xref: /freebsd/sys/cam/ctl/ctl.c (revision aeaed508982227551b2748339033bb2483382b4d)
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) and
324  * Logical Block Provisioning (0xB2)
325  */
326 #define SCSI_EVPD_NUM_SUPPORTED_PAGES	7
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_lbp(struct ctl_scsiio *ctsio, int alloc_len);
387 static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
388 static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
389 static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint32_t *len);
390 static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2);
391 static ctl_action ctl_check_for_blockage(union ctl_io *pending_io,
392 					 union ctl_io *ooa_io);
393 static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
394 				union ctl_io *starting_io);
395 static int ctl_check_blocked(struct ctl_lun *lun);
396 static int ctl_scsiio_lun_check(struct ctl_softc *ctl_softc,
397 				struct ctl_lun *lun,
398 				const struct ctl_cmd_entry *entry,
399 				struct ctl_scsiio *ctsio);
400 //static int ctl_check_rtr(union ctl_io *pending_io, struct ctl_softc *softc);
401 static void ctl_failover(void);
402 static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
403 			       struct ctl_scsiio *ctsio);
404 static int ctl_scsiio(struct ctl_scsiio *ctsio);
405 
406 static int ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
407 static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
408 			    ctl_ua_type ua_type);
409 static int ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io,
410 			 ctl_ua_type ua_type);
411 static int ctl_abort_task(union ctl_io *io);
412 static int ctl_abort_task_set(union ctl_io *io);
413 static int ctl_i_t_nexus_reset(union ctl_io *io);
414 static void ctl_run_task(union ctl_io *io);
415 #ifdef CTL_IO_DELAY
416 static void ctl_datamove_timer_wakeup(void *arg);
417 static void ctl_done_timer_wakeup(void *arg);
418 #endif /* CTL_IO_DELAY */
419 
420 static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
421 static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
422 static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
423 static void ctl_datamove_remote_write(union ctl_io *io);
424 static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
425 static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
426 static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
427 static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
428 				    ctl_ha_dt_cb callback);
429 static void ctl_datamove_remote_read(union ctl_io *io);
430 static void ctl_datamove_remote(union ctl_io *io);
431 static int ctl_process_done(union ctl_io *io);
432 static void ctl_lun_thread(void *arg);
433 static void ctl_work_thread(void *arg);
434 static void ctl_enqueue_incoming(union ctl_io *io);
435 static void ctl_enqueue_rtr(union ctl_io *io);
436 static void ctl_enqueue_done(union ctl_io *io);
437 static void ctl_enqueue_isc(union ctl_io *io);
438 static const struct ctl_cmd_entry *
439     ctl_get_cmd_entry(struct ctl_scsiio *ctsio);
440 static const struct ctl_cmd_entry *
441     ctl_validate_command(struct ctl_scsiio *ctsio);
442 static int ctl_cmd_applicable(uint8_t lun_type,
443     const struct ctl_cmd_entry *entry);
444 
445 /*
446  * Load the serialization table.  This isn't very pretty, but is probably
447  * the easiest way to do it.
448  */
449 #include "ctl_ser_table.c"
450 
451 /*
452  * We only need to define open, close and ioctl routines for this driver.
453  */
454 static struct cdevsw ctl_cdevsw = {
455 	.d_version =	D_VERSION,
456 	.d_flags =	0,
457 	.d_open =	ctl_open,
458 	.d_close =	ctl_close,
459 	.d_ioctl =	ctl_ioctl,
460 	.d_name =	"ctl",
461 };
462 
463 
464 MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
465 MALLOC_DEFINE(M_CTLIO, "ctlio", "Memory used for CTL requests");
466 
467 static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
468 
469 static moduledata_t ctl_moduledata = {
470 	"ctl",
471 	ctl_module_event_handler,
472 	NULL
473 };
474 
475 DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
476 MODULE_VERSION(ctl, 1);
477 
478 static struct ctl_frontend ioctl_frontend =
479 {
480 	.name = "ioctl",
481 };
482 
483 static void
484 ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
485 			    union ctl_ha_msg *msg_info)
486 {
487 	struct ctl_scsiio *ctsio;
488 
489 	if (msg_info->hdr.original_sc == NULL) {
490 		printf("%s: original_sc == NULL!\n", __func__);
491 		/* XXX KDM now what? */
492 		return;
493 	}
494 
495 	ctsio = &msg_info->hdr.original_sc->scsiio;
496 	ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
497 	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
498 	ctsio->io_hdr.status = msg_info->hdr.status;
499 	ctsio->scsi_status = msg_info->scsi.scsi_status;
500 	ctsio->sense_len = msg_info->scsi.sense_len;
501 	ctsio->sense_residual = msg_info->scsi.sense_residual;
502 	ctsio->residual = msg_info->scsi.residual;
503 	memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
504 	       sizeof(ctsio->sense_data));
505 	memcpy(&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
506 	       &msg_info->scsi.lbalen, sizeof(msg_info->scsi.lbalen));
507 	ctl_enqueue_isc((union ctl_io *)ctsio);
508 }
509 
510 static void
511 ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
512 				union ctl_ha_msg *msg_info)
513 {
514 	struct ctl_scsiio *ctsio;
515 
516 	if (msg_info->hdr.serializing_sc == NULL) {
517 		printf("%s: serializing_sc == NULL!\n", __func__);
518 		/* XXX KDM now what? */
519 		return;
520 	}
521 
522 	ctsio = &msg_info->hdr.serializing_sc->scsiio;
523 #if 0
524 	/*
525 	 * Attempt to catch the situation where an I/O has
526 	 * been freed, and we're using it again.
527 	 */
528 	if (ctsio->io_hdr.io_type == 0xff) {
529 		union ctl_io *tmp_io;
530 		tmp_io = (union ctl_io *)ctsio;
531 		printf("%s: %p use after free!\n", __func__,
532 		       ctsio);
533 		printf("%s: type %d msg %d cdb %x iptl: "
534 		       "%d:%d:%d:%d tag 0x%04x "
535 		       "flag %#x status %x\n",
536 			__func__,
537 			tmp_io->io_hdr.io_type,
538 			tmp_io->io_hdr.msg_type,
539 			tmp_io->scsiio.cdb[0],
540 			tmp_io->io_hdr.nexus.initid.id,
541 			tmp_io->io_hdr.nexus.targ_port,
542 			tmp_io->io_hdr.nexus.targ_target.id,
543 			tmp_io->io_hdr.nexus.targ_lun,
544 			(tmp_io->io_hdr.io_type ==
545 			CTL_IO_TASK) ?
546 			tmp_io->taskio.tag_num :
547 			tmp_io->scsiio.tag_num,
548 		        tmp_io->io_hdr.flags,
549 			tmp_io->io_hdr.status);
550 	}
551 #endif
552 	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
553 	ctl_enqueue_isc((union ctl_io *)ctsio);
554 }
555 
556 /*
557  * ISC (Inter Shelf Communication) event handler.  Events from the HA
558  * subsystem come in here.
559  */
560 static void
561 ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
562 {
563 	struct ctl_softc *ctl_softc;
564 	union ctl_io *io;
565 	struct ctl_prio *presio;
566 	ctl_ha_status isc_status;
567 
568 	ctl_softc = control_softc;
569 	io = NULL;
570 
571 
572 #if 0
573 	printf("CTL: Isc Msg event %d\n", event);
574 #endif
575 	if (event == CTL_HA_EVT_MSG_RECV) {
576 		union ctl_ha_msg msg_info;
577 
578 		isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
579 					     sizeof(msg_info), /*wait*/ 0);
580 #if 0
581 		printf("CTL: msg_type %d\n", msg_info.msg_type);
582 #endif
583 		if (isc_status != 0) {
584 			printf("Error receiving message, status = %d\n",
585 			       isc_status);
586 			return;
587 		}
588 
589 		switch (msg_info.hdr.msg_type) {
590 		case CTL_MSG_SERIALIZE:
591 #if 0
592 			printf("Serialize\n");
593 #endif
594 			io = ctl_alloc_io((void *)ctl_softc->othersc_pool);
595 			if (io == NULL) {
596 				printf("ctl_isc_event_handler: can't allocate "
597 				       "ctl_io!\n");
598 				/* Bad Juju */
599 				/* Need to set busy and send msg back */
600 				msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
601 				msg_info.hdr.status = CTL_SCSI_ERROR;
602 				msg_info.scsi.scsi_status = SCSI_STATUS_BUSY;
603 				msg_info.scsi.sense_len = 0;
604 			        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
605 				    sizeof(msg_info), 0) > CTL_HA_STATUS_SUCCESS){
606 				}
607 				goto bailout;
608 			}
609 			ctl_zero_io(io);
610 			// populate ctsio from msg_info
611 			io->io_hdr.io_type = CTL_IO_SCSI;
612 			io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
613 			io->io_hdr.original_sc = msg_info.hdr.original_sc;
614 #if 0
615 			printf("pOrig %x\n", (int)msg_info.original_sc);
616 #endif
617 			io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
618 					    CTL_FLAG_IO_ACTIVE;
619 			/*
620 			 * If we're in serialization-only mode, we don't
621 			 * want to go through full done processing.  Thus
622 			 * the COPY flag.
623 			 *
624 			 * XXX KDM add another flag that is more specific.
625 			 */
626 			if (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)
627 				io->io_hdr.flags |= CTL_FLAG_INT_COPY;
628 			io->io_hdr.nexus = msg_info.hdr.nexus;
629 #if 0
630 			printf("targ %d, port %d, iid %d, lun %d\n",
631 			       io->io_hdr.nexus.targ_target.id,
632 			       io->io_hdr.nexus.targ_port,
633 			       io->io_hdr.nexus.initid.id,
634 			       io->io_hdr.nexus.targ_lun);
635 #endif
636 			io->scsiio.tag_num = msg_info.scsi.tag_num;
637 			io->scsiio.tag_type = msg_info.scsi.tag_type;
638 			memcpy(io->scsiio.cdb, msg_info.scsi.cdb,
639 			       CTL_MAX_CDBLEN);
640 			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
641 				const struct ctl_cmd_entry *entry;
642 
643 				entry = ctl_get_cmd_entry(&io->scsiio);
644 				io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
645 				io->io_hdr.flags |=
646 					entry->flags & CTL_FLAG_DATA_MASK;
647 			}
648 			ctl_enqueue_isc(io);
649 			break;
650 
651 		/* Performed on the Originating SC, XFER mode only */
652 		case CTL_MSG_DATAMOVE: {
653 			struct ctl_sg_entry *sgl;
654 			int i, j;
655 
656 			io = msg_info.hdr.original_sc;
657 			if (io == NULL) {
658 				printf("%s: original_sc == NULL!\n", __func__);
659 				/* XXX KDM do something here */
660 				break;
661 			}
662 			io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
663 			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
664 			/*
665 			 * Keep track of this, we need to send it back over
666 			 * when the datamove is complete.
667 			 */
668 			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
669 
670 			if (msg_info.dt.sg_sequence == 0) {
671 				/*
672 				 * XXX KDM we use the preallocated S/G list
673 				 * here, but we'll need to change this to
674 				 * dynamic allocation if we need larger S/G
675 				 * lists.
676 				 */
677 				if (msg_info.dt.kern_sg_entries >
678 				    sizeof(io->io_hdr.remote_sglist) /
679 				    sizeof(io->io_hdr.remote_sglist[0])) {
680 					printf("%s: number of S/G entries "
681 					    "needed %u > allocated num %zd\n",
682 					    __func__,
683 					    msg_info.dt.kern_sg_entries,
684 					    sizeof(io->io_hdr.remote_sglist)/
685 					    sizeof(io->io_hdr.remote_sglist[0]));
686 
687 					/*
688 					 * XXX KDM send a message back to
689 					 * the other side to shut down the
690 					 * DMA.  The error will come back
691 					 * through via the normal channel.
692 					 */
693 					break;
694 				}
695 				sgl = io->io_hdr.remote_sglist;
696 				memset(sgl, 0,
697 				       sizeof(io->io_hdr.remote_sglist));
698 
699 				io->scsiio.kern_data_ptr = (uint8_t *)sgl;
700 
701 				io->scsiio.kern_sg_entries =
702 					msg_info.dt.kern_sg_entries;
703 				io->scsiio.rem_sg_entries =
704 					msg_info.dt.kern_sg_entries;
705 				io->scsiio.kern_data_len =
706 					msg_info.dt.kern_data_len;
707 				io->scsiio.kern_total_len =
708 					msg_info.dt.kern_total_len;
709 				io->scsiio.kern_data_resid =
710 					msg_info.dt.kern_data_resid;
711 				io->scsiio.kern_rel_offset =
712 					msg_info.dt.kern_rel_offset;
713 				/*
714 				 * Clear out per-DMA flags.
715 				 */
716 				io->io_hdr.flags &= ~CTL_FLAG_RDMA_MASK;
717 				/*
718 				 * Add per-DMA flags that are set for this
719 				 * particular DMA request.
720 				 */
721 				io->io_hdr.flags |= msg_info.dt.flags &
722 						    CTL_FLAG_RDMA_MASK;
723 			} else
724 				sgl = (struct ctl_sg_entry *)
725 					io->scsiio.kern_data_ptr;
726 
727 			for (i = msg_info.dt.sent_sg_entries, j = 0;
728 			     i < (msg_info.dt.sent_sg_entries +
729 			     msg_info.dt.cur_sg_entries); i++, j++) {
730 				sgl[i].addr = msg_info.dt.sg_list[j].addr;
731 				sgl[i].len = msg_info.dt.sg_list[j].len;
732 
733 #if 0
734 				printf("%s: L: %p,%d -> %p,%d j=%d, i=%d\n",
735 				       __func__,
736 				       msg_info.dt.sg_list[j].addr,
737 				       msg_info.dt.sg_list[j].len,
738 				       sgl[i].addr, sgl[i].len, j, i);
739 #endif
740 			}
741 #if 0
742 			memcpy(&sgl[msg_info.dt.sent_sg_entries],
743 			       msg_info.dt.sg_list,
744 			       sizeof(*sgl) * msg_info.dt.cur_sg_entries);
745 #endif
746 
747 			/*
748 			 * If this is the last piece of the I/O, we've got
749 			 * the full S/G list.  Queue processing in the thread.
750 			 * Otherwise wait for the next piece.
751 			 */
752 			if (msg_info.dt.sg_last != 0)
753 				ctl_enqueue_isc(io);
754 			break;
755 		}
756 		/* Performed on the Serializing (primary) SC, XFER mode only */
757 		case CTL_MSG_DATAMOVE_DONE: {
758 			if (msg_info.hdr.serializing_sc == NULL) {
759 				printf("%s: serializing_sc == NULL!\n",
760 				       __func__);
761 				/* XXX KDM now what? */
762 				break;
763 			}
764 			/*
765 			 * We grab the sense information here in case
766 			 * there was a failure, so we can return status
767 			 * back to the initiator.
768 			 */
769 			io = msg_info.hdr.serializing_sc;
770 			io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
771 			io->io_hdr.status = msg_info.hdr.status;
772 			io->scsiio.scsi_status = msg_info.scsi.scsi_status;
773 			io->scsiio.sense_len = msg_info.scsi.sense_len;
774 			io->scsiio.sense_residual =msg_info.scsi.sense_residual;
775 			io->io_hdr.port_status = msg_info.scsi.fetd_status;
776 			io->scsiio.residual = msg_info.scsi.residual;
777 			memcpy(&io->scsiio.sense_data,&msg_info.scsi.sense_data,
778 			       sizeof(io->scsiio.sense_data));
779 			ctl_enqueue_isc(io);
780 			break;
781 		}
782 
783 		/* Preformed on Originating SC, SER_ONLY mode */
784 		case CTL_MSG_R2R:
785 			io = msg_info.hdr.original_sc;
786 			if (io == NULL) {
787 				printf("%s: Major Bummer\n", __func__);
788 				return;
789 			} else {
790 #if 0
791 				printf("pOrig %x\n",(int) ctsio);
792 #endif
793 			}
794 			io->io_hdr.msg_type = CTL_MSG_R2R;
795 			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
796 			ctl_enqueue_isc(io);
797 			break;
798 
799 		/*
800 		 * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
801 		 * mode.
802 		 * Performed on the Originating (i.e. secondary) SC in XFER
803 		 * mode
804 		 */
805 		case CTL_MSG_FINISH_IO:
806 			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER)
807 				ctl_isc_handler_finish_xfer(ctl_softc,
808 							    &msg_info);
809 			else
810 				ctl_isc_handler_finish_ser_only(ctl_softc,
811 								&msg_info);
812 			break;
813 
814 		/* Preformed on Originating SC */
815 		case CTL_MSG_BAD_JUJU:
816 			io = msg_info.hdr.original_sc;
817 			if (io == NULL) {
818 				printf("%s: Bad JUJU!, original_sc is NULL!\n",
819 				       __func__);
820 				break;
821 			}
822 			ctl_copy_sense_data(&msg_info, io);
823 			/*
824 			 * IO should have already been cleaned up on other
825 			 * SC so clear this flag so we won't send a message
826 			 * back to finish the IO there.
827 			 */
828 			io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
829 			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
830 
831 			/* io = msg_info.hdr.serializing_sc; */
832 			io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
833 			ctl_enqueue_isc(io);
834 			break;
835 
836 		/* Handle resets sent from the other side */
837 		case CTL_MSG_MANAGE_TASKS: {
838 			struct ctl_taskio *taskio;
839 			taskio = (struct ctl_taskio *)ctl_alloc_io(
840 				(void *)ctl_softc->othersc_pool);
841 			if (taskio == NULL) {
842 				printf("ctl_isc_event_handler: can't allocate "
843 				       "ctl_io!\n");
844 				/* Bad Juju */
845 				/* should I just call the proper reset func
846 				   here??? */
847 				goto bailout;
848 			}
849 			ctl_zero_io((union ctl_io *)taskio);
850 			taskio->io_hdr.io_type = CTL_IO_TASK;
851 			taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
852 			taskio->io_hdr.nexus = msg_info.hdr.nexus;
853 			taskio->task_action = msg_info.task.task_action;
854 			taskio->tag_num = msg_info.task.tag_num;
855 			taskio->tag_type = msg_info.task.tag_type;
856 #ifdef CTL_TIME_IO
857 			taskio->io_hdr.start_time = time_uptime;
858 			getbintime(&taskio->io_hdr.start_bt);
859 #if 0
860 			cs_prof_gettime(&taskio->io_hdr.start_ticks);
861 #endif
862 #endif /* CTL_TIME_IO */
863 			ctl_run_task((union ctl_io *)taskio);
864 			break;
865 		}
866 		/* Persistent Reserve action which needs attention */
867 		case CTL_MSG_PERS_ACTION:
868 			presio = (struct ctl_prio *)ctl_alloc_io(
869 				(void *)ctl_softc->othersc_pool);
870 			if (presio == NULL) {
871 				printf("ctl_isc_event_handler: can't allocate "
872 				       "ctl_io!\n");
873 				/* Bad Juju */
874 				/* Need to set busy and send msg back */
875 				goto bailout;
876 			}
877 			ctl_zero_io((union ctl_io *)presio);
878 			presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
879 			presio->pr_msg = msg_info.pr;
880 			ctl_enqueue_isc((union ctl_io *)presio);
881 			break;
882 		case CTL_MSG_SYNC_FE:
883 			rcv_sync_msg = 1;
884 			break;
885 		case CTL_MSG_APS_LOCK: {
886 			// It's quicker to execute this then to
887 			// queue it.
888 			struct ctl_lun *lun;
889 			struct ctl_page_index *page_index;
890 			struct copan_aps_subpage *current_sp;
891 			uint32_t targ_lun;
892 
893 			targ_lun = msg_info.hdr.nexus.targ_mapped_lun;
894 			lun = ctl_softc->ctl_luns[targ_lun];
895 			mtx_lock(&lun->lun_lock);
896 			page_index = &lun->mode_pages.index[index_to_aps_page];
897 			current_sp = (struct copan_aps_subpage *)
898 				     (page_index->page_data +
899 				     (page_index->page_len * CTL_PAGE_CURRENT));
900 
901 			current_sp->lock_active = msg_info.aps.lock_flag;
902 			mtx_unlock(&lun->lun_lock);
903 		        break;
904 		}
905 		default:
906 		        printf("How did I get here?\n");
907 		}
908 	} else if (event == CTL_HA_EVT_MSG_SENT) {
909 		if (param != CTL_HA_STATUS_SUCCESS) {
910 			printf("Bad status from ctl_ha_msg_send status %d\n",
911 			       param);
912 		}
913 		return;
914 	} else if (event == CTL_HA_EVT_DISCONNECT) {
915 		printf("CTL: Got a disconnect from Isc\n");
916 		return;
917 	} else {
918 		printf("ctl_isc_event_handler: Unknown event %d\n", event);
919 		return;
920 	}
921 
922 bailout:
923 	return;
924 }
925 
926 static void
927 ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
928 {
929 	struct scsi_sense_data *sense;
930 
931 	sense = &dest->scsiio.sense_data;
932 	bcopy(&src->scsi.sense_data, sense, sizeof(*sense));
933 	dest->scsiio.scsi_status = src->scsi.scsi_status;
934 	dest->scsiio.sense_len = src->scsi.sense_len;
935 	dest->io_hdr.status = src->hdr.status;
936 }
937 
938 static int
939 ctl_init(void)
940 {
941 	struct ctl_softc *softc;
942 	struct ctl_io_pool *internal_pool, *emergency_pool, *other_pool;
943 	struct ctl_port *port;
944         uint8_t sc_id =0;
945 	int i, error, retval;
946 	//int isc_retval;
947 
948 	retval = 0;
949 	ctl_pause_rtr = 0;
950         rcv_sync_msg = 0;
951 
952 	control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
953 			       M_WAITOK | M_ZERO);
954 	softc = control_softc;
955 
956 	softc->dev = make_dev(&ctl_cdevsw, 0, UID_ROOT, GID_OPERATOR, 0600,
957 			      "cam/ctl");
958 
959 	softc->dev->si_drv1 = softc;
960 
961 	/*
962 	 * By default, return a "bad LUN" peripheral qualifier for unknown
963 	 * LUNs.  The user can override this default using the tunable or
964 	 * sysctl.  See the comment in ctl_inquiry_std() for more details.
965 	 */
966 	softc->inquiry_pq_no_lun = 1;
967 	TUNABLE_INT_FETCH("kern.cam.ctl.inquiry_pq_no_lun",
968 			  &softc->inquiry_pq_no_lun);
969 	sysctl_ctx_init(&softc->sysctl_ctx);
970 	softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
971 		SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
972 		CTLFLAG_RD, 0, "CAM Target Layer");
973 
974 	if (softc->sysctl_tree == NULL) {
975 		printf("%s: unable to allocate sysctl tree\n", __func__);
976 		destroy_dev(softc->dev);
977 		free(control_softc, M_DEVBUF);
978 		control_softc = NULL;
979 		return (ENOMEM);
980 	}
981 
982 	SYSCTL_ADD_INT(&softc->sysctl_ctx,
983 		       SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO,
984 		       "inquiry_pq_no_lun", CTLFLAG_RW,
985 		       &softc->inquiry_pq_no_lun, 0,
986 		       "Report no lun possible for invalid LUNs");
987 
988 	mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
989 	mtx_init(&softc->pool_lock, "CTL pool mutex", NULL, MTX_DEF);
990 	softc->open_count = 0;
991 
992 	/*
993 	 * Default to actually sending a SYNCHRONIZE CACHE command down to
994 	 * the drive.
995 	 */
996 	softc->flags = CTL_FLAG_REAL_SYNC;
997 
998 	/*
999 	 * In Copan's HA scheme, the "master" and "slave" roles are
1000 	 * figured out through the slot the controller is in.  Although it
1001 	 * is an active/active system, someone has to be in charge.
1002  	 */
1003 #ifdef NEEDTOPORT
1004         scmicro_rw(SCMICRO_GET_SHELF_ID, &sc_id);
1005 #endif
1006 
1007         if (sc_id == 0) {
1008 		softc->flags |= CTL_FLAG_MASTER_SHELF;
1009 		persis_offset = 0;
1010 	} else
1011 		persis_offset = CTL_MAX_INITIATORS;
1012 
1013 	/*
1014 	 * XXX KDM need to figure out where we want to get our target ID
1015 	 * and WWID.  Is it different on each port?
1016 	 */
1017 	softc->target.id = 0;
1018 	softc->target.wwid[0] = 0x12345678;
1019 	softc->target.wwid[1] = 0x87654321;
1020 	STAILQ_INIT(&softc->lun_list);
1021 	STAILQ_INIT(&softc->pending_lun_queue);
1022 	STAILQ_INIT(&softc->fe_list);
1023 	STAILQ_INIT(&softc->port_list);
1024 	STAILQ_INIT(&softc->be_list);
1025 	STAILQ_INIT(&softc->io_pools);
1026 	ctl_tpc_init(softc);
1027 
1028 	if (ctl_pool_create(softc, CTL_POOL_INTERNAL, CTL_POOL_ENTRIES_INTERNAL,
1029 			    &internal_pool)!= 0){
1030 		printf("ctl: can't allocate %d entry internal pool, "
1031 		       "exiting\n", CTL_POOL_ENTRIES_INTERNAL);
1032 		return (ENOMEM);
1033 	}
1034 
1035 	if (ctl_pool_create(softc, CTL_POOL_EMERGENCY,
1036 			    CTL_POOL_ENTRIES_EMERGENCY, &emergency_pool) != 0) {
1037 		printf("ctl: can't allocate %d entry emergency pool, "
1038 		       "exiting\n", CTL_POOL_ENTRIES_EMERGENCY);
1039 		ctl_pool_free(internal_pool);
1040 		return (ENOMEM);
1041 	}
1042 
1043 	if (ctl_pool_create(softc, CTL_POOL_4OTHERSC, CTL_POOL_ENTRIES_OTHER_SC,
1044 	                    &other_pool) != 0)
1045 	{
1046 		printf("ctl: can't allocate %d entry other SC pool, "
1047 		       "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1048 		ctl_pool_free(internal_pool);
1049 		ctl_pool_free(emergency_pool);
1050 		return (ENOMEM);
1051 	}
1052 
1053 	softc->internal_pool = internal_pool;
1054 	softc->emergency_pool = emergency_pool;
1055 	softc->othersc_pool = other_pool;
1056 
1057 	if (worker_threads <= 0)
1058 		worker_threads = max(1, mp_ncpus / 4);
1059 	if (worker_threads > CTL_MAX_THREADS)
1060 		worker_threads = CTL_MAX_THREADS;
1061 
1062 	for (i = 0; i < worker_threads; i++) {
1063 		struct ctl_thread *thr = &softc->threads[i];
1064 
1065 		mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1066 		thr->ctl_softc = softc;
1067 		STAILQ_INIT(&thr->incoming_queue);
1068 		STAILQ_INIT(&thr->rtr_queue);
1069 		STAILQ_INIT(&thr->done_queue);
1070 		STAILQ_INIT(&thr->isc_queue);
1071 
1072 		error = kproc_kthread_add(ctl_work_thread, thr,
1073 		    &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1074 		if (error != 0) {
1075 			printf("error creating CTL work thread!\n");
1076 			ctl_pool_free(internal_pool);
1077 			ctl_pool_free(emergency_pool);
1078 			ctl_pool_free(other_pool);
1079 			return (error);
1080 		}
1081 	}
1082 	error = kproc_kthread_add(ctl_lun_thread, softc,
1083 	    &softc->ctl_proc, NULL, 0, 0, "ctl", "lun");
1084 	if (error != 0) {
1085 		printf("error creating CTL lun thread!\n");
1086 		ctl_pool_free(internal_pool);
1087 		ctl_pool_free(emergency_pool);
1088 		ctl_pool_free(other_pool);
1089 		return (error);
1090 	}
1091 	if (bootverbose)
1092 		printf("ctl: CAM Target Layer loaded\n");
1093 
1094 	/*
1095 	 * Initialize the ioctl front end.
1096 	 */
1097 	ctl_frontend_register(&ioctl_frontend);
1098 	port = &softc->ioctl_info.port;
1099 	port->frontend = &ioctl_frontend;
1100 	sprintf(softc->ioctl_info.port_name, "ioctl");
1101 	port->port_type = CTL_PORT_IOCTL;
1102 	port->num_requested_ctl_io = 100;
1103 	port->port_name = softc->ioctl_info.port_name;
1104 	port->port_online = ctl_ioctl_online;
1105 	port->port_offline = ctl_ioctl_offline;
1106 	port->onoff_arg = &softc->ioctl_info;
1107 	port->lun_enable = ctl_ioctl_lun_enable;
1108 	port->lun_disable = ctl_ioctl_lun_disable;
1109 	port->targ_lun_arg = &softc->ioctl_info;
1110 	port->fe_datamove = ctl_ioctl_datamove;
1111 	port->fe_done = ctl_ioctl_done;
1112 	port->max_targets = 15;
1113 	port->max_target_id = 15;
1114 
1115 	if (ctl_port_register(&softc->ioctl_info.port,
1116 	                  (softc->flags & CTL_FLAG_MASTER_SHELF)) != 0) {
1117 		printf("ctl: ioctl front end registration failed, will "
1118 		       "continue anyway\n");
1119 	}
1120 
1121 #ifdef CTL_IO_DELAY
1122 	if (sizeof(struct callout) > CTL_TIMER_BYTES) {
1123 		printf("sizeof(struct callout) %zd > CTL_TIMER_BYTES %zd\n",
1124 		       sizeof(struct callout), CTL_TIMER_BYTES);
1125 		return (EINVAL);
1126 	}
1127 #endif /* CTL_IO_DELAY */
1128 
1129 	return (0);
1130 }
1131 
1132 void
1133 ctl_shutdown(void)
1134 {
1135 	struct ctl_softc *softc;
1136 	struct ctl_lun *lun, *next_lun;
1137 	struct ctl_io_pool *pool;
1138 
1139 	softc = (struct ctl_softc *)control_softc;
1140 
1141 	if (ctl_port_deregister(&softc->ioctl_info.port) != 0)
1142 		printf("ctl: ioctl front end deregistration failed\n");
1143 
1144 	mtx_lock(&softc->ctl_lock);
1145 
1146 	/*
1147 	 * Free up each LUN.
1148 	 */
1149 	for (lun = STAILQ_FIRST(&softc->lun_list); lun != NULL; lun = next_lun){
1150 		next_lun = STAILQ_NEXT(lun, links);
1151 		ctl_free_lun(lun);
1152 	}
1153 
1154 	mtx_unlock(&softc->ctl_lock);
1155 
1156 	ctl_frontend_deregister(&ioctl_frontend);
1157 
1158 	/*
1159 	 * This will rip the rug out from under any FETDs or anyone else
1160 	 * that has a pool allocated.  Since we increment our module
1161 	 * refcount any time someone outside the main CTL module allocates
1162 	 * a pool, we shouldn't have any problems here.  The user won't be
1163 	 * able to unload the CTL module until client modules have
1164 	 * successfully unloaded.
1165 	 */
1166 	while ((pool = STAILQ_FIRST(&softc->io_pools)) != NULL)
1167 		ctl_pool_free(pool);
1168 
1169 #if 0
1170 	ctl_shutdown_thread(softc->work_thread);
1171 	mtx_destroy(&softc->queue_lock);
1172 #endif
1173 
1174 	ctl_tpc_shutdown(softc);
1175 	mtx_destroy(&softc->pool_lock);
1176 	mtx_destroy(&softc->ctl_lock);
1177 
1178 	destroy_dev(softc->dev);
1179 
1180 	sysctl_ctx_free(&softc->sysctl_ctx);
1181 
1182 	free(control_softc, M_DEVBUF);
1183 	control_softc = NULL;
1184 
1185 	if (bootverbose)
1186 		printf("ctl: CAM Target Layer unloaded\n");
1187 }
1188 
1189 static int
1190 ctl_module_event_handler(module_t mod, int what, void *arg)
1191 {
1192 
1193 	switch (what) {
1194 	case MOD_LOAD:
1195 		return (ctl_init());
1196 	case MOD_UNLOAD:
1197 		return (EBUSY);
1198 	default:
1199 		return (EOPNOTSUPP);
1200 	}
1201 }
1202 
1203 /*
1204  * XXX KDM should we do some access checks here?  Bump a reference count to
1205  * prevent a CTL module from being unloaded while someone has it open?
1206  */
1207 static int
1208 ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1209 {
1210 	return (0);
1211 }
1212 
1213 static int
1214 ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1215 {
1216 	return (0);
1217 }
1218 
1219 int
1220 ctl_port_enable(ctl_port_type port_type)
1221 {
1222 	struct ctl_softc *softc;
1223 	struct ctl_port *port;
1224 
1225 	if (ctl_is_single == 0) {
1226 		union ctl_ha_msg msg_info;
1227 		int isc_retval;
1228 
1229 #if 0
1230 		printf("%s: HA mode, synchronizing frontend enable\n",
1231 		        __func__);
1232 #endif
1233 		msg_info.hdr.msg_type = CTL_MSG_SYNC_FE;
1234 	        if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1235 		        sizeof(msg_info), 1 )) > CTL_HA_STATUS_SUCCESS) {
1236 			printf("Sync msg send error retval %d\n", isc_retval);
1237 		}
1238 		if (!rcv_sync_msg) {
1239 			isc_retval=ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
1240 			        sizeof(msg_info), 1);
1241 		}
1242 #if 0
1243         	printf("CTL:Frontend Enable\n");
1244 	} else {
1245 		printf("%s: single mode, skipping frontend synchronization\n",
1246 		        __func__);
1247 #endif
1248 	}
1249 
1250 	softc = control_softc;
1251 
1252 	STAILQ_FOREACH(port, &softc->port_list, links) {
1253 		if (port_type & port->port_type)
1254 		{
1255 #if 0
1256 			printf("port %d\n", port->targ_port);
1257 #endif
1258 			ctl_port_online(port);
1259 		}
1260 	}
1261 
1262 	return (0);
1263 }
1264 
1265 int
1266 ctl_port_disable(ctl_port_type port_type)
1267 {
1268 	struct ctl_softc *softc;
1269 	struct ctl_port *port;
1270 
1271 	softc = control_softc;
1272 
1273 	STAILQ_FOREACH(port, &softc->port_list, links) {
1274 		if (port_type & port->port_type)
1275 			ctl_port_offline(port);
1276 	}
1277 
1278 	return (0);
1279 }
1280 
1281 /*
1282  * Returns 0 for success, 1 for failure.
1283  * Currently the only failure mode is if there aren't enough entries
1284  * allocated.  So, in case of a failure, look at num_entries_dropped,
1285  * reallocate and try again.
1286  */
1287 int
1288 ctl_port_list(struct ctl_port_entry *entries, int num_entries_alloced,
1289 	      int *num_entries_filled, int *num_entries_dropped,
1290 	      ctl_port_type port_type, int no_virtual)
1291 {
1292 	struct ctl_softc *softc;
1293 	struct ctl_port *port;
1294 	int entries_dropped, entries_filled;
1295 	int retval;
1296 	int i;
1297 
1298 	softc = control_softc;
1299 
1300 	retval = 0;
1301 	entries_filled = 0;
1302 	entries_dropped = 0;
1303 
1304 	i = 0;
1305 	mtx_lock(&softc->ctl_lock);
1306 	STAILQ_FOREACH(port, &softc->port_list, links) {
1307 		struct ctl_port_entry *entry;
1308 
1309 		if ((port->port_type & port_type) == 0)
1310 			continue;
1311 
1312 		if ((no_virtual != 0)
1313 		 && (port->virtual_port != 0))
1314 			continue;
1315 
1316 		if (entries_filled >= num_entries_alloced) {
1317 			entries_dropped++;
1318 			continue;
1319 		}
1320 		entry = &entries[i];
1321 
1322 		entry->port_type = port->port_type;
1323 		strlcpy(entry->port_name, port->port_name,
1324 			sizeof(entry->port_name));
1325 		entry->physical_port = port->physical_port;
1326 		entry->virtual_port = port->virtual_port;
1327 		entry->wwnn = port->wwnn;
1328 		entry->wwpn = port->wwpn;
1329 
1330 		i++;
1331 		entries_filled++;
1332 	}
1333 
1334 	mtx_unlock(&softc->ctl_lock);
1335 
1336 	if (entries_dropped > 0)
1337 		retval = 1;
1338 
1339 	*num_entries_dropped = entries_dropped;
1340 	*num_entries_filled = entries_filled;
1341 
1342 	return (retval);
1343 }
1344 
1345 static void
1346 ctl_ioctl_online(void *arg)
1347 {
1348 	struct ctl_ioctl_info *ioctl_info;
1349 
1350 	ioctl_info = (struct ctl_ioctl_info *)arg;
1351 
1352 	ioctl_info->flags |= CTL_IOCTL_FLAG_ENABLED;
1353 }
1354 
1355 static void
1356 ctl_ioctl_offline(void *arg)
1357 {
1358 	struct ctl_ioctl_info *ioctl_info;
1359 
1360 	ioctl_info = (struct ctl_ioctl_info *)arg;
1361 
1362 	ioctl_info->flags &= ~CTL_IOCTL_FLAG_ENABLED;
1363 }
1364 
1365 /*
1366  * Remove an initiator by port number and initiator ID.
1367  * Returns 0 for success, -1 for failure.
1368  */
1369 int
1370 ctl_remove_initiator(struct ctl_port *port, int iid)
1371 {
1372 	struct ctl_softc *softc = control_softc;
1373 
1374 	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1375 
1376 	if (iid > CTL_MAX_INIT_PER_PORT) {
1377 		printf("%s: initiator ID %u > maximun %u!\n",
1378 		       __func__, iid, CTL_MAX_INIT_PER_PORT);
1379 		return (-1);
1380 	}
1381 
1382 	mtx_lock(&softc->ctl_lock);
1383 	port->wwpn_iid[iid].in_use--;
1384 	port->wwpn_iid[iid].last_use = time_uptime;
1385 	mtx_unlock(&softc->ctl_lock);
1386 
1387 	return (0);
1388 }
1389 
1390 /*
1391  * Add an initiator to the initiator map.
1392  * Returns iid for success, < 0 for failure.
1393  */
1394 int
1395 ctl_add_initiator(struct ctl_port *port, int iid, uint64_t wwpn, char *name)
1396 {
1397 	struct ctl_softc *softc = control_softc;
1398 	time_t best_time;
1399 	int i, best;
1400 
1401 	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1402 
1403 	if (iid >= CTL_MAX_INIT_PER_PORT) {
1404 		printf("%s: WWPN %#jx initiator ID %u > maximum %u!\n",
1405 		       __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
1406 		free(name, M_CTL);
1407 		return (-1);
1408 	}
1409 
1410 	mtx_lock(&softc->ctl_lock);
1411 
1412 	if (iid < 0 && (wwpn != 0 || name != NULL)) {
1413 		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1414 			if (wwpn != 0 && wwpn == port->wwpn_iid[i].wwpn) {
1415 				iid = i;
1416 				break;
1417 			}
1418 			if (name != NULL && port->wwpn_iid[i].name != NULL &&
1419 			    strcmp(name, port->wwpn_iid[i].name) == 0) {
1420 				iid = i;
1421 				break;
1422 			}
1423 		}
1424 	}
1425 
1426 	if (iid < 0) {
1427 		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1428 			if (port->wwpn_iid[i].in_use == 0 &&
1429 			    port->wwpn_iid[i].wwpn == 0 &&
1430 			    port->wwpn_iid[i].name == NULL) {
1431 				iid = i;
1432 				break;
1433 			}
1434 		}
1435 	}
1436 
1437 	if (iid < 0) {
1438 		best = -1;
1439 		best_time = INT32_MAX;
1440 		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1441 			if (port->wwpn_iid[i].in_use == 0) {
1442 				if (port->wwpn_iid[i].last_use < best_time) {
1443 					best = i;
1444 					best_time = port->wwpn_iid[i].last_use;
1445 				}
1446 			}
1447 		}
1448 		iid = best;
1449 	}
1450 
1451 	if (iid < 0) {
1452 		mtx_unlock(&softc->ctl_lock);
1453 		free(name, M_CTL);
1454 		return (-2);
1455 	}
1456 
1457 	if (port->wwpn_iid[iid].in_use > 0 && (wwpn != 0 || name != NULL)) {
1458 		/*
1459 		 * This is not an error yet.
1460 		 */
1461 		if (wwpn != 0 && wwpn == port->wwpn_iid[iid].wwpn) {
1462 #if 0
1463 			printf("%s: port %d iid %u WWPN %#jx arrived"
1464 			    " again\n", __func__, port->targ_port,
1465 			    iid, (uintmax_t)wwpn);
1466 #endif
1467 			goto take;
1468 		}
1469 		if (name != NULL && port->wwpn_iid[iid].name != NULL &&
1470 		    strcmp(name, port->wwpn_iid[iid].name) == 0) {
1471 #if 0
1472 			printf("%s: port %d iid %u name '%s' arrived"
1473 			    " again\n", __func__, port->targ_port,
1474 			    iid, name);
1475 #endif
1476 			goto take;
1477 		}
1478 
1479 		/*
1480 		 * This is an error, but what do we do about it?  The
1481 		 * driver is telling us we have a new WWPN for this
1482 		 * initiator ID, so we pretty much need to use it.
1483 		 */
1484 		printf("%s: port %d iid %u WWPN %#jx '%s' arrived,"
1485 		    " but WWPN %#jx '%s' is still at that address\n",
1486 		    __func__, port->targ_port, iid, wwpn, name,
1487 		    (uintmax_t)port->wwpn_iid[iid].wwpn,
1488 		    port->wwpn_iid[iid].name);
1489 
1490 		/*
1491 		 * XXX KDM clear have_ca and ua_pending on each LUN for
1492 		 * this initiator.
1493 		 */
1494 	}
1495 take:
1496 	free(port->wwpn_iid[iid].name, M_CTL);
1497 	port->wwpn_iid[iid].name = name;
1498 	port->wwpn_iid[iid].wwpn = wwpn;
1499 	port->wwpn_iid[iid].in_use++;
1500 	mtx_unlock(&softc->ctl_lock);
1501 
1502 	return (iid);
1503 }
1504 
1505 static int
1506 ctl_create_iid(struct ctl_port *port, int iid, uint8_t *buf)
1507 {
1508 	int len;
1509 
1510 	switch (port->port_type) {
1511 	case CTL_PORT_FC:
1512 	{
1513 		struct scsi_transportid_fcp *id =
1514 		    (struct scsi_transportid_fcp *)buf;
1515 		if (port->wwpn_iid[iid].wwpn == 0)
1516 			return (0);
1517 		memset(id, 0, sizeof(*id));
1518 		id->format_protocol = SCSI_PROTO_FC;
1519 		scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->n_port_name);
1520 		return (sizeof(*id));
1521 	}
1522 	case CTL_PORT_ISCSI:
1523 	{
1524 		struct scsi_transportid_iscsi_port *id =
1525 		    (struct scsi_transportid_iscsi_port *)buf;
1526 		if (port->wwpn_iid[iid].name == NULL)
1527 			return (0);
1528 		memset(id, 0, 256);
1529 		id->format_protocol = SCSI_TRN_ISCSI_FORMAT_PORT |
1530 		    SCSI_PROTO_ISCSI;
1531 		len = strlcpy(id->iscsi_name, port->wwpn_iid[iid].name, 252) + 1;
1532 		len = roundup2(min(len, 252), 4);
1533 		scsi_ulto2b(len, id->additional_length);
1534 		return (sizeof(*id) + len);
1535 	}
1536 	case CTL_PORT_SAS:
1537 	{
1538 		struct scsi_transportid_sas *id =
1539 		    (struct scsi_transportid_sas *)buf;
1540 		if (port->wwpn_iid[iid].wwpn == 0)
1541 			return (0);
1542 		memset(id, 0, sizeof(*id));
1543 		id->format_protocol = SCSI_PROTO_SAS;
1544 		scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->sas_address);
1545 		return (sizeof(*id));
1546 	}
1547 	default:
1548 	{
1549 		struct scsi_transportid_spi *id =
1550 		    (struct scsi_transportid_spi *)buf;
1551 		memset(id, 0, sizeof(*id));
1552 		id->format_protocol = SCSI_PROTO_SPI;
1553 		scsi_ulto2b(iid, id->scsi_addr);
1554 		scsi_ulto2b(port->targ_port, id->rel_trgt_port_id);
1555 		return (sizeof(*id));
1556 	}
1557 	}
1558 }
1559 
1560 static int
1561 ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id)
1562 {
1563 	return (0);
1564 }
1565 
1566 static int
1567 ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id)
1568 {
1569 	return (0);
1570 }
1571 
1572 /*
1573  * Data movement routine for the CTL ioctl frontend port.
1574  */
1575 static int
1576 ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio)
1577 {
1578 	struct ctl_sg_entry *ext_sglist, *kern_sglist;
1579 	struct ctl_sg_entry ext_entry, kern_entry;
1580 	int ext_sglen, ext_sg_entries, kern_sg_entries;
1581 	int ext_sg_start, ext_offset;
1582 	int len_to_copy, len_copied;
1583 	int kern_watermark, ext_watermark;
1584 	int ext_sglist_malloced;
1585 	int i, j;
1586 
1587 	ext_sglist_malloced = 0;
1588 	ext_sg_start = 0;
1589 	ext_offset = 0;
1590 
1591 	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove\n"));
1592 
1593 	/*
1594 	 * If this flag is set, fake the data transfer.
1595 	 */
1596 	if (ctsio->io_hdr.flags & CTL_FLAG_NO_DATAMOVE) {
1597 		ctsio->ext_data_filled = ctsio->ext_data_len;
1598 		goto bailout;
1599 	}
1600 
1601 	/*
1602 	 * To simplify things here, if we have a single buffer, stick it in
1603 	 * a S/G entry and just make it a single entry S/G list.
1604 	 */
1605 	if (ctsio->io_hdr.flags & CTL_FLAG_EDPTR_SGLIST) {
1606 		int len_seen;
1607 
1608 		ext_sglen = ctsio->ext_sg_entries * sizeof(*ext_sglist);
1609 
1610 		ext_sglist = (struct ctl_sg_entry *)malloc(ext_sglen, M_CTL,
1611 							   M_WAITOK);
1612 		ext_sglist_malloced = 1;
1613 		if (copyin(ctsio->ext_data_ptr, ext_sglist,
1614 				   ext_sglen) != 0) {
1615 			ctl_set_internal_failure(ctsio,
1616 						 /*sks_valid*/ 0,
1617 						 /*retry_count*/ 0);
1618 			goto bailout;
1619 		}
1620 		ext_sg_entries = ctsio->ext_sg_entries;
1621 		len_seen = 0;
1622 		for (i = 0; i < ext_sg_entries; i++) {
1623 			if ((len_seen + ext_sglist[i].len) >=
1624 			     ctsio->ext_data_filled) {
1625 				ext_sg_start = i;
1626 				ext_offset = ctsio->ext_data_filled - len_seen;
1627 				break;
1628 			}
1629 			len_seen += ext_sglist[i].len;
1630 		}
1631 	} else {
1632 		ext_sglist = &ext_entry;
1633 		ext_sglist->addr = ctsio->ext_data_ptr;
1634 		ext_sglist->len = ctsio->ext_data_len;
1635 		ext_sg_entries = 1;
1636 		ext_sg_start = 0;
1637 		ext_offset = ctsio->ext_data_filled;
1638 	}
1639 
1640 	if (ctsio->kern_sg_entries > 0) {
1641 		kern_sglist = (struct ctl_sg_entry *)ctsio->kern_data_ptr;
1642 		kern_sg_entries = ctsio->kern_sg_entries;
1643 	} else {
1644 		kern_sglist = &kern_entry;
1645 		kern_sglist->addr = ctsio->kern_data_ptr;
1646 		kern_sglist->len = ctsio->kern_data_len;
1647 		kern_sg_entries = 1;
1648 	}
1649 
1650 
1651 	kern_watermark = 0;
1652 	ext_watermark = ext_offset;
1653 	len_copied = 0;
1654 	for (i = ext_sg_start, j = 0;
1655 	     i < ext_sg_entries && j < kern_sg_entries;) {
1656 		uint8_t *ext_ptr, *kern_ptr;
1657 
1658 		len_to_copy = ctl_min(ext_sglist[i].len - ext_watermark,
1659 				      kern_sglist[j].len - kern_watermark);
1660 
1661 		ext_ptr = (uint8_t *)ext_sglist[i].addr;
1662 		ext_ptr = ext_ptr + ext_watermark;
1663 		if (ctsio->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
1664 			/*
1665 			 * XXX KDM fix this!
1666 			 */
1667 			panic("need to implement bus address support");
1668 #if 0
1669 			kern_ptr = bus_to_virt(kern_sglist[j].addr);
1670 #endif
1671 		} else
1672 			kern_ptr = (uint8_t *)kern_sglist[j].addr;
1673 		kern_ptr = kern_ptr + kern_watermark;
1674 
1675 		kern_watermark += len_to_copy;
1676 		ext_watermark += len_to_copy;
1677 
1678 		if ((ctsio->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
1679 		     CTL_FLAG_DATA_IN) {
1680 			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1681 					 "bytes to user\n", len_to_copy));
1682 			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1683 					 "to %p\n", kern_ptr, ext_ptr));
1684 			if (copyout(kern_ptr, ext_ptr, len_to_copy) != 0) {
1685 				ctl_set_internal_failure(ctsio,
1686 							 /*sks_valid*/ 0,
1687 							 /*retry_count*/ 0);
1688 				goto bailout;
1689 			}
1690 		} else {
1691 			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1692 					 "bytes from user\n", len_to_copy));
1693 			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1694 					 "to %p\n", ext_ptr, kern_ptr));
1695 			if (copyin(ext_ptr, kern_ptr, len_to_copy)!= 0){
1696 				ctl_set_internal_failure(ctsio,
1697 							 /*sks_valid*/ 0,
1698 							 /*retry_count*/0);
1699 				goto bailout;
1700 			}
1701 		}
1702 
1703 		len_copied += len_to_copy;
1704 
1705 		if (ext_sglist[i].len == ext_watermark) {
1706 			i++;
1707 			ext_watermark = 0;
1708 		}
1709 
1710 		if (kern_sglist[j].len == kern_watermark) {
1711 			j++;
1712 			kern_watermark = 0;
1713 		}
1714 	}
1715 
1716 	ctsio->ext_data_filled += len_copied;
1717 
1718 	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_sg_entries: %d, "
1719 			 "kern_sg_entries: %d\n", ext_sg_entries,
1720 			 kern_sg_entries));
1721 	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_data_len = %d, "
1722 			 "kern_data_len = %d\n", ctsio->ext_data_len,
1723 			 ctsio->kern_data_len));
1724 
1725 
1726 	/* XXX KDM set residual?? */
1727 bailout:
1728 
1729 	if (ext_sglist_malloced != 0)
1730 		free(ext_sglist, M_CTL);
1731 
1732 	return (CTL_RETVAL_COMPLETE);
1733 }
1734 
1735 /*
1736  * Serialize a command that went down the "wrong" side, and so was sent to
1737  * this controller for execution.  The logic is a little different than the
1738  * standard case in ctl_scsiio_precheck().  Errors in this case need to get
1739  * sent back to the other side, but in the success case, we execute the
1740  * command on this side (XFER mode) or tell the other side to execute it
1741  * (SER_ONLY mode).
1742  */
1743 static int
1744 ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
1745 {
1746 	struct ctl_softc *ctl_softc;
1747 	union ctl_ha_msg msg_info;
1748 	struct ctl_lun *lun;
1749 	int retval = 0;
1750 	uint32_t targ_lun;
1751 
1752 	ctl_softc = control_softc;
1753 
1754 	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
1755 	lun = ctl_softc->ctl_luns[targ_lun];
1756 	if (lun==NULL)
1757 	{
1758 		/*
1759 		 * Why isn't LUN defined? The other side wouldn't
1760 		 * send a cmd if the LUN is undefined.
1761 		 */
1762 		printf("%s: Bad JUJU!, LUN is NULL!\n", __func__);
1763 
1764 		/* "Logical unit not supported" */
1765 		ctl_set_sense_data(&msg_info.scsi.sense_data,
1766 				   lun,
1767 				   /*sense_format*/SSD_TYPE_NONE,
1768 				   /*current_error*/ 1,
1769 				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1770 				   /*asc*/ 0x25,
1771 				   /*ascq*/ 0x00,
1772 				   SSD_ELEM_NONE);
1773 
1774 		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1775 		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1776 		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1777 		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1778 		msg_info.hdr.serializing_sc = NULL;
1779 		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1780 	        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1781 				sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1782 		}
1783 		return(1);
1784 
1785 	}
1786 
1787 	mtx_lock(&lun->lun_lock);
1788     	TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1789 
1790 	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
1791 		(union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
1792 		 ooa_links))) {
1793 	case CTL_ACTION_BLOCK:
1794 		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
1795 		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
1796 				  blocked_links);
1797 		break;
1798 	case CTL_ACTION_PASS:
1799 	case CTL_ACTION_SKIP:
1800 		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
1801 			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
1802 			ctl_enqueue_rtr((union ctl_io *)ctsio);
1803 		} else {
1804 
1805 			/* send msg back to other side */
1806 			msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1807 			msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
1808 			msg_info.hdr.msg_type = CTL_MSG_R2R;
1809 #if 0
1810 			printf("2. pOrig %x\n", (int)msg_info.hdr.original_sc);
1811 #endif
1812 		        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1813 			    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1814 			}
1815 		}
1816 		break;
1817 	case CTL_ACTION_OVERLAP:
1818 		/* OVERLAPPED COMMANDS ATTEMPTED */
1819 		ctl_set_sense_data(&msg_info.scsi.sense_data,
1820 				   lun,
1821 				   /*sense_format*/SSD_TYPE_NONE,
1822 				   /*current_error*/ 1,
1823 				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1824 				   /*asc*/ 0x4E,
1825 				   /*ascq*/ 0x00,
1826 				   SSD_ELEM_NONE);
1827 
1828 		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1829 		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1830 		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1831 		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1832 		msg_info.hdr.serializing_sc = NULL;
1833 		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1834 #if 0
1835 		printf("BAD JUJU:Major Bummer Overlap\n");
1836 #endif
1837 		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1838 		retval = 1;
1839 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1840 		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1841 		}
1842 		break;
1843 	case CTL_ACTION_OVERLAP_TAG:
1844 		/* TAGGED OVERLAPPED COMMANDS (NN = QUEUE TAG) */
1845 		ctl_set_sense_data(&msg_info.scsi.sense_data,
1846 				   lun,
1847 				   /*sense_format*/SSD_TYPE_NONE,
1848 				   /*current_error*/ 1,
1849 				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1850 				   /*asc*/ 0x4D,
1851 				   /*ascq*/ ctsio->tag_num & 0xff,
1852 				   SSD_ELEM_NONE);
1853 
1854 		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1855 		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1856 		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1857 		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1858 		msg_info.hdr.serializing_sc = NULL;
1859 		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1860 #if 0
1861 		printf("BAD JUJU:Major Bummer Overlap Tag\n");
1862 #endif
1863 		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1864 		retval = 1;
1865 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1866 		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1867 		}
1868 		break;
1869 	case CTL_ACTION_ERROR:
1870 	default:
1871 		/* "Internal target failure" */
1872 		ctl_set_sense_data(&msg_info.scsi.sense_data,
1873 				   lun,
1874 				   /*sense_format*/SSD_TYPE_NONE,
1875 				   /*current_error*/ 1,
1876 				   /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
1877 				   /*asc*/ 0x44,
1878 				   /*ascq*/ 0x00,
1879 				   SSD_ELEM_NONE);
1880 
1881 		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1882 		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1883 		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1884 		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1885 		msg_info.hdr.serializing_sc = NULL;
1886 		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1887 #if 0
1888 		printf("BAD JUJU:Major Bummer HW Error\n");
1889 #endif
1890 		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1891 		retval = 1;
1892 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1893 		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1894 		}
1895 		break;
1896 	}
1897 	mtx_unlock(&lun->lun_lock);
1898 	return (retval);
1899 }
1900 
1901 static int
1902 ctl_ioctl_submit_wait(union ctl_io *io)
1903 {
1904 	struct ctl_fe_ioctl_params params;
1905 	ctl_fe_ioctl_state last_state;
1906 	int done, retval;
1907 
1908 	retval = 0;
1909 
1910 	bzero(&params, sizeof(params));
1911 
1912 	mtx_init(&params.ioctl_mtx, "ctliocmtx", NULL, MTX_DEF);
1913 	cv_init(&params.sem, "ctlioccv");
1914 	params.state = CTL_IOCTL_INPROG;
1915 	last_state = params.state;
1916 
1917 	io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = &params;
1918 
1919 	CTL_DEBUG_PRINT(("ctl_ioctl_submit_wait\n"));
1920 
1921 	/* This shouldn't happen */
1922 	if ((retval = ctl_queue(io)) != CTL_RETVAL_COMPLETE)
1923 		return (retval);
1924 
1925 	done = 0;
1926 
1927 	do {
1928 		mtx_lock(&params.ioctl_mtx);
1929 		/*
1930 		 * Check the state here, and don't sleep if the state has
1931 		 * already changed (i.e. wakeup has already occured, but we
1932 		 * weren't waiting yet).
1933 		 */
1934 		if (params.state == last_state) {
1935 			/* XXX KDM cv_wait_sig instead? */
1936 			cv_wait(&params.sem, &params.ioctl_mtx);
1937 		}
1938 		last_state = params.state;
1939 
1940 		switch (params.state) {
1941 		case CTL_IOCTL_INPROG:
1942 			/* Why did we wake up? */
1943 			/* XXX KDM error here? */
1944 			mtx_unlock(&params.ioctl_mtx);
1945 			break;
1946 		case CTL_IOCTL_DATAMOVE:
1947 			CTL_DEBUG_PRINT(("got CTL_IOCTL_DATAMOVE\n"));
1948 
1949 			/*
1950 			 * change last_state back to INPROG to avoid
1951 			 * deadlock on subsequent data moves.
1952 			 */
1953 			params.state = last_state = CTL_IOCTL_INPROG;
1954 
1955 			mtx_unlock(&params.ioctl_mtx);
1956 			ctl_ioctl_do_datamove(&io->scsiio);
1957 			/*
1958 			 * Note that in some cases, most notably writes,
1959 			 * this will queue the I/O and call us back later.
1960 			 * In other cases, generally reads, this routine
1961 			 * will immediately call back and wake us up,
1962 			 * probably using our own context.
1963 			 */
1964 			io->scsiio.be_move_done(io);
1965 			break;
1966 		case CTL_IOCTL_DONE:
1967 			mtx_unlock(&params.ioctl_mtx);
1968 			CTL_DEBUG_PRINT(("got CTL_IOCTL_DONE\n"));
1969 			done = 1;
1970 			break;
1971 		default:
1972 			mtx_unlock(&params.ioctl_mtx);
1973 			/* XXX KDM error here? */
1974 			break;
1975 		}
1976 	} while (done == 0);
1977 
1978 	mtx_destroy(&params.ioctl_mtx);
1979 	cv_destroy(&params.sem);
1980 
1981 	return (CTL_RETVAL_COMPLETE);
1982 }
1983 
1984 static void
1985 ctl_ioctl_datamove(union ctl_io *io)
1986 {
1987 	struct ctl_fe_ioctl_params *params;
1988 
1989 	params = (struct ctl_fe_ioctl_params *)
1990 		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
1991 
1992 	mtx_lock(&params->ioctl_mtx);
1993 	params->state = CTL_IOCTL_DATAMOVE;
1994 	cv_broadcast(&params->sem);
1995 	mtx_unlock(&params->ioctl_mtx);
1996 }
1997 
1998 static void
1999 ctl_ioctl_done(union ctl_io *io)
2000 {
2001 	struct ctl_fe_ioctl_params *params;
2002 
2003 	params = (struct ctl_fe_ioctl_params *)
2004 		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
2005 
2006 	mtx_lock(&params->ioctl_mtx);
2007 	params->state = CTL_IOCTL_DONE;
2008 	cv_broadcast(&params->sem);
2009 	mtx_unlock(&params->ioctl_mtx);
2010 }
2011 
2012 static void
2013 ctl_ioctl_hard_startstop_callback(void *arg, struct cfi_metatask *metatask)
2014 {
2015 	struct ctl_fe_ioctl_startstop_info *sd_info;
2016 
2017 	sd_info = (struct ctl_fe_ioctl_startstop_info *)arg;
2018 
2019 	sd_info->hs_info.status = metatask->status;
2020 	sd_info->hs_info.total_luns = metatask->taskinfo.startstop.total_luns;
2021 	sd_info->hs_info.luns_complete =
2022 		metatask->taskinfo.startstop.luns_complete;
2023 	sd_info->hs_info.luns_failed = metatask->taskinfo.startstop.luns_failed;
2024 
2025 	cv_broadcast(&sd_info->sem);
2026 }
2027 
2028 static void
2029 ctl_ioctl_bbrread_callback(void *arg, struct cfi_metatask *metatask)
2030 {
2031 	struct ctl_fe_ioctl_bbrread_info *fe_bbr_info;
2032 
2033 	fe_bbr_info = (struct ctl_fe_ioctl_bbrread_info *)arg;
2034 
2035 	mtx_lock(fe_bbr_info->lock);
2036 	fe_bbr_info->bbr_info->status = metatask->status;
2037 	fe_bbr_info->bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2038 	fe_bbr_info->wakeup_done = 1;
2039 	mtx_unlock(fe_bbr_info->lock);
2040 
2041 	cv_broadcast(&fe_bbr_info->sem);
2042 }
2043 
2044 /*
2045  * Returns 0 for success, errno for failure.
2046  */
2047 static int
2048 ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
2049 		   struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
2050 {
2051 	union ctl_io *io;
2052 	int retval;
2053 
2054 	retval = 0;
2055 
2056 	mtx_lock(&lun->lun_lock);
2057 	for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
2058 	     (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2059 	     ooa_links)) {
2060 		struct ctl_ooa_entry *entry;
2061 
2062 		/*
2063 		 * If we've got more than we can fit, just count the
2064 		 * remaining entries.
