xref: /freebsd/sys/cam/ctl/ctl.c (revision 9124ddeb4a551977cf6b2218291e7c666ce25f47)
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 SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD, 0, "CAM Target Layer");
312 
313 /*
314  * XXX KDM move these into the softc.
315  */
316 static int rcv_sync_msg;
317 static int persis_offset;
318 static uint8_t ctl_pause_rtr;
319 static int     ctl_is_single;
320 static int     index_to_aps_page;
321 
322 
323 /*
324  * Serial number (0x80), device id (0x83), and supported pages (0x00)
325  */
326 #define SCSI_EVPD_NUM_SUPPORTED_PAGES	3
327 
328 static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
329 				  int param);
330 static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
331 static void ctl_init(void);
332 void ctl_shutdown(void);
333 static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
334 static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
335 static void ctl_ioctl_online(void *arg);
336 static void ctl_ioctl_offline(void *arg);
337 static int ctl_ioctl_targ_enable(void *arg, struct ctl_id targ_id);
338 static int ctl_ioctl_targ_disable(void *arg, struct ctl_id targ_id);
339 static int ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id);
340 static int ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id);
341 static int ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio);
342 static int ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio, int have_lock);
343 static int ctl_ioctl_submit_wait(union ctl_io *io);
344 static void ctl_ioctl_datamove(union ctl_io *io);
345 static void ctl_ioctl_done(union ctl_io *io);
346 static void ctl_ioctl_hard_startstop_callback(void *arg,
347 					      struct cfi_metatask *metatask);
348 static void ctl_ioctl_bbrread_callback(void *arg,struct cfi_metatask *metatask);
349 static int ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
350 			      struct ctl_ooa *ooa_hdr);
351 static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
352 		     struct thread *td);
353 uint32_t ctl_get_resindex(struct ctl_nexus *nexus);
354 uint32_t ctl_port_idx(int port_num);
355 #ifdef unused
356 static union ctl_io *ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port,
357 				   uint32_t targ_target, uint32_t targ_lun,
358 				   int can_wait);
359 static void ctl_kfree_io(union ctl_io *io);
360 #endif /* unused */
361 static void ctl_free_io_internal(union ctl_io *io, int have_lock);
362 static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
363 			 struct ctl_be_lun *be_lun, struct ctl_id target_id);
364 static int ctl_free_lun(struct ctl_lun *lun);
365 static void ctl_create_lun(struct ctl_be_lun *be_lun);
366 /**
367 static void ctl_failover_change_pages(struct ctl_softc *softc,
368 				      struct ctl_scsiio *ctsio, int master);
369 **/
370 
371 static int ctl_do_mode_select(union ctl_io *io);
372 static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
373 			   uint64_t res_key, uint64_t sa_res_key,
374 			   uint8_t type, uint32_t residx,
375 			   struct ctl_scsiio *ctsio,
376 			   struct scsi_per_res_out *cdb,
377 			   struct scsi_per_res_out_parms* param);
378 static void ctl_pro_preempt_other(struct ctl_lun *lun,
379 				  union ctl_ha_msg *msg);
380 static void ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg);
381 static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
382 static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
383 static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
384 static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
385 static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
386 static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint32_t *len);
387 static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2);
388 static ctl_action ctl_check_for_blockage(union ctl_io *pending_io,
389 					 union ctl_io *ooa_io);
390 static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
391 				union ctl_io *starting_io);
392 static int ctl_check_blocked(struct ctl_lun *lun);
393 static int ctl_scsiio_lun_check(struct ctl_softc *ctl_softc,
394 				struct ctl_lun *lun,
395 				struct ctl_cmd_entry *entry,
396 				struct ctl_scsiio *ctsio);
397 //static int ctl_check_rtr(union ctl_io *pending_io, struct ctl_softc *softc);
398 static void ctl_failover(void);
399 static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
400 			       struct ctl_scsiio *ctsio);
401 static int ctl_scsiio(struct ctl_scsiio *ctsio);
402 
403 static int ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
404 static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
405 			    ctl_ua_type ua_type);
406 static int ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io,
407 			 ctl_ua_type ua_type);
408 static int ctl_abort_task(union ctl_io *io);
409 static void ctl_run_task_queue(struct ctl_softc *ctl_softc);
410 #ifdef CTL_IO_DELAY
411 static void ctl_datamove_timer_wakeup(void *arg);
412 static void ctl_done_timer_wakeup(void *arg);
413 #endif /* CTL_IO_DELAY */
414 
415 static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
416 static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
417 static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
418 static void ctl_datamove_remote_write(union ctl_io *io);
419 static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
420 static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
421 static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
422 static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
423 				    ctl_ha_dt_cb callback);
424 static void ctl_datamove_remote_read(union ctl_io *io);
425 static void ctl_datamove_remote(union ctl_io *io);
426 static int ctl_process_done(union ctl_io *io, int have_lock);
427 static void ctl_work_thread(void *arg);
428 
429 /*
430  * Load the serialization table.  This isn't very pretty, but is probably
431  * the easiest way to do it.
432  */
433 #include "ctl_ser_table.c"
434 
435 /*
436  * We only need to define open, close and ioctl routines for this driver.
437  */
438 static struct cdevsw ctl_cdevsw = {
439 	.d_version =	D_VERSION,
440 	.d_flags =	0,
441 	.d_open =	ctl_open,
442 	.d_close =	ctl_close,
443 	.d_ioctl =	ctl_ioctl,
444 	.d_name =	"ctl",
445 };
446 
447 
448 MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
449 
450 /*
451  * If we have the CAM SIM, we may or may not have another SIM that will
452  * cause CTL to get initialized.  If not, we need to initialize it.
453  */
454 SYSINIT(ctl_init, SI_SUB_CONFIGURE, SI_ORDER_THIRD, ctl_init, NULL);
455 
456 static void
457 ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
458 			    union ctl_ha_msg *msg_info)
459 {
460 	struct ctl_scsiio *ctsio;
461 
462 	if (msg_info->hdr.original_sc == NULL) {
463 		printf("%s: original_sc == NULL!\n", __func__);
464 		/* XXX KDM now what? */
465 		return;
466 	}
467 
468 	ctsio = &msg_info->hdr.original_sc->scsiio;
469 	ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
470 	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
471 	ctsio->io_hdr.status = msg_info->hdr.status;
472 	ctsio->scsi_status = msg_info->scsi.scsi_status;
473 	ctsio->sense_len = msg_info->scsi.sense_len;
474 	ctsio->sense_residual = msg_info->scsi.sense_residual;
475 	ctsio->residual = msg_info->scsi.residual;
476 	memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
477 	       sizeof(ctsio->sense_data));
478 	memcpy(&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
479 	       &msg_info->scsi.lbalen, sizeof(msg_info->scsi.lbalen));;
480 	STAILQ_INSERT_TAIL(&ctl_softc->isc_queue, &ctsio->io_hdr, links);
481 	ctl_wakeup_thread();
482 }
483 
484 static void
485 ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
486 				union ctl_ha_msg *msg_info)
487 {
488 	struct ctl_scsiio *ctsio;
489 
490 	if (msg_info->hdr.serializing_sc == NULL) {
491 		printf("%s: serializing_sc == NULL!\n", __func__);
492 		/* XXX KDM now what? */
493 		return;
494 	}
495 
496 	ctsio = &msg_info->hdr.serializing_sc->scsiio;
497 #if 0
498 	/*
499 	 * Attempt to catch the situation where an I/O has
500 	 * been freed, and we're using it again.
501 	 */
502 	if (ctsio->io_hdr.io_type == 0xff) {
503 		union ctl_io *tmp_io;
504 		tmp_io = (union ctl_io *)ctsio;
505 		printf("%s: %p use after free!\n", __func__,
506 		       ctsio);
507 		printf("%s: type %d msg %d cdb %x iptl: "
508 		       "%d:%d:%d:%d tag 0x%04x "
509 		       "flag %#x status %x\n",
510 			__func__,
511 			tmp_io->io_hdr.io_type,
512 			tmp_io->io_hdr.msg_type,
513 			tmp_io->scsiio.cdb[0],
514 			tmp_io->io_hdr.nexus.initid.id,
515 			tmp_io->io_hdr.nexus.targ_port,
516 			tmp_io->io_hdr.nexus.targ_target.id,
517 			tmp_io->io_hdr.nexus.targ_lun,
518 			(tmp_io->io_hdr.io_type ==
519 			CTL_IO_TASK) ?
520 			tmp_io->taskio.tag_num :
521 			tmp_io->scsiio.tag_num,
522 		        tmp_io->io_hdr.flags,
523 			tmp_io->io_hdr.status);
524 	}
525 #endif
526 	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
527 	STAILQ_INSERT_TAIL(&ctl_softc->isc_queue, &ctsio->io_hdr, links);
528 	ctl_wakeup_thread();
529 }
530 
531 /*
532  * ISC (Inter Shelf Communication) event handler.  Events from the HA
533  * subsystem come in here.
534  */
535 static void
536 ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
537 {
538 	struct ctl_softc *ctl_softc;
539 	union ctl_io *io;
540 	struct ctl_prio *presio;
541 	ctl_ha_status isc_status;
542 
543 	ctl_softc = control_softc;
544 	io = NULL;
545 
546 
547 #if 0
548 	printf("CTL: Isc Msg event %d\n", event);
549 #endif
550 	if (event == CTL_HA_EVT_MSG_RECV) {
551 		union ctl_ha_msg msg_info;
552 
553 		isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
554 					     sizeof(msg_info), /*wait*/ 0);
555 #if 0
556 		printf("CTL: msg_type %d\n", msg_info.msg_type);
557 #endif
558 		if (isc_status != 0) {
559 			printf("Error receiving message, status = %d\n",
560 			       isc_status);
561 			return;
562 		}
563 		mtx_lock(&ctl_softc->ctl_lock);
564 
565 		switch (msg_info.hdr.msg_type) {
566 		case CTL_MSG_SERIALIZE:
567 #if 0
568 			printf("Serialize\n");
569 #endif
570 			io = ctl_alloc_io((void *)ctl_softc->othersc_pool);
571 			if (io == NULL) {
572 				printf("ctl_isc_event_handler: can't allocate "
573 				       "ctl_io!\n");
574 				/* Bad Juju */
575 				/* Need to set busy and send msg back */
576 				mtx_unlock(&ctl_softc->ctl_lock);
577 				msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
578 				msg_info.hdr.status = CTL_SCSI_ERROR;
579 				msg_info.scsi.scsi_status = SCSI_STATUS_BUSY;
580 				msg_info.scsi.sense_len = 0;
581 			        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
582 				    sizeof(msg_info), 0) > CTL_HA_STATUS_SUCCESS){
583 				}
584 				goto bailout;
585 			}
586 			ctl_zero_io(io);
587 			// populate ctsio from msg_info
588 			io->io_hdr.io_type = CTL_IO_SCSI;
589 			io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
590 			io->io_hdr.original_sc = msg_info.hdr.original_sc;
591 #if 0
592 			printf("pOrig %x\n", (int)msg_info.original_sc);
593 #endif
594 			io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
595 					    CTL_FLAG_IO_ACTIVE;
596 			/*
597 			 * If we're in serialization-only mode, we don't
598 			 * want to go through full done processing.  Thus
599 			 * the COPY flag.
600 			 *
601 			 * XXX KDM add another flag that is more specific.
602 			 */
603 			if (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)
604 				io->io_hdr.flags |= CTL_FLAG_INT_COPY;
605 			io->io_hdr.nexus = msg_info.hdr.nexus;
606 #if 0
607 			printf("targ %d, port %d, iid %d, lun %d\n",
608 			       io->io_hdr.nexus.targ_target.id,
609 			       io->io_hdr.nexus.targ_port,
610 			       io->io_hdr.nexus.initid.id,
611 			       io->io_hdr.nexus.targ_lun);
612 #endif
613 			io->scsiio.tag_num = msg_info.scsi.tag_num;
614 			io->scsiio.tag_type = msg_info.scsi.tag_type;
615 			memcpy(io->scsiio.cdb, msg_info.scsi.cdb,
616 			       CTL_MAX_CDBLEN);
617 			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
618 				struct ctl_cmd_entry *entry;
619 				uint8_t opcode;
620 
621 				opcode = io->scsiio.cdb[0];
622 				entry = &ctl_cmd_table[opcode];
623 				io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
624 				io->io_hdr.flags |=
625 					entry->flags & CTL_FLAG_DATA_MASK;
626 			}
627 			STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
628 					   &io->io_hdr, links);
629 			ctl_wakeup_thread();
630 			break;
631 
632 		/* Performed on the Originating SC, XFER mode only */
633 		case CTL_MSG_DATAMOVE: {
634 			struct ctl_sg_entry *sgl;
635 			int i, j;
636 
637 			io = msg_info.hdr.original_sc;
638 			if (io == NULL) {
639 				printf("%s: original_sc == NULL!\n", __func__);
640 				/* XXX KDM do something here */
641 				break;
642 			}
643 			io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
644 			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
645 			/*
646 			 * Keep track of this, we need to send it back over
647 			 * when the datamove is complete.
648 			 */
649 			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
650 
651 			if (msg_info.dt.sg_sequence == 0) {
652 				/*
653 				 * XXX KDM we use the preallocated S/G list
654 				 * here, but we'll need to change this to
655 				 * dynamic allocation if we need larger S/G
656 				 * lists.
657 				 */
658 				if (msg_info.dt.kern_sg_entries >
659 				    sizeof(io->io_hdr.remote_sglist) /
660 				    sizeof(io->io_hdr.remote_sglist[0])) {
661 					printf("%s: number of S/G entries "
662 					    "needed %u > allocated num %zd\n",
663 					    __func__,
664 					    msg_info.dt.kern_sg_entries,
665 					    sizeof(io->io_hdr.remote_sglist)/
666 					    sizeof(io->io_hdr.remote_sglist[0]));
667 
668 					/*
669 					 * XXX KDM send a message back to
670 					 * the other side to shut down the
671 					 * DMA.  The error will come back
672 					 * through via the normal channel.
673 					 */
674 					break;
675 				}
676 				sgl = io->io_hdr.remote_sglist;
677 				memset(sgl, 0,
678 				       sizeof(io->io_hdr.remote_sglist));
679 
680 				io->scsiio.kern_data_ptr = (uint8_t *)sgl;
681 
682 				io->scsiio.kern_sg_entries =
683 					msg_info.dt.kern_sg_entries;
684 				io->scsiio.rem_sg_entries =
685 					msg_info.dt.kern_sg_entries;
686 				io->scsiio.kern_data_len =
687 					msg_info.dt.kern_data_len;
688 				io->scsiio.kern_total_len =
689 					msg_info.dt.kern_total_len;
690 				io->scsiio.kern_data_resid =
691 					msg_info.dt.kern_data_resid;
692 				io->scsiio.kern_rel_offset =
693 					msg_info.dt.kern_rel_offset;
694 				/*
695 				 * Clear out per-DMA flags.
696 				 */
697 				io->io_hdr.flags &= ~CTL_FLAG_RDMA_MASK;
698 				/*
699 				 * Add per-DMA flags that are set for this
700 				 * particular DMA request.
701 				 */
702 				io->io_hdr.flags |= msg_info.dt.flags &
703 						    CTL_FLAG_RDMA_MASK;
704 			} else
705 				sgl = (struct ctl_sg_entry *)
706 					io->scsiio.kern_data_ptr;
707 
708 			for (i = msg_info.dt.sent_sg_entries, j = 0;
709 			     i < (msg_info.dt.sent_sg_entries +
710 			     msg_info.dt.cur_sg_entries); i++, j++) {
711 				sgl[i].addr = msg_info.dt.sg_list[j].addr;
712 				sgl[i].len = msg_info.dt.sg_list[j].len;
713 
714 #if 0
715 				printf("%s: L: %p,%d -> %p,%d j=%d, i=%d\n",
716 				       __func__,
717 				       msg_info.dt.sg_list[j].addr,
718 				       msg_info.dt.sg_list[j].len,
719 				       sgl[i].addr, sgl[i].len, j, i);
720 #endif
721 			}
722 #if 0
723 			memcpy(&sgl[msg_info.dt.sent_sg_entries],
724 			       msg_info.dt.sg_list,
725 			       sizeof(*sgl) * msg_info.dt.cur_sg_entries);
726 #endif
727 
728 			/*
729 			 * If this is the last piece of the I/O, we've got
730 			 * the full S/G list.  Queue processing in the thread.
731 			 * Otherwise wait for the next piece.
732 			 */
733 			if (msg_info.dt.sg_last != 0) {
734 				STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
735 						   &io->io_hdr, links);
736 				ctl_wakeup_thread();
737 			}
738 			break;
739 		}
740 		/* Performed on the Serializing (primary) SC, XFER mode only */
741 		case CTL_MSG_DATAMOVE_DONE: {
742 			if (msg_info.hdr.serializing_sc == NULL) {
743 				printf("%s: serializing_sc == NULL!\n",
744 				       __func__);
745 				/* XXX KDM now what? */
746 				break;
747 			}
748 			/*
749 			 * We grab the sense information here in case
750 			 * there was a failure, so we can return status
751 			 * back to the initiator.
752 			 */
753 			io = msg_info.hdr.serializing_sc;
754 			io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
755 			io->io_hdr.status = msg_info.hdr.status;
756 			io->scsiio.scsi_status = msg_info.scsi.scsi_status;
757 			io->scsiio.sense_len = msg_info.scsi.sense_len;
758 			io->scsiio.sense_residual =msg_info.scsi.sense_residual;
759 			io->io_hdr.port_status = msg_info.scsi.fetd_status;
760 			io->scsiio.residual = msg_info.scsi.residual;
761 			memcpy(&io->scsiio.sense_data,&msg_info.scsi.sense_data,
762 			       sizeof(io->scsiio.sense_data));
763 
764 			STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
765 					   &io->io_hdr, links);
766 			ctl_wakeup_thread();
767 			break;
768 		}
769 
770 		/* Preformed on Originating SC, SER_ONLY mode */
771 		case CTL_MSG_R2R:
772 			io = msg_info.hdr.original_sc;
773 			if (io == NULL) {
774 				printf("%s: Major Bummer\n", __func__);
775 				mtx_unlock(&ctl_softc->ctl_lock);
776 				return;
777 			} else {
778 #if 0
779 				printf("pOrig %x\n",(int) ctsio);
780 #endif
781 			}
782 			io->io_hdr.msg_type = CTL_MSG_R2R;
783 			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
784 			STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
785 					   &io->io_hdr, links);
786 			ctl_wakeup_thread();
787 			break;
788 
789 		/*
790 		 * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
791 		 * mode.
792 		 * Performed on the Originating (i.e. secondary) SC in XFER
793 		 * mode
794 		 */
795 		case CTL_MSG_FINISH_IO:
796 			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER)
797 				ctl_isc_handler_finish_xfer(ctl_softc,
798 							    &msg_info);
799 			else
800 				ctl_isc_handler_finish_ser_only(ctl_softc,
801 								&msg_info);
802 			break;
803 
804 		/* Preformed on Originating SC */
805 		case CTL_MSG_BAD_JUJU:
806 			io = msg_info.hdr.original_sc;
807 			if (io == NULL) {
808 				printf("%s: Bad JUJU!, original_sc is NULL!\n",
809 				       __func__);
810 				break;
811 			}
812 			ctl_copy_sense_data(&msg_info, io);
813 			/*
814 			 * IO should have already been cleaned up on other
815 			 * SC so clear this flag so we won't send a message
816 			 * back to finish the IO there.
817 			 */
818 			io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
819 			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
820 
821 			/* io = msg_info.hdr.serializing_sc; */
822 			io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
823 		        STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
824 					   &io->io_hdr, links);
825 			ctl_wakeup_thread();
826 			break;
827 
828 		/* Handle resets sent from the other side */
829 		case CTL_MSG_MANAGE_TASKS: {
830 			struct ctl_taskio *taskio;
831 			taskio = (struct ctl_taskio *)ctl_alloc_io(
832 				(void *)ctl_softc->othersc_pool);
833 			if (taskio == NULL) {
834 				printf("ctl_isc_event_handler: can't allocate "
835 				       "ctl_io!\n");
836 				/* Bad Juju */
837 				/* should I just call the proper reset func
838 				   here??? */
839 				mtx_unlock(&ctl_softc->ctl_lock);
840 				goto bailout;
841 			}
842 			ctl_zero_io((union ctl_io *)taskio);
843 			taskio->io_hdr.io_type = CTL_IO_TASK;
844 			taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
845 			taskio->io_hdr.nexus = msg_info.hdr.nexus;
846 			taskio->task_action = msg_info.task.task_action;
847 			taskio->tag_num = msg_info.task.tag_num;
848 			taskio->tag_type = msg_info.task.tag_type;
849 #ifdef CTL_TIME_IO
850 			taskio->io_hdr.start_time = time_uptime;
851 			getbintime(&taskio->io_hdr.start_bt);
852 #if 0
853 			cs_prof_gettime(&taskio->io_hdr.start_ticks);
854 #endif
855 #endif /* CTL_TIME_IO */
856 		        STAILQ_INSERT_TAIL(&ctl_softc->task_queue,
857 					   &taskio->io_hdr, links);
858 			ctl_softc->flags |= CTL_FLAG_TASK_PENDING;
859 			ctl_wakeup_thread();
860 			break;
861 		}
862 		/* Persistent Reserve action which needs attention */
863 		case CTL_MSG_PERS_ACTION:
864 			presio = (struct ctl_prio *)ctl_alloc_io(
865 				(void *)ctl_softc->othersc_pool);
866 			if (presio == NULL) {
867 				printf("ctl_isc_event_handler: can't allocate "
868 				       "ctl_io!\n");
869 				/* Bad Juju */
870 				/* Need to set busy and send msg back */
871 				mtx_unlock(&ctl_softc->ctl_lock);
872 				goto bailout;
873 			}
874 			ctl_zero_io((union ctl_io *)presio);
875 			presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
876 			presio->pr_msg = msg_info.pr;
877 		        STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
878 					   &presio->io_hdr, links);
879 			ctl_wakeup_thread();
880 			break;
881 		case CTL_MSG_SYNC_FE:
882 			rcv_sync_msg = 1;
883 			break;
884 		case CTL_MSG_APS_LOCK: {
885 			// It's quicker to execute this then to
886 			// queue it.
887 			struct ctl_lun *lun;
888 			struct ctl_page_index *page_index;
889 			struct copan_aps_subpage *current_sp;
890 
891 			lun = ctl_softc->ctl_luns[msg_info.hdr.nexus.targ_lun];
892 			page_index = &lun->mode_pages.index[index_to_aps_page];
893 			current_sp = (struct copan_aps_subpage *)
894 				     (page_index->page_data +
895 				     (page_index->page_len * CTL_PAGE_CURRENT));
896 
897 			current_sp->lock_active = msg_info.aps.lock_flag;
898 		        break;
899 		}
900 		default:
901 		        printf("How did I get here?\n");
902 		}
903 		mtx_unlock(&ctl_softc->ctl_lock);
904 	} else if (event == CTL_HA_EVT_MSG_SENT) {
905 		if (param != CTL_HA_STATUS_SUCCESS) {
906 			printf("Bad status from ctl_ha_msg_send status %d\n",
907 			       param);
908 		}
909 		return;
910 	} else if (event == CTL_HA_EVT_DISCONNECT) {
911 		printf("CTL: Got a disconnect from Isc\n");
912 		return;
913 	} else {
914 		printf("ctl_isc_event_handler: Unknown event %d\n", event);
915 		return;
916 	}
917 
918 bailout:
919 	return;
920 }
921 
922 static void
923 ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
924 {
925 	struct scsi_sense_data *sense;
926 
927 	sense = &dest->scsiio.sense_data;
928 	bcopy(&src->scsi.sense_data, sense, sizeof(*sense));
929 	dest->scsiio.scsi_status = src->scsi.scsi_status;
930 	dest->scsiio.sense_len = src->scsi.sense_len;
931 	dest->io_hdr.status = src->hdr.status;
932 }
933 
934 static void
935 ctl_init(void)
936 {
937 	struct ctl_softc *softc;
938 	struct ctl_io_pool *internal_pool, *emergency_pool, *other_pool;
939 	struct ctl_frontend *fe;
940 	struct ctl_lun *lun;
941         uint8_t sc_id =0;
942 #if 0
943 	int i;
944 #endif
945 	int retval;
946 	//int isc_retval;
947 
948 	retval = 0;
949 	ctl_pause_rtr = 0;
950         rcv_sync_msg = 0;
951 
952 	control_softc = malloc(sizeof(*control_softc), M_DEVBUF, M_WAITOK);
953 	softc = control_softc;
954 
955 	memset(softc, 0, sizeof(*softc));
956 
957 	softc->dev = make_dev(&ctl_cdevsw, 0, UID_ROOT, GID_OPERATOR, 0600,
958 			      "cam/ctl");
959 
960 	softc->dev->si_drv1 = softc;
961 
962 	mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
963 	softc->open_count = 0;
964 
965 	/*
966 	 * Default to actually sending a SYNCHRONIZE CACHE command down to
967 	 * the drive.
968 	 */
969 	softc->flags = CTL_FLAG_REAL_SYNC;
970 
971 	/*
972 	 * In Copan's HA scheme, the "master" and "slave" roles are
973 	 * figured out through the slot the controller is in.  Although it
974 	 * is an active/active system, someone has to be in charge.
975  	 */
976 #ifdef NEEDTOPORT
977         scmicro_rw(SCMICRO_GET_SHELF_ID, &sc_id);
978 #endif
979 
980         if (sc_id == 0) {
981 		softc->flags |= CTL_FLAG_MASTER_SHELF;
982 		persis_offset = 0;
983 	} else
984 		persis_offset = CTL_MAX_INITIATORS;
985 
986 	/*
987 	 * XXX KDM need to figure out where we want to get our target ID
988 	 * and WWID.  Is it different on each port?
989 	 */
990 	softc->target.id = 0;
991 	softc->target.wwid[0] = 0x12345678;
992 	softc->target.wwid[1] = 0x87654321;
993 	STAILQ_INIT(&softc->lun_list);
994 	STAILQ_INIT(&softc->pending_lun_queue);
995 	STAILQ_INIT(&softc->task_queue);
996 	STAILQ_INIT(&softc->incoming_queue);
997 	STAILQ_INIT(&softc->rtr_queue);
998 	STAILQ_INIT(&softc->done_queue);
999 	STAILQ_INIT(&softc->isc_queue);
1000 	STAILQ_INIT(&softc->fe_list);
1001 	STAILQ_INIT(&softc->be_list);
1002 	STAILQ_INIT(&softc->io_pools);
1003 
1004 	lun = &softc->lun;
1005 
1006 	/*
1007 	 * We don't bother calling these with ctl_lock held here, because,
1008 	 * in theory, no one else can try to do anything while we're in our
1009 	 * module init routine.
1010 	 */
1011 	if (ctl_pool_create(softc, CTL_POOL_INTERNAL, CTL_POOL_ENTRIES_INTERNAL,
1012 			    &internal_pool)!= 0){
1013 		printf("ctl: can't allocate %d entry internal pool, "
1014 		       "exiting\n", CTL_POOL_ENTRIES_INTERNAL);
1015 		return;
1016 	}
1017 
1018 	if (ctl_pool_create(softc, CTL_POOL_EMERGENCY,
1019 			    CTL_POOL_ENTRIES_EMERGENCY, &emergency_pool) != 0) {
1020 		printf("ctl: can't allocate %d entry emergency pool, "
1021 		       "exiting\n", CTL_POOL_ENTRIES_EMERGENCY);
1022 		ctl_pool_free(softc, internal_pool);
1023 		return;
1024 	}
1025 
1026 	if (ctl_pool_create(softc, CTL_POOL_4OTHERSC, CTL_POOL_ENTRIES_OTHER_SC,
1027 	                    &other_pool) != 0)
1028 	{
1029 		printf("ctl: can't allocate %d entry other SC pool, "
1030 		       "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1031 		ctl_pool_free(softc, internal_pool);
1032 		ctl_pool_free(softc, emergency_pool);
1033 		return;
1034 	}
1035 
1036 	softc->internal_pool = internal_pool;
1037 	softc->emergency_pool = emergency_pool;
1038 	softc->othersc_pool = other_pool;
1039 
1040 	ctl_pool_acquire(internal_pool);
1041 	ctl_pool_acquire(emergency_pool);
1042 	ctl_pool_acquire(other_pool);
1043 
1044 	/*
1045 	 * We used to allocate a processor LUN here.  The new scheme is to
1046 	 * just let the user allocate LUNs as he sees fit.
1047 	 */
1048 #if 0
1049 	mtx_lock(&softc->ctl_lock);
1050 	ctl_alloc_lun(softc, lun, /*be_lun*/NULL, /*target*/softc->target);
1051 	mtx_unlock(&softc->ctl_lock);
1052 #endif
1053 
1054 	if (kproc_create(ctl_work_thread, softc, &softc->work_thread, 0, 0,
1055 			 "ctl_thrd") != 0) {
1056 		printf("error creating CTL work thread!\n");
1057 		ctl_free_lun(lun);
1058 		ctl_pool_free(softc, internal_pool);
1059 		ctl_pool_free(softc, emergency_pool);
1060 		ctl_pool_free(softc, other_pool);
1061 		return;
1062 	}
1063 	printf("ctl: CAM Target Layer loaded\n");
1064 
1065 	/*
1066 	 * Initialize the initiator and portname mappings
1067 	 */
1068 	memset(softc->wwpn_iid, 0, sizeof(softc->wwpn_iid));
1069 
1070 	/*
1071 	 * Initialize the ioctl front end.
1072 	 */
1073 	fe = &softc->ioctl_info.fe;
1074 	sprintf(softc->ioctl_info.port_name, "CTL ioctl");
1075 	fe->port_type = CTL_PORT_IOCTL;
1076 	fe->num_requested_ctl_io = 100;
1077 	fe->port_name = softc->ioctl_info.port_name;
1078 	fe->port_online = ctl_ioctl_online;
1079 	fe->port_offline = ctl_ioctl_offline;
1080 	fe->onoff_arg = &softc->ioctl_info;
1081 	fe->targ_enable = ctl_ioctl_targ_enable;
1082 	fe->targ_disable = ctl_ioctl_targ_disable;
1083 	fe->lun_enable = ctl_ioctl_lun_enable;
1084 	fe->lun_disable = ctl_ioctl_lun_disable;
1085 	fe->targ_lun_arg = &softc->ioctl_info;
1086 	fe->fe_datamove = ctl_ioctl_datamove;
1087 	fe->fe_done = ctl_ioctl_done;
1088 	fe->max_targets = 15;
1089 	fe->max_target_id = 15;
1090 
1091 	if (ctl_frontend_register(&softc->ioctl_info.fe,
1092 	                  (softc->flags & CTL_FLAG_MASTER_SHELF)) != 0) {
1093 		printf("ctl: ioctl front end registration failed, will "
1094 		       "continue anyway\n");
1095 	}
1096 
1097 #ifdef CTL_IO_DELAY
1098 	if (sizeof(struct callout) > CTL_TIMER_BYTES) {
1099 		printf("sizeof(struct callout) %zd > CTL_TIMER_BYTES %zd\n",
1100 		       sizeof(struct callout), CTL_TIMER_BYTES);
1101 		return;
1102 	}
1103 #endif /* CTL_IO_DELAY */
1104 
1105 }
1106 
1107 void
1108 ctl_shutdown(void)
1109 {
1110 	struct ctl_softc *softc;
1111 	struct ctl_lun *lun, *next_lun;
1112 	struct ctl_io_pool *pool, *next_pool;
1113 
1114 	softc = (struct ctl_softc *)control_softc;
1115 
1116 	if (ctl_frontend_deregister(&softc->ioctl_info.fe) != 0)
1117 		printf("ctl: ioctl front end deregistration failed\n");
1118 
1119 	mtx_lock(&softc->ctl_lock);
1120 
1121 	/*
1122 	 * Free up each LUN.
1123 	 */
1124 	for (lun = STAILQ_FIRST(&softc->lun_list); lun != NULL; lun = next_lun){
1125 		next_lun = STAILQ_NEXT(lun, links);
1126 		ctl_free_lun(lun);
1127 	}
1128 
1129 	/*
1130 	 * This will rip the rug out from under any FETDs or anyone else
1131 	 * that has a pool allocated.  Since we increment our module
1132 	 * refcount any time someone outside the main CTL module allocates
1133 	 * a pool, we shouldn't have any problems here.  The user won't be
1134 	 * able to unload the CTL module until client modules have
1135 	 * successfully unloaded.
1136 	 */
1137 	for (pool = STAILQ_FIRST(&softc->io_pools); pool != NULL;
1138 	     pool = next_pool) {
1139 		next_pool = STAILQ_NEXT(pool, links);
1140 		ctl_pool_free(softc, pool);
1141 	}
1142 
1143 	mtx_unlock(&softc->ctl_lock);
1144 
1145 #if 0
1146 	ctl_shutdown_thread(softc->work_thread);
1147 #endif
1148 
1149 	mtx_destroy(&softc->ctl_lock);
1150 
1151 	destroy_dev(softc->dev);
1152 
1153 	printf("ctl: CAM Target Layer unloaded\n");
1154 }
1155 
1156 /*
1157  * XXX KDM should we do some access checks here?  Bump a reference count to
1158  * prevent a CTL module from being unloaded while someone has it open?
1159  */
1160 static int
1161 ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1162 {
1163 	return (0);
1164 }
1165 
1166 static int
1167 ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1168 {
1169 	return (0);
1170 }
1171 
1172 int
1173 ctl_port_enable(ctl_port_type port_type)
1174 {
1175 	struct ctl_softc *softc;
1176 	struct ctl_frontend *fe;
1177 
1178 	if (ctl_is_single == 0) {
1179 		union ctl_ha_msg msg_info;
1180 		int isc_retval;
1181 
1182 #if 0
1183 		printf("%s: HA mode, synchronizing frontend enable\n",
1184 		        __func__);
1185 #endif
1186 		msg_info.hdr.msg_type = CTL_MSG_SYNC_FE;
1187 	        if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1188 		        sizeof(msg_info), 1 )) > CTL_HA_STATUS_SUCCESS) {
1189 			printf("Sync msg send error retval %d\n", isc_retval);
1190 		}
1191 		if (!rcv_sync_msg) {
1192 			isc_retval=ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
1193 			        sizeof(msg_info), 1);
1194 		}
1195 #if 0
1196         	printf("CTL:Frontend Enable\n");
1197 	} else {
1198 		printf("%s: single mode, skipping frontend synchronization\n",
1199 		        __func__);
1200 #endif
1201 	}
1202 
1203 	softc = control_softc;
1204 
1205 	STAILQ_FOREACH(fe, &softc->fe_list, links) {
1206 		if (port_type & fe->port_type)
1207 		{
1208 #if 0
1209 			printf("port %d\n", fe->targ_port);
1210 #endif
1211 			ctl_frontend_online(fe);
1212 		}
1213 	}
1214 
1215 	return (0);
1216 }
1217 
1218 int
1219 ctl_port_disable(ctl_port_type port_type)
1220 {
1221 	struct ctl_softc *softc;
1222 	struct ctl_frontend *fe;
1223 
1224 	softc = control_softc;
1225 
1226 	STAILQ_FOREACH(fe, &softc->fe_list, links) {
1227 		if (port_type & fe->port_type)
1228 			ctl_frontend_offline(fe);
1229 	}
1230 
1231 	return (0);
1232 }
1233 
1234 /*
1235  * Returns 0 for success, 1 for failure.
1236  * Currently the only failure mode is if there aren't enough entries
1237  * allocated.  So, in case of a failure, look at num_entries_dropped,
1238  * reallocate and try again.
1239  */
1240 int
1241 ctl_port_list(struct ctl_port_entry *entries, int num_entries_alloced,
1242 	      int *num_entries_filled, int *num_entries_dropped,
1243 	      ctl_port_type port_type, int no_virtual)
1244 {
1245 	struct ctl_softc *softc;
1246 	struct ctl_frontend *fe;
1247 	int entries_dropped, entries_filled;
1248 	int retval;
1249 	int i;
1250 
1251 	softc = control_softc;
1252 
1253 	retval = 0;
1254 	entries_filled = 0;
1255 	entries_dropped = 0;
1256 
1257 	i = 0;
1258 	mtx_lock(&softc->ctl_lock);
1259 	STAILQ_FOREACH(fe, &softc->fe_list, links) {
1260 		struct ctl_port_entry *entry;
1261 
1262 		if ((fe->port_type & port_type) == 0)
1263 			continue;
1264 
1265 		if ((no_virtual != 0)
1266 		 && (fe->virtual_port != 0))
1267 			continue;
1268 
1269 		if (entries_filled >= num_entries_alloced) {
1270 			entries_dropped++;
1271 			continue;
1272 		}
1273 		entry = &entries[i];
1274 
1275 		entry->port_type = fe->port_type;
1276 		strlcpy(entry->port_name, fe->port_name,
1277 			sizeof(entry->port_name));
1278 		entry->physical_port = fe->physical_port;
1279 		entry->virtual_port = fe->virtual_port;
1280 		entry->wwnn = fe->wwnn;
1281 		entry->wwpn = fe->wwpn;
1282 
1283 		i++;
1284 		entries_filled++;
1285 	}
1286 
1287 	mtx_unlock(&softc->ctl_lock);
1288 
1289 	if (entries_dropped > 0)
1290 		retval = 1;
1291 
1292 	*num_entries_dropped = entries_dropped;
1293 	*num_entries_filled = entries_filled;
1294 
1295 	return (retval);
1296 }
1297 
1298 static void
1299 ctl_ioctl_online(void *arg)
1300 {
1301 	struct ctl_ioctl_info *ioctl_info;
1302 
1303 	ioctl_info = (struct ctl_ioctl_info *)arg;
1304 
1305 	ioctl_info->flags |= CTL_IOCTL_FLAG_ENABLED;
1306 }
1307 
1308 static void
1309 ctl_ioctl_offline(void *arg)
1310 {
1311 	struct ctl_ioctl_info *ioctl_info;
1312 
1313 	ioctl_info = (struct ctl_ioctl_info *)arg;
1314 
1315 	ioctl_info->flags &= ~CTL_IOCTL_FLAG_ENABLED;
1316 }
1317 
1318 /*
1319  * Remove an initiator by port number and initiator ID.
1320  * Returns 0 for success, 1 for failure.
1321  * Assumes the caller does NOT hold the CTL lock.
1322  */
1323 int
1324 ctl_remove_initiator(int32_t targ_port, uint32_t iid)
1325 {
1326 	struct ctl_softc *softc;
1327 
1328 	softc = control_softc;
1329 
1330 	if ((targ_port < 0)
1331 	 || (targ_port > CTL_MAX_PORTS)) {
1332 		printf("%s: invalid port number %d\n", __func__, targ_port);
1333 		return (1);
1334 	}
1335 	if (iid > CTL_MAX_INIT_PER_PORT) {
1336 		printf("%s: initiator ID %u > maximun %u!\n",
1337 		       __func__, iid, CTL_MAX_INIT_PER_PORT);
1338 		return (1);
1339 	}
1340 
1341 	mtx_lock(&softc->ctl_lock);
1342 
1343 	softc->wwpn_iid[targ_port][iid].in_use = 0;
1344 
1345 	mtx_unlock(&softc->ctl_lock);
1346 
1347 	return (0);
1348 }
1349 
1350 /*
1351  * Add an initiator to the initiator map.
1352  * Returns 0 for success, 1 for failure.
1353  * Assumes the caller does NOT hold the CTL lock.
1354  */
1355 int
1356 ctl_add_initiator(uint64_t wwpn, int32_t targ_port, uint32_t iid)
1357 {
1358 	struct ctl_softc *softc;
1359 	int retval;
1360 
1361 	softc = control_softc;
1362 
1363 	retval = 0;
1364 
1365 	if ((targ_port < 0)
1366 	 || (targ_port > CTL_MAX_PORTS)) {
1367 		printf("%s: invalid port number %d\n", __func__, targ_port);
1368 		return (1);
1369 	}
1370 	if (iid > CTL_MAX_INIT_PER_PORT) {
1371 		printf("%s: WWPN %#jx initiator ID %u > maximun %u!\n",
1372 		       __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
1373 		return (1);
1374 	}
1375 
1376 	mtx_lock(&softc->ctl_lock);
1377 
1378 	if (softc->wwpn_iid[targ_port][iid].in_use != 0) {
1379 		/*
1380 		 * We don't treat this as an error.
1381 		 */
1382 		if (softc->wwpn_iid[targ_port][iid].wwpn == wwpn) {
1383 			printf("%s: port %d iid %u WWPN %#jx arrived again?\n",
1384 			       __func__, targ_port, iid, (uintmax_t)wwpn);
1385 			goto bailout;
1386 		}
1387 
1388 		/*
1389 		 * This is an error, but what do we do about it?  The
1390 		 * driver is telling us we have a new WWPN for this
1391 		 * initiator ID, so we pretty much need to use it.
1392 		 */
1393 		printf("%s: port %d iid %u WWPN %#jx arrived, WWPN %#jx is "
1394 		       "still at that address\n", __func__, targ_port, iid,
1395 		       (uintmax_t)wwpn,
1396 		       (uintmax_t)softc->wwpn_iid[targ_port][iid].wwpn);
1397 
1398 		/*
1399 		 * XXX KDM clear have_ca and ua_pending on each LUN for
1400 		 * this initiator.
1401 		 */
1402 	}
1403 	softc->wwpn_iid[targ_port][iid].in_use = 1;
1404 	softc->wwpn_iid[targ_port][iid].iid = iid;
1405 	softc->wwpn_iid[targ_port][iid].wwpn = wwpn;
1406 	softc->wwpn_iid[targ_port][iid].port = targ_port;
1407 
1408 bailout:
1409 
1410 	mtx_unlock(&softc->ctl_lock);
1411 
1412 	return (retval);
1413 }
1414 
1415 /*
1416  * XXX KDM should we pretend to do something in the target/lun
1417  * enable/disable functions?
1418  */
1419 static int
1420 ctl_ioctl_targ_enable(void *arg, struct ctl_id targ_id)
1421 {
1422 	return (0);
1423 }
1424 
1425 static int
1426 ctl_ioctl_targ_disable(void *arg, struct ctl_id targ_id)
1427 {
1428 	return (0);
1429 }
1430 
1431 static int
1432 ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id)
1433 {
1434 	return (0);
1435 }
1436 
1437 static int
1438 ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id)
1439 {
1440 	return (0);
1441 }
1442 
1443 /*
1444  * Data movement routine for the CTL ioctl frontend port.
1445  */
1446 static int
1447 ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio)
1448 {
1449 	struct ctl_sg_entry *ext_sglist, *kern_sglist;
1450 	struct ctl_sg_entry ext_entry, kern_entry;
1451 	int ext_sglen, ext_sg_entries, kern_sg_entries;
1452 	int ext_sg_start, ext_offset;
1453 	int len_to_copy, len_copied;
1454 	int kern_watermark, ext_watermark;
1455 	int ext_sglist_malloced;
1456 	int i, j;
1457 
1458 	ext_sglist_malloced = 0;
1459 	ext_sg_start = 0;
1460 	ext_offset = 0;
1461 
1462 	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove\n"));
1463 
1464 	/*
1465 	 * If this flag is set, fake the data transfer.
1466 	 */
1467 	if (ctsio->io_hdr.flags & CTL_FLAG_NO_DATAMOVE) {
1468 		ctsio->ext_data_filled = ctsio->ext_data_len;
1469 		goto bailout;
1470 	}
1471 
1472 	/*
1473 	 * To simplify things here, if we have a single buffer, stick it in
1474 	 * a S/G entry and just make it a single entry S/G list.
1475 	 */
1476 	if (ctsio->io_hdr.flags & CTL_FLAG_EDPTR_SGLIST) {
1477 		int len_seen;
1478 
1479 		ext_sglen = ctsio->ext_sg_entries * sizeof(*ext_sglist);
1480 
1481 		ext_sglist = (struct ctl_sg_entry *)malloc(ext_sglen, M_CTL,
1482 							   M_WAITOK);
1483 		if (ext_sglist == NULL) {
1484 			ctl_set_internal_failure(ctsio,
1485 						 /*sks_valid*/ 0,
1486 						 /*retry_count*/ 0);
1487 			return (CTL_RETVAL_COMPLETE);
1488 		}
1489 		ext_sglist_malloced = 1;
1490 		if (copyin(ctsio->ext_data_ptr, ext_sglist,
1491 				   ext_sglen) != 0) {
1492 			ctl_set_internal_failure(ctsio,
1493 						 /*sks_valid*/ 0,
1494 						 /*retry_count*/ 0);
1495 			goto bailout;
1496 		}
1497 		ext_sg_entries = ctsio->ext_sg_entries;
1498 		len_seen = 0;
1499 		for (i = 0; i < ext_sg_entries; i++) {
1500 			if ((len_seen + ext_sglist[i].len) >=
1501 			     ctsio->ext_data_filled) {
1502 				ext_sg_start = i;
1503 				ext_offset = ctsio->ext_data_filled - len_seen;
1504 				break;
1505 			}
1506 			len_seen += ext_sglist[i].len;
1507 		}
1508 	} else {
1509 		ext_sglist = &ext_entry;
1510 		ext_sglist->addr = ctsio->ext_data_ptr;
1511 		ext_sglist->len = ctsio->ext_data_len;
1512 		ext_sg_entries = 1;
1513 		ext_sg_start = 0;
1514 		ext_offset = ctsio->ext_data_filled;
1515 	}
1516 
1517 	if (ctsio->kern_sg_entries > 0) {
1518 		kern_sglist = (struct ctl_sg_entry *)ctsio->kern_data_ptr;
1519 		kern_sg_entries = ctsio->kern_sg_entries;
1520 	} else {
1521 		kern_sglist = &kern_entry;
1522 		kern_sglist->addr = ctsio->kern_data_ptr;
1523 		kern_sglist->len = ctsio->kern_data_len;
1524 		kern_sg_entries = 1;
1525 	}
1526 
1527 
1528 	kern_watermark = 0;
1529 	ext_watermark = ext_offset;
1530 	len_copied = 0;
1531 	for (i = ext_sg_start, j = 0;
1532 	     i < ext_sg_entries && j < kern_sg_entries;) {
1533 		uint8_t *ext_ptr, *kern_ptr;
1534 
1535 		len_to_copy = ctl_min(ext_sglist[i].len - ext_watermark,
1536 				      kern_sglist[j].len - kern_watermark);
1537 
1538 		ext_ptr = (uint8_t *)ext_sglist[i].addr;
1539 		ext_ptr = ext_ptr + ext_watermark;
1540 		if (ctsio->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
1541 			/*
1542 			 * XXX KDM fix this!
1543 			 */
1544 			panic("need to implement bus address support");
1545 #if 0
1546 			kern_ptr = bus_to_virt(kern_sglist[j].addr);
1547 #endif
1548 		} else
1549 			kern_ptr = (uint8_t *)kern_sglist[j].addr;
1550 		kern_ptr = kern_ptr + kern_watermark;
1551 
1552 		kern_watermark += len_to_copy;
1553 		ext_watermark += len_to_copy;
1554 
1555 		if ((ctsio->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
1556 		     CTL_FLAG_DATA_IN) {
1557 			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1558 					 "bytes to user\n", len_to_copy));
1559 			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1560 					 "to %p\n", kern_ptr, ext_ptr));
1561 			if (copyout(kern_ptr, ext_ptr, len_to_copy) != 0) {
1562 				ctl_set_internal_failure(ctsio,
1563 							 /*sks_valid*/ 0,
1564 							 /*retry_count*/ 0);
1565 				goto bailout;
1566 			}
1567 		} else {
1568 			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1569 					 "bytes from user\n", len_to_copy));
1570 			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1571 					 "to %p\n", ext_ptr, kern_ptr));
1572 			if (copyin(ext_ptr, kern_ptr, len_to_copy)!= 0){
1573 				ctl_set_internal_failure(ctsio,
1574 							 /*sks_valid*/ 0,
1575 							 /*retry_count*/0);
1576 				goto bailout;
1577 			}
1578 		}
1579 
1580 		len_copied += len_to_copy;
1581 
1582 		if (ext_sglist[i].len == ext_watermark) {
1583 			i++;
1584 			ext_watermark = 0;
1585 		}
1586 
1587 		if (kern_sglist[j].len == kern_watermark) {
1588 			j++;
1589 			kern_watermark = 0;
1590 		}
1591 	}
1592 
1593 	ctsio->ext_data_filled += len_copied;
1594 
1595 	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_sg_entries: %d, "
1596 			 "kern_sg_entries: %d\n", ext_sg_entries,
1597 			 kern_sg_entries));
1598 	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_data_len = %d, "
1599 			 "kern_data_len = %d\n", ctsio->ext_data_len,
1600 			 ctsio->kern_data_len));
1601 
1602 
1603 	/* XXX KDM set residual?? */
1604 bailout:
1605 
1606 	if (ext_sglist_malloced != 0)
1607 		free(ext_sglist, M_CTL);
1608 
1609 	return (CTL_RETVAL_COMPLETE);
1610 }
1611 
1612 /*
1613  * Serialize a command that went down the "wrong" side, and so was sent to
1614  * this controller for execution.  The logic is a little different than the
1615  * standard case in ctl_scsiio_precheck().  Errors in this case need to get
1616  * sent back to the other side, but in the success case, we execute the
1617  * command on this side (XFER mode) or tell the other side to execute it
1618  * (SER_ONLY mode).
1619  */
1620 static int
1621 ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio, int have_lock)
1622 {
1623 	struct ctl_softc *ctl_softc;
1624 	union ctl_ha_msg msg_info;
1625 	struct ctl_lun *lun;
1626 	int retval = 0;
1627 
1628 	ctl_softc = control_softc;
1629 	if (have_lock == 0)
1630 		mtx_lock(&ctl_softc->ctl_lock);
1631 
1632 	lun = ctl_softc->ctl_luns[ctsio->io_hdr.nexus.targ_lun];
1633 	if (lun==NULL)
1634 	{
1635 		/*
1636 		 * Why isn't LUN defined? The other side wouldn't
1637 		 * send a cmd if the LUN is undefined.
1638 		 */
1639 		printf("%s: Bad JUJU!, LUN is NULL!\n", __func__);
1640 
1641 		/* "Logical unit not supported" */
1642 		ctl_set_sense_data(&msg_info.scsi.sense_data,
1643 				   lun,
1644 				   /*sense_format*/SSD_TYPE_NONE,
1645 				   /*current_error*/ 1,
1646 				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1647 				   /*asc*/ 0x25,
1648 				   /*ascq*/ 0x00,
1649 				   SSD_ELEM_NONE);
1650 
1651 		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1652 		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1653 		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1654 		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1655 		msg_info.hdr.serializing_sc = NULL;
1656 		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1657 	        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1658 				sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1659 		}
1660 		if (have_lock == 0)
1661 			mtx_unlock(&ctl_softc->ctl_lock);
1662 		return(1);
1663 
1664 	}
1665 
1666     	TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1667 
1668 	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
1669 		(union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
1670 		 ooa_links))) {
1671 	case CTL_ACTION_BLOCK:
1672 		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
1673 		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
1674 				  blocked_links);
1675 		break;
1676 	case CTL_ACTION_PASS:
1677 	case CTL_ACTION_SKIP:
1678 		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
1679 			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
1680 			STAILQ_INSERT_TAIL(&ctl_softc->rtr_queue,
1681 					   &ctsio->io_hdr, links);
1682 		} else {
1683 
1684 			/* send msg back to other side */
1685 			msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1686 			msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
1687 			msg_info.hdr.msg_type = CTL_MSG_R2R;
1688 #if 0
1689 			printf("2. pOrig %x\n", (int)msg_info.hdr.original_sc);
1690 #endif
1691 		        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1692 			    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1693 			}
1694 		}
1695 		break;
1696 	case CTL_ACTION_OVERLAP:
1697 		/* OVERLAPPED COMMANDS ATTEMPTED */
1698 		ctl_set_sense_data(&msg_info.scsi.sense_data,
1699 				   lun,
1700 				   /*sense_format*/SSD_TYPE_NONE,
1701 				   /*current_error*/ 1,
1702 				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1703 				   /*asc*/ 0x4E,
1704 				   /*ascq*/ 0x00,
1705 				   SSD_ELEM_NONE);
1706 
1707 		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1708 		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1709 		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1710 		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1711 		msg_info.hdr.serializing_sc = NULL;
1712 		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1713 #if 0
1714 		printf("BAD JUJU:Major Bummer Overlap\n");
1715 #endif
1716 		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1717 		retval = 1;
1718 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1719 		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1720 		}
1721 		break;
1722 	case CTL_ACTION_OVERLAP_TAG:
1723 		/* TAGGED OVERLAPPED COMMANDS (NN = QUEUE TAG) */
1724 		ctl_set_sense_data(&msg_info.scsi.sense_data,
1725 				   lun,
1726 				   /*sense_format*/SSD_TYPE_NONE,
1727 				   /*current_error*/ 1,
1728 				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1729 				   /*asc*/ 0x4D,
1730 				   /*ascq*/ ctsio->tag_num & 0xff,
1731 				   SSD_ELEM_NONE);
1732 
1733 		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1734 		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1735 		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1736 		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1737 		msg_info.hdr.serializing_sc = NULL;
1738 		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1739 #if 0
1740 		printf("BAD JUJU:Major Bummer Overlap Tag\n");
1741 #endif
1742 		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1743 		retval = 1;
1744 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1745 		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1746 		}
1747 		break;
1748 	case CTL_ACTION_ERROR:
1749 	default:
1750 		/* "Internal target failure" */
1751 		ctl_set_sense_data(&msg_info.scsi.sense_data,
1752 				   lun,
1753 				   /*sense_format*/SSD_TYPE_NONE,
1754 				   /*current_error*/ 1,
1755 				   /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
1756 				   /*asc*/ 0x44,
1757 				   /*ascq*/ 0x00,
1758 				   SSD_ELEM_NONE);
1759 
1760 		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1761 		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1762 		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1763 		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1764 		msg_info.hdr.serializing_sc = NULL;
1765 		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1766 #if 0
1767 		printf("BAD JUJU:Major Bummer HW Error\n");
1768 #endif
1769 		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1770 		retval = 1;
1771 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1772 		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1773 		}
1774 		break;
1775 	}
1776 	if (have_lock == 0)
1777 		mtx_unlock(&ctl_softc->ctl_lock);
1778 	return (retval);
1779 }
1780 
1781 static int
1782 ctl_ioctl_submit_wait(union ctl_io *io)
1783 {
1784 	struct ctl_fe_ioctl_params params;
1785 	ctl_fe_ioctl_state last_state;
1786 	int done, retval;
1787 
1788 	retval = 0;
1789 
1790 	bzero(&params, sizeof(params));
1791 
1792 	mtx_init(&params.ioctl_mtx, "ctliocmtx", NULL, MTX_DEF);
1793 	cv_init(&params.sem, "ctlioccv");
1794 	params.state = CTL_IOCTL_INPROG;
1795 	last_state = params.state;
1796 
1797 	io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = &params;
1798 
1799 	CTL_DEBUG_PRINT(("ctl_ioctl_submit_wait\n"));
1800 
1801 	/* This shouldn't happen */
1802 	if ((retval = ctl_queue(io)) != CTL_RETVAL_COMPLETE)
1803 		return (retval);
1804 
1805 	done = 0;
1806 
1807 	do {
1808 		mtx_lock(&params.ioctl_mtx);
1809 		/*
1810 		 * Check the state here, and don't sleep if the state has
1811 		 * already changed (i.e. wakeup has already occured, but we
1812 		 * weren't waiting yet).
1813 		 */
1814 		if (params.state == last_state) {
1815 			/* XXX KDM cv_wait_sig instead? */
1816 			cv_wait(&params.sem, &params.ioctl_mtx);
1817 		}
1818 		last_state = params.state;
1819 
1820 		switch (params.state) {
1821 		case CTL_IOCTL_INPROG:
1822 			/* Why did we wake up? */
1823 			/* XXX KDM error here? */
1824 			mtx_unlock(&params.ioctl_mtx);
1825 			break;
1826 		case CTL_IOCTL_DATAMOVE:
1827 			CTL_DEBUG_PRINT(("got CTL_IOCTL_DATAMOVE\n"));
1828 
1829 			/*
1830 			 * change last_state back to INPROG to avoid
1831 			 * deadlock on subsequent data moves.
1832 			 */
1833 			params.state = last_state = CTL_IOCTL_INPROG;
1834 
1835 			mtx_unlock(&params.ioctl_mtx);
1836 			ctl_ioctl_do_datamove(&io->scsiio);
1837 			/*
1838 			 * Note that in some cases, most notably writes,
1839 			 * this will queue the I/O and call us back later.
1840 			 * In other cases, generally reads, this routine
1841 			 * will immediately call back and wake us up,
1842 			 * probably using our own context.
1843 			 */
1844 			io->scsiio.be_move_done(io);
1845 			break;
1846 		case CTL_IOCTL_DONE:
1847 			mtx_unlock(&params.ioctl_mtx);
1848 			CTL_DEBUG_PRINT(("got CTL_IOCTL_DONE\n"));
1849 			done = 1;
1850 			break;
1851 		default:
1852 			mtx_unlock(&params.ioctl_mtx);
1853 			/* XXX KDM error here? */
1854 			break;
1855 		}
1856 	} while (done == 0);
1857 
1858 	mtx_destroy(&params.ioctl_mtx);
1859 	cv_destroy(&params.sem);
1860 
1861 	return (CTL_RETVAL_COMPLETE);
1862 }
1863 
1864 static void
1865 ctl_ioctl_datamove(union ctl_io *io)
1866 {
1867 	struct ctl_fe_ioctl_params *params;
1868 
1869 	params = (struct ctl_fe_ioctl_params *)
1870 		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
1871 
1872 	mtx_lock(&params->ioctl_mtx);
1873 	params->state = CTL_IOCTL_DATAMOVE;
1874 	cv_broadcast(&params->sem);
1875 	mtx_unlock(&params->ioctl_mtx);
1876 }
1877 
1878 static void
1879 ctl_ioctl_done(union ctl_io *io)
1880 {
1881 	struct ctl_fe_ioctl_params *params;
1882 
1883 	params = (struct ctl_fe_ioctl_params *)
1884 		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
1885 
1886 	mtx_lock(&params->ioctl_mtx);
1887 	params->state = CTL_IOCTL_DONE;
1888 	cv_broadcast(&params->sem);
1889 	mtx_unlock(&params->ioctl_mtx);
1890 }
1891 
1892 static void
1893 ctl_ioctl_hard_startstop_callback(void *arg, struct cfi_metatask *metatask)
1894 {
1895 	struct ctl_fe_ioctl_startstop_info *sd_info;
1896 
1897 	sd_info = (struct ctl_fe_ioctl_startstop_info *)arg;
1898 
1899 	sd_info->hs_info.status = metatask->status;
1900 	sd_info->hs_info.total_luns = metatask->taskinfo.startstop.total_luns;
1901 	sd_info->hs_info.luns_complete =
1902 		metatask->taskinfo.startstop.luns_complete;
1903 	sd_info->hs_info.luns_failed = metatask->taskinfo.startstop.luns_failed;
1904 
1905 	cv_broadcast(&sd_info->sem);
1906 }
1907 
1908 static void
1909 ctl_ioctl_bbrread_callback(void *arg, struct cfi_metatask *metatask)
1910 {
1911 	struct ctl_fe_ioctl_bbrread_info *fe_bbr_info;
1912 
1913 	fe_bbr_info = (struct ctl_fe_ioctl_bbrread_info *)arg;
1914 
1915 	mtx_lock(fe_bbr_info->lock);
1916 	fe_bbr_info->bbr_info->status = metatask->status;
1917 	fe_bbr_info->bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
1918 	fe_bbr_info->wakeup_done = 1;
1919 	mtx_unlock(fe_bbr_info->lock);
1920 
1921 	cv_broadcast(&fe_bbr_info->sem);
1922 }
1923 
1924 /*
1925  * Must be called with the ctl_lock held.
