xref: /freebsd/sys/cam/ctl/ctl.c (revision 8ec81e392654c03ebba7792e47bf3744d623cf54)
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/mutex.h>
56 #include <sys/condvar.h>
57 #include <sys/malloc.h>
58 #include <sys/conf.h>
59 #include <sys/ioccom.h>
60 #include <sys/queue.h>
61 #include <sys/sbuf.h>
62 #include <sys/endian.h>
63 #include <sys/sysctl.h>
64 
65 #include <cam/cam.h>
66 #include <cam/scsi/scsi_all.h>
67 #include <cam/scsi/scsi_da.h>
68 #include <cam/ctl/ctl_io.h>
69 #include <cam/ctl/ctl.h>
70 #include <cam/ctl/ctl_frontend.h>
71 #include <cam/ctl/ctl_frontend_internal.h>
72 #include <cam/ctl/ctl_util.h>
73 #include <cam/ctl/ctl_backend.h>
74 #include <cam/ctl/ctl_ioctl.h>
75 #include <cam/ctl/ctl_ha.h>
76 #include <cam/ctl/ctl_private.h>
77 #include <cam/ctl/ctl_debug.h>
78 #include <cam/ctl/ctl_scsi_all.h>
79 #include <cam/ctl/ctl_error.h>
80 
81 struct ctl_softc *control_softc = NULL;
82 
83 /*
84  * The default is to run with CTL_DONE_THREAD turned on.  Completed
85  * transactions are queued for processing by the CTL work thread.  When
86  * CTL_DONE_THREAD is not defined, completed transactions are processed in
87  * the caller's context.
88  */
89 #define CTL_DONE_THREAD
90 
91 /*
92  *  * Use the serial number and device ID provided by the backend, rather than
93  *   * making up our own.
94  *    */
95 #define CTL_USE_BACKEND_SN
96 
97 /*
98  * Size and alignment macros needed for Copan-specific HA hardware.  These
99  * can go away when the HA code is re-written, and uses busdma for any
100  * hardware.
101  */
102 #define	CTL_ALIGN_8B(target, source, type)				\
103 	if (((uint32_t)source & 0x7) != 0)				\
104 		target = (type)(source + (0x8 - ((uint32_t)source & 0x7)));\
105 	else								\
106 		target = (type)source;
107 
108 #define	CTL_SIZE_8B(target, size)					\
109 	if ((size & 0x7) != 0)						\
110 		target = size + (0x8 - (size & 0x7));			\
111 	else								\
112 		target = size;
113 
114 #define CTL_ALIGN_8B_MARGIN	16
115 
116 /*
117  * Template mode pages.
118  */
119 
120 /*
121  * Note that these are default values only.  The actual values will be
122  * filled in when the user does a mode sense.
123  */
124 static struct copan_power_subpage power_page_default = {
125 	/*page_code*/ PWR_PAGE_CODE | SMPH_SPF,
126 	/*subpage*/ PWR_SUBPAGE_CODE,
127 	/*page_length*/ {(sizeof(struct copan_power_subpage) - 4) & 0xff00,
128 			 (sizeof(struct copan_power_subpage) - 4) & 0x00ff},
129 	/*page_version*/ PWR_VERSION,
130 	/* total_luns */ 26,
131 	/* max_active_luns*/ PWR_DFLT_MAX_LUNS,
132 	/*reserved*/ {0, 0, 0, 0, 0, 0, 0, 0, 0,
133 		      0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
134 		      0, 0, 0, 0, 0, 0}
135 };
136 
137 static struct copan_power_subpage power_page_changeable = {
138 	/*page_code*/ PWR_PAGE_CODE | SMPH_SPF,
139 	/*subpage*/ PWR_SUBPAGE_CODE,
140 	/*page_length*/ {(sizeof(struct copan_power_subpage) - 4) & 0xff00,
141 			 (sizeof(struct copan_power_subpage) - 4) & 0x00ff},
142 	/*page_version*/ 0,
143 	/* total_luns */ 0,
144 	/* max_active_luns*/ 0,
145 	/*reserved*/ {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, 0}
148 };
149 
150 static struct copan_aps_subpage aps_page_default = {
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 	APS_VERSION, //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_aps_subpage aps_page_changeable = {
163 	APS_PAGE_CODE | SMPH_SPF, //page_code
164 	APS_SUBPAGE_CODE, //subpage
165 	{(sizeof(struct copan_aps_subpage) - 4) & 0xff00,
166 	 (sizeof(struct copan_aps_subpage) - 4) & 0x00ff}, //page_length
167 	0, //page_version
168 	0, //lock_active
169 	{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
170 	0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
171 	0, 0, 0, 0, 0} //reserved
172 };
173 
174 static struct copan_debugconf_subpage debugconf_page_default = {
175 	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
176 	DBGCNF_SUBPAGE_CODE,		/* subpage */
177 	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
178 	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
179 	DBGCNF_VERSION,			/* page_version */
180 	{CTL_TIME_IO_DEFAULT_SECS>>8,
181 	 CTL_TIME_IO_DEFAULT_SECS>>0},	/* ctl_time_io_secs */
182 };
183 
184 static struct copan_debugconf_subpage debugconf_page_changeable = {
185 	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
186 	DBGCNF_SUBPAGE_CODE,		/* subpage */
187 	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
188 	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
189 	0,				/* page_version */
190 	{0xff,0xff},			/* ctl_time_io_secs */
191 };
192 
193 static struct scsi_format_page format_page_default = {
194 	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
195 	/*page_length*/sizeof(struct scsi_format_page) - 2,
196 	/*tracks_per_zone*/ {0, 0},
197 	/*alt_sectors_per_zone*/ {0, 0},
198 	/*alt_tracks_per_zone*/ {0, 0},
199 	/*alt_tracks_per_lun*/ {0, 0},
200 	/*sectors_per_track*/ {(CTL_DEFAULT_SECTORS_PER_TRACK >> 8) & 0xff,
201 			        CTL_DEFAULT_SECTORS_PER_TRACK & 0xff},
202 	/*bytes_per_sector*/ {0, 0},
203 	/*interleave*/ {0, 0},
204 	/*track_skew*/ {0, 0},
205 	/*cylinder_skew*/ {0, 0},
206 	/*flags*/ SFP_HSEC,
207 	/*reserved*/ {0, 0, 0}
208 };
209 
210 static struct scsi_format_page format_page_changeable = {
211 	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
212 	/*page_length*/sizeof(struct scsi_format_page) - 2,
213 	/*tracks_per_zone*/ {0, 0},
214 	/*alt_sectors_per_zone*/ {0, 0},
215 	/*alt_tracks_per_zone*/ {0, 0},
216 	/*alt_tracks_per_lun*/ {0, 0},
217 	/*sectors_per_track*/ {0, 0},
218 	/*bytes_per_sector*/ {0, 0},
219 	/*interleave*/ {0, 0},
220 	/*track_skew*/ {0, 0},
221 	/*cylinder_skew*/ {0, 0},
222 	/*flags*/ 0,
223 	/*reserved*/ {0, 0, 0}
224 };
225 
226 static struct scsi_rigid_disk_page rigid_disk_page_default = {
227 	/*page_code*/SMS_RIGID_DISK_PAGE,
228 	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
229 	/*cylinders*/ {0, 0, 0},
230 	/*heads*/ CTL_DEFAULT_HEADS,
231 	/*start_write_precomp*/ {0, 0, 0},
232 	/*start_reduced_current*/ {0, 0, 0},
233 	/*step_rate*/ {0, 0},
234 	/*landing_zone_cylinder*/ {0, 0, 0},
235 	/*rpl*/ SRDP_RPL_DISABLED,
236 	/*rotational_offset*/ 0,
237 	/*reserved1*/ 0,
238 	/*rotation_rate*/ {(CTL_DEFAULT_ROTATION_RATE >> 8) & 0xff,
239 			   CTL_DEFAULT_ROTATION_RATE & 0xff},
240 	/*reserved2*/ {0, 0}
241 };
242 
243 static struct scsi_rigid_disk_page rigid_disk_page_changeable = {
244 	/*page_code*/SMS_RIGID_DISK_PAGE,
245 	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
246 	/*cylinders*/ {0, 0, 0},
247 	/*heads*/ 0,
248 	/*start_write_precomp*/ {0, 0, 0},
249 	/*start_reduced_current*/ {0, 0, 0},
250 	/*step_rate*/ {0, 0},
251 	/*landing_zone_cylinder*/ {0, 0, 0},
252 	/*rpl*/ 0,
253 	/*rotational_offset*/ 0,
254 	/*reserved1*/ 0,
255 	/*rotation_rate*/ {0, 0},
256 	/*reserved2*/ {0, 0}
257 };
258 
259 static struct scsi_caching_page caching_page_default = {
260 	/*page_code*/SMS_CACHING_PAGE,
261 	/*page_length*/sizeof(struct scsi_caching_page) - 2,
262 	/*flags1*/ SCP_DISC | SCP_WCE,
263 	/*ret_priority*/ 0,
264 	/*disable_pf_transfer_len*/ {0xff, 0xff},
265 	/*min_prefetch*/ {0, 0},
266 	/*max_prefetch*/ {0xff, 0xff},
267 	/*max_pf_ceiling*/ {0xff, 0xff},
268 	/*flags2*/ 0,
269 	/*cache_segments*/ 0,
270 	/*cache_seg_size*/ {0, 0},
271 	/*reserved*/ 0,
272 	/*non_cache_seg_size*/ {0, 0, 0}
273 };
274 
275 static struct scsi_caching_page caching_page_changeable = {
276 	/*page_code*/SMS_CACHING_PAGE,
277 	/*page_length*/sizeof(struct scsi_caching_page) - 2,
278 	/*flags1*/ 0,
279 	/*ret_priority*/ 0,
280 	/*disable_pf_transfer_len*/ {0, 0},
281 	/*min_prefetch*/ {0, 0},
282 	/*max_prefetch*/ {0, 0},
283 	/*max_pf_ceiling*/ {0, 0},
284 	/*flags2*/ 0,
285 	/*cache_segments*/ 0,
286 	/*cache_seg_size*/ {0, 0},
287 	/*reserved*/ 0,
288 	/*non_cache_seg_size*/ {0, 0, 0}
289 };
290 
291 static struct scsi_control_page control_page_default = {
292 	/*page_code*/SMS_CONTROL_MODE_PAGE,
293 	/*page_length*/sizeof(struct scsi_control_page) - 2,
294 	/*rlec*/0,
295 	/*queue_flags*/0,
296 	/*eca_and_aen*/0,
297 	/*reserved*/0,
298 	/*aen_holdoff_period*/{0, 0}
299 };
300 
301 static struct scsi_control_page control_page_changeable = {
302 	/*page_code*/SMS_CONTROL_MODE_PAGE,
303 	/*page_length*/sizeof(struct scsi_control_page) - 2,
304 	/*rlec*/SCP_DSENSE,
305 	/*queue_flags*/0,
306 	/*eca_and_aen*/0,
307 	/*reserved*/0,
308 	/*aen_holdoff_period*/{0, 0}
309 };
310 
311 
312 /*
313  * XXX KDM move these into the softc.
314  */
315 static int rcv_sync_msg;
316 static int persis_offset;
317 static uint8_t ctl_pause_rtr;
318 static int     ctl_is_single;
319 static int     index_to_aps_page;
320 int	   ctl_disable = 0;
321 
322 SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD, 0, "CAM Target Layer");
323 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, disable, CTLFLAG_RDTUN, &ctl_disable, 0,
324 	   "Disable CTL");
325 TUNABLE_INT("kern.cam.ctl.disable", &ctl_disable);
326 
327 /*
328  * Serial number (0x80), device id (0x83), and supported pages (0x00)
329  */
330 #define SCSI_EVPD_NUM_SUPPORTED_PAGES	3
331 
332 static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
333 				  int param);
334 static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
335 static void ctl_init(void);
336 void ctl_shutdown(void);
337 static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
338 static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
339 static void ctl_ioctl_online(void *arg);
340 static void ctl_ioctl_offline(void *arg);
341 static int ctl_ioctl_targ_enable(void *arg, struct ctl_id targ_id);
342 static int ctl_ioctl_targ_disable(void *arg, struct ctl_id targ_id);
343 static int ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id);
344 static int ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id);
345 static int ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio);
346 static int ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio, int have_lock);
347 static int ctl_ioctl_submit_wait(union ctl_io *io);
348 static void ctl_ioctl_datamove(union ctl_io *io);
349 static void ctl_ioctl_done(union ctl_io *io);
350 static void ctl_ioctl_hard_startstop_callback(void *arg,
351 					      struct cfi_metatask *metatask);
352 static void ctl_ioctl_bbrread_callback(void *arg,struct cfi_metatask *metatask);
353 static int ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
354 			      struct ctl_ooa *ooa_hdr,
355 			      struct ctl_ooa_entry *kern_entries);
356 static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
357 		     struct thread *td);
358 uint32_t ctl_get_resindex(struct ctl_nexus *nexus);
359 uint32_t ctl_port_idx(int port_num);
360 #ifdef unused
361 static union ctl_io *ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port,
362 				   uint32_t targ_target, uint32_t targ_lun,
363 				   int can_wait);
364 static void ctl_kfree_io(union ctl_io *io);
365 #endif /* unused */
366 static void ctl_free_io_internal(union ctl_io *io, int have_lock);
367 static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
368 			 struct ctl_be_lun *be_lun, struct ctl_id target_id);
369 static int ctl_free_lun(struct ctl_lun *lun);
370 static void ctl_create_lun(struct ctl_be_lun *be_lun);
371 /**
372 static void ctl_failover_change_pages(struct ctl_softc *softc,
373 				      struct ctl_scsiio *ctsio, int master);
374 **/
375 
376 static int ctl_do_mode_select(union ctl_io *io);
377 static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
378 			   uint64_t res_key, uint64_t sa_res_key,
379 			   uint8_t type, uint32_t residx,
380 			   struct ctl_scsiio *ctsio,
381 			   struct scsi_per_res_out *cdb,
382 			   struct scsi_per_res_out_parms* param);
383 static void ctl_pro_preempt_other(struct ctl_lun *lun,
384 				  union ctl_ha_msg *msg);
385 static void ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg);
386 static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
387 static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
388 static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
389 static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
390 static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
391 static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint32_t *len);
392 static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2);
393 static ctl_action ctl_check_for_blockage(union ctl_io *pending_io,
394 					 union ctl_io *ooa_io);
395 static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
396 				union ctl_io *starting_io);
397 static int ctl_check_blocked(struct ctl_lun *lun);
398 static int ctl_scsiio_lun_check(struct ctl_softc *ctl_softc,
399 				struct ctl_lun *lun,
400 				struct ctl_cmd_entry *entry,
401 				struct ctl_scsiio *ctsio);
402 //static int ctl_check_rtr(union ctl_io *pending_io, struct ctl_softc *softc);
403 static void ctl_failover(void);
404 static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
405 			       struct ctl_scsiio *ctsio);
406 static int ctl_scsiio(struct ctl_scsiio *ctsio);
407 
408 static int ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
409 static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
410 			    ctl_ua_type ua_type);
411 static int ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io,
412 			 ctl_ua_type ua_type);
413 static int ctl_abort_task(union ctl_io *io);
414 static void ctl_run_task_queue(struct ctl_softc *ctl_softc);
415 #ifdef CTL_IO_DELAY
416 static void ctl_datamove_timer_wakeup(void *arg);
417 static void ctl_done_timer_wakeup(void *arg);
418 #endif /* CTL_IO_DELAY */
419 
420 static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
421 static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
422 static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
423 static void ctl_datamove_remote_write(union ctl_io *io);
424 static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
425 static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
426 static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
427 static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
428 				    ctl_ha_dt_cb callback);
429 static void ctl_datamove_remote_read(union ctl_io *io);
430 static void ctl_datamove_remote(union ctl_io *io);
431 static int ctl_process_done(union ctl_io *io, int have_lock);
432 static void ctl_work_thread(void *arg);
433 
434 /*
435  * Load the serialization table.  This isn't very pretty, but is probably
436  * the easiest way to do it.
437  */
438 #include "ctl_ser_table.c"
439 
440 /*
441  * We only need to define open, close and ioctl routines for this driver.
442  */
443 static struct cdevsw ctl_cdevsw = {
444 	.d_version =	D_VERSION,
445 	.d_flags =	0,
446 	.d_open =	ctl_open,
447 	.d_close =	ctl_close,
448 	.d_ioctl =	ctl_ioctl,
449 	.d_name =	"ctl",
450 };
451 
452 
453 MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
454 
455 /*
456  * If we have the CAM SIM, we may or may not have another SIM that will
457  * cause CTL to get initialized.  If not, we need to initialize it.
458  */
459 SYSINIT(ctl_init, SI_SUB_CONFIGURE, SI_ORDER_THIRD, ctl_init, NULL);
460 
461 static void
462 ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
463 			    union ctl_ha_msg *msg_info)
464 {
465 	struct ctl_scsiio *ctsio;
466 
467 	if (msg_info->hdr.original_sc == NULL) {
468 		printf("%s: original_sc == NULL!\n", __func__);
469 		/* XXX KDM now what? */
470 		return;
471 	}
472 
473 	ctsio = &msg_info->hdr.original_sc->scsiio;
474 	ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
475 	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
476 	ctsio->io_hdr.status = msg_info->hdr.status;
477 	ctsio->scsi_status = msg_info->scsi.scsi_status;
478 	ctsio->sense_len = msg_info->scsi.sense_len;
479 	ctsio->sense_residual = msg_info->scsi.sense_residual;
480 	ctsio->residual = msg_info->scsi.residual;
481 	memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
482 	       sizeof(ctsio->sense_data));
483 	memcpy(&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
484 	       &msg_info->scsi.lbalen, sizeof(msg_info->scsi.lbalen));
485 	STAILQ_INSERT_TAIL(&ctl_softc->isc_queue, &ctsio->io_hdr, links);
486 	ctl_wakeup_thread();
487 }
488 
489 static void
490 ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
491 				union ctl_ha_msg *msg_info)
492 {
493 	struct ctl_scsiio *ctsio;
494 
495 	if (msg_info->hdr.serializing_sc == NULL) {
496 		printf("%s: serializing_sc == NULL!\n", __func__);
497 		/* XXX KDM now what? */
498 		return;
499 	}
500 
501 	ctsio = &msg_info->hdr.serializing_sc->scsiio;
502 #if 0
503 	/*
504 	 * Attempt to catch the situation where an I/O has
505 	 * been freed, and we're using it again.
506 	 */
507 	if (ctsio->io_hdr.io_type == 0xff) {
508 		union ctl_io *tmp_io;
509 		tmp_io = (union ctl_io *)ctsio;
510 		printf("%s: %p use after free!\n", __func__,
511 		       ctsio);
512 		printf("%s: type %d msg %d cdb %x iptl: "
513 		       "%d:%d:%d:%d tag 0x%04x "
514 		       "flag %#x status %x\n",
515 			__func__,
516 			tmp_io->io_hdr.io_type,
517 			tmp_io->io_hdr.msg_type,
518 			tmp_io->scsiio.cdb[0],
519 			tmp_io->io_hdr.nexus.initid.id,
520 			tmp_io->io_hdr.nexus.targ_port,
521 			tmp_io->io_hdr.nexus.targ_target.id,
522 			tmp_io->io_hdr.nexus.targ_lun,
523 			(tmp_io->io_hdr.io_type ==
524 			CTL_IO_TASK) ?
525 			tmp_io->taskio.tag_num :
526 			tmp_io->scsiio.tag_num,
527 		        tmp_io->io_hdr.flags,
528 			tmp_io->io_hdr.status);
529 	}
530 #endif
531 	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
532 	STAILQ_INSERT_TAIL(&ctl_softc->isc_queue, &ctsio->io_hdr, links);
533 	ctl_wakeup_thread();
534 }
535 
536 /*
537  * ISC (Inter Shelf Communication) event handler.  Events from the HA
538  * subsystem come in here.
539  */
540 static void
541 ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
542 {
543 	struct ctl_softc *ctl_softc;
544 	union ctl_io *io;
545 	struct ctl_prio *presio;
546 	ctl_ha_status isc_status;
547 
548 	ctl_softc = control_softc;
549 	io = NULL;
550 
551 
552 #if 0
553 	printf("CTL: Isc Msg event %d\n", event);
554 #endif
555 	if (event == CTL_HA_EVT_MSG_RECV) {
556 		union ctl_ha_msg msg_info;
557 
558 		isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
559 					     sizeof(msg_info), /*wait*/ 0);
560 #if 0
561 		printf("CTL: msg_type %d\n", msg_info.msg_type);
562 #endif
563 		if (isc_status != 0) {
564 			printf("Error receiving message, status = %d\n",
565 			       isc_status);
566 			return;
567 		}
568 		mtx_lock(&ctl_softc->ctl_lock);
569 
570 		switch (msg_info.hdr.msg_type) {
571 		case CTL_MSG_SERIALIZE:
572 #if 0
573 			printf("Serialize\n");
574 #endif
575 			io = ctl_alloc_io((void *)ctl_softc->othersc_pool);
576 			if (io == NULL) {
577 				printf("ctl_isc_event_handler: can't allocate "
578 				       "ctl_io!\n");
579 				/* Bad Juju */
580 				/* Need to set busy and send msg back */
581 				mtx_unlock(&ctl_softc->ctl_lock);
582 				msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
583 				msg_info.hdr.status = CTL_SCSI_ERROR;
584 				msg_info.scsi.scsi_status = SCSI_STATUS_BUSY;
585 				msg_info.scsi.sense_len = 0;
586 			        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
587 				    sizeof(msg_info), 0) > CTL_HA_STATUS_SUCCESS){
588 				}
589 				goto bailout;
590 			}
591 			ctl_zero_io(io);
592 			// populate ctsio from msg_info
593 			io->io_hdr.io_type = CTL_IO_SCSI;
594 			io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
595 			io->io_hdr.original_sc = msg_info.hdr.original_sc;
596 #if 0
597 			printf("pOrig %x\n", (int)msg_info.original_sc);
598 #endif
599 			io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
600 					    CTL_FLAG_IO_ACTIVE;
601 			/*
602 			 * If we're in serialization-only mode, we don't
603 			 * want to go through full done processing.  Thus
604 			 * the COPY flag.
605 			 *
606 			 * XXX KDM add another flag that is more specific.
607 			 */
608 			if (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)
609 				io->io_hdr.flags |= CTL_FLAG_INT_COPY;
610 			io->io_hdr.nexus = msg_info.hdr.nexus;
611 #if 0
612 			printf("targ %d, port %d, iid %d, lun %d\n",
613 			       io->io_hdr.nexus.targ_target.id,
614 			       io->io_hdr.nexus.targ_port,
615 			       io->io_hdr.nexus.initid.id,
616 			       io->io_hdr.nexus.targ_lun);
617 #endif
618 			io->scsiio.tag_num = msg_info.scsi.tag_num;
619 			io->scsiio.tag_type = msg_info.scsi.tag_type;
620 			memcpy(io->scsiio.cdb, msg_info.scsi.cdb,
621 			       CTL_MAX_CDBLEN);
622 			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
623 				struct ctl_cmd_entry *entry;
624 				uint8_t opcode;
625 
626 				opcode = io->scsiio.cdb[0];
627 				entry = &ctl_cmd_table[opcode];
628 				io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
629 				io->io_hdr.flags |=
630 					entry->flags & CTL_FLAG_DATA_MASK;
631 			}
632 			STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
633 					   &io->io_hdr, links);
634 			ctl_wakeup_thread();
635 			break;
636 
637 		/* Performed on the Originating SC, XFER mode only */
638 		case CTL_MSG_DATAMOVE: {
639 			struct ctl_sg_entry *sgl;
640 			int i, j;
641 
642 			io = msg_info.hdr.original_sc;
643 			if (io == NULL) {
644 				printf("%s: original_sc == NULL!\n", __func__);
645 				/* XXX KDM do something here */
646 				break;
647 			}
648 			io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
649 			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
650 			/*
651 			 * Keep track of this, we need to send it back over
652 			 * when the datamove is complete.
653 			 */
654 			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
655 
656 			if (msg_info.dt.sg_sequence == 0) {
657 				/*
658 				 * XXX KDM we use the preallocated S/G list
659 				 * here, but we'll need to change this to
660 				 * dynamic allocation if we need larger S/G
661 				 * lists.
662 				 */
663 				if (msg_info.dt.kern_sg_entries >
664 				    sizeof(io->io_hdr.remote_sglist) /
665 				    sizeof(io->io_hdr.remote_sglist[0])) {
666 					printf("%s: number of S/G entries "
667 					    "needed %u > allocated num %zd\n",
668 					    __func__,
669 					    msg_info.dt.kern_sg_entries,
670 					    sizeof(io->io_hdr.remote_sglist)/
671 					    sizeof(io->io_hdr.remote_sglist[0]));
672 
673 					/*
674 					 * XXX KDM send a message back to
675 					 * the other side to shut down the
676 					 * DMA.  The error will come back
677 					 * through via the normal channel.
678 					 */
679 					break;
680 				}
681 				sgl = io->io_hdr.remote_sglist;
682 				memset(sgl, 0,
683 				       sizeof(io->io_hdr.remote_sglist));
684 
685 				io->scsiio.kern_data_ptr = (uint8_t *)sgl;
686 
687 				io->scsiio.kern_sg_entries =
688 					msg_info.dt.kern_sg_entries;
689 				io->scsiio.rem_sg_entries =
690 					msg_info.dt.kern_sg_entries;
691 				io->scsiio.kern_data_len =
692 					msg_info.dt.kern_data_len;
693 				io->scsiio.kern_total_len =
694 					msg_info.dt.kern_total_len;
695 				io->scsiio.kern_data_resid =
696 					msg_info.dt.kern_data_resid;
697 				io->scsiio.kern_rel_offset =
698 					msg_info.dt.kern_rel_offset;
699 				/*
700 				 * Clear out per-DMA flags.
701 				 */
702 				io->io_hdr.flags &= ~CTL_FLAG_RDMA_MASK;
703 				/*
704 				 * Add per-DMA flags that are set for this
705 				 * particular DMA request.
706 				 */
707 				io->io_hdr.flags |= msg_info.dt.flags &
708 						    CTL_FLAG_RDMA_MASK;
709 			} else
710 				sgl = (struct ctl_sg_entry *)
711 					io->scsiio.kern_data_ptr;
712 
713 			for (i = msg_info.dt.sent_sg_entries, j = 0;
714 			     i < (msg_info.dt.sent_sg_entries +
715 			     msg_info.dt.cur_sg_entries); i++, j++) {
716 				sgl[i].addr = msg_info.dt.sg_list[j].addr;
717 				sgl[i].len = msg_info.dt.sg_list[j].len;
718 
719 #if 0
720 				printf("%s: L: %p,%d -> %p,%d j=%d, i=%d\n",
721 				       __func__,
722 				       msg_info.dt.sg_list[j].addr,
723 				       msg_info.dt.sg_list[j].len,
724 				       sgl[i].addr, sgl[i].len, j, i);
725 #endif
726 			}
727 #if 0
728 			memcpy(&sgl[msg_info.dt.sent_sg_entries],
729 			       msg_info.dt.sg_list,
730 			       sizeof(*sgl) * msg_info.dt.cur_sg_entries);
731 #endif
732 
733 			/*
734 			 * If this is the last piece of the I/O, we've got
735 			 * the full S/G list.  Queue processing in the thread.
736 			 * Otherwise wait for the next piece.
737 			 */
738 			if (msg_info.dt.sg_last != 0) {
739 				STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
740 						   &io->io_hdr, links);
741 				ctl_wakeup_thread();
742 			}
743 			break;
744 		}
745 		/* Performed on the Serializing (primary) SC, XFER mode only */
746 		case CTL_MSG_DATAMOVE_DONE: {
747 			if (msg_info.hdr.serializing_sc == NULL) {
748 				printf("%s: serializing_sc == NULL!\n",
749 				       __func__);
750 				/* XXX KDM now what? */
751 				break;
752 			}
753 			/*
754 			 * We grab the sense information here in case
755 			 * there was a failure, so we can return status
756 			 * back to the initiator.
757 			 */
758 			io = msg_info.hdr.serializing_sc;
759 			io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
760 			io->io_hdr.status = msg_info.hdr.status;
761 			io->scsiio.scsi_status = msg_info.scsi.scsi_status;
762 			io->scsiio.sense_len = msg_info.scsi.sense_len;
763 			io->scsiio.sense_residual =msg_info.scsi.sense_residual;
764 			io->io_hdr.port_status = msg_info.scsi.fetd_status;
765 			io->scsiio.residual = msg_info.scsi.residual;
766 			memcpy(&io->scsiio.sense_data,&msg_info.scsi.sense_data,
767 			       sizeof(io->scsiio.sense_data));
768 
769 			STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
770 					   &io->io_hdr, links);
771 			ctl_wakeup_thread();
772 			break;
773 		}
774 
775 		/* Preformed on Originating SC, SER_ONLY mode */
776 		case CTL_MSG_R2R:
777 			io = msg_info.hdr.original_sc;
778 			if (io == NULL) {
779 				printf("%s: Major Bummer\n", __func__);
780 				mtx_unlock(&ctl_softc->ctl_lock);
781 				return;
782 			} else {
783 #if 0
784 				printf("pOrig %x\n",(int) ctsio);
785 #endif
786 			}
787 			io->io_hdr.msg_type = CTL_MSG_R2R;
788 			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
789 			STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
790 					   &io->io_hdr, links);
791 			ctl_wakeup_thread();
792 			break;
793 
794 		/*
795 		 * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
796 		 * mode.
797 		 * Performed on the Originating (i.e. secondary) SC in XFER
798 		 * mode
799 		 */
800 		case CTL_MSG_FINISH_IO:
801 			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER)
802 				ctl_isc_handler_finish_xfer(ctl_softc,
803 							    &msg_info);
804 			else
805 				ctl_isc_handler_finish_ser_only(ctl_softc,
806 								&msg_info);
807 			break;
808 
809 		/* Preformed on Originating SC */
810 		case CTL_MSG_BAD_JUJU:
811 			io = msg_info.hdr.original_sc;
812 			if (io == NULL) {
813 				printf("%s: Bad JUJU!, original_sc is NULL!\n",
814 				       __func__);
815 				break;
816 			}
817 			ctl_copy_sense_data(&msg_info, io);
818 			/*
819 			 * IO should have already been cleaned up on other
820 			 * SC so clear this flag so we won't send a message
821 			 * back to finish the IO there.
822 			 */
823 			io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
824 			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
825 
826 			/* io = msg_info.hdr.serializing_sc; */
827 			io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
828 		        STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
829 					   &io->io_hdr, links);
830 			ctl_wakeup_thread();
831 			break;
832 
833 		/* Handle resets sent from the other side */
834 		case CTL_MSG_MANAGE_TASKS: {
835 			struct ctl_taskio *taskio;
836 			taskio = (struct ctl_taskio *)ctl_alloc_io(
837 				(void *)ctl_softc->othersc_pool);
838 			if (taskio == NULL) {
839 				printf("ctl_isc_event_handler: can't allocate "
840 				       "ctl_io!\n");
841 				/* Bad Juju */
842 				/* should I just call the proper reset func
843 				   here??? */
844 				mtx_unlock(&ctl_softc->ctl_lock);
845 				goto bailout;
846 			}
847 			ctl_zero_io((union ctl_io *)taskio);
848 			taskio->io_hdr.io_type = CTL_IO_TASK;
849 			taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
850 			taskio->io_hdr.nexus = msg_info.hdr.nexus;
851 			taskio->task_action = msg_info.task.task_action;
852 			taskio->tag_num = msg_info.task.tag_num;
853 			taskio->tag_type = msg_info.task.tag_type;
854 #ifdef CTL_TIME_IO
855 			taskio->io_hdr.start_time = time_uptime;
856 			getbintime(&taskio->io_hdr.start_bt);
857 #if 0
858 			cs_prof_gettime(&taskio->io_hdr.start_ticks);
859 #endif
860 #endif /* CTL_TIME_IO */
861 		        STAILQ_INSERT_TAIL(&ctl_softc->task_queue,
862 					   &taskio->io_hdr, links);
863 			ctl_softc->flags |= CTL_FLAG_TASK_PENDING;
864 			ctl_wakeup_thread();
865 			break;
866 		}
867 		/* Persistent Reserve action which needs attention */
868 		case CTL_MSG_PERS_ACTION:
869 			presio = (struct ctl_prio *)ctl_alloc_io(
870 				(void *)ctl_softc->othersc_pool);
871 			if (presio == NULL) {
872 				printf("ctl_isc_event_handler: can't allocate "
873 				       "ctl_io!\n");
874 				/* Bad Juju */
875 				/* Need to set busy and send msg back */
876 				mtx_unlock(&ctl_softc->ctl_lock);
877 				goto bailout;
878 			}
879 			ctl_zero_io((union ctl_io *)presio);
880 			presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
881 			presio->pr_msg = msg_info.pr;
882 		        STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
883 					   &presio->io_hdr, links);
884 			ctl_wakeup_thread();
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 
896 			lun = ctl_softc->ctl_luns[msg_info.hdr.nexus.targ_lun];
897 			page_index = &lun->mode_pages.index[index_to_aps_page];
898 			current_sp = (struct copan_aps_subpage *)
899 				     (page_index->page_data +
900 				     (page_index->page_len * CTL_PAGE_CURRENT));
901 
902 			current_sp->lock_active = msg_info.aps.lock_flag;
903 		        break;
904 		}
905 		default:
906 		        printf("How did I get here?\n");
907 		}
908 		mtx_unlock(&ctl_softc->ctl_lock);
909 	} else if (event == CTL_HA_EVT_MSG_SENT) {
910 		if (param != CTL_HA_STATUS_SUCCESS) {
911 			printf("Bad status from ctl_ha_msg_send status %d\n",
912 			       param);
913 		}
914 		return;
915 	} else if (event == CTL_HA_EVT_DISCONNECT) {
916 		printf("CTL: Got a disconnect from Isc\n");
917 		return;
918 	} else {
919 		printf("ctl_isc_event_handler: Unknown event %d\n", event);
920 		return;
921 	}
922 
923 bailout:
924 	return;
925 }
926 
927 static void
928 ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
929 {
930 	struct scsi_sense_data *sense;
931 
932 	sense = &dest->scsiio.sense_data;
933 	bcopy(&src->scsi.sense_data, sense, sizeof(*sense));
934 	dest->scsiio.scsi_status = src->scsi.scsi_status;
935 	dest->scsiio.sense_len = src->scsi.sense_len;
936 	dest->io_hdr.status = src->hdr.status;
937 }
938 
939 static void
940 ctl_init(void)
941 {
942 	struct ctl_softc *softc;
943 	struct ctl_io_pool *internal_pool, *emergency_pool, *other_pool;
944 	struct ctl_frontend *fe;
945 	struct ctl_lun *lun;
946         uint8_t sc_id =0;
947 #if 0
948 	int i;
949 #endif
950 	int retval;
951 	//int isc_retval;
952 
953 	retval = 0;
954 	ctl_pause_rtr = 0;
955         rcv_sync_msg = 0;
956 
957 	/* If we're disabled, don't initialize. */
958 	if (ctl_disable != 0)
959 		return;
960 
961 	control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
962 			       M_WAITOK | M_ZERO);
963 	softc = control_softc;
964 
965 	softc->dev = make_dev(&ctl_cdevsw, 0, UID_ROOT, GID_OPERATOR, 0600,
966 			      "cam/ctl");
967 
968 	softc->dev->si_drv1 = softc;
969 
970 	/*
971 	 * By default, return a "bad LUN" peripheral qualifier for unknown
972 	 * LUNs.  The user can override this default using the tunable or
973 	 * sysctl.  See the comment in ctl_inquiry_std() for more details.
974 	 */
975 	softc->inquiry_pq_no_lun = 1;
976 	TUNABLE_INT_FETCH("kern.cam.ctl.inquiry_pq_no_lun",
977 			  &softc->inquiry_pq_no_lun);
978 	sysctl_ctx_init(&softc->sysctl_ctx);
979 	softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
980 		SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
981 		CTLFLAG_RD, 0, "CAM Target Layer");
982 
983 	if (softc->sysctl_tree == NULL) {
984 		printf("%s: unable to allocate sysctl tree\n", __func__);
985 		destroy_dev(softc->dev);
986 		free(control_softc, M_DEVBUF);
987 		control_softc = NULL;
988 		return;
989 	}
990 
991 	SYSCTL_ADD_INT(&softc->sysctl_ctx,
992 		       SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO,
993 		       "inquiry_pq_no_lun", CTLFLAG_RW,
994 		       &softc->inquiry_pq_no_lun, 0,
995 		       "Report no lun possible for invalid LUNs");
996 
997 	mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
998 	softc->open_count = 0;
999 
1000 	/*
1001 	 * Default to actually sending a SYNCHRONIZE CACHE command down to
1002 	 * the drive.
1003 	 */
1004 	softc->flags = CTL_FLAG_REAL_SYNC;
1005 
1006 	/*
1007 	 * In Copan's HA scheme, the "master" and "slave" roles are
1008 	 * figured out through the slot the controller is in.  Although it
1009 	 * is an active/active system, someone has to be in charge.
1010  	 */
1011 #ifdef NEEDTOPORT
1012         scmicro_rw(SCMICRO_GET_SHELF_ID, &sc_id);
1013 #endif
1014 
1015         if (sc_id == 0) {
1016 		softc->flags |= CTL_FLAG_MASTER_SHELF;
1017 		persis_offset = 0;
1018 	} else
1019 		persis_offset = CTL_MAX_INITIATORS;
1020 
1021 	/*
1022 	 * XXX KDM need to figure out where we want to get our target ID
1023 	 * and WWID.  Is it different on each port?
1024 	 */
1025 	softc->target.id = 0;
1026 	softc->target.wwid[0] = 0x12345678;
1027 	softc->target.wwid[1] = 0x87654321;
1028 	STAILQ_INIT(&softc->lun_list);
1029 	STAILQ_INIT(&softc->pending_lun_queue);
1030 	STAILQ_INIT(&softc->task_queue);
1031 	STAILQ_INIT(&softc->incoming_queue);
1032 	STAILQ_INIT(&softc->rtr_queue);
1033 	STAILQ_INIT(&softc->done_queue);
1034 	STAILQ_INIT(&softc->isc_queue);
1035 	STAILQ_INIT(&softc->fe_list);
1036 	STAILQ_INIT(&softc->be_list);
1037 	STAILQ_INIT(&softc->io_pools);
1038 
1039 	lun = &softc->lun;
1040 
1041 	/*
1042 	 * We don't bother calling these with ctl_lock held here, because,
1043 	 * in theory, no one else can try to do anything while we're in our
1044 	 * module init routine.
1045 	 */
1046 	if (ctl_pool_create(softc, CTL_POOL_INTERNAL, CTL_POOL_ENTRIES_INTERNAL,
1047 			    &internal_pool)!= 0){
1048 		printf("ctl: can't allocate %d entry internal pool, "
1049 		       "exiting\n", CTL_POOL_ENTRIES_INTERNAL);
1050 		return;
1051 	}
1052 
1053 	if (ctl_pool_create(softc, CTL_POOL_EMERGENCY,
1054 			    CTL_POOL_ENTRIES_EMERGENCY, &emergency_pool) != 0) {
1055 		printf("ctl: can't allocate %d entry emergency pool, "
1056 		       "exiting\n", CTL_POOL_ENTRIES_EMERGENCY);
1057 		ctl_pool_free(softc, internal_pool);
1058 		return;
1059 	}
1060 
1061 	if (ctl_pool_create(softc, CTL_POOL_4OTHERSC, CTL_POOL_ENTRIES_OTHER_SC,
1062 	                    &other_pool) != 0)
1063 	{
1064 		printf("ctl: can't allocate %d entry other SC pool, "
1065 		       "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1066 		ctl_pool_free(softc, internal_pool);
1067 		ctl_pool_free(softc, emergency_pool);
1068 		return;
1069 	}
1070 
1071 	softc->internal_pool = internal_pool;
1072 	softc->emergency_pool = emergency_pool;
1073 	softc->othersc_pool = other_pool;
1074 
1075 	ctl_pool_acquire(internal_pool);
1076 	ctl_pool_acquire(emergency_pool);
1077 	ctl_pool_acquire(other_pool);
1078 
1079 	/*
1080 	 * We used to allocate a processor LUN here.  The new scheme is to
1081 	 * just let the user allocate LUNs as he sees fit.
1082 	 */
1083 #if 0
1084 	mtx_lock(&softc->ctl_lock);
1085 	ctl_alloc_lun(softc, lun, /*be_lun*/NULL, /*target*/softc->target);
1086 	mtx_unlock(&softc->ctl_lock);
1087 #endif
1088 
1089 	if (kproc_create(ctl_work_thread, softc, &softc->work_thread, 0, 0,
1090 			 "ctl_thrd") != 0) {
1091 		printf("error creating CTL work thread!\n");
1092 		ctl_free_lun(lun);
1093 		ctl_pool_free(softc, internal_pool);
1094 		ctl_pool_free(softc, emergency_pool);
1095 		ctl_pool_free(softc, other_pool);
1096 		return;
1097 	}
1098 	printf("ctl: CAM Target Layer loaded\n");
1099 
1100 	/*
1101 	 * Initialize the initiator and portname mappings
1102 	 */
1103 	memset(softc->wwpn_iid, 0, sizeof(softc->wwpn_iid));
1104 
1105 	/*
1106 	 * Initialize the ioctl front end.
1107 	 */
1108 	fe = &softc->ioctl_info.fe;
1109 	sprintf(softc->ioctl_info.port_name, "CTL ioctl");
1110 	fe->port_type = CTL_PORT_IOCTL;
1111 	fe->num_requested_ctl_io = 100;
1112 	fe->port_name = softc->ioctl_info.port_name;
1113 	fe->port_online = ctl_ioctl_online;
1114 	fe->port_offline = ctl_ioctl_offline;
1115 	fe->onoff_arg = &softc->ioctl_info;
1116 	fe->targ_enable = ctl_ioctl_targ_enable;
1117 	fe->targ_disable = ctl_ioctl_targ_disable;
1118 	fe->lun_enable = ctl_ioctl_lun_enable;
1119 	fe->lun_disable = ctl_ioctl_lun_disable;
1120 	fe->targ_lun_arg = &softc->ioctl_info;
1121 	fe->fe_datamove = ctl_ioctl_datamove;
1122 	fe->fe_done = ctl_ioctl_done;
1123 	fe->max_targets = 15;
1124 	fe->max_target_id = 15;
1125 
1126 	if (ctl_frontend_register(&softc->ioctl_info.fe,
1127 	                  (softc->flags & CTL_FLAG_MASTER_SHELF)) != 0) {
1128 		printf("ctl: ioctl front end registration failed, will "
1129 		       "continue anyway\n");
1130 	}
1131 
1132 #ifdef CTL_IO_DELAY
1133 	if (sizeof(struct callout) > CTL_TIMER_BYTES) {
1134 		printf("sizeof(struct callout) %zd > CTL_TIMER_BYTES %zd\n",
1135 		       sizeof(struct callout), CTL_TIMER_BYTES);
1136 		return;
1137 	}
1138 #endif /* CTL_IO_DELAY */
1139 
1140 }
1141 
1142 void
1143 ctl_shutdown(void)
1144 {
1145 	struct ctl_softc *softc;
1146 	struct ctl_lun *lun, *next_lun;
1147 	struct ctl_io_pool *pool, *next_pool;
1148 
1149 	softc = (struct ctl_softc *)control_softc;
1150 
1151 	if (ctl_frontend_deregister(&softc->ioctl_info.fe) != 0)
1152 		printf("ctl: ioctl front end deregistration failed\n");
1153 
1154 	mtx_lock(&softc->ctl_lock);
1155 
1156 	/*
1157 	 * Free up each LUN.
1158 	 */
1159 	for (lun = STAILQ_FIRST(&softc->lun_list); lun != NULL; lun = next_lun){
1160 		next_lun = STAILQ_NEXT(lun, links);
1161 		ctl_free_lun(lun);
1162 	}
1163 
1164 	/*
1165 	 * This will rip the rug out from under any FETDs or anyone else
1166 	 * that has a pool allocated.  Since we increment our module
1167 	 * refcount any time someone outside the main CTL module allocates
1168 	 * a pool, we shouldn't have any problems here.  The user won't be
1169 	 * able to unload the CTL module until client modules have
1170 	 * successfully unloaded.
1171 	 */
1172 	for (pool = STAILQ_FIRST(&softc->io_pools); pool != NULL;
1173 	     pool = next_pool) {
1174 		next_pool = STAILQ_NEXT(pool, links);
1175 		ctl_pool_free(softc, pool);
1176 	}
1177 
1178 	mtx_unlock(&softc->ctl_lock);
1179 
1180 #if 0
1181 	ctl_shutdown_thread(softc->work_thread);
1182 #endif
1183 
1184 	mtx_destroy(&softc->ctl_lock);
1185 
1186 	destroy_dev(softc->dev);
1187 
1188 	sysctl_ctx_free(&softc->sysctl_ctx);
1189 
1190 	free(control_softc, M_DEVBUF);
1191 	control_softc = NULL;
1192 
1193 	printf("ctl: CAM Target Layer unloaded\n");
1194 }
1195 
1196 /*
1197  * XXX KDM should we do some access checks here?  Bump a reference count to
1198  * prevent a CTL module from being unloaded while someone has it open?
1199  */
1200 static int
1201 ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1202 {
1203 	return (0);
1204 }
1205 
1206 static int
1207 ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1208 {
1209 	return (0);
1210 }
1211 
1212 int
1213 ctl_port_enable(ctl_port_type port_type)
1214 {
1215 	struct ctl_softc *softc;
1216 	struct ctl_frontend *fe;
1217 
1218 	if (ctl_is_single == 0) {
1219 		union ctl_ha_msg msg_info;
1220 		int isc_retval;
1221 
1222 #if 0
1223 		printf("%s: HA mode, synchronizing frontend enable\n",
1224 		        __func__);
1225 #endif
1226 		msg_info.hdr.msg_type = CTL_MSG_SYNC_FE;
1227 	        if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1228 		        sizeof(msg_info), 1 )) > CTL_HA_STATUS_SUCCESS) {
1229 			printf("Sync msg send error retval %d\n", isc_retval);
1230 		}
1231 		if (!rcv_sync_msg) {
1232 			isc_retval=ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
1233 			        sizeof(msg_info), 1);
1234 		}
1235 #if 0
1236         	printf("CTL:Frontend Enable\n");
1237 	} else {
1238 		printf("%s: single mode, skipping frontend synchronization\n",
1239 		        __func__);
1240 #endif
1241 	}
1242 
1243 	softc = control_softc;
1244 
1245 	STAILQ_FOREACH(fe, &softc->fe_list, links) {
1246 		if (port_type & fe->port_type)
1247 		{
1248 #if 0
1249 			printf("port %d\n", fe->targ_port);
1250 #endif
1251 			ctl_frontend_online(fe);
1252 		}
1253 	}
1254 
1255 	return (0);
1256 }
1257 
1258 int
1259 ctl_port_disable(ctl_port_type port_type)
1260 {
1261 	struct ctl_softc *softc;
1262 	struct ctl_frontend *fe;
1263 
1264 	softc = control_softc;
1265 
1266 	STAILQ_FOREACH(fe, &softc->fe_list, links) {
1267 		if (port_type & fe->port_type)
1268 			ctl_frontend_offline(fe);
1269 	}
1270 
1271 	return (0);
1272 }
1273 
1274 /*
1275  * Returns 0 for success, 1 for failure.
1276  * Currently the only failure mode is if there aren't enough entries
1277  * allocated.  So, in case of a failure, look at num_entries_dropped,
1278  * reallocate and try again.
1279  */
1280 int
1281 ctl_port_list(struct ctl_port_entry *entries, int num_entries_alloced,
1282 	      int *num_entries_filled, int *num_entries_dropped,
1283 	      ctl_port_type port_type, int no_virtual)
1284 {
1285 	struct ctl_softc *softc;
1286 	struct ctl_frontend *fe;
1287 	int entries_dropped, entries_filled;
1288 	int retval;
1289 	int i;
1290 
1291 	softc = control_softc;
1292 
1293 	retval = 0;
1294 	entries_filled = 0;
1295 	entries_dropped = 0;
1296 
1297 	i = 0;
1298 	mtx_lock(&softc->ctl_lock);
1299 	STAILQ_FOREACH(fe, &softc->fe_list, links) {
1300 		struct ctl_port_entry *entry;
1301 
1302 		if ((fe->port_type & port_type) == 0)
1303 			continue;
1304 
1305 		if ((no_virtual != 0)
1306 		 && (fe->virtual_port != 0))
1307 			continue;
1308 
1309 		if (entries_filled >= num_entries_alloced) {
1310 			entries_dropped++;
1311 			continue;
1312 		}
1313 		entry = &entries[i];
1314 
1315 		entry->port_type = fe->port_type;
1316 		strlcpy(entry->port_name, fe->port_name,
1317 			sizeof(entry->port_name));
1318 		entry->physical_port = fe->physical_port;
1319 		entry->virtual_port = fe->virtual_port;
1320 		entry->wwnn = fe->wwnn;
1321 		entry->wwpn = fe->wwpn;
1322 
1323 		i++;
1324 		entries_filled++;
1325 	}
1326 
1327 	mtx_unlock(&softc->ctl_lock);
1328 
1329 	if (entries_dropped > 0)
1330 		retval = 1;
1331 
1332 	*num_entries_dropped = entries_dropped;
1333 	*num_entries_filled = entries_filled;
1334 
1335 	return (retval);
1336 }
1337 
1338 static void
1339 ctl_ioctl_online(void *arg)
1340 {
1341 	struct ctl_ioctl_info *ioctl_info;
1342 
1343 	ioctl_info = (struct ctl_ioctl_info *)arg;
1344 
1345 	ioctl_info->flags |= CTL_IOCTL_FLAG_ENABLED;
1346 }
1347 
1348 static void
1349 ctl_ioctl_offline(void *arg)
1350 {
1351 	struct ctl_ioctl_info *ioctl_info;
1352 
1353 	ioctl_info = (struct ctl_ioctl_info *)arg;
1354 
1355 	ioctl_info->flags &= ~CTL_IOCTL_FLAG_ENABLED;
1356 }
1357 
1358 /*
1359  * Remove an initiator by port number and initiator ID.
1360  * Returns 0 for success, 1 for failure.
1361  * Assumes the caller does NOT hold the CTL lock.
1362  */
1363 int
1364 ctl_remove_initiator(int32_t targ_port, uint32_t iid)
1365 {
1366 	struct ctl_softc *softc;
1367 
1368 	softc = control_softc;
1369 
1370 	if ((targ_port < 0)
1371 	 || (targ_port > CTL_MAX_PORTS)) {
1372 		printf("%s: invalid port number %d\n", __func__, targ_port);
1373 		return (1);
1374 	}
1375 	if (iid > CTL_MAX_INIT_PER_PORT) {
1376 		printf("%s: initiator ID %u > maximun %u!\n",
1377 		       __func__, iid, CTL_MAX_INIT_PER_PORT);
1378 		return (1);
1379 	}
1380 
1381 	mtx_lock(&softc->ctl_lock);
1382 
1383 	softc->wwpn_iid[targ_port][iid].in_use = 0;
1384 
1385 	mtx_unlock(&softc->ctl_lock);
1386 
1387 	return (0);
1388 }
1389 
1390 /*
1391  * Add an initiator to the initiator map.
1392  * Returns 0 for success, 1 for failure.
1393  * Assumes the caller does NOT hold the CTL lock.
1394  */
1395 int
1396 ctl_add_initiator(uint64_t wwpn, int32_t targ_port, uint32_t iid)
1397 {
1398 	struct ctl_softc *softc;
1399 	int retval;
1400 
1401 	softc = control_softc;
1402 
1403 	retval = 0;
1404 
1405 	if ((targ_port < 0)
1406 	 || (targ_port > CTL_MAX_PORTS)) {
1407 		printf("%s: invalid port number %d\n", __func__, targ_port);
1408 		return (1);
1409 	}
1410 	if (iid > CTL_MAX_INIT_PER_PORT) {
1411 		printf("%s: WWPN %#jx initiator ID %u > maximun %u!\n",
1412 		       __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
1413 		return (1);
1414 	}
1415 
1416 	mtx_lock(&softc->ctl_lock);
1417 
1418 	if (softc->wwpn_iid[targ_port][iid].in_use != 0) {
1419 		/*
1420 		 * We don't treat this as an error.
1421 		 */
1422 		if (softc->wwpn_iid[targ_port][iid].wwpn == wwpn) {
1423 			printf("%s: port %d iid %u WWPN %#jx arrived again?\n",
1424 			       __func__, targ_port, iid, (uintmax_t)wwpn);
1425 			goto bailout;
1426 		}
1427 
1428 		/*
1429 		 * This is an error, but what do we do about it?  The
1430 		 * driver is telling us we have a new WWPN for this
1431 		 * initiator ID, so we pretty much need to use it.
1432 		 */
1433 		printf("%s: port %d iid %u WWPN %#jx arrived, WWPN %#jx is "
1434 		       "still at that address\n", __func__, targ_port, iid,
1435 		       (uintmax_t)wwpn,
1436 		       (uintmax_t)softc->wwpn_iid[targ_port][iid].wwpn);
1437 
1438 		/*
1439 		 * XXX KDM clear have_ca and ua_pending on each LUN for
1440 		 * this initiator.
1441 		 */
1442 	}
1443 	softc->wwpn_iid[targ_port][iid].in_use = 1;
1444 	softc->wwpn_iid[targ_port][iid].iid = iid;
1445 	softc->wwpn_iid[targ_port][iid].wwpn = wwpn;
1446 	softc->wwpn_iid[targ_port][iid].port = targ_port;
1447 
1448 bailout:
1449 
1450 	mtx_unlock(&softc->ctl_lock);
1451 
1452 	return (retval);
1453 }
1454 
1455 /*
1456  * XXX KDM should we pretend to do something in the target/lun
1457  * enable/disable functions?
1458  */
1459 static int
1460 ctl_ioctl_targ_enable(void *arg, struct ctl_id targ_id)
1461 {
1462 	return (0);
1463 }
1464 
1465 static int
1466 ctl_ioctl_targ_disable(void *arg, struct ctl_id targ_id)
1467 {
1468 	return (0);
1469 }
1470 
1471 static int
1472 ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id)
1473 {
1474 	return (0);
1475 }
1476 
1477 static int
1478 ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id)
1479 {
1480 	return (0);
1481 }
1482 
1483 /*
1484  * Data movement routine for the CTL ioctl frontend port.
1485  */
1486 static int
1487 ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio)
1488 {
1489 	struct ctl_sg_entry *ext_sglist, *kern_sglist;
1490 	struct ctl_sg_entry ext_entry, kern_entry;
1491 	int ext_sglen, ext_sg_entries, kern_sg_entries;
1492 	int ext_sg_start, ext_offset;
1493 	int len_to_copy, len_copied;
1494 	int kern_watermark, ext_watermark;
1495 	int ext_sglist_malloced;
1496 	int i, j;
1497 
1498 	ext_sglist_malloced = 0;
1499 	ext_sg_start = 0;
1500 	ext_offset = 0;
1501 
1502 	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove\n"));
1503 
1504 	/*
1505 	 * If this flag is set, fake the data transfer.
1506 	 */
1507 	if (ctsio->io_hdr.flags & CTL_FLAG_NO_DATAMOVE) {
1508 		ctsio->ext_data_filled = ctsio->ext_data_len;
1509 		goto bailout;
1510 	}
1511 
1512 	/*
1513 	 * To simplify things here, if we have a single buffer, stick it in
1514 	 * a S/G entry and just make it a single entry S/G list.
1515 	 */
1516 	if (ctsio->io_hdr.flags & CTL_FLAG_EDPTR_SGLIST) {
1517 		int len_seen;
1518 
1519 		ext_sglen = ctsio->ext_sg_entries * sizeof(*ext_sglist);
1520 
1521 		ext_sglist = (struct ctl_sg_entry *)malloc(ext_sglen, M_CTL,
1522 							   M_WAITOK);
1523 		ext_sglist_malloced = 1;
1524 		if (copyin(ctsio->ext_data_ptr, ext_sglist,
1525 				   ext_sglen) != 0) {
1526 			ctl_set_internal_failure(ctsio,
1527 						 /*sks_valid*/ 0,
1528 						 /*retry_count*/ 0);
1529 			goto bailout;
1530 		}
1531 		ext_sg_entries = ctsio->ext_sg_entries;
1532 		len_seen = 0;
1533 		for (i = 0; i < ext_sg_entries; i++) {
1534 			if ((len_seen + ext_sglist[i].len) >=
1535 			     ctsio->ext_data_filled) {
1536 				ext_sg_start = i;
1537 				ext_offset = ctsio->ext_data_filled - len_seen;
1538 				break;
1539 			}
1540 			len_seen += ext_sglist[i].len;
1541 		}
1542 	} else {
1543 		ext_sglist = &ext_entry;
1544 		ext_sglist->addr = ctsio->ext_data_ptr;
1545 		ext_sglist->len = ctsio->ext_data_len;
1546 		ext_sg_entries = 1;
1547 		ext_sg_start = 0;
1548 		ext_offset = ctsio->ext_data_filled;
1549 	}
1550 
1551 	if (ctsio->kern_sg_entries > 0) {
1552 		kern_sglist = (struct ctl_sg_entry *)ctsio->kern_data_ptr;
1553 		kern_sg_entries = ctsio->kern_sg_entries;
1554 	} else {
1555 		kern_sglist = &kern_entry;
1556 		kern_sglist->addr = ctsio->kern_data_ptr;
1557 		kern_sglist->len = ctsio->kern_data_len;
1558 		kern_sg_entries = 1;
1559 	}
1560 
1561 
1562 	kern_watermark = 0;
1563 	ext_watermark = ext_offset;
1564 	len_copied = 0;
1565 	for (i = ext_sg_start, j = 0;
1566 	     i < ext_sg_entries && j < kern_sg_entries;) {
1567 		uint8_t *ext_ptr, *kern_ptr;
1568 
1569 		len_to_copy = ctl_min(ext_sglist[i].len - ext_watermark,
1570 				      kern_sglist[j].len - kern_watermark);
1571 
1572 		ext_ptr = (uint8_t *)ext_sglist[i].addr;
1573 		ext_ptr = ext_ptr + ext_watermark;
1574 		if (ctsio->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
1575 			/*
1576 			 * XXX KDM fix this!
1577 			 */
1578 			panic("need to implement bus address support");
1579 #if 0
1580 			kern_ptr = bus_to_virt(kern_sglist[j].addr);
1581 #endif
1582 		} else
1583 			kern_ptr = (uint8_t *)kern_sglist[j].addr;
1584 		kern_ptr = kern_ptr + kern_watermark;
1585 
1586 		kern_watermark += len_to_copy;
1587 		ext_watermark += len_to_copy;
1588 
1589 		if ((ctsio->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
1590 		     CTL_FLAG_DATA_IN) {
1591 			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1592 					 "bytes to user\n", len_to_copy));
1593 			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1594 					 "to %p\n", kern_ptr, ext_ptr));
1595 			if (copyout(kern_ptr, ext_ptr, len_to_copy) != 0) {
1596 				ctl_set_internal_failure(ctsio,
1597 							 /*sks_valid*/ 0,
1598 							 /*retry_count*/ 0);
1599 				goto bailout;
1600 			}
1601 		} else {
1602 			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1603 					 "bytes from user\n", len_to_copy));
1604 			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1605 					 "to %p\n", ext_ptr, kern_ptr));
1606 			if (copyin(ext_ptr, kern_ptr, len_to_copy)!= 0){
1607 				ctl_set_internal_failure(ctsio,
1608 							 /*sks_valid*/ 0,
1609 							 /*retry_count*/0);
1610 				goto bailout;
1611 			}
1612 		}
1613 
1614 		len_copied += len_to_copy;
1615 
1616 		if (ext_sglist[i].len == ext_watermark) {
1617 			i++;
1618 			ext_watermark = 0;
1619 		}
1620 
1621 		if (kern_sglist[j].len == kern_watermark) {
1622 			j++;
1623 			kern_watermark = 0;
1624 		}
1625 	}
1626 
1627 	ctsio->ext_data_filled += len_copied;
1628 
1629 	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_sg_entries: %d, "
1630 			 "kern_sg_entries: %d\n", ext_sg_entries,
1631 			 kern_sg_entries));
1632 	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_data_len = %d, "
1633 			 "kern_data_len = %d\n", ctsio->ext_data_len,
1634 			 ctsio->kern_data_len));
1635 
1636 
1637 	/* XXX KDM set residual?? */
1638 bailout:
1639 
1640 	if (ext_sglist_malloced != 0)
1641 		free(ext_sglist, M_CTL);
1642 
1643 	return (CTL_RETVAL_COMPLETE);
1644 }
1645 
1646 /*
1647  * Serialize a command that went down the "wrong" side, and so was sent to
1648  * this controller for execution.  The logic is a little different than the
1649  * standard case in ctl_scsiio_precheck().  Errors in this case need to get
1650  * sent back to the other side, but in the success case, we execute the
1651  * command on this side (XFER mode) or tell the other side to execute it
1652  * (SER_ONLY mode).