2065 		 */
2066 		if (*cur_fill_num >= ooa_hdr->alloc_num)
2067 			continue;
2068 
2069 		entry = &kern_entries[*cur_fill_num];
2070 
2071 		entry->tag_num = io->scsiio.tag_num;
2072 		entry->lun_num = lun->lun;
2073 #ifdef CTL_TIME_IO
2074 		entry->start_bt = io->io_hdr.start_bt;
2075 #endif
2076 		bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2077 		entry->cdb_len = io->scsiio.cdb_len;
2078 		if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
2079 			entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2080 
2081 		if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2082 			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2083 
2084 		if (io->io_hdr.flags & CTL_FLAG_ABORT)
2085 			entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2086 
2087 		if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2088 			entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2089 
2090 		if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2091 			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2092 	}
2093 	mtx_unlock(&lun->lun_lock);
2094 
2095 	return (retval);
2096 }
2097 
2098 static void *
2099 ctl_copyin_alloc(void *user_addr, int len, char *error_str,
2100 		 size_t error_str_len)
2101 {
2102 	void *kptr;
2103 
2104 	kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2105 
2106 	if (copyin(user_addr, kptr, len) != 0) {
2107 		snprintf(error_str, error_str_len, "Error copying %d bytes "
2108 			 "from user address %p to kernel address %p", len,
2109 			 user_addr, kptr);
2110 		free(kptr, M_CTL);
2111 		return (NULL);
2112 	}
2113 
2114 	return (kptr);
2115 }
2116 
2117 static void
2118 ctl_free_args(int num_args, struct ctl_be_arg *args)
2119 {
2120 	int i;
2121 
2122 	if (args == NULL)
2123 		return;
2124 
2125 	for (i = 0; i < num_args; i++) {
2126 		free(args[i].kname, M_CTL);
2127 		free(args[i].kvalue, M_CTL);
2128 	}
2129 
2130 	free(args, M_CTL);
2131 }
2132 
2133 static struct ctl_be_arg *
2134 ctl_copyin_args(int num_args, struct ctl_be_arg *uargs,
2135 		char *error_str, size_t error_str_len)
2136 {
2137 	struct ctl_be_arg *args;
2138 	int i;
2139 
2140 	args = ctl_copyin_alloc(uargs, num_args * sizeof(*args),
2141 				error_str, error_str_len);
2142 
2143 	if (args == NULL)
2144 		goto bailout;
2145 
2146 	for (i = 0; i < num_args; i++) {
2147 		args[i].kname = NULL;
2148 		args[i].kvalue = NULL;
2149 	}
2150 
2151 	for (i = 0; i < num_args; i++) {
2152 		uint8_t *tmpptr;
2153 
2154 		args[i].kname = ctl_copyin_alloc(args[i].name,
2155 			args[i].namelen, error_str, error_str_len);
2156 		if (args[i].kname == NULL)
2157 			goto bailout;
2158 
2159 		if (args[i].kname[args[i].namelen - 1] != '\0') {
2160 			snprintf(error_str, error_str_len, "Argument %d "
2161 				 "name is not NUL-terminated", i);
2162 			goto bailout;
2163 		}
2164 
2165 		if (args[i].flags & CTL_BEARG_RD) {
2166 			tmpptr = ctl_copyin_alloc(args[i].value,
2167 				args[i].vallen, error_str, error_str_len);
2168 			if (tmpptr == NULL)
2169 				goto bailout;
2170 			if ((args[i].flags & CTL_BEARG_ASCII)
2171 			 && (tmpptr[args[i].vallen - 1] != '\0')) {
2172 				snprintf(error_str, error_str_len, "Argument "
2173 				    "%d value is not NUL-terminated", i);
2174 				goto bailout;
2175 			}
2176 			args[i].kvalue = tmpptr;
2177 		} else {
2178 			args[i].kvalue = malloc(args[i].vallen,
2179 			    M_CTL, M_WAITOK | M_ZERO);
2180 		}
2181 	}
2182 
2183 	return (args);
2184 bailout:
2185 
2186 	ctl_free_args(num_args, args);
2187 
2188 	return (NULL);
2189 }
2190 
2191 static void
2192 ctl_copyout_args(int num_args, struct ctl_be_arg *args)
2193 {
2194 	int i;
2195 
2196 	for (i = 0; i < num_args; i++) {
2197 		if (args[i].flags & CTL_BEARG_WR)
2198 			copyout(args[i].kvalue, args[i].value, args[i].vallen);
2199 	}
2200 }
2201 
2202 /*
2203  * Escape characters that are illegal or not recommended in XML.
2204  */
2205 int
2206 ctl_sbuf_printf_esc(struct sbuf *sb, char *str)
2207 {
2208 	int retval;
2209 
2210 	retval = 0;
2211 
2212 	for (; *str; str++) {
2213 		switch (*str) {
2214 		case '&':
2215 			retval = sbuf_printf(sb, "&amp;");
2216 			break;
2217 		case '>':
2218 			retval = sbuf_printf(sb, "&gt;");
2219 			break;
2220 		case '<':
2221 			retval = sbuf_printf(sb, "&lt;");
2222 			break;
2223 		default:
2224 			retval = sbuf_putc(sb, *str);
2225 			break;
2226 		}
2227 
2228 		if (retval != 0)
2229 			break;
2230 
2231 	}
2232 
2233 	return (retval);
2234 }
2235 
2236 static int
2237 ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2238 	  struct thread *td)
2239 {
2240 	struct ctl_softc *softc;
2241 	int retval;
2242 
2243 	softc = control_softc;
2244 
2245 	retval = 0;
2246 
2247 	switch (cmd) {
2248 	case CTL_IO: {
2249 		union ctl_io *io;
2250 		void *pool_tmp;
2251 
2252 		/*
2253 		 * If we haven't been "enabled", don't allow any SCSI I/O
2254 		 * to this FETD.
2255 		 */
2256 		if ((softc->ioctl_info.flags & CTL_IOCTL_FLAG_ENABLED) == 0) {
2257 			retval = EPERM;
2258 			break;
2259 		}
2260 
2261 		io = ctl_alloc_io(softc->ioctl_info.port.ctl_pool_ref);
2262 		if (io == NULL) {
2263 			printf("ctl_ioctl: can't allocate ctl_io!\n");
2264 			retval = ENOSPC;
2265 			break;
2266 		}
2267 
2268 		/*
2269 		 * Need to save the pool reference so it doesn't get
2270 		 * spammed by the user's ctl_io.
2271 		 */
2272 		pool_tmp = io->io_hdr.pool;
2273 
2274 		memcpy(io, (void *)addr, sizeof(*io));
2275 
2276 		io->io_hdr.pool = pool_tmp;
2277 		/*
2278 		 * No status yet, so make sure the status is set properly.
2279 		 */
2280 		io->io_hdr.status = CTL_STATUS_NONE;
2281 
2282 		/*
2283 		 * The user sets the initiator ID, target and LUN IDs.
2284 		 */
2285 		io->io_hdr.nexus.targ_port = softc->ioctl_info.port.targ_port;
2286 		io->io_hdr.flags |= CTL_FLAG_USER_REQ;
2287 		if ((io->io_hdr.io_type == CTL_IO_SCSI)
2288 		 && (io->scsiio.tag_type != CTL_TAG_UNTAGGED))
2289 			io->scsiio.tag_num = softc->ioctl_info.cur_tag_num++;
2290 
2291 		retval = ctl_ioctl_submit_wait(io);
2292 
2293 		if (retval != 0) {
2294 			ctl_free_io(io);
2295 			break;
2296 		}
2297 
2298 		memcpy((void *)addr, io, sizeof(*io));
2299 
2300 		/* return this to our pool */
2301 		ctl_free_io(io);
2302 
2303 		break;
2304 	}
2305 	case CTL_ENABLE_PORT:
2306 	case CTL_DISABLE_PORT:
2307 	case CTL_SET_PORT_WWNS: {
2308 		struct ctl_port *port;
2309 		struct ctl_port_entry *entry;
2310 
2311 		entry = (struct ctl_port_entry *)addr;
2312 
2313 		mtx_lock(&softc->ctl_lock);
2314 		STAILQ_FOREACH(port, &softc->port_list, links) {
2315 			int action, done;
2316 
2317 			action = 0;
2318 			done = 0;
2319 
2320 			if ((entry->port_type == CTL_PORT_NONE)
2321 			 && (entry->targ_port == port->targ_port)) {
2322 				/*
2323 				 * If the user only wants to enable or
2324 				 * disable or set WWNs on a specific port,
2325 				 * do the operation and we're done.
2326 				 */
2327 				action = 1;
2328 				done = 1;
2329 			} else if (entry->port_type & port->port_type) {
2330 				/*
2331 				 * Compare the user's type mask with the
2332 				 * particular frontend type to see if we
2333 				 * have a match.
2334 				 */
2335 				action = 1;
2336 				done = 0;
2337 
2338 				/*
2339 				 * Make sure the user isn't trying to set
2340 				 * WWNs on multiple ports at the same time.
2341 				 */
2342 				if (cmd == CTL_SET_PORT_WWNS) {
2343 					printf("%s: Can't set WWNs on "
2344 					       "multiple ports\n", __func__);
2345 					retval = EINVAL;
2346 					break;
2347 				}
2348 			}
2349 			if (action != 0) {
2350 				/*
2351 				 * XXX KDM we have to drop the lock here,
2352 				 * because the online/offline operations
2353 				 * can potentially block.  We need to
2354 				 * reference count the frontends so they
2355 				 * can't go away,
2356 				 */
2357 				mtx_unlock(&softc->ctl_lock);
2358 
2359 				if (cmd == CTL_ENABLE_PORT) {
2360 					struct ctl_lun *lun;
2361 
2362 					STAILQ_FOREACH(lun, &softc->lun_list,
2363 						       links) {
2364 						port->lun_enable(port->targ_lun_arg,
2365 						    lun->target,
2366 						    lun->lun);
2367 					}
2368 
2369 					ctl_port_online(port);
2370 				} else if (cmd == CTL_DISABLE_PORT) {
2371 					struct ctl_lun *lun;
2372 
2373 					ctl_port_offline(port);
2374 
2375 					STAILQ_FOREACH(lun, &softc->lun_list,
2376 						       links) {
2377 						port->lun_disable(
2378 						    port->targ_lun_arg,
2379 						    lun->target,
2380 						    lun->lun);
2381 					}
2382 				}
2383 
2384 				mtx_lock(&softc->ctl_lock);
2385 
2386 				if (cmd == CTL_SET_PORT_WWNS)
2387 					ctl_port_set_wwns(port,
2388 					    (entry->flags & CTL_PORT_WWNN_VALID) ?
2389 					    1 : 0, entry->wwnn,
2390 					    (entry->flags & CTL_PORT_WWPN_VALID) ?
2391 					    1 : 0, entry->wwpn);
2392 			}
2393 			if (done != 0)
2394 				break;
2395 		}
2396 		mtx_unlock(&softc->ctl_lock);
2397 		break;
2398 	}
2399 	case CTL_GET_PORT_LIST: {
2400 		struct ctl_port *port;
2401 		struct ctl_port_list *list;
2402 		int i;
2403 
2404 		list = (struct ctl_port_list *)addr;
2405 
2406 		if (list->alloc_len != (list->alloc_num *
2407 		    sizeof(struct ctl_port_entry))) {
2408 			printf("%s: CTL_GET_PORT_LIST: alloc_len %u != "
2409 			       "alloc_num %u * sizeof(struct ctl_port_entry) "
2410 			       "%zu\n", __func__, list->alloc_len,
2411 			       list->alloc_num, sizeof(struct ctl_port_entry));
2412 			retval = EINVAL;
2413 			break;
2414 		}
2415 		list->fill_len = 0;
2416 		list->fill_num = 0;
2417 		list->dropped_num = 0;
2418 		i = 0;
2419 		mtx_lock(&softc->ctl_lock);
2420 		STAILQ_FOREACH(port, &softc->port_list, links) {
2421 			struct ctl_port_entry entry, *list_entry;
2422 
2423 			if (list->fill_num >= list->alloc_num) {
2424 				list->dropped_num++;
2425 				continue;
2426 			}
2427 
2428 			entry.port_type = port->port_type;
2429 			strlcpy(entry.port_name, port->port_name,
2430 				sizeof(entry.port_name));
2431 			entry.targ_port = port->targ_port;
2432 			entry.physical_port = port->physical_port;
2433 			entry.virtual_port = port->virtual_port;
2434 			entry.wwnn = port->wwnn;
2435 			entry.wwpn = port->wwpn;
2436 			if (port->status & CTL_PORT_STATUS_ONLINE)
2437 				entry.online = 1;
2438 			else
2439 				entry.online = 0;
2440 
2441 			list_entry = &list->entries[i];
2442 
2443 			retval = copyout(&entry, list_entry, sizeof(entry));
2444 			if (retval != 0) {
2445 				printf("%s: CTL_GET_PORT_LIST: copyout "
2446 				       "returned %d\n", __func__, retval);
2447 				break;
2448 			}
2449 			i++;
2450 			list->fill_num++;
2451 			list->fill_len += sizeof(entry);
2452 		}
2453 		mtx_unlock(&softc->ctl_lock);
2454 
2455 		/*
2456 		 * If this is non-zero, we had a copyout fault, so there's
2457 		 * probably no point in attempting to set the status inside
2458 		 * the structure.
2459 		 */
2460 		if (retval != 0)
2461 			break;
2462 
2463 		if (list->dropped_num > 0)
2464 			list->status = CTL_PORT_LIST_NEED_MORE_SPACE;
2465 		else
2466 			list->status = CTL_PORT_LIST_OK;
2467 		break;
2468 	}
2469 	case CTL_DUMP_OOA: {
2470 		struct ctl_lun *lun;
2471 		union ctl_io *io;
2472 		char printbuf[128];
2473 		struct sbuf sb;
2474 
2475 		mtx_lock(&softc->ctl_lock);
2476 		printf("Dumping OOA queues:\n");
2477 		STAILQ_FOREACH(lun, &softc->lun_list, links) {
2478 			mtx_lock(&lun->lun_lock);
2479 			for (io = (union ctl_io *)TAILQ_FIRST(
2480 			     &lun->ooa_queue); io != NULL;
2481 			     io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2482 			     ooa_links)) {
2483 				sbuf_new(&sb, printbuf, sizeof(printbuf),
2484 					 SBUF_FIXEDLEN);
2485 				sbuf_printf(&sb, "LUN %jd tag 0x%04x%s%s%s%s: ",
2486 					    (intmax_t)lun->lun,
2487 					    io->scsiio.tag_num,
2488 					    (io->io_hdr.flags &
2489 					    CTL_FLAG_BLOCKED) ? "" : " BLOCKED",
2490 					    (io->io_hdr.flags &
2491 					    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
2492 					    (io->io_hdr.flags &
2493 					    CTL_FLAG_ABORT) ? " ABORT" : "",
2494 			                    (io->io_hdr.flags &
2495 		                        CTL_FLAG_IS_WAS_ON_RTR) ? " RTR" : "");
2496 				ctl_scsi_command_string(&io->scsiio, NULL, &sb);
2497 				sbuf_finish(&sb);
2498 				printf("%s\n", sbuf_data(&sb));
2499 			}
2500 			mtx_unlock(&lun->lun_lock);
2501 		}
2502 		printf("OOA queues dump done\n");
2503 		mtx_unlock(&softc->ctl_lock);
2504 		break;
2505 	}
2506 	case CTL_GET_OOA: {
2507 		struct ctl_lun *lun;
2508 		struct ctl_ooa *ooa_hdr;
2509 		struct ctl_ooa_entry *entries;
2510 		uint32_t cur_fill_num;
2511 
2512 		ooa_hdr = (struct ctl_ooa *)addr;
2513 
2514 		if ((ooa_hdr->alloc_len == 0)
2515 		 || (ooa_hdr->alloc_num == 0)) {
2516 			printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2517 			       "must be non-zero\n", __func__,
2518 			       ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2519 			retval = EINVAL;
2520 			break;
2521 		}
2522 
2523 		if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2524 		    sizeof(struct ctl_ooa_entry))) {
2525 			printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2526 			       "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2527 			       __func__, ooa_hdr->alloc_len,
2528 			       ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2529 			retval = EINVAL;
2530 			break;
2531 		}
2532 
2533 		entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2534 		if (entries == NULL) {
2535 			printf("%s: could not allocate %d bytes for OOA "
2536 			       "dump\n", __func__, ooa_hdr->alloc_len);
2537 			retval = ENOMEM;
2538 			break;
2539 		}
2540 
2541 		mtx_lock(&softc->ctl_lock);
2542 		if (((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0)
2543 		 && ((ooa_hdr->lun_num > CTL_MAX_LUNS)
2544 		  || (softc->ctl_luns[ooa_hdr->lun_num] == NULL))) {
2545 			mtx_unlock(&softc->ctl_lock);
2546 			free(entries, M_CTL);
2547 			printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2548 			       __func__, (uintmax_t)ooa_hdr->lun_num);
2549 			retval = EINVAL;
2550 			break;
2551 		}
2552 
2553 		cur_fill_num = 0;
2554 
2555 		if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2556 			STAILQ_FOREACH(lun, &softc->lun_list, links) {
2557 				retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2558 					ooa_hdr, entries);
2559 				if (retval != 0)
2560 					break;
2561 			}
2562 			if (retval != 0) {
2563 				mtx_unlock(&softc->ctl_lock);
2564 				free(entries, M_CTL);
2565 				break;
2566 			}
2567 		} else {
2568 			lun = softc->ctl_luns[ooa_hdr->lun_num];
2569 
2570 			retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,ooa_hdr,
2571 						    entries);
2572 		}
2573 		mtx_unlock(&softc->ctl_lock);
2574 
2575 		ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2576 		ooa_hdr->fill_len = ooa_hdr->fill_num *
2577 			sizeof(struct ctl_ooa_entry);
2578 		retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2579 		if (retval != 0) {
2580 			printf("%s: error copying out %d bytes for OOA dump\n",
2581 			       __func__, ooa_hdr->fill_len);
2582 		}
2583 
2584 		getbintime(&ooa_hdr->cur_bt);
2585 
2586 		if (cur_fill_num > ooa_hdr->alloc_num) {
2587 			ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2588 			ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2589 		} else {
2590 			ooa_hdr->dropped_num = 0;
2591 			ooa_hdr->status = CTL_OOA_OK;
2592 		}
2593 
2594 		free(entries, M_CTL);
2595 		break;
2596 	}
2597 	case CTL_CHECK_OOA: {
2598 		union ctl_io *io;
2599 		struct ctl_lun *lun;
2600 		struct ctl_ooa_info *ooa_info;
2601 
2602 
2603 		ooa_info = (struct ctl_ooa_info *)addr;
2604 
2605 		if (ooa_info->lun_id >= CTL_MAX_LUNS) {
2606 			ooa_info->status = CTL_OOA_INVALID_LUN;
2607 			break;
2608 		}
2609 		mtx_lock(&softc->ctl_lock);
2610 		lun = softc->ctl_luns[ooa_info->lun_id];
2611 		if (lun == NULL) {
2612 			mtx_unlock(&softc->ctl_lock);
2613 			ooa_info->status = CTL_OOA_INVALID_LUN;
2614 			break;
2615 		}
2616 		mtx_lock(&lun->lun_lock);
2617 		mtx_unlock(&softc->ctl_lock);
2618 		ooa_info->num_entries = 0;
2619 		for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
2620 		     io != NULL; io = (union ctl_io *)TAILQ_NEXT(
2621 		     &io->io_hdr, ooa_links)) {
2622 			ooa_info->num_entries++;
2623 		}
2624 		mtx_unlock(&lun->lun_lock);
2625 
2626 		ooa_info->status = CTL_OOA_SUCCESS;
2627 
2628 		break;
2629 	}
2630 	case CTL_HARD_START:
2631 	case CTL_HARD_STOP: {
2632 		struct ctl_fe_ioctl_startstop_info ss_info;
2633 		struct cfi_metatask *metatask;
2634 		struct mtx hs_mtx;
2635 
2636 		mtx_init(&hs_mtx, "HS Mutex", NULL, MTX_DEF);
2637 
2638 		cv_init(&ss_info.sem, "hard start/stop cv" );
2639 
2640 		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2641 		if (metatask == NULL) {
2642 			retval = ENOMEM;
2643 			mtx_destroy(&hs_mtx);
2644 			break;
2645 		}
2646 
2647 		if (cmd == CTL_HARD_START)
2648 			metatask->tasktype = CFI_TASK_STARTUP;
2649 		else
2650 			metatask->tasktype = CFI_TASK_SHUTDOWN;
2651 
2652 		metatask->callback = ctl_ioctl_hard_startstop_callback;
2653 		metatask->callback_arg = &ss_info;
2654 
2655 		cfi_action(metatask);
2656 
2657 		/* Wait for the callback */
2658 		mtx_lock(&hs_mtx);
2659 		cv_wait_sig(&ss_info.sem, &hs_mtx);
2660 		mtx_unlock(&hs_mtx);
2661 
2662 		/*
2663 		 * All information has been copied from the metatask by the
2664 		 * time cv_broadcast() is called, so we free the metatask here.
2665 		 */
2666 		cfi_free_metatask(metatask);
2667 
2668 		memcpy((void *)addr, &ss_info.hs_info, sizeof(ss_info.hs_info));
2669 
2670 		mtx_destroy(&hs_mtx);
2671 		break;
2672 	}
2673 	case CTL_BBRREAD: {
2674 		struct ctl_bbrread_info *bbr_info;
2675 		struct ctl_fe_ioctl_bbrread_info fe_bbr_info;
2676 		struct mtx bbr_mtx;
2677 		struct cfi_metatask *metatask;
2678 
2679 		bbr_info = (struct ctl_bbrread_info *)addr;
2680 
2681 		bzero(&fe_bbr_info, sizeof(fe_bbr_info));
2682 
2683 		bzero(&bbr_mtx, sizeof(bbr_mtx));
2684 		mtx_init(&bbr_mtx, "BBR Mutex", NULL, MTX_DEF);
2685 
2686 		fe_bbr_info.bbr_info = bbr_info;
2687 		fe_bbr_info.lock = &bbr_mtx;
2688 
2689 		cv_init(&fe_bbr_info.sem, "BBR read cv");
2690 		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2691 
2692 		if (metatask == NULL) {
2693 			mtx_destroy(&bbr_mtx);
2694 			cv_destroy(&fe_bbr_info.sem);
2695 			retval = ENOMEM;
2696 			break;
2697 		}
2698 		metatask->tasktype = CFI_TASK_BBRREAD;
2699 		metatask->callback = ctl_ioctl_bbrread_callback;
2700 		metatask->callback_arg = &fe_bbr_info;
2701 		metatask->taskinfo.bbrread.lun_num = bbr_info->lun_num;
2702 		metatask->taskinfo.bbrread.lba = bbr_info->lba;
2703 		metatask->taskinfo.bbrread.len = bbr_info->len;
2704 
2705 		cfi_action(metatask);
2706 
2707 		mtx_lock(&bbr_mtx);
2708 		while (fe_bbr_info.wakeup_done == 0)
2709 			cv_wait_sig(&fe_bbr_info.sem, &bbr_mtx);
2710 		mtx_unlock(&bbr_mtx);
2711 
2712 		bbr_info->status = metatask->status;
2713 		bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2714 		bbr_info->scsi_status = metatask->taskinfo.bbrread.scsi_status;
2715 		memcpy(&bbr_info->sense_data,
2716 		       &metatask->taskinfo.bbrread.sense_data,
2717 		       ctl_min(sizeof(bbr_info->sense_data),
2718 			       sizeof(metatask->taskinfo.bbrread.sense_data)));
2719 
2720 		cfi_free_metatask(metatask);
2721 
2722 		mtx_destroy(&bbr_mtx);
2723 		cv_destroy(&fe_bbr_info.sem);
2724 
2725 		break;
2726 	}
2727 	case CTL_DELAY_IO: {
2728 		struct ctl_io_delay_info *delay_info;
2729 #ifdef CTL_IO_DELAY
2730 		struct ctl_lun *lun;
2731 #endif /* CTL_IO_DELAY */
2732 
2733 		delay_info = (struct ctl_io_delay_info *)addr;
2734 
2735 #ifdef CTL_IO_DELAY
2736 		mtx_lock(&softc->ctl_lock);
2737 
2738 		if ((delay_info->lun_id > CTL_MAX_LUNS)
2739 		 || (softc->ctl_luns[delay_info->lun_id] == NULL)) {
2740 			delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2741 		} else {
2742 			lun = softc->ctl_luns[delay_info->lun_id];
2743 			mtx_lock(&lun->lun_lock);
2744 
2745 			delay_info->status = CTL_DELAY_STATUS_OK;
2746 
2747 			switch (delay_info->delay_type) {
2748 			case CTL_DELAY_TYPE_CONT:
2749 				break;
2750 			case CTL_DELAY_TYPE_ONESHOT:
2751 				break;
2752 			default:
2753 				delay_info->status =
2754 					CTL_DELAY_STATUS_INVALID_TYPE;
2755 				break;
2756 			}
2757 
2758 			switch (delay_info->delay_loc) {
2759 			case CTL_DELAY_LOC_DATAMOVE:
2760 				lun->delay_info.datamove_type =
2761 					delay_info->delay_type;
2762 				lun->delay_info.datamove_delay =
2763 					delay_info->delay_secs;
2764 				break;
2765 			case CTL_DELAY_LOC_DONE:
2766 				lun->delay_info.done_type =
2767 					delay_info->delay_type;
2768 				lun->delay_info.done_delay =
2769 					delay_info->delay_secs;
2770 				break;
2771 			default:
2772 				delay_info->status =
2773 					CTL_DELAY_STATUS_INVALID_LOC;
2774 				break;
2775 			}
2776 			mtx_unlock(&lun->lun_lock);
2777 		}
2778 
2779 		mtx_unlock(&softc->ctl_lock);
2780 #else
2781 		delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2782 #endif /* CTL_IO_DELAY */
2783 		break;
2784 	}
2785 	case CTL_REALSYNC_SET: {
2786 		int *syncstate;
2787 
2788 		syncstate = (int *)addr;
2789 
2790 		mtx_lock(&softc->ctl_lock);
2791 		switch (*syncstate) {
2792 		case 0:
2793 			softc->flags &= ~CTL_FLAG_REAL_SYNC;
2794 			break;
2795 		case 1:
2796 			softc->flags |= CTL_FLAG_REAL_SYNC;
2797 			break;
2798 		default:
2799 			retval = EINVAL;
2800 			break;
2801 		}
2802 		mtx_unlock(&softc->ctl_lock);
2803 		break;
2804 	}
2805 	case CTL_REALSYNC_GET: {
2806 		int *syncstate;
2807 
2808 		syncstate = (int*)addr;
2809 
2810 		mtx_lock(&softc->ctl_lock);
2811 		if (softc->flags & CTL_FLAG_REAL_SYNC)
2812 			*syncstate = 1;
2813 		else
2814 			*syncstate = 0;
2815 		mtx_unlock(&softc->ctl_lock);
2816 
2817 		break;
2818 	}
2819 	case CTL_SETSYNC:
2820 	case CTL_GETSYNC: {
2821 		struct ctl_sync_info *sync_info;
2822 		struct ctl_lun *lun;
2823 
2824 		sync_info = (struct ctl_sync_info *)addr;
2825 
2826 		mtx_lock(&softc->ctl_lock);
2827 		lun = softc->ctl_luns[sync_info->lun_id];
2828 		if (lun == NULL) {
2829 			mtx_unlock(&softc->ctl_lock);
2830 			sync_info->status = CTL_GS_SYNC_NO_LUN;
2831 		}
2832 		/*
2833 		 * Get or set the sync interval.  We're not bounds checking
2834 		 * in the set case, hopefully the user won't do something
2835 		 * silly.
2836 		 */
2837 		mtx_lock(&lun->lun_lock);
2838 		mtx_unlock(&softc->ctl_lock);
2839 		if (cmd == CTL_GETSYNC)
2840 			sync_info->sync_interval = lun->sync_interval;
2841 		else
2842 			lun->sync_interval = sync_info->sync_interval;
2843 		mtx_unlock(&lun->lun_lock);
2844 
2845 		sync_info->status = CTL_GS_SYNC_OK;
2846 
2847 		break;
2848 	}
2849 	case CTL_GETSTATS: {
2850 		struct ctl_stats *stats;
2851 		struct ctl_lun *lun;
2852 		int i;
2853 
2854 		stats = (struct ctl_stats *)addr;
2855 
2856 		if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) >
2857 		     stats->alloc_len) {
2858 			stats->status = CTL_SS_NEED_MORE_SPACE;
2859 			stats->num_luns = softc->num_luns;
2860 			break;
2861 		}
2862 		/*
2863 		 * XXX KDM no locking here.  If the LUN list changes,
2864 		 * things can blow up.
2865 		 */
2866 		for (i = 0, lun = STAILQ_FIRST(&softc->lun_list); lun != NULL;
2867 		     i++, lun = STAILQ_NEXT(lun, links)) {
2868 			retval = copyout(&lun->stats, &stats->lun_stats[i],
2869 					 sizeof(lun->stats));
2870 			if (retval != 0)
2871 				break;
2872 		}
2873 		stats->num_luns = softc->num_luns;
2874 		stats->fill_len = sizeof(struct ctl_lun_io_stats) *
2875 				 softc->num_luns;
2876 		stats->status = CTL_SS_OK;
2877 #ifdef CTL_TIME_IO
2878 		stats->flags = CTL_STATS_FLAG_TIME_VALID;
2879 #else
2880 		stats->flags = CTL_STATS_FLAG_NONE;
2881 #endif
2882 		getnanouptime(&stats->timestamp);
2883 		break;
2884 	}
2885 	case CTL_ERROR_INJECT: {
2886 		struct ctl_error_desc *err_desc, *new_err_desc;
2887 		struct ctl_lun *lun;
2888 
2889 		err_desc = (struct ctl_error_desc *)addr;
2890 
2891 		new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2892 				      M_WAITOK | M_ZERO);
2893 		bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2894 
2895 		mtx_lock(&softc->ctl_lock);
2896 		lun = softc->ctl_luns[err_desc->lun_id];
2897 		if (lun == NULL) {
2898 			mtx_unlock(&softc->ctl_lock);
2899 			printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2900 			       __func__, (uintmax_t)err_desc->lun_id);
2901 			retval = EINVAL;
2902 			break;
2903 		}
2904 		mtx_lock(&lun->lun_lock);
2905 		mtx_unlock(&softc->ctl_lock);
2906 
2907 		/*
2908 		 * We could do some checking here to verify the validity
2909 		 * of the request, but given the complexity of error
2910 		 * injection requests, the checking logic would be fairly
2911 		 * complex.
2912 		 *
2913 		 * For now, if the request is invalid, it just won't get
2914 		 * executed and might get deleted.
2915 		 */
2916 		STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2917 
2918 		/*
2919 		 * XXX KDM check to make sure the serial number is unique,
2920 		 * in case we somehow manage to wrap.  That shouldn't
2921 		 * happen for a very long time, but it's the right thing to
2922 		 * do.
2923 		 */
2924 		new_err_desc->serial = lun->error_serial;
2925 		err_desc->serial = lun->error_serial;
2926 		lun->error_serial++;
2927 
2928 		mtx_unlock(&lun->lun_lock);
2929 		break;
2930 	}
2931 	case CTL_ERROR_INJECT_DELETE: {
2932 		struct ctl_error_desc *delete_desc, *desc, *desc2;
2933 		struct ctl_lun *lun;
2934 		int delete_done;
2935 
2936 		delete_desc = (struct ctl_error_desc *)addr;
2937 		delete_done = 0;
2938 
2939 		mtx_lock(&softc->ctl_lock);
2940 		lun = softc->ctl_luns[delete_desc->lun_id];
2941 		if (lun == NULL) {
2942 			mtx_unlock(&softc->ctl_lock);
2943 			printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2944 			       __func__, (uintmax_t)delete_desc->lun_id);
2945 			retval = EINVAL;
2946 			break;
2947 		}
2948 		mtx_lock(&lun->lun_lock);
2949 		mtx_unlock(&softc->ctl_lock);
2950 		STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2951 			if (desc->serial != delete_desc->serial)
2952 				continue;
2953 
2954 			STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2955 				      links);
2956 			free(desc, M_CTL);
2957 			delete_done = 1;
2958 		}
2959 		mtx_unlock(&lun->lun_lock);
2960 		if (delete_done == 0) {
2961 			printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2962 			       "error serial %ju on LUN %u\n", __func__,
2963 			       delete_desc->serial, delete_desc->lun_id);
2964 			retval = EINVAL;
2965 			break;
2966 		}
2967 		break;
2968 	}
2969 	case CTL_DUMP_STRUCTS: {
2970 		int i, j, k, idx;
2971 		struct ctl_port *port;
2972 		struct ctl_frontend *fe;
2973 
2974 		mtx_lock(&softc->ctl_lock);
2975 		printf("CTL Persistent Reservation information start:\n");
2976 		for (i = 0; i < CTL_MAX_LUNS; i++) {
2977 			struct ctl_lun *lun;
2978 
2979 			lun = softc->ctl_luns[i];
2980 
2981 			if ((lun == NULL)
2982 			 || ((lun->flags & CTL_LUN_DISABLED) != 0))
2983 				continue;
2984 
2985 			for (j = 0; j < (CTL_MAX_PORTS * 2); j++) {
2986 				for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2987 					idx = j * CTL_MAX_INIT_PER_PORT + k;
2988 					if (lun->per_res[idx].registered == 0)
2989 						continue;
2990 					printf("  LUN %d port %d iid %d key "
2991 					       "%#jx\n", i, j, k,
2992 					       (uintmax_t)scsi_8btou64(
2993 					       lun->per_res[idx].res_key.key));
2994 				}
2995 			}
2996 		}
2997 		printf("CTL Persistent Reservation information end\n");
2998 		printf("CTL Ports:\n");
2999 		STAILQ_FOREACH(port, &softc->port_list, links) {
3000 			printf("  Port %d '%s' Frontend '%s' Type %u pp %d vp %d WWNN "
3001 			       "%#jx WWPN %#jx\n", port->targ_port, port->port_name,
3002 			       port->frontend->name, port->port_type,
3003 			       port->physical_port, port->virtual_port,
3004 			       (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
3005 			for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3006 				if (port->wwpn_iid[j].in_use == 0 &&
3007 				    port->wwpn_iid[j].wwpn == 0 &&
3008 				    port->wwpn_iid[j].name == NULL)
3009 					continue;
3010 
3011 				printf("    iid %u use %d WWPN %#jx '%s'\n",
3012 				    j, port->wwpn_iid[j].in_use,
3013 				    (uintmax_t)port->wwpn_iid[j].wwpn,
3014 				    port->wwpn_iid[j].name);
3015 			}
3016 		}
3017 		printf("CTL Port information end\n");
3018 		mtx_unlock(&softc->ctl_lock);
3019 		/*
3020 		 * XXX KDM calling this without a lock.  We'd likely want
3021 		 * to drop the lock before calling the frontend's dump
3022 		 * routine anyway.
3023 		 */
3024 		printf("CTL Frontends:\n");
3025 		STAILQ_FOREACH(fe, &softc->fe_list, links) {
3026 			printf("  Frontend '%s'\n", fe->name);
3027 			if (fe->fe_dump != NULL)
3028 				fe->fe_dump();
3029 		}
3030 		printf("CTL Frontend information end\n");
3031 		break;
3032 	}
3033 	case CTL_LUN_REQ: {
3034 		struct ctl_lun_req *lun_req;
3035 		struct ctl_backend_driver *backend;
3036 
3037 		lun_req = (struct ctl_lun_req *)addr;
3038 
3039 		backend = ctl_backend_find(lun_req->backend);
3040 		if (backend == NULL) {
3041 			lun_req->status = CTL_LUN_ERROR;
3042 			snprintf(lun_req->error_str,
3043 				 sizeof(lun_req->error_str),
3044 				 "Backend \"%s\" not found.",
3045 				 lun_req->backend);
3046 			break;
3047 		}
3048 		if (lun_req->num_be_args > 0) {
3049 			lun_req->kern_be_args = ctl_copyin_args(
3050 				lun_req->num_be_args,
3051 				lun_req->be_args,
3052 				lun_req->error_str,
3053 				sizeof(lun_req->error_str));
3054 			if (lun_req->kern_be_args == NULL) {
3055 				lun_req->status = CTL_LUN_ERROR;
3056 				break;
3057 			}
3058 		}
3059 
3060 		retval = backend->ioctl(dev, cmd, addr, flag, td);
3061 
3062 		if (lun_req->num_be_args > 0) {
3063 			ctl_copyout_args(lun_req->num_be_args,
3064 				      lun_req->kern_be_args);
3065 			ctl_free_args(lun_req->num_be_args,
3066 				      lun_req->kern_be_args);
3067 		}
3068 		break;
3069 	}
3070 	case CTL_LUN_LIST: {
3071 		struct sbuf *sb;
3072 		struct ctl_lun *lun;
3073 		struct ctl_lun_list *list;
3074 		struct ctl_option *opt;
3075 
3076 		list = (struct ctl_lun_list *)addr;
3077 
3078 		/*
3079 		 * Allocate a fixed length sbuf here, based on the length
3080 		 * of the user's buffer.  We could allocate an auto-extending
3081 		 * buffer, and then tell the user how much larger our
3082 		 * amount of data is than his buffer, but that presents
3083 		 * some problems:
3084 		 *
3085 		 * 1.  The sbuf(9) routines use a blocking malloc, and so
3086 		 *     we can't hold a lock while calling them with an
3087 		 *     auto-extending buffer.
3088  		 *
3089 		 * 2.  There is not currently a LUN reference counting
3090 		 *     mechanism, outside of outstanding transactions on
3091 		 *     the LUN's OOA queue.  So a LUN could go away on us
3092 		 *     while we're getting the LUN number, backend-specific
3093 		 *     information, etc.  Thus, given the way things
3094 		 *     currently work, we need to hold the CTL lock while
3095 		 *     grabbing LUN information.
3096 		 *
3097 		 * So, from the user's standpoint, the best thing to do is
3098 		 * allocate what he thinks is a reasonable buffer length,
3099 		 * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
3100 		 * double the buffer length and try again.  (And repeat
3101 		 * that until he succeeds.)
3102 		 */
3103 		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3104 		if (sb == NULL) {
3105 			list->status = CTL_LUN_LIST_ERROR;
3106 			snprintf(list->error_str, sizeof(list->error_str),
3107 				 "Unable to allocate %d bytes for LUN list",
3108 				 list->alloc_len);
3109 			break;
3110 		}
3111 
3112 		sbuf_printf(sb, "<ctllunlist>\n");
3113 
3114 		mtx_lock(&softc->ctl_lock);
3115 		STAILQ_FOREACH(lun, &softc->lun_list, links) {
3116 			mtx_lock(&lun->lun_lock);
3117 			retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3118 					     (uintmax_t)lun->lun);
3119 
3120 			/*
3121 			 * Bail out as soon as we see that we've overfilled
3122 			 * the buffer.
3123 			 */
3124 			if (retval != 0)
3125 				break;
3126 
3127 			retval = sbuf_printf(sb, "\t<backend_type>%s"
3128 					     "</backend_type>\n",
3129 					     (lun->backend == NULL) ?  "none" :
3130 					     lun->backend->name);
3131 
3132 			if (retval != 0)
3133 				break;
3134 
3135 			retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
3136 					     lun->be_lun->lun_type);
3137 
3138 			if (retval != 0)
3139 				break;
3140 
3141 			if (lun->backend == NULL) {
3142 				retval = sbuf_printf(sb, "</lun>\n");
3143 				if (retval != 0)
3144 					break;
3145 				continue;
3146 			}
3147 
3148 			retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
3149 					     (lun->be_lun->maxlba > 0) ?
3150 					     lun->be_lun->maxlba + 1 : 0);
3151 
3152 			if (retval != 0)
3153 				break;
3154 
3155 			retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
3156 					     lun->be_lun->blocksize);
3157 
3158 			if (retval != 0)
3159 				break;
3160 
3161 			retval = sbuf_printf(sb, "\t<serial_number>");
3162 
3163 			if (retval != 0)
3164 				break;
3165 
3166 			retval = ctl_sbuf_printf_esc(sb,
3167 						     lun->be_lun->serial_num);
3168 
3169 			if (retval != 0)
3170 				break;
3171 
3172 			retval = sbuf_printf(sb, "</serial_number>\n");
3173 
3174 			if (retval != 0)
3175 				break;
3176 
3177 			retval = sbuf_printf(sb, "\t<device_id>");
3178 
3179 			if (retval != 0)
3180 				break;
3181 
3182 			retval = ctl_sbuf_printf_esc(sb,lun->be_lun->device_id);
3183 
3184 			if (retval != 0)
3185 				break;
3186 
3187 			retval = sbuf_printf(sb, "</device_id>\n");
3188 
3189 			if (retval != 0)
3190 				break;
3191 
3192 			if (lun->backend->lun_info != NULL) {
3193 				retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3194 				if (retval != 0)
3195 					break;
3196 			}
3197 			STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
3198 				retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3199 				    opt->name, opt->value, opt->name);
3200 				if (retval != 0)
3201 					break;
3202 			}
3203 
3204 			retval = sbuf_printf(sb, "</lun>\n");
3205 
3206 			if (retval != 0)
3207 				break;
3208 			mtx_unlock(&lun->lun_lock);
3209 		}
3210 		if (lun != NULL)
3211 			mtx_unlock(&lun->lun_lock);
3212 		mtx_unlock(&softc->ctl_lock);
3213 
3214 		if ((retval != 0)
3215 		 || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3216 			retval = 0;
3217 			sbuf_delete(sb);
3218 			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3219 			snprintf(list->error_str, sizeof(list->error_str),
3220 				 "Out of space, %d bytes is too small",
3221 				 list->alloc_len);
3222 			break;
3223 		}
3224 
3225 		sbuf_finish(sb);
3226 
3227 		retval = copyout(sbuf_data(sb), list->lun_xml,
3228 				 sbuf_len(sb) + 1);
3229 
3230 		list->fill_len = sbuf_len(sb) + 1;
3231 		list->status = CTL_LUN_LIST_OK;
3232 		sbuf_delete(sb);
3233 		break;
3234 	}
3235 	case CTL_ISCSI: {
3236 		struct ctl_iscsi *ci;
3237 		struct ctl_frontend *fe;
3238 
3239 		ci = (struct ctl_iscsi *)addr;
3240 
3241 		fe = ctl_frontend_find("iscsi");
3242 		if (fe == NULL) {
3243 			ci->status = CTL_ISCSI_ERROR;
3244 			snprintf(ci->error_str, sizeof(ci->error_str),
3245 			    "Frontend \"iscsi\" not found.");
3246 			break;
3247 		}
3248 
3249 		retval = fe->ioctl(dev, cmd, addr, flag, td);
3250 		break;
3251 	}
3252 	case CTL_PORT_REQ: {
3253 		struct ctl_req *req;
3254 		struct ctl_frontend *fe;
3255 
3256 		req = (struct ctl_req *)addr;
3257 
3258 		fe = ctl_frontend_find(req->driver);
3259 		if (fe == NULL) {
3260 			req->status = CTL_LUN_ERROR;
3261 			snprintf(req->error_str, sizeof(req->error_str),
3262 			    "Frontend \"%s\" not found.", req->driver);
3263 			break;
3264 		}
3265 		if (req->num_args > 0) {
3266 			req->kern_args = ctl_copyin_args(req->num_args,
3267 			    req->args, req->error_str, sizeof(req->error_str));
3268 			if (req->kern_args == NULL) {
3269 				req->status = CTL_LUN_ERROR;
3270 				break;
3271 			}
3272 		}
3273 
3274 		retval = fe->ioctl(dev, cmd, addr, flag, td);
3275 
3276 		if (req->num_args > 0) {
3277 			ctl_copyout_args(req->num_args, req->kern_args);
3278 			ctl_free_args(req->num_args, req->kern_args);
3279 		}
3280 		break;
3281 	}
3282 	case CTL_PORT_LIST: {
3283 		struct sbuf *sb;
3284 		struct ctl_port *port;
3285 		struct ctl_lun_list *list;
3286 		struct ctl_option *opt;
3287 
3288 		list = (struct ctl_lun_list *)addr;
3289 
3290 		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3291 		if (sb == NULL) {
3292 			list->status = CTL_LUN_LIST_ERROR;
3293 			snprintf(list->error_str, sizeof(list->error_str),
3294 				 "Unable to allocate %d bytes for LUN list",
3295 				 list->alloc_len);
3296 			break;
3297 		}
3298 
3299 		sbuf_printf(sb, "<ctlportlist>\n");
3300 
3301 		mtx_lock(&softc->ctl_lock);
3302 		STAILQ_FOREACH(port, &softc->port_list, links) {
3303 			retval = sbuf_printf(sb, "<targ_port id=\"%ju\">\n",
3304 					     (uintmax_t)port->targ_port);
3305 
3306 			/*
3307 			 * Bail out as soon as we see that we've overfilled
3308 			 * the buffer.
3309 			 */
3310 			if (retval != 0)
3311 				break;
3312 
3313 			retval = sbuf_printf(sb, "\t<frontend_type>%s"
3314 			    "</frontend_type>\n", port->frontend->name);
3315 			if (retval != 0)
3316 				break;
3317 
3318 			retval = sbuf_printf(sb, "\t<port_type>%d</port_type>\n",
3319 					     port->port_type);
3320 			if (retval != 0)
3321 				break;
3322 
3323 			retval = sbuf_printf(sb, "\t<online>%s</online>\n",
3324 			    (port->status & CTL_PORT_STATUS_ONLINE) ? "YES" : "NO");
3325 			if (retval != 0)
3326 				break;
3327 
3328 			retval = sbuf_printf(sb, "\t<port_name>%s</port_name>\n",
3329 			    port->port_name);
3330 			if (retval != 0)
3331 				break;
3332 
3333 			retval = sbuf_printf(sb, "\t<physical_port>%d</physical_port>\n",
3334 			    port->physical_port);
3335 			if (retval != 0)
3336 				break;
3337 
3338 			retval = sbuf_printf(sb, "\t<virtual_port>%d</virtual_port>\n",
3339 			    port->virtual_port);
3340 			if (retval != 0)
3341 				break;
3342 
3343 			retval = sbuf_printf(sb, "\t<wwnn>%#jx</wwnn>\n",
3344 			    (uintmax_t)port->wwnn);
3345 			if (retval != 0)
3346 				break;
3347 
3348 			retval = sbuf_printf(sb, "\t<wwpn>%#jx</wwpn>\n",
3349 			    (uintmax_t)port->wwpn);
3350 			if (retval != 0)
3351 				break;
3352 
3353 			if (port->port_info != NULL) {
3354 				retval = port->port_info(port->onoff_arg, sb);
3355 				if (retval != 0)
3356 					break;
3357 			}
3358 			STAILQ_FOREACH(opt, &port->options, links) {
3359 				retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3360 				    opt->name, opt->value, opt->name);
3361 				if (retval != 0)
3362 					break;
3363 			}
3364 
3365 			retval = sbuf_printf(sb, "</targ_port>\n");
3366 			if (retval != 0)
3367 				break;
3368 		}
3369 		mtx_unlock(&softc->ctl_lock);
3370 
3371 		if ((retval != 0)
3372 		 || ((retval = sbuf_printf(sb, "</ctlportlist>\n")) != 0)) {
3373 			retval = 0;
3374 			sbuf_delete(sb);
3375 			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3376 			snprintf(list->error_str, sizeof(list->error_str),
3377 				 "Out of space, %d bytes is too small",
3378 				 list->alloc_len);
3379 			break;
3380 		}
3381 
3382 		sbuf_finish(sb);
3383 
3384 		retval = copyout(sbuf_data(sb), list->lun_xml,
3385 				 sbuf_len(sb) + 1);
3386 
3387 		list->fill_len = sbuf_len(sb) + 1;
3388 		list->status = CTL_LUN_LIST_OK;
3389 		sbuf_delete(sb);
3390 		break;
3391 	}
3392 	default: {
3393 		/* XXX KDM should we fix this? */
3394 #if 0
3395 		struct ctl_backend_driver *backend;
3396 		unsigned int type;
3397 		int found;
3398 
3399 		found = 0;
3400 
3401 		/*
3402 		 * We encode the backend type as the ioctl type for backend
3403 		 * ioctls.  So parse it out here, and then search for a
3404 		 * backend of this type.
3405 		 */
3406 		type = _IOC_TYPE(cmd);
3407 
3408 		STAILQ_FOREACH(backend, &softc->be_list, links) {
3409 			if (backend->type == type) {
3410 				found = 1;
3411 				break;
3412 			}
3413 		}
3414 		if (found == 0) {
3415 			printf("ctl: unknown ioctl command %#lx or backend "
3416 			       "%d\n", cmd, type);
3417 			retval = EINVAL;
3418 			break;
3419 		}
3420 		retval = backend->ioctl(dev, cmd, addr, flag, td);
3421 #endif
3422 		retval = ENOTTY;
3423 		break;
3424 	}
3425 	}
3426 	return (retval);
3427 }
3428 
3429 uint32_t
3430 ctl_get_initindex(struct ctl_nexus *nexus)
3431 {
3432 	if (nexus->targ_port < CTL_MAX_PORTS)
3433 		return (nexus->initid.id +
3434 			(nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3435 	else
3436 		return (nexus->initid.id +
3437 		       ((nexus->targ_port - CTL_MAX_PORTS) *
3438 			CTL_MAX_INIT_PER_PORT));
3439 }
3440 
3441 uint32_t
3442 ctl_get_resindex(struct ctl_nexus *nexus)
3443 {
3444 	return (nexus->initid.id + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3445 }
3446 
3447 uint32_t
3448 ctl_port_idx(int port_num)
3449 {
3450 	if (port_num < CTL_MAX_PORTS)
3451 		return(port_num);
3452 	else
3453 		return(port_num - CTL_MAX_PORTS);
3454 }
3455 
3456 static uint32_t
3457 ctl_map_lun(int port_num, uint32_t lun_id)
3458 {
3459 	struct ctl_port *port;
3460 
3461 	port = control_softc->ctl_ports[ctl_port_idx(port_num)];
3462 	if (port == NULL)
3463 		return (UINT32_MAX);
3464 	if (port->lun_map == NULL)
3465 		return (lun_id);
3466 	return (port->lun_map(port->targ_lun_arg, lun_id));
3467 }
3468 
3469 static uint32_t
3470 ctl_map_lun_back(int port_num, uint32_t lun_id)
3471 {
3472 	struct ctl_port *port;
3473 	uint32_t i;
3474 
3475 	port = control_softc->ctl_ports[ctl_port_idx(port_num)];
3476 	if (port->lun_map == NULL)
3477 		return (lun_id);
3478 	for (i = 0; i < CTL_MAX_LUNS; i++) {
3479 		if (port->lun_map(port->targ_lun_arg, i) == lun_id)
3480 			return (i);
3481 	}
3482 	return (UINT32_MAX);
3483 }
3484 
3485 /*
3486  * Note:  This only works for bitmask sizes that are at least 32 bits, and
3487  * that are a power of 2.
3488  */
3489 int
3490 ctl_ffz(uint32_t *mask, uint32_t size)
3491 {
3492 	uint32_t num_chunks, num_pieces;
3493 	int i, j;
3494 
3495 	num_chunks = (size >> 5);
3496 	if (num_chunks == 0)
3497 		num_chunks++;
3498 	num_pieces = ctl_min((sizeof(uint32_t) * 8), size);
3499 
3500 	for (i = 0; i < num_chunks; i++) {
3501 		for (j = 0; j < num_pieces; j++) {
3502 			if ((mask[i] & (1 << j)) == 0)
3503 				return ((i << 5) + j);
3504 		}
3505 	}
3506 
3507 	return (-1);
3508 }
3509 
3510 int
3511 ctl_set_mask(uint32_t *mask, uint32_t bit)
3512 {
3513 	uint32_t chunk, piece;
3514 
3515 	chunk = bit >> 5;
3516 	piece = bit % (sizeof(uint32_t) * 8);
3517 
3518 	if ((mask[chunk] & (1 << piece)) != 0)
3519 		return (-1);
3520 	else
3521 		mask[chunk] |= (1 << piece);
3522 
3523 	return (0);
3524 }
3525 
3526 int
3527 ctl_clear_mask(uint32_t *mask, uint32_t bit)
3528 {
3529 	uint32_t chunk, piece;
3530 
3531 	chunk = bit >> 5;
3532 	piece = bit % (sizeof(uint32_t) * 8);
3533 
3534 	if ((mask[chunk] & (1 << piece)) == 0)
3535 		return (-1);
3536 	else
3537 		mask[chunk] &= ~(1 << piece);
3538 
3539 	return (0);
3540 }
3541 
3542 int
3543 ctl_is_set(uint32_t *mask, uint32_t bit)
3544 {
3545 	uint32_t chunk, piece;
3546 
3547 	chunk = bit >> 5;
3548 	piece = bit % (sizeof(uint32_t) * 8);
3549 
3550 	if ((mask[chunk] & (1 << piece)) == 0)
3551 		return (0);
3552 	else
3553 		return (1);
3554 }
3555 
3556 #ifdef unused
3557 /*
3558  * The bus, target and lun are optional, they can be filled in later.
3559  * can_wait is used to determine whether we can wait on the malloc or not.
3560  */
3561 union ctl_io*
3562 ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port, uint32_t targ_target,
3563 	      uint32_t targ_lun, int can_wait)
3564 {
3565 	union ctl_io *io;
3566 
3567 	if (can_wait)
3568 		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_WAITOK);
3569 	else
3570 		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_NOWAIT);
3571 
3572 	if (io != NULL) {
3573 		io->io_hdr.io_type = io_type;
3574 		io->io_hdr.targ_port = targ_port;
3575 		/*
3576 		 * XXX KDM this needs to change/go away.  We need to move
3577 		 * to a preallocated pool of ctl_scsiio structures.
3578 		 */
3579 		io->io_hdr.nexus.targ_target.id = targ_target;
3580 		io->io_hdr.nexus.targ_lun = targ_lun;
3581 	}
3582 
3583 	return (io);
3584 }
3585 
3586 void
3587 ctl_kfree_io(union ctl_io *io)
3588 {
3589 	free(io, M_CTL);
3590 }
3591 #endif /* unused */
3592 
3593 /*
3594  * ctl_softc, pool_type, total_ctl_io are passed in.
3595  * npool is passed out.
3596  */
3597 int
3598 ctl_pool_create(struct ctl_softc *ctl_softc, ctl_pool_type pool_type,
3599 		uint32_t total_ctl_io, struct ctl_io_pool **npool)
3600 {
3601 	uint32_t i;
3602 	union ctl_io *cur_io, *next_io;
3603 	struct ctl_io_pool *pool;
3604 	int retval;
3605 
3606 	retval = 0;
3607 
3608 	pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3609 					    M_NOWAIT | M_ZERO);
3610 	if (pool == NULL) {
3611 		retval = ENOMEM;
3612 		goto bailout;
3613 	}
3614 
3615 	pool->type = pool_type;
3616 	pool->ctl_softc = ctl_softc;
3617 
3618 	mtx_lock(&ctl_softc->pool_lock);
3619 	pool->id = ctl_softc->cur_pool_id++;
3620 	mtx_unlock(&ctl_softc->pool_lock);
3621 
3622 	pool->flags = CTL_POOL_FLAG_NONE;
3623 	pool->refcount = 1;		/* Reference for validity. */
3624 	STAILQ_INIT(&pool->free_queue);
3625 
3626 	/*
3627 	 * XXX KDM other options here:
3628 	 * - allocate a page at a time
3629 	 * - allocate one big chunk of memory.
3630 	 * Page allocation might work well, but would take a little more
3631 	 * tracking.
3632 	 */
3633 	for (i = 0; i < total_ctl_io; i++) {
3634 		cur_io = (union ctl_io *)malloc(sizeof(*cur_io), M_CTLIO,
3635 						M_NOWAIT);
3636 		if (cur_io == NULL) {
3637 			retval = ENOMEM;
3638 			break;
3639 		}
3640 		cur_io->io_hdr.pool = pool;
3641 		STAILQ_INSERT_TAIL(&pool->free_queue, &cur_io->io_hdr, links);
3642 		pool->total_ctl_io++;
3643 		pool->free_ctl_io++;
3644 	}
3645 
3646 	if (retval != 0) {
3647 		for (cur_io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3648 		     cur_io != NULL; cur_io = next_io) {
3649 			next_io = (union ctl_io *)STAILQ_NEXT(&cur_io->io_hdr,
3650 							      links);
3651 			STAILQ_REMOVE(&pool->free_queue, &cur_io->io_hdr,
3652 				      ctl_io_hdr, links);
3653 			free(cur_io, M_CTLIO);
3654 		}
3655 
3656 		free(pool, M_CTL);
3657 		goto bailout;
3658 	}
3659 	mtx_lock(&ctl_softc->pool_lock);
3660 	ctl_softc->num_pools++;
3661 	STAILQ_INSERT_TAIL(&ctl_softc->io_pools, pool, links);
3662 	/*
3663 	 * Increment our usage count if this is an external consumer, so we
3664 	 * can't get unloaded until the external consumer (most likely a
3665 	 * FETD) unloads and frees his pool.
3666 	 *
3667 	 * XXX KDM will this increment the caller's module use count, or
3668 	 * mine?
3669 	 */
3670 #if 0
3671 	if ((pool_type != CTL_POOL_EMERGENCY)
3672 	 && (pool_type != CTL_POOL_INTERNAL)
3673 	 && (pool_type != CTL_POOL_4OTHERSC))
3674 		MOD_INC_USE_COUNT;
3675 #endif
3676 
3677 	mtx_unlock(&ctl_softc->pool_lock);
3678 
3679 	*npool = pool;
3680 
3681 bailout:
3682 
3683 	return (retval);
3684 }
3685 
3686 static int
3687 ctl_pool_acquire(struct ctl_io_pool *pool)
3688 {
3689 
3690 	mtx_assert(&pool->ctl_softc->pool_lock, MA_OWNED);
3691 
3692 	if (pool->flags & CTL_POOL_FLAG_INVALID)
3693 		return (EINVAL);
3694 
3695 	pool->refcount++;
3696 
3697 	return (0);
3698 }
3699 
3700 static void
3701 ctl_pool_release(struct ctl_io_pool *pool)
3702 {
3703 	struct ctl_softc *ctl_softc = pool->ctl_softc;
3704 	union ctl_io *io;
3705 
3706 	mtx_assert(&ctl_softc->pool_lock, MA_OWNED);
3707 
3708 	if (--pool->refcount != 0)
3709 		return;
3710 
3711 	while ((io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue)) != NULL) {
3712 		STAILQ_REMOVE(&pool->free_queue, &io->io_hdr, ctl_io_hdr,
3713 			      links);
3714 		free(io, M_CTLIO);
3715 	}
3716 
3717 	STAILQ_REMOVE(&ctl_softc->io_pools, pool, ctl_io_pool, links);
3718 	ctl_softc->num_pools--;
3719 
3720 	/*
3721 	 * XXX KDM will this decrement the caller's usage count or mine?
3722 	 */
3723 #if 0
3724 	if ((pool->type != CTL_POOL_EMERGENCY)
3725 	 && (pool->type != CTL_POOL_INTERNAL)
3726 	 && (pool->type != CTL_POOL_4OTHERSC))
3727 		MOD_DEC_USE_COUNT;
3728 #endif
3729 
3730 	free(pool, M_CTL);
3731 }
3732 
3733 void
3734 ctl_pool_free(struct ctl_io_pool *pool)
3735 {
3736 	struct ctl_softc *ctl_softc;
3737 
3738 	if (pool == NULL)
3739 		return;
3740 
3741 	ctl_softc = pool->ctl_softc;
3742 	mtx_lock(&ctl_softc->pool_lock);
3743 	pool->flags |= CTL_POOL_FLAG_INVALID;
3744 	ctl_pool_release(pool);
3745 	mtx_unlock(&ctl_softc->pool_lock);
3746 }
3747 
3748 /*
3749  * This routine does not block (except for spinlocks of course).
3750  * It tries to allocate a ctl_io union from the caller's pool as quickly as
3751  * possible.
3752  */
3753 union ctl_io *
3754 ctl_alloc_io(void *pool_ref)
3755 {
3756 	union ctl_io *io;
3757 	struct ctl_softc *ctl_softc;
3758 	struct ctl_io_pool *pool, *npool;
3759 	struct ctl_io_pool *emergency_pool;
3760 
3761 	pool = (struct ctl_io_pool *)pool_ref;
3762 
3763 	if (pool == NULL) {
3764 		printf("%s: pool is NULL\n", __func__);
3765 		return (NULL);
3766 	}
3767 
3768 	emergency_pool = NULL;
3769 
3770 	ctl_softc = pool->ctl_softc;
3771 
3772 	mtx_lock(&ctl_softc->pool_lock);
3773 	/*
3774 	 * First, try to get the io structure from the user's pool.
3775 	 */
3776 	if (ctl_pool_acquire(pool) == 0) {
3777 		io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3778 		if (io != NULL) {
3779 			STAILQ_REMOVE_HEAD(&pool->free_queue, links);
3780 			pool->total_allocated++;
3781 			pool->free_ctl_io--;
3782 			mtx_unlock(&ctl_softc->pool_lock);
3783 			return (io);
3784 		} else
3785 			ctl_pool_release(pool);
3786 	}
3787 	/*
3788 	 * If he doesn't have any io structures left, search for an
3789 	 * emergency pool and grab one from there.
3790 	 */
3791 	STAILQ_FOREACH(npool, &ctl_softc->io_pools, links) {
3792 		if (npool->type != CTL_POOL_EMERGENCY)
3793 			continue;
3794 
3795 		if (ctl_pool_acquire(npool) != 0)
3796 			continue;
3797 
3798 		emergency_pool = npool;
3799 
3800 		io = (union ctl_io *)STAILQ_FIRST(&npool->free_queue);
3801 		if (io != NULL) {
3802 			STAILQ_REMOVE_HEAD(&npool->free_queue, links);
3803 			npool->total_allocated++;
3804 			npool->free_ctl_io--;
3805 			mtx_unlock(&ctl_softc->pool_lock);
3806 			return (io);
3807 		} else
3808 			ctl_pool_release(npool);
3809 	}
3810 
3811 	/* Drop the spinlock before we malloc */
3812 	mtx_unlock(&ctl_softc->pool_lock);
3813 
3814 	/*
3815 	 * The emergency pool (if it exists) didn't have one, so try an
3816 	 * atomic (i.e. nonblocking) malloc and see if we get lucky.
