xref: /freebsd/sys/cam/ctl/ctl.c (revision 5bd73b51076b5cb5a2c9810f76c1d7ed20c4460e)
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*/ SCP_WCE | SCP_RCD,
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*/SCP_QUEUE_ALG_RESTRICTED,
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*/SCP_QUEUE_ALG_MASK,
296 	/*eca_and_aen*/SCP_SWP,
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  * Extended INQUIRY Data (0x86), Mode Page Policy (0x87),
324  * SCSI Ports (0x88), Third-party Copy (0x8F), Block limits (0xB0),
325  * Block Device Characteristics (0xB1) and Logical Block Provisioning (0xB2)
326  */
327 #define SCSI_EVPD_NUM_SUPPORTED_PAGES	10
328 
329 static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
330 				  int param);
331 static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
332 static int ctl_init(void);
333 void ctl_shutdown(void);
334 static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
335 static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
336 static void ctl_ioctl_online(void *arg);
337 static void ctl_ioctl_offline(void *arg);
338 static int ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id);
339 static int ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id);
340 static int ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio);
341 static int ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio);
342 static int ctl_ioctl_submit_wait(union ctl_io *io);
343 static void ctl_ioctl_datamove(union ctl_io *io);
344 static void ctl_ioctl_done(union ctl_io *io);
345 static void ctl_ioctl_hard_startstop_callback(void *arg,
346 					      struct cfi_metatask *metatask);
347 static void ctl_ioctl_bbrread_callback(void *arg,struct cfi_metatask *metatask);
348 static int ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
349 			      struct ctl_ooa *ooa_hdr,
350 			      struct ctl_ooa_entry *kern_entries);
351 static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
352 		     struct thread *td);
353 static uint32_t ctl_map_lun(int port_num, uint32_t lun);
354 static uint32_t ctl_map_lun_back(int port_num, uint32_t lun);
355 #ifdef unused
356 static union ctl_io *ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port,
357 				   uint32_t targ_target, uint32_t targ_lun,
358 				   int can_wait);
359 static void ctl_kfree_io(union ctl_io *io);
360 #endif /* unused */
361 static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
362 			 struct ctl_be_lun *be_lun, struct ctl_id target_id);
363 static int ctl_free_lun(struct ctl_lun *lun);
364 static void ctl_create_lun(struct ctl_be_lun *be_lun);
365 /**
366 static void ctl_failover_change_pages(struct ctl_softc *softc,
367 				      struct ctl_scsiio *ctsio, int master);
368 **/
369 
370 static int ctl_do_mode_select(union ctl_io *io);
371 static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
372 			   uint64_t res_key, uint64_t sa_res_key,
373 			   uint8_t type, uint32_t residx,
374 			   struct ctl_scsiio *ctsio,
375 			   struct scsi_per_res_out *cdb,
376 			   struct scsi_per_res_out_parms* param);
377 static void ctl_pro_preempt_other(struct ctl_lun *lun,
378 				  union ctl_ha_msg *msg);
379 static void ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg);
380 static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
381 static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
382 static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
383 static int ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len);
384 static int ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len);
385 static int ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio,
386 					 int alloc_len);
387 static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
388 					 int alloc_len);
389 static int ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len);
390 static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
391 static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
392 static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
393 static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len);
394 static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2);
395 static ctl_action ctl_check_for_blockage(struct ctl_lun *lun,
396     union ctl_io *pending_io, union ctl_io *ooa_io);
397 static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
398 				union ctl_io *starting_io);
399 static int ctl_check_blocked(struct ctl_lun *lun);
400 static int ctl_scsiio_lun_check(struct ctl_softc *ctl_softc,
401 				struct ctl_lun *lun,
402 				const struct ctl_cmd_entry *entry,
403 				struct ctl_scsiio *ctsio);
404 //static int ctl_check_rtr(union ctl_io *pending_io, struct ctl_softc *softc);
405 static void ctl_failover(void);
406 static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
407 			       struct ctl_scsiio *ctsio);
408 static int ctl_scsiio(struct ctl_scsiio *ctsio);
409 
410 static int ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
411 static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
412 			    ctl_ua_type ua_type);
413 static int ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io,
414 			 ctl_ua_type ua_type);
415 static int ctl_abort_task(union ctl_io *io);
416 static int ctl_abort_task_set(union ctl_io *io);
417 static int ctl_i_t_nexus_reset(union ctl_io *io);
418 static void ctl_run_task(union ctl_io *io);
419 #ifdef CTL_IO_DELAY
420 static void ctl_datamove_timer_wakeup(void *arg);
421 static void ctl_done_timer_wakeup(void *arg);
422 #endif /* CTL_IO_DELAY */
423 
424 static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
425 static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
426 static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
427 static void ctl_datamove_remote_write(union ctl_io *io);
428 static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
429 static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
430 static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
431 static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
432 				    ctl_ha_dt_cb callback);
433 static void ctl_datamove_remote_read(union ctl_io *io);
434 static void ctl_datamove_remote(union ctl_io *io);
435 static int ctl_process_done(union ctl_io *io);
436 static void ctl_lun_thread(void *arg);
437 static void ctl_work_thread(void *arg);
438 static void ctl_enqueue_incoming(union ctl_io *io);
439 static void ctl_enqueue_rtr(union ctl_io *io);
440 static void ctl_enqueue_done(union ctl_io *io);
441 static void ctl_enqueue_isc(union ctl_io *io);
442 static const struct ctl_cmd_entry *
443     ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa);
444 static const struct ctl_cmd_entry *
445     ctl_validate_command(struct ctl_scsiio *ctsio);
446 static int ctl_cmd_applicable(uint8_t lun_type,
447     const struct ctl_cmd_entry *entry);
448 
449 /*
450  * Load the serialization table.  This isn't very pretty, but is probably
451  * the easiest way to do it.
452  */
453 #include "ctl_ser_table.c"
454 
455 /*
456  * We only need to define open, close and ioctl routines for this driver.
457  */
458 static struct cdevsw ctl_cdevsw = {
459 	.d_version =	D_VERSION,
460 	.d_flags =	0,
461 	.d_open =	ctl_open,
462 	.d_close =	ctl_close,
463 	.d_ioctl =	ctl_ioctl,
464 	.d_name =	"ctl",
465 };
466 
467 
468 MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
469 MALLOC_DEFINE(M_CTLIO, "ctlio", "Memory used for CTL requests");
470 
471 static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
472 
473 static moduledata_t ctl_moduledata = {
474 	"ctl",
475 	ctl_module_event_handler,
476 	NULL
477 };
478 
479 DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
480 MODULE_VERSION(ctl, 1);
481 
482 static struct ctl_frontend ioctl_frontend =
483 {
484 	.name = "ioctl",
485 };
486 
487 static void
488 ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
489 			    union ctl_ha_msg *msg_info)
490 {
491 	struct ctl_scsiio *ctsio;
492 
493 	if (msg_info->hdr.original_sc == NULL) {
494 		printf("%s: original_sc == NULL!\n", __func__);
495 		/* XXX KDM now what? */
496 		return;
497 	}
498 
499 	ctsio = &msg_info->hdr.original_sc->scsiio;
500 	ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
501 	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
502 	ctsio->io_hdr.status = msg_info->hdr.status;
503 	ctsio->scsi_status = msg_info->scsi.scsi_status;
504 	ctsio->sense_len = msg_info->scsi.sense_len;
505 	ctsio->sense_residual = msg_info->scsi.sense_residual;
506 	ctsio->residual = msg_info->scsi.residual;
507 	memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
508 	       sizeof(ctsio->sense_data));
509 	memcpy(&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
510 	       &msg_info->scsi.lbalen, sizeof(msg_info->scsi.lbalen));
511 	ctl_enqueue_isc((union ctl_io *)ctsio);
512 }
513 
514 static void
515 ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
516 				union ctl_ha_msg *msg_info)
517 {
518 	struct ctl_scsiio *ctsio;
519 
520 	if (msg_info->hdr.serializing_sc == NULL) {
521 		printf("%s: serializing_sc == NULL!\n", __func__);
522 		/* XXX KDM now what? */
523 		return;
524 	}
525 
526 	ctsio = &msg_info->hdr.serializing_sc->scsiio;
527 #if 0
528 	/*
529 	 * Attempt to catch the situation where an I/O has
530 	 * been freed, and we're using it again.
531 	 */
532 	if (ctsio->io_hdr.io_type == 0xff) {
533 		union ctl_io *tmp_io;
534 		tmp_io = (union ctl_io *)ctsio;
535 		printf("%s: %p use after free!\n", __func__,
536 		       ctsio);
537 		printf("%s: type %d msg %d cdb %x iptl: "
538 		       "%d:%d:%d:%d tag 0x%04x "
539 		       "flag %#x status %x\n",
540 			__func__,
541 			tmp_io->io_hdr.io_type,
542 			tmp_io->io_hdr.msg_type,
543 			tmp_io->scsiio.cdb[0],
544 			tmp_io->io_hdr.nexus.initid.id,
545 			tmp_io->io_hdr.nexus.targ_port,
546 			tmp_io->io_hdr.nexus.targ_target.id,
547 			tmp_io->io_hdr.nexus.targ_lun,
548 			(tmp_io->io_hdr.io_type ==
549 			CTL_IO_TASK) ?
550 			tmp_io->taskio.tag_num :
551 			tmp_io->scsiio.tag_num,
552 		        tmp_io->io_hdr.flags,
553 			tmp_io->io_hdr.status);
554 	}
555 #endif
556 	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
557 	ctl_enqueue_isc((union ctl_io *)ctsio);
558 }
559 
560 /*
561  * ISC (Inter Shelf Communication) event handler.  Events from the HA
562  * subsystem come in here.
563  */
564 static void
565 ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
566 {
567 	struct ctl_softc *ctl_softc;
568 	union ctl_io *io;
569 	struct ctl_prio *presio;
570 	ctl_ha_status isc_status;
571 
572 	ctl_softc = control_softc;
573 	io = NULL;
574 
575 
576 #if 0
577 	printf("CTL: Isc Msg event %d\n", event);
578 #endif
579 	if (event == CTL_HA_EVT_MSG_RECV) {
580 		union ctl_ha_msg msg_info;
581 
582 		isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
583 					     sizeof(msg_info), /*wait*/ 0);
584 #if 0
585 		printf("CTL: msg_type %d\n", msg_info.msg_type);
586 #endif
587 		if (isc_status != 0) {
588 			printf("Error receiving message, status = %d\n",
589 			       isc_status);
590 			return;
591 		}
592 
593 		switch (msg_info.hdr.msg_type) {
594 		case CTL_MSG_SERIALIZE:
595 #if 0
596 			printf("Serialize\n");
597 #endif
598 			io = ctl_alloc_io((void *)ctl_softc->othersc_pool);
599 			if (io == NULL) {
600 				printf("ctl_isc_event_handler: can't allocate "
601 				       "ctl_io!\n");
602 				/* Bad Juju */
603 				/* Need to set busy and send msg back */
604 				msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
605 				msg_info.hdr.status = CTL_SCSI_ERROR;
606 				msg_info.scsi.scsi_status = SCSI_STATUS_BUSY;
607 				msg_info.scsi.sense_len = 0;
608 			        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
609 				    sizeof(msg_info), 0) > CTL_HA_STATUS_SUCCESS){
610 				}
611 				goto bailout;
612 			}
613 			ctl_zero_io(io);
614 			// populate ctsio from msg_info
615 			io->io_hdr.io_type = CTL_IO_SCSI;
616 			io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
617 			io->io_hdr.original_sc = msg_info.hdr.original_sc;
618 #if 0
619 			printf("pOrig %x\n", (int)msg_info.original_sc);
620 #endif
621 			io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
622 					    CTL_FLAG_IO_ACTIVE;
623 			/*
624 			 * If we're in serialization-only mode, we don't
625 			 * want to go through full done processing.  Thus
626 			 * the COPY flag.
627 			 *
628 			 * XXX KDM add another flag that is more specific.
629 			 */
630 			if (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)
631 				io->io_hdr.flags |= CTL_FLAG_INT_COPY;
632 			io->io_hdr.nexus = msg_info.hdr.nexus;
633 #if 0
634 			printf("targ %d, port %d, iid %d, lun %d\n",
635 			       io->io_hdr.nexus.targ_target.id,
636 			       io->io_hdr.nexus.targ_port,
637 			       io->io_hdr.nexus.initid.id,
638 			       io->io_hdr.nexus.targ_lun);
639 #endif
640 			io->scsiio.tag_num = msg_info.scsi.tag_num;
641 			io->scsiio.tag_type = msg_info.scsi.tag_type;
642 			memcpy(io->scsiio.cdb, msg_info.scsi.cdb,
643 			       CTL_MAX_CDBLEN);
644 			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
645 				const struct ctl_cmd_entry *entry;
646 
647 				entry = ctl_get_cmd_entry(&io->scsiio, NULL);
648 				io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
649 				io->io_hdr.flags |=
650 					entry->flags & CTL_FLAG_DATA_MASK;
651 			}
652 			ctl_enqueue_isc(io);
653 			break;
654 
655 		/* Performed on the Originating SC, XFER mode only */
656 		case CTL_MSG_DATAMOVE: {
657 			struct ctl_sg_entry *sgl;
658 			int i, j;
659 
660 			io = msg_info.hdr.original_sc;
661 			if (io == NULL) {
662 				printf("%s: original_sc == NULL!\n", __func__);
663 				/* XXX KDM do something here */
664 				break;
665 			}
666 			io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
667 			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
668 			/*
669 			 * Keep track of this, we need to send it back over
670 			 * when the datamove is complete.
671 			 */
672 			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
673 
674 			if (msg_info.dt.sg_sequence == 0) {
675 				/*
676 				 * XXX KDM we use the preallocated S/G list
677 				 * here, but we'll need to change this to
678 				 * dynamic allocation if we need larger S/G
679 				 * lists.
680 				 */
681 				if (msg_info.dt.kern_sg_entries >
682 				    sizeof(io->io_hdr.remote_sglist) /
683 				    sizeof(io->io_hdr.remote_sglist[0])) {
684 					printf("%s: number of S/G entries "
685 					    "needed %u > allocated num %zd\n",
686 					    __func__,
687 					    msg_info.dt.kern_sg_entries,
688 					    sizeof(io->io_hdr.remote_sglist)/
689 					    sizeof(io->io_hdr.remote_sglist[0]));
690 
691 					/*
692 					 * XXX KDM send a message back to
693 					 * the other side to shut down the
694 					 * DMA.  The error will come back
695 					 * through via the normal channel.
696 					 */
697 					break;
698 				}
699 				sgl = io->io_hdr.remote_sglist;
700 				memset(sgl, 0,
701 				       sizeof(io->io_hdr.remote_sglist));
702 
703 				io->scsiio.kern_data_ptr = (uint8_t *)sgl;
704 
705 				io->scsiio.kern_sg_entries =
706 					msg_info.dt.kern_sg_entries;
707 				io->scsiio.rem_sg_entries =
708 					msg_info.dt.kern_sg_entries;
709 				io->scsiio.kern_data_len =
710 					msg_info.dt.kern_data_len;
711 				io->scsiio.kern_total_len =
712 					msg_info.dt.kern_total_len;
713 				io->scsiio.kern_data_resid =
714 					msg_info.dt.kern_data_resid;
715 				io->scsiio.kern_rel_offset =
716 					msg_info.dt.kern_rel_offset;
717 				/*
718 				 * Clear out per-DMA flags.
719 				 */
720 				io->io_hdr.flags &= ~CTL_FLAG_RDMA_MASK;
721 				/*
722 				 * Add per-DMA flags that are set for this
723 				 * particular DMA request.
724 				 */
725 				io->io_hdr.flags |= msg_info.dt.flags &
726 						    CTL_FLAG_RDMA_MASK;
727 			} else
728 				sgl = (struct ctl_sg_entry *)
729 					io->scsiio.kern_data_ptr;
730 
731 			for (i = msg_info.dt.sent_sg_entries, j = 0;
732 			     i < (msg_info.dt.sent_sg_entries +
733 			     msg_info.dt.cur_sg_entries); i++, j++) {
734 				sgl[i].addr = msg_info.dt.sg_list[j].addr;
735 				sgl[i].len = msg_info.dt.sg_list[j].len;
736 
737 #if 0
738 				printf("%s: L: %p,%d -> %p,%d j=%d, i=%d\n",
739 				       __func__,
740 				       msg_info.dt.sg_list[j].addr,
741 				       msg_info.dt.sg_list[j].len,
742 				       sgl[i].addr, sgl[i].len, j, i);
743 #endif
744 			}
745 #if 0
746 			memcpy(&sgl[msg_info.dt.sent_sg_entries],
747 			       msg_info.dt.sg_list,
748 			       sizeof(*sgl) * msg_info.dt.cur_sg_entries);
749 #endif
750 
751 			/*
752 			 * If this is the last piece of the I/O, we've got
753 			 * the full S/G list.  Queue processing in the thread.
754 			 * Otherwise wait for the next piece.
755 			 */
756 			if (msg_info.dt.sg_last != 0)
757 				ctl_enqueue_isc(io);
758 			break;
759 		}
760 		/* Performed on the Serializing (primary) SC, XFER mode only */
761 		case CTL_MSG_DATAMOVE_DONE: {
762 			if (msg_info.hdr.serializing_sc == NULL) {
763 				printf("%s: serializing_sc == NULL!\n",
764 				       __func__);
765 				/* XXX KDM now what? */
766 				break;
767 			}
768 			/*
769 			 * We grab the sense information here in case
770 			 * there was a failure, so we can return status
771 			 * back to the initiator.
772 			 */
773 			io = msg_info.hdr.serializing_sc;
774 			io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
775 			io->io_hdr.status = msg_info.hdr.status;
776 			io->scsiio.scsi_status = msg_info.scsi.scsi_status;
777 			io->scsiio.sense_len = msg_info.scsi.sense_len;
778 			io->scsiio.sense_residual =msg_info.scsi.sense_residual;
779 			io->io_hdr.port_status = msg_info.scsi.fetd_status;
780 			io->scsiio.residual = msg_info.scsi.residual;
781 			memcpy(&io->scsiio.sense_data,&msg_info.scsi.sense_data,
782 			       sizeof(io->scsiio.sense_data));
783 			ctl_enqueue_isc(io);
784 			break;
785 		}
786 
787 		/* Preformed on Originating SC, SER_ONLY mode */
788 		case CTL_MSG_R2R:
789 			io = msg_info.hdr.original_sc;
790 			if (io == NULL) {
791 				printf("%s: Major Bummer\n", __func__);
792 				return;
793 			} else {
794 #if 0
795 				printf("pOrig %x\n",(int) ctsio);
796 #endif
797 			}
798 			io->io_hdr.msg_type = CTL_MSG_R2R;
799 			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
800 			ctl_enqueue_isc(io);
801 			break;
802 
803 		/*
804 		 * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
805 		 * mode.
806 		 * Performed on the Originating (i.e. secondary) SC in XFER
807 		 * mode
808 		 */
809 		case CTL_MSG_FINISH_IO:
810 			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER)
811 				ctl_isc_handler_finish_xfer(ctl_softc,
812 							    &msg_info);
813 			else
814 				ctl_isc_handler_finish_ser_only(ctl_softc,
815 								&msg_info);
816 			break;
817 
818 		/* Preformed on Originating SC */
819 		case CTL_MSG_BAD_JUJU:
820 			io = msg_info.hdr.original_sc;
821 			if (io == NULL) {
822 				printf("%s: Bad JUJU!, original_sc is NULL!\n",
823 				       __func__);
824 				break;
825 			}
826 			ctl_copy_sense_data(&msg_info, io);
827 			/*
828 			 * IO should have already been cleaned up on other
829 			 * SC so clear this flag so we won't send a message
830 			 * back to finish the IO there.
831 			 */
832 			io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
833 			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
834 
835 			/* io = msg_info.hdr.serializing_sc; */
836 			io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
837 			ctl_enqueue_isc(io);
838 			break;
839 
840 		/* Handle resets sent from the other side */
841 		case CTL_MSG_MANAGE_TASKS: {
842 			struct ctl_taskio *taskio;
843 			taskio = (struct ctl_taskio *)ctl_alloc_io(
844 				(void *)ctl_softc->othersc_pool);
845 			if (taskio == NULL) {
846 				printf("ctl_isc_event_handler: can't allocate "
847 				       "ctl_io!\n");
848 				/* Bad Juju */
849 				/* should I just call the proper reset func
850 				   here??? */
851 				goto bailout;
852 			}
853 			ctl_zero_io((union ctl_io *)taskio);
854 			taskio->io_hdr.io_type = CTL_IO_TASK;
855 			taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
856 			taskio->io_hdr.nexus = msg_info.hdr.nexus;
857 			taskio->task_action = msg_info.task.task_action;
858 			taskio->tag_num = msg_info.task.tag_num;
859 			taskio->tag_type = msg_info.task.tag_type;
860 #ifdef CTL_TIME_IO
861 			taskio->io_hdr.start_time = time_uptime;
862 			getbintime(&taskio->io_hdr.start_bt);
863 #if 0
864 			cs_prof_gettime(&taskio->io_hdr.start_ticks);
865 #endif
866 #endif /* CTL_TIME_IO */
867 			ctl_run_task((union ctl_io *)taskio);
868 			break;
869 		}
870 		/* Persistent Reserve action which needs attention */
871 		case CTL_MSG_PERS_ACTION:
872 			presio = (struct ctl_prio *)ctl_alloc_io(
873 				(void *)ctl_softc->othersc_pool);
874 			if (presio == NULL) {
875 				printf("ctl_isc_event_handler: can't allocate "
876 				       "ctl_io!\n");
877 				/* Bad Juju */
878 				/* Need to set busy and send msg back */
879 				goto bailout;
880 			}
881 			ctl_zero_io((union ctl_io *)presio);
882 			presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
883 			presio->pr_msg = msg_info.pr;
884 			ctl_enqueue_isc((union ctl_io *)presio);
885 			break;
886 		case CTL_MSG_SYNC_FE:
887 			rcv_sync_msg = 1;
888 			break;
889 		case CTL_MSG_APS_LOCK: {
890 			// It's quicker to execute this then to
891 			// queue it.
892 			struct ctl_lun *lun;
893 			struct ctl_page_index *page_index;
894 			struct copan_aps_subpage *current_sp;
895 			uint32_t targ_lun;
896 
897 			targ_lun = msg_info.hdr.nexus.targ_mapped_lun;
898 			lun = ctl_softc->ctl_luns[targ_lun];
899 			mtx_lock(&lun->lun_lock);
900 			page_index = &lun->mode_pages.index[index_to_aps_page];
901 			current_sp = (struct copan_aps_subpage *)
902 				     (page_index->page_data +
903 				     (page_index->page_len * CTL_PAGE_CURRENT));
904 
905 			current_sp->lock_active = msg_info.aps.lock_flag;
906 			mtx_unlock(&lun->lun_lock);
907 		        break;
908 		}
909 		default:
910 		        printf("How did I get here?\n");
911 		}
912 	} else if (event == CTL_HA_EVT_MSG_SENT) {
913 		if (param != CTL_HA_STATUS_SUCCESS) {
914 			printf("Bad status from ctl_ha_msg_send status %d\n",
915 			       param);
916 		}
917 		return;
918 	} else if (event == CTL_HA_EVT_DISCONNECT) {
919 		printf("CTL: Got a disconnect from Isc\n");
920 		return;
921 	} else {
922 		printf("ctl_isc_event_handler: Unknown event %d\n", event);
923 		return;
924 	}
925 
926 bailout:
927 	return;
928 }
929 
930 static void
931 ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
932 {
933 	struct scsi_sense_data *sense;
934 
935 	sense = &dest->scsiio.sense_data;
936 	bcopy(&src->scsi.sense_data, sense, sizeof(*sense));
937 	dest->scsiio.scsi_status = src->scsi.scsi_status;
938 	dest->scsiio.sense_len = src->scsi.sense_len;
939 	dest->io_hdr.status = src->hdr.status;
940 }
941 
942 static int
943 ctl_init(void)
944 {
945 	struct ctl_softc *softc;
946 	struct ctl_io_pool *internal_pool, *emergency_pool, *other_pool;
947 	struct ctl_port *port;
948         uint8_t sc_id =0;
949 	int i, error, retval;
950 	//int isc_retval;
951 
952 	retval = 0;
953 	ctl_pause_rtr = 0;
954         rcv_sync_msg = 0;
955 
956 	control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
957 			       M_WAITOK | M_ZERO);
958 	softc = control_softc;
959 
960 	softc->dev = make_dev(&ctl_cdevsw, 0, UID_ROOT, GID_OPERATOR, 0600,
961 			      "cam/ctl");
962 
963 	softc->dev->si_drv1 = softc;
964 
965 	/*
966 	 * By default, return a "bad LUN" peripheral qualifier for unknown
967 	 * LUNs.  The user can override this default using the tunable or
968 	 * sysctl.  See the comment in ctl_inquiry_std() for more details.
969 	 */
970 	softc->inquiry_pq_no_lun = 1;
971 	TUNABLE_INT_FETCH("kern.cam.ctl.inquiry_pq_no_lun",
972 			  &softc->inquiry_pq_no_lun);
973 	sysctl_ctx_init(&softc->sysctl_ctx);
974 	softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
975 		SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
976 		CTLFLAG_RD, 0, "CAM Target Layer");
977 
978 	if (softc->sysctl_tree == NULL) {
979 		printf("%s: unable to allocate sysctl tree\n", __func__);
980 		destroy_dev(softc->dev);
981 		free(control_softc, M_DEVBUF);
982 		control_softc = NULL;
983 		return (ENOMEM);
984 	}
985 
986 	SYSCTL_ADD_INT(&softc->sysctl_ctx,
987 		       SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO,
988 		       "inquiry_pq_no_lun", CTLFLAG_RW,
989 		       &softc->inquiry_pq_no_lun, 0,
990 		       "Report no lun possible for invalid LUNs");
991 
992 	mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
993 	mtx_init(&softc->pool_lock, "CTL pool mutex", NULL, MTX_DEF);
994 	softc->open_count = 0;
995 
996 	/*
997 	 * Default to actually sending a SYNCHRONIZE CACHE command down to
998 	 * the drive.
999 	 */
1000 	softc->flags = CTL_FLAG_REAL_SYNC;
1001 
1002 	/*
1003 	 * In Copan's HA scheme, the "master" and "slave" roles are
1004 	 * figured out through the slot the controller is in.  Although it
1005 	 * is an active/active system, someone has to be in charge.
1006  	 */
1007 #ifdef NEEDTOPORT
1008         scmicro_rw(SCMICRO_GET_SHELF_ID, &sc_id);
1009 #endif
1010 
1011         if (sc_id == 0) {
1012 		softc->flags |= CTL_FLAG_MASTER_SHELF;
1013 		persis_offset = 0;
1014 	} else
1015 		persis_offset = CTL_MAX_INITIATORS;
1016 
1017 	/*
1018 	 * XXX KDM need to figure out where we want to get our target ID
1019 	 * and WWID.  Is it different on each port?
1020 	 */
1021 	softc->target.id = 0;
1022 	softc->target.wwid[0] = 0x12345678;
1023 	softc->target.wwid[1] = 0x87654321;
1024 	STAILQ_INIT(&softc->lun_list);
1025 	STAILQ_INIT(&softc->pending_lun_queue);
1026 	STAILQ_INIT(&softc->fe_list);
1027 	STAILQ_INIT(&softc->port_list);
1028 	STAILQ_INIT(&softc->be_list);
1029 	STAILQ_INIT(&softc->io_pools);
1030 	ctl_tpc_init(softc);
1031 
1032 	if (ctl_pool_create(softc, CTL_POOL_INTERNAL, CTL_POOL_ENTRIES_INTERNAL,
1033 			    &internal_pool)!= 0){
1034 		printf("ctl: can't allocate %d entry internal pool, "
1035 		       "exiting\n", CTL_POOL_ENTRIES_INTERNAL);
1036 		return (ENOMEM);
1037 	}
1038 
1039 	if (ctl_pool_create(softc, CTL_POOL_EMERGENCY,
1040 			    CTL_POOL_ENTRIES_EMERGENCY, &emergency_pool) != 0) {
1041 		printf("ctl: can't allocate %d entry emergency pool, "
1042 		       "exiting\n", CTL_POOL_ENTRIES_EMERGENCY);
1043 		ctl_pool_free(internal_pool);
1044 		return (ENOMEM);
1045 	}
1046 
1047 	if (ctl_pool_create(softc, CTL_POOL_4OTHERSC, CTL_POOL_ENTRIES_OTHER_SC,
1048 	                    &other_pool) != 0)
1049 	{
1050 		printf("ctl: can't allocate %d entry other SC pool, "
1051 		       "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1052 		ctl_pool_free(internal_pool);
1053 		ctl_pool_free(emergency_pool);
1054 		return (ENOMEM);
1055 	}
1056 
1057 	softc->internal_pool = internal_pool;
1058 	softc->emergency_pool = emergency_pool;
1059 	softc->othersc_pool = other_pool;
1060 
1061 	if (worker_threads <= 0)
1062 		worker_threads = max(1, mp_ncpus / 4);
1063 	if (worker_threads > CTL_MAX_THREADS)
1064 		worker_threads = CTL_MAX_THREADS;
1065 
1066 	for (i = 0; i < worker_threads; i++) {
1067 		struct ctl_thread *thr = &softc->threads[i];
1068 
1069 		mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1070 		thr->ctl_softc = softc;
1071 		STAILQ_INIT(&thr->incoming_queue);
1072 		STAILQ_INIT(&thr->rtr_queue);
1073 		STAILQ_INIT(&thr->done_queue);
1074 		STAILQ_INIT(&thr->isc_queue);
1075 
1076 		error = kproc_kthread_add(ctl_work_thread, thr,
1077 		    &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1078 		if (error != 0) {
1079 			printf("error creating CTL work thread!\n");
1080 			ctl_pool_free(internal_pool);
1081 			ctl_pool_free(emergency_pool);
1082 			ctl_pool_free(other_pool);
1083 			return (error);
1084 		}
1085 	}
1086 	error = kproc_kthread_add(ctl_lun_thread, softc,
1087 	    &softc->ctl_proc, NULL, 0, 0, "ctl", "lun");
1088 	if (error != 0) {
1089 		printf("error creating CTL lun thread!\n");
1090 		ctl_pool_free(internal_pool);
1091 		ctl_pool_free(emergency_pool);
1092 		ctl_pool_free(other_pool);
1093 		return (error);
1094 	}
1095 	if (bootverbose)
1096 		printf("ctl: CAM Target Layer loaded\n");
1097 
1098 	/*
1099 	 * Initialize the ioctl front end.
1100 	 */
1101 	ctl_frontend_register(&ioctl_frontend);
1102 	port = &softc->ioctl_info.port;
1103 	port->frontend = &ioctl_frontend;
1104 	sprintf(softc->ioctl_info.port_name, "ioctl");
1105 	port->port_type = CTL_PORT_IOCTL;
1106 	port->num_requested_ctl_io = 100;
1107 	port->port_name = softc->ioctl_info.port_name;
1108 	port->port_online = ctl_ioctl_online;
1109 	port->port_offline = ctl_ioctl_offline;
1110 	port->onoff_arg = &softc->ioctl_info;
1111 	port->lun_enable = ctl_ioctl_lun_enable;
1112 	port->lun_disable = ctl_ioctl_lun_disable;
1113 	port->targ_lun_arg = &softc->ioctl_info;
1114 	port->fe_datamove = ctl_ioctl_datamove;
1115 	port->fe_done = ctl_ioctl_done;
1116 	port->max_targets = 15;
1117 	port->max_target_id = 15;
1118 
1119 	if (ctl_port_register(&softc->ioctl_info.port,
1120 	                  (softc->flags & CTL_FLAG_MASTER_SHELF)) != 0) {
1121 		printf("ctl: ioctl front end registration failed, will "
1122 		       "continue anyway\n");
1123 	}
1124 
1125 #ifdef CTL_IO_DELAY
1126 	if (sizeof(struct callout) > CTL_TIMER_BYTES) {
1127 		printf("sizeof(struct callout) %zd > CTL_TIMER_BYTES %zd\n",
1128 		       sizeof(struct callout), CTL_TIMER_BYTES);
1129 		return (EINVAL);
1130 	}
1131 #endif /* CTL_IO_DELAY */
1132 
1133 	return (0);
1134 }
1135 
1136 void
1137 ctl_shutdown(void)
1138 {
1139 	struct ctl_softc *softc;
1140 	struct ctl_lun *lun, *next_lun;
1141 	struct ctl_io_pool *pool;
1142 
1143 	softc = (struct ctl_softc *)control_softc;
1144 
1145 	if (ctl_port_deregister(&softc->ioctl_info.port) != 0)
1146 		printf("ctl: ioctl front end deregistration failed\n");
1147 
1148 	mtx_lock(&softc->ctl_lock);
1149 
1150 	/*
1151 	 * Free up each LUN.
1152 	 */
1153 	for (lun = STAILQ_FIRST(&softc->lun_list); lun != NULL; lun = next_lun){
1154 		next_lun = STAILQ_NEXT(lun, links);
1155 		ctl_free_lun(lun);
1156 	}
1157 
1158 	mtx_unlock(&softc->ctl_lock);
1159 
1160 	ctl_frontend_deregister(&ioctl_frontend);
1161 
1162 	/*
1163 	 * This will rip the rug out from under any FETDs or anyone else
1164 	 * that has a pool allocated.  Since we increment our module
1165 	 * refcount any time someone outside the main CTL module allocates
1166 	 * a pool, we shouldn't have any problems here.  The user won't be
1167 	 * able to unload the CTL module until client modules have
1168 	 * successfully unloaded.
1169 	 */
1170 	while ((pool = STAILQ_FIRST(&softc->io_pools)) != NULL)
1171 		ctl_pool_free(pool);
1172 
1173 #if 0
1174 	ctl_shutdown_thread(softc->work_thread);
1175 	mtx_destroy(&softc->queue_lock);
1176 #endif
1177 
1178 	ctl_tpc_shutdown(softc);
1179 	mtx_destroy(&softc->pool_lock);
1180 	mtx_destroy(&softc->ctl_lock);
1181 
1182 	destroy_dev(softc->dev);
1183 
1184 	sysctl_ctx_free(&softc->sysctl_ctx);
1185 
1186 	free(control_softc, M_DEVBUF);
1187 	control_softc = NULL;
1188 
1189 	if (bootverbose)
1190 		printf("ctl: CAM Target Layer unloaded\n");
1191 }
1192 
1193 static int
1194 ctl_module_event_handler(module_t mod, int what, void *arg)
1195 {
1196 
1197 	switch (what) {
1198 	case MOD_LOAD:
1199 		return (ctl_init());
1200 	case MOD_UNLOAD:
1201 		return (EBUSY);
1202 	default:
1203 		return (EOPNOTSUPP);
1204 	}
1205 }
1206 
1207 /*
1208  * XXX KDM should we do some access checks here?  Bump a reference count to
1209  * prevent a CTL module from being unloaded while someone has it open?
1210  */
1211 static int
1212 ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1213 {
1214 	return (0);
1215 }
1216 
1217 static int
1218 ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1219 {
1220 	return (0);
1221 }
1222 
1223 int
1224 ctl_port_enable(ctl_port_type port_type)
1225 {
1226 	struct ctl_softc *softc;
1227 	struct ctl_port *port;
1228 
1229 	if (ctl_is_single == 0) {
1230 		union ctl_ha_msg msg_info;
1231 		int isc_retval;
1232 
1233 #if 0
1234 		printf("%s: HA mode, synchronizing frontend enable\n",
1235 		        __func__);
1236 #endif
1237 		msg_info.hdr.msg_type = CTL_MSG_SYNC_FE;
1238 	        if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1239 		        sizeof(msg_info), 1 )) > CTL_HA_STATUS_SUCCESS) {
1240 			printf("Sync msg send error retval %d\n", isc_retval);
1241 		}
1242 		if (!rcv_sync_msg) {
1243 			isc_retval=ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
1244 			        sizeof(msg_info), 1);
1245 		}
1246 #if 0
1247         	printf("CTL:Frontend Enable\n");
1248 	} else {
1249 		printf("%s: single mode, skipping frontend synchronization\n",
1250 		        __func__);
1251 #endif
1252 	}
1253 
1254 	softc = control_softc;
1255 
1256 	STAILQ_FOREACH(port, &softc->port_list, links) {
1257 		if (port_type & port->port_type)
1258 		{
1259 #if 0
1260 			printf("port %d\n", port->targ_port);
1261 #endif
1262 			ctl_port_online(port);
1263 		}
1264 	}
1265 
1266 	return (0);
1267 }
1268 
1269 int
1270 ctl_port_disable(ctl_port_type port_type)
1271 {
1272 	struct ctl_softc *softc;
1273 	struct ctl_port *port;
1274 
1275 	softc = control_softc;
1276 
1277 	STAILQ_FOREACH(port, &softc->port_list, links) {
1278 		if (port_type & port->port_type)
1279 			ctl_port_offline(port);
1280 	}
1281 
1282 	return (0);
1283 }
1284 
1285 /*
1286  * Returns 0 for success, 1 for failure.
1287  * Currently the only failure mode is if there aren't enough entries
1288  * allocated.  So, in case of a failure, look at num_entries_dropped,
1289  * reallocate and try again.
1290  */
1291 int
1292 ctl_port_list(struct ctl_port_entry *entries, int num_entries_alloced,
1293 	      int *num_entries_filled, int *num_entries_dropped,
1294 	      ctl_port_type port_type, int no_virtual)
1295 {
1296 	struct ctl_softc *softc;
1297 	struct ctl_port *port;
1298 	int entries_dropped, entries_filled;
1299 	int retval;
1300 	int i;
1301 
1302 	softc = control_softc;
1303 
1304 	retval = 0;
1305 	entries_filled = 0;
1306 	entries_dropped = 0;
1307 
1308 	i = 0;
1309 	mtx_lock(&softc->ctl_lock);
1310 	STAILQ_FOREACH(port, &softc->port_list, links) {
1311 		struct ctl_port_entry *entry;
1312 
1313 		if ((port->port_type & port_type) == 0)
1314 			continue;
1315 
1316 		if ((no_virtual != 0)
1317 		 && (port->virtual_port != 0))
1318 			continue;
1319 
1320 		if (entries_filled >= num_entries_alloced) {
1321 			entries_dropped++;
1322 			continue;
1323 		}
1324 		entry = &entries[i];
1325 
1326 		entry->port_type = port->port_type;
1327 		strlcpy(entry->port_name, port->port_name,
1328 			sizeof(entry->port_name));
1329 		entry->physical_port = port->physical_port;
1330 		entry->virtual_port = port->virtual_port;
1331 		entry->wwnn = port->wwnn;
1332 		entry->wwpn = port->wwpn;
1333 
1334 		i++;
1335 		entries_filled++;
1336 	}
1337 
1338 	mtx_unlock(&softc->ctl_lock);
1339 
1340 	if (entries_dropped > 0)
1341 		retval = 1;
1342 
1343 	*num_entries_dropped = entries_dropped;
1344 	*num_entries_filled = entries_filled;
1345 
1346 	return (retval);
1347 }
1348 
1349 static void
1350 ctl_ioctl_online(void *arg)
1351 {
1352 	struct ctl_ioctl_info *ioctl_info;
1353 
1354 	ioctl_info = (struct ctl_ioctl_info *)arg;
1355 
1356 	ioctl_info->flags |= CTL_IOCTL_FLAG_ENABLED;
1357 }
1358 
1359 static void
1360 ctl_ioctl_offline(void *arg)
1361 {
1362 	struct ctl_ioctl_info *ioctl_info;
1363 
1364 	ioctl_info = (struct ctl_ioctl_info *)arg;
1365 
1366 	ioctl_info->flags &= ~CTL_IOCTL_FLAG_ENABLED;
1367 }
1368 
1369 /*
1370  * Remove an initiator by port number and initiator ID.
1371  * Returns 0 for success, -1 for failure.
1372  */
1373 int
1374 ctl_remove_initiator(struct ctl_port *port, int iid)
1375 {
1376 	struct ctl_softc *softc = control_softc;
1377 
1378 	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1379 
1380 	if (iid > CTL_MAX_INIT_PER_PORT) {
1381 		printf("%s: initiator ID %u > maximun %u!\n",
1382 		       __func__, iid, CTL_MAX_INIT_PER_PORT);
1383 		return (-1);
1384 	}
1385 
1386 	mtx_lock(&softc->ctl_lock);
1387 	port->wwpn_iid[iid].in_use--;
1388 	port->wwpn_iid[iid].last_use = time_uptime;
1389 	mtx_unlock(&softc->ctl_lock);
1390 
1391 	return (0);
1392 }
1393 
1394 /*
1395  * Add an initiator to the initiator map.
1396  * Returns iid for success, < 0 for failure.
1397  */
1398 int
1399 ctl_add_initiator(struct ctl_port *port, int iid, uint64_t wwpn, char *name)
1400 {
1401 	struct ctl_softc *softc = control_softc;
1402 	time_t best_time;
1403 	int i, best;
1404 
1405 	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1406 
1407 	if (iid >= CTL_MAX_INIT_PER_PORT) {
1408 		printf("%s: WWPN %#jx initiator ID %u > maximum %u!\n",
1409 		       __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
1410 		free(name, M_CTL);
1411 		return (-1);
1412 	}
1413 
1414 	mtx_lock(&softc->ctl_lock);
1415 
1416 	if (iid < 0 && (wwpn != 0 || name != NULL)) {
1417 		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1418 			if (wwpn != 0 && wwpn == port->wwpn_iid[i].wwpn) {
1419 				iid = i;
1420 				break;
1421 			}
1422 			if (name != NULL && port->wwpn_iid[i].name != NULL &&
1423 			    strcmp(name, port->wwpn_iid[i].name) == 0) {
1424 				iid = i;
1425 				break;
1426 			}
1427 		}
1428 	}
1429 
1430 	if (iid < 0) {
1431 		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1432 			if (port->wwpn_iid[i].in_use == 0 &&
1433 			    port->wwpn_iid[i].wwpn == 0 &&
1434 			    port->wwpn_iid[i].name == NULL) {
1435 				iid = i;
1436 				break;
1437 			}
1438 		}
1439 	}
1440 
1441 	if (iid < 0) {
1442 		best = -1;
1443 		best_time = INT32_MAX;
1444 		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1445 			if (port->wwpn_iid[i].in_use == 0) {
1446 				if (port->wwpn_iid[i].last_use < best_time) {
1447 					best = i;
1448 					best_time = port->wwpn_iid[i].last_use;
1449 				}
1450 			}
1451 		}
1452 		iid = best;
1453 	}
1454 
1455 	if (iid < 0) {
1456 		mtx_unlock(&softc->ctl_lock);
1457 		free(name, M_CTL);
1458 		return (-2);
1459 	}
1460 
1461 	if (port->wwpn_iid[iid].in_use > 0 && (wwpn != 0 || name != NULL)) {
1462 		/*
1463 		 * This is not an error yet.
1464 		 */
1465 		if (wwpn != 0 && wwpn == port->wwpn_iid[iid].wwpn) {
1466 #if 0
1467 			printf("%s: port %d iid %u WWPN %#jx arrived"
1468 			    " again\n", __func__, port->targ_port,
1469 			    iid, (uintmax_t)wwpn);
1470 #endif
1471 			goto take;
1472 		}
1473 		if (name != NULL && port->wwpn_iid[iid].name != NULL &&
1474 		    strcmp(name, port->wwpn_iid[iid].name) == 0) {
1475 #if 0
1476 			printf("%s: port %d iid %u name '%s' arrived"
1477 			    " again\n", __func__, port->targ_port,
1478 			    iid, name);
1479 #endif
1480 			goto take;
1481 		}
1482 
1483 		/*
1484 		 * This is an error, but what do we do about it?  The
1485 		 * driver is telling us we have a new WWPN for this
1486 		 * initiator ID, so we pretty much need to use it.
1487 		 */
1488 		printf("%s: port %d iid %u WWPN %#jx '%s' arrived,"
1489 		    " but WWPN %#jx '%s' is still at that address\n",
1490 		    __func__, port->targ_port, iid, wwpn, name,
1491 		    (uintmax_t)port->wwpn_iid[iid].wwpn,
1492 		    port->wwpn_iid[iid].name);
1493 
1494 		/*
1495 		 * XXX KDM clear have_ca and ua_pending on each LUN for
1496 		 * this initiator.
1497 		 */
1498 	}
1499 take:
1500 	free(port->wwpn_iid[iid].name, M_CTL);
1501 	port->wwpn_iid[iid].name = name;
1502 	port->wwpn_iid[iid].wwpn = wwpn;
1503 	port->wwpn_iid[iid].in_use++;
1504 	mtx_unlock(&softc->ctl_lock);
1505 
1506 	return (iid);
1507 }
1508 
1509 static int
1510 ctl_create_iid(struct ctl_port *port, int iid, uint8_t *buf)
1511 {
1512 	int len;
1513 
1514 	switch (port->port_type) {
1515 	case CTL_PORT_FC:
1516 	{
1517 		struct scsi_transportid_fcp *id =
1518 		    (struct scsi_transportid_fcp *)buf;
1519 		if (port->wwpn_iid[iid].wwpn == 0)
1520 			return (0);
1521 		memset(id, 0, sizeof(*id));
1522 		id->format_protocol = SCSI_PROTO_FC;
1523 		scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->n_port_name);
1524 		return (sizeof(*id));
1525 	}
1526 	case CTL_PORT_ISCSI:
1527 	{
1528 		struct scsi_transportid_iscsi_port *id =
1529 		    (struct scsi_transportid_iscsi_port *)buf;
1530 		if (port->wwpn_iid[iid].name == NULL)
1531 			return (0);
1532 		memset(id, 0, 256);
1533 		id->format_protocol = SCSI_TRN_ISCSI_FORMAT_PORT |
1534 		    SCSI_PROTO_ISCSI;
1535 		len = strlcpy(id->iscsi_name, port->wwpn_iid[iid].name, 252) + 1;
1536 		len = roundup2(min(len, 252), 4);
1537 		scsi_ulto2b(len, id->additional_length);
1538 		return (sizeof(*id) + len);
1539 	}
1540 	case CTL_PORT_SAS:
1541 	{
1542 		struct scsi_transportid_sas *id =
1543 		    (struct scsi_transportid_sas *)buf;
1544 		if (port->wwpn_iid[iid].wwpn == 0)
1545 			return (0);
1546 		memset(id, 0, sizeof(*id));
1547 		id->format_protocol = SCSI_PROTO_SAS;
1548 		scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->sas_address);
1549 		return (sizeof(*id));
1550 	}
1551 	default:
1552 	{
1553 		struct scsi_transportid_spi *id =
1554 		    (struct scsi_transportid_spi *)buf;
1555 		memset(id, 0, sizeof(*id));
1556 		id->format_protocol = SCSI_PROTO_SPI;
1557 		scsi_ulto2b(iid, id->scsi_addr);
1558 		scsi_ulto2b(port->targ_port, id->rel_trgt_port_id);
1559 		return (sizeof(*id));
1560 	}
1561 	}
1562 }
1563 
1564 static int
1565 ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id)
1566 {
1567 	return (0);
1568 }
1569 
1570 static int
1571 ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id)
1572 {
1573 	return (0);
1574 }
1575 
1576 /*
1577  * Data movement routine for the CTL ioctl frontend port.
1578  */
1579 static int
1580 ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio)
1581 {
1582 	struct ctl_sg_entry *ext_sglist, *kern_sglist;
1583 	struct ctl_sg_entry ext_entry, kern_entry;
1584 	int ext_sglen, ext_sg_entries, kern_sg_entries;
1585 	int ext_sg_start, ext_offset;
1586 	int len_to_copy, len_copied;
1587 	int kern_watermark, ext_watermark;
1588 	int ext_sglist_malloced;
1589 	int i, j;
1590 
1591 	ext_sglist_malloced = 0;
1592 	ext_sg_start = 0;
1593 	ext_offset = 0;
1594 
1595 	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove\n"));
1596 
1597 	/*
1598 	 * If this flag is set, fake the data transfer.
1599 	 */
1600 	if (ctsio->io_hdr.flags & CTL_FLAG_NO_DATAMOVE) {
1601 		ctsio->ext_data_filled = ctsio->ext_data_len;
1602 		goto bailout;
1603 	}
1604 
1605 	/*
1606 	 * To simplify things here, if we have a single buffer, stick it in
1607 	 * a S/G entry and just make it a single entry S/G list.
1608 	 */
1609 	if (ctsio->io_hdr.flags & CTL_FLAG_EDPTR_SGLIST) {
1610 		int len_seen;
1611 
1612 		ext_sglen = ctsio->ext_sg_entries * sizeof(*ext_sglist);
1613 
1614 		ext_sglist = (struct ctl_sg_entry *)malloc(ext_sglen, M_CTL,
1615 							   M_WAITOK);
1616 		ext_sglist_malloced = 1;
1617 		if (copyin(ctsio->ext_data_ptr, ext_sglist,
1618 				   ext_sglen) != 0) {
1619 			ctl_set_internal_failure(ctsio,
1620 						 /*sks_valid*/ 0,
1621 						 /*retry_count*/ 0);
1622 			goto bailout;
1623 		}
1624 		ext_sg_entries = ctsio->ext_sg_entries;
1625 		len_seen = 0;
1626 		for (i = 0; i < ext_sg_entries; i++) {
1627 			if ((len_seen + ext_sglist[i].len) >=
1628 			     ctsio->ext_data_filled) {
1629 				ext_sg_start = i;
1630 				ext_offset = ctsio->ext_data_filled - len_seen;
1631 				break;
1632 			}
1633 			len_seen += ext_sglist[i].len;
1634 		}
1635 	} else {
1636 		ext_sglist = &ext_entry;
1637 		ext_sglist->addr = ctsio->ext_data_ptr;
1638 		ext_sglist->len = ctsio->ext_data_len;
1639 		ext_sg_entries = 1;
1640 		ext_sg_start = 0;
1641 		ext_offset = ctsio->ext_data_filled;
1642 	}
1643 
1644 	if (ctsio->kern_sg_entries > 0) {
1645 		kern_sglist = (struct ctl_sg_entry *)ctsio->kern_data_ptr;
1646 		kern_sg_entries = ctsio->kern_sg_entries;
1647 	} else {
1648 		kern_sglist = &kern_entry;
1649 		kern_sglist->addr = ctsio->kern_data_ptr;
1650 		kern_sglist->len = ctsio->kern_data_len;
1651 		kern_sg_entries = 1;
1652 	}
1653 
1654 
1655 	kern_watermark = 0;
1656 	ext_watermark = ext_offset;
1657 	len_copied = 0;
1658 	for (i = ext_sg_start, j = 0;
1659 	     i < ext_sg_entries && j < kern_sg_entries;) {
1660 		uint8_t *ext_ptr, *kern_ptr;
1661 
1662 		len_to_copy = ctl_min(ext_sglist[i].len - ext_watermark,
1663 				      kern_sglist[j].len - kern_watermark);
1664 
1665 		ext_ptr = (uint8_t *)ext_sglist[i].addr;
1666 		ext_ptr = ext_ptr + ext_watermark;
1667 		if (ctsio->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
1668 			/*
1669 			 * XXX KDM fix this!
1670 			 */
1671 			panic("need to implement bus address support");
1672 #if 0
1673 			kern_ptr = bus_to_virt(kern_sglist[j].addr);
1674 #endif
1675 		} else
1676 			kern_ptr = (uint8_t *)kern_sglist[j].addr;
1677 		kern_ptr = kern_ptr + kern_watermark;
1678 
1679 		kern_watermark += len_to_copy;
1680 		ext_watermark += len_to_copy;
1681 
1682 		if ((ctsio->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
1683 		     CTL_FLAG_DATA_IN) {
1684 			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1685 					 "bytes to user\n", len_to_copy));
1686 			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1687 					 "to %p\n", kern_ptr, ext_ptr));
1688 			if (copyout(kern_ptr, ext_ptr, len_to_copy) != 0) {
1689 				ctl_set_internal_failure(ctsio,
1690 							 /*sks_valid*/ 0,
1691 							 /*retry_count*/ 0);
1692 				goto bailout;
1693 			}
1694 		} else {
1695 			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1696 					 "bytes from user\n", len_to_copy));
1697 			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1698 					 "to %p\n", ext_ptr, kern_ptr));
1699 			if (copyin(ext_ptr, kern_ptr, len_to_copy)!= 0){
1700 				ctl_set_internal_failure(ctsio,
1701 							 /*sks_valid*/ 0,
1702 							 /*retry_count*/0);
1703 				goto bailout;
1704 			}
1705 		}
1706 
1707 		len_copied += len_to_copy;
1708 
1709 		if (ext_sglist[i].len == ext_watermark) {
1710 			i++;
1711 			ext_watermark = 0;
1712 		}
1713 
1714 		if (kern_sglist[j].len == kern_watermark) {
1715 			j++;
1716 			kern_watermark = 0;
1717 		}
1718 	}
1719 
1720 	ctsio->ext_data_filled += len_copied;
1721 
1722 	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_sg_entries: %d, "
1723 			 "kern_sg_entries: %d\n", ext_sg_entries,
1724 			 kern_sg_entries));
1725 	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_data_len = %d, "
1726 			 "kern_data_len = %d\n", ctsio->ext_data_len,
1727 			 ctsio->kern_data_len));
1728 
1729 
1730 	/* XXX KDM set residual?? */
1731 bailout:
1732 
1733 	if (ext_sglist_malloced != 0)
1734 		free(ext_sglist, M_CTL);
1735 
1736 	return (CTL_RETVAL_COMPLETE);
1737 }
1738 
1739 /*
1740  * Serialize a command that went down the "wrong" side, and so was sent to
1741  * this controller for execution.  The logic is a little different than the
1742  * standard case in ctl_scsiio_precheck().  Errors in this case need to get
1743  * sent back to the other side, but in the success case, we execute the
1744  * command on this side (XFER mode) or tell the other side to execute it
1745  * (SER_ONLY mode).
1746  */
1747 static int
1748 ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
1749 {
1750 	struct ctl_softc *ctl_softc;
1751 	union ctl_ha_msg msg_info;
1752 	struct ctl_lun *lun;
1753 	int retval = 0;
1754 	uint32_t targ_lun;
1755 
1756 	ctl_softc = control_softc;
1757 
1758 	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
1759 	lun = ctl_softc->ctl_luns[targ_lun];
1760 	if (lun==NULL)
1761 	{
1762 		/*
1763 		 * Why isn't LUN defined? The other side wouldn't
1764 		 * send a cmd if the LUN is undefined.
1765 		 */
1766 		printf("%s: Bad JUJU!, LUN is NULL!\n", __func__);
1767 
1768 		/* "Logical unit not supported" */
1769 		ctl_set_sense_data(&msg_info.scsi.sense_data,
1770 				   lun,
1771 				   /*sense_format*/SSD_TYPE_NONE,
1772 				   /*current_error*/ 1,
1773 				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1774 				   /*asc*/ 0x25,
1775 				   /*ascq*/ 0x00,
1776 				   SSD_ELEM_NONE);
1777 
1778 		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1779 		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1780 		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1781 		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1782 		msg_info.hdr.serializing_sc = NULL;
1783 		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1784 	        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1785 				sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1786 		}
1787 		return(1);
1788 
1789 	}
1790 
1791 	mtx_lock(&lun->lun_lock);
1792     	TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1793 
1794 	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
1795 		(union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
1796 		 ooa_links))) {
1797 	case CTL_ACTION_BLOCK:
1798 		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
1799 		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
1800 				  blocked_links);
1801 		break;
1802 	case CTL_ACTION_PASS:
1803 	case CTL_ACTION_SKIP:
1804 		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
1805 			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
1806 			ctl_enqueue_rtr((union ctl_io *)ctsio);
1807 		} else {
1808 
1809 			/* send msg back to other side */
1810 			msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1811 			msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
1812 			msg_info.hdr.msg_type = CTL_MSG_R2R;
1813 #if 0
1814 			printf("2. pOrig %x\n", (int)msg_info.hdr.original_sc);
1815 #endif
1816 		        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1817 			    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1818 			}
1819 		}
1820 		break;
1821 	case CTL_ACTION_OVERLAP:
1822 		/* OVERLAPPED COMMANDS ATTEMPTED */
1823 		ctl_set_sense_data(&msg_info.scsi.sense_data,
1824 				   lun,
1825 				   /*sense_format*/SSD_TYPE_NONE,
1826 				   /*current_error*/ 1,
1827 				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1828 				   /*asc*/ 0x4E,
1829 				   /*ascq*/ 0x00,
1830 				   SSD_ELEM_NONE);
1831 
1832 		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1833 		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1834 		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1835 		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1836 		msg_info.hdr.serializing_sc = NULL;
1837 		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1838 #if 0
1839 		printf("BAD JUJU:Major Bummer Overlap\n");
1840 #endif
1841 		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1842 		retval = 1;
1843 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1844 		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1845 		}
1846 		break;
1847 	case CTL_ACTION_OVERLAP_TAG:
1848 		/* TAGGED OVERLAPPED COMMANDS (NN = QUEUE TAG) */
1849 		ctl_set_sense_data(&msg_info.scsi.sense_data,
1850 				   lun,
1851 				   /*sense_format*/SSD_TYPE_NONE,
1852 				   /*current_error*/ 1,
1853 				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1854 				   /*asc*/ 0x4D,
1855 				   /*ascq*/ ctsio->tag_num & 0xff,
1856 				   SSD_ELEM_NONE);
1857 
1858 		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1859 		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1860 		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1861 		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1862 		msg_info.hdr.serializing_sc = NULL;
1863 		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1864 #if 0
1865 		printf("BAD JUJU:Major Bummer Overlap Tag\n");
1866 #endif
1867 		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1868 		retval = 1;
1869 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1870 		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1871 		}
1872 		break;
1873 	case CTL_ACTION_ERROR:
1874 	default:
1875 		/* "Internal target failure" */
1876 		ctl_set_sense_data(&msg_info.scsi.sense_data,
1877 				   lun,
1878 				   /*sense_format*/SSD_TYPE_NONE,
1879 				   /*current_error*/ 1,
1880 				   /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
1881 				   /*asc*/ 0x44,
1882 				   /*ascq*/ 0x00,
1883 				   SSD_ELEM_NONE);
1884 
1885 		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1886 		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1887 		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1888 		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1889 		msg_info.hdr.serializing_sc = NULL;
1890 		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1891 #if 0
1892 		printf("BAD JUJU:Major Bummer HW Error\n");
1893 #endif
1894 		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1895 		retval = 1;
1896 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1897 		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1898 		}
1899 		break;
1900 	}
1901 	mtx_unlock(&lun->lun_lock);
1902 	return (retval);
1903 }
1904 
1905 static int
1906 ctl_ioctl_submit_wait(union ctl_io *io)
1907 {
1908 	struct ctl_fe_ioctl_params params;
1909 	ctl_fe_ioctl_state last_state;
1910 	int done, retval;
1911 
1912 	retval = 0;
1913 
1914 	bzero(&params, sizeof(params));
1915 
1916 	mtx_init(&params.ioctl_mtx, "ctliocmtx", NULL, MTX_DEF);
1917 	cv_init(&params.sem, "ctlioccv");
1918 	params.state = CTL_IOCTL_INPROG;
1919 	last_state = params.state;
1920 
1921 	io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = &params;
1922 
1923 	CTL_DEBUG_PRINT(("ctl_ioctl_submit_wait\n"));
1924 
1925 	/* This shouldn't happen */
1926 	if ((retval = ctl_queue(io)) != CTL_RETVAL_COMPLETE)
1927 		return (retval);
1928 
1929 	done = 0;
1930 
1931 	do {
1932 		mtx_lock(&params.ioctl_mtx);
1933 		/*
1934 		 * Check the state here, and don't sleep if the state has
1935 		 * already changed (i.e. wakeup has already occured, but we
1936 		 * weren't waiting yet).
1937 		 */
1938 		if (params.state == last_state) {
1939 			/* XXX KDM cv_wait_sig instead? */
1940 			cv_wait(&params.sem, &params.ioctl_mtx);
1941 		}
1942 		last_state = params.state;
1943 
1944 		switch (params.state) {
1945 		case CTL_IOCTL_INPROG:
1946 			/* Why did we wake up? */
1947 			/* XXX KDM error here? */
1948 			mtx_unlock(&params.ioctl_mtx);
1949 			break;
1950 		case CTL_IOCTL_DATAMOVE:
1951 			CTL_DEBUG_PRINT(("got CTL_IOCTL_DATAMOVE\n"));
1952 
1953 			/*
1954 			 * change last_state back to INPROG to avoid
1955 			 * deadlock on subsequent data moves.
1956 			 */
1957 			params.state = last_state = CTL_IOCTL_INPROG;
1958 
1959 			mtx_unlock(&params.ioctl_mtx);
1960 			ctl_ioctl_do_datamove(&io->scsiio);
1961 			/*
1962 			 * Note that in some cases, most notably writes,
1963 			 * this will queue the I/O and call us back later.
1964 			 * In other cases, generally reads, this routine
1965 			 * will immediately call back and wake us up,
1966 			 * probably using our own context.
1967 			 */
1968 			io->scsiio.be_move_done(io);
1969 			break;
1970 		case CTL_IOCTL_DONE:
1971 			mtx_unlock(&params.ioctl_mtx);
1972 			CTL_DEBUG_PRINT(("got CTL_IOCTL_DONE\n"));
1973 			done = 1;
1974 			break;
1975 		default:
1976 			mtx_unlock(&params.ioctl_mtx);
1977 			/* XXX KDM error here? */
1978 			break;
1979 		}
1980 	} while (done == 0);
1981 
1982 	mtx_destroy(&params.ioctl_mtx);
1983 	cv_destroy(&params.sem);
1984 
1985 	return (CTL_RETVAL_COMPLETE);
1986 }
1987 
1988 static void
1989 ctl_ioctl_datamove(union ctl_io *io)
1990 {
1991 	struct ctl_fe_ioctl_params *params;
1992 
1993 	params = (struct ctl_fe_ioctl_params *)
1994 		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
1995 
1996 	mtx_lock(&params->ioctl_mtx);
1997 	params->state = CTL_IOCTL_DATAMOVE;
1998 	cv_broadcast(&params->sem);
1999 	mtx_unlock(&params->ioctl_mtx);
2000 }
2001 
2002 static void
2003 ctl_ioctl_done(union ctl_io *io)
2004 {
2005 	struct ctl_fe_ioctl_params *params;
2006 
2007 	params = (struct ctl_fe_ioctl_params *)
2008 		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
2009 
2010 	mtx_lock(&params->ioctl_mtx);
2011 	params->state = CTL_IOCTL_DONE;
2012 	cv_broadcast(&params->sem);
2013 	mtx_unlock(&params->ioctl_mtx);
2014 }
2015 
2016 static void
2017 ctl_ioctl_hard_startstop_callback(void *arg, struct cfi_metatask *metatask)
2018 {
2019 	struct ctl_fe_ioctl_startstop_info *sd_info;
2020 
2021 	sd_info = (struct ctl_fe_ioctl_startstop_info *)arg;
2022 
2023 	sd_info->hs_info.status = metatask->status;
2024 	sd_info->hs_info.total_luns = metatask->taskinfo.startstop.total_luns;
2025 	sd_info->hs_info.luns_complete =
2026 		metatask->taskinfo.startstop.luns_complete;
2027 	sd_info->hs_info.luns_failed = metatask->taskinfo.startstop.luns_failed;
2028 
2029 	cv_broadcast(&sd_info->sem);
2030 }
2031 
2032 static void
2033 ctl_ioctl_bbrread_callback(void *arg, struct cfi_metatask *metatask)
2034 {
2035 	struct ctl_fe_ioctl_bbrread_info *fe_bbr_info;
2036 
2037 	fe_bbr_info = (struct ctl_fe_ioctl_bbrread_info *)arg;
2038 
2039 	mtx_lock(fe_bbr_info->lock);
2040 	fe_bbr_info->bbr_info->status = metatask->status;
2041 	fe_bbr_info->bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2042 	fe_bbr_info->wakeup_done = 1;
2043 	mtx_unlock(fe_bbr_info->lock);
2044 
2045 	cv_broadcast(&fe_bbr_info->sem);
2046 }
2047 
2048 /*
2049  * Returns 0 for success, errno for failure.