1653  */
1654 static int
1655 ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio, int have_lock)
1656 {
1657 	struct ctl_softc *ctl_softc;
1658 	union ctl_ha_msg msg_info;
1659 	struct ctl_lun *lun;
1660 	int retval = 0;
1661 
1662 	ctl_softc = control_softc;
1663 	if (have_lock == 0)
1664 		mtx_lock(&ctl_softc->ctl_lock);
1665 
1666 	lun = ctl_softc->ctl_luns[ctsio->io_hdr.nexus.targ_lun];
1667 	if (lun==NULL)
1668 	{
1669 		/*
1670 		 * Why isn't LUN defined? The other side wouldn't
1671 		 * send a cmd if the LUN is undefined.
1672 		 */
1673 		printf("%s: Bad JUJU!, LUN is NULL!\n", __func__);
1674 
1675 		/* "Logical unit not supported" */
1676 		ctl_set_sense_data(&msg_info.scsi.sense_data,
1677 				   lun,
1678 				   /*sense_format*/SSD_TYPE_NONE,
1679 				   /*current_error*/ 1,
1680 				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1681 				   /*asc*/ 0x25,
1682 				   /*ascq*/ 0x00,
1683 				   SSD_ELEM_NONE);
1684 
1685 		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1686 		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1687 		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1688 		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1689 		msg_info.hdr.serializing_sc = NULL;
1690 		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1691 	        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1692 				sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1693 		}
1694 		if (have_lock == 0)
1695 			mtx_unlock(&ctl_softc->ctl_lock);
1696 		return(1);
1697 
1698 	}
1699 
1700     	TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1701 
1702 	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
1703 		(union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
1704 		 ooa_links))) {
1705 	case CTL_ACTION_BLOCK:
1706 		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
1707 		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
1708 				  blocked_links);
1709 		break;
1710 	case CTL_ACTION_PASS:
1711 	case CTL_ACTION_SKIP:
1712 		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
1713 			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
1714 			STAILQ_INSERT_TAIL(&ctl_softc->rtr_queue,
1715 					   &ctsio->io_hdr, links);
1716 		} else {
1717 
1718 			/* send msg back to other side */
1719 			msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1720 			msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
1721 			msg_info.hdr.msg_type = CTL_MSG_R2R;
1722 #if 0
1723 			printf("2. pOrig %x\n", (int)msg_info.hdr.original_sc);
1724 #endif
1725 		        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1726 			    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1727 			}
1728 		}
1729 		break;
1730 	case CTL_ACTION_OVERLAP:
1731 		/* OVERLAPPED COMMANDS ATTEMPTED */
1732 		ctl_set_sense_data(&msg_info.scsi.sense_data,
1733 				   lun,
1734 				   /*sense_format*/SSD_TYPE_NONE,
1735 				   /*current_error*/ 1,
1736 				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1737 				   /*asc*/ 0x4E,
1738 				   /*ascq*/ 0x00,
1739 				   SSD_ELEM_NONE);
1740 
1741 		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1742 		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1743 		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1744 		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1745 		msg_info.hdr.serializing_sc = NULL;
1746 		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1747 #if 0
1748 		printf("BAD JUJU:Major Bummer Overlap\n");
1749 #endif
1750 		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1751 		retval = 1;
1752 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1753 		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1754 		}
1755 		break;
1756 	case CTL_ACTION_OVERLAP_TAG:
1757 		/* TAGGED OVERLAPPED COMMANDS (NN = QUEUE TAG) */
1758 		ctl_set_sense_data(&msg_info.scsi.sense_data,
1759 				   lun,
1760 				   /*sense_format*/SSD_TYPE_NONE,
1761 				   /*current_error*/ 1,
1762 				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1763 				   /*asc*/ 0x4D,
1764 				   /*ascq*/ ctsio->tag_num & 0xff,
1765 				   SSD_ELEM_NONE);
1766 
1767 		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1768 		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1769 		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1770 		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1771 		msg_info.hdr.serializing_sc = NULL;
1772 		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1773 #if 0
1774 		printf("BAD JUJU:Major Bummer Overlap Tag\n");
1775 #endif
1776 		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1777 		retval = 1;
1778 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1779 		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1780 		}
1781 		break;
1782 	case CTL_ACTION_ERROR:
1783 	default:
1784 		/* "Internal target failure" */
1785 		ctl_set_sense_data(&msg_info.scsi.sense_data,
1786 				   lun,
1787 				   /*sense_format*/SSD_TYPE_NONE,
1788 				   /*current_error*/ 1,
1789 				   /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
1790 				   /*asc*/ 0x44,
1791 				   /*ascq*/ 0x00,
1792 				   SSD_ELEM_NONE);
1793 
1794 		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1795 		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1796 		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1797 		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1798 		msg_info.hdr.serializing_sc = NULL;
1799 		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1800 #if 0
1801 		printf("BAD JUJU:Major Bummer HW Error\n");
1802 #endif
1803 		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1804 		retval = 1;
1805 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1806 		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1807 		}
1808 		break;
1809 	}
1810 	if (have_lock == 0)
1811 		mtx_unlock(&ctl_softc->ctl_lock);
1812 	return (retval);
1813 }
1814 
1815 static int
1816 ctl_ioctl_submit_wait(union ctl_io *io)
1817 {
1818 	struct ctl_fe_ioctl_params params;
1819 	ctl_fe_ioctl_state last_state;
1820 	int done, retval;
1821 
1822 	retval = 0;
1823 
1824 	bzero(&params, sizeof(params));
1825 
1826 	mtx_init(&params.ioctl_mtx, "ctliocmtx", NULL, MTX_DEF);
1827 	cv_init(&params.sem, "ctlioccv");
1828 	params.state = CTL_IOCTL_INPROG;
1829 	last_state = params.state;
1830 
1831 	io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = &params;
1832 
1833 	CTL_DEBUG_PRINT(("ctl_ioctl_submit_wait\n"));
1834 
1835 	/* This shouldn't happen */
1836 	if ((retval = ctl_queue(io)) != CTL_RETVAL_COMPLETE)
1837 		return (retval);
1838 
1839 	done = 0;
1840 
1841 	do {
1842 		mtx_lock(&params.ioctl_mtx);
1843 		/*
1844 		 * Check the state here, and don't sleep if the state has
1845 		 * already changed (i.e. wakeup has already occured, but we
1846 		 * weren't waiting yet).
1847 		 */
1848 		if (params.state == last_state) {
1849 			/* XXX KDM cv_wait_sig instead? */
1850 			cv_wait(&params.sem, &params.ioctl_mtx);
1851 		}
1852 		last_state = params.state;
1853 
1854 		switch (params.state) {
1855 		case CTL_IOCTL_INPROG:
1856 			/* Why did we wake up? */
1857 			/* XXX KDM error here? */
1858 			mtx_unlock(&params.ioctl_mtx);
1859 			break;
1860 		case CTL_IOCTL_DATAMOVE:
1861 			CTL_DEBUG_PRINT(("got CTL_IOCTL_DATAMOVE\n"));
1862 
1863 			/*
1864 			 * change last_state back to INPROG to avoid
1865 			 * deadlock on subsequent data moves.
1866 			 */
1867 			params.state = last_state = CTL_IOCTL_INPROG;
1868 
1869 			mtx_unlock(&params.ioctl_mtx);
1870 			ctl_ioctl_do_datamove(&io->scsiio);
1871 			/*
1872 			 * Note that in some cases, most notably writes,
1873 			 * this will queue the I/O and call us back later.
1874 			 * In other cases, generally reads, this routine
1875 			 * will immediately call back and wake us up,
1876 			 * probably using our own context.
1877 			 */
1878 			io->scsiio.be_move_done(io);
1879 			break;
1880 		case CTL_IOCTL_DONE:
1881 			mtx_unlock(&params.ioctl_mtx);
1882 			CTL_DEBUG_PRINT(("got CTL_IOCTL_DONE\n"));
1883 			done = 1;
1884 			break;
1885 		default:
1886 			mtx_unlock(&params.ioctl_mtx);
1887 			/* XXX KDM error here? */
1888 			break;
1889 		}
1890 	} while (done == 0);
1891 
1892 	mtx_destroy(&params.ioctl_mtx);
1893 	cv_destroy(&params.sem);
1894 
1895 	return (CTL_RETVAL_COMPLETE);
1896 }
1897 
1898 static void
1899 ctl_ioctl_datamove(union ctl_io *io)
1900 {
1901 	struct ctl_fe_ioctl_params *params;
1902 
1903 	params = (struct ctl_fe_ioctl_params *)
1904 		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
1905 
1906 	mtx_lock(&params->ioctl_mtx);
1907 	params->state = CTL_IOCTL_DATAMOVE;
1908 	cv_broadcast(&params->sem);
1909 	mtx_unlock(&params->ioctl_mtx);
1910 }
1911 
1912 static void
1913 ctl_ioctl_done(union ctl_io *io)
1914 {
1915 	struct ctl_fe_ioctl_params *params;
1916 
1917 	params = (struct ctl_fe_ioctl_params *)
1918 		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
1919 
1920 	mtx_lock(&params->ioctl_mtx);
1921 	params->state = CTL_IOCTL_DONE;
1922 	cv_broadcast(&params->sem);
1923 	mtx_unlock(&params->ioctl_mtx);
1924 }
1925 
1926 static void
1927 ctl_ioctl_hard_startstop_callback(void *arg, struct cfi_metatask *metatask)
1928 {
1929 	struct ctl_fe_ioctl_startstop_info *sd_info;
1930 
1931 	sd_info = (struct ctl_fe_ioctl_startstop_info *)arg;
1932 
1933 	sd_info->hs_info.status = metatask->status;
1934 	sd_info->hs_info.total_luns = metatask->taskinfo.startstop.total_luns;
1935 	sd_info->hs_info.luns_complete =
1936 		metatask->taskinfo.startstop.luns_complete;
1937 	sd_info->hs_info.luns_failed = metatask->taskinfo.startstop.luns_failed;
1938 
1939 	cv_broadcast(&sd_info->sem);
1940 }
1941 
1942 static void
1943 ctl_ioctl_bbrread_callback(void *arg, struct cfi_metatask *metatask)
1944 {
1945 	struct ctl_fe_ioctl_bbrread_info *fe_bbr_info;
1946 
1947 	fe_bbr_info = (struct ctl_fe_ioctl_bbrread_info *)arg;
1948 
1949 	mtx_lock(fe_bbr_info->lock);
1950 	fe_bbr_info->bbr_info->status = metatask->status;
1951 	fe_bbr_info->bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
1952 	fe_bbr_info->wakeup_done = 1;
1953 	mtx_unlock(fe_bbr_info->lock);
1954 
1955 	cv_broadcast(&fe_bbr_info->sem);
1956 }
1957 
1958 /*
1959  * Must be called with the ctl_lock held.
1960  * Returns 0 for success, errno for failure.
1961  */
1962 static int
1963 ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
1964 		   struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
1965 {
1966 	union ctl_io *io;
1967 	int retval;
1968 
1969 	retval = 0;
1970 
1971 	for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
1972 	     (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
1973 	     ooa_links)) {
1974 		struct ctl_ooa_entry *entry;
1975 
1976 		/*
1977 		 * If we've got more than we can fit, just count the
1978 		 * remaining entries.
1979 		 */
1980 		if (*cur_fill_num >= ooa_hdr->alloc_num)
1981 			continue;
1982 
1983 		entry = &kern_entries[*cur_fill_num];
1984 
1985 		entry->tag_num = io->scsiio.tag_num;
1986 		entry->lun_num = lun->lun;
1987 #ifdef CTL_TIME_IO
1988 		entry->start_bt = io->io_hdr.start_bt;
1989 #endif
1990 		bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
1991 		entry->cdb_len = io->scsiio.cdb_len;
1992 		if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
1993 			entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
1994 
1995 		if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
1996 			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
1997 
1998 		if (io->io_hdr.flags & CTL_FLAG_ABORT)
1999 			entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2000 
2001 		if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2002 			entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2003 
2004 		if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2005 			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2006 	}
2007 
2008 	return (retval);
2009 }
2010 
2011 static void *
2012 ctl_copyin_alloc(void *user_addr, int len, char *error_str,
2013 		 size_t error_str_len)
2014 {
2015 	void *kptr;
2016 
2017 	kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2018 
2019 	if (copyin(user_addr, kptr, len) != 0) {
2020 		snprintf(error_str, error_str_len, "Error copying %d bytes "
2021 			 "from user address %p to kernel address %p", len,
2022 			 user_addr, kptr);
2023 		free(kptr, M_CTL);
2024 		return (NULL);
2025 	}
2026 
2027 	return (kptr);
2028 }
2029 
2030 static void
2031 ctl_free_args(int num_be_args, struct ctl_be_arg *be_args)
2032 {
2033 	int i;
2034 
2035 	if (be_args == NULL)
2036 		return;
2037 
2038 	for (i = 0; i < num_be_args; i++) {
2039 		free(be_args[i].kname, M_CTL);
2040 		free(be_args[i].kvalue, M_CTL);
2041 	}
2042 
2043 	free(be_args, M_CTL);
2044 }
2045 
2046 static struct ctl_be_arg *
2047 ctl_copyin_args(int num_be_args, struct ctl_be_arg *be_args,
2048 		char *error_str, size_t error_str_len)
2049 {
2050 	struct ctl_be_arg *args;
2051 	int i;
2052 
2053 	args = ctl_copyin_alloc(be_args, num_be_args * sizeof(*be_args),
2054 				error_str, error_str_len);
2055 
2056 	if (args == NULL)
2057 		goto bailout;
2058 
2059 	for (i = 0; i < num_be_args; i++) {
2060 		args[i].kname = NULL;
2061 		args[i].kvalue = NULL;
2062 	}
2063 
2064 	for (i = 0; i < num_be_args; i++) {
2065 		uint8_t *tmpptr;
2066 
2067 		args[i].kname = ctl_copyin_alloc(args[i].name,
2068 			args[i].namelen, error_str, error_str_len);
2069 		if (args[i].kname == NULL)
2070 			goto bailout;
2071 
2072 		if (args[i].kname[args[i].namelen - 1] != '\0') {
2073 			snprintf(error_str, error_str_len, "Argument %d "
2074 				 "name is not NUL-terminated", i);
2075 			goto bailout;
2076 		}
2077 
2078 		args[i].kvalue = NULL;
2079 
2080 		tmpptr = ctl_copyin_alloc(args[i].value,
2081 			args[i].vallen, error_str, error_str_len);
2082 		if (tmpptr == NULL)
2083 			goto bailout;
2084 
2085 		args[i].kvalue = tmpptr;
2086 
2087 		if ((args[i].flags & CTL_BEARG_ASCII)
2088 		 && (tmpptr[args[i].vallen - 1] != '\0')) {
2089 			snprintf(error_str, error_str_len, "Argument %d "
2090 				 "value is not NUL-terminated", i);
2091 			goto bailout;
2092 		}
2093 	}
2094 
2095 	return (args);
2096 bailout:
2097 
2098 	ctl_free_args(num_be_args, args);
2099 
2100 	return (NULL);
2101 }
2102 
2103 /*
2104  * Escape characters that are illegal or not recommended in XML.
2105  */
2106 int
2107 ctl_sbuf_printf_esc(struct sbuf *sb, char *str)
2108 {
2109 	int retval;
2110 
2111 	retval = 0;
2112 
2113 	for (; *str; str++) {
2114 		switch (*str) {
2115 		case '&':
2116 			retval = sbuf_printf(sb, "&amp;");
2117 			break;
2118 		case '>':
2119 			retval = sbuf_printf(sb, "&gt;");
2120 			break;
2121 		case '<':
2122 			retval = sbuf_printf(sb, "&lt;");
2123 			break;
2124 		default:
2125 			retval = sbuf_putc(sb, *str);
2126 			break;
2127 		}
2128 
2129 		if (retval != 0)
2130 			break;
2131 
2132 	}
2133 
2134 	return (retval);
2135 }
2136 
2137 static int
2138 ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2139 	  struct thread *td)
2140 {
2141 	struct ctl_softc *softc;
2142 	int retval;
2143 
2144 	softc = control_softc;
2145 
2146 	retval = 0;
2147 
2148 	switch (cmd) {
2149 	case CTL_IO: {
2150 		union ctl_io *io;
2151 		void *pool_tmp;
2152 
2153 		/*
2154 		 * If we haven't been "enabled", don't allow any SCSI I/O
2155 		 * to this FETD.
2156 		 */
2157 		if ((softc->ioctl_info.flags & CTL_IOCTL_FLAG_ENABLED) == 0) {
2158 			retval = -EPERM;
2159 			break;
2160 		}
2161 
2162 		io = ctl_alloc_io(softc->ioctl_info.fe.ctl_pool_ref);
2163 		if (io == NULL) {
2164 			printf("ctl_ioctl: can't allocate ctl_io!\n");
2165 			retval = -ENOSPC;
2166 			break;
2167 		}
2168 
2169 		/*
2170 		 * Need to save the pool reference so it doesn't get
2171 		 * spammed by the user's ctl_io.
2172 		 */
2173 		pool_tmp = io->io_hdr.pool;
2174 
2175 		memcpy(io, (void *)addr, sizeof(*io));
2176 
2177 		io->io_hdr.pool = pool_tmp;
2178 		/*
2179 		 * No status yet, so make sure the status is set properly.
2180 		 */
2181 		io->io_hdr.status = CTL_STATUS_NONE;
2182 
2183 		/*
2184 		 * The user sets the initiator ID, target and LUN IDs.
2185 		 */
2186 		io->io_hdr.nexus.targ_port = softc->ioctl_info.fe.targ_port;
2187 		io->io_hdr.flags |= CTL_FLAG_USER_REQ;
2188 		if ((io->io_hdr.io_type == CTL_IO_SCSI)
2189 		 && (io->scsiio.tag_type != CTL_TAG_UNTAGGED))
2190 			io->scsiio.tag_num = softc->ioctl_info.cur_tag_num++;
2191 
2192 		retval = ctl_ioctl_submit_wait(io);
2193 
2194 		if (retval != 0) {
2195 			ctl_free_io(io);
2196 			break;
2197 		}
2198 
2199 		memcpy((void *)addr, io, sizeof(*io));
2200 
2201 		/* return this to our pool */
2202 		ctl_free_io(io);
2203 
2204 		break;
2205 	}
2206 	case CTL_ENABLE_PORT:
2207 	case CTL_DISABLE_PORT:
2208 	case CTL_SET_PORT_WWNS: {
2209 		struct ctl_frontend *fe;
2210 		struct ctl_port_entry *entry;
2211 
2212 		entry = (struct ctl_port_entry *)addr;
2213 
2214 		mtx_lock(&softc->ctl_lock);
2215 		STAILQ_FOREACH(fe, &softc->fe_list, links) {
2216 			int action, done;
2217 
2218 			action = 0;
2219 			done = 0;
2220 
2221 			if ((entry->port_type == CTL_PORT_NONE)
2222 			 && (entry->targ_port == fe->targ_port)) {
2223 				/*
2224 				 * If the user only wants to enable or
2225 				 * disable or set WWNs on a specific port,
2226 				 * do the operation and we're done.
2227 				 */
2228 				action = 1;
2229 				done = 1;
2230 			} else if (entry->port_type & fe->port_type) {
2231 				/*
2232 				 * Compare the user's type mask with the
2233 				 * particular frontend type to see if we
2234 				 * have a match.
2235 				 */
2236 				action = 1;
2237 				done = 0;
2238 
2239 				/*
2240 				 * Make sure the user isn't trying to set
2241 				 * WWNs on multiple ports at the same time.
2242 				 */
2243 				if (cmd == CTL_SET_PORT_WWNS) {
2244 					printf("%s: Can't set WWNs on "
2245 					       "multiple ports\n", __func__);
2246 					retval = EINVAL;
2247 					break;
2248 				}
2249 			}
2250 			if (action != 0) {
2251 				/*
2252 				 * XXX KDM we have to drop the lock here,
2253 				 * because the online/offline operations
2254 				 * can potentially block.  We need to
2255 				 * reference count the frontends so they
2256 				 * can't go away,
2257 				 */
2258 				mtx_unlock(&softc->ctl_lock);
2259 
2260 				if (cmd == CTL_ENABLE_PORT) {
2261 					struct ctl_lun *lun;
2262 
2263 					STAILQ_FOREACH(lun, &softc->lun_list,
2264 						       links) {
2265 						fe->lun_enable(fe->targ_lun_arg,
2266 						    lun->target,
2267 						    lun->lun);
2268 					}
2269 
2270 					ctl_frontend_online(fe);
2271 				} else if (cmd == CTL_DISABLE_PORT) {
2272 					struct ctl_lun *lun;
2273 
2274 					ctl_frontend_offline(fe);
2275 
2276 					STAILQ_FOREACH(lun, &softc->lun_list,
2277 						       links) {
2278 						fe->lun_disable(
2279 						    fe->targ_lun_arg,
2280 						    lun->target,
2281 						    lun->lun);
2282 					}
2283 				}
2284 
2285 				mtx_lock(&softc->ctl_lock);
2286 
2287 				if (cmd == CTL_SET_PORT_WWNS)
2288 					ctl_frontend_set_wwns(fe,
2289 					    (entry->flags & CTL_PORT_WWNN_VALID) ?
2290 					    1 : 0, entry->wwnn,
2291 					    (entry->flags & CTL_PORT_WWPN_VALID) ?
2292 					    1 : 0, entry->wwpn);
2293 			}
2294 			if (done != 0)
2295 				break;
2296 		}
2297 		mtx_unlock(&softc->ctl_lock);
2298 		break;
2299 	}
2300 	case CTL_GET_PORT_LIST: {
2301 		struct ctl_frontend *fe;
2302 		struct ctl_port_list *list;
2303 		int i;
2304 
2305 		list = (struct ctl_port_list *)addr;
2306 
2307 		if (list->alloc_len != (list->alloc_num *
2308 		    sizeof(struct ctl_port_entry))) {
2309 			printf("%s: CTL_GET_PORT_LIST: alloc_len %u != "
2310 			       "alloc_num %u * sizeof(struct ctl_port_entry) "
2311 			       "%zu\n", __func__, list->alloc_len,
2312 			       list->alloc_num, sizeof(struct ctl_port_entry));
2313 			retval = EINVAL;
2314 			break;
2315 		}
2316 		list->fill_len = 0;
2317 		list->fill_num = 0;
2318 		list->dropped_num = 0;
2319 		i = 0;
2320 		mtx_lock(&softc->ctl_lock);
2321 		STAILQ_FOREACH(fe, &softc->fe_list, links) {
2322 			struct ctl_port_entry entry, *list_entry;
2323 
2324 			if (list->fill_num >= list->alloc_num) {
2325 				list->dropped_num++;
2326 				continue;
2327 			}
2328 
2329 			entry.port_type = fe->port_type;
2330 			strlcpy(entry.port_name, fe->port_name,
2331 				sizeof(entry.port_name));
2332 			entry.targ_port = fe->targ_port;
2333 			entry.physical_port = fe->physical_port;
2334 			entry.virtual_port = fe->virtual_port;
2335 			entry.wwnn = fe->wwnn;
2336 			entry.wwpn = fe->wwpn;
2337 			if (fe->status & CTL_PORT_STATUS_ONLINE)
2338 				entry.online = 1;
2339 			else
2340 				entry.online = 0;
2341 
2342 			list_entry = &list->entries[i];
2343 
2344 			retval = copyout(&entry, list_entry, sizeof(entry));
2345 			if (retval != 0) {
2346 				printf("%s: CTL_GET_PORT_LIST: copyout "
2347 				       "returned %d\n", __func__, retval);
2348 				break;
2349 			}
2350 			i++;
2351 			list->fill_num++;
2352 			list->fill_len += sizeof(entry);
2353 		}
2354 		mtx_unlock(&softc->ctl_lock);
2355 
2356 		/*
2357 		 * If this is non-zero, we had a copyout fault, so there's
2358 		 * probably no point in attempting to set the status inside
2359 		 * the structure.
2360 		 */
2361 		if (retval != 0)
2362 			break;
2363 
2364 		if (list->dropped_num > 0)
2365 			list->status = CTL_PORT_LIST_NEED_MORE_SPACE;
2366 		else
2367 			list->status = CTL_PORT_LIST_OK;
2368 		break;
2369 	}
2370 	case CTL_DUMP_OOA: {
2371 		struct ctl_lun *lun;
2372 		union ctl_io *io;
2373 		char printbuf[128];
2374 		struct sbuf sb;
2375 
2376 		mtx_lock(&softc->ctl_lock);
2377 		printf("Dumping OOA queues:\n");
2378 		STAILQ_FOREACH(lun, &softc->lun_list, links) {
2379 			for (io = (union ctl_io *)TAILQ_FIRST(
2380 			     &lun->ooa_queue); io != NULL;
2381 			     io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2382 			     ooa_links)) {
2383 				sbuf_new(&sb, printbuf, sizeof(printbuf),
2384 					 SBUF_FIXEDLEN);
2385 				sbuf_printf(&sb, "LUN %jd tag 0x%04x%s%s%s%s: ",
2386 					    (intmax_t)lun->lun,
2387 					    io->scsiio.tag_num,
2388 					    (io->io_hdr.flags &
2389 					    CTL_FLAG_BLOCKED) ? "" : " BLOCKED",
2390 					    (io->io_hdr.flags &
2391 					    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
2392 					    (io->io_hdr.flags &
2393 					    CTL_FLAG_ABORT) ? " ABORT" : "",
2394 			                    (io->io_hdr.flags &
2395 		                        CTL_FLAG_IS_WAS_ON_RTR) ? " RTR" : "");
2396 				ctl_scsi_command_string(&io->scsiio, NULL, &sb);
2397 				sbuf_finish(&sb);
2398 				printf("%s\n", sbuf_data(&sb));
2399 			}
2400 		}
2401 		printf("OOA queues dump done\n");
2402 		mtx_unlock(&softc->ctl_lock);
2403 		break;
2404 	}
2405 	case CTL_GET_OOA: {
2406 		struct ctl_lun *lun;
2407 		struct ctl_ooa *ooa_hdr;
2408 		struct ctl_ooa_entry *entries;
2409 		uint32_t cur_fill_num;
2410 
2411 		ooa_hdr = (struct ctl_ooa *)addr;
2412 
2413 		if ((ooa_hdr->alloc_len == 0)
2414 		 || (ooa_hdr->alloc_num == 0)) {
2415 			printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2416 			       "must be non-zero\n", __func__,
2417 			       ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2418 			retval = EINVAL;
2419 			break;
2420 		}
2421 
2422 		if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2423 		    sizeof(struct ctl_ooa_entry))) {
2424 			printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2425 			       "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2426 			       __func__, ooa_hdr->alloc_len,
2427 			       ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2428 			retval = EINVAL;
2429 			break;
2430 		}
2431 
2432 		entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2433 		if (entries == NULL) {
2434 			printf("%s: could not allocate %d bytes for OOA "
2435 			       "dump\n", __func__, ooa_hdr->alloc_len);
2436 			retval = ENOMEM;
2437 			break;
2438 		}
2439 
2440 		mtx_lock(&softc->ctl_lock);
2441 		if (((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0)
2442 		 && ((ooa_hdr->lun_num > CTL_MAX_LUNS)
2443 		  || (softc->ctl_luns[ooa_hdr->lun_num] == NULL))) {
2444 			mtx_unlock(&softc->ctl_lock);
2445 			free(entries, M_CTL);
2446 			printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2447 			       __func__, (uintmax_t)ooa_hdr->lun_num);
2448 			retval = EINVAL;
2449 			break;
2450 		}
2451 
2452 		cur_fill_num = 0;
2453 
2454 		if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2455 			STAILQ_FOREACH(lun, &softc->lun_list, links) {
2456 				retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2457 					ooa_hdr, entries);
2458 				if (retval != 0)
2459 					break;
2460 			}
2461 			if (retval != 0) {
2462 				mtx_unlock(&softc->ctl_lock);
2463 				free(entries, M_CTL);
2464 				break;
2465 			}
2466 		} else {
2467 			lun = softc->ctl_luns[ooa_hdr->lun_num];
2468 
2469 			retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,ooa_hdr,
2470 						    entries);
2471 		}
2472 		mtx_unlock(&softc->ctl_lock);
2473 
2474 		ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2475 		ooa_hdr->fill_len = ooa_hdr->fill_num *
2476 			sizeof(struct ctl_ooa_entry);
2477 		retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2478 		if (retval != 0) {
2479 			printf("%s: error copying out %d bytes for OOA dump\n",
2480 			       __func__, ooa_hdr->fill_len);
2481 		}
2482 
2483 		getbintime(&ooa_hdr->cur_bt);
2484 
2485 		if (cur_fill_num > ooa_hdr->alloc_num) {
2486 			ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2487 			ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2488 		} else {
2489 			ooa_hdr->dropped_num = 0;
2490 			ooa_hdr->status = CTL_OOA_OK;
2491 		}
2492 
2493 		free(entries, M_CTL);
2494 		break;
2495 	}
2496 	case CTL_CHECK_OOA: {
2497 		union ctl_io *io;
2498 		struct ctl_lun *lun;
2499 		struct ctl_ooa_info *ooa_info;
2500 
2501 
2502 		ooa_info = (struct ctl_ooa_info *)addr;
2503 
2504 		if (ooa_info->lun_id >= CTL_MAX_LUNS) {
2505 			ooa_info->status = CTL_OOA_INVALID_LUN;
2506 			break;
2507 		}
2508 		mtx_lock(&softc->ctl_lock);
2509 		lun = softc->ctl_luns[ooa_info->lun_id];
2510 		if (lun == NULL) {
2511 			mtx_unlock(&softc->ctl_lock);
2512 			ooa_info->status = CTL_OOA_INVALID_LUN;
2513 			break;
2514 		}
2515 
2516 		ooa_info->num_entries = 0;
2517 		for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
2518 		     io != NULL; io = (union ctl_io *)TAILQ_NEXT(
2519 		     &io->io_hdr, ooa_links)) {
2520 			ooa_info->num_entries++;
2521 		}
2522 
2523 		mtx_unlock(&softc->ctl_lock);
2524 		ooa_info->status = CTL_OOA_SUCCESS;
2525 
2526 		break;
2527 	}
2528 	case CTL_HARD_START:
2529 	case CTL_HARD_STOP: {
2530 		struct ctl_fe_ioctl_startstop_info ss_info;
2531 		struct cfi_metatask *metatask;
2532 		struct mtx hs_mtx;
2533 
2534 		mtx_init(&hs_mtx, "HS Mutex", NULL, MTX_DEF);
2535 
2536 		cv_init(&ss_info.sem, "hard start/stop cv" );
2537 
2538 		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2539 		if (metatask == NULL) {
2540 			retval = ENOMEM;
2541 			mtx_destroy(&hs_mtx);
2542 			break;
2543 		}
2544 
2545 		if (cmd == CTL_HARD_START)
2546 			metatask->tasktype = CFI_TASK_STARTUP;
2547 		else
2548 			metatask->tasktype = CFI_TASK_SHUTDOWN;
2549 
2550 		metatask->callback = ctl_ioctl_hard_startstop_callback;
2551 		metatask->callback_arg = &ss_info;
2552 
2553 		cfi_action(metatask);
2554 
2555 		/* Wait for the callback */
2556 		mtx_lock(&hs_mtx);
2557 		cv_wait_sig(&ss_info.sem, &hs_mtx);
2558 		mtx_unlock(&hs_mtx);
2559 
2560 		/*
2561 		 * All information has been copied from the metatask by the
2562 		 * time cv_broadcast() is called, so we free the metatask here.
2563 		 */
2564 		cfi_free_metatask(metatask);
2565 
2566 		memcpy((void *)addr, &ss_info.hs_info, sizeof(ss_info.hs_info));
2567 
2568 		mtx_destroy(&hs_mtx);
2569 		break;
2570 	}
2571 	case CTL_BBRREAD: {
2572 		struct ctl_bbrread_info *bbr_info;
2573 		struct ctl_fe_ioctl_bbrread_info fe_bbr_info;
2574 		struct mtx bbr_mtx;
2575 		struct cfi_metatask *metatask;
2576 
2577 		bbr_info = (struct ctl_bbrread_info *)addr;
2578 
2579 		bzero(&fe_bbr_info, sizeof(fe_bbr_info));
2580 
2581 		bzero(&bbr_mtx, sizeof(bbr_mtx));
2582 		mtx_init(&bbr_mtx, "BBR Mutex", NULL, MTX_DEF);
2583 
2584 		fe_bbr_info.bbr_info = bbr_info;
2585 		fe_bbr_info.lock = &bbr_mtx;
2586 
2587 		cv_init(&fe_bbr_info.sem, "BBR read cv");
2588 		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2589 
2590 		if (metatask == NULL) {
2591 			mtx_destroy(&bbr_mtx);
2592 			cv_destroy(&fe_bbr_info.sem);
2593 			retval = ENOMEM;
2594 			break;
2595 		}
2596 		metatask->tasktype = CFI_TASK_BBRREAD;
2597 		metatask->callback = ctl_ioctl_bbrread_callback;
2598 		metatask->callback_arg = &fe_bbr_info;
2599 		metatask->taskinfo.bbrread.lun_num = bbr_info->lun_num;
2600 		metatask->taskinfo.bbrread.lba = bbr_info->lba;
2601 		metatask->taskinfo.bbrread.len = bbr_info->len;
2602 
2603 		cfi_action(metatask);
2604 
2605 		mtx_lock(&bbr_mtx);
2606 		while (fe_bbr_info.wakeup_done == 0)
2607 			cv_wait_sig(&fe_bbr_info.sem, &bbr_mtx);
2608 		mtx_unlock(&bbr_mtx);
2609 
2610 		bbr_info->status = metatask->status;
2611 		bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2612 		bbr_info->scsi_status = metatask->taskinfo.bbrread.scsi_status;
2613 		memcpy(&bbr_info->sense_data,
2614 		       &metatask->taskinfo.bbrread.sense_data,
2615 		       ctl_min(sizeof(bbr_info->sense_data),
2616 			       sizeof(metatask->taskinfo.bbrread.sense_data)));
2617 
2618 		cfi_free_metatask(metatask);
2619 
2620 		mtx_destroy(&bbr_mtx);
2621 		cv_destroy(&fe_bbr_info.sem);
2622 
2623 		break;
2624 	}
2625 	case CTL_DELAY_IO: {
2626 		struct ctl_io_delay_info *delay_info;
2627 #ifdef CTL_IO_DELAY
2628 		struct ctl_lun *lun;
2629 #endif /* CTL_IO_DELAY */
2630 
2631 		delay_info = (struct ctl_io_delay_info *)addr;
2632 
2633 #ifdef CTL_IO_DELAY
2634 		mtx_lock(&softc->ctl_lock);
2635 
2636 		if ((delay_info->lun_id > CTL_MAX_LUNS)
2637 		 || (softc->ctl_luns[delay_info->lun_id] == NULL)) {
2638 			delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2639 		} else {
2640 			lun = softc->ctl_luns[delay_info->lun_id];
2641 
2642 			delay_info->status = CTL_DELAY_STATUS_OK;
2643 
2644 			switch (delay_info->delay_type) {
2645 			case CTL_DELAY_TYPE_CONT:
2646 				break;
2647 			case CTL_DELAY_TYPE_ONESHOT:
2648 				break;
2649 			default:
2650 				delay_info->status =
2651 					CTL_DELAY_STATUS_INVALID_TYPE;
2652 				break;
2653 			}
2654 
2655 			switch (delay_info->delay_loc) {
2656 			case CTL_DELAY_LOC_DATAMOVE:
2657 				lun->delay_info.datamove_type =
2658 					delay_info->delay_type;
2659 				lun->delay_info.datamove_delay =
2660 					delay_info->delay_secs;
2661 				break;
2662 			case CTL_DELAY_LOC_DONE:
2663 				lun->delay_info.done_type =
2664 					delay_info->delay_type;
2665 				lun->delay_info.done_delay =
2666 					delay_info->delay_secs;
2667 				break;
2668 			default:
2669 				delay_info->status =
2670 					CTL_DELAY_STATUS_INVALID_LOC;
2671 				break;
2672 			}
2673 		}
2674 
2675 		mtx_unlock(&softc->ctl_lock);
2676 #else
2677 		delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2678 #endif /* CTL_IO_DELAY */
2679 		break;
2680 	}
2681 	case CTL_REALSYNC_SET: {
2682 		int *syncstate;
2683 
2684 		syncstate = (int *)addr;
2685 
2686 		mtx_lock(&softc->ctl_lock);
2687 		switch (*syncstate) {
2688 		case 0:
2689 			softc->flags &= ~CTL_FLAG_REAL_SYNC;
2690 			break;
2691 		case 1:
2692 			softc->flags |= CTL_FLAG_REAL_SYNC;
2693 			break;
2694 		default:
2695 			retval = -EINVAL;
2696 			break;
2697 		}
2698 		mtx_unlock(&softc->ctl_lock);
2699 		break;
2700 	}
2701 	case CTL_REALSYNC_GET: {
2702 		int *syncstate;
2703 
2704 		syncstate = (int*)addr;
2705 
2706 		mtx_lock(&softc->ctl_lock);
2707 		if (softc->flags & CTL_FLAG_REAL_SYNC)
2708 			*syncstate = 1;
2709 		else
2710 			*syncstate = 0;
2711 		mtx_unlock(&softc->ctl_lock);
2712 
2713 		break;
2714 	}
2715 	case CTL_SETSYNC:
2716 	case CTL_GETSYNC: {
2717 		struct ctl_sync_info *sync_info;
2718 		struct ctl_lun *lun;
2719 
2720 		sync_info = (struct ctl_sync_info *)addr;
2721 
2722 		mtx_lock(&softc->ctl_lock);
2723 		lun = softc->ctl_luns[sync_info->lun_id];
2724 		if (lun == NULL) {
2725 			mtx_unlock(&softc->ctl_lock);
2726 			sync_info->status = CTL_GS_SYNC_NO_LUN;
2727 		}
2728 		/*
2729 		 * Get or set the sync interval.  We're not bounds checking
2730 		 * in the set case, hopefully the user won't do something
2731 		 * silly.
2732 		 */
2733 		if (cmd == CTL_GETSYNC)
2734 			sync_info->sync_interval = lun->sync_interval;
2735 		else
2736 			lun->sync_interval = sync_info->sync_interval;
2737 
2738 		mtx_unlock(&softc->ctl_lock);
2739 
2740 		sync_info->status = CTL_GS_SYNC_OK;
2741 
2742 		break;
2743 	}
2744 	case CTL_GETSTATS: {
2745 		struct ctl_stats *stats;
2746 		struct ctl_lun *lun;
2747 		int i;
2748 
2749 		stats = (struct ctl_stats *)addr;
2750 
2751 		if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) >
2752 		     stats->alloc_len) {
2753 			stats->status = CTL_SS_NEED_MORE_SPACE;
2754 			stats->num_luns = softc->num_luns;
2755 			break;
2756 		}
2757 		/*
2758 		 * XXX KDM no locking here.  If the LUN list changes,
2759 		 * things can blow up.
2760 		 */
2761 		for (i = 0, lun = STAILQ_FIRST(&softc->lun_list); lun != NULL;
2762 		     i++, lun = STAILQ_NEXT(lun, links)) {
2763 			retval = copyout(&lun->stats, &stats->lun_stats[i],
2764 					 sizeof(lun->stats));
2765 			if (retval != 0)
2766 				break;
2767 		}
2768 		stats->num_luns = softc->num_luns;
2769 		stats->fill_len = sizeof(struct ctl_lun_io_stats) *
2770 				 softc->num_luns;
2771 		stats->status = CTL_SS_OK;
2772 #ifdef CTL_TIME_IO
2773 		stats->flags = CTL_STATS_FLAG_TIME_VALID;
2774 #else
2775 		stats->flags = CTL_STATS_FLAG_NONE;
2776 #endif
2777 		getnanouptime(&stats->timestamp);
2778 		break;
2779 	}
2780 	case CTL_ERROR_INJECT: {
2781 		struct ctl_error_desc *err_desc, *new_err_desc;
2782 		struct ctl_lun *lun;
2783 
2784 		err_desc = (struct ctl_error_desc *)addr;
2785 
2786 		new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2787 				      M_WAITOK | M_ZERO);
2788 		bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2789 
2790 		mtx_lock(&softc->ctl_lock);
2791 		lun = softc->ctl_luns[err_desc->lun_id];
2792 		if (lun == NULL) {
2793 			mtx_unlock(&softc->ctl_lock);
2794 			printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2795 			       __func__, (uintmax_t)err_desc->lun_id);
2796 			retval = EINVAL;
2797 			break;
2798 		}
2799 
2800 		/*
2801 		 * We could do some checking here to verify the validity
2802 		 * of the request, but given the complexity of error
2803 		 * injection requests, the checking logic would be fairly
2804 		 * complex.
2805 		 *
2806 		 * For now, if the request is invalid, it just won't get
2807 		 * executed and might get deleted.
2808 		 */
2809 		STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2810 
2811 		/*
2812 		 * XXX KDM check to make sure the serial number is unique,
2813 		 * in case we somehow manage to wrap.  That shouldn't
2814 		 * happen for a very long time, but it's the right thing to
2815 		 * do.
2816 		 */
2817 		new_err_desc->serial = lun->error_serial;
2818 		err_desc->serial = lun->error_serial;
2819 		lun->error_serial++;
2820 
2821 		mtx_unlock(&softc->ctl_lock);
2822 		break;
2823 	}
2824 	case CTL_ERROR_INJECT_DELETE: {
2825 		struct ctl_error_desc *delete_desc, *desc, *desc2;
2826 		struct ctl_lun *lun;
2827 		int delete_done;
2828 
2829 		delete_desc = (struct ctl_error_desc *)addr;
2830 		delete_done = 0;
2831 
2832 		mtx_lock(&softc->ctl_lock);
2833 		lun = softc->ctl_luns[delete_desc->lun_id];
2834 		if (lun == NULL) {
2835 			mtx_unlock(&softc->ctl_lock);
2836 			printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2837 			       __func__, (uintmax_t)delete_desc->lun_id);
2838 			retval = EINVAL;
2839 			break;
2840 		}
2841 		STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2842 			if (desc->serial != delete_desc->serial)
2843 				continue;
2844 
2845 			STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2846 				      links);
2847 			free(desc, M_CTL);
2848 			delete_done = 1;
2849 		}
2850 		mtx_unlock(&softc->ctl_lock);
2851 		if (delete_done == 0) {
2852 			printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2853 			       "error serial %ju on LUN %u\n", __func__,
2854 			       delete_desc->serial, delete_desc->lun_id);
2855 			retval = EINVAL;
2856 			break;
2857 		}
2858 		break;
2859 	}
2860 	case CTL_DUMP_STRUCTS: {
2861 		int i, j, k;
2862 		struct ctl_frontend *fe;
2863 
2864 		printf("CTL IID to WWPN map start:\n");
2865 		for (i = 0; i < CTL_MAX_PORTS; i++) {
2866 			for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
2867 				if (softc->wwpn_iid[i][j].in_use == 0)
2868 					continue;
2869 
2870 				printf("port %d iid %u WWPN %#jx\n",
2871 				       softc->wwpn_iid[i][j].port,
2872 				       softc->wwpn_iid[i][j].iid,
2873 				       (uintmax_t)softc->wwpn_iid[i][j].wwpn);
2874 			}
2875 		}
2876 		printf("CTL IID to WWPN map end\n");
2877 		printf("CTL Persistent Reservation information start:\n");
2878 		for (i = 0; i < CTL_MAX_LUNS; i++) {
2879 			struct ctl_lun *lun;
2880 
2881 			lun = softc->ctl_luns[i];
2882 
2883 			if ((lun == NULL)
2884 			 || ((lun->flags & CTL_LUN_DISABLED) != 0))
2885 				continue;
2886 
2887 			for (j = 0; j < (CTL_MAX_PORTS * 2); j++) {
2888 				for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2889 					if (lun->per_res[j+k].registered == 0)
2890 						continue;
2891 					printf("LUN %d port %d iid %d key "
2892 					       "%#jx\n", i, j, k,
2893 					       (uintmax_t)scsi_8btou64(
2894 					       lun->per_res[j+k].res_key.key));
2895 				}
2896 			}
2897 		}
2898 		printf("CTL Persistent Reservation information end\n");
2899 		printf("CTL Frontends:\n");
2900 		/*
2901 		 * XXX KDM calling this without a lock.  We'd likely want
2902 		 * to drop the lock before calling the frontend's dump
2903 		 * routine anyway.
2904 		 */
2905 		STAILQ_FOREACH(fe, &softc->fe_list, links) {
2906 			printf("Frontend %s Type %u pport %d vport %d WWNN "
2907 			       "%#jx WWPN %#jx\n", fe->port_name, fe->port_type,
2908 			       fe->physical_port, fe->virtual_port,
2909 			       (uintmax_t)fe->wwnn, (uintmax_t)fe->wwpn);
2910 
2911 			/*
2912 			 * Frontends are not required to support the dump
2913 			 * routine.
2914 			 */
2915 			if (fe->fe_dump == NULL)
2916 				continue;
2917 
2918 			fe->fe_dump();
2919 		}
2920 		printf("CTL Frontend information end\n");
2921 		break;
2922 	}
2923 	case CTL_LUN_REQ: {
2924 		struct ctl_lun_req *lun_req;
2925 		struct ctl_backend_driver *backend;
2926 
2927 		lun_req = (struct ctl_lun_req *)addr;
2928 
2929 		backend = ctl_backend_find(lun_req->backend);
2930 		if (backend == NULL) {
2931 			lun_req->status = CTL_LUN_ERROR;
2932 			snprintf(lun_req->error_str,
2933 				 sizeof(lun_req->error_str),
2934 				 "Backend \"%s\" not found.",
2935 				 lun_req->backend);
2936 			break;
2937 		}
2938 		if (lun_req->num_be_args > 0) {
2939 			lun_req->kern_be_args = ctl_copyin_args(
2940 				lun_req->num_be_args,
2941 				lun_req->be_args,
2942 				lun_req->error_str,
2943 				sizeof(lun_req->error_str));
2944 			if (lun_req->kern_be_args == NULL) {
2945 				lun_req->status = CTL_LUN_ERROR;
2946 				break;
2947 			}
2948 		}
2949 
2950 		retval = backend->ioctl(dev, cmd, addr, flag, td);
2951 
2952 		if (lun_req->num_be_args > 0) {
2953 			ctl_free_args(lun_req->num_be_args,
2954 				      lun_req->kern_be_args);
2955 		}
2956 		break;
2957 	}
2958 	case CTL_LUN_LIST: {
2959 		struct sbuf *sb;
2960 		struct ctl_lun *lun;
2961 		struct ctl_lun_list *list;
2962 
2963 		list = (struct ctl_lun_list *)addr;
2964 
2965 		/*
2966 		 * Allocate a fixed length sbuf here, based on the length
2967 		 * of the user's buffer.  We could allocate an auto-extending
2968 		 * buffer, and then tell the user how much larger our
2969 		 * amount of data is than his buffer, but that presents
2970 		 * some problems:
2971 		 *
2972 		 * 1.  The sbuf(9) routines use a blocking malloc, and so
2973 		 *     we can't hold a lock while calling them with an
2974 		 *     auto-extending buffer.
2975  		 *
2976 		 * 2.  There is not currently a LUN reference counting
2977 		 *     mechanism, outside of outstanding transactions on
2978 		 *     the LUN's OOA queue.  So a LUN could go away on us
2979 		 *     while we're getting the LUN number, backend-specific
2980 		 *     information, etc.  Thus, given the way things
2981 		 *     currently work, we need to hold the CTL lock while
2982 		 *     grabbing LUN information.
2983 		 *
2984 		 * So, from the user's standpoint, the best thing to do is
2985 		 * allocate what he thinks is a reasonable buffer length,
2986 		 * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
2987 		 * double the buffer length and try again.  (And repeat
2988 		 * that until he succeeds.)
2989 		 */
2990 		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
2991 		if (sb == NULL) {
2992 			list->status = CTL_LUN_LIST_ERROR;
2993 			snprintf(list->error_str, sizeof(list->error_str),
2994 				 "Unable to allocate %d bytes for LUN list",
2995 				 list->alloc_len);
2996 			break;
2997 		}
2998 
2999 		sbuf_printf(sb, "<ctllunlist>\n");
3000 
3001 		mtx_lock(&softc->ctl_lock);
3002 
3003 		STAILQ_FOREACH(lun, &softc->lun_list, links) {
3004 			retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3005 					     (uintmax_t)lun->lun);
3006 
3007 			/*
3008 			 * Bail out as soon as we see that we've overfilled
3009 			 * the buffer.
3010 			 */
3011 			if (retval != 0)
3012 				break;
3013 
3014 			retval = sbuf_printf(sb, "<backend_type>%s"
3015 					     "</backend_type>\n",
3016 					     (lun->backend == NULL) ?  "none" :
3017 					     lun->backend->name);
3018 
3019 			if (retval != 0)
3020 				break;
3021 
3022 			retval = sbuf_printf(sb, "<lun_type>%d</lun_type>\n",
3023 					     lun->be_lun->lun_type);
3024 
3025 			if (retval != 0)
3026 				break;
3027 
3028 			if (lun->backend == NULL) {
3029 				retval = sbuf_printf(sb, "</lun>\n");
3030 				if (retval != 0)
3031 					break;
3032 				continue;
3033 			}
3034 
3035 			retval = sbuf_printf(sb, "<size>%ju</size>\n",
3036 					     (lun->be_lun->maxlba > 0) ?
3037 					     lun->be_lun->maxlba + 1 : 0);
3038 
3039 			if (retval != 0)
3040 				break;
3041 
3042 			retval = sbuf_printf(sb, "<blocksize>%u</blocksize>\n",
3043 					     lun->be_lun->blocksize);
3044 
3045 			if (retval != 0)
3046 				break;
3047 
3048 			retval = sbuf_printf(sb, "<serial_number>");
3049 
3050 			if (retval != 0)
3051 				break;
3052 
3053 			retval = ctl_sbuf_printf_esc(sb,
3054 						     lun->be_lun->serial_num);
3055 
3056 			if (retval != 0)
3057 				break;
3058 
3059 			retval = sbuf_printf(sb, "</serial_number>\n");
3060 
3061 			if (retval != 0)
3062 				break;
3063 
3064 			retval = sbuf_printf(sb, "<device_id>");
3065 
3066 			if (retval != 0)
3067 				break;
3068 
3069 			retval = ctl_sbuf_printf_esc(sb,lun->be_lun->device_id);
3070 
3071 			if (retval != 0)
3072 				break;
3073 
3074 			retval = sbuf_printf(sb, "</device_id>\n");
3075 
3076 			if (retval != 0)
3077 				break;
3078 
3079 			if (lun->backend->lun_info == NULL) {
3080 				retval = sbuf_printf(sb, "</lun>\n");
3081 				if (retval != 0)
3082 					break;
3083 				continue;
3084 			}
3085 
3086 			retval =lun->backend->lun_info(lun->be_lun->be_lun, sb);
3087 
3088 			if (retval != 0)
3089 				break;
3090 
3091 			retval = sbuf_printf(sb, "</lun>\n");
3092 
3093 			if (retval != 0)
3094 				break;
3095 		}
3096 		mtx_unlock(&softc->ctl_lock);
3097 
3098 		if ((retval != 0)
3099 		 || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3100 			retval = 0;
3101 			sbuf_delete(sb);
3102 			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3103 			snprintf(list->error_str, sizeof(list->error_str),
3104 				 "Out of space, %d bytes is too small",
3105 				 list->alloc_len);
3106 			break;
3107 		}
3108 
3109 		sbuf_finish(sb);
3110 
3111 		retval = copyout(sbuf_data(sb), list->lun_xml,
3112 				 sbuf_len(sb) + 1);
3113 
3114 		list->fill_len = sbuf_len(sb) + 1;
3115 		list->status = CTL_LUN_LIST_OK;
3116 		sbuf_delete(sb);
3117 		break;
3118 	}
3119 	default: {
3120 		/* XXX KDM should we fix this? */
3121 #if 0
3122 		struct ctl_backend_driver *backend;
3123 		unsigned int type;
3124 		int found;
3125 
3126 		found = 0;
3127 
3128 		/*
3129 		 * We encode the backend type as the ioctl type for backend
3130 		 * ioctls.  So parse it out here, and then search for a
3131 		 * backend of this type.
3132 		 */
3133 		type = _IOC_TYPE(cmd);
3134 
3135 		STAILQ_FOREACH(backend, &softc->be_list, links) {
3136 			if (backend->type == type) {
3137 				found = 1;
3138 				break;
3139 			}
3140 		}
3141 		if (found == 0) {
3142 			printf("ctl: unknown ioctl command %#lx or backend "
3143 			       "%d\n", cmd, type);
3144 			retval = -EINVAL;
3145 			break;
3146 		}
3147 		retval = backend->ioctl(dev, cmd, addr, flag, td);
3148 #endif
3149 		retval = ENOTTY;
3150 		break;
3151 	}
3152 	}
3153 	return (retval);
3154 }
3155 
3156 uint32_t
3157 ctl_get_initindex(struct ctl_nexus *nexus)
3158 {
3159 	if (nexus->targ_port < CTL_MAX_PORTS)
3160 		return (nexus->initid.id +
3161 			(nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3162 	else
3163 		return (nexus->initid.id +
3164 		       ((nexus->targ_port - CTL_MAX_PORTS) *
3165 			CTL_MAX_INIT_PER_PORT));
3166 }
3167 
3168 uint32_t
3169 ctl_get_resindex(struct ctl_nexus *nexus)
3170 {
3171 	return (nexus->initid.id + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3172 }
3173 
3174 uint32_t
3175 ctl_port_idx(int port_num)
3176 {
3177 	if (port_num < CTL_MAX_PORTS)
3178 		return(port_num);
3179 	else
3180 		return(port_num - CTL_MAX_PORTS);
3181 }
3182 
3183 /*
3184  * Note:  This only works for bitmask sizes that are at least 32 bits, and
3185  * that are a power of 2.
3186  */
3187 int
3188 ctl_ffz(uint32_t *mask, uint32_t size)
3189 {
3190 	uint32_t num_chunks, num_pieces;
3191 	int i, j;
3192 
3193 	num_chunks = (size >> 5);
3194 	if (num_chunks == 0)
3195 		num_chunks++;
3196 	num_pieces = ctl_min((sizeof(uint32_t) * 8), size);
3197 
3198 	for (i = 0; i < num_chunks; i++) {
3199 		for (j = 0; j < num_pieces; j++) {
3200 			if ((mask[i] & (1 << j)) == 0)
3201 				return ((i << 5) + j);
3202 		}
3203 	}
3204 
3205 	return (-1);
3206 }
3207 
3208 int
3209 ctl_set_mask(uint32_t *mask, uint32_t bit)
3210 {
3211 	uint32_t chunk, piece;
3212 
3213 	chunk = bit >> 5;
3214 	piece = bit % (sizeof(uint32_t) * 8);
3215 
3216 	if ((mask[chunk] & (1 << piece)) != 0)
3217 		return (-1);
3218 	else
3219 		mask[chunk] |= (1 << piece);
3220 
3221 	return (0);
3222 }
3223 
3224 int
3225 ctl_clear_mask(uint32_t *mask, uint32_t bit)
3226 {
3227 	uint32_t chunk, piece;
3228 
3229 	chunk = bit >> 5;
3230 	piece = bit % (sizeof(uint32_t) * 8);
3231 
3232 	if ((mask[chunk] & (1 << piece)) == 0)
3233 		return (-1);
3234 	else
3235 		mask[chunk] &= ~(1 << piece);
3236 
3237 	return (0);
3238 }
3239 
3240 int
3241 ctl_is_set(uint32_t *mask, uint32_t bit)
3242 {
3243 	uint32_t chunk, piece;
3244 
3245 	chunk = bit >> 5;
3246 	piece = bit % (sizeof(uint32_t) * 8);
3247 
3248 	if ((mask[chunk] & (1 << piece)) == 0)
3249 		return (0);
3250 	else
3251 		return (1);
3252 }
3253 
3254 #ifdef unused
3255 /*
3256  * The bus, target and lun are optional, they can be filled in later.
3257  * can_wait is used to determine whether we can wait on the malloc or not.
3258  */
3259 union ctl_io*
3260 ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port, uint32_t targ_target,
3261 	      uint32_t targ_lun, int can_wait)
3262 {
3263 	union ctl_io *io;
3264 
3265 	if (can_wait)
3266 		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_WAITOK);
3267 	else
3268 		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_NOWAIT);
3269 
3270 	if (io != NULL) {
3271 		io->io_hdr.io_type = io_type;
3272 		io->io_hdr.targ_port = targ_port;
3273 		/*
3274 		 * XXX KDM this needs to change/go away.  We need to move
3275 		 * to a preallocated pool of ctl_scsiio structures.
3276 		 */
3277 		io->io_hdr.nexus.targ_target.id = targ_target;
3278 		io->io_hdr.nexus.targ_lun = targ_lun;
3279 	}
3280 
3281 	return (io);
3282 }
3283 
3284 void
3285 ctl_kfree_io(union ctl_io *io)
3286 {
3287 	free(io, M_CTL);
3288 }
3289 #endif /* unused */
3290 
3291 /*
3292  * ctl_softc, pool_type, total_ctl_io are passed in.
3293  * npool is passed out.
3294  */
3295 int
3296 ctl_pool_create(struct ctl_softc *ctl_softc, ctl_pool_type pool_type,
3297 		uint32_t total_ctl_io, struct ctl_io_pool **npool)
3298 {
3299 	uint32_t i;
3300 	union ctl_io *cur_io, *next_io;
3301 	struct ctl_io_pool *pool;
3302 	int retval;
3303 
3304 	retval = 0;
3305 
3306 	pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3307 					    M_NOWAIT | M_ZERO);
3308 	if (pool == NULL) {
3309 		retval = -ENOMEM;
3310 		goto bailout;
3311 	}
3312 
3313 	pool->type = pool_type;
3314 	pool->ctl_softc = ctl_softc;
3315 
3316 	mtx_lock(&ctl_softc->ctl_lock);
3317 	pool->id = ctl_softc->cur_pool_id++;
3318 	mtx_unlock(&ctl_softc->ctl_lock);
3319 
3320 	pool->flags = CTL_POOL_FLAG_NONE;
3321 	STAILQ_INIT(&pool->free_queue);
3322 
3323 	/*
3324 	 * XXX KDM other options here:
3325 	 * - allocate a page at a time
3326 	 * - allocate one big chunk of memory.
3327 	 * Page allocation might work well, but would take a little more
3328 	 * tracking.
3329 	 */
3330 	for (i = 0; i < total_ctl_io; i++) {
3331 		cur_io = (union ctl_io *)malloc(sizeof(*cur_io), M_CTL,
3332 						M_NOWAIT);
3333 		if (cur_io == NULL) {
3334 			retval = ENOMEM;
3335 			break;
3336 		}
3337 		cur_io->io_hdr.pool = pool;
3338 		STAILQ_INSERT_TAIL(&pool->free_queue, &cur_io->io_hdr, links);
3339 		pool->total_ctl_io++;
3340 		pool->free_ctl_io++;
3341 	}
3342 
3343 	if (retval != 0) {
3344 		for (cur_io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3345 		     cur_io != NULL; cur_io = next_io) {
3346 			next_io = (union ctl_io *)STAILQ_NEXT(&cur_io->io_hdr,
3347 							      links);
3348 			STAILQ_REMOVE(&pool->free_queue, &cur_io->io_hdr,
3349 				      ctl_io_hdr, links);
3350 			free(cur_io, M_CTL);
3351 		}
3352 
3353 		free(pool, M_CTL);
3354 		goto bailout;
3355 	}
3356 	mtx_lock(&ctl_softc->ctl_lock);
3357 	ctl_softc->num_pools++;
3358 	STAILQ_INSERT_TAIL(&ctl_softc->io_pools, pool, links);
3359 	/*
3360 	 * Increment our usage count if this is an external consumer, so we
3361 	 * can't get unloaded until the external consumer (most likely a
3362 	 * FETD) unloads and frees his pool.
3363 	 *
3364 	 * XXX KDM will this increment the caller's module use count, or
3365 	 * mine?