3817 	 */
3818 	io = (union ctl_io *)malloc(sizeof(*io), M_CTLIO, M_NOWAIT);
3819 	if (io != NULL) {
3820 		/*
3821 		 * If the emergency pool exists but is empty, add this
3822 		 * ctl_io to its list when it gets freed.
3823 		 */
3824 		if (emergency_pool != NULL) {
3825 			mtx_lock(&ctl_softc->pool_lock);
3826 			if (ctl_pool_acquire(emergency_pool) == 0) {
3827 				io->io_hdr.pool = emergency_pool;
3828 				emergency_pool->total_ctl_io++;
3829 				/*
3830 				 * Need to bump this, otherwise
3831 				 * total_allocated and total_freed won't
3832 				 * match when we no longer have anything
3833 				 * outstanding.
3834 				 */
3835 				emergency_pool->total_allocated++;
3836 			}
3837 			mtx_unlock(&ctl_softc->pool_lock);
3838 		} else
3839 			io->io_hdr.pool = NULL;
3840 	}
3841 
3842 	return (io);
3843 }
3844 
3845 void
3846 ctl_free_io(union ctl_io *io)
3847 {
3848 	if (io == NULL)
3849 		return;
3850 
3851 	/*
3852 	 * If this ctl_io has a pool, return it to that pool.
3853 	 */
3854 	if (io->io_hdr.pool != NULL) {
3855 		struct ctl_io_pool *pool;
3856 
3857 		pool = (struct ctl_io_pool *)io->io_hdr.pool;
3858 		mtx_lock(&pool->ctl_softc->pool_lock);
3859 		io->io_hdr.io_type = 0xff;
3860 		STAILQ_INSERT_TAIL(&pool->free_queue, &io->io_hdr, links);
3861 		pool->total_freed++;
3862 		pool->free_ctl_io++;
3863 		ctl_pool_release(pool);
3864 		mtx_unlock(&pool->ctl_softc->pool_lock);
3865 	} else {
3866 		/*
3867 		 * Otherwise, just free it.  We probably malloced it and
3868 		 * the emergency pool wasn't available.
3869 		 */
3870 		free(io, M_CTLIO);
3871 	}
3872 
3873 }
3874 
3875 void
3876 ctl_zero_io(union ctl_io *io)
3877 {
3878 	void *pool_ref;
3879 
3880 	if (io == NULL)
3881 		return;
3882 
3883 	/*
3884 	 * May need to preserve linked list pointers at some point too.
3885 	 */
3886 	pool_ref = io->io_hdr.pool;
3887 
3888 	memset(io, 0, sizeof(*io));
3889 
3890 	io->io_hdr.pool = pool_ref;
3891 }
3892 
3893 /*
3894  * This routine is currently used for internal copies of ctl_ios that need
3895  * to persist for some reason after we've already returned status to the
3896  * FETD.  (Thus the flag set.)
3897  *
3898  * XXX XXX
3899  * Note that this makes a blind copy of all fields in the ctl_io, except
3900  * for the pool reference.  This includes any memory that has been
3901  * allocated!  That memory will no longer be valid after done has been
3902  * called, so this would be VERY DANGEROUS for command that actually does
3903  * any reads or writes.  Right now (11/7/2005), this is only used for immediate
3904  * start and stop commands, which don't transfer any data, so this is not a
3905  * problem.  If it is used for anything else, the caller would also need to
3906  * allocate data buffer space and this routine would need to be modified to
3907  * copy the data buffer(s) as well.
3908  */
3909 void
3910 ctl_copy_io(union ctl_io *src, union ctl_io *dest)
3911 {
3912 	void *pool_ref;
3913 
3914 	if ((src == NULL)
3915 	 || (dest == NULL))
3916 		return;
3917 
3918 	/*
3919 	 * May need to preserve linked list pointers at some point too.
3920 	 */
3921 	pool_ref = dest->io_hdr.pool;
3922 
3923 	memcpy(dest, src, ctl_min(sizeof(*src), sizeof(*dest)));
3924 
3925 	dest->io_hdr.pool = pool_ref;
3926 	/*
3927 	 * We need to know that this is an internal copy, and doesn't need
3928 	 * to get passed back to the FETD that allocated it.
3929 	 */
3930 	dest->io_hdr.flags |= CTL_FLAG_INT_COPY;
3931 }
3932 
3933 #ifdef NEEDTOPORT
3934 static void
3935 ctl_update_power_subpage(struct copan_power_subpage *page)
3936 {
3937 	int num_luns, num_partitions, config_type;
3938 	struct ctl_softc *softc;
3939 	cs_BOOL_t aor_present, shelf_50pct_power;
3940 	cs_raidset_personality_t rs_type;
3941 	int max_active_luns;
3942 
3943 	softc = control_softc;
3944 
3945 	/* subtract out the processor LUN */
3946 	num_luns = softc->num_luns - 1;
3947 	/*
3948 	 * Default to 7 LUNs active, which was the only number we allowed
3949 	 * in the past.
3950 	 */
3951 	max_active_luns = 7;
3952 
3953 	num_partitions = config_GetRsPartitionInfo();
3954 	config_type = config_GetConfigType();
3955 	shelf_50pct_power = config_GetShelfPowerMode();
3956 	aor_present = config_IsAorRsPresent();
3957 
3958 	rs_type = ddb_GetRsRaidType(1);
3959 	if ((rs_type != CS_RAIDSET_PERSONALITY_RAID5)
3960 	 && (rs_type != CS_RAIDSET_PERSONALITY_RAID1)) {
3961 		EPRINT(0, "Unsupported RS type %d!", rs_type);
3962 	}
3963 
3964 
3965 	page->total_luns = num_luns;
3966 
3967 	switch (config_type) {
3968 	case 40:
3969 		/*
3970 		 * In a 40 drive configuration, it doesn't matter what DC
3971 		 * cards we have, whether we have AOR enabled or not,
3972 		 * partitioning or not, or what type of RAIDset we have.
3973 		 * In that scenario, we can power up every LUN we present
3974 		 * to the user.
3975 		 */
3976 		max_active_luns = num_luns;
3977 
3978 		break;
3979 	case 64:
3980 		if (shelf_50pct_power == CS_FALSE) {
3981 			/* 25% power */
3982 			if (aor_present == CS_TRUE) {
3983 				if (rs_type ==
3984 				     CS_RAIDSET_PERSONALITY_RAID5) {
3985 					max_active_luns = 7;
3986 				} else if (rs_type ==
3987 					 CS_RAIDSET_PERSONALITY_RAID1){
3988 					max_active_luns = 14;
3989 				} else {
3990 					/* XXX KDM now what?? */
3991 				}
3992 			} else {
3993 				if (rs_type ==
3994 				     CS_RAIDSET_PERSONALITY_RAID5) {
3995 					max_active_luns = 8;
3996 				} else if (rs_type ==
3997 					 CS_RAIDSET_PERSONALITY_RAID1){
3998 					max_active_luns = 16;
3999 				} else {
4000 					/* XXX KDM now what?? */
4001 				}
4002 			}
4003 		} else {
4004 			/* 50% power */
4005 			/*
4006 			 * With 50% power in a 64 drive configuration, we
4007 			 * can power all LUNs we present.
4008 			 */
4009 			max_active_luns = num_luns;
4010 		}
4011 		break;
4012 	case 112:
4013 		if (shelf_50pct_power == CS_FALSE) {
4014 			/* 25% power */
4015 			if (aor_present == CS_TRUE) {
4016 				if (rs_type ==
4017 				     CS_RAIDSET_PERSONALITY_RAID5) {
4018 					max_active_luns = 7;
4019 				} else if (rs_type ==
4020 					 CS_RAIDSET_PERSONALITY_RAID1){
4021 					max_active_luns = 14;
4022 				} else {
4023 					/* XXX KDM now what?? */
4024 				}
4025 			} else {
4026 				if (rs_type ==
4027 				     CS_RAIDSET_PERSONALITY_RAID5) {
4028 					max_active_luns = 8;
4029 				} else if (rs_type ==
4030 					 CS_RAIDSET_PERSONALITY_RAID1){
4031 					max_active_luns = 16;
4032 				} else {
4033 					/* XXX KDM now what?? */
4034 				}
4035 			}
4036 		} else {
4037 			/* 50% power */
4038 			if (aor_present == CS_TRUE) {
4039 				if (rs_type ==
4040 				     CS_RAIDSET_PERSONALITY_RAID5) {
4041 					max_active_luns = 14;
4042 				} else if (rs_type ==
4043 					 CS_RAIDSET_PERSONALITY_RAID1){
4044 					/*
4045 					 * We're assuming here that disk
4046 					 * caching is enabled, and so we're
4047 					 * able to power up half of each
4048 					 * LUN, and cache all writes.
4049 					 */
4050 					max_active_luns = num_luns;
4051 				} else {
4052 					/* XXX KDM now what?? */
4053 				}
4054 			} else {
4055 				if (rs_type ==
4056 				     CS_RAIDSET_PERSONALITY_RAID5) {
4057 					max_active_luns = 15;
4058 				} else if (rs_type ==
4059 					 CS_RAIDSET_PERSONALITY_RAID1){
4060 					max_active_luns = 30;
4061 				} else {
4062 					/* XXX KDM now what?? */
4063 				}
4064 			}
4065 		}
4066 		break;
4067 	default:
4068 		/*
4069 		 * In this case, we have an unknown configuration, so we
4070 		 * just use the default from above.
4071 		 */
4072 		break;
4073 	}
4074 
4075 	page->max_active_luns = max_active_luns;
4076 #if 0
4077 	printk("%s: total_luns = %d, max_active_luns = %d\n", __func__,
4078 	       page->total_luns, page->max_active_luns);
4079 #endif
4080 }
4081 #endif /* NEEDTOPORT */
4082 
4083 /*
4084  * This routine could be used in the future to load default and/or saved
4085  * mode page parameters for a particuar lun.
4086  */
4087 static int
4088 ctl_init_page_index(struct ctl_lun *lun)
4089 {
4090 	int i;
4091 	struct ctl_page_index *page_index;
4092 	struct ctl_softc *softc;
4093 
4094 	memcpy(&lun->mode_pages.index, page_index_template,
4095 	       sizeof(page_index_template));
4096 
4097 	softc = lun->ctl_softc;
4098 
4099 	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
4100 
4101 		page_index = &lun->mode_pages.index[i];
4102 		/*
4103 		 * If this is a disk-only mode page, there's no point in
4104 		 * setting it up.  For some pages, we have to have some
4105 		 * basic information about the disk in order to calculate the
4106 		 * mode page data.
4107 		 */
4108 		if ((lun->be_lun->lun_type != T_DIRECT)
4109 		 && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
4110 			continue;
4111 
4112 		switch (page_index->page_code & SMPH_PC_MASK) {
4113 		case SMS_FORMAT_DEVICE_PAGE: {
4114 			struct scsi_format_page *format_page;
4115 
4116 			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4117 				panic("subpage is incorrect!");
4118 
4119 			/*
4120 			 * Sectors per track are set above.  Bytes per
4121 			 * sector need to be set here on a per-LUN basis.
4122 			 */
4123 			memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
4124 			       &format_page_default,
4125 			       sizeof(format_page_default));
4126 			memcpy(&lun->mode_pages.format_page[
4127 			       CTL_PAGE_CHANGEABLE], &format_page_changeable,
4128 			       sizeof(format_page_changeable));
4129 			memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
4130 			       &format_page_default,
4131 			       sizeof(format_page_default));
4132 			memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
4133 			       &format_page_default,
4134 			       sizeof(format_page_default));
4135 
4136 			format_page = &lun->mode_pages.format_page[
4137 				CTL_PAGE_CURRENT];
4138 			scsi_ulto2b(lun->be_lun->blocksize,
4139 				    format_page->bytes_per_sector);
4140 
4141 			format_page = &lun->mode_pages.format_page[
4142 				CTL_PAGE_DEFAULT];
4143 			scsi_ulto2b(lun->be_lun->blocksize,
4144 				    format_page->bytes_per_sector);
4145 
4146 			format_page = &lun->mode_pages.format_page[
4147 				CTL_PAGE_SAVED];
4148 			scsi_ulto2b(lun->be_lun->blocksize,
4149 				    format_page->bytes_per_sector);
4150 
4151 			page_index->page_data =
4152 				(uint8_t *)lun->mode_pages.format_page;
4153 			break;
4154 		}
4155 		case SMS_RIGID_DISK_PAGE: {
4156 			struct scsi_rigid_disk_page *rigid_disk_page;
4157 			uint32_t sectors_per_cylinder;
4158 			uint64_t cylinders;
4159 #ifndef	__XSCALE__
4160 			int shift;
4161 #endif /* !__XSCALE__ */
4162 
4163 			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4164 				panic("invalid subpage value %d",
4165 				      page_index->subpage);
4166 
4167 			/*
4168 			 * Rotation rate and sectors per track are set
4169 			 * above.  We calculate the cylinders here based on
4170 			 * capacity.  Due to the number of heads and
4171 			 * sectors per track we're using, smaller arrays
4172 			 * may turn out to have 0 cylinders.  Linux and
4173 			 * FreeBSD don't pay attention to these mode pages
4174 			 * to figure out capacity, but Solaris does.  It
4175 			 * seems to deal with 0 cylinders just fine, and
4176 			 * works out a fake geometry based on the capacity.
4177 			 */
4178 			memcpy(&lun->mode_pages.rigid_disk_page[
4179 			       CTL_PAGE_CURRENT], &rigid_disk_page_default,
4180 			       sizeof(rigid_disk_page_default));
4181 			memcpy(&lun->mode_pages.rigid_disk_page[
4182 			       CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
4183 			       sizeof(rigid_disk_page_changeable));
4184 			memcpy(&lun->mode_pages.rigid_disk_page[
4185 			       CTL_PAGE_DEFAULT], &rigid_disk_page_default,
4186 			       sizeof(rigid_disk_page_default));
4187 			memcpy(&lun->mode_pages.rigid_disk_page[
4188 			       CTL_PAGE_SAVED], &rigid_disk_page_default,
4189 			       sizeof(rigid_disk_page_default));
4190 
4191 			sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
4192 				CTL_DEFAULT_HEADS;
4193 
4194 			/*
4195 			 * The divide method here will be more accurate,
4196 			 * probably, but results in floating point being
4197 			 * used in the kernel on i386 (__udivdi3()).  On the
4198 			 * XScale, though, __udivdi3() is implemented in
4199 			 * software.
4200 			 *
4201 			 * The shift method for cylinder calculation is
4202 			 * accurate if sectors_per_cylinder is a power of
4203 			 * 2.  Otherwise it might be slightly off -- you
4204 			 * might have a bit of a truncation problem.
4205 			 */
4206 #ifdef	__XSCALE__
4207 			cylinders = (lun->be_lun->maxlba + 1) /
4208 				sectors_per_cylinder;
4209 #else
4210 			for (shift = 31; shift > 0; shift--) {
4211 				if (sectors_per_cylinder & (1 << shift))
4212 					break;
4213 			}
4214 			cylinders = (lun->be_lun->maxlba + 1) >> shift;
4215 #endif
4216 
4217 			/*
4218 			 * We've basically got 3 bytes, or 24 bits for the
4219 			 * cylinder size in the mode page.  If we're over,
4220 			 * just round down to 2^24.
4221 			 */
4222 			if (cylinders > 0xffffff)
4223 				cylinders = 0xffffff;
4224 
4225 			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4226 				CTL_PAGE_CURRENT];
4227 			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4228 
4229 			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4230 				CTL_PAGE_DEFAULT];
4231 			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4232 
4233 			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4234 				CTL_PAGE_SAVED];
4235 			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4236 
4237 			page_index->page_data =
4238 				(uint8_t *)lun->mode_pages.rigid_disk_page;
4239 			break;
4240 		}
4241 		case SMS_CACHING_PAGE: {
4242 
4243 			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4244 				panic("invalid subpage value %d",
4245 				      page_index->subpage);
4246 			/*
4247 			 * Defaults should be okay here, no calculations
4248 			 * needed.
4249 			 */
4250 			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4251 			       &caching_page_default,
4252 			       sizeof(caching_page_default));
4253 			memcpy(&lun->mode_pages.caching_page[
4254 			       CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4255 			       sizeof(caching_page_changeable));
4256 			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4257 			       &caching_page_default,
4258 			       sizeof(caching_page_default));
4259 			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4260 			       &caching_page_default,
4261 			       sizeof(caching_page_default));
4262 			page_index->page_data =
4263 				(uint8_t *)lun->mode_pages.caching_page;
4264 			break;
4265 		}
4266 		case SMS_CONTROL_MODE_PAGE: {
4267 
4268 			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4269 				panic("invalid subpage value %d",
4270 				      page_index->subpage);
4271 
4272 			/*
4273 			 * Defaults should be okay here, no calculations
4274 			 * needed.
4275 			 */
4276 			memcpy(&lun->mode_pages.control_page[CTL_PAGE_CURRENT],
4277 			       &control_page_default,
4278 			       sizeof(control_page_default));
4279 			memcpy(&lun->mode_pages.control_page[
4280 			       CTL_PAGE_CHANGEABLE], &control_page_changeable,
4281 			       sizeof(control_page_changeable));
4282 			memcpy(&lun->mode_pages.control_page[CTL_PAGE_DEFAULT],
4283 			       &control_page_default,
4284 			       sizeof(control_page_default));
4285 			memcpy(&lun->mode_pages.control_page[CTL_PAGE_SAVED],
4286 			       &control_page_default,
4287 			       sizeof(control_page_default));
4288 			page_index->page_data =
4289 				(uint8_t *)lun->mode_pages.control_page;
4290 			break;
4291 
4292 		}
4293 		case SMS_VENDOR_SPECIFIC_PAGE:{
4294 			switch (page_index->subpage) {
4295 			case PWR_SUBPAGE_CODE: {
4296 				struct copan_power_subpage *current_page,
4297 							   *saved_page;
4298 
4299 				memcpy(&lun->mode_pages.power_subpage[
4300 				       CTL_PAGE_CURRENT],
4301 				       &power_page_default,
4302 				       sizeof(power_page_default));
4303 				memcpy(&lun->mode_pages.power_subpage[
4304 				       CTL_PAGE_CHANGEABLE],
4305 				       &power_page_changeable,
4306 				       sizeof(power_page_changeable));
4307 				memcpy(&lun->mode_pages.power_subpage[
4308 				       CTL_PAGE_DEFAULT],
4309 				       &power_page_default,
4310 				       sizeof(power_page_default));
4311 				memcpy(&lun->mode_pages.power_subpage[
4312 				       CTL_PAGE_SAVED],
4313 				       &power_page_default,
4314 				       sizeof(power_page_default));
4315 				page_index->page_data =
4316 				    (uint8_t *)lun->mode_pages.power_subpage;
4317 
4318 				current_page = (struct copan_power_subpage *)
4319 					(page_index->page_data +
4320 					 (page_index->page_len *
4321 					  CTL_PAGE_CURRENT));
4322 			        saved_page = (struct copan_power_subpage *)
4323 				        (page_index->page_data +
4324 					 (page_index->page_len *
4325 					  CTL_PAGE_SAVED));
4326 				break;
4327 			}
4328 			case APS_SUBPAGE_CODE: {
4329 				struct copan_aps_subpage *current_page,
4330 							 *saved_page;
4331 
4332 				// This gets set multiple times but
4333 				// it should always be the same. It's
4334 				// only done during init so who cares.
4335 				index_to_aps_page = i;
4336 
4337 				memcpy(&lun->mode_pages.aps_subpage[
4338 				       CTL_PAGE_CURRENT],
4339 				       &aps_page_default,
4340 				       sizeof(aps_page_default));
4341 				memcpy(&lun->mode_pages.aps_subpage[
4342 				       CTL_PAGE_CHANGEABLE],
4343 				       &aps_page_changeable,
4344 				       sizeof(aps_page_changeable));
4345 				memcpy(&lun->mode_pages.aps_subpage[
4346 				       CTL_PAGE_DEFAULT],
4347 				       &aps_page_default,
4348 				       sizeof(aps_page_default));
4349 				memcpy(&lun->mode_pages.aps_subpage[
4350 				       CTL_PAGE_SAVED],
4351 				       &aps_page_default,
4352 				       sizeof(aps_page_default));
4353 				page_index->page_data =
4354 					(uint8_t *)lun->mode_pages.aps_subpage;
4355 
4356 				current_page = (struct copan_aps_subpage *)
4357 					(page_index->page_data +
4358 					 (page_index->page_len *
4359 					  CTL_PAGE_CURRENT));
4360 				saved_page = (struct copan_aps_subpage *)
4361 					(page_index->page_data +
4362 					 (page_index->page_len *
4363 					  CTL_PAGE_SAVED));
4364 				break;
4365 			}
4366 			case DBGCNF_SUBPAGE_CODE: {
4367 				struct copan_debugconf_subpage *current_page,
4368 							       *saved_page;
4369 
4370 				memcpy(&lun->mode_pages.debugconf_subpage[
4371 				       CTL_PAGE_CURRENT],
4372 				       &debugconf_page_default,
4373 				       sizeof(debugconf_page_default));
4374 				memcpy(&lun->mode_pages.debugconf_subpage[
4375 				       CTL_PAGE_CHANGEABLE],
4376 				       &debugconf_page_changeable,
4377 				       sizeof(debugconf_page_changeable));
4378 				memcpy(&lun->mode_pages.debugconf_subpage[
4379 				       CTL_PAGE_DEFAULT],
4380 				       &debugconf_page_default,
4381 				       sizeof(debugconf_page_default));
4382 				memcpy(&lun->mode_pages.debugconf_subpage[
4383 				       CTL_PAGE_SAVED],
4384 				       &debugconf_page_default,
4385 				       sizeof(debugconf_page_default));
4386 				page_index->page_data =
4387 					(uint8_t *)lun->mode_pages.debugconf_subpage;
4388 
4389 				current_page = (struct copan_debugconf_subpage *)
4390 					(page_index->page_data +
4391 					 (page_index->page_len *
4392 					  CTL_PAGE_CURRENT));
4393 				saved_page = (struct copan_debugconf_subpage *)
4394 					(page_index->page_data +
4395 					 (page_index->page_len *
4396 					  CTL_PAGE_SAVED));
4397 				break;
4398 			}
4399 			default:
4400 				panic("invalid subpage value %d",
4401 				      page_index->subpage);
4402 				break;
4403 			}
4404    			break;
4405 		}
4406 		default:
4407 			panic("invalid page value %d",
4408 			      page_index->page_code & SMPH_PC_MASK);
4409 			break;
4410     	}
4411 	}
4412 
4413 	return (CTL_RETVAL_COMPLETE);
4414 }
4415 
4416 /*
4417  * LUN allocation.
4418  *
4419  * Requirements:
4420  * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4421  *   wants us to allocate the LUN and he can block.
4422  * - ctl_softc is always set
4423  * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4424  *
4425  * Returns 0 for success, non-zero (errno) for failure.
4426  */
4427 static int
4428 ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4429 	      struct ctl_be_lun *const be_lun, struct ctl_id target_id)
4430 {
4431 	struct ctl_lun *nlun, *lun;
4432 	struct ctl_port *port;
4433 	struct scsi_vpd_id_descriptor *desc;
4434 	struct scsi_vpd_id_t10 *t10id;
4435 	const char *eui, *naa, *scsiname, *vendor;
4436 	int lun_number, i, lun_malloced;
4437 	int devidlen, idlen1, idlen2 = 0, len;
4438 
4439 	if (be_lun == NULL)
4440 		return (EINVAL);
4441 
4442 	/*
4443 	 * We currently only support Direct Access or Processor LUN types.
4444 	 */
4445 	switch (be_lun->lun_type) {
4446 	case T_DIRECT:
4447 		break;
4448 	case T_PROCESSOR:
4449 		break;
4450 	case T_SEQUENTIAL:
4451 	case T_CHANGER:
4452 	default:
4453 		be_lun->lun_config_status(be_lun->be_lun,
4454 					  CTL_LUN_CONFIG_FAILURE);
4455 		break;
4456 	}
4457 	if (ctl_lun == NULL) {
4458 		lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4459 		lun_malloced = 1;
4460 	} else {
4461 		lun_malloced = 0;
4462 		lun = ctl_lun;
4463 	}
4464 
4465 	memset(lun, 0, sizeof(*lun));
4466 	if (lun_malloced)
4467 		lun->flags = CTL_LUN_MALLOCED;
4468 
4469 	/* Generate LUN ID. */
4470 	devidlen = max(CTL_DEVID_MIN_LEN,
4471 	    strnlen(be_lun->device_id, CTL_DEVID_LEN));
4472 	idlen1 = sizeof(*t10id) + devidlen;
4473 	len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4474 	scsiname = ctl_get_opt(&be_lun->options, "scsiname");
4475 	if (scsiname != NULL) {
4476 		idlen2 = roundup2(strlen(scsiname) + 1, 4);
4477 		len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4478 	}
4479 	eui = ctl_get_opt(&be_lun->options, "eui");
4480 	if (eui != NULL) {
4481 		len += sizeof(struct scsi_vpd_id_descriptor) + 8;
4482 	}
4483 	naa = ctl_get_opt(&be_lun->options, "naa");
4484 	if (naa != NULL) {
4485 		len += sizeof(struct scsi_vpd_id_descriptor) + 8;
4486 	}
4487 	lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4488 	    M_CTL, M_WAITOK | M_ZERO);
4489 	lun->lun_devid->len = len;
4490 	desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4491 	desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4492 	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4493 	desc->length = idlen1;
4494 	t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4495 	memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4496 	if ((vendor = ctl_get_opt(&be_lun->options, "vendor")) == NULL) {
4497 		strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4498 	} else {
4499 		strncpy(t10id->vendor, vendor,
4500 		    min(sizeof(t10id->vendor), strlen(vendor)));
4501 	}
4502 	strncpy((char *)t10id->vendor_spec_id,
4503 	    (char *)be_lun->device_id, devidlen);
4504 	if (scsiname != NULL) {
4505 		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4506 		    desc->length);
4507 		desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4508 		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4509 		    SVPD_ID_TYPE_SCSI_NAME;
4510 		desc->length = idlen2;
4511 		strlcpy(desc->identifier, scsiname, idlen2);
4512 	}
4513 	if (eui != NULL) {
4514 		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4515 		    desc->length);
4516 		desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4517 		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4518 		    SVPD_ID_TYPE_EUI64;
4519 		desc->length = 8;
4520 		scsi_u64to8b(strtouq(eui, NULL, 0), desc->identifier);
4521 	}
4522 	if (naa != NULL) {
4523 		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4524 		    desc->length);
4525 		desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4526 		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4527 		    SVPD_ID_TYPE_NAA;
4528 		desc->length = 8;
4529 		scsi_u64to8b(strtouq(naa, NULL, 0), desc->identifier);
4530 	}
4531 
4532 	mtx_lock(&ctl_softc->ctl_lock);
4533 	/*
4534 	 * See if the caller requested a particular LUN number.  If so, see
4535 	 * if it is available.  Otherwise, allocate the first available LUN.
4536 	 */
4537 	if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4538 		if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4539 		 || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4540 			mtx_unlock(&ctl_softc->ctl_lock);
4541 			if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4542 				printf("ctl: requested LUN ID %d is higher "
4543 				       "than CTL_MAX_LUNS - 1 (%d)\n",
4544 				       be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4545 			} else {
4546 				/*
4547 				 * XXX KDM return an error, or just assign
4548 				 * another LUN ID in this case??
4549 				 */
4550 				printf("ctl: requested LUN ID %d is already "
4551 				       "in use\n", be_lun->req_lun_id);
4552 			}
4553 			if (lun->flags & CTL_LUN_MALLOCED)
4554 				free(lun, M_CTL);
4555 			be_lun->lun_config_status(be_lun->be_lun,
4556 						  CTL_LUN_CONFIG_FAILURE);
4557 			return (ENOSPC);
4558 		}
4559 		lun_number = be_lun->req_lun_id;
4560 	} else {
4561 		lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, CTL_MAX_LUNS);
4562 		if (lun_number == -1) {
4563 			mtx_unlock(&ctl_softc->ctl_lock);
4564 			printf("ctl: can't allocate LUN on target %ju, out of "
4565 			       "LUNs\n", (uintmax_t)target_id.id);
4566 			if (lun->flags & CTL_LUN_MALLOCED)
4567 				free(lun, M_CTL);
4568 			be_lun->lun_config_status(be_lun->be_lun,
4569 						  CTL_LUN_CONFIG_FAILURE);
4570 			return (ENOSPC);
4571 		}
4572 	}
4573 	ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4574 
4575 	mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4576 	lun->target = target_id;
4577 	lun->lun = lun_number;
4578 	lun->be_lun = be_lun;
4579 	/*
4580 	 * The processor LUN is always enabled.  Disk LUNs come on line
4581 	 * disabled, and must be enabled by the backend.
4582 	 */
4583 	lun->flags |= CTL_LUN_DISABLED;
4584 	lun->backend = be_lun->be;
4585 	be_lun->ctl_lun = lun;
4586 	be_lun->lun_id = lun_number;
4587 	atomic_add_int(&be_lun->be->num_luns, 1);
4588 	if (be_lun->flags & CTL_LUN_FLAG_POWERED_OFF)
4589 		lun->flags |= CTL_LUN_STOPPED;
4590 
4591 	if (be_lun->flags & CTL_LUN_FLAG_INOPERABLE)
4592 		lun->flags |= CTL_LUN_INOPERABLE;
4593 
4594 	if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4595 		lun->flags |= CTL_LUN_PRIMARY_SC;
4596 
4597 	lun->ctl_softc = ctl_softc;
4598 	TAILQ_INIT(&lun->ooa_queue);
4599 	TAILQ_INIT(&lun->blocked_queue);
4600 	STAILQ_INIT(&lun->error_list);
4601 	ctl_tpc_lun_init(lun);
4602 
4603 	/*
4604 	 * Initialize the mode page index.
4605 	 */
4606 	ctl_init_page_index(lun);
4607 
4608 	/*
4609 	 * Set the poweron UA for all initiators on this LUN only.
4610 	 */
4611 	for (i = 0; i < CTL_MAX_INITIATORS; i++)
4612 		lun->pending_ua[i] = CTL_UA_POWERON;
4613 
4614 	/*
4615 	 * Now, before we insert this lun on the lun list, set the lun
4616 	 * inventory changed UA for all other luns.
4617 	 */
4618 	STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4619 		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4620 			nlun->pending_ua[i] |= CTL_UA_LUN_CHANGE;
4621 		}
4622 	}
4623 
4624 	STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4625 
4626 	ctl_softc->ctl_luns[lun_number] = lun;
4627 
4628 	ctl_softc->num_luns++;
4629 
4630 	/* Setup statistics gathering */
4631 	lun->stats.device_type = be_lun->lun_type;
4632 	lun->stats.lun_number = lun_number;
4633 	if (lun->stats.device_type == T_DIRECT)
4634 		lun->stats.blocksize = be_lun->blocksize;
4635 	else
4636 		lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4637 	for (i = 0;i < CTL_MAX_PORTS;i++)
4638 		lun->stats.ports[i].targ_port = i;
4639 
4640 	mtx_unlock(&ctl_softc->ctl_lock);
4641 
4642 	lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4643 
4644 	/*
4645 	 * Run through each registered FETD and bring it online if it isn't
4646 	 * already.  Enable the target ID if it hasn't been enabled, and
4647 	 * enable this particular LUN.
4648 	 */
4649 	STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4650 		int retval;
4651 
4652 		retval = port->lun_enable(port->targ_lun_arg, target_id,lun_number);
4653 		if (retval != 0) {
4654 			printf("ctl_alloc_lun: FETD %s port %d returned error "
4655 			       "%d for lun_enable on target %ju lun %d\n",
4656 			       port->port_name, port->targ_port, retval,
4657 			       (uintmax_t)target_id.id, lun_number);
4658 		} else
4659 			port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4660 	}
4661 	return (0);
4662 }
4663 
4664 /*
4665  * Delete a LUN.
4666  * Assumptions:
4667  * - LUN has already been marked invalid and any pending I/O has been taken
4668  *   care of.
4669  */
4670 static int
4671 ctl_free_lun(struct ctl_lun *lun)
4672 {
4673 	struct ctl_softc *softc;
4674 #if 0
4675 	struct ctl_port *port;
4676 #endif
4677 	struct ctl_lun *nlun;
4678 	int i;
4679 
4680 	softc = lun->ctl_softc;
4681 
4682 	mtx_assert(&softc->ctl_lock, MA_OWNED);
4683 
4684 	STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4685 
4686 	ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4687 
4688 	softc->ctl_luns[lun->lun] = NULL;
4689 
4690 	if (!TAILQ_EMPTY(&lun->ooa_queue))
4691 		panic("Freeing a LUN %p with outstanding I/O!!\n", lun);
4692 
4693 	softc->num_luns--;
4694 
4695 	/*
4696 	 * XXX KDM this scheme only works for a single target/multiple LUN
4697 	 * setup.  It needs to be revamped for a multiple target scheme.
4698 	 *
4699 	 * XXX KDM this results in port->lun_disable() getting called twice,
4700 	 * once when ctl_disable_lun() is called, and a second time here.
4701 	 * We really need to re-think the LUN disable semantics.  There
4702 	 * should probably be several steps/levels to LUN removal:
4703 	 *  - disable
4704 	 *  - invalidate
4705 	 *  - free
4706  	 *
4707 	 * Right now we only have a disable method when communicating to
4708 	 * the front end ports, at least for individual LUNs.
4709 	 */
4710 #if 0
4711 	STAILQ_FOREACH(port, &softc->port_list, links) {
4712 		int retval;
4713 
4714 		retval = port->lun_disable(port->targ_lun_arg, lun->target,
4715 					 lun->lun);
4716 		if (retval != 0) {
4717 			printf("ctl_free_lun: FETD %s port %d returned error "
4718 			       "%d for lun_disable on target %ju lun %jd\n",
4719 			       port->port_name, port->targ_port, retval,
4720 			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4721 		}
4722 
4723 		if (STAILQ_FIRST(&softc->lun_list) == NULL) {
4724 			port->status &= ~CTL_PORT_STATUS_LUN_ONLINE;
4725 
4726 			retval = port->targ_disable(port->targ_lun_arg,lun->target);
4727 			if (retval != 0) {
4728 				printf("ctl_free_lun: FETD %s port %d "
4729 				       "returned error %d for targ_disable on "
4730 				       "target %ju\n", port->port_name,
4731 				       port->targ_port, retval,
4732 				       (uintmax_t)lun->target.id);
4733 			} else
4734 				port->status &= ~CTL_PORT_STATUS_TARG_ONLINE;
4735 
4736 			if ((port->status & CTL_PORT_STATUS_TARG_ONLINE) != 0)
4737 				continue;
4738 
4739 #if 0
4740 			port->port_offline(port->onoff_arg);
4741 			port->status &= ~CTL_PORT_STATUS_ONLINE;
4742 #endif
4743 		}
4744 	}
4745 #endif
4746 
4747 	/*
4748 	 * Tell the backend to free resources, if this LUN has a backend.
4749 	 */
4750 	atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4751 	lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4752 
4753 	ctl_tpc_lun_shutdown(lun);
4754 	mtx_destroy(&lun->lun_lock);
4755 	free(lun->lun_devid, M_CTL);
4756 	if (lun->flags & CTL_LUN_MALLOCED)
4757 		free(lun, M_CTL);
4758 
4759 	STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4760 		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4761 			nlun->pending_ua[i] |= CTL_UA_LUN_CHANGE;
4762 		}
4763 	}
4764 
4765 	return (0);
4766 }
4767 
4768 static void
4769 ctl_create_lun(struct ctl_be_lun *be_lun)
4770 {
4771 	struct ctl_softc *ctl_softc;
4772 
4773 	ctl_softc = control_softc;
4774 
4775 	/*
4776 	 * ctl_alloc_lun() should handle all potential failure cases.
4777 	 */
4778 	ctl_alloc_lun(ctl_softc, NULL, be_lun, ctl_softc->target);
4779 }
4780 
4781 int
4782 ctl_add_lun(struct ctl_be_lun *be_lun)
4783 {
4784 	struct ctl_softc *ctl_softc = control_softc;
4785 
4786 	mtx_lock(&ctl_softc->ctl_lock);
4787 	STAILQ_INSERT_TAIL(&ctl_softc->pending_lun_queue, be_lun, links);
4788 	mtx_unlock(&ctl_softc->ctl_lock);
4789 	wakeup(&ctl_softc->pending_lun_queue);
4790 
4791 	return (0);
4792 }
4793 
4794 int
4795 ctl_enable_lun(struct ctl_be_lun *be_lun)
4796 {
4797 	struct ctl_softc *ctl_softc;
4798 	struct ctl_port *port, *nport;
4799 	struct ctl_lun *lun;
4800 	int retval;
4801 
4802 	ctl_softc = control_softc;
4803 
4804 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4805 
4806 	mtx_lock(&ctl_softc->ctl_lock);
4807 	mtx_lock(&lun->lun_lock);
4808 	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4809 		/*
4810 		 * eh?  Why did we get called if the LUN is already
4811 		 * enabled?
4812 		 */
4813 		mtx_unlock(&lun->lun_lock);
4814 		mtx_unlock(&ctl_softc->ctl_lock);
4815 		return (0);
4816 	}
4817 	lun->flags &= ~CTL_LUN_DISABLED;
4818 	mtx_unlock(&lun->lun_lock);
4819 
4820 	for (port = STAILQ_FIRST(&ctl_softc->port_list); port != NULL; port = nport) {
4821 		nport = STAILQ_NEXT(port, links);
4822 
4823 		/*
4824 		 * Drop the lock while we call the FETD's enable routine.
4825 		 * This can lead to a callback into CTL (at least in the
4826 		 * case of the internal initiator frontend.
4827 		 */
4828 		mtx_unlock(&ctl_softc->ctl_lock);
4829 		retval = port->lun_enable(port->targ_lun_arg, lun->target,lun->lun);
4830 		mtx_lock(&ctl_softc->ctl_lock);
4831 		if (retval != 0) {
4832 			printf("%s: FETD %s port %d returned error "
4833 			       "%d for lun_enable on target %ju lun %jd\n",
4834 			       __func__, port->port_name, port->targ_port, retval,
4835 			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4836 		}
4837 #if 0
4838 		 else {
4839             /* NOTE:  TODO:  why does lun enable affect port status? */
4840 			port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4841 		}
4842 #endif
4843 	}
4844 
4845 	mtx_unlock(&ctl_softc->ctl_lock);
4846 
4847 	return (0);
4848 }
4849 
4850 int
4851 ctl_disable_lun(struct ctl_be_lun *be_lun)
4852 {
4853 	struct ctl_softc *ctl_softc;
4854 	struct ctl_port *port;
4855 	struct ctl_lun *lun;
4856 	int retval;
4857 
4858 	ctl_softc = control_softc;
4859 
4860 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4861 
4862 	mtx_lock(&ctl_softc->ctl_lock);
4863 	mtx_lock(&lun->lun_lock);
4864 	if (lun->flags & CTL_LUN_DISABLED) {
4865 		mtx_unlock(&lun->lun_lock);
4866 		mtx_unlock(&ctl_softc->ctl_lock);
4867 		return (0);
4868 	}
4869 	lun->flags |= CTL_LUN_DISABLED;
4870 	mtx_unlock(&lun->lun_lock);
4871 
4872 	STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4873 		mtx_unlock(&ctl_softc->ctl_lock);
4874 		/*
4875 		 * Drop the lock before we call the frontend's disable
4876 		 * routine, to avoid lock order reversals.
4877 		 *
4878 		 * XXX KDM what happens if the frontend list changes while
4879 		 * we're traversing it?  It's unlikely, but should be handled.
4880 		 */
4881 		retval = port->lun_disable(port->targ_lun_arg, lun->target,
4882 					 lun->lun);
4883 		mtx_lock(&ctl_softc->ctl_lock);
4884 		if (retval != 0) {
4885 			printf("ctl_alloc_lun: FETD %s port %d returned error "
4886 			       "%d for lun_disable on target %ju lun %jd\n",
4887 			       port->port_name, port->targ_port, retval,
4888 			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4889 		}
4890 	}
4891 
4892 	mtx_unlock(&ctl_softc->ctl_lock);
4893 
4894 	return (0);
4895 }
4896 
4897 int
4898 ctl_start_lun(struct ctl_be_lun *be_lun)
4899 {
4900 	struct ctl_softc *ctl_softc;
4901 	struct ctl_lun *lun;
4902 
4903 	ctl_softc = control_softc;
4904 
4905 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4906 
4907 	mtx_lock(&lun->lun_lock);
4908 	lun->flags &= ~CTL_LUN_STOPPED;
4909 	mtx_unlock(&lun->lun_lock);
4910 
4911 	return (0);
4912 }
4913 
4914 int
4915 ctl_stop_lun(struct ctl_be_lun *be_lun)
4916 {
4917 	struct ctl_softc *ctl_softc;
4918 	struct ctl_lun *lun;
4919 
4920 	ctl_softc = control_softc;
4921 
4922 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4923 
4924 	mtx_lock(&lun->lun_lock);
4925 	lun->flags |= CTL_LUN_STOPPED;
4926 	mtx_unlock(&lun->lun_lock);
4927 
4928 	return (0);
4929 }
4930 
4931 int
4932 ctl_lun_offline(struct ctl_be_lun *be_lun)
4933 {
4934 	struct ctl_softc *ctl_softc;
4935 	struct ctl_lun *lun;
4936 
4937 	ctl_softc = control_softc;
4938 
4939 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4940 
4941 	mtx_lock(&lun->lun_lock);
4942 	lun->flags |= CTL_LUN_OFFLINE;
4943 	mtx_unlock(&lun->lun_lock);
4944 
4945 	return (0);
4946 }
4947 
4948 int
4949 ctl_lun_online(struct ctl_be_lun *be_lun)
4950 {
4951 	struct ctl_softc *ctl_softc;
4952 	struct ctl_lun *lun;
4953 
4954 	ctl_softc = control_softc;
4955 
4956 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4957 
4958 	mtx_lock(&lun->lun_lock);
4959 	lun->flags &= ~CTL_LUN_OFFLINE;
4960 	mtx_unlock(&lun->lun_lock);
4961 
4962 	return (0);
4963 }
4964 
4965 int
4966 ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4967 {
4968 	struct ctl_softc *ctl_softc;
4969 	struct ctl_lun *lun;
4970 
4971 	ctl_softc = control_softc;
4972 
4973 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4974 
4975 	mtx_lock(&lun->lun_lock);
4976 
4977 	/*
4978 	 * The LUN needs to be disabled before it can be marked invalid.
4979 	 */
4980 	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4981 		mtx_unlock(&lun->lun_lock);
4982 		return (-1);
4983 	}
4984 	/*
4985 	 * Mark the LUN invalid.
4986 	 */
4987 	lun->flags |= CTL_LUN_INVALID;
4988 
4989 	/*
4990 	 * If there is nothing in the OOA queue, go ahead and free the LUN.
4991 	 * If we have something in the OOA queue, we'll free it when the
4992 	 * last I/O completes.
4993 	 */
4994 	if (TAILQ_EMPTY(&lun->ooa_queue)) {
4995 		mtx_unlock(&lun->lun_lock);
4996 		mtx_lock(&ctl_softc->ctl_lock);
4997 		ctl_free_lun(lun);
4998 		mtx_unlock(&ctl_softc->ctl_lock);
4999 	} else
5000 		mtx_unlock(&lun->lun_lock);
5001 
5002 	return (0);
5003 }
5004 
5005 int
5006 ctl_lun_inoperable(struct ctl_be_lun *be_lun)
5007 {
5008 	struct ctl_softc *ctl_softc;
5009 	struct ctl_lun *lun;
5010 
5011 	ctl_softc = control_softc;
5012 	lun = (struct ctl_lun *)be_lun->ctl_lun;
5013 
5014 	mtx_lock(&lun->lun_lock);
5015 	lun->flags |= CTL_LUN_INOPERABLE;
5016 	mtx_unlock(&lun->lun_lock);
5017 
5018 	return (0);
5019 }
5020 
5021 int
5022 ctl_lun_operable(struct ctl_be_lun *be_lun)
5023 {
5024 	struct ctl_softc *ctl_softc;
5025 	struct ctl_lun *lun;
5026 
5027 	ctl_softc = control_softc;
5028 	lun = (struct ctl_lun *)be_lun->ctl_lun;
5029 
5030 	mtx_lock(&lun->lun_lock);
5031 	lun->flags &= ~CTL_LUN_INOPERABLE;
5032 	mtx_unlock(&lun->lun_lock);
5033 
5034 	return (0);
5035 }
5036 
5037 int
5038 ctl_lun_power_lock(struct ctl_be_lun *be_lun, struct ctl_nexus *nexus,
5039 		   int lock)
5040 {
5041 	struct ctl_softc *softc;
5042 	struct ctl_lun *lun;
5043 	struct copan_aps_subpage *current_sp;
5044 	struct ctl_page_index *page_index;
5045 	int i;
5046 
5047 	softc = control_softc;
5048 
5049 	mtx_lock(&softc->ctl_lock);
5050 
5051 	lun = (struct ctl_lun *)be_lun->ctl_lun;
5052 	mtx_lock(&lun->lun_lock);
5053 
5054 	page_index = NULL;
5055 	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
5056 		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
5057 		     APS_PAGE_CODE)
5058 			continue;
5059 
5060 		if (lun->mode_pages.index[i].subpage != APS_SUBPAGE_CODE)
5061 			continue;
5062 		page_index = &lun->mode_pages.index[i];
5063 	}
5064 
5065 	if (page_index == NULL) {
5066 		mtx_unlock(&lun->lun_lock);
5067 		mtx_unlock(&softc->ctl_lock);
5068 		printf("%s: APS subpage not found for lun %ju!\n", __func__,
5069 		       (uintmax_t)lun->lun);
5070 		return (1);
5071 	}
5072 #if 0
5073 	if ((softc->aps_locked_lun != 0)
5074 	 && (softc->aps_locked_lun != lun->lun)) {
5075 		printf("%s: attempt to lock LUN %llu when %llu is already "
5076 		       "locked\n");
5077 		mtx_unlock(&lun->lun_lock);
5078 		mtx_unlock(&softc->ctl_lock);
5079 		return (1);
5080 	}
5081 #endif
5082 
5083 	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
5084 		(page_index->page_len * CTL_PAGE_CURRENT));
5085 
5086 	if (lock != 0) {
5087 		current_sp->lock_active = APS_LOCK_ACTIVE;
5088 		softc->aps_locked_lun = lun->lun;
5089 	} else {
5090 		current_sp->lock_active = 0;
5091 		softc->aps_locked_lun = 0;
5092 	}
5093 
5094 
5095 	/*
5096 	 * If we're in HA mode, try to send the lock message to the other
5097 	 * side.
5098 	 */
5099 	if (ctl_is_single == 0) {
5100 		int isc_retval;
5101 		union ctl_ha_msg lock_msg;
5102 
5103 		lock_msg.hdr.nexus = *nexus;
5104 		lock_msg.hdr.msg_type = CTL_MSG_APS_LOCK;
5105 		if (lock != 0)
5106 			lock_msg.aps.lock_flag = 1;
5107 		else
5108 			lock_msg.aps.lock_flag = 0;
5109 		isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &lock_msg,
5110 					 sizeof(lock_msg), 0);
5111 		if (isc_retval > CTL_HA_STATUS_SUCCESS) {
5112 			printf("%s: APS (lock=%d) error returned from "
5113 			       "ctl_ha_msg_send: %d\n", __func__, lock, isc_retval);
5114 			mtx_unlock(&lun->lun_lock);
5115 			mtx_unlock(&softc->ctl_lock);
5116 			return (1);
5117 		}
5118 	}
5119 
5120 	mtx_unlock(&lun->lun_lock);
5121 	mtx_unlock(&softc->ctl_lock);
5122 
5123 	return (0);
5124 }
5125 
5126 void
5127 ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
5128 {
5129 	struct ctl_lun *lun;
5130 	struct ctl_softc *softc;
5131 	int i;
5132 
5133 	softc = control_softc;
5134 
5135 	lun = (struct ctl_lun *)be_lun->ctl_lun;
5136 
5137 	mtx_lock(&lun->lun_lock);
5138 
5139 	for (i = 0; i < CTL_MAX_INITIATORS; i++)
5140 		lun->pending_ua[i] |= CTL_UA_CAPACITY_CHANGED;
5141 
5142 	mtx_unlock(&lun->lun_lock);
5143 }
5144 
5145 /*
5146  * Backend "memory move is complete" callback for requests that never
5147  * make it down to say RAIDCore's configuration code.
5148  */
5149 int
5150 ctl_config_move_done(union ctl_io *io)
5151 {
5152 	int retval;
5153 
5154 	retval = CTL_RETVAL_COMPLETE;
5155 
5156 
5157 	CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
5158 	/*
5159 	 * XXX KDM this shouldn't happen, but what if it does?
5160 	 */
5161 	if (io->io_hdr.io_type != CTL_IO_SCSI)
5162 		panic("I/O type isn't CTL_IO_SCSI!");
5163 
5164 	if ((io->io_hdr.port_status == 0)
5165 	 && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
5166 	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE))
5167 		io->io_hdr.status = CTL_SUCCESS;
5168 	else if ((io->io_hdr.port_status != 0)
5169 	      && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
5170 	      && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)){
5171 		/*
5172 		 * For hardware error sense keys, the sense key
5173 		 * specific value is defined to be a retry count,
5174 		 * but we use it to pass back an internal FETD
5175 		 * error code.  XXX KDM  Hopefully the FETD is only
5176 		 * using 16 bits for an error code, since that's
5177 		 * all the space we have in the sks field.
5178 		 */
5179 		ctl_set_internal_failure(&io->scsiio,
5180 					 /*sks_valid*/ 1,
5181 					 /*retry_count*/
5182 					 io->io_hdr.port_status);
5183 		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5184 			free(io->scsiio.kern_data_ptr, M_CTL);
5185 		ctl_done(io);
5186 		goto bailout;
5187 	}
5188 
5189 	if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
5190 	 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
5191 	 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
5192 		/*
5193 		 * XXX KDM just assuming a single pointer here, and not a
5194 		 * S/G list.  If we start using S/G lists for config data,
5195 		 * we'll need to know how to clean them up here as well.
5196 		 */
5197 		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5198 			free(io->scsiio.kern_data_ptr, M_CTL);
5199 		/* Hopefully the user has already set the status... */
5200 		ctl_done(io);
5201 	} else {
5202 		/*
5203 		 * XXX KDM now we need to continue data movement.  Some
5204 		 * options:
5205 		 * - call ctl_scsiio() again?  We don't do this for data
5206 		 *   writes, because for those at least we know ahead of
5207 		 *   time where the write will go and how long it is.  For
5208 		 *   config writes, though, that information is largely
5209 		 *   contained within the write itself, thus we need to
5210 		 *   parse out the data again.
5211 		 *
5212 		 * - Call some other function once the data is in?
5213 		 */
5214 
5215 		/*
5216 		 * XXX KDM call ctl_scsiio() again for now, and check flag
5217 		 * bits to see whether we're allocated or not.
5218 		 */
5219 		retval = ctl_scsiio(&io->scsiio);
5220 	}
5221 bailout:
5222 	return (retval);
5223 }
5224 
5225 /*
5226  * This gets called by a backend driver when it is done with a
5227  * data_submit method.
5228  */
5229 void
5230 ctl_data_submit_done(union ctl_io *io)
5231 {
5232 	/*
5233 	 * If the IO_CONT flag is set, we need to call the supplied
5234 	 * function to continue processing the I/O, instead of completing
5235 	 * the I/O just yet.
5236 	 *
5237 	 * If there is an error, though, we don't want to keep processing.
5238 	 * Instead, just send status back to the initiator.
5239 	 */
5240 	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5241 	    (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5242 	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5243 	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5244 		io->scsiio.io_cont(io);
5245 		return;
5246 	}
5247 	ctl_done(io);
5248 }
5249 
5250 /*
5251  * This gets called by a backend driver when it is done with a
5252  * configuration write.
5253  */
5254 void
5255 ctl_config_write_done(union ctl_io *io)
5256 {
5257 	/*
5258 	 * If the IO_CONT flag is set, we need to call the supplied
5259 	 * function to continue processing the I/O, instead of completing
5260 	 * the I/O just yet.
5261 	 *
5262 	 * If there is an error, though, we don't want to keep processing.
5263 	 * Instead, just send status back to the initiator.
5264 	 */
5265 	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT)
5266 	 && (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)
5267 	  || ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))) {
5268 		io->scsiio.io_cont(io);
5269 		return;
5270 	}
5271 	/*
5272 	 * Since a configuration write can be done for commands that actually
5273 	 * have data allocated, like write buffer, and commands that have
5274 	 * no data, like start/stop unit, we need to check here.
5275 	 */
5276 	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT)
5277 		free(io->scsiio.kern_data_ptr, M_CTL);
5278 	ctl_done(io);
5279 }
5280 
5281 /*
5282  * SCSI release command.
5283  */
5284 int
5285 ctl_scsi_release(struct ctl_scsiio *ctsio)
5286 {
5287 	int length, longid, thirdparty_id, resv_id;
5288 	struct ctl_softc *ctl_softc;
5289 	struct ctl_lun *lun;
5290 
5291 	length = 0;
5292 	resv_id = 0;
5293 
5294 	CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5295 
5296 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5297 	ctl_softc = control_softc;
5298 
5299 	switch (ctsio->cdb[0]) {
5300 	case RELEASE_10: {
5301 		struct scsi_release_10 *cdb;
5302 
5303 		cdb = (struct scsi_release_10 *)ctsio->cdb;
5304 
5305 		if (cdb->byte2 & SR10_LONGID)
5306 			longid = 1;
5307 		else
5308 			thirdparty_id = cdb->thirdparty_id;
5309 
5310 		resv_id = cdb->resv_id;
5311 		length = scsi_2btoul(cdb->length);
5312 		break;
5313 	}
5314 	}
5315 
5316 
5317 	/*
5318 	 * XXX KDM right now, we only support LUN reservation.  We don't
5319 	 * support 3rd party reservations, or extent reservations, which
5320 	 * might actually need the parameter list.  If we've gotten this
5321 	 * far, we've got a LUN reservation.  Anything else got kicked out
5322 	 * above.  So, according to SPC, ignore the length.
5323 	 */
5324 	length = 0;
5325 
5326 	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5327 	 && (length > 0)) {
5328 		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5329 		ctsio->kern_data_len = length;
5330 		ctsio->kern_total_len = length;
5331 		ctsio->kern_data_resid = 0;
5332 		ctsio->kern_rel_offset = 0;
5333 		ctsio->kern_sg_entries = 0;
5334 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5335 		ctsio->be_move_done = ctl_config_move_done;
5336 		ctl_datamove((union ctl_io *)ctsio);
5337 
5338 		return (CTL_RETVAL_COMPLETE);
5339 	}
5340 
5341 	if (length > 0)
5342 		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5343 
5344 	mtx_lock(&lun->lun_lock);
5345 
5346 	/*
5347 	 * According to SPC, it is not an error for an intiator to attempt
5348 	 * to release a reservation on a LUN that isn't reserved, or that
5349 	 * is reserved by another initiator.  The reservation can only be
5350 	 * released, though, by the initiator who made it or by one of
5351 	 * several reset type events.
5352 	 */
5353 	if (lun->flags & CTL_LUN_RESERVED) {
5354 		if ((ctsio->io_hdr.nexus.initid.id == lun->rsv_nexus.initid.id)
5355 		 && (ctsio->io_hdr.nexus.targ_port == lun->rsv_nexus.targ_port)
5356 		 && (ctsio->io_hdr.nexus.targ_target.id ==
5357 		     lun->rsv_nexus.targ_target.id)) {
5358 			lun->flags &= ~CTL_LUN_RESERVED;
5359 		}
5360 	}
5361 
5362 	mtx_unlock(&lun->lun_lock);
5363 
5364 	ctsio->scsi_status = SCSI_STATUS_OK;
5365 	ctsio->io_hdr.status = CTL_SUCCESS;
5366 
5367 	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5368 		free(ctsio->kern_data_ptr, M_CTL);
5369 		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5370 	}
5371 
5372 	ctl_done((union ctl_io *)ctsio);
5373 	return (CTL_RETVAL_COMPLETE);
5374 }
5375 
5376 int
5377 ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5378 {
5379 	int extent, thirdparty, longid;
5380 	int resv_id, length;
5381 	uint64_t thirdparty_id;
5382 	struct ctl_softc *ctl_softc;
5383 	struct ctl_lun *lun;
5384 
5385 	extent = 0;
5386 	thirdparty = 0;
5387 	longid = 0;
5388 	resv_id = 0;
5389 	length = 0;
5390 	thirdparty_id = 0;
5391 
5392 	CTL_DEBUG_PRINT(("ctl_reserve\n"));
5393 
5394 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5395 	ctl_softc = control_softc;
5396 
5397 	switch (ctsio->cdb[0]) {
5398 	case RESERVE_10: {
5399 		struct scsi_reserve_10 *cdb;
5400 
5401 		cdb = (struct scsi_reserve_10 *)ctsio->cdb;
5402 
5403 		if (cdb->byte2 & SR10_LONGID)
5404 			longid = 1;
5405 		else
5406 			thirdparty_id = cdb->thirdparty_id;
5407 
5408 		resv_id = cdb->resv_id;
5409 		length = scsi_2btoul(cdb->length);
5410 		break;
5411 	}
5412 	}
5413 
5414 	/*
5415 	 * XXX KDM right now, we only support LUN reservation.  We don't
5416 	 * support 3rd party reservations, or extent reservations, which
5417 	 * might actually need the parameter list.  If we've gotten this
5418 	 * far, we've got a LUN reservation.  Anything else got kicked out
5419 	 * above.  So, according to SPC, ignore the length.
5420 	 */
5421 	length = 0;
5422 
5423 	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5424 	 && (length > 0)) {
5425 		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5426 		ctsio->kern_data_len = length;
5427 		ctsio->kern_total_len = length;
5428 		ctsio->kern_data_resid = 0;
5429 		ctsio->kern_rel_offset = 0;
5430 		ctsio->kern_sg_entries = 0;
5431 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5432 		ctsio->be_move_done = ctl_config_move_done;
5433 		ctl_datamove((union ctl_io *)ctsio);
5434 
5435 		return (CTL_RETVAL_COMPLETE);
5436 	}
5437 
5438 	if (length > 0)
5439 		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5440 
5441 	mtx_lock(&lun->lun_lock);
5442 	if (lun->flags & CTL_LUN_RESERVED) {
5443 		if ((ctsio->io_hdr.nexus.initid.id != lun->rsv_nexus.initid.id)
5444 		 || (ctsio->io_hdr.nexus.targ_port != lun->rsv_nexus.targ_port)
5445 		 || (ctsio->io_hdr.nexus.targ_target.id !=
5446 		     lun->rsv_nexus.targ_target.id)) {
5447 			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
5448 			ctsio->io_hdr.status = CTL_SCSI_ERROR;
5449 			goto bailout;
5450 		}
5451 	}
5452 
5453 	lun->flags |= CTL_LUN_RESERVED;
5454 	lun->rsv_nexus = ctsio->io_hdr.nexus;
5455 
5456 	ctsio->scsi_status = SCSI_STATUS_OK;
5457 	ctsio->io_hdr.status = CTL_SUCCESS;
5458 
5459 bailout:
5460 	mtx_unlock(&lun->lun_lock);
5461 
5462 	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5463 		free(ctsio->kern_data_ptr, M_CTL);
5464 		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5465 	}
5466 
5467 	ctl_done((union ctl_io *)ctsio);
5468 	return (CTL_RETVAL_COMPLETE);
5469 }
5470 
5471 int
5472 ctl_start_stop(struct ctl_scsiio *ctsio)
5473 {
5474 	struct scsi_start_stop_unit *cdb;
5475 	struct ctl_lun *lun;
5476 	struct ctl_softc *ctl_softc;
5477 	int retval;
5478 
5479 	CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5480 
5481 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5482 	ctl_softc = control_softc;
5483 	retval = 0;
5484 
5485 	cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5486 
5487 	/*
5488 	 * XXX KDM
5489 	 * We don't support the immediate bit on a stop unit.  In order to
5490 	 * do that, we would need to code up a way to know that a stop is
5491 	 * pending, and hold off any new commands until it completes, one
5492 	 * way or another.  Then we could accept or reject those commands
5493 	 * depending on its status.  We would almost need to do the reverse
5494 	 * of what we do below for an immediate start -- return the copy of
5495 	 * the ctl_io to the FETD with status to send to the host (and to
5496 	 * free the copy!) and then free the original I/O once the stop
5497 	 * actually completes.  That way, the OOA queue mechanism can work
5498 	 * to block commands that shouldn't proceed.  Another alternative
5499 	 * would be to put the copy in the queue in place of the original,
5500 	 * and return the original back to the caller.  That could be
5501 	 * slightly safer..
5502 	 */
5503 	if ((cdb->byte2 & SSS_IMMED)
5504 	 && ((cdb->how & SSS_START) == 0)) {
5505 		ctl_set_invalid_field(ctsio,
5506 				      /*sks_valid*/ 1,
5507 				      /*command*/ 1,
5508 				      /*field*/ 1,
5509 				      /*bit_valid*/ 1,
5510 				      /*bit*/ 0);
5511 		ctl_done((union ctl_io *)ctsio);
5512 		return (CTL_RETVAL_COMPLETE);
5513 	}
5514 
5515 	if ((lun->flags & CTL_LUN_PR_RESERVED)
5516 	 && ((cdb->how & SSS_START)==0)) {
5517 		uint32_t residx;
5518 
5519 		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5520 		if (!lun->per_res[residx].registered
5521 		 || (lun->pr_res_idx!=residx && lun->res_type < 4)) {
5522 
5523 			ctl_set_reservation_conflict(ctsio);
5524 			ctl_done((union ctl_io *)ctsio);
5525 			return (CTL_RETVAL_COMPLETE);
5526 		}
5527 	}
5528 
5529 	/*
5530 	 * If there is no backend on this device, we can't start or stop
5531 	 * it.  In theory we shouldn't get any start/stop commands in the
5532 	 * first place at this level if the LUN doesn't have a backend.