2050  */
2051 static int
2052 ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
2053 		   struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
2054 {
2055 	union ctl_io *io;
2056 	int retval;
2057 
2058 	retval = 0;
2059 
2060 	mtx_lock(&lun->lun_lock);
2061 	for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
2062 	     (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2063 	     ooa_links)) {
2064 		struct ctl_ooa_entry *entry;
2065 
2066 		/*
2067 		 * If we've got more than we can fit, just count the
2068 		 * remaining entries.
2069 		 */
2070 		if (*cur_fill_num >= ooa_hdr->alloc_num)
2071 			continue;
2072 
2073 		entry = &kern_entries[*cur_fill_num];
2074 
2075 		entry->tag_num = io->scsiio.tag_num;
2076 		entry->lun_num = lun->lun;
2077 #ifdef CTL_TIME_IO
2078 		entry->start_bt = io->io_hdr.start_bt;
2079 #endif
2080 		bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2081 		entry->cdb_len = io->scsiio.cdb_len;
2082 		if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
2083 			entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2084 
2085 		if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2086 			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2087 
2088 		if (io->io_hdr.flags & CTL_FLAG_ABORT)
2089 			entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2090 
2091 		if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2092 			entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2093 
2094 		if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2095 			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2096 	}
2097 	mtx_unlock(&lun->lun_lock);
2098 
2099 	return (retval);
2100 }
2101 
2102 static void *
2103 ctl_copyin_alloc(void *user_addr, int len, char *error_str,
2104 		 size_t error_str_len)
2105 {
2106 	void *kptr;
2107 
2108 	kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2109 
2110 	if (copyin(user_addr, kptr, len) != 0) {
2111 		snprintf(error_str, error_str_len, "Error copying %d bytes "
2112 			 "from user address %p to kernel address %p", len,
2113 			 user_addr, kptr);
2114 		free(kptr, M_CTL);
2115 		return (NULL);
2116 	}
2117 
2118 	return (kptr);
2119 }
2120 
2121 static void
2122 ctl_free_args(int num_args, struct ctl_be_arg *args)
2123 {
2124 	int i;
2125 
2126 	if (args == NULL)
2127 		return;
2128 
2129 	for (i = 0; i < num_args; i++) {
2130 		free(args[i].kname, M_CTL);
2131 		free(args[i].kvalue, M_CTL);
2132 	}
2133 
2134 	free(args, M_CTL);
2135 }
2136 
2137 static struct ctl_be_arg *
2138 ctl_copyin_args(int num_args, struct ctl_be_arg *uargs,
2139 		char *error_str, size_t error_str_len)
2140 {
2141 	struct ctl_be_arg *args;
2142 	int i;
2143 
2144 	args = ctl_copyin_alloc(uargs, num_args * sizeof(*args),
2145 				error_str, error_str_len);
2146 
2147 	if (args == NULL)
2148 		goto bailout;
2149 
2150 	for (i = 0; i < num_args; i++) {
2151 		args[i].kname = NULL;
2152 		args[i].kvalue = NULL;
2153 	}
2154 
2155 	for (i = 0; i < num_args; i++) {
2156 		uint8_t *tmpptr;
2157 
2158 		args[i].kname = ctl_copyin_alloc(args[i].name,
2159 			args[i].namelen, error_str, error_str_len);
2160 		if (args[i].kname == NULL)
2161 			goto bailout;
2162 
2163 		if (args[i].kname[args[i].namelen - 1] != '\0') {
2164 			snprintf(error_str, error_str_len, "Argument %d "
2165 				 "name is not NUL-terminated", i);
2166 			goto bailout;
2167 		}
2168 
2169 		if (args[i].flags & CTL_BEARG_RD) {
2170 			tmpptr = ctl_copyin_alloc(args[i].value,
2171 				args[i].vallen, error_str, error_str_len);
2172 			if (tmpptr == NULL)
2173 				goto bailout;
2174 			if ((args[i].flags & CTL_BEARG_ASCII)
2175 			 && (tmpptr[args[i].vallen - 1] != '\0')) {
2176 				snprintf(error_str, error_str_len, "Argument "
2177 				    "%d value is not NUL-terminated", i);
2178 				goto bailout;
2179 			}
2180 			args[i].kvalue = tmpptr;
2181 		} else {
2182 			args[i].kvalue = malloc(args[i].vallen,
2183 			    M_CTL, M_WAITOK | M_ZERO);
2184 		}
2185 	}
2186 
2187 	return (args);
2188 bailout:
2189 
2190 	ctl_free_args(num_args, args);
2191 
2192 	return (NULL);
2193 }
2194 
2195 static void
2196 ctl_copyout_args(int num_args, struct ctl_be_arg *args)
2197 {
2198 	int i;
2199 
2200 	for (i = 0; i < num_args; i++) {
2201 		if (args[i].flags & CTL_BEARG_WR)
2202 			copyout(args[i].kvalue, args[i].value, args[i].vallen);
2203 	}
2204 }
2205 
2206 /*
2207  * Escape characters that are illegal or not recommended in XML.
2208  */
2209 int
2210 ctl_sbuf_printf_esc(struct sbuf *sb, char *str)
2211 {
2212 	int retval;
2213 
2214 	retval = 0;
2215 
2216 	for (; *str; str++) {
2217 		switch (*str) {
2218 		case '&':
2219 			retval = sbuf_printf(sb, "&amp;");
2220 			break;
2221 		case '>':
2222 			retval = sbuf_printf(sb, "&gt;");
2223 			break;
2224 		case '<':
2225 			retval = sbuf_printf(sb, "&lt;");
2226 			break;
2227 		default:
2228 			retval = sbuf_putc(sb, *str);
2229 			break;
2230 		}
2231 
2232 		if (retval != 0)
2233 			break;
2234 
2235 	}
2236 
2237 	return (retval);
2238 }
2239 
2240 static void
2241 ctl_id_sbuf(struct ctl_devid *id, struct sbuf *sb)
2242 {
2243 	struct scsi_vpd_id_descriptor *desc;
2244 	int i;
2245 
2246 	if (id == NULL || id->len < 4)
2247 		return;
2248 	desc = (struct scsi_vpd_id_descriptor *)id->data;
2249 	switch (desc->id_type & SVPD_ID_TYPE_MASK) {
2250 	case SVPD_ID_TYPE_T10:
2251 		sbuf_printf(sb, "t10.");
2252 		break;
2253 	case SVPD_ID_TYPE_EUI64:
2254 		sbuf_printf(sb, "eui.");
2255 		break;
2256 	case SVPD_ID_TYPE_NAA:
2257 		sbuf_printf(sb, "naa.");
2258 		break;
2259 	case SVPD_ID_TYPE_SCSI_NAME:
2260 		break;
2261 	}
2262 	switch (desc->proto_codeset & SVPD_ID_CODESET_MASK) {
2263 	case SVPD_ID_CODESET_BINARY:
2264 		for (i = 0; i < desc->length; i++)
2265 			sbuf_printf(sb, "%02x", desc->identifier[i]);
2266 		break;
2267 	case SVPD_ID_CODESET_ASCII:
2268 		sbuf_printf(sb, "%.*s", (int)desc->length,
2269 		    (char *)desc->identifier);
2270 		break;
2271 	case SVPD_ID_CODESET_UTF8:
2272 		sbuf_printf(sb, "%s", (char *)desc->identifier);
2273 		break;
2274 	}
2275 }
2276 
2277 static int
2278 ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2279 	  struct thread *td)
2280 {
2281 	struct ctl_softc *softc;
2282 	int retval;
2283 
2284 	softc = control_softc;
2285 
2286 	retval = 0;
2287 
2288 	switch (cmd) {
2289 	case CTL_IO: {
2290 		union ctl_io *io;
2291 		void *pool_tmp;
2292 
2293 		/*
2294 		 * If we haven't been "enabled", don't allow any SCSI I/O
2295 		 * to this FETD.
2296 		 */
2297 		if ((softc->ioctl_info.flags & CTL_IOCTL_FLAG_ENABLED) == 0) {
2298 			retval = EPERM;
2299 			break;
2300 		}
2301 
2302 		io = ctl_alloc_io(softc->ioctl_info.port.ctl_pool_ref);
2303 		if (io == NULL) {
2304 			printf("ctl_ioctl: can't allocate ctl_io!\n");
2305 			retval = ENOSPC;
2306 			break;
2307 		}
2308 
2309 		/*
2310 		 * Need to save the pool reference so it doesn't get
2311 		 * spammed by the user's ctl_io.
2312 		 */
2313 		pool_tmp = io->io_hdr.pool;
2314 
2315 		memcpy(io, (void *)addr, sizeof(*io));
2316 
2317 		io->io_hdr.pool = pool_tmp;
2318 		/*
2319 		 * No status yet, so make sure the status is set properly.
2320 		 */
2321 		io->io_hdr.status = CTL_STATUS_NONE;
2322 
2323 		/*
2324 		 * The user sets the initiator ID, target and LUN IDs.
2325 		 */
2326 		io->io_hdr.nexus.targ_port = softc->ioctl_info.port.targ_port;
2327 		io->io_hdr.flags |= CTL_FLAG_USER_REQ;
2328 		if ((io->io_hdr.io_type == CTL_IO_SCSI)
2329 		 && (io->scsiio.tag_type != CTL_TAG_UNTAGGED))
2330 			io->scsiio.tag_num = softc->ioctl_info.cur_tag_num++;
2331 
2332 		retval = ctl_ioctl_submit_wait(io);
2333 
2334 		if (retval != 0) {
2335 			ctl_free_io(io);
2336 			break;
2337 		}
2338 
2339 		memcpy((void *)addr, io, sizeof(*io));
2340 
2341 		/* return this to our pool */
2342 		ctl_free_io(io);
2343 
2344 		break;
2345 	}
2346 	case CTL_ENABLE_PORT:
2347 	case CTL_DISABLE_PORT:
2348 	case CTL_SET_PORT_WWNS: {
2349 		struct ctl_port *port;
2350 		struct ctl_port_entry *entry;
2351 
2352 		entry = (struct ctl_port_entry *)addr;
2353 
2354 		mtx_lock(&softc->ctl_lock);
2355 		STAILQ_FOREACH(port, &softc->port_list, links) {
2356 			int action, done;
2357 
2358 			action = 0;
2359 			done = 0;
2360 
2361 			if ((entry->port_type == CTL_PORT_NONE)
2362 			 && (entry->targ_port == port->targ_port)) {
2363 				/*
2364 				 * If the user only wants to enable or
2365 				 * disable or set WWNs on a specific port,
2366 				 * do the operation and we're done.
2367 				 */
2368 				action = 1;
2369 				done = 1;
2370 			} else if (entry->port_type & port->port_type) {
2371 				/*
2372 				 * Compare the user's type mask with the
2373 				 * particular frontend type to see if we
2374 				 * have a match.
2375 				 */
2376 				action = 1;
2377 				done = 0;
2378 
2379 				/*
2380 				 * Make sure the user isn't trying to set
2381 				 * WWNs on multiple ports at the same time.
2382 				 */
2383 				if (cmd == CTL_SET_PORT_WWNS) {
2384 					printf("%s: Can't set WWNs on "
2385 					       "multiple ports\n", __func__);
2386 					retval = EINVAL;
2387 					break;
2388 				}
2389 			}
2390 			if (action != 0) {
2391 				/*
2392 				 * XXX KDM we have to drop the lock here,
2393 				 * because the online/offline operations
2394 				 * can potentially block.  We need to
2395 				 * reference count the frontends so they
2396 				 * can't go away,
2397 				 */
2398 				mtx_unlock(&softc->ctl_lock);
2399 
2400 				if (cmd == CTL_ENABLE_PORT) {
2401 					struct ctl_lun *lun;
2402 
2403 					STAILQ_FOREACH(lun, &softc->lun_list,
2404 						       links) {
2405 						port->lun_enable(port->targ_lun_arg,
2406 						    lun->target,
2407 						    lun->lun);
2408 					}
2409 
2410 					ctl_port_online(port);
2411 				} else if (cmd == CTL_DISABLE_PORT) {
2412 					struct ctl_lun *lun;
2413 
2414 					ctl_port_offline(port);
2415 
2416 					STAILQ_FOREACH(lun, &softc->lun_list,
2417 						       links) {
2418 						port->lun_disable(
2419 						    port->targ_lun_arg,
2420 						    lun->target,
2421 						    lun->lun);
2422 					}
2423 				}
2424 
2425 				mtx_lock(&softc->ctl_lock);
2426 
2427 				if (cmd == CTL_SET_PORT_WWNS)
2428 					ctl_port_set_wwns(port,
2429 					    (entry->flags & CTL_PORT_WWNN_VALID) ?
2430 					    1 : 0, entry->wwnn,
2431 					    (entry->flags & CTL_PORT_WWPN_VALID) ?
2432 					    1 : 0, entry->wwpn);
2433 			}
2434 			if (done != 0)
2435 				break;
2436 		}
2437 		mtx_unlock(&softc->ctl_lock);
2438 		break;
2439 	}
2440 	case CTL_GET_PORT_LIST: {
2441 		struct ctl_port *port;
2442 		struct ctl_port_list *list;
2443 		int i;
2444 
2445 		list = (struct ctl_port_list *)addr;
2446 
2447 		if (list->alloc_len != (list->alloc_num *
2448 		    sizeof(struct ctl_port_entry))) {
2449 			printf("%s: CTL_GET_PORT_LIST: alloc_len %u != "
2450 			       "alloc_num %u * sizeof(struct ctl_port_entry) "
2451 			       "%zu\n", __func__, list->alloc_len,
2452 			       list->alloc_num, sizeof(struct ctl_port_entry));
2453 			retval = EINVAL;
2454 			break;
2455 		}
2456 		list->fill_len = 0;
2457 		list->fill_num = 0;
2458 		list->dropped_num = 0;
2459 		i = 0;
2460 		mtx_lock(&softc->ctl_lock);
2461 		STAILQ_FOREACH(port, &softc->port_list, links) {
2462 			struct ctl_port_entry entry, *list_entry;
2463 
2464 			if (list->fill_num >= list->alloc_num) {
2465 				list->dropped_num++;
2466 				continue;
2467 			}
2468 
2469 			entry.port_type = port->port_type;
2470 			strlcpy(entry.port_name, port->port_name,
2471 				sizeof(entry.port_name));
2472 			entry.targ_port = port->targ_port;
2473 			entry.physical_port = port->physical_port;
2474 			entry.virtual_port = port->virtual_port;
2475 			entry.wwnn = port->wwnn;
2476 			entry.wwpn = port->wwpn;
2477 			if (port->status & CTL_PORT_STATUS_ONLINE)
2478 				entry.online = 1;
2479 			else
2480 				entry.online = 0;
2481 
2482 			list_entry = &list->entries[i];
2483 
2484 			retval = copyout(&entry, list_entry, sizeof(entry));
2485 			if (retval != 0) {
2486 				printf("%s: CTL_GET_PORT_LIST: copyout "
2487 				       "returned %d\n", __func__, retval);
2488 				break;
2489 			}
2490 			i++;
2491 			list->fill_num++;
2492 			list->fill_len += sizeof(entry);
2493 		}
2494 		mtx_unlock(&softc->ctl_lock);
2495 
2496 		/*
2497 		 * If this is non-zero, we had a copyout fault, so there's
2498 		 * probably no point in attempting to set the status inside
2499 		 * the structure.
2500 		 */
2501 		if (retval != 0)
2502 			break;
2503 
2504 		if (list->dropped_num > 0)
2505 			list->status = CTL_PORT_LIST_NEED_MORE_SPACE;
2506 		else
2507 			list->status = CTL_PORT_LIST_OK;
2508 		break;
2509 	}
2510 	case CTL_DUMP_OOA: {
2511 		struct ctl_lun *lun;
2512 		union ctl_io *io;
2513 		char printbuf[128];
2514 		struct sbuf sb;
2515 
2516 		mtx_lock(&softc->ctl_lock);
2517 		printf("Dumping OOA queues:\n");
2518 		STAILQ_FOREACH(lun, &softc->lun_list, links) {
2519 			mtx_lock(&lun->lun_lock);
2520 			for (io = (union ctl_io *)TAILQ_FIRST(
2521 			     &lun->ooa_queue); io != NULL;
2522 			     io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2523 			     ooa_links)) {
2524 				sbuf_new(&sb, printbuf, sizeof(printbuf),
2525 					 SBUF_FIXEDLEN);
2526 				sbuf_printf(&sb, "LUN %jd tag 0x%04x%s%s%s%s: ",
2527 					    (intmax_t)lun->lun,
2528 					    io->scsiio.tag_num,
2529 					    (io->io_hdr.flags &
2530 					    CTL_FLAG_BLOCKED) ? "" : " BLOCKED",
2531 					    (io->io_hdr.flags &
2532 					    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
2533 					    (io->io_hdr.flags &
2534 					    CTL_FLAG_ABORT) ? " ABORT" : "",
2535 			                    (io->io_hdr.flags &
2536 		                        CTL_FLAG_IS_WAS_ON_RTR) ? " RTR" : "");
2537 				ctl_scsi_command_string(&io->scsiio, NULL, &sb);
2538 				sbuf_finish(&sb);
2539 				printf("%s\n", sbuf_data(&sb));
2540 			}
2541 			mtx_unlock(&lun->lun_lock);
2542 		}
2543 		printf("OOA queues dump done\n");
2544 		mtx_unlock(&softc->ctl_lock);
2545 		break;
2546 	}
2547 	case CTL_GET_OOA: {
2548 		struct ctl_lun *lun;
2549 		struct ctl_ooa *ooa_hdr;
2550 		struct ctl_ooa_entry *entries;
2551 		uint32_t cur_fill_num;
2552 
2553 		ooa_hdr = (struct ctl_ooa *)addr;
2554 
2555 		if ((ooa_hdr->alloc_len == 0)
2556 		 || (ooa_hdr->alloc_num == 0)) {
2557 			printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2558 			       "must be non-zero\n", __func__,
2559 			       ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2560 			retval = EINVAL;
2561 			break;
2562 		}
2563 
2564 		if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2565 		    sizeof(struct ctl_ooa_entry))) {
2566 			printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2567 			       "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2568 			       __func__, ooa_hdr->alloc_len,
2569 			       ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2570 			retval = EINVAL;
2571 			break;
2572 		}
2573 
2574 		entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2575 		if (entries == NULL) {
2576 			printf("%s: could not allocate %d bytes for OOA "
2577 			       "dump\n", __func__, ooa_hdr->alloc_len);
2578 			retval = ENOMEM;
2579 			break;
2580 		}
2581 
2582 		mtx_lock(&softc->ctl_lock);
2583 		if (((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0)
2584 		 && ((ooa_hdr->lun_num >= CTL_MAX_LUNS)
2585 		  || (softc->ctl_luns[ooa_hdr->lun_num] == NULL))) {
2586 			mtx_unlock(&softc->ctl_lock);
2587 			free(entries, M_CTL);
2588 			printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2589 			       __func__, (uintmax_t)ooa_hdr->lun_num);
2590 			retval = EINVAL;
2591 			break;
2592 		}
2593 
2594 		cur_fill_num = 0;
2595 
2596 		if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2597 			STAILQ_FOREACH(lun, &softc->lun_list, links) {
2598 				retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2599 					ooa_hdr, entries);
2600 				if (retval != 0)
2601 					break;
2602 			}
2603 			if (retval != 0) {
2604 				mtx_unlock(&softc->ctl_lock);
2605 				free(entries, M_CTL);
2606 				break;
2607 			}
2608 		} else {
2609 			lun = softc->ctl_luns[ooa_hdr->lun_num];
2610 
2611 			retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,ooa_hdr,
2612 						    entries);
2613 		}
2614 		mtx_unlock(&softc->ctl_lock);
2615 
2616 		ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2617 		ooa_hdr->fill_len = ooa_hdr->fill_num *
2618 			sizeof(struct ctl_ooa_entry);
2619 		retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2620 		if (retval != 0) {
2621 			printf("%s: error copying out %d bytes for OOA dump\n",
2622 			       __func__, ooa_hdr->fill_len);
2623 		}
2624 
2625 		getbintime(&ooa_hdr->cur_bt);
2626 
2627 		if (cur_fill_num > ooa_hdr->alloc_num) {
2628 			ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2629 			ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2630 		} else {
2631 			ooa_hdr->dropped_num = 0;
2632 			ooa_hdr->status = CTL_OOA_OK;
2633 		}
2634 
2635 		free(entries, M_CTL);
2636 		break;
2637 	}
2638 	case CTL_CHECK_OOA: {
2639 		union ctl_io *io;
2640 		struct ctl_lun *lun;
2641 		struct ctl_ooa_info *ooa_info;
2642 
2643 
2644 		ooa_info = (struct ctl_ooa_info *)addr;
2645 
2646 		if (ooa_info->lun_id >= CTL_MAX_LUNS) {
2647 			ooa_info->status = CTL_OOA_INVALID_LUN;
2648 			break;
2649 		}
2650 		mtx_lock(&softc->ctl_lock);
2651 		lun = softc->ctl_luns[ooa_info->lun_id];
2652 		if (lun == NULL) {
2653 			mtx_unlock(&softc->ctl_lock);
2654 			ooa_info->status = CTL_OOA_INVALID_LUN;
2655 			break;
2656 		}
2657 		mtx_lock(&lun->lun_lock);
2658 		mtx_unlock(&softc->ctl_lock);
2659 		ooa_info->num_entries = 0;
2660 		for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
2661 		     io != NULL; io = (union ctl_io *)TAILQ_NEXT(
2662 		     &io->io_hdr, ooa_links)) {
2663 			ooa_info->num_entries++;
2664 		}
2665 		mtx_unlock(&lun->lun_lock);
2666 
2667 		ooa_info->status = CTL_OOA_SUCCESS;
2668 
2669 		break;
2670 	}
2671 	case CTL_HARD_START:
2672 	case CTL_HARD_STOP: {
2673 		struct ctl_fe_ioctl_startstop_info ss_info;
2674 		struct cfi_metatask *metatask;
2675 		struct mtx hs_mtx;
2676 
2677 		mtx_init(&hs_mtx, "HS Mutex", NULL, MTX_DEF);
2678 
2679 		cv_init(&ss_info.sem, "hard start/stop cv" );
2680 
2681 		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2682 		if (metatask == NULL) {
2683 			retval = ENOMEM;
2684 			mtx_destroy(&hs_mtx);
2685 			break;
2686 		}
2687 
2688 		if (cmd == CTL_HARD_START)
2689 			metatask->tasktype = CFI_TASK_STARTUP;
2690 		else
2691 			metatask->tasktype = CFI_TASK_SHUTDOWN;
2692 
2693 		metatask->callback = ctl_ioctl_hard_startstop_callback;
2694 		metatask->callback_arg = &ss_info;
2695 
2696 		cfi_action(metatask);
2697 
2698 		/* Wait for the callback */
2699 		mtx_lock(&hs_mtx);
2700 		cv_wait_sig(&ss_info.sem, &hs_mtx);
2701 		mtx_unlock(&hs_mtx);
2702 
2703 		/*
2704 		 * All information has been copied from the metatask by the
2705 		 * time cv_broadcast() is called, so we free the metatask here.
2706 		 */
2707 		cfi_free_metatask(metatask);
2708 
2709 		memcpy((void *)addr, &ss_info.hs_info, sizeof(ss_info.hs_info));
2710 
2711 		mtx_destroy(&hs_mtx);
2712 		break;
2713 	}
2714 	case CTL_BBRREAD: {
2715 		struct ctl_bbrread_info *bbr_info;
2716 		struct ctl_fe_ioctl_bbrread_info fe_bbr_info;
2717 		struct mtx bbr_mtx;
2718 		struct cfi_metatask *metatask;
2719 
2720 		bbr_info = (struct ctl_bbrread_info *)addr;
2721 
2722 		bzero(&fe_bbr_info, sizeof(fe_bbr_info));
2723 
2724 		bzero(&bbr_mtx, sizeof(bbr_mtx));
2725 		mtx_init(&bbr_mtx, "BBR Mutex", NULL, MTX_DEF);
2726 
2727 		fe_bbr_info.bbr_info = bbr_info;
2728 		fe_bbr_info.lock = &bbr_mtx;
2729 
2730 		cv_init(&fe_bbr_info.sem, "BBR read cv");
2731 		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2732 
2733 		if (metatask == NULL) {
2734 			mtx_destroy(&bbr_mtx);
2735 			cv_destroy(&fe_bbr_info.sem);
2736 			retval = ENOMEM;
2737 			break;
2738 		}
2739 		metatask->tasktype = CFI_TASK_BBRREAD;
2740 		metatask->callback = ctl_ioctl_bbrread_callback;
2741 		metatask->callback_arg = &fe_bbr_info;
2742 		metatask->taskinfo.bbrread.lun_num = bbr_info->lun_num;
2743 		metatask->taskinfo.bbrread.lba = bbr_info->lba;
2744 		metatask->taskinfo.bbrread.len = bbr_info->len;
2745 
2746 		cfi_action(metatask);
2747 
2748 		mtx_lock(&bbr_mtx);
2749 		while (fe_bbr_info.wakeup_done == 0)
2750 			cv_wait_sig(&fe_bbr_info.sem, &bbr_mtx);
2751 		mtx_unlock(&bbr_mtx);
2752 
2753 		bbr_info->status = metatask->status;
2754 		bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2755 		bbr_info->scsi_status = metatask->taskinfo.bbrread.scsi_status;
2756 		memcpy(&bbr_info->sense_data,
2757 		       &metatask->taskinfo.bbrread.sense_data,
2758 		       ctl_min(sizeof(bbr_info->sense_data),
2759 			       sizeof(metatask->taskinfo.bbrread.sense_data)));
2760 
2761 		cfi_free_metatask(metatask);
2762 
2763 		mtx_destroy(&bbr_mtx);
2764 		cv_destroy(&fe_bbr_info.sem);
2765 
2766 		break;
2767 	}
2768 	case CTL_DELAY_IO: {
2769 		struct ctl_io_delay_info *delay_info;
2770 #ifdef CTL_IO_DELAY
2771 		struct ctl_lun *lun;
2772 #endif /* CTL_IO_DELAY */
2773 
2774 		delay_info = (struct ctl_io_delay_info *)addr;
2775 
2776 #ifdef CTL_IO_DELAY
2777 		mtx_lock(&softc->ctl_lock);
2778 
2779 		if ((delay_info->lun_id >= CTL_MAX_LUNS)
2780 		 || (softc->ctl_luns[delay_info->lun_id] == NULL)) {
2781 			delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2782 		} else {
2783 			lun = softc->ctl_luns[delay_info->lun_id];
2784 			mtx_lock(&lun->lun_lock);
2785 
2786 			delay_info->status = CTL_DELAY_STATUS_OK;
2787 
2788 			switch (delay_info->delay_type) {
2789 			case CTL_DELAY_TYPE_CONT:
2790 				break;
2791 			case CTL_DELAY_TYPE_ONESHOT:
2792 				break;
2793 			default:
2794 				delay_info->status =
2795 					CTL_DELAY_STATUS_INVALID_TYPE;
2796 				break;
2797 			}
2798 
2799 			switch (delay_info->delay_loc) {
2800 			case CTL_DELAY_LOC_DATAMOVE:
2801 				lun->delay_info.datamove_type =
2802 					delay_info->delay_type;
2803 				lun->delay_info.datamove_delay =
2804 					delay_info->delay_secs;
2805 				break;
2806 			case CTL_DELAY_LOC_DONE:
2807 				lun->delay_info.done_type =
2808 					delay_info->delay_type;
2809 				lun->delay_info.done_delay =
2810 					delay_info->delay_secs;
2811 				break;
2812 			default:
2813 				delay_info->status =
2814 					CTL_DELAY_STATUS_INVALID_LOC;
2815 				break;
2816 			}
2817 			mtx_unlock(&lun->lun_lock);
2818 		}
2819 
2820 		mtx_unlock(&softc->ctl_lock);
2821 #else
2822 		delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2823 #endif /* CTL_IO_DELAY */
2824 		break;
2825 	}
2826 	case CTL_REALSYNC_SET: {
2827 		int *syncstate;
2828 
2829 		syncstate = (int *)addr;
2830 
2831 		mtx_lock(&softc->ctl_lock);
2832 		switch (*syncstate) {
2833 		case 0:
2834 			softc->flags &= ~CTL_FLAG_REAL_SYNC;
2835 			break;
2836 		case 1:
2837 			softc->flags |= CTL_FLAG_REAL_SYNC;
2838 			break;
2839 		default:
2840 			retval = EINVAL;
2841 			break;
2842 		}
2843 		mtx_unlock(&softc->ctl_lock);
2844 		break;
2845 	}
2846 	case CTL_REALSYNC_GET: {
2847 		int *syncstate;
2848 
2849 		syncstate = (int*)addr;
2850 
2851 		mtx_lock(&softc->ctl_lock);
2852 		if (softc->flags & CTL_FLAG_REAL_SYNC)
2853 			*syncstate = 1;
2854 		else
2855 			*syncstate = 0;
2856 		mtx_unlock(&softc->ctl_lock);
2857 
2858 		break;
2859 	}
2860 	case CTL_SETSYNC:
2861 	case CTL_GETSYNC: {
2862 		struct ctl_sync_info *sync_info;
2863 		struct ctl_lun *lun;
2864 
2865 		sync_info = (struct ctl_sync_info *)addr;
2866 
2867 		mtx_lock(&softc->ctl_lock);
2868 		lun = softc->ctl_luns[sync_info->lun_id];
2869 		if (lun == NULL) {
2870 			mtx_unlock(&softc->ctl_lock);
2871 			sync_info->status = CTL_GS_SYNC_NO_LUN;
2872 		}
2873 		/*
2874 		 * Get or set the sync interval.  We're not bounds checking
2875 		 * in the set case, hopefully the user won't do something
2876 		 * silly.
2877 		 */
2878 		mtx_lock(&lun->lun_lock);
2879 		mtx_unlock(&softc->ctl_lock);
2880 		if (cmd == CTL_GETSYNC)
2881 			sync_info->sync_interval = lun->sync_interval;
2882 		else
2883 			lun->sync_interval = sync_info->sync_interval;
2884 		mtx_unlock(&lun->lun_lock);
2885 
2886 		sync_info->status = CTL_GS_SYNC_OK;
2887 
2888 		break;
2889 	}
2890 	case CTL_GETSTATS: {
2891 		struct ctl_stats *stats;
2892 		struct ctl_lun *lun;
2893 		int i;
2894 
2895 		stats = (struct ctl_stats *)addr;
2896 
2897 		if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) >
2898 		     stats->alloc_len) {
2899 			stats->status = CTL_SS_NEED_MORE_SPACE;
2900 			stats->num_luns = softc->num_luns;
2901 			break;
2902 		}
2903 		/*
2904 		 * XXX KDM no locking here.  If the LUN list changes,
2905 		 * things can blow up.
2906 		 */
2907 		for (i = 0, lun = STAILQ_FIRST(&softc->lun_list); lun != NULL;
2908 		     i++, lun = STAILQ_NEXT(lun, links)) {
2909 			retval = copyout(&lun->stats, &stats->lun_stats[i],
2910 					 sizeof(lun->stats));
2911 			if (retval != 0)
2912 				break;
2913 		}
2914 		stats->num_luns = softc->num_luns;
2915 		stats->fill_len = sizeof(struct ctl_lun_io_stats) *
2916 				 softc->num_luns;
2917 		stats->status = CTL_SS_OK;
2918 #ifdef CTL_TIME_IO
2919 		stats->flags = CTL_STATS_FLAG_TIME_VALID;
2920 #else
2921 		stats->flags = CTL_STATS_FLAG_NONE;
2922 #endif
2923 		getnanouptime(&stats->timestamp);
2924 		break;
2925 	}
2926 	case CTL_ERROR_INJECT: {
2927 		struct ctl_error_desc *err_desc, *new_err_desc;
2928 		struct ctl_lun *lun;
2929 
2930 		err_desc = (struct ctl_error_desc *)addr;
2931 
2932 		new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2933 				      M_WAITOK | M_ZERO);
2934 		bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2935 
2936 		mtx_lock(&softc->ctl_lock);
2937 		lun = softc->ctl_luns[err_desc->lun_id];
2938 		if (lun == NULL) {
2939 			mtx_unlock(&softc->ctl_lock);
2940 			free(new_err_desc, M_CTL);
2941 			printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2942 			       __func__, (uintmax_t)err_desc->lun_id);
2943 			retval = EINVAL;
2944 			break;
2945 		}
2946 		mtx_lock(&lun->lun_lock);
2947 		mtx_unlock(&softc->ctl_lock);
2948 
2949 		/*
2950 		 * We could do some checking here to verify the validity
2951 		 * of the request, but given the complexity of error
2952 		 * injection requests, the checking logic would be fairly
2953 		 * complex.
2954 		 *
2955 		 * For now, if the request is invalid, it just won't get
2956 		 * executed and might get deleted.
2957 		 */
2958 		STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2959 
2960 		/*
2961 		 * XXX KDM check to make sure the serial number is unique,
2962 		 * in case we somehow manage to wrap.  That shouldn't
2963 		 * happen for a very long time, but it's the right thing to
2964 		 * do.
2965 		 */
2966 		new_err_desc->serial = lun->error_serial;
2967 		err_desc->serial = lun->error_serial;
2968 		lun->error_serial++;
2969 
2970 		mtx_unlock(&lun->lun_lock);
2971 		break;
2972 	}
2973 	case CTL_ERROR_INJECT_DELETE: {
2974 		struct ctl_error_desc *delete_desc, *desc, *desc2;
2975 		struct ctl_lun *lun;
2976 		int delete_done;
2977 
2978 		delete_desc = (struct ctl_error_desc *)addr;
2979 		delete_done = 0;
2980 
2981 		mtx_lock(&softc->ctl_lock);
2982 		lun = softc->ctl_luns[delete_desc->lun_id];
2983 		if (lun == NULL) {
2984 			mtx_unlock(&softc->ctl_lock);
2985 			printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2986 			       __func__, (uintmax_t)delete_desc->lun_id);
2987 			retval = EINVAL;
2988 			break;
2989 		}
2990 		mtx_lock(&lun->lun_lock);
2991 		mtx_unlock(&softc->ctl_lock);
2992 		STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2993 			if (desc->serial != delete_desc->serial)
2994 				continue;
2995 
2996 			STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2997 				      links);
2998 			free(desc, M_CTL);
2999 			delete_done = 1;
3000 		}
3001 		mtx_unlock(&lun->lun_lock);
3002 		if (delete_done == 0) {
3003 			printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
3004 			       "error serial %ju on LUN %u\n", __func__,
3005 			       delete_desc->serial, delete_desc->lun_id);
3006 			retval = EINVAL;
3007 			break;
3008 		}
3009 		break;
3010 	}
3011 	case CTL_DUMP_STRUCTS: {
3012 		int i, j, k, idx;
3013 		struct ctl_port *port;
3014 		struct ctl_frontend *fe;
3015 
3016 		mtx_lock(&softc->ctl_lock);
3017 		printf("CTL Persistent Reservation information start:\n");
3018 		for (i = 0; i < CTL_MAX_LUNS; i++) {
3019 			struct ctl_lun *lun;
3020 
3021 			lun = softc->ctl_luns[i];
3022 
3023 			if ((lun == NULL)
3024 			 || ((lun->flags & CTL_LUN_DISABLED) != 0))
3025 				continue;
3026 
3027 			for (j = 0; j < (CTL_MAX_PORTS * 2); j++) {
3028 				for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
3029 					idx = j * CTL_MAX_INIT_PER_PORT + k;
3030 					if (lun->pr_keys[idx] == 0)
3031 						continue;
3032 					printf("  LUN %d port %d iid %d key "
3033 					       "%#jx\n", i, j, k,
3034 					       (uintmax_t)lun->pr_keys[idx]);
3035 				}
3036 			}
3037 		}
3038 		printf("CTL Persistent Reservation information end\n");
3039 		printf("CTL Ports:\n");
3040 		STAILQ_FOREACH(port, &softc->port_list, links) {
3041 			printf("  Port %d '%s' Frontend '%s' Type %u pp %d vp %d WWNN "
3042 			       "%#jx WWPN %#jx\n", port->targ_port, port->port_name,
3043 			       port->frontend->name, port->port_type,
3044 			       port->physical_port, port->virtual_port,
3045 			       (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
3046 			for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3047 				if (port->wwpn_iid[j].in_use == 0 &&
3048 				    port->wwpn_iid[j].wwpn == 0 &&
3049 				    port->wwpn_iid[j].name == NULL)
3050 					continue;
3051 
3052 				printf("    iid %u use %d WWPN %#jx '%s'\n",
3053 				    j, port->wwpn_iid[j].in_use,
3054 				    (uintmax_t)port->wwpn_iid[j].wwpn,
3055 				    port->wwpn_iid[j].name);
3056 			}
3057 		}
3058 		printf("CTL Port information end\n");
3059 		mtx_unlock(&softc->ctl_lock);
3060 		/*
3061 		 * XXX KDM calling this without a lock.  We'd likely want
3062 		 * to drop the lock before calling the frontend's dump
3063 		 * routine anyway.
3064 		 */
3065 		printf("CTL Frontends:\n");
3066 		STAILQ_FOREACH(fe, &softc->fe_list, links) {
3067 			printf("  Frontend '%s'\n", fe->name);
3068 			if (fe->fe_dump != NULL)
3069 				fe->fe_dump();
3070 		}
3071 		printf("CTL Frontend information end\n");
3072 		break;
3073 	}
3074 	case CTL_LUN_REQ: {
3075 		struct ctl_lun_req *lun_req;
3076 		struct ctl_backend_driver *backend;
3077 
3078 		lun_req = (struct ctl_lun_req *)addr;
3079 
3080 		backend = ctl_backend_find(lun_req->backend);
3081 		if (backend == NULL) {
3082 			lun_req->status = CTL_LUN_ERROR;
3083 			snprintf(lun_req->error_str,
3084 				 sizeof(lun_req->error_str),
3085 				 "Backend \"%s\" not found.",
3086 				 lun_req->backend);
3087 			break;
3088 		}
3089 		if (lun_req->num_be_args > 0) {
3090 			lun_req->kern_be_args = ctl_copyin_args(
3091 				lun_req->num_be_args,
3092 				lun_req->be_args,
3093 				lun_req->error_str,
3094 				sizeof(lun_req->error_str));
3095 			if (lun_req->kern_be_args == NULL) {
3096 				lun_req->status = CTL_LUN_ERROR;
3097 				break;
3098 			}
3099 		}
3100 
3101 		retval = backend->ioctl(dev, cmd, addr, flag, td);
3102 
3103 		if (lun_req->num_be_args > 0) {
3104 			ctl_copyout_args(lun_req->num_be_args,
3105 				      lun_req->kern_be_args);
3106 			ctl_free_args(lun_req->num_be_args,
3107 				      lun_req->kern_be_args);
3108 		}
3109 		break;
3110 	}
3111 	case CTL_LUN_LIST: {
3112 		struct sbuf *sb;
3113 		struct ctl_lun *lun;
3114 		struct ctl_lun_list *list;
3115 		struct ctl_option *opt;
3116 
3117 		list = (struct ctl_lun_list *)addr;
3118 
3119 		/*
3120 		 * Allocate a fixed length sbuf here, based on the length
3121 		 * of the user's buffer.  We could allocate an auto-extending
3122 		 * buffer, and then tell the user how much larger our
3123 		 * amount of data is than his buffer, but that presents
3124 		 * some problems:
3125 		 *
3126 		 * 1.  The sbuf(9) routines use a blocking malloc, and so
3127 		 *     we can't hold a lock while calling them with an
3128 		 *     auto-extending buffer.
3129  		 *
3130 		 * 2.  There is not currently a LUN reference counting
3131 		 *     mechanism, outside of outstanding transactions on
3132 		 *     the LUN's OOA queue.  So a LUN could go away on us
3133 		 *     while we're getting the LUN number, backend-specific
3134 		 *     information, etc.  Thus, given the way things
3135 		 *     currently work, we need to hold the CTL lock while
3136 		 *     grabbing LUN information.
3137 		 *
3138 		 * So, from the user's standpoint, the best thing to do is
3139 		 * allocate what he thinks is a reasonable buffer length,
3140 		 * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
3141 		 * double the buffer length and try again.  (And repeat
3142 		 * that until he succeeds.)
3143 		 */
3144 		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3145 		if (sb == NULL) {
3146 			list->status = CTL_LUN_LIST_ERROR;
3147 			snprintf(list->error_str, sizeof(list->error_str),
3148 				 "Unable to allocate %d bytes for LUN list",
3149 				 list->alloc_len);
3150 			break;
3151 		}
3152 
3153 		sbuf_printf(sb, "<ctllunlist>\n");
3154 
3155 		mtx_lock(&softc->ctl_lock);
3156 		STAILQ_FOREACH(lun, &softc->lun_list, links) {
3157 			mtx_lock(&lun->lun_lock);
3158 			retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3159 					     (uintmax_t)lun->lun);
3160 
3161 			/*
3162 			 * Bail out as soon as we see that we've overfilled
3163 			 * the buffer.
3164 			 */
3165 			if (retval != 0)
3166 				break;
3167 
3168 			retval = sbuf_printf(sb, "\t<backend_type>%s"
3169 					     "</backend_type>\n",
3170 					     (lun->backend == NULL) ?  "none" :
3171 					     lun->backend->name);
3172 
3173 			if (retval != 0)
3174 				break;
3175 
3176 			retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
3177 					     lun->be_lun->lun_type);
3178 
3179 			if (retval != 0)
3180 				break;
3181 
3182 			if (lun->backend == NULL) {
3183 				retval = sbuf_printf(sb, "</lun>\n");
3184 				if (retval != 0)
3185 					break;
3186 				continue;
3187 			}
3188 
3189 			retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
3190 					     (lun->be_lun->maxlba > 0) ?
3191 					     lun->be_lun->maxlba + 1 : 0);
3192 
3193 			if (retval != 0)
3194 				break;
3195 
3196 			retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
3197 					     lun->be_lun->blocksize);
3198 
3199 			if (retval != 0)
3200 				break;
3201 
3202 			retval = sbuf_printf(sb, "\t<serial_number>");
3203 
3204 			if (retval != 0)
3205 				break;
3206 
3207 			retval = ctl_sbuf_printf_esc(sb,
3208 						     lun->be_lun->serial_num);
3209 
3210 			if (retval != 0)
3211 				break;
3212 
3213 			retval = sbuf_printf(sb, "</serial_number>\n");
3214 
3215 			if (retval != 0)
3216 				break;
3217 
3218 			retval = sbuf_printf(sb, "\t<device_id>");
3219 
3220 			if (retval != 0)
3221 				break;
3222 
3223 			retval = ctl_sbuf_printf_esc(sb,lun->be_lun->device_id);
3224 
3225 			if (retval != 0)
3226 				break;
3227 
3228 			retval = sbuf_printf(sb, "</device_id>\n");
3229 
3230 			if (retval != 0)
3231 				break;
3232 
3233 			if (lun->backend->lun_info != NULL) {
3234 				retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3235 				if (retval != 0)
3236 					break;
3237 			}
3238 			STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
3239 				retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3240 				    opt->name, opt->value, opt->name);
3241 				if (retval != 0)
3242 					break;
3243 			}
3244 
3245 			retval = sbuf_printf(sb, "</lun>\n");
3246 
3247 			if (retval != 0)
3248 				break;
3249 			mtx_unlock(&lun->lun_lock);
3250 		}
3251 		if (lun != NULL)
3252 			mtx_unlock(&lun->lun_lock);
3253 		mtx_unlock(&softc->ctl_lock);
3254 
3255 		if ((retval != 0)
3256 		 || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3257 			retval = 0;
3258 			sbuf_delete(sb);
3259 			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3260 			snprintf(list->error_str, sizeof(list->error_str),
3261 				 "Out of space, %d bytes is too small",
3262 				 list->alloc_len);
3263 			break;
3264 		}
3265 
3266 		sbuf_finish(sb);
3267 
3268 		retval = copyout(sbuf_data(sb), list->lun_xml,
3269 				 sbuf_len(sb) + 1);
3270 
3271 		list->fill_len = sbuf_len(sb) + 1;
3272 		list->status = CTL_LUN_LIST_OK;
3273 		sbuf_delete(sb);
3274 		break;
3275 	}
3276 	case CTL_ISCSI: {
3277 		struct ctl_iscsi *ci;
3278 		struct ctl_frontend *fe;
3279 
3280 		ci = (struct ctl_iscsi *)addr;
3281 
3282 		fe = ctl_frontend_find("iscsi");
3283 		if (fe == NULL) {
3284 			ci->status = CTL_ISCSI_ERROR;
3285 			snprintf(ci->error_str, sizeof(ci->error_str),
3286 			    "Frontend \"iscsi\" not found.");
3287 			break;
3288 		}
3289 
3290 		retval = fe->ioctl(dev, cmd, addr, flag, td);
3291 		break;
3292 	}
3293 	case CTL_PORT_REQ: {
3294 		struct ctl_req *req;
3295 		struct ctl_frontend *fe;
3296 
3297 		req = (struct ctl_req *)addr;
3298 
3299 		fe = ctl_frontend_find(req->driver);
3300 		if (fe == NULL) {
3301 			req->status = CTL_LUN_ERROR;
3302 			snprintf(req->error_str, sizeof(req->error_str),
3303 			    "Frontend \"%s\" not found.", req->driver);
3304 			break;
3305 		}
3306 		if (req->num_args > 0) {
3307 			req->kern_args = ctl_copyin_args(req->num_args,
3308 			    req->args, req->error_str, sizeof(req->error_str));
3309 			if (req->kern_args == NULL) {
3310 				req->status = CTL_LUN_ERROR;
3311 				break;
3312 			}
3313 		}
3314 
3315 		retval = fe->ioctl(dev, cmd, addr, flag, td);
3316 
3317 		if (req->num_args > 0) {
3318 			ctl_copyout_args(req->num_args, req->kern_args);
3319 			ctl_free_args(req->num_args, req->kern_args);
3320 		}
3321 		break;
3322 	}
3323 	case CTL_PORT_LIST: {
3324 		struct sbuf *sb;
3325 		struct ctl_port *port;
3326 		struct ctl_lun_list *list;
3327 		struct ctl_option *opt;
3328 		int j;
3329 
3330 		list = (struct ctl_lun_list *)addr;
3331 
3332 		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3333 		if (sb == NULL) {
3334 			list->status = CTL_LUN_LIST_ERROR;
3335 			snprintf(list->error_str, sizeof(list->error_str),
3336 				 "Unable to allocate %d bytes for LUN list",
3337 				 list->alloc_len);
3338 			break;
3339 		}
3340 
3341 		sbuf_printf(sb, "<ctlportlist>\n");
3342 
3343 		mtx_lock(&softc->ctl_lock);
3344 		STAILQ_FOREACH(port, &softc->port_list, links) {
3345 			retval = sbuf_printf(sb, "<targ_port id=\"%ju\">\n",
3346 					     (uintmax_t)port->targ_port);
3347 
3348 			/*
3349 			 * Bail out as soon as we see that we've overfilled
3350 			 * the buffer.
3351 			 */
3352 			if (retval != 0)
3353 				break;
3354 
3355 			retval = sbuf_printf(sb, "\t<frontend_type>%s"
3356 			    "</frontend_type>\n", port->frontend->name);
3357 			if (retval != 0)
3358 				break;
3359 
3360 			retval = sbuf_printf(sb, "\t<port_type>%d</port_type>\n",
3361 					     port->port_type);
3362 			if (retval != 0)
3363 				break;
3364 
3365 			retval = sbuf_printf(sb, "\t<online>%s</online>\n",
3366 			    (port->status & CTL_PORT_STATUS_ONLINE) ? "YES" : "NO");
3367 			if (retval != 0)
3368 				break;
3369 
3370 			retval = sbuf_printf(sb, "\t<port_name>%s</port_name>\n",
3371 			    port->port_name);
3372 			if (retval != 0)
3373 				break;
3374 
3375 			retval = sbuf_printf(sb, "\t<physical_port>%d</physical_port>\n",
3376 			    port->physical_port);
3377 			if (retval != 0)
3378 				break;
3379 
3380 			retval = sbuf_printf(sb, "\t<virtual_port>%d</virtual_port>\n",
3381 			    port->virtual_port);
3382 			if (retval != 0)
3383 				break;
3384 
3385 			if (port->target_devid != NULL) {
3386 				sbuf_printf(sb, "\t<target>");
3387 				ctl_id_sbuf(port->target_devid, sb);
3388 				sbuf_printf(sb, "</target>\n");
3389 			}
3390 
3391 			if (port->port_devid != NULL) {
3392 				sbuf_printf(sb, "\t<port>");
3393 				ctl_id_sbuf(port->port_devid, sb);
3394 				sbuf_printf(sb, "</port>\n");
3395 			}
3396 
3397 			if (port->port_info != NULL) {
3398 				retval = port->port_info(port->onoff_arg, sb);
3399 				if (retval != 0)
3400 					break;
3401 			}
3402 			STAILQ_FOREACH(opt, &port->options, links) {
3403 				retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3404 				    opt->name, opt->value, opt->name);
3405 				if (retval != 0)
3406 					break;
3407 			}
3408 
3409 			for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3410 				if (port->wwpn_iid[j].in_use == 0 ||
3411 				    (port->wwpn_iid[j].wwpn == 0 &&
3412 				     port->wwpn_iid[j].name == NULL))
3413 					continue;
3414 
3415 				if (port->wwpn_iid[j].name != NULL)
3416 					retval = sbuf_printf(sb,
3417 					    "\t<initiator>%u %s</initiator>\n",
3418 					    j, port->wwpn_iid[j].name);
3419 				else
3420 					retval = sbuf_printf(sb,
3421 					    "\t<initiator>%u naa.%08jx</initiator>\n",
3422 					    j, port->wwpn_iid[j].wwpn);
3423 				if (retval != 0)
3424 					break;
3425 			}
3426 			if (retval != 0)
3427 				break;
3428 
3429 			retval = sbuf_printf(sb, "</targ_port>\n");
3430 			if (retval != 0)
3431 				break;
3432 		}
3433 		mtx_unlock(&softc->ctl_lock);
3434 
3435 		if ((retval != 0)
3436 		 || ((retval = sbuf_printf(sb, "</ctlportlist>\n")) != 0)) {
3437 			retval = 0;
3438 			sbuf_delete(sb);
3439 			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3440 			snprintf(list->error_str, sizeof(list->error_str),
3441 				 "Out of space, %d bytes is too small",
3442 				 list->alloc_len);
3443 			break;
3444 		}
3445 
3446 		sbuf_finish(sb);
3447 
3448 		retval = copyout(sbuf_data(sb), list->lun_xml,
3449 				 sbuf_len(sb) + 1);
3450 
3451 		list->fill_len = sbuf_len(sb) + 1;
3452 		list->status = CTL_LUN_LIST_OK;
3453 		sbuf_delete(sb);
3454 		break;
3455 	}
3456 	default: {
3457 		/* XXX KDM should we fix this? */
3458 #if 0
3459 		struct ctl_backend_driver *backend;
3460 		unsigned int type;
3461 		int found;
3462 
3463 		found = 0;
3464 
3465 		/*
3466 		 * We encode the backend type as the ioctl type for backend
3467 		 * ioctls.  So parse it out here, and then search for a
3468 		 * backend of this type.
3469 		 */
3470 		type = _IOC_TYPE(cmd);
3471 
3472 		STAILQ_FOREACH(backend, &softc->be_list, links) {
3473 			if (backend->type == type) {
3474 				found = 1;
3475 				break;
3476 			}
3477 		}
3478 		if (found == 0) {
3479 			printf("ctl: unknown ioctl command %#lx or backend "
3480 			       "%d\n", cmd, type);
3481 			retval = EINVAL;
3482 			break;
3483 		}
3484 		retval = backend->ioctl(dev, cmd, addr, flag, td);
3485 #endif
3486 		retval = ENOTTY;
3487 		break;
3488 	}
3489 	}
3490 	return (retval);
3491 }
3492 
3493 uint32_t
3494 ctl_get_initindex(struct ctl_nexus *nexus)
3495 {
3496 	if (nexus->targ_port < CTL_MAX_PORTS)
3497 		return (nexus->initid.id +
3498 			(nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3499 	else
3500 		return (nexus->initid.id +
3501 		       ((nexus->targ_port - CTL_MAX_PORTS) *
3502 			CTL_MAX_INIT_PER_PORT));
3503 }
3504 
3505 uint32_t
3506 ctl_get_resindex(struct ctl_nexus *nexus)
3507 {
3508 	return (nexus->initid.id + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3509 }
3510 
3511 uint32_t
3512 ctl_port_idx(int port_num)
3513 {
3514 	if (port_num < CTL_MAX_PORTS)
3515 		return(port_num);
3516 	else
3517 		return(port_num - CTL_MAX_PORTS);
3518 }
3519 
3520 static uint32_t
3521 ctl_map_lun(int port_num, uint32_t lun_id)
3522 {
3523 	struct ctl_port *port;
3524 
3525 	port = control_softc->ctl_ports[ctl_port_idx(port_num)];
3526 	if (port == NULL)
3527 		return (UINT32_MAX);
3528 	if (port->lun_map == NULL)
3529 		return (lun_id);
3530 	return (port->lun_map(port->targ_lun_arg, lun_id));
3531 }
3532 
3533 static uint32_t
3534 ctl_map_lun_back(int port_num, uint32_t lun_id)
3535 {
3536 	struct ctl_port *port;
3537 	uint32_t i;
3538 
3539 	port = control_softc->ctl_ports[ctl_port_idx(port_num)];
3540 	if (port->lun_map == NULL)
3541 		return (lun_id);
3542 	for (i = 0; i < CTL_MAX_LUNS; i++) {
3543 		if (port->lun_map(port->targ_lun_arg, i) == lun_id)
3544 			return (i);
3545 	}
3546 	return (UINT32_MAX);
3547 }
3548 
3549 /*
3550  * Note:  This only works for bitmask sizes that are at least 32 bits, and
3551  * that are a power of 2.
3552  */
3553 int
3554 ctl_ffz(uint32_t *mask, uint32_t size)
3555 {
3556 	uint32_t num_chunks, num_pieces;
3557 	int i, j;
3558 
3559 	num_chunks = (size >> 5);
3560 	if (num_chunks == 0)
3561 		num_chunks++;
3562 	num_pieces = ctl_min((sizeof(uint32_t) * 8), size);
3563 
3564 	for (i = 0; i < num_chunks; i++) {
3565 		for (j = 0; j < num_pieces; j++) {
3566 			if ((mask[i] & (1 << j)) == 0)
3567 				return ((i << 5) + j);
3568 		}
3569 	}
3570 
3571 	return (-1);
3572 }
3573 
3574 int
3575 ctl_set_mask(uint32_t *mask, uint32_t bit)
3576 {
3577 	uint32_t chunk, piece;
3578 
3579 	chunk = bit >> 5;
3580 	piece = bit % (sizeof(uint32_t) * 8);
3581 
3582 	if ((mask[chunk] & (1 << piece)) != 0)
3583 		return (-1);
3584 	else
3585 		mask[chunk] |= (1 << piece);
3586 
3587 	return (0);
3588 }
3589 
3590 int
3591 ctl_clear_mask(uint32_t *mask, uint32_t bit)
3592 {
3593 	uint32_t chunk, piece;
3594 
3595 	chunk = bit >> 5;
3596 	piece = bit % (sizeof(uint32_t) * 8);
3597 
3598 	if ((mask[chunk] & (1 << piece)) == 0)
3599 		return (-1);
3600 	else
3601 		mask[chunk] &= ~(1 << piece);
3602 
3603 	return (0);
3604 }
3605 
3606 int
3607 ctl_is_set(uint32_t *mask, uint32_t bit)
3608 {
3609 	uint32_t chunk, piece;
3610 
3611 	chunk = bit >> 5;
3612 	piece = bit % (sizeof(uint32_t) * 8);
3613 
3614 	if ((mask[chunk] & (1 << piece)) == 0)
3615 		return (0);
3616 	else
3617 		return (1);
3618 }
3619 
3620 #ifdef unused
3621 /*
3622  * The bus, target and lun are optional, they can be filled in later.
3623  * can_wait is used to determine whether we can wait on the malloc or not.
3624  */
3625 union ctl_io*
3626 ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port, uint32_t targ_target,
3627 	      uint32_t targ_lun, int can_wait)
3628 {
3629 	union ctl_io *io;
3630 
3631 	if (can_wait)
3632 		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_WAITOK);
3633 	else
3634 		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_NOWAIT);
3635 
3636 	if (io != NULL) {
3637 		io->io_hdr.io_type = io_type;
3638 		io->io_hdr.targ_port = targ_port;
3639 		/*
3640 		 * XXX KDM this needs to change/go away.  We need to move
3641 		 * to a preallocated pool of ctl_scsiio structures.
3642 		 */
3643 		io->io_hdr.nexus.targ_target.id = targ_target;
3644 		io->io_hdr.nexus.targ_lun = targ_lun;
3645 	}
3646 
3647 	return (io);
3648 }
3649 
3650 void
3651 ctl_kfree_io(union ctl_io *io)
3652 {
3653 	free(io, M_CTL);
3654 }
3655 #endif /* unused */
3656 
3657 /*
3658  * ctl_softc, pool_type, total_ctl_io are passed in.
3659  * npool is passed out.
3660  */
3661 int
3662 ctl_pool_create(struct ctl_softc *ctl_softc, ctl_pool_type pool_type,
3663 		uint32_t total_ctl_io, struct ctl_io_pool **npool)
3664 {
3665 	uint32_t i;
3666 	union ctl_io *cur_io, *next_io;
3667 	struct ctl_io_pool *pool;
3668 	int retval;
3669 
3670 	retval = 0;
3671 
3672 	pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3673 					    M_NOWAIT | M_ZERO);
3674 	if (pool == NULL) {
3675 		retval = ENOMEM;
3676 		goto bailout;
3677 	}
3678 
3679 	pool->type = pool_type;
3680 	pool->ctl_softc = ctl_softc;
3681 
3682 	mtx_lock(&ctl_softc->pool_lock);
3683 	pool->id = ctl_softc->cur_pool_id++;
3684 	mtx_unlock(&ctl_softc->pool_lock);
3685 
3686 	pool->flags = CTL_POOL_FLAG_NONE;
3687 	pool->refcount = 1;		/* Reference for validity. */
3688 	STAILQ_INIT(&pool->free_queue);
3689 
3690 	/*
3691 	 * XXX KDM other options here:
3692 	 * - allocate a page at a time
3693 	 * - allocate one big chunk of memory.
3694 	 * Page allocation might work well, but would take a little more
3695 	 * tracking.
3696 	 */
3697 	for (i = 0; i < total_ctl_io; i++) {
3698 		cur_io = (union ctl_io *)malloc(sizeof(*cur_io), M_CTLIO,
3699 						M_NOWAIT);
3700 		if (cur_io == NULL) {
3701 			retval = ENOMEM;
3702 			break;
3703 		}
3704 		cur_io->io_hdr.pool = pool;
3705 		STAILQ_INSERT_TAIL(&pool->free_queue, &cur_io->io_hdr, links);
3706 		pool->total_ctl_io++;
3707 		pool->free_ctl_io++;
3708 	}
3709 
3710 	if (retval != 0) {
3711 		for (cur_io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3712 		     cur_io != NULL; cur_io = next_io) {
3713 			next_io = (union ctl_io *)STAILQ_NEXT(&cur_io->io_hdr,
3714 							      links);
3715 			STAILQ_REMOVE(&pool->free_queue, &cur_io->io_hdr,
3716 				      ctl_io_hdr, links);
3717 			free(cur_io, M_CTLIO);
3718 		}
3719 
3720 		free(pool, M_CTL);
3721 		goto bailout;
3722 	}
3723 	mtx_lock(&ctl_softc->pool_lock);
3724 	ctl_softc->num_pools++;
3725 	STAILQ_INSERT_TAIL(&ctl_softc->io_pools, pool, links);
3726 	/*
3727 	 * Increment our usage count if this is an external consumer, so we
3728 	 * can't get unloaded until the external consumer (most likely a
3729 	 * FETD) unloads and frees his pool.
3730 	 *
3731 	 * XXX KDM will this increment the caller's module use count, or
3732 	 * mine?
3733 	 */
3734 #if 0
3735 	if ((pool_type != CTL_POOL_EMERGENCY)
3736 	 && (pool_type != CTL_POOL_INTERNAL)
3737 	 && (pool_type != CTL_POOL_4OTHERSC))
3738 		MOD_INC_USE_COUNT;
3739 #endif
3740 
3741 	mtx_unlock(&ctl_softc->pool_lock);
3742 
3743 	*npool = pool;
3744 
3745 bailout:
3746 
3747 	return (retval);
3748 }
3749 
3750 static int
3751 ctl_pool_acquire(struct ctl_io_pool *pool)
3752 {
3753 
3754 	mtx_assert(&pool->ctl_softc->pool_lock, MA_OWNED);
3755 
3756 	if (pool->flags & CTL_POOL_FLAG_INVALID)
3757 		return (EINVAL);
3758 
3759 	pool->refcount++;
3760 
3761 	return (0);
3762 }
3763 
3764 static void
3765 ctl_pool_release(struct ctl_io_pool *pool)
3766 {
3767 	struct ctl_softc *ctl_softc = pool->ctl_softc;
3768 	union ctl_io *io;
3769 
3770 	mtx_assert(&ctl_softc->pool_lock, MA_OWNED);
3771 
3772 	if (--pool->refcount != 0)
3773 		return;
3774 
3775 	while ((io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue)) != NULL) {
3776 		STAILQ_REMOVE(&pool->free_queue, &io->io_hdr, ctl_io_hdr,
3777 			      links);
3778 		free(io, M_CTLIO);
3779 	}
3780 
3781 	STAILQ_REMOVE(&ctl_softc->io_pools, pool, ctl_io_pool, links);
3782 	ctl_softc->num_pools--;
3783 
3784 	/*
3785 	 * XXX KDM will this decrement the caller's usage count or mine?
3786 	 */
3787 #if 0
3788 	if ((pool->type != CTL_POOL_EMERGENCY)
3789 	 && (pool->type != CTL_POOL_INTERNAL)
3790 	 && (pool->type != CTL_POOL_4OTHERSC))
3791 		MOD_DEC_USE_COUNT;
3792 #endif
3793 
3794 	free(pool, M_CTL);
3795 }
3796 
3797 void
3798 ctl_pool_free(struct ctl_io_pool *pool)
3799 {
3800 	struct ctl_softc *ctl_softc;
3801 
3802 	if (pool == NULL)
3803 		return;
3804 
3805 	ctl_softc = pool->ctl_softc;
3806 	mtx_lock(&ctl_softc->pool_lock);
3807 	pool->flags |= CTL_POOL_FLAG_INVALID;
3808 	ctl_pool_release(pool);
3809 	mtx_unlock(&ctl_softc->pool_lock);
3810 }
3811 
3812 /*
3813  * This routine does not block (except for spinlocks of course).
3814  * It tries to allocate a ctl_io union from the caller's pool as quickly as
3815  * possible.
3816  */
3817 union ctl_io *
3818 ctl_alloc_io(void *pool_ref)
3819 {
3820 	union ctl_io *io;
3821 	struct ctl_softc *ctl_softc;
3822 	struct ctl_io_pool *pool, *npool;
3823 	struct ctl_io_pool *emergency_pool;
3824 
3825 	pool = (struct ctl_io_pool *)pool_ref;
3826 
3827 	if (pool == NULL) {
3828 		printf("%s: pool is NULL\n", __func__);
3829 		return (NULL);
3830 	}
3831 
3832 	emergency_pool = NULL;
3833 
3834 	ctl_softc = pool->ctl_softc;
3835 
3836 	mtx_lock(&ctl_softc->pool_lock);
3837 	/*
3838 	 * First, try to get the io structure from the user's pool.
