xref: /freebsd/sys/cam/ctl/ctl.c (revision c6ec7d31830ab1c80edae95ad5e4b9dba10c47ac)
1 /*-
2  * Copyright (c) 2003-2009 Silicon Graphics International Corp.
3  * Copyright (c) 2012 The FreeBSD Foundation
4  * All rights reserved.
5  *
6  * Portions of this software were developed by Edward Tomasz Napierala
7  * under sponsorship from the FreeBSD Foundation.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions, and the following disclaimer,
14  *    without modification.
15  * 2. Redistributions in binary form must reproduce at minimum a disclaimer
16  *    substantially similar to the "NO WARRANTY" disclaimer below
17  *    ("Disclaimer") and any redistribution must be conditioned upon
18  *    including a substantially similar Disclaimer requirement for further
19  *    binary redistribution.
20  *
21  * NO WARRANTY
22  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
23  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
24  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR
25  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
26  * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
30  * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
31  * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
32  * POSSIBILITY OF SUCH DAMAGES.
33  *
34  * $Id: //depot/users/kenm/FreeBSD-test2/sys/cam/ctl/ctl.c#8 $
35  */
36 /*
37  * CAM Target Layer, a SCSI device emulation subsystem.
38  *
39  * Author: Ken Merry <ken@FreeBSD.org>
40  */
41 
42 #define _CTL_C
43 
44 #include <sys/cdefs.h>
45 __FBSDID("$FreeBSD$");
46 
47 #include <sys/param.h>
48 #include <sys/systm.h>
49 #include <sys/kernel.h>
50 #include <sys/types.h>
51 #include <sys/kthread.h>
52 #include <sys/bio.h>
53 #include <sys/fcntl.h>
54 #include <sys/lock.h>
55 #include <sys/mutex.h>
56 #include <sys/condvar.h>
57 #include <sys/malloc.h>
58 #include <sys/conf.h>
59 #include <sys/ioccom.h>
60 #include <sys/queue.h>
61 #include <sys/sbuf.h>
62 #include <sys/endian.h>
63 #include <sys/sysctl.h>
64 
65 #include <cam/cam.h>
66 #include <cam/scsi/scsi_all.h>
67 #include <cam/scsi/scsi_da.h>
68 #include <cam/ctl/ctl_io.h>
69 #include <cam/ctl/ctl.h>
70 #include <cam/ctl/ctl_frontend.h>
71 #include <cam/ctl/ctl_frontend_internal.h>
72 #include <cam/ctl/ctl_util.h>
73 #include <cam/ctl/ctl_backend.h>
74 #include <cam/ctl/ctl_ioctl.h>
75 #include <cam/ctl/ctl_ha.h>
76 #include <cam/ctl/ctl_private.h>
77 #include <cam/ctl/ctl_debug.h>
78 #include <cam/ctl/ctl_scsi_all.h>
79 #include <cam/ctl/ctl_error.h>
80 
81 struct ctl_softc *control_softc = NULL;
82 
83 /*
84  * The default is to run with CTL_DONE_THREAD turned on.  Completed
85  * transactions are queued for processing by the CTL work thread.  When
86  * CTL_DONE_THREAD is not defined, completed transactions are processed in
87  * the caller's context.
88  */
89 #define CTL_DONE_THREAD
90 
91 /*
92  *  * Use the serial number and device ID provided by the backend, rather than
93  *   * making up our own.
94  *    */
95 #define CTL_USE_BACKEND_SN
96 
97 /*
98  * Size and alignment macros needed for Copan-specific HA hardware.  These
99  * can go away when the HA code is re-written, and uses busdma for any
100  * hardware.
101  */
102 #define	CTL_ALIGN_8B(target, source, type)				\
103 	if (((uint32_t)source & 0x7) != 0)				\
104 		target = (type)(source + (0x8 - ((uint32_t)source & 0x7)));\
105 	else								\
106 		target = (type)source;
107 
108 #define	CTL_SIZE_8B(target, size)					\
109 	if ((size & 0x7) != 0)						\
110 		target = size + (0x8 - (size & 0x7));			\
111 	else								\
112 		target = size;
113 
114 #define CTL_ALIGN_8B_MARGIN	16
115 
116 /*
117  * Template mode pages.
118  */
119 
120 /*
121  * Note that these are default values only.  The actual values will be
122  * filled in when the user does a mode sense.
123  */
124 static struct copan_power_subpage power_page_default = {
125 	/*page_code*/ PWR_PAGE_CODE | SMPH_SPF,
126 	/*subpage*/ PWR_SUBPAGE_CODE,
127 	/*page_length*/ {(sizeof(struct copan_power_subpage) - 4) & 0xff00,
128 			 (sizeof(struct copan_power_subpage) - 4) & 0x00ff},
129 	/*page_version*/ PWR_VERSION,
130 	/* total_luns */ 26,
131 	/* max_active_luns*/ PWR_DFLT_MAX_LUNS,
132 	/*reserved*/ {0, 0, 0, 0, 0, 0, 0, 0, 0,
133 		      0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
134 		      0, 0, 0, 0, 0, 0}
135 };
136 
137 static struct copan_power_subpage power_page_changeable = {
138 	/*page_code*/ PWR_PAGE_CODE | SMPH_SPF,
139 	/*subpage*/ PWR_SUBPAGE_CODE,
140 	/*page_length*/ {(sizeof(struct copan_power_subpage) - 4) & 0xff00,
141 			 (sizeof(struct copan_power_subpage) - 4) & 0x00ff},
142 	/*page_version*/ 0,
143 	/* total_luns */ 0,
144 	/* max_active_luns*/ 0,
145 	/*reserved*/ {0, 0, 0, 0, 0, 0, 0, 0, 0,
146 		      0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
147 		      0, 0, 0, 0, 0, 0}
148 };
149 
150 static struct copan_aps_subpage aps_page_default = {
151 	APS_PAGE_CODE | SMPH_SPF, //page_code
152 	APS_SUBPAGE_CODE, //subpage
153 	{(sizeof(struct copan_aps_subpage) - 4) & 0xff00,
154 	 (sizeof(struct copan_aps_subpage) - 4) & 0x00ff}, //page_length
155 	APS_VERSION, //page_version
156 	0, //lock_active
157 	{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
158 	0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
159 	0, 0, 0, 0, 0} //reserved
160 };
161 
162 static struct copan_aps_subpage aps_page_changeable = {
163 	APS_PAGE_CODE | SMPH_SPF, //page_code
164 	APS_SUBPAGE_CODE, //subpage
165 	{(sizeof(struct copan_aps_subpage) - 4) & 0xff00,
166 	 (sizeof(struct copan_aps_subpage) - 4) & 0x00ff}, //page_length
167 	0, //page_version
168 	0, //lock_active
169 	{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
170 	0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
171 	0, 0, 0, 0, 0} //reserved
172 };
173 
174 static struct copan_debugconf_subpage debugconf_page_default = {
175 	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
176 	DBGCNF_SUBPAGE_CODE,		/* subpage */
177 	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
178 	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
179 	DBGCNF_VERSION,			/* page_version */
180 	{CTL_TIME_IO_DEFAULT_SECS>>8,
181 	 CTL_TIME_IO_DEFAULT_SECS>>0},	/* ctl_time_io_secs */
182 };
183 
184 static struct copan_debugconf_subpage debugconf_page_changeable = {
185 	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
186 	DBGCNF_SUBPAGE_CODE,		/* subpage */
187 	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
188 	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
189 	0,				/* page_version */
190 	{0xff,0xff},			/* ctl_time_io_secs */
191 };
192 
193 static struct scsi_format_page format_page_default = {
194 	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
195 	/*page_length*/sizeof(struct scsi_format_page) - 2,
196 	/*tracks_per_zone*/ {0, 0},
197 	/*alt_sectors_per_zone*/ {0, 0},
198 	/*alt_tracks_per_zone*/ {0, 0},
199 	/*alt_tracks_per_lun*/ {0, 0},
200 	/*sectors_per_track*/ {(CTL_DEFAULT_SECTORS_PER_TRACK >> 8) & 0xff,
201 			        CTL_DEFAULT_SECTORS_PER_TRACK & 0xff},
202 	/*bytes_per_sector*/ {0, 0},
203 	/*interleave*/ {0, 0},
204 	/*track_skew*/ {0, 0},
205 	/*cylinder_skew*/ {0, 0},
206 	/*flags*/ SFP_HSEC,
207 	/*reserved*/ {0, 0, 0}
208 };
209 
210 static struct scsi_format_page format_page_changeable = {
211 	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
212 	/*page_length*/sizeof(struct scsi_format_page) - 2,
213 	/*tracks_per_zone*/ {0, 0},
214 	/*alt_sectors_per_zone*/ {0, 0},
215 	/*alt_tracks_per_zone*/ {0, 0},
216 	/*alt_tracks_per_lun*/ {0, 0},
217 	/*sectors_per_track*/ {0, 0},
218 	/*bytes_per_sector*/ {0, 0},
219 	/*interleave*/ {0, 0},
220 	/*track_skew*/ {0, 0},
221 	/*cylinder_skew*/ {0, 0},
222 	/*flags*/ 0,
223 	/*reserved*/ {0, 0, 0}
224 };
225 
226 static struct scsi_rigid_disk_page rigid_disk_page_default = {
227 	/*page_code*/SMS_RIGID_DISK_PAGE,
228 	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
229 	/*cylinders*/ {0, 0, 0},
230 	/*heads*/ CTL_DEFAULT_HEADS,
231 	/*start_write_precomp*/ {0, 0, 0},
232 	/*start_reduced_current*/ {0, 0, 0},
233 	/*step_rate*/ {0, 0},
234 	/*landing_zone_cylinder*/ {0, 0, 0},
235 	/*rpl*/ SRDP_RPL_DISABLED,
236 	/*rotational_offset*/ 0,
237 	/*reserved1*/ 0,
238 	/*rotation_rate*/ {(CTL_DEFAULT_ROTATION_RATE >> 8) & 0xff,
239 			   CTL_DEFAULT_ROTATION_RATE & 0xff},
240 	/*reserved2*/ {0, 0}
241 };
242 
243 static struct scsi_rigid_disk_page rigid_disk_page_changeable = {
244 	/*page_code*/SMS_RIGID_DISK_PAGE,
245 	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
246 	/*cylinders*/ {0, 0, 0},
247 	/*heads*/ 0,
248 	/*start_write_precomp*/ {0, 0, 0},
249 	/*start_reduced_current*/ {0, 0, 0},
250 	/*step_rate*/ {0, 0},
251 	/*landing_zone_cylinder*/ {0, 0, 0},
252 	/*rpl*/ 0,
253 	/*rotational_offset*/ 0,
254 	/*reserved1*/ 0,
255 	/*rotation_rate*/ {0, 0},
256 	/*reserved2*/ {0, 0}
257 };
258 
259 static struct scsi_caching_page caching_page_default = {
260 	/*page_code*/SMS_CACHING_PAGE,
261 	/*page_length*/sizeof(struct scsi_caching_page) - 2,
262 	/*flags1*/ SCP_DISC | SCP_WCE,
263 	/*ret_priority*/ 0,
264 	/*disable_pf_transfer_len*/ {0xff, 0xff},
265 	/*min_prefetch*/ {0, 0},
266 	/*max_prefetch*/ {0xff, 0xff},
267 	/*max_pf_ceiling*/ {0xff, 0xff},
268 	/*flags2*/ 0,
269 	/*cache_segments*/ 0,
270 	/*cache_seg_size*/ {0, 0},
271 	/*reserved*/ 0,
272 	/*non_cache_seg_size*/ {0, 0, 0}
273 };
274 
275 static struct scsi_caching_page caching_page_changeable = {
276 	/*page_code*/SMS_CACHING_PAGE,
277 	/*page_length*/sizeof(struct scsi_caching_page) - 2,
278 	/*flags1*/ 0,
279 	/*ret_priority*/ 0,
280 	/*disable_pf_transfer_len*/ {0, 0},
281 	/*min_prefetch*/ {0, 0},
282 	/*max_prefetch*/ {0, 0},
283 	/*max_pf_ceiling*/ {0, 0},
284 	/*flags2*/ 0,
285 	/*cache_segments*/ 0,
286 	/*cache_seg_size*/ {0, 0},
287 	/*reserved*/ 0,
288 	/*non_cache_seg_size*/ {0, 0, 0}
289 };
290 
291 static struct scsi_control_page control_page_default = {
292 	/*page_code*/SMS_CONTROL_MODE_PAGE,
293 	/*page_length*/sizeof(struct scsi_control_page) - 2,
294 	/*rlec*/0,
295 	/*queue_flags*/0,
296 	/*eca_and_aen*/0,
297 	/*reserved*/0,
298 	/*aen_holdoff_period*/{0, 0}
299 };
300 
301 static struct scsi_control_page control_page_changeable = {
302 	/*page_code*/SMS_CONTROL_MODE_PAGE,
303 	/*page_length*/sizeof(struct scsi_control_page) - 2,
304 	/*rlec*/SCP_DSENSE,
305 	/*queue_flags*/0,
306 	/*eca_and_aen*/0,
307 	/*reserved*/0,
308 	/*aen_holdoff_period*/{0, 0}
309 };
310 
311 
312 /*
313  * XXX KDM move these into the softc.
314  */
315 static int rcv_sync_msg;
316 static int persis_offset;
317 static uint8_t ctl_pause_rtr;
318 static int     ctl_is_single;
319 static int     index_to_aps_page;
320 int	   ctl_disable = 0;
321 
322 SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD, 0, "CAM Target Layer");
323 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, disable, CTLFLAG_RDTUN, &ctl_disable, 0,
324 	   "Disable CTL");
325 TUNABLE_INT("kern.cam.ctl.disable", &ctl_disable);
326 
327 /*
328  * Serial number (0x80), device id (0x83), and supported pages (0x00)
329  */
330 #define SCSI_EVPD_NUM_SUPPORTED_PAGES	3
331 
332 static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
333 				  int param);
334 static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
335 static void ctl_init(void);
336 void ctl_shutdown(void);
337 static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
338 static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
339 static void ctl_ioctl_online(void *arg);
340 static void ctl_ioctl_offline(void *arg);
341 static int ctl_ioctl_targ_enable(void *arg, struct ctl_id targ_id);
342 static int ctl_ioctl_targ_disable(void *arg, struct ctl_id targ_id);
343 static int ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id);
344 static int ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id);
345 static int ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio);
346 static int ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio, int have_lock);
347 static int ctl_ioctl_submit_wait(union ctl_io *io);
348 static void ctl_ioctl_datamove(union ctl_io *io);
349 static void ctl_ioctl_done(union ctl_io *io);
350 static void ctl_ioctl_hard_startstop_callback(void *arg,
351 					      struct cfi_metatask *metatask);
352 static void ctl_ioctl_bbrread_callback(void *arg,struct cfi_metatask *metatask);
353 static int ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
354 			      struct ctl_ooa *ooa_hdr,
355 			      struct ctl_ooa_entry *kern_entries);
356 static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
357 		     struct thread *td);
358 uint32_t ctl_get_resindex(struct ctl_nexus *nexus);
359 uint32_t ctl_port_idx(int port_num);
360 #ifdef unused
361 static union ctl_io *ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port,
362 				   uint32_t targ_target, uint32_t targ_lun,
363 				   int can_wait);
364 static void ctl_kfree_io(union ctl_io *io);
365 #endif /* unused */
366 static void ctl_free_io_internal(union ctl_io *io, int have_lock);
367 static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
368 			 struct ctl_be_lun *be_lun, struct ctl_id target_id);
369 static int ctl_free_lun(struct ctl_lun *lun);
370 static void ctl_create_lun(struct ctl_be_lun *be_lun);
371 /**
372 static void ctl_failover_change_pages(struct ctl_softc *softc,
373 				      struct ctl_scsiio *ctsio, int master);
374 **/
375 
376 static int ctl_do_mode_select(union ctl_io *io);
377 static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
378 			   uint64_t res_key, uint64_t sa_res_key,
379 			   uint8_t type, uint32_t residx,
380 			   struct ctl_scsiio *ctsio,
381 			   struct scsi_per_res_out *cdb,
382 			   struct scsi_per_res_out_parms* param);
383 static void ctl_pro_preempt_other(struct ctl_lun *lun,
384 				  union ctl_ha_msg *msg);
385 static void ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg);
386 static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
387 static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
388 static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
389 static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
390 static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
391 static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint32_t *len);
392 static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2);
393 static ctl_action ctl_check_for_blockage(union ctl_io *pending_io,
394 					 union ctl_io *ooa_io);
395 static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
396 				union ctl_io *starting_io);
397 static int ctl_check_blocked(struct ctl_lun *lun);
398 static int ctl_scsiio_lun_check(struct ctl_softc *ctl_softc,
399 				struct ctl_lun *lun,
400 				struct ctl_cmd_entry *entry,
401 				struct ctl_scsiio *ctsio);
402 //static int ctl_check_rtr(union ctl_io *pending_io, struct ctl_softc *softc);
403 static void ctl_failover(void);
404 static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
405 			       struct ctl_scsiio *ctsio);
406 static int ctl_scsiio(struct ctl_scsiio *ctsio);
407 
408 static int ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
409 static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
410 			    ctl_ua_type ua_type);
411 static int ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io,
412 			 ctl_ua_type ua_type);
413 static int ctl_abort_task(union ctl_io *io);
414 static void ctl_run_task_queue(struct ctl_softc *ctl_softc);
415 #ifdef CTL_IO_DELAY
416 static void ctl_datamove_timer_wakeup(void *arg);
417 static void ctl_done_timer_wakeup(void *arg);
418 #endif /* CTL_IO_DELAY */
419 
420 static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
421 static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
422 static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
423 static void ctl_datamove_remote_write(union ctl_io *io);
424 static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
425 static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
426 static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
427 static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
428 				    ctl_ha_dt_cb callback);
429 static void ctl_datamove_remote_read(union ctl_io *io);
430 static void ctl_datamove_remote(union ctl_io *io);
431 static int ctl_process_done(union ctl_io *io, int have_lock);
432 static void ctl_work_thread(void *arg);
433 
434 /*
435  * Load the serialization table.  This isn't very pretty, but is probably
436  * the easiest way to do it.
437  */
438 #include "ctl_ser_table.c"
439 
440 /*
441  * We only need to define open, close and ioctl routines for this driver.
442  */
443 static struct cdevsw ctl_cdevsw = {
444 	.d_version =	D_VERSION,
445 	.d_flags =	0,
446 	.d_open =	ctl_open,
447 	.d_close =	ctl_close,
448 	.d_ioctl =	ctl_ioctl,
449 	.d_name =	"ctl",
450 };
451 
452 
453 MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
454 
455 /*
456  * If we have the CAM SIM, we may or may not have another SIM that will
457  * cause CTL to get initialized.  If not, we need to initialize it.
458  */
459 SYSINIT(ctl_init, SI_SUB_CONFIGURE, SI_ORDER_THIRD, ctl_init, NULL);
460 
461 static void
462 ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
463 			    union ctl_ha_msg *msg_info)
464 {
465 	struct ctl_scsiio *ctsio;
466 
467 	if (msg_info->hdr.original_sc == NULL) {
468 		printf("%s: original_sc == NULL!\n", __func__);
469 		/* XXX KDM now what? */
470 		return;
471 	}
472 
473 	ctsio = &msg_info->hdr.original_sc->scsiio;
474 	ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
475 	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
476 	ctsio->io_hdr.status = msg_info->hdr.status;
477 	ctsio->scsi_status = msg_info->scsi.scsi_status;
478 	ctsio->sense_len = msg_info->scsi.sense_len;
479 	ctsio->sense_residual = msg_info->scsi.sense_residual;
480 	ctsio->residual = msg_info->scsi.residual;
481 	memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
482 	       sizeof(ctsio->sense_data));
483 	memcpy(&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
484 	       &msg_info->scsi.lbalen, sizeof(msg_info->scsi.lbalen));
485 	STAILQ_INSERT_TAIL(&ctl_softc->isc_queue, &ctsio->io_hdr, links);
486 	ctl_wakeup_thread();
487 }
488 
489 static void
490 ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
491 				union ctl_ha_msg *msg_info)
492 {
493 	struct ctl_scsiio *ctsio;
494 
495 	if (msg_info->hdr.serializing_sc == NULL) {
496 		printf("%s: serializing_sc == NULL!\n", __func__);
497 		/* XXX KDM now what? */
498 		return;
499 	}
500 
501 	ctsio = &msg_info->hdr.serializing_sc->scsiio;
502 #if 0
503 	/*
504 	 * Attempt to catch the situation where an I/O has
505 	 * been freed, and we're using it again.
506 	 */
507 	if (ctsio->io_hdr.io_type == 0xff) {
508 		union ctl_io *tmp_io;
509 		tmp_io = (union ctl_io *)ctsio;
510 		printf("%s: %p use after free!\n", __func__,
511 		       ctsio);
512 		printf("%s: type %d msg %d cdb %x iptl: "
513 		       "%d:%d:%d:%d tag 0x%04x "
514 		       "flag %#x status %x\n",
515 			__func__,
516 			tmp_io->io_hdr.io_type,
517 			tmp_io->io_hdr.msg_type,
518 			tmp_io->scsiio.cdb[0],
519 			tmp_io->io_hdr.nexus.initid.id,
520 			tmp_io->io_hdr.nexus.targ_port,
521 			tmp_io->io_hdr.nexus.targ_target.id,
522 			tmp_io->io_hdr.nexus.targ_lun,
523 			(tmp_io->io_hdr.io_type ==
524 			CTL_IO_TASK) ?
525 			tmp_io->taskio.tag_num :
526 			tmp_io->scsiio.tag_num,
527 		        tmp_io->io_hdr.flags,
528 			tmp_io->io_hdr.status);
529 	}
530 #endif
531 	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
532 	STAILQ_INSERT_TAIL(&ctl_softc->isc_queue, &ctsio->io_hdr, links);
533 	ctl_wakeup_thread();
534 }
535 
536 /*
537  * ISC (Inter Shelf Communication) event handler.  Events from the HA
538  * subsystem come in here.
539  */
540 static void
541 ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
542 {
543 	struct ctl_softc *ctl_softc;
544 	union ctl_io *io;
545 	struct ctl_prio *presio;
546 	ctl_ha_status isc_status;
547 
548 	ctl_softc = control_softc;
549 	io = NULL;
550 
551 
552 #if 0
553 	printf("CTL: Isc Msg event %d\n", event);
554 #endif
555 	if (event == CTL_HA_EVT_MSG_RECV) {
556 		union ctl_ha_msg msg_info;
557 
558 		isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
559 					     sizeof(msg_info), /*wait*/ 0);
560 #if 0
561 		printf("CTL: msg_type %d\n", msg_info.msg_type);
562 #endif
563 		if (isc_status != 0) {
564 			printf("Error receiving message, status = %d\n",
565 			       isc_status);
566 			return;
567 		}
568 		mtx_lock(&ctl_softc->ctl_lock);
569 
570 		switch (msg_info.hdr.msg_type) {
571 		case CTL_MSG_SERIALIZE:
572 #if 0
573 			printf("Serialize\n");
574 #endif
575 			io = ctl_alloc_io((void *)ctl_softc->othersc_pool);
576 			if (io == NULL) {
577 				printf("ctl_isc_event_handler: can't allocate "
578 				       "ctl_io!\n");
579 				/* Bad Juju */
580 				/* Need to set busy and send msg back */
581 				mtx_unlock(&ctl_softc->ctl_lock);
582 				msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
583 				msg_info.hdr.status = CTL_SCSI_ERROR;
584 				msg_info.scsi.scsi_status = SCSI_STATUS_BUSY;
585 				msg_info.scsi.sense_len = 0;
586 			        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
587 				    sizeof(msg_info), 0) > CTL_HA_STATUS_SUCCESS){
588 				}
589 				goto bailout;
590 			}
591 			ctl_zero_io(io);
592 			// populate ctsio from msg_info
593 			io->io_hdr.io_type = CTL_IO_SCSI;
594 			io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
595 			io->io_hdr.original_sc = msg_info.hdr.original_sc;
596 #if 0
597 			printf("pOrig %x\n", (int)msg_info.original_sc);
598 #endif
599 			io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
600 					    CTL_FLAG_IO_ACTIVE;
601 			/*
602 			 * If we're in serialization-only mode, we don't
603 			 * want to go through full done processing.  Thus
604 			 * the COPY flag.
605 			 *
606 			 * XXX KDM add another flag that is more specific.
607 			 */
608 			if (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)
609 				io->io_hdr.flags |= CTL_FLAG_INT_COPY;
610 			io->io_hdr.nexus = msg_info.hdr.nexus;
611 #if 0
612 			printf("targ %d, port %d, iid %d, lun %d\n",
613 			       io->io_hdr.nexus.targ_target.id,
614 			       io->io_hdr.nexus.targ_port,
615 			       io->io_hdr.nexus.initid.id,
616 			       io->io_hdr.nexus.targ_lun);
617 #endif
618 			io->scsiio.tag_num = msg_info.scsi.tag_num;
619 			io->scsiio.tag_type = msg_info.scsi.tag_type;
620 			memcpy(io->scsiio.cdb, msg_info.scsi.cdb,
621 			       CTL_MAX_CDBLEN);
622 			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
623 				struct ctl_cmd_entry *entry;
624 				uint8_t opcode;
625 
626 				opcode = io->scsiio.cdb[0];
627 				entry = &ctl_cmd_table[opcode];
628 				io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
629 				io->io_hdr.flags |=
630 					entry->flags & CTL_FLAG_DATA_MASK;
631 			}
632 			STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
633 					   &io->io_hdr, links);
634 			ctl_wakeup_thread();
635 			break;
636 
637 		/* Performed on the Originating SC, XFER mode only */
638 		case CTL_MSG_DATAMOVE: {
639 			struct ctl_sg_entry *sgl;
640 			int i, j;
641 
642 			io = msg_info.hdr.original_sc;
643 			if (io == NULL) {
644 				printf("%s: original_sc == NULL!\n", __func__);
645 				/* XXX KDM do something here */
646 				break;
647 			}
648 			io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
649 			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
650 			/*
651 			 * Keep track of this, we need to send it back over
652 			 * when the datamove is complete.
653 			 */
654 			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
655 
656 			if (msg_info.dt.sg_sequence == 0) {
657 				/*
658 				 * XXX KDM we use the preallocated S/G list
659 				 * here, but we'll need to change this to
660 				 * dynamic allocation if we need larger S/G
661 				 * lists.
662 				 */
663 				if (msg_info.dt.kern_sg_entries >
664 				    sizeof(io->io_hdr.remote_sglist) /
665 				    sizeof(io->io_hdr.remote_sglist[0])) {
666 					printf("%s: number of S/G entries "
667 					    "needed %u > allocated num %zd\n",
668 					    __func__,
669 					    msg_info.dt.kern_sg_entries,
670 					    sizeof(io->io_hdr.remote_sglist)/
671 					    sizeof(io->io_hdr.remote_sglist[0]));
672 
673 					/*
674 					 * XXX KDM send a message back to
675 					 * the other side to shut down the
676 					 * DMA.  The error will come back
677 					 * through via the normal channel.
678 					 */
679 					break;
680 				}
681 				sgl = io->io_hdr.remote_sglist;
682 				memset(sgl, 0,
683 				       sizeof(io->io_hdr.remote_sglist));
684 
685 				io->scsiio.kern_data_ptr = (uint8_t *)sgl;
686 
687 				io->scsiio.kern_sg_entries =
688 					msg_info.dt.kern_sg_entries;
689 				io->scsiio.rem_sg_entries =
690 					msg_info.dt.kern_sg_entries;
691 				io->scsiio.kern_data_len =
692 					msg_info.dt.kern_data_len;
693 				io->scsiio.kern_total_len =
694 					msg_info.dt.kern_total_len;
695 				io->scsiio.kern_data_resid =
696 					msg_info.dt.kern_data_resid;
697 				io->scsiio.kern_rel_offset =
698 					msg_info.dt.kern_rel_offset;
699 				/*
700 				 * Clear out per-DMA flags.
701 				 */
702 				io->io_hdr.flags &= ~CTL_FLAG_RDMA_MASK;
703 				/*
704 				 * Add per-DMA flags that are set for this
705 				 * particular DMA request.
706 				 */
707 				io->io_hdr.flags |= msg_info.dt.flags &
708 						    CTL_FLAG_RDMA_MASK;
709 			} else
710 				sgl = (struct ctl_sg_entry *)
711 					io->scsiio.kern_data_ptr;
712 
713 			for (i = msg_info.dt.sent_sg_entries, j = 0;
714 			     i < (msg_info.dt.sent_sg_entries +
715 			     msg_info.dt.cur_sg_entries); i++, j++) {
716 				sgl[i].addr = msg_info.dt.sg_list[j].addr;
717 				sgl[i].len = msg_info.dt.sg_list[j].len;
718 
719 #if 0
720 				printf("%s: L: %p,%d -> %p,%d j=%d, i=%d\n",
721 				       __func__,
722 				       msg_info.dt.sg_list[j].addr,
723 				       msg_info.dt.sg_list[j].len,
724 				       sgl[i].addr, sgl[i].len, j, i);
725 #endif
726 			}
727 #if 0
728 			memcpy(&sgl[msg_info.dt.sent_sg_entries],
729 			       msg_info.dt.sg_list,
730 			       sizeof(*sgl) * msg_info.dt.cur_sg_entries);
731 #endif
732 
733 			/*
734 			 * If this is the last piece of the I/O, we've got
735 			 * the full S/G list.  Queue processing in the thread.
736 			 * Otherwise wait for the next piece.
737 			 */
738 			if (msg_info.dt.sg_last != 0) {
739 				STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
740 						   &io->io_hdr, links);
741 				ctl_wakeup_thread();
742 			}
743 			break;
744 		}
745 		/* Performed on the Serializing (primary) SC, XFER mode only */
746 		case CTL_MSG_DATAMOVE_DONE: {
747 			if (msg_info.hdr.serializing_sc == NULL) {
748 				printf("%s: serializing_sc == NULL!\n",
749 				       __func__);
750 				/* XXX KDM now what? */
751 				break;
752 			}
753 			/*
754 			 * We grab the sense information here in case
755 			 * there was a failure, so we can return status
756 			 * back to the initiator.
757 			 */
758 			io = msg_info.hdr.serializing_sc;
759 			io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
760 			io->io_hdr.status = msg_info.hdr.status;
761 			io->scsiio.scsi_status = msg_info.scsi.scsi_status;
762 			io->scsiio.sense_len = msg_info.scsi.sense_len;
763 			io->scsiio.sense_residual =msg_info.scsi.sense_residual;
764 			io->io_hdr.port_status = msg_info.scsi.fetd_status;
765 			io->scsiio.residual = msg_info.scsi.residual;
766 			memcpy(&io->scsiio.sense_data,&msg_info.scsi.sense_data,
767 			       sizeof(io->scsiio.sense_data));
768 
769 			STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
770 					   &io->io_hdr, links);
771 			ctl_wakeup_thread();
772 			break;
773 		}
774 
775 		/* Preformed on Originating SC, SER_ONLY mode */
776 		case CTL_MSG_R2R:
777 			io = msg_info.hdr.original_sc;
778 			if (io == NULL) {
779 				printf("%s: Major Bummer\n", __func__);
780 				mtx_unlock(&ctl_softc->ctl_lock);
781 				return;
782 			} else {
783 #if 0
784 				printf("pOrig %x\n",(int) ctsio);
785 #endif
786 			}
787 			io->io_hdr.msg_type = CTL_MSG_R2R;
788 			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
789 			STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
790 					   &io->io_hdr, links);
791 			ctl_wakeup_thread();
792 			break;
793 
794 		/*
795 		 * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
796 		 * mode.
797 		 * Performed on the Originating (i.e. secondary) SC in XFER
798 		 * mode
799 		 */
800 		case CTL_MSG_FINISH_IO:
801 			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER)
802 				ctl_isc_handler_finish_xfer(ctl_softc,
803 							    &msg_info);
804 			else
805 				ctl_isc_handler_finish_ser_only(ctl_softc,
806 								&msg_info);
807 			break;
808 
809 		/* Preformed on Originating SC */
810 		case CTL_MSG_BAD_JUJU:
811 			io = msg_info.hdr.original_sc;
812 			if (io == NULL) {
813 				printf("%s: Bad JUJU!, original_sc is NULL!\n",
814 				       __func__);
815 				break;
816 			}
817 			ctl_copy_sense_data(&msg_info, io);
818 			/*
819 			 * IO should have already been cleaned up on other
820 			 * SC so clear this flag so we won't send a message
821 			 * back to finish the IO there.
822 			 */
823 			io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
824 			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
825 
826 			/* io = msg_info.hdr.serializing_sc; */
827 			io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
828 		        STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
829 					   &io->io_hdr, links);
830 			ctl_wakeup_thread();
831 			break;
832 
833 		/* Handle resets sent from the other side */
834 		case CTL_MSG_MANAGE_TASKS: {
835 			struct ctl_taskio *taskio;
836 			taskio = (struct ctl_taskio *)ctl_alloc_io(
837 				(void *)ctl_softc->othersc_pool);
838 			if (taskio == NULL) {
839 				printf("ctl_isc_event_handler: can't allocate "
840 				       "ctl_io!\n");
841 				/* Bad Juju */
842 				/* should I just call the proper reset func
843 				   here??? */
844 				mtx_unlock(&ctl_softc->ctl_lock);
845 				goto bailout;
846 			}
847 			ctl_zero_io((union ctl_io *)taskio);
848 			taskio->io_hdr.io_type = CTL_IO_TASK;
849 			taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
850 			taskio->io_hdr.nexus = msg_info.hdr.nexus;
851 			taskio->task_action = msg_info.task.task_action;
852 			taskio->tag_num = msg_info.task.tag_num;
853 			taskio->tag_type = msg_info.task.tag_type;
854 #ifdef CTL_TIME_IO
855 			taskio->io_hdr.start_time = time_uptime;
856 			getbintime(&taskio->io_hdr.start_bt);
857 #if 0
858 			cs_prof_gettime(&taskio->io_hdr.start_ticks);
859 #endif
860 #endif /* CTL_TIME_IO */
861 		        STAILQ_INSERT_TAIL(&ctl_softc->task_queue,
862 					   &taskio->io_hdr, links);
863 			ctl_softc->flags |= CTL_FLAG_TASK_PENDING;
864 			ctl_wakeup_thread();
865 			break;
866 		}
867 		/* Persistent Reserve action which needs attention */
868 		case CTL_MSG_PERS_ACTION:
869 			presio = (struct ctl_prio *)ctl_alloc_io(
870 				(void *)ctl_softc->othersc_pool);
871 			if (presio == NULL) {
872 				printf("ctl_isc_event_handler: can't allocate "
873 				       "ctl_io!\n");
874 				/* Bad Juju */
875 				/* Need to set busy and send msg back */
876 				mtx_unlock(&ctl_softc->ctl_lock);
877 				goto bailout;
878 			}
879 			ctl_zero_io((union ctl_io *)presio);
880 			presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
881 			presio->pr_msg = msg_info.pr;
882 		        STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
883 					   &presio->io_hdr, links);
884 			ctl_wakeup_thread();
885 			break;
886 		case CTL_MSG_SYNC_FE:
887 			rcv_sync_msg = 1;
888 			break;
889 		case CTL_MSG_APS_LOCK: {
890 			// It's quicker to execute this then to
891 			// queue it.