3366 	 */
3367 #if 0
3368 	if ((pool_type != CTL_POOL_EMERGENCY)
3369 	 && (pool_type != CTL_POOL_INTERNAL)
3370 	 && (pool_type != CTL_POOL_IOCTL)
3371 	 && (pool_type != CTL_POOL_4OTHERSC))
3372 		MOD_INC_USE_COUNT;
3373 #endif
3374 
3375 	mtx_unlock(&ctl_softc->ctl_lock);
3376 
3377 	*npool = pool;
3378 
3379 bailout:
3380 
3381 	return (retval);
3382 }
3383 
3384 /*
3385  * Caller must hold ctl_softc->ctl_lock.
3386  */
3387 int
3388 ctl_pool_acquire(struct ctl_io_pool *pool)
3389 {
3390 	if (pool == NULL)
3391 		return (-EINVAL);
3392 
3393 	if (pool->flags & CTL_POOL_FLAG_INVALID)
3394 		return (-EINVAL);
3395 
3396 	pool->refcount++;
3397 
3398 	return (0);
3399 }
3400 
3401 /*
3402  * Caller must hold ctl_softc->ctl_lock.
3403  */
3404 int
3405 ctl_pool_invalidate(struct ctl_io_pool *pool)
3406 {
3407 	if (pool == NULL)
3408 		return (-EINVAL);
3409 
3410 	pool->flags |= CTL_POOL_FLAG_INVALID;
3411 
3412 	return (0);
3413 }
3414 
3415 /*
3416  * Caller must hold ctl_softc->ctl_lock.
3417  */
3418 int
3419 ctl_pool_release(struct ctl_io_pool *pool)
3420 {
3421 	if (pool == NULL)
3422 		return (-EINVAL);
3423 
3424 	if ((--pool->refcount == 0)
3425 	 && (pool->flags & CTL_POOL_FLAG_INVALID)) {
3426 		ctl_pool_free(pool->ctl_softc, pool);
3427 	}
3428 
3429 	return (0);
3430 }
3431 
3432 /*
3433  * Must be called with ctl_softc->ctl_lock held.
3434  */
3435 void
3436 ctl_pool_free(struct ctl_softc *ctl_softc, struct ctl_io_pool *pool)
3437 {
3438 	union ctl_io *cur_io, *next_io;
3439 
3440 	for (cur_io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3441 	     cur_io != NULL; cur_io = next_io) {
3442 		next_io = (union ctl_io *)STAILQ_NEXT(&cur_io->io_hdr,
3443 						      links);
3444 		STAILQ_REMOVE(&pool->free_queue, &cur_io->io_hdr, ctl_io_hdr,
3445 			      links);
3446 		free(cur_io, M_CTL);
3447 	}
3448 
3449 	STAILQ_REMOVE(&ctl_softc->io_pools, pool, ctl_io_pool, links);
3450 	ctl_softc->num_pools--;
3451 
3452 	/*
3453 	 * XXX KDM will this decrement the caller's usage count or mine?
3454 	 */
3455 #if 0
3456 	if ((pool->type != CTL_POOL_EMERGENCY)
3457 	 && (pool->type != CTL_POOL_INTERNAL)
3458 	 && (pool->type != CTL_POOL_IOCTL))
3459 		MOD_DEC_USE_COUNT;
3460 #endif
3461 
3462 	free(pool, M_CTL);
3463 }
3464 
3465 /*
3466  * This routine does not block (except for spinlocks of course).
3467  * It tries to allocate a ctl_io union from the caller's pool as quickly as
3468  * possible.
3469  */
3470 union ctl_io *
3471 ctl_alloc_io(void *pool_ref)
3472 {
3473 	union ctl_io *io;
3474 	struct ctl_softc *ctl_softc;
3475 	struct ctl_io_pool *pool, *npool;
3476 	struct ctl_io_pool *emergency_pool;
3477 
3478 	pool = (struct ctl_io_pool *)pool_ref;
3479 
3480 	if (pool == NULL) {
3481 		printf("%s: pool is NULL\n", __func__);
3482 		return (NULL);
3483 	}
3484 
3485 	emergency_pool = NULL;
3486 
3487 	ctl_softc = pool->ctl_softc;
3488 
3489 	mtx_lock(&ctl_softc->ctl_lock);
3490 	/*
3491 	 * First, try to get the io structure from the user's pool.
3492 	 */
3493 	if (ctl_pool_acquire(pool) == 0) {
3494 		io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3495 		if (io != NULL) {
3496 			STAILQ_REMOVE_HEAD(&pool->free_queue, links);
3497 			pool->total_allocated++;
3498 			pool->free_ctl_io--;
3499 			mtx_unlock(&ctl_softc->ctl_lock);
3500 			return (io);
3501 		} else
3502 			ctl_pool_release(pool);
3503 	}
3504 	/*
3505 	 * If he doesn't have any io structures left, search for an
3506 	 * emergency pool and grab one from there.
3507 	 */
3508 	STAILQ_FOREACH(npool, &ctl_softc->io_pools, links) {
3509 		if (npool->type != CTL_POOL_EMERGENCY)
3510 			continue;
3511 
3512 		if (ctl_pool_acquire(npool) != 0)
3513 			continue;
3514 
3515 		emergency_pool = npool;
3516 
3517 		io = (union ctl_io *)STAILQ_FIRST(&npool->free_queue);
3518 		if (io != NULL) {
3519 			STAILQ_REMOVE_HEAD(&npool->free_queue, links);
3520 			npool->total_allocated++;
3521 			npool->free_ctl_io--;
3522 			mtx_unlock(&ctl_softc->ctl_lock);
3523 			return (io);
3524 		} else
3525 			ctl_pool_release(npool);
3526 	}
3527 
3528 	/* Drop the spinlock before we malloc */
3529 	mtx_unlock(&ctl_softc->ctl_lock);
3530 
3531 	/*
3532 	 * The emergency pool (if it exists) didn't have one, so try an
3533 	 * atomic (i.e. nonblocking) malloc and see if we get lucky.
3534 	 */
3535 	io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_NOWAIT);
3536 	if (io != NULL) {
3537 		/*
3538 		 * If the emergency pool exists but is empty, add this
3539 		 * ctl_io to its list when it gets freed.
3540 		 */
3541 		if (emergency_pool != NULL) {
3542 			mtx_lock(&ctl_softc->ctl_lock);
3543 			if (ctl_pool_acquire(emergency_pool) == 0) {
3544 				io->io_hdr.pool = emergency_pool;
3545 				emergency_pool->total_ctl_io++;
3546 				/*
3547 				 * Need to bump this, otherwise
3548 				 * total_allocated and total_freed won't
3549 				 * match when we no longer have anything
3550 				 * outstanding.
3551 				 */
3552 				emergency_pool->total_allocated++;
3553 			}
3554 			mtx_unlock(&ctl_softc->ctl_lock);
3555 		} else
3556 			io->io_hdr.pool = NULL;
3557 	}
3558 
3559 	return (io);
3560 }
3561 
3562 static void
3563 ctl_free_io_internal(union ctl_io *io, int have_lock)
3564 {
3565 	if (io == NULL)
3566 		return;
3567 
3568 	/*
3569 	 * If this ctl_io has a pool, return it to that pool.
3570 	 */
3571 	if (io->io_hdr.pool != NULL) {
3572 		struct ctl_io_pool *pool;
3573 #if 0
3574 		struct ctl_softc *ctl_softc;
3575 		union ctl_io *tmp_io;
3576 		unsigned long xflags;
3577 		int i;
3578 
3579 		ctl_softc = control_softc;
3580 #endif
3581 
3582 		pool = (struct ctl_io_pool *)io->io_hdr.pool;
3583 
3584 		if (have_lock == 0)
3585 			mtx_lock(&pool->ctl_softc->ctl_lock);
3586 #if 0
3587 		save_flags(xflags);
3588 
3589 		for (i = 0, tmp_io = (union ctl_io *)STAILQ_FIRST(
3590 		     &ctl_softc->task_queue); tmp_io != NULL; i++,
3591 		     tmp_io = (union ctl_io *)STAILQ_NEXT(&tmp_io->io_hdr,
3592 		     links)) {
3593 			if (tmp_io == io) {
3594 				printf("%s: %p is still on the task queue!\n",
3595 				       __func__, tmp_io);
3596 				printf("%s: (%d): type %d "
3597 				       "msg %d cdb %x iptl: "
3598 				       "%d:%d:%d:%d tag 0x%04x "
3599 				       "flg %#lx\n",
3600 					__func__, i,
3601 					tmp_io->io_hdr.io_type,
3602 					tmp_io->io_hdr.msg_type,
3603 					tmp_io->scsiio.cdb[0],
3604 					tmp_io->io_hdr.nexus.initid.id,
3605 					tmp_io->io_hdr.nexus.targ_port,
3606 					tmp_io->io_hdr.nexus.targ_target.id,
3607 					tmp_io->io_hdr.nexus.targ_lun,
3608 					(tmp_io->io_hdr.io_type ==
3609 					CTL_IO_TASK) ?
3610 					tmp_io->taskio.tag_num :
3611 					tmp_io->scsiio.tag_num,
3612 					xflags);
3613 				panic("I/O still on the task queue!");
3614 			}
3615 		}
3616 #endif
3617 		io->io_hdr.io_type = 0xff;
3618 		STAILQ_INSERT_TAIL(&pool->free_queue, &io->io_hdr, links);
3619 		pool->total_freed++;
3620 		pool->free_ctl_io++;
3621 		ctl_pool_release(pool);
3622 		if (have_lock == 0)
3623 			mtx_unlock(&pool->ctl_softc->ctl_lock);
3624 	} else {
3625 		/*
3626 		 * Otherwise, just free it.  We probably malloced it and
3627 		 * the emergency pool wasn't available.
3628 		 */
3629 		free(io, M_CTL);
3630 	}
3631 
3632 }
3633 
3634 void
3635 ctl_free_io(union ctl_io *io)
3636 {
3637 	ctl_free_io_internal(io, /*have_lock*/ 0);
3638 }
3639 
3640 void
3641 ctl_zero_io(union ctl_io *io)
3642 {
3643 	void *pool_ref;
3644 
3645 	if (io == NULL)
3646 		return;
3647 
3648 	/*
3649 	 * May need to preserve linked list pointers at some point too.
3650 	 */
3651 	pool_ref = io->io_hdr.pool;
3652 
3653 	memset(io, 0, sizeof(*io));
3654 
3655 	io->io_hdr.pool = pool_ref;
3656 }
3657 
3658 /*
3659  * This routine is currently used for internal copies of ctl_ios that need
3660  * to persist for some reason after we've already returned status to the
3661  * FETD.  (Thus the flag set.)
3662  *
3663  * XXX XXX
3664  * Note that this makes a blind copy of all fields in the ctl_io, except
3665  * for the pool reference.  This includes any memory that has been
3666  * allocated!  That memory will no longer be valid after done has been
3667  * called, so this would be VERY DANGEROUS for command that actually does
3668  * any reads or writes.  Right now (11/7/2005), this is only used for immediate
3669  * start and stop commands, which don't transfer any data, so this is not a
3670  * problem.  If it is used for anything else, the caller would also need to
3671  * allocate data buffer space and this routine would need to be modified to
3672  * copy the data buffer(s) as well.
3673  */
3674 void
3675 ctl_copy_io(union ctl_io *src, union ctl_io *dest)
3676 {
3677 	void *pool_ref;
3678 
3679 	if ((src == NULL)
3680 	 || (dest == NULL))
3681 		return;
3682 
3683 	/*
3684 	 * May need to preserve linked list pointers at some point too.
3685 	 */
3686 	pool_ref = dest->io_hdr.pool;
3687 
3688 	memcpy(dest, src, ctl_min(sizeof(*src), sizeof(*dest)));
3689 
3690 	dest->io_hdr.pool = pool_ref;
3691 	/*
3692 	 * We need to know that this is an internal copy, and doesn't need
3693 	 * to get passed back to the FETD that allocated it.
3694 	 */
3695 	dest->io_hdr.flags |= CTL_FLAG_INT_COPY;
3696 }
3697 
3698 #ifdef NEEDTOPORT
3699 static void
3700 ctl_update_power_subpage(struct copan_power_subpage *page)
3701 {
3702 	int num_luns, num_partitions, config_type;
3703 	struct ctl_softc *softc;
3704 	cs_BOOL_t aor_present, shelf_50pct_power;
3705 	cs_raidset_personality_t rs_type;
3706 	int max_active_luns;
3707 
3708 	softc = control_softc;
3709 
3710 	/* subtract out the processor LUN */
3711 	num_luns = softc->num_luns - 1;
3712 	/*
3713 	 * Default to 7 LUNs active, which was the only number we allowed
3714 	 * in the past.
3715 	 */
3716 	max_active_luns = 7;
3717 
3718 	num_partitions = config_GetRsPartitionInfo();
3719 	config_type = config_GetConfigType();
3720 	shelf_50pct_power = config_GetShelfPowerMode();
3721 	aor_present = config_IsAorRsPresent();
3722 
3723 	rs_type = ddb_GetRsRaidType(1);
3724 	if ((rs_type != CS_RAIDSET_PERSONALITY_RAID5)
3725 	 && (rs_type != CS_RAIDSET_PERSONALITY_RAID1)) {
3726 		EPRINT(0, "Unsupported RS type %d!", rs_type);
3727 	}
3728 
3729 
3730 	page->total_luns = num_luns;
3731 
3732 	switch (config_type) {
3733 	case 40:
3734 		/*
3735 		 * In a 40 drive configuration, it doesn't matter what DC
3736 		 * cards we have, whether we have AOR enabled or not,
3737 		 * partitioning or not, or what type of RAIDset we have.
3738 		 * In that scenario, we can power up every LUN we present
3739 		 * to the user.
3740 		 */
3741 		max_active_luns = num_luns;
3742 
3743 		break;
3744 	case 64:
3745 		if (shelf_50pct_power == CS_FALSE) {
3746 			/* 25% power */
3747 			if (aor_present == CS_TRUE) {
3748 				if (rs_type ==
3749 				     CS_RAIDSET_PERSONALITY_RAID5) {
3750 					max_active_luns = 7;
3751 				} else if (rs_type ==
3752 					 CS_RAIDSET_PERSONALITY_RAID1){
3753 					max_active_luns = 14;
3754 				} else {
3755 					/* XXX KDM now what?? */
3756 				}
3757 			} else {
3758 				if (rs_type ==
3759 				     CS_RAIDSET_PERSONALITY_RAID5) {
3760 					max_active_luns = 8;
3761 				} else if (rs_type ==
3762 					 CS_RAIDSET_PERSONALITY_RAID1){
3763 					max_active_luns = 16;
3764 				} else {
3765 					/* XXX KDM now what?? */
3766 				}
3767 			}
3768 		} else {
3769 			/* 50% power */
3770 			/*
3771 			 * With 50% power in a 64 drive configuration, we
3772 			 * can power all LUNs we present.
3773 			 */
3774 			max_active_luns = num_luns;
3775 		}
3776 		break;
3777 	case 112:
3778 		if (shelf_50pct_power == CS_FALSE) {
3779 			/* 25% power */
3780 			if (aor_present == CS_TRUE) {
3781 				if (rs_type ==
3782 				     CS_RAIDSET_PERSONALITY_RAID5) {
3783 					max_active_luns = 7;
3784 				} else if (rs_type ==
3785 					 CS_RAIDSET_PERSONALITY_RAID1){
3786 					max_active_luns = 14;
3787 				} else {
3788 					/* XXX KDM now what?? */
3789 				}
3790 			} else {
3791 				if (rs_type ==
3792 				     CS_RAIDSET_PERSONALITY_RAID5) {
3793 					max_active_luns = 8;
3794 				} else if (rs_type ==
3795 					 CS_RAIDSET_PERSONALITY_RAID1){
3796 					max_active_luns = 16;
3797 				} else {
3798 					/* XXX KDM now what?? */
3799 				}
3800 			}
3801 		} else {
3802 			/* 50% power */
3803 			if (aor_present == CS_TRUE) {
3804 				if (rs_type ==
3805 				     CS_RAIDSET_PERSONALITY_RAID5) {
3806 					max_active_luns = 14;
3807 				} else if (rs_type ==
3808 					 CS_RAIDSET_PERSONALITY_RAID1){
3809 					/*
3810 					 * We're assuming here that disk
3811 					 * caching is enabled, and so we're
3812 					 * able to power up half of each
3813 					 * LUN, and cache all writes.
3814 					 */
3815 					max_active_luns = num_luns;
3816 				} else {
3817 					/* XXX KDM now what?? */
3818 				}
3819 			} else {
3820 				if (rs_type ==
3821 				     CS_RAIDSET_PERSONALITY_RAID5) {
3822 					max_active_luns = 15;
3823 				} else if (rs_type ==
3824 					 CS_RAIDSET_PERSONALITY_RAID1){
3825 					max_active_luns = 30;
3826 				} else {
3827 					/* XXX KDM now what?? */
3828 				}
3829 			}
3830 		}
3831 		break;
3832 	default:
3833 		/*
3834 		 * In this case, we have an unknown configuration, so we
3835 		 * just use the default from above.
3836 		 */
3837 		break;
3838 	}
3839 
3840 	page->max_active_luns = max_active_luns;
3841 #if 0
3842 	printk("%s: total_luns = %d, max_active_luns = %d\n", __func__,
3843 	       page->total_luns, page->max_active_luns);
3844 #endif
3845 }
3846 #endif /* NEEDTOPORT */
3847 
3848 /*
3849  * This routine could be used in the future to load default and/or saved
3850  * mode page parameters for a particuar lun.
3851  */
3852 static int
3853 ctl_init_page_index(struct ctl_lun *lun)
3854 {
3855 	int i;
3856 	struct ctl_page_index *page_index;
3857 	struct ctl_softc *softc;
3858 
3859 	memcpy(&lun->mode_pages.index, page_index_template,
3860 	       sizeof(page_index_template));
3861 
3862 	softc = lun->ctl_softc;
3863 
3864 	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
3865 
3866 		page_index = &lun->mode_pages.index[i];
3867 		/*
3868 		 * If this is a disk-only mode page, there's no point in
3869 		 * setting it up.  For some pages, we have to have some
3870 		 * basic information about the disk in order to calculate the
3871 		 * mode page data.
3872 		 */
3873 		if ((lun->be_lun->lun_type != T_DIRECT)
3874 		 && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
3875 			continue;
3876 
3877 		switch (page_index->page_code & SMPH_PC_MASK) {
3878 		case SMS_FORMAT_DEVICE_PAGE: {
3879 			struct scsi_format_page *format_page;
3880 
3881 			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3882 				panic("subpage is incorrect!");
3883 
3884 			/*
3885 			 * Sectors per track are set above.  Bytes per
3886 			 * sector need to be set here on a per-LUN basis.
3887 			 */
3888 			memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
3889 			       &format_page_default,
3890 			       sizeof(format_page_default));
3891 			memcpy(&lun->mode_pages.format_page[
3892 			       CTL_PAGE_CHANGEABLE], &format_page_changeable,
3893 			       sizeof(format_page_changeable));
3894 			memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
3895 			       &format_page_default,
3896 			       sizeof(format_page_default));
3897 			memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
3898 			       &format_page_default,
3899 			       sizeof(format_page_default));
3900 
3901 			format_page = &lun->mode_pages.format_page[
3902 				CTL_PAGE_CURRENT];
3903 			scsi_ulto2b(lun->be_lun->blocksize,
3904 				    format_page->bytes_per_sector);
3905 
3906 			format_page = &lun->mode_pages.format_page[
3907 				CTL_PAGE_DEFAULT];
3908 			scsi_ulto2b(lun->be_lun->blocksize,
3909 				    format_page->bytes_per_sector);
3910 
3911 			format_page = &lun->mode_pages.format_page[
3912 				CTL_PAGE_SAVED];
3913 			scsi_ulto2b(lun->be_lun->blocksize,
3914 				    format_page->bytes_per_sector);
3915 
3916 			page_index->page_data =
3917 				(uint8_t *)lun->mode_pages.format_page;
3918 			break;
3919 		}
3920 		case SMS_RIGID_DISK_PAGE: {
3921 			struct scsi_rigid_disk_page *rigid_disk_page;
3922 			uint32_t sectors_per_cylinder;
3923 			uint64_t cylinders;
3924 #ifndef	__XSCALE__
3925 			int shift;
3926 #endif /* !__XSCALE__ */
3927 
3928 			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3929 				panic("invalid subpage value %d",
3930 				      page_index->subpage);
3931 
3932 			/*
3933 			 * Rotation rate and sectors per track are set
3934 			 * above.  We calculate the cylinders here based on
3935 			 * capacity.  Due to the number of heads and
3936 			 * sectors per track we're using, smaller arrays
3937 			 * may turn out to have 0 cylinders.  Linux and
3938 			 * FreeBSD don't pay attention to these mode pages
3939 			 * to figure out capacity, but Solaris does.  It
3940 			 * seems to deal with 0 cylinders just fine, and
3941 			 * works out a fake geometry based on the capacity.
3942 			 */
3943 			memcpy(&lun->mode_pages.rigid_disk_page[
3944 			       CTL_PAGE_CURRENT], &rigid_disk_page_default,
3945 			       sizeof(rigid_disk_page_default));
3946 			memcpy(&lun->mode_pages.rigid_disk_page[
3947 			       CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
3948 			       sizeof(rigid_disk_page_changeable));
3949 			memcpy(&lun->mode_pages.rigid_disk_page[
3950 			       CTL_PAGE_DEFAULT], &rigid_disk_page_default,
3951 			       sizeof(rigid_disk_page_default));
3952 			memcpy(&lun->mode_pages.rigid_disk_page[
3953 			       CTL_PAGE_SAVED], &rigid_disk_page_default,
3954 			       sizeof(rigid_disk_page_default));
3955 
3956 			sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
3957 				CTL_DEFAULT_HEADS;
3958 
3959 			/*
3960 			 * The divide method here will be more accurate,
3961 			 * probably, but results in floating point being
3962 			 * used in the kernel on i386 (__udivdi3()).  On the
3963 			 * XScale, though, __udivdi3() is implemented in
3964 			 * software.
3965 			 *
3966 			 * The shift method for cylinder calculation is
3967 			 * accurate if sectors_per_cylinder is a power of
3968 			 * 2.  Otherwise it might be slightly off -- you
3969 			 * might have a bit of a truncation problem.
3970 			 */
3971 #ifdef	__XSCALE__
3972 			cylinders = (lun->be_lun->maxlba + 1) /
3973 				sectors_per_cylinder;
3974 #else
3975 			for (shift = 31; shift > 0; shift--) {
3976 				if (sectors_per_cylinder & (1 << shift))
3977 					break;
3978 			}
3979 			cylinders = (lun->be_lun->maxlba + 1) >> shift;
3980 #endif
3981 
3982 			/*
3983 			 * We've basically got 3 bytes, or 24 bits for the
3984 			 * cylinder size in the mode page.  If we're over,
3985 			 * just round down to 2^24.
3986 			 */
3987 			if (cylinders > 0xffffff)
3988 				cylinders = 0xffffff;
3989 
3990 			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
3991 				CTL_PAGE_CURRENT];
3992 			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
3993 
3994 			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
3995 				CTL_PAGE_DEFAULT];
3996 			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
3997 
3998 			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
3999 				CTL_PAGE_SAVED];
4000 			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4001 
4002 			page_index->page_data =
4003 				(uint8_t *)lun->mode_pages.rigid_disk_page;
4004 			break;
4005 		}
4006 		case SMS_CACHING_PAGE: {
4007 
4008 			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4009 				panic("invalid subpage value %d",
4010 				      page_index->subpage);
4011 			/*
4012 			 * Defaults should be okay here, no calculations
4013 			 * needed.
4014 			 */
4015 			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4016 			       &caching_page_default,
4017 			       sizeof(caching_page_default));
4018 			memcpy(&lun->mode_pages.caching_page[
4019 			       CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4020 			       sizeof(caching_page_changeable));
4021 			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4022 			       &caching_page_default,
4023 			       sizeof(caching_page_default));
4024 			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4025 			       &caching_page_default,
4026 			       sizeof(caching_page_default));
4027 			page_index->page_data =
4028 				(uint8_t *)lun->mode_pages.caching_page;
4029 			break;
4030 		}
4031 		case SMS_CONTROL_MODE_PAGE: {
4032 
4033 			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4034 				panic("invalid subpage value %d",
4035 				      page_index->subpage);
4036 
4037 			/*
4038 			 * Defaults should be okay here, no calculations
4039 			 * needed.
4040 			 */
4041 			memcpy(&lun->mode_pages.control_page[CTL_PAGE_CURRENT],
4042 			       &control_page_default,
4043 			       sizeof(control_page_default));
4044 			memcpy(&lun->mode_pages.control_page[
4045 			       CTL_PAGE_CHANGEABLE], &control_page_changeable,
4046 			       sizeof(control_page_changeable));
4047 			memcpy(&lun->mode_pages.control_page[CTL_PAGE_DEFAULT],
4048 			       &control_page_default,
4049 			       sizeof(control_page_default));
4050 			memcpy(&lun->mode_pages.control_page[CTL_PAGE_SAVED],
4051 			       &control_page_default,
4052 			       sizeof(control_page_default));
4053 			page_index->page_data =
4054 				(uint8_t *)lun->mode_pages.control_page;
4055 			break;
4056 
4057 		}
4058 		case SMS_VENDOR_SPECIFIC_PAGE:{
4059 			switch (page_index->subpage) {
4060 			case PWR_SUBPAGE_CODE: {
4061 				struct copan_power_subpage *current_page,
4062 							   *saved_page;
4063 
4064 				memcpy(&lun->mode_pages.power_subpage[
4065 				       CTL_PAGE_CURRENT],
4066 				       &power_page_default,
4067 				       sizeof(power_page_default));
4068 				memcpy(&lun->mode_pages.power_subpage[
4069 				       CTL_PAGE_CHANGEABLE],
4070 				       &power_page_changeable,
4071 				       sizeof(power_page_changeable));
4072 				memcpy(&lun->mode_pages.power_subpage[
4073 				       CTL_PAGE_DEFAULT],
4074 				       &power_page_default,
4075 				       sizeof(power_page_default));
4076 				memcpy(&lun->mode_pages.power_subpage[
4077 				       CTL_PAGE_SAVED],
4078 				       &power_page_default,
4079 				       sizeof(power_page_default));
4080 				page_index->page_data =
4081 				    (uint8_t *)lun->mode_pages.power_subpage;
4082 
4083 				current_page = (struct copan_power_subpage *)
4084 					(page_index->page_data +
4085 					 (page_index->page_len *
4086 					  CTL_PAGE_CURRENT));
4087 			        saved_page = (struct copan_power_subpage *)
4088 				        (page_index->page_data +
4089 					 (page_index->page_len *
4090 					  CTL_PAGE_SAVED));
4091 				break;
4092 			}
4093 			case APS_SUBPAGE_CODE: {
4094 				struct copan_aps_subpage *current_page,
4095 							 *saved_page;
4096 
4097 				// This gets set multiple times but
4098 				// it should always be the same. It's
4099 				// only done during init so who cares.
4100 				index_to_aps_page = i;
4101 
4102 				memcpy(&lun->mode_pages.aps_subpage[
4103 				       CTL_PAGE_CURRENT],
4104 				       &aps_page_default,
4105 				       sizeof(aps_page_default));
4106 				memcpy(&lun->mode_pages.aps_subpage[
4107 				       CTL_PAGE_CHANGEABLE],
4108 				       &aps_page_changeable,
4109 				       sizeof(aps_page_changeable));
4110 				memcpy(&lun->mode_pages.aps_subpage[
4111 				       CTL_PAGE_DEFAULT],
4112 				       &aps_page_default,
4113 				       sizeof(aps_page_default));
4114 				memcpy(&lun->mode_pages.aps_subpage[
4115 				       CTL_PAGE_SAVED],
4116 				       &aps_page_default,
4117 				       sizeof(aps_page_default));
4118 				page_index->page_data =
4119 					(uint8_t *)lun->mode_pages.aps_subpage;
4120 
4121 				current_page = (struct copan_aps_subpage *)
4122 					(page_index->page_data +
4123 					 (page_index->page_len *
4124 					  CTL_PAGE_CURRENT));
4125 				saved_page = (struct copan_aps_subpage *)
4126 					(page_index->page_data +
4127 					 (page_index->page_len *
4128 					  CTL_PAGE_SAVED));
4129 				break;
4130 			}
4131 			case DBGCNF_SUBPAGE_CODE: {
4132 				struct copan_debugconf_subpage *current_page,
4133 							       *saved_page;
4134 
4135 				memcpy(&lun->mode_pages.debugconf_subpage[
4136 				       CTL_PAGE_CURRENT],
4137 				       &debugconf_page_default,
4138 				       sizeof(debugconf_page_default));
4139 				memcpy(&lun->mode_pages.debugconf_subpage[
4140 				       CTL_PAGE_CHANGEABLE],
4141 				       &debugconf_page_changeable,
4142 				       sizeof(debugconf_page_changeable));
4143 				memcpy(&lun->mode_pages.debugconf_subpage[
4144 				       CTL_PAGE_DEFAULT],
4145 				       &debugconf_page_default,
4146 				       sizeof(debugconf_page_default));
4147 				memcpy(&lun->mode_pages.debugconf_subpage[
4148 				       CTL_PAGE_SAVED],
4149 				       &debugconf_page_default,
4150 				       sizeof(debugconf_page_default));
4151 				page_index->page_data =
4152 					(uint8_t *)lun->mode_pages.debugconf_subpage;
4153 
4154 				current_page = (struct copan_debugconf_subpage *)
4155 					(page_index->page_data +
4156 					 (page_index->page_len *
4157 					  CTL_PAGE_CURRENT));
4158 				saved_page = (struct copan_debugconf_subpage *)
4159 					(page_index->page_data +
4160 					 (page_index->page_len *
4161 					  CTL_PAGE_SAVED));
4162 				break;
4163 			}
4164 			default:
4165 				panic("invalid subpage value %d",
4166 				      page_index->subpage);
4167 				break;
4168 			}
4169    			break;
4170 		}
4171 		default:
4172 			panic("invalid page value %d",
4173 			      page_index->page_code & SMPH_PC_MASK);
4174 			break;
4175     	}
4176 	}
4177 
4178 	return (CTL_RETVAL_COMPLETE);
4179 }
4180 
4181 /*
4182  * LUN allocation.
4183  *
4184  * Requirements:
4185  * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4186  *   wants us to allocate the LUN and he can block.
4187  * - ctl_softc is always set
4188  * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4189  *
4190  * Returns 0 for success, non-zero (errno) for failure.
4191  */
4192 static int
4193 ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4194 	      struct ctl_be_lun *const be_lun, struct ctl_id target_id)
4195 {
4196 	struct ctl_lun *nlun, *lun;
4197 	struct ctl_frontend *fe;
4198 	int lun_number, i;
4199 
4200 	if (be_lun == NULL)
4201 		return (EINVAL);
4202 
4203 	/*
4204 	 * We currently only support Direct Access or Processor LUN types.
4205 	 */
4206 	switch (be_lun->lun_type) {
4207 	case T_DIRECT:
4208 		break;
4209 	case T_PROCESSOR:
4210 		break;
4211 	case T_SEQUENTIAL:
4212 	case T_CHANGER:
4213 	default:
4214 		be_lun->lun_config_status(be_lun->be_lun,
4215 					  CTL_LUN_CONFIG_FAILURE);
4216 		break;
4217 	}
4218 	if (ctl_lun == NULL) {
4219 		lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4220 		lun->flags = CTL_LUN_MALLOCED;
4221 	} else
4222 		lun = ctl_lun;
4223 
4224 	memset(lun, 0, sizeof(*lun));
4225 
4226 	mtx_lock(&ctl_softc->ctl_lock);
4227 	/*
4228 	 * See if the caller requested a particular LUN number.  If so, see
4229 	 * if it is available.  Otherwise, allocate the first available LUN.
4230 	 */
4231 	if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4232 		if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4233 		 || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4234 			mtx_unlock(&ctl_softc->ctl_lock);
4235 			if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4236 				printf("ctl: requested LUN ID %d is higher "
4237 				       "than CTL_MAX_LUNS - 1 (%d)\n",
4238 				       be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4239 			} else {
4240 				/*
4241 				 * XXX KDM return an error, or just assign
4242 				 * another LUN ID in this case??
4243 				 */
4244 				printf("ctl: requested LUN ID %d is already "
4245 				       "in use\n", be_lun->req_lun_id);
4246 			}
4247 			if (lun->flags & CTL_LUN_MALLOCED)
4248 				free(lun, M_CTL);
4249 			be_lun->lun_config_status(be_lun->be_lun,
4250 						  CTL_LUN_CONFIG_FAILURE);
4251 			return (ENOSPC);
4252 		}
4253 		lun_number = be_lun->req_lun_id;
4254 	} else {
4255 		lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, CTL_MAX_LUNS);
4256 		if (lun_number == -1) {
4257 			mtx_unlock(&ctl_softc->ctl_lock);
4258 			printf("ctl: can't allocate LUN on target %ju, out of "
4259 			       "LUNs\n", (uintmax_t)target_id.id);
4260 			if (lun->flags & CTL_LUN_MALLOCED)
4261 				free(lun, M_CTL);
4262 			be_lun->lun_config_status(be_lun->be_lun,
4263 						  CTL_LUN_CONFIG_FAILURE);
4264 			return (ENOSPC);
4265 		}
4266 	}
4267 	ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4268 
4269 	lun->target = target_id;
4270 	lun->lun = lun_number;
4271 	lun->be_lun = be_lun;
4272 	/*
4273 	 * The processor LUN is always enabled.  Disk LUNs come on line
4274 	 * disabled, and must be enabled by the backend.
4275 	 */
4276 	lun->flags = CTL_LUN_DISABLED;
4277 	lun->backend = be_lun->be;
4278 	be_lun->ctl_lun = lun;
4279 	be_lun->lun_id = lun_number;
4280 	atomic_add_int(&be_lun->be->num_luns, 1);
4281 	if (be_lun->flags & CTL_LUN_FLAG_POWERED_OFF)
4282 		lun->flags |= CTL_LUN_STOPPED;
4283 
4284 	if (be_lun->flags & CTL_LUN_FLAG_INOPERABLE)
4285 		lun->flags |= CTL_LUN_INOPERABLE;
4286 
4287 	if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4288 		lun->flags |= CTL_LUN_PRIMARY_SC;
4289 
4290 	lun->ctl_softc = ctl_softc;
4291 	TAILQ_INIT(&lun->ooa_queue);
4292 	TAILQ_INIT(&lun->blocked_queue);
4293 	STAILQ_INIT(&lun->error_list);
4294 
4295 	/*
4296 	 * Initialize the mode page index.
4297 	 */
4298 	ctl_init_page_index(lun);
4299 
4300 	/*
4301 	 * Set the poweron UA for all initiators on this LUN only.
4302 	 */
4303 	for (i = 0; i < CTL_MAX_INITIATORS; i++)
4304 		lun->pending_sense[i].ua_pending = CTL_UA_POWERON;
4305 
4306 	/*
4307 	 * Now, before we insert this lun on the lun list, set the lun
4308 	 * inventory changed UA for all other luns.
4309 	 */
4310 	STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4311 		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4312 			nlun->pending_sense[i].ua_pending |= CTL_UA_LUN_CHANGE;
4313 		}
4314 	}
4315 
4316 	STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4317 
4318 	ctl_softc->ctl_luns[lun_number] = lun;
4319 
4320 	ctl_softc->num_luns++;
4321 
4322 	/* Setup statistics gathering */
4323 	lun->stats.device_type = be_lun->lun_type;
4324 	lun->stats.lun_number = lun_number;
4325 	if (lun->stats.device_type == T_DIRECT)
4326 		lun->stats.blocksize = be_lun->blocksize;
4327 	else
4328 		lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4329 	for (i = 0;i < CTL_MAX_PORTS;i++)
4330 		lun->stats.ports[i].targ_port = i;
4331 
4332 	mtx_unlock(&ctl_softc->ctl_lock);
4333 
4334 	lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4335 
4336 	/*
4337 	 * Run through each registered FETD and bring it online if it isn't
4338 	 * already.  Enable the target ID if it hasn't been enabled, and
4339 	 * enable this particular LUN.
4340 	 */
4341 	STAILQ_FOREACH(fe, &ctl_softc->fe_list, links) {
4342 		int retval;
4343 
4344 		/*
4345 		 * XXX KDM this only works for ONE TARGET ID.  We'll need
4346 		 * to do things differently if we go to a multiple target
4347 		 * ID scheme.
4348 		 */
4349 		if ((fe->status & CTL_PORT_STATUS_TARG_ONLINE) == 0) {
4350 
4351 			retval = fe->targ_enable(fe->targ_lun_arg, target_id);
4352 			if (retval != 0) {
4353 				printf("ctl_alloc_lun: FETD %s port %d "
4354 				       "returned error %d for targ_enable on "
4355 				       "target %ju\n", fe->port_name,
4356 				       fe->targ_port, retval,
4357 				       (uintmax_t)target_id.id);
4358 			} else
4359 				fe->status |= CTL_PORT_STATUS_TARG_ONLINE;
4360 		}
4361 
4362 		retval = fe->lun_enable(fe->targ_lun_arg, target_id,lun_number);
4363 		if (retval != 0) {
4364 			printf("ctl_alloc_lun: FETD %s port %d returned error "
4365 			       "%d for lun_enable on target %ju lun %d\n",
4366 			       fe->port_name, fe->targ_port, retval,
4367 			       (uintmax_t)target_id.id, lun_number);
4368 		} else
4369 			fe->status |= CTL_PORT_STATUS_LUN_ONLINE;
4370 	}
4371 	return (0);
4372 }
4373 
4374 /*
4375  * Delete a LUN.
4376  * Assumptions:
4377  * - caller holds ctl_softc->ctl_lock.
4378  * - LUN has already been marked invalid and any pending I/O has been taken
4379  *   care of.
4380  */
4381 static int
4382 ctl_free_lun(struct ctl_lun *lun)
4383 {
4384 	struct ctl_softc *softc;
4385 #if 0
4386 	struct ctl_frontend *fe;
4387 #endif
4388 	struct ctl_lun *nlun;
4389 	union ctl_io *io, *next_io;
4390 	int i;
4391 
4392 	softc = lun->ctl_softc;
4393 
4394 	STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4395 
4396 	ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4397 
4398 	softc->ctl_luns[lun->lun] = NULL;
4399 
4400 	if (TAILQ_FIRST(&lun->ooa_queue) != NULL) {
4401 		printf("ctl_free_lun: aieee!! freeing a LUN with "
4402 		       "outstanding I/O!!\n");
4403 	}
4404 
4405 	/*
4406 	 * If we have anything pending on the RtR queue, remove it.
4407 	 */
4408 	for (io = (union ctl_io *)STAILQ_FIRST(&softc->rtr_queue); io != NULL;
4409 	     io = next_io) {
4410 		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
4411 		if ((io->io_hdr.nexus.targ_target.id == lun->target.id)
4412 		 && (io->io_hdr.nexus.targ_lun == lun->lun))
4413 			STAILQ_REMOVE(&softc->rtr_queue, &io->io_hdr,
4414 				      ctl_io_hdr, links);
4415 	}
4416 
4417 	/*
4418 	 * Then remove everything from the blocked queue.
4419 	 */
4420 	for (io = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue); io != NULL;
4421 	     io = next_io) {
4422 		next_io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,blocked_links);
4423 		TAILQ_REMOVE(&lun->blocked_queue, &io->io_hdr, blocked_links);
4424 		io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
4425 	}
4426 
4427 	/*
4428 	 * Now clear out the OOA queue, and free all the I/O.
4429 	 * XXX KDM should we notify the FETD here?  We probably need to
4430 	 * quiesce the LUN before deleting it.
4431 	 */
4432 	for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); io != NULL;
4433 	     io = next_io) {
4434 		next_io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr, ooa_links);
4435 		TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
4436 		ctl_free_io_internal(io, /*have_lock*/ 1);
4437 	}
4438 
4439 	softc->num_luns--;
4440 
4441 	/*
4442 	 * XXX KDM this scheme only works for a single target/multiple LUN
4443 	 * setup.  It needs to be revamped for a multiple target scheme.
4444 	 *
4445 	 * XXX KDM this results in fe->lun_disable() getting called twice,
4446 	 * once when ctl_disable_lun() is called, and a second time here.
4447 	 * We really need to re-think the LUN disable semantics.  There
4448 	 * should probably be several steps/levels to LUN removal:
4449 	 *  - disable
4450 	 *  - invalidate
4451 	 *  - free
4452  	 *
4453 	 * Right now we only have a disable method when communicating to
4454 	 * the front end ports, at least for individual LUNs.
4455 	 */
4456 #if 0
4457 	STAILQ_FOREACH(fe, &softc->fe_list, links) {
4458 		int retval;
4459 
4460 		retval = fe->lun_disable(fe->targ_lun_arg, lun->target,
4461 					 lun->lun);
4462 		if (retval != 0) {
4463 			printf("ctl_free_lun: FETD %s port %d returned error "
4464 			       "%d for lun_disable on target %ju lun %jd\n",
4465 			       fe->port_name, fe->targ_port, retval,
4466 			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4467 		}
4468 
4469 		if (STAILQ_FIRST(&softc->lun_list) == NULL) {
4470 			fe->status &= ~CTL_PORT_STATUS_LUN_ONLINE;
4471 
4472 			retval = fe->targ_disable(fe->targ_lun_arg,lun->target);
4473 			if (retval != 0) {
4474 				printf("ctl_free_lun: FETD %s port %d "
4475 				       "returned error %d for targ_disable on "
4476 				       "target %ju\n", fe->port_name,
4477 				       fe->targ_port, retval,
4478 				       (uintmax_t)lun->target.id);
4479 			} else
4480 				fe->status &= ~CTL_PORT_STATUS_TARG_ONLINE;
4481 
4482 			if ((fe->status & CTL_PORT_STATUS_TARG_ONLINE) != 0)
4483 				continue;
4484 
4485 #if 0
4486 			fe->port_offline(fe->onoff_arg);
4487 			fe->status &= ~CTL_PORT_STATUS_ONLINE;
4488 #endif
4489 		}
4490 	}
4491 #endif
4492 
4493 	/*
4494 	 * Tell the backend to free resources, if this LUN has a backend.
4495 	 */
4496 	atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4497 	lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4498 
4499 	if (lun->flags & CTL_LUN_MALLOCED)
4500 		free(lun, M_CTL);
4501 
4502 	STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4503 		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4504 			nlun->pending_sense[i].ua_pending |= CTL_UA_LUN_CHANGE;
4505 		}
4506 	}
4507 
4508 	return (0);
4509 }
4510 
4511 static void
4512 ctl_create_lun(struct ctl_be_lun *be_lun)
4513 {
4514 	struct ctl_softc *ctl_softc;
4515 
4516 	ctl_softc = control_softc;
4517 
4518 	/*
4519 	 * ctl_alloc_lun() should handle all potential failure cases.
4520 	 */
4521 	ctl_alloc_lun(ctl_softc, NULL, be_lun, ctl_softc->target);
4522 }
4523 
4524 int
4525 ctl_add_lun(struct ctl_be_lun *be_lun)
4526 {
4527 	struct ctl_softc *ctl_softc;
4528 
4529 	ctl_softc = control_softc;
4530 
4531 	mtx_lock(&ctl_softc->ctl_lock);
4532 	STAILQ_INSERT_TAIL(&ctl_softc->pending_lun_queue, be_lun, links);
4533 	mtx_unlock(&ctl_softc->ctl_lock);
4534 
4535 	ctl_wakeup_thread();
4536 
4537 	return (0);
4538 }
4539 
4540 int
4541 ctl_enable_lun(struct ctl_be_lun *be_lun)
4542 {
4543 	struct ctl_softc *ctl_softc;
4544 	struct ctl_frontend *fe, *nfe;
4545 	struct ctl_lun *lun;
4546 	int retval;
4547 
4548 	ctl_softc = control_softc;
4549 
4550 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4551 
4552 	mtx_lock(&ctl_softc->ctl_lock);
4553 	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4554 		/*
4555 		 * eh?  Why did we get called if the LUN is already
4556 		 * enabled?
4557 		 */
4558 		mtx_unlock(&ctl_softc->ctl_lock);
4559 		return (0);
4560 	}
4561 	lun->flags &= ~CTL_LUN_DISABLED;
4562 
4563 	for (fe = STAILQ_FIRST(&ctl_softc->fe_list); fe != NULL; fe = nfe) {
4564 		nfe = STAILQ_NEXT(fe, links);
4565 
4566 		/*
4567 		 * Drop the lock while we call the FETD's enable routine.
4568 		 * This can lead to a callback into CTL (at least in the
4569 		 * case of the internal initiator frontend.
4570 		 */
4571 		mtx_unlock(&ctl_softc->ctl_lock);
4572 		retval = fe->lun_enable(fe->targ_lun_arg, lun->target,lun->lun);
4573 		mtx_lock(&ctl_softc->ctl_lock);
4574 		if (retval != 0) {
4575 			printf("%s: FETD %s port %d returned error "
4576 			       "%d for lun_enable on target %ju lun %jd\n",
4577 			       __func__, fe->port_name, fe->targ_port, retval,
4578 			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4579 		}
4580 #if 0
4581 		 else {
4582             /* NOTE:  TODO:  why does lun enable affect port status? */
4583 			fe->status |= CTL_PORT_STATUS_LUN_ONLINE;
4584 		}
4585 #endif
4586 	}
4587 
4588 	mtx_unlock(&ctl_softc->ctl_lock);
4589 
4590 	return (0);
4591 }
4592 
4593 int
4594 ctl_disable_lun(struct ctl_be_lun *be_lun)
4595 {
4596 	struct ctl_softc *ctl_softc;
4597 	struct ctl_frontend *fe;
4598 	struct ctl_lun *lun;
4599 	int retval;
4600 
4601 	ctl_softc = control_softc;
4602 
4603 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4604 
4605 	mtx_lock(&ctl_softc->ctl_lock);
4606 
4607 	if (lun->flags & CTL_LUN_DISABLED) {
4608 		mtx_unlock(&ctl_softc->ctl_lock);
4609 		return (0);
4610 	}
4611 	lun->flags |= CTL_LUN_DISABLED;
4612 
4613 	STAILQ_FOREACH(fe, &ctl_softc->fe_list, links) {
4614 		mtx_unlock(&ctl_softc->ctl_lock);
4615 		/*
4616 		 * Drop the lock before we call the frontend's disable
4617 		 * routine, to avoid lock order reversals.
4618 		 *
4619 		 * XXX KDM what happens if the frontend list changes while
4620 		 * we're traversing it?  It's unlikely, but should be handled.
4621 		 */
4622 		retval = fe->lun_disable(fe->targ_lun_arg, lun->target,
4623 					 lun->lun);
4624 		mtx_lock(&ctl_softc->ctl_lock);
4625 		if (retval != 0) {
4626 			printf("ctl_alloc_lun: FETD %s port %d returned error "
4627 			       "%d for lun_disable on target %ju lun %jd\n",
4628 			       fe->port_name, fe->targ_port, retval,
4629 			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4630 		}
4631 	}
4632 
4633 	mtx_unlock(&ctl_softc->ctl_lock);
4634 
4635 	return (0);
4636 }
4637 
4638 int
4639 ctl_start_lun(struct ctl_be_lun *be_lun)
4640 {
4641 	struct ctl_softc *ctl_softc;
4642 	struct ctl_lun *lun;
4643 
4644 	ctl_softc = control_softc;
4645 
4646 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4647 
4648 	mtx_lock(&ctl_softc->ctl_lock);
4649 	lun->flags &= ~CTL_LUN_STOPPED;
4650 	mtx_unlock(&ctl_softc->ctl_lock);
4651 
4652 	return (0);
4653 }
4654 
4655 int
4656 ctl_stop_lun(struct ctl_be_lun *be_lun)
4657 {
4658 	struct ctl_softc *ctl_softc;
4659 	struct ctl_lun *lun;
4660 
4661 	ctl_softc = control_softc;
4662 
4663 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4664 
4665 	mtx_lock(&ctl_softc->ctl_lock);
4666 	lun->flags |= CTL_LUN_STOPPED;
4667 	mtx_unlock(&ctl_softc->ctl_lock);
4668 
4669 	return (0);
4670 }
4671 
4672 int
4673 ctl_lun_offline(struct ctl_be_lun *be_lun)
4674 {
4675 	struct ctl_softc *ctl_softc;
4676 	struct ctl_lun *lun;
4677 
4678 	ctl_softc = control_softc;
4679 
4680 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4681 
4682 	mtx_lock(&ctl_softc->ctl_lock);
4683 	lun->flags |= CTL_LUN_OFFLINE;
4684 	mtx_unlock(&ctl_softc->ctl_lock);
4685 
4686 	return (0);
4687 }
4688 
4689 int
4690 ctl_lun_online(struct ctl_be_lun *be_lun)
4691 {
4692 	struct ctl_softc *ctl_softc;
4693 	struct ctl_lun *lun;
4694 
4695 	ctl_softc = control_softc;
4696 
4697 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4698 
4699 	mtx_lock(&ctl_softc->ctl_lock);
4700 	lun->flags &= ~CTL_LUN_OFFLINE;
4701 	mtx_unlock(&ctl_softc->ctl_lock);
4702 
4703 	return (0);
4704 }
4705 
4706 int
4707 ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4708 {
4709 	struct ctl_softc *ctl_softc;
4710 	struct ctl_lun *lun;
4711 
4712 	ctl_softc = control_softc;
4713 
4714 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4715 
4716 	mtx_lock(&ctl_softc->ctl_lock);
4717 
4718 	/*
4719 	 * The LUN needs to be disabled before it can be marked invalid.
4720 	 */
4721 	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4722 		mtx_unlock(&ctl_softc->ctl_lock);
4723 		return (-1);
4724 	}
4725 	/*
4726 	 * Mark the LUN invalid.
4727 	 */
4728 	lun->flags |= CTL_LUN_INVALID;
4729 
4730 	/*
4731 	 * If there is nothing in the OOA queue, go ahead and free the LUN.
4732 	 * If we have something in the OOA queue, we'll free it when the
4733 	 * last I/O completes.
4734 	 */
4735 	if (TAILQ_FIRST(&lun->ooa_queue) == NULL)
4736 		ctl_free_lun(lun);
4737 	mtx_unlock(&ctl_softc->ctl_lock);
4738 
4739 	return (0);
4740 }
4741 
4742 int
4743 ctl_lun_inoperable(struct ctl_be_lun *be_lun)
4744 {
4745 	struct ctl_softc *ctl_softc;
4746 	struct ctl_lun *lun;
4747 
4748 	ctl_softc = control_softc;
4749 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4750 
4751 	mtx_lock(&ctl_softc->ctl_lock);
4752 	lun->flags |= CTL_LUN_INOPERABLE;
4753 	mtx_unlock(&ctl_softc->ctl_lock);
4754 
4755 	return (0);
4756 }
4757 
4758 int
4759 ctl_lun_operable(struct ctl_be_lun *be_lun)
4760 {
4761 	struct ctl_softc *ctl_softc;
4762 	struct ctl_lun *lun;
4763 
4764 	ctl_softc = control_softc;
4765 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4766 
4767 	mtx_lock(&ctl_softc->ctl_lock);
4768 	lun->flags &= ~CTL_LUN_INOPERABLE;
4769 	mtx_unlock(&ctl_softc->ctl_lock);
4770 
4771 	return (0);
4772 }
4773 
4774 int
4775 ctl_lun_power_lock(struct ctl_be_lun *be_lun, struct ctl_nexus *nexus,
4776 		   int lock)
4777 {
4778 	struct ctl_softc *softc;
4779 	struct ctl_lun *lun;
4780 	struct copan_aps_subpage *current_sp;
4781 	struct ctl_page_index *page_index;
4782 	int i;
4783 
4784 	softc = control_softc;
4785 
4786 	mtx_lock(&softc->ctl_lock);
4787 
4788 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4789 
4790 	page_index = NULL;
4791 	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
4792 		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
4793 		     APS_PAGE_CODE)
4794 			continue;
4795 
4796 		if (lun->mode_pages.index[i].subpage != APS_SUBPAGE_CODE)
4797 			continue;
4798 		page_index = &lun->mode_pages.index[i];
4799 	}
4800 
4801 	if (page_index == NULL) {
4802 		mtx_unlock(&softc->ctl_lock);
4803 		printf("%s: APS subpage not found for lun %ju!\n", __func__,
4804 		       (uintmax_t)lun->lun);
4805 		return (1);
4806 	}
4807 #if 0
4808 	if ((softc->aps_locked_lun != 0)
4809 	 && (softc->aps_locked_lun != lun->lun)) {
4810 		printf("%s: attempt to lock LUN %llu when %llu is already "
4811 		       "locked\n");
4812 		mtx_unlock(&softc->ctl_lock);
4813 		return (1);
4814 	}
4815 #endif
4816 
4817 	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
4818 		(page_index->page_len * CTL_PAGE_CURRENT));
4819 
4820 	if (lock != 0) {
4821 		current_sp->lock_active = APS_LOCK_ACTIVE;
4822 		softc->aps_locked_lun = lun->lun;
4823 	} else {
4824 		current_sp->lock_active = 0;
4825 		softc->aps_locked_lun = 0;
4826 	}
4827 
4828 
4829 	/*
4830 	 * If we're in HA mode, try to send the lock message to the other
4831 	 * side.
4832 	 */
4833 	if (ctl_is_single == 0) {
4834 		int isc_retval;
4835 		union ctl_ha_msg lock_msg;
4836 
4837 		lock_msg.hdr.nexus = *nexus;
4838 		lock_msg.hdr.msg_type = CTL_MSG_APS_LOCK;
4839 		if (lock != 0)
4840 			lock_msg.aps.lock_flag = 1;
4841 		else
4842 			lock_msg.aps.lock_flag = 0;
4843 		isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &lock_msg,
4844 					 sizeof(lock_msg), 0);
4845 		if (isc_retval > CTL_HA_STATUS_SUCCESS) {
4846 			printf("%s: APS (lock=%d) error returned from "
4847 			       "ctl_ha_msg_send: %d\n", __func__, lock, isc_retval);
4848 			mtx_unlock(&softc->ctl_lock);
4849 			return (1);
4850 		}
4851 	}
4852 
4853 	mtx_unlock(&softc->ctl_lock);
4854 
4855 	return (0);
4856 }
4857 
4858 void
4859 ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
4860 {
4861 	struct ctl_lun *lun;
4862 	struct ctl_softc *softc;
4863 	int i;
4864 
4865 	softc = control_softc;
4866 
4867 	mtx_lock(&softc->ctl_lock);
4868 
4869 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4870 
4871 	for (i = 0; i < CTL_MAX_INITIATORS; i++)
4872 		lun->pending_sense[i].ua_pending |= CTL_UA_CAPACITY_CHANGED;
4873 
4874 	mtx_unlock(&softc->ctl_lock);
4875 }
4876 
4877 /*
4878  * Backend "memory move is complete" callback for requests that never
4879  * make it down to say RAIDCore's configuration code.
4880  */
4881 int
4882 ctl_config_move_done(union ctl_io *io)
4883 {
4884 	int retval;
4885 
4886 	retval = CTL_RETVAL_COMPLETE;
4887 
4888 
4889 	CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
4890 	/*
4891 	 * XXX KDM this shouldn't happen, but what if it does?
4892 	 */
4893 	if (io->io_hdr.io_type != CTL_IO_SCSI)
4894 		panic("I/O type isn't CTL_IO_SCSI!");
4895 
4896 	if ((io->io_hdr.port_status == 0)
4897 	 && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
4898 	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE))
4899 		io->io_hdr.status = CTL_SUCCESS;
4900 	else if ((io->io_hdr.port_status != 0)
4901 	      && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
4902 	      && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)){
4903 		/*
4904 		 * For hardware error sense keys, the sense key
4905 		 * specific value is defined to be a retry count,
4906 		 * but we use it to pass back an internal FETD
4907 		 * error code.  XXX KDM  Hopefully the FETD is only
4908 		 * using 16 bits for an error code, since that's
4909 		 * all the space we have in the sks field.
4910 		 */
4911 		ctl_set_internal_failure(&io->scsiio,
4912 					 /*sks_valid*/ 1,
4913 					 /*retry_count*/
4914 					 io->io_hdr.port_status);
4915 		free(io->scsiio.kern_data_ptr, M_CTL);
4916 		ctl_done(io);
4917 		goto bailout;
4918 	}
4919 
4920 	if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
4921 	 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
4922 	 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
4923 		/*
4924 		 * XXX KDM just assuming a single pointer here, and not a
4925 		 * S/G list.  If we start using S/G lists for config data,
4926 		 * we'll need to know how to clean them up here as well.
4927 		 */
4928 		free(io->scsiio.kern_data_ptr, M_CTL);
4929 		/* Hopefully the user has already set the status... */
4930 		ctl_done(io);
4931 	} else {
4932 		/*
4933 		 * XXX KDM now we need to continue data movement.  Some
4934 		 * options:
4935 		 * - call ctl_scsiio() again?  We don't do this for data
4936 		 *   writes, because for those at least we know ahead of
4937 		 *   time where the write will go and how long it is.  For
4938 		 *   config writes, though, that information is largely
4939 		 *   contained within the write itself, thus we need to
4940 		 *   parse out the data again.
4941 		 *
4942 		 * - Call some other function once the data is in?
4943 		 */
4944 
4945 		/*
4946 		 * XXX KDM call ctl_scsiio() again for now, and check flag
4947 		 * bits to see whether we're allocated or not.
4948 		 */
4949 		retval = ctl_scsiio(&io->scsiio);
4950 	}
4951 bailout:
4952 	return (retval);
4953 }
4954 
4955 /*
4956  * This gets called by a backend driver when it is done with a
4957  * configuration write.
4958  */
4959 void
4960 ctl_config_write_done(union ctl_io *io)
4961 {
4962 	/*
4963 	 * If the IO_CONT flag is set, we need to call the supplied
4964 	 * function to continue processing the I/O, instead of completing
4965 	 * the I/O just yet.
4966 	 *
4967 	 * If there is an error, though, we don't want to keep processing.
4968 	 * Instead, just send status back to the initiator.
4969 	 */
4970 	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT)
4971 	 && (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)
4972 	  || ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))) {
4973 		io->scsiio.io_cont(io);
4974 		return;
4975 	}
4976 	/*
4977 	 * Since a configuration write can be done for commands that actually
4978 	 * have data allocated, like write buffer, and commands that have
4979 	 * no data, like start/stop unit, we need to check here.
4980 	 */
4981 	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT)
4982 		free(io->scsiio.kern_data_ptr, M_CTL);
4983 	ctl_done(io);
4984 }
4985 
4986 /*
4987  * SCSI release command.
4988  */
4989 int
4990 ctl_scsi_release(struct ctl_scsiio *ctsio)
4991 {
4992 	int length, longid, thirdparty_id, resv_id;
4993 	struct ctl_softc *ctl_softc;
4994 	struct ctl_lun *lun;
4995 
4996 	length = 0;
4997 	resv_id = 0;
4998 
4999 	CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5000 
5001 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5002 	ctl_softc = control_softc;
5003 
5004 	switch (ctsio->cdb[0]) {
5005 	case RELEASE: {
5006 		struct scsi_release *cdb;
5007 
5008 		cdb = (struct scsi_release *)ctsio->cdb;
5009 		if ((cdb->byte2 & 0x1f) != 0) {
5010 			ctl_set_invalid_field(ctsio,
5011 					      /*sks_valid*/ 1,
5012 					      /*command*/ 1,
5013 					      /*field*/ 1,
5014 					      /*bit_valid*/ 0,
5015 					      /*bit*/ 0);
5016 			ctl_done((union ctl_io *)ctsio);
5017 			return (CTL_RETVAL_COMPLETE);
5018 		}
5019 		break;
5020 	}
5021 	case RELEASE_10: {
5022 		struct scsi_release_10 *cdb;
5023 
5024 		cdb = (struct scsi_release_10 *)ctsio->cdb;
5025 
5026 		if ((cdb->byte2 & SR10_EXTENT) != 0) {
5027 			ctl_set_invalid_field(ctsio,
5028 					      /*sks_valid*/ 1,
5029 					      /*command*/ 1,
5030 					      /*field*/ 1,
5031 					      /*bit_valid*/ 1,
5032 					      /*bit*/ 0);
5033 			ctl_done((union ctl_io *)ctsio);
5034 			return (CTL_RETVAL_COMPLETE);
5035 
5036 		}
5037 
5038 		if ((cdb->byte2 & SR10_3RDPTY) != 0) {
5039 			ctl_set_invalid_field(ctsio,
5040 					      /*sks_valid*/ 1,
5041 					      /*command*/ 1,
5042 					      /*field*/ 1,
5043 					      /*bit_valid*/ 1,
5044 					      /*bit*/ 4);
5045 			ctl_done((union ctl_io *)ctsio);
5046 			return (CTL_RETVAL_COMPLETE);
5047 		}
5048 
5049 		if (cdb->byte2 & SR10_LONGID)
5050 			longid = 1;
5051 		else
5052 			thirdparty_id = cdb->thirdparty_id;
5053 
5054 		resv_id = cdb->resv_id;
5055 		length = scsi_2btoul(cdb->length);
5056 		break;
5057 	}
5058 	}
5059 
5060 
5061 	/*
5062 	 * XXX KDM right now, we only support LUN reservation.  We don't
5063 	 * support 3rd party reservations, or extent reservations, which
5064 	 * might actually need the parameter list.  If we've gotten this
5065 	 * far, we've got a LUN reservation.  Anything else got kicked out
5066 	 * above.  So, according to SPC, ignore the length.
