6218 			set_ua = 1;
6219 		} else {
6220 #ifdef NEEDTOPORT
6221 			csevent_log(CSC_CTL | CSC_SHELF_SW |
6222 				    CTL_TAG_TO_TAG,
6223 				    csevent_LogType_Trace,
6224 				    csevent_Severity_Information,
6225 				    csevent_AlertLevel_Green,
6226 				    csevent_FRU_Firmware,
6227 				    csevent_FRU_Unknown,
6228 				    "Received tagged queueing to tagged "
6229 				    "queueing transition");
6230 #endif /* NEEDTOPORT */
6231 		}
6232 	}
6233 	if (set_ua != 0) {
6234 		int i;
6235 		/*
6236 		 * Let other initiators know that the mode
6237 		 * parameters for this LUN have changed.
6238 		 */
6239 		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
6240 			if (i == initidx)
6241 				continue;
6242 
6243 			lun->pending_ua[i] |= CTL_UA_MODE_CHANGE;
6244 		}
6245 	}
6246 	mtx_unlock(&lun->lun_lock);
6247 
6248 	return (0);
6249 }
6250 
6251 int
6252 ctl_power_sp_handler(struct ctl_scsiio *ctsio,
6253 		     struct ctl_page_index *page_index, uint8_t *page_ptr)
6254 {
6255 	return (0);
6256 }
6257 
6258 int
6259 ctl_power_sp_sense_handler(struct ctl_scsiio *ctsio,
6260 			   struct ctl_page_index *page_index, int pc)
6261 {
6262 	struct copan_power_subpage *page;
6263 
6264 	page = (struct copan_power_subpage *)page_index->page_data +
6265 		(page_index->page_len * pc);
6266 
6267 	switch (pc) {
6268 	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6269 		/*
6270 		 * We don't update the changable bits for this page.
6271 		 */
6272 		break;
6273 	case SMS_PAGE_CTRL_CURRENT >> 6:
6274 	case SMS_PAGE_CTRL_DEFAULT >> 6:
6275 	case SMS_PAGE_CTRL_SAVED >> 6:
6276 #ifdef NEEDTOPORT
6277 		ctl_update_power_subpage(page);
6278 #endif
6279 		break;
6280 	default:
6281 #ifdef NEEDTOPORT
6282 		EPRINT(0, "Invalid PC %d!!", pc);
6283 #endif
6284 		break;
6285 	}
6286 	return (0);
6287 }
6288 
6289 
6290 int
6291 ctl_aps_sp_handler(struct ctl_scsiio *ctsio,
6292 		   struct ctl_page_index *page_index, uint8_t *page_ptr)
6293 {
6294 	struct copan_aps_subpage *user_sp;
6295 	struct copan_aps_subpage *current_sp;
6296 	union ctl_modepage_info *modepage_info;
6297 	struct ctl_softc *softc;
6298 	struct ctl_lun *lun;
6299 	int retval;
6300 
6301 	retval = CTL_RETVAL_COMPLETE;
6302 	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
6303 		     (page_index->page_len * CTL_PAGE_CURRENT));
6304 	softc = control_softc;
6305 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6306 
6307 	user_sp = (struct copan_aps_subpage *)page_ptr;
6308 
6309 	modepage_info = (union ctl_modepage_info *)
6310 		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6311 
6312 	modepage_info->header.page_code = page_index->page_code & SMPH_PC_MASK;
6313 	modepage_info->header.subpage = page_index->subpage;
6314 	modepage_info->aps.lock_active = user_sp->lock_active;
6315 
6316 	mtx_lock(&softc->ctl_lock);
6317 
6318 	/*
6319 	 * If there is a request to lock the LUN and another LUN is locked
6320 	 * this is an error. If the requested LUN is already locked ignore
6321 	 * the request. If no LUN is locked attempt to lock it.
6322 	 * if there is a request to unlock the LUN and the LUN is currently
6323 	 * locked attempt to unlock it. Otherwise ignore the request. i.e.
6324 	 * if another LUN is locked or no LUN is locked.
6325 	 */
6326 	if (user_sp->lock_active & APS_LOCK_ACTIVE) {
6327 		if (softc->aps_locked_lun == lun->lun) {
6328 			/*
6329 			 * This LUN is already locked, so we're done.
6330 			 */
6331 			retval = CTL_RETVAL_COMPLETE;
6332 		} else if (softc->aps_locked_lun == 0) {
6333 			/*
6334 			 * No one has the lock, pass the request to the
6335 			 * backend.
6336 			 */
6337 			retval = lun->backend->config_write(
6338 				(union ctl_io *)ctsio);
6339 		} else {
6340 			/*
6341 			 * Someone else has the lock, throw out the request.
6342 			 */
6343 			ctl_set_already_locked(ctsio);
6344 			free(ctsio->kern_data_ptr, M_CTL);
6345 			ctl_done((union ctl_io *)ctsio);
6346 
6347 			/*
6348 			 * Set the return value so that ctl_do_mode_select()
6349 			 * won't try to complete the command.  We already
6350 			 * completed it here.
6351 			 */
6352 			retval = CTL_RETVAL_ERROR;
6353 		}
6354 	} else if (softc->aps_locked_lun == lun->lun) {
6355 		/*
6356 		 * This LUN is locked, so pass the unlock request to the
6357 		 * backend.
6358 		 */
6359 		retval = lun->backend->config_write((union ctl_io *)ctsio);
6360 	}
6361 	mtx_unlock(&softc->ctl_lock);
6362 
6363 	return (retval);
6364 }
6365 
6366 int
6367 ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio,
6368 				struct ctl_page_index *page_index,
6369 				uint8_t *page_ptr)
6370 {
6371 	uint8_t *c;
6372 	int i;
6373 
6374 	c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs;
6375 	ctl_time_io_secs =
6376 		(c[0] << 8) |
6377 		(c[1] << 0) |
6378 		0;
6379 	CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs));
6380 	printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs);
6381 	printf("page data:");
6382 	for (i=0; i<8; i++)
6383 		printf(" %.2x",page_ptr[i]);
6384 	printf("\n");
6385 	return (0);
6386 }
6387 
6388 int
6389 ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio,
6390 			       struct ctl_page_index *page_index,
6391 			       int pc)
6392 {
6393 	struct copan_debugconf_subpage *page;
6394 
6395 	page = (struct copan_debugconf_subpage *)page_index->page_data +
6396 		(page_index->page_len * pc);
6397 
6398 	switch (pc) {
6399 	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6400 	case SMS_PAGE_CTRL_DEFAULT >> 6:
6401 	case SMS_PAGE_CTRL_SAVED >> 6:
6402 		/*
6403 		 * We don't update the changable or default bits for this page.
6404 		 */
6405 		break;
6406 	case SMS_PAGE_CTRL_CURRENT >> 6:
6407 		page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8;
6408 		page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0;
6409 		break;
6410 	default:
6411 #ifdef NEEDTOPORT
6412 		EPRINT(0, "Invalid PC %d!!", pc);
6413 #endif /* NEEDTOPORT */
6414 		break;
6415 	}
6416 	return (0);
6417 }
6418 
6419 
6420 static int
6421 ctl_do_mode_select(union ctl_io *io)
6422 {
6423 	struct scsi_mode_page_header *page_header;
6424 	struct ctl_page_index *page_index;
6425 	struct ctl_scsiio *ctsio;
6426 	int control_dev, page_len;
6427 	int page_len_offset, page_len_size;
6428 	union ctl_modepage_info *modepage_info;
6429 	struct ctl_lun *lun;
6430 	int *len_left, *len_used;
6431 	int retval, i;
6432 
6433 	ctsio = &io->scsiio;
6434 	page_index = NULL;
6435 	page_len = 0;
6436 	retval = CTL_RETVAL_COMPLETE;
6437 
6438 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6439 
6440 	if (lun->be_lun->lun_type != T_DIRECT)
6441 		control_dev = 1;
6442 	else
6443 		control_dev = 0;
6444 
6445 	modepage_info = (union ctl_modepage_info *)
6446 		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6447 	len_left = &modepage_info->header.len_left;
6448 	len_used = &modepage_info->header.len_used;
6449 
6450 do_next_page:
6451 
6452 	page_header = (struct scsi_mode_page_header *)
6453 		(ctsio->kern_data_ptr + *len_used);
6454 
6455 	if (*len_left == 0) {
6456 		free(ctsio->kern_data_ptr, M_CTL);
6457 		ctl_set_success(ctsio);
6458 		ctl_done((union ctl_io *)ctsio);
6459 		return (CTL_RETVAL_COMPLETE);
6460 	} else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6461 
6462 		free(ctsio->kern_data_ptr, M_CTL);
6463 		ctl_set_param_len_error(ctsio);
6464 		ctl_done((union ctl_io *)ctsio);
6465 		return (CTL_RETVAL_COMPLETE);
6466 
6467 	} else if ((page_header->page_code & SMPH_SPF)
6468 		&& (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6469 
6470 		free(ctsio->kern_data_ptr, M_CTL);
6471 		ctl_set_param_len_error(ctsio);
6472 		ctl_done((union ctl_io *)ctsio);
6473 		return (CTL_RETVAL_COMPLETE);
6474 	}
6475 
6476 
6477 	/*
6478 	 * XXX KDM should we do something with the block descriptor?
6479 	 */
6480 	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6481 
6482 		if ((control_dev != 0)
6483 		 && (lun->mode_pages.index[i].page_flags &
6484 		     CTL_PAGE_FLAG_DISK_ONLY))
6485 			continue;
6486 
6487 		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
6488 		    (page_header->page_code & SMPH_PC_MASK))
6489 			continue;
6490 
6491 		/*
6492 		 * If neither page has a subpage code, then we've got a
6493 		 * match.
6494 		 */
6495 		if (((lun->mode_pages.index[i].page_code & SMPH_SPF) == 0)
6496 		 && ((page_header->page_code & SMPH_SPF) == 0)) {
6497 			page_index = &lun->mode_pages.index[i];
6498 			page_len = page_header->page_length;
6499 			break;
6500 		}
6501 
6502 		/*
6503 		 * If both pages have subpages, then the subpage numbers
6504 		 * have to match.
6505 		 */
6506 		if ((lun->mode_pages.index[i].page_code & SMPH_SPF)
6507 		  && (page_header->page_code & SMPH_SPF)) {
6508 			struct scsi_mode_page_header_sp *sph;
6509 
6510 			sph = (struct scsi_mode_page_header_sp *)page_header;
6511 
6512 			if (lun->mode_pages.index[i].subpage ==
6513 			    sph->subpage) {
6514 				page_index = &lun->mode_pages.index[i];
6515 				page_len = scsi_2btoul(sph->page_length);
6516 				break;
6517 			}
6518 		}
6519 	}
6520 
6521 	/*
6522 	 * If we couldn't find the page, or if we don't have a mode select
6523 	 * handler for it, send back an error to the user.
6524 	 */
6525 	if ((page_index == NULL)
6526 	 || (page_index->select_handler == NULL)) {
6527 		ctl_set_invalid_field(ctsio,
6528 				      /*sks_valid*/ 1,
6529 				      /*command*/ 0,
6530 				      /*field*/ *len_used,
6531 				      /*bit_valid*/ 0,
6532 				      /*bit*/ 0);
6533 		free(ctsio->kern_data_ptr, M_CTL);
6534 		ctl_done((union ctl_io *)ctsio);
6535 		return (CTL_RETVAL_COMPLETE);
6536 	}
6537 
6538 	if (page_index->page_code & SMPH_SPF) {
6539 		page_len_offset = 2;
6540 		page_len_size = 2;
6541 	} else {
6542 		page_len_size = 1;
6543 		page_len_offset = 1;
6544 	}
6545 
6546 	/*
6547 	 * If the length the initiator gives us isn't the one we specify in
6548 	 * the mode page header, or if they didn't specify enough data in
6549 	 * the CDB to avoid truncating this page, kick out the request.
6550 	 */
6551 	if ((page_len != (page_index->page_len - page_len_offset -
6552 			  page_len_size))
6553 	 || (*len_left < page_index->page_len)) {
6554 
6555 
6556 		ctl_set_invalid_field(ctsio,
6557 				      /*sks_valid*/ 1,
6558 				      /*command*/ 0,
6559 				      /*field*/ *len_used + page_len_offset,
6560 				      /*bit_valid*/ 0,
6561 				      /*bit*/ 0);
6562 		free(ctsio->kern_data_ptr, M_CTL);
6563 		ctl_done((union ctl_io *)ctsio);
6564 		return (CTL_RETVAL_COMPLETE);
6565 	}
6566 
6567 	/*
6568 	 * Run through the mode page, checking to make sure that the bits
6569 	 * the user changed are actually legal for him to change.
6570 	 */
6571 	for (i = 0; i < page_index->page_len; i++) {
6572 		uint8_t *user_byte, *change_mask, *current_byte;
6573 		int bad_bit;
6574 		int j;
6575 
6576 		user_byte = (uint8_t *)page_header + i;
6577 		change_mask = page_index->page_data +
6578 			      (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6579 		current_byte = page_index->page_data +
6580 			       (page_index->page_len * CTL_PAGE_CURRENT) + i;
6581 
6582 		/*
6583 		 * Check to see whether the user set any bits in this byte
6584 		 * that he is not allowed to set.
6585 		 */
6586 		if ((*user_byte & ~(*change_mask)) ==
6587 		    (*current_byte & ~(*change_mask)))
6588 			continue;
6589 
6590 		/*
6591 		 * Go through bit by bit to determine which one is illegal.
6592 		 */
6593 		bad_bit = 0;
6594 		for (j = 7; j >= 0; j--) {
6595 			if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6596 			    (((1 << i) & ~(*change_mask)) & *current_byte)) {
6597 				bad_bit = i;
6598 				break;
6599 			}
6600 		}
6601 		ctl_set_invalid_field(ctsio,
6602 				      /*sks_valid*/ 1,
6603 				      /*command*/ 0,
6604 				      /*field*/ *len_used + i,
6605 				      /*bit_valid*/ 1,
6606 				      /*bit*/ bad_bit);
6607 		free(ctsio->kern_data_ptr, M_CTL);
6608 		ctl_done((union ctl_io *)ctsio);
6609 		return (CTL_RETVAL_COMPLETE);
6610 	}
6611 
6612 	/*
6613 	 * Decrement these before we call the page handler, since we may
6614 	 * end up getting called back one way or another before the handler
6615 	 * returns to this context.
6616 	 */
6617 	*len_left -= page_index->page_len;
6618 	*len_used += page_index->page_len;
6619 
6620 	retval = page_index->select_handler(ctsio, page_index,
6621 					    (uint8_t *)page_header);
6622 
6623 	/*
6624 	 * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6625 	 * wait until this queued command completes to finish processing
6626 	 * the mode page.  If it returns anything other than
6627 	 * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6628 	 * already set the sense information, freed the data pointer, and
6629 	 * completed the io for us.
6630 	 */
6631 	if (retval != CTL_RETVAL_COMPLETE)
6632 		goto bailout_no_done;
6633 
6634 	/*
6635 	 * If the initiator sent us more than one page, parse the next one.
6636 	 */
6637 	if (*len_left > 0)
6638 		goto do_next_page;
6639 
6640 	ctl_set_success(ctsio);
6641 	free(ctsio->kern_data_ptr, M_CTL);
6642 	ctl_done((union ctl_io *)ctsio);
6643 
6644 bailout_no_done:
6645 
6646 	return (CTL_RETVAL_COMPLETE);
6647 
6648 }
6649 
6650 int
6651 ctl_mode_select(struct ctl_scsiio *ctsio)
6652 {
6653 	int param_len, pf, sp;
6654 	int header_size, bd_len;
6655 	int len_left, len_used;
6656 	struct ctl_page_index *page_index;
6657 	struct ctl_lun *lun;
6658 	int control_dev, page_len;
6659 	union ctl_modepage_info *modepage_info;
6660 	int retval;
6661 
6662 	pf = 0;
6663 	sp = 0;
6664 	page_len = 0;
6665 	len_used = 0;
6666 	len_left = 0;
6667 	retval = 0;
6668 	bd_len = 0;
6669 	page_index = NULL;
6670 
6671 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6672 
6673 	if (lun->be_lun->lun_type != T_DIRECT)
6674 		control_dev = 1;
6675 	else
6676 		control_dev = 0;
6677 
6678 	switch (ctsio->cdb[0]) {
6679 	case MODE_SELECT_6: {
6680 		struct scsi_mode_select_6 *cdb;
6681 
6682 		cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6683 
6684 		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6685 		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6686 
6687 		param_len = cdb->length;
6688 		header_size = sizeof(struct scsi_mode_header_6);
6689 		break;
6690 	}
6691 	case MODE_SELECT_10: {
6692 		struct scsi_mode_select_10 *cdb;
6693 
6694 		cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6695 
6696 		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6697 		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6698 
6699 		param_len = scsi_2btoul(cdb->length);
6700 		header_size = sizeof(struct scsi_mode_header_10);
6701 		break;
6702 	}
6703 	default:
6704 		ctl_set_invalid_opcode(ctsio);
6705 		ctl_done((union ctl_io *)ctsio);
6706 		return (CTL_RETVAL_COMPLETE);
6707 		break; /* NOTREACHED */
6708 	}
6709 
6710 	/*
6711 	 * From SPC-3:
6712 	 * "A parameter list length of zero indicates that the Data-Out Buffer
6713 	 * shall be empty. This condition shall not be considered as an error."
6714 	 */
6715 	if (param_len == 0) {
6716 		ctl_set_success(ctsio);
6717 		ctl_done((union ctl_io *)ctsio);
6718 		return (CTL_RETVAL_COMPLETE);
6719 	}
6720 
6721 	/*
6722 	 * Since we'll hit this the first time through, prior to
6723 	 * allocation, we don't need to free a data buffer here.
6724 	 */
6725 	if (param_len < header_size) {
6726 		ctl_set_param_len_error(ctsio);
6727 		ctl_done((union ctl_io *)ctsio);
6728 		return (CTL_RETVAL_COMPLETE);
6729 	}
6730 
6731 	/*
6732 	 * Allocate the data buffer and grab the user's data.  In theory,
6733 	 * we shouldn't have to sanity check the parameter list length here
6734 	 * because the maximum size is 64K.  We should be able to malloc
6735 	 * that much without too many problems.
6736 	 */
6737 	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6738 		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6739 		ctsio->kern_data_len = param_len;
6740 		ctsio->kern_total_len = param_len;
6741 		ctsio->kern_data_resid = 0;
6742 		ctsio->kern_rel_offset = 0;
6743 		ctsio->kern_sg_entries = 0;
6744 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6745 		ctsio->be_move_done = ctl_config_move_done;
6746 		ctl_datamove((union ctl_io *)ctsio);
6747 
6748 		return (CTL_RETVAL_COMPLETE);
6749 	}
6750 
6751 	switch (ctsio->cdb[0]) {
6752 	case MODE_SELECT_6: {
6753 		struct scsi_mode_header_6 *mh6;
6754 
6755 		mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6756 		bd_len = mh6->blk_desc_len;
6757 		break;
6758 	}
6759 	case MODE_SELECT_10: {
6760 		struct scsi_mode_header_10 *mh10;
6761 
6762 		mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6763 		bd_len = scsi_2btoul(mh10->blk_desc_len);
6764 		break;
6765 	}
6766 	default:
6767 		panic("Invalid CDB type %#x", ctsio->cdb[0]);
6768 		break;
6769 	}
6770 
6771 	if (param_len < (header_size + bd_len)) {
6772 		free(ctsio->kern_data_ptr, M_CTL);
6773 		ctl_set_param_len_error(ctsio);
6774 		ctl_done((union ctl_io *)ctsio);
6775 		return (CTL_RETVAL_COMPLETE);
6776 	}
6777 
6778 	/*
6779 	 * Set the IO_CONT flag, so that if this I/O gets passed to
6780 	 * ctl_config_write_done(), it'll get passed back to
6781 	 * ctl_do_mode_select() for further processing, or completion if
6782 	 * we're all done.
6783 	 */
6784 	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6785 	ctsio->io_cont = ctl_do_mode_select;
6786 
6787 	modepage_info = (union ctl_modepage_info *)
6788 		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6789 
6790 	memset(modepage_info, 0, sizeof(*modepage_info));
6791 
6792 	len_left = param_len - header_size - bd_len;
6793 	len_used = header_size + bd_len;
6794 
6795 	modepage_info->header.len_left = len_left;
6796 	modepage_info->header.len_used = len_used;
6797 
6798 	return (ctl_do_mode_select((union ctl_io *)ctsio));
6799 }
6800 
6801 int
6802 ctl_mode_sense(struct ctl_scsiio *ctsio)
6803 {
6804 	struct ctl_lun *lun;
6805 	int pc, page_code, dbd, llba, subpage;
6806 	int alloc_len, page_len, header_len, total_len;
6807 	struct scsi_mode_block_descr *block_desc;
6808 	struct ctl_page_index *page_index;
6809 	int control_dev;
6810 
6811 	dbd = 0;
6812 	llba = 0;
6813 	block_desc = NULL;
6814 	page_index = NULL;
6815 
6816 	CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6817 
6818 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6819 
6820 	if (lun->be_lun->lun_type != T_DIRECT)
6821 		control_dev = 1;
6822 	else
6823 		control_dev = 0;
6824 
6825 	if (lun->flags & CTL_LUN_PR_RESERVED) {
6826 		uint32_t residx;
6827 
6828 		/*
6829 		 * XXX KDM need a lock here.
6830 		 */
6831 		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
6832 		if ((lun->res_type == SPR_TYPE_EX_AC
6833 		  && residx != lun->pr_res_idx)
6834 		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
6835 		   || lun->res_type == SPR_TYPE_EX_AC_AR)
6836 		  && !lun->per_res[residx].registered)) {
6837 			ctl_set_reservation_conflict(ctsio);
6838 			ctl_done((union ctl_io *)ctsio);
6839 			return (CTL_RETVAL_COMPLETE);
6840 		}
6841 	}
6842 
6843 	switch (ctsio->cdb[0]) {
6844 	case MODE_SENSE_6: {
6845 		struct scsi_mode_sense_6 *cdb;
6846 
6847 		cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6848 
6849 		header_len = sizeof(struct scsi_mode_hdr_6);
6850 		if (cdb->byte2 & SMS_DBD)
6851 			dbd = 1;
6852 		else
6853 			header_len += sizeof(struct scsi_mode_block_descr);
6854 
6855 		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6856 		page_code = cdb->page & SMS_PAGE_CODE;
6857 		subpage = cdb->subpage;
6858 		alloc_len = cdb->length;
6859 		break;
6860 	}
6861 	case MODE_SENSE_10: {
6862 		struct scsi_mode_sense_10 *cdb;
6863 
6864 		cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6865 
6866 		header_len = sizeof(struct scsi_mode_hdr_10);
6867 
6868 		if (cdb->byte2 & SMS_DBD)
6869 			dbd = 1;
6870 		else
6871 			header_len += sizeof(struct scsi_mode_block_descr);
6872 		if (cdb->byte2 & SMS10_LLBAA)
6873 			llba = 1;
6874 		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6875 		page_code = cdb->page & SMS_PAGE_CODE;
6876 		subpage = cdb->subpage;
6877 		alloc_len = scsi_2btoul(cdb->length);
6878 		break;
6879 	}
6880 	default:
6881 		ctl_set_invalid_opcode(ctsio);
6882 		ctl_done((union ctl_io *)ctsio);
6883 		return (CTL_RETVAL_COMPLETE);
6884 		break; /* NOTREACHED */
6885 	}
6886 
6887 	/*
6888 	 * We have to make a first pass through to calculate the size of
6889 	 * the pages that match the user's query.  Then we allocate enough
6890 	 * memory to hold it, and actually copy the data into the buffer.
6891 	 */
6892 	switch (page_code) {
6893 	case SMS_ALL_PAGES_PAGE: {
6894 		int i;
6895 
6896 		page_len = 0;
6897 
6898 		/*
6899 		 * At the moment, values other than 0 and 0xff here are
6900 		 * reserved according to SPC-3.
6901 		 */
6902 		if ((subpage != SMS_SUBPAGE_PAGE_0)
6903 		 && (subpage != SMS_SUBPAGE_ALL)) {
6904 			ctl_set_invalid_field(ctsio,
6905 					      /*sks_valid*/ 1,
6906 					      /*command*/ 1,
6907 					      /*field*/ 3,
6908 					      /*bit_valid*/ 0,
6909 					      /*bit*/ 0);
6910 			ctl_done((union ctl_io *)ctsio);
6911 			return (CTL_RETVAL_COMPLETE);
6912 		}
6913 
6914 		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6915 			if ((control_dev != 0)
6916 			 && (lun->mode_pages.index[i].page_flags &
6917 			     CTL_PAGE_FLAG_DISK_ONLY))
6918 				continue;
6919 
6920 			/*
6921 			 * We don't use this subpage if the user didn't
6922 			 * request all subpages.
6923 			 */
6924 			if ((lun->mode_pages.index[i].subpage != 0)
6925 			 && (subpage == SMS_SUBPAGE_PAGE_0))
6926 				continue;
6927 
6928 #if 0
6929 			printf("found page %#x len %d\n",
6930 			       lun->mode_pages.index[i].page_code &
6931 			       SMPH_PC_MASK,
6932 			       lun->mode_pages.index[i].page_len);
6933 #endif
6934 			page_len += lun->mode_pages.index[i].page_len;
6935 		}
6936 		break;
6937 	}
6938 	default: {
6939 		int i;
6940 
6941 		page_len = 0;
6942 
6943 		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6944 			/* Look for the right page code */
6945 			if ((lun->mode_pages.index[i].page_code &
6946 			     SMPH_PC_MASK) != page_code)
6947 				continue;
6948 
6949 			/* Look for the right subpage or the subpage wildcard*/
6950 			if ((lun->mode_pages.index[i].subpage != subpage)
6951 			 && (subpage != SMS_SUBPAGE_ALL))
6952 				continue;
6953 
6954 			/* Make sure the page is supported for this dev type */
6955 			if ((control_dev != 0)
6956 			 && (lun->mode_pages.index[i].page_flags &
6957 			     CTL_PAGE_FLAG_DISK_ONLY))
6958 				continue;
6959 
6960 #if 0
6961 			printf("found page %#x len %d\n",
6962 			       lun->mode_pages.index[i].page_code &
6963 			       SMPH_PC_MASK,
6964 			       lun->mode_pages.index[i].page_len);
6965 #endif
6966 
6967 			page_len += lun->mode_pages.index[i].page_len;
6968 		}
6969 
6970 		if (page_len == 0) {
6971 			ctl_set_invalid_field(ctsio,
6972 					      /*sks_valid*/ 1,
6973 					      /*command*/ 1,
6974 					      /*field*/ 2,
6975 					      /*bit_valid*/ 1,
6976 					      /*bit*/ 5);
6977 			ctl_done((union ctl_io *)ctsio);
6978 			return (CTL_RETVAL_COMPLETE);
6979 		}
6980 		break;
6981 	}
6982 	}
6983 
6984 	total_len = header_len + page_len;
6985 #if 0
6986 	printf("header_len = %d, page_len = %d, total_len = %d\n",
6987 	       header_len, page_len, total_len);
6988 #endif
6989 
6990 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6991 	ctsio->kern_sg_entries = 0;
6992 	ctsio->kern_data_resid = 0;
6993 	ctsio->kern_rel_offset = 0;
6994 	if (total_len < alloc_len) {
6995 		ctsio->residual = alloc_len - total_len;
6996 		ctsio->kern_data_len = total_len;
6997 		ctsio->kern_total_len = total_len;
6998 	} else {
6999 		ctsio->residual = 0;
7000 		ctsio->kern_data_len = alloc_len;
7001 		ctsio->kern_total_len = alloc_len;
7002 	}
7003 
7004 	switch (ctsio->cdb[0]) {
7005 	case MODE_SENSE_6: {
7006 		struct scsi_mode_hdr_6 *header;
7007 
7008 		header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
7009 
7010 		header->datalen = ctl_min(total_len - 1, 254);
7011 
7012 		if (dbd)
7013 			header->block_descr_len = 0;
7014 		else
7015 			header->block_descr_len =
7016 				sizeof(struct scsi_mode_block_descr);
7017 		block_desc = (struct scsi_mode_block_descr *)&header[1];
7018 		break;
7019 	}
7020 	case MODE_SENSE_10: {
7021 		struct scsi_mode_hdr_10 *header;
7022 		int datalen;
7023 
7024 		header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
7025 
7026 		datalen = ctl_min(total_len - 2, 65533);
7027 		scsi_ulto2b(datalen, header->datalen);
7028 		if (dbd)
7029 			scsi_ulto2b(0, header->block_descr_len);
7030 		else
7031 			scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
7032 				    header->block_descr_len);
7033 		block_desc = (struct scsi_mode_block_descr *)&header[1];
7034 		break;
7035 	}
7036 	default:
7037 		panic("invalid CDB type %#x", ctsio->cdb[0]);
7038 		break; /* NOTREACHED */
7039 	}
7040 
7041 	/*
7042 	 * If we've got a disk, use its blocksize in the block
7043 	 * descriptor.  Otherwise, just set it to 0.
7044 	 */
7045 	if (dbd == 0) {
7046 		if (control_dev != 0)
7047 			scsi_ulto3b(lun->be_lun->blocksize,
7048 				    block_desc->block_len);
7049 		else
7050 			scsi_ulto3b(0, block_desc->block_len);
7051 	}
7052 
7053 	switch (page_code) {
7054 	case SMS_ALL_PAGES_PAGE: {
7055 		int i, data_used;
7056 
7057 		data_used = header_len;
7058 		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
7059 			struct ctl_page_index *page_index;
7060 
7061 			page_index = &lun->mode_pages.index[i];
7062 
7063 			if ((control_dev != 0)
7064 			 && (page_index->page_flags &
7065 			    CTL_PAGE_FLAG_DISK_ONLY))
7066 				continue;
7067 
7068 			/*
7069 			 * We don't use this subpage if the user didn't
7070 			 * request all subpages.  We already checked (above)
7071 			 * to make sure the user only specified a subpage
7072 			 * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
7073 			 */
7074 			if ((page_index->subpage != 0)
7075 			 && (subpage == SMS_SUBPAGE_PAGE_0))
7076 				continue;
7077 
7078 			/*
7079 			 * Call the handler, if it exists, to update the
7080 			 * page to the latest values.
7081 			 */
7082 			if (page_index->sense_handler != NULL)
7083 				page_index->sense_handler(ctsio, page_index,pc);
7084 
7085 			memcpy(ctsio->kern_data_ptr + data_used,
7086 			       page_index->page_data +
7087 			       (page_index->page_len * pc),
7088 			       page_index->page_len);
7089 			data_used += page_index->page_len;
7090 		}
7091 		break;
7092 	}
7093 	default: {
7094 		int i, data_used;
7095 
7096 		data_used = header_len;
7097 
7098 		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
7099 			struct ctl_page_index *page_index;
7100 
7101 			page_index = &lun->mode_pages.index[i];
7102 
7103 			/* Look for the right page code */
7104 			if ((page_index->page_code & SMPH_PC_MASK) != page_code)
7105 				continue;
7106 
7107 			/* Look for the right subpage or the subpage wildcard*/
7108 			if ((page_index->subpage != subpage)
7109 			 && (subpage != SMS_SUBPAGE_ALL))
7110 				continue;
7111 
7112 			/* Make sure the page is supported for this dev type */
7113 			if ((control_dev != 0)
7114 			 && (page_index->page_flags &
7115 			     CTL_PAGE_FLAG_DISK_ONLY))
7116 				continue;
7117 
7118 			/*
7119 			 * Call the handler, if it exists, to update the
7120 			 * page to the latest values.
7121 			 */
7122 			if (page_index->sense_handler != NULL)
7123 				page_index->sense_handler(ctsio, page_index,pc);
7124 
7125 			memcpy(ctsio->kern_data_ptr + data_used,
7126 			       page_index->page_data +
7127 			       (page_index->page_len * pc),
7128 			       page_index->page_len);
7129 			data_used += page_index->page_len;
7130 		}
7131 		break;
7132 	}
7133 	}
7134 
7135 	ctsio->scsi_status = SCSI_STATUS_OK;
7136 
7137 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7138 	ctsio->be_move_done = ctl_config_move_done;
7139 	ctl_datamove((union ctl_io *)ctsio);
7140 
7141 	return (CTL_RETVAL_COMPLETE);
7142 }
7143 
7144 int
7145 ctl_read_capacity(struct ctl_scsiio *ctsio)
7146 {
7147 	struct scsi_read_capacity *cdb;
7148 	struct scsi_read_capacity_data *data;
7149 	struct ctl_lun *lun;
7150 	uint32_t lba;
7151 
7152 	CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
7153 
7154 	cdb = (struct scsi_read_capacity *)ctsio->cdb;
7155 
7156 	lba = scsi_4btoul(cdb->addr);
7157 	if (((cdb->pmi & SRC_PMI) == 0)
7158 	 && (lba != 0)) {
7159 		ctl_set_invalid_field(/*ctsio*/ ctsio,
7160 				      /*sks_valid*/ 1,
7161 				      /*command*/ 1,
7162 				      /*field*/ 2,
7163 				      /*bit_valid*/ 0,
7164 				      /*bit*/ 0);
7165 		ctl_done((union ctl_io *)ctsio);
7166 		return (CTL_RETVAL_COMPLETE);
7167 	}
7168 
7169 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7170 
7171 	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7172 	data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
7173 	ctsio->residual = 0;
7174 	ctsio->kern_data_len = sizeof(*data);
7175 	ctsio->kern_total_len = sizeof(*data);
7176 	ctsio->kern_data_resid = 0;
7177 	ctsio->kern_rel_offset = 0;
7178 	ctsio->kern_sg_entries = 0;
7179 
7180 	/*
7181 	 * If the maximum LBA is greater than 0xfffffffe, the user must
7182 	 * issue a SERVICE ACTION IN (16) command, with the read capacity
7183 	 * serivce action set.
7184 	 */
7185 	if (lun->be_lun->maxlba > 0xfffffffe)
7186 		scsi_ulto4b(0xffffffff, data->addr);
7187 	else
7188 		scsi_ulto4b(lun->be_lun->maxlba, data->addr);
7189 
7190 	/*
7191 	 * XXX KDM this may not be 512 bytes...