5533 	 * That should get stopped by the command decode code.
5534 	 */
5535 	if (lun->backend == NULL) {
5536 		ctl_set_invalid_opcode(ctsio);
5537 		ctl_done((union ctl_io *)ctsio);
5538 		return (CTL_RETVAL_COMPLETE);
5539 	}
5540 
5541 	/*
5542 	 * XXX KDM Copan-specific offline behavior.
5543 	 * Figure out a reasonable way to port this?
5544 	 */
5545 #ifdef NEEDTOPORT
5546 	mtx_lock(&lun->lun_lock);
5547 
5548 	if (((cdb->byte2 & SSS_ONOFFLINE) == 0)
5549 	 && (lun->flags & CTL_LUN_OFFLINE)) {
5550 		/*
5551 		 * If the LUN is offline, and the on/offline bit isn't set,
5552 		 * reject the start or stop.  Otherwise, let it through.
5553 		 */
5554 		mtx_unlock(&lun->lun_lock);
5555 		ctl_set_lun_not_ready(ctsio);
5556 		ctl_done((union ctl_io *)ctsio);
5557 	} else {
5558 		mtx_unlock(&lun->lun_lock);
5559 #endif /* NEEDTOPORT */
5560 		/*
5561 		 * This could be a start or a stop when we're online,
5562 		 * or a stop/offline or start/online.  A start or stop when
5563 		 * we're offline is covered in the case above.
5564 		 */
5565 		/*
5566 		 * In the non-immediate case, we send the request to
5567 		 * the backend and return status to the user when
5568 		 * it is done.
5569 		 *
5570 		 * In the immediate case, we allocate a new ctl_io
5571 		 * to hold a copy of the request, and send that to
5572 		 * the backend.  We then set good status on the
5573 		 * user's request and return it immediately.
5574 		 */
5575 		if (cdb->byte2 & SSS_IMMED) {
5576 			union ctl_io *new_io;
5577 
5578 			new_io = ctl_alloc_io(ctsio->io_hdr.pool);
5579 			if (new_io == NULL) {
5580 				ctl_set_busy(ctsio);
5581 				ctl_done((union ctl_io *)ctsio);
5582 			} else {
5583 				ctl_copy_io((union ctl_io *)ctsio,
5584 					    new_io);
5585 				retval = lun->backend->config_write(new_io);
5586 				ctl_set_success(ctsio);
5587 				ctl_done((union ctl_io *)ctsio);
5588 			}
5589 		} else {
5590 			retval = lun->backend->config_write(
5591 				(union ctl_io *)ctsio);
5592 		}
5593 #ifdef NEEDTOPORT
5594 	}
5595 #endif
5596 	return (retval);
5597 }
5598 
5599 /*
5600  * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5601  * we don't really do anything with the LBA and length fields if the user
5602  * passes them in.  Instead we'll just flush out the cache for the entire
5603  * LUN.
5604  */
5605 int
5606 ctl_sync_cache(struct ctl_scsiio *ctsio)
5607 {
5608 	struct ctl_lun *lun;
5609 	struct ctl_softc *ctl_softc;
5610 	uint64_t starting_lba;
5611 	uint32_t block_count;
5612 	int retval;
5613 
5614 	CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5615 
5616 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5617 	ctl_softc = control_softc;
5618 	retval = 0;
5619 
5620 	switch (ctsio->cdb[0]) {
5621 	case SYNCHRONIZE_CACHE: {
5622 		struct scsi_sync_cache *cdb;
5623 		cdb = (struct scsi_sync_cache *)ctsio->cdb;
5624 
5625 		starting_lba = scsi_4btoul(cdb->begin_lba);
5626 		block_count = scsi_2btoul(cdb->lb_count);
5627 		break;
5628 	}
5629 	case SYNCHRONIZE_CACHE_16: {
5630 		struct scsi_sync_cache_16 *cdb;
5631 		cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5632 
5633 		starting_lba = scsi_8btou64(cdb->begin_lba);
5634 		block_count = scsi_4btoul(cdb->lb_count);
5635 		break;
5636 	}
5637 	default:
5638 		ctl_set_invalid_opcode(ctsio);
5639 		ctl_done((union ctl_io *)ctsio);
5640 		goto bailout;
5641 		break; /* NOTREACHED */
5642 	}
5643 
5644 	/*
5645 	 * We check the LBA and length, but don't do anything with them.
5646 	 * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5647 	 * get flushed.  This check will just help satisfy anyone who wants
5648 	 * to see an error for an out of range LBA.
5649 	 */
5650 	if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5651 		ctl_set_lba_out_of_range(ctsio);
5652 		ctl_done((union ctl_io *)ctsio);
5653 		goto bailout;
5654 	}
5655 
5656 	/*
5657 	 * If this LUN has no backend, we can't flush the cache anyway.
5658 	 */
5659 	if (lun->backend == NULL) {
5660 		ctl_set_invalid_opcode(ctsio);
5661 		ctl_done((union ctl_io *)ctsio);
5662 		goto bailout;
5663 	}
5664 
5665 	/*
5666 	 * Check to see whether we're configured to send the SYNCHRONIZE
5667 	 * CACHE command directly to the back end.
5668 	 */
5669 	mtx_lock(&lun->lun_lock);
5670 	if ((ctl_softc->flags & CTL_FLAG_REAL_SYNC)
5671 	 && (++(lun->sync_count) >= lun->sync_interval)) {
5672 		lun->sync_count = 0;
5673 		mtx_unlock(&lun->lun_lock);
5674 		retval = lun->backend->config_write((union ctl_io *)ctsio);
5675 	} else {
5676 		mtx_unlock(&lun->lun_lock);
5677 		ctl_set_success(ctsio);
5678 		ctl_done((union ctl_io *)ctsio);
5679 	}
5680 
5681 bailout:
5682 
5683 	return (retval);
5684 }
5685 
5686 int
5687 ctl_format(struct ctl_scsiio *ctsio)
5688 {
5689 	struct scsi_format *cdb;
5690 	struct ctl_lun *lun;
5691 	struct ctl_softc *ctl_softc;
5692 	int length, defect_list_len;
5693 
5694 	CTL_DEBUG_PRINT(("ctl_format\n"));
5695 
5696 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5697 	ctl_softc = control_softc;
5698 
5699 	cdb = (struct scsi_format *)ctsio->cdb;
5700 
5701 	length = 0;
5702 	if (cdb->byte2 & SF_FMTDATA) {
5703 		if (cdb->byte2 & SF_LONGLIST)
5704 			length = sizeof(struct scsi_format_header_long);
5705 		else
5706 			length = sizeof(struct scsi_format_header_short);
5707 	}
5708 
5709 	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5710 	 && (length > 0)) {
5711 		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5712 		ctsio->kern_data_len = length;
5713 		ctsio->kern_total_len = length;
5714 		ctsio->kern_data_resid = 0;
5715 		ctsio->kern_rel_offset = 0;
5716 		ctsio->kern_sg_entries = 0;
5717 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5718 		ctsio->be_move_done = ctl_config_move_done;
5719 		ctl_datamove((union ctl_io *)ctsio);
5720 
5721 		return (CTL_RETVAL_COMPLETE);
5722 	}
5723 
5724 	defect_list_len = 0;
5725 
5726 	if (cdb->byte2 & SF_FMTDATA) {
5727 		if (cdb->byte2 & SF_LONGLIST) {
5728 			struct scsi_format_header_long *header;
5729 
5730 			header = (struct scsi_format_header_long *)
5731 				ctsio->kern_data_ptr;
5732 
5733 			defect_list_len = scsi_4btoul(header->defect_list_len);
5734 			if (defect_list_len != 0) {
5735 				ctl_set_invalid_field(ctsio,
5736 						      /*sks_valid*/ 1,
5737 						      /*command*/ 0,
5738 						      /*field*/ 2,
5739 						      /*bit_valid*/ 0,
5740 						      /*bit*/ 0);
5741 				goto bailout;
5742 			}
5743 		} else {
5744 			struct scsi_format_header_short *header;
5745 
5746 			header = (struct scsi_format_header_short *)
5747 				ctsio->kern_data_ptr;
5748 
5749 			defect_list_len = scsi_2btoul(header->defect_list_len);
5750 			if (defect_list_len != 0) {
5751 				ctl_set_invalid_field(ctsio,
5752 						      /*sks_valid*/ 1,
5753 						      /*command*/ 0,
5754 						      /*field*/ 2,
5755 						      /*bit_valid*/ 0,
5756 						      /*bit*/ 0);
5757 				goto bailout;
5758 			}
5759 		}
5760 	}
5761 
5762 	/*
5763 	 * The format command will clear out the "Medium format corrupted"
5764 	 * status if set by the configuration code.  That status is really
5765 	 * just a way to notify the host that we have lost the media, and
5766 	 * get them to issue a command that will basically make them think
5767 	 * they're blowing away the media.
5768 	 */
5769 	mtx_lock(&lun->lun_lock);
5770 	lun->flags &= ~CTL_LUN_INOPERABLE;
5771 	mtx_unlock(&lun->lun_lock);
5772 
5773 	ctsio->scsi_status = SCSI_STATUS_OK;
5774 	ctsio->io_hdr.status = CTL_SUCCESS;
5775 bailout:
5776 
5777 	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5778 		free(ctsio->kern_data_ptr, M_CTL);
5779 		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5780 	}
5781 
5782 	ctl_done((union ctl_io *)ctsio);
5783 	return (CTL_RETVAL_COMPLETE);
5784 }
5785 
5786 int
5787 ctl_read_buffer(struct ctl_scsiio *ctsio)
5788 {
5789 	struct scsi_read_buffer *cdb;
5790 	struct ctl_lun *lun;
5791 	int buffer_offset, len;
5792 	static uint8_t descr[4];
5793 	static uint8_t echo_descr[4] = { 0 };
5794 
5795 	CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5796 
5797 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5798 	cdb = (struct scsi_read_buffer *)ctsio->cdb;
5799 
5800 	if (lun->flags & CTL_LUN_PR_RESERVED) {
5801 		uint32_t residx;
5802 
5803 		/*
5804 		 * XXX KDM need a lock here.
5805 		 */
5806 		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5807 		if ((lun->res_type == SPR_TYPE_EX_AC
5808 		  && residx != lun->pr_res_idx)
5809 		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
5810 		   || lun->res_type == SPR_TYPE_EX_AC_AR)
5811 		  && !lun->per_res[residx].registered)) {
5812 			ctl_set_reservation_conflict(ctsio);
5813 			ctl_done((union ctl_io *)ctsio);
5814 			return (CTL_RETVAL_COMPLETE);
5815 	        }
5816 	}
5817 
5818 	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA &&
5819 	    (cdb->byte2 & RWB_MODE) != RWB_MODE_ECHO_DESCR &&
5820 	    (cdb->byte2 & RWB_MODE) != RWB_MODE_DESCR) {
5821 		ctl_set_invalid_field(ctsio,
5822 				      /*sks_valid*/ 1,
5823 				      /*command*/ 1,
5824 				      /*field*/ 1,
5825 				      /*bit_valid*/ 1,
5826 				      /*bit*/ 4);
5827 		ctl_done((union ctl_io *)ctsio);
5828 		return (CTL_RETVAL_COMPLETE);
5829 	}
5830 
5831 	len = scsi_3btoul(cdb->length);
5832 	buffer_offset = scsi_3btoul(cdb->offset);
5833 
5834 	if (buffer_offset + len > sizeof(lun->write_buffer)) {
5835 		ctl_set_invalid_field(ctsio,
5836 				      /*sks_valid*/ 1,
5837 				      /*command*/ 1,
5838 				      /*field*/ 6,
5839 				      /*bit_valid*/ 0,
5840 				      /*bit*/ 0);
5841 		ctl_done((union ctl_io *)ctsio);
5842 		return (CTL_RETVAL_COMPLETE);
5843 	}
5844 
5845 	if ((cdb->byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5846 		descr[0] = 0;
5847 		scsi_ulto3b(sizeof(lun->write_buffer), &descr[1]);
5848 		ctsio->kern_data_ptr = descr;
5849 		len = min(len, sizeof(descr));
5850 	} else if ((cdb->byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5851 		ctsio->kern_data_ptr = echo_descr;
5852 		len = min(len, sizeof(echo_descr));
5853 	} else
5854 		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5855 	ctsio->kern_data_len = len;
5856 	ctsio->kern_total_len = len;
5857 	ctsio->kern_data_resid = 0;
5858 	ctsio->kern_rel_offset = 0;
5859 	ctsio->kern_sg_entries = 0;
5860 	ctsio->be_move_done = ctl_config_move_done;
5861 	ctl_datamove((union ctl_io *)ctsio);
5862 
5863 	return (CTL_RETVAL_COMPLETE);
5864 }
5865 
5866 int
5867 ctl_write_buffer(struct ctl_scsiio *ctsio)
5868 {
5869 	struct scsi_write_buffer *cdb;
5870 	struct ctl_lun *lun;
5871 	int buffer_offset, len;
5872 
5873 	CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5874 
5875 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5876 	cdb = (struct scsi_write_buffer *)ctsio->cdb;
5877 
5878 	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5879 		ctl_set_invalid_field(ctsio,
5880 				      /*sks_valid*/ 1,
5881 				      /*command*/ 1,
5882 				      /*field*/ 1,
5883 				      /*bit_valid*/ 1,
5884 				      /*bit*/ 4);
5885 		ctl_done((union ctl_io *)ctsio);
5886 		return (CTL_RETVAL_COMPLETE);
5887 	}
5888 
5889 	len = scsi_3btoul(cdb->length);
5890 	buffer_offset = scsi_3btoul(cdb->offset);
5891 
5892 	if (buffer_offset + len > sizeof(lun->write_buffer)) {
5893 		ctl_set_invalid_field(ctsio,
5894 				      /*sks_valid*/ 1,
5895 				      /*command*/ 1,
5896 				      /*field*/ 6,
5897 				      /*bit_valid*/ 0,
5898 				      /*bit*/ 0);
5899 		ctl_done((union ctl_io *)ctsio);
5900 		return (CTL_RETVAL_COMPLETE);
5901 	}
5902 
5903 	/*
5904 	 * If we've got a kernel request that hasn't been malloced yet,
5905 	 * malloc it and tell the caller the data buffer is here.
5906 	 */
5907 	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5908 		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5909 		ctsio->kern_data_len = len;
5910 		ctsio->kern_total_len = len;
5911 		ctsio->kern_data_resid = 0;
5912 		ctsio->kern_rel_offset = 0;
5913 		ctsio->kern_sg_entries = 0;
5914 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5915 		ctsio->be_move_done = ctl_config_move_done;
5916 		ctl_datamove((union ctl_io *)ctsio);
5917 
5918 		return (CTL_RETVAL_COMPLETE);
5919 	}
5920 
5921 	ctl_done((union ctl_io *)ctsio);
5922 
5923 	return (CTL_RETVAL_COMPLETE);
5924 }
5925 
5926 int
5927 ctl_write_same(struct ctl_scsiio *ctsio)
5928 {
5929 	struct ctl_lun *lun;
5930 	struct ctl_lba_len_flags *lbalen;
5931 	uint64_t lba;
5932 	uint32_t num_blocks;
5933 	int len, retval;
5934 	uint8_t byte2;
5935 
5936 	retval = CTL_RETVAL_COMPLETE;
5937 
5938 	CTL_DEBUG_PRINT(("ctl_write_same\n"));
5939 
5940 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5941 
5942 	switch (ctsio->cdb[0]) {
5943 	case WRITE_SAME_10: {
5944 		struct scsi_write_same_10 *cdb;
5945 
5946 		cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5947 
5948 		lba = scsi_4btoul(cdb->addr);
5949 		num_blocks = scsi_2btoul(cdb->length);
5950 		byte2 = cdb->byte2;
5951 		break;
5952 	}
5953 	case WRITE_SAME_16: {
5954 		struct scsi_write_same_16 *cdb;
5955 
5956 		cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5957 
5958 		lba = scsi_8btou64(cdb->addr);
5959 		num_blocks = scsi_4btoul(cdb->length);
5960 		byte2 = cdb->byte2;
5961 		break;
5962 	}
5963 	default:
5964 		/*
5965 		 * We got a command we don't support.  This shouldn't
5966 		 * happen, commands should be filtered out above us.
5967 		 */
5968 		ctl_set_invalid_opcode(ctsio);
5969 		ctl_done((union ctl_io *)ctsio);
5970 
5971 		return (CTL_RETVAL_COMPLETE);
5972 		break; /* NOTREACHED */
5973 	}
5974 
5975 	/*
5976 	 * The first check is to make sure we're in bounds, the second
5977 	 * check is to catch wrap-around problems.  If the lba + num blocks
5978 	 * is less than the lba, then we've wrapped around and the block
5979 	 * range is invalid anyway.
5980 	 */
5981 	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5982 	 || ((lba + num_blocks) < lba)) {
5983 		ctl_set_lba_out_of_range(ctsio);
5984 		ctl_done((union ctl_io *)ctsio);
5985 		return (CTL_RETVAL_COMPLETE);
5986 	}
5987 
5988 	/* Zero number of blocks means "to the last logical block" */
5989 	if (num_blocks == 0) {
5990 		if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5991 			ctl_set_invalid_field(ctsio,
5992 					      /*sks_valid*/ 0,
5993 					      /*command*/ 1,
5994 					      /*field*/ 0,
5995 					      /*bit_valid*/ 0,
5996 					      /*bit*/ 0);
5997 			ctl_done((union ctl_io *)ctsio);
5998 			return (CTL_RETVAL_COMPLETE);
5999 		}
6000 		num_blocks = (lun->be_lun->maxlba + 1) - lba;
6001 	}
6002 
6003 	len = lun->be_lun->blocksize;
6004 
6005 	/*
6006 	 * If we've got a kernel request that hasn't been malloced yet,
6007 	 * malloc it and tell the caller the data buffer is here.
6008 	 */
6009 	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6010 		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
6011 		ctsio->kern_data_len = len;
6012 		ctsio->kern_total_len = len;
6013 		ctsio->kern_data_resid = 0;
6014 		ctsio->kern_rel_offset = 0;
6015 		ctsio->kern_sg_entries = 0;
6016 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6017 		ctsio->be_move_done = ctl_config_move_done;
6018 		ctl_datamove((union ctl_io *)ctsio);
6019 
6020 		return (CTL_RETVAL_COMPLETE);
6021 	}
6022 
6023 	lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
6024 	lbalen->lba = lba;
6025 	lbalen->len = num_blocks;
6026 	lbalen->flags = byte2;
6027 	retval = lun->backend->config_write((union ctl_io *)ctsio);
6028 
6029 	return (retval);
6030 }
6031 
6032 int
6033 ctl_unmap(struct ctl_scsiio *ctsio)
6034 {
6035 	struct ctl_lun *lun;
6036 	struct scsi_unmap *cdb;
6037 	struct ctl_ptr_len_flags *ptrlen;
6038 	struct scsi_unmap_header *hdr;
6039 	struct scsi_unmap_desc *buf, *end;
6040 	uint64_t lba;
6041 	uint32_t num_blocks;
6042 	int len, retval;
6043 	uint8_t byte2;
6044 
6045 	retval = CTL_RETVAL_COMPLETE;
6046 
6047 	CTL_DEBUG_PRINT(("ctl_unmap\n"));
6048 
6049 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6050 	cdb = (struct scsi_unmap *)ctsio->cdb;
6051 
6052 	len = scsi_2btoul(cdb->length);
6053 	byte2 = cdb->byte2;
6054 
6055 	/*
6056 	 * If we've got a kernel request that hasn't been malloced yet,
6057 	 * malloc it and tell the caller the data buffer is here.
6058 	 */
6059 	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6060 		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
6061 		ctsio->kern_data_len = len;
6062 		ctsio->kern_total_len = len;
6063 		ctsio->kern_data_resid = 0;
6064 		ctsio->kern_rel_offset = 0;
6065 		ctsio->kern_sg_entries = 0;
6066 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6067 		ctsio->be_move_done = ctl_config_move_done;
6068 		ctl_datamove((union ctl_io *)ctsio);
6069 
6070 		return (CTL_RETVAL_COMPLETE);
6071 	}
6072 
6073 	len = ctsio->kern_total_len - ctsio->kern_data_resid;
6074 	hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
6075 	if (len < sizeof (*hdr) ||
6076 	    len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
6077 	    len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
6078 	    scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
6079 		ctl_set_invalid_field(ctsio,
6080 				      /*sks_valid*/ 0,
6081 				      /*command*/ 0,
6082 				      /*field*/ 0,
6083 				      /*bit_valid*/ 0,
6084 				      /*bit*/ 0);
6085 		ctl_done((union ctl_io *)ctsio);
6086 		return (CTL_RETVAL_COMPLETE);
6087 	}
6088 	len = scsi_2btoul(hdr->desc_length);
6089 	buf = (struct scsi_unmap_desc *)(hdr + 1);
6090 	end = buf + len / sizeof(*buf);
6091 
6092 	ptrlen = (struct ctl_ptr_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
6093 	ptrlen->ptr = (void *)buf;
6094 	ptrlen->len = len;
6095 	ptrlen->flags = byte2;
6096 
6097 	for (; buf < end; buf++) {
6098 		lba = scsi_8btou64(buf->lba);
6099 		num_blocks = scsi_4btoul(buf->length);
6100 		if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
6101 		 || ((lba + num_blocks) < lba)) {
6102 			ctl_set_lba_out_of_range(ctsio);
6103 			ctl_done((union ctl_io *)ctsio);
6104 			return (CTL_RETVAL_COMPLETE);
6105 		}
6106 	}
6107 
6108 	retval = lun->backend->config_write((union ctl_io *)ctsio);
6109 
6110 	return (retval);
6111 }
6112 
6113 /*
6114  * Note that this function currently doesn't actually do anything inside
6115  * CTL to enforce things if the DQue bit is turned on.
6116  *
6117  * Also note that this function can't be used in the default case, because
6118  * the DQue bit isn't set in the changeable mask for the control mode page
6119  * anyway.  This is just here as an example for how to implement a page
6120  * handler, and a placeholder in case we want to allow the user to turn
6121  * tagged queueing on and off.
6122  *
6123  * The D_SENSE bit handling is functional, however, and will turn
6124  * descriptor sense on and off for a given LUN.
6125  */
6126 int
6127 ctl_control_page_handler(struct ctl_scsiio *ctsio,
6128 			 struct ctl_page_index *page_index, uint8_t *page_ptr)
6129 {
6130 	struct scsi_control_page *current_cp, *saved_cp, *user_cp;
6131 	struct ctl_lun *lun;
6132 	struct ctl_softc *softc;
6133 	int set_ua;
6134 	uint32_t initidx;
6135 
6136 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6137 	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6138 	set_ua = 0;
6139 
6140 	user_cp = (struct scsi_control_page *)page_ptr;
6141 	current_cp = (struct scsi_control_page *)
6142 		(page_index->page_data + (page_index->page_len *
6143 		CTL_PAGE_CURRENT));
6144 	saved_cp = (struct scsi_control_page *)
6145 		(page_index->page_data + (page_index->page_len *
6146 		CTL_PAGE_SAVED));
6147 
6148 	softc = control_softc;
6149 
6150 	mtx_lock(&lun->lun_lock);
6151 	if (((current_cp->rlec & SCP_DSENSE) == 0)
6152 	 && ((user_cp->rlec & SCP_DSENSE) != 0)) {
6153 		/*
6154 		 * Descriptor sense is currently turned off and the user
6155 		 * wants to turn it on.
6156 		 */
6157 		current_cp->rlec |= SCP_DSENSE;
6158 		saved_cp->rlec |= SCP_DSENSE;
6159 		lun->flags |= CTL_LUN_SENSE_DESC;
6160 		set_ua = 1;
6161 	} else if (((current_cp->rlec & SCP_DSENSE) != 0)
6162 		&& ((user_cp->rlec & SCP_DSENSE) == 0)) {
6163 		/*
6164 		 * Descriptor sense is currently turned on, and the user
6165 		 * wants to turn it off.
6166 		 */
6167 		current_cp->rlec &= ~SCP_DSENSE;
6168 		saved_cp->rlec &= ~SCP_DSENSE;
6169 		lun->flags &= ~CTL_LUN_SENSE_DESC;
6170 		set_ua = 1;
6171 	}
6172 	if (current_cp->queue_flags & SCP_QUEUE_DQUE) {
6173 		if (user_cp->queue_flags & SCP_QUEUE_DQUE) {
6174 #ifdef NEEDTOPORT
6175 			csevent_log(CSC_CTL | CSC_SHELF_SW |
6176 				    CTL_UNTAG_TO_UNTAG,
6177 				    csevent_LogType_Trace,
6178 				    csevent_Severity_Information,
6179 				    csevent_AlertLevel_Green,
6180 				    csevent_FRU_Firmware,
6181 				    csevent_FRU_Unknown,
6182 				    "Received untagged to untagged transition");
6183 #endif /* NEEDTOPORT */
6184 		} else {
6185 #ifdef NEEDTOPORT
6186 			csevent_log(CSC_CTL | CSC_SHELF_SW |
6187 				    CTL_UNTAG_TO_TAG,
6188 				    csevent_LogType_ConfigChange,
6189 				    csevent_Severity_Information,
6190 				    csevent_AlertLevel_Green,
6191 				    csevent_FRU_Firmware,
6192 				    csevent_FRU_Unknown,
6193 				    "Received untagged to tagged "
6194 				    "queueing transition");
6195 #endif /* NEEDTOPORT */
6196 
6197 			current_cp->queue_flags &= ~SCP_QUEUE_DQUE;
6198 			saved_cp->queue_flags &= ~SCP_QUEUE_DQUE;
6199 			set_ua = 1;
6200 		}
6201 	} else {
6202 		if (user_cp->queue_flags & SCP_QUEUE_DQUE) {
6203 #ifdef NEEDTOPORT
6204 			csevent_log(CSC_CTL | CSC_SHELF_SW |
6205 				    CTL_TAG_TO_UNTAG,
6206 				    csevent_LogType_ConfigChange,
6207 				    csevent_Severity_Warning,
6208 				    csevent_AlertLevel_Yellow,
6209 				    csevent_FRU_Firmware,
6210 				    csevent_FRU_Unknown,
6211 				    "Received tagged queueing to untagged "
6212 				    "transition");
6213 #endif /* NEEDTOPORT */
6214 
6215 			current_cp->queue_flags |= SCP_QUEUE_DQUE;
6216 			saved_cp->queue_flags |= SCP_QUEUE_DQUE;
6217 			set_ua = 1;
6218 		} else {
6219 #ifdef NEEDTOPORT
6220 			csevent_log(CSC_CTL | CSC_SHELF_SW |
6221 				    CTL_TAG_TO_TAG,
6222 				    csevent_LogType_Trace,
6223 				    csevent_Severity_Information,
6224 				    csevent_AlertLevel_Green,
6225 				    csevent_FRU_Firmware,
6226 				    csevent_FRU_Unknown,
6227 				    "Received tagged queueing to tagged "
6228 				    "queueing transition");
6229 #endif /* NEEDTOPORT */
6230 		}
6231 	}
6232 	if (set_ua != 0) {
6233 		int i;
6234 		/*
6235 		 * Let other initiators know that the mode
6236 		 * parameters for this LUN have changed.
6237 		 */
6238 		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
6239 			if (i == initidx)
6240 				continue;
6241 
6242 			lun->pending_ua[i] |= CTL_UA_MODE_CHANGE;
6243 		}
6244 	}
6245 	mtx_unlock(&lun->lun_lock);
6246 
6247 	return (0);
6248 }
6249 
6250 int
6251 ctl_power_sp_handler(struct ctl_scsiio *ctsio,
6252 		     struct ctl_page_index *page_index, uint8_t *page_ptr)
6253 {
6254 	return (0);
6255 }
6256 
6257 int
6258 ctl_power_sp_sense_handler(struct ctl_scsiio *ctsio,
6259 			   struct ctl_page_index *page_index, int pc)
6260 {
6261 	struct copan_power_subpage *page;
6262 
6263 	page = (struct copan_power_subpage *)page_index->page_data +
6264 		(page_index->page_len * pc);
6265 
6266 	switch (pc) {
6267 	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6268 		/*
6269 		 * We don't update the changable bits for this page.
6270 		 */
6271 		break;
6272 	case SMS_PAGE_CTRL_CURRENT >> 6:
6273 	case SMS_PAGE_CTRL_DEFAULT >> 6:
6274 	case SMS_PAGE_CTRL_SAVED >> 6:
6275 #ifdef NEEDTOPORT
6276 		ctl_update_power_subpage(page);
6277 #endif
6278 		break;
6279 	default:
6280 #ifdef NEEDTOPORT
6281 		EPRINT(0, "Invalid PC %d!!", pc);
6282 #endif
6283 		break;
6284 	}
6285 	return (0);
6286 }
6287 
6288 
6289 int
6290 ctl_aps_sp_handler(struct ctl_scsiio *ctsio,
6291 		   struct ctl_page_index *page_index, uint8_t *page_ptr)
6292 {
6293 	struct copan_aps_subpage *user_sp;
6294 	struct copan_aps_subpage *current_sp;
6295 	union ctl_modepage_info *modepage_info;
6296 	struct ctl_softc *softc;
6297 	struct ctl_lun *lun;
6298 	int retval;
6299 
6300 	retval = CTL_RETVAL_COMPLETE;
6301 	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
6302 		     (page_index->page_len * CTL_PAGE_CURRENT));
6303 	softc = control_softc;
6304 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6305 
6306 	user_sp = (struct copan_aps_subpage *)page_ptr;
6307 
6308 	modepage_info = (union ctl_modepage_info *)
6309 		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6310 
6311 	modepage_info->header.page_code = page_index->page_code & SMPH_PC_MASK;
6312 	modepage_info->header.subpage = page_index->subpage;
6313 	modepage_info->aps.lock_active = user_sp->lock_active;
6314 
6315 	mtx_lock(&softc->ctl_lock);
6316 
6317 	/*
6318 	 * If there is a request to lock the LUN and another LUN is locked
6319 	 * this is an error. If the requested LUN is already locked ignore
6320 	 * the request. If no LUN is locked attempt to lock it.
6321 	 * if there is a request to unlock the LUN and the LUN is currently
6322 	 * locked attempt to unlock it. Otherwise ignore the request. i.e.
6323 	 * if another LUN is locked or no LUN is locked.
6324 	 */
6325 	if (user_sp->lock_active & APS_LOCK_ACTIVE) {
6326 		if (softc->aps_locked_lun == lun->lun) {
6327 			/*
6328 			 * This LUN is already locked, so we're done.
6329 			 */
6330 			retval = CTL_RETVAL_COMPLETE;
6331 		} else if (softc->aps_locked_lun == 0) {
6332 			/*
6333 			 * No one has the lock, pass the request to the
6334 			 * backend.
6335 			 */
6336 			retval = lun->backend->config_write(
6337 				(union ctl_io *)ctsio);
6338 		} else {
6339 			/*
6340 			 * Someone else has the lock, throw out the request.
6341 			 */
6342 			ctl_set_already_locked(ctsio);
6343 			free(ctsio->kern_data_ptr, M_CTL);
6344 			ctl_done((union ctl_io *)ctsio);
6345 
6346 			/*
6347 			 * Set the return value so that ctl_do_mode_select()
6348 			 * won't try to complete the command.  We already
6349 			 * completed it here.
6350 			 */
6351 			retval = CTL_RETVAL_ERROR;
6352 		}
6353 	} else if (softc->aps_locked_lun == lun->lun) {
6354 		/*
6355 		 * This LUN is locked, so pass the unlock request to the
6356 		 * backend.
6357 		 */
6358 		retval = lun->backend->config_write((union ctl_io *)ctsio);
6359 	}
6360 	mtx_unlock(&softc->ctl_lock);
6361 
6362 	return (retval);
6363 }
6364 
6365 int
6366 ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio,
6367 				struct ctl_page_index *page_index,
6368 				uint8_t *page_ptr)
6369 {
6370 	uint8_t *c;
6371 	int i;
6372 
6373 	c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs;
6374 	ctl_time_io_secs =
6375 		(c[0] << 8) |
6376 		(c[1] << 0) |
6377 		0;
6378 	CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs));
6379 	printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs);
6380 	printf("page data:");
6381 	for (i=0; i<8; i++)
6382 		printf(" %.2x",page_ptr[i]);
6383 	printf("\n");
6384 	return (0);
6385 }
6386 
6387 int
6388 ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio,
6389 			       struct ctl_page_index *page_index,
6390 			       int pc)
6391 {
6392 	struct copan_debugconf_subpage *page;
6393 
6394 	page = (struct copan_debugconf_subpage *)page_index->page_data +
6395 		(page_index->page_len * pc);
6396 
6397 	switch (pc) {
6398 	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6399 	case SMS_PAGE_CTRL_DEFAULT >> 6:
6400 	case SMS_PAGE_CTRL_SAVED >> 6:
6401 		/*
6402 		 * We don't update the changable or default bits for this page.
6403 		 */
6404 		break;
6405 	case SMS_PAGE_CTRL_CURRENT >> 6:
6406 		page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8;
6407 		page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0;
6408 		break;
6409 	default:
6410 #ifdef NEEDTOPORT
6411 		EPRINT(0, "Invalid PC %d!!", pc);
6412 #endif /* NEEDTOPORT */
6413 		break;
6414 	}
6415 	return (0);
6416 }
6417 
6418 
6419 static int
6420 ctl_do_mode_select(union ctl_io *io)
6421 {
6422 	struct scsi_mode_page_header *page_header;
6423 	struct ctl_page_index *page_index;
6424 	struct ctl_scsiio *ctsio;
6425 	int control_dev, page_len;
6426 	int page_len_offset, page_len_size;
6427 	union ctl_modepage_info *modepage_info;
6428 	struct ctl_lun *lun;
6429 	int *len_left, *len_used;
6430 	int retval, i;
6431 
6432 	ctsio = &io->scsiio;
6433 	page_index = NULL;
6434 	page_len = 0;
6435 	retval = CTL_RETVAL_COMPLETE;
6436 
6437 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6438 
6439 	if (lun->be_lun->lun_type != T_DIRECT)
6440 		control_dev = 1;
6441 	else
6442 		control_dev = 0;
6443 
6444 	modepage_info = (union ctl_modepage_info *)
6445 		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6446 	len_left = &modepage_info->header.len_left;
6447 	len_used = &modepage_info->header.len_used;
6448 
6449 do_next_page:
6450 
6451 	page_header = (struct scsi_mode_page_header *)
6452 		(ctsio->kern_data_ptr + *len_used);
6453 
6454 	if (*len_left == 0) {
6455 		free(ctsio->kern_data_ptr, M_CTL);
6456 		ctl_set_success(ctsio);
6457 		ctl_done((union ctl_io *)ctsio);
6458 		return (CTL_RETVAL_COMPLETE);
6459 	} else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6460 
6461 		free(ctsio->kern_data_ptr, M_CTL);
6462 		ctl_set_param_len_error(ctsio);
6463 		ctl_done((union ctl_io *)ctsio);
6464 		return (CTL_RETVAL_COMPLETE);
6465 
6466 	} else if ((page_header->page_code & SMPH_SPF)
6467 		&& (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6468 
6469 		free(ctsio->kern_data_ptr, M_CTL);
6470 		ctl_set_param_len_error(ctsio);
6471 		ctl_done((union ctl_io *)ctsio);
6472 		return (CTL_RETVAL_COMPLETE);
6473 	}
6474 
6475 
6476 	/*
6477 	 * XXX KDM should we do something with the block descriptor?
6478 	 */
6479 	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6480 
6481 		if ((control_dev != 0)
6482 		 && (lun->mode_pages.index[i].page_flags &
6483 		     CTL_PAGE_FLAG_DISK_ONLY))
6484 			continue;
6485 
6486 		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
6487 		    (page_header->page_code & SMPH_PC_MASK))
6488 			continue;
6489 
6490 		/*
6491 		 * If neither page has a subpage code, then we've got a
6492 		 * match.
6493 		 */
6494 		if (((lun->mode_pages.index[i].page_code & SMPH_SPF) == 0)
6495 		 && ((page_header->page_code & SMPH_SPF) == 0)) {
6496 			page_index = &lun->mode_pages.index[i];
6497 			page_len = page_header->page_length;
6498 			break;
6499 		}
6500 
6501 		/*
6502 		 * If both pages have subpages, then the subpage numbers
6503 		 * have to match.
6504 		 */
6505 		if ((lun->mode_pages.index[i].page_code & SMPH_SPF)
6506 		  && (page_header->page_code & SMPH_SPF)) {
6507 			struct scsi_mode_page_header_sp *sph;
6508 
6509 			sph = (struct scsi_mode_page_header_sp *)page_header;
6510 
6511 			if (lun->mode_pages.index[i].subpage ==
6512 			    sph->subpage) {
6513 				page_index = &lun->mode_pages.index[i];
6514 				page_len = scsi_2btoul(sph->page_length);
6515 				break;
6516 			}
6517 		}
6518 	}
6519 
6520 	/*
6521 	 * If we couldn't find the page, or if we don't have a mode select
6522 	 * handler for it, send back an error to the user.
6523 	 */
6524 	if ((page_index == NULL)
6525 	 || (page_index->select_handler == NULL)) {
6526 		ctl_set_invalid_field(ctsio,
6527 				      /*sks_valid*/ 1,
6528 				      /*command*/ 0,
6529 				      /*field*/ *len_used,
6530 				      /*bit_valid*/ 0,
6531 				      /*bit*/ 0);
6532 		free(ctsio->kern_data_ptr, M_CTL);
6533 		ctl_done((union ctl_io *)ctsio);
6534 		return (CTL_RETVAL_COMPLETE);
6535 	}
6536 
6537 	if (page_index->page_code & SMPH_SPF) {
6538 		page_len_offset = 2;
6539 		page_len_size = 2;
6540 	} else {
6541 		page_len_size = 1;
6542 		page_len_offset = 1;
6543 	}
6544 
6545 	/*
6546 	 * If the length the initiator gives us isn't the one we specify in
6547 	 * the mode page header, or if they didn't specify enough data in
6548 	 * the CDB to avoid truncating this page, kick out the request.
6549 	 */
6550 	if ((page_len != (page_index->page_len - page_len_offset -
6551 			  page_len_size))
6552 	 || (*len_left < page_index->page_len)) {
6553 
6554 
6555 		ctl_set_invalid_field(ctsio,
6556 				      /*sks_valid*/ 1,
6557 				      /*command*/ 0,
6558 				      /*field*/ *len_used + page_len_offset,
6559 				      /*bit_valid*/ 0,
6560 				      /*bit*/ 0);
6561 		free(ctsio->kern_data_ptr, M_CTL);
6562 		ctl_done((union ctl_io *)ctsio);
6563 		return (CTL_RETVAL_COMPLETE);
6564 	}
6565 
6566 	/*
6567 	 * Run through the mode page, checking to make sure that the bits
6568 	 * the user changed are actually legal for him to change.
6569 	 */
6570 	for (i = 0; i < page_index->page_len; i++) {
6571 		uint8_t *user_byte, *change_mask, *current_byte;
6572 		int bad_bit;
6573 		int j;
6574 
6575 		user_byte = (uint8_t *)page_header + i;
6576 		change_mask = page_index->page_data +
6577 			      (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6578 		current_byte = page_index->page_data +
6579 			       (page_index->page_len * CTL_PAGE_CURRENT) + i;
6580 
6581 		/*
6582 		 * Check to see whether the user set any bits in this byte
6583 		 * that he is not allowed to set.
6584 		 */
6585 		if ((*user_byte & ~(*change_mask)) ==
6586 		    (*current_byte & ~(*change_mask)))
6587 			continue;
6588 
6589 		/*
6590 		 * Go through bit by bit to determine which one is illegal.
6591 		 */
6592 		bad_bit = 0;
6593 		for (j = 7; j >= 0; j--) {
6594 			if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6595 			    (((1 << i) & ~(*change_mask)) & *current_byte)) {
6596 				bad_bit = i;
6597 				break;
6598 			}
6599 		}
6600 		ctl_set_invalid_field(ctsio,
6601 				      /*sks_valid*/ 1,
6602 				      /*command*/ 0,
6603 				      /*field*/ *len_used + i,
6604 				      /*bit_valid*/ 1,
6605 				      /*bit*/ bad_bit);
6606 		free(ctsio->kern_data_ptr, M_CTL);
6607 		ctl_done((union ctl_io *)ctsio);
6608 		return (CTL_RETVAL_COMPLETE);
6609 	}
6610 
6611 	/*
6612 	 * Decrement these before we call the page handler, since we may
6613 	 * end up getting called back one way or another before the handler
6614 	 * returns to this context.
6615 	 */
6616 	*len_left -= page_index->page_len;
6617 	*len_used += page_index->page_len;
6618 
6619 	retval = page_index->select_handler(ctsio, page_index,
6620 					    (uint8_t *)page_header);
6621 
6622 	/*
6623 	 * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6624 	 * wait until this queued command completes to finish processing
6625 	 * the mode page.  If it returns anything other than
6626 	 * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6627 	 * already set the sense information, freed the data pointer, and
6628 	 * completed the io for us.
6629 	 */
6630 	if (retval != CTL_RETVAL_COMPLETE)
6631 		goto bailout_no_done;
6632 
6633 	/*
6634 	 * If the initiator sent us more than one page, parse the next one.
6635 	 */
6636 	if (*len_left > 0)
6637 		goto do_next_page;
6638 
6639 	ctl_set_success(ctsio);
6640 	free(ctsio->kern_data_ptr, M_CTL);
6641 	ctl_done((union ctl_io *)ctsio);
6642 
6643 bailout_no_done:
6644 
6645 	return (CTL_RETVAL_COMPLETE);
6646 
6647 }
6648 
6649 int
6650 ctl_mode_select(struct ctl_scsiio *ctsio)
6651 {
6652 	int param_len, pf, sp;
6653 	int header_size, bd_len;
6654 	int len_left, len_used;
6655 	struct ctl_page_index *page_index;
6656 	struct ctl_lun *lun;
6657 	int control_dev, page_len;
6658 	union ctl_modepage_info *modepage_info;
6659 	int retval;
6660 
6661 	pf = 0;
6662 	sp = 0;
6663 	page_len = 0;
6664 	len_used = 0;
6665 	len_left = 0;
6666 	retval = 0;
6667 	bd_len = 0;
6668 	page_index = NULL;
6669 
6670 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6671 
6672 	if (lun->be_lun->lun_type != T_DIRECT)
6673 		control_dev = 1;
6674 	else
6675 		control_dev = 0;
6676 
6677 	switch (ctsio->cdb[0]) {
6678 	case MODE_SELECT_6: {
6679 		struct scsi_mode_select_6 *cdb;
6680 
6681 		cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6682 
6683 		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6684 		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6685 
6686 		param_len = cdb->length;
6687 		header_size = sizeof(struct scsi_mode_header_6);
6688 		break;
6689 	}
6690 	case MODE_SELECT_10: {
6691 		struct scsi_mode_select_10 *cdb;
6692 
6693 		cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6694 
6695 		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6696 		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6697 
6698 		param_len = scsi_2btoul(cdb->length);
6699 		header_size = sizeof(struct scsi_mode_header_10);
6700 		break;
6701 	}
6702 	default:
6703 		ctl_set_invalid_opcode(ctsio);
6704 		ctl_done((union ctl_io *)ctsio);
6705 		return (CTL_RETVAL_COMPLETE);
6706 		break; /* NOTREACHED */
6707 	}
6708 
6709 	/*
6710 	 * From SPC-3:
6711 	 * "A parameter list length of zero indicates that the Data-Out Buffer
6712 	 * shall be empty. This condition shall not be considered as an error."
6713 	 */
6714 	if (param_len == 0) {
6715 		ctl_set_success(ctsio);
6716 		ctl_done((union ctl_io *)ctsio);
6717 		return (CTL_RETVAL_COMPLETE);
6718 	}
6719 
6720 	/*
6721 	 * Since we'll hit this the first time through, prior to
6722 	 * allocation, we don't need to free a data buffer here.
6723 	 */
6724 	if (param_len < header_size) {
6725 		ctl_set_param_len_error(ctsio);
6726 		ctl_done((union ctl_io *)ctsio);
6727 		return (CTL_RETVAL_COMPLETE);
6728 	}
6729 
6730 	/*
6731 	 * Allocate the data buffer and grab the user's data.  In theory,
6732 	 * we shouldn't have to sanity check the parameter list length here
6733 	 * because the maximum size is 64K.  We should be able to malloc
6734 	 * that much without too many problems.
6735 	 */
6736 	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6737 		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6738 		ctsio->kern_data_len = param_len;
6739 		ctsio->kern_total_len = param_len;
6740 		ctsio->kern_data_resid = 0;
6741 		ctsio->kern_rel_offset = 0;
6742 		ctsio->kern_sg_entries = 0;
6743 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6744 		ctsio->be_move_done = ctl_config_move_done;
6745 		ctl_datamove((union ctl_io *)ctsio);
6746 
6747 		return (CTL_RETVAL_COMPLETE);
6748 	}
6749 
6750 	switch (ctsio->cdb[0]) {
6751 	case MODE_SELECT_6: {
6752 		struct scsi_mode_header_6 *mh6;
6753 
6754 		mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6755 		bd_len = mh6->blk_desc_len;
6756 		break;
6757 	}
6758 	case MODE_SELECT_10: {
6759 		struct scsi_mode_header_10 *mh10;
6760 
6761 		mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6762 		bd_len = scsi_2btoul(mh10->blk_desc_len);
6763 		break;
6764 	}
6765 	default:
6766 		panic("Invalid CDB type %#x", ctsio->cdb[0]);
6767 		break;
6768 	}
6769 
6770 	if (param_len < (header_size + bd_len)) {
6771 		free(ctsio->kern_data_ptr, M_CTL);
6772 		ctl_set_param_len_error(ctsio);
6773 		ctl_done((union ctl_io *)ctsio);
6774 		return (CTL_RETVAL_COMPLETE);
6775 	}
6776 
6777 	/*
6778 	 * Set the IO_CONT flag, so that if this I/O gets passed to
6779 	 * ctl_config_write_done(), it'll get passed back to
6780 	 * ctl_do_mode_select() for further processing, or completion if
6781 	 * we're all done.
6782 	 */
6783 	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6784 	ctsio->io_cont = ctl_do_mode_select;
6785 
6786 	modepage_info = (union ctl_modepage_info *)
6787 		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6788 
6789 	memset(modepage_info, 0, sizeof(*modepage_info));
6790 
6791 	len_left = param_len - header_size - bd_len;
6792 	len_used = header_size + bd_len;
6793 
6794 	modepage_info->header.len_left = len_left;
6795 	modepage_info->header.len_used = len_used;
6796 
6797 	return (ctl_do_mode_select((union ctl_io *)ctsio));
6798 }
6799 
6800 int
6801 ctl_mode_sense(struct ctl_scsiio *ctsio)
6802 {
6803 	struct ctl_lun *lun;
6804 	int pc, page_code, dbd, llba, subpage;
6805 	int alloc_len, page_len, header_len, total_len;
6806 	struct scsi_mode_block_descr *block_desc;
6807 	struct ctl_page_index *page_index;
6808 	int control_dev;
6809 
6810 	dbd = 0;
6811 	llba = 0;
6812 	block_desc = NULL;
6813 	page_index = NULL;
6814 
6815 	CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6816 
6817 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6818 
6819 	if (lun->be_lun->lun_type != T_DIRECT)
6820 		control_dev = 1;
6821 	else
6822 		control_dev = 0;
6823 
6824 	if (lun->flags & CTL_LUN_PR_RESERVED) {
6825 		uint32_t residx;
6826 
6827 		/*
6828 		 * XXX KDM need a lock here.
6829 		 */
6830 		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
6831 		if ((lun->res_type == SPR_TYPE_EX_AC
6832 		  && residx != lun->pr_res_idx)
6833 		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
6834 		   || lun->res_type == SPR_TYPE_EX_AC_AR)
6835 		  && !lun->per_res[residx].registered)) {
6836 			ctl_set_reservation_conflict(ctsio);
6837 			ctl_done((union ctl_io *)ctsio);
6838 			return (CTL_RETVAL_COMPLETE);
6839 		}
6840 	}
6841 
6842 	switch (ctsio->cdb[0]) {
6843 	case MODE_SENSE_6: {
6844 		struct scsi_mode_sense_6 *cdb;
6845 
6846 		cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6847 
6848 		header_len = sizeof(struct scsi_mode_hdr_6);
6849 		if (cdb->byte2 & SMS_DBD)
6850 			dbd = 1;
6851 		else
6852 			header_len += sizeof(struct scsi_mode_block_descr);
6853 
6854 		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6855 		page_code = cdb->page & SMS_PAGE_CODE;
6856 		subpage = cdb->subpage;
6857 		alloc_len = cdb->length;
6858 		break;
6859 	}
6860 	case MODE_SENSE_10: {
6861 		struct scsi_mode_sense_10 *cdb;
6862 
6863 		cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6864 
6865 		header_len = sizeof(struct scsi_mode_hdr_10);
6866 
6867 		if (cdb->byte2 & SMS_DBD)
6868 			dbd = 1;
6869 		else
6870 			header_len += sizeof(struct scsi_mode_block_descr);
6871 		if (cdb->byte2 & SMS10_LLBAA)
6872 			llba = 1;
6873 		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6874 		page_code = cdb->page & SMS_PAGE_CODE;
6875 		subpage = cdb->subpage;
6876 		alloc_len = scsi_2btoul(cdb->length);
6877 		break;
6878 	}
6879 	default:
6880 		ctl_set_invalid_opcode(ctsio);
6881 		ctl_done((union ctl_io *)ctsio);
6882 		return (CTL_RETVAL_COMPLETE);
6883 		break; /* NOTREACHED */
6884 	}
6885 
6886 	/*
6887 	 * We have to make a first pass through to calculate the size of
6888 	 * the pages that match the user's query.  Then we allocate enough
6889 	 * memory to hold it, and actually copy the data into the buffer.
6890 	 */
6891 	switch (page_code) {
6892 	case SMS_ALL_PAGES_PAGE: {
6893 		int i;
6894 
6895 		page_len = 0;
6896 
6897 		/*
6898 		 * At the moment, values other than 0 and 0xff here are
6899 		 * reserved according to SPC-3.
6900 		 */
6901 		if ((subpage != SMS_SUBPAGE_PAGE_0)
6902 		 && (subpage != SMS_SUBPAGE_ALL)) {
6903 			ctl_set_invalid_field(ctsio,
6904 					      /*sks_valid*/ 1,
6905 					      /*command*/ 1,
6906 					      /*field*/ 3,
6907 					      /*bit_valid*/ 0,
6908 					      /*bit*/ 0);
6909 			ctl_done((union ctl_io *)ctsio);
6910 			return (CTL_RETVAL_COMPLETE);
6911 		}
6912 
6913 		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6914 			if ((control_dev != 0)
6915 			 && (lun->mode_pages.index[i].page_flags &
6916 			     CTL_PAGE_FLAG_DISK_ONLY))
6917 				continue;
6918 
6919 			/*
6920 			 * We don't use this subpage if the user didn't
6921 			 * request all subpages.
6922 			 */
6923 			if ((lun->mode_pages.index[i].subpage != 0)
6924 			 && (subpage == SMS_SUBPAGE_PAGE_0))
6925 				continue;
6926 
6927 #if 0
6928 			printf("found page %#x len %d\n",
6929 			       lun->mode_pages.index[i].page_code &
6930 			       SMPH_PC_MASK,
6931 			       lun->mode_pages.index[i].page_len);
6932 #endif
6933 			page_len += lun->mode_pages.index[i].page_len;
6934 		}
6935 		break;
6936 	}
6937 	default: {
6938 		int i;
6939 
6940 		page_len = 0;
6941 
6942 		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6943 			/* Look for the right page code */
6944 			if ((lun->mode_pages.index[i].page_code &
6945 			     SMPH_PC_MASK) != page_code)
6946 				continue;
6947 
6948 			/* Look for the right subpage or the subpage wildcard*/
6949 			if ((lun->mode_pages.index[i].subpage != subpage)
6950 			 && (subpage != SMS_SUBPAGE_ALL))
6951 				continue;
6952 
6953 			/* Make sure the page is supported for this dev type */
6954 			if ((control_dev != 0)
6955 			 && (lun->mode_pages.index[i].page_flags &
6956 			     CTL_PAGE_FLAG_DISK_ONLY))
6957 				continue;
6958 
6959 #if 0
6960 			printf("found page %#x len %d\n",
6961 			       lun->mode_pages.index[i].page_code &
6962 			       SMPH_PC_MASK,
6963 			       lun->mode_pages.index[i].page_len);
6964 #endif
6965 
6966 			page_len += lun->mode_pages.index[i].page_len;
6967 		}
6968 
6969 		if (page_len == 0) {
6970 			ctl_set_invalid_field(ctsio,
6971 					      /*sks_valid*/ 1,
6972 					      /*command*/ 1,
6973 					      /*field*/ 2,
6974 					      /*bit_valid*/ 1,
6975 					      /*bit*/ 5);
6976 			ctl_done((union ctl_io *)ctsio);
6977 			return (CTL_RETVAL_COMPLETE);
6978 		}
6979 		break;
6980 	}
6981 	}
6982 
6983 	total_len = header_len + page_len;
6984 #if 0
6985 	printf("header_len = %d, page_len = %d, total_len = %d\n",
6986 	       header_len, page_len, total_len);
6987 #endif
6988 
6989 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6990 	ctsio->kern_sg_entries = 0;
6991 	ctsio->kern_data_resid = 0;
6992 	ctsio->kern_rel_offset = 0;
6993 	if (total_len < alloc_len) {
6994 		ctsio->residual = alloc_len - total_len;
6995 		ctsio->kern_data_len = total_len;
6996 		ctsio->kern_total_len = total_len;
6997 	} else {
6998 		ctsio->residual = 0;
6999 		ctsio->kern_data_len = alloc_len;
7000 		ctsio->kern_total_len = alloc_len;
7001 	}
7002 
7003 	switch (ctsio->cdb[0]) {
7004 	case MODE_SENSE_6: {
7005 		struct scsi_mode_hdr_6 *header;
7006 
7007 		header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
7008 
7009 		header->datalen = ctl_min(total_len - 1, 254);
7010 
7011 		if (dbd)
7012 			header->block_descr_len = 0;
7013 		else
7014 			header->block_descr_len =
7015 				sizeof(struct scsi_mode_block_descr);
7016 		block_desc = (struct scsi_mode_block_descr *)&header[1];
7017 		break;
7018 	}
7019 	case MODE_SENSE_10: {
7020 		struct scsi_mode_hdr_10 *header;
7021 		int datalen;
7022 
7023 		header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
7024 
7025 		datalen = ctl_min(total_len - 2, 65533);
7026 		scsi_ulto2b(datalen, header->datalen);
7027 		if (dbd)
7028 			scsi_ulto2b(0, header->block_descr_len);
7029 		else
7030 			scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
7031 				    header->block_descr_len);
7032 		block_desc = (struct scsi_mode_block_descr *)&header[1];
7033 		break;
7034 	}
7035 	default:
7036 		panic("invalid CDB type %#x", ctsio->cdb[0]);
7037 		break; /* NOTREACHED */
7038 	}
7039 
7040 	/*
7041 	 * If we've got a disk, use its blocksize in the block
7042 	 * descriptor.  Otherwise, just set it to 0.
7043 	 */
7044 	if (dbd == 0) {
7045 		if (control_dev != 0)
7046 			scsi_ulto3b(lun->be_lun->blocksize,
7047 				    block_desc->block_len);
7048 		else
7049 			scsi_ulto3b(0, block_desc->block_len);
7050 	}
7051 
7052 	switch (page_code) {
7053 	case SMS_ALL_PAGES_PAGE: {
7054 		int i, data_used;
7055 
7056 		data_used = header_len;
7057 		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
7058 			struct ctl_page_index *page_index;
7059 
7060 			page_index = &lun->mode_pages.index[i];
7061 
7062 			if ((control_dev != 0)
7063 			 && (page_index->page_flags &
7064 			    CTL_PAGE_FLAG_DISK_ONLY))
7065 				continue;
7066 
7067 			/*
7068 			 * We don't use this subpage if the user didn't
7069 			 * request all subpages.  We already checked (above)
7070 			 * to make sure the user only specified a subpage
7071 			 * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
7072 			 */
7073 			if ((page_index->subpage != 0)
7074 			 && (subpage == SMS_SUBPAGE_PAGE_0))
7075 				continue;
7076 
7077 			/*
7078 			 * Call the handler, if it exists, to update the
7079 			 * page to the latest values.
7080 			 */
7081 			if (page_index->sense_handler != NULL)
7082 				page_index->sense_handler(ctsio, page_index,pc);
7083 
7084 			memcpy(ctsio->kern_data_ptr + data_used,
7085 			       page_index->page_data +
7086 			       (page_index->page_len * pc),
7087 			       page_index->page_len);
7088 			data_used += page_index->page_len;
7089 		}
7090 		break;
7091 	}
7092 	default: {
7093 		int i, data_used;
7094 
7095 		data_used = header_len;
7096 
7097 		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
7098 			struct ctl_page_index *page_index;
7099 
7100 			page_index = &lun->mode_pages.index[i];
7101 
7102 			/* Look for the right page code */
7103 			if ((page_index->page_code & SMPH_PC_MASK) != page_code)
7104 				continue;
7105 
7106 			/* Look for the right subpage or the subpage wildcard*/
7107 			if ((page_index->subpage != subpage)
7108 			 && (subpage != SMS_SUBPAGE_ALL))
7109 				continue;
7110 
7111 			/* Make sure the page is supported for this dev type */
7112 			if ((control_dev != 0)
7113 			 && (page_index->page_flags &
7114 			     CTL_PAGE_FLAG_DISK_ONLY))
7115 				continue;
7116 
7117 			/*
7118 			 * Call the handler, if it exists, to update the
7119 			 * page to the latest values.
7120 			 */
7121 			if (page_index->sense_handler != NULL)
7122 				page_index->sense_handler(ctsio, page_index,pc);
7123 
7124 			memcpy(ctsio->kern_data_ptr + data_used,
7125 			       page_index->page_data +
7126 			       (page_index->page_len * pc),
7127 			       page_index->page_len);
7128 			data_used += page_index->page_len;
7129 		}
7130 		break;
7131 	}
7132 	}
7133 
7134 	ctsio->scsi_status = SCSI_STATUS_OK;
7135 
7136 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7137 	ctsio->be_move_done = ctl_config_move_done;
7138 	ctl_datamove((union ctl_io *)ctsio);
7139 
7140 	return (CTL_RETVAL_COMPLETE);
7141 }
7142 
7143 int
7144 ctl_read_capacity(struct ctl_scsiio *ctsio)
7145 {
7146 	struct scsi_read_capacity *cdb;
7147 	struct scsi_read_capacity_data *data;
7148 	struct ctl_lun *lun;
7149 	uint32_t lba;
7150 
7151 	CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
7152 
7153 	cdb = (struct scsi_read_capacity *)ctsio->cdb;
7154 
7155 	lba = scsi_4btoul(cdb->addr);
7156 	if (((cdb->pmi & SRC_PMI) == 0)
7157 	 && (lba != 0)) {
7158 		ctl_set_invalid_field(/*ctsio*/ ctsio,
7159 				      /*sks_valid*/ 1,
7160 				      /*command*/ 1,
7161 				      /*field*/ 2,
7162 				      /*bit_valid*/ 0,
7163 				      /*bit*/ 0);
7164 		ctl_done((union ctl_io *)ctsio);
7165 		return (CTL_RETVAL_COMPLETE);
7166 	}
7167 
7168 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7169 
7170 	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7171 	data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
7172 	ctsio->residual = 0;
7173 	ctsio->kern_data_len = sizeof(*data);
7174 	ctsio->kern_total_len = sizeof(*data);
7175 	ctsio->kern_data_resid = 0;
7176 	ctsio->kern_rel_offset = 0;
7177 	ctsio->kern_sg_entries = 0;
7178 
7179 	/*
7180 	 * If the maximum LBA is greater than 0xfffffffe, the user must
7181 	 * issue a SERVICE ACTION IN (16) command, with the read capacity
7182 	 * serivce action set.
7183 	 */
7184 	if (lun->be_lun->maxlba > 0xfffffffe)
7185 		scsi_ulto4b(0xffffffff, data->addr);
7186 	else
7187 		scsi_ulto4b(lun->be_lun->maxlba, data->addr);
7188 
7189 	/*
7190 	 * XXX KDM this may not be 512 bytes...