892 			struct ctl_lun *lun;
893 			struct ctl_page_index *page_index;
894 			struct copan_aps_subpage *current_sp;
895 
896 			lun = ctl_softc->ctl_luns[msg_info.hdr.nexus.targ_lun];
897 			page_index = &lun->mode_pages.index[index_to_aps_page];
898 			current_sp = (struct copan_aps_subpage *)
899 				     (page_index->page_data +
900 				     (page_index->page_len * CTL_PAGE_CURRENT));
901 
902 			current_sp->lock_active = msg_info.aps.lock_flag;
903 		        break;
904 		}
905 		default:
906 		        printf("How did I get here?\n");
907 		}
908 		mtx_unlock(&ctl_softc->ctl_lock);
909 	} else if (event == CTL_HA_EVT_MSG_SENT) {
910 		if (param != CTL_HA_STATUS_SUCCESS) {
911 			printf("Bad status from ctl_ha_msg_send status %d\n",
912 			       param);
913 		}
914 		return;
915 	} else if (event == CTL_HA_EVT_DISCONNECT) {
916 		printf("CTL: Got a disconnect from Isc\n");
917 		return;
918 	} else {
919 		printf("ctl_isc_event_handler: Unknown event %d\n", event);
920 		return;
921 	}
922 
923 bailout:
924 	return;
925 }
926 
927 static void
928 ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
929 {
930 	struct scsi_sense_data *sense;
931 
932 	sense = &dest->scsiio.sense_data;
933 	bcopy(&src->scsi.sense_data, sense, sizeof(*sense));
934 	dest->scsiio.scsi_status = src->scsi.scsi_status;
935 	dest->scsiio.sense_len = src->scsi.sense_len;
936 	dest->io_hdr.status = src->hdr.status;
937 }
938 
939 static void
940 ctl_init(void)
941 {
942 	struct ctl_softc *softc;
943 	struct ctl_io_pool *internal_pool, *emergency_pool, *other_pool;
944 	struct ctl_frontend *fe;
945 	struct ctl_lun *lun;
946         uint8_t sc_id =0;
947 #if 0
948 	int i;
949 #endif
950 	int retval;
951 	//int isc_retval;
952 
953 	retval = 0;
954 	ctl_pause_rtr = 0;
955         rcv_sync_msg = 0;
956 
957 	/* If we're disabled, don't initialize. */
958 	if (ctl_disable != 0)
959 		return;
960 
961 	control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
962 			       M_WAITOK | M_ZERO);
963 	softc = control_softc;
964 
965 	softc->dev = make_dev(&ctl_cdevsw, 0, UID_ROOT, GID_OPERATOR, 0600,
966 			      "cam/ctl");
967 
968 	softc->dev->si_drv1 = softc;
969 
970 	/*
971 	 * By default, return a "bad LUN" peripheral qualifier for unknown
972 	 * LUNs.  The user can override this default using the tunable or
973 	 * sysctl.  See the comment in ctl_inquiry_std() for more details.
974 	 */
975 	softc->inquiry_pq_no_lun = 1;
976 	TUNABLE_INT_FETCH("kern.cam.ctl.inquiry_pq_no_lun",
977 			  &softc->inquiry_pq_no_lun);
978 	sysctl_ctx_init(&softc->sysctl_ctx);
979 	softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
980 		SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
981 		CTLFLAG_RD, 0, "CAM Target Layer");
982 
983 	if (softc->sysctl_tree == NULL) {
984 		printf("%s: unable to allocate sysctl tree\n", __func__);
985 		destroy_dev(softc->dev);
986 		free(control_softc, M_DEVBUF);
987 		control_softc = NULL;
988 		return;
989 	}
990 
991 	SYSCTL_ADD_INT(&softc->sysctl_ctx,
992 		       SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO,
993 		       "inquiry_pq_no_lun", CTLFLAG_RW,
994 		       &softc->inquiry_pq_no_lun, 0,
995 		       "Report no lun possible for invalid LUNs");
996 
997 	mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
998 	softc->open_count = 0;
999 
1000 	/*
1001 	 * Default to actually sending a SYNCHRONIZE CACHE command down to
1002 	 * the drive.
1003 	 */
1004 	softc->flags = CTL_FLAG_REAL_SYNC;
1005 
1006 	/*
1007 	 * In Copan's HA scheme, the "master" and "slave" roles are
1008 	 * figured out through the slot the controller is in.  Although it
1009 	 * is an active/active system, someone has to be in charge.
1010  	 */
1011 #ifdef NEEDTOPORT
1012         scmicro_rw(SCMICRO_GET_SHELF_ID, &sc_id);
1013 #endif
1014 
1015         if (sc_id == 0) {
1016 		softc->flags |= CTL_FLAG_MASTER_SHELF;
1017 		persis_offset = 0;
1018 	} else
1019 		persis_offset = CTL_MAX_INITIATORS;
1020 
1021 	/*
1022 	 * XXX KDM need to figure out where we want to get our target ID
1023 	 * and WWID.  Is it different on each port?
1024 	 */
1025 	softc->target.id = 0;
1026 	softc->target.wwid[0] = 0x12345678;
1027 	softc->target.wwid[1] = 0x87654321;
1028 	STAILQ_INIT(&softc->lun_list);
1029 	STAILQ_INIT(&softc->pending_lun_queue);
1030 	STAILQ_INIT(&softc->task_queue);
1031 	STAILQ_INIT(&softc->incoming_queue);
1032 	STAILQ_INIT(&softc->rtr_queue);
1033 	STAILQ_INIT(&softc->done_queue);
1034 	STAILQ_INIT(&softc->isc_queue);
1035 	STAILQ_INIT(&softc->fe_list);
1036 	STAILQ_INIT(&softc->be_list);
1037 	STAILQ_INIT(&softc->io_pools);
1038 
1039 	lun = &softc->lun;
1040 
1041 	/*
1042 	 * We don't bother calling these with ctl_lock held here, because,
1043 	 * in theory, no one else can try to do anything while we're in our
1044 	 * module init routine.
1045 	 */
1046 	if (ctl_pool_create(softc, CTL_POOL_INTERNAL, CTL_POOL_ENTRIES_INTERNAL,
1047 			    &internal_pool)!= 0){
1048 		printf("ctl: can't allocate %d entry internal pool, "
1049 		       "exiting\n", CTL_POOL_ENTRIES_INTERNAL);
1050 		return;
1051 	}
1052 
1053 	if (ctl_pool_create(softc, CTL_POOL_EMERGENCY,
1054 			    CTL_POOL_ENTRIES_EMERGENCY, &emergency_pool) != 0) {
1055 		printf("ctl: can't allocate %d entry emergency pool, "
1056 		       "exiting\n", CTL_POOL_ENTRIES_EMERGENCY);
1057 		ctl_pool_free(softc, internal_pool);
1058 		return;
1059 	}
1060 
1061 	if (ctl_pool_create(softc, CTL_POOL_4OTHERSC, CTL_POOL_ENTRIES_OTHER_SC,
1062 	                    &other_pool) != 0)
1063 	{
1064 		printf("ctl: can't allocate %d entry other SC pool, "
1065 		       "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1066 		ctl_pool_free(softc, internal_pool);
1067 		ctl_pool_free(softc, emergency_pool);
1068 		return;
1069 	}
1070 
1071 	softc->internal_pool = internal_pool;
1072 	softc->emergency_pool = emergency_pool;
1073 	softc->othersc_pool = other_pool;
1074 
1075 	ctl_pool_acquire(internal_pool);
1076 	ctl_pool_acquire(emergency_pool);
1077 	ctl_pool_acquire(other_pool);
1078 
1079 	/*
1080 	 * We used to allocate a processor LUN here.  The new scheme is to
1081 	 * just let the user allocate LUNs as he sees fit.
1082 	 */
1083 #if 0
1084 	mtx_lock(&softc->ctl_lock);
1085 	ctl_alloc_lun(softc, lun, /*be_lun*/NULL, /*target*/softc->target);
1086 	mtx_unlock(&softc->ctl_lock);
1087 #endif
1088 
1089 	if (kproc_create(ctl_work_thread, softc, &softc->work_thread, 0, 0,
1090 			 "ctl_thrd") != 0) {
1091 		printf("error creating CTL work thread!\n");
1092 		ctl_free_lun(lun);
1093 		ctl_pool_free(softc, internal_pool);
1094 		ctl_pool_free(softc, emergency_pool);
1095 		ctl_pool_free(softc, other_pool);
1096 		return;
1097 	}
1098 	printf("ctl: CAM Target Layer loaded\n");
1099 
1100 	/*
1101 	 * Initialize the initiator and portname mappings
1102 	 */
1103 	memset(softc->wwpn_iid, 0, sizeof(softc->wwpn_iid));
1104 
1105 	/*
1106 	 * Initialize the ioctl front end.
1107 	 */
1108 	fe = &softc->ioctl_info.fe;
1109 	sprintf(softc->ioctl_info.port_name, "CTL ioctl");
1110 	fe->port_type = CTL_PORT_IOCTL;
1111 	fe->num_requested_ctl_io = 100;
1112 	fe->port_name = softc->ioctl_info.port_name;
1113 	fe->port_online = ctl_ioctl_online;
1114 	fe->port_offline = ctl_ioctl_offline;
1115 	fe->onoff_arg = &softc->ioctl_info;
1116 	fe->targ_enable = ctl_ioctl_targ_enable;
1117 	fe->targ_disable = ctl_ioctl_targ_disable;
1118 	fe->lun_enable = ctl_ioctl_lun_enable;
1119 	fe->lun_disable = ctl_ioctl_lun_disable;
1120 	fe->targ_lun_arg = &softc->ioctl_info;
1121 	fe->fe_datamove = ctl_ioctl_datamove;
1122 	fe->fe_done = ctl_ioctl_done;
1123 	fe->max_targets = 15;
1124 	fe->max_target_id = 15;
1125 
1126 	if (ctl_frontend_register(&softc->ioctl_info.fe,
1127 	                  (softc->flags & CTL_FLAG_MASTER_SHELF)) != 0) {
1128 		printf("ctl: ioctl front end registration failed, will "
1129 		       "continue anyway\n");
1130 	}
1131 
1132 #ifdef CTL_IO_DELAY
1133 	if (sizeof(struct callout) > CTL_TIMER_BYTES) {
1134 		printf("sizeof(struct callout) %zd > CTL_TIMER_BYTES %zd\n",
1135 		       sizeof(struct callout), CTL_TIMER_BYTES);
1136 		return;
1137 	}
1138 #endif /* CTL_IO_DELAY */
1139 
1140 }
1141 
1142 void
1143 ctl_shutdown(void)
1144 {
1145 	struct ctl_softc *softc;
1146 	struct ctl_lun *lun, *next_lun;
1147 	struct ctl_io_pool *pool, *next_pool;
1148 
1149 	softc = (struct ctl_softc *)control_softc;
1150 
1151 	if (ctl_frontend_deregister(&softc->ioctl_info.fe) != 0)
1152 		printf("ctl: ioctl front end deregistration failed\n");
1153 
1154 	mtx_lock(&softc->ctl_lock);
1155 
1156 	/*
1157 	 * Free up each LUN.
1158 	 */
1159 	for (lun = STAILQ_FIRST(&softc->lun_list); lun != NULL; lun = next_lun){
1160 		next_lun = STAILQ_NEXT(lun, links);
1161 		ctl_free_lun(lun);
1162 	}
1163 
1164 	/*
1165 	 * This will rip the rug out from under any FETDs or anyone else
1166 	 * that has a pool allocated.  Since we increment our module
1167 	 * refcount any time someone outside the main CTL module allocates
1168 	 * a pool, we shouldn't have any problems here.  The user won't be
1169 	 * able to unload the CTL module until client modules have
1170 	 * successfully unloaded.
1171 	 */
1172 	for (pool = STAILQ_FIRST(&softc->io_pools); pool != NULL;
1173 	     pool = next_pool) {
1174 		next_pool = STAILQ_NEXT(pool, links);
1175 		ctl_pool_free(softc, pool);
1176 	}
1177 
1178 	mtx_unlock(&softc->ctl_lock);
1179 
1180 #if 0
1181 	ctl_shutdown_thread(softc->work_thread);
1182 #endif
1183 
1184 	mtx_destroy(&softc->ctl_lock);
1185 
1186 	destroy_dev(softc->dev);
1187 
1188 	sysctl_ctx_free(&softc->sysctl_ctx);
1189 
1190 	free(control_softc, M_DEVBUF);
1191 	control_softc = NULL;
1192 
1193 	printf("ctl: CAM Target Layer unloaded\n");
1194 }
1195 
1196 /*
1197  * XXX KDM should we do some access checks here?  Bump a reference count to
1198  * prevent a CTL module from being unloaded while someone has it open?
1199  */
1200 static int
1201 ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1202 {
1203 	return (0);
1204 }
1205 
1206 static int
1207 ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1208 {
1209 	return (0);
1210 }
1211 
1212 int
1213 ctl_port_enable(ctl_port_type port_type)
1214 {
1215 	struct ctl_softc *softc;
1216 	struct ctl_frontend *fe;
1217 
1218 	if (ctl_is_single == 0) {
1219 		union ctl_ha_msg msg_info;
1220 		int isc_retval;
1221 
1222 #if 0
1223 		printf("%s: HA mode, synchronizing frontend enable\n",
1224 		        __func__);
1225 #endif
1226 		msg_info.hdr.msg_type = CTL_MSG_SYNC_FE;
1227 	        if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1228 		        sizeof(msg_info), 1 )) > CTL_HA_STATUS_SUCCESS) {
1229 			printf("Sync msg send error retval %d\n", isc_retval);
1230 		}
1231 		if (!rcv_sync_msg) {
1232 			isc_retval=ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
1233 			        sizeof(msg_info), 1);
1234 		}
1235 #if 0
1236         	printf("CTL:Frontend Enable\n");
1237 	} else {
1238 		printf("%s: single mode, skipping frontend synchronization\n",
1239 		        __func__);
1240 #endif
1241 	}
1242 
1243 	softc = control_softc;
1244 
1245 	STAILQ_FOREACH(fe, &softc->fe_list, links) {
1246 		if (port_type & fe->port_type)
1247 		{
1248 #if 0
1249 			printf("port %d\n", fe->targ_port);
1250 #endif
1251 			ctl_frontend_online(fe);
1252 		}
1253 	}
1254 
1255 	return (0);
1256 }
1257 
1258 int
1259 ctl_port_disable(ctl_port_type port_type)
1260 {
1261 	struct ctl_softc *softc;
1262 	struct ctl_frontend *fe;
1263 
1264 	softc = control_softc;
1265 
1266 	STAILQ_FOREACH(fe, &softc->fe_list, links) {
1267 		if (port_type & fe->port_type)
1268 			ctl_frontend_offline(fe);
1269 	}
1270 
1271 	return (0);
1272 }
1273 
1274 /*
1275  * Returns 0 for success, 1 for failure.
1276  * Currently the only failure mode is if there aren't enough entries
1277  * allocated.  So, in case of a failure, look at num_entries_dropped,
1278  * reallocate and try again.
1279  */
1280 int
1281 ctl_port_list(struct ctl_port_entry *entries, int num_entries_alloced,
1282 	      int *num_entries_filled, int *num_entries_dropped,
1283 	      ctl_port_type port_type, int no_virtual)
1284 {
1285 	struct ctl_softc *softc;
1286 	struct ctl_frontend *fe;
1287 	int entries_dropped, entries_filled;
1288 	int retval;
1289 	int i;
1290 
1291 	softc = control_softc;
1292 
1293 	retval = 0;
1294 	entries_filled = 0;
1295 	entries_dropped = 0;
1296 
1297 	i = 0;
1298 	mtx_lock(&softc->ctl_lock);
1299 	STAILQ_FOREACH(fe, &softc->fe_list, links) {
1300 		struct ctl_port_entry *entry;
1301 
1302 		if ((fe->port_type & port_type) == 0)
1303 			continue;
1304 
1305 		if ((no_virtual != 0)
1306 		 && (fe->virtual_port != 0))
1307 			continue;
1308 
1309 		if (entries_filled >= num_entries_alloced) {
1310 			entries_dropped++;
1311 			continue;
1312 		}
1313 		entry = &entries[i];
1314 
1315 		entry->port_type = fe->port_type;
1316 		strlcpy(entry->port_name, fe->port_name,
1317 			sizeof(entry->port_name));
1318 		entry->physical_port = fe->physical_port;
1319 		entry->virtual_port = fe->virtual_port;
1320 		entry->wwnn = fe->wwnn;
1321 		entry->wwpn = fe->wwpn;
1322 
1323 		i++;
1324 		entries_filled++;
1325 	}
1326 
1327 	mtx_unlock(&softc->ctl_lock);
1328 
1329 	if (entries_dropped > 0)
1330 		retval = 1;
1331 
1332 	*num_entries_dropped = entries_dropped;
1333 	*num_entries_filled = entries_filled;
1334 
1335 	return (retval);
1336 }
1337 
1338 static void
1339 ctl_ioctl_online(void *arg)
1340 {
1341 	struct ctl_ioctl_info *ioctl_info;
1342 
1343 	ioctl_info = (struct ctl_ioctl_info *)arg;
1344 
1345 	ioctl_info->flags |= CTL_IOCTL_FLAG_ENABLED;
1346 }
1347 
1348 static void
1349 ctl_ioctl_offline(void *arg)
1350 {
1351 	struct ctl_ioctl_info *ioctl_info;
1352 
1353 	ioctl_info = (struct ctl_ioctl_info *)arg;
1354 
1355 	ioctl_info->flags &= ~CTL_IOCTL_FLAG_ENABLED;
1356 }
1357 
1358 /*
1359  * Remove an initiator by port number and initiator ID.
1360  * Returns 0 for success, 1 for failure.
1361  * Assumes the caller does NOT hold the CTL lock.
1362  */
1363 int
1364 ctl_remove_initiator(int32_t targ_port, uint32_t iid)
1365 {
1366 	struct ctl_softc *softc;
1367 
1368 	softc = control_softc;
1369 
1370 	if ((targ_port < 0)
1371 	 || (targ_port > CTL_MAX_PORTS)) {
1372 		printf("%s: invalid port number %d\n", __func__, targ_port);
1373 		return (1);
1374 	}
1375 	if (iid > CTL_MAX_INIT_PER_PORT) {
1376 		printf("%s: initiator ID %u > maximun %u!\n",
1377 		       __func__, iid, CTL_MAX_INIT_PER_PORT);
1378 		return (1);
1379 	}
1380 
1381 	mtx_lock(&softc->ctl_lock);
1382 
1383 	softc->wwpn_iid[targ_port][iid].in_use = 0;
1384 
1385 	mtx_unlock(&softc->ctl_lock);
1386 
1387 	return (0);
1388 }
1389 
1390 /*
1391  * Add an initiator to the initiator map.
1392  * Returns 0 for success, 1 for failure.
1393  * Assumes the caller does NOT hold the CTL lock.
1394  */
1395 int
1396 ctl_add_initiator(uint64_t wwpn, int32_t targ_port, uint32_t iid)
1397 {
1398 	struct ctl_softc *softc;
1399 	int retval;
1400 
1401 	softc = control_softc;
1402 
1403 	retval = 0;
1404 
1405 	if ((targ_port < 0)
1406 	 || (targ_port > CTL_MAX_PORTS)) {
1407 		printf("%s: invalid port number %d\n", __func__, targ_port);
1408 		return (1);
1409 	}
1410 	if (iid > CTL_MAX_INIT_PER_PORT) {
1411 		printf("%s: WWPN %#jx initiator ID %u > maximun %u!\n",
1412 		       __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
1413 		return (1);
1414 	}
1415 
1416 	mtx_lock(&softc->ctl_lock);
1417 
1418 	if (softc->wwpn_iid[targ_port][iid].in_use != 0) {
1419 		/*
1420 		 * We don't treat this as an error.
1421 		 */
1422 		if (softc->wwpn_iid[targ_port][iid].wwpn == wwpn) {
1423 			printf("%s: port %d iid %u WWPN %#jx arrived again?\n",
1424 			       __func__, targ_port, iid, (uintmax_t)wwpn);
1425 			goto bailout;
1426 		}
1427 
1428 		/*
1429 		 * This is an error, but what do we do about it?  The
1430 		 * driver is telling us we have a new WWPN for this
1431 		 * initiator ID, so we pretty much need to use it.
1432 		 */
1433 		printf("%s: port %d iid %u WWPN %#jx arrived, WWPN %#jx is "
1434 		       "still at that address\n", __func__, targ_port, iid,
1435 		       (uintmax_t)wwpn,
1436 		       (uintmax_t)softc->wwpn_iid[targ_port][iid].wwpn);
1437 
1438 		/*
1439 		 * XXX KDM clear have_ca and ua_pending on each LUN for
1440 		 * this initiator.
1441 		 */
1442 	}
1443 	softc->wwpn_iid[targ_port][iid].in_use = 1;
1444 	softc->wwpn_iid[targ_port][iid].iid = iid;
1445 	softc->wwpn_iid[targ_port][iid].wwpn = wwpn;
1446 	softc->wwpn_iid[targ_port][iid].port = targ_port;
1447 
1448 bailout:
1449 
1450 	mtx_unlock(&softc->ctl_lock);
1451 
1452 	return (retval);
1453 }
1454 
1455 /*
1456  * XXX KDM should we pretend to do something in the target/lun
1457  * enable/disable functions?
1458  */
1459 static int
1460 ctl_ioctl_targ_enable(void *arg, struct ctl_id targ_id)
1461 {
1462 	return (0);
1463 }
1464 
1465 static int
1466 ctl_ioctl_targ_disable(void *arg, struct ctl_id targ_id)
1467 {
1468 	return (0);
1469 }
1470 
1471 static int
1472 ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id)
1473 {
1474 	return (0);
1475 }
1476 
1477 static int
1478 ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id)
1479 {
1480 	return (0);
1481 }
1482 
1483 /*
1484  * Data movement routine for the CTL ioctl frontend port.
1485  */
1486 static int
1487 ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio)
1488 {
1489 	struct ctl_sg_entry *ext_sglist, *kern_sglist;
1490 	struct ctl_sg_entry ext_entry, kern_entry;
1491 	int ext_sglen, ext_sg_entries, kern_sg_entries;
1492 	int ext_sg_start, ext_offset;
1493 	int len_to_copy, len_copied;
1494 	int kern_watermark, ext_watermark;
1495 	int ext_sglist_malloced;
1496 	int i, j;
1497 
1498 	ext_sglist_malloced = 0;
1499 	ext_sg_start = 0;
1500 	ext_offset = 0;
1501 
1502 	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove\n"));
1503 
1504 	/*
1505 	 * If this flag is set, fake the data transfer.
1506 	 */
1507 	if (ctsio->io_hdr.flags & CTL_FLAG_NO_DATAMOVE) {
1508 		ctsio->ext_data_filled = ctsio->ext_data_len;
1509 		goto bailout;
1510 	}
1511 
1512 	/*
1513 	 * To simplify things here, if we have a single buffer, stick it in
1514 	 * a S/G entry and just make it a single entry S/G list.
1515 	 */
1516 	if (ctsio->io_hdr.flags & CTL_FLAG_EDPTR_SGLIST) {
1517 		int len_seen;
1518 
1519 		ext_sglen = ctsio->ext_sg_entries * sizeof(*ext_sglist);
1520 
1521 		ext_sglist = (struct ctl_sg_entry *)malloc(ext_sglen, M_CTL,
1522 							   M_WAITOK);
1523 		ext_sglist_malloced = 1;
1524 		if (copyin(ctsio->ext_data_ptr, ext_sglist,
1525 				   ext_sglen) != 0) {
1526 			ctl_set_internal_failure(ctsio,
1527 						 /*sks_valid*/ 0,
1528 						 /*retry_count*/ 0);
1529 			goto bailout;
1530 		}
1531 		ext_sg_entries = ctsio->ext_sg_entries;
1532 		len_seen = 0;
1533 		for (i = 0; i < ext_sg_entries; i++) {
1534 			if ((len_seen + ext_sglist[i].len) >=
1535 			     ctsio->ext_data_filled) {
1536 				ext_sg_start = i;
1537 				ext_offset = ctsio->ext_data_filled - len_seen;
1538 				break;
1539 			}
1540 			len_seen += ext_sglist[i].len;
1541 		}
1542 	} else {
1543 		ext_sglist = &ext_entry;
1544 		ext_sglist->addr = ctsio->ext_data_ptr;
1545 		ext_sglist->len = ctsio->ext_data_len;
1546 		ext_sg_entries = 1;
1547 		ext_sg_start = 0;
1548 		ext_offset = ctsio->ext_data_filled;
1549 	}
1550 
1551 	if (ctsio->kern_sg_entries > 0) {
1552 		kern_sglist = (struct ctl_sg_entry *)ctsio->kern_data_ptr;
1553 		kern_sg_entries = ctsio->kern_sg_entries;
1554 	} else {
1555 		kern_sglist = &kern_entry;
1556 		kern_sglist->addr = ctsio->kern_data_ptr;
1557 		kern_sglist->len = ctsio->kern_data_len;
1558 		kern_sg_entries = 1;
1559 	}
1560 
1561 
1562 	kern_watermark = 0;
1563 	ext_watermark = ext_offset;
1564 	len_copied = 0;
1565 	for (i = ext_sg_start, j = 0;
1566 	     i < ext_sg_entries && j < kern_sg_entries;) {
1567 		uint8_t *ext_ptr, *kern_ptr;
1568 
1569 		len_to_copy = ctl_min(ext_sglist[i].len - ext_watermark,
1570 				      kern_sglist[j].len - kern_watermark);
1571 
1572 		ext_ptr = (uint8_t *)ext_sglist[i].addr;
1573 		ext_ptr = ext_ptr + ext_watermark;
1574 		if (ctsio->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
1575 			/*
1576 			 * XXX KDM fix this!
1577 			 */
1578 			panic("need to implement bus address support");
1579 #if 0
1580 			kern_ptr = bus_to_virt(kern_sglist[j].addr);
1581 #endif
1582 		} else
1583 			kern_ptr = (uint8_t *)kern_sglist[j].addr;
1584 		kern_ptr = kern_ptr + kern_watermark;
1585 
1586 		kern_watermark += len_to_copy;
1587 		ext_watermark += len_to_copy;
1588 
1589 		if ((ctsio->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
1590 		     CTL_FLAG_DATA_IN) {
1591 			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1592 					 "bytes to user\n", len_to_copy));
1593 			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1594 					 "to %p\n", kern_ptr, ext_ptr));
1595 			if (copyout(kern_ptr, ext_ptr, len_to_copy) != 0) {
1596 				ctl_set_internal_failure(ctsio,
1597 							 /*sks_valid*/ 0,
1598 							 /*retry_count*/ 0);
1599 				goto bailout;
1600 			}
1601 		} else {
1602 			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1603 					 "bytes from user\n", len_to_copy));
1604 			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1605 					 "to %p\n", ext_ptr, kern_ptr));
1606 			if (copyin(ext_ptr, kern_ptr, len_to_copy)!= 0){
1607 				ctl_set_internal_failure(ctsio,
1608 							 /*sks_valid*/ 0,
1609 							 /*retry_count*/0);
1610 				goto bailout;
1611 			}
1612 		}
1613 
1614 		len_copied += len_to_copy;
1615 
1616 		if (ext_sglist[i].len == ext_watermark) {
1617 			i++;
1618 			ext_watermark = 0;
1619 		}
1620 
1621 		if (kern_sglist[j].len == kern_watermark) {
1622 			j++;
1623 			kern_watermark = 0;
1624 		}
1625 	}
1626 
1627 	ctsio->ext_data_filled += len_copied;
1628 
1629 	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_sg_entries: %d, "
1630 			 "kern_sg_entries: %d\n", ext_sg_entries,
1631 			 kern_sg_entries));
1632 	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_data_len = %d, "
1633 			 "kern_data_len = %d\n", ctsio->ext_data_len,
1634 			 ctsio->kern_data_len));
1635 
1636 
1637 	/* XXX KDM set residual?? */
1638 bailout:
1639 
1640 	if (ext_sglist_malloced != 0)
1641 		free(ext_sglist, M_CTL);
1642 
1643 	return (CTL_RETVAL_COMPLETE);
1644 }
1645 
1646 /*
1647  * Serialize a command that went down the "wrong" side, and so was sent to
1648  * this controller for execution.  The logic is a little different than the
1649  * standard case in ctl_scsiio_precheck().  Errors in this case need to get
1650  * sent back to the other side, but in the success case, we execute the
1651  * command on this side (XFER mode) or tell the other side to execute it
1652  * (SER_ONLY mode).
1653  */
1654 static int
1655 ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio, int have_lock)
1656 {
1657 	struct ctl_softc *ctl_softc;
1658 	union ctl_ha_msg msg_info;
1659 	struct ctl_lun *lun;
1660 	int retval = 0;
1661 
1662 	ctl_softc = control_softc;
1663 	if (have_lock == 0)
1664 		mtx_lock(&ctl_softc->ctl_lock);
1665 
1666 	lun = ctl_softc->ctl_luns[ctsio->io_hdr.nexus.targ_lun];
1667 	if (lun==NULL)
1668 	{
1669 		/*
1670 		 * Why isn't LUN defined? The other side wouldn't
1671 		 * send a cmd if the LUN is undefined.
1672 		 */
1673 		printf("%s: Bad JUJU!, LUN is NULL!\n", __func__);
1674 
1675 		/* "Logical unit not supported" */
1676 		ctl_set_sense_data(&msg_info.scsi.sense_data,
1677 				   lun,
1678 				   /*sense_format*/SSD_TYPE_NONE,
1679 				   /*current_error*/ 1,
1680 				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1681 				   /*asc*/ 0x25,
1682 				   /*ascq*/ 0x00,
1683 				   SSD_ELEM_NONE);
1684 
1685 		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1686 		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1687 		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1688 		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1689 		msg_info.hdr.serializing_sc = NULL;
1690 		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1691 	        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1692 				sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1693 		}
1694 		if (have_lock == 0)
1695 			mtx_unlock(&ctl_softc->ctl_lock);
1696 		return(1);
1697 
1698 	}
1699 
1700     	TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1701 
1702 	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
1703 		(union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
1704 		 ooa_links))) {
1705 	case CTL_ACTION_BLOCK:
1706 		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
1707 		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
1708 				  blocked_links);
1709 		break;
1710 	case CTL_ACTION_PASS:
1711 	case CTL_ACTION_SKIP:
1712 		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
1713 			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
1714 			STAILQ_INSERT_TAIL(&ctl_softc->rtr_queue,
1715 					   &ctsio->io_hdr, links);
1716 		} else {
1717 
1718 			/* send msg back to other side */
1719 			msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1720 			msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
1721 			msg_info.hdr.msg_type = CTL_MSG_R2R;
1722 #if 0
1723 			printf("2. pOrig %x\n", (int)msg_info.hdr.original_sc);
1724 #endif
1725 		        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1726 			    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1727 			}
1728 		}
1729 		break;
1730 	case CTL_ACTION_OVERLAP:
1731 		/* OVERLAPPED COMMANDS ATTEMPTED */
1732 		ctl_set_sense_data(&msg_info.scsi.sense_data,
1733 				   lun,
1734 				   /*sense_format*/SSD_TYPE_NONE,
1735 				   /*current_error*/ 1,
1736 				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1737 				   /*asc*/ 0x4E,
1738 				   /*ascq*/ 0x00,
1739 				   SSD_ELEM_NONE);
1740 
1741 		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1742 		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1743 		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1744 		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1745 		msg_info.hdr.serializing_sc = NULL;
1746 		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1747 #if 0
1748 		printf("BAD JUJU:Major Bummer Overlap\n");
1749 #endif
1750 		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1751 		retval = 1;
1752 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1753 		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1754 		}
1755 		break;
1756 	case CTL_ACTION_OVERLAP_TAG:
1757 		/* TAGGED OVERLAPPED COMMANDS (NN = QUEUE TAG) */
1758 		ctl_set_sense_data(&msg_info.scsi.sense_data,
1759 				   lun,
1760 				   /*sense_format*/SSD_TYPE_NONE,
1761 				   /*current_error*/ 1,
1762 				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1763 				   /*asc*/ 0x4D,
1764 				   /*ascq*/ ctsio->tag_num & 0xff,
1765 				   SSD_ELEM_NONE);
1766 
1767 		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1768 		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1769 		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1770 		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1771 		msg_info.hdr.serializing_sc = NULL;
1772 		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1773 #if 0
1774 		printf("BAD JUJU:Major Bummer Overlap Tag\n");
1775 #endif
1776 		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1777 		retval = 1;
1778 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1779 		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1780 		}
1781 		break;
1782 	case CTL_ACTION_ERROR:
1783 	default:
1784 		/* "Internal target failure" */
1785 		ctl_set_sense_data(&msg_info.scsi.sense_data,
1786 				   lun,
1787 				   /*sense_format*/SSD_TYPE_NONE,
1788 				   /*current_error*/ 1,
1789 				   /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
1790 				   /*asc*/ 0x44,
1791 				   /*ascq*/ 0x00,
1792 				   SSD_ELEM_NONE);
1793 
1794 		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1795 		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1796 		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1797 		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1798 		msg_info.hdr.serializing_sc = NULL;
1799 		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1800 #if 0
1801 		printf("BAD JUJU:Major Bummer HW Error\n");
1802 #endif
1803 		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1804 		retval = 1;
1805 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1806 		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1807 		}
1808 		break;
1809 	}
1810 	if (have_lock == 0)
1811 		mtx_unlock(&ctl_softc->ctl_lock);
1812 	return (retval);
1813 }
1814 
1815 static int
1816 ctl_ioctl_submit_wait(union ctl_io *io)
1817 {
1818 	struct ctl_fe_ioctl_params params;
1819 	ctl_fe_ioctl_state last_state;
1820 	int done, retval;
1821 
1822 	retval = 0;
1823 
1824 	bzero(&params, sizeof(params));
1825 
1826 	mtx_init(&params.ioctl_mtx, "ctliocmtx", NULL, MTX_DEF);
1827 	cv_init(&params.sem, "ctlioccv");
1828 	params.state = CTL_IOCTL_INPROG;
1829 	last_state = params.state;
1830 
1831 	io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = &params;
1832 
1833 	CTL_DEBUG_PRINT(("ctl_ioctl_submit_wait\n"));
1834 
1835 	/* This shouldn't happen */
1836 	if ((retval = ctl_queue(io)) != CTL_RETVAL_COMPLETE)
1837 		return (retval);
1838 
1839 	done = 0;
1840 
1841 	do {
1842 		mtx_lock(&params.ioctl_mtx);
1843 		/*
1844 		 * Check the state here, and don't sleep if the state has
1845 		 * already changed (i.e. wakeup has already occured, but we
1846 		 * weren't waiting yet).