7192 	 */
7193 	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7194 
7195 	ctsio->scsi_status = SCSI_STATUS_OK;
7196 
7197 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7198 	ctsio->be_move_done = ctl_config_move_done;
7199 	ctl_datamove((union ctl_io *)ctsio);
7200 
7201 	return (CTL_RETVAL_COMPLETE);
7202 }
7203 
7204 int
7205 ctl_read_capacity_16(struct ctl_scsiio *ctsio)
7206 {
7207 	struct scsi_read_capacity_16 *cdb;
7208 	struct scsi_read_capacity_data_long *data;
7209 	struct ctl_lun *lun;
7210 	uint64_t lba;
7211 	uint32_t alloc_len;
7212 
7213 	CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
7214 
7215 	cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
7216 
7217 	alloc_len = scsi_4btoul(cdb->alloc_len);
7218 	lba = scsi_8btou64(cdb->addr);
7219 
7220 	if ((cdb->reladr & SRC16_PMI)
7221 	 && (lba != 0)) {
7222 		ctl_set_invalid_field(/*ctsio*/ ctsio,
7223 				      /*sks_valid*/ 1,
7224 				      /*command*/ 1,
7225 				      /*field*/ 2,
7226 				      /*bit_valid*/ 0,
7227 				      /*bit*/ 0);
7228 		ctl_done((union ctl_io *)ctsio);
7229 		return (CTL_RETVAL_COMPLETE);
7230 	}
7231 
7232 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7233 
7234 	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7235 	data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
7236 
7237 	if (sizeof(*data) < alloc_len) {
7238 		ctsio->residual = alloc_len - sizeof(*data);
7239 		ctsio->kern_data_len = sizeof(*data);
7240 		ctsio->kern_total_len = sizeof(*data);
7241 	} else {
7242 		ctsio->residual = 0;
7243 		ctsio->kern_data_len = alloc_len;
7244 		ctsio->kern_total_len = alloc_len;
7245 	}
7246 	ctsio->kern_data_resid = 0;
7247 	ctsio->kern_rel_offset = 0;
7248 	ctsio->kern_sg_entries = 0;
7249 
7250 	scsi_u64to8b(lun->be_lun->maxlba, data->addr);
7251 	/* XXX KDM this may not be 512 bytes... */
7252 	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7253 	data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
7254 	scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
7255 	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
7256 		data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ;
7257 
7258 	ctsio->scsi_status = SCSI_STATUS_OK;
7259 
7260 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7261 	ctsio->be_move_done = ctl_config_move_done;
7262 	ctl_datamove((union ctl_io *)ctsio);
7263 
7264 	return (CTL_RETVAL_COMPLETE);
7265 }
7266 
7267 int
7268 ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7269 {
7270 	struct scsi_maintenance_in *cdb;
7271 	int retval;
7272 	int alloc_len, ext, total_len = 0, g, p, pc, pg;
7273 	int num_target_port_groups, num_target_ports, single;
7274 	struct ctl_lun *lun;
7275 	struct ctl_softc *softc;
7276 	struct ctl_port *port;
7277 	struct scsi_target_group_data *rtg_ptr;
7278 	struct scsi_target_group_data_extended *rtg_ext_ptr;
7279 	struct scsi_target_port_group_descriptor *tpg_desc;
7280 
7281 	CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7282 
7283 	cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7284 	softc = control_softc;
7285 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7286 
7287 	retval = CTL_RETVAL_COMPLETE;
7288 
7289 	switch (cdb->byte2 & STG_PDF_MASK) {
7290 	case STG_PDF_LENGTH:
7291 		ext = 0;
7292 		break;
7293 	case STG_PDF_EXTENDED:
7294 		ext = 1;
7295 		break;
7296 	default:
7297 		ctl_set_invalid_field(/*ctsio*/ ctsio,
7298 				      /*sks_valid*/ 1,
7299 				      /*command*/ 1,
7300 				      /*field*/ 2,
7301 				      /*bit_valid*/ 1,
7302 				      /*bit*/ 5);
7303 		ctl_done((union ctl_io *)ctsio);
7304 		return(retval);
7305 	}
7306 
7307 	single = ctl_is_single;
7308 	if (single)
7309 		num_target_port_groups = 1;
7310 	else
7311 		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
7312 	num_target_ports = 0;
7313 	mtx_lock(&softc->ctl_lock);
7314 	STAILQ_FOREACH(port, &softc->port_list, links) {
7315 		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7316 			continue;
7317 		if (ctl_map_lun_back(port->targ_port, lun->lun) >= CTL_MAX_LUNS)
7318 			continue;
7319 		num_target_ports++;
7320 	}
7321 	mtx_unlock(&softc->ctl_lock);
7322 
7323 	if (ext)
7324 		total_len = sizeof(struct scsi_target_group_data_extended);
7325 	else
7326 		total_len = sizeof(struct scsi_target_group_data);
7327 	total_len += sizeof(struct scsi_target_port_group_descriptor) *
7328 		num_target_port_groups +
7329 	    sizeof(struct scsi_target_port_descriptor) *
7330 		num_target_ports * num_target_port_groups;
7331 
7332 	alloc_len = scsi_4btoul(cdb->length);
7333 
7334 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7335 
7336 	ctsio->kern_sg_entries = 0;
7337 
7338 	if (total_len < alloc_len) {
7339 		ctsio->residual = alloc_len - total_len;
7340 		ctsio->kern_data_len = total_len;
7341 		ctsio->kern_total_len = total_len;
7342 	} else {
7343 		ctsio->residual = 0;
7344 		ctsio->kern_data_len = alloc_len;
7345 		ctsio->kern_total_len = alloc_len;
7346 	}
7347 	ctsio->kern_data_resid = 0;
7348 	ctsio->kern_rel_offset = 0;
7349 
7350 	if (ext) {
7351 		rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7352 		    ctsio->kern_data_ptr;
7353 		scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7354 		rtg_ext_ptr->format_type = 0x10;
7355 		rtg_ext_ptr->implicit_transition_time = 0;
7356 		tpg_desc = &rtg_ext_ptr->groups[0];
7357 	} else {
7358 		rtg_ptr = (struct scsi_target_group_data *)
7359 		    ctsio->kern_data_ptr;
7360 		scsi_ulto4b(total_len - 4, rtg_ptr->length);
7361 		tpg_desc = &rtg_ptr->groups[0];
7362 	}
7363 
7364 	pg = ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS;
7365 	mtx_lock(&softc->ctl_lock);
7366 	for (g = 0; g < num_target_port_groups; g++) {
7367 		if (g == pg)
7368 			tpg_desc->pref_state = TPG_PRIMARY |
7369 			    TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7370 		else
7371 			tpg_desc->pref_state =
7372 			    TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7373 		tpg_desc->support = TPG_AO_SUP;
7374 		if (!single)
7375 			tpg_desc->support |= TPG_AN_SUP;
7376 		scsi_ulto2b(g + 1, tpg_desc->target_port_group);
7377 		tpg_desc->status = TPG_IMPLICIT;
7378 		pc = 0;
7379 		STAILQ_FOREACH(port, &softc->port_list, links) {
7380 			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7381 				continue;
7382 			if (ctl_map_lun_back(port->targ_port, lun->lun) >=
7383 			    CTL_MAX_LUNS)
7384 				continue;
7385 			p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
7386 			scsi_ulto2b(p, tpg_desc->descriptors[pc].
7387 			    relative_target_port_identifier);
7388 			pc++;
7389 		}
7390 		tpg_desc->target_port_count = pc;
7391 		tpg_desc = (struct scsi_target_port_group_descriptor *)
7392 		    &tpg_desc->descriptors[pc];
7393 	}
7394 	mtx_unlock(&softc->ctl_lock);
7395 
7396 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7397 	ctsio->be_move_done = ctl_config_move_done;
7398 
7399 	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7400 			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7401 			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7402 			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7403 			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7404 
7405 	ctl_datamove((union ctl_io *)ctsio);
7406 	return(retval);
7407 }
7408 
7409 int
7410 ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7411 {
7412 	struct ctl_lun *lun;
7413 	struct scsi_report_supported_opcodes *cdb;
7414 	const struct ctl_cmd_entry *entry, *sentry;
7415 	struct scsi_report_supported_opcodes_all *all;
7416 	struct scsi_report_supported_opcodes_descr *descr;
7417 	struct scsi_report_supported_opcodes_one *one;
7418 	int retval;
7419 	int alloc_len, total_len;
7420 	int opcode, service_action, i, j, num;
7421 
7422 	CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7423 
7424 	cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7425 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7426 
7427 	retval = CTL_RETVAL_COMPLETE;
7428 
7429 	opcode = cdb->requested_opcode;
7430 	service_action = scsi_2btoul(cdb->requested_service_action);
7431 	switch (cdb->options & RSO_OPTIONS_MASK) {
7432 	case RSO_OPTIONS_ALL:
7433 		num = 0;
7434 		for (i = 0; i < 256; i++) {
7435 			entry = &ctl_cmd_table[i];
7436 			if (entry->flags & CTL_CMD_FLAG_SA5) {
7437 				for (j = 0; j < 32; j++) {
7438 					sentry = &((const struct ctl_cmd_entry *)
7439 					    entry->execute)[j];
7440 					if (ctl_cmd_applicable(
7441 					    lun->be_lun->lun_type, sentry))
7442 						num++;
7443 				}
7444 			} else {
7445 				if (ctl_cmd_applicable(lun->be_lun->lun_type,
7446 				    entry))
7447 					num++;
7448 			}
7449 		}
7450 		total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7451 		    num * sizeof(struct scsi_report_supported_opcodes_descr);
7452 		break;
7453 	case RSO_OPTIONS_OC:
7454 		if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7455 			ctl_set_invalid_field(/*ctsio*/ ctsio,
7456 					      /*sks_valid*/ 1,
7457 					      /*command*/ 1,
7458 					      /*field*/ 2,
7459 					      /*bit_valid*/ 1,
7460 					      /*bit*/ 2);
7461 			ctl_done((union ctl_io *)ctsio);
7462 			return (CTL_RETVAL_COMPLETE);
7463 		}
7464 		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7465 		break;
7466 	case RSO_OPTIONS_OC_SA:
7467 		if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7468 		    service_action >= 32) {
7469 			ctl_set_invalid_field(/*ctsio*/ ctsio,
7470 					      /*sks_valid*/ 1,
7471 					      /*command*/ 1,
7472 					      /*field*/ 2,
7473 					      /*bit_valid*/ 1,
7474 					      /*bit*/ 2);
7475 			ctl_done((union ctl_io *)ctsio);
7476 			return (CTL_RETVAL_COMPLETE);
7477 		}
7478 		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7479 		break;
7480 	default:
7481 		ctl_set_invalid_field(/*ctsio*/ ctsio,
7482 				      /*sks_valid*/ 1,
7483 				      /*command*/ 1,
7484 				      /*field*/ 2,
7485 				      /*bit_valid*/ 1,
7486 				      /*bit*/ 2);
7487 		ctl_done((union ctl_io *)ctsio);
7488 		return (CTL_RETVAL_COMPLETE);
7489 	}
7490 
7491 	alloc_len = scsi_4btoul(cdb->length);
7492 
7493 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7494 
7495 	ctsio->kern_sg_entries = 0;
7496 
7497 	if (total_len < alloc_len) {
7498 		ctsio->residual = alloc_len - total_len;
7499 		ctsio->kern_data_len = total_len;
7500 		ctsio->kern_total_len = total_len;
7501 	} else {
7502 		ctsio->residual = 0;
7503 		ctsio->kern_data_len = alloc_len;
7504 		ctsio->kern_total_len = alloc_len;
7505 	}
7506 	ctsio->kern_data_resid = 0;
7507 	ctsio->kern_rel_offset = 0;
7508 
7509 	switch (cdb->options & RSO_OPTIONS_MASK) {
7510 	case RSO_OPTIONS_ALL:
7511 		all = (struct scsi_report_supported_opcodes_all *)
7512 		    ctsio->kern_data_ptr;
7513 		num = 0;
7514 		for (i = 0; i < 256; i++) {
7515 			entry = &ctl_cmd_table[i];
7516 			if (entry->flags & CTL_CMD_FLAG_SA5) {
7517 				for (j = 0; j < 32; j++) {
7518 					sentry = &((const struct ctl_cmd_entry *)
7519 					    entry->execute)[j];
7520 					if (!ctl_cmd_applicable(
7521 					    lun->be_lun->lun_type, sentry))
7522 						continue;
7523 					descr = &all->descr[num++];
7524 					descr->opcode = i;
7525 					scsi_ulto2b(j, descr->service_action);
7526 					descr->flags = RSO_SERVACTV;
7527 					scsi_ulto2b(sentry->length,
7528 					    descr->cdb_length);
7529 				}
7530 			} else {
7531 				if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7532 				    entry))
7533 					continue;
7534 				descr = &all->descr[num++];
7535 				descr->opcode = i;
7536 				scsi_ulto2b(0, descr->service_action);
7537 				descr->flags = 0;
7538 				scsi_ulto2b(entry->length, descr->cdb_length);
7539 			}
7540 		}
7541 		scsi_ulto4b(
7542 		    num * sizeof(struct scsi_report_supported_opcodes_descr),
7543 		    all->length);
7544 		break;
7545 	case RSO_OPTIONS_OC:
7546 		one = (struct scsi_report_supported_opcodes_one *)
7547 		    ctsio->kern_data_ptr;
7548 		entry = &ctl_cmd_table[opcode];
7549 		goto fill_one;
7550 	case RSO_OPTIONS_OC_SA:
7551 		one = (struct scsi_report_supported_opcodes_one *)
7552 		    ctsio->kern_data_ptr;
7553 		entry = &ctl_cmd_table[opcode];
7554 		entry = &((const struct ctl_cmd_entry *)
7555 		    entry->execute)[service_action];
7556 fill_one:
7557 		if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7558 			one->support = 3;
7559 			scsi_ulto2b(entry->length, one->cdb_length);
7560 			one->cdb_usage[0] = opcode;
7561 			memcpy(&one->cdb_usage[1], entry->usage,
7562 			    entry->length - 1);
7563 		} else
7564 			one->support = 1;
7565 		break;
7566 	}
7567 
7568 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7569 	ctsio->be_move_done = ctl_config_move_done;
7570 
7571 	ctl_datamove((union ctl_io *)ctsio);
7572 	return(retval);
7573 }
7574 
7575 int
7576 ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7577 {
7578 	struct ctl_lun *lun;
7579 	struct scsi_report_supported_tmf *cdb;
7580 	struct scsi_report_supported_tmf_data *data;
7581 	int retval;
7582 	int alloc_len, total_len;
7583 
7584 	CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7585 
7586 	cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7587 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7588 
7589 	retval = CTL_RETVAL_COMPLETE;
7590 
7591 	total_len = sizeof(struct scsi_report_supported_tmf_data);
7592 	alloc_len = scsi_4btoul(cdb->length);
7593 
7594 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7595 
7596 	ctsio->kern_sg_entries = 0;
7597 
7598 	if (total_len < alloc_len) {
7599 		ctsio->residual = alloc_len - total_len;
7600 		ctsio->kern_data_len = total_len;
7601 		ctsio->kern_total_len = total_len;
7602 	} else {
7603 		ctsio->residual = 0;
7604 		ctsio->kern_data_len = alloc_len;
7605 		ctsio->kern_total_len = alloc_len;
7606 	}
7607 	ctsio->kern_data_resid = 0;
7608 	ctsio->kern_rel_offset = 0;
7609 
7610 	data = (struct scsi_report_supported_tmf_data *)ctsio->kern_data_ptr;
7611 	data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_TRS;
7612 	data->byte2 |= RST_ITNRS;
7613 
7614 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7615 	ctsio->be_move_done = ctl_config_move_done;
7616 
7617 	ctl_datamove((union ctl_io *)ctsio);
7618 	return (retval);
7619 }
7620 
7621 int
7622 ctl_report_timestamp(struct ctl_scsiio *ctsio)
7623 {
7624 	struct ctl_lun *lun;
7625 	struct scsi_report_timestamp *cdb;
7626 	struct scsi_report_timestamp_data *data;
7627 	struct timeval tv;
7628 	int64_t timestamp;
7629 	int retval;
7630 	int alloc_len, total_len;
7631 
7632 	CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7633 
7634 	cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7635 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7636 
7637 	retval = CTL_RETVAL_COMPLETE;
7638 
7639 	total_len = sizeof(struct scsi_report_timestamp_data);
7640 	alloc_len = scsi_4btoul(cdb->length);
7641 
7642 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7643 
7644 	ctsio->kern_sg_entries = 0;
7645 
7646 	if (total_len < alloc_len) {
7647 		ctsio->residual = alloc_len - total_len;
7648 		ctsio->kern_data_len = total_len;
7649 		ctsio->kern_total_len = total_len;
7650 	} else {
7651 		ctsio->residual = 0;
7652 		ctsio->kern_data_len = alloc_len;
7653 		ctsio->kern_total_len = alloc_len;
7654 	}
7655 	ctsio->kern_data_resid = 0;
7656 	ctsio->kern_rel_offset = 0;
7657 
7658 	data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7659 	scsi_ulto2b(sizeof(*data) - 2, data->length);
7660 	data->origin = RTS_ORIG_OUTSIDE;
7661 	getmicrotime(&tv);
7662 	timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7663 	scsi_ulto4b(timestamp >> 16, data->timestamp);
7664 	scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7665 
7666 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7667 	ctsio->be_move_done = ctl_config_move_done;
7668 
7669 	ctl_datamove((union ctl_io *)ctsio);
7670 	return (retval);
7671 }
7672 
7673 int
7674 ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7675 {
7676 	struct scsi_per_res_in *cdb;
7677 	int alloc_len, total_len = 0;
7678 	/* struct scsi_per_res_in_rsrv in_data; */
7679 	struct ctl_lun *lun;
7680 	struct ctl_softc *softc;
7681 
7682 	CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7683 
7684 	softc = control_softc;
7685 
7686 	cdb = (struct scsi_per_res_in *)ctsio->cdb;
7687 
7688 	alloc_len = scsi_2btoul(cdb->length);
7689 
7690 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7691 
7692 retry:
7693 	mtx_lock(&lun->lun_lock);
7694 	switch (cdb->action) {
7695 	case SPRI_RK: /* read keys */
7696 		total_len = sizeof(struct scsi_per_res_in_keys) +
7697 			lun->pr_key_count *
7698 			sizeof(struct scsi_per_res_key);
7699 		break;
7700 	case SPRI_RR: /* read reservation */
7701 		if (lun->flags & CTL_LUN_PR_RESERVED)
7702 			total_len = sizeof(struct scsi_per_res_in_rsrv);
7703 		else
7704 			total_len = sizeof(struct scsi_per_res_in_header);
7705 		break;
7706 	case SPRI_RC: /* report capabilities */
7707 		total_len = sizeof(struct scsi_per_res_cap);
7708 		break;
7709 	case SPRI_RS: /* read full status */
7710 		total_len = sizeof(struct scsi_per_res_in_header) +
7711 		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7712 		    lun->pr_key_count;
7713 		break;
7714 	default:
7715 		panic("Invalid PR type %x", cdb->action);
7716 	}
7717 	mtx_unlock(&lun->lun_lock);
7718 
7719 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7720 
7721 	if (total_len < alloc_len) {
7722 		ctsio->residual = alloc_len - total_len;
7723 		ctsio->kern_data_len = total_len;
7724 		ctsio->kern_total_len = total_len;
7725 	} else {
7726 		ctsio->residual = 0;
7727 		ctsio->kern_data_len = alloc_len;
7728 		ctsio->kern_total_len = alloc_len;
7729 	}
7730 
7731 	ctsio->kern_data_resid = 0;
7732 	ctsio->kern_rel_offset = 0;
7733 	ctsio->kern_sg_entries = 0;
7734 
7735 	mtx_lock(&lun->lun_lock);
7736 	switch (cdb->action) {
7737 	case SPRI_RK: { // read keys
7738         struct scsi_per_res_in_keys *res_keys;
7739 		int i, key_count;
7740 
7741 		res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7742 
7743 		/*
7744 		 * We had to drop the lock to allocate our buffer, which
7745 		 * leaves time for someone to come in with another
7746 		 * persistent reservation.  (That is unlikely, though,
7747 		 * since this should be the only persistent reservation
7748 		 * command active right now.)
7749 		 */
7750 		if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7751 		    (lun->pr_key_count *
7752 		     sizeof(struct scsi_per_res_key)))){
7753 			mtx_unlock(&lun->lun_lock);
7754 			free(ctsio->kern_data_ptr, M_CTL);
7755 			printf("%s: reservation length changed, retrying\n",
7756 			       __func__);
7757 			goto retry;
7758 		}
7759 
7760 		scsi_ulto4b(lun->PRGeneration, res_keys->header.generation);
7761 
7762 		scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7763 			     lun->pr_key_count, res_keys->header.length);
7764 
7765 		for (i = 0, key_count = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7766 			if (!lun->per_res[i].registered)
7767 				continue;
7768 
7769 			/*
7770 			 * We used lun->pr_key_count to calculate the
7771 			 * size to allocate.  If it turns out the number of
7772 			 * initiators with the registered flag set is
7773 			 * larger than that (i.e. they haven't been kept in
7774 			 * sync), we've got a problem.
7775 			 */
7776 			if (key_count >= lun->pr_key_count) {
7777 #ifdef NEEDTOPORT
7778 				csevent_log(CSC_CTL | CSC_SHELF_SW |
7779 					    CTL_PR_ERROR,
7780 					    csevent_LogType_Fault,
7781 					    csevent_AlertLevel_Yellow,
7782 					    csevent_FRU_ShelfController,
7783 					    csevent_FRU_Firmware,
7784 				        csevent_FRU_Unknown,
7785 					    "registered keys %d >= key "
7786 					    "count %d", key_count,
7787 					    lun->pr_key_count);
7788 #endif
7789 				key_count++;
7790 				continue;
7791 			}
7792 			memcpy(res_keys->keys[key_count].key,
7793 			       lun->per_res[i].res_key.key,
7794 			       ctl_min(sizeof(res_keys->keys[key_count].key),
7795 			       sizeof(lun->per_res[i].res_key)));
7796 			key_count++;
7797 		}
7798 		break;
7799 	}
7800 	case SPRI_RR: { // read reservation
7801 		struct scsi_per_res_in_rsrv *res;
7802 		int tmp_len, header_only;
7803 
7804 		res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7805 
7806 		scsi_ulto4b(lun->PRGeneration, res->header.generation);
7807 
7808 		if (lun->flags & CTL_LUN_PR_RESERVED)
7809 		{
7810 			tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7811 			scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7812 				    res->header.length);
7813 			header_only = 0;
7814 		} else {
7815 			tmp_len = sizeof(struct scsi_per_res_in_header);
7816 			scsi_ulto4b(0, res->header.length);
7817 			header_only = 1;
7818 		}
7819 
7820 		/*
7821 		 * We had to drop the lock to allocate our buffer, which
7822 		 * leaves time for someone to come in with another
7823 		 * persistent reservation.  (That is unlikely, though,
7824 		 * since this should be the only persistent reservation
7825 		 * command active right now.)
7826 		 */
7827 		if (tmp_len != total_len) {
7828 			mtx_unlock(&lun->lun_lock);
7829 			free(ctsio->kern_data_ptr, M_CTL);
7830 			printf("%s: reservation status changed, retrying\n",
7831 			       __func__);
7832 			goto retry;
7833 		}
7834 
7835 		/*
7836 		 * No reservation held, so we're done.
7837 		 */
7838 		if (header_only != 0)
7839 			break;
7840 
7841 		/*
7842 		 * If the registration is an All Registrants type, the key
7843 		 * is 0, since it doesn't really matter.
7844 		 */
7845 		if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7846 			memcpy(res->data.reservation,
7847 			       &lun->per_res[lun->pr_res_idx].res_key,
7848 			       sizeof(struct scsi_per_res_key));
7849 		}
7850 		res->data.scopetype = lun->res_type;
7851 		break;
7852 	}
7853 	case SPRI_RC:     //report capabilities
7854 	{
7855 		struct scsi_per_res_cap *res_cap;
7856 		uint16_t type_mask;
7857 
7858 		res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7859 		scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7860 		res_cap->flags2 |= SPRI_TMV | SPRI_ALLOW_3;
7861 		type_mask = SPRI_TM_WR_EX_AR |
7862 			    SPRI_TM_EX_AC_RO |
7863 			    SPRI_TM_WR_EX_RO |
7864 			    SPRI_TM_EX_AC |
7865 			    SPRI_TM_WR_EX |
7866 			    SPRI_TM_EX_AC_AR;
7867 		scsi_ulto2b(type_mask, res_cap->type_mask);
7868 		break;
7869 	}
7870 	case SPRI_RS: { // read full status
7871 		struct scsi_per_res_in_full *res_status;
7872 		struct scsi_per_res_in_full_desc *res_desc;
7873 		struct ctl_port *port;
7874 		int i, len;
7875 
7876 		res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7877 
7878 		/*
7879 		 * We had to drop the lock to allocate our buffer, which
7880 		 * leaves time for someone to come in with another
7881 		 * persistent reservation.  (That is unlikely, though,
7882 		 * since this should be the only persistent reservation
7883 		 * command active right now.)
7884 		 */
7885 		if (total_len < (sizeof(struct scsi_per_res_in_header) +
7886 		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7887 		     lun->pr_key_count)){
7888 			mtx_unlock(&lun->lun_lock);
7889 			free(ctsio->kern_data_ptr, M_CTL);
7890 			printf("%s: reservation length changed, retrying\n",
7891 			       __func__);
7892 			goto retry;
7893 		}
7894 
7895 		scsi_ulto4b(lun->PRGeneration, res_status->header.generation);
7896 
7897 		res_desc = &res_status->desc[0];
7898 		for (i = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7899 			if (!lun->per_res[i].registered)
7900 				continue;
7901 
7902 			memcpy(&res_desc->res_key, &lun->per_res[i].res_key.key,
7903 			    sizeof(res_desc->res_key));
7904 			if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7905 			    (lun->pr_res_idx == i ||
7906 			     lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7907 				res_desc->flags = SPRI_FULL_R_HOLDER;
7908 				res_desc->scopetype = lun->res_type;
7909 			}
7910 			scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7911 			    res_desc->rel_trgt_port_id);
7912 			len = 0;
7913 			port = softc->ctl_ports[i / CTL_MAX_INIT_PER_PORT];
7914 			if (port != NULL)
7915 				len = ctl_create_iid(port,
7916 				    i % CTL_MAX_INIT_PER_PORT,
7917 				    res_desc->transport_id);
7918 			scsi_ulto4b(len, res_desc->additional_length);
7919 			res_desc = (struct scsi_per_res_in_full_desc *)
7920 			    &res_desc->transport_id[len];
7921 		}
7922 		scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7923 		    res_status->header.length);
7924 		break;
7925 	}
7926 	default:
7927 		/*
7928 		 * This is a bug, because we just checked for this above,
7929 		 * and should have returned an error.
7930 		 */
7931 		panic("Invalid PR type %x", cdb->action);
7932 		break; /* NOTREACHED */
7933 	}
7934 	mtx_unlock(&lun->lun_lock);
7935 
7936 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7937 	ctsio->be_move_done = ctl_config_move_done;
7938 
7939 	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7940 			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7941 			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7942 			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7943 			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7944 
7945 	ctl_datamove((union ctl_io *)ctsio);
7946 
7947 	return (CTL_RETVAL_COMPLETE);
7948 }
7949 
7950 /*
7951  * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7952  * it should return.
7953  */
7954 static int
7955 ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7956 		uint64_t sa_res_key, uint8_t type, uint32_t residx,
7957 		struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7958 		struct scsi_per_res_out_parms* param)
7959 {
7960 	union ctl_ha_msg persis_io;
7961 	int retval, i;
7962 	int isc_retval;
7963 
7964 	retval = 0;
7965 
7966 	mtx_lock(&lun->lun_lock);
7967 	if (sa_res_key == 0) {
7968 		if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7969 			/* validate scope and type */
7970 			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7971 			     SPR_LU_SCOPE) {
7972 				mtx_unlock(&lun->lun_lock);
7973 				ctl_set_invalid_field(/*ctsio*/ ctsio,
7974 						      /*sks_valid*/ 1,
7975 						      /*command*/ 1,
7976 						      /*field*/ 2,
7977 						      /*bit_valid*/ 1,
7978 						      /*bit*/ 4);
7979 				ctl_done((union ctl_io *)ctsio);
7980 				return (1);
7981 			}
7982 
7983 		        if (type>8 || type==2 || type==4 || type==0) {
7984 				mtx_unlock(&lun->lun_lock);
7985 				ctl_set_invalid_field(/*ctsio*/ ctsio,
7986        	           				      /*sks_valid*/ 1,
7987 						      /*command*/ 1,
7988 						      /*field*/ 2,
7989 						      /*bit_valid*/ 1,
7990 						      /*bit*/ 0);
7991 				ctl_done((union ctl_io *)ctsio);
7992 				return (1);
7993 		        }
7994 
7995 			/* temporarily unregister this nexus */
7996 			lun->per_res[residx].registered = 0;
7997 
7998 			/*
7999 			 * Unregister everybody else and build UA for
8000 			 * them
8001 			 */
8002 			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8003 				if (lun->per_res[i].registered == 0)
8004 					continue;
8005 
8006 				if (!persis_offset
8007 				 && i <CTL_MAX_INITIATORS)
8008 					lun->pending_ua[i] |=
8009 						CTL_UA_REG_PREEMPT;
8010 				else if (persis_offset
8011 				      && i >= persis_offset)
8012 					lun->pending_ua[i-persis_offset] |=
8013 						CTL_UA_REG_PREEMPT;
8014 				lun->per_res[i].registered = 0;
8015 				memset(&lun->per_res[i].res_key, 0,
8016 				       sizeof(struct scsi_per_res_key));
8017 			}
8018 			lun->per_res[residx].registered = 1;
8019 			lun->pr_key_count = 1;
8020 			lun->res_type = type;
8021 			if (lun->res_type != SPR_TYPE_WR_EX_AR
8022 			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8023 				lun->pr_res_idx = residx;
8024 
8025 			/* send msg to other side */
8026 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8027 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8028 			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8029 			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8030 			persis_io.pr.pr_info.res_type = type;
8031 			memcpy(persis_io.pr.pr_info.sa_res_key,
8032 			       param->serv_act_res_key,
8033 			       sizeof(param->serv_act_res_key));
8034 			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8035 			     &persis_io, sizeof(persis_io), 0)) >
8036 			     CTL_HA_STATUS_SUCCESS) {
8037 				printf("CTL:Persis Out error returned "
8038 				       "from ctl_ha_msg_send %d\n",
8039 				       isc_retval);
8040 			}
8041 		} else {
8042 			/* not all registrants */
8043 			mtx_unlock(&lun->lun_lock);
8044 			free(ctsio->kern_data_ptr, M_CTL);
8045 			ctl_set_invalid_field(ctsio,
8046 					      /*sks_valid*/ 1,
8047 					      /*command*/ 0,
8048 					      /*field*/ 8,
8049 					      /*bit_valid*/ 0,
8050 					      /*bit*/ 0);
8051 			ctl_done((union ctl_io *)ctsio);
8052 			return (1);
8053 		}
8054 	} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8055 		|| !(lun->flags & CTL_LUN_PR_RESERVED)) {
8056 		int found = 0;
8057 
8058 		if (res_key == sa_res_key) {
8059 			/* special case */
8060 			/*
8061 			 * The spec implies this is not good but doesn't
8062 			 * say what to do. There are two choices either
8063 			 * generate a res conflict or check condition
8064 			 * with illegal field in parameter data. Since
8065 			 * that is what is done when the sa_res_key is
8066 			 * zero I'll take that approach since this has
8067 			 * to do with the sa_res_key.
8068 			 */
8069 			mtx_unlock(&lun->lun_lock);
8070 			free(ctsio->kern_data_ptr, M_CTL);
8071 			ctl_set_invalid_field(ctsio,
8072 					      /*sks_valid*/ 1,
8073 					      /*command*/ 0,
8074 					      /*field*/ 8,
8075 					      /*bit_valid*/ 0,
8076 					      /*bit*/ 0);
8077 			ctl_done((union ctl_io *)ctsio);
8078 			return (1);
8079 		}
8080 
8081 		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8082 			if (lun->per_res[i].registered
8083 			 && memcmp(param->serv_act_res_key,
8084 			    lun->per_res[i].res_key.key,
8085 			    sizeof(struct scsi_per_res_key)) != 0)
8086 				continue;
8087 
8088 			found = 1;
8089 			lun->per_res[i].registered = 0;
8090 			memset(&lun->per_res[i].res_key, 0,
8091 			       sizeof(struct scsi_per_res_key));
8092 			lun->pr_key_count--;
8093 
8094 			if (!persis_offset && i < CTL_MAX_INITIATORS)
8095 				lun->pending_ua[i] |= CTL_UA_REG_PREEMPT;
8096 			else if (persis_offset && i >= persis_offset)
8097 				lun->pending_ua[i-persis_offset] |=
8098 					CTL_UA_REG_PREEMPT;
8099 		}
8100 		if (!found) {
8101 			mtx_unlock(&lun->lun_lock);
8102 			free(ctsio->kern_data_ptr, M_CTL);
8103 			ctl_set_reservation_conflict(ctsio);
8104 			ctl_done((union ctl_io *)ctsio);
8105 			return (CTL_RETVAL_COMPLETE);
8106 		}
8107 		/* send msg to other side */
8108 		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8109 		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8110 		persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8111 		persis_io.pr.pr_info.residx = lun->pr_res_idx;
8112 		persis_io.pr.pr_info.res_type = type;
8113 		memcpy(persis_io.pr.pr_info.sa_res_key,
8114 		       param->serv_act_res_key,
8115 		       sizeof(param->serv_act_res_key));
8116 		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8117 		     &persis_io, sizeof(persis_io), 0)) >
8118 		     CTL_HA_STATUS_SUCCESS) {
8119 			printf("CTL:Persis Out error returned from "
8120 			       "ctl_ha_msg_send %d\n", isc_retval);
8121 		}
8122 	} else {
8123 		/* Reserved but not all registrants */
8124 		/* sa_res_key is res holder */
8125 		if (memcmp(param->serv_act_res_key,
8126                    lun->per_res[lun->pr_res_idx].res_key.key,
8127                    sizeof(struct scsi_per_res_key)) == 0) {
8128 			/* validate scope and type */
8129 			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
8130 			     SPR_LU_SCOPE) {
8131 				mtx_unlock(&lun->lun_lock);
8132 				ctl_set_invalid_field(/*ctsio*/ ctsio,
8133 						      /*sks_valid*/ 1,
8134 						      /*command*/ 1,
8135 						      /*field*/ 2,
8136 						      /*bit_valid*/ 1,
8137 						      /*bit*/ 4);
8138 				ctl_done((union ctl_io *)ctsio);
8139 				return (1);
8140 			}
8141 
8142 			if (type>8 || type==2 || type==4 || type==0) {
8143 				mtx_unlock(&lun->lun_lock);
8144 				ctl_set_invalid_field(/*ctsio*/ ctsio,
8145 						      /*sks_valid*/ 1,
8146 						      /*command*/ 1,
8147 						      /*field*/ 2,
8148 						      /*bit_valid*/ 1,
8149 						      /*bit*/ 0);
8150 				ctl_done((union ctl_io *)ctsio);
8151 				return (1);
8152 			}
8153 
8154 			/*
8155 			 * Do the following:
8156 			 * if sa_res_key != res_key remove all
8157 			 * registrants w/sa_res_key and generate UA
8158 			 * for these registrants(Registrations
8159 			 * Preempted) if it wasn't an exclusive
8160 			 * reservation generate UA(Reservations
8161 			 * Preempted) for all other registered nexuses
8162 			 * if the type has changed. Establish the new
8163 			 * reservation and holder. If res_key and
8164 			 * sa_res_key are the same do the above
8165 			 * except don't unregister the res holder.