7191 	 */
7192 	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7193 
7194 	ctsio->scsi_status = SCSI_STATUS_OK;
7195 
7196 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7197 	ctsio->be_move_done = ctl_config_move_done;
7198 	ctl_datamove((union ctl_io *)ctsio);
7199 
7200 	return (CTL_RETVAL_COMPLETE);
7201 }
7202 
7203 int
7204 ctl_read_capacity_16(struct ctl_scsiio *ctsio)
7205 {
7206 	struct scsi_read_capacity_16 *cdb;
7207 	struct scsi_read_capacity_data_long *data;
7208 	struct ctl_lun *lun;
7209 	uint64_t lba;
7210 	uint32_t alloc_len;
7211 
7212 	CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
7213 
7214 	cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
7215 
7216 	alloc_len = scsi_4btoul(cdb->alloc_len);
7217 	lba = scsi_8btou64(cdb->addr);
7218 
7219 	if ((cdb->reladr & SRC16_PMI)
7220 	 && (lba != 0)) {
7221 		ctl_set_invalid_field(/*ctsio*/ ctsio,
7222 				      /*sks_valid*/ 1,
7223 				      /*command*/ 1,
7224 				      /*field*/ 2,
7225 				      /*bit_valid*/ 0,
7226 				      /*bit*/ 0);
7227 		ctl_done((union ctl_io *)ctsio);
7228 		return (CTL_RETVAL_COMPLETE);
7229 	}
7230 
7231 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7232 
7233 	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7234 	data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
7235 
7236 	if (sizeof(*data) < alloc_len) {
7237 		ctsio->residual = alloc_len - sizeof(*data);
7238 		ctsio->kern_data_len = sizeof(*data);
7239 		ctsio->kern_total_len = sizeof(*data);
7240 	} else {
7241 		ctsio->residual = 0;
7242 		ctsio->kern_data_len = alloc_len;
7243 		ctsio->kern_total_len = alloc_len;
7244 	}
7245 	ctsio->kern_data_resid = 0;
7246 	ctsio->kern_rel_offset = 0;
7247 	ctsio->kern_sg_entries = 0;
7248 
7249 	scsi_u64to8b(lun->be_lun->maxlba, data->addr);
7250 	/* XXX KDM this may not be 512 bytes... */
7251 	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7252 	data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
7253 	scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
7254 	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
7255 		data->lalba_lbp[0] |= SRC16_LBPME;
7256 
7257 	ctsio->scsi_status = SCSI_STATUS_OK;
7258 
7259 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7260 	ctsio->be_move_done = ctl_config_move_done;
7261 	ctl_datamove((union ctl_io *)ctsio);
7262 
7263 	return (CTL_RETVAL_COMPLETE);
7264 }
7265 
7266 int
7267 ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7268 {
7269 	struct scsi_maintenance_in *cdb;
7270 	int retval;
7271 	int alloc_len, ext, total_len = 0, g, p, pc, pg;
7272 	int num_target_port_groups, num_target_ports, single;
7273 	struct ctl_lun *lun;
7274 	struct ctl_softc *softc;
7275 	struct ctl_port *port;
7276 	struct scsi_target_group_data *rtg_ptr;
7277 	struct scsi_target_group_data_extended *rtg_ext_ptr;
7278 	struct scsi_target_port_group_descriptor *tpg_desc;
7279 
7280 	CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7281 
7282 	cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7283 	softc = control_softc;
7284 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7285 
7286 	retval = CTL_RETVAL_COMPLETE;
7287 
7288 	switch (cdb->byte2 & STG_PDF_MASK) {
7289 	case STG_PDF_LENGTH:
7290 		ext = 0;
7291 		break;
7292 	case STG_PDF_EXTENDED:
7293 		ext = 1;
7294 		break;
7295 	default:
7296 		ctl_set_invalid_field(/*ctsio*/ ctsio,
7297 				      /*sks_valid*/ 1,
7298 				      /*command*/ 1,
7299 				      /*field*/ 2,
7300 				      /*bit_valid*/ 1,
7301 				      /*bit*/ 5);
7302 		ctl_done((union ctl_io *)ctsio);
7303 		return(retval);
7304 	}
7305 
7306 	single = ctl_is_single;
7307 	if (single)
7308 		num_target_port_groups = 1;
7309 	else
7310 		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
7311 	num_target_ports = 0;
7312 	mtx_lock(&softc->ctl_lock);
7313 	STAILQ_FOREACH(port, &softc->port_list, links) {
7314 		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7315 			continue;
7316 		if (ctl_map_lun_back(port->targ_port, lun->lun) >= CTL_MAX_LUNS)
7317 			continue;
7318 		num_target_ports++;
7319 	}
7320 	mtx_unlock(&softc->ctl_lock);
7321 
7322 	if (ext)
7323 		total_len = sizeof(struct scsi_target_group_data_extended);
7324 	else
7325 		total_len = sizeof(struct scsi_target_group_data);
7326 	total_len += sizeof(struct scsi_target_port_group_descriptor) *
7327 		num_target_port_groups +
7328 	    sizeof(struct scsi_target_port_descriptor) *
7329 		num_target_ports * num_target_port_groups;
7330 
7331 	alloc_len = scsi_4btoul(cdb->length);
7332 
7333 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7334 
7335 	ctsio->kern_sg_entries = 0;
7336 
7337 	if (total_len < alloc_len) {
7338 		ctsio->residual = alloc_len - total_len;
7339 		ctsio->kern_data_len = total_len;
7340 		ctsio->kern_total_len = total_len;
7341 	} else {
7342 		ctsio->residual = 0;
7343 		ctsio->kern_data_len = alloc_len;
7344 		ctsio->kern_total_len = alloc_len;
7345 	}
7346 	ctsio->kern_data_resid = 0;
7347 	ctsio->kern_rel_offset = 0;
7348 
7349 	if (ext) {
7350 		rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7351 		    ctsio->kern_data_ptr;
7352 		scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7353 		rtg_ext_ptr->format_type = 0x10;
7354 		rtg_ext_ptr->implicit_transition_time = 0;
7355 		tpg_desc = &rtg_ext_ptr->groups[0];
7356 	} else {
7357 		rtg_ptr = (struct scsi_target_group_data *)
7358 		    ctsio->kern_data_ptr;
7359 		scsi_ulto4b(total_len - 4, rtg_ptr->length);
7360 		tpg_desc = &rtg_ptr->groups[0];
7361 	}
7362 
7363 	pg = ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS;
7364 	mtx_lock(&softc->ctl_lock);
7365 	for (g = 0; g < num_target_port_groups; g++) {
7366 		if (g == pg)
7367 			tpg_desc->pref_state = TPG_PRIMARY |
7368 			    TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7369 		else
7370 			tpg_desc->pref_state =
7371 			    TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7372 		tpg_desc->support = TPG_AO_SUP;
7373 		if (!single)
7374 			tpg_desc->support |= TPG_AN_SUP;
7375 		scsi_ulto2b(g + 1, tpg_desc->target_port_group);
7376 		tpg_desc->status = TPG_IMPLICIT;
7377 		pc = 0;
7378 		STAILQ_FOREACH(port, &softc->port_list, links) {
7379 			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7380 				continue;
7381 			if (ctl_map_lun_back(port->targ_port, lun->lun) >=
7382 			    CTL_MAX_LUNS)
7383 				continue;
7384 			p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
7385 			scsi_ulto2b(p, tpg_desc->descriptors[pc].
7386 			    relative_target_port_identifier);
7387 			pc++;
7388 		}
7389 		tpg_desc->target_port_count = pc;
7390 		tpg_desc = (struct scsi_target_port_group_descriptor *)
7391 		    &tpg_desc->descriptors[pc];
7392 	}
7393 	mtx_unlock(&softc->ctl_lock);
7394 
7395 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7396 	ctsio->be_move_done = ctl_config_move_done;
7397 
7398 	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7399 			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7400 			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7401 			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7402 			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7403 
7404 	ctl_datamove((union ctl_io *)ctsio);
7405 	return(retval);
7406 }
7407 
7408 int
7409 ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7410 {
7411 	struct ctl_lun *lun;
7412 	struct scsi_report_supported_opcodes *cdb;
7413 	const struct ctl_cmd_entry *entry, *sentry;
7414 	struct scsi_report_supported_opcodes_all *all;
7415 	struct scsi_report_supported_opcodes_descr *descr;
7416 	struct scsi_report_supported_opcodes_one *one;
7417 	int retval;
7418 	int alloc_len, total_len;
7419 	int opcode, service_action, i, j, num;
7420 
7421 	CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7422 
7423 	cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7424 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7425 
7426 	retval = CTL_RETVAL_COMPLETE;
7427 
7428 	opcode = cdb->requested_opcode;
7429 	service_action = scsi_2btoul(cdb->requested_service_action);
7430 	switch (cdb->options & RSO_OPTIONS_MASK) {
7431 	case RSO_OPTIONS_ALL:
7432 		num = 0;
7433 		for (i = 0; i < 256; i++) {
7434 			entry = &ctl_cmd_table[i];
7435 			if (entry->flags & CTL_CMD_FLAG_SA5) {
7436 				for (j = 0; j < 32; j++) {
7437 					sentry = &((const struct ctl_cmd_entry *)
7438 					    entry->execute)[j];
7439 					if (ctl_cmd_applicable(
7440 					    lun->be_lun->lun_type, sentry))
7441 						num++;
7442 				}
7443 			} else {
7444 				if (ctl_cmd_applicable(lun->be_lun->lun_type,
7445 				    entry))
7446 					num++;
7447 			}
7448 		}
7449 		total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7450 		    num * sizeof(struct scsi_report_supported_opcodes_descr);
7451 		break;
7452 	case RSO_OPTIONS_OC:
7453 		if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7454 			ctl_set_invalid_field(/*ctsio*/ ctsio,
7455 					      /*sks_valid*/ 1,
7456 					      /*command*/ 1,
7457 					      /*field*/ 2,
7458 					      /*bit_valid*/ 1,
7459 					      /*bit*/ 2);
7460 			ctl_done((union ctl_io *)ctsio);
7461 			return (CTL_RETVAL_COMPLETE);
7462 		}
7463 		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7464 		break;
7465 	case RSO_OPTIONS_OC_SA:
7466 		if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7467 		    service_action >= 32) {
7468 			ctl_set_invalid_field(/*ctsio*/ ctsio,
7469 					      /*sks_valid*/ 1,
7470 					      /*command*/ 1,
7471 					      /*field*/ 2,
7472 					      /*bit_valid*/ 1,
7473 					      /*bit*/ 2);
7474 			ctl_done((union ctl_io *)ctsio);
7475 			return (CTL_RETVAL_COMPLETE);
7476 		}
7477 		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7478 		break;
7479 	default:
7480 		ctl_set_invalid_field(/*ctsio*/ ctsio,
7481 				      /*sks_valid*/ 1,
7482 				      /*command*/ 1,
7483 				      /*field*/ 2,
7484 				      /*bit_valid*/ 1,
7485 				      /*bit*/ 2);
7486 		ctl_done((union ctl_io *)ctsio);
7487 		return (CTL_RETVAL_COMPLETE);
7488 	}
7489 
7490 	alloc_len = scsi_4btoul(cdb->length);
7491 
7492 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7493 
7494 	ctsio->kern_sg_entries = 0;
7495 
7496 	if (total_len < alloc_len) {
7497 		ctsio->residual = alloc_len - total_len;
7498 		ctsio->kern_data_len = total_len;
7499 		ctsio->kern_total_len = total_len;
7500 	} else {
7501 		ctsio->residual = 0;
7502 		ctsio->kern_data_len = alloc_len;
7503 		ctsio->kern_total_len = alloc_len;
7504 	}
7505 	ctsio->kern_data_resid = 0;
7506 	ctsio->kern_rel_offset = 0;
7507 
7508 	switch (cdb->options & RSO_OPTIONS_MASK) {
7509 	case RSO_OPTIONS_ALL:
7510 		all = (struct scsi_report_supported_opcodes_all *)
7511 		    ctsio->kern_data_ptr;
7512 		num = 0;
7513 		for (i = 0; i < 256; i++) {
7514 			entry = &ctl_cmd_table[i];
7515 			if (entry->flags & CTL_CMD_FLAG_SA5) {
7516 				for (j = 0; j < 32; j++) {
7517 					sentry = &((const struct ctl_cmd_entry *)
7518 					    entry->execute)[j];
7519 					if (!ctl_cmd_applicable(
7520 					    lun->be_lun->lun_type, sentry))
7521 						continue;
7522 					descr = &all->descr[num++];
7523 					descr->opcode = i;
7524 					scsi_ulto2b(j, descr->service_action);
7525 					descr->flags = RSO_SERVACTV;
7526 					scsi_ulto2b(sentry->length,
7527 					    descr->cdb_length);
7528 				}
7529 			} else {
7530 				if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7531 				    entry))
7532 					continue;
7533 				descr = &all->descr[num++];
7534 				descr->opcode = i;
7535 				scsi_ulto2b(0, descr->service_action);
7536 				descr->flags = 0;
7537 				scsi_ulto2b(entry->length, descr->cdb_length);
7538 			}
7539 		}
7540 		scsi_ulto4b(
7541 		    num * sizeof(struct scsi_report_supported_opcodes_descr),
7542 		    all->length);
7543 		break;
7544 	case RSO_OPTIONS_OC:
7545 		one = (struct scsi_report_supported_opcodes_one *)
7546 		    ctsio->kern_data_ptr;
7547 		entry = &ctl_cmd_table[opcode];
7548 		goto fill_one;
7549 	case RSO_OPTIONS_OC_SA:
7550 		one = (struct scsi_report_supported_opcodes_one *)
7551 		    ctsio->kern_data_ptr;
7552 		entry = &ctl_cmd_table[opcode];
7553 		entry = &((const struct ctl_cmd_entry *)
7554 		    entry->execute)[service_action];
7555 fill_one:
7556 		if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7557 			one->support = 3;
7558 			scsi_ulto2b(entry->length, one->cdb_length);
7559 			one->cdb_usage[0] = opcode;
7560 			memcpy(&one->cdb_usage[1], entry->usage,
7561 			    entry->length - 1);
7562 		} else
7563 			one->support = 1;
7564 		break;
7565 	}
7566 
7567 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7568 	ctsio->be_move_done = ctl_config_move_done;
7569 
7570 	ctl_datamove((union ctl_io *)ctsio);
7571 	return(retval);
7572 }
7573 
7574 int
7575 ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7576 {
7577 	struct ctl_lun *lun;
7578 	struct scsi_report_supported_tmf *cdb;
7579 	struct scsi_report_supported_tmf_data *data;
7580 	int retval;
7581 	int alloc_len, total_len;
7582 
7583 	CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7584 
7585 	cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7586 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7587 
7588 	retval = CTL_RETVAL_COMPLETE;
7589 
7590 	total_len = sizeof(struct scsi_report_supported_tmf_data);
7591 	alloc_len = scsi_4btoul(cdb->length);
7592 
7593 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7594 
7595 	ctsio->kern_sg_entries = 0;
7596 
7597 	if (total_len < alloc_len) {
7598 		ctsio->residual = alloc_len - total_len;
7599 		ctsio->kern_data_len = total_len;
7600 		ctsio->kern_total_len = total_len;
7601 	} else {
7602 		ctsio->residual = 0;
7603 		ctsio->kern_data_len = alloc_len;
7604 		ctsio->kern_total_len = alloc_len;
7605 	}
7606 	ctsio->kern_data_resid = 0;
7607 	ctsio->kern_rel_offset = 0;
7608 
7609 	data = (struct scsi_report_supported_tmf_data *)ctsio->kern_data_ptr;
7610 	data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_TRS;
7611 	data->byte2 |= RST_ITNRS;
7612 
7613 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7614 	ctsio->be_move_done = ctl_config_move_done;
7615 
7616 	ctl_datamove((union ctl_io *)ctsio);
7617 	return (retval);
7618 }
7619 
7620 int
7621 ctl_report_timestamp(struct ctl_scsiio *ctsio)
7622 {
7623 	struct ctl_lun *lun;
7624 	struct scsi_report_timestamp *cdb;
7625 	struct scsi_report_timestamp_data *data;
7626 	struct timeval tv;
7627 	int64_t timestamp;
7628 	int retval;
7629 	int alloc_len, total_len;
7630 
7631 	CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7632 
7633 	cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7634 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7635 
7636 	retval = CTL_RETVAL_COMPLETE;
7637 
7638 	total_len = sizeof(struct scsi_report_timestamp_data);
7639 	alloc_len = scsi_4btoul(cdb->length);
7640 
7641 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7642 
7643 	ctsio->kern_sg_entries = 0;
7644 
7645 	if (total_len < alloc_len) {
7646 		ctsio->residual = alloc_len - total_len;
7647 		ctsio->kern_data_len = total_len;
7648 		ctsio->kern_total_len = total_len;
7649 	} else {
7650 		ctsio->residual = 0;
7651 		ctsio->kern_data_len = alloc_len;
7652 		ctsio->kern_total_len = alloc_len;
7653 	}
7654 	ctsio->kern_data_resid = 0;
7655 	ctsio->kern_rel_offset = 0;
7656 
7657 	data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7658 	scsi_ulto2b(sizeof(*data) - 2, data->length);
7659 	data->origin = RTS_ORIG_OUTSIDE;
7660 	getmicrotime(&tv);
7661 	timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7662 	scsi_ulto4b(timestamp >> 16, data->timestamp);
7663 	scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7664 
7665 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7666 	ctsio->be_move_done = ctl_config_move_done;
7667 
7668 	ctl_datamove((union ctl_io *)ctsio);
7669 	return (retval);
7670 }
7671 
7672 int
7673 ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7674 {
7675 	struct scsi_per_res_in *cdb;
7676 	int alloc_len, total_len = 0;
7677 	/* struct scsi_per_res_in_rsrv in_data; */
7678 	struct ctl_lun *lun;
7679 	struct ctl_softc *softc;
7680 
7681 	CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7682 
7683 	softc = control_softc;
7684 
7685 	cdb = (struct scsi_per_res_in *)ctsio->cdb;
7686 
7687 	alloc_len = scsi_2btoul(cdb->length);
7688 
7689 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7690 
7691 retry:
7692 	mtx_lock(&lun->lun_lock);
7693 	switch (cdb->action) {
7694 	case SPRI_RK: /* read keys */
7695 		total_len = sizeof(struct scsi_per_res_in_keys) +
7696 			lun->pr_key_count *
7697 			sizeof(struct scsi_per_res_key);
7698 		break;
7699 	case SPRI_RR: /* read reservation */
7700 		if (lun->flags & CTL_LUN_PR_RESERVED)
7701 			total_len = sizeof(struct scsi_per_res_in_rsrv);
7702 		else
7703 			total_len = sizeof(struct scsi_per_res_in_header);
7704 		break;
7705 	case SPRI_RC: /* report capabilities */
7706 		total_len = sizeof(struct scsi_per_res_cap);
7707 		break;
7708 	case SPRI_RS: /* read full status */
7709 		total_len = sizeof(struct scsi_per_res_in_header) +
7710 		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7711 		    lun->pr_key_count;
7712 		break;
7713 	default:
7714 		panic("Invalid PR type %x", cdb->action);
7715 	}
7716 	mtx_unlock(&lun->lun_lock);
7717 
7718 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7719 
7720 	if (total_len < alloc_len) {
7721 		ctsio->residual = alloc_len - total_len;
7722 		ctsio->kern_data_len = total_len;
7723 		ctsio->kern_total_len = total_len;
7724 	} else {
7725 		ctsio->residual = 0;
7726 		ctsio->kern_data_len = alloc_len;
7727 		ctsio->kern_total_len = alloc_len;
7728 	}
7729 
7730 	ctsio->kern_data_resid = 0;
7731 	ctsio->kern_rel_offset = 0;
7732 	ctsio->kern_sg_entries = 0;
7733 
7734 	mtx_lock(&lun->lun_lock);
7735 	switch (cdb->action) {
7736 	case SPRI_RK: { // read keys
7737         struct scsi_per_res_in_keys *res_keys;
7738 		int i, key_count;
7739 
7740 		res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7741 
7742 		/*
7743 		 * We had to drop the lock to allocate our buffer, which
7744 		 * leaves time for someone to come in with another
7745 		 * persistent reservation.  (That is unlikely, though,
7746 		 * since this should be the only persistent reservation
7747 		 * command active right now.)
7748 		 */
7749 		if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7750 		    (lun->pr_key_count *
7751 		     sizeof(struct scsi_per_res_key)))){
7752 			mtx_unlock(&lun->lun_lock);
7753 			free(ctsio->kern_data_ptr, M_CTL);
7754 			printf("%s: reservation length changed, retrying\n",
7755 			       __func__);
7756 			goto retry;
7757 		}
7758 
7759 		scsi_ulto4b(lun->PRGeneration, res_keys->header.generation);
7760 
7761 		scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7762 			     lun->pr_key_count, res_keys->header.length);
7763 
7764 		for (i = 0, key_count = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7765 			if (!lun->per_res[i].registered)
7766 				continue;
7767 
7768 			/*
7769 			 * We used lun->pr_key_count to calculate the
7770 			 * size to allocate.  If it turns out the number of
7771 			 * initiators with the registered flag set is
7772 			 * larger than that (i.e. they haven't been kept in
7773 			 * sync), we've got a problem.
7774 			 */
7775 			if (key_count >= lun->pr_key_count) {
7776 #ifdef NEEDTOPORT
7777 				csevent_log(CSC_CTL | CSC_SHELF_SW |
7778 					    CTL_PR_ERROR,
7779 					    csevent_LogType_Fault,
7780 					    csevent_AlertLevel_Yellow,
7781 					    csevent_FRU_ShelfController,
7782 					    csevent_FRU_Firmware,
7783 				        csevent_FRU_Unknown,
7784 					    "registered keys %d >= key "
7785 					    "count %d", key_count,
7786 					    lun->pr_key_count);
7787 #endif
7788 				key_count++;
7789 				continue;
7790 			}
7791 			memcpy(res_keys->keys[key_count].key,
7792 			       lun->per_res[i].res_key.key,
7793 			       ctl_min(sizeof(res_keys->keys[key_count].key),
7794 			       sizeof(lun->per_res[i].res_key)));
7795 			key_count++;
7796 		}
7797 		break;
7798 	}
7799 	case SPRI_RR: { // read reservation
7800 		struct scsi_per_res_in_rsrv *res;
7801 		int tmp_len, header_only;
7802 
7803 		res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7804 
7805 		scsi_ulto4b(lun->PRGeneration, res->header.generation);
7806 
7807 		if (lun->flags & CTL_LUN_PR_RESERVED)
7808 		{
7809 			tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7810 			scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7811 				    res->header.length);
7812 			header_only = 0;
7813 		} else {
7814 			tmp_len = sizeof(struct scsi_per_res_in_header);
7815 			scsi_ulto4b(0, res->header.length);
7816 			header_only = 1;
7817 		}
7818 
7819 		/*
7820 		 * We had to drop the lock to allocate our buffer, which
7821 		 * leaves time for someone to come in with another
7822 		 * persistent reservation.  (That is unlikely, though,
7823 		 * since this should be the only persistent reservation
7824 		 * command active right now.)
7825 		 */
7826 		if (tmp_len != total_len) {
7827 			mtx_unlock(&lun->lun_lock);
7828 			free(ctsio->kern_data_ptr, M_CTL);
7829 			printf("%s: reservation status changed, retrying\n",
7830 			       __func__);
7831 			goto retry;
7832 		}
7833 
7834 		/*
7835 		 * No reservation held, so we're done.
7836 		 */
7837 		if (header_only != 0)
7838 			break;
7839 
7840 		/*
7841 		 * If the registration is an All Registrants type, the key
7842 		 * is 0, since it doesn't really matter.
7843 		 */
7844 		if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7845 			memcpy(res->data.reservation,
7846 			       &lun->per_res[lun->pr_res_idx].res_key,
7847 			       sizeof(struct scsi_per_res_key));
7848 		}
7849 		res->data.scopetype = lun->res_type;
7850 		break;
7851 	}
7852 	case SPRI_RC:     //report capabilities
7853 	{
7854 		struct scsi_per_res_cap *res_cap;
7855 		uint16_t type_mask;
7856 
7857 		res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7858 		scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7859 		res_cap->flags2 |= SPRI_TMV | SPRI_ALLOW_3;
7860 		type_mask = SPRI_TM_WR_EX_AR |
7861 			    SPRI_TM_EX_AC_RO |
7862 			    SPRI_TM_WR_EX_RO |
7863 			    SPRI_TM_EX_AC |
7864 			    SPRI_TM_WR_EX |
7865 			    SPRI_TM_EX_AC_AR;
7866 		scsi_ulto2b(type_mask, res_cap->type_mask);
7867 		break;
7868 	}
7869 	case SPRI_RS: { // read full status
7870 		struct scsi_per_res_in_full *res_status;
7871 		struct scsi_per_res_in_full_desc *res_desc;
7872 		struct ctl_port *port;
7873 		int i, len;
7874 
7875 		res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7876 
7877 		/*
7878 		 * We had to drop the lock to allocate our buffer, which
7879 		 * leaves time for someone to come in with another
7880 		 * persistent reservation.  (That is unlikely, though,
7881 		 * since this should be the only persistent reservation
7882 		 * command active right now.)
7883 		 */
7884 		if (total_len < (sizeof(struct scsi_per_res_in_header) +
7885 		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7886 		     lun->pr_key_count)){
7887 			mtx_unlock(&lun->lun_lock);
7888 			free(ctsio->kern_data_ptr, M_CTL);
7889 			printf("%s: reservation length changed, retrying\n",
7890 			       __func__);
7891 			goto retry;
7892 		}
7893 
7894 		scsi_ulto4b(lun->PRGeneration, res_status->header.generation);
7895 
7896 		res_desc = &res_status->desc[0];
7897 		for (i = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7898 			if (!lun->per_res[i].registered)
7899 				continue;
7900 
7901 			memcpy(&res_desc->res_key, &lun->per_res[i].res_key.key,
7902 			    sizeof(res_desc->res_key));
7903 			if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7904 			    (lun->pr_res_idx == i ||
7905 			     lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7906 				res_desc->flags = SPRI_FULL_R_HOLDER;
7907 				res_desc->scopetype = lun->res_type;
7908 			}
7909 			scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7910 			    res_desc->rel_trgt_port_id);
7911 			len = 0;
7912 			port = softc->ctl_ports[i / CTL_MAX_INIT_PER_PORT];
7913 			if (port != NULL)
7914 				len = ctl_create_iid(port,
7915 				    i % CTL_MAX_INIT_PER_PORT,
7916 				    res_desc->transport_id);
7917 			scsi_ulto4b(len, res_desc->additional_length);
7918 			res_desc = (struct scsi_per_res_in_full_desc *)
7919 			    &res_desc->transport_id[len];
7920 		}
7921 		scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7922 		    res_status->header.length);
7923 		break;
7924 	}
7925 	default:
7926 		/*
7927 		 * This is a bug, because we just checked for this above,
7928 		 * and should have returned an error.
7929 		 */
7930 		panic("Invalid PR type %x", cdb->action);
7931 		break; /* NOTREACHED */
7932 	}
7933 	mtx_unlock(&lun->lun_lock);
7934 
7935 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7936 	ctsio->be_move_done = ctl_config_move_done;
7937 
7938 	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7939 			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7940 			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7941 			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7942 			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7943 
7944 	ctl_datamove((union ctl_io *)ctsio);
7945 
7946 	return (CTL_RETVAL_COMPLETE);
7947 }
7948 
7949 /*
7950  * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7951  * it should return.
7952  */
7953 static int
7954 ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7955 		uint64_t sa_res_key, uint8_t type, uint32_t residx,
7956 		struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7957 		struct scsi_per_res_out_parms* param)
7958 {
7959 	union ctl_ha_msg persis_io;
7960 	int retval, i;
7961 	int isc_retval;
7962 
7963 	retval = 0;
7964 
7965 	mtx_lock(&lun->lun_lock);
7966 	if (sa_res_key == 0) {
7967 		if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7968 			/* validate scope and type */
7969 			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7970 			     SPR_LU_SCOPE) {
7971 				mtx_unlock(&lun->lun_lock);
7972 				ctl_set_invalid_field(/*ctsio*/ ctsio,
7973 						      /*sks_valid*/ 1,
7974 						      /*command*/ 1,
7975 						      /*field*/ 2,
7976 						      /*bit_valid*/ 1,
7977 						      /*bit*/ 4);
7978 				ctl_done((union ctl_io *)ctsio);
7979 				return (1);
7980 			}
7981 
7982 		        if (type>8 || type==2 || type==4 || type==0) {
7983 				mtx_unlock(&lun->lun_lock);
7984 				ctl_set_invalid_field(/*ctsio*/ ctsio,
7985        	           				      /*sks_valid*/ 1,
7986 						      /*command*/ 1,
7987 						      /*field*/ 2,
7988 						      /*bit_valid*/ 1,
7989 						      /*bit*/ 0);
7990 				ctl_done((union ctl_io *)ctsio);
7991 				return (1);
7992 		        }
7993 
7994 			/* temporarily unregister this nexus */
7995 			lun->per_res[residx].registered = 0;
7996 
7997 			/*
7998 			 * Unregister everybody else and build UA for
7999 			 * them
8000 			 */
8001 			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8002 				if (lun->per_res[i].registered == 0)
8003 					continue;
8004 
8005 				if (!persis_offset
8006 				 && i <CTL_MAX_INITIATORS)
8007 					lun->pending_ua[i] |=
8008 						CTL_UA_REG_PREEMPT;
8009 				else if (persis_offset
8010 				      && i >= persis_offset)
8011 					lun->pending_ua[i-persis_offset] |=
8012 						CTL_UA_REG_PREEMPT;
8013 				lun->per_res[i].registered = 0;
8014 				memset(&lun->per_res[i].res_key, 0,
8015 				       sizeof(struct scsi_per_res_key));
8016 			}
8017 			lun->per_res[residx].registered = 1;
8018 			lun->pr_key_count = 1;
8019 			lun->res_type = type;
8020 			if (lun->res_type != SPR_TYPE_WR_EX_AR
8021 			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8022 				lun->pr_res_idx = residx;
8023 
8024 			/* send msg to other side */
8025 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8026 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8027 			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8028 			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8029 			persis_io.pr.pr_info.res_type = type;
8030 			memcpy(persis_io.pr.pr_info.sa_res_key,
8031 			       param->serv_act_res_key,
8032 			       sizeof(param->serv_act_res_key));
8033 			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8034 			     &persis_io, sizeof(persis_io), 0)) >
8035 			     CTL_HA_STATUS_SUCCESS) {
8036 				printf("CTL:Persis Out error returned "
8037 				       "from ctl_ha_msg_send %d\n",
8038 				       isc_retval);
8039 			}
8040 		} else {
8041 			/* not all registrants */
8042 			mtx_unlock(&lun->lun_lock);
8043 			free(ctsio->kern_data_ptr, M_CTL);
8044 			ctl_set_invalid_field(ctsio,
8045 					      /*sks_valid*/ 1,
8046 					      /*command*/ 0,
8047 					      /*field*/ 8,
8048 					      /*bit_valid*/ 0,
8049 					      /*bit*/ 0);
8050 			ctl_done((union ctl_io *)ctsio);
8051 			return (1);
8052 		}
8053 	} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8054 		|| !(lun->flags & CTL_LUN_PR_RESERVED)) {
8055 		int found = 0;
8056 
8057 		if (res_key == sa_res_key) {
8058 			/* special case */
8059 			/*
8060 			 * The spec implies this is not good but doesn't
8061 			 * say what to do. There are two choices either
8062 			 * generate a res conflict or check condition
8063 			 * with illegal field in parameter data. Since
8064 			 * that is what is done when the sa_res_key is
8065 			 * zero I'll take that approach since this has
8066 			 * to do with the sa_res_key.
8067 			 */
8068 			mtx_unlock(&lun->lun_lock);
8069 			free(ctsio->kern_data_ptr, M_CTL);
8070 			ctl_set_invalid_field(ctsio,
8071 					      /*sks_valid*/ 1,
8072 					      /*command*/ 0,
8073 					      /*field*/ 8,
8074 					      /*bit_valid*/ 0,
8075 					      /*bit*/ 0);
8076 			ctl_done((union ctl_io *)ctsio);
8077 			return (1);
8078 		}
8079 
8080 		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8081 			if (lun->per_res[i].registered
8082 			 && memcmp(param->serv_act_res_key,
8083 			    lun->per_res[i].res_key.key,
8084 			    sizeof(struct scsi_per_res_key)) != 0)
8085 				continue;
8086 
8087 			found = 1;
8088 			lun->per_res[i].registered = 0;
8089 			memset(&lun->per_res[i].res_key, 0,
8090 			       sizeof(struct scsi_per_res_key));
8091 			lun->pr_key_count--;
8092 
8093 			if (!persis_offset && i < CTL_MAX_INITIATORS)
8094 				lun->pending_ua[i] |= CTL_UA_REG_PREEMPT;
8095 			else if (persis_offset && i >= persis_offset)
8096 				lun->pending_ua[i-persis_offset] |=
8097 					CTL_UA_REG_PREEMPT;
8098 		}
8099 		if (!found) {
8100 			mtx_unlock(&lun->lun_lock);
8101 			free(ctsio->kern_data_ptr, M_CTL);
8102 			ctl_set_reservation_conflict(ctsio);
8103 			ctl_done((union ctl_io *)ctsio);
8104 			return (CTL_RETVAL_COMPLETE);
8105 		}
8106 		/* send msg to other side */
8107 		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8108 		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8109 		persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8110 		persis_io.pr.pr_info.residx = lun->pr_res_idx;
8111 		persis_io.pr.pr_info.res_type = type;
8112 		memcpy(persis_io.pr.pr_info.sa_res_key,
8113 		       param->serv_act_res_key,
8114 		       sizeof(param->serv_act_res_key));
8115 		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8116 		     &persis_io, sizeof(persis_io), 0)) >
8117 		     CTL_HA_STATUS_SUCCESS) {
8118 			printf("CTL:Persis Out error returned from "
8119 			       "ctl_ha_msg_send %d\n", isc_retval);
8120 		}
8121 	} else {
8122 		/* Reserved but not all registrants */
8123 		/* sa_res_key is res holder */
8124 		if (memcmp(param->serv_act_res_key,
8125                    lun->per_res[lun->pr_res_idx].res_key.key,
8126                    sizeof(struct scsi_per_res_key)) == 0) {
8127 			/* validate scope and type */
8128 			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
8129 			     SPR_LU_SCOPE) {
8130 				mtx_unlock(&lun->lun_lock);
8131 				ctl_set_invalid_field(/*ctsio*/ ctsio,
8132 						      /*sks_valid*/ 1,
8133 						      /*command*/ 1,
8134 						      /*field*/ 2,
8135 						      /*bit_valid*/ 1,
8136 						      /*bit*/ 4);
8137 				ctl_done((union ctl_io *)ctsio);
8138 				return (1);
8139 			}
8140 
8141 			if (type>8 || type==2 || type==4 || type==0) {
8142 				mtx_unlock(&lun->lun_lock);
8143 				ctl_set_invalid_field(/*ctsio*/ ctsio,
8144 						      /*sks_valid*/ 1,
8145 						      /*command*/ 1,
8146 						      /*field*/ 2,
8147 						      /*bit_valid*/ 1,
8148 						      /*bit*/ 0);
8149 				ctl_done((union ctl_io *)ctsio);
8150 				return (1);
8151 			}
8152 
8153 			/*
8154 			 * Do the following:
8155 			 * if sa_res_key != res_key remove all
8156 			 * registrants w/sa_res_key and generate UA
8157 			 * for these registrants(Registrations
8158 			 * Preempted) if it wasn't an exclusive
8159 			 * reservation generate UA(Reservations
8160 			 * Preempted) for all other registered nexuses
8161 			 * if the type has changed. Establish the new
8162 			 * reservation and holder. If res_key and
8163 			 * sa_res_key are the same do the above
8164 			 * except don't unregister the res holder.
8165 			 */
8166 
8167 			/*
8168 			 * Temporarily unregister so it won't get
8169 			 * removed or UA generated
8170 			 */
8171 			lun->per_res[residx].registered = 0;
8172 			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8173 				if (lun->per_res[i].registered == 0)
8174 					continue;
8175 
8176 				if (memcmp(param->serv_act_res_key,
8177 				    lun->per_res[i].res_key.key,
8178 				    sizeof(struct scsi_per_res_key)) == 0) {
8179 					lun->per_res[i].registered = 0;
8180 					memset(&lun->per_res[i].res_key,
8181 					       0,
8182 					       sizeof(struct scsi_per_res_key));
8183 					lun->pr_key_count--;
8184 
8185 					if (!persis_offset
8186 					 && i < CTL_MAX_INITIATORS)
8187 						lun->pending_ua[i] |=
8188 							CTL_UA_REG_PREEMPT;
8189 					else if (persis_offset
8190 					      && i >= persis_offset)
8191 						lun->pending_ua[i-persis_offset] |=
8192 						  CTL_UA_REG_PREEMPT;
8193 				} else if (type != lun->res_type
8194 					&& (lun->res_type == SPR_TYPE_WR_EX_RO
8195 					 || lun->res_type ==SPR_TYPE_EX_AC_RO)){
8196 						if (!persis_offset
8197 						 && i < CTL_MAX_INITIATORS)
8198 							lun->pending_ua[i] |=
8199 							CTL_UA_RES_RELEASE;
8200 						else if (persis_offset
8201 						      && i >= persis_offset)
8202 							lun->pending_ua[
8203 							i-persis_offset] |=
8204 							CTL_UA_RES_RELEASE;
8205 				}
8206 			}
8207 			lun->per_res[residx].registered = 1;
8208 			lun->res_type = type;
8209 			if (lun->res_type != SPR_TYPE_WR_EX_AR
8210 			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8211 				lun->pr_res_idx = residx;
8212 			else
8213 				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8214 
8215 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8216 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8217 			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8218 			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8219 			persis_io.pr.pr_info.res_type = type;
8220 			memcpy(persis_io.pr.pr_info.sa_res_key,
8221 			       param->serv_act_res_key,
8222 			       sizeof(param->serv_act_res_key));
8223 			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8224 			     &persis_io, sizeof(persis_io), 0)) >
8225 			     CTL_HA_STATUS_SUCCESS) {
8226 				printf("CTL:Persis Out error returned "
8227 				       "from ctl_ha_msg_send %d\n",
8228 				       isc_retval);
8229 			}
8230 		} else {
8231 			/*
8232 			 * sa_res_key is not the res holder just
8233 			 * remove registrants
8234 			 */
8235 			int found=0;
8236 
8237 			for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8238 				if (memcmp(param->serv_act_res_key,
8239 				    lun->per_res[i].res_key.key,
8240 				    sizeof(struct scsi_per_res_key)) != 0)
8241 					continue;
8242 
8243 				found = 1;
8244 				lun->per_res[i].registered = 0;
8245 				memset(&lun->per_res[i].res_key, 0,
8246 				       sizeof(struct scsi_per_res_key));
8247 				lun->pr_key_count--;
8248 
8249 				if (!persis_offset
8250 				 && i < CTL_MAX_INITIATORS)
8251 					lun->pending_ua[i] |=
8252 						CTL_UA_REG_PREEMPT;
8253 				else if (persis_offset
8254 				      && i >= persis_offset)
8255 					lun->pending_ua[i-persis_offset] |=
8256 						CTL_UA_REG_PREEMPT;
8257 			}
8258 
8259 			if (!found) {
8260 				mtx_unlock(&lun->lun_lock);
8261 				free(ctsio->kern_data_ptr, M_CTL);
8262 				ctl_set_reservation_conflict(ctsio);
8263 				ctl_done((union ctl_io *)ctsio);
8264 		        	return (1);
8265 			}
8266 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8267 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8268 			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8269 			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8270 			persis_io.pr.pr_info.res_type = type;
8271 			memcpy(persis_io.pr.pr_info.sa_res_key,
8272 			       param->serv_act_res_key,
8273 			       sizeof(param->serv_act_res_key));
8274 			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8275 			     &persis_io, sizeof(persis_io), 0)) >
8276 			     CTL_HA_STATUS_SUCCESS) {
8277 				printf("CTL:Persis Out error returned "
8278 				       "from ctl_ha_msg_send %d\n",
8279 				isc_retval);
8280 			}
8281 		}
8282 	}
8283 
8284 	lun->PRGeneration++;
8285 	mtx_unlock(&lun->lun_lock);
8286 
8287 	return (retval);
8288 }
8289 
8290 static void
8291 ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
8292 {
8293 	int i;
8294 
8295 	if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8296 	 || lun->pr_res_idx == CTL_PR_NO_RESERVATION
8297 	 || memcmp(&lun->per_res[lun->pr_res_idx].res_key,
8298 		   msg->pr.pr_info.sa_res_key,
8299 		   sizeof(struct scsi_per_res_key)) != 0) {
8300 		uint64_t sa_res_key;
8301 		sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
8302 
8303 		if (sa_res_key == 0) {
8304 			/* temporarily unregister this nexus */
8305 			lun->per_res[msg->pr.pr_info.residx].registered = 0;
8306 
8307 			/*
8308 			 * Unregister everybody else and build UA for
8309 			 * them
8310 			 */
8311 			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8312 				if (lun->per_res[i].registered == 0)
8313 					continue;
8314 
8315 				if (!persis_offset
8316 				 && i < CTL_MAX_INITIATORS)
8317 					lun->pending_ua[i] |=
8318 						CTL_UA_REG_PREEMPT;
8319 				else if (persis_offset && i >= persis_offset)
8320 					lun->pending_ua[i - persis_offset] |=
8321 						CTL_UA_REG_PREEMPT;
8322 				lun->per_res[i].registered = 0;
8323 				memset(&lun->per_res[i].res_key, 0,
8324 				       sizeof(struct scsi_per_res_key));
8325 			}
8326 
8327 			lun->per_res[msg->pr.pr_info.residx].registered = 1;
8328 			lun->pr_key_count = 1;
8329 			lun->res_type = msg->pr.pr_info.res_type;
8330 			if (lun->res_type != SPR_TYPE_WR_EX_AR
8331 			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8332 				lun->pr_res_idx = msg->pr.pr_info.residx;
8333 		} else {
8334 		        for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8335 				if (memcmp(msg->pr.pr_info.sa_res_key,
8336 		                   lun->per_res[i].res_key.key,
8337 		                   sizeof(struct scsi_per_res_key)) != 0)
8338 					continue;
8339 
8340 				lun->per_res[i].registered = 0;
8341 				memset(&lun->per_res[i].res_key, 0,
8342 				       sizeof(struct scsi_per_res_key));
8343 				lun->pr_key_count--;
8344 
8345 				if (!persis_offset
8346 				 && i < persis_offset)
8347 					lun->pending_ua[i] |=
8348 						CTL_UA_REG_PREEMPT;
8349 				else if (persis_offset
8350 				      && i >= persis_offset)
8351 					lun->pending_ua[i - persis_offset] |=
8352 						CTL_UA_REG_PREEMPT;
8353 			}
8354 		}
8355 	} else {
8356 		/*
8357 		 * Temporarily unregister so it won't get removed
8358 		 * or UA generated
8359 		 */
8360 		lun->per_res[msg->pr.pr_info.residx].registered = 0;
8361 		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8362 			if (lun->per_res[i].registered == 0)
8363 				continue;
8364 
8365 			if (memcmp(msg->pr.pr_info.sa_res_key,
8366 	                   lun->per_res[i].res_key.key,
8367 	                   sizeof(struct scsi_per_res_key)) == 0) {
8368 				lun->per_res[i].registered = 0;
8369 				memset(&lun->per_res[i].res_key, 0,
8370 				       sizeof(struct scsi_per_res_key));
8371 				lun->pr_key_count--;
8372 				if (!persis_offset
8373 				 && i < CTL_MAX_INITIATORS)
8374 					lun->pending_ua[i] |=
8375 						CTL_UA_REG_PREEMPT;
8376 				else if (persis_offset
8377 				      && i >= persis_offset)
8378 					lun->pending_ua[i - persis_offset] |=
8379 						CTL_UA_REG_PREEMPT;
8380 			} else if (msg->pr.pr_info.res_type != lun->res_type
8381 				&& (lun->res_type == SPR_TYPE_WR_EX_RO
8382 				 || lun->res_type == SPR_TYPE_EX_AC_RO)) {
8383 					if (!persis_offset
8384 					 && i < persis_offset)
8385 						lun->pending_ua[i] |=
8386 							CTL_UA_RES_RELEASE;
8387 					else if (persis_offset
8388 					      && i >= persis_offset)
8389 					lun->pending_ua[i - persis_offset] |=
8390 						CTL_UA_RES_RELEASE;
8391 			}
8392 		}
8393 		lun->per_res[msg->pr.pr_info.residx].registered = 1;
8394 		lun->res_type = msg->pr.pr_info.res_type;
8395 		if (lun->res_type != SPR_TYPE_WR_EX_AR
8396 		 && lun->res_type != SPR_TYPE_EX_AC_AR)
8397 			lun->pr_res_idx = msg->pr.pr_info.residx;
8398 		else
8399 			lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8400 	}
8401 	lun->PRGeneration++;
8402 
8403 }
8404 
8405 
8406 int
8407 ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8408 {
8409 	int retval;
8410 	int isc_retval;
8411 	u_int32_t param_len;
8412 	struct scsi_per_res_out *cdb;
8413 	struct ctl_lun *lun;
8414 	struct scsi_per_res_out_parms* param;
8415 	struct ctl_softc *softc;
8416 	uint32_t residx;
8417 	uint64_t res_key, sa_res_key;
8418 	uint8_t type;
8419 	union ctl_ha_msg persis_io;
8420 	int    i;
8421 
8422 	CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8423 
8424 	retval = CTL_RETVAL_COMPLETE;
8425 
8426 	softc = control_softc;
8427 
8428 	cdb = (struct scsi_per_res_out *)ctsio->cdb;
8429 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8430 
8431 	/*
8432 	 * We only support whole-LUN scope.  The scope & type are ignored for
8433 	 * register, register and ignore existing key and clear.
8434 	 * We sometimes ignore scope and type on preempts too!!
8435 	 * Verify reservation type here as well.
8436 	 */
8437 	type = cdb->scope_type & SPR_TYPE_MASK;
8438 	if ((cdb->action == SPRO_RESERVE)
8439 	 || (cdb->action == SPRO_RELEASE)) {
8440 		if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8441 			ctl_set_invalid_field(/*ctsio*/ ctsio,
8442 					      /*sks_valid*/ 1,
8443 					      /*command*/ 1,
8444 					      /*field*/ 2,
8445 					      /*bit_valid*/ 1,
8446 					      /*bit*/ 4);
8447 			ctl_done((union ctl_io *)ctsio);
8448 			return (CTL_RETVAL_COMPLETE);
8449 		}
8450 
8451 		if (type>8 || type==2 || type==4 || type==0) {
8452 			ctl_set_invalid_field(/*ctsio*/ ctsio,
8453 					      /*sks_valid*/ 1,
8454 					      /*command*/ 1,
8455 					      /*field*/ 2,
8456 					      /*bit_valid*/ 1,
8457 					      /*bit*/ 0);
8458 			ctl_done((union ctl_io *)ctsio);
8459 			return (CTL_RETVAL_COMPLETE);
8460 		}
8461 	}
8462 
8463 	param_len = scsi_4btoul(cdb->length);
8464 
8465 	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8466 		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8467 		ctsio->kern_data_len = param_len;
8468 		ctsio->kern_total_len = param_len;
8469 		ctsio->kern_data_resid = 0;
8470 		ctsio->kern_rel_offset = 0;
8471 		ctsio->kern_sg_entries = 0;
8472 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8473 		ctsio->be_move_done = ctl_config_move_done;
8474 		ctl_datamove((union ctl_io *)ctsio);
8475 
8476 		return (CTL_RETVAL_COMPLETE);
8477 	}
8478 
8479 	param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8480 
8481 	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8482 	res_key = scsi_8btou64(param->res_key.key);
8483 	sa_res_key = scsi_8btou64(param->serv_act_res_key);
8484 
8485 	/*
8486 	 * Validate the reservation key here except for SPRO_REG_IGNO
8487 	 * This must be done for all other service actions
8488 	 */
8489 	if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8490 		mtx_lock(&lun->lun_lock);
8491 		if (lun->per_res[residx].registered) {
8492 		    if (memcmp(param->res_key.key,
8493 			       lun->per_res[residx].res_key.key,
8494 			       ctl_min(sizeof(param->res_key),
8495 			       sizeof(lun->per_res[residx].res_key))) != 0) {
8496 				/*
8497 				 * The current key passed in doesn't match
8498 				 * the one the initiator previously
8499 				 * registered.
8500 				 */
8501 				mtx_unlock(&lun->lun_lock);
8502 				free(ctsio->kern_data_ptr, M_CTL);
8503 				ctl_set_reservation_conflict(ctsio);
8504 				ctl_done((union ctl_io *)ctsio);
8505 				return (CTL_RETVAL_COMPLETE);
8506 			}
8507 		} else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8508 			/*
8509 			 * We are not registered
8510 			 */
8511 			mtx_unlock(&lun->lun_lock);
8512 			free(ctsio->kern_data_ptr, M_CTL);
8513 			ctl_set_reservation_conflict(ctsio);
8514 			ctl_done((union ctl_io *)ctsio);
8515 			return (CTL_RETVAL_COMPLETE);
8516 		} else if (res_key != 0) {
8517 			/*
8518 			 * We are not registered and trying to register but
8519 			 * the register key isn't zero.
8520 			 */
8521 			mtx_unlock(&lun->lun_lock);
8522 			free(ctsio->kern_data_ptr, M_CTL);
8523 			ctl_set_reservation_conflict(ctsio);
8524 			ctl_done((union ctl_io *)ctsio);
8525 			return (CTL_RETVAL_COMPLETE);
8526 		}
8527 		mtx_unlock(&lun->lun_lock);
8528 	}
8529 
8530 	switch (cdb->action & SPRO_ACTION_MASK) {
8531 	case SPRO_REGISTER:
8532 	case SPRO_REG_IGNO: {
8533 
8534 #if 0
8535 		printf("Registration received\n");
8536 #endif
8537 
8538 		/*
8539 		 * We don't support any of these options, as we report in
8540 		 * the read capabilities request (see
8541 		 * ctl_persistent_reserve_in(), above).
8542 		 */
8543 		if ((param->flags & SPR_SPEC_I_PT)
8544 		 || (param->flags & SPR_ALL_TG_PT)
8545 		 || (param->flags & SPR_APTPL)) {
8546 			int bit_ptr;
8547 
8548 			if (param->flags & SPR_APTPL)
8549 				bit_ptr = 0;
8550 			else if (param->flags & SPR_ALL_TG_PT)
8551 				bit_ptr = 2;
8552 			else /* SPR_SPEC_I_PT */
8553 				bit_ptr = 3;
8554 
8555 			free(ctsio->kern_data_ptr, M_CTL);
8556 			ctl_set_invalid_field(ctsio,
8557 					      /*sks_valid*/ 1,
8558 					      /*command*/ 0,
8559 					      /*field*/ 20,
8560 					      /*bit_valid*/ 1,
8561 					      /*bit*/ bit_ptr);
8562 			ctl_done((union ctl_io *)ctsio);
8563 			return (CTL_RETVAL_COMPLETE);
8564 		}
8565 
8566 		mtx_lock(&lun->lun_lock);
8567 
8568 		/*
8569 		 * The initiator wants to clear the
8570 		 * key/unregister.
8571 		 */
8572 		if (sa_res_key == 0) {
8573 			if ((res_key == 0
8574 			  && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8575 			 || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8576 			  && !lun->per_res[residx].registered)) {
8577 				mtx_unlock(&lun->lun_lock);
8578 				goto done;
8579 			}
8580 
8581 			lun->per_res[residx].registered = 0;
8582 			memset(&lun->per_res[residx].res_key,
8583 			       0, sizeof(lun->per_res[residx].res_key));
8584 			lun->pr_key_count--;
8585 
8586 			if (residx == lun->pr_res_idx) {
8587 				lun->flags &= ~CTL_LUN_PR_RESERVED;
8588 				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8589 
8590 				if ((lun->res_type == SPR_TYPE_WR_EX_RO
8591 				  || lun->res_type == SPR_TYPE_EX_AC_RO)
8592 				 && lun->pr_key_count) {
8593 					/*
8594 					 * If the reservation is a registrants
8595 					 * only type we need to generate a UA
8596 					 * for other registered inits.  The
8597 					 * sense code should be RESERVATIONS
8598 					 * RELEASED
8599 					 */
8600 
8601 					for (i = 0; i < CTL_MAX_INITIATORS;i++){
8602 						if (lun->per_res[
8603 						    i+persis_offset].registered
8604 						    == 0)
8605 							continue;
8606 						lun->pending_ua[i] |=
8607 							CTL_UA_RES_RELEASE;
8608 					}
8609 				}
8610 				lun->res_type = 0;
8611 			} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8612 				if (lun->pr_key_count==0) {
8613 					lun->flags &= ~CTL_LUN_PR_RESERVED;
8614 					lun->res_type = 0;
8615 					lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8616 				}
8617 			}
8618 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8619 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8620 			persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8621 			persis_io.pr.pr_info.residx = residx;
8622 			if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8623 			     &persis_io, sizeof(persis_io), 0 )) >
8624 			     CTL_HA_STATUS_SUCCESS) {
8625 				printf("CTL:Persis Out error returned from "
8626 				       "ctl_ha_msg_send %d\n", isc_retval);
8627 			}
8628 		} else /* sa_res_key != 0 */ {
8629 
8630 			/*
8631 			 * If we aren't registered currently then increment
8632 			 * the key count and set the registered flag.
8633 			 */
8634 			if (!lun->per_res[residx].registered) {
8635 				lun->pr_key_count++;
8636 				lun->per_res[residx].registered = 1;
8637 			}
8638 
8639 			memcpy(&lun->per_res[residx].res_key,
8640 			       param->serv_act_res_key,
8641 			       ctl_min(sizeof(param->serv_act_res_key),
8642 			       sizeof(lun->per_res[residx].res_key)));
8643 
8644 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8645 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8646 			persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8647 			persis_io.pr.pr_info.residx = residx;
8648 			memcpy(persis_io.pr.pr_info.sa_res_key,
8649 			       param->serv_act_res_key,
8650 			       sizeof(param->serv_act_res_key));
8651 			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8652 			     &persis_io, sizeof(persis_io), 0)) >
8653 			     CTL_HA_STATUS_SUCCESS) {
8654 				printf("CTL:Persis Out error returned from "
8655 				       "ctl_ha_msg_send %d\n", isc_retval);
8656 			}
8657 		}
8658 		lun->PRGeneration++;
8659 		mtx_unlock(&lun->lun_lock);
8660 
8661 		break;
8662 	}
8663 	case SPRO_RESERVE:
8664 #if 0
8665                 printf("Reserve executed type %d\n", type);
8666 #endif
8667 		mtx_lock(&lun->lun_lock);
8668 		if (lun->flags & CTL_LUN_PR_RESERVED) {
8669 			/*
8670 			 * if this isn't the reservation holder and it's
8671 			 * not a "all registrants" type or if the type is
8672 			 * different then we have a conflict
8673 			 */
8674 			if ((lun->pr_res_idx != residx
8675 			  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8676 			 || lun->res_type != type) {
8677 				mtx_unlock(&lun->lun_lock);
8678 				free(ctsio->kern_data_ptr, M_CTL);
8679 				ctl_set_reservation_conflict(ctsio);
8680 				ctl_done((union ctl_io *)ctsio);
8681 				return (CTL_RETVAL_COMPLETE);
8682 			}
8683 			mtx_unlock(&lun->lun_lock);
8684 		} else /* create a reservation */ {
8685 			/*
8686 			 * If it's not an "all registrants" type record
8687 			 * reservation holder
8688 			 */
8689 			if (type != SPR_TYPE_WR_EX_AR
8690 			 && type != SPR_TYPE_EX_AC_AR)
8691 				lun->pr_res_idx = residx; /* Res holder */
8692 			else
8693 				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8694 
8695 			lun->flags |= CTL_LUN_PR_RESERVED;
8696 			lun->res_type = type;
8697 
8698 			mtx_unlock(&lun->lun_lock);
8699 
8700 			/* send msg to other side */
8701 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8702 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8703 			persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8704 			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8705 			persis_io.pr.pr_info.res_type = type;
8706 			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8707 			     &persis_io, sizeof(persis_io), 0)) >
8708 			     CTL_HA_STATUS_SUCCESS) {
8709 				printf("CTL:Persis Out error returned from "
8710 				       "ctl_ha_msg_send %d\n", isc_retval);
8711 			}
8712 		}
8713 		break;
8714 
8715 	case SPRO_RELEASE:
8716 		mtx_lock(&lun->lun_lock);
8717 		if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8718 			/* No reservation exists return good status */
8719 			mtx_unlock(&lun->lun_lock);
8720 			goto done;
8721 		}
8722 		/*
8723 		 * Is this nexus a reservation holder?
8724 		 */
8725 		if (lun->pr_res_idx != residx
8726 		 && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8727 			/*
8728 			 * not a res holder return good status but
8729 			 * do nothing
8730 			 */
8731 			mtx_unlock(&lun->lun_lock);
8732 			goto done;
8733 		}
8734 
8735 		if (lun->res_type != type) {
8736 			mtx_unlock(&lun->lun_lock);
8737 			free(ctsio->kern_data_ptr, M_CTL);
8738 			ctl_set_illegal_pr_release(ctsio);
8739 			ctl_done((union ctl_io *)ctsio);
8740 			return (CTL_RETVAL_COMPLETE);
8741 		}
8742 
8743 		/* okay to release */
8744 		lun->flags &= ~CTL_LUN_PR_RESERVED;
8745 		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8746 		lun->res_type = 0;
8747 
8748 		/*
8749 		 * if this isn't an exclusive access
8750 		 * res generate UA for all other
8751 		 * registrants.
8752 		 */
8753 		if (type != SPR_TYPE_EX_AC
8754 		 && type != SPR_TYPE_WR_EX) {
8755 			/*
8756 			 * temporarily unregister so we don't generate UA
8757 			 */
8758 			lun->per_res[residx].registered = 0;
8759 
8760 			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8761 				if (lun->per_res[i+persis_offset].registered
8762 				    == 0)
8763 					continue;
8764 				lun->pending_ua[i] |=
8765 					CTL_UA_RES_RELEASE;
8766 			}
8767 
8768 			lun->per_res[residx].registered = 1;
8769 		}
8770 		mtx_unlock(&lun->lun_lock);
8771 		/* Send msg to other side */
8772 		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8773 		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8774 		persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8775 		if ((isc_retval=ctl_ha_msg_send( CTL_HA_CHAN_CTL, &persis_io,
8776 		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8777 			printf("CTL:Persis Out error returned from "
8778 			       "ctl_ha_msg_send %d\n", isc_retval);
8779 		}
8780 		break;
8781 
8782 	case SPRO_CLEAR:
8783 		/* send msg to other side */
8784 
8785 		mtx_lock(&lun->lun_lock);
8786 		lun->flags &= ~CTL_LUN_PR_RESERVED;
8787 		lun->res_type = 0;
8788 		lun->pr_key_count = 0;
8789 		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8790 
8791 
8792 		memset(&lun->per_res[residx].res_key,
8793 		       0, sizeof(lun->per_res[residx].res_key));
8794 		lun->per_res[residx].registered = 0;
8795 
8796 		for (i=0; i < 2*CTL_MAX_INITIATORS; i++)
8797 			if (lun->per_res[i].registered) {
8798 				if (!persis_offset && i < CTL_MAX_INITIATORS)
8799 					lun->pending_ua[i] |=
8800 						CTL_UA_RES_PREEMPT;
8801 				else if (persis_offset && i >= persis_offset)
8802 					lun->pending_ua[i-persis_offset] |=
8803 					    CTL_UA_RES_PREEMPT;
8804 
8805 				memset(&lun->per_res[i].res_key,
8806 				       0, sizeof(struct scsi_per_res_key));
8807 				lun->per_res[i].registered = 0;
8808 			}
8809 		lun->PRGeneration++;
8810 		mtx_unlock(&lun->lun_lock);
8811 		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8812 		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8813 		persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8814 		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8815 		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8816 			printf("CTL:Persis Out error returned from "
8817 			       "ctl_ha_msg_send %d\n", isc_retval);
8818 		}
8819 		break;
8820 
8821 	case SPRO_PREEMPT: {
8822 		int nretval;
8823 
8824 		nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8825 					  residx, ctsio, cdb, param);
8826 		if (nretval != 0)
8827 			return (CTL_RETVAL_COMPLETE);
8828 		break;
8829 	}
8830 	default:
8831 		panic("Invalid PR type %x", cdb->action);
8832 	}
8833 
8834 done:
8835 	free(ctsio->kern_data_ptr, M_CTL);
8836 	ctl_set_success(ctsio);
8837 	ctl_done((union ctl_io *)ctsio);
8838 
8839 	return (retval);
8840 }
8841 
8842 /*
8843  * This routine is for handling a message from the other SC pertaining to
8844  * persistent reserve out. All the error checking will have been done
8845  * so only perorming the action need be done here to keep the two
8846  * in sync.
8847  */
8848 static void
8849 ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8850 {
8851 	struct ctl_lun *lun;
8852 	struct ctl_softc *softc;
8853 	int i;
8854 	uint32_t targ_lun;
8855 
8856 	softc = control_softc;
8857 
8858 	targ_lun = msg->hdr.nexus.targ_mapped_lun;
8859 	lun = softc->ctl_luns[targ_lun];
8860 	mtx_lock(&lun->lun_lock);
8861 	switch(msg->pr.pr_info.action) {
8862 	case CTL_PR_REG_KEY:
8863 		if (!lun->per_res[msg->pr.pr_info.residx].registered) {
8864 			lun->per_res[msg->pr.pr_info.residx].registered = 1;
8865 			lun->pr_key_count++;
8866 		}
8867 		lun->PRGeneration++;
8868 		memcpy(&lun->per_res[msg->pr.pr_info.residx].res_key,
8869 		       msg->pr.pr_info.sa_res_key,
8870 		       sizeof(struct scsi_per_res_key));
8871 		break;
8872 
8873 	case CTL_PR_UNREG_KEY:
8874 		lun->per_res[msg->pr.pr_info.residx].registered = 0;
8875 		memset(&lun->per_res[msg->pr.pr_info.residx].res_key,
8876 		       0, sizeof(struct scsi_per_res_key));
8877 		lun->pr_key_count--;
8878 
8879 		/* XXX Need to see if the reservation has been released */
8880 		/* if so do we need to generate UA? */
8881 		if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8882 			lun->flags &= ~CTL_LUN_PR_RESERVED;
8883 			lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8884 
8885 			if ((lun->res_type == SPR_TYPE_WR_EX_RO
8886 			  || lun->res_type == SPR_TYPE_EX_AC_RO)
8887 			 && lun->pr_key_count) {
8888 				/*
8889 				 * If the reservation is a registrants
8890 				 * only type we need to generate a UA
8891 				 * for other registered inits.  The
8892 				 * sense code should be RESERVATIONS
8893 				 * RELEASED
8894 				 */
8895 
8896 				for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8897 					if (lun->per_res[i+
8898 					    persis_offset].registered == 0)
8899 						continue;
8900 
8901 					lun->pending_ua[i] |=
8902 						CTL_UA_RES_RELEASE;
8903 				}
8904 			}
8905 			lun->res_type = 0;
8906 		} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8907 			if (lun->pr_key_count==0) {
8908 				lun->flags &= ~CTL_LUN_PR_RESERVED;
8909 				lun->res_type = 0;
8910 				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8911 			}
8912 		}
8913 		lun->PRGeneration++;
8914 		break;
8915 
8916 	case CTL_PR_RESERVE:
8917 		lun->flags |= CTL_LUN_PR_RESERVED;
8918 		lun->res_type = msg->pr.pr_info.res_type;
8919 		lun->pr_res_idx = msg->pr.pr_info.residx;
8920 
8921 		break;
8922 
8923 	case CTL_PR_RELEASE:
8924 		/*
8925 		 * if this isn't an exclusive access res generate UA for all
8926 		 * other registrants.