3839 	 */
3840 	if (ctl_pool_acquire(pool) == 0) {
3841 		io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3842 		if (io != NULL) {
3843 			STAILQ_REMOVE_HEAD(&pool->free_queue, links);
3844 			pool->total_allocated++;
3845 			pool->free_ctl_io--;
3846 			mtx_unlock(&ctl_softc->pool_lock);
3847 			return (io);
3848 		} else
3849 			ctl_pool_release(pool);
3850 	}
3851 	/*
3852 	 * If he doesn't have any io structures left, search for an
3853 	 * emergency pool and grab one from there.
3854 	 */
3855 	STAILQ_FOREACH(npool, &ctl_softc->io_pools, links) {
3856 		if (npool->type != CTL_POOL_EMERGENCY)
3857 			continue;
3858 
3859 		if (ctl_pool_acquire(npool) != 0)
3860 			continue;
3861 
3862 		emergency_pool = npool;
3863 
3864 		io = (union ctl_io *)STAILQ_FIRST(&npool->free_queue);
3865 		if (io != NULL) {
3866 			STAILQ_REMOVE_HEAD(&npool->free_queue, links);
3867 			npool->total_allocated++;
3868 			npool->free_ctl_io--;
3869 			mtx_unlock(&ctl_softc->pool_lock);
3870 			return (io);
3871 		} else
3872 			ctl_pool_release(npool);
3873 	}
3874 
3875 	/* Drop the spinlock before we malloc */
3876 	mtx_unlock(&ctl_softc->pool_lock);
3877 
3878 	/*
3879 	 * The emergency pool (if it exists) didn't have one, so try an
3880 	 * atomic (i.e. nonblocking) malloc and see if we get lucky.
3881 	 */
3882 	io = (union ctl_io *)malloc(sizeof(*io), M_CTLIO, M_NOWAIT);
3883 	if (io != NULL) {
3884 		/*
3885 		 * If the emergency pool exists but is empty, add this
3886 		 * ctl_io to its list when it gets freed.
3887 		 */
3888 		if (emergency_pool != NULL) {
3889 			mtx_lock(&ctl_softc->pool_lock);
3890 			if (ctl_pool_acquire(emergency_pool) == 0) {
3891 				io->io_hdr.pool = emergency_pool;
3892 				emergency_pool->total_ctl_io++;
3893 				/*
3894 				 * Need to bump this, otherwise
3895 				 * total_allocated and total_freed won't
3896 				 * match when we no longer have anything
3897 				 * outstanding.
3898 				 */
3899 				emergency_pool->total_allocated++;
3900 			}
3901 			mtx_unlock(&ctl_softc->pool_lock);
3902 		} else
3903 			io->io_hdr.pool = NULL;
3904 	}
3905 
3906 	return (io);
3907 }
3908 
3909 void
3910 ctl_free_io(union ctl_io *io)
3911 {
3912 	if (io == NULL)
3913 		return;
3914 
3915 	/*
3916 	 * If this ctl_io has a pool, return it to that pool.
3917 	 */
3918 	if (io->io_hdr.pool != NULL) {
3919 		struct ctl_io_pool *pool;
3920 
3921 		pool = (struct ctl_io_pool *)io->io_hdr.pool;
3922 		mtx_lock(&pool->ctl_softc->pool_lock);
3923 		io->io_hdr.io_type = 0xff;
3924 		STAILQ_INSERT_TAIL(&pool->free_queue, &io->io_hdr, links);
3925 		pool->total_freed++;
3926 		pool->free_ctl_io++;
3927 		ctl_pool_release(pool);
3928 		mtx_unlock(&pool->ctl_softc->pool_lock);
3929 	} else {
3930 		/*
3931 		 * Otherwise, just free it.  We probably malloced it and
3932 		 * the emergency pool wasn't available.
3933 		 */
3934 		free(io, M_CTLIO);
3935 	}
3936 
3937 }
3938 
3939 void
3940 ctl_zero_io(union ctl_io *io)
3941 {
3942 	void *pool_ref;
3943 
3944 	if (io == NULL)
3945 		return;
3946 
3947 	/*
3948 	 * May need to preserve linked list pointers at some point too.
3949 	 */
3950 	pool_ref = io->io_hdr.pool;
3951 
3952 	memset(io, 0, sizeof(*io));
3953 
3954 	io->io_hdr.pool = pool_ref;
3955 }
3956 
3957 /*
3958  * This routine is currently used for internal copies of ctl_ios that need
3959  * to persist for some reason after we've already returned status to the
3960  * FETD.  (Thus the flag set.)
3961  *
3962  * XXX XXX
3963  * Note that this makes a blind copy of all fields in the ctl_io, except
3964  * for the pool reference.  This includes any memory that has been
3965  * allocated!  That memory will no longer be valid after done has been
3966  * called, so this would be VERY DANGEROUS for command that actually does
3967  * any reads or writes.  Right now (11/7/2005), this is only used for immediate
3968  * start and stop commands, which don't transfer any data, so this is not a
3969  * problem.  If it is used for anything else, the caller would also need to
3970  * allocate data buffer space and this routine would need to be modified to
3971  * copy the data buffer(s) as well.
3972  */
3973 void
3974 ctl_copy_io(union ctl_io *src, union ctl_io *dest)
3975 {
3976 	void *pool_ref;
3977 
3978 	if ((src == NULL)
3979 	 || (dest == NULL))
3980 		return;
3981 
3982 	/*
3983 	 * May need to preserve linked list pointers at some point too.
3984 	 */
3985 	pool_ref = dest->io_hdr.pool;
3986 
3987 	memcpy(dest, src, ctl_min(sizeof(*src), sizeof(*dest)));
3988 
3989 	dest->io_hdr.pool = pool_ref;
3990 	/*
3991 	 * We need to know that this is an internal copy, and doesn't need
3992 	 * to get passed back to the FETD that allocated it.
3993 	 */
3994 	dest->io_hdr.flags |= CTL_FLAG_INT_COPY;
3995 }
3996 
3997 #ifdef NEEDTOPORT
3998 static void
3999 ctl_update_power_subpage(struct copan_power_subpage *page)
4000 {
4001 	int num_luns, num_partitions, config_type;
4002 	struct ctl_softc *softc;
4003 	cs_BOOL_t aor_present, shelf_50pct_power;
4004 	cs_raidset_personality_t rs_type;
4005 	int max_active_luns;
4006 
4007 	softc = control_softc;
4008 
4009 	/* subtract out the processor LUN */
4010 	num_luns = softc->num_luns - 1;
4011 	/*
4012 	 * Default to 7 LUNs active, which was the only number we allowed
4013 	 * in the past.
4014 	 */
4015 	max_active_luns = 7;
4016 
4017 	num_partitions = config_GetRsPartitionInfo();
4018 	config_type = config_GetConfigType();
4019 	shelf_50pct_power = config_GetShelfPowerMode();
4020 	aor_present = config_IsAorRsPresent();
4021 
4022 	rs_type = ddb_GetRsRaidType(1);
4023 	if ((rs_type != CS_RAIDSET_PERSONALITY_RAID5)
4024 	 && (rs_type != CS_RAIDSET_PERSONALITY_RAID1)) {
4025 		EPRINT(0, "Unsupported RS type %d!", rs_type);
4026 	}
4027 
4028 
4029 	page->total_luns = num_luns;
4030 
4031 	switch (config_type) {
4032 	case 40:
4033 		/*
4034 		 * In a 40 drive configuration, it doesn't matter what DC
4035 		 * cards we have, whether we have AOR enabled or not,
4036 		 * partitioning or not, or what type of RAIDset we have.
4037 		 * In that scenario, we can power up every LUN we present
4038 		 * to the user.
4039 		 */
4040 		max_active_luns = num_luns;
4041 
4042 		break;
4043 	case 64:
4044 		if (shelf_50pct_power == CS_FALSE) {
4045 			/* 25% power */
4046 			if (aor_present == CS_TRUE) {
4047 				if (rs_type ==
4048 				     CS_RAIDSET_PERSONALITY_RAID5) {
4049 					max_active_luns = 7;
4050 				} else if (rs_type ==
4051 					 CS_RAIDSET_PERSONALITY_RAID1){
4052 					max_active_luns = 14;
4053 				} else {
4054 					/* XXX KDM now what?? */
4055 				}
4056 			} else {
4057 				if (rs_type ==
4058 				     CS_RAIDSET_PERSONALITY_RAID5) {
4059 					max_active_luns = 8;
4060 				} else if (rs_type ==
4061 					 CS_RAIDSET_PERSONALITY_RAID1){
4062 					max_active_luns = 16;
4063 				} else {
4064 					/* XXX KDM now what?? */
4065 				}
4066 			}
4067 		} else {
4068 			/* 50% power */
4069 			/*
4070 			 * With 50% power in a 64 drive configuration, we
4071 			 * can power all LUNs we present.
4072 			 */
4073 			max_active_luns = num_luns;
4074 		}
4075 		break;
4076 	case 112:
4077 		if (shelf_50pct_power == CS_FALSE) {
4078 			/* 25% power */
4079 			if (aor_present == CS_TRUE) {
4080 				if (rs_type ==
4081 				     CS_RAIDSET_PERSONALITY_RAID5) {
4082 					max_active_luns = 7;
4083 				} else if (rs_type ==
4084 					 CS_RAIDSET_PERSONALITY_RAID1){
4085 					max_active_luns = 14;
4086 				} else {
4087 					/* XXX KDM now what?? */
4088 				}
4089 			} else {
4090 				if (rs_type ==
4091 				     CS_RAIDSET_PERSONALITY_RAID5) {
4092 					max_active_luns = 8;
4093 				} else if (rs_type ==
4094 					 CS_RAIDSET_PERSONALITY_RAID1){
4095 					max_active_luns = 16;
4096 				} else {
4097 					/* XXX KDM now what?? */
4098 				}
4099 			}
4100 		} else {
4101 			/* 50% power */
4102 			if (aor_present == CS_TRUE) {
4103 				if (rs_type ==
4104 				     CS_RAIDSET_PERSONALITY_RAID5) {
4105 					max_active_luns = 14;
4106 				} else if (rs_type ==
4107 					 CS_RAIDSET_PERSONALITY_RAID1){
4108 					/*
4109 					 * We're assuming here that disk
4110 					 * caching is enabled, and so we're
4111 					 * able to power up half of each
4112 					 * LUN, and cache all writes.
4113 					 */
4114 					max_active_luns = num_luns;
4115 				} else {
4116 					/* XXX KDM now what?? */
4117 				}
4118 			} else {
4119 				if (rs_type ==
4120 				     CS_RAIDSET_PERSONALITY_RAID5) {
4121 					max_active_luns = 15;
4122 				} else if (rs_type ==
4123 					 CS_RAIDSET_PERSONALITY_RAID1){
4124 					max_active_luns = 30;
4125 				} else {
4126 					/* XXX KDM now what?? */
4127 				}
4128 			}
4129 		}
4130 		break;
4131 	default:
4132 		/*
4133 		 * In this case, we have an unknown configuration, so we
4134 		 * just use the default from above.
4135 		 */
4136 		break;
4137 	}
4138 
4139 	page->max_active_luns = max_active_luns;
4140 #if 0
4141 	printk("%s: total_luns = %d, max_active_luns = %d\n", __func__,
4142 	       page->total_luns, page->max_active_luns);
4143 #endif
4144 }
4145 #endif /* NEEDTOPORT */
4146 
4147 /*
4148  * This routine could be used in the future to load default and/or saved
4149  * mode page parameters for a particuar lun.
4150  */
4151 static int
4152 ctl_init_page_index(struct ctl_lun *lun)
4153 {
4154 	int i;
4155 	struct ctl_page_index *page_index;
4156 	struct ctl_softc *softc;
4157 	const char *value;
4158 
4159 	memcpy(&lun->mode_pages.index, page_index_template,
4160 	       sizeof(page_index_template));
4161 
4162 	softc = lun->ctl_softc;
4163 
4164 	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
4165 
4166 		page_index = &lun->mode_pages.index[i];
4167 		/*
4168 		 * If this is a disk-only mode page, there's no point in
4169 		 * setting it up.  For some pages, we have to have some
4170 		 * basic information about the disk in order to calculate the
4171 		 * mode page data.
4172 		 */
4173 		if ((lun->be_lun->lun_type != T_DIRECT)
4174 		 && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
4175 			continue;
4176 
4177 		switch (page_index->page_code & SMPH_PC_MASK) {
4178 		case SMS_FORMAT_DEVICE_PAGE: {
4179 			struct scsi_format_page *format_page;
4180 
4181 			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4182 				panic("subpage is incorrect!");
4183 
4184 			/*
4185 			 * Sectors per track are set above.  Bytes per
4186 			 * sector need to be set here on a per-LUN basis.
4187 			 */
4188 			memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
4189 			       &format_page_default,
4190 			       sizeof(format_page_default));
4191 			memcpy(&lun->mode_pages.format_page[
4192 			       CTL_PAGE_CHANGEABLE], &format_page_changeable,
4193 			       sizeof(format_page_changeable));
4194 			memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
4195 			       &format_page_default,
4196 			       sizeof(format_page_default));
4197 			memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
4198 			       &format_page_default,
4199 			       sizeof(format_page_default));
4200 
4201 			format_page = &lun->mode_pages.format_page[
4202 				CTL_PAGE_CURRENT];
4203 			scsi_ulto2b(lun->be_lun->blocksize,
4204 				    format_page->bytes_per_sector);
4205 
4206 			format_page = &lun->mode_pages.format_page[
4207 				CTL_PAGE_DEFAULT];
4208 			scsi_ulto2b(lun->be_lun->blocksize,
4209 				    format_page->bytes_per_sector);
4210 
4211 			format_page = &lun->mode_pages.format_page[
4212 				CTL_PAGE_SAVED];
4213 			scsi_ulto2b(lun->be_lun->blocksize,
4214 				    format_page->bytes_per_sector);
4215 
4216 			page_index->page_data =
4217 				(uint8_t *)lun->mode_pages.format_page;
4218 			break;
4219 		}
4220 		case SMS_RIGID_DISK_PAGE: {
4221 			struct scsi_rigid_disk_page *rigid_disk_page;
4222 			uint32_t sectors_per_cylinder;
4223 			uint64_t cylinders;
4224 #ifndef	__XSCALE__
4225 			int shift;
4226 #endif /* !__XSCALE__ */
4227 
4228 			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4229 				panic("invalid subpage value %d",
4230 				      page_index->subpage);
4231 
4232 			/*
4233 			 * Rotation rate and sectors per track are set
4234 			 * above.  We calculate the cylinders here based on
4235 			 * capacity.  Due to the number of heads and
4236 			 * sectors per track we're using, smaller arrays
4237 			 * may turn out to have 0 cylinders.  Linux and
4238 			 * FreeBSD don't pay attention to these mode pages
4239 			 * to figure out capacity, but Solaris does.  It
4240 			 * seems to deal with 0 cylinders just fine, and
4241 			 * works out a fake geometry based on the capacity.
4242 			 */
4243 			memcpy(&lun->mode_pages.rigid_disk_page[
4244 			       CTL_PAGE_CURRENT], &rigid_disk_page_default,
4245 			       sizeof(rigid_disk_page_default));
4246 			memcpy(&lun->mode_pages.rigid_disk_page[
4247 			       CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
4248 			       sizeof(rigid_disk_page_changeable));
4249 			memcpy(&lun->mode_pages.rigid_disk_page[
4250 			       CTL_PAGE_DEFAULT], &rigid_disk_page_default,
4251 			       sizeof(rigid_disk_page_default));
4252 			memcpy(&lun->mode_pages.rigid_disk_page[
4253 			       CTL_PAGE_SAVED], &rigid_disk_page_default,
4254 			       sizeof(rigid_disk_page_default));
4255 
4256 			sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
4257 				CTL_DEFAULT_HEADS;
4258 
4259 			/*
4260 			 * The divide method here will be more accurate,
4261 			 * probably, but results in floating point being
4262 			 * used in the kernel on i386 (__udivdi3()).  On the
4263 			 * XScale, though, __udivdi3() is implemented in
4264 			 * software.
4265 			 *
4266 			 * The shift method for cylinder calculation is
4267 			 * accurate if sectors_per_cylinder is a power of
4268 			 * 2.  Otherwise it might be slightly off -- you
4269 			 * might have a bit of a truncation problem.
4270 			 */
4271 #ifdef	__XSCALE__
4272 			cylinders = (lun->be_lun->maxlba + 1) /
4273 				sectors_per_cylinder;
4274 #else
4275 			for (shift = 31; shift > 0; shift--) {
4276 				if (sectors_per_cylinder & (1 << shift))
4277 					break;
4278 			}
4279 			cylinders = (lun->be_lun->maxlba + 1) >> shift;
4280 #endif
4281 
4282 			/*
4283 			 * We've basically got 3 bytes, or 24 bits for the
4284 			 * cylinder size in the mode page.  If we're over,
4285 			 * just round down to 2^24.
4286 			 */
4287 			if (cylinders > 0xffffff)
4288 				cylinders = 0xffffff;
4289 
4290 			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4291 				CTL_PAGE_CURRENT];
4292 			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4293 
4294 			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4295 				CTL_PAGE_DEFAULT];
4296 			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4297 
4298 			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4299 				CTL_PAGE_SAVED];
4300 			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4301 
4302 			page_index->page_data =
4303 				(uint8_t *)lun->mode_pages.rigid_disk_page;
4304 			break;
4305 		}
4306 		case SMS_CACHING_PAGE: {
4307 			struct scsi_caching_page *caching_page;
4308 
4309 			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4310 				panic("invalid subpage value %d",
4311 				      page_index->subpage);
4312 			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4313 			       &caching_page_default,
4314 			       sizeof(caching_page_default));
4315 			memcpy(&lun->mode_pages.caching_page[
4316 			       CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4317 			       sizeof(caching_page_changeable));
4318 			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4319 			       &caching_page_default,
4320 			       sizeof(caching_page_default));
4321 			caching_page = &lun->mode_pages.caching_page[
4322 			    CTL_PAGE_SAVED];
4323 			value = ctl_get_opt(&lun->be_lun->options, "writecache");
4324 			if (value != NULL && strcmp(value, "off") == 0)
4325 				caching_page->flags1 &= ~SCP_WCE;
4326 			value = ctl_get_opt(&lun->be_lun->options, "readcache");
4327 			if (value != NULL && strcmp(value, "off") == 0)
4328 				caching_page->flags1 |= SCP_RCD;
4329 			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4330 			       &lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4331 			       sizeof(caching_page_default));
4332 			page_index->page_data =
4333 				(uint8_t *)lun->mode_pages.caching_page;
4334 			break;
4335 		}
4336 		case SMS_CONTROL_MODE_PAGE: {
4337 			struct scsi_control_page *control_page;
4338 
4339 			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4340 				panic("invalid subpage value %d",
4341 				      page_index->subpage);
4342 
4343 			memcpy(&lun->mode_pages.control_page[CTL_PAGE_DEFAULT],
4344 			       &control_page_default,
4345 			       sizeof(control_page_default));
4346 			memcpy(&lun->mode_pages.control_page[
4347 			       CTL_PAGE_CHANGEABLE], &control_page_changeable,
4348 			       sizeof(control_page_changeable));
4349 			memcpy(&lun->mode_pages.control_page[CTL_PAGE_SAVED],
4350 			       &control_page_default,
4351 			       sizeof(control_page_default));
4352 			control_page = &lun->mode_pages.control_page[
4353 			    CTL_PAGE_SAVED];
4354 			value = ctl_get_opt(&lun->be_lun->options, "reordering");
4355 			if (value != NULL && strcmp(value, "unrestricted") == 0) {
4356 				control_page->queue_flags &= ~SCP_QUEUE_ALG_MASK;
4357 				control_page->queue_flags |= SCP_QUEUE_ALG_UNRESTRICTED;
4358 			}
4359 			memcpy(&lun->mode_pages.control_page[CTL_PAGE_CURRENT],
4360 			       &lun->mode_pages.control_page[CTL_PAGE_SAVED],
4361 			       sizeof(control_page_default));
4362 			page_index->page_data =
4363 				(uint8_t *)lun->mode_pages.control_page;
4364 			break;
4365 
4366 		}
4367 		case SMS_VENDOR_SPECIFIC_PAGE:{
4368 			switch (page_index->subpage) {
4369 			case PWR_SUBPAGE_CODE: {
4370 				struct copan_power_subpage *current_page,
4371 							   *saved_page;
4372 
4373 				memcpy(&lun->mode_pages.power_subpage[
4374 				       CTL_PAGE_CURRENT],
4375 				       &power_page_default,
4376 				       sizeof(power_page_default));
4377 				memcpy(&lun->mode_pages.power_subpage[
4378 				       CTL_PAGE_CHANGEABLE],
4379 				       &power_page_changeable,
4380 				       sizeof(power_page_changeable));
4381 				memcpy(&lun->mode_pages.power_subpage[
4382 				       CTL_PAGE_DEFAULT],
4383 				       &power_page_default,
4384 				       sizeof(power_page_default));
4385 				memcpy(&lun->mode_pages.power_subpage[
4386 				       CTL_PAGE_SAVED],
4387 				       &power_page_default,
4388 				       sizeof(power_page_default));
4389 				page_index->page_data =
4390 				    (uint8_t *)lun->mode_pages.power_subpage;
4391 
4392 				current_page = (struct copan_power_subpage *)
4393 					(page_index->page_data +
4394 					 (page_index->page_len *
4395 					  CTL_PAGE_CURRENT));
4396 			        saved_page = (struct copan_power_subpage *)
4397 				        (page_index->page_data +
4398 					 (page_index->page_len *
4399 					  CTL_PAGE_SAVED));
4400 				break;
4401 			}
4402 			case APS_SUBPAGE_CODE: {
4403 				struct copan_aps_subpage *current_page,
4404 							 *saved_page;
4405 
4406 				// This gets set multiple times but
4407 				// it should always be the same. It's
4408 				// only done during init so who cares.
4409 				index_to_aps_page = i;
4410 
4411 				memcpy(&lun->mode_pages.aps_subpage[
4412 				       CTL_PAGE_CURRENT],
4413 				       &aps_page_default,
4414 				       sizeof(aps_page_default));
4415 				memcpy(&lun->mode_pages.aps_subpage[
4416 				       CTL_PAGE_CHANGEABLE],
4417 				       &aps_page_changeable,
4418 				       sizeof(aps_page_changeable));
4419 				memcpy(&lun->mode_pages.aps_subpage[
4420 				       CTL_PAGE_DEFAULT],
4421 				       &aps_page_default,
4422 				       sizeof(aps_page_default));
4423 				memcpy(&lun->mode_pages.aps_subpage[
4424 				       CTL_PAGE_SAVED],
4425 				       &aps_page_default,
4426 				       sizeof(aps_page_default));
4427 				page_index->page_data =
4428 					(uint8_t *)lun->mode_pages.aps_subpage;
4429 
4430 				current_page = (struct copan_aps_subpage *)
4431 					(page_index->page_data +
4432 					 (page_index->page_len *
4433 					  CTL_PAGE_CURRENT));
4434 				saved_page = (struct copan_aps_subpage *)
4435 					(page_index->page_data +
4436 					 (page_index->page_len *
4437 					  CTL_PAGE_SAVED));
4438 				break;
4439 			}
4440 			case DBGCNF_SUBPAGE_CODE: {
4441 				struct copan_debugconf_subpage *current_page,
4442 							       *saved_page;
4443 
4444 				memcpy(&lun->mode_pages.debugconf_subpage[
4445 				       CTL_PAGE_CURRENT],
4446 				       &debugconf_page_default,
4447 				       sizeof(debugconf_page_default));
4448 				memcpy(&lun->mode_pages.debugconf_subpage[
4449 				       CTL_PAGE_CHANGEABLE],
4450 				       &debugconf_page_changeable,
4451 				       sizeof(debugconf_page_changeable));
4452 				memcpy(&lun->mode_pages.debugconf_subpage[
4453 				       CTL_PAGE_DEFAULT],
4454 				       &debugconf_page_default,
4455 				       sizeof(debugconf_page_default));
4456 				memcpy(&lun->mode_pages.debugconf_subpage[
4457 				       CTL_PAGE_SAVED],
4458 				       &debugconf_page_default,
4459 				       sizeof(debugconf_page_default));
4460 				page_index->page_data =
4461 					(uint8_t *)lun->mode_pages.debugconf_subpage;
4462 
4463 				current_page = (struct copan_debugconf_subpage *)
4464 					(page_index->page_data +
4465 					 (page_index->page_len *
4466 					  CTL_PAGE_CURRENT));
4467 				saved_page = (struct copan_debugconf_subpage *)
4468 					(page_index->page_data +
4469 					 (page_index->page_len *
4470 					  CTL_PAGE_SAVED));
4471 				break;
4472 			}
4473 			default:
4474 				panic("invalid subpage value %d",
4475 				      page_index->subpage);
4476 				break;
4477 			}
4478    			break;
4479 		}
4480 		default:
4481 			panic("invalid page value %d",
4482 			      page_index->page_code & SMPH_PC_MASK);
4483 			break;
4484     	}
4485 	}
4486 
4487 	return (CTL_RETVAL_COMPLETE);
4488 }
4489 
4490 /*
4491  * LUN allocation.
4492  *
4493  * Requirements:
4494  * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4495  *   wants us to allocate the LUN and he can block.
4496  * - ctl_softc is always set
4497  * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4498  *
4499  * Returns 0 for success, non-zero (errno) for failure.
4500  */
4501 static int
4502 ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4503 	      struct ctl_be_lun *const be_lun, struct ctl_id target_id)
4504 {
4505 	struct ctl_lun *nlun, *lun;
4506 	struct ctl_port *port;
4507 	struct scsi_vpd_id_descriptor *desc;
4508 	struct scsi_vpd_id_t10 *t10id;
4509 	const char *eui, *naa, *scsiname, *vendor, *value;
4510 	int lun_number, i, lun_malloced;
4511 	int devidlen, idlen1, idlen2 = 0, len;
4512 
4513 	if (be_lun == NULL)
4514 		return (EINVAL);
4515 
4516 	/*
4517 	 * We currently only support Direct Access or Processor LUN types.
4518 	 */
4519 	switch (be_lun->lun_type) {
4520 	case T_DIRECT:
4521 		break;
4522 	case T_PROCESSOR:
4523 		break;
4524 	case T_SEQUENTIAL:
4525 	case T_CHANGER:
4526 	default:
4527 		be_lun->lun_config_status(be_lun->be_lun,
4528 					  CTL_LUN_CONFIG_FAILURE);
4529 		break;
4530 	}
4531 	if (ctl_lun == NULL) {
4532 		lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4533 		lun_malloced = 1;
4534 	} else {
4535 		lun_malloced = 0;
4536 		lun = ctl_lun;
4537 	}
4538 
4539 	memset(lun, 0, sizeof(*lun));
4540 	if (lun_malloced)
4541 		lun->flags = CTL_LUN_MALLOCED;
4542 
4543 	/* Generate LUN ID. */
4544 	devidlen = max(CTL_DEVID_MIN_LEN,
4545 	    strnlen(be_lun->device_id, CTL_DEVID_LEN));
4546 	idlen1 = sizeof(*t10id) + devidlen;
4547 	len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4548 	scsiname = ctl_get_opt(&be_lun->options, "scsiname");
4549 	if (scsiname != NULL) {
4550 		idlen2 = roundup2(strlen(scsiname) + 1, 4);
4551 		len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4552 	}
4553 	eui = ctl_get_opt(&be_lun->options, "eui");
4554 	if (eui != NULL) {
4555 		len += sizeof(struct scsi_vpd_id_descriptor) + 8;
4556 	}
4557 	naa = ctl_get_opt(&be_lun->options, "naa");
4558 	if (naa != NULL) {
4559 		len += sizeof(struct scsi_vpd_id_descriptor) + 8;
4560 	}
4561 	lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4562 	    M_CTL, M_WAITOK | M_ZERO);
4563 	lun->lun_devid->len = len;
4564 	desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4565 	desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4566 	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4567 	desc->length = idlen1;
4568 	t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4569 	memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4570 	if ((vendor = ctl_get_opt(&be_lun->options, "vendor")) == NULL) {
4571 		strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4572 	} else {
4573 		strncpy(t10id->vendor, vendor,
4574 		    min(sizeof(t10id->vendor), strlen(vendor)));
4575 	}
4576 	strncpy((char *)t10id->vendor_spec_id,
4577 	    (char *)be_lun->device_id, devidlen);
4578 	if (scsiname != NULL) {
4579 		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4580 		    desc->length);
4581 		desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4582 		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4583 		    SVPD_ID_TYPE_SCSI_NAME;
4584 		desc->length = idlen2;
4585 		strlcpy(desc->identifier, scsiname, idlen2);
4586 	}
4587 	if (eui != NULL) {
4588 		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4589 		    desc->length);
4590 		desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4591 		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4592 		    SVPD_ID_TYPE_EUI64;
4593 		desc->length = 8;
4594 		scsi_u64to8b(strtouq(eui, NULL, 0), desc->identifier);
4595 	}
4596 	if (naa != NULL) {
4597 		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4598 		    desc->length);
4599 		desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4600 		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4601 		    SVPD_ID_TYPE_NAA;
4602 		desc->length = 8;
4603 		scsi_u64to8b(strtouq(naa, NULL, 0), desc->identifier);
4604 	}
4605 
4606 	mtx_lock(&ctl_softc->ctl_lock);
4607 	/*
4608 	 * See if the caller requested a particular LUN number.  If so, see
4609 	 * if it is available.  Otherwise, allocate the first available LUN.
4610 	 */
4611 	if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4612 		if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4613 		 || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4614 			mtx_unlock(&ctl_softc->ctl_lock);
4615 			if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4616 				printf("ctl: requested LUN ID %d is higher "
4617 				       "than CTL_MAX_LUNS - 1 (%d)\n",
4618 				       be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4619 			} else {
4620 				/*
4621 				 * XXX KDM return an error, or just assign
4622 				 * another LUN ID in this case??
4623 				 */
4624 				printf("ctl: requested LUN ID %d is already "
4625 				       "in use\n", be_lun->req_lun_id);
4626 			}
4627 			if (lun->flags & CTL_LUN_MALLOCED)
4628 				free(lun, M_CTL);
4629 			be_lun->lun_config_status(be_lun->be_lun,
4630 						  CTL_LUN_CONFIG_FAILURE);
4631 			return (ENOSPC);
4632 		}
4633 		lun_number = be_lun->req_lun_id;
4634 	} else {
4635 		lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, CTL_MAX_LUNS);
4636 		if (lun_number == -1) {
4637 			mtx_unlock(&ctl_softc->ctl_lock);
4638 			printf("ctl: can't allocate LUN on target %ju, out of "
4639 			       "LUNs\n", (uintmax_t)target_id.id);
4640 			if (lun->flags & CTL_LUN_MALLOCED)
4641 				free(lun, M_CTL);
4642 			be_lun->lun_config_status(be_lun->be_lun,
4643 						  CTL_LUN_CONFIG_FAILURE);
4644 			return (ENOSPC);
4645 		}
4646 	}
4647 	ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4648 
4649 	mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4650 	lun->target = target_id;
4651 	lun->lun = lun_number;
4652 	lun->be_lun = be_lun;
4653 	/*
4654 	 * The processor LUN is always enabled.  Disk LUNs come on line
4655 	 * disabled, and must be enabled by the backend.
4656 	 */
4657 	lun->flags |= CTL_LUN_DISABLED;
4658 	lun->backend = be_lun->be;
4659 	be_lun->ctl_lun = lun;
4660 	be_lun->lun_id = lun_number;
4661 	atomic_add_int(&be_lun->be->num_luns, 1);
4662 	if (be_lun->flags & CTL_LUN_FLAG_OFFLINE)
4663 		lun->flags |= CTL_LUN_OFFLINE;
4664 
4665 	if (be_lun->flags & CTL_LUN_FLAG_POWERED_OFF)
4666 		lun->flags |= CTL_LUN_STOPPED;
4667 
4668 	if (be_lun->flags & CTL_LUN_FLAG_INOPERABLE)
4669 		lun->flags |= CTL_LUN_INOPERABLE;
4670 
4671 	if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4672 		lun->flags |= CTL_LUN_PRIMARY_SC;
4673 
4674 	value = ctl_get_opt(&be_lun->options, "readonly");
4675 	if (value != NULL && strcmp(value, "on") == 0)
4676 		lun->flags |= CTL_LUN_READONLY;
4677 
4678 	lun->ctl_softc = ctl_softc;
4679 	TAILQ_INIT(&lun->ooa_queue);
4680 	TAILQ_INIT(&lun->blocked_queue);
4681 	STAILQ_INIT(&lun->error_list);
4682 	ctl_tpc_lun_init(lun);
4683 
4684 	/*
4685 	 * Initialize the mode page index.
4686 	 */
4687 	ctl_init_page_index(lun);
4688 
4689 	/*
4690 	 * Set the poweron UA for all initiators on this LUN only.
4691 	 */
4692 	for (i = 0; i < CTL_MAX_INITIATORS; i++)
4693 		lun->pending_ua[i] = CTL_UA_POWERON;
4694 
4695 	/*
4696 	 * Now, before we insert this lun on the lun list, set the lun
4697 	 * inventory changed UA for all other luns.
4698 	 */
4699 	STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4700 		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4701 			nlun->pending_ua[i] |= CTL_UA_LUN_CHANGE;
4702 		}
4703 	}
4704 
4705 	STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4706 
4707 	ctl_softc->ctl_luns[lun_number] = lun;
4708 
4709 	ctl_softc->num_luns++;
4710 
4711 	/* Setup statistics gathering */
4712 	lun->stats.device_type = be_lun->lun_type;
4713 	lun->stats.lun_number = lun_number;
4714 	if (lun->stats.device_type == T_DIRECT)
4715 		lun->stats.blocksize = be_lun->blocksize;
4716 	else
4717 		lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4718 	for (i = 0;i < CTL_MAX_PORTS;i++)
4719 		lun->stats.ports[i].targ_port = i;
4720 
4721 	mtx_unlock(&ctl_softc->ctl_lock);
4722 
4723 	lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4724 
4725 	/*
4726 	 * Run through each registered FETD and bring it online if it isn't
4727 	 * already.  Enable the target ID if it hasn't been enabled, and
4728 	 * enable this particular LUN.
4729 	 */
4730 	STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4731 		int retval;
4732 
4733 		retval = port->lun_enable(port->targ_lun_arg, target_id,lun_number);
4734 		if (retval != 0) {
4735 			printf("ctl_alloc_lun: FETD %s port %d returned error "
4736 			       "%d for lun_enable on target %ju lun %d\n",
4737 			       port->port_name, port->targ_port, retval,
4738 			       (uintmax_t)target_id.id, lun_number);
4739 		} else
4740 			port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4741 	}
4742 	return (0);
4743 }
4744 
4745 /*
4746  * Delete a LUN.
4747  * Assumptions:
4748  * - LUN has already been marked invalid and any pending I/O has been taken
4749  *   care of.
4750  */
4751 static int
4752 ctl_free_lun(struct ctl_lun *lun)
4753 {
4754 	struct ctl_softc *softc;
4755 #if 0
4756 	struct ctl_port *port;
4757 #endif
4758 	struct ctl_lun *nlun;
4759 	int i;
4760 
4761 	softc = lun->ctl_softc;
4762 
4763 	mtx_assert(&softc->ctl_lock, MA_OWNED);
4764 
4765 	STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4766 
4767 	ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4768 
4769 	softc->ctl_luns[lun->lun] = NULL;
4770 
4771 	if (!TAILQ_EMPTY(&lun->ooa_queue))
4772 		panic("Freeing a LUN %p with outstanding I/O!!\n", lun);
4773 
4774 	softc->num_luns--;
4775 
4776 	/*
4777 	 * XXX KDM this scheme only works for a single target/multiple LUN
4778 	 * setup.  It needs to be revamped for a multiple target scheme.
4779 	 *
4780 	 * XXX KDM this results in port->lun_disable() getting called twice,
4781 	 * once when ctl_disable_lun() is called, and a second time here.
4782 	 * We really need to re-think the LUN disable semantics.  There
4783 	 * should probably be several steps/levels to LUN removal:
4784 	 *  - disable
4785 	 *  - invalidate
4786 	 *  - free
4787  	 *
4788 	 * Right now we only have a disable method when communicating to
4789 	 * the front end ports, at least for individual LUNs.
4790 	 */
4791 #if 0
4792 	STAILQ_FOREACH(port, &softc->port_list, links) {
4793 		int retval;
4794 
4795 		retval = port->lun_disable(port->targ_lun_arg, lun->target,
4796 					 lun->lun);
4797 		if (retval != 0) {
4798 			printf("ctl_free_lun: FETD %s port %d returned error "
4799 			       "%d for lun_disable on target %ju lun %jd\n",
4800 			       port->port_name, port->targ_port, retval,
4801 			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4802 		}
4803 
4804 		if (STAILQ_FIRST(&softc->lun_list) == NULL) {
4805 			port->status &= ~CTL_PORT_STATUS_LUN_ONLINE;
4806 
4807 			retval = port->targ_disable(port->targ_lun_arg,lun->target);
4808 			if (retval != 0) {
4809 				printf("ctl_free_lun: FETD %s port %d "
4810 				       "returned error %d for targ_disable on "
4811 				       "target %ju\n", port->port_name,
4812 				       port->targ_port, retval,
4813 				       (uintmax_t)lun->target.id);
4814 			} else
4815 				port->status &= ~CTL_PORT_STATUS_TARG_ONLINE;
4816 
4817 			if ((port->status & CTL_PORT_STATUS_TARG_ONLINE) != 0)
4818 				continue;
4819 
4820 #if 0
4821 			port->port_offline(port->onoff_arg);
4822 			port->status &= ~CTL_PORT_STATUS_ONLINE;
4823 #endif
4824 		}
4825 	}
4826 #endif
4827 
4828 	/*
4829 	 * Tell the backend to free resources, if this LUN has a backend.
4830 	 */
4831 	atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4832 	lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4833 
4834 	ctl_tpc_lun_shutdown(lun);
4835 	mtx_destroy(&lun->lun_lock);
4836 	free(lun->lun_devid, M_CTL);
4837 	if (lun->flags & CTL_LUN_MALLOCED)
4838 		free(lun, M_CTL);
4839 
4840 	STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4841 		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4842 			nlun->pending_ua[i] |= CTL_UA_LUN_CHANGE;
4843 		}
4844 	}
4845 
4846 	return (0);
4847 }
4848 
4849 static void
4850 ctl_create_lun(struct ctl_be_lun *be_lun)
4851 {
4852 	struct ctl_softc *ctl_softc;
4853 
4854 	ctl_softc = control_softc;
4855 
4856 	/*
4857 	 * ctl_alloc_lun() should handle all potential failure cases.
4858 	 */
4859 	ctl_alloc_lun(ctl_softc, NULL, be_lun, ctl_softc->target);
4860 }
4861 
4862 int
4863 ctl_add_lun(struct ctl_be_lun *be_lun)
4864 {
4865 	struct ctl_softc *ctl_softc = control_softc;
4866 
4867 	mtx_lock(&ctl_softc->ctl_lock);
4868 	STAILQ_INSERT_TAIL(&ctl_softc->pending_lun_queue, be_lun, links);
4869 	mtx_unlock(&ctl_softc->ctl_lock);
4870 	wakeup(&ctl_softc->pending_lun_queue);
4871 
4872 	return (0);
4873 }
4874 
4875 int
4876 ctl_enable_lun(struct ctl_be_lun *be_lun)
4877 {
4878 	struct ctl_softc *ctl_softc;
4879 	struct ctl_port *port, *nport;
4880 	struct ctl_lun *lun;
4881 	int retval;
4882 
4883 	ctl_softc = control_softc;
4884 
4885 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4886 
4887 	mtx_lock(&ctl_softc->ctl_lock);
4888 	mtx_lock(&lun->lun_lock);
4889 	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4890 		/*
4891 		 * eh?  Why did we get called if the LUN is already
4892 		 * enabled?
4893 		 */
4894 		mtx_unlock(&lun->lun_lock);
4895 		mtx_unlock(&ctl_softc->ctl_lock);
4896 		return (0);
4897 	}
4898 	lun->flags &= ~CTL_LUN_DISABLED;
4899 	mtx_unlock(&lun->lun_lock);
4900 
4901 	for (port = STAILQ_FIRST(&ctl_softc->port_list); port != NULL; port = nport) {
4902 		nport = STAILQ_NEXT(port, links);
4903 
4904 		/*
4905 		 * Drop the lock while we call the FETD's enable routine.
4906 		 * This can lead to a callback into CTL (at least in the
4907 		 * case of the internal initiator frontend.
4908 		 */
4909 		mtx_unlock(&ctl_softc->ctl_lock);
4910 		retval = port->lun_enable(port->targ_lun_arg, lun->target,lun->lun);
4911 		mtx_lock(&ctl_softc->ctl_lock);
4912 		if (retval != 0) {
4913 			printf("%s: FETD %s port %d returned error "
4914 			       "%d for lun_enable on target %ju lun %jd\n",
4915 			       __func__, port->port_name, port->targ_port, retval,
4916 			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4917 		}
4918 #if 0
4919 		 else {
4920             /* NOTE:  TODO:  why does lun enable affect port status? */
4921 			port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4922 		}
4923 #endif
4924 	}
4925 
4926 	mtx_unlock(&ctl_softc->ctl_lock);
4927 
4928 	return (0);
4929 }
4930 
4931 int
4932 ctl_disable_lun(struct ctl_be_lun *be_lun)
4933 {
4934 	struct ctl_softc *ctl_softc;
4935 	struct ctl_port *port;
4936 	struct ctl_lun *lun;
4937 	int retval;
4938 
4939 	ctl_softc = control_softc;
4940 
4941 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4942 
4943 	mtx_lock(&ctl_softc->ctl_lock);
4944 	mtx_lock(&lun->lun_lock);
4945 	if (lun->flags & CTL_LUN_DISABLED) {
4946 		mtx_unlock(&lun->lun_lock);
4947 		mtx_unlock(&ctl_softc->ctl_lock);
4948 		return (0);
4949 	}
4950 	lun->flags |= CTL_LUN_DISABLED;
4951 	mtx_unlock(&lun->lun_lock);
4952 
4953 	STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4954 		mtx_unlock(&ctl_softc->ctl_lock);
4955 		/*
4956 		 * Drop the lock before we call the frontend's disable
4957 		 * routine, to avoid lock order reversals.
4958 		 *
4959 		 * XXX KDM what happens if the frontend list changes while
4960 		 * we're traversing it?  It's unlikely, but should be handled.
4961 		 */
4962 		retval = port->lun_disable(port->targ_lun_arg, lun->target,
4963 					 lun->lun);
4964 		mtx_lock(&ctl_softc->ctl_lock);
4965 		if (retval != 0) {
4966 			printf("ctl_alloc_lun: FETD %s port %d returned error "
4967 			       "%d for lun_disable on target %ju lun %jd\n",
4968 			       port->port_name, port->targ_port, retval,
4969 			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4970 		}
4971 	}
4972 
4973 	mtx_unlock(&ctl_softc->ctl_lock);
4974 
4975 	return (0);
4976 }
4977 
4978 int
4979 ctl_start_lun(struct ctl_be_lun *be_lun)
4980 {
4981 	struct ctl_softc *ctl_softc;
4982 	struct ctl_lun *lun;
4983 
4984 	ctl_softc = control_softc;
4985 
4986 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4987 
4988 	mtx_lock(&lun->lun_lock);
4989 	lun->flags &= ~CTL_LUN_STOPPED;
4990 	mtx_unlock(&lun->lun_lock);
4991 
4992 	return (0);
4993 }
4994 
4995 int
4996 ctl_stop_lun(struct ctl_be_lun *be_lun)
4997 {
4998 	struct ctl_softc *ctl_softc;
4999 	struct ctl_lun *lun;
5000 
5001 	ctl_softc = control_softc;
5002 
5003 	lun = (struct ctl_lun *)be_lun->ctl_lun;
5004 
5005 	mtx_lock(&lun->lun_lock);
5006 	lun->flags |= CTL_LUN_STOPPED;
5007 	mtx_unlock(&lun->lun_lock);
5008 
5009 	return (0);
5010 }
5011 
5012 int
5013 ctl_lun_offline(struct ctl_be_lun *be_lun)
5014 {
5015 	struct ctl_softc *ctl_softc;
5016 	struct ctl_lun *lun;
5017 
5018 	ctl_softc = control_softc;
5019 
5020 	lun = (struct ctl_lun *)be_lun->ctl_lun;
5021 
5022 	mtx_lock(&lun->lun_lock);
5023 	lun->flags |= CTL_LUN_OFFLINE;
5024 	mtx_unlock(&lun->lun_lock);
5025 
5026 	return (0);
5027 }
5028 
5029 int
5030 ctl_lun_online(struct ctl_be_lun *be_lun)
5031 {
5032 	struct ctl_softc *ctl_softc;
5033 	struct ctl_lun *lun;
5034 
5035 	ctl_softc = control_softc;
5036 
5037 	lun = (struct ctl_lun *)be_lun->ctl_lun;
5038 
5039 	mtx_lock(&lun->lun_lock);
5040 	lun->flags &= ~CTL_LUN_OFFLINE;
5041 	mtx_unlock(&lun->lun_lock);
5042 
5043 	return (0);
5044 }
5045 
5046 int
5047 ctl_invalidate_lun(struct ctl_be_lun *be_lun)
5048 {
5049 	struct ctl_softc *ctl_softc;
5050 	struct ctl_lun *lun;
5051 
5052 	ctl_softc = control_softc;
5053 
5054 	lun = (struct ctl_lun *)be_lun->ctl_lun;
5055 
5056 	mtx_lock(&lun->lun_lock);
5057 
5058 	/*
5059 	 * The LUN needs to be disabled before it can be marked invalid.
5060 	 */
5061 	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
5062 		mtx_unlock(&lun->lun_lock);
5063 		return (-1);
5064 	}
5065 	/*
5066 	 * Mark the LUN invalid.
5067 	 */
5068 	lun->flags |= CTL_LUN_INVALID;
5069 
5070 	/*
5071 	 * If there is nothing in the OOA queue, go ahead and free the LUN.
5072 	 * If we have something in the OOA queue, we'll free it when the
5073 	 * last I/O completes.
5074 	 */
5075 	if (TAILQ_EMPTY(&lun->ooa_queue)) {
5076 		mtx_unlock(&lun->lun_lock);
5077 		mtx_lock(&ctl_softc->ctl_lock);
5078 		ctl_free_lun(lun);
5079 		mtx_unlock(&ctl_softc->ctl_lock);
5080 	} else
5081 		mtx_unlock(&lun->lun_lock);
5082 
5083 	return (0);
5084 }
5085 
5086 int
5087 ctl_lun_inoperable(struct ctl_be_lun *be_lun)
5088 {
5089 	struct ctl_softc *ctl_softc;
5090 	struct ctl_lun *lun;
5091 
5092 	ctl_softc = control_softc;
5093 	lun = (struct ctl_lun *)be_lun->ctl_lun;
5094 
5095 	mtx_lock(&lun->lun_lock);
5096 	lun->flags |= CTL_LUN_INOPERABLE;
5097 	mtx_unlock(&lun->lun_lock);
5098 
5099 	return (0);
5100 }
5101 
5102 int
5103 ctl_lun_operable(struct ctl_be_lun *be_lun)
5104 {
5105 	struct ctl_softc *ctl_softc;
5106 	struct ctl_lun *lun;
5107 
5108 	ctl_softc = control_softc;
5109 	lun = (struct ctl_lun *)be_lun->ctl_lun;
5110 
5111 	mtx_lock(&lun->lun_lock);
5112 	lun->flags &= ~CTL_LUN_INOPERABLE;
5113 	mtx_unlock(&lun->lun_lock);
5114 
5115 	return (0);
5116 }
5117 
5118 int
5119 ctl_lun_power_lock(struct ctl_be_lun *be_lun, struct ctl_nexus *nexus,
5120 		   int lock)
5121 {
5122 	struct ctl_softc *softc;
5123 	struct ctl_lun *lun;
5124 	struct copan_aps_subpage *current_sp;
5125 	struct ctl_page_index *page_index;
5126 	int i;
5127 
5128 	softc = control_softc;
5129 
5130 	mtx_lock(&softc->ctl_lock);
5131 
5132 	lun = (struct ctl_lun *)be_lun->ctl_lun;
5133 	mtx_lock(&lun->lun_lock);
5134 
5135 	page_index = NULL;
5136 	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
5137 		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
5138 		     APS_PAGE_CODE)
5139 			continue;
5140 
5141 		if (lun->mode_pages.index[i].subpage != APS_SUBPAGE_CODE)
5142 			continue;
5143 		page_index = &lun->mode_pages.index[i];
5144 	}
5145 
5146 	if (page_index == NULL) {
5147 		mtx_unlock(&lun->lun_lock);
5148 		mtx_unlock(&softc->ctl_lock);
5149 		printf("%s: APS subpage not found for lun %ju!\n", __func__,
5150 		       (uintmax_t)lun->lun);
5151 		return (1);
5152 	}
5153 #if 0
5154 	if ((softc->aps_locked_lun != 0)
5155 	 && (softc->aps_locked_lun != lun->lun)) {
5156 		printf("%s: attempt to lock LUN %llu when %llu is already "
5157 		       "locked\n");
5158 		mtx_unlock(&lun->lun_lock);
5159 		mtx_unlock(&softc->ctl_lock);
5160 		return (1);
5161 	}
5162 #endif
5163 
5164 	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
5165 		(page_index->page_len * CTL_PAGE_CURRENT));
5166 
5167 	if (lock != 0) {
5168 		current_sp->lock_active = APS_LOCK_ACTIVE;
5169 		softc->aps_locked_lun = lun->lun;
5170 	} else {
5171 		current_sp->lock_active = 0;
5172 		softc->aps_locked_lun = 0;
5173 	}
5174 
5175 
5176 	/*
5177 	 * If we're in HA mode, try to send the lock message to the other
5178 	 * side.
5179 	 */
5180 	if (ctl_is_single == 0) {
5181 		int isc_retval;
5182 		union ctl_ha_msg lock_msg;
5183 
5184 		lock_msg.hdr.nexus = *nexus;
5185 		lock_msg.hdr.msg_type = CTL_MSG_APS_LOCK;
5186 		if (lock != 0)
5187 			lock_msg.aps.lock_flag = 1;
5188 		else
5189 			lock_msg.aps.lock_flag = 0;
5190 		isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &lock_msg,
5191 					 sizeof(lock_msg), 0);
5192 		if (isc_retval > CTL_HA_STATUS_SUCCESS) {
5193 			printf("%s: APS (lock=%d) error returned from "
5194 			       "ctl_ha_msg_send: %d\n", __func__, lock, isc_retval);
5195 			mtx_unlock(&lun->lun_lock);
5196 			mtx_unlock(&softc->ctl_lock);
5197 			return (1);
5198 		}
5199 	}
5200 
5201 	mtx_unlock(&lun->lun_lock);
5202 	mtx_unlock(&softc->ctl_lock);
5203 
5204 	return (0);
5205 }
5206 
5207 void
5208 ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
5209 {
5210 	struct ctl_lun *lun;
5211 	struct ctl_softc *softc;
5212 	int i;
5213 
5214 	softc = control_softc;
5215 
5216 	lun = (struct ctl_lun *)be_lun->ctl_lun;
5217 
5218 	mtx_lock(&lun->lun_lock);
5219 
5220 	for (i = 0; i < CTL_MAX_INITIATORS; i++)
5221 		lun->pending_ua[i] |= CTL_UA_CAPACITY_CHANGED;
5222 
5223 	mtx_unlock(&lun->lun_lock);
5224 }
5225 
5226 /*
5227  * Backend "memory move is complete" callback for requests that never
5228  * make it down to say RAIDCore's configuration code.
5229  */
5230 int
5231 ctl_config_move_done(union ctl_io *io)
5232 {
5233 	int retval;
5234 
5235 	retval = CTL_RETVAL_COMPLETE;
5236 
5237 
5238 	CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
5239 	/*
5240 	 * XXX KDM this shouldn't happen, but what if it does?
5241 	 */
5242 	if (io->io_hdr.io_type != CTL_IO_SCSI)
5243 		panic("I/O type isn't CTL_IO_SCSI!");
5244 
5245 	if ((io->io_hdr.port_status == 0)
5246 	 && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
5247 	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE))
5248 		io->io_hdr.status = CTL_SUCCESS;
5249 	else if ((io->io_hdr.port_status != 0)
5250 	      && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
5251 	      && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)){
5252 		/*
5253 		 * For hardware error sense keys, the sense key
5254 		 * specific value is defined to be a retry count,
5255 		 * but we use it to pass back an internal FETD
5256 		 * error code.  XXX KDM  Hopefully the FETD is only
5257 		 * using 16 bits for an error code, since that's
5258 		 * all the space we have in the sks field.
5259 		 */
5260 		ctl_set_internal_failure(&io->scsiio,
5261 					 /*sks_valid*/ 1,
5262 					 /*retry_count*/
5263 					 io->io_hdr.port_status);
5264 		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5265 			free(io->scsiio.kern_data_ptr, M_CTL);
5266 		ctl_done(io);
5267 		goto bailout;
5268 	}
5269 
5270 	if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
5271 	 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
5272 	 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
5273 		/*
5274 		 * XXX KDM just assuming a single pointer here, and not a
5275 		 * S/G list.  If we start using S/G lists for config data,
5276 		 * we'll need to know how to clean them up here as well.
5277 		 */
5278 		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5279 			free(io->scsiio.kern_data_ptr, M_CTL);
5280 		/* Hopefully the user has already set the status... */
5281 		ctl_done(io);
5282 	} else {
5283 		/*
5284 		 * XXX KDM now we need to continue data movement.  Some
5285 		 * options:
5286 		 * - call ctl_scsiio() again?  We don't do this for data
5287 		 *   writes, because for those at least we know ahead of
5288 		 *   time where the write will go and how long it is.  For
5289 		 *   config writes, though, that information is largely
5290 		 *   contained within the write itself, thus we need to
5291 		 *   parse out the data again.
5292 		 *
5293 		 * - Call some other function once the data is in?
5294 		 */
5295 
5296 		/*
5297 		 * XXX KDM call ctl_scsiio() again for now, and check flag
5298 		 * bits to see whether we're allocated or not.
5299 		 */
5300 		retval = ctl_scsiio(&io->scsiio);
5301 	}
5302 bailout:
5303 	return (retval);
5304 }
5305 
5306 /*
5307  * This gets called by a backend driver when it is done with a
5308  * data_submit method.
5309  */
5310 void
5311 ctl_data_submit_done(union ctl_io *io)
5312 {
5313 	/*
5314 	 * If the IO_CONT flag is set, we need to call the supplied
5315 	 * function to continue processing the I/O, instead of completing
5316 	 * the I/O just yet.
5317 	 *
5318 	 * If there is an error, though, we don't want to keep processing.
5319 	 * Instead, just send status back to the initiator.
5320 	 */
5321 	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5322 	    (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5323 	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5324 	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5325 		io->scsiio.io_cont(io);
5326 		return;
5327 	}
5328 	ctl_done(io);
5329 }
5330 
5331 /*
5332  * This gets called by a backend driver when it is done with a
5333  * configuration write.
5334  */
5335 void
5336 ctl_config_write_done(union ctl_io *io)
5337 {
5338 	uint8_t *buf;
5339 
5340 	/*
5341 	 * If the IO_CONT flag is set, we need to call the supplied
5342 	 * function to continue processing the I/O, instead of completing
5343 	 * the I/O just yet.
5344 	 *
5345 	 * If there is an error, though, we don't want to keep processing.
5346 	 * Instead, just send status back to the initiator.
5347 	 */
5348 	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5349 	    (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5350 	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5351 	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5352 		io->scsiio.io_cont(io);
5353 		return;
5354 	}
5355 	/*
5356 	 * Since a configuration write can be done for commands that actually
5357 	 * have data allocated, like write buffer, and commands that have
5358 	 * no data, like start/stop unit, we need to check here.
5359 	 */
5360 	if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5361 		buf = io->scsiio.kern_data_ptr;
5362 	else
5363 		buf = NULL;
5364 	ctl_done(io);
5365 	if (buf)
5366 		free(buf, M_CTL);
5367 }
5368 
5369 /*
5370  * SCSI release command.
5371  */
5372 int
5373 ctl_scsi_release(struct ctl_scsiio *ctsio)
5374 {
5375 	int length, longid, thirdparty_id, resv_id;
5376 	struct ctl_softc *ctl_softc;
5377 	struct ctl_lun *lun;
5378 	uint32_t residx;
5379 
5380 	length = 0;
5381 	resv_id = 0;
5382 
5383 	CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5384 
5385 	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5386 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5387 	ctl_softc = control_softc;
5388 
5389 	switch (ctsio->cdb[0]) {
5390 	case RELEASE_10: {
5391 		struct scsi_release_10 *cdb;
5392 
5393 		cdb = (struct scsi_release_10 *)ctsio->cdb;
5394 
5395 		if (cdb->byte2 & SR10_LONGID)
5396 			longid = 1;
5397 		else
5398 			thirdparty_id = cdb->thirdparty_id;
5399 
5400 		resv_id = cdb->resv_id;
5401 		length = scsi_2btoul(cdb->length);
5402 		break;
5403 	}
5404 	}
5405 
5406 
5407 	/*
5408 	 * XXX KDM right now, we only support LUN reservation.  We don't
5409 	 * support 3rd party reservations, or extent reservations, which
5410 	 * might actually need the parameter list.  If we've gotten this
5411 	 * far, we've got a LUN reservation.  Anything else got kicked out
5412 	 * above.  So, according to SPC, ignore the length.
5413 	 */
5414 	length = 0;
5415 
5416 	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5417 	 && (length > 0)) {
5418 		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5419 		ctsio->kern_data_len = length;
5420 		ctsio->kern_total_len = length;
5421 		ctsio->kern_data_resid = 0;
5422 		ctsio->kern_rel_offset = 0;
5423 		ctsio->kern_sg_entries = 0;
5424 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5425 		ctsio->be_move_done = ctl_config_move_done;
5426 		ctl_datamove((union ctl_io *)ctsio);
5427 
5428 		return (CTL_RETVAL_COMPLETE);
5429 	}
5430 
5431 	if (length > 0)
5432 		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5433 
5434 	mtx_lock(&lun->lun_lock);
5435 
5436 	/*
5437 	 * According to SPC, it is not an error for an intiator to attempt
5438 	 * to release a reservation on a LUN that isn't reserved, or that
5439 	 * is reserved by another initiator.  The reservation can only be
5440 	 * released, though, by the initiator who made it or by one of
5441 	 * several reset type events.
5442 	 */
5443 	if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
5444 			lun->flags &= ~CTL_LUN_RESERVED;
5445 
5446 	mtx_unlock(&lun->lun_lock);
5447 
5448 	ctsio->scsi_status = SCSI_STATUS_OK;
5449 	ctsio->io_hdr.status = CTL_SUCCESS;
5450 
5451 	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5452 		free(ctsio->kern_data_ptr, M_CTL);
5453 		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5454 	}
5455 
5456 	ctl_done((union ctl_io *)ctsio);
5457 	return (CTL_RETVAL_COMPLETE);
5458 }
5459 
5460 int
5461 ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5462 {
5463 	int extent, thirdparty, longid;
5464 	int resv_id, length;
5465 	uint64_t thirdparty_id;
5466 	struct ctl_softc *ctl_softc;
5467 	struct ctl_lun *lun;
5468 	uint32_t residx;
5469 
5470 	extent = 0;
5471 	thirdparty = 0;
5472 	longid = 0;
5473 	resv_id = 0;
5474 	length = 0;
5475 	thirdparty_id = 0;
5476 
5477 	CTL_DEBUG_PRINT(("ctl_reserve\n"));
5478 
5479 	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5480 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5481 	ctl_softc = control_softc;
5482 
5483 	switch (ctsio->cdb[0]) {
5484 	case RESERVE_10: {
5485 		struct scsi_reserve_10 *cdb;
5486 
5487 		cdb = (struct scsi_reserve_10 *)ctsio->cdb;
5488 
5489 		if (cdb->byte2 & SR10_LONGID)
5490 			longid = 1;
5491 		else
5492 			thirdparty_id = cdb->thirdparty_id;
5493 
5494 		resv_id = cdb->resv_id;
5495 		length = scsi_2btoul(cdb->length);
5496 		break;
5497 	}
5498 	}
5499 
5500 	/*
5501 	 * XXX KDM right now, we only support LUN reservation.  We don't
5502 	 * support 3rd party reservations, or extent reservations, which
5503 	 * might actually need the parameter list.  If we've gotten this
5504 	 * far, we've got a LUN reservation.  Anything else got kicked out
5505 	 * above.  So, according to SPC, ignore the length.
5506 	 */
5507 	length = 0;
5508 
5509 	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5510 	 && (length > 0)) {
5511 		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5512 		ctsio->kern_data_len = length;
5513 		ctsio->kern_total_len = length;
5514 		ctsio->kern_data_resid = 0;
5515 		ctsio->kern_rel_offset = 0;
5516 		ctsio->kern_sg_entries = 0;
5517 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5518 		ctsio->be_move_done = ctl_config_move_done;
5519 		ctl_datamove((union ctl_io *)ctsio);
5520 
5521 		return (CTL_RETVAL_COMPLETE);
5522 	}
5523 
5524 	if (length > 0)
5525 		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5526 
5527 	mtx_lock(&lun->lun_lock);
5528 	if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx != residx)) {
5529 		ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
5530 		ctsio->io_hdr.status = CTL_SCSI_ERROR;
5531 		goto bailout;
5532 	}
5533 
5534 	lun->flags |= CTL_LUN_RESERVED;
5535 	lun->res_idx = residx;
5536 
5537 	ctsio->scsi_status = SCSI_STATUS_OK;
5538 	ctsio->io_hdr.status = CTL_SUCCESS;
5539 
5540 bailout:
5541 	mtx_unlock(&lun->lun_lock);
5542 
5543 	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5544 		free(ctsio->kern_data_ptr, M_CTL);
5545 		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5546 	}
5547 
5548 	ctl_done((union ctl_io *)ctsio);
5549 	return (CTL_RETVAL_COMPLETE);
5550 }
5551 
5552 int
5553 ctl_start_stop(struct ctl_scsiio *ctsio)
5554 {
5555 	struct scsi_start_stop_unit *cdb;
5556 	struct ctl_lun *lun;
5557 	struct ctl_softc *ctl_softc;
5558 	int retval;
5559 
5560 	CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5561 
5562 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5563 	ctl_softc = control_softc;
5564 	retval = 0;
5565 
5566 	cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5567 
5568 	/*
5569 	 * XXX KDM
5570 	 * We don't support the immediate bit on a stop unit.  In order to
5571 	 * do that, we would need to code up a way to know that a stop is
5572 	 * pending, and hold off any new commands until it completes, one
5573 	 * way or another.  Then we could accept or reject those commands
5574 	 * depending on its status.  We would almost need to do the reverse
5575 	 * of what we do below for an immediate start -- return the copy of
5576 	 * the ctl_io to the FETD with status to send to the host (and to
5577 	 * free the copy!) and then free the original I/O once the stop
5578 	 * actually completes.  That way, the OOA queue mechanism can work
5579 	 * to block commands that shouldn't proceed.  Another alternative
5580 	 * would be to put the copy in the queue in place of the original,
5581 	 * and return the original back to the caller.  That could be
5582 	 * slightly safer..
5583 	 */
5584 	if ((cdb->byte2 & SSS_IMMED)
5585 	 && ((cdb->how & SSS_START) == 0)) {
5586 		ctl_set_invalid_field(ctsio,
5587 				      /*sks_valid*/ 1,
5588 				      /*command*/ 1,
5589 				      /*field*/ 1,
5590 				      /*bit_valid*/ 1,
5591 				      /*bit*/ 0);
5592 		ctl_done((union ctl_io *)ctsio);
5593 		return (CTL_RETVAL_COMPLETE);
5594 	}
5595 
5596 	if ((lun->flags & CTL_LUN_PR_RESERVED)
5597 	 && ((cdb->how & SSS_START)==0)) {
5598 		uint32_t residx;
5599 
5600 		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5601 		if (lun->pr_keys[residx] == 0
5602 		 || (lun->pr_res_idx!=residx && lun->res_type < 4)) {
5603 
5604 			ctl_set_reservation_conflict(ctsio);
5605 			ctl_done((union ctl_io *)ctsio);
5606 			return (CTL_RETVAL_COMPLETE);
5607 		}
5608 	}
5609 
5610 	/*
5611 	 * If there is no backend on this device, we can't start or stop
5612 	 * it.  In theory we shouldn't get any start/stop commands in the
5613 	 * first place at this level if the LUN doesn't have a backend.