8166 			 */
8167 
8168 			/*
8169 			 * Temporarily unregister so it won't get
8170 			 * removed or UA generated
8171 			 */
8172 			lun->per_res[residx].registered = 0;
8173 			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8174 				if (lun->per_res[i].registered == 0)
8175 					continue;
8176 
8177 				if (memcmp(param->serv_act_res_key,
8178 				    lun->per_res[i].res_key.key,
8179 				    sizeof(struct scsi_per_res_key)) == 0) {
8180 					lun->per_res[i].registered = 0;
8181 					memset(&lun->per_res[i].res_key,
8182 					       0,
8183 					       sizeof(struct scsi_per_res_key));
8184 					lun->pr_key_count--;
8185 
8186 					if (!persis_offset
8187 					 && i < CTL_MAX_INITIATORS)
8188 						lun->pending_ua[i] |=
8189 							CTL_UA_REG_PREEMPT;
8190 					else if (persis_offset
8191 					      && i >= persis_offset)
8192 						lun->pending_ua[i-persis_offset] |=
8193 						  CTL_UA_REG_PREEMPT;
8194 				} else if (type != lun->res_type
8195 					&& (lun->res_type == SPR_TYPE_WR_EX_RO
8196 					 || lun->res_type ==SPR_TYPE_EX_AC_RO)){
8197 						if (!persis_offset
8198 						 && i < CTL_MAX_INITIATORS)
8199 							lun->pending_ua[i] |=
8200 							CTL_UA_RES_RELEASE;
8201 						else if (persis_offset
8202 						      && i >= persis_offset)
8203 							lun->pending_ua[
8204 							i-persis_offset] |=
8205 							CTL_UA_RES_RELEASE;
8206 				}
8207 			}
8208 			lun->per_res[residx].registered = 1;
8209 			lun->res_type = type;
8210 			if (lun->res_type != SPR_TYPE_WR_EX_AR
8211 			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8212 				lun->pr_res_idx = residx;
8213 			else
8214 				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8215 
8216 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8217 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8218 			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8219 			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8220 			persis_io.pr.pr_info.res_type = type;
8221 			memcpy(persis_io.pr.pr_info.sa_res_key,
8222 			       param->serv_act_res_key,
8223 			       sizeof(param->serv_act_res_key));
8224 			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8225 			     &persis_io, sizeof(persis_io), 0)) >
8226 			     CTL_HA_STATUS_SUCCESS) {
8227 				printf("CTL:Persis Out error returned "
8228 				       "from ctl_ha_msg_send %d\n",
8229 				       isc_retval);
8230 			}
8231 		} else {
8232 			/*
8233 			 * sa_res_key is not the res holder just
8234 			 * remove registrants
8235 			 */
8236 			int found=0;
8237 
8238 			for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8239 				if (memcmp(param->serv_act_res_key,
8240 				    lun->per_res[i].res_key.key,
8241 				    sizeof(struct scsi_per_res_key)) != 0)
8242 					continue;
8243 
8244 				found = 1;
8245 				lun->per_res[i].registered = 0;
8246 				memset(&lun->per_res[i].res_key, 0,
8247 				       sizeof(struct scsi_per_res_key));
8248 				lun->pr_key_count--;
8249 
8250 				if (!persis_offset
8251 				 && i < CTL_MAX_INITIATORS)
8252 					lun->pending_ua[i] |=
8253 						CTL_UA_REG_PREEMPT;
8254 				else if (persis_offset
8255 				      && i >= persis_offset)
8256 					lun->pending_ua[i-persis_offset] |=
8257 						CTL_UA_REG_PREEMPT;
8258 			}
8259 
8260 			if (!found) {
8261 				mtx_unlock(&lun->lun_lock);
8262 				free(ctsio->kern_data_ptr, M_CTL);
8263 				ctl_set_reservation_conflict(ctsio);
8264 				ctl_done((union ctl_io *)ctsio);
8265 		        	return (1);
8266 			}
8267 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8268 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8269 			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8270 			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8271 			persis_io.pr.pr_info.res_type = type;
8272 			memcpy(persis_io.pr.pr_info.sa_res_key,
8273 			       param->serv_act_res_key,
8274 			       sizeof(param->serv_act_res_key));
8275 			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8276 			     &persis_io, sizeof(persis_io), 0)) >
8277 			     CTL_HA_STATUS_SUCCESS) {
8278 				printf("CTL:Persis Out error returned "
8279 				       "from ctl_ha_msg_send %d\n",
8280 				isc_retval);
8281 			}
8282 		}
8283 	}
8284 
8285 	lun->PRGeneration++;
8286 	mtx_unlock(&lun->lun_lock);
8287 
8288 	return (retval);
8289 }
8290 
8291 static void
8292 ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
8293 {
8294 	int i;
8295 
8296 	if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8297 	 || lun->pr_res_idx == CTL_PR_NO_RESERVATION
8298 	 || memcmp(&lun->per_res[lun->pr_res_idx].res_key,
8299 		   msg->pr.pr_info.sa_res_key,
8300 		   sizeof(struct scsi_per_res_key)) != 0) {
8301 		uint64_t sa_res_key;
8302 		sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
8303 
8304 		if (sa_res_key == 0) {
8305 			/* temporarily unregister this nexus */
8306 			lun->per_res[msg->pr.pr_info.residx].registered = 0;
8307 
8308 			/*
8309 			 * Unregister everybody else and build UA for
8310 			 * them
8311 			 */
8312 			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8313 				if (lun->per_res[i].registered == 0)
8314 					continue;
8315 
8316 				if (!persis_offset
8317 				 && i < CTL_MAX_INITIATORS)
8318 					lun->pending_ua[i] |=
8319 						CTL_UA_REG_PREEMPT;
8320 				else if (persis_offset && i >= persis_offset)
8321 					lun->pending_ua[i - persis_offset] |=
8322 						CTL_UA_REG_PREEMPT;
8323 				lun->per_res[i].registered = 0;
8324 				memset(&lun->per_res[i].res_key, 0,
8325 				       sizeof(struct scsi_per_res_key));
8326 			}
8327 
8328 			lun->per_res[msg->pr.pr_info.residx].registered = 1;
8329 			lun->pr_key_count = 1;
8330 			lun->res_type = msg->pr.pr_info.res_type;
8331 			if (lun->res_type != SPR_TYPE_WR_EX_AR
8332 			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8333 				lun->pr_res_idx = msg->pr.pr_info.residx;
8334 		} else {
8335 		        for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8336 				if (memcmp(msg->pr.pr_info.sa_res_key,
8337 		                   lun->per_res[i].res_key.key,
8338 		                   sizeof(struct scsi_per_res_key)) != 0)
8339 					continue;
8340 
8341 				lun->per_res[i].registered = 0;
8342 				memset(&lun->per_res[i].res_key, 0,
8343 				       sizeof(struct scsi_per_res_key));
8344 				lun->pr_key_count--;
8345 
8346 				if (!persis_offset
8347 				 && i < persis_offset)
8348 					lun->pending_ua[i] |=
8349 						CTL_UA_REG_PREEMPT;
8350 				else if (persis_offset
8351 				      && i >= persis_offset)
8352 					lun->pending_ua[i - persis_offset] |=
8353 						CTL_UA_REG_PREEMPT;
8354 			}
8355 		}
8356 	} else {
8357 		/*
8358 		 * Temporarily unregister so it won't get removed
8359 		 * or UA generated
8360 		 */
8361 		lun->per_res[msg->pr.pr_info.residx].registered = 0;
8362 		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8363 			if (lun->per_res[i].registered == 0)
8364 				continue;
8365 
8366 			if (memcmp(msg->pr.pr_info.sa_res_key,
8367 	                   lun->per_res[i].res_key.key,
8368 	                   sizeof(struct scsi_per_res_key)) == 0) {
8369 				lun->per_res[i].registered = 0;
8370 				memset(&lun->per_res[i].res_key, 0,
8371 				       sizeof(struct scsi_per_res_key));
8372 				lun->pr_key_count--;
8373 				if (!persis_offset
8374 				 && i < CTL_MAX_INITIATORS)
8375 					lun->pending_ua[i] |=
8376 						CTL_UA_REG_PREEMPT;
8377 				else if (persis_offset
8378 				      && i >= persis_offset)
8379 					lun->pending_ua[i - persis_offset] |=
8380 						CTL_UA_REG_PREEMPT;
8381 			} else if (msg->pr.pr_info.res_type != lun->res_type
8382 				&& (lun->res_type == SPR_TYPE_WR_EX_RO
8383 				 || lun->res_type == SPR_TYPE_EX_AC_RO)) {
8384 					if (!persis_offset
8385 					 && i < persis_offset)
8386 						lun->pending_ua[i] |=
8387 							CTL_UA_RES_RELEASE;
8388 					else if (persis_offset
8389 					      && i >= persis_offset)
8390 					lun->pending_ua[i - persis_offset] |=
8391 						CTL_UA_RES_RELEASE;
8392 			}
8393 		}
8394 		lun->per_res[msg->pr.pr_info.residx].registered = 1;
8395 		lun->res_type = msg->pr.pr_info.res_type;
8396 		if (lun->res_type != SPR_TYPE_WR_EX_AR
8397 		 && lun->res_type != SPR_TYPE_EX_AC_AR)
8398 			lun->pr_res_idx = msg->pr.pr_info.residx;
8399 		else
8400 			lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8401 	}
8402 	lun->PRGeneration++;
8403 
8404 }
8405 
8406 
8407 int
8408 ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8409 {
8410 	int retval;
8411 	int isc_retval;
8412 	u_int32_t param_len;
8413 	struct scsi_per_res_out *cdb;
8414 	struct ctl_lun *lun;
8415 	struct scsi_per_res_out_parms* param;
8416 	struct ctl_softc *softc;
8417 	uint32_t residx;
8418 	uint64_t res_key, sa_res_key;
8419 	uint8_t type;
8420 	union ctl_ha_msg persis_io;
8421 	int    i;
8422 
8423 	CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8424 
8425 	retval = CTL_RETVAL_COMPLETE;
8426 
8427 	softc = control_softc;
8428 
8429 	cdb = (struct scsi_per_res_out *)ctsio->cdb;
8430 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8431 
8432 	/*
8433 	 * We only support whole-LUN scope.  The scope & type are ignored for
8434 	 * register, register and ignore existing key and clear.
8435 	 * We sometimes ignore scope and type on preempts too!!
8436 	 * Verify reservation type here as well.
8437 	 */
8438 	type = cdb->scope_type & SPR_TYPE_MASK;
8439 	if ((cdb->action == SPRO_RESERVE)
8440 	 || (cdb->action == SPRO_RELEASE)) {
8441 		if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8442 			ctl_set_invalid_field(/*ctsio*/ ctsio,
8443 					      /*sks_valid*/ 1,
8444 					      /*command*/ 1,
8445 					      /*field*/ 2,
8446 					      /*bit_valid*/ 1,
8447 					      /*bit*/ 4);
8448 			ctl_done((union ctl_io *)ctsio);
8449 			return (CTL_RETVAL_COMPLETE);
8450 		}
8451 
8452 		if (type>8 || type==2 || type==4 || type==0) {
8453 			ctl_set_invalid_field(/*ctsio*/ ctsio,
8454 					      /*sks_valid*/ 1,
8455 					      /*command*/ 1,
8456 					      /*field*/ 2,
8457 					      /*bit_valid*/ 1,
8458 					      /*bit*/ 0);
8459 			ctl_done((union ctl_io *)ctsio);
8460 			return (CTL_RETVAL_COMPLETE);
8461 		}
8462 	}
8463 
8464 	param_len = scsi_4btoul(cdb->length);
8465 
8466 	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8467 		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8468 		ctsio->kern_data_len = param_len;
8469 		ctsio->kern_total_len = param_len;
8470 		ctsio->kern_data_resid = 0;
8471 		ctsio->kern_rel_offset = 0;
8472 		ctsio->kern_sg_entries = 0;
8473 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8474 		ctsio->be_move_done = ctl_config_move_done;
8475 		ctl_datamove((union ctl_io *)ctsio);
8476 
8477 		return (CTL_RETVAL_COMPLETE);
8478 	}
8479 
8480 	param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8481 
8482 	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8483 	res_key = scsi_8btou64(param->res_key.key);
8484 	sa_res_key = scsi_8btou64(param->serv_act_res_key);
8485 
8486 	/*
8487 	 * Validate the reservation key here except for SPRO_REG_IGNO
8488 	 * This must be done for all other service actions
8489 	 */
8490 	if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8491 		mtx_lock(&lun->lun_lock);
8492 		if (lun->per_res[residx].registered) {
8493 		    if (memcmp(param->res_key.key,
8494 			       lun->per_res[residx].res_key.key,
8495 			       ctl_min(sizeof(param->res_key),
8496 			       sizeof(lun->per_res[residx].res_key))) != 0) {
8497 				/*
8498 				 * The current key passed in doesn't match
8499 				 * the one the initiator previously
8500 				 * registered.
8501 				 */
8502 				mtx_unlock(&lun->lun_lock);
8503 				free(ctsio->kern_data_ptr, M_CTL);
8504 				ctl_set_reservation_conflict(ctsio);
8505 				ctl_done((union ctl_io *)ctsio);
8506 				return (CTL_RETVAL_COMPLETE);
8507 			}
8508 		} else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8509 			/*
8510 			 * We are not registered
8511 			 */
8512 			mtx_unlock(&lun->lun_lock);
8513 			free(ctsio->kern_data_ptr, M_CTL);
8514 			ctl_set_reservation_conflict(ctsio);
8515 			ctl_done((union ctl_io *)ctsio);
8516 			return (CTL_RETVAL_COMPLETE);
8517 		} else if (res_key != 0) {
8518 			/*
8519 			 * We are not registered and trying to register but
8520 			 * the register key isn't zero.
8521 			 */
8522 			mtx_unlock(&lun->lun_lock);
8523 			free(ctsio->kern_data_ptr, M_CTL);
8524 			ctl_set_reservation_conflict(ctsio);
8525 			ctl_done((union ctl_io *)ctsio);
8526 			return (CTL_RETVAL_COMPLETE);
8527 		}
8528 		mtx_unlock(&lun->lun_lock);
8529 	}
8530 
8531 	switch (cdb->action & SPRO_ACTION_MASK) {
8532 	case SPRO_REGISTER:
8533 	case SPRO_REG_IGNO: {
8534 
8535 #if 0
8536 		printf("Registration received\n");
8537 #endif
8538 
8539 		/*
8540 		 * We don't support any of these options, as we report in
8541 		 * the read capabilities request (see
8542 		 * ctl_persistent_reserve_in(), above).
8543 		 */
8544 		if ((param->flags & SPR_SPEC_I_PT)
8545 		 || (param->flags & SPR_ALL_TG_PT)
8546 		 || (param->flags & SPR_APTPL)) {
8547 			int bit_ptr;
8548 
8549 			if (param->flags & SPR_APTPL)
8550 				bit_ptr = 0;
8551 			else if (param->flags & SPR_ALL_TG_PT)
8552 				bit_ptr = 2;
8553 			else /* SPR_SPEC_I_PT */
8554 				bit_ptr = 3;
8555 
8556 			free(ctsio->kern_data_ptr, M_CTL);
8557 			ctl_set_invalid_field(ctsio,
8558 					      /*sks_valid*/ 1,
8559 					      /*command*/ 0,
8560 					      /*field*/ 20,
8561 					      /*bit_valid*/ 1,
8562 					      /*bit*/ bit_ptr);
8563 			ctl_done((union ctl_io *)ctsio);
8564 			return (CTL_RETVAL_COMPLETE);
8565 		}
8566 
8567 		mtx_lock(&lun->lun_lock);
8568 
8569 		/*
8570 		 * The initiator wants to clear the
8571 		 * key/unregister.
8572 		 */
8573 		if (sa_res_key == 0) {
8574 			if ((res_key == 0
8575 			  && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8576 			 || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8577 			  && !lun->per_res[residx].registered)) {
8578 				mtx_unlock(&lun->lun_lock);
8579 				goto done;
8580 			}
8581 
8582 			lun->per_res[residx].registered = 0;
8583 			memset(&lun->per_res[residx].res_key,
8584 			       0, sizeof(lun->per_res[residx].res_key));
8585 			lun->pr_key_count--;
8586 
8587 			if (residx == lun->pr_res_idx) {
8588 				lun->flags &= ~CTL_LUN_PR_RESERVED;
8589 				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8590 
8591 				if ((lun->res_type == SPR_TYPE_WR_EX_RO
8592 				  || lun->res_type == SPR_TYPE_EX_AC_RO)
8593 				 && lun->pr_key_count) {
8594 					/*
8595 					 * If the reservation is a registrants
8596 					 * only type we need to generate a UA
8597 					 * for other registered inits.  The
8598 					 * sense code should be RESERVATIONS
8599 					 * RELEASED
8600 					 */
8601 
8602 					for (i = 0; i < CTL_MAX_INITIATORS;i++){
8603 						if (lun->per_res[
8604 						    i+persis_offset].registered
8605 						    == 0)
8606 							continue;
8607 						lun->pending_ua[i] |=
8608 							CTL_UA_RES_RELEASE;
8609 					}
8610 				}
8611 				lun->res_type = 0;
8612 			} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8613 				if (lun->pr_key_count==0) {
8614 					lun->flags &= ~CTL_LUN_PR_RESERVED;
8615 					lun->res_type = 0;
8616 					lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8617 				}
8618 			}
8619 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8620 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8621 			persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8622 			persis_io.pr.pr_info.residx = residx;
8623 			if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8624 			     &persis_io, sizeof(persis_io), 0 )) >
8625 			     CTL_HA_STATUS_SUCCESS) {
8626 				printf("CTL:Persis Out error returned from "
8627 				       "ctl_ha_msg_send %d\n", isc_retval);
8628 			}
8629 		} else /* sa_res_key != 0 */ {
8630 
8631 			/*
8632 			 * If we aren't registered currently then increment
8633 			 * the key count and set the registered flag.
8634 			 */
8635 			if (!lun->per_res[residx].registered) {
8636 				lun->pr_key_count++;
8637 				lun->per_res[residx].registered = 1;
8638 			}
8639 
8640 			memcpy(&lun->per_res[residx].res_key,
8641 			       param->serv_act_res_key,
8642 			       ctl_min(sizeof(param->serv_act_res_key),
8643 			       sizeof(lun->per_res[residx].res_key)));
8644 
8645 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8646 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8647 			persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8648 			persis_io.pr.pr_info.residx = residx;
8649 			memcpy(persis_io.pr.pr_info.sa_res_key,
8650 			       param->serv_act_res_key,
8651 			       sizeof(param->serv_act_res_key));
8652 			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8653 			     &persis_io, sizeof(persis_io), 0)) >
8654 			     CTL_HA_STATUS_SUCCESS) {
8655 				printf("CTL:Persis Out error returned from "
8656 				       "ctl_ha_msg_send %d\n", isc_retval);
8657 			}
8658 		}
8659 		lun->PRGeneration++;
8660 		mtx_unlock(&lun->lun_lock);
8661 
8662 		break;
8663 	}
8664 	case SPRO_RESERVE:
8665 #if 0
8666                 printf("Reserve executed type %d\n", type);
8667 #endif
8668 		mtx_lock(&lun->lun_lock);
8669 		if (lun->flags & CTL_LUN_PR_RESERVED) {
8670 			/*
8671 			 * if this isn't the reservation holder and it's
8672 			 * not a "all registrants" type or if the type is
8673 			 * different then we have a conflict
8674 			 */
8675 			if ((lun->pr_res_idx != residx
8676 			  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8677 			 || lun->res_type != type) {
8678 				mtx_unlock(&lun->lun_lock);
8679 				free(ctsio->kern_data_ptr, M_CTL);
8680 				ctl_set_reservation_conflict(ctsio);
8681 				ctl_done((union ctl_io *)ctsio);
8682 				return (CTL_RETVAL_COMPLETE);
8683 			}
8684 			mtx_unlock(&lun->lun_lock);
8685 		} else /* create a reservation */ {
8686 			/*
8687 			 * If it's not an "all registrants" type record
8688 			 * reservation holder
8689 			 */
8690 			if (type != SPR_TYPE_WR_EX_AR
8691 			 && type != SPR_TYPE_EX_AC_AR)
8692 				lun->pr_res_idx = residx; /* Res holder */
8693 			else
8694 				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8695 
8696 			lun->flags |= CTL_LUN_PR_RESERVED;
8697 			lun->res_type = type;
8698 
8699 			mtx_unlock(&lun->lun_lock);
8700 
8701 			/* send msg to other side */
8702 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8703 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8704 			persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8705 			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8706 			persis_io.pr.pr_info.res_type = type;
8707 			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8708 			     &persis_io, sizeof(persis_io), 0)) >
8709 			     CTL_HA_STATUS_SUCCESS) {
8710 				printf("CTL:Persis Out error returned from "
8711 				       "ctl_ha_msg_send %d\n", isc_retval);
8712 			}
8713 		}
8714 		break;
8715 
8716 	case SPRO_RELEASE:
8717 		mtx_lock(&lun->lun_lock);
8718 		if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8719 			/* No reservation exists return good status */
8720 			mtx_unlock(&lun->lun_lock);
8721 			goto done;
8722 		}
8723 		/*
8724 		 * Is this nexus a reservation holder?
8725 		 */
8726 		if (lun->pr_res_idx != residx
8727 		 && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8728 			/*
8729 			 * not a res holder return good status but
8730 			 * do nothing
8731 			 */
8732 			mtx_unlock(&lun->lun_lock);
8733 			goto done;
8734 		}
8735 
8736 		if (lun->res_type != type) {
8737 			mtx_unlock(&lun->lun_lock);
8738 			free(ctsio->kern_data_ptr, M_CTL);
8739 			ctl_set_illegal_pr_release(ctsio);
8740 			ctl_done((union ctl_io *)ctsio);
8741 			return (CTL_RETVAL_COMPLETE);
8742 		}
8743 
8744 		/* okay to release */
8745 		lun->flags &= ~CTL_LUN_PR_RESERVED;
8746 		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8747 		lun->res_type = 0;
8748 
8749 		/*
8750 		 * if this isn't an exclusive access
8751 		 * res generate UA for all other
8752 		 * registrants.
8753 		 */
8754 		if (type != SPR_TYPE_EX_AC
8755 		 && type != SPR_TYPE_WR_EX) {
8756 			/*
8757 			 * temporarily unregister so we don't generate UA
8758 			 */
8759 			lun->per_res[residx].registered = 0;
8760 
8761 			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8762 				if (lun->per_res[i+persis_offset].registered
8763 				    == 0)
8764 					continue;
8765 				lun->pending_ua[i] |=
8766 					CTL_UA_RES_RELEASE;
8767 			}
8768 
8769 			lun->per_res[residx].registered = 1;
8770 		}
8771 		mtx_unlock(&lun->lun_lock);
8772 		/* Send msg to other side */
8773 		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8774 		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8775 		persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8776 		if ((isc_retval=ctl_ha_msg_send( CTL_HA_CHAN_CTL, &persis_io,
8777 		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8778 			printf("CTL:Persis Out error returned from "
8779 			       "ctl_ha_msg_send %d\n", isc_retval);
8780 		}
8781 		break;
8782 
8783 	case SPRO_CLEAR:
8784 		/* send msg to other side */
8785 
8786 		mtx_lock(&lun->lun_lock);
8787 		lun->flags &= ~CTL_LUN_PR_RESERVED;
8788 		lun->res_type = 0;
8789 		lun->pr_key_count = 0;
8790 		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8791 
8792 
8793 		memset(&lun->per_res[residx].res_key,
8794 		       0, sizeof(lun->per_res[residx].res_key));
8795 		lun->per_res[residx].registered = 0;
8796 
8797 		for (i=0; i < 2*CTL_MAX_INITIATORS; i++)
8798 			if (lun->per_res[i].registered) {
8799 				if (!persis_offset && i < CTL_MAX_INITIATORS)
8800 					lun->pending_ua[i] |=
8801 						CTL_UA_RES_PREEMPT;
8802 				else if (persis_offset && i >= persis_offset)
8803 					lun->pending_ua[i-persis_offset] |=
8804 					    CTL_UA_RES_PREEMPT;
8805 
8806 				memset(&lun->per_res[i].res_key,
8807 				       0, sizeof(struct scsi_per_res_key));
8808 				lun->per_res[i].registered = 0;
8809 			}
8810 		lun->PRGeneration++;
8811 		mtx_unlock(&lun->lun_lock);
8812 		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8813 		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8814 		persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8815 		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8816 		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8817 			printf("CTL:Persis Out error returned from "
8818 			       "ctl_ha_msg_send %d\n", isc_retval);
8819 		}
8820 		break;
8821 
8822 	case SPRO_PREEMPT: {
8823 		int nretval;
8824 
8825 		nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8826 					  residx, ctsio, cdb, param);
8827 		if (nretval != 0)
8828 			return (CTL_RETVAL_COMPLETE);
8829 		break;
8830 	}
8831 	default:
8832 		panic("Invalid PR type %x", cdb->action);
8833 	}
8834 
8835 done:
8836 	free(ctsio->kern_data_ptr, M_CTL);
8837 	ctl_set_success(ctsio);
8838 	ctl_done((union ctl_io *)ctsio);
8839 
8840 	return (retval);
8841 }
8842 
8843 /*
8844  * This routine is for handling a message from the other SC pertaining to
8845  * persistent reserve out. All the error checking will have been done
8846  * so only perorming the action need be done here to keep the two
8847  * in sync.
8848  */
8849 static void
8850 ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8851 {
8852 	struct ctl_lun *lun;
8853 	struct ctl_softc *softc;
8854 	int i;
8855 	uint32_t targ_lun;
8856 
8857 	softc = control_softc;
8858 
8859 	targ_lun = msg->hdr.nexus.targ_mapped_lun;
8860 	lun = softc->ctl_luns[targ_lun];
8861 	mtx_lock(&lun->lun_lock);
8862 	switch(msg->pr.pr_info.action) {
8863 	case CTL_PR_REG_KEY:
8864 		if (!lun->per_res[msg->pr.pr_info.residx].registered) {
8865 			lun->per_res[msg->pr.pr_info.residx].registered = 1;
8866 			lun->pr_key_count++;
8867 		}
8868 		lun->PRGeneration++;
8869 		memcpy(&lun->per_res[msg->pr.pr_info.residx].res_key,
8870 		       msg->pr.pr_info.sa_res_key,
8871 		       sizeof(struct scsi_per_res_key));
8872 		break;
8873 
8874 	case CTL_PR_UNREG_KEY:
8875 		lun->per_res[msg->pr.pr_info.residx].registered = 0;
8876 		memset(&lun->per_res[msg->pr.pr_info.residx].res_key,
8877 		       0, sizeof(struct scsi_per_res_key));
8878 		lun->pr_key_count--;
8879 
8880 		/* XXX Need to see if the reservation has been released */
8881 		/* if so do we need to generate UA? */
8882 		if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8883 			lun->flags &= ~CTL_LUN_PR_RESERVED;
8884 			lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8885 
8886 			if ((lun->res_type == SPR_TYPE_WR_EX_RO
8887 			  || lun->res_type == SPR_TYPE_EX_AC_RO)
8888 			 && lun->pr_key_count) {
8889 				/*
8890 				 * If the reservation is a registrants
8891 				 * only type we need to generate a UA
8892 				 * for other registered inits.  The
8893 				 * sense code should be RESERVATIONS
8894 				 * RELEASED
8895 				 */
8896 
8897 				for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8898 					if (lun->per_res[i+
8899 					    persis_offset].registered == 0)
8900 						continue;
8901 
8902 					lun->pending_ua[i] |=
8903 						CTL_UA_RES_RELEASE;
8904 				}
8905 			}
8906 			lun->res_type = 0;
8907 		} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8908 			if (lun->pr_key_count==0) {
8909 				lun->flags &= ~CTL_LUN_PR_RESERVED;
8910 				lun->res_type = 0;
8911 				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8912 			}
8913 		}
8914 		lun->PRGeneration++;
8915 		break;
8916 
8917 	case CTL_PR_RESERVE:
8918 		lun->flags |= CTL_LUN_PR_RESERVED;
8919 		lun->res_type = msg->pr.pr_info.res_type;
8920 		lun->pr_res_idx = msg->pr.pr_info.residx;
8921 
8922 		break;
8923 
8924 	case CTL_PR_RELEASE:
8925 		/*
8926 		 * if this isn't an exclusive access res generate UA for all
8927 		 * other registrants.
8928 		 */
8929 		if (lun->res_type != SPR_TYPE_EX_AC
8930 		 && lun->res_type != SPR_TYPE_WR_EX) {
8931 			for (i = 0; i < CTL_MAX_INITIATORS; i++)
8932 				if (lun->per_res[i+persis_offset].registered)
8933 					lun->pending_ua[i] |=
8934 						CTL_UA_RES_RELEASE;
8935 		}
8936 
8937 		lun->flags &= ~CTL_LUN_PR_RESERVED;
8938 		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8939 		lun->res_type = 0;
8940 		break;
8941 
8942 	case CTL_PR_PREEMPT:
8943 		ctl_pro_preempt_other(lun, msg);
8944 		break;
8945 	case CTL_PR_CLEAR:
8946 		lun->flags &= ~CTL_LUN_PR_RESERVED;
8947 		lun->res_type = 0;
8948 		lun->pr_key_count = 0;
8949 		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8950 
8951 		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8952 			if (lun->per_res[i].registered == 0)
8953 				continue;
8954 			if (!persis_offset
8955 			 && i < CTL_MAX_INITIATORS)
8956 				lun->pending_ua[i] |= CTL_UA_RES_PREEMPT;
8957 			else if (persis_offset
8958 			      && i >= persis_offset)
8959 				lun->pending_ua[i-persis_offset] |=
8960 					CTL_UA_RES_PREEMPT;
8961 			memset(&lun->per_res[i].res_key, 0,
8962 			       sizeof(struct scsi_per_res_key));
8963 			lun->per_res[i].registered = 0;
8964 		}
8965 		lun->PRGeneration++;
8966 		break;
8967 	}
8968 
8969 	mtx_unlock(&lun->lun_lock);
8970 }
8971 
8972 int
8973 ctl_read_write(struct ctl_scsiio *ctsio)
8974 {
8975 	struct ctl_lun *lun;
8976 	struct ctl_lba_len_flags *lbalen;
8977 	uint64_t lba;
8978 	uint32_t num_blocks;
8979 	int fua, dpo;
8980 	int retval;
8981 	int isread;
8982 
8983 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8984 
8985 	CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8986 
8987 	fua = 0;
8988 	dpo = 0;
8989 
8990 	retval = CTL_RETVAL_COMPLETE;
8991 
8992 	isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8993 	      || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8994 	if (lun->flags & CTL_LUN_PR_RESERVED && isread) {
8995 		uint32_t residx;
8996 
8997 		/*
8998 		 * XXX KDM need a lock here.
8999 		 */
9000 		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
9001 		if ((lun->res_type == SPR_TYPE_EX_AC
9002 		  && residx != lun->pr_res_idx)
9003 		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
9004 		   || lun->res_type == SPR_TYPE_EX_AC_AR)
9005 		  && !lun->per_res[residx].registered)) {
9006 			ctl_set_reservation_conflict(ctsio);
9007 			ctl_done((union ctl_io *)ctsio);
9008 			return (CTL_RETVAL_COMPLETE);
9009 	        }
9010 	}
9011 
9012 	switch (ctsio->cdb[0]) {
9013 	case READ_6:
9014 	case WRITE_6: {
9015 		struct scsi_rw_6 *cdb;
9016 
9017 		cdb = (struct scsi_rw_6 *)ctsio->cdb;
9018 
9019 		lba = scsi_3btoul(cdb->addr);
9020 		/* only 5 bits are valid in the most significant address byte */
9021 		lba &= 0x1fffff;
9022 		num_blocks = cdb->length;
9023 		/*
9024 		 * This is correct according to SBC-2.
9025 		 */
9026 		if (num_blocks == 0)
9027 			num_blocks = 256;
9028 		break;
9029 	}
9030 	case READ_10:
9031 	case WRITE_10: {
9032 		struct scsi_rw_10 *cdb;
9033 
9034 		cdb = (struct scsi_rw_10 *)ctsio->cdb;
9035 
9036 		if (cdb->byte2 & SRW10_FUA)
9037 			fua = 1;
9038 		if (cdb->byte2 & SRW10_DPO)
9039 			dpo = 1;
9040 
9041 		lba = scsi_4btoul(cdb->addr);
9042 		num_blocks = scsi_2btoul(cdb->length);
9043 		break;
9044 	}
9045 	case WRITE_VERIFY_10: {
9046 		struct scsi_write_verify_10 *cdb;
9047 
9048 		cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
9049 
9050 		/*
9051 		 * XXX KDM we should do actual write verify support at some
9052 		 * point.  This is obviously fake, we're just translating
9053 		 * things to a write.  So we don't even bother checking the
9054 		 * BYTCHK field, since we don't do any verification.  If
9055 		 * the user asks for it, we'll just pretend we did it.
9056 		 */
9057 		if (cdb->byte2 & SWV_DPO)
9058 			dpo = 1;
9059 
9060 		lba = scsi_4btoul(cdb->addr);
9061 		num_blocks = scsi_2btoul(cdb->length);
9062 		break;
9063 	}
9064 	case READ_12:
9065 	case WRITE_12: {
9066 		struct scsi_rw_12 *cdb;
9067 
9068 		cdb = (struct scsi_rw_12 *)ctsio->cdb;
9069 
9070 		if (cdb->byte2 & SRW12_FUA)
9071 			fua = 1;
9072 		if (cdb->byte2 & SRW12_DPO)
9073 			dpo = 1;
9074 		lba = scsi_4btoul(cdb->addr);
9075 		num_blocks = scsi_4btoul(cdb->length);
9076 		break;
9077 	}
9078 	case WRITE_VERIFY_12: {
9079 		struct scsi_write_verify_12 *cdb;
9080 
9081 		cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
9082 
9083 		if (cdb->byte2 & SWV_DPO)
9084 			dpo = 1;
9085 
9086 		lba = scsi_4btoul(cdb->addr);
9087 		num_blocks = scsi_4btoul(cdb->length);
9088 
9089 		break;
9090 	}
9091 	case READ_16:
9092 	case WRITE_16: {
9093 		struct scsi_rw_16 *cdb;
9094 
9095 		cdb = (struct scsi_rw_16 *)ctsio->cdb;
9096 
9097 		if (cdb->byte2 & SRW12_FUA)
9098 			fua = 1;
9099 		if (cdb->byte2 & SRW12_DPO)
9100 			dpo = 1;
9101 
9102 		lba = scsi_8btou64(cdb->addr);
9103 		num_blocks = scsi_4btoul(cdb->length);
9104 		break;
9105 	}
9106 	case WRITE_VERIFY_16: {
9107 		struct scsi_write_verify_16 *cdb;
9108 
9109 		cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
9110 
9111 		if (cdb->byte2 & SWV_DPO)
9112 			dpo = 1;
9113 
9114 		lba = scsi_8btou64(cdb->addr);
9115 		num_blocks = scsi_4btoul(cdb->length);
9116 		break;
9117 	}
9118 	default:
9119 		/*
9120 		 * We got a command we don't support.  This shouldn't
9121 		 * happen, commands should be filtered out above us.
9122 		 */
9123 		ctl_set_invalid_opcode(ctsio);
9124 		ctl_done((union ctl_io *)ctsio);
9125 
9126 		return (CTL_RETVAL_COMPLETE);
9127 		break; /* NOTREACHED */
9128 	}
9129 
9130 	/*
9131 	 * XXX KDM what do we do with the DPO and FUA bits?  FUA might be
9132 	 * interesting for us, but if RAIDCore is in write-back mode,
9133 	 * getting it to do write-through for a particular transaction may
9134 	 * not be possible.
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 	 * Note that this cannot happen with WRITE(6) or READ(6), since 0
9153 	 * translates to 256 blocks for those commands.
9154 	 */
9155 	if (num_blocks == 0) {
9156 		ctl_set_success(ctsio);
9157 		ctl_done((union ctl_io *)ctsio);
9158 		return (CTL_RETVAL_COMPLETE);
9159 	}
9160 
9161 	lbalen = (struct ctl_lba_len_flags *)
9162 	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9163 	lbalen->lba = lba;
9164 	lbalen->len = num_blocks;
9165 	lbalen->flags = isread ? CTL_LLF_READ : CTL_LLF_WRITE;
9166 
9167 	ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9168 	ctsio->kern_rel_offset = 0;
9169 
9170 	CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
9171 
9172 	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9173 
9174 	return (retval);
9175 }
9176 
9177 static int
9178 ctl_cnw_cont(union ctl_io *io)
9179 {
9180 	struct ctl_scsiio *ctsio;
9181 	struct ctl_lun *lun;
9182 	struct ctl_lba_len_flags *lbalen;
9183 	int retval;
9184 
9185 	ctsio = &io->scsiio;
9186 	ctsio->io_hdr.status = CTL_STATUS_NONE;
9187 	ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
9188 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9189 	lbalen = (struct ctl_lba_len_flags *)
9190 	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9191 	lbalen->flags = CTL_LLF_WRITE;
9192 
9193 	CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
9194 	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9195 	return (retval);
9196 }
9197 
9198 int
9199 ctl_cnw(struct ctl_scsiio *ctsio)
9200 {
9201 	struct ctl_lun *lun;
9202 	struct ctl_lba_len_flags *lbalen;
9203 	uint64_t lba;
9204 	uint32_t num_blocks;
9205 	int fua, dpo;
9206 	int retval;
9207 
9208 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9209 
9210 	CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
9211 
9212 	fua = 0;
9213 	dpo = 0;
9214 
9215 	retval = CTL_RETVAL_COMPLETE;
9216 
9217 	switch (ctsio->cdb[0]) {
9218 	case COMPARE_AND_WRITE: {
9219 		struct scsi_compare_and_write *cdb;
9220 
9221 		cdb = (struct scsi_compare_and_write *)ctsio->cdb;
9222 
9223 		if (cdb->byte2 & SRW10_FUA)
9224 			fua = 1;
9225 		if (cdb->byte2 & SRW10_DPO)
9226 			dpo = 1;
9227 		lba = scsi_8btou64(cdb->addr);
9228 		num_blocks = cdb->length;
9229 		break;
9230 	}
9231 	default:
9232 		/*
9233 		 * We got a command we don't support.  This shouldn't
9234 		 * happen, commands should be filtered out above us.