5067 	 */
5068 	length = 0;
5069 
5070 	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5071 	 && (length > 0)) {
5072 		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5073 		ctsio->kern_data_len = length;
5074 		ctsio->kern_total_len = length;
5075 		ctsio->kern_data_resid = 0;
5076 		ctsio->kern_rel_offset = 0;
5077 		ctsio->kern_sg_entries = 0;
5078 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5079 		ctsio->be_move_done = ctl_config_move_done;
5080 		ctl_datamove((union ctl_io *)ctsio);
5081 
5082 		return (CTL_RETVAL_COMPLETE);
5083 	}
5084 
5085 	if (length > 0)
5086 		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5087 
5088 	mtx_lock(&ctl_softc->ctl_lock);
5089 
5090 	/*
5091 	 * According to SPC, it is not an error for an intiator to attempt
5092 	 * to release a reservation on a LUN that isn't reserved, or that
5093 	 * is reserved by another initiator.  The reservation can only be
5094 	 * released, though, by the initiator who made it or by one of
5095 	 * several reset type events.
5096 	 */
5097 	if (lun->flags & CTL_LUN_RESERVED) {
5098 		if ((ctsio->io_hdr.nexus.initid.id == lun->rsv_nexus.initid.id)
5099 		 && (ctsio->io_hdr.nexus.targ_port == lun->rsv_nexus.targ_port)
5100 		 && (ctsio->io_hdr.nexus.targ_target.id ==
5101 		     lun->rsv_nexus.targ_target.id)) {
5102 			lun->flags &= ~CTL_LUN_RESERVED;
5103 		}
5104 	}
5105 
5106 	ctsio->scsi_status = SCSI_STATUS_OK;
5107 	ctsio->io_hdr.status = CTL_SUCCESS;
5108 
5109 	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5110 		free(ctsio->kern_data_ptr, M_CTL);
5111 		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5112 	}
5113 
5114 	mtx_unlock(&ctl_softc->ctl_lock);
5115 
5116 	ctl_done((union ctl_io *)ctsio);
5117 	return (CTL_RETVAL_COMPLETE);
5118 }
5119 
5120 int
5121 ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5122 {
5123 	int extent, thirdparty, longid;
5124 	int resv_id, length;
5125 	uint64_t thirdparty_id;
5126 	struct ctl_softc *ctl_softc;
5127 	struct ctl_lun *lun;
5128 
5129 	extent = 0;
5130 	thirdparty = 0;
5131 	longid = 0;
5132 	resv_id = 0;
5133 	length = 0;
5134 	thirdparty_id = 0;
5135 
5136 	CTL_DEBUG_PRINT(("ctl_reserve\n"));
5137 
5138 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5139 	ctl_softc = control_softc;
5140 
5141 	switch (ctsio->cdb[0]) {
5142 	case RESERVE: {
5143 		struct scsi_reserve *cdb;
5144 
5145 		cdb = (struct scsi_reserve *)ctsio->cdb;
5146 		if ((cdb->byte2 & 0x1f) != 0) {
5147 			ctl_set_invalid_field(ctsio,
5148 					      /*sks_valid*/ 1,
5149 					      /*command*/ 1,
5150 					      /*field*/ 1,
5151 					      /*bit_valid*/ 0,
5152 					      /*bit*/ 0);
5153 			ctl_done((union ctl_io *)ctsio);
5154 			return (CTL_RETVAL_COMPLETE);
5155 		}
5156 		resv_id = cdb->resv_id;
5157 		length = scsi_2btoul(cdb->length);
5158 		break;
5159 	}
5160 	case RESERVE_10: {
5161 		struct scsi_reserve_10 *cdb;
5162 
5163 		cdb = (struct scsi_reserve_10 *)ctsio->cdb;
5164 
5165 		if ((cdb->byte2 & SR10_EXTENT) != 0) {
5166 			ctl_set_invalid_field(ctsio,
5167 					      /*sks_valid*/ 1,
5168 					      /*command*/ 1,
5169 					      /*field*/ 1,
5170 					      /*bit_valid*/ 1,
5171 					      /*bit*/ 0);
5172 			ctl_done((union ctl_io *)ctsio);
5173 			return (CTL_RETVAL_COMPLETE);
5174 		}
5175 		if ((cdb->byte2 & SR10_3RDPTY) != 0) {
5176 			ctl_set_invalid_field(ctsio,
5177 					      /*sks_valid*/ 1,
5178 					      /*command*/ 1,
5179 					      /*field*/ 1,
5180 					      /*bit_valid*/ 1,
5181 					      /*bit*/ 4);
5182 			ctl_done((union ctl_io *)ctsio);
5183 			return (CTL_RETVAL_COMPLETE);
5184 		}
5185 		if (cdb->byte2 & SR10_LONGID)
5186 			longid = 1;
5187 		else
5188 			thirdparty_id = cdb->thirdparty_id;
5189 
5190 		resv_id = cdb->resv_id;
5191 		length = scsi_2btoul(cdb->length);
5192 		break;
5193 	}
5194 	}
5195 
5196 	/*
5197 	 * XXX KDM right now, we only support LUN reservation.  We don't
5198 	 * support 3rd party reservations, or extent reservations, which
5199 	 * might actually need the parameter list.  If we've gotten this
5200 	 * far, we've got a LUN reservation.  Anything else got kicked out
5201 	 * above.  So, according to SPC, ignore the length.
5202 	 */
5203 	length = 0;
5204 
5205 	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5206 	 && (length > 0)) {
5207 		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5208 		ctsio->kern_data_len = length;
5209 		ctsio->kern_total_len = length;
5210 		ctsio->kern_data_resid = 0;
5211 		ctsio->kern_rel_offset = 0;
5212 		ctsio->kern_sg_entries = 0;
5213 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5214 		ctsio->be_move_done = ctl_config_move_done;
5215 		ctl_datamove((union ctl_io *)ctsio);
5216 
5217 		return (CTL_RETVAL_COMPLETE);
5218 	}
5219 
5220 	if (length > 0)
5221 		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5222 
5223 	mtx_lock(&ctl_softc->ctl_lock);
5224 	if (lun->flags & CTL_LUN_RESERVED) {
5225 		if ((ctsio->io_hdr.nexus.initid.id != lun->rsv_nexus.initid.id)
5226 		 || (ctsio->io_hdr.nexus.targ_port != lun->rsv_nexus.targ_port)
5227 		 || (ctsio->io_hdr.nexus.targ_target.id !=
5228 		     lun->rsv_nexus.targ_target.id)) {
5229 			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
5230 			ctsio->io_hdr.status = CTL_SCSI_ERROR;
5231 			goto bailout;
5232 		}
5233 	}
5234 
5235 	lun->flags |= CTL_LUN_RESERVED;
5236 	lun->rsv_nexus = ctsio->io_hdr.nexus;
5237 
5238 	ctsio->scsi_status = SCSI_STATUS_OK;
5239 	ctsio->io_hdr.status = CTL_SUCCESS;
5240 
5241 bailout:
5242 	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5243 		free(ctsio->kern_data_ptr, M_CTL);
5244 		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5245 	}
5246 
5247 	mtx_unlock(&ctl_softc->ctl_lock);
5248 
5249 	ctl_done((union ctl_io *)ctsio);
5250 	return (CTL_RETVAL_COMPLETE);
5251 }
5252 
5253 int
5254 ctl_start_stop(struct ctl_scsiio *ctsio)
5255 {
5256 	struct scsi_start_stop_unit *cdb;
5257 	struct ctl_lun *lun;
5258 	struct ctl_softc *ctl_softc;
5259 	int retval;
5260 
5261 	CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5262 
5263 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5264 	ctl_softc = control_softc;
5265 	retval = 0;
5266 
5267 	cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5268 
5269 	/*
5270 	 * XXX KDM
5271 	 * We don't support the immediate bit on a stop unit.  In order to
5272 	 * do that, we would need to code up a way to know that a stop is
5273 	 * pending, and hold off any new commands until it completes, one
5274 	 * way or another.  Then we could accept or reject those commands
5275 	 * depending on its status.  We would almost need to do the reverse
5276 	 * of what we do below for an immediate start -- return the copy of
5277 	 * the ctl_io to the FETD with status to send to the host (and to
5278 	 * free the copy!) and then free the original I/O once the stop
5279 	 * actually completes.  That way, the OOA queue mechanism can work
5280 	 * to block commands that shouldn't proceed.  Another alternative
5281 	 * would be to put the copy in the queue in place of the original,
5282 	 * and return the original back to the caller.  That could be
5283 	 * slightly safer..
5284 	 */
5285 	if ((cdb->byte2 & SSS_IMMED)
5286 	 && ((cdb->how & SSS_START) == 0)) {
5287 		ctl_set_invalid_field(ctsio,
5288 				      /*sks_valid*/ 1,
5289 				      /*command*/ 1,
5290 				      /*field*/ 1,
5291 				      /*bit_valid*/ 1,
5292 				      /*bit*/ 0);
5293 		ctl_done((union ctl_io *)ctsio);
5294 		return (CTL_RETVAL_COMPLETE);
5295 	}
5296 
5297 	/*
5298 	 * We don't support the power conditions field.  We need to check
5299 	 * this prior to checking the load/eject and start/stop bits.
5300 	 */
5301 	if ((cdb->how & SSS_PC_MASK) != SSS_PC_START_VALID) {
5302 		ctl_set_invalid_field(ctsio,
5303 				      /*sks_valid*/ 1,
5304 				      /*command*/ 1,
5305 				      /*field*/ 4,
5306 				      /*bit_valid*/ 1,
5307 				      /*bit*/ 4);
5308 		ctl_done((union ctl_io *)ctsio);
5309 		return (CTL_RETVAL_COMPLETE);
5310 	}
5311 
5312 	/*
5313 	 * Media isn't removable, so we can't load or eject it.
5314 	 */
5315 	if ((cdb->how & SSS_LOEJ) != 0) {
5316 		ctl_set_invalid_field(ctsio,
5317 				      /*sks_valid*/ 1,
5318 				      /*command*/ 1,
5319 				      /*field*/ 4,
5320 				      /*bit_valid*/ 1,
5321 				      /*bit*/ 1);
5322 		ctl_done((union ctl_io *)ctsio);
5323 		return (CTL_RETVAL_COMPLETE);
5324 	}
5325 
5326 	if ((lun->flags & CTL_LUN_PR_RESERVED)
5327 	 && ((cdb->how & SSS_START)==0)) {
5328 		uint32_t residx;
5329 
5330 		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5331 		if (!lun->per_res[residx].registered
5332 		 || (lun->pr_res_idx!=residx && lun->res_type < 4)) {
5333 
5334 			ctl_set_reservation_conflict(ctsio);
5335 			ctl_done((union ctl_io *)ctsio);
5336 			return (CTL_RETVAL_COMPLETE);
5337 		}
5338 	}
5339 
5340 	/*
5341 	 * If there is no backend on this device, we can't start or stop
5342 	 * it.  In theory we shouldn't get any start/stop commands in the
5343 	 * first place at this level if the LUN doesn't have a backend.
5344 	 * That should get stopped by the command decode code.
5345 	 */
5346 	if (lun->backend == NULL) {
5347 		ctl_set_invalid_opcode(ctsio);
5348 		ctl_done((union ctl_io *)ctsio);
5349 		return (CTL_RETVAL_COMPLETE);
5350 	}
5351 
5352 	/*
5353 	 * XXX KDM Copan-specific offline behavior.
5354 	 * Figure out a reasonable way to port this?
5355 	 */
5356 #ifdef NEEDTOPORT
5357 	mtx_lock(&ctl_softc->ctl_lock);
5358 
5359 	if (((cdb->byte2 & SSS_ONOFFLINE) == 0)
5360 	 && (lun->flags & CTL_LUN_OFFLINE)) {
5361 		/*
5362 		 * If the LUN is offline, and the on/offline bit isn't set,
5363 		 * reject the start or stop.  Otherwise, let it through.
5364 		 */
5365 		mtx_unlock(&ctl_softc->ctl_lock);
5366 		ctl_set_lun_not_ready(ctsio);
5367 		ctl_done((union ctl_io *)ctsio);
5368 	} else {
5369 		mtx_unlock(&ctl_softc->ctl_lock);
5370 #endif /* NEEDTOPORT */
5371 		/*
5372 		 * This could be a start or a stop when we're online,
5373 		 * or a stop/offline or start/online.  A start or stop when
5374 		 * we're offline is covered in the case above.
5375 		 */
5376 		/*
5377 		 * In the non-immediate case, we send the request to
5378 		 * the backend and return status to the user when
5379 		 * it is done.
5380 		 *
5381 		 * In the immediate case, we allocate a new ctl_io
5382 		 * to hold a copy of the request, and send that to
5383 		 * the backend.  We then set good status on the
5384 		 * user's request and return it immediately.
5385 		 */
5386 		if (cdb->byte2 & SSS_IMMED) {
5387 			union ctl_io *new_io;
5388 
5389 			new_io = ctl_alloc_io(ctsio->io_hdr.pool);
5390 			if (new_io == NULL) {
5391 				ctl_set_busy(ctsio);
5392 				ctl_done((union ctl_io *)ctsio);
5393 			} else {
5394 				ctl_copy_io((union ctl_io *)ctsio,
5395 					    new_io);
5396 				retval = lun->backend->config_write(new_io);
5397 				ctl_set_success(ctsio);
5398 				ctl_done((union ctl_io *)ctsio);
5399 			}
5400 		} else {
5401 			retval = lun->backend->config_write(
5402 				(union ctl_io *)ctsio);
5403 		}
5404 #ifdef NEEDTOPORT
5405 	}
5406 #endif
5407 	return (retval);
5408 }
5409 
5410 /*
5411  * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5412  * we don't really do anything with the LBA and length fields if the user
5413  * passes them in.  Instead we'll just flush out the cache for the entire
5414  * LUN.
5415  */
5416 int
5417 ctl_sync_cache(struct ctl_scsiio *ctsio)
5418 {
5419 	struct ctl_lun *lun;
5420 	struct ctl_softc *ctl_softc;
5421 	uint64_t starting_lba;
5422 	uint32_t block_count;
5423 	int reladr, immed;
5424 	int retval;
5425 
5426 	CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5427 
5428 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5429 	ctl_softc = control_softc;
5430 	retval = 0;
5431 	reladr = 0;
5432 	immed = 0;
5433 
5434 	switch (ctsio->cdb[0]) {
5435 	case SYNCHRONIZE_CACHE: {
5436 		struct scsi_sync_cache *cdb;
5437 		cdb = (struct scsi_sync_cache *)ctsio->cdb;
5438 
5439 		if (cdb->byte2 & SSC_RELADR)
5440 			reladr = 1;
5441 
5442 		if (cdb->byte2 & SSC_IMMED)
5443 			immed = 1;
5444 
5445 		starting_lba = scsi_4btoul(cdb->begin_lba);
5446 		block_count = scsi_2btoul(cdb->lb_count);
5447 		break;
5448 	}
5449 	case SYNCHRONIZE_CACHE_16: {
5450 		struct scsi_sync_cache_16 *cdb;
5451 		cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5452 
5453 		if (cdb->byte2 & SSC_RELADR)
5454 			reladr = 1;
5455 
5456 		if (cdb->byte2 & SSC_IMMED)
5457 			immed = 1;
5458 
5459 		starting_lba = scsi_8btou64(cdb->begin_lba);
5460 		block_count = scsi_4btoul(cdb->lb_count);
5461 		break;
5462 	}
5463 	default:
5464 		ctl_set_invalid_opcode(ctsio);
5465 		ctl_done((union ctl_io *)ctsio);
5466 		goto bailout;
5467 		break; /* NOTREACHED */
5468 	}
5469 
5470 	if (immed) {
5471 		/*
5472 		 * We don't support the immediate bit.  Since it's in the
5473 		 * same place for the 10 and 16 byte SYNCHRONIZE CACHE
5474 		 * commands, we can just return the same error in either
5475 		 * case.
5476 		 */
5477 		ctl_set_invalid_field(ctsio,
5478 				      /*sks_valid*/ 1,
5479 				      /*command*/ 1,
5480 				      /*field*/ 1,
5481 				      /*bit_valid*/ 1,
5482 				      /*bit*/ 1);
5483 		ctl_done((union ctl_io *)ctsio);
5484 		goto bailout;
5485 	}
5486 
5487 	if (reladr) {
5488 		/*
5489 		 * We don't support the reladr bit either.  It can only be
5490 		 * used with linked commands, and we don't support linked
5491 		 * commands.  Since the bit is in the same place for the
5492 		 * 10 and 16 byte SYNCHRONIZE CACHE * commands, we can
5493 		 * just return the same error in either case.
5494 		 */
5495 		ctl_set_invalid_field(ctsio,
5496 				      /*sks_valid*/ 1,
5497 				      /*command*/ 1,
5498 				      /*field*/ 1,
5499 				      /*bit_valid*/ 1,
5500 				      /*bit*/ 0);
5501 		ctl_done((union ctl_io *)ctsio);
5502 		goto bailout;
5503 	}
5504 
5505 	/*
5506 	 * We check the LBA and length, but don't do anything with them.
5507 	 * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5508 	 * get flushed.  This check will just help satisfy anyone who wants
5509 	 * to see an error for an out of range LBA.
5510 	 */
5511 	if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5512 		ctl_set_lba_out_of_range(ctsio);
5513 		ctl_done((union ctl_io *)ctsio);
5514 		goto bailout;
5515 	}
5516 
5517 	/*
5518 	 * If this LUN has no backend, we can't flush the cache anyway.
5519 	 */
5520 	if (lun->backend == NULL) {
5521 		ctl_set_invalid_opcode(ctsio);
5522 		ctl_done((union ctl_io *)ctsio);
5523 		goto bailout;
5524 	}
5525 
5526 	/*
5527 	 * Check to see whether we're configured to send the SYNCHRONIZE
5528 	 * CACHE command directly to the back end.
5529 	 */
5530 	mtx_lock(&ctl_softc->ctl_lock);
5531 	if ((ctl_softc->flags & CTL_FLAG_REAL_SYNC)
5532 	 && (++(lun->sync_count) >= lun->sync_interval)) {
5533 		lun->sync_count = 0;
5534 		mtx_unlock(&ctl_softc->ctl_lock);
5535 		retval = lun->backend->config_write((union ctl_io *)ctsio);
5536 	} else {
5537 		mtx_unlock(&ctl_softc->ctl_lock);
5538 		ctl_set_success(ctsio);
5539 		ctl_done((union ctl_io *)ctsio);
5540 	}
5541 
5542 bailout:
5543 
5544 	return (retval);
5545 }
5546 
5547 int
5548 ctl_format(struct ctl_scsiio *ctsio)
5549 {
5550 	struct scsi_format *cdb;
5551 	struct ctl_lun *lun;
5552 	struct ctl_softc *ctl_softc;
5553 	int length, defect_list_len;
5554 
5555 	CTL_DEBUG_PRINT(("ctl_format\n"));
5556 
5557 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5558 	ctl_softc = control_softc;
5559 
5560 	cdb = (struct scsi_format *)ctsio->cdb;
5561 
5562 	length = 0;
5563 	if (cdb->byte2 & SF_FMTDATA) {
5564 		if (cdb->byte2 & SF_LONGLIST)
5565 			length = sizeof(struct scsi_format_header_long);
5566 		else
5567 			length = sizeof(struct scsi_format_header_short);
5568 	}
5569 
5570 	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5571 	 && (length > 0)) {
5572 		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5573 		ctsio->kern_data_len = length;
5574 		ctsio->kern_total_len = length;
5575 		ctsio->kern_data_resid = 0;
5576 		ctsio->kern_rel_offset = 0;
5577 		ctsio->kern_sg_entries = 0;
5578 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5579 		ctsio->be_move_done = ctl_config_move_done;
5580 		ctl_datamove((union ctl_io *)ctsio);
5581 
5582 		return (CTL_RETVAL_COMPLETE);
5583 	}
5584 
5585 	defect_list_len = 0;
5586 
5587 	if (cdb->byte2 & SF_FMTDATA) {
5588 		if (cdb->byte2 & SF_LONGLIST) {
5589 			struct scsi_format_header_long *header;
5590 
5591 			header = (struct scsi_format_header_long *)
5592 				ctsio->kern_data_ptr;
5593 
5594 			defect_list_len = scsi_4btoul(header->defect_list_len);
5595 			if (defect_list_len != 0) {
5596 				ctl_set_invalid_field(ctsio,
5597 						      /*sks_valid*/ 1,
5598 						      /*command*/ 0,
5599 						      /*field*/ 2,
5600 						      /*bit_valid*/ 0,
5601 						      /*bit*/ 0);
5602 				goto bailout;
5603 			}
5604 		} else {
5605 			struct scsi_format_header_short *header;
5606 
5607 			header = (struct scsi_format_header_short *)
5608 				ctsio->kern_data_ptr;
5609 
5610 			defect_list_len = scsi_2btoul(header->defect_list_len);
5611 			if (defect_list_len != 0) {
5612 				ctl_set_invalid_field(ctsio,
5613 						      /*sks_valid*/ 1,
5614 						      /*command*/ 0,
5615 						      /*field*/ 2,
5616 						      /*bit_valid*/ 0,
5617 						      /*bit*/ 0);
5618 				goto bailout;
5619 			}
5620 		}
5621 	}
5622 
5623 	/*
5624 	 * The format command will clear out the "Medium format corrupted"
5625 	 * status if set by the configuration code.  That status is really
5626 	 * just a way to notify the host that we have lost the media, and
5627 	 * get them to issue a command that will basically make them think
5628 	 * they're blowing away the media.
5629 	 */
5630 	mtx_lock(&ctl_softc->ctl_lock);
5631 	lun->flags &= ~CTL_LUN_INOPERABLE;
5632 	mtx_unlock(&ctl_softc->ctl_lock);
5633 
5634 	ctsio->scsi_status = SCSI_STATUS_OK;
5635 	ctsio->io_hdr.status = CTL_SUCCESS;
5636 bailout:
5637 
5638 	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5639 		free(ctsio->kern_data_ptr, M_CTL);
5640 		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5641 	}
5642 
5643 	ctl_done((union ctl_io *)ctsio);
5644 	return (CTL_RETVAL_COMPLETE);
5645 }
5646 
5647 int
5648 ctl_write_buffer(struct ctl_scsiio *ctsio)
5649 {
5650 	struct scsi_write_buffer *cdb;
5651 	struct copan_page_header *header;
5652 	struct ctl_lun *lun;
5653 	struct ctl_softc *ctl_softc;
5654 	int buffer_offset, len;
5655 	int retval;
5656 
5657 	header = NULL;
5658 
5659 	retval = CTL_RETVAL_COMPLETE;
5660 
5661 	CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5662 
5663 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5664 	ctl_softc = control_softc;
5665 	cdb = (struct scsi_write_buffer *)ctsio->cdb;
5666 
5667 	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5668 		ctl_set_invalid_field(ctsio,
5669 				      /*sks_valid*/ 1,
5670 				      /*command*/ 1,
5671 				      /*field*/ 1,
5672 				      /*bit_valid*/ 1,
5673 				      /*bit*/ 4);
5674 		ctl_done((union ctl_io *)ctsio);
5675 		return (CTL_RETVAL_COMPLETE);
5676 	}
5677 	if (cdb->buffer_id != 0) {
5678 		ctl_set_invalid_field(ctsio,
5679 				      /*sks_valid*/ 1,
5680 				      /*command*/ 1,
5681 				      /*field*/ 2,
5682 				      /*bit_valid*/ 0,
5683 				      /*bit*/ 0);
5684 		ctl_done((union ctl_io *)ctsio);
5685 		return (CTL_RETVAL_COMPLETE);
5686 	}
5687 
5688 	len = scsi_3btoul(cdb->length);
5689 	buffer_offset = scsi_3btoul(cdb->offset);
5690 
5691 	if (len > sizeof(lun->write_buffer)) {
5692 		ctl_set_invalid_field(ctsio,
5693 				      /*sks_valid*/ 1,
5694 				      /*command*/ 1,
5695 				      /*field*/ 6,
5696 				      /*bit_valid*/ 0,
5697 				      /*bit*/ 0);
5698 		ctl_done((union ctl_io *)ctsio);
5699 		return (CTL_RETVAL_COMPLETE);
5700 	}
5701 
5702 	if (buffer_offset != 0) {
5703 		ctl_set_invalid_field(ctsio,
5704 				      /*sks_valid*/ 1,
5705 				      /*command*/ 1,
5706 				      /*field*/ 3,
5707 				      /*bit_valid*/ 0,
5708 				      /*bit*/ 0);
5709 		ctl_done((union ctl_io *)ctsio);
5710 		return (CTL_RETVAL_COMPLETE);
5711 	}
5712 
5713 	/*
5714 	 * If we've got a kernel request that hasn't been malloced yet,
5715 	 * malloc it and tell the caller the data buffer is here.
5716 	 */
5717 	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5718 		ctsio->kern_data_ptr = lun->write_buffer;
5719 		ctsio->kern_data_len = len;
5720 		ctsio->kern_total_len = len;
5721 		ctsio->kern_data_resid = 0;
5722 		ctsio->kern_rel_offset = 0;
5723 		ctsio->kern_sg_entries = 0;
5724 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5725 		ctsio->be_move_done = ctl_config_move_done;
5726 		ctl_datamove((union ctl_io *)ctsio);
5727 
5728 		return (CTL_RETVAL_COMPLETE);
5729 	}
5730 
5731 	ctl_done((union ctl_io *)ctsio);
5732 
5733 	return (CTL_RETVAL_COMPLETE);
5734 }
5735 
5736 /*
5737  * Note that this function currently doesn't actually do anything inside
5738  * CTL to enforce things if the DQue bit is turned on.
5739  *
5740  * Also note that this function can't be used in the default case, because
5741  * the DQue bit isn't set in the changeable mask for the control mode page
5742  * anyway.  This is just here as an example for how to implement a page
5743  * handler, and a placeholder in case we want to allow the user to turn
5744  * tagged queueing on and off.
5745  *
5746  * The D_SENSE bit handling is functional, however, and will turn
5747  * descriptor sense on and off for a given LUN.
5748  */
5749 int
5750 ctl_control_page_handler(struct ctl_scsiio *ctsio,
5751 			 struct ctl_page_index *page_index, uint8_t *page_ptr)
5752 {
5753 	struct scsi_control_page *current_cp, *saved_cp, *user_cp;
5754 	struct ctl_lun *lun;
5755 	struct ctl_softc *softc;
5756 	int set_ua;
5757 	uint32_t initidx;
5758 
5759 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5760 	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5761 	set_ua = 0;
5762 
5763 	user_cp = (struct scsi_control_page *)page_ptr;
5764 	current_cp = (struct scsi_control_page *)
5765 		(page_index->page_data + (page_index->page_len *
5766 		CTL_PAGE_CURRENT));
5767 	saved_cp = (struct scsi_control_page *)
5768 		(page_index->page_data + (page_index->page_len *
5769 		CTL_PAGE_SAVED));
5770 
5771 	softc = control_softc;
5772 
5773 	mtx_lock(&softc->ctl_lock);
5774 	if (((current_cp->rlec & SCP_DSENSE) == 0)
5775 	 && ((user_cp->rlec & SCP_DSENSE) != 0)) {
5776 		/*
5777 		 * Descriptor sense is currently turned off and the user
5778 		 * wants to turn it on.
5779 		 */
5780 		current_cp->rlec |= SCP_DSENSE;
5781 		saved_cp->rlec |= SCP_DSENSE;
5782 		lun->flags |= CTL_LUN_SENSE_DESC;
5783 		set_ua = 1;
5784 	} else if (((current_cp->rlec & SCP_DSENSE) != 0)
5785 		&& ((user_cp->rlec & SCP_DSENSE) == 0)) {
5786 		/*
5787 		 * Descriptor sense is currently turned on, and the user
5788 		 * wants to turn it off.
5789 		 */
5790 		current_cp->rlec &= ~SCP_DSENSE;
5791 		saved_cp->rlec &= ~SCP_DSENSE;
5792 		lun->flags &= ~CTL_LUN_SENSE_DESC;
5793 		set_ua = 1;
5794 	}
5795 	if (current_cp->queue_flags & SCP_QUEUE_DQUE) {
5796 		if (user_cp->queue_flags & SCP_QUEUE_DQUE) {
5797 #ifdef NEEDTOPORT
5798 			csevent_log(CSC_CTL | CSC_SHELF_SW |
5799 				    CTL_UNTAG_TO_UNTAG,
5800 				    csevent_LogType_Trace,
5801 				    csevent_Severity_Information,
5802 				    csevent_AlertLevel_Green,
5803 				    csevent_FRU_Firmware,
5804 				    csevent_FRU_Unknown,
5805 				    "Received untagged to untagged transition");
5806 #endif /* NEEDTOPORT */
5807 		} else {
5808 #ifdef NEEDTOPORT
5809 			csevent_log(CSC_CTL | CSC_SHELF_SW |
5810 				    CTL_UNTAG_TO_TAG,
5811 				    csevent_LogType_ConfigChange,
5812 				    csevent_Severity_Information,
5813 				    csevent_AlertLevel_Green,
5814 				    csevent_FRU_Firmware,
5815 				    csevent_FRU_Unknown,
5816 				    "Received untagged to tagged "
5817 				    "queueing transition");
5818 #endif /* NEEDTOPORT */
5819 
5820 			current_cp->queue_flags &= ~SCP_QUEUE_DQUE;
5821 			saved_cp->queue_flags &= ~SCP_QUEUE_DQUE;
5822 			set_ua = 1;
5823 		}
5824 	} else {
5825 		if (user_cp->queue_flags & SCP_QUEUE_DQUE) {
5826 #ifdef NEEDTOPORT
5827 			csevent_log(CSC_CTL | CSC_SHELF_SW |
5828 				    CTL_TAG_TO_UNTAG,
5829 				    csevent_LogType_ConfigChange,
5830 				    csevent_Severity_Warning,
5831 				    csevent_AlertLevel_Yellow,
5832 				    csevent_FRU_Firmware,
5833 				    csevent_FRU_Unknown,
5834 				    "Received tagged queueing to untagged "
5835 				    "transition");
5836 #endif /* NEEDTOPORT */
5837 
5838 			current_cp->queue_flags |= SCP_QUEUE_DQUE;
5839 			saved_cp->queue_flags |= SCP_QUEUE_DQUE;
5840 			set_ua = 1;
5841 		} else {
5842 #ifdef NEEDTOPORT
5843 			csevent_log(CSC_CTL | CSC_SHELF_SW |
5844 				    CTL_TAG_TO_TAG,
5845 				    csevent_LogType_Trace,
5846 				    csevent_Severity_Information,
5847 				    csevent_AlertLevel_Green,
5848 				    csevent_FRU_Firmware,
5849 				    csevent_FRU_Unknown,
5850 				    "Received tagged queueing to tagged "
5851 				    "queueing transition");
5852 #endif /* NEEDTOPORT */
5853 		}
5854 	}
5855 	if (set_ua != 0) {
5856 		int i;
5857 		/*
5858 		 * Let other initiators know that the mode
5859 		 * parameters for this LUN have changed.
5860 		 */
5861 		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
5862 			if (i == initidx)
5863 				continue;
5864 
5865 			lun->pending_sense[i].ua_pending |=
5866 				CTL_UA_MODE_CHANGE;
5867 		}
5868 	}
5869 	mtx_unlock(&softc->ctl_lock);
5870 
5871 	return (0);
5872 }
5873 
5874 int
5875 ctl_power_sp_handler(struct ctl_scsiio *ctsio,
5876 		     struct ctl_page_index *page_index, uint8_t *page_ptr)
5877 {
5878 	return (0);
5879 }
5880 
5881 int
5882 ctl_power_sp_sense_handler(struct ctl_scsiio *ctsio,
5883 			   struct ctl_page_index *page_index, int pc)
5884 {
5885 	struct copan_power_subpage *page;
5886 
5887 	page = (struct copan_power_subpage *)page_index->page_data +
5888 		(page_index->page_len * pc);
5889 
5890 	switch (pc) {
5891 	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
5892 		/*
5893 		 * We don't update the changable bits for this page.
5894 		 */
5895 		break;
5896 	case SMS_PAGE_CTRL_CURRENT >> 6:
5897 	case SMS_PAGE_CTRL_DEFAULT >> 6:
5898 	case SMS_PAGE_CTRL_SAVED >> 6:
5899 #ifdef NEEDTOPORT
5900 		ctl_update_power_subpage(page);
5901 #endif
5902 		break;
5903 	default:
5904 #ifdef NEEDTOPORT
5905 		EPRINT(0, "Invalid PC %d!!", pc);
5906 #endif
5907 		break;
5908 	}
5909 	return (0);
5910 }
5911 
5912 
5913 int
5914 ctl_aps_sp_handler(struct ctl_scsiio *ctsio,
5915 		   struct ctl_page_index *page_index, uint8_t *page_ptr)
5916 {
5917 	struct copan_aps_subpage *user_sp;
5918 	struct copan_aps_subpage *current_sp;
5919 	union ctl_modepage_info *modepage_info;
5920 	struct ctl_softc *softc;
5921 	struct ctl_lun *lun;
5922 	int retval;
5923 
5924 	retval = CTL_RETVAL_COMPLETE;
5925 	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
5926 		     (page_index->page_len * CTL_PAGE_CURRENT));
5927 	softc = control_softc;
5928 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5929 
5930 	user_sp = (struct copan_aps_subpage *)page_ptr;
5931 
5932 	modepage_info = (union ctl_modepage_info *)
5933 		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
5934 
5935 	modepage_info->header.page_code = page_index->page_code & SMPH_PC_MASK;
5936 	modepage_info->header.subpage = page_index->subpage;
5937 	modepage_info->aps.lock_active = user_sp->lock_active;
5938 
5939 	mtx_lock(&softc->ctl_lock);
5940 
5941 	/*
5942 	 * If there is a request to lock the LUN and another LUN is locked
5943 	 * this is an error. If the requested LUN is already locked ignore
5944 	 * the request. If no LUN is locked attempt to lock it.
5945 	 * if there is a request to unlock the LUN and the LUN is currently
5946 	 * locked attempt to unlock it. Otherwise ignore the request. i.e.
5947 	 * if another LUN is locked or no LUN is locked.
5948 	 */
5949 	if (user_sp->lock_active & APS_LOCK_ACTIVE) {
5950 		if (softc->aps_locked_lun == lun->lun) {
5951 			/*
5952 			 * This LUN is already locked, so we're done.
5953 			 */
5954 			retval = CTL_RETVAL_COMPLETE;
5955 		} else if (softc->aps_locked_lun == 0) {
5956 			/*
5957 			 * No one has the lock, pass the request to the
5958 			 * backend.
5959 			 */
5960 			retval = lun->backend->config_write(
5961 				(union ctl_io *)ctsio);
5962 		} else {
5963 			/*
5964 			 * Someone else has the lock, throw out the request.
5965 			 */
5966 			ctl_set_already_locked(ctsio);
5967 			free(ctsio->kern_data_ptr, M_CTL);
5968 			ctl_done((union ctl_io *)ctsio);
5969 
5970 			/*
5971 			 * Set the return value so that ctl_do_mode_select()
5972 			 * won't try to complete the command.  We already
5973 			 * completed it here.
5974 			 */
5975 			retval = CTL_RETVAL_ERROR;
5976 		}
5977 	} else if (softc->aps_locked_lun == lun->lun) {
5978 		/*
5979 		 * This LUN is locked, so pass the unlock request to the
5980 		 * backend.
5981 		 */
5982 		retval = lun->backend->config_write((union ctl_io *)ctsio);
5983 	}
5984 	mtx_unlock(&softc->ctl_lock);
5985 
5986 	return (retval);
5987 }
5988 
5989 int
5990 ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio,
5991 				struct ctl_page_index *page_index,
5992 				uint8_t *page_ptr)
5993 {
5994 	uint8_t *c;
5995 	int i;
5996 
5997 	c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs;
5998 	ctl_time_io_secs =
5999 		(c[0] << 8) |
6000 		(c[1] << 0) |
6001 		0;
6002 	CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs));
6003 	printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs);
6004 	printf("page data:");
6005 	for (i=0; i<8; i++)
6006 		printf(" %.2x",page_ptr[i]);
6007 	printf("\n");
6008 	return (0);
6009 }
6010 
6011 int
6012 ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio,
6013 			       struct ctl_page_index *page_index,
6014 			       int pc)
6015 {
6016 	struct copan_debugconf_subpage *page;
6017 
6018 	page = (struct copan_debugconf_subpage *)page_index->page_data +
6019 		(page_index->page_len * pc);
6020 
6021 	switch (pc) {
6022 	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6023 	case SMS_PAGE_CTRL_DEFAULT >> 6:
6024 	case SMS_PAGE_CTRL_SAVED >> 6:
6025 		/*
6026 		 * We don't update the changable or default bits for this page.
6027 		 */
6028 		break;
6029 	case SMS_PAGE_CTRL_CURRENT >> 6:
6030 		page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8;
6031 		page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0;
6032 		break;
6033 	default:
6034 #ifdef NEEDTOPORT
6035 		EPRINT(0, "Invalid PC %d!!", pc);
6036 #endif /* NEEDTOPORT */
6037 		break;
6038 	}
6039 	return (0);
6040 }
6041 
6042 
6043 static int
6044 ctl_do_mode_select(union ctl_io *io)
6045 {
6046 	struct scsi_mode_page_header *page_header;
6047 	struct ctl_page_index *page_index;
6048 	struct ctl_scsiio *ctsio;
6049 	int control_dev, page_len;
6050 	int page_len_offset, page_len_size;
6051 	union ctl_modepage_info *modepage_info;
6052 	struct ctl_lun *lun;
6053 	int *len_left, *len_used;
6054 	int retval, i;
6055 
6056 	ctsio = &io->scsiio;
6057 	page_index = NULL;
6058 	page_len = 0;
6059 	retval = CTL_RETVAL_COMPLETE;
6060 
6061 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6062 
6063 	if (lun->be_lun->lun_type != T_DIRECT)
6064 		control_dev = 1;
6065 	else
6066 		control_dev = 0;
6067 
6068 	modepage_info = (union ctl_modepage_info *)
6069 		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6070 	len_left = &modepage_info->header.len_left;
6071 	len_used = &modepage_info->header.len_used;
6072 
6073 do_next_page:
6074 
6075 	page_header = (struct scsi_mode_page_header *)
6076 		(ctsio->kern_data_ptr + *len_used);
6077 
6078 	if (*len_left == 0) {
6079 		free(ctsio->kern_data_ptr, M_CTL);
6080 		ctl_set_success(ctsio);
6081 		ctl_done((union ctl_io *)ctsio);
6082 		return (CTL_RETVAL_COMPLETE);
6083 	} else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6084 
6085 		free(ctsio->kern_data_ptr, M_CTL);
6086 		ctl_set_param_len_error(ctsio);
6087 		ctl_done((union ctl_io *)ctsio);
6088 		return (CTL_RETVAL_COMPLETE);
6089 
6090 	} else if ((page_header->page_code & SMPH_SPF)
6091 		&& (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6092 
6093 		free(ctsio->kern_data_ptr, M_CTL);
6094 		ctl_set_param_len_error(ctsio);
6095 		ctl_done((union ctl_io *)ctsio);
6096 		return (CTL_RETVAL_COMPLETE);
6097 	}
6098 
6099 
6100 	/*
6101 	 * XXX KDM should we do something with the block descriptor?
6102 	 */
6103 	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6104 
6105 		if ((control_dev != 0)
6106 		 && (lun->mode_pages.index[i].page_flags &
6107 		     CTL_PAGE_FLAG_DISK_ONLY))
6108 			continue;
6109 
6110 		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
6111 		    (page_header->page_code & SMPH_PC_MASK))
6112 			continue;
6113 
6114 		/*
6115 		 * If neither page has a subpage code, then we've got a
6116 		 * match.
6117 		 */
6118 		if (((lun->mode_pages.index[i].page_code & SMPH_SPF) == 0)
6119 		 && ((page_header->page_code & SMPH_SPF) == 0)) {
6120 			page_index = &lun->mode_pages.index[i];
6121 			page_len = page_header->page_length;
6122 			break;
6123 		}
6124 
6125 		/*
6126 		 * If both pages have subpages, then the subpage numbers
6127 		 * have to match.
6128 		 */
6129 		if ((lun->mode_pages.index[i].page_code & SMPH_SPF)
6130 		  && (page_header->page_code & SMPH_SPF)) {
6131 			struct scsi_mode_page_header_sp *sph;
6132 
6133 			sph = (struct scsi_mode_page_header_sp *)page_header;
6134 
6135 			if (lun->mode_pages.index[i].subpage ==
6136 			    sph->subpage) {
6137 				page_index = &lun->mode_pages.index[i];
6138 				page_len = scsi_2btoul(sph->page_length);
6139 				break;
6140 			}
6141 		}
6142 	}
6143 
6144 	/*
6145 	 * If we couldn't find the page, or if we don't have a mode select
6146 	 * handler for it, send back an error to the user.
6147 	 */
6148 	if ((page_index == NULL)
6149 	 || (page_index->select_handler == NULL)) {
6150 		ctl_set_invalid_field(ctsio,
6151 				      /*sks_valid*/ 1,
6152 				      /*command*/ 0,
6153 				      /*field*/ *len_used,
6154 				      /*bit_valid*/ 0,
6155 				      /*bit*/ 0);
6156 		free(ctsio->kern_data_ptr, M_CTL);
6157 		ctl_done((union ctl_io *)ctsio);
6158 		return (CTL_RETVAL_COMPLETE);
6159 	}
6160 
6161 	if (page_index->page_code & SMPH_SPF) {
6162 		page_len_offset = 2;
6163 		page_len_size = 2;
6164 	} else {
6165 		page_len_size = 1;
6166 		page_len_offset = 1;
6167 	}
6168 
6169 	/*
6170 	 * If the length the initiator gives us isn't the one we specify in
6171 	 * the mode page header, or if they didn't specify enough data in
6172 	 * the CDB to avoid truncating this page, kick out the request.
6173 	 */
6174 	if ((page_len != (page_index->page_len - page_len_offset -
6175 			  page_len_size))
6176 	 || (*len_left < page_index->page_len)) {
6177 
6178 
6179 		ctl_set_invalid_field(ctsio,
6180 				      /*sks_valid*/ 1,
6181 				      /*command*/ 0,
6182 				      /*field*/ *len_used + page_len_offset,
6183 				      /*bit_valid*/ 0,
6184 				      /*bit*/ 0);
6185 		free(ctsio->kern_data_ptr, M_CTL);
6186 		ctl_done((union ctl_io *)ctsio);
6187 		return (CTL_RETVAL_COMPLETE);
6188 	}
6189 
6190 	/*
6191 	 * Run through the mode page, checking to make sure that the bits
6192 	 * the user changed are actually legal for him to change.
6193 	 */
6194 	for (i = 0; i < page_index->page_len; i++) {
6195 		uint8_t *user_byte, *change_mask, *current_byte;
6196 		int bad_bit;
6197 		int j;
6198 
6199 		user_byte = (uint8_t *)page_header + i;
6200 		change_mask = page_index->page_data +
6201 			      (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6202 		current_byte = page_index->page_data +
6203 			       (page_index->page_len * CTL_PAGE_CURRENT) + i;
6204 
6205 		/*
6206 		 * Check to see whether the user set any bits in this byte
6207 		 * that he is not allowed to set.
6208 		 */
6209 		if ((*user_byte & ~(*change_mask)) ==
6210 		    (*current_byte & ~(*change_mask)))
6211 			continue;
6212 
6213 		/*
6214 		 * Go through bit by bit to determine which one is illegal.
6215 		 */
6216 		bad_bit = 0;
6217 		for (j = 7; j >= 0; j--) {
6218 			if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6219 			    (((1 << i) & ~(*change_mask)) & *current_byte)) {
6220 				bad_bit = i;
6221 				break;
6222 			}
6223 		}
6224 		ctl_set_invalid_field(ctsio,
6225 				      /*sks_valid*/ 1,
6226 				      /*command*/ 0,
6227 				      /*field*/ *len_used + i,
6228 				      /*bit_valid*/ 1,
6229 				      /*bit*/ bad_bit);
6230 		free(ctsio->kern_data_ptr, M_CTL);
6231 		ctl_done((union ctl_io *)ctsio);
6232 		return (CTL_RETVAL_COMPLETE);
6233 	}
6234 
6235 	/*
6236 	 * Decrement these before we call the page handler, since we may
6237 	 * end up getting called back one way or another before the handler
6238 	 * returns to this context.
6239 	 */
6240 	*len_left -= page_index->page_len;
6241 	*len_used += page_index->page_len;
6242 
6243 	retval = page_index->select_handler(ctsio, page_index,
6244 					    (uint8_t *)page_header);
6245 
6246 	/*
6247 	 * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6248 	 * wait until this queued command completes to finish processing
6249 	 * the mode page.  If it returns anything other than
6250 	 * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6251 	 * already set the sense information, freed the data pointer, and
6252 	 * completed the io for us.
6253 	 */
6254 	if (retval != CTL_RETVAL_COMPLETE)
6255 		goto bailout_no_done;
6256 
6257 	/*
6258 	 * If the initiator sent us more than one page, parse the next one.
6259 	 */
6260 	if (*len_left > 0)
6261 		goto do_next_page;
6262 
6263 	ctl_set_success(ctsio);
6264 	free(ctsio->kern_data_ptr, M_CTL);
6265 	ctl_done((union ctl_io *)ctsio);
6266 
6267 bailout_no_done:
6268 
6269 	return (CTL_RETVAL_COMPLETE);
6270 
6271 }
6272 
6273 int
6274 ctl_mode_select(struct ctl_scsiio *ctsio)
6275 {
6276 	int param_len, pf, sp;
6277 	int header_size, bd_len;
6278 	int len_left, len_used;
6279 	struct ctl_page_index *page_index;
6280 	struct ctl_lun *lun;
6281 	int control_dev, page_len;
6282 	union ctl_modepage_info *modepage_info;
6283 	int retval;
6284 
6285 	pf = 0;
6286 	sp = 0;
6287 	page_len = 0;
6288 	len_used = 0;
6289 	len_left = 0;
6290 	retval = 0;
6291 	bd_len = 0;
6292 	page_index = NULL;
6293 
6294 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6295 
6296 	if (lun->be_lun->lun_type != T_DIRECT)
6297 		control_dev = 1;
6298 	else
6299 		control_dev = 0;
6300 
6301 	switch (ctsio->cdb[0]) {
6302 	case MODE_SELECT_6: {
6303 		struct scsi_mode_select_6 *cdb;
6304 
6305 		cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6306 
6307 		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6308 		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6309 
6310 		param_len = cdb->length;
6311 		header_size = sizeof(struct scsi_mode_header_6);
6312 		break;
6313 	}
6314 	case MODE_SELECT_10: {
6315 		struct scsi_mode_select_10 *cdb;
6316 
6317 		cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6318 
6319 		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6320 		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6321 
6322 		param_len = scsi_2btoul(cdb->length);
6323 		header_size = sizeof(struct scsi_mode_header_10);
6324 		break;
6325 	}
6326 	default:
6327 		ctl_set_invalid_opcode(ctsio);
6328 		ctl_done((union ctl_io *)ctsio);
6329 		return (CTL_RETVAL_COMPLETE);
6330 		break; /* NOTREACHED */
6331 	}
6332 
6333 	/*
6334 	 * From SPC-3:
6335 	 * "A parameter list length of zero indicates that the Data-Out Buffer
6336 	 * shall be empty. This condition shall not be considered as an error."
6337 	 */
6338 	if (param_len == 0) {
6339 		ctl_set_success(ctsio);
6340 		ctl_done((union ctl_io *)ctsio);
6341 		return (CTL_RETVAL_COMPLETE);
6342 	}
6343 
6344 	/*
6345 	 * Since we'll hit this the first time through, prior to
6346 	 * allocation, we don't need to free a data buffer here.
6347 	 */
6348 	if (param_len < header_size) {
6349 		ctl_set_param_len_error(ctsio);
6350 		ctl_done((union ctl_io *)ctsio);
6351 		return (CTL_RETVAL_COMPLETE);
6352 	}
6353 
6354 	/*
6355 	 * Allocate the data buffer and grab the user's data.  In theory,
6356 	 * we shouldn't have to sanity check the parameter list length here
6357 	 * because the maximum size is 64K.  We should be able to malloc
6358 	 * that much without too many problems.
6359 	 */
6360 	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6361 		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6362 		ctsio->kern_data_len = param_len;
6363 		ctsio->kern_total_len = param_len;
6364 		ctsio->kern_data_resid = 0;
6365 		ctsio->kern_rel_offset = 0;
6366 		ctsio->kern_sg_entries = 0;
6367 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6368 		ctsio->be_move_done = ctl_config_move_done;
6369 		ctl_datamove((union ctl_io *)ctsio);
6370 
6371 		return (CTL_RETVAL_COMPLETE);
6372 	}
6373 
6374 	switch (ctsio->cdb[0]) {
6375 	case MODE_SELECT_6: {
6376 		struct scsi_mode_header_6 *mh6;
6377 
6378 		mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6379 		bd_len = mh6->blk_desc_len;
6380 		break;
6381 	}
6382 	case MODE_SELECT_10: {
6383 		struct scsi_mode_header_10 *mh10;
6384 
6385 		mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6386 		bd_len = scsi_2btoul(mh10->blk_desc_len);
6387 		break;
6388 	}
6389 	default:
6390 		panic("Invalid CDB type %#x", ctsio->cdb[0]);
6391 		break;
6392 	}
6393 
6394 	if (param_len < (header_size + bd_len)) {
6395 		free(ctsio->kern_data_ptr, M_CTL);
6396 		ctl_set_param_len_error(ctsio);
6397 		ctl_done((union ctl_io *)ctsio);
6398 		return (CTL_RETVAL_COMPLETE);
6399 	}
6400 
6401 	/*
6402 	 * Set the IO_CONT flag, so that if this I/O gets passed to
6403 	 * ctl_config_write_done(), it'll get passed back to
6404 	 * ctl_do_mode_select() for further processing, or completion if
6405 	 * we're all done.
6406 	 */
6407 	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6408 	ctsio->io_cont = ctl_do_mode_select;
6409 
6410 	modepage_info = (union ctl_modepage_info *)
6411 		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6412 
6413 	memset(modepage_info, 0, sizeof(*modepage_info));
6414 
6415 	len_left = param_len - header_size - bd_len;
6416 	len_used = header_size + bd_len;
6417 
6418 	modepage_info->header.len_left = len_left;
6419 	modepage_info->header.len_used = len_used;
6420 
6421 	return (ctl_do_mode_select((union ctl_io *)ctsio));
6422 }
6423 
6424 int
6425 ctl_mode_sense(struct ctl_scsiio *ctsio)
6426 {
6427 	struct ctl_lun *lun;
6428 	int pc, page_code, dbd, llba, subpage;
6429 	int alloc_len, page_len, header_len, total_len;
6430 	struct scsi_mode_block_descr *block_desc;
6431 	struct ctl_page_index *page_index;
6432 	int control_dev;
6433 
6434 	dbd = 0;
6435 	llba = 0;
6436 	block_desc = NULL;
6437 	page_index = NULL;
6438 
6439 	CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6440 
6441 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6442 
6443 	if (lun->be_lun->lun_type != T_DIRECT)
6444 		control_dev = 1;
6445 	else
6446 		control_dev = 0;
6447 
6448 	switch (ctsio->cdb[0]) {
6449 	case MODE_SENSE_6: {
6450 		struct scsi_mode_sense_6 *cdb;
6451 
6452 		cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6453 
6454 		header_len = sizeof(struct scsi_mode_hdr_6);
6455 		if (cdb->byte2 & SMS_DBD)
6456 			dbd = 1;
6457 		else
6458 			header_len += sizeof(struct scsi_mode_block_descr);
6459 
6460 		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6461 		page_code = cdb->page & SMS_PAGE_CODE;
6462 		subpage = cdb->subpage;
6463 		alloc_len = cdb->length;
6464 		break;
6465 	}
6466 	case MODE_SENSE_10: {
6467 		struct scsi_mode_sense_10 *cdb;
6468 
6469 		cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6470 
6471 		header_len = sizeof(struct scsi_mode_hdr_10);
6472 
6473 		if (cdb->byte2 & SMS_DBD)
6474 			dbd = 1;
6475 		else
6476 			header_len += sizeof(struct scsi_mode_block_descr);
6477 		if (cdb->byte2 & SMS10_LLBAA)
6478 			llba = 1;
6479 		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6480 		page_code = cdb->page & SMS_PAGE_CODE;
6481 		subpage = cdb->subpage;
6482 		alloc_len = scsi_2btoul(cdb->length);
6483 		break;
6484 	}
6485 	default:
6486 		ctl_set_invalid_opcode(ctsio);
6487 		ctl_done((union ctl_io *)ctsio);
6488 		return (CTL_RETVAL_COMPLETE);
6489 		break; /* NOTREACHED */
6490 	}
6491 
6492 	/*
6493 	 * We have to make a first pass through to calculate the size of
6494 	 * the pages that match the user's query.  Then we allocate enough
6495 	 * memory to hold it, and actually copy the data into the buffer.
6496 	 */
6497 	switch (page_code) {
6498 	case SMS_ALL_PAGES_PAGE: {
6499 		int i;
6500 
6501 		page_len = 0;
6502 
6503 		/*
6504 		 * At the moment, values other than 0 and 0xff here are
6505 		 * reserved according to SPC-3.
6506 		 */
6507 		if ((subpage != SMS_SUBPAGE_PAGE_0)
6508 		 && (subpage != SMS_SUBPAGE_ALL)) {
6509 			ctl_set_invalid_field(ctsio,
6510 					      /*sks_valid*/ 1,
6511 					      /*command*/ 1,
6512 					      /*field*/ 3,
6513 					      /*bit_valid*/ 0,
6514 					      /*bit*/ 0);
6515 			ctl_done((union ctl_io *)ctsio);
6516 			return (CTL_RETVAL_COMPLETE);
6517 		}
6518 
6519 		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6520 			if ((control_dev != 0)
6521 			 && (lun->mode_pages.index[i].page_flags &
6522 			     CTL_PAGE_FLAG_DISK_ONLY))
6523 				continue;
6524 
6525 			/*
6526 			 * We don't use this subpage if the user didn't
6527 			 * request all subpages.
6528 			 */
6529 			if ((lun->mode_pages.index[i].subpage != 0)
6530 			 && (subpage == SMS_SUBPAGE_PAGE_0))
6531 				continue;
6532 
6533 #if 0
6534 			printf("found page %#x len %d\n",
6535 			       lun->mode_pages.index[i].page_code &
6536 			       SMPH_PC_MASK,
6537 			       lun->mode_pages.index[i].page_len);
6538 #endif
6539 			page_len += lun->mode_pages.index[i].page_len;
6540 		}
6541 		break;
6542 	}
6543 	default: {
6544 		int i;
6545 
6546 		page_len = 0;
6547 
6548 		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6549 			/* Look for the right page code */
6550 			if ((lun->mode_pages.index[i].page_code &
6551 			     SMPH_PC_MASK) != page_code)
6552 				continue;
6553 
6554 			/* Look for the right subpage or the subpage wildcard*/
6555 			if ((lun->mode_pages.index[i].subpage != subpage)
6556 			 && (subpage != SMS_SUBPAGE_ALL))
6557 				continue;
6558 
6559 			/* Make sure the page is supported for this dev type */
6560 			if ((control_dev != 0)
6561 			 && (lun->mode_pages.index[i].page_flags &
6562 			     CTL_PAGE_FLAG_DISK_ONLY))
6563 				continue;
6564 
6565 #if 0
6566 			printf("found page %#x len %d\n",
6567 			       lun->mode_pages.index[i].page_code &
6568 			       SMPH_PC_MASK,
6569 			       lun->mode_pages.index[i].page_len);
6570 #endif
6571 
6572 			page_len += lun->mode_pages.index[i].page_len;
6573 		}
6574 
6575 		if (page_len == 0) {
6576 			ctl_set_invalid_field(ctsio,
6577 					      /*sks_valid*/ 1,
6578 					      /*command*/ 1,
6579 					      /*field*/ 2,
6580 					      /*bit_valid*/ 1,
6581 					      /*bit*/ 5);
6582 			ctl_done((union ctl_io *)ctsio);
6583 			return (CTL_RETVAL_COMPLETE);
6584 		}
6585 		break;
6586 	}
6587 	}
6588 
6589 	total_len = header_len + page_len;
6590 #if 0
6591 	printf("header_len = %d, page_len = %d, total_len = %d\n",
6592 	       header_len, page_len, total_len);
6593 #endif
6594 
6595 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6596 	ctsio->kern_sg_entries = 0;
6597 	ctsio->kern_data_resid = 0;
6598 	ctsio->kern_rel_offset = 0;
6599 	if (total_len < alloc_len) {
6600 		ctsio->residual = alloc_len - total_len;
6601 		ctsio->kern_data_len = total_len;
6602 		ctsio->kern_total_len = total_len;
6603 	} else {
6604 		ctsio->residual = 0;
6605 		ctsio->kern_data_len = alloc_len;
6606 		ctsio->kern_total_len = alloc_len;
6607 	}
6608 
6609 	switch (ctsio->cdb[0]) {
6610 	case MODE_SENSE_6: {
6611 		struct scsi_mode_hdr_6 *header;
6612 
6613 		header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6614 
6615 		header->datalen = ctl_min(total_len - 1, 254);
6616 
6617 		if (dbd)
6618 			header->block_descr_len = 0;
6619 		else
6620 			header->block_descr_len =
6621 				sizeof(struct scsi_mode_block_descr);
6622 		block_desc = (struct scsi_mode_block_descr *)&header[1];
6623 		break;
6624 	}
6625 	case MODE_SENSE_10: {
6626 		struct scsi_mode_hdr_10 *header;
6627 		int datalen;
6628 
6629 		header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6630 
6631 		datalen = ctl_min(total_len - 2, 65533);
6632 		scsi_ulto2b(datalen, header->datalen);
6633 		if (dbd)
6634 			scsi_ulto2b(0, header->block_descr_len);
6635 		else
6636 			scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
6637 				    header->block_descr_len);
6638 		block_desc = (struct scsi_mode_block_descr *)&header[1];
6639 		break;
6640 	}
6641 	default:
6642 		panic("invalid CDB type %#x", ctsio->cdb[0]);
6643 		break; /* NOTREACHED */
6644 	}
6645 
6646 	/*
6647 	 * If we've got a disk, use its blocksize in the block
6648 	 * descriptor.  Otherwise, just set it to 0.
6649 	 */
6650 	if (dbd == 0) {
6651 		if (control_dev != 0)
6652 			scsi_ulto3b(lun->be_lun->blocksize,
6653 				    block_desc->block_len);
6654 		else
6655 			scsi_ulto3b(0, block_desc->block_len);
6656 	}
6657 
6658 	switch (page_code) {
6659 	case SMS_ALL_PAGES_PAGE: {
6660 		int i, data_used;
6661 
6662 		data_used = header_len;
6663 		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6664 			struct ctl_page_index *page_index;
6665 
6666 			page_index = &lun->mode_pages.index[i];
6667 
6668 			if ((control_dev != 0)
6669 			 && (page_index->page_flags &
6670 			    CTL_PAGE_FLAG_DISK_ONLY))
6671 				continue;
6672 
6673 			/*
6674 			 * We don't use this subpage if the user didn't
6675 			 * request all subpages.  We already checked (above)
6676 			 * to make sure the user only specified a subpage
6677 			 * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
6678 			 */
6679 			if ((page_index->subpage != 0)
6680 			 && (subpage == SMS_SUBPAGE_PAGE_0))
6681 				continue;
6682 
6683 			/*
6684 			 * Call the handler, if it exists, to update the
6685 			 * page to the latest values.