5614 	 * That should get stopped by the command decode code.
5615 	 */
5616 	if (lun->backend == NULL) {
5617 		ctl_set_invalid_opcode(ctsio);
5618 		ctl_done((union ctl_io *)ctsio);
5619 		return (CTL_RETVAL_COMPLETE);
5620 	}
5621 
5622 	/*
5623 	 * XXX KDM Copan-specific offline behavior.
5624 	 * Figure out a reasonable way to port this?
5625 	 */
5626 #ifdef NEEDTOPORT
5627 	mtx_lock(&lun->lun_lock);
5628 
5629 	if (((cdb->byte2 & SSS_ONOFFLINE) == 0)
5630 	 && (lun->flags & CTL_LUN_OFFLINE)) {
5631 		/*
5632 		 * If the LUN is offline, and the on/offline bit isn't set,
5633 		 * reject the start or stop.  Otherwise, let it through.
5634 		 */
5635 		mtx_unlock(&lun->lun_lock);
5636 		ctl_set_lun_not_ready(ctsio);
5637 		ctl_done((union ctl_io *)ctsio);
5638 	} else {
5639 		mtx_unlock(&lun->lun_lock);
5640 #endif /* NEEDTOPORT */
5641 		/*
5642 		 * This could be a start or a stop when we're online,
5643 		 * or a stop/offline or start/online.  A start or stop when
5644 		 * we're offline is covered in the case above.
5645 		 */
5646 		/*
5647 		 * In the non-immediate case, we send the request to
5648 		 * the backend and return status to the user when
5649 		 * it is done.
5650 		 *
5651 		 * In the immediate case, we allocate a new ctl_io
5652 		 * to hold a copy of the request, and send that to
5653 		 * the backend.  We then set good status on the
5654 		 * user's request and return it immediately.
5655 		 */
5656 		if (cdb->byte2 & SSS_IMMED) {
5657 			union ctl_io *new_io;
5658 
5659 			new_io = ctl_alloc_io(ctsio->io_hdr.pool);
5660 			if (new_io == NULL) {
5661 				ctl_set_busy(ctsio);
5662 				ctl_done((union ctl_io *)ctsio);
5663 			} else {
5664 				ctl_copy_io((union ctl_io *)ctsio,
5665 					    new_io);
5666 				retval = lun->backend->config_write(new_io);
5667 				ctl_set_success(ctsio);
5668 				ctl_done((union ctl_io *)ctsio);
5669 			}
5670 		} else {
5671 			retval = lun->backend->config_write(
5672 				(union ctl_io *)ctsio);
5673 		}
5674 #ifdef NEEDTOPORT
5675 	}
5676 #endif
5677 	return (retval);
5678 }
5679 
5680 /*
5681  * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5682  * we don't really do anything with the LBA and length fields if the user
5683  * passes them in.  Instead we'll just flush out the cache for the entire
5684  * LUN.
5685  */
5686 int
5687 ctl_sync_cache(struct ctl_scsiio *ctsio)
5688 {
5689 	struct ctl_lun *lun;
5690 	struct ctl_softc *ctl_softc;
5691 	uint64_t starting_lba;
5692 	uint32_t block_count;
5693 	int retval;
5694 
5695 	CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5696 
5697 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5698 	ctl_softc = control_softc;
5699 	retval = 0;
5700 
5701 	switch (ctsio->cdb[0]) {
5702 	case SYNCHRONIZE_CACHE: {
5703 		struct scsi_sync_cache *cdb;
5704 		cdb = (struct scsi_sync_cache *)ctsio->cdb;
5705 
5706 		starting_lba = scsi_4btoul(cdb->begin_lba);
5707 		block_count = scsi_2btoul(cdb->lb_count);
5708 		break;
5709 	}
5710 	case SYNCHRONIZE_CACHE_16: {
5711 		struct scsi_sync_cache_16 *cdb;
5712 		cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5713 
5714 		starting_lba = scsi_8btou64(cdb->begin_lba);
5715 		block_count = scsi_4btoul(cdb->lb_count);
5716 		break;
5717 	}
5718 	default:
5719 		ctl_set_invalid_opcode(ctsio);
5720 		ctl_done((union ctl_io *)ctsio);
5721 		goto bailout;
5722 		break; /* NOTREACHED */
5723 	}
5724 
5725 	/*
5726 	 * We check the LBA and length, but don't do anything with them.
5727 	 * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5728 	 * get flushed.  This check will just help satisfy anyone who wants
5729 	 * to see an error for an out of range LBA.
5730 	 */
5731 	if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5732 		ctl_set_lba_out_of_range(ctsio);
5733 		ctl_done((union ctl_io *)ctsio);
5734 		goto bailout;
5735 	}
5736 
5737 	/*
5738 	 * If this LUN has no backend, we can't flush the cache anyway.
5739 	 */
5740 	if (lun->backend == NULL) {
5741 		ctl_set_invalid_opcode(ctsio);
5742 		ctl_done((union ctl_io *)ctsio);
5743 		goto bailout;
5744 	}
5745 
5746 	/*
5747 	 * Check to see whether we're configured to send the SYNCHRONIZE
5748 	 * CACHE command directly to the back end.
5749 	 */
5750 	mtx_lock(&lun->lun_lock);
5751 	if ((ctl_softc->flags & CTL_FLAG_REAL_SYNC)
5752 	 && (++(lun->sync_count) >= lun->sync_interval)) {
5753 		lun->sync_count = 0;
5754 		mtx_unlock(&lun->lun_lock);
5755 		retval = lun->backend->config_write((union ctl_io *)ctsio);
5756 	} else {
5757 		mtx_unlock(&lun->lun_lock);
5758 		ctl_set_success(ctsio);
5759 		ctl_done((union ctl_io *)ctsio);
5760 	}
5761 
5762 bailout:
5763 
5764 	return (retval);
5765 }
5766 
5767 int
5768 ctl_format(struct ctl_scsiio *ctsio)
5769 {
5770 	struct scsi_format *cdb;
5771 	struct ctl_lun *lun;
5772 	struct ctl_softc *ctl_softc;
5773 	int length, defect_list_len;
5774 
5775 	CTL_DEBUG_PRINT(("ctl_format\n"));
5776 
5777 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5778 	ctl_softc = control_softc;
5779 
5780 	cdb = (struct scsi_format *)ctsio->cdb;
5781 
5782 	length = 0;
5783 	if (cdb->byte2 & SF_FMTDATA) {
5784 		if (cdb->byte2 & SF_LONGLIST)
5785 			length = sizeof(struct scsi_format_header_long);
5786 		else
5787 			length = sizeof(struct scsi_format_header_short);
5788 	}
5789 
5790 	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5791 	 && (length > 0)) {
5792 		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5793 		ctsio->kern_data_len = length;
5794 		ctsio->kern_total_len = length;
5795 		ctsio->kern_data_resid = 0;
5796 		ctsio->kern_rel_offset = 0;
5797 		ctsio->kern_sg_entries = 0;
5798 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5799 		ctsio->be_move_done = ctl_config_move_done;
5800 		ctl_datamove((union ctl_io *)ctsio);
5801 
5802 		return (CTL_RETVAL_COMPLETE);
5803 	}
5804 
5805 	defect_list_len = 0;
5806 
5807 	if (cdb->byte2 & SF_FMTDATA) {
5808 		if (cdb->byte2 & SF_LONGLIST) {
5809 			struct scsi_format_header_long *header;
5810 
5811 			header = (struct scsi_format_header_long *)
5812 				ctsio->kern_data_ptr;
5813 
5814 			defect_list_len = scsi_4btoul(header->defect_list_len);
5815 			if (defect_list_len != 0) {
5816 				ctl_set_invalid_field(ctsio,
5817 						      /*sks_valid*/ 1,
5818 						      /*command*/ 0,
5819 						      /*field*/ 2,
5820 						      /*bit_valid*/ 0,
5821 						      /*bit*/ 0);
5822 				goto bailout;
5823 			}
5824 		} else {
5825 			struct scsi_format_header_short *header;
5826 
5827 			header = (struct scsi_format_header_short *)
5828 				ctsio->kern_data_ptr;
5829 
5830 			defect_list_len = scsi_2btoul(header->defect_list_len);
5831 			if (defect_list_len != 0) {
5832 				ctl_set_invalid_field(ctsio,
5833 						      /*sks_valid*/ 1,
5834 						      /*command*/ 0,
5835 						      /*field*/ 2,
5836 						      /*bit_valid*/ 0,
5837 						      /*bit*/ 0);
5838 				goto bailout;
5839 			}
5840 		}
5841 	}
5842 
5843 	/*
5844 	 * The format command will clear out the "Medium format corrupted"
5845 	 * status if set by the configuration code.  That status is really
5846 	 * just a way to notify the host that we have lost the media, and
5847 	 * get them to issue a command that will basically make them think
5848 	 * they're blowing away the media.
5849 	 */
5850 	mtx_lock(&lun->lun_lock);
5851 	lun->flags &= ~CTL_LUN_INOPERABLE;
5852 	mtx_unlock(&lun->lun_lock);
5853 
5854 	ctsio->scsi_status = SCSI_STATUS_OK;
5855 	ctsio->io_hdr.status = CTL_SUCCESS;
5856 bailout:
5857 
5858 	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5859 		free(ctsio->kern_data_ptr, M_CTL);
5860 		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5861 	}
5862 
5863 	ctl_done((union ctl_io *)ctsio);
5864 	return (CTL_RETVAL_COMPLETE);
5865 }
5866 
5867 int
5868 ctl_read_buffer(struct ctl_scsiio *ctsio)
5869 {
5870 	struct scsi_read_buffer *cdb;
5871 	struct ctl_lun *lun;
5872 	int buffer_offset, len;
5873 	static uint8_t descr[4];
5874 	static uint8_t echo_descr[4] = { 0 };
5875 
5876 	CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5877 
5878 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5879 	cdb = (struct scsi_read_buffer *)ctsio->cdb;
5880 
5881 	if (lun->flags & CTL_LUN_PR_RESERVED) {
5882 		uint32_t residx;
5883 
5884 		/*
5885 		 * XXX KDM need a lock here.
5886 		 */
5887 		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5888 		if ((lun->res_type == SPR_TYPE_EX_AC
5889 		  && residx != lun->pr_res_idx)
5890 		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
5891 		   || lun->res_type == SPR_TYPE_EX_AC_AR)
5892 		  && lun->pr_keys[residx] == 0)) {
5893 			ctl_set_reservation_conflict(ctsio);
5894 			ctl_done((union ctl_io *)ctsio);
5895 			return (CTL_RETVAL_COMPLETE);
5896 	        }
5897 	}
5898 
5899 	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA &&
5900 	    (cdb->byte2 & RWB_MODE) != RWB_MODE_ECHO_DESCR &&
5901 	    (cdb->byte2 & RWB_MODE) != RWB_MODE_DESCR) {
5902 		ctl_set_invalid_field(ctsio,
5903 				      /*sks_valid*/ 1,
5904 				      /*command*/ 1,
5905 				      /*field*/ 1,
5906 				      /*bit_valid*/ 1,
5907 				      /*bit*/ 4);
5908 		ctl_done((union ctl_io *)ctsio);
5909 		return (CTL_RETVAL_COMPLETE);
5910 	}
5911 
5912 	len = scsi_3btoul(cdb->length);
5913 	buffer_offset = scsi_3btoul(cdb->offset);
5914 
5915 	if (buffer_offset + len > sizeof(lun->write_buffer)) {
5916 		ctl_set_invalid_field(ctsio,
5917 				      /*sks_valid*/ 1,
5918 				      /*command*/ 1,
5919 				      /*field*/ 6,
5920 				      /*bit_valid*/ 0,
5921 				      /*bit*/ 0);
5922 		ctl_done((union ctl_io *)ctsio);
5923 		return (CTL_RETVAL_COMPLETE);
5924 	}
5925 
5926 	if ((cdb->byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5927 		descr[0] = 0;
5928 		scsi_ulto3b(sizeof(lun->write_buffer), &descr[1]);
5929 		ctsio->kern_data_ptr = descr;
5930 		len = min(len, sizeof(descr));
5931 	} else if ((cdb->byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5932 		ctsio->kern_data_ptr = echo_descr;
5933 		len = min(len, sizeof(echo_descr));
5934 	} else
5935 		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5936 	ctsio->kern_data_len = len;
5937 	ctsio->kern_total_len = len;
5938 	ctsio->kern_data_resid = 0;
5939 	ctsio->kern_rel_offset = 0;
5940 	ctsio->kern_sg_entries = 0;
5941 	ctsio->be_move_done = ctl_config_move_done;
5942 	ctl_datamove((union ctl_io *)ctsio);
5943 
5944 	return (CTL_RETVAL_COMPLETE);
5945 }
5946 
5947 int
5948 ctl_write_buffer(struct ctl_scsiio *ctsio)
5949 {
5950 	struct scsi_write_buffer *cdb;
5951 	struct ctl_lun *lun;
5952 	int buffer_offset, len;
5953 
5954 	CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5955 
5956 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5957 	cdb = (struct scsi_write_buffer *)ctsio->cdb;
5958 
5959 	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5960 		ctl_set_invalid_field(ctsio,
5961 				      /*sks_valid*/ 1,
5962 				      /*command*/ 1,
5963 				      /*field*/ 1,
5964 				      /*bit_valid*/ 1,
5965 				      /*bit*/ 4);
5966 		ctl_done((union ctl_io *)ctsio);
5967 		return (CTL_RETVAL_COMPLETE);
5968 	}
5969 
5970 	len = scsi_3btoul(cdb->length);
5971 	buffer_offset = scsi_3btoul(cdb->offset);
5972 
5973 	if (buffer_offset + len > sizeof(lun->write_buffer)) {
5974 		ctl_set_invalid_field(ctsio,
5975 				      /*sks_valid*/ 1,
5976 				      /*command*/ 1,
5977 				      /*field*/ 6,
5978 				      /*bit_valid*/ 0,
5979 				      /*bit*/ 0);
5980 		ctl_done((union ctl_io *)ctsio);
5981 		return (CTL_RETVAL_COMPLETE);
5982 	}
5983 
5984 	/*
5985 	 * If we've got a kernel request that hasn't been malloced yet,
5986 	 * malloc it and tell the caller the data buffer is here.
5987 	 */
5988 	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5989 		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5990 		ctsio->kern_data_len = len;
5991 		ctsio->kern_total_len = len;
5992 		ctsio->kern_data_resid = 0;
5993 		ctsio->kern_rel_offset = 0;
5994 		ctsio->kern_sg_entries = 0;
5995 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5996 		ctsio->be_move_done = ctl_config_move_done;
5997 		ctl_datamove((union ctl_io *)ctsio);
5998 
5999 		return (CTL_RETVAL_COMPLETE);
6000 	}
6001 
6002 	ctl_done((union ctl_io *)ctsio);
6003 
6004 	return (CTL_RETVAL_COMPLETE);
6005 }
6006 
6007 int
6008 ctl_write_same(struct ctl_scsiio *ctsio)
6009 {
6010 	struct ctl_lun *lun;
6011 	struct ctl_lba_len_flags *lbalen;
6012 	uint64_t lba;
6013 	uint32_t num_blocks;
6014 	int len, retval;
6015 	uint8_t byte2;
6016 
6017 	retval = CTL_RETVAL_COMPLETE;
6018 
6019 	CTL_DEBUG_PRINT(("ctl_write_same\n"));
6020 
6021 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6022 
6023 	switch (ctsio->cdb[0]) {
6024 	case WRITE_SAME_10: {
6025 		struct scsi_write_same_10 *cdb;
6026 
6027 		cdb = (struct scsi_write_same_10 *)ctsio->cdb;
6028 
6029 		lba = scsi_4btoul(cdb->addr);
6030 		num_blocks = scsi_2btoul(cdb->length);
6031 		byte2 = cdb->byte2;
6032 		break;
6033 	}
6034 	case WRITE_SAME_16: {
6035 		struct scsi_write_same_16 *cdb;
6036 
6037 		cdb = (struct scsi_write_same_16 *)ctsio->cdb;
6038 
6039 		lba = scsi_8btou64(cdb->addr);
6040 		num_blocks = scsi_4btoul(cdb->length);
6041 		byte2 = cdb->byte2;
6042 		break;
6043 	}
6044 	default:
6045 		/*
6046 		 * We got a command we don't support.  This shouldn't
6047 		 * happen, commands should be filtered out above us.
6048 		 */
6049 		ctl_set_invalid_opcode(ctsio);
6050 		ctl_done((union ctl_io *)ctsio);
6051 
6052 		return (CTL_RETVAL_COMPLETE);
6053 		break; /* NOTREACHED */
6054 	}
6055 
6056 	/* NDOB and ANCHOR flags can be used only together with UNMAP */
6057 	if ((byte2 & SWS_UNMAP) == 0 &&
6058 	    (byte2 & (SWS_NDOB | SWS_ANCHOR)) != 0) {
6059 		ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
6060 		    /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
6061 		ctl_done((union ctl_io *)ctsio);
6062 		return (CTL_RETVAL_COMPLETE);
6063 	}
6064 
6065 	/*
6066 	 * The first check is to make sure we're in bounds, the second
6067 	 * check is to catch wrap-around problems.  If the lba + num blocks
6068 	 * is less than the lba, then we've wrapped around and the block
6069 	 * range is invalid anyway.
6070 	 */
6071 	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
6072 	 || ((lba + num_blocks) < lba)) {
6073 		ctl_set_lba_out_of_range(ctsio);
6074 		ctl_done((union ctl_io *)ctsio);
6075 		return (CTL_RETVAL_COMPLETE);
6076 	}
6077 
6078 	/* Zero number of blocks means "to the last logical block" */
6079 	if (num_blocks == 0) {
6080 		if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
6081 			ctl_set_invalid_field(ctsio,
6082 					      /*sks_valid*/ 0,
6083 					      /*command*/ 1,
6084 					      /*field*/ 0,
6085 					      /*bit_valid*/ 0,
6086 					      /*bit*/ 0);
6087 			ctl_done((union ctl_io *)ctsio);
6088 			return (CTL_RETVAL_COMPLETE);
6089 		}
6090 		num_blocks = (lun->be_lun->maxlba + 1) - lba;
6091 	}
6092 
6093 	len = lun->be_lun->blocksize;
6094 
6095 	/*
6096 	 * If we've got a kernel request that hasn't been malloced yet,
6097 	 * malloc it and tell the caller the data buffer is here.
6098 	 */
6099 	if ((byte2 & SWS_NDOB) == 0 &&
6100 	    (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6101 		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
6102 		ctsio->kern_data_len = len;
6103 		ctsio->kern_total_len = len;
6104 		ctsio->kern_data_resid = 0;
6105 		ctsio->kern_rel_offset = 0;
6106 		ctsio->kern_sg_entries = 0;
6107 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6108 		ctsio->be_move_done = ctl_config_move_done;
6109 		ctl_datamove((union ctl_io *)ctsio);
6110 
6111 		return (CTL_RETVAL_COMPLETE);
6112 	}
6113 
6114 	lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
6115 	lbalen->lba = lba;
6116 	lbalen->len = num_blocks;
6117 	lbalen->flags = byte2;
6118 	retval = lun->backend->config_write((union ctl_io *)ctsio);
6119 
6120 	return (retval);
6121 }
6122 
6123 int
6124 ctl_unmap(struct ctl_scsiio *ctsio)
6125 {
6126 	struct ctl_lun *lun;
6127 	struct scsi_unmap *cdb;
6128 	struct ctl_ptr_len_flags *ptrlen;
6129 	struct scsi_unmap_header *hdr;
6130 	struct scsi_unmap_desc *buf, *end, *endnz, *range;
6131 	uint64_t lba;
6132 	uint32_t num_blocks;
6133 	int len, retval;
6134 	uint8_t byte2;
6135 
6136 	retval = CTL_RETVAL_COMPLETE;
6137 
6138 	CTL_DEBUG_PRINT(("ctl_unmap\n"));
6139 
6140 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6141 	cdb = (struct scsi_unmap *)ctsio->cdb;
6142 
6143 	len = scsi_2btoul(cdb->length);
6144 	byte2 = cdb->byte2;
6145 
6146 	/*
6147 	 * If we've got a kernel request that hasn't been malloced yet,
6148 	 * malloc it and tell the caller the data buffer is here.
6149 	 */
6150 	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6151 		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
6152 		ctsio->kern_data_len = len;
6153 		ctsio->kern_total_len = len;
6154 		ctsio->kern_data_resid = 0;
6155 		ctsio->kern_rel_offset = 0;
6156 		ctsio->kern_sg_entries = 0;
6157 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6158 		ctsio->be_move_done = ctl_config_move_done;
6159 		ctl_datamove((union ctl_io *)ctsio);
6160 
6161 		return (CTL_RETVAL_COMPLETE);
6162 	}
6163 
6164 	len = ctsio->kern_total_len - ctsio->kern_data_resid;
6165 	hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
6166 	if (len < sizeof (*hdr) ||
6167 	    len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
6168 	    len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
6169 	    scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
6170 		ctl_set_invalid_field(ctsio,
6171 				      /*sks_valid*/ 0,
6172 				      /*command*/ 0,
6173 				      /*field*/ 0,
6174 				      /*bit_valid*/ 0,
6175 				      /*bit*/ 0);
6176 		ctl_done((union ctl_io *)ctsio);
6177 		return (CTL_RETVAL_COMPLETE);
6178 	}
6179 	len = scsi_2btoul(hdr->desc_length);
6180 	buf = (struct scsi_unmap_desc *)(hdr + 1);
6181 	end = buf + len / sizeof(*buf);
6182 
6183 	endnz = buf;
6184 	for (range = buf; range < end; range++) {
6185 		lba = scsi_8btou64(range->lba);
6186 		num_blocks = scsi_4btoul(range->length);
6187 		if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
6188 		 || ((lba + num_blocks) < lba)) {
6189 			ctl_set_lba_out_of_range(ctsio);
6190 			ctl_done((union ctl_io *)ctsio);
6191 			return (CTL_RETVAL_COMPLETE);
6192 		}
6193 		if (num_blocks != 0)
6194 			endnz = range + 1;
6195 	}
6196 
6197 	/*
6198 	 * Block backend can not handle zero last range.
6199 	 * Filter it out and return if there is nothing left.
6200 	 */
6201 	len = (uint8_t *)endnz - (uint8_t *)buf;
6202 	if (len == 0) {
6203 		ctl_set_success(ctsio);
6204 		ctl_done((union ctl_io *)ctsio);
6205 		return (CTL_RETVAL_COMPLETE);
6206 	}
6207 
6208 	mtx_lock(&lun->lun_lock);
6209 	ptrlen = (struct ctl_ptr_len_flags *)
6210 	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
6211 	ptrlen->ptr = (void *)buf;
6212 	ptrlen->len = len;
6213 	ptrlen->flags = byte2;
6214 	ctl_check_blocked(lun);
6215 	mtx_unlock(&lun->lun_lock);
6216 
6217 	retval = lun->backend->config_write((union ctl_io *)ctsio);
6218 	return (retval);
6219 }
6220 
6221 /*
6222  * Note that this function currently doesn't actually do anything inside
6223  * CTL to enforce things if the DQue bit is turned on.
6224  *
6225  * Also note that this function can't be used in the default case, because
6226  * the DQue bit isn't set in the changeable mask for the control mode page
6227  * anyway.  This is just here as an example for how to implement a page
6228  * handler, and a placeholder in case we want to allow the user to turn
6229  * tagged queueing on and off.
6230  *
6231  * The D_SENSE bit handling is functional, however, and will turn
6232  * descriptor sense on and off for a given LUN.
6233  */
6234 int
6235 ctl_control_page_handler(struct ctl_scsiio *ctsio,
6236 			 struct ctl_page_index *page_index, uint8_t *page_ptr)
6237 {
6238 	struct scsi_control_page *current_cp, *saved_cp, *user_cp;
6239 	struct ctl_lun *lun;
6240 	struct ctl_softc *softc;
6241 	int set_ua;
6242 	uint32_t initidx;
6243 
6244 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6245 	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6246 	set_ua = 0;
6247 
6248 	user_cp = (struct scsi_control_page *)page_ptr;
6249 	current_cp = (struct scsi_control_page *)
6250 		(page_index->page_data + (page_index->page_len *
6251 		CTL_PAGE_CURRENT));
6252 	saved_cp = (struct scsi_control_page *)
6253 		(page_index->page_data + (page_index->page_len *
6254 		CTL_PAGE_SAVED));
6255 
6256 	softc = control_softc;
6257 
6258 	mtx_lock(&lun->lun_lock);
6259 	if (((current_cp->rlec & SCP_DSENSE) == 0)
6260 	 && ((user_cp->rlec & SCP_DSENSE) != 0)) {
6261 		/*
6262 		 * Descriptor sense is currently turned off and the user
6263 		 * wants to turn it on.
6264 		 */
6265 		current_cp->rlec |= SCP_DSENSE;
6266 		saved_cp->rlec |= SCP_DSENSE;
6267 		lun->flags |= CTL_LUN_SENSE_DESC;
6268 		set_ua = 1;
6269 	} else if (((current_cp->rlec & SCP_DSENSE) != 0)
6270 		&& ((user_cp->rlec & SCP_DSENSE) == 0)) {
6271 		/*
6272 		 * Descriptor sense is currently turned on, and the user
6273 		 * wants to turn it off.
6274 		 */
6275 		current_cp->rlec &= ~SCP_DSENSE;
6276 		saved_cp->rlec &= ~SCP_DSENSE;
6277 		lun->flags &= ~CTL_LUN_SENSE_DESC;
6278 		set_ua = 1;
6279 	}
6280 	if ((current_cp->queue_flags & SCP_QUEUE_ALG_MASK) !=
6281 	    (user_cp->queue_flags & SCP_QUEUE_ALG_MASK)) {
6282 		current_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
6283 		current_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
6284 		saved_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
6285 		saved_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
6286 		set_ua = 1;
6287 	}
6288 	if ((current_cp->eca_and_aen & SCP_SWP) !=
6289 	    (user_cp->eca_and_aen & SCP_SWP)) {
6290 		current_cp->eca_and_aen &= ~SCP_SWP;
6291 		current_cp->eca_and_aen |= user_cp->eca_and_aen & SCP_SWP;
6292 		saved_cp->eca_and_aen &= ~SCP_SWP;
6293 		saved_cp->eca_and_aen |= user_cp->eca_and_aen & SCP_SWP;
6294 		set_ua = 1;
6295 	}
6296 	if (set_ua != 0) {
6297 		int i;
6298 		/*
6299 		 * Let other initiators know that the mode
6300 		 * parameters for this LUN have changed.
6301 		 */
6302 		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
6303 			if (i == initidx)
6304 				continue;
6305 
6306 			lun->pending_ua[i] |= CTL_UA_MODE_CHANGE;
6307 		}
6308 	}
6309 	mtx_unlock(&lun->lun_lock);
6310 
6311 	return (0);
6312 }
6313 
6314 int
6315 ctl_caching_sp_handler(struct ctl_scsiio *ctsio,
6316 		     struct ctl_page_index *page_index, uint8_t *page_ptr)
6317 {
6318 	struct scsi_caching_page *current_cp, *saved_cp, *user_cp;
6319 	struct ctl_lun *lun;
6320 	int set_ua;
6321 	uint32_t initidx;
6322 
6323 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6324 	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6325 	set_ua = 0;
6326 
6327 	user_cp = (struct scsi_caching_page *)page_ptr;
6328 	current_cp = (struct scsi_caching_page *)
6329 		(page_index->page_data + (page_index->page_len *
6330 		CTL_PAGE_CURRENT));
6331 	saved_cp = (struct scsi_caching_page *)
6332 		(page_index->page_data + (page_index->page_len *
6333 		CTL_PAGE_SAVED));
6334 
6335 	mtx_lock(&lun->lun_lock);
6336 	if ((current_cp->flags1 & (SCP_WCE | SCP_RCD)) !=
6337 	    (user_cp->flags1 & (SCP_WCE | SCP_RCD))) {
6338 		current_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
6339 		current_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
6340 		saved_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
6341 		saved_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
6342 		set_ua = 1;
6343 	}
6344 	if (set_ua != 0) {
6345 		int i;
6346 		/*
6347 		 * Let other initiators know that the mode
6348 		 * parameters for this LUN have changed.
6349 		 */
6350 		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
6351 			if (i == initidx)
6352 				continue;
6353 
6354 			lun->pending_ua[i] |= CTL_UA_MODE_CHANGE;
6355 		}
6356 	}
6357 	mtx_unlock(&lun->lun_lock);
6358 
6359 	return (0);
6360 }
6361 
6362 int
6363 ctl_power_sp_handler(struct ctl_scsiio *ctsio,
6364 		     struct ctl_page_index *page_index, uint8_t *page_ptr)
6365 {
6366 	return (0);
6367 }
6368 
6369 int
6370 ctl_power_sp_sense_handler(struct ctl_scsiio *ctsio,
6371 			   struct ctl_page_index *page_index, int pc)
6372 {
6373 	struct copan_power_subpage *page;
6374 
6375 	page = (struct copan_power_subpage *)page_index->page_data +
6376 		(page_index->page_len * pc);
6377 
6378 	switch (pc) {
6379 	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6380 		/*
6381 		 * We don't update the changable bits for this page.
6382 		 */
6383 		break;
6384 	case SMS_PAGE_CTRL_CURRENT >> 6:
6385 	case SMS_PAGE_CTRL_DEFAULT >> 6:
6386 	case SMS_PAGE_CTRL_SAVED >> 6:
6387 #ifdef NEEDTOPORT
6388 		ctl_update_power_subpage(page);
6389 #endif
6390 		break;
6391 	default:
6392 #ifdef NEEDTOPORT
6393 		EPRINT(0, "Invalid PC %d!!", pc);
6394 #endif
6395 		break;
6396 	}
6397 	return (0);
6398 }
6399 
6400 
6401 int
6402 ctl_aps_sp_handler(struct ctl_scsiio *ctsio,
6403 		   struct ctl_page_index *page_index, uint8_t *page_ptr)
6404 {
6405 	struct copan_aps_subpage *user_sp;
6406 	struct copan_aps_subpage *current_sp;
6407 	union ctl_modepage_info *modepage_info;
6408 	struct ctl_softc *softc;
6409 	struct ctl_lun *lun;
6410 	int retval;
6411 
6412 	retval = CTL_RETVAL_COMPLETE;
6413 	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
6414 		     (page_index->page_len * CTL_PAGE_CURRENT));
6415 	softc = control_softc;
6416 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6417 
6418 	user_sp = (struct copan_aps_subpage *)page_ptr;
6419 
6420 	modepage_info = (union ctl_modepage_info *)
6421 		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6422 
6423 	modepage_info->header.page_code = page_index->page_code & SMPH_PC_MASK;
6424 	modepage_info->header.subpage = page_index->subpage;
6425 	modepage_info->aps.lock_active = user_sp->lock_active;
6426 
6427 	mtx_lock(&softc->ctl_lock);
6428 
6429 	/*
6430 	 * If there is a request to lock the LUN and another LUN is locked
6431 	 * this is an error. If the requested LUN is already locked ignore
6432 	 * the request. If no LUN is locked attempt to lock it.
6433 	 * if there is a request to unlock the LUN and the LUN is currently
6434 	 * locked attempt to unlock it. Otherwise ignore the request. i.e.
6435 	 * if another LUN is locked or no LUN is locked.
6436 	 */
6437 	if (user_sp->lock_active & APS_LOCK_ACTIVE) {
6438 		if (softc->aps_locked_lun == lun->lun) {
6439 			/*
6440 			 * This LUN is already locked, so we're done.
6441 			 */
6442 			retval = CTL_RETVAL_COMPLETE;
6443 		} else if (softc->aps_locked_lun == 0) {
6444 			/*
6445 			 * No one has the lock, pass the request to the
6446 			 * backend.
6447 			 */
6448 			retval = lun->backend->config_write(
6449 				(union ctl_io *)ctsio);
6450 		} else {
6451 			/*
6452 			 * Someone else has the lock, throw out the request.
6453 			 */
6454 			ctl_set_already_locked(ctsio);
6455 			free(ctsio->kern_data_ptr, M_CTL);
6456 			ctl_done((union ctl_io *)ctsio);
6457 
6458 			/*
6459 			 * Set the return value so that ctl_do_mode_select()
6460 			 * won't try to complete the command.  We already
6461 			 * completed it here.
6462 			 */
6463 			retval = CTL_RETVAL_ERROR;
6464 		}
6465 	} else if (softc->aps_locked_lun == lun->lun) {
6466 		/*
6467 		 * This LUN is locked, so pass the unlock request to the
6468 		 * backend.
6469 		 */
6470 		retval = lun->backend->config_write((union ctl_io *)ctsio);
6471 	}
6472 	mtx_unlock(&softc->ctl_lock);
6473 
6474 	return (retval);
6475 }
6476 
6477 int
6478 ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio,
6479 				struct ctl_page_index *page_index,
6480 				uint8_t *page_ptr)
6481 {
6482 	uint8_t *c;
6483 	int i;
6484 
6485 	c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs;
6486 	ctl_time_io_secs =
6487 		(c[0] << 8) |
6488 		(c[1] << 0) |
6489 		0;
6490 	CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs));
6491 	printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs);
6492 	printf("page data:");
6493 	for (i=0; i<8; i++)
6494 		printf(" %.2x",page_ptr[i]);
6495 	printf("\n");
6496 	return (0);
6497 }
6498 
6499 int
6500 ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio,
6501 			       struct ctl_page_index *page_index,
6502 			       int pc)
6503 {
6504 	struct copan_debugconf_subpage *page;
6505 
6506 	page = (struct copan_debugconf_subpage *)page_index->page_data +
6507 		(page_index->page_len * pc);
6508 
6509 	switch (pc) {
6510 	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6511 	case SMS_PAGE_CTRL_DEFAULT >> 6:
6512 	case SMS_PAGE_CTRL_SAVED >> 6:
6513 		/*
6514 		 * We don't update the changable or default bits for this page.
6515 		 */
6516 		break;
6517 	case SMS_PAGE_CTRL_CURRENT >> 6:
6518 		page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8;
6519 		page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0;
6520 		break;
6521 	default:
6522 #ifdef NEEDTOPORT
6523 		EPRINT(0, "Invalid PC %d!!", pc);
6524 #endif /* NEEDTOPORT */
6525 		break;
6526 	}
6527 	return (0);
6528 }
6529 
6530 
6531 static int
6532 ctl_do_mode_select(union ctl_io *io)
6533 {
6534 	struct scsi_mode_page_header *page_header;
6535 	struct ctl_page_index *page_index;
6536 	struct ctl_scsiio *ctsio;
6537 	int control_dev, page_len;
6538 	int page_len_offset, page_len_size;
6539 	union ctl_modepage_info *modepage_info;
6540 	struct ctl_lun *lun;
6541 	int *len_left, *len_used;
6542 	int retval, i;
6543 
6544 	ctsio = &io->scsiio;
6545 	page_index = NULL;
6546 	page_len = 0;
6547 	retval = CTL_RETVAL_COMPLETE;
6548 
6549 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6550 
6551 	if (lun->be_lun->lun_type != T_DIRECT)
6552 		control_dev = 1;
6553 	else
6554 		control_dev = 0;
6555 
6556 	modepage_info = (union ctl_modepage_info *)
6557 		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6558 	len_left = &modepage_info->header.len_left;
6559 	len_used = &modepage_info->header.len_used;
6560 
6561 do_next_page:
6562 
6563 	page_header = (struct scsi_mode_page_header *)
6564 		(ctsio->kern_data_ptr + *len_used);
6565 
6566 	if (*len_left == 0) {
6567 		free(ctsio->kern_data_ptr, M_CTL);
6568 		ctl_set_success(ctsio);
6569 		ctl_done((union ctl_io *)ctsio);
6570 		return (CTL_RETVAL_COMPLETE);
6571 	} else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6572 
6573 		free(ctsio->kern_data_ptr, M_CTL);
6574 		ctl_set_param_len_error(ctsio);
6575 		ctl_done((union ctl_io *)ctsio);
6576 		return (CTL_RETVAL_COMPLETE);
6577 
6578 	} else if ((page_header->page_code & SMPH_SPF)
6579 		&& (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6580 
6581 		free(ctsio->kern_data_ptr, M_CTL);
6582 		ctl_set_param_len_error(ctsio);
6583 		ctl_done((union ctl_io *)ctsio);
6584 		return (CTL_RETVAL_COMPLETE);
6585 	}
6586 
6587 
6588 	/*
6589 	 * XXX KDM should we do something with the block descriptor?
6590 	 */
6591 	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6592 
6593 		if ((control_dev != 0)
6594 		 && (lun->mode_pages.index[i].page_flags &
6595 		     CTL_PAGE_FLAG_DISK_ONLY))
6596 			continue;
6597 
6598 		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
6599 		    (page_header->page_code & SMPH_PC_MASK))
6600 			continue;
6601 
6602 		/*
6603 		 * If neither page has a subpage code, then we've got a
6604 		 * match.
6605 		 */
6606 		if (((lun->mode_pages.index[i].page_code & SMPH_SPF) == 0)
6607 		 && ((page_header->page_code & SMPH_SPF) == 0)) {
6608 			page_index = &lun->mode_pages.index[i];
6609 			page_len = page_header->page_length;
6610 			break;
6611 		}
6612 
6613 		/*
6614 		 * If both pages have subpages, then the subpage numbers
6615 		 * have to match.
6616 		 */
6617 		if ((lun->mode_pages.index[i].page_code & SMPH_SPF)
6618 		  && (page_header->page_code & SMPH_SPF)) {
6619 			struct scsi_mode_page_header_sp *sph;
6620 
6621 			sph = (struct scsi_mode_page_header_sp *)page_header;
6622 
6623 			if (lun->mode_pages.index[i].subpage ==
6624 			    sph->subpage) {
6625 				page_index = &lun->mode_pages.index[i];
6626 				page_len = scsi_2btoul(sph->page_length);
6627 				break;
6628 			}
6629 		}
6630 	}
6631 
6632 	/*
6633 	 * If we couldn't find the page, or if we don't have a mode select
6634 	 * handler for it, send back an error to the user.
6635 	 */
6636 	if ((page_index == NULL)
6637 	 || (page_index->select_handler == NULL)) {
6638 		ctl_set_invalid_field(ctsio,
6639 				      /*sks_valid*/ 1,
6640 				      /*command*/ 0,
6641 				      /*field*/ *len_used,
6642 				      /*bit_valid*/ 0,
6643 				      /*bit*/ 0);
6644 		free(ctsio->kern_data_ptr, M_CTL);
6645 		ctl_done((union ctl_io *)ctsio);
6646 		return (CTL_RETVAL_COMPLETE);
6647 	}
6648 
6649 	if (page_index->page_code & SMPH_SPF) {
6650 		page_len_offset = 2;
6651 		page_len_size = 2;
6652 	} else {
6653 		page_len_size = 1;
6654 		page_len_offset = 1;
6655 	}
6656 
6657 	/*
6658 	 * If the length the initiator gives us isn't the one we specify in
6659 	 * the mode page header, or if they didn't specify enough data in
6660 	 * the CDB to avoid truncating this page, kick out the request.
6661 	 */
6662 	if ((page_len != (page_index->page_len - page_len_offset -
6663 			  page_len_size))
6664 	 || (*len_left < page_index->page_len)) {
6665 
6666 
6667 		ctl_set_invalid_field(ctsio,
6668 				      /*sks_valid*/ 1,
6669 				      /*command*/ 0,
6670 				      /*field*/ *len_used + page_len_offset,
6671 				      /*bit_valid*/ 0,
6672 				      /*bit*/ 0);
6673 		free(ctsio->kern_data_ptr, M_CTL);
6674 		ctl_done((union ctl_io *)ctsio);
6675 		return (CTL_RETVAL_COMPLETE);
6676 	}
6677 
6678 	/*
6679 	 * Run through the mode page, checking to make sure that the bits
6680 	 * the user changed are actually legal for him to change.
6681 	 */
6682 	for (i = 0; i < page_index->page_len; i++) {
6683 		uint8_t *user_byte, *change_mask, *current_byte;
6684 		int bad_bit;
6685 		int j;
6686 
6687 		user_byte = (uint8_t *)page_header + i;
6688 		change_mask = page_index->page_data +
6689 			      (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6690 		current_byte = page_index->page_data +
6691 			       (page_index->page_len * CTL_PAGE_CURRENT) + i;
6692 
6693 		/*
6694 		 * Check to see whether the user set any bits in this byte
6695 		 * that he is not allowed to set.
6696 		 */
6697 		if ((*user_byte & ~(*change_mask)) ==
6698 		    (*current_byte & ~(*change_mask)))
6699 			continue;
6700 
6701 		/*
6702 		 * Go through bit by bit to determine which one is illegal.
6703 		 */
6704 		bad_bit = 0;
6705 		for (j = 7; j >= 0; j--) {
6706 			if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6707 			    (((1 << i) & ~(*change_mask)) & *current_byte)) {
6708 				bad_bit = i;
6709 				break;
6710 			}
6711 		}
6712 		ctl_set_invalid_field(ctsio,
6713 				      /*sks_valid*/ 1,
6714 				      /*command*/ 0,
6715 				      /*field*/ *len_used + i,
6716 				      /*bit_valid*/ 1,
6717 				      /*bit*/ bad_bit);
6718 		free(ctsio->kern_data_ptr, M_CTL);
6719 		ctl_done((union ctl_io *)ctsio);
6720 		return (CTL_RETVAL_COMPLETE);
6721 	}
6722 
6723 	/*
6724 	 * Decrement these before we call the page handler, since we may
6725 	 * end up getting called back one way or another before the handler
6726 	 * returns to this context.
6727 	 */
6728 	*len_left -= page_index->page_len;
6729 	*len_used += page_index->page_len;
6730 
6731 	retval = page_index->select_handler(ctsio, page_index,
6732 					    (uint8_t *)page_header);
6733 
6734 	/*
6735 	 * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6736 	 * wait until this queued command completes to finish processing
6737 	 * the mode page.  If it returns anything other than
6738 	 * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6739 	 * already set the sense information, freed the data pointer, and
6740 	 * completed the io for us.
6741 	 */
6742 	if (retval != CTL_RETVAL_COMPLETE)
6743 		goto bailout_no_done;
6744 
6745 	/*
6746 	 * If the initiator sent us more than one page, parse the next one.
6747 	 */
6748 	if (*len_left > 0)
6749 		goto do_next_page;
6750 
6751 	ctl_set_success(ctsio);
6752 	free(ctsio->kern_data_ptr, M_CTL);
6753 	ctl_done((union ctl_io *)ctsio);
6754 
6755 bailout_no_done:
6756 
6757 	return (CTL_RETVAL_COMPLETE);
6758 
6759 }
6760 
6761 int
6762 ctl_mode_select(struct ctl_scsiio *ctsio)
6763 {
6764 	int param_len, pf, sp;
6765 	int header_size, bd_len;
6766 	int len_left, len_used;
6767 	struct ctl_page_index *page_index;
6768 	struct ctl_lun *lun;
6769 	int control_dev, page_len;
6770 	union ctl_modepage_info *modepage_info;
6771 	int retval;
6772 
6773 	pf = 0;
6774 	sp = 0;
6775 	page_len = 0;
6776 	len_used = 0;
6777 	len_left = 0;
6778 	retval = 0;
6779 	bd_len = 0;
6780 	page_index = NULL;
6781 
6782 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6783 
6784 	if (lun->be_lun->lun_type != T_DIRECT)
6785 		control_dev = 1;
6786 	else
6787 		control_dev = 0;
6788 
6789 	switch (ctsio->cdb[0]) {
6790 	case MODE_SELECT_6: {
6791 		struct scsi_mode_select_6 *cdb;
6792 
6793 		cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6794 
6795 		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6796 		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6797 
6798 		param_len = cdb->length;
6799 		header_size = sizeof(struct scsi_mode_header_6);
6800 		break;
6801 	}
6802 	case MODE_SELECT_10: {
6803 		struct scsi_mode_select_10 *cdb;
6804 
6805 		cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6806 
6807 		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6808 		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6809 
6810 		param_len = scsi_2btoul(cdb->length);
6811 		header_size = sizeof(struct scsi_mode_header_10);
6812 		break;
6813 	}
6814 	default:
6815 		ctl_set_invalid_opcode(ctsio);
6816 		ctl_done((union ctl_io *)ctsio);
6817 		return (CTL_RETVAL_COMPLETE);
6818 		break; /* NOTREACHED */
6819 	}
6820 
6821 	/*
6822 	 * From SPC-3:
6823 	 * "A parameter list length of zero indicates that the Data-Out Buffer
6824 	 * shall be empty. This condition shall not be considered as an error."
6825 	 */
6826 	if (param_len == 0) {
6827 		ctl_set_success(ctsio);
6828 		ctl_done((union ctl_io *)ctsio);
6829 		return (CTL_RETVAL_COMPLETE);
6830 	}
6831 
6832 	/*
6833 	 * Since we'll hit this the first time through, prior to
6834 	 * allocation, we don't need to free a data buffer here.
6835 	 */
6836 	if (param_len < header_size) {
6837 		ctl_set_param_len_error(ctsio);
6838 		ctl_done((union ctl_io *)ctsio);
6839 		return (CTL_RETVAL_COMPLETE);
6840 	}
6841 
6842 	/*
6843 	 * Allocate the data buffer and grab the user's data.  In theory,
6844 	 * we shouldn't have to sanity check the parameter list length here
6845 	 * because the maximum size is 64K.  We should be able to malloc
6846 	 * that much without too many problems.
6847 	 */
6848 	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6849 		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6850 		ctsio->kern_data_len = param_len;
6851 		ctsio->kern_total_len = param_len;
6852 		ctsio->kern_data_resid = 0;
6853 		ctsio->kern_rel_offset = 0;
6854 		ctsio->kern_sg_entries = 0;
6855 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6856 		ctsio->be_move_done = ctl_config_move_done;
6857 		ctl_datamove((union ctl_io *)ctsio);
6858 
6859 		return (CTL_RETVAL_COMPLETE);
6860 	}
6861 
6862 	switch (ctsio->cdb[0]) {
6863 	case MODE_SELECT_6: {
6864 		struct scsi_mode_header_6 *mh6;
6865 
6866 		mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6867 		bd_len = mh6->blk_desc_len;
6868 		break;
6869 	}
6870 	case MODE_SELECT_10: {
6871 		struct scsi_mode_header_10 *mh10;
6872 
6873 		mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6874 		bd_len = scsi_2btoul(mh10->blk_desc_len);
6875 		break;
6876 	}
6877 	default:
6878 		panic("Invalid CDB type %#x", ctsio->cdb[0]);
6879 		break;
6880 	}
6881 
6882 	if (param_len < (header_size + bd_len)) {
6883 		free(ctsio->kern_data_ptr, M_CTL);
6884 		ctl_set_param_len_error(ctsio);
6885 		ctl_done((union ctl_io *)ctsio);
6886 		return (CTL_RETVAL_COMPLETE);
6887 	}
6888 
6889 	/*
6890 	 * Set the IO_CONT flag, so that if this I/O gets passed to
6891 	 * ctl_config_write_done(), it'll get passed back to
6892 	 * ctl_do_mode_select() for further processing, or completion if
6893 	 * we're all done.
6894 	 */
6895 	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6896 	ctsio->io_cont = ctl_do_mode_select;
6897 
6898 	modepage_info = (union ctl_modepage_info *)
6899 		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6900 
6901 	memset(modepage_info, 0, sizeof(*modepage_info));
6902 
6903 	len_left = param_len - header_size - bd_len;
6904 	len_used = header_size + bd_len;
6905 
6906 	modepage_info->header.len_left = len_left;
6907 	modepage_info->header.len_used = len_used;
6908 
6909 	return (ctl_do_mode_select((union ctl_io *)ctsio));
6910 }
6911 
6912 int
6913 ctl_mode_sense(struct ctl_scsiio *ctsio)
6914 {
6915 	struct ctl_lun *lun;
6916 	int pc, page_code, dbd, llba, subpage;
6917 	int alloc_len, page_len, header_len, total_len;
6918 	struct scsi_mode_block_descr *block_desc;
6919 	struct ctl_page_index *page_index;
6920 	int control_dev;
6921 
6922 	dbd = 0;
6923 	llba = 0;
6924 	block_desc = NULL;
6925 	page_index = NULL;
6926 
6927 	CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6928 
6929 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6930 
6931 	if (lun->be_lun->lun_type != T_DIRECT)
6932 		control_dev = 1;
6933 	else
6934 		control_dev = 0;
6935 
6936 	if (lun->flags & CTL_LUN_PR_RESERVED) {
6937 		uint32_t residx;
6938 
6939 		/*
6940 		 * XXX KDM need a lock here.
6941 		 */
6942 		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
6943 		if ((lun->res_type == SPR_TYPE_EX_AC
6944 		  && residx != lun->pr_res_idx)
6945 		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
6946 		   || lun->res_type == SPR_TYPE_EX_AC_AR)
6947 		  && lun->pr_keys[residx] == 0)) {
6948 			ctl_set_reservation_conflict(ctsio);
6949 			ctl_done((union ctl_io *)ctsio);
6950 			return (CTL_RETVAL_COMPLETE);
6951 		}
6952 	}
6953 
6954 	switch (ctsio->cdb[0]) {
6955 	case MODE_SENSE_6: {
6956 		struct scsi_mode_sense_6 *cdb;
6957 
6958 		cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6959 
6960 		header_len = sizeof(struct scsi_mode_hdr_6);
6961 		if (cdb->byte2 & SMS_DBD)
6962 			dbd = 1;
6963 		else
6964 			header_len += sizeof(struct scsi_mode_block_descr);
6965 
6966 		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6967 		page_code = cdb->page & SMS_PAGE_CODE;
6968 		subpage = cdb->subpage;
6969 		alloc_len = cdb->length;
6970 		break;
6971 	}
6972 	case MODE_SENSE_10: {
6973 		struct scsi_mode_sense_10 *cdb;
6974 
6975 		cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6976 
6977 		header_len = sizeof(struct scsi_mode_hdr_10);
6978 
6979 		if (cdb->byte2 & SMS_DBD)
6980 			dbd = 1;
6981 		else
6982 			header_len += sizeof(struct scsi_mode_block_descr);
6983 		if (cdb->byte2 & SMS10_LLBAA)
6984 			llba = 1;
6985 		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6986 		page_code = cdb->page & SMS_PAGE_CODE;
6987 		subpage = cdb->subpage;
6988 		alloc_len = scsi_2btoul(cdb->length);
6989 		break;
6990 	}
6991 	default:
6992 		ctl_set_invalid_opcode(ctsio);
6993 		ctl_done((union ctl_io *)ctsio);
6994 		return (CTL_RETVAL_COMPLETE);
6995 		break; /* NOTREACHED */
6996 	}
6997 
6998 	/*
6999 	 * We have to make a first pass through to calculate the size of
7000 	 * the pages that match the user's query.  Then we allocate enough
7001 	 * memory to hold it, and actually copy the data into the buffer.
7002 	 */
7003 	switch (page_code) {
7004 	case SMS_ALL_PAGES_PAGE: {
7005 		int i;
7006 
7007 		page_len = 0;
7008 
7009 		/*
7010 		 * At the moment, values other than 0 and 0xff here are
7011 		 * reserved according to SPC-3.
7012 		 */
7013 		if ((subpage != SMS_SUBPAGE_PAGE_0)
7014 		 && (subpage != SMS_SUBPAGE_ALL)) {
7015 			ctl_set_invalid_field(ctsio,
7016 					      /*sks_valid*/ 1,
7017 					      /*command*/ 1,
7018 					      /*field*/ 3,
7019 					      /*bit_valid*/ 0,
7020 					      /*bit*/ 0);
7021 			ctl_done((union ctl_io *)ctsio);
7022 			return (CTL_RETVAL_COMPLETE);
7023 		}
7024 
7025 		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
7026 			if ((control_dev != 0)
7027 			 && (lun->mode_pages.index[i].page_flags &
7028 			     CTL_PAGE_FLAG_DISK_ONLY))
7029 				continue;
7030 
7031 			/*
7032 			 * We don't use this subpage if the user didn't
7033 			 * request all subpages.
7034 			 */
7035 			if ((lun->mode_pages.index[i].subpage != 0)
7036 			 && (subpage == SMS_SUBPAGE_PAGE_0))
7037 				continue;
7038 
7039 #if 0
7040 			printf("found page %#x len %d\n",
7041 			       lun->mode_pages.index[i].page_code &
7042 			       SMPH_PC_MASK,
7043 			       lun->mode_pages.index[i].page_len);
7044 #endif
7045 			page_len += lun->mode_pages.index[i].page_len;
7046 		}
7047 		break;
7048 	}
7049 	default: {
7050 		int i;
7051 
7052 		page_len = 0;
7053 
7054 		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
7055 			/* Look for the right page code */
7056 			if ((lun->mode_pages.index[i].page_code &
7057 			     SMPH_PC_MASK) != page_code)
7058 				continue;
7059 
7060 			/* Look for the right subpage or the subpage wildcard*/
7061 			if ((lun->mode_pages.index[i].subpage != subpage)
7062 			 && (subpage != SMS_SUBPAGE_ALL))
7063 				continue;
7064 
7065 			/* Make sure the page is supported for this dev type */
7066 			if ((control_dev != 0)
7067 			 && (lun->mode_pages.index[i].page_flags &
7068 			     CTL_PAGE_FLAG_DISK_ONLY))
7069 				continue;
7070 
7071 #if 0
7072 			printf("found page %#x len %d\n",
7073 			       lun->mode_pages.index[i].page_code &
7074 			       SMPH_PC_MASK,
7075 			       lun->mode_pages.index[i].page_len);
7076 #endif
7077 
7078 			page_len += lun->mode_pages.index[i].page_len;
7079 		}
7080 
7081 		if (page_len == 0) {
7082 			ctl_set_invalid_field(ctsio,
7083 					      /*sks_valid*/ 1,
7084 					      /*command*/ 1,
7085 					      /*field*/ 2,
7086 					      /*bit_valid*/ 1,
7087 					      /*bit*/ 5);
7088 			ctl_done((union ctl_io *)ctsio);
7089 			return (CTL_RETVAL_COMPLETE);
7090 		}
7091 		break;
7092 	}
7093 	}
7094 
7095 	total_len = header_len + page_len;
7096 #if 0
7097 	printf("header_len = %d, page_len = %d, total_len = %d\n",
7098 	       header_len, page_len, total_len);
7099 #endif
7100 
7101 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7102 	ctsio->kern_sg_entries = 0;
7103 	ctsio->kern_data_resid = 0;
7104 	ctsio->kern_rel_offset = 0;
7105 	if (total_len < alloc_len) {
7106 		ctsio->residual = alloc_len - total_len;
7107 		ctsio->kern_data_len = total_len;
7108 		ctsio->kern_total_len = total_len;
7109 	} else {
7110 		ctsio->residual = 0;
7111 		ctsio->kern_data_len = alloc_len;
7112 		ctsio->kern_total_len = alloc_len;
7113 	}
7114 
7115 	switch (ctsio->cdb[0]) {
7116 	case MODE_SENSE_6: {
7117 		struct scsi_mode_hdr_6 *header;
7118 
7119 		header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
7120 
7121 		header->datalen = ctl_min(total_len - 1, 254);
7122 		if (control_dev == 0) {
7123 			header->dev_specific = 0x10; /* DPOFUA */
7124 			if ((lun->flags & CTL_LUN_READONLY) ||
7125 			    (lun->mode_pages.control_page[CTL_PAGE_CURRENT]
7126 			    .eca_and_aen & SCP_SWP) != 0)
7127 				    header->dev_specific |= 0x80; /* WP */
7128 		}
7129 		if (dbd)
7130 			header->block_descr_len = 0;
7131 		else
7132 			header->block_descr_len =
7133 				sizeof(struct scsi_mode_block_descr);
7134 		block_desc = (struct scsi_mode_block_descr *)&header[1];
7135 		break;
7136 	}
7137 	case MODE_SENSE_10: {
7138 		struct scsi_mode_hdr_10 *header;
7139 		int datalen;
7140 
7141 		header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
7142 
7143 		datalen = ctl_min(total_len - 2, 65533);
7144 		scsi_ulto2b(datalen, header->datalen);
7145 		if (control_dev == 0) {
7146 			header->dev_specific = 0x10; /* DPOFUA */
7147 			if ((lun->flags & CTL_LUN_READONLY) ||
7148 			    (lun->mode_pages.control_page[CTL_PAGE_CURRENT]
7149 			    .eca_and_aen & SCP_SWP) != 0)
7150 				    header->dev_specific |= 0x80; /* WP */
7151 		}
7152 		if (dbd)
7153 			scsi_ulto2b(0, header->block_descr_len);
7154 		else
7155 			scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
7156 				    header->block_descr_len);
7157 		block_desc = (struct scsi_mode_block_descr *)&header[1];
7158 		break;
7159 	}
7160 	default:
7161 		panic("invalid CDB type %#x", ctsio->cdb[0]);
7162 		break; /* NOTREACHED */
7163 	}
7164 
7165 	/*
7166 	 * If we've got a disk, use its blocksize in the block
7167 	 * descriptor.  Otherwise, just set it to 0.
7168 	 */
7169 	if (dbd == 0) {
7170 		if (control_dev == 0)
7171 			scsi_ulto3b(lun->be_lun->blocksize,
7172 				    block_desc->block_len);
7173 		else
7174 			scsi_ulto3b(0, block_desc->block_len);
7175 	}
7176 
7177 	switch (page_code) {
7178 	case SMS_ALL_PAGES_PAGE: {
7179 		int i, data_used;
7180 
7181 		data_used = header_len;
7182 		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
7183 			struct ctl_page_index *page_index;
7184 
7185 			page_index = &lun->mode_pages.index[i];
7186 
7187 			if ((control_dev != 0)
7188 			 && (page_index->page_flags &
7189 			    CTL_PAGE_FLAG_DISK_ONLY))
7190 				continue;
7191 
7192 			/*
7193 			 * We don't use this subpage if the user didn't
7194 			 * request all subpages.  We already checked (above)
7195 			 * to make sure the user only specified a subpage
7196 			 * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
7197 			 */
7198 			if ((page_index->subpage != 0)
7199 			 && (subpage == SMS_SUBPAGE_PAGE_0))
7200 				continue;
7201 
7202 			/*
7203 			 * Call the handler, if it exists, to update the
7204 			 * page to the latest values.
7205 			 */
7206 			if (page_index->sense_handler != NULL)
7207 				page_index->sense_handler(ctsio, page_index,pc);
7208 
7209 			memcpy(ctsio->kern_data_ptr + data_used,
7210 			       page_index->page_data +
7211 			       (page_index->page_len * pc),
7212 			       page_index->page_len);
7213 			data_used += page_index->page_len;
7214 		}
7215 		break;
7216 	}
7217 	default: {
7218 		int i, data_used;
7219 
7220 		data_used = header_len;
7221 
7222 		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
7223 			struct ctl_page_index *page_index;
7224 
7225 			page_index = &lun->mode_pages.index[i];
7226 
7227 			/* Look for the right page code */
7228 			if ((page_index->page_code & SMPH_PC_MASK) != page_code)
7229 				continue;
7230 
7231 			/* Look for the right subpage or the subpage wildcard*/
7232 			if ((page_index->subpage != subpage)
7233 			 && (subpage != SMS_SUBPAGE_ALL))
7234 				continue;
7235 
7236 			/* Make sure the page is supported for this dev type */
7237 			if ((control_dev != 0)
7238 			 && (page_index->page_flags &
7239 			     CTL_PAGE_FLAG_DISK_ONLY))
7240 				continue;
7241 
7242 			/*
7243 			 * Call the handler, if it exists, to update the
7244 			 * page to the latest values.
7245 			 */
7246 			if (page_index->sense_handler != NULL)
7247 				page_index->sense_handler(ctsio, page_index,pc);
7248 
7249 			memcpy(ctsio->kern_data_ptr + data_used,
7250 			       page_index->page_data +
7251 			       (page_index->page_len * pc),
7252 			       page_index->page_len);
7253 			data_used += page_index->page_len;
7254 		}
7255 		break;
7256 	}
7257 	}
7258 
7259 	ctsio->scsi_status = SCSI_STATUS_OK;
7260 
7261 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7262 	ctsio->be_move_done = ctl_config_move_done;
7263 	ctl_datamove((union ctl_io *)ctsio);
7264 
7265 	return (CTL_RETVAL_COMPLETE);
7266 }
7267 
7268 int
7269 ctl_read_capacity(struct ctl_scsiio *ctsio)
7270 {
7271 	struct scsi_read_capacity *cdb;
7272 	struct scsi_read_capacity_data *data;
7273 	struct ctl_lun *lun;
7274 	uint32_t lba;
7275 
7276 	CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
7277 
7278 	cdb = (struct scsi_read_capacity *)ctsio->cdb;
7279 
7280 	lba = scsi_4btoul(cdb->addr);
7281 	if (((cdb->pmi & SRC_PMI) == 0)
7282 	 && (lba != 0)) {
7283 		ctl_set_invalid_field(/*ctsio*/ ctsio,
7284 				      /*sks_valid*/ 1,
7285 				      /*command*/ 1,
7286 				      /*field*/ 2,
7287 				      /*bit_valid*/ 0,
7288 				      /*bit*/ 0);
7289 		ctl_done((union ctl_io *)ctsio);
7290 		return (CTL_RETVAL_COMPLETE);
7291 	}
7292 
7293 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7294 
7295 	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7296 	data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
7297 	ctsio->residual = 0;
7298 	ctsio->kern_data_len = sizeof(*data);
7299 	ctsio->kern_total_len = sizeof(*data);
7300 	ctsio->kern_data_resid = 0;
7301 	ctsio->kern_rel_offset = 0;
7302 	ctsio->kern_sg_entries = 0;
7303 
7304 	/*
7305 	 * If the maximum LBA is greater than 0xfffffffe, the user must
7306 	 * issue a SERVICE ACTION IN (16) command, with the read capacity
7307 	 * serivce action set.
7308 	 */
7309 	if (lun->be_lun->maxlba > 0xfffffffe)
7310 		scsi_ulto4b(0xffffffff, data->addr);
7311 	else
7312 		scsi_ulto4b(lun->be_lun->maxlba, data->addr);
7313 
7314 	/*
7315 	 * XXX KDM this may not be 512 bytes...