9235 		 */
9236 		ctl_set_invalid_opcode(ctsio);
9237 		ctl_done((union ctl_io *)ctsio);
9238 
9239 		return (CTL_RETVAL_COMPLETE);
9240 		break; /* NOTREACHED */
9241 	}
9242 
9243 	/*
9244 	 * XXX KDM what do we do with the DPO and FUA bits?  FUA might be
9245 	 * interesting for us, but if RAIDCore is in write-back mode,
9246 	 * getting it to do write-through for a particular transaction may
9247 	 * not be possible.
9248 	 */
9249 
9250 	/*
9251 	 * The first check is to make sure we're in bounds, the second
9252 	 * check is to catch wrap-around problems.  If the lba + num blocks
9253 	 * is less than the lba, then we've wrapped around and the block
9254 	 * range is invalid anyway.
9255 	 */
9256 	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9257 	 || ((lba + num_blocks) < lba)) {
9258 		ctl_set_lba_out_of_range(ctsio);
9259 		ctl_done((union ctl_io *)ctsio);
9260 		return (CTL_RETVAL_COMPLETE);
9261 	}
9262 
9263 	/*
9264 	 * According to SBC-3, a transfer length of 0 is not an error.
9265 	 */
9266 	if (num_blocks == 0) {
9267 		ctl_set_success(ctsio);
9268 		ctl_done((union ctl_io *)ctsio);
9269 		return (CTL_RETVAL_COMPLETE);
9270 	}
9271 
9272 	ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
9273 	ctsio->kern_rel_offset = 0;
9274 
9275 	/*
9276 	 * Set the IO_CONT flag, so that if this I/O gets passed to
9277 	 * ctl_data_submit_done(), it'll get passed back to
9278 	 * ctl_ctl_cnw_cont() for further processing.
9279 	 */
9280 	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
9281 	ctsio->io_cont = ctl_cnw_cont;
9282 
9283 	lbalen = (struct ctl_lba_len_flags *)
9284 	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9285 	lbalen->lba = lba;
9286 	lbalen->len = num_blocks;
9287 	lbalen->flags = CTL_LLF_COMPARE;
9288 
9289 	CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
9290 	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9291 	return (retval);
9292 }
9293 
9294 int
9295 ctl_verify(struct ctl_scsiio *ctsio)
9296 {
9297 	struct ctl_lun *lun;
9298 	struct ctl_lba_len_flags *lbalen;
9299 	uint64_t lba;
9300 	uint32_t num_blocks;
9301 	int bytchk, dpo;
9302 	int retval;
9303 
9304 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9305 
9306 	CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
9307 
9308 	bytchk = 0;
9309 	dpo = 0;
9310 	retval = CTL_RETVAL_COMPLETE;
9311 
9312 	switch (ctsio->cdb[0]) {
9313 	case VERIFY_10: {
9314 		struct scsi_verify_10 *cdb;
9315 
9316 		cdb = (struct scsi_verify_10 *)ctsio->cdb;
9317 		if (cdb->byte2 & SVFY_BYTCHK)
9318 			bytchk = 1;
9319 		if (cdb->byte2 & SVFY_DPO)
9320 			dpo = 1;
9321 		lba = scsi_4btoul(cdb->addr);
9322 		num_blocks = scsi_2btoul(cdb->length);
9323 		break;
9324 	}
9325 	case VERIFY_12: {
9326 		struct scsi_verify_12 *cdb;
9327 
9328 		cdb = (struct scsi_verify_12 *)ctsio->cdb;
9329 		if (cdb->byte2 & SVFY_BYTCHK)
9330 			bytchk = 1;
9331 		if (cdb->byte2 & SVFY_DPO)
9332 			dpo = 1;
9333 		lba = scsi_4btoul(cdb->addr);
9334 		num_blocks = scsi_4btoul(cdb->length);
9335 		break;
9336 	}
9337 	case VERIFY_16: {
9338 		struct scsi_rw_16 *cdb;
9339 
9340 		cdb = (struct scsi_rw_16 *)ctsio->cdb;
9341 		if (cdb->byte2 & SVFY_BYTCHK)
9342 			bytchk = 1;
9343 		if (cdb->byte2 & SVFY_DPO)
9344 			dpo = 1;
9345 		lba = scsi_8btou64(cdb->addr);
9346 		num_blocks = scsi_4btoul(cdb->length);
9347 		break;
9348 	}
9349 	default:
9350 		/*
9351 		 * We got a command we don't support.  This shouldn't
9352 		 * happen, commands should be filtered out above us.
9353 		 */
9354 		ctl_set_invalid_opcode(ctsio);
9355 		ctl_done((union ctl_io *)ctsio);
9356 		return (CTL_RETVAL_COMPLETE);
9357 	}
9358 
9359 	/*
9360 	 * The first check is to make sure we're in bounds, the second
9361 	 * check is to catch wrap-around problems.  If the lba + num blocks
9362 	 * is less than the lba, then we've wrapped around and the block
9363 	 * range is invalid anyway.
9364 	 */
9365 	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9366 	 || ((lba + num_blocks) < lba)) {
9367 		ctl_set_lba_out_of_range(ctsio);
9368 		ctl_done((union ctl_io *)ctsio);
9369 		return (CTL_RETVAL_COMPLETE);
9370 	}
9371 
9372 	/*
9373 	 * According to SBC-3, a transfer length of 0 is not an error.
9374 	 */
9375 	if (num_blocks == 0) {
9376 		ctl_set_success(ctsio);
9377 		ctl_done((union ctl_io *)ctsio);
9378 		return (CTL_RETVAL_COMPLETE);
9379 	}
9380 
9381 	lbalen = (struct ctl_lba_len_flags *)
9382 	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9383 	lbalen->lba = lba;
9384 	lbalen->len = num_blocks;
9385 	if (bytchk) {
9386 		lbalen->flags = CTL_LLF_COMPARE;
9387 		ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9388 	} else {
9389 		lbalen->flags = CTL_LLF_VERIFY;
9390 		ctsio->kern_total_len = 0;
9391 	}
9392 	ctsio->kern_rel_offset = 0;
9393 
9394 	CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9395 	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9396 	return (retval);
9397 }
9398 
9399 int
9400 ctl_report_luns(struct ctl_scsiio *ctsio)
9401 {
9402 	struct scsi_report_luns *cdb;
9403 	struct scsi_report_luns_data *lun_data;
9404 	struct ctl_lun *lun, *request_lun;
9405 	int num_luns, retval;
9406 	uint32_t alloc_len, lun_datalen;
9407 	int num_filled, well_known;
9408 	uint32_t initidx, targ_lun_id, lun_id;
9409 
9410 	retval = CTL_RETVAL_COMPLETE;
9411 	well_known = 0;
9412 
9413 	cdb = (struct scsi_report_luns *)ctsio->cdb;
9414 
9415 	CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9416 
9417 	mtx_lock(&control_softc->ctl_lock);
9418 	num_luns = control_softc->num_luns;
9419 	mtx_unlock(&control_softc->ctl_lock);
9420 
9421 	switch (cdb->select_report) {
9422 	case RPL_REPORT_DEFAULT:
9423 	case RPL_REPORT_ALL:
9424 		break;
9425 	case RPL_REPORT_WELLKNOWN:
9426 		well_known = 1;
9427 		num_luns = 0;
9428 		break;
9429 	default:
9430 		ctl_set_invalid_field(ctsio,
9431 				      /*sks_valid*/ 1,
9432 				      /*command*/ 1,
9433 				      /*field*/ 2,
9434 				      /*bit_valid*/ 0,
9435 				      /*bit*/ 0);
9436 		ctl_done((union ctl_io *)ctsio);
9437 		return (retval);
9438 		break; /* NOTREACHED */
9439 	}
9440 
9441 	alloc_len = scsi_4btoul(cdb->length);
9442 	/*
9443 	 * The initiator has to allocate at least 16 bytes for this request,
9444 	 * so he can at least get the header and the first LUN.  Otherwise
9445 	 * we reject the request (per SPC-3 rev 14, section 6.21).
9446 	 */
9447 	if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9448 	    sizeof(struct scsi_report_luns_lundata))) {
9449 		ctl_set_invalid_field(ctsio,
9450 				      /*sks_valid*/ 1,
9451 				      /*command*/ 1,
9452 				      /*field*/ 6,
9453 				      /*bit_valid*/ 0,
9454 				      /*bit*/ 0);
9455 		ctl_done((union ctl_io *)ctsio);
9456 		return (retval);
9457 	}
9458 
9459 	request_lun = (struct ctl_lun *)
9460 		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9461 
9462 	lun_datalen = sizeof(*lun_data) +
9463 		(num_luns * sizeof(struct scsi_report_luns_lundata));
9464 
9465 	ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9466 	lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9467 	ctsio->kern_sg_entries = 0;
9468 
9469 	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9470 
9471 	mtx_lock(&control_softc->ctl_lock);
9472 	for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
9473 		lun_id = ctl_map_lun(ctsio->io_hdr.nexus.targ_port, targ_lun_id);
9474 		if (lun_id >= CTL_MAX_LUNS)
9475 			continue;
9476 		lun = control_softc->ctl_luns[lun_id];
9477 		if (lun == NULL)
9478 			continue;
9479 
9480 		if (targ_lun_id <= 0xff) {
9481 			/*
9482 			 * Peripheral addressing method, bus number 0.
9483 			 */
9484 			lun_data->luns[num_filled].lundata[0] =
9485 				RPL_LUNDATA_ATYP_PERIPH;
9486 			lun_data->luns[num_filled].lundata[1] = targ_lun_id;
9487 			num_filled++;
9488 		} else if (targ_lun_id <= 0x3fff) {
9489 			/*
9490 			 * Flat addressing method.
9491 			 */
9492 			lun_data->luns[num_filled].lundata[0] =
9493 				RPL_LUNDATA_ATYP_FLAT |
9494 				(targ_lun_id & RPL_LUNDATA_FLAT_LUN_MASK);
9495 #ifdef OLDCTLHEADERS
9496 				(SRLD_ADDR_FLAT << SRLD_ADDR_SHIFT) |
9497 				(targ_lun_id & SRLD_BUS_LUN_MASK);
9498 #endif
9499 			lun_data->luns[num_filled].lundata[1] =
9500 #ifdef OLDCTLHEADERS
9501 				targ_lun_id >> SRLD_BUS_LUN_BITS;
9502 #endif
9503 				targ_lun_id >> RPL_LUNDATA_FLAT_LUN_BITS;
9504 			num_filled++;
9505 		} else {
9506 			printf("ctl_report_luns: bogus LUN number %jd, "
9507 			       "skipping\n", (intmax_t)targ_lun_id);
9508 		}
9509 		/*
9510 		 * According to SPC-3, rev 14 section 6.21:
9511 		 *
9512 		 * "The execution of a REPORT LUNS command to any valid and
9513 		 * installed logical unit shall clear the REPORTED LUNS DATA
9514 		 * HAS CHANGED unit attention condition for all logical
9515 		 * units of that target with respect to the requesting
9516 		 * initiator. A valid and installed logical unit is one
9517 		 * having a PERIPHERAL QUALIFIER of 000b in the standard
9518 		 * INQUIRY data (see 6.4.2)."
9519 		 *
9520 		 * If request_lun is NULL, the LUN this report luns command
9521 		 * was issued to is either disabled or doesn't exist. In that
9522 		 * case, we shouldn't clear any pending lun change unit
9523 		 * attention.
9524 		 */
9525 		if (request_lun != NULL) {
9526 			mtx_lock(&lun->lun_lock);
9527 			lun->pending_ua[initidx] &= ~CTL_UA_LUN_CHANGE;
9528 			mtx_unlock(&lun->lun_lock);
9529 		}
9530 	}
9531 	mtx_unlock(&control_softc->ctl_lock);
9532 
9533 	/*
9534 	 * It's quite possible that we've returned fewer LUNs than we allocated
9535 	 * space for.  Trim it.
9536 	 */
9537 	lun_datalen = sizeof(*lun_data) +
9538 		(num_filled * sizeof(struct scsi_report_luns_lundata));
9539 
9540 	if (lun_datalen < alloc_len) {
9541 		ctsio->residual = alloc_len - lun_datalen;
9542 		ctsio->kern_data_len = lun_datalen;
9543 		ctsio->kern_total_len = lun_datalen;
9544 	} else {
9545 		ctsio->residual = 0;
9546 		ctsio->kern_data_len = alloc_len;
9547 		ctsio->kern_total_len = alloc_len;
9548 	}
9549 	ctsio->kern_data_resid = 0;
9550 	ctsio->kern_rel_offset = 0;
9551 	ctsio->kern_sg_entries = 0;
9552 
9553 	/*
9554 	 * We set this to the actual data length, regardless of how much
9555 	 * space we actually have to return results.  If the user looks at
9556 	 * this value, he'll know whether or not he allocated enough space
9557 	 * and reissue the command if necessary.  We don't support well
9558 	 * known logical units, so if the user asks for that, return none.
9559 	 */
9560 	scsi_ulto4b(lun_datalen - 8, lun_data->length);
9561 
9562 	/*
9563 	 * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9564 	 * this request.
9565 	 */
9566 	ctsio->scsi_status = SCSI_STATUS_OK;
9567 
9568 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9569 	ctsio->be_move_done = ctl_config_move_done;
9570 	ctl_datamove((union ctl_io *)ctsio);
9571 
9572 	return (retval);
9573 }
9574 
9575 int
9576 ctl_request_sense(struct ctl_scsiio *ctsio)
9577 {
9578 	struct scsi_request_sense *cdb;
9579 	struct scsi_sense_data *sense_ptr;
9580 	struct ctl_lun *lun;
9581 	uint32_t initidx;
9582 	int have_error;
9583 	scsi_sense_data_type sense_format;
9584 
9585 	cdb = (struct scsi_request_sense *)ctsio->cdb;
9586 
9587 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9588 
9589 	CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9590 
9591 	/*
9592 	 * Determine which sense format the user wants.
9593 	 */
9594 	if (cdb->byte2 & SRS_DESC)
9595 		sense_format = SSD_TYPE_DESC;
9596 	else
9597 		sense_format = SSD_TYPE_FIXED;
9598 
9599 	ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9600 	sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9601 	ctsio->kern_sg_entries = 0;
9602 
9603 	/*
9604 	 * struct scsi_sense_data, which is currently set to 256 bytes, is
9605 	 * larger than the largest allowed value for the length field in the
9606 	 * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9607 	 */
9608 	ctsio->residual = 0;
9609 	ctsio->kern_data_len = cdb->length;
9610 	ctsio->kern_total_len = cdb->length;
9611 
9612 	ctsio->kern_data_resid = 0;
9613 	ctsio->kern_rel_offset = 0;
9614 	ctsio->kern_sg_entries = 0;
9615 
9616 	/*
9617 	 * If we don't have a LUN, we don't have any pending sense.
9618 	 */
9619 	if (lun == NULL)
9620 		goto no_sense;
9621 
9622 	have_error = 0;
9623 	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9624 	/*
9625 	 * Check for pending sense, and then for pending unit attentions.
9626 	 * Pending sense gets returned first, then pending unit attentions.
9627 	 */
9628 	mtx_lock(&lun->lun_lock);
9629 #ifdef CTL_WITH_CA
9630 	if (ctl_is_set(lun->have_ca, initidx)) {
9631 		scsi_sense_data_type stored_format;
9632 
9633 		/*
9634 		 * Check to see which sense format was used for the stored
9635 		 * sense data.
9636 		 */
9637 		stored_format = scsi_sense_type(&lun->pending_sense[initidx]);
9638 
9639 		/*
9640 		 * If the user requested a different sense format than the
9641 		 * one we stored, then we need to convert it to the other
9642 		 * format.  If we're going from descriptor to fixed format
9643 		 * sense data, we may lose things in translation, depending
9644 		 * on what options were used.
9645 		 *
9646 		 * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9647 		 * for some reason we'll just copy it out as-is.
9648 		 */
9649 		if ((stored_format == SSD_TYPE_FIXED)
9650 		 && (sense_format == SSD_TYPE_DESC))
9651 			ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9652 			    &lun->pending_sense[initidx],
9653 			    (struct scsi_sense_data_desc *)sense_ptr);
9654 		else if ((stored_format == SSD_TYPE_DESC)
9655 		      && (sense_format == SSD_TYPE_FIXED))
9656 			ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9657 			    &lun->pending_sense[initidx],
9658 			    (struct scsi_sense_data_fixed *)sense_ptr);
9659 		else
9660 			memcpy(sense_ptr, &lun->pending_sense[initidx],
9661 			       ctl_min(sizeof(*sense_ptr),
9662 			       sizeof(lun->pending_sense[initidx])));
9663 
9664 		ctl_clear_mask(lun->have_ca, initidx);
9665 		have_error = 1;
9666 	} else
9667 #endif
9668 	if (lun->pending_ua[initidx] != CTL_UA_NONE) {
9669 		ctl_ua_type ua_type;
9670 
9671 		ua_type = ctl_build_ua(lun->pending_ua[initidx],
9672 				       sense_ptr, sense_format);
9673 		if (ua_type != CTL_UA_NONE) {
9674 			have_error = 1;
9675 			/* We're reporting this UA, so clear it */
9676 			lun->pending_ua[initidx] &= ~ua_type;
9677 		}
9678 	}
9679 	mtx_unlock(&lun->lun_lock);
9680 
9681 	/*
9682 	 * We already have a pending error, return it.
9683 	 */
9684 	if (have_error != 0) {
9685 		/*
9686 		 * We report the SCSI status as OK, since the status of the
9687 		 * request sense command itself is OK.
9688 		 */
9689 		ctsio->scsi_status = SCSI_STATUS_OK;
9690 
9691 		/*
9692 		 * We report 0 for the sense length, because we aren't doing
9693 		 * autosense in this case.  We're reporting sense as
9694 		 * parameter data.
9695 		 */
9696 		ctsio->sense_len = 0;
9697 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9698 		ctsio->be_move_done = ctl_config_move_done;
9699 		ctl_datamove((union ctl_io *)ctsio);
9700 
9701 		return (CTL_RETVAL_COMPLETE);
9702 	}
9703 
9704 no_sense:
9705 
9706 	/*
9707 	 * No sense information to report, so we report that everything is
9708 	 * okay.
9709 	 */
9710 	ctl_set_sense_data(sense_ptr,
9711 			   lun,
9712 			   sense_format,
9713 			   /*current_error*/ 1,
9714 			   /*sense_key*/ SSD_KEY_NO_SENSE,
9715 			   /*asc*/ 0x00,
9716 			   /*ascq*/ 0x00,
9717 			   SSD_ELEM_NONE);
9718 
9719 	ctsio->scsi_status = SCSI_STATUS_OK;
9720 
9721 	/*
9722 	 * We report 0 for the sense length, because we aren't doing
9723 	 * autosense in this case.  We're reporting sense as parameter data.
9724 	 */
9725 	ctsio->sense_len = 0;
9726 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9727 	ctsio->be_move_done = ctl_config_move_done;
9728 	ctl_datamove((union ctl_io *)ctsio);
9729 
9730 	return (CTL_RETVAL_COMPLETE);
9731 }
9732 
9733 int
9734 ctl_tur(struct ctl_scsiio *ctsio)
9735 {
9736 	struct ctl_lun *lun;
9737 
9738 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9739 
9740 	CTL_DEBUG_PRINT(("ctl_tur\n"));
9741 
9742 	if (lun == NULL)
9743 		return (EINVAL);
9744 
9745 	ctsio->scsi_status = SCSI_STATUS_OK;
9746 	ctsio->io_hdr.status = CTL_SUCCESS;
9747 
9748 	ctl_done((union ctl_io *)ctsio);
9749 
9750 	return (CTL_RETVAL_COMPLETE);
9751 }
9752 
9753 #ifdef notyet
9754 static int
9755 ctl_cmddt_inquiry(struct ctl_scsiio *ctsio)
9756 {
9757 
9758 }
9759 #endif
9760 
9761 static int
9762 ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9763 {
9764 	struct scsi_vpd_supported_pages *pages;
9765 	int sup_page_size;
9766 	struct ctl_lun *lun;
9767 
9768 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9769 
9770 	sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9771 	    SCSI_EVPD_NUM_SUPPORTED_PAGES;
9772 	ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9773 	pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9774 	ctsio->kern_sg_entries = 0;
9775 
9776 	if (sup_page_size < alloc_len) {
9777 		ctsio->residual = alloc_len - sup_page_size;
9778 		ctsio->kern_data_len = sup_page_size;
9779 		ctsio->kern_total_len = sup_page_size;
9780 	} else {
9781 		ctsio->residual = 0;
9782 		ctsio->kern_data_len = alloc_len;
9783 		ctsio->kern_total_len = alloc_len;
9784 	}
9785 	ctsio->kern_data_resid = 0;
9786 	ctsio->kern_rel_offset = 0;
9787 	ctsio->kern_sg_entries = 0;
9788 
9789 	/*
9790 	 * The control device is always connected.  The disk device, on the
9791 	 * other hand, may not be online all the time.  Need to change this
9792 	 * to figure out whether the disk device is actually online or not.
9793 	 */
9794 	if (lun != NULL)
9795 		pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9796 				lun->be_lun->lun_type;
9797 	else
9798 		pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9799 
9800 	pages->length = SCSI_EVPD_NUM_SUPPORTED_PAGES;
9801 	/* Supported VPD pages */
9802 	pages->page_list[0] = SVPD_SUPPORTED_PAGES;
9803 	/* Serial Number */
9804 	pages->page_list[1] = SVPD_UNIT_SERIAL_NUMBER;
9805 	/* Device Identification */
9806 	pages->page_list[2] = SVPD_DEVICE_ID;
9807 	/* SCSI Ports */
9808 	pages->page_list[3] = SVPD_SCSI_PORTS;
9809 	/* Third-party Copy */
9810 	pages->page_list[4] = SVPD_SCSI_TPC;
9811 	/* Block limits */
9812 	pages->page_list[5] = SVPD_BLOCK_LIMITS;
9813 	/* Block Device Characteristics */
9814 	pages->page_list[6] = SVPD_BDC;
9815 	/* Logical Block Provisioning */
9816 	pages->page_list[7] = SVPD_LBP;
9817 
9818 	ctsio->scsi_status = SCSI_STATUS_OK;
9819 
9820 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9821 	ctsio->be_move_done = ctl_config_move_done;
9822 	ctl_datamove((union ctl_io *)ctsio);
9823 
9824 	return (CTL_RETVAL_COMPLETE);
9825 }
9826 
9827 static int
9828 ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9829 {
9830 	struct scsi_vpd_unit_serial_number *sn_ptr;
9831 	struct ctl_lun *lun;
9832 
9833 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9834 
9835 	ctsio->kern_data_ptr = malloc(sizeof(*sn_ptr), M_CTL, M_WAITOK | M_ZERO);
9836 	sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9837 	ctsio->kern_sg_entries = 0;
9838 
9839 	if (sizeof(*sn_ptr) < alloc_len) {
9840 		ctsio->residual = alloc_len - sizeof(*sn_ptr);
9841 		ctsio->kern_data_len = sizeof(*sn_ptr);
9842 		ctsio->kern_total_len = sizeof(*sn_ptr);
9843 	} else {
9844 		ctsio->residual = 0;
9845 		ctsio->kern_data_len = alloc_len;
9846 		ctsio->kern_total_len = alloc_len;
9847 	}
9848 	ctsio->kern_data_resid = 0;
9849 	ctsio->kern_rel_offset = 0;
9850 	ctsio->kern_sg_entries = 0;
9851 
9852 	/*
9853 	 * The control device is always connected.  The disk device, on the
9854 	 * other hand, may not be online all the time.  Need to change this
9855 	 * to figure out whether the disk device is actually online or not.
9856 	 */
9857 	if (lun != NULL)
9858 		sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9859 				  lun->be_lun->lun_type;
9860 	else
9861 		sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9862 
9863 	sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9864 	sn_ptr->length = ctl_min(sizeof(*sn_ptr) - 4, CTL_SN_LEN);
9865 	/*
9866 	 * If we don't have a LUN, we just leave the serial number as
9867 	 * all spaces.
9868 	 */
9869 	memset(sn_ptr->serial_num, 0x20, sizeof(sn_ptr->serial_num));
9870 	if (lun != NULL) {
9871 		strncpy((char *)sn_ptr->serial_num,
9872 			(char *)lun->be_lun->serial_num, CTL_SN_LEN);
9873 	}
9874 	ctsio->scsi_status = SCSI_STATUS_OK;
9875 
9876 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9877 	ctsio->be_move_done = ctl_config_move_done;
9878 	ctl_datamove((union ctl_io *)ctsio);
9879 
9880 	return (CTL_RETVAL_COMPLETE);
9881 }
9882 
9883 
9884 static int
9885 ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9886 {
9887 	struct scsi_vpd_device_id *devid_ptr;
9888 	struct scsi_vpd_id_descriptor *desc;
9889 	struct ctl_softc *ctl_softc;
9890 	struct ctl_lun *lun;
9891 	struct ctl_port *port;
9892 	int data_len;
9893 	uint8_t proto;
9894 
9895 	ctl_softc = control_softc;
9896 
9897 	port = ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)];
9898 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9899 
9900 	data_len = sizeof(struct scsi_vpd_device_id) +
9901 	    sizeof(struct scsi_vpd_id_descriptor) +
9902 		sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9903 	    sizeof(struct scsi_vpd_id_descriptor) +
9904 		sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9905 	if (lun && lun->lun_devid)
9906 		data_len += lun->lun_devid->len;
9907 	if (port->port_devid)
9908 		data_len += port->port_devid->len;
9909 	if (port->target_devid)
9910 		data_len += port->target_devid->len;
9911 
9912 	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9913 	devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9914 	ctsio->kern_sg_entries = 0;
9915 
9916 	if (data_len < alloc_len) {
9917 		ctsio->residual = alloc_len - data_len;
9918 		ctsio->kern_data_len = data_len;
9919 		ctsio->kern_total_len = data_len;
9920 	} else {
9921 		ctsio->residual = 0;
9922 		ctsio->kern_data_len = alloc_len;
9923 		ctsio->kern_total_len = alloc_len;
9924 	}
9925 	ctsio->kern_data_resid = 0;
9926 	ctsio->kern_rel_offset = 0;
9927 	ctsio->kern_sg_entries = 0;
9928 
9929 	/*
9930 	 * The control device is always connected.  The disk device, on the
9931 	 * other hand, may not be online all the time.
9932 	 */
9933 	if (lun != NULL)
9934 		devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9935 				     lun->be_lun->lun_type;
9936 	else
9937 		devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9938 	devid_ptr->page_code = SVPD_DEVICE_ID;
9939 	scsi_ulto2b(data_len - 4, devid_ptr->length);
9940 
9941 	if (port->port_type == CTL_PORT_FC)
9942 		proto = SCSI_PROTO_FC << 4;
9943 	else if (port->port_type == CTL_PORT_ISCSI)
9944 		proto = SCSI_PROTO_ISCSI << 4;
9945 	else
9946 		proto = SCSI_PROTO_SPI << 4;
9947 	desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9948 
9949 	/*
9950 	 * We're using a LUN association here.  i.e., this device ID is a
9951 	 * per-LUN identifier.
9952 	 */
9953 	if (lun && lun->lun_devid) {
9954 		memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
9955 		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9956 		    lun->lun_devid->len);
9957 	}
9958 
9959 	/*
9960 	 * This is for the WWPN which is a port association.
9961 	 */
9962 	if (port->port_devid) {
9963 		memcpy(desc, port->port_devid->data, port->port_devid->len);
9964 		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9965 		    port->port_devid->len);
9966 	}
9967 
9968 	/*
9969 	 * This is for the Relative Target Port(type 4h) identifier
9970 	 */
9971 	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9972 	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9973 	    SVPD_ID_TYPE_RELTARG;
9974 	desc->length = 4;
9975 	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
9976 	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9977 	    sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9978 
9979 	/*
9980 	 * This is for the Target Port Group(type 5h) identifier
9981 	 */
9982 	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9983 	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9984 	    SVPD_ID_TYPE_TPORTGRP;
9985 	desc->length = 4;
9986 	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS + 1,
9987 	    &desc->identifier[2]);
9988 	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9989 	    sizeof(struct scsi_vpd_id_trgt_port_grp_id));
9990 
9991 	/*
9992 	 * This is for the Target identifier
9993 	 */
9994 	if (port->target_devid) {
9995 		memcpy(desc, port->target_devid->data, port->target_devid->len);
9996 	}
9997 
9998 	ctsio->scsi_status = SCSI_STATUS_OK;
9999 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10000 	ctsio->be_move_done = ctl_config_move_done;
10001 	ctl_datamove((union ctl_io *)ctsio);
10002 
10003 	return (CTL_RETVAL_COMPLETE);
10004 }
10005 
10006 static int
10007 ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
10008 {
10009 	struct ctl_softc *softc = control_softc;
10010 	struct scsi_vpd_scsi_ports *sp;
10011 	struct scsi_vpd_port_designation *pd;
10012 	struct scsi_vpd_port_designation_cont *pdc;
10013 	struct ctl_lun *lun;
10014 	struct ctl_port *port;
10015 	int data_len, num_target_ports, iid_len, id_len, g, pg, p;
10016 	int num_target_port_groups, single;
10017 
10018 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10019 
10020 	single = ctl_is_single;
10021 	if (single)
10022 		num_target_port_groups = 1;
10023 	else
10024 		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
10025 	num_target_ports = 0;
10026 	iid_len = 0;
10027 	id_len = 0;
10028 	mtx_lock(&softc->ctl_lock);
10029 	STAILQ_FOREACH(port, &softc->port_list, links) {
10030 		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
10031 			continue;
10032 		if (lun != NULL &&
10033 		    ctl_map_lun_back(port->targ_port, lun->lun) >=
10034 		    CTL_MAX_LUNS)
10035 			continue;
10036 		num_target_ports++;
10037 		if (port->init_devid)
10038 			iid_len += port->init_devid->len;
10039 		if (port->port_devid)
10040 			id_len += port->port_devid->len;
10041 	}
10042 	mtx_unlock(&softc->ctl_lock);
10043 
10044 	data_len = sizeof(struct scsi_vpd_scsi_ports) + num_target_port_groups *
10045 	    num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
10046 	     sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
10047 	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10048 	sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
10049 	ctsio->kern_sg_entries = 0;
10050 
10051 	if (data_len < alloc_len) {
10052 		ctsio->residual = alloc_len - data_len;
10053 		ctsio->kern_data_len = data_len;
10054 		ctsio->kern_total_len = data_len;
10055 	} else {
10056 		ctsio->residual = 0;
10057 		ctsio->kern_data_len = alloc_len;
10058 		ctsio->kern_total_len = alloc_len;
10059 	}
10060 	ctsio->kern_data_resid = 0;
10061 	ctsio->kern_rel_offset = 0;
10062 	ctsio->kern_sg_entries = 0;
10063 
10064 	/*
10065 	 * The control device is always connected.  The disk device, on the
10066 	 * other hand, may not be online all the time.  Need to change this
10067 	 * to figure out whether the disk device is actually online or not.
10068 	 */
10069 	if (lun != NULL)
10070 		sp->device = (SID_QUAL_LU_CONNECTED << 5) |
10071 				  lun->be_lun->lun_type;
10072 	else
10073 		sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10074 
10075 	sp->page_code = SVPD_SCSI_PORTS;
10076 	scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
10077 	    sp->page_length);
10078 	pd = &sp->design[0];
10079 
10080 	mtx_lock(&softc->ctl_lock);
10081 	if (softc->flags & CTL_FLAG_MASTER_SHELF)
10082 		pg = 0;
10083 	else
10084 		pg = 1;
10085 	for (g = 0; g < num_target_port_groups; g++) {
10086 		STAILQ_FOREACH(port, &softc->port_list, links) {
10087 			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
10088 				continue;
10089 			if (lun != NULL &&
10090 			    ctl_map_lun_back(port->targ_port, lun->lun) >=
10091 			    CTL_MAX_LUNS)
10092 				continue;
10093 			p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
10094 			scsi_ulto2b(p, pd->relative_port_id);
10095 			if (port->init_devid && g == pg) {
10096 				iid_len = port->init_devid->len;
10097 				memcpy(pd->initiator_transportid,
10098 				    port->init_devid->data, port->init_devid->len);
10099 			} else
10100 				iid_len = 0;
10101 			scsi_ulto2b(iid_len, pd->initiator_transportid_length);
10102 			pdc = (struct scsi_vpd_port_designation_cont *)
10103 			    (&pd->initiator_transportid[iid_len]);
10104 			if (port->port_devid && g == pg) {
10105 				id_len = port->port_devid->len;
10106 				memcpy(pdc->target_port_descriptors,
10107 				    port->port_devid->data, port->port_devid->len);
10108 			} else
10109 				id_len = 0;
10110 			scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
10111 			pd = (struct scsi_vpd_port_designation *)
10112 			    ((uint8_t *)pdc->target_port_descriptors + id_len);
10113 		}
10114 	}
10115 	mtx_unlock(&softc->ctl_lock);
10116 
10117 	ctsio->scsi_status = SCSI_STATUS_OK;
10118 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10119 	ctsio->be_move_done = ctl_config_move_done;
10120 	ctl_datamove((union ctl_io *)ctsio);
10121 
10122 	return (CTL_RETVAL_COMPLETE);
10123 }
10124 
10125 static int
10126 ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
10127 {
10128 	struct scsi_vpd_block_limits *bl_ptr;
10129 	struct ctl_lun *lun;
10130 	int bs;
10131 
10132 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10133 
10134 	ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
10135 	bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
10136 	ctsio->kern_sg_entries = 0;
10137 
10138 	if (sizeof(*bl_ptr) < alloc_len) {
10139 		ctsio->residual = alloc_len - sizeof(*bl_ptr);
10140 		ctsio->kern_data_len = sizeof(*bl_ptr);
10141 		ctsio->kern_total_len = sizeof(*bl_ptr);
10142 	} else {
10143 		ctsio->residual = 0;
10144 		ctsio->kern_data_len = alloc_len;
10145 		ctsio->kern_total_len = alloc_len;
10146 	}
10147 	ctsio->kern_data_resid = 0;
10148 	ctsio->kern_rel_offset = 0;
10149 	ctsio->kern_sg_entries = 0;
10150 
10151 	/*
10152 	 * The control device is always connected.  The disk device, on the
10153 	 * other hand, may not be online all the time.  Need to change this
10154 	 * to figure out whether the disk device is actually online or not.