6686 			 */
6687 			if (page_index->sense_handler != NULL)
6688 				page_index->sense_handler(ctsio, page_index,pc);
6689 
6690 			memcpy(ctsio->kern_data_ptr + data_used,
6691 			       page_index->page_data +
6692 			       (page_index->page_len * pc),
6693 			       page_index->page_len);
6694 			data_used += page_index->page_len;
6695 		}
6696 		break;
6697 	}
6698 	default: {
6699 		int i, data_used;
6700 
6701 		data_used = header_len;
6702 
6703 		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6704 			struct ctl_page_index *page_index;
6705 
6706 			page_index = &lun->mode_pages.index[i];
6707 
6708 			/* Look for the right page code */
6709 			if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6710 				continue;
6711 
6712 			/* Look for the right subpage or the subpage wildcard*/
6713 			if ((page_index->subpage != subpage)
6714 			 && (subpage != SMS_SUBPAGE_ALL))
6715 				continue;
6716 
6717 			/* Make sure the page is supported for this dev type */
6718 			if ((control_dev != 0)
6719 			 && (page_index->page_flags &
6720 			     CTL_PAGE_FLAG_DISK_ONLY))
6721 				continue;
6722 
6723 			/*
6724 			 * Call the handler, if it exists, to update the
6725 			 * page to the latest values.
6726 			 */
6727 			if (page_index->sense_handler != NULL)
6728 				page_index->sense_handler(ctsio, page_index,pc);
6729 
6730 			memcpy(ctsio->kern_data_ptr + data_used,
6731 			       page_index->page_data +
6732 			       (page_index->page_len * pc),
6733 			       page_index->page_len);
6734 			data_used += page_index->page_len;
6735 		}
6736 		break;
6737 	}
6738 	}
6739 
6740 	ctsio->scsi_status = SCSI_STATUS_OK;
6741 
6742 	ctsio->be_move_done = ctl_config_move_done;
6743 	ctl_datamove((union ctl_io *)ctsio);
6744 
6745 	return (CTL_RETVAL_COMPLETE);
6746 }
6747 
6748 int
6749 ctl_read_capacity(struct ctl_scsiio *ctsio)
6750 {
6751 	struct scsi_read_capacity *cdb;
6752 	struct scsi_read_capacity_data *data;
6753 	struct ctl_lun *lun;
6754 	uint32_t lba;
6755 
6756 	CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
6757 
6758 	cdb = (struct scsi_read_capacity *)ctsio->cdb;
6759 
6760 	lba = scsi_4btoul(cdb->addr);
6761 	if (((cdb->pmi & SRC_PMI) == 0)
6762 	 && (lba != 0)) {
6763 		ctl_set_invalid_field(/*ctsio*/ ctsio,
6764 				      /*sks_valid*/ 1,
6765 				      /*command*/ 1,
6766 				      /*field*/ 2,
6767 				      /*bit_valid*/ 0,
6768 				      /*bit*/ 0);
6769 		ctl_done((union ctl_io *)ctsio);
6770 		return (CTL_RETVAL_COMPLETE);
6771 	}
6772 
6773 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6774 
6775 	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6776 	data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
6777 	ctsio->residual = 0;
6778 	ctsio->kern_data_len = sizeof(*data);
6779 	ctsio->kern_total_len = sizeof(*data);
6780 	ctsio->kern_data_resid = 0;
6781 	ctsio->kern_rel_offset = 0;
6782 	ctsio->kern_sg_entries = 0;
6783 
6784 	/*
6785 	 * If the maximum LBA is greater than 0xfffffffe, the user must
6786 	 * issue a SERVICE ACTION IN (16) command, with the read capacity
6787 	 * serivce action set.
6788 	 */
6789 	if (lun->be_lun->maxlba > 0xfffffffe)
6790 		scsi_ulto4b(0xffffffff, data->addr);
6791 	else
6792 		scsi_ulto4b(lun->be_lun->maxlba, data->addr);
6793 
6794 	/*
6795 	 * XXX KDM this may not be 512 bytes...
6796 	 */
6797 	scsi_ulto4b(lun->be_lun->blocksize, data->length);
6798 
6799 	ctsio->scsi_status = SCSI_STATUS_OK;
6800 
6801 	ctsio->be_move_done = ctl_config_move_done;
6802 	ctl_datamove((union ctl_io *)ctsio);
6803 
6804 	return (CTL_RETVAL_COMPLETE);
6805 }
6806 
6807 static int
6808 ctl_read_capacity_16(struct ctl_scsiio *ctsio)
6809 {
6810 	struct scsi_read_capacity_16 *cdb;
6811 	struct scsi_read_capacity_data_long *data;
6812 	struct ctl_lun *lun;
6813 	uint64_t lba;
6814 	uint32_t alloc_len;
6815 
6816 	CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
6817 
6818 	cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
6819 
6820 	alloc_len = scsi_4btoul(cdb->alloc_len);
6821 	lba = scsi_8btou64(cdb->addr);
6822 
6823 	if ((cdb->reladr & SRC16_PMI)
6824 	 && (lba != 0)) {
6825 		ctl_set_invalid_field(/*ctsio*/ ctsio,
6826 				      /*sks_valid*/ 1,
6827 				      /*command*/ 1,
6828 				      /*field*/ 2,
6829 				      /*bit_valid*/ 0,
6830 				      /*bit*/ 0);
6831 		ctl_done((union ctl_io *)ctsio);
6832 		return (CTL_RETVAL_COMPLETE);
6833 	}
6834 
6835 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6836 
6837 	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6838 	data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
6839 
6840 	if (sizeof(*data) < alloc_len) {
6841 		ctsio->residual = alloc_len - sizeof(*data);
6842 		ctsio->kern_data_len = sizeof(*data);
6843 		ctsio->kern_total_len = sizeof(*data);
6844 	} else {
6845 		ctsio->residual = 0;
6846 		ctsio->kern_data_len = alloc_len;
6847 		ctsio->kern_total_len = alloc_len;
6848 	}
6849 	ctsio->kern_data_resid = 0;
6850 	ctsio->kern_rel_offset = 0;
6851 	ctsio->kern_sg_entries = 0;
6852 
6853 	scsi_u64to8b(lun->be_lun->maxlba, data->addr);
6854 	/* XXX KDM this may not be 512 bytes... */
6855 	scsi_ulto4b(lun->be_lun->blocksize, data->length);
6856 
6857 	ctsio->scsi_status = SCSI_STATUS_OK;
6858 
6859 	ctsio->be_move_done = ctl_config_move_done;
6860 	ctl_datamove((union ctl_io *)ctsio);
6861 
6862 	return (CTL_RETVAL_COMPLETE);
6863 }
6864 
6865 int
6866 ctl_service_action_in(struct ctl_scsiio *ctsio)
6867 {
6868 	struct scsi_service_action_in *cdb;
6869 	int retval;
6870 
6871 	CTL_DEBUG_PRINT(("ctl_service_action_in\n"));
6872 
6873 	cdb = (struct scsi_service_action_in *)ctsio->cdb;
6874 
6875 	retval = CTL_RETVAL_COMPLETE;
6876 
6877 	switch (cdb->service_action) {
6878 	case SRC16_SERVICE_ACTION:
6879 		retval = ctl_read_capacity_16(ctsio);
6880 		break;
6881 	default:
6882 		ctl_set_invalid_field(/*ctsio*/ ctsio,
6883 				      /*sks_valid*/ 1,
6884 				      /*command*/ 1,
6885 				      /*field*/ 1,
6886 				      /*bit_valid*/ 1,
6887 				      /*bit*/ 4);
6888 		ctl_done((union ctl_io *)ctsio);
6889 		break;
6890 	}
6891 
6892 	return (retval);
6893 }
6894 
6895 int
6896 ctl_maintenance_in(struct ctl_scsiio *ctsio)
6897 {
6898 	struct scsi_maintenance_in *cdb;
6899 	int retval;
6900 	int alloc_len, total_len = 0;
6901 	int num_target_port_groups;
6902 	struct ctl_lun *lun;
6903 	struct ctl_softc *softc;
6904 	struct scsi_target_group_data *rtg_ptr;
6905 	struct scsi_target_port_group_descriptor *tpg_desc_ptr1, *tpg_desc_ptr2;
6906 	struct scsi_target_port_descriptor  *tp_desc_ptr1_1, *tp_desc_ptr1_2,
6907 	                                    *tp_desc_ptr2_1, *tp_desc_ptr2_2;
6908 
6909 	CTL_DEBUG_PRINT(("ctl_maintenance_in\n"));
6910 
6911 	cdb = (struct scsi_maintenance_in *)ctsio->cdb;
6912 	softc = control_softc;
6913 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6914 
6915 	retval = CTL_RETVAL_COMPLETE;
6916 	mtx_lock(&softc->ctl_lock);
6917 
6918 	if ((cdb->byte2 & SERVICE_ACTION_MASK) != SA_RPRT_TRGT_GRP) {
6919 		ctl_set_invalid_field(/*ctsio*/ ctsio,
6920 				      /*sks_valid*/ 1,
6921 				      /*command*/ 1,
6922 				      /*field*/ 1,
6923 				      /*bit_valid*/ 1,
6924 				      /*bit*/ 4);
6925 		ctl_done((union ctl_io *)ctsio);
6926 		return(retval);
6927 	}
6928 
6929 	if (ctl_is_single)
6930         	num_target_port_groups = NUM_TARGET_PORT_GROUPS - 1;
6931 	else
6932         	num_target_port_groups = NUM_TARGET_PORT_GROUPS;
6933 
6934 	total_len = sizeof(struct scsi_target_group_data) +
6935 		sizeof(struct scsi_target_port_group_descriptor) *
6936 		num_target_port_groups +
6937 		sizeof(struct scsi_target_port_descriptor) *
6938 		NUM_PORTS_PER_GRP * num_target_port_groups;
6939 
6940 	alloc_len = scsi_4btoul(cdb->length);
6941 
6942 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6943 
6944 	ctsio->kern_sg_entries = 0;
6945 
6946 	if (total_len < alloc_len) {
6947 		ctsio->residual = alloc_len - total_len;
6948 		ctsio->kern_data_len = total_len;
6949 		ctsio->kern_total_len = total_len;
6950 	} else {
6951 		ctsio->residual = 0;
6952 		ctsio->kern_data_len = alloc_len;
6953 		ctsio->kern_total_len = alloc_len;
6954 	}
6955 	ctsio->kern_data_resid = 0;
6956 	ctsio->kern_rel_offset = 0;
6957 
6958 	rtg_ptr = (struct scsi_target_group_data *)ctsio->kern_data_ptr;
6959 
6960 	tpg_desc_ptr1 = &rtg_ptr->groups[0];
6961 	tp_desc_ptr1_1 = &tpg_desc_ptr1->descriptors[0];
6962 	tp_desc_ptr1_2 = (struct scsi_target_port_descriptor *)
6963 	        &tp_desc_ptr1_1->desc_list[0];
6964 
6965 
6966 
6967 	if (ctl_is_single == 0) {
6968 		tpg_desc_ptr2 = (struct scsi_target_port_group_descriptor *)
6969 	                &tp_desc_ptr1_2->desc_list[0];
6970 		tp_desc_ptr2_1 = &tpg_desc_ptr2->descriptors[0];
6971 		tp_desc_ptr2_2 = (struct scsi_target_port_descriptor *)
6972 	        	&tp_desc_ptr2_1->desc_list[0];
6973         } else {
6974 		tpg_desc_ptr2 = NULL;
6975 		tp_desc_ptr2_1 = NULL;
6976 		tp_desc_ptr2_2 = NULL;
6977 	}
6978 
6979 	scsi_ulto4b(total_len - 4, rtg_ptr->length);
6980 	if (ctl_is_single == 0) {
6981         	if (ctsio->io_hdr.nexus.targ_port < CTL_MAX_PORTS) {
6982 			if (lun->flags & CTL_LUN_PRIMARY_SC) {
6983 				tpg_desc_ptr1->pref_state = TPG_PRIMARY;
6984 				tpg_desc_ptr2->pref_state =
6985 					TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
6986 			} else {
6987 				tpg_desc_ptr1->pref_state =
6988 					TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
6989 				tpg_desc_ptr2->pref_state = TPG_PRIMARY;
6990 			}
6991 		} else {
6992 			if (lun->flags & CTL_LUN_PRIMARY_SC) {
6993 				tpg_desc_ptr1->pref_state =
6994 					TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
6995 				tpg_desc_ptr2->pref_state = TPG_PRIMARY;
6996 			} else {
6997 				tpg_desc_ptr1->pref_state = TPG_PRIMARY;
6998 				tpg_desc_ptr2->pref_state =
6999 					TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7000 			}
7001 		}
7002 	} else {
7003 		tpg_desc_ptr1->pref_state = TPG_PRIMARY;
7004 	}
7005 	tpg_desc_ptr1->support = 0;
7006 	tpg_desc_ptr1->target_port_group[1] = 1;
7007 	tpg_desc_ptr1->status = TPG_IMPLICIT;
7008 	tpg_desc_ptr1->target_port_count= NUM_PORTS_PER_GRP;
7009 
7010 	if (ctl_is_single == 0) {
7011 		tpg_desc_ptr2->support = 0;
7012 		tpg_desc_ptr2->target_port_group[1] = 2;
7013 		tpg_desc_ptr2->status = TPG_IMPLICIT;
7014 		tpg_desc_ptr2->target_port_count = NUM_PORTS_PER_GRP;
7015 
7016 		tp_desc_ptr1_1->relative_target_port_identifier[1] = 1;
7017 		tp_desc_ptr1_2->relative_target_port_identifier[1] = 2;
7018 
7019 		tp_desc_ptr2_1->relative_target_port_identifier[1] = 9;
7020 		tp_desc_ptr2_2->relative_target_port_identifier[1] = 10;
7021 	} else {
7022         	if (ctsio->io_hdr.nexus.targ_port < CTL_MAX_PORTS) {
7023 			tp_desc_ptr1_1->relative_target_port_identifier[1] = 1;
7024 			tp_desc_ptr1_2->relative_target_port_identifier[1] = 2;
7025 		} else {
7026 			tp_desc_ptr1_1->relative_target_port_identifier[1] = 9;
7027 			tp_desc_ptr1_2->relative_target_port_identifier[1] = 10;
7028 		}
7029 	}
7030 
7031 	mtx_unlock(&softc->ctl_lock);
7032 
7033 	ctsio->be_move_done = ctl_config_move_done;
7034 
7035 	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7036 			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7037 			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7038 			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7039 			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7040 
7041 	ctl_datamove((union ctl_io *)ctsio);
7042 	return(retval);
7043 }
7044 
7045 int
7046 ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7047 {
7048 	struct scsi_per_res_in *cdb;
7049 	int alloc_len, total_len = 0;
7050 	/* struct scsi_per_res_in_rsrv in_data; */
7051 	struct ctl_lun *lun;
7052 	struct ctl_softc *softc;
7053 
7054 	CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7055 
7056 	softc = control_softc;
7057 
7058 	cdb = (struct scsi_per_res_in *)ctsio->cdb;
7059 
7060 	alloc_len = scsi_2btoul(cdb->length);
7061 
7062 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7063 
7064 retry:
7065 	mtx_lock(&softc->ctl_lock);
7066 	switch (cdb->action) {
7067 	case SPRI_RK: /* read keys */
7068 		total_len = sizeof(struct scsi_per_res_in_keys) +
7069 			lun->pr_key_count *
7070 			sizeof(struct scsi_per_res_key);
7071 		break;
7072 	case SPRI_RR: /* read reservation */
7073 		if (lun->flags & CTL_LUN_PR_RESERVED)
7074 			total_len = sizeof(struct scsi_per_res_in_rsrv);
7075 		else
7076 			total_len = sizeof(struct scsi_per_res_in_header);
7077 		break;
7078 	case SPRI_RC: /* report capabilities */
7079 		total_len = sizeof(struct scsi_per_res_cap);
7080 		break;
7081 	case SPRI_RS: /* read full status */
7082 	default:
7083 		mtx_unlock(&softc->ctl_lock);
7084 		ctl_set_invalid_field(ctsio,
7085 				      /*sks_valid*/ 1,
7086 				      /*command*/ 1,
7087 				      /*field*/ 1,
7088 				      /*bit_valid*/ 1,
7089 				      /*bit*/ 0);
7090 		ctl_done((union ctl_io *)ctsio);
7091 		return (CTL_RETVAL_COMPLETE);
7092 		break; /* NOTREACHED */
7093 	}
7094 	mtx_unlock(&softc->ctl_lock);
7095 
7096 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7097 
7098 	if (total_len < alloc_len) {
7099 		ctsio->residual = alloc_len - total_len;
7100 		ctsio->kern_data_len = total_len;
7101 		ctsio->kern_total_len = total_len;
7102 	} else {
7103 		ctsio->residual = 0;
7104 		ctsio->kern_data_len = alloc_len;
7105 		ctsio->kern_total_len = alloc_len;
7106 	}
7107 
7108 	ctsio->kern_data_resid = 0;
7109 	ctsio->kern_rel_offset = 0;
7110 	ctsio->kern_sg_entries = 0;
7111 
7112 	mtx_lock(&softc->ctl_lock);
7113 	switch (cdb->action) {
7114 	case SPRI_RK: { // read keys
7115         struct scsi_per_res_in_keys *res_keys;
7116 		int i, key_count;
7117 
7118 		res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7119 
7120 		/*
7121 		 * We had to drop the lock to allocate our buffer, which
7122 		 * leaves time for someone to come in with another
7123 		 * persistent reservation.  (That is unlikely, though,
7124 		 * since this should be the only persistent reservation
7125 		 * command active right now.)
7126 		 */
7127 		if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7128 		    (lun->pr_key_count *
7129 		     sizeof(struct scsi_per_res_key)))){
7130 			mtx_unlock(&softc->ctl_lock);
7131 			free(ctsio->kern_data_ptr, M_CTL);
7132 			printf("%s: reservation length changed, retrying\n",
7133 			       __func__);
7134 			goto retry;
7135 		}
7136 
7137 		scsi_ulto4b(lun->PRGeneration, res_keys->header.generation);
7138 
7139 		scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7140 			     lun->pr_key_count, res_keys->header.length);
7141 
7142 		for (i = 0, key_count = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7143 			if (!lun->per_res[i].registered)
7144 				continue;
7145 
7146 			/*
7147 			 * We used lun->pr_key_count to calculate the
7148 			 * size to allocate.  If it turns out the number of
7149 			 * initiators with the registered flag set is
7150 			 * larger than that (i.e. they haven't been kept in
7151 			 * sync), we've got a problem.
7152 			 */
7153 			if (key_count >= lun->pr_key_count) {
7154 #ifdef NEEDTOPORT
7155 				csevent_log(CSC_CTL | CSC_SHELF_SW |
7156 					    CTL_PR_ERROR,
7157 					    csevent_LogType_Fault,
7158 					    csevent_AlertLevel_Yellow,
7159 					    csevent_FRU_ShelfController,
7160 					    csevent_FRU_Firmware,
7161 				        csevent_FRU_Unknown,
7162 					    "registered keys %d >= key "
7163 					    "count %d", key_count,
7164 					    lun->pr_key_count);
7165 #endif
7166 				key_count++;
7167 				continue;
7168 			}
7169 			memcpy(res_keys->keys[key_count].key,
7170 			       lun->per_res[i].res_key.key,
7171 			       ctl_min(sizeof(res_keys->keys[key_count].key),
7172 			       sizeof(lun->per_res[i].res_key)));
7173 			key_count++;
7174 		}
7175 		break;
7176 	}
7177 	case SPRI_RR: { // read reservation
7178 		struct scsi_per_res_in_rsrv *res;
7179 		int tmp_len, header_only;
7180 
7181 		res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7182 
7183 		scsi_ulto4b(lun->PRGeneration, res->header.generation);
7184 
7185 		if (lun->flags & CTL_LUN_PR_RESERVED)
7186 		{
7187 			tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7188 			scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7189 				    res->header.length);
7190 			header_only = 0;
7191 		} else {
7192 			tmp_len = sizeof(struct scsi_per_res_in_header);
7193 			scsi_ulto4b(0, res->header.length);
7194 			header_only = 1;
7195 		}
7196 
7197 		/*
7198 		 * We had to drop the lock to allocate our buffer, which
7199 		 * leaves time for someone to come in with another
7200 		 * persistent reservation.  (That is unlikely, though,
7201 		 * since this should be the only persistent reservation
7202 		 * command active right now.)
7203 		 */
7204 		if (tmp_len != total_len) {
7205 			mtx_unlock(&softc->ctl_lock);
7206 			free(ctsio->kern_data_ptr, M_CTL);
7207 			printf("%s: reservation status changed, retrying\n",
7208 			       __func__);
7209 			goto retry;
7210 		}
7211 
7212 		/*
7213 		 * No reservation held, so we're done.
7214 		 */
7215 		if (header_only != 0)
7216 			break;
7217 
7218 		/*
7219 		 * If the registration is an All Registrants type, the key
7220 		 * is 0, since it doesn't really matter.
7221 		 */
7222 		if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7223 			memcpy(res->data.reservation,
7224 			       &lun->per_res[lun->pr_res_idx].res_key,
7225 			       sizeof(struct scsi_per_res_key));
7226 		}
7227 		res->data.scopetype = lun->res_type;
7228 		break;
7229 	}
7230 	case SPRI_RC:     //report capabilities
7231 	{
7232 		struct scsi_per_res_cap *res_cap;
7233 		uint16_t type_mask;
7234 
7235 		res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7236 		scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7237 		res_cap->flags2 |= SPRI_TMV;
7238 		type_mask = SPRI_TM_WR_EX_AR |
7239 			    SPRI_TM_EX_AC_RO |
7240 			    SPRI_TM_WR_EX_RO |
7241 			    SPRI_TM_EX_AC |
7242 			    SPRI_TM_WR_EX |
7243 			    SPRI_TM_EX_AC_AR;
7244 		scsi_ulto2b(type_mask, res_cap->type_mask);
7245 		break;
7246 	}
7247 	case SPRI_RS: //read full status
7248 	default:
7249 		/*
7250 		 * This is a bug, because we just checked for this above,
7251 		 * and should have returned an error.
7252 		 */
7253 		panic("Invalid PR type %x", cdb->action);
7254 		break; /* NOTREACHED */
7255 	}
7256 	mtx_unlock(&softc->ctl_lock);
7257 
7258 	ctsio->be_move_done = ctl_config_move_done;
7259 
7260 	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7261 			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7262 			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7263 			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7264 			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7265 
7266 	ctl_datamove((union ctl_io *)ctsio);
7267 
7268 	return (CTL_RETVAL_COMPLETE);
7269 }
7270 
7271 /*
7272  * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7273  * it should return.
7274  */
7275 static int
7276 ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7277 		uint64_t sa_res_key, uint8_t type, uint32_t residx,
7278 		struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7279 		struct scsi_per_res_out_parms* param)
7280 {
7281 	union ctl_ha_msg persis_io;
7282 	int retval, i;
7283 	int isc_retval;
7284 
7285 	retval = 0;
7286 
7287 	if (sa_res_key == 0) {
7288 		mtx_lock(&softc->ctl_lock);
7289 		if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7290 			/* validate scope and type */
7291 			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7292 			     SPR_LU_SCOPE) {
7293 				mtx_unlock(&softc->ctl_lock);
7294 				ctl_set_invalid_field(/*ctsio*/ ctsio,
7295 						      /*sks_valid*/ 1,
7296 						      /*command*/ 1,
7297 						      /*field*/ 2,
7298 						      /*bit_valid*/ 1,
7299 						      /*bit*/ 4);
7300 				ctl_done((union ctl_io *)ctsio);
7301 				return (1);
7302 			}
7303 
7304 		        if (type>8 || type==2 || type==4 || type==0) {
7305 				mtx_unlock(&softc->ctl_lock);
7306 				ctl_set_invalid_field(/*ctsio*/ ctsio,
7307        	           				      /*sks_valid*/ 1,
7308 						      /*command*/ 1,
7309 						      /*field*/ 2,
7310 						      /*bit_valid*/ 1,
7311 						      /*bit*/ 0);
7312 				ctl_done((union ctl_io *)ctsio);
7313 				return (1);
7314 		        }
7315 
7316 			/* temporarily unregister this nexus */
7317 			lun->per_res[residx].registered = 0;
7318 
7319 			/*
7320 			 * Unregister everybody else and build UA for
7321 			 * them
7322 			 */
7323 			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7324 				if (lun->per_res[i].registered == 0)
7325 					continue;
7326 
7327 				if (!persis_offset
7328 				 && i <CTL_MAX_INITIATORS)
7329 					lun->pending_sense[i].ua_pending |=
7330 						CTL_UA_REG_PREEMPT;
7331 				else if (persis_offset
7332 				      && i >= persis_offset)
7333 					lun->pending_sense[i-persis_offset
7334 						].ua_pending |=
7335 						CTL_UA_REG_PREEMPT;
7336 				lun->per_res[i].registered = 0;
7337 				memset(&lun->per_res[i].res_key, 0,
7338 				       sizeof(struct scsi_per_res_key));
7339 			}
7340 			lun->per_res[residx].registered = 1;
7341 			lun->pr_key_count = 1;
7342 			lun->res_type = type;
7343 			if (lun->res_type != SPR_TYPE_WR_EX_AR
7344 			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7345 				lun->pr_res_idx = residx;
7346 
7347 			mtx_unlock(&softc->ctl_lock);
7348 			/* send msg to other side */
7349 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7350 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7351 			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7352 			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7353 			persis_io.pr.pr_info.res_type = type;
7354 			memcpy(persis_io.pr.pr_info.sa_res_key,
7355 			       param->serv_act_res_key,
7356 			       sizeof(param->serv_act_res_key));
7357 			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7358 			     &persis_io, sizeof(persis_io), 0)) >
7359 			     CTL_HA_STATUS_SUCCESS) {
7360 				printf("CTL:Persis Out error returned "
7361 				       "from ctl_ha_msg_send %d\n",
7362 				       isc_retval);
7363 			}
7364 		} else {
7365 			/* not all registrants */
7366 			mtx_unlock(&softc->ctl_lock);
7367 			free(ctsio->kern_data_ptr, M_CTL);
7368 			ctl_set_invalid_field(ctsio,
7369 					      /*sks_valid*/ 1,
7370 					      /*command*/ 0,
7371 					      /*field*/ 8,
7372 					      /*bit_valid*/ 0,
7373 					      /*bit*/ 0);
7374 			ctl_done((union ctl_io *)ctsio);
7375 			return (1);
7376 		}
7377 	} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7378 		|| !(lun->flags & CTL_LUN_PR_RESERVED)) {
7379 		int found = 0;
7380 
7381 		mtx_lock(&softc->ctl_lock);
7382 		if (res_key == sa_res_key) {
7383 			/* special case */
7384 			/*
7385 			 * The spec implies this is not good but doesn't
7386 			 * say what to do. There are two choices either
7387 			 * generate a res conflict or check condition
7388 			 * with illegal field in parameter data. Since
7389 			 * that is what is done when the sa_res_key is
7390 			 * zero I'll take that approach since this has
7391 			 * to do with the sa_res_key.
7392 			 */
7393 			mtx_unlock(&softc->ctl_lock);
7394 			free(ctsio->kern_data_ptr, M_CTL);
7395 			ctl_set_invalid_field(ctsio,
7396 					      /*sks_valid*/ 1,
7397 					      /*command*/ 0,
7398 					      /*field*/ 8,
7399 					      /*bit_valid*/ 0,
7400 					      /*bit*/ 0);
7401 			ctl_done((union ctl_io *)ctsio);
7402 			return (1);
7403 		}
7404 
7405 		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7406 			if (lun->per_res[i].registered
7407 			 && memcmp(param->serv_act_res_key,
7408 			    lun->per_res[i].res_key.key,
7409 			    sizeof(struct scsi_per_res_key)) != 0)
7410 				continue;
7411 
7412 			found = 1;
7413 			lun->per_res[i].registered = 0;
7414 			memset(&lun->per_res[i].res_key, 0,
7415 			       sizeof(struct scsi_per_res_key));
7416 			lun->pr_key_count--;
7417 
7418 			if (!persis_offset
7419 			 && i < CTL_MAX_INITIATORS)
7420 				lun->pending_sense[i].ua_pending |=
7421 					CTL_UA_REG_PREEMPT;
7422 			else if (persis_offset
7423 			      && i >= persis_offset)
7424 				lun->pending_sense[i-persis_offset].ua_pending|=
7425 					CTL_UA_REG_PREEMPT;
7426 		}
7427 		mtx_unlock(&softc->ctl_lock);
7428 		if (!found) {
7429 			free(ctsio->kern_data_ptr, M_CTL);
7430 			ctl_set_reservation_conflict(ctsio);
7431 			ctl_done((union ctl_io *)ctsio);
7432 			return (CTL_RETVAL_COMPLETE);
7433 		}
7434 		/* send msg to other side */
7435 		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7436 		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7437 		persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7438 		persis_io.pr.pr_info.residx = lun->pr_res_idx;
7439 		persis_io.pr.pr_info.res_type = type;
7440 		memcpy(persis_io.pr.pr_info.sa_res_key,
7441 		       param->serv_act_res_key,
7442 		       sizeof(param->serv_act_res_key));
7443 		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7444 		     &persis_io, sizeof(persis_io), 0)) >
7445 		     CTL_HA_STATUS_SUCCESS) {
7446 			printf("CTL:Persis Out error returned from "
7447 			       "ctl_ha_msg_send %d\n", isc_retval);
7448 		}
7449 	} else {
7450 		/* Reserved but not all registrants */
7451 		/* sa_res_key is res holder */
7452 		if (memcmp(param->serv_act_res_key,
7453                    lun->per_res[lun->pr_res_idx].res_key.key,
7454                    sizeof(struct scsi_per_res_key)) == 0) {
7455 			/* validate scope and type */
7456 			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7457 			     SPR_LU_SCOPE) {
7458 				ctl_set_invalid_field(/*ctsio*/ ctsio,
7459 						      /*sks_valid*/ 1,
7460 						      /*command*/ 1,
7461 						      /*field*/ 2,
7462 						      /*bit_valid*/ 1,
7463 						      /*bit*/ 4);
7464 				ctl_done((union ctl_io *)ctsio);
7465 				return (1);
7466 			}
7467 
7468 			if (type>8 || type==2 || type==4 || type==0) {
7469 				ctl_set_invalid_field(/*ctsio*/ ctsio,
7470 						      /*sks_valid*/ 1,
7471 						      /*command*/ 1,
7472 						      /*field*/ 2,
7473 						      /*bit_valid*/ 1,
7474 						      /*bit*/ 0);
7475 				ctl_done((union ctl_io *)ctsio);
7476 				return (1);
7477 			}
7478 
7479 			/*
7480 			 * Do the following:
7481 			 * if sa_res_key != res_key remove all
7482 			 * registrants w/sa_res_key and generate UA
7483 			 * for these registrants(Registrations
7484 			 * Preempted) if it wasn't an exclusive
7485 			 * reservation generate UA(Reservations
7486 			 * Preempted) for all other registered nexuses
7487 			 * if the type has changed. Establish the new
7488 			 * reservation and holder. If res_key and
7489 			 * sa_res_key are the same do the above
7490 			 * except don't unregister the res holder.
7491 			 */
7492 
7493 			/*
7494 			 * Temporarily unregister so it won't get
7495 			 * removed or UA generated
7496 			 */
7497 			lun->per_res[residx].registered = 0;
7498 			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7499 				if (lun->per_res[i].registered == 0)
7500 					continue;
7501 
7502 				if (memcmp(param->serv_act_res_key,
7503 				    lun->per_res[i].res_key.key,
7504 				    sizeof(struct scsi_per_res_key)) == 0) {
7505 					lun->per_res[i].registered = 0;
7506 					memset(&lun->per_res[i].res_key,
7507 					       0,
7508 					       sizeof(struct scsi_per_res_key));
7509 					lun->pr_key_count--;
7510 
7511 					if (!persis_offset
7512 					 && i < CTL_MAX_INITIATORS)
7513 						lun->pending_sense[i
7514 							].ua_pending |=
7515 							CTL_UA_REG_PREEMPT;
7516 					else if (persis_offset
7517 					      && i >= persis_offset)
7518 						lun->pending_sense[
7519 						  i-persis_offset].ua_pending |=
7520 						  CTL_UA_REG_PREEMPT;
7521 				} else if (type != lun->res_type
7522 					&& (lun->res_type == SPR_TYPE_WR_EX_RO
7523 					 || lun->res_type ==SPR_TYPE_EX_AC_RO)){
7524 						if (!persis_offset
7525 						 && i < CTL_MAX_INITIATORS)
7526 							lun->pending_sense[i
7527 							].ua_pending |=
7528 							CTL_UA_RES_RELEASE;
7529 						else if (persis_offset
7530 						      && i >= persis_offset)
7531 							lun->pending_sense[
7532 							i-persis_offset
7533 							].ua_pending |=
7534 							CTL_UA_RES_RELEASE;
7535 				}
7536 			}
7537 			lun->per_res[residx].registered = 1;
7538 			lun->res_type = type;
7539 			if (lun->res_type != SPR_TYPE_WR_EX_AR
7540 			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7541 				lun->pr_res_idx = residx;
7542 			else
7543 				lun->pr_res_idx =
7544 					CTL_PR_ALL_REGISTRANTS;
7545 
7546 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7547 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7548 			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7549 			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7550 			persis_io.pr.pr_info.res_type = type;
7551 			memcpy(persis_io.pr.pr_info.sa_res_key,
7552 			       param->serv_act_res_key,
7553 			       sizeof(param->serv_act_res_key));
7554 			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7555 			     &persis_io, sizeof(persis_io), 0)) >
7556 			     CTL_HA_STATUS_SUCCESS) {
7557 				printf("CTL:Persis Out error returned "
7558 				       "from ctl_ha_msg_send %d\n",
7559 				       isc_retval);
7560 			}
7561 		} else {
7562 			/*
7563 			 * sa_res_key is not the res holder just
7564 			 * remove registrants
7565 			 */
7566 			int found=0;
7567 			mtx_lock(&softc->ctl_lock);
7568 
7569 			for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7570 				if (memcmp(param->serv_act_res_key,
7571 				    lun->per_res[i].res_key.key,
7572 				    sizeof(struct scsi_per_res_key)) != 0)
7573 					continue;
7574 
7575 				found = 1;
7576 				lun->per_res[i].registered = 0;
7577 				memset(&lun->per_res[i].res_key, 0,
7578 				       sizeof(struct scsi_per_res_key));
7579 				lun->pr_key_count--;
7580 
7581 				if (!persis_offset
7582 				 && i < CTL_MAX_INITIATORS)
7583 					lun->pending_sense[i].ua_pending |=
7584 						CTL_UA_REG_PREEMPT;
7585 				else if (persis_offset
7586 				      && i >= persis_offset)
7587 					lun->pending_sense[
7588 						i-persis_offset].ua_pending |=
7589 						CTL_UA_REG_PREEMPT;
7590 			}
7591 
7592 			if (!found) {
7593 				mtx_unlock(&softc->ctl_lock);
7594 				free(ctsio->kern_data_ptr, M_CTL);
7595 				ctl_set_reservation_conflict(ctsio);
7596 				ctl_done((union ctl_io *)ctsio);
7597 		        	return (1);
7598 			}
7599 			mtx_unlock(&softc->ctl_lock);
7600 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7601 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7602 			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7603 			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7604 			persis_io.pr.pr_info.res_type = type;
7605 			memcpy(persis_io.pr.pr_info.sa_res_key,
7606 			       param->serv_act_res_key,
7607 			       sizeof(param->serv_act_res_key));
7608 			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7609 			     &persis_io, sizeof(persis_io), 0)) >
7610 			     CTL_HA_STATUS_SUCCESS) {
7611 				printf("CTL:Persis Out error returned "
7612 				       "from ctl_ha_msg_send %d\n",
7613 				isc_retval);
7614 			}
7615 		}
7616 	}
7617 
7618 	lun->PRGeneration++;
7619 
7620 	return (retval);
7621 }
7622 
7623 static void
7624 ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
7625 {
7626 	int i;
7627 
7628 	if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7629 	 || lun->pr_res_idx == CTL_PR_NO_RESERVATION
7630 	 || memcmp(&lun->per_res[lun->pr_res_idx].res_key,
7631 		   msg->pr.pr_info.sa_res_key,
7632 		   sizeof(struct scsi_per_res_key)) != 0) {
7633 		uint64_t sa_res_key;
7634 		sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
7635 
7636 		if (sa_res_key == 0) {
7637 			/* temporarily unregister this nexus */
7638 			lun->per_res[msg->pr.pr_info.residx].registered = 0;
7639 
7640 			/*
7641 			 * Unregister everybody else and build UA for
7642 			 * them
7643 			 */
7644 			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7645 				if (lun->per_res[i].registered == 0)
7646 					continue;
7647 
7648 				if (!persis_offset
7649 				 && i < CTL_MAX_INITIATORS)
7650 					lun->pending_sense[i].ua_pending |=
7651 						CTL_UA_REG_PREEMPT;
7652 				else if (persis_offset && i >= persis_offset)
7653 					lun->pending_sense[i -
7654 						persis_offset].ua_pending |=
7655 						CTL_UA_REG_PREEMPT;
7656 				lun->per_res[i].registered = 0;
7657 				memset(&lun->per_res[i].res_key, 0,
7658 				       sizeof(struct scsi_per_res_key));
7659 			}
7660 
7661 			lun->per_res[msg->pr.pr_info.residx].registered = 1;
7662 			lun->pr_key_count = 1;
7663 			lun->res_type = msg->pr.pr_info.res_type;
7664 			if (lun->res_type != SPR_TYPE_WR_EX_AR
7665 			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7666 				lun->pr_res_idx = msg->pr.pr_info.residx;
7667 		} else {
7668 		        for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7669 				if (memcmp(msg->pr.pr_info.sa_res_key,
7670 		                   lun->per_res[i].res_key.key,
7671 		                   sizeof(struct scsi_per_res_key)) != 0)
7672 					continue;
7673 
7674 				lun->per_res[i].registered = 0;
7675 				memset(&lun->per_res[i].res_key, 0,
7676 				       sizeof(struct scsi_per_res_key));
7677 				lun->pr_key_count--;
7678 
7679 				if (!persis_offset
7680 				 && i < persis_offset)
7681 					lun->pending_sense[i].ua_pending |=
7682 						CTL_UA_REG_PREEMPT;
7683 				else if (persis_offset
7684 				      && i >= persis_offset)
7685 					lun->pending_sense[i -
7686 						persis_offset].ua_pending |=
7687 						CTL_UA_REG_PREEMPT;
7688 			}
7689 		}
7690 	} else {
7691 		/*
7692 		 * Temporarily unregister so it won't get removed
7693 		 * or UA generated
7694 		 */
7695 		lun->per_res[msg->pr.pr_info.residx].registered = 0;
7696 		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7697 			if (lun->per_res[i].registered == 0)
7698 				continue;
7699 
7700 			if (memcmp(msg->pr.pr_info.sa_res_key,
7701 	                   lun->per_res[i].res_key.key,
7702 	                   sizeof(struct scsi_per_res_key)) == 0) {
7703 				lun->per_res[i].registered = 0;
7704 				memset(&lun->per_res[i].res_key, 0,
7705 				       sizeof(struct scsi_per_res_key));
7706 				lun->pr_key_count--;
7707 				if (!persis_offset
7708 				 && i < CTL_MAX_INITIATORS)
7709 					lun->pending_sense[i].ua_pending |=
7710 						CTL_UA_REG_PREEMPT;
7711 				else if (persis_offset
7712 				      && i >= persis_offset)
7713 					lun->pending_sense[i -
7714 						persis_offset].ua_pending |=
7715 						CTL_UA_REG_PREEMPT;
7716 			} else if (msg->pr.pr_info.res_type != lun->res_type
7717 				&& (lun->res_type == SPR_TYPE_WR_EX_RO
7718 				 || lun->res_type == SPR_TYPE_EX_AC_RO)) {
7719 					if (!persis_offset
7720 					 && i < persis_offset)
7721 						lun->pending_sense[i
7722 							].ua_pending |=
7723 							CTL_UA_RES_RELEASE;
7724 					else if (persis_offset
7725 					      && i >= persis_offset)
7726 					lun->pending_sense[i -
7727 						persis_offset].ua_pending |=
7728 						CTL_UA_RES_RELEASE;
7729 			}
7730 		}
7731 		lun->per_res[msg->pr.pr_info.residx].registered = 1;
7732 		lun->res_type = msg->pr.pr_info.res_type;
7733 		if (lun->res_type != SPR_TYPE_WR_EX_AR
7734 		 && lun->res_type != SPR_TYPE_EX_AC_AR)
7735 			lun->pr_res_idx = msg->pr.pr_info.residx;
7736 		else
7737 			lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
7738 	}
7739 	lun->PRGeneration++;
7740 
7741 }
7742 
7743 
7744 int
7745 ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
7746 {
7747 	int retval;
7748 	int isc_retval;
7749 	u_int32_t param_len;
7750 	struct scsi_per_res_out *cdb;
7751 	struct ctl_lun *lun;
7752 	struct scsi_per_res_out_parms* param;
7753 	struct ctl_softc *softc;
7754 	uint32_t residx;
7755 	uint64_t res_key, sa_res_key;
7756 	uint8_t type;
7757 	union ctl_ha_msg persis_io;
7758 	int    i;
7759 
7760 	CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
7761 
7762 	retval = CTL_RETVAL_COMPLETE;
7763 
7764 	softc = control_softc;
7765 
7766 	cdb = (struct scsi_per_res_out *)ctsio->cdb;
7767 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7768 
7769 	/*
7770 	 * We only support whole-LUN scope.  The scope & type are ignored for
7771 	 * register, register and ignore existing key and clear.
7772 	 * We sometimes ignore scope and type on preempts too!!
7773 	 * Verify reservation type here as well.
7774 	 */
7775 	type = cdb->scope_type & SPR_TYPE_MASK;
7776 	if ((cdb->action == SPRO_RESERVE)
7777 	 || (cdb->action == SPRO_RELEASE)) {
7778 		if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
7779 			ctl_set_invalid_field(/*ctsio*/ ctsio,
7780 					      /*sks_valid*/ 1,
7781 					      /*command*/ 1,
7782 					      /*field*/ 2,
7783 					      /*bit_valid*/ 1,
7784 					      /*bit*/ 4);
7785 			ctl_done((union ctl_io *)ctsio);
7786 			return (CTL_RETVAL_COMPLETE);
7787 		}
7788 
7789 		if (type>8 || type==2 || type==4 || type==0) {
7790 			ctl_set_invalid_field(/*ctsio*/ ctsio,
7791 					      /*sks_valid*/ 1,
7792 					      /*command*/ 1,
7793 					      /*field*/ 2,
7794 					      /*bit_valid*/ 1,
7795 					      /*bit*/ 0);
7796 			ctl_done((union ctl_io *)ctsio);
7797 			return (CTL_RETVAL_COMPLETE);
7798 		}
7799 	}
7800 
7801 	switch (cdb->action & SPRO_ACTION_MASK) {
7802 	case SPRO_REGISTER:
7803 	case SPRO_RESERVE:
7804 	case SPRO_RELEASE:
7805 	case SPRO_CLEAR:
7806 	case SPRO_PREEMPT:
7807 	case SPRO_REG_IGNO:
7808 		break;
7809 	case SPRO_REG_MOVE:
7810 	case SPRO_PRE_ABO:
7811 	default:
7812 		ctl_set_invalid_field(/*ctsio*/ ctsio,
7813 				      /*sks_valid*/ 1,
7814 				      /*command*/ 1,
7815 				      /*field*/ 1,
7816 				      /*bit_valid*/ 1,
7817 				      /*bit*/ 0);
7818 		ctl_done((union ctl_io *)ctsio);
7819 		return (CTL_RETVAL_COMPLETE);
7820 		break; /* NOTREACHED */
7821 	}
7822 
7823 	param_len = scsi_4btoul(cdb->length);
7824 
7825 	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
7826 		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
7827 		ctsio->kern_data_len = param_len;
7828 		ctsio->kern_total_len = param_len;
7829 		ctsio->kern_data_resid = 0;
7830 		ctsio->kern_rel_offset = 0;
7831 		ctsio->kern_sg_entries = 0;
7832 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7833 		ctsio->be_move_done = ctl_config_move_done;
7834 		ctl_datamove((union ctl_io *)ctsio);
7835 
7836 		return (CTL_RETVAL_COMPLETE);
7837 	}
7838 
7839 	param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
7840 
7841 	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
7842 	res_key = scsi_8btou64(param->res_key.key);
7843 	sa_res_key = scsi_8btou64(param->serv_act_res_key);
7844 
7845 	/*
7846 	 * Validate the reservation key here except for SPRO_REG_IGNO
7847 	 * This must be done for all other service actions
7848 	 */
7849 	if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
7850 		mtx_lock(&softc->ctl_lock);
7851 		if (lun->per_res[residx].registered) {
7852 		    if (memcmp(param->res_key.key,
7853 			       lun->per_res[residx].res_key.key,
7854 			       ctl_min(sizeof(param->res_key),
7855 			       sizeof(lun->per_res[residx].res_key))) != 0) {
7856 				/*
7857 				 * The current key passed in doesn't match
7858 				 * the one the initiator previously
7859 				 * registered.
7860 				 */
7861 				mtx_unlock(&softc->ctl_lock);
7862 				free(ctsio->kern_data_ptr, M_CTL);
7863 				ctl_set_reservation_conflict(ctsio);
7864 				ctl_done((union ctl_io *)ctsio);
7865 				return (CTL_RETVAL_COMPLETE);
7866 			}
7867 		} else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
7868 		    /*
7869 			 * We are not registered
7870 			 */
7871 			mtx_unlock(&softc->ctl_lock);
7872 			free(ctsio->kern_data_ptr, M_CTL);
7873 			ctl_set_reservation_conflict(ctsio);
7874 			ctl_done((union ctl_io *)ctsio);
7875 			return (CTL_RETVAL_COMPLETE);
7876 		} else if (res_key != 0) {
7877 			/*
7878 			 * We are not registered and trying to register but
7879 			 * the register key isn't zero.
7880 			 */
7881 			mtx_unlock(&softc->ctl_lock);
7882 			free(ctsio->kern_data_ptr, M_CTL);
7883 			ctl_set_reservation_conflict(ctsio);
7884 			ctl_done((union ctl_io *)ctsio);
7885 			return (CTL_RETVAL_COMPLETE);
7886 		}
7887 		mtx_unlock(&softc->ctl_lock);
7888 	}
7889 
7890 	switch (cdb->action & SPRO_ACTION_MASK) {
7891 	case SPRO_REGISTER:
7892 	case SPRO_REG_IGNO: {
7893 
7894 #if 0
7895 		printf("Registration received\n");
7896 #endif
7897 
7898 		/*
7899 		 * We don't support any of these options, as we report in
7900 		 * the read capabilities request (see
7901 		 * ctl_persistent_reserve_in(), above).
7902 		 */
7903 		if ((param->flags & SPR_SPEC_I_PT)
7904 		 || (param->flags & SPR_ALL_TG_PT)
7905 		 || (param->flags & SPR_APTPL)) {
7906 			int bit_ptr;
7907 
7908 			if (param->flags & SPR_APTPL)
7909 				bit_ptr = 0;
7910 			else if (param->flags & SPR_ALL_TG_PT)
7911 				bit_ptr = 2;
7912 			else /* SPR_SPEC_I_PT */
7913 				bit_ptr = 3;
7914 
7915 			free(ctsio->kern_data_ptr, M_CTL);
7916 			ctl_set_invalid_field(ctsio,
7917 					      /*sks_valid*/ 1,
7918 					      /*command*/ 0,
7919 					      /*field*/ 20,
7920 					      /*bit_valid*/ 1,
7921 					      /*bit*/ bit_ptr);
7922 			ctl_done((union ctl_io *)ctsio);
7923 			return (CTL_RETVAL_COMPLETE);
7924 		}
7925 
7926 		mtx_lock(&softc->ctl_lock);
7927 
7928 		/*
7929 		 * The initiator wants to clear the
7930 		 * key/unregister.
7931 		 */
7932 		if (sa_res_key == 0) {
7933 			if ((res_key == 0
7934 			  && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
7935 			 || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
7936 			  && !lun->per_res[residx].registered)) {
7937 				mtx_unlock(&softc->ctl_lock);
7938 				goto done;
7939 			}
7940 
7941 			lun->per_res[residx].registered = 0;
7942 			memset(&lun->per_res[residx].res_key,
7943 			       0, sizeof(lun->per_res[residx].res_key));
7944 			lun->pr_key_count--;
7945 
7946 			if (residx == lun->pr_res_idx) {
7947 				lun->flags &= ~CTL_LUN_PR_RESERVED;
7948 				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
7949 
7950 				if ((lun->res_type == SPR_TYPE_WR_EX_RO
7951 				  || lun->res_type == SPR_TYPE_EX_AC_RO)
7952 				 && lun->pr_key_count) {
7953 					/*
7954 					 * If the reservation is a registrants
7955 					 * only type we need to generate a UA
7956 					 * for other registered inits.  The
7957 					 * sense code should be RESERVATIONS
7958 					 * RELEASED
7959 					 */
7960 
7961 					for (i = 0; i < CTL_MAX_INITIATORS;i++){
7962 						if (lun->per_res[
7963 						    i+persis_offset].registered
7964 						    == 0)
7965 							continue;
7966 						lun->pending_sense[i
7967 							].ua_pending |=
7968 							CTL_UA_RES_RELEASE;
7969 					}
7970 				}
7971 				lun->res_type = 0;
7972 			} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7973 				if (lun->pr_key_count==0) {
7974 					lun->flags &= ~CTL_LUN_PR_RESERVED;
7975 					lun->res_type = 0;
7976 					lun->pr_res_idx = CTL_PR_NO_RESERVATION;
7977 				}
7978 			}
7979 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7980 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7981 			persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
7982 			persis_io.pr.pr_info.residx = residx;
7983 			if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7984 			     &persis_io, sizeof(persis_io), 0 )) >
7985 			     CTL_HA_STATUS_SUCCESS) {
7986 				printf("CTL:Persis Out error returned from "
7987 				       "ctl_ha_msg_send %d\n", isc_retval);
7988 			}
7989 			mtx_unlock(&softc->ctl_lock);
7990 		} else /* sa_res_key != 0 */ {
7991 
7992 			/*
7993 			 * If we aren't registered currently then increment
7994 			 * the key count and set the registered flag.
7995 			 */
7996 			if (!lun->per_res[residx].registered) {
7997 				lun->pr_key_count++;
7998 				lun->per_res[residx].registered = 1;
7999 			}
8000 
8001 			memcpy(&lun->per_res[residx].res_key,
8002 			       param->serv_act_res_key,
8003 			       ctl_min(sizeof(param->serv_act_res_key),
8004 			       sizeof(lun->per_res[residx].res_key)));
8005 
8006 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8007 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8008 			persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8009 			persis_io.pr.pr_info.residx = residx;
8010 			memcpy(persis_io.pr.pr_info.sa_res_key,
8011 			       param->serv_act_res_key,
8012 			       sizeof(param->serv_act_res_key));
8013 			mtx_unlock(&softc->ctl_lock);
8014 			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8015 			     &persis_io, sizeof(persis_io), 0)) >
8016 			     CTL_HA_STATUS_SUCCESS) {
8017 				printf("CTL:Persis Out error returned from "
8018 				       "ctl_ha_msg_send %d\n", isc_retval);
8019 			}
8020 		}
8021 		lun->PRGeneration++;
8022 
8023 		break;
8024 	}
8025 	case SPRO_RESERVE:
8026 #if 0
8027                 printf("Reserve executed type %d\n", type);
8028 #endif
8029 		mtx_lock(&softc->ctl_lock);
8030 		if (lun->flags & CTL_LUN_PR_RESERVED) {
8031 			/*
8032 			 * if this isn't the reservation holder and it's
8033 			 * not a "all registrants" type or if the type is
8034 			 * different then we have a conflict
8035 			 */
8036 			if ((lun->pr_res_idx != residx
8037 			  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8038 			 || lun->res_type != type) {
8039 				mtx_unlock(&softc->ctl_lock);
8040 				free(ctsio->kern_data_ptr, M_CTL);
8041 				ctl_set_reservation_conflict(ctsio);
8042 				ctl_done((union ctl_io *)ctsio);
8043 				return (CTL_RETVAL_COMPLETE);
8044 			}
8045 		} else /* create a reservation */ {
8046 			/*
8047 			 * If it's not an "all registrants" type record
8048 			 * reservation holder
8049 			 */
8050 			if (type != SPR_TYPE_WR_EX_AR
8051 			 && type != SPR_TYPE_EX_AC_AR)
8052 				lun->pr_res_idx = residx; /* Res holder */
8053 			else
8054 				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8055 
8056 			lun->flags |= CTL_LUN_PR_RESERVED;
8057 			lun->res_type = type;
8058 
8059 			mtx_unlock(&softc->ctl_lock);
8060 
8061 			/* send msg to other side */
8062 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8063 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8064 			persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8065 			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8066 			persis_io.pr.pr_info.res_type = type;
8067 			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8068 			     &persis_io, sizeof(persis_io), 0)) >
8069 			     CTL_HA_STATUS_SUCCESS) {
8070 				printf("CTL:Persis Out error returned from "
8071 				       "ctl_ha_msg_send %d\n", isc_retval);
8072 			}
8073 		}
8074 		break;
8075 
8076 	case SPRO_RELEASE:
8077 		mtx_lock(&softc->ctl_lock);
8078 		if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8079 			/* No reservation exists return good status */
8080 			mtx_unlock(&softc->ctl_lock);
8081 			goto done;
8082 		}
8083 		/*
8084 		 * Is this nexus a reservation holder?
8085 		 */
8086 		if (lun->pr_res_idx != residx
8087 		 && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8088 			/*
8089 			 * not a res holder return good status but
8090 			 * do nothing
8091 			 */
8092 			mtx_unlock(&softc->ctl_lock);
8093 			goto done;
8094 		}
8095 
8096 		if (lun->res_type != type) {
8097 			mtx_unlock(&softc->ctl_lock);
8098 			free(ctsio->kern_data_ptr, M_CTL);
8099 			ctl_set_illegal_pr_release(ctsio);
8100 			ctl_done((union ctl_io *)ctsio);
8101 			return (CTL_RETVAL_COMPLETE);
8102 		}
8103 
8104 		/* okay to release */
8105 		lun->flags &= ~CTL_LUN_PR_RESERVED;
8106 		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8107 		lun->res_type = 0;
8108 
8109 		/*
8110 		 * if this isn't an exclusive access
8111 		 * res generate UA for all other
8112 		 * registrants.
8113 		 */
8114 		if (type != SPR_TYPE_EX_AC
8115 		 && type != SPR_TYPE_WR_EX) {
8116 			/*
8117 			 * temporarily unregister so we don't generate UA
8118 			 */
8119 			lun->per_res[residx].registered = 0;
8120 
8121 			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8122 				if (lun->per_res[i+persis_offset].registered
8123 				    == 0)
8124 					continue;
8125 				lun->pending_sense[i].ua_pending |=
8126 					CTL_UA_RES_RELEASE;
8127 			}
8128 
8129 			lun->per_res[residx].registered = 1;
8130 		}
8131 		mtx_unlock(&softc->ctl_lock);
8132 		/* Send msg to other side */
8133 		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8134 		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8135 		persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8136 		if ((isc_retval=ctl_ha_msg_send( CTL_HA_CHAN_CTL, &persis_io,
8137 		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8138 			printf("CTL:Persis Out error returned from "
8139 			       "ctl_ha_msg_send %d\n", isc_retval);
8140 		}
8141 		break;
8142 
8143 	case SPRO_CLEAR:
8144 		/* send msg to other side */
8145 
8146 		mtx_lock(&softc->ctl_lock);
8147 		lun->flags &= ~CTL_LUN_PR_RESERVED;
8148 		lun->res_type = 0;
8149 		lun->pr_key_count = 0;
8150 		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8151 
8152 
8153 		memset(&lun->per_res[residx].res_key,
8154 		       0, sizeof(lun->per_res[residx].res_key));
8155 		lun->per_res[residx].registered = 0;
8156 
8157 		for (i=0; i < 2*CTL_MAX_INITIATORS; i++)
8158 			if (lun->per_res[i].registered) {
8159 				if (!persis_offset && i < CTL_MAX_INITIATORS)
8160 					lun->pending_sense[i].ua_pending |=
8161 						CTL_UA_RES_PREEMPT;
8162 				else if (persis_offset && i >= persis_offset)
8163 					lun->pending_sense[i-persis_offset
8164 					    ].ua_pending |= CTL_UA_RES_PREEMPT;
8165 
8166 				memset(&lun->per_res[i].res_key,
8167 				       0, sizeof(struct scsi_per_res_key));
8168 				lun->per_res[i].registered = 0;
8169 			}
8170 		lun->PRGeneration++;
8171 		mtx_unlock(&softc->ctl_lock);
8172 		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8173 		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8174 		persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8175 		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8176 		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8177 			printf("CTL:Persis Out error returned from "
8178 			       "ctl_ha_msg_send %d\n", isc_retval);
8179 		}
8180 		break;
8181 
8182 	case SPRO_PREEMPT: {
8183 		int nretval;
8184 
8185 		nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8186 					  residx, ctsio, cdb, param);
8187 		if (nretval != 0)
8188 			return (CTL_RETVAL_COMPLETE);
8189 		break;
8190 	}
8191 	case SPRO_REG_MOVE:
8192 	case SPRO_PRE_ABO:
8193 	default:
8194 		free(ctsio->kern_data_ptr, M_CTL);
8195 		ctl_set_invalid_field(/*ctsio*/ ctsio,
8196 				      /*sks_valid*/ 1,
8197 				      /*command*/ 1,
8198 				      /*field*/ 1,
8199 				      /*bit_valid*/ 1,
8200 				      /*bit*/ 0);
8201 		ctl_done((union ctl_io *)ctsio);
8202 		return (CTL_RETVAL_COMPLETE);
8203 		break; /* NOTREACHED */
8204 	}
8205 
8206 done:
8207 	free(ctsio->kern_data_ptr, M_CTL);
8208 	ctl_set_success(ctsio);
8209 	ctl_done((union ctl_io *)ctsio);
8210 
8211 	return (retval);
8212 }
8213 
8214 /*
8215  * This routine is for handling a message from the other SC pertaining to
8216  * persistent reserve out. All the error checking will have been done
8217  * so only perorming the action need be done here to keep the two
8218  * in sync.
8219  */
8220 static void
8221 ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8222 {
8223 	struct ctl_lun *lun;
8224 	struct ctl_softc *softc;
8225 	int i;
8226 
8227 	softc = control_softc;
8228 
8229 	mtx_lock(&softc->ctl_lock);
8230 
8231 	lun = softc->ctl_luns[msg->hdr.nexus.targ_lun];
8232 	switch(msg->pr.pr_info.action) {
8233 	case CTL_PR_REG_KEY:
8234 		if (!lun->per_res[msg->pr.pr_info.residx].registered) {
8235 			lun->per_res[msg->pr.pr_info.residx].registered = 1;
8236 			lun->pr_key_count++;
8237 		}
8238 		lun->PRGeneration++;
8239 		memcpy(&lun->per_res[msg->pr.pr_info.residx].res_key,
8240 		       msg->pr.pr_info.sa_res_key,
8241 		       sizeof(struct scsi_per_res_key));
8242 		break;
8243 
8244 	case CTL_PR_UNREG_KEY:
8245 		lun->per_res[msg->pr.pr_info.residx].registered = 0;
8246 		memset(&lun->per_res[msg->pr.pr_info.residx].res_key,
8247 		       0, sizeof(struct scsi_per_res_key));
8248 		lun->pr_key_count--;
8249 
8250 		/* XXX Need to see if the reservation has been released */
8251 		/* if so do we need to generate UA? */
8252 		if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8253 			lun->flags &= ~CTL_LUN_PR_RESERVED;
8254 			lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8255 
8256 			if ((lun->res_type == SPR_TYPE_WR_EX_RO
8257 			  || lun->res_type == SPR_TYPE_EX_AC_RO)
8258 			 && lun->pr_key_count) {
8259 				/*
8260 				 * If the reservation is a registrants
8261 				 * only type we need to generate a UA
8262 				 * for other registered inits.  The
8263 				 * sense code should be RESERVATIONS
8264 				 * RELEASED
8265 				 */
8266 
8267 				for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8268 					if (lun->per_res[i+
8269 					    persis_offset].registered == 0)
8270 						continue;
8271 
8272 					lun->pending_sense[i
8273 						].ua_pending |=
8274 						CTL_UA_RES_RELEASE;
8275 				}
8276 			}
8277 			lun->res_type = 0;
8278 		} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8279 			if (lun->pr_key_count==0) {
8280 				lun->flags &= ~CTL_LUN_PR_RESERVED;
8281 				lun->res_type = 0;
8282 				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8283 			}
8284 		}
8285 		lun->PRGeneration++;
8286 		break;
8287 
8288 	case CTL_PR_RESERVE:
8289 		lun->flags |= CTL_LUN_PR_RESERVED;
8290 		lun->res_type = msg->pr.pr_info.res_type;
8291 		lun->pr_res_idx = msg->pr.pr_info.residx;
8292 
8293 		break;
8294 
8295 	case CTL_PR_RELEASE:
8296 		/*
8297 		 * if this isn't an exclusive access res generate UA for all
8298 		 * other registrants.