1847 		 */
1848 		if (params.state == last_state) {
1849 			/* XXX KDM cv_wait_sig instead? */
1850 			cv_wait(&params.sem, &params.ioctl_mtx);
1851 		}
1852 		last_state = params.state;
1853 
1854 		switch (params.state) {
1855 		case CTL_IOCTL_INPROG:
1856 			/* Why did we wake up? */
1857 			/* XXX KDM error here? */
1858 			mtx_unlock(&params.ioctl_mtx);
1859 			break;
1860 		case CTL_IOCTL_DATAMOVE:
1861 			CTL_DEBUG_PRINT(("got CTL_IOCTL_DATAMOVE\n"));
1862 
1863 			/*
1864 			 * change last_state back to INPROG to avoid
1865 			 * deadlock on subsequent data moves.
1866 			 */
1867 			params.state = last_state = CTL_IOCTL_INPROG;
1868 
1869 			mtx_unlock(&params.ioctl_mtx);
1870 			ctl_ioctl_do_datamove(&io->scsiio);
1871 			/*
1872 			 * Note that in some cases, most notably writes,
1873 			 * this will queue the I/O and call us back later.
1874 			 * In other cases, generally reads, this routine
1875 			 * will immediately call back and wake us up,
1876 			 * probably using our own context.
1877 			 */
1878 			io->scsiio.be_move_done(io);
1879 			break;
1880 		case CTL_IOCTL_DONE:
1881 			mtx_unlock(&params.ioctl_mtx);
1882 			CTL_DEBUG_PRINT(("got CTL_IOCTL_DONE\n"));
1883 			done = 1;
1884 			break;
1885 		default:
1886 			mtx_unlock(&params.ioctl_mtx);
1887 			/* XXX KDM error here? */
1888 			break;
1889 		}
1890 	} while (done == 0);
1891 
1892 	mtx_destroy(&params.ioctl_mtx);
1893 	cv_destroy(&params.sem);
1894 
1895 	return (CTL_RETVAL_COMPLETE);
1896 }
1897 
1898 static void
1899 ctl_ioctl_datamove(union ctl_io *io)
1900 {
1901 	struct ctl_fe_ioctl_params *params;
1902 
1903 	params = (struct ctl_fe_ioctl_params *)
1904 		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
1905 
1906 	mtx_lock(&params->ioctl_mtx);
1907 	params->state = CTL_IOCTL_DATAMOVE;
1908 	cv_broadcast(&params->sem);
1909 	mtx_unlock(&params->ioctl_mtx);
1910 }
1911 
1912 static void
1913 ctl_ioctl_done(union ctl_io *io)
1914 {
1915 	struct ctl_fe_ioctl_params *params;
1916 
1917 	params = (struct ctl_fe_ioctl_params *)
1918 		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
1919 
1920 	mtx_lock(&params->ioctl_mtx);
1921 	params->state = CTL_IOCTL_DONE;
1922 	cv_broadcast(&params->sem);
1923 	mtx_unlock(&params->ioctl_mtx);
1924 }
1925 
1926 static void
1927 ctl_ioctl_hard_startstop_callback(void *arg, struct cfi_metatask *metatask)
1928 {
1929 	struct ctl_fe_ioctl_startstop_info *sd_info;
1930 
1931 	sd_info = (struct ctl_fe_ioctl_startstop_info *)arg;
1932 
1933 	sd_info->hs_info.status = metatask->status;
1934 	sd_info->hs_info.total_luns = metatask->taskinfo.startstop.total_luns;
1935 	sd_info->hs_info.luns_complete =
1936 		metatask->taskinfo.startstop.luns_complete;
1937 	sd_info->hs_info.luns_failed = metatask->taskinfo.startstop.luns_failed;
1938 
1939 	cv_broadcast(&sd_info->sem);
1940 }
1941 
1942 static void
1943 ctl_ioctl_bbrread_callback(void *arg, struct cfi_metatask *metatask)
1944 {
1945 	struct ctl_fe_ioctl_bbrread_info *fe_bbr_info;
1946 
1947 	fe_bbr_info = (struct ctl_fe_ioctl_bbrread_info *)arg;
1948 
1949 	mtx_lock(fe_bbr_info->lock);
1950 	fe_bbr_info->bbr_info->status = metatask->status;
1951 	fe_bbr_info->bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
1952 	fe_bbr_info->wakeup_done = 1;
1953 	mtx_unlock(fe_bbr_info->lock);
1954 
1955 	cv_broadcast(&fe_bbr_info->sem);
1956 }
1957 
1958 /*
1959  * Must be called with the ctl_lock held.
1960  * Returns 0 for success, errno for failure.
1961  */
1962 static int
1963 ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
1964 		   struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
1965 {
1966 	union ctl_io *io;
1967 	int retval;
1968 
1969 	retval = 0;
1970 
1971 	for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
1972 	     (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
1973 	     ooa_links)) {
1974 		struct ctl_ooa_entry *entry;
1975 
1976 		/*
1977 		 * If we've got more than we can fit, just count the
1978 		 * remaining entries.
1979 		 */
1980 		if (*cur_fill_num >= ooa_hdr->alloc_num)
1981 			continue;
1982 
1983 		entry = &kern_entries[*cur_fill_num];
1984 
1985 		entry->tag_num = io->scsiio.tag_num;
1986 		entry->lun_num = lun->lun;
1987 #ifdef CTL_TIME_IO
1988 		entry->start_bt = io->io_hdr.start_bt;
1989 #endif
1990 		bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
1991 		entry->cdb_len = io->scsiio.cdb_len;
1992 		if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
1993 			entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
1994 
1995 		if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
1996 			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
1997 
1998 		if (io->io_hdr.flags & CTL_FLAG_ABORT)
1999 			entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2000 
2001 		if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2002 			entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2003 
2004 		if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2005 			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2006 	}
2007 
2008 	return (retval);
2009 }
2010 
2011 static void *
2012 ctl_copyin_alloc(void *user_addr, int len, char *error_str,
2013 		 size_t error_str_len)
2014 {
2015 	void *kptr;
2016 
2017 	kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2018 
2019 	if (copyin(user_addr, kptr, len) != 0) {
2020 		snprintf(error_str, error_str_len, "Error copying %d bytes "
2021 			 "from user address %p to kernel address %p", len,
2022 			 user_addr, kptr);
2023 		free(kptr, M_CTL);
2024 		return (NULL);
2025 	}
2026 
2027 	return (kptr);
2028 }
2029 
2030 static void
2031 ctl_free_args(int num_be_args, struct ctl_be_arg *be_args)
2032 {
2033 	int i;
2034 
2035 	if (be_args == NULL)
2036 		return;
2037 
2038 	for (i = 0; i < num_be_args; i++) {
2039 		free(be_args[i].kname, M_CTL);
2040 		free(be_args[i].kvalue, M_CTL);
2041 	}
2042 
2043 	free(be_args, M_CTL);
2044 }
2045 
2046 static struct ctl_be_arg *
2047 ctl_copyin_args(int num_be_args, struct ctl_be_arg *be_args,
2048 		char *error_str, size_t error_str_len)
2049 {
2050 	struct ctl_be_arg *args;
2051 	int i;
2052 
2053 	args = ctl_copyin_alloc(be_args, num_be_args * sizeof(*be_args),
2054 				error_str, error_str_len);
2055 
2056 	if (args == NULL)
2057 		goto bailout;
2058 
2059 	for (i = 0; i < num_be_args; i++) {
2060 		args[i].kname = NULL;
2061 		args[i].kvalue = NULL;
2062 	}
2063 
2064 	for (i = 0; i < num_be_args; i++) {
2065 		uint8_t *tmpptr;
2066 
2067 		args[i].kname = ctl_copyin_alloc(args[i].name,
2068 			args[i].namelen, error_str, error_str_len);
2069 		if (args[i].kname == NULL)
2070 			goto bailout;
2071 
2072 		if (args[i].kname[args[i].namelen - 1] != '\0') {
2073 			snprintf(error_str, error_str_len, "Argument %d "
2074 				 "name is not NUL-terminated", i);
2075 			goto bailout;
2076 		}
2077 
2078 		args[i].kvalue = NULL;
2079 
2080 		tmpptr = ctl_copyin_alloc(args[i].value,
2081 			args[i].vallen, error_str, error_str_len);
2082 		if (tmpptr == NULL)
2083 			goto bailout;
2084 
2085 		args[i].kvalue = tmpptr;
2086 
2087 		if ((args[i].flags & CTL_BEARG_ASCII)
2088 		 && (tmpptr[args[i].vallen - 1] != '\0')) {
2089 			snprintf(error_str, error_str_len, "Argument %d "
2090 				 "value is not NUL-terminated", i);
2091 			goto bailout;
2092 		}
2093 	}
2094 
2095 	return (args);
2096 bailout:
2097 
2098 	ctl_free_args(num_be_args, args);
2099 
2100 	return (NULL);
2101 }
2102 
2103 /*
2104  * Escape characters that are illegal or not recommended in XML.
2105  */
2106 int
2107 ctl_sbuf_printf_esc(struct sbuf *sb, char *str)
2108 {
2109 	int retval;
2110 
2111 	retval = 0;
2112 
2113 	for (; *str; str++) {
2114 		switch (*str) {
2115 		case '&':
2116 			retval = sbuf_printf(sb, "&amp;");
2117 			break;
2118 		case '>':
2119 			retval = sbuf_printf(sb, "&gt;");
2120 			break;
2121 		case '<':
2122 			retval = sbuf_printf(sb, "&lt;");
2123 			break;
2124 		default:
2125 			retval = sbuf_putc(sb, *str);
2126 			break;
2127 		}
2128 
2129 		if (retval != 0)
2130 			break;
2131 
2132 	}
2133 
2134 	return (retval);
2135 }
2136 
2137 static int
2138 ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2139 	  struct thread *td)
2140 {
2141 	struct ctl_softc *softc;
2142 	int retval;
2143 
2144 	softc = control_softc;
2145 
2146 	retval = 0;
2147 
2148 	switch (cmd) {
2149 	case CTL_IO: {
2150 		union ctl_io *io;
2151 		void *pool_tmp;
2152 
2153 		/*
2154 		 * If we haven't been "enabled", don't allow any SCSI I/O
2155 		 * to this FETD.
2156 		 */
2157 		if ((softc->ioctl_info.flags & CTL_IOCTL_FLAG_ENABLED) == 0) {
2158 			retval = -EPERM;
2159 			break;
2160 		}
2161 
2162 		io = ctl_alloc_io(softc->ioctl_info.fe.ctl_pool_ref);
2163 		if (io == NULL) {
2164 			printf("ctl_ioctl: can't allocate ctl_io!\n");
2165 			retval = -ENOSPC;
2166 			break;
2167 		}
2168 
2169 		/*
2170 		 * Need to save the pool reference so it doesn't get
2171 		 * spammed by the user's ctl_io.
2172 		 */
2173 		pool_tmp = io->io_hdr.pool;
2174 
2175 		memcpy(io, (void *)addr, sizeof(*io));
2176 
2177 		io->io_hdr.pool = pool_tmp;
2178 		/*
2179 		 * No status yet, so make sure the status is set properly.
2180 		 */
2181 		io->io_hdr.status = CTL_STATUS_NONE;
2182 
2183 		/*
2184 		 * The user sets the initiator ID, target and LUN IDs.
2185 		 */
2186 		io->io_hdr.nexus.targ_port = softc->ioctl_info.fe.targ_port;
2187 		io->io_hdr.flags |= CTL_FLAG_USER_REQ;
2188 		if ((io->io_hdr.io_type == CTL_IO_SCSI)
2189 		 && (io->scsiio.tag_type != CTL_TAG_UNTAGGED))
2190 			io->scsiio.tag_num = softc->ioctl_info.cur_tag_num++;
2191 
2192 		retval = ctl_ioctl_submit_wait(io);
2193 
2194 		if (retval != 0) {
2195 			ctl_free_io(io);
2196 			break;
2197 		}
2198 
2199 		memcpy((void *)addr, io, sizeof(*io));
2200 
2201 		/* return this to our pool */
2202 		ctl_free_io(io);
2203 
2204 		break;
2205 	}
2206 	case CTL_ENABLE_PORT:
2207 	case CTL_DISABLE_PORT:
2208 	case CTL_SET_PORT_WWNS: {
2209 		struct ctl_frontend *fe;
2210 		struct ctl_port_entry *entry;
2211 
2212 		entry = (struct ctl_port_entry *)addr;
2213 
2214 		mtx_lock(&softc->ctl_lock);
2215 		STAILQ_FOREACH(fe, &softc->fe_list, links) {
2216 			int action, done;
2217 
2218 			action = 0;
2219 			done = 0;
2220 
2221 			if ((entry->port_type == CTL_PORT_NONE)
2222 			 && (entry->targ_port == fe->targ_port)) {
2223 				/*
2224 				 * If the user only wants to enable or
2225 				 * disable or set WWNs on a specific port,
2226 				 * do the operation and we're done.
2227 				 */
2228 				action = 1;
2229 				done = 1;
2230 			} else if (entry->port_type & fe->port_type) {
2231 				/*
2232 				 * Compare the user's type mask with the
2233 				 * particular frontend type to see if we
2234 				 * have a match.
2235 				 */
2236 				action = 1;
2237 				done = 0;
2238 
2239 				/*
2240 				 * Make sure the user isn't trying to set
2241 				 * WWNs on multiple ports at the same time.
2242 				 */
2243 				if (cmd == CTL_SET_PORT_WWNS) {
2244 					printf("%s: Can't set WWNs on "
2245 					       "multiple ports\n", __func__);
2246 					retval = EINVAL;
2247 					break;
2248 				}
2249 			}
2250 			if (action != 0) {
2251 				/*
2252 				 * XXX KDM we have to drop the lock here,
2253 				 * because the online/offline operations
2254 				 * can potentially block.  We need to
2255 				 * reference count the frontends so they
2256 				 * can't go away,
2257 				 */
2258 				mtx_unlock(&softc->ctl_lock);
2259 
2260 				if (cmd == CTL_ENABLE_PORT)
2261 					ctl_frontend_online(fe);
2262 				else if (cmd == CTL_DISABLE_PORT)
2263 					ctl_frontend_offline(fe);
2264 
2265 				mtx_lock(&softc->ctl_lock);
2266 
2267 				if (cmd == CTL_SET_PORT_WWNS)
2268 					ctl_frontend_set_wwns(fe,
2269 					    (entry->flags & CTL_PORT_WWNN_VALID) ?
2270 					    1 : 0, entry->wwnn,
2271 					    (entry->flags & CTL_PORT_WWPN_VALID) ?
2272 					    1 : 0, entry->wwpn);
2273 			}
2274 			if (done != 0)
2275 				break;
2276 		}
2277 		mtx_unlock(&softc->ctl_lock);
2278 		break;
2279 	}
2280 	case CTL_GET_PORT_LIST: {
2281 		struct ctl_frontend *fe;
2282 		struct ctl_port_list *list;
2283 		int i;
2284 
2285 		list = (struct ctl_port_list *)addr;
2286 
2287 		if (list->alloc_len != (list->alloc_num *
2288 		    sizeof(struct ctl_port_entry))) {
2289 			printf("%s: CTL_GET_PORT_LIST: alloc_len %u != "
2290 			       "alloc_num %u * sizeof(struct ctl_port_entry) "
2291 			       "%zu\n", __func__, list->alloc_len,
2292 			       list->alloc_num, sizeof(struct ctl_port_entry));
2293 			retval = EINVAL;
2294 			break;
2295 		}
2296 		list->fill_len = 0;
2297 		list->fill_num = 0;
2298 		list->dropped_num = 0;
2299 		i = 0;
2300 		mtx_lock(&softc->ctl_lock);
2301 		STAILQ_FOREACH(fe, &softc->fe_list, links) {
2302 			struct ctl_port_entry entry, *list_entry;
2303 
2304 			if (list->fill_num >= list->alloc_num) {
2305 				list->dropped_num++;
2306 				continue;
2307 			}
2308 
2309 			entry.port_type = fe->port_type;
2310 			strlcpy(entry.port_name, fe->port_name,
2311 				sizeof(entry.port_name));
2312 			entry.targ_port = fe->targ_port;
2313 			entry.physical_port = fe->physical_port;
2314 			entry.virtual_port = fe->virtual_port;
2315 			entry.wwnn = fe->wwnn;
2316 			entry.wwpn = fe->wwpn;
2317 			if (fe->status & CTL_PORT_STATUS_ONLINE)
2318 				entry.online = 1;
2319 			else
2320 				entry.online = 0;
2321 
2322 			list_entry = &list->entries[i];
2323 
2324 			retval = copyout(&entry, list_entry, sizeof(entry));
2325 			if (retval != 0) {
2326 				printf("%s: CTL_GET_PORT_LIST: copyout "
2327 				       "returned %d\n", __func__, retval);
2328 				break;
2329 			}
2330 			i++;
2331 			list->fill_num++;
2332 			list->fill_len += sizeof(entry);
2333 		}
2334 		mtx_unlock(&softc->ctl_lock);
2335 
2336 		/*
2337 		 * If this is non-zero, we had a copyout fault, so there's
2338 		 * probably no point in attempting to set the status inside
2339 		 * the structure.
2340 		 */
2341 		if (retval != 0)
2342 			break;
2343 
2344 		if (list->dropped_num > 0)
2345 			list->status = CTL_PORT_LIST_NEED_MORE_SPACE;
2346 		else
2347 			list->status = CTL_PORT_LIST_OK;
2348 		break;
2349 	}
2350 	case CTL_DUMP_OOA: {
2351 		struct ctl_lun *lun;
2352 		union ctl_io *io;
2353 		char printbuf[128];
2354 		struct sbuf sb;
2355 
2356 		mtx_lock(&softc->ctl_lock);
2357 		printf("Dumping OOA queues:\n");
2358 		STAILQ_FOREACH(lun, &softc->lun_list, links) {
2359 			for (io = (union ctl_io *)TAILQ_FIRST(
2360 			     &lun->ooa_queue); io != NULL;
2361 			     io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2362 			     ooa_links)) {
2363 				sbuf_new(&sb, printbuf, sizeof(printbuf),
2364 					 SBUF_FIXEDLEN);
2365 				sbuf_printf(&sb, "LUN %jd tag 0x%04x%s%s%s%s: ",
2366 					    (intmax_t)lun->lun,
2367 					    io->scsiio.tag_num,
2368 					    (io->io_hdr.flags &
2369 					    CTL_FLAG_BLOCKED) ? "" : " BLOCKED",
2370 					    (io->io_hdr.flags &
2371 					    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
2372 					    (io->io_hdr.flags &
2373 					    CTL_FLAG_ABORT) ? " ABORT" : "",
2374 			                    (io->io_hdr.flags &
2375 		                        CTL_FLAG_IS_WAS_ON_RTR) ? " RTR" : "");
2376 				ctl_scsi_command_string(&io->scsiio, NULL, &sb);
2377 				sbuf_finish(&sb);
2378 				printf("%s\n", sbuf_data(&sb));
2379 			}
2380 		}
2381 		printf("OOA queues dump done\n");
2382 		mtx_unlock(&softc->ctl_lock);
2383 		break;
2384 	}
2385 	case CTL_GET_OOA: {
2386 		struct ctl_lun *lun;
2387 		struct ctl_ooa *ooa_hdr;
2388 		struct ctl_ooa_entry *entries;
2389 		uint32_t cur_fill_num;
2390 
2391 		ooa_hdr = (struct ctl_ooa *)addr;
2392 
2393 		if ((ooa_hdr->alloc_len == 0)
2394 		 || (ooa_hdr->alloc_num == 0)) {
2395 			printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2396 			       "must be non-zero\n", __func__,
2397 			       ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2398 			retval = EINVAL;
2399 			break;
2400 		}
2401 
2402 		if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2403 		    sizeof(struct ctl_ooa_entry))) {
2404 			printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2405 			       "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2406 			       __func__, ooa_hdr->alloc_len,
2407 			       ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2408 			retval = EINVAL;
2409 			break;
2410 		}
2411 
2412 		entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2413 		if (entries == NULL) {
2414 			printf("%s: could not allocate %d bytes for OOA "
2415 			       "dump\n", __func__, ooa_hdr->alloc_len);
2416 			retval = ENOMEM;
2417 			break;
2418 		}
2419 
2420 		mtx_lock(&softc->ctl_lock);
2421 		if (((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0)
2422 		 && ((ooa_hdr->lun_num > CTL_MAX_LUNS)
2423 		  || (softc->ctl_luns[ooa_hdr->lun_num] == NULL))) {
2424 			mtx_unlock(&softc->ctl_lock);
2425 			free(entries, M_CTL);
2426 			printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2427 			       __func__, (uintmax_t)ooa_hdr->lun_num);
2428 			retval = EINVAL;
2429 			break;
2430 		}
2431 
2432 		cur_fill_num = 0;
2433 
2434 		if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2435 			STAILQ_FOREACH(lun, &softc->lun_list, links) {
2436 				retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2437 					ooa_hdr, entries);
2438 				if (retval != 0)
2439 					break;
2440 			}
2441 			if (retval != 0) {
2442 				mtx_unlock(&softc->ctl_lock);
2443 				free(entries, M_CTL);
2444 				break;
2445 			}
2446 		} else {
2447 			lun = softc->ctl_luns[ooa_hdr->lun_num];
2448 
2449 			retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,ooa_hdr,
2450 						    entries);
2451 		}
2452 		mtx_unlock(&softc->ctl_lock);
2453 
2454 		ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2455 		ooa_hdr->fill_len = ooa_hdr->fill_num *
2456 			sizeof(struct ctl_ooa_entry);
2457 		retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2458 		if (retval != 0) {
2459 			printf("%s: error copying out %d bytes for OOA dump\n",
2460 			       __func__, ooa_hdr->fill_len);
2461 		}
2462 
2463 		getbintime(&ooa_hdr->cur_bt);
2464 
2465 		if (cur_fill_num > ooa_hdr->alloc_num) {
2466 			ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2467 			ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2468 		} else {
2469 			ooa_hdr->dropped_num = 0;
2470 			ooa_hdr->status = CTL_OOA_OK;
2471 		}
2472 
2473 		free(entries, M_CTL);
2474 		break;
2475 	}
2476 	case CTL_CHECK_OOA: {
2477 		union ctl_io *io;
2478 		struct ctl_lun *lun;
2479 		struct ctl_ooa_info *ooa_info;
2480 
2481 
2482 		ooa_info = (struct ctl_ooa_info *)addr;
2483 
2484 		if (ooa_info->lun_id >= CTL_MAX_LUNS) {
2485 			ooa_info->status = CTL_OOA_INVALID_LUN;
2486 			break;
2487 		}
2488 		mtx_lock(&softc->ctl_lock);
2489 		lun = softc->ctl_luns[ooa_info->lun_id];
2490 		if (lun == NULL) {
2491 			mtx_unlock(&softc->ctl_lock);
2492 			ooa_info->status = CTL_OOA_INVALID_LUN;
2493 			break;
2494 		}
2495 
2496 		ooa_info->num_entries = 0;
2497 		for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
2498 		     io != NULL; io = (union ctl_io *)TAILQ_NEXT(
2499 		     &io->io_hdr, ooa_links)) {
2500 			ooa_info->num_entries++;
2501 		}
2502 
2503 		mtx_unlock(&softc->ctl_lock);
2504 		ooa_info->status = CTL_OOA_SUCCESS;
2505 
2506 		break;
2507 	}
2508 	case CTL_HARD_START:
2509 	case CTL_HARD_STOP: {
2510 		struct ctl_fe_ioctl_startstop_info ss_info;
2511 		struct cfi_metatask *metatask;
2512 		struct mtx hs_mtx;
2513 
2514 		mtx_init(&hs_mtx, "HS Mutex", NULL, MTX_DEF);
2515 
2516 		cv_init(&ss_info.sem, "hard start/stop cv" );
2517 
2518 		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2519 		if (metatask == NULL) {
2520 			retval = ENOMEM;
2521 			mtx_destroy(&hs_mtx);
2522 			break;
2523 		}
2524 
2525 		if (cmd == CTL_HARD_START)
2526 			metatask->tasktype = CFI_TASK_STARTUP;
2527 		else
2528 			metatask->tasktype = CFI_TASK_SHUTDOWN;
2529 
2530 		metatask->callback = ctl_ioctl_hard_startstop_callback;
2531 		metatask->callback_arg = &ss_info;
2532 
2533 		cfi_action(metatask);
2534 
2535 		/* Wait for the callback */
2536 		mtx_lock(&hs_mtx);
2537 		cv_wait_sig(&ss_info.sem, &hs_mtx);
2538 		mtx_unlock(&hs_mtx);
2539 
2540 		/*
2541 		 * All information has been copied from the metatask by the
2542 		 * time cv_broadcast() is called, so we free the metatask here.
2543 		 */
2544 		cfi_free_metatask(metatask);
2545 
2546 		memcpy((void *)addr, &ss_info.hs_info, sizeof(ss_info.hs_info));
2547 
2548 		mtx_destroy(&hs_mtx);
2549 		break;
2550 	}
2551 	case CTL_BBRREAD: {
2552 		struct ctl_bbrread_info *bbr_info;
2553 		struct ctl_fe_ioctl_bbrread_info fe_bbr_info;
2554 		struct mtx bbr_mtx;
2555 		struct cfi_metatask *metatask;
2556 
2557 		bbr_info = (struct ctl_bbrread_info *)addr;
2558 
2559 		bzero(&fe_bbr_info, sizeof(fe_bbr_info));
2560 
2561 		bzero(&bbr_mtx, sizeof(bbr_mtx));
2562 		mtx_init(&bbr_mtx, "BBR Mutex", NULL, MTX_DEF);
2563 
2564 		fe_bbr_info.bbr_info = bbr_info;
2565 		fe_bbr_info.lock = &bbr_mtx;
2566 
2567 		cv_init(&fe_bbr_info.sem, "BBR read cv");
2568 		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2569 
2570 		if (metatask == NULL) {
2571 			mtx_destroy(&bbr_mtx);
2572 			cv_destroy(&fe_bbr_info.sem);
2573 			retval = ENOMEM;
2574 			break;
2575 		}
2576 		metatask->tasktype = CFI_TASK_BBRREAD;
2577 		metatask->callback = ctl_ioctl_bbrread_callback;
2578 		metatask->callback_arg = &fe_bbr_info;
2579 		metatask->taskinfo.bbrread.lun_num = bbr_info->lun_num;
2580 		metatask->taskinfo.bbrread.lba = bbr_info->lba;
2581 		metatask->taskinfo.bbrread.len = bbr_info->len;
2582 
2583 		cfi_action(metatask);
2584 
2585 		mtx_lock(&bbr_mtx);
2586 		while (fe_bbr_info.wakeup_done == 0)
2587 			cv_wait_sig(&fe_bbr_info.sem, &bbr_mtx);
2588 		mtx_unlock(&bbr_mtx);
2589 
2590 		bbr_info->status = metatask->status;
2591 		bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2592 		bbr_info->scsi_status = metatask->taskinfo.bbrread.scsi_status;
2593 		memcpy(&bbr_info->sense_data,
2594 		       &metatask->taskinfo.bbrread.sense_data,
2595 		       ctl_min(sizeof(bbr_info->sense_data),
2596 			       sizeof(metatask->taskinfo.bbrread.sense_data)));
2597 
2598 		cfi_free_metatask(metatask);
2599 
2600 		mtx_destroy(&bbr_mtx);
2601 		cv_destroy(&fe_bbr_info.sem);
2602 
2603 		break;
2604 	}
2605 	case CTL_DELAY_IO: {
2606 		struct ctl_io_delay_info *delay_info;
2607 #ifdef CTL_IO_DELAY
2608 		struct ctl_lun *lun;
2609 #endif /* CTL_IO_DELAY */
2610 
2611 		delay_info = (struct ctl_io_delay_info *)addr;
2612 
2613 #ifdef CTL_IO_DELAY
2614 		mtx_lock(&softc->ctl_lock);
2615 
2616 		if ((delay_info->lun_id > CTL_MAX_LUNS)
2617 		 || (softc->ctl_luns[delay_info->lun_id] == NULL)) {
2618 			delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2619 		} else {
2620 			lun = softc->ctl_luns[delay_info->lun_id];
2621 
2622 			delay_info->status = CTL_DELAY_STATUS_OK;
2623 
2624 			switch (delay_info->delay_type) {
2625 			case CTL_DELAY_TYPE_CONT:
2626 				break;
2627 			case CTL_DELAY_TYPE_ONESHOT:
2628 				break;
2629 			default:
2630 				delay_info->status =
2631 					CTL_DELAY_STATUS_INVALID_TYPE;
2632 				break;
2633 			}
2634 
2635 			switch (delay_info->delay_loc) {
2636 			case CTL_DELAY_LOC_DATAMOVE:
2637 				lun->delay_info.datamove_type =
2638 					delay_info->delay_type;
2639 				lun->delay_info.datamove_delay =
2640 					delay_info->delay_secs;
2641 				break;
2642 			case CTL_DELAY_LOC_DONE:
2643 				lun->delay_info.done_type =
2644 					delay_info->delay_type;
2645 				lun->delay_info.done_delay =
2646 					delay_info->delay_secs;
2647 				break;
2648 			default:
2649 				delay_info->status =
2650 					CTL_DELAY_STATUS_INVALID_LOC;
2651 				break;
2652 			}
2653 		}
2654 
2655 		mtx_unlock(&softc->ctl_lock);
2656 #else
2657 		delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2658 #endif /* CTL_IO_DELAY */
2659 		break;
2660 	}
2661 	case CTL_REALSYNC_SET: {
2662 		int *syncstate;
2663 
2664 		syncstate = (int *)addr;
2665 
2666 		mtx_lock(&softc->ctl_lock);
2667 		switch (*syncstate) {
2668 		case 0:
2669 			softc->flags &= ~CTL_FLAG_REAL_SYNC;
2670 			break;
2671 		case 1:
2672 			softc->flags |= CTL_FLAG_REAL_SYNC;
2673 			break;
2674 		default:
2675 			retval = -EINVAL;
2676 			break;
2677 		}
2678 		mtx_unlock(&softc->ctl_lock);
2679 		break;
2680 	}
2681 	case CTL_REALSYNC_GET: {
2682 		int *syncstate;
2683 
2684 		syncstate = (int*)addr;
2685 
2686 		mtx_lock(&softc->ctl_lock);
2687 		if (softc->flags & CTL_FLAG_REAL_SYNC)
2688 			*syncstate = 1;
2689 		else
2690 			*syncstate = 0;
2691 		mtx_unlock(&softc->ctl_lock);
2692 
2693 		break;
2694 	}
2695 	case CTL_SETSYNC:
2696 	case CTL_GETSYNC: {
2697 		struct ctl_sync_info *sync_info;
2698 		struct ctl_lun *lun;
2699 
2700 		sync_info = (struct ctl_sync_info *)addr;
2701 
2702 		mtx_lock(&softc->ctl_lock);
2703 		lun = softc->ctl_luns[sync_info->lun_id];
2704 		if (lun == NULL) {
2705 			mtx_unlock(&softc->ctl_lock);
2706 			sync_info->status = CTL_GS_SYNC_NO_LUN;
2707 		}
2708 		/*
2709 		 * Get or set the sync interval.  We're not bounds checking
2710 		 * in the set case, hopefully the user won't do something
2711 		 * silly.
2712 		 */
2713 		if (cmd == CTL_GETSYNC)
2714 			sync_info->sync_interval = lun->sync_interval;
2715 		else
2716 			lun->sync_interval = sync_info->sync_interval;
2717 
2718 		mtx_unlock(&softc->ctl_lock);
2719 
2720 		sync_info->status = CTL_GS_SYNC_OK;
2721 
2722 		break;
2723 	}
2724 	case CTL_GETSTATS: {
2725 		struct ctl_stats *stats;
2726 		struct ctl_lun *lun;
2727 		int i;
2728 
2729 		stats = (struct ctl_stats *)addr;
2730 
2731 		if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) >
2732 		     stats->alloc_len) {
2733 			stats->status = CTL_SS_NEED_MORE_SPACE;
2734 			stats->num_luns = softc->num_luns;
2735 			break;
2736 		}
2737 		/*
2738 		 * XXX KDM no locking here.  If the LUN list changes,
2739 		 * things can blow up.