1926  * Returns 0 for success, errno for failure.
1927  */
1928 static int
1929 ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
1930 		   struct ctl_ooa *ooa_hdr)
1931 {
1932 	union ctl_io *io;
1933 	int retval;
1934 
1935 	retval = 0;
1936 
1937 	for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
1938 	     (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
1939 	     ooa_links)) {
1940 		struct ctl_ooa_entry *cur_entry, entry;
1941 
1942 		/*
1943 		 * If we've got more than we can fit, just count the
1944 		 * remaining entries.
1945 		 */
1946 		if (*cur_fill_num >= ooa_hdr->alloc_num)
1947 			continue;
1948 
1949 		cur_entry = &ooa_hdr->entries[*cur_fill_num];
1950 
1951 		bzero(&entry, sizeof(entry));
1952 
1953 		entry.tag_num = io->scsiio.tag_num;
1954 		entry.lun_num = lun->lun;
1955 #ifdef CTL_TIME_IO
1956 		entry.start_bt = io->io_hdr.start_bt;
1957 #endif
1958 		bcopy(io->scsiio.cdb, entry.cdb, io->scsiio.cdb_len);
1959 		entry.cdb_len = io->scsiio.cdb_len;
1960 		if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
1961 			entry.cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
1962 
1963 		if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
1964 			entry.cmd_flags |= CTL_OOACMD_FLAG_DMA;
1965 
1966 		if (io->io_hdr.flags & CTL_FLAG_ABORT)
1967 			entry.cmd_flags |= CTL_OOACMD_FLAG_ABORT;
1968 
1969 		if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
1970 			entry.cmd_flags |= CTL_OOACMD_FLAG_RTR;
1971 
1972 		if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
1973 			entry.cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
1974 
1975 		retval = copyout(&entry, cur_entry, sizeof(entry));
1976 
1977 		if (retval != 0)
1978 			break;
1979 	}
1980 
1981 	return (retval);
1982 }
1983 
1984 static void *
1985 ctl_copyin_alloc(void *user_addr, int len, char *error_str,
1986 		 size_t error_str_len)
1987 {
1988 	void *kptr;
1989 
1990 	kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
1991 	if (kptr == NULL) {
1992 		snprintf(error_str, error_str_len, "Cannot allocate %d bytes",
1993 			 len);
1994 		return (NULL);
1995 	}
1996 
1997 	if (copyin(user_addr, kptr, len) != 0) {
1998 		snprintf(error_str, error_str_len, "Error copying %d bytes "
1999 			 "from user address %p to kernel address %p", len,
2000 			 user_addr, kptr);
2001 		free(kptr, M_CTL);
2002 		return (NULL);
2003 	}
2004 
2005 	return (kptr);
2006 }
2007 
2008 static void
2009 ctl_free_args(int num_be_args, struct ctl_be_arg *be_args)
2010 {
2011 	int i;
2012 
2013 	if (be_args == NULL)
2014 		return;
2015 
2016 	for (i = 0; i < num_be_args; i++) {
2017 		free(be_args[i].kname, M_CTL);
2018 		free(be_args[i].kvalue, M_CTL);
2019 	}
2020 
2021 	free(be_args, M_CTL);
2022 }
2023 
2024 static struct ctl_be_arg *
2025 ctl_copyin_args(int num_be_args, struct ctl_be_arg *be_args,
2026 		char *error_str, size_t error_str_len)
2027 {
2028 	struct ctl_be_arg *args;
2029 	int i;
2030 
2031 	args = ctl_copyin_alloc(be_args, num_be_args * sizeof(*be_args),
2032 				error_str, error_str_len);
2033 
2034 	if (args == NULL)
2035 		goto bailout;
2036 
2037 	for (i = 0; i < num_be_args; i++) {
2038 		uint8_t *tmpptr;
2039 
2040 		args[i].kname = ctl_copyin_alloc(args[i].name,
2041 			args[i].namelen, error_str, error_str_len);
2042 		if (args[i].kname == NULL)
2043 			goto bailout;
2044 
2045 		if (args[i].kname[args[i].namelen - 1] != '\0') {
2046 			snprintf(error_str, error_str_len, "Argument %d "
2047 				 "name is not NUL-terminated", i);
2048 			goto bailout;
2049 		}
2050 
2051 		args[i].kvalue = NULL;
2052 
2053 		tmpptr = ctl_copyin_alloc(args[i].value,
2054 			args[i].vallen, error_str, error_str_len);
2055 		if (tmpptr == NULL)
2056 			goto bailout;
2057 
2058 		args[i].kvalue = tmpptr;
2059 
2060 		if ((args[i].flags & CTL_BEARG_ASCII)
2061 		 && (tmpptr[args[i].vallen - 1] != '\0')) {
2062 			snprintf(error_str, error_str_len, "Argument %d "
2063 				 "value is not NUL-terminated", i);
2064 			goto bailout;
2065 		}
2066 	}
2067 
2068 	return (args);
2069 bailout:
2070 
2071 	ctl_free_args(num_be_args, args);
2072 
2073 	return (NULL);
2074 }
2075 
2076 /*
2077  * Escape characters that are illegal or not recommended in XML.
2078  */
2079 int
2080 ctl_sbuf_printf_esc(struct sbuf *sb, char *str)
2081 {
2082 	int retval;
2083 
2084 	retval = 0;
2085 
2086 	for (; *str; str++) {
2087 		switch (*str) {
2088 		case '&':
2089 			retval = sbuf_printf(sb, "&amp;");
2090 			break;
2091 		case '>':
2092 			retval = sbuf_printf(sb, "&gt;");
2093 			break;
2094 		case '<':
2095 			retval = sbuf_printf(sb, "&lt;");
2096 			break;
2097 		default:
2098 			retval = sbuf_putc(sb, *str);
2099 			break;
2100 		}
2101 
2102 		if (retval != 0)
2103 			break;
2104 
2105 	}
2106 
2107 	return (retval);
2108 }
2109 
2110 static int
2111 ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2112 	  struct thread *td)
2113 {
2114 	struct ctl_softc *softc;
2115 	int retval;
2116 
2117 	softc = control_softc;
2118 
2119 	retval = 0;
2120 
2121 	switch (cmd) {
2122 	case CTL_IO: {
2123 		union ctl_io *io;
2124 		void *pool_tmp;
2125 
2126 		/*
2127 		 * If we haven't been "enabled", don't allow any SCSI I/O
2128 		 * to this FETD.
2129 		 */
2130 		if ((softc->ioctl_info.flags & CTL_IOCTL_FLAG_ENABLED) == 0) {
2131 			retval = -EPERM;
2132 			break;
2133 		}
2134 
2135 		io = ctl_alloc_io(softc->ioctl_info.fe.ctl_pool_ref);
2136 		if (io == NULL) {
2137 			printf("ctl_ioctl: can't allocate ctl_io!\n");
2138 			retval = -ENOSPC;
2139 			break;
2140 		}
2141 
2142 		/*
2143 		 * Need to save the pool reference so it doesn't get
2144 		 * spammed by the user's ctl_io.
2145 		 */
2146 		pool_tmp = io->io_hdr.pool;
2147 
2148 		memcpy(io, (void *)addr, sizeof(*io));
2149 
2150 		io->io_hdr.pool = pool_tmp;
2151 		/*
2152 		 * No status yet, so make sure the status is set properly.
2153 		 */
2154 		io->io_hdr.status = CTL_STATUS_NONE;
2155 
2156 		/*
2157 		 * The user sets the initiator ID, target and LUN IDs.
2158 		 */
2159 		io->io_hdr.nexus.targ_port = softc->ioctl_info.fe.targ_port;
2160 		io->io_hdr.flags |= CTL_FLAG_USER_REQ;
2161 		if ((io->io_hdr.io_type == CTL_IO_SCSI)
2162 		 && (io->scsiio.tag_type != CTL_TAG_UNTAGGED))
2163 			io->scsiio.tag_num = softc->ioctl_info.cur_tag_num++;
2164 
2165 		retval = ctl_ioctl_submit_wait(io);
2166 
2167 		if (retval != 0) {
2168 			ctl_free_io(io);
2169 			break;
2170 		}
2171 
2172 		memcpy((void *)addr, io, sizeof(*io));
2173 
2174 		/* return this to our pool */
2175 		ctl_free_io(io);
2176 
2177 		break;
2178 	}
2179 	case CTL_ENABLE_PORT:
2180 	case CTL_DISABLE_PORT:
2181 	case CTL_SET_PORT_WWNS: {
2182 		struct ctl_frontend *fe;
2183 		struct ctl_port_entry *entry;
2184 
2185 		entry = (struct ctl_port_entry *)addr;
2186 
2187 		mtx_lock(&softc->ctl_lock);
2188 		STAILQ_FOREACH(fe, &softc->fe_list, links) {
2189 			int action, done;
2190 
2191 			action = 0;
2192 			done = 0;
2193 
2194 			if ((entry->port_type == CTL_PORT_NONE)
2195 			 && (entry->targ_port == fe->targ_port)) {
2196 				/*
2197 				 * If the user only wants to enable or
2198 				 * disable or set WWNs on a specific port,
2199 				 * do the operation and we're done.
2200 				 */
2201 				action = 1;
2202 				done = 1;
2203 			} else if (entry->port_type & fe->port_type) {
2204 				/*
2205 				 * Compare the user's type mask with the
2206 				 * particular frontend type to see if we
2207 				 * have a match.
2208 				 */
2209 				action = 1;
2210 				done = 0;
2211 
2212 				/*
2213 				 * Make sure the user isn't trying to set
2214 				 * WWNs on multiple ports at the same time.
2215 				 */
2216 				if (cmd == CTL_SET_PORT_WWNS) {
2217 					printf("%s: Can't set WWNs on "
2218 					       "multiple ports\n", __func__);
2219 					retval = EINVAL;
2220 					break;
2221 				}
2222 			}
2223 			if (action != 0) {
2224 				/*
2225 				 * XXX KDM we have to drop the lock here,
2226 				 * because the online/offline operations
2227 				 * can potentially block.  We need to
2228 				 * reference count the frontends so they
2229 				 * can't go away,
2230 				 */
2231 				mtx_unlock(&softc->ctl_lock);
2232 
2233 				if (cmd == CTL_ENABLE_PORT)
2234 					ctl_frontend_online(fe);
2235 				else if (cmd == CTL_DISABLE_PORT)
2236 					ctl_frontend_offline(fe);
2237 
2238 				mtx_lock(&softc->ctl_lock);
2239 
2240 				if (cmd == CTL_SET_PORT_WWNS)
2241 					ctl_frontend_set_wwns(fe,
2242 					    (entry->flags & CTL_PORT_WWNN_VALID) ?
2243 					    1 : 0, entry->wwnn,
2244 					    (entry->flags & CTL_PORT_WWPN_VALID) ?
2245 					    1 : 0, entry->wwpn);
2246 			}
2247 			if (done != 0)
2248 				break;
2249 		}
2250 		mtx_unlock(&softc->ctl_lock);
2251 		break;
2252 	}
2253 	case CTL_GET_PORT_LIST: {
2254 		struct ctl_frontend *fe;
2255 		struct ctl_port_list *list;
2256 		int i;
2257 
2258 		list = (struct ctl_port_list *)addr;
2259 
2260 		if (list->alloc_len != (list->alloc_num *
2261 		    sizeof(struct ctl_port_entry))) {
2262 			printf("%s: CTL_GET_PORT_LIST: alloc_len %u != "
2263 			       "alloc_num %u * sizeof(struct ctl_port_entry) "
2264 			       "%zu\n", __func__, list->alloc_len,
2265 			       list->alloc_num, sizeof(struct ctl_port_entry));
2266 			retval = EINVAL;
2267 			break;
2268 		}
2269 		list->fill_len = 0;
2270 		list->fill_num = 0;
2271 		list->dropped_num = 0;
2272 		i = 0;
2273 		mtx_lock(&softc->ctl_lock);
2274 		STAILQ_FOREACH(fe, &softc->fe_list, links) {
2275 			struct ctl_port_entry entry, *list_entry;
2276 
2277 			if (list->fill_num >= list->alloc_num) {
2278 				list->dropped_num++;
2279 				continue;
2280 			}
2281 
2282 			entry.port_type = fe->port_type;
2283 			strlcpy(entry.port_name, fe->port_name,
2284 				sizeof(entry.port_name));
2285 			entry.targ_port = fe->targ_port;
2286 			entry.physical_port = fe->physical_port;
2287 			entry.virtual_port = fe->virtual_port;
2288 			entry.wwnn = fe->wwnn;
2289 			entry.wwpn = fe->wwpn;
2290 			if (fe->status & CTL_PORT_STATUS_ONLINE)
2291 				entry.online = 1;
2292 			else
2293 				entry.online = 0;
2294 
2295 			list_entry = &list->entries[i];
2296 
2297 			retval = copyout(&entry, list_entry, sizeof(entry));
2298 			if (retval != 0) {
2299 				printf("%s: CTL_GET_PORT_LIST: copyout "
2300 				       "returned %d\n", __func__, retval);
2301 				break;
2302 			}
2303 			i++;
2304 			list->fill_num++;
2305 			list->fill_len += sizeof(entry);
2306 		}
2307 		mtx_unlock(&softc->ctl_lock);
2308 
2309 		/*
2310 		 * If this is non-zero, we had a copyout fault, so there's
2311 		 * probably no point in attempting to set the status inside
2312 		 * the structure.
2313 		 */
2314 		if (retval != 0)
2315 			break;
2316 
2317 		if (list->dropped_num > 0)
2318 			list->status = CTL_PORT_LIST_NEED_MORE_SPACE;
2319 		else
2320 			list->status = CTL_PORT_LIST_OK;
2321 		break;
2322 	}
2323 	case CTL_DUMP_OOA: {
2324 		struct ctl_lun *lun;
2325 		union ctl_io *io;
2326 		char printbuf[128];
2327 		struct sbuf sb;
2328 
2329 		mtx_lock(&softc->ctl_lock);
2330 		printf("Dumping OOA queues:\n");
2331 		STAILQ_FOREACH(lun, &softc->lun_list, links) {
2332 			for (io = (union ctl_io *)TAILQ_FIRST(
2333 			     &lun->ooa_queue); io != NULL;
2334 			     io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2335 			     ooa_links)) {
2336 				sbuf_new(&sb, printbuf, sizeof(printbuf),
2337 					 SBUF_FIXEDLEN);
2338 				sbuf_printf(&sb, "LUN %jd tag 0x%04x%s%s%s%s: ",
2339 					    (intmax_t)lun->lun,
2340 					    io->scsiio.tag_num,
2341 					    (io->io_hdr.flags &
2342 					    CTL_FLAG_BLOCKED) ? "" : " BLOCKED",
2343 					    (io->io_hdr.flags &
2344 					    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
2345 					    (io->io_hdr.flags &
2346 					    CTL_FLAG_ABORT) ? " ABORT" : "",
2347 			                    (io->io_hdr.flags &
2348 		                        CTL_FLAG_IS_WAS_ON_RTR) ? " RTR" : "");
2349 				ctl_scsi_command_string(&io->scsiio, NULL, &sb);
2350 				sbuf_finish(&sb);
2351 				printf("%s\n", sbuf_data(&sb));
2352 			}
2353 		}
2354 		printf("OOA queues dump done\n");
2355 		mtx_unlock(&softc->ctl_lock);
2356 		break;
2357 	}
2358 	case CTL_GET_OOA: {
2359 		struct ctl_lun *lun;
2360 		struct ctl_ooa *ooa_hdr;
2361 		uint32_t cur_fill_num;
2362 
2363 		ooa_hdr = (struct ctl_ooa *)addr;
2364 
2365 		if ((ooa_hdr->alloc_len == 0)
2366 		 || (ooa_hdr->alloc_num == 0)) {
2367 			printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2368 			       "must be non-zero\n", __func__,
2369 			       ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2370 			retval = EINVAL;
2371 			break;
2372 		}
2373 
2374 		if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2375 		    sizeof(struct ctl_ooa_entry))) {
2376 			printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2377 			       "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2378 			       __func__, ooa_hdr->alloc_len,
2379 			       ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2380 			retval = EINVAL;
2381 			break;
2382 		}
2383 
2384 		mtx_lock(&softc->ctl_lock);
2385 		if (((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0)
2386 		 && ((ooa_hdr->lun_num > CTL_MAX_LUNS)
2387 		  || (softc->ctl_luns[ooa_hdr->lun_num] == NULL))) {
2388 			mtx_unlock(&softc->ctl_lock);
2389 			printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2390 			       __func__, (uintmax_t)ooa_hdr->lun_num);
2391 			retval = EINVAL;
2392 			break;
2393 		}
2394 
2395 		cur_fill_num = 0;
2396 
2397 		if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2398 			STAILQ_FOREACH(lun, &softc->lun_list, links) {
2399 				retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2400 					ooa_hdr);
2401 				if (retval != 0)
2402 					break;
2403 			}
2404 			if (retval != 0) {
2405 				mtx_unlock(&softc->ctl_lock);
2406 				break;
2407 			}
2408 		} else {
2409 			lun = softc->ctl_luns[ooa_hdr->lun_num];
2410 
2411 			retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,ooa_hdr);
2412 		}
2413 		mtx_unlock(&softc->ctl_lock);
2414 
2415 		ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2416 		ooa_hdr->fill_len = ooa_hdr->fill_num *
2417 			sizeof(struct ctl_ooa_entry);
2418 
2419 		getbintime(&ooa_hdr->cur_bt);
2420 
2421 		if (cur_fill_num > ooa_hdr->alloc_num) {
2422 			ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2423 			ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2424 		} else {
2425 			ooa_hdr->dropped_num = 0;
2426 			ooa_hdr->status = CTL_OOA_OK;
2427 		}
2428 		break;
2429 	}
2430 	case CTL_CHECK_OOA: {
2431 		union ctl_io *io;
2432 		struct ctl_lun *lun;
2433 		struct ctl_ooa_info *ooa_info;
2434 
2435 
2436 		ooa_info = (struct ctl_ooa_info *)addr;
2437 
2438 		if (ooa_info->lun_id >= CTL_MAX_LUNS) {
2439 			ooa_info->status = CTL_OOA_INVALID_LUN;
2440 			break;
2441 		}
2442 		mtx_lock(&softc->ctl_lock);
2443 		lun = softc->ctl_luns[ooa_info->lun_id];
2444 		if (lun == NULL) {
2445 			mtx_unlock(&softc->ctl_lock);
2446 			ooa_info->status = CTL_OOA_INVALID_LUN;
2447 			break;
2448 		}
2449 
2450 		ooa_info->num_entries = 0;
2451 		for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
2452 		     io != NULL; io = (union ctl_io *)TAILQ_NEXT(
2453 		     &io->io_hdr, ooa_links)) {
2454 			ooa_info->num_entries++;
2455 		}
2456 
2457 		mtx_unlock(&softc->ctl_lock);
2458 		ooa_info->status = CTL_OOA_SUCCESS;
2459 
2460 		break;
2461 	}
2462 	case CTL_HARD_START:
2463 	case CTL_HARD_STOP: {
2464 		struct ctl_fe_ioctl_startstop_info ss_info;
2465 		struct cfi_metatask *metatask;
2466 		struct mtx hs_mtx;
2467 
2468 		mtx_init(&hs_mtx, "HS Mutex", NULL, MTX_DEF);
2469 
2470 		cv_init(&ss_info.sem, "hard start/stop cv" );
2471 
2472 		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2473 		if (metatask == NULL) {
2474 			retval = ENOMEM;
2475 			mtx_destroy(&hs_mtx);
2476 			break;
2477 		}
2478 
2479 		if (cmd == CTL_HARD_START)
2480 			metatask->tasktype = CFI_TASK_STARTUP;
2481 		else
2482 			metatask->tasktype = CFI_TASK_SHUTDOWN;
2483 
2484 		metatask->callback = ctl_ioctl_hard_startstop_callback;
2485 		metatask->callback_arg = &ss_info;
2486 
2487 		cfi_action(metatask);
2488 
2489 		/* Wait for the callback */
2490 		mtx_lock(&hs_mtx);
2491 		cv_wait_sig(&ss_info.sem, &hs_mtx);
2492 		mtx_unlock(&hs_mtx);
2493 
2494 		/*
2495 		 * All information has been copied from the metatask by the
2496 		 * time cv_broadcast() is called, so we free the metatask here.
2497 		 */
2498 		cfi_free_metatask(metatask);
2499 
2500 		memcpy((void *)addr, &ss_info.hs_info, sizeof(ss_info.hs_info));
2501 
2502 		mtx_destroy(&hs_mtx);
2503 		break;
2504 	}
2505 	case CTL_BBRREAD: {
2506 		struct ctl_bbrread_info *bbr_info;
2507 		struct ctl_fe_ioctl_bbrread_info fe_bbr_info;
2508 		struct mtx bbr_mtx;
2509 		struct cfi_metatask *metatask;
2510 
2511 		bbr_info = (struct ctl_bbrread_info *)addr;
2512 
2513 		bzero(&fe_bbr_info, sizeof(fe_bbr_info));
2514 
2515 		bzero(&bbr_mtx, sizeof(bbr_mtx));
2516 		mtx_init(&bbr_mtx, "BBR Mutex", NULL, MTX_DEF);
2517 
2518 		fe_bbr_info.bbr_info = bbr_info;
2519 		fe_bbr_info.lock = &bbr_mtx;
2520 
2521 		cv_init(&fe_bbr_info.sem, "BBR read cv");
2522 		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2523 
2524 		if (metatask == NULL) {
2525 			mtx_destroy(&bbr_mtx);
2526 			cv_destroy(&fe_bbr_info.sem);
2527 			retval = ENOMEM;
2528 			break;
2529 		}
2530 		metatask->tasktype = CFI_TASK_BBRREAD;
2531 		metatask->callback = ctl_ioctl_bbrread_callback;
2532 		metatask->callback_arg = &fe_bbr_info;
2533 		metatask->taskinfo.bbrread.lun_num = bbr_info->lun_num;
2534 		metatask->taskinfo.bbrread.lba = bbr_info->lba;
2535 		metatask->taskinfo.bbrread.len = bbr_info->len;
2536 
2537 		cfi_action(metatask);
2538 
2539 		mtx_lock(&bbr_mtx);
2540 		while (fe_bbr_info.wakeup_done == 0)
2541 			cv_wait_sig(&fe_bbr_info.sem, &bbr_mtx);
2542 		mtx_unlock(&bbr_mtx);
2543 
2544 		bbr_info->status = metatask->status;
2545 		bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2546 		bbr_info->scsi_status = metatask->taskinfo.bbrread.scsi_status;
2547 		memcpy(&bbr_info->sense_data,
2548 		       &metatask->taskinfo.bbrread.sense_data,
2549 		       ctl_min(sizeof(bbr_info->sense_data),
2550 			       sizeof(metatask->taskinfo.bbrread.sense_data)));
2551 
2552 		cfi_free_metatask(metatask);
2553 
2554 		mtx_destroy(&bbr_mtx);
2555 		cv_destroy(&fe_bbr_info.sem);
2556 
2557 		break;
2558 	}
2559 	case CTL_DELAY_IO: {
2560 		struct ctl_io_delay_info *delay_info;
2561 #ifdef CTL_IO_DELAY
2562 		struct ctl_lun *lun;
2563 #endif /* CTL_IO_DELAY */
2564 
2565 		delay_info = (struct ctl_io_delay_info *)addr;
2566 
2567 #ifdef CTL_IO_DELAY
2568 		mtx_lock(&softc->ctl_lock);
2569 
2570 		if ((delay_info->lun_id > CTL_MAX_LUNS)
2571 		 || (softc->ctl_luns[delay_info->lun_id] == NULL)) {
2572 			delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2573 		} else {
2574 			lun = softc->ctl_luns[delay_info->lun_id];
2575 
2576 			delay_info->status = CTL_DELAY_STATUS_OK;
2577 
2578 			switch (delay_info->delay_type) {
2579 			case CTL_DELAY_TYPE_CONT:
2580 				break;
2581 			case CTL_DELAY_TYPE_ONESHOT:
2582 				break;
2583 			default:
2584 				delay_info->status =
2585 					CTL_DELAY_STATUS_INVALID_TYPE;
2586 				break;
2587 			}
2588 
2589 			switch (delay_info->delay_loc) {
2590 			case CTL_DELAY_LOC_DATAMOVE:
2591 				lun->delay_info.datamove_type =
2592 					delay_info->delay_type;
2593 				lun->delay_info.datamove_delay =
2594 					delay_info->delay_secs;
2595 				break;
2596 			case CTL_DELAY_LOC_DONE:
2597 				lun->delay_info.done_type =
2598 					delay_info->delay_type;
2599 				lun->delay_info.done_delay =
2600 					delay_info->delay_secs;
2601 				break;
2602 			default:
2603 				delay_info->status =
2604 					CTL_DELAY_STATUS_INVALID_LOC;
2605 				break;
2606 			}
2607 		}
2608 
2609 		mtx_unlock(&softc->ctl_lock);
2610 #else
2611 		delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2612 #endif /* CTL_IO_DELAY */
2613 		break;
2614 	}
2615 	case CTL_REALSYNC_SET: {
2616 		int *syncstate;
2617 
2618 		syncstate = (int *)addr;
2619 
2620 		mtx_lock(&softc->ctl_lock);
2621 		switch (*syncstate) {
2622 		case 0:
2623 			softc->flags &= ~CTL_FLAG_REAL_SYNC;
2624 			break;
2625 		case 1:
2626 			softc->flags |= CTL_FLAG_REAL_SYNC;
2627 			break;
2628 		default:
2629 			retval = -EINVAL;
2630 			break;
2631 		}
2632 		mtx_unlock(&softc->ctl_lock);
2633 		break;
2634 	}
2635 	case CTL_REALSYNC_GET: {
2636 		int *syncstate;
2637 
2638 		syncstate = (int*)addr;
2639 
2640 		mtx_lock(&softc->ctl_lock);
2641 		if (softc->flags & CTL_FLAG_REAL_SYNC)
2642 			*syncstate = 1;
2643 		else
2644 			*syncstate = 0;
2645 		mtx_unlock(&softc->ctl_lock);
2646 
2647 		break;
2648 	}
2649 	case CTL_SETSYNC:
2650 	case CTL_GETSYNC: {
2651 		struct ctl_sync_info *sync_info;
2652 		struct ctl_lun *lun;
2653 
2654 		sync_info = (struct ctl_sync_info *)addr;
2655 
2656 		mtx_lock(&softc->ctl_lock);
2657 		lun = softc->ctl_luns[sync_info->lun_id];
2658 		if (lun == NULL) {
2659 			mtx_unlock(&softc->ctl_lock);
2660 			sync_info->status = CTL_GS_SYNC_NO_LUN;
2661 		}
2662 		/*
2663 		 * Get or set the sync interval.  We're not bounds checking
2664 		 * in the set case, hopefully the user won't do something
2665 		 * silly.
2666 		 */
2667 		if (cmd == CTL_GETSYNC)
2668 			sync_info->sync_interval = lun->sync_interval;
2669 		else
2670 			lun->sync_interval = sync_info->sync_interval;
2671 
2672 		mtx_unlock(&softc->ctl_lock);
2673 
2674 		sync_info->status = CTL_GS_SYNC_OK;
2675 
2676 		break;
2677 	}
2678 	case CTL_GETSTATS: {
2679 		struct ctl_stats *stats;
2680 		struct ctl_lun *lun;
2681 		int i;
2682 
2683 		stats = (struct ctl_stats *)addr;
2684 
2685 		if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) >
2686 		     stats->alloc_len) {
2687 			stats->status = CTL_SS_NEED_MORE_SPACE;
2688 			stats->num_luns = softc->num_luns;
2689 			break;
2690 		}
2691 		/*
2692 		 * XXX KDM no locking here.  If the LUN list changes,
2693 		 * things can blow up.
2694 		 */
2695 		for (i = 0, lun = STAILQ_FIRST(&softc->lun_list); lun != NULL;
2696 		     i++, lun = STAILQ_NEXT(lun, links)) {
2697 			retval = copyout(&lun->stats, &stats->lun_stats[i],
2698 					 sizeof(lun->stats));
2699 			if (retval != 0)
2700 				break;
2701 		}
2702 		stats->num_luns = softc->num_luns;
2703 		stats->fill_len = sizeof(struct ctl_lun_io_stats) *
2704 				 softc->num_luns;
2705 		stats->status = CTL_SS_OK;
2706 #ifdef CTL_TIME_IO
2707 		stats->flags = CTL_STATS_FLAG_TIME_VALID;
2708 #else
2709 		stats->flags = CTL_STATS_FLAG_NONE;
2710 #endif
2711 		getnanouptime(&stats->timestamp);
2712 		break;
2713 	}
2714 	case CTL_ERROR_INJECT: {
2715 		struct ctl_error_desc *err_desc, *new_err_desc;
2716 		struct ctl_lun *lun;
2717 
2718 		err_desc = (struct ctl_error_desc *)addr;
2719 
2720 		new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2721 				      M_WAITOK | M_ZERO);
2722 		if (new_err_desc == NULL) {
2723 			printf("%s: CTL_ERROR_INJECT: error allocating %zu "
2724 			       "bytes\n", __func__, sizeof(*new_err_desc));
2725 			retval = ENOMEM;
2726 			break;
2727 		}
2728 		bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2729 
2730 		mtx_lock(&softc->ctl_lock);
2731 		lun = softc->ctl_luns[err_desc->lun_id];
2732 		if (lun == NULL) {
2733 			mtx_unlock(&softc->ctl_lock);
2734 			printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2735 			       __func__, (uintmax_t)err_desc->lun_id);
2736 			retval = EINVAL;
2737 			break;
2738 		}
2739 
2740 		/*
2741 		 * We could do some checking here to verify the validity
2742 		 * of the request, but given the complexity of error
2743 		 * injection requests, the checking logic would be fairly
2744 		 * complex.
2745 		 *
2746 		 * For now, if the request is invalid, it just won't get
2747 		 * executed and might get deleted.
2748 		 */
2749 		STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2750 
2751 		/*
2752 		 * XXX KDM check to make sure the serial number is unique,
2753 		 * in case we somehow manage to wrap.  That shouldn't
2754 		 * happen for a very long time, but it's the right thing to
2755 		 * do.
2756 		 */
2757 		new_err_desc->serial = lun->error_serial;
2758 		err_desc->serial = lun->error_serial;
2759 		lun->error_serial++;
2760 
2761 		mtx_unlock(&softc->ctl_lock);
2762 		break;
2763 	}
2764 	case CTL_ERROR_INJECT_DELETE: {
2765 		struct ctl_error_desc *delete_desc, *desc, *desc2;
2766 		struct ctl_lun *lun;
2767 		int delete_done;
2768 
2769 		delete_desc = (struct ctl_error_desc *)addr;
2770 		delete_done = 0;
2771 
2772 		mtx_lock(&softc->ctl_lock);
2773 		lun = softc->ctl_luns[delete_desc->lun_id];
2774 		if (lun == NULL) {
2775 			mtx_unlock(&softc->ctl_lock);
2776 			printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2777 			       __func__, (uintmax_t)delete_desc->lun_id);
2778 			retval = EINVAL;
2779 			break;
2780 		}
2781 		STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2782 			if (desc->serial != delete_desc->serial)
2783 				continue;
2784 
2785 			STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2786 				      links);
2787 			free(desc, M_CTL);
2788 			delete_done = 1;
2789 		}
2790 		mtx_unlock(&softc->ctl_lock);
2791 		if (delete_done == 0) {
2792 			printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2793 			       "error serial %ju on LUN %u\n", __func__,
2794 			       delete_desc->serial, delete_desc->lun_id);
2795 			retval = EINVAL;
2796 			break;
2797 		}
2798 		break;
2799 	}
2800 	case CTL_DUMP_STRUCTS: {
2801 		int i, j, k;
2802 		struct ctl_frontend *fe;
2803 
2804 		printf("CTL IID to WWPN map start:\n");
2805 		for (i = 0; i < CTL_MAX_PORTS; i++) {
2806 			for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
2807 				if (softc->wwpn_iid[i][j].in_use == 0)
2808 					continue;
2809 
2810 				printf("port %d iid %u WWPN %#jx\n",
2811 				       softc->wwpn_iid[i][j].port,
2812 				       softc->wwpn_iid[i][j].iid,
2813 				       (uintmax_t)softc->wwpn_iid[i][j].wwpn);
2814 			}
2815 		}
2816 		printf("CTL IID to WWPN map end\n");
2817 		printf("CTL Persistent Reservation information start:\n");
2818 		for (i = 0; i < CTL_MAX_LUNS; i++) {
2819 			struct ctl_lun *lun;
2820 
2821 			lun = softc->ctl_luns[i];
2822 
2823 			if ((lun == NULL)
2824 			 || ((lun->flags & CTL_LUN_DISABLED) != 0))
2825 				continue;
2826 
2827 			for (j = 0; j < (CTL_MAX_PORTS * 2); j++) {
2828 				for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2829 					if (lun->per_res[j+k].registered == 0)
2830 						continue;
2831 					printf("LUN %d port %d iid %d key "
2832 					       "%#jx\n", i, j, k,
2833 					       (uintmax_t)scsi_8btou64(
2834 					       lun->per_res[j+k].res_key.key));
2835 				}
2836 			}
2837 		}
2838 		printf("CTL Persistent Reservation information end\n");
2839 		printf("CTL Frontends:\n");
2840 		/*
2841 		 * XXX KDM calling this without a lock.  We'd likely want
2842 		 * to drop the lock before calling the frontend's dump
2843 		 * routine anyway.
2844 		 */
2845 		STAILQ_FOREACH(fe, &softc->fe_list, links) {
2846 			printf("Frontend %s Type %u pport %d vport %d WWNN "
2847 			       "%#jx WWPN %#jx\n", fe->port_name, fe->port_type,
2848 			       fe->physical_port, fe->virtual_port,
2849 			       (uintmax_t)fe->wwnn, (uintmax_t)fe->wwpn);
2850 
2851 			/*
2852 			 * Frontends are not required to support the dump
2853 			 * routine.
2854 			 */
2855 			if (fe->fe_dump == NULL)
2856 				continue;
2857 
2858 			fe->fe_dump();
2859 		}
2860 		printf("CTL Frontend information end\n");
2861 		break;
2862 	}
2863 	case CTL_LUN_REQ: {
2864 		struct ctl_lun_req *lun_req;
2865 		struct ctl_backend_driver *backend;
2866 
2867 		lun_req = (struct ctl_lun_req *)addr;
2868 
2869 		backend = ctl_backend_find(lun_req->backend);
2870 		if (backend == NULL) {
2871 			lun_req->status = CTL_LUN_ERROR;
2872 			snprintf(lun_req->error_str,
2873 				 sizeof(lun_req->error_str),
2874 				 "Backend \"%s\" not found.",
2875 				 lun_req->backend);
2876 			break;
2877 		}
2878 		if (lun_req->num_be_args > 0) {
2879 			lun_req->kern_be_args = ctl_copyin_args(
2880 				lun_req->num_be_args,
2881 				lun_req->be_args,
2882 				lun_req->error_str,
2883 				sizeof(lun_req->error_str));
2884 			if (lun_req->kern_be_args == NULL) {
2885 				lun_req->status = CTL_LUN_ERROR;
2886 				break;
2887 			}
2888 		}
2889 
2890 		retval = backend->ioctl(dev, cmd, addr, flag, td);
2891 
2892 		if (lun_req->num_be_args > 0) {
2893 			ctl_free_args(lun_req->num_be_args,
2894 				      lun_req->kern_be_args);
2895 		}
2896 		break;
2897 	}
2898 	case CTL_LUN_LIST: {
2899 		struct sbuf *sb;
2900 		struct ctl_lun *lun;
2901 		struct ctl_lun_list *list;
2902 
2903 		list = (struct ctl_lun_list *)addr;
2904 
2905 		/*
2906 		 * Allocate a fixed length sbuf here, based on the length
2907 		 * of the user's buffer.  We could allocate an auto-extending
2908 		 * buffer, and then tell the user how much larger our
2909 		 * amount of data is than his buffer, but that presents
2910 		 * some problems:
2911 		 *
2912 		 * 1.  The sbuf(9) routines use a blocking malloc, and so
2913 		 *     we can't hold a lock while calling them with an
2914 		 *     auto-extending buffer.
2915  		 *
2916 		 * 2.  There is not currently a LUN reference counting
2917 		 *     mechanism, outside of outstanding transactions on
2918 		 *     the LUN's OOA queue.  So a LUN could go away on us
2919 		 *     while we're getting the LUN number, backend-specific
2920 		 *     information, etc.  Thus, given the way things
2921 		 *     currently work, we need to hold the CTL lock while
2922 		 *     grabbing LUN information.
2923 		 *
2924 		 * So, from the user's standpoint, the best thing to do is
2925 		 * allocate what he thinks is a reasonable buffer length,
2926 		 * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
2927 		 * double the buffer length and try again.  (And repeat
2928 		 * that until he succeeds.)
2929 		 */
2930 		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
2931 		if (sb == NULL) {
2932 			list->status = CTL_LUN_LIST_ERROR;
2933 			snprintf(list->error_str, sizeof(list->error_str),
2934 				 "Unable to allocate %d bytes for LUN list",
2935 				 list->alloc_len);
2936 			break;
2937 		}
2938 
2939 		sbuf_printf(sb, "<ctllunlist>\n");
2940 
2941 		mtx_lock(&softc->ctl_lock);
2942 
2943 		STAILQ_FOREACH(lun, &softc->lun_list, links) {
2944 			retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
2945 					     (uintmax_t)lun->lun);
2946 
2947 			/*
2948 			 * Bail out as soon as we see that we've overfilled
2949 			 * the buffer.
2950 			 */
2951 			if (retval != 0)
2952 				break;
2953 
2954 			retval = sbuf_printf(sb, "<backend_type>%s"
2955 					     "</backend_type>\n",
2956 					     (lun->backend == NULL) ?  "none" :
2957 					     lun->backend->name);
2958 
2959 			if (retval != 0)
2960 				break;
2961 
2962 			retval = sbuf_printf(sb, "<lun_type>%d</lun_type>\n",
2963 					     lun->be_lun->lun_type);
2964 
2965 			if (retval != 0)
2966 				break;
2967 
2968 			if (lun->backend == NULL) {
2969 				retval = sbuf_printf(sb, "</lun>\n");
2970 				if (retval != 0)
2971 					break;
2972 				continue;
2973 			}
2974 
2975 			retval = sbuf_printf(sb, "<size>%ju</size>\n",
2976 					     (lun->be_lun->maxlba > 0) ?
2977 					     lun->be_lun->maxlba + 1 : 0);
2978 
2979 			if (retval != 0)
2980 				break;
2981 
2982 			retval = sbuf_printf(sb, "<blocksize>%u</blocksize>\n",
2983 					     lun->be_lun->blocksize);
2984 
2985 			if (retval != 0)
2986 				break;
2987 
2988 			retval = sbuf_printf(sb, "<serial_number>");
2989 
2990 			if (retval != 0)
2991 				break;
2992 
2993 			retval = ctl_sbuf_printf_esc(sb,
2994 						     lun->be_lun->serial_num);
2995 
2996 			if (retval != 0)
2997 				break;
2998 
2999 			retval = sbuf_printf(sb, "</serial_number>\n");
3000 
3001 			if (retval != 0)
3002 				break;
3003 
3004 			retval = sbuf_printf(sb, "<device_id>");
3005 
3006 			if (retval != 0)
3007 				break;
3008 
3009 			retval = ctl_sbuf_printf_esc(sb,lun->be_lun->device_id);
3010 
3011 			if (retval != 0)
3012 				break;
3013 
3014 			retval = sbuf_printf(sb, "</device_id>\n");
3015 
3016 			if (retval != 0)
3017 				break;
3018 
3019 			if (lun->backend->lun_info == NULL) {
3020 				retval = sbuf_printf(sb, "</lun>\n");
3021 				if (retval != 0)
3022 					break;
3023 				continue;
3024 			}
3025 
3026 			retval =lun->backend->lun_info(lun->be_lun->be_lun, sb);
3027 
3028 			if (retval != 0)
3029 				break;
3030 
3031 			retval = sbuf_printf(sb, "</lun>\n");
3032 
3033 			if (retval != 0)
3034 				break;
3035 		}
3036 		mtx_unlock(&softc->ctl_lock);
3037 
3038 		if ((retval != 0)
3039 		 || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3040 			retval = 0;
3041 			sbuf_delete(sb);
3042 			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3043 			snprintf(list->error_str, sizeof(list->error_str),
3044 				 "Out of space, %d bytes is too small",
3045 				 list->alloc_len);
3046 			break;
3047 		}
3048 
3049 		sbuf_finish(sb);
3050 
3051 		retval = copyout(sbuf_data(sb), list->lun_xml,
3052 				 sbuf_len(sb) + 1);
3053 
3054 		list->fill_len = sbuf_len(sb) + 1;
3055 		list->status = CTL_LUN_LIST_OK;
3056 		sbuf_delete(sb);
3057 		break;
3058 	}
3059 	default: {
3060 		/* XXX KDM should we fix this? */
3061 #if 0
3062 		struct ctl_backend_driver *backend;
3063 		unsigned int type;
3064 		int found;
3065 
3066 		found = 0;
3067 
3068 		/*
3069 		 * We encode the backend type as the ioctl type for backend
3070 		 * ioctls.  So parse it out here, and then search for a
3071 		 * backend of this type.
3072 		 */
3073 		type = _IOC_TYPE(cmd);
3074 
3075 		STAILQ_FOREACH(backend, &softc->be_list, links) {
3076 			if (backend->type == type) {
3077 				found = 1;
3078 				break;
3079 			}
3080 		}
3081 		if (found == 0) {
3082 			printf("ctl: unknown ioctl command %#lx or backend "
3083 			       "%d\n", cmd, type);
3084 			retval = -EINVAL;
3085 			break;
3086 		}
3087 		retval = backend->ioctl(dev, cmd, addr, flag, td);
3088 #endif
3089 		retval = ENOTTY;
3090 		break;
3091 	}
3092 	}
3093 	return (retval);
3094 }
3095 
3096 uint32_t
3097 ctl_get_initindex(struct ctl_nexus *nexus)
3098 {
3099 	if (nexus->targ_port < CTL_MAX_PORTS)
3100 		return (nexus->initid.id +
3101 			(nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3102 	else
3103 		return (nexus->initid.id +
3104 		       ((nexus->targ_port - CTL_MAX_PORTS) *
3105 			CTL_MAX_INIT_PER_PORT));
3106 }
3107 
3108 uint32_t
3109 ctl_get_resindex(struct ctl_nexus *nexus)
3110 {
3111 	return (nexus->initid.id + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3112 }
3113 
3114 uint32_t
3115 ctl_port_idx(int port_num)
3116 {
3117 	if (port_num < CTL_MAX_PORTS)
3118 		return(port_num);
3119 	else
3120 		return(port_num - CTL_MAX_PORTS);
3121 }
3122 
3123 /*
3124  * Note:  This only works for bitmask sizes that are at least 32 bits, and
3125  * that are a power of 2.
3126  */
3127 int
3128 ctl_ffz(uint32_t *mask, uint32_t size)
3129 {
3130 	uint32_t num_chunks, num_pieces;
3131 	int i, j;
3132 
3133 	num_chunks = (size >> 5);
3134 	if (num_chunks == 0)
3135 		num_chunks++;
3136 	num_pieces = ctl_min((sizeof(uint32_t) * 8), size);
3137 
3138 	for (i = 0; i < num_chunks; i++) {
3139 		for (j = 0; j < num_pieces; j++) {
3140 			if ((mask[i] & (1 << j)) == 0)
3141 				return ((i << 5) + j);
3142 		}
3143 	}
3144 
3145 	return (-1);
3146 }
3147 
3148 int
3149 ctl_set_mask(uint32_t *mask, uint32_t bit)
3150 {
3151 	uint32_t chunk, piece;
3152 
3153 	chunk = bit >> 5;
3154 	piece = bit % (sizeof(uint32_t) * 8);
3155 
3156 	if ((mask[chunk] & (1 << piece)) != 0)
3157 		return (-1);
3158 	else
3159 		mask[chunk] |= (1 << piece);
3160 
3161 	return (0);
3162 }
3163 
3164 int
3165 ctl_clear_mask(uint32_t *mask, uint32_t bit)
3166 {
3167 	uint32_t chunk, piece;
3168 
3169 	chunk = bit >> 5;
3170 	piece = bit % (sizeof(uint32_t) * 8);
3171 
3172 	if ((mask[chunk] & (1 << piece)) == 0)
3173 		return (-1);
3174 	else
3175 		mask[chunk] &= ~(1 << piece);
3176 
3177 	return (0);
3178 }
3179 
3180 int
3181 ctl_is_set(uint32_t *mask, uint32_t bit)
3182 {
3183 	uint32_t chunk, piece;
3184 
3185 	chunk = bit >> 5;
3186 	piece = bit % (sizeof(uint32_t) * 8);
3187 
3188 	if ((mask[chunk] & (1 << piece)) == 0)
3189 		return (0);
3190 	else
3191 		return (1);
3192 }
3193 
3194 #ifdef unused
3195 /*
3196  * The bus, target and lun are optional, they can be filled in later.
3197  * can_wait is used to determine whether we can wait on the malloc or not.
3198  */
3199 union ctl_io*
3200 ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port, uint32_t targ_target,
3201 	      uint32_t targ_lun, int can_wait)
3202 {
3203 	union ctl_io *io;
3204 
3205 	if (can_wait)
3206 		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_WAITOK);
3207 	else
3208 		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_NOWAIT);
3209 
3210 	if (io != NULL) {
3211 		io->io_hdr.io_type = io_type;
3212 		io->io_hdr.targ_port = targ_port;
3213 		/*
3214 		 * XXX KDM this needs to change/go away.  We need to move
3215 		 * to a preallocated pool of ctl_scsiio structures.
3216 		 */
3217 		io->io_hdr.nexus.targ_target.id = targ_target;
3218 		io->io_hdr.nexus.targ_lun = targ_lun;
3219 	}
3220 
3221 	return (io);
3222 }
3223 
3224 void
3225 ctl_kfree_io(union ctl_io *io)
3226 {
3227 	free(io, M_CTL);
3228 }
3229 #endif /* unused */
3230 
3231 /*
3232  * ctl_softc, pool_type, total_ctl_io are passed in.
3233  * npool is passed out.
3234  */
3235 int
3236 ctl_pool_create(struct ctl_softc *ctl_softc, ctl_pool_type pool_type,
3237 		uint32_t total_ctl_io, struct ctl_io_pool **npool)
3238 {
3239 	uint32_t i;
3240 	union ctl_io *cur_io, *next_io;
3241 	struct ctl_io_pool *pool;
3242 	int retval;
3243 
3244 	retval = 0;
3245 
3246 	pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL, M_NOWAIT);
3247 	if (pool == NULL) {
3248 		retval = -ENOMEM;
3249 		goto bailout;
3250 	}
3251 
3252 	memset(pool, 0, sizeof(*pool));
3253 
3254 	pool->type = pool_type;
3255 	pool->ctl_softc = ctl_softc;
3256 
3257 	mtx_lock(&ctl_softc->ctl_lock);
3258 	pool->id = ctl_softc->cur_pool_id++;
3259 	mtx_unlock(&ctl_softc->ctl_lock);
3260 
3261 	pool->flags = CTL_POOL_FLAG_NONE;
3262 	STAILQ_INIT(&pool->free_queue);
3263 
3264 	/*
3265 	 * XXX KDM other options here:
3266 	 * - allocate a page at a time
3267 	 * - allocate one big chunk of memory.
3268 	 * Page allocation might work well, but would take a little more
3269 	 * tracking.
3270 	 */
3271 	for (i = 0; i < total_ctl_io; i++) {
3272 		cur_io = (union ctl_io *)malloc(sizeof(*cur_io), M_CTL,
3273 						M_NOWAIT);
3274 		if (cur_io == NULL) {
3275 			retval = ENOMEM;
3276 			break;
3277 		}
3278 		cur_io->io_hdr.pool = pool;
3279 		STAILQ_INSERT_TAIL(&pool->free_queue, &cur_io->io_hdr, links);
3280 		pool->total_ctl_io++;
3281 		pool->free_ctl_io++;
3282 	}
3283 
3284 	if (retval != 0) {
3285 		for (cur_io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3286 		     cur_io != NULL; cur_io = next_io) {
3287 			next_io = (union ctl_io *)STAILQ_NEXT(&cur_io->io_hdr,
3288 							      links);
3289 			STAILQ_REMOVE(&pool->free_queue, &cur_io->io_hdr,
3290 				      ctl_io_hdr, links);
3291 			free(cur_io, M_CTL);
3292 		}
3293 
3294 		free(pool, M_CTL);
3295 		goto bailout;
3296 	}
3297 	mtx_lock(&ctl_softc->ctl_lock);
3298 	ctl_softc->num_pools++;
3299 	STAILQ_INSERT_TAIL(&ctl_softc->io_pools, pool, links);
3300 	/*
3301 	 * Increment our usage count if this is an external consumer, so we
3302 	 * can't get unloaded until the external consumer (most likely a
3303 	 * FETD) unloads and frees his pool.
3304 	 *
3305 	 * XXX KDM will this increment the caller's module use count, or
3306 	 * mine?
3307 	 */
3308 #if 0
3309 	if ((pool_type != CTL_POOL_EMERGENCY)
3310 	 && (pool_type != CTL_POOL_INTERNAL)
3311 	 && (pool_type != CTL_POOL_IOCTL)
3312 	 && (pool_type != CTL_POOL_4OTHERSC))
3313 		MOD_INC_USE_COUNT;
3314 #endif
3315 
3316 	mtx_unlock(&ctl_softc->ctl_lock);
3317 
3318 	*npool = pool;
3319 
3320 bailout:
3321 
3322 	return (retval);
3323 }
3324 
3325 /*
3326  * Caller must hold ctl_softc->ctl_lock.
3327  */
3328 int
3329 ctl_pool_acquire(struct ctl_io_pool *pool)
3330 {
3331 	if (pool == NULL)
3332 		return (-EINVAL);
3333 
3334 	if (pool->flags & CTL_POOL_FLAG_INVALID)
3335 		return (-EINVAL);
3336 
3337 	pool->refcount++;
3338 
3339 	return (0);
3340 }
3341 
3342 /*
3343  * Caller must hold ctl_softc->ctl_lock.
3344  */
3345 int
3346 ctl_pool_invalidate(struct ctl_io_pool *pool)
3347 {
3348 	if (pool == NULL)
3349 		return (-EINVAL);
3350 
3351 	pool->flags |= CTL_POOL_FLAG_INVALID;
3352 
3353 	return (0);
3354 }
3355 
3356 /*
3357  * Caller must hold ctl_softc->ctl_lock.
3358  */
3359 int
3360 ctl_pool_release(struct ctl_io_pool *pool)
3361 {
3362 	if (pool == NULL)
3363 		return (-EINVAL);
3364 
3365 	if ((--pool->refcount == 0)
3366 	 && (pool->flags & CTL_POOL_FLAG_INVALID)) {
3367 		ctl_pool_free(pool->ctl_softc, pool);
3368 	}
3369 
3370 	return (0);
3371 }
3372 
3373 /*
3374  * Must be called with ctl_softc->ctl_lock held.
3375  */
3376 void
3377 ctl_pool_free(struct ctl_softc *ctl_softc, struct ctl_io_pool *pool)
3378 {
3379 	union ctl_io *cur_io, *next_io;
3380 
3381 	for (cur_io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3382 	     cur_io != NULL; cur_io = next_io) {
3383 		next_io = (union ctl_io *)STAILQ_NEXT(&cur_io->io_hdr,
3384 						      links);
3385 		STAILQ_REMOVE(&pool->free_queue, &cur_io->io_hdr, ctl_io_hdr,
3386 			      links);
3387 		free(cur_io, M_CTL);
3388 	}
3389 
3390 	STAILQ_REMOVE(&ctl_softc->io_pools, pool, ctl_io_pool, links);
3391 	ctl_softc->num_pools--;
3392 
3393 	/*
3394 	 * XXX KDM will this decrement the caller's usage count or mine?
3395 	 */
3396 #if 0
3397 	if ((pool->type != CTL_POOL_EMERGENCY)
3398 	 && (pool->type != CTL_POOL_INTERNAL)
3399 	 && (pool->type != CTL_POOL_IOCTL))
3400 		MOD_DEC_USE_COUNT;
3401 #endif
3402 
3403 	free(pool, M_CTL);
3404 }
3405 
3406 /*
3407  * This routine does not block (except for spinlocks of course).
3408  * It tries to allocate a ctl_io union from the caller's pool as quickly as
3409  * possible.
3410  */
3411 union ctl_io *
3412 ctl_alloc_io(void *pool_ref)
3413 {
3414 	union ctl_io *io;
3415 	struct ctl_softc *ctl_softc;
3416 	struct ctl_io_pool *pool, *npool;
3417 	struct ctl_io_pool *emergency_pool;
3418 
3419 	pool = (struct ctl_io_pool *)pool_ref;
3420 
3421 	if (pool == NULL) {
3422 		printf("%s: pool is NULL\n", __func__);
3423 		return (NULL);
3424 	}
3425 
3426 	emergency_pool = NULL;
3427 
3428 	ctl_softc = pool->ctl_softc;
3429 
3430 	mtx_lock(&ctl_softc->ctl_lock);
3431 	/*
3432 	 * First, try to get the io structure from the user's pool.
3433 	 */
3434 	if (ctl_pool_acquire(pool) == 0) {
3435 		io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3436 		if (io != NULL) {
3437 			STAILQ_REMOVE_HEAD(&pool->free_queue, links);
3438 			pool->total_allocated++;
3439 			pool->free_ctl_io--;
3440 			mtx_unlock(&ctl_softc->ctl_lock);
3441 			return (io);
3442 		} else
3443 			ctl_pool_release(pool);
3444 	}
3445 	/*
3446 	 * If he doesn't have any io structures left, search for an
3447 	 * emergency pool and grab one from there.
3448 	 */
3449 	STAILQ_FOREACH(npool, &ctl_softc->io_pools, links) {
3450 		if (npool->type != CTL_POOL_EMERGENCY)
3451 			continue;
3452 
3453 		if (ctl_pool_acquire(npool) != 0)
3454 			continue;
3455 
3456 		emergency_pool = npool;
3457 
3458 		io = (union ctl_io *)STAILQ_FIRST(&npool->free_queue);
3459 		if (io != NULL) {
3460 			STAILQ_REMOVE_HEAD(&npool->free_queue, links);
3461 			npool->total_allocated++;
3462 			npool->free_ctl_io--;
3463 			mtx_unlock(&ctl_softc->ctl_lock);
3464 			return (io);
3465 		} else
3466 			ctl_pool_release(npool);
3467 	}
3468 
3469 	/* Drop the spinlock before we malloc */
3470 	mtx_unlock(&ctl_softc->ctl_lock);
3471 
3472 	/*
3473 	 * The emergency pool (if it exists) didn't have one, so try an
3474 	 * atomic (i.e. nonblocking) malloc and see if we get lucky.
3475 	 */
3476 	io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_NOWAIT);
3477 	if (io != NULL) {
3478 		/*
3479 		 * If the emergency pool exists but is empty, add this
3480 		 * ctl_io to its list when it gets freed.
3481 		 */
3482 		if (emergency_pool != NULL) {
3483 			mtx_lock(&ctl_softc->ctl_lock);
3484 			if (ctl_pool_acquire(emergency_pool) == 0) {
3485 				io->io_hdr.pool = emergency_pool;
3486 				emergency_pool->total_ctl_io++;
3487 				/*
3488 				 * Need to bump this, otherwise
3489 				 * total_allocated and total_freed won't
3490 				 * match when we no longer have anything
3491 				 * outstanding.
3492 				 */
3493 				emergency_pool->total_allocated++;
3494 			}
3495 			mtx_unlock(&ctl_softc->ctl_lock);
3496 		} else
3497 			io->io_hdr.pool = NULL;
3498 	}
3499 
3500 	return (io);
3501 }
3502 
3503 static void
3504 ctl_free_io_internal(union ctl_io *io, int have_lock)
3505 {
3506 	if (io == NULL)
3507 		return;
3508 
3509 	/*
3510 	 * If this ctl_io has a pool, return it to that pool.
3511 	 */
3512 	if (io->io_hdr.pool != NULL) {
3513 		struct ctl_io_pool *pool;
3514 #if 0
3515 		struct ctl_softc *ctl_softc;
3516 		union ctl_io *tmp_io;
3517 		unsigned long xflags;
3518 		int i;
3519 
3520 		ctl_softc = control_softc;
3521 #endif
3522 
3523 		pool = (struct ctl_io_pool *)io->io_hdr.pool;
3524 
3525 		if (have_lock == 0)
3526 			mtx_lock(&pool->ctl_softc->ctl_lock);
3527 #if 0
3528 		save_flags(xflags);
3529 
3530 		for (i = 0, tmp_io = (union ctl_io *)STAILQ_FIRST(
3531 		     &ctl_softc->task_queue); tmp_io != NULL; i++,
3532 		     tmp_io = (union ctl_io *)STAILQ_NEXT(&tmp_io->io_hdr,
3533 		     links)) {
3534 			if (tmp_io == io) {
3535 				printf("%s: %p is still on the task queue!\n",
3536 				       __func__, tmp_io);
3537 				printf("%s: (%d): type %d "
3538 				       "msg %d cdb %x iptl: "
3539 				       "%d:%d:%d:%d tag 0x%04x "
3540 				       "flg %#lx\n",
3541 					__func__, i,
3542 					tmp_io->io_hdr.io_type,
3543 					tmp_io->io_hdr.msg_type,
3544 					tmp_io->scsiio.cdb[0],
3545 					tmp_io->io_hdr.nexus.initid.id,
3546 					tmp_io->io_hdr.nexus.targ_port,
3547 					tmp_io->io_hdr.nexus.targ_target.id,
3548 					tmp_io->io_hdr.nexus.targ_lun,
3549 					(tmp_io->io_hdr.io_type ==
3550 					CTL_IO_TASK) ?
3551 					tmp_io->taskio.tag_num :
3552 					tmp_io->scsiio.tag_num,
3553 					xflags);
3554 				panic("I/O still on the task queue!");
3555 			}
3556 		}
3557 #endif
3558 		io->io_hdr.io_type = 0xff;
3559 		STAILQ_INSERT_TAIL(&pool->free_queue, &io->io_hdr, links);
3560 		pool->total_freed++;
3561 		pool->free_ctl_io++;
3562 		ctl_pool_release(pool);
3563 		if (have_lock == 0)
3564 			mtx_unlock(&pool->ctl_softc->ctl_lock);
3565 	} else {
3566 		/*
3567 		 * Otherwise, just free it.  We probably malloced it and
3568 		 * the emergency pool wasn't available.
3569 		 */
3570 		free(io, M_CTL);
3571 	}
3572 
3573 }
3574 
3575 void
3576 ctl_free_io(union ctl_io *io)
3577 {
3578 	ctl_free_io_internal(io, /*have_lock*/ 0);
3579 }
3580 
3581 void
3582 ctl_zero_io(union ctl_io *io)
3583 {
3584 	void *pool_ref;
3585 
3586 	if (io == NULL)
3587 		return;
3588 
3589 	/*
3590 	 * May need to preserve linked list pointers at some point too.