7316 	 */
7317 	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7318 
7319 	ctsio->scsi_status = SCSI_STATUS_OK;
7320 
7321 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7322 	ctsio->be_move_done = ctl_config_move_done;
7323 	ctl_datamove((union ctl_io *)ctsio);
7324 
7325 	return (CTL_RETVAL_COMPLETE);
7326 }
7327 
7328 int
7329 ctl_read_capacity_16(struct ctl_scsiio *ctsio)
7330 {
7331 	struct scsi_read_capacity_16 *cdb;
7332 	struct scsi_read_capacity_data_long *data;
7333 	struct ctl_lun *lun;
7334 	uint64_t lba;
7335 	uint32_t alloc_len;
7336 
7337 	CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
7338 
7339 	cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
7340 
7341 	alloc_len = scsi_4btoul(cdb->alloc_len);
7342 	lba = scsi_8btou64(cdb->addr);
7343 
7344 	if ((cdb->reladr & SRC16_PMI)
7345 	 && (lba != 0)) {
7346 		ctl_set_invalid_field(/*ctsio*/ ctsio,
7347 				      /*sks_valid*/ 1,
7348 				      /*command*/ 1,
7349 				      /*field*/ 2,
7350 				      /*bit_valid*/ 0,
7351 				      /*bit*/ 0);
7352 		ctl_done((union ctl_io *)ctsio);
7353 		return (CTL_RETVAL_COMPLETE);
7354 	}
7355 
7356 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7357 
7358 	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7359 	data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
7360 
7361 	if (sizeof(*data) < alloc_len) {
7362 		ctsio->residual = alloc_len - sizeof(*data);
7363 		ctsio->kern_data_len = sizeof(*data);
7364 		ctsio->kern_total_len = sizeof(*data);
7365 	} else {
7366 		ctsio->residual = 0;
7367 		ctsio->kern_data_len = alloc_len;
7368 		ctsio->kern_total_len = alloc_len;
7369 	}
7370 	ctsio->kern_data_resid = 0;
7371 	ctsio->kern_rel_offset = 0;
7372 	ctsio->kern_sg_entries = 0;
7373 
7374 	scsi_u64to8b(lun->be_lun->maxlba, data->addr);
7375 	/* XXX KDM this may not be 512 bytes... */
7376 	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7377 	data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
7378 	scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
7379 	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
7380 		data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ;
7381 
7382 	ctsio->scsi_status = SCSI_STATUS_OK;
7383 
7384 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7385 	ctsio->be_move_done = ctl_config_move_done;
7386 	ctl_datamove((union ctl_io *)ctsio);
7387 
7388 	return (CTL_RETVAL_COMPLETE);
7389 }
7390 
7391 int
7392 ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7393 {
7394 	struct scsi_maintenance_in *cdb;
7395 	int retval;
7396 	int alloc_len, ext, total_len = 0, g, p, pc, pg;
7397 	int num_target_port_groups, num_target_ports, single;
7398 	struct ctl_lun *lun;
7399 	struct ctl_softc *softc;
7400 	struct ctl_port *port;
7401 	struct scsi_target_group_data *rtg_ptr;
7402 	struct scsi_target_group_data_extended *rtg_ext_ptr;
7403 	struct scsi_target_port_group_descriptor *tpg_desc;
7404 
7405 	CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7406 
7407 	cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7408 	softc = control_softc;
7409 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7410 
7411 	retval = CTL_RETVAL_COMPLETE;
7412 
7413 	switch (cdb->byte2 & STG_PDF_MASK) {
7414 	case STG_PDF_LENGTH:
7415 		ext = 0;
7416 		break;
7417 	case STG_PDF_EXTENDED:
7418 		ext = 1;
7419 		break;
7420 	default:
7421 		ctl_set_invalid_field(/*ctsio*/ ctsio,
7422 				      /*sks_valid*/ 1,
7423 				      /*command*/ 1,
7424 				      /*field*/ 2,
7425 				      /*bit_valid*/ 1,
7426 				      /*bit*/ 5);
7427 		ctl_done((union ctl_io *)ctsio);
7428 		return(retval);
7429 	}
7430 
7431 	single = ctl_is_single;
7432 	if (single)
7433 		num_target_port_groups = 1;
7434 	else
7435 		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
7436 	num_target_ports = 0;
7437 	mtx_lock(&softc->ctl_lock);
7438 	STAILQ_FOREACH(port, &softc->port_list, links) {
7439 		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7440 			continue;
7441 		if (ctl_map_lun_back(port->targ_port, lun->lun) >= CTL_MAX_LUNS)
7442 			continue;
7443 		num_target_ports++;
7444 	}
7445 	mtx_unlock(&softc->ctl_lock);
7446 
7447 	if (ext)
7448 		total_len = sizeof(struct scsi_target_group_data_extended);
7449 	else
7450 		total_len = sizeof(struct scsi_target_group_data);
7451 	total_len += sizeof(struct scsi_target_port_group_descriptor) *
7452 		num_target_port_groups +
7453 	    sizeof(struct scsi_target_port_descriptor) *
7454 		num_target_ports * num_target_port_groups;
7455 
7456 	alloc_len = scsi_4btoul(cdb->length);
7457 
7458 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7459 
7460 	ctsio->kern_sg_entries = 0;
7461 
7462 	if (total_len < alloc_len) {
7463 		ctsio->residual = alloc_len - total_len;
7464 		ctsio->kern_data_len = total_len;
7465 		ctsio->kern_total_len = total_len;
7466 	} else {
7467 		ctsio->residual = 0;
7468 		ctsio->kern_data_len = alloc_len;
7469 		ctsio->kern_total_len = alloc_len;
7470 	}
7471 	ctsio->kern_data_resid = 0;
7472 	ctsio->kern_rel_offset = 0;
7473 
7474 	if (ext) {
7475 		rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7476 		    ctsio->kern_data_ptr;
7477 		scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7478 		rtg_ext_ptr->format_type = 0x10;
7479 		rtg_ext_ptr->implicit_transition_time = 0;
7480 		tpg_desc = &rtg_ext_ptr->groups[0];
7481 	} else {
7482 		rtg_ptr = (struct scsi_target_group_data *)
7483 		    ctsio->kern_data_ptr;
7484 		scsi_ulto4b(total_len - 4, rtg_ptr->length);
7485 		tpg_desc = &rtg_ptr->groups[0];
7486 	}
7487 
7488 	pg = ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS;
7489 	mtx_lock(&softc->ctl_lock);
7490 	for (g = 0; g < num_target_port_groups; g++) {
7491 		if (g == pg)
7492 			tpg_desc->pref_state = TPG_PRIMARY |
7493 			    TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7494 		else
7495 			tpg_desc->pref_state =
7496 			    TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7497 		tpg_desc->support = TPG_AO_SUP;
7498 		if (!single)
7499 			tpg_desc->support |= TPG_AN_SUP;
7500 		scsi_ulto2b(g + 1, tpg_desc->target_port_group);
7501 		tpg_desc->status = TPG_IMPLICIT;
7502 		pc = 0;
7503 		STAILQ_FOREACH(port, &softc->port_list, links) {
7504 			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7505 				continue;
7506 			if (ctl_map_lun_back(port->targ_port, lun->lun) >=
7507 			    CTL_MAX_LUNS)
7508 				continue;
7509 			p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
7510 			scsi_ulto2b(p, tpg_desc->descriptors[pc].
7511 			    relative_target_port_identifier);
7512 			pc++;
7513 		}
7514 		tpg_desc->target_port_count = pc;
7515 		tpg_desc = (struct scsi_target_port_group_descriptor *)
7516 		    &tpg_desc->descriptors[pc];
7517 	}
7518 	mtx_unlock(&softc->ctl_lock);
7519 
7520 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7521 	ctsio->be_move_done = ctl_config_move_done;
7522 
7523 	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7524 			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7525 			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7526 			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7527 			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7528 
7529 	ctl_datamove((union ctl_io *)ctsio);
7530 	return(retval);
7531 }
7532 
7533 int
7534 ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7535 {
7536 	struct ctl_lun *lun;
7537 	struct scsi_report_supported_opcodes *cdb;
7538 	const struct ctl_cmd_entry *entry, *sentry;
7539 	struct scsi_report_supported_opcodes_all *all;
7540 	struct scsi_report_supported_opcodes_descr *descr;
7541 	struct scsi_report_supported_opcodes_one *one;
7542 	int retval;
7543 	int alloc_len, total_len;
7544 	int opcode, service_action, i, j, num;
7545 
7546 	CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7547 
7548 	cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7549 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7550 
7551 	retval = CTL_RETVAL_COMPLETE;
7552 
7553 	opcode = cdb->requested_opcode;
7554 	service_action = scsi_2btoul(cdb->requested_service_action);
7555 	switch (cdb->options & RSO_OPTIONS_MASK) {
7556 	case RSO_OPTIONS_ALL:
7557 		num = 0;
7558 		for (i = 0; i < 256; i++) {
7559 			entry = &ctl_cmd_table[i];
7560 			if (entry->flags & CTL_CMD_FLAG_SA5) {
7561 				for (j = 0; j < 32; j++) {
7562 					sentry = &((const struct ctl_cmd_entry *)
7563 					    entry->execute)[j];
7564 					if (ctl_cmd_applicable(
7565 					    lun->be_lun->lun_type, sentry))
7566 						num++;
7567 				}
7568 			} else {
7569 				if (ctl_cmd_applicable(lun->be_lun->lun_type,
7570 				    entry))
7571 					num++;
7572 			}
7573 		}
7574 		total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7575 		    num * sizeof(struct scsi_report_supported_opcodes_descr);
7576 		break;
7577 	case RSO_OPTIONS_OC:
7578 		if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7579 			ctl_set_invalid_field(/*ctsio*/ ctsio,
7580 					      /*sks_valid*/ 1,
7581 					      /*command*/ 1,
7582 					      /*field*/ 2,
7583 					      /*bit_valid*/ 1,
7584 					      /*bit*/ 2);
7585 			ctl_done((union ctl_io *)ctsio);
7586 			return (CTL_RETVAL_COMPLETE);
7587 		}
7588 		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7589 		break;
7590 	case RSO_OPTIONS_OC_SA:
7591 		if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7592 		    service_action >= 32) {
7593 			ctl_set_invalid_field(/*ctsio*/ ctsio,
7594 					      /*sks_valid*/ 1,
7595 					      /*command*/ 1,
7596 					      /*field*/ 2,
7597 					      /*bit_valid*/ 1,
7598 					      /*bit*/ 2);
7599 			ctl_done((union ctl_io *)ctsio);
7600 			return (CTL_RETVAL_COMPLETE);
7601 		}
7602 		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7603 		break;
7604 	default:
7605 		ctl_set_invalid_field(/*ctsio*/ ctsio,
7606 				      /*sks_valid*/ 1,
7607 				      /*command*/ 1,
7608 				      /*field*/ 2,
7609 				      /*bit_valid*/ 1,
7610 				      /*bit*/ 2);
7611 		ctl_done((union ctl_io *)ctsio);
7612 		return (CTL_RETVAL_COMPLETE);
7613 	}
7614 
7615 	alloc_len = scsi_4btoul(cdb->length);
7616 
7617 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7618 
7619 	ctsio->kern_sg_entries = 0;
7620 
7621 	if (total_len < alloc_len) {
7622 		ctsio->residual = alloc_len - total_len;
7623 		ctsio->kern_data_len = total_len;
7624 		ctsio->kern_total_len = total_len;
7625 	} else {
7626 		ctsio->residual = 0;
7627 		ctsio->kern_data_len = alloc_len;
7628 		ctsio->kern_total_len = alloc_len;
7629 	}
7630 	ctsio->kern_data_resid = 0;
7631 	ctsio->kern_rel_offset = 0;
7632 
7633 	switch (cdb->options & RSO_OPTIONS_MASK) {
7634 	case RSO_OPTIONS_ALL:
7635 		all = (struct scsi_report_supported_opcodes_all *)
7636 		    ctsio->kern_data_ptr;
7637 		num = 0;
7638 		for (i = 0; i < 256; i++) {
7639 			entry = &ctl_cmd_table[i];
7640 			if (entry->flags & CTL_CMD_FLAG_SA5) {
7641 				for (j = 0; j < 32; j++) {
7642 					sentry = &((const struct ctl_cmd_entry *)
7643 					    entry->execute)[j];
7644 					if (!ctl_cmd_applicable(
7645 					    lun->be_lun->lun_type, sentry))
7646 						continue;
7647 					descr = &all->descr[num++];
7648 					descr->opcode = i;
7649 					scsi_ulto2b(j, descr->service_action);
7650 					descr->flags = RSO_SERVACTV;
7651 					scsi_ulto2b(sentry->length,
7652 					    descr->cdb_length);
7653 				}
7654 			} else {
7655 				if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7656 				    entry))
7657 					continue;
7658 				descr = &all->descr[num++];
7659 				descr->opcode = i;
7660 				scsi_ulto2b(0, descr->service_action);
7661 				descr->flags = 0;
7662 				scsi_ulto2b(entry->length, descr->cdb_length);
7663 			}
7664 		}
7665 		scsi_ulto4b(
7666 		    num * sizeof(struct scsi_report_supported_opcodes_descr),
7667 		    all->length);
7668 		break;
7669 	case RSO_OPTIONS_OC:
7670 		one = (struct scsi_report_supported_opcodes_one *)
7671 		    ctsio->kern_data_ptr;
7672 		entry = &ctl_cmd_table[opcode];
7673 		goto fill_one;
7674 	case RSO_OPTIONS_OC_SA:
7675 		one = (struct scsi_report_supported_opcodes_one *)
7676 		    ctsio->kern_data_ptr;
7677 		entry = &ctl_cmd_table[opcode];
7678 		entry = &((const struct ctl_cmd_entry *)
7679 		    entry->execute)[service_action];
7680 fill_one:
7681 		if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7682 			one->support = 3;
7683 			scsi_ulto2b(entry->length, one->cdb_length);
7684 			one->cdb_usage[0] = opcode;
7685 			memcpy(&one->cdb_usage[1], entry->usage,
7686 			    entry->length - 1);
7687 		} else
7688 			one->support = 1;
7689 		break;
7690 	}
7691 
7692 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7693 	ctsio->be_move_done = ctl_config_move_done;
7694 
7695 	ctl_datamove((union ctl_io *)ctsio);
7696 	return(retval);
7697 }
7698 
7699 int
7700 ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7701 {
7702 	struct ctl_lun *lun;
7703 	struct scsi_report_supported_tmf *cdb;
7704 	struct scsi_report_supported_tmf_data *data;
7705 	int retval;
7706 	int alloc_len, total_len;
7707 
7708 	CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7709 
7710 	cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7711 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7712 
7713 	retval = CTL_RETVAL_COMPLETE;
7714 
7715 	total_len = sizeof(struct scsi_report_supported_tmf_data);
7716 	alloc_len = scsi_4btoul(cdb->length);
7717 
7718 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7719 
7720 	ctsio->kern_sg_entries = 0;
7721 
7722 	if (total_len < alloc_len) {
7723 		ctsio->residual = alloc_len - total_len;
7724 		ctsio->kern_data_len = total_len;
7725 		ctsio->kern_total_len = total_len;
7726 	} else {
7727 		ctsio->residual = 0;
7728 		ctsio->kern_data_len = alloc_len;
7729 		ctsio->kern_total_len = alloc_len;
7730 	}
7731 	ctsio->kern_data_resid = 0;
7732 	ctsio->kern_rel_offset = 0;
7733 
7734 	data = (struct scsi_report_supported_tmf_data *)ctsio->kern_data_ptr;
7735 	data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_TRS;
7736 	data->byte2 |= RST_ITNRS;
7737 
7738 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7739 	ctsio->be_move_done = ctl_config_move_done;
7740 
7741 	ctl_datamove((union ctl_io *)ctsio);
7742 	return (retval);
7743 }
7744 
7745 int
7746 ctl_report_timestamp(struct ctl_scsiio *ctsio)
7747 {
7748 	struct ctl_lun *lun;
7749 	struct scsi_report_timestamp *cdb;
7750 	struct scsi_report_timestamp_data *data;
7751 	struct timeval tv;
7752 	int64_t timestamp;
7753 	int retval;
7754 	int alloc_len, total_len;
7755 
7756 	CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7757 
7758 	cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7759 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7760 
7761 	retval = CTL_RETVAL_COMPLETE;
7762 
7763 	total_len = sizeof(struct scsi_report_timestamp_data);
7764 	alloc_len = scsi_4btoul(cdb->length);
7765 
7766 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7767 
7768 	ctsio->kern_sg_entries = 0;
7769 
7770 	if (total_len < alloc_len) {
7771 		ctsio->residual = alloc_len - total_len;
7772 		ctsio->kern_data_len = total_len;
7773 		ctsio->kern_total_len = total_len;
7774 	} else {
7775 		ctsio->residual = 0;
7776 		ctsio->kern_data_len = alloc_len;
7777 		ctsio->kern_total_len = alloc_len;
7778 	}
7779 	ctsio->kern_data_resid = 0;
7780 	ctsio->kern_rel_offset = 0;
7781 
7782 	data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7783 	scsi_ulto2b(sizeof(*data) - 2, data->length);
7784 	data->origin = RTS_ORIG_OUTSIDE;
7785 	getmicrotime(&tv);
7786 	timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7787 	scsi_ulto4b(timestamp >> 16, data->timestamp);
7788 	scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7789 
7790 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7791 	ctsio->be_move_done = ctl_config_move_done;
7792 
7793 	ctl_datamove((union ctl_io *)ctsio);
7794 	return (retval);
7795 }
7796 
7797 int
7798 ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7799 {
7800 	struct scsi_per_res_in *cdb;
7801 	int alloc_len, total_len = 0;
7802 	/* struct scsi_per_res_in_rsrv in_data; */
7803 	struct ctl_lun *lun;
7804 	struct ctl_softc *softc;
7805 
7806 	CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7807 
7808 	softc = control_softc;
7809 
7810 	cdb = (struct scsi_per_res_in *)ctsio->cdb;
7811 
7812 	alloc_len = scsi_2btoul(cdb->length);
7813 
7814 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7815 
7816 retry:
7817 	mtx_lock(&lun->lun_lock);
7818 	switch (cdb->action) {
7819 	case SPRI_RK: /* read keys */
7820 		total_len = sizeof(struct scsi_per_res_in_keys) +
7821 			lun->pr_key_count *
7822 			sizeof(struct scsi_per_res_key);
7823 		break;
7824 	case SPRI_RR: /* read reservation */
7825 		if (lun->flags & CTL_LUN_PR_RESERVED)
7826 			total_len = sizeof(struct scsi_per_res_in_rsrv);
7827 		else
7828 			total_len = sizeof(struct scsi_per_res_in_header);
7829 		break;
7830 	case SPRI_RC: /* report capabilities */
7831 		total_len = sizeof(struct scsi_per_res_cap);
7832 		break;
7833 	case SPRI_RS: /* read full status */
7834 		total_len = sizeof(struct scsi_per_res_in_header) +
7835 		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7836 		    lun->pr_key_count;
7837 		break;
7838 	default:
7839 		panic("Invalid PR type %x", cdb->action);
7840 	}
7841 	mtx_unlock(&lun->lun_lock);
7842 
7843 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7844 
7845 	if (total_len < alloc_len) {
7846 		ctsio->residual = alloc_len - total_len;
7847 		ctsio->kern_data_len = total_len;
7848 		ctsio->kern_total_len = total_len;
7849 	} else {
7850 		ctsio->residual = 0;
7851 		ctsio->kern_data_len = alloc_len;
7852 		ctsio->kern_total_len = alloc_len;
7853 	}
7854 
7855 	ctsio->kern_data_resid = 0;
7856 	ctsio->kern_rel_offset = 0;
7857 	ctsio->kern_sg_entries = 0;
7858 
7859 	mtx_lock(&lun->lun_lock);
7860 	switch (cdb->action) {
7861 	case SPRI_RK: { // read keys
7862         struct scsi_per_res_in_keys *res_keys;
7863 		int i, key_count;
7864 
7865 		res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7866 
7867 		/*
7868 		 * We had to drop the lock to allocate our buffer, which
7869 		 * leaves time for someone to come in with another
7870 		 * persistent reservation.  (That is unlikely, though,
7871 		 * since this should be the only persistent reservation
7872 		 * command active right now.)
7873 		 */
7874 		if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7875 		    (lun->pr_key_count *
7876 		     sizeof(struct scsi_per_res_key)))){
7877 			mtx_unlock(&lun->lun_lock);
7878 			free(ctsio->kern_data_ptr, M_CTL);
7879 			printf("%s: reservation length changed, retrying\n",
7880 			       __func__);
7881 			goto retry;
7882 		}
7883 
7884 		scsi_ulto4b(lun->PRGeneration, res_keys->header.generation);
7885 
7886 		scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7887 			     lun->pr_key_count, res_keys->header.length);
7888 
7889 		for (i = 0, key_count = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7890 			if (lun->pr_keys[i] == 0)
7891 				continue;
7892 
7893 			/*
7894 			 * We used lun->pr_key_count to calculate the
7895 			 * size to allocate.  If it turns out the number of
7896 			 * initiators with the registered flag set is
7897 			 * larger than that (i.e. they haven't been kept in
7898 			 * sync), we've got a problem.
7899 			 */
7900 			if (key_count >= lun->pr_key_count) {
7901 #ifdef NEEDTOPORT
7902 				csevent_log(CSC_CTL | CSC_SHELF_SW |
7903 					    CTL_PR_ERROR,
7904 					    csevent_LogType_Fault,
7905 					    csevent_AlertLevel_Yellow,
7906 					    csevent_FRU_ShelfController,
7907 					    csevent_FRU_Firmware,
7908 				        csevent_FRU_Unknown,
7909 					    "registered keys %d >= key "
7910 					    "count %d", key_count,
7911 					    lun->pr_key_count);
7912 #endif
7913 				key_count++;
7914 				continue;
7915 			}
7916 			scsi_u64to8b(lun->pr_keys[i],
7917 			    res_keys->keys[key_count].key);
7918 			key_count++;
7919 		}
7920 		break;
7921 	}
7922 	case SPRI_RR: { // read reservation
7923 		struct scsi_per_res_in_rsrv *res;
7924 		int tmp_len, header_only;
7925 
7926 		res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7927 
7928 		scsi_ulto4b(lun->PRGeneration, res->header.generation);
7929 
7930 		if (lun->flags & CTL_LUN_PR_RESERVED)
7931 		{
7932 			tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7933 			scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7934 				    res->header.length);
7935 			header_only = 0;
7936 		} else {
7937 			tmp_len = sizeof(struct scsi_per_res_in_header);
7938 			scsi_ulto4b(0, res->header.length);
7939 			header_only = 1;
7940 		}
7941 
7942 		/*
7943 		 * We had to drop the lock to allocate our buffer, which
7944 		 * leaves time for someone to come in with another
7945 		 * persistent reservation.  (That is unlikely, though,
7946 		 * since this should be the only persistent reservation
7947 		 * command active right now.)
7948 		 */
7949 		if (tmp_len != total_len) {
7950 			mtx_unlock(&lun->lun_lock);
7951 			free(ctsio->kern_data_ptr, M_CTL);
7952 			printf("%s: reservation status changed, retrying\n",
7953 			       __func__);
7954 			goto retry;
7955 		}
7956 
7957 		/*
7958 		 * No reservation held, so we're done.
7959 		 */
7960 		if (header_only != 0)
7961 			break;
7962 
7963 		/*
7964 		 * If the registration is an All Registrants type, the key
7965 		 * is 0, since it doesn't really matter.
7966 		 */
7967 		if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7968 			scsi_u64to8b(lun->pr_keys[lun->pr_res_idx],
7969 			    res->data.reservation);
7970 		}
7971 		res->data.scopetype = lun->res_type;
7972 		break;
7973 	}
7974 	case SPRI_RC:     //report capabilities
7975 	{
7976 		struct scsi_per_res_cap *res_cap;
7977 		uint16_t type_mask;
7978 
7979 		res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7980 		scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7981 		res_cap->flags2 |= SPRI_TMV | SPRI_ALLOW_5;
7982 		type_mask = SPRI_TM_WR_EX_AR |
7983 			    SPRI_TM_EX_AC_RO |
7984 			    SPRI_TM_WR_EX_RO |
7985 			    SPRI_TM_EX_AC |
7986 			    SPRI_TM_WR_EX |
7987 			    SPRI_TM_EX_AC_AR;
7988 		scsi_ulto2b(type_mask, res_cap->type_mask);
7989 		break;
7990 	}
7991 	case SPRI_RS: { // read full status
7992 		struct scsi_per_res_in_full *res_status;
7993 		struct scsi_per_res_in_full_desc *res_desc;
7994 		struct ctl_port *port;
7995 		int i, len;
7996 
7997 		res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7998 
7999 		/*
8000 		 * We had to drop the lock to allocate our buffer, which
8001 		 * leaves time for someone to come in with another
8002 		 * persistent reservation.  (That is unlikely, though,
8003 		 * since this should be the only persistent reservation
8004 		 * command active right now.)
8005 		 */
8006 		if (total_len < (sizeof(struct scsi_per_res_in_header) +
8007 		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
8008 		     lun->pr_key_count)){
8009 			mtx_unlock(&lun->lun_lock);
8010 			free(ctsio->kern_data_ptr, M_CTL);
8011 			printf("%s: reservation length changed, retrying\n",
8012 			       __func__);
8013 			goto retry;
8014 		}
8015 
8016 		scsi_ulto4b(lun->PRGeneration, res_status->header.generation);
8017 
8018 		res_desc = &res_status->desc[0];
8019 		for (i = 0; i < 2*CTL_MAX_INITIATORS; i++) {
8020 			if (lun->pr_keys[i] == 0)
8021 				continue;
8022 
8023 			scsi_u64to8b(lun->pr_keys[i], res_desc->res_key.key);
8024 			if ((lun->flags & CTL_LUN_PR_RESERVED) &&
8025 			    (lun->pr_res_idx == i ||
8026 			     lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
8027 				res_desc->flags = SPRI_FULL_R_HOLDER;
8028 				res_desc->scopetype = lun->res_type;
8029 			}
8030 			scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
8031 			    res_desc->rel_trgt_port_id);
8032 			len = 0;
8033 			port = softc->ctl_ports[
8034 			    ctl_port_idx(i / CTL_MAX_INIT_PER_PORT)];
8035 			if (port != NULL)
8036 				len = ctl_create_iid(port,
8037 				    i % CTL_MAX_INIT_PER_PORT,
8038 				    res_desc->transport_id);
8039 			scsi_ulto4b(len, res_desc->additional_length);
8040 			res_desc = (struct scsi_per_res_in_full_desc *)
8041 			    &res_desc->transport_id[len];
8042 		}
8043 		scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
8044 		    res_status->header.length);
8045 		break;
8046 	}
8047 	default:
8048 		/*
8049 		 * This is a bug, because we just checked for this above,
8050 		 * and should have returned an error.
8051 		 */
8052 		panic("Invalid PR type %x", cdb->action);
8053 		break; /* NOTREACHED */
8054 	}
8055 	mtx_unlock(&lun->lun_lock);
8056 
8057 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8058 	ctsio->be_move_done = ctl_config_move_done;
8059 
8060 	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
8061 			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
8062 			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
8063 			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
8064 			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
8065 
8066 	ctl_datamove((union ctl_io *)ctsio);
8067 
8068 	return (CTL_RETVAL_COMPLETE);
8069 }
8070 
8071 /*
8072  * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
8073  * it should return.
8074  */
8075 static int
8076 ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
8077 		uint64_t sa_res_key, uint8_t type, uint32_t residx,
8078 		struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
8079 		struct scsi_per_res_out_parms* param)
8080 {
8081 	union ctl_ha_msg persis_io;
8082 	int retval, i;
8083 	int isc_retval;
8084 
8085 	retval = 0;
8086 
8087 	mtx_lock(&lun->lun_lock);
8088 	if (sa_res_key == 0) {
8089 		if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8090 			/* validate scope and type */
8091 			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
8092 			     SPR_LU_SCOPE) {
8093 				mtx_unlock(&lun->lun_lock);
8094 				ctl_set_invalid_field(/*ctsio*/ ctsio,
8095 						      /*sks_valid*/ 1,
8096 						      /*command*/ 1,
8097 						      /*field*/ 2,
8098 						      /*bit_valid*/ 1,
8099 						      /*bit*/ 4);
8100 				ctl_done((union ctl_io *)ctsio);
8101 				return (1);
8102 			}
8103 
8104 		        if (type>8 || type==2 || type==4 || type==0) {
8105 				mtx_unlock(&lun->lun_lock);
8106 				ctl_set_invalid_field(/*ctsio*/ ctsio,
8107        	           				      /*sks_valid*/ 1,
8108 						      /*command*/ 1,
8109 						      /*field*/ 2,
8110 						      /*bit_valid*/ 1,
8111 						      /*bit*/ 0);
8112 				ctl_done((union ctl_io *)ctsio);
8113 				return (1);
8114 		        }
8115 
8116 			/*
8117 			 * Unregister everybody else and build UA for
8118 			 * them
8119 			 */
8120 			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8121 				if (i == residx || lun->pr_keys[i] == 0)
8122 					continue;
8123 
8124 				if (!persis_offset
8125 				 && i <CTL_MAX_INITIATORS)
8126 					lun->pending_ua[i] |=
8127 						CTL_UA_REG_PREEMPT;
8128 				else if (persis_offset
8129 				      && i >= persis_offset)
8130 					lun->pending_ua[i-persis_offset] |=
8131 						CTL_UA_REG_PREEMPT;
8132 				lun->pr_keys[i] = 0;
8133 			}
8134 			lun->pr_key_count = 1;
8135 			lun->res_type = type;
8136 			if (lun->res_type != SPR_TYPE_WR_EX_AR
8137 			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8138 				lun->pr_res_idx = residx;
8139 
8140 			/* send msg to other side */
8141 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8142 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8143 			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8144 			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8145 			persis_io.pr.pr_info.res_type = type;
8146 			memcpy(persis_io.pr.pr_info.sa_res_key,
8147 			       param->serv_act_res_key,
8148 			       sizeof(param->serv_act_res_key));
8149 			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8150 			     &persis_io, sizeof(persis_io), 0)) >
8151 			     CTL_HA_STATUS_SUCCESS) {
8152 				printf("CTL:Persis Out error returned "
8153 				       "from ctl_ha_msg_send %d\n",
8154 				       isc_retval);
8155 			}
8156 		} else {
8157 			/* not all registrants */
8158 			mtx_unlock(&lun->lun_lock);
8159 			free(ctsio->kern_data_ptr, M_CTL);
8160 			ctl_set_invalid_field(ctsio,
8161 					      /*sks_valid*/ 1,
8162 					      /*command*/ 0,
8163 					      /*field*/ 8,
8164 					      /*bit_valid*/ 0,
8165 					      /*bit*/ 0);
8166 			ctl_done((union ctl_io *)ctsio);
8167 			return (1);
8168 		}
8169 	} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8170 		|| !(lun->flags & CTL_LUN_PR_RESERVED)) {
8171 		int found = 0;
8172 
8173 		if (res_key == sa_res_key) {
8174 			/* special case */
8175 			/*
8176 			 * The spec implies this is not good but doesn't
8177 			 * say what to do. There are two choices either
8178 			 * generate a res conflict or check condition
8179 			 * with illegal field in parameter data. Since
8180 			 * that is what is done when the sa_res_key is
8181 			 * zero I'll take that approach since this has
8182 			 * to do with the sa_res_key.
8183 			 */
8184 			mtx_unlock(&lun->lun_lock);
8185 			free(ctsio->kern_data_ptr, M_CTL);
8186 			ctl_set_invalid_field(ctsio,
8187 					      /*sks_valid*/ 1,
8188 					      /*command*/ 0,
8189 					      /*field*/ 8,
8190 					      /*bit_valid*/ 0,
8191 					      /*bit*/ 0);
8192 			ctl_done((union ctl_io *)ctsio);
8193 			return (1);
8194 		}
8195 
8196 		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8197 			if (lun->pr_keys[i] != sa_res_key)
8198 				continue;
8199 
8200 			found = 1;
8201 			lun->pr_keys[i] = 0;
8202 			lun->pr_key_count--;
8203 
8204 			if (!persis_offset && i < CTL_MAX_INITIATORS)
8205 				lun->pending_ua[i] |= CTL_UA_REG_PREEMPT;
8206 			else if (persis_offset && i >= persis_offset)
8207 				lun->pending_ua[i-persis_offset] |=
8208 					CTL_UA_REG_PREEMPT;
8209 		}
8210 		if (!found) {
8211 			mtx_unlock(&lun->lun_lock);
8212 			free(ctsio->kern_data_ptr, M_CTL);
8213 			ctl_set_reservation_conflict(ctsio);
8214 			ctl_done((union ctl_io *)ctsio);
8215 			return (CTL_RETVAL_COMPLETE);
8216 		}
8217 		/* send msg to other side */
8218 		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8219 		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8220 		persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8221 		persis_io.pr.pr_info.residx = lun->pr_res_idx;
8222 		persis_io.pr.pr_info.res_type = type;
8223 		memcpy(persis_io.pr.pr_info.sa_res_key,
8224 		       param->serv_act_res_key,
8225 		       sizeof(param->serv_act_res_key));
8226 		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8227 		     &persis_io, sizeof(persis_io), 0)) >
8228 		     CTL_HA_STATUS_SUCCESS) {
8229 			printf("CTL:Persis Out error returned from "
8230 			       "ctl_ha_msg_send %d\n", isc_retval);
8231 		}
8232 	} else {
8233 		/* Reserved but not all registrants */
8234 		/* sa_res_key is res holder */
8235 		if (sa_res_key == lun->pr_keys[lun->pr_res_idx]) {
8236 			/* validate scope and type */
8237 			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
8238 			     SPR_LU_SCOPE) {
8239 				mtx_unlock(&lun->lun_lock);
8240 				ctl_set_invalid_field(/*ctsio*/ ctsio,
8241 						      /*sks_valid*/ 1,
8242 						      /*command*/ 1,
8243 						      /*field*/ 2,
8244 						      /*bit_valid*/ 1,
8245 						      /*bit*/ 4);
8246 				ctl_done((union ctl_io *)ctsio);
8247 				return (1);
8248 			}
8249 
8250 			if (type>8 || type==2 || type==4 || type==0) {
8251 				mtx_unlock(&lun->lun_lock);
8252 				ctl_set_invalid_field(/*ctsio*/ ctsio,
8253 						      /*sks_valid*/ 1,
8254 						      /*command*/ 1,
8255 						      /*field*/ 2,
8256 						      /*bit_valid*/ 1,
8257 						      /*bit*/ 0);
8258 				ctl_done((union ctl_io *)ctsio);
8259 				return (1);
8260 			}
8261 
8262 			/*
8263 			 * Do the following:
8264 			 * if sa_res_key != res_key remove all
8265 			 * registrants w/sa_res_key and generate UA
8266 			 * for these registrants(Registrations
8267 			 * Preempted) if it wasn't an exclusive
8268 			 * reservation generate UA(Reservations
8269 			 * Preempted) for all other registered nexuses
8270 			 * if the type has changed. Establish the new
8271 			 * reservation and holder. If res_key and
8272 			 * sa_res_key are the same do the above
8273 			 * except don't unregister the res holder.
8274 			 */
8275 
8276 			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8277 				if (i == residx || lun->pr_keys[i] == 0)
8278 					continue;
8279 
8280 				if (sa_res_key == lun->pr_keys[i]) {
8281 					lun->pr_keys[i] = 0;
8282 					lun->pr_key_count--;
8283 
8284 					if (!persis_offset
8285 					 && i < CTL_MAX_INITIATORS)
8286 						lun->pending_ua[i] |=
8287 							CTL_UA_REG_PREEMPT;
8288 					else if (persis_offset
8289 					      && i >= persis_offset)
8290 						lun->pending_ua[i-persis_offset] |=
8291 						  CTL_UA_REG_PREEMPT;
8292 				} else if (type != lun->res_type
8293 					&& (lun->res_type == SPR_TYPE_WR_EX_RO
8294 					 || lun->res_type ==SPR_TYPE_EX_AC_RO)){
8295 						if (!persis_offset
8296 						 && i < CTL_MAX_INITIATORS)
8297 							lun->pending_ua[i] |=
8298 							CTL_UA_RES_RELEASE;
8299 						else if (persis_offset
8300 						      && i >= persis_offset)
8301 							lun->pending_ua[
8302 							i-persis_offset] |=
8303 							CTL_UA_RES_RELEASE;
8304 				}
8305 			}
8306 			lun->res_type = type;
8307 			if (lun->res_type != SPR_TYPE_WR_EX_AR
8308 			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8309 				lun->pr_res_idx = residx;
8310 			else
8311 				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8312 
8313 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8314 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8315 			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8316 			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8317 			persis_io.pr.pr_info.res_type = type;
8318 			memcpy(persis_io.pr.pr_info.sa_res_key,
8319 			       param->serv_act_res_key,
8320 			       sizeof(param->serv_act_res_key));
8321 			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8322 			     &persis_io, sizeof(persis_io), 0)) >
8323 			     CTL_HA_STATUS_SUCCESS) {
8324 				printf("CTL:Persis Out error returned "
8325 				       "from ctl_ha_msg_send %d\n",
8326 				       isc_retval);
8327 			}
8328 		} else {
8329 			/*
8330 			 * sa_res_key is not the res holder just
8331 			 * remove registrants
8332 			 */
8333 			int found=0;
8334 
8335 			for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8336 				if (sa_res_key != lun->pr_keys[i])
8337 					continue;
8338 
8339 				found = 1;
8340 				lun->pr_keys[i] = 0;
8341 				lun->pr_key_count--;
8342 
8343 				if (!persis_offset
8344 				 && i < CTL_MAX_INITIATORS)
8345 					lun->pending_ua[i] |=
8346 						CTL_UA_REG_PREEMPT;
8347 				else if (persis_offset
8348 				      && i >= persis_offset)
8349 					lun->pending_ua[i-persis_offset] |=
8350 						CTL_UA_REG_PREEMPT;
8351 			}
8352 
8353 			if (!found) {
8354 				mtx_unlock(&lun->lun_lock);
8355 				free(ctsio->kern_data_ptr, M_CTL);
8356 				ctl_set_reservation_conflict(ctsio);
8357 				ctl_done((union ctl_io *)ctsio);
8358 		        	return (1);
8359 			}
8360 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8361 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8362 			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8363 			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8364 			persis_io.pr.pr_info.res_type = type;
8365 			memcpy(persis_io.pr.pr_info.sa_res_key,
8366 			       param->serv_act_res_key,
8367 			       sizeof(param->serv_act_res_key));
8368 			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8369 			     &persis_io, sizeof(persis_io), 0)) >
8370 			     CTL_HA_STATUS_SUCCESS) {
8371 				printf("CTL:Persis Out error returned "
8372 				       "from ctl_ha_msg_send %d\n",
8373 				isc_retval);
8374 			}
8375 		}
8376 	}
8377 
8378 	lun->PRGeneration++;
8379 	mtx_unlock(&lun->lun_lock);
8380 
8381 	return (retval);
8382 }
8383 
8384 static void
8385 ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
8386 {
8387 	uint64_t sa_res_key;
8388 	int i;
8389 
8390 	sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
8391 
8392 	if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8393 	 || lun->pr_res_idx == CTL_PR_NO_RESERVATION
8394 	 || sa_res_key != lun->pr_keys[lun->pr_res_idx]) {
8395 		if (sa_res_key == 0) {
8396 			/*
8397 			 * Unregister everybody else and build UA for
8398 			 * them
8399 			 */
8400 			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8401 				if (i == msg->pr.pr_info.residx ||
8402 				    lun->pr_keys[i] == 0)
8403 					continue;
8404 
8405 				if (!persis_offset
8406 				 && i < CTL_MAX_INITIATORS)
8407 					lun->pending_ua[i] |=
8408 						CTL_UA_REG_PREEMPT;
8409 				else if (persis_offset && i >= persis_offset)
8410 					lun->pending_ua[i - persis_offset] |=
8411 						CTL_UA_REG_PREEMPT;
8412 				lun->pr_keys[i] = 0;
8413 			}
8414 
8415 			lun->pr_key_count = 1;
8416 			lun->res_type = msg->pr.pr_info.res_type;
8417 			if (lun->res_type != SPR_TYPE_WR_EX_AR
8418 			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8419 				lun->pr_res_idx = msg->pr.pr_info.residx;
8420 		} else {
8421 		        for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8422 				if (sa_res_key == lun->pr_keys[i])
8423 					continue;
8424 
8425 				lun->pr_keys[i] = 0;
8426 				lun->pr_key_count--;
8427 
8428 				if (!persis_offset
8429 				 && i < persis_offset)
8430 					lun->pending_ua[i] |=
8431 						CTL_UA_REG_PREEMPT;
8432 				else if (persis_offset
8433 				      && i >= persis_offset)
8434 					lun->pending_ua[i - persis_offset] |=
8435 						CTL_UA_REG_PREEMPT;
8436 			}
8437 		}
8438 	} else {
8439 		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8440 			if (i == msg->pr.pr_info.residx ||
8441 			    lun->pr_keys[i] == 0)
8442 				continue;
8443 
8444 			if (sa_res_key == lun->pr_keys[i]) {
8445 				lun->pr_keys[i] = 0;
8446 				lun->pr_key_count--;
8447 				if (!persis_offset
8448 				 && i < CTL_MAX_INITIATORS)
8449 					lun->pending_ua[i] |=
8450 						CTL_UA_REG_PREEMPT;
8451 				else if (persis_offset
8452 				      && i >= persis_offset)
8453 					lun->pending_ua[i - persis_offset] |=
8454 						CTL_UA_REG_PREEMPT;
8455 			} else if (msg->pr.pr_info.res_type != lun->res_type
8456 				&& (lun->res_type == SPR_TYPE_WR_EX_RO
8457 				 || lun->res_type == SPR_TYPE_EX_AC_RO)) {
8458 					if (!persis_offset
8459 					 && i < persis_offset)
8460 						lun->pending_ua[i] |=
8461 							CTL_UA_RES_RELEASE;
8462 					else if (persis_offset
8463 					      && i >= persis_offset)
8464 					lun->pending_ua[i - persis_offset] |=
8465 						CTL_UA_RES_RELEASE;
8466 			}
8467 		}
8468 		lun->res_type = msg->pr.pr_info.res_type;
8469 		if (lun->res_type != SPR_TYPE_WR_EX_AR
8470 		 && lun->res_type != SPR_TYPE_EX_AC_AR)
8471 			lun->pr_res_idx = msg->pr.pr_info.residx;
8472 		else
8473 			lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8474 	}
8475 	lun->PRGeneration++;
8476 
8477 }
8478 
8479 
8480 int
8481 ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8482 {
8483 	int retval;
8484 	int isc_retval;
8485 	u_int32_t param_len;
8486 	struct scsi_per_res_out *cdb;
8487 	struct ctl_lun *lun;
8488 	struct scsi_per_res_out_parms* param;
8489 	struct ctl_softc *softc;
8490 	uint32_t residx;
8491 	uint64_t res_key, sa_res_key;
8492 	uint8_t type;
8493 	union ctl_ha_msg persis_io;
8494 	int    i;
8495 
8496 	CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8497 
8498 	retval = CTL_RETVAL_COMPLETE;
8499 
8500 	softc = control_softc;
8501 
8502 	cdb = (struct scsi_per_res_out *)ctsio->cdb;
8503 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8504 
8505 	/*
8506 	 * We only support whole-LUN scope.  The scope & type are ignored for
8507 	 * register, register and ignore existing key and clear.
8508 	 * We sometimes ignore scope and type on preempts too!!
8509 	 * Verify reservation type here as well.
8510 	 */
8511 	type = cdb->scope_type & SPR_TYPE_MASK;
8512 	if ((cdb->action == SPRO_RESERVE)
8513 	 || (cdb->action == SPRO_RELEASE)) {
8514 		if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8515 			ctl_set_invalid_field(/*ctsio*/ ctsio,
8516 					      /*sks_valid*/ 1,
8517 					      /*command*/ 1,
8518 					      /*field*/ 2,
8519 					      /*bit_valid*/ 1,
8520 					      /*bit*/ 4);
8521 			ctl_done((union ctl_io *)ctsio);
8522 			return (CTL_RETVAL_COMPLETE);
8523 		}
8524 
8525 		if (type>8 || type==2 || type==4 || type==0) {
8526 			ctl_set_invalid_field(/*ctsio*/ ctsio,
8527 					      /*sks_valid*/ 1,
8528 					      /*command*/ 1,
8529 					      /*field*/ 2,
8530 					      /*bit_valid*/ 1,
8531 					      /*bit*/ 0);
8532 			ctl_done((union ctl_io *)ctsio);
8533 			return (CTL_RETVAL_COMPLETE);
8534 		}
8535 	}
8536 
8537 	param_len = scsi_4btoul(cdb->length);
8538 
8539 	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8540 		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8541 		ctsio->kern_data_len = param_len;
8542 		ctsio->kern_total_len = param_len;
8543 		ctsio->kern_data_resid = 0;
8544 		ctsio->kern_rel_offset = 0;
8545 		ctsio->kern_sg_entries = 0;
8546 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8547 		ctsio->be_move_done = ctl_config_move_done;
8548 		ctl_datamove((union ctl_io *)ctsio);
8549 
8550 		return (CTL_RETVAL_COMPLETE);
8551 	}
8552 
8553 	param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8554 
8555 	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8556 	res_key = scsi_8btou64(param->res_key.key);
8557 	sa_res_key = scsi_8btou64(param->serv_act_res_key);
8558 
8559 	/*
8560 	 * Validate the reservation key here except for SPRO_REG_IGNO
8561 	 * This must be done for all other service actions
8562 	 */
8563 	if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8564 		mtx_lock(&lun->lun_lock);
8565 		if (lun->pr_keys[residx] != 0) {
8566 		    if (res_key != lun->pr_keys[residx]) {
8567 				/*
8568 				 * The current key passed in doesn't match
8569 				 * the one the initiator previously
8570 				 * registered.
8571 				 */
8572 				mtx_unlock(&lun->lun_lock);
8573 				free(ctsio->kern_data_ptr, M_CTL);
8574 				ctl_set_reservation_conflict(ctsio);
8575 				ctl_done((union ctl_io *)ctsio);
8576 				return (CTL_RETVAL_COMPLETE);
8577 			}
8578 		} else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8579 			/*
8580 			 * We are not registered
8581 			 */
8582 			mtx_unlock(&lun->lun_lock);
8583 			free(ctsio->kern_data_ptr, M_CTL);
8584 			ctl_set_reservation_conflict(ctsio);
8585 			ctl_done((union ctl_io *)ctsio);
8586 			return (CTL_RETVAL_COMPLETE);
8587 		} else if (res_key != 0) {
8588 			/*
8589 			 * We are not registered and trying to register but
8590 			 * the register key isn't zero.
8591 			 */
8592 			mtx_unlock(&lun->lun_lock);
8593 			free(ctsio->kern_data_ptr, M_CTL);
8594 			ctl_set_reservation_conflict(ctsio);
8595 			ctl_done((union ctl_io *)ctsio);
8596 			return (CTL_RETVAL_COMPLETE);
8597 		}
8598 		mtx_unlock(&lun->lun_lock);
8599 	}
8600 
8601 	switch (cdb->action & SPRO_ACTION_MASK) {
8602 	case SPRO_REGISTER:
8603 	case SPRO_REG_IGNO: {
8604 
8605 #if 0
8606 		printf("Registration received\n");
8607 #endif
8608 
8609 		/*
8610 		 * We don't support any of these options, as we report in
8611 		 * the read capabilities request (see
8612 		 * ctl_persistent_reserve_in(), above).
8613 		 */
8614 		if ((param->flags & SPR_SPEC_I_PT)
8615 		 || (param->flags & SPR_ALL_TG_PT)
8616 		 || (param->flags & SPR_APTPL)) {
8617 			int bit_ptr;
8618 
8619 			if (param->flags & SPR_APTPL)
8620 				bit_ptr = 0;
8621 			else if (param->flags & SPR_ALL_TG_PT)
8622 				bit_ptr = 2;
8623 			else /* SPR_SPEC_I_PT */
8624 				bit_ptr = 3;
8625 
8626 			free(ctsio->kern_data_ptr, M_CTL);
8627 			ctl_set_invalid_field(ctsio,
8628 					      /*sks_valid*/ 1,
8629 					      /*command*/ 0,
8630 					      /*field*/ 20,
8631 					      /*bit_valid*/ 1,
8632 					      /*bit*/ bit_ptr);
8633 			ctl_done((union ctl_io *)ctsio);
8634 			return (CTL_RETVAL_COMPLETE);
8635 		}
8636 
8637 		mtx_lock(&lun->lun_lock);
8638 
8639 		/*
8640 		 * The initiator wants to clear the
8641 		 * key/unregister.
8642 		 */
8643 		if (sa_res_key == 0) {
8644 			if ((res_key == 0
8645 			  && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8646 			 || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8647 			  && lun->pr_keys[residx] == 0)) {
8648 				mtx_unlock(&lun->lun_lock);
8649 				goto done;
8650 			}
8651 
8652 			lun->pr_keys[residx] = 0;
8653 			lun->pr_key_count--;
8654 
8655 			if (residx == lun->pr_res_idx) {
8656 				lun->flags &= ~CTL_LUN_PR_RESERVED;
8657 				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8658 
8659 				if ((lun->res_type == SPR_TYPE_WR_EX_RO
8660 				  || lun->res_type == SPR_TYPE_EX_AC_RO)
8661 				 && lun->pr_key_count) {
8662 					/*
8663 					 * If the reservation is a registrants
8664 					 * only type we need to generate a UA
8665 					 * for other registered inits.  The
8666 					 * sense code should be RESERVATIONS
8667 					 * RELEASED
8668 					 */
8669 
8670 					for (i = 0; i < CTL_MAX_INITIATORS;i++){
8671 						if (lun->pr_keys[
8672 						    i + persis_offset] == 0)
8673 							continue;
8674 						lun->pending_ua[i] |=
8675 							CTL_UA_RES_RELEASE;
8676 					}
8677 				}
8678 				lun->res_type = 0;
8679 			} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8680 				if (lun->pr_key_count==0) {
8681 					lun->flags &= ~CTL_LUN_PR_RESERVED;
8682 					lun->res_type = 0;
8683 					lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8684 				}
8685 			}
8686 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8687 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8688 			persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8689 			persis_io.pr.pr_info.residx = residx;
8690 			if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8691 			     &persis_io, sizeof(persis_io), 0 )) >
8692 			     CTL_HA_STATUS_SUCCESS) {
8693 				printf("CTL:Persis Out error returned from "
8694 				       "ctl_ha_msg_send %d\n", isc_retval);
8695 			}
8696 		} else /* sa_res_key != 0 */ {
8697 
8698 			/*
8699 			 * If we aren't registered currently then increment
8700 			 * the key count and set the registered flag.
8701 			 */
8702 			if (lun->pr_keys[residx] == 0)
8703 				lun->pr_key_count++;
8704 			lun->pr_keys[residx] = sa_res_key;
8705 
8706 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8707 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8708 			persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8709 			persis_io.pr.pr_info.residx = residx;
8710 			memcpy(persis_io.pr.pr_info.sa_res_key,
8711 			       param->serv_act_res_key,
8712 			       sizeof(param->serv_act_res_key));
8713 			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8714 			     &persis_io, sizeof(persis_io), 0)) >
8715 			     CTL_HA_STATUS_SUCCESS) {
8716 				printf("CTL:Persis Out error returned from "
8717 				       "ctl_ha_msg_send %d\n", isc_retval);
8718 			}
8719 		}
8720 		lun->PRGeneration++;
8721 		mtx_unlock(&lun->lun_lock);
8722 
8723 		break;
8724 	}
8725 	case SPRO_RESERVE:
8726 #if 0
8727                 printf("Reserve executed type %d\n", type);
8728 #endif
8729 		mtx_lock(&lun->lun_lock);
8730 		if (lun->flags & CTL_LUN_PR_RESERVED) {
8731 			/*
8732 			 * if this isn't the reservation holder and it's
8733 			 * not a "all registrants" type or if the type is
8734 			 * different then we have a conflict
8735 			 */
8736 			if ((lun->pr_res_idx != residx
8737 			  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8738 			 || lun->res_type != type) {
8739 				mtx_unlock(&lun->lun_lock);
8740 				free(ctsio->kern_data_ptr, M_CTL);
8741 				ctl_set_reservation_conflict(ctsio);
8742 				ctl_done((union ctl_io *)ctsio);
8743 				return (CTL_RETVAL_COMPLETE);
8744 			}
8745 			mtx_unlock(&lun->lun_lock);
8746 		} else /* create a reservation */ {
8747 			/*
8748 			 * If it's not an "all registrants" type record
8749 			 * reservation holder
8750 			 */
8751 			if (type != SPR_TYPE_WR_EX_AR
8752 			 && type != SPR_TYPE_EX_AC_AR)
8753 				lun->pr_res_idx = residx; /* Res holder */
8754 			else
8755 				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8756 
8757 			lun->flags |= CTL_LUN_PR_RESERVED;
8758 			lun->res_type = type;
8759 
8760 			mtx_unlock(&lun->lun_lock);
8761 
8762 			/* send msg to other side */
8763 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8764 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8765 			persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8766 			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8767 			persis_io.pr.pr_info.res_type = type;
8768 			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8769 			     &persis_io, sizeof(persis_io), 0)) >
8770 			     CTL_HA_STATUS_SUCCESS) {
8771 				printf("CTL:Persis Out error returned from "
8772 				       "ctl_ha_msg_send %d\n", isc_retval);
8773 			}
8774 		}
8775 		break;
8776 
8777 	case SPRO_RELEASE:
8778 		mtx_lock(&lun->lun_lock);
8779 		if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8780 			/* No reservation exists return good status */
8781 			mtx_unlock(&lun->lun_lock);
8782 			goto done;
8783 		}
8784 		/*
8785 		 * Is this nexus a reservation holder?
8786 		 */
8787 		if (lun->pr_res_idx != residx
8788 		 && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8789 			/*
8790 			 * not a res holder return good status but
8791 			 * do nothing
8792 			 */
8793 			mtx_unlock(&lun->lun_lock);
8794 			goto done;
8795 		}
8796 
8797 		if (lun->res_type != type) {
8798 			mtx_unlock(&lun->lun_lock);
8799 			free(ctsio->kern_data_ptr, M_CTL);
8800 			ctl_set_illegal_pr_release(ctsio);
8801 			ctl_done((union ctl_io *)ctsio);
8802 			return (CTL_RETVAL_COMPLETE);
8803 		}
8804 
8805 		/* okay to release */
8806 		lun->flags &= ~CTL_LUN_PR_RESERVED;
8807 		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8808 		lun->res_type = 0;
8809 
8810 		/*
8811 		 * if this isn't an exclusive access
8812 		 * res generate UA for all other
8813 		 * registrants.
8814 		 */
8815 		if (type != SPR_TYPE_EX_AC
8816 		 && type != SPR_TYPE_WR_EX) {
8817 			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8818 				if (i == residx ||
8819 				    lun->pr_keys[i + persis_offset] == 0)
8820 					continue;
8821 				lun->pending_ua[i] |= CTL_UA_RES_RELEASE;
8822 			}
8823 		}
8824 		mtx_unlock(&lun->lun_lock);
8825 		/* Send msg to other side */
8826 		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8827 		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8828 		persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8829 		if ((isc_retval=ctl_ha_msg_send( CTL_HA_CHAN_CTL, &persis_io,
8830 		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8831 			printf("CTL:Persis Out error returned from "
8832 			       "ctl_ha_msg_send %d\n", isc_retval);
8833 		}
8834 		break;
8835 
8836 	case SPRO_CLEAR:
8837 		/* send msg to other side */
8838 
8839 		mtx_lock(&lun->lun_lock);
8840 		lun->flags &= ~CTL_LUN_PR_RESERVED;
8841 		lun->res_type = 0;
8842 		lun->pr_key_count = 0;
8843 		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8844 
8845 		lun->pr_keys[residx] = 0;
8846 
8847 		for (i=0; i < 2*CTL_MAX_INITIATORS; i++)
8848 			if (lun->pr_keys[i] != 0) {
8849 				if (!persis_offset && i < CTL_MAX_INITIATORS)
8850 					lun->pending_ua[i] |=
8851 						CTL_UA_RES_PREEMPT;
8852 				else if (persis_offset && i >= persis_offset)
8853 					lun->pending_ua[i-persis_offset] |=
8854 					    CTL_UA_RES_PREEMPT;
8855 
8856 				lun->pr_keys[i] = 0;
8857 			}
8858 		lun->PRGeneration++;
8859 		mtx_unlock(&lun->lun_lock);
8860 		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8861 		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8862 		persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8863 		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8864 		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8865 			printf("CTL:Persis Out error returned from "
8866 			       "ctl_ha_msg_send %d\n", isc_retval);
8867 		}
8868 		break;
8869 
8870 	case SPRO_PREEMPT: {
8871 		int nretval;
8872 
8873 		nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8874 					  residx, ctsio, cdb, param);
8875 		if (nretval != 0)
8876 			return (CTL_RETVAL_COMPLETE);
8877 		break;
8878 	}
8879 	default:
8880 		panic("Invalid PR type %x", cdb->action);
8881 	}
8882 
8883 done:
8884 	free(ctsio->kern_data_ptr, M_CTL);
8885 	ctl_set_success(ctsio);
8886 	ctl_done((union ctl_io *)ctsio);
8887 
8888 	return (retval);
8889 }
8890 
8891 /*
8892  * This routine is for handling a message from the other SC pertaining to
8893  * persistent reserve out. All the error checking will have been done
8894  * so only perorming the action need be done here to keep the two
8895  * in sync.
8896  */
8897 static void
8898 ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8899 {
8900 	struct ctl_lun *lun;
8901 	struct ctl_softc *softc;
8902 	int i;
8903 	uint32_t targ_lun;
8904 
8905 	softc = control_softc;
8906 
8907 	targ_lun = msg->hdr.nexus.targ_mapped_lun;
8908 	lun = softc->ctl_luns[targ_lun];
8909 	mtx_lock(&lun->lun_lock);
8910 	switch(msg->pr.pr_info.action) {
8911 	case CTL_PR_REG_KEY:
8912 		if (lun->pr_keys[msg->pr.pr_info.residx] == 0)
8913 			lun->pr_key_count++;
8914 		lun->pr_keys[msg->pr.pr_info.residx] =
8915 		    scsi_8btou64(msg->pr.pr_info.sa_res_key);
8916 		lun->PRGeneration++;
8917 		break;
8918 
8919 	case CTL_PR_UNREG_KEY:
8920 		lun->pr_keys[msg->pr.pr_info.residx] = 0;
8921 		lun->pr_key_count--;
8922 
8923 		/* XXX Need to see if the reservation has been released */
8924 		/* if so do we need to generate UA? */
8925 		if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8926 			lun->flags &= ~CTL_LUN_PR_RESERVED;
8927 			lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8928 
8929 			if ((lun->res_type == SPR_TYPE_WR_EX_RO
8930 			  || lun->res_type == SPR_TYPE_EX_AC_RO)
8931 			 && lun->pr_key_count) {
8932 				/*
8933 				 * If the reservation is a registrants
8934 				 * only type we need to generate a UA
8935 				 * for other registered inits.  The
8936 				 * sense code should be RESERVATIONS
8937 				 * RELEASED
8938 				 */
8939 
8940 				for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8941 					if (lun->pr_keys[i+
8942 					    persis_offset] == 0)
8943 						continue;
8944 
8945 					lun->pending_ua[i] |=
8946 						CTL_UA_RES_RELEASE;
8947 				}
8948 			}
8949 			lun->res_type = 0;
8950 		} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8951 			if (lun->pr_key_count==0) {
8952 				lun->flags &= ~CTL_LUN_PR_RESERVED;
8953 				lun->res_type = 0;
8954 				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8955 			}
8956 		}
8957 		lun->PRGeneration++;
8958 		break;
8959 
8960 	case CTL_PR_RESERVE:
8961 		lun->flags |= CTL_LUN_PR_RESERVED;
8962 		lun->res_type = msg->pr.pr_info.res_type;
8963 		lun->pr_res_idx = msg->pr.pr_info.residx;
8964 
8965 		break;
8966 
8967 	case CTL_PR_RELEASE:
8968 		/*
8969 		 * if this isn't an exclusive access res generate UA for all
8970 		 * other registrants.
8971 		 */
8972 		if (lun->res_type != SPR_TYPE_EX_AC
8973 		 && lun->res_type != SPR_TYPE_WR_EX) {
8974 			for (i = 0; i < CTL_MAX_INITIATORS; i++)
8975 				if (lun->pr_keys[i+persis_offset] != 0)
8976 					lun->pending_ua[i] |=
8977 						CTL_UA_RES_RELEASE;
8978 		}
8979 
8980 		lun->flags &= ~CTL_LUN_PR_RESERVED;
8981 		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8982 		lun->res_type = 0;
8983 		break;
8984 
8985 	case CTL_PR_PREEMPT:
8986 		ctl_pro_preempt_other(lun, msg);
8987 		break;
8988 	case CTL_PR_CLEAR:
8989 		lun->flags &= ~CTL_LUN_PR_RESERVED;
8990 		lun->res_type = 0;
8991 		lun->pr_key_count = 0;
8992 		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8993 
8994 		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8995 			if (lun->pr_keys[i] == 0)
8996 				continue;
8997 			if (!persis_offset
8998 			 && i < CTL_MAX_INITIATORS)
8999 				lun->pending_ua[i] |= CTL_UA_RES_PREEMPT;
9000 			else if (persis_offset
9001 			      && i >= persis_offset)
9002 				lun->pending_ua[i-persis_offset] |=
9003 					CTL_UA_RES_PREEMPT;
9004 			lun->pr_keys[i] = 0;
9005 		}
9006 		lun->PRGeneration++;
9007 		break;
9008 	}
9009 
9010 	mtx_unlock(&lun->lun_lock);
9011 }
9012 
9013 int
9014 ctl_read_write(struct ctl_scsiio *ctsio)
9015 {
9016 	struct ctl_lun *lun;
9017 	struct ctl_lba_len_flags *lbalen;
9018 	uint64_t lba;
9019 	uint32_t num_blocks;
9020 	int flags, retval;
9021 	int isread;
9022 
9023 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9024 
9025 	CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
9026 
9027 	flags = 0;
9028 	retval = CTL_RETVAL_COMPLETE;
9029 
9030 	isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
9031 	      || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
9032 	if (lun->flags & CTL_LUN_PR_RESERVED && isread) {
9033 		uint32_t residx;
9034 
9035 		/*
9036 		 * XXX KDM need a lock here.
9037 		 */
9038 		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
9039 		if ((lun->res_type == SPR_TYPE_EX_AC
9040 		  && residx != lun->pr_res_idx)
9041 		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
9042 		   || lun->res_type == SPR_TYPE_EX_AC_AR)
9043 		  && lun->pr_keys[residx] == 0)) {
9044 			ctl_set_reservation_conflict(ctsio);
9045 			ctl_done((union ctl_io *)ctsio);
9046 			return (CTL_RETVAL_COMPLETE);
9047 	        }
9048 	}
9049 
9050 	switch (ctsio->cdb[0]) {
9051 	case READ_6:
9052 	case WRITE_6: {
9053 		struct scsi_rw_6 *cdb;
9054 
9055 		cdb = (struct scsi_rw_6 *)ctsio->cdb;
9056 
9057 		lba = scsi_3btoul(cdb->addr);
9058 		/* only 5 bits are valid in the most significant address byte */
9059 		lba &= 0x1fffff;
9060 		num_blocks = cdb->length;
9061 		/*
9062 		 * This is correct according to SBC-2.