2740 		 */
2741 		for (i = 0, lun = STAILQ_FIRST(&softc->lun_list); lun != NULL;
2742 		     i++, lun = STAILQ_NEXT(lun, links)) {
2743 			retval = copyout(&lun->stats, &stats->lun_stats[i],
2744 					 sizeof(lun->stats));
2745 			if (retval != 0)
2746 				break;
2747 		}
2748 		stats->num_luns = softc->num_luns;
2749 		stats->fill_len = sizeof(struct ctl_lun_io_stats) *
2750 				 softc->num_luns;
2751 		stats->status = CTL_SS_OK;
2752 #ifdef CTL_TIME_IO
2753 		stats->flags = CTL_STATS_FLAG_TIME_VALID;
2754 #else
2755 		stats->flags = CTL_STATS_FLAG_NONE;
2756 #endif
2757 		getnanouptime(&stats->timestamp);
2758 		break;
2759 	}
2760 	case CTL_ERROR_INJECT: {
2761 		struct ctl_error_desc *err_desc, *new_err_desc;
2762 		struct ctl_lun *lun;
2763 
2764 		err_desc = (struct ctl_error_desc *)addr;
2765 
2766 		new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2767 				      M_WAITOK | M_ZERO);
2768 		bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2769 
2770 		mtx_lock(&softc->ctl_lock);
2771 		lun = softc->ctl_luns[err_desc->lun_id];
2772 		if (lun == NULL) {
2773 			mtx_unlock(&softc->ctl_lock);
2774 			printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2775 			       __func__, (uintmax_t)err_desc->lun_id);
2776 			retval = EINVAL;
2777 			break;
2778 		}
2779 
2780 		/*
2781 		 * We could do some checking here to verify the validity
2782 		 * of the request, but given the complexity of error
2783 		 * injection requests, the checking logic would be fairly
2784 		 * complex.
2785 		 *
2786 		 * For now, if the request is invalid, it just won't get
2787 		 * executed and might get deleted.
2788 		 */
2789 		STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2790 
2791 		/*
2792 		 * XXX KDM check to make sure the serial number is unique,
2793 		 * in case we somehow manage to wrap.  That shouldn't
2794 		 * happen for a very long time, but it's the right thing to
2795 		 * do.
2796 		 */
2797 		new_err_desc->serial = lun->error_serial;
2798 		err_desc->serial = lun->error_serial;
2799 		lun->error_serial++;
2800 
2801 		mtx_unlock(&softc->ctl_lock);
2802 		break;
2803 	}
2804 	case CTL_ERROR_INJECT_DELETE: {
2805 		struct ctl_error_desc *delete_desc, *desc, *desc2;
2806 		struct ctl_lun *lun;
2807 		int delete_done;
2808 
2809 		delete_desc = (struct ctl_error_desc *)addr;
2810 		delete_done = 0;
2811 
2812 		mtx_lock(&softc->ctl_lock);
2813 		lun = softc->ctl_luns[delete_desc->lun_id];
2814 		if (lun == NULL) {
2815 			mtx_unlock(&softc->ctl_lock);
2816 			printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2817 			       __func__, (uintmax_t)delete_desc->lun_id);
2818 			retval = EINVAL;
2819 			break;
2820 		}
2821 		STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2822 			if (desc->serial != delete_desc->serial)
2823 				continue;
2824 
2825 			STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2826 				      links);
2827 			free(desc, M_CTL);
2828 			delete_done = 1;
2829 		}
2830 		mtx_unlock(&softc->ctl_lock);
2831 		if (delete_done == 0) {
2832 			printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2833 			       "error serial %ju on LUN %u\n", __func__,
2834 			       delete_desc->serial, delete_desc->lun_id);
2835 			retval = EINVAL;
2836 			break;
2837 		}
2838 		break;
2839 	}
2840 	case CTL_DUMP_STRUCTS: {
2841 		int i, j, k;
2842 		struct ctl_frontend *fe;
2843 
2844 		printf("CTL IID to WWPN map start:\n");
2845 		for (i = 0; i < CTL_MAX_PORTS; i++) {
2846 			for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
2847 				if (softc->wwpn_iid[i][j].in_use == 0)
2848 					continue;
2849 
2850 				printf("port %d iid %u WWPN %#jx\n",
2851 				       softc->wwpn_iid[i][j].port,
2852 				       softc->wwpn_iid[i][j].iid,
2853 				       (uintmax_t)softc->wwpn_iid[i][j].wwpn);
2854 			}
2855 		}
2856 		printf("CTL IID to WWPN map end\n");
2857 		printf("CTL Persistent Reservation information start:\n");
2858 		for (i = 0; i < CTL_MAX_LUNS; i++) {
2859 			struct ctl_lun *lun;
2860 
2861 			lun = softc->ctl_luns[i];
2862 
2863 			if ((lun == NULL)
2864 			 || ((lun->flags & CTL_LUN_DISABLED) != 0))
2865 				continue;
2866 
2867 			for (j = 0; j < (CTL_MAX_PORTS * 2); j++) {
2868 				for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2869 					if (lun->per_res[j+k].registered == 0)
2870 						continue;
2871 					printf("LUN %d port %d iid %d key "
2872 					       "%#jx\n", i, j, k,
2873 					       (uintmax_t)scsi_8btou64(
2874 					       lun->per_res[j+k].res_key.key));
2875 				}
2876 			}
2877 		}
2878 		printf("CTL Persistent Reservation information end\n");
2879 		printf("CTL Frontends:\n");
2880 		/*
2881 		 * XXX KDM calling this without a lock.  We'd likely want
2882 		 * to drop the lock before calling the frontend's dump
2883 		 * routine anyway.
2884 		 */
2885 		STAILQ_FOREACH(fe, &softc->fe_list, links) {
2886 			printf("Frontend %s Type %u pport %d vport %d WWNN "
2887 			       "%#jx WWPN %#jx\n", fe->port_name, fe->port_type,
2888 			       fe->physical_port, fe->virtual_port,
2889 			       (uintmax_t)fe->wwnn, (uintmax_t)fe->wwpn);
2890 
2891 			/*
2892 			 * Frontends are not required to support the dump
2893 			 * routine.
2894 			 */
2895 			if (fe->fe_dump == NULL)
2896 				continue;
2897 
2898 			fe->fe_dump();
2899 		}
2900 		printf("CTL Frontend information end\n");
2901 		break;
2902 	}
2903 	case CTL_LUN_REQ: {
2904 		struct ctl_lun_req *lun_req;
2905 		struct ctl_backend_driver *backend;
2906 
2907 		lun_req = (struct ctl_lun_req *)addr;
2908 
2909 		backend = ctl_backend_find(lun_req->backend);
2910 		if (backend == NULL) {
2911 			lun_req->status = CTL_LUN_ERROR;
2912 			snprintf(lun_req->error_str,
2913 				 sizeof(lun_req->error_str),
2914 				 "Backend \"%s\" not found.",
2915 				 lun_req->backend);
2916 			break;
2917 		}
2918 		if (lun_req->num_be_args > 0) {
2919 			lun_req->kern_be_args = ctl_copyin_args(
2920 				lun_req->num_be_args,
2921 				lun_req->be_args,
2922 				lun_req->error_str,
2923 				sizeof(lun_req->error_str));
2924 			if (lun_req->kern_be_args == NULL) {
2925 				lun_req->status = CTL_LUN_ERROR;
2926 				break;
2927 			}
2928 		}
2929 
2930 		retval = backend->ioctl(dev, cmd, addr, flag, td);
2931 
2932 		if (lun_req->num_be_args > 0) {
2933 			ctl_free_args(lun_req->num_be_args,
2934 				      lun_req->kern_be_args);
2935 		}
2936 		break;
2937 	}
2938 	case CTL_LUN_LIST: {
2939 		struct sbuf *sb;
2940 		struct ctl_lun *lun;
2941 		struct ctl_lun_list *list;
2942 
2943 		list = (struct ctl_lun_list *)addr;
2944 
2945 		/*
2946 		 * Allocate a fixed length sbuf here, based on the length
2947 		 * of the user's buffer.  We could allocate an auto-extending
2948 		 * buffer, and then tell the user how much larger our
2949 		 * amount of data is than his buffer, but that presents
2950 		 * some problems:
2951 		 *
2952 		 * 1.  The sbuf(9) routines use a blocking malloc, and so
2953 		 *     we can't hold a lock while calling them with an
2954 		 *     auto-extending buffer.
2955  		 *
2956 		 * 2.  There is not currently a LUN reference counting
2957 		 *     mechanism, outside of outstanding transactions on
2958 		 *     the LUN's OOA queue.  So a LUN could go away on us
2959 		 *     while we're getting the LUN number, backend-specific
2960 		 *     information, etc.  Thus, given the way things
2961 		 *     currently work, we need to hold the CTL lock while
2962 		 *     grabbing LUN information.
2963 		 *
2964 		 * So, from the user's standpoint, the best thing to do is
2965 		 * allocate what he thinks is a reasonable buffer length,
2966 		 * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
2967 		 * double the buffer length and try again.  (And repeat
2968 		 * that until he succeeds.)
2969 		 */
2970 		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
2971 		if (sb == NULL) {
2972 			list->status = CTL_LUN_LIST_ERROR;
2973 			snprintf(list->error_str, sizeof(list->error_str),
2974 				 "Unable to allocate %d bytes for LUN list",
2975 				 list->alloc_len);
2976 			break;
2977 		}
2978 
2979 		sbuf_printf(sb, "<ctllunlist>\n");
2980 
2981 		mtx_lock(&softc->ctl_lock);
2982 
2983 		STAILQ_FOREACH(lun, &softc->lun_list, links) {
2984 			retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
2985 					     (uintmax_t)lun->lun);
2986 
2987 			/*
2988 			 * Bail out as soon as we see that we've overfilled
2989 			 * the buffer.
2990 			 */
2991 			if (retval != 0)
2992 				break;
2993 
2994 			retval = sbuf_printf(sb, "<backend_type>%s"
2995 					     "</backend_type>\n",
2996 					     (lun->backend == NULL) ?  "none" :
2997 					     lun->backend->name);
2998 
2999 			if (retval != 0)
3000 				break;
3001 
3002 			retval = sbuf_printf(sb, "<lun_type>%d</lun_type>\n",
3003 					     lun->be_lun->lun_type);
3004 
3005 			if (retval != 0)
3006 				break;
3007 
3008 			if (lun->backend == NULL) {
3009 				retval = sbuf_printf(sb, "</lun>\n");
3010 				if (retval != 0)
3011 					break;
3012 				continue;
3013 			}
3014 
3015 			retval = sbuf_printf(sb, "<size>%ju</size>\n",
3016 					     (lun->be_lun->maxlba > 0) ?
3017 					     lun->be_lun->maxlba + 1 : 0);
3018 
3019 			if (retval != 0)
3020 				break;
3021 
3022 			retval = sbuf_printf(sb, "<blocksize>%u</blocksize>\n",
3023 					     lun->be_lun->blocksize);
3024 
3025 			if (retval != 0)
3026 				break;
3027 
3028 			retval = sbuf_printf(sb, "<serial_number>");
3029 
3030 			if (retval != 0)
3031 				break;
3032 
3033 			retval = ctl_sbuf_printf_esc(sb,
3034 						     lun->be_lun->serial_num);
3035 
3036 			if (retval != 0)
3037 				break;
3038 
3039 			retval = sbuf_printf(sb, "</serial_number>\n");
3040 
3041 			if (retval != 0)
3042 				break;
3043 
3044 			retval = sbuf_printf(sb, "<device_id>");
3045 
3046 			if (retval != 0)
3047 				break;
3048 
3049 			retval = ctl_sbuf_printf_esc(sb,lun->be_lun->device_id);
3050 
3051 			if (retval != 0)
3052 				break;
3053 
3054 			retval = sbuf_printf(sb, "</device_id>\n");
3055 
3056 			if (retval != 0)
3057 				break;
3058 
3059 			if (lun->backend->lun_info == NULL) {
3060 				retval = sbuf_printf(sb, "</lun>\n");
3061 				if (retval != 0)
3062 					break;
3063 				continue;
3064 			}
3065 
3066 			retval =lun->backend->lun_info(lun->be_lun->be_lun, sb);
3067 
3068 			if (retval != 0)
3069 				break;
3070 
3071 			retval = sbuf_printf(sb, "</lun>\n");
3072 
3073 			if (retval != 0)
3074 				break;
3075 		}
3076 		mtx_unlock(&softc->ctl_lock);
3077 
3078 		if ((retval != 0)
3079 		 || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3080 			retval = 0;
3081 			sbuf_delete(sb);
3082 			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3083 			snprintf(list->error_str, sizeof(list->error_str),
3084 				 "Out of space, %d bytes is too small",
3085 				 list->alloc_len);
3086 			break;
3087 		}
3088 
3089 		sbuf_finish(sb);
3090 
3091 		retval = copyout(sbuf_data(sb), list->lun_xml,
3092 				 sbuf_len(sb) + 1);
3093 
3094 		list->fill_len = sbuf_len(sb) + 1;
3095 		list->status = CTL_LUN_LIST_OK;
3096 		sbuf_delete(sb);
3097 		break;
3098 	}
3099 	default: {
3100 		/* XXX KDM should we fix this? */
3101 #if 0
3102 		struct ctl_backend_driver *backend;
3103 		unsigned int type;
3104 		int found;
3105 
3106 		found = 0;
3107 
3108 		/*
3109 		 * We encode the backend type as the ioctl type for backend
3110 		 * ioctls.  So parse it out here, and then search for a
3111 		 * backend of this type.
3112 		 */
3113 		type = _IOC_TYPE(cmd);
3114 
3115 		STAILQ_FOREACH(backend, &softc->be_list, links) {
3116 			if (backend->type == type) {
3117 				found = 1;
3118 				break;
3119 			}
3120 		}
3121 		if (found == 0) {
3122 			printf("ctl: unknown ioctl command %#lx or backend "
3123 			       "%d\n", cmd, type);
3124 			retval = -EINVAL;
3125 			break;
3126 		}
3127 		retval = backend->ioctl(dev, cmd, addr, flag, td);
3128 #endif
3129 		retval = ENOTTY;
3130 		break;
3131 	}
3132 	}
3133 	return (retval);
3134 }
3135 
3136 uint32_t
3137 ctl_get_initindex(struct ctl_nexus *nexus)
3138 {
3139 	if (nexus->targ_port < CTL_MAX_PORTS)
3140 		return (nexus->initid.id +
3141 			(nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3142 	else
3143 		return (nexus->initid.id +
3144 		       ((nexus->targ_port - CTL_MAX_PORTS) *
3145 			CTL_MAX_INIT_PER_PORT));
3146 }
3147 
3148 uint32_t
3149 ctl_get_resindex(struct ctl_nexus *nexus)
3150 {
3151 	return (nexus->initid.id + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3152 }
3153 
3154 uint32_t
3155 ctl_port_idx(int port_num)
3156 {
3157 	if (port_num < CTL_MAX_PORTS)
3158 		return(port_num);
3159 	else
3160 		return(port_num - CTL_MAX_PORTS);
3161 }
3162 
3163 /*
3164  * Note:  This only works for bitmask sizes that are at least 32 bits, and
3165  * that are a power of 2.
3166  */
3167 int
3168 ctl_ffz(uint32_t *mask, uint32_t size)
3169 {
3170 	uint32_t num_chunks, num_pieces;
3171 	int i, j;
3172 
3173 	num_chunks = (size >> 5);
3174 	if (num_chunks == 0)
3175 		num_chunks++;
3176 	num_pieces = ctl_min((sizeof(uint32_t) * 8), size);
3177 
3178 	for (i = 0; i < num_chunks; i++) {
3179 		for (j = 0; j < num_pieces; j++) {
3180 			if ((mask[i] & (1 << j)) == 0)
3181 				return ((i << 5) + j);
3182 		}
3183 	}
3184 
3185 	return (-1);
3186 }
3187 
3188 int
3189 ctl_set_mask(uint32_t *mask, uint32_t bit)
3190 {
3191 	uint32_t chunk, piece;
3192 
3193 	chunk = bit >> 5;
3194 	piece = bit % (sizeof(uint32_t) * 8);
3195 
3196 	if ((mask[chunk] & (1 << piece)) != 0)
3197 		return (-1);
3198 	else
3199 		mask[chunk] |= (1 << piece);
3200 
3201 	return (0);
3202 }
3203 
3204 int
3205 ctl_clear_mask(uint32_t *mask, uint32_t bit)
3206 {
3207 	uint32_t chunk, piece;
3208 
3209 	chunk = bit >> 5;
3210 	piece = bit % (sizeof(uint32_t) * 8);
3211 
3212 	if ((mask[chunk] & (1 << piece)) == 0)
3213 		return (-1);
3214 	else
3215 		mask[chunk] &= ~(1 << piece);
3216 
3217 	return (0);
3218 }
3219 
3220 int
3221 ctl_is_set(uint32_t *mask, uint32_t bit)
3222 {
3223 	uint32_t chunk, piece;
3224 
3225 	chunk = bit >> 5;
3226 	piece = bit % (sizeof(uint32_t) * 8);
3227 
3228 	if ((mask[chunk] & (1 << piece)) == 0)
3229 		return (0);
3230 	else
3231 		return (1);
3232 }
3233 
3234 #ifdef unused
3235 /*
3236  * The bus, target and lun are optional, they can be filled in later.
3237  * can_wait is used to determine whether we can wait on the malloc or not.
3238  */
3239 union ctl_io*
3240 ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port, uint32_t targ_target,
3241 	      uint32_t targ_lun, int can_wait)
3242 {
3243 	union ctl_io *io;
3244 
3245 	if (can_wait)
3246 		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_WAITOK);
3247 	else
3248 		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_NOWAIT);
3249 
3250 	if (io != NULL) {
3251 		io->io_hdr.io_type = io_type;
3252 		io->io_hdr.targ_port = targ_port;
3253 		/*
3254 		 * XXX KDM this needs to change/go away.  We need to move
3255 		 * to a preallocated pool of ctl_scsiio structures.
3256 		 */
3257 		io->io_hdr.nexus.targ_target.id = targ_target;
3258 		io->io_hdr.nexus.targ_lun = targ_lun;
3259 	}
3260 
3261 	return (io);
3262 }
3263 
3264 void
3265 ctl_kfree_io(union ctl_io *io)
3266 {
3267 	free(io, M_CTL);
3268 }
3269 #endif /* unused */
3270 
3271 /*
3272  * ctl_softc, pool_type, total_ctl_io are passed in.
3273  * npool is passed out.
3274  */
3275 int
3276 ctl_pool_create(struct ctl_softc *ctl_softc, ctl_pool_type pool_type,
3277 		uint32_t total_ctl_io, struct ctl_io_pool **npool)
3278 {
3279 	uint32_t i;
3280 	union ctl_io *cur_io, *next_io;
3281 	struct ctl_io_pool *pool;
3282 	int retval;
3283 
3284 	retval = 0;
3285 
3286 	pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3287 					    M_NOWAIT | M_ZERO);
3288 	if (pool == NULL) {
3289 		retval = -ENOMEM;
3290 		goto bailout;
3291 	}
3292 
3293 	pool->type = pool_type;
3294 	pool->ctl_softc = ctl_softc;
3295 
3296 	mtx_lock(&ctl_softc->ctl_lock);
3297 	pool->id = ctl_softc->cur_pool_id++;
3298 	mtx_unlock(&ctl_softc->ctl_lock);
3299 
3300 	pool->flags = CTL_POOL_FLAG_NONE;
3301 	STAILQ_INIT(&pool->free_queue);
3302 
3303 	/*
3304 	 * XXX KDM other options here:
3305 	 * - allocate a page at a time
3306 	 * - allocate one big chunk of memory.
3307 	 * Page allocation might work well, but would take a little more
3308 	 * tracking.
3309 	 */
3310 	for (i = 0; i < total_ctl_io; i++) {
3311 		cur_io = (union ctl_io *)malloc(sizeof(*cur_io), M_CTL,
3312 						M_NOWAIT);
3313 		if (cur_io == NULL) {
3314 			retval = ENOMEM;
3315 			break;
3316 		}
3317 		cur_io->io_hdr.pool = pool;
3318 		STAILQ_INSERT_TAIL(&pool->free_queue, &cur_io->io_hdr, links);
3319 		pool->total_ctl_io++;
3320 		pool->free_ctl_io++;
3321 	}
3322 
3323 	if (retval != 0) {
3324 		for (cur_io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3325 		     cur_io != NULL; cur_io = next_io) {
3326 			next_io = (union ctl_io *)STAILQ_NEXT(&cur_io->io_hdr,
3327 							      links);
3328 			STAILQ_REMOVE(&pool->free_queue, &cur_io->io_hdr,
3329 				      ctl_io_hdr, links);
3330 			free(cur_io, M_CTL);
3331 		}
3332 
3333 		free(pool, M_CTL);
3334 		goto bailout;
3335 	}
3336 	mtx_lock(&ctl_softc->ctl_lock);
3337 	ctl_softc->num_pools++;
3338 	STAILQ_INSERT_TAIL(&ctl_softc->io_pools, pool, links);
3339 	/*
3340 	 * Increment our usage count if this is an external consumer, so we
3341 	 * can't get unloaded until the external consumer (most likely a
3342 	 * FETD) unloads and frees his pool.
3343 	 *
3344 	 * XXX KDM will this increment the caller's module use count, or
3345 	 * mine?
3346 	 */
3347 #if 0
3348 	if ((pool_type != CTL_POOL_EMERGENCY)
3349 	 && (pool_type != CTL_POOL_INTERNAL)
3350 	 && (pool_type != CTL_POOL_IOCTL)
3351 	 && (pool_type != CTL_POOL_4OTHERSC))
3352 		MOD_INC_USE_COUNT;
3353 #endif
3354 
3355 	mtx_unlock(&ctl_softc->ctl_lock);
3356 
3357 	*npool = pool;
3358 
3359 bailout:
3360 
3361 	return (retval);
3362 }
3363 
3364 /*
3365  * Caller must hold ctl_softc->ctl_lock.
3366  */
3367 int
3368 ctl_pool_acquire(struct ctl_io_pool *pool)
3369 {
3370 	if (pool == NULL)
3371 		return (-EINVAL);
3372 
3373 	if (pool->flags & CTL_POOL_FLAG_INVALID)
3374 		return (-EINVAL);
3375 
3376 	pool->refcount++;
3377 
3378 	return (0);
3379 }
3380 
3381 /*
3382  * Caller must hold ctl_softc->ctl_lock.
3383  */
3384 int
3385 ctl_pool_invalidate(struct ctl_io_pool *pool)
3386 {
3387 	if (pool == NULL)
3388 		return (-EINVAL);
3389 
3390 	pool->flags |= CTL_POOL_FLAG_INVALID;
3391 
3392 	return (0);
3393 }
3394 
3395 /*
3396  * Caller must hold ctl_softc->ctl_lock.
3397  */
3398 int
3399 ctl_pool_release(struct ctl_io_pool *pool)
3400 {
3401 	if (pool == NULL)
3402 		return (-EINVAL);
3403 
3404 	if ((--pool->refcount == 0)
3405 	 && (pool->flags & CTL_POOL_FLAG_INVALID)) {
3406 		ctl_pool_free(pool->ctl_softc, pool);
3407 	}
3408 
3409 	return (0);
3410 }
3411 
3412 /*
3413  * Must be called with ctl_softc->ctl_lock held.
3414  */
3415 void
3416 ctl_pool_free(struct ctl_softc *ctl_softc, struct ctl_io_pool *pool)
3417 {
3418 	union ctl_io *cur_io, *next_io;
3419 
3420 	for (cur_io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3421 	     cur_io != NULL; cur_io = next_io) {
3422 		next_io = (union ctl_io *)STAILQ_NEXT(&cur_io->io_hdr,
3423 						      links);
3424 		STAILQ_REMOVE(&pool->free_queue, &cur_io->io_hdr, ctl_io_hdr,
3425 			      links);
3426 		free(cur_io, M_CTL);
3427 	}
3428 
3429 	STAILQ_REMOVE(&ctl_softc->io_pools, pool, ctl_io_pool, links);
3430 	ctl_softc->num_pools--;
3431 
3432 	/*
3433 	 * XXX KDM will this decrement the caller's usage count or mine?
3434 	 */
3435 #if 0
3436 	if ((pool->type != CTL_POOL_EMERGENCY)
3437 	 && (pool->type != CTL_POOL_INTERNAL)
3438 	 && (pool->type != CTL_POOL_IOCTL))
3439 		MOD_DEC_USE_COUNT;
3440 #endif
3441 
3442 	free(pool, M_CTL);
3443 }
3444 
3445 /*
3446  * This routine does not block (except for spinlocks of course).
3447  * It tries to allocate a ctl_io union from the caller's pool as quickly as
3448  * possible.
3449  */
3450 union ctl_io *
3451 ctl_alloc_io(void *pool_ref)
3452 {
3453 	union ctl_io *io;
3454 	struct ctl_softc *ctl_softc;
3455 	struct ctl_io_pool *pool, *npool;
3456 	struct ctl_io_pool *emergency_pool;
3457 
3458 	pool = (struct ctl_io_pool *)pool_ref;
3459 
3460 	if (pool == NULL) {
3461 		printf("%s: pool is NULL\n", __func__);
3462 		return (NULL);
3463 	}
3464 
3465 	emergency_pool = NULL;
3466 
3467 	ctl_softc = pool->ctl_softc;
3468 
3469 	mtx_lock(&ctl_softc->ctl_lock);
3470 	/*
3471 	 * First, try to get the io structure from the user's pool.
3472 	 */
3473 	if (ctl_pool_acquire(pool) == 0) {
3474 		io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3475 		if (io != NULL) {
3476 			STAILQ_REMOVE_HEAD(&pool->free_queue, links);
3477 			pool->total_allocated++;
3478 			pool->free_ctl_io--;
3479 			mtx_unlock(&ctl_softc->ctl_lock);
3480 			return (io);
3481 		} else
3482 			ctl_pool_release(pool);
3483 	}
3484 	/*
3485 	 * If he doesn't have any io structures left, search for an
3486 	 * emergency pool and grab one from there.
3487 	 */
3488 	STAILQ_FOREACH(npool, &ctl_softc->io_pools, links) {
3489 		if (npool->type != CTL_POOL_EMERGENCY)
3490 			continue;
3491 
3492 		if (ctl_pool_acquire(npool) != 0)
3493 			continue;
3494 
3495 		emergency_pool = npool;
3496 
3497 		io = (union ctl_io *)STAILQ_FIRST(&npool->free_queue);
3498 		if (io != NULL) {
3499 			STAILQ_REMOVE_HEAD(&npool->free_queue, links);
3500 			npool->total_allocated++;
3501 			npool->free_ctl_io--;
3502 			mtx_unlock(&ctl_softc->ctl_lock);
3503 			return (io);
3504 		} else
3505 			ctl_pool_release(npool);
3506 	}
3507 
3508 	/* Drop the spinlock before we malloc */
3509 	mtx_unlock(&ctl_softc->ctl_lock);
3510 
3511 	/*
3512 	 * The emergency pool (if it exists) didn't have one, so try an
3513 	 * atomic (i.e. nonblocking) malloc and see if we get lucky.
3514 	 */
3515 	io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_NOWAIT);
3516 	if (io != NULL) {
3517 		/*
3518 		 * If the emergency pool exists but is empty, add this
3519 		 * ctl_io to its list when it gets freed.
3520 		 */
3521 		if (emergency_pool != NULL) {
3522 			mtx_lock(&ctl_softc->ctl_lock);
3523 			if (ctl_pool_acquire(emergency_pool) == 0) {
3524 				io->io_hdr.pool = emergency_pool;
3525 				emergency_pool->total_ctl_io++;
3526 				/*
3527 				 * Need to bump this, otherwise
3528 				 * total_allocated and total_freed won't
3529 				 * match when we no longer have anything
3530 				 * outstanding.
3531 				 */
3532 				emergency_pool->total_allocated++;
3533 			}
3534 			mtx_unlock(&ctl_softc->ctl_lock);
3535 		} else
3536 			io->io_hdr.pool = NULL;
3537 	}
3538 
3539 	return (io);
3540 }
3541 
3542 static void
3543 ctl_free_io_internal(union ctl_io *io, int have_lock)
3544 {
3545 	if (io == NULL)
3546 		return;
3547 
3548 	/*
3549 	 * If this ctl_io has a pool, return it to that pool.
3550 	 */
3551 	if (io->io_hdr.pool != NULL) {
3552 		struct ctl_io_pool *pool;
3553 #if 0
3554 		struct ctl_softc *ctl_softc;
3555 		union ctl_io *tmp_io;
3556 		unsigned long xflags;
3557 		int i;
3558 
3559 		ctl_softc = control_softc;
3560 #endif
3561 
3562 		pool = (struct ctl_io_pool *)io->io_hdr.pool;
3563 
3564 		if (have_lock == 0)
3565 			mtx_lock(&pool->ctl_softc->ctl_lock);
3566 #if 0
3567 		save_flags(xflags);
3568 
3569 		for (i = 0, tmp_io = (union ctl_io *)STAILQ_FIRST(
3570 		     &ctl_softc->task_queue); tmp_io != NULL; i++,
3571 		     tmp_io = (union ctl_io *)STAILQ_NEXT(&tmp_io->io_hdr,
3572 		     links)) {
3573 			if (tmp_io == io) {
3574 				printf("%s: %p is still on the task queue!\n",
3575 				       __func__, tmp_io);
3576 				printf("%s: (%d): type %d "
3577 				       "msg %d cdb %x iptl: "
3578 				       "%d:%d:%d:%d tag 0x%04x "
3579 				       "flg %#lx\n",
3580 					__func__, i,
3581 					tmp_io->io_hdr.io_type,
3582 					tmp_io->io_hdr.msg_type,
3583 					tmp_io->scsiio.cdb[0],
3584 					tmp_io->io_hdr.nexus.initid.id,
3585 					tmp_io->io_hdr.nexus.targ_port,
3586 					tmp_io->io_hdr.nexus.targ_target.id,
3587 					tmp_io->io_hdr.nexus.targ_lun,
3588 					(tmp_io->io_hdr.io_type ==
3589 					CTL_IO_TASK) ?
3590 					tmp_io->taskio.tag_num :
3591 					tmp_io->scsiio.tag_num,
3592 					xflags);
3593 				panic("I/O still on the task queue!");
3594 			}
3595 		}
3596 #endif
3597 		io->io_hdr.io_type = 0xff;
3598 		STAILQ_INSERT_TAIL(&pool->free_queue, &io->io_hdr, links);
3599 		pool->total_freed++;
3600 		pool->free_ctl_io++;
3601 		ctl_pool_release(pool);
3602 		if (have_lock == 0)
3603 			mtx_unlock(&pool->ctl_softc->ctl_lock);
3604 	} else {
3605 		/*
3606 		 * Otherwise, just free it.  We probably malloced it and
3607 		 * the emergency pool wasn't available.
3608 		 */
3609 		free(io, M_CTL);
3610 	}
3611 
3612 }
3613 
3614 void
3615 ctl_free_io(union ctl_io *io)
3616 {
3617 	ctl_free_io_internal(io, /*have_lock*/ 0);
3618 }
3619 
3620 void
3621 ctl_zero_io(union ctl_io *io)
3622 {
3623 	void *pool_ref;
3624 
3625 	if (io == NULL)
3626 		return;
3627 
3628 	/*
3629 	 * May need to preserve linked list pointers at some point too.
3630 	 */
3631 	pool_ref = io->io_hdr.pool;
3632 
3633 	memset(io, 0, sizeof(*io));
3634 
3635 	io->io_hdr.pool = pool_ref;
3636 }
3637 
3638 /*
3639  * This routine is currently used for internal copies of ctl_ios that need
3640  * to persist for some reason after we've already returned status to the
3641  * FETD.  (Thus the flag set.)
3642  *
3643  * XXX XXX
3644  * Note that this makes a blind copy of all fields in the ctl_io, except
3645  * for the pool reference.  This includes any memory that has been
3646  * allocated!  That memory will no longer be valid after done has been
3647  * called, so this would be VERY DANGEROUS for command that actually does
3648  * any reads or writes.  Right now (11/7/2005), this is only used for immediate
3649  * start and stop commands, which don't transfer any data, so this is not a
3650  * problem.  If it is used for anything else, the caller would also need to
3651  * allocate data buffer space and this routine would need to be modified to
3652  * copy the data buffer(s) as well.
3653  */
3654 void
3655 ctl_copy_io(union ctl_io *src, union ctl_io *dest)
3656 {
3657 	void *pool_ref;
3658 
3659 	if ((src == NULL)
3660 	 || (dest == NULL))
3661 		return;
3662 
3663 	/*
3664 	 * May need to preserve linked list pointers at some point too.
3665 	 */
3666 	pool_ref = dest->io_hdr.pool;
3667 
3668 	memcpy(dest, src, ctl_min(sizeof(*src), sizeof(*dest)));
3669 
3670 	dest->io_hdr.pool = pool_ref;
3671 	/*
3672 	 * We need to know that this is an internal copy, and doesn't need
3673 	 * to get passed back to the FETD that allocated it.
3674 	 */
3675 	dest->io_hdr.flags |= CTL_FLAG_INT_COPY;
3676 }
3677 
3678 #ifdef NEEDTOPORT
3679 static void
3680 ctl_update_power_subpage(struct copan_power_subpage *page)
3681 {
3682 	int num_luns, num_partitions, config_type;
3683 	struct ctl_softc *softc;
3684 	cs_BOOL_t aor_present, shelf_50pct_power;
3685 	cs_raidset_personality_t rs_type;
3686 	int max_active_luns;
3687 
3688 	softc = control_softc;
3689 
3690 	/* subtract out the processor LUN */
3691 	num_luns = softc->num_luns - 1;
3692 	/*
3693 	 * Default to 7 LUNs active, which was the only number we allowed
3694 	 * in the past.