8299 		 */
8300 		if (lun->res_type != SPR_TYPE_EX_AC
8301 		 && lun->res_type != SPR_TYPE_WR_EX) {
8302 			for (i = 0; i < CTL_MAX_INITIATORS; i++)
8303 				if (lun->per_res[i+persis_offset].registered)
8304 					lun->pending_sense[i].ua_pending |=
8305 						CTL_UA_RES_RELEASE;
8306 		}
8307 
8308 		lun->flags &= ~CTL_LUN_PR_RESERVED;
8309 		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8310 		lun->res_type = 0;
8311 		break;
8312 
8313 	case CTL_PR_PREEMPT:
8314 		ctl_pro_preempt_other(lun, msg);
8315 		break;
8316 	case CTL_PR_CLEAR:
8317 		lun->flags &= ~CTL_LUN_PR_RESERVED;
8318 		lun->res_type = 0;
8319 		lun->pr_key_count = 0;
8320 		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8321 
8322 		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8323 			if (lun->per_res[i].registered == 0)
8324 				continue;
8325 			if (!persis_offset
8326 			 && i < CTL_MAX_INITIATORS)
8327 				lun->pending_sense[i].ua_pending |=
8328 					CTL_UA_RES_PREEMPT;
8329 			else if (persis_offset
8330 			      && i >= persis_offset)
8331    				lun->pending_sense[i-persis_offset].ua_pending|=
8332 					CTL_UA_RES_PREEMPT;
8333 			memset(&lun->per_res[i].res_key, 0,
8334 			       sizeof(struct scsi_per_res_key));
8335 			lun->per_res[i].registered = 0;
8336 		}
8337 		lun->PRGeneration++;
8338 		break;
8339 	}
8340 
8341 	mtx_unlock(&softc->ctl_lock);
8342 }
8343 
8344 int
8345 ctl_read_write(struct ctl_scsiio *ctsio)
8346 {
8347 	struct ctl_lun *lun;
8348 	struct ctl_lba_len lbalen;
8349 	uint64_t lba;
8350 	uint32_t num_blocks;
8351 	int reladdr, fua, dpo, ebp;
8352 	int retval;
8353 	int isread;
8354 
8355 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8356 
8357 	CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8358 
8359 	reladdr = 0;
8360 	fua = 0;
8361 	dpo = 0;
8362 	ebp = 0;
8363 
8364 	retval = CTL_RETVAL_COMPLETE;
8365 
8366 	isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8367 	      || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8368 	if (lun->flags & CTL_LUN_PR_RESERVED && isread) {
8369 		uint32_t residx;
8370 
8371 		/*
8372 		 * XXX KDM need a lock here.
8373 		 */
8374 		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8375 		if ((lun->res_type == SPR_TYPE_EX_AC
8376 		  && residx != lun->pr_res_idx)
8377 		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
8378 		   || lun->res_type == SPR_TYPE_EX_AC_AR)
8379 		  && !lun->per_res[residx].registered)) {
8380 			ctl_set_reservation_conflict(ctsio);
8381 			ctl_done((union ctl_io *)ctsio);
8382 			return (CTL_RETVAL_COMPLETE);
8383 	        }
8384 	}
8385 
8386 	switch (ctsio->cdb[0]) {
8387 	case READ_6:
8388 	case WRITE_6: {
8389 		struct scsi_rw_6 *cdb;
8390 
8391 		cdb = (struct scsi_rw_6 *)ctsio->cdb;
8392 
8393 		lba = scsi_3btoul(cdb->addr);
8394 		/* only 5 bits are valid in the most significant address byte */
8395 		lba &= 0x1fffff;
8396 		num_blocks = cdb->length;
8397 		/*
8398 		 * This is correct according to SBC-2.
8399 		 */
8400 		if (num_blocks == 0)
8401 			num_blocks = 256;
8402 		break;
8403 	}
8404 	case READ_10:
8405 	case WRITE_10: {
8406 		struct scsi_rw_10 *cdb;
8407 
8408 		cdb = (struct scsi_rw_10 *)ctsio->cdb;
8409 
8410 		if (cdb->byte2 & SRW10_RELADDR)
8411 			reladdr = 1;
8412 		if (cdb->byte2 & SRW10_FUA)
8413 			fua = 1;
8414 		if (cdb->byte2 & SRW10_DPO)
8415 			dpo = 1;
8416 
8417 		if ((cdb->opcode == WRITE_10)
8418 		 && (cdb->byte2 & SRW10_EBP))
8419 			ebp = 1;
8420 
8421 		lba = scsi_4btoul(cdb->addr);
8422 		num_blocks = scsi_2btoul(cdb->length);
8423 		break;
8424 	}
8425 	case WRITE_VERIFY_10: {
8426 		struct scsi_write_verify_10 *cdb;
8427 
8428 		cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8429 
8430 		/*
8431 		 * XXX KDM we should do actual write verify support at some
8432 		 * point.  This is obviously fake, we're just translating
8433 		 * things to a write.  So we don't even bother checking the
8434 		 * BYTCHK field, since we don't do any verification.  If
8435 		 * the user asks for it, we'll just pretend we did it.
8436 		 */
8437 		if (cdb->byte2 & SWV_DPO)
8438 			dpo = 1;
8439 
8440 		lba = scsi_4btoul(cdb->addr);
8441 		num_blocks = scsi_2btoul(cdb->length);
8442 		break;
8443 	}
8444 	case READ_12:
8445 	case WRITE_12: {
8446 		struct scsi_rw_12 *cdb;
8447 
8448 		cdb = (struct scsi_rw_12 *)ctsio->cdb;
8449 
8450 		if (cdb->byte2 & SRW12_RELADDR)
8451 			reladdr = 1;
8452 		if (cdb->byte2 & SRW12_FUA)
8453 			fua = 1;
8454 		if (cdb->byte2 & SRW12_DPO)
8455 			dpo = 1;
8456 		lba = scsi_4btoul(cdb->addr);
8457 		num_blocks = scsi_4btoul(cdb->length);
8458 		break;
8459 	}
8460 	case WRITE_VERIFY_12: {
8461 		struct scsi_write_verify_12 *cdb;
8462 
8463 		cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8464 
8465 		if (cdb->byte2 & SWV_DPO)
8466 			dpo = 1;
8467 
8468 		lba = scsi_4btoul(cdb->addr);
8469 		num_blocks = scsi_4btoul(cdb->length);
8470 
8471 		break;
8472 	}
8473 	case READ_16:
8474 	case WRITE_16: {
8475 		struct scsi_rw_16 *cdb;
8476 
8477 		cdb = (struct scsi_rw_16 *)ctsio->cdb;
8478 
8479 		if (cdb->byte2 & SRW12_RELADDR)
8480 			reladdr = 1;
8481 		if (cdb->byte2 & SRW12_FUA)
8482 			fua = 1;
8483 		if (cdb->byte2 & SRW12_DPO)
8484 			dpo = 1;
8485 
8486 		lba = scsi_8btou64(cdb->addr);
8487 		num_blocks = scsi_4btoul(cdb->length);
8488 		break;
8489 	}
8490 	case WRITE_VERIFY_16: {
8491 		struct scsi_write_verify_16 *cdb;
8492 
8493 		cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
8494 
8495 		if (cdb->byte2 & SWV_DPO)
8496 			dpo = 1;
8497 
8498 		lba = scsi_8btou64(cdb->addr);
8499 		num_blocks = scsi_4btoul(cdb->length);
8500 		break;
8501 	}
8502 	default:
8503 		/*
8504 		 * We got a command we don't support.  This shouldn't
8505 		 * happen, commands should be filtered out above us.
8506 		 */
8507 		ctl_set_invalid_opcode(ctsio);
8508 		ctl_done((union ctl_io *)ctsio);
8509 
8510 		return (CTL_RETVAL_COMPLETE);
8511 		break; /* NOTREACHED */
8512 	}
8513 
8514 	/*
8515 	 * XXX KDM what do we do with the DPO and FUA bits?  FUA might be
8516 	 * interesting for us, but if RAIDCore is in write-back mode,
8517 	 * getting it to do write-through for a particular transaction may
8518 	 * not be possible.
8519 	 */
8520 	/*
8521 	 * We don't support relative addressing.  That also requires
8522 	 * supporting linked commands, which we don't do.
8523 	 */
8524 	if (reladdr != 0) {
8525 		ctl_set_invalid_field(ctsio,
8526 				      /*sks_valid*/ 1,
8527 				      /*command*/ 1,
8528 				      /*field*/ 1,
8529 				      /*bit_valid*/ 1,
8530 				      /*bit*/ 0);
8531 		ctl_done((union ctl_io *)ctsio);
8532 		return (CTL_RETVAL_COMPLETE);
8533 	}
8534 
8535 	/*
8536 	 * The first check is to make sure we're in bounds, the second
8537 	 * check is to catch wrap-around problems.  If the lba + num blocks
8538 	 * is less than the lba, then we've wrapped around and the block
8539 	 * range is invalid anyway.
8540 	 */
8541 	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8542 	 || ((lba + num_blocks) < lba)) {
8543 		ctl_set_lba_out_of_range(ctsio);
8544 		ctl_done((union ctl_io *)ctsio);
8545 		return (CTL_RETVAL_COMPLETE);
8546 	}
8547 
8548 	/*
8549 	 * According to SBC-3, a transfer length of 0 is not an error.
8550 	 * Note that this cannot happen with WRITE(6) or READ(6), since 0
8551 	 * translates to 256 blocks for those commands.
8552 	 */
8553 	if (num_blocks == 0) {
8554 		ctl_set_success(ctsio);
8555 		ctl_done((union ctl_io *)ctsio);
8556 		return (CTL_RETVAL_COMPLETE);
8557 	}
8558 
8559 	lbalen.lba = lba;
8560 	lbalen.len = num_blocks;
8561 	memcpy(ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes, &lbalen,
8562 	       sizeof(lbalen));
8563 
8564 	CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
8565 
8566 	retval = lun->backend->data_submit((union ctl_io *)ctsio);
8567 
8568 	return (retval);
8569 }
8570 
8571 int
8572 ctl_report_luns(struct ctl_scsiio *ctsio)
8573 {
8574 	struct scsi_report_luns *cdb;
8575 	struct scsi_report_luns_data *lun_data;
8576 	struct ctl_lun *lun, *request_lun;
8577 	int num_luns, retval;
8578 	uint32_t alloc_len, lun_datalen;
8579 	int num_filled, well_known;
8580 	uint32_t initidx;
8581 
8582 	retval = CTL_RETVAL_COMPLETE;
8583 	well_known = 0;
8584 
8585 	cdb = (struct scsi_report_luns *)ctsio->cdb;
8586 
8587 	CTL_DEBUG_PRINT(("ctl_report_luns\n"));
8588 
8589 	mtx_lock(&control_softc->ctl_lock);
8590 	num_luns = control_softc->num_luns;
8591 	mtx_unlock(&control_softc->ctl_lock);
8592 
8593 	switch (cdb->select_report) {
8594 	case RPL_REPORT_DEFAULT:
8595 	case RPL_REPORT_ALL:
8596 		break;
8597 	case RPL_REPORT_WELLKNOWN:
8598 		well_known = 1;
8599 		num_luns = 0;
8600 		break;
8601 	default:
8602 		ctl_set_invalid_field(ctsio,
8603 				      /*sks_valid*/ 1,
8604 				      /*command*/ 1,
8605 				      /*field*/ 2,
8606 				      /*bit_valid*/ 0,
8607 				      /*bit*/ 0);
8608 		ctl_done((union ctl_io *)ctsio);
8609 		return (retval);
8610 		break; /* NOTREACHED */
8611 	}
8612 
8613 	alloc_len = scsi_4btoul(cdb->length);
8614 	/*
8615 	 * The initiator has to allocate at least 16 bytes for this request,
8616 	 * so he can at least get the header and the first LUN.  Otherwise
8617 	 * we reject the request (per SPC-3 rev 14, section 6.21).
8618 	 */
8619 	if (alloc_len < (sizeof(struct scsi_report_luns_data) +
8620 	    sizeof(struct scsi_report_luns_lundata))) {
8621 		ctl_set_invalid_field(ctsio,
8622 				      /*sks_valid*/ 1,
8623 				      /*command*/ 1,
8624 				      /*field*/ 6,
8625 				      /*bit_valid*/ 0,
8626 				      /*bit*/ 0);
8627 		ctl_done((union ctl_io *)ctsio);
8628 		return (retval);
8629 	}
8630 
8631 	request_lun = (struct ctl_lun *)
8632 		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8633 
8634 	lun_datalen = sizeof(*lun_data) +
8635 		(num_luns * sizeof(struct scsi_report_luns_lundata));
8636 
8637 	ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
8638 	lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
8639 	ctsio->kern_sg_entries = 0;
8640 
8641 	if (lun_datalen < alloc_len) {
8642 		ctsio->residual = alloc_len - lun_datalen;
8643 		ctsio->kern_data_len = lun_datalen;
8644 		ctsio->kern_total_len = lun_datalen;
8645 	} else {
8646 		ctsio->residual = 0;
8647 		ctsio->kern_data_len = alloc_len;
8648 		ctsio->kern_total_len = alloc_len;
8649 	}
8650 	ctsio->kern_data_resid = 0;
8651 	ctsio->kern_rel_offset = 0;
8652 	ctsio->kern_sg_entries = 0;
8653 
8654 	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
8655 
8656 	/*
8657 	 * We set this to the actual data length, regardless of how much
8658 	 * space we actually have to return results.  If the user looks at
8659 	 * this value, he'll know whether or not he allocated enough space
8660 	 * and reissue the command if necessary.  We don't support well
8661 	 * known logical units, so if the user asks for that, return none.
8662 	 */
8663 	scsi_ulto4b(lun_datalen - 8, lun_data->length);
8664 
8665 	mtx_lock(&control_softc->ctl_lock);
8666 	for (num_filled = 0, lun = STAILQ_FIRST(&control_softc->lun_list);
8667 	     (lun != NULL) && (num_filled < num_luns);
8668 	     lun = STAILQ_NEXT(lun, links)) {
8669 
8670 		if (lun->lun <= 0xff) {
8671 			/*
8672 			 * Peripheral addressing method, bus number 0.
8673 			 */
8674 			lun_data->luns[num_filled].lundata[0] =
8675 				RPL_LUNDATA_ATYP_PERIPH;
8676 			lun_data->luns[num_filled].lundata[1] = lun->lun;
8677 			num_filled++;
8678 		} else if (lun->lun <= 0x3fff) {
8679 			/*
8680 			 * Flat addressing method.
8681 			 */
8682 			lun_data->luns[num_filled].lundata[0] =
8683 				RPL_LUNDATA_ATYP_FLAT |
8684 				(lun->lun & RPL_LUNDATA_FLAT_LUN_MASK);
8685 #ifdef OLDCTLHEADERS
8686 				(SRLD_ADDR_FLAT << SRLD_ADDR_SHIFT) |
8687 				(lun->lun & SRLD_BUS_LUN_MASK);
8688 #endif
8689 			lun_data->luns[num_filled].lundata[1] =
8690 #ifdef OLDCTLHEADERS
8691 				lun->lun >> SRLD_BUS_LUN_BITS;
8692 #endif
8693 				lun->lun >> RPL_LUNDATA_FLAT_LUN_BITS;
8694 			num_filled++;
8695 		} else {
8696 			printf("ctl_report_luns: bogus LUN number %jd, "
8697 			       "skipping\n", (intmax_t)lun->lun);
8698 		}
8699 		/*
8700 		 * According to SPC-3, rev 14 section 6.21:
8701 		 *
8702 		 * "The execution of a REPORT LUNS command to any valid and
8703 		 * installed logical unit shall clear the REPORTED LUNS DATA
8704 		 * HAS CHANGED unit attention condition for all logical
8705 		 * units of that target with respect to the requesting
8706 		 * initiator. A valid and installed logical unit is one
8707 		 * having a PERIPHERAL QUALIFIER of 000b in the standard
8708 		 * INQUIRY data (see 6.4.2)."
8709 		 *
8710 		 * If request_lun is NULL, the LUN this report luns command
8711 		 * was issued to is either disabled or doesn't exist. In that
8712 		 * case, we shouldn't clear any pending lun change unit
8713 		 * attention.
8714 		 */
8715 		if (request_lun != NULL)
8716 			lun->pending_sense[initidx].ua_pending &=
8717 				~CTL_UA_LUN_CHANGE;
8718 	}
8719 	mtx_unlock(&control_softc->ctl_lock);
8720 
8721 	/*
8722 	 * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
8723 	 * this request.
8724 	 */
8725 	ctsio->scsi_status = SCSI_STATUS_OK;
8726 
8727 	ctsio->be_move_done = ctl_config_move_done;
8728 	ctl_datamove((union ctl_io *)ctsio);
8729 
8730 	return (retval);
8731 }
8732 
8733 int
8734 ctl_request_sense(struct ctl_scsiio *ctsio)
8735 {
8736 	struct scsi_request_sense *cdb;
8737 	struct scsi_sense_data *sense_ptr;
8738 	struct ctl_lun *lun;
8739 	uint32_t initidx;
8740 	int have_error;
8741 	scsi_sense_data_type sense_format;
8742 
8743 	cdb = (struct scsi_request_sense *)ctsio->cdb;
8744 
8745 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8746 
8747 	CTL_DEBUG_PRINT(("ctl_request_sense\n"));
8748 
8749 	/*
8750 	 * Determine which sense format the user wants.
8751 	 */
8752 	if (cdb->byte2 & SRS_DESC)
8753 		sense_format = SSD_TYPE_DESC;
8754 	else
8755 		sense_format = SSD_TYPE_FIXED;
8756 
8757 	ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
8758 	sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
8759 	ctsio->kern_sg_entries = 0;
8760 
8761 	/*
8762 	 * struct scsi_sense_data, which is currently set to 256 bytes, is
8763 	 * larger than the largest allowed value for the length field in the
8764 	 * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
8765 	 */
8766 	ctsio->residual = 0;
8767 	ctsio->kern_data_len = cdb->length;
8768 	ctsio->kern_total_len = cdb->length;
8769 
8770 	ctsio->kern_data_resid = 0;
8771 	ctsio->kern_rel_offset = 0;
8772 	ctsio->kern_sg_entries = 0;
8773 
8774 	/*
8775 	 * If we don't have a LUN, we don't have any pending sense.
8776 	 */
8777 	if (lun == NULL)
8778 		goto no_sense;
8779 
8780 	have_error = 0;
8781 	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
8782 	/*
8783 	 * Check for pending sense, and then for pending unit attentions.
8784 	 * Pending sense gets returned first, then pending unit attentions.
8785 	 */
8786 	mtx_lock(&lun->ctl_softc->ctl_lock);
8787 	if (ctl_is_set(lun->have_ca, initidx)) {
8788 		scsi_sense_data_type stored_format;
8789 
8790 		/*
8791 		 * Check to see which sense format was used for the stored
8792 		 * sense data.
8793 		 */
8794 		stored_format = scsi_sense_type(
8795 		    &lun->pending_sense[initidx].sense);
8796 
8797 		/*
8798 		 * If the user requested a different sense format than the
8799 		 * one we stored, then we need to convert it to the other
8800 		 * format.  If we're going from descriptor to fixed format
8801 		 * sense data, we may lose things in translation, depending
8802 		 * on what options were used.
8803 		 *
8804 		 * If the stored format is SSD_TYPE_NONE (i.e. invalid),
8805 		 * for some reason we'll just copy it out as-is.
8806 		 */
8807 		if ((stored_format == SSD_TYPE_FIXED)
8808 		 && (sense_format == SSD_TYPE_DESC))
8809 			ctl_sense_to_desc((struct scsi_sense_data_fixed *)
8810 			    &lun->pending_sense[initidx].sense,
8811 			    (struct scsi_sense_data_desc *)sense_ptr);
8812 		else if ((stored_format == SSD_TYPE_DESC)
8813 		      && (sense_format == SSD_TYPE_FIXED))
8814 			ctl_sense_to_fixed((struct scsi_sense_data_desc *)
8815 			    &lun->pending_sense[initidx].sense,
8816 			    (struct scsi_sense_data_fixed *)sense_ptr);
8817 		else
8818 			memcpy(sense_ptr, &lun->pending_sense[initidx].sense,
8819 			       ctl_min(sizeof(*sense_ptr),
8820 			       sizeof(lun->pending_sense[initidx].sense)));
8821 
8822 		ctl_clear_mask(lun->have_ca, initidx);
8823 		have_error = 1;
8824 	} else if (lun->pending_sense[initidx].ua_pending != CTL_UA_NONE) {
8825 		ctl_ua_type ua_type;
8826 
8827 		ua_type = ctl_build_ua(lun->pending_sense[initidx].ua_pending,
8828 				       sense_ptr, sense_format);
8829 		if (ua_type != CTL_UA_NONE) {
8830 			have_error = 1;
8831 			/* We're reporting this UA, so clear it */
8832 			lun->pending_sense[initidx].ua_pending &= ~ua_type;
8833 		}
8834 	}
8835 	mtx_unlock(&lun->ctl_softc->ctl_lock);
8836 
8837 	/*
8838 	 * We already have a pending error, return it.
8839 	 */
8840 	if (have_error != 0) {
8841 		/*
8842 		 * We report the SCSI status as OK, since the status of the
8843 		 * request sense command itself is OK.
8844 		 */
8845 		ctsio->scsi_status = SCSI_STATUS_OK;
8846 
8847 		/*
8848 		 * We report 0 for the sense length, because we aren't doing
8849 		 * autosense in this case.  We're reporting sense as
8850 		 * parameter data.
8851 		 */
8852 		ctsio->sense_len = 0;
8853 
8854 		ctsio->be_move_done = ctl_config_move_done;
8855 		ctl_datamove((union ctl_io *)ctsio);
8856 
8857 		return (CTL_RETVAL_COMPLETE);
8858 	}
8859 
8860 no_sense:
8861 
8862 	/*
8863 	 * No sense information to report, so we report that everything is
8864 	 * okay.
8865 	 */
8866 	ctl_set_sense_data(sense_ptr,
8867 			   lun,
8868 			   sense_format,
8869 			   /*current_error*/ 1,
8870 			   /*sense_key*/ SSD_KEY_NO_SENSE,
8871 			   /*asc*/ 0x00,
8872 			   /*ascq*/ 0x00,
8873 			   SSD_ELEM_NONE);
8874 
8875 	ctsio->scsi_status = SCSI_STATUS_OK;
8876 
8877 	/*
8878 	 * We report 0 for the sense length, because we aren't doing
8879 	 * autosense in this case.  We're reporting sense as parameter data.
8880 	 */
8881 	ctsio->sense_len = 0;
8882 	ctsio->be_move_done = ctl_config_move_done;
8883 	ctl_datamove((union ctl_io *)ctsio);
8884 
8885 	return (CTL_RETVAL_COMPLETE);
8886 }
8887 
8888 int
8889 ctl_tur(struct ctl_scsiio *ctsio)
8890 {
8891 	struct ctl_lun *lun;
8892 
8893 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8894 
8895 	CTL_DEBUG_PRINT(("ctl_tur\n"));
8896 
8897 	if (lun == NULL)
8898 		return (-EINVAL);
8899 
8900 	ctsio->scsi_status = SCSI_STATUS_OK;
8901 	ctsio->io_hdr.status = CTL_SUCCESS;
8902 
8903 	ctl_done((union ctl_io *)ctsio);
8904 
8905 	return (CTL_RETVAL_COMPLETE);
8906 }
8907 
8908 #ifdef notyet
8909 static int
8910 ctl_cmddt_inquiry(struct ctl_scsiio *ctsio)
8911 {
8912 
8913 }
8914 #endif
8915 
8916 static int
8917 ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
8918 {
8919 	struct scsi_vpd_supported_pages *pages;
8920 	int sup_page_size;
8921 	struct ctl_lun *lun;
8922 
8923 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8924 
8925 	sup_page_size = sizeof(struct scsi_vpd_supported_pages) +
8926 		SCSI_EVPD_NUM_SUPPORTED_PAGES;
8927 	/*
8928 	 * XXX KDM GFP_???  We probably don't want to wait here,
8929 	 * unless we end up having a process/thread context.
8930 	 */
8931 	ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
8932 	if (ctsio->kern_data_ptr == NULL) {
8933 		ctsio->io_hdr.status = CTL_SCSI_ERROR;
8934 		ctsio->scsi_status = SCSI_STATUS_BUSY;
8935 		ctl_done((union ctl_io *)ctsio);
8936 		return (CTL_RETVAL_COMPLETE);
8937 	}
8938 	pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
8939 	ctsio->kern_sg_entries = 0;
8940 
8941 	if (sup_page_size < alloc_len) {
8942 		ctsio->residual = alloc_len - sup_page_size;
8943 		ctsio->kern_data_len = sup_page_size;
8944 		ctsio->kern_total_len = sup_page_size;
8945 	} else {
8946 		ctsio->residual = 0;
8947 		ctsio->kern_data_len = alloc_len;
8948 		ctsio->kern_total_len = alloc_len;
8949 	}
8950 	ctsio->kern_data_resid = 0;
8951 	ctsio->kern_rel_offset = 0;
8952 	ctsio->kern_sg_entries = 0;
8953 
8954 	/*
8955 	 * The control device is always connected.  The disk device, on the
8956 	 * other hand, may not be online all the time.  Need to change this
8957 	 * to figure out whether the disk device is actually online or not.
8958 	 */
8959 	if (lun != NULL)
8960 		pages->device = (SID_QUAL_LU_CONNECTED << 5) |
8961 				lun->be_lun->lun_type;
8962 	else
8963 		pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
8964 
8965 	pages->length = SCSI_EVPD_NUM_SUPPORTED_PAGES;
8966 	/* Supported VPD pages */
8967 	pages->page_list[0] = SVPD_SUPPORTED_PAGES;
8968 	/* Serial Number */
8969 	pages->page_list[1] = SVPD_UNIT_SERIAL_NUMBER;
8970 	/* Device Identification */
8971 	pages->page_list[2] = SVPD_DEVICE_ID;
8972 
8973 	ctsio->scsi_status = SCSI_STATUS_OK;
8974 
8975 	ctsio->be_move_done = ctl_config_move_done;
8976 	ctl_datamove((union ctl_io *)ctsio);
8977 
8978 	return (CTL_RETVAL_COMPLETE);
8979 }
8980 
8981 static int
8982 ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
8983 {
8984 	struct scsi_vpd_unit_serial_number *sn_ptr;
8985 	struct ctl_lun *lun;
8986 #ifndef CTL_USE_BACKEND_SN
8987 	char tmpstr[32];
8988 #endif
8989 
8990 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8991 
8992 	/* XXX KDM which malloc flags here?? */
8993 	ctsio->kern_data_ptr = malloc(sizeof(*sn_ptr), M_CTL, M_WAITOK | M_ZERO);
8994 	if (ctsio->kern_data_ptr == NULL) {
8995 		ctsio->io_hdr.status = CTL_SCSI_ERROR;
8996 		ctsio->scsi_status = SCSI_STATUS_BUSY;
8997 		ctl_done((union ctl_io *)ctsio);
8998 		return (CTL_RETVAL_COMPLETE);
8999 	}
9000 	sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9001 	ctsio->kern_sg_entries = 0;
9002 
9003 	if (sizeof(*sn_ptr) < alloc_len) {
9004 		ctsio->residual = alloc_len - sizeof(*sn_ptr);
9005 		ctsio->kern_data_len = sizeof(*sn_ptr);
9006 		ctsio->kern_total_len = sizeof(*sn_ptr);
9007 	} else {
9008 		ctsio->residual = 0;
9009 		ctsio->kern_data_len = alloc_len;
9010 		ctsio->kern_total_len = alloc_len;
9011 	}
9012 	ctsio->kern_data_resid = 0;
9013 	ctsio->kern_rel_offset = 0;
9014 	ctsio->kern_sg_entries = 0;
9015 
9016 	/*
9017 	 * The control device is always connected.  The disk device, on the
9018 	 * other hand, may not be online all the time.  Need to change this
9019 	 * to figure out whether the disk device is actually online or not.
9020 	 */
9021 	if (lun != NULL)
9022 		sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9023 				  lun->be_lun->lun_type;
9024 	else
9025 		sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9026 
9027 	sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9028 	sn_ptr->length = ctl_min(sizeof(*sn_ptr) - 4, CTL_SN_LEN);
9029 #ifdef CTL_USE_BACKEND_SN
9030 	/*
9031 	 * If we don't have a LUN, we just leave the serial number as
9032 	 * all spaces.
9033 	 */
9034 	memset(sn_ptr->serial_num, 0x20, sizeof(sn_ptr->serial_num));
9035 	if (lun != NULL) {
9036 		strncpy((char *)sn_ptr->serial_num,
9037 			(char *)lun->be_lun->serial_num, CTL_SN_LEN);
9038 	}
9039 #else
9040 	/*
9041 	 * Note that we're using a non-unique serial number here,
9042 	 */
9043 	snprintf(tmpstr, sizeof(tmpstr), "MYSERIALNUMIS000");
9044 	memset(sn_ptr->serial_num, 0x20, sizeof(sn_ptr->serial_num));
9045 	strncpy(sn_ptr->serial_num, tmpstr, ctl_min(CTL_SN_LEN,
9046 		ctl_min(sizeof(tmpstr), sizeof(*sn_ptr) - 4)));
9047 #endif
9048 	ctsio->scsi_status = SCSI_STATUS_OK;
9049 
9050 	ctsio->be_move_done = ctl_config_move_done;
9051 	ctl_datamove((union ctl_io *)ctsio);
9052 
9053 	return (CTL_RETVAL_COMPLETE);
9054 }
9055 
9056 
9057 static int
9058 ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9059 {
9060 	struct scsi_vpd_device_id *devid_ptr;
9061 	struct scsi_vpd_id_descriptor *desc, *desc1;
9062 	struct scsi_vpd_id_descriptor *desc2, *desc3; /* for types 4h and 5h */
9063 	struct scsi_vpd_id_t10 *t10id;
9064 	struct ctl_softc *ctl_softc;
9065 	struct ctl_lun *lun;
9066 	struct ctl_frontend *fe;
9067 #ifndef CTL_USE_BACKEND_SN
9068 	char tmpstr[32];
9069 #endif /* CTL_USE_BACKEND_SN */
9070 	int devid_len;
9071 
9072 	ctl_softc = control_softc;
9073 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9074 
9075 	devid_len = sizeof(struct scsi_vpd_device_id) +
9076 		sizeof(struct scsi_vpd_id_descriptor) +
9077 		sizeof(struct scsi_vpd_id_t10) + CTL_DEVID_LEN +
9078 		sizeof(struct scsi_vpd_id_descriptor) + CTL_WWPN_LEN +
9079 		sizeof(struct scsi_vpd_id_descriptor) +
9080 		sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9081 		sizeof(struct scsi_vpd_id_descriptor) +
9082 		sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9083 
9084 	/* XXX KDM which malloc flags here ?? */
9085 	ctsio->kern_data_ptr = malloc(devid_len, M_CTL, M_WAITOK | M_ZERO);
9086 	if (ctsio->kern_data_ptr == NULL) {
9087 		ctsio->io_hdr.status = CTL_SCSI_ERROR;
9088 		ctsio->scsi_status = SCSI_STATUS_BUSY;
9089 		ctl_done((union ctl_io *)ctsio);
9090 		return (CTL_RETVAL_COMPLETE);
9091 	}
9092 	devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9093 	ctsio->kern_sg_entries = 0;
9094 
9095 	if (devid_len < alloc_len) {
9096 		ctsio->residual = alloc_len - devid_len;
9097 		ctsio->kern_data_len = devid_len;
9098 		ctsio->kern_total_len = devid_len;
9099 	} else {
9100 		ctsio->residual = 0;
9101 		ctsio->kern_data_len = alloc_len;
9102 		ctsio->kern_total_len = alloc_len;
9103 	}
9104 	ctsio->kern_data_resid = 0;
9105 	ctsio->kern_rel_offset = 0;
9106 	ctsio->kern_sg_entries = 0;
9107 
9108 	desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9109 	t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
9110 	desc1 = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9111 		sizeof(struct scsi_vpd_id_t10) + CTL_DEVID_LEN);
9112 	desc2 = (struct scsi_vpd_id_descriptor *)(&desc1->identifier[0] +
9113 	          CTL_WWPN_LEN);
9114 	desc3 = (struct scsi_vpd_id_descriptor *)(&desc2->identifier[0] +
9115 	         sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9116 
9117 	/*
9118 	 * The control device is always connected.  The disk device, on the
9119 	 * other hand, may not be online all the time.
9120 	 */
9121 	if (lun != NULL)
9122 		devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9123 				     lun->be_lun->lun_type;
9124 	else
9125 		devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9126 
9127 	devid_ptr->page_code = SVPD_DEVICE_ID;
9128 
9129 	scsi_ulto2b(devid_len - 4, devid_ptr->length);
9130 
9131 	mtx_lock(&ctl_softc->ctl_lock);
9132 
9133 	fe = ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)];
9134 
9135 	/*
9136 	 * For Fibre channel,
9137 	 */
9138 	if (fe->port_type == CTL_PORT_FC)
9139 	{
9140 		desc->proto_codeset = (SCSI_PROTO_FC << 4) |
9141 				      SVPD_ID_CODESET_ASCII;
9142         	desc1->proto_codeset = (SCSI_PROTO_FC << 4) |
9143 		              SVPD_ID_CODESET_BINARY;
9144 	}
9145 	else
9146 	{
9147 		desc->proto_codeset = (SCSI_PROTO_SPI << 4) |
9148 				      SVPD_ID_CODESET_ASCII;
9149         	desc1->proto_codeset = (SCSI_PROTO_SPI << 4) |
9150 		              SVPD_ID_CODESET_BINARY;
9151 	}
9152 	desc2->proto_codeset = desc3->proto_codeset = desc1->proto_codeset;
9153 	mtx_unlock(&ctl_softc->ctl_lock);
9154 
9155 	/*
9156 	 * We're using a LUN association here.  i.e., this device ID is a
9157 	 * per-LUN identifier.
9158 	 */
9159 	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
9160 	desc->length = sizeof(*t10id) + CTL_DEVID_LEN;
9161 	strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
9162 
9163 	/*
9164 	 * desc1 is for the WWPN which is a port asscociation.
9165 	 */
9166 	desc1->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT | SVPD_ID_TYPE_NAA;
9167 	desc1->length = CTL_WWPN_LEN;
9168 	/* XXX Call Reggie's get_WWNN func here then add port # to the end */
9169 	/* For testing just create the WWPN */
9170 #if 0
9171 	ddb_GetWWNN((char *)desc1->identifier);
9172 
9173 	/* NOTE: if the port is 0 or 8 we don't want to subtract 1 */
9174 	/* This is so Copancontrol will return something sane */
9175 	if (ctsio->io_hdr.nexus.targ_port!=0 &&
9176 	    ctsio->io_hdr.nexus.targ_port!=8)
9177 		desc1->identifier[7] += ctsio->io_hdr.nexus.targ_port-1;
9178 	else
9179 		desc1->identifier[7] += ctsio->io_hdr.nexus.targ_port;
9180 #endif
9181 
9182 	be64enc(desc1->identifier, fe->wwpn);
9183 
9184 	/*
9185 	 * desc2 is for the Relative Target Port(type 4h) identifier
9186 	 */
9187 	desc2->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT
9188 	                 | SVPD_ID_TYPE_RELTARG;
9189 	desc2->length = 4;
9190 //#if 0
9191 	/* NOTE: if the port is 0 or 8 we don't want to subtract 1 */
9192 	/* This is so Copancontrol will return something sane */
9193 	if (ctsio->io_hdr.nexus.targ_port!=0 &&
9194 	    ctsio->io_hdr.nexus.targ_port!=8)
9195 		desc2->identifier[3] = ctsio->io_hdr.nexus.targ_port - 1;
9196 	else
9197 	        desc2->identifier[3] = ctsio->io_hdr.nexus.targ_port;
9198 //#endif
9199 
9200 	/*
9201 	 * desc3 is for the Target Port Group(type 5h) identifier
9202 	 */
9203 	desc3->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT
9204 	                 | SVPD_ID_TYPE_TPORTGRP;
9205 	desc3->length = 4;
9206 	if (ctsio->io_hdr.nexus.targ_port < CTL_MAX_PORTS || ctl_is_single)
9207 		desc3->identifier[3] = 1;
9208 	else
9209 		desc3->identifier[3] = 2;
9210 
9211 #ifdef CTL_USE_BACKEND_SN
9212 	/*
9213 	 * If we've actually got a backend, copy the device id from the
9214 	 * per-LUN data.  Otherwise, set it to all spaces.
9215 	 */
9216 	if (lun != NULL) {
9217 		/*
9218 		 * Copy the backend's LUN ID.
9219 		 */
9220 		strncpy((char *)t10id->vendor_spec_id,
9221 			(char *)lun->be_lun->device_id, CTL_DEVID_LEN);
9222 	} else {
9223 		/*
9224 		 * No backend, set this to spaces.
9225 		 */
9226 		memset(t10id->vendor_spec_id, 0x20, CTL_DEVID_LEN);
9227 	}
9228 #else
9229 	snprintf(tmpstr, sizeof(tmpstr), "MYDEVICEIDIS%4d",
9230 		 (lun != NULL) ?  (int)lun->lun : 0);
9231 	strncpy(t10id->vendor_spec_id, tmpstr, ctl_min(CTL_DEVID_LEN,
9232 		sizeof(tmpstr)));
9233 #endif
9234 
9235 	ctsio->scsi_status = SCSI_STATUS_OK;
9236 
9237 	ctsio->be_move_done = ctl_config_move_done;
9238 	ctl_datamove((union ctl_io *)ctsio);
9239 
9240 	return (CTL_RETVAL_COMPLETE);
9241 }
9242 
9243 static int
9244 ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
9245 {
9246 	struct scsi_inquiry *cdb;
9247 	int alloc_len, retval;
9248 
9249 	cdb = (struct scsi_inquiry *)ctsio->cdb;
9250 
9251 	retval = CTL_RETVAL_COMPLETE;
9252 
9253 	alloc_len = scsi_2btoul(cdb->length);
9254 
9255 	switch (cdb->page_code) {
9256 	case SVPD_SUPPORTED_PAGES:
9257 		retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
9258 		break;
9259 	case SVPD_UNIT_SERIAL_NUMBER:
9260 		retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
9261 		break;
9262 	case SVPD_DEVICE_ID:
9263 		retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
9264 		break;
9265 	default:
9266 		ctl_set_invalid_field(ctsio,
9267 				      /*sks_valid*/ 1,
9268 				      /*command*/ 1,
9269 				      /*field*/ 2,
9270 				      /*bit_valid*/ 0,
9271 				      /*bit*/ 0);
9272 		ctl_done((union ctl_io *)ctsio);
9273 		retval = CTL_RETVAL_COMPLETE;
9274 		break;
9275 	}
9276 
9277 	return (retval);
9278 }
9279 
9280 static int
9281 ctl_inquiry_std(struct ctl_scsiio *ctsio)
9282 {
9283 	struct scsi_inquiry_data *inq_ptr;
9284 	struct scsi_inquiry *cdb;
9285 	struct ctl_softc *ctl_softc;
9286 	struct ctl_lun *lun;
9287 	uint32_t alloc_len;
9288 	int is_fc;
9289 
9290 	ctl_softc = control_softc;
9291 
9292 	/*
9293 	 * Figure out whether we're talking to a Fibre Channel port or not.
9294 	 * We treat the ioctl front end, and any SCSI adapters, as packetized
9295 	 * SCSI front ends.
9296 	 */
9297 	mtx_lock(&ctl_softc->ctl_lock);
9298 	if (ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)]->port_type !=
9299 	    CTL_PORT_FC)
9300 		is_fc = 0;
9301 	else
9302 		is_fc = 1;
9303 	mtx_unlock(&ctl_softc->ctl_lock);
9304 
9305 	lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9306 	cdb = (struct scsi_inquiry *)ctsio->cdb;
9307 	alloc_len = scsi_2btoul(cdb->length);
9308 
9309 	/*
9310 	 * We malloc the full inquiry data size here and fill it
9311 	 * in.  If the user only asks for less, we'll give him
9312 	 * that much.
9313 	 */
9314 	/* XXX KDM what malloc flags should we use here?? */
9315 	ctsio->kern_data_ptr = malloc(sizeof(*inq_ptr), M_CTL, M_WAITOK | M_ZERO);
9316 	if (ctsio->kern_data_ptr == NULL) {
9317 		ctsio->io_hdr.status = CTL_SCSI_ERROR;
9318 		ctsio->scsi_status = SCSI_STATUS_BUSY;
9319 		ctl_done((union ctl_io *)ctsio);
9320 		return (CTL_RETVAL_COMPLETE);
9321 	}
9322 	inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
9323 	ctsio->kern_sg_entries = 0;
9324 	ctsio->kern_data_resid = 0;
9325 	ctsio->kern_rel_offset = 0;
9326 
9327 	if (sizeof(*inq_ptr) < alloc_len) {
9328 		ctsio->residual = alloc_len - sizeof(*inq_ptr);
9329 		ctsio->kern_data_len = sizeof(*inq_ptr);
9330 		ctsio->kern_total_len = sizeof(*inq_ptr);
9331 	} else {
9332 		ctsio->residual = 0;
9333 		ctsio->kern_data_len = alloc_len;
9334 		ctsio->kern_total_len = alloc_len;
9335 	}
9336 
9337 	/*
9338 	 * If we have a LUN configured, report it as connected.  Otherwise,
9339 	 * report that it is offline or no device is supported, depending
9340 	 * on the value of inquiry_pq_no_lun.
9341 	 *
9342 	 * According to the spec (SPC-4 r34), the peripheral qualifier
9343 	 * SID_QUAL_LU_OFFLINE (001b) is used in the following scenario:
9344 	 *
9345 	 * "A peripheral device having the specified peripheral device type
9346 	 * is not connected to this logical unit. However, the device
9347 	 * server is capable of supporting the specified peripheral device
9348 	 * type on this logical unit."
9349 	 *
9350 	 * According to the same spec, the peripheral qualifier
9351 	 * SID_QUAL_BAD_LU (011b) is used in this scenario:
9352 	 *
9353 	 * "The device server is not capable of supporting a peripheral
9354 	 * device on this logical unit. For this peripheral qualifier the
9355 	 * peripheral device type shall be set to 1Fh. All other peripheral
9356 	 * device type values are reserved for this peripheral qualifier."
9357 	 *
9358 	 * Given the text, it would seem that we probably want to report that
9359 	 * the LUN is offline here.  There is no LUN connected, but we can
9360 	 * support a LUN at the given LUN number.
9361 	 *
9362 	 * In the real world, though, it sounds like things are a little
9363 	 * different:
9364 	 *
9365 	 * - Linux, when presented with a LUN with the offline peripheral
9366 	 *   qualifier, will create an sg driver instance for it.  So when
9367 	 *   you attach it to CTL, you wind up with a ton of sg driver
9368 	 *   instances.  (One for every LUN that Linux bothered to probe.)
9369 	 *   Linux does this despite the fact that it issues a REPORT LUNs
9370 	 *   to LUN 0 to get the inventory of supported LUNs.
9371 	 *
9372 	 * - There is other anecdotal evidence (from Emulex folks) about
9373 	 *   arrays that use the offline peripheral qualifier for LUNs that
9374 	 *   are on the "passive" path in an active/passive array.
9375 	 *
9376 	 * So the solution is provide a hopefully reasonable default
9377 	 * (return bad/no LUN) and allow the user to change the behavior
9378 	 * with a tunable/sysctl variable.
9379 	 */
9380 	if (lun != NULL)
9381 		inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9382 				  lun->be_lun->lun_type;
9383 	else if (ctl_softc->inquiry_pq_no_lun == 0)
9384 		inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9385 	else
9386 		inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
9387 
9388 	/* RMB in byte 2 is 0 */
9389 	inq_ptr->version = SCSI_REV_SPC3;
9390 
9391 	/*
9392 	 * According to SAM-3, even if a device only supports a single
9393 	 * level of LUN addressing, it should still set the HISUP bit:
9394 	 *
9395 	 * 4.9.1 Logical unit numbers overview
9396 	 *
9397 	 * All logical unit number formats described in this standard are
9398 	 * hierarchical in structure even when only a single level in that
9399 	 * hierarchy is used. The HISUP bit shall be set to one in the
9400 	 * standard INQUIRY data (see SPC-2) when any logical unit number
9401 	 * format described in this standard is used.  Non-hierarchical
9402 	 * formats are outside the scope of this standard.
9403 	 *
9404 	 * Therefore we set the HiSup bit here.
9405 	 *
9406 	 * The reponse format is 2, per SPC-3.
9407 	 */
9408 	inq_ptr->response_format = SID_HiSup | 2;
9409 
9410 	inq_ptr->additional_length = sizeof(*inq_ptr) - 4;
9411 	CTL_DEBUG_PRINT(("additional_length = %d\n",
9412 			 inq_ptr->additional_length));
9413 
9414 	inq_ptr->spc3_flags = SPC3_SID_TPGS_IMPLICIT;
9415 	/* 16 bit addressing */
9416 	if (is_fc == 0)
9417 		inq_ptr->spc2_flags = SPC2_SID_ADDR16;
9418 	/* XXX set the SID_MultiP bit here if we're actually going to
9419 	   respond on multiple ports */
9420 	inq_ptr->spc2_flags |= SPC2_SID_MultiP;
9421 
9422 	/* 16 bit data bus, synchronous transfers */
9423 	/* XXX these flags don't apply for FC */
9424 	if (is_fc == 0)
9425 		inq_ptr->flags = SID_WBus16 | SID_Sync;
9426 	/*
9427 	 * XXX KDM do we want to support tagged queueing on the control
9428 	 * device at all?
9429 	 */
9430 	if ((lun == NULL)
9431 	 || (lun->be_lun->lun_type != T_PROCESSOR))
9432 		inq_ptr->flags |= SID_CmdQue;
9433 	/*
9434 	 * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
9435 	 * We have 8 bytes for the vendor name, and 16 bytes for the device
9436 	 * name and 4 bytes for the revision.
9437 	 */
9438 	strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
9439 	if (lun == NULL) {
9440 		strcpy(inq_ptr->product, CTL_DIRECT_PRODUCT);
9441 	} else {
9442 		switch (lun->be_lun->lun_type) {
9443 		case T_DIRECT:
9444 			strcpy(inq_ptr->product, CTL_DIRECT_PRODUCT);
9445 			break;
9446 		case T_PROCESSOR:
9447 			strcpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT);
9448 			break;
9449 		default:
9450 			strcpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT);
9451 			break;
9452 		}
9453 	}
9454 
9455 	/*
9456 	 * XXX make this a macro somewhere so it automatically gets
9457 	 * incremented when we make changes.
9458 	 */
9459 	strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
9460 
9461 	/*
9462 	 * For parallel SCSI, we support double transition and single
9463 	 * transition clocking.  We also support QAS (Quick Arbitration
9464 	 * and Selection) and Information Unit transfers on both the
9465 	 * control and array devices.
9466 	 */
9467 	if (is_fc == 0)
9468 		inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
9469 				    SID_SPI_IUS;
9470 
9471 	/* SAM-3 */
9472 	scsi_ulto2b(0x0060, inq_ptr->version1);
9473 	/* SPC-3 (no version claimed) XXX should we claim a version? */
9474 	scsi_ulto2b(0x0300, inq_ptr->version2);
9475 	if (is_fc) {
9476 		/* FCP-2 ANSI INCITS.350:2003 */
9477 		scsi_ulto2b(0x0917, inq_ptr->version3);
9478 	} else {
9479 		/* SPI-4 ANSI INCITS.362:200x */
9480 		scsi_ulto2b(0x0B56, inq_ptr->version3);
9481 	}
9482 
9483 	if (lun == NULL) {
9484 		/* SBC-2 (no version claimed) XXX should we claim a version? */
9485 		scsi_ulto2b(0x0320, inq_ptr->version4);
9486 	} else {
9487 		switch (lun->be_lun->lun_type) {
9488 		case T_DIRECT:
9489 			/*
9490 			 * SBC-2 (no version claimed) XXX should we claim a
9491 			 * version?
9492 			 */
9493 			scsi_ulto2b(0x0320, inq_ptr->version4);
9494 			break;
9495 		case T_PROCESSOR:
9496 		default:
9497 			break;
9498 		}
9499 	}
9500 
9501 	ctsio->scsi_status = SCSI_STATUS_OK;
9502 	if (ctsio->kern_data_len > 0) {
9503 		ctsio->be_move_done = ctl_config_move_done;
9504 		ctl_datamove((union ctl_io *)ctsio);
9505 	} else {
9506 		ctsio->io_hdr.status = CTL_SUCCESS;
9507 		ctl_done((union ctl_io *)ctsio);
9508 	}
9509 
9510 	return (CTL_RETVAL_COMPLETE);
9511 }
9512 
9513 int
9514 ctl_inquiry(struct ctl_scsiio *ctsio)
9515 {
9516 	struct scsi_inquiry *cdb;
9517 	int retval;
9518 
9519 	cdb = (struct scsi_inquiry *)ctsio->cdb;
9520 
9521 	retval = 0;
9522 
9523 	CTL_DEBUG_PRINT(("ctl_inquiry\n"));
9524 
9525 	/*
9526 	 * Right now, we don't support the CmdDt inquiry information.
9527 	 * This would be nice to support in the future.  When we do
9528 	 * support it, we should change this test so that it checks to make
9529 	 * sure SI_EVPD and SI_CMDDT aren't both set at the same time.
9530 	 */
9531 #ifdef notyet
9532 	if (((cdb->byte2 & SI_EVPD)
9533 	 && (cdb->byte2 & SI_CMDDT)))
9534 #endif
9535 	if (cdb->byte2 & SI_CMDDT) {
9536 		/*
9537 		 * Point to the SI_CMDDT bit.  We might change this
9538 		 * when we support SI_CMDDT, but since both bits would be
9539 		 * "wrong", this should probably just stay as-is then.
9540 		 */
9541 		ctl_set_invalid_field(ctsio,
9542 				      /*sks_valid*/ 1,
9543 				      /*command*/ 1,
9544 				      /*field*/ 1,
9545 				      /*bit_valid*/ 1,
9546 				      /*bit*/ 1);
9547 		ctl_done((union ctl_io *)ctsio);
9548 		return (CTL_RETVAL_COMPLETE);
9549 	}
9550 	if (cdb->byte2 & SI_EVPD)
9551 		retval = ctl_inquiry_evpd(ctsio);
9552 #ifdef notyet
9553 	else if (cdb->byte2 & SI_CMDDT)
9554 		retval = ctl_inquiry_cmddt(ctsio);
9555 #endif
9556 	else
9557 		retval = ctl_inquiry_std(ctsio);
9558 
9559 	return (retval);
9560 }
9561 
9562 /*
9563  * For known CDB types, parse the LBA and length.
9564  */
9565 static int
9566 ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint32_t *len)
9567 {
9568 	if (io->io_hdr.io_type != CTL_IO_SCSI)
9569 		return (1);
9570 
9571 	switch (io->scsiio.cdb[0]) {
9572 	case READ_6:
9573 	case WRITE_6: {
9574 		struct scsi_rw_6 *cdb;
9575 
9576 		cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
9577 
9578 		*lba = scsi_3btoul(cdb->addr);
9579 		/* only 5 bits are valid in the most significant address byte */
9580 		*lba &= 0x1fffff;
9581 		*len = cdb->length;
9582 		break;
9583 	}
9584 	case READ_10:
9585 	case WRITE_10: {
9586 		struct scsi_rw_10 *cdb;
9587 
9588 		cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
9589 
9590 		*lba = scsi_4btoul(cdb->addr);
9591 		*len = scsi_2btoul(cdb->length);
9592 		break;
9593 	}
9594 	case WRITE_VERIFY_10: {
9595 		struct scsi_write_verify_10 *cdb;
9596 
9597 		cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
9598 
9599 		*lba = scsi_4btoul(cdb->addr);
9600 		*len = scsi_2btoul(cdb->length);
9601 		break;
9602 	}
9603 	case READ_12:
9604 	case WRITE_12: {
9605 		struct scsi_rw_12 *cdb;
9606 
9607 		cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
9608 
9609 		*lba = scsi_4btoul(cdb->addr);
9610 		*len = scsi_4btoul(cdb->length);
9611 		break;
9612 	}
9613 	case WRITE_VERIFY_12: {
9614 		struct scsi_write_verify_12 *cdb;
9615 
9616 		cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
9617 
9618 		*lba = scsi_4btoul(cdb->addr);
9619 		*len = scsi_4btoul(cdb->length);
9620 		break;
9621 	}
9622 	case READ_16:
9623 	case WRITE_16: {
9624 		struct scsi_rw_16 *cdb;
9625 
9626 		cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
9627 
9628 		*lba = scsi_8btou64(cdb->addr);
9629 		*len = scsi_4btoul(cdb->length);
9630 		break;
9631 	}
9632 	case WRITE_VERIFY_16: {
9633 		struct scsi_write_verify_16 *cdb;
9634 
9635 		cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
9636 
9637 
9638 		*lba = scsi_8btou64(cdb->addr);
9639 		*len = scsi_4btoul(cdb->length);
9640 		break;
9641 	}
9642 	default:
9643 		return (1);
9644 		break; /* NOTREACHED */
9645 	}
9646 
9647 	return (0);
9648 }
9649 
9650 static ctl_action
9651 ctl_extent_check_lba(uint64_t lba1, uint32_t len1, uint64_t lba2, uint32_t len2)
9652 {
9653 	uint64_t endlba1, endlba2;
9654 
9655 	endlba1 = lba1 + len1 - 1;
9656 	endlba2 = lba2 + len2 - 1;
9657 
9658 	if ((endlba1 < lba2)
9659 	 || (endlba2 < lba1))
9660 		return (CTL_ACTION_PASS);
9661 	else
9662 		return (CTL_ACTION_BLOCK);
9663 }
9664 
9665 static ctl_action
9666 ctl_extent_check(union ctl_io *io1, union ctl_io *io2)
9667 {
9668 	uint64_t lba1, lba2;
9669 	uint32_t len1, len2;
9670 	int retval;
9671 
9672 	retval = ctl_get_lba_len(io1, &lba1, &len1);
9673 	if (retval != 0)
9674 		return (CTL_ACTION_ERROR);
9675 
9676 	retval = ctl_get_lba_len(io2, &lba2, &len2);
9677 	if (retval != 0)
9678 		return (CTL_ACTION_ERROR);
9679 
9680 	return (ctl_extent_check_lba(lba1, len1, lba2, len2));
9681 }
9682 
9683 static ctl_action
9684 ctl_check_for_blockage(union ctl_io *pending_io, union ctl_io *ooa_io)
9685 {
9686 	struct ctl_cmd_entry *pending_entry, *ooa_entry;
9687 	ctl_serialize_action *serialize_row;
9688 
9689 	/*
9690 	 * The initiator attempted multiple untagged commands at the same
9691 	 * time.  Can't do that.
9692 	 */
9693 	if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
9694 	 && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
9695 	 && ((pending_io->io_hdr.nexus.targ_port ==
9696 	      ooa_io->io_hdr.nexus.targ_port)
9697 	  && (pending_io->io_hdr.nexus.initid.id ==
9698 	      ooa_io->io_hdr.nexus.initid.id))
9699 	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
9700 		return (CTL_ACTION_OVERLAP);
9701 
9702 	/*
9703 	 * The initiator attempted to send multiple tagged commands with
9704 	 * the same ID.  (It's fine if different initiators have the same
9705 	 * tag ID.)
9706 	 *
9707 	 * Even if all of those conditions are true, we don't kill the I/O
9708 	 * if the command ahead of us has been aborted.  We won't end up
9709 	 * sending it to the FETD, and it's perfectly legal to resend a
9710 	 * command with the same tag number as long as the previous
9711 	 * instance of this tag number has been aborted somehow.
9712 	 */
9713 	if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
9714 	 && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
9715 	 && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
9716 	 && ((pending_io->io_hdr.nexus.targ_port ==
9717 	      ooa_io->io_hdr.nexus.targ_port)
9718 	  && (pending_io->io_hdr.nexus.initid.id ==
9719 	      ooa_io->io_hdr.nexus.initid.id))
9720 	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
9721 		return (CTL_ACTION_OVERLAP_TAG);
9722 
9723 	/*
9724 	 * If we get a head of queue tag, SAM-3 says that we should
9725 	 * immediately execute it.
9726 	 *
9727 	 * What happens if this command would normally block for some other
9728 	 * reason?  e.g. a request sense with a head of queue tag
9729 	 * immediately after a write.  Normally that would block, but this
9730 	 * will result in its getting executed immediately...
9731 	 *
9732 	 * We currently return "pass" instead of "skip", so we'll end up
9733 	 * going through the rest of the queue to check for overlapped tags.
9734 	 *
9735 	 * XXX KDM check for other types of blockage first??
9736 	 */
9737 	if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
9738 		return (CTL_ACTION_PASS);
9739 
9740 	/*
9741 	 * Ordered tags have to block until all items ahead of them
9742 	 * have completed.  If we get called with an ordered tag, we always
9743 	 * block, if something else is ahead of us in the queue.
9744 	 */
9745 	if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
9746 		return (CTL_ACTION_BLOCK);
9747 
9748 	/*
9749 	 * Simple tags get blocked until all head of queue and ordered tags
9750 	 * ahead of them have completed.  I'm lumping untagged commands in
9751 	 * with simple tags here.  XXX KDM is that the right thing to do?
9752 	 */
9753 	if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
9754 	  || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
9755 	 && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
9756 	  || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
9757 		return (CTL_ACTION_BLOCK);
9758 
9759 	pending_entry = &ctl_cmd_table[pending_io->scsiio.cdb[0]];
9760 	ooa_entry = &ctl_cmd_table[ooa_io->scsiio.cdb[0]];
9761 
9762 	serialize_row = ctl_serialize_table[ooa_entry->seridx];
9763 
9764 	switch (serialize_row[pending_entry->seridx]) {
9765 	case CTL_SER_BLOCK:
9766 		return (CTL_ACTION_BLOCK);
9767 		break; /* NOTREACHED */
9768 	case CTL_SER_EXTENT:
9769 		return (ctl_extent_check(pending_io, ooa_io));
9770 		break; /* NOTREACHED */
9771 	case CTL_SER_PASS:
9772 		return (CTL_ACTION_PASS);
9773 		break; /* NOTREACHED */
9774 	case CTL_SER_SKIP:
9775 		return (CTL_ACTION_SKIP);
9776 		break;
9777 	default:
9778 		panic("invalid serialization value %d",
9779 		      serialize_row[pending_entry->seridx]);
9780 		break; /* NOTREACHED */
9781 	}
9782 
9783 	return (CTL_ACTION_ERROR);
9784 }
9785 
9786 /*
9787  * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
9788  * Assumptions:
9789  * - caller holds ctl_lock
9790  * - pending_io is generally either incoming, or on the blocked queue
9791  * - starting I/O is the I/O we want to start the check with.
9792  */
9793 static ctl_action
9794 ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
9795 	      union ctl_io *starting_io)
9796 {
9797 	union ctl_io *ooa_io;
9798 	ctl_action action;
9799 
9800 	/*
9801 	 * Run back along the OOA queue, starting with the current
9802 	 * blocked I/O and going through every I/O before it on the
9803 	 * queue.  If starting_io is NULL, we'll just end up returning
9804 	 * CTL_ACTION_PASS.
9805 	 */
9806 	for (ooa_io = starting_io; ooa_io != NULL;
9807 	     ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
9808 	     ooa_links)){
9809 
9810 		/*
9811 		 * This routine just checks to see whether
9812 		 * cur_blocked is blocked by ooa_io, which is ahead
9813 		 * of it in the queue.  It doesn't queue/dequeue
9814 		 * cur_blocked.