10155 	 */
10156 	if (lun != NULL)
10157 		bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10158 				  lun->be_lun->lun_type;
10159 	else
10160 		bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10161 
10162 	bl_ptr->page_code = SVPD_BLOCK_LIMITS;
10163 	scsi_ulto2b(sizeof(*bl_ptr), bl_ptr->page_length);
10164 	bl_ptr->max_cmp_write_len = 0xff;
10165 	scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
10166 	if (lun != NULL) {
10167 		bs = lun->be_lun->blocksize;
10168 		scsi_ulto4b(MAXPHYS / bs, bl_ptr->opt_txfer_len);
10169 		if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10170 			scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_lba_cnt);
10171 			scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_blk_cnt);
10172 			if (lun->be_lun->pblockexp != 0) {
10173 				scsi_ulto4b((1 << lun->be_lun->pblockexp),
10174 				    bl_ptr->opt_unmap_grain);
10175 				scsi_ulto4b(0x80000000 | lun->be_lun->pblockoff,
10176 				    bl_ptr->unmap_grain_align);
10177 			}
10178 		}
10179 	}
10180 	scsi_u64to8b(UINT64_MAX, bl_ptr->max_write_same_length);
10181 
10182 	ctsio->scsi_status = SCSI_STATUS_OK;
10183 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10184 	ctsio->be_move_done = ctl_config_move_done;
10185 	ctl_datamove((union ctl_io *)ctsio);
10186 
10187 	return (CTL_RETVAL_COMPLETE);
10188 }
10189 
10190 static int
10191 ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len)
10192 {
10193 	struct scsi_vpd_block_device_characteristics *bdc_ptr;
10194 	struct ctl_lun *lun;
10195 
10196 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10197 
10198 	ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO);
10199 	bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr;
10200 	ctsio->kern_sg_entries = 0;
10201 
10202 	if (sizeof(*bdc_ptr) < alloc_len) {
10203 		ctsio->residual = alloc_len - sizeof(*bdc_ptr);
10204 		ctsio->kern_data_len = sizeof(*bdc_ptr);
10205 		ctsio->kern_total_len = sizeof(*bdc_ptr);
10206 	} else {
10207 		ctsio->residual = 0;
10208 		ctsio->kern_data_len = alloc_len;
10209 		ctsio->kern_total_len = alloc_len;
10210 	}
10211 	ctsio->kern_data_resid = 0;
10212 	ctsio->kern_rel_offset = 0;
10213 	ctsio->kern_sg_entries = 0;
10214 
10215 	/*
10216 	 * The control device is always connected.  The disk device, on the
10217 	 * other hand, may not be online all the time.  Need to change this
10218 	 * to figure out whether the disk device is actually online or not.
10219 	 */
10220 	if (lun != NULL)
10221 		bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10222 				  lun->be_lun->lun_type;
10223 	else
10224 		bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10225 	bdc_ptr->page_code = SVPD_BDC;
10226 	scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length);
10227 	scsi_ulto2b(SVPD_NON_ROTATING, bdc_ptr->medium_rotation_rate);
10228 	bdc_ptr->flags = SVPD_FUAB | SVPD_VBULS;
10229 
10230 	ctsio->scsi_status = SCSI_STATUS_OK;
10231 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10232 	ctsio->be_move_done = ctl_config_move_done;
10233 	ctl_datamove((union ctl_io *)ctsio);
10234 
10235 	return (CTL_RETVAL_COMPLETE);
10236 }
10237 
10238 static int
10239 ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
10240 {
10241 	struct scsi_vpd_logical_block_prov *lbp_ptr;
10242 	struct ctl_lun *lun;
10243 
10244 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10245 
10246 	ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
10247 	lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
10248 	ctsio->kern_sg_entries = 0;
10249 
10250 	if (sizeof(*lbp_ptr) < alloc_len) {
10251 		ctsio->residual = alloc_len - sizeof(*lbp_ptr);
10252 		ctsio->kern_data_len = sizeof(*lbp_ptr);
10253 		ctsio->kern_total_len = sizeof(*lbp_ptr);
10254 	} else {
10255 		ctsio->residual = 0;
10256 		ctsio->kern_data_len = alloc_len;
10257 		ctsio->kern_total_len = alloc_len;
10258 	}
10259 	ctsio->kern_data_resid = 0;
10260 	ctsio->kern_rel_offset = 0;
10261 	ctsio->kern_sg_entries = 0;
10262 
10263 	/*
10264 	 * The control device is always connected.  The disk device, on the
10265 	 * other hand, may not be online all the time.  Need to change this
10266 	 * to figure out whether the disk device is actually online or not.
10267 	 */
10268 	if (lun != NULL)
10269 		lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10270 				  lun->be_lun->lun_type;
10271 	else
10272 		lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10273 
10274 	lbp_ptr->page_code = SVPD_LBP;
10275 	scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length);
10276 	if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10277 		lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 |
10278 		    SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP;
10279 		lbp_ptr->prov_type = SVPD_LBP_RESOURCE;
10280 	}
10281 
10282 	ctsio->scsi_status = SCSI_STATUS_OK;
10283 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10284 	ctsio->be_move_done = ctl_config_move_done;
10285 	ctl_datamove((union ctl_io *)ctsio);
10286 
10287 	return (CTL_RETVAL_COMPLETE);
10288 }
10289 
10290 static int
10291 ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
10292 {
10293 	struct scsi_inquiry *cdb;
10294 	struct ctl_lun *lun;
10295 	int alloc_len, retval;
10296 
10297 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10298 	cdb = (struct scsi_inquiry *)ctsio->cdb;
10299 
10300 	retval = CTL_RETVAL_COMPLETE;
10301 
10302 	alloc_len = scsi_2btoul(cdb->length);
10303 
10304 	switch (cdb->page_code) {
10305 	case SVPD_SUPPORTED_PAGES:
10306 		retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
10307 		break;
10308 	case SVPD_UNIT_SERIAL_NUMBER:
10309 		retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
10310 		break;
10311 	case SVPD_DEVICE_ID:
10312 		retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
10313 		break;
10314 	case SVPD_SCSI_PORTS:
10315 		retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
10316 		break;
10317 	case SVPD_SCSI_TPC:
10318 		retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
10319 		break;
10320 	case SVPD_BLOCK_LIMITS:
10321 		retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
10322 		break;
10323 	case SVPD_BDC:
10324 		retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len);
10325 		break;
10326 	case SVPD_LBP:
10327 		retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
10328 		break;
10329 	default:
10330 		ctl_set_invalid_field(ctsio,
10331 				      /*sks_valid*/ 1,
10332 				      /*command*/ 1,
10333 				      /*field*/ 2,
10334 				      /*bit_valid*/ 0,
10335 				      /*bit*/ 0);
10336 		ctl_done((union ctl_io *)ctsio);
10337 		retval = CTL_RETVAL_COMPLETE;
10338 		break;
10339 	}
10340 
10341 	return (retval);
10342 }
10343 
10344 static int
10345 ctl_inquiry_std(struct ctl_scsiio *ctsio)
10346 {
10347 	struct scsi_inquiry_data *inq_ptr;
10348 	struct scsi_inquiry *cdb;
10349 	struct ctl_softc *ctl_softc;
10350 	struct ctl_lun *lun;
10351 	char *val;
10352 	uint32_t alloc_len;
10353 	ctl_port_type port_type;
10354 
10355 	ctl_softc = control_softc;
10356 
10357 	/*
10358 	 * Figure out whether we're talking to a Fibre Channel port or not.
10359 	 * We treat the ioctl front end, and any SCSI adapters, as packetized
10360 	 * SCSI front ends.
10361 	 */
10362 	port_type = ctl_softc->ctl_ports[
10363 	    ctl_port_idx(ctsio->io_hdr.nexus.targ_port)]->port_type;
10364 	if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
10365 		port_type = CTL_PORT_SCSI;
10366 
10367 	lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10368 	cdb = (struct scsi_inquiry *)ctsio->cdb;
10369 	alloc_len = scsi_2btoul(cdb->length);
10370 
10371 	/*
10372 	 * We malloc the full inquiry data size here and fill it
10373 	 * in.  If the user only asks for less, we'll give him
10374 	 * that much.
10375 	 */
10376 	ctsio->kern_data_ptr = malloc(sizeof(*inq_ptr), M_CTL, M_WAITOK | M_ZERO);
10377 	inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
10378 	ctsio->kern_sg_entries = 0;
10379 	ctsio->kern_data_resid = 0;
10380 	ctsio->kern_rel_offset = 0;
10381 
10382 	if (sizeof(*inq_ptr) < alloc_len) {
10383 		ctsio->residual = alloc_len - sizeof(*inq_ptr);
10384 		ctsio->kern_data_len = sizeof(*inq_ptr);
10385 		ctsio->kern_total_len = sizeof(*inq_ptr);
10386 	} else {
10387 		ctsio->residual = 0;
10388 		ctsio->kern_data_len = alloc_len;
10389 		ctsio->kern_total_len = alloc_len;
10390 	}
10391 
10392 	/*
10393 	 * If we have a LUN configured, report it as connected.  Otherwise,
10394 	 * report that it is offline or no device is supported, depending
10395 	 * on the value of inquiry_pq_no_lun.
10396 	 *
10397 	 * According to the spec (SPC-4 r34), the peripheral qualifier
10398 	 * SID_QUAL_LU_OFFLINE (001b) is used in the following scenario:
10399 	 *
10400 	 * "A peripheral device having the specified peripheral device type
10401 	 * is not connected to this logical unit. However, the device
10402 	 * server is capable of supporting the specified peripheral device
10403 	 * type on this logical unit."
10404 	 *
10405 	 * According to the same spec, the peripheral qualifier
10406 	 * SID_QUAL_BAD_LU (011b) is used in this scenario:
10407 	 *
10408 	 * "The device server is not capable of supporting a peripheral
10409 	 * device on this logical unit. For this peripheral qualifier the
10410 	 * peripheral device type shall be set to 1Fh. All other peripheral
10411 	 * device type values are reserved for this peripheral qualifier."
10412 	 *
10413 	 * Given the text, it would seem that we probably want to report that
10414 	 * the LUN is offline here.  There is no LUN connected, but we can
10415 	 * support a LUN at the given LUN number.
10416 	 *
10417 	 * In the real world, though, it sounds like things are a little
10418 	 * different:
10419 	 *
10420 	 * - Linux, when presented with a LUN with the offline peripheral
10421 	 *   qualifier, will create an sg driver instance for it.  So when
10422 	 *   you attach it to CTL, you wind up with a ton of sg driver
10423 	 *   instances.  (One for every LUN that Linux bothered to probe.)
10424 	 *   Linux does this despite the fact that it issues a REPORT LUNs
10425 	 *   to LUN 0 to get the inventory of supported LUNs.
10426 	 *
10427 	 * - There is other anecdotal evidence (from Emulex folks) about
10428 	 *   arrays that use the offline peripheral qualifier for LUNs that
10429 	 *   are on the "passive" path in an active/passive array.
10430 	 *
10431 	 * So the solution is provide a hopefully reasonable default
10432 	 * (return bad/no LUN) and allow the user to change the behavior
10433 	 * with a tunable/sysctl variable.
10434 	 */
10435 	if (lun != NULL)
10436 		inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10437 				  lun->be_lun->lun_type;
10438 	else if (ctl_softc->inquiry_pq_no_lun == 0)
10439 		inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10440 	else
10441 		inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10442 
10443 	/* RMB in byte 2 is 0 */
10444 	inq_ptr->version = SCSI_REV_SPC4;
10445 
10446 	/*
10447 	 * According to SAM-3, even if a device only supports a single
10448 	 * level of LUN addressing, it should still set the HISUP bit:
10449 	 *
10450 	 * 4.9.1 Logical unit numbers overview
10451 	 *
10452 	 * All logical unit number formats described in this standard are
10453 	 * hierarchical in structure even when only a single level in that
10454 	 * hierarchy is used. The HISUP bit shall be set to one in the
10455 	 * standard INQUIRY data (see SPC-2) when any logical unit number
10456 	 * format described in this standard is used.  Non-hierarchical
10457 	 * formats are outside the scope of this standard.
10458 	 *
10459 	 * Therefore we set the HiSup bit here.
10460 	 *
10461 	 * The reponse format is 2, per SPC-3.
10462 	 */
10463 	inq_ptr->response_format = SID_HiSup | 2;
10464 
10465 	inq_ptr->additional_length =
10466 	    offsetof(struct scsi_inquiry_data, vendor_specific1) -
10467 	    (offsetof(struct scsi_inquiry_data, additional_length) + 1);
10468 	CTL_DEBUG_PRINT(("additional_length = %d\n",
10469 			 inq_ptr->additional_length));
10470 
10471 	inq_ptr->spc3_flags = SPC3_SID_3PC;
10472 	if (!ctl_is_single)
10473 		inq_ptr->spc3_flags |= SPC3_SID_TPGS_IMPLICIT;
10474 	/* 16 bit addressing */
10475 	if (port_type == CTL_PORT_SCSI)
10476 		inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10477 	/* XXX set the SID_MultiP bit here if we're actually going to
10478 	   respond on multiple ports */
10479 	inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10480 
10481 	/* 16 bit data bus, synchronous transfers */
10482 	if (port_type == CTL_PORT_SCSI)
10483 		inq_ptr->flags = SID_WBus16 | SID_Sync;
10484 	/*
10485 	 * XXX KDM do we want to support tagged queueing on the control
10486 	 * device at all?
10487 	 */
10488 	if ((lun == NULL)
10489 	 || (lun->be_lun->lun_type != T_PROCESSOR))
10490 		inq_ptr->flags |= SID_CmdQue;
10491 	/*
10492 	 * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10493 	 * We have 8 bytes for the vendor name, and 16 bytes for the device
10494 	 * name and 4 bytes for the revision.
10495 	 */
10496 	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10497 	    "vendor")) == NULL) {
10498 		strcpy(inq_ptr->vendor, CTL_VENDOR);
10499 	} else {
10500 		memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10501 		strncpy(inq_ptr->vendor, val,
10502 		    min(sizeof(inq_ptr->vendor), strlen(val)));
10503 	}
10504 	if (lun == NULL) {
10505 		strcpy(inq_ptr->product, CTL_DIRECT_PRODUCT);
10506 	} else if ((val = ctl_get_opt(&lun->be_lun->options, "product")) == NULL) {
10507 		switch (lun->be_lun->lun_type) {
10508 		case T_DIRECT:
10509 			strcpy(inq_ptr->product, CTL_DIRECT_PRODUCT);
10510 			break;
10511 		case T_PROCESSOR:
10512 			strcpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT);
10513 			break;
10514 		default:
10515 			strcpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT);
10516 			break;
10517 		}
10518 	} else {
10519 		memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10520 		strncpy(inq_ptr->product, val,
10521 		    min(sizeof(inq_ptr->product), strlen(val)));
10522 	}
10523 
10524 	/*
10525 	 * XXX make this a macro somewhere so it automatically gets
10526 	 * incremented when we make changes.
10527 	 */
10528 	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10529 	    "revision")) == NULL) {
10530 		strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10531 	} else {
10532 		memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10533 		strncpy(inq_ptr->revision, val,
10534 		    min(sizeof(inq_ptr->revision), strlen(val)));
10535 	}
10536 
10537 	/*
10538 	 * For parallel SCSI, we support double transition and single
10539 	 * transition clocking.  We also support QAS (Quick Arbitration
10540 	 * and Selection) and Information Unit transfers on both the
10541 	 * control and array devices.
10542 	 */
10543 	if (port_type == CTL_PORT_SCSI)
10544 		inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10545 				    SID_SPI_IUS;
10546 
10547 	/* SAM-5 (no version claimed) */
10548 	scsi_ulto2b(0x00A0, inq_ptr->version1);
10549 	/* SPC-4 (no version claimed) */
10550 	scsi_ulto2b(0x0460, inq_ptr->version2);
10551 	if (port_type == CTL_PORT_FC) {
10552 		/* FCP-2 ANSI INCITS.350:2003 */
10553 		scsi_ulto2b(0x0917, inq_ptr->version3);
10554 	} else if (port_type == CTL_PORT_SCSI) {
10555 		/* SPI-4 ANSI INCITS.362:200x */
10556 		scsi_ulto2b(0x0B56, inq_ptr->version3);
10557 	} else if (port_type == CTL_PORT_ISCSI) {
10558 		/* iSCSI (no version claimed) */
10559 		scsi_ulto2b(0x0960, inq_ptr->version3);
10560 	} else if (port_type == CTL_PORT_SAS) {
10561 		/* SAS (no version claimed) */
10562 		scsi_ulto2b(0x0BE0, inq_ptr->version3);
10563 	}
10564 
10565 	if (lun == NULL) {
10566 		/* SBC-3 (no version claimed) */
10567 		scsi_ulto2b(0x04C0, inq_ptr->version4);
10568 	} else {
10569 		switch (lun->be_lun->lun_type) {
10570 		case T_DIRECT:
10571 			/* SBC-3 (no version claimed) */
10572 			scsi_ulto2b(0x04C0, inq_ptr->version4);
10573 			break;
10574 		case T_PROCESSOR:
10575 		default:
10576 			break;
10577 		}
10578 	}
10579 
10580 	ctsio->scsi_status = SCSI_STATUS_OK;
10581 	if (ctsio->kern_data_len > 0) {
10582 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10583 		ctsio->be_move_done = ctl_config_move_done;
10584 		ctl_datamove((union ctl_io *)ctsio);
10585 	} else {
10586 		ctsio->io_hdr.status = CTL_SUCCESS;
10587 		ctl_done((union ctl_io *)ctsio);
10588 	}
10589 
10590 	return (CTL_RETVAL_COMPLETE);
10591 }
10592 
10593 int
10594 ctl_inquiry(struct ctl_scsiio *ctsio)
10595 {
10596 	struct scsi_inquiry *cdb;
10597 	int retval;
10598 
10599 	cdb = (struct scsi_inquiry *)ctsio->cdb;
10600 
10601 	retval = 0;
10602 
10603 	CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10604 
10605 	/*
10606 	 * Right now, we don't support the CmdDt inquiry information.
10607 	 * This would be nice to support in the future.  When we do
10608 	 * support it, we should change this test so that it checks to make
10609 	 * sure SI_EVPD and SI_CMDDT aren't both set at the same time.
10610 	 */
10611 #ifdef notyet
10612 	if (((cdb->byte2 & SI_EVPD)
10613 	 && (cdb->byte2 & SI_CMDDT)))
10614 #endif
10615 	if (cdb->byte2 & SI_CMDDT) {
10616 		/*
10617 		 * Point to the SI_CMDDT bit.  We might change this
10618 		 * when we support SI_CMDDT, but since both bits would be
10619 		 * "wrong", this should probably just stay as-is then.
10620 		 */
10621 		ctl_set_invalid_field(ctsio,
10622 				      /*sks_valid*/ 1,
10623 				      /*command*/ 1,
10624 				      /*field*/ 1,
10625 				      /*bit_valid*/ 1,
10626 				      /*bit*/ 1);
10627 		ctl_done((union ctl_io *)ctsio);
10628 		return (CTL_RETVAL_COMPLETE);
10629 	}
10630 	if (cdb->byte2 & SI_EVPD)
10631 		retval = ctl_inquiry_evpd(ctsio);
10632 #ifdef notyet
10633 	else if (cdb->byte2 & SI_CMDDT)
10634 		retval = ctl_inquiry_cmddt(ctsio);
10635 #endif
10636 	else
10637 		retval = ctl_inquiry_std(ctsio);
10638 
10639 	return (retval);
10640 }
10641 
10642 /*
10643  * For known CDB types, parse the LBA and length.
10644  */
10645 static int
10646 ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint32_t *len)
10647 {
10648 	if (io->io_hdr.io_type != CTL_IO_SCSI)
10649 		return (1);
10650 
10651 	switch (io->scsiio.cdb[0]) {
10652 	case COMPARE_AND_WRITE: {
10653 		struct scsi_compare_and_write *cdb;
10654 
10655 		cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10656 
10657 		*lba = scsi_8btou64(cdb->addr);
10658 		*len = cdb->length;
10659 		break;
10660 	}
10661 	case READ_6:
10662 	case WRITE_6: {
10663 		struct scsi_rw_6 *cdb;
10664 
10665 		cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10666 
10667 		*lba = scsi_3btoul(cdb->addr);
10668 		/* only 5 bits are valid in the most significant address byte */
10669 		*lba &= 0x1fffff;
10670 		*len = cdb->length;
10671 		break;
10672 	}
10673 	case READ_10:
10674 	case WRITE_10: {
10675 		struct scsi_rw_10 *cdb;
10676 
10677 		cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10678 
10679 		*lba = scsi_4btoul(cdb->addr);
10680 		*len = scsi_2btoul(cdb->length);
10681 		break;
10682 	}
10683 	case WRITE_VERIFY_10: {
10684 		struct scsi_write_verify_10 *cdb;
10685 
10686 		cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10687 
10688 		*lba = scsi_4btoul(cdb->addr);
10689 		*len = scsi_2btoul(cdb->length);
10690 		break;
10691 	}
10692 	case READ_12:
10693 	case WRITE_12: {
10694 		struct scsi_rw_12 *cdb;
10695 
10696 		cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10697 
10698 		*lba = scsi_4btoul(cdb->addr);
10699 		*len = scsi_4btoul(cdb->length);
10700 		break;
10701 	}
10702 	case WRITE_VERIFY_12: {
10703 		struct scsi_write_verify_12 *cdb;
10704 
10705 		cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10706 
10707 		*lba = scsi_4btoul(cdb->addr);
10708 		*len = scsi_4btoul(cdb->length);
10709 		break;
10710 	}
10711 	case READ_16:
10712 	case WRITE_16: {
10713 		struct scsi_rw_16 *cdb;
10714 
10715 		cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10716 
10717 		*lba = scsi_8btou64(cdb->addr);
10718 		*len = scsi_4btoul(cdb->length);
10719 		break;
10720 	}
10721 	case WRITE_VERIFY_16: {
10722 		struct scsi_write_verify_16 *cdb;
10723 
10724 		cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10725 
10726 
10727 		*lba = scsi_8btou64(cdb->addr);
10728 		*len = scsi_4btoul(cdb->length);
10729 		break;
10730 	}
10731 	case WRITE_SAME_10: {
10732 		struct scsi_write_same_10 *cdb;
10733 
10734 		cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10735 
10736 		*lba = scsi_4btoul(cdb->addr);
10737 		*len = scsi_2btoul(cdb->length);
10738 		break;
10739 	}
10740 	case WRITE_SAME_16: {
10741 		struct scsi_write_same_16 *cdb;
10742 
10743 		cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10744 
10745 		*lba = scsi_8btou64(cdb->addr);
10746 		*len = scsi_4btoul(cdb->length);
10747 		break;
10748 	}
10749 	case VERIFY_10: {
10750 		struct scsi_verify_10 *cdb;
10751 
10752 		cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10753 
10754 		*lba = scsi_4btoul(cdb->addr);
10755 		*len = scsi_2btoul(cdb->length);
10756 		break;
10757 	}
10758 	case VERIFY_12: {
10759 		struct scsi_verify_12 *cdb;
10760 
10761 		cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10762 
10763 		*lba = scsi_4btoul(cdb->addr);
10764 		*len = scsi_4btoul(cdb->length);
10765 		break;
10766 	}
10767 	case VERIFY_16: {
10768 		struct scsi_verify_16 *cdb;
10769 
10770 		cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10771 
10772 		*lba = scsi_8btou64(cdb->addr);
10773 		*len = scsi_4btoul(cdb->length);
10774 		break;
10775 	}
10776 	default:
10777 		return (1);
10778 		break; /* NOTREACHED */
10779 	}
10780 
10781 	return (0);
10782 }
10783 
10784 static ctl_action
10785 ctl_extent_check_lba(uint64_t lba1, uint32_t len1, uint64_t lba2, uint32_t len2)
10786 {
10787 	uint64_t endlba1, endlba2;
10788 
10789 	endlba1 = lba1 + len1 - 1;
10790 	endlba2 = lba2 + len2 - 1;
10791 
10792 	if ((endlba1 < lba2)
10793 	 || (endlba2 < lba1))
10794 		return (CTL_ACTION_PASS);
10795 	else
10796 		return (CTL_ACTION_BLOCK);
10797 }
10798 
10799 static ctl_action
10800 ctl_extent_check(union ctl_io *io1, union ctl_io *io2)
10801 {
10802 	uint64_t lba1, lba2;
10803 	uint32_t len1, len2;
10804 	int retval;
10805 
10806 	retval = ctl_get_lba_len(io1, &lba1, &len1);
10807 	if (retval != 0)
10808 		return (CTL_ACTION_ERROR);
10809 
10810 	retval = ctl_get_lba_len(io2, &lba2, &len2);
10811 	if (retval != 0)
10812 		return (CTL_ACTION_ERROR);
10813 
10814 	return (ctl_extent_check_lba(lba1, len1, lba2, len2));
10815 }
10816 
10817 static ctl_action
10818 ctl_check_for_blockage(union ctl_io *pending_io, union ctl_io *ooa_io)
10819 {
10820 	const struct ctl_cmd_entry *pending_entry, *ooa_entry;
10821 	ctl_serialize_action *serialize_row;
10822 
10823 	/*
10824 	 * The initiator attempted multiple untagged commands at the same
10825 	 * time.  Can't do that.
10826 	 */
10827 	if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10828 	 && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10829 	 && ((pending_io->io_hdr.nexus.targ_port ==
10830 	      ooa_io->io_hdr.nexus.targ_port)
10831 	  && (pending_io->io_hdr.nexus.initid.id ==
10832 	      ooa_io->io_hdr.nexus.initid.id))
10833 	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10834 		return (CTL_ACTION_OVERLAP);
10835 
10836 	/*
10837 	 * The initiator attempted to send multiple tagged commands with
10838 	 * the same ID.  (It's fine if different initiators have the same
10839 	 * tag ID.)
10840 	 *
10841 	 * Even if all of those conditions are true, we don't kill the I/O
10842 	 * if the command ahead of us has been aborted.  We won't end up
10843 	 * sending it to the FETD, and it's perfectly legal to resend a
10844 	 * command with the same tag number as long as the previous
10845 	 * instance of this tag number has been aborted somehow.
10846 	 */
10847 	if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10848 	 && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10849 	 && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
10850 	 && ((pending_io->io_hdr.nexus.targ_port ==
10851 	      ooa_io->io_hdr.nexus.targ_port)
10852 	  && (pending_io->io_hdr.nexus.initid.id ==
10853 	      ooa_io->io_hdr.nexus.initid.id))
10854 	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10855 		return (CTL_ACTION_OVERLAP_TAG);
10856 
10857 	/*
10858 	 * If we get a head of queue tag, SAM-3 says that we should
10859 	 * immediately execute it.
10860 	 *
10861 	 * What happens if this command would normally block for some other
10862 	 * reason?  e.g. a request sense with a head of queue tag
10863 	 * immediately after a write.  Normally that would block, but this
10864 	 * will result in its getting executed immediately...
10865 	 *
10866 	 * We currently return "pass" instead of "skip", so we'll end up
10867 	 * going through the rest of the queue to check for overlapped tags.
10868 	 *
10869 	 * XXX KDM check for other types of blockage first??
10870 	 */
10871 	if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10872 		return (CTL_ACTION_PASS);
10873 
10874 	/*
10875 	 * Ordered tags have to block until all items ahead of them
10876 	 * have completed.  If we get called with an ordered tag, we always
10877 	 * block, if something else is ahead of us in the queue.
10878 	 */
10879 	if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
10880 		return (CTL_ACTION_BLOCK);
10881 
10882 	/*
10883 	 * Simple tags get blocked until all head of queue and ordered tags
10884 	 * ahead of them have completed.  I'm lumping untagged commands in
10885 	 * with simple tags here.  XXX KDM is that the right thing to do?
10886 	 */
10887 	if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10888 	  || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
10889 	 && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10890 	  || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
10891 		return (CTL_ACTION_BLOCK);
10892 
10893 	pending_entry = ctl_get_cmd_entry(&pending_io->scsiio);
10894 	ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio);
10895 
10896 	serialize_row = ctl_serialize_table[ooa_entry->seridx];
10897 
10898 	switch (serialize_row[pending_entry->seridx]) {
10899 	case CTL_SER_BLOCK:
10900 		return (CTL_ACTION_BLOCK);
10901 		break; /* NOTREACHED */
10902 	case CTL_SER_EXTENT:
10903 		return (ctl_extent_check(pending_io, ooa_io));
10904 		break; /* NOTREACHED */
10905 	case CTL_SER_PASS:
10906 		return (CTL_ACTION_PASS);
10907 		break; /* NOTREACHED */
10908 	case CTL_SER_SKIP:
10909 		return (CTL_ACTION_SKIP);
10910 		break;
10911 	default:
10912 		panic("invalid serialization value %d",
10913 		      serialize_row[pending_entry->seridx]);
10914 		break; /* NOTREACHED */
10915 	}
10916 
10917 	return (CTL_ACTION_ERROR);
10918 }
10919 
10920 /*
10921  * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
10922  * Assumptions:
10923  * - pending_io is generally either incoming, or on the blocked queue
10924  * - starting I/O is the I/O we want to start the check with.
10925  */
10926 static ctl_action
10927 ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
10928 	      union ctl_io *starting_io)
10929 {
10930 	union ctl_io *ooa_io;
10931 	ctl_action action;
10932 
10933 	mtx_assert(&lun->lun_lock, MA_OWNED);
10934 
10935 	/*
10936 	 * Run back along the OOA queue, starting with the current
10937 	 * blocked I/O and going through every I/O before it on the
10938 	 * queue.  If starting_io is NULL, we'll just end up returning
10939 	 * CTL_ACTION_PASS.
10940 	 */
10941 	for (ooa_io = starting_io; ooa_io != NULL;
10942 	     ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
10943 	     ooa_links)){
10944 
10945 		/*
10946 		 * This routine just checks to see whether
10947 		 * cur_blocked is blocked by ooa_io, which is ahead
10948 		 * of it in the queue.  It doesn't queue/dequeue
10949 		 * cur_blocked.
10950 		 */
10951 		action = ctl_check_for_blockage(pending_io, ooa_io);
10952 		switch (action) {
10953 		case CTL_ACTION_BLOCK:
10954 		case CTL_ACTION_OVERLAP:
10955 		case CTL_ACTION_OVERLAP_TAG:
10956 		case CTL_ACTION_SKIP:
10957 		case CTL_ACTION_ERROR:
10958 			return (action);
10959 			break; /* NOTREACHED */
10960 		case CTL_ACTION_PASS:
10961 			break;
10962 		default:
10963 			panic("invalid action %d", action);
10964 			break;  /* NOTREACHED */
10965 		}
10966 	}
10967 
10968 	return (CTL_ACTION_PASS);
10969 }
10970 
10971 /*
10972  * Assumptions:
10973  * - An I/O has just completed, and has been removed from the per-LUN OOA
10974  *   queue, so some items on the blocked queue may now be unblocked.
10975  */
10976 static int
10977 ctl_check_blocked(struct ctl_lun *lun)
10978 {
10979 	union ctl_io *cur_blocked, *next_blocked;
10980 
10981 	mtx_assert(&lun->lun_lock, MA_OWNED);
10982 
10983 	/*
10984 	 * Run forward from the head of the blocked queue, checking each
10985 	 * entry against the I/Os prior to it on the OOA queue to see if
10986 	 * there is still any blockage.
10987 	 *
10988 	 * We cannot use the TAILQ_FOREACH() macro, because it can't deal
10989 	 * with our removing a variable on it while it is traversing the
10990 	 * list.
10991 	 */
10992 	for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
10993 	     cur_blocked != NULL; cur_blocked = next_blocked) {
10994 		union ctl_io *prev_ooa;
10995 		ctl_action action;
10996 
10997 		next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
10998 							  blocked_links);
10999 
11000 		prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
11001 						      ctl_ooaq, ooa_links);
11002 
11003 		/*
11004 		 * If cur_blocked happens to be the first item in the OOA
11005 		 * queue now, prev_ooa will be NULL, and the action
11006 		 * returned will just be CTL_ACTION_PASS.
11007 		 */
11008 		action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
11009 
11010 		switch (action) {
11011 		case CTL_ACTION_BLOCK:
11012 			/* Nothing to do here, still blocked */
11013 			break;
11014 		case CTL_ACTION_OVERLAP:
11015 		case CTL_ACTION_OVERLAP_TAG:
11016 			/*
11017 			 * This shouldn't happen!  In theory we've already
11018 			 * checked this command for overlap...
11019 			 */
11020 			break;
11021 		case CTL_ACTION_PASS:
11022 		case CTL_ACTION_SKIP: {
11023 			struct ctl_softc *softc;
11024 			const struct ctl_cmd_entry *entry;
11025 			uint32_t initidx;
11026 			int isc_retval;
11027 
11028 			/*
11029 			 * The skip case shouldn't happen, this transaction
11030 			 * should have never made it onto the blocked queue.
11031 			 */
11032 			/*
11033 			 * This I/O is no longer blocked, we can remove it
11034 			 * from the blocked queue.  Since this is a TAILQ
11035 			 * (doubly linked list), we can do O(1) removals
11036 			 * from any place on the list.
11037 			 */
11038 			TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
11039 				     blocked_links);
11040 			cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11041 
11042 			if (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC){
11043 				/*
11044 				 * Need to send IO back to original side to
11045 				 * run
11046 				 */
11047 				union ctl_ha_msg msg_info;
11048 
11049 				msg_info.hdr.original_sc =
11050 					cur_blocked->io_hdr.original_sc;
11051 				msg_info.hdr.serializing_sc = cur_blocked;
11052 				msg_info.hdr.msg_type = CTL_MSG_R2R;
11053 				if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11054 				     &msg_info, sizeof(msg_info), 0)) >
11055 				     CTL_HA_STATUS_SUCCESS) {
11056 					printf("CTL:Check Blocked error from "
11057 					       "ctl_ha_msg_send %d\n",
11058 					       isc_retval);
11059 				}
11060 				break;
11061 			}
11062 			entry = ctl_get_cmd_entry(&cur_blocked->scsiio);
11063 			softc = control_softc;
11064 
11065 			initidx = ctl_get_initindex(&cur_blocked->io_hdr.nexus);
11066 
11067 			/*
11068 			 * Check this I/O for LUN state changes that may
11069 			 * have happened while this command was blocked.
11070 			 * The LUN state may have been changed by a command
11071 			 * ahead of us in the queue, so we need to re-check
11072 			 * for any states that can be caused by SCSI
11073 			 * commands.
11074 			 */
11075 			if (ctl_scsiio_lun_check(softc, lun, entry,
11076 						 &cur_blocked->scsiio) == 0) {
11077 				cur_blocked->io_hdr.flags |=
11078 				                      CTL_FLAG_IS_WAS_ON_RTR;
11079 				ctl_enqueue_rtr(cur_blocked);
11080 			} else
11081 				ctl_done(cur_blocked);
11082 			break;
11083 		}
11084 		default:
11085 			/*
11086 			 * This probably shouldn't happen -- we shouldn't
11087 			 * get CTL_ACTION_ERROR, or anything else.
11088 			 */
11089 			break;
11090 		}
11091 	}
11092 
11093 	return (CTL_RETVAL_COMPLETE);
11094 }
11095 
11096 /*
11097  * This routine (with one exception) checks LUN flags that can be set by
11098  * commands ahead of us in the OOA queue.  These flags have to be checked
11099  * when a command initially comes in, and when we pull a command off the
11100  * blocked queue and are preparing to execute it.  The reason we have to
11101  * check these flags for commands on the blocked queue is that the LUN
11102  * state may have been changed by a command ahead of us while we're on the
11103  * blocked queue.
11104  *
11105  * Ordering is somewhat important with these checks, so please pay
11106  * careful attention to the placement of any new checks.
11107  */
11108 static int
11109 ctl_scsiio_lun_check(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
11110     const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
11111 {
11112 	int retval;
11113 
11114 	retval = 0;
11115 
11116 	mtx_assert(&lun->lun_lock, MA_OWNED);
11117 
11118 	/*
11119 	 * If this shelf is a secondary shelf controller, we have to reject
11120 	 * any media access commands.
11121 	 */
11122 #if 0
11123 	/* No longer needed for HA */
11124 	if (((ctl_softc->flags & CTL_FLAG_MASTER_SHELF) == 0)
11125 	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_SECONDARY) == 0)) {
11126 		ctl_set_lun_standby(ctsio);
11127 		retval = 1;
11128 		goto bailout;
11129 	}
11130 #endif
11131 
11132 	/*
11133 	 * Check for a reservation conflict.  If this command isn't allowed
11134 	 * even on reserved LUNs, and if this initiator isn't the one who
11135 	 * reserved us, reject the command with a reservation conflict.