3695 	 */
3696 	max_active_luns = 7;
3697 
3698 	num_partitions = config_GetRsPartitionInfo();
3699 	config_type = config_GetConfigType();
3700 	shelf_50pct_power = config_GetShelfPowerMode();
3701 	aor_present = config_IsAorRsPresent();
3702 
3703 	rs_type = ddb_GetRsRaidType(1);
3704 	if ((rs_type != CS_RAIDSET_PERSONALITY_RAID5)
3705 	 && (rs_type != CS_RAIDSET_PERSONALITY_RAID1)) {
3706 		EPRINT(0, "Unsupported RS type %d!", rs_type);
3707 	}
3708 
3709 
3710 	page->total_luns = num_luns;
3711 
3712 	switch (config_type) {
3713 	case 40:
3714 		/*
3715 		 * In a 40 drive configuration, it doesn't matter what DC
3716 		 * cards we have, whether we have AOR enabled or not,
3717 		 * partitioning or not, or what type of RAIDset we have.
3718 		 * In that scenario, we can power up every LUN we present
3719 		 * to the user.
3720 		 */
3721 		max_active_luns = num_luns;
3722 
3723 		break;
3724 	case 64:
3725 		if (shelf_50pct_power == CS_FALSE) {
3726 			/* 25% power */
3727 			if (aor_present == CS_TRUE) {
3728 				if (rs_type ==
3729 				     CS_RAIDSET_PERSONALITY_RAID5) {
3730 					max_active_luns = 7;
3731 				} else if (rs_type ==
3732 					 CS_RAIDSET_PERSONALITY_RAID1){
3733 					max_active_luns = 14;
3734 				} else {
3735 					/* XXX KDM now what?? */
3736 				}
3737 			} else {
3738 				if (rs_type ==
3739 				     CS_RAIDSET_PERSONALITY_RAID5) {
3740 					max_active_luns = 8;
3741 				} else if (rs_type ==
3742 					 CS_RAIDSET_PERSONALITY_RAID1){
3743 					max_active_luns = 16;
3744 				} else {
3745 					/* XXX KDM now what?? */
3746 				}
3747 			}
3748 		} else {
3749 			/* 50% power */
3750 			/*
3751 			 * With 50% power in a 64 drive configuration, we
3752 			 * can power all LUNs we present.
3753 			 */
3754 			max_active_luns = num_luns;
3755 		}
3756 		break;
3757 	case 112:
3758 		if (shelf_50pct_power == CS_FALSE) {
3759 			/* 25% power */
3760 			if (aor_present == CS_TRUE) {
3761 				if (rs_type ==
3762 				     CS_RAIDSET_PERSONALITY_RAID5) {
3763 					max_active_luns = 7;
3764 				} else if (rs_type ==
3765 					 CS_RAIDSET_PERSONALITY_RAID1){
3766 					max_active_luns = 14;
3767 				} else {
3768 					/* XXX KDM now what?? */
3769 				}
3770 			} else {
3771 				if (rs_type ==
3772 				     CS_RAIDSET_PERSONALITY_RAID5) {
3773 					max_active_luns = 8;
3774 				} else if (rs_type ==
3775 					 CS_RAIDSET_PERSONALITY_RAID1){
3776 					max_active_luns = 16;
3777 				} else {
3778 					/* XXX KDM now what?? */
3779 				}
3780 			}
3781 		} else {
3782 			/* 50% power */
3783 			if (aor_present == CS_TRUE) {
3784 				if (rs_type ==
3785 				     CS_RAIDSET_PERSONALITY_RAID5) {
3786 					max_active_luns = 14;
3787 				} else if (rs_type ==
3788 					 CS_RAIDSET_PERSONALITY_RAID1){
3789 					/*
3790 					 * We're assuming here that disk
3791 					 * caching is enabled, and so we're
3792 					 * able to power up half of each
3793 					 * LUN, and cache all writes.
3794 					 */
3795 					max_active_luns = num_luns;
3796 				} else {
3797 					/* XXX KDM now what?? */
3798 				}
3799 			} else {
3800 				if (rs_type ==
3801 				     CS_RAIDSET_PERSONALITY_RAID5) {
3802 					max_active_luns = 15;
3803 				} else if (rs_type ==
3804 					 CS_RAIDSET_PERSONALITY_RAID1){
3805 					max_active_luns = 30;
3806 				} else {
3807 					/* XXX KDM now what?? */
3808 				}
3809 			}
3810 		}
3811 		break;
3812 	default:
3813 		/*
3814 		 * In this case, we have an unknown configuration, so we
3815 		 * just use the default from above.
3816 		 */
3817 		break;
3818 	}
3819 
3820 	page->max_active_luns = max_active_luns;
3821 #if 0
3822 	printk("%s: total_luns = %d, max_active_luns = %d\n", __func__,
3823 	       page->total_luns, page->max_active_luns);
3824 #endif
3825 }
3826 #endif /* NEEDTOPORT */
3827 
3828 /*
3829  * This routine could be used in the future to load default and/or saved
3830  * mode page parameters for a particuar lun.
3831  */
3832 static int
3833 ctl_init_page_index(struct ctl_lun *lun)
3834 {
3835 	int i;
3836 	struct ctl_page_index *page_index;
3837 	struct ctl_softc *softc;
3838 
3839 	memcpy(&lun->mode_pages.index, page_index_template,
3840 	       sizeof(page_index_template));
3841 
3842 	softc = lun->ctl_softc;
3843 
3844 	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
3845 
3846 		page_index = &lun->mode_pages.index[i];
3847 		/*
3848 		 * If this is a disk-only mode page, there's no point in
3849 		 * setting it up.  For some pages, we have to have some
3850 		 * basic information about the disk in order to calculate the
3851 		 * mode page data.
3852 		 */
3853 		if ((lun->be_lun->lun_type != T_DIRECT)
3854 		 && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
3855 			continue;
3856 
3857 		switch (page_index->page_code & SMPH_PC_MASK) {
3858 		case SMS_FORMAT_DEVICE_PAGE: {
3859 			struct scsi_format_page *format_page;
3860 
3861 			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3862 				panic("subpage is incorrect!");
3863 
3864 			/*
3865 			 * Sectors per track are set above.  Bytes per
3866 			 * sector need to be set here on a per-LUN basis.
3867 			 */
3868 			memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
3869 			       &format_page_default,
3870 			       sizeof(format_page_default));
3871 			memcpy(&lun->mode_pages.format_page[
3872 			       CTL_PAGE_CHANGEABLE], &format_page_changeable,
3873 			       sizeof(format_page_changeable));
3874 			memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
3875 			       &format_page_default,
3876 			       sizeof(format_page_default));
3877 			memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
3878 			       &format_page_default,
3879 			       sizeof(format_page_default));
3880 
3881 			format_page = &lun->mode_pages.format_page[
3882 				CTL_PAGE_CURRENT];
3883 			scsi_ulto2b(lun->be_lun->blocksize,
3884 				    format_page->bytes_per_sector);
3885 
3886 			format_page = &lun->mode_pages.format_page[
3887 				CTL_PAGE_DEFAULT];
3888 			scsi_ulto2b(lun->be_lun->blocksize,
3889 				    format_page->bytes_per_sector);
3890 
3891 			format_page = &lun->mode_pages.format_page[
3892 				CTL_PAGE_SAVED];
3893 			scsi_ulto2b(lun->be_lun->blocksize,
3894 				    format_page->bytes_per_sector);
3895 
3896 			page_index->page_data =
3897 				(uint8_t *)lun->mode_pages.format_page;
3898 			break;
3899 		}
3900 		case SMS_RIGID_DISK_PAGE: {
3901 			struct scsi_rigid_disk_page *rigid_disk_page;
3902 			uint32_t sectors_per_cylinder;
3903 			uint64_t cylinders;
3904 #ifndef	__XSCALE__
3905 			int shift;
3906 #endif /* !__XSCALE__ */
3907 
3908 			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3909 				panic("invalid subpage value %d",
3910 				      page_index->subpage);
3911 
3912 			/*
3913 			 * Rotation rate and sectors per track are set
3914 			 * above.  We calculate the cylinders here based on
3915 			 * capacity.  Due to the number of heads and
3916 			 * sectors per track we're using, smaller arrays
3917 			 * may turn out to have 0 cylinders.  Linux and
3918 			 * FreeBSD don't pay attention to these mode pages
3919 			 * to figure out capacity, but Solaris does.  It
3920 			 * seems to deal with 0 cylinders just fine, and
3921 			 * works out a fake geometry based on the capacity.
3922 			 */
3923 			memcpy(&lun->mode_pages.rigid_disk_page[
3924 			       CTL_PAGE_CURRENT], &rigid_disk_page_default,
3925 			       sizeof(rigid_disk_page_default));
3926 			memcpy(&lun->mode_pages.rigid_disk_page[
3927 			       CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
3928 			       sizeof(rigid_disk_page_changeable));
3929 			memcpy(&lun->mode_pages.rigid_disk_page[
3930 			       CTL_PAGE_DEFAULT], &rigid_disk_page_default,
3931 			       sizeof(rigid_disk_page_default));
3932 			memcpy(&lun->mode_pages.rigid_disk_page[
3933 			       CTL_PAGE_SAVED], &rigid_disk_page_default,
3934 			       sizeof(rigid_disk_page_default));
3935 
3936 			sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
3937 				CTL_DEFAULT_HEADS;
3938 
3939 			/*
3940 			 * The divide method here will be more accurate,
3941 			 * probably, but results in floating point being
3942 			 * used in the kernel on i386 (__udivdi3()).  On the
3943 			 * XScale, though, __udivdi3() is implemented in
3944 			 * software.
3945 			 *
3946 			 * The shift method for cylinder calculation is
3947 			 * accurate if sectors_per_cylinder is a power of
3948 			 * 2.  Otherwise it might be slightly off -- you
3949 			 * might have a bit of a truncation problem.
3950 			 */
3951 #ifdef	__XSCALE__
3952 			cylinders = (lun->be_lun->maxlba + 1) /
3953 				sectors_per_cylinder;
3954 #else
3955 			for (shift = 31; shift > 0; shift--) {
3956 				if (sectors_per_cylinder & (1 << shift))
3957 					break;
3958 			}
3959 			cylinders = (lun->be_lun->maxlba + 1) >> shift;
3960 #endif
3961 
3962 			/*
3963 			 * We've basically got 3 bytes, or 24 bits for the
3964 			 * cylinder size in the mode page.  If we're over,
3965 			 * just round down to 2^24.
3966 			 */
3967 			if (cylinders > 0xffffff)
3968 				cylinders = 0xffffff;
3969 
3970 			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
3971 				CTL_PAGE_CURRENT];
3972 			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
3973 
3974 			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
3975 				CTL_PAGE_DEFAULT];
3976 			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
3977 
3978 			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
3979 				CTL_PAGE_SAVED];
3980 			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
3981 
3982 			page_index->page_data =
3983 				(uint8_t *)lun->mode_pages.rigid_disk_page;
3984 			break;
3985 		}
3986 		case SMS_CACHING_PAGE: {
3987 
3988 			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3989 				panic("invalid subpage value %d",
3990 				      page_index->subpage);
3991 			/*
3992 			 * Defaults should be okay here, no calculations
3993 			 * needed.
3994 			 */
3995 			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
3996 			       &caching_page_default,
3997 			       sizeof(caching_page_default));
3998 			memcpy(&lun->mode_pages.caching_page[
3999 			       CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4000 			       sizeof(caching_page_changeable));
4001 			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4002 			       &caching_page_default,
4003 			       sizeof(caching_page_default));
4004 			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4005 			       &caching_page_default,
4006 			       sizeof(caching_page_default));
4007 			page_index->page_data =
4008 				(uint8_t *)lun->mode_pages.caching_page;
4009 			break;
4010 		}
4011 		case SMS_CONTROL_MODE_PAGE: {
4012 
4013 			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4014 				panic("invalid subpage value %d",
4015 				      page_index->subpage);
4016 
4017 			/*
4018 			 * Defaults should be okay here, no calculations
4019 			 * needed.
4020 			 */
4021 			memcpy(&lun->mode_pages.control_page[CTL_PAGE_CURRENT],
4022 			       &control_page_default,
4023 			       sizeof(control_page_default));
4024 			memcpy(&lun->mode_pages.control_page[
4025 			       CTL_PAGE_CHANGEABLE], &control_page_changeable,
4026 			       sizeof(control_page_changeable));
4027 			memcpy(&lun->mode_pages.control_page[CTL_PAGE_DEFAULT],
4028 			       &control_page_default,
4029 			       sizeof(control_page_default));
4030 			memcpy(&lun->mode_pages.control_page[CTL_PAGE_SAVED],
4031 			       &control_page_default,
4032 			       sizeof(control_page_default));
4033 			page_index->page_data =
4034 				(uint8_t *)lun->mode_pages.control_page;
4035 			break;
4036 
4037 		}
4038 		case SMS_VENDOR_SPECIFIC_PAGE:{
4039 			switch (page_index->subpage) {
4040 			case PWR_SUBPAGE_CODE: {
4041 				struct copan_power_subpage *current_page,
4042 							   *saved_page;
4043 
4044 				memcpy(&lun->mode_pages.power_subpage[
4045 				       CTL_PAGE_CURRENT],
4046 				       &power_page_default,
4047 				       sizeof(power_page_default));
4048 				memcpy(&lun->mode_pages.power_subpage[
4049 				       CTL_PAGE_CHANGEABLE],
4050 				       &power_page_changeable,
4051 				       sizeof(power_page_changeable));
4052 				memcpy(&lun->mode_pages.power_subpage[
4053 				       CTL_PAGE_DEFAULT],
4054 				       &power_page_default,
4055 				       sizeof(power_page_default));
4056 				memcpy(&lun->mode_pages.power_subpage[
4057 				       CTL_PAGE_SAVED],
4058 				       &power_page_default,
4059 				       sizeof(power_page_default));
4060 				page_index->page_data =
4061 				    (uint8_t *)lun->mode_pages.power_subpage;
4062 
4063 				current_page = (struct copan_power_subpage *)
4064 					(page_index->page_data +
4065 					 (page_index->page_len *
4066 					  CTL_PAGE_CURRENT));
4067 			        saved_page = (struct copan_power_subpage *)
4068 				        (page_index->page_data +
4069 					 (page_index->page_len *
4070 					  CTL_PAGE_SAVED));
4071 				break;
4072 			}
4073 			case APS_SUBPAGE_CODE: {
4074 				struct copan_aps_subpage *current_page,
4075 							 *saved_page;
4076 
4077 				// This gets set multiple times but
4078 				// it should always be the same. It's
4079 				// only done during init so who cares.
4080 				index_to_aps_page = i;
4081 
4082 				memcpy(&lun->mode_pages.aps_subpage[
4083 				       CTL_PAGE_CURRENT],
4084 				       &aps_page_default,
4085 				       sizeof(aps_page_default));
4086 				memcpy(&lun->mode_pages.aps_subpage[
4087 				       CTL_PAGE_CHANGEABLE],
4088 				       &aps_page_changeable,
4089 				       sizeof(aps_page_changeable));
4090 				memcpy(&lun->mode_pages.aps_subpage[
4091 				       CTL_PAGE_DEFAULT],
4092 				       &aps_page_default,
4093 				       sizeof(aps_page_default));
4094 				memcpy(&lun->mode_pages.aps_subpage[
4095 				       CTL_PAGE_SAVED],
4096 				       &aps_page_default,
4097 				       sizeof(aps_page_default));
4098 				page_index->page_data =
4099 					(uint8_t *)lun->mode_pages.aps_subpage;
4100 
4101 				current_page = (struct copan_aps_subpage *)
4102 					(page_index->page_data +
4103 					 (page_index->page_len *
4104 					  CTL_PAGE_CURRENT));
4105 				saved_page = (struct copan_aps_subpage *)
4106 					(page_index->page_data +
4107 					 (page_index->page_len *
4108 					  CTL_PAGE_SAVED));
4109 				break;
4110 			}
4111 			case DBGCNF_SUBPAGE_CODE: {
4112 				struct copan_debugconf_subpage *current_page,
4113 							       *saved_page;
4114 
4115 				memcpy(&lun->mode_pages.debugconf_subpage[
4116 				       CTL_PAGE_CURRENT],
4117 				       &debugconf_page_default,
4118 				       sizeof(debugconf_page_default));
4119 				memcpy(&lun->mode_pages.debugconf_subpage[
4120 				       CTL_PAGE_CHANGEABLE],
4121 				       &debugconf_page_changeable,
4122 				       sizeof(debugconf_page_changeable));
4123 				memcpy(&lun->mode_pages.debugconf_subpage[
4124 				       CTL_PAGE_DEFAULT],
4125 				       &debugconf_page_default,
4126 				       sizeof(debugconf_page_default));
4127 				memcpy(&lun->mode_pages.debugconf_subpage[
4128 				       CTL_PAGE_SAVED],
4129 				       &debugconf_page_default,
4130 				       sizeof(debugconf_page_default));
4131 				page_index->page_data =
4132 					(uint8_t *)lun->mode_pages.debugconf_subpage;
4133 
4134 				current_page = (struct copan_debugconf_subpage *)
4135 					(page_index->page_data +
4136 					 (page_index->page_len *
4137 					  CTL_PAGE_CURRENT));
4138 				saved_page = (struct copan_debugconf_subpage *)
4139 					(page_index->page_data +
4140 					 (page_index->page_len *
4141 					  CTL_PAGE_SAVED));
4142 				break;
4143 			}
4144 			default:
4145 				panic("invalid subpage value %d",
4146 				      page_index->subpage);
4147 				break;
4148 			}
4149    			break;
4150 		}
4151 		default:
4152 			panic("invalid page value %d",
4153 			      page_index->page_code & SMPH_PC_MASK);
4154 			break;
4155     	}
4156 	}
4157 
4158 	return (CTL_RETVAL_COMPLETE);
4159 }
4160 
4161 /*
4162  * LUN allocation.
4163  *
4164  * Requirements:
4165  * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4166  *   wants us to allocate the LUN and he can block.
4167  * - ctl_softc is always set
4168  * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4169  *
4170  * Returns 0 for success, non-zero (errno) for failure.
4171  */
4172 static int
4173 ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4174 	      struct ctl_be_lun *const be_lun, struct ctl_id target_id)
4175 {
4176 	struct ctl_lun *nlun, *lun;
4177 	struct ctl_frontend *fe;
4178 	int lun_number, i;
4179 
4180 	if (be_lun == NULL)
4181 		return (EINVAL);
4182 
4183 	/*
4184 	 * We currently only support Direct Access or Processor LUN types.
4185 	 */
4186 	switch (be_lun->lun_type) {
4187 	case T_DIRECT:
4188 		break;
4189 	case T_PROCESSOR:
4190 		break;
4191 	case T_SEQUENTIAL:
4192 	case T_CHANGER:
4193 	default:
4194 		be_lun->lun_config_status(be_lun->be_lun,
4195 					  CTL_LUN_CONFIG_FAILURE);
4196 		break;
4197 	}
4198 	if (ctl_lun == NULL) {
4199 		lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4200 		lun->flags = CTL_LUN_MALLOCED;
4201 	} else
4202 		lun = ctl_lun;
4203 
4204 	memset(lun, 0, sizeof(*lun));
4205 
4206 	mtx_lock(&ctl_softc->ctl_lock);
4207 	/*
4208 	 * See if the caller requested a particular LUN number.  If so, see
4209 	 * if it is available.  Otherwise, allocate the first available LUN.
4210 	 */
4211 	if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4212 		if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4213 		 || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4214 			mtx_unlock(&ctl_softc->ctl_lock);
4215 			if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4216 				printf("ctl: requested LUN ID %d is higher "
4217 				       "than CTL_MAX_LUNS - 1 (%d)\n",
4218 				       be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4219 			} else {
4220 				/*
4221 				 * XXX KDM return an error, or just assign
4222 				 * another LUN ID in this case??
4223 				 */
4224 				printf("ctl: requested LUN ID %d is already "
4225 				       "in use\n", be_lun->req_lun_id);
4226 			}
4227 			if (lun->flags & CTL_LUN_MALLOCED)
4228 				free(lun, M_CTL);
4229 			be_lun->lun_config_status(be_lun->be_lun,
4230 						  CTL_LUN_CONFIG_FAILURE);
4231 			return (ENOSPC);
4232 		}
4233 		lun_number = be_lun->req_lun_id;
4234 	} else {
4235 		lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, CTL_MAX_LUNS);
4236 		if (lun_number == -1) {
4237 			mtx_unlock(&ctl_softc->ctl_lock);
4238 			printf("ctl: can't allocate LUN on target %ju, out of "
4239 			       "LUNs\n", (uintmax_t)target_id.id);
4240 			if (lun->flags & CTL_LUN_MALLOCED)
4241 				free(lun, M_CTL);
4242 			be_lun->lun_config_status(be_lun->be_lun,
4243 						  CTL_LUN_CONFIG_FAILURE);
4244 			return (ENOSPC);
4245 		}
4246 	}
4247 	ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4248 
4249 	lun->target = target_id;
4250 	lun->lun = lun_number;
4251 	lun->be_lun = be_lun;
4252 	/*
4253 	 * The processor LUN is always enabled.  Disk LUNs come on line
4254 	 * disabled, and must be enabled by the backend.
4255 	 */
4256 	lun->flags = CTL_LUN_DISABLED;
4257 	lun->backend = be_lun->be;
4258 	be_lun->ctl_lun = lun;
4259 	be_lun->lun_id = lun_number;
4260 	atomic_add_int(&be_lun->be->num_luns, 1);
4261 	if (be_lun->flags & CTL_LUN_FLAG_POWERED_OFF)
4262 		lun->flags |= CTL_LUN_STOPPED;
4263 
4264 	if (be_lun->flags & CTL_LUN_FLAG_INOPERABLE)
4265 		lun->flags |= CTL_LUN_INOPERABLE;
4266 
4267 	if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4268 		lun->flags |= CTL_LUN_PRIMARY_SC;
4269 
4270 	lun->ctl_softc = ctl_softc;
4271 	TAILQ_INIT(&lun->ooa_queue);
4272 	TAILQ_INIT(&lun->blocked_queue);
4273 	STAILQ_INIT(&lun->error_list);
4274 
4275 	/*
4276 	 * Initialize the mode page index.
4277 	 */
4278 	ctl_init_page_index(lun);
4279 
4280 	/*
4281 	 * Set the poweron UA for all initiators on this LUN only.
4282 	 */
4283 	for (i = 0; i < CTL_MAX_INITIATORS; i++)
4284 		lun->pending_sense[i].ua_pending = CTL_UA_POWERON;
4285 
4286 	/*
4287 	 * Now, before we insert this lun on the lun list, set the lun
4288 	 * inventory changed UA for all other luns.
4289 	 */
4290 	STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4291 		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4292 			nlun->pending_sense[i].ua_pending |= CTL_UA_LUN_CHANGE;
4293 		}
4294 	}
4295 
4296 	STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4297 
4298 	ctl_softc->ctl_luns[lun_number] = lun;
4299 
4300 	ctl_softc->num_luns++;
4301 
4302 	/* Setup statistics gathering */
4303 	lun->stats.device_type = be_lun->lun_type;
4304 	lun->stats.lun_number = lun_number;
4305 	if (lun->stats.device_type == T_DIRECT)
4306 		lun->stats.blocksize = be_lun->blocksize;
4307 	else
4308 		lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4309 	for (i = 0;i < CTL_MAX_PORTS;i++)
4310 		lun->stats.ports[i].targ_port = i;
4311 
4312 	mtx_unlock(&ctl_softc->ctl_lock);
4313 
4314 	lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4315 
4316 	/*
4317 	 * Run through each registered FETD and bring it online if it isn't
4318 	 * already.  Enable the target ID if it hasn't been enabled, and
4319 	 * enable this particular LUN.
4320 	 */
4321 	STAILQ_FOREACH(fe, &ctl_softc->fe_list, links) {
4322 		int retval;
4323 
4324 		/*
4325 		 * XXX KDM this only works for ONE TARGET ID.  We'll need
4326 		 * to do things differently if we go to a multiple target
4327 		 * ID scheme.
4328 		 */
4329 		if ((fe->status & CTL_PORT_STATUS_TARG_ONLINE) == 0) {
4330 
4331 			retval = fe->targ_enable(fe->targ_lun_arg, target_id);
4332 			if (retval != 0) {
4333 				printf("ctl_alloc_lun: FETD %s port %d "
4334 				       "returned error %d for targ_enable on "
4335 				       "target %ju\n", fe->port_name,
4336 				       fe->targ_port, retval,
4337 				       (uintmax_t)target_id.id);
4338 			} else
4339 				fe->status |= CTL_PORT_STATUS_TARG_ONLINE;
4340 		}
4341 
4342 		retval = fe->lun_enable(fe->targ_lun_arg, target_id,lun_number);
4343 		if (retval != 0) {
4344 			printf("ctl_alloc_lun: FETD %s port %d returned error "
4345 			       "%d for lun_enable on target %ju lun %d\n",
4346 			       fe->port_name, fe->targ_port, retval,
4347 			       (uintmax_t)target_id.id, lun_number);
4348 		} else
4349 			fe->status |= CTL_PORT_STATUS_LUN_ONLINE;
4350 	}
4351 	return (0);
4352 }
4353 
4354 /*
4355  * Delete a LUN.
4356  * Assumptions:
4357  * - caller holds ctl_softc->ctl_lock.
4358  * - LUN has already been marked invalid and any pending I/O has been taken
4359  *   care of.
4360  */
4361 static int
4362 ctl_free_lun(struct ctl_lun *lun)
4363 {
4364 	struct ctl_softc *softc;
4365 #if 0
4366 	struct ctl_frontend *fe;
4367 #endif
4368 	struct ctl_lun *nlun;
4369 	union ctl_io *io, *next_io;
4370 	int i;
4371 
4372 	softc = lun->ctl_softc;
4373 
4374 	STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4375 
4376 	ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4377 
4378 	softc->ctl_luns[lun->lun] = NULL;
4379 
4380 	if (TAILQ_FIRST(&lun->ooa_queue) != NULL) {
4381 		printf("ctl_free_lun: aieee!! freeing a LUN with "
4382 		       "outstanding I/O!!\n");
4383 	}
4384 
4385 	/*
4386 	 * If we have anything pending on the RtR queue, remove it.
4387 	 */
4388 	for (io = (union ctl_io *)STAILQ_FIRST(&softc->rtr_queue); io != NULL;
4389 	     io = next_io) {
4390 		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
4391 		if ((io->io_hdr.nexus.targ_target.id == lun->target.id)
4392 		 && (io->io_hdr.nexus.targ_lun == lun->lun))
4393 			STAILQ_REMOVE(&softc->rtr_queue, &io->io_hdr,
4394 				      ctl_io_hdr, links);
4395 	}
4396 
4397 	/*
4398 	 * Then remove everything from the blocked queue.
4399 	 */
4400 	for (io = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue); io != NULL;
4401 	     io = next_io) {
4402 		next_io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,blocked_links);
4403 		TAILQ_REMOVE(&lun->blocked_queue, &io->io_hdr, blocked_links);
4404 		io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
4405 	}
4406 
4407 	/*
4408 	 * Now clear out the OOA queue, and free all the I/O.
4409 	 * XXX KDM should we notify the FETD here?  We probably need to
4410 	 * quiesce the LUN before deleting it.
4411 	 */
4412 	for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); io != NULL;
4413 	     io = next_io) {
4414 		next_io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr, ooa_links);
4415 		TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
4416 		ctl_free_io_internal(io, /*have_lock*/ 1);
4417 	}
4418 
4419 	softc->num_luns--;
4420 
4421 	/*
4422 	 * XXX KDM this scheme only works for a single target/multiple LUN
4423 	 * setup.  It needs to be revamped for a multiple target scheme.
4424 	 *
4425 	 * XXX KDM this results in fe->lun_disable() getting called twice,
4426 	 * once when ctl_disable_lun() is called, and a second time here.
4427 	 * We really need to re-think the LUN disable semantics.  There
4428 	 * should probably be several steps/levels to LUN removal:
4429 	 *  - disable
4430 	 *  - invalidate
4431 	 *  - free
4432  	 *
4433 	 * Right now we only have a disable method when communicating to
4434 	 * the front end ports, at least for individual LUNs.
4435 	 */
4436 #if 0
4437 	STAILQ_FOREACH(fe, &softc->fe_list, links) {
4438 		int retval;
4439 
4440 		retval = fe->lun_disable(fe->targ_lun_arg, lun->target,
4441 					 lun->lun);
4442 		if (retval != 0) {
4443 			printf("ctl_free_lun: FETD %s port %d returned error "
4444 			       "%d for lun_disable on target %ju lun %jd\n",
4445 			       fe->port_name, fe->targ_port, retval,
4446 			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4447 		}
4448 
4449 		if (STAILQ_FIRST(&softc->lun_list) == NULL) {
4450 			fe->status &= ~CTL_PORT_STATUS_LUN_ONLINE;
4451 
4452 			retval = fe->targ_disable(fe->targ_lun_arg,lun->target);
4453 			if (retval != 0) {
4454 				printf("ctl_free_lun: FETD %s port %d "
4455 				       "returned error %d for targ_disable on "
4456 				       "target %ju\n", fe->port_name,
4457 				       fe->targ_port, retval,
4458 				       (uintmax_t)lun->target.id);
4459 			} else
4460 				fe->status &= ~CTL_PORT_STATUS_TARG_ONLINE;
4461 
4462 			if ((fe->status & CTL_PORT_STATUS_TARG_ONLINE) != 0)
4463 				continue;
4464 
4465 #if 0
4466 			fe->port_offline(fe->onoff_arg);
4467 			fe->status &= ~CTL_PORT_STATUS_ONLINE;
4468 #endif
4469 		}
4470 	}
4471 #endif
4472 
4473 	/*
4474 	 * Tell the backend to free resources, if this LUN has a backend.
4475 	 */
4476 	atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4477 	lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4478 
4479 	if (lun->flags & CTL_LUN_MALLOCED)
4480 		free(lun, M_CTL);
4481 
4482 	STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4483 		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4484 			nlun->pending_sense[i].ua_pending |= CTL_UA_LUN_CHANGE;
4485 		}
4486 	}
4487 
4488 	return (0);
4489 }
4490 
4491 static void
4492 ctl_create_lun(struct ctl_be_lun *be_lun)
4493 {
4494 	struct ctl_softc *ctl_softc;
4495 
4496 	ctl_softc = control_softc;
4497 
4498 	/*
4499 	 * ctl_alloc_lun() should handle all potential failure cases.
4500 	 */
4501 	ctl_alloc_lun(ctl_softc, NULL, be_lun, ctl_softc->target);
4502 }
4503 
4504 int
4505 ctl_add_lun(struct ctl_be_lun *be_lun)
4506 {
4507 	struct ctl_softc *ctl_softc;
4508 
4509 	ctl_softc = control_softc;
4510 
4511 	mtx_lock(&ctl_softc->ctl_lock);
4512 	STAILQ_INSERT_TAIL(&ctl_softc->pending_lun_queue, be_lun, links);
4513 	mtx_unlock(&ctl_softc->ctl_lock);
4514 
4515 	ctl_wakeup_thread();
4516 
4517 	return (0);
4518 }
4519 
4520 int
4521 ctl_enable_lun(struct ctl_be_lun *be_lun)
4522 {
4523 	struct ctl_softc *ctl_softc;
4524 	struct ctl_frontend *fe, *nfe;
4525 	struct ctl_lun *lun;
4526 	int retval;
4527 
4528 	ctl_softc = control_softc;
4529 
4530 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4531 
4532 	mtx_lock(&ctl_softc->ctl_lock);
4533 	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4534 		/*
4535 		 * eh?  Why did we get called if the LUN is already
4536 		 * enabled?
4537 		 */
4538 		mtx_unlock(&ctl_softc->ctl_lock);
4539 		return (0);
4540 	}
4541 	lun->flags &= ~CTL_LUN_DISABLED;
4542 
4543 	for (fe = STAILQ_FIRST(&ctl_softc->fe_list); fe != NULL; fe = nfe) {
4544 		nfe = STAILQ_NEXT(fe, links);
4545 
4546 		/*
4547 		 * Drop the lock while we call the FETD's enable routine.
4548 		 * This can lead to a callback into CTL (at least in the
4549 		 * case of the internal initiator frontend.
4550 		 */
4551 		mtx_unlock(&ctl_softc->ctl_lock);
4552 		retval = fe->lun_enable(fe->targ_lun_arg, lun->target,lun->lun);
4553 		mtx_lock(&ctl_softc->ctl_lock);
4554 		if (retval != 0) {
4555 			printf("%s: FETD %s port %d returned error "
4556 			       "%d for lun_enable on target %ju lun %jd\n",
4557 			       __func__, fe->port_name, fe->targ_port, retval,
4558 			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4559 		}
4560 #if 0
4561 		 else {
4562             /* NOTE:  TODO:  why does lun enable affect port status? */
4563 			fe->status |= CTL_PORT_STATUS_LUN_ONLINE;
4564 		}
4565 #endif
4566 	}
4567 
4568 	mtx_unlock(&ctl_softc->ctl_lock);
4569 
4570 	return (0);
4571 }
4572 
4573 int
4574 ctl_disable_lun(struct ctl_be_lun *be_lun)
4575 {
4576 	struct ctl_softc *ctl_softc;
4577 	struct ctl_frontend *fe;
4578 	struct ctl_lun *lun;
4579 	int retval;
4580 
4581 	ctl_softc = control_softc;
4582 
4583 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4584 
4585 	mtx_lock(&ctl_softc->ctl_lock);
4586 
4587 	if (lun->flags & CTL_LUN_DISABLED) {
4588 		mtx_unlock(&ctl_softc->ctl_lock);
4589 		return (0);
4590 	}
4591 	lun->flags |= CTL_LUN_DISABLED;
4592 
4593 	STAILQ_FOREACH(fe, &ctl_softc->fe_list, links) {
4594 		mtx_unlock(&ctl_softc->ctl_lock);
4595 		/*
4596 		 * Drop the lock before we call the frontend's disable
4597 		 * routine, to avoid lock order reversals.
4598 		 *
4599 		 * XXX KDM what happens if the frontend list changes while
4600 		 * we're traversing it?  It's unlikely, but should be handled.