9815 		 */
9816 		action = ctl_check_for_blockage(pending_io, ooa_io);
9817 		switch (action) {
9818 		case CTL_ACTION_BLOCK:
9819 		case CTL_ACTION_OVERLAP:
9820 		case CTL_ACTION_OVERLAP_TAG:
9821 		case CTL_ACTION_SKIP:
9822 		case CTL_ACTION_ERROR:
9823 			return (action);
9824 			break; /* NOTREACHED */
9825 		case CTL_ACTION_PASS:
9826 			break;
9827 		default:
9828 			panic("invalid action %d", action);
9829 			break;  /* NOTREACHED */
9830 		}
9831 	}
9832 
9833 	return (CTL_ACTION_PASS);
9834 }
9835 
9836 /*
9837  * Assumptions:
9838  * - An I/O has just completed, and has been removed from the per-LUN OOA
9839  *   queue, so some items on the blocked queue may now be unblocked.
9840  * - The caller holds ctl_softc->ctl_lock
9841  */
9842 static int
9843 ctl_check_blocked(struct ctl_lun *lun)
9844 {
9845 	union ctl_io *cur_blocked, *next_blocked;
9846 
9847 	/*
9848 	 * Run forward from the head of the blocked queue, checking each
9849 	 * entry against the I/Os prior to it on the OOA queue to see if
9850 	 * there is still any blockage.
9851 	 *
9852 	 * We cannot use the TAILQ_FOREACH() macro, because it can't deal
9853 	 * with our removing a variable on it while it is traversing the
9854 	 * list.
9855 	 */
9856 	for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
9857 	     cur_blocked != NULL; cur_blocked = next_blocked) {
9858 		union ctl_io *prev_ooa;
9859 		ctl_action action;
9860 
9861 		next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
9862 							  blocked_links);
9863 
9864 		prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
9865 						      ctl_ooaq, ooa_links);
9866 
9867 		/*
9868 		 * If cur_blocked happens to be the first item in the OOA
9869 		 * queue now, prev_ooa will be NULL, and the action
9870 		 * returned will just be CTL_ACTION_PASS.
9871 		 */
9872 		action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
9873 
9874 		switch (action) {
9875 		case CTL_ACTION_BLOCK:
9876 			/* Nothing to do here, still blocked */
9877 			break;
9878 		case CTL_ACTION_OVERLAP:
9879 		case CTL_ACTION_OVERLAP_TAG:
9880 			/*
9881 			 * This shouldn't happen!  In theory we've already
9882 			 * checked this command for overlap...
9883 			 */
9884 			break;
9885 		case CTL_ACTION_PASS:
9886 		case CTL_ACTION_SKIP: {
9887 			struct ctl_softc *softc;
9888 			struct ctl_cmd_entry *entry;
9889 			uint32_t initidx;
9890 			uint8_t opcode;
9891 			int isc_retval;
9892 
9893 			/*
9894 			 * The skip case shouldn't happen, this transaction
9895 			 * should have never made it onto the blocked queue.
9896 			 */
9897 			/*
9898 			 * This I/O is no longer blocked, we can remove it
9899 			 * from the blocked queue.  Since this is a TAILQ
9900 			 * (doubly linked list), we can do O(1) removals
9901 			 * from any place on the list.
9902 			 */
9903 			TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
9904 				     blocked_links);
9905 			cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
9906 
9907 			if (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC){
9908 				/*
9909 				 * Need to send IO back to original side to
9910 				 * run
9911 				 */
9912 				union ctl_ha_msg msg_info;
9913 
9914 				msg_info.hdr.original_sc =
9915 					cur_blocked->io_hdr.original_sc;
9916 				msg_info.hdr.serializing_sc = cur_blocked;
9917 				msg_info.hdr.msg_type = CTL_MSG_R2R;
9918 				if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
9919 				     &msg_info, sizeof(msg_info), 0)) >
9920 				     CTL_HA_STATUS_SUCCESS) {
9921 					printf("CTL:Check Blocked error from "
9922 					       "ctl_ha_msg_send %d\n",
9923 					       isc_retval);
9924 				}
9925 				break;
9926 			}
9927 			opcode = cur_blocked->scsiio.cdb[0];
9928 			entry = &ctl_cmd_table[opcode];
9929 			softc = control_softc;
9930 
9931 			initidx = ctl_get_initindex(&cur_blocked->io_hdr.nexus);
9932 
9933 			/*
9934 			 * Check this I/O for LUN state changes that may
9935 			 * have happened while this command was blocked.
9936 			 * The LUN state may have been changed by a command
9937 			 * ahead of us in the queue, so we need to re-check
9938 			 * for any states that can be caused by SCSI
9939 			 * commands.
9940 			 */
9941 			if (ctl_scsiio_lun_check(softc, lun, entry,
9942 						 &cur_blocked->scsiio) == 0) {
9943 				cur_blocked->io_hdr.flags |=
9944 				                      CTL_FLAG_IS_WAS_ON_RTR;
9945 				STAILQ_INSERT_TAIL(&lun->ctl_softc->rtr_queue,
9946 						   &cur_blocked->io_hdr, links);
9947 				/*
9948 				 * In the non CTL_DONE_THREAD case, we need
9949 				 * to wake up the work thread here.  When
9950 				 * we're processing completed requests from
9951 				 * the work thread context, we'll pop back
9952 				 * around and end up pulling things off the
9953 				 * RtR queue.  When we aren't processing
9954 				 * things from the work thread context,
9955 				 * though, we won't ever check the RtR queue.
9956 				 * So we need to wake up the thread to clear
9957 				 * things off the queue.  Otherwise this
9958 				 * transaction will just sit on the RtR queue
9959 				 * until a new I/O comes in.  (Which may or
9960 				 * may not happen...)
9961 				 */
9962 #ifndef CTL_DONE_THREAD
9963 				ctl_wakeup_thread();
9964 #endif
9965 			} else
9966 				ctl_done_lock(cur_blocked, /*have_lock*/ 1);
9967 			break;
9968 		}
9969 		default:
9970 			/*
9971 			 * This probably shouldn't happen -- we shouldn't
9972 			 * get CTL_ACTION_ERROR, or anything else.
9973 			 */
9974 			break;
9975 		}
9976 	}
9977 
9978 	return (CTL_RETVAL_COMPLETE);
9979 }
9980 
9981 /*
9982  * This routine (with one exception) checks LUN flags that can be set by
9983  * commands ahead of us in the OOA queue.  These flags have to be checked
9984  * when a command initially comes in, and when we pull a command off the
9985  * blocked queue and are preparing to execute it.  The reason we have to
9986  * check these flags for commands on the blocked queue is that the LUN
9987  * state may have been changed by a command ahead of us while we're on the
9988  * blocked queue.
9989  *
9990  * Ordering is somewhat important with these checks, so please pay
9991  * careful attention to the placement of any new checks.
9992  */
9993 static int
9994 ctl_scsiio_lun_check(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
9995 		     struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
9996 {
9997 	int retval;
9998 
9999 	retval = 0;
10000 
10001 	/*
10002 	 * If this shelf is a secondary shelf controller, we have to reject
10003 	 * any media access commands.
10004 	 */
10005 #if 0
10006 	/* No longer needed for HA */
10007 	if (((ctl_softc->flags & CTL_FLAG_MASTER_SHELF) == 0)
10008 	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_SECONDARY) == 0)) {
10009 		ctl_set_lun_standby(ctsio);
10010 		retval = 1;
10011 		goto bailout;
10012 	}
10013 #endif
10014 
10015 	/*
10016 	 * Check for a reservation conflict.  If this command isn't allowed
10017 	 * even on reserved LUNs, and if this initiator isn't the one who
10018 	 * reserved us, reject the command with a reservation conflict.
10019 	 */
10020 	if ((lun->flags & CTL_LUN_RESERVED)
10021 	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
10022 		if ((ctsio->io_hdr.nexus.initid.id != lun->rsv_nexus.initid.id)
10023 		 || (ctsio->io_hdr.nexus.targ_port != lun->rsv_nexus.targ_port)
10024 		 || (ctsio->io_hdr.nexus.targ_target.id !=
10025 		     lun->rsv_nexus.targ_target.id)) {
10026 			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
10027 			ctsio->io_hdr.status = CTL_SCSI_ERROR;
10028 			retval = 1;
10029 			goto bailout;
10030 		}
10031 	}
10032 
10033 	if ( (lun->flags & CTL_LUN_PR_RESERVED)
10034 	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV) == 0)) {
10035 		uint32_t residx;
10036 
10037 		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
10038 		/*
10039 		 * if we aren't registered or it's a res holder type
10040 		 * reservation and this isn't the res holder then set a
10041 		 * conflict.
10042 		 * NOTE: Commands which might be allowed on write exclusive
10043 		 * type reservations are checked in the particular command
10044 		 * for a conflict. Read and SSU are the only ones.
10045 		 */
10046 		if (!lun->per_res[residx].registered
10047 		 || (residx != lun->pr_res_idx && lun->res_type < 4)) {
10048 			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
10049 			ctsio->io_hdr.status = CTL_SCSI_ERROR;
10050 			retval = 1;
10051 			goto bailout;
10052 		}
10053 
10054 	}
10055 
10056 	if ((lun->flags & CTL_LUN_OFFLINE)
10057 	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_OFFLINE) == 0)) {
10058 		ctl_set_lun_not_ready(ctsio);
10059 		retval = 1;
10060 		goto bailout;
10061 	}
10062 
10063 	/*
10064 	 * If the LUN is stopped, see if this particular command is allowed
10065 	 * for a stopped lun.  Otherwise, reject it with 0x04,0x02.
10066 	 */
10067 	if ((lun->flags & CTL_LUN_STOPPED)
10068 	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_STOPPED) == 0)) {
10069 		/* "Logical unit not ready, initializing cmd. required" */
10070 		ctl_set_lun_stopped(ctsio);
10071 		retval = 1;
10072 		goto bailout;
10073 	}
10074 
10075 	if ((lun->flags & CTL_LUN_INOPERABLE)
10076 	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_INOPERABLE) == 0)) {
10077 		/* "Medium format corrupted" */
10078 		ctl_set_medium_format_corrupted(ctsio);
10079 		retval = 1;
10080 		goto bailout;
10081 	}
10082 
10083 bailout:
10084 	return (retval);
10085 
10086 }
10087 
10088 static void
10089 ctl_failover_io(union ctl_io *io, int have_lock)
10090 {
10091 	ctl_set_busy(&io->scsiio);
10092 	ctl_done_lock(io, have_lock);
10093 }
10094 
10095 static void
10096 ctl_failover(void)
10097 {
10098 	struct ctl_lun *lun;
10099 	struct ctl_softc *ctl_softc;
10100 	union ctl_io *next_io, *pending_io;
10101 	union ctl_io *io;
10102 	int lun_idx;
10103 	int i;
10104 
10105 	ctl_softc = control_softc;
10106 
10107 	mtx_lock(&ctl_softc->ctl_lock);
10108 	/*
10109 	 * Remove any cmds from the other SC from the rtr queue.  These
10110 	 * will obviously only be for LUNs for which we're the primary.
10111 	 * We can't send status or get/send data for these commands.
10112 	 * Since they haven't been executed yet, we can just remove them.
10113 	 * We'll either abort them or delete them below, depending on
10114 	 * which HA mode we're in.
10115 	 */
10116 	for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->rtr_queue);
10117 	     io != NULL; io = next_io) {
10118 		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
10119 		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
10120 			STAILQ_REMOVE(&ctl_softc->rtr_queue, &io->io_hdr,
10121 				      ctl_io_hdr, links);
10122 	}
10123 
10124 	for (lun_idx=0; lun_idx < ctl_softc->num_luns; lun_idx++) {
10125 		lun = ctl_softc->ctl_luns[lun_idx];
10126 		if (lun==NULL)
10127 			continue;
10128 
10129 		/*
10130 		 * Processor LUNs are primary on both sides.
10131 		 * XXX will this always be true?
10132 		 */
10133 		if (lun->be_lun->lun_type == T_PROCESSOR)
10134 			continue;
10135 
10136 		if ((lun->flags & CTL_LUN_PRIMARY_SC)
10137 		 && (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
10138 			printf("FAILOVER: primary lun %d\n", lun_idx);
10139 		        /*
10140 			 * Remove all commands from the other SC. First from the
10141 			 * blocked queue then from the ooa queue. Once we have
10142 			 * removed them. Call ctl_check_blocked to see if there
10143 			 * is anything that can run.
10144 			 */
10145 			for (io = (union ctl_io *)TAILQ_FIRST(
10146 			     &lun->blocked_queue); io != NULL; io = next_io) {
10147 
10148 		        	next_io = (union ctl_io *)TAILQ_NEXT(
10149 				    &io->io_hdr, blocked_links);
10150 
10151 				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
10152 					TAILQ_REMOVE(&lun->blocked_queue,
10153 						     &io->io_hdr,blocked_links);
10154 					io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
10155 					TAILQ_REMOVE(&lun->ooa_queue,
10156 						     &io->io_hdr, ooa_links);
10157 
10158 					ctl_free_io_internal(io, 1);
10159 				}
10160 			}
10161 
10162 			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
10163 	     		     io != NULL; io = next_io) {
10164 
10165 		        	next_io = (union ctl_io *)TAILQ_NEXT(
10166 				    &io->io_hdr, ooa_links);
10167 
10168 				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
10169 
10170 					TAILQ_REMOVE(&lun->ooa_queue,
10171 						&io->io_hdr,
10172 					     	ooa_links);
10173 
10174 					ctl_free_io_internal(io, 1);
10175 				}
10176 			}
10177 			ctl_check_blocked(lun);
10178 		} else if ((lun->flags & CTL_LUN_PRIMARY_SC)
10179 			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
10180 
10181 			printf("FAILOVER: primary lun %d\n", lun_idx);
10182 			/*
10183 			 * Abort all commands from the other SC.  We can't
10184 			 * send status back for them now.  These should get
10185 			 * cleaned up when they are completed or come out
10186 			 * for a datamove operation.
10187 			 */
10188 			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
10189 	     		     io != NULL; io = next_io) {
10190 		        	next_io = (union ctl_io *)TAILQ_NEXT(
10191 					&io->io_hdr, ooa_links);
10192 
10193 				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
10194 					io->io_hdr.flags |= CTL_FLAG_ABORT;
10195 			}
10196 		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
10197 			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
10198 
10199 			printf("FAILOVER: secondary lun %d\n", lun_idx);
10200 
10201 			lun->flags |= CTL_LUN_PRIMARY_SC;
10202 
10203 			/*
10204 			 * We send all I/O that was sent to this controller
10205 			 * and redirected to the other side back with
10206 			 * busy status, and have the initiator retry it.
10207 			 * Figuring out how much data has been transferred,
10208 			 * etc. and picking up where we left off would be
10209 			 * very tricky.
10210 			 *
10211 			 * XXX KDM need to remove I/O from the blocked
10212 			 * queue as well!
10213 			 */
10214 			for (pending_io = (union ctl_io *)TAILQ_FIRST(
10215 			     &lun->ooa_queue); pending_io != NULL;
10216 			     pending_io = next_io) {
10217 
10218 				next_io =  (union ctl_io *)TAILQ_NEXT(
10219 					&pending_io->io_hdr, ooa_links);
10220 
10221 				pending_io->io_hdr.flags &=
10222 					~CTL_FLAG_SENT_2OTHER_SC;
10223 
10224 				if (pending_io->io_hdr.flags &
10225 				    CTL_FLAG_IO_ACTIVE) {
10226 					pending_io->io_hdr.flags |=
10227 						CTL_FLAG_FAILOVER;
10228 				} else {
10229 					ctl_set_busy(&pending_io->scsiio);
10230 					ctl_done_lock(pending_io,
10231 						      /*have_lock*/1);
10232 				}
10233 			}
10234 
10235 			/*
10236 			 * Build Unit Attention
10237 			 */
10238 			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
10239 				lun->pending_sense[i].ua_pending |=
10240 				                     CTL_UA_ASYM_ACC_CHANGE;
10241 			}
10242 		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
10243 			&& (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
10244 			printf("FAILOVER: secondary lun %d\n", lun_idx);
10245 			/*
10246 			 * if the first io on the OOA is not on the RtR queue
10247 			 * add it.
10248 			 */
10249 			lun->flags |= CTL_LUN_PRIMARY_SC;
10250 
10251 			pending_io = (union ctl_io *)TAILQ_FIRST(
10252 			    &lun->ooa_queue);
10253 			if (pending_io==NULL) {
10254 				printf("Nothing on OOA queue\n");
10255 				continue;
10256 			}
10257 
10258 			pending_io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
10259 			if ((pending_io->io_hdr.flags &
10260 			     CTL_FLAG_IS_WAS_ON_RTR) == 0) {
10261 				pending_io->io_hdr.flags |=
10262 				    CTL_FLAG_IS_WAS_ON_RTR;
10263 				STAILQ_INSERT_TAIL(&ctl_softc->rtr_queue,
10264 						   &pending_io->io_hdr, links);
10265 			}
10266 #if 0
10267 			else
10268 			{
10269 				printf("Tag 0x%04x is running\n",
10270 				      pending_io->scsiio.tag_num);
10271 			}
10272 #endif
10273 
10274 			next_io = (union ctl_io *)TAILQ_NEXT(
10275 			    &pending_io->io_hdr, ooa_links);
10276 			for (pending_io=next_io; pending_io != NULL;
10277 			     pending_io = next_io) {
10278 				pending_io->io_hdr.flags &=
10279 				    ~CTL_FLAG_SENT_2OTHER_SC;
10280 				next_io = (union ctl_io *)TAILQ_NEXT(
10281 					&pending_io->io_hdr, ooa_links);
10282 				if (pending_io->io_hdr.flags &
10283 				    CTL_FLAG_IS_WAS_ON_RTR) {
10284 #if 0
10285 				        printf("Tag 0x%04x is running\n",
10286 				      		pending_io->scsiio.tag_num);
10287 #endif
10288 					continue;
10289 				}
10290 
10291 				switch (ctl_check_ooa(lun, pending_io,
10292 			            (union ctl_io *)TAILQ_PREV(
10293 				    &pending_io->io_hdr, ctl_ooaq,
10294 				    ooa_links))) {
10295 
10296 				case CTL_ACTION_BLOCK:
10297 					TAILQ_INSERT_TAIL(&lun->blocked_queue,
10298 							  &pending_io->io_hdr,
10299 							  blocked_links);
10300 					pending_io->io_hdr.flags |=
10301 					    CTL_FLAG_BLOCKED;
10302 					break;
10303 				case CTL_ACTION_PASS:
10304 				case CTL_ACTION_SKIP:
10305 					pending_io->io_hdr.flags |=
10306 					    CTL_FLAG_IS_WAS_ON_RTR;
10307 					STAILQ_INSERT_TAIL(
10308 					    &ctl_softc->rtr_queue,
10309 					    &pending_io->io_hdr, links);
10310 					break;
10311 				case CTL_ACTION_OVERLAP:
10312 					ctl_set_overlapped_cmd(
10313 					    (struct ctl_scsiio *)pending_io);
10314 					ctl_done_lock(pending_io,
10315 						      /*have_lock*/ 1);
10316 					break;
10317 				case CTL_ACTION_OVERLAP_TAG:
10318 					ctl_set_overlapped_tag(
10319 					    (struct ctl_scsiio *)pending_io,
10320 					    pending_io->scsiio.tag_num & 0xff);
10321 					ctl_done_lock(pending_io,
10322 						      /*have_lock*/ 1);
10323 					break;
10324 				case CTL_ACTION_ERROR:
10325 				default:
10326 					ctl_set_internal_failure(
10327 						(struct ctl_scsiio *)pending_io,
10328 						0,  // sks_valid
10329 						0); //retry count
10330 					ctl_done_lock(pending_io,
10331 						      /*have_lock*/ 1);
10332 					break;
10333 				}
10334 			}
10335 
10336 			/*
10337 			 * Build Unit Attention
10338 			 */
10339 			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
10340 				lun->pending_sense[i].ua_pending |=
10341 				                     CTL_UA_ASYM_ACC_CHANGE;
10342 			}
10343 		} else {
10344 			panic("Unhandled HA mode failover, LUN flags = %#x, "
10345 			      "ha_mode = #%x", lun->flags, ctl_softc->ha_mode);
10346 		}
10347 	}
10348 	ctl_pause_rtr = 0;
10349 	mtx_unlock(&ctl_softc->ctl_lock);
10350 }
10351 
10352 static int
10353 ctl_scsiio_precheck(struct ctl_softc *ctl_softc, struct ctl_scsiio *ctsio)
10354 {
10355 	struct ctl_lun *lun;
10356 	struct ctl_cmd_entry *entry;
10357 	uint8_t opcode;
10358 	uint32_t initidx;
10359 	int retval;
10360 
10361 	retval = 0;
10362 
10363 	lun = NULL;
10364 
10365 	opcode = ctsio->cdb[0];
10366 
10367 	mtx_lock(&ctl_softc->ctl_lock);
10368 
10369 	if ((ctsio->io_hdr.nexus.targ_lun < CTL_MAX_LUNS)
10370 	 && (ctl_softc->ctl_luns[ctsio->io_hdr.nexus.targ_lun] != NULL)) {
10371 		lun = ctl_softc->ctl_luns[ctsio->io_hdr.nexus.targ_lun];
10372 		/*
10373 		 * If the LUN is invalid, pretend that it doesn't exist.
10374 		 * It will go away as soon as all pending I/O has been
10375 		 * completed.
10376 		 */
10377 		if (lun->flags & CTL_LUN_DISABLED) {
10378 			lun = NULL;
10379 		} else {
10380 			ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
10381 			ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
10382 				lun->be_lun;
10383 			if (lun->be_lun->lun_type == T_PROCESSOR) {
10384 				ctsio->io_hdr.flags |= CTL_FLAG_CONTROL_DEV;
10385 			}
10386 		}
10387 	} else {
10388 		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
10389 		ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
10390 	}
10391 
10392 	entry = &ctl_cmd_table[opcode];
10393 
10394 	ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
10395 	ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
10396 
10397 	/*
10398 	 * Check to see whether we can send this command to LUNs that don't
10399 	 * exist.  This should pretty much only be the case for inquiry
10400 	 * and request sense.  Further checks, below, really require having
10401 	 * a LUN, so we can't really check the command anymore.  Just put
10402 	 * it on the rtr queue.
10403 	 */
10404 	if (lun == NULL) {
10405 		if (entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS)
10406 			goto queue_rtr;
10407 
10408 		ctl_set_unsupported_lun(ctsio);
10409 		mtx_unlock(&ctl_softc->ctl_lock);
10410 		ctl_done((union ctl_io *)ctsio);
10411 		goto bailout;
10412 	} else {
10413 		/*
10414 		 * Every I/O goes into the OOA queue for a particular LUN, and
10415 		 * stays there until completion.
10416 		 */
10417 		TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
10418 
10419 		/*
10420 		 * Make sure we support this particular command on this LUN.
10421 		 * e.g., we don't support writes to the control LUN.
10422 		 */
10423 		switch (lun->be_lun->lun_type) {
10424 		case T_PROCESSOR:
10425 		 	if (((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0)
10426 			 && ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS)
10427 			      == 0)) {
10428 				ctl_set_invalid_opcode(ctsio);
10429 				mtx_unlock(&ctl_softc->ctl_lock);
10430 				ctl_done((union ctl_io *)ctsio);
10431 				goto bailout;
10432 			}
10433 			break;
10434 		case T_DIRECT:
10435 			if (((entry->flags & CTL_CMD_FLAG_OK_ON_SLUN) == 0)
10436 			 && ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS)
10437 			      == 0)){
10438 				ctl_set_invalid_opcode(ctsio);
10439 				mtx_unlock(&ctl_softc->ctl_lock);
10440 				ctl_done((union ctl_io *)ctsio);
10441 				goto bailout;
10442 			}
10443 			break;
10444 		default:
10445 			printf("Unsupported CTL LUN type %d\n",
10446 			       lun->be_lun->lun_type);
10447 			panic("Unsupported CTL LUN type %d\n",
10448 			      lun->be_lun->lun_type);
10449 			break; /* NOTREACHED */
10450 		}
10451 	}
10452 
10453 	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
10454 
10455 	/*
10456 	 * If we've got a request sense, it'll clear the contingent
10457 	 * allegiance condition.  Otherwise, if we have a CA condition for
10458 	 * this initiator, clear it, because it sent down a command other
10459 	 * than request sense.
10460 	 */
10461 	if ((opcode != REQUEST_SENSE)
10462 	 && (ctl_is_set(lun->have_ca, initidx)))
10463 		ctl_clear_mask(lun->have_ca, initidx);
10464 
10465 	/*
10466 	 * If the command has this flag set, it handles its own unit
10467 	 * attention reporting, we shouldn't do anything.  Otherwise we
10468 	 * check for any pending unit attentions, and send them back to the
10469 	 * initiator.  We only do this when a command initially comes in,
10470 	 * not when we pull it off the blocked queue.
10471 	 *
10472 	 * According to SAM-3, section 5.3.2, the order that things get
10473 	 * presented back to the host is basically unit attentions caused
10474 	 * by some sort of reset event, busy status, reservation conflicts
10475 	 * or task set full, and finally any other status.
10476 	 *
10477 	 * One issue here is that some of the unit attentions we report
10478 	 * don't fall into the "reset" category (e.g. "reported luns data
10479 	 * has changed").  So reporting it here, before the reservation
10480 	 * check, may be technically wrong.  I guess the only thing to do
10481 	 * would be to check for and report the reset events here, and then
10482 	 * check for the other unit attention types after we check for a
10483 	 * reservation conflict.
10484 	 *
10485 	 * XXX KDM need to fix this
10486 	 */
10487 	if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
10488 		ctl_ua_type ua_type;
10489 
10490 		ua_type = lun->pending_sense[initidx].ua_pending;
10491 		if (ua_type != CTL_UA_NONE) {
10492 			scsi_sense_data_type sense_format;
10493 
10494 			if (lun != NULL)
10495 				sense_format = (lun->flags &
10496 				    CTL_LUN_SENSE_DESC) ? SSD_TYPE_DESC :
10497 				    SSD_TYPE_FIXED;
10498 			else
10499 				sense_format = SSD_TYPE_FIXED;
10500 
10501 			ua_type = ctl_build_ua(ua_type, &ctsio->sense_data,
10502 					       sense_format);
10503 			if (ua_type != CTL_UA_NONE) {
10504 				ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
10505 				ctsio->io_hdr.status = CTL_SCSI_ERROR |
10506 						       CTL_AUTOSENSE;
10507 				ctsio->sense_len = SSD_FULL_SIZE;
10508 				lun->pending_sense[initidx].ua_pending &=
10509 					~ua_type;
10510 				mtx_unlock(&ctl_softc->ctl_lock);
10511 				ctl_done((union ctl_io *)ctsio);
10512 				goto bailout;
10513 			}
10514 		}
10515 	}
10516 
10517 
10518 	if (ctl_scsiio_lun_check(ctl_softc, lun, entry, ctsio) != 0) {
10519 		mtx_unlock(&ctl_softc->ctl_lock);
10520 		ctl_done((union ctl_io *)ctsio);
10521 		goto bailout;
10522 	}
10523 
10524 	/*
10525 	 * XXX CHD this is where we want to send IO to other side if
10526 	 * this LUN is secondary on this SC. We will need to make a copy
10527 	 * of the IO and flag the IO on this side as SENT_2OTHER and the flag
10528 	 * the copy we send as FROM_OTHER.
10529 	 * We also need to stuff the address of the original IO so we can
10530 	 * find it easily. Something similar will need be done on the other
10531 	 * side so when we are done we can find the copy.
10532 	 */
10533 	if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
10534 		union ctl_ha_msg msg_info;
10535 		int isc_retval;
10536 
10537 		ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
10538 
10539 		msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
10540 		msg_info.hdr.original_sc = (union ctl_io *)ctsio;
10541 #if 0
10542 		printf("1. ctsio %p\n", ctsio);
10543 #endif
10544 		msg_info.hdr.serializing_sc = NULL;
10545 		msg_info.hdr.nexus = ctsio->io_hdr.nexus;
10546 		msg_info.scsi.tag_num = ctsio->tag_num;
10547 		msg_info.scsi.tag_type = ctsio->tag_type;
10548 		memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
10549 
10550 		ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
10551 
10552 		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
10553 		    (void *)&msg_info, sizeof(msg_info), 0)) >
10554 		    CTL_HA_STATUS_SUCCESS) {
10555 			printf("CTL:precheck, ctl_ha_msg_send returned %d\n",
10556 			       isc_retval);
10557 			printf("CTL:opcode is %x\n",opcode);
10558 		} else {
10559 #if 0
10560 			printf("CTL:Precheck sent msg, opcode is %x\n",opcode);
10561 #endif
10562 		}
10563 
10564 		/*
10565 		 * XXX KDM this I/O is off the incoming queue, but hasn't
10566 		 * been inserted on any other queue.  We may need to come
10567 		 * up with a holding queue while we wait for serialization
10568 		 * so that we have an idea of what we're waiting for from
10569 		 * the other side.
10570 		 */
10571 		goto bailout_unlock;
10572 	}
10573 
10574 	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
10575 			      (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
10576 			      ctl_ooaq, ooa_links))) {
10577 	case CTL_ACTION_BLOCK:
10578 		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
10579 		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
10580 				  blocked_links);
10581 		goto bailout_unlock;
10582 		break; /* NOTREACHED */
10583 	case CTL_ACTION_PASS:
10584 	case CTL_ACTION_SKIP:
10585 		goto queue_rtr;
10586 		break; /* NOTREACHED */
10587 	case CTL_ACTION_OVERLAP:
10588 		ctl_set_overlapped_cmd(ctsio);
10589 		mtx_unlock(&ctl_softc->ctl_lock);
10590 		ctl_done((union ctl_io *)ctsio);
10591 		goto bailout;
10592 		break; /* NOTREACHED */
10593 	case CTL_ACTION_OVERLAP_TAG:
10594 		ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
10595 		mtx_unlock(&ctl_softc->ctl_lock);
10596 		ctl_done((union ctl_io *)ctsio);
10597 		goto bailout;
10598 		break; /* NOTREACHED */
10599 	case CTL_ACTION_ERROR:
10600 	default:
10601 		ctl_set_internal_failure(ctsio,
10602 					 /*sks_valid*/ 0,
10603 					 /*retry_count*/ 0);
10604 		mtx_unlock(&ctl_softc->ctl_lock);
10605 		ctl_done((union ctl_io *)ctsio);
10606 		goto bailout;
10607 		break; /* NOTREACHED */
10608 	}
10609 
10610 	goto bailout_unlock;
10611 
10612 queue_rtr:
10613 	ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
10614 	STAILQ_INSERT_TAIL(&ctl_softc->rtr_queue, &ctsio->io_hdr, links);
10615 
10616 bailout_unlock:
10617 	mtx_unlock(&ctl_softc->ctl_lock);
10618 
10619 bailout:
10620 	return (retval);
10621 }
10622 
10623 static int
10624 ctl_scsiio(struct ctl_scsiio *ctsio)
10625 {
10626 	int retval;
10627 	struct ctl_cmd_entry *entry;
10628 
10629 	retval = CTL_RETVAL_COMPLETE;
10630 
10631 	CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
10632 
10633 	entry = &ctl_cmd_table[ctsio->cdb[0]];
10634 
10635 	/*
10636 	 * If this I/O has been aborted, just send it straight to
10637 	 * ctl_done() without executing it.
10638 	 */
10639 	if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
10640 		ctl_done((union ctl_io *)ctsio);
10641 		goto bailout;
10642 	}
10643 
10644 	/*
10645 	 * All the checks should have been handled by ctl_scsiio_precheck().
10646 	 * We should be clear now to just execute the I/O.
10647 	 */
10648 	retval = entry->execute(ctsio);
10649 
10650 bailout:
10651 	return (retval);
10652 }
10653 
10654 /*
10655  * Since we only implement one target right now, a bus reset simply resets
10656  * our single target.
10657  */
10658 static int
10659 ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io)
10660 {
10661 	return(ctl_target_reset(ctl_softc, io, CTL_UA_BUS_RESET));
10662 }
10663 
10664 static int
10665 ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
10666 		 ctl_ua_type ua_type)
10667 {
10668 	struct ctl_lun *lun;
10669 	int retval;
10670 
10671 	if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
10672 		union ctl_ha_msg msg_info;
10673 
10674 		io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
10675 		msg_info.hdr.nexus = io->io_hdr.nexus;
10676 		if (ua_type==CTL_UA_TARG_RESET)
10677 			msg_info.task.task_action = CTL_TASK_TARGET_RESET;
10678 		else
10679 			msg_info.task.task_action = CTL_TASK_BUS_RESET;
10680 		msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
10681 		msg_info.hdr.original_sc = NULL;
10682 		msg_info.hdr.serializing_sc = NULL;
10683 		if (CTL_HA_STATUS_SUCCESS != ctl_ha_msg_send(CTL_HA_CHAN_CTL,
10684 		    (void *)&msg_info, sizeof(msg_info), 0)) {
10685 		}
10686 	}
10687 	retval = 0;
10688 
10689 	STAILQ_FOREACH(lun, &ctl_softc->lun_list, links)
10690 		retval += ctl_lun_reset(lun, io, ua_type);
10691 
10692 	return (retval);
10693 }
10694 
10695 /*
10696  * The LUN should always be set.  The I/O is optional, and is used to
10697  * distinguish between I/Os sent by this initiator, and by other
10698  * initiators.  We set unit attention for initiators other than this one.
10699  * SAM-3 is vague on this point.  It does say that a unit attention should
10700  * be established for other initiators when a LUN is reset (see section
10701  * 5.7.3), but it doesn't specifically say that the unit attention should
10702  * be established for this particular initiator when a LUN is reset.  Here
10703  * is the relevant text, from SAM-3 rev 8:
10704  *
10705  * 5.7.2 When a SCSI initiator port aborts its own tasks
10706  *
10707  * When a SCSI initiator port causes its own task(s) to be aborted, no
10708  * notification that the task(s) have been aborted shall be returned to
10709  * the SCSI initiator port other than the completion response for the
10710  * command or task management function action that caused the task(s) to
10711  * be aborted and notification(s) associated with related effects of the
10712  * action (e.g., a reset unit attention condition).
10713  *
10714  * XXX KDM for now, we're setting unit attention for all initiators.
10715  */
10716 static int
10717 ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
10718 {
10719 	union ctl_io *xio;
10720 #if 0
10721 	uint32_t initindex;
10722 #endif
10723 	int i;
10724 
10725 	/*
10726 	 * Run through the OOA queue and abort each I/O.
10727 	 */
10728 #if 0
10729 	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
10730 #endif
10731 	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
10732 	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
10733 		xio->io_hdr.flags |= CTL_FLAG_ABORT;
10734 	}
10735 
10736 	/*
10737 	 * This version sets unit attention for every
10738 	 */
10739 #if 0
10740 	initindex = ctl_get_initindex(&io->io_hdr.nexus);
10741 	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
10742 		if (initindex == i)
10743 			continue;
10744 		lun->pending_sense[i].ua_pending |= ua_type;
10745 	}
10746 #endif
10747 
10748 	/*
10749 	 * A reset (any kind, really) clears reservations established with
10750 	 * RESERVE/RELEASE.  It does not clear reservations established
10751 	 * with PERSISTENT RESERVE OUT, but we don't support that at the
10752 	 * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
10753 	 * reservations made with the RESERVE/RELEASE commands, because
10754 	 * those commands are obsolete in SPC-3.
10755 	 */
10756 	lun->flags &= ~CTL_LUN_RESERVED;
10757 
10758 	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
10759 		ctl_clear_mask(lun->have_ca, i);
10760 		lun->pending_sense[i].ua_pending |= ua_type;
10761 	}
10762 
10763 	return (0);
10764 }
10765 
10766 static int
10767 ctl_abort_task(union ctl_io *io)
10768 {
10769 	union ctl_io *xio;
10770 	struct ctl_lun *lun;
10771 	struct ctl_softc *ctl_softc;
10772 #if 0
10773 	struct sbuf sb;
10774 	char printbuf[128];
10775 #endif
10776 	int found;
10777 
10778 	ctl_softc = control_softc;
10779 	found = 0;
10780 
10781 	/*
10782 	 * Look up the LUN.
10783 	 */
10784 	if ((io->io_hdr.nexus.targ_lun < CTL_MAX_LUNS)
10785 	 && (ctl_softc->ctl_luns[io->io_hdr.nexus.targ_lun] != NULL))
10786 		lun = ctl_softc->ctl_luns[io->io_hdr.nexus.targ_lun];
10787 	else
10788 		goto bailout;
10789 
10790 #if 0
10791 	printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
10792 	       lun->lun, io->taskio.tag_num, io->taskio.tag_type);
10793 #endif
10794 
10795 	/*
10796 	 * Run through the OOA queue and attempt to find the given I/O.
10797 	 * The target port, initiator ID, tag type and tag number have to
10798 	 * match the values that we got from the initiator.  If we have an
10799 	 * untagged command to abort, simply abort the first untagged command
10800 	 * we come to.  We only allow one untagged command at a time of course.
10801 	 */
10802 #if 0
10803 	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
10804 #endif
10805 	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
10806 	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
10807 #if 0
10808 		sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
10809 
10810 		sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
10811 			    lun->lun, xio->scsiio.tag_num,
10812 			    xio->scsiio.tag_type,
10813 			    (xio->io_hdr.blocked_links.tqe_prev
10814 			    == NULL) ? "" : " BLOCKED",
10815 			    (xio->io_hdr.flags &
10816 			    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
10817 			    (xio->io_hdr.flags &
10818 			    CTL_FLAG_ABORT) ? " ABORT" : "",
10819 			    (xio->io_hdr.flags &
10820 			    CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
10821 		ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
10822 		sbuf_finish(&sb);
10823 		printf("%s\n", sbuf_data(&sb));
10824 #endif
10825 
10826 		if ((xio->io_hdr.nexus.targ_port == io->io_hdr.nexus.targ_port)
10827 		 && (xio->io_hdr.nexus.initid.id ==
10828 		     io->io_hdr.nexus.initid.id)) {
10829 			/*
10830 			 * If the abort says that the task is untagged, the
10831 			 * task in the queue must be untagged.  Otherwise,
10832 			 * we just check to see whether the tag numbers
10833 			 * match.  This is because the QLogic firmware
10834 			 * doesn't pass back the tag type in an abort
10835 			 * request.
10836 			 */
10837 #if 0
10838 			if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
10839 			  && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
10840 			 || (xio->scsiio.tag_num == io->taskio.tag_num)) {
10841 #endif
10842 			/*
10843 			 * XXX KDM we've got problems with FC, because it
10844 			 * doesn't send down a tag type with aborts.  So we
10845 			 * can only really go by the tag number...
10846 			 * This may cause problems with parallel SCSI.
10847 			 * Need to figure that out!!
10848 			 */
10849 			if (xio->scsiio.tag_num == io->taskio.tag_num) {
10850 				xio->io_hdr.flags |= CTL_FLAG_ABORT;
10851 				found = 1;
10852 				if ((io->io_hdr.flags &
10853 				     CTL_FLAG_FROM_OTHER_SC) == 0 &&
10854 				    !(lun->flags & CTL_LUN_PRIMARY_SC)) {
10855 					union ctl_ha_msg msg_info;
10856 
10857 					io->io_hdr.flags |=
10858 					                CTL_FLAG_SENT_2OTHER_SC;
10859 					msg_info.hdr.nexus = io->io_hdr.nexus;
10860 					msg_info.task.task_action =
10861 						CTL_TASK_ABORT_TASK;
10862 					msg_info.task.tag_num =
10863 						io->taskio.tag_num;
10864 					msg_info.task.tag_type =
10865 						io->taskio.tag_type;
10866 					msg_info.hdr.msg_type =
10867 						CTL_MSG_MANAGE_TASKS;
10868 					msg_info.hdr.original_sc = NULL;
10869 					msg_info.hdr.serializing_sc = NULL;
10870 #if 0
10871 					printf("Sent Abort to other side\n");
10872 #endif
10873 					if (CTL_HA_STATUS_SUCCESS !=
10874 					        ctl_ha_msg_send(CTL_HA_CHAN_CTL,
10875 		    				(void *)&msg_info,
10876 						sizeof(msg_info), 0)) {
10877 					}
10878 				}
10879 #if 0
10880 				printf("ctl_abort_task: found I/O to abort\n");
10881 #endif
10882 				break;
10883 			}
10884 		}
10885 	}
10886 
10887 bailout:
10888 
10889 	if (found == 0) {
10890 		/*
10891 		 * This isn't really an error.  It's entirely possible for
10892 		 * the abort and command completion to cross on the wire.
10893 		 * This is more of an informative/diagnostic error.
10894 		 */
10895 #if 0
10896 		printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
10897 		       "%d:%d:%d:%d tag %d type %d\n",
10898 		       io->io_hdr.nexus.initid.id,
10899 		       io->io_hdr.nexus.targ_port,
10900 		       io->io_hdr.nexus.targ_target.id,
10901 		       io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
10902 		       io->taskio.tag_type);
10903 #endif
10904 		return (1);
10905 	} else
10906 		return (0);
10907 }
10908 
10909 /*
10910  * Assumptions:  caller holds ctl_softc->ctl_lock
10911  *
10912  * This routine cannot block!  It must be callable from an interrupt
10913  * handler as well as from the work thread.
10914  */
10915 static void
10916 ctl_run_task_queue(struct ctl_softc *ctl_softc)
10917 {
10918 	union ctl_io *io, *next_io;
10919 
10920 	CTL_DEBUG_PRINT(("ctl_run_task_queue\n"));
10921 
10922 	for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->task_queue);
10923 	     io != NULL; io = next_io) {
10924 		int retval;
10925 		const char *task_desc;
10926 
10927 		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
10928 
10929 		retval = 0;
10930 
10931 		switch (io->io_hdr.io_type) {
10932 		case CTL_IO_TASK: {
10933 			task_desc = ctl_scsi_task_string(&io->taskio);
10934 			if (task_desc != NULL) {
10935 #ifdef NEEDTOPORT
10936 				csevent_log(CSC_CTL | CSC_SHELF_SW |
10937 					    CTL_TASK_REPORT,
10938 					    csevent_LogType_Trace,
10939 					    csevent_Severity_Information,
10940 					    csevent_AlertLevel_Green,
10941 					    csevent_FRU_Firmware,
10942 					    csevent_FRU_Unknown,
10943 					    "CTL: received task: %s",task_desc);
10944 #endif
10945 			} else {
10946 #ifdef NEEDTOPORT
10947 				csevent_log(CSC_CTL | CSC_SHELF_SW |
10948 					    CTL_TASK_REPORT,
10949 					    csevent_LogType_Trace,
10950 					    csevent_Severity_Information,
10951 					    csevent_AlertLevel_Green,
10952 					    csevent_FRU_Firmware,
10953 					    csevent_FRU_Unknown,
10954 					    "CTL: received unknown task "
10955 					    "type: %d (%#x)",
10956 					    io->taskio.task_action,
10957 					    io->taskio.task_action);
10958 #endif
10959 			}
10960 			switch (io->taskio.task_action) {
10961 			case CTL_TASK_ABORT_TASK:
10962 				retval = ctl_abort_task(io);
10963 				break;
10964 			case CTL_TASK_ABORT_TASK_SET:
10965 				break;
10966 			case CTL_TASK_CLEAR_ACA:
10967 				break;
10968 			case CTL_TASK_CLEAR_TASK_SET:
10969 				break;
10970 			case CTL_TASK_LUN_RESET: {
10971 				struct ctl_lun *lun;
10972 				uint32_t targ_lun;
10973 				int retval;
10974 
10975 				targ_lun = io->io_hdr.nexus.targ_lun;
10976 
10977 				if ((targ_lun < CTL_MAX_LUNS)
10978 				 && (ctl_softc->ctl_luns[targ_lun] != NULL))
10979 					lun = ctl_softc->ctl_luns[targ_lun];
10980 				else {
10981 					retval = 1;
10982 					break;
10983 				}
10984 
10985 				if (!(io->io_hdr.flags &
10986 				    CTL_FLAG_FROM_OTHER_SC)) {
10987 					union ctl_ha_msg msg_info;
10988 
10989 					io->io_hdr.flags |=
10990 						CTL_FLAG_SENT_2OTHER_SC;
10991 					msg_info.hdr.msg_type =
10992 						CTL_MSG_MANAGE_TASKS;
10993 					msg_info.hdr.nexus = io->io_hdr.nexus;
10994 					msg_info.task.task_action =
10995 						CTL_TASK_LUN_RESET;
10996 					msg_info.hdr.original_sc = NULL;
10997 					msg_info.hdr.serializing_sc = NULL;
10998 					if (CTL_HA_STATUS_SUCCESS !=
10999 					    ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11000 					    (void *)&msg_info,
11001 					    sizeof(msg_info), 0)) {
11002 					}
11003 				}
11004 
11005 				retval = ctl_lun_reset(lun, io,
11006 						       CTL_UA_LUN_RESET);
11007 				break;
11008 			}
11009 			case CTL_TASK_TARGET_RESET:
11010 				retval = ctl_target_reset(ctl_softc, io,
11011 							  CTL_UA_TARG_RESET);
11012 				break;
11013 			case CTL_TASK_BUS_RESET:
11014 				retval = ctl_bus_reset(ctl_softc, io);
11015 				break;
11016 			case CTL_TASK_PORT_LOGIN:
11017 				break;
11018 			case CTL_TASK_PORT_LOGOUT:
11019 				break;
11020 			default:
11021 				printf("ctl_run_task_queue: got unknown task "
11022 				       "management event %d\n",
11023 				       io->taskio.task_action);
11024 				break;
11025 			}
11026 			if (retval == 0)
11027 				io->io_hdr.status = CTL_SUCCESS;
11028 			else
11029 				io->io_hdr.status = CTL_ERROR;
11030 
11031 			STAILQ_REMOVE(&ctl_softc->task_queue, &io->io_hdr,
11032 				      ctl_io_hdr, links);
11033 			/*
11034 			 * This will queue this I/O to the done queue, but the
11035 			 * work thread won't be able to process it until we
11036 			 * return and the lock is released.
11037 			 */
11038 			ctl_done_lock(io, /*have_lock*/ 1);
11039 			break;
11040 		}
11041 		default: {
11042 
11043 			printf("%s: invalid I/O type %d msg %d cdb %x"
11044 			       " iptl: %ju:%d:%ju:%d tag 0x%04x\n",
11045 			       __func__, io->io_hdr.io_type,
11046 			       io->io_hdr.msg_type, io->scsiio.cdb[0],
11047 			       (uintmax_t)io->io_hdr.nexus.initid.id,
11048 			       io->io_hdr.nexus.targ_port,
11049 			       (uintmax_t)io->io_hdr.nexus.targ_target.id,
11050 			       io->io_hdr.nexus.targ_lun,
11051 			       (io->io_hdr.io_type == CTL_IO_TASK) ?
11052 			       io->taskio.tag_num : io->scsiio.tag_num);
11053 			STAILQ_REMOVE(&ctl_softc->task_queue, &io->io_hdr,
11054 				      ctl_io_hdr, links);
11055 			ctl_free_io_internal(io, 1);
11056 			break;
11057 		}
11058 		}
11059 	}
11060 
11061 	ctl_softc->flags &= ~CTL_FLAG_TASK_PENDING;
11062 }
11063 
11064 /*
11065  * For HA operation.  Handle commands that come in from the other
11066  * controller.
11067  */
11068 static void
11069 ctl_handle_isc(union ctl_io *io)
11070 {
11071 	int free_io;
11072 	struct ctl_lun *lun;
11073 	struct ctl_softc *ctl_softc;
11074 
11075 	ctl_softc = control_softc;
11076 
11077 	lun = ctl_softc->ctl_luns[io->io_hdr.nexus.targ_lun];
11078 
11079 	switch (io->io_hdr.msg_type) {
11080 	case CTL_MSG_SERIALIZE:
11081 		free_io = ctl_serialize_other_sc_cmd(&io->scsiio,
11082 						     /*have_lock*/ 0);
11083 		break;
11084 	case CTL_MSG_R2R: {
11085 		uint8_t opcode;
11086 		struct ctl_cmd_entry *entry;
11087 
11088 		/*
11089 		 * This is only used in SER_ONLY mode.
11090 		 */
11091 		free_io = 0;
11092 		opcode = io->scsiio.cdb[0];
11093 		entry = &ctl_cmd_table[opcode];
11094 		mtx_lock(&ctl_softc->ctl_lock);
11095 		if (ctl_scsiio_lun_check(ctl_softc, lun,
11096 		    entry, (struct ctl_scsiio *)io) != 0) {
11097 			ctl_done_lock(io, /*have_lock*/ 1);
11098 			mtx_unlock(&ctl_softc->ctl_lock);
11099 			break;
11100 		}
11101 		io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11102 		STAILQ_INSERT_TAIL(&ctl_softc->rtr_queue,
11103 				   &io->io_hdr, links);
11104 		mtx_unlock(&ctl_softc->ctl_lock);
11105 		break;
11106 	}
11107 	case CTL_MSG_FINISH_IO:
11108 		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
11109 			free_io = 0;
11110 			ctl_done_lock(io, /*have_lock*/ 0);
11111 		} else {
11112 			free_io = 1;
11113 			mtx_lock(&ctl_softc->ctl_lock);
11114 			TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
11115 				     ooa_links);
11116 			STAILQ_REMOVE(&ctl_softc->task_queue,
11117 				      &io->io_hdr, ctl_io_hdr, links);
11118 			ctl_check_blocked(lun);
11119 			mtx_unlock(&ctl_softc->ctl_lock);
11120 		}
11121 		break;
11122 	case CTL_MSG_PERS_ACTION:
11123 		ctl_hndl_per_res_out_on_other_sc(
11124 			(union ctl_ha_msg *)&io->presio.pr_msg);
11125 		free_io = 1;
11126 		break;
11127 	case CTL_MSG_BAD_JUJU:
11128 		free_io = 0;
11129 		ctl_done_lock(io, /*have_lock*/ 0);
11130 		break;
11131 	case CTL_MSG_DATAMOVE:
11132 		/* Only used in XFER mode */
11133 		free_io = 0;
11134 		ctl_datamove_remote(io);
11135 		break;
11136 	case CTL_MSG_DATAMOVE_DONE:
11137 		/* Only used in XFER mode */
11138 		free_io = 0;
11139 		io->scsiio.be_move_done(io);
11140 		break;
11141 	default:
11142 		free_io = 1;
11143 		printf("%s: Invalid message type %d\n",
11144 		       __func__, io->io_hdr.msg_type);
11145 		break;
11146 	}
11147 	if (free_io)
11148 		ctl_free_io_internal(io, 0);
11149 
11150 }
11151 
11152 
11153 /*
11154  * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
11155  * there is no match.
11156  */
11157 static ctl_lun_error_pattern
11158 ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
11159 {
11160 	struct ctl_cmd_entry *entry;
11161 	ctl_lun_error_pattern filtered_pattern, pattern;
11162 	uint8_t opcode;
11163 
11164 	pattern = desc->error_pattern;
11165 
11166 	/*
11167 	 * XXX KDM we need more data passed into this function to match a
11168 	 * custom pattern, and we actually need to implement custom pattern
11169 	 * matching.
11170 	 */
11171 	if (pattern & CTL_LUN_PAT_CMD)
11172 		return (CTL_LUN_PAT_CMD);
11173 
11174 	if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
11175 		return (CTL_LUN_PAT_ANY);
11176 
11177 	opcode = ctsio->cdb[0];
11178 	entry = &ctl_cmd_table[opcode];
11179 
11180 	filtered_pattern = entry->pattern & pattern;
11181 
11182 	/*
11183 	 * If the user requested specific flags in the pattern (e.g.
11184 	 * CTL_LUN_PAT_RANGE), make sure the command supports all of those
11185 	 * flags.
11186 	 *
11187 	 * If the user did not specify any flags, it doesn't matter whether
11188 	 * or not the command supports the flags.
11189 	 */
11190 	if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
11191 	     (pattern & ~CTL_LUN_PAT_MASK))
11192 		return (CTL_LUN_PAT_NONE);
11193 
11194 	/*
11195 	 * If the user asked for a range check, see if the requested LBA
11196 	 * range overlaps with this command's LBA range.
11197 	 */
11198 	if (filtered_pattern & CTL_LUN_PAT_RANGE) {
11199 		uint64_t lba1;
11200 		uint32_t len1;
11201 		ctl_action action;
11202 		int retval;
11203 
11204 		retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
11205 		if (retval != 0)
11206 			return (CTL_LUN_PAT_NONE);
11207 
11208 		action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
11209 					      desc->lba_range.len);
11210 		/*
11211 		 * A "pass" means that the LBA ranges don't overlap, so
11212 		 * this doesn't match the user's range criteria.
11213 		 */
11214 		if (action == CTL_ACTION_PASS)
11215 			return (CTL_LUN_PAT_NONE);
11216 	}
11217 
11218 	return (filtered_pattern);
11219 }
11220 
11221 /*
11222  * Called with the CTL lock held.
11223  */
11224 static void
11225 ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
11226 {
11227 	struct ctl_error_desc *desc, *desc2;
11228 
11229 	STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
11230 		ctl_lun_error_pattern pattern;
11231 		/*
11232 		 * Check to see whether this particular command matches
11233 		 * the pattern in the descriptor.
11234 		 */
11235 		pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
11236 		if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
11237 			continue;
11238 
11239 		switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
11240 		case CTL_LUN_INJ_ABORTED:
11241 			ctl_set_aborted(&io->scsiio);
11242 			break;
11243 		case CTL_LUN_INJ_MEDIUM_ERR:
11244 			ctl_set_medium_error(&io->scsiio);
11245 			break;
11246 		case CTL_LUN_INJ_UA:
11247 			/* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
11248 			 * OCCURRED */
11249 			ctl_set_ua(&io->scsiio, 0x29, 0x00);
11250 			break;
11251 		case CTL_LUN_INJ_CUSTOM:
11252 			/*
11253 			 * We're assuming the user knows what he is doing.
11254 			 * Just copy the sense information without doing
11255 			 * checks.
11256 			 */
11257 			bcopy(&desc->custom_sense, &io->scsiio.sense_data,
11258 			      ctl_min(sizeof(desc->custom_sense),
11259 				      sizeof(io->scsiio.sense_data)));
11260 			io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
11261 			io->scsiio.sense_len = SSD_FULL_SIZE;
11262 			io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
11263 			break;
11264 		case CTL_LUN_INJ_NONE:
11265 		default:
11266 			/*
11267 			 * If this is an error injection type we don't know
11268 			 * about, clear the continuous flag (if it is set)
11269 			 * so it will get deleted below.
11270 			 */
11271 			desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
11272 			break;
11273 		}
11274 		/*
11275 		 * By default, each error injection action is a one-shot
11276 		 */
11277 		if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
11278 			continue;
11279 
11280 		STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
11281 
11282 		free(desc, M_CTL);
11283 	}
11284 }
11285 
11286 #ifdef CTL_IO_DELAY
11287 static void
11288 ctl_datamove_timer_wakeup(void *arg)
11289 {
11290 	union ctl_io *io;
11291 
11292 	io = (union ctl_io *)arg;
11293 
11294 	ctl_datamove(io);
11295 }
11296 #endif /* CTL_IO_DELAY */
11297 
11298 /*
11299  * Assumption:  caller does NOT hold ctl_lock
11300  */
11301 void
11302 ctl_datamove(union ctl_io *io)
11303 {
11304 	void (*fe_datamove)(union ctl_io *io);
11305 
11306 	CTL_DEBUG_PRINT(("ctl_datamove\n"));
11307 
11308 #ifdef CTL_TIME_IO
11309 	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
11310 		char str[256];
11311 		char path_str[64];
11312 		struct sbuf sb;
11313 
11314 		ctl_scsi_path_string(io, path_str, sizeof(path_str));
11315 		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
11316 
11317 		sbuf_cat(&sb, path_str);
11318 		switch (io->io_hdr.io_type) {
11319 		case CTL_IO_SCSI:
11320 			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
11321 			sbuf_printf(&sb, "\n");
11322 			sbuf_cat(&sb, path_str);
11323 			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
11324 				    io->scsiio.tag_num, io->scsiio.tag_type);
11325 			break;
11326 		case CTL_IO_TASK:
11327 			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
11328 				    "Tag Type: %d\n", io->taskio.task_action,
11329 				    io->taskio.tag_num, io->taskio.tag_type);
11330 			break;
11331 		default:
11332 			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
11333 			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
11334 			break;
11335 		}
11336 		sbuf_cat(&sb, path_str);
11337 		sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
11338 			    (intmax_t)time_uptime - io->io_hdr.start_time);
11339 		sbuf_finish(&sb);
11340 		printf("%s", sbuf_data(&sb));
11341 	}
11342 #endif /* CTL_TIME_IO */
11343 
11344 	mtx_lock(&control_softc->ctl_lock);
11345 #ifdef CTL_IO_DELAY
11346 	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
11347 		struct ctl_lun *lun;
11348 
11349 		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
11350 
11351 		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
11352 	} else {
11353 		struct ctl_lun *lun;
11354 
11355 		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
11356 		if ((lun != NULL)
11357 		 && (lun->delay_info.datamove_delay > 0)) {
11358 			struct callout *callout;
11359 
11360 			callout = (struct callout *)&io->io_hdr.timer_bytes;
11361 			callout_init(callout, /*mpsafe*/ 1);
11362 			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
11363 			callout_reset(callout,
11364 				      lun->delay_info.datamove_delay * hz,
11365 				      ctl_datamove_timer_wakeup, io);
11366 			if (lun->delay_info.datamove_type ==
11367 			    CTL_DELAY_TYPE_ONESHOT)
11368 				lun->delay_info.datamove_delay = 0;
11369 			mtx_unlock(&control_softc->ctl_lock);
11370 			return;
11371 		}
11372 	}
11373 #endif
11374 	/*
11375 	 * If we have any pending task management commands, process them
11376 	 * first.  This is necessary to eliminate a race condition with the
11377 	 * FETD:
11378 	 *
11379 	 * - FETD submits a task management command, like an abort.
11380 	 * - Back end calls fe_datamove() to move the data for the aborted
11381 	 *   command.  The FETD can't really accept it, but if it did, it
11382 	 *   would end up transmitting data for a command that the initiator
11383 	 *   told us to abort.
11384 	 *
11385 	 * We close the race by processing all pending task management
11386 	 * commands here (we can't block!), and then check this I/O to see
11387 	 * if it has been aborted.  If so, return it to the back end with
11388 	 * bad status, so the back end can say return an error to the back end
11389 	 * and then when the back end returns an error, we can return the
11390 	 * aborted command to the FETD, so it can clean up its resources.
11391 	 */
11392 	if (control_softc->flags & CTL_FLAG_TASK_PENDING)
11393 		ctl_run_task_queue(control_softc);
11394 
11395 	/*
11396 	 * This command has been aborted.  Set the port status, so we fail
11397 	 * the data move.
11398 	 */
11399 	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
11400 		printf("ctl_datamove: tag 0x%04x on (%ju:%d:%ju:%d) aborted\n",
11401 		       io->scsiio.tag_num,(uintmax_t)io->io_hdr.nexus.initid.id,
11402 		       io->io_hdr.nexus.targ_port,
11403 		       (uintmax_t)io->io_hdr.nexus.targ_target.id,
11404 		       io->io_hdr.nexus.targ_lun);
11405 		io->io_hdr.status = CTL_CMD_ABORTED;
11406 		io->io_hdr.port_status = 31337;
11407 		mtx_unlock(&control_softc->ctl_lock);
11408 		/*
11409 		 * Note that the backend, in this case, will get the
11410 		 * callback in its context.  In other cases it may get
11411 		 * called in the frontend's interrupt thread context.
11412 		 */
11413 		io->scsiio.be_move_done(io);
11414 		return;
11415 	}
11416 
11417 	/*
11418 	 * If we're in XFER mode and this I/O is from the other shelf
11419 	 * controller, we need to send the DMA to the other side to
11420 	 * actually transfer the data to/from the host.  In serialize only
11421 	 * mode the transfer happens below CTL and ctl_datamove() is only
11422 	 * called on the machine that originally received the I/O.
11423 	 */
11424 	if ((control_softc->ha_mode == CTL_HA_MODE_XFER)
11425 	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11426 		union ctl_ha_msg msg;
11427 		uint32_t sg_entries_sent;
11428 		int do_sg_copy;
11429 		int i;
11430 
11431 		memset(&msg, 0, sizeof(msg));
11432 		msg.hdr.msg_type = CTL_MSG_DATAMOVE;
11433 		msg.hdr.original_sc = io->io_hdr.original_sc;
11434 		msg.hdr.serializing_sc = io;
11435 		msg.hdr.nexus = io->io_hdr.nexus;
11436 		msg.dt.flags = io->io_hdr.flags;
11437 		/*
11438 		 * We convert everything into a S/G list here.  We can't
11439 		 * pass by reference, only by value between controllers.