3591 	 */
3592 	pool_ref = io->io_hdr.pool;
3593 
3594 	memset(io, 0, sizeof(*io));
3595 
3596 	io->io_hdr.pool = pool_ref;
3597 }
3598 
3599 /*
3600  * This routine is currently used for internal copies of ctl_ios that need
3601  * to persist for some reason after we've already returned status to the
3602  * FETD.  (Thus the flag set.)
3603  *
3604  * XXX XXX
3605  * Note that this makes a blind copy of all fields in the ctl_io, except
3606  * for the pool reference.  This includes any memory that has been
3607  * allocated!  That memory will no longer be valid after done has been
3608  * called, so this would be VERY DANGEROUS for command that actually does
3609  * any reads or writes.  Right now (11/7/2005), this is only used for immediate
3610  * start and stop commands, which don't transfer any data, so this is not a
3611  * problem.  If it is used for anything else, the caller would also need to
3612  * allocate data buffer space and this routine would need to be modified to
3613  * copy the data buffer(s) as well.
3614  */
3615 void
3616 ctl_copy_io(union ctl_io *src, union ctl_io *dest)
3617 {
3618 	void *pool_ref;
3619 
3620 	if ((src == NULL)
3621 	 || (dest == NULL))
3622 		return;
3623 
3624 	/*
3625 	 * May need to preserve linked list pointers at some point too.
3626 	 */
3627 	pool_ref = dest->io_hdr.pool;
3628 
3629 	memcpy(dest, src, ctl_min(sizeof(*src), sizeof(*dest)));
3630 
3631 	dest->io_hdr.pool = pool_ref;
3632 	/*
3633 	 * We need to know that this is an internal copy, and doesn't need
3634 	 * to get passed back to the FETD that allocated it.
3635 	 */
3636 	dest->io_hdr.flags |= CTL_FLAG_INT_COPY;
3637 }
3638 
3639 #ifdef NEEDTOPORT
3640 static void
3641 ctl_update_power_subpage(struct copan_power_subpage *page)
3642 {
3643 	int num_luns, num_partitions, config_type;
3644 	struct ctl_softc *softc;
3645 	cs_BOOL_t aor_present, shelf_50pct_power;
3646 	cs_raidset_personality_t rs_type;
3647 	int max_active_luns;
3648 
3649 	softc = control_softc;
3650 
3651 	/* subtract out the processor LUN */
3652 	num_luns = softc->num_luns - 1;
3653 	/*
3654 	 * Default to 7 LUNs active, which was the only number we allowed
3655 	 * in the past.
3656 	 */
3657 	max_active_luns = 7;
3658 
3659 	num_partitions = config_GetRsPartitionInfo();
3660 	config_type = config_GetConfigType();
3661 	shelf_50pct_power = config_GetShelfPowerMode();
3662 	aor_present = config_IsAorRsPresent();
3663 
3664 	rs_type = ddb_GetRsRaidType(1);
3665 	if ((rs_type != CS_RAIDSET_PERSONALITY_RAID5)
3666 	 && (rs_type != CS_RAIDSET_PERSONALITY_RAID1)) {
3667 		EPRINT(0, "Unsupported RS type %d!", rs_type);
3668 	}
3669 
3670 
3671 	page->total_luns = num_luns;
3672 
3673 	switch (config_type) {
3674 	case 40:
3675 		/*
3676 		 * In a 40 drive configuration, it doesn't matter what DC
3677 		 * cards we have, whether we have AOR enabled or not,
3678 		 * partitioning or not, or what type of RAIDset we have.
3679 		 * In that scenario, we can power up every LUN we present
3680 		 * to the user.
3681 		 */
3682 		max_active_luns = num_luns;
3683 
3684 		break;
3685 	case 64:
3686 		if (shelf_50pct_power == CS_FALSE) {
3687 			/* 25% power */
3688 			if (aor_present == CS_TRUE) {
3689 				if (rs_type ==
3690 				     CS_RAIDSET_PERSONALITY_RAID5) {
3691 					max_active_luns = 7;
3692 				} else if (rs_type ==
3693 					 CS_RAIDSET_PERSONALITY_RAID1){
3694 					max_active_luns = 14;
3695 				} else {
3696 					/* XXX KDM now what?? */
3697 				}
3698 			} else {
3699 				if (rs_type ==
3700 				     CS_RAIDSET_PERSONALITY_RAID5) {
3701 					max_active_luns = 8;
3702 				} else if (rs_type ==
3703 					 CS_RAIDSET_PERSONALITY_RAID1){
3704 					max_active_luns = 16;
3705 				} else {
3706 					/* XXX KDM now what?? */
3707 				}
3708 			}
3709 		} else {
3710 			/* 50% power */
3711 			/*
3712 			 * With 50% power in a 64 drive configuration, we
3713 			 * can power all LUNs we present.
3714 			 */
3715 			max_active_luns = num_luns;
3716 		}
3717 		break;
3718 	case 112:
3719 		if (shelf_50pct_power == CS_FALSE) {
3720 			/* 25% power */
3721 			if (aor_present == CS_TRUE) {
3722 				if (rs_type ==
3723 				     CS_RAIDSET_PERSONALITY_RAID5) {
3724 					max_active_luns = 7;
3725 				} else if (rs_type ==
3726 					 CS_RAIDSET_PERSONALITY_RAID1){
3727 					max_active_luns = 14;
3728 				} else {
3729 					/* XXX KDM now what?? */
3730 				}
3731 			} else {
3732 				if (rs_type ==
3733 				     CS_RAIDSET_PERSONALITY_RAID5) {
3734 					max_active_luns = 8;
3735 				} else if (rs_type ==
3736 					 CS_RAIDSET_PERSONALITY_RAID1){
3737 					max_active_luns = 16;
3738 				} else {
3739 					/* XXX KDM now what?? */
3740 				}
3741 			}
3742 		} else {
3743 			/* 50% power */
3744 			if (aor_present == CS_TRUE) {
3745 				if (rs_type ==
3746 				     CS_RAIDSET_PERSONALITY_RAID5) {
3747 					max_active_luns = 14;
3748 				} else if (rs_type ==
3749 					 CS_RAIDSET_PERSONALITY_RAID1){
3750 					/*
3751 					 * We're assuming here that disk
3752 					 * caching is enabled, and so we're
3753 					 * able to power up half of each
3754 					 * LUN, and cache all writes.
3755 					 */
3756 					max_active_luns = num_luns;
3757 				} else {
3758 					/* XXX KDM now what?? */
3759 				}
3760 			} else {
3761 				if (rs_type ==
3762 				     CS_RAIDSET_PERSONALITY_RAID5) {
3763 					max_active_luns = 15;
3764 				} else if (rs_type ==
3765 					 CS_RAIDSET_PERSONALITY_RAID1){
3766 					max_active_luns = 30;
3767 				} else {
3768 					/* XXX KDM now what?? */
3769 				}
3770 			}
3771 		}
3772 		break;
3773 	default:
3774 		/*
3775 		 * In this case, we have an unknown configuration, so we
3776 		 * just use the default from above.
3777 		 */
3778 		break;
3779 	}
3780 
3781 	page->max_active_luns = max_active_luns;
3782 #if 0
3783 	printk("%s: total_luns = %d, max_active_luns = %d\n", __func__,
3784 	       page->total_luns, page->max_active_luns);
3785 #endif
3786 }
3787 #endif /* NEEDTOPORT */
3788 
3789 /*
3790  * This routine could be used in the future to load default and/or saved
3791  * mode page parameters for a particuar lun.
3792  */
3793 static int
3794 ctl_init_page_index(struct ctl_lun *lun)
3795 {
3796 	int i;
3797 	struct ctl_page_index *page_index;
3798 	struct ctl_softc *softc;
3799 
3800 	memcpy(&lun->mode_pages.index, page_index_template,
3801 	       sizeof(page_index_template));
3802 
3803 	softc = lun->ctl_softc;
3804 
3805 	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
3806 
3807 		page_index = &lun->mode_pages.index[i];
3808 		/*
3809 		 * If this is a disk-only mode page, there's no point in
3810 		 * setting it up.  For some pages, we have to have some
3811 		 * basic information about the disk in order to calculate the
3812 		 * mode page data.
3813 		 */
3814 		if ((lun->be_lun->lun_type != T_DIRECT)
3815 		 && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
3816 			continue;
3817 
3818 		switch (page_index->page_code & SMPH_PC_MASK) {
3819 		case SMS_FORMAT_DEVICE_PAGE: {
3820 			struct scsi_format_page *format_page;
3821 
3822 			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3823 				panic("subpage is incorrect!");
3824 
3825 			/*
3826 			 * Sectors per track are set above.  Bytes per
3827 			 * sector need to be set here on a per-LUN basis.
3828 			 */
3829 			memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
3830 			       &format_page_default,
3831 			       sizeof(format_page_default));
3832 			memcpy(&lun->mode_pages.format_page[
3833 			       CTL_PAGE_CHANGEABLE], &format_page_changeable,
3834 			       sizeof(format_page_changeable));
3835 			memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
3836 			       &format_page_default,
3837 			       sizeof(format_page_default));
3838 			memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
3839 			       &format_page_default,
3840 			       sizeof(format_page_default));
3841 
3842 			format_page = &lun->mode_pages.format_page[
3843 				CTL_PAGE_CURRENT];
3844 			scsi_ulto2b(lun->be_lun->blocksize,
3845 				    format_page->bytes_per_sector);
3846 
3847 			format_page = &lun->mode_pages.format_page[
3848 				CTL_PAGE_DEFAULT];
3849 			scsi_ulto2b(lun->be_lun->blocksize,
3850 				    format_page->bytes_per_sector);
3851 
3852 			format_page = &lun->mode_pages.format_page[
3853 				CTL_PAGE_SAVED];
3854 			scsi_ulto2b(lun->be_lun->blocksize,
3855 				    format_page->bytes_per_sector);
3856 
3857 			page_index->page_data =
3858 				(uint8_t *)lun->mode_pages.format_page;
3859 			break;
3860 		}
3861 		case SMS_RIGID_DISK_PAGE: {
3862 			struct scsi_rigid_disk_page *rigid_disk_page;
3863 			uint32_t sectors_per_cylinder;
3864 			uint64_t cylinders;
3865 #ifndef	__XSCALE__
3866 			int shift;
3867 #endif /* !__XSCALE__ */
3868 
3869 			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3870 				panic("invalid subpage value %d",
3871 				      page_index->subpage);
3872 
3873 			/*
3874 			 * Rotation rate and sectors per track are set
3875 			 * above.  We calculate the cylinders here based on
3876 			 * capacity.  Due to the number of heads and
3877 			 * sectors per track we're using, smaller arrays
3878 			 * may turn out to have 0 cylinders.  Linux and
3879 			 * FreeBSD don't pay attention to these mode pages
3880 			 * to figure out capacity, but Solaris does.  It
3881 			 * seems to deal with 0 cylinders just fine, and
3882 			 * works out a fake geometry based on the capacity.
3883 			 */
3884 			memcpy(&lun->mode_pages.rigid_disk_page[
3885 			       CTL_PAGE_CURRENT], &rigid_disk_page_default,
3886 			       sizeof(rigid_disk_page_default));
3887 			memcpy(&lun->mode_pages.rigid_disk_page[
3888 			       CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
3889 			       sizeof(rigid_disk_page_changeable));
3890 			memcpy(&lun->mode_pages.rigid_disk_page[
3891 			       CTL_PAGE_DEFAULT], &rigid_disk_page_default,
3892 			       sizeof(rigid_disk_page_default));
3893 			memcpy(&lun->mode_pages.rigid_disk_page[
3894 			       CTL_PAGE_SAVED], &rigid_disk_page_default,
3895 			       sizeof(rigid_disk_page_default));
3896 
3897 			sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
3898 				CTL_DEFAULT_HEADS;
3899 
3900 			/*
3901 			 * The divide method here will be more accurate,
3902 			 * probably, but results in floating point being
3903 			 * used in the kernel on i386 (__udivdi3()).  On the
3904 			 * XScale, though, __udivdi3() is implemented in
3905 			 * software.
3906 			 *
3907 			 * The shift method for cylinder calculation is
3908 			 * accurate if sectors_per_cylinder is a power of
3909 			 * 2.  Otherwise it might be slightly off -- you
3910 			 * might have a bit of a truncation problem.
3911 			 */
3912 #ifdef	__XSCALE__
3913 			cylinders = (lun->be_lun->maxlba + 1) /
3914 				sectors_per_cylinder;
3915 #else
3916 			for (shift = 31; shift > 0; shift--) {
3917 				if (sectors_per_cylinder & (1 << shift))
3918 					break;
3919 			}
3920 			cylinders = (lun->be_lun->maxlba + 1) >> shift;
3921 #endif
3922 
3923 			/*
3924 			 * We've basically got 3 bytes, or 24 bits for the
3925 			 * cylinder size in the mode page.  If we're over,
3926 			 * just round down to 2^24.
3927 			 */
3928 			if (cylinders > 0xffffff)
3929 				cylinders = 0xffffff;
3930 
3931 			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
3932 				CTL_PAGE_CURRENT];
3933 			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
3934 
3935 			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
3936 				CTL_PAGE_DEFAULT];
3937 			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
3938 
3939 			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
3940 				CTL_PAGE_SAVED];
3941 			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
3942 
3943 			page_index->page_data =
3944 				(uint8_t *)lun->mode_pages.rigid_disk_page;
3945 			break;
3946 		}
3947 		case SMS_CACHING_PAGE: {
3948 
3949 			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3950 				panic("invalid subpage value %d",
3951 				      page_index->subpage);
3952 			/*
3953 			 * Defaults should be okay here, no calculations
3954 			 * needed.
3955 			 */
3956 			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
3957 			       &caching_page_default,
3958 			       sizeof(caching_page_default));
3959 			memcpy(&lun->mode_pages.caching_page[
3960 			       CTL_PAGE_CHANGEABLE], &caching_page_changeable,
3961 			       sizeof(caching_page_changeable));
3962 			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
3963 			       &caching_page_default,
3964 			       sizeof(caching_page_default));
3965 			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
3966 			       &caching_page_default,
3967 			       sizeof(caching_page_default));
3968 			page_index->page_data =
3969 				(uint8_t *)lun->mode_pages.caching_page;
3970 			break;
3971 		}
3972 		case SMS_CONTROL_MODE_PAGE: {
3973 
3974 			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3975 				panic("invalid subpage value %d",
3976 				      page_index->subpage);
3977 
3978 			/*
3979 			 * Defaults should be okay here, no calculations
3980 			 * needed.
3981 			 */
3982 			memcpy(&lun->mode_pages.control_page[CTL_PAGE_CURRENT],
3983 			       &control_page_default,
3984 			       sizeof(control_page_default));
3985 			memcpy(&lun->mode_pages.control_page[
3986 			       CTL_PAGE_CHANGEABLE], &control_page_changeable,
3987 			       sizeof(control_page_changeable));
3988 			memcpy(&lun->mode_pages.control_page[CTL_PAGE_DEFAULT],
3989 			       &control_page_default,
3990 			       sizeof(control_page_default));
3991 			memcpy(&lun->mode_pages.control_page[CTL_PAGE_SAVED],
3992 			       &control_page_default,
3993 			       sizeof(control_page_default));
3994 			page_index->page_data =
3995 				(uint8_t *)lun->mode_pages.control_page;
3996 			break;
3997 
3998 		}
3999 		case SMS_VENDOR_SPECIFIC_PAGE:{
4000 			switch (page_index->subpage) {
4001 			case PWR_SUBPAGE_CODE: {
4002 				struct copan_power_subpage *current_page,
4003 							   *saved_page;
4004 
4005 				memcpy(&lun->mode_pages.power_subpage[
4006 				       CTL_PAGE_CURRENT],
4007 				       &power_page_default,
4008 				       sizeof(power_page_default));
4009 				memcpy(&lun->mode_pages.power_subpage[
4010 				       CTL_PAGE_CHANGEABLE],
4011 				       &power_page_changeable,
4012 				       sizeof(power_page_changeable));
4013 				memcpy(&lun->mode_pages.power_subpage[
4014 				       CTL_PAGE_DEFAULT],
4015 				       &power_page_default,
4016 				       sizeof(power_page_default));
4017 				memcpy(&lun->mode_pages.power_subpage[
4018 				       CTL_PAGE_SAVED],
4019 				       &power_page_default,
4020 				       sizeof(power_page_default));
4021 				page_index->page_data =
4022 				    (uint8_t *)lun->mode_pages.power_subpage;
4023 
4024 				current_page = (struct copan_power_subpage *)
4025 					(page_index->page_data +
4026 					 (page_index->page_len *
4027 					  CTL_PAGE_CURRENT));
4028 			        saved_page = (struct copan_power_subpage *)
4029 				        (page_index->page_data +
4030 					 (page_index->page_len *
4031 					  CTL_PAGE_SAVED));
4032 				break;
4033 			}
4034 			case APS_SUBPAGE_CODE: {
4035 				struct copan_aps_subpage *current_page,
4036 							 *saved_page;
4037 
4038 				// This gets set multiple times but
4039 				// it should always be the same. It's
4040 				// only done during init so who cares.
4041 				index_to_aps_page = i;
4042 
4043 				memcpy(&lun->mode_pages.aps_subpage[
4044 				       CTL_PAGE_CURRENT],
4045 				       &aps_page_default,
4046 				       sizeof(aps_page_default));
4047 				memcpy(&lun->mode_pages.aps_subpage[
4048 				       CTL_PAGE_CHANGEABLE],
4049 				       &aps_page_changeable,
4050 				       sizeof(aps_page_changeable));
4051 				memcpy(&lun->mode_pages.aps_subpage[
4052 				       CTL_PAGE_DEFAULT],
4053 				       &aps_page_default,
4054 				       sizeof(aps_page_default));
4055 				memcpy(&lun->mode_pages.aps_subpage[
4056 				       CTL_PAGE_SAVED],
4057 				       &aps_page_default,
4058 				       sizeof(aps_page_default));
4059 				page_index->page_data =
4060 					(uint8_t *)lun->mode_pages.aps_subpage;
4061 
4062 				current_page = (struct copan_aps_subpage *)
4063 					(page_index->page_data +
4064 					 (page_index->page_len *
4065 					  CTL_PAGE_CURRENT));
4066 				saved_page = (struct copan_aps_subpage *)
4067 					(page_index->page_data +
4068 					 (page_index->page_len *
4069 					  CTL_PAGE_SAVED));
4070 				break;
4071 			}
4072 			case DBGCNF_SUBPAGE_CODE: {
4073 				struct copan_debugconf_subpage *current_page,
4074 							       *saved_page;
4075 
4076 				memcpy(&lun->mode_pages.debugconf_subpage[
4077 				       CTL_PAGE_CURRENT],
4078 				       &debugconf_page_default,
4079 				       sizeof(debugconf_page_default));
4080 				memcpy(&lun->mode_pages.debugconf_subpage[
4081 				       CTL_PAGE_CHANGEABLE],
4082 				       &debugconf_page_changeable,
4083 				       sizeof(debugconf_page_changeable));
4084 				memcpy(&lun->mode_pages.debugconf_subpage[
4085 				       CTL_PAGE_DEFAULT],
4086 				       &debugconf_page_default,
4087 				       sizeof(debugconf_page_default));
4088 				memcpy(&lun->mode_pages.debugconf_subpage[
4089 				       CTL_PAGE_SAVED],
4090 				       &debugconf_page_default,
4091 				       sizeof(debugconf_page_default));
4092 				page_index->page_data =
4093 					(uint8_t *)lun->mode_pages.debugconf_subpage;
4094 
4095 				current_page = (struct copan_debugconf_subpage *)
4096 					(page_index->page_data +
4097 					 (page_index->page_len *
4098 					  CTL_PAGE_CURRENT));
4099 				saved_page = (struct copan_debugconf_subpage *)
4100 					(page_index->page_data +
4101 					 (page_index->page_len *
4102 					  CTL_PAGE_SAVED));
4103 				break;
4104 			}
4105 			default:
4106 				panic("invalid subpage value %d",
4107 				      page_index->subpage);
4108 				break;
4109 			}
4110    			break;
4111 		}
4112 		default:
4113 			panic("invalid page value %d",
4114 			      page_index->page_code & SMPH_PC_MASK);
4115 			break;
4116     	}
4117 	}
4118 
4119 	return (CTL_RETVAL_COMPLETE);
4120 }
4121 
4122 /*
4123  * LUN allocation.
4124  *
4125  * Requirements:
4126  * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4127  *   wants us to allocate the LUN and he can block.
4128  * - ctl_softc is always set
4129  * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4130  *
4131  * Returns 0 for success, non-zero (errno) for failure.
4132  */
4133 static int
4134 ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4135 	      struct ctl_be_lun *const be_lun, struct ctl_id target_id)
4136 {
4137 	struct ctl_lun *nlun, *lun;
4138 	struct ctl_frontend *fe;
4139 	int lun_number, i;
4140 
4141 	if (be_lun == NULL)
4142 		return (EINVAL);
4143 
4144 	/*
4145 	 * We currently only support Direct Access or Processor LUN types.
4146 	 */
4147 	switch (be_lun->lun_type) {
4148 	case T_DIRECT:
4149 		break;
4150 	case T_PROCESSOR:
4151 		break;
4152 	case T_SEQUENTIAL:
4153 	case T_CHANGER:
4154 	default:
4155 		be_lun->lun_config_status(be_lun->be_lun,
4156 					  CTL_LUN_CONFIG_FAILURE);
4157 		break;
4158 	}
4159 	if (ctl_lun == NULL) {
4160 		lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4161 		if (lun == NULL) {
4162 			be_lun->lun_config_status(lun->be_lun->be_lun,
4163 						  CTL_LUN_CONFIG_FAILURE);
4164 			return (-ENOMEM);
4165 		}
4166 		lun->flags = CTL_LUN_MALLOCED;
4167 	} else
4168 		lun = ctl_lun;
4169 
4170 	memset(lun, 0, sizeof(*lun));
4171 
4172 	mtx_lock(&ctl_softc->ctl_lock);
4173 	/*
4174 	 * See if the caller requested a particular LUN number.  If so, see
4175 	 * if it is available.  Otherwise, allocate the first available LUN.
4176 	 */
4177 	if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4178 		if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4179 		 || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4180 			mtx_unlock(&ctl_softc->ctl_lock);
4181 			if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4182 				printf("ctl: requested LUN ID %d is higher "
4183 				       "than CTL_MAX_LUNS - 1 (%d)\n",
4184 				       be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4185 			} else {
4186 				/*
4187 				 * XXX KDM return an error, or just assign
4188 				 * another LUN ID in this case??
4189 				 */
4190 				printf("ctl: requested LUN ID %d is already "
4191 				       "in use\n", be_lun->req_lun_id);
4192 			}
4193 			if (lun->flags & CTL_LUN_MALLOCED)
4194 				free(lun, M_CTL);
4195 			be_lun->lun_config_status(be_lun->be_lun,
4196 						  CTL_LUN_CONFIG_FAILURE);
4197 			return (ENOSPC);
4198 		}
4199 		lun_number = be_lun->req_lun_id;
4200 	} else {
4201 		lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, CTL_MAX_LUNS);
4202 		if (lun_number == -1) {
4203 			mtx_unlock(&ctl_softc->ctl_lock);
4204 			printf("ctl: can't allocate LUN on target %ju, out of "
4205 			       "LUNs\n", (uintmax_t)target_id.id);
4206 			if (lun->flags & CTL_LUN_MALLOCED)
4207 				free(lun, M_CTL);
4208 			be_lun->lun_config_status(be_lun->be_lun,
4209 						  CTL_LUN_CONFIG_FAILURE);
4210 			return (ENOSPC);
4211 		}
4212 	}
4213 	ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4214 
4215 	lun->target = target_id;
4216 	lun->lun = lun_number;
4217 	lun->be_lun = be_lun;
4218 	/*
4219 	 * The processor LUN is always enabled.  Disk LUNs come on line
4220 	 * disabled, and must be enabled by the backend.
4221 	 */
4222 	lun->flags = CTL_LUN_DISABLED;
4223 	lun->backend = be_lun->be;
4224 	be_lun->ctl_lun = lun;
4225 	be_lun->lun_id = lun_number;
4226 	atomic_add_int(&be_lun->be->num_luns, 1);
4227 	if (be_lun->flags & CTL_LUN_FLAG_POWERED_OFF)
4228 		lun->flags |= CTL_LUN_STOPPED;
4229 
4230 	if (be_lun->flags & CTL_LUN_FLAG_INOPERABLE)
4231 		lun->flags |= CTL_LUN_INOPERABLE;
4232 
4233 	if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4234 		lun->flags |= CTL_LUN_PRIMARY_SC;
4235 
4236 	lun->ctl_softc = ctl_softc;
4237 	TAILQ_INIT(&lun->ooa_queue);
4238 	TAILQ_INIT(&lun->blocked_queue);
4239 	STAILQ_INIT(&lun->error_list);
4240 
4241 	/*
4242 	 * Initialize the mode page index.
4243 	 */
4244 	ctl_init_page_index(lun);
4245 
4246 	/*
4247 	 * Set the poweron UA for all initiators on this LUN only.
4248 	 */
4249 	for (i = 0; i < CTL_MAX_INITIATORS; i++)
4250 		lun->pending_sense[i].ua_pending = CTL_UA_POWERON;
4251 
4252 	/*
4253 	 * Now, before we insert this lun on the lun list, set the lun
4254 	 * inventory changed UA for all other luns.
4255 	 */
4256 	STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4257 		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4258 			nlun->pending_sense[i].ua_pending |= CTL_UA_LUN_CHANGE;
4259 		}
4260 	}
4261 
4262 	STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4263 
4264 	ctl_softc->ctl_luns[lun_number] = lun;
4265 
4266 	ctl_softc->num_luns++;
4267 
4268 	/* Setup statistics gathering */
4269 	lun->stats.device_type = be_lun->lun_type;
4270 	lun->stats.lun_number = lun_number;
4271 	if (lun->stats.device_type == T_DIRECT)
4272 		lun->stats.blocksize = be_lun->blocksize;
4273 	else
4274 		lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4275 	for (i = 0;i < CTL_MAX_PORTS;i++)
4276 		lun->stats.ports[i].targ_port = i;
4277 
4278 	mtx_unlock(&ctl_softc->ctl_lock);
4279 
4280 	lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4281 
4282 	/*
4283 	 * Run through each registered FETD and bring it online if it isn't
4284 	 * already.  Enable the target ID if it hasn't been enabled, and
4285 	 * enable this particular LUN.
4286 	 */
4287 	STAILQ_FOREACH(fe, &ctl_softc->fe_list, links) {
4288 		int retval;
4289 
4290 		/*
4291 		 * XXX KDM this only works for ONE TARGET ID.  We'll need
4292 		 * to do things differently if we go to a multiple target
4293 		 * ID scheme.
4294 		 */
4295 		if ((fe->status & CTL_PORT_STATUS_TARG_ONLINE) == 0) {
4296 
4297 			retval = fe->targ_enable(fe->targ_lun_arg, target_id);
4298 			if (retval != 0) {
4299 				printf("ctl_alloc_lun: FETD %s port %d "
4300 				       "returned error %d for targ_enable on "
4301 				       "target %ju\n", fe->port_name,
4302 				       fe->targ_port, retval,
4303 				       (uintmax_t)target_id.id);
4304 			} else
4305 				fe->status |= CTL_PORT_STATUS_TARG_ONLINE;
4306 		}
4307 
4308 		retval = fe->lun_enable(fe->targ_lun_arg, target_id,lun_number);
4309 		if (retval != 0) {
4310 			printf("ctl_alloc_lun: FETD %s port %d returned error "
4311 			       "%d for lun_enable on target %ju lun %d\n",
4312 			       fe->port_name, fe->targ_port, retval,
4313 			       (uintmax_t)target_id.id, lun_number);
4314 		} else
4315 			fe->status |= CTL_PORT_STATUS_LUN_ONLINE;
4316 	}
4317 	return (0);
4318 }
4319 
4320 /*
4321  * Delete a LUN.
4322  * Assumptions:
4323  * - caller holds ctl_softc->ctl_lock.
4324  * - LUN has already been marked invalid and any pending I/O has been taken
4325  *   care of.
4326  */
4327 static int
4328 ctl_free_lun(struct ctl_lun *lun)
4329 {
4330 	struct ctl_softc *softc;
4331 #if 0
4332 	struct ctl_frontend *fe;
4333 #endif
4334 	struct ctl_lun *nlun;
4335 	union ctl_io *io, *next_io;
4336 	int i;
4337 
4338 	softc = lun->ctl_softc;
4339 
4340 	STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4341 
4342 	ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4343 
4344 	softc->ctl_luns[lun->lun] = NULL;
4345 
4346 	if (TAILQ_FIRST(&lun->ooa_queue) != NULL) {
4347 		printf("ctl_free_lun: aieee!! freeing a LUN with "
4348 		       "outstanding I/O!!\n");
4349 	}
4350 
4351 	/*
4352 	 * If we have anything pending on the RtR queue, remove it.
4353 	 */
4354 	for (io = (union ctl_io *)STAILQ_FIRST(&softc->rtr_queue); io != NULL;
4355 	     io = next_io) {
4356 		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
4357 		if ((io->io_hdr.nexus.targ_target.id == lun->target.id)
4358 		 && (io->io_hdr.nexus.targ_lun == lun->lun))
4359 			STAILQ_REMOVE(&softc->rtr_queue, &io->io_hdr,
4360 				      ctl_io_hdr, links);
4361 	}
4362 
4363 	/*
4364 	 * Then remove everything from the blocked queue.
4365 	 */
4366 	for (io = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue); io != NULL;
4367 	     io = next_io) {
4368 		next_io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,blocked_links);
4369 		TAILQ_REMOVE(&lun->blocked_queue, &io->io_hdr, blocked_links);
4370 		io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
4371 	}
4372 
4373 	/*
4374 	 * Now clear out the OOA queue, and free all the I/O.
4375 	 * XXX KDM should we notify the FETD here?  We probably need to
4376 	 * quiesce the LUN before deleting it.
4377 	 */
4378 	for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); io != NULL;
4379 	     io = next_io) {
4380 		next_io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr, ooa_links);
4381 		TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
4382 		ctl_free_io_internal(io, /*have_lock*/ 1);
4383 	}
4384 
4385 	softc->num_luns--;
4386 
4387 	/*
4388 	 * XXX KDM this scheme only works for a single target/multiple LUN
4389 	 * setup.  It needs to be revamped for a multiple target scheme.
4390 	 *
4391 	 * XXX KDM this results in fe->lun_disable() getting called twice,
4392 	 * once when ctl_disable_lun() is called, and a second time here.
4393 	 * We really need to re-think the LUN disable semantics.  There
4394 	 * should probably be several steps/levels to LUN removal:
4395 	 *  - disable
4396 	 *  - invalidate
4397 	 *  - free
4398  	 *
4399 	 * Right now we only have a disable method when communicating to
4400 	 * the front end ports, at least for individual LUNs.
4401 	 */
4402 #if 0
4403 	STAILQ_FOREACH(fe, &softc->fe_list, links) {
4404 		int retval;
4405 
4406 		retval = fe->lun_disable(fe->targ_lun_arg, lun->target,
4407 					 lun->lun);
4408 		if (retval != 0) {
4409 			printf("ctl_free_lun: FETD %s port %d returned error "
4410 			       "%d for lun_disable on target %ju lun %jd\n",
4411 			       fe->port_name, fe->targ_port, retval,
4412 			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4413 		}
4414 
4415 		if (STAILQ_FIRST(&softc->lun_list) == NULL) {
4416 			fe->status &= ~CTL_PORT_STATUS_LUN_ONLINE;
4417 
4418 			retval = fe->targ_disable(fe->targ_lun_arg,lun->target);
4419 			if (retval != 0) {
4420 				printf("ctl_free_lun: FETD %s port %d "
4421 				       "returned error %d for targ_disable on "
4422 				       "target %ju\n", fe->port_name,
4423 				       fe->targ_port, retval,
4424 				       (uintmax_t)lun->target.id);
4425 			} else
4426 				fe->status &= ~CTL_PORT_STATUS_TARG_ONLINE;
4427 
4428 			if ((fe->status & CTL_PORT_STATUS_TARG_ONLINE) != 0)
4429 				continue;
4430 
4431 #if 0
4432 			fe->port_offline(fe->onoff_arg);
4433 			fe->status &= ~CTL_PORT_STATUS_ONLINE;
4434 #endif
4435 		}
4436 	}
4437 #endif
4438 
4439 	/*
4440 	 * Tell the backend to free resources, if this LUN has a backend.
4441 	 */
4442 	atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4443 	lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4444 
4445 	if (lun->flags & CTL_LUN_MALLOCED)
4446 		free(lun, M_CTL);
4447 
4448 	STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4449 		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4450 			nlun->pending_sense[i].ua_pending |= CTL_UA_LUN_CHANGE;
4451 		}
4452 	}
4453 
4454 	return (0);
4455 }
4456 
4457 static void
4458 ctl_create_lun(struct ctl_be_lun *be_lun)
4459 {
4460 	struct ctl_softc *ctl_softc;
4461 
4462 	ctl_softc = control_softc;
4463 
4464 	/*
4465 	 * ctl_alloc_lun() should handle all potential failure cases.
4466 	 */
4467 	ctl_alloc_lun(ctl_softc, NULL, be_lun, ctl_softc->target);
4468 }
4469 
4470 int
4471 ctl_add_lun(struct ctl_be_lun *be_lun)
4472 {
4473 	struct ctl_softc *ctl_softc;
4474 
4475 	ctl_softc = control_softc;
4476 
4477 	mtx_lock(&ctl_softc->ctl_lock);
4478 	STAILQ_INSERT_TAIL(&ctl_softc->pending_lun_queue, be_lun, links);
4479 	mtx_unlock(&ctl_softc->ctl_lock);
4480 
4481 	ctl_wakeup_thread();
4482 
4483 	return (0);
4484 }
4485 
4486 int
4487 ctl_enable_lun(struct ctl_be_lun *be_lun)
4488 {
4489 	struct ctl_softc *ctl_softc;
4490 	struct ctl_frontend *fe, *nfe;
4491 	struct ctl_lun *lun;
4492 	int retval;
4493 
4494 	ctl_softc = control_softc;
4495 
4496 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4497 
4498 	mtx_lock(&ctl_softc->ctl_lock);
4499 	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4500 		/*
4501 		 * eh?  Why did we get called if the LUN is already
4502 		 * enabled?
4503 		 */
4504 		mtx_unlock(&ctl_softc->ctl_lock);
4505 		return (0);
4506 	}
4507 	lun->flags &= ~CTL_LUN_DISABLED;
4508 
4509 	for (fe = STAILQ_FIRST(&ctl_softc->fe_list); fe != NULL; fe = nfe) {
4510 		nfe = STAILQ_NEXT(fe, links);
4511 
4512 		/*
4513 		 * Drop the lock while we call the FETD's enable routine.
4514 		 * This can lead to a callback into CTL (at least in the
4515 		 * case of the internal initiator frontend.
4516 		 */
4517 		mtx_unlock(&ctl_softc->ctl_lock);
4518 		retval = fe->lun_enable(fe->targ_lun_arg, lun->target,lun->lun);
4519 		mtx_lock(&ctl_softc->ctl_lock);
4520 		if (retval != 0) {
4521 			printf("%s: FETD %s port %d returned error "
4522 			       "%d for lun_enable on target %ju lun %jd\n",
4523 			       __func__, fe->port_name, fe->targ_port, retval,
4524 			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4525 		}
4526 #if 0
4527 		 else {
4528             /* NOTE:  TODO:  why does lun enable affect port status? */
4529 			fe->status |= CTL_PORT_STATUS_LUN_ONLINE;
4530 		}
4531 #endif
4532 	}
4533 
4534 	mtx_unlock(&ctl_softc->ctl_lock);
4535 
4536 	return (0);
4537 }
4538 
4539 int
4540 ctl_disable_lun(struct ctl_be_lun *be_lun)
4541 {
4542 	struct ctl_softc *ctl_softc;
4543 	struct ctl_frontend *fe;
4544 	struct ctl_lun *lun;
4545 	int retval;
4546 
4547 	ctl_softc = control_softc;
4548 
4549 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4550 
4551 	mtx_lock(&ctl_softc->ctl_lock);
4552 
4553 	if (lun->flags & CTL_LUN_DISABLED) {
4554 		mtx_unlock(&ctl_softc->ctl_lock);
4555 		return (0);
4556 	}
4557 	lun->flags |= CTL_LUN_DISABLED;
4558 
4559 	STAILQ_FOREACH(fe, &ctl_softc->fe_list, links) {
4560 		mtx_unlock(&ctl_softc->ctl_lock);
4561 		/*
4562 		 * Drop the lock before we call the frontend's disable
4563 		 * routine, to avoid lock order reversals.
4564 		 *
4565 		 * XXX KDM what happens if the frontend list changes while
4566 		 * we're traversing it?  It's unlikely, but should be handled.
4567 		 */
4568 		retval = fe->lun_disable(fe->targ_lun_arg, lun->target,
4569 					 lun->lun);
4570 		mtx_lock(&ctl_softc->ctl_lock);
4571 		if (retval != 0) {
4572 			printf("ctl_alloc_lun: FETD %s port %d returned error "
4573 			       "%d for lun_disable on target %ju lun %jd\n",
4574 			       fe->port_name, fe->targ_port, retval,
4575 			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4576 		}
4577 	}
4578 
4579 	mtx_unlock(&ctl_softc->ctl_lock);
4580 
4581 	return (0);
4582 }
4583 
4584 int
4585 ctl_start_lun(struct ctl_be_lun *be_lun)
4586 {
4587 	struct ctl_softc *ctl_softc;
4588 	struct ctl_lun *lun;
4589 
4590 	ctl_softc = control_softc;
4591 
4592 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4593 
4594 	mtx_lock(&ctl_softc->ctl_lock);
4595 	lun->flags &= ~CTL_LUN_STOPPED;
4596 	mtx_unlock(&ctl_softc->ctl_lock);
4597 
4598 	return (0);
4599 }
4600 
4601 int
4602 ctl_stop_lun(struct ctl_be_lun *be_lun)
4603 {
4604 	struct ctl_softc *ctl_softc;
4605 	struct ctl_lun *lun;
4606 
4607 	ctl_softc = control_softc;
4608 
4609 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4610 
4611 	mtx_lock(&ctl_softc->ctl_lock);
4612 	lun->flags |= CTL_LUN_STOPPED;
4613 	mtx_unlock(&ctl_softc->ctl_lock);
4614 
4615 	return (0);
4616 }
4617 
4618 int
4619 ctl_lun_offline(struct ctl_be_lun *be_lun)
4620 {
4621 	struct ctl_softc *ctl_softc;
4622 	struct ctl_lun *lun;
4623 
4624 	ctl_softc = control_softc;
4625 
4626 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4627 
4628 	mtx_lock(&ctl_softc->ctl_lock);
4629 	lun->flags |= CTL_LUN_OFFLINE;
4630 	mtx_unlock(&ctl_softc->ctl_lock);
4631 
4632 	return (0);
4633 }
4634 
4635 int
4636 ctl_lun_online(struct ctl_be_lun *be_lun)
4637 {
4638 	struct ctl_softc *ctl_softc;
4639 	struct ctl_lun *lun;
4640 
4641 	ctl_softc = control_softc;
4642 
4643 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4644 
4645 	mtx_lock(&ctl_softc->ctl_lock);
4646 	lun->flags &= ~CTL_LUN_OFFLINE;
4647 	mtx_unlock(&ctl_softc->ctl_lock);
4648 
4649 	return (0);
4650 }
4651 
4652 int
4653 ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4654 {
4655 	struct ctl_softc *ctl_softc;
4656 	struct ctl_lun *lun;
4657 
4658 	ctl_softc = control_softc;
4659 
4660 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4661 
4662 	mtx_lock(&ctl_softc->ctl_lock);
4663 
4664 	/*
4665 	 * The LUN needs to be disabled before it can be marked invalid.
4666 	 */
4667 	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4668 		mtx_unlock(&ctl_softc->ctl_lock);
4669 		return (-1);
4670 	}
4671 	/*
4672 	 * Mark the LUN invalid.
4673 	 */
4674 	lun->flags |= CTL_LUN_INVALID;
4675 
4676 	/*
4677 	 * If there is nothing in the OOA queue, go ahead and free the LUN.
4678 	 * If we have something in the OOA queue, we'll free it when the
4679 	 * last I/O completes.
4680 	 */
4681 	if (TAILQ_FIRST(&lun->ooa_queue) == NULL)
4682 		ctl_free_lun(lun);
4683 	mtx_unlock(&ctl_softc->ctl_lock);
4684 
4685 	return (0);
4686 }
4687 
4688 int
4689 ctl_lun_inoperable(struct ctl_be_lun *be_lun)
4690 {
4691 	struct ctl_softc *ctl_softc;
4692 	struct ctl_lun *lun;
4693 
4694 	ctl_softc = control_softc;
4695 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4696 
4697 	mtx_lock(&ctl_softc->ctl_lock);
4698 	lun->flags |= CTL_LUN_INOPERABLE;
4699 	mtx_unlock(&ctl_softc->ctl_lock);
4700 
4701 	return (0);
4702 }
4703 
4704 int
4705 ctl_lun_operable(struct ctl_be_lun *be_lun)
4706 {
4707 	struct ctl_softc *ctl_softc;
4708 	struct ctl_lun *lun;
4709 
4710 	ctl_softc = control_softc;
4711 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4712 
4713 	mtx_lock(&ctl_softc->ctl_lock);
4714 	lun->flags &= ~CTL_LUN_INOPERABLE;
4715 	mtx_unlock(&ctl_softc->ctl_lock);
4716 
4717 	return (0);
4718 }
4719 
4720 int
4721 ctl_lun_power_lock(struct ctl_be_lun *be_lun, struct ctl_nexus *nexus,
4722 		   int lock)
4723 {
4724 	struct ctl_softc *softc;
4725 	struct ctl_lun *lun;
4726 	struct copan_aps_subpage *current_sp;
4727 	struct ctl_page_index *page_index;
4728 	int i;
4729 
4730 	softc = control_softc;
4731 
4732 	mtx_lock(&softc->ctl_lock);
4733 
4734 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4735 
4736 	page_index = NULL;
4737 	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
4738 		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
4739 		     APS_PAGE_CODE)
4740 			continue;
4741 
4742 		if (lun->mode_pages.index[i].subpage != APS_SUBPAGE_CODE)
4743 			continue;
4744 		page_index = &lun->mode_pages.index[i];
4745 	}
4746 
4747 	if (page_index == NULL) {
4748 		mtx_unlock(&softc->ctl_lock);
4749 		printf("%s: APS subpage not found for lun %ju!\n", __func__,
4750 		       (uintmax_t)lun->lun);
4751 		return (1);
4752 	}
4753 #if 0
4754 	if ((softc->aps_locked_lun != 0)
4755 	 && (softc->aps_locked_lun != lun->lun)) {
4756 		printf("%s: attempt to lock LUN %llu when %llu is already "
4757 		       "locked\n");
4758 		mtx_unlock(&softc->ctl_lock);
4759 		return (1);
4760 	}
4761 #endif
4762 
4763 	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
4764 		(page_index->page_len * CTL_PAGE_CURRENT));
4765 
4766 	if (lock != 0) {
4767 		current_sp->lock_active = APS_LOCK_ACTIVE;
4768 		softc->aps_locked_lun = lun->lun;
4769 	} else {
4770 		current_sp->lock_active = 0;
4771 		softc->aps_locked_lun = 0;
4772 	}
4773 
4774 
4775 	/*
4776 	 * If we're in HA mode, try to send the lock message to the other
4777 	 * side.
4778 	 */
4779 	if (ctl_is_single == 0) {
4780 		int isc_retval;
4781 		union ctl_ha_msg lock_msg;
4782 
4783 		lock_msg.hdr.nexus = *nexus;
4784 		lock_msg.hdr.msg_type = CTL_MSG_APS_LOCK;
4785 		if (lock != 0)
4786 			lock_msg.aps.lock_flag = 1;
4787 		else
4788 			lock_msg.aps.lock_flag = 0;
4789 		isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &lock_msg,
4790 					 sizeof(lock_msg), 0);
4791 		if (isc_retval > CTL_HA_STATUS_SUCCESS) {
4792 			printf("%s: APS (lock=%d) error returned from "
4793 			       "ctl_ha_msg_send: %d\n", __func__, lock, isc_retval);
4794 			mtx_unlock(&softc->ctl_lock);
4795 			return (1);
4796 		}
4797 	}
4798 
4799 	mtx_unlock(&softc->ctl_lock);
4800 
4801 	return (0);
4802 }
4803 
4804 void
4805 ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
4806 {
4807 	struct ctl_lun *lun;
4808 	struct ctl_softc *softc;
4809 	int i;
4810 
4811 	softc = control_softc;
4812 
4813 	mtx_lock(&softc->ctl_lock);
4814 
4815 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4816 
4817 	for (i = 0; i < CTL_MAX_INITIATORS; i++)
4818 		lun->pending_sense[i].ua_pending |= CTL_UA_CAPACITY_CHANGED;
4819 
4820 	mtx_unlock(&softc->ctl_lock);
4821 }
4822 
4823 /*
4824  * Backend "memory move is complete" callback for requests that never
4825  * make it down to say RAIDCore's configuration code.
4826  */
4827 int
4828 ctl_config_move_done(union ctl_io *io)
4829 {
4830 	int retval;
4831 
4832 	retval = CTL_RETVAL_COMPLETE;
4833 
4834 
4835 	CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
4836 	/*
4837 	 * XXX KDM this shouldn't happen, but what if it does?
4838 	 */
4839 	if (io->io_hdr.io_type != CTL_IO_SCSI)
4840 		panic("I/O type isn't CTL_IO_SCSI!");
4841 
4842 	if ((io->io_hdr.port_status == 0)
4843 	 && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
4844 	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE))
4845 		io->io_hdr.status = CTL_SUCCESS;
4846 	else if ((io->io_hdr.port_status != 0)
4847 	      && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
4848 	      && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)){
4849 		/*
4850 		 * For hardware error sense keys, the sense key
4851 		 * specific value is defined to be a retry count,
4852 		 * but we use it to pass back an internal FETD
4853 		 * error code.  XXX KDM  Hopefully the FETD is only
4854 		 * using 16 bits for an error code, since that's
4855 		 * all the space we have in the sks field.
4856 		 */
4857 		ctl_set_internal_failure(&io->scsiio,
4858 					 /*sks_valid*/ 1,
4859 					 /*retry_count*/
4860 					 io->io_hdr.port_status);
4861 		free(io->scsiio.kern_data_ptr, M_CTL);
4862 		ctl_done(io);
4863 		goto bailout;
4864 	}
4865 
4866 	if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
4867 	 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
4868 	 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
4869 		/*
4870 		 * XXX KDM just assuming a single pointer here, and not a
4871 		 * S/G list.  If we start using S/G lists for config data,
4872 		 * we'll need to know how to clean them up here as well.
4873 		 */
4874 		free(io->scsiio.kern_data_ptr, M_CTL);
4875 		/* Hopefully the user has already set the status... */
4876 		ctl_done(io);
4877 	} else {
4878 		/*
4879 		 * XXX KDM now we need to continue data movement.  Some
4880 		 * options:
4881 		 * - call ctl_scsiio() again?  We don't do this for data
4882 		 *   writes, because for those at least we know ahead of
4883 		 *   time where the write will go and how long it is.  For
4884 		 *   config writes, though, that information is largely
4885 		 *   contained within the write itself, thus we need to
4886 		 *   parse out the data again.
4887 		 *
4888 		 * - Call some other function once the data is in?
4889 		 */
4890 
4891 		/*
4892 		 * XXX KDM call ctl_scsiio() again for now, and check flag
4893 		 * bits to see whether we're allocated or not.
4894 		 */
4895 		retval = ctl_scsiio(&io->scsiio);
4896 	}
4897 bailout:
4898 	return (retval);
4899 }
4900 
4901 /*
4902  * This gets called by a backend driver when it is done with a
4903  * configuration write.
4904  */
4905 void
4906 ctl_config_write_done(union ctl_io *io)
4907 {
4908 	/*
4909 	 * If the IO_CONT flag is set, we need to call the supplied
4910 	 * function to continue processing the I/O, instead of completing
4911 	 * the I/O just yet.
4912 	 *
4913 	 * If there is an error, though, we don't want to keep processing.
4914 	 * Instead, just send status back to the initiator.
4915 	 */
4916 	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT)
4917 	 && (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)
4918 	  || ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))) {
4919 		io->scsiio.io_cont(io);
4920 		return;
4921 	}
4922 	/*
4923 	 * Since a configuration write can be done for commands that actually
4924 	 * have data allocated, like write buffer, and commands that have
4925 	 * no data, like start/stop unit, we need to check here.
4926 	 */
4927 	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT)
4928 		free(io->scsiio.kern_data_ptr, M_CTL);
4929 	ctl_done(io);
4930 }
4931 
4932 /*
4933  * SCSI release command.
4934  */
4935 int
4936 ctl_scsi_release(struct ctl_scsiio *ctsio)
4937 {
4938 	int length, longid, thirdparty_id, resv_id;
4939 	struct ctl_softc *ctl_softc;
4940 	struct ctl_lun *lun;
4941 
4942 	length = 0;
4943 	resv_id = 0;
4944 
4945 	CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
4946 
4947 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
4948 	ctl_softc = control_softc;
4949 
4950 	switch (ctsio->cdb[0]) {
4951 	case RELEASE: {
4952 		struct scsi_release *cdb;
4953 
4954 		cdb = (struct scsi_release *)ctsio->cdb;
4955 		if ((cdb->byte2 & 0x1f) != 0) {
4956 			ctl_set_invalid_field(ctsio,
4957 					      /*sks_valid*/ 1,
4958 					      /*command*/ 1,
4959 					      /*field*/ 1,
4960 					      /*bit_valid*/ 0,
4961 					      /*bit*/ 0);
4962 			ctl_done((union ctl_io *)ctsio);
4963 			return (CTL_RETVAL_COMPLETE);
4964 		}
4965 		break;
4966 	}
4967 	case RELEASE_10: {
4968 		struct scsi_release_10 *cdb;
4969 
4970 		cdb = (struct scsi_release_10 *)ctsio->cdb;
4971 
4972 		if ((cdb->byte2 & SR10_EXTENT) != 0) {
4973 			ctl_set_invalid_field(ctsio,
4974 					      /*sks_valid*/ 1,
4975 					      /*command*/ 1,
4976 					      /*field*/ 1,
4977 					      /*bit_valid*/ 1,
4978 					      /*bit*/ 0);
4979 			ctl_done((union ctl_io *)ctsio);
4980 			return (CTL_RETVAL_COMPLETE);
4981 
4982 		}
4983 
4984 		if ((cdb->byte2 & SR10_3RDPTY) != 0) {
4985 			ctl_set_invalid_field(ctsio,
4986 					      /*sks_valid*/ 1,
4987 					      /*command*/ 1,
4988 					      /*field*/ 1,
4989 					      /*bit_valid*/ 1,
4990 					      /*bit*/ 4);
4991 			ctl_done((union ctl_io *)ctsio);
4992 			return (CTL_RETVAL_COMPLETE);
4993 		}
4994 
4995 		if (cdb->byte2 & SR10_LONGID)
4996 			longid = 1;
4997 		else
4998 			thirdparty_id = cdb->thirdparty_id;
4999 
5000 		resv_id = cdb->resv_id;
5001 		length = scsi_2btoul(cdb->length);
5002 		break;
5003 	}
5004 	}
5005 
5006 
5007 	/*
5008 	 * XXX KDM right now, we only support LUN reservation.  We don't
5009 	 * support 3rd party reservations, or extent reservations, which
5010 	 * might actually need the parameter list.  If we've gotten this
5011 	 * far, we've got a LUN reservation.  Anything else got kicked out
5012 	 * above.  So, according to SPC, ignore the length.
5013 	 */
5014 	length = 0;
5015 
5016 	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5017 	 && (length > 0)) {
5018 		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5019 		if (ctsio->kern_data_ptr == NULL) {
5020 			ctsio->io_hdr.status = CTL_SCSI_ERROR;
5021 			ctsio->io_hdr.status = SCSI_STATUS_BUSY;
5022 			ctl_done((union ctl_io *)ctsio);
5023 			return (CTL_RETVAL_COMPLETE);
5024 		}
5025 		ctsio->kern_data_len = length;
5026 		ctsio->kern_total_len = length;
5027 		ctsio->kern_data_resid = 0;
5028 		ctsio->kern_rel_offset = 0;
5029 		ctsio->kern_sg_entries = 0;
5030 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5031 		ctsio->be_move_done = ctl_config_move_done;
5032 		ctl_datamove((union ctl_io *)ctsio);
5033 
5034 		return (CTL_RETVAL_COMPLETE);
5035 	}
5036 
5037 	if (length > 0)
5038 		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5039 
5040 	mtx_lock(&ctl_softc->ctl_lock);
5041 
5042 	/*
5043 	 * According to SPC, it is not an error for an intiator to attempt
5044 	 * to release a reservation on a LUN that isn't reserved, or that
5045 	 * is reserved by another initiator.  The reservation can only be
5046 	 * released, though, by the initiator who made it or by one of
5047 	 * several reset type events.
5048 	 */
5049 	if (lun->flags & CTL_LUN_RESERVED) {
5050 		if ((ctsio->io_hdr.nexus.initid.id == lun->rsv_nexus.initid.id)
5051 		 && (ctsio->io_hdr.nexus.targ_port == lun->rsv_nexus.targ_port)
5052 		 && (ctsio->io_hdr.nexus.targ_target.id ==
5053 		     lun->rsv_nexus.targ_target.id)) {
5054 			lun->flags &= ~CTL_LUN_RESERVED;
5055 		}
5056 	}
5057 
5058 	ctsio->scsi_status = SCSI_STATUS_OK;
5059 	ctsio->io_hdr.status = CTL_SUCCESS;
5060 
5061 	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5062 		free(ctsio->kern_data_ptr, M_CTL);
5063 		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5064 	}
5065 
5066 	mtx_unlock(&ctl_softc->ctl_lock);
5067 
5068 	ctl_done((union ctl_io *)ctsio);
5069 	return (CTL_RETVAL_COMPLETE);
5070 }
5071 
5072 int
5073 ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5074 {
5075 	int extent, thirdparty, longid;
5076 	int resv_id, length;
5077 	uint64_t thirdparty_id;
5078 	struct ctl_softc *ctl_softc;
5079 	struct ctl_lun *lun;
5080 
5081 	extent = 0;
5082 	thirdparty = 0;
5083 	longid = 0;
5084 	resv_id = 0;
5085 	length = 0;
5086 	thirdparty_id = 0;
5087 
5088 	CTL_DEBUG_PRINT(("ctl_reserve\n"));
5089 
5090 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5091 	ctl_softc = control_softc;
5092 
5093 	switch (ctsio->cdb[0]) {
5094 	case RESERVE: {
5095 		struct scsi_reserve *cdb;
5096 
5097 		cdb = (struct scsi_reserve *)ctsio->cdb;
5098 		if ((cdb->byte2 & 0x1f) != 0) {
5099 			ctl_set_invalid_field(ctsio,
5100 					      /*sks_valid*/ 1,
5101 					      /*command*/ 1,
5102 					      /*field*/ 1,
5103 					      /*bit_valid*/ 0,
5104 					      /*bit*/ 0);
5105 			ctl_done((union ctl_io *)ctsio);
5106 			return (CTL_RETVAL_COMPLETE);
5107 		}
5108 		resv_id = cdb->resv_id;
5109 		length = scsi_2btoul(cdb->length);
5110 		break;
5111 	}
5112 	case RESERVE_10: {
5113 		struct scsi_reserve_10 *cdb;
5114 
5115 		cdb = (struct scsi_reserve_10 *)ctsio->cdb;
5116 
5117 		if ((cdb->byte2 & SR10_EXTENT) != 0) {
5118 			ctl_set_invalid_field(ctsio,
5119 					      /*sks_valid*/ 1,
5120 					      /*command*/ 1,
5121 					      /*field*/ 1,
5122 					      /*bit_valid*/ 1,
5123 					      /*bit*/ 0);
5124 			ctl_done((union ctl_io *)ctsio);
5125 			return (CTL_RETVAL_COMPLETE);
5126 		}
5127 		if ((cdb->byte2 & SR10_3RDPTY) != 0) {
5128 			ctl_set_invalid_field(ctsio,
5129 					      /*sks_valid*/ 1,
5130 					      /*command*/ 1,
5131 					      /*field*/ 1,
5132 					      /*bit_valid*/ 1,
5133 					      /*bit*/ 4);
5134 			ctl_done((union ctl_io *)ctsio);
5135 			return (CTL_RETVAL_COMPLETE);
5136 		}
5137 		if (cdb->byte2 & SR10_LONGID)
5138 			longid = 1;
5139 		else
5140 			thirdparty_id = cdb->thirdparty_id;
5141 
5142 		resv_id = cdb->resv_id;
5143 		length = scsi_2btoul(cdb->length);
5144 		break;
5145 	}
5146 	}
5147 
5148 	/*
5149 	 * XXX KDM right now, we only support LUN reservation.  We don't
5150 	 * support 3rd party reservations, or extent reservations, which
5151 	 * might actually need the parameter list.  If we've gotten this
5152 	 * far, we've got a LUN reservation.  Anything else got kicked out
5153 	 * above.  So, according to SPC, ignore the length.
5154 	 */
5155 	length = 0;
5156 
5157 	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5158 	 && (length > 0)) {
5159 		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5160 		if (ctsio->kern_data_ptr == NULL) {
5161 			ctsio->io_hdr.status = CTL_SCSI_ERROR;
5162 			ctsio->io_hdr.status = SCSI_STATUS_BUSY;
5163 			ctl_done((union ctl_io *)ctsio);
5164 			return (CTL_RETVAL_COMPLETE);
5165 		}
5166 		ctsio->kern_data_len = length;
5167 		ctsio->kern_total_len = length;
5168 		ctsio->kern_data_resid = 0;
5169 		ctsio->kern_rel_offset = 0;
5170 		ctsio->kern_sg_entries = 0;
5171 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5172 		ctsio->be_move_done = ctl_config_move_done;
5173 		ctl_datamove((union ctl_io *)ctsio);
5174 
5175 		return (CTL_RETVAL_COMPLETE);
5176 	}
5177 
5178 	if (length > 0)
5179 		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5180 
5181 	mtx_lock(&ctl_softc->ctl_lock);
5182 	if (lun->flags & CTL_LUN_RESERVED) {
5183 		if ((ctsio->io_hdr.nexus.initid.id != lun->rsv_nexus.initid.id)
5184 		 || (ctsio->io_hdr.nexus.targ_port != lun->rsv_nexus.targ_port)
5185 		 || (ctsio->io_hdr.nexus.targ_target.id !=
5186 		     lun->rsv_nexus.targ_target.id)) {
5187 			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
5188 			ctsio->io_hdr.status = CTL_SCSI_ERROR;
5189 			goto bailout;
5190 		}
5191 	}
5192 
5193 	lun->flags |= CTL_LUN_RESERVED;
5194 	lun->rsv_nexus = ctsio->io_hdr.nexus;
5195 
5196 	ctsio->scsi_status = SCSI_STATUS_OK;
5197 	ctsio->io_hdr.status = CTL_SUCCESS;
5198 
5199 bailout:
5200 	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5201 		free(ctsio->kern_data_ptr, M_CTL);
5202 		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5203 	}
5204 
5205 	mtx_unlock(&ctl_softc->ctl_lock);
5206 
5207 	ctl_done((union ctl_io *)ctsio);
5208 	return (CTL_RETVAL_COMPLETE);
5209 }
5210 
5211 int
5212 ctl_start_stop(struct ctl_scsiio *ctsio)
5213 {
5214 	struct scsi_start_stop_unit *cdb;
5215 	struct ctl_lun *lun;
5216 	struct ctl_softc *ctl_softc;
5217 	int retval;
5218 
5219 	CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5220 
5221 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5222 	ctl_softc = control_softc;
5223 	retval = 0;
5224 
5225 	cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5226 
5227 	/*
5228 	 * XXX KDM
5229 	 * We don't support the immediate bit on a stop unit.  In order to
5230 	 * do that, we would need to code up a way to know that a stop is
5231 	 * pending, and hold off any new commands until it completes, one
5232 	 * way or another.  Then we could accept or reject those commands
5233 	 * depending on its status.  We would almost need to do the reverse
5234 	 * of what we do below for an immediate start -- return the copy of
5235 	 * the ctl_io to the FETD with status to send to the host (and to
5236 	 * free the copy!) and then free the original I/O once the stop
5237 	 * actually completes.  That way, the OOA queue mechanism can work
5238 	 * to block commands that shouldn't proceed.  Another alternative
5239 	 * would be to put the copy in the queue in place of the original,
5240 	 * and return the original back to the caller.  That could be
5241 	 * slightly safer..