9063 		 */
9064 		if (num_blocks == 0)
9065 			num_blocks = 256;
9066 		break;
9067 	}
9068 	case READ_10:
9069 	case WRITE_10: {
9070 		struct scsi_rw_10 *cdb;
9071 
9072 		cdb = (struct scsi_rw_10 *)ctsio->cdb;
9073 		if (cdb->byte2 & SRW10_FUA)
9074 			flags |= CTL_LLF_FUA;
9075 		if (cdb->byte2 & SRW10_DPO)
9076 			flags |= CTL_LLF_DPO;
9077 		lba = scsi_4btoul(cdb->addr);
9078 		num_blocks = scsi_2btoul(cdb->length);
9079 		break;
9080 	}
9081 	case WRITE_VERIFY_10: {
9082 		struct scsi_write_verify_10 *cdb;
9083 
9084 		cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
9085 		flags |= CTL_LLF_FUA;
9086 		if (cdb->byte2 & SWV_DPO)
9087 			flags |= CTL_LLF_DPO;
9088 		lba = scsi_4btoul(cdb->addr);
9089 		num_blocks = scsi_2btoul(cdb->length);
9090 		break;
9091 	}
9092 	case READ_12:
9093 	case WRITE_12: {
9094 		struct scsi_rw_12 *cdb;
9095 
9096 		cdb = (struct scsi_rw_12 *)ctsio->cdb;
9097 		if (cdb->byte2 & SRW12_FUA)
9098 			flags |= CTL_LLF_FUA;
9099 		if (cdb->byte2 & SRW12_DPO)
9100 			flags |= CTL_LLF_DPO;
9101 		lba = scsi_4btoul(cdb->addr);
9102 		num_blocks = scsi_4btoul(cdb->length);
9103 		break;
9104 	}
9105 	case WRITE_VERIFY_12: {
9106 		struct scsi_write_verify_12 *cdb;
9107 
9108 		cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
9109 		flags |= CTL_LLF_FUA;
9110 		if (cdb->byte2 & SWV_DPO)
9111 			flags |= CTL_LLF_DPO;
9112 		lba = scsi_4btoul(cdb->addr);
9113 		num_blocks = scsi_4btoul(cdb->length);
9114 		break;
9115 	}
9116 	case READ_16:
9117 	case WRITE_16: {
9118 		struct scsi_rw_16 *cdb;
9119 
9120 		cdb = (struct scsi_rw_16 *)ctsio->cdb;
9121 		if (cdb->byte2 & SRW12_FUA)
9122 			flags |= CTL_LLF_FUA;
9123 		if (cdb->byte2 & SRW12_DPO)
9124 			flags |= CTL_LLF_DPO;
9125 		lba = scsi_8btou64(cdb->addr);
9126 		num_blocks = scsi_4btoul(cdb->length);
9127 		break;
9128 	}
9129 	case WRITE_ATOMIC_16: {
9130 		struct scsi_rw_16 *cdb;
9131 
9132 		if (lun->be_lun->atomicblock == 0) {
9133 			ctl_set_invalid_opcode(ctsio);
9134 			ctl_done((union ctl_io *)ctsio);
9135 			return (CTL_RETVAL_COMPLETE);
9136 		}
9137 
9138 		cdb = (struct scsi_rw_16 *)ctsio->cdb;
9139 		if (cdb->byte2 & SRW12_FUA)
9140 			flags |= CTL_LLF_FUA;
9141 		if (cdb->byte2 & SRW12_DPO)
9142 			flags |= CTL_LLF_DPO;
9143 		lba = scsi_8btou64(cdb->addr);
9144 		num_blocks = scsi_4btoul(cdb->length);
9145 		if (num_blocks > lun->be_lun->atomicblock) {
9146 			ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
9147 			    /*command*/ 1, /*field*/ 12, /*bit_valid*/ 0,
9148 			    /*bit*/ 0);
9149 			ctl_done((union ctl_io *)ctsio);
9150 			return (CTL_RETVAL_COMPLETE);
9151 		}
9152 		break;
9153 	}
9154 	case WRITE_VERIFY_16: {
9155 		struct scsi_write_verify_16 *cdb;
9156 
9157 		cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
9158 		flags |= CTL_LLF_FUA;
9159 		if (cdb->byte2 & SWV_DPO)
9160 			flags |= CTL_LLF_DPO;
9161 		lba = scsi_8btou64(cdb->addr);
9162 		num_blocks = scsi_4btoul(cdb->length);
9163 		break;
9164 	}
9165 	default:
9166 		/*
9167 		 * We got a command we don't support.  This shouldn't
9168 		 * happen, commands should be filtered out above us.
9169 		 */
9170 		ctl_set_invalid_opcode(ctsio);
9171 		ctl_done((union ctl_io *)ctsio);
9172 
9173 		return (CTL_RETVAL_COMPLETE);
9174 		break; /* NOTREACHED */
9175 	}
9176 
9177 	/*
9178 	 * The first check is to make sure we're in bounds, the second
9179 	 * check is to catch wrap-around problems.  If the lba + num blocks
9180 	 * is less than the lba, then we've wrapped around and the block
9181 	 * range is invalid anyway.
9182 	 */
9183 	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9184 	 || ((lba + num_blocks) < lba)) {
9185 		ctl_set_lba_out_of_range(ctsio);
9186 		ctl_done((union ctl_io *)ctsio);
9187 		return (CTL_RETVAL_COMPLETE);
9188 	}
9189 
9190 	/*
9191 	 * According to SBC-3, a transfer length of 0 is not an error.
9192 	 * Note that this cannot happen with WRITE(6) or READ(6), since 0
9193 	 * translates to 256 blocks for those commands.
9194 	 */
9195 	if (num_blocks == 0) {
9196 		ctl_set_success(ctsio);
9197 		ctl_done((union ctl_io *)ctsio);
9198 		return (CTL_RETVAL_COMPLETE);
9199 	}
9200 
9201 	/* Set FUA and/or DPO if caches are disabled. */
9202 	if (isread) {
9203 		if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9204 		    SCP_RCD) != 0)
9205 			flags |= CTL_LLF_FUA | CTL_LLF_DPO;
9206 	} else {
9207 		if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9208 		    SCP_WCE) == 0)
9209 			flags |= CTL_LLF_FUA;
9210 	}
9211 
9212 	lbalen = (struct ctl_lba_len_flags *)
9213 	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9214 	lbalen->lba = lba;
9215 	lbalen->len = num_blocks;
9216 	lbalen->flags = (isread ? CTL_LLF_READ : CTL_LLF_WRITE) | flags;
9217 
9218 	ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9219 	ctsio->kern_rel_offset = 0;
9220 
9221 	CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
9222 
9223 	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9224 
9225 	return (retval);
9226 }
9227 
9228 static int
9229 ctl_cnw_cont(union ctl_io *io)
9230 {
9231 	struct ctl_scsiio *ctsio;
9232 	struct ctl_lun *lun;
9233 	struct ctl_lba_len_flags *lbalen;
9234 	int retval;
9235 
9236 	ctsio = &io->scsiio;
9237 	ctsio->io_hdr.status = CTL_STATUS_NONE;
9238 	ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
9239 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9240 	lbalen = (struct ctl_lba_len_flags *)
9241 	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9242 	lbalen->flags &= ~CTL_LLF_COMPARE;
9243 	lbalen->flags |= CTL_LLF_WRITE;
9244 
9245 	CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
9246 	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9247 	return (retval);
9248 }
9249 
9250 int
9251 ctl_cnw(struct ctl_scsiio *ctsio)
9252 {
9253 	struct ctl_lun *lun;
9254 	struct ctl_lba_len_flags *lbalen;
9255 	uint64_t lba;
9256 	uint32_t num_blocks;
9257 	int flags, retval;
9258 
9259 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9260 
9261 	CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
9262 
9263 	flags = 0;
9264 	retval = CTL_RETVAL_COMPLETE;
9265 
9266 	switch (ctsio->cdb[0]) {
9267 	case COMPARE_AND_WRITE: {
9268 		struct scsi_compare_and_write *cdb;
9269 
9270 		cdb = (struct scsi_compare_and_write *)ctsio->cdb;
9271 		if (cdb->byte2 & SRW10_FUA)
9272 			flags |= CTL_LLF_FUA;
9273 		if (cdb->byte2 & SRW10_DPO)
9274 			flags |= CTL_LLF_DPO;
9275 		lba = scsi_8btou64(cdb->addr);
9276 		num_blocks = cdb->length;
9277 		break;
9278 	}
9279 	default:
9280 		/*
9281 		 * We got a command we don't support.  This shouldn't
9282 		 * happen, commands should be filtered out above us.
9283 		 */
9284 		ctl_set_invalid_opcode(ctsio);
9285 		ctl_done((union ctl_io *)ctsio);
9286 
9287 		return (CTL_RETVAL_COMPLETE);
9288 		break; /* NOTREACHED */
9289 	}
9290 
9291 	/*
9292 	 * The first check is to make sure we're in bounds, the second
9293 	 * check is to catch wrap-around problems.  If the lba + num blocks
9294 	 * is less than the lba, then we've wrapped around and the block
9295 	 * range is invalid anyway.
9296 	 */
9297 	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9298 	 || ((lba + num_blocks) < lba)) {
9299 		ctl_set_lba_out_of_range(ctsio);
9300 		ctl_done((union ctl_io *)ctsio);
9301 		return (CTL_RETVAL_COMPLETE);
9302 	}
9303 
9304 	/*
9305 	 * According to SBC-3, a transfer length of 0 is not an error.
9306 	 */
9307 	if (num_blocks == 0) {
9308 		ctl_set_success(ctsio);
9309 		ctl_done((union ctl_io *)ctsio);
9310 		return (CTL_RETVAL_COMPLETE);
9311 	}
9312 
9313 	/* Set FUA if write cache is disabled. */
9314 	if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9315 	    SCP_WCE) == 0)
9316 		flags |= CTL_LLF_FUA;
9317 
9318 	ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
9319 	ctsio->kern_rel_offset = 0;
9320 
9321 	/*
9322 	 * Set the IO_CONT flag, so that if this I/O gets passed to
9323 	 * ctl_data_submit_done(), it'll get passed back to
9324 	 * ctl_ctl_cnw_cont() for further processing.
9325 	 */
9326 	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
9327 	ctsio->io_cont = ctl_cnw_cont;
9328 
9329 	lbalen = (struct ctl_lba_len_flags *)
9330 	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9331 	lbalen->lba = lba;
9332 	lbalen->len = num_blocks;
9333 	lbalen->flags = CTL_LLF_COMPARE | flags;
9334 
9335 	CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
9336 	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9337 	return (retval);
9338 }
9339 
9340 int
9341 ctl_verify(struct ctl_scsiio *ctsio)
9342 {
9343 	struct ctl_lun *lun;
9344 	struct ctl_lba_len_flags *lbalen;
9345 	uint64_t lba;
9346 	uint32_t num_blocks;
9347 	int bytchk, flags;
9348 	int retval;
9349 
9350 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9351 
9352 	CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
9353 
9354 	bytchk = 0;
9355 	flags = CTL_LLF_FUA;
9356 	retval = CTL_RETVAL_COMPLETE;
9357 
9358 	switch (ctsio->cdb[0]) {
9359 	case VERIFY_10: {
9360 		struct scsi_verify_10 *cdb;
9361 
9362 		cdb = (struct scsi_verify_10 *)ctsio->cdb;
9363 		if (cdb->byte2 & SVFY_BYTCHK)
9364 			bytchk = 1;
9365 		if (cdb->byte2 & SVFY_DPO)
9366 			flags |= CTL_LLF_DPO;
9367 		lba = scsi_4btoul(cdb->addr);
9368 		num_blocks = scsi_2btoul(cdb->length);
9369 		break;
9370 	}
9371 	case VERIFY_12: {
9372 		struct scsi_verify_12 *cdb;
9373 
9374 		cdb = (struct scsi_verify_12 *)ctsio->cdb;
9375 		if (cdb->byte2 & SVFY_BYTCHK)
9376 			bytchk = 1;
9377 		if (cdb->byte2 & SVFY_DPO)
9378 			flags |= CTL_LLF_DPO;
9379 		lba = scsi_4btoul(cdb->addr);
9380 		num_blocks = scsi_4btoul(cdb->length);
9381 		break;
9382 	}
9383 	case VERIFY_16: {
9384 		struct scsi_rw_16 *cdb;
9385 
9386 		cdb = (struct scsi_rw_16 *)ctsio->cdb;
9387 		if (cdb->byte2 & SVFY_BYTCHK)
9388 			bytchk = 1;
9389 		if (cdb->byte2 & SVFY_DPO)
9390 			flags |= CTL_LLF_DPO;
9391 		lba = scsi_8btou64(cdb->addr);
9392 		num_blocks = scsi_4btoul(cdb->length);
9393 		break;
9394 	}
9395 	default:
9396 		/*
9397 		 * We got a command we don't support.  This shouldn't
9398 		 * happen, commands should be filtered out above us.
9399 		 */
9400 		ctl_set_invalid_opcode(ctsio);
9401 		ctl_done((union ctl_io *)ctsio);
9402 		return (CTL_RETVAL_COMPLETE);
9403 	}
9404 
9405 	/*
9406 	 * The first check is to make sure we're in bounds, the second
9407 	 * check is to catch wrap-around problems.  If the lba + num blocks
9408 	 * is less than the lba, then we've wrapped around and the block
9409 	 * range is invalid anyway.
9410 	 */
9411 	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9412 	 || ((lba + num_blocks) < lba)) {
9413 		ctl_set_lba_out_of_range(ctsio);
9414 		ctl_done((union ctl_io *)ctsio);
9415 		return (CTL_RETVAL_COMPLETE);
9416 	}
9417 
9418 	/*
9419 	 * According to SBC-3, a transfer length of 0 is not an error.
9420 	 */
9421 	if (num_blocks == 0) {
9422 		ctl_set_success(ctsio);
9423 		ctl_done((union ctl_io *)ctsio);
9424 		return (CTL_RETVAL_COMPLETE);
9425 	}
9426 
9427 	lbalen = (struct ctl_lba_len_flags *)
9428 	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9429 	lbalen->lba = lba;
9430 	lbalen->len = num_blocks;
9431 	if (bytchk) {
9432 		lbalen->flags = CTL_LLF_COMPARE | flags;
9433 		ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9434 	} else {
9435 		lbalen->flags = CTL_LLF_VERIFY | flags;
9436 		ctsio->kern_total_len = 0;
9437 	}
9438 	ctsio->kern_rel_offset = 0;
9439 
9440 	CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9441 	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9442 	return (retval);
9443 }
9444 
9445 int
9446 ctl_report_luns(struct ctl_scsiio *ctsio)
9447 {
9448 	struct scsi_report_luns *cdb;
9449 	struct scsi_report_luns_data *lun_data;
9450 	struct ctl_lun *lun, *request_lun;
9451 	int num_luns, retval;
9452 	uint32_t alloc_len, lun_datalen;
9453 	int num_filled, well_known;
9454 	uint32_t initidx, targ_lun_id, lun_id;
9455 
9456 	retval = CTL_RETVAL_COMPLETE;
9457 	well_known = 0;
9458 
9459 	cdb = (struct scsi_report_luns *)ctsio->cdb;
9460 
9461 	CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9462 
9463 	mtx_lock(&control_softc->ctl_lock);
9464 	num_luns = control_softc->num_luns;
9465 	mtx_unlock(&control_softc->ctl_lock);
9466 
9467 	switch (cdb->select_report) {
9468 	case RPL_REPORT_DEFAULT:
9469 	case RPL_REPORT_ALL:
9470 		break;
9471 	case RPL_REPORT_WELLKNOWN:
9472 		well_known = 1;
9473 		num_luns = 0;
9474 		break;
9475 	default:
9476 		ctl_set_invalid_field(ctsio,
9477 				      /*sks_valid*/ 1,
9478 				      /*command*/ 1,
9479 				      /*field*/ 2,
9480 				      /*bit_valid*/ 0,
9481 				      /*bit*/ 0);
9482 		ctl_done((union ctl_io *)ctsio);
9483 		return (retval);
9484 		break; /* NOTREACHED */
9485 	}
9486 
9487 	alloc_len = scsi_4btoul(cdb->length);
9488 	/*
9489 	 * The initiator has to allocate at least 16 bytes for this request,
9490 	 * so he can at least get the header and the first LUN.  Otherwise
9491 	 * we reject the request (per SPC-3 rev 14, section 6.21).
9492 	 */
9493 	if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9494 	    sizeof(struct scsi_report_luns_lundata))) {
9495 		ctl_set_invalid_field(ctsio,
9496 				      /*sks_valid*/ 1,
9497 				      /*command*/ 1,
9498 				      /*field*/ 6,
9499 				      /*bit_valid*/ 0,
9500 				      /*bit*/ 0);
9501 		ctl_done((union ctl_io *)ctsio);
9502 		return (retval);
9503 	}
9504 
9505 	request_lun = (struct ctl_lun *)
9506 		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9507 
9508 	lun_datalen = sizeof(*lun_data) +
9509 		(num_luns * sizeof(struct scsi_report_luns_lundata));
9510 
9511 	ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9512 	lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9513 	ctsio->kern_sg_entries = 0;
9514 
9515 	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9516 
9517 	mtx_lock(&control_softc->ctl_lock);
9518 	for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
9519 		lun_id = ctl_map_lun(ctsio->io_hdr.nexus.targ_port, targ_lun_id);
9520 		if (lun_id >= CTL_MAX_LUNS)
9521 			continue;
9522 		lun = control_softc->ctl_luns[lun_id];
9523 		if (lun == NULL)
9524 			continue;
9525 
9526 		if (targ_lun_id <= 0xff) {
9527 			/*
9528 			 * Peripheral addressing method, bus number 0.
9529 			 */
9530 			lun_data->luns[num_filled].lundata[0] =
9531 				RPL_LUNDATA_ATYP_PERIPH;
9532 			lun_data->luns[num_filled].lundata[1] = targ_lun_id;
9533 			num_filled++;
9534 		} else if (targ_lun_id <= 0x3fff) {
9535 			/*
9536 			 * Flat addressing method.
9537 			 */
9538 			lun_data->luns[num_filled].lundata[0] =
9539 				RPL_LUNDATA_ATYP_FLAT |
9540 				(targ_lun_id & RPL_LUNDATA_FLAT_LUN_MASK);
9541 #ifdef OLDCTLHEADERS
9542 				(SRLD_ADDR_FLAT << SRLD_ADDR_SHIFT) |
9543 				(targ_lun_id & SRLD_BUS_LUN_MASK);
9544 #endif
9545 			lun_data->luns[num_filled].lundata[1] =
9546 #ifdef OLDCTLHEADERS
9547 				targ_lun_id >> SRLD_BUS_LUN_BITS;
9548 #endif
9549 				targ_lun_id >> RPL_LUNDATA_FLAT_LUN_BITS;
9550 			num_filled++;
9551 		} else {
9552 			printf("ctl_report_luns: bogus LUN number %jd, "
9553 			       "skipping\n", (intmax_t)targ_lun_id);
9554 		}
9555 		/*
9556 		 * According to SPC-3, rev 14 section 6.21:
9557 		 *
9558 		 * "The execution of a REPORT LUNS command to any valid and
9559 		 * installed logical unit shall clear the REPORTED LUNS DATA
9560 		 * HAS CHANGED unit attention condition for all logical
9561 		 * units of that target with respect to the requesting
9562 		 * initiator. A valid and installed logical unit is one
9563 		 * having a PERIPHERAL QUALIFIER of 000b in the standard
9564 		 * INQUIRY data (see 6.4.2)."
9565 		 *
9566 		 * If request_lun is NULL, the LUN this report luns command
9567 		 * was issued to is either disabled or doesn't exist. In that
9568 		 * case, we shouldn't clear any pending lun change unit
9569 		 * attention.
9570 		 */
9571 		if (request_lun != NULL) {
9572 			mtx_lock(&lun->lun_lock);
9573 			lun->pending_ua[initidx] &= ~CTL_UA_LUN_CHANGE;
9574 			mtx_unlock(&lun->lun_lock);
9575 		}
9576 	}
9577 	mtx_unlock(&control_softc->ctl_lock);
9578 
9579 	/*
9580 	 * It's quite possible that we've returned fewer LUNs than we allocated
9581 	 * space for.  Trim it.
9582 	 */
9583 	lun_datalen = sizeof(*lun_data) +
9584 		(num_filled * sizeof(struct scsi_report_luns_lundata));
9585 
9586 	if (lun_datalen < alloc_len) {
9587 		ctsio->residual = alloc_len - lun_datalen;
9588 		ctsio->kern_data_len = lun_datalen;
9589 		ctsio->kern_total_len = lun_datalen;
9590 	} else {
9591 		ctsio->residual = 0;
9592 		ctsio->kern_data_len = alloc_len;
9593 		ctsio->kern_total_len = alloc_len;
9594 	}
9595 	ctsio->kern_data_resid = 0;
9596 	ctsio->kern_rel_offset = 0;
9597 	ctsio->kern_sg_entries = 0;
9598 
9599 	/*
9600 	 * We set this to the actual data length, regardless of how much
9601 	 * space we actually have to return results.  If the user looks at
9602 	 * this value, he'll know whether or not he allocated enough space
9603 	 * and reissue the command if necessary.  We don't support well
9604 	 * known logical units, so if the user asks for that, return none.
9605 	 */
9606 	scsi_ulto4b(lun_datalen - 8, lun_data->length);
9607 
9608 	/*
9609 	 * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9610 	 * this request.
9611 	 */
9612 	ctsio->scsi_status = SCSI_STATUS_OK;
9613 
9614 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9615 	ctsio->be_move_done = ctl_config_move_done;
9616 	ctl_datamove((union ctl_io *)ctsio);
9617 
9618 	return (retval);
9619 }
9620 
9621 int
9622 ctl_request_sense(struct ctl_scsiio *ctsio)
9623 {
9624 	struct scsi_request_sense *cdb;
9625 	struct scsi_sense_data *sense_ptr;
9626 	struct ctl_lun *lun;
9627 	uint32_t initidx;
9628 	int have_error;
9629 	scsi_sense_data_type sense_format;
9630 
9631 	cdb = (struct scsi_request_sense *)ctsio->cdb;
9632 
9633 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9634 
9635 	CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9636 
9637 	/*
9638 	 * Determine which sense format the user wants.
9639 	 */
9640 	if (cdb->byte2 & SRS_DESC)
9641 		sense_format = SSD_TYPE_DESC;
9642 	else
9643 		sense_format = SSD_TYPE_FIXED;
9644 
9645 	ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9646 	sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9647 	ctsio->kern_sg_entries = 0;
9648 
9649 	/*
9650 	 * struct scsi_sense_data, which is currently set to 256 bytes, is
9651 	 * larger than the largest allowed value for the length field in the
9652 	 * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9653 	 */
9654 	ctsio->residual = 0;
9655 	ctsio->kern_data_len = cdb->length;
9656 	ctsio->kern_total_len = cdb->length;
9657 
9658 	ctsio->kern_data_resid = 0;
9659 	ctsio->kern_rel_offset = 0;
9660 	ctsio->kern_sg_entries = 0;
9661 
9662 	/*
9663 	 * If we don't have a LUN, we don't have any pending sense.
9664 	 */
9665 	if (lun == NULL)
9666 		goto no_sense;
9667 
9668 	have_error = 0;
9669 	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9670 	/*
9671 	 * Check for pending sense, and then for pending unit attentions.
9672 	 * Pending sense gets returned first, then pending unit attentions.
9673 	 */
9674 	mtx_lock(&lun->lun_lock);
9675 #ifdef CTL_WITH_CA
9676 	if (ctl_is_set(lun->have_ca, initidx)) {
9677 		scsi_sense_data_type stored_format;
9678 
9679 		/*
9680 		 * Check to see which sense format was used for the stored
9681 		 * sense data.
9682 		 */
9683 		stored_format = scsi_sense_type(&lun->pending_sense[initidx]);
9684 
9685 		/*
9686 		 * If the user requested a different sense format than the
9687 		 * one we stored, then we need to convert it to the other
9688 		 * format.  If we're going from descriptor to fixed format
9689 		 * sense data, we may lose things in translation, depending
9690 		 * on what options were used.
9691 		 *
9692 		 * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9693 		 * for some reason we'll just copy it out as-is.
9694 		 */
9695 		if ((stored_format == SSD_TYPE_FIXED)
9696 		 && (sense_format == SSD_TYPE_DESC))
9697 			ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9698 			    &lun->pending_sense[initidx],
9699 			    (struct scsi_sense_data_desc *)sense_ptr);
9700 		else if ((stored_format == SSD_TYPE_DESC)
9701 		      && (sense_format == SSD_TYPE_FIXED))
9702 			ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9703 			    &lun->pending_sense[initidx],
9704 			    (struct scsi_sense_data_fixed *)sense_ptr);
9705 		else
9706 			memcpy(sense_ptr, &lun->pending_sense[initidx],
9707 			       ctl_min(sizeof(*sense_ptr),
9708 			       sizeof(lun->pending_sense[initidx])));
9709 
9710 		ctl_clear_mask(lun->have_ca, initidx);
9711 		have_error = 1;
9712 	} else
9713 #endif
9714 	if (lun->pending_ua[initidx] != CTL_UA_NONE) {
9715 		ctl_ua_type ua_type;
9716 
9717 		ua_type = ctl_build_ua(&lun->pending_ua[initidx],
9718 				       sense_ptr, sense_format);
9719 		if (ua_type != CTL_UA_NONE)
9720 			have_error = 1;
9721 	}
9722 	mtx_unlock(&lun->lun_lock);
9723 
9724 	/*
9725 	 * We already have a pending error, return it.
9726 	 */
9727 	if (have_error != 0) {
9728 		/*
9729 		 * We report the SCSI status as OK, since the status of the
9730 		 * request sense command itself is OK.
9731 		 */
9732 		ctsio->scsi_status = SCSI_STATUS_OK;
9733 
9734 		/*
9735 		 * We report 0 for the sense length, because we aren't doing
9736 		 * autosense in this case.  We're reporting sense as
9737 		 * parameter data.
9738 		 */
9739 		ctsio->sense_len = 0;
9740 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9741 		ctsio->be_move_done = ctl_config_move_done;
9742 		ctl_datamove((union ctl_io *)ctsio);
9743 
9744 		return (CTL_RETVAL_COMPLETE);
9745 	}
9746 
9747 no_sense:
9748 
9749 	/*
9750 	 * No sense information to report, so we report that everything is
9751 	 * okay.
9752 	 */
9753 	ctl_set_sense_data(sense_ptr,
9754 			   lun,
9755 			   sense_format,
9756 			   /*current_error*/ 1,
9757 			   /*sense_key*/ SSD_KEY_NO_SENSE,
9758 			   /*asc*/ 0x00,
9759 			   /*ascq*/ 0x00,
9760 			   SSD_ELEM_NONE);
9761 
9762 	ctsio->scsi_status = SCSI_STATUS_OK;
9763 
9764 	/*
9765 	 * We report 0 for the sense length, because we aren't doing
9766 	 * autosense in this case.  We're reporting sense as parameter data.
9767 	 */
9768 	ctsio->sense_len = 0;
9769 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9770 	ctsio->be_move_done = ctl_config_move_done;
9771 	ctl_datamove((union ctl_io *)ctsio);
9772 
9773 	return (CTL_RETVAL_COMPLETE);
9774 }
9775 
9776 int
9777 ctl_tur(struct ctl_scsiio *ctsio)
9778 {
9779 	struct ctl_lun *lun;
9780 
9781 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9782 
9783 	CTL_DEBUG_PRINT(("ctl_tur\n"));
9784 
9785 	if (lun == NULL)
9786 		return (EINVAL);
9787 
9788 	ctsio->scsi_status = SCSI_STATUS_OK;
9789 	ctsio->io_hdr.status = CTL_SUCCESS;
9790 
9791 	ctl_done((union ctl_io *)ctsio);
9792 
9793 	return (CTL_RETVAL_COMPLETE);
9794 }
9795 
9796 #ifdef notyet
9797 static int
9798 ctl_cmddt_inquiry(struct ctl_scsiio *ctsio)
9799 {
9800 
9801 }
9802 #endif
9803 
9804 static int
9805 ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9806 {
9807 	struct scsi_vpd_supported_pages *pages;
9808 	int sup_page_size;
9809 	struct ctl_lun *lun;
9810 
9811 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9812 
9813 	sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9814 	    SCSI_EVPD_NUM_SUPPORTED_PAGES;
9815 	ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9816 	pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9817 	ctsio->kern_sg_entries = 0;
9818 
9819 	if (sup_page_size < alloc_len) {
9820 		ctsio->residual = alloc_len - sup_page_size;
9821 		ctsio->kern_data_len = sup_page_size;
9822 		ctsio->kern_total_len = sup_page_size;
9823 	} else {
9824 		ctsio->residual = 0;
9825 		ctsio->kern_data_len = alloc_len;
9826 		ctsio->kern_total_len = alloc_len;
9827 	}
9828 	ctsio->kern_data_resid = 0;
9829 	ctsio->kern_rel_offset = 0;
9830 	ctsio->kern_sg_entries = 0;
9831 
9832 	/*
9833 	 * The control device is always connected.  The disk device, on the
9834 	 * other hand, may not be online all the time.  Need to change this
9835 	 * to figure out whether the disk device is actually online or not.
9836 	 */
9837 	if (lun != NULL)
9838 		pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9839 				lun->be_lun->lun_type;
9840 	else
9841 		pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9842 
9843 	pages->length = SCSI_EVPD_NUM_SUPPORTED_PAGES;
9844 	/* Supported VPD pages */
9845 	pages->page_list[0] = SVPD_SUPPORTED_PAGES;
9846 	/* Serial Number */
9847 	pages->page_list[1] = SVPD_UNIT_SERIAL_NUMBER;
9848 	/* Device Identification */
9849 	pages->page_list[2] = SVPD_DEVICE_ID;
9850 	/* Extended INQUIRY Data */
9851 	pages->page_list[3] = SVPD_EXTENDED_INQUIRY_DATA;
9852 	/* Mode Page Policy */
9853 	pages->page_list[4] = SVPD_MODE_PAGE_POLICY;
9854 	/* SCSI Ports */
9855 	pages->page_list[5] = SVPD_SCSI_PORTS;
9856 	/* Third-party Copy */
9857 	pages->page_list[6] = SVPD_SCSI_TPC;
9858 	/* Block limits */
9859 	pages->page_list[7] = SVPD_BLOCK_LIMITS;
9860 	/* Block Device Characteristics */
9861 	pages->page_list[8] = SVPD_BDC;
9862 	/* Logical Block Provisioning */
9863 	pages->page_list[9] = SVPD_LBP;
9864 
9865 	ctsio->scsi_status = SCSI_STATUS_OK;
9866 
9867 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9868 	ctsio->be_move_done = ctl_config_move_done;
9869 	ctl_datamove((union ctl_io *)ctsio);
9870 
9871 	return (CTL_RETVAL_COMPLETE);
9872 }
9873 
9874 static int
9875 ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9876 {
9877 	struct scsi_vpd_unit_serial_number *sn_ptr;
9878 	struct ctl_lun *lun;
9879 
9880 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9881 
9882 	ctsio->kern_data_ptr = malloc(sizeof(*sn_ptr), M_CTL, M_WAITOK | M_ZERO);
9883 	sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9884 	ctsio->kern_sg_entries = 0;
9885 
9886 	if (sizeof(*sn_ptr) < alloc_len) {
9887 		ctsio->residual = alloc_len - sizeof(*sn_ptr);
9888 		ctsio->kern_data_len = sizeof(*sn_ptr);
9889 		ctsio->kern_total_len = sizeof(*sn_ptr);
9890 	} else {
9891 		ctsio->residual = 0;
9892 		ctsio->kern_data_len = alloc_len;
9893 		ctsio->kern_total_len = alloc_len;
9894 	}
9895 	ctsio->kern_data_resid = 0;
9896 	ctsio->kern_rel_offset = 0;
9897 	ctsio->kern_sg_entries = 0;
9898 
9899 	/*
9900 	 * The control device is always connected.  The disk device, on the
9901 	 * other hand, may not be online all the time.  Need to change this
9902 	 * to figure out whether the disk device is actually online or not.
9903 	 */
9904 	if (lun != NULL)
9905 		sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9906 				  lun->be_lun->lun_type;
9907 	else
9908 		sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9909 
9910 	sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9911 	sn_ptr->length = ctl_min(sizeof(*sn_ptr) - 4, CTL_SN_LEN);
9912 	/*
9913 	 * If we don't have a LUN, we just leave the serial number as
9914 	 * all spaces.
9915 	 */
9916 	memset(sn_ptr->serial_num, 0x20, sizeof(sn_ptr->serial_num));
9917 	if (lun != NULL) {
9918 		strncpy((char *)sn_ptr->serial_num,
9919 			(char *)lun->be_lun->serial_num, CTL_SN_LEN);
9920 	}
9921 	ctsio->scsi_status = SCSI_STATUS_OK;
9922 
9923 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9924 	ctsio->be_move_done = ctl_config_move_done;
9925 	ctl_datamove((union ctl_io *)ctsio);
9926 
9927 	return (CTL_RETVAL_COMPLETE);
9928 }
9929 
9930 
9931 static int
9932 ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len)
9933 {
9934 	struct scsi_vpd_extended_inquiry_data *eid_ptr;
9935 	struct ctl_lun *lun;
9936 	int data_len;
9937 
9938 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9939 
9940 	data_len = sizeof(struct scsi_vpd_extended_inquiry_data);
9941 	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9942 	eid_ptr = (struct scsi_vpd_extended_inquiry_data *)ctsio->kern_data_ptr;
9943 	ctsio->kern_sg_entries = 0;
9944 
9945 	if (data_len < alloc_len) {
9946 		ctsio->residual = alloc_len - data_len;
9947 		ctsio->kern_data_len = data_len;
9948 		ctsio->kern_total_len = data_len;
9949 	} else {
9950 		ctsio->residual = 0;
9951 		ctsio->kern_data_len = alloc_len;
9952 		ctsio->kern_total_len = alloc_len;
9953 	}
9954 	ctsio->kern_data_resid = 0;
9955 	ctsio->kern_rel_offset = 0;
9956 	ctsio->kern_sg_entries = 0;
9957 
9958 	/*
9959 	 * The control device is always connected.  The disk device, on the
9960 	 * other hand, may not be online all the time.
9961 	 */
9962 	if (lun != NULL)
9963 		eid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9964 				     lun->be_lun->lun_type;
9965 	else
9966 		eid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9967 	eid_ptr->page_code = SVPD_EXTENDED_INQUIRY_DATA;
9968 	eid_ptr->page_length = data_len - 4;
9969 	eid_ptr->flags2 = SVPD_EID_HEADSUP | SVPD_EID_ORDSUP | SVPD_EID_SIMPSUP;
9970 	eid_ptr->flags3 = SVPD_EID_V_SUP;
9971 
9972 	ctsio->scsi_status = SCSI_STATUS_OK;
9973 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9974 	ctsio->be_move_done = ctl_config_move_done;
9975 	ctl_datamove((union ctl_io *)ctsio);
9976 
9977 	return (CTL_RETVAL_COMPLETE);
9978 }
9979 
9980 static int
9981 ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len)
9982 {
9983 	struct scsi_vpd_mode_page_policy *mpp_ptr;
9984 	struct ctl_lun *lun;
9985 	int data_len;
9986 
9987 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9988 
9989 	data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9990 	    sizeof(struct scsi_vpd_mode_page_policy_descr);
9991 
9992 	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9993 	mpp_ptr = (struct scsi_vpd_mode_page_policy *)ctsio->kern_data_ptr;
9994 	ctsio->kern_sg_entries = 0;
9995 
9996 	if (data_len < alloc_len) {
9997 		ctsio->residual = alloc_len - data_len;
9998 		ctsio->kern_data_len = data_len;
9999 		ctsio->kern_total_len = data_len;
10000 	} else {
10001 		ctsio->residual = 0;
10002 		ctsio->kern_data_len = alloc_len;
10003 		ctsio->kern_total_len = alloc_len;
10004 	}
10005 	ctsio->kern_data_resid = 0;
10006 	ctsio->kern_rel_offset = 0;
10007 	ctsio->kern_sg_entries = 0;
10008 
10009 	/*
10010 	 * The control device is always connected.  The disk device, on the
10011 	 * other hand, may not be online all the time.
10012 	 */
10013 	if (lun != NULL)
10014 		mpp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10015 				     lun->be_lun->lun_type;
10016 	else
10017 		mpp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10018 	mpp_ptr->page_code = SVPD_MODE_PAGE_POLICY;
10019 	scsi_ulto2b(data_len - 4, mpp_ptr->page_length);
10020 	mpp_ptr->descr[0].page_code = 0x3f;
10021 	mpp_ptr->descr[0].subpage_code = 0xff;
10022 	mpp_ptr->descr[0].policy = SVPD_MPP_SHARED;
10023 
10024 	ctsio->scsi_status = SCSI_STATUS_OK;
10025 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10026 	ctsio->be_move_done = ctl_config_move_done;
10027 	ctl_datamove((union ctl_io *)ctsio);
10028 
10029 	return (CTL_RETVAL_COMPLETE);
10030 }
10031 
10032 static int
10033 ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
10034 {
10035 	struct scsi_vpd_device_id *devid_ptr;
10036 	struct scsi_vpd_id_descriptor *desc;
10037 	struct ctl_softc *ctl_softc;
10038 	struct ctl_lun *lun;
10039 	struct ctl_port *port;
10040 	int data_len;
10041 	uint8_t proto;
10042 
10043 	ctl_softc = control_softc;
10044 
10045 	port = ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)];
10046 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10047 
10048 	data_len = sizeof(struct scsi_vpd_device_id) +
10049 	    sizeof(struct scsi_vpd_id_descriptor) +
10050 		sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
10051 	    sizeof(struct scsi_vpd_id_descriptor) +
10052 		sizeof(struct scsi_vpd_id_trgt_port_grp_id);
10053 	if (lun && lun->lun_devid)
10054 		data_len += lun->lun_devid->len;
10055 	if (port->port_devid)
10056 		data_len += port->port_devid->len;
10057 	if (port->target_devid)
10058 		data_len += port->target_devid->len;
10059 
10060 	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10061 	devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
10062 	ctsio->kern_sg_entries = 0;
10063 
10064 	if (data_len < alloc_len) {
10065 		ctsio->residual = alloc_len - data_len;
10066 		ctsio->kern_data_len = data_len;
10067 		ctsio->kern_total_len = data_len;
10068 	} else {
10069 		ctsio->residual = 0;
10070 		ctsio->kern_data_len = alloc_len;
10071 		ctsio->kern_total_len = alloc_len;
10072 	}
10073 	ctsio->kern_data_resid = 0;
10074 	ctsio->kern_rel_offset = 0;
10075 	ctsio->kern_sg_entries = 0;
10076 
10077 	/*
10078 	 * The control device is always connected.  The disk device, on the
10079 	 * other hand, may not be online all the time.
10080 	 */
10081 	if (lun != NULL)
10082 		devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10083 				     lun->be_lun->lun_type;
10084 	else
10085 		devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10086 	devid_ptr->page_code = SVPD_DEVICE_ID;
10087 	scsi_ulto2b(data_len - 4, devid_ptr->length);
10088 
10089 	if (port->port_type == CTL_PORT_FC)
10090 		proto = SCSI_PROTO_FC << 4;
10091 	else if (port->port_type == CTL_PORT_ISCSI)
10092 		proto = SCSI_PROTO_ISCSI << 4;
10093 	else
10094 		proto = SCSI_PROTO_SPI << 4;
10095 	desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
10096 
10097 	/*
10098 	 * We're using a LUN association here.  i.e., this device ID is a
10099 	 * per-LUN identifier.
10100 	 */
10101 	if (lun && lun->lun_devid) {
10102 		memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
10103 		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
10104 		    lun->lun_devid->len);
10105 	}
10106 
10107 	/*
10108 	 * This is for the WWPN which is a port association.
10109 	 */
10110 	if (port->port_devid) {
10111 		memcpy(desc, port->port_devid->data, port->port_devid->len);
10112 		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
10113 		    port->port_devid->len);
10114 	}
10115 
10116 	/*
10117 	 * This is for the Relative Target Port(type 4h) identifier
10118 	 */
10119 	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
10120 	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
10121 	    SVPD_ID_TYPE_RELTARG;
10122 	desc->length = 4;
10123 	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
10124 	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
10125 	    sizeof(struct scsi_vpd_id_rel_trgt_port_id));
10126 
10127 	/*
10128 	 * This is for the Target Port Group(type 5h) identifier
10129 	 */
10130 	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
10131 	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
10132 	    SVPD_ID_TYPE_TPORTGRP;
10133 	desc->length = 4;
10134 	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS + 1,
10135 	    &desc->identifier[2]);
10136 	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
10137 	    sizeof(struct scsi_vpd_id_trgt_port_grp_id));
10138 
10139 	/*
10140 	 * This is for the Target identifier
10141 	 */
10142 	if (port->target_devid) {
10143 		memcpy(desc, port->target_devid->data, port->target_devid->len);
10144 	}
10145 
10146 	ctsio->scsi_status = SCSI_STATUS_OK;
10147 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10148 	ctsio->be_move_done = ctl_config_move_done;
10149 	ctl_datamove((union ctl_io *)ctsio);
10150 
10151 	return (CTL_RETVAL_COMPLETE);
10152 }
10153 
10154 static int
10155 ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
10156 {
10157 	struct ctl_softc *softc = control_softc;
10158 	struct scsi_vpd_scsi_ports *sp;
10159 	struct scsi_vpd_port_designation *pd;
10160 	struct scsi_vpd_port_designation_cont *pdc;
10161 	struct ctl_lun *lun;
10162 	struct ctl_port *port;
10163 	int data_len, num_target_ports, iid_len, id_len, g, pg, p;
10164 	int num_target_port_groups, single;
10165 
10166 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10167 
10168 	single = ctl_is_single;
10169 	if (single)
10170 		num_target_port_groups = 1;
10171 	else
10172 		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
10173 	num_target_ports = 0;
10174 	iid_len = 0;
10175 	id_len = 0;
10176 	mtx_lock(&softc->ctl_lock);
10177 	STAILQ_FOREACH(port, &softc->port_list, links) {
10178 		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
10179 			continue;
10180 		if (lun != NULL &&
10181 		    ctl_map_lun_back(port->targ_port, lun->lun) >=
10182 		    CTL_MAX_LUNS)
10183 			continue;
10184 		num_target_ports++;
10185 		if (port->init_devid)
10186 			iid_len += port->init_devid->len;
10187 		if (port->port_devid)
10188 			id_len += port->port_devid->len;
10189 	}
10190 	mtx_unlock(&softc->ctl_lock);
10191 
10192 	data_len = sizeof(struct scsi_vpd_scsi_ports) + num_target_port_groups *
10193 	    num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
10194 	     sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
10195 	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10196 	sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
10197 	ctsio->kern_sg_entries = 0;
10198 
10199 	if (data_len < alloc_len) {
10200 		ctsio->residual = alloc_len - data_len;
10201 		ctsio->kern_data_len = data_len;
10202 		ctsio->kern_total_len = data_len;
10203 	} else {
10204 		ctsio->residual = 0;
10205 		ctsio->kern_data_len = alloc_len;
10206 		ctsio->kern_total_len = alloc_len;
10207 	}
10208 	ctsio->kern_data_resid = 0;
10209 	ctsio->kern_rel_offset = 0;
10210 	ctsio->kern_sg_entries = 0;
10211 
10212 	/*
10213 	 * The control device is always connected.  The disk device, on the
10214 	 * other hand, may not be online all the time.  Need to change this
10215 	 * to figure out whether the disk device is actually online or not.
10216 	 */
10217 	if (lun != NULL)
10218 		sp->device = (SID_QUAL_LU_CONNECTED << 5) |
10219 				  lun->be_lun->lun_type;
10220 	else
10221 		sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10222 
10223 	sp->page_code = SVPD_SCSI_PORTS;
10224 	scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
10225 	    sp->page_length);
10226 	pd = &sp->design[0];
10227 
10228 	mtx_lock(&softc->ctl_lock);
10229 	if (softc->flags & CTL_FLAG_MASTER_SHELF)
10230 		pg = 0;
10231 	else
10232 		pg = 1;
10233 	for (g = 0; g < num_target_port_groups; g++) {
10234 		STAILQ_FOREACH(port, &softc->port_list, links) {
10235 			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
10236 				continue;
10237 			if (lun != NULL &&
10238 			    ctl_map_lun_back(port->targ_port, lun->lun) >=
10239 			    CTL_MAX_LUNS)
10240 				continue;
10241 			p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
10242 			scsi_ulto2b(p, pd->relative_port_id);
10243 			if (port->init_devid && g == pg) {
10244 				iid_len = port->init_devid->len;
10245 				memcpy(pd->initiator_transportid,
10246 				    port->init_devid->data, port->init_devid->len);
10247 			} else
10248 				iid_len = 0;
10249 			scsi_ulto2b(iid_len, pd->initiator_transportid_length);
10250 			pdc = (struct scsi_vpd_port_designation_cont *)
10251 			    (&pd->initiator_transportid[iid_len]);
10252 			if (port->port_devid && g == pg) {
10253 				id_len = port->port_devid->len;
10254 				memcpy(pdc->target_port_descriptors,
10255 				    port->port_devid->data, port->port_devid->len);
10256 			} else
10257 				id_len = 0;
10258 			scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
10259 			pd = (struct scsi_vpd_port_designation *)
10260 			    ((uint8_t *)pdc->target_port_descriptors + id_len);
10261 		}
10262 	}
10263 	mtx_unlock(&softc->ctl_lock);
10264 
10265 	ctsio->scsi_status = SCSI_STATUS_OK;
10266 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10267 	ctsio->be_move_done = ctl_config_move_done;
10268 	ctl_datamove((union ctl_io *)ctsio);
10269 
10270 	return (CTL_RETVAL_COMPLETE);
10271 }
10272 
10273 static int
10274 ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
10275 {
10276 	struct scsi_vpd_block_limits *bl_ptr;
10277 	struct ctl_lun *lun;
10278 	int bs;
10279 
10280 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10281 
10282 	ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
10283 	bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
10284 	ctsio->kern_sg_entries = 0;
10285 
10286 	if (sizeof(*bl_ptr) < alloc_len) {
10287 		ctsio->residual = alloc_len - sizeof(*bl_ptr);
10288 		ctsio->kern_data_len = sizeof(*bl_ptr);
10289 		ctsio->kern_total_len = sizeof(*bl_ptr);
10290 	} else {
10291 		ctsio->residual = 0;
10292 		ctsio->kern_data_len = alloc_len;
10293 		ctsio->kern_total_len = alloc_len;
10294 	}
10295 	ctsio->kern_data_resid = 0;
10296 	ctsio->kern_rel_offset = 0;
10297 	ctsio->kern_sg_entries = 0;
10298 
10299 	/*
10300 	 * The control device is always connected.  The disk device, on the
10301 	 * other hand, may not be online all the time.  Need to change this
10302 	 * to figure out whether the disk device is actually online or not.
10303 	 */
10304 	if (lun != NULL)
10305 		bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10306 				  lun->be_lun->lun_type;
10307 	else
10308 		bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10309 
10310 	bl_ptr->page_code = SVPD_BLOCK_LIMITS;
10311 	scsi_ulto2b(sizeof(*bl_ptr) - 4, bl_ptr->page_length);
10312 	bl_ptr->max_cmp_write_len = 0xff;
10313 	scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
10314 	if (lun != NULL) {
10315 		bs = lun->be_lun->blocksize;
10316 		scsi_ulto4b(MAXPHYS / bs, bl_ptr->opt_txfer_len);
10317 		if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10318 			scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_lba_cnt);
10319 			scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_blk_cnt);
10320 			if (lun->be_lun->pblockexp != 0) {
10321 				scsi_ulto4b((1 << lun->be_lun->pblockexp),
10322 				    bl_ptr->opt_unmap_grain);
10323 				scsi_ulto4b(0x80000000 | lun->be_lun->pblockoff,
10324 				    bl_ptr->unmap_grain_align);
10325 			}
10326 		}
10327 		scsi_ulto4b(lun->be_lun->atomicblock,
10328 		    bl_ptr->max_atomic_transfer_length);
10329 		scsi_ulto4b(0, bl_ptr->atomic_alignment);
10330 		scsi_ulto4b(0, bl_ptr->atomic_transfer_length_granularity);
10331 	}
10332 	scsi_u64to8b(UINT64_MAX, bl_ptr->max_write_same_length);
10333 
10334 	ctsio->scsi_status = SCSI_STATUS_OK;
10335 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10336 	ctsio->be_move_done = ctl_config_move_done;
10337 	ctl_datamove((union ctl_io *)ctsio);
10338 
10339 	return (CTL_RETVAL_COMPLETE);
10340 }
10341 
10342 static int
10343 ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len)
10344 {
10345 	struct scsi_vpd_block_device_characteristics *bdc_ptr;
10346 	struct ctl_lun *lun;
10347 
10348 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10349 
10350 	ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO);
10351 	bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr;
10352 	ctsio->kern_sg_entries = 0;
10353 
10354 	if (sizeof(*bdc_ptr) < alloc_len) {
10355 		ctsio->residual = alloc_len - sizeof(*bdc_ptr);
10356 		ctsio->kern_data_len = sizeof(*bdc_ptr);
10357 		ctsio->kern_total_len = sizeof(*bdc_ptr);
10358 	} else {
10359 		ctsio->residual = 0;
10360 		ctsio->kern_data_len = alloc_len;
10361 		ctsio->kern_total_len = alloc_len;
10362 	}
10363 	ctsio->kern_data_resid = 0;
10364 	ctsio->kern_rel_offset = 0;
10365 	ctsio->kern_sg_entries = 0;
10366 
10367 	/*
10368 	 * The control device is always connected.  The disk device, on the
10369 	 * other hand, may not be online all the time.  Need to change this
10370 	 * to figure out whether the disk device is actually online or not.
10371 	 */
10372 	if (lun != NULL)
10373 		bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10374 				  lun->be_lun->lun_type;
10375 	else
10376 		bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10377 	bdc_ptr->page_code = SVPD_BDC;
10378 	scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length);
10379 	scsi_ulto2b(SVPD_NON_ROTATING, bdc_ptr->medium_rotation_rate);
10380 	bdc_ptr->flags = SVPD_FUAB | SVPD_VBULS;
10381 
10382 	ctsio->scsi_status = SCSI_STATUS_OK;
10383 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10384 	ctsio->be_move_done = ctl_config_move_done;
10385 	ctl_datamove((union ctl_io *)ctsio);
10386 
10387 	return (CTL_RETVAL_COMPLETE);
10388 }
10389 
10390 static int
10391 ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
10392 {
10393 	struct scsi_vpd_logical_block_prov *lbp_ptr;
10394 	struct ctl_lun *lun;
10395 
10396 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10397 
10398 	ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
10399 	lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
10400 	ctsio->kern_sg_entries = 0;
10401 
10402 	if (sizeof(*lbp_ptr) < alloc_len) {
10403 		ctsio->residual = alloc_len - sizeof(*lbp_ptr);
10404 		ctsio->kern_data_len = sizeof(*lbp_ptr);
10405 		ctsio->kern_total_len = sizeof(*lbp_ptr);
10406 	} else {
10407 		ctsio->residual = 0;
10408 		ctsio->kern_data_len = alloc_len;
10409 		ctsio->kern_total_len = alloc_len;
10410 	}
10411 	ctsio->kern_data_resid = 0;
10412 	ctsio->kern_rel_offset = 0;
10413 	ctsio->kern_sg_entries = 0;
10414 
10415 	/*
10416 	 * The control device is always connected.  The disk device, on the
10417 	 * other hand, may not be online all the time.  Need to change this
10418 	 * to figure out whether the disk device is actually online or not.
10419 	 */
10420 	if (lun != NULL)
10421 		lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10422 				  lun->be_lun->lun_type;
10423 	else
10424 		lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10425 
10426 	lbp_ptr->page_code = SVPD_LBP;
10427 	scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length);
10428 	if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10429 		lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 |
10430 		    SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP;
10431 		lbp_ptr->prov_type = SVPD_LBP_RESOURCE;
10432 	}
10433 
10434 	ctsio->scsi_status = SCSI_STATUS_OK;
10435 	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10436 	ctsio->be_move_done = ctl_config_move_done;
10437 	ctl_datamove((union ctl_io *)ctsio);
10438 
10439 	return (CTL_RETVAL_COMPLETE);
10440 }
10441 
10442 static int
10443 ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
10444 {
10445 	struct scsi_inquiry *cdb;
10446 	struct ctl_lun *lun;
10447 	int alloc_len, retval;
10448 
10449 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10450 	cdb = (struct scsi_inquiry *)ctsio->cdb;
10451 
10452 	retval = CTL_RETVAL_COMPLETE;
10453 
10454 	alloc_len = scsi_2btoul(cdb->length);
10455 
10456 	switch (cdb->page_code) {
10457 	case SVPD_SUPPORTED_PAGES:
10458 		retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
10459 		break;
10460 	case SVPD_UNIT_SERIAL_NUMBER:
10461 		retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
10462 		break;
10463 	case SVPD_DEVICE_ID:
10464 		retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
10465 		break;
10466 	case SVPD_EXTENDED_INQUIRY_DATA:
10467 		retval = ctl_inquiry_evpd_eid(ctsio, alloc_len);
10468 		break;
10469 	case SVPD_MODE_PAGE_POLICY:
10470 		retval = ctl_inquiry_evpd_mpp(ctsio, alloc_len);
10471 		break;
10472 	case SVPD_SCSI_PORTS:
10473 		retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
10474 		break;
10475 	case SVPD_SCSI_TPC:
10476 		retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
10477 		break;
10478 	case SVPD_BLOCK_LIMITS:
10479 		retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
10480 		break;
10481 	case SVPD_BDC:
10482 		retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len);
10483 		break;
10484 	case SVPD_LBP:
10485 		retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
10486 		break;
10487 	default:
10488 		ctl_set_invalid_field(ctsio,
10489 				      /*sks_valid*/ 1,
10490 				      /*command*/ 1,
10491 				      /*field*/ 2,
10492 				      /*bit_valid*/ 0,
10493 				      /*bit*/ 0);
10494 		ctl_done((union ctl_io *)ctsio);
10495 		retval = CTL_RETVAL_COMPLETE;
10496 		break;
10497 	}
10498 
10499 	return (retval);
10500 }
10501 
10502 static int
10503 ctl_inquiry_std(struct ctl_scsiio *ctsio)
10504 {
10505 	struct scsi_inquiry_data *inq_ptr;
10506 	struct scsi_inquiry *cdb;
10507 	struct ctl_softc *ctl_softc;
10508 	struct ctl_lun *lun;
10509 	char *val;
10510 	uint32_t alloc_len, data_len;
10511 	ctl_port_type port_type;
10512 
10513 	ctl_softc = control_softc;
10514 
10515 	/*
10516 	 * Figure out whether we're talking to a Fibre Channel port or not.
10517 	 * We treat the ioctl front end, and any SCSI adapters, as packetized
10518 	 * SCSI front ends.
10519 	 */
10520 	port_type = ctl_softc->ctl_ports[
10521 	    ctl_port_idx(ctsio->io_hdr.nexus.targ_port)]->port_type;
10522 	if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
10523 		port_type = CTL_PORT_SCSI;
10524 
10525 	lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10526 	cdb = (struct scsi_inquiry *)ctsio->cdb;
10527 	alloc_len = scsi_2btoul(cdb->length);
10528 
10529 	/*
10530 	 * We malloc the full inquiry data size here and fill it
10531 	 * in.  If the user only asks for less, we'll give him
10532 	 * that much.
10533 	 */
10534 	data_len = offsetof(struct scsi_inquiry_data, vendor_specific1);
10535 	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10536 	inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
10537 	ctsio->kern_sg_entries = 0;
10538 	ctsio->kern_data_resid = 0;
10539 	ctsio->kern_rel_offset = 0;
10540 
10541 	if (data_len < alloc_len) {
10542 		ctsio->residual = alloc_len - data_len;
10543 		ctsio->kern_data_len = data_len;
10544 		ctsio->kern_total_len = data_len;
10545 	} else {
10546 		ctsio->residual = 0;
10547 		ctsio->kern_data_len = alloc_len;
10548 		ctsio->kern_total_len = alloc_len;
10549 	}
10550 
10551 	/*
10552 	 * If we have a LUN configured, report it as connected.  Otherwise,
10553 	 * report that it is offline or no device is supported, depending
10554 	 * on the value of inquiry_pq_no_lun.
10555 	 *
10556 	 * According to the spec (SPC-4 r34), the peripheral qualifier
10557 	 * SID_QUAL_LU_OFFLINE (001b) is used in the following scenario:
10558 	 *
10559 	 * "A peripheral device having the specified peripheral device type
10560 	 * is not connected to this logical unit. However, the device
10561 	 * server is capable of supporting the specified peripheral device
10562 	 * type on this logical unit."
10563 	 *
10564 	 * According to the same spec, the peripheral qualifier
10565 	 * SID_QUAL_BAD_LU (011b) is used in this scenario:
10566 	 *
10567 	 * "The device server is not capable of supporting a peripheral
10568 	 * device on this logical unit. For this peripheral qualifier the
10569 	 * peripheral device type shall be set to 1Fh. All other peripheral
10570 	 * device type values are reserved for this peripheral qualifier."
10571 	 *
10572 	 * Given the text, it would seem that we probably want to report that
10573 	 * the LUN is offline here.  There is no LUN connected, but we can
10574 	 * support a LUN at the given LUN number.
10575 	 *
10576 	 * In the real world, though, it sounds like things are a little
10577 	 * different:
10578 	 *
10579 	 * - Linux, when presented with a LUN with the offline peripheral
10580 	 *   qualifier, will create an sg driver instance for it.  So when
10581 	 *   you attach it to CTL, you wind up with a ton of sg driver
10582 	 *   instances.  (One for every LUN that Linux bothered to probe.)
10583 	 *   Linux does this despite the fact that it issues a REPORT LUNs
10584 	 *   to LUN 0 to get the inventory of supported LUNs.
10585 	 *
10586 	 * - There is other anecdotal evidence (from Emulex folks) about
10587 	 *   arrays that use the offline peripheral qualifier for LUNs that
10588 	 *   are on the "passive" path in an active/passive array.
10589 	 *
10590 	 * So the solution is provide a hopefully reasonable default
10591 	 * (return bad/no LUN) and allow the user to change the behavior
10592 	 * with a tunable/sysctl variable.
10593 	 */
10594 	if (lun != NULL)
10595 		inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10596 				  lun->be_lun->lun_type;
10597 	else if (ctl_softc->inquiry_pq_no_lun == 0)
10598 		inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10599 	else
10600 		inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10601 
10602 	/* RMB in byte 2 is 0 */
10603 	inq_ptr->version = SCSI_REV_SPC4;
10604 
10605 	/*
10606 	 * According to SAM-3, even if a device only supports a single
10607 	 * level of LUN addressing, it should still set the HISUP bit:
10608 	 *
10609 	 * 4.9.1 Logical unit numbers overview
10610 	 *
10611 	 * All logical unit number formats described in this standard are
10612 	 * hierarchical in structure even when only a single level in that
10613 	 * hierarchy is used. The HISUP bit shall be set to one in the
10614 	 * standard INQUIRY data (see SPC-2) when any logical unit number
10615 	 * format described in this standard is used.  Non-hierarchical
10616 	 * formats are outside the scope of this standard.
10617 	 *
10618 	 * Therefore we set the HiSup bit here.
10619 	 *
10620 	 * The reponse format is 2, per SPC-3.
10621 	 */
10622 	inq_ptr->response_format = SID_HiSup | 2;
10623 
10624 	inq_ptr->additional_length = data_len -
10625 	    (offsetof(struct scsi_inquiry_data, additional_length) + 1);
10626 	CTL_DEBUG_PRINT(("additional_length = %d\n",
10627 			 inq_ptr->additional_length));
10628 
10629 	inq_ptr->spc3_flags = SPC3_SID_3PC | SPC3_SID_TPGS_IMPLICIT;
10630 	/* 16 bit addressing */
10631 	if (port_type == CTL_PORT_SCSI)
10632 		inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10633 	/* XXX set the SID_MultiP bit here if we're actually going to
10634 	   respond on multiple ports */
10635 	inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10636 
10637 	/* 16 bit data bus, synchronous transfers */
10638 	if (port_type == CTL_PORT_SCSI)
10639 		inq_ptr->flags = SID_WBus16 | SID_Sync;
10640 	/*
10641 	 * XXX KDM do we want to support tagged queueing on the control
10642 	 * device at all?
10643 	 */
10644 	if ((lun == NULL)
10645 	 || (lun->be_lun->lun_type != T_PROCESSOR))
10646 		inq_ptr->flags |= SID_CmdQue;
10647 	/*
10648 	 * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10649 	 * We have 8 bytes for the vendor name, and 16 bytes for the device
10650 	 * name and 4 bytes for the revision.
10651 	 */
10652 	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10653 	    "vendor")) == NULL) {
10654 		strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
10655 	} else {
10656 		memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10657 		strncpy(inq_ptr->vendor, val,
10658 		    min(sizeof(inq_ptr->vendor), strlen(val)));
10659 	}
10660 	if (lun == NULL) {
10661 		strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10662 		    sizeof(inq_ptr->product));
10663 	} else if ((val = ctl_get_opt(&lun->be_lun->options, "product")) == NULL) {
10664 		switch (lun->be_lun->lun_type) {
10665 		case T_DIRECT:
10666 			strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10667 			    sizeof(inq_ptr->product));
10668 			break;
10669 		case T_PROCESSOR:
10670 			strncpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT,
10671 			    sizeof(inq_ptr->product));
10672 			break;
10673 		default:
10674 			strncpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT,
10675 			    sizeof(inq_ptr->product));
10676 			break;
10677 		}
10678 	} else {
10679 		memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10680 		strncpy(inq_ptr->product, val,
10681 		    min(sizeof(inq_ptr->product), strlen(val)));
10682 	}
10683 
10684 	/*
10685 	 * XXX make this a macro somewhere so it automatically gets
10686 	 * incremented when we make changes.
10687 	 */
10688 	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10689 	    "revision")) == NULL) {
10690 		strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10691 	} else {
10692 		memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10693 		strncpy(inq_ptr->revision, val,
10694 		    min(sizeof(inq_ptr->revision), strlen(val)));
10695 	}
10696 
10697 	/*
10698 	 * For parallel SCSI, we support double transition and single
10699 	 * transition clocking.  We also support QAS (Quick Arbitration
10700 	 * and Selection) and Information Unit transfers on both the
10701 	 * control and array devices.
10702 	 */
10703 	if (port_type == CTL_PORT_SCSI)
10704 		inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10705 				    SID_SPI_IUS;
10706 
10707 	/* SAM-5 (no version claimed) */
10708 	scsi_ulto2b(0x00A0, inq_ptr->version1);
10709 	/* SPC-4 (no version claimed) */
10710 	scsi_ulto2b(0x0460, inq_ptr->version2);
10711 	if (port_type == CTL_PORT_FC) {
10712 		/* FCP-2 ANSI INCITS.350:2003 */
10713 		scsi_ulto2b(0x0917, inq_ptr->version3);
10714 	} else if (port_type == CTL_PORT_SCSI) {
10715 		/* SPI-4 ANSI INCITS.362:200x */
10716 		scsi_ulto2b(0x0B56, inq_ptr->version3);
10717 	} else if (port_type == CTL_PORT_ISCSI) {
10718 		/* iSCSI (no version claimed) */
10719 		scsi_ulto2b(0x0960, inq_ptr->version3);
10720 	} else if (port_type == CTL_PORT_SAS) {
10721 		/* SAS (no version claimed) */
10722 		scsi_ulto2b(0x0BE0, inq_ptr->version3);
10723 	}
10724 
10725 	if (lun == NULL) {
10726 		/* SBC-4 (no version claimed) */
10727 		scsi_ulto2b(0x0600, inq_ptr->version4);
10728 	} else {
10729 		switch (lun->be_lun->lun_type) {
10730 		case T_DIRECT:
10731 			/* SBC-4 (no version claimed) */
10732 			scsi_ulto2b(0x0600, inq_ptr->version4);
10733 			break;
10734 		case T_PROCESSOR:
10735 		default:
10736 			break;
10737 		}
10738 	}
10739 
10740 	ctsio->scsi_status = SCSI_STATUS_OK;
10741 	if (ctsio->kern_data_len > 0) {
10742 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10743 		ctsio->be_move_done = ctl_config_move_done;
10744 		ctl_datamove((union ctl_io *)ctsio);
10745 	} else {
10746 		ctsio->io_hdr.status = CTL_SUCCESS;
10747 		ctl_done((union ctl_io *)ctsio);
10748 	}
10749 
10750 	return (CTL_RETVAL_COMPLETE);
10751 }
10752 
10753 int
10754 ctl_inquiry(struct ctl_scsiio *ctsio)
10755 {
10756 	struct scsi_inquiry *cdb;
10757 	int retval;
10758 
10759 	CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10760 
10761 	cdb = (struct scsi_inquiry *)ctsio->cdb;
10762 	if (cdb->byte2 & SI_EVPD)
10763 		retval = ctl_inquiry_evpd(ctsio);
10764 	else if (cdb->page_code == 0)
10765 		retval = ctl_inquiry_std(ctsio);
10766 	else {
10767 		ctl_set_invalid_field(ctsio,
10768 				      /*sks_valid*/ 1,
10769 				      /*command*/ 1,
10770 				      /*field*/ 2,
10771 				      /*bit_valid*/ 0,
10772 				      /*bit*/ 0);
10773 		ctl_done((union ctl_io *)ctsio);
10774 		return (CTL_RETVAL_COMPLETE);
10775 	}
10776 
10777 	return (retval);
10778 }
10779 
10780 /*
10781  * For known CDB types, parse the LBA and length.