8927 		 */
8928 		if (lun->res_type != SPR_TYPE_EX_AC
8929 		 && lun->res_type != SPR_TYPE_WR_EX) {
8930 			for (i = 0; i < CTL_MAX_INITIATORS; i++)
8931 				if (lun->per_res[i+persis_offset].registered)
8932 					lun->pending_ua[i] |=
8933 						CTL_UA_RES_RELEASE;
8934 		}
8935 
8936 		lun->flags &= ~CTL_LUN_PR_RESERVED;
8937 		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8938 		lun->res_type = 0;
8939 		break;
8940 
8941 	case CTL_PR_PREEMPT:
8942 		ctl_pro_preempt_other(lun, msg);
8943 		break;
8944 	case CTL_PR_CLEAR:
8945 		lun->flags &= ~CTL_LUN_PR_RESERVED;
8946 		lun->res_type = 0;
8947 		lun->pr_key_count = 0;
8948 		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8949 
8950 		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8951 			if (lun->per_res[i].registered == 0)
8952 				continue;
8953 			if (!persis_offset
8954 			 && i < CTL_MAX_INITIATORS)
8955 				lun->pending_ua[i] |= CTL_UA_RES_PREEMPT;
8956 			else if (persis_offset
8957 			      && i >= persis_offset)
8958 				lun->pending_ua[i-persis_offset] |=
8959 					CTL_UA_RES_PREEMPT;
8960 			memset(&lun->per_res[i].res_key, 0,
8961 			       sizeof(struct scsi_per_res_key));
8962 			lun->per_res[i].registered = 0;
8963 		}
8964 		lun->PRGeneration++;
8965 		break;
8966 	}
8967 
8968 	mtx_unlock(&lun->lun_lock);
8969 }
8970 
8971 int
8972 ctl_read_write(struct ctl_scsiio *ctsio)
8973 {
8974 	struct ctl_lun *lun;
8975 	struct ctl_lba_len_flags *lbalen;
8976 	uint64_t lba;
8977 	uint32_t num_blocks;
8978 	int fua, dpo;
8979 	int retval;
8980 	int isread;
8981 
8982 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8983 
8984 	CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8985 
8986 	fua = 0;
8987 	dpo = 0;
8988 
8989 	retval = CTL_RETVAL_COMPLETE;
8990 
8991 	isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8992 	      || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8993 	if (lun->flags & CTL_LUN_PR_RESERVED && isread) {
8994 		uint32_t residx;
8995 
8996 		/*
8997 		 * XXX KDM need a lock here.
8998 		 */
8999 		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
9000 		if ((lun->res_type == SPR_TYPE_EX_AC
9001 		  && residx != lun->pr_res_idx)
9002 		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
9003 		   || lun->res_type == SPR_TYPE_EX_AC_AR)
9004 		  && !lun->per_res[residx].registered)) {
9005 			ctl_set_reservation_conflict(ctsio);
9006 			ctl_done((union ctl_io *)ctsio);
9007 			return (CTL_RETVAL_COMPLETE);
9008 	        }
9009 	}
9010 
9011 	switch (ctsio->cdb[0]) {
9012 	case READ_6:
9013 	case WRITE_6: {
9014 		struct scsi_rw_6 *cdb;
9015 
9016 		cdb = (struct scsi_rw_6 *)ctsio->cdb;
9017 
9018 		lba = scsi_3btoul(cdb->addr);
9019 		/* only 5 bits are valid in the most significant address byte */
9020 		lba &= 0x1fffff;
9021 		num_blocks = cdb->length;
9022 		/*
9023 		 * This is correct according to SBC-2.
9024 		 */
9025 		if (num_blocks == 0)
9026 			num_blocks = 256;
9027 		break;
9028 	}
9029 	case READ_10:
9030 	case WRITE_10: {
9031 		struct scsi_rw_10 *cdb;
9032 
9033 		cdb = (struct scsi_rw_10 *)ctsio->cdb;
9034 
9035 		if (cdb->byte2 & SRW10_FUA)
9036 			fua = 1;
9037 		if (cdb->byte2 & SRW10_DPO)
9038 			dpo = 1;
9039 
9040 		lba = scsi_4btoul(cdb->addr);
9041 		num_blocks = scsi_2btoul(cdb->length);
9042 		break;
9043 	}
9044 	case WRITE_VERIFY_10: {
9045 		struct scsi_write_verify_10 *cdb;
9046 
9047 		cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
9048 
9049 		/*
9050 		 * XXX KDM we should do actual write verify support at some
9051 		 * point.  This is obviously fake, we're just translating
9052 		 * things to a write.  So we don't even bother checking the
9053 		 * BYTCHK field, since we don't do any verification.  If
9054 		 * the user asks for it, we'll just pretend we did it.
9055 		 */
9056 		if (cdb->byte2 & SWV_DPO)
9057 			dpo = 1;
9058 
9059 		lba = scsi_4btoul(cdb->addr);
9060 		num_blocks = scsi_2btoul(cdb->length);
9061 		break;
9062 	}
9063 	case READ_12:
9064 	case WRITE_12: {
9065 		struct scsi_rw_12 *cdb;
9066 
9067 		cdb = (struct scsi_rw_12 *)ctsio->cdb;
9068 
9069 		if (cdb->byte2 & SRW12_FUA)
9070 			fua = 1;
9071 		if (cdb->byte2 & SRW12_DPO)
9072 			dpo = 1;
9073 		lba = scsi_4btoul(cdb->addr);
9074 		num_blocks = scsi_4btoul(cdb->length);
9075 		break;
9076 	}
9077 	case WRITE_VERIFY_12: {
9078 		struct scsi_write_verify_12 *cdb;
9079 
9080 		cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
9081 
9082 		if (cdb->byte2 & SWV_DPO)
9083 			dpo = 1;
9084 
9085 		lba = scsi_4btoul(cdb->addr);
9086 		num_blocks = scsi_4btoul(cdb->length);
9087 
9088 		break;
9089 	}
9090 	case READ_16:
9091 	case WRITE_16: {
9092 		struct scsi_rw_16 *cdb;
9093 
9094 		cdb = (struct scsi_rw_16 *)ctsio->cdb;
9095 
9096 		if (cdb->byte2 & SRW12_FUA)
9097 			fua = 1;
9098 		if (cdb->byte2 & SRW12_DPO)
9099 			dpo = 1;
9100 
9101 		lba = scsi_8btou64(cdb->addr);
9102 		num_blocks = scsi_4btoul(cdb->length);
9103 		break;
9104 	}
9105 	case WRITE_VERIFY_16: {
9106 		struct scsi_write_verify_16 *cdb;
9107 
9108 		cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
9109 
9110 		if (cdb->byte2 & SWV_DPO)
9111 			dpo = 1;
9112 
9113 		lba = scsi_8btou64(cdb->addr);
9114 		num_blocks = scsi_4btoul(cdb->length);
9115 		break;
9116 	}
9117 	default:
9118 		/*
9119 		 * We got a command we don't support.  This shouldn't
9120 		 * happen, commands should be filtered out above us.
9121 		 */
9122 		ctl_set_invalid_opcode(ctsio);
9123 		ctl_done((union ctl_io *)ctsio);
9124 
9125 		return (CTL_RETVAL_COMPLETE);
9126 		break; /* NOTREACHED */
9127 	}
9128 
9129 	/*
9130 	 * XXX KDM what do we do with the DPO and FUA bits?  FUA might be
9131 	 * interesting for us, but if RAIDCore is in write-back mode,
9132 	 * getting it to do write-through for a particular transaction may
9133 	 * not be possible.
9134 	 */
9135 
9136 	/*
9137 	 * The first check is to make sure we're in bounds, the second
9138 	 * check is to catch wrap-around problems.  If the lba + num blocks
9139 	 * is less than the lba, then we've wrapped around and the block
9140 	 * range is invalid anyway.
9141 	 */
9142 	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9143 	 || ((lba + num_blocks) < lba)) {
9144 		ctl_set_lba_out_of_range(ctsio);
9145 		ctl_done((union ctl_io *)ctsio);
9146 		return (CTL_RETVAL_COMPLETE);
9147 	}
9148 
9149 	/*
9150 	 * According to SBC-3, a transfer length of 0 is not an error.
9151 	 * Note that this cannot happen with WRITE(6) or READ(6), since 0
9152 	 * translates to 256 blocks for those commands.
9153 	 */
9154 	if (num_blocks == 0) {
9155 		ctl_set_success(ctsio);
9156 		ctl_done((union ctl_io *)ctsio);
9157 		return (CTL_RETVAL_COMPLETE);
9158 	}
9159 
9160 	lbalen = (struct ctl_lba_len_flags *)
9161 	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9162 	lbalen->lba = lba;
9163 	lbalen->len = num_blocks;
9164 	lbalen->flags = isread ? CTL_LLF_READ : CTL_LLF_WRITE;
9165 
9166 	ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9167 	ctsio->kern_rel_offset = 0;
9168 
9169 	CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
9170 
9171 	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9172 
9173 	return (retval);
9174 }
9175 
9176 static int
9177 ctl_cnw_cont(union ctl_io *io)
9178 {
9179 	struct ctl_scsiio *ctsio;
9180 	struct ctl_lun *lun;
9181 	struct ctl_lba_len_flags *lbalen;
9182 	int retval;
9183 
9184 	ctsio = &io->scsiio;
9185 	ctsio->io_hdr.status = CTL_STATUS_NONE;
9186 	ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
9187 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9188 	lbalen = (struct ctl_lba_len_flags *)
9189 	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9190 	lbalen->flags = CTL_LLF_WRITE;
9191 
9192 	CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
9193 	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9194 	return (retval);
9195 }
9196 
9197 int
9198 ctl_cnw(struct ctl_scsiio *ctsio)
9199 {
9200 	struct ctl_lun *lun;
9201 	struct ctl_lba_len_flags *lbalen;
9202 	uint64_t lba;
9203 	uint32_t num_blocks;
9204 	int fua, dpo;
9205 	int retval;
9206 
9207 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9208 
9209 	CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
9210 
9211 	fua = 0;
9212 	dpo = 0;
9213 
9214 	retval = CTL_RETVAL_COMPLETE;
9215 
9216 	switch (ctsio->cdb[0]) {
9217 	case COMPARE_AND_WRITE: {
9218 		struct scsi_compare_and_write *cdb;
9219 
9220 		cdb = (struct scsi_compare_and_write *)ctsio->cdb;
9221 
9222 		if (cdb->byte2 & SRW10_FUA)
9223 			fua = 1;
9224 		if (cdb->byte2 & SRW10_DPO)
9225 			dpo = 1;
9226 		lba = scsi_8btou64(cdb->addr);
9227 		num_blocks = cdb->length;
9228 		break;
9229 	}
9230 	default:
9231 		/*
9232 		 * We got a command we don't support.  This shouldn't
9233 		 * happen, commands should be filtered out above us.
9234 		 */
9235 		ctl_set_invalid_opcode(ctsio);
9236 		ctl_done((union ctl_io *)ctsio);
9237 
9238 		return (CTL_RETVAL_COMPLETE);
9239 		break; /* NOTREACHED */
9240 	}
9241 
9242 	/*
9243 	 * XXX KDM what do we do with the DPO and FUA bits?  FUA might be
9244 	 * interesting for us, but if RAIDCore is in write-back mode,
9245 	 * getting it to do write-through for a particular transaction may
9246 	 * not be possible.
9247 	 */
9248 
9249 	/*
9250 	 * The first check is to make sure we're in bounds, the second
9251 	 * check is to catch wrap-around problems.  If the lba + num blocks
9252 	 * is less than the lba, then we've wrapped around and the block
9253 	 * range is invalid anyway.
9254 	 */
9255 	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9256 	 || ((lba + num_blocks) < lba)) {
9257 		ctl_set_lba_out_of_range(ctsio);
9258 		ctl_done((union ctl_io *)ctsio);
9259 		return (CTL_RETVAL_COMPLETE);
9260 	}
9261 
9262 	/*
9263 	 * According to SBC-3, a transfer length of 0 is not an error.
9264 	 */
9265 	if (num_blocks == 0) {
9266 		ctl_set_success(ctsio);
9267 		ctl_done((union ctl_io *)ctsio);
9268 		return (CTL_RETVAL_COMPLETE);
9269 	}
9270 
9271 	ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
9272 	ctsio->kern_rel_offset = 0;
9273 
9274 	/*
9275 	 * Set the IO_CONT flag, so that if this I/O gets passed to
9276 	 * ctl_data_submit_done(), it'll get passed back to
9277 	 * ctl_ctl_cnw_cont() for further processing.
9278 	 */
9279 	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
9280 	ctsio->io_cont = ctl_cnw_cont;
9281 
9282 	lbalen = (struct ctl_lba_len_flags *)
9283 	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9284 	lbalen->lba = lba;
9285 	lbalen->len = num_blocks;
9286 	lbalen->flags = CTL_LLF_COMPARE;
9287 
9288 	CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
9289 	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9290 	return (retval);
9291 }
9292 
9293 int
9294 ctl_verify(struct ctl_scsiio *ctsio)
9295 {
9296 	struct ctl_lun *lun;
9297 	struct ctl_lba_len_flags *lbalen;
9298 	uint64_t lba;
9299 	uint32_t num_blocks;
9300 	int bytchk, dpo;
9301 	int retval;
9302 
9303 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9304 
9305 	CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
9306 
9307 	bytchk = 0;
9308 	dpo = 0;
9309 	retval = CTL_RETVAL_COMPLETE;
9310 
9311 	switch (ctsio->cdb[0]) {
9312 	case VERIFY_10: {
9313 		struct scsi_verify_10 *cdb;
9314 
9315 		cdb = (struct scsi_verify_10 *)ctsio->cdb;
9316 		if (cdb->byte2 & SVFY_BYTCHK)
9317 			bytchk = 1;
9318 		if (cdb->byte2 & SVFY_DPO)
9319 			dpo = 1;
9320 		lba = scsi_4btoul(cdb->addr);
9321 		num_blocks = scsi_2btoul(cdb->length);
9322 		break;
9323 	}
9324 	case VERIFY_12: {
9325 		struct scsi_verify_12 *cdb;
9326 
9327 		cdb = (struct scsi_verify_12 *)ctsio->cdb;
9328 		if (cdb->byte2 & SVFY_BYTCHK)
9329 			bytchk = 1;
9330 		if (cdb->byte2 & SVFY_DPO)
9331 			dpo = 1;
9332 		lba = scsi_4btoul(cdb->addr);
9333 		num_blocks = scsi_4btoul(cdb->length);
9334 		break;
9335 	}
9336 	case VERIFY_16: {
9337 		struct scsi_rw_16 *cdb;
9338 
9339 		cdb = (struct scsi_rw_16 *)ctsio->cdb;
9340 		if (cdb->byte2 & SVFY_BYTCHK)
9341 			bytchk = 1;
9342 		if (cdb->byte2 & SVFY_DPO)
9343 			dpo = 1;
9344 		lba = scsi_8btou64(cdb->addr);
9345 		num_blocks = scsi_4btoul(cdb->length);
9346 		break;
9347 	}
9348 	default:
9349 		/*
9350 		 * We got a command we don't support.  This shouldn't
9351 		 * happen, commands should be filtered out above us.
9352 		 */
9353 		ctl_set_invalid_opcode(ctsio);
9354 		ctl_done((union ctl_io *)ctsio);
9355 		return (CTL_RETVAL_COMPLETE);
9356 	}
9357 
9358 	/*
9359 	 * The first check is to make sure we're in bounds, the second
9360 	 * check is to catch wrap-around problems.  If the lba + num blocks
9361 	 * is less than the lba, then we've wrapped around and the block
9362 	 * range is invalid anyway.
9363 	 */
9364 	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9365 	 || ((lba + num_blocks) < lba)) {
9366 		ctl_set_lba_out_of_range(ctsio);
9367 		ctl_done((union ctl_io *)ctsio);
9368 		return (CTL_RETVAL_COMPLETE);
9369 	}
9370 
9371 	/*
9372 	 * According to SBC-3, a transfer length of 0 is not an error.
9373 	 */
9374 	if (num_blocks == 0) {
9375 		ctl_set_success(ctsio);
9376 		ctl_done((union ctl_io *)ctsio);
9377 		return (CTL_RETVAL_COMPLETE);
9378 	}
9379 
9380 	lbalen = (struct ctl_lba_len_flags *)
9381 	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9382 	lbalen->lba = lba;
9383 	lbalen->len = num_blocks;
9384 	if (bytchk) {
9385 		lbalen->flags = CTL_LLF_COMPARE;
9386 		ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9387 	} else {
9388 		lbalen->flags = CTL_LLF_VERIFY;
9389 		ctsio->kern_total_len = 0;
9390 	}
9391 	ctsio->kern_rel_offset = 0;
9392 
9393 	CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9394 	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9395 	return (retval);
9396 }
9397 
9398 int
9399 ctl_report_luns(struct ctl_scsiio *ctsio)
9400 {
9401 	struct scsi_report_luns *cdb;
9402 	struct scsi_report_luns_data *lun_data;
9403 	struct ctl_lun *lun, *request_lun;
9404 	int num_luns, retval;
9405 	uint32_t alloc_len, lun_datalen;
9406 	int num_filled, well_known;
9407 	uint32_t initidx, targ_lun_id, lun_id;
9408 
9409 	retval = CTL_RETVAL_COMPLETE;
9410 	well_known = 0;
9411 
9412 	cdb = (struct scsi_report_luns *)ctsio->cdb;
9413 
9414 	CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9415 
9416 	mtx_lock(&control_softc->ctl_lock);
9417 	num_luns = control_softc->num_luns;
9418 	mtx_unlock(&control_softc->ctl_lock);
9419 
9420 	switch (cdb->select_report) {
9421 	case RPL_REPORT_DEFAULT:
9422 	case RPL_REPORT_ALL:
9423 		break;
9424 	case RPL_REPORT_WELLKNOWN:
9425 		well_known = 1;
9426 		num_luns = 0;
9427 		break;
9428 	default:
9429 		ctl_set_invalid_field(ctsio,
9430 				      /*sks_valid*/ 1,
9431 				      /*command*/ 1,
9432 				      /*field*/ 2,
9433 				      /*bit_valid*/ 0,
9434 				      /*bit*/ 0);
9435 		ctl_done((union ctl_io *)ctsio);
9436 		return (retval);
9437 		break; /* NOTREACHED */
9438 	}
9439 
9440 	alloc_len = scsi_4btoul(cdb->length);
9441 	/*
9442 	 * The initiator has to allocate at least 16 bytes for this request,
9443 	 * so he can at least get the header and the first LUN.  Otherwise
9444 	 * we reject the request (per SPC-3 rev 14, section 6.21).
9445 	 */
9446 	if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9447 	    sizeof(struct scsi_report_luns_lundata))) {
9448 		ctl_set_invalid_field(ctsio,
9449 				      /*sks_valid*/ 1,
9450 				      /*command*/ 1,
9451 				      /*field*/ 6,
9452 				      /*bit_valid*/ 0,
9453 				      /*bit*/ 0);
9454 		ctl_done((union ctl_io *)ctsio);
9455 		return (retval);
9456 	}
9457 
9458 	request_lun = (struct ctl_lun *)
9459 		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9460 
9461 	lun_datalen = sizeof(*lun_data) +
9462 		(num_luns * sizeof(struct scsi_report_luns_lundata));
9463 
9464 	ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9465 	lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9466 	ctsio->kern_sg_entries = 0;
9467 
9468 	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9469 
9470 	mtx_lock(&control_softc->ctl_lock);
9471 	for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
9472 		lun_id = ctl_map_lun(ctsio->io_hdr.nexus.targ_port, targ_lun_id);
9473 		if (lun_id >= CTL_MAX_LUNS)
9474 			continue;
9475 		lun = control_softc->ctl_luns[lun_id];
9476 		if (lun == NULL)
9477 			continue;
9478 
9479 		if (targ_lun_id <= 0xff) {
9480 			/*
9481 			 * Peripheral addressing method, bus number 0.
9482 			 */
9483 			lun_data->luns[num_filled].lundata[0] =
9484 				RPL_LUNDATA_ATYP_PERIPH;
9485 			lun_data->luns[num_filled].lundata[1] = targ_lun_id;
9486 			num_filled++;
9487 		} else if (targ_lun_id <= 0x3fff) {
9488 			/*
9489 			 * Flat addressing method.
9490 			 */
9491 			lun_data->luns[num_filled].lundata[0] =
9492 				RPL_LUNDATA_ATYP_FLAT |
9493 				(targ_lun_id & RPL_LUNDATA_FLAT_LUN_MASK);
9494 #ifdef OLDCTLHEADERS
9495 				(SRLD_ADDR_FLAT << SRLD_ADDR_SHIFT) |
9496 				(targ_lun_id & SRLD_BUS_LUN_MASK);
9497 #endif
9498 			lun_data->luns[num_filled].lundata[1] =
9499 #ifdef OLDCTLHEADERS
9500 				targ_lun_id >> SRLD_BUS_LUN_BITS;
9501 #endif
9502 				targ_lun_id >> RPL_LUNDATA_FLAT_LUN_BITS;
9503 			num_filled++;
9504 		} else {
9505 			printf("ctl_report_luns: bogus LUN number %jd, "
9506 			       "skipping\n", (intmax_t)targ_lun_id);
9507 		}
9508 		/*
9509 		 * According to SPC-3, rev 14 section 6.21:
9510 		 *
9511 		 * "The execution of a REPORT LUNS command to any valid and
9512 		 * installed logical unit shall clear the REPORTED LUNS DATA
9513 		 * HAS CHANGED unit attention condition for all logical
9514 		 * units of that target with respect to the requesting
9515 		 * initiator. A valid and installed logical unit is one
9516 		 * having a PERIPHERAL QUALIFIER of 000b in the standard
9517 		 * INQUIRY data (see 6.4.2)."
9518 		 *
9519 		 * If request_lun is NULL, the LUN this report luns command
9520 		 * was issued to is either disabled or doesn't exist. In that
9521 		 * case, we shouldn't clear any pending lun change unit
9522 		 * attention.
9523 		 */
9524 		if (request_lun != NULL) {
9525 			mtx_lock(&lun->lun_lock);
9526 			lun->pending_ua[initidx] &= ~CTL_UA_LUN_CHANGE;
9527 			mtx_unlock(&lun->lun_lock);
9528 		}
9529 	}
9530 	mtx_unlock(&control_softc->ctl_lock);
9531 
9532 	/*
9533 	 * It's quite possible that we've returned fewer LUNs than we allocated
9534 	 * space for.  Trim it.
9535 	 */
9536 	lun_datalen = sizeof(*lun_data) +
9537 		(num_filled * sizeof(struct scsi_report_luns_lundata));
9538 
9539 	if (lun_datalen < alloc_len) {
9540 		ctsio->residual = alloc_len - lun_datalen;
9541 		ctsio->kern_data_len = lun_datalen;
9542 		ctsio->kern_total_len = lun_datalen;
9543 	} else {
9544 		ctsio->residual = 0;
9545 		ctsio->kern_data_len = alloc_len;
9546 		ctsio->kern_total_len = alloc_len;
9547 	}
9548 	ctsio->kern_data_resid = 0;
9549 	ctsio->kern_rel_offset = 0;
9550 	ctsio->kern_sg_entries = 0;
9551 
9552 	/*
9553 	 * We set this to the actual data length, regardless of how much
9554 	 * space we actually have to return results.  If the user looks at
9555 	 * this value, he'll know whether or not he allocated enough space
9556 	 * and reissue the command if necessary.  We don't support well
9557 	 * known logical units, so if the user asks for that, return none.
9558 	 */
9559 	scsi_ulto4b(lun_datalen - 8, lun_data->length);
9560 
9561 	/*
9562 	 * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9563 	 * this request.
9564 	 */
9565 	ctsio->scsi_status = SCSI_STATUS_OK;
9566 
9567 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9568 	ctsio->be_move_done = ctl_config_move_done;
9569 	ctl_datamove((union ctl_io *)ctsio);
9570 
9571 	return (retval);
9572 }
9573 
9574 int
9575 ctl_request_sense(struct ctl_scsiio *ctsio)
9576 {
9577 	struct scsi_request_sense *cdb;
9578 	struct scsi_sense_data *sense_ptr;
9579 	struct ctl_lun *lun;
9580 	uint32_t initidx;
9581 	int have_error;
9582 	scsi_sense_data_type sense_format;
9583 
9584 	cdb = (struct scsi_request_sense *)ctsio->cdb;
9585 
9586 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9587 
9588 	CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9589 
9590 	/*
9591 	 * Determine which sense format the user wants.
9592 	 */
9593 	if (cdb->byte2 & SRS_DESC)
9594 		sense_format = SSD_TYPE_DESC;
9595 	else
9596 		sense_format = SSD_TYPE_FIXED;
9597 
9598 	ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9599 	sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9600 	ctsio->kern_sg_entries = 0;
9601 
9602 	/*
9603 	 * struct scsi_sense_data, which is currently set to 256 bytes, is
9604 	 * larger than the largest allowed value for the length field in the
9605 	 * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9606 	 */
9607 	ctsio->residual = 0;
9608 	ctsio->kern_data_len = cdb->length;
9609 	ctsio->kern_total_len = cdb->length;
9610 
9611 	ctsio->kern_data_resid = 0;
9612 	ctsio->kern_rel_offset = 0;
9613 	ctsio->kern_sg_entries = 0;
9614 
9615 	/*
9616 	 * If we don't have a LUN, we don't have any pending sense.
9617 	 */
9618 	if (lun == NULL)
9619 		goto no_sense;
9620 
9621 	have_error = 0;
9622 	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9623 	/*
9624 	 * Check for pending sense, and then for pending unit attentions.
9625 	 * Pending sense gets returned first, then pending unit attentions.
9626 	 */
9627 	mtx_lock(&lun->lun_lock);
9628 #ifdef CTL_WITH_CA
9629 	if (ctl_is_set(lun->have_ca, initidx)) {
9630 		scsi_sense_data_type stored_format;
9631 
9632 		/*
9633 		 * Check to see which sense format was used for the stored
9634 		 * sense data.
9635 		 */
9636 		stored_format = scsi_sense_type(&lun->pending_sense[initidx]);
9637 
9638 		/*
9639 		 * If the user requested a different sense format than the
9640 		 * one we stored, then we need to convert it to the other
9641 		 * format.  If we're going from descriptor to fixed format
9642 		 * sense data, we may lose things in translation, depending
9643 		 * on what options were used.
9644 		 *
9645 		 * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9646 		 * for some reason we'll just copy it out as-is.
9647 		 */
9648 		if ((stored_format == SSD_TYPE_FIXED)
9649 		 && (sense_format == SSD_TYPE_DESC))
9650 			ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9651 			    &lun->pending_sense[initidx],
9652 			    (struct scsi_sense_data_desc *)sense_ptr);
9653 		else if ((stored_format == SSD_TYPE_DESC)
9654 		      && (sense_format == SSD_TYPE_FIXED))
9655 			ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9656 			    &lun->pending_sense[initidx],
9657 			    (struct scsi_sense_data_fixed *)sense_ptr);
9658 		else
9659 			memcpy(sense_ptr, &lun->pending_sense[initidx],
9660 			       ctl_min(sizeof(*sense_ptr),
9661 			       sizeof(lun->pending_sense[initidx])));
9662 
9663 		ctl_clear_mask(lun->have_ca, initidx);
9664 		have_error = 1;
9665 	} else
9666 #endif
9667 	if (lun->pending_ua[initidx] != CTL_UA_NONE) {
9668 		ctl_ua_type ua_type;
9669 
9670 		ua_type = ctl_build_ua(lun->pending_ua[initidx],
9671 				       sense_ptr, sense_format);
9672 		if (ua_type != CTL_UA_NONE) {
9673 			have_error = 1;
9674 			/* We're reporting this UA, so clear it */
9675 			lun->pending_ua[initidx] &= ~ua_type;
9676 		}
9677 	}
9678 	mtx_unlock(&lun->lun_lock);
9679 
9680 	/*
9681 	 * We already have a pending error, return it.
9682 	 */
9683 	if (have_error != 0) {
9684 		/*
9685 		 * We report the SCSI status as OK, since the status of the
9686 		 * request sense command itself is OK.
9687 		 */
9688 		ctsio->scsi_status = SCSI_STATUS_OK;
9689 
9690 		/*
9691 		 * We report 0 for the sense length, because we aren't doing
9692 		 * autosense in this case.  We're reporting sense as
9693 		 * parameter data.
9694 		 */
9695 		ctsio->sense_len = 0;
9696 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9697 		ctsio->be_move_done = ctl_config_move_done;
9698 		ctl_datamove((union ctl_io *)ctsio);
9699 
9700 		return (CTL_RETVAL_COMPLETE);
9701 	}
9702 
9703 no_sense:
9704 
9705 	/*
9706 	 * No sense information to report, so we report that everything is
9707 	 * okay.
9708 	 */
9709 	ctl_set_sense_data(sense_ptr,
9710 			   lun,
9711 			   sense_format,
9712 			   /*current_error*/ 1,
9713 			   /*sense_key*/ SSD_KEY_NO_SENSE,
9714 			   /*asc*/ 0x00,
9715 			   /*ascq*/ 0x00,
9716 			   SSD_ELEM_NONE);
9717 
9718 	ctsio->scsi_status = SCSI_STATUS_OK;
9719 
9720 	/*
9721 	 * We report 0 for the sense length, because we aren't doing
9722 	 * autosense in this case.  We're reporting sense as parameter data.
9723 	 */
9724 	ctsio->sense_len = 0;
9725 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9726 	ctsio->be_move_done = ctl_config_move_done;
9727 	ctl_datamove((union ctl_io *)ctsio);
9728 
9729 	return (CTL_RETVAL_COMPLETE);
9730 }
9731 
9732 int
9733 ctl_tur(struct ctl_scsiio *ctsio)
9734 {
9735 	struct ctl_lun *lun;
9736 
9737 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9738 
9739 	CTL_DEBUG_PRINT(("ctl_tur\n"));
9740 
9741 	if (lun == NULL)
9742 		return (EINVAL);
9743 
9744 	ctsio->scsi_status = SCSI_STATUS_OK;
9745 	ctsio->io_hdr.status = CTL_SUCCESS;
9746 
9747 	ctl_done((union ctl_io *)ctsio);
9748 
9749 	return (CTL_RETVAL_COMPLETE);
9750 }
9751 
9752 #ifdef notyet
9753 static int
9754 ctl_cmddt_inquiry(struct ctl_scsiio *ctsio)
9755 {
9756 
9757 }
9758 #endif
9759 
9760 static int
9761 ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9762 {
9763 	struct scsi_vpd_supported_pages *pages;
9764 	int sup_page_size;
9765 	struct ctl_lun *lun;
9766 
9767 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9768 
9769 	sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9770 	    SCSI_EVPD_NUM_SUPPORTED_PAGES;
9771 	ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9772 	pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9773 	ctsio->kern_sg_entries = 0;
9774 
9775 	if (sup_page_size < alloc_len) {
9776 		ctsio->residual = alloc_len - sup_page_size;
9777 		ctsio->kern_data_len = sup_page_size;
9778 		ctsio->kern_total_len = sup_page_size;
9779 	} else {
9780 		ctsio->residual = 0;
9781 		ctsio->kern_data_len = alloc_len;
9782 		ctsio->kern_total_len = alloc_len;
9783 	}
9784 	ctsio->kern_data_resid = 0;
9785 	ctsio->kern_rel_offset = 0;
9786 	ctsio->kern_sg_entries = 0;
9787 
9788 	/*
9789 	 * The control device is always connected.  The disk device, on the
9790 	 * other hand, may not be online all the time.  Need to change this
9791 	 * to figure out whether the disk device is actually online or not.
9792 	 */
9793 	if (lun != NULL)
9794 		pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9795 				lun->be_lun->lun_type;
9796 	else
9797 		pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9798 
9799 	pages->length = SCSI_EVPD_NUM_SUPPORTED_PAGES;
9800 	/* Supported VPD pages */
9801 	pages->page_list[0] = SVPD_SUPPORTED_PAGES;
9802 	/* Serial Number */
9803 	pages->page_list[1] = SVPD_UNIT_SERIAL_NUMBER;
9804 	/* Device Identification */
9805 	pages->page_list[2] = SVPD_DEVICE_ID;
9806 	/* SCSI Ports */
9807 	pages->page_list[3] = SVPD_SCSI_PORTS;
9808 	/* Third-party Copy */
9809 	pages->page_list[4] = SVPD_SCSI_TPC;
9810 	/* Block limits */
9811 	pages->page_list[5] = SVPD_BLOCK_LIMITS;
9812 	/* Logical Block Provisioning */
9813 	pages->page_list[6] = SVPD_LBP;
9814 
9815 	ctsio->scsi_status = SCSI_STATUS_OK;
9816 
9817 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9818 	ctsio->be_move_done = ctl_config_move_done;
9819 	ctl_datamove((union ctl_io *)ctsio);
9820 
9821 	return (CTL_RETVAL_COMPLETE);
9822 }
9823 
9824 static int
9825 ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9826 {
9827 	struct scsi_vpd_unit_serial_number *sn_ptr;
9828 	struct ctl_lun *lun;
9829 
9830 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9831 
9832 	ctsio->kern_data_ptr = malloc(sizeof(*sn_ptr), M_CTL, M_WAITOK | M_ZERO);
9833 	sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9834 	ctsio->kern_sg_entries = 0;
9835 
9836 	if (sizeof(*sn_ptr) < alloc_len) {
9837 		ctsio->residual = alloc_len - sizeof(*sn_ptr);
9838 		ctsio->kern_data_len = sizeof(*sn_ptr);
9839 		ctsio->kern_total_len = sizeof(*sn_ptr);
9840 	} else {
9841 		ctsio->residual = 0;
9842 		ctsio->kern_data_len = alloc_len;
9843 		ctsio->kern_total_len = alloc_len;
9844 	}
9845 	ctsio->kern_data_resid = 0;
9846 	ctsio->kern_rel_offset = 0;
9847 	ctsio->kern_sg_entries = 0;
9848 
9849 	/*
9850 	 * The control device is always connected.  The disk device, on the
9851 	 * other hand, may not be online all the time.  Need to change this
9852 	 * to figure out whether the disk device is actually online or not.
9853 	 */
9854 	if (lun != NULL)
9855 		sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9856 				  lun->be_lun->lun_type;
9857 	else
9858 		sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9859 
9860 	sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9861 	sn_ptr->length = ctl_min(sizeof(*sn_ptr) - 4, CTL_SN_LEN);
9862 	/*
9863 	 * If we don't have a LUN, we just leave the serial number as
9864 	 * all spaces.
9865 	 */
9866 	memset(sn_ptr->serial_num, 0x20, sizeof(sn_ptr->serial_num));
9867 	if (lun != NULL) {
9868 		strncpy((char *)sn_ptr->serial_num,
9869 			(char *)lun->be_lun->serial_num, CTL_SN_LEN);
9870 	}
9871 	ctsio->scsi_status = SCSI_STATUS_OK;
9872 
9873 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9874 	ctsio->be_move_done = ctl_config_move_done;
9875 	ctl_datamove((union ctl_io *)ctsio);
9876 
9877 	return (CTL_RETVAL_COMPLETE);
9878 }
9879 
9880 
9881 static int
9882 ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9883 {
9884 	struct scsi_vpd_device_id *devid_ptr;
9885 	struct scsi_vpd_id_descriptor *desc;
9886 	struct ctl_softc *ctl_softc;
9887 	struct ctl_lun *lun;
9888 	struct ctl_port *port;
9889 	int data_len;
9890 	uint8_t proto;
9891 
9892 	ctl_softc = control_softc;
9893 
9894 	port = ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)];
9895 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9896 
9897 	data_len = sizeof(struct scsi_vpd_device_id) +
9898 	    sizeof(struct scsi_vpd_id_descriptor) +
9899 		sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9900 	    sizeof(struct scsi_vpd_id_descriptor) +
9901 		sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9902 	if (lun && lun->lun_devid)
9903 		data_len += lun->lun_devid->len;
9904 	if (port->port_devid)
9905 		data_len += port->port_devid->len;
9906 	if (port->target_devid)
9907 		data_len += port->target_devid->len;
9908 
9909 	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9910 	devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9911 	ctsio->kern_sg_entries = 0;
9912 
9913 	if (data_len < alloc_len) {
9914 		ctsio->residual = alloc_len - data_len;
9915 		ctsio->kern_data_len = data_len;
9916 		ctsio->kern_total_len = data_len;
9917 	} else {
9918 		ctsio->residual = 0;
9919 		ctsio->kern_data_len = alloc_len;
9920 		ctsio->kern_total_len = alloc_len;
9921 	}
9922 	ctsio->kern_data_resid = 0;
9923 	ctsio->kern_rel_offset = 0;
9924 	ctsio->kern_sg_entries = 0;
9925 
9926 	/*
9927 	 * The control device is always connected.  The disk device, on the
9928 	 * other hand, may not be online all the time.
9929 	 */
9930 	if (lun != NULL)
9931 		devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9932 				     lun->be_lun->lun_type;
9933 	else
9934 		devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9935 	devid_ptr->page_code = SVPD_DEVICE_ID;
9936 	scsi_ulto2b(data_len - 4, devid_ptr->length);
9937 
9938 	if (port->port_type == CTL_PORT_FC)
9939 		proto = SCSI_PROTO_FC << 4;
9940 	else if (port->port_type == CTL_PORT_ISCSI)
9941 		proto = SCSI_PROTO_ISCSI << 4;
9942 	else
9943 		proto = SCSI_PROTO_SPI << 4;
9944 	desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9945 
9946 	/*
9947 	 * We're using a LUN association here.  i.e., this device ID is a
9948 	 * per-LUN identifier.
9949 	 */
9950 	if (lun && lun->lun_devid) {
9951 		memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
9952 		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9953 		    lun->lun_devid->len);
9954 	}
9955 
9956 	/*
9957 	 * This is for the WWPN which is a port association.
9958 	 */
9959 	if (port->port_devid) {
9960 		memcpy(desc, port->port_devid->data, port->port_devid->len);
9961 		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9962 		    port->port_devid->len);
9963 	}
9964 
9965 	/*
9966 	 * This is for the Relative Target Port(type 4h) identifier
9967 	 */
9968 	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9969 	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9970 	    SVPD_ID_TYPE_RELTARG;
9971 	desc->length = 4;
9972 	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
9973 	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9974 	    sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9975 
9976 	/*
9977 	 * This is for the Target Port Group(type 5h) identifier
9978 	 */
9979 	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9980 	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9981 	    SVPD_ID_TYPE_TPORTGRP;
9982 	desc->length = 4;
9983 	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS + 1,
9984 	    &desc->identifier[2]);
9985 	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9986 	    sizeof(struct scsi_vpd_id_trgt_port_grp_id));
9987 
9988 	/*
9989 	 * This is for the Target identifier
9990 	 */
9991 	if (port->target_devid) {
9992 		memcpy(desc, port->target_devid->data, port->target_devid->len);
9993 	}
9994 
9995 	ctsio->scsi_status = SCSI_STATUS_OK;
9996 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9997 	ctsio->be_move_done = ctl_config_move_done;
9998 	ctl_datamove((union ctl_io *)ctsio);
9999 
10000 	return (CTL_RETVAL_COMPLETE);
10001 }
10002 
10003 static int
10004 ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
10005 {
10006 	struct ctl_softc *softc = control_softc;
10007 	struct scsi_vpd_scsi_ports *sp;
10008 	struct scsi_vpd_port_designation *pd;
10009 	struct scsi_vpd_port_designation_cont *pdc;
10010 	struct ctl_lun *lun;
10011 	struct ctl_port *port;
10012 	int data_len, num_target_ports, iid_len, id_len, g, pg, p;
10013 	int num_target_port_groups, single;
10014 
10015 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10016 
10017 	single = ctl_is_single;
10018 	if (single)
10019 		num_target_port_groups = 1;
10020 	else
10021 		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
10022 	num_target_ports = 0;
10023 	iid_len = 0;
10024 	id_len = 0;
10025 	mtx_lock(&softc->ctl_lock);
10026 	STAILQ_FOREACH(port, &softc->port_list, links) {
10027 		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
10028 			continue;
10029 		if (lun != NULL &&
10030 		    ctl_map_lun_back(port->targ_port, lun->lun) >=
10031 		    CTL_MAX_LUNS)
10032 			continue;
10033 		num_target_ports++;
10034 		if (port->init_devid)
10035 			iid_len += port->init_devid->len;
10036 		if (port->port_devid)
10037 			id_len += port->port_devid->len;
10038 	}
10039 	mtx_unlock(&softc->ctl_lock);
10040 
10041 	data_len = sizeof(struct scsi_vpd_scsi_ports) + num_target_port_groups *
10042 	    num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
10043 	     sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
10044 	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10045 	sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
10046 	ctsio->kern_sg_entries = 0;
10047 
10048 	if (data_len < alloc_len) {
10049 		ctsio->residual = alloc_len - data_len;
10050 		ctsio->kern_data_len = data_len;
10051 		ctsio->kern_total_len = data_len;
10052 	} else {
10053 		ctsio->residual = 0;
10054 		ctsio->kern_data_len = alloc_len;
10055 		ctsio->kern_total_len = alloc_len;
10056 	}
10057 	ctsio->kern_data_resid = 0;
10058 	ctsio->kern_rel_offset = 0;
10059 	ctsio->kern_sg_entries = 0;
10060 
10061 	/*
10062 	 * The control device is always connected.  The disk device, on the
10063 	 * other hand, may not be online all the time.  Need to change this
10064 	 * to figure out whether the disk device is actually online or not.
10065 	 */
10066 	if (lun != NULL)
10067 		sp->device = (SID_QUAL_LU_CONNECTED << 5) |
10068 				  lun->be_lun->lun_type;
10069 	else
10070 		sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10071 
10072 	sp->page_code = SVPD_SCSI_PORTS;
10073 	scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
10074 	    sp->page_length);
10075 	pd = &sp->design[0];
10076 
10077 	mtx_lock(&softc->ctl_lock);
10078 	if (softc->flags & CTL_FLAG_MASTER_SHELF)
10079 		pg = 0;
10080 	else
10081 		pg = 1;
10082 	for (g = 0; g < num_target_port_groups; g++) {
10083 		STAILQ_FOREACH(port, &softc->port_list, links) {
10084 			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
10085 				continue;
10086 			if (lun != NULL &&
10087 			    ctl_map_lun_back(port->targ_port, lun->lun) >=
10088 			    CTL_MAX_LUNS)
10089 				continue;
10090 			p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
10091 			scsi_ulto2b(p, pd->relative_port_id);
10092 			if (port->init_devid && g == pg) {
10093 				iid_len = port->init_devid->len;
10094 				memcpy(pd->initiator_transportid,
10095 				    port->init_devid->data, port->init_devid->len);
10096 			} else
10097 				iid_len = 0;
10098 			scsi_ulto2b(iid_len, pd->initiator_transportid_length);
10099 			pdc = (struct scsi_vpd_port_designation_cont *)
10100 			    (&pd->initiator_transportid[iid_len]);
10101 			if (port->port_devid && g == pg) {
10102 				id_len = port->port_devid->len;
10103 				memcpy(pdc->target_port_descriptors,
10104 				    port->port_devid->data, port->port_devid->len);
10105 			} else
10106 				id_len = 0;
10107 			scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
10108 			pd = (struct scsi_vpd_port_designation *)
10109 			    ((uint8_t *)pdc->target_port_descriptors + id_len);
10110 		}
10111 	}
10112 	mtx_unlock(&softc->ctl_lock);
10113 
10114 	ctsio->scsi_status = SCSI_STATUS_OK;
10115 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10116 	ctsio->be_move_done = ctl_config_move_done;
10117 	ctl_datamove((union ctl_io *)ctsio);
10118 
10119 	return (CTL_RETVAL_COMPLETE);
10120 }
10121 
10122 static int
10123 ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
10124 {
10125 	struct scsi_vpd_block_limits *bl_ptr;
10126 	struct ctl_lun *lun;
10127 	int bs;
10128 
10129 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10130 
10131 	ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
10132 	bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
10133 	ctsio->kern_sg_entries = 0;
10134 
10135 	if (sizeof(*bl_ptr) < alloc_len) {
10136 		ctsio->residual = alloc_len - sizeof(*bl_ptr);
10137 		ctsio->kern_data_len = sizeof(*bl_ptr);
10138 		ctsio->kern_total_len = sizeof(*bl_ptr);
10139 	} else {
10140 		ctsio->residual = 0;
10141 		ctsio->kern_data_len = alloc_len;
10142 		ctsio->kern_total_len = alloc_len;
10143 	}
10144 	ctsio->kern_data_resid = 0;
10145 	ctsio->kern_rel_offset = 0;
10146 	ctsio->kern_sg_entries = 0;
10147 
10148 	/*
10149 	 * The control device is always connected.  The disk device, on the
10150 	 * other hand, may not be online all the time.  Need to change this
10151 	 * to figure out whether the disk device is actually online or not.
10152 	 */
10153 	if (lun != NULL)
10154 		bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10155 				  lun->be_lun->lun_type;
10156 	else
10157 		bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10158 
10159 	bl_ptr->page_code = SVPD_BLOCK_LIMITS;
10160 	scsi_ulto2b(sizeof(*bl_ptr), bl_ptr->page_length);
10161 	bl_ptr->max_cmp_write_len = 0xff;
10162 	scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
10163 	if (lun != NULL) {
10164 		bs = lun->be_lun->blocksize;
10165 		scsi_ulto4b(MAXPHYS / bs, bl_ptr->opt_txfer_len);
10166 		if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10167 			scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_lba_cnt);
10168 			scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_blk_cnt);
10169 		}
10170 	}
10171 	scsi_u64to8b(UINT64_MAX, bl_ptr->max_write_same_length);
10172 
10173 	ctsio->scsi_status = SCSI_STATUS_OK;
10174 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10175 	ctsio->be_move_done = ctl_config_move_done;
10176 	ctl_datamove((union ctl_io *)ctsio);
10177 
10178 	return (CTL_RETVAL_COMPLETE);
10179 }
10180 
10181 static int
10182 ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
10183 {
10184 	struct scsi_vpd_logical_block_prov *lbp_ptr;
10185 	struct ctl_lun *lun;
10186 
10187 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10188 
10189 	ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
10190 	lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
10191 	ctsio->kern_sg_entries = 0;
10192 
10193 	if (sizeof(*lbp_ptr) < alloc_len) {
10194 		ctsio->residual = alloc_len - sizeof(*lbp_ptr);
10195 		ctsio->kern_data_len = sizeof(*lbp_ptr);
10196 		ctsio->kern_total_len = sizeof(*lbp_ptr);
10197 	} else {
10198 		ctsio->residual = 0;
10199 		ctsio->kern_data_len = alloc_len;
10200 		ctsio->kern_total_len = alloc_len;
10201 	}
10202 	ctsio->kern_data_resid = 0;
10203 	ctsio->kern_rel_offset = 0;
10204 	ctsio->kern_sg_entries = 0;
10205 
10206 	/*
10207 	 * The control device is always connected.  The disk device, on the
10208 	 * other hand, may not be online all the time.  Need to change this
10209 	 * to figure out whether the disk device is actually online or not.
10210 	 */
10211 	if (lun != NULL)
10212 		lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10213 				  lun->be_lun->lun_type;
10214 	else
10215 		lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10216 
10217 	lbp_ptr->page_code = SVPD_LBP;
10218 	if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
10219 		lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 | SVPD_LBP_WS10;
10220 
10221 	ctsio->scsi_status = SCSI_STATUS_OK;
10222 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10223 	ctsio->be_move_done = ctl_config_move_done;
10224 	ctl_datamove((union ctl_io *)ctsio);
10225 
10226 	return (CTL_RETVAL_COMPLETE);
10227 }
10228 
10229 static int
10230 ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
10231 {
10232 	struct scsi_inquiry *cdb;
10233 	struct ctl_lun *lun;
10234 	int alloc_len, retval;
10235 
10236 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10237 	cdb = (struct scsi_inquiry *)ctsio->cdb;
10238 
10239 	retval = CTL_RETVAL_COMPLETE;
10240 
10241 	alloc_len = scsi_2btoul(cdb->length);
10242 
10243 	switch (cdb->page_code) {
10244 	case SVPD_SUPPORTED_PAGES:
10245 		retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
10246 		break;
10247 	case SVPD_UNIT_SERIAL_NUMBER:
10248 		retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
10249 		break;
10250 	case SVPD_DEVICE_ID:
10251 		retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
10252 		break;
10253 	case SVPD_SCSI_PORTS:
10254 		retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
10255 		break;
10256 	case SVPD_SCSI_TPC:
10257 		retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
10258 		break;
10259 	case SVPD_BLOCK_LIMITS:
10260 		retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
10261 		break;
10262 	case SVPD_LBP:
10263 		retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
10264 		break;
10265 	default:
10266 		ctl_set_invalid_field(ctsio,
10267 				      /*sks_valid*/ 1,
10268 				      /*command*/ 1,
10269 				      /*field*/ 2,
10270 				      /*bit_valid*/ 0,
10271 				      /*bit*/ 0);
10272 		ctl_done((union ctl_io *)ctsio);
10273 		retval = CTL_RETVAL_COMPLETE;
10274 		break;
10275 	}
10276 
10277 	return (retval);
10278 }
10279 
10280 static int
10281 ctl_inquiry_std(struct ctl_scsiio *ctsio)
10282 {
10283 	struct scsi_inquiry_data *inq_ptr;
10284 	struct scsi_inquiry *cdb;
10285 	struct ctl_softc *ctl_softc;
10286 	struct ctl_lun *lun;
10287 	char *val;
10288 	uint32_t alloc_len;
10289 	ctl_port_type port_type;
10290 
10291 	ctl_softc = control_softc;
10292 
10293 	/*
10294 	 * Figure out whether we're talking to a Fibre Channel port or not.
10295 	 * We treat the ioctl front end, and any SCSI adapters, as packetized
10296 	 * SCSI front ends.
10297 	 */
10298 	port_type = ctl_softc->ctl_ports[
10299 	    ctl_port_idx(ctsio->io_hdr.nexus.targ_port)]->port_type;
10300 	if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
10301 		port_type = CTL_PORT_SCSI;
10302 
10303 	lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10304 	cdb = (struct scsi_inquiry *)ctsio->cdb;
10305 	alloc_len = scsi_2btoul(cdb->length);
10306 
10307 	/*
10308 	 * We malloc the full inquiry data size here and fill it
10309 	 * in.  If the user only asks for less, we'll give him
10310 	 * that much.
10311 	 */
10312 	ctsio->kern_data_ptr = malloc(sizeof(*inq_ptr), M_CTL, M_WAITOK | M_ZERO);
10313 	inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
10314 	ctsio->kern_sg_entries = 0;
10315 	ctsio->kern_data_resid = 0;
10316 	ctsio->kern_rel_offset = 0;
10317 
10318 	if (sizeof(*inq_ptr) < alloc_len) {
10319 		ctsio->residual = alloc_len - sizeof(*inq_ptr);
10320 		ctsio->kern_data_len = sizeof(*inq_ptr);
10321 		ctsio->kern_total_len = sizeof(*inq_ptr);
10322 	} else {
10323 		ctsio->residual = 0;
10324 		ctsio->kern_data_len = alloc_len;
10325 		ctsio->kern_total_len = alloc_len;
10326 	}
10327 
10328 	/*
10329 	 * If we have a LUN configured, report it as connected.  Otherwise,
10330 	 * report that it is offline or no device is supported, depending
10331 	 * on the value of inquiry_pq_no_lun.
10332 	 *
10333 	 * According to the spec (SPC-4 r34), the peripheral qualifier
10334 	 * SID_QUAL_LU_OFFLINE (001b) is used in the following scenario:
10335 	 *
10336 	 * "A peripheral device having the specified peripheral device type
10337 	 * is not connected to this logical unit. However, the device
10338 	 * server is capable of supporting the specified peripheral device
10339 	 * type on this logical unit."
10340 	 *
10341 	 * According to the same spec, the peripheral qualifier
10342 	 * SID_QUAL_BAD_LU (011b) is used in this scenario:
10343 	 *
10344 	 * "The device server is not capable of supporting a peripheral
10345 	 * device on this logical unit. For this peripheral qualifier the
10346 	 * peripheral device type shall be set to 1Fh. All other peripheral
10347 	 * device type values are reserved for this peripheral qualifier."
10348 	 *
10349 	 * Given the text, it would seem that we probably want to report that
10350 	 * the LUN is offline here.  There is no LUN connected, but we can
10351 	 * support a LUN at the given LUN number.
10352 	 *
10353 	 * In the real world, though, it sounds like things are a little
10354 	 * different:
10355 	 *
10356 	 * - Linux, when presented with a LUN with the offline peripheral
10357 	 *   qualifier, will create an sg driver instance for it.  So when
10358 	 *   you attach it to CTL, you wind up with a ton of sg driver
10359 	 *   instances.  (One for every LUN that Linux bothered to probe.)
10360 	 *   Linux does this despite the fact that it issues a REPORT LUNs
10361 	 *   to LUN 0 to get the inventory of supported LUNs.
10362 	 *
10363 	 * - There is other anecdotal evidence (from Emulex folks) about
10364 	 *   arrays that use the offline peripheral qualifier for LUNs that
10365 	 *   are on the "passive" path in an active/passive array.
10366 	 *
10367 	 * So the solution is provide a hopefully reasonable default
10368 	 * (return bad/no LUN) and allow the user to change the behavior
10369 	 * with a tunable/sysctl variable.
10370 	 */
10371 	if (lun != NULL)
10372 		inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10373 				  lun->be_lun->lun_type;
10374 	else if (ctl_softc->inquiry_pq_no_lun == 0)
10375 		inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10376 	else
10377 		inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10378 
10379 	/* RMB in byte 2 is 0 */
10380 	inq_ptr->version = SCSI_REV_SPC4;
10381 
10382 	/*
10383 	 * According to SAM-3, even if a device only supports a single
10384 	 * level of LUN addressing, it should still set the HISUP bit:
10385 	 *
10386 	 * 4.9.1 Logical unit numbers overview
10387 	 *
10388 	 * All logical unit number formats described in this standard are
10389 	 * hierarchical in structure even when only a single level in that
10390 	 * hierarchy is used. The HISUP bit shall be set to one in the
10391 	 * standard INQUIRY data (see SPC-2) when any logical unit number
10392 	 * format described in this standard is used.  Non-hierarchical
10393 	 * formats are outside the scope of this standard.
10394 	 *
10395 	 * Therefore we set the HiSup bit here.
10396 	 *
10397 	 * The reponse format is 2, per SPC-3.
10398 	 */
10399 	inq_ptr->response_format = SID_HiSup | 2;
10400 
10401 	inq_ptr->additional_length = sizeof(*inq_ptr) - 4;
10402 	CTL_DEBUG_PRINT(("additional_length = %d\n",
10403 			 inq_ptr->additional_length));
10404 
10405 	inq_ptr->spc3_flags = SPC3_SID_3PC;
10406 	if (!ctl_is_single)
10407 		inq_ptr->spc3_flags |= SPC3_SID_TPGS_IMPLICIT;
10408 	/* 16 bit addressing */
10409 	if (port_type == CTL_PORT_SCSI)
10410 		inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10411 	/* XXX set the SID_MultiP bit here if we're actually going to
10412 	   respond on multiple ports */
10413 	inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10414 
10415 	/* 16 bit data bus, synchronous transfers */
10416 	if (port_type == CTL_PORT_SCSI)
10417 		inq_ptr->flags = SID_WBus16 | SID_Sync;
10418 	/*
10419 	 * XXX KDM do we want to support tagged queueing on the control
10420 	 * device at all?
10421 	 */
10422 	if ((lun == NULL)
10423 	 || (lun->be_lun->lun_type != T_PROCESSOR))
10424 		inq_ptr->flags |= SID_CmdQue;
10425 	/*
10426 	 * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10427 	 * We have 8 bytes for the vendor name, and 16 bytes for the device
10428 	 * name and 4 bytes for the revision.
10429 	 */
10430 	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10431 	    "vendor")) == NULL) {
10432 		strcpy(inq_ptr->vendor, CTL_VENDOR);
10433 	} else {
10434 		memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10435 		strncpy(inq_ptr->vendor, val,
10436 		    min(sizeof(inq_ptr->vendor), strlen(val)));
10437 	}
10438 	if (lun == NULL) {
10439 		strcpy(inq_ptr->product, CTL_DIRECT_PRODUCT);
10440 	} else if ((val = ctl_get_opt(&lun->be_lun->options, "product")) == NULL) {
10441 		switch (lun->be_lun->lun_type) {
10442 		case T_DIRECT:
10443 			strcpy(inq_ptr->product, CTL_DIRECT_PRODUCT);
10444 			break;
10445 		case T_PROCESSOR:
10446 			strcpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT);
10447 			break;
10448 		default:
10449 			strcpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT);
10450 			break;
10451 		}
10452 	} else {
10453 		memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10454 		strncpy(inq_ptr->product, val,
10455 		    min(sizeof(inq_ptr->product), strlen(val)));
10456 	}
10457 
10458 	/*
10459 	 * XXX make this a macro somewhere so it automatically gets
10460 	 * incremented when we make changes.
10461 	 */
10462 	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10463 	    "revision")) == NULL) {
10464 		strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10465 	} else {
10466 		memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10467 		strncpy(inq_ptr->revision, val,
10468 		    min(sizeof(inq_ptr->revision), strlen(val)));
10469 	}
10470 
10471 	/*
10472 	 * For parallel SCSI, we support double transition and single
10473 	 * transition clocking.  We also support QAS (Quick Arbitration
10474 	 * and Selection) and Information Unit transfers on both the
10475 	 * control and array devices.
10476 	 */
10477 	if (port_type == CTL_PORT_SCSI)
10478 		inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10479 				    SID_SPI_IUS;
10480 
10481 	/* SAM-5 (no version claimed) */
10482 	scsi_ulto2b(0x00A0, inq_ptr->version1);
10483 	/* SPC-4 (no version claimed) */
10484 	scsi_ulto2b(0x0460, inq_ptr->version2);
10485 	if (port_type == CTL_PORT_FC) {
10486 		/* FCP-2 ANSI INCITS.350:2003 */
10487 		scsi_ulto2b(0x0917, inq_ptr->version3);
10488 	} else if (port_type == CTL_PORT_SCSI) {
10489 		/* SPI-4 ANSI INCITS.362:200x */
10490 		scsi_ulto2b(0x0B56, inq_ptr->version3);
10491 	} else if (port_type == CTL_PORT_ISCSI) {
10492 		/* iSCSI (no version claimed) */
10493 		scsi_ulto2b(0x0960, inq_ptr->version3);
10494 	} else if (port_type == CTL_PORT_SAS) {
10495 		/* SAS (no version claimed) */
10496 		scsi_ulto2b(0x0BE0, inq_ptr->version3);
10497 	}
10498 
10499 	if (lun == NULL) {
10500 		/* SBC-3 (no version claimed) */
10501 		scsi_ulto2b(0x04C0, inq_ptr->version4);
10502 	} else {
10503 		switch (lun->be_lun->lun_type) {
10504 		case T_DIRECT:
10505 			/* SBC-3 (no version claimed) */
10506 			scsi_ulto2b(0x04C0, inq_ptr->version4);
10507 			break;
10508 		case T_PROCESSOR:
10509 		default:
10510 			break;
10511 		}
10512 	}
10513 
10514 	ctsio->scsi_status = SCSI_STATUS_OK;
10515 	if (ctsio->kern_data_len > 0) {
10516 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10517 		ctsio->be_move_done = ctl_config_move_done;
10518 		ctl_datamove((union ctl_io *)ctsio);
10519 	} else {
10520 		ctsio->io_hdr.status = CTL_SUCCESS;
10521 		ctl_done((union ctl_io *)ctsio);
10522 	}
10523 
10524 	return (CTL_RETVAL_COMPLETE);
10525 }
10526 
10527 int
10528 ctl_inquiry(struct ctl_scsiio *ctsio)
10529 {
10530 	struct scsi_inquiry *cdb;
10531 	int retval;
10532 
10533 	cdb = (struct scsi_inquiry *)ctsio->cdb;
10534 
10535 	retval = 0;
10536 
10537 	CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10538 
10539 	/*
10540 	 * Right now, we don't support the CmdDt inquiry information.
10541 	 * This would be nice to support in the future.  When we do
10542 	 * support it, we should change this test so that it checks to make
10543 	 * sure SI_EVPD and SI_CMDDT aren't both set at the same time.
10544 	 */
10545 #ifdef notyet
10546 	if (((cdb->byte2 & SI_EVPD)
10547 	 && (cdb->byte2 & SI_CMDDT)))
10548 #endif
10549 	if (cdb->byte2 & SI_CMDDT) {
10550 		/*
10551 		 * Point to the SI_CMDDT bit.  We might change this
10552 		 * when we support SI_CMDDT, but since both bits would be
10553 		 * "wrong", this should probably just stay as-is then.
10554 		 */
10555 		ctl_set_invalid_field(ctsio,
10556 				      /*sks_valid*/ 1,
10557 				      /*command*/ 1,
10558 				      /*field*/ 1,
10559 				      /*bit_valid*/ 1,
10560 				      /*bit*/ 1);
10561 		ctl_done((union ctl_io *)ctsio);
10562 		return (CTL_RETVAL_COMPLETE);
10563 	}
10564 	if (cdb->byte2 & SI_EVPD)
10565 		retval = ctl_inquiry_evpd(ctsio);
10566 #ifdef notyet
10567 	else if (cdb->byte2 & SI_CMDDT)
10568 		retval = ctl_inquiry_cmddt(ctsio);
10569 #endif
10570 	else
10571 		retval = ctl_inquiry_std(ctsio);
10572 
10573 	return (retval);
10574 }
10575 
10576 /*
10577  * For known CDB types, parse the LBA and length.