11136 	 */
11137 	if ((lun->flags & CTL_LUN_RESERVED)
11138 	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
11139 		if ((ctsio->io_hdr.nexus.initid.id != lun->rsv_nexus.initid.id)
11140 		 || (ctsio->io_hdr.nexus.targ_port != lun->rsv_nexus.targ_port)
11141 		 || (ctsio->io_hdr.nexus.targ_target.id !=
11142 		     lun->rsv_nexus.targ_target.id)) {
11143 			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
11144 			ctsio->io_hdr.status = CTL_SCSI_ERROR;
11145 			retval = 1;
11146 			goto bailout;
11147 		}
11148 	}
11149 
11150 	if ( (lun->flags & CTL_LUN_PR_RESERVED)
11151 	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV) == 0)) {
11152 		uint32_t residx;
11153 
11154 		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
11155 		/*
11156 		 * if we aren't registered or it's a res holder type
11157 		 * reservation and this isn't the res holder then set a
11158 		 * conflict.
11159 		 * NOTE: Commands which might be allowed on write exclusive
11160 		 * type reservations are checked in the particular command
11161 		 * for a conflict. Read and SSU are the only ones.
11162 		 */
11163 		if (!lun->per_res[residx].registered
11164 		 || (residx != lun->pr_res_idx && lun->res_type < 4)) {
11165 			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
11166 			ctsio->io_hdr.status = CTL_SCSI_ERROR;
11167 			retval = 1;
11168 			goto bailout;
11169 		}
11170 
11171 	}
11172 
11173 	if ((lun->flags & CTL_LUN_OFFLINE)
11174 	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_OFFLINE) == 0)) {
11175 		ctl_set_lun_not_ready(ctsio);
11176 		retval = 1;
11177 		goto bailout;
11178 	}
11179 
11180 	/*
11181 	 * If the LUN is stopped, see if this particular command is allowed
11182 	 * for a stopped lun.  Otherwise, reject it with 0x04,0x02.
11183 	 */
11184 	if ((lun->flags & CTL_LUN_STOPPED)
11185 	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_STOPPED) == 0)) {
11186 		/* "Logical unit not ready, initializing cmd. required" */
11187 		ctl_set_lun_stopped(ctsio);
11188 		retval = 1;
11189 		goto bailout;
11190 	}
11191 
11192 	if ((lun->flags & CTL_LUN_INOPERABLE)
11193 	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_INOPERABLE) == 0)) {
11194 		/* "Medium format corrupted" */
11195 		ctl_set_medium_format_corrupted(ctsio);
11196 		retval = 1;
11197 		goto bailout;
11198 	}
11199 
11200 bailout:
11201 	return (retval);
11202 
11203 }
11204 
11205 static void
11206 ctl_failover_io(union ctl_io *io, int have_lock)
11207 {
11208 	ctl_set_busy(&io->scsiio);
11209 	ctl_done(io);
11210 }
11211 
11212 static void
11213 ctl_failover(void)
11214 {
11215 	struct ctl_lun *lun;
11216 	struct ctl_softc *ctl_softc;
11217 	union ctl_io *next_io, *pending_io;
11218 	union ctl_io *io;
11219 	int lun_idx;
11220 	int i;
11221 
11222 	ctl_softc = control_softc;
11223 
11224 	mtx_lock(&ctl_softc->ctl_lock);
11225 	/*
11226 	 * Remove any cmds from the other SC from the rtr queue.  These
11227 	 * will obviously only be for LUNs for which we're the primary.
11228 	 * We can't send status or get/send data for these commands.
11229 	 * Since they haven't been executed yet, we can just remove them.
11230 	 * We'll either abort them or delete them below, depending on
11231 	 * which HA mode we're in.
11232 	 */
11233 #ifdef notyet
11234 	mtx_lock(&ctl_softc->queue_lock);
11235 	for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->rtr_queue);
11236 	     io != NULL; io = next_io) {
11237 		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
11238 		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
11239 			STAILQ_REMOVE(&ctl_softc->rtr_queue, &io->io_hdr,
11240 				      ctl_io_hdr, links);
11241 	}
11242 	mtx_unlock(&ctl_softc->queue_lock);
11243 #endif
11244 
11245 	for (lun_idx=0; lun_idx < ctl_softc->num_luns; lun_idx++) {
11246 		lun = ctl_softc->ctl_luns[lun_idx];
11247 		if (lun==NULL)
11248 			continue;
11249 
11250 		/*
11251 		 * Processor LUNs are primary on both sides.
11252 		 * XXX will this always be true?
11253 		 */
11254 		if (lun->be_lun->lun_type == T_PROCESSOR)
11255 			continue;
11256 
11257 		if ((lun->flags & CTL_LUN_PRIMARY_SC)
11258 		 && (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11259 			printf("FAILOVER: primary lun %d\n", lun_idx);
11260 		        /*
11261 			 * Remove all commands from the other SC. First from the
11262 			 * blocked queue then from the ooa queue. Once we have
11263 			 * removed them. Call ctl_check_blocked to see if there
11264 			 * is anything that can run.
11265 			 */
11266 			for (io = (union ctl_io *)TAILQ_FIRST(
11267 			     &lun->blocked_queue); io != NULL; io = next_io) {
11268 
11269 		        	next_io = (union ctl_io *)TAILQ_NEXT(
11270 				    &io->io_hdr, blocked_links);
11271 
11272 				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11273 					TAILQ_REMOVE(&lun->blocked_queue,
11274 						     &io->io_hdr,blocked_links);
11275 					io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11276 					TAILQ_REMOVE(&lun->ooa_queue,
11277 						     &io->io_hdr, ooa_links);
11278 
11279 					ctl_free_io(io);
11280 				}
11281 			}
11282 
11283 			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11284 	     		     io != NULL; io = next_io) {
11285 
11286 		        	next_io = (union ctl_io *)TAILQ_NEXT(
11287 				    &io->io_hdr, ooa_links);
11288 
11289 				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11290 
11291 					TAILQ_REMOVE(&lun->ooa_queue,
11292 						&io->io_hdr,
11293 					     	ooa_links);
11294 
11295 					ctl_free_io(io);
11296 				}
11297 			}
11298 			ctl_check_blocked(lun);
11299 		} else if ((lun->flags & CTL_LUN_PRIMARY_SC)
11300 			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
11301 
11302 			printf("FAILOVER: primary lun %d\n", lun_idx);
11303 			/*
11304 			 * Abort all commands from the other SC.  We can't
11305 			 * send status back for them now.  These should get
11306 			 * cleaned up when they are completed or come out
11307 			 * for a datamove operation.
11308 			 */
11309 			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11310 	     		     io != NULL; io = next_io) {
11311 		        	next_io = (union ctl_io *)TAILQ_NEXT(
11312 					&io->io_hdr, ooa_links);
11313 
11314 				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
11315 					io->io_hdr.flags |= CTL_FLAG_ABORT;
11316 			}
11317 		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11318 			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
11319 
11320 			printf("FAILOVER: secondary lun %d\n", lun_idx);
11321 
11322 			lun->flags |= CTL_LUN_PRIMARY_SC;
11323 
11324 			/*
11325 			 * We send all I/O that was sent to this controller
11326 			 * and redirected to the other side back with
11327 			 * busy status, and have the initiator retry it.
11328 			 * Figuring out how much data has been transferred,
11329 			 * etc. and picking up where we left off would be
11330 			 * very tricky.
11331 			 *
11332 			 * XXX KDM need to remove I/O from the blocked
11333 			 * queue as well!
11334 			 */
11335 			for (pending_io = (union ctl_io *)TAILQ_FIRST(
11336 			     &lun->ooa_queue); pending_io != NULL;
11337 			     pending_io = next_io) {
11338 
11339 				next_io =  (union ctl_io *)TAILQ_NEXT(
11340 					&pending_io->io_hdr, ooa_links);
11341 
11342 				pending_io->io_hdr.flags &=
11343 					~CTL_FLAG_SENT_2OTHER_SC;
11344 
11345 				if (pending_io->io_hdr.flags &
11346 				    CTL_FLAG_IO_ACTIVE) {
11347 					pending_io->io_hdr.flags |=
11348 						CTL_FLAG_FAILOVER;
11349 				} else {
11350 					ctl_set_busy(&pending_io->scsiio);
11351 					ctl_done(pending_io);
11352 				}
11353 			}
11354 
11355 			/*
11356 			 * Build Unit Attention
11357 			 */
11358 			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11359 				lun->pending_ua[i] |=
11360 				                     CTL_UA_ASYM_ACC_CHANGE;
11361 			}
11362 		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11363 			&& (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11364 			printf("FAILOVER: secondary lun %d\n", lun_idx);
11365 			/*
11366 			 * if the first io on the OOA is not on the RtR queue
11367 			 * add it.
11368 			 */
11369 			lun->flags |= CTL_LUN_PRIMARY_SC;
11370 
11371 			pending_io = (union ctl_io *)TAILQ_FIRST(
11372 			    &lun->ooa_queue);
11373 			if (pending_io==NULL) {
11374 				printf("Nothing on OOA queue\n");
11375 				continue;
11376 			}
11377 
11378 			pending_io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11379 			if ((pending_io->io_hdr.flags &
11380 			     CTL_FLAG_IS_WAS_ON_RTR) == 0) {
11381 				pending_io->io_hdr.flags |=
11382 				    CTL_FLAG_IS_WAS_ON_RTR;
11383 				ctl_enqueue_rtr(pending_io);
11384 			}
11385 #if 0
11386 			else
11387 			{
11388 				printf("Tag 0x%04x is running\n",
11389 				      pending_io->scsiio.tag_num);
11390 			}
11391 #endif
11392 
11393 			next_io = (union ctl_io *)TAILQ_NEXT(
11394 			    &pending_io->io_hdr, ooa_links);
11395 			for (pending_io=next_io; pending_io != NULL;
11396 			     pending_io = next_io) {
11397 				pending_io->io_hdr.flags &=
11398 				    ~CTL_FLAG_SENT_2OTHER_SC;
11399 				next_io = (union ctl_io *)TAILQ_NEXT(
11400 					&pending_io->io_hdr, ooa_links);
11401 				if (pending_io->io_hdr.flags &
11402 				    CTL_FLAG_IS_WAS_ON_RTR) {
11403 #if 0
11404 				        printf("Tag 0x%04x is running\n",
11405 				      		pending_io->scsiio.tag_num);
11406 #endif
11407 					continue;
11408 				}
11409 
11410 				switch (ctl_check_ooa(lun, pending_io,
11411 			            (union ctl_io *)TAILQ_PREV(
11412 				    &pending_io->io_hdr, ctl_ooaq,
11413 				    ooa_links))) {
11414 
11415 				case CTL_ACTION_BLOCK:
11416 					TAILQ_INSERT_TAIL(&lun->blocked_queue,
11417 							  &pending_io->io_hdr,
11418 							  blocked_links);
11419 					pending_io->io_hdr.flags |=
11420 					    CTL_FLAG_BLOCKED;
11421 					break;
11422 				case CTL_ACTION_PASS:
11423 				case CTL_ACTION_SKIP:
11424 					pending_io->io_hdr.flags |=
11425 					    CTL_FLAG_IS_WAS_ON_RTR;
11426 					ctl_enqueue_rtr(pending_io);
11427 					break;
11428 				case CTL_ACTION_OVERLAP:
11429 					ctl_set_overlapped_cmd(
11430 					    (struct ctl_scsiio *)pending_io);
11431 					ctl_done(pending_io);
11432 					break;
11433 				case CTL_ACTION_OVERLAP_TAG:
11434 					ctl_set_overlapped_tag(
11435 					    (struct ctl_scsiio *)pending_io,
11436 					    pending_io->scsiio.tag_num & 0xff);
11437 					ctl_done(pending_io);
11438 					break;
11439 				case CTL_ACTION_ERROR:
11440 				default:
11441 					ctl_set_internal_failure(
11442 						(struct ctl_scsiio *)pending_io,
11443 						0,  // sks_valid
11444 						0); //retry count
11445 					ctl_done(pending_io);
11446 					break;
11447 				}
11448 			}
11449 
11450 			/*
11451 			 * Build Unit Attention
11452 			 */
11453 			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11454 				lun->pending_ua[i] |=
11455 				                     CTL_UA_ASYM_ACC_CHANGE;
11456 			}
11457 		} else {
11458 			panic("Unhandled HA mode failover, LUN flags = %#x, "
11459 			      "ha_mode = #%x", lun->flags, ctl_softc->ha_mode);
11460 		}
11461 	}
11462 	ctl_pause_rtr = 0;
11463 	mtx_unlock(&ctl_softc->ctl_lock);
11464 }
11465 
11466 static int
11467 ctl_scsiio_precheck(struct ctl_softc *ctl_softc, struct ctl_scsiio *ctsio)
11468 {
11469 	struct ctl_lun *lun;
11470 	const struct ctl_cmd_entry *entry;
11471 	uint32_t initidx, targ_lun;
11472 	int retval;
11473 
11474 	retval = 0;
11475 
11476 	lun = NULL;
11477 
11478 	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11479 	if ((targ_lun < CTL_MAX_LUNS)
11480 	 && (ctl_softc->ctl_luns[targ_lun] != NULL)) {
11481 		lun = ctl_softc->ctl_luns[targ_lun];
11482 		/*
11483 		 * If the LUN is invalid, pretend that it doesn't exist.
11484 		 * It will go away as soon as all pending I/O has been
11485 		 * completed.
11486 		 */
11487 		if (lun->flags & CTL_LUN_DISABLED) {
11488 			lun = NULL;
11489 		} else {
11490 			ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
11491 			ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
11492 				lun->be_lun;
11493 			if (lun->be_lun->lun_type == T_PROCESSOR) {
11494 				ctsio->io_hdr.flags |= CTL_FLAG_CONTROL_DEV;
11495 			}
11496 
11497 			/*
11498 			 * Every I/O goes into the OOA queue for a
11499 			 * particular LUN, and stays there until completion.
11500 			 */
11501 			mtx_lock(&lun->lun_lock);
11502 			TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr,
11503 			    ooa_links);
11504 		}
11505 	} else {
11506 		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11507 		ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11508 	}
11509 
11510 	/* Get command entry and return error if it is unsuppotyed. */
11511 	entry = ctl_validate_command(ctsio);
11512 	if (entry == NULL) {
11513 		if (lun)
11514 			mtx_unlock(&lun->lun_lock);
11515 		return (retval);
11516 	}
11517 
11518 	ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11519 	ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11520 
11521 	/*
11522 	 * Check to see whether we can send this command to LUNs that don't
11523 	 * exist.  This should pretty much only be the case for inquiry
11524 	 * and request sense.  Further checks, below, really require having
11525 	 * a LUN, so we can't really check the command anymore.  Just put
11526 	 * it on the rtr queue.
11527 	 */
11528 	if (lun == NULL) {
11529 		if (entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) {
11530 			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11531 			ctl_enqueue_rtr((union ctl_io *)ctsio);
11532 			return (retval);
11533 		}
11534 
11535 		ctl_set_unsupported_lun(ctsio);
11536 		ctl_done((union ctl_io *)ctsio);
11537 		CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11538 		return (retval);
11539 	} else {
11540 		/*
11541 		 * Make sure we support this particular command on this LUN.
11542 		 * e.g., we don't support writes to the control LUN.
11543 		 */
11544 		if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11545 			mtx_unlock(&lun->lun_lock);
11546 			ctl_set_invalid_opcode(ctsio);
11547 			ctl_done((union ctl_io *)ctsio);
11548 			return (retval);
11549 		}
11550 	}
11551 
11552 	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11553 
11554 #ifdef CTL_WITH_CA
11555 	/*
11556 	 * If we've got a request sense, it'll clear the contingent
11557 	 * allegiance condition.  Otherwise, if we have a CA condition for
11558 	 * this initiator, clear it, because it sent down a command other
11559 	 * than request sense.
11560 	 */
11561 	if ((ctsio->cdb[0] != REQUEST_SENSE)
11562 	 && (ctl_is_set(lun->have_ca, initidx)))
11563 		ctl_clear_mask(lun->have_ca, initidx);
11564 #endif
11565 
11566 	/*
11567 	 * If the command has this flag set, it handles its own unit
11568 	 * attention reporting, we shouldn't do anything.  Otherwise we
11569 	 * check for any pending unit attentions, and send them back to the
11570 	 * initiator.  We only do this when a command initially comes in,
11571 	 * not when we pull it off the blocked queue.
11572 	 *
11573 	 * According to SAM-3, section 5.3.2, the order that things get
11574 	 * presented back to the host is basically unit attentions caused
11575 	 * by some sort of reset event, busy status, reservation conflicts
11576 	 * or task set full, and finally any other status.
11577 	 *
11578 	 * One issue here is that some of the unit attentions we report
11579 	 * don't fall into the "reset" category (e.g. "reported luns data
11580 	 * has changed").  So reporting it here, before the reservation
11581 	 * check, may be technically wrong.  I guess the only thing to do
11582 	 * would be to check for and report the reset events here, and then
11583 	 * check for the other unit attention types after we check for a
11584 	 * reservation conflict.
11585 	 *
11586 	 * XXX KDM need to fix this
11587 	 */
11588 	if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11589 		ctl_ua_type ua_type;
11590 
11591 		ua_type = lun->pending_ua[initidx];
11592 		if (ua_type != CTL_UA_NONE) {
11593 			scsi_sense_data_type sense_format;
11594 
11595 			if (lun != NULL)
11596 				sense_format = (lun->flags &
11597 				    CTL_LUN_SENSE_DESC) ? SSD_TYPE_DESC :
11598 				    SSD_TYPE_FIXED;
11599 			else
11600 				sense_format = SSD_TYPE_FIXED;
11601 
11602 			ua_type = ctl_build_ua(ua_type, &ctsio->sense_data,
11603 					       sense_format);
11604 			if (ua_type != CTL_UA_NONE) {
11605 				ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11606 				ctsio->io_hdr.status = CTL_SCSI_ERROR |
11607 						       CTL_AUTOSENSE;
11608 				ctsio->sense_len = SSD_FULL_SIZE;
11609 				lun->pending_ua[initidx] &= ~ua_type;
11610 				mtx_unlock(&lun->lun_lock);
11611 				ctl_done((union ctl_io *)ctsio);
11612 				return (retval);
11613 			}
11614 		}
11615 	}
11616 
11617 
11618 	if (ctl_scsiio_lun_check(ctl_softc, lun, entry, ctsio) != 0) {
11619 		mtx_unlock(&lun->lun_lock);
11620 		ctl_done((union ctl_io *)ctsio);
11621 		return (retval);
11622 	}
11623 
11624 	/*
11625 	 * XXX CHD this is where we want to send IO to other side if
11626 	 * this LUN is secondary on this SC. We will need to make a copy
11627 	 * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11628 	 * the copy we send as FROM_OTHER.
11629 	 * We also need to stuff the address of the original IO so we can
11630 	 * find it easily. Something similar will need be done on the other
11631 	 * side so when we are done we can find the copy.
11632 	 */
11633 	if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11634 		union ctl_ha_msg msg_info;
11635 		int isc_retval;
11636 
11637 		ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11638 
11639 		msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11640 		msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11641 #if 0
11642 		printf("1. ctsio %p\n", ctsio);
11643 #endif
11644 		msg_info.hdr.serializing_sc = NULL;
11645 		msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11646 		msg_info.scsi.tag_num = ctsio->tag_num;
11647 		msg_info.scsi.tag_type = ctsio->tag_type;
11648 		memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11649 
11650 		ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11651 
11652 		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11653 		    (void *)&msg_info, sizeof(msg_info), 0)) >
11654 		    CTL_HA_STATUS_SUCCESS) {
11655 			printf("CTL:precheck, ctl_ha_msg_send returned %d\n",
11656 			       isc_retval);
11657 			printf("CTL:opcode is %x\n", ctsio->cdb[0]);
11658 		} else {
11659 #if 0
11660 			printf("CTL:Precheck sent msg, opcode is %x\n",opcode);
11661 #endif
11662 		}
11663 
11664 		/*
11665 		 * XXX KDM this I/O is off the incoming queue, but hasn't
11666 		 * been inserted on any other queue.  We may need to come
11667 		 * up with a holding queue while we wait for serialization
11668 		 * so that we have an idea of what we're waiting for from
11669 		 * the other side.
11670 		 */
11671 		mtx_unlock(&lun->lun_lock);
11672 		return (retval);
11673 	}
11674 
11675 	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11676 			      (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11677 			      ctl_ooaq, ooa_links))) {
11678 	case CTL_ACTION_BLOCK:
11679 		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11680 		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11681 				  blocked_links);
11682 		mtx_unlock(&lun->lun_lock);
11683 		return (retval);
11684 	case CTL_ACTION_PASS:
11685 	case CTL_ACTION_SKIP:
11686 		ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11687 		mtx_unlock(&lun->lun_lock);
11688 		ctl_enqueue_rtr((union ctl_io *)ctsio);
11689 		break;
11690 	case CTL_ACTION_OVERLAP:
11691 		mtx_unlock(&lun->lun_lock);
11692 		ctl_set_overlapped_cmd(ctsio);
11693 		ctl_done((union ctl_io *)ctsio);
11694 		break;
11695 	case CTL_ACTION_OVERLAP_TAG:
11696 		mtx_unlock(&lun->lun_lock);
11697 		ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11698 		ctl_done((union ctl_io *)ctsio);
11699 		break;
11700 	case CTL_ACTION_ERROR:
11701 	default:
11702 		mtx_unlock(&lun->lun_lock);
11703 		ctl_set_internal_failure(ctsio,
11704 					 /*sks_valid*/ 0,
11705 					 /*retry_count*/ 0);
11706 		ctl_done((union ctl_io *)ctsio);
11707 		break;
11708 	}
11709 	return (retval);
11710 }
11711 
11712 const struct ctl_cmd_entry *
11713 ctl_get_cmd_entry(struct ctl_scsiio *ctsio)
11714 {
11715 	const struct ctl_cmd_entry *entry;
11716 	int service_action;
11717 
11718 	entry = &ctl_cmd_table[ctsio->cdb[0]];
11719 	if (entry->flags & CTL_CMD_FLAG_SA5) {
11720 		service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11721 		entry = &((const struct ctl_cmd_entry *)
11722 		    entry->execute)[service_action];
11723 	}
11724 	return (entry);
11725 }
11726 
11727 const struct ctl_cmd_entry *
11728 ctl_validate_command(struct ctl_scsiio *ctsio)
11729 {
11730 	const struct ctl_cmd_entry *entry;
11731 	int i;
11732 	uint8_t diff;
11733 
11734 	entry = ctl_get_cmd_entry(ctsio);
11735 	if (entry->execute == NULL) {
11736 		ctl_set_invalid_opcode(ctsio);
11737 		ctl_done((union ctl_io *)ctsio);
11738 		return (NULL);
11739 	}
11740 	KASSERT(entry->length > 0,
11741 	    ("Not defined length for command 0x%02x/0x%02x",
11742 	     ctsio->cdb[0], ctsio->cdb[1]));
11743 	for (i = 1; i < entry->length; i++) {
11744 		diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11745 		if (diff == 0)
11746 			continue;
11747 		ctl_set_invalid_field(ctsio,
11748 				      /*sks_valid*/ 1,
11749 				      /*command*/ 1,
11750 				      /*field*/ i,
11751 				      /*bit_valid*/ 1,
11752 				      /*bit*/ fls(diff) - 1);
11753 		ctl_done((union ctl_io *)ctsio);
11754 		return (NULL);
11755 	}
11756 	return (entry);
11757 }
11758 
11759 static int
11760 ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11761 {
11762 
11763 	switch (lun_type) {
11764 	case T_PROCESSOR:
11765 		if (((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0) &&
11766 		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11767 			return (0);
11768 		break;
11769 	case T_DIRECT:
11770 		if (((entry->flags & CTL_CMD_FLAG_OK_ON_SLUN) == 0) &&
11771 		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11772 			return (0);
11773 		break;
11774 	default:
11775 		return (0);
11776 	}
11777 	return (1);
11778 }
11779 
11780 static int
11781 ctl_scsiio(struct ctl_scsiio *ctsio)
11782 {
11783 	int retval;
11784 	const struct ctl_cmd_entry *entry;
11785 
11786 	retval = CTL_RETVAL_COMPLETE;
11787 
11788 	CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11789 
11790 	entry = ctl_get_cmd_entry(ctsio);
11791 
11792 	/*
11793 	 * If this I/O has been aborted, just send it straight to
11794 	 * ctl_done() without executing it.
11795 	 */
11796 	if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11797 		ctl_done((union ctl_io *)ctsio);
11798 		goto bailout;
11799 	}
11800 
11801 	/*
11802 	 * All the checks should have been handled by ctl_scsiio_precheck().
11803 	 * We should be clear now to just execute the I/O.
11804 	 */
11805 	retval = entry->execute(ctsio);
11806 
11807 bailout:
11808 	return (retval);
11809 }
11810 
11811 /*
11812  * Since we only implement one target right now, a bus reset simply resets
11813  * our single target.
11814  */
11815 static int
11816 ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io)
11817 {
11818 	return(ctl_target_reset(ctl_softc, io, CTL_UA_BUS_RESET));
11819 }
11820 
11821 static int
11822 ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
11823 		 ctl_ua_type ua_type)
11824 {
11825 	struct ctl_lun *lun;
11826 	int retval;
11827 
11828 	if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11829 		union ctl_ha_msg msg_info;
11830 
11831 		io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11832 		msg_info.hdr.nexus = io->io_hdr.nexus;
11833 		if (ua_type==CTL_UA_TARG_RESET)
11834 			msg_info.task.task_action = CTL_TASK_TARGET_RESET;
11835 		else
11836 			msg_info.task.task_action = CTL_TASK_BUS_RESET;
11837 		msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11838 		msg_info.hdr.original_sc = NULL;
11839 		msg_info.hdr.serializing_sc = NULL;
11840 		if (CTL_HA_STATUS_SUCCESS != ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11841 		    (void *)&msg_info, sizeof(msg_info), 0)) {
11842 		}
11843 	}
11844 	retval = 0;
11845 
11846 	mtx_lock(&ctl_softc->ctl_lock);
11847 	STAILQ_FOREACH(lun, &ctl_softc->lun_list, links)
11848 		retval += ctl_lun_reset(lun, io, ua_type);
11849 	mtx_unlock(&ctl_softc->ctl_lock);
11850 
11851 	return (retval);
11852 }
11853 
11854 /*
11855  * The LUN should always be set.  The I/O is optional, and is used to
11856  * distinguish between I/Os sent by this initiator, and by other
11857  * initiators.  We set unit attention for initiators other than this one.
11858  * SAM-3 is vague on this point.  It does say that a unit attention should
11859  * be established for other initiators when a LUN is reset (see section
11860  * 5.7.3), but it doesn't specifically say that the unit attention should
11861  * be established for this particular initiator when a LUN is reset.  Here
11862  * is the relevant text, from SAM-3 rev 8:
11863  *
11864  * 5.7.2 When a SCSI initiator port aborts its own tasks
11865  *
11866  * When a SCSI initiator port causes its own task(s) to be aborted, no
11867  * notification that the task(s) have been aborted shall be returned to
11868  * the SCSI initiator port other than the completion response for the
11869  * command or task management function action that caused the task(s) to
11870  * be aborted and notification(s) associated with related effects of the
11871  * action (e.g., a reset unit attention condition).
11872  *
11873  * XXX KDM for now, we're setting unit attention for all initiators.
11874  */
11875 static int
11876 ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
11877 {
11878 	union ctl_io *xio;
11879 #if 0
11880 	uint32_t initindex;
11881 #endif
11882 	int i;
11883 
11884 	mtx_lock(&lun->lun_lock);
11885 	/*
11886 	 * Run through the OOA queue and abort each I/O.
11887 	 */
11888 #if 0
11889 	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
11890 #endif
11891 	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11892 	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11893 		xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
11894 	}
11895 
11896 	/*
11897 	 * This version sets unit attention for every
11898 	 */
11899 #if 0
11900 	initindex = ctl_get_initindex(&io->io_hdr.nexus);
11901 	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11902 		if (initindex == i)
11903 			continue;
11904 		lun->pending_ua[i] |= ua_type;
11905 	}
11906 #endif
11907 
11908 	/*
11909 	 * A reset (any kind, really) clears reservations established with
11910 	 * RESERVE/RELEASE.  It does not clear reservations established
11911 	 * with PERSISTENT RESERVE OUT, but we don't support that at the
11912 	 * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
11913 	 * reservations made with the RESERVE/RELEASE commands, because
11914 	 * those commands are obsolete in SPC-3.
11915 	 */
11916 	lun->flags &= ~CTL_LUN_RESERVED;
11917 
11918 	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11919 #ifdef CTL_WITH_CA
11920 		ctl_clear_mask(lun->have_ca, i);
11921 #endif
11922 		lun->pending_ua[i] |= ua_type;
11923 	}
11924 	mtx_unlock(&lun->lun_lock);
11925 
11926 	return (0);
11927 }
11928 
11929 static int
11930 ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
11931     int other_sc)
11932 {
11933 	union ctl_io *xio;
11934 	int found;
11935 
11936 	mtx_assert(&lun->lun_lock, MA_OWNED);
11937 
11938 	/*
11939 	 * Run through the OOA queue and attempt to find the given I/O.
11940 	 * The target port, initiator ID, tag type and tag number have to
11941 	 * match the values that we got from the initiator.  If we have an
11942 	 * untagged command to abort, simply abort the first untagged command
11943 	 * we come to.  We only allow one untagged command at a time of course.
11944 	 */
11945 	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11946 	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11947 
11948 		if ((targ_port == UINT32_MAX ||
11949 		     targ_port == xio->io_hdr.nexus.targ_port) &&
11950 		    (init_id == UINT32_MAX ||
11951 		     init_id == xio->io_hdr.nexus.initid.id)) {
11952 			if (targ_port != xio->io_hdr.nexus.targ_port ||
11953 			    init_id != xio->io_hdr.nexus.initid.id)
11954 				xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS;
11955 			xio->io_hdr.flags |= CTL_FLAG_ABORT;
11956 			found = 1;
11957 			if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11958 				union ctl_ha_msg msg_info;
11959 
11960 				msg_info.hdr.nexus = xio->io_hdr.nexus;
11961 				msg_info.task.task_action = CTL_TASK_ABORT_TASK;
11962 				msg_info.task.tag_num = xio->scsiio.tag_num;
11963 				msg_info.task.tag_type = xio->scsiio.tag_type;
11964 				msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11965 				msg_info.hdr.original_sc = NULL;
11966 				msg_info.hdr.serializing_sc = NULL;
11967 				ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11968 				    (void *)&msg_info, sizeof(msg_info), 0);
11969 			}
11970 		}
11971 	}
11972 	return (found);
11973 }
11974 
11975 static int
11976 ctl_abort_task_set(union ctl_io *io)
11977 {
11978 	struct ctl_softc *softc = control_softc;
11979 	struct ctl_lun *lun;
11980 	uint32_t targ_lun;
11981 
11982 	/*
11983 	 * Look up the LUN.
11984 	 */
11985 	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11986 	mtx_lock(&softc->ctl_lock);
11987 	if ((targ_lun < CTL_MAX_LUNS) && (softc->ctl_luns[targ_lun] != NULL))
11988 		lun = softc->ctl_luns[targ_lun];
11989 	else {
11990 		mtx_unlock(&softc->ctl_lock);
11991 		return (1);
11992 	}
11993 
11994 	mtx_lock(&lun->lun_lock);
11995 	mtx_unlock(&softc->ctl_lock);
11996 	if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
11997 		ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
11998 		    io->io_hdr.nexus.initid.id,
11999 		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12000 	} else { /* CTL_TASK_CLEAR_TASK_SET */
12001 		ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
12002 		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12003 	}
12004 	mtx_unlock(&lun->lun_lock);
12005 	return (0);
12006 }
12007 
12008 static int
12009 ctl_i_t_nexus_reset(union ctl_io *io)
12010 {
12011 	struct ctl_softc *softc = control_softc;
12012 	struct ctl_lun *lun;
12013 	uint32_t initindex;
12014 
12015 	initindex = ctl_get_initindex(&io->io_hdr.nexus);
12016 	mtx_lock(&softc->ctl_lock);
12017 	STAILQ_FOREACH(lun, &softc->lun_list, links) {
12018 		mtx_lock(&lun->lun_lock);
12019 		ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12020 		    io->io_hdr.nexus.initid.id,
12021 		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12022 #ifdef CTL_WITH_CA
12023 		ctl_clear_mask(lun->have_ca, initindex);
12024 #endif
12025 		lun->pending_ua[initindex] |= CTL_UA_I_T_NEXUS_LOSS;
12026 		mtx_unlock(&lun->lun_lock);
12027 	}
12028 	mtx_unlock(&softc->ctl_lock);
12029 	return (0);
12030 }
12031 
12032 static int
12033 ctl_abort_task(union ctl_io *io)
12034 {
12035 	union ctl_io *xio;
12036 	struct ctl_lun *lun;
12037 	struct ctl_softc *ctl_softc;
12038 #if 0
12039 	struct sbuf sb;
12040 	char printbuf[128];
12041 #endif
12042 	int found;
12043 	uint32_t targ_lun;
12044 
12045 	ctl_softc = control_softc;
12046 	found = 0;
12047 
12048 	/*
12049 	 * Look up the LUN.
12050 	 */
12051 	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12052 	mtx_lock(&ctl_softc->ctl_lock);
12053 	if ((targ_lun < CTL_MAX_LUNS)
12054 	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
12055 		lun = ctl_softc->ctl_luns[targ_lun];
12056 	else {
12057 		mtx_unlock(&ctl_softc->ctl_lock);
12058 		return (1);
12059 	}
12060 
12061 #if 0
12062 	printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
12063 	       lun->lun, io->taskio.tag_num, io->taskio.tag_type);
12064 #endif
12065 
12066 	mtx_lock(&lun->lun_lock);
12067 	mtx_unlock(&ctl_softc->ctl_lock);
12068 	/*
12069 	 * Run through the OOA queue and attempt to find the given I/O.
12070 	 * The target port, initiator ID, tag type and tag number have to
12071 	 * match the values that we got from the initiator.  If we have an
12072 	 * untagged command to abort, simply abort the first untagged command
12073 	 * we come to.  We only allow one untagged command at a time of course.
12074 	 */
12075 #if 0
12076 	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
12077 #endif
12078 	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12079 	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12080 #if 0
12081 		sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
12082 
12083 		sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
12084 			    lun->lun, xio->scsiio.tag_num,
12085 			    xio->scsiio.tag_type,
12086 			    (xio->io_hdr.blocked_links.tqe_prev
12087 			    == NULL) ? "" : " BLOCKED",
12088 			    (xio->io_hdr.flags &
12089 			    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
12090 			    (xio->io_hdr.flags &
12091 			    CTL_FLAG_ABORT) ? " ABORT" : "",
12092 			    (xio->io_hdr.flags &
12093 			    CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
12094 		ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
12095 		sbuf_finish(&sb);
12096 		printf("%s\n", sbuf_data(&sb));
12097 #endif
12098 
12099 		if ((xio->io_hdr.nexus.targ_port == io->io_hdr.nexus.targ_port)
12100 		 && (xio->io_hdr.nexus.initid.id ==
12101 		     io->io_hdr.nexus.initid.id)) {
12102 			/*
12103 			 * If the abort says that the task is untagged, the
12104 			 * task in the queue must be untagged.  Otherwise,
12105 			 * we just check to see whether the tag numbers
12106 			 * match.  This is because the QLogic firmware
12107 			 * doesn't pass back the tag type in an abort
12108 			 * request.
12109 			 */
12110 #if 0
12111 			if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
12112 			  && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
12113 			 || (xio->scsiio.tag_num == io->taskio.tag_num)) {
12114 #endif
12115 			/*
12116 			 * XXX KDM we've got problems with FC, because it
12117 			 * doesn't send down a tag type with aborts.  So we
12118 			 * can only really go by the tag number...
12119 			 * This may cause problems with parallel SCSI.
12120 			 * Need to figure that out!!