10782  */
10783 static int
10784 ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len)
10785 {
10786 	if (io->io_hdr.io_type != CTL_IO_SCSI)
10787 		return (1);
10788 
10789 	switch (io->scsiio.cdb[0]) {
10790 	case COMPARE_AND_WRITE: {
10791 		struct scsi_compare_and_write *cdb;
10792 
10793 		cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10794 
10795 		*lba = scsi_8btou64(cdb->addr);
10796 		*len = cdb->length;
10797 		break;
10798 	}
10799 	case READ_6:
10800 	case WRITE_6: {
10801 		struct scsi_rw_6 *cdb;
10802 
10803 		cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10804 
10805 		*lba = scsi_3btoul(cdb->addr);
10806 		/* only 5 bits are valid in the most significant address byte */
10807 		*lba &= 0x1fffff;
10808 		*len = cdb->length;
10809 		break;
10810 	}
10811 	case READ_10:
10812 	case WRITE_10: {
10813 		struct scsi_rw_10 *cdb;
10814 
10815 		cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10816 
10817 		*lba = scsi_4btoul(cdb->addr);
10818 		*len = scsi_2btoul(cdb->length);
10819 		break;
10820 	}
10821 	case WRITE_VERIFY_10: {
10822 		struct scsi_write_verify_10 *cdb;
10823 
10824 		cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10825 
10826 		*lba = scsi_4btoul(cdb->addr);
10827 		*len = scsi_2btoul(cdb->length);
10828 		break;
10829 	}
10830 	case READ_12:
10831 	case WRITE_12: {
10832 		struct scsi_rw_12 *cdb;
10833 
10834 		cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10835 
10836 		*lba = scsi_4btoul(cdb->addr);
10837 		*len = scsi_4btoul(cdb->length);
10838 		break;
10839 	}
10840 	case WRITE_VERIFY_12: {
10841 		struct scsi_write_verify_12 *cdb;
10842 
10843 		cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10844 
10845 		*lba = scsi_4btoul(cdb->addr);
10846 		*len = scsi_4btoul(cdb->length);
10847 		break;
10848 	}
10849 	case READ_16:
10850 	case WRITE_16:
10851 	case WRITE_ATOMIC_16: {
10852 		struct scsi_rw_16 *cdb;
10853 
10854 		cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10855 
10856 		*lba = scsi_8btou64(cdb->addr);
10857 		*len = scsi_4btoul(cdb->length);
10858 		break;
10859 	}
10860 	case WRITE_VERIFY_16: {
10861 		struct scsi_write_verify_16 *cdb;
10862 
10863 		cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10864 
10865 		*lba = scsi_8btou64(cdb->addr);
10866 		*len = scsi_4btoul(cdb->length);
10867 		break;
10868 	}
10869 	case WRITE_SAME_10: {
10870 		struct scsi_write_same_10 *cdb;
10871 
10872 		cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10873 
10874 		*lba = scsi_4btoul(cdb->addr);
10875 		*len = scsi_2btoul(cdb->length);
10876 		break;
10877 	}
10878 	case WRITE_SAME_16: {
10879 		struct scsi_write_same_16 *cdb;
10880 
10881 		cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10882 
10883 		*lba = scsi_8btou64(cdb->addr);
10884 		*len = scsi_4btoul(cdb->length);
10885 		break;
10886 	}
10887 	case VERIFY_10: {
10888 		struct scsi_verify_10 *cdb;
10889 
10890 		cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10891 
10892 		*lba = scsi_4btoul(cdb->addr);
10893 		*len = scsi_2btoul(cdb->length);
10894 		break;
10895 	}
10896 	case VERIFY_12: {
10897 		struct scsi_verify_12 *cdb;
10898 
10899 		cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10900 
10901 		*lba = scsi_4btoul(cdb->addr);
10902 		*len = scsi_4btoul(cdb->length);
10903 		break;
10904 	}
10905 	case VERIFY_16: {
10906 		struct scsi_verify_16 *cdb;
10907 
10908 		cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10909 
10910 		*lba = scsi_8btou64(cdb->addr);
10911 		*len = scsi_4btoul(cdb->length);
10912 		break;
10913 	}
10914 	case UNMAP: {
10915 		*lba = 0;
10916 		*len = UINT64_MAX;
10917 		break;
10918 	}
10919 	default:
10920 		return (1);
10921 		break; /* NOTREACHED */
10922 	}
10923 
10924 	return (0);
10925 }
10926 
10927 static ctl_action
10928 ctl_extent_check_lba(uint64_t lba1, uint64_t len1, uint64_t lba2, uint64_t len2)
10929 {
10930 	uint64_t endlba1, endlba2;
10931 
10932 	endlba1 = lba1 + len1 - 1;
10933 	endlba2 = lba2 + len2 - 1;
10934 
10935 	if ((endlba1 < lba2)
10936 	 || (endlba2 < lba1))
10937 		return (CTL_ACTION_PASS);
10938 	else
10939 		return (CTL_ACTION_BLOCK);
10940 }
10941 
10942 static int
10943 ctl_extent_check_unmap(union ctl_io *io, uint64_t lba2, uint64_t len2)
10944 {
10945 	struct ctl_ptr_len_flags *ptrlen;
10946 	struct scsi_unmap_desc *buf, *end, *range;
10947 	uint64_t lba;
10948 	uint32_t len;
10949 
10950 	/* If not UNMAP -- go other way. */
10951 	if (io->io_hdr.io_type != CTL_IO_SCSI ||
10952 	    io->scsiio.cdb[0] != UNMAP)
10953 		return (CTL_ACTION_ERROR);
10954 
10955 	/* If UNMAP without data -- block and wait for data. */
10956 	ptrlen = (struct ctl_ptr_len_flags *)
10957 	    &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
10958 	if ((io->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0 ||
10959 	    ptrlen->ptr == NULL)
10960 		return (CTL_ACTION_BLOCK);
10961 
10962 	/* UNMAP with data -- check for collision. */
10963 	buf = (struct scsi_unmap_desc *)ptrlen->ptr;
10964 	end = buf + ptrlen->len / sizeof(*buf);
10965 	for (range = buf; range < end; range++) {
10966 		lba = scsi_8btou64(range->lba);
10967 		len = scsi_4btoul(range->length);
10968 		if ((lba < lba2 + len2) && (lba + len > lba2))
10969 			return (CTL_ACTION_BLOCK);
10970 	}
10971 	return (CTL_ACTION_PASS);
10972 }
10973 
10974 static ctl_action
10975 ctl_extent_check(union ctl_io *io1, union ctl_io *io2)
10976 {
10977 	uint64_t lba1, lba2;
10978 	uint64_t len1, len2;
10979 	int retval;
10980 
10981 	if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10982 		return (CTL_ACTION_ERROR);
10983 
10984 	retval = ctl_extent_check_unmap(io2, lba1, len1);
10985 	if (retval != CTL_ACTION_ERROR)
10986 		return (retval);
10987 
10988 	if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10989 		return (CTL_ACTION_ERROR);
10990 
10991 	return (ctl_extent_check_lba(lba1, len1, lba2, len2));
10992 }
10993 
10994 static ctl_action
10995 ctl_check_for_blockage(struct ctl_lun *lun, union ctl_io *pending_io,
10996     union ctl_io *ooa_io)
10997 {
10998 	const struct ctl_cmd_entry *pending_entry, *ooa_entry;
10999 	ctl_serialize_action *serialize_row;
11000 
11001 	/*
11002 	 * The initiator attempted multiple untagged commands at the same
11003 	 * time.  Can't do that.
11004 	 */
11005 	if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
11006 	 && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
11007 	 && ((pending_io->io_hdr.nexus.targ_port ==
11008 	      ooa_io->io_hdr.nexus.targ_port)
11009 	  && (pending_io->io_hdr.nexus.initid.id ==
11010 	      ooa_io->io_hdr.nexus.initid.id))
11011 	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
11012 		return (CTL_ACTION_OVERLAP);
11013 
11014 	/*
11015 	 * The initiator attempted to send multiple tagged commands with
11016 	 * the same ID.  (It's fine if different initiators have the same
11017 	 * tag ID.)
11018 	 *
11019 	 * Even if all of those conditions are true, we don't kill the I/O
11020 	 * if the command ahead of us has been aborted.  We won't end up
11021 	 * sending it to the FETD, and it's perfectly legal to resend a
11022 	 * command with the same tag number as long as the previous
11023 	 * instance of this tag number has been aborted somehow.
11024 	 */
11025 	if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
11026 	 && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
11027 	 && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
11028 	 && ((pending_io->io_hdr.nexus.targ_port ==
11029 	      ooa_io->io_hdr.nexus.targ_port)
11030 	  && (pending_io->io_hdr.nexus.initid.id ==
11031 	      ooa_io->io_hdr.nexus.initid.id))
11032 	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
11033 		return (CTL_ACTION_OVERLAP_TAG);
11034 
11035 	/*
11036 	 * If we get a head of queue tag, SAM-3 says that we should
11037 	 * immediately execute it.
11038 	 *
11039 	 * What happens if this command would normally block for some other
11040 	 * reason?  e.g. a request sense with a head of queue tag
11041 	 * immediately after a write.  Normally that would block, but this
11042 	 * will result in its getting executed immediately...
11043 	 *
11044 	 * We currently return "pass" instead of "skip", so we'll end up
11045 	 * going through the rest of the queue to check for overlapped tags.
11046 	 *
11047 	 * XXX KDM check for other types of blockage first??
11048 	 */
11049 	if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
11050 		return (CTL_ACTION_PASS);
11051 
11052 	/*
11053 	 * Ordered tags have to block until all items ahead of them
11054 	 * have completed.  If we get called with an ordered tag, we always
11055 	 * block, if something else is ahead of us in the queue.
11056 	 */
11057 	if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
11058 		return (CTL_ACTION_BLOCK);
11059 
11060 	/*
11061 	 * Simple tags get blocked until all head of queue and ordered tags
11062 	 * ahead of them have completed.  I'm lumping untagged commands in
11063 	 * with simple tags here.  XXX KDM is that the right thing to do?
11064 	 */
11065 	if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
11066 	  || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
11067 	 && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
11068 	  || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
11069 		return (CTL_ACTION_BLOCK);
11070 
11071 	pending_entry = ctl_get_cmd_entry(&pending_io->scsiio, NULL);
11072 	ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio, NULL);
11073 
11074 	serialize_row = ctl_serialize_table[ooa_entry->seridx];
11075 
11076 	switch (serialize_row[pending_entry->seridx]) {
11077 	case CTL_SER_BLOCK:
11078 		return (CTL_ACTION_BLOCK);
11079 	case CTL_SER_EXTENT:
11080 		return (ctl_extent_check(pending_io, ooa_io));
11081 	case CTL_SER_EXTENTOPT:
11082 		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
11083 		    & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
11084 			return (ctl_extent_check(pending_io, ooa_io));
11085 		/* FALLTHROUGH */
11086 	case CTL_SER_PASS:
11087 		return (CTL_ACTION_PASS);
11088 	case CTL_SER_BLOCKOPT:
11089 		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
11090 		    & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
11091 			return (CTL_ACTION_BLOCK);
11092 		return (CTL_ACTION_PASS);
11093 	case CTL_SER_SKIP:
11094 		return (CTL_ACTION_SKIP);
11095 	default:
11096 		panic("invalid serialization value %d",
11097 		      serialize_row[pending_entry->seridx]);
11098 	}
11099 
11100 	return (CTL_ACTION_ERROR);
11101 }
11102 
11103 /*
11104  * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
11105  * Assumptions:
11106  * - pending_io is generally either incoming, or on the blocked queue
11107  * - starting I/O is the I/O we want to start the check with.
11108  */
11109 static ctl_action
11110 ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
11111 	      union ctl_io *starting_io)
11112 {
11113 	union ctl_io *ooa_io;
11114 	ctl_action action;
11115 
11116 	mtx_assert(&lun->lun_lock, MA_OWNED);
11117 
11118 	/*
11119 	 * Run back along the OOA queue, starting with the current
11120 	 * blocked I/O and going through every I/O before it on the
11121 	 * queue.  If starting_io is NULL, we'll just end up returning
11122 	 * CTL_ACTION_PASS.
11123 	 */
11124 	for (ooa_io = starting_io; ooa_io != NULL;
11125 	     ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
11126 	     ooa_links)){
11127 
11128 		/*
11129 		 * This routine just checks to see whether
11130 		 * cur_blocked is blocked by ooa_io, which is ahead
11131 		 * of it in the queue.  It doesn't queue/dequeue
11132 		 * cur_blocked.
11133 		 */
11134 		action = ctl_check_for_blockage(lun, pending_io, ooa_io);
11135 		switch (action) {
11136 		case CTL_ACTION_BLOCK:
11137 		case CTL_ACTION_OVERLAP:
11138 		case CTL_ACTION_OVERLAP_TAG:
11139 		case CTL_ACTION_SKIP:
11140 		case CTL_ACTION_ERROR:
11141 			return (action);
11142 			break; /* NOTREACHED */
11143 		case CTL_ACTION_PASS:
11144 			break;
11145 		default:
11146 			panic("invalid action %d", action);
11147 			break;  /* NOTREACHED */
11148 		}
11149 	}
11150 
11151 	return (CTL_ACTION_PASS);
11152 }
11153 
11154 /*
11155  * Assumptions:
11156  * - An I/O has just completed, and has been removed from the per-LUN OOA
11157  *   queue, so some items on the blocked queue may now be unblocked.
11158  */
11159 static int
11160 ctl_check_blocked(struct ctl_lun *lun)
11161 {
11162 	union ctl_io *cur_blocked, *next_blocked;
11163 
11164 	mtx_assert(&lun->lun_lock, MA_OWNED);
11165 
11166 	/*
11167 	 * Run forward from the head of the blocked queue, checking each
11168 	 * entry against the I/Os prior to it on the OOA queue to see if
11169 	 * there is still any blockage.
11170 	 *
11171 	 * We cannot use the TAILQ_FOREACH() macro, because it can't deal
11172 	 * with our removing a variable on it while it is traversing the
11173 	 * list.
11174 	 */
11175 	for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
11176 	     cur_blocked != NULL; cur_blocked = next_blocked) {
11177 		union ctl_io *prev_ooa;
11178 		ctl_action action;
11179 
11180 		next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
11181 							  blocked_links);
11182 
11183 		prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
11184 						      ctl_ooaq, ooa_links);
11185 
11186 		/*
11187 		 * If cur_blocked happens to be the first item in the OOA
11188 		 * queue now, prev_ooa will be NULL, and the action
11189 		 * returned will just be CTL_ACTION_PASS.
11190 		 */
11191 		action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
11192 
11193 		switch (action) {
11194 		case CTL_ACTION_BLOCK:
11195 			/* Nothing to do here, still blocked */
11196 			break;
11197 		case CTL_ACTION_OVERLAP:
11198 		case CTL_ACTION_OVERLAP_TAG:
11199 			/*
11200 			 * This shouldn't happen!  In theory we've already
11201 			 * checked this command for overlap...
11202 			 */
11203 			break;
11204 		case CTL_ACTION_PASS:
11205 		case CTL_ACTION_SKIP: {
11206 			struct ctl_softc *softc;
11207 			const struct ctl_cmd_entry *entry;
11208 			uint32_t initidx;
11209 			int isc_retval;
11210 
11211 			/*
11212 			 * The skip case shouldn't happen, this transaction
11213 			 * should have never made it onto the blocked queue.
11214 			 */
11215 			/*
11216 			 * This I/O is no longer blocked, we can remove it
11217 			 * from the blocked queue.  Since this is a TAILQ
11218 			 * (doubly linked list), we can do O(1) removals
11219 			 * from any place on the list.
11220 			 */
11221 			TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
11222 				     blocked_links);
11223 			cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11224 
11225 			if (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC){
11226 				/*
11227 				 * Need to send IO back to original side to
11228 				 * run
11229 				 */
11230 				union ctl_ha_msg msg_info;
11231 
11232 				msg_info.hdr.original_sc =
11233 					cur_blocked->io_hdr.original_sc;
11234 				msg_info.hdr.serializing_sc = cur_blocked;
11235 				msg_info.hdr.msg_type = CTL_MSG_R2R;
11236 				if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11237 				     &msg_info, sizeof(msg_info), 0)) >
11238 				     CTL_HA_STATUS_SUCCESS) {
11239 					printf("CTL:Check Blocked error from "
11240 					       "ctl_ha_msg_send %d\n",
11241 					       isc_retval);
11242 				}
11243 				break;
11244 			}
11245 			entry = ctl_get_cmd_entry(&cur_blocked->scsiio, NULL);
11246 			softc = control_softc;
11247 
11248 			initidx = ctl_get_initindex(&cur_blocked->io_hdr.nexus);
11249 
11250 			/*
11251 			 * Check this I/O for LUN state changes that may
11252 			 * have happened while this command was blocked.
11253 			 * The LUN state may have been changed by a command
11254 			 * ahead of us in the queue, so we need to re-check
11255 			 * for any states that can be caused by SCSI
11256 			 * commands.
11257 			 */
11258 			if (ctl_scsiio_lun_check(softc, lun, entry,
11259 						 &cur_blocked->scsiio) == 0) {
11260 				cur_blocked->io_hdr.flags |=
11261 				                      CTL_FLAG_IS_WAS_ON_RTR;
11262 				ctl_enqueue_rtr(cur_blocked);
11263 			} else
11264 				ctl_done(cur_blocked);
11265 			break;
11266 		}
11267 		default:
11268 			/*
11269 			 * This probably shouldn't happen -- we shouldn't
11270 			 * get CTL_ACTION_ERROR, or anything else.
11271 			 */
11272 			break;
11273 		}
11274 	}
11275 
11276 	return (CTL_RETVAL_COMPLETE);
11277 }
11278 
11279 /*
11280  * This routine (with one exception) checks LUN flags that can be set by
11281  * commands ahead of us in the OOA queue.  These flags have to be checked
11282  * when a command initially comes in, and when we pull a command off the
11283  * blocked queue and are preparing to execute it.  The reason we have to
11284  * check these flags for commands on the blocked queue is that the LUN
11285  * state may have been changed by a command ahead of us while we're on the
11286  * blocked queue.
11287  *
11288  * Ordering is somewhat important with these checks, so please pay
11289  * careful attention to the placement of any new checks.
11290  */
11291 static int
11292 ctl_scsiio_lun_check(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
11293     const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
11294 {
11295 	int retval;
11296 	uint32_t residx;
11297 
11298 	retval = 0;
11299 
11300 	mtx_assert(&lun->lun_lock, MA_OWNED);
11301 
11302 	/*
11303 	 * If this shelf is a secondary shelf controller, we have to reject
11304 	 * any media access commands.
11305 	 */
11306 #if 0
11307 	/* No longer needed for HA */
11308 	if (((ctl_softc->flags & CTL_FLAG_MASTER_SHELF) == 0)
11309 	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_SECONDARY) == 0)) {
11310 		ctl_set_lun_standby(ctsio);
11311 		retval = 1;
11312 		goto bailout;
11313 	}
11314 #endif
11315 
11316 	if (entry->pattern & CTL_LUN_PAT_WRITE) {
11317 		if (lun->flags & CTL_LUN_READONLY) {
11318 			ctl_set_sense(ctsio, /*current_error*/ 1,
11319 			    /*sense_key*/ SSD_KEY_DATA_PROTECT,
11320 			    /*asc*/ 0x27, /*ascq*/ 0x01, SSD_ELEM_NONE);
11321 			retval = 1;
11322 			goto bailout;
11323 		}
11324 		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT]
11325 		    .eca_and_aen & SCP_SWP) != 0) {
11326 			ctl_set_sense(ctsio, /*current_error*/ 1,
11327 			    /*sense_key*/ SSD_KEY_DATA_PROTECT,
11328 			    /*asc*/ 0x27, /*ascq*/ 0x02, SSD_ELEM_NONE);
11329 			retval = 1;
11330 			goto bailout;
11331 		}
11332 	}
11333 
11334 	/*
11335 	 * Check for a reservation conflict.  If this command isn't allowed
11336 	 * even on reserved LUNs, and if this initiator isn't the one who
11337 	 * reserved us, reject the command with a reservation conflict.
11338 	 */
11339 	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
11340 	if ((lun->flags & CTL_LUN_RESERVED)
11341 	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
11342 		if (lun->res_idx != residx) {
11343 			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
11344 			ctsio->io_hdr.status = CTL_SCSI_ERROR;
11345 			retval = 1;
11346 			goto bailout;
11347 		}
11348 	}
11349 
11350 	if ((lun->flags & CTL_LUN_PR_RESERVED)
11351 	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV) == 0)) {
11352 		/*
11353 		 * if we aren't registered or it's a res holder type
11354 		 * reservation and this isn't the res holder then set a
11355 		 * conflict.
11356 		 * NOTE: Commands which might be allowed on write exclusive
11357 		 * type reservations are checked in the particular command
11358 		 * for a conflict. Read and SSU are the only ones.
11359 		 */
11360 		if (lun->pr_keys[residx] == 0
11361 		 || (residx != lun->pr_res_idx && lun->res_type < 4)) {
11362 			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
11363 			ctsio->io_hdr.status = CTL_SCSI_ERROR;
11364 			retval = 1;
11365 			goto bailout;
11366 		}
11367 
11368 	}
11369 
11370 	if ((lun->flags & CTL_LUN_OFFLINE)
11371 	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_OFFLINE) == 0)) {
11372 		ctl_set_lun_not_ready(ctsio);
11373 		retval = 1;
11374 		goto bailout;
11375 	}
11376 
11377 	/*
11378 	 * If the LUN is stopped, see if this particular command is allowed
11379 	 * for a stopped lun.  Otherwise, reject it with 0x04,0x02.
11380 	 */
11381 	if ((lun->flags & CTL_LUN_STOPPED)
11382 	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_STOPPED) == 0)) {
11383 		/* "Logical unit not ready, initializing cmd. required" */
11384 		ctl_set_lun_stopped(ctsio);
11385 		retval = 1;
11386 		goto bailout;
11387 	}
11388 
11389 	if ((lun->flags & CTL_LUN_INOPERABLE)
11390 	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_INOPERABLE) == 0)) {
11391 		/* "Medium format corrupted" */
11392 		ctl_set_medium_format_corrupted(ctsio);
11393 		retval = 1;
11394 		goto bailout;
11395 	}
11396 
11397 bailout:
11398 	return (retval);
11399 
11400 }
11401 
11402 static void
11403 ctl_failover_io(union ctl_io *io, int have_lock)
11404 {
11405 	ctl_set_busy(&io->scsiio);
11406 	ctl_done(io);
11407 }
11408 
11409 static void
11410 ctl_failover(void)
11411 {
11412 	struct ctl_lun *lun;
11413 	struct ctl_softc *ctl_softc;
11414 	union ctl_io *next_io, *pending_io;
11415 	union ctl_io *io;
11416 	int lun_idx;
11417 	int i;
11418 
11419 	ctl_softc = control_softc;
11420 
11421 	mtx_lock(&ctl_softc->ctl_lock);
11422 	/*
11423 	 * Remove any cmds from the other SC from the rtr queue.  These
11424 	 * will obviously only be for LUNs for which we're the primary.
11425 	 * We can't send status or get/send data for these commands.
11426 	 * Since they haven't been executed yet, we can just remove them.
11427 	 * We'll either abort them or delete them below, depending on
11428 	 * which HA mode we're in.
11429 	 */
11430 #ifdef notyet
11431 	mtx_lock(&ctl_softc->queue_lock);
11432 	for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->rtr_queue);
11433 	     io != NULL; io = next_io) {
11434 		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
11435 		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
11436 			STAILQ_REMOVE(&ctl_softc->rtr_queue, &io->io_hdr,
11437 				      ctl_io_hdr, links);
11438 	}
11439 	mtx_unlock(&ctl_softc->queue_lock);
11440 #endif
11441 
11442 	for (lun_idx=0; lun_idx < ctl_softc->num_luns; lun_idx++) {
11443 		lun = ctl_softc->ctl_luns[lun_idx];
11444 		if (lun==NULL)
11445 			continue;
11446 
11447 		/*
11448 		 * Processor LUNs are primary on both sides.
11449 		 * XXX will this always be true?
11450 		 */
11451 		if (lun->be_lun->lun_type == T_PROCESSOR)
11452 			continue;
11453 
11454 		if ((lun->flags & CTL_LUN_PRIMARY_SC)
11455 		 && (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11456 			printf("FAILOVER: primary lun %d\n", lun_idx);
11457 		        /*
11458 			 * Remove all commands from the other SC. First from the
11459 			 * blocked queue then from the ooa queue. Once we have
11460 			 * removed them. Call ctl_check_blocked to see if there
11461 			 * is anything that can run.
11462 			 */
11463 			for (io = (union ctl_io *)TAILQ_FIRST(
11464 			     &lun->blocked_queue); io != NULL; io = next_io) {
11465 
11466 		        	next_io = (union ctl_io *)TAILQ_NEXT(
11467 				    &io->io_hdr, blocked_links);
11468 
11469 				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11470 					TAILQ_REMOVE(&lun->blocked_queue,
11471 						     &io->io_hdr,blocked_links);
11472 					io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11473 					TAILQ_REMOVE(&lun->ooa_queue,
11474 						     &io->io_hdr, ooa_links);
11475 
11476 					ctl_free_io(io);
11477 				}
11478 			}
11479 
11480 			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11481 	     		     io != NULL; io = next_io) {
11482 
11483 		        	next_io = (union ctl_io *)TAILQ_NEXT(
11484 				    &io->io_hdr, ooa_links);
11485 
11486 				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11487 
11488 					TAILQ_REMOVE(&lun->ooa_queue,
11489 						&io->io_hdr,
11490 					     	ooa_links);
11491 
11492 					ctl_free_io(io);
11493 				}
11494 			}
11495 			ctl_check_blocked(lun);
11496 		} else if ((lun->flags & CTL_LUN_PRIMARY_SC)
11497 			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
11498 
11499 			printf("FAILOVER: primary lun %d\n", lun_idx);
11500 			/*
11501 			 * Abort all commands from the other SC.  We can't
11502 			 * send status back for them now.  These should get
11503 			 * cleaned up when they are completed or come out
11504 			 * for a datamove operation.
11505 			 */
11506 			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11507 	     		     io != NULL; io = next_io) {
11508 		        	next_io = (union ctl_io *)TAILQ_NEXT(
11509 					&io->io_hdr, ooa_links);
11510 
11511 				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
11512 					io->io_hdr.flags |= CTL_FLAG_ABORT;
11513 			}
11514 		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11515 			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
11516 
11517 			printf("FAILOVER: secondary lun %d\n", lun_idx);
11518 
11519 			lun->flags |= CTL_LUN_PRIMARY_SC;
11520 
11521 			/*
11522 			 * We send all I/O that was sent to this controller
11523 			 * and redirected to the other side back with
11524 			 * busy status, and have the initiator retry it.
11525 			 * Figuring out how much data has been transferred,
11526 			 * etc. and picking up where we left off would be
11527 			 * very tricky.
11528 			 *
11529 			 * XXX KDM need to remove I/O from the blocked
11530 			 * queue as well!
11531 			 */
11532 			for (pending_io = (union ctl_io *)TAILQ_FIRST(
11533 			     &lun->ooa_queue); pending_io != NULL;
11534 			     pending_io = next_io) {
11535 
11536 				next_io =  (union ctl_io *)TAILQ_NEXT(
11537 					&pending_io->io_hdr, ooa_links);
11538 
11539 				pending_io->io_hdr.flags &=
11540 					~CTL_FLAG_SENT_2OTHER_SC;
11541 
11542 				if (pending_io->io_hdr.flags &
11543 				    CTL_FLAG_IO_ACTIVE) {
11544 					pending_io->io_hdr.flags |=
11545 						CTL_FLAG_FAILOVER;
11546 				} else {
11547 					ctl_set_busy(&pending_io->scsiio);
11548 					ctl_done(pending_io);
11549 				}
11550 			}
11551 
11552 			/*
11553 			 * Build Unit Attention
11554 			 */
11555 			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11556 				lun->pending_ua[i] |=
11557 				                     CTL_UA_ASYM_ACC_CHANGE;
11558 			}
11559 		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11560 			&& (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11561 			printf("FAILOVER: secondary lun %d\n", lun_idx);
11562 			/*
11563 			 * if the first io on the OOA is not on the RtR queue
11564 			 * add it.
11565 			 */
11566 			lun->flags |= CTL_LUN_PRIMARY_SC;
11567 
11568 			pending_io = (union ctl_io *)TAILQ_FIRST(
11569 			    &lun->ooa_queue);
11570 			if (pending_io==NULL) {
11571 				printf("Nothing on OOA queue\n");
11572 				continue;
11573 			}
11574 
11575 			pending_io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11576 			if ((pending_io->io_hdr.flags &
11577 			     CTL_FLAG_IS_WAS_ON_RTR) == 0) {
11578 				pending_io->io_hdr.flags |=
11579 				    CTL_FLAG_IS_WAS_ON_RTR;
11580 				ctl_enqueue_rtr(pending_io);
11581 			}
11582 #if 0
11583 			else
11584 			{
11585 				printf("Tag 0x%04x is running\n",
11586 				      pending_io->scsiio.tag_num);
11587 			}
11588 #endif
11589 
11590 			next_io = (union ctl_io *)TAILQ_NEXT(
11591 			    &pending_io->io_hdr, ooa_links);
11592 			for (pending_io=next_io; pending_io != NULL;
11593 			     pending_io = next_io) {
11594 				pending_io->io_hdr.flags &=
11595 				    ~CTL_FLAG_SENT_2OTHER_SC;
11596 				next_io = (union ctl_io *)TAILQ_NEXT(
11597 					&pending_io->io_hdr, ooa_links);
11598 				if (pending_io->io_hdr.flags &
11599 				    CTL_FLAG_IS_WAS_ON_RTR) {
11600 #if 0
11601 				        printf("Tag 0x%04x is running\n",
11602 				      		pending_io->scsiio.tag_num);
11603 #endif
11604 					continue;
11605 				}
11606 
11607 				switch (ctl_check_ooa(lun, pending_io,
11608 			            (union ctl_io *)TAILQ_PREV(
11609 				    &pending_io->io_hdr, ctl_ooaq,
11610 				    ooa_links))) {
11611 
11612 				case CTL_ACTION_BLOCK:
11613 					TAILQ_INSERT_TAIL(&lun->blocked_queue,
11614 							  &pending_io->io_hdr,
11615 							  blocked_links);
11616 					pending_io->io_hdr.flags |=
11617 					    CTL_FLAG_BLOCKED;
11618 					break;
11619 				case CTL_ACTION_PASS:
11620 				case CTL_ACTION_SKIP:
11621 					pending_io->io_hdr.flags |=
11622 					    CTL_FLAG_IS_WAS_ON_RTR;
11623 					ctl_enqueue_rtr(pending_io);
11624 					break;
11625 				case CTL_ACTION_OVERLAP:
11626 					ctl_set_overlapped_cmd(
11627 					    (struct ctl_scsiio *)pending_io);
11628 					ctl_done(pending_io);
11629 					break;
11630 				case CTL_ACTION_OVERLAP_TAG:
11631 					ctl_set_overlapped_tag(
11632 					    (struct ctl_scsiio *)pending_io,
11633 					    pending_io->scsiio.tag_num & 0xff);
11634 					ctl_done(pending_io);
11635 					break;
11636 				case CTL_ACTION_ERROR:
11637 				default:
11638 					ctl_set_internal_failure(
11639 						(struct ctl_scsiio *)pending_io,
11640 						0,  // sks_valid
11641 						0); //retry count
11642 					ctl_done(pending_io);
11643 					break;
11644 				}
11645 			}
11646 
11647 			/*
11648 			 * Build Unit Attention
11649 			 */
11650 			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11651 				lun->pending_ua[i] |=
11652 				                     CTL_UA_ASYM_ACC_CHANGE;
11653 			}
11654 		} else {
11655 			panic("Unhandled HA mode failover, LUN flags = %#x, "
11656 			      "ha_mode = #%x", lun->flags, ctl_softc->ha_mode);
11657 		}
11658 	}
11659 	ctl_pause_rtr = 0;
11660 	mtx_unlock(&ctl_softc->ctl_lock);
11661 }
11662 
11663 static int
11664 ctl_scsiio_precheck(struct ctl_softc *ctl_softc, struct ctl_scsiio *ctsio)
11665 {
11666 	struct ctl_lun *lun;
11667 	const struct ctl_cmd_entry *entry;
11668 	uint32_t initidx, targ_lun;
11669 	int retval;
11670 
11671 	retval = 0;
11672 
11673 	lun = NULL;
11674 
11675 	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11676 	if ((targ_lun < CTL_MAX_LUNS)
11677 	 && (ctl_softc->ctl_luns[targ_lun] != NULL)) {
11678 		lun = ctl_softc->ctl_luns[targ_lun];
11679 		/*
11680 		 * If the LUN is invalid, pretend that it doesn't exist.
11681 		 * It will go away as soon as all pending I/O has been
11682 		 * completed.
11683 		 */
11684 		if (lun->flags & CTL_LUN_DISABLED) {
11685 			lun = NULL;
11686 		} else {
11687 			ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
11688 			ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
11689 				lun->be_lun;
11690 			if (lun->be_lun->lun_type == T_PROCESSOR) {
11691 				ctsio->io_hdr.flags |= CTL_FLAG_CONTROL_DEV;
11692 			}
11693 
11694 			/*
11695 			 * Every I/O goes into the OOA queue for a
11696 			 * particular LUN, and stays there until completion.
11697 			 */
11698 			mtx_lock(&lun->lun_lock);
11699 			TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr,
11700 			    ooa_links);
11701 		}
11702 	} else {
11703 		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11704 		ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11705 	}
11706 
11707 	/* Get command entry and return error if it is unsuppotyed. */
11708 	entry = ctl_validate_command(ctsio);
11709 	if (entry == NULL) {
11710 		if (lun)
11711 			mtx_unlock(&lun->lun_lock);
11712 		return (retval);
11713 	}
11714 
11715 	ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11716 	ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11717 
11718 	/*
11719 	 * Check to see whether we can send this command to LUNs that don't
11720 	 * exist.  This should pretty much only be the case for inquiry
11721 	 * and request sense.  Further checks, below, really require having
11722 	 * a LUN, so we can't really check the command anymore.  Just put
11723 	 * it on the rtr queue.
11724 	 */
11725 	if (lun == NULL) {
11726 		if (entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) {
11727 			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11728 			ctl_enqueue_rtr((union ctl_io *)ctsio);
11729 			return (retval);
11730 		}
11731 
11732 		ctl_set_unsupported_lun(ctsio);
11733 		ctl_done((union ctl_io *)ctsio);
11734 		CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11735 		return (retval);
11736 	} else {
11737 		/*
11738 		 * Make sure we support this particular command on this LUN.
11739 		 * e.g., we don't support writes to the control LUN.
11740 		 */
11741 		if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11742 			mtx_unlock(&lun->lun_lock);
11743 			ctl_set_invalid_opcode(ctsio);
11744 			ctl_done((union ctl_io *)ctsio);
11745 			return (retval);
11746 		}
11747 	}
11748 
11749 	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11750 
11751 #ifdef CTL_WITH_CA
11752 	/*
11753 	 * If we've got a request sense, it'll clear the contingent
11754 	 * allegiance condition.  Otherwise, if we have a CA condition for
11755 	 * this initiator, clear it, because it sent down a command other
11756 	 * than request sense.
11757 	 */
11758 	if ((ctsio->cdb[0] != REQUEST_SENSE)
11759 	 && (ctl_is_set(lun->have_ca, initidx)))
11760 		ctl_clear_mask(lun->have_ca, initidx);
11761 #endif
11762 
11763 	/*
11764 	 * If the command has this flag set, it handles its own unit
11765 	 * attention reporting, we shouldn't do anything.  Otherwise we
11766 	 * check for any pending unit attentions, and send them back to the
11767 	 * initiator.  We only do this when a command initially comes in,
11768 	 * not when we pull it off the blocked queue.
11769 	 *
11770 	 * According to SAM-3, section 5.3.2, the order that things get
11771 	 * presented back to the host is basically unit attentions caused
11772 	 * by some sort of reset event, busy status, reservation conflicts
11773 	 * or task set full, and finally any other status.
11774 	 *
11775 	 * One issue here is that some of the unit attentions we report
11776 	 * don't fall into the "reset" category (e.g. "reported luns data
11777 	 * has changed").  So reporting it here, before the reservation
11778 	 * check, may be technically wrong.  I guess the only thing to do
11779 	 * would be to check for and report the reset events here, and then
11780 	 * check for the other unit attention types after we check for a
11781 	 * reservation conflict.
11782 	 *
11783 	 * XXX KDM need to fix this
11784 	 */
11785 	if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11786 		ctl_ua_type ua_type;
11787 
11788 		if (lun->pending_ua[initidx] != CTL_UA_NONE) {
11789 			scsi_sense_data_type sense_format;
11790 
11791 			if (lun != NULL)
11792 				sense_format = (lun->flags &
11793 				    CTL_LUN_SENSE_DESC) ? SSD_TYPE_DESC :
11794 				    SSD_TYPE_FIXED;
11795 			else
11796 				sense_format = SSD_TYPE_FIXED;
11797 
11798 			ua_type = ctl_build_ua(&lun->pending_ua[initidx],
11799 			    &ctsio->sense_data, sense_format);
11800 			if (ua_type != CTL_UA_NONE) {
11801 				ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11802 				ctsio->io_hdr.status = CTL_SCSI_ERROR |
11803 						       CTL_AUTOSENSE;
11804 				ctsio->sense_len = SSD_FULL_SIZE;
11805 				mtx_unlock(&lun->lun_lock);
11806 				ctl_done((union ctl_io *)ctsio);
11807 				return (retval);
11808 			}
11809 		}
11810 	}
11811 
11812 
11813 	if (ctl_scsiio_lun_check(ctl_softc, lun, entry, ctsio) != 0) {
11814 		mtx_unlock(&lun->lun_lock);
11815 		ctl_done((union ctl_io *)ctsio);
11816 		return (retval);
11817 	}
11818 
11819 	/*
11820 	 * XXX CHD this is where we want to send IO to other side if
11821 	 * this LUN is secondary on this SC. We will need to make a copy
11822 	 * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11823 	 * the copy we send as FROM_OTHER.
11824 	 * We also need to stuff the address of the original IO so we can
11825 	 * find it easily. Something similar will need be done on the other
11826 	 * side so when we are done we can find the copy.
11827 	 */
11828 	if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11829 		union ctl_ha_msg msg_info;
11830 		int isc_retval;
11831 
11832 		ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11833 
11834 		msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11835 		msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11836 #if 0
11837 		printf("1. ctsio %p\n", ctsio);
11838 #endif
11839 		msg_info.hdr.serializing_sc = NULL;
11840 		msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11841 		msg_info.scsi.tag_num = ctsio->tag_num;
11842 		msg_info.scsi.tag_type = ctsio->tag_type;
11843 		memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11844 
11845 		ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11846 
11847 		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11848 		    (void *)&msg_info, sizeof(msg_info), 0)) >
11849 		    CTL_HA_STATUS_SUCCESS) {
11850 			printf("CTL:precheck, ctl_ha_msg_send returned %d\n",
11851 			       isc_retval);
11852 			printf("CTL:opcode is %x\n", ctsio->cdb[0]);
11853 		} else {
11854 #if 0
11855 			printf("CTL:Precheck sent msg, opcode is %x\n",opcode);
11856 #endif
11857 		}
11858 
11859 		/*
11860 		 * XXX KDM this I/O is off the incoming queue, but hasn't
11861 		 * been inserted on any other queue.  We may need to come
11862 		 * up with a holding queue while we wait for serialization
11863 		 * so that we have an idea of what we're waiting for from
11864 		 * the other side.
11865 		 */
11866 		mtx_unlock(&lun->lun_lock);
11867 		return (retval);
11868 	}
11869 
11870 	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11871 			      (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11872 			      ctl_ooaq, ooa_links))) {
11873 	case CTL_ACTION_BLOCK:
11874 		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11875 		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11876 				  blocked_links);
11877 		mtx_unlock(&lun->lun_lock);
11878 		return (retval);
11879 	case CTL_ACTION_PASS:
11880 	case CTL_ACTION_SKIP:
11881 		ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11882 		mtx_unlock(&lun->lun_lock);
11883 		ctl_enqueue_rtr((union ctl_io *)ctsio);
11884 		break;
11885 	case CTL_ACTION_OVERLAP:
11886 		mtx_unlock(&lun->lun_lock);
11887 		ctl_set_overlapped_cmd(ctsio);
11888 		ctl_done((union ctl_io *)ctsio);
11889 		break;
11890 	case CTL_ACTION_OVERLAP_TAG:
11891 		mtx_unlock(&lun->lun_lock);
11892 		ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11893 		ctl_done((union ctl_io *)ctsio);
11894 		break;
11895 	case CTL_ACTION_ERROR:
11896 	default:
11897 		mtx_unlock(&lun->lun_lock);
11898 		ctl_set_internal_failure(ctsio,
11899 					 /*sks_valid*/ 0,
11900 					 /*retry_count*/ 0);
11901 		ctl_done((union ctl_io *)ctsio);
11902 		break;
11903 	}
11904 	return (retval);
11905 }
11906 
11907 const struct ctl_cmd_entry *
11908 ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa)
11909 {
11910 	const struct ctl_cmd_entry *entry;
11911 	int service_action;
11912 
11913 	entry = &ctl_cmd_table[ctsio->cdb[0]];
11914 	if (sa)
11915 		*sa = ((entry->flags & CTL_CMD_FLAG_SA5) != 0);
11916 	if (entry->flags & CTL_CMD_FLAG_SA5) {
11917 		service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11918 		entry = &((const struct ctl_cmd_entry *)
11919 		    entry->execute)[service_action];
11920 	}
11921 	return (entry);
11922 }
11923 
11924 const struct ctl_cmd_entry *
11925 ctl_validate_command(struct ctl_scsiio *ctsio)
11926 {
11927 	const struct ctl_cmd_entry *entry;
11928 	int i, sa;
11929 	uint8_t diff;
11930 
11931 	entry = ctl_get_cmd_entry(ctsio, &sa);
11932 	if (entry->execute == NULL) {
11933 		if (sa)
11934 			ctl_set_invalid_field(ctsio,
11935 					      /*sks_valid*/ 1,
11936 					      /*command*/ 1,
11937 					      /*field*/ 1,
11938 					      /*bit_valid*/ 1,
11939 					      /*bit*/ 4);
11940 		else
11941 			ctl_set_invalid_opcode(ctsio);
11942 		ctl_done((union ctl_io *)ctsio);
11943 		return (NULL);
11944 	}
11945 	KASSERT(entry->length > 0,
11946 	    ("Not defined length for command 0x%02x/0x%02x",
11947 	     ctsio->cdb[0], ctsio->cdb[1]));
11948 	for (i = 1; i < entry->length; i++) {
11949 		diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11950 		if (diff == 0)
11951 			continue;
11952 		ctl_set_invalid_field(ctsio,
11953 				      /*sks_valid*/ 1,
11954 				      /*command*/ 1,
11955 				      /*field*/ i,
11956 				      /*bit_valid*/ 1,
11957 				      /*bit*/ fls(diff) - 1);
11958 		ctl_done((union ctl_io *)ctsio);
11959 		return (NULL);
11960 	}
11961 	return (entry);
11962 }
11963 
11964 static int
11965 ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11966 {
11967 
11968 	switch (lun_type) {
11969 	case T_PROCESSOR:
11970 		if (((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0) &&
11971 		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11972 			return (0);
11973 		break;
11974 	case T_DIRECT:
11975 		if (((entry->flags & CTL_CMD_FLAG_OK_ON_SLUN) == 0) &&
11976 		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11977 			return (0);
11978 		break;
11979 	default:
11980 		return (0);
11981 	}
11982 	return (1);
11983 }
11984 
11985 static int
11986 ctl_scsiio(struct ctl_scsiio *ctsio)
11987 {
11988 	int retval;
11989 	const struct ctl_cmd_entry *entry;
11990 
11991 	retval = CTL_RETVAL_COMPLETE;
11992 
11993 	CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11994 
11995 	entry = ctl_get_cmd_entry(ctsio, NULL);
11996 
11997 	/*
11998 	 * If this I/O has been aborted, just send it straight to
11999 	 * ctl_done() without executing it.
12000 	 */
12001 	if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
12002 		ctl_done((union ctl_io *)ctsio);
12003 		goto bailout;
12004 	}
12005 
12006 	/*
12007 	 * All the checks should have been handled by ctl_scsiio_precheck().
12008 	 * We should be clear now to just execute the I/O.
12009 	 */
12010 	retval = entry->execute(ctsio);
12011 
12012 bailout:
12013 	return (retval);
12014 }
12015 
12016 /*
12017  * Since we only implement one target right now, a bus reset simply resets
12018  * our single target.
12019  */
12020 static int
12021 ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io)
12022 {
12023 	return(ctl_target_reset(ctl_softc, io, CTL_UA_BUS_RESET));
12024 }
12025 
12026 static int
12027 ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
12028 		 ctl_ua_type ua_type)
12029 {
12030 	struct ctl_lun *lun;
12031 	int retval;
12032 
12033 	if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12034 		union ctl_ha_msg msg_info;
12035 
12036 		io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
12037 		msg_info.hdr.nexus = io->io_hdr.nexus;
12038 		if (ua_type==CTL_UA_TARG_RESET)
12039 			msg_info.task.task_action = CTL_TASK_TARGET_RESET;
12040 		else
12041 			msg_info.task.task_action = CTL_TASK_BUS_RESET;
12042 		msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12043 		msg_info.hdr.original_sc = NULL;
12044 		msg_info.hdr.serializing_sc = NULL;
12045 		if (CTL_HA_STATUS_SUCCESS != ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12046 		    (void *)&msg_info, sizeof(msg_info), 0)) {
12047 		}
12048 	}
12049 	retval = 0;
12050 
12051 	mtx_lock(&ctl_softc->ctl_lock);
12052 	STAILQ_FOREACH(lun, &ctl_softc->lun_list, links)
12053 		retval += ctl_lun_reset(lun, io, ua_type);
12054 	mtx_unlock(&ctl_softc->ctl_lock);
12055 
12056 	return (retval);
12057 }
12058 
12059 /*
12060  * The LUN should always be set.  The I/O is optional, and is used to
12061  * distinguish between I/Os sent by this initiator, and by other
12062  * initiators.  We set unit attention for initiators other than this one.
12063  * SAM-3 is vague on this point.  It does say that a unit attention should
12064  * be established for other initiators when a LUN is reset (see section
12065  * 5.7.3), but it doesn't specifically say that the unit attention should
12066  * be established for this particular initiator when a LUN is reset.  Here
12067  * is the relevant text, from SAM-3 rev 8:
12068  *
12069  * 5.7.2 When a SCSI initiator port aborts its own tasks
12070  *
12071  * When a SCSI initiator port causes its own task(s) to be aborted, no
12072  * notification that the task(s) have been aborted shall be returned to
12073  * the SCSI initiator port other than the completion response for the
12074  * command or task management function action that caused the task(s) to
12075  * be aborted and notification(s) associated with related effects of the
12076  * action (e.g., a reset unit attention condition).
12077  *
12078  * XXX KDM for now, we're setting unit attention for all initiators.
12079  */
12080 static int
12081 ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
12082 {
12083 	union ctl_io *xio;
12084 #if 0
12085 	uint32_t initindex;
12086 #endif
12087 	int i;
12088 
12089 	mtx_lock(&lun->lun_lock);
12090 	/*
12091 	 * Run through the OOA queue and abort each I/O.
12092 	 */
12093 #if 0
12094 	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
12095 #endif
12096 	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12097 	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12098 		xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
12099 	}
12100 
12101 	/*
12102 	 * This version sets unit attention for every
12103 	 */
12104 #if 0
12105 	initindex = ctl_get_initindex(&io->io_hdr.nexus);
12106 	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
12107 		if (initindex == i)
12108 			continue;
12109 		lun->pending_ua[i] |= ua_type;
12110 	}
12111 #endif
12112 
12113 	/*
12114 	 * A reset (any kind, really) clears reservations established with
12115 	 * RESERVE/RELEASE.  It does not clear reservations established
12116 	 * with PERSISTENT RESERVE OUT, but we don't support that at the
12117 	 * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
12118 	 * reservations made with the RESERVE/RELEASE commands, because
12119 	 * those commands are obsolete in SPC-3.
12120 	 */
12121 	lun->flags &= ~CTL_LUN_RESERVED;
12122 
12123 	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
12124 #ifdef CTL_WITH_CA
12125 		ctl_clear_mask(lun->have_ca, i);
12126 #endif
12127 		lun->pending_ua[i] |= ua_type;
12128 	}
12129 	mtx_unlock(&lun->lun_lock);
12130 
12131 	return (0);
12132 }
12133 
12134 static void
12135 ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
12136     int other_sc)
12137 {
12138 	union ctl_io *xio;
12139 
12140 	mtx_assert(&lun->lun_lock, MA_OWNED);
12141 
12142 	/*
12143 	 * Run through the OOA queue and attempt to find the given I/O.
12144 	 * The target port, initiator ID, tag type and tag number have to
12145 	 * match the values that we got from the initiator.  If we have an
12146 	 * untagged command to abort, simply abort the first untagged command
12147 	 * we come to.  We only allow one untagged command at a time of course.
12148 	 */
12149 	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12150 	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12151 
12152 		if ((targ_port == UINT32_MAX ||
12153 		     targ_port == xio->io_hdr.nexus.targ_port) &&
12154 		    (init_id == UINT32_MAX ||
12155 		     init_id == xio->io_hdr.nexus.initid.id)) {
12156 			if (targ_port != xio->io_hdr.nexus.targ_port ||
12157 			    init_id != xio->io_hdr.nexus.initid.id)
12158 				xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS;
12159 			xio->io_hdr.flags |= CTL_FLAG_ABORT;
12160 			if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12161 				union ctl_ha_msg msg_info;
12162 
12163 				msg_info.hdr.nexus = xio->io_hdr.nexus;
12164 				msg_info.task.task_action = CTL_TASK_ABORT_TASK;
12165 				msg_info.task.tag_num = xio->scsiio.tag_num;
12166 				msg_info.task.tag_type = xio->scsiio.tag_type;
12167 				msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12168 				msg_info.hdr.original_sc = NULL;
12169 				msg_info.hdr.serializing_sc = NULL;
12170 				ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12171 				    (void *)&msg_info, sizeof(msg_info), 0);
12172 			}
12173 		}
12174 	}
12175 }
12176 
12177 static int
12178 ctl_abort_task_set(union ctl_io *io)
12179 {
12180 	struct ctl_softc *softc = control_softc;
12181 	struct ctl_lun *lun;
12182 	uint32_t targ_lun;
12183 
12184 	/*
12185 	 * Look up the LUN.
12186 	 */
12187 	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12188 	mtx_lock(&softc->ctl_lock);
12189 	if ((targ_lun < CTL_MAX_LUNS) && (softc->ctl_luns[targ_lun] != NULL))
12190 		lun = softc->ctl_luns[targ_lun];
12191 	else {
12192 		mtx_unlock(&softc->ctl_lock);
12193 		return (1);
12194 	}
12195 
12196 	mtx_lock(&lun->lun_lock);
12197 	mtx_unlock(&softc->ctl_lock);
12198 	if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
12199 		ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12200 		    io->io_hdr.nexus.initid.id,
12201 		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12202 	} else { /* CTL_TASK_CLEAR_TASK_SET */
12203 		ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
12204 		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12205 	}
12206 	mtx_unlock(&lun->lun_lock);
12207 	return (0);
12208 }
12209 
12210 static int
12211 ctl_i_t_nexus_reset(union ctl_io *io)
12212 {
12213 	struct ctl_softc *softc = control_softc;
12214 	struct ctl_lun *lun;
12215 	uint32_t initindex, residx;
12216 
12217 	initindex = ctl_get_initindex(&io->io_hdr.nexus);
12218 	residx = ctl_get_resindex(&io->io_hdr.nexus);
12219 	mtx_lock(&softc->ctl_lock);
12220 	STAILQ_FOREACH(lun, &softc->lun_list, links) {
12221 		mtx_lock(&lun->lun_lock);
12222 		ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12223 		    io->io_hdr.nexus.initid.id,
12224 		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12225 #ifdef CTL_WITH_CA
12226 		ctl_clear_mask(lun->have_ca, initindex);
12227 #endif
12228 		if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
12229 			lun->flags &= ~CTL_LUN_RESERVED;
12230 		lun->pending_ua[initindex] |= CTL_UA_I_T_NEXUS_LOSS;
12231 		mtx_unlock(&lun->lun_lock);
12232 	}
12233 	mtx_unlock(&softc->ctl_lock);
12234 	return (0);
12235 }
12236 
12237 static int
12238 ctl_abort_task(union ctl_io *io)
12239 {
12240 	union ctl_io *xio;
12241 	struct ctl_lun *lun;
12242 	struct ctl_softc *ctl_softc;
12243 #if 0
12244 	struct sbuf sb;
12245 	char printbuf[128];
12246 #endif
12247 	int found;
12248 	uint32_t targ_lun;
12249 
12250 	ctl_softc = control_softc;
12251 	found = 0;
12252 
12253 	/*
12254 	 * Look up the LUN.
12255 	 */
12256 	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12257 	mtx_lock(&ctl_softc->ctl_lock);
12258 	if ((targ_lun < CTL_MAX_LUNS)
12259 	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
12260 		lun = ctl_softc->ctl_luns[targ_lun];
12261 	else {
12262 		mtx_unlock(&ctl_softc->ctl_lock);
12263 		return (1);
12264 	}
12265 
12266 #if 0
12267 	printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
12268 	       lun->lun, io->taskio.tag_num, io->taskio.tag_type);
12269 #endif
12270 
12271 	mtx_lock(&lun->lun_lock);
12272 	mtx_unlock(&ctl_softc->ctl_lock);
12273 	/*
12274 	 * Run through the OOA queue and attempt to find the given I/O.
12275 	 * The target port, initiator ID, tag type and tag number have to
12276 	 * match the values that we got from the initiator.  If we have an
12277 	 * untagged command to abort, simply abort the first untagged command
12278 	 * we come to.  We only allow one untagged command at a time of course.
12279 	 */
12280 #if 0
12281 	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
12282 #endif
12283 	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12284 	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12285 #if 0
12286 		sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
12287 
12288 		sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
12289 			    lun->lun, xio->scsiio.tag_num,
12290 			    xio->scsiio.tag_type,
12291 			    (xio->io_hdr.blocked_links.tqe_prev
12292 			    == NULL) ? "" : " BLOCKED",
12293 			    (xio->io_hdr.flags &
12294 			    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
12295 			    (xio->io_hdr.flags &
12296 			    CTL_FLAG_ABORT) ? " ABORT" : "",
12297 			    (xio->io_hdr.flags &
12298 			    CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
12299 		ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
12300 		sbuf_finish(&sb);
12301 		printf("%s\n", sbuf_data(&sb));
12302 #endif
12303 
12304 		if ((xio->io_hdr.nexus.targ_port == io->io_hdr.nexus.targ_port)
12305 		 && (xio->io_hdr.nexus.initid.id ==
12306 		     io->io_hdr.nexus.initid.id)) {
12307 			/*
12308 			 * If the abort says that the task is untagged, the
12309 			 * task in the queue must be untagged.  Otherwise,
12310 			 * we just check to see whether the tag numbers
12311 			 * match.  This is because the QLogic firmware
12312 			 * doesn't pass back the tag type in an abort
12313 			 * request.
12314 			 */
12315 #if 0
12316 			if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
12317 			  && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
12318 			 || (xio->scsiio.tag_num == io->taskio.tag_num)) {
12319 #endif
12320 			/*
12321 			 * XXX KDM we've got problems with FC, because it
12322 			 * doesn't send down a tag type with aborts.  So we
12323 			 * can only really go by the tag number...
12324 			 * This may cause problems with parallel SCSI.
12325 			 * Need to figure that out!!
12326 			 */
12327 			if (xio->scsiio.tag_num == io->taskio.tag_num) {
12328 				xio->io_hdr.flags |= CTL_FLAG_ABORT;
12329 				found = 1;
12330 				if ((io->io_hdr.flags &
12331 				     CTL_FLAG_FROM_OTHER_SC) == 0 &&
12332 				    !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12333 					union ctl_ha_msg msg_info;
12334 
12335 					io->io_hdr.flags |=
12336 					                CTL_FLAG_SENT_2OTHER_SC;
12337 					msg_info.hdr.nexus = io->io_hdr.nexus;
12338 					msg_info.task.task_action =
12339 						CTL_TASK_ABORT_TASK;
12340 					msg_info.task.tag_num =
12341 						io->taskio.tag_num;
12342 					msg_info.task.tag_type =
12343 						io->taskio.tag_type;
12344 					msg_info.hdr.msg_type =
12345 						CTL_MSG_MANAGE_TASKS;
12346 					msg_info.hdr.original_sc = NULL;
12347 					msg_info.hdr.serializing_sc = NULL;
12348 #if 0
12349 					printf("Sent Abort to other side\n");
12350 #endif
12351 					if (CTL_HA_STATUS_SUCCESS !=
12352 					        ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12353 		    				(void *)&msg_info,
12354 						sizeof(msg_info), 0)) {
12355 					}
12356 				}
12357 #if 0
12358 				printf("ctl_abort_task: found I/O to abort\n");
12359 #endif
12360 				break;
12361 			}
12362 		}
12363 	}
12364 	mtx_unlock(&lun->lun_lock);
12365 
12366 	if (found == 0) {
12367 		/*
12368 		 * This isn't really an error.  It's entirely possible for
12369 		 * the abort and command completion to cross on the wire.
12370 		 * This is more of an informative/diagnostic error.
12371 		 */
12372 #if 0
12373 		printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
12374 		       "%d:%d:%d:%d tag %d type %d\n",
12375 		       io->io_hdr.nexus.initid.id,
12376 		       io->io_hdr.nexus.targ_port,
12377 		       io->io_hdr.nexus.targ_target.id,
12378 		       io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
12379 		       io->taskio.tag_type);
12380 #endif
12381 	}
12382 	return (0);
12383 }
12384 
12385 static void
12386 ctl_run_task(union ctl_io *io)
12387 {
12388 	struct ctl_softc *ctl_softc = control_softc;
12389 	int retval = 1;
12390 	const char *task_desc;
12391 
12392 	CTL_DEBUG_PRINT(("ctl_run_task\n"));
12393 
12394 	KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
12395 	    ("ctl_run_task: Unextected io_type %d\n",
12396 	     io->io_hdr.io_type));
12397 
12398 	task_desc = ctl_scsi_task_string(&io->taskio);
12399 	if (task_desc != NULL) {
12400 #ifdef NEEDTOPORT
12401 		csevent_log(CSC_CTL | CSC_SHELF_SW |
12402 			    CTL_TASK_REPORT,
12403 			    csevent_LogType_Trace,
12404 			    csevent_Severity_Information,
12405 			    csevent_AlertLevel_Green,
12406 			    csevent_FRU_Firmware,
12407 			    csevent_FRU_Unknown,
12408 			    "CTL: received task: %s",task_desc);
12409 #endif
12410 	} else {
12411 #ifdef NEEDTOPORT
12412 		csevent_log(CSC_CTL | CSC_SHELF_SW |
12413 			    CTL_TASK_REPORT,
12414 			    csevent_LogType_Trace,
12415 			    csevent_Severity_Information,
12416 			    csevent_AlertLevel_Green,
12417 			    csevent_FRU_Firmware,
12418 			    csevent_FRU_Unknown,
12419 			    "CTL: received unknown task "
12420 			    "type: %d (%#x)",
12421 			    io->taskio.task_action,
12422 			    io->taskio.task_action);
12423 #endif
12424 	}
12425 	switch (io->taskio.task_action) {
12426 	case CTL_TASK_ABORT_TASK:
12427 		retval = ctl_abort_task(io);
12428 		break;
12429 	case CTL_TASK_ABORT_TASK_SET:
12430 	case CTL_TASK_CLEAR_TASK_SET:
12431 		retval = ctl_abort_task_set(io);
12432 		break;
12433 	case CTL_TASK_CLEAR_ACA:
12434 		break;
12435 	case CTL_TASK_I_T_NEXUS_RESET:
12436 		retval = ctl_i_t_nexus_reset(io);
12437 		break;
12438 	case CTL_TASK_LUN_RESET: {
12439 		struct ctl_lun *lun;
12440 		uint32_t targ_lun;
12441 
12442 		targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12443 		mtx_lock(&ctl_softc->ctl_lock);
12444 		if ((targ_lun < CTL_MAX_LUNS)
12445 		 && (ctl_softc->ctl_luns[targ_lun] != NULL))
12446 			lun = ctl_softc->ctl_luns[targ_lun];
12447 		else {
12448 			mtx_unlock(&ctl_softc->ctl_lock);
12449 			retval = 1;
12450 			break;
12451 		}
12452 
12453 		if (!(io->io_hdr.flags &
12454 		    CTL_FLAG_FROM_OTHER_SC)) {
12455 			union ctl_ha_msg msg_info;
12456 
12457 			io->io_hdr.flags |=
12458 				CTL_FLAG_SENT_2OTHER_SC;
12459 			msg_info.hdr.msg_type =
12460 				CTL_MSG_MANAGE_TASKS;
12461 			msg_info.hdr.nexus = io->io_hdr.nexus;
12462 			msg_info.task.task_action =
12463 				CTL_TASK_LUN_RESET;
12464 			msg_info.hdr.original_sc = NULL;
12465 			msg_info.hdr.serializing_sc = NULL;
12466 			if (CTL_HA_STATUS_SUCCESS !=
12467 			    ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12468 			    (void *)&msg_info,
12469 			    sizeof(msg_info), 0)) {
12470 			}
12471 		}
12472 
12473 		retval = ctl_lun_reset(lun, io,
12474 				       CTL_UA_LUN_RESET);
12475 		mtx_unlock(&ctl_softc->ctl_lock);
12476 		break;
12477 	}
12478 	case CTL_TASK_TARGET_RESET:
12479 		retval = ctl_target_reset(ctl_softc, io, CTL_UA_TARG_RESET);
12480 		break;
12481 	case CTL_TASK_BUS_RESET:
12482 		retval = ctl_bus_reset(ctl_softc, io);
12483 		break;
12484 	case CTL_TASK_PORT_LOGIN:
12485 		break;
12486 	case CTL_TASK_PORT_LOGOUT:
12487 		break;
12488 	default:
12489 		printf("ctl_run_task: got unknown task management event %d\n",
12490 		       io->taskio.task_action);
12491 		break;
12492 	}
12493 	if (retval == 0)
12494 		io->io_hdr.status = CTL_SUCCESS;
12495 	else
12496 		io->io_hdr.status = CTL_ERROR;
12497 	ctl_done(io);
12498 }
12499 
12500 /*
12501  * For HA operation.  Handle commands that come in from the other
12502  * controller.
12503  */
12504 static void
12505 ctl_handle_isc(union ctl_io *io)
12506 {
12507 	int free_io;
12508 	struct ctl_lun *lun;
12509 	struct ctl_softc *ctl_softc;
12510 	uint32_t targ_lun;
12511 
12512 	ctl_softc = control_softc;
12513 
12514 	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12515 	lun = ctl_softc->ctl_luns[targ_lun];
12516 
12517 	switch (io->io_hdr.msg_type) {
12518 	case CTL_MSG_SERIALIZE:
12519 		free_io = ctl_serialize_other_sc_cmd(&io->scsiio);
12520 		break;
12521 	case CTL_MSG_R2R: {
12522 		const struct ctl_cmd_entry *entry;
12523 
12524 		/*
12525 		 * This is only used in SER_ONLY mode.