4601 		 */
4602 		retval = fe->lun_disable(fe->targ_lun_arg, lun->target,
4603 					 lun->lun);
4604 		mtx_lock(&ctl_softc->ctl_lock);
4605 		if (retval != 0) {
4606 			printf("ctl_alloc_lun: FETD %s port %d returned error "
4607 			       "%d for lun_disable on target %ju lun %jd\n",
4608 			       fe->port_name, fe->targ_port, retval,
4609 			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4610 		}
4611 	}
4612 
4613 	mtx_unlock(&ctl_softc->ctl_lock);
4614 
4615 	return (0);
4616 }
4617 
4618 int
4619 ctl_start_lun(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_STOPPED;
4630 	mtx_unlock(&ctl_softc->ctl_lock);
4631 
4632 	return (0);
4633 }
4634 
4635 int
4636 ctl_stop_lun(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_STOPPED;
4647 	mtx_unlock(&ctl_softc->ctl_lock);
4648 
4649 	return (0);
4650 }
4651 
4652 int
4653 ctl_lun_offline(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 	lun->flags |= CTL_LUN_OFFLINE;
4664 	mtx_unlock(&ctl_softc->ctl_lock);
4665 
4666 	return (0);
4667 }
4668 
4669 int
4670 ctl_lun_online(struct ctl_be_lun *be_lun)
4671 {
4672 	struct ctl_softc *ctl_softc;
4673 	struct ctl_lun *lun;
4674 
4675 	ctl_softc = control_softc;
4676 
4677 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4678 
4679 	mtx_lock(&ctl_softc->ctl_lock);
4680 	lun->flags &= ~CTL_LUN_OFFLINE;
4681 	mtx_unlock(&ctl_softc->ctl_lock);
4682 
4683 	return (0);
4684 }
4685 
4686 int
4687 ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4688 {
4689 	struct ctl_softc *ctl_softc;
4690 	struct ctl_lun *lun;
4691 
4692 	ctl_softc = control_softc;
4693 
4694 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4695 
4696 	mtx_lock(&ctl_softc->ctl_lock);
4697 
4698 	/*
4699 	 * The LUN needs to be disabled before it can be marked invalid.
4700 	 */
4701 	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4702 		mtx_unlock(&ctl_softc->ctl_lock);
4703 		return (-1);
4704 	}
4705 	/*
4706 	 * Mark the LUN invalid.
4707 	 */
4708 	lun->flags |= CTL_LUN_INVALID;
4709 
4710 	/*
4711 	 * If there is nothing in the OOA queue, go ahead and free the LUN.
4712 	 * If we have something in the OOA queue, we'll free it when the
4713 	 * last I/O completes.
4714 	 */
4715 	if (TAILQ_FIRST(&lun->ooa_queue) == NULL)
4716 		ctl_free_lun(lun);
4717 	mtx_unlock(&ctl_softc->ctl_lock);
4718 
4719 	return (0);
4720 }
4721 
4722 int
4723 ctl_lun_inoperable(struct ctl_be_lun *be_lun)
4724 {
4725 	struct ctl_softc *ctl_softc;
4726 	struct ctl_lun *lun;
4727 
4728 	ctl_softc = control_softc;
4729 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4730 
4731 	mtx_lock(&ctl_softc->ctl_lock);
4732 	lun->flags |= CTL_LUN_INOPERABLE;
4733 	mtx_unlock(&ctl_softc->ctl_lock);
4734 
4735 	return (0);
4736 }
4737 
4738 int
4739 ctl_lun_operable(struct ctl_be_lun *be_lun)
4740 {
4741 	struct ctl_softc *ctl_softc;
4742 	struct ctl_lun *lun;
4743 
4744 	ctl_softc = control_softc;
4745 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4746 
4747 	mtx_lock(&ctl_softc->ctl_lock);
4748 	lun->flags &= ~CTL_LUN_INOPERABLE;
4749 	mtx_unlock(&ctl_softc->ctl_lock);
4750 
4751 	return (0);
4752 }
4753 
4754 int
4755 ctl_lun_power_lock(struct ctl_be_lun *be_lun, struct ctl_nexus *nexus,
4756 		   int lock)
4757 {
4758 	struct ctl_softc *softc;
4759 	struct ctl_lun *lun;
4760 	struct copan_aps_subpage *current_sp;
4761 	struct ctl_page_index *page_index;
4762 	int i;
4763 
4764 	softc = control_softc;
4765 
4766 	mtx_lock(&softc->ctl_lock);
4767 
4768 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4769 
4770 	page_index = NULL;
4771 	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
4772 		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
4773 		     APS_PAGE_CODE)
4774 			continue;
4775 
4776 		if (lun->mode_pages.index[i].subpage != APS_SUBPAGE_CODE)
4777 			continue;
4778 		page_index = &lun->mode_pages.index[i];
4779 	}
4780 
4781 	if (page_index == NULL) {
4782 		mtx_unlock(&softc->ctl_lock);
4783 		printf("%s: APS subpage not found for lun %ju!\n", __func__,
4784 		       (uintmax_t)lun->lun);
4785 		return (1);
4786 	}
4787 #if 0
4788 	if ((softc->aps_locked_lun != 0)
4789 	 && (softc->aps_locked_lun != lun->lun)) {
4790 		printf("%s: attempt to lock LUN %llu when %llu is already "
4791 		       "locked\n");
4792 		mtx_unlock(&softc->ctl_lock);
4793 		return (1);
4794 	}
4795 #endif
4796 
4797 	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
4798 		(page_index->page_len * CTL_PAGE_CURRENT));
4799 
4800 	if (lock != 0) {
4801 		current_sp->lock_active = APS_LOCK_ACTIVE;
4802 		softc->aps_locked_lun = lun->lun;
4803 	} else {
4804 		current_sp->lock_active = 0;
4805 		softc->aps_locked_lun = 0;
4806 	}
4807 
4808 
4809 	/*
4810 	 * If we're in HA mode, try to send the lock message to the other
4811 	 * side.
4812 	 */
4813 	if (ctl_is_single == 0) {
4814 		int isc_retval;
4815 		union ctl_ha_msg lock_msg;
4816 
4817 		lock_msg.hdr.nexus = *nexus;
4818 		lock_msg.hdr.msg_type = CTL_MSG_APS_LOCK;
4819 		if (lock != 0)
4820 			lock_msg.aps.lock_flag = 1;
4821 		else
4822 			lock_msg.aps.lock_flag = 0;
4823 		isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &lock_msg,
4824 					 sizeof(lock_msg), 0);
4825 		if (isc_retval > CTL_HA_STATUS_SUCCESS) {
4826 			printf("%s: APS (lock=%d) error returned from "
4827 			       "ctl_ha_msg_send: %d\n", __func__, lock, isc_retval);
4828 			mtx_unlock(&softc->ctl_lock);
4829 			return (1);
4830 		}
4831 	}
4832 
4833 	mtx_unlock(&softc->ctl_lock);
4834 
4835 	return (0);
4836 }
4837 
4838 void
4839 ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
4840 {
4841 	struct ctl_lun *lun;
4842 	struct ctl_softc *softc;
4843 	int i;
4844 
4845 	softc = control_softc;
4846 
4847 	mtx_lock(&softc->ctl_lock);
4848 
4849 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4850 
4851 	for (i = 0; i < CTL_MAX_INITIATORS; i++)
4852 		lun->pending_sense[i].ua_pending |= CTL_UA_CAPACITY_CHANGED;
4853 
4854 	mtx_unlock(&softc->ctl_lock);
4855 }
4856 
4857 /*
4858  * Backend "memory move is complete" callback for requests that never
4859  * make it down to say RAIDCore's configuration code.
4860  */
4861 int
4862 ctl_config_move_done(union ctl_io *io)
4863 {
4864 	int retval;
4865 
4866 	retval = CTL_RETVAL_COMPLETE;
4867 
4868 
4869 	CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
4870 	/*
4871 	 * XXX KDM this shouldn't happen, but what if it does?
4872 	 */
4873 	if (io->io_hdr.io_type != CTL_IO_SCSI)
4874 		panic("I/O type isn't CTL_IO_SCSI!");
4875 
4876 	if ((io->io_hdr.port_status == 0)
4877 	 && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
4878 	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE))
4879 		io->io_hdr.status = CTL_SUCCESS;
4880 	else if ((io->io_hdr.port_status != 0)
4881 	      && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
4882 	      && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)){
4883 		/*
4884 		 * For hardware error sense keys, the sense key
4885 		 * specific value is defined to be a retry count,
4886 		 * but we use it to pass back an internal FETD
4887 		 * error code.  XXX KDM  Hopefully the FETD is only
4888 		 * using 16 bits for an error code, since that's
4889 		 * all the space we have in the sks field.
4890 		 */
4891 		ctl_set_internal_failure(&io->scsiio,
4892 					 /*sks_valid*/ 1,
4893 					 /*retry_count*/
4894 					 io->io_hdr.port_status);
4895 		free(io->scsiio.kern_data_ptr, M_CTL);
4896 		ctl_done(io);
4897 		goto bailout;
4898 	}
4899 
4900 	if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
4901 	 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
4902 	 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
4903 		/*
4904 		 * XXX KDM just assuming a single pointer here, and not a
4905 		 * S/G list.  If we start using S/G lists for config data,
4906 		 * we'll need to know how to clean them up here as well.
4907 		 */
4908 		free(io->scsiio.kern_data_ptr, M_CTL);
4909 		/* Hopefully the user has already set the status... */
4910 		ctl_done(io);
4911 	} else {
4912 		/*
4913 		 * XXX KDM now we need to continue data movement.  Some
4914 		 * options:
4915 		 * - call ctl_scsiio() again?  We don't do this for data
4916 		 *   writes, because for those at least we know ahead of
4917 		 *   time where the write will go and how long it is.  For
4918 		 *   config writes, though, that information is largely
4919 		 *   contained within the write itself, thus we need to
4920 		 *   parse out the data again.
4921 		 *
4922 		 * - Call some other function once the data is in?
4923 		 */
4924 
4925 		/*
4926 		 * XXX KDM call ctl_scsiio() again for now, and check flag
4927 		 * bits to see whether we're allocated or not.
4928 		 */
4929 		retval = ctl_scsiio(&io->scsiio);
4930 	}
4931 bailout:
4932 	return (retval);
4933 }
4934 
4935 /*
4936  * This gets called by a backend driver when it is done with a
4937  * configuration write.
4938  */
4939 void
4940 ctl_config_write_done(union ctl_io *io)
4941 {
4942 	/*
4943 	 * If the IO_CONT flag is set, we need to call the supplied
4944 	 * function to continue processing the I/O, instead of completing
4945 	 * the I/O just yet.
4946 	 *
4947 	 * If there is an error, though, we don't want to keep processing.
4948 	 * Instead, just send status back to the initiator.
4949 	 */
4950 	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT)
4951 	 && (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)
4952 	  || ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))) {
4953 		io->scsiio.io_cont(io);
4954 		return;
4955 	}
4956 	/*
4957 	 * Since a configuration write can be done for commands that actually
4958 	 * have data allocated, like write buffer, and commands that have
4959 	 * no data, like start/stop unit, we need to check here.
4960 	 */
4961 	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT)
4962 		free(io->scsiio.kern_data_ptr, M_CTL);
4963 	ctl_done(io);
4964 }
4965 
4966 /*
4967  * SCSI release command.
4968  */
4969 int
4970 ctl_scsi_release(struct ctl_scsiio *ctsio)
4971 {
4972 	int length, longid, thirdparty_id, resv_id;
4973 	struct ctl_softc *ctl_softc;
4974 	struct ctl_lun *lun;
4975 
4976 	length = 0;
4977 	resv_id = 0;
4978 
4979 	CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
4980 
4981 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
4982 	ctl_softc = control_softc;
4983 
4984 	switch (ctsio->cdb[0]) {
4985 	case RELEASE: {
4986 		struct scsi_release *cdb;
4987 
4988 		cdb = (struct scsi_release *)ctsio->cdb;
4989 		if ((cdb->byte2 & 0x1f) != 0) {
4990 			ctl_set_invalid_field(ctsio,
4991 					      /*sks_valid*/ 1,
4992 					      /*command*/ 1,
4993 					      /*field*/ 1,
4994 					      /*bit_valid*/ 0,
4995 					      /*bit*/ 0);
4996 			ctl_done((union ctl_io *)ctsio);
4997 			return (CTL_RETVAL_COMPLETE);
4998 		}
4999 		break;
5000 	}
5001 	case RELEASE_10: {
5002 		struct scsi_release_10 *cdb;
5003 
5004 		cdb = (struct scsi_release_10 *)ctsio->cdb;
5005 
5006 		if ((cdb->byte2 & SR10_EXTENT) != 0) {
5007 			ctl_set_invalid_field(ctsio,
5008 					      /*sks_valid*/ 1,
5009 					      /*command*/ 1,
5010 					      /*field*/ 1,
5011 					      /*bit_valid*/ 1,
5012 					      /*bit*/ 0);
5013 			ctl_done((union ctl_io *)ctsio);
5014 			return (CTL_RETVAL_COMPLETE);
5015 
5016 		}
5017 
5018 		if ((cdb->byte2 & SR10_3RDPTY) != 0) {
5019 			ctl_set_invalid_field(ctsio,
5020 					      /*sks_valid*/ 1,
5021 					      /*command*/ 1,
5022 					      /*field*/ 1,
5023 					      /*bit_valid*/ 1,
5024 					      /*bit*/ 4);
5025 			ctl_done((union ctl_io *)ctsio);
5026 			return (CTL_RETVAL_COMPLETE);
5027 		}
5028 
5029 		if (cdb->byte2 & SR10_LONGID)
5030 			longid = 1;
5031 		else
5032 			thirdparty_id = cdb->thirdparty_id;
5033 
5034 		resv_id = cdb->resv_id;
5035 		length = scsi_2btoul(cdb->length);
5036 		break;
5037 	}
5038 	}
5039 
5040 
5041 	/*
5042 	 * XXX KDM right now, we only support LUN reservation.  We don't
5043 	 * support 3rd party reservations, or extent reservations, which
5044 	 * might actually need the parameter list.  If we've gotten this
5045 	 * far, we've got a LUN reservation.  Anything else got kicked out
5046 	 * above.  So, according to SPC, ignore the length.
5047 	 */
5048 	length = 0;
5049 
5050 	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5051 	 && (length > 0)) {
5052 		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5053 		ctsio->kern_data_len = length;
5054 		ctsio->kern_total_len = length;
5055 		ctsio->kern_data_resid = 0;
5056 		ctsio->kern_rel_offset = 0;
5057 		ctsio->kern_sg_entries = 0;
5058 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5059 		ctsio->be_move_done = ctl_config_move_done;
5060 		ctl_datamove((union ctl_io *)ctsio);
5061 
5062 		return (CTL_RETVAL_COMPLETE);
5063 	}
5064 
5065 	if (length > 0)
5066 		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5067 
5068 	mtx_lock(&ctl_softc->ctl_lock);
5069 
5070 	/*
5071 	 * According to SPC, it is not an error for an intiator to attempt
5072 	 * to release a reservation on a LUN that isn't reserved, or that
5073 	 * is reserved by another initiator.  The reservation can only be
5074 	 * released, though, by the initiator who made it or by one of
5075 	 * several reset type events.
5076 	 */
5077 	if (lun->flags & CTL_LUN_RESERVED) {
5078 		if ((ctsio->io_hdr.nexus.initid.id == lun->rsv_nexus.initid.id)
5079 		 && (ctsio->io_hdr.nexus.targ_port == lun->rsv_nexus.targ_port)
5080 		 && (ctsio->io_hdr.nexus.targ_target.id ==
5081 		     lun->rsv_nexus.targ_target.id)) {
5082 			lun->flags &= ~CTL_LUN_RESERVED;
5083 		}
5084 	}
5085 
5086 	ctsio->scsi_status = SCSI_STATUS_OK;
5087 	ctsio->io_hdr.status = CTL_SUCCESS;
5088 
5089 	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5090 		free(ctsio->kern_data_ptr, M_CTL);
5091 		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5092 	}
5093 
5094 	mtx_unlock(&ctl_softc->ctl_lock);
5095 
5096 	ctl_done((union ctl_io *)ctsio);
5097 	return (CTL_RETVAL_COMPLETE);
5098 }
5099 
5100 int
5101 ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5102 {
5103 	int extent, thirdparty, longid;
5104 	int resv_id, length;
5105 	uint64_t thirdparty_id;
5106 	struct ctl_softc *ctl_softc;
5107 	struct ctl_lun *lun;
5108 
5109 	extent = 0;
5110 	thirdparty = 0;
5111 	longid = 0;
5112 	resv_id = 0;
5113 	length = 0;
5114 	thirdparty_id = 0;
5115 
5116 	CTL_DEBUG_PRINT(("ctl_reserve\n"));
5117 
5118 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5119 	ctl_softc = control_softc;
5120 
5121 	switch (ctsio->cdb[0]) {
5122 	case RESERVE: {
5123 		struct scsi_reserve *cdb;
5124 
5125 		cdb = (struct scsi_reserve *)ctsio->cdb;
5126 		if ((cdb->byte2 & 0x1f) != 0) {
5127 			ctl_set_invalid_field(ctsio,
5128 					      /*sks_valid*/ 1,
5129 					      /*command*/ 1,
5130 					      /*field*/ 1,
5131 					      /*bit_valid*/ 0,
5132 					      /*bit*/ 0);
5133 			ctl_done((union ctl_io *)ctsio);
5134 			return (CTL_RETVAL_COMPLETE);
5135 		}
5136 		resv_id = cdb->resv_id;
5137 		length = scsi_2btoul(cdb->length);
5138 		break;
5139 	}
5140 	case RESERVE_10: {
5141 		struct scsi_reserve_10 *cdb;
5142 
5143 		cdb = (struct scsi_reserve_10 *)ctsio->cdb;
5144 
5145 		if ((cdb->byte2 & SR10_EXTENT) != 0) {
5146 			ctl_set_invalid_field(ctsio,
5147 					      /*sks_valid*/ 1,
5148 					      /*command*/ 1,
5149 					      /*field*/ 1,
5150 					      /*bit_valid*/ 1,
5151 					      /*bit*/ 0);
5152 			ctl_done((union ctl_io *)ctsio);
5153 			return (CTL_RETVAL_COMPLETE);
5154 		}
5155 		if ((cdb->byte2 & SR10_3RDPTY) != 0) {
5156 			ctl_set_invalid_field(ctsio,
5157 					      /*sks_valid*/ 1,
5158 					      /*command*/ 1,
5159 					      /*field*/ 1,
5160 					      /*bit_valid*/ 1,
5161 					      /*bit*/ 4);
5162 			ctl_done((union ctl_io *)ctsio);
5163 			return (CTL_RETVAL_COMPLETE);
5164 		}
5165 		if (cdb->byte2 & SR10_LONGID)
5166 			longid = 1;
5167 		else
5168 			thirdparty_id = cdb->thirdparty_id;
5169 
5170 		resv_id = cdb->resv_id;
5171 		length = scsi_2btoul(cdb->length);
5172 		break;
5173 	}
5174 	}
5175 
5176 	/*
5177 	 * XXX KDM right now, we only support LUN reservation.  We don't
5178 	 * support 3rd party reservations, or extent reservations, which
5179 	 * might actually need the parameter list.  If we've gotten this
5180 	 * far, we've got a LUN reservation.  Anything else got kicked out
5181 	 * above.  So, according to SPC, ignore the length.
5182 	 */
5183 	length = 0;
5184 
5185 	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5186 	 && (length > 0)) {
5187 		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5188 		ctsio->kern_data_len = length;
5189 		ctsio->kern_total_len = length;
5190 		ctsio->kern_data_resid = 0;
5191 		ctsio->kern_rel_offset = 0;
5192 		ctsio->kern_sg_entries = 0;
5193 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5194 		ctsio->be_move_done = ctl_config_move_done;
5195 		ctl_datamove((union ctl_io *)ctsio);
5196 
5197 		return (CTL_RETVAL_COMPLETE);
5198 	}
5199 
5200 	if (length > 0)
5201 		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5202 
5203 	mtx_lock(&ctl_softc->ctl_lock);
5204 	if (lun->flags & CTL_LUN_RESERVED) {
5205 		if ((ctsio->io_hdr.nexus.initid.id != lun->rsv_nexus.initid.id)
5206 		 || (ctsio->io_hdr.nexus.targ_port != lun->rsv_nexus.targ_port)
5207 		 || (ctsio->io_hdr.nexus.targ_target.id !=
5208 		     lun->rsv_nexus.targ_target.id)) {
5209 			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
5210 			ctsio->io_hdr.status = CTL_SCSI_ERROR;
5211 			goto bailout;
5212 		}
5213 	}
5214 
5215 	lun->flags |= CTL_LUN_RESERVED;
5216 	lun->rsv_nexus = ctsio->io_hdr.nexus;
5217 
5218 	ctsio->scsi_status = SCSI_STATUS_OK;
5219 	ctsio->io_hdr.status = CTL_SUCCESS;
5220 
5221 bailout:
5222 	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5223 		free(ctsio->kern_data_ptr, M_CTL);
5224 		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5225 	}
5226 
5227 	mtx_unlock(&ctl_softc->ctl_lock);
5228 
5229 	ctl_done((union ctl_io *)ctsio);
5230 	return (CTL_RETVAL_COMPLETE);
5231 }
5232 
5233 int
5234 ctl_start_stop(struct ctl_scsiio *ctsio)
5235 {
5236 	struct scsi_start_stop_unit *cdb;
5237 	struct ctl_lun *lun;
5238 	struct ctl_softc *ctl_softc;
5239 	int retval;
5240 
5241 	CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5242 
5243 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5244 	ctl_softc = control_softc;
5245 	retval = 0;
5246 
5247 	cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5248 
5249 	/*
5250 	 * XXX KDM
5251 	 * We don't support the immediate bit on a stop unit.  In order to
5252 	 * do that, we would need to code up a way to know that a stop is
5253 	 * pending, and hold off any new commands until it completes, one
5254 	 * way or another.  Then we could accept or reject those commands
5255 	 * depending on its status.  We would almost need to do the reverse
5256 	 * of what we do below for an immediate start -- return the copy of
5257 	 * the ctl_io to the FETD with status to send to the host (and to
5258 	 * free the copy!) and then free the original I/O once the stop
5259 	 * actually completes.  That way, the OOA queue mechanism can work
5260 	 * to block commands that shouldn't proceed.  Another alternative
5261 	 * would be to put the copy in the queue in place of the original,
5262 	 * and return the original back to the caller.  That could be
5263 	 * slightly safer..
5264 	 */
5265 	if ((cdb->byte2 & SSS_IMMED)
5266 	 && ((cdb->how & SSS_START) == 0)) {
5267 		ctl_set_invalid_field(ctsio,
5268 				      /*sks_valid*/ 1,
5269 				      /*command*/ 1,
5270 				      /*field*/ 1,
5271 				      /*bit_valid*/ 1,
5272 				      /*bit*/ 0);
5273 		ctl_done((union ctl_io *)ctsio);
5274 		return (CTL_RETVAL_COMPLETE);
5275 	}
5276 
5277 	/*
5278 	 * We don't support the power conditions field.  We need to check
5279 	 * this prior to checking the load/eject and start/stop bits.
5280 	 */
5281 	if ((cdb->how & SSS_PC_MASK) != SSS_PC_START_VALID) {
5282 		ctl_set_invalid_field(ctsio,
5283 				      /*sks_valid*/ 1,
5284 				      /*command*/ 1,
5285 				      /*field*/ 4,
5286 				      /*bit_valid*/ 1,
5287 				      /*bit*/ 4);
5288 		ctl_done((union ctl_io *)ctsio);
5289 		return (CTL_RETVAL_COMPLETE);
5290 	}
5291 
5292 	/*
5293 	 * Media isn't removable, so we can't load or eject it.
5294 	 */
5295 	if ((cdb->how & SSS_LOEJ) != 0) {
5296 		ctl_set_invalid_field(ctsio,
5297 				      /*sks_valid*/ 1,
5298 				      /*command*/ 1,
5299 				      /*field*/ 4,
5300 				      /*bit_valid*/ 1,
5301 				      /*bit*/ 1);
5302 		ctl_done((union ctl_io *)ctsio);
5303 		return (CTL_RETVAL_COMPLETE);
5304 	}
5305 
5306 	if ((lun->flags & CTL_LUN_PR_RESERVED)
5307 	 && ((cdb->how & SSS_START)==0)) {
5308 		uint32_t residx;
5309 
5310 		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5311 		if (!lun->per_res[residx].registered
5312 		 || (lun->pr_res_idx!=residx && lun->res_type < 4)) {
5313 
5314 			ctl_set_reservation_conflict(ctsio);
5315 			ctl_done((union ctl_io *)ctsio);
5316 			return (CTL_RETVAL_COMPLETE);
5317 		}
5318 	}
5319 
5320 	/*
5321 	 * If there is no backend on this device, we can't start or stop
5322 	 * it.  In theory we shouldn't get any start/stop commands in the
5323 	 * first place at this level if the LUN doesn't have a backend.
5324 	 * That should get stopped by the command decode code.
5325 	 */
5326 	if (lun->backend == NULL) {
5327 		ctl_set_invalid_opcode(ctsio);
5328 		ctl_done((union ctl_io *)ctsio);
5329 		return (CTL_RETVAL_COMPLETE);
5330 	}
5331 
5332 	/*
5333 	 * XXX KDM Copan-specific offline behavior.
5334 	 * Figure out a reasonable way to port this?
5335 	 */
5336 #ifdef NEEDTOPORT
5337 	mtx_lock(&ctl_softc->ctl_lock);
5338 
5339 	if (((cdb->byte2 & SSS_ONOFFLINE) == 0)
5340 	 && (lun->flags & CTL_LUN_OFFLINE)) {
5341 		/*
5342 		 * If the LUN is offline, and the on/offline bit isn't set,
5343 		 * reject the start or stop.  Otherwise, let it through.
5344 		 */
5345 		mtx_unlock(&ctl_softc->ctl_lock);
5346 		ctl_set_lun_not_ready(ctsio);
5347 		ctl_done((union ctl_io *)ctsio);
5348 	} else {
5349 		mtx_unlock(&ctl_softc->ctl_lock);
5350 #endif /* NEEDTOPORT */
5351 		/*
5352 		 * This could be a start or a stop when we're online,
5353 		 * or a stop/offline or start/online.  A start or stop when
5354 		 * we're offline is covered in the case above.
5355 		 */
5356 		/*
5357 		 * In the non-immediate case, we send the request to
5358 		 * the backend and return status to the user when
5359 		 * it is done.
5360 		 *
5361 		 * In the immediate case, we allocate a new ctl_io
5362 		 * to hold a copy of the request, and send that to
5363 		 * the backend.  We then set good status on the
5364 		 * user's request and return it immediately.
5365 		 */
5366 		if (cdb->byte2 & SSS_IMMED) {
5367 			union ctl_io *new_io;
5368 
5369 			new_io = ctl_alloc_io(ctsio->io_hdr.pool);
5370 			if (new_io == NULL) {
5371 				ctl_set_busy(ctsio);
5372 				ctl_done((union ctl_io *)ctsio);
5373 			} else {
5374 				ctl_copy_io((union ctl_io *)ctsio,
5375 					    new_io);
5376 				retval = lun->backend->config_write(new_io);
5377 				ctl_set_success(ctsio);
5378 				ctl_done((union ctl_io *)ctsio);
5379 			}
5380 		} else {
5381 			retval = lun->backend->config_write(
5382 				(union ctl_io *)ctsio);
5383 		}
5384 #ifdef NEEDTOPORT
5385 	}
5386 #endif
5387 	return (retval);
5388 }
5389 
5390 /*
5391  * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5392  * we don't really do anything with the LBA and length fields if the user
5393  * passes them in.  Instead we'll just flush out the cache for the entire
5394  * LUN.
5395  */
5396 int
5397 ctl_sync_cache(struct ctl_scsiio *ctsio)
5398 {
5399 	struct ctl_lun *lun;
5400 	struct ctl_softc *ctl_softc;
5401 	uint64_t starting_lba;
5402 	uint32_t block_count;
5403 	int reladr, immed;
5404 	int retval;
5405 
5406 	CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5407 
5408 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5409 	ctl_softc = control_softc;
5410 	retval = 0;
5411 	reladr = 0;
5412 	immed = 0;
5413 
5414 	switch (ctsio->cdb[0]) {
5415 	case SYNCHRONIZE_CACHE: {
5416 		struct scsi_sync_cache *cdb;
5417 		cdb = (struct scsi_sync_cache *)ctsio->cdb;
5418 
5419 		if (cdb->byte2 & SSC_RELADR)
5420 			reladr = 1;
5421 
5422 		if (cdb->byte2 & SSC_IMMED)
5423 			immed = 1;
5424 
5425 		starting_lba = scsi_4btoul(cdb->begin_lba);
5426 		block_count = scsi_2btoul(cdb->lb_count);
5427 		break;
5428 	}
5429 	case SYNCHRONIZE_CACHE_16: {
5430 		struct scsi_sync_cache_16 *cdb;
5431 		cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5432 
5433 		if (cdb->byte2 & SSC_RELADR)
5434 			reladr = 1;
5435 
5436 		if (cdb->byte2 & SSC_IMMED)
5437 			immed = 1;
5438 
5439 		starting_lba = scsi_8btou64(cdb->begin_lba);
5440 		block_count = scsi_4btoul(cdb->lb_count);
5441 		break;
5442 	}
5443 	default:
5444 		ctl_set_invalid_opcode(ctsio);
5445 		ctl_done((union ctl_io *)ctsio);
5446 		goto bailout;
5447 		break; /* NOTREACHED */
5448 	}
5449 
5450 	if (immed) {
5451 		/*
5452 		 * We don't support the immediate bit.  Since it's in the
5453 		 * same place for the 10 and 16 byte SYNCHRONIZE CACHE
5454 		 * commands, we can just return the same error in either
5455 		 * case.
5456 		 */
5457 		ctl_set_invalid_field(ctsio,
5458 				      /*sks_valid*/ 1,
5459 				      /*command*/ 1,
5460 				      /*field*/ 1,
5461 				      /*bit_valid*/ 1,
5462 				      /*bit*/ 1);
5463 		ctl_done((union ctl_io *)ctsio);
5464 		goto bailout;
5465 	}
5466 
5467 	if (reladr) {
5468 		/*
5469 		 * We don't support the reladr bit either.  It can only be
5470 		 * used with linked commands, and we don't support linked
5471 		 * commands.  Since the bit is in the same place for the
5472 		 * 10 and 16 byte SYNCHRONIZE CACHE * commands, we can
5473 		 * just return the same error in either case.
5474 		 */
5475 		ctl_set_invalid_field(ctsio,
5476 				      /*sks_valid*/ 1,
5477 				      /*command*/ 1,
5478 				      /*field*/ 1,
5479 				      /*bit_valid*/ 1,
5480 				      /*bit*/ 0);
5481 		ctl_done((union ctl_io *)ctsio);
5482 		goto bailout;
5483 	}
5484 
5485 	/*
5486 	 * We check the LBA and length, but don't do anything with them.
5487 	 * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5488 	 * get flushed.  This check will just help satisfy anyone who wants
5489 	 * to see an error for an out of range LBA.
5490 	 */
5491 	if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5492 		ctl_set_lba_out_of_range(ctsio);
5493 		ctl_done((union ctl_io *)ctsio);
5494 		goto bailout;
5495 	}
5496 
5497 	/*
5498 	 * If this LUN has no backend, we can't flush the cache anyway.
5499 	 */
5500 	if (lun->backend == NULL) {
5501 		ctl_set_invalid_opcode(ctsio);
5502 		ctl_done((union ctl_io *)ctsio);
5503 		goto bailout;
5504 	}
5505 
5506 	/*
5507 	 * Check to see whether we're configured to send the SYNCHRONIZE
5508 	 * CACHE command directly to the back end.
5509 	 */
5510 	mtx_lock(&ctl_softc->ctl_lock);
5511 	if ((ctl_softc->flags & CTL_FLAG_REAL_SYNC)
5512 	 && (++(lun->sync_count) >= lun->sync_interval)) {
5513 		lun->sync_count = 0;
5514 		mtx_unlock(&ctl_softc->ctl_lock);
5515 		retval = lun->backend->config_write((union ctl_io *)ctsio);
5516 	} else {
5517 		mtx_unlock(&ctl_softc->ctl_lock);
5518 		ctl_set_success(ctsio);
5519 		ctl_done((union ctl_io *)ctsio);
5520 	}
5521 
5522 bailout:
5523 
5524 	return (retval);
5525 }
5526 
5527 int
5528 ctl_format(struct ctl_scsiio *ctsio)
5529 {
5530 	struct scsi_format *cdb;
5531 	struct ctl_lun *lun;
5532 	struct ctl_softc *ctl_softc;
5533 	int length, defect_list_len;
5534 
5535 	CTL_DEBUG_PRINT(("ctl_format\n"));
5536 
5537 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5538 	ctl_softc = control_softc;
5539 
5540 	cdb = (struct scsi_format *)ctsio->cdb;
5541 
5542 	length = 0;
5543 	if (cdb->byte2 & SF_FMTDATA) {
5544 		if (cdb->byte2 & SF_LONGLIST)
5545 			length = sizeof(struct scsi_format_header_long);
5546 		else
5547 			length = sizeof(struct scsi_format_header_short);
5548 	}
5549 
5550 	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5551 	 && (length > 0)) {
5552 		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5553 		ctsio->kern_data_len = length;
5554 		ctsio->kern_total_len = length;
5555 		ctsio->kern_data_resid = 0;
5556 		ctsio->kern_rel_offset = 0;
5557 		ctsio->kern_sg_entries = 0;
5558 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5559 		ctsio->be_move_done = ctl_config_move_done;
5560 		ctl_datamove((union ctl_io *)ctsio);
5561 
5562 		return (CTL_RETVAL_COMPLETE);
5563 	}
5564 
5565 	defect_list_len = 0;
5566 
5567 	if (cdb->byte2 & SF_FMTDATA) {
5568 		if (cdb->byte2 & SF_LONGLIST) {
5569 			struct scsi_format_header_long *header;
5570 
5571 			header = (struct scsi_format_header_long *)
5572 				ctsio->kern_data_ptr;
5573 
5574 			defect_list_len = scsi_4btoul(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 		} else {
5585 			struct scsi_format_header_short *header;
5586 
5587 			header = (struct scsi_format_header_short *)
5588 				ctsio->kern_data_ptr;
5589 
5590 			defect_list_len = scsi_2btoul(header->defect_list_len);
5591 			if (defect_list_len != 0) {
5592 				ctl_set_invalid_field(ctsio,
5593 						      /*sks_valid*/ 1,
5594 						      /*command*/ 0,
5595 						      /*field*/ 2,
5596 						      /*bit_valid*/ 0,
5597 						      /*bit*/ 0);
5598 				goto bailout;
5599 			}
5600 		}
5601 	}
5602 
5603 	/*
5604 	 * The format command will clear out the "Medium format corrupted"
5605 	 * status if set by the configuration code.  That status is really
5606 	 * just a way to notify the host that we have lost the media, and
5607 	 * get them to issue a command that will basically make them think
5608 	 * they're blowing away the media.