5242 	 */
5243 	if ((cdb->byte2 & SSS_IMMED)
5244 	 && ((cdb->how & SSS_START) == 0)) {
5245 		ctl_set_invalid_field(ctsio,
5246 				      /*sks_valid*/ 1,
5247 				      /*command*/ 1,
5248 				      /*field*/ 1,
5249 				      /*bit_valid*/ 1,
5250 				      /*bit*/ 0);
5251 		ctl_done((union ctl_io *)ctsio);
5252 		return (CTL_RETVAL_COMPLETE);
5253 	}
5254 
5255 	/*
5256 	 * We don't support the power conditions field.  We need to check
5257 	 * this prior to checking the load/eject and start/stop bits.
5258 	 */
5259 	if ((cdb->how & SSS_PC_MASK) != SSS_PC_START_VALID) {
5260 		ctl_set_invalid_field(ctsio,
5261 				      /*sks_valid*/ 1,
5262 				      /*command*/ 1,
5263 				      /*field*/ 4,
5264 				      /*bit_valid*/ 1,
5265 				      /*bit*/ 4);
5266 		ctl_done((union ctl_io *)ctsio);
5267 		return (CTL_RETVAL_COMPLETE);
5268 	}
5269 
5270 	/*
5271 	 * Media isn't removable, so we can't load or eject it.
5272 	 */
5273 	if ((cdb->how & SSS_LOEJ) != 0) {
5274 		ctl_set_invalid_field(ctsio,
5275 				      /*sks_valid*/ 1,
5276 				      /*command*/ 1,
5277 				      /*field*/ 4,
5278 				      /*bit_valid*/ 1,
5279 				      /*bit*/ 1);
5280 		ctl_done((union ctl_io *)ctsio);
5281 		return (CTL_RETVAL_COMPLETE);
5282 	}
5283 
5284 	if ((lun->flags & CTL_LUN_PR_RESERVED)
5285 	 && ((cdb->how & SSS_START)==0)) {
5286 		uint32_t residx;
5287 
5288 		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5289 		if (!lun->per_res[residx].registered
5290 		 || (lun->pr_res_idx!=residx && lun->res_type < 4)) {
5291 
5292 			ctl_set_reservation_conflict(ctsio);
5293 			ctl_done((union ctl_io *)ctsio);
5294 			return (CTL_RETVAL_COMPLETE);
5295 		}
5296 	}
5297 
5298 	/*
5299 	 * If there is no backend on this device, we can't start or stop
5300 	 * it.  In theory we shouldn't get any start/stop commands in the
5301 	 * first place at this level if the LUN doesn't have a backend.
5302 	 * That should get stopped by the command decode code.
5303 	 */
5304 	if (lun->backend == NULL) {
5305 		ctl_set_invalid_opcode(ctsio);
5306 		ctl_done((union ctl_io *)ctsio);
5307 		return (CTL_RETVAL_COMPLETE);
5308 	}
5309 
5310 	/*
5311 	 * XXX KDM Copan-specific offline behavior.
5312 	 * Figure out a reasonable way to port this?
5313 	 */
5314 #ifdef NEEDTOPORT
5315 	mtx_lock(&ctl_softc->ctl_lock);
5316 
5317 	if (((cdb->byte2 & SSS_ONOFFLINE) == 0)
5318 	 && (lun->flags & CTL_LUN_OFFLINE)) {
5319 		/*
5320 		 * If the LUN is offline, and the on/offline bit isn't set,
5321 		 * reject the start or stop.  Otherwise, let it through.
5322 		 */
5323 		mtx_unlock(&ctl_softc->ctl_lock);
5324 		ctl_set_lun_not_ready(ctsio);
5325 		ctl_done((union ctl_io *)ctsio);
5326 	} else {
5327 		mtx_unlock(&ctl_softc->ctl_lock);
5328 #endif /* NEEDTOPORT */
5329 		/*
5330 		 * This could be a start or a stop when we're online,
5331 		 * or a stop/offline or start/online.  A start or stop when
5332 		 * we're offline is covered in the case above.
5333 		 */
5334 		/*
5335 		 * In the non-immediate case, we send the request to
5336 		 * the backend and return status to the user when
5337 		 * it is done.
5338 		 *
5339 		 * In the immediate case, we allocate a new ctl_io
5340 		 * to hold a copy of the request, and send that to
5341 		 * the backend.  We then set good status on the
5342 		 * user's request and return it immediately.
5343 		 */
5344 		if (cdb->byte2 & SSS_IMMED) {
5345 			union ctl_io *new_io;
5346 
5347 			new_io = ctl_alloc_io(ctsio->io_hdr.pool);
5348 			if (new_io == NULL) {
5349 				ctl_set_busy(ctsio);
5350 				ctl_done((union ctl_io *)ctsio);
5351 			} else {
5352 				ctl_copy_io((union ctl_io *)ctsio,
5353 					    new_io);
5354 				retval = lun->backend->config_write(new_io);
5355 				ctl_set_success(ctsio);
5356 				ctl_done((union ctl_io *)ctsio);
5357 			}
5358 		} else {
5359 			retval = lun->backend->config_write(
5360 				(union ctl_io *)ctsio);
5361 		}
5362 #ifdef NEEDTOPORT
5363 	}
5364 #endif
5365 	return (retval);
5366 }
5367 
5368 /*
5369  * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5370  * we don't really do anything with the LBA and length fields if the user
5371  * passes them in.  Instead we'll just flush out the cache for the entire
5372  * LUN.
5373  */
5374 int
5375 ctl_sync_cache(struct ctl_scsiio *ctsio)
5376 {
5377 	struct ctl_lun *lun;
5378 	struct ctl_softc *ctl_softc;
5379 	uint64_t starting_lba;
5380 	uint32_t block_count;
5381 	int reladr, immed;
5382 	int retval;
5383 
5384 	CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5385 
5386 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5387 	ctl_softc = control_softc;
5388 	retval = 0;
5389 	reladr = 0;
5390 	immed = 0;
5391 
5392 	switch (ctsio->cdb[0]) {
5393 	case SYNCHRONIZE_CACHE: {
5394 		struct scsi_sync_cache *cdb;
5395 		cdb = (struct scsi_sync_cache *)ctsio->cdb;
5396 
5397 		if (cdb->byte2 & SSC_RELADR)
5398 			reladr = 1;
5399 
5400 		if (cdb->byte2 & SSC_IMMED)
5401 			immed = 1;
5402 
5403 		starting_lba = scsi_4btoul(cdb->begin_lba);
5404 		block_count = scsi_2btoul(cdb->lb_count);
5405 		break;
5406 	}
5407 	case SYNCHRONIZE_CACHE_16: {
5408 		struct scsi_sync_cache_16 *cdb;
5409 		cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5410 
5411 		if (cdb->byte2 & SSC_RELADR)
5412 			reladr = 1;
5413 
5414 		if (cdb->byte2 & SSC_IMMED)
5415 			immed = 1;
5416 
5417 		starting_lba = scsi_8btou64(cdb->begin_lba);
5418 		block_count = scsi_4btoul(cdb->lb_count);
5419 		break;
5420 	}
5421 	default:
5422 		ctl_set_invalid_opcode(ctsio);
5423 		ctl_done((union ctl_io *)ctsio);
5424 		goto bailout;
5425 		break; /* NOTREACHED */
5426 	}
5427 
5428 	if (immed) {
5429 		/*
5430 		 * We don't support the immediate bit.  Since it's in the
5431 		 * same place for the 10 and 16 byte SYNCHRONIZE CACHE
5432 		 * commands, we can just return the same error in either
5433 		 * case.
5434 		 */
5435 		ctl_set_invalid_field(ctsio,
5436 				      /*sks_valid*/ 1,
5437 				      /*command*/ 1,
5438 				      /*field*/ 1,
5439 				      /*bit_valid*/ 1,
5440 				      /*bit*/ 1);
5441 		ctl_done((union ctl_io *)ctsio);
5442 		goto bailout;
5443 	}
5444 
5445 	if (reladr) {
5446 		/*
5447 		 * We don't support the reladr bit either.  It can only be
5448 		 * used with linked commands, and we don't support linked
5449 		 * commands.  Since the bit is in the same place for the
5450 		 * 10 and 16 byte SYNCHRONIZE CACHE * commands, we can
5451 		 * just return the same error in either case.
5452 		 */
5453 		ctl_set_invalid_field(ctsio,
5454 				      /*sks_valid*/ 1,
5455 				      /*command*/ 1,
5456 				      /*field*/ 1,
5457 				      /*bit_valid*/ 1,
5458 				      /*bit*/ 0);
5459 		ctl_done((union ctl_io *)ctsio);
5460 		goto bailout;
5461 	}
5462 
5463 	/*
5464 	 * We check the LBA and length, but don't do anything with them.
5465 	 * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5466 	 * get flushed.  This check will just help satisfy anyone who wants
5467 	 * to see an error for an out of range LBA.
5468 	 */
5469 	if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5470 		ctl_set_lba_out_of_range(ctsio);
5471 		ctl_done((union ctl_io *)ctsio);
5472 		goto bailout;
5473 	}
5474 
5475 	/*
5476 	 * If this LUN has no backend, we can't flush the cache anyway.
5477 	 */
5478 	if (lun->backend == NULL) {
5479 		ctl_set_invalid_opcode(ctsio);
5480 		ctl_done((union ctl_io *)ctsio);
5481 		goto bailout;
5482 	}
5483 
5484 	/*
5485 	 * Check to see whether we're configured to send the SYNCHRONIZE
5486 	 * CACHE command directly to the back end.
5487 	 */
5488 	mtx_lock(&ctl_softc->ctl_lock);
5489 	if ((ctl_softc->flags & CTL_FLAG_REAL_SYNC)
5490 	 && (++(lun->sync_count) >= lun->sync_interval)) {
5491 		lun->sync_count = 0;
5492 		mtx_unlock(&ctl_softc->ctl_lock);
5493 		retval = lun->backend->config_write((union ctl_io *)ctsio);
5494 	} else {
5495 		mtx_unlock(&ctl_softc->ctl_lock);
5496 		ctl_set_success(ctsio);
5497 		ctl_done((union ctl_io *)ctsio);
5498 	}
5499 
5500 bailout:
5501 
5502 	return (retval);
5503 }
5504 
5505 int
5506 ctl_format(struct ctl_scsiio *ctsio)
5507 {
5508 	struct scsi_format *cdb;
5509 	struct ctl_lun *lun;
5510 	struct ctl_softc *ctl_softc;
5511 	int length, defect_list_len;
5512 
5513 	CTL_DEBUG_PRINT(("ctl_format\n"));
5514 
5515 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5516 	ctl_softc = control_softc;
5517 
5518 	cdb = (struct scsi_format *)ctsio->cdb;
5519 
5520 	length = 0;
5521 	if (cdb->byte2 & SF_FMTDATA) {
5522 		if (cdb->byte2 & SF_LONGLIST)
5523 			length = sizeof(struct scsi_format_header_long);
5524 		else
5525 			length = sizeof(struct scsi_format_header_short);
5526 	}
5527 
5528 	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5529 	 && (length > 0)) {
5530 		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5531 		if (ctsio->kern_data_ptr == NULL) {
5532 			ctsio->io_hdr.status = CTL_SCSI_ERROR;
5533 			ctsio->io_hdr.status = SCSI_STATUS_BUSY;
5534 			ctl_done((union ctl_io *)ctsio);
5535 			return (CTL_RETVAL_COMPLETE);
5536 		}
5537 		ctsio->kern_data_len = length;
5538 		ctsio->kern_total_len = length;
5539 		ctsio->kern_data_resid = 0;
5540 		ctsio->kern_rel_offset = 0;
5541 		ctsio->kern_sg_entries = 0;
5542 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5543 		ctsio->be_move_done = ctl_config_move_done;
5544 		ctl_datamove((union ctl_io *)ctsio);
5545 
5546 		return (CTL_RETVAL_COMPLETE);
5547 	}
5548 
5549 	defect_list_len = 0;
5550 
5551 	if (cdb->byte2 & SF_FMTDATA) {
5552 		if (cdb->byte2 & SF_LONGLIST) {
5553 			struct scsi_format_header_long *header;
5554 
5555 			header = (struct scsi_format_header_long *)
5556 				ctsio->kern_data_ptr;
5557 
5558 			defect_list_len = scsi_4btoul(header->defect_list_len);
5559 			if (defect_list_len != 0) {
5560 				ctl_set_invalid_field(ctsio,
5561 						      /*sks_valid*/ 1,
5562 						      /*command*/ 0,
5563 						      /*field*/ 2,
5564 						      /*bit_valid*/ 0,
5565 						      /*bit*/ 0);
5566 				goto bailout;
5567 			}
5568 		} else {
5569 			struct scsi_format_header_short *header;
5570 
5571 			header = (struct scsi_format_header_short *)
5572 				ctsio->kern_data_ptr;
5573 
5574 			defect_list_len = scsi_2btoul(header->defect_list_len);
5575 			if (defect_list_len != 0) {
5576 				ctl_set_invalid_field(ctsio,
5577 						      /*sks_valid*/ 1,
5578 						      /*command*/ 0,
5579 						      /*field*/ 2,
5580 						      /*bit_valid*/ 0,
5581 						      /*bit*/ 0);
5582 				goto bailout;
5583 			}
5584 		}
5585 	}
5586 
5587 	/*
5588 	 * The format command will clear out the "Medium format corrupted"
5589 	 * status if set by the configuration code.  That status is really
5590 	 * just a way to notify the host that we have lost the media, and
5591 	 * get them to issue a command that will basically make them think
5592 	 * they're blowing away the media.
5593 	 */
5594 	mtx_lock(&ctl_softc->ctl_lock);
5595 	lun->flags &= ~CTL_LUN_INOPERABLE;
5596 	mtx_unlock(&ctl_softc->ctl_lock);
5597 
5598 	ctsio->scsi_status = SCSI_STATUS_OK;
5599 	ctsio->io_hdr.status = CTL_SUCCESS;
5600 bailout:
5601 
5602 	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5603 		free(ctsio->kern_data_ptr, M_CTL);
5604 		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5605 	}
5606 
5607 	ctl_done((union ctl_io *)ctsio);
5608 	return (CTL_RETVAL_COMPLETE);
5609 }
5610 
5611 int
5612 ctl_write_buffer(struct ctl_scsiio *ctsio)
5613 {
5614 	struct scsi_write_buffer *cdb;
5615 	struct copan_page_header *header;
5616 	struct ctl_lun *lun;
5617 	struct ctl_softc *ctl_softc;
5618 	int buffer_offset, len;
5619 	int retval;
5620 
5621 	header = NULL;
5622 
5623 	retval = CTL_RETVAL_COMPLETE;
5624 
5625 	CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5626 
5627 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5628 	ctl_softc = control_softc;
5629 	cdb = (struct scsi_write_buffer *)ctsio->cdb;
5630 
5631 	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5632 		ctl_set_invalid_field(ctsio,
5633 				      /*sks_valid*/ 1,
5634 				      /*command*/ 1,
5635 				      /*field*/ 1,
5636 				      /*bit_valid*/ 1,
5637 				      /*bit*/ 4);
5638 		ctl_done((union ctl_io *)ctsio);
5639 		return (CTL_RETVAL_COMPLETE);
5640 	}
5641 	if (cdb->buffer_id != 0) {
5642 		ctl_set_invalid_field(ctsio,
5643 				      /*sks_valid*/ 1,
5644 				      /*command*/ 1,
5645 				      /*field*/ 2,
5646 				      /*bit_valid*/ 0,
5647 				      /*bit*/ 0);
5648 		ctl_done((union ctl_io *)ctsio);
5649 		return (CTL_RETVAL_COMPLETE);
5650 	}
5651 
5652 	len = scsi_3btoul(cdb->length);
5653 	buffer_offset = scsi_3btoul(cdb->offset);
5654 
5655 	if (len > sizeof(lun->write_buffer)) {
5656 		ctl_set_invalid_field(ctsio,
5657 				      /*sks_valid*/ 1,
5658 				      /*command*/ 1,
5659 				      /*field*/ 6,
5660 				      /*bit_valid*/ 0,
5661 				      /*bit*/ 0);
5662 		ctl_done((union ctl_io *)ctsio);
5663 		return (CTL_RETVAL_COMPLETE);
5664 	}
5665 
5666 	if (buffer_offset != 0) {
5667 		ctl_set_invalid_field(ctsio,
5668 				      /*sks_valid*/ 1,
5669 				      /*command*/ 1,
5670 				      /*field*/ 3,
5671 				      /*bit_valid*/ 0,
5672 				      /*bit*/ 0);
5673 		ctl_done((union ctl_io *)ctsio);
5674 		return (CTL_RETVAL_COMPLETE);
5675 	}
5676 
5677 	/*
5678 	 * If we've got a kernel request that hasn't been malloced yet,
5679 	 * malloc it and tell the caller the data buffer is here.
5680 	 */
5681 	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5682 		ctsio->kern_data_ptr = lun->write_buffer;
5683 		ctsio->kern_data_len = len;
5684 		ctsio->kern_total_len = len;
5685 		ctsio->kern_data_resid = 0;
5686 		ctsio->kern_rel_offset = 0;
5687 		ctsio->kern_sg_entries = 0;
5688 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5689 		ctsio->be_move_done = ctl_config_move_done;
5690 		ctl_datamove((union ctl_io *)ctsio);
5691 
5692 		return (CTL_RETVAL_COMPLETE);
5693 	}
5694 
5695 	ctl_done((union ctl_io *)ctsio);
5696 
5697 	return (CTL_RETVAL_COMPLETE);
5698 }
5699 
5700 /*
5701  * Note that this function currently doesn't actually do anything inside
5702  * CTL to enforce things if the DQue bit is turned on.
5703  *
5704  * Also note that this function can't be used in the default case, because
5705  * the DQue bit isn't set in the changeable mask for the control mode page
5706  * anyway.  This is just here as an example for how to implement a page
5707  * handler, and a placeholder in case we want to allow the user to turn
5708  * tagged queueing on and off.
5709  *
5710  * The D_SENSE bit handling is functional, however, and will turn
5711  * descriptor sense on and off for a given LUN.
5712  */
5713 int
5714 ctl_control_page_handler(struct ctl_scsiio *ctsio,
5715 			 struct ctl_page_index *page_index, uint8_t *page_ptr)
5716 {
5717 	struct scsi_control_page *current_cp, *saved_cp, *user_cp;
5718 	struct ctl_lun *lun;
5719 	struct ctl_softc *softc;
5720 	int set_ua;
5721 	uint32_t initidx;
5722 
5723 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5724 	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5725 	set_ua = 0;
5726 
5727 	user_cp = (struct scsi_control_page *)page_ptr;
5728 	current_cp = (struct scsi_control_page *)
5729 		(page_index->page_data + (page_index->page_len *
5730 		CTL_PAGE_CURRENT));
5731 	saved_cp = (struct scsi_control_page *)
5732 		(page_index->page_data + (page_index->page_len *
5733 		CTL_PAGE_SAVED));
5734 
5735 	softc = control_softc;
5736 
5737 	mtx_lock(&softc->ctl_lock);
5738 	if (((current_cp->rlec & SCP_DSENSE) == 0)
5739 	 && ((user_cp->rlec & SCP_DSENSE) != 0)) {
5740 		/*
5741 		 * Descriptor sense is currently turned off and the user
5742 		 * wants to turn it on.
5743 		 */
5744 		current_cp->rlec |= SCP_DSENSE;
5745 		saved_cp->rlec |= SCP_DSENSE;
5746 		lun->flags |= CTL_LUN_SENSE_DESC;
5747 		set_ua = 1;
5748 	} else if (((current_cp->rlec & SCP_DSENSE) != 0)
5749 		&& ((user_cp->rlec & SCP_DSENSE) == 0)) {
5750 		/*
5751 		 * Descriptor sense is currently turned on, and the user
5752 		 * wants to turn it off.
5753 		 */
5754 		current_cp->rlec &= ~SCP_DSENSE;
5755 		saved_cp->rlec &= ~SCP_DSENSE;
5756 		lun->flags &= ~CTL_LUN_SENSE_DESC;
5757 		set_ua = 1;
5758 	}
5759 	if (current_cp->queue_flags & SCP_QUEUE_DQUE) {
5760 		if (user_cp->queue_flags & SCP_QUEUE_DQUE) {
5761 #ifdef NEEDTOPORT
5762 			csevent_log(CSC_CTL | CSC_SHELF_SW |
5763 				    CTL_UNTAG_TO_UNTAG,
5764 				    csevent_LogType_Trace,
5765 				    csevent_Severity_Information,
5766 				    csevent_AlertLevel_Green,
5767 				    csevent_FRU_Firmware,
5768 				    csevent_FRU_Unknown,
5769 				    "Received untagged to untagged transition");
5770 #endif /* NEEDTOPORT */
5771 		} else {
5772 #ifdef NEEDTOPORT
5773 			csevent_log(CSC_CTL | CSC_SHELF_SW |
5774 				    CTL_UNTAG_TO_TAG,
5775 				    csevent_LogType_ConfigChange,
5776 				    csevent_Severity_Information,
5777 				    csevent_AlertLevel_Green,
5778 				    csevent_FRU_Firmware,
5779 				    csevent_FRU_Unknown,
5780 				    "Received untagged to tagged "
5781 				    "queueing transition");
5782 #endif /* NEEDTOPORT */
5783 
5784 			current_cp->queue_flags &= ~SCP_QUEUE_DQUE;
5785 			saved_cp->queue_flags &= ~SCP_QUEUE_DQUE;
5786 			set_ua = 1;
5787 		}
5788 	} else {
5789 		if (user_cp->queue_flags & SCP_QUEUE_DQUE) {
5790 #ifdef NEEDTOPORT
5791 			csevent_log(CSC_CTL | CSC_SHELF_SW |
5792 				    CTL_TAG_TO_UNTAG,
5793 				    csevent_LogType_ConfigChange,
5794 				    csevent_Severity_Warning,
5795 				    csevent_AlertLevel_Yellow,
5796 				    csevent_FRU_Firmware,
5797 				    csevent_FRU_Unknown,
5798 				    "Received tagged queueing to untagged "
5799 				    "transition");
5800 #endif /* NEEDTOPORT */
5801 
5802 			current_cp->queue_flags |= SCP_QUEUE_DQUE;
5803 			saved_cp->queue_flags |= SCP_QUEUE_DQUE;
5804 			set_ua = 1;
5805 		} else {
5806 #ifdef NEEDTOPORT
5807 			csevent_log(CSC_CTL | CSC_SHELF_SW |
5808 				    CTL_TAG_TO_TAG,
5809 				    csevent_LogType_Trace,
5810 				    csevent_Severity_Information,
5811 				    csevent_AlertLevel_Green,
5812 				    csevent_FRU_Firmware,
5813 				    csevent_FRU_Unknown,
5814 				    "Received tagged queueing to tagged "
5815 				    "queueing transition");
5816 #endif /* NEEDTOPORT */
5817 		}
5818 	}
5819 	if (set_ua != 0) {
5820 		int i;
5821 		/*
5822 		 * Let other initiators know that the mode
5823 		 * parameters for this LUN have changed.
5824 		 */
5825 		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
5826 			if (i == initidx)
5827 				continue;
5828 
5829 			lun->pending_sense[i].ua_pending |=
5830 				CTL_UA_MODE_CHANGE;
5831 		}
5832 	}
5833 	mtx_unlock(&softc->ctl_lock);
5834 
5835 	return (0);
5836 }
5837 
5838 int
5839 ctl_power_sp_handler(struct ctl_scsiio *ctsio,
5840 		     struct ctl_page_index *page_index, uint8_t *page_ptr)
5841 {
5842 	return (0);
5843 }
5844 
5845 int
5846 ctl_power_sp_sense_handler(struct ctl_scsiio *ctsio,
5847 			   struct ctl_page_index *page_index, int pc)
5848 {
5849 	struct copan_power_subpage *page;
5850 
5851 	page = (struct copan_power_subpage *)page_index->page_data +
5852 		(page_index->page_len * pc);
5853 
5854 	switch (pc) {
5855 	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
5856 		/*
5857 		 * We don't update the changable bits for this page.
5858 		 */
5859 		break;
5860 	case SMS_PAGE_CTRL_CURRENT >> 6:
5861 	case SMS_PAGE_CTRL_DEFAULT >> 6:
5862 	case SMS_PAGE_CTRL_SAVED >> 6:
5863 #ifdef NEEDTOPORT
5864 		ctl_update_power_subpage(page);
5865 #endif
5866 		break;
5867 	default:
5868 #ifdef NEEDTOPORT
5869 		EPRINT(0, "Invalid PC %d!!", pc);
5870 #endif
5871 		break;
5872 	}
5873 	return (0);
5874 }
5875 
5876 
5877 int
5878 ctl_aps_sp_handler(struct ctl_scsiio *ctsio,
5879 		   struct ctl_page_index *page_index, uint8_t *page_ptr)
5880 {
5881 	struct copan_aps_subpage *user_sp;
5882 	struct copan_aps_subpage *current_sp;
5883 	union ctl_modepage_info *modepage_info;
5884 	struct ctl_softc *softc;
5885 	struct ctl_lun *lun;
5886 	int retval;
5887 
5888 	retval = CTL_RETVAL_COMPLETE;
5889 	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
5890 		     (page_index->page_len * CTL_PAGE_CURRENT));
5891 	softc = control_softc;
5892 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5893 
5894 	user_sp = (struct copan_aps_subpage *)page_ptr;
5895 
5896 	modepage_info = (union ctl_modepage_info *)
5897 		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
5898 
5899 	modepage_info->header.page_code = page_index->page_code & SMPH_PC_MASK;
5900 	modepage_info->header.subpage = page_index->subpage;
5901 	modepage_info->aps.lock_active = user_sp->lock_active;
5902 
5903 	mtx_lock(&softc->ctl_lock);
5904 
5905 	/*
5906 	 * If there is a request to lock the LUN and another LUN is locked
5907 	 * this is an error. If the requested LUN is already locked ignore
5908 	 * the request. If no LUN is locked attempt to lock it.
5909 	 * if there is a request to unlock the LUN and the LUN is currently
5910 	 * locked attempt to unlock it. Otherwise ignore the request. i.e.
5911 	 * if another LUN is locked or no LUN is locked.
5912 	 */
5913 	if (user_sp->lock_active & APS_LOCK_ACTIVE) {
5914 		if (softc->aps_locked_lun == lun->lun) {
5915 			/*
5916 			 * This LUN is already locked, so we're done.
5917 			 */
5918 			retval = CTL_RETVAL_COMPLETE;
5919 		} else if (softc->aps_locked_lun == 0) {
5920 			/*
5921 			 * No one has the lock, pass the request to the
5922 			 * backend.
5923 			 */
5924 			retval = lun->backend->config_write(
5925 				(union ctl_io *)ctsio);
5926 		} else {
5927 			/*
5928 			 * Someone else has the lock, throw out the request.
5929 			 */
5930 			ctl_set_already_locked(ctsio);
5931 			free(ctsio->kern_data_ptr, M_CTL);
5932 			ctl_done((union ctl_io *)ctsio);
5933 
5934 			/*
5935 			 * Set the return value so that ctl_do_mode_select()
5936 			 * won't try to complete the command.  We already
5937 			 * completed it here.
5938 			 */
5939 			retval = CTL_RETVAL_ERROR;
5940 		}
5941 	} else if (softc->aps_locked_lun == lun->lun) {
5942 		/*
5943 		 * This LUN is locked, so pass the unlock request to the
5944 		 * backend.
5945 		 */
5946 		retval = lun->backend->config_write((union ctl_io *)ctsio);
5947 	}
5948 	mtx_unlock(&softc->ctl_lock);
5949 
5950 	return (retval);
5951 }
5952 
5953 int
5954 ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio,
5955 				struct ctl_page_index *page_index,
5956 				uint8_t *page_ptr)
5957 {
5958 	uint8_t *c;
5959 	int i;
5960 
5961 	c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs;
5962 	ctl_time_io_secs =
5963 		(c[0] << 8) |
5964 		(c[1] << 0) |
5965 		0;
5966 	CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs));
5967 	printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs);
5968 	printf("page data:");
5969 	for (i=0; i<8; i++)
5970 		printf(" %.2x",page_ptr[i]);
5971 	printf("\n");
5972 	return (0);
5973 }
5974 
5975 int
5976 ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio,
5977 			       struct ctl_page_index *page_index,
5978 			       int pc)
5979 {
5980 	struct copan_debugconf_subpage *page;
5981 
5982 	page = (struct copan_debugconf_subpage *)page_index->page_data +
5983 		(page_index->page_len * pc);
5984 
5985 	switch (pc) {
5986 	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
5987 	case SMS_PAGE_CTRL_DEFAULT >> 6:
5988 	case SMS_PAGE_CTRL_SAVED >> 6:
5989 		/*
5990 		 * We don't update the changable or default bits for this page.
5991 		 */
5992 		break;
5993 	case SMS_PAGE_CTRL_CURRENT >> 6:
5994 		page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8;
5995 		page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0;
5996 		break;
5997 	default:
5998 #ifdef NEEDTOPORT
5999 		EPRINT(0, "Invalid PC %d!!", pc);
6000 #endif /* NEEDTOPORT */
6001 		break;
6002 	}
6003 	return (0);
6004 }
6005 
6006 
6007 static int
6008 ctl_do_mode_select(union ctl_io *io)
6009 {
6010 	struct scsi_mode_page_header *page_header;
6011 	struct ctl_page_index *page_index;
6012 	struct ctl_scsiio *ctsio;
6013 	int control_dev, page_len;
6014 	int page_len_offset, page_len_size;
6015 	union ctl_modepage_info *modepage_info;
6016 	struct ctl_lun *lun;
6017 	int *len_left, *len_used;
6018 	int retval, i;
6019 
6020 	ctsio = &io->scsiio;
6021 	page_index = NULL;
6022 	page_len = 0;
6023 	retval = CTL_RETVAL_COMPLETE;
6024 
6025 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6026 
6027 	if (lun->be_lun->lun_type != T_DIRECT)
6028 		control_dev = 1;
6029 	else
6030 		control_dev = 0;
6031 
6032 	modepage_info = (union ctl_modepage_info *)
6033 		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6034 	len_left = &modepage_info->header.len_left;
6035 	len_used = &modepage_info->header.len_used;
6036 
6037 do_next_page:
6038 
6039 	page_header = (struct scsi_mode_page_header *)
6040 		(ctsio->kern_data_ptr + *len_used);
6041 
6042 	if (*len_left == 0) {
6043 		free(ctsio->kern_data_ptr, M_CTL);
6044 		ctl_set_success(ctsio);
6045 		ctl_done((union ctl_io *)ctsio);
6046 		return (CTL_RETVAL_COMPLETE);
6047 	} else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6048 
6049 		free(ctsio->kern_data_ptr, M_CTL);
6050 		ctl_set_param_len_error(ctsio);
6051 		ctl_done((union ctl_io *)ctsio);
6052 		return (CTL_RETVAL_COMPLETE);
6053 
6054 	} else if ((page_header->page_code & SMPH_SPF)
6055 		&& (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6056 
6057 		free(ctsio->kern_data_ptr, M_CTL);
6058 		ctl_set_param_len_error(ctsio);
6059 		ctl_done((union ctl_io *)ctsio);
6060 		return (CTL_RETVAL_COMPLETE);
6061 	}
6062 
6063 
6064 	/*
6065 	 * XXX KDM should we do something with the block descriptor?
6066 	 */
6067 	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6068 
6069 		if ((control_dev != 0)
6070 		 && (lun->mode_pages.index[i].page_flags &
6071 		     CTL_PAGE_FLAG_DISK_ONLY))
6072 			continue;
6073 
6074 		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
6075 		    (page_header->page_code & SMPH_PC_MASK))
6076 			continue;
6077 
6078 		/*
6079 		 * If neither page has a subpage code, then we've got a
6080 		 * match.
6081 		 */
6082 		if (((lun->mode_pages.index[i].page_code & SMPH_SPF) == 0)
6083 		 && ((page_header->page_code & SMPH_SPF) == 0)) {
6084 			page_index = &lun->mode_pages.index[i];
6085 			page_len = page_header->page_length;
6086 			break;
6087 		}
6088 
6089 		/*
6090 		 * If both pages have subpages, then the subpage numbers
6091 		 * have to match.
6092 		 */
6093 		if ((lun->mode_pages.index[i].page_code & SMPH_SPF)
6094 		  && (page_header->page_code & SMPH_SPF)) {
6095 			struct scsi_mode_page_header_sp *sph;
6096 
6097 			sph = (struct scsi_mode_page_header_sp *)page_header;
6098 
6099 			if (lun->mode_pages.index[i].subpage ==
6100 			    sph->subpage) {
6101 				page_index = &lun->mode_pages.index[i];
6102 				page_len = scsi_2btoul(sph->page_length);
6103 				break;
6104 			}
6105 		}
6106 	}
6107 
6108 	/*
6109 	 * If we couldn't find the page, or if we don't have a mode select
6110 	 * handler for it, send back an error to the user.
6111 	 */
6112 	if ((page_index == NULL)
6113 	 || (page_index->select_handler == NULL)) {
6114 		ctl_set_invalid_field(ctsio,
6115 				      /*sks_valid*/ 1,
6116 				      /*command*/ 0,
6117 				      /*field*/ *len_used,
6118 				      /*bit_valid*/ 0,
6119 				      /*bit*/ 0);
6120 		free(ctsio->kern_data_ptr, M_CTL);
6121 		ctl_done((union ctl_io *)ctsio);
6122 		return (CTL_RETVAL_COMPLETE);
6123 	}
6124 
6125 	if (page_index->page_code & SMPH_SPF) {
6126 		page_len_offset = 2;
6127 		page_len_size = 2;
6128 	} else {
6129 		page_len_size = 1;
6130 		page_len_offset = 1;
6131 	}
6132 
6133 	/*
6134 	 * If the length the initiator gives us isn't the one we specify in
6135 	 * the mode page header, or if they didn't specify enough data in
6136 	 * the CDB to avoid truncating this page, kick out the request.
6137 	 */
6138 	if ((page_len != (page_index->page_len - page_len_offset -
6139 			  page_len_size))
6140 	 || (*len_left < page_index->page_len)) {
6141 
6142 
6143 		ctl_set_invalid_field(ctsio,
6144 				      /*sks_valid*/ 1,
6145 				      /*command*/ 0,
6146 				      /*field*/ *len_used + page_len_offset,
6147 				      /*bit_valid*/ 0,
6148 				      /*bit*/ 0);
6149 		free(ctsio->kern_data_ptr, M_CTL);
6150 		ctl_done((union ctl_io *)ctsio);
6151 		return (CTL_RETVAL_COMPLETE);
6152 	}
6153 
6154 	/*
6155 	 * Run through the mode page, checking to make sure that the bits
6156 	 * the user changed are actually legal for him to change.
6157 	 */
6158 	for (i = 0; i < page_index->page_len; i++) {
6159 		uint8_t *user_byte, *change_mask, *current_byte;
6160 		int bad_bit;
6161 		int j;
6162 
6163 		user_byte = (uint8_t *)page_header + i;
6164 		change_mask = page_index->page_data +
6165 			      (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6166 		current_byte = page_index->page_data +
6167 			       (page_index->page_len * CTL_PAGE_CURRENT) + i;
6168 
6169 		/*
6170 		 * Check to see whether the user set any bits in this byte
6171 		 * that he is not allowed to set.
6172 		 */
6173 		if ((*user_byte & ~(*change_mask)) ==
6174 		    (*current_byte & ~(*change_mask)))
6175 			continue;
6176 
6177 		/*
6178 		 * Go through bit by bit to determine which one is illegal.
6179 		 */
6180 		bad_bit = 0;
6181 		for (j = 7; j >= 0; j--) {
6182 			if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6183 			    (((1 << i) & ~(*change_mask)) & *current_byte)) {
6184 				bad_bit = i;
6185 				break;
6186 			}
6187 		}
6188 		ctl_set_invalid_field(ctsio,
6189 				      /*sks_valid*/ 1,
6190 				      /*command*/ 0,
6191 				      /*field*/ *len_used + i,
6192 				      /*bit_valid*/ 1,
6193 				      /*bit*/ bad_bit);
6194 		free(ctsio->kern_data_ptr, M_CTL);
6195 		ctl_done((union ctl_io *)ctsio);
6196 		return (CTL_RETVAL_COMPLETE);
6197 	}
6198 
6199 	/*
6200 	 * Decrement these before we call the page handler, since we may
6201 	 * end up getting called back one way or another before the handler
6202 	 * returns to this context.
6203 	 */
6204 	*len_left -= page_index->page_len;
6205 	*len_used += page_index->page_len;
6206 
6207 	retval = page_index->select_handler(ctsio, page_index,
6208 					    (uint8_t *)page_header);
6209 
6210 	/*
6211 	 * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6212 	 * wait until this queued command completes to finish processing
6213 	 * the mode page.  If it returns anything other than
6214 	 * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6215 	 * already set the sense information, freed the data pointer, and
6216 	 * completed the io for us.
6217 	 */
6218 	if (retval != CTL_RETVAL_COMPLETE)
6219 		goto bailout_no_done;
6220 
6221 	/*
6222 	 * If the initiator sent us more than one page, parse the next one.
6223 	 */
6224 	if (*len_left > 0)
6225 		goto do_next_page;
6226 
6227 	ctl_set_success(ctsio);
6228 	free(ctsio->kern_data_ptr, M_CTL);
6229 	ctl_done((union ctl_io *)ctsio);
6230 
6231 bailout_no_done:
6232 
6233 	return (CTL_RETVAL_COMPLETE);
6234 
6235 }
6236 
6237 int
6238 ctl_mode_select(struct ctl_scsiio *ctsio)
6239 {
6240 	int param_len, pf, sp;
6241 	int header_size, bd_len;
6242 	int len_left, len_used;
6243 	struct ctl_page_index *page_index;
6244 	struct ctl_lun *lun;
6245 	int control_dev, page_len;
6246 	union ctl_modepage_info *modepage_info;
6247 	int retval;
6248 
6249 	pf = 0;
6250 	sp = 0;
6251 	page_len = 0;
6252 	len_used = 0;
6253 	len_left = 0;
6254 	retval = 0;
6255 	bd_len = 0;
6256 	page_index = NULL;
6257 
6258 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6259 
6260 	if (lun->be_lun->lun_type != T_DIRECT)
6261 		control_dev = 1;
6262 	else
6263 		control_dev = 0;
6264 
6265 	switch (ctsio->cdb[0]) {
6266 	case MODE_SELECT_6: {
6267 		struct scsi_mode_select_6 *cdb;
6268 
6269 		cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6270 
6271 		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6272 		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6273 
6274 		param_len = cdb->length;
6275 		header_size = sizeof(struct scsi_mode_header_6);
6276 		break;
6277 	}
6278 	case MODE_SELECT_10: {
6279 		struct scsi_mode_select_10 *cdb;
6280 
6281 		cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6282 
6283 		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6284 		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6285 
6286 		param_len = scsi_2btoul(cdb->length);
6287 		header_size = sizeof(struct scsi_mode_header_10);
6288 		break;
6289 	}
6290 	default:
6291 		ctl_set_invalid_opcode(ctsio);
6292 		ctl_done((union ctl_io *)ctsio);
6293 		return (CTL_RETVAL_COMPLETE);
6294 		break; /* NOTREACHED */
6295 	}
6296 
6297 	/*
6298 	 * From SPC-3:
6299 	 * "A parameter list length of zero indicates that the Data-Out Buffer
6300 	 * shall be empty. This condition shall not be considered as an error."
6301 	 */
6302 	if (param_len == 0) {
6303 		ctl_set_success(ctsio);
6304 		ctl_done((union ctl_io *)ctsio);
6305 		return (CTL_RETVAL_COMPLETE);
6306 	}
6307 
6308 	/*
6309 	 * Since we'll hit this the first time through, prior to
6310 	 * allocation, we don't need to free a data buffer here.
6311 	 */
6312 	if (param_len < header_size) {
6313 		ctl_set_param_len_error(ctsio);
6314 		ctl_done((union ctl_io *)ctsio);
6315 		return (CTL_RETVAL_COMPLETE);
6316 	}
6317 
6318 	/*
6319 	 * Allocate the data buffer and grab the user's data.  In theory,
6320 	 * we shouldn't have to sanity check the parameter list length here
6321 	 * because the maximum size is 64K.  We should be able to malloc
6322 	 * that much without too many problems.
6323 	 */
6324 	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6325 		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6326 		if (ctsio->kern_data_ptr == NULL) {
6327 			ctl_set_busy(ctsio);
6328 			ctl_done((union ctl_io *)ctsio);
6329 			return (CTL_RETVAL_COMPLETE);
6330 		}
6331 		ctsio->kern_data_len = param_len;
6332 		ctsio->kern_total_len = param_len;
6333 		ctsio->kern_data_resid = 0;
6334 		ctsio->kern_rel_offset = 0;
6335 		ctsio->kern_sg_entries = 0;
6336 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6337 		ctsio->be_move_done = ctl_config_move_done;
6338 		ctl_datamove((union ctl_io *)ctsio);
6339 
6340 		return (CTL_RETVAL_COMPLETE);
6341 	}
6342 
6343 	switch (ctsio->cdb[0]) {
6344 	case MODE_SELECT_6: {
6345 		struct scsi_mode_header_6 *mh6;
6346 
6347 		mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6348 		bd_len = mh6->blk_desc_len;
6349 		break;
6350 	}
6351 	case MODE_SELECT_10: {
6352 		struct scsi_mode_header_10 *mh10;
6353 
6354 		mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6355 		bd_len = scsi_2btoul(mh10->blk_desc_len);
6356 		break;
6357 	}
6358 	default:
6359 		panic("Invalid CDB type %#x", ctsio->cdb[0]);
6360 		break;
6361 	}
6362 
6363 	if (param_len < (header_size + bd_len)) {
6364 		free(ctsio->kern_data_ptr, M_CTL);
6365 		ctl_set_param_len_error(ctsio);
6366 		ctl_done((union ctl_io *)ctsio);
6367 		return (CTL_RETVAL_COMPLETE);
6368 	}
6369 
6370 	/*
6371 	 * Set the IO_CONT flag, so that if this I/O gets passed to
6372 	 * ctl_config_write_done(), it'll get passed back to
6373 	 * ctl_do_mode_select() for further processing, or completion if
6374 	 * we're all done.
6375 	 */
6376 	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6377 	ctsio->io_cont = ctl_do_mode_select;
6378 
6379 	modepage_info = (union ctl_modepage_info *)
6380 		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6381 
6382 	memset(modepage_info, 0, sizeof(*modepage_info));
6383 
6384 	len_left = param_len - header_size - bd_len;
6385 	len_used = header_size + bd_len;
6386 
6387 	modepage_info->header.len_left = len_left;
6388 	modepage_info->header.len_used = len_used;
6389 
6390 	return (ctl_do_mode_select((union ctl_io *)ctsio));
6391 }
6392 
6393 int
6394 ctl_mode_sense(struct ctl_scsiio *ctsio)
6395 {
6396 	struct ctl_lun *lun;
6397 	int pc, page_code, dbd, llba, subpage;
6398 	int alloc_len, page_len, header_len, total_len;
6399 	struct scsi_mode_block_descr *block_desc;
6400 	struct ctl_page_index *page_index;
6401 	int control_dev;
6402 
6403 	dbd = 0;
6404 	llba = 0;
6405 	block_desc = NULL;
6406 	page_index = NULL;
6407 
6408 	CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6409 
6410 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6411 
6412 	if (lun->be_lun->lun_type != T_DIRECT)
6413 		control_dev = 1;
6414 	else
6415 		control_dev = 0;
6416 
6417 	switch (ctsio->cdb[0]) {
6418 	case MODE_SENSE_6: {
6419 		struct scsi_mode_sense_6 *cdb;
6420 
6421 		cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6422 
6423 		header_len = sizeof(struct scsi_mode_hdr_6);
6424 		if (cdb->byte2 & SMS_DBD)
6425 			dbd = 1;
6426 		else
6427 			header_len += sizeof(struct scsi_mode_block_descr);
6428 
6429 		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6430 		page_code = cdb->page & SMS_PAGE_CODE;
6431 		subpage = cdb->subpage;
6432 		alloc_len = cdb->length;
6433 		break;
6434 	}
6435 	case MODE_SENSE_10: {
6436 		struct scsi_mode_sense_10 *cdb;
6437 
6438 		cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6439 
6440 		header_len = sizeof(struct scsi_mode_hdr_10);
6441 
6442 		if (cdb->byte2 & SMS_DBD)
6443 			dbd = 1;
6444 		else
6445 			header_len += sizeof(struct scsi_mode_block_descr);
6446 		if (cdb->byte2 & SMS10_LLBAA)
6447 			llba = 1;
6448 		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6449 		page_code = cdb->page & SMS_PAGE_CODE;
6450 		subpage = cdb->subpage;
6451 		alloc_len = scsi_2btoul(cdb->length);
6452 		break;
6453 	}
6454 	default:
6455 		ctl_set_invalid_opcode(ctsio);
6456 		ctl_done((union ctl_io *)ctsio);
6457 		return (CTL_RETVAL_COMPLETE);
6458 		break; /* NOTREACHED */
6459 	}
6460 
6461 	/*
6462 	 * We have to make a first pass through to calculate the size of
6463 	 * the pages that match the user's query.  Then we allocate enough
6464 	 * memory to hold it, and actually copy the data into the buffer.
6465 	 */
6466 	switch (page_code) {
6467 	case SMS_ALL_PAGES_PAGE: {
6468 		int i;
6469 
6470 		page_len = 0;
6471 
6472 		/*
6473 		 * At the moment, values other than 0 and 0xff here are
6474 		 * reserved according to SPC-3.
6475 		 */
6476 		if ((subpage != SMS_SUBPAGE_PAGE_0)
6477 		 && (subpage != SMS_SUBPAGE_ALL)) {
6478 			ctl_set_invalid_field(ctsio,
6479 					      /*sks_valid*/ 1,
6480 					      /*command*/ 1,
6481 					      /*field*/ 3,
6482 					      /*bit_valid*/ 0,
6483 					      /*bit*/ 0);
6484 			ctl_done((union ctl_io *)ctsio);
6485 			return (CTL_RETVAL_COMPLETE);
6486 		}
6487 
6488 		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6489 			if ((control_dev != 0)
6490 			 && (lun->mode_pages.index[i].page_flags &
6491 			     CTL_PAGE_FLAG_DISK_ONLY))
6492 				continue;
6493 
6494 			/*
6495 			 * We don't use this subpage if the user didn't
6496 			 * request all subpages.
6497 			 */
6498 			if ((lun->mode_pages.index[i].subpage != 0)
6499 			 && (subpage == SMS_SUBPAGE_PAGE_0))
6500 				continue;
6501 
6502 #if 0
6503 			printf("found page %#x len %d\n",
6504 			       lun->mode_pages.index[i].page_code &
6505 			       SMPH_PC_MASK,
6506 			       lun->mode_pages.index[i].page_len);
6507 #endif
6508 			page_len += lun->mode_pages.index[i].page_len;
6509 		}
6510 		break;
6511 	}
6512 	default: {
6513 		int i;
6514 
6515 		page_len = 0;
6516 
6517 		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6518 			/* Look for the right page code */
6519 			if ((lun->mode_pages.index[i].page_code &
6520 			     SMPH_PC_MASK) != page_code)
6521 				continue;
6522 
6523 			/* Look for the right subpage or the subpage wildcard*/
6524 			if ((lun->mode_pages.index[i].subpage != subpage)
6525 			 && (subpage != SMS_SUBPAGE_ALL))
6526 				continue;
6527 
6528 			/* Make sure the page is supported for this dev type */
6529 			if ((control_dev != 0)
6530 			 && (lun->mode_pages.index[i].page_flags &
6531 			     CTL_PAGE_FLAG_DISK_ONLY))
6532 				continue;
6533 
6534 #if 0
6535 			printf("found page %#x len %d\n",
6536 			       lun->mode_pages.index[i].page_code &
6537 			       SMPH_PC_MASK,
6538 			       lun->mode_pages.index[i].page_len);
6539 #endif
6540 
6541 			page_len += lun->mode_pages.index[i].page_len;
6542 		}
6543 
6544 		if (page_len == 0) {
6545 			ctl_set_invalid_field(ctsio,
6546 					      /*sks_valid*/ 1,
6547 					      /*command*/ 1,
6548 					      /*field*/ 2,
6549 					      /*bit_valid*/ 1,
6550 					      /*bit*/ 5);
6551 			ctl_done((union ctl_io *)ctsio);
6552 			return (CTL_RETVAL_COMPLETE);
6553 		}
6554 		break;
6555 	}
6556 	}
6557 
6558 	total_len = header_len + page_len;
6559 #if 0
6560 	printf("header_len = %d, page_len = %d, total_len = %d\n",
6561 	       header_len, page_len, total_len);
6562 #endif
6563 
6564 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK);
6565 	if (ctsio->kern_data_ptr == NULL) {
6566 		ctsio->io_hdr.status = CTL_SCSI_ERROR;
6567 		ctsio->scsi_status = SCSI_STATUS_BUSY;
6568 		ctl_done((union ctl_io *)ctsio);
6569 		return (CTL_RETVAL_COMPLETE);
6570 	}
6571 	ctsio->kern_sg_entries = 0;
6572 	ctsio->kern_data_resid = 0;
6573 	ctsio->kern_rel_offset = 0;
6574 	if (total_len < alloc_len) {
6575 		ctsio->residual = alloc_len - total_len;
6576 		ctsio->kern_data_len = total_len;
6577 		ctsio->kern_total_len = total_len;
6578 	} else {
6579 		ctsio->residual = 0;
6580 		ctsio->kern_data_len = alloc_len;
6581 		ctsio->kern_total_len = alloc_len;
6582 	}
6583 	memset(ctsio->kern_data_ptr, 0, total_len);
6584 
6585 	switch (ctsio->cdb[0]) {
6586 	case MODE_SENSE_6: {
6587 		struct scsi_mode_hdr_6 *header;
6588 
6589 		header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6590 
6591 		header->datalen = ctl_min(total_len - 1, 254);
6592 
6593 		if (dbd)
6594 			header->block_descr_len = 0;
6595 		else
6596 			header->block_descr_len =
6597 				sizeof(struct scsi_mode_block_descr);
6598 		block_desc = (struct scsi_mode_block_descr *)&header[1];
6599 		break;
6600 	}
6601 	case MODE_SENSE_10: {
6602 		struct scsi_mode_hdr_10 *header;
6603 		int datalen;
6604 
6605 		header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6606 
6607 		datalen = ctl_min(total_len - 2, 65533);
6608 		scsi_ulto2b(datalen, header->datalen);
6609 		if (dbd)
6610 			scsi_ulto2b(0, header->block_descr_len);
6611 		else
6612 			scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
6613 				    header->block_descr_len);
6614 		block_desc = (struct scsi_mode_block_descr *)&header[1];
6615 		break;
6616 	}
6617 	default:
6618 		panic("invalid CDB type %#x", ctsio->cdb[0]);
6619 		break; /* NOTREACHED */
6620 	}
6621 
6622 	/*
6623 	 * If we've got a disk, use its blocksize in the block
6624 	 * descriptor.  Otherwise, just set it to 0.
6625 	 */
6626 	if (dbd == 0) {
6627 		if (control_dev != 0)
6628 			scsi_ulto3b(lun->be_lun->blocksize,
6629 				    block_desc->block_len);
6630 		else
6631 			scsi_ulto3b(0, block_desc->block_len);
6632 	}
6633 
6634 	switch (page_code) {
6635 	case SMS_ALL_PAGES_PAGE: {
6636 		int i, data_used;
6637 
6638 		data_used = header_len;
6639 		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6640 			struct ctl_page_index *page_index;
6641 
6642 			page_index = &lun->mode_pages.index[i];
6643 
6644 			if ((control_dev != 0)
6645 			 && (page_index->page_flags &
6646 			    CTL_PAGE_FLAG_DISK_ONLY))
6647 				continue;
6648 
6649 			/*
6650 			 * We don't use this subpage if the user didn't
6651 			 * request all subpages.  We already checked (above)
6652 			 * to make sure the user only specified a subpage
6653 			 * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
6654 			 */
6655 			if ((page_index->subpage != 0)
6656 			 && (subpage == SMS_SUBPAGE_PAGE_0))
6657 				continue;
6658 
6659 			/*
6660 			 * Call the handler, if it exists, to update the
6661 			 * page to the latest values.
6662 			 */
6663 			if (page_index->sense_handler != NULL)
6664 				page_index->sense_handler(ctsio, page_index,pc);
6665 
6666 			memcpy(ctsio->kern_data_ptr + data_used,
6667 			       page_index->page_data +
6668 			       (page_index->page_len * pc),
6669 			       page_index->page_len);
6670 			data_used += page_index->page_len;
6671 		}
6672 		break;
6673 	}
6674 	default: {
6675 		int i, data_used;
6676 
6677 		data_used = header_len;
6678 
6679 		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6680 			struct ctl_page_index *page_index;
6681 
6682 			page_index = &lun->mode_pages.index[i];
6683 
6684 			/* Look for the right page code */
6685 			if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6686 				continue;
6687 
6688 			/* Look for the right subpage or the subpage wildcard*/
6689 			if ((page_index->subpage != subpage)
6690 			 && (subpage != SMS_SUBPAGE_ALL))
6691 				continue;
6692 
6693 			/* Make sure the page is supported for this dev type */
6694 			if ((control_dev != 0)
6695 			 && (page_index->page_flags &
6696 			     CTL_PAGE_FLAG_DISK_ONLY))
6697 				continue;
6698 
6699 			/*
6700 			 * Call the handler, if it exists, to update the
6701 			 * page to the latest values.
6702 			 */
6703 			if (page_index->sense_handler != NULL)
6704 				page_index->sense_handler(ctsio, page_index,pc);
6705 
6706 			memcpy(ctsio->kern_data_ptr + data_used,
6707 			       page_index->page_data +
6708 			       (page_index->page_len * pc),
6709 			       page_index->page_len);
6710 			data_used += page_index->page_len;
6711 		}
6712 		break;
6713 	}
6714 	}
6715 
6716 	ctsio->scsi_status = SCSI_STATUS_OK;
6717 
6718 	ctsio->be_move_done = ctl_config_move_done;
6719 	ctl_datamove((union ctl_io *)ctsio);
6720 
6721 	return (CTL_RETVAL_COMPLETE);
6722 }
6723 
6724 int
6725 ctl_read_capacity(struct ctl_scsiio *ctsio)
6726 {
6727 	struct scsi_read_capacity *cdb;
6728 	struct scsi_read_capacity_data *data;
6729 	struct ctl_lun *lun;
6730 	uint32_t lba;
6731 
6732 	CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
6733 
6734 	cdb = (struct scsi_read_capacity *)ctsio->cdb;
6735 
6736 	lba = scsi_4btoul(cdb->addr);
6737 	if (((cdb->pmi & SRC_PMI) == 0)
6738 	 && (lba != 0)) {
6739 		ctl_set_invalid_field(/*ctsio*/ ctsio,
6740 				      /*sks_valid*/ 1,
6741 				      /*command*/ 1,
6742 				      /*field*/ 2,
6743 				      /*bit_valid*/ 0,
6744 				      /*bit*/ 0);
6745 		ctl_done((union ctl_io *)ctsio);
6746 		return (CTL_RETVAL_COMPLETE);
6747 	}
6748 
6749 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6750 
6751 	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK);
6752 	if (ctsio->kern_data_ptr == NULL) {
6753 		ctsio->io_hdr.status = CTL_SCSI_ERROR;
6754 		ctsio->scsi_status = SCSI_STATUS_BUSY;
6755 		ctl_done((union ctl_io *)ctsio);
6756 		return (CTL_RETVAL_COMPLETE);
6757 	}
6758 	data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
6759 	ctsio->residual = 0;
6760 	ctsio->kern_data_len = sizeof(*data);
6761 	ctsio->kern_total_len = sizeof(*data);
6762 	ctsio->kern_data_resid = 0;
6763 	ctsio->kern_rel_offset = 0;
6764 	ctsio->kern_sg_entries = 0;
6765 
6766 	memset(data, 0, sizeof(*data));
6767 
6768 	/*
6769 	 * If the maximum LBA is greater than 0xfffffffe, the user must
6770 	 * issue a SERVICE ACTION IN (16) command, with the read capacity
6771 	 * serivce action set.
6772 	 */
6773 	if (lun->be_lun->maxlba > 0xfffffffe)
6774 		scsi_ulto4b(0xffffffff, data->addr);
6775 	else
6776 		scsi_ulto4b(lun->be_lun->maxlba, data->addr);
6777 
6778 	/*
6779 	 * XXX KDM this may not be 512 bytes...