12526 		 */
12527 		free_io = 0;
12528 		entry = ctl_get_cmd_entry(&io->scsiio, NULL);
12529 		mtx_lock(&lun->lun_lock);
12530 		if (ctl_scsiio_lun_check(ctl_softc, lun,
12531 		    entry, (struct ctl_scsiio *)io) != 0) {
12532 			mtx_unlock(&lun->lun_lock);
12533 			ctl_done(io);
12534 			break;
12535 		}
12536 		io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
12537 		mtx_unlock(&lun->lun_lock);
12538 		ctl_enqueue_rtr(io);
12539 		break;
12540 	}
12541 	case CTL_MSG_FINISH_IO:
12542 		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
12543 			free_io = 0;
12544 			ctl_done(io);
12545 		} else {
12546 			free_io = 1;
12547 			mtx_lock(&lun->lun_lock);
12548 			TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
12549 				     ooa_links);
12550 			ctl_check_blocked(lun);
12551 			mtx_unlock(&lun->lun_lock);
12552 		}
12553 		break;
12554 	case CTL_MSG_PERS_ACTION:
12555 		ctl_hndl_per_res_out_on_other_sc(
12556 			(union ctl_ha_msg *)&io->presio.pr_msg);
12557 		free_io = 1;
12558 		break;
12559 	case CTL_MSG_BAD_JUJU:
12560 		free_io = 0;
12561 		ctl_done(io);
12562 		break;
12563 	case CTL_MSG_DATAMOVE:
12564 		/* Only used in XFER mode */
12565 		free_io = 0;
12566 		ctl_datamove_remote(io);
12567 		break;
12568 	case CTL_MSG_DATAMOVE_DONE:
12569 		/* Only used in XFER mode */
12570 		free_io = 0;
12571 		io->scsiio.be_move_done(io);
12572 		break;
12573 	default:
12574 		free_io = 1;
12575 		printf("%s: Invalid message type %d\n",
12576 		       __func__, io->io_hdr.msg_type);
12577 		break;
12578 	}
12579 	if (free_io)
12580 		ctl_free_io(io);
12581 
12582 }
12583 
12584 
12585 /*
12586  * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12587  * there is no match.
12588  */
12589 static ctl_lun_error_pattern
12590 ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12591 {
12592 	const struct ctl_cmd_entry *entry;
12593 	ctl_lun_error_pattern filtered_pattern, pattern;
12594 
12595 	pattern = desc->error_pattern;
12596 
12597 	/*
12598 	 * XXX KDM we need more data passed into this function to match a
12599 	 * custom pattern, and we actually need to implement custom pattern
12600 	 * matching.
12601 	 */
12602 	if (pattern & CTL_LUN_PAT_CMD)
12603 		return (CTL_LUN_PAT_CMD);
12604 
12605 	if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12606 		return (CTL_LUN_PAT_ANY);
12607 
12608 	entry = ctl_get_cmd_entry(ctsio, NULL);
12609 
12610 	filtered_pattern = entry->pattern & pattern;
12611 
12612 	/*
12613 	 * If the user requested specific flags in the pattern (e.g.
12614 	 * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12615 	 * flags.
12616 	 *
12617 	 * If the user did not specify any flags, it doesn't matter whether
12618 	 * or not the command supports the flags.
12619 	 */
12620 	if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12621 	     (pattern & ~CTL_LUN_PAT_MASK))
12622 		return (CTL_LUN_PAT_NONE);
12623 
12624 	/*
12625 	 * If the user asked for a range check, see if the requested LBA
12626 	 * range overlaps with this command's LBA range.
12627 	 */
12628 	if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12629 		uint64_t lba1;
12630 		uint64_t len1;
12631 		ctl_action action;
12632 		int retval;
12633 
12634 		retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12635 		if (retval != 0)
12636 			return (CTL_LUN_PAT_NONE);
12637 
12638 		action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12639 					      desc->lba_range.len);
12640 		/*
12641 		 * A "pass" means that the LBA ranges don't overlap, so
12642 		 * this doesn't match the user's range criteria.
12643 		 */
12644 		if (action == CTL_ACTION_PASS)
12645 			return (CTL_LUN_PAT_NONE);
12646 	}
12647 
12648 	return (filtered_pattern);
12649 }
12650 
12651 static void
12652 ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12653 {
12654 	struct ctl_error_desc *desc, *desc2;
12655 
12656 	mtx_assert(&lun->lun_lock, MA_OWNED);
12657 
12658 	STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12659 		ctl_lun_error_pattern pattern;
12660 		/*
12661 		 * Check to see whether this particular command matches
12662 		 * the pattern in the descriptor.
12663 		 */
12664 		pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12665 		if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12666 			continue;
12667 
12668 		switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12669 		case CTL_LUN_INJ_ABORTED:
12670 			ctl_set_aborted(&io->scsiio);
12671 			break;
12672 		case CTL_LUN_INJ_MEDIUM_ERR:
12673 			ctl_set_medium_error(&io->scsiio);
12674 			break;
12675 		case CTL_LUN_INJ_UA:
12676 			/* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12677 			 * OCCURRED */
12678 			ctl_set_ua(&io->scsiio, 0x29, 0x00);
12679 			break;
12680 		case CTL_LUN_INJ_CUSTOM:
12681 			/*
12682 			 * We're assuming the user knows what he is doing.
12683 			 * Just copy the sense information without doing
12684 			 * checks.
12685 			 */
12686 			bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12687 			      ctl_min(sizeof(desc->custom_sense),
12688 				      sizeof(io->scsiio.sense_data)));
12689 			io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12690 			io->scsiio.sense_len = SSD_FULL_SIZE;
12691 			io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12692 			break;
12693 		case CTL_LUN_INJ_NONE:
12694 		default:
12695 			/*
12696 			 * If this is an error injection type we don't know
12697 			 * about, clear the continuous flag (if it is set)
12698 			 * so it will get deleted below.
12699 			 */
12700 			desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12701 			break;
12702 		}
12703 		/*
12704 		 * By default, each error injection action is a one-shot
12705 		 */
12706 		if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12707 			continue;
12708 
12709 		STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12710 
12711 		free(desc, M_CTL);
12712 	}
12713 }
12714 
12715 #ifdef CTL_IO_DELAY
12716 static void
12717 ctl_datamove_timer_wakeup(void *arg)
12718 {
12719 	union ctl_io *io;
12720 
12721 	io = (union ctl_io *)arg;
12722 
12723 	ctl_datamove(io);
12724 }
12725 #endif /* CTL_IO_DELAY */
12726 
12727 void
12728 ctl_datamove(union ctl_io *io)
12729 {
12730 	void (*fe_datamove)(union ctl_io *io);
12731 
12732 	mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
12733 
12734 	CTL_DEBUG_PRINT(("ctl_datamove\n"));
12735 
12736 #ifdef CTL_TIME_IO
12737 	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12738 		char str[256];
12739 		char path_str[64];
12740 		struct sbuf sb;
12741 
12742 		ctl_scsi_path_string(io, path_str, sizeof(path_str));
12743 		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12744 
12745 		sbuf_cat(&sb, path_str);
12746 		switch (io->io_hdr.io_type) {
12747 		case CTL_IO_SCSI:
12748 			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12749 			sbuf_printf(&sb, "\n");
12750 			sbuf_cat(&sb, path_str);
12751 			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12752 				    io->scsiio.tag_num, io->scsiio.tag_type);
12753 			break;
12754 		case CTL_IO_TASK:
12755 			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12756 				    "Tag Type: %d\n", io->taskio.task_action,
12757 				    io->taskio.tag_num, io->taskio.tag_type);
12758 			break;
12759 		default:
12760 			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12761 			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12762 			break;
12763 		}
12764 		sbuf_cat(&sb, path_str);
12765 		sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12766 			    (intmax_t)time_uptime - io->io_hdr.start_time);
12767 		sbuf_finish(&sb);
12768 		printf("%s", sbuf_data(&sb));
12769 	}
12770 #endif /* CTL_TIME_IO */
12771 
12772 #ifdef CTL_IO_DELAY
12773 	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12774 		struct ctl_lun *lun;
12775 
12776 		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12777 
12778 		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12779 	} else {
12780 		struct ctl_lun *lun;
12781 
12782 		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12783 		if ((lun != NULL)
12784 		 && (lun->delay_info.datamove_delay > 0)) {
12785 			struct callout *callout;
12786 
12787 			callout = (struct callout *)&io->io_hdr.timer_bytes;
12788 			callout_init(callout, /*mpsafe*/ 1);
12789 			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12790 			callout_reset(callout,
12791 				      lun->delay_info.datamove_delay * hz,
12792 				      ctl_datamove_timer_wakeup, io);
12793 			if (lun->delay_info.datamove_type ==
12794 			    CTL_DELAY_TYPE_ONESHOT)
12795 				lun->delay_info.datamove_delay = 0;
12796 			return;
12797 		}
12798 	}
12799 #endif
12800 
12801 	/*
12802 	 * This command has been aborted.  Set the port status, so we fail
12803 	 * the data move.
12804 	 */
12805 	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12806 		printf("ctl_datamove: tag 0x%04x on (%ju:%d:%ju:%d) aborted\n",
12807 		       io->scsiio.tag_num,(uintmax_t)io->io_hdr.nexus.initid.id,
12808 		       io->io_hdr.nexus.targ_port,
12809 		       (uintmax_t)io->io_hdr.nexus.targ_target.id,
12810 		       io->io_hdr.nexus.targ_lun);
12811 		io->io_hdr.port_status = 31337;
12812 		/*
12813 		 * Note that the backend, in this case, will get the
12814 		 * callback in its context.  In other cases it may get
12815 		 * called in the frontend's interrupt thread context.
12816 		 */
12817 		io->scsiio.be_move_done(io);
12818 		return;
12819 	}
12820 
12821 	/*
12822 	 * If we're in XFER mode and this I/O is from the other shelf
12823 	 * controller, we need to send the DMA to the other side to
12824 	 * actually transfer the data to/from the host.  In serialize only
12825 	 * mode the transfer happens below CTL and ctl_datamove() is only
12826 	 * called on the machine that originally received the I/O.
12827 	 */
12828 	if ((control_softc->ha_mode == CTL_HA_MODE_XFER)
12829 	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12830 		union ctl_ha_msg msg;
12831 		uint32_t sg_entries_sent;
12832 		int do_sg_copy;
12833 		int i;
12834 
12835 		memset(&msg, 0, sizeof(msg));
12836 		msg.hdr.msg_type = CTL_MSG_DATAMOVE;
12837 		msg.hdr.original_sc = io->io_hdr.original_sc;
12838 		msg.hdr.serializing_sc = io;
12839 		msg.hdr.nexus = io->io_hdr.nexus;
12840 		msg.dt.flags = io->io_hdr.flags;
12841 		/*
12842 		 * We convert everything into a S/G list here.  We can't
12843 		 * pass by reference, only by value between controllers.
12844 		 * So we can't pass a pointer to the S/G list, only as many
12845 		 * S/G entries as we can fit in here.  If it's possible for
12846 		 * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
12847 		 * then we need to break this up into multiple transfers.
12848 		 */
12849 		if (io->scsiio.kern_sg_entries == 0) {
12850 			msg.dt.kern_sg_entries = 1;
12851 			/*
12852 			 * If this is in cached memory, flush the cache
12853 			 * before we send the DMA request to the other
12854 			 * controller.  We want to do this in either the
12855 			 * read or the write case.  The read case is
12856 			 * straightforward.  In the write case, we want to
12857 			 * make sure nothing is in the local cache that
12858 			 * could overwrite the DMAed data.
12859 			 */
12860 			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12861 				/*
12862 				 * XXX KDM use bus_dmamap_sync() here.
12863 				 */
12864 			}
12865 
12866 			/*
12867 			 * Convert to a physical address if this is a
12868 			 * virtual address.
12869 			 */
12870 			if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
12871 				msg.dt.sg_list[0].addr =
12872 					io->scsiio.kern_data_ptr;
12873 			} else {
12874 				/*
12875 				 * XXX KDM use busdma here!
12876 				 */
12877 #if 0
12878 				msg.dt.sg_list[0].addr = (void *)
12879 					vtophys(io->scsiio.kern_data_ptr);
12880 #endif
12881 			}
12882 
12883 			msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
12884 			do_sg_copy = 0;
12885 		} else {
12886 			struct ctl_sg_entry *sgl;
12887 
12888 			do_sg_copy = 1;
12889 			msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
12890 			sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
12891 			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12892 				/*
12893 				 * XXX KDM use bus_dmamap_sync() here.
12894 				 */
12895 			}
12896 		}
12897 
12898 		msg.dt.kern_data_len = io->scsiio.kern_data_len;
12899 		msg.dt.kern_total_len = io->scsiio.kern_total_len;
12900 		msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
12901 		msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
12902 		msg.dt.sg_sequence = 0;
12903 
12904 		/*
12905 		 * Loop until we've sent all of the S/G entries.  On the
12906 		 * other end, we'll recompose these S/G entries into one
12907 		 * contiguous list before passing it to the
12908 		 */
12909 		for (sg_entries_sent = 0; sg_entries_sent <
12910 		     msg.dt.kern_sg_entries; msg.dt.sg_sequence++) {
12911 			msg.dt.cur_sg_entries = ctl_min((sizeof(msg.dt.sg_list)/
12912 				sizeof(msg.dt.sg_list[0])),
12913 				msg.dt.kern_sg_entries - sg_entries_sent);
12914 
12915 			if (do_sg_copy != 0) {
12916 				struct ctl_sg_entry *sgl;
12917 				int j;
12918 
12919 				sgl = (struct ctl_sg_entry *)
12920 					io->scsiio.kern_data_ptr;
12921 				/*
12922 				 * If this is in cached memory, flush the cache
12923 				 * before we send the DMA request to the other
12924 				 * controller.  We want to do this in either
12925 				 * the * read or the write case.  The read
12926 				 * case is straightforward.  In the write
12927 				 * case, we want to make sure nothing is
12928 				 * in the local cache that could overwrite
12929 				 * the DMAed data.
12930 				 */
12931 
12932 				for (i = sg_entries_sent, j = 0;
12933 				     i < msg.dt.cur_sg_entries; i++, j++) {
12934 					if ((io->io_hdr.flags &
12935 					     CTL_FLAG_NO_DATASYNC) == 0) {
12936 						/*
12937 						 * XXX KDM use bus_dmamap_sync()
12938 						 */
12939 					}
12940 					if ((io->io_hdr.flags &
12941 					     CTL_FLAG_BUS_ADDR) == 0) {
12942 						/*
12943 						 * XXX KDM use busdma.
12944 						 */
12945 #if 0
12946 						msg.dt.sg_list[j].addr =(void *)
12947 						       vtophys(sgl[i].addr);
12948 #endif
12949 					} else {
12950 						msg.dt.sg_list[j].addr =
12951 							sgl[i].addr;
12952 					}
12953 					msg.dt.sg_list[j].len = sgl[i].len;
12954 				}
12955 			}
12956 
12957 			sg_entries_sent += msg.dt.cur_sg_entries;
12958 			if (sg_entries_sent >= msg.dt.kern_sg_entries)
12959 				msg.dt.sg_last = 1;
12960 			else
12961 				msg.dt.sg_last = 0;
12962 
12963 			/*
12964 			 * XXX KDM drop and reacquire the lock here?
12965 			 */
12966 			if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12967 			    sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
12968 				/*
12969 				 * XXX do something here.
12970 				 */
12971 			}
12972 
12973 			msg.dt.sent_sg_entries = sg_entries_sent;
12974 		}
12975 		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12976 		if (io->io_hdr.flags & CTL_FLAG_FAILOVER)
12977 			ctl_failover_io(io, /*have_lock*/ 0);
12978 
12979 	} else {
12980 
12981 		/*
12982 		 * Lookup the fe_datamove() function for this particular
12983 		 * front end.
12984 		 */
12985 		fe_datamove =
12986 		    control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12987 
12988 		fe_datamove(io);
12989 	}
12990 }
12991 
12992 static void
12993 ctl_send_datamove_done(union ctl_io *io, int have_lock)
12994 {
12995 	union ctl_ha_msg msg;
12996 	int isc_status;
12997 
12998 	memset(&msg, 0, sizeof(msg));
12999 
13000 	msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
13001 	msg.hdr.original_sc = io;
13002 	msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
13003 	msg.hdr.nexus = io->io_hdr.nexus;
13004 	msg.hdr.status = io->io_hdr.status;
13005 	msg.scsi.tag_num = io->scsiio.tag_num;
13006 	msg.scsi.tag_type = io->scsiio.tag_type;
13007 	msg.scsi.scsi_status = io->scsiio.scsi_status;
13008 	memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
13009 	       sizeof(io->scsiio.sense_data));
13010 	msg.scsi.sense_len = io->scsiio.sense_len;
13011 	msg.scsi.sense_residual = io->scsiio.sense_residual;
13012 	msg.scsi.fetd_status = io->io_hdr.port_status;
13013 	msg.scsi.residual = io->scsiio.residual;
13014 	io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
13015 
13016 	if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
13017 		ctl_failover_io(io, /*have_lock*/ have_lock);
13018 		return;
13019 	}
13020 
13021 	isc_status = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0);
13022 	if (isc_status > CTL_HA_STATUS_SUCCESS) {
13023 		/* XXX do something if this fails */
13024 	}
13025 
13026 }
13027 
13028 /*
13029  * The DMA to the remote side is done, now we need to tell the other side
13030  * we're done so it can continue with its data movement.
13031  */
13032 static void
13033 ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
13034 {
13035 	union ctl_io *io;
13036 
13037 	io = rq->context;
13038 
13039 	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
13040 		printf("%s: ISC DMA write failed with error %d", __func__,
13041 		       rq->ret);
13042 		ctl_set_internal_failure(&io->scsiio,
13043 					 /*sks_valid*/ 1,
13044 					 /*retry_count*/ rq->ret);
13045 	}
13046 
13047 	ctl_dt_req_free(rq);
13048 
13049 	/*
13050 	 * In this case, we had to malloc the memory locally.  Free it.
13051 	 */
13052 	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
13053 		int i;
13054 		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13055 			free(io->io_hdr.local_sglist[i].addr, M_CTL);
13056 	}
13057 	/*
13058 	 * The data is in local and remote memory, so now we need to send
13059 	 * status (good or back) back to the other side.
13060 	 */
13061 	ctl_send_datamove_done(io, /*have_lock*/ 0);
13062 }
13063 
13064 /*
13065  * We've moved the data from the host/controller into local memory.  Now we
13066  * need to push it over to the remote controller's memory.
13067  */
13068 static int
13069 ctl_datamove_remote_dm_write_cb(union ctl_io *io)
13070 {
13071 	int retval;
13072 
13073 	retval = 0;
13074 
13075 	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
13076 					  ctl_datamove_remote_write_cb);
13077 
13078 	return (retval);
13079 }
13080 
13081 static void
13082 ctl_datamove_remote_write(union ctl_io *io)
13083 {
13084 	int retval;
13085 	void (*fe_datamove)(union ctl_io *io);
13086 
13087 	/*
13088 	 * - Get the data from the host/HBA into local memory.
13089 	 * - DMA memory from the local controller to the remote controller.
13090 	 * - Send status back to the remote controller.
13091 	 */
13092 
13093 	retval = ctl_datamove_remote_sgl_setup(io);
13094 	if (retval != 0)
13095 		return;
13096 
13097 	/* Switch the pointer over so the FETD knows what to do */
13098 	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
13099 
13100 	/*
13101 	 * Use a custom move done callback, since we need to send completion
13102 	 * back to the other controller, not to the backend on this side.
13103 	 */
13104 	io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
13105 
13106 	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
13107 
13108 	fe_datamove(io);
13109 
13110 	return;
13111 
13112 }
13113 
13114 static int
13115 ctl_datamove_remote_dm_read_cb(union ctl_io *io)
13116 {
13117 #if 0
13118 	char str[256];
13119 	char path_str[64];
13120 	struct sbuf sb;
13121 #endif
13122 
13123 	/*
13124 	 * In this case, we had to malloc the memory locally.  Free it.
13125 	 */
13126 	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
13127 		int i;
13128 		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13129 			free(io->io_hdr.local_sglist[i].addr, M_CTL);
13130 	}
13131 
13132 #if 0
13133 	scsi_path_string(io, path_str, sizeof(path_str));
13134 	sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13135 	sbuf_cat(&sb, path_str);
13136 	scsi_command_string(&io->scsiio, NULL, &sb);
13137 	sbuf_printf(&sb, "\n");
13138 	sbuf_cat(&sb, path_str);
13139 	sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13140 		    io->scsiio.tag_num, io->scsiio.tag_type);
13141 	sbuf_cat(&sb, path_str);
13142 	sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
13143 		    io->io_hdr.flags, io->io_hdr.status);
13144 	sbuf_finish(&sb);
13145 	printk("%s", sbuf_data(&sb));
13146 #endif
13147 
13148 
13149 	/*
13150 	 * The read is done, now we need to send status (good or bad) back
13151 	 * to the other side.
13152 	 */
13153 	ctl_send_datamove_done(io, /*have_lock*/ 0);
13154 
13155 	return (0);
13156 }
13157 
13158 static void
13159 ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
13160 {
13161 	union ctl_io *io;
13162 	void (*fe_datamove)(union ctl_io *io);
13163 
13164 	io = rq->context;
13165 
13166 	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
13167 		printf("%s: ISC DMA read failed with error %d", __func__,
13168 		       rq->ret);
13169 		ctl_set_internal_failure(&io->scsiio,
13170 					 /*sks_valid*/ 1,
13171 					 /*retry_count*/ rq->ret);
13172 	}
13173 
13174 	ctl_dt_req_free(rq);
13175 
13176 	/* Switch the pointer over so the FETD knows what to do */
13177 	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
13178 
13179 	/*
13180 	 * Use a custom move done callback, since we need to send completion
13181 	 * back to the other controller, not to the backend on this side.
13182 	 */
13183 	io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
13184 
13185 	/* XXX KDM add checks like the ones in ctl_datamove? */
13186 
13187 	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
13188 
13189 	fe_datamove(io);
13190 }
13191 
13192 static int
13193 ctl_datamove_remote_sgl_setup(union ctl_io *io)
13194 {
13195 	struct ctl_sg_entry *local_sglist, *remote_sglist;
13196 	struct ctl_sg_entry *local_dma_sglist, *remote_dma_sglist;
13197 	struct ctl_softc *softc;
13198 	int retval;
13199 	int i;
13200 
13201 	retval = 0;
13202 	softc = control_softc;
13203 
13204 	local_sglist = io->io_hdr.local_sglist;
13205 	local_dma_sglist = io->io_hdr.local_dma_sglist;
13206 	remote_sglist = io->io_hdr.remote_sglist;
13207 	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
13208 
13209 	if (io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) {
13210 		for (i = 0; i < io->scsiio.kern_sg_entries; i++) {
13211 			local_sglist[i].len = remote_sglist[i].len;
13212 
13213 			/*
13214 			 * XXX Detect the situation where the RS-level I/O
13215 			 * redirector on the other side has already read the
13216 			 * data off of the AOR RS on this side, and
13217 			 * transferred it to remote (mirror) memory on the
13218 			 * other side.  Since we already have the data in
13219 			 * memory here, we just need to use it.
13220 			 *
13221 			 * XXX KDM this can probably be removed once we
13222 			 * get the cache device code in and take the
13223 			 * current AOR implementation out.
13224 			 */
13225 #ifdef NEEDTOPORT
13226 			if ((remote_sglist[i].addr >=
13227 			     (void *)vtophys(softc->mirr->addr))
13228 			 && (remote_sglist[i].addr <
13229 			     ((void *)vtophys(softc->mirr->addr) +
13230 			     CacheMirrorOffset))) {
13231 				local_sglist[i].addr = remote_sglist[i].addr -
13232 					CacheMirrorOffset;
13233 				if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13234 				     CTL_FLAG_DATA_IN)
13235 					io->io_hdr.flags |= CTL_FLAG_REDIR_DONE;
13236 			} else {
13237 				local_sglist[i].addr = remote_sglist[i].addr +
13238 					CacheMirrorOffset;
13239 			}
13240 #endif
13241 #if 0
13242 			printf("%s: local %p, remote %p, len %d\n",
13243 			       __func__, local_sglist[i].addr,
13244 			       remote_sglist[i].addr, local_sglist[i].len);
13245 #endif
13246 		}
13247 	} else {
13248 		uint32_t len_to_go;
13249 
13250 		/*
13251 		 * In this case, we don't have automatically allocated
13252 		 * memory for this I/O on this controller.  This typically
13253 		 * happens with internal CTL I/O -- e.g. inquiry, mode
13254 		 * sense, etc.  Anything coming from RAIDCore will have
13255 		 * a mirror area available.
13256 		 */
13257 		len_to_go = io->scsiio.kern_data_len;
13258 
13259 		/*
13260 		 * Clear the no datasync flag, we have to use malloced
13261 		 * buffers.
13262 		 */
13263 		io->io_hdr.flags &= ~CTL_FLAG_NO_DATASYNC;
13264 
13265 		/*
13266 		 * The difficult thing here is that the size of the various
13267 		 * S/G segments may be different than the size from the
13268 		 * remote controller.  That'll make it harder when DMAing
13269 		 * the data back to the other side.
13270 		 */
13271 		for (i = 0; (i < sizeof(io->io_hdr.remote_sglist) /
13272 		     sizeof(io->io_hdr.remote_sglist[0])) &&
13273 		     (len_to_go > 0); i++) {
13274 			local_sglist[i].len = ctl_min(len_to_go, 131072);
13275 			CTL_SIZE_8B(local_dma_sglist[i].len,
13276 				    local_sglist[i].len);
13277 			local_sglist[i].addr =
13278 				malloc(local_dma_sglist[i].len, M_CTL,M_WAITOK);
13279 
13280 			local_dma_sglist[i].addr = local_sglist[i].addr;
13281 
13282 			if (local_sglist[i].addr == NULL) {
13283 				int j;
13284 
13285 				printf("malloc failed for %zd bytes!",
13286 				       local_dma_sglist[i].len);
13287 				for (j = 0; j < i; j++) {
13288 					free(local_sglist[j].addr, M_CTL);
13289 				}
13290 				ctl_set_internal_failure(&io->scsiio,
13291 							 /*sks_valid*/ 1,
13292 							 /*retry_count*/ 4857);
13293 				retval = 1;
13294 				goto bailout_error;
13295 
13296 			}
13297 			/* XXX KDM do we need a sync here? */
13298 
13299 			len_to_go -= local_sglist[i].len;
13300 		}
13301 		/*
13302 		 * Reset the number of S/G entries accordingly.  The
13303 		 * original number of S/G entries is available in
13304 		 * rem_sg_entries.
13305 		 */
13306 		io->scsiio.kern_sg_entries = i;
13307 
13308 #if 0
13309 		printf("%s: kern_sg_entries = %d\n", __func__,
13310 		       io->scsiio.kern_sg_entries);
13311 		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13312 			printf("%s: sg[%d] = %p, %d (DMA: %d)\n", __func__, i,
13313 			       local_sglist[i].addr, local_sglist[i].len,
13314 			       local_dma_sglist[i].len);
13315 #endif
13316 	}
13317 
13318 
13319 	return (retval);
13320 
13321 bailout_error:
13322 
13323 	ctl_send_datamove_done(io, /*have_lock*/ 0);
13324 
13325 	return (retval);
13326 }
13327 
13328 static int
13329 ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
13330 			 ctl_ha_dt_cb callback)
13331 {
13332 	struct ctl_ha_dt_req *rq;
13333 	struct ctl_sg_entry *remote_sglist, *local_sglist;
13334 	struct ctl_sg_entry *remote_dma_sglist, *local_dma_sglist;
13335 	uint32_t local_used, remote_used, total_used;
13336 	int retval;
13337 	int i, j;
13338 
13339 	retval = 0;
13340 
13341 	rq = ctl_dt_req_alloc();
13342 
13343 	/*
13344 	 * If we failed to allocate the request, and if the DMA didn't fail
13345 	 * anyway, set busy status.  This is just a resource allocation
13346 	 * failure.
13347 	 */
13348 	if ((rq == NULL)
13349 	 && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE))
13350 		ctl_set_busy(&io->scsiio);
13351 
13352 	if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) {
13353 
13354 		if (rq != NULL)
13355 			ctl_dt_req_free(rq);
13356 
13357 		/*
13358 		 * The data move failed.  We need to return status back
13359 		 * to the other controller.  No point in trying to DMA
13360 		 * data to the remote controller.
13361 		 */
13362 
13363 		ctl_send_datamove_done(io, /*have_lock*/ 0);
13364 
13365 		retval = 1;
13366 
13367 		goto bailout;
13368 	}
13369 
13370 	local_sglist = io->io_hdr.local_sglist;
13371 	local_dma_sglist = io->io_hdr.local_dma_sglist;
13372 	remote_sglist = io->io_hdr.remote_sglist;
13373 	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
13374 	local_used = 0;
13375 	remote_used = 0;
13376 	total_used = 0;
13377 
13378 	if (io->io_hdr.flags & CTL_FLAG_REDIR_DONE) {
13379 		rq->ret = CTL_HA_STATUS_SUCCESS;
13380 		rq->context = io;
13381 		callback(rq);
13382 		goto bailout;
13383 	}
13384 
13385 	/*
13386 	 * Pull/push the data over the wire from/to the other controller.
13387 	 * This takes into account the possibility that the local and
13388 	 * remote sglists may not be identical in terms of the size of
13389 	 * the elements and the number of elements.
13390 	 *
13391 	 * One fundamental assumption here is that the length allocated for
13392 	 * both the local and remote sglists is identical.  Otherwise, we've
13393 	 * essentially got a coding error of some sort.
13394 	 */
13395 	for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
13396 		int isc_ret;
13397 		uint32_t cur_len, dma_length;
13398 		uint8_t *tmp_ptr;
13399 
13400 		rq->id = CTL_HA_DATA_CTL;
13401 		rq->command = command;
13402 		rq->context = io;
13403 
13404 		/*
13405 		 * Both pointers should be aligned.  But it is possible
13406 		 * that the allocation length is not.  They should both
13407 		 * also have enough slack left over at the end, though,
13408 		 * to round up to the next 8 byte boundary.
13409 		 */
13410 		cur_len = ctl_min(local_sglist[i].len - local_used,
13411 				  remote_sglist[j].len - remote_used);
13412 
13413 		/*
13414 		 * In this case, we have a size issue and need to decrease
13415 		 * the size, except in the case where we actually have less
13416 		 * than 8 bytes left.  In that case, we need to increase
13417 		 * the DMA length to get the last bit.
13418 		 */
13419 		if ((cur_len & 0x7) != 0) {
13420 			if (cur_len > 0x7) {
13421 				cur_len = cur_len - (cur_len & 0x7);
13422 				dma_length = cur_len;
13423 			} else {
13424 				CTL_SIZE_8B(dma_length, cur_len);
13425 			}
13426 
13427 		} else
13428 			dma_length = cur_len;
13429 
13430 		/*
13431 		 * If we had to allocate memory for this I/O, instead of using
13432 		 * the non-cached mirror memory, we'll need to flush the cache
13433 		 * before trying to DMA to the other controller.
13434 		 *
13435 		 * We could end up doing this multiple times for the same
13436 		 * segment if we have a larger local segment than remote
13437 		 * segment.  That shouldn't be an issue.
13438 		 */
13439 		if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
13440 			/*
13441 			 * XXX KDM use bus_dmamap_sync() here.
13442 			 */
13443 		}
13444 
13445 		rq->size = dma_length;
13446 
13447 		tmp_ptr = (uint8_t *)local_sglist[i].addr;
13448 		tmp_ptr += local_used;
13449 
13450 		/* Use physical addresses when talking to ISC hardware */
13451 		if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
13452 			/* XXX KDM use busdma */
13453 #if 0
13454 			rq->local = vtophys(tmp_ptr);
13455 #endif
13456 		} else
13457 			rq->local = tmp_ptr;
13458 
13459 		tmp_ptr = (uint8_t *)remote_sglist[j].addr;
13460 		tmp_ptr += remote_used;
13461 		rq->remote = tmp_ptr;
13462 
13463 		rq->callback = NULL;
13464 
13465 		local_used += cur_len;
13466 		if (local_used >= local_sglist[i].len) {
13467 			i++;
13468 			local_used = 0;
13469 		}
13470 
13471 		remote_used += cur_len;
13472 		if (remote_used >= remote_sglist[j].len) {
13473 			j++;
13474 			remote_used = 0;
13475 		}
13476 		total_used += cur_len;
13477 
13478 		if (total_used >= io->scsiio.kern_data_len)
13479 			rq->callback = callback;
13480 
13481 		if ((rq->size & 0x7) != 0) {
13482 			printf("%s: warning: size %d is not on 8b boundary\n",
13483 			       __func__, rq->size);
13484 		}
13485 		if (((uintptr_t)rq->local & 0x7) != 0) {
13486 			printf("%s: warning: local %p not on 8b boundary\n",
13487 			       __func__, rq->local);
13488 		}
13489 		if (((uintptr_t)rq->remote & 0x7) != 0) {
13490 			printf("%s: warning: remote %p not on 8b boundary\n",
13491 			       __func__, rq->local);
13492 		}
13493 #if 0
13494 		printf("%s: %s: local %#x remote %#x size %d\n", __func__,
13495 		       (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
13496 		       rq->local, rq->remote, rq->size);
13497 #endif
13498 
13499 		isc_ret = ctl_dt_single(rq);
13500 		if (isc_ret == CTL_HA_STATUS_WAIT)
13501 			continue;
13502 
13503 		if (isc_ret == CTL_HA_STATUS_DISCONNECT) {
13504 			rq->ret = CTL_HA_STATUS_SUCCESS;
13505 		} else {
13506 			rq->ret = isc_ret;
13507 		}
13508 		callback(rq);
13509 		goto bailout;
13510 	}
13511 
13512 bailout:
13513 	return (retval);
13514 
13515 }
13516 
13517 static void
13518 ctl_datamove_remote_read(union ctl_io *io)
13519 {
13520 	int retval;
13521 	int i;
13522 
13523 	/*
13524 	 * This will send an error to the other controller in the case of a
13525 	 * failure.
13526 	 */
13527 	retval = ctl_datamove_remote_sgl_setup(io);
13528 	if (retval != 0)
13529 		return;
13530 
13531 	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
13532 					  ctl_datamove_remote_read_cb);
13533 	if ((retval != 0)
13534 	 && ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0)) {
13535 		/*
13536 		 * Make sure we free memory if there was an error..  The
13537 		 * ctl_datamove_remote_xfer() function will send the
13538 		 * datamove done message, or call the callback with an
13539 		 * error if there is a problem.
13540 		 */
13541 		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13542 			free(io->io_hdr.local_sglist[i].addr, M_CTL);
13543 	}
13544 
13545 	return;
13546 }
13547 
13548 /*
13549  * Process a datamove request from the other controller.  This is used for
13550  * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
13551  * first.  Once that is complete, the data gets DMAed into the remote
13552  * controller's memory.  For reads, we DMA from the remote controller's
13553  * memory into our memory first, and then move it out to the FETD.
13554  */
13555 static void
13556 ctl_datamove_remote(union ctl_io *io)
13557 {
13558 	struct ctl_softc *softc;
13559 
13560 	softc = control_softc;
13561 
13562 	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
13563 
13564 	/*
13565 	 * Note that we look for an aborted I/O here, but don't do some of
13566 	 * the other checks that ctl_datamove() normally does.
13567 	 * We don't need to run the datamove delay code, since that should
13568 	 * have been done if need be on the other controller.
13569 	 */
13570 	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
13571 		printf("%s: tag 0x%04x on (%d:%d:%d:%d) aborted\n", __func__,
13572 		       io->scsiio.tag_num, io->io_hdr.nexus.initid.id,
13573 		       io->io_hdr.nexus.targ_port,
13574 		       io->io_hdr.nexus.targ_target.id,
13575 		       io->io_hdr.nexus.targ_lun);
13576 		io->io_hdr.port_status = 31338;
13577 		ctl_send_datamove_done(io, /*have_lock*/ 0);
13578 		return;
13579 	}
13580 
13581 	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT) {
13582 		ctl_datamove_remote_write(io);
13583 	} else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN){
13584 		ctl_datamove_remote_read(io);
13585 	} else {
13586 		union ctl_ha_msg msg;
13587 		struct scsi_sense_data *sense;
13588 		uint8_t sks[3];
13589 		int retry_count;
13590 
13591 		memset(&msg, 0, sizeof(msg));
13592 
13593 		msg.hdr.msg_type = CTL_MSG_BAD_JUJU;
13594 		msg.hdr.status = CTL_SCSI_ERROR;
13595 		msg.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
13596 
13597 		retry_count = 4243;
13598 
13599 		sense = &msg.scsi.sense_data;
13600 		sks[0] = SSD_SCS_VALID;
13601 		sks[1] = (retry_count >> 8) & 0xff;
13602 		sks[2] = retry_count & 0xff;
13603 
13604 		/* "Internal target failure" */
13605 		scsi_set_sense_data(sense,
13606 				    /*sense_format*/ SSD_TYPE_NONE,
13607 				    /*current_error*/ 1,
13608 				    /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
13609 				    /*asc*/ 0x44,
13610 				    /*ascq*/ 0x00,
13611 				    /*type*/ SSD_ELEM_SKS,
13612 				    /*size*/ sizeof(sks),
13613 				    /*data*/ sks,
13614 				    SSD_ELEM_NONE);
13615 
13616 		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
13617 		if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
13618 			ctl_failover_io(io, /*have_lock*/ 1);
13619 			return;
13620 		}
13621 
13622 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0) >
13623 		    CTL_HA_STATUS_SUCCESS) {
13624 			/* XXX KDM what to do if this fails? */
13625 		}
13626 		return;
13627 	}
13628 
13629 }
13630 
13631 static int
13632 ctl_process_done(union ctl_io *io)
13633 {
13634 	struct ctl_lun *lun;
13635 	struct ctl_softc *ctl_softc;
13636 	void (*fe_done)(union ctl_io *io);
13637 	uint32_t targ_port = ctl_port_idx(io->io_hdr.nexus.targ_port);
13638 
13639 	CTL_DEBUG_PRINT(("ctl_process_done\n"));
13640 
13641 	fe_done =
13642 	    control_softc->ctl_ports[targ_port]->fe_done;
13643 
13644 #ifdef CTL_TIME_IO
13645 	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
13646 		char str[256];
13647 		char path_str[64];
13648 		struct sbuf sb;
13649 
13650 		ctl_scsi_path_string(io, path_str, sizeof(path_str));
13651 		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13652 
13653 		sbuf_cat(&sb, path_str);
13654 		switch (io->io_hdr.io_type) {
13655 		case CTL_IO_SCSI:
13656 			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
13657 			sbuf_printf(&sb, "\n");
13658 			sbuf_cat(&sb, path_str);
13659 			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13660 				    io->scsiio.tag_num, io->scsiio.tag_type);
13661 			break;
13662 		case CTL_IO_TASK:
13663 			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
13664 				    "Tag Type: %d\n", io->taskio.task_action,
13665 				    io->taskio.tag_num, io->taskio.tag_type);
13666 			break;
13667 		default:
13668 			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13669 			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13670 			break;
13671 		}
13672 		sbuf_cat(&sb, path_str);
13673 		sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
13674 			    (intmax_t)time_uptime - io->io_hdr.start_time);
13675 		sbuf_finish(&sb);
13676 		printf("%s", sbuf_data(&sb));
13677 	}
13678 #endif /* CTL_TIME_IO */
13679 
13680 	switch (io->io_hdr.io_type) {
13681 	case CTL_IO_SCSI:
13682 		break;
13683 	case CTL_IO_TASK:
13684 		if (bootverbose || verbose > 0)
13685 			ctl_io_error_print(io, NULL);
13686 		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
13687 			ctl_free_io(io);
13688 		else
13689 			fe_done(io);
13690 		return (CTL_RETVAL_COMPLETE);
13691 		break;
13692 	default:
13693 		printf("ctl_process_done: invalid io type %d\n",
13694 		       io->io_hdr.io_type);
13695 		panic("ctl_process_done: invalid io type %d\n",
13696 		      io->io_hdr.io_type);
13697 		break; /* NOTREACHED */
13698 	}
13699 
13700 	lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13701 	if (lun == NULL) {
13702 		CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
13703 				 io->io_hdr.nexus.targ_mapped_lun));
13704 		fe_done(io);
13705 		goto bailout;
13706 	}
13707 	ctl_softc = lun->ctl_softc;
13708 
13709 	mtx_lock(&lun->lun_lock);
13710 
13711 	/*
13712 	 * Check to see if we have any errors to inject here.  We only
13713 	 * inject errors for commands that don't already have errors set.
13714 	 */
13715 	if ((STAILQ_FIRST(&lun->error_list) != NULL)
13716 	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))
13717 		ctl_inject_error(lun, io);
13718 
13719 	/*
13720 	 * XXX KDM how do we treat commands that aren't completed
13721 	 * successfully?
13722 	 *
13723 	 * XXX KDM should we also track I/O latency?
13724 	 */
13725 	if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
13726 	    io->io_hdr.io_type == CTL_IO_SCSI) {
13727 #ifdef CTL_TIME_IO
13728 		struct bintime cur_bt;
13729 #endif
13730 		int type;
13731 
13732 		if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13733 		    CTL_FLAG_DATA_IN)
13734 			type = CTL_STATS_READ;
13735 		else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13736 		    CTL_FLAG_DATA_OUT)
13737 			type = CTL_STATS_WRITE;
13738 		else
13739 			type = CTL_STATS_NO_IO;
13740 
13741 		lun->stats.ports[targ_port].bytes[type] +=
13742 		    io->scsiio.kern_total_len;
13743 		lun->stats.ports[targ_port].operations[type]++;
13744 #ifdef CTL_TIME_IO
13745 		bintime_add(&lun->stats.ports[targ_port].dma_time[type],
13746 		   &io->io_hdr.dma_bt);
13747 		lun->stats.ports[targ_port].num_dmas[type] +=
13748 		    io->io_hdr.num_dmas;
13749 		getbintime(&cur_bt);
13750 		bintime_sub(&cur_bt, &io->io_hdr.start_bt);
13751 		bintime_add(&lun->stats.ports[targ_port].time[type], &cur_bt);
13752 #endif
13753 	}
13754 
13755 	/*
13756 	 * Remove this from the OOA queue.
13757 	 */
13758 	TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13759 
13760 	/*
13761 	 * Run through the blocked queue on this LUN and see if anything
13762 	 * has become unblocked, now that this transaction is done.
13763 	 */
13764 	ctl_check_blocked(lun);
13765 
13766 	/*
13767 	 * If the LUN has been invalidated, free it if there is nothing
13768 	 * left on its OOA queue.
13769 	 */
13770 	if ((lun->flags & CTL_LUN_INVALID)
13771 	 && TAILQ_EMPTY(&lun->ooa_queue)) {
13772 		mtx_unlock(&lun->lun_lock);
13773 		mtx_lock(&ctl_softc->ctl_lock);
13774 		ctl_free_lun(lun);
13775 		mtx_unlock(&ctl_softc->ctl_lock);
13776 	} else
13777 		mtx_unlock(&lun->lun_lock);
13778 
13779 	/*
13780 	 * If this command has been aborted, make sure we set the status
13781 	 * properly.  The FETD is responsible for freeing the I/O and doing
13782 	 * whatever it needs to do to clean up its state.
13783 	 */
13784 	if (io->io_hdr.flags & CTL_FLAG_ABORT)
13785 		ctl_set_task_aborted(&io->scsiio);
13786 
13787 	/*
13788 	 * We print out status for every task management command.  For SCSI
13789 	 * commands, we filter out any unit attention errors; they happen
13790 	 * on every boot, and would clutter up the log.  Note:  task
13791 	 * management commands aren't printed here, they are printed above,
13792 	 * since they should never even make it down here.
13793 	 */
13794 	switch (io->io_hdr.io_type) {
13795 	case CTL_IO_SCSI: {
13796 		int error_code, sense_key, asc, ascq;
13797 
13798 		sense_key = 0;
13799 
13800 		if (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SCSI_ERROR)
13801 		 && (io->scsiio.scsi_status == SCSI_STATUS_CHECK_COND)) {
13802 			/*
13803 			 * Since this is just for printing, no need to
13804 			 * show errors here.
13805 			 */
13806 			scsi_extract_sense_len(&io->scsiio.sense_data,
13807 					       io->scsiio.sense_len,
13808 					       &error_code,
13809 					       &sense_key,
13810 					       &asc,
13811 					       &ascq,
13812 					       /*show_errors*/ 0);
13813 		}
13814 
13815 		if (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
13816 		 && (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SCSI_ERROR)
13817 		  || (io->scsiio.scsi_status != SCSI_STATUS_CHECK_COND)
13818 		  || (sense_key != SSD_KEY_UNIT_ATTENTION))) {
13819 
13820 			if ((time_uptime - ctl_softc->last_print_jiffies) <= 0){
13821 				ctl_softc->skipped_prints++;
13822 			} else {
13823 				uint32_t skipped_prints;
13824 
13825 				skipped_prints = ctl_softc->skipped_prints;
13826 
13827 				ctl_softc->skipped_prints = 0;
13828 				ctl_softc->last_print_jiffies = time_uptime;
13829 
13830 				if (skipped_prints > 0) {
13831 #ifdef NEEDTOPORT
13832 					csevent_log(CSC_CTL | CSC_SHELF_SW |
13833 					    CTL_ERROR_REPORT,
13834 					    csevent_LogType_Trace,
13835 					    csevent_Severity_Information,
13836 					    csevent_AlertLevel_Green,
13837 					    csevent_FRU_Firmware,
13838 					    csevent_FRU_Unknown,
13839 					    "High CTL error volume, %d prints "
13840 					    "skipped", skipped_prints);
13841 #endif
13842 				}
13843 				if (bootverbose || verbose > 0)
13844 					ctl_io_error_print(io, NULL);
13845 			}
13846 		}
13847 		break;
13848 	}
13849 	case CTL_IO_TASK:
13850 		if (bootverbose || verbose > 0)
13851 			ctl_io_error_print(io, NULL);
13852 		break;
13853 	default:
13854 		break;
13855 	}
13856 
13857 	/*
13858 	 * Tell the FETD or the other shelf controller we're done with this
13859 	 * command.  Note that only SCSI commands get to this point.  Task
13860 	 * management commands are completed above.
13861 	 *
13862 	 * We only send status to the other controller if we're in XFER
13863 	 * mode.  In SER_ONLY mode, the I/O is done on the controller that
13864 	 * received the I/O (from CTL's perspective), and so the status is
13865 	 * generated there.
13866 	 *
13867 	 * XXX KDM if we hold the lock here, we could cause a deadlock
13868 	 * if the frontend comes back in in this context to queue
13869 	 * something.
13870 	 */
13871 	if ((ctl_softc->ha_mode == CTL_HA_MODE_XFER)
13872 	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
13873 		union ctl_ha_msg msg;
13874 
13875 		memset(&msg, 0, sizeof(msg));
13876 		msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13877 		msg.hdr.original_sc = io->io_hdr.original_sc;
13878 		msg.hdr.nexus = io->io_hdr.nexus;
13879 		msg.hdr.status = io->io_hdr.status;
13880 		msg.scsi.scsi_status = io->scsiio.scsi_status;
13881 		msg.scsi.tag_num = io->scsiio.tag_num;
13882 		msg.scsi.tag_type = io->scsiio.tag_type;
13883 		msg.scsi.sense_len = io->scsiio.sense_len;
13884 		msg.scsi.sense_residual = io->scsiio.sense_residual;
13885 		msg.scsi.residual = io->scsiio.residual;
13886 		memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
13887 		       sizeof(io->scsiio.sense_data));
13888 		/*
13889 		 * We copy this whether or not this is an I/O-related
13890 		 * command.  Otherwise, we'd have to go and check to see
13891 		 * whether it's a read/write command, and it really isn't
13892 		 * worth it.
13893 		 */
13894 		memcpy(&msg.scsi.lbalen,
13895 		       &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
13896 		       sizeof(msg.scsi.lbalen));
13897 
13898 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13899 				sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
13900 			/* XXX do something here */
13901 		}
13902 
13903 		ctl_free_io(io);
13904 	} else
13905 		fe_done(io);
13906 
13907 bailout:
13908 
13909 	return (CTL_RETVAL_COMPLETE);
13910 }
13911 
13912 #ifdef CTL_WITH_CA
13913 /*
13914  * Front end should call this if it doesn't do autosense.  When the request
13915  * sense comes back in from the initiator, we'll dequeue this and send it.
13916  */
13917 int
13918 ctl_queue_sense(union ctl_io *io)
13919 {
13920 	struct ctl_lun *lun;
13921 	struct ctl_softc *ctl_softc;
13922 	uint32_t initidx, targ_lun;
13923 
13924 	ctl_softc = control_softc;
13925 
13926 	CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13927 
13928 	/*
13929 	 * LUN lookup will likely move to the ctl_work_thread() once we
13930 	 * have our new queueing infrastructure (that doesn't put things on
13931 	 * a per-LUN queue initially).  That is so that we can handle
13932 	 * things like an INQUIRY to a LUN that we don't have enabled.  We
13933 	 * can't deal with that right now.
13934 	 */
13935 	mtx_lock(&ctl_softc->ctl_lock);
13936 
13937 	/*
13938 	 * If we don't have a LUN for this, just toss the sense
13939 	 * information.
13940 	 */
13941 	targ_lun = io->io_hdr.nexus.targ_lun;
13942 	targ_lun = ctl_map_lun(io->io_hdr.nexus.targ_port, targ_lun);
13943 	if ((targ_lun < CTL_MAX_LUNS)
13944 	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
13945 		lun = ctl_softc->ctl_luns[targ_lun];
13946 	else
13947 		goto bailout;
13948 
13949 	initidx = ctl_get_initindex(&io->io_hdr.nexus);
13950 
13951 	mtx_lock(&lun->lun_lock);
13952 	/*
13953 	 * Already have CA set for this LUN...toss the sense information.
13954 	 */
13955 	if (ctl_is_set(lun->have_ca, initidx)) {
13956 		mtx_unlock(&lun->lun_lock);
13957 		goto bailout;
13958 	}
13959 
13960 	memcpy(&lun->pending_sense[initidx], &io->scsiio.sense_data,
13961 	       ctl_min(sizeof(lun->pending_sense[initidx]),
13962 	       sizeof(io->scsiio.sense_data)));
13963 	ctl_set_mask(lun->have_ca, initidx);
13964 	mtx_unlock(&lun->lun_lock);
13965 
13966 bailout:
13967 	mtx_unlock(&ctl_softc->ctl_lock);
13968 
13969 	ctl_free_io(io);
13970 
13971 	return (CTL_RETVAL_COMPLETE);
13972 }
13973 #endif
13974 
13975 /*
13976  * Primary command inlet from frontend ports.  All SCSI and task I/O
13977  * requests must go through this function.
13978  */
13979 int
13980 ctl_queue(union ctl_io *io)
13981 {
13982 	struct ctl_softc *ctl_softc;
13983 
13984 	CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13985 
13986 	ctl_softc = control_softc;
13987 
13988 #ifdef CTL_TIME_IO
13989 	io->io_hdr.start_time = time_uptime;
13990 	getbintime(&io->io_hdr.start_bt);
13991 #endif /* CTL_TIME_IO */
13992 
13993 	/* Map FE-specific LUN ID into global one. */
13994 	io->io_hdr.nexus.targ_mapped_lun =
13995 	    ctl_map_lun(io->io_hdr.nexus.targ_port, io->io_hdr.nexus.targ_lun);
13996 
13997 	switch (io->io_hdr.io_type) {
13998 	case CTL_IO_SCSI:
13999 	case CTL_IO_TASK:
14000 		ctl_enqueue_incoming(io);
14001 		break;
14002 	default:
14003 		printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
14004 		return (EINVAL);
14005 	}
14006 
14007 	return (CTL_RETVAL_COMPLETE);
14008 }
14009 
14010 #ifdef CTL_IO_DELAY
14011 static void
14012 ctl_done_timer_wakeup(void *arg)
14013 {
14014 	union ctl_io *io;
14015 
14016 	io = (union ctl_io *)arg;
14017 	ctl_done(io);
14018 }
14019 #endif /* CTL_IO_DELAY */
14020 
14021 void
14022 ctl_done(union ctl_io *io)
14023 {
14024 	struct ctl_softc *ctl_softc;
14025 
14026 	ctl_softc = control_softc;
14027 
14028 	/*
14029 	 * Enable this to catch duplicate completion issues.
14030 	 */
14031 #if 0
14032 	if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
14033 		printf("%s: type %d msg %d cdb %x iptl: "
14034 		       "%d:%d:%d:%d tag 0x%04x "
14035 		       "flag %#x status %x\n",
14036 			__func__,
14037 			io->io_hdr.io_type,
14038 			io->io_hdr.msg_type,
14039 			io->scsiio.cdb[0],
14040 			io->io_hdr.nexus.initid.id,
14041 			io->io_hdr.nexus.targ_port,
14042 			io->io_hdr.nexus.targ_target.id,
14043 			io->io_hdr.nexus.targ_lun,
14044 			(io->io_hdr.io_type ==
14045 			CTL_IO_TASK) ?
14046 			io->taskio.tag_num :
14047 			io->scsiio.tag_num,
14048 		        io->io_hdr.flags,
14049 			io->io_hdr.status);
14050 	} else
14051 		io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
14052 #endif
14053 
14054 	/*
14055 	 * This is an internal copy of an I/O, and should not go through
14056 	 * the normal done processing logic.
14057 	 */
14058 	if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
14059 		return;
14060 
14061 	/*
14062 	 * We need to send a msg to the serializing shelf to finish the IO
14063 	 * as well.  We don't send a finish message to the other shelf if
14064 	 * this is a task management command.  Task management commands
14065 	 * aren't serialized in the OOA queue, but rather just executed on
14066 	 * both shelf controllers for commands that originated on that
14067 	 * controller.
14068 	 */
14069 	if ((io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)
14070 	 && (io->io_hdr.io_type != CTL_IO_TASK)) {
14071 		union ctl_ha_msg msg_io;
14072 
14073 		msg_io.hdr.msg_type = CTL_MSG_FINISH_IO;
14074 		msg_io.hdr.serializing_sc = io->io_hdr.serializing_sc;
14075 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_io,
14076 		    sizeof(msg_io), 0 ) != CTL_HA_STATUS_SUCCESS) {
14077 		}
14078 		/* continue on to finish IO */
14079 	}
14080 #ifdef CTL_IO_DELAY
14081 	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
14082 		struct ctl_lun *lun;
14083 
14084 		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
14085 
14086 		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
14087 	} else {
14088 		struct ctl_lun *lun;
14089 
14090 		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
14091 
14092 		if ((lun != NULL)
14093 		 && (lun->delay_info.done_delay > 0)) {
14094 			struct callout *callout;
14095 
14096 			callout = (struct callout *)&io->io_hdr.timer_bytes;
14097 			callout_init(callout, /*mpsafe*/ 1);
14098 			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
14099 			callout_reset(callout,
14100 				      lun->delay_info.done_delay * hz,
14101 				      ctl_done_timer_wakeup, io);
14102 			if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
14103 				lun->delay_info.done_delay = 0;
14104 			return;
14105 		}
14106 	}
14107 #endif /* CTL_IO_DELAY */
14108 
14109 	ctl_enqueue_done(io);
14110 }
14111 
14112 int
14113 ctl_isc(struct ctl_scsiio *ctsio)
14114 {
14115 	struct ctl_lun *lun;
14116 	int retval;
14117 
14118 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
14119 
14120 	CTL_DEBUG_PRINT(("ctl_isc: command: %02x\n", ctsio->cdb[0]));
14121 
14122 	CTL_DEBUG_PRINT(("ctl_isc: calling data_submit()\n"));
14123 
14124 	retval = lun->backend->data_submit((union ctl_io *)ctsio);
14125 
14126 	return (retval);
14127 }
14128 
14129 
14130 static void
14131 ctl_work_thread(void *arg)
14132 {
14133 	struct ctl_thread *thr = (struct ctl_thread *)arg;
14134 	struct ctl_softc *softc = thr->ctl_softc;
14135 	union ctl_io *io;
14136 	int retval;
14137 
14138 	CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
14139 
14140 	for (;;) {
14141 		retval = 0;
14142 
14143 		/*
14144 		 * We handle the queues in this order:
14145 		 * - ISC
14146 		 * - done queue (to free up resources, unblock other commands)
14147 		 * - RtR queue
14148 		 * - incoming queue
14149 		 *
14150 		 * If those queues are empty, we break out of the loop and
14151 		 * go to sleep.
14152 		 */
14153 		mtx_lock(&thr->queue_lock);
14154 		io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
14155 		if (io != NULL) {
14156 			STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
14157 			mtx_unlock(&thr->queue_lock);
14158 			ctl_handle_isc(io);
14159 			continue;
14160 		}
14161 		io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
14162 		if (io != NULL) {
14163 			STAILQ_REMOVE_HEAD(&thr->done_queue, links);
14164 			/* clear any blocked commands, call fe_done */
14165 			mtx_unlock(&thr->queue_lock);
14166 			retval = ctl_process_done(io);
14167 			continue;
14168 		}
14169 		io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
14170 		if (io != NULL) {
14171 			STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
14172 			mtx_unlock(&thr->queue_lock);
14173 			if (io->io_hdr.io_type == CTL_IO_TASK)
14174 				ctl_run_task(io);
14175 			else
14176 				ctl_scsiio_precheck(softc, &io->scsiio);
14177 			continue;
14178 		}
14179 		if (!ctl_pause_rtr) {
14180 			io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
14181 			if (io != NULL) {
14182 				STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
14183 				mtx_unlock(&thr->queue_lock);
14184 				retval = ctl_scsiio(&io->scsiio);
14185 				if (retval != CTL_RETVAL_COMPLETE)
14186 					CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
14187 				continue;
14188 			}
14189 		}
14190 
14191 		/* Sleep until we have something to do. */
14192 		mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
14193 	}
14194 }
14195 
14196 static void
14197 ctl_lun_thread(void *arg)
14198 {
14199 	struct ctl_softc *softc = (struct ctl_softc *)arg;
14200 	struct ctl_be_lun *be_lun;
14201 	int retval;
14202 
14203 	CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
14204 
14205 	for (;;) {
14206 		retval = 0;
14207 		mtx_lock(&softc->ctl_lock);
14208 		be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
14209 		if (be_lun != NULL) {
14210 			STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
14211 			mtx_unlock(&softc->ctl_lock);
14212 			ctl_create_lun(be_lun);
14213 			continue;
14214 		}
14215 
14216 		/* Sleep until we have something to do. */
14217 		mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
14218 		    PDROP | PRIBIO, "-", 0);
14219 	}
14220 }
14221 
14222 static void
14223 ctl_enqueue_incoming(union ctl_io *io)
14224 {
14225 	struct ctl_softc *softc = control_softc;
14226 	struct ctl_thread *thr;
14227 	u_int idx;
14228 
14229 	idx = (io->io_hdr.nexus.targ_port * 127 +
14230 	       io->io_hdr.nexus.initid.id) % worker_threads;
14231 	thr = &softc->threads[idx];
14232 	mtx_lock(&thr->queue_lock);
14233 	STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
14234 	mtx_unlock(&thr->queue_lock);
14235 	wakeup(thr);
14236 }
14237 
14238 static void
14239 ctl_enqueue_rtr(union ctl_io *io)
14240 {
14241 	struct ctl_softc *softc = control_softc;
14242 	struct ctl_thread *thr;
14243 
14244 	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14245 	mtx_lock(&thr->queue_lock);
14246 	STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
14247 	mtx_unlock(&thr->queue_lock);
14248 	wakeup(thr);
14249 }
14250 
14251 static void
14252 ctl_enqueue_done(union ctl_io *io)
14253 {
14254 	struct ctl_softc *softc = control_softc;
14255 	struct ctl_thread *thr;
14256 
14257 	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14258 	mtx_lock(&thr->queue_lock);
14259 	STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
14260 	mtx_unlock(&thr->queue_lock);
14261 	wakeup(thr);
14262 }
14263 
14264 static void
14265 ctl_enqueue_isc(union ctl_io *io)
14266 {
14267 	struct ctl_softc *softc = control_softc;
14268 	struct ctl_thread *thr;
14269 
14270 	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14271 	mtx_lock(&thr->queue_lock);
14272 	STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
14273 	mtx_unlock(&thr->queue_lock);
14274 	wakeup(thr);
14275 }
14276 
14277 /* Initialization and failover */
14278 
14279 void
14280 ctl_init_isc_msg(void)
14281 {
14282 	printf("CTL: Still calling this thing\n");
14283 }
14284 
14285 /*
14286  * Init component
14287  * 	Initializes component into configuration defined by bootMode
14288  *	(see hasc-sv.c)
14289  *  	returns hasc_Status:
14290  * 		OK
14291  *		ERROR - fatal error
14292  */
14293 static ctl_ha_comp_status
14294 ctl_isc_init(struct ctl_ha_component *c)
14295 {
14296 	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
14297 
14298 	c->status = ret;
14299 	return ret;
14300 }
14301 
14302 /* Start component
14303  * 	Starts component in state requested. If component starts successfully,
14304  *	it must set its own state to the requestrd state
14305  *	When requested state is HASC_STATE_HA, the component may refine it
14306  * 	by adding _SLAVE or _MASTER flags.
14307  *	Currently allowed state transitions are:
14308  *	UNKNOWN->HA		- initial startup
14309  *	UNKNOWN->SINGLE - initial startup when no parter detected
14310  *	HA->SINGLE		- failover
14311  * returns ctl_ha_comp_status:
14312  * 		OK	- component successfully started in requested state
14313  *		FAILED  - could not start the requested state, failover may
14314  * 			  be possible
14315  *		ERROR	- fatal error detected, no future startup possible
14316  */
14317 static ctl_ha_comp_status
14318 ctl_isc_start(struct ctl_ha_component *c, ctl_ha_state state)
14319 {
14320 	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
14321 
14322 	printf("%s: go\n", __func__);
14323 
14324 	// UNKNOWN->HA or UNKNOWN->SINGLE (bootstrap)
14325 	if (c->state == CTL_HA_STATE_UNKNOWN ) {
14326 		ctl_is_single = 0;
14327 		if (ctl_ha_msg_create(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
14328 		    != CTL_HA_STATUS_SUCCESS) {
14329 			printf("ctl_isc_start: ctl_ha_msg_create failed.\n");
14330 			ret = CTL_HA_COMP_STATUS_ERROR;
14331 		}
14332 	} else if (CTL_HA_STATE_IS_HA(c->state)
14333 		&& CTL_HA_STATE_IS_SINGLE(state)){
14334 		// HA->SINGLE transition
14335 	        ctl_failover();
14336 		ctl_is_single = 1;
14337 	} else {
14338 		printf("ctl_isc_start:Invalid state transition %X->%X\n",
14339 		       c->state, state);
14340 		ret = CTL_HA_COMP_STATUS_ERROR;
14341 	}
14342 	if (CTL_HA_STATE_IS_SINGLE(state))
14343 		ctl_is_single = 1;
14344 
14345 	c->state = state;
14346 	c->status = ret;
14347 	return ret;
14348 }
14349 
14350 /*
14351  * Quiesce component
14352  * The component must clear any error conditions (set status to OK) and
14353  * prepare itself to another Start call
14354  * returns ctl_ha_comp_status:
14355  * 	OK
14356  *	ERROR
14357  */
14358 static ctl_ha_comp_status
14359 ctl_isc_quiesce(struct ctl_ha_component *c)
14360 {
14361 	int ret = CTL_HA_COMP_STATUS_OK;
14362 
14363 	ctl_pause_rtr = 1;
14364 	c->status = ret;
14365 	return ret;
14366 }
14367 
14368 struct ctl_ha_component ctl_ha_component_ctlisc =
14369 {
14370 	.name = "CTL ISC",
14371 	.state = CTL_HA_STATE_UNKNOWN,
14372 	.init = ctl_isc_init,
14373 	.start = ctl_isc_start,
14374 	.quiesce = ctl_isc_quiesce
14375 };
14376 
14377 /*
14378  *  vim: ts=8
14379  */
14380