5609 	 */
5610 	mtx_lock(&ctl_softc->ctl_lock);
5611 	lun->flags &= ~CTL_LUN_INOPERABLE;
5612 	mtx_unlock(&ctl_softc->ctl_lock);
5613 
5614 	ctsio->scsi_status = SCSI_STATUS_OK;
5615 	ctsio->io_hdr.status = CTL_SUCCESS;
5616 bailout:
5617 
5618 	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5619 		free(ctsio->kern_data_ptr, M_CTL);
5620 		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5621 	}
5622 
5623 	ctl_done((union ctl_io *)ctsio);
5624 	return (CTL_RETVAL_COMPLETE);
5625 }
5626 
5627 int
5628 ctl_write_buffer(struct ctl_scsiio *ctsio)
5629 {
5630 	struct scsi_write_buffer *cdb;
5631 	struct copan_page_header *header;
5632 	struct ctl_lun *lun;
5633 	struct ctl_softc *ctl_softc;
5634 	int buffer_offset, len;
5635 	int retval;
5636 
5637 	header = NULL;
5638 
5639 	retval = CTL_RETVAL_COMPLETE;
5640 
5641 	CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5642 
5643 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5644 	ctl_softc = control_softc;
5645 	cdb = (struct scsi_write_buffer *)ctsio->cdb;
5646 
5647 	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5648 		ctl_set_invalid_field(ctsio,
5649 				      /*sks_valid*/ 1,
5650 				      /*command*/ 1,
5651 				      /*field*/ 1,
5652 				      /*bit_valid*/ 1,
5653 				      /*bit*/ 4);
5654 		ctl_done((union ctl_io *)ctsio);
5655 		return (CTL_RETVAL_COMPLETE);
5656 	}
5657 	if (cdb->buffer_id != 0) {
5658 		ctl_set_invalid_field(ctsio,
5659 				      /*sks_valid*/ 1,
5660 				      /*command*/ 1,
5661 				      /*field*/ 2,
5662 				      /*bit_valid*/ 0,
5663 				      /*bit*/ 0);
5664 		ctl_done((union ctl_io *)ctsio);
5665 		return (CTL_RETVAL_COMPLETE);
5666 	}
5667 
5668 	len = scsi_3btoul(cdb->length);
5669 	buffer_offset = scsi_3btoul(cdb->offset);
5670 
5671 	if (len > sizeof(lun->write_buffer)) {
5672 		ctl_set_invalid_field(ctsio,
5673 				      /*sks_valid*/ 1,
5674 				      /*command*/ 1,
5675 				      /*field*/ 6,
5676 				      /*bit_valid*/ 0,
5677 				      /*bit*/ 0);
5678 		ctl_done((union ctl_io *)ctsio);
5679 		return (CTL_RETVAL_COMPLETE);
5680 	}
5681 
5682 	if (buffer_offset != 0) {
5683 		ctl_set_invalid_field(ctsio,
5684 				      /*sks_valid*/ 1,
5685 				      /*command*/ 1,
5686 				      /*field*/ 3,
5687 				      /*bit_valid*/ 0,
5688 				      /*bit*/ 0);
5689 		ctl_done((union ctl_io *)ctsio);
5690 		return (CTL_RETVAL_COMPLETE);
5691 	}
5692 
5693 	/*
5694 	 * If we've got a kernel request that hasn't been malloced yet,
5695 	 * malloc it and tell the caller the data buffer is here.
5696 	 */
5697 	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5698 		ctsio->kern_data_ptr = lun->write_buffer;
5699 		ctsio->kern_data_len = len;
5700 		ctsio->kern_total_len = len;
5701 		ctsio->kern_data_resid = 0;
5702 		ctsio->kern_rel_offset = 0;
5703 		ctsio->kern_sg_entries = 0;
5704 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5705 		ctsio->be_move_done = ctl_config_move_done;
5706 		ctl_datamove((union ctl_io *)ctsio);
5707 
5708 		return (CTL_RETVAL_COMPLETE);
5709 	}
5710 
5711 	ctl_done((union ctl_io *)ctsio);
5712 
5713 	return (CTL_RETVAL_COMPLETE);
5714 }
5715 
5716 /*
5717  * Note that this function currently doesn't actually do anything inside
5718  * CTL to enforce things if the DQue bit is turned on.
5719  *
5720  * Also note that this function can't be used in the default case, because
5721  * the DQue bit isn't set in the changeable mask for the control mode page
5722  * anyway.  This is just here as an example for how to implement a page
5723  * handler, and a placeholder in case we want to allow the user to turn
5724  * tagged queueing on and off.
5725  *
5726  * The D_SENSE bit handling is functional, however, and will turn
5727  * descriptor sense on and off for a given LUN.
5728  */
5729 int
5730 ctl_control_page_handler(struct ctl_scsiio *ctsio,
5731 			 struct ctl_page_index *page_index, uint8_t *page_ptr)
5732 {
5733 	struct scsi_control_page *current_cp, *saved_cp, *user_cp;
5734 	struct ctl_lun *lun;
5735 	struct ctl_softc *softc;
5736 	int set_ua;
5737 	uint32_t initidx;
5738 
5739 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5740 	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5741 	set_ua = 0;
5742 
5743 	user_cp = (struct scsi_control_page *)page_ptr;
5744 	current_cp = (struct scsi_control_page *)
5745 		(page_index->page_data + (page_index->page_len *
5746 		CTL_PAGE_CURRENT));
5747 	saved_cp = (struct scsi_control_page *)
5748 		(page_index->page_data + (page_index->page_len *
5749 		CTL_PAGE_SAVED));
5750 
5751 	softc = control_softc;
5752 
5753 	mtx_lock(&softc->ctl_lock);
5754 	if (((current_cp->rlec & SCP_DSENSE) == 0)
5755 	 && ((user_cp->rlec & SCP_DSENSE) != 0)) {
5756 		/*
5757 		 * Descriptor sense is currently turned off and the user
5758 		 * wants to turn it on.
5759 		 */
5760 		current_cp->rlec |= SCP_DSENSE;
5761 		saved_cp->rlec |= SCP_DSENSE;
5762 		lun->flags |= CTL_LUN_SENSE_DESC;
5763 		set_ua = 1;
5764 	} else if (((current_cp->rlec & SCP_DSENSE) != 0)
5765 		&& ((user_cp->rlec & SCP_DSENSE) == 0)) {
5766 		/*
5767 		 * Descriptor sense is currently turned on, and the user
5768 		 * wants to turn it off.
5769 		 */
5770 		current_cp->rlec &= ~SCP_DSENSE;
5771 		saved_cp->rlec &= ~SCP_DSENSE;
5772 		lun->flags &= ~CTL_LUN_SENSE_DESC;
5773 		set_ua = 1;
5774 	}
5775 	if (current_cp->queue_flags & SCP_QUEUE_DQUE) {
5776 		if (user_cp->queue_flags & SCP_QUEUE_DQUE) {
5777 #ifdef NEEDTOPORT
5778 			csevent_log(CSC_CTL | CSC_SHELF_SW |
5779 				    CTL_UNTAG_TO_UNTAG,
5780 				    csevent_LogType_Trace,
5781 				    csevent_Severity_Information,
5782 				    csevent_AlertLevel_Green,
5783 				    csevent_FRU_Firmware,
5784 				    csevent_FRU_Unknown,
5785 				    "Received untagged to untagged transition");
5786 #endif /* NEEDTOPORT */
5787 		} else {
5788 #ifdef NEEDTOPORT
5789 			csevent_log(CSC_CTL | CSC_SHELF_SW |
5790 				    CTL_UNTAG_TO_TAG,
5791 				    csevent_LogType_ConfigChange,
5792 				    csevent_Severity_Information,
5793 				    csevent_AlertLevel_Green,
5794 				    csevent_FRU_Firmware,
5795 				    csevent_FRU_Unknown,
5796 				    "Received untagged to tagged "
5797 				    "queueing transition");
5798 #endif /* NEEDTOPORT */
5799 
5800 			current_cp->queue_flags &= ~SCP_QUEUE_DQUE;
5801 			saved_cp->queue_flags &= ~SCP_QUEUE_DQUE;
5802 			set_ua = 1;
5803 		}
5804 	} else {
5805 		if (user_cp->queue_flags & SCP_QUEUE_DQUE) {
5806 #ifdef NEEDTOPORT
5807 			csevent_log(CSC_CTL | CSC_SHELF_SW |
5808 				    CTL_TAG_TO_UNTAG,
5809 				    csevent_LogType_ConfigChange,
5810 				    csevent_Severity_Warning,
5811 				    csevent_AlertLevel_Yellow,
5812 				    csevent_FRU_Firmware,
5813 				    csevent_FRU_Unknown,
5814 				    "Received tagged queueing to untagged "
5815 				    "transition");
5816 #endif /* NEEDTOPORT */
5817 
5818 			current_cp->queue_flags |= SCP_QUEUE_DQUE;
5819 			saved_cp->queue_flags |= SCP_QUEUE_DQUE;
5820 			set_ua = 1;
5821 		} else {
5822 #ifdef NEEDTOPORT
5823 			csevent_log(CSC_CTL | CSC_SHELF_SW |
5824 				    CTL_TAG_TO_TAG,
5825 				    csevent_LogType_Trace,
5826 				    csevent_Severity_Information,
5827 				    csevent_AlertLevel_Green,
5828 				    csevent_FRU_Firmware,
5829 				    csevent_FRU_Unknown,
5830 				    "Received tagged queueing to tagged "
5831 				    "queueing transition");
5832 #endif /* NEEDTOPORT */
5833 		}
5834 	}
5835 	if (set_ua != 0) {
5836 		int i;
5837 		/*
5838 		 * Let other initiators know that the mode
5839 		 * parameters for this LUN have changed.
5840 		 */
5841 		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
5842 			if (i == initidx)
5843 				continue;
5844 
5845 			lun->pending_sense[i].ua_pending |=
5846 				CTL_UA_MODE_CHANGE;
5847 		}
5848 	}
5849 	mtx_unlock(&softc->ctl_lock);
5850 
5851 	return (0);
5852 }
5853 
5854 int
5855 ctl_power_sp_handler(struct ctl_scsiio *ctsio,
5856 		     struct ctl_page_index *page_index, uint8_t *page_ptr)
5857 {
5858 	return (0);
5859 }
5860 
5861 int
5862 ctl_power_sp_sense_handler(struct ctl_scsiio *ctsio,
5863 			   struct ctl_page_index *page_index, int pc)
5864 {
5865 	struct copan_power_subpage *page;
5866 
5867 	page = (struct copan_power_subpage *)page_index->page_data +
5868 		(page_index->page_len * pc);
5869 
5870 	switch (pc) {
5871 	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
5872 		/*
5873 		 * We don't update the changable bits for this page.
5874 		 */
5875 		break;
5876 	case SMS_PAGE_CTRL_CURRENT >> 6:
5877 	case SMS_PAGE_CTRL_DEFAULT >> 6:
5878 	case SMS_PAGE_CTRL_SAVED >> 6:
5879 #ifdef NEEDTOPORT
5880 		ctl_update_power_subpage(page);
5881 #endif
5882 		break;
5883 	default:
5884 #ifdef NEEDTOPORT
5885 		EPRINT(0, "Invalid PC %d!!", pc);
5886 #endif
5887 		break;
5888 	}
5889 	return (0);
5890 }
5891 
5892 
5893 int
5894 ctl_aps_sp_handler(struct ctl_scsiio *ctsio,
5895 		   struct ctl_page_index *page_index, uint8_t *page_ptr)
5896 {
5897 	struct copan_aps_subpage *user_sp;
5898 	struct copan_aps_subpage *current_sp;
5899 	union ctl_modepage_info *modepage_info;
5900 	struct ctl_softc *softc;
5901 	struct ctl_lun *lun;
5902 	int retval;
5903 
5904 	retval = CTL_RETVAL_COMPLETE;
5905 	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
5906 		     (page_index->page_len * CTL_PAGE_CURRENT));
5907 	softc = control_softc;
5908 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5909 
5910 	user_sp = (struct copan_aps_subpage *)page_ptr;
5911 
5912 	modepage_info = (union ctl_modepage_info *)
5913 		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
5914 
5915 	modepage_info->header.page_code = page_index->page_code & SMPH_PC_MASK;
5916 	modepage_info->header.subpage = page_index->subpage;
5917 	modepage_info->aps.lock_active = user_sp->lock_active;
5918 
5919 	mtx_lock(&softc->ctl_lock);
5920 
5921 	/*
5922 	 * If there is a request to lock the LUN and another LUN is locked
5923 	 * this is an error. If the requested LUN is already locked ignore
5924 	 * the request. If no LUN is locked attempt to lock it.
5925 	 * if there is a request to unlock the LUN and the LUN is currently
5926 	 * locked attempt to unlock it. Otherwise ignore the request. i.e.
5927 	 * if another LUN is locked or no LUN is locked.
5928 	 */
5929 	if (user_sp->lock_active & APS_LOCK_ACTIVE) {
5930 		if (softc->aps_locked_lun == lun->lun) {
5931 			/*
5932 			 * This LUN is already locked, so we're done.
5933 			 */
5934 			retval = CTL_RETVAL_COMPLETE;
5935 		} else if (softc->aps_locked_lun == 0) {
5936 			/*
5937 			 * No one has the lock, pass the request to the
5938 			 * backend.
5939 			 */
5940 			retval = lun->backend->config_write(
5941 				(union ctl_io *)ctsio);
5942 		} else {
5943 			/*
5944 			 * Someone else has the lock, throw out the request.
5945 			 */
5946 			ctl_set_already_locked(ctsio);
5947 			free(ctsio->kern_data_ptr, M_CTL);
5948 			ctl_done((union ctl_io *)ctsio);
5949 
5950 			/*
5951 			 * Set the return value so that ctl_do_mode_select()
5952 			 * won't try to complete the command.  We already
5953 			 * completed it here.
5954 			 */
5955 			retval = CTL_RETVAL_ERROR;
5956 		}
5957 	} else if (softc->aps_locked_lun == lun->lun) {
5958 		/*
5959 		 * This LUN is locked, so pass the unlock request to the
5960 		 * backend.
5961 		 */
5962 		retval = lun->backend->config_write((union ctl_io *)ctsio);
5963 	}
5964 	mtx_unlock(&softc->ctl_lock);
5965 
5966 	return (retval);
5967 }
5968 
5969 int
5970 ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio,
5971 				struct ctl_page_index *page_index,
5972 				uint8_t *page_ptr)
5973 {
5974 	uint8_t *c;
5975 	int i;
5976 
5977 	c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs;
5978 	ctl_time_io_secs =
5979 		(c[0] << 8) |
5980 		(c[1] << 0) |
5981 		0;
5982 	CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs));
5983 	printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs);
5984 	printf("page data:");
5985 	for (i=0; i<8; i++)
5986 		printf(" %.2x",page_ptr[i]);
5987 	printf("\n");
5988 	return (0);
5989 }
5990 
5991 int
5992 ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio,
5993 			       struct ctl_page_index *page_index,
5994 			       int pc)
5995 {
5996 	struct copan_debugconf_subpage *page;
5997 
5998 	page = (struct copan_debugconf_subpage *)page_index->page_data +
5999 		(page_index->page_len * pc);
6000 
6001 	switch (pc) {
6002 	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6003 	case SMS_PAGE_CTRL_DEFAULT >> 6:
6004 	case SMS_PAGE_CTRL_SAVED >> 6:
6005 		/*
6006 		 * We don't update the changable or default bits for this page.
6007 		 */
6008 		break;
6009 	case SMS_PAGE_CTRL_CURRENT >> 6:
6010 		page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8;
6011 		page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0;
6012 		break;
6013 	default:
6014 #ifdef NEEDTOPORT
6015 		EPRINT(0, "Invalid PC %d!!", pc);
6016 #endif /* NEEDTOPORT */
6017 		break;
6018 	}
6019 	return (0);
6020 }
6021 
6022 
6023 static int
6024 ctl_do_mode_select(union ctl_io *io)
6025 {
6026 	struct scsi_mode_page_header *page_header;
6027 	struct ctl_page_index *page_index;
6028 	struct ctl_scsiio *ctsio;
6029 	int control_dev, page_len;
6030 	int page_len_offset, page_len_size;
6031 	union ctl_modepage_info *modepage_info;
6032 	struct ctl_lun *lun;
6033 	int *len_left, *len_used;
6034 	int retval, i;
6035 
6036 	ctsio = &io->scsiio;
6037 	page_index = NULL;
6038 	page_len = 0;
6039 	retval = CTL_RETVAL_COMPLETE;
6040 
6041 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6042 
6043 	if (lun->be_lun->lun_type != T_DIRECT)
6044 		control_dev = 1;
6045 	else
6046 		control_dev = 0;
6047 
6048 	modepage_info = (union ctl_modepage_info *)
6049 		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6050 	len_left = &modepage_info->header.len_left;
6051 	len_used = &modepage_info->header.len_used;
6052 
6053 do_next_page:
6054 
6055 	page_header = (struct scsi_mode_page_header *)
6056 		(ctsio->kern_data_ptr + *len_used);
6057 
6058 	if (*len_left == 0) {
6059 		free(ctsio->kern_data_ptr, M_CTL);
6060 		ctl_set_success(ctsio);
6061 		ctl_done((union ctl_io *)ctsio);
6062 		return (CTL_RETVAL_COMPLETE);
6063 	} else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6064 
6065 		free(ctsio->kern_data_ptr, M_CTL);
6066 		ctl_set_param_len_error(ctsio);
6067 		ctl_done((union ctl_io *)ctsio);
6068 		return (CTL_RETVAL_COMPLETE);
6069 
6070 	} else if ((page_header->page_code & SMPH_SPF)
6071 		&& (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6072 
6073 		free(ctsio->kern_data_ptr, M_CTL);
6074 		ctl_set_param_len_error(ctsio);
6075 		ctl_done((union ctl_io *)ctsio);
6076 		return (CTL_RETVAL_COMPLETE);
6077 	}
6078 
6079 
6080 	/*
6081 	 * XXX KDM should we do something with the block descriptor?
6082 	 */
6083 	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6084 
6085 		if ((control_dev != 0)
6086 		 && (lun->mode_pages.index[i].page_flags &
6087 		     CTL_PAGE_FLAG_DISK_ONLY))
6088 			continue;
6089 
6090 		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
6091 		    (page_header->page_code & SMPH_PC_MASK))
6092 			continue;
6093 
6094 		/*
6095 		 * If neither page has a subpage code, then we've got a
6096 		 * match.
6097 		 */
6098 		if (((lun->mode_pages.index[i].page_code & SMPH_SPF) == 0)
6099 		 && ((page_header->page_code & SMPH_SPF) == 0)) {
6100 			page_index = &lun->mode_pages.index[i];
6101 			page_len = page_header->page_length;
6102 			break;
6103 		}
6104 
6105 		/*
6106 		 * If both pages have subpages, then the subpage numbers
6107 		 * have to match.
6108 		 */
6109 		if ((lun->mode_pages.index[i].page_code & SMPH_SPF)
6110 		  && (page_header->page_code & SMPH_SPF)) {
6111 			struct scsi_mode_page_header_sp *sph;
6112 
6113 			sph = (struct scsi_mode_page_header_sp *)page_header;
6114 
6115 			if (lun->mode_pages.index[i].subpage ==
6116 			    sph->subpage) {
6117 				page_index = &lun->mode_pages.index[i];
6118 				page_len = scsi_2btoul(sph->page_length);
6119 				break;
6120 			}
6121 		}
6122 	}
6123 
6124 	/*
6125 	 * If we couldn't find the page, or if we don't have a mode select
6126 	 * handler for it, send back an error to the user.
6127 	 */
6128 	if ((page_index == NULL)
6129 	 || (page_index->select_handler == NULL)) {
6130 		ctl_set_invalid_field(ctsio,
6131 				      /*sks_valid*/ 1,
6132 				      /*command*/ 0,
6133 				      /*field*/ *len_used,
6134 				      /*bit_valid*/ 0,
6135 				      /*bit*/ 0);
6136 		free(ctsio->kern_data_ptr, M_CTL);
6137 		ctl_done((union ctl_io *)ctsio);
6138 		return (CTL_RETVAL_COMPLETE);
6139 	}
6140 
6141 	if (page_index->page_code & SMPH_SPF) {
6142 		page_len_offset = 2;
6143 		page_len_size = 2;
6144 	} else {
6145 		page_len_size = 1;
6146 		page_len_offset = 1;
6147 	}
6148 
6149 	/*
6150 	 * If the length the initiator gives us isn't the one we specify in
6151 	 * the mode page header, or if they didn't specify enough data in
6152 	 * the CDB to avoid truncating this page, kick out the request.
6153 	 */
6154 	if ((page_len != (page_index->page_len - page_len_offset -
6155 			  page_len_size))
6156 	 || (*len_left < page_index->page_len)) {
6157 
6158 
6159 		ctl_set_invalid_field(ctsio,
6160 				      /*sks_valid*/ 1,
6161 				      /*command*/ 0,
6162 				      /*field*/ *len_used + page_len_offset,
6163 				      /*bit_valid*/ 0,
6164 				      /*bit*/ 0);
6165 		free(ctsio->kern_data_ptr, M_CTL);
6166 		ctl_done((union ctl_io *)ctsio);
6167 		return (CTL_RETVAL_COMPLETE);
6168 	}
6169 
6170 	/*
6171 	 * Run through the mode page, checking to make sure that the bits
6172 	 * the user changed are actually legal for him to change.
6173 	 */
6174 	for (i = 0; i < page_index->page_len; i++) {
6175 		uint8_t *user_byte, *change_mask, *current_byte;
6176 		int bad_bit;
6177 		int j;
6178 
6179 		user_byte = (uint8_t *)page_header + i;
6180 		change_mask = page_index->page_data +
6181 			      (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6182 		current_byte = page_index->page_data +
6183 			       (page_index->page_len * CTL_PAGE_CURRENT) + i;
6184 
6185 		/*
6186 		 * Check to see whether the user set any bits in this byte
6187 		 * that he is not allowed to set.
6188 		 */
6189 		if ((*user_byte & ~(*change_mask)) ==
6190 		    (*current_byte & ~(*change_mask)))
6191 			continue;
6192 
6193 		/*
6194 		 * Go through bit by bit to determine which one is illegal.
6195 		 */
6196 		bad_bit = 0;
6197 		for (j = 7; j >= 0; j--) {
6198 			if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6199 			    (((1 << i) & ~(*change_mask)) & *current_byte)) {
6200 				bad_bit = i;
6201 				break;
6202 			}
6203 		}
6204 		ctl_set_invalid_field(ctsio,
6205 				      /*sks_valid*/ 1,
6206 				      /*command*/ 0,
6207 				      /*field*/ *len_used + i,
6208 				      /*bit_valid*/ 1,
6209 				      /*bit*/ bad_bit);
6210 		free(ctsio->kern_data_ptr, M_CTL);
6211 		ctl_done((union ctl_io *)ctsio);
6212 		return (CTL_RETVAL_COMPLETE);
6213 	}
6214 
6215 	/*
6216 	 * Decrement these before we call the page handler, since we may
6217 	 * end up getting called back one way or another before the handler
6218 	 * returns to this context.
6219 	 */
6220 	*len_left -= page_index->page_len;
6221 	*len_used += page_index->page_len;
6222 
6223 	retval = page_index->select_handler(ctsio, page_index,
6224 					    (uint8_t *)page_header);
6225 
6226 	/*
6227 	 * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6228 	 * wait until this queued command completes to finish processing
6229 	 * the mode page.  If it returns anything other than
6230 	 * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6231 	 * already set the sense information, freed the data pointer, and
6232 	 * completed the io for us.
6233 	 */
6234 	if (retval != CTL_RETVAL_COMPLETE)
6235 		goto bailout_no_done;
6236 
6237 	/*
6238 	 * If the initiator sent us more than one page, parse the next one.
6239 	 */
6240 	if (*len_left > 0)
6241 		goto do_next_page;
6242 
6243 	ctl_set_success(ctsio);
6244 	free(ctsio->kern_data_ptr, M_CTL);
6245 	ctl_done((union ctl_io *)ctsio);
6246 
6247 bailout_no_done:
6248 
6249 	return (CTL_RETVAL_COMPLETE);
6250 
6251 }
6252 
6253 int
6254 ctl_mode_select(struct ctl_scsiio *ctsio)
6255 {
6256 	int param_len, pf, sp;
6257 	int header_size, bd_len;
6258 	int len_left, len_used;
6259 	struct ctl_page_index *page_index;
6260 	struct ctl_lun *lun;
6261 	int control_dev, page_len;
6262 	union ctl_modepage_info *modepage_info;
6263 	int retval;
6264 
6265 	pf = 0;
6266 	sp = 0;
6267 	page_len = 0;
6268 	len_used = 0;
6269 	len_left = 0;
6270 	retval = 0;
6271 	bd_len = 0;
6272 	page_index = NULL;
6273 
6274 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6275 
6276 	if (lun->be_lun->lun_type != T_DIRECT)
6277 		control_dev = 1;
6278 	else
6279 		control_dev = 0;
6280 
6281 	switch (ctsio->cdb[0]) {
6282 	case MODE_SELECT_6: {
6283 		struct scsi_mode_select_6 *cdb;
6284 
6285 		cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6286 
6287 		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6288 		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6289 
6290 		param_len = cdb->length;
6291 		header_size = sizeof(struct scsi_mode_header_6);
6292 		break;
6293 	}
6294 	case MODE_SELECT_10: {
6295 		struct scsi_mode_select_10 *cdb;
6296 
6297 		cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6298 
6299 		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6300 		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6301 
6302 		param_len = scsi_2btoul(cdb->length);
6303 		header_size = sizeof(struct scsi_mode_header_10);
6304 		break;
6305 	}
6306 	default:
6307 		ctl_set_invalid_opcode(ctsio);
6308 		ctl_done((union ctl_io *)ctsio);
6309 		return (CTL_RETVAL_COMPLETE);
6310 		break; /* NOTREACHED */
6311 	}
6312 
6313 	/*
6314 	 * From SPC-3:
6315 	 * "A parameter list length of zero indicates that the Data-Out Buffer
6316 	 * shall be empty. This condition shall not be considered as an error."
6317 	 */
6318 	if (param_len == 0) {
6319 		ctl_set_success(ctsio);
6320 		ctl_done((union ctl_io *)ctsio);
6321 		return (CTL_RETVAL_COMPLETE);
6322 	}
6323 
6324 	/*
6325 	 * Since we'll hit this the first time through, prior to
6326 	 * allocation, we don't need to free a data buffer here.
6327 	 */
6328 	if (param_len < header_size) {
6329 		ctl_set_param_len_error(ctsio);
6330 		ctl_done((union ctl_io *)ctsio);
6331 		return (CTL_RETVAL_COMPLETE);
6332 	}
6333 
6334 	/*
6335 	 * Allocate the data buffer and grab the user's data.  In theory,
6336 	 * we shouldn't have to sanity check the parameter list length here
6337 	 * because the maximum size is 64K.  We should be able to malloc
6338 	 * that much without too many problems.
6339 	 */
6340 	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6341 		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6342 		ctsio->kern_data_len = param_len;
6343 		ctsio->kern_total_len = param_len;
6344 		ctsio->kern_data_resid = 0;
6345 		ctsio->kern_rel_offset = 0;
6346 		ctsio->kern_sg_entries = 0;
6347 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6348 		ctsio->be_move_done = ctl_config_move_done;
6349 		ctl_datamove((union ctl_io *)ctsio);
6350 
6351 		return (CTL_RETVAL_COMPLETE);
6352 	}
6353 
6354 	switch (ctsio->cdb[0]) {
6355 	case MODE_SELECT_6: {
6356 		struct scsi_mode_header_6 *mh6;
6357 
6358 		mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6359 		bd_len = mh6->blk_desc_len;
6360 		break;
6361 	}
6362 	case MODE_SELECT_10: {
6363 		struct scsi_mode_header_10 *mh10;
6364 
6365 		mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6366 		bd_len = scsi_2btoul(mh10->blk_desc_len);
6367 		break;
6368 	}
6369 	default:
6370 		panic("Invalid CDB type %#x", ctsio->cdb[0]);
6371 		break;
6372 	}
6373 
6374 	if (param_len < (header_size + bd_len)) {
6375 		free(ctsio->kern_data_ptr, M_CTL);
6376 		ctl_set_param_len_error(ctsio);
6377 		ctl_done((union ctl_io *)ctsio);
6378 		return (CTL_RETVAL_COMPLETE);
6379 	}
6380 
6381 	/*
6382 	 * Set the IO_CONT flag, so that if this I/O gets passed to
6383 	 * ctl_config_write_done(), it'll get passed back to
6384 	 * ctl_do_mode_select() for further processing, or completion if
6385 	 * we're all done.
6386 	 */
6387 	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6388 	ctsio->io_cont = ctl_do_mode_select;
6389 
6390 	modepage_info = (union ctl_modepage_info *)
6391 		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6392 
6393 	memset(modepage_info, 0, sizeof(*modepage_info));
6394 
6395 	len_left = param_len - header_size - bd_len;
6396 	len_used = header_size + bd_len;
6397 
6398 	modepage_info->header.len_left = len_left;
6399 	modepage_info->header.len_used = len_used;
6400 
6401 	return (ctl_do_mode_select((union ctl_io *)ctsio));
6402 }
6403 
6404 int
6405 ctl_mode_sense(struct ctl_scsiio *ctsio)
6406 {
6407 	struct ctl_lun *lun;
6408 	int pc, page_code, dbd, llba, subpage;
6409 	int alloc_len, page_len, header_len, total_len;
6410 	struct scsi_mode_block_descr *block_desc;
6411 	struct ctl_page_index *page_index;
6412 	int control_dev;
6413 
6414 	dbd = 0;
6415 	llba = 0;
6416 	block_desc = NULL;
6417 	page_index = NULL;
6418 
6419 	CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6420 
6421 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6422 
6423 	if (lun->be_lun->lun_type != T_DIRECT)
6424 		control_dev = 1;
6425 	else
6426 		control_dev = 0;
6427 
6428 	switch (ctsio->cdb[0]) {
6429 	case MODE_SENSE_6: {
6430 		struct scsi_mode_sense_6 *cdb;
6431 
6432 		cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6433 
6434 		header_len = sizeof(struct scsi_mode_hdr_6);
6435 		if (cdb->byte2 & SMS_DBD)
6436 			dbd = 1;
6437 		else
6438 			header_len += sizeof(struct scsi_mode_block_descr);
6439 
6440 		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6441 		page_code = cdb->page & SMS_PAGE_CODE;
6442 		subpage = cdb->subpage;
6443 		alloc_len = cdb->length;
6444 		break;
6445 	}
6446 	case MODE_SENSE_10: {
6447 		struct scsi_mode_sense_10 *cdb;
6448 
6449 		cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6450 
6451 		header_len = sizeof(struct scsi_mode_hdr_10);
6452 
6453 		if (cdb->byte2 & SMS_DBD)
6454 			dbd = 1;
6455 		else
6456 			header_len += sizeof(struct scsi_mode_block_descr);
6457 		if (cdb->byte2 & SMS10_LLBAA)
6458 			llba = 1;
6459 		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6460 		page_code = cdb->page & SMS_PAGE_CODE;
6461 		subpage = cdb->subpage;
6462 		alloc_len = scsi_2btoul(cdb->length);
6463 		break;
6464 	}
6465 	default:
6466 		ctl_set_invalid_opcode(ctsio);
6467 		ctl_done((union ctl_io *)ctsio);
6468 		return (CTL_RETVAL_COMPLETE);
6469 		break; /* NOTREACHED */
6470 	}
6471 
6472 	/*
6473 	 * We have to make a first pass through to calculate the size of
6474 	 * the pages that match the user's query.  Then we allocate enough
6475 	 * memory to hold it, and actually copy the data into the buffer.
6476 	 */
6477 	switch (page_code) {
6478 	case SMS_ALL_PAGES_PAGE: {
6479 		int i;
6480 
6481 		page_len = 0;
6482 
6483 		/*
6484 		 * At the moment, values other than 0 and 0xff here are
6485 		 * reserved according to SPC-3.
6486 		 */
6487 		if ((subpage != SMS_SUBPAGE_PAGE_0)
6488 		 && (subpage != SMS_SUBPAGE_ALL)) {
6489 			ctl_set_invalid_field(ctsio,
6490 					      /*sks_valid*/ 1,
6491 					      /*command*/ 1,
6492 					      /*field*/ 3,
6493 					      /*bit_valid*/ 0,
6494 					      /*bit*/ 0);
6495 			ctl_done((union ctl_io *)ctsio);
6496 			return (CTL_RETVAL_COMPLETE);
6497 		}
6498 
6499 		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6500 			if ((control_dev != 0)
6501 			 && (lun->mode_pages.index[i].page_flags &
6502 			     CTL_PAGE_FLAG_DISK_ONLY))
6503 				continue;
6504 
6505 			/*
6506 			 * We don't use this subpage if the user didn't
6507 			 * request all subpages.