6780 	 */
6781 	scsi_ulto4b(lun->be_lun->blocksize, data->length);
6782 
6783 	ctsio->scsi_status = SCSI_STATUS_OK;
6784 
6785 	ctsio->be_move_done = ctl_config_move_done;
6786 	ctl_datamove((union ctl_io *)ctsio);
6787 
6788 	return (CTL_RETVAL_COMPLETE);
6789 }
6790 
6791 static int
6792 ctl_read_capacity_16(struct ctl_scsiio *ctsio)
6793 {
6794 	struct scsi_read_capacity_16 *cdb;
6795 	struct scsi_read_capacity_data_long *data;
6796 	struct ctl_lun *lun;
6797 	uint64_t lba;
6798 	uint32_t alloc_len;
6799 
6800 	CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
6801 
6802 	cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
6803 
6804 	alloc_len = scsi_4btoul(cdb->alloc_len);
6805 	lba = scsi_8btou64(cdb->addr);
6806 
6807 	if ((cdb->reladr & SRC16_PMI)
6808 	 && (lba != 0)) {
6809 		ctl_set_invalid_field(/*ctsio*/ ctsio,
6810 				      /*sks_valid*/ 1,
6811 				      /*command*/ 1,
6812 				      /*field*/ 2,
6813 				      /*bit_valid*/ 0,
6814 				      /*bit*/ 0);
6815 		ctl_done((union ctl_io *)ctsio);
6816 		return (CTL_RETVAL_COMPLETE);
6817 	}
6818 
6819 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6820 
6821 	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK);
6822 	if (ctsio->kern_data_ptr == NULL) {
6823 		ctsio->io_hdr.status = CTL_SCSI_ERROR;
6824 		ctsio->scsi_status = SCSI_STATUS_BUSY;
6825 		ctl_done((union ctl_io *)ctsio);
6826 		return (CTL_RETVAL_COMPLETE);
6827 	}
6828 	data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
6829 
6830 	if (sizeof(*data) < alloc_len) {
6831 		ctsio->residual = alloc_len - sizeof(*data);
6832 		ctsio->kern_data_len = sizeof(*data);
6833 		ctsio->kern_total_len = sizeof(*data);
6834 	} else {
6835 		ctsio->residual = 0;
6836 		ctsio->kern_data_len = alloc_len;
6837 		ctsio->kern_total_len = alloc_len;
6838 	}
6839 	ctsio->kern_data_resid = 0;
6840 	ctsio->kern_rel_offset = 0;
6841 	ctsio->kern_sg_entries = 0;
6842 
6843 	memset(data, 0, sizeof(*data));
6844 
6845 	scsi_u64to8b(lun->be_lun->maxlba, data->addr);
6846 	/* XXX KDM this may not be 512 bytes... */
6847 	scsi_ulto4b(lun->be_lun->blocksize, data->length);
6848 
6849 	ctsio->scsi_status = SCSI_STATUS_OK;
6850 
6851 	ctsio->be_move_done = ctl_config_move_done;
6852 	ctl_datamove((union ctl_io *)ctsio);
6853 
6854 	return (CTL_RETVAL_COMPLETE);
6855 }
6856 
6857 int
6858 ctl_service_action_in(struct ctl_scsiio *ctsio)
6859 {
6860 	struct scsi_service_action_in *cdb;
6861 	int retval;
6862 
6863 	CTL_DEBUG_PRINT(("ctl_service_action_in\n"));
6864 
6865 	cdb = (struct scsi_service_action_in *)ctsio->cdb;
6866 
6867 	retval = CTL_RETVAL_COMPLETE;
6868 
6869 	switch (cdb->service_action) {
6870 	case SRC16_SERVICE_ACTION:
6871 		retval = ctl_read_capacity_16(ctsio);
6872 		break;
6873 	default:
6874 		ctl_set_invalid_field(/*ctsio*/ ctsio,
6875 				      /*sks_valid*/ 1,
6876 				      /*command*/ 1,
6877 				      /*field*/ 1,
6878 				      /*bit_valid*/ 1,
6879 				      /*bit*/ 4);
6880 		ctl_done((union ctl_io *)ctsio);
6881 		break;
6882 	}
6883 
6884 	return (retval);
6885 }
6886 
6887 int
6888 ctl_maintenance_in(struct ctl_scsiio *ctsio)
6889 {
6890 	struct scsi_maintenance_in *cdb;
6891 	int retval;
6892 	int alloc_len, total_len = 0;
6893 	int num_target_port_groups;
6894 	struct ctl_lun *lun;
6895 	struct ctl_softc *softc;
6896 	struct scsi_target_group_data *rtg_ptr;
6897 	struct scsi_target_port_group_descriptor *tpg_desc_ptr1, *tpg_desc_ptr2;
6898 	struct scsi_target_port_descriptor  *tp_desc_ptr1_1, *tp_desc_ptr1_2,
6899 	                                    *tp_desc_ptr2_1, *tp_desc_ptr2_2;
6900 
6901 	CTL_DEBUG_PRINT(("ctl_maintenance_in\n"));
6902 
6903 	cdb = (struct scsi_maintenance_in *)ctsio->cdb;
6904 	softc = control_softc;
6905 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6906 
6907 	retval = CTL_RETVAL_COMPLETE;
6908 	mtx_lock(&softc->ctl_lock);
6909 
6910 	if ((cdb->byte2 & SERVICE_ACTION_MASK) != SA_RPRT_TRGT_GRP) {
6911 		ctl_set_invalid_field(/*ctsio*/ ctsio,
6912 				      /*sks_valid*/ 1,
6913 				      /*command*/ 1,
6914 				      /*field*/ 1,
6915 				      /*bit_valid*/ 1,
6916 				      /*bit*/ 4);
6917 		ctl_done((union ctl_io *)ctsio);
6918 		return(retval);
6919 	}
6920 
6921 	if (ctl_is_single)
6922         	num_target_port_groups = NUM_TARGET_PORT_GROUPS - 1;
6923 	else
6924         	num_target_port_groups = NUM_TARGET_PORT_GROUPS;
6925 
6926 	total_len = sizeof(struct scsi_target_group_data) +
6927 		sizeof(struct scsi_target_port_group_descriptor) *
6928 		num_target_port_groups +
6929 		sizeof(struct scsi_target_port_descriptor) *
6930 		NUM_PORTS_PER_GRP * num_target_port_groups;
6931 
6932 	alloc_len = scsi_4btoul(cdb->length);
6933 
6934 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK);
6935 	if (ctsio->kern_data_ptr == NULL) {
6936 		ctsio->io_hdr.status = CTL_SCSI_ERROR;
6937 		ctsio->scsi_status = SCSI_STATUS_BUSY;
6938 		ctl_done((union ctl_io *)ctsio);
6939 		return (CTL_RETVAL_COMPLETE);
6940 	}
6941 	memset(ctsio->kern_data_ptr, 0, total_len);
6942 
6943 	ctsio->kern_sg_entries = 0;
6944 
6945 	if (total_len < alloc_len) {
6946 		ctsio->residual = alloc_len - total_len;
6947 		ctsio->kern_data_len = total_len;
6948 		ctsio->kern_total_len = total_len;
6949 	} else {
6950 		ctsio->residual = 0;
6951 		ctsio->kern_data_len = alloc_len;
6952 		ctsio->kern_total_len = alloc_len;
6953 	}
6954 	ctsio->kern_data_resid = 0;
6955 	ctsio->kern_rel_offset = 0;
6956 
6957 	rtg_ptr = (struct scsi_target_group_data *)ctsio->kern_data_ptr;
6958 
6959 	tpg_desc_ptr1 = &rtg_ptr->groups[0];
6960 	tp_desc_ptr1_1 = &tpg_desc_ptr1->descriptors[0];
6961 	tp_desc_ptr1_2 = (struct scsi_target_port_descriptor *)
6962 	        &tp_desc_ptr1_1->desc_list[0];
6963 
6964 
6965 
6966 	if (ctl_is_single == 0) {
6967 		tpg_desc_ptr2 = (struct scsi_target_port_group_descriptor *)
6968 	                &tp_desc_ptr1_2->desc_list[0];
6969 		tp_desc_ptr2_1 = &tpg_desc_ptr2->descriptors[0];
6970 		tp_desc_ptr2_2 = (struct scsi_target_port_descriptor *)
6971 	        	&tp_desc_ptr2_1->desc_list[0];
6972         } else {
6973 		tpg_desc_ptr2 = NULL;
6974 		tp_desc_ptr2_1 = NULL;
6975 		tp_desc_ptr2_2 = NULL;
6976 	}
6977 
6978 	scsi_ulto4b(total_len - 4, rtg_ptr->length);
6979 	if (ctl_is_single == 0) {
6980         	if (ctsio->io_hdr.nexus.targ_port < CTL_MAX_PORTS) {
6981 			if (lun->flags & CTL_LUN_PRIMARY_SC) {
6982 				tpg_desc_ptr1->pref_state = TPG_PRIMARY;
6983 				tpg_desc_ptr2->pref_state =
6984 					TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
6985 			} else {
6986 				tpg_desc_ptr1->pref_state =
6987 					TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
6988 				tpg_desc_ptr2->pref_state = TPG_PRIMARY;
6989 			}
6990 		} else {
6991 			if (lun->flags & CTL_LUN_PRIMARY_SC) {
6992 				tpg_desc_ptr1->pref_state =
6993 					TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
6994 				tpg_desc_ptr2->pref_state = TPG_PRIMARY;
6995 			} else {
6996 				tpg_desc_ptr1->pref_state = TPG_PRIMARY;
6997 				tpg_desc_ptr2->pref_state =
6998 					TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
6999 			}
7000 		}
7001 	} else {
7002 		tpg_desc_ptr1->pref_state = TPG_PRIMARY;
7003 	}
7004 	tpg_desc_ptr1->support = 0;
7005 	tpg_desc_ptr1->target_port_group[1] = 1;
7006 	tpg_desc_ptr1->status = TPG_IMPLICIT;
7007 	tpg_desc_ptr1->target_port_count= NUM_PORTS_PER_GRP;
7008 
7009 	if (ctl_is_single == 0) {
7010 		tpg_desc_ptr2->support = 0;
7011 		tpg_desc_ptr2->target_port_group[1] = 2;
7012 		tpg_desc_ptr2->status = TPG_IMPLICIT;
7013 		tpg_desc_ptr2->target_port_count = NUM_PORTS_PER_GRP;
7014 
7015 		tp_desc_ptr1_1->relative_target_port_identifier[1] = 1;
7016 		tp_desc_ptr1_2->relative_target_port_identifier[1] = 2;
7017 
7018 		tp_desc_ptr2_1->relative_target_port_identifier[1] = 9;
7019 		tp_desc_ptr2_2->relative_target_port_identifier[1] = 10;
7020 	} else {
7021         	if (ctsio->io_hdr.nexus.targ_port < CTL_MAX_PORTS) {
7022 			tp_desc_ptr1_1->relative_target_port_identifier[1] = 1;
7023 			tp_desc_ptr1_2->relative_target_port_identifier[1] = 2;
7024 		} else {
7025 			tp_desc_ptr1_1->relative_target_port_identifier[1] = 9;
7026 			tp_desc_ptr1_2->relative_target_port_identifier[1] = 10;
7027 		}
7028 	}
7029 
7030 	mtx_unlock(&softc->ctl_lock);
7031 
7032 	ctsio->be_move_done = ctl_config_move_done;
7033 
7034 	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7035 			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7036 			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7037 			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7038 			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7039 
7040 	ctl_datamove((union ctl_io *)ctsio);
7041 	return(retval);
7042 }
7043 
7044 int
7045 ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7046 {
7047 	struct scsi_per_res_in *cdb;
7048 	int alloc_len, total_len = 0;
7049 	/* struct scsi_per_res_in_rsrv in_data; */
7050 	struct ctl_lun *lun;
7051 	struct ctl_softc *softc;
7052 
7053 	CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7054 
7055 	softc = control_softc;
7056 
7057 	cdb = (struct scsi_per_res_in *)ctsio->cdb;
7058 
7059 	alloc_len = scsi_2btoul(cdb->length);
7060 
7061 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7062 
7063 retry:
7064 	mtx_lock(&softc->ctl_lock);
7065 	switch (cdb->action) {
7066 	case SPRI_RK: /* read keys */
7067 		total_len = sizeof(struct scsi_per_res_in_keys) +
7068 			lun->pr_key_count *
7069 			sizeof(struct scsi_per_res_key);
7070 		break;
7071 	case SPRI_RR: /* read reservation */
7072 		if (lun->flags & CTL_LUN_PR_RESERVED)
7073 			total_len = sizeof(struct scsi_per_res_in_rsrv);
7074 		else
7075 			total_len = sizeof(struct scsi_per_res_in_header);
7076 		break;
7077 	case SPRI_RC: /* report capabilities */
7078 		total_len = sizeof(struct scsi_per_res_cap);
7079 		break;
7080 	case SPRI_RS: /* read full status */
7081 	default:
7082 		mtx_unlock(&softc->ctl_lock);
7083 		ctl_set_invalid_field(ctsio,
7084 				      /*sks_valid*/ 1,
7085 				      /*command*/ 1,
7086 				      /*field*/ 1,
7087 				      /*bit_valid*/ 1,
7088 				      /*bit*/ 0);
7089 		ctl_done((union ctl_io *)ctsio);
7090 		return (CTL_RETVAL_COMPLETE);
7091 		break; /* NOTREACHED */
7092 	}
7093 	mtx_unlock(&softc->ctl_lock);
7094 
7095 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK);
7096 	if (ctsio->kern_data_ptr == NULL) {
7097 		ctsio->io_hdr.status = CTL_SCSI_ERROR;
7098 		ctsio->scsi_status = SCSI_STATUS_BUSY;
7099 		ctl_done((union ctl_io *)ctsio);
7100 		return (CTL_RETVAL_COMPLETE);
7101 	}
7102 
7103 	if (total_len < alloc_len) {
7104 		ctsio->residual = alloc_len - total_len;
7105 		ctsio->kern_data_len = total_len;
7106 		ctsio->kern_total_len = total_len;
7107 	} else {
7108 		ctsio->residual = 0;
7109 		ctsio->kern_data_len = alloc_len;
7110 		ctsio->kern_total_len = alloc_len;
7111 	}
7112 
7113 	ctsio->kern_data_resid = 0;
7114 	ctsio->kern_rel_offset = 0;
7115 	ctsio->kern_sg_entries = 0;
7116 
7117 	memset(ctsio->kern_data_ptr, 0, total_len);
7118 
7119 	mtx_lock(&softc->ctl_lock);
7120 	switch (cdb->action) {
7121 	case SPRI_RK: { // read keys
7122         struct scsi_per_res_in_keys *res_keys;
7123 		int i, key_count;
7124 
7125 		res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7126 
7127 		/*
7128 		 * We had to drop the lock to allocate our buffer, which
7129 		 * leaves time for someone to come in with another
7130 		 * persistent reservation.  (That is unlikely, though,
7131 		 * since this should be the only persistent reservation
7132 		 * command active right now.)
7133 		 */
7134 		if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7135 		    (lun->pr_key_count *
7136 		     sizeof(struct scsi_per_res_key)))){
7137 			mtx_unlock(&softc->ctl_lock);
7138 			free(ctsio->kern_data_ptr, M_CTL);
7139 			printf("%s: reservation length changed, retrying\n",
7140 			       __func__);
7141 			goto retry;
7142 		}
7143 
7144 		scsi_ulto4b(lun->PRGeneration, res_keys->header.generation);
7145 
7146 		scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7147 			     lun->pr_key_count, res_keys->header.length);
7148 
7149 		for (i = 0, key_count = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7150 			if (!lun->per_res[i].registered)
7151 				continue;
7152 
7153 			/*
7154 			 * We used lun->pr_key_count to calculate the
7155 			 * size to allocate.  If it turns out the number of
7156 			 * initiators with the registered flag set is
7157 			 * larger than that (i.e. they haven't been kept in
7158 			 * sync), we've got a problem.
7159 			 */
7160 			if (key_count >= lun->pr_key_count) {
7161 #ifdef NEEDTOPORT
7162 				csevent_log(CSC_CTL | CSC_SHELF_SW |
7163 					    CTL_PR_ERROR,
7164 					    csevent_LogType_Fault,
7165 					    csevent_AlertLevel_Yellow,
7166 					    csevent_FRU_ShelfController,
7167 					    csevent_FRU_Firmware,
7168 				        csevent_FRU_Unknown,
7169 					    "registered keys %d >= key "
7170 					    "count %d", key_count,
7171 					    lun->pr_key_count);
7172 #endif
7173 				key_count++;
7174 				continue;
7175 			}
7176 			memcpy(res_keys->keys[key_count].key,
7177 			       lun->per_res[i].res_key.key,
7178 			       ctl_min(sizeof(res_keys->keys[key_count].key),
7179 			       sizeof(lun->per_res[i].res_key)));
7180 			key_count++;
7181 		}
7182 		break;
7183 	}
7184 	case SPRI_RR: { // read reservation
7185 		struct scsi_per_res_in_rsrv *res;
7186 		int tmp_len, header_only;
7187 
7188 		res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7189 
7190 		scsi_ulto4b(lun->PRGeneration, res->header.generation);
7191 
7192 		if (lun->flags & CTL_LUN_PR_RESERVED)
7193 		{
7194 			tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7195 			scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7196 				    res->header.length);
7197 			header_only = 0;
7198 		} else {
7199 			tmp_len = sizeof(struct scsi_per_res_in_header);
7200 			scsi_ulto4b(0, res->header.length);
7201 			header_only = 1;
7202 		}
7203 
7204 		/*
7205 		 * We had to drop the lock to allocate our buffer, which
7206 		 * leaves time for someone to come in with another
7207 		 * persistent reservation.  (That is unlikely, though,
7208 		 * since this should be the only persistent reservation
7209 		 * command active right now.)
7210 		 */
7211 		if (tmp_len != total_len) {
7212 			mtx_unlock(&softc->ctl_lock);
7213 			free(ctsio->kern_data_ptr, M_CTL);
7214 			printf("%s: reservation status changed, retrying\n",
7215 			       __func__);
7216 			goto retry;
7217 		}
7218 
7219 		/*
7220 		 * No reservation held, so we're done.
7221 		 */
7222 		if (header_only != 0)
7223 			break;
7224 
7225 		/*
7226 		 * If the registration is an All Registrants type, the key
7227 		 * is 0, since it doesn't really matter.
7228 		 */
7229 		if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7230 			memcpy(res->data.reservation,
7231 			       &lun->per_res[lun->pr_res_idx].res_key,
7232 			       sizeof(struct scsi_per_res_key));
7233 		}
7234 		res->data.scopetype = lun->res_type;
7235 		break;
7236 	}
7237 	case SPRI_RC:     //report capabilities
7238 	{
7239 		struct scsi_per_res_cap *res_cap;
7240 		uint16_t type_mask;
7241 
7242 		res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7243 		scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7244 		res_cap->flags2 |= SPRI_TMV;
7245 		type_mask = SPRI_TM_WR_EX_AR |
7246 			    SPRI_TM_EX_AC_RO |
7247 			    SPRI_TM_WR_EX_RO |
7248 			    SPRI_TM_EX_AC |
7249 			    SPRI_TM_WR_EX |
7250 			    SPRI_TM_EX_AC_AR;
7251 		scsi_ulto2b(type_mask, res_cap->type_mask);
7252 		break;
7253 	}
7254 	case SPRI_RS: //read full status
7255 	default:
7256 		/*
7257 		 * This is a bug, because we just checked for this above,
7258 		 * and should have returned an error.
7259 		 */
7260 		panic("Invalid PR type %x", cdb->action);
7261 		break; /* NOTREACHED */
7262 	}
7263 	mtx_unlock(&softc->ctl_lock);
7264 
7265 	ctsio->be_move_done = ctl_config_move_done;
7266 
7267 	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7268 			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7269 			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7270 			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7271 			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7272 
7273 	ctl_datamove((union ctl_io *)ctsio);
7274 
7275 	return (CTL_RETVAL_COMPLETE);
7276 }
7277 
7278 /*
7279  * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7280  * it should return.
7281  */
7282 static int
7283 ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7284 		uint64_t sa_res_key, uint8_t type, uint32_t residx,
7285 		struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7286 		struct scsi_per_res_out_parms* param)
7287 {
7288 	union ctl_ha_msg persis_io;
7289 	int retval, i;
7290 	int isc_retval;
7291 
7292 	retval = 0;
7293 
7294 	if (sa_res_key == 0) {
7295 		mtx_lock(&softc->ctl_lock);
7296 		if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7297 			/* validate scope and type */
7298 			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7299 			     SPR_LU_SCOPE) {
7300 				mtx_unlock(&softc->ctl_lock);
7301 				ctl_set_invalid_field(/*ctsio*/ ctsio,
7302 						      /*sks_valid*/ 1,
7303 						      /*command*/ 1,
7304 						      /*field*/ 2,
7305 						      /*bit_valid*/ 1,
7306 						      /*bit*/ 4);
7307 				ctl_done((union ctl_io *)ctsio);
7308 				return (1);
7309 			}
7310 
7311 		        if (type>8 || type==2 || type==4 || type==0) {
7312 				mtx_unlock(&softc->ctl_lock);
7313 				ctl_set_invalid_field(/*ctsio*/ ctsio,
7314        	           				      /*sks_valid*/ 1,
7315 						      /*command*/ 1,
7316 						      /*field*/ 2,
7317 						      /*bit_valid*/ 1,
7318 						      /*bit*/ 0);
7319 				ctl_done((union ctl_io *)ctsio);
7320 				return (1);
7321 		        }
7322 
7323 			/* temporarily unregister this nexus */
7324 			lun->per_res[residx].registered = 0;
7325 
7326 			/*
7327 			 * Unregister everybody else and build UA for
7328 			 * them
7329 			 */
7330 			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7331 				if (lun->per_res[i].registered == 0)
7332 					continue;
7333 
7334 				if (!persis_offset
7335 				 && i <CTL_MAX_INITIATORS)
7336 					lun->pending_sense[i].ua_pending |=
7337 						CTL_UA_REG_PREEMPT;
7338 				else if (persis_offset
7339 				      && i >= persis_offset)
7340 					lun->pending_sense[i-persis_offset
7341 						].ua_pending |=
7342 						CTL_UA_REG_PREEMPT;
7343 				lun->per_res[i].registered = 0;
7344 				memset(&lun->per_res[i].res_key, 0,
7345 				       sizeof(struct scsi_per_res_key));
7346 			}
7347 			lun->per_res[residx].registered = 1;
7348 			lun->pr_key_count = 1;
7349 			lun->res_type = type;
7350 			if (lun->res_type != SPR_TYPE_WR_EX_AR
7351 			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7352 				lun->pr_res_idx = residx;
7353 
7354 			mtx_unlock(&softc->ctl_lock);
7355 			/* send msg to other side */
7356 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7357 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7358 			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7359 			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7360 			persis_io.pr.pr_info.res_type = type;
7361 			memcpy(persis_io.pr.pr_info.sa_res_key,
7362 			       param->serv_act_res_key,
7363 			       sizeof(param->serv_act_res_key));
7364 			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7365 			     &persis_io, sizeof(persis_io), 0)) >
7366 			     CTL_HA_STATUS_SUCCESS) {
7367 				printf("CTL:Persis Out error returned "
7368 				       "from ctl_ha_msg_send %d\n",
7369 				       isc_retval);
7370 			}
7371 		} else {
7372 			/* not all registrants */
7373 			mtx_unlock(&softc->ctl_lock);
7374 			free(ctsio->kern_data_ptr, M_CTL);
7375 			ctl_set_invalid_field(ctsio,
7376 					      /*sks_valid*/ 1,
7377 					      /*command*/ 0,
7378 					      /*field*/ 8,
7379 					      /*bit_valid*/ 0,
7380 					      /*bit*/ 0);
7381 			ctl_done((union ctl_io *)ctsio);
7382 			return (1);
7383 		}
7384 	} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7385 		|| !(lun->flags & CTL_LUN_PR_RESERVED)) {
7386 		int found = 0;
7387 
7388 		mtx_lock(&softc->ctl_lock);
7389 		if (res_key == sa_res_key) {
7390 			/* special case */
7391 			/*
7392 			 * The spec implies this is not good but doesn't
7393 			 * say what to do. There are two choices either
7394 			 * generate a res conflict or check condition
7395 			 * with illegal field in parameter data. Since
7396 			 * that is what is done when the sa_res_key is
7397 			 * zero I'll take that approach since this has
7398 			 * to do with the sa_res_key.
7399 			 */
7400 			mtx_unlock(&softc->ctl_lock);
7401 			free(ctsio->kern_data_ptr, M_CTL);
7402 			ctl_set_invalid_field(ctsio,
7403 					      /*sks_valid*/ 1,
7404 					      /*command*/ 0,
7405 					      /*field*/ 8,
7406 					      /*bit_valid*/ 0,
7407 					      /*bit*/ 0);
7408 			ctl_done((union ctl_io *)ctsio);
7409 			return (1);
7410 		}
7411 
7412 		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7413 			if (lun->per_res[i].registered
7414 			 && memcmp(param->serv_act_res_key,
7415 			    lun->per_res[i].res_key.key,
7416 			    sizeof(struct scsi_per_res_key)) != 0)
7417 				continue;
7418 
7419 			found = 1;
7420 			lun->per_res[i].registered = 0;
7421 			memset(&lun->per_res[i].res_key, 0,
7422 			       sizeof(struct scsi_per_res_key));
7423 			lun->pr_key_count--;
7424 
7425 			if (!persis_offset
7426 			 && i < CTL_MAX_INITIATORS)
7427 				lun->pending_sense[i].ua_pending |=
7428 					CTL_UA_REG_PREEMPT;
7429 			else if (persis_offset
7430 			      && i >= persis_offset)
7431 				lun->pending_sense[i-persis_offset].ua_pending|=
7432 					CTL_UA_REG_PREEMPT;
7433 		}
7434 		mtx_unlock(&softc->ctl_lock);
7435 		if (!found) {
7436 			free(ctsio->kern_data_ptr, M_CTL);
7437 			ctl_set_reservation_conflict(ctsio);
7438 			ctl_done((union ctl_io *)ctsio);
7439 			return (CTL_RETVAL_COMPLETE);
7440 		}
7441 		/* send msg to other side */
7442 		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7443 		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7444 		persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7445 		persis_io.pr.pr_info.residx = lun->pr_res_idx;
7446 		persis_io.pr.pr_info.res_type = type;
7447 		memcpy(persis_io.pr.pr_info.sa_res_key,
7448 		       param->serv_act_res_key,
7449 		       sizeof(param->serv_act_res_key));
7450 		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7451 		     &persis_io, sizeof(persis_io), 0)) >
7452 		     CTL_HA_STATUS_SUCCESS) {
7453 			printf("CTL:Persis Out error returned from "
7454 			       "ctl_ha_msg_send %d\n", isc_retval);
7455 		}
7456 	} else {
7457 		/* Reserved but not all registrants */
7458 		/* sa_res_key is res holder */
7459 		if (memcmp(param->serv_act_res_key,
7460                    lun->per_res[lun->pr_res_idx].res_key.key,
7461                    sizeof(struct scsi_per_res_key)) == 0) {
7462 			/* validate scope and type */
7463 			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7464 			     SPR_LU_SCOPE) {
7465 				ctl_set_invalid_field(/*ctsio*/ ctsio,
7466 						      /*sks_valid*/ 1,
7467 						      /*command*/ 1,
7468 						      /*field*/ 2,
7469 						      /*bit_valid*/ 1,
7470 						      /*bit*/ 4);
7471 				ctl_done((union ctl_io *)ctsio);
7472 				return (1);
7473 			}
7474 
7475 			if (type>8 || type==2 || type==4 || type==0) {
7476 				ctl_set_invalid_field(/*ctsio*/ ctsio,
7477 						      /*sks_valid*/ 1,
7478 						      /*command*/ 1,
7479 						      /*field*/ 2,
7480 						      /*bit_valid*/ 1,
7481 						      /*bit*/ 0);
7482 				ctl_done((union ctl_io *)ctsio);
7483 				return (1);
7484 			}
7485 
7486 			/*
7487 			 * Do the following:
7488 			 * if sa_res_key != res_key remove all
7489 			 * registrants w/sa_res_key and generate UA
7490 			 * for these registrants(Registrations
7491 			 * Preempted) if it wasn't an exclusive
7492 			 * reservation generate UA(Reservations
7493 			 * Preempted) for all other registered nexuses
7494 			 * if the type has changed. Establish the new
7495 			 * reservation and holder. If res_key and
7496 			 * sa_res_key are the same do the above
7497 			 * except don't unregister the res holder.
7498 			 */
7499 
7500 			/*
7501 			 * Temporarily unregister so it won't get
7502 			 * removed or UA generated
7503 			 */
7504 			lun->per_res[residx].registered = 0;
7505 			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7506 				if (lun->per_res[i].registered == 0)
7507 					continue;
7508 
7509 				if (memcmp(param->serv_act_res_key,
7510 				    lun->per_res[i].res_key.key,
7511 				    sizeof(struct scsi_per_res_key)) == 0) {
7512 					lun->per_res[i].registered = 0;
7513 					memset(&lun->per_res[i].res_key,
7514 					       0,
7515 					       sizeof(struct scsi_per_res_key));
7516 					lun->pr_key_count--;
7517 
7518 					if (!persis_offset
7519 					 && i < CTL_MAX_INITIATORS)
7520 						lun->pending_sense[i
7521 							].ua_pending |=
7522 							CTL_UA_REG_PREEMPT;
7523 					else if (persis_offset
7524 					      && i >= persis_offset)
7525 						lun->pending_sense[
7526 						  i-persis_offset].ua_pending |=
7527 						  CTL_UA_REG_PREEMPT;
7528 				} else if (type != lun->res_type
7529 					&& (lun->res_type == SPR_TYPE_WR_EX_RO
7530 					 || lun->res_type ==SPR_TYPE_EX_AC_RO)){
7531 						if (!persis_offset
7532 						 && i < CTL_MAX_INITIATORS)
7533 							lun->pending_sense[i
7534 							].ua_pending |=
7535 							CTL_UA_RES_RELEASE;
7536 						else if (persis_offset
7537 						      && i >= persis_offset)
7538 							lun->pending_sense[
7539 							i-persis_offset
7540 							].ua_pending |=
7541 							CTL_UA_RES_RELEASE;
7542 				}
7543 			}
7544 			lun->per_res[residx].registered = 1;
7545 			lun->res_type = type;
7546 			if (lun->res_type != SPR_TYPE_WR_EX_AR
7547 			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7548 				lun->pr_res_idx = residx;
7549 			else
7550 				lun->pr_res_idx =
7551 					CTL_PR_ALL_REGISTRANTS;
7552 
7553 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7554 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7555 			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7556 			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7557 			persis_io.pr.pr_info.res_type = type;
7558 			memcpy(persis_io.pr.pr_info.sa_res_key,
7559 			       param->serv_act_res_key,
7560 			       sizeof(param->serv_act_res_key));
7561 			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7562 			     &persis_io, sizeof(persis_io), 0)) >
7563 			     CTL_HA_STATUS_SUCCESS) {
7564 				printf("CTL:Persis Out error returned "
7565 				       "from ctl_ha_msg_send %d\n",
7566 				       isc_retval);
7567 			}
7568 		} else {
7569 			/*
7570 			 * sa_res_key is not the res holder just
7571 			 * remove registrants
7572 			 */
7573 			int found=0;
7574 			mtx_lock(&softc->ctl_lock);
7575 
7576 			for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7577 				if (memcmp(param->serv_act_res_key,
7578 				    lun->per_res[i].res_key.key,
7579 				    sizeof(struct scsi_per_res_key)) != 0)
7580 					continue;
7581 
7582 				found = 1;
7583 				lun->per_res[i].registered = 0;
7584 				memset(&lun->per_res[i].res_key, 0,
7585 				       sizeof(struct scsi_per_res_key));
7586 				lun->pr_key_count--;
7587 
7588 				if (!persis_offset
7589 				 && i < CTL_MAX_INITIATORS)
7590 					lun->pending_sense[i].ua_pending |=
7591 						CTL_UA_REG_PREEMPT;
7592 				else if (persis_offset
7593 				      && i >= persis_offset)
7594 					lun->pending_sense[
7595 						i-persis_offset].ua_pending |=
7596 						CTL_UA_REG_PREEMPT;
7597 			}
7598 
7599 			if (!found) {
7600 				mtx_unlock(&softc->ctl_lock);
7601 				free(ctsio->kern_data_ptr, M_CTL);
7602 				ctl_set_reservation_conflict(ctsio);
7603 				ctl_done((union ctl_io *)ctsio);
7604 		        	return (1);
7605 			}
7606 			mtx_unlock(&softc->ctl_lock);
7607 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7608 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7609 			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7610 			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7611 			persis_io.pr.pr_info.res_type = type;
7612 			memcpy(persis_io.pr.pr_info.sa_res_key,
7613 			       param->serv_act_res_key,
7614 			       sizeof(param->serv_act_res_key));
7615 			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7616 			     &persis_io, sizeof(persis_io), 0)) >
7617 			     CTL_HA_STATUS_SUCCESS) {
7618 				printf("CTL:Persis Out error returned "
7619 				       "from ctl_ha_msg_send %d\n",
7620 				isc_retval);
7621 			}
7622 		}
7623 	}
7624 
7625 	lun->PRGeneration++;
7626 
7627 	return (retval);
7628 }
7629 
7630 static void
7631 ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
7632 {
7633 	int i;
7634 
7635 	if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7636 	 || lun->pr_res_idx == CTL_PR_NO_RESERVATION
7637 	 || memcmp(&lun->per_res[lun->pr_res_idx].res_key,
7638 		   msg->pr.pr_info.sa_res_key,
7639 		   sizeof(struct scsi_per_res_key)) != 0) {
7640 		uint64_t sa_res_key;
7641 		sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
7642 
7643 		if (sa_res_key == 0) {
7644 			/* temporarily unregister this nexus */
7645 			lun->per_res[msg->pr.pr_info.residx].registered = 0;
7646 
7647 			/*
7648 			 * Unregister everybody else and build UA for
7649 			 * them
7650 			 */
7651 			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7652 				if (lun->per_res[i].registered == 0)
7653 					continue;
7654 
7655 				if (!persis_offset
7656 				 && i < CTL_MAX_INITIATORS)
7657 					lun->pending_sense[i].ua_pending |=
7658 						CTL_UA_REG_PREEMPT;
7659 				else if (persis_offset && i >= persis_offset)
7660 					lun->pending_sense[i -
7661 						persis_offset].ua_pending |=
7662 						CTL_UA_REG_PREEMPT;
7663 				lun->per_res[i].registered = 0;
7664 				memset(&lun->per_res[i].res_key, 0,
7665 				       sizeof(struct scsi_per_res_key));
7666 			}
7667 
7668 			lun->per_res[msg->pr.pr_info.residx].registered = 1;
7669 			lun->pr_key_count = 1;
7670 			lun->res_type = msg->pr.pr_info.res_type;
7671 			if (lun->res_type != SPR_TYPE_WR_EX_AR
7672 			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7673 				lun->pr_res_idx = msg->pr.pr_info.residx;
7674 		} else {
7675 		        for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7676 				if (memcmp(msg->pr.pr_info.sa_res_key,
7677 		                   lun->per_res[i].res_key.key,
7678 		                   sizeof(struct scsi_per_res_key)) != 0)
7679 					continue;
7680 
7681 				lun->per_res[i].registered = 0;
7682 				memset(&lun->per_res[i].res_key, 0,
7683 				       sizeof(struct scsi_per_res_key));
7684 				lun->pr_key_count--;
7685 
7686 				if (!persis_offset
7687 				 && i < persis_offset)
7688 					lun->pending_sense[i].ua_pending |=
7689 						CTL_UA_REG_PREEMPT;
7690 				else if (persis_offset
7691 				      && i >= persis_offset)
7692 					lun->pending_sense[i -
7693 						persis_offset].ua_pending |=
7694 						CTL_UA_REG_PREEMPT;
7695 			}
7696 		}
7697 	} else {
7698 		/*
7699 		 * Temporarily unregister so it won't get removed
7700 		 * or UA generated
7701 		 */
7702 		lun->per_res[msg->pr.pr_info.residx].registered = 0;
7703 		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7704 			if (lun->per_res[i].registered == 0)
7705 				continue;
7706 
7707 			if (memcmp(msg->pr.pr_info.sa_res_key,
7708 	                   lun->per_res[i].res_key.key,
7709 	                   sizeof(struct scsi_per_res_key)) == 0) {
7710 				lun->per_res[i].registered = 0;
7711 				memset(&lun->per_res[i].res_key, 0,
7712 				       sizeof(struct scsi_per_res_key));
7713 				lun->pr_key_count--;
7714 				if (!persis_offset
7715 				 && i < CTL_MAX_INITIATORS)
7716 					lun->pending_sense[i].ua_pending |=
7717 						CTL_UA_REG_PREEMPT;
7718 				else if (persis_offset
7719 				      && i >= persis_offset)
7720 					lun->pending_sense[i -
7721 						persis_offset].ua_pending |=
7722 						CTL_UA_REG_PREEMPT;
7723 			} else if (msg->pr.pr_info.res_type != lun->res_type
7724 				&& (lun->res_type == SPR_TYPE_WR_EX_RO
7725 				 || lun->res_type == SPR_TYPE_EX_AC_RO)) {
7726 					if (!persis_offset
7727 					 && i < persis_offset)
7728 						lun->pending_sense[i
7729 							].ua_pending |=
7730 							CTL_UA_RES_RELEASE;
7731 					else if (persis_offset
7732 					      && i >= persis_offset)
7733 					lun->pending_sense[i -
7734 						persis_offset].ua_pending |=
7735 						CTL_UA_RES_RELEASE;
7736 			}
7737 		}
7738 		lun->per_res[msg->pr.pr_info.residx].registered = 1;
7739 		lun->res_type = msg->pr.pr_info.res_type;
7740 		if (lun->res_type != SPR_TYPE_WR_EX_AR
7741 		 && lun->res_type != SPR_TYPE_EX_AC_AR)
7742 			lun->pr_res_idx = msg->pr.pr_info.residx;
7743 		else
7744 			lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
7745 	}
7746 	lun->PRGeneration++;
7747 
7748 }
7749 
7750 
7751 int
7752 ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
7753 {
7754 	int retval;
7755 	int isc_retval;
7756 	u_int32_t param_len;
7757 	struct scsi_per_res_out *cdb;
7758 	struct ctl_lun *lun;
7759 	struct scsi_per_res_out_parms* param;
7760 	struct ctl_softc *softc;
7761 	uint32_t residx;
7762 	uint64_t res_key, sa_res_key;
7763 	uint8_t type;
7764 	union ctl_ha_msg persis_io;
7765 	int    i;
7766 
7767 	CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
7768 
7769 	retval = CTL_RETVAL_COMPLETE;
7770 
7771 	softc = control_softc;
7772 
7773 	cdb = (struct scsi_per_res_out *)ctsio->cdb;
7774 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7775 
7776 	/*
7777 	 * We only support whole-LUN scope.  The scope & type are ignored for
7778 	 * register, register and ignore existing key and clear.
7779 	 * We sometimes ignore scope and type on preempts too!!
7780 	 * Verify reservation type here as well.
7781 	 */
7782 	type = cdb->scope_type & SPR_TYPE_MASK;
7783 	if ((cdb->action == SPRO_RESERVE)
7784 	 || (cdb->action == SPRO_RELEASE)) {
7785 		if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
7786 			ctl_set_invalid_field(/*ctsio*/ ctsio,
7787 					      /*sks_valid*/ 1,
7788 					      /*command*/ 1,
7789 					      /*field*/ 2,
7790 					      /*bit_valid*/ 1,
7791 					      /*bit*/ 4);
7792 			ctl_done((union ctl_io *)ctsio);
7793 			return (CTL_RETVAL_COMPLETE);
7794 		}
7795 
7796 		if (type>8 || type==2 || type==4 || type==0) {
7797 			ctl_set_invalid_field(/*ctsio*/ ctsio,
7798 					      /*sks_valid*/ 1,
7799 					      /*command*/ 1,
7800 					      /*field*/ 2,
7801 					      /*bit_valid*/ 1,
7802 					      /*bit*/ 0);
7803 			ctl_done((union ctl_io *)ctsio);
7804 			return (CTL_RETVAL_COMPLETE);
7805 		}
7806 	}
7807 
7808 	switch (cdb->action & SPRO_ACTION_MASK) {
7809 	case SPRO_REGISTER:
7810 	case SPRO_RESERVE:
7811 	case SPRO_RELEASE:
7812 	case SPRO_CLEAR:
7813 	case SPRO_PREEMPT:
7814 	case SPRO_REG_IGNO:
7815 		break;
7816 	case SPRO_REG_MOVE:
7817 	case SPRO_PRE_ABO:
7818 	default:
7819 		ctl_set_invalid_field(/*ctsio*/ ctsio,
7820 				      /*sks_valid*/ 1,
7821 				      /*command*/ 1,
7822 				      /*field*/ 1,
7823 				      /*bit_valid*/ 1,
7824 				      /*bit*/ 0);
7825 		ctl_done((union ctl_io *)ctsio);
7826 		return (CTL_RETVAL_COMPLETE);
7827 		break; /* NOTREACHED */
7828 	}
7829 
7830 	param_len = scsi_4btoul(cdb->length);
7831 
7832 	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
7833 		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
7834 		if (ctsio->kern_data_ptr == NULL) {
7835 			ctl_set_busy(ctsio);
7836 			ctl_done((union ctl_io *)ctsio);
7837 			return (CTL_RETVAL_COMPLETE);
7838 		}
7839 		ctsio->kern_data_len = param_len;
7840 		ctsio->kern_total_len = param_len;
7841 		ctsio->kern_data_resid = 0;
7842 		ctsio->kern_rel_offset = 0;
7843 		ctsio->kern_sg_entries = 0;
7844 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7845 		ctsio->be_move_done = ctl_config_move_done;
7846 		ctl_datamove((union ctl_io *)ctsio);
7847 
7848 		return (CTL_RETVAL_COMPLETE);
7849 	}
7850 
7851 	param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
7852 
7853 	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
7854 	res_key = scsi_8btou64(param->res_key.key);
7855 	sa_res_key = scsi_8btou64(param->serv_act_res_key);
7856 
7857 	/*
7858 	 * Validate the reservation key here except for SPRO_REG_IGNO
7859 	 * This must be done for all other service actions
7860 	 */
7861 	if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
7862 		mtx_lock(&softc->ctl_lock);
7863 		if (lun->per_res[residx].registered) {
7864 		    if (memcmp(param->res_key.key,
7865 			       lun->per_res[residx].res_key.key,
7866 			       ctl_min(sizeof(param->res_key),
7867 			       sizeof(lun->per_res[residx].res_key))) != 0) {
7868 				/*
7869 				 * The current key passed in doesn't match
7870 				 * the one the initiator previously
7871 				 * registered.
7872 				 */
7873 				mtx_unlock(&softc->ctl_lock);
7874 				free(ctsio->kern_data_ptr, M_CTL);
7875 				ctl_set_reservation_conflict(ctsio);
7876 				ctl_done((union ctl_io *)ctsio);
7877 				return (CTL_RETVAL_COMPLETE);
7878 			}
7879 		} else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
7880 		    /*
7881 			 * We are not registered
7882 			 */
7883 			mtx_unlock(&softc->ctl_lock);
7884 			free(ctsio->kern_data_ptr, M_CTL);
7885 			ctl_set_reservation_conflict(ctsio);
7886 			ctl_done((union ctl_io *)ctsio);
7887 			return (CTL_RETVAL_COMPLETE);
7888 		} else if (res_key != 0) {
7889 			/*
7890 			 * We are not registered and trying to register but
7891 			 * the register key isn't zero.
7892 			 */
7893 			mtx_unlock(&softc->ctl_lock);
7894 			free(ctsio->kern_data_ptr, M_CTL);
7895 			ctl_set_reservation_conflict(ctsio);
7896 			ctl_done((union ctl_io *)ctsio);
7897 			return (CTL_RETVAL_COMPLETE);
7898 		}
7899 		mtx_unlock(&softc->ctl_lock);
7900 	}
7901 
7902 	switch (cdb->action & SPRO_ACTION_MASK) {
7903 	case SPRO_REGISTER:
7904 	case SPRO_REG_IGNO: {
7905 
7906 #if 0
7907 		printf("Registration received\n");
7908 #endif
7909 
7910 		/*
7911 		 * We don't support any of these options, as we report in
7912 		 * the read capabilities request (see
7913 		 * ctl_persistent_reserve_in(), above).
7914 		 */
7915 		if ((param->flags & SPR_SPEC_I_PT)
7916 		 || (param->flags & SPR_ALL_TG_PT)
7917 		 || (param->flags & SPR_APTPL)) {
7918 			int bit_ptr;
7919 
7920 			if (param->flags & SPR_APTPL)
7921 				bit_ptr = 0;
7922 			else if (param->flags & SPR_ALL_TG_PT)
7923 				bit_ptr = 2;
7924 			else /* SPR_SPEC_I_PT */
7925 				bit_ptr = 3;
7926 
7927 			free(ctsio->kern_data_ptr, M_CTL);
7928 			ctl_set_invalid_field(ctsio,
7929 					      /*sks_valid*/ 1,
7930 					      /*command*/ 0,
7931 					      /*field*/ 20,
7932 					      /*bit_valid*/ 1,
7933 					      /*bit*/ bit_ptr);
7934 			ctl_done((union ctl_io *)ctsio);
7935 			return (CTL_RETVAL_COMPLETE);
7936 		}
7937 
7938 		mtx_lock(&softc->ctl_lock);
7939 
7940 		/*
7941 		 * The initiator wants to clear the
7942 		 * key/unregister.
7943 		 */
7944 		if (sa_res_key == 0) {
7945 			if ((res_key == 0
7946 			  && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
7947 			 || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
7948 			  && !lun->per_res[residx].registered)) {
7949 				mtx_unlock(&softc->ctl_lock);
7950 				goto done;
7951 			}
7952 
7953 			lun->per_res[residx].registered = 0;
7954 			memset(&lun->per_res[residx].res_key,
7955 			       0, sizeof(lun->per_res[residx].res_key));
7956 			lun->pr_key_count--;
7957 
7958 			if (residx == lun->pr_res_idx) {
7959 				lun->flags &= ~CTL_LUN_PR_RESERVED;
7960 				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
7961 
7962 				if ((lun->res_type == SPR_TYPE_WR_EX_RO
7963 				  || lun->res_type == SPR_TYPE_EX_AC_RO)
7964 				 && lun->pr_key_count) {
7965 					/*
7966 					 * If the reservation is a registrants
7967 					 * only type we need to generate a UA
7968 					 * for other registered inits.  The
7969 					 * sense code should be RESERVATIONS
7970 					 * RELEASED
7971 					 */
7972 
7973 					for (i = 0; i < CTL_MAX_INITIATORS;i++){
7974 						if (lun->per_res[
7975 						    i+persis_offset].registered
7976 						    == 0)
7977 							continue;
7978 						lun->pending_sense[i
7979 							].ua_pending |=
7980 							CTL_UA_RES_RELEASE;
7981 					}
7982 				}
7983 				lun->res_type = 0;
7984 			} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7985 				if (lun->pr_key_count==0) {
7986 					lun->flags &= ~CTL_LUN_PR_RESERVED;
7987 					lun->res_type = 0;
7988 					lun->pr_res_idx = CTL_PR_NO_RESERVATION;
7989 				}
7990 			}
7991 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7992 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7993 			persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
7994 			persis_io.pr.pr_info.residx = residx;
7995 			if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7996 			     &persis_io, sizeof(persis_io), 0 )) >
7997 			     CTL_HA_STATUS_SUCCESS) {
7998 				printf("CTL:Persis Out error returned from "
7999 				       "ctl_ha_msg_send %d\n", isc_retval);
8000 			}
8001 			mtx_unlock(&softc->ctl_lock);
8002 		} else /* sa_res_key != 0 */ {
8003 
8004 			/*
8005 			 * If we aren't registered currently then increment
8006 			 * the key count and set the registered flag.
8007 			 */
8008 			if (!lun->per_res[residx].registered) {
8009 				lun->pr_key_count++;
8010 				lun->per_res[residx].registered = 1;
8011 			}
8012 
8013 			memcpy(&lun->per_res[residx].res_key,
8014 			       param->serv_act_res_key,
8015 			       ctl_min(sizeof(param->serv_act_res_key),
8016 			       sizeof(lun->per_res[residx].res_key)));
8017 
8018 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8019 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8020 			persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8021 			persis_io.pr.pr_info.residx = residx;
8022 			memcpy(persis_io.pr.pr_info.sa_res_key,
8023 			       param->serv_act_res_key,
8024 			       sizeof(param->serv_act_res_key));
8025 			mtx_unlock(&softc->ctl_lock);
8026 			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8027 			     &persis_io, sizeof(persis_io), 0)) >
8028 			     CTL_HA_STATUS_SUCCESS) {
8029 				printf("CTL:Persis Out error returned from "
8030 				       "ctl_ha_msg_send %d\n", isc_retval);
8031 			}
8032 		}
8033 		lun->PRGeneration++;
8034 
8035 		break;
8036 	}
8037 	case SPRO_RESERVE:
8038 #if 0
8039                 printf("Reserve executed type %d\n", type);
8040 #endif
8041 		mtx_lock(&softc->ctl_lock);
8042 		if (lun->flags & CTL_LUN_PR_RESERVED) {
8043 			/*
8044 			 * if this isn't the reservation holder and it's
8045 			 * not a "all registrants" type or if the type is
8046 			 * different then we have a conflict
8047 			 */
8048 			if ((lun->pr_res_idx != residx
8049 			  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8050 			 || lun->res_type != type) {
8051 				mtx_unlock(&softc->ctl_lock);
8052 				free(ctsio->kern_data_ptr, M_CTL);
8053 				ctl_set_reservation_conflict(ctsio);
8054 				ctl_done((union ctl_io *)ctsio);
8055 				return (CTL_RETVAL_COMPLETE);
8056 			}
8057 		} else /* create a reservation */ {
8058 			/*
8059 			 * If it's not an "all registrants" type record
8060 			 * reservation holder
8061 			 */
8062 			if (type != SPR_TYPE_WR_EX_AR
8063 			 && type != SPR_TYPE_EX_AC_AR)
8064 				lun->pr_res_idx = residx; /* Res holder */
8065 			else
8066 				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8067 
8068 			lun->flags |= CTL_LUN_PR_RESERVED;
8069 			lun->res_type = type;
8070 
8071 			mtx_unlock(&softc->ctl_lock);
8072 
8073 			/* send msg to other side */
8074 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8075 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8076 			persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8077 			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8078 			persis_io.pr.pr_info.res_type = type;
8079 			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8080 			     &persis_io, sizeof(persis_io), 0)) >
8081 			     CTL_HA_STATUS_SUCCESS) {
8082 				printf("CTL:Persis Out error returned from "
8083 				       "ctl_ha_msg_send %d\n", isc_retval);
8084 			}
8085 		}
8086 		break;
8087 
8088 	case SPRO_RELEASE:
8089 		mtx_lock(&softc->ctl_lock);
8090 		if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8091 			/* No reservation exists return good status */
8092 			mtx_unlock(&softc->ctl_lock);
8093 			goto done;
8094 		}
8095 		/*
8096 		 * Is this nexus a reservation holder?
8097 		 */
8098 		if (lun->pr_res_idx != residx
8099 		 && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8100 			/*
8101 			 * not a res holder return good status but
8102 			 * do nothing
8103 			 */
8104 			mtx_unlock(&softc->ctl_lock);
8105 			goto done;
8106 		}
8107 
8108 		if (lun->res_type != type) {
8109 			mtx_unlock(&softc->ctl_lock);
8110 			free(ctsio->kern_data_ptr, M_CTL);
8111 			ctl_set_illegal_pr_release(ctsio);
8112 			ctl_done((union ctl_io *)ctsio);
8113 			return (CTL_RETVAL_COMPLETE);
8114 		}
8115 
8116 		/* okay to release */
8117 		lun->flags &= ~CTL_LUN_PR_RESERVED;
8118 		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8119 		lun->res_type = 0;
8120 
8121 		/*
8122 		 * if this isn't an exclusive access
8123 		 * res generate UA for all other
8124 		 * registrants.
8125 		 */
8126 		if (type != SPR_TYPE_EX_AC
8127 		 && type != SPR_TYPE_WR_EX) {
8128 			/*
8129 			 * temporarily unregister so we don't generate UA
8130 			 */
8131 			lun->per_res[residx].registered = 0;
8132 
8133 			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8134 				if (lun->per_res[i+persis_offset].registered
8135 				    == 0)
8136 					continue;
8137 				lun->pending_sense[i].ua_pending |=
8138 					CTL_UA_RES_RELEASE;
8139 			}
8140 
8141 			lun->per_res[residx].registered = 1;
8142 		}
8143 		mtx_unlock(&softc->ctl_lock);
8144 		/* Send msg to other side */
8145 		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8146 		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8147 		persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8148 		if ((isc_retval=ctl_ha_msg_send( CTL_HA_CHAN_CTL, &persis_io,
8149 		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8150 			printf("CTL:Persis Out error returned from "
8151 			       "ctl_ha_msg_send %d\n", isc_retval);
8152 		}
8153 		break;
8154 
8155 	case SPRO_CLEAR:
8156 		/* send msg to other side */
8157 
8158 		mtx_lock(&softc->ctl_lock);
8159 		lun->flags &= ~CTL_LUN_PR_RESERVED;
8160 		lun->res_type = 0;
8161 		lun->pr_key_count = 0;
8162 		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8163 
8164 
8165 		memset(&lun->per_res[residx].res_key,
8166 		       0, sizeof(lun->per_res[residx].res_key));
8167 		lun->per_res[residx].registered = 0;
8168 
8169 		for (i=0; i < 2*CTL_MAX_INITIATORS; i++)
8170 			if (lun->per_res[i].registered) {
8171 				if (!persis_offset && i < CTL_MAX_INITIATORS)
8172 					lun->pending_sense[i].ua_pending |=
8173 						CTL_UA_RES_PREEMPT;
8174 				else if (persis_offset && i >= persis_offset)
8175 					lun->pending_sense[i-persis_offset
8176 					    ].ua_pending |= CTL_UA_RES_PREEMPT;
8177 
8178 				memset(&lun->per_res[i].res_key,
8179 				       0, sizeof(struct scsi_per_res_key));
8180 				lun->per_res[i].registered = 0;
8181 			}
8182 		lun->PRGeneration++;
8183 		mtx_unlock(&softc->ctl_lock);
8184 		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8185 		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8186 		persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8187 		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8188 		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8189 			printf("CTL:Persis Out error returned from "
8190 			       "ctl_ha_msg_send %d\n", isc_retval);
8191 		}
8192 		break;
8193 
8194 	case SPRO_PREEMPT: {
8195 		int nretval;
8196 
8197 		nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8198 					  residx, ctsio, cdb, param);
8199 		if (nretval != 0)
8200 			return (CTL_RETVAL_COMPLETE);
8201 		break;
8202 	}
8203 	case SPRO_REG_MOVE:
8204 	case SPRO_PRE_ABO:
8205 	default:
8206 		free(ctsio->kern_data_ptr, M_CTL);
8207 		ctl_set_invalid_field(/*ctsio*/ ctsio,
8208 				      /*sks_valid*/ 1,
8209 				      /*command*/ 1,
8210 				      /*field*/ 1,
8211 				      /*bit_valid*/ 1,
8212 				      /*bit*/ 0);
8213 		ctl_done((union ctl_io *)ctsio);
8214 		return (CTL_RETVAL_COMPLETE);
8215 		break; /* NOTREACHED */
8216 	}
8217 
8218 done:
8219 	free(ctsio->kern_data_ptr, M_CTL);
8220 	ctl_set_success(ctsio);
8221 	ctl_done((union ctl_io *)ctsio);
8222 
8223 	return (retval);
8224 }
8225 
8226 /*
8227  * This routine is for handling a message from the other SC pertaining to
8228  * persistent reserve out. All the error checking will have been done
8229  * so only perorming the action need be done here to keep the two
8230  * in sync.
8231  */
8232 static void
8233 ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8234 {
8235 	struct ctl_lun *lun;
8236 	struct ctl_softc *softc;
8237 	int i;
8238 
8239 	softc = control_softc;
8240 
8241 	mtx_lock(&softc->ctl_lock);
8242 
8243 	lun = softc->ctl_luns[msg->hdr.nexus.targ_lun];
8244 	switch(msg->pr.pr_info.action) {
8245 	case CTL_PR_REG_KEY:
8246 		if (!lun->per_res[msg->pr.pr_info.residx].registered) {
8247 			lun->per_res[msg->pr.pr_info.residx].registered = 1;
8248 			lun->pr_key_count++;
8249 		}
8250 		lun->PRGeneration++;
8251 		memcpy(&lun->per_res[msg->pr.pr_info.residx].res_key,
8252 		       msg->pr.pr_info.sa_res_key,
8253 		       sizeof(struct scsi_per_res_key));
8254 		break;
8255 
8256 	case CTL_PR_UNREG_KEY:
8257 		lun->per_res[msg->pr.pr_info.residx].registered = 0;
8258 		memset(&lun->per_res[msg->pr.pr_info.residx].res_key,
8259 		       0, sizeof(struct scsi_per_res_key));
8260 		lun->pr_key_count--;
8261 
8262 		/* XXX Need to see if the reservation has been released */
8263 		/* if so do we need to generate UA? */
8264 		if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8265 			lun->flags &= ~CTL_LUN_PR_RESERVED;
8266 			lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8267 
8268 			if ((lun->res_type == SPR_TYPE_WR_EX_RO
8269 			  || lun->res_type == SPR_TYPE_EX_AC_RO)
8270 			 && lun->pr_key_count) {
8271 				/*
8272 				 * If the reservation is a registrants
8273 				 * only type we need to generate a UA
8274 				 * for other registered inits.  The
8275 				 * sense code should be RESERVATIONS
8276 				 * RELEASED
8277 				 */
8278 
8279 				for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8280 					if (lun->per_res[i+
8281 					    persis_offset].registered == 0)
8282 						continue;
8283 
8284 					lun->pending_sense[i
8285 						].ua_pending |=
8286 						CTL_UA_RES_RELEASE;
8287 				}
8288 			}
8289 			lun->res_type = 0;
8290 		} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8291 			if (lun->pr_key_count==0) {
8292 				lun->flags &= ~CTL_LUN_PR_RESERVED;
8293 				lun->res_type = 0;
8294 				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8295 			}
8296 		}
8297 		lun->PRGeneration++;
8298 		break;
8299 
8300 	case CTL_PR_RESERVE:
8301 		lun->flags |= CTL_LUN_PR_RESERVED;
8302 		lun->res_type = msg->pr.pr_info.res_type;
8303 		lun->pr_res_idx = msg->pr.pr_info.residx;
8304 
8305 		break;
8306 
8307 	case CTL_PR_RELEASE:
8308 		/*
8309 		 * if this isn't an exclusive access res generate UA for all
8310 		 * other registrants.
8311 		 */
8312 		if (lun->res_type != SPR_TYPE_EX_AC
8313 		 && lun->res_type != SPR_TYPE_WR_EX) {
8314 			for (i = 0; i < CTL_MAX_INITIATORS; i++)
8315 				if (lun->per_res[i+persis_offset].registered)
8316 					lun->pending_sense[i].ua_pending |=
8317 						CTL_UA_RES_RELEASE;
8318 		}
8319 
8320 		lun->flags &= ~CTL_LUN_PR_RESERVED;
8321 		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8322 		lun->res_type = 0;
8323 		break;
8324 
8325 	case CTL_PR_PREEMPT:
8326 		ctl_pro_preempt_other(lun, msg);
8327 		break;
8328 	case CTL_PR_CLEAR:
8329 		lun->flags &= ~CTL_LUN_PR_RESERVED;
8330 		lun->res_type = 0;
8331 		lun->pr_key_count = 0;
8332 		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8333 
8334 		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8335 			if (lun->per_res[i].registered == 0)
8336 				continue;
8337 			if (!persis_offset
8338 			 && i < CTL_MAX_INITIATORS)
8339 				lun->pending_sense[i].ua_pending |=
8340 					CTL_UA_RES_PREEMPT;
8341 			else if (persis_offset
8342 			      && i >= persis_offset)
8343    				lun->pending_sense[i-persis_offset].ua_pending|=
8344 					CTL_UA_RES_PREEMPT;
8345 			memset(&lun->per_res[i].res_key, 0,
8346 			       sizeof(struct scsi_per_res_key));
8347 			lun->per_res[i].registered = 0;
8348 		}
8349 		lun->PRGeneration++;
8350 		break;
8351 	}
8352 
8353 	mtx_unlock(&softc->ctl_lock);
8354 }
8355 
8356 int
8357 ctl_read_write(struct ctl_scsiio *ctsio)
8358 {
8359 	struct ctl_lun *lun;
8360 	struct ctl_lba_len lbalen;
8361 	uint64_t lba;
8362 	uint32_t num_blocks;
8363 	int reladdr, fua, dpo, ebp;
8364 	int retval;
8365 	int isread;
8366 
8367 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8368 
8369 	CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8370 
8371 	reladdr = 0;
8372 	fua = 0;
8373 	dpo = 0;
8374 	ebp = 0;
8375 
8376 	retval = CTL_RETVAL_COMPLETE;
8377 
8378 	isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8379 	      || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8380 	if (lun->flags & CTL_LUN_PR_RESERVED && isread) {
8381 		uint32_t residx;
8382 
8383 		/*
8384 		 * XXX KDM need a lock here.