10578  */
10579 static int
10580 ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint32_t *len)
10581 {
10582 	if (io->io_hdr.io_type != CTL_IO_SCSI)
10583 		return (1);
10584 
10585 	switch (io->scsiio.cdb[0]) {
10586 	case COMPARE_AND_WRITE: {
10587 		struct scsi_compare_and_write *cdb;
10588 
10589 		cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10590 
10591 		*lba = scsi_8btou64(cdb->addr);
10592 		*len = cdb->length;
10593 		break;
10594 	}
10595 	case READ_6:
10596 	case WRITE_6: {
10597 		struct scsi_rw_6 *cdb;
10598 
10599 		cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10600 
10601 		*lba = scsi_3btoul(cdb->addr);
10602 		/* only 5 bits are valid in the most significant address byte */
10603 		*lba &= 0x1fffff;
10604 		*len = cdb->length;
10605 		break;
10606 	}
10607 	case READ_10:
10608 	case WRITE_10: {
10609 		struct scsi_rw_10 *cdb;
10610 
10611 		cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10612 
10613 		*lba = scsi_4btoul(cdb->addr);
10614 		*len = scsi_2btoul(cdb->length);
10615 		break;
10616 	}
10617 	case WRITE_VERIFY_10: {
10618 		struct scsi_write_verify_10 *cdb;
10619 
10620 		cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10621 
10622 		*lba = scsi_4btoul(cdb->addr);
10623 		*len = scsi_2btoul(cdb->length);
10624 		break;
10625 	}
10626 	case READ_12:
10627 	case WRITE_12: {
10628 		struct scsi_rw_12 *cdb;
10629 
10630 		cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10631 
10632 		*lba = scsi_4btoul(cdb->addr);
10633 		*len = scsi_4btoul(cdb->length);
10634 		break;
10635 	}
10636 	case WRITE_VERIFY_12: {
10637 		struct scsi_write_verify_12 *cdb;
10638 
10639 		cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10640 
10641 		*lba = scsi_4btoul(cdb->addr);
10642 		*len = scsi_4btoul(cdb->length);
10643 		break;
10644 	}
10645 	case READ_16:
10646 	case WRITE_16: {
10647 		struct scsi_rw_16 *cdb;
10648 
10649 		cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10650 
10651 		*lba = scsi_8btou64(cdb->addr);
10652 		*len = scsi_4btoul(cdb->length);
10653 		break;
10654 	}
10655 	case WRITE_VERIFY_16: {
10656 		struct scsi_write_verify_16 *cdb;
10657 
10658 		cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10659 
10660 
10661 		*lba = scsi_8btou64(cdb->addr);
10662 		*len = scsi_4btoul(cdb->length);
10663 		break;
10664 	}
10665 	case WRITE_SAME_10: {
10666 		struct scsi_write_same_10 *cdb;
10667 
10668 		cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10669 
10670 		*lba = scsi_4btoul(cdb->addr);
10671 		*len = scsi_2btoul(cdb->length);
10672 		break;
10673 	}
10674 	case WRITE_SAME_16: {
10675 		struct scsi_write_same_16 *cdb;
10676 
10677 		cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10678 
10679 		*lba = scsi_8btou64(cdb->addr);
10680 		*len = scsi_4btoul(cdb->length);
10681 		break;
10682 	}
10683 	case VERIFY_10: {
10684 		struct scsi_verify_10 *cdb;
10685 
10686 		cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10687 
10688 		*lba = scsi_4btoul(cdb->addr);
10689 		*len = scsi_2btoul(cdb->length);
10690 		break;
10691 	}
10692 	case VERIFY_12: {
10693 		struct scsi_verify_12 *cdb;
10694 
10695 		cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10696 
10697 		*lba = scsi_4btoul(cdb->addr);
10698 		*len = scsi_4btoul(cdb->length);
10699 		break;
10700 	}
10701 	case VERIFY_16: {
10702 		struct scsi_verify_16 *cdb;
10703 
10704 		cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10705 
10706 		*lba = scsi_8btou64(cdb->addr);
10707 		*len = scsi_4btoul(cdb->length);
10708 		break;
10709 	}
10710 	default:
10711 		return (1);
10712 		break; /* NOTREACHED */
10713 	}
10714 
10715 	return (0);
10716 }
10717 
10718 static ctl_action
10719 ctl_extent_check_lba(uint64_t lba1, uint32_t len1, uint64_t lba2, uint32_t len2)
10720 {
10721 	uint64_t endlba1, endlba2;
10722 
10723 	endlba1 = lba1 + len1 - 1;
10724 	endlba2 = lba2 + len2 - 1;
10725 
10726 	if ((endlba1 < lba2)
10727 	 || (endlba2 < lba1))
10728 		return (CTL_ACTION_PASS);
10729 	else
10730 		return (CTL_ACTION_BLOCK);
10731 }
10732 
10733 static ctl_action
10734 ctl_extent_check(union ctl_io *io1, union ctl_io *io2)
10735 {
10736 	uint64_t lba1, lba2;
10737 	uint32_t len1, len2;
10738 	int retval;
10739 
10740 	retval = ctl_get_lba_len(io1, &lba1, &len1);
10741 	if (retval != 0)
10742 		return (CTL_ACTION_ERROR);
10743 
10744 	retval = ctl_get_lba_len(io2, &lba2, &len2);
10745 	if (retval != 0)
10746 		return (CTL_ACTION_ERROR);
10747 
10748 	return (ctl_extent_check_lba(lba1, len1, lba2, len2));
10749 }
10750 
10751 static ctl_action
10752 ctl_check_for_blockage(union ctl_io *pending_io, union ctl_io *ooa_io)
10753 {
10754 	const struct ctl_cmd_entry *pending_entry, *ooa_entry;
10755 	ctl_serialize_action *serialize_row;
10756 
10757 	/*
10758 	 * The initiator attempted multiple untagged commands at the same
10759 	 * time.  Can't do that.
10760 	 */
10761 	if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10762 	 && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10763 	 && ((pending_io->io_hdr.nexus.targ_port ==
10764 	      ooa_io->io_hdr.nexus.targ_port)
10765 	  && (pending_io->io_hdr.nexus.initid.id ==
10766 	      ooa_io->io_hdr.nexus.initid.id))
10767 	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10768 		return (CTL_ACTION_OVERLAP);
10769 
10770 	/*
10771 	 * The initiator attempted to send multiple tagged commands with
10772 	 * the same ID.  (It's fine if different initiators have the same
10773 	 * tag ID.)
10774 	 *
10775 	 * Even if all of those conditions are true, we don't kill the I/O
10776 	 * if the command ahead of us has been aborted.  We won't end up
10777 	 * sending it to the FETD, and it's perfectly legal to resend a
10778 	 * command with the same tag number as long as the previous
10779 	 * instance of this tag number has been aborted somehow.
10780 	 */
10781 	if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10782 	 && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10783 	 && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
10784 	 && ((pending_io->io_hdr.nexus.targ_port ==
10785 	      ooa_io->io_hdr.nexus.targ_port)
10786 	  && (pending_io->io_hdr.nexus.initid.id ==
10787 	      ooa_io->io_hdr.nexus.initid.id))
10788 	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10789 		return (CTL_ACTION_OVERLAP_TAG);
10790 
10791 	/*
10792 	 * If we get a head of queue tag, SAM-3 says that we should
10793 	 * immediately execute it.
10794 	 *
10795 	 * What happens if this command would normally block for some other
10796 	 * reason?  e.g. a request sense with a head of queue tag
10797 	 * immediately after a write.  Normally that would block, but this
10798 	 * will result in its getting executed immediately...
10799 	 *
10800 	 * We currently return "pass" instead of "skip", so we'll end up
10801 	 * going through the rest of the queue to check for overlapped tags.
10802 	 *
10803 	 * XXX KDM check for other types of blockage first??
10804 	 */
10805 	if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10806 		return (CTL_ACTION_PASS);
10807 
10808 	/*
10809 	 * Ordered tags have to block until all items ahead of them
10810 	 * have completed.  If we get called with an ordered tag, we always
10811 	 * block, if something else is ahead of us in the queue.
10812 	 */
10813 	if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
10814 		return (CTL_ACTION_BLOCK);
10815 
10816 	/*
10817 	 * Simple tags get blocked until all head of queue and ordered tags
10818 	 * ahead of them have completed.  I'm lumping untagged commands in
10819 	 * with simple tags here.  XXX KDM is that the right thing to do?
10820 	 */
10821 	if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10822 	  || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
10823 	 && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10824 	  || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
10825 		return (CTL_ACTION_BLOCK);
10826 
10827 	pending_entry = ctl_get_cmd_entry(&pending_io->scsiio);
10828 	ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio);
10829 
10830 	serialize_row = ctl_serialize_table[ooa_entry->seridx];
10831 
10832 	switch (serialize_row[pending_entry->seridx]) {
10833 	case CTL_SER_BLOCK:
10834 		return (CTL_ACTION_BLOCK);
10835 		break; /* NOTREACHED */
10836 	case CTL_SER_EXTENT:
10837 		return (ctl_extent_check(pending_io, ooa_io));
10838 		break; /* NOTREACHED */
10839 	case CTL_SER_PASS:
10840 		return (CTL_ACTION_PASS);
10841 		break; /* NOTREACHED */
10842 	case CTL_SER_SKIP:
10843 		return (CTL_ACTION_SKIP);
10844 		break;
10845 	default:
10846 		panic("invalid serialization value %d",
10847 		      serialize_row[pending_entry->seridx]);
10848 		break; /* NOTREACHED */
10849 	}
10850 
10851 	return (CTL_ACTION_ERROR);
10852 }
10853 
10854 /*
10855  * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
10856  * Assumptions:
10857  * - pending_io is generally either incoming, or on the blocked queue
10858  * - starting I/O is the I/O we want to start the check with.
10859  */
10860 static ctl_action
10861 ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
10862 	      union ctl_io *starting_io)
10863 {
10864 	union ctl_io *ooa_io;
10865 	ctl_action action;
10866 
10867 	mtx_assert(&lun->lun_lock, MA_OWNED);
10868 
10869 	/*
10870 	 * Run back along the OOA queue, starting with the current
10871 	 * blocked I/O and going through every I/O before it on the
10872 	 * queue.  If starting_io is NULL, we'll just end up returning
10873 	 * CTL_ACTION_PASS.
10874 	 */
10875 	for (ooa_io = starting_io; ooa_io != NULL;
10876 	     ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
10877 	     ooa_links)){
10878 
10879 		/*
10880 		 * This routine just checks to see whether
10881 		 * cur_blocked is blocked by ooa_io, which is ahead
10882 		 * of it in the queue.  It doesn't queue/dequeue
10883 		 * cur_blocked.
10884 		 */
10885 		action = ctl_check_for_blockage(pending_io, ooa_io);
10886 		switch (action) {
10887 		case CTL_ACTION_BLOCK:
10888 		case CTL_ACTION_OVERLAP:
10889 		case CTL_ACTION_OVERLAP_TAG:
10890 		case CTL_ACTION_SKIP:
10891 		case CTL_ACTION_ERROR:
10892 			return (action);
10893 			break; /* NOTREACHED */
10894 		case CTL_ACTION_PASS:
10895 			break;
10896 		default:
10897 			panic("invalid action %d", action);
10898 			break;  /* NOTREACHED */
10899 		}
10900 	}
10901 
10902 	return (CTL_ACTION_PASS);
10903 }
10904 
10905 /*
10906  * Assumptions:
10907  * - An I/O has just completed, and has been removed from the per-LUN OOA
10908  *   queue, so some items on the blocked queue may now be unblocked.
10909  */
10910 static int
10911 ctl_check_blocked(struct ctl_lun *lun)
10912 {
10913 	union ctl_io *cur_blocked, *next_blocked;
10914 
10915 	mtx_assert(&lun->lun_lock, MA_OWNED);
10916 
10917 	/*
10918 	 * Run forward from the head of the blocked queue, checking each
10919 	 * entry against the I/Os prior to it on the OOA queue to see if
10920 	 * there is still any blockage.
10921 	 *
10922 	 * We cannot use the TAILQ_FOREACH() macro, because it can't deal
10923 	 * with our removing a variable on it while it is traversing the
10924 	 * list.
10925 	 */
10926 	for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
10927 	     cur_blocked != NULL; cur_blocked = next_blocked) {
10928 		union ctl_io *prev_ooa;
10929 		ctl_action action;
10930 
10931 		next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
10932 							  blocked_links);
10933 
10934 		prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
10935 						      ctl_ooaq, ooa_links);
10936 
10937 		/*
10938 		 * If cur_blocked happens to be the first item in the OOA
10939 		 * queue now, prev_ooa will be NULL, and the action
10940 		 * returned will just be CTL_ACTION_PASS.
10941 		 */
10942 		action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
10943 
10944 		switch (action) {
10945 		case CTL_ACTION_BLOCK:
10946 			/* Nothing to do here, still blocked */
10947 			break;
10948 		case CTL_ACTION_OVERLAP:
10949 		case CTL_ACTION_OVERLAP_TAG:
10950 			/*
10951 			 * This shouldn't happen!  In theory we've already
10952 			 * checked this command for overlap...
10953 			 */
10954 			break;
10955 		case CTL_ACTION_PASS:
10956 		case CTL_ACTION_SKIP: {
10957 			struct ctl_softc *softc;
10958 			const struct ctl_cmd_entry *entry;
10959 			uint32_t initidx;
10960 			int isc_retval;
10961 
10962 			/*
10963 			 * The skip case shouldn't happen, this transaction
10964 			 * should have never made it onto the blocked queue.
10965 			 */
10966 			/*
10967 			 * This I/O is no longer blocked, we can remove it
10968 			 * from the blocked queue.  Since this is a TAILQ
10969 			 * (doubly linked list), we can do O(1) removals
10970 			 * from any place on the list.
10971 			 */
10972 			TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
10973 				     blocked_links);
10974 			cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
10975 
10976 			if (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC){
10977 				/*
10978 				 * Need to send IO back to original side to
10979 				 * run
10980 				 */
10981 				union ctl_ha_msg msg_info;
10982 
10983 				msg_info.hdr.original_sc =
10984 					cur_blocked->io_hdr.original_sc;
10985 				msg_info.hdr.serializing_sc = cur_blocked;
10986 				msg_info.hdr.msg_type = CTL_MSG_R2R;
10987 				if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
10988 				     &msg_info, sizeof(msg_info), 0)) >
10989 				     CTL_HA_STATUS_SUCCESS) {
10990 					printf("CTL:Check Blocked error from "
10991 					       "ctl_ha_msg_send %d\n",
10992 					       isc_retval);
10993 				}
10994 				break;
10995 			}
10996 			entry = ctl_get_cmd_entry(&cur_blocked->scsiio);
10997 			softc = control_softc;
10998 
10999 			initidx = ctl_get_initindex(&cur_blocked->io_hdr.nexus);
11000 
11001 			/*
11002 			 * Check this I/O for LUN state changes that may
11003 			 * have happened while this command was blocked.
11004 			 * The LUN state may have been changed by a command
11005 			 * ahead of us in the queue, so we need to re-check
11006 			 * for any states that can be caused by SCSI
11007 			 * commands.
11008 			 */
11009 			if (ctl_scsiio_lun_check(softc, lun, entry,
11010 						 &cur_blocked->scsiio) == 0) {
11011 				cur_blocked->io_hdr.flags |=
11012 				                      CTL_FLAG_IS_WAS_ON_RTR;
11013 				ctl_enqueue_rtr(cur_blocked);
11014 			} else
11015 				ctl_done(cur_blocked);
11016 			break;
11017 		}
11018 		default:
11019 			/*
11020 			 * This probably shouldn't happen -- we shouldn't
11021 			 * get CTL_ACTION_ERROR, or anything else.
11022 			 */
11023 			break;
11024 		}
11025 	}
11026 
11027 	return (CTL_RETVAL_COMPLETE);
11028 }
11029 
11030 /*
11031  * This routine (with one exception) checks LUN flags that can be set by
11032  * commands ahead of us in the OOA queue.  These flags have to be checked
11033  * when a command initially comes in, and when we pull a command off the
11034  * blocked queue and are preparing to execute it.  The reason we have to
11035  * check these flags for commands on the blocked queue is that the LUN
11036  * state may have been changed by a command ahead of us while we're on the
11037  * blocked queue.
11038  *
11039  * Ordering is somewhat important with these checks, so please pay
11040  * careful attention to the placement of any new checks.
11041  */
11042 static int
11043 ctl_scsiio_lun_check(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
11044     const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
11045 {
11046 	int retval;
11047 
11048 	retval = 0;
11049 
11050 	mtx_assert(&lun->lun_lock, MA_OWNED);
11051 
11052 	/*
11053 	 * If this shelf is a secondary shelf controller, we have to reject
11054 	 * any media access commands.
11055 	 */
11056 #if 0
11057 	/* No longer needed for HA */
11058 	if (((ctl_softc->flags & CTL_FLAG_MASTER_SHELF) == 0)
11059 	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_SECONDARY) == 0)) {
11060 		ctl_set_lun_standby(ctsio);
11061 		retval = 1;
11062 		goto bailout;
11063 	}
11064 #endif
11065 
11066 	/*
11067 	 * Check for a reservation conflict.  If this command isn't allowed
11068 	 * even on reserved LUNs, and if this initiator isn't the one who
11069 	 * reserved us, reject the command with a reservation conflict.
11070 	 */
11071 	if ((lun->flags & CTL_LUN_RESERVED)
11072 	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
11073 		if ((ctsio->io_hdr.nexus.initid.id != lun->rsv_nexus.initid.id)
11074 		 || (ctsio->io_hdr.nexus.targ_port != lun->rsv_nexus.targ_port)
11075 		 || (ctsio->io_hdr.nexus.targ_target.id !=
11076 		     lun->rsv_nexus.targ_target.id)) {
11077 			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
11078 			ctsio->io_hdr.status = CTL_SCSI_ERROR;
11079 			retval = 1;
11080 			goto bailout;
11081 		}
11082 	}
11083 
11084 	if ( (lun->flags & CTL_LUN_PR_RESERVED)
11085 	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV) == 0)) {
11086 		uint32_t residx;
11087 
11088 		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
11089 		/*
11090 		 * if we aren't registered or it's a res holder type
11091 		 * reservation and this isn't the res holder then set a
11092 		 * conflict.
11093 		 * NOTE: Commands which might be allowed on write exclusive
11094 		 * type reservations are checked in the particular command
11095 		 * for a conflict. Read and SSU are the only ones.
11096 		 */
11097 		if (!lun->per_res[residx].registered
11098 		 || (residx != lun->pr_res_idx && lun->res_type < 4)) {
11099 			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
11100 			ctsio->io_hdr.status = CTL_SCSI_ERROR;
11101 			retval = 1;
11102 			goto bailout;
11103 		}
11104 
11105 	}
11106 
11107 	if ((lun->flags & CTL_LUN_OFFLINE)
11108 	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_OFFLINE) == 0)) {
11109 		ctl_set_lun_not_ready(ctsio);
11110 		retval = 1;
11111 		goto bailout;
11112 	}
11113 
11114 	/*
11115 	 * If the LUN is stopped, see if this particular command is allowed
11116 	 * for a stopped lun.  Otherwise, reject it with 0x04,0x02.
11117 	 */
11118 	if ((lun->flags & CTL_LUN_STOPPED)
11119 	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_STOPPED) == 0)) {
11120 		/* "Logical unit not ready, initializing cmd. required" */
11121 		ctl_set_lun_stopped(ctsio);
11122 		retval = 1;
11123 		goto bailout;
11124 	}
11125 
11126 	if ((lun->flags & CTL_LUN_INOPERABLE)
11127 	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_INOPERABLE) == 0)) {
11128 		/* "Medium format corrupted" */
11129 		ctl_set_medium_format_corrupted(ctsio);
11130 		retval = 1;
11131 		goto bailout;
11132 	}
11133 
11134 bailout:
11135 	return (retval);
11136 
11137 }
11138 
11139 static void
11140 ctl_failover_io(union ctl_io *io, int have_lock)
11141 {
11142 	ctl_set_busy(&io->scsiio);
11143 	ctl_done(io);
11144 }
11145 
11146 static void
11147 ctl_failover(void)
11148 {
11149 	struct ctl_lun *lun;
11150 	struct ctl_softc *ctl_softc;
11151 	union ctl_io *next_io, *pending_io;
11152 	union ctl_io *io;
11153 	int lun_idx;
11154 	int i;
11155 
11156 	ctl_softc = control_softc;
11157 
11158 	mtx_lock(&ctl_softc->ctl_lock);
11159 	/*
11160 	 * Remove any cmds from the other SC from the rtr queue.  These
11161 	 * will obviously only be for LUNs for which we're the primary.
11162 	 * We can't send status or get/send data for these commands.
11163 	 * Since they haven't been executed yet, we can just remove them.
11164 	 * We'll either abort them or delete them below, depending on
11165 	 * which HA mode we're in.
11166 	 */
11167 #ifdef notyet
11168 	mtx_lock(&ctl_softc->queue_lock);
11169 	for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->rtr_queue);
11170 	     io != NULL; io = next_io) {
11171 		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
11172 		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
11173 			STAILQ_REMOVE(&ctl_softc->rtr_queue, &io->io_hdr,
11174 				      ctl_io_hdr, links);
11175 	}
11176 	mtx_unlock(&ctl_softc->queue_lock);
11177 #endif
11178 
11179 	for (lun_idx=0; lun_idx < ctl_softc->num_luns; lun_idx++) {
11180 		lun = ctl_softc->ctl_luns[lun_idx];
11181 		if (lun==NULL)
11182 			continue;
11183 
11184 		/*
11185 		 * Processor LUNs are primary on both sides.
11186 		 * XXX will this always be true?
11187 		 */
11188 		if (lun->be_lun->lun_type == T_PROCESSOR)
11189 			continue;
11190 
11191 		if ((lun->flags & CTL_LUN_PRIMARY_SC)
11192 		 && (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11193 			printf("FAILOVER: primary lun %d\n", lun_idx);
11194 		        /*
11195 			 * Remove all commands from the other SC. First from the
11196 			 * blocked queue then from the ooa queue. Once we have
11197 			 * removed them. Call ctl_check_blocked to see if there
11198 			 * is anything that can run.
11199 			 */
11200 			for (io = (union ctl_io *)TAILQ_FIRST(
11201 			     &lun->blocked_queue); io != NULL; io = next_io) {
11202 
11203 		        	next_io = (union ctl_io *)TAILQ_NEXT(
11204 				    &io->io_hdr, blocked_links);
11205 
11206 				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11207 					TAILQ_REMOVE(&lun->blocked_queue,
11208 						     &io->io_hdr,blocked_links);
11209 					io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11210 					TAILQ_REMOVE(&lun->ooa_queue,
11211 						     &io->io_hdr, ooa_links);
11212 
11213 					ctl_free_io(io);
11214 				}
11215 			}
11216 
11217 			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11218 	     		     io != NULL; io = next_io) {
11219 
11220 		        	next_io = (union ctl_io *)TAILQ_NEXT(
11221 				    &io->io_hdr, ooa_links);
11222 
11223 				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11224 
11225 					TAILQ_REMOVE(&lun->ooa_queue,
11226 						&io->io_hdr,
11227 					     	ooa_links);
11228 
11229 					ctl_free_io(io);
11230 				}
11231 			}
11232 			ctl_check_blocked(lun);
11233 		} else if ((lun->flags & CTL_LUN_PRIMARY_SC)
11234 			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
11235 
11236 			printf("FAILOVER: primary lun %d\n", lun_idx);
11237 			/*
11238 			 * Abort all commands from the other SC.  We can't
11239 			 * send status back for them now.  These should get
11240 			 * cleaned up when they are completed or come out
11241 			 * for a datamove operation.
11242 			 */
11243 			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11244 	     		     io != NULL; io = next_io) {
11245 		        	next_io = (union ctl_io *)TAILQ_NEXT(
11246 					&io->io_hdr, ooa_links);
11247 
11248 				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
11249 					io->io_hdr.flags |= CTL_FLAG_ABORT;
11250 			}
11251 		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11252 			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
11253 
11254 			printf("FAILOVER: secondary lun %d\n", lun_idx);
11255 
11256 			lun->flags |= CTL_LUN_PRIMARY_SC;
11257 
11258 			/*
11259 			 * We send all I/O that was sent to this controller
11260 			 * and redirected to the other side back with
11261 			 * busy status, and have the initiator retry it.
11262 			 * Figuring out how much data has been transferred,
11263 			 * etc. and picking up where we left off would be
11264 			 * very tricky.
11265 			 *
11266 			 * XXX KDM need to remove I/O from the blocked
11267 			 * queue as well!
11268 			 */
11269 			for (pending_io = (union ctl_io *)TAILQ_FIRST(
11270 			     &lun->ooa_queue); pending_io != NULL;
11271 			     pending_io = next_io) {
11272 
11273 				next_io =  (union ctl_io *)TAILQ_NEXT(
11274 					&pending_io->io_hdr, ooa_links);
11275 
11276 				pending_io->io_hdr.flags &=
11277 					~CTL_FLAG_SENT_2OTHER_SC;
11278 
11279 				if (pending_io->io_hdr.flags &
11280 				    CTL_FLAG_IO_ACTIVE) {
11281 					pending_io->io_hdr.flags |=
11282 						CTL_FLAG_FAILOVER;
11283 				} else {
11284 					ctl_set_busy(&pending_io->scsiio);
11285 					ctl_done(pending_io);
11286 				}
11287 			}
11288 
11289 			/*
11290 			 * Build Unit Attention
11291 			 */
11292 			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11293 				lun->pending_ua[i] |=
11294 				                     CTL_UA_ASYM_ACC_CHANGE;
11295 			}
11296 		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11297 			&& (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11298 			printf("FAILOVER: secondary lun %d\n", lun_idx);
11299 			/*
11300 			 * if the first io on the OOA is not on the RtR queue
11301 			 * add it.
11302 			 */
11303 			lun->flags |= CTL_LUN_PRIMARY_SC;
11304 
11305 			pending_io = (union ctl_io *)TAILQ_FIRST(
11306 			    &lun->ooa_queue);
11307 			if (pending_io==NULL) {
11308 				printf("Nothing on OOA queue\n");
11309 				continue;
11310 			}
11311 
11312 			pending_io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11313 			if ((pending_io->io_hdr.flags &
11314 			     CTL_FLAG_IS_WAS_ON_RTR) == 0) {
11315 				pending_io->io_hdr.flags |=
11316 				    CTL_FLAG_IS_WAS_ON_RTR;
11317 				ctl_enqueue_rtr(pending_io);
11318 			}
11319 #if 0
11320 			else
11321 			{
11322 				printf("Tag 0x%04x is running\n",
11323 				      pending_io->scsiio.tag_num);
11324 			}
11325 #endif
11326 
11327 			next_io = (union ctl_io *)TAILQ_NEXT(
11328 			    &pending_io->io_hdr, ooa_links);
11329 			for (pending_io=next_io; pending_io != NULL;
11330 			     pending_io = next_io) {
11331 				pending_io->io_hdr.flags &=
11332 				    ~CTL_FLAG_SENT_2OTHER_SC;
11333 				next_io = (union ctl_io *)TAILQ_NEXT(
11334 					&pending_io->io_hdr, ooa_links);
11335 				if (pending_io->io_hdr.flags &
11336 				    CTL_FLAG_IS_WAS_ON_RTR) {
11337 #if 0
11338 				        printf("Tag 0x%04x is running\n",
11339 				      		pending_io->scsiio.tag_num);
11340 #endif
11341 					continue;
11342 				}
11343 
11344 				switch (ctl_check_ooa(lun, pending_io,
11345 			            (union ctl_io *)TAILQ_PREV(
11346 				    &pending_io->io_hdr, ctl_ooaq,
11347 				    ooa_links))) {
11348 
11349 				case CTL_ACTION_BLOCK:
11350 					TAILQ_INSERT_TAIL(&lun->blocked_queue,
11351 							  &pending_io->io_hdr,
11352 							  blocked_links);
11353 					pending_io->io_hdr.flags |=
11354 					    CTL_FLAG_BLOCKED;
11355 					break;
11356 				case CTL_ACTION_PASS:
11357 				case CTL_ACTION_SKIP:
11358 					pending_io->io_hdr.flags |=
11359 					    CTL_FLAG_IS_WAS_ON_RTR;
11360 					ctl_enqueue_rtr(pending_io);
11361 					break;
11362 				case CTL_ACTION_OVERLAP:
11363 					ctl_set_overlapped_cmd(
11364 					    (struct ctl_scsiio *)pending_io);
11365 					ctl_done(pending_io);
11366 					break;
11367 				case CTL_ACTION_OVERLAP_TAG:
11368 					ctl_set_overlapped_tag(
11369 					    (struct ctl_scsiio *)pending_io,
11370 					    pending_io->scsiio.tag_num & 0xff);
11371 					ctl_done(pending_io);
11372 					break;
11373 				case CTL_ACTION_ERROR:
11374 				default:
11375 					ctl_set_internal_failure(
11376 						(struct ctl_scsiio *)pending_io,
11377 						0,  // sks_valid
11378 						0); //retry count
11379 					ctl_done(pending_io);
11380 					break;
11381 				}
11382 			}
11383 
11384 			/*
11385 			 * Build Unit Attention
11386 			 */
11387 			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11388 				lun->pending_ua[i] |=
11389 				                     CTL_UA_ASYM_ACC_CHANGE;
11390 			}
11391 		} else {
11392 			panic("Unhandled HA mode failover, LUN flags = %#x, "
11393 			      "ha_mode = #%x", lun->flags, ctl_softc->ha_mode);
11394 		}
11395 	}
11396 	ctl_pause_rtr = 0;
11397 	mtx_unlock(&ctl_softc->ctl_lock);
11398 }
11399 
11400 static int
11401 ctl_scsiio_precheck(struct ctl_softc *ctl_softc, struct ctl_scsiio *ctsio)
11402 {
11403 	struct ctl_lun *lun;
11404 	const struct ctl_cmd_entry *entry;
11405 	uint32_t initidx, targ_lun;
11406 	int retval;
11407 
11408 	retval = 0;
11409 
11410 	lun = NULL;
11411 
11412 	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11413 	if ((targ_lun < CTL_MAX_LUNS)
11414 	 && (ctl_softc->ctl_luns[targ_lun] != NULL)) {
11415 		lun = ctl_softc->ctl_luns[targ_lun];
11416 		/*
11417 		 * If the LUN is invalid, pretend that it doesn't exist.
11418 		 * It will go away as soon as all pending I/O has been
11419 		 * completed.
11420 		 */
11421 		if (lun->flags & CTL_LUN_DISABLED) {
11422 			lun = NULL;
11423 		} else {
11424 			ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
11425 			ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
11426 				lun->be_lun;
11427 			if (lun->be_lun->lun_type == T_PROCESSOR) {
11428 				ctsio->io_hdr.flags |= CTL_FLAG_CONTROL_DEV;
11429 			}
11430 
11431 			/*
11432 			 * Every I/O goes into the OOA queue for a
11433 			 * particular LUN, and stays there until completion.
11434 			 */
11435 			mtx_lock(&lun->lun_lock);
11436 			TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr,
11437 			    ooa_links);
11438 		}
11439 	} else {
11440 		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11441 		ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11442 	}
11443 
11444 	/* Get command entry and return error if it is unsuppotyed. */
11445 	entry = ctl_validate_command(ctsio);
11446 	if (entry == NULL) {
11447 		if (lun)
11448 			mtx_unlock(&lun->lun_lock);
11449 		return (retval);
11450 	}
11451 
11452 	ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11453 	ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11454 
11455 	/*
11456 	 * Check to see whether we can send this command to LUNs that don't
11457 	 * exist.  This should pretty much only be the case for inquiry
11458 	 * and request sense.  Further checks, below, really require having
11459 	 * a LUN, so we can't really check the command anymore.  Just put
11460 	 * it on the rtr queue.
11461 	 */
11462 	if (lun == NULL) {
11463 		if (entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) {
11464 			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11465 			ctl_enqueue_rtr((union ctl_io *)ctsio);
11466 			return (retval);
11467 		}
11468 
11469 		ctl_set_unsupported_lun(ctsio);
11470 		ctl_done((union ctl_io *)ctsio);
11471 		CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11472 		return (retval);
11473 	} else {
11474 		/*
11475 		 * Make sure we support this particular command on this LUN.
11476 		 * e.g., we don't support writes to the control LUN.
11477 		 */
11478 		if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11479 			mtx_unlock(&lun->lun_lock);
11480 			ctl_set_invalid_opcode(ctsio);
11481 			ctl_done((union ctl_io *)ctsio);
11482 			return (retval);
11483 		}
11484 	}
11485 
11486 	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11487 
11488 #ifdef CTL_WITH_CA
11489 	/*
11490 	 * If we've got a request sense, it'll clear the contingent
11491 	 * allegiance condition.  Otherwise, if we have a CA condition for
11492 	 * this initiator, clear it, because it sent down a command other
11493 	 * than request sense.
11494 	 */
11495 	if ((ctsio->cdb[0] != REQUEST_SENSE)
11496 	 && (ctl_is_set(lun->have_ca, initidx)))
11497 		ctl_clear_mask(lun->have_ca, initidx);
11498 #endif
11499 
11500 	/*
11501 	 * If the command has this flag set, it handles its own unit
11502 	 * attention reporting, we shouldn't do anything.  Otherwise we
11503 	 * check for any pending unit attentions, and send them back to the
11504 	 * initiator.  We only do this when a command initially comes in,
11505 	 * not when we pull it off the blocked queue.
11506 	 *
11507 	 * According to SAM-3, section 5.3.2, the order that things get
11508 	 * presented back to the host is basically unit attentions caused
11509 	 * by some sort of reset event, busy status, reservation conflicts
11510 	 * or task set full, and finally any other status.
11511 	 *
11512 	 * One issue here is that some of the unit attentions we report
11513 	 * don't fall into the "reset" category (e.g. "reported luns data
11514 	 * has changed").  So reporting it here, before the reservation
11515 	 * check, may be technically wrong.  I guess the only thing to do
11516 	 * would be to check for and report the reset events here, and then
11517 	 * check for the other unit attention types after we check for a
11518 	 * reservation conflict.
11519 	 *
11520 	 * XXX KDM need to fix this
11521 	 */
11522 	if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11523 		ctl_ua_type ua_type;
11524 
11525 		ua_type = lun->pending_ua[initidx];
11526 		if (ua_type != CTL_UA_NONE) {
11527 			scsi_sense_data_type sense_format;
11528 
11529 			if (lun != NULL)
11530 				sense_format = (lun->flags &
11531 				    CTL_LUN_SENSE_DESC) ? SSD_TYPE_DESC :
11532 				    SSD_TYPE_FIXED;
11533 			else
11534 				sense_format = SSD_TYPE_FIXED;
11535 
11536 			ua_type = ctl_build_ua(ua_type, &ctsio->sense_data,
11537 					       sense_format);
11538 			if (ua_type != CTL_UA_NONE) {
11539 				ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11540 				ctsio->io_hdr.status = CTL_SCSI_ERROR |
11541 						       CTL_AUTOSENSE;
11542 				ctsio->sense_len = SSD_FULL_SIZE;
11543 				lun->pending_ua[initidx] &= ~ua_type;
11544 				mtx_unlock(&lun->lun_lock);
11545 				ctl_done((union ctl_io *)ctsio);
11546 				return (retval);
11547 			}
11548 		}
11549 	}
11550 
11551 
11552 	if (ctl_scsiio_lun_check(ctl_softc, lun, entry, ctsio) != 0) {
11553 		mtx_unlock(&lun->lun_lock);
11554 		ctl_done((union ctl_io *)ctsio);
11555 		return (retval);
11556 	}
11557 
11558 	/*
11559 	 * XXX CHD this is where we want to send IO to other side if
11560 	 * this LUN is secondary on this SC. We will need to make a copy
11561 	 * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11562 	 * the copy we send as FROM_OTHER.
11563 	 * We also need to stuff the address of the original IO so we can
11564 	 * find it easily. Something similar will need be done on the other
11565 	 * side so when we are done we can find the copy.
11566 	 */
11567 	if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11568 		union ctl_ha_msg msg_info;
11569 		int isc_retval;
11570 
11571 		ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11572 
11573 		msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11574 		msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11575 #if 0
11576 		printf("1. ctsio %p\n", ctsio);
11577 #endif
11578 		msg_info.hdr.serializing_sc = NULL;
11579 		msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11580 		msg_info.scsi.tag_num = ctsio->tag_num;
11581 		msg_info.scsi.tag_type = ctsio->tag_type;
11582 		memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11583 
11584 		ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11585 
11586 		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11587 		    (void *)&msg_info, sizeof(msg_info), 0)) >
11588 		    CTL_HA_STATUS_SUCCESS) {
11589 			printf("CTL:precheck, ctl_ha_msg_send returned %d\n",
11590 			       isc_retval);
11591 			printf("CTL:opcode is %x\n", ctsio->cdb[0]);
11592 		} else {
11593 #if 0
11594 			printf("CTL:Precheck sent msg, opcode is %x\n",opcode);
11595 #endif
11596 		}
11597 
11598 		/*
11599 		 * XXX KDM this I/O is off the incoming queue, but hasn't
11600 		 * been inserted on any other queue.  We may need to come
11601 		 * up with a holding queue while we wait for serialization
11602 		 * so that we have an idea of what we're waiting for from
11603 		 * the other side.
11604 		 */
11605 		mtx_unlock(&lun->lun_lock);
11606 		return (retval);
11607 	}
11608 
11609 	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11610 			      (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11611 			      ctl_ooaq, ooa_links))) {
11612 	case CTL_ACTION_BLOCK:
11613 		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11614 		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11615 				  blocked_links);
11616 		mtx_unlock(&lun->lun_lock);
11617 		return (retval);
11618 	case CTL_ACTION_PASS:
11619 	case CTL_ACTION_SKIP:
11620 		ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11621 		mtx_unlock(&lun->lun_lock);
11622 		ctl_enqueue_rtr((union ctl_io *)ctsio);
11623 		break;
11624 	case CTL_ACTION_OVERLAP:
11625 		mtx_unlock(&lun->lun_lock);
11626 		ctl_set_overlapped_cmd(ctsio);
11627 		ctl_done((union ctl_io *)ctsio);
11628 		break;
11629 	case CTL_ACTION_OVERLAP_TAG:
11630 		mtx_unlock(&lun->lun_lock);
11631 		ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11632 		ctl_done((union ctl_io *)ctsio);
11633 		break;
11634 	case CTL_ACTION_ERROR:
11635 	default:
11636 		mtx_unlock(&lun->lun_lock);
11637 		ctl_set_internal_failure(ctsio,
11638 					 /*sks_valid*/ 0,
11639 					 /*retry_count*/ 0);
11640 		ctl_done((union ctl_io *)ctsio);
11641 		break;
11642 	}
11643 	return (retval);
11644 }
11645 
11646 const struct ctl_cmd_entry *
11647 ctl_get_cmd_entry(struct ctl_scsiio *ctsio)
11648 {
11649 	const struct ctl_cmd_entry *entry;
11650 	int service_action;
11651 
11652 	entry = &ctl_cmd_table[ctsio->cdb[0]];
11653 	if (entry->flags & CTL_CMD_FLAG_SA5) {
11654 		service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11655 		entry = &((const struct ctl_cmd_entry *)
11656 		    entry->execute)[service_action];
11657 	}
11658 	return (entry);
11659 }
11660 
11661 const struct ctl_cmd_entry *
11662 ctl_validate_command(struct ctl_scsiio *ctsio)
11663 {
11664 	const struct ctl_cmd_entry *entry;
11665 	int i;
11666 	uint8_t diff;
11667 
11668 	entry = ctl_get_cmd_entry(ctsio);
11669 	if (entry->execute == NULL) {
11670 		ctl_set_invalid_opcode(ctsio);
11671 		ctl_done((union ctl_io *)ctsio);
11672 		return (NULL);
11673 	}
11674 	KASSERT(entry->length > 0,
11675 	    ("Not defined length for command 0x%02x/0x%02x",
11676 	     ctsio->cdb[0], ctsio->cdb[1]));
11677 	for (i = 1; i < entry->length; i++) {
11678 		diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11679 		if (diff == 0)
11680 			continue;
11681 		ctl_set_invalid_field(ctsio,
11682 				      /*sks_valid*/ 1,
11683 				      /*command*/ 1,
11684 				      /*field*/ i,
11685 				      /*bit_valid*/ 1,
11686 				      /*bit*/ fls(diff) - 1);
11687 		ctl_done((union ctl_io *)ctsio);
11688 		return (NULL);
11689 	}
11690 	return (entry);
11691 }
11692 
11693 static int
11694 ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11695 {
11696 
11697 	switch (lun_type) {
11698 	case T_PROCESSOR:
11699 		if (((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0) &&
11700 		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11701 			return (0);
11702 		break;
11703 	case T_DIRECT:
11704 		if (((entry->flags & CTL_CMD_FLAG_OK_ON_SLUN) == 0) &&
11705 		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11706 			return (0);
11707 		break;
11708 	default:
11709 		return (0);
11710 	}
11711 	return (1);
11712 }
11713 
11714 static int
11715 ctl_scsiio(struct ctl_scsiio *ctsio)
11716 {
11717 	int retval;
11718 	const struct ctl_cmd_entry *entry;
11719 
11720 	retval = CTL_RETVAL_COMPLETE;
11721 
11722 	CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11723 
11724 	entry = ctl_get_cmd_entry(ctsio);
11725 
11726 	/*
11727 	 * If this I/O has been aborted, just send it straight to
11728 	 * ctl_done() without executing it.
11729 	 */
11730 	if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11731 		ctl_done((union ctl_io *)ctsio);
11732 		goto bailout;
11733 	}
11734 
11735 	/*
11736 	 * All the checks should have been handled by ctl_scsiio_precheck().
11737 	 * We should be clear now to just execute the I/O.
11738 	 */
11739 	retval = entry->execute(ctsio);
11740 
11741 bailout:
11742 	return (retval);
11743 }
11744 
11745 /*
11746  * Since we only implement one target right now, a bus reset simply resets
11747  * our single target.
11748  */
11749 static int
11750 ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io)
11751 {
11752 	return(ctl_target_reset(ctl_softc, io, CTL_UA_BUS_RESET));
11753 }
11754 
11755 static int
11756 ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
11757 		 ctl_ua_type ua_type)
11758 {
11759 	struct ctl_lun *lun;
11760 	int retval;
11761 
11762 	if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11763 		union ctl_ha_msg msg_info;
11764 
11765 		io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11766 		msg_info.hdr.nexus = io->io_hdr.nexus;
11767 		if (ua_type==CTL_UA_TARG_RESET)
11768 			msg_info.task.task_action = CTL_TASK_TARGET_RESET;
11769 		else
11770 			msg_info.task.task_action = CTL_TASK_BUS_RESET;
11771 		msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11772 		msg_info.hdr.original_sc = NULL;
11773 		msg_info.hdr.serializing_sc = NULL;
11774 		if (CTL_HA_STATUS_SUCCESS != ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11775 		    (void *)&msg_info, sizeof(msg_info), 0)) {
11776 		}
11777 	}
11778 	retval = 0;
11779 
11780 	mtx_lock(&ctl_softc->ctl_lock);
11781 	STAILQ_FOREACH(lun, &ctl_softc->lun_list, links)
11782 		retval += ctl_lun_reset(lun, io, ua_type);
11783 	mtx_unlock(&ctl_softc->ctl_lock);
11784 
11785 	return (retval);
11786 }
11787 
11788 /*
11789  * The LUN should always be set.  The I/O is optional, and is used to
11790  * distinguish between I/Os sent by this initiator, and by other
11791  * initiators.  We set unit attention for initiators other than this one.
11792  * SAM-3 is vague on this point.  It does say that a unit attention should
11793  * be established for other initiators when a LUN is reset (see section
11794  * 5.7.3), but it doesn't specifically say that the unit attention should
11795  * be established for this particular initiator when a LUN is reset.  Here
11796  * is the relevant text, from SAM-3 rev 8:
11797  *
11798  * 5.7.2 When a SCSI initiator port aborts its own tasks
11799  *
11800  * When a SCSI initiator port causes its own task(s) to be aborted, no
11801  * notification that the task(s) have been aborted shall be returned to
11802  * the SCSI initiator port other than the completion response for the
11803  * command or task management function action that caused the task(s) to
11804  * be aborted and notification(s) associated with related effects of the
11805  * action (e.g., a reset unit attention condition).
11806  *
11807  * XXX KDM for now, we're setting unit attention for all initiators.
11808  */
11809 static int
11810 ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
11811 {
11812 	union ctl_io *xio;
11813 #if 0
11814 	uint32_t initindex;
11815 #endif
11816 	int i;
11817 
11818 	mtx_lock(&lun->lun_lock);
11819 	/*
11820 	 * Run through the OOA queue and abort each I/O.
11821 	 */
11822 #if 0
11823 	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
11824 #endif
11825 	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11826 	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11827 		xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
11828 	}
11829 
11830 	/*
11831 	 * This version sets unit attention for every
11832 	 */
11833 #if 0
11834 	initindex = ctl_get_initindex(&io->io_hdr.nexus);
11835 	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11836 		if (initindex == i)
11837 			continue;
11838 		lun->pending_ua[i] |= ua_type;
11839 	}
11840 #endif
11841 
11842 	/*
11843 	 * A reset (any kind, really) clears reservations established with
11844 	 * RESERVE/RELEASE.  It does not clear reservations established
11845 	 * with PERSISTENT RESERVE OUT, but we don't support that at the
11846 	 * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
11847 	 * reservations made with the RESERVE/RELEASE commands, because
11848 	 * those commands are obsolete in SPC-3.
11849 	 */
11850 	lun->flags &= ~CTL_LUN_RESERVED;
11851 
11852 	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11853 #ifdef CTL_WITH_CA
11854 		ctl_clear_mask(lun->have_ca, i);
11855 #endif
11856 		lun->pending_ua[i] |= ua_type;
11857 	}
11858 	mtx_unlock(&lun->lun_lock);
11859 
11860 	return (0);
11861 }
11862 
11863 static int
11864 ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
11865     int other_sc)
11866 {
11867 	union ctl_io *xio;
11868 	int found;
11869 
11870 	mtx_assert(&lun->lun_lock, MA_OWNED);
11871 
11872 	/*
11873 	 * Run through the OOA queue and attempt to find the given I/O.
11874 	 * The target port, initiator ID, tag type and tag number have to
11875 	 * match the values that we got from the initiator.  If we have an
11876 	 * untagged command to abort, simply abort the first untagged command
11877 	 * we come to.  We only allow one untagged command at a time of course.
11878 	 */
11879 	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11880 	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11881 
11882 		if ((targ_port == UINT32_MAX ||
11883 		     targ_port == xio->io_hdr.nexus.targ_port) &&
11884 		    (init_id == UINT32_MAX ||
11885 		     init_id == xio->io_hdr.nexus.initid.id)) {
11886 			if (targ_port != xio->io_hdr.nexus.targ_port ||
11887 			    init_id != xio->io_hdr.nexus.initid.id)
11888 				xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS;
11889 			xio->io_hdr.flags |= CTL_FLAG_ABORT;
11890 			found = 1;
11891 			if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11892 				union ctl_ha_msg msg_info;
11893 
11894 				msg_info.hdr.nexus = xio->io_hdr.nexus;
11895 				msg_info.task.task_action = CTL_TASK_ABORT_TASK;
11896 				msg_info.task.tag_num = xio->scsiio.tag_num;
11897 				msg_info.task.tag_type = xio->scsiio.tag_type;
11898 				msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11899 				msg_info.hdr.original_sc = NULL;
11900 				msg_info.hdr.serializing_sc = NULL;
11901 				ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11902 				    (void *)&msg_info, sizeof(msg_info), 0);
11903 			}
11904 		}
11905 	}
11906 	return (found);
11907 }
11908 
11909 static int
11910 ctl_abort_task_set(union ctl_io *io)
11911 {
11912 	struct ctl_softc *softc = control_softc;
11913 	struct ctl_lun *lun;
11914 	uint32_t targ_lun;
11915 
11916 	/*
11917 	 * Look up the LUN.
11918 	 */
11919 	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11920 	mtx_lock(&softc->ctl_lock);
11921 	if ((targ_lun < CTL_MAX_LUNS) && (softc->ctl_luns[targ_lun] != NULL))
11922 		lun = softc->ctl_luns[targ_lun];
11923 	else {
11924 		mtx_unlock(&softc->ctl_lock);
11925 		return (1);
11926 	}
11927 
11928 	mtx_lock(&lun->lun_lock);
11929 	mtx_unlock(&softc->ctl_lock);
11930 	if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
11931 		ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
11932 		    io->io_hdr.nexus.initid.id,
11933 		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11934 	} else { /* CTL_TASK_CLEAR_TASK_SET */
11935 		ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
11936 		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11937 	}
11938 	mtx_unlock(&lun->lun_lock);
11939 	return (0);
11940 }
11941 
11942 static int
11943 ctl_i_t_nexus_reset(union ctl_io *io)
11944 {
11945 	struct ctl_softc *softc = control_softc;
11946 	struct ctl_lun *lun;
11947 	uint32_t initindex;
11948 
11949 	initindex = ctl_get_initindex(&io->io_hdr.nexus);
11950 	mtx_lock(&softc->ctl_lock);
11951 	STAILQ_FOREACH(lun, &softc->lun_list, links) {
11952 		mtx_lock(&lun->lun_lock);
11953 		ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
11954 		    io->io_hdr.nexus.initid.id,
11955 		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11956 #ifdef CTL_WITH_CA
11957 		ctl_clear_mask(lun->have_ca, initindex);
11958 #endif
11959 		lun->pending_ua[initindex] |= CTL_UA_I_T_NEXUS_LOSS;
11960 		mtx_unlock(&lun->lun_lock);
11961 	}
11962 	mtx_unlock(&softc->ctl_lock);
11963 	return (0);
11964 }
11965 
11966 static int
11967 ctl_abort_task(union ctl_io *io)
11968 {
11969 	union ctl_io *xio;
11970 	struct ctl_lun *lun;
11971 	struct ctl_softc *ctl_softc;
11972 #if 0
11973 	struct sbuf sb;
11974 	char printbuf[128];
11975 #endif
11976 	int found;
11977 	uint32_t targ_lun;
11978 
11979 	ctl_softc = control_softc;
11980 	found = 0;
11981 
11982 	/*
11983 	 * Look up the LUN.
11984 	 */
11985 	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11986 	mtx_lock(&ctl_softc->ctl_lock);
11987 	if ((targ_lun < CTL_MAX_LUNS)
11988 	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
11989 		lun = ctl_softc->ctl_luns[targ_lun];
11990 	else {
11991 		mtx_unlock(&ctl_softc->ctl_lock);
11992 		return (1);
11993 	}
11994 
11995 #if 0
11996 	printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
11997 	       lun->lun, io->taskio.tag_num, io->taskio.tag_type);
11998 #endif
11999 
12000 	mtx_lock(&lun->lun_lock);
12001 	mtx_unlock(&ctl_softc->ctl_lock);
12002 	/*
12003 	 * Run through the OOA queue and attempt to find the given I/O.
12004 	 * The target port, initiator ID, tag type and tag number have to
12005 	 * match the values that we got from the initiator.  If we have an
12006 	 * untagged command to abort, simply abort the first untagged command
12007 	 * we come to.  We only allow one untagged command at a time of course.
12008 	 */
12009 #if 0
12010 	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
12011 #endif
12012 	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12013 	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12014 #if 0
12015 		sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
12016 
12017 		sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
12018 			    lun->lun, xio->scsiio.tag_num,
12019 			    xio->scsiio.tag_type,
12020 			    (xio->io_hdr.blocked_links.tqe_prev
12021 			    == NULL) ? "" : " BLOCKED",
12022 			    (xio->io_hdr.flags &
12023 			    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
12024 			    (xio->io_hdr.flags &
12025 			    CTL_FLAG_ABORT) ? " ABORT" : "",
12026 			    (xio->io_hdr.flags &
12027 			    CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
12028 		ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
12029 		sbuf_finish(&sb);
12030 		printf("%s\n", sbuf_data(&sb));
12031 #endif
12032 
12033 		if ((xio->io_hdr.nexus.targ_port == io->io_hdr.nexus.targ_port)
12034 		 && (xio->io_hdr.nexus.initid.id ==
12035 		     io->io_hdr.nexus.initid.id)) {
12036 			/*
12037 			 * If the abort says that the task is untagged, the
12038 			 * task in the queue must be untagged.  Otherwise,
12039 			 * we just check to see whether the tag numbers
12040 			 * match.  This is because the QLogic firmware
12041 			 * doesn't pass back the tag type in an abort
12042 			 * request.
12043 			 */
12044 #if 0
12045 			if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
12046 			  && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
12047 			 || (xio->scsiio.tag_num == io->taskio.tag_num)) {
12048 #endif
12049 			/*
12050 			 * XXX KDM we've got problems with FC, because it
12051 			 * doesn't send down a tag type with aborts.  So we
12052 			 * can only really go by the tag number...
12053 			 * This may cause problems with parallel SCSI.
12054 			 * Need to figure that out!!
12055 			 */
12056 			if (xio->scsiio.tag_num == io->taskio.tag_num) {
12057 				xio->io_hdr.flags |= CTL_FLAG_ABORT;
12058 				found = 1;
12059 				if ((io->io_hdr.flags &
12060 				     CTL_FLAG_FROM_OTHER_SC) == 0 &&
12061 				    !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12062 					union ctl_ha_msg msg_info;
12063 
12064 					io->io_hdr.flags |=
12065 					                CTL_FLAG_SENT_2OTHER_SC;
12066 					msg_info.hdr.nexus = io->io_hdr.nexus;
12067 					msg_info.task.task_action =
12068 						CTL_TASK_ABORT_TASK;
12069 					msg_info.task.tag_num =
12070 						io->taskio.tag_num;
12071 					msg_info.task.tag_type =
12072 						io->taskio.tag_type;
12073 					msg_info.hdr.msg_type =
12074 						CTL_MSG_MANAGE_TASKS;
12075 					msg_info.hdr.original_sc = NULL;
12076 					msg_info.hdr.serializing_sc = NULL;
12077 #if 0
12078 					printf("Sent Abort to other side\n");
12079 #endif
12080 					if (CTL_HA_STATUS_SUCCESS !=
12081 					        ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12082 		    				(void *)&msg_info,
12083 						sizeof(msg_info), 0)) {
12084 					}
12085 				}
12086 #if 0
12087 				printf("ctl_abort_task: found I/O to abort\n");
12088 #endif
12089 				break;
12090 			}
12091 		}
12092 	}
12093 	mtx_unlock(&lun->lun_lock);
12094 
12095 	if (found == 0) {
12096 		/*
12097 		 * This isn't really an error.  It's entirely possible for
12098 		 * the abort and command completion to cross on the wire.
12099 		 * This is more of an informative/diagnostic error.
12100 		 */
12101 #if 0
12102 		printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
12103 		       "%d:%d:%d:%d tag %d type %d\n",
12104 		       io->io_hdr.nexus.initid.id,
12105 		       io->io_hdr.nexus.targ_port,
12106 		       io->io_hdr.nexus.targ_target.id,
12107 		       io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
12108 		       io->taskio.tag_type);
12109 #endif
12110 	}
12111 	return (0);
12112 }
12113 
12114 static void
12115 ctl_run_task(union ctl_io *io)
12116 {
12117 	struct ctl_softc *ctl_softc = control_softc;
12118 	int retval = 1;
12119 	const char *task_desc;
12120 
12121 	CTL_DEBUG_PRINT(("ctl_run_task\n"));
12122 
12123 	KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
12124 	    ("ctl_run_task: Unextected io_type %d\n",
12125 	     io->io_hdr.io_type));
12126 
12127 	task_desc = ctl_scsi_task_string(&io->taskio);
12128 	if (task_desc != NULL) {
12129 #ifdef NEEDTOPORT
12130 		csevent_log(CSC_CTL | CSC_SHELF_SW |
12131 			    CTL_TASK_REPORT,
12132 			    csevent_LogType_Trace,
12133 			    csevent_Severity_Information,
12134 			    csevent_AlertLevel_Green,
12135 			    csevent_FRU_Firmware,
12136 			    csevent_FRU_Unknown,
12137 			    "CTL: received task: %s",task_desc);
12138 #endif
12139 	} else {
12140 #ifdef NEEDTOPORT
12141 		csevent_log(CSC_CTL | CSC_SHELF_SW |
12142 			    CTL_TASK_REPORT,
12143 			    csevent_LogType_Trace,
12144 			    csevent_Severity_Information,
12145 			    csevent_AlertLevel_Green,
12146 			    csevent_FRU_Firmware,
12147 			    csevent_FRU_Unknown,
12148 			    "CTL: received unknown task "
12149 			    "type: %d (%#x)",
12150 			    io->taskio.task_action,
12151 			    io->taskio.task_action);
12152 #endif
12153 	}
12154 	switch (io->taskio.task_action) {
12155 	case CTL_TASK_ABORT_TASK:
12156 		retval = ctl_abort_task(io);
12157 		break;
12158 	case CTL_TASK_ABORT_TASK_SET:
12159 	case CTL_TASK_CLEAR_TASK_SET:
12160 		retval = ctl_abort_task_set(io);
12161 		break;
12162 	case CTL_TASK_CLEAR_ACA:
12163 		break;
12164 	case CTL_TASK_I_T_NEXUS_RESET:
12165 		retval = ctl_i_t_nexus_reset(io);
12166 		break;
12167 	case CTL_TASK_LUN_RESET: {
12168 		struct ctl_lun *lun;
12169 		uint32_t targ_lun;
12170 
12171 		targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12172 		mtx_lock(&ctl_softc->ctl_lock);
12173 		if ((targ_lun < CTL_MAX_LUNS)
12174 		 && (ctl_softc->ctl_luns[targ_lun] != NULL))
12175 			lun = ctl_softc->ctl_luns[targ_lun];
12176 		else {
12177 			mtx_unlock(&ctl_softc->ctl_lock);
12178 			retval = 1;
12179 			break;
12180 		}
12181 
12182 		if (!(io->io_hdr.flags &
12183 		    CTL_FLAG_FROM_OTHER_SC)) {
12184 			union ctl_ha_msg msg_info;
12185 
12186 			io->io_hdr.flags |=
12187 				CTL_FLAG_SENT_2OTHER_SC;
12188 			msg_info.hdr.msg_type =
12189 				CTL_MSG_MANAGE_TASKS;
12190 			msg_info.hdr.nexus = io->io_hdr.nexus;
12191 			msg_info.task.task_action =
12192 				CTL_TASK_LUN_RESET;
12193 			msg_info.hdr.original_sc = NULL;
12194 			msg_info.hdr.serializing_sc = NULL;
12195 			if (CTL_HA_STATUS_SUCCESS !=
12196 			    ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12197 			    (void *)&msg_info,
12198 			    sizeof(msg_info), 0)) {
12199 			}
12200 		}
12201 
12202 		retval = ctl_lun_reset(lun, io,
12203 				       CTL_UA_LUN_RESET);
12204 		mtx_unlock(&ctl_softc->ctl_lock);
12205 		break;
12206 	}
12207 	case CTL_TASK_TARGET_RESET:
12208 		retval = ctl_target_reset(ctl_softc, io, CTL_UA_TARG_RESET);
12209 		break;
12210 	case CTL_TASK_BUS_RESET:
12211 		retval = ctl_bus_reset(ctl_softc, io);
12212 		break;
12213 	case CTL_TASK_PORT_LOGIN:
12214 		break;
12215 	case CTL_TASK_PORT_LOGOUT:
12216 		break;
12217 	default:
12218 		printf("ctl_run_task: got unknown task management event %d\n",
12219 		       io->taskio.task_action);
12220 		break;
12221 	}
12222 	if (retval == 0)
12223 		io->io_hdr.status = CTL_SUCCESS;
12224 	else
12225 		io->io_hdr.status = CTL_ERROR;
12226 	ctl_done(io);
12227 }
12228 
12229 /*
12230  * For HA operation.  Handle commands that come in from the other
12231  * controller.
12232  */
12233 static void
12234 ctl_handle_isc(union ctl_io *io)
12235 {
12236 	int free_io;
12237 	struct ctl_lun *lun;
12238 	struct ctl_softc *ctl_softc;
12239 	uint32_t targ_lun;
12240 
12241 	ctl_softc = control_softc;
12242 
12243 	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12244 	lun = ctl_softc->ctl_luns[targ_lun];
12245 
12246 	switch (io->io_hdr.msg_type) {
12247 	case CTL_MSG_SERIALIZE:
12248 		free_io = ctl_serialize_other_sc_cmd(&io->scsiio);
12249 		break;
12250 	case CTL_MSG_R2R: {
12251 		const struct ctl_cmd_entry *entry;
12252 
12253 		/*
12254 		 * This is only used in SER_ONLY mode.
12255 		 */
12256 		free_io = 0;
12257 		entry = ctl_get_cmd_entry(&io->scsiio);
12258 		mtx_lock(&lun->lun_lock);
12259 		if (ctl_scsiio_lun_check(ctl_softc, lun,
12260 		    entry, (struct ctl_scsiio *)io) != 0) {
12261 			mtx_unlock(&lun->lun_lock);
12262 			ctl_done(io);
12263 			break;
12264 		}
12265 		io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
12266 		mtx_unlock(&lun->lun_lock);
12267 		ctl_enqueue_rtr(io);
12268 		break;
12269 	}
12270 	case CTL_MSG_FINISH_IO:
12271 		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
12272 			free_io = 0;
12273 			ctl_done(io);
12274 		} else {
12275 			free_io = 1;
12276 			mtx_lock(&lun->lun_lock);
12277 			TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
12278 				     ooa_links);
12279 			ctl_check_blocked(lun);
12280 			mtx_unlock(&lun->lun_lock);
12281 		}
12282 		break;
12283 	case CTL_MSG_PERS_ACTION:
12284 		ctl_hndl_per_res_out_on_other_sc(
12285 			(union ctl_ha_msg *)&io->presio.pr_msg);
12286 		free_io = 1;
12287 		break;
12288 	case CTL_MSG_BAD_JUJU:
12289 		free_io = 0;
12290 		ctl_done(io);
12291 		break;
12292 	case CTL_MSG_DATAMOVE:
12293 		/* Only used in XFER mode */
12294 		free_io = 0;
12295 		ctl_datamove_remote(io);
12296 		break;
12297 	case CTL_MSG_DATAMOVE_DONE:
12298 		/* Only used in XFER mode */
12299 		free_io = 0;
12300 		io->scsiio.be_move_done(io);
12301 		break;
12302 	default:
12303 		free_io = 1;
12304 		printf("%s: Invalid message type %d\n",
12305 		       __func__, io->io_hdr.msg_type);
12306 		break;
12307 	}
12308 	if (free_io)
12309 		ctl_free_io(io);
12310 
12311 }
12312 
12313 
12314 /*
12315  * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12316  * there is no match.