11440 		 * So we can't pass a pointer to the S/G list, only as many
11441 		 * S/G entries as we can fit in here.  If it's possible for
11442 		 * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
11443 		 * then we need to break this up into multiple transfers.
11444 		 */
11445 		if (io->scsiio.kern_sg_entries == 0) {
11446 			msg.dt.kern_sg_entries = 1;
11447 			/*
11448 			 * If this is in cached memory, flush the cache
11449 			 * before we send the DMA request to the other
11450 			 * controller.  We want to do this in either the
11451 			 * read or the write case.  The read case is
11452 			 * straightforward.  In the write case, we want to
11453 			 * make sure nothing is in the local cache that
11454 			 * could overwrite the DMAed data.
11455 			 */
11456 			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
11457 				/*
11458 				 * XXX KDM use bus_dmamap_sync() here.
11459 				 */
11460 			}
11461 
11462 			/*
11463 			 * Convert to a physical address if this is a
11464 			 * virtual address.
11465 			 */
11466 			if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
11467 				msg.dt.sg_list[0].addr =
11468 					io->scsiio.kern_data_ptr;
11469 			} else {
11470 				/*
11471 				 * XXX KDM use busdma here!
11472 				 */
11473 #if 0
11474 				msg.dt.sg_list[0].addr = (void *)
11475 					vtophys(io->scsiio.kern_data_ptr);
11476 #endif
11477 			}
11478 
11479 			msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
11480 			do_sg_copy = 0;
11481 		} else {
11482 			struct ctl_sg_entry *sgl;
11483 
11484 			do_sg_copy = 1;
11485 			msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
11486 			sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
11487 			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
11488 				/*
11489 				 * XXX KDM use bus_dmamap_sync() here.
11490 				 */
11491 			}
11492 		}
11493 
11494 		msg.dt.kern_data_len = io->scsiio.kern_data_len;
11495 		msg.dt.kern_total_len = io->scsiio.kern_total_len;
11496 		msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
11497 		msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
11498 		msg.dt.sg_sequence = 0;
11499 
11500 		/*
11501 		 * Loop until we've sent all of the S/G entries.  On the
11502 		 * other end, we'll recompose these S/G entries into one
11503 		 * contiguous list before passing it to the
11504 		 */
11505 		for (sg_entries_sent = 0; sg_entries_sent <
11506 		     msg.dt.kern_sg_entries; msg.dt.sg_sequence++) {
11507 			msg.dt.cur_sg_entries = ctl_min((sizeof(msg.dt.sg_list)/
11508 				sizeof(msg.dt.sg_list[0])),
11509 				msg.dt.kern_sg_entries - sg_entries_sent);
11510 
11511 			if (do_sg_copy != 0) {
11512 				struct ctl_sg_entry *sgl;
11513 				int j;
11514 
11515 				sgl = (struct ctl_sg_entry *)
11516 					io->scsiio.kern_data_ptr;
11517 				/*
11518 				 * If this is in cached memory, flush the cache
11519 				 * before we send the DMA request to the other
11520 				 * controller.  We want to do this in either
11521 				 * the * read or the write case.  The read
11522 				 * case is straightforward.  In the write
11523 				 * case, we want to make sure nothing is
11524 				 * in the local cache that could overwrite
11525 				 * the DMAed data.
11526 				 */
11527 
11528 				for (i = sg_entries_sent, j = 0;
11529 				     i < msg.dt.cur_sg_entries; i++, j++) {
11530 					if ((io->io_hdr.flags &
11531 					     CTL_FLAG_NO_DATASYNC) == 0) {
11532 						/*
11533 						 * XXX KDM use bus_dmamap_sync()
11534 						 */
11535 					}
11536 					if ((io->io_hdr.flags &
11537 					     CTL_FLAG_BUS_ADDR) == 0) {
11538 						/*
11539 						 * XXX KDM use busdma.
11540 						 */
11541 #if 0
11542 						msg.dt.sg_list[j].addr =(void *)
11543 						       vtophys(sgl[i].addr);
11544 #endif
11545 					} else {
11546 						msg.dt.sg_list[j].addr =
11547 							sgl[i].addr;
11548 					}
11549 					msg.dt.sg_list[j].len = sgl[i].len;
11550 				}
11551 			}
11552 
11553 			sg_entries_sent += msg.dt.cur_sg_entries;
11554 			if (sg_entries_sent >= msg.dt.kern_sg_entries)
11555 				msg.dt.sg_last = 1;
11556 			else
11557 				msg.dt.sg_last = 0;
11558 
11559 			/*
11560 			 * XXX KDM drop and reacquire the lock here?
11561 			 */
11562 			if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
11563 			    sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
11564 				/*
11565 				 * XXX do something here.
11566 				 */
11567 			}
11568 
11569 			msg.dt.sent_sg_entries = sg_entries_sent;
11570 		}
11571 		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11572 		if (io->io_hdr.flags & CTL_FLAG_FAILOVER)
11573 			ctl_failover_io(io, /*have_lock*/ 1);
11574 
11575 	} else {
11576 
11577 		/*
11578 		 * Lookup the fe_datamove() function for this particular
11579 		 * front end.
11580 		 */
11581 		fe_datamove =
11582 		    control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
11583 		mtx_unlock(&control_softc->ctl_lock);
11584 
11585 		fe_datamove(io);
11586 	}
11587 }
11588 
11589 static void
11590 ctl_send_datamove_done(union ctl_io *io, int have_lock)
11591 {
11592 	union ctl_ha_msg msg;
11593 	int isc_status;
11594 
11595 	memset(&msg, 0, sizeof(msg));
11596 
11597 	msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
11598 	msg.hdr.original_sc = io;
11599 	msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
11600 	msg.hdr.nexus = io->io_hdr.nexus;
11601 	msg.hdr.status = io->io_hdr.status;
11602 	msg.scsi.tag_num = io->scsiio.tag_num;
11603 	msg.scsi.tag_type = io->scsiio.tag_type;
11604 	msg.scsi.scsi_status = io->scsiio.scsi_status;
11605 	memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
11606 	       sizeof(io->scsiio.sense_data));
11607 	msg.scsi.sense_len = io->scsiio.sense_len;
11608 	msg.scsi.sense_residual = io->scsiio.sense_residual;
11609 	msg.scsi.fetd_status = io->io_hdr.port_status;
11610 	msg.scsi.residual = io->scsiio.residual;
11611 	io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11612 
11613 	if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
11614 		ctl_failover_io(io, /*have_lock*/ have_lock);
11615 		return;
11616 	}
11617 
11618 	isc_status = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0);
11619 	if (isc_status > CTL_HA_STATUS_SUCCESS) {
11620 		/* XXX do something if this fails */
11621 	}
11622 
11623 }
11624 
11625 /*
11626  * The DMA to the remote side is done, now we need to tell the other side
11627  * we're done so it can continue with its data movement.
11628  */
11629 static void
11630 ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
11631 {
11632 	union ctl_io *io;
11633 
11634 	io = rq->context;
11635 
11636 	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
11637 		printf("%s: ISC DMA write failed with error %d", __func__,
11638 		       rq->ret);
11639 		ctl_set_internal_failure(&io->scsiio,
11640 					 /*sks_valid*/ 1,
11641 					 /*retry_count*/ rq->ret);
11642 	}
11643 
11644 	ctl_dt_req_free(rq);
11645 
11646 	/*
11647 	 * In this case, we had to malloc the memory locally.  Free it.
11648 	 */
11649 	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
11650 		int i;
11651 		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
11652 			free(io->io_hdr.local_sglist[i].addr, M_CTL);
11653 	}
11654 	/*
11655 	 * The data is in local and remote memory, so now we need to send
11656 	 * status (good or back) back to the other side.
11657 	 */
11658 	ctl_send_datamove_done(io, /*have_lock*/ 0);
11659 }
11660 
11661 /*
11662  * We've moved the data from the host/controller into local memory.  Now we
11663  * need to push it over to the remote controller's memory.
11664  */
11665 static int
11666 ctl_datamove_remote_dm_write_cb(union ctl_io *io)
11667 {
11668 	int retval;
11669 
11670 	retval = 0;
11671 
11672 	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
11673 					  ctl_datamove_remote_write_cb);
11674 
11675 	return (retval);
11676 }
11677 
11678 static void
11679 ctl_datamove_remote_write(union ctl_io *io)
11680 {
11681 	int retval;
11682 	void (*fe_datamove)(union ctl_io *io);
11683 
11684 	/*
11685 	 * - Get the data from the host/HBA into local memory.
11686 	 * - DMA memory from the local controller to the remote controller.
11687 	 * - Send status back to the remote controller.
11688 	 */
11689 
11690 	retval = ctl_datamove_remote_sgl_setup(io);
11691 	if (retval != 0)
11692 		return;
11693 
11694 	/* Switch the pointer over so the FETD knows what to do */
11695 	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
11696 
11697 	/*
11698 	 * Use a custom move done callback, since we need to send completion
11699 	 * back to the other controller, not to the backend on this side.
11700 	 */
11701 	io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
11702 
11703 	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
11704 
11705 	fe_datamove(io);
11706 
11707 	return;
11708 
11709 }
11710 
11711 static int
11712 ctl_datamove_remote_dm_read_cb(union ctl_io *io)
11713 {
11714 #if 0
11715 	char str[256];
11716 	char path_str[64];
11717 	struct sbuf sb;
11718 #endif
11719 
11720 	/*
11721 	 * In this case, we had to malloc the memory locally.  Free it.
11722 	 */
11723 	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
11724 		int i;
11725 		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
11726 			free(io->io_hdr.local_sglist[i].addr, M_CTL);
11727 	}
11728 
11729 #if 0
11730 	scsi_path_string(io, path_str, sizeof(path_str));
11731 	sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
11732 	sbuf_cat(&sb, path_str);
11733 	scsi_command_string(&io->scsiio, NULL, &sb);
11734 	sbuf_printf(&sb, "\n");
11735 	sbuf_cat(&sb, path_str);
11736 	sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
11737 		    io->scsiio.tag_num, io->scsiio.tag_type);
11738 	sbuf_cat(&sb, path_str);
11739 	sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
11740 		    io->io_hdr.flags, io->io_hdr.status);
11741 	sbuf_finish(&sb);
11742 	printk("%s", sbuf_data(&sb));
11743 #endif
11744 
11745 
11746 	/*
11747 	 * The read is done, now we need to send status (good or bad) back
11748 	 * to the other side.
11749 	 */
11750 	ctl_send_datamove_done(io, /*have_lock*/ 0);
11751 
11752 	return (0);
11753 }
11754 
11755 static void
11756 ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
11757 {
11758 	union ctl_io *io;
11759 	void (*fe_datamove)(union ctl_io *io);
11760 
11761 	io = rq->context;
11762 
11763 	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
11764 		printf("%s: ISC DMA read failed with error %d", __func__,
11765 		       rq->ret);
11766 		ctl_set_internal_failure(&io->scsiio,
11767 					 /*sks_valid*/ 1,
11768 					 /*retry_count*/ rq->ret);
11769 	}
11770 
11771 	ctl_dt_req_free(rq);
11772 
11773 	/* Switch the pointer over so the FETD knows what to do */
11774 	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
11775 
11776 	/*
11777 	 * Use a custom move done callback, since we need to send completion
11778 	 * back to the other controller, not to the backend on this side.
11779 	 */
11780 	io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
11781 
11782 	/* XXX KDM add checks like the ones in ctl_datamove? */
11783 
11784 	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
11785 
11786 	fe_datamove(io);
11787 }
11788 
11789 static int
11790 ctl_datamove_remote_sgl_setup(union ctl_io *io)
11791 {
11792 	struct ctl_sg_entry *local_sglist, *remote_sglist;
11793 	struct ctl_sg_entry *local_dma_sglist, *remote_dma_sglist;
11794 	struct ctl_softc *softc;
11795 	int retval;
11796 	int i;
11797 
11798 	retval = 0;
11799 	softc = control_softc;
11800 
11801 	local_sglist = io->io_hdr.local_sglist;
11802 	local_dma_sglist = io->io_hdr.local_dma_sglist;
11803 	remote_sglist = io->io_hdr.remote_sglist;
11804 	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
11805 
11806 	if (io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) {
11807 		for (i = 0; i < io->scsiio.kern_sg_entries; i++) {
11808 			local_sglist[i].len = remote_sglist[i].len;
11809 
11810 			/*
11811 			 * XXX Detect the situation where the RS-level I/O
11812 			 * redirector on the other side has already read the
11813 			 * data off of the AOR RS on this side, and
11814 			 * transferred it to remote (mirror) memory on the
11815 			 * other side.  Since we already have the data in
11816 			 * memory here, we just need to use it.
11817 			 *
11818 			 * XXX KDM this can probably be removed once we
11819 			 * get the cache device code in and take the
11820 			 * current AOR implementation out.
11821 			 */
11822 #ifdef NEEDTOPORT
11823 			if ((remote_sglist[i].addr >=
11824 			     (void *)vtophys(softc->mirr->addr))
11825 			 && (remote_sglist[i].addr <
11826 			     ((void *)vtophys(softc->mirr->addr) +
11827 			     CacheMirrorOffset))) {
11828 				local_sglist[i].addr = remote_sglist[i].addr -
11829 					CacheMirrorOffset;
11830 				if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
11831 				     CTL_FLAG_DATA_IN)
11832 					io->io_hdr.flags |= CTL_FLAG_REDIR_DONE;
11833 			} else {
11834 				local_sglist[i].addr = remote_sglist[i].addr +
11835 					CacheMirrorOffset;
11836 			}
11837 #endif
11838 #if 0
11839 			printf("%s: local %p, remote %p, len %d\n",
11840 			       __func__, local_sglist[i].addr,
11841 			       remote_sglist[i].addr, local_sglist[i].len);
11842 #endif
11843 		}
11844 	} else {
11845 		uint32_t len_to_go;
11846 
11847 		/*
11848 		 * In this case, we don't have automatically allocated
11849 		 * memory for this I/O on this controller.  This typically
11850 		 * happens with internal CTL I/O -- e.g. inquiry, mode
11851 		 * sense, etc.  Anything coming from RAIDCore will have
11852 		 * a mirror area available.
11853 		 */
11854 		len_to_go = io->scsiio.kern_data_len;
11855 
11856 		/*
11857 		 * Clear the no datasync flag, we have to use malloced
11858 		 * buffers.
11859 		 */
11860 		io->io_hdr.flags &= ~CTL_FLAG_NO_DATASYNC;
11861 
11862 		/*
11863 		 * The difficult thing here is that the size of the various
11864 		 * S/G segments may be different than the size from the
11865 		 * remote controller.  That'll make it harder when DMAing
11866 		 * the data back to the other side.
11867 		 */
11868 		for (i = 0; (i < sizeof(io->io_hdr.remote_sglist) /
11869 		     sizeof(io->io_hdr.remote_sglist[0])) &&
11870 		     (len_to_go > 0); i++) {
11871 			local_sglist[i].len = ctl_min(len_to_go, 131072);
11872 			CTL_SIZE_8B(local_dma_sglist[i].len,
11873 				    local_sglist[i].len);
11874 			local_sglist[i].addr =
11875 				malloc(local_dma_sglist[i].len, M_CTL,M_WAITOK);
11876 
11877 			local_dma_sglist[i].addr = local_sglist[i].addr;
11878 
11879 			if (local_sglist[i].addr == NULL) {
11880 				int j;
11881 
11882 				printf("malloc failed for %zd bytes!",
11883 				       local_dma_sglist[i].len);
11884 				for (j = 0; j < i; j++) {
11885 					free(local_sglist[j].addr, M_CTL);
11886 				}
11887 				ctl_set_internal_failure(&io->scsiio,
11888 							 /*sks_valid*/ 1,
11889 							 /*retry_count*/ 4857);
11890 				retval = 1;
11891 				goto bailout_error;
11892 
11893 			}
11894 			/* XXX KDM do we need a sync here? */
11895 
11896 			len_to_go -= local_sglist[i].len;
11897 		}
11898 		/*
11899 		 * Reset the number of S/G entries accordingly.  The
11900 		 * original number of S/G entries is available in
11901 		 * rem_sg_entries.
11902 		 */
11903 		io->scsiio.kern_sg_entries = i;
11904 
11905 #if 0
11906 		printf("%s: kern_sg_entries = %d\n", __func__,
11907 		       io->scsiio.kern_sg_entries);
11908 		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
11909 			printf("%s: sg[%d] = %p, %d (DMA: %d)\n", __func__, i,
11910 			       local_sglist[i].addr, local_sglist[i].len,
11911 			       local_dma_sglist[i].len);
11912 #endif
11913 	}
11914 
11915 
11916 	return (retval);
11917 
11918 bailout_error:
11919 
11920 	ctl_send_datamove_done(io, /*have_lock*/ 0);
11921 
11922 	return (retval);
11923 }
11924 
11925 static int
11926 ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
11927 			 ctl_ha_dt_cb callback)
11928 {
11929 	struct ctl_ha_dt_req *rq;
11930 	struct ctl_sg_entry *remote_sglist, *local_sglist;
11931 	struct ctl_sg_entry *remote_dma_sglist, *local_dma_sglist;
11932 	uint32_t local_used, remote_used, total_used;
11933 	int retval;
11934 	int i, j;
11935 
11936 	retval = 0;
11937 
11938 	rq = ctl_dt_req_alloc();
11939 
11940 	/*
11941 	 * If we failed to allocate the request, and if the DMA didn't fail
11942 	 * anyway, set busy status.  This is just a resource allocation
11943 	 * failure.
11944 	 */
11945 	if ((rq == NULL)
11946 	 && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE))
11947 		ctl_set_busy(&io->scsiio);
11948 
11949 	if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) {
11950 
11951 		if (rq != NULL)
11952 			ctl_dt_req_free(rq);
11953 
11954 		/*
11955 		 * The data move failed.  We need to return status back
11956 		 * to the other controller.  No point in trying to DMA
11957 		 * data to the remote controller.
11958 		 */
11959 
11960 		ctl_send_datamove_done(io, /*have_lock*/ 0);
11961 
11962 		retval = 1;
11963 
11964 		goto bailout;
11965 	}
11966 
11967 	local_sglist = io->io_hdr.local_sglist;
11968 	local_dma_sglist = io->io_hdr.local_dma_sglist;
11969 	remote_sglist = io->io_hdr.remote_sglist;
11970 	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
11971 	local_used = 0;
11972 	remote_used = 0;
11973 	total_used = 0;
11974 
11975 	if (io->io_hdr.flags & CTL_FLAG_REDIR_DONE) {
11976 		rq->ret = CTL_HA_STATUS_SUCCESS;
11977 		rq->context = io;
11978 		callback(rq);
11979 		goto bailout;
11980 	}
11981 
11982 	/*
11983 	 * Pull/push the data over the wire from/to the other controller.
11984 	 * This takes into account the possibility that the local and
11985 	 * remote sglists may not be identical in terms of the size of
11986 	 * the elements and the number of elements.
11987 	 *
11988 	 * One fundamental assumption here is that the length allocated for
11989 	 * both the local and remote sglists is identical.  Otherwise, we've
11990 	 * essentially got a coding error of some sort.
11991 	 */
11992 	for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
11993 		int isc_ret;
11994 		uint32_t cur_len, dma_length;
11995 		uint8_t *tmp_ptr;
11996 
11997 		rq->id = CTL_HA_DATA_CTL;
11998 		rq->command = command;
11999 		rq->context = io;
12000 
12001 		/*
12002 		 * Both pointers should be aligned.  But it is possible
12003 		 * that the allocation length is not.  They should both
12004 		 * also have enough slack left over at the end, though,
12005 		 * to round up to the next 8 byte boundary.
12006 		 */
12007 		cur_len = ctl_min(local_sglist[i].len - local_used,
12008 				  remote_sglist[j].len - remote_used);
12009 
12010 		/*
12011 		 * In this case, we have a size issue and need to decrease
12012 		 * the size, except in the case where we actually have less
12013 		 * than 8 bytes left.  In that case, we need to increase
12014 		 * the DMA length to get the last bit.
12015 		 */
12016 		if ((cur_len & 0x7) != 0) {
12017 			if (cur_len > 0x7) {
12018 				cur_len = cur_len - (cur_len & 0x7);
12019 				dma_length = cur_len;
12020 			} else {
12021 				CTL_SIZE_8B(dma_length, cur_len);
12022 			}
12023 
12024 		} else
12025 			dma_length = cur_len;
12026 
12027 		/*
12028 		 * If we had to allocate memory for this I/O, instead of using
12029 		 * the non-cached mirror memory, we'll need to flush the cache
12030 		 * before trying to DMA to the other controller.
12031 		 *
12032 		 * We could end up doing this multiple times for the same
12033 		 * segment if we have a larger local segment than remote
12034 		 * segment.  That shouldn't be an issue.
12035 		 */
12036 		if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12037 			/*
12038 			 * XXX KDM use bus_dmamap_sync() here.
12039 			 */
12040 		}
12041 
12042 		rq->size = dma_length;
12043 
12044 		tmp_ptr = (uint8_t *)local_sglist[i].addr;
12045 		tmp_ptr += local_used;
12046 
12047 		/* Use physical addresses when talking to ISC hardware */
12048 		if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
12049 			/* XXX KDM use busdma */
12050 #if 0
12051 			rq->local = vtophys(tmp_ptr);
12052 #endif
12053 		} else
12054 			rq->local = tmp_ptr;
12055 
12056 		tmp_ptr = (uint8_t *)remote_sglist[j].addr;
12057 		tmp_ptr += remote_used;
12058 		rq->remote = tmp_ptr;
12059 
12060 		rq->callback = NULL;
12061 
12062 		local_used += cur_len;
12063 		if (local_used >= local_sglist[i].len) {
12064 			i++;
12065 			local_used = 0;
12066 		}
12067 
12068 		remote_used += cur_len;
12069 		if (remote_used >= remote_sglist[j].len) {
12070 			j++;
12071 			remote_used = 0;
12072 		}
12073 		total_used += cur_len;
12074 
12075 		if (total_used >= io->scsiio.kern_data_len)
12076 			rq->callback = callback;
12077 
12078 		if ((rq->size & 0x7) != 0) {
12079 			printf("%s: warning: size %d is not on 8b boundary\n",
12080 			       __func__, rq->size);
12081 		}
12082 		if (((uintptr_t)rq->local & 0x7) != 0) {
12083 			printf("%s: warning: local %p not on 8b boundary\n",
12084 			       __func__, rq->local);
12085 		}
12086 		if (((uintptr_t)rq->remote & 0x7) != 0) {
12087 			printf("%s: warning: remote %p not on 8b boundary\n",
12088 			       __func__, rq->local);
12089 		}
12090 #if 0
12091 		printf("%s: %s: local %#x remote %#x size %d\n", __func__,
12092 		       (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
12093 		       rq->local, rq->remote, rq->size);
12094 #endif
12095 
12096 		isc_ret = ctl_dt_single(rq);
12097 		if (isc_ret == CTL_HA_STATUS_WAIT)
12098 			continue;
12099 
12100 		if (isc_ret == CTL_HA_STATUS_DISCONNECT) {
12101 			rq->ret = CTL_HA_STATUS_SUCCESS;
12102 		} else {
12103 			rq->ret = isc_ret;
12104 		}
12105 		callback(rq);
12106 		goto bailout;
12107 	}
12108 
12109 bailout:
12110 	return (retval);
12111 
12112 }
12113 
12114 static void
12115 ctl_datamove_remote_read(union ctl_io *io)
12116 {
12117 	int retval;
12118 	int i;
12119 
12120 	/*
12121 	 * This will send an error to the other controller in the case of a
12122 	 * failure.
12123 	 */
12124 	retval = ctl_datamove_remote_sgl_setup(io);
12125 	if (retval != 0)
12126 		return;
12127 
12128 	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
12129 					  ctl_datamove_remote_read_cb);
12130 	if ((retval != 0)
12131 	 && ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0)) {
12132 		/*
12133 		 * Make sure we free memory if there was an error..  The
12134 		 * ctl_datamove_remote_xfer() function will send the
12135 		 * datamove done message, or call the callback with an
12136 		 * error if there is a problem.
12137 		 */
12138 		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12139 			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12140 	}
12141 
12142 	return;
12143 }
12144 
12145 /*
12146  * Process a datamove request from the other controller.  This is used for
12147  * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
12148  * first.  Once that is complete, the data gets DMAed into the remote
12149  * controller's memory.  For reads, we DMA from the remote controller's
12150  * memory into our memory first, and then move it out to the FETD.
12151  *
12152  * Should be called without the ctl_lock held.
12153  */
12154 static void
12155 ctl_datamove_remote(union ctl_io *io)
12156 {
12157 	struct ctl_softc *softc;
12158 
12159 	softc = control_softc;
12160 
12161 	/*
12162 	 * Note that we look for an aborted I/O here, but don't do some of
12163 	 * the other checks that ctl_datamove() normally does.  We don't
12164 	 * need to run the task queue, because this I/O is on the ISC
12165 	 * queue, which is executed by the work thread after the task queue.
12166 	 * We don't need to run the datamove delay code, since that should
12167 	 * have been done if need be on the other controller.
12168 	 */
12169 	mtx_lock(&softc->ctl_lock);
12170 
12171 	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12172 
12173 		printf("%s: tag 0x%04x on (%d:%d:%d:%d) aborted\n", __func__,
12174 		       io->scsiio.tag_num, io->io_hdr.nexus.initid.id,
12175 		       io->io_hdr.nexus.targ_port,
12176 		       io->io_hdr.nexus.targ_target.id,
12177 		       io->io_hdr.nexus.targ_lun);
12178 		io->io_hdr.status = CTL_CMD_ABORTED;
12179 		io->io_hdr.port_status = 31338;
12180 
12181 		mtx_unlock(&softc->ctl_lock);
12182 
12183 		ctl_send_datamove_done(io, /*have_lock*/ 0);
12184 
12185 		return;
12186 	}
12187 
12188 	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT) {
12189 		mtx_unlock(&softc->ctl_lock);
12190 		ctl_datamove_remote_write(io);
12191 	} else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN){
12192 		mtx_unlock(&softc->ctl_lock);
12193 		ctl_datamove_remote_read(io);
12194 	} else {
12195 		union ctl_ha_msg msg;
12196 		struct scsi_sense_data *sense;
12197 		uint8_t sks[3];
12198 		int retry_count;
12199 
12200 		memset(&msg, 0, sizeof(msg));
12201 
12202 		msg.hdr.msg_type = CTL_MSG_BAD_JUJU;
12203 		msg.hdr.status = CTL_SCSI_ERROR;
12204 		msg.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
12205 
12206 		retry_count = 4243;
12207 
12208 		sense = &msg.scsi.sense_data;
12209 		sks[0] = SSD_SCS_VALID;
12210 		sks[1] = (retry_count >> 8) & 0xff;
12211 		sks[2] = retry_count & 0xff;
12212 
12213 		/* "Internal target failure" */
12214 		scsi_set_sense_data(sense,
12215 				    /*sense_format*/ SSD_TYPE_NONE,
12216 				    /*current_error*/ 1,
12217 				    /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
12218 				    /*asc*/ 0x44,
12219 				    /*ascq*/ 0x00,
12220 				    /*type*/ SSD_ELEM_SKS,
12221 				    /*size*/ sizeof(sks),
12222 				    /*data*/ sks,
12223 				    SSD_ELEM_NONE);
12224 
12225 		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12226 		if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12227 			ctl_failover_io(io, /*have_lock*/ 1);
12228 			mtx_unlock(&softc->ctl_lock);
12229 			return;
12230 		}
12231 
12232 		mtx_unlock(&softc->ctl_lock);
12233 
12234 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0) >
12235 		    CTL_HA_STATUS_SUCCESS) {
12236 			/* XXX KDM what to do if this fails? */
12237 		}
12238 		return;
12239 	}
12240 
12241 }
12242 
12243 static int
12244 ctl_process_done(union ctl_io *io, int have_lock)
12245 {
12246 	struct ctl_lun *lun;
12247 	struct ctl_softc *ctl_softc;
12248 	void (*fe_done)(union ctl_io *io);
12249 	uint32_t targ_port = ctl_port_idx(io->io_hdr.nexus.targ_port);
12250 
12251 	CTL_DEBUG_PRINT(("ctl_process_done\n"));
12252 
12253 	fe_done =
12254 	    control_softc->ctl_ports[targ_port]->fe_done;
12255 
12256 #ifdef CTL_TIME_IO
12257 	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12258 		char str[256];
12259 		char path_str[64];
12260 		struct sbuf sb;
12261 
12262 		ctl_scsi_path_string(io, path_str, sizeof(path_str));
12263 		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12264 
12265 		sbuf_cat(&sb, path_str);
12266 		switch (io->io_hdr.io_type) {
12267 		case CTL_IO_SCSI:
12268 			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12269 			sbuf_printf(&sb, "\n");
12270 			sbuf_cat(&sb, path_str);
12271 			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12272 				    io->scsiio.tag_num, io->scsiio.tag_type);
12273 			break;
12274 		case CTL_IO_TASK:
12275 			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12276 				    "Tag Type: %d\n", io->taskio.task_action,
12277 				    io->taskio.tag_num, io->taskio.tag_type);
12278 			break;
12279 		default:
12280 			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12281 			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12282 			break;
12283 		}
12284 		sbuf_cat(&sb, path_str);
12285 		sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
12286 			    (intmax_t)time_uptime - io->io_hdr.start_time);
12287 		sbuf_finish(&sb);
12288 		printf("%s", sbuf_data(&sb));
12289 	}
12290 #endif /* CTL_TIME_IO */
12291 
12292 	switch (io->io_hdr.io_type) {
12293 	case CTL_IO_SCSI:
12294 		break;
12295 	case CTL_IO_TASK:
12296 		ctl_io_error_print(io, NULL);
12297 		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
12298 			ctl_free_io_internal(io, /*have_lock*/ 0);
12299 		else
12300 			fe_done(io);
12301 		return (CTL_RETVAL_COMPLETE);
12302 		break;
12303 	default:
12304 		printf("ctl_process_done: invalid io type %d\n",
12305 		       io->io_hdr.io_type);
12306 		panic("ctl_process_done: invalid io type %d\n",
12307 		      io->io_hdr.io_type);
12308 		break; /* NOTREACHED */
12309 	}
12310 
12311 	lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12312 	if (lun == NULL) {
12313 		CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
12314 				 io->io_hdr.nexus.targ_lun));
12315 		fe_done(io);
12316 		goto bailout;
12317 	}
12318 	ctl_softc = lun->ctl_softc;
12319 
12320 	/*
12321 	 * Remove this from the OOA queue.
12322 	 */
12323 	if (have_lock == 0)
12324 		mtx_lock(&ctl_softc->ctl_lock);
12325 
12326 	/*
12327 	 * Check to see if we have any errors to inject here.  We only
12328 	 * inject errors for commands that don't already have errors set.
12329 	 */
12330 	if ((STAILQ_FIRST(&lun->error_list) != NULL)
12331 	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))
12332 		ctl_inject_error(lun, io);
12333 
12334 	/*
12335 	 * XXX KDM how do we treat commands that aren't completed
12336 	 * successfully?
12337 	 *
12338 	 * XXX KDM should we also track I/O latency?
12339 	 */
12340 	if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) {
12341 		uint32_t blocksize;
12342 #ifdef CTL_TIME_IO
12343 		struct bintime cur_bt;
12344 #endif
12345 
12346 		if ((lun->be_lun != NULL)
12347 		 && (lun->be_lun->blocksize != 0))
12348 			blocksize = lun->be_lun->blocksize;
12349 		else
12350 			blocksize = 512;
12351 
12352 		switch (io->io_hdr.io_type) {
12353 		case CTL_IO_SCSI: {
12354 			int isread;
12355 			struct ctl_lba_len lbalen;
12356 
12357 			isread = 0;
12358 			switch (io->scsiio.cdb[0]) {
12359 			case READ_6:
12360 			case READ_10:
12361 			case READ_12:
12362 			case READ_16:
12363 				isread = 1;
12364 				/* FALLTHROUGH */
12365 			case WRITE_6:
12366 			case WRITE_10:
12367 			case WRITE_12:
12368 			case WRITE_16:
12369 			case WRITE_VERIFY_10:
12370 			case WRITE_VERIFY_12:
12371 			case WRITE_VERIFY_16:
12372 				memcpy(&lbalen, io->io_hdr.ctl_private[
12373 				       CTL_PRIV_LBA_LEN].bytes, sizeof(lbalen));
12374 
12375 				if (isread) {
12376 					lun->stats.ports[targ_port].bytes[CTL_STATS_READ] +=
12377 						lbalen.len * blocksize;
12378 					lun->stats.ports[targ_port].operations[CTL_STATS_READ]++;
12379 
12380 #ifdef CTL_TIME_IO
12381 					bintime_add(
12382 					   &lun->stats.ports[targ_port].dma_time[CTL_STATS_READ],
12383 					   &io->io_hdr.dma_bt);
12384 					lun->stats.ports[targ_port].num_dmas[CTL_STATS_READ] +=
12385 						io->io_hdr.num_dmas;
12386 					getbintime(&cur_bt);
12387 					bintime_sub(&cur_bt,
12388 						    &io->io_hdr.start_bt);
12389 
12390 					bintime_add(
12391 					    &lun->stats.ports[targ_port].time[CTL_STATS_READ],
12392 					    &cur_bt);
12393 
12394 #if 0
12395 					cs_prof_gettime(&cur_ticks);
12396 					lun->stats.time[CTL_STATS_READ] +=
12397 						cur_ticks -
12398 						io->io_hdr.start_ticks;
12399 #endif
12400 #if 0
12401 					lun->stats.time[CTL_STATS_READ] +=
12402 						jiffies - io->io_hdr.start_time;
12403 #endif
12404 #endif /* CTL_TIME_IO */
12405 				} else {
12406 					lun->stats.ports[targ_port].bytes[CTL_STATS_WRITE] +=
12407 						lbalen.len * blocksize;
12408 					lun->stats.ports[targ_port].operations[
12409 						CTL_STATS_WRITE]++;
12410 
12411 #ifdef CTL_TIME_IO
12412 					bintime_add(
12413 					  &lun->stats.ports[targ_port].dma_time[CTL_STATS_WRITE],
12414 					  &io->io_hdr.dma_bt);
12415 					lun->stats.ports[targ_port].num_dmas[CTL_STATS_WRITE] +=
12416 						io->io_hdr.num_dmas;
12417 					getbintime(&cur_bt);
12418 					bintime_sub(&cur_bt,
12419 						    &io->io_hdr.start_bt);
12420 
12421 					bintime_add(
12422 					    &lun->stats.ports[targ_port].time[CTL_STATS_WRITE],
12423 					    &cur_bt);
12424 #if 0
12425 					cs_prof_gettime(&cur_ticks);
12426 					lun->stats.ports[targ_port].time[CTL_STATS_WRITE] +=
12427 						cur_ticks -
12428 						io->io_hdr.start_ticks;
12429 					lun->stats.ports[targ_port].time[CTL_STATS_WRITE] +=
12430 						jiffies - io->io_hdr.start_time;
12431 #endif
12432 #endif /* CTL_TIME_IO */
12433 				}
12434 				break;
12435 			default:
12436 				lun->stats.ports[targ_port].operations[CTL_STATS_NO_IO]++;
12437 
12438 #ifdef CTL_TIME_IO
12439 				bintime_add(
12440 				  &lun->stats.ports[targ_port].dma_time[CTL_STATS_NO_IO],
12441 				  &io->io_hdr.dma_bt);
12442 				lun->stats.ports[targ_port].num_dmas[CTL_STATS_NO_IO] +=
12443 					io->io_hdr.num_dmas;
12444 				getbintime(&cur_bt);
12445 				bintime_sub(&cur_bt, &io->io_hdr.start_bt);
12446 
12447 				bintime_add(&lun->stats.ports[targ_port].time[CTL_STATS_NO_IO],
12448 					    &cur_bt);
12449 
12450 #if 0
12451 				cs_prof_gettime(&cur_ticks);
12452 				lun->stats.ports[targ_port].time[CTL_STATS_NO_IO] +=
12453 					cur_ticks -
12454 					io->io_hdr.start_ticks;
12455 				lun->stats.ports[targ_port].time[CTL_STATS_NO_IO] +=
12456 					jiffies - io->io_hdr.start_time;
12457 #endif
12458 #endif /* CTL_TIME_IO */
12459 				break;
12460 			}
12461 			break;
12462 		}
12463 		default:
12464 			break;
12465 		}
12466 	}
12467 
12468 	TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
12469 
12470 	/*
12471 	 * Run through the blocked queue on this LUN and see if anything
12472 	 * has become unblocked, now that this transaction is done.
12473 	 */
12474 	ctl_check_blocked(lun);
12475 
12476 	/*
12477 	 * If the LUN has been invalidated, free it if there is nothing
12478 	 * left on its OOA queue.
12479 	 */
12480 	if ((lun->flags & CTL_LUN_INVALID)
12481 	 && (TAILQ_FIRST(&lun->ooa_queue) == NULL))
12482 		ctl_free_lun(lun);
12483 
12484 	/*
12485 	 * If this command has been aborted, make sure we set the status
12486 	 * properly.  The FETD is responsible for freeing the I/O and doing
12487 	 * whatever it needs to do to clean up its state.
12488 	 */
12489 	if (io->io_hdr.flags & CTL_FLAG_ABORT)
12490 		io->io_hdr.status = CTL_CMD_ABORTED;
12491 
12492 	/*
12493 	 * We print out status for every task management command.  For SCSI
12494 	 * commands, we filter out any unit attention errors; they happen
12495 	 * on every boot, and would clutter up the log.  Note:  task
12496 	 * management commands aren't printed here, they are printed above,
12497 	 * since they should never even make it down here.
12498 	 */
12499 	switch (io->io_hdr.io_type) {
12500 	case CTL_IO_SCSI: {
12501 		int error_code, sense_key, asc, ascq;
12502 
12503 		sense_key = 0;
12504 
12505 		if (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SCSI_ERROR)
12506 		 && (io->scsiio.scsi_status == SCSI_STATUS_CHECK_COND)) {
12507 			/*
12508 			 * Since this is just for printing, no need to
12509 			 * show errors here.
12510 			 */
12511 			scsi_extract_sense_len(&io->scsiio.sense_data,
12512 					       io->scsiio.sense_len,
12513 					       &error_code,
12514 					       &sense_key,
12515 					       &asc,
12516 					       &ascq,
12517 					       /*show_errors*/ 0);
12518 		}
12519 
12520 		if (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
12521 		 && (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SCSI_ERROR)
12522 		  || (io->scsiio.scsi_status != SCSI_STATUS_CHECK_COND)
12523 		  || (sense_key != SSD_KEY_UNIT_ATTENTION))) {
12524 
12525 			if ((time_uptime - ctl_softc->last_print_jiffies) <= 0){
12526 				ctl_softc->skipped_prints++;
12527 				if (have_lock == 0)
12528 					mtx_unlock(&ctl_softc->ctl_lock);
12529 			} else {
12530 				uint32_t skipped_prints;
12531 
12532 				skipped_prints = ctl_softc->skipped_prints;
12533 
12534 				ctl_softc->skipped_prints = 0;
12535 				ctl_softc->last_print_jiffies = time_uptime;
12536 
12537 				if (have_lock == 0)
12538 					mtx_unlock(&ctl_softc->ctl_lock);
12539 				if (skipped_prints > 0) {
12540 #ifdef NEEDTOPORT
12541 					csevent_log(CSC_CTL | CSC_SHELF_SW |
12542 					    CTL_ERROR_REPORT,
12543 					    csevent_LogType_Trace,
12544 					    csevent_Severity_Information,
12545 					    csevent_AlertLevel_Green,
12546 					    csevent_FRU_Firmware,
12547 					    csevent_FRU_Unknown,
12548 					    "High CTL error volume, %d prints "
12549 					    "skipped", skipped_prints);
12550 #endif
12551 				}
12552 				ctl_io_error_print(io, NULL);
12553 			}
12554 		} else {
12555 			if (have_lock == 0)
12556 				mtx_unlock(&ctl_softc->ctl_lock);
12557 		}
12558 		break;
12559 	}
12560 	case CTL_IO_TASK:
12561 		if (have_lock == 0)
12562 			mtx_unlock(&ctl_softc->ctl_lock);
12563 		ctl_io_error_print(io, NULL);
12564 		break;
12565 	default:
12566 		if (have_lock == 0)
12567 			mtx_unlock(&ctl_softc->ctl_lock);
12568 		break;
12569 	}
12570 
12571 	/*
12572 	 * Tell the FETD or the other shelf controller we're done with this
12573 	 * command.  Note that only SCSI commands get to this point.  Task
12574 	 * management commands are completed above.
12575 	 *
12576 	 * We only send status to the other controller if we're in XFER
12577 	 * mode.  In SER_ONLY mode, the I/O is done on the controller that
12578 	 * received the I/O (from CTL's perspective), and so the status is
12579 	 * generated there.
12580 	 *
12581 	 * XXX KDM if we hold the lock here, we could cause a deadlock
12582 	 * if the frontend comes back in in this context to queue
12583 	 * something.
12584 	 */
12585 	if ((ctl_softc->ha_mode == CTL_HA_MODE_XFER)
12586 	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12587 		union ctl_ha_msg msg;
12588 
12589 		memset(&msg, 0, sizeof(msg));
12590 		msg.hdr.msg_type = CTL_MSG_FINISH_IO;
12591 		msg.hdr.original_sc = io->io_hdr.original_sc;
12592 		msg.hdr.nexus = io->io_hdr.nexus;
12593 		msg.hdr.status = io->io_hdr.status;
12594 		msg.scsi.scsi_status = io->scsiio.scsi_status;
12595 		msg.scsi.tag_num = io->scsiio.tag_num;
12596 		msg.scsi.tag_type = io->scsiio.tag_type;
12597 		msg.scsi.sense_len = io->scsiio.sense_len;
12598 		msg.scsi.sense_residual = io->scsiio.sense_residual;
12599 		msg.scsi.residual = io->scsiio.residual;
12600 		memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12601 		       sizeof(io->scsiio.sense_data));
12602 		/*
12603 		 * We copy this whether or not this is an I/O-related
12604 		 * command.  Otherwise, we'd have to go and check to see
12605 		 * whether it's a read/write command, and it really isn't
12606 		 * worth it.
12607 		 */
12608 		memcpy(&msg.scsi.lbalen,
12609 		       &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
12610 		       sizeof(msg.scsi.lbalen));
12611 
12612 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12613 				sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
12614 			/* XXX do something here */
12615 		}
12616 
12617 		ctl_free_io_internal(io, /*have_lock*/ 0);
12618 	} else
12619 		fe_done(io);
12620 
12621 bailout:
12622 
12623 	return (CTL_RETVAL_COMPLETE);
12624 }
12625 
12626 /*
12627  * Front end should call this if it doesn't do autosense.  When the request
12628  * sense comes back in from the initiator, we'll dequeue this and send it.
12629  */
12630 int
12631 ctl_queue_sense(union ctl_io *io)
12632 {
12633 	struct ctl_lun *lun;
12634 	struct ctl_softc *ctl_softc;
12635 	uint32_t initidx;
12636 
12637 	ctl_softc = control_softc;
12638 
12639 	CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
12640 
12641 	/*
12642 	 * LUN lookup will likely move to the ctl_work_thread() once we
12643 	 * have our new queueing infrastructure (that doesn't put things on
12644 	 * a per-LUN queue initially).  That is so that we can handle
12645 	 * things like an INQUIRY to a LUN that we don't have enabled.  We
12646 	 * can't deal with that right now.
12647 	 */
12648 	mtx_lock(&ctl_softc->ctl_lock);
12649 
12650 	/*
12651 	 * If we don't have a LUN for this, just toss the sense
12652 	 * information.
12653 	 */
12654 	if ((io->io_hdr.nexus.targ_lun < CTL_MAX_LUNS)
12655 	 && (ctl_softc->ctl_luns[io->io_hdr.nexus.targ_lun] != NULL))
12656 		lun = ctl_softc->ctl_luns[io->io_hdr.nexus.targ_lun];
12657 	else
12658 		goto bailout;
12659 
12660 	initidx = ctl_get_initindex(&io->io_hdr.nexus);
12661 
12662 	/*
12663 	 * Already have CA set for this LUN...toss the sense information.
12664 	 */
12665 	if (ctl_is_set(lun->have_ca, initidx))
12666 		goto bailout;
12667 
12668 	memcpy(&lun->pending_sense[initidx].sense, &io->scsiio.sense_data,
12669 	       ctl_min(sizeof(lun->pending_sense[initidx].sense),
12670 	       sizeof(io->scsiio.sense_data)));
12671 	ctl_set_mask(lun->have_ca, initidx);
12672 
12673 bailout:
12674 	mtx_unlock(&ctl_softc->ctl_lock);
12675 
12676 	ctl_free_io(io);
12677 
12678 	return (CTL_RETVAL_COMPLETE);
12679 }
12680 
12681 /*
12682  * Primary command inlet from frontend ports.  All SCSI and task I/O
12683  * requests must go through this function.
12684  */
12685 int
12686 ctl_queue(union ctl_io *io)
12687 {
12688 	struct ctl_softc *ctl_softc;
12689 
12690 	CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
12691 
12692 	ctl_softc = control_softc;
12693 
12694 #ifdef CTL_TIME_IO
12695 	io->io_hdr.start_time = time_uptime;
12696 	getbintime(&io->io_hdr.start_bt);
12697 #endif /* CTL_TIME_IO */
12698 
12699 	mtx_lock(&ctl_softc->ctl_lock);
12700 
12701 	switch (io->io_hdr.io_type) {
12702 	case CTL_IO_SCSI:
12703 		STAILQ_INSERT_TAIL(&ctl_softc->incoming_queue, &io->io_hdr,
12704 				   links);
12705 		break;
12706 	case CTL_IO_TASK:
12707 		STAILQ_INSERT_TAIL(&ctl_softc->task_queue, &io->io_hdr, links);
12708 		/*
12709 		 * Set the task pending flag.  This is necessary to close a
12710 		 * race condition with the FETD:
12711 		 *
12712 		 * - FETD submits a task management command, like an abort.
12713 		 * - Back end calls fe_datamove() to move the data for the
12714 		 *   aborted command.  The FETD can't really accept it, but
12715 		 *   if it did, it would end up transmitting data for a
12716 		 *   command that the initiator told us to abort.
12717 		 *
12718 		 * We close the race condition by setting the flag here,
12719 		 * and checking it in ctl_datamove(), before calling the
12720 		 * FETD's fe_datamove routine.  If we've got a task
12721 		 * pending, we run the task queue and then check to see
12722 		 * whether our particular I/O has been aborted.
12723 		 */
12724 		ctl_softc->flags |= CTL_FLAG_TASK_PENDING;
12725 		break;
12726 	default:
12727 		mtx_unlock(&ctl_softc->ctl_lock);
12728 		printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
12729 		return (-EINVAL);
12730 		break; /* NOTREACHED */
12731 	}
12732 	mtx_unlock(&ctl_softc->ctl_lock);
12733 
12734 	ctl_wakeup_thread();
12735 
12736 	return (CTL_RETVAL_COMPLETE);
12737 }
12738 
12739 #ifdef CTL_IO_DELAY
12740 static void
12741 ctl_done_timer_wakeup(void *arg)
12742 {
12743 	union ctl_io *io;
12744 
12745 	io = (union ctl_io *)arg;
12746 	ctl_done_lock(io, /*have_lock*/ 0);
12747 }
12748 #endif /* CTL_IO_DELAY */
12749 
12750 void
12751 ctl_done_lock(union ctl_io *io, int have_lock)
12752 {
12753 	struct ctl_softc *ctl_softc;
12754 #ifndef CTL_DONE_THREAD
12755 	union ctl_io *xio;
12756 #endif /* !CTL_DONE_THREAD */
12757 
12758 	ctl_softc = control_softc;
12759 
12760 	if (have_lock == 0)
12761 		mtx_lock(&ctl_softc->ctl_lock);
12762 
12763 	/*
12764 	 * Enable this to catch duplicate completion issues.
12765 	 */
12766 #if 0
12767 	if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
12768 		printf("%s: type %d msg %d cdb %x iptl: "
12769 		       "%d:%d:%d:%d tag 0x%04x "
12770 		       "flag %#x status %x\n",
12771 			__func__,
12772 			io->io_hdr.io_type,
12773 			io->io_hdr.msg_type,
12774 			io->scsiio.cdb[0],
12775 			io->io_hdr.nexus.initid.id,
12776 			io->io_hdr.nexus.targ_port,
12777 			io->io_hdr.nexus.targ_target.id,
12778 			io->io_hdr.nexus.targ_lun,
12779 			(io->io_hdr.io_type ==
12780 			CTL_IO_TASK) ?
12781 			io->taskio.tag_num :
12782 			io->scsiio.tag_num,
12783 		        io->io_hdr.flags,
12784 			io->io_hdr.status);
12785 	} else
12786 		io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
12787 #endif
12788 
12789 	/*
12790 	 * This is an internal copy of an I/O, and should not go through
12791 	 * the normal done processing logic.
12792 	 */
12793 	if (io->io_hdr.flags & CTL_FLAG_INT_COPY) {
12794 		if (have_lock == 0)
12795 			mtx_unlock(&ctl_softc->ctl_lock);
12796 		return;
12797 	}
12798 
12799 	/*
12800 	 * We need to send a msg to the serializing shelf to finish the IO
12801 	 * as well.  We don't send a finish message to the other shelf if
12802 	 * this is a task management command.  Task management commands
12803 	 * aren't serialized in the OOA queue, but rather just executed on
12804 	 * both shelf controllers for commands that originated on that
12805 	 * controller.
12806 	 */
12807 	if ((io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)
12808 	 && (io->io_hdr.io_type != CTL_IO_TASK)) {
12809 		union ctl_ha_msg msg_io;
12810 
12811 		msg_io.hdr.msg_type = CTL_MSG_FINISH_IO;
12812 		msg_io.hdr.serializing_sc = io->io_hdr.serializing_sc;
12813 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_io,
12814 		    sizeof(msg_io), 0 ) != CTL_HA_STATUS_SUCCESS) {
12815 		}
12816 		/* continue on to finish IO */
12817 	}
12818 #ifdef CTL_IO_DELAY
12819 	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12820 		struct ctl_lun *lun;
12821 
12822 		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12823 
12824 		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12825 	} else {
12826 		struct ctl_lun *lun;
12827 
12828 		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12829 
12830 		if ((lun != NULL)
12831 		 && (lun->delay_info.done_delay > 0)) {
12832 			struct callout *callout;
12833 
12834 			callout = (struct callout *)&io->io_hdr.timer_bytes;
12835 			callout_init(callout, /*mpsafe*/ 1);
12836 			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12837 			callout_reset(callout,
12838 				      lun->delay_info.done_delay * hz,
12839 				      ctl_done_timer_wakeup, io);
12840 			if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
12841 				lun->delay_info.done_delay = 0;
12842 			if (have_lock == 0)
12843 				mtx_unlock(&ctl_softc->ctl_lock);
12844 			return;
12845 		}
12846 	}
12847 #endif /* CTL_IO_DELAY */
12848 
12849 	STAILQ_INSERT_TAIL(&ctl_softc->done_queue, &io->io_hdr, links);
12850 
12851 #ifdef CTL_DONE_THREAD
12852 	if (have_lock == 0)
12853 		mtx_unlock(&ctl_softc->ctl_lock);
12854 
12855 	ctl_wakeup_thread();
12856 #else /* CTL_DONE_THREAD */
12857 	for (xio = (union ctl_io *)STAILQ_FIRST(&ctl_softc->done_queue);
12858 	     xio != NULL;
12859 	     xio =(union ctl_io *)STAILQ_FIRST(&ctl_softc->done_queue)) {
12860 
12861 		STAILQ_REMOVE_HEAD(&ctl_softc->done_queue, links);
12862 
12863 		ctl_process_done(xio, /*have_lock*/ 1);
12864 	}
12865 	if (have_lock == 0)
12866 		mtx_unlock(&ctl_softc->ctl_lock);
12867 #endif /* CTL_DONE_THREAD */
12868 }
12869 
12870 void
12871 ctl_done(union ctl_io *io)
12872 {
12873 	ctl_done_lock(io, /*have_lock*/ 0);
12874 }
12875 
12876 int
12877 ctl_isc(struct ctl_scsiio *ctsio)
12878 {
12879 	struct ctl_lun *lun;
12880 	int retval;
12881 
12882 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12883 
12884 	CTL_DEBUG_PRINT(("ctl_isc: command: %02x\n", ctsio->cdb[0]));
12885 
12886 	CTL_DEBUG_PRINT(("ctl_isc: calling data_submit()\n"));
12887 
12888 	retval = lun->backend->data_submit((union ctl_io *)ctsio);
12889 
12890 	return (retval);
12891 }
12892 
12893 
12894 static void
12895 ctl_work_thread(void *arg)
12896 {
12897 	struct ctl_softc *softc;
12898 	union ctl_io *io;
12899 	struct ctl_be_lun *be_lun;
12900 	int retval;
12901 
12902 	CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
12903 
12904 	softc = (struct ctl_softc *)arg;
12905 	if (softc == NULL)
12906 		return;
12907 
12908 	mtx_lock(&softc->ctl_lock);
12909 	for (;;) {
12910 		retval = 0;
12911 
12912 		/*
12913 		 * We handle the queues in this order:
12914 		 * - task management
12915 		 * - ISC
12916 		 * - done queue (to free up resources, unblock other commands)
12917 		 * - RtR queue
12918 		 * - incoming queue
12919 		 *
12920 		 * If those queues are empty, we break out of the loop and
12921 		 * go to sleep.
12922 		 */
12923 		io = (union ctl_io *)STAILQ_FIRST(&softc->task_queue);
12924 		if (io != NULL) {
12925 			ctl_run_task_queue(softc);
12926 			continue;
12927 		}
12928 		io = (union ctl_io *)STAILQ_FIRST(&softc->isc_queue);
12929 		if (io != NULL) {
12930 			STAILQ_REMOVE_HEAD(&softc->isc_queue, links);
12931 			ctl_handle_isc(io);
12932 			continue;
12933 		}
12934 		io = (union ctl_io *)STAILQ_FIRST(&softc->done_queue);
12935 		if (io != NULL) {
12936 			STAILQ_REMOVE_HEAD(&softc->done_queue, links);
12937 			/* clear any blocked commands, call fe_done */
12938 			mtx_unlock(&softc->ctl_lock);
12939 			/*
12940 			 * XXX KDM
12941 			 * Call this without a lock for now.  This will
12942 			 * depend on whether there is any way the FETD can
12943 			 * sleep or deadlock if called with the CTL lock
12944 			 * held.
12945 			 */
12946 			retval = ctl_process_done(io, /*have_lock*/ 0);
12947 			mtx_lock(&softc->ctl_lock);
12948 			continue;
12949 		}
12950 		if (!ctl_pause_rtr) {
12951 			io = (union ctl_io *)STAILQ_FIRST(&softc->rtr_queue);
12952 			if (io != NULL) {
12953 				STAILQ_REMOVE_HEAD(&softc->rtr_queue, links);
12954 				mtx_unlock(&softc->ctl_lock);
12955 				goto execute;
12956 			}
12957 		}
12958 		io = (union ctl_io *)STAILQ_FIRST(&softc->incoming_queue);
12959 		if (io != NULL) {
12960 			STAILQ_REMOVE_HEAD(&softc->incoming_queue, links);
12961 			mtx_unlock(&softc->ctl_lock);
12962 			ctl_scsiio_precheck(softc, &io->scsiio);
12963 			mtx_lock(&softc->ctl_lock);
12964 			continue;
12965 		}
12966 		/*
12967 		 * We might want to move this to a separate thread, so that
12968 		 * configuration requests (in this case LUN creations)
12969 		 * won't impact the I/O path.
12970 		 */
12971 		be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
12972 		if (be_lun != NULL) {
12973 			STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
12974 			mtx_unlock(&softc->ctl_lock);
12975 			ctl_create_lun(be_lun);
12976 			mtx_lock(&softc->ctl_lock);
12977 			continue;
12978 		}
12979 
12980 		/* XXX KDM use the PDROP flag?? */
12981 		/* Sleep until we have something to do. */
12982 		mtx_sleep(softc, &softc->ctl_lock, PRIBIO, "ctl_work", 0);
12983 
12984 		/* Back to the top of the loop to see what woke us up. */
12985 		continue;
12986 
12987 execute:
12988 		retval = ctl_scsiio(&io->scsiio);
12989 		switch (retval) {
12990 		case CTL_RETVAL_COMPLETE:
12991 			break;
12992 		default:
12993 			/*
12994 			 * Probably need to make sure this doesn't happen.
12995 			 */
12996 			break;
12997 		}
12998 		mtx_lock(&softc->ctl_lock);
12999 	}
13000 }
13001 
13002 void
13003 ctl_wakeup_thread()
13004 {
13005 	struct ctl_softc *softc;
13006 
13007 	softc = control_softc;
13008 
13009 	wakeup(softc);
13010 }
13011 
13012 /* Initialization and failover */
13013 
13014 void
13015 ctl_init_isc_msg(void)
13016 {
13017 	printf("CTL: Still calling this thing\n");
13018 }
13019 
13020 /*
13021  * Init component
13022  * 	Initializes component into configuration defined by bootMode
13023  *	(see hasc-sv.c)
13024  *  	returns hasc_Status:
13025  * 		OK
13026  *		ERROR - fatal error
13027  */
13028 static ctl_ha_comp_status
13029 ctl_isc_init(struct ctl_ha_component *c)
13030 {
13031 	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
13032 
13033 	c->status = ret;
13034 	return ret;
13035 }
13036 
13037 /* Start component
13038  * 	Starts component in state requested. If component starts successfully,
13039  *	it must set its own state to the requestrd state
13040  *	When requested state is HASC_STATE_HA, the component may refine it
13041  * 	by adding _SLAVE or _MASTER flags.
13042  *	Currently allowed state transitions are:
13043  *	UNKNOWN->HA		- initial startup
13044  *	UNKNOWN->SINGLE - initial startup when no parter detected
13045  *	HA->SINGLE		- failover
13046  * returns ctl_ha_comp_status:
13047  * 		OK	- component successfully started in requested state
13048  *		FAILED  - could not start the requested state, failover may
13049  * 			  be possible
13050  *		ERROR	- fatal error detected, no future startup possible
13051  */
13052 static ctl_ha_comp_status
13053 ctl_isc_start(struct ctl_ha_component *c, ctl_ha_state state)
13054 {
13055 	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
13056 
13057 	// UNKNOWN->HA or UNKNOWN->SINGLE (bootstrap)
13058 	if (c->state == CTL_HA_STATE_UNKNOWN ) {
13059 		ctl_is_single = 0;
13060 		if (ctl_ha_msg_create(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
13061 		    != CTL_HA_STATUS_SUCCESS) {
13062 			printf("ctl_isc_start: ctl_ha_msg_create failed.\n");
13063 			ret = CTL_HA_COMP_STATUS_ERROR;
13064 		}
13065 	} else if (CTL_HA_STATE_IS_HA(c->state)
13066 		&& CTL_HA_STATE_IS_SINGLE(state)){
13067 		// HA->SINGLE transition
13068 	        ctl_failover();
13069 		ctl_is_single = 1;
13070 	} else {
13071 		printf("ctl_isc_start:Invalid state transition %X->%X\n",
13072 		       c->state, state);
13073 		ret = CTL_HA_COMP_STATUS_ERROR;
13074 	}
13075 	if (CTL_HA_STATE_IS_SINGLE(state))
13076 		ctl_is_single = 1;
13077 
13078 	c->state = state;
13079 	c->status = ret;
13080 	return ret;
13081 }
13082 
13083 /*
13084  * Quiesce component
13085  * The component must clear any error conditions (set status to OK) and
13086  * prepare itself to another Start call
13087  * returns ctl_ha_comp_status:
13088  * 	OK
13089  *	ERROR
13090  */
13091 static ctl_ha_comp_status
13092 ctl_isc_quiesce(struct ctl_ha_component *c)
13093 {
13094 	int ret = CTL_HA_COMP_STATUS_OK;
13095 
13096 	ctl_pause_rtr = 1;
13097 	c->status = ret;
13098 	return ret;
13099 }
13100 
13101 struct ctl_ha_component ctl_ha_component_ctlisc =
13102 {
13103 	.name = "CTL ISC",
13104 	.state = CTL_HA_STATE_UNKNOWN,
13105 	.init = ctl_isc_init,
13106 	.start = ctl_isc_start,
13107 	.quiesce = ctl_isc_quiesce
13108 };
13109 
13110 /*
13111  *  vim: ts=8
13112  */
13113