8385 		 */
8386 		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8387 		if ((lun->res_type == SPR_TYPE_EX_AC
8388 		  && residx != lun->pr_res_idx)
8389 		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
8390 		   || lun->res_type == SPR_TYPE_EX_AC_AR)
8391 		  && !lun->per_res[residx].registered)) {
8392 			ctl_set_reservation_conflict(ctsio);
8393 			ctl_done((union ctl_io *)ctsio);
8394 			return (CTL_RETVAL_COMPLETE);
8395 	        }
8396 	}
8397 
8398 	switch (ctsio->cdb[0]) {
8399 	case READ_6:
8400 	case WRITE_6: {
8401 		struct scsi_rw_6 *cdb;
8402 
8403 		cdb = (struct scsi_rw_6 *)ctsio->cdb;
8404 
8405 		lba = scsi_3btoul(cdb->addr);
8406 		/* only 5 bits are valid in the most significant address byte */
8407 		lba &= 0x1fffff;
8408 		num_blocks = cdb->length;
8409 		/*
8410 		 * This is correct according to SBC-2.
8411 		 */
8412 		if (num_blocks == 0)
8413 			num_blocks = 256;
8414 		break;
8415 	}
8416 	case READ_10:
8417 	case WRITE_10: {
8418 		struct scsi_rw_10 *cdb;
8419 
8420 		cdb = (struct scsi_rw_10 *)ctsio->cdb;
8421 
8422 		if (cdb->byte2 & SRW10_RELADDR)
8423 			reladdr = 1;
8424 		if (cdb->byte2 & SRW10_FUA)
8425 			fua = 1;
8426 		if (cdb->byte2 & SRW10_DPO)
8427 			dpo = 1;
8428 
8429 		if ((cdb->opcode == WRITE_10)
8430 		 && (cdb->byte2 & SRW10_EBP))
8431 			ebp = 1;
8432 
8433 		lba = scsi_4btoul(cdb->addr);
8434 		num_blocks = scsi_2btoul(cdb->length);
8435 		break;
8436 	}
8437 	case WRITE_VERIFY_10: {
8438 		struct scsi_write_verify_10 *cdb;
8439 
8440 		cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8441 
8442 		/*
8443 		 * XXX KDM we should do actual write verify support at some
8444 		 * point.  This is obviously fake, we're just translating
8445 		 * things to a write.  So we don't even bother checking the
8446 		 * BYTCHK field, since we don't do any verification.  If
8447 		 * the user asks for it, we'll just pretend we did it.
8448 		 */
8449 		if (cdb->byte2 & SWV_DPO)
8450 			dpo = 1;
8451 
8452 		lba = scsi_4btoul(cdb->addr);
8453 		num_blocks = scsi_2btoul(cdb->length);
8454 		break;
8455 	}
8456 	case READ_12:
8457 	case WRITE_12: {
8458 		struct scsi_rw_12 *cdb;
8459 
8460 		cdb = (struct scsi_rw_12 *)ctsio->cdb;
8461 
8462 		if (cdb->byte2 & SRW12_RELADDR)
8463 			reladdr = 1;
8464 		if (cdb->byte2 & SRW12_FUA)
8465 			fua = 1;
8466 		if (cdb->byte2 & SRW12_DPO)
8467 			dpo = 1;
8468 		lba = scsi_4btoul(cdb->addr);
8469 		num_blocks = scsi_4btoul(cdb->length);
8470 		break;
8471 	}
8472 	case WRITE_VERIFY_12: {
8473 		struct scsi_write_verify_12 *cdb;
8474 
8475 		cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8476 
8477 		if (cdb->byte2 & SWV_DPO)
8478 			dpo = 1;
8479 
8480 		lba = scsi_4btoul(cdb->addr);
8481 		num_blocks = scsi_4btoul(cdb->length);
8482 
8483 		break;
8484 	}
8485 	case READ_16:
8486 	case WRITE_16: {
8487 		struct scsi_rw_16 *cdb;
8488 
8489 		cdb = (struct scsi_rw_16 *)ctsio->cdb;
8490 
8491 		if (cdb->byte2 & SRW12_RELADDR)
8492 			reladdr = 1;
8493 		if (cdb->byte2 & SRW12_FUA)
8494 			fua = 1;
8495 		if (cdb->byte2 & SRW12_DPO)
8496 			dpo = 1;
8497 
8498 		lba = scsi_8btou64(cdb->addr);
8499 		num_blocks = scsi_4btoul(cdb->length);
8500 		break;
8501 	}
8502 	case WRITE_VERIFY_16: {
8503 		struct scsi_write_verify_16 *cdb;
8504 
8505 		cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
8506 
8507 		if (cdb->byte2 & SWV_DPO)
8508 			dpo = 1;
8509 
8510 		lba = scsi_8btou64(cdb->addr);
8511 		num_blocks = scsi_4btoul(cdb->length);
8512 		break;
8513 	}
8514 	default:
8515 		/*
8516 		 * We got a command we don't support.  This shouldn't
8517 		 * happen, commands should be filtered out above us.
8518 		 */
8519 		ctl_set_invalid_opcode(ctsio);
8520 		ctl_done((union ctl_io *)ctsio);
8521 
8522 		return (CTL_RETVAL_COMPLETE);
8523 		break; /* NOTREACHED */
8524 	}
8525 
8526 	/*
8527 	 * XXX KDM what do we do with the DPO and FUA bits?  FUA might be
8528 	 * interesting for us, but if RAIDCore is in write-back mode,
8529 	 * getting it to do write-through for a particular transaction may
8530 	 * not be possible.
8531 	 */
8532 	/*
8533 	 * We don't support relative addressing.  That also requires
8534 	 * supporting linked commands, which we don't do.
8535 	 */
8536 	if (reladdr != 0) {
8537 		ctl_set_invalid_field(ctsio,
8538 				      /*sks_valid*/ 1,
8539 				      /*command*/ 1,
8540 				      /*field*/ 1,
8541 				      /*bit_valid*/ 1,
8542 				      /*bit*/ 0);
8543 		ctl_done((union ctl_io *)ctsio);
8544 		return (CTL_RETVAL_COMPLETE);
8545 	}
8546 
8547 	/*
8548 	 * The first check is to make sure we're in bounds, the second
8549 	 * check is to catch wrap-around problems.  If the lba + num blocks
8550 	 * is less than the lba, then we've wrapped around and the block
8551 	 * range is invalid anyway.
8552 	 */
8553 	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8554 	 || ((lba + num_blocks) < lba)) {
8555 		ctl_set_lba_out_of_range(ctsio);
8556 		ctl_done((union ctl_io *)ctsio);
8557 		return (CTL_RETVAL_COMPLETE);
8558 	}
8559 
8560 	/*
8561 	 * According to SBC-3, a transfer length of 0 is not an error.
8562 	 * Note that this cannot happen with WRITE(6) or READ(6), since 0
8563 	 * translates to 256 blocks for those commands.
8564 	 */
8565 	if (num_blocks == 0) {
8566 		ctl_set_success(ctsio);
8567 		ctl_done((union ctl_io *)ctsio);
8568 		return (CTL_RETVAL_COMPLETE);
8569 	}
8570 
8571 	lbalen.lba = lba;
8572 	lbalen.len = num_blocks;
8573 	memcpy(ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes, &lbalen,
8574 	       sizeof(lbalen));
8575 
8576 	CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
8577 
8578 	retval = lun->backend->data_submit((union ctl_io *)ctsio);
8579 
8580 	return (retval);
8581 }
8582 
8583 int
8584 ctl_report_luns(struct ctl_scsiio *ctsio)
8585 {
8586 	struct scsi_report_luns *cdb;
8587 	struct scsi_report_luns_data *lun_data;
8588 	struct ctl_lun *lun, *request_lun;
8589 	int num_luns, retval;
8590 	uint32_t alloc_len, lun_datalen;
8591 	int num_filled, well_known;
8592 	uint32_t initidx;
8593 
8594 	retval = CTL_RETVAL_COMPLETE;
8595 	well_known = 0;
8596 
8597 	cdb = (struct scsi_report_luns *)ctsio->cdb;
8598 
8599 	CTL_DEBUG_PRINT(("ctl_report_luns\n"));
8600 
8601 	mtx_lock(&control_softc->ctl_lock);
8602 	num_luns = control_softc->num_luns;
8603 	mtx_unlock(&control_softc->ctl_lock);
8604 
8605 	switch (cdb->select_report) {
8606 	case RPL_REPORT_DEFAULT:
8607 	case RPL_REPORT_ALL:
8608 		break;
8609 	case RPL_REPORT_WELLKNOWN:
8610 		well_known = 1;
8611 		num_luns = 0;
8612 		break;
8613 	default:
8614 		ctl_set_invalid_field(ctsio,
8615 				      /*sks_valid*/ 1,
8616 				      /*command*/ 1,
8617 				      /*field*/ 2,
8618 				      /*bit_valid*/ 0,
8619 				      /*bit*/ 0);
8620 		ctl_done((union ctl_io *)ctsio);
8621 		return (retval);
8622 		break; /* NOTREACHED */
8623 	}
8624 
8625 	alloc_len = scsi_4btoul(cdb->length);
8626 	/*
8627 	 * The initiator has to allocate at least 16 bytes for this request,
8628 	 * so he can at least get the header and the first LUN.  Otherwise
8629 	 * we reject the request (per SPC-3 rev 14, section 6.21).
8630 	 */
8631 	if (alloc_len < (sizeof(struct scsi_report_luns_data) +
8632 	    sizeof(struct scsi_report_luns_lundata))) {
8633 		ctl_set_invalid_field(ctsio,
8634 				      /*sks_valid*/ 1,
8635 				      /*command*/ 1,
8636 				      /*field*/ 6,
8637 				      /*bit_valid*/ 0,
8638 				      /*bit*/ 0);
8639 		ctl_done((union ctl_io *)ctsio);
8640 		return (retval);
8641 	}
8642 
8643 	request_lun = (struct ctl_lun *)
8644 		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8645 
8646 	lun_datalen = sizeof(*lun_data) +
8647 		(num_luns * sizeof(struct scsi_report_luns_lundata));
8648 
8649 	ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK);
8650 	if (ctsio->kern_data_ptr == NULL) {
8651 		ctsio->io_hdr.status = CTL_SCSI_ERROR;
8652 		ctsio->scsi_status = SCSI_STATUS_BUSY;
8653 		ctl_done((union ctl_io *)ctsio);
8654 		return (CTL_RETVAL_COMPLETE);
8655 	}
8656 
8657 	lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
8658 	ctsio->kern_sg_entries = 0;
8659 
8660 	if (lun_datalen < alloc_len) {
8661 		ctsio->residual = alloc_len - lun_datalen;
8662 		ctsio->kern_data_len = lun_datalen;
8663 		ctsio->kern_total_len = lun_datalen;
8664 	} else {
8665 		ctsio->residual = 0;
8666 		ctsio->kern_data_len = alloc_len;
8667 		ctsio->kern_total_len = alloc_len;
8668 	}
8669 	ctsio->kern_data_resid = 0;
8670 	ctsio->kern_rel_offset = 0;
8671 	ctsio->kern_sg_entries = 0;
8672 
8673 	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
8674 
8675 	memset(lun_data, 0, lun_datalen);
8676 
8677 	/*
8678 	 * We set this to the actual data length, regardless of how much
8679 	 * space we actually have to return results.  If the user looks at
8680 	 * this value, he'll know whether or not he allocated enough space
8681 	 * and reissue the command if necessary.  We don't support well
8682 	 * known logical units, so if the user asks for that, return none.
8683 	 */
8684 	scsi_ulto4b(lun_datalen - 8, lun_data->length);
8685 
8686 	mtx_lock(&control_softc->ctl_lock);
8687 	for (num_filled = 0, lun = STAILQ_FIRST(&control_softc->lun_list);
8688 	     (lun != NULL) && (num_filled < num_luns);
8689 	     lun = STAILQ_NEXT(lun, links)) {
8690 
8691 		if (lun->lun <= 0xff) {
8692 			/*
8693 			 * Peripheral addressing method, bus number 0.
8694 			 */
8695 			lun_data->luns[num_filled].lundata[0] =
8696 				RPL_LUNDATA_ATYP_PERIPH;
8697 			lun_data->luns[num_filled].lundata[1] = lun->lun;
8698 			num_filled++;
8699 		} else if (lun->lun <= 0x3fff) {
8700 			/*
8701 			 * Flat addressing method.
8702 			 */
8703 			lun_data->luns[num_filled].lundata[0] =
8704 				RPL_LUNDATA_ATYP_FLAT |
8705 				(lun->lun & RPL_LUNDATA_FLAT_LUN_MASK);
8706 #ifdef OLDCTLHEADERS
8707 				(SRLD_ADDR_FLAT << SRLD_ADDR_SHIFT) |
8708 				(lun->lun & SRLD_BUS_LUN_MASK);
8709 #endif
8710 			lun_data->luns[num_filled].lundata[1] =
8711 #ifdef OLDCTLHEADERS
8712 				lun->lun >> SRLD_BUS_LUN_BITS;
8713 #endif
8714 				lun->lun >> RPL_LUNDATA_FLAT_LUN_BITS;
8715 			num_filled++;
8716 		} else {
8717 			printf("ctl_report_luns: bogus LUN number %jd, "
8718 			       "skipping\n", (intmax_t)lun->lun);
8719 		}
8720 		/*
8721 		 * According to SPC-3, rev 14 section 6.21:
8722 		 *
8723 		 * "The execution of a REPORT LUNS command to any valid and
8724 		 * installed logical unit shall clear the REPORTED LUNS DATA
8725 		 * HAS CHANGED unit attention condition for all logical
8726 		 * units of that target with respect to the requesting
8727 		 * initiator. A valid and installed logical unit is one
8728 		 * having a PERIPHERAL QUALIFIER of 000b in the standard
8729 		 * INQUIRY data (see 6.4.2)."
8730 		 *
8731 		 * If request_lun is NULL, the LUN this report luns command
8732 		 * was issued to is either disabled or doesn't exist. In that
8733 		 * case, we shouldn't clear any pending lun change unit
8734 		 * attention.
8735 		 */
8736 		if (request_lun != NULL)
8737 			lun->pending_sense[initidx].ua_pending &=
8738 				~CTL_UA_LUN_CHANGE;
8739 	}
8740 	mtx_unlock(&control_softc->ctl_lock);
8741 
8742 	/*
8743 	 * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
8744 	 * this request.
8745 	 */
8746 	ctsio->scsi_status = SCSI_STATUS_OK;
8747 
8748 	ctsio->be_move_done = ctl_config_move_done;
8749 	ctl_datamove((union ctl_io *)ctsio);
8750 
8751 	return (retval);
8752 }
8753 
8754 int
8755 ctl_request_sense(struct ctl_scsiio *ctsio)
8756 {
8757 	struct scsi_request_sense *cdb;
8758 	struct scsi_sense_data *sense_ptr;
8759 	struct ctl_lun *lun;
8760 	uint32_t initidx;
8761 	int have_error;
8762 	scsi_sense_data_type sense_format;
8763 
8764 	cdb = (struct scsi_request_sense *)ctsio->cdb;
8765 
8766 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8767 
8768 	CTL_DEBUG_PRINT(("ctl_request_sense\n"));
8769 
8770 	/*
8771 	 * Determine which sense format the user wants.
8772 	 */
8773 	if (cdb->byte2 & SRS_DESC)
8774 		sense_format = SSD_TYPE_DESC;
8775 	else
8776 		sense_format = SSD_TYPE_FIXED;
8777 
8778 	ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
8779 	if (ctsio->kern_data_ptr == NULL) {
8780 		ctsio->io_hdr.status = CTL_SCSI_ERROR;
8781 		ctsio->scsi_status = SCSI_STATUS_BUSY;
8782 		ctl_done((union ctl_io *)ctsio);
8783 		return (CTL_RETVAL_COMPLETE);
8784 	}
8785 	sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
8786 	ctsio->kern_sg_entries = 0;
8787 
8788 	/*
8789 	 * struct scsi_sense_data, which is currently set to 256 bytes, is
8790 	 * larger than the largest allowed value for the length field in the
8791 	 * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
8792 	 */
8793 	ctsio->residual = 0;
8794 	ctsio->kern_data_len = cdb->length;
8795 	ctsio->kern_total_len = cdb->length;
8796 
8797 	ctsio->kern_data_resid = 0;
8798 	ctsio->kern_rel_offset = 0;
8799 	ctsio->kern_sg_entries = 0;
8800 
8801 	/*
8802 	 * If we don't have a LUN, we don't have any pending sense.
8803 	 */
8804 	if (lun == NULL)
8805 		goto no_sense;
8806 
8807 	have_error = 0;
8808 	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
8809 	/*
8810 	 * Check for pending sense, and then for pending unit attentions.
8811 	 * Pending sense gets returned first, then pending unit attentions.
8812 	 */
8813 	mtx_lock(&lun->ctl_softc->ctl_lock);
8814 	if (ctl_is_set(lun->have_ca, initidx)) {
8815 		scsi_sense_data_type stored_format;
8816 
8817 		/*
8818 		 * Check to see which sense format was used for the stored
8819 		 * sense data.
8820 		 */
8821 		stored_format = scsi_sense_type(
8822 		    &lun->pending_sense[initidx].sense);
8823 
8824 		/*
8825 		 * If the user requested a different sense format than the
8826 		 * one we stored, then we need to convert it to the other
8827 		 * format.  If we're going from descriptor to fixed format
8828 		 * sense data, we may lose things in translation, depending
8829 		 * on what options were used.
8830 		 *
8831 		 * If the stored format is SSD_TYPE_NONE (i.e. invalid),
8832 		 * for some reason we'll just copy it out as-is.
8833 		 */
8834 		if ((stored_format == SSD_TYPE_FIXED)
8835 		 && (sense_format == SSD_TYPE_DESC))
8836 			ctl_sense_to_desc((struct scsi_sense_data_fixed *)
8837 			    &lun->pending_sense[initidx].sense,
8838 			    (struct scsi_sense_data_desc *)sense_ptr);
8839 		else if ((stored_format == SSD_TYPE_DESC)
8840 		      && (sense_format == SSD_TYPE_FIXED))
8841 			ctl_sense_to_fixed((struct scsi_sense_data_desc *)
8842 			    &lun->pending_sense[initidx].sense,
8843 			    (struct scsi_sense_data_fixed *)sense_ptr);
8844 		else
8845 			memcpy(sense_ptr, &lun->pending_sense[initidx].sense,
8846 			       ctl_min(sizeof(*sense_ptr),
8847 			       sizeof(lun->pending_sense[initidx].sense)));
8848 
8849 		ctl_clear_mask(lun->have_ca, initidx);
8850 		have_error = 1;
8851 	} else if (lun->pending_sense[initidx].ua_pending != CTL_UA_NONE) {
8852 		ctl_ua_type ua_type;
8853 
8854 		ua_type = ctl_build_ua(lun->pending_sense[initidx].ua_pending,
8855 				       sense_ptr, sense_format);
8856 		if (ua_type != CTL_UA_NONE) {
8857 			have_error = 1;
8858 			/* We're reporting this UA, so clear it */
8859 			lun->pending_sense[initidx].ua_pending &= ~ua_type;
8860 		}
8861 	}
8862 	mtx_unlock(&lun->ctl_softc->ctl_lock);
8863 
8864 	/*
8865 	 * We already have a pending error, return it.
8866 	 */
8867 	if (have_error != 0) {
8868 		/*
8869 		 * We report the SCSI status as OK, since the status of the
8870 		 * request sense command itself is OK.
8871 		 */
8872 		ctsio->scsi_status = SCSI_STATUS_OK;
8873 
8874 		/*
8875 		 * We report 0 for the sense length, because we aren't doing
8876 		 * autosense in this case.  We're reporting sense as
8877 		 * parameter data.
8878 		 */
8879 		ctsio->sense_len = 0;
8880 
8881 		ctsio->be_move_done = ctl_config_move_done;
8882 		ctl_datamove((union ctl_io *)ctsio);
8883 
8884 		return (CTL_RETVAL_COMPLETE);
8885 	}
8886 
8887 no_sense:
8888 
8889 	/*
8890 	 * No sense information to report, so we report that everything is
8891 	 * okay.
8892 	 */
8893 	ctl_set_sense_data(sense_ptr,
8894 			   lun,
8895 			   sense_format,
8896 			   /*current_error*/ 1,
8897 			   /*sense_key*/ SSD_KEY_NO_SENSE,
8898 			   /*asc*/ 0x00,
8899 			   /*ascq*/ 0x00,
8900 			   SSD_ELEM_NONE);
8901 
8902 	ctsio->scsi_status = SCSI_STATUS_OK;
8903 
8904 	/*
8905 	 * We report 0 for the sense length, because we aren't doing
8906 	 * autosense in this case.  We're reporting sense as parameter data.
8907 	 */
8908 	ctsio->sense_len = 0;
8909 	ctsio->be_move_done = ctl_config_move_done;
8910 	ctl_datamove((union ctl_io *)ctsio);
8911 
8912 	return (CTL_RETVAL_COMPLETE);
8913 }
8914 
8915 int
8916 ctl_tur(struct ctl_scsiio *ctsio)
8917 {
8918 	struct ctl_lun *lun;
8919 
8920 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8921 
8922 	CTL_DEBUG_PRINT(("ctl_tur\n"));
8923 
8924 	if (lun == NULL)
8925 		return (-EINVAL);
8926 
8927 	ctsio->scsi_status = SCSI_STATUS_OK;
8928 	ctsio->io_hdr.status = CTL_SUCCESS;
8929 
8930 	ctl_done((union ctl_io *)ctsio);
8931 
8932 	return (CTL_RETVAL_COMPLETE);
8933 }
8934 
8935 #ifdef notyet
8936 static int
8937 ctl_cmddt_inquiry(struct ctl_scsiio *ctsio)
8938 {
8939 
8940 }
8941 #endif
8942 
8943 static int
8944 ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
8945 {
8946 	struct scsi_vpd_supported_pages *pages;
8947 	int sup_page_size;
8948 	struct ctl_lun *lun;
8949 
8950 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8951 
8952 	sup_page_size = sizeof(struct scsi_vpd_supported_pages) +
8953 		SCSI_EVPD_NUM_SUPPORTED_PAGES;
8954 	/*
8955 	 * XXX KDM GFP_???  We probably don't want to wait here,
8956 	 * unless we end up having a process/thread context.
8957 	 */
8958 	ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK);
8959 	if (ctsio->kern_data_ptr == NULL) {
8960 		ctsio->io_hdr.status = CTL_SCSI_ERROR;
8961 		ctsio->scsi_status = SCSI_STATUS_BUSY;
8962 		ctl_done((union ctl_io *)ctsio);
8963 		return (CTL_RETVAL_COMPLETE);
8964 	}
8965 	pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
8966 	ctsio->kern_sg_entries = 0;
8967 
8968 	if (sup_page_size < alloc_len) {
8969 		ctsio->residual = alloc_len - sup_page_size;
8970 		ctsio->kern_data_len = sup_page_size;
8971 		ctsio->kern_total_len = sup_page_size;
8972 	} else {
8973 		ctsio->residual = 0;
8974 		ctsio->kern_data_len = alloc_len;
8975 		ctsio->kern_total_len = alloc_len;
8976 	}
8977 	ctsio->kern_data_resid = 0;
8978 	ctsio->kern_rel_offset = 0;
8979 	ctsio->kern_sg_entries = 0;
8980 
8981 	memset(pages, 0, sup_page_size);
8982 
8983 	/*
8984 	 * The control device is always connected.  The disk device, on the
8985 	 * other hand, may not be online all the time.  Need to change this
8986 	 * to figure out whether the disk device is actually online or not.
8987 	 */
8988 	if (lun != NULL)
8989 		pages->device = (SID_QUAL_LU_CONNECTED << 5) |
8990 				lun->be_lun->lun_type;
8991 	else
8992 		pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
8993 
8994 	pages->length = SCSI_EVPD_NUM_SUPPORTED_PAGES;
8995 	/* Supported VPD pages */
8996 	pages->page_list[0] = SVPD_SUPPORTED_PAGES;
8997 	/* Serial Number */
8998 	pages->page_list[1] = SVPD_UNIT_SERIAL_NUMBER;
8999 	/* Device Identification */
9000 	pages->page_list[2] = SVPD_DEVICE_ID;
9001 
9002 	ctsio->scsi_status = SCSI_STATUS_OK;
9003 
9004 	ctsio->be_move_done = ctl_config_move_done;
9005 	ctl_datamove((union ctl_io *)ctsio);
9006 
9007 	return (CTL_RETVAL_COMPLETE);
9008 }
9009 
9010 static int
9011 ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9012 {
9013 	struct scsi_vpd_unit_serial_number *sn_ptr;
9014 	struct ctl_lun *lun;
9015 #ifndef CTL_USE_BACKEND_SN
9016 	char tmpstr[32];
9017 #endif
9018 
9019 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9020 
9021 	/* XXX KDM which malloc flags here?? */
9022 	ctsio->kern_data_ptr = malloc(sizeof(*sn_ptr), M_CTL, M_WAITOK);
9023 	if (ctsio->kern_data_ptr == NULL) {
9024 		ctsio->io_hdr.status = CTL_SCSI_ERROR;
9025 		ctsio->scsi_status = SCSI_STATUS_BUSY;
9026 		ctl_done((union ctl_io *)ctsio);
9027 		return (CTL_RETVAL_COMPLETE);
9028 	}
9029 	sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9030 	ctsio->kern_sg_entries = 0;
9031 
9032 	if (sizeof(*sn_ptr) < alloc_len) {
9033 		ctsio->residual = alloc_len - sizeof(*sn_ptr);
9034 		ctsio->kern_data_len = sizeof(*sn_ptr);
9035 		ctsio->kern_total_len = sizeof(*sn_ptr);
9036 	} else {
9037 		ctsio->residual = 0;
9038 		ctsio->kern_data_len = alloc_len;
9039 		ctsio->kern_total_len = alloc_len;
9040 	}
9041 	ctsio->kern_data_resid = 0;
9042 	ctsio->kern_rel_offset = 0;
9043 	ctsio->kern_sg_entries = 0;
9044 
9045 	memset(sn_ptr, 0, sizeof(*sn_ptr));
9046 
9047 	/*
9048 	 * The control device is always connected.  The disk device, on the
9049 	 * other hand, may not be online all the time.  Need to change this
9050 	 * to figure out whether the disk device is actually online or not.
9051 	 */
9052 	if (lun != NULL)
9053 		sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9054 				  lun->be_lun->lun_type;
9055 	else
9056 		sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9057 
9058 	sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9059 	sn_ptr->length = ctl_min(sizeof(*sn_ptr) - 4, CTL_SN_LEN);
9060 #ifdef CTL_USE_BACKEND_SN
9061 	/*
9062 	 * If we don't have a LUN, we just leave the serial number as
9063 	 * all spaces.
9064 	 */
9065 	memset(sn_ptr->serial_num, 0x20, sizeof(sn_ptr->serial_num));
9066 	if (lun != NULL) {
9067 		strncpy((char *)sn_ptr->serial_num,
9068 			(char *)lun->be_lun->serial_num, CTL_SN_LEN);
9069 	}
9070 #else
9071 	/*
9072 	 * Note that we're using a non-unique serial number here,
9073 	 */
9074 	snprintf(tmpstr, sizeof(tmpstr), "MYSERIALNUMIS000");
9075 	memset(sn_ptr->serial_num, 0x20, sizeof(sn_ptr->serial_num));
9076 	strncpy(sn_ptr->serial_num, tmpstr, ctl_min(CTL_SN_LEN,
9077 		ctl_min(sizeof(tmpstr), sizeof(*sn_ptr) - 4)));
9078 #endif
9079 	ctsio->scsi_status = SCSI_STATUS_OK;
9080 
9081 	ctsio->be_move_done = ctl_config_move_done;
9082 	ctl_datamove((union ctl_io *)ctsio);
9083 
9084 	return (CTL_RETVAL_COMPLETE);
9085 }
9086 
9087 
9088 static int
9089 ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9090 {
9091 	struct scsi_vpd_device_id *devid_ptr;
9092 	struct scsi_vpd_id_descriptor *desc, *desc1;
9093 	struct scsi_vpd_id_descriptor *desc2, *desc3; /* for types 4h and 5h */
9094 	struct scsi_vpd_id_t10 *t10id;
9095 	struct ctl_softc *ctl_softc;
9096 	struct ctl_lun *lun;
9097 	struct ctl_frontend *fe;
9098 #ifndef CTL_USE_BACKEND_SN
9099 	char tmpstr[32];
9100 #endif /* CTL_USE_BACKEND_SN */
9101 	int devid_len;
9102 
9103 	ctl_softc = control_softc;
9104 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9105 
9106 	devid_len = sizeof(struct scsi_vpd_device_id) +
9107 		sizeof(struct scsi_vpd_id_descriptor) +
9108 		sizeof(struct scsi_vpd_id_t10) + CTL_DEVID_LEN +
9109 		sizeof(struct scsi_vpd_id_descriptor) + CTL_WWPN_LEN +
9110 		sizeof(struct scsi_vpd_id_descriptor) +
9111 		sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9112 		sizeof(struct scsi_vpd_id_descriptor) +
9113 		sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9114 
9115 	/* XXX KDM which malloc flags here ?? */
9116 	ctsio->kern_data_ptr = malloc(devid_len, M_CTL, M_WAITOK);
9117 	if (ctsio->kern_data_ptr == NULL) {
9118 		ctsio->io_hdr.status = CTL_SCSI_ERROR;
9119 		ctsio->scsi_status = SCSI_STATUS_BUSY;
9120 		ctl_done((union ctl_io *)ctsio);
9121 		return (CTL_RETVAL_COMPLETE);
9122 	}
9123 	devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9124 	ctsio->kern_sg_entries = 0;
9125 
9126 	if (devid_len < alloc_len) {
9127 		ctsio->residual = alloc_len - devid_len;
9128 		ctsio->kern_data_len = devid_len;
9129 		ctsio->kern_total_len = devid_len;
9130 	} else {
9131 		ctsio->residual = 0;
9132 		ctsio->kern_data_len = alloc_len;
9133 		ctsio->kern_total_len = alloc_len;
9134 	}
9135 	ctsio->kern_data_resid = 0;
9136 	ctsio->kern_rel_offset = 0;
9137 	ctsio->kern_sg_entries = 0;
9138 
9139 	desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9140 	t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
9141 	desc1 = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9142 		sizeof(struct scsi_vpd_id_t10) + CTL_DEVID_LEN);
9143 	desc2 = (struct scsi_vpd_id_descriptor *)(&desc1->identifier[0] +
9144 	          CTL_WWPN_LEN);
9145 	desc3 = (struct scsi_vpd_id_descriptor *)(&desc2->identifier[0] +
9146 	         sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9147 	memset(devid_ptr, 0, devid_len);
9148 
9149 	/*
9150 	 * The control device is always connected.  The disk device, on the
9151 	 * other hand, may not be online all the time.
9152 	 */
9153 	if (lun != NULL)
9154 		devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9155 				     lun->be_lun->lun_type;
9156 	else
9157 		devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9158 
9159 	devid_ptr->page_code = SVPD_DEVICE_ID;
9160 
9161 	scsi_ulto2b(devid_len - 4, devid_ptr->length);
9162 
9163 	mtx_lock(&ctl_softc->ctl_lock);
9164 
9165 	fe = ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)];
9166 
9167 	/*
9168 	 * For Fibre channel,
9169 	 */
9170 	if (fe->port_type == CTL_PORT_FC)
9171 	{
9172 		desc->proto_codeset = (SCSI_PROTO_FC << 4) |
9173 				      SVPD_ID_CODESET_ASCII;
9174         	desc1->proto_codeset = (SCSI_PROTO_FC << 4) |
9175 		              SVPD_ID_CODESET_BINARY;
9176 	}
9177 	else
9178 	{
9179 		desc->proto_codeset = (SCSI_PROTO_SPI << 4) |
9180 				      SVPD_ID_CODESET_ASCII;
9181         	desc1->proto_codeset = (SCSI_PROTO_SPI << 4) |
9182 		              SVPD_ID_CODESET_BINARY;
9183 	}
9184 	desc2->proto_codeset = desc3->proto_codeset = desc1->proto_codeset;
9185 	mtx_unlock(&ctl_softc->ctl_lock);
9186 
9187 	/*
9188 	 * We're using a LUN association here.  i.e., this device ID is a
9189 	 * per-LUN identifier.
9190 	 */
9191 	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
9192 	desc->length = sizeof(*t10id) + CTL_DEVID_LEN;
9193 	strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
9194 
9195 	/*
9196 	 * desc1 is for the WWPN which is a port asscociation.
9197 	 */
9198 	desc1->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT | SVPD_ID_TYPE_NAA;
9199 	desc1->length = CTL_WWPN_LEN;
9200 	/* XXX Call Reggie's get_WWNN func here then add port # to the end */
9201 	/* For testing just create the WWPN */
9202 #if 0
9203 	ddb_GetWWNN((char *)desc1->identifier);
9204 
9205 	/* NOTE: if the port is 0 or 8 we don't want to subtract 1 */
9206 	/* This is so Copancontrol will return something sane */
9207 	if (ctsio->io_hdr.nexus.targ_port!=0 &&
9208 	    ctsio->io_hdr.nexus.targ_port!=8)
9209 		desc1->identifier[7] += ctsio->io_hdr.nexus.targ_port-1;
9210 	else
9211 		desc1->identifier[7] += ctsio->io_hdr.nexus.targ_port;
9212 #endif
9213 
9214 	be64enc(desc1->identifier, fe->wwpn);
9215 
9216 	/*
9217 	 * desc2 is for the Relative Target Port(type 4h) identifier
9218 	 */
9219 	desc2->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT
9220 	                 | SVPD_ID_TYPE_RELTARG;
9221 	desc2->length = 4;
9222 //#if 0
9223 	/* NOTE: if the port is 0 or 8 we don't want to subtract 1 */
9224 	/* This is so Copancontrol will return something sane */
9225 	if (ctsio->io_hdr.nexus.targ_port!=0 &&
9226 	    ctsio->io_hdr.nexus.targ_port!=8)
9227 		desc2->identifier[3] = ctsio->io_hdr.nexus.targ_port - 1;
9228 	else
9229 	        desc2->identifier[3] = ctsio->io_hdr.nexus.targ_port;
9230 //#endif
9231 
9232 	/*
9233 	 * desc3 is for the Target Port Group(type 5h) identifier
9234 	 */
9235 	desc3->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT
9236 	                 | SVPD_ID_TYPE_TPORTGRP;
9237 	desc3->length = 4;
9238 	if (ctsio->io_hdr.nexus.targ_port < CTL_MAX_PORTS || ctl_is_single)
9239 		desc3->identifier[3] = 1;
9240 	else
9241 		desc3->identifier[3] = 2;
9242 
9243 #ifdef CTL_USE_BACKEND_SN
9244 	/*
9245 	 * If we've actually got a backend, copy the device id from the
9246 	 * per-LUN data.  Otherwise, set it to all spaces.
9247 	 */
9248 	if (lun != NULL) {
9249 		/*
9250 		 * Copy the backend's LUN ID.
9251 		 */
9252 		strncpy((char *)t10id->vendor_spec_id,
9253 			(char *)lun->be_lun->device_id, CTL_DEVID_LEN);
9254 	} else {
9255 		/*
9256 		 * No backend, set this to spaces.
9257 		 */
9258 		memset(t10id->vendor_spec_id, 0x20, CTL_DEVID_LEN);
9259 	}
9260 #else
9261 	snprintf(tmpstr, sizeof(tmpstr), "MYDEVICEIDIS%4d",
9262 		 (lun != NULL) ?  (int)lun->lun : 0);
9263 	strncpy(t10id->vendor_spec_id, tmpstr, ctl_min(CTL_DEVID_LEN,
9264 		sizeof(tmpstr)));
9265 #endif
9266 
9267 	ctsio->scsi_status = SCSI_STATUS_OK;
9268 
9269 	ctsio->be_move_done = ctl_config_move_done;
9270 	ctl_datamove((union ctl_io *)ctsio);
9271 
9272 	return (CTL_RETVAL_COMPLETE);
9273 }
9274 
9275 static int
9276 ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
9277 {
9278 	struct scsi_inquiry *cdb;
9279 	int alloc_len, retval;
9280 
9281 	cdb = (struct scsi_inquiry *)ctsio->cdb;
9282 
9283 	retval = CTL_RETVAL_COMPLETE;
9284 
9285 	alloc_len = scsi_2btoul(cdb->length);
9286 
9287 	switch (cdb->page_code) {
9288 	case SVPD_SUPPORTED_PAGES:
9289 		retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
9290 		break;
9291 	case SVPD_UNIT_SERIAL_NUMBER:
9292 		retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
9293 		break;
9294 	case SVPD_DEVICE_ID:
9295 		retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
9296 		break;
9297 	default:
9298 		ctl_set_invalid_field(ctsio,
9299 				      /*sks_valid*/ 1,
9300 				      /*command*/ 1,
9301 				      /*field*/ 2,
9302 				      /*bit_valid*/ 0,
9303 				      /*bit*/ 0);
9304 		ctl_done((union ctl_io *)ctsio);
9305 		retval = CTL_RETVAL_COMPLETE;
9306 		break;
9307 	}
9308 
9309 	return (retval);
9310 }
9311 
9312 static int
9313 ctl_inquiry_std(struct ctl_scsiio *ctsio)
9314 {
9315 	struct scsi_inquiry_data *inq_ptr;
9316 	struct scsi_inquiry *cdb;
9317 	struct ctl_softc *ctl_softc;
9318 	struct ctl_lun *lun;
9319 	uint32_t alloc_len;
9320 	int is_fc;
9321 
9322 	ctl_softc = control_softc;
9323 
9324 	/*
9325 	 * Figure out whether we're talking to a Fibre Channel port or not.
9326 	 * We treat the ioctl front end, and any SCSI adapters, as packetized
9327 	 * SCSI front ends.
9328 	 */
9329 	mtx_lock(&ctl_softc->ctl_lock);
9330 	if (ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)]->port_type !=
9331 	    CTL_PORT_FC)
9332 		is_fc = 0;
9333 	else
9334 		is_fc = 1;
9335 	mtx_unlock(&ctl_softc->ctl_lock);
9336 
9337 	lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9338 	cdb = (struct scsi_inquiry *)ctsio->cdb;
9339 	alloc_len = scsi_2btoul(cdb->length);
9340 
9341 	/*
9342 	 * We malloc the full inquiry data size here and fill it
9343 	 * in.  If the user only asks for less, we'll give him
9344 	 * that much.
9345 	 */
9346 	/* XXX KDM what malloc flags should we use here?? */
9347 	ctsio->kern_data_ptr = malloc(sizeof(*inq_ptr), M_CTL, M_WAITOK);
9348 	if (ctsio->kern_data_ptr == NULL) {
9349 		ctsio->io_hdr.status = CTL_SCSI_ERROR;
9350 		ctsio->scsi_status = SCSI_STATUS_BUSY;
9351 		ctl_done((union ctl_io *)ctsio);
9352 		return (CTL_RETVAL_COMPLETE);
9353 	}
9354 	inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
9355 	ctsio->kern_sg_entries = 0;
9356 	ctsio->kern_data_resid = 0;
9357 	ctsio->kern_rel_offset = 0;
9358 
9359 	if (sizeof(*inq_ptr) < alloc_len) {
9360 		ctsio->residual = alloc_len - sizeof(*inq_ptr);
9361 		ctsio->kern_data_len = sizeof(*inq_ptr);
9362 		ctsio->kern_total_len = sizeof(*inq_ptr);
9363 	} else {
9364 		ctsio->residual = 0;
9365 		ctsio->kern_data_len = alloc_len;
9366 		ctsio->kern_total_len = alloc_len;
9367 	}
9368 
9369 	memset(inq_ptr, 0, sizeof(*inq_ptr));
9370 
9371 	/*
9372 	 * The control device is always connected.  The disk device, on the
9373 	 * other hand, may not be online all the time.  If we don't have a
9374 	 * LUN mapping, we'll just say it's offline.
9375 	 */
9376 	if (lun != NULL)
9377 		inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9378 				  lun->be_lun->lun_type;
9379 	else
9380 		inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9381 
9382 	/* RMB in byte 2 is 0 */
9383 	inq_ptr->version = SCSI_REV_SPC3;
9384 
9385 	/*
9386 	 * According to SAM-3, even if a device only supports a single
9387 	 * level of LUN addressing, it should still set the HISUP bit:
9388 	 *
9389 	 * 4.9.1 Logical unit numbers overview
9390 	 *
9391 	 * All logical unit number formats described in this standard are
9392 	 * hierarchical in structure even when only a single level in that
9393 	 * hierarchy is used. The HISUP bit shall be set to one in the
9394 	 * standard INQUIRY data (see SPC-2) when any logical unit number
9395 	 * format described in this standard is used.  Non-hierarchical
9396 	 * formats are outside the scope of this standard.
9397 	 *
9398 	 * Therefore we set the HiSup bit here.
9399 	 *
9400 	 * The reponse format is 2, per SPC-3.
9401 	 */
9402 	inq_ptr->response_format = SID_HiSup | 2;
9403 
9404 	inq_ptr->additional_length = sizeof(*inq_ptr) - 4;
9405 	CTL_DEBUG_PRINT(("additional_length = %d\n",
9406 			 inq_ptr->additional_length));
9407 
9408 	inq_ptr->spc3_flags = SPC3_SID_TPGS_IMPLICIT;
9409 	/* 16 bit addressing */
9410 	if (is_fc == 0)
9411 		inq_ptr->spc2_flags = SPC2_SID_ADDR16;
9412 	/* XXX set the SID_MultiP bit here if we're actually going to
9413 	   respond on multiple ports */
9414 	inq_ptr->spc2_flags |= SPC2_SID_MultiP;
9415 
9416 	/* 16 bit data bus, synchronous transfers */
9417 	/* XXX these flags don't apply for FC */
9418 	if (is_fc == 0)
9419 		inq_ptr->flags = SID_WBus16 | SID_Sync;
9420 	/*
9421 	 * XXX KDM do we want to support tagged queueing on the control
9422 	 * device at all?
9423 	 */
9424 	if ((lun == NULL)
9425 	 || (lun->be_lun->lun_type != T_PROCESSOR))
9426 		inq_ptr->flags |= SID_CmdQue;
9427 	/*
9428 	 * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
9429 	 * We have 8 bytes for the vendor name, and 16 bytes for the device
9430 	 * name and 4 bytes for the revision.
9431 	 */
9432 	strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
9433 	if (lun == NULL) {
9434 		strcpy(inq_ptr->product, CTL_DIRECT_PRODUCT);
9435 	} else {
9436 		switch (lun->be_lun->lun_type) {
9437 		case T_DIRECT:
9438 			strcpy(inq_ptr->product, CTL_DIRECT_PRODUCT);
9439 			break;
9440 		case T_PROCESSOR:
9441 			strcpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT);
9442 			break;
9443 		default:
9444 			strcpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT);
9445 			break;
9446 		}
9447 	}
9448 
9449 	/*
9450 	 * XXX make this a macro somewhere so it automatically gets
9451 	 * incremented when we make changes.
9452 	 */
9453 	strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
9454 
9455 	/*
9456 	 * For parallel SCSI, we support double transition and single
9457 	 * transition clocking.  We also support QAS (Quick Arbitration
9458 	 * and Selection) and Information Unit transfers on both the
9459 	 * control and array devices.
9460 	 */
9461 	if (is_fc == 0)
9462 		inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
9463 				    SID_SPI_IUS;
9464 
9465 	/* SAM-3 */
9466 	scsi_ulto2b(0x0060, inq_ptr->version1);
9467 	/* SPC-3 (no version claimed) XXX should we claim a version? */
9468 	scsi_ulto2b(0x0300, inq_ptr->version2);
9469 	if (is_fc) {
9470 		/* FCP-2 ANSI INCITS.350:2003 */
9471 		scsi_ulto2b(0x0917, inq_ptr->version3);
9472 	} else {
9473 		/* SPI-4 ANSI INCITS.362:200x */
9474 		scsi_ulto2b(0x0B56, inq_ptr->version3);
9475 	}
9476 
9477 	if (lun == NULL) {
9478 		/* SBC-2 (no version claimed) XXX should we claim a version? */
9479 		scsi_ulto2b(0x0320, inq_ptr->version4);
9480 	} else {
9481 		switch (lun->be_lun->lun_type) {
9482 		case T_DIRECT:
9483 			/*
9484 			 * SBC-2 (no version claimed) XXX should we claim a
9485 			 * version?
9486 			 */
9487 			scsi_ulto2b(0x0320, inq_ptr->version4);
9488 			break;
9489 		case T_PROCESSOR:
9490 		default:
9491 			break;
9492 		}
9493 	}
9494 	sprintf((char *)inq_ptr->vendor_specific1, "Copyright (C) 2004, COPAN "
9495 		"Systems, Inc.  All Rights Reserved.");
9496 
9497 	ctsio->scsi_status = SCSI_STATUS_OK;
9498 	if (ctsio->kern_data_len > 0) {
9499 		ctsio->be_move_done = ctl_config_move_done;
9500 		ctl_datamove((union ctl_io *)ctsio);
9501 	} else {
9502 		ctsio->io_hdr.status = CTL_SUCCESS;
9503 		ctl_done((union ctl_io *)ctsio);
9504 	}
9505 
9506 	return (CTL_RETVAL_COMPLETE);
9507 }
9508 
9509 int
9510 ctl_inquiry(struct ctl_scsiio *ctsio)
9511 {
9512 	struct scsi_inquiry *cdb;
9513 	int retval;
9514 
9515 	cdb = (struct scsi_inquiry *)ctsio->cdb;
9516 
9517 	retval = 0;
9518 
9519 	CTL_DEBUG_PRINT(("ctl_inquiry\n"));
9520 
9521 	/*
9522 	 * Right now, we don't support the CmdDt inquiry information.
9523 	 * This would be nice to support in the future.  When we do
9524 	 * support it, we should change this test so that it checks to make
9525 	 * sure SI_EVPD and SI_CMDDT aren't both set at the same time.
9526 	 */
9527 #ifdef notyet
9528 	if (((cdb->byte2 & SI_EVPD)
9529 	 && (cdb->byte2 & SI_CMDDT)))
9530 #endif
9531 	if (cdb->byte2 & SI_CMDDT) {
9532 		/*
9533 		 * Point to the SI_CMDDT bit.  We might change this
9534 		 * when we support SI_CMDDT, but since both bits would be
9535 		 * "wrong", this should probably just stay as-is then.
9536 		 */
9537 		ctl_set_invalid_field(ctsio,
9538 				      /*sks_valid*/ 1,
9539 				      /*command*/ 1,
9540 				      /*field*/ 1,
9541 				      /*bit_valid*/ 1,
9542 				      /*bit*/ 1);
9543 		ctl_done((union ctl_io *)ctsio);
9544 		return (CTL_RETVAL_COMPLETE);
9545 	}
9546 	if (cdb->byte2 & SI_EVPD)
9547 		retval = ctl_inquiry_evpd(ctsio);
9548 #ifdef notyet
9549 	else if (cdb->byte2 & SI_CMDDT)
9550 		retval = ctl_inquiry_cmddt(ctsio);
9551 #endif
9552 	else
9553 		retval = ctl_inquiry_std(ctsio);
9554 
9555 	return (retval);
9556 }
9557 
9558 /*
9559  * For known CDB types, parse the LBA and length.
9560  */
9561 static int
9562 ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint32_t *len)
9563 {
9564 	if (io->io_hdr.io_type != CTL_IO_SCSI)
9565 		return (1);
9566 
9567 	switch (io->scsiio.cdb[0]) {
9568 	case READ_6:
9569 	case WRITE_6: {
9570 		struct scsi_rw_6 *cdb;
9571 
9572 		cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
9573 
9574 		*lba = scsi_3btoul(cdb->addr);
9575 		/* only 5 bits are valid in the most significant address byte */
9576 		*lba &= 0x1fffff;
9577 		*len = cdb->length;
9578 		break;
9579 	}
9580 	case READ_10:
9581 	case WRITE_10: {
9582 		struct scsi_rw_10 *cdb;
9583 
9584 		cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
9585 
9586 		*lba = scsi_4btoul(cdb->addr);
9587 		*len = scsi_2btoul(cdb->length);
9588 		break;
9589 	}
9590 	case WRITE_VERIFY_10: {
9591 		struct scsi_write_verify_10 *cdb;
9592 
9593 		cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
9594 
9595 		*lba = scsi_4btoul(cdb->addr);
9596 		*len = scsi_2btoul(cdb->length);
9597 		break;
9598 	}
9599 	case READ_12:
9600 	case WRITE_12: {
9601 		struct scsi_rw_12 *cdb;
9602 
9603 		cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
9604 
9605 		*lba = scsi_4btoul(cdb->addr);
9606 		*len = scsi_4btoul(cdb->length);
9607 		break;
9608 	}
9609 	case WRITE_VERIFY_12: {
9610 		struct scsi_write_verify_12 *cdb;
9611 
9612 		cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
9613 
9614 		*lba = scsi_4btoul(cdb->addr);
9615 		*len = scsi_4btoul(cdb->length);
9616 		break;
9617 	}
9618 	case READ_16:
9619 	case WRITE_16: {
9620 		struct scsi_rw_16 *cdb;
9621 
9622 		cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
9623 
9624 		*lba = scsi_8btou64(cdb->addr);
9625 		*len = scsi_4btoul(cdb->length);
9626 		break;
9627 	}
9628 	case WRITE_VERIFY_16: {
9629 		struct scsi_write_verify_16 *cdb;
9630 
9631 		cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
9632 
9633 
9634 		*lba = scsi_8btou64(cdb->addr);
9635 		*len = scsi_4btoul(cdb->length);
9636 		break;
9637 	}
9638 	default:
9639 		return (1);
9640 		break; /* NOTREACHED */
9641 	}
9642 
9643 	return (0);
9644 }
9645 
9646 static ctl_action
9647 ctl_extent_check_lba(uint64_t lba1, uint32_t len1, uint64_t lba2, uint32_t len2)
9648 {
9649 	uint64_t endlba1, endlba2;
9650 
9651 	endlba1 = lba1 + len1 - 1;
9652 	endlba2 = lba2 + len2 - 1;
9653 
9654 	if ((endlba1 < lba2)
9655 	 || (endlba2 < lba1))
9656 		return (CTL_ACTION_PASS);
9657 	else
9658 		return (CTL_ACTION_BLOCK);
9659 }
9660 
9661 static ctl_action
9662 ctl_extent_check(union ctl_io *io1, union ctl_io *io2)
9663 {
9664 	uint64_t lba1, lba2;
9665 	uint32_t len1, len2;
9666 	int retval;
9667 
9668 	retval = ctl_get_lba_len(io1, &lba1, &len1);
9669 	if (retval != 0)
9670 		return (CTL_ACTION_ERROR);
9671 
9672 	retval = ctl_get_lba_len(io2, &lba2, &len2);
9673 	if (retval != 0)
9674 		return (CTL_ACTION_ERROR);
9675 
9676 	return (ctl_extent_check_lba(lba1, len1, lba2, len2));
9677 }
9678 
9679 static ctl_action
9680 ctl_check_for_blockage(union ctl_io *pending_io, union ctl_io *ooa_io)
9681 {
9682 	struct ctl_cmd_entry *pending_entry, *ooa_entry;
9683 	ctl_serialize_action *serialize_row;
9684 
9685 	/*
9686 	 * The initiator attempted multiple untagged commands at the same
9687 	 * time.  Can't do that.
9688 	 */
9689 	if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
9690 	 && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
9691 	 && ((pending_io->io_hdr.nexus.targ_port ==
9692 	      ooa_io->io_hdr.nexus.targ_port)
9693 	  && (pending_io->io_hdr.nexus.initid.id ==
9694 	      ooa_io->io_hdr.nexus.initid.id))
9695 	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
9696 		return (CTL_ACTION_OVERLAP);
9697 
9698 	/*
9699 	 * The initiator attempted to send multiple tagged commands with
9700 	 * the same ID.  (It's fine if different initiators have the same
9701 	 * tag ID.)
9702 	 *
9703 	 * Even if all of those conditions are true, we don't kill the I/O
9704 	 * if the command ahead of us has been aborted.  We won't end up
9705 	 * sending it to the FETD, and it's perfectly legal to resend a
9706 	 * command with the same tag number as long as the previous
9707 	 * instance of this tag number has been aborted somehow.
9708 	 */
9709 	if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
9710 	 && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
9711 	 && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
9712 	 && ((pending_io->io_hdr.nexus.targ_port ==
9713 	      ooa_io->io_hdr.nexus.targ_port)
9714 	  && (pending_io->io_hdr.nexus.initid.id ==
9715 	      ooa_io->io_hdr.nexus.initid.id))
9716 	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
9717 		return (CTL_ACTION_OVERLAP_TAG);
9718 
9719 	/*
9720 	 * If we get a head of queue tag, SAM-3 says that we should
9721 	 * immediately execute it.
9722 	 *
9723 	 * What happens if this command would normally block for some other
9724 	 * reason?  e.g. a request sense with a head of queue tag
9725 	 * immediately after a write.  Normally that would block, but this
9726 	 * will result in its getting executed immediately...
9727 	 *
9728 	 * We currently return "pass" instead of "skip", so we'll end up
9729 	 * going through the rest of the queue to check for overlapped tags.
9730 	 *
9731 	 * XXX KDM check for other types of blockage first??
9732 	 */
9733 	if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
9734 		return (CTL_ACTION_PASS);
9735 
9736 	/*
9737 	 * Ordered tags have to block until all items ahead of them
9738 	 * have completed.  If we get called with an ordered tag, we always
9739 	 * block, if something else is ahead of us in the queue.
9740 	 */
9741 	if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
9742 		return (CTL_ACTION_BLOCK);
9743 
9744 	/*
9745 	 * Simple tags get blocked until all head of queue and ordered tags
9746 	 * ahead of them have completed.  I'm lumping untagged commands in
9747 	 * with simple tags here.  XXX KDM is that the right thing to do?
9748 	 */
9749 	if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
9750 	  || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
9751 	 && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
9752 	  || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
9753 		return (CTL_ACTION_BLOCK);
9754 
9755 	pending_entry = &ctl_cmd_table[pending_io->scsiio.cdb[0]];
9756 	ooa_entry = &ctl_cmd_table[ooa_io->scsiio.cdb[0]];
9757 
9758 	serialize_row = ctl_serialize_table[ooa_entry->seridx];
9759 
9760 	switch (serialize_row[pending_entry->seridx]) {
9761 	case CTL_SER_BLOCK:
9762 		return (CTL_ACTION_BLOCK);
9763 		break; /* NOTREACHED */
9764 	case CTL_SER_EXTENT:
9765 		return (ctl_extent_check(pending_io, ooa_io));
9766 		break; /* NOTREACHED */
9767 	case CTL_SER_PASS:
9768 		return (CTL_ACTION_PASS);
9769 		break; /* NOTREACHED */
9770 	case CTL_SER_SKIP:
9771 		return (CTL_ACTION_SKIP);
9772 		break;
9773 	default:
9774 		panic("invalid serialization value %d",
9775 		      serialize_row[pending_entry->seridx]);
9776 		break; /* NOTREACHED */
9777 	}
9778 
9779 	return (CTL_ACTION_ERROR);
9780 }
9781 
9782 /*
9783  * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
9784  * Assumptions:
9785  * - caller holds ctl_lock
9786  * - pending_io is generally either incoming, or on the blocked queue
9787  * - starting I/O is the I/O we want to start the check with.
9788  */
9789 static ctl_action
9790 ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
9791 	      union ctl_io *starting_io)
9792 {
9793 	union ctl_io *ooa_io;
9794 	ctl_action action;
9795 
9796 	/*
9797 	 * Run back along the OOA queue, starting with the current
9798 	 * blocked I/O and going through every I/O before it on the
9799 	 * queue.  If starting_io is NULL, we'll just end up returning
9800 	 * CTL_ACTION_PASS.
9801 	 */
9802 	for (ooa_io = starting_io; ooa_io != NULL;
9803 	     ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
9804 	     ooa_links)){
9805 
9806 		/*
9807 		 * This routine just checks to see whether
9808 		 * cur_blocked is blocked by ooa_io, which is ahead
9809 		 * of it in the queue.  It doesn't queue/dequeue
9810 		 * cur_blocked.
9811 		 */
9812 		action = ctl_check_for_blockage(pending_io, ooa_io);
9813 		switch (action) {
9814 		case CTL_ACTION_BLOCK:
9815 		case CTL_ACTION_OVERLAP:
9816 		case CTL_ACTION_OVERLAP_TAG:
9817 		case CTL_ACTION_SKIP:
9818 		case CTL_ACTION_ERROR:
9819 			return (action);
9820 			break; /* NOTREACHED */
9821 		case CTL_ACTION_PASS:
9822 			break;
9823 		default:
9824 			panic("invalid action %d", action);
9825 			break;  /* NOTREACHED */
9826 		}
9827 	}
9828 
9829 	return (CTL_ACTION_PASS);
9830 }
9831 
9832 /*
9833  * Assumptions:
9834  * - An I/O has just completed, and has been removed from the per-LUN OOA
9835  *   queue, so some items on the blocked queue may now be unblocked.
9836  * - The caller holds ctl_softc->ctl_lock
9837  */
9838 static int
9839 ctl_check_blocked(struct ctl_lun *lun)
9840 {
9841 	union ctl_io *cur_blocked, *next_blocked;
9842 
9843 	/*
9844 	 * Run forward from the head of the blocked queue, checking each
9845 	 * entry against the I/Os prior to it on the OOA queue to see if
9846 	 * there is still any blockage.
9847 	 *
9848 	 * We cannot use the TAILQ_FOREACH() macro, because it can't deal
9849 	 * with our removing a variable on it while it is traversing the
9850 	 * list.
9851 	 */
9852 	for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
9853 	     cur_blocked != NULL; cur_blocked = next_blocked) {
9854 		union ctl_io *prev_ooa;
9855 		ctl_action action;
9856 
9857 		next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
9858 							  blocked_links);
9859 
9860 		prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
9861 						      ctl_ooaq, ooa_links);
9862 
9863 		/*
9864 		 * If cur_blocked happens to be the first item in the OOA
9865 		 * queue now, prev_ooa will be NULL, and the action
9866 		 * returned will just be CTL_ACTION_PASS.
9867 		 */
9868 		action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
9869 
9870 		switch (action) {
9871 		case CTL_ACTION_BLOCK:
9872 			/* Nothing to do here, still blocked */
9873 			break;
9874 		case CTL_ACTION_OVERLAP:
9875 		case CTL_ACTION_OVERLAP_TAG:
9876 			/*
9877 			 * This shouldn't happen!  In theory we've already
9878 			 * checked this command for overlap...