12317  */
12318 static ctl_lun_error_pattern
12319 ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12320 {
12321 	const struct ctl_cmd_entry *entry;
12322 	ctl_lun_error_pattern filtered_pattern, pattern;
12323 
12324 	pattern = desc->error_pattern;
12325 
12326 	/*
12327 	 * XXX KDM we need more data passed into this function to match a
12328 	 * custom pattern, and we actually need to implement custom pattern
12329 	 * matching.
12330 	 */
12331 	if (pattern & CTL_LUN_PAT_CMD)
12332 		return (CTL_LUN_PAT_CMD);
12333 
12334 	if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12335 		return (CTL_LUN_PAT_ANY);
12336 
12337 	entry = ctl_get_cmd_entry(ctsio);
12338 
12339 	filtered_pattern = entry->pattern & pattern;
12340 
12341 	/*
12342 	 * If the user requested specific flags in the pattern (e.g.
12343 	 * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12344 	 * flags.
12345 	 *
12346 	 * If the user did not specify any flags, it doesn't matter whether
12347 	 * or not the command supports the flags.
12348 	 */
12349 	if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12350 	     (pattern & ~CTL_LUN_PAT_MASK))
12351 		return (CTL_LUN_PAT_NONE);
12352 
12353 	/*
12354 	 * If the user asked for a range check, see if the requested LBA
12355 	 * range overlaps with this command's LBA range.
12356 	 */
12357 	if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12358 		uint64_t lba1;
12359 		uint32_t len1;
12360 		ctl_action action;
12361 		int retval;
12362 
12363 		retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12364 		if (retval != 0)
12365 			return (CTL_LUN_PAT_NONE);
12366 
12367 		action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12368 					      desc->lba_range.len);
12369 		/*
12370 		 * A "pass" means that the LBA ranges don't overlap, so
12371 		 * this doesn't match the user's range criteria.
12372 		 */
12373 		if (action == CTL_ACTION_PASS)
12374 			return (CTL_LUN_PAT_NONE);
12375 	}
12376 
12377 	return (filtered_pattern);
12378 }
12379 
12380 static void
12381 ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12382 {
12383 	struct ctl_error_desc *desc, *desc2;
12384 
12385 	mtx_assert(&lun->lun_lock, MA_OWNED);
12386 
12387 	STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12388 		ctl_lun_error_pattern pattern;
12389 		/*
12390 		 * Check to see whether this particular command matches
12391 		 * the pattern in the descriptor.
12392 		 */
12393 		pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12394 		if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12395 			continue;
12396 
12397 		switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12398 		case CTL_LUN_INJ_ABORTED:
12399 			ctl_set_aborted(&io->scsiio);
12400 			break;
12401 		case CTL_LUN_INJ_MEDIUM_ERR:
12402 			ctl_set_medium_error(&io->scsiio);
12403 			break;
12404 		case CTL_LUN_INJ_UA:
12405 			/* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12406 			 * OCCURRED */
12407 			ctl_set_ua(&io->scsiio, 0x29, 0x00);
12408 			break;
12409 		case CTL_LUN_INJ_CUSTOM:
12410 			/*
12411 			 * We're assuming the user knows what he is doing.
12412 			 * Just copy the sense information without doing
12413 			 * checks.
12414 			 */
12415 			bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12416 			      ctl_min(sizeof(desc->custom_sense),
12417 				      sizeof(io->scsiio.sense_data)));
12418 			io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12419 			io->scsiio.sense_len = SSD_FULL_SIZE;
12420 			io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12421 			break;
12422 		case CTL_LUN_INJ_NONE:
12423 		default:
12424 			/*
12425 			 * If this is an error injection type we don't know
12426 			 * about, clear the continuous flag (if it is set)
12427 			 * so it will get deleted below.
12428 			 */
12429 			desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12430 			break;
12431 		}
12432 		/*
12433 		 * By default, each error injection action is a one-shot
12434 		 */
12435 		if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12436 			continue;
12437 
12438 		STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12439 
12440 		free(desc, M_CTL);
12441 	}
12442 }
12443 
12444 #ifdef CTL_IO_DELAY
12445 static void
12446 ctl_datamove_timer_wakeup(void *arg)
12447 {
12448 	union ctl_io *io;
12449 
12450 	io = (union ctl_io *)arg;
12451 
12452 	ctl_datamove(io);
12453 }
12454 #endif /* CTL_IO_DELAY */
12455 
12456 void
12457 ctl_datamove(union ctl_io *io)
12458 {
12459 	void (*fe_datamove)(union ctl_io *io);
12460 
12461 	mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
12462 
12463 	CTL_DEBUG_PRINT(("ctl_datamove\n"));
12464 
12465 #ifdef CTL_TIME_IO
12466 	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12467 		char str[256];
12468 		char path_str[64];
12469 		struct sbuf sb;
12470 
12471 		ctl_scsi_path_string(io, path_str, sizeof(path_str));
12472 		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12473 
12474 		sbuf_cat(&sb, path_str);
12475 		switch (io->io_hdr.io_type) {
12476 		case CTL_IO_SCSI:
12477 			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12478 			sbuf_printf(&sb, "\n");
12479 			sbuf_cat(&sb, path_str);
12480 			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12481 				    io->scsiio.tag_num, io->scsiio.tag_type);
12482 			break;
12483 		case CTL_IO_TASK:
12484 			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12485 				    "Tag Type: %d\n", io->taskio.task_action,
12486 				    io->taskio.tag_num, io->taskio.tag_type);
12487 			break;
12488 		default:
12489 			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12490 			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12491 			break;
12492 		}
12493 		sbuf_cat(&sb, path_str);
12494 		sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12495 			    (intmax_t)time_uptime - io->io_hdr.start_time);
12496 		sbuf_finish(&sb);
12497 		printf("%s", sbuf_data(&sb));
12498 	}
12499 #endif /* CTL_TIME_IO */
12500 
12501 #ifdef CTL_IO_DELAY
12502 	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12503 		struct ctl_lun *lun;
12504 
12505 		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12506 
12507 		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12508 	} else {
12509 		struct ctl_lun *lun;
12510 
12511 		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12512 		if ((lun != NULL)
12513 		 && (lun->delay_info.datamove_delay > 0)) {
12514 			struct callout *callout;
12515 
12516 			callout = (struct callout *)&io->io_hdr.timer_bytes;
12517 			callout_init(callout, /*mpsafe*/ 1);
12518 			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12519 			callout_reset(callout,
12520 				      lun->delay_info.datamove_delay * hz,
12521 				      ctl_datamove_timer_wakeup, io);
12522 			if (lun->delay_info.datamove_type ==
12523 			    CTL_DELAY_TYPE_ONESHOT)
12524 				lun->delay_info.datamove_delay = 0;
12525 			return;
12526 		}
12527 	}
12528 #endif
12529 
12530 	/*
12531 	 * This command has been aborted.  Set the port status, so we fail
12532 	 * the data move.
12533 	 */
12534 	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12535 		printf("ctl_datamove: tag 0x%04x on (%ju:%d:%ju:%d) aborted\n",
12536 		       io->scsiio.tag_num,(uintmax_t)io->io_hdr.nexus.initid.id,
12537 		       io->io_hdr.nexus.targ_port,
12538 		       (uintmax_t)io->io_hdr.nexus.targ_target.id,
12539 		       io->io_hdr.nexus.targ_lun);
12540 		io->io_hdr.port_status = 31337;
12541 		/*
12542 		 * Note that the backend, in this case, will get the
12543 		 * callback in its context.  In other cases it may get
12544 		 * called in the frontend's interrupt thread context.
12545 		 */
12546 		io->scsiio.be_move_done(io);
12547 		return;
12548 	}
12549 
12550 	/*
12551 	 * If we're in XFER mode and this I/O is from the other shelf
12552 	 * controller, we need to send the DMA to the other side to
12553 	 * actually transfer the data to/from the host.  In serialize only
12554 	 * mode the transfer happens below CTL and ctl_datamove() is only
12555 	 * called on the machine that originally received the I/O.
12556 	 */
12557 	if ((control_softc->ha_mode == CTL_HA_MODE_XFER)
12558 	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12559 		union ctl_ha_msg msg;
12560 		uint32_t sg_entries_sent;
12561 		int do_sg_copy;
12562 		int i;
12563 
12564 		memset(&msg, 0, sizeof(msg));
12565 		msg.hdr.msg_type = CTL_MSG_DATAMOVE;
12566 		msg.hdr.original_sc = io->io_hdr.original_sc;
12567 		msg.hdr.serializing_sc = io;
12568 		msg.hdr.nexus = io->io_hdr.nexus;
12569 		msg.dt.flags = io->io_hdr.flags;
12570 		/*
12571 		 * We convert everything into a S/G list here.  We can't
12572 		 * pass by reference, only by value between controllers.
12573 		 * So we can't pass a pointer to the S/G list, only as many
12574 		 * S/G entries as we can fit in here.  If it's possible for
12575 		 * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
12576 		 * then we need to break this up into multiple transfers.
12577 		 */
12578 		if (io->scsiio.kern_sg_entries == 0) {
12579 			msg.dt.kern_sg_entries = 1;
12580 			/*
12581 			 * If this is in cached memory, flush the cache
12582 			 * before we send the DMA request to the other
12583 			 * controller.  We want to do this in either the
12584 			 * read or the write case.  The read case is
12585 			 * straightforward.  In the write case, we want to
12586 			 * make sure nothing is in the local cache that
12587 			 * could overwrite the DMAed data.
12588 			 */
12589 			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12590 				/*
12591 				 * XXX KDM use bus_dmamap_sync() here.
12592 				 */
12593 			}
12594 
12595 			/*
12596 			 * Convert to a physical address if this is a
12597 			 * virtual address.
12598 			 */
12599 			if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
12600 				msg.dt.sg_list[0].addr =
12601 					io->scsiio.kern_data_ptr;
12602 			} else {
12603 				/*
12604 				 * XXX KDM use busdma here!
12605 				 */
12606 #if 0
12607 				msg.dt.sg_list[0].addr = (void *)
12608 					vtophys(io->scsiio.kern_data_ptr);
12609 #endif
12610 			}
12611 
12612 			msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
12613 			do_sg_copy = 0;
12614 		} else {
12615 			struct ctl_sg_entry *sgl;
12616 
12617 			do_sg_copy = 1;
12618 			msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
12619 			sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
12620 			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12621 				/*
12622 				 * XXX KDM use bus_dmamap_sync() here.
12623 				 */
12624 			}
12625 		}
12626 
12627 		msg.dt.kern_data_len = io->scsiio.kern_data_len;
12628 		msg.dt.kern_total_len = io->scsiio.kern_total_len;
12629 		msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
12630 		msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
12631 		msg.dt.sg_sequence = 0;
12632 
12633 		/*
12634 		 * Loop until we've sent all of the S/G entries.  On the
12635 		 * other end, we'll recompose these S/G entries into one
12636 		 * contiguous list before passing it to the
12637 		 */
12638 		for (sg_entries_sent = 0; sg_entries_sent <
12639 		     msg.dt.kern_sg_entries; msg.dt.sg_sequence++) {
12640 			msg.dt.cur_sg_entries = ctl_min((sizeof(msg.dt.sg_list)/
12641 				sizeof(msg.dt.sg_list[0])),
12642 				msg.dt.kern_sg_entries - sg_entries_sent);
12643 
12644 			if (do_sg_copy != 0) {
12645 				struct ctl_sg_entry *sgl;
12646 				int j;
12647 
12648 				sgl = (struct ctl_sg_entry *)
12649 					io->scsiio.kern_data_ptr;
12650 				/*
12651 				 * If this is in cached memory, flush the cache
12652 				 * before we send the DMA request to the other
12653 				 * controller.  We want to do this in either
12654 				 * the * read or the write case.  The read
12655 				 * case is straightforward.  In the write
12656 				 * case, we want to make sure nothing is
12657 				 * in the local cache that could overwrite
12658 				 * the DMAed data.
12659 				 */
12660 
12661 				for (i = sg_entries_sent, j = 0;
12662 				     i < msg.dt.cur_sg_entries; i++, j++) {
12663 					if ((io->io_hdr.flags &
12664 					     CTL_FLAG_NO_DATASYNC) == 0) {
12665 						/*
12666 						 * XXX KDM use bus_dmamap_sync()
12667 						 */
12668 					}
12669 					if ((io->io_hdr.flags &
12670 					     CTL_FLAG_BUS_ADDR) == 0) {
12671 						/*
12672 						 * XXX KDM use busdma.
12673 						 */
12674 #if 0
12675 						msg.dt.sg_list[j].addr =(void *)
12676 						       vtophys(sgl[i].addr);
12677 #endif
12678 					} else {
12679 						msg.dt.sg_list[j].addr =
12680 							sgl[i].addr;
12681 					}
12682 					msg.dt.sg_list[j].len = sgl[i].len;
12683 				}
12684 			}
12685 
12686 			sg_entries_sent += msg.dt.cur_sg_entries;
12687 			if (sg_entries_sent >= msg.dt.kern_sg_entries)
12688 				msg.dt.sg_last = 1;
12689 			else
12690 				msg.dt.sg_last = 0;
12691 
12692 			/*
12693 			 * XXX KDM drop and reacquire the lock here?
12694 			 */
12695 			if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12696 			    sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
12697 				/*
12698 				 * XXX do something here.
12699 				 */
12700 			}
12701 
12702 			msg.dt.sent_sg_entries = sg_entries_sent;
12703 		}
12704 		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12705 		if (io->io_hdr.flags & CTL_FLAG_FAILOVER)
12706 			ctl_failover_io(io, /*have_lock*/ 0);
12707 
12708 	} else {
12709 
12710 		/*
12711 		 * Lookup the fe_datamove() function for this particular
12712 		 * front end.
12713 		 */
12714 		fe_datamove =
12715 		    control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12716 
12717 		fe_datamove(io);
12718 	}
12719 }
12720 
12721 static void
12722 ctl_send_datamove_done(union ctl_io *io, int have_lock)
12723 {
12724 	union ctl_ha_msg msg;
12725 	int isc_status;
12726 
12727 	memset(&msg, 0, sizeof(msg));
12728 
12729 	msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12730 	msg.hdr.original_sc = io;
12731 	msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12732 	msg.hdr.nexus = io->io_hdr.nexus;
12733 	msg.hdr.status = io->io_hdr.status;
12734 	msg.scsi.tag_num = io->scsiio.tag_num;
12735 	msg.scsi.tag_type = io->scsiio.tag_type;
12736 	msg.scsi.scsi_status = io->scsiio.scsi_status;
12737 	memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12738 	       sizeof(io->scsiio.sense_data));
12739 	msg.scsi.sense_len = io->scsiio.sense_len;
12740 	msg.scsi.sense_residual = io->scsiio.sense_residual;
12741 	msg.scsi.fetd_status = io->io_hdr.port_status;
12742 	msg.scsi.residual = io->scsiio.residual;
12743 	io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12744 
12745 	if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12746 		ctl_failover_io(io, /*have_lock*/ have_lock);
12747 		return;
12748 	}
12749 
12750 	isc_status = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0);
12751 	if (isc_status > CTL_HA_STATUS_SUCCESS) {
12752 		/* XXX do something if this fails */
12753 	}
12754 
12755 }
12756 
12757 /*
12758  * The DMA to the remote side is done, now we need to tell the other side
12759  * we're done so it can continue with its data movement.
12760  */
12761 static void
12762 ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12763 {
12764 	union ctl_io *io;
12765 
12766 	io = rq->context;
12767 
12768 	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12769 		printf("%s: ISC DMA write failed with error %d", __func__,
12770 		       rq->ret);
12771 		ctl_set_internal_failure(&io->scsiio,
12772 					 /*sks_valid*/ 1,
12773 					 /*retry_count*/ rq->ret);
12774 	}
12775 
12776 	ctl_dt_req_free(rq);
12777 
12778 	/*
12779 	 * In this case, we had to malloc the memory locally.  Free it.
12780 	 */
12781 	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12782 		int i;
12783 		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12784 			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12785 	}
12786 	/*
12787 	 * The data is in local and remote memory, so now we need to send
12788 	 * status (good or back) back to the other side.
12789 	 */
12790 	ctl_send_datamove_done(io, /*have_lock*/ 0);
12791 }
12792 
12793 /*
12794  * We've moved the data from the host/controller into local memory.  Now we
12795  * need to push it over to the remote controller's memory.
12796  */
12797 static int
12798 ctl_datamove_remote_dm_write_cb(union ctl_io *io)
12799 {
12800 	int retval;
12801 
12802 	retval = 0;
12803 
12804 	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12805 					  ctl_datamove_remote_write_cb);
12806 
12807 	return (retval);
12808 }
12809 
12810 static void
12811 ctl_datamove_remote_write(union ctl_io *io)
12812 {
12813 	int retval;
12814 	void (*fe_datamove)(union ctl_io *io);
12815 
12816 	/*
12817 	 * - Get the data from the host/HBA into local memory.
12818 	 * - DMA memory from the local controller to the remote controller.
12819 	 * - Send status back to the remote controller.
12820 	 */
12821 
12822 	retval = ctl_datamove_remote_sgl_setup(io);
12823 	if (retval != 0)
12824 		return;
12825 
12826 	/* Switch the pointer over so the FETD knows what to do */
12827 	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12828 
12829 	/*
12830 	 * Use a custom move done callback, since we need to send completion
12831 	 * back to the other controller, not to the backend on this side.
12832 	 */
12833 	io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12834 
12835 	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12836 
12837 	fe_datamove(io);
12838 
12839 	return;
12840 
12841 }
12842 
12843 static int
12844 ctl_datamove_remote_dm_read_cb(union ctl_io *io)
12845 {
12846 #if 0
12847 	char str[256];
12848 	char path_str[64];
12849 	struct sbuf sb;
12850 #endif
12851 
12852 	/*
12853 	 * In this case, we had to malloc the memory locally.  Free it.
12854 	 */
12855 	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12856 		int i;
12857 		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12858 			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12859 	}
12860 
12861 #if 0
12862 	scsi_path_string(io, path_str, sizeof(path_str));
12863 	sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12864 	sbuf_cat(&sb, path_str);
12865 	scsi_command_string(&io->scsiio, NULL, &sb);
12866 	sbuf_printf(&sb, "\n");
12867 	sbuf_cat(&sb, path_str);
12868 	sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12869 		    io->scsiio.tag_num, io->scsiio.tag_type);
12870 	sbuf_cat(&sb, path_str);
12871 	sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
12872 		    io->io_hdr.flags, io->io_hdr.status);
12873 	sbuf_finish(&sb);
12874 	printk("%s", sbuf_data(&sb));
12875 #endif
12876 
12877 
12878 	/*
12879 	 * The read is done, now we need to send status (good or bad) back
12880 	 * to the other side.
12881 	 */
12882 	ctl_send_datamove_done(io, /*have_lock*/ 0);
12883 
12884 	return (0);
12885 }
12886 
12887 static void
12888 ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
12889 {
12890 	union ctl_io *io;
12891 	void (*fe_datamove)(union ctl_io *io);
12892 
12893 	io = rq->context;
12894 
12895 	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12896 		printf("%s: ISC DMA read failed with error %d", __func__,
12897 		       rq->ret);
12898 		ctl_set_internal_failure(&io->scsiio,
12899 					 /*sks_valid*/ 1,
12900 					 /*retry_count*/ rq->ret);
12901 	}
12902 
12903 	ctl_dt_req_free(rq);
12904 
12905 	/* Switch the pointer over so the FETD knows what to do */
12906 	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12907 
12908 	/*
12909 	 * Use a custom move done callback, since we need to send completion
12910 	 * back to the other controller, not to the backend on this side.
12911 	 */
12912 	io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
12913 
12914 	/* XXX KDM add checks like the ones in ctl_datamove? */
12915 
12916 	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12917 
12918 	fe_datamove(io);
12919 }
12920 
12921 static int
12922 ctl_datamove_remote_sgl_setup(union ctl_io *io)
12923 {
12924 	struct ctl_sg_entry *local_sglist, *remote_sglist;
12925 	struct ctl_sg_entry *local_dma_sglist, *remote_dma_sglist;
12926 	struct ctl_softc *softc;
12927 	int retval;
12928 	int i;
12929 
12930 	retval = 0;
12931 	softc = control_softc;
12932 
12933 	local_sglist = io->io_hdr.local_sglist;
12934 	local_dma_sglist = io->io_hdr.local_dma_sglist;
12935 	remote_sglist = io->io_hdr.remote_sglist;
12936 	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
12937 
12938 	if (io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) {
12939 		for (i = 0; i < io->scsiio.kern_sg_entries; i++) {
12940 			local_sglist[i].len = remote_sglist[i].len;
12941 
12942 			/*
12943 			 * XXX Detect the situation where the RS-level I/O
12944 			 * redirector on the other side has already read the
12945 			 * data off of the AOR RS on this side, and
12946 			 * transferred it to remote (mirror) memory on the
12947 			 * other side.  Since we already have the data in
12948 			 * memory here, we just need to use it.
12949 			 *
12950 			 * XXX KDM this can probably be removed once we
12951 			 * get the cache device code in and take the
12952 			 * current AOR implementation out.
12953 			 */
12954 #ifdef NEEDTOPORT
12955 			if ((remote_sglist[i].addr >=
12956 			     (void *)vtophys(softc->mirr->addr))
12957 			 && (remote_sglist[i].addr <
12958 			     ((void *)vtophys(softc->mirr->addr) +
12959 			     CacheMirrorOffset))) {
12960 				local_sglist[i].addr = remote_sglist[i].addr -
12961 					CacheMirrorOffset;
12962 				if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
12963 				     CTL_FLAG_DATA_IN)
12964 					io->io_hdr.flags |= CTL_FLAG_REDIR_DONE;
12965 			} else {
12966 				local_sglist[i].addr = remote_sglist[i].addr +
12967 					CacheMirrorOffset;
12968 			}
12969 #endif
12970 #if 0
12971 			printf("%s: local %p, remote %p, len %d\n",
12972 			       __func__, local_sglist[i].addr,
12973 			       remote_sglist[i].addr, local_sglist[i].len);
12974 #endif
12975 		}
12976 	} else {
12977 		uint32_t len_to_go;
12978 
12979 		/*
12980 		 * In this case, we don't have automatically allocated
12981 		 * memory for this I/O on this controller.  This typically
12982 		 * happens with internal CTL I/O -- e.g. inquiry, mode
12983 		 * sense, etc.  Anything coming from RAIDCore will have
12984 		 * a mirror area available.
12985 		 */
12986 		len_to_go = io->scsiio.kern_data_len;
12987 
12988 		/*
12989 		 * Clear the no datasync flag, we have to use malloced
12990 		 * buffers.
12991 		 */
12992 		io->io_hdr.flags &= ~CTL_FLAG_NO_DATASYNC;
12993 
12994 		/*
12995 		 * The difficult thing here is that the size of the various
12996 		 * S/G segments may be different than the size from the
12997 		 * remote controller.  That'll make it harder when DMAing
12998 		 * the data back to the other side.
12999 		 */
13000 		for (i = 0; (i < sizeof(io->io_hdr.remote_sglist) /
13001 		     sizeof(io->io_hdr.remote_sglist[0])) &&
13002 		     (len_to_go > 0); i++) {
13003 			local_sglist[i].len = ctl_min(len_to_go, 131072);
13004 			CTL_SIZE_8B(local_dma_sglist[i].len,
13005 				    local_sglist[i].len);
13006 			local_sglist[i].addr =
13007 				malloc(local_dma_sglist[i].len, M_CTL,M_WAITOK);
13008 
13009 			local_dma_sglist[i].addr = local_sglist[i].addr;
13010 
13011 			if (local_sglist[i].addr == NULL) {
13012 				int j;
13013 
13014 				printf("malloc failed for %zd bytes!",
13015 				       local_dma_sglist[i].len);
13016 				for (j = 0; j < i; j++) {
13017 					free(local_sglist[j].addr, M_CTL);
13018 				}
13019 				ctl_set_internal_failure(&io->scsiio,
13020 							 /*sks_valid*/ 1,
13021 							 /*retry_count*/ 4857);
13022 				retval = 1;
13023 				goto bailout_error;
13024 
13025 			}
13026 			/* XXX KDM do we need a sync here? */
13027 
13028 			len_to_go -= local_sglist[i].len;
13029 		}
13030 		/*
13031 		 * Reset the number of S/G entries accordingly.  The
13032 		 * original number of S/G entries is available in
13033 		 * rem_sg_entries.
13034 		 */
13035 		io->scsiio.kern_sg_entries = i;
13036 
13037 #if 0
13038 		printf("%s: kern_sg_entries = %d\n", __func__,
13039 		       io->scsiio.kern_sg_entries);
13040 		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13041 			printf("%s: sg[%d] = %p, %d (DMA: %d)\n", __func__, i,
13042 			       local_sglist[i].addr, local_sglist[i].len,
13043 			       local_dma_sglist[i].len);
13044 #endif
13045 	}
13046 
13047 
13048 	return (retval);
13049 
13050 bailout_error:
13051 
13052 	ctl_send_datamove_done(io, /*have_lock*/ 0);
13053 
13054 	return (retval);
13055 }
13056 
13057 static int
13058 ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
13059 			 ctl_ha_dt_cb callback)
13060 {
13061 	struct ctl_ha_dt_req *rq;
13062 	struct ctl_sg_entry *remote_sglist, *local_sglist;
13063 	struct ctl_sg_entry *remote_dma_sglist, *local_dma_sglist;
13064 	uint32_t local_used, remote_used, total_used;
13065 	int retval;
13066 	int i, j;
13067 
13068 	retval = 0;
13069 
13070 	rq = ctl_dt_req_alloc();
13071 
13072 	/*
13073 	 * If we failed to allocate the request, and if the DMA didn't fail
13074 	 * anyway, set busy status.  This is just a resource allocation
13075 	 * failure.
13076 	 */
13077 	if ((rq == NULL)
13078 	 && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE))
13079 		ctl_set_busy(&io->scsiio);
13080 
13081 	if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) {
13082 
13083 		if (rq != NULL)
13084 			ctl_dt_req_free(rq);
13085 
13086 		/*
13087 		 * The data move failed.  We need to return status back
13088 		 * to the other controller.  No point in trying to DMA
13089 		 * data to the remote controller.
13090 		 */
13091 
13092 		ctl_send_datamove_done(io, /*have_lock*/ 0);
13093 
13094 		retval = 1;
13095 
13096 		goto bailout;
13097 	}
13098 
13099 	local_sglist = io->io_hdr.local_sglist;
13100 	local_dma_sglist = io->io_hdr.local_dma_sglist;
13101 	remote_sglist = io->io_hdr.remote_sglist;
13102 	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
13103 	local_used = 0;
13104 	remote_used = 0;
13105 	total_used = 0;
13106 
13107 	if (io->io_hdr.flags & CTL_FLAG_REDIR_DONE) {
13108 		rq->ret = CTL_HA_STATUS_SUCCESS;
13109 		rq->context = io;
13110 		callback(rq);
13111 		goto bailout;
13112 	}
13113 
13114 	/*
13115 	 * Pull/push the data over the wire from/to the other controller.
13116 	 * This takes into account the possibility that the local and
13117 	 * remote sglists may not be identical in terms of the size of
13118 	 * the elements and the number of elements.
13119 	 *
13120 	 * One fundamental assumption here is that the length allocated for
13121 	 * both the local and remote sglists is identical.  Otherwise, we've
13122 	 * essentially got a coding error of some sort.
13123 	 */
13124 	for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
13125 		int isc_ret;
13126 		uint32_t cur_len, dma_length;
13127 		uint8_t *tmp_ptr;
13128 
13129 		rq->id = CTL_HA_DATA_CTL;
13130 		rq->command = command;
13131 		rq->context = io;
13132 
13133 		/*
13134 		 * Both pointers should be aligned.  But it is possible
13135 		 * that the allocation length is not.  They should both
13136 		 * also have enough slack left over at the end, though,
13137 		 * to round up to the next 8 byte boundary.
13138 		 */
13139 		cur_len = ctl_min(local_sglist[i].len - local_used,
13140 				  remote_sglist[j].len - remote_used);
13141 
13142 		/*
13143 		 * In this case, we have a size issue and need to decrease
13144 		 * the size, except in the case where we actually have less
13145 		 * than 8 bytes left.  In that case, we need to increase
13146 		 * the DMA length to get the last bit.
13147 		 */
13148 		if ((cur_len & 0x7) != 0) {
13149 			if (cur_len > 0x7) {
13150 				cur_len = cur_len - (cur_len & 0x7);
13151 				dma_length = cur_len;
13152 			} else {
13153 				CTL_SIZE_8B(dma_length, cur_len);
13154 			}
13155 
13156 		} else
13157 			dma_length = cur_len;
13158 
13159 		/*
13160 		 * If we had to allocate memory for this I/O, instead of using
13161 		 * the non-cached mirror memory, we'll need to flush the cache
13162 		 * before trying to DMA to the other controller.
13163 		 *
13164 		 * We could end up doing this multiple times for the same
13165 		 * segment if we have a larger local segment than remote
13166 		 * segment.  That shouldn't be an issue.
13167 		 */
13168 		if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
13169 			/*
13170 			 * XXX KDM use bus_dmamap_sync() here.
13171 			 */
13172 		}
13173 
13174 		rq->size = dma_length;
13175 
13176 		tmp_ptr = (uint8_t *)local_sglist[i].addr;
13177 		tmp_ptr += local_used;
13178 
13179 		/* Use physical addresses when talking to ISC hardware */
13180 		if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
13181 			/* XXX KDM use busdma */
13182 #if 0
13183 			rq->local = vtophys(tmp_ptr);
13184 #endif
13185 		} else
13186 			rq->local = tmp_ptr;
13187 
13188 		tmp_ptr = (uint8_t *)remote_sglist[j].addr;
13189 		tmp_ptr += remote_used;
13190 		rq->remote = tmp_ptr;
13191 
13192 		rq->callback = NULL;
13193 
13194 		local_used += cur_len;
13195 		if (local_used >= local_sglist[i].len) {
13196 			i++;
13197 			local_used = 0;
13198 		}
13199 
13200 		remote_used += cur_len;
13201 		if (remote_used >= remote_sglist[j].len) {
13202 			j++;
13203 			remote_used = 0;
13204 		}
13205 		total_used += cur_len;
13206 
13207 		if (total_used >= io->scsiio.kern_data_len)
13208 			rq->callback = callback;
13209 
13210 		if ((rq->size & 0x7) != 0) {
13211 			printf("%s: warning: size %d is not on 8b boundary\n",
13212 			       __func__, rq->size);
13213 		}
13214 		if (((uintptr_t)rq->local & 0x7) != 0) {
13215 			printf("%s: warning: local %p not on 8b boundary\n",
13216 			       __func__, rq->local);
13217 		}
13218 		if (((uintptr_t)rq->remote & 0x7) != 0) {
13219 			printf("%s: warning: remote %p not on 8b boundary\n",
13220 			       __func__, rq->local);
13221 		}
13222 #if 0
13223 		printf("%s: %s: local %#x remote %#x size %d\n", __func__,
13224 		       (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
13225 		       rq->local, rq->remote, rq->size);
13226 #endif
13227 
13228 		isc_ret = ctl_dt_single(rq);
13229 		if (isc_ret == CTL_HA_STATUS_WAIT)
13230 			continue;
13231 
13232 		if (isc_ret == CTL_HA_STATUS_DISCONNECT) {
13233 			rq->ret = CTL_HA_STATUS_SUCCESS;
13234 		} else {
13235 			rq->ret = isc_ret;
13236 		}
13237 		callback(rq);
13238 		goto bailout;
13239 	}
13240 
13241 bailout:
13242 	return (retval);
13243 
13244 }
13245 
13246 static void
13247 ctl_datamove_remote_read(union ctl_io *io)
13248 {
13249 	int retval;
13250 	int i;
13251 
13252 	/*
13253 	 * This will send an error to the other controller in the case of a
13254 	 * failure.
13255 	 */
13256 	retval = ctl_datamove_remote_sgl_setup(io);
13257 	if (retval != 0)
13258 		return;
13259 
13260 	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
13261 					  ctl_datamove_remote_read_cb);
13262 	if ((retval != 0)
13263 	 && ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0)) {
13264 		/*
13265 		 * Make sure we free memory if there was an error..  The
13266 		 * ctl_datamove_remote_xfer() function will send the
13267 		 * datamove done message, or call the callback with an
13268 		 * error if there is a problem.
13269 		 */
13270 		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13271 			free(io->io_hdr.local_sglist[i].addr, M_CTL);
13272 	}
13273 
13274 	return;
13275 }
13276 
13277 /*
13278  * Process a datamove request from the other controller.  This is used for
13279  * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
13280  * first.  Once that is complete, the data gets DMAed into the remote
13281  * controller's memory.  For reads, we DMA from the remote controller's
13282  * memory into our memory first, and then move it out to the FETD.
13283  */
13284 static void
13285 ctl_datamove_remote(union ctl_io *io)
13286 {
13287 	struct ctl_softc *softc;
13288 
13289 	softc = control_softc;
13290 
13291 	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
13292 
13293 	/*
13294 	 * Note that we look for an aborted I/O here, but don't do some of
13295 	 * the other checks that ctl_datamove() normally does.
13296 	 * We don't need to run the datamove delay code, since that should
13297 	 * have been done if need be on the other controller.
13298 	 */
13299 	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
13300 		printf("%s: tag 0x%04x on (%d:%d:%d:%d) aborted\n", __func__,
13301 		       io->scsiio.tag_num, io->io_hdr.nexus.initid.id,
13302 		       io->io_hdr.nexus.targ_port,
13303 		       io->io_hdr.nexus.targ_target.id,
13304 		       io->io_hdr.nexus.targ_lun);
13305 		io->io_hdr.port_status = 31338;
13306 		ctl_send_datamove_done(io, /*have_lock*/ 0);
13307 		return;
13308 	}
13309 
13310 	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT) {
13311 		ctl_datamove_remote_write(io);
13312 	} else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN){
13313 		ctl_datamove_remote_read(io);
13314 	} else {
13315 		union ctl_ha_msg msg;
13316 		struct scsi_sense_data *sense;
13317 		uint8_t sks[3];
13318 		int retry_count;
13319 
13320 		memset(&msg, 0, sizeof(msg));
13321 
13322 		msg.hdr.msg_type = CTL_MSG_BAD_JUJU;
13323 		msg.hdr.status = CTL_SCSI_ERROR;
13324 		msg.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
13325 
13326 		retry_count = 4243;
13327 
13328 		sense = &msg.scsi.sense_data;
13329 		sks[0] = SSD_SCS_VALID;
13330 		sks[1] = (retry_count >> 8) & 0xff;
13331 		sks[2] = retry_count & 0xff;
13332 
13333 		/* "Internal target failure" */
13334 		scsi_set_sense_data(sense,
13335 				    /*sense_format*/ SSD_TYPE_NONE,
13336 				    /*current_error*/ 1,
13337 				    /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
13338 				    /*asc*/ 0x44,
13339 				    /*ascq*/ 0x00,
13340 				    /*type*/ SSD_ELEM_SKS,
13341 				    /*size*/ sizeof(sks),
13342 				    /*data*/ sks,
13343 				    SSD_ELEM_NONE);
13344 
13345 		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
13346 		if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
13347 			ctl_failover_io(io, /*have_lock*/ 1);
13348 			return;
13349 		}
13350 
13351 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0) >
13352 		    CTL_HA_STATUS_SUCCESS) {
13353 			/* XXX KDM what to do if this fails? */
13354 		}
13355 		return;
13356 	}
13357 
13358 }
13359 
13360 static int
13361 ctl_process_done(union ctl_io *io)
13362 {
13363 	struct ctl_lun *lun;
13364 	struct ctl_softc *ctl_softc;
13365 	void (*fe_done)(union ctl_io *io);
13366 	uint32_t targ_port = ctl_port_idx(io->io_hdr.nexus.targ_port);
13367 
13368 	CTL_DEBUG_PRINT(("ctl_process_done\n"));
13369 
13370 	fe_done =
13371 	    control_softc->ctl_ports[targ_port]->fe_done;
13372 
13373 #ifdef CTL_TIME_IO
13374 	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
13375 		char str[256];
13376 		char path_str[64];
13377 		struct sbuf sb;
13378 
13379 		ctl_scsi_path_string(io, path_str, sizeof(path_str));
13380 		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13381 
13382 		sbuf_cat(&sb, path_str);
13383 		switch (io->io_hdr.io_type) {
13384 		case CTL_IO_SCSI:
13385 			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
13386 			sbuf_printf(&sb, "\n");
13387 			sbuf_cat(&sb, path_str);
13388 			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13389 				    io->scsiio.tag_num, io->scsiio.tag_type);
13390 			break;
13391 		case CTL_IO_TASK:
13392 			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
13393 				    "Tag Type: %d\n", io->taskio.task_action,
13394 				    io->taskio.tag_num, io->taskio.tag_type);
13395 			break;
13396 		default:
13397 			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13398 			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13399 			break;
13400 		}
13401 		sbuf_cat(&sb, path_str);
13402 		sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
13403 			    (intmax_t)time_uptime - io->io_hdr.start_time);
13404 		sbuf_finish(&sb);
13405 		printf("%s", sbuf_data(&sb));
13406 	}
13407 #endif /* CTL_TIME_IO */
13408 
13409 	switch (io->io_hdr.io_type) {
13410 	case CTL_IO_SCSI:
13411 		break;
13412 	case CTL_IO_TASK:
13413 		if (bootverbose || verbose > 0)
13414 			ctl_io_error_print(io, NULL);
13415 		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
13416 			ctl_free_io(io);
13417 		else
13418 			fe_done(io);
13419 		return (CTL_RETVAL_COMPLETE);
13420 		break;
13421 	default:
13422 		printf("ctl_process_done: invalid io type %d\n",
13423 		       io->io_hdr.io_type);
13424 		panic("ctl_process_done: invalid io type %d\n",
13425 		      io->io_hdr.io_type);
13426 		break; /* NOTREACHED */
13427 	}
13428 
13429 	lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13430 	if (lun == NULL) {
13431 		CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
13432 				 io->io_hdr.nexus.targ_mapped_lun));
13433 		fe_done(io);
13434 		goto bailout;
13435 	}
13436 	ctl_softc = lun->ctl_softc;
13437 
13438 	mtx_lock(&lun->lun_lock);
13439 
13440 	/*
13441 	 * Check to see if we have any errors to inject here.  We only
13442 	 * inject errors for commands that don't already have errors set.
13443 	 */
13444 	if ((STAILQ_FIRST(&lun->error_list) != NULL)
13445 	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))
13446 		ctl_inject_error(lun, io);
13447 
13448 	/*
13449 	 * XXX KDM how do we treat commands that aren't completed
13450 	 * successfully?
13451 	 *
13452 	 * XXX KDM should we also track I/O latency?
13453 	 */
13454 	if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
13455 	    io->io_hdr.io_type == CTL_IO_SCSI) {
13456 #ifdef CTL_TIME_IO
13457 		struct bintime cur_bt;
13458 #endif
13459 		int type;
13460 
13461 		if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13462 		    CTL_FLAG_DATA_IN)
13463 			type = CTL_STATS_READ;
13464 		else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13465 		    CTL_FLAG_DATA_OUT)
13466 			type = CTL_STATS_WRITE;
13467 		else
13468 			type = CTL_STATS_NO_IO;
13469 
13470 		lun->stats.ports[targ_port].bytes[type] +=
13471 		    io->scsiio.kern_total_len;
13472 		lun->stats.ports[targ_port].operations[type]++;
13473 #ifdef CTL_TIME_IO
13474 		bintime_add(&lun->stats.ports[targ_port].dma_time[type],
13475 		   &io->io_hdr.dma_bt);
13476 		lun->stats.ports[targ_port].num_dmas[type] +=
13477 		    io->io_hdr.num_dmas;
13478 		getbintime(&cur_bt);
13479 		bintime_sub(&cur_bt, &io->io_hdr.start_bt);
13480 		bintime_add(&lun->stats.ports[targ_port].time[type], &cur_bt);
13481 #endif
13482 	}
13483 
13484 	/*
13485 	 * Remove this from the OOA queue.
13486 	 */
13487 	TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13488 
13489 	/*
13490 	 * Run through the blocked queue on this LUN and see if anything
13491 	 * has become unblocked, now that this transaction is done.
13492 	 */
13493 	ctl_check_blocked(lun);
13494 
13495 	/*
13496 	 * If the LUN has been invalidated, free it if there is nothing
13497 	 * left on its OOA queue.
13498 	 */
13499 	if ((lun->flags & CTL_LUN_INVALID)
13500 	 && TAILQ_EMPTY(&lun->ooa_queue)) {
13501 		mtx_unlock(&lun->lun_lock);
13502 		mtx_lock(&ctl_softc->ctl_lock);
13503 		ctl_free_lun(lun);
13504 		mtx_unlock(&ctl_softc->ctl_lock);
13505 	} else
13506 		mtx_unlock(&lun->lun_lock);
13507 
13508 	/*
13509 	 * If this command has been aborted, make sure we set the status
13510 	 * properly.  The FETD is responsible for freeing the I/O and doing
13511 	 * whatever it needs to do to clean up its state.
13512 	 */
13513 	if (io->io_hdr.flags & CTL_FLAG_ABORT)
13514 		ctl_set_task_aborted(&io->scsiio);
13515 
13516 	/*
13517 	 * We print out status for every task management command.  For SCSI
13518 	 * commands, we filter out any unit attention errors; they happen
13519 	 * on every boot, and would clutter up the log.  Note:  task
13520 	 * management commands aren't printed here, they are printed above,
13521 	 * since they should never even make it down here.
13522 	 */
13523 	switch (io->io_hdr.io_type) {
13524 	case CTL_IO_SCSI: {
13525 		int error_code, sense_key, asc, ascq;
13526 
13527 		sense_key = 0;
13528 
13529 		if (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SCSI_ERROR)
13530 		 && (io->scsiio.scsi_status == SCSI_STATUS_CHECK_COND)) {
13531 			/*
13532 			 * Since this is just for printing, no need to
13533 			 * show errors here.
13534 			 */
13535 			scsi_extract_sense_len(&io->scsiio.sense_data,
13536 					       io->scsiio.sense_len,
13537 					       &error_code,
13538 					       &sense_key,
13539 					       &asc,
13540 					       &ascq,
13541 					       /*show_errors*/ 0);
13542 		}
13543 
13544 		if (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
13545 		 && (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SCSI_ERROR)
13546 		  || (io->scsiio.scsi_status != SCSI_STATUS_CHECK_COND)
13547 		  || (sense_key != SSD_KEY_UNIT_ATTENTION))) {
13548 
13549 			if ((time_uptime - ctl_softc->last_print_jiffies) <= 0){
13550 				ctl_softc->skipped_prints++;
13551 			} else {
13552 				uint32_t skipped_prints;
13553 
13554 				skipped_prints = ctl_softc->skipped_prints;
13555 
13556 				ctl_softc->skipped_prints = 0;
13557 				ctl_softc->last_print_jiffies = time_uptime;
13558 
13559 				if (skipped_prints > 0) {
13560 #ifdef NEEDTOPORT
13561 					csevent_log(CSC_CTL | CSC_SHELF_SW |
13562 					    CTL_ERROR_REPORT,
13563 					    csevent_LogType_Trace,
13564 					    csevent_Severity_Information,
13565 					    csevent_AlertLevel_Green,
13566 					    csevent_FRU_Firmware,
13567 					    csevent_FRU_Unknown,
13568 					    "High CTL error volume, %d prints "
13569 					    "skipped", skipped_prints);
13570 #endif
13571 				}
13572 				if (bootverbose || verbose > 0)
13573 					ctl_io_error_print(io, NULL);
13574 			}
13575 		}
13576 		break;
13577 	}
13578 	case CTL_IO_TASK:
13579 		if (bootverbose || verbose > 0)
13580 			ctl_io_error_print(io, NULL);
13581 		break;
13582 	default:
13583 		break;
13584 	}
13585 
13586 	/*
13587 	 * Tell the FETD or the other shelf controller we're done with this
13588 	 * command.  Note that only SCSI commands get to this point.  Task
13589 	 * management commands are completed above.
13590 	 *
13591 	 * We only send status to the other controller if we're in XFER
13592 	 * mode.  In SER_ONLY mode, the I/O is done on the controller that
13593 	 * received the I/O (from CTL's perspective), and so the status is
13594 	 * generated there.
13595 	 *
13596 	 * XXX KDM if we hold the lock here, we could cause a deadlock
13597 	 * if the frontend comes back in in this context to queue
13598 	 * something.
13599 	 */
13600 	if ((ctl_softc->ha_mode == CTL_HA_MODE_XFER)
13601 	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
13602 		union ctl_ha_msg msg;
13603 
13604 		memset(&msg, 0, sizeof(msg));
13605 		msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13606 		msg.hdr.original_sc = io->io_hdr.original_sc;
13607 		msg.hdr.nexus = io->io_hdr.nexus;
13608 		msg.hdr.status = io->io_hdr.status;
13609 		msg.scsi.scsi_status = io->scsiio.scsi_status;
13610 		msg.scsi.tag_num = io->scsiio.tag_num;
13611 		msg.scsi.tag_type = io->scsiio.tag_type;
13612 		msg.scsi.sense_len = io->scsiio.sense_len;
13613 		msg.scsi.sense_residual = io->scsiio.sense_residual;
13614 		msg.scsi.residual = io->scsiio.residual;
13615 		memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
13616 		       sizeof(io->scsiio.sense_data));
13617 		/*
13618 		 * We copy this whether or not this is an I/O-related
13619 		 * command.  Otherwise, we'd have to go and check to see
13620 		 * whether it's a read/write command, and it really isn't
13621 		 * worth it.
13622 		 */
13623 		memcpy(&msg.scsi.lbalen,
13624 		       &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
13625 		       sizeof(msg.scsi.lbalen));
13626 
13627 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13628 				sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
13629 			/* XXX do something here */
13630 		}
13631 
13632 		ctl_free_io(io);
13633 	} else
13634 		fe_done(io);
13635 
13636 bailout:
13637 
13638 	return (CTL_RETVAL_COMPLETE);
13639 }
13640 
13641 #ifdef CTL_WITH_CA
13642 /*
13643  * Front end should call this if it doesn't do autosense.  When the request
13644  * sense comes back in from the initiator, we'll dequeue this and send it.
13645  */
13646 int
13647 ctl_queue_sense(union ctl_io *io)
13648 {
13649 	struct ctl_lun *lun;
13650 	struct ctl_softc *ctl_softc;
13651 	uint32_t initidx, targ_lun;
13652 
13653 	ctl_softc = control_softc;
13654 
13655 	CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13656 
13657 	/*
13658 	 * LUN lookup will likely move to the ctl_work_thread() once we
13659 	 * have our new queueing infrastructure (that doesn't put things on
13660 	 * a per-LUN queue initially).  That is so that we can handle
13661 	 * things like an INQUIRY to a LUN that we don't have enabled.  We
13662 	 * can't deal with that right now.
13663 	 */
13664 	mtx_lock(&ctl_softc->ctl_lock);
13665 
13666 	/*
13667 	 * If we don't have a LUN for this, just toss the sense
13668 	 * information.
13669 	 */
13670 	targ_lun = io->io_hdr.nexus.targ_lun;
13671 	targ_lun = ctl_map_lun(io->io_hdr.nexus.targ_port, targ_lun);
13672 	if ((targ_lun < CTL_MAX_LUNS)
13673 	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
13674 		lun = ctl_softc->ctl_luns[targ_lun];
13675 	else
13676 		goto bailout;
13677 
13678 	initidx = ctl_get_initindex(&io->io_hdr.nexus);
13679 
13680 	mtx_lock(&lun->lun_lock);
13681 	/*
13682 	 * Already have CA set for this LUN...toss the sense information.
13683 	 */
13684 	if (ctl_is_set(lun->have_ca, initidx)) {
13685 		mtx_unlock(&lun->lun_lock);
13686 		goto bailout;
13687 	}
13688 
13689 	memcpy(&lun->pending_sense[initidx], &io->scsiio.sense_data,
13690 	       ctl_min(sizeof(lun->pending_sense[initidx]),
13691 	       sizeof(io->scsiio.sense_data)));
13692 	ctl_set_mask(lun->have_ca, initidx);
13693 	mtx_unlock(&lun->lun_lock);
13694 
13695 bailout:
13696 	mtx_unlock(&ctl_softc->ctl_lock);
13697 
13698 	ctl_free_io(io);
13699 
13700 	return (CTL_RETVAL_COMPLETE);
13701 }
13702 #endif
13703 
13704 /*
13705  * Primary command inlet from frontend ports.  All SCSI and task I/O
13706  * requests must go through this function.
13707  */
13708 int
13709 ctl_queue(union ctl_io *io)
13710 {
13711 	struct ctl_softc *ctl_softc;
13712 
13713 	CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13714 
13715 	ctl_softc = control_softc;
13716 
13717 #ifdef CTL_TIME_IO
13718 	io->io_hdr.start_time = time_uptime;
13719 	getbintime(&io->io_hdr.start_bt);
13720 #endif /* CTL_TIME_IO */
13721 
13722 	/* Map FE-specific LUN ID into global one. */
13723 	io->io_hdr.nexus.targ_mapped_lun =
13724 	    ctl_map_lun(io->io_hdr.nexus.targ_port, io->io_hdr.nexus.targ_lun);
13725 
13726 	switch (io->io_hdr.io_type) {
13727 	case CTL_IO_SCSI:
13728 	case CTL_IO_TASK:
13729 		ctl_enqueue_incoming(io);
13730 		break;
13731 	default:
13732 		printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13733 		return (EINVAL);
13734 	}
13735 
13736 	return (CTL_RETVAL_COMPLETE);
13737 }
13738 
13739 #ifdef CTL_IO_DELAY
13740 static void
13741 ctl_done_timer_wakeup(void *arg)
13742 {
13743 	union ctl_io *io;
13744 
13745 	io = (union ctl_io *)arg;
13746 	ctl_done(io);
13747 }
13748 #endif /* CTL_IO_DELAY */
13749 
13750 void
13751 ctl_done(union ctl_io *io)
13752 {
13753 	struct ctl_softc *ctl_softc;
13754 
13755 	ctl_softc = control_softc;
13756 
13757 	/*
13758 	 * Enable this to catch duplicate completion issues.
13759 	 */
13760 #if 0
13761 	if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13762 		printf("%s: type %d msg %d cdb %x iptl: "
13763 		       "%d:%d:%d:%d tag 0x%04x "
13764 		       "flag %#x status %x\n",
13765 			__func__,
13766 			io->io_hdr.io_type,
13767 			io->io_hdr.msg_type,
13768 			io->scsiio.cdb[0],
13769 			io->io_hdr.nexus.initid.id,
13770 			io->io_hdr.nexus.targ_port,
13771 			io->io_hdr.nexus.targ_target.id,
13772 			io->io_hdr.nexus.targ_lun,
13773 			(io->io_hdr.io_type ==
13774 			CTL_IO_TASK) ?
13775 			io->taskio.tag_num :
13776 			io->scsiio.tag_num,
13777 		        io->io_hdr.flags,
13778 			io->io_hdr.status);
13779 	} else
13780 		io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13781 #endif
13782 
13783 	/*
13784 	 * This is an internal copy of an I/O, and should not go through
13785 	 * the normal done processing logic.
13786 	 */
13787 	if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
13788 		return;
13789 
13790 	/*
13791 	 * We need to send a msg to the serializing shelf to finish the IO
13792 	 * as well.  We don't send a finish message to the other shelf if
13793 	 * this is a task management command.  Task management commands
13794 	 * aren't serialized in the OOA queue, but rather just executed on
13795 	 * both shelf controllers for commands that originated on that
13796 	 * controller.
13797 	 */
13798 	if ((io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)
13799 	 && (io->io_hdr.io_type != CTL_IO_TASK)) {
13800 		union ctl_ha_msg msg_io;
13801 
13802 		msg_io.hdr.msg_type = CTL_MSG_FINISH_IO;
13803 		msg_io.hdr.serializing_sc = io->io_hdr.serializing_sc;
13804 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_io,
13805 		    sizeof(msg_io), 0 ) != CTL_HA_STATUS_SUCCESS) {
13806 		}
13807 		/* continue on to finish IO */
13808 	}
13809 #ifdef CTL_IO_DELAY
13810 	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13811 		struct ctl_lun *lun;
13812 
13813 		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13814 
13815 		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13816 	} else {
13817 		struct ctl_lun *lun;
13818 
13819 		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13820 
13821 		if ((lun != NULL)
13822 		 && (lun->delay_info.done_delay > 0)) {
13823 			struct callout *callout;
13824 
13825 			callout = (struct callout *)&io->io_hdr.timer_bytes;
13826 			callout_init(callout, /*mpsafe*/ 1);
13827 			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13828 			callout_reset(callout,
13829 				      lun->delay_info.done_delay * hz,
13830 				      ctl_done_timer_wakeup, io);
13831 			if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13832 				lun->delay_info.done_delay = 0;
13833 			return;
13834 		}
13835 	}
13836 #endif /* CTL_IO_DELAY */
13837 
13838 	ctl_enqueue_done(io);
13839 }
13840 
13841 int
13842 ctl_isc(struct ctl_scsiio *ctsio)
13843 {
13844 	struct ctl_lun *lun;
13845 	int retval;
13846 
13847 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13848 
13849 	CTL_DEBUG_PRINT(("ctl_isc: command: %02x\n", ctsio->cdb[0]));
13850 
13851 	CTL_DEBUG_PRINT(("ctl_isc: calling data_submit()\n"));
13852 
13853 	retval = lun->backend->data_submit((union ctl_io *)ctsio);
13854 
13855 	return (retval);
13856 }
13857 
13858 
13859 static void
13860 ctl_work_thread(void *arg)
13861 {
13862 	struct ctl_thread *thr = (struct ctl_thread *)arg;
13863 	struct ctl_softc *softc = thr->ctl_softc;
13864 	union ctl_io *io;
13865 	int retval;
13866 
13867 	CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13868 
13869 	for (;;) {
13870 		retval = 0;
13871 
13872 		/*
13873 		 * We handle the queues in this order:
13874 		 * - ISC
13875 		 * - done queue (to free up resources, unblock other commands)
13876 		 * - RtR queue
13877 		 * - incoming queue
13878 		 *
13879 		 * If those queues are empty, we break out of the loop and
13880 		 * go to sleep.
13881 		 */
13882 		mtx_lock(&thr->queue_lock);
13883 		io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
13884 		if (io != NULL) {
13885 			STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
13886 			mtx_unlock(&thr->queue_lock);
13887 			ctl_handle_isc(io);
13888 			continue;
13889 		}
13890 		io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
13891 		if (io != NULL) {
13892 			STAILQ_REMOVE_HEAD(&thr->done_queue, links);
13893 			/* clear any blocked commands, call fe_done */
13894 			mtx_unlock(&thr->queue_lock);
13895 			retval = ctl_process_done(io);
13896 			continue;
13897 		}
13898 		io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
13899 		if (io != NULL) {
13900 			STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
13901 			mtx_unlock(&thr->queue_lock);
13902 			if (io->io_hdr.io_type == CTL_IO_TASK)
13903 				ctl_run_task(io);
13904 			else
13905 				ctl_scsiio_precheck(softc, &io->scsiio);
13906 			continue;
13907 		}
13908 		if (!ctl_pause_rtr) {
13909 			io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
13910 			if (io != NULL) {
13911 				STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
13912 				mtx_unlock(&thr->queue_lock);
13913 				retval = ctl_scsiio(&io->scsiio);
13914 				if (retval != CTL_RETVAL_COMPLETE)
13915 					CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13916 				continue;
13917 			}
13918 		}
13919 
13920 		/* Sleep until we have something to do. */
13921 		mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
13922 	}
13923 }
13924 
13925 static void
13926 ctl_lun_thread(void *arg)
13927 {
13928 	struct ctl_softc *softc = (struct ctl_softc *)arg;
13929 	struct ctl_be_lun *be_lun;
13930 	int retval;
13931 
13932 	CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
13933 
13934 	for (;;) {
13935 		retval = 0;
13936 		mtx_lock(&softc->ctl_lock);
13937 		be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
13938 		if (be_lun != NULL) {
13939 			STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
13940 			mtx_unlock(&softc->ctl_lock);
13941 			ctl_create_lun(be_lun);
13942 			continue;
13943 		}
13944 
13945 		/* Sleep until we have something to do. */
13946 		mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
13947 		    PDROP | PRIBIO, "-", 0);
13948 	}
13949 }
13950 
13951 static void
13952 ctl_enqueue_incoming(union ctl_io *io)
13953 {
13954 	struct ctl_softc *softc = control_softc;
13955 	struct ctl_thread *thr;
13956 	u_int idx;
13957 
13958 	idx = (io->io_hdr.nexus.targ_port * 127 +
13959 	       io->io_hdr.nexus.initid.id) % worker_threads;
13960 	thr = &softc->threads[idx];
13961 	mtx_lock(&thr->queue_lock);
13962 	STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
13963 	mtx_unlock(&thr->queue_lock);
13964 	wakeup(thr);
13965 }
13966 
13967 static void
13968 ctl_enqueue_rtr(union ctl_io *io)
13969 {
13970 	struct ctl_softc *softc = control_softc;
13971 	struct ctl_thread *thr;
13972 
13973 	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13974 	mtx_lock(&thr->queue_lock);
13975 	STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
13976 	mtx_unlock(&thr->queue_lock);
13977 	wakeup(thr);
13978 }
13979 
13980 static void
13981 ctl_enqueue_done(union ctl_io *io)
13982 {
13983 	struct ctl_softc *softc = control_softc;
13984 	struct ctl_thread *thr;
13985 
13986 	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13987 	mtx_lock(&thr->queue_lock);
13988 	STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
13989 	mtx_unlock(&thr->queue_lock);
13990 	wakeup(thr);
13991 }
13992 
13993 static void
13994 ctl_enqueue_isc(union ctl_io *io)
13995 {
13996 	struct ctl_softc *softc = control_softc;
13997 	struct ctl_thread *thr;
13998 
13999 	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14000 	mtx_lock(&thr->queue_lock);
14001 	STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
14002 	mtx_unlock(&thr->queue_lock);
14003 	wakeup(thr);
14004 }
14005 
14006 /* Initialization and failover */
14007 
14008 void
14009 ctl_init_isc_msg(void)
14010 {
14011 	printf("CTL: Still calling this thing\n");
14012 }
14013 
14014 /*
14015  * Init component
14016  * 	Initializes component into configuration defined by bootMode
14017  *	(see hasc-sv.c)
14018  *  	returns hasc_Status:
14019  * 		OK
14020  *		ERROR - fatal error
14021  */
14022 static ctl_ha_comp_status
14023 ctl_isc_init(struct ctl_ha_component *c)
14024 {
14025 	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
14026 
14027 	c->status = ret;
14028 	return ret;
14029 }
14030 
14031 /* Start component
14032  * 	Starts component in state requested. If component starts successfully,
14033  *	it must set its own state to the requestrd state
14034  *	When requested state is HASC_STATE_HA, the component may refine it
14035  * 	by adding _SLAVE or _MASTER flags.
14036  *	Currently allowed state transitions are:
14037  *	UNKNOWN->HA		- initial startup
14038  *	UNKNOWN->SINGLE - initial startup when no parter detected
14039  *	HA->SINGLE		- failover
14040  * returns ctl_ha_comp_status:
14041  * 		OK	- component successfully started in requested state
14042  *		FAILED  - could not start the requested state, failover may
14043  * 			  be possible
14044  *		ERROR	- fatal error detected, no future startup possible
14045  */
14046 static ctl_ha_comp_status
14047 ctl_isc_start(struct ctl_ha_component *c, ctl_ha_state state)
14048 {
14049 	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
14050 
14051 	printf("%s: go\n", __func__);
14052 
14053 	// UNKNOWN->HA or UNKNOWN->SINGLE (bootstrap)
14054 	if (c->state == CTL_HA_STATE_UNKNOWN ) {
14055 		ctl_is_single = 0;
14056 		if (ctl_ha_msg_create(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
14057 		    != CTL_HA_STATUS_SUCCESS) {
14058 			printf("ctl_isc_start: ctl_ha_msg_create failed.\n");
14059 			ret = CTL_HA_COMP_STATUS_ERROR;
14060 		}
14061 	} else if (CTL_HA_STATE_IS_HA(c->state)
14062 		&& CTL_HA_STATE_IS_SINGLE(state)){
14063 		// HA->SINGLE transition
14064 	        ctl_failover();
14065 		ctl_is_single = 1;
14066 	} else {
14067 		printf("ctl_isc_start:Invalid state transition %X->%X\n",
14068 		       c->state, state);
14069 		ret = CTL_HA_COMP_STATUS_ERROR;
14070 	}
14071 	if (CTL_HA_STATE_IS_SINGLE(state))
14072 		ctl_is_single = 1;
14073 
14074 	c->state = state;
14075 	c->status = ret;
14076 	return ret;
14077 }
14078 
14079 /*
14080  * Quiesce component
14081  * The component must clear any error conditions (set status to OK) and
14082  * prepare itself to another Start call
14083  * returns ctl_ha_comp_status:
14084  * 	OK
14085  *	ERROR
14086  */
14087 static ctl_ha_comp_status
14088 ctl_isc_quiesce(struct ctl_ha_component *c)
14089 {
14090 	int ret = CTL_HA_COMP_STATUS_OK;
14091 
14092 	ctl_pause_rtr = 1;
14093 	c->status = ret;
14094 	return ret;
14095 }
14096 
14097 struct ctl_ha_component ctl_ha_component_ctlisc =
14098 {
14099 	.name = "CTL ISC",
14100 	.state = CTL_HA_STATE_UNKNOWN,
14101 	.init = ctl_isc_init,
14102 	.start = ctl_isc_start,
14103 	.quiesce = ctl_isc_quiesce
14104 };
14105 
14106 /*
14107  *  vim: ts=8
14108  */
14109