12121 			 */
12122 			if (xio->scsiio.tag_num == io->taskio.tag_num) {
12123 				xio->io_hdr.flags |= CTL_FLAG_ABORT;
12124 				found = 1;
12125 				if ((io->io_hdr.flags &
12126 				     CTL_FLAG_FROM_OTHER_SC) == 0 &&
12127 				    !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12128 					union ctl_ha_msg msg_info;
12129 
12130 					io->io_hdr.flags |=
12131 					                CTL_FLAG_SENT_2OTHER_SC;
12132 					msg_info.hdr.nexus = io->io_hdr.nexus;
12133 					msg_info.task.task_action =
12134 						CTL_TASK_ABORT_TASK;
12135 					msg_info.task.tag_num =
12136 						io->taskio.tag_num;
12137 					msg_info.task.tag_type =
12138 						io->taskio.tag_type;
12139 					msg_info.hdr.msg_type =
12140 						CTL_MSG_MANAGE_TASKS;
12141 					msg_info.hdr.original_sc = NULL;
12142 					msg_info.hdr.serializing_sc = NULL;
12143 #if 0
12144 					printf("Sent Abort to other side\n");
12145 #endif
12146 					if (CTL_HA_STATUS_SUCCESS !=
12147 					        ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12148 		    				(void *)&msg_info,
12149 						sizeof(msg_info), 0)) {
12150 					}
12151 				}
12152 #if 0
12153 				printf("ctl_abort_task: found I/O to abort\n");
12154 #endif
12155 				break;
12156 			}
12157 		}
12158 	}
12159 	mtx_unlock(&lun->lun_lock);
12160 
12161 	if (found == 0) {
12162 		/*
12163 		 * This isn't really an error.  It's entirely possible for
12164 		 * the abort and command completion to cross on the wire.
12165 		 * This is more of an informative/diagnostic error.
12166 		 */
12167 #if 0
12168 		printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
12169 		       "%d:%d:%d:%d tag %d type %d\n",
12170 		       io->io_hdr.nexus.initid.id,
12171 		       io->io_hdr.nexus.targ_port,
12172 		       io->io_hdr.nexus.targ_target.id,
12173 		       io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
12174 		       io->taskio.tag_type);
12175 #endif
12176 	}
12177 	return (0);
12178 }
12179 
12180 static void
12181 ctl_run_task(union ctl_io *io)
12182 {
12183 	struct ctl_softc *ctl_softc = control_softc;
12184 	int retval = 1;
12185 	const char *task_desc;
12186 
12187 	CTL_DEBUG_PRINT(("ctl_run_task\n"));
12188 
12189 	KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
12190 	    ("ctl_run_task: Unextected io_type %d\n",
12191 	     io->io_hdr.io_type));
12192 
12193 	task_desc = ctl_scsi_task_string(&io->taskio);
12194 	if (task_desc != NULL) {
12195 #ifdef NEEDTOPORT
12196 		csevent_log(CSC_CTL | CSC_SHELF_SW |
12197 			    CTL_TASK_REPORT,
12198 			    csevent_LogType_Trace,
12199 			    csevent_Severity_Information,
12200 			    csevent_AlertLevel_Green,
12201 			    csevent_FRU_Firmware,
12202 			    csevent_FRU_Unknown,
12203 			    "CTL: received task: %s",task_desc);
12204 #endif
12205 	} else {
12206 #ifdef NEEDTOPORT
12207 		csevent_log(CSC_CTL | CSC_SHELF_SW |
12208 			    CTL_TASK_REPORT,
12209 			    csevent_LogType_Trace,
12210 			    csevent_Severity_Information,
12211 			    csevent_AlertLevel_Green,
12212 			    csevent_FRU_Firmware,
12213 			    csevent_FRU_Unknown,
12214 			    "CTL: received unknown task "
12215 			    "type: %d (%#x)",
12216 			    io->taskio.task_action,
12217 			    io->taskio.task_action);
12218 #endif
12219 	}
12220 	switch (io->taskio.task_action) {
12221 	case CTL_TASK_ABORT_TASK:
12222 		retval = ctl_abort_task(io);
12223 		break;
12224 	case CTL_TASK_ABORT_TASK_SET:
12225 	case CTL_TASK_CLEAR_TASK_SET:
12226 		retval = ctl_abort_task_set(io);
12227 		break;
12228 	case CTL_TASK_CLEAR_ACA:
12229 		break;
12230 	case CTL_TASK_I_T_NEXUS_RESET:
12231 		retval = ctl_i_t_nexus_reset(io);
12232 		break;
12233 	case CTL_TASK_LUN_RESET: {
12234 		struct ctl_lun *lun;
12235 		uint32_t targ_lun;
12236 
12237 		targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12238 		mtx_lock(&ctl_softc->ctl_lock);
12239 		if ((targ_lun < CTL_MAX_LUNS)
12240 		 && (ctl_softc->ctl_luns[targ_lun] != NULL))
12241 			lun = ctl_softc->ctl_luns[targ_lun];
12242 		else {
12243 			mtx_unlock(&ctl_softc->ctl_lock);
12244 			retval = 1;
12245 			break;
12246 		}
12247 
12248 		if (!(io->io_hdr.flags &
12249 		    CTL_FLAG_FROM_OTHER_SC)) {
12250 			union ctl_ha_msg msg_info;
12251 
12252 			io->io_hdr.flags |=
12253 				CTL_FLAG_SENT_2OTHER_SC;
12254 			msg_info.hdr.msg_type =
12255 				CTL_MSG_MANAGE_TASKS;
12256 			msg_info.hdr.nexus = io->io_hdr.nexus;
12257 			msg_info.task.task_action =
12258 				CTL_TASK_LUN_RESET;
12259 			msg_info.hdr.original_sc = NULL;
12260 			msg_info.hdr.serializing_sc = NULL;
12261 			if (CTL_HA_STATUS_SUCCESS !=
12262 			    ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12263 			    (void *)&msg_info,
12264 			    sizeof(msg_info), 0)) {
12265 			}
12266 		}
12267 
12268 		retval = ctl_lun_reset(lun, io,
12269 				       CTL_UA_LUN_RESET);
12270 		mtx_unlock(&ctl_softc->ctl_lock);
12271 		break;
12272 	}
12273 	case CTL_TASK_TARGET_RESET:
12274 		retval = ctl_target_reset(ctl_softc, io, CTL_UA_TARG_RESET);
12275 		break;
12276 	case CTL_TASK_BUS_RESET:
12277 		retval = ctl_bus_reset(ctl_softc, io);
12278 		break;
12279 	case CTL_TASK_PORT_LOGIN:
12280 		break;
12281 	case CTL_TASK_PORT_LOGOUT:
12282 		break;
12283 	default:
12284 		printf("ctl_run_task: got unknown task management event %d\n",
12285 		       io->taskio.task_action);
12286 		break;
12287 	}
12288 	if (retval == 0)
12289 		io->io_hdr.status = CTL_SUCCESS;
12290 	else
12291 		io->io_hdr.status = CTL_ERROR;
12292 	ctl_done(io);
12293 }
12294 
12295 /*
12296  * For HA operation.  Handle commands that come in from the other
12297  * controller.
12298  */
12299 static void
12300 ctl_handle_isc(union ctl_io *io)
12301 {
12302 	int free_io;
12303 	struct ctl_lun *lun;
12304 	struct ctl_softc *ctl_softc;
12305 	uint32_t targ_lun;
12306 
12307 	ctl_softc = control_softc;
12308 
12309 	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12310 	lun = ctl_softc->ctl_luns[targ_lun];
12311 
12312 	switch (io->io_hdr.msg_type) {
12313 	case CTL_MSG_SERIALIZE:
12314 		free_io = ctl_serialize_other_sc_cmd(&io->scsiio);
12315 		break;
12316 	case CTL_MSG_R2R: {
12317 		const struct ctl_cmd_entry *entry;
12318 
12319 		/*
12320 		 * This is only used in SER_ONLY mode.
12321 		 */
12322 		free_io = 0;
12323 		entry = ctl_get_cmd_entry(&io->scsiio);
12324 		mtx_lock(&lun->lun_lock);
12325 		if (ctl_scsiio_lun_check(ctl_softc, lun,
12326 		    entry, (struct ctl_scsiio *)io) != 0) {
12327 			mtx_unlock(&lun->lun_lock);
12328 			ctl_done(io);
12329 			break;
12330 		}
12331 		io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
12332 		mtx_unlock(&lun->lun_lock);
12333 		ctl_enqueue_rtr(io);
12334 		break;
12335 	}
12336 	case CTL_MSG_FINISH_IO:
12337 		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
12338 			free_io = 0;
12339 			ctl_done(io);
12340 		} else {
12341 			free_io = 1;
12342 			mtx_lock(&lun->lun_lock);
12343 			TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
12344 				     ooa_links);
12345 			ctl_check_blocked(lun);
12346 			mtx_unlock(&lun->lun_lock);
12347 		}
12348 		break;
12349 	case CTL_MSG_PERS_ACTION:
12350 		ctl_hndl_per_res_out_on_other_sc(
12351 			(union ctl_ha_msg *)&io->presio.pr_msg);
12352 		free_io = 1;
12353 		break;
12354 	case CTL_MSG_BAD_JUJU:
12355 		free_io = 0;
12356 		ctl_done(io);
12357 		break;
12358 	case CTL_MSG_DATAMOVE:
12359 		/* Only used in XFER mode */
12360 		free_io = 0;
12361 		ctl_datamove_remote(io);
12362 		break;
12363 	case CTL_MSG_DATAMOVE_DONE:
12364 		/* Only used in XFER mode */
12365 		free_io = 0;
12366 		io->scsiio.be_move_done(io);
12367 		break;
12368 	default:
12369 		free_io = 1;
12370 		printf("%s: Invalid message type %d\n",
12371 		       __func__, io->io_hdr.msg_type);
12372 		break;
12373 	}
12374 	if (free_io)
12375 		ctl_free_io(io);
12376 
12377 }
12378 
12379 
12380 /*
12381  * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12382  * there is no match.
12383  */
12384 static ctl_lun_error_pattern
12385 ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12386 {
12387 	const struct ctl_cmd_entry *entry;
12388 	ctl_lun_error_pattern filtered_pattern, pattern;
12389 
12390 	pattern = desc->error_pattern;
12391 
12392 	/*
12393 	 * XXX KDM we need more data passed into this function to match a
12394 	 * custom pattern, and we actually need to implement custom pattern
12395 	 * matching.
12396 	 */
12397 	if (pattern & CTL_LUN_PAT_CMD)
12398 		return (CTL_LUN_PAT_CMD);
12399 
12400 	if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12401 		return (CTL_LUN_PAT_ANY);
12402 
12403 	entry = ctl_get_cmd_entry(ctsio);
12404 
12405 	filtered_pattern = entry->pattern & pattern;
12406 
12407 	/*
12408 	 * If the user requested specific flags in the pattern (e.g.
12409 	 * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12410 	 * flags.
12411 	 *
12412 	 * If the user did not specify any flags, it doesn't matter whether
12413 	 * or not the command supports the flags.
12414 	 */
12415 	if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12416 	     (pattern & ~CTL_LUN_PAT_MASK))
12417 		return (CTL_LUN_PAT_NONE);
12418 
12419 	/*
12420 	 * If the user asked for a range check, see if the requested LBA
12421 	 * range overlaps with this command's LBA range.
12422 	 */
12423 	if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12424 		uint64_t lba1;
12425 		uint32_t len1;
12426 		ctl_action action;
12427 		int retval;
12428 
12429 		retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12430 		if (retval != 0)
12431 			return (CTL_LUN_PAT_NONE);
12432 
12433 		action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12434 					      desc->lba_range.len);
12435 		/*
12436 		 * A "pass" means that the LBA ranges don't overlap, so
12437 		 * this doesn't match the user's range criteria.
12438 		 */
12439 		if (action == CTL_ACTION_PASS)
12440 			return (CTL_LUN_PAT_NONE);
12441 	}
12442 
12443 	return (filtered_pattern);
12444 }
12445 
12446 static void
12447 ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12448 {
12449 	struct ctl_error_desc *desc, *desc2;
12450 
12451 	mtx_assert(&lun->lun_lock, MA_OWNED);
12452 
12453 	STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12454 		ctl_lun_error_pattern pattern;
12455 		/*
12456 		 * Check to see whether this particular command matches
12457 		 * the pattern in the descriptor.
12458 		 */
12459 		pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12460 		if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12461 			continue;
12462 
12463 		switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12464 		case CTL_LUN_INJ_ABORTED:
12465 			ctl_set_aborted(&io->scsiio);
12466 			break;
12467 		case CTL_LUN_INJ_MEDIUM_ERR:
12468 			ctl_set_medium_error(&io->scsiio);
12469 			break;
12470 		case CTL_LUN_INJ_UA:
12471 			/* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12472 			 * OCCURRED */
12473 			ctl_set_ua(&io->scsiio, 0x29, 0x00);
12474 			break;
12475 		case CTL_LUN_INJ_CUSTOM:
12476 			/*
12477 			 * We're assuming the user knows what he is doing.
12478 			 * Just copy the sense information without doing
12479 			 * checks.
12480 			 */
12481 			bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12482 			      ctl_min(sizeof(desc->custom_sense),
12483 				      sizeof(io->scsiio.sense_data)));
12484 			io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12485 			io->scsiio.sense_len = SSD_FULL_SIZE;
12486 			io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12487 			break;
12488 		case CTL_LUN_INJ_NONE:
12489 		default:
12490 			/*
12491 			 * If this is an error injection type we don't know
12492 			 * about, clear the continuous flag (if it is set)
12493 			 * so it will get deleted below.
12494 			 */
12495 			desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12496 			break;
12497 		}
12498 		/*
12499 		 * By default, each error injection action is a one-shot
12500 		 */
12501 		if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12502 			continue;
12503 
12504 		STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12505 
12506 		free(desc, M_CTL);
12507 	}
12508 }
12509 
12510 #ifdef CTL_IO_DELAY
12511 static void
12512 ctl_datamove_timer_wakeup(void *arg)
12513 {
12514 	union ctl_io *io;
12515 
12516 	io = (union ctl_io *)arg;
12517 
12518 	ctl_datamove(io);
12519 }
12520 #endif /* CTL_IO_DELAY */
12521 
12522 void
12523 ctl_datamove(union ctl_io *io)
12524 {
12525 	void (*fe_datamove)(union ctl_io *io);
12526 
12527 	mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
12528 
12529 	CTL_DEBUG_PRINT(("ctl_datamove\n"));
12530 
12531 #ifdef CTL_TIME_IO
12532 	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12533 		char str[256];
12534 		char path_str[64];
12535 		struct sbuf sb;
12536 
12537 		ctl_scsi_path_string(io, path_str, sizeof(path_str));
12538 		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12539 
12540 		sbuf_cat(&sb, path_str);
12541 		switch (io->io_hdr.io_type) {
12542 		case CTL_IO_SCSI:
12543 			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12544 			sbuf_printf(&sb, "\n");
12545 			sbuf_cat(&sb, path_str);
12546 			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12547 				    io->scsiio.tag_num, io->scsiio.tag_type);
12548 			break;
12549 		case CTL_IO_TASK:
12550 			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12551 				    "Tag Type: %d\n", io->taskio.task_action,
12552 				    io->taskio.tag_num, io->taskio.tag_type);
12553 			break;
12554 		default:
12555 			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12556 			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12557 			break;
12558 		}
12559 		sbuf_cat(&sb, path_str);
12560 		sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12561 			    (intmax_t)time_uptime - io->io_hdr.start_time);
12562 		sbuf_finish(&sb);
12563 		printf("%s", sbuf_data(&sb));
12564 	}
12565 #endif /* CTL_TIME_IO */
12566 
12567 #ifdef CTL_IO_DELAY
12568 	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12569 		struct ctl_lun *lun;
12570 
12571 		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12572 
12573 		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12574 	} else {
12575 		struct ctl_lun *lun;
12576 
12577 		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12578 		if ((lun != NULL)
12579 		 && (lun->delay_info.datamove_delay > 0)) {
12580 			struct callout *callout;
12581 
12582 			callout = (struct callout *)&io->io_hdr.timer_bytes;
12583 			callout_init(callout, /*mpsafe*/ 1);
12584 			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12585 			callout_reset(callout,
12586 				      lun->delay_info.datamove_delay * hz,
12587 				      ctl_datamove_timer_wakeup, io);
12588 			if (lun->delay_info.datamove_type ==
12589 			    CTL_DELAY_TYPE_ONESHOT)
12590 				lun->delay_info.datamove_delay = 0;
12591 			return;
12592 		}
12593 	}
12594 #endif
12595 
12596 	/*
12597 	 * This command has been aborted.  Set the port status, so we fail
12598 	 * the data move.
12599 	 */
12600 	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12601 		printf("ctl_datamove: tag 0x%04x on (%ju:%d:%ju:%d) aborted\n",
12602 		       io->scsiio.tag_num,(uintmax_t)io->io_hdr.nexus.initid.id,
12603 		       io->io_hdr.nexus.targ_port,
12604 		       (uintmax_t)io->io_hdr.nexus.targ_target.id,
12605 		       io->io_hdr.nexus.targ_lun);
12606 		io->io_hdr.port_status = 31337;
12607 		/*
12608 		 * Note that the backend, in this case, will get the
12609 		 * callback in its context.  In other cases it may get
12610 		 * called in the frontend's interrupt thread context.
12611 		 */
12612 		io->scsiio.be_move_done(io);
12613 		return;
12614 	}
12615 
12616 	/*
12617 	 * If we're in XFER mode and this I/O is from the other shelf
12618 	 * controller, we need to send the DMA to the other side to
12619 	 * actually transfer the data to/from the host.  In serialize only
12620 	 * mode the transfer happens below CTL and ctl_datamove() is only
12621 	 * called on the machine that originally received the I/O.
12622 	 */
12623 	if ((control_softc->ha_mode == CTL_HA_MODE_XFER)
12624 	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12625 		union ctl_ha_msg msg;
12626 		uint32_t sg_entries_sent;
12627 		int do_sg_copy;
12628 		int i;
12629 
12630 		memset(&msg, 0, sizeof(msg));
12631 		msg.hdr.msg_type = CTL_MSG_DATAMOVE;
12632 		msg.hdr.original_sc = io->io_hdr.original_sc;
12633 		msg.hdr.serializing_sc = io;
12634 		msg.hdr.nexus = io->io_hdr.nexus;
12635 		msg.dt.flags = io->io_hdr.flags;
12636 		/*
12637 		 * We convert everything into a S/G list here.  We can't
12638 		 * pass by reference, only by value between controllers.
12639 		 * So we can't pass a pointer to the S/G list, only as many
12640 		 * S/G entries as we can fit in here.  If it's possible for
12641 		 * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
12642 		 * then we need to break this up into multiple transfers.
12643 		 */
12644 		if (io->scsiio.kern_sg_entries == 0) {
12645 			msg.dt.kern_sg_entries = 1;
12646 			/*
12647 			 * If this is in cached memory, flush the cache
12648 			 * before we send the DMA request to the other
12649 			 * controller.  We want to do this in either the
12650 			 * read or the write case.  The read case is
12651 			 * straightforward.  In the write case, we want to
12652 			 * make sure nothing is in the local cache that
12653 			 * could overwrite the DMAed data.
12654 			 */
12655 			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12656 				/*
12657 				 * XXX KDM use bus_dmamap_sync() here.
12658 				 */
12659 			}
12660 
12661 			/*
12662 			 * Convert to a physical address if this is a
12663 			 * virtual address.
12664 			 */
12665 			if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
12666 				msg.dt.sg_list[0].addr =
12667 					io->scsiio.kern_data_ptr;
12668 			} else {
12669 				/*
12670 				 * XXX KDM use busdma here!
12671 				 */
12672 #if 0
12673 				msg.dt.sg_list[0].addr = (void *)
12674 					vtophys(io->scsiio.kern_data_ptr);
12675 #endif
12676 			}
12677 
12678 			msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
12679 			do_sg_copy = 0;
12680 		} else {
12681 			struct ctl_sg_entry *sgl;
12682 
12683 			do_sg_copy = 1;
12684 			msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
12685 			sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
12686 			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12687 				/*
12688 				 * XXX KDM use bus_dmamap_sync() here.
12689 				 */
12690 			}
12691 		}
12692 
12693 		msg.dt.kern_data_len = io->scsiio.kern_data_len;
12694 		msg.dt.kern_total_len = io->scsiio.kern_total_len;
12695 		msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
12696 		msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
12697 		msg.dt.sg_sequence = 0;
12698 
12699 		/*
12700 		 * Loop until we've sent all of the S/G entries.  On the
12701 		 * other end, we'll recompose these S/G entries into one
12702 		 * contiguous list before passing it to the
12703 		 */
12704 		for (sg_entries_sent = 0; sg_entries_sent <
12705 		     msg.dt.kern_sg_entries; msg.dt.sg_sequence++) {
12706 			msg.dt.cur_sg_entries = ctl_min((sizeof(msg.dt.sg_list)/
12707 				sizeof(msg.dt.sg_list[0])),
12708 				msg.dt.kern_sg_entries - sg_entries_sent);
12709 
12710 			if (do_sg_copy != 0) {
12711 				struct ctl_sg_entry *sgl;
12712 				int j;
12713 
12714 				sgl = (struct ctl_sg_entry *)
12715 					io->scsiio.kern_data_ptr;
12716 				/*
12717 				 * If this is in cached memory, flush the cache
12718 				 * before we send the DMA request to the other
12719 				 * controller.  We want to do this in either
12720 				 * the * read or the write case.  The read
12721 				 * case is straightforward.  In the write
12722 				 * case, we want to make sure nothing is
12723 				 * in the local cache that could overwrite
12724 				 * the DMAed data.
12725 				 */
12726 
12727 				for (i = sg_entries_sent, j = 0;
12728 				     i < msg.dt.cur_sg_entries; i++, j++) {
12729 					if ((io->io_hdr.flags &
12730 					     CTL_FLAG_NO_DATASYNC) == 0) {
12731 						/*
12732 						 * XXX KDM use bus_dmamap_sync()
12733 						 */
12734 					}
12735 					if ((io->io_hdr.flags &
12736 					     CTL_FLAG_BUS_ADDR) == 0) {
12737 						/*
12738 						 * XXX KDM use busdma.
12739 						 */
12740 #if 0
12741 						msg.dt.sg_list[j].addr =(void *)
12742 						       vtophys(sgl[i].addr);
12743 #endif
12744 					} else {
12745 						msg.dt.sg_list[j].addr =
12746 							sgl[i].addr;
12747 					}
12748 					msg.dt.sg_list[j].len = sgl[i].len;
12749 				}
12750 			}
12751 
12752 			sg_entries_sent += msg.dt.cur_sg_entries;
12753 			if (sg_entries_sent >= msg.dt.kern_sg_entries)
12754 				msg.dt.sg_last = 1;
12755 			else
12756 				msg.dt.sg_last = 0;
12757 
12758 			/*
12759 			 * XXX KDM drop and reacquire the lock here?
12760 			 */
12761 			if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12762 			    sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
12763 				/*
12764 				 * XXX do something here.
12765 				 */
12766 			}
12767 
12768 			msg.dt.sent_sg_entries = sg_entries_sent;
12769 		}
12770 		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12771 		if (io->io_hdr.flags & CTL_FLAG_FAILOVER)
12772 			ctl_failover_io(io, /*have_lock*/ 0);
12773 
12774 	} else {
12775 
12776 		/*
12777 		 * Lookup the fe_datamove() function for this particular
12778 		 * front end.
12779 		 */
12780 		fe_datamove =
12781 		    control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12782 
12783 		fe_datamove(io);
12784 	}
12785 }
12786 
12787 static void
12788 ctl_send_datamove_done(union ctl_io *io, int have_lock)
12789 {
12790 	union ctl_ha_msg msg;
12791 	int isc_status;
12792 
12793 	memset(&msg, 0, sizeof(msg));
12794 
12795 	msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12796 	msg.hdr.original_sc = io;
12797 	msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12798 	msg.hdr.nexus = io->io_hdr.nexus;
12799 	msg.hdr.status = io->io_hdr.status;
12800 	msg.scsi.tag_num = io->scsiio.tag_num;
12801 	msg.scsi.tag_type = io->scsiio.tag_type;
12802 	msg.scsi.scsi_status = io->scsiio.scsi_status;
12803 	memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12804 	       sizeof(io->scsiio.sense_data));
12805 	msg.scsi.sense_len = io->scsiio.sense_len;
12806 	msg.scsi.sense_residual = io->scsiio.sense_residual;
12807 	msg.scsi.fetd_status = io->io_hdr.port_status;
12808 	msg.scsi.residual = io->scsiio.residual;
12809 	io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12810 
12811 	if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12812 		ctl_failover_io(io, /*have_lock*/ have_lock);
12813 		return;
12814 	}
12815 
12816 	isc_status = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0);
12817 	if (isc_status > CTL_HA_STATUS_SUCCESS) {
12818 		/* XXX do something if this fails */
12819 	}
12820 
12821 }
12822 
12823 /*
12824  * The DMA to the remote side is done, now we need to tell the other side
12825  * we're done so it can continue with its data movement.
12826  */
12827 static void
12828 ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12829 {
12830 	union ctl_io *io;
12831 
12832 	io = rq->context;
12833 
12834 	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12835 		printf("%s: ISC DMA write failed with error %d", __func__,
12836 		       rq->ret);
12837 		ctl_set_internal_failure(&io->scsiio,
12838 					 /*sks_valid*/ 1,
12839 					 /*retry_count*/ rq->ret);
12840 	}
12841 
12842 	ctl_dt_req_free(rq);
12843 
12844 	/*
12845 	 * In this case, we had to malloc the memory locally.  Free it.
12846 	 */
12847 	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12848 		int i;
12849 		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12850 			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12851 	}
12852 	/*
12853 	 * The data is in local and remote memory, so now we need to send
12854 	 * status (good or back) back to the other side.
12855 	 */
12856 	ctl_send_datamove_done(io, /*have_lock*/ 0);
12857 }
12858 
12859 /*
12860  * We've moved the data from the host/controller into local memory.  Now we
12861  * need to push it over to the remote controller's memory.
12862  */
12863 static int
12864 ctl_datamove_remote_dm_write_cb(union ctl_io *io)
12865 {
12866 	int retval;
12867 
12868 	retval = 0;
12869 
12870 	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12871 					  ctl_datamove_remote_write_cb);
12872 
12873 	return (retval);
12874 }
12875 
12876 static void
12877 ctl_datamove_remote_write(union ctl_io *io)
12878 {
12879 	int retval;
12880 	void (*fe_datamove)(union ctl_io *io);
12881 
12882 	/*
12883 	 * - Get the data from the host/HBA into local memory.
12884 	 * - DMA memory from the local controller to the remote controller.
12885 	 * - Send status back to the remote controller.
12886 	 */
12887 
12888 	retval = ctl_datamove_remote_sgl_setup(io);
12889 	if (retval != 0)
12890 		return;
12891 
12892 	/* Switch the pointer over so the FETD knows what to do */
12893 	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12894 
12895 	/*
12896 	 * Use a custom move done callback, since we need to send completion
12897 	 * back to the other controller, not to the backend on this side.
12898 	 */
12899 	io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12900 
12901 	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12902 
12903 	fe_datamove(io);
12904 
12905 	return;
12906 
12907 }
12908 
12909 static int
12910 ctl_datamove_remote_dm_read_cb(union ctl_io *io)
12911 {
12912 #if 0
12913 	char str[256];
12914 	char path_str[64];
12915 	struct sbuf sb;
12916 #endif
12917 
12918 	/*
12919 	 * In this case, we had to malloc the memory locally.  Free it.
12920 	 */
12921 	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12922 		int i;
12923 		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12924 			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12925 	}
12926 
12927 #if 0
12928 	scsi_path_string(io, path_str, sizeof(path_str));
12929 	sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12930 	sbuf_cat(&sb, path_str);
12931 	scsi_command_string(&io->scsiio, NULL, &sb);
12932 	sbuf_printf(&sb, "\n");
12933 	sbuf_cat(&sb, path_str);
12934 	sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12935 		    io->scsiio.tag_num, io->scsiio.tag_type);
12936 	sbuf_cat(&sb, path_str);
12937 	sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
12938 		    io->io_hdr.flags, io->io_hdr.status);
12939 	sbuf_finish(&sb);
12940 	printk("%s", sbuf_data(&sb));
12941 #endif
12942 
12943 
12944 	/*
12945 	 * The read is done, now we need to send status (good or bad) back
12946 	 * to the other side.
12947 	 */
12948 	ctl_send_datamove_done(io, /*have_lock*/ 0);
12949 
12950 	return (0);
12951 }
12952 
12953 static void
12954 ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
12955 {
12956 	union ctl_io *io;
12957 	void (*fe_datamove)(union ctl_io *io);
12958 
12959 	io = rq->context;
12960 
12961 	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12962 		printf("%s: ISC DMA read failed with error %d", __func__,
12963 		       rq->ret);
12964 		ctl_set_internal_failure(&io->scsiio,
12965 					 /*sks_valid*/ 1,
12966 					 /*retry_count*/ rq->ret);
12967 	}
12968 
12969 	ctl_dt_req_free(rq);
12970 
12971 	/* Switch the pointer over so the FETD knows what to do */
12972 	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12973 
12974 	/*
12975 	 * Use a custom move done callback, since we need to send completion
12976 	 * back to the other controller, not to the backend on this side.
12977 	 */
12978 	io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
12979 
12980 	/* XXX KDM add checks like the ones in ctl_datamove? */
12981 
12982 	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12983 
12984 	fe_datamove(io);
12985 }
12986 
12987 static int
12988 ctl_datamove_remote_sgl_setup(union ctl_io *io)
12989 {
12990 	struct ctl_sg_entry *local_sglist, *remote_sglist;
12991 	struct ctl_sg_entry *local_dma_sglist, *remote_dma_sglist;
12992 	struct ctl_softc *softc;
12993 	int retval;
12994 	int i;
12995 
12996 	retval = 0;
12997 	softc = control_softc;
12998 
12999 	local_sglist = io->io_hdr.local_sglist;
13000 	local_dma_sglist = io->io_hdr.local_dma_sglist;
13001 	remote_sglist = io->io_hdr.remote_sglist;
13002 	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
13003 
13004 	if (io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) {
13005 		for (i = 0; i < io->scsiio.kern_sg_entries; i++) {
13006 			local_sglist[i].len = remote_sglist[i].len;
13007 
13008 			/*
13009 			 * XXX Detect the situation where the RS-level I/O
13010 			 * redirector on the other side has already read the
13011 			 * data off of the AOR RS on this side, and
13012 			 * transferred it to remote (mirror) memory on the
13013 			 * other side.  Since we already have the data in
13014 			 * memory here, we just need to use it.
13015 			 *
13016 			 * XXX KDM this can probably be removed once we
13017 			 * get the cache device code in and take the
13018 			 * current AOR implementation out.
13019 			 */
13020 #ifdef NEEDTOPORT
13021 			if ((remote_sglist[i].addr >=
13022 			     (void *)vtophys(softc->mirr->addr))
13023 			 && (remote_sglist[i].addr <
13024 			     ((void *)vtophys(softc->mirr->addr) +
13025 			     CacheMirrorOffset))) {
13026 				local_sglist[i].addr = remote_sglist[i].addr -
13027 					CacheMirrorOffset;
13028 				if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13029 				     CTL_FLAG_DATA_IN)
13030 					io->io_hdr.flags |= CTL_FLAG_REDIR_DONE;
13031 			} else {
13032 				local_sglist[i].addr = remote_sglist[i].addr +
13033 					CacheMirrorOffset;
13034 			}
13035 #endif
13036 #if 0
13037 			printf("%s: local %p, remote %p, len %d\n",
13038 			       __func__, local_sglist[i].addr,
13039 			       remote_sglist[i].addr, local_sglist[i].len);
13040 #endif
13041 		}
13042 	} else {
13043 		uint32_t len_to_go;
13044 
13045 		/*
13046 		 * In this case, we don't have automatically allocated
13047 		 * memory for this I/O on this controller.  This typically
13048 		 * happens with internal CTL I/O -- e.g. inquiry, mode
13049 		 * sense, etc.  Anything coming from RAIDCore will have
13050 		 * a mirror area available.
13051 		 */
13052 		len_to_go = io->scsiio.kern_data_len;
13053 
13054 		/*
13055 		 * Clear the no datasync flag, we have to use malloced
13056 		 * buffers.
13057 		 */
13058 		io->io_hdr.flags &= ~CTL_FLAG_NO_DATASYNC;
13059 
13060 		/*
13061 		 * The difficult thing here is that the size of the various
13062 		 * S/G segments may be different than the size from the
13063 		 * remote controller.  That'll make it harder when DMAing
13064 		 * the data back to the other side.
13065 		 */
13066 		for (i = 0; (i < sizeof(io->io_hdr.remote_sglist) /
13067 		     sizeof(io->io_hdr.remote_sglist[0])) &&
13068 		     (len_to_go > 0); i++) {
13069 			local_sglist[i].len = ctl_min(len_to_go, 131072);
13070 			CTL_SIZE_8B(local_dma_sglist[i].len,
13071 				    local_sglist[i].len);
13072 			local_sglist[i].addr =
13073 				malloc(local_dma_sglist[i].len, M_CTL,M_WAITOK);
13074 
13075 			local_dma_sglist[i].addr = local_sglist[i].addr;
13076 
13077 			if (local_sglist[i].addr == NULL) {
13078 				int j;
13079 
13080 				printf("malloc failed for %zd bytes!",
13081 				       local_dma_sglist[i].len);
13082 				for (j = 0; j < i; j++) {
13083 					free(local_sglist[j].addr, M_CTL);
13084 				}
13085 				ctl_set_internal_failure(&io->scsiio,
13086 							 /*sks_valid*/ 1,
13087 							 /*retry_count*/ 4857);
13088 				retval = 1;
13089 				goto bailout_error;
13090 
13091 			}
13092 			/* XXX KDM do we need a sync here? */
13093 
13094 			len_to_go -= local_sglist[i].len;
13095 		}
13096 		/*
13097 		 * Reset the number of S/G entries accordingly.  The
13098 		 * original number of S/G entries is available in
13099 		 * rem_sg_entries.
13100 		 */
13101 		io->scsiio.kern_sg_entries = i;
13102 
13103 #if 0
13104 		printf("%s: kern_sg_entries = %d\n", __func__,
13105 		       io->scsiio.kern_sg_entries);
13106 		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13107 			printf("%s: sg[%d] = %p, %d (DMA: %d)\n", __func__, i,
13108 			       local_sglist[i].addr, local_sglist[i].len,
13109 			       local_dma_sglist[i].len);
13110 #endif
13111 	}
13112 
13113 
13114 	return (retval);
13115 
13116 bailout_error:
13117 
13118 	ctl_send_datamove_done(io, /*have_lock*/ 0);
13119 
13120 	return (retval);
13121 }
13122 
13123 static int
13124 ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
13125 			 ctl_ha_dt_cb callback)
13126 {
13127 	struct ctl_ha_dt_req *rq;
13128 	struct ctl_sg_entry *remote_sglist, *local_sglist;
13129 	struct ctl_sg_entry *remote_dma_sglist, *local_dma_sglist;
13130 	uint32_t local_used, remote_used, total_used;
13131 	int retval;
13132 	int i, j;
13133 
13134 	retval = 0;
13135 
13136 	rq = ctl_dt_req_alloc();
13137 
13138 	/*
13139 	 * If we failed to allocate the request, and if the DMA didn't fail
13140 	 * anyway, set busy status.  This is just a resource allocation
13141 	 * failure.
13142 	 */
13143 	if ((rq == NULL)
13144 	 && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE))
13145 		ctl_set_busy(&io->scsiio);
13146 
13147 	if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) {
13148 
13149 		if (rq != NULL)
13150 			ctl_dt_req_free(rq);
13151 
13152 		/*
13153 		 * The data move failed.  We need to return status back
13154 		 * to the other controller.  No point in trying to DMA
13155 		 * data to the remote controller.