9879 			 */
9880 			break;
9881 		case CTL_ACTION_PASS:
9882 		case CTL_ACTION_SKIP: {
9883 			struct ctl_softc *softc;
9884 			struct ctl_cmd_entry *entry;
9885 			uint32_t initidx;
9886 			uint8_t opcode;
9887 			int isc_retval;
9888 
9889 			/*
9890 			 * The skip case shouldn't happen, this transaction
9891 			 * should have never made it onto the blocked queue.
9892 			 */
9893 			/*
9894 			 * This I/O is no longer blocked, we can remove it
9895 			 * from the blocked queue.  Since this is a TAILQ
9896 			 * (doubly linked list), we can do O(1) removals
9897 			 * from any place on the list.
9898 			 */
9899 			TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
9900 				     blocked_links);
9901 			cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
9902 
9903 			if (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC){
9904 				/*
9905 				 * Need to send IO back to original side to
9906 				 * run
9907 				 */
9908 				union ctl_ha_msg msg_info;
9909 
9910 				msg_info.hdr.original_sc =
9911 					cur_blocked->io_hdr.original_sc;
9912 				msg_info.hdr.serializing_sc = cur_blocked;
9913 				msg_info.hdr.msg_type = CTL_MSG_R2R;
9914 				if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
9915 				     &msg_info, sizeof(msg_info), 0)) >
9916 				     CTL_HA_STATUS_SUCCESS) {
9917 					printf("CTL:Check Blocked error from "
9918 					       "ctl_ha_msg_send %d\n",
9919 					       isc_retval);
9920 				}
9921 				break;
9922 			}
9923 			opcode = cur_blocked->scsiio.cdb[0];
9924 			entry = &ctl_cmd_table[opcode];
9925 			softc = control_softc;
9926 
9927 			initidx = ctl_get_initindex(&cur_blocked->io_hdr.nexus);
9928 
9929 			/*
9930 			 * Check this I/O for LUN state changes that may
9931 			 * have happened while this command was blocked.
9932 			 * The LUN state may have been changed by a command
9933 			 * ahead of us in the queue, so we need to re-check
9934 			 * for any states that can be caused by SCSI
9935 			 * commands.
9936 			 */
9937 			if (ctl_scsiio_lun_check(softc, lun, entry,
9938 						 &cur_blocked->scsiio) == 0) {
9939 				cur_blocked->io_hdr.flags |=
9940 				                      CTL_FLAG_IS_WAS_ON_RTR;
9941 				STAILQ_INSERT_TAIL(&lun->ctl_softc->rtr_queue,
9942 						   &cur_blocked->io_hdr, links);
9943 				/*
9944 				 * In the non CTL_DONE_THREAD case, we need
9945 				 * to wake up the work thread here.  When
9946 				 * we're processing completed requests from
9947 				 * the work thread context, we'll pop back
9948 				 * around and end up pulling things off the
9949 				 * RtR queue.  When we aren't processing
9950 				 * things from the work thread context,
9951 				 * though, we won't ever check the RtR queue.
9952 				 * So we need to wake up the thread to clear
9953 				 * things off the queue.  Otherwise this
9954 				 * transaction will just sit on the RtR queue
9955 				 * until a new I/O comes in.  (Which may or
9956 				 * may not happen...)
9957 				 */
9958 #ifndef CTL_DONE_THREAD
9959 				ctl_wakeup_thread();
9960 #endif
9961 			} else
9962 				ctl_done_lock(cur_blocked, /*have_lock*/ 1);
9963 			break;
9964 		}
9965 		default:
9966 			/*
9967 			 * This probably shouldn't happen -- we shouldn't
9968 			 * get CTL_ACTION_ERROR, or anything else.
9969 			 */
9970 			break;
9971 		}
9972 	}
9973 
9974 	return (CTL_RETVAL_COMPLETE);
9975 }
9976 
9977 /*
9978  * This routine (with one exception) checks LUN flags that can be set by
9979  * commands ahead of us in the OOA queue.  These flags have to be checked
9980  * when a command initially comes in, and when we pull a command off the
9981  * blocked queue and are preparing to execute it.  The reason we have to
9982  * check these flags for commands on the blocked queue is that the LUN
9983  * state may have been changed by a command ahead of us while we're on the
9984  * blocked queue.
9985  *
9986  * Ordering is somewhat important with these checks, so please pay
9987  * careful attention to the placement of any new checks.
9988  */
9989 static int
9990 ctl_scsiio_lun_check(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
9991 		     struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
9992 {
9993 	int retval;
9994 
9995 	retval = 0;
9996 
9997 	/*
9998 	 * If this shelf is a secondary shelf controller, we have to reject
9999 	 * any media access commands.
10000 	 */
10001 #if 0
10002 	/* No longer needed for HA */
10003 	if (((ctl_softc->flags & CTL_FLAG_MASTER_SHELF) == 0)
10004 	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_SECONDARY) == 0)) {
10005 		ctl_set_lun_standby(ctsio);
10006 		retval = 1;
10007 		goto bailout;
10008 	}
10009 #endif
10010 
10011 	/*
10012 	 * Check for a reservation conflict.  If this command isn't allowed
10013 	 * even on reserved LUNs, and if this initiator isn't the one who
10014 	 * reserved us, reject the command with a reservation conflict.
10015 	 */
10016 	if ((lun->flags & CTL_LUN_RESERVED)
10017 	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
10018 		if ((ctsio->io_hdr.nexus.initid.id != lun->rsv_nexus.initid.id)
10019 		 || (ctsio->io_hdr.nexus.targ_port != lun->rsv_nexus.targ_port)
10020 		 || (ctsio->io_hdr.nexus.targ_target.id !=
10021 		     lun->rsv_nexus.targ_target.id)) {
10022 			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
10023 			ctsio->io_hdr.status = CTL_SCSI_ERROR;
10024 			retval = 1;
10025 			goto bailout;
10026 		}
10027 	}
10028 
10029 	if ( (lun->flags & CTL_LUN_PR_RESERVED)
10030 	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV) == 0)) {
10031 		uint32_t residx;
10032 
10033 		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
10034 		/*
10035 		 * if we aren't registered or it's a res holder type
10036 		 * reservation and this isn't the res holder then set a
10037 		 * conflict.
10038 		 * NOTE: Commands which might be allowed on write exclusive
10039 		 * type reservations are checked in the particular command
10040 		 * for a conflict. Read and SSU are the only ones.
10041 		 */
10042 		if (!lun->per_res[residx].registered
10043 		 || (residx != lun->pr_res_idx && lun->res_type < 4)) {
10044 			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
10045 			ctsio->io_hdr.status = CTL_SCSI_ERROR;
10046 			retval = 1;
10047 			goto bailout;
10048 		}
10049 
10050 	}
10051 
10052 	if ((lun->flags & CTL_LUN_OFFLINE)
10053 	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_OFFLINE) == 0)) {
10054 		ctl_set_lun_not_ready(ctsio);
10055 		retval = 1;
10056 		goto bailout;
10057 	}
10058 
10059 	/*
10060 	 * If the LUN is stopped, see if this particular command is allowed
10061 	 * for a stopped lun.  Otherwise, reject it with 0x04,0x02.
10062 	 */
10063 	if ((lun->flags & CTL_LUN_STOPPED)
10064 	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_STOPPED) == 0)) {
10065 		/* "Logical unit not ready, initializing cmd. required" */
10066 		ctl_set_lun_stopped(ctsio);
10067 		retval = 1;
10068 		goto bailout;
10069 	}
10070 
10071 	if ((lun->flags & CTL_LUN_INOPERABLE)
10072 	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_INOPERABLE) == 0)) {
10073 		/* "Medium format corrupted" */
10074 		ctl_set_medium_format_corrupted(ctsio);
10075 		retval = 1;
10076 		goto bailout;
10077 	}
10078 
10079 bailout:
10080 	return (retval);
10081 
10082 }
10083 
10084 static void
10085 ctl_failover_io(union ctl_io *io, int have_lock)
10086 {
10087 	ctl_set_busy(&io->scsiio);
10088 	ctl_done_lock(io, have_lock);
10089 }
10090 
10091 static void
10092 ctl_failover(void)
10093 {
10094 	struct ctl_lun *lun;
10095 	struct ctl_softc *ctl_softc;
10096 	union ctl_io *next_io, *pending_io;
10097 	union ctl_io *io;
10098 	int lun_idx;
10099 	int i;
10100 
10101 	ctl_softc = control_softc;
10102 
10103 	mtx_lock(&ctl_softc->ctl_lock);
10104 	/*
10105 	 * Remove any cmds from the other SC from the rtr queue.  These
10106 	 * will obviously only be for LUNs for which we're the primary.
10107 	 * We can't send status or get/send data for these commands.
10108 	 * Since they haven't been executed yet, we can just remove them.
10109 	 * We'll either abort them or delete them below, depending on
10110 	 * which HA mode we're in.
10111 	 */
10112 	for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->rtr_queue);
10113 	     io != NULL; io = next_io) {
10114 		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
10115 		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
10116 			STAILQ_REMOVE(&ctl_softc->rtr_queue, &io->io_hdr,
10117 				      ctl_io_hdr, links);
10118 	}
10119 
10120 	for (lun_idx=0; lun_idx < ctl_softc->num_luns; lun_idx++) {
10121 		lun = ctl_softc->ctl_luns[lun_idx];
10122 		if (lun==NULL)
10123 			continue;
10124 
10125 		/*
10126 		 * Processor LUNs are primary on both sides.
10127 		 * XXX will this always be true?
10128 		 */
10129 		if (lun->be_lun->lun_type == T_PROCESSOR)
10130 			continue;
10131 
10132 		if ((lun->flags & CTL_LUN_PRIMARY_SC)
10133 		 && (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
10134 			printf("FAILOVER: primary lun %d\n", lun_idx);
10135 		        /*
10136 			 * Remove all commands from the other SC. First from the
10137 			 * blocked queue then from the ooa queue. Once we have
10138 			 * removed them. Call ctl_check_blocked to see if there
10139 			 * is anything that can run.
10140 			 */
10141 			for (io = (union ctl_io *)TAILQ_FIRST(
10142 			     &lun->blocked_queue); io != NULL; io = next_io) {
10143 
10144 		        	next_io = (union ctl_io *)TAILQ_NEXT(
10145 				    &io->io_hdr, blocked_links);
10146 
10147 				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
10148 					TAILQ_REMOVE(&lun->blocked_queue,
10149 						     &io->io_hdr,blocked_links);
10150 					io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
10151 					TAILQ_REMOVE(&lun->ooa_queue,
10152 						     &io->io_hdr, ooa_links);
10153 
10154 					ctl_free_io_internal(io, 1);
10155 				}
10156 			}
10157 
10158 			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
10159 	     		     io != NULL; io = next_io) {
10160 
10161 		        	next_io = (union ctl_io *)TAILQ_NEXT(
10162 				    &io->io_hdr, ooa_links);
10163 
10164 				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
10165 
10166 					TAILQ_REMOVE(&lun->ooa_queue,
10167 						&io->io_hdr,
10168 					     	ooa_links);
10169 
10170 					ctl_free_io_internal(io, 1);
10171 				}
10172 			}
10173 			ctl_check_blocked(lun);
10174 		} else if ((lun->flags & CTL_LUN_PRIMARY_SC)
10175 			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
10176 
10177 			printf("FAILOVER: primary lun %d\n", lun_idx);
10178 			/*
10179 			 * Abort all commands from the other SC.  We can't
10180 			 * send status back for them now.  These should get
10181 			 * cleaned up when they are completed or come out
10182 			 * for a datamove operation.
10183 			 */
10184 			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
10185 	     		     io != NULL; io = next_io) {
10186 		        	next_io = (union ctl_io *)TAILQ_NEXT(
10187 					&io->io_hdr, ooa_links);
10188 
10189 				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
10190 					io->io_hdr.flags |= CTL_FLAG_ABORT;
10191 			}
10192 		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
10193 			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
10194 
10195 			printf("FAILOVER: secondary lun %d\n", lun_idx);
10196 
10197 			lun->flags |= CTL_LUN_PRIMARY_SC;
10198 
10199 			/*
10200 			 * We send all I/O that was sent to this controller
10201 			 * and redirected to the other side back with
10202 			 * busy status, and have the initiator retry it.
10203 			 * Figuring out how much data has been transferred,
10204 			 * etc. and picking up where we left off would be
10205 			 * very tricky.
10206 			 *
10207 			 * XXX KDM need to remove I/O from the blocked
10208 			 * queue as well!
10209 			 */
10210 			for (pending_io = (union ctl_io *)TAILQ_FIRST(
10211 			     &lun->ooa_queue); pending_io != NULL;
10212 			     pending_io = next_io) {
10213 
10214 				next_io =  (union ctl_io *)TAILQ_NEXT(
10215 					&pending_io->io_hdr, ooa_links);
10216 
10217 				pending_io->io_hdr.flags &=
10218 					~CTL_FLAG_SENT_2OTHER_SC;
10219 
10220 				if (pending_io->io_hdr.flags &
10221 				    CTL_FLAG_IO_ACTIVE) {
10222 					pending_io->io_hdr.flags |=
10223 						CTL_FLAG_FAILOVER;
10224 				} else {
10225 					ctl_set_busy(&pending_io->scsiio);
10226 					ctl_done_lock(pending_io,
10227 						      /*have_lock*/1);
10228 				}
10229 			}
10230 
10231 			/*
10232 			 * Build Unit Attention
10233 			 */
10234 			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
10235 				lun->pending_sense[i].ua_pending |=
10236 				                     CTL_UA_ASYM_ACC_CHANGE;
10237 			}
10238 		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
10239 			&& (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
10240 			printf("FAILOVER: secondary lun %d\n", lun_idx);
10241 			/*
10242 			 * if the first io on the OOA is not on the RtR queue
10243 			 * add it.
10244 			 */
10245 			lun->flags |= CTL_LUN_PRIMARY_SC;
10246 
10247 			pending_io = (union ctl_io *)TAILQ_FIRST(
10248 			    &lun->ooa_queue);
10249 			if (pending_io==NULL) {
10250 				printf("Nothing on OOA queue\n");
10251 				continue;
10252 			}
10253 
10254 			pending_io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
10255 			if ((pending_io->io_hdr.flags &
10256 			     CTL_FLAG_IS_WAS_ON_RTR) == 0) {
10257 				pending_io->io_hdr.flags |=
10258 				    CTL_FLAG_IS_WAS_ON_RTR;
10259 				STAILQ_INSERT_TAIL(&ctl_softc->rtr_queue,
10260 						   &pending_io->io_hdr, links);
10261 			}
10262 #if 0
10263 			else
10264 			{
10265 				printf("Tag 0x%04x is running\n",
10266 				      pending_io->scsiio.tag_num);
10267 			}
10268 #endif
10269 
10270 			next_io = (union ctl_io *)TAILQ_NEXT(
10271 			    &pending_io->io_hdr, ooa_links);
10272 			for (pending_io=next_io; pending_io != NULL;
10273 			     pending_io = next_io) {
10274 				pending_io->io_hdr.flags &=
10275 				    ~CTL_FLAG_SENT_2OTHER_SC;
10276 				next_io = (union ctl_io *)TAILQ_NEXT(
10277 					&pending_io->io_hdr, ooa_links);
10278 				if (pending_io->io_hdr.flags &
10279 				    CTL_FLAG_IS_WAS_ON_RTR) {
10280 #if 0
10281 				        printf("Tag 0x%04x is running\n",
10282 				      		pending_io->scsiio.tag_num);
10283 #endif
10284 					continue;
10285 				}
10286 
10287 				switch (ctl_check_ooa(lun, pending_io,
10288 			            (union ctl_io *)TAILQ_PREV(
10289 				    &pending_io->io_hdr, ctl_ooaq,
10290 				    ooa_links))) {
10291 
10292 				case CTL_ACTION_BLOCK:
10293 					TAILQ_INSERT_TAIL(&lun->blocked_queue,
10294 							  &pending_io->io_hdr,
10295 							  blocked_links);
10296 					pending_io->io_hdr.flags |=
10297 					    CTL_FLAG_BLOCKED;
10298 					break;
10299 				case CTL_ACTION_PASS:
10300 				case CTL_ACTION_SKIP:
10301 					pending_io->io_hdr.flags |=
10302 					    CTL_FLAG_IS_WAS_ON_RTR;
10303 					STAILQ_INSERT_TAIL(
10304 					    &ctl_softc->rtr_queue,
10305 					    &pending_io->io_hdr, links);
10306 					break;
10307 				case CTL_ACTION_OVERLAP:
10308 					ctl_set_overlapped_cmd(
10309 					    (struct ctl_scsiio *)pending_io);
10310 					ctl_done_lock(pending_io,
10311 						      /*have_lock*/ 1);
10312 					break;
10313 				case CTL_ACTION_OVERLAP_TAG:
10314 					ctl_set_overlapped_tag(
10315 					    (struct ctl_scsiio *)pending_io,
10316 					    pending_io->scsiio.tag_num & 0xff);
10317 					ctl_done_lock(pending_io,
10318 						      /*have_lock*/ 1);
10319 					break;
10320 				case CTL_ACTION_ERROR:
10321 				default:
10322 					ctl_set_internal_failure(
10323 						(struct ctl_scsiio *)pending_io,
10324 						0,  // sks_valid
10325 						0); //retry count
10326 					ctl_done_lock(pending_io,
10327 						      /*have_lock*/ 1);
10328 					break;
10329 				}
10330 			}
10331 
10332 			/*
10333 			 * Build Unit Attention
10334 			 */
10335 			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
10336 				lun->pending_sense[i].ua_pending |=
10337 				                     CTL_UA_ASYM_ACC_CHANGE;
10338 			}
10339 		} else {
10340 			panic("Unhandled HA mode failover, LUN flags = %#x, "
10341 			      "ha_mode = #%x", lun->flags, ctl_softc->ha_mode);
10342 		}
10343 	}
10344 	ctl_pause_rtr = 0;
10345 	mtx_unlock(&ctl_softc->ctl_lock);
10346 }
10347 
10348 static int
10349 ctl_scsiio_precheck(struct ctl_softc *ctl_softc, struct ctl_scsiio *ctsio)
10350 {
10351 	struct ctl_lun *lun;
10352 	struct ctl_cmd_entry *entry;
10353 	uint8_t opcode;
10354 	uint32_t initidx;
10355 	int retval;
10356 
10357 	retval = 0;
10358 
10359 	lun = NULL;
10360 
10361 	opcode = ctsio->cdb[0];
10362 
10363 	mtx_lock(&ctl_softc->ctl_lock);
10364 
10365 	if ((ctsio->io_hdr.nexus.targ_lun < CTL_MAX_LUNS)
10366 	 && (ctl_softc->ctl_luns[ctsio->io_hdr.nexus.targ_lun] != NULL)) {
10367 		lun = ctl_softc->ctl_luns[ctsio->io_hdr.nexus.targ_lun];
10368 		/*
10369 		 * If the LUN is invalid, pretend that it doesn't exist.
10370 		 * It will go away as soon as all pending I/O has been
10371 		 * completed.
10372 		 */
10373 		if (lun->flags & CTL_LUN_DISABLED) {
10374 			lun = NULL;
10375 		} else {
10376 			ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
10377 			ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
10378 				lun->be_lun;
10379 			if (lun->be_lun->lun_type == T_PROCESSOR) {
10380 				ctsio->io_hdr.flags |= CTL_FLAG_CONTROL_DEV;
10381 			}
10382 		}
10383 	} else {
10384 		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
10385 		ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
10386 	}
10387 
10388 	entry = &ctl_cmd_table[opcode];
10389 
10390 	ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
10391 	ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
10392 
10393 	/*
10394 	 * Check to see whether we can send this command to LUNs that don't
10395 	 * exist.  This should pretty much only be the case for inquiry
10396 	 * and request sense.  Further checks, below, really require having
10397 	 * a LUN, so we can't really check the command anymore.  Just put
10398 	 * it on the rtr queue.
10399 	 */
10400 	if (lun == NULL) {
10401 		if (entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS)
10402 			goto queue_rtr;
10403 
10404 		ctl_set_unsupported_lun(ctsio);
10405 		mtx_unlock(&ctl_softc->ctl_lock);
10406 		ctl_done((union ctl_io *)ctsio);
10407 		goto bailout;
10408 	} else {
10409 		/*
10410 		 * Every I/O goes into the OOA queue for a particular LUN, and
10411 		 * stays there until completion.
10412 		 */
10413 		TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
10414 
10415 		/*
10416 		 * Make sure we support this particular command on this LUN.
10417 		 * e.g., we don't support writes to the control LUN.
10418 		 */
10419 		switch (lun->be_lun->lun_type) {
10420 		case T_PROCESSOR:
10421 		 	if (((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0)
10422 			 && ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS)
10423 			      == 0)) {
10424 				ctl_set_invalid_opcode(ctsio);
10425 				mtx_unlock(&ctl_softc->ctl_lock);
10426 				ctl_done((union ctl_io *)ctsio);
10427 				goto bailout;
10428 			}
10429 			break;
10430 		case T_DIRECT:
10431 			if (((entry->flags & CTL_CMD_FLAG_OK_ON_SLUN) == 0)
10432 			 && ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS)
10433 			      == 0)){
10434 				ctl_set_invalid_opcode(ctsio);
10435 				mtx_unlock(&ctl_softc->ctl_lock);
10436 				ctl_done((union ctl_io *)ctsio);
10437 				goto bailout;
10438 			}
10439 			break;
10440 		default:
10441 			printf("Unsupported CTL LUN type %d\n",
10442 			       lun->be_lun->lun_type);
10443 			panic("Unsupported CTL LUN type %d\n",
10444 			      lun->be_lun->lun_type);
10445 			break; /* NOTREACHED */
10446 		}
10447 	}
10448 
10449 	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
10450 
10451 	/*
10452 	 * If we've got a request sense, it'll clear the contingent
10453 	 * allegiance condition.  Otherwise, if we have a CA condition for
10454 	 * this initiator, clear it, because it sent down a command other
10455 	 * than request sense.
10456 	 */
10457 	if ((opcode != REQUEST_SENSE)
10458 	 && (ctl_is_set(lun->have_ca, initidx)))
10459 		ctl_clear_mask(lun->have_ca, initidx);
10460 
10461 	/*
10462 	 * If the command has this flag set, it handles its own unit
10463 	 * attention reporting, we shouldn't do anything.  Otherwise we
10464 	 * check for any pending unit attentions, and send them back to the
10465 	 * initiator.  We only do this when a command initially comes in,
10466 	 * not when we pull it off the blocked queue.
10467 	 *
10468 	 * According to SAM-3, section 5.3.2, the order that things get
10469 	 * presented back to the host is basically unit attentions caused
10470 	 * by some sort of reset event, busy status, reservation conflicts
10471 	 * or task set full, and finally any other status.
10472 	 *
10473 	 * One issue here is that some of the unit attentions we report
10474 	 * don't fall into the "reset" category (e.g. "reported luns data
10475 	 * has changed").  So reporting it here, before the reservation
10476 	 * check, may be technically wrong.  I guess the only thing to do
10477 	 * would be to check for and report the reset events here, and then
10478 	 * check for the other unit attention types after we check for a
10479 	 * reservation conflict.
10480 	 *
10481 	 * XXX KDM need to fix this
10482 	 */
10483 	if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
10484 		ctl_ua_type ua_type;
10485 
10486 		ua_type = lun->pending_sense[initidx].ua_pending;
10487 		if (ua_type != CTL_UA_NONE) {
10488 			scsi_sense_data_type sense_format;
10489 
10490 			if (lun != NULL)
10491 				sense_format = (lun->flags &
10492 				    CTL_LUN_SENSE_DESC) ? SSD_TYPE_DESC :
10493 				    SSD_TYPE_FIXED;
10494 			else
10495 				sense_format = SSD_TYPE_FIXED;
10496 
10497 			ua_type = ctl_build_ua(ua_type, &ctsio->sense_data,
10498 					       sense_format);
10499 			if (ua_type != CTL_UA_NONE) {
10500 				ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
10501 				ctsio->io_hdr.status = CTL_SCSI_ERROR |
10502 						       CTL_AUTOSENSE;
10503 				ctsio->sense_len = SSD_FULL_SIZE;
10504 				lun->pending_sense[initidx].ua_pending &=
10505 					~ua_type;
10506 				mtx_unlock(&ctl_softc->ctl_lock);
10507 				ctl_done((union ctl_io *)ctsio);
10508 				goto bailout;
10509 			}
10510 		}
10511 	}
10512 
10513 
10514 	if (ctl_scsiio_lun_check(ctl_softc, lun, entry, ctsio) != 0) {
10515 		mtx_unlock(&ctl_softc->ctl_lock);
10516 		ctl_done((union ctl_io *)ctsio);
10517 		goto bailout;
10518 	}
10519 
10520 	/*
10521 	 * XXX CHD this is where we want to send IO to other side if
10522 	 * this LUN is secondary on this SC. We will need to make a copy
10523 	 * of the IO and flag the IO on this side as SENT_2OTHER and the flag
10524 	 * the copy we send as FROM_OTHER.
10525 	 * We also need to stuff the address of the original IO so we can
10526 	 * find it easily. Something similar will need be done on the other
10527 	 * side so when we are done we can find the copy.
10528 	 */
10529 	if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
10530 		union ctl_ha_msg msg_info;
10531 		int isc_retval;
10532 
10533 		ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
10534 
10535 		msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
10536 		msg_info.hdr.original_sc = (union ctl_io *)ctsio;
10537 #if 0
10538 		printf("1. ctsio %p\n", ctsio);
10539 #endif
10540 		msg_info.hdr.serializing_sc = NULL;
10541 		msg_info.hdr.nexus = ctsio->io_hdr.nexus;
10542 		msg_info.scsi.tag_num = ctsio->tag_num;
10543 		msg_info.scsi.tag_type = ctsio->tag_type;
10544 		memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
10545 
10546 		ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
10547 
10548 		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
10549 		    (void *)&msg_info, sizeof(msg_info), 0)) >
10550 		    CTL_HA_STATUS_SUCCESS) {
10551 			printf("CTL:precheck, ctl_ha_msg_send returned %d\n",
10552 			       isc_retval);
10553 			printf("CTL:opcode is %x\n",opcode);
10554 		} else {
10555 #if 0
10556 			printf("CTL:Precheck sent msg, opcode is %x\n",opcode);
10557 #endif
10558 		}
10559 
10560 		/*
10561 		 * XXX KDM this I/O is off the incoming queue, but hasn't
10562 		 * been inserted on any other queue.  We may need to come
10563 		 * up with a holding queue while we wait for serialization
10564 		 * so that we have an idea of what we're waiting for from
10565 		 * the other side.
10566 		 */
10567 		goto bailout_unlock;
10568 	}
10569 
10570 	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
10571 			      (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
10572 			      ctl_ooaq, ooa_links))) {
10573 	case CTL_ACTION_BLOCK:
10574 		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
10575 		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
10576 				  blocked_links);
10577 		goto bailout_unlock;
10578 		break; /* NOTREACHED */
10579 	case CTL_ACTION_PASS:
10580 	case CTL_ACTION_SKIP:
10581 		goto queue_rtr;
10582 		break; /* NOTREACHED */
10583 	case CTL_ACTION_OVERLAP:
10584 		ctl_set_overlapped_cmd(ctsio);
10585 		mtx_unlock(&ctl_softc->ctl_lock);
10586 		ctl_done((union ctl_io *)ctsio);
10587 		goto bailout;
10588 		break; /* NOTREACHED */
10589 	case CTL_ACTION_OVERLAP_TAG:
10590 		ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
10591 		mtx_unlock(&ctl_softc->ctl_lock);
10592 		ctl_done((union ctl_io *)ctsio);
10593 		goto bailout;
10594 		break; /* NOTREACHED */
10595 	case CTL_ACTION_ERROR:
10596 	default:
10597 		ctl_set_internal_failure(ctsio,
10598 					 /*sks_valid*/ 0,
10599 					 /*retry_count*/ 0);
10600 		mtx_unlock(&ctl_softc->ctl_lock);
10601 		ctl_done((union ctl_io *)ctsio);
10602 		goto bailout;
10603 		break; /* NOTREACHED */
10604 	}
10605 
10606 	goto bailout_unlock;
10607 
10608 queue_rtr:
10609 	ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
10610 	STAILQ_INSERT_TAIL(&ctl_softc->rtr_queue, &ctsio->io_hdr, links);
10611 
10612 bailout_unlock:
10613 	mtx_unlock(&ctl_softc->ctl_lock);
10614 
10615 bailout:
10616 	return (retval);
10617 }
10618 
10619 static int
10620 ctl_scsiio(struct ctl_scsiio *ctsio)
10621 {
10622 	int retval;
10623 	struct ctl_cmd_entry *entry;
10624 
10625 	retval = CTL_RETVAL_COMPLETE;
10626 
10627 	CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
10628 
10629 	entry = &ctl_cmd_table[ctsio->cdb[0]];
10630 
10631 	/*
10632 	 * If this I/O has been aborted, just send it straight to
10633 	 * ctl_done() without executing it.
10634 	 */
10635 	if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
10636 		ctl_done((union ctl_io *)ctsio);
10637 		goto bailout;
10638 	}
10639 
10640 	/*
10641 	 * All the checks should have been handled by ctl_scsiio_precheck().
10642 	 * We should be clear now to just execute the I/O.
10643 	 */
10644 	retval = entry->execute(ctsio);
10645 
10646 bailout:
10647 	return (retval);
10648 }
10649 
10650 /*
10651  * Since we only implement one target right now, a bus reset simply resets
10652  * our single target.
10653  */
10654 static int
10655 ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io)
10656 {
10657 	return(ctl_target_reset(ctl_softc, io, CTL_UA_BUS_RESET));
10658 }
10659 
10660 static int
10661 ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
10662 		 ctl_ua_type ua_type)
10663 {
10664 	struct ctl_lun *lun;
10665 	int retval;
10666 
10667 	if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
10668 		union ctl_ha_msg msg_info;
10669 
10670 		io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
10671 		msg_info.hdr.nexus = io->io_hdr.nexus;
10672 		if (ua_type==CTL_UA_TARG_RESET)
10673 			msg_info.task.task_action = CTL_TASK_TARGET_RESET;
10674 		else
10675 			msg_info.task.task_action = CTL_TASK_BUS_RESET;
10676 		msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
10677 		msg_info.hdr.original_sc = NULL;
10678 		msg_info.hdr.serializing_sc = NULL;
10679 		if (CTL_HA_STATUS_SUCCESS != ctl_ha_msg_send(CTL_HA_CHAN_CTL,
10680 		    (void *)&msg_info, sizeof(msg_info), 0)) {
10681 		}
10682 	}
10683 	retval = 0;
10684 
10685 	STAILQ_FOREACH(lun, &ctl_softc->lun_list, links)
10686 		retval += ctl_lun_reset(lun, io, ua_type);
10687 
10688 	return (retval);
10689 }
10690 
10691 /*
10692  * The LUN should always be set.  The I/O is optional, and is used to
10693  * distinguish between I/Os sent by this initiator, and by other
10694  * initiators.  We set unit attention for initiators other than this one.
10695  * SAM-3 is vague on this point.  It does say that a unit attention should
10696  * be established for other initiators when a LUN is reset (see section
10697  * 5.7.3), but it doesn't specifically say that the unit attention should
10698  * be established for this particular initiator when a LUN is reset.  Here
10699  * is the relevant text, from SAM-3 rev 8:
10700  *
10701  * 5.7.2 When a SCSI initiator port aborts its own tasks
10702  *
10703  * When a SCSI initiator port causes its own task(s) to be aborted, no
10704  * notification that the task(s) have been aborted shall be returned to
10705  * the SCSI initiator port other than the completion response for the
10706  * command or task management function action that caused the task(s) to
10707  * be aborted and notification(s) associated with related effects of the
10708  * action (e.g., a reset unit attention condition).
10709  *
10710  * XXX KDM for now, we're setting unit attention for all initiators.
10711  */
10712 static int
10713 ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
10714 {
10715 	union ctl_io *xio;
10716 #if 0
10717 	uint32_t initindex;
10718 #endif
10719 	int i;
10720 
10721 	/*
10722 	 * Run through the OOA queue and abort each I/O.
10723 	 */
10724 #if 0
10725 	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
10726 #endif
10727 	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
10728 	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
10729 		xio->io_hdr.flags |= CTL_FLAG_ABORT;
10730 	}
10731 
10732 	/*
10733 	 * This version sets unit attention for every
10734 	 */
10735 #if 0
10736 	initindex = ctl_get_initindex(&io->io_hdr.nexus);
10737 	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
10738 		if (initindex == i)
10739 			continue;
10740 		lun->pending_sense[i].ua_pending |= ua_type;
10741 	}
10742 #endif
10743 
10744 	/*
10745 	 * A reset (any kind, really) clears reservations established with
10746 	 * RESERVE/RELEASE.  It does not clear reservations established
10747 	 * with PERSISTENT RESERVE OUT, but we don't support that at the
10748 	 * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
10749 	 * reservations made with the RESERVE/RELEASE commands, because
10750 	 * those commands are obsolete in SPC-3.
10751 	 */
10752 	lun->flags &= ~CTL_LUN_RESERVED;
10753 
10754 	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
10755 		ctl_clear_mask(lun->have_ca, i);
10756 		lun->pending_sense[i].ua_pending |= ua_type;
10757 	}
10758 
10759 	return (0);
10760 }
10761 
10762 static int
10763 ctl_abort_task(union ctl_io *io)
10764 {
10765 	union ctl_io *xio;
10766 	struct ctl_lun *lun;
10767 	struct ctl_softc *ctl_softc;
10768 #if 0
10769 	struct sbuf sb;
10770 	char printbuf[128];
10771 #endif
10772 	int found;
10773 
10774 	ctl_softc = control_softc;
10775 	found = 0;
10776 
10777 	/*
10778 	 * Look up the LUN.
10779 	 */
10780 	if ((io->io_hdr.nexus.targ_lun < CTL_MAX_LUNS)
10781 	 && (ctl_softc->ctl_luns[io->io_hdr.nexus.targ_lun] != NULL))
10782 		lun = ctl_softc->ctl_luns[io->io_hdr.nexus.targ_lun];
10783 	else
10784 		goto bailout;
10785 
10786 #if 0
10787 	printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
10788 	       lun->lun, io->taskio.tag_num, io->taskio.tag_type);
10789 #endif
10790 
10791 	/*
10792 	 * Run through the OOA queue and attempt to find the given I/O.
10793 	 * The target port, initiator ID, tag type and tag number have to
10794 	 * match the values that we got from the initiator.  If we have an
10795 	 * untagged command to abort, simply abort the first untagged command
10796 	 * we come to.  We only allow one untagged command at a time of course.
10797 	 */
10798 #if 0
10799 	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
10800 #endif
10801 	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
10802 	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
10803 #if 0
10804 		sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
10805 
10806 		sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
10807 			    lun->lun, xio->scsiio.tag_num,
10808 			    xio->scsiio.tag_type,
10809 			    (xio->io_hdr.blocked_links.tqe_prev
10810 			    == NULL) ? "" : " BLOCKED",
10811 			    (xio->io_hdr.flags &
10812 			    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
10813 			    (xio->io_hdr.flags &
10814 			    CTL_FLAG_ABORT) ? " ABORT" : ""),
10815 			    (xio->io_hdr.flags &
10816 			    CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : "");
10817 		ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
10818 		sbuf_finish(&sb);
10819 		printf("%s\n", sbuf_data(&sb));
10820 #endif
10821 
10822 		if ((xio->io_hdr.nexus.targ_port == io->io_hdr.nexus.targ_port)
10823 		 && (xio->io_hdr.nexus.initid.id ==
10824 		     io->io_hdr.nexus.initid.id)) {
10825 			/*
10826 			 * If the abort says that the task is untagged, the
10827 			 * task in the queue must be untagged.  Otherwise,
10828 			 * we just check to see whether the tag numbers
10829 			 * match.  This is because the QLogic firmware
10830 			 * doesn't pass back the tag type in an abort
10831 			 * request.
10832 			 */
10833 #if 0
10834 			if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
10835 			  && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
10836 			 || (xio->scsiio.tag_num == io->taskio.tag_num)) {
10837 #endif
10838 			/*
10839 			 * XXX KDM we've got problems with FC, because it
10840 			 * doesn't send down a tag type with aborts.  So we
10841 			 * can only really go by the tag number...
10842 			 * This may cause problems with parallel SCSI.
10843 			 * Need to figure that out!!
10844 			 */
10845 			if (xio->scsiio.tag_num == io->taskio.tag_num) {
10846 				xio->io_hdr.flags |= CTL_FLAG_ABORT;
10847 				found = 1;
10848 				if ((io->io_hdr.flags &
10849 				     CTL_FLAG_FROM_OTHER_SC) == 0 &&
10850 				    !(lun->flags & CTL_LUN_PRIMARY_SC)) {
10851 					union ctl_ha_msg msg_info;
10852 
10853 					io->io_hdr.flags |=
10854 					                CTL_FLAG_SENT_2OTHER_SC;
10855 					msg_info.hdr.nexus = io->io_hdr.nexus;
10856 					msg_info.task.task_action =
10857 						CTL_TASK_ABORT_TASK;
10858 					msg_info.task.tag_num =
10859 						io->taskio.tag_num;
10860 					msg_info.task.tag_type =
10861 						io->taskio.tag_type;
10862 					msg_info.hdr.msg_type =
10863 						CTL_MSG_MANAGE_TASKS;
10864 					msg_info.hdr.original_sc = NULL;
10865 					msg_info.hdr.serializing_sc = NULL;
10866 #if 0
10867 					printf("Sent Abort to other side\n");
10868 #endif
10869 					if (CTL_HA_STATUS_SUCCESS !=
10870 					        ctl_ha_msg_send(CTL_HA_CHAN_CTL,
10871 		    				(void *)&msg_info,
10872 						sizeof(msg_info), 0)) {
10873 					}
10874 				}
10875 #if 0
10876 				printf("ctl_abort_task: found I/O to abort\n");
10877 #endif
10878 				break;
10879 			}
10880 		}
10881 	}
10882 
10883 bailout:
10884 
10885 	if (found == 0) {
10886 		/*
10887 		 * This isn't really an error.  It's entirely possible for
10888 		 * the abort and command completion to cross on the wire.
10889 		 * This is more of an informative/diagnostic error.
10890 		 */
10891 #if 0
10892 		printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
10893 		       "%d:%d:%d:%d tag %d type %d\n",
10894 		       io->io_hdr.nexus.initid.id,
10895 		       io->io_hdr.nexus.targ_port,
10896 		       io->io_hdr.nexus.targ_target.id,
10897 		       io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
10898 		       io->taskio.tag_type);
10899 #endif
10900 		return (1);
10901 	} else
10902 		return (0);
10903 }
10904 
10905 /*
10906  * Assumptions:  caller holds ctl_softc->ctl_lock
10907  *
10908  * This routine cannot block!  It must be callable from an interrupt
10909  * handler as well as from the work thread.
10910  */
10911 static void
10912 ctl_run_task_queue(struct ctl_softc *ctl_softc)
10913 {
10914 	union ctl_io *io, *next_io;
10915 
10916 	CTL_DEBUG_PRINT(("ctl_run_task_queue\n"));
10917 
10918 	for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->task_queue);
10919 	     io != NULL; io = next_io) {
10920 		int retval;
10921 		const char *task_desc;
10922 
10923 		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
10924 
10925 		retval = 0;
10926 
10927 		switch (io->io_hdr.io_type) {
10928 		case CTL_IO_TASK: {
10929 			task_desc = ctl_scsi_task_string(&io->taskio);
10930 			if (task_desc != NULL) {
10931 #ifdef NEEDTOPORT
10932 				csevent_log(CSC_CTL | CSC_SHELF_SW |
10933 					    CTL_TASK_REPORT,
10934 					    csevent_LogType_Trace,
10935 					    csevent_Severity_Information,
10936 					    csevent_AlertLevel_Green,
10937 					    csevent_FRU_Firmware,
10938 					    csevent_FRU_Unknown,
10939 					    "CTL: received task: %s",task_desc);
10940 #endif
10941 			} else {
10942 #ifdef NEEDTOPORT
10943 				csevent_log(CSC_CTL | CSC_SHELF_SW |
10944 					    CTL_TASK_REPORT,
10945 					    csevent_LogType_Trace,
10946 					    csevent_Severity_Information,
10947 					    csevent_AlertLevel_Green,
10948 					    csevent_FRU_Firmware,
10949 					    csevent_FRU_Unknown,
10950 					    "CTL: received unknown task "
10951 					    "type: %d (%#x)",
10952 					    io->taskio.task_action,
10953 					    io->taskio.task_action);
10954 #endif
10955 			}
10956 			switch (io->taskio.task_action) {
10957 			case CTL_TASK_ABORT_TASK:
10958 				retval = ctl_abort_task(io);
10959 				break;
10960 			case CTL_TASK_ABORT_TASK_SET:
10961 				break;
10962 			case CTL_TASK_CLEAR_ACA:
10963 				break;
10964 			case CTL_TASK_CLEAR_TASK_SET:
10965 				break;
10966 			case CTL_TASK_LUN_RESET: {
10967 				struct ctl_lun *lun;
10968 				uint32_t targ_lun;
10969 				int retval;
10970 
10971 				targ_lun = io->io_hdr.nexus.targ_lun;
10972 
10973 				if ((targ_lun < CTL_MAX_LUNS)
10974 				 && (ctl_softc->ctl_luns[targ_lun] != NULL))
10975 					lun = ctl_softc->ctl_luns[targ_lun];
10976 				else {
10977 					retval = 1;
10978 					break;
10979 				}
10980 
10981 				if (!(io->io_hdr.flags &
10982 				    CTL_FLAG_FROM_OTHER_SC)) {
10983 					union ctl_ha_msg msg_info;
10984 
10985 					io->io_hdr.flags |=
10986 						CTL_FLAG_SENT_2OTHER_SC;
10987 					msg_info.hdr.msg_type =
10988 						CTL_MSG_MANAGE_TASKS;
10989 					msg_info.hdr.nexus = io->io_hdr.nexus;
10990 					msg_info.task.task_action =
10991 						CTL_TASK_LUN_RESET;
10992 					msg_info.hdr.original_sc = NULL;
10993 					msg_info.hdr.serializing_sc = NULL;
10994 					if (CTL_HA_STATUS_SUCCESS !=
10995 					    ctl_ha_msg_send(CTL_HA_CHAN_CTL,
10996 					    (void *)&msg_info,
10997 					    sizeof(msg_info), 0)) {
10998 					}
10999 				}
11000 
11001 				retval = ctl_lun_reset(lun, io,
11002 						       CTL_UA_LUN_RESET);
11003 				break;
11004 			}
11005 			case CTL_TASK_TARGET_RESET:
11006 				retval = ctl_target_reset(ctl_softc, io,
11007 							  CTL_UA_TARG_RESET);
11008 				break;
11009 			case CTL_TASK_BUS_RESET:
11010 				retval = ctl_bus_reset(ctl_softc, io);
11011 				break;
11012 			case CTL_TASK_PORT_LOGIN:
11013 				break;
11014 			case CTL_TASK_PORT_LOGOUT:
11015 				break;
11016 			default:
11017 				printf("ctl_run_task_queue: got unknown task "
11018 				       "management event %d\n",
11019 				       io->taskio.task_action);
11020 				break;
11021 			}
11022 			if (retval == 0)
11023 				io->io_hdr.status = CTL_SUCCESS;
11024 			else
11025 				io->io_hdr.status = CTL_ERROR;
11026 
11027 			STAILQ_REMOVE(&ctl_softc->task_queue, &io->io_hdr,
11028 				      ctl_io_hdr, links);
11029 			/*
11030 			 * This will queue this I/O to the done queue, but the
11031 			 * work thread won't be able to process it until we
11032 			 * return and the lock is released.
11033 			 */
11034 			ctl_done_lock(io, /*have_lock*/ 1);
11035 			break;
11036 		}
11037 		default: {
11038 
11039 			printf("%s: invalid I/O type %d msg %d cdb %x"
11040 			       " iptl: %ju:%d:%ju:%d tag 0x%04x\n",
11041 			       __func__, io->io_hdr.io_type,
11042 			       io->io_hdr.msg_type, io->scsiio.cdb[0],
11043 			       (uintmax_t)io->io_hdr.nexus.initid.id,
11044 			       io->io_hdr.nexus.targ_port,
11045 			       (uintmax_t)io->io_hdr.nexus.targ_target.id,
11046 			       io->io_hdr.nexus.targ_lun,
11047 			       (io->io_hdr.io_type == CTL_IO_TASK) ?
11048 			       io->taskio.tag_num : io->scsiio.tag_num);
11049 			STAILQ_REMOVE(&ctl_softc->task_queue, &io->io_hdr,
11050 				      ctl_io_hdr, links);
11051 			ctl_free_io_internal(io, 1);
11052 			break;
11053 		}
11054 		}
11055 	}
11056 
11057 	ctl_softc->flags &= ~CTL_FLAG_TASK_PENDING;
11058 }
11059 
11060 /*
11061  * For HA operation.  Handle commands that come in from the other
11062  * controller.
11063  */
11064 static void
11065 ctl_handle_isc(union ctl_io *io)
11066 {
11067 	int free_io;
11068 	struct ctl_lun *lun;
11069 	struct ctl_softc *ctl_softc;
11070 
11071 	ctl_softc = control_softc;
11072 
11073 	lun = ctl_softc->ctl_luns[io->io_hdr.nexus.targ_lun];
11074 
11075 	switch (io->io_hdr.msg_type) {
11076 	case CTL_MSG_SERIALIZE:
11077 		free_io = ctl_serialize_other_sc_cmd(&io->scsiio,
11078 						     /*have_lock*/ 0);
11079 		break;
11080 	case CTL_MSG_R2R: {
11081 		uint8_t opcode;
11082 		struct ctl_cmd_entry *entry;
11083 
11084 		/*
11085 		 * This is only used in SER_ONLY mode.
11086 		 */
11087 		free_io = 0;
11088 		opcode = io->scsiio.cdb[0];
11089 		entry = &ctl_cmd_table[opcode];
11090 		mtx_lock(&ctl_softc->ctl_lock);
11091 		if (ctl_scsiio_lun_check(ctl_softc, lun,
11092 		    entry, (struct ctl_scsiio *)io) != 0) {
11093 			ctl_done_lock(io, /*have_lock*/ 1);
11094 			mtx_unlock(&ctl_softc->ctl_lock);
11095 			break;
11096 		}
11097 		io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11098 		STAILQ_INSERT_TAIL(&ctl_softc->rtr_queue,
11099 				   &io->io_hdr, links);
11100 		mtx_unlock(&ctl_softc->ctl_lock);
11101 		break;
11102 	}
11103 	case CTL_MSG_FINISH_IO:
11104 		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
11105 			free_io = 0;
11106 			ctl_done_lock(io, /*have_lock*/ 0);
11107 		} else {
11108 			free_io = 1;
11109 			mtx_lock(&ctl_softc->ctl_lock);
11110 			TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
11111 				     ooa_links);
11112 			STAILQ_REMOVE(&ctl_softc->task_queue,
11113 				      &io->io_hdr, ctl_io_hdr, links);
11114 			ctl_check_blocked(lun);
11115 			mtx_unlock(&ctl_softc->ctl_lock);
11116 		}
11117 		break;
11118 	case CTL_MSG_PERS_ACTION:
11119 		ctl_hndl_per_res_out_on_other_sc(
11120 			(union ctl_ha_msg *)&io->presio.pr_msg);
11121 		free_io = 1;
11122 		break;
11123 	case CTL_MSG_BAD_JUJU:
11124 		free_io = 0;
11125 		ctl_done_lock(io, /*have_lock*/ 0);
11126 		break;
11127 	case CTL_MSG_DATAMOVE:
11128 		/* Only used in XFER mode */
11129 		free_io = 0;
11130 		ctl_datamove_remote(io);
11131 		break;
11132 	case CTL_MSG_DATAMOVE_DONE:
11133 		/* Only used in XFER mode */
11134 		free_io = 0;
11135 		io->scsiio.be_move_done(io);
11136 		break;
11137 	default:
11138 		free_io = 1;
11139 		printf("%s: Invalid message type %d\n",
11140 		       __func__, io->io_hdr.msg_type);
11141 		break;
11142 	}
11143 	if (free_io)
11144 		ctl_free_io_internal(io, 0);
11145 
11146 }
11147 
11148 
11149 /*
11150  * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
11151  * there is no match.
11152  */
11153 static ctl_lun_error_pattern
11154 ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
11155 {
11156 	struct ctl_cmd_entry *entry;
11157 	ctl_lun_error_pattern filtered_pattern, pattern;
11158 	uint8_t opcode;
11159 
11160 	pattern = desc->error_pattern;
11161 
11162 	/*
11163 	 * XXX KDM we need more data passed into this function to match a
11164 	 * custom pattern, and we actually need to implement custom pattern
11165 	 * matching.
11166 	 */
11167 	if (pattern & CTL_LUN_PAT_CMD)
11168 		return (CTL_LUN_PAT_CMD);
11169 
11170 	if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
11171 		return (CTL_LUN_PAT_ANY);
11172 
11173 	opcode = ctsio->cdb[0];
11174 	entry = &ctl_cmd_table[opcode];
11175 
11176 	filtered_pattern = entry->pattern & pattern;
11177 
11178 	/*
11179 	 * If the user requested specific flags in the pattern (e.g.
11180 	 * CTL_LUN_PAT_RANGE), make sure the command supports all of those
11181 	 * flags.
11182 	 *
11183 	 * If the user did not specify any flags, it doesn't matter whether
11184 	 * or not the command supports the flags.
11185 	 */
11186 	if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
11187 	     (pattern & ~CTL_LUN_PAT_MASK))
11188 		return (CTL_LUN_PAT_NONE);
11189 
11190 	/*
11191 	 * If the user asked for a range check, see if the requested LBA
11192 	 * range overlaps with this command's LBA range.
11193 	 */
11194 	if (filtered_pattern & CTL_LUN_PAT_RANGE) {
11195 		uint64_t lba1;
11196 		uint32_t len1;
11197 		ctl_action action;
11198 		int retval;
11199 
11200 		retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
11201 		if (retval != 0)
11202 			return (CTL_LUN_PAT_NONE);
11203 
11204 		action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
11205 					      desc->lba_range.len);
11206 		/*
11207 		 * A "pass" means that the LBA ranges don't overlap, so
11208 		 * this doesn't match the user's range criteria.
11209 		 */
11210 		if (action == CTL_ACTION_PASS)
11211 			return (CTL_LUN_PAT_NONE);
11212 	}
11213 
11214 	return (filtered_pattern);
11215 }
11216 
11217 /*
11218  * Called with the CTL lock held.
11219  */
11220 static void
11221 ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
11222 {
11223 	struct ctl_error_desc *desc, *desc2;
11224 
11225 	STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
11226 		ctl_lun_error_pattern pattern;
11227 		/*
11228 		 * Check to see whether this particular command matches
11229 		 * the pattern in the descriptor.
11230 		 */
11231 		pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
11232 		if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
11233 			continue;
11234 
11235 		switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
11236 		case CTL_LUN_INJ_ABORTED:
11237 			ctl_set_aborted(&io->scsiio);
11238 			break;
11239 		case CTL_LUN_INJ_MEDIUM_ERR:
11240 			ctl_set_medium_error(&io->scsiio);
11241 			break;
11242 		case CTL_LUN_INJ_UA:
11243 			/* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
11244 			 * OCCURRED */
11245 			ctl_set_ua(&io->scsiio, 0x29, 0x00);
11246 			break;
11247 		case CTL_LUN_INJ_CUSTOM:
11248 			/*
11249 			 * We're assuming the user knows what he is doing.
11250 			 * Just copy the sense information without doing
11251 			 * checks.
11252 			 */
11253 			bcopy(&desc->custom_sense, &io->scsiio.sense_data,
11254 			      ctl_min(sizeof(desc->custom_sense),
11255 				      sizeof(io->scsiio.sense_data)));
11256 			io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
11257 			io->scsiio.sense_len = SSD_FULL_SIZE;
11258 			io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
11259 			break;
11260 		case CTL_LUN_INJ_NONE:
11261 		default:
11262 			/*
11263 			 * If this is an error injection type we don't know
11264 			 * about, clear the continuous flag (if it is set)
11265 			 * so it will get deleted below.
11266 			 */
11267 			desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
11268 			break;
11269 		}
11270 		/*
11271 		 * By default, each error injection action is a one-shot
11272 		 */
11273 		if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
11274 			continue;
11275 
11276 		STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
11277 
11278 		free(desc, M_CTL);
11279 	}
11280 }
11281 
11282 #ifdef CTL_IO_DELAY
11283 static void
11284 ctl_datamove_timer_wakeup(void *arg)
11285 {
11286 	union ctl_io *io;
11287 
11288 	io = (union ctl_io *)arg;
11289 
11290 	ctl_datamove(io);
11291 }
11292 #endif /* CTL_IO_DELAY */
11293 
11294 /*
11295  * Assumption:  caller does NOT hold ctl_lock
11296  */
11297 void
11298 ctl_datamove(union ctl_io *io)
11299 {
11300 	void (*fe_datamove)(union ctl_io *io);
11301 
11302 	CTL_DEBUG_PRINT(("ctl_datamove\n"));
11303 
11304 #ifdef CTL_TIME_IO
11305 	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
11306 		char str[256];
11307 		char path_str[64];
11308 		struct sbuf sb;
11309 
11310 		ctl_scsi_path_string(io, path_str, sizeof(path_str));
11311 		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
11312 
11313 		sbuf_cat(&sb, path_str);
11314 		switch (io->io_hdr.io_type) {
11315 		case CTL_IO_SCSI:
11316 			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
11317 			sbuf_printf(&sb, "\n");
11318 			sbuf_cat(&sb, path_str);
11319 			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
11320 				    io->scsiio.tag_num, io->scsiio.tag_type);
11321 			break;
11322 		case CTL_IO_TASK:
11323 			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
11324 				    "Tag Type: %d\n", io->taskio.task_action,
11325 				    io->taskio.tag_num, io->taskio.tag_type);
11326 			break;
11327 		default:
11328 			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
11329 			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
11330 			break;
11331 		}
11332 		sbuf_cat(&sb, path_str);
11333 		sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
11334 			    (intmax_t)time_uptime - io->io_hdr.start_time);
11335 		sbuf_finish(&sb);
11336 		printf("%s", sbuf_data(&sb));
11337 	}
11338 #endif /* CTL_TIME_IO */
11339 
11340 	mtx_lock(&control_softc->ctl_lock);
11341 #ifdef CTL_IO_DELAY
11342 	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
11343 		struct ctl_lun *lun;
11344 
11345 		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
11346 
11347 		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
11348 	} else {
11349 		struct ctl_lun *lun;
11350 
11351 		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
11352 		if ((lun != NULL)
11353 		 && (lun->delay_info.datamove_delay > 0)) {
11354 			struct callout *callout;
11355 
11356 			callout = (struct callout *)&io->io_hdr.timer_bytes;
11357 			callout_init(callout, /*mpsafe*/ 1);
11358 			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
11359 			callout_reset(callout,
11360 				      lun->delay_info.datamove_delay * hz,
11361 				      ctl_datamove_timer_wakeup, io);
11362 			if (lun->delay_info.datamove_type ==
11363 			    CTL_DELAY_TYPE_ONESHOT)
11364 				lun->delay_info.datamove_delay = 0;
11365 			mtx_unlock(&control_softc->ctl_lock);
11366 			return;
11367 		}
11368 	}
11369 #endif
11370 	/*
11371 	 * If we have any pending task management commands, process them
11372 	 * first.  This is necessary to eliminate a race condition with the
11373 	 * FETD:
11374 	 *
11375 	 * - FETD submits a task management command, like an abort.
11376 	 * - Back end calls fe_datamove() to move the data for the aborted
11377 	 *   command.  The FETD can't really accept it, but if it did, it
11378 	 *   would end up transmitting data for a command that the initiator
11379 	 *   told us to abort.
11380 	 *
11381 	 * We close the race by processing all pending task management
11382 	 * commands here (we can't block!), and then check this I/O to see
11383 	 * if it has been aborted.  If so, return it to the back end with
11384 	 * bad status, so the back end can say return an error to the back end
11385 	 * and then when the back end returns an error, we can return the
11386 	 * aborted command to the FETD, so it can clean up its resources.
11387 	 */
11388 	if (control_softc->flags & CTL_FLAG_TASK_PENDING)
11389 		ctl_run_task_queue(control_softc);
11390 
11391 	/*
11392 	 * This command has been aborted.  Set the port status, so we fail
11393 	 * the data move.
11394 	 */
11395 	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
11396 		printf("ctl_datamove: tag 0x%04x on (%ju:%d:%ju:%d) aborted\n",
11397 		       io->scsiio.tag_num,(uintmax_t)io->io_hdr.nexus.initid.id,
11398 		       io->io_hdr.nexus.targ_port,
11399 		       (uintmax_t)io->io_hdr.nexus.targ_target.id,
11400 		       io->io_hdr.nexus.targ_lun);
11401 		io->io_hdr.status = CTL_CMD_ABORTED;
11402 		io->io_hdr.port_status = 31337;
11403 		mtx_unlock(&control_softc->ctl_lock);
11404 		/*
11405 		 * Note that the backend, in this case, will get the
11406 		 * callback in its context.  In other cases it may get
11407 		 * called in the frontend's interrupt thread context.
11408 		 */
11409 		io->scsiio.be_move_done(io);
11410 		return;
11411 	}
11412 
11413 	/*
11414 	 * If we're in XFER mode and this I/O is from the other shelf
11415 	 * controller, we need to send the DMA to the other side to
11416 	 * actually transfer the data to/from the host.  In serialize only
11417 	 * mode the transfer happens below CTL and ctl_datamove() is only
11418 	 * called on the machine that originally received the I/O.
11419 	 */
11420 	if ((control_softc->ha_mode == CTL_HA_MODE_XFER)
11421 	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11422 		union ctl_ha_msg msg;
11423 		uint32_t sg_entries_sent;
11424 		int do_sg_copy;
11425 		int i;
11426 
11427 		memset(&msg, 0, sizeof(msg));
11428 		msg.hdr.msg_type = CTL_MSG_DATAMOVE;
11429 		msg.hdr.original_sc = io->io_hdr.original_sc;
11430 		msg.hdr.serializing_sc = io;
11431 		msg.hdr.nexus = io->io_hdr.nexus;
11432 		msg.dt.flags = io->io_hdr.flags;
11433 		/*
11434 		 * We convert everything into a S/G list here.  We can't
11435 		 * pass by reference, only by value between controllers.
11436 		 * So we can't pass a pointer to the S/G list, only as many
11437 		 * S/G entries as we can fit in here.  If it's possible for
11438 		 * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
11439 		 * then we need to break this up into multiple transfers.
11440 		 */
11441 		if (io->scsiio.kern_sg_entries == 0) {
11442 			msg.dt.kern_sg_entries = 1;
11443 			/*
11444 			 * If this is in cached memory, flush the cache
11445 			 * before we send the DMA request to the other
11446 			 * controller.  We want to do this in either the
11447 			 * read or the write case.  The read case is
11448 			 * straightforward.  In the write case, we want to
11449 			 * make sure nothing is in the local cache that
11450 			 * could overwrite the DMAed data.
11451 			 */
11452 			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
11453 				/*
11454 				 * XXX KDM use bus_dmamap_sync() here.
11455 				 */
11456 			}
11457 
11458 			/*
11459 			 * Convert to a physical address if this is a
11460 			 * virtual address.
11461 			 */
11462 			if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
11463 				msg.dt.sg_list[0].addr =
11464 					io->scsiio.kern_data_ptr;
11465 			} else {
11466 				/*
11467 				 * XXX KDM use busdma here!
11468 				 */
11469 #if 0
11470 				msg.dt.sg_list[0].addr = (void *)
11471 					vtophys(io->scsiio.kern_data_ptr);
11472 #endif
11473 			}
11474 
11475 			msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
11476 			do_sg_copy = 0;
11477 		} else {
11478 			struct ctl_sg_entry *sgl;
11479 
11480 			do_sg_copy = 1;
11481 			msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
11482 			sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
11483 			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
11484 				/*
11485 				 * XXX KDM use bus_dmamap_sync() here.
11486 				 */
11487 			}
11488 		}
11489 
11490 		msg.dt.kern_data_len = io->scsiio.kern_data_len;
11491 		msg.dt.kern_total_len = io->scsiio.kern_total_len;
11492 		msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
11493 		msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
11494 		msg.dt.sg_sequence = 0;
11495 
11496 		/*
11497 		 * Loop until we've sent all of the S/G entries.  On the
11498 		 * other end, we'll recompose these S/G entries into one
11499 		 * contiguous list before passing it to the
11500 		 */
11501 		for (sg_entries_sent = 0; sg_entries_sent <
11502 		     msg.dt.kern_sg_entries; msg.dt.sg_sequence++) {
11503 			msg.dt.cur_sg_entries = ctl_min((sizeof(msg.dt.sg_list)/
11504 				sizeof(msg.dt.sg_list[0])),
11505 				msg.dt.kern_sg_entries - sg_entries_sent);
11506 
11507 			if (do_sg_copy != 0) {
11508 				struct ctl_sg_entry *sgl;
11509 				int j;
11510 
11511 				sgl = (struct ctl_sg_entry *)
11512 					io->scsiio.kern_data_ptr;
11513 				/*
11514 				 * If this is in cached memory, flush the cache
11515 				 * before we send the DMA request to the other
11516 				 * controller.  We want to do this in either
11517 				 * the * read or the write case.  The read
11518 				 * case is straightforward.  In the write
11519 				 * case, we want to make sure nothing is
11520 				 * in the local cache that could overwrite
11521 				 * the DMAed data.
11522 				 */
11523 
11524 				for (i = sg_entries_sent, j = 0;
11525 				     i < msg.dt.cur_sg_entries; i++, j++) {
11526 					if ((io->io_hdr.flags &
11527 					     CTL_FLAG_NO_DATASYNC) == 0) {
11528 						/*
11529 						 * XXX KDM use bus_dmamap_sync()
11530 						 */
11531 					}
11532 					if ((io->io_hdr.flags &
11533 					     CTL_FLAG_BUS_ADDR) == 0) {
11534 						/*
11535 						 * XXX KDM use busdma.
11536 						 */
11537 #if 0
11538 						msg.dt.sg_list[j].addr =(void *)
11539 						       vtophys(sgl[i].addr);
11540 #endif
11541 					} else {
11542 						msg.dt.sg_list[j].addr =
11543 							sgl[i].addr;
11544 					}
11545 					msg.dt.sg_list[j].len = sgl[i].len;
11546 				}
11547 			}
11548 
11549 			sg_entries_sent += msg.dt.cur_sg_entries;
11550 			if (sg_entries_sent >= msg.dt.kern_sg_entries)
11551 				msg.dt.sg_last = 1;
11552 			else
11553 				msg.dt.sg_last = 0;
11554 
11555 			/*
11556 			 * XXX KDM drop and reacquire the lock here?
11557 			 */
11558 			if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
11559 			    sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
11560 				/*
11561 				 * XXX do something here.
11562 				 */
11563 			}
11564 
11565 			msg.dt.sent_sg_entries = sg_entries_sent;
11566 		}
11567 		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11568 		if (io->io_hdr.flags & CTL_FLAG_FAILOVER)
11569 			ctl_failover_io(io, /*have_lock*/ 1);
11570 
11571 	} else {
11572 
11573 		/*
11574 		 * Lookup the fe_datamove() function for this particular
11575 		 * front end.
11576 		 */
11577 		fe_datamove =
11578 		    control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
11579 		mtx_unlock(&control_softc->ctl_lock);
11580 
11581 		fe_datamove(io);
11582 	}
11583 }
11584 
11585 static void
11586 ctl_send_datamove_done(union ctl_io *io, int have_lock)
11587 {
11588 	union ctl_ha_msg msg;
11589 	int isc_status;
11590 
11591 	memset(&msg, 0, sizeof(msg));
11592 
11593 	msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
11594 	msg.hdr.original_sc = io;
11595 	msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
11596 	msg.hdr.nexus = io->io_hdr.nexus;
11597 	msg.hdr.status = io->io_hdr.status;
11598 	msg.scsi.tag_num = io->scsiio.tag_num;
11599 	msg.scsi.tag_type = io->scsiio.tag_type;
11600 	msg.scsi.scsi_status = io->scsiio.scsi_status;
11601 	memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
11602 	       sizeof(io->scsiio.sense_data));
11603 	msg.scsi.sense_len = io->scsiio.sense_len;
11604 	msg.scsi.sense_residual = io->scsiio.sense_residual;
11605 	msg.scsi.fetd_status = io->io_hdr.port_status;
11606 	msg.scsi.residual = io->scsiio.residual;
11607 	io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11608 
11609 	if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
11610 		ctl_failover_io(io, /*have_lock*/ have_lock);
11611 		return;
11612 	}
11613 
11614 	isc_status = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0);
11615 	if (isc_status > CTL_HA_STATUS_SUCCESS) {
11616 		/* XXX do something if this fails */
11617 	}
11618 
11619 }
11620 
11621 /*
11622  * The DMA to the remote side is done, now we need to tell the other side
11623  * we're done so it can continue with its data movement.
11624  */
11625 static void
11626 ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
11627 {
11628 	union ctl_io *io;
11629 
11630 	io = rq->context;
11631 
11632 	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
11633 		printf("%s: ISC DMA write failed with error %d", __func__,
11634 		       rq->ret);
11635 		ctl_set_internal_failure(&io->scsiio,
11636 					 /*sks_valid*/ 1,
11637 					 /*retry_count*/ rq->ret);
11638 	}
11639 
11640 	ctl_dt_req_free(rq);
11641 
11642 	/*
11643 	 * In this case, we had to malloc the memory locally.  Free it.
11644 	 */
11645 	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
11646 		int i;
11647 		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
11648 			free(io->io_hdr.local_sglist[i].addr, M_CTL);
11649 	}
11650 	/*
11651 	 * The data is in local and remote memory, so now we need to send
11652 	 * status (good or back) back to the other side.
11653 	 */
11654 	ctl_send_datamove_done(io, /*have_lock*/ 0);
11655 }
11656 
11657 /*
11658  * We've moved the data from the host/controller into local memory.  Now we
11659  * need to push it over to the remote controller's memory.
11660  */
11661 static int
11662 ctl_datamove_remote_dm_write_cb(union ctl_io *io)
11663 {
11664 	int retval;
11665 
11666 	retval = 0;
11667 
11668 	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
11669 					  ctl_datamove_remote_write_cb);
11670 
11671 	return (retval);
11672 }
11673 
11674 static void
11675 ctl_datamove_remote_write(union ctl_io *io)
11676 {
11677 	int retval;
11678 	void (*fe_datamove)(union ctl_io *io);
11679 
11680 	/*
11681 	 * - Get the data from the host/HBA into local memory.
11682 	 * - DMA memory from the local controller to the remote controller.
11683 	 * - Send status back to the remote controller.
11684 	 */
11685 
11686 	retval = ctl_datamove_remote_sgl_setup(io);
11687 	if (retval != 0)
11688 		return;
11689 
11690 	/* Switch the pointer over so the FETD knows what to do */
11691 	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
11692 
11693 	/*
11694 	 * Use a custom move done callback, since we need to send completion
11695 	 * back to the other controller, not to the backend on this side.
11696 	 */
11697 	io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
11698 
11699 	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
11700 
11701 	fe_datamove(io);
11702 
11703 	return;
11704 
11705 }
11706 
11707 static int
11708 ctl_datamove_remote_dm_read_cb(union ctl_io *io)
11709 {
11710 #if 0
11711 	char str[256];
11712 	char path_str[64];
11713 	struct sbuf sb;
11714 #endif
11715 
11716 	/*
11717 	 * In this case, we had to malloc the memory locally.  Free it.
11718 	 */
11719 	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
11720 		int i;
11721 		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
11722 			free(io->io_hdr.local_sglist[i].addr, M_CTL);
11723 	}
11724 
11725 #if 0
11726 	scsi_path_string(io, path_str, sizeof(path_str));
11727 	sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
11728 	sbuf_cat(&sb, path_str);
11729 	scsi_command_string(&io->scsiio, NULL, &sb);
11730 	sbuf_printf(&sb, "\n");
11731 	sbuf_cat(&sb, path_str);
11732 	sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
11733 		    io->scsiio.tag_num, io->scsiio.tag_type);
11734 	sbuf_cat(&sb, path_str);
11735 	sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
11736 		    io->io_hdr.flags, io->io_hdr.status);
11737 	sbuf_finish(&sb);
11738 	printk("%s", sbuf_data(&sb));
11739 #endif
11740 
11741 
11742 	/*
11743 	 * The read is done, now we need to send status (good or bad) back
11744 	 * to the other side.
11745 	 */
11746 	ctl_send_datamove_done(io, /*have_lock*/ 0);
11747 
11748 	return (0);
11749 }
11750 
11751 static void
11752 ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
11753 {
11754 	union ctl_io *io;
11755 	void (*fe_datamove)(union ctl_io *io);
11756 
11757 	io = rq->context;
11758 
11759 	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
11760 		printf("%s: ISC DMA read failed with error %d", __func__,
11761 		       rq->ret);
11762 		ctl_set_internal_failure(&io->scsiio,
11763 					 /*sks_valid*/ 1,
11764 					 /*retry_count*/ rq->ret);
11765 	}
11766 
11767 	ctl_dt_req_free(rq);
11768 
11769 	/* Switch the pointer over so the FETD knows what to do */
11770 	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
11771 
11772 	/*
11773 	 * Use a custom move done callback, since we need to send completion
11774 	 * back to the other controller, not to the backend on this side.
11775 	 */
11776 	io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
11777 
11778 	/* XXX KDM add checks like the ones in ctl_datamove? */
11779 
11780 	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
11781 
11782 	fe_datamove(io);
11783 }
11784 
11785 static int
11786 ctl_datamove_remote_sgl_setup(union ctl_io *io)
11787 {
11788 	struct ctl_sg_entry *local_sglist, *remote_sglist;
11789 	struct ctl_sg_entry *local_dma_sglist, *remote_dma_sglist;
11790 	struct ctl_softc *softc;
11791 	int retval;
11792 	int i;
11793 
11794 	retval = 0;
11795 	softc = control_softc;
11796 
11797 	local_sglist = io->io_hdr.local_sglist;
11798 	local_dma_sglist = io->io_hdr.local_dma_sglist;
11799 	remote_sglist = io->io_hdr.remote_sglist;
11800 	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
11801 
11802 	if (io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) {
11803 		for (i = 0; i < io->scsiio.kern_sg_entries; i++) {
11804 			local_sglist[i].len = remote_sglist[i].len;
11805 
11806 			/*
11807 			 * XXX Detect the situation where the RS-level I/O
11808 			 * redirector on the other side has already read the
11809 			 * data off of the AOR RS on this side, and
11810 			 * transferred it to remote (mirror) memory on the
11811 			 * other side.  Since we already have the data in
11812 			 * memory here, we just need to use it.
11813 			 *
11814 			 * XXX KDM this can probably be removed once we
11815 			 * get the cache device code in and take the
11816 			 * current AOR implementation out.
11817 			 */
11818 #ifdef NEEDTOPORT
11819 			if ((remote_sglist[i].addr >=
11820 			     (void *)vtophys(softc->mirr->addr))
11821 			 && (remote_sglist[i].addr <
11822 			     ((void *)vtophys(softc->mirr->addr) +
11823 			     CacheMirrorOffset))) {
11824 				local_sglist[i].addr = remote_sglist[i].addr -
11825 					CacheMirrorOffset;
11826 				if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
11827 				     CTL_FLAG_DATA_IN)
11828 					io->io_hdr.flags |= CTL_FLAG_REDIR_DONE;
11829 			} else {
11830 				local_sglist[i].addr = remote_sglist[i].addr +
11831 					CacheMirrorOffset;
11832 			}
11833 #endif
11834 #if 0
11835 			printf("%s: local %p, remote %p, len %d\n",
11836 			       __func__, local_sglist[i].addr,
11837 			       remote_sglist[i].addr, local_sglist[i].len);
11838 #endif
11839 		}
11840 	} else {
11841 		uint32_t len_to_go;
11842 
11843 		/*
11844 		 * In this case, we don't have automatically allocated
11845 		 * memory for this I/O on this controller.  This typically
11846 		 * happens with internal CTL I/O -- e.g. inquiry, mode
11847 		 * sense, etc.  Anything coming from RAIDCore will have
11848 		 * a mirror area available.
11849 		 */
11850 		len_to_go = io->scsiio.kern_data_len;
11851 
11852 		/*
11853 		 * Clear the no datasync flag, we have to use malloced
11854 		 * buffers.
11855 		 */
11856 		io->io_hdr.flags &= ~CTL_FLAG_NO_DATASYNC;
11857 
11858 		/*
11859 		 * The difficult thing here is that the size of the various
11860 		 * S/G segments may be different than the size from the
11861 		 * remote controller.  That'll make it harder when DMAing
11862 		 * the data back to the other side.
11863 		 */
11864 		for (i = 0; (i < sizeof(io->io_hdr.remote_sglist) /
11865 		     sizeof(io->io_hdr.remote_sglist[0])) &&
11866 		     (len_to_go > 0); i++) {
11867 			local_sglist[i].len = ctl_min(len_to_go, 131072);
11868 			CTL_SIZE_8B(local_dma_sglist[i].len,
11869 				    local_sglist[i].len);
11870 			local_sglist[i].addr =
11871 				malloc(local_dma_sglist[i].len, M_CTL,M_WAITOK);
11872 
11873 			local_dma_sglist[i].addr = local_sglist[i].addr;
11874 
11875 			if (local_sglist[i].addr == NULL) {
11876 				int j;
11877 
11878 				printf("malloc failed for %zd bytes!",
11879 				       local_dma_sglist[i].len);
11880 				for (j = 0; j < i; j++) {
11881 					free(local_sglist[j].addr, M_CTL);
11882 				}
11883 				ctl_set_internal_failure(&io->scsiio,
11884 							 /*sks_valid*/ 1,
11885 							 /*retry_count*/ 4857);
11886 				retval = 1;
11887 				goto bailout_error;
11888 
11889 			}
11890 			/* XXX KDM do we need a sync here? */
11891 
11892 			len_to_go -= local_sglist[i].len;
11893 		}
11894 		/*
11895 		 * Reset the number of S/G entries accordingly.  The
11896 		 * original number of S/G entries is available in
11897 		 * rem_sg_entries.
11898 		 */
11899 		io->scsiio.kern_sg_entries = i;
11900 
11901 #if 0
11902 		printf("%s: kern_sg_entries = %d\n", __func__,
11903 		       io->scsiio.kern_sg_entries);
11904 		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
11905 			printf("%s: sg[%d] = %p, %d (DMA: %d)\n", __func__, i,
11906 			       local_sglist[i].addr, local_sglist[i].len,
11907 			       local_dma_sglist[i].len);
11908 #endif
11909 	}
11910 
11911 
11912 	return (retval);
11913 
11914 bailout_error:
11915 
11916 	ctl_send_datamove_done(io, /*have_lock*/ 0);
11917 
11918 	return (retval);
11919 }
11920 
11921 static int
11922 ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
11923 			 ctl_ha_dt_cb callback)
11924 {
11925 	struct ctl_ha_dt_req *rq;
11926 	struct ctl_sg_entry *remote_sglist, *local_sglist;
11927 	struct ctl_sg_entry *remote_dma_sglist, *local_dma_sglist;
11928 	uint32_t local_used, remote_used, total_used;
11929 	int retval;
11930 	int i, j;
11931 
11932 	retval = 0;
11933 
11934 	rq = ctl_dt_req_alloc();
11935 
11936 	/*
11937 	 * If we failed to allocate the request, and if the DMA didn't fail
11938 	 * anyway, set busy status.  This is just a resource allocation
11939 	 * failure.
11940 	 */
11941 	if ((rq == NULL)
11942 	 && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE))
11943 		ctl_set_busy(&io->scsiio);
11944 
11945 	if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) {
11946 
11947 		if (rq != NULL)
11948 			ctl_dt_req_free(rq);
11949 
11950 		/*
11951 		 * The data move failed.  We need to return status back
11952 		 * to the other controller.  No point in trying to DMA
11953 		 * data to the remote controller.
11954 		 */
11955 
11956 		ctl_send_datamove_done(io, /*have_lock*/ 0);
11957 
11958 		retval = 1;
11959 
11960 		goto bailout;
11961 	}
11962 
11963 	local_sglist = io->io_hdr.local_sglist;
11964 	local_dma_sglist = io->io_hdr.local_dma_sglist;
11965 	remote_sglist = io->io_hdr.remote_sglist;
11966 	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
11967 	local_used = 0;
11968 	remote_used = 0;
11969 	total_used = 0;
11970 
11971 	if (io->io_hdr.flags & CTL_FLAG_REDIR_DONE) {
11972 		rq->ret = CTL_HA_STATUS_SUCCESS;
11973 		rq->context = io;
11974 		callback(rq);
11975 		goto bailout;
11976 	}
11977 
11978 	/*
11979 	 * Pull/push the data over the wire from/to the other controller.
11980 	 * This takes into account the possibility that the local and
11981 	 * remote sglists may not be identical in terms of the size of
11982 	 * the elements and the number of elements.
11983 	 *
11984 	 * One fundamental assumption here is that the length allocated for
11985 	 * both the local and remote sglists is identical.  Otherwise, we've
11986 	 * essentially got a coding error of some sort.
11987 	 */
11988 	for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
11989 		int isc_ret;
11990 		uint32_t cur_len, dma_length;
11991 		uint8_t *tmp_ptr;
11992 
11993 		rq->id = CTL_HA_DATA_CTL;
11994 		rq->command = command;
11995 		rq->context = io;
11996 
11997 		/*
11998 		 * Both pointers should be aligned.  But it is possible
11999 		 * that the allocation length is not.  They should both
12000 		 * also have enough slack left over at the end, though,
12001 		 * to round up to the next 8 byte boundary.
12002 		 */
12003 		cur_len = ctl_min(local_sglist[i].len - local_used,
12004 				  remote_sglist[j].len - remote_used);
12005 
12006 		/*
12007 		 * In this case, we have a size issue and need to decrease
12008 		 * the size, except in the case where we actually have less
12009 		 * than 8 bytes left.  In that case, we need to increase
12010 		 * the DMA length to get the last bit.
12011 		 */
12012 		if ((cur_len & 0x7) != 0) {
12013 			if (cur_len > 0x7) {
12014 				cur_len = cur_len - (cur_len & 0x7);
12015 				dma_length = cur_len;
12016 			} else {
12017 				CTL_SIZE_8B(dma_length, cur_len);
12018 			}
12019 
12020 		} else
12021 			dma_length = cur_len;
12022 
12023 		/*
12024 		 * If we had to allocate memory for this I/O, instead of using
12025 		 * the non-cached mirror memory, we'll need to flush the cache
12026 		 * before trying to DMA to the other controller.
12027 		 *
12028 		 * We could end up doing this multiple times for the same
12029 		 * segment if we have a larger local segment than remote
12030 		 * segment.  That shouldn't be an issue.
12031 		 */
12032 		if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12033 			/*
12034 			 * XXX KDM use bus_dmamap_sync() here.
12035 			 */
12036 		}
12037 
12038 		rq->size = dma_length;
12039 
12040 		tmp_ptr = (uint8_t *)local_sglist[i].addr;
12041 		tmp_ptr += local_used;
12042 
12043 		/* Use physical addresses when talking to ISC hardware */
12044 		if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
12045 			/* XXX KDM use busdma */
12046 #if 0
12047 			rq->local = vtophys(tmp_ptr);
12048 #endif
12049 		} else
12050 			rq->local = tmp_ptr;
12051 
12052 		tmp_ptr = (uint8_t *)remote_sglist[j].addr;
12053 		tmp_ptr += remote_used;
12054 		rq->remote = tmp_ptr;
12055 
12056 		rq->callback = NULL;
12057 
12058 		local_used += cur_len;
12059 		if (local_used >= local_sglist[i].len) {
12060 			i++;
12061 			local_used = 0;
12062 		}
12063 
12064 		remote_used += cur_len;
12065 		if (remote_used >= remote_sglist[j].len) {
12066 			j++;
12067 			remote_used = 0;
12068 		}
12069 		total_used += cur_len;
12070 
12071 		if (total_used >= io->scsiio.kern_data_len)
12072 			rq->callback = callback;
12073 
12074 		if ((rq->size & 0x7) != 0) {
12075 			printf("%s: warning: size %d is not on 8b boundary\n",
12076 			       __func__, rq->size);
12077 		}
12078 		if (((uintptr_t)rq->local & 0x7) != 0) {
12079 			printf("%s: warning: local %p not on 8b boundary\n",
12080 			       __func__, rq->local);
12081 		}
12082 		if (((uintptr_t)rq->remote & 0x7) != 0) {
12083 			printf("%s: warning: remote %p not on 8b boundary\n",
12084 			       __func__, rq->local);
12085 		}
12086 #if 0
12087 		printf("%s: %s: local %#x remote %#x size %d\n", __func__,
12088 		       (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
12089 		       rq->local, rq->remote, rq->size);
12090 #endif
12091 
12092 		isc_ret = ctl_dt_single(rq);
12093 		if (isc_ret == CTL_HA_STATUS_WAIT)
12094 			continue;
12095 
12096 		if (isc_ret == CTL_HA_STATUS_DISCONNECT) {
12097 			rq->ret = CTL_HA_STATUS_SUCCESS;
12098 		} else {
12099 			rq->ret = isc_ret;
12100 		}
12101 		callback(rq);
12102 		goto bailout;
12103 	}
12104 
12105 bailout:
12106 	return (retval);
12107 
12108 }
12109 
12110 static void
12111 ctl_datamove_remote_read(union ctl_io *io)
12112 {
12113 	int retval;
12114 	int i;
12115 
12116 	/*
12117 	 * This will send an error to the other controller in the case of a
12118 	 * failure.
12119 	 */
12120 	retval = ctl_datamove_remote_sgl_setup(io);
12121 	if (retval != 0)
12122 		return;
12123 
12124 	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
12125 					  ctl_datamove_remote_read_cb);
12126 	if ((retval != 0)
12127 	 && ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0)) {
12128 		/*
12129 		 * Make sure we free memory if there was an error..  The
12130 		 * ctl_datamove_remote_xfer() function will send the
12131 		 * datamove done message, or call the callback with an
12132 		 * error if there is a problem.
12133 		 */
12134 		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12135 			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12136 	}
12137 
12138 	return;
12139 }
12140 
12141 /*
12142  * Process a datamove request from the other controller.  This is used for
12143  * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
12144  * first.  Once that is complete, the data gets DMAed into the remote
12145  * controller's memory.  For reads, we DMA from the remote controller's
12146  * memory into our memory first, and then move it out to the FETD.
12147  *
12148  * Should be called without the ctl_lock held.
12149  */
12150 static void
12151 ctl_datamove_remote(union ctl_io *io)
12152 {
12153 	struct ctl_softc *softc;
12154 
12155 	softc = control_softc;
12156 
12157 	/*
12158 	 * Note that we look for an aborted I/O here, but don't do some of
12159 	 * the other checks that ctl_datamove() normally does.  We don't
12160 	 * need to run the task queue, because this I/O is on the ISC
12161 	 * queue, which is executed by the work thread after the task queue.
12162 	 * We don't need to run the datamove delay code, since that should
12163 	 * have been done if need be on the other controller.
12164 	 */
12165 	mtx_lock(&softc->ctl_lock);
12166 
12167 	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12168 
12169 		printf("%s: tag 0x%04x on (%d:%d:%d:%d) aborted\n", __func__,
12170 		       io->scsiio.tag_num, io->io_hdr.nexus.initid.id,
12171 		       io->io_hdr.nexus.targ_port,
12172 		       io->io_hdr.nexus.targ_target.id,
12173 		       io->io_hdr.nexus.targ_lun);
12174 		io->io_hdr.status = CTL_CMD_ABORTED;
12175 		io->io_hdr.port_status = 31338;
12176 
12177 		mtx_unlock(&softc->ctl_lock);
12178 
12179 		ctl_send_datamove_done(io, /*have_lock*/ 0);
12180 
12181 		return;
12182 	}
12183 
12184 	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT) {
12185 		mtx_unlock(&softc->ctl_lock);
12186 		ctl_datamove_remote_write(io);
12187 	} else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN){
12188 		mtx_unlock(&softc->ctl_lock);
12189 		ctl_datamove_remote_read(io);
12190 	} else {
12191 		union ctl_ha_msg msg;
12192 		struct scsi_sense_data *sense;
12193 		uint8_t sks[3];
12194 		int retry_count;
12195 
12196 		memset(&msg, 0, sizeof(msg));
12197 
12198 		msg.hdr.msg_type = CTL_MSG_BAD_JUJU;
12199 		msg.hdr.status = CTL_SCSI_ERROR;
12200 		msg.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
12201 
12202 		retry_count = 4243;
12203 
12204 		sense = &msg.scsi.sense_data;
12205 		sks[0] = SSD_SCS_VALID;
12206 		sks[1] = (retry_count >> 8) & 0xff;
12207 		sks[2] = retry_count & 0xff;
12208 
12209 		/* "Internal target failure" */
12210 		scsi_set_sense_data(sense,
12211 				    /*sense_format*/ SSD_TYPE_NONE,
12212 				    /*current_error*/ 1,
12213 				    /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
12214 				    /*asc*/ 0x44,
12215 				    /*ascq*/ 0x00,
12216 				    /*type*/ SSD_ELEM_SKS,
12217 				    /*size*/ sizeof(sks),
12218 				    /*data*/ sks,
12219 				    SSD_ELEM_NONE);
12220 
12221 		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12222 		if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12223 			ctl_failover_io(io, /*have_lock*/ 1);
12224 			mtx_unlock(&softc->ctl_lock);
12225 			return;
12226 		}
12227 
12228 		mtx_unlock(&softc->ctl_lock);
12229 
12230 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0) >
12231 		    CTL_HA_STATUS_SUCCESS) {
12232 			/* XXX KDM what to do if this fails? */
12233 		}
12234 		return;
12235 	}
12236 
12237 }
12238 
12239 static int
12240 ctl_process_done(union ctl_io *io, int have_lock)
12241 {
12242 	struct ctl_lun *lun;
12243 	struct ctl_softc *ctl_softc;
12244 	void (*fe_done)(union ctl_io *io);
12245 	uint32_t targ_port = ctl_port_idx(io->io_hdr.nexus.targ_port);
12246 
12247 	CTL_DEBUG_PRINT(("ctl_process_done\n"));
12248 
12249 	fe_done =
12250 	    control_softc->ctl_ports[targ_port]->fe_done;
12251 
12252 #ifdef CTL_TIME_IO
12253 	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12254 		char str[256];
12255 		char path_str[64];
12256 		struct sbuf sb;
12257 
12258 		ctl_scsi_path_string(io, path_str, sizeof(path_str));
12259 		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12260 
12261 		sbuf_cat(&sb, path_str);
12262 		switch (io->io_hdr.io_type) {
12263 		case CTL_IO_SCSI:
12264 			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12265 			sbuf_printf(&sb, "\n");
12266 			sbuf_cat(&sb, path_str);
12267 			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12268 				    io->scsiio.tag_num, io->scsiio.tag_type);
12269 			break;
12270 		case CTL_IO_TASK:
12271 			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12272 				    "Tag Type: %d\n", io->taskio.task_action,
12273 				    io->taskio.tag_num, io->taskio.tag_type);
12274 			break;
12275 		default:
12276 			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12277 			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12278 			break;
12279 		}
12280 		sbuf_cat(&sb, path_str);
12281 		sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
12282 			    (intmax_t)time_uptime - io->io_hdr.start_time);
12283 		sbuf_finish(&sb);
12284 		printf("%s", sbuf_data(&sb));
12285 	}
12286 #endif /* CTL_TIME_IO */
12287 
12288 	switch (io->io_hdr.io_type) {
12289 	case CTL_IO_SCSI:
12290 		break;
12291 	case CTL_IO_TASK:
12292 		ctl_io_error_print(io, NULL);
12293 		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
12294 			ctl_free_io_internal(io, /*have_lock*/ 0);
12295 		else
12296 			fe_done(io);
12297 		return (CTL_RETVAL_COMPLETE);
12298 		break;
12299 	default:
12300 		printf("ctl_process_done: invalid io type %d\n",
12301 		       io->io_hdr.io_type);
12302 		panic("ctl_process_done: invalid io type %d\n",
12303 		      io->io_hdr.io_type);
12304 		break; /* NOTREACHED */
12305 	}
12306 
12307 	lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12308 	if (lun == NULL) {
12309 		CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
12310 				 io->io_hdr.nexus.targ_lun));
12311 		fe_done(io);
12312 		goto bailout;
12313 	}
12314 	ctl_softc = lun->ctl_softc;
12315 
12316 	/*
12317 	 * Remove this from the OOA queue.
12318 	 */
12319 	if (have_lock == 0)
12320 		mtx_lock(&ctl_softc->ctl_lock);
12321 
12322 	/*
12323 	 * Check to see if we have any errors to inject here.  We only
12324 	 * inject errors for commands that don't already have errors set.
12325 	 */
12326 	if ((STAILQ_FIRST(&lun->error_list) != NULL)
12327 	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))
12328 		ctl_inject_error(lun, io);
12329 
12330 	/*
12331 	 * XXX KDM how do we treat commands that aren't completed
12332 	 * successfully?
12333 	 *
12334 	 * XXX KDM should we also track I/O latency?
12335 	 */
12336 	if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) {
12337 		uint32_t blocksize;
12338 #ifdef CTL_TIME_IO
12339 		struct bintime cur_bt;
12340 #endif
12341 
12342 		if ((lun->be_lun != NULL)
12343 		 && (lun->be_lun->blocksize != 0))
12344 			blocksize = lun->be_lun->blocksize;
12345 		else
12346 			blocksize = 512;
12347 
12348 		switch (io->io_hdr.io_type) {
12349 		case CTL_IO_SCSI: {
12350 			int isread;
12351 			struct ctl_lba_len lbalen;
12352 
12353 			isread = 0;
12354 			switch (io->scsiio.cdb[0]) {
12355 			case READ_6:
12356 			case READ_10:
12357 			case READ_12:
12358 			case READ_16:
12359 				isread = 1;
12360 				/* FALLTHROUGH */
12361 			case WRITE_6:
12362 			case WRITE_10:
12363 			case WRITE_12:
12364 			case WRITE_16:
12365 			case WRITE_VERIFY_10:
12366 			case WRITE_VERIFY_12:
12367 			case WRITE_VERIFY_16:
12368 				memcpy(&lbalen, io->io_hdr.ctl_private[
12369 				       CTL_PRIV_LBA_LEN].bytes, sizeof(lbalen));
12370 
12371 				if (isread) {
12372 					lun->stats.ports[targ_port].bytes[CTL_STATS_READ] +=
12373 						lbalen.len * blocksize;
12374 					lun->stats.ports[targ_port].operations[CTL_STATS_READ]++;
12375 
12376 #ifdef CTL_TIME_IO
12377 					bintime_add(
12378 					   &lun->stats.ports[targ_port].dma_time[CTL_STATS_READ],
12379 					   &io->io_hdr.dma_bt);
12380 					lun->stats.ports[targ_port].num_dmas[CTL_STATS_READ] +=
12381 						io->io_hdr.num_dmas;
12382 					getbintime(&cur_bt);
12383 					bintime_sub(&cur_bt,
12384 						    &io->io_hdr.start_bt);
12385 
12386 					bintime_add(
12387 					    &lun->stats.ports[targ_port].time[CTL_STATS_READ],
12388 					    &cur_bt);
12389 
12390 #if 0
12391 					cs_prof_gettime(&cur_ticks);
12392 					lun->stats.time[CTL_STATS_READ] +=
12393 						cur_ticks -
12394 						io->io_hdr.start_ticks;
12395 #endif
12396 #if 0
12397 					lun->stats.time[CTL_STATS_READ] +=
12398 						jiffies - io->io_hdr.start_time;
12399 #endif
12400 #endif /* CTL_TIME_IO */
12401 				} else {
12402 					lun->stats.ports[targ_port].bytes[CTL_STATS_WRITE] +=
12403 						lbalen.len * blocksize;
12404 					lun->stats.ports[targ_port].operations[
12405 						CTL_STATS_WRITE]++;
12406 
12407 #ifdef CTL_TIME_IO
12408 					bintime_add(
12409 					  &lun->stats.ports[targ_port].dma_time[CTL_STATS_WRITE],
12410 					  &io->io_hdr.dma_bt);
12411 					lun->stats.ports[targ_port].num_dmas[CTL_STATS_WRITE] +=
12412 						io->io_hdr.num_dmas;
12413 					getbintime(&cur_bt);
12414 					bintime_sub(&cur_bt,
12415 						    &io->io_hdr.start_bt);
12416 
12417 					bintime_add(
12418 					    &lun->stats.ports[targ_port].time[CTL_STATS_WRITE],
12419 					    &cur_bt);
12420 #if 0
12421 					cs_prof_gettime(&cur_ticks);
12422 					lun->stats.ports[targ_port].time[CTL_STATS_WRITE] +=
12423 						cur_ticks -
12424 						io->io_hdr.start_ticks;
12425 					lun->stats.ports[targ_port].time[CTL_STATS_WRITE] +=
12426 						jiffies - io->io_hdr.start_time;
12427 #endif
12428 #endif /* CTL_TIME_IO */
12429 				}
12430 				break;
12431 			default:
12432 				lun->stats.ports[targ_port].operations[CTL_STATS_NO_IO]++;
12433 
12434 #ifdef CTL_TIME_IO
12435 				bintime_add(
12436 				  &lun->stats.ports[targ_port].dma_time[CTL_STATS_NO_IO],
12437 				  &io->io_hdr.dma_bt);
12438 				lun->stats.ports[targ_port].num_dmas[CTL_STATS_NO_IO] +=
12439 					io->io_hdr.num_dmas;
12440 				getbintime(&cur_bt);
12441 				bintime_sub(&cur_bt, &io->io_hdr.start_bt);
12442 
12443 				bintime_add(&lun->stats.ports[targ_port].time[CTL_STATS_NO_IO],
12444 					    &cur_bt);
12445 
12446 #if 0
12447 				cs_prof_gettime(&cur_ticks);
12448 				lun->stats.ports[targ_port].time[CTL_STATS_NO_IO] +=
12449 					cur_ticks -
12450 					io->io_hdr.start_ticks;
12451 				lun->stats.ports[targ_port].time[CTL_STATS_NO_IO] +=
12452 					jiffies - io->io_hdr.start_time;
12453 #endif
12454 #endif /* CTL_TIME_IO */
12455 				break;
12456 			}
12457 			break;
12458 		}
12459 		default:
12460 			break;
12461 		}
12462 	}
12463 
12464 	TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
12465 
12466 	/*
12467 	 * Run through the blocked queue on this LUN and see if anything
12468 	 * has become unblocked, now that this transaction is done.
12469 	 */
12470 	ctl_check_blocked(lun);
12471 
12472 	/*
12473 	 * If the LUN has been invalidated, free it if there is nothing
12474 	 * left on its OOA queue.
12475 	 */
12476 	if ((lun->flags & CTL_LUN_INVALID)
12477 	 && (TAILQ_FIRST(&lun->ooa_queue) == NULL))
12478 		ctl_free_lun(lun);
12479 
12480 	/*
12481 	 * If this command has been aborted, make sure we set the status
12482 	 * properly.  The FETD is responsible for freeing the I/O and doing
12483 	 * whatever it needs to do to clean up its state.
12484 	 */
12485 	if (io->io_hdr.flags & CTL_FLAG_ABORT)
12486 		io->io_hdr.status = CTL_CMD_ABORTED;
12487 
12488 	/*
12489 	 * We print out status for every task management command.  For SCSI
12490 	 * commands, we filter out any unit attention errors; they happen
12491 	 * on every boot, and would clutter up the log.  Note:  task
12492 	 * management commands aren't printed here, they are printed above,
12493 	 * since they should never even make it down here.
12494 	 */
12495 	switch (io->io_hdr.io_type) {
12496 	case CTL_IO_SCSI: {
12497 		int error_code, sense_key, asc, ascq;
12498 
12499 		sense_key = 0;
12500 
12501 		if (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SCSI_ERROR)
12502 		 && (io->scsiio.scsi_status == SCSI_STATUS_CHECK_COND)) {
12503 			/*
12504 			 * Since this is just for printing, no need to
12505 			 * show errors here.
12506 			 */
12507 			scsi_extract_sense_len(&io->scsiio.sense_data,
12508 					       io->scsiio.sense_len,
12509 					       &error_code,
12510 					       &sense_key,
12511 					       &asc,
12512 					       &ascq,
12513 					       /*show_errors*/ 0);
12514 		}
12515 
12516 		if (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
12517 		 && (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SCSI_ERROR)
12518 		  || (io->scsiio.scsi_status != SCSI_STATUS_CHECK_COND)
12519 		  || (sense_key != SSD_KEY_UNIT_ATTENTION))) {
12520 
12521 			if ((time_uptime - ctl_softc->last_print_jiffies) <= 0){
12522 				ctl_softc->skipped_prints++;
12523 				if (have_lock == 0)
12524 					mtx_unlock(&ctl_softc->ctl_lock);
12525 			} else {
12526 				uint32_t skipped_prints;
12527 
12528 				skipped_prints = ctl_softc->skipped_prints;
12529 
12530 				ctl_softc->skipped_prints = 0;
12531 				ctl_softc->last_print_jiffies = time_uptime;
12532 
12533 				if (have_lock == 0)
12534 					mtx_unlock(&ctl_softc->ctl_lock);
12535 				if (skipped_prints > 0) {
12536 #ifdef NEEDTOPORT
12537 					csevent_log(CSC_CTL | CSC_SHELF_SW |
12538 					    CTL_ERROR_REPORT,
12539 					    csevent_LogType_Trace,
12540 					    csevent_Severity_Information,
12541 					    csevent_AlertLevel_Green,
12542 					    csevent_FRU_Firmware,
12543 					    csevent_FRU_Unknown,
12544 					    "High CTL error volume, %d prints "
12545 					    "skipped", skipped_prints);
12546 #endif
12547 				}
12548 				ctl_io_error_print(io, NULL);
12549 			}
12550 		} else {
12551 			if (have_lock == 0)
12552 				mtx_unlock(&ctl_softc->ctl_lock);
12553 		}
12554 		break;
12555 	}
12556 	case CTL_IO_TASK:
12557 		if (have_lock == 0)
12558 			mtx_unlock(&ctl_softc->ctl_lock);
12559 		ctl_io_error_print(io, NULL);
12560 		break;
12561 	default:
12562 		if (have_lock == 0)
12563 			mtx_unlock(&ctl_softc->ctl_lock);
12564 		break;
12565 	}
12566 
12567 	/*
12568 	 * Tell the FETD or the other shelf controller we're done with this
12569 	 * command.  Note that only SCSI commands get to this point.  Task
12570 	 * management commands are completed above.
12571 	 *
12572 	 * We only send status to the other controller if we're in XFER
12573 	 * mode.  In SER_ONLY mode, the I/O is done on the controller that
12574 	 * received the I/O (from CTL's perspective), and so the status is
12575 	 * generated there.
12576 	 *
12577 	 * XXX KDM if we hold the lock here, we could cause a deadlock
12578 	 * if the frontend comes back in in this context to queue
12579 	 * something.
12580 	 */
12581 	if ((ctl_softc->ha_mode == CTL_HA_MODE_XFER)
12582 	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12583 		union ctl_ha_msg msg;
12584 
12585 		memset(&msg, 0, sizeof(msg));
12586 		msg.hdr.msg_type = CTL_MSG_FINISH_IO;
12587 		msg.hdr.original_sc = io->io_hdr.original_sc;
12588 		msg.hdr.nexus = io->io_hdr.nexus;
12589 		msg.hdr.status = io->io_hdr.status;
12590 		msg.scsi.scsi_status = io->scsiio.scsi_status;
12591 		msg.scsi.tag_num = io->scsiio.tag_num;
12592 		msg.scsi.tag_type = io->scsiio.tag_type;
12593 		msg.scsi.sense_len = io->scsiio.sense_len;
12594 		msg.scsi.sense_residual = io->scsiio.sense_residual;
12595 		msg.scsi.residual = io->scsiio.residual;
12596 		memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12597 		       sizeof(io->scsiio.sense_data));
12598 		/*
12599 		 * We copy this whether or not this is an I/O-related
12600 		 * command.  Otherwise, we'd have to go and check to see
12601 		 * whether it's a read/write command, and it really isn't
12602 		 * worth it.
12603 		 */
12604 		memcpy(&msg.scsi.lbalen,
12605 		       &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
12606 		       sizeof(msg.scsi.lbalen));;
12607 
12608 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12609 				sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
12610 			/* XXX do something here */
12611 		}
12612 
12613 		ctl_free_io_internal(io, /*have_lock*/ 0);
12614 	} else
12615 		fe_done(io);
12616 
12617 bailout:
12618 
12619 	return (CTL_RETVAL_COMPLETE);
12620 }
12621 
12622 /*
12623  * Front end should call this if it doesn't do autosense.  When the request
12624  * sense comes back in from the initiator, we'll dequeue this and send it.
12625  */
12626 int
12627 ctl_queue_sense(union ctl_io *io)
12628 {
12629 	struct ctl_lun *lun;
12630 	struct ctl_softc *ctl_softc;
12631 	uint32_t initidx;
12632 
12633 	ctl_softc = control_softc;
12634 
12635 	CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
12636 
12637 	/*
12638 	 * LUN lookup will likely move to the ctl_work_thread() once we
12639 	 * have our new queueing infrastructure (that doesn't put things on
12640 	 * a per-LUN queue initially).  That is so that we can handle
12641 	 * things like an INQUIRY to a LUN that we don't have enabled.  We
12642 	 * can't deal with that right now.
12643 	 */
12644 	mtx_lock(&ctl_softc->ctl_lock);
12645 
12646 	/*
12647 	 * If we don't have a LUN for this, just toss the sense
12648 	 * information.
12649 	 */
12650 	if ((io->io_hdr.nexus.targ_lun < CTL_MAX_LUNS)
12651 	 && (ctl_softc->ctl_luns[io->io_hdr.nexus.targ_lun] != NULL))
12652 		lun = ctl_softc->ctl_luns[io->io_hdr.nexus.targ_lun];
12653 	else
12654 		goto bailout;
12655 
12656 	initidx = ctl_get_initindex(&io->io_hdr.nexus);
12657 
12658 	/*
12659 	 * Already have CA set for this LUN...toss the sense information.
12660 	 */
12661 	if (ctl_is_set(lun->have_ca, initidx))
12662 		goto bailout;
12663 
12664 	memcpy(&lun->pending_sense[initidx].sense, &io->scsiio.sense_data,
12665 	       ctl_min(sizeof(lun->pending_sense[initidx].sense),
12666 	       sizeof(io->scsiio.sense_data)));
12667 	ctl_set_mask(lun->have_ca, initidx);
12668 
12669 bailout:
12670 	mtx_unlock(&ctl_softc->ctl_lock);
12671 
12672 	ctl_free_io(io);
12673 
12674 	return (CTL_RETVAL_COMPLETE);
12675 }
12676 
12677 /*
12678  * Primary command inlet from frontend ports.  All SCSI and task I/O
12679  * requests must go through this function.
12680  */
12681 int
12682 ctl_queue(union ctl_io *io)
12683 {
12684 	struct ctl_softc *ctl_softc;
12685 
12686 	CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
12687 
12688 	ctl_softc = control_softc;
12689 
12690 #ifdef CTL_TIME_IO
12691 	io->io_hdr.start_time = time_uptime;
12692 	getbintime(&io->io_hdr.start_bt);
12693 #endif /* CTL_TIME_IO */
12694 
12695 	mtx_lock(&ctl_softc->ctl_lock);
12696 
12697 	switch (io->io_hdr.io_type) {
12698 	case CTL_IO_SCSI:
12699 		STAILQ_INSERT_TAIL(&ctl_softc->incoming_queue, &io->io_hdr,
12700 				   links);
12701 		break;
12702 	case CTL_IO_TASK:
12703 		STAILQ_INSERT_TAIL(&ctl_softc->task_queue, &io->io_hdr, links);
12704 		/*
12705 		 * Set the task pending flag.  This is necessary to close a
12706 		 * race condition with the FETD:
12707 		 *
12708 		 * - FETD submits a task management command, like an abort.
12709 		 * - Back end calls fe_datamove() to move the data for the
12710 		 *   aborted command.  The FETD can't really accept it, but
12711 		 *   if it did, it would end up transmitting data for a
12712 		 *   command that the initiator told us to abort.
12713 		 *
12714 		 * We close the race condition by setting the flag here,
12715 		 * and checking it in ctl_datamove(), before calling the
12716 		 * FETD's fe_datamove routine.  If we've got a task
12717 		 * pending, we run the task queue and then check to see
12718 		 * whether our particular I/O has been aborted.
12719 		 */
12720 		ctl_softc->flags |= CTL_FLAG_TASK_PENDING;
12721 		break;
12722 	default:
12723 		mtx_unlock(&ctl_softc->ctl_lock);
12724 		printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
12725 		return (-EINVAL);
12726 		break; /* NOTREACHED */
12727 	}
12728 	mtx_unlock(&ctl_softc->ctl_lock);
12729 
12730 	ctl_wakeup_thread();
12731 
12732 	return (CTL_RETVAL_COMPLETE);
12733 }
12734 
12735 #ifdef CTL_IO_DELAY
12736 static void
12737 ctl_done_timer_wakeup(void *arg)
12738 {
12739 	union ctl_io *io;
12740 
12741 	io = (union ctl_io *)arg;
12742 	ctl_done_lock(io, /*have_lock*/ 0);
12743 }
12744 #endif /* CTL_IO_DELAY */
12745 
12746 void
12747 ctl_done_lock(union ctl_io *io, int have_lock)
12748 {
12749 	struct ctl_softc *ctl_softc;
12750 #ifndef CTL_DONE_THREAD
12751 	union ctl_io *xio;
12752 #endif /* !CTL_DONE_THREAD */
12753 
12754 	ctl_softc = control_softc;
12755 
12756 	if (have_lock == 0)
12757 		mtx_lock(&ctl_softc->ctl_lock);
12758 
12759 	/*
12760 	 * Enable this to catch duplicate completion issues.
12761 	 */
12762 #if 0
12763 	if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
12764 		printf("%s: type %d msg %d cdb %x iptl: "
12765 		       "%d:%d:%d:%d tag 0x%04x "
12766 		       "flag %#x status %x\n",
12767 			__func__,
12768 			io->io_hdr.io_type,
12769 			io->io_hdr.msg_type,
12770 			io->scsiio.cdb[0],
12771 			io->io_hdr.nexus.initid.id,
12772 			io->io_hdr.nexus.targ_port,
12773 			io->io_hdr.nexus.targ_target.id,
12774 			io->io_hdr.nexus.targ_lun,
12775 			(io->io_hdr.io_type ==
12776 			CTL_IO_TASK) ?
12777 			io->taskio.tag_num :
12778 			io->scsiio.tag_num,
12779 		        io->io_hdr.flags,
12780 			io->io_hdr.status);
12781 	} else
12782 		io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
12783 #endif
12784 
12785 	/*
12786 	 * This is an internal copy of an I/O, and should not go through
12787 	 * the normal done processing logic.
12788 	 */
12789 	if (io->io_hdr.flags & CTL_FLAG_INT_COPY) {
12790 		if (have_lock == 0)
12791 			mtx_unlock(&ctl_softc->ctl_lock);
12792 		return;
12793 	}
12794 
12795 	/*
12796 	 * We need to send a msg to the serializing shelf to finish the IO
12797 	 * as well.  We don't send a finish message to the other shelf if
12798 	 * this is a task management command.  Task management commands
12799 	 * aren't serialized in the OOA queue, but rather just executed on
12800 	 * both shelf controllers for commands that originated on that
12801 	 * controller.
12802 	 */
12803 	if ((io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)
12804 	 && (io->io_hdr.io_type != CTL_IO_TASK)) {
12805 		union ctl_ha_msg msg_io;
12806 
12807 		msg_io.hdr.msg_type = CTL_MSG_FINISH_IO;
12808 		msg_io.hdr.serializing_sc = io->io_hdr.serializing_sc;
12809 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_io,
12810 		    sizeof(msg_io), 0 ) != CTL_HA_STATUS_SUCCESS) {
12811 		}
12812 		/* continue on to finish IO */
12813 	}
12814 #ifdef CTL_IO_DELAY
12815 	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12816 		struct ctl_lun *lun;
12817 
12818 		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12819 
12820 		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12821 	} else {
12822 		struct ctl_lun *lun;
12823 
12824 		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12825 
12826 		if ((lun != NULL)
12827 		 && (lun->delay_info.done_delay > 0)) {
12828 			struct callout *callout;
12829 
12830 			callout = (struct callout *)&io->io_hdr.timer_bytes;
12831 			callout_init(callout, /*mpsafe*/ 1);
12832 			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12833 			callout_reset(callout,
12834 				      lun->delay_info.done_delay * hz,
12835 				      ctl_done_timer_wakeup, io);
12836 			if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
12837 				lun->delay_info.done_delay = 0;
12838 			if (have_lock == 0)
12839 				mtx_unlock(&ctl_softc->ctl_lock);
12840 			return;
12841 		}
12842 	}
12843 #endif /* CTL_IO_DELAY */
12844 
12845 	STAILQ_INSERT_TAIL(&ctl_softc->done_queue, &io->io_hdr, links);
12846 
12847 #ifdef CTL_DONE_THREAD
12848 	if (have_lock == 0)
12849 		mtx_unlock(&ctl_softc->ctl_lock);
12850 
12851 	ctl_wakeup_thread();
12852 #else /* CTL_DONE_THREAD */
12853 	for (xio = (union ctl_io *)STAILQ_FIRST(&ctl_softc->done_queue);
12854 	     xio != NULL;
12855 	     xio =(union ctl_io *)STAILQ_FIRST(&ctl_softc->done_queue)) {
12856 
12857 		STAILQ_REMOVE_HEAD(&ctl_softc->done_queue, links);
12858 
12859 		ctl_process_done(xio, /*have_lock*/ 1);
12860 	}
12861 	if (have_lock == 0)
12862 		mtx_unlock(&ctl_softc->ctl_lock);
12863 #endif /* CTL_DONE_THREAD */
12864 }
12865 
12866 void
12867 ctl_done(union ctl_io *io)
12868 {
12869 	ctl_done_lock(io, /*have_lock*/ 0);
12870 }
12871 
12872 int
12873 ctl_isc(struct ctl_scsiio *ctsio)
12874 {
12875 	struct ctl_lun *lun;
12876 	int retval;
12877 
12878 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12879 
12880 	CTL_DEBUG_PRINT(("ctl_isc: command: %02x\n", ctsio->cdb[0]));
12881 
12882 	CTL_DEBUG_PRINT(("ctl_isc: calling data_submit()\n"));
12883 
12884 	retval = lun->backend->data_submit((union ctl_io *)ctsio);
12885 
12886 	return (retval);
12887 }
12888 
12889 
12890 static void
12891 ctl_work_thread(void *arg)
12892 {
12893 	struct ctl_softc *softc;
12894 	union ctl_io *io;
12895 	struct ctl_be_lun *be_lun;
12896 	int retval;
12897 
12898 	CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
12899 
12900 	softc = (struct ctl_softc *)arg;
12901 	if (softc == NULL)
12902 		return;
12903 
12904 	mtx_lock(&softc->ctl_lock);
12905 	for (;;) {
12906 		retval = 0;
12907 
12908 		/*
12909 		 * We handle the queues in this order:
12910 		 * - task management
12911 		 * - ISC
12912 		 * - done queue (to free up resources, unblock other commands)
12913 		 * - RtR queue
12914 		 * - incoming queue
12915 		 *
12916 		 * If those queues are empty, we break out of the loop and
12917 		 * go to sleep.
12918 		 */
12919 		io = (union ctl_io *)STAILQ_FIRST(&softc->task_queue);
12920 		if (io != NULL) {
12921 			ctl_run_task_queue(softc);
12922 			continue;
12923 		}
12924 		io = (union ctl_io *)STAILQ_FIRST(&softc->isc_queue);
12925 		if (io != NULL) {
12926 			STAILQ_REMOVE_HEAD(&softc->isc_queue, links);
12927 			ctl_handle_isc(io);
12928 			continue;
12929 		}
12930 		io = (union ctl_io *)STAILQ_FIRST(&softc->done_queue);
12931 		if (io != NULL) {
12932 			STAILQ_REMOVE_HEAD(&softc->done_queue, links);
12933 			/* clear any blocked commands, call fe_done */
12934 			mtx_unlock(&softc->ctl_lock);
12935 			/*
12936 			 * XXX KDM
12937 			 * Call this without a lock for now.  This will
12938 			 * depend on whether there is any way the FETD can
12939 			 * sleep or deadlock if called with the CTL lock
12940 			 * held.
12941 			 */
12942 			retval = ctl_process_done(io, /*have_lock*/ 0);
12943 			mtx_lock(&softc->ctl_lock);
12944 			continue;
12945 		}
12946 		if (!ctl_pause_rtr) {
12947 			io = (union ctl_io *)STAILQ_FIRST(&softc->rtr_queue);
12948 			if (io != NULL) {
12949 				STAILQ_REMOVE_HEAD(&softc->rtr_queue, links);
12950 				mtx_unlock(&softc->ctl_lock);
12951 				goto execute;
12952 			}
12953 		}
12954 		io = (union ctl_io *)STAILQ_FIRST(&softc->incoming_queue);
12955 		if (io != NULL) {
12956 			STAILQ_REMOVE_HEAD(&softc->incoming_queue, links);
12957 			mtx_unlock(&softc->ctl_lock);
12958 			ctl_scsiio_precheck(softc, &io->scsiio);
12959 			mtx_lock(&softc->ctl_lock);
12960 			continue;
12961 		}
12962 		/*
12963 		 * We might want to move this to a separate thread, so that
12964 		 * configuration requests (in this case LUN creations)
12965 		 * won't impact the I/O path.
12966 		 */
12967 		be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
12968 		if (be_lun != NULL) {
12969 			STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
12970 			mtx_unlock(&softc->ctl_lock);
12971 			ctl_create_lun(be_lun);
12972 			mtx_lock(&softc->ctl_lock);
12973 			continue;
12974 		}
12975 
12976 		/* XXX KDM use the PDROP flag?? */
12977 		/* Sleep until we have something to do. */
12978 		mtx_sleep(softc, &softc->ctl_lock, PRIBIO, "ctl_work", 0);
12979 
12980 		/* Back to the top of the loop to see what woke us up. */
12981 		continue;
12982 
12983 execute:
12984 		retval = ctl_scsiio(&io->scsiio);
12985 		switch (retval) {
12986 		case CTL_RETVAL_COMPLETE:
12987 			break;
12988 		default:
12989 			/*
12990 			 * Probably need to make sure this doesn't happen.
12991 			 */
12992 			break;
12993 		}
12994 		mtx_lock(&softc->ctl_lock);
12995 	}
12996 }
12997 
12998 void
12999 ctl_wakeup_thread()
13000 {
13001 	struct ctl_softc *softc;
13002 
13003 	softc = control_softc;
13004 
13005 	wakeup(softc);
13006 }
13007 
13008 /* Initialization and failover */
13009 
13010 void
13011 ctl_init_isc_msg(void)
13012 {
13013 	printf("CTL: Still calling this thing\n");
13014 }
13015 
13016 /*
13017  * Init component
13018  * 	Initializes component into configuration defined by bootMode
13019  *	(see hasc-sv.c)
13020  *  	returns hasc_Status:
13021  * 		OK
13022  *		ERROR - fatal error
13023  */
13024 static ctl_ha_comp_status
13025 ctl_isc_init(struct ctl_ha_component *c)
13026 {
13027 	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
13028 
13029 	c->status = ret;
13030 	return ret;
13031 }
13032 
13033 /* Start component
13034  * 	Starts component in state requested. If component starts successfully,
13035  *	it must set its own state to the requestrd state
13036  *	When requested state is HASC_STATE_HA, the component may refine it
13037  * 	by adding _SLAVE or _MASTER flags.
13038  *	Currently allowed state transitions are:
13039  *	UNKNOWN->HA		- initial startup
13040  *	UNKNOWN->SINGLE - initial startup when no parter detected
13041  *	HA->SINGLE		- failover
13042  * returns ctl_ha_comp_status:
13043  * 		OK	- component successfully started in requested state
13044  *		FAILED  - could not start the requested state, failover may
13045  * 			  be possible
13046  *		ERROR	- fatal error detected, no future startup possible
13047  */
13048 static ctl_ha_comp_status
13049 ctl_isc_start(struct ctl_ha_component *c, ctl_ha_state state)
13050 {
13051 	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
13052 
13053 	// UNKNOWN->HA or UNKNOWN->SINGLE (bootstrap)
13054 	if (c->state == CTL_HA_STATE_UNKNOWN ) {
13055 		ctl_is_single = 0;
13056 		if (ctl_ha_msg_create(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
13057 		    != CTL_HA_STATUS_SUCCESS) {
13058 			printf("ctl_isc_start: ctl_ha_msg_create failed.\n");
13059 			ret = CTL_HA_COMP_STATUS_ERROR;
13060 		}
13061 	} else if (CTL_HA_STATE_IS_HA(c->state)
13062 		&& CTL_HA_STATE_IS_SINGLE(state)){
13063 		// HA->SINGLE transition
13064 	        ctl_failover();
13065 		ctl_is_single = 1;
13066 	} else {
13067 		printf("ctl_isc_start:Invalid state transition %X->%X\n",
13068 		       c->state, state);
13069 		ret = CTL_HA_COMP_STATUS_ERROR;
13070 	}
13071 	if (CTL_HA_STATE_IS_SINGLE(state))
13072 		ctl_is_single = 1;
13073 
13074 	c->state = state;
13075 	c->status = ret;
13076 	return ret;
13077 }
13078 
13079 /*
13080  * Quiesce component
13081  * The component must clear any error conditions (set status to OK) and
13082  * prepare itself to another Start call
13083  * returns ctl_ha_comp_status:
13084  * 	OK
13085  *	ERROR
13086  */
13087 static ctl_ha_comp_status
13088 ctl_isc_quiesce(struct ctl_ha_component *c)
13089 {
13090 	int ret = CTL_HA_COMP_STATUS_OK;
13091 
13092 	ctl_pause_rtr = 1;
13093 	c->status = ret;
13094 	return ret;
13095 }
13096 
13097 struct ctl_ha_component ctl_ha_component_ctlisc =
13098 {
13099 	.name = "CTL ISC",
13100 	.state = CTL_HA_STATE_UNKNOWN,
13101 	.init = ctl_isc_init,
13102 	.start = ctl_isc_start,
13103 	.quiesce = ctl_isc_quiesce
13104 };
13105 
13106 /*
13107  *  vim: ts=8
13108  */
13109