13156 		 */
13157 
13158 		ctl_send_datamove_done(io, /*have_lock*/ 0);
13159 
13160 		retval = 1;
13161 
13162 		goto bailout;
13163 	}
13164 
13165 	local_sglist = io->io_hdr.local_sglist;
13166 	local_dma_sglist = io->io_hdr.local_dma_sglist;
13167 	remote_sglist = io->io_hdr.remote_sglist;
13168 	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
13169 	local_used = 0;
13170 	remote_used = 0;
13171 	total_used = 0;
13172 
13173 	if (io->io_hdr.flags & CTL_FLAG_REDIR_DONE) {
13174 		rq->ret = CTL_HA_STATUS_SUCCESS;
13175 		rq->context = io;
13176 		callback(rq);
13177 		goto bailout;
13178 	}
13179 
13180 	/*
13181 	 * Pull/push the data over the wire from/to the other controller.
13182 	 * This takes into account the possibility that the local and
13183 	 * remote sglists may not be identical in terms of the size of
13184 	 * the elements and the number of elements.
13185 	 *
13186 	 * One fundamental assumption here is that the length allocated for
13187 	 * both the local and remote sglists is identical.  Otherwise, we've
13188 	 * essentially got a coding error of some sort.
13189 	 */
13190 	for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
13191 		int isc_ret;
13192 		uint32_t cur_len, dma_length;
13193 		uint8_t *tmp_ptr;
13194 
13195 		rq->id = CTL_HA_DATA_CTL;
13196 		rq->command = command;
13197 		rq->context = io;
13198 
13199 		/*
13200 		 * Both pointers should be aligned.  But it is possible
13201 		 * that the allocation length is not.  They should both
13202 		 * also have enough slack left over at the end, though,
13203 		 * to round up to the next 8 byte boundary.
13204 		 */
13205 		cur_len = ctl_min(local_sglist[i].len - local_used,
13206 				  remote_sglist[j].len - remote_used);
13207 
13208 		/*
13209 		 * In this case, we have a size issue and need to decrease
13210 		 * the size, except in the case where we actually have less
13211 		 * than 8 bytes left.  In that case, we need to increase
13212 		 * the DMA length to get the last bit.
13213 		 */
13214 		if ((cur_len & 0x7) != 0) {
13215 			if (cur_len > 0x7) {
13216 				cur_len = cur_len - (cur_len & 0x7);
13217 				dma_length = cur_len;
13218 			} else {
13219 				CTL_SIZE_8B(dma_length, cur_len);
13220 			}
13221 
13222 		} else
13223 			dma_length = cur_len;
13224 
13225 		/*
13226 		 * If we had to allocate memory for this I/O, instead of using
13227 		 * the non-cached mirror memory, we'll need to flush the cache
13228 		 * before trying to DMA to the other controller.
13229 		 *
13230 		 * We could end up doing this multiple times for the same
13231 		 * segment if we have a larger local segment than remote
13232 		 * segment.  That shouldn't be an issue.
13233 		 */
13234 		if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
13235 			/*
13236 			 * XXX KDM use bus_dmamap_sync() here.
13237 			 */
13238 		}
13239 
13240 		rq->size = dma_length;
13241 
13242 		tmp_ptr = (uint8_t *)local_sglist[i].addr;
13243 		tmp_ptr += local_used;
13244 
13245 		/* Use physical addresses when talking to ISC hardware */
13246 		if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
13247 			/* XXX KDM use busdma */
13248 #if 0
13249 			rq->local = vtophys(tmp_ptr);
13250 #endif
13251 		} else
13252 			rq->local = tmp_ptr;
13253 
13254 		tmp_ptr = (uint8_t *)remote_sglist[j].addr;
13255 		tmp_ptr += remote_used;
13256 		rq->remote = tmp_ptr;
13257 
13258 		rq->callback = NULL;
13259 
13260 		local_used += cur_len;
13261 		if (local_used >= local_sglist[i].len) {
13262 			i++;
13263 			local_used = 0;
13264 		}
13265 
13266 		remote_used += cur_len;
13267 		if (remote_used >= remote_sglist[j].len) {
13268 			j++;
13269 			remote_used = 0;
13270 		}
13271 		total_used += cur_len;
13272 
13273 		if (total_used >= io->scsiio.kern_data_len)
13274 			rq->callback = callback;
13275 
13276 		if ((rq->size & 0x7) != 0) {
13277 			printf("%s: warning: size %d is not on 8b boundary\n",
13278 			       __func__, rq->size);
13279 		}
13280 		if (((uintptr_t)rq->local & 0x7) != 0) {
13281 			printf("%s: warning: local %p not on 8b boundary\n",
13282 			       __func__, rq->local);
13283 		}
13284 		if (((uintptr_t)rq->remote & 0x7) != 0) {
13285 			printf("%s: warning: remote %p not on 8b boundary\n",
13286 			       __func__, rq->local);
13287 		}
13288 #if 0
13289 		printf("%s: %s: local %#x remote %#x size %d\n", __func__,
13290 		       (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
13291 		       rq->local, rq->remote, rq->size);
13292 #endif
13293 
13294 		isc_ret = ctl_dt_single(rq);
13295 		if (isc_ret == CTL_HA_STATUS_WAIT)
13296 			continue;
13297 
13298 		if (isc_ret == CTL_HA_STATUS_DISCONNECT) {
13299 			rq->ret = CTL_HA_STATUS_SUCCESS;
13300 		} else {
13301 			rq->ret = isc_ret;
13302 		}
13303 		callback(rq);
13304 		goto bailout;
13305 	}
13306 
13307 bailout:
13308 	return (retval);
13309 
13310 }
13311 
13312 static void
13313 ctl_datamove_remote_read(union ctl_io *io)
13314 {
13315 	int retval;
13316 	int i;
13317 
13318 	/*
13319 	 * This will send an error to the other controller in the case of a
13320 	 * failure.
13321 	 */
13322 	retval = ctl_datamove_remote_sgl_setup(io);
13323 	if (retval != 0)
13324 		return;
13325 
13326 	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
13327 					  ctl_datamove_remote_read_cb);
13328 	if ((retval != 0)
13329 	 && ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0)) {
13330 		/*
13331 		 * Make sure we free memory if there was an error..  The
13332 		 * ctl_datamove_remote_xfer() function will send the
13333 		 * datamove done message, or call the callback with an
13334 		 * error if there is a problem.
13335 		 */
13336 		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13337 			free(io->io_hdr.local_sglist[i].addr, M_CTL);
13338 	}
13339 
13340 	return;
13341 }
13342 
13343 /*
13344  * Process a datamove request from the other controller.  This is used for
13345  * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
13346  * first.  Once that is complete, the data gets DMAed into the remote
13347  * controller's memory.  For reads, we DMA from the remote controller's
13348  * memory into our memory first, and then move it out to the FETD.
13349  */
13350 static void
13351 ctl_datamove_remote(union ctl_io *io)
13352 {
13353 	struct ctl_softc *softc;
13354 
13355 	softc = control_softc;
13356 
13357 	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
13358 
13359 	/*
13360 	 * Note that we look for an aborted I/O here, but don't do some of
13361 	 * the other checks that ctl_datamove() normally does.
13362 	 * We don't need to run the datamove delay code, since that should
13363 	 * have been done if need be on the other controller.
13364 	 */
13365 	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
13366 		printf("%s: tag 0x%04x on (%d:%d:%d:%d) aborted\n", __func__,
13367 		       io->scsiio.tag_num, io->io_hdr.nexus.initid.id,
13368 		       io->io_hdr.nexus.targ_port,
13369 		       io->io_hdr.nexus.targ_target.id,
13370 		       io->io_hdr.nexus.targ_lun);
13371 		io->io_hdr.port_status = 31338;
13372 		ctl_send_datamove_done(io, /*have_lock*/ 0);
13373 		return;
13374 	}
13375 
13376 	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT) {
13377 		ctl_datamove_remote_write(io);
13378 	} else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN){
13379 		ctl_datamove_remote_read(io);
13380 	} else {
13381 		union ctl_ha_msg msg;
13382 		struct scsi_sense_data *sense;
13383 		uint8_t sks[3];
13384 		int retry_count;
13385 
13386 		memset(&msg, 0, sizeof(msg));
13387 
13388 		msg.hdr.msg_type = CTL_MSG_BAD_JUJU;
13389 		msg.hdr.status = CTL_SCSI_ERROR;
13390 		msg.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
13391 
13392 		retry_count = 4243;
13393 
13394 		sense = &msg.scsi.sense_data;
13395 		sks[0] = SSD_SCS_VALID;
13396 		sks[1] = (retry_count >> 8) & 0xff;
13397 		sks[2] = retry_count & 0xff;
13398 
13399 		/* "Internal target failure" */
13400 		scsi_set_sense_data(sense,
13401 				    /*sense_format*/ SSD_TYPE_NONE,
13402 				    /*current_error*/ 1,
13403 				    /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
13404 				    /*asc*/ 0x44,
13405 				    /*ascq*/ 0x00,
13406 				    /*type*/ SSD_ELEM_SKS,
13407 				    /*size*/ sizeof(sks),
13408 				    /*data*/ sks,
13409 				    SSD_ELEM_NONE);
13410 
13411 		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
13412 		if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
13413 			ctl_failover_io(io, /*have_lock*/ 1);
13414 			return;
13415 		}
13416 
13417 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0) >
13418 		    CTL_HA_STATUS_SUCCESS) {
13419 			/* XXX KDM what to do if this fails? */
13420 		}
13421 		return;
13422 	}
13423 
13424 }
13425 
13426 static int
13427 ctl_process_done(union ctl_io *io)
13428 {
13429 	struct ctl_lun *lun;
13430 	struct ctl_softc *ctl_softc;
13431 	void (*fe_done)(union ctl_io *io);
13432 	uint32_t targ_port = ctl_port_idx(io->io_hdr.nexus.targ_port);
13433 
13434 	CTL_DEBUG_PRINT(("ctl_process_done\n"));
13435 
13436 	fe_done =
13437 	    control_softc->ctl_ports[targ_port]->fe_done;
13438 
13439 #ifdef CTL_TIME_IO
13440 	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
13441 		char str[256];
13442 		char path_str[64];
13443 		struct sbuf sb;
13444 
13445 		ctl_scsi_path_string(io, path_str, sizeof(path_str));
13446 		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13447 
13448 		sbuf_cat(&sb, path_str);
13449 		switch (io->io_hdr.io_type) {
13450 		case CTL_IO_SCSI:
13451 			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
13452 			sbuf_printf(&sb, "\n");
13453 			sbuf_cat(&sb, path_str);
13454 			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13455 				    io->scsiio.tag_num, io->scsiio.tag_type);
13456 			break;
13457 		case CTL_IO_TASK:
13458 			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
13459 				    "Tag Type: %d\n", io->taskio.task_action,
13460 				    io->taskio.tag_num, io->taskio.tag_type);
13461 			break;
13462 		default:
13463 			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13464 			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13465 			break;
13466 		}
13467 		sbuf_cat(&sb, path_str);
13468 		sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
13469 			    (intmax_t)time_uptime - io->io_hdr.start_time);
13470 		sbuf_finish(&sb);
13471 		printf("%s", sbuf_data(&sb));
13472 	}
13473 #endif /* CTL_TIME_IO */
13474 
13475 	switch (io->io_hdr.io_type) {
13476 	case CTL_IO_SCSI:
13477 		break;
13478 	case CTL_IO_TASK:
13479 		if (bootverbose || verbose > 0)
13480 			ctl_io_error_print(io, NULL);
13481 		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
13482 			ctl_free_io(io);
13483 		else
13484 			fe_done(io);
13485 		return (CTL_RETVAL_COMPLETE);
13486 		break;
13487 	default:
13488 		printf("ctl_process_done: invalid io type %d\n",
13489 		       io->io_hdr.io_type);
13490 		panic("ctl_process_done: invalid io type %d\n",
13491 		      io->io_hdr.io_type);
13492 		break; /* NOTREACHED */
13493 	}
13494 
13495 	lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13496 	if (lun == NULL) {
13497 		CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
13498 				 io->io_hdr.nexus.targ_mapped_lun));
13499 		fe_done(io);
13500 		goto bailout;
13501 	}
13502 	ctl_softc = lun->ctl_softc;
13503 
13504 	mtx_lock(&lun->lun_lock);
13505 
13506 	/*
13507 	 * Check to see if we have any errors to inject here.  We only
13508 	 * inject errors for commands that don't already have errors set.
13509 	 */
13510 	if ((STAILQ_FIRST(&lun->error_list) != NULL)
13511 	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))
13512 		ctl_inject_error(lun, io);
13513 
13514 	/*
13515 	 * XXX KDM how do we treat commands that aren't completed
13516 	 * successfully?
13517 	 *
13518 	 * XXX KDM should we also track I/O latency?
13519 	 */
13520 	if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
13521 	    io->io_hdr.io_type == CTL_IO_SCSI) {
13522 #ifdef CTL_TIME_IO
13523 		struct bintime cur_bt;
13524 #endif
13525 		int type;
13526 
13527 		if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13528 		    CTL_FLAG_DATA_IN)
13529 			type = CTL_STATS_READ;
13530 		else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13531 		    CTL_FLAG_DATA_OUT)
13532 			type = CTL_STATS_WRITE;
13533 		else
13534 			type = CTL_STATS_NO_IO;
13535 
13536 		lun->stats.ports[targ_port].bytes[type] +=
13537 		    io->scsiio.kern_total_len;
13538 		lun->stats.ports[targ_port].operations[type]++;
13539 #ifdef CTL_TIME_IO
13540 		bintime_add(&lun->stats.ports[targ_port].dma_time[type],
13541 		   &io->io_hdr.dma_bt);
13542 		lun->stats.ports[targ_port].num_dmas[type] +=
13543 		    io->io_hdr.num_dmas;
13544 		getbintime(&cur_bt);
13545 		bintime_sub(&cur_bt, &io->io_hdr.start_bt);
13546 		bintime_add(&lun->stats.ports[targ_port].time[type], &cur_bt);
13547 #endif
13548 	}
13549 
13550 	/*
13551 	 * Remove this from the OOA queue.
13552 	 */
13553 	TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13554 
13555 	/*
13556 	 * Run through the blocked queue on this LUN and see if anything
13557 	 * has become unblocked, now that this transaction is done.
13558 	 */
13559 	ctl_check_blocked(lun);
13560 
13561 	/*
13562 	 * If the LUN has been invalidated, free it if there is nothing
13563 	 * left on its OOA queue.
13564 	 */
13565 	if ((lun->flags & CTL_LUN_INVALID)
13566 	 && TAILQ_EMPTY(&lun->ooa_queue)) {
13567 		mtx_unlock(&lun->lun_lock);
13568 		mtx_lock(&ctl_softc->ctl_lock);
13569 		ctl_free_lun(lun);
13570 		mtx_unlock(&ctl_softc->ctl_lock);
13571 	} else
13572 		mtx_unlock(&lun->lun_lock);
13573 
13574 	/*
13575 	 * If this command has been aborted, make sure we set the status
13576 	 * properly.  The FETD is responsible for freeing the I/O and doing
13577 	 * whatever it needs to do to clean up its state.
13578 	 */
13579 	if (io->io_hdr.flags & CTL_FLAG_ABORT)
13580 		ctl_set_task_aborted(&io->scsiio);
13581 
13582 	/*
13583 	 * We print out status for every task management command.  For SCSI
13584 	 * commands, we filter out any unit attention errors; they happen
13585 	 * on every boot, and would clutter up the log.  Note:  task
13586 	 * management commands aren't printed here, they are printed above,
13587 	 * since they should never even make it down here.
13588 	 */
13589 	switch (io->io_hdr.io_type) {
13590 	case CTL_IO_SCSI: {
13591 		int error_code, sense_key, asc, ascq;
13592 
13593 		sense_key = 0;
13594 
13595 		if (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SCSI_ERROR)
13596 		 && (io->scsiio.scsi_status == SCSI_STATUS_CHECK_COND)) {
13597 			/*
13598 			 * Since this is just for printing, no need to
13599 			 * show errors here.
13600 			 */
13601 			scsi_extract_sense_len(&io->scsiio.sense_data,
13602 					       io->scsiio.sense_len,
13603 					       &error_code,
13604 					       &sense_key,
13605 					       &asc,
13606 					       &ascq,
13607 					       /*show_errors*/ 0);
13608 		}
13609 
13610 		if (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
13611 		 && (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SCSI_ERROR)
13612 		  || (io->scsiio.scsi_status != SCSI_STATUS_CHECK_COND)
13613 		  || (sense_key != SSD_KEY_UNIT_ATTENTION))) {
13614 
13615 			if ((time_uptime - ctl_softc->last_print_jiffies) <= 0){
13616 				ctl_softc->skipped_prints++;
13617 			} else {
13618 				uint32_t skipped_prints;
13619 
13620 				skipped_prints = ctl_softc->skipped_prints;
13621 
13622 				ctl_softc->skipped_prints = 0;
13623 				ctl_softc->last_print_jiffies = time_uptime;
13624 
13625 				if (skipped_prints > 0) {
13626 #ifdef NEEDTOPORT
13627 					csevent_log(CSC_CTL | CSC_SHELF_SW |
13628 					    CTL_ERROR_REPORT,
13629 					    csevent_LogType_Trace,
13630 					    csevent_Severity_Information,
13631 					    csevent_AlertLevel_Green,
13632 					    csevent_FRU_Firmware,
13633 					    csevent_FRU_Unknown,
13634 					    "High CTL error volume, %d prints "
13635 					    "skipped", skipped_prints);
13636 #endif
13637 				}
13638 				if (bootverbose || verbose > 0)
13639 					ctl_io_error_print(io, NULL);
13640 			}
13641 		}
13642 		break;
13643 	}
13644 	case CTL_IO_TASK:
13645 		if (bootverbose || verbose > 0)
13646 			ctl_io_error_print(io, NULL);
13647 		break;
13648 	default:
13649 		break;
13650 	}
13651 
13652 	/*
13653 	 * Tell the FETD or the other shelf controller we're done with this
13654 	 * command.  Note that only SCSI commands get to this point.  Task
13655 	 * management commands are completed above.
13656 	 *
13657 	 * We only send status to the other controller if we're in XFER
13658 	 * mode.  In SER_ONLY mode, the I/O is done on the controller that
13659 	 * received the I/O (from CTL's perspective), and so the status is
13660 	 * generated there.
13661 	 *
13662 	 * XXX KDM if we hold the lock here, we could cause a deadlock
13663 	 * if the frontend comes back in in this context to queue
13664 	 * something.
13665 	 */
13666 	if ((ctl_softc->ha_mode == CTL_HA_MODE_XFER)
13667 	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
13668 		union ctl_ha_msg msg;
13669 
13670 		memset(&msg, 0, sizeof(msg));
13671 		msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13672 		msg.hdr.original_sc = io->io_hdr.original_sc;
13673 		msg.hdr.nexus = io->io_hdr.nexus;
13674 		msg.hdr.status = io->io_hdr.status;
13675 		msg.scsi.scsi_status = io->scsiio.scsi_status;
13676 		msg.scsi.tag_num = io->scsiio.tag_num;
13677 		msg.scsi.tag_type = io->scsiio.tag_type;
13678 		msg.scsi.sense_len = io->scsiio.sense_len;
13679 		msg.scsi.sense_residual = io->scsiio.sense_residual;
13680 		msg.scsi.residual = io->scsiio.residual;
13681 		memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
13682 		       sizeof(io->scsiio.sense_data));
13683 		/*
13684 		 * We copy this whether or not this is an I/O-related
13685 		 * command.  Otherwise, we'd have to go and check to see
13686 		 * whether it's a read/write command, and it really isn't
13687 		 * worth it.
13688 		 */
13689 		memcpy(&msg.scsi.lbalen,
13690 		       &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
13691 		       sizeof(msg.scsi.lbalen));
13692 
13693 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13694 				sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
13695 			/* XXX do something here */
13696 		}
13697 
13698 		ctl_free_io(io);
13699 	} else
13700 		fe_done(io);
13701 
13702 bailout:
13703 
13704 	return (CTL_RETVAL_COMPLETE);
13705 }
13706 
13707 #ifdef CTL_WITH_CA
13708 /*
13709  * Front end should call this if it doesn't do autosense.  When the request
13710  * sense comes back in from the initiator, we'll dequeue this and send it.
13711  */
13712 int
13713 ctl_queue_sense(union ctl_io *io)
13714 {
13715 	struct ctl_lun *lun;
13716 	struct ctl_softc *ctl_softc;
13717 	uint32_t initidx, targ_lun;
13718 
13719 	ctl_softc = control_softc;
13720 
13721 	CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13722 
13723 	/*
13724 	 * LUN lookup will likely move to the ctl_work_thread() once we
13725 	 * have our new queueing infrastructure (that doesn't put things on
13726 	 * a per-LUN queue initially).  That is so that we can handle
13727 	 * things like an INQUIRY to a LUN that we don't have enabled.  We
13728 	 * can't deal with that right now.
13729 	 */
13730 	mtx_lock(&ctl_softc->ctl_lock);
13731 
13732 	/*
13733 	 * If we don't have a LUN for this, just toss the sense
13734 	 * information.
13735 	 */
13736 	targ_lun = io->io_hdr.nexus.targ_lun;
13737 	targ_lun = ctl_map_lun(io->io_hdr.nexus.targ_port, targ_lun);
13738 	if ((targ_lun < CTL_MAX_LUNS)
13739 	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
13740 		lun = ctl_softc->ctl_luns[targ_lun];
13741 	else
13742 		goto bailout;
13743 
13744 	initidx = ctl_get_initindex(&io->io_hdr.nexus);
13745 
13746 	mtx_lock(&lun->lun_lock);
13747 	/*
13748 	 * Already have CA set for this LUN...toss the sense information.
13749 	 */
13750 	if (ctl_is_set(lun->have_ca, initidx)) {
13751 		mtx_unlock(&lun->lun_lock);
13752 		goto bailout;
13753 	}
13754 
13755 	memcpy(&lun->pending_sense[initidx], &io->scsiio.sense_data,
13756 	       ctl_min(sizeof(lun->pending_sense[initidx]),
13757 	       sizeof(io->scsiio.sense_data)));
13758 	ctl_set_mask(lun->have_ca, initidx);
13759 	mtx_unlock(&lun->lun_lock);
13760 
13761 bailout:
13762 	mtx_unlock(&ctl_softc->ctl_lock);
13763 
13764 	ctl_free_io(io);
13765 
13766 	return (CTL_RETVAL_COMPLETE);
13767 }
13768 #endif
13769 
13770 /*
13771  * Primary command inlet from frontend ports.  All SCSI and task I/O
13772  * requests must go through this function.
13773  */
13774 int
13775 ctl_queue(union ctl_io *io)
13776 {
13777 	struct ctl_softc *ctl_softc;
13778 
13779 	CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13780 
13781 	ctl_softc = control_softc;
13782 
13783 #ifdef CTL_TIME_IO
13784 	io->io_hdr.start_time = time_uptime;
13785 	getbintime(&io->io_hdr.start_bt);
13786 #endif /* CTL_TIME_IO */
13787 
13788 	/* Map FE-specific LUN ID into global one. */
13789 	io->io_hdr.nexus.targ_mapped_lun =
13790 	    ctl_map_lun(io->io_hdr.nexus.targ_port, io->io_hdr.nexus.targ_lun);
13791 
13792 	switch (io->io_hdr.io_type) {
13793 	case CTL_IO_SCSI:
13794 	case CTL_IO_TASK:
13795 		ctl_enqueue_incoming(io);
13796 		break;
13797 	default:
13798 		printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13799 		return (EINVAL);
13800 	}
13801 
13802 	return (CTL_RETVAL_COMPLETE);
13803 }
13804 
13805 #ifdef CTL_IO_DELAY
13806 static void
13807 ctl_done_timer_wakeup(void *arg)
13808 {
13809 	union ctl_io *io;
13810 
13811 	io = (union ctl_io *)arg;
13812 	ctl_done(io);
13813 }
13814 #endif /* CTL_IO_DELAY */
13815 
13816 void
13817 ctl_done(union ctl_io *io)
13818 {
13819 	struct ctl_softc *ctl_softc;
13820 
13821 	ctl_softc = control_softc;
13822 
13823 	/*
13824 	 * Enable this to catch duplicate completion issues.
13825 	 */
13826 #if 0
13827 	if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13828 		printf("%s: type %d msg %d cdb %x iptl: "
13829 		       "%d:%d:%d:%d tag 0x%04x "
13830 		       "flag %#x status %x\n",
13831 			__func__,
13832 			io->io_hdr.io_type,
13833 			io->io_hdr.msg_type,
13834 			io->scsiio.cdb[0],
13835 			io->io_hdr.nexus.initid.id,
13836 			io->io_hdr.nexus.targ_port,
13837 			io->io_hdr.nexus.targ_target.id,
13838 			io->io_hdr.nexus.targ_lun,
13839 			(io->io_hdr.io_type ==
13840 			CTL_IO_TASK) ?
13841 			io->taskio.tag_num :
13842 			io->scsiio.tag_num,
13843 		        io->io_hdr.flags,
13844 			io->io_hdr.status);
13845 	} else
13846 		io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13847 #endif
13848 
13849 	/*
13850 	 * This is an internal copy of an I/O, and should not go through
13851 	 * the normal done processing logic.
13852 	 */
13853 	if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
13854 		return;
13855 
13856 	/*
13857 	 * We need to send a msg to the serializing shelf to finish the IO
13858 	 * as well.  We don't send a finish message to the other shelf if
13859 	 * this is a task management command.  Task management commands
13860 	 * aren't serialized in the OOA queue, but rather just executed on
13861 	 * both shelf controllers for commands that originated on that
13862 	 * controller.
13863 	 */
13864 	if ((io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)
13865 	 && (io->io_hdr.io_type != CTL_IO_TASK)) {
13866 		union ctl_ha_msg msg_io;
13867 
13868 		msg_io.hdr.msg_type = CTL_MSG_FINISH_IO;
13869 		msg_io.hdr.serializing_sc = io->io_hdr.serializing_sc;
13870 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_io,
13871 		    sizeof(msg_io), 0 ) != CTL_HA_STATUS_SUCCESS) {
13872 		}
13873 		/* continue on to finish IO */
13874 	}
13875 #ifdef CTL_IO_DELAY
13876 	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13877 		struct ctl_lun *lun;
13878 
13879 		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13880 
13881 		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13882 	} else {
13883 		struct ctl_lun *lun;
13884 
13885 		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13886 
13887 		if ((lun != NULL)
13888 		 && (lun->delay_info.done_delay > 0)) {
13889 			struct callout *callout;
13890 
13891 			callout = (struct callout *)&io->io_hdr.timer_bytes;
13892 			callout_init(callout, /*mpsafe*/ 1);
13893 			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13894 			callout_reset(callout,
13895 				      lun->delay_info.done_delay * hz,
13896 				      ctl_done_timer_wakeup, io);
13897 			if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13898 				lun->delay_info.done_delay = 0;
13899 			return;
13900 		}
13901 	}
13902 #endif /* CTL_IO_DELAY */
13903 
13904 	ctl_enqueue_done(io);
13905 }
13906 
13907 int
13908 ctl_isc(struct ctl_scsiio *ctsio)
13909 {
13910 	struct ctl_lun *lun;
13911 	int retval;
13912 
13913 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13914 
13915 	CTL_DEBUG_PRINT(("ctl_isc: command: %02x\n", ctsio->cdb[0]));
13916 
13917 	CTL_DEBUG_PRINT(("ctl_isc: calling data_submit()\n"));
13918 
13919 	retval = lun->backend->data_submit((union ctl_io *)ctsio);
13920 
13921 	return (retval);
13922 }
13923 
13924 
13925 static void
13926 ctl_work_thread(void *arg)
13927 {
13928 	struct ctl_thread *thr = (struct ctl_thread *)arg;
13929 	struct ctl_softc *softc = thr->ctl_softc;
13930 	union ctl_io *io;
13931 	int retval;
13932 
13933 	CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13934 
13935 	for (;;) {
13936 		retval = 0;
13937 
13938 		/*
13939 		 * We handle the queues in this order:
13940 		 * - ISC
13941 		 * - done queue (to free up resources, unblock other commands)
13942 		 * - RtR queue
13943 		 * - incoming queue
13944 		 *
13945 		 * If those queues are empty, we break out of the loop and
13946 		 * go to sleep.
13947 		 */
13948 		mtx_lock(&thr->queue_lock);
13949 		io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
13950 		if (io != NULL) {
13951 			STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
13952 			mtx_unlock(&thr->queue_lock);
13953 			ctl_handle_isc(io);
13954 			continue;
13955 		}
13956 		io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
13957 		if (io != NULL) {
13958 			STAILQ_REMOVE_HEAD(&thr->done_queue, links);
13959 			/* clear any blocked commands, call fe_done */
13960 			mtx_unlock(&thr->queue_lock);
13961 			retval = ctl_process_done(io);
13962 			continue;
13963 		}
13964 		io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
13965 		if (io != NULL) {
13966 			STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
13967 			mtx_unlock(&thr->queue_lock);
13968 			if (io->io_hdr.io_type == CTL_IO_TASK)
13969 				ctl_run_task(io);
13970 			else
13971 				ctl_scsiio_precheck(softc, &io->scsiio);
13972 			continue;
13973 		}
13974 		if (!ctl_pause_rtr) {
13975 			io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
13976 			if (io != NULL) {
13977 				STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
13978 				mtx_unlock(&thr->queue_lock);
13979 				retval = ctl_scsiio(&io->scsiio);
13980 				if (retval != CTL_RETVAL_COMPLETE)
13981 					CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13982 				continue;
13983 			}
13984 		}
13985 
13986 		/* Sleep until we have something to do. */
13987 		mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
13988 	}
13989 }
13990 
13991 static void
13992 ctl_lun_thread(void *arg)
13993 {
13994 	struct ctl_softc *softc = (struct ctl_softc *)arg;
13995 	struct ctl_be_lun *be_lun;
13996 	int retval;
13997 
13998 	CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
13999 
14000 	for (;;) {
14001 		retval = 0;
14002 		mtx_lock(&softc->ctl_lock);
14003 		be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
14004 		if (be_lun != NULL) {
14005 			STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
14006 			mtx_unlock(&softc->ctl_lock);
14007 			ctl_create_lun(be_lun);
14008 			continue;
14009 		}
14010 
14011 		/* Sleep until we have something to do. */
14012 		mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
14013 		    PDROP | PRIBIO, "-", 0);
14014 	}
14015 }
14016 
14017 static void
14018 ctl_enqueue_incoming(union ctl_io *io)
14019 {
14020 	struct ctl_softc *softc = control_softc;
14021 	struct ctl_thread *thr;
14022 	u_int idx;
14023 
14024 	idx = (io->io_hdr.nexus.targ_port * 127 +
14025 	       io->io_hdr.nexus.initid.id) % worker_threads;
14026 	thr = &softc->threads[idx];
14027 	mtx_lock(&thr->queue_lock);
14028 	STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
14029 	mtx_unlock(&thr->queue_lock);
14030 	wakeup(thr);
14031 }
14032 
14033 static void
14034 ctl_enqueue_rtr(union ctl_io *io)
14035 {
14036 	struct ctl_softc *softc = control_softc;
14037 	struct ctl_thread *thr;
14038 
14039 	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14040 	mtx_lock(&thr->queue_lock);
14041 	STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
14042 	mtx_unlock(&thr->queue_lock);
14043 	wakeup(thr);
14044 }
14045 
14046 static void
14047 ctl_enqueue_done(union ctl_io *io)
14048 {
14049 	struct ctl_softc *softc = control_softc;
14050 	struct ctl_thread *thr;
14051 
14052 	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14053 	mtx_lock(&thr->queue_lock);
14054 	STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
14055 	mtx_unlock(&thr->queue_lock);
14056 	wakeup(thr);
14057 }
14058 
14059 static void
14060 ctl_enqueue_isc(union ctl_io *io)
14061 {
14062 	struct ctl_softc *softc = control_softc;
14063 	struct ctl_thread *thr;
14064 
14065 	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14066 	mtx_lock(&thr->queue_lock);
14067 	STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
14068 	mtx_unlock(&thr->queue_lock);
14069 	wakeup(thr);
14070 }
14071 
14072 /* Initialization and failover */
14073 
14074 void
14075 ctl_init_isc_msg(void)
14076 {
14077 	printf("CTL: Still calling this thing\n");
14078 }
14079 
14080 /*
14081  * Init component
14082  * 	Initializes component into configuration defined by bootMode
14083  *	(see hasc-sv.c)
14084  *  	returns hasc_Status:
14085  * 		OK
14086  *		ERROR - fatal error
14087  */
14088 static ctl_ha_comp_status
14089 ctl_isc_init(struct ctl_ha_component *c)
14090 {
14091 	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
14092 
14093 	c->status = ret;
14094 	return ret;
14095 }
14096 
14097 /* Start component
14098  * 	Starts component in state requested. If component starts successfully,
14099  *	it must set its own state to the requestrd state
14100  *	When requested state is HASC_STATE_HA, the component may refine it
14101  * 	by adding _SLAVE or _MASTER flags.
14102  *	Currently allowed state transitions are:
14103  *	UNKNOWN->HA		- initial startup
14104  *	UNKNOWN->SINGLE - initial startup when no parter detected
14105  *	HA->SINGLE		- failover
14106  * returns ctl_ha_comp_status:
14107  * 		OK	- component successfully started in requested state
14108  *		FAILED  - could not start the requested state, failover may
14109  * 			  be possible
14110  *		ERROR	- fatal error detected, no future startup possible
14111  */
14112 static ctl_ha_comp_status
14113 ctl_isc_start(struct ctl_ha_component *c, ctl_ha_state state)
14114 {
14115 	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
14116 
14117 	printf("%s: go\n", __func__);
14118 
14119 	// UNKNOWN->HA or UNKNOWN->SINGLE (bootstrap)
14120 	if (c->state == CTL_HA_STATE_UNKNOWN ) {
14121 		ctl_is_single = 0;
14122 		if (ctl_ha_msg_create(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
14123 		    != CTL_HA_STATUS_SUCCESS) {
14124 			printf("ctl_isc_start: ctl_ha_msg_create failed.\n");
14125 			ret = CTL_HA_COMP_STATUS_ERROR;
14126 		}
14127 	} else if (CTL_HA_STATE_IS_HA(c->state)
14128 		&& CTL_HA_STATE_IS_SINGLE(state)){
14129 		// HA->SINGLE transition
14130 	        ctl_failover();
14131 		ctl_is_single = 1;
14132 	} else {
14133 		printf("ctl_isc_start:Invalid state transition %X->%X\n",
14134 		       c->state, state);
14135 		ret = CTL_HA_COMP_STATUS_ERROR;
14136 	}
14137 	if (CTL_HA_STATE_IS_SINGLE(state))
14138 		ctl_is_single = 1;
14139 
14140 	c->state = state;
14141 	c->status = ret;
14142 	return ret;
14143 }
14144 
14145 /*
14146  * Quiesce component
14147  * The component must clear any error conditions (set status to OK) and
14148  * prepare itself to another Start call
14149  * returns ctl_ha_comp_status:
14150  * 	OK
14151  *	ERROR
14152  */
14153 static ctl_ha_comp_status
14154 ctl_isc_quiesce(struct ctl_ha_component *c)
14155 {
14156 	int ret = CTL_HA_COMP_STATUS_OK;
14157 
14158 	ctl_pause_rtr = 1;
14159 	c->status = ret;
14160 	return ret;
14161 }
14162 
14163 struct ctl_ha_component ctl_ha_component_ctlisc =
14164 {
14165 	.name = "CTL ISC",
14166 	.state = CTL_HA_STATE_UNKNOWN,
14167 	.init = ctl_isc_init,
14168 	.start = ctl_isc_start,
14169 	.quiesce = ctl_isc_quiesce
14170 };
14171 
14172 /*
14173  *  vim: ts=8
14174  */
14175