6508 			 */
6509 			if ((lun->mode_pages.index[i].subpage != 0)
6510 			 && (subpage == SMS_SUBPAGE_PAGE_0))
6511 				continue;
6512 
6513 #if 0
6514 			printf("found page %#x len %d\n",
6515 			       lun->mode_pages.index[i].page_code &
6516 			       SMPH_PC_MASK,
6517 			       lun->mode_pages.index[i].page_len);
6518 #endif
6519 			page_len += lun->mode_pages.index[i].page_len;
6520 		}
6521 		break;
6522 	}
6523 	default: {
6524 		int i;
6525 
6526 		page_len = 0;
6527 
6528 		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6529 			/* Look for the right page code */
6530 			if ((lun->mode_pages.index[i].page_code &
6531 			     SMPH_PC_MASK) != page_code)
6532 				continue;
6533 
6534 			/* Look for the right subpage or the subpage wildcard*/
6535 			if ((lun->mode_pages.index[i].subpage != subpage)
6536 			 && (subpage != SMS_SUBPAGE_ALL))
6537 				continue;
6538 
6539 			/* Make sure the page is supported for this dev type */
6540 			if ((control_dev != 0)
6541 			 && (lun->mode_pages.index[i].page_flags &
6542 			     CTL_PAGE_FLAG_DISK_ONLY))
6543 				continue;
6544 
6545 #if 0
6546 			printf("found page %#x len %d\n",
6547 			       lun->mode_pages.index[i].page_code &
6548 			       SMPH_PC_MASK,
6549 			       lun->mode_pages.index[i].page_len);
6550 #endif
6551 
6552 			page_len += lun->mode_pages.index[i].page_len;
6553 		}
6554 
6555 		if (page_len == 0) {
6556 			ctl_set_invalid_field(ctsio,
6557 					      /*sks_valid*/ 1,
6558 					      /*command*/ 1,
6559 					      /*field*/ 2,
6560 					      /*bit_valid*/ 1,
6561 					      /*bit*/ 5);
6562 			ctl_done((union ctl_io *)ctsio);
6563 			return (CTL_RETVAL_COMPLETE);
6564 		}
6565 		break;
6566 	}
6567 	}
6568 
6569 	total_len = header_len + page_len;
6570 #if 0
6571 	printf("header_len = %d, page_len = %d, total_len = %d\n",
6572 	       header_len, page_len, total_len);
6573 #endif
6574 
6575 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6576 	ctsio->kern_sg_entries = 0;
6577 	ctsio->kern_data_resid = 0;
6578 	ctsio->kern_rel_offset = 0;
6579 	if (total_len < alloc_len) {
6580 		ctsio->residual = alloc_len - total_len;
6581 		ctsio->kern_data_len = total_len;
6582 		ctsio->kern_total_len = total_len;
6583 	} else {
6584 		ctsio->residual = 0;
6585 		ctsio->kern_data_len = alloc_len;
6586 		ctsio->kern_total_len = alloc_len;
6587 	}
6588 
6589 	switch (ctsio->cdb[0]) {
6590 	case MODE_SENSE_6: {
6591 		struct scsi_mode_hdr_6 *header;
6592 
6593 		header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6594 
6595 		header->datalen = ctl_min(total_len - 1, 254);
6596 
6597 		if (dbd)
6598 			header->block_descr_len = 0;
6599 		else
6600 			header->block_descr_len =
6601 				sizeof(struct scsi_mode_block_descr);
6602 		block_desc = (struct scsi_mode_block_descr *)&header[1];
6603 		break;
6604 	}
6605 	case MODE_SENSE_10: {
6606 		struct scsi_mode_hdr_10 *header;
6607 		int datalen;
6608 
6609 		header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6610 
6611 		datalen = ctl_min(total_len - 2, 65533);
6612 		scsi_ulto2b(datalen, header->datalen);
6613 		if (dbd)
6614 			scsi_ulto2b(0, header->block_descr_len);
6615 		else
6616 			scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
6617 				    header->block_descr_len);
6618 		block_desc = (struct scsi_mode_block_descr *)&header[1];
6619 		break;
6620 	}
6621 	default:
6622 		panic("invalid CDB type %#x", ctsio->cdb[0]);
6623 		break; /* NOTREACHED */
6624 	}
6625 
6626 	/*
6627 	 * If we've got a disk, use its blocksize in the block
6628 	 * descriptor.  Otherwise, just set it to 0.
6629 	 */
6630 	if (dbd == 0) {
6631 		if (control_dev != 0)
6632 			scsi_ulto3b(lun->be_lun->blocksize,
6633 				    block_desc->block_len);
6634 		else
6635 			scsi_ulto3b(0, block_desc->block_len);
6636 	}
6637 
6638 	switch (page_code) {
6639 	case SMS_ALL_PAGES_PAGE: {
6640 		int i, data_used;
6641 
6642 		data_used = header_len;
6643 		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6644 			struct ctl_page_index *page_index;
6645 
6646 			page_index = &lun->mode_pages.index[i];
6647 
6648 			if ((control_dev != 0)
6649 			 && (page_index->page_flags &
6650 			    CTL_PAGE_FLAG_DISK_ONLY))
6651 				continue;
6652 
6653 			/*
6654 			 * We don't use this subpage if the user didn't
6655 			 * request all subpages.  We already checked (above)
6656 			 * to make sure the user only specified a subpage
6657 			 * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
6658 			 */
6659 			if ((page_index->subpage != 0)
6660 			 && (subpage == SMS_SUBPAGE_PAGE_0))
6661 				continue;
6662 
6663 			/*
6664 			 * Call the handler, if it exists, to update the
6665 			 * page to the latest values.
6666 			 */
6667 			if (page_index->sense_handler != NULL)
6668 				page_index->sense_handler(ctsio, page_index,pc);
6669 
6670 			memcpy(ctsio->kern_data_ptr + data_used,
6671 			       page_index->page_data +
6672 			       (page_index->page_len * pc),
6673 			       page_index->page_len);
6674 			data_used += page_index->page_len;
6675 		}
6676 		break;
6677 	}
6678 	default: {
6679 		int i, data_used;
6680 
6681 		data_used = header_len;
6682 
6683 		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6684 			struct ctl_page_index *page_index;
6685 
6686 			page_index = &lun->mode_pages.index[i];
6687 
6688 			/* Look for the right page code */
6689 			if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6690 				continue;
6691 
6692 			/* Look for the right subpage or the subpage wildcard*/
6693 			if ((page_index->subpage != subpage)
6694 			 && (subpage != SMS_SUBPAGE_ALL))
6695 				continue;
6696 
6697 			/* Make sure the page is supported for this dev type */
6698 			if ((control_dev != 0)
6699 			 && (page_index->page_flags &
6700 			     CTL_PAGE_FLAG_DISK_ONLY))
6701 				continue;
6702 
6703 			/*
6704 			 * Call the handler, if it exists, to update the
6705 			 * page to the latest values.
6706 			 */
6707 			if (page_index->sense_handler != NULL)
6708 				page_index->sense_handler(ctsio, page_index,pc);
6709 
6710 			memcpy(ctsio->kern_data_ptr + data_used,
6711 			       page_index->page_data +
6712 			       (page_index->page_len * pc),
6713 			       page_index->page_len);
6714 			data_used += page_index->page_len;
6715 		}
6716 		break;
6717 	}
6718 	}
6719 
6720 	ctsio->scsi_status = SCSI_STATUS_OK;
6721 
6722 	ctsio->be_move_done = ctl_config_move_done;
6723 	ctl_datamove((union ctl_io *)ctsio);
6724 
6725 	return (CTL_RETVAL_COMPLETE);
6726 }
6727 
6728 int
6729 ctl_read_capacity(struct ctl_scsiio *ctsio)
6730 {
6731 	struct scsi_read_capacity *cdb;
6732 	struct scsi_read_capacity_data *data;
6733 	struct ctl_lun *lun;
6734 	uint32_t lba;
6735 
6736 	CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
6737 
6738 	cdb = (struct scsi_read_capacity *)ctsio->cdb;
6739 
6740 	lba = scsi_4btoul(cdb->addr);
6741 	if (((cdb->pmi & SRC_PMI) == 0)
6742 	 && (lba != 0)) {
6743 		ctl_set_invalid_field(/*ctsio*/ ctsio,
6744 				      /*sks_valid*/ 1,
6745 				      /*command*/ 1,
6746 				      /*field*/ 2,
6747 				      /*bit_valid*/ 0,
6748 				      /*bit*/ 0);
6749 		ctl_done((union ctl_io *)ctsio);
6750 		return (CTL_RETVAL_COMPLETE);
6751 	}
6752 
6753 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6754 
6755 	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6756 	data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
6757 	ctsio->residual = 0;
6758 	ctsio->kern_data_len = sizeof(*data);
6759 	ctsio->kern_total_len = sizeof(*data);
6760 	ctsio->kern_data_resid = 0;
6761 	ctsio->kern_rel_offset = 0;
6762 	ctsio->kern_sg_entries = 0;
6763 
6764 	/*
6765 	 * If the maximum LBA is greater than 0xfffffffe, the user must
6766 	 * issue a SERVICE ACTION IN (16) command, with the read capacity
6767 	 * serivce action set.
6768 	 */
6769 	if (lun->be_lun->maxlba > 0xfffffffe)
6770 		scsi_ulto4b(0xffffffff, data->addr);
6771 	else
6772 		scsi_ulto4b(lun->be_lun->maxlba, data->addr);
6773 
6774 	/*
6775 	 * XXX KDM this may not be 512 bytes...
6776 	 */
6777 	scsi_ulto4b(lun->be_lun->blocksize, data->length);
6778 
6779 	ctsio->scsi_status = SCSI_STATUS_OK;
6780 
6781 	ctsio->be_move_done = ctl_config_move_done;
6782 	ctl_datamove((union ctl_io *)ctsio);
6783 
6784 	return (CTL_RETVAL_COMPLETE);
6785 }
6786 
6787 static int
6788 ctl_read_capacity_16(struct ctl_scsiio *ctsio)
6789 {
6790 	struct scsi_read_capacity_16 *cdb;
6791 	struct scsi_read_capacity_data_long *data;
6792 	struct ctl_lun *lun;
6793 	uint64_t lba;
6794 	uint32_t alloc_len;
6795 
6796 	CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
6797 
6798 	cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
6799 
6800 	alloc_len = scsi_4btoul(cdb->alloc_len);
6801 	lba = scsi_8btou64(cdb->addr);
6802 
6803 	if ((cdb->reladr & SRC16_PMI)
6804 	 && (lba != 0)) {
6805 		ctl_set_invalid_field(/*ctsio*/ ctsio,
6806 				      /*sks_valid*/ 1,
6807 				      /*command*/ 1,
6808 				      /*field*/ 2,
6809 				      /*bit_valid*/ 0,
6810 				      /*bit*/ 0);
6811 		ctl_done((union ctl_io *)ctsio);
6812 		return (CTL_RETVAL_COMPLETE);
6813 	}
6814 
6815 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6816 
6817 	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6818 	data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
6819 
6820 	if (sizeof(*data) < alloc_len) {
6821 		ctsio->residual = alloc_len - sizeof(*data);
6822 		ctsio->kern_data_len = sizeof(*data);
6823 		ctsio->kern_total_len = sizeof(*data);
6824 	} else {
6825 		ctsio->residual = 0;
6826 		ctsio->kern_data_len = alloc_len;
6827 		ctsio->kern_total_len = alloc_len;
6828 	}
6829 	ctsio->kern_data_resid = 0;
6830 	ctsio->kern_rel_offset = 0;
6831 	ctsio->kern_sg_entries = 0;
6832 
6833 	scsi_u64to8b(lun->be_lun->maxlba, data->addr);
6834 	/* XXX KDM this may not be 512 bytes... */
6835 	scsi_ulto4b(lun->be_lun->blocksize, data->length);
6836 
6837 	ctsio->scsi_status = SCSI_STATUS_OK;
6838 
6839 	ctsio->be_move_done = ctl_config_move_done;
6840 	ctl_datamove((union ctl_io *)ctsio);
6841 
6842 	return (CTL_RETVAL_COMPLETE);
6843 }
6844 
6845 int
6846 ctl_service_action_in(struct ctl_scsiio *ctsio)
6847 {
6848 	struct scsi_service_action_in *cdb;
6849 	int retval;
6850 
6851 	CTL_DEBUG_PRINT(("ctl_service_action_in\n"));
6852 
6853 	cdb = (struct scsi_service_action_in *)ctsio->cdb;
6854 
6855 	retval = CTL_RETVAL_COMPLETE;
6856 
6857 	switch (cdb->service_action) {
6858 	case SRC16_SERVICE_ACTION:
6859 		retval = ctl_read_capacity_16(ctsio);
6860 		break;
6861 	default:
6862 		ctl_set_invalid_field(/*ctsio*/ ctsio,
6863 				      /*sks_valid*/ 1,
6864 				      /*command*/ 1,
6865 				      /*field*/ 1,
6866 				      /*bit_valid*/ 1,
6867 				      /*bit*/ 4);
6868 		ctl_done((union ctl_io *)ctsio);
6869 		break;
6870 	}
6871 
6872 	return (retval);
6873 }
6874 
6875 int
6876 ctl_maintenance_in(struct ctl_scsiio *ctsio)
6877 {
6878 	struct scsi_maintenance_in *cdb;
6879 	int retval;
6880 	int alloc_len, total_len = 0;
6881 	int num_target_port_groups;
6882 	struct ctl_lun *lun;
6883 	struct ctl_softc *softc;
6884 	struct scsi_target_group_data *rtg_ptr;
6885 	struct scsi_target_port_group_descriptor *tpg_desc_ptr1, *tpg_desc_ptr2;
6886 	struct scsi_target_port_descriptor  *tp_desc_ptr1_1, *tp_desc_ptr1_2,
6887 	                                    *tp_desc_ptr2_1, *tp_desc_ptr2_2;
6888 
6889 	CTL_DEBUG_PRINT(("ctl_maintenance_in\n"));
6890 
6891 	cdb = (struct scsi_maintenance_in *)ctsio->cdb;
6892 	softc = control_softc;
6893 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6894 
6895 	retval = CTL_RETVAL_COMPLETE;
6896 	mtx_lock(&softc->ctl_lock);
6897 
6898 	if ((cdb->byte2 & SERVICE_ACTION_MASK) != SA_RPRT_TRGT_GRP) {
6899 		ctl_set_invalid_field(/*ctsio*/ ctsio,
6900 				      /*sks_valid*/ 1,
6901 				      /*command*/ 1,
6902 				      /*field*/ 1,
6903 				      /*bit_valid*/ 1,
6904 				      /*bit*/ 4);
6905 		ctl_done((union ctl_io *)ctsio);
6906 		return(retval);
6907 	}
6908 
6909 	if (ctl_is_single)
6910         	num_target_port_groups = NUM_TARGET_PORT_GROUPS - 1;
6911 	else
6912         	num_target_port_groups = NUM_TARGET_PORT_GROUPS;
6913 
6914 	total_len = sizeof(struct scsi_target_group_data) +
6915 		sizeof(struct scsi_target_port_group_descriptor) *
6916 		num_target_port_groups +
6917 		sizeof(struct scsi_target_port_descriptor) *
6918 		NUM_PORTS_PER_GRP * num_target_port_groups;
6919 
6920 	alloc_len = scsi_4btoul(cdb->length);
6921 
6922 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6923 
6924 	ctsio->kern_sg_entries = 0;
6925 
6926 	if (total_len < alloc_len) {
6927 		ctsio->residual = alloc_len - total_len;
6928 		ctsio->kern_data_len = total_len;
6929 		ctsio->kern_total_len = total_len;
6930 	} else {
6931 		ctsio->residual = 0;
6932 		ctsio->kern_data_len = alloc_len;
6933 		ctsio->kern_total_len = alloc_len;
6934 	}
6935 	ctsio->kern_data_resid = 0;
6936 	ctsio->kern_rel_offset = 0;
6937 
6938 	rtg_ptr = (struct scsi_target_group_data *)ctsio->kern_data_ptr;
6939 
6940 	tpg_desc_ptr1 = &rtg_ptr->groups[0];
6941 	tp_desc_ptr1_1 = &tpg_desc_ptr1->descriptors[0];
6942 	tp_desc_ptr1_2 = (struct scsi_target_port_descriptor *)
6943 	        &tp_desc_ptr1_1->desc_list[0];
6944 
6945 
6946 
6947 	if (ctl_is_single == 0) {
6948 		tpg_desc_ptr2 = (struct scsi_target_port_group_descriptor *)
6949 	                &tp_desc_ptr1_2->desc_list[0];
6950 		tp_desc_ptr2_1 = &tpg_desc_ptr2->descriptors[0];
6951 		tp_desc_ptr2_2 = (struct scsi_target_port_descriptor *)
6952 	        	&tp_desc_ptr2_1->desc_list[0];
6953         } else {
6954 		tpg_desc_ptr2 = NULL;
6955 		tp_desc_ptr2_1 = NULL;
6956 		tp_desc_ptr2_2 = NULL;
6957 	}
6958 
6959 	scsi_ulto4b(total_len - 4, rtg_ptr->length);
6960 	if (ctl_is_single == 0) {
6961         	if (ctsio->io_hdr.nexus.targ_port < CTL_MAX_PORTS) {
6962 			if (lun->flags & CTL_LUN_PRIMARY_SC) {
6963 				tpg_desc_ptr1->pref_state = TPG_PRIMARY;
6964 				tpg_desc_ptr2->pref_state =
6965 					TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
6966 			} else {
6967 				tpg_desc_ptr1->pref_state =
6968 					TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
6969 				tpg_desc_ptr2->pref_state = TPG_PRIMARY;
6970 			}
6971 		} else {
6972 			if (lun->flags & CTL_LUN_PRIMARY_SC) {
6973 				tpg_desc_ptr1->pref_state =
6974 					TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
6975 				tpg_desc_ptr2->pref_state = TPG_PRIMARY;
6976 			} else {
6977 				tpg_desc_ptr1->pref_state = TPG_PRIMARY;
6978 				tpg_desc_ptr2->pref_state =
6979 					TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
6980 			}
6981 		}
6982 	} else {
6983 		tpg_desc_ptr1->pref_state = TPG_PRIMARY;
6984 	}
6985 	tpg_desc_ptr1->support = 0;
6986 	tpg_desc_ptr1->target_port_group[1] = 1;
6987 	tpg_desc_ptr1->status = TPG_IMPLICIT;
6988 	tpg_desc_ptr1->target_port_count= NUM_PORTS_PER_GRP;
6989 
6990 	if (ctl_is_single == 0) {
6991 		tpg_desc_ptr2->support = 0;
6992 		tpg_desc_ptr2->target_port_group[1] = 2;
6993 		tpg_desc_ptr2->status = TPG_IMPLICIT;
6994 		tpg_desc_ptr2->target_port_count = NUM_PORTS_PER_GRP;
6995 
6996 		tp_desc_ptr1_1->relative_target_port_identifier[1] = 1;
6997 		tp_desc_ptr1_2->relative_target_port_identifier[1] = 2;
6998 
6999 		tp_desc_ptr2_1->relative_target_port_identifier[1] = 9;
7000 		tp_desc_ptr2_2->relative_target_port_identifier[1] = 10;
7001 	} else {
7002         	if (ctsio->io_hdr.nexus.targ_port < CTL_MAX_PORTS) {
7003 			tp_desc_ptr1_1->relative_target_port_identifier[1] = 1;
7004 			tp_desc_ptr1_2->relative_target_port_identifier[1] = 2;
7005 		} else {
7006 			tp_desc_ptr1_1->relative_target_port_identifier[1] = 9;
7007 			tp_desc_ptr1_2->relative_target_port_identifier[1] = 10;
7008 		}
7009 	}
7010 
7011 	mtx_unlock(&softc->ctl_lock);
7012 
7013 	ctsio->be_move_done = ctl_config_move_done;
7014 
7015 	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7016 			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7017 			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7018 			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7019 			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7020 
7021 	ctl_datamove((union ctl_io *)ctsio);
7022 	return(retval);
7023 }
7024 
7025 int
7026 ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7027 {
7028 	struct scsi_per_res_in *cdb;
7029 	int alloc_len, total_len = 0;
7030 	/* struct scsi_per_res_in_rsrv in_data; */
7031 	struct ctl_lun *lun;
7032 	struct ctl_softc *softc;
7033 
7034 	CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7035 
7036 	softc = control_softc;
7037 
7038 	cdb = (struct scsi_per_res_in *)ctsio->cdb;
7039 
7040 	alloc_len = scsi_2btoul(cdb->length);
7041 
7042 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7043 
7044 retry:
7045 	mtx_lock(&softc->ctl_lock);
7046 	switch (cdb->action) {
7047 	case SPRI_RK: /* read keys */
7048 		total_len = sizeof(struct scsi_per_res_in_keys) +
7049 			lun->pr_key_count *
7050 			sizeof(struct scsi_per_res_key);
7051 		break;
7052 	case SPRI_RR: /* read reservation */
7053 		if (lun->flags & CTL_LUN_PR_RESERVED)
7054 			total_len = sizeof(struct scsi_per_res_in_rsrv);
7055 		else
7056 			total_len = sizeof(struct scsi_per_res_in_header);
7057 		break;
7058 	case SPRI_RC: /* report capabilities */
7059 		total_len = sizeof(struct scsi_per_res_cap);
7060 		break;
7061 	case SPRI_RS: /* read full status */
7062 	default:
7063 		mtx_unlock(&softc->ctl_lock);
7064 		ctl_set_invalid_field(ctsio,
7065 				      /*sks_valid*/ 1,
7066 				      /*command*/ 1,
7067 				      /*field*/ 1,
7068 				      /*bit_valid*/ 1,
7069 				      /*bit*/ 0);
7070 		ctl_done((union ctl_io *)ctsio);
7071 		return (CTL_RETVAL_COMPLETE);
7072 		break; /* NOTREACHED */
7073 	}
7074 	mtx_unlock(&softc->ctl_lock);
7075 
7076 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7077 
7078 	if (total_len < alloc_len) {
7079 		ctsio->residual = alloc_len - total_len;
7080 		ctsio->kern_data_len = total_len;
7081 		ctsio->kern_total_len = total_len;
7082 	} else {
7083 		ctsio->residual = 0;
7084 		ctsio->kern_data_len = alloc_len;
7085 		ctsio->kern_total_len = alloc_len;
7086 	}
7087 
7088 	ctsio->kern_data_resid = 0;
7089 	ctsio->kern_rel_offset = 0;
7090 	ctsio->kern_sg_entries = 0;
7091 
7092 	mtx_lock(&softc->ctl_lock);
7093 	switch (cdb->action) {
7094 	case SPRI_RK: { // read keys
7095         struct scsi_per_res_in_keys *res_keys;
7096 		int i, key_count;
7097 
7098 		res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7099 
7100 		/*
7101 		 * We had to drop the lock to allocate our buffer, which
7102 		 * leaves time for someone to come in with another
7103 		 * persistent reservation.  (That is unlikely, though,
7104 		 * since this should be the only persistent reservation
7105 		 * command active right now.)
7106 		 */
7107 		if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7108 		    (lun->pr_key_count *
7109 		     sizeof(struct scsi_per_res_key)))){
7110 			mtx_unlock(&softc->ctl_lock);
7111 			free(ctsio->kern_data_ptr, M_CTL);
7112 			printf("%s: reservation length changed, retrying\n",
7113 			       __func__);
7114 			goto retry;
7115 		}
7116 
7117 		scsi_ulto4b(lun->PRGeneration, res_keys->header.generation);
7118 
7119 		scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7120 			     lun->pr_key_count, res_keys->header.length);
7121 
7122 		for (i = 0, key_count = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7123 			if (!lun->per_res[i].registered)
7124 				continue;
7125 
7126 			/*
7127 			 * We used lun->pr_key_count to calculate the
7128 			 * size to allocate.  If it turns out the number of
7129 			 * initiators with the registered flag set is
7130 			 * larger than that (i.e. they haven't been kept in
7131 			 * sync), we've got a problem.
7132 			 */
7133 			if (key_count >= lun->pr_key_count) {
7134 #ifdef NEEDTOPORT
7135 				csevent_log(CSC_CTL | CSC_SHELF_SW |
7136 					    CTL_PR_ERROR,
7137 					    csevent_LogType_Fault,
7138 					    csevent_AlertLevel_Yellow,
7139 					    csevent_FRU_ShelfController,
7140 					    csevent_FRU_Firmware,
7141 				        csevent_FRU_Unknown,
7142 					    "registered keys %d >= key "
7143 					    "count %d", key_count,
7144 					    lun->pr_key_count);
7145 #endif
7146 				key_count++;
7147 				continue;
7148 			}
7149 			memcpy(res_keys->keys[key_count].key,
7150 			       lun->per_res[i].res_key.key,
7151 			       ctl_min(sizeof(res_keys->keys[key_count].key),
7152 			       sizeof(lun->per_res[i].res_key)));
7153 			key_count++;
7154 		}
7155 		break;
7156 	}
7157 	case SPRI_RR: { // read reservation
7158 		struct scsi_per_res_in_rsrv *res;
7159 		int tmp_len, header_only;
7160 
7161 		res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7162 
7163 		scsi_ulto4b(lun->PRGeneration, res->header.generation);
7164 
7165 		if (lun->flags & CTL_LUN_PR_RESERVED)
7166 		{
7167 			tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7168 			scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7169 				    res->header.length);
7170 			header_only = 0;
7171 		} else {
7172 			tmp_len = sizeof(struct scsi_per_res_in_header);
7173 			scsi_ulto4b(0, res->header.length);
7174 			header_only = 1;
7175 		}
7176 
7177 		/*
7178 		 * We had to drop the lock to allocate our buffer, which
7179 		 * leaves time for someone to come in with another
7180 		 * persistent reservation.  (That is unlikely, though,
7181 		 * since this should be the only persistent reservation
7182 		 * command active right now.)
7183 		 */
7184 		if (tmp_len != total_len) {
7185 			mtx_unlock(&softc->ctl_lock);
7186 			free(ctsio->kern_data_ptr, M_CTL);
7187 			printf("%s: reservation status changed, retrying\n",
7188 			       __func__);
7189 			goto retry;
7190 		}
7191 
7192 		/*
7193 		 * No reservation held, so we're done.
7194 		 */
7195 		if (header_only != 0)
7196 			break;
7197 
7198 		/*
7199 		 * If the registration is an All Registrants type, the key
7200 		 * is 0, since it doesn't really matter.
7201 		 */
7202 		if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7203 			memcpy(res->data.reservation,
7204 			       &lun->per_res[lun->pr_res_idx].res_key,
7205 			       sizeof(struct scsi_per_res_key));
7206 		}
7207 		res->data.scopetype = lun->res_type;
7208 		break;
7209 	}
7210 	case SPRI_RC:     //report capabilities
7211 	{
7212 		struct scsi_per_res_cap *res_cap;
7213 		uint16_t type_mask;
7214 
7215 		res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7216 		scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7217 		res_cap->flags2 |= SPRI_TMV;
7218 		type_mask = SPRI_TM_WR_EX_AR |
7219 			    SPRI_TM_EX_AC_RO |
7220 			    SPRI_TM_WR_EX_RO |
7221 			    SPRI_TM_EX_AC |
7222 			    SPRI_TM_WR_EX |
7223 			    SPRI_TM_EX_AC_AR;
7224 		scsi_ulto2b(type_mask, res_cap->type_mask);
7225 		break;
7226 	}
7227 	case SPRI_RS: //read full status
7228 	default:
7229 		/*
7230 		 * This is a bug, because we just checked for this above,
7231 		 * and should have returned an error.
7232 		 */
7233 		panic("Invalid PR type %x", cdb->action);
7234 		break; /* NOTREACHED */
7235 	}
7236 	mtx_unlock(&softc->ctl_lock);
7237 
7238 	ctsio->be_move_done = ctl_config_move_done;
7239 
7240 	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7241 			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7242 			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7243 			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7244 			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7245 
7246 	ctl_datamove((union ctl_io *)ctsio);
7247 
7248 	return (CTL_RETVAL_COMPLETE);
7249 }
7250 
7251 /*
7252  * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7253  * it should return.
7254  */
7255 static int
7256 ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7257 		uint64_t sa_res_key, uint8_t type, uint32_t residx,
7258 		struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7259 		struct scsi_per_res_out_parms* param)
7260 {
7261 	union ctl_ha_msg persis_io;
7262 	int retval, i;
7263 	int isc_retval;
7264 
7265 	retval = 0;
7266 
7267 	if (sa_res_key == 0) {
7268 		mtx_lock(&softc->ctl_lock);
7269 		if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7270 			/* validate scope and type */
7271 			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7272 			     SPR_LU_SCOPE) {
7273 				mtx_unlock(&softc->ctl_lock);
7274 				ctl_set_invalid_field(/*ctsio*/ ctsio,
7275 						      /*sks_valid*/ 1,
7276 						      /*command*/ 1,
7277 						      /*field*/ 2,
7278 						      /*bit_valid*/ 1,
7279 						      /*bit*/ 4);
7280 				ctl_done((union ctl_io *)ctsio);
7281 				return (1);
7282 			}
7283 
7284 		        if (type>8 || type==2 || type==4 || type==0) {
7285 				mtx_unlock(&softc->ctl_lock);
7286 				ctl_set_invalid_field(/*ctsio*/ ctsio,
7287        	           				      /*sks_valid*/ 1,
7288 						      /*command*/ 1,
7289 						      /*field*/ 2,
7290 						      /*bit_valid*/ 1,
7291 						      /*bit*/ 0);
7292 				ctl_done((union ctl_io *)ctsio);
7293 				return (1);
7294 		        }
7295 
7296 			/* temporarily unregister this nexus */
7297 			lun->per_res[residx].registered = 0;
7298 
7299 			/*
7300 			 * Unregister everybody else and build UA for
7301 			 * them
7302 			 */
7303 			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7304 				if (lun->per_res[i].registered == 0)
7305 					continue;
7306 
7307 				if (!persis_offset
7308 				 && i <CTL_MAX_INITIATORS)
7309 					lun->pending_sense[i].ua_pending |=
7310 						CTL_UA_REG_PREEMPT;
7311 				else if (persis_offset
7312 				      && i >= persis_offset)
7313 					lun->pending_sense[i-persis_offset
7314 						].ua_pending |=
7315 						CTL_UA_REG_PREEMPT;
7316 				lun->per_res[i].registered = 0;
7317 				memset(&lun->per_res[i].res_key, 0,
7318 				       sizeof(struct scsi_per_res_key));
7319 			}
7320 			lun->per_res[residx].registered = 1;
7321 			lun->pr_key_count = 1;
7322 			lun->res_type = type;
7323 			if (lun->res_type != SPR_TYPE_WR_EX_AR
7324 			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7325 				lun->pr_res_idx = residx;
7326 
7327 			mtx_unlock(&softc->ctl_lock);
7328 			/* send msg to other side */
7329 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7330 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7331 			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7332 			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7333 			persis_io.pr.pr_info.res_type = type;
7334 			memcpy(persis_io.pr.pr_info.sa_res_key,
7335 			       param->serv_act_res_key,
7336 			       sizeof(param->serv_act_res_key));
7337 			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7338 			     &persis_io, sizeof(persis_io), 0)) >
7339 			     CTL_HA_STATUS_SUCCESS) {
7340 				printf("CTL:Persis Out error returned "
7341 				       "from ctl_ha_msg_send %d\n",
7342 				       isc_retval);
7343 			}
7344 		} else {
7345 			/* not all registrants */
7346 			mtx_unlock(&softc->ctl_lock);
7347 			free(ctsio->kern_data_ptr, M_CTL);
7348 			ctl_set_invalid_field(ctsio,
7349 					      /*sks_valid*/ 1,
7350 					      /*command*/ 0,
7351 					      /*field*/ 8,
7352 					      /*bit_valid*/ 0,
7353 					      /*bit*/ 0);
7354 			ctl_done((union ctl_io *)ctsio);
7355 			return (1);
7356 		}
7357 	} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7358 		|| !(lun->flags & CTL_LUN_PR_RESERVED)) {
7359 		int found = 0;
7360 
7361 		mtx_lock(&softc->ctl_lock);
7362 		if (res_key == sa_res_key) {
7363 			/* special case */
7364 			/*
7365 			 * The spec implies this is not good but doesn't
7366 			 * say what to do. There are two choices either
7367 			 * generate a res conflict or check condition
7368 			 * with illegal field in parameter data. Since
7369 			 * that is what is done when the sa_res_key is
7370 			 * zero I'll take that approach since this has
7371 			 * to do with the sa_res_key.
7372 			 */
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 
7385 		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7386 			if (lun->per_res[i].registered
7387 			 && memcmp(param->serv_act_res_key,
7388 			    lun->per_res[i].res_key.key,
7389 			    sizeof(struct scsi_per_res_key)) != 0)
7390 				continue;
7391 
7392 			found = 1;
7393 			lun->per_res[i].registered = 0;
7394 			memset(&lun->per_res[i].res_key, 0,
7395 			       sizeof(struct scsi_per_res_key));
7396 			lun->pr_key_count--;
7397 
7398 			if (!persis_offset
7399 			 && i < CTL_MAX_INITIATORS)
7400 				lun->pending_sense[i].ua_pending |=
7401 					CTL_UA_REG_PREEMPT;
7402 			else if (persis_offset
7403 			      && i >= persis_offset)
7404 				lun->pending_sense[i-persis_offset].ua_pending|=
7405 					CTL_UA_REG_PREEMPT;
7406 		}
7407 		mtx_unlock(&softc->ctl_lock);
7408 		if (!found) {
7409 			free(ctsio->kern_data_ptr, M_CTL);
7410 			ctl_set_reservation_conflict(ctsio);
7411 			ctl_done((union ctl_io *)ctsio);
7412 			return (CTL_RETVAL_COMPLETE);
7413 		}
7414 		/* send msg to other side */
7415 		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7416 		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7417 		persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7418 		persis_io.pr.pr_info.residx = lun->pr_res_idx;
7419 		persis_io.pr.pr_info.res_type = type;
7420 		memcpy(persis_io.pr.pr_info.sa_res_key,
7421 		       param->serv_act_res_key,
7422 		       sizeof(param->serv_act_res_key));
7423 		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7424 		     &persis_io, sizeof(persis_io), 0)) >
7425 		     CTL_HA_STATUS_SUCCESS) {
7426 			printf("CTL:Persis Out error returned from "
7427 			       "ctl_ha_msg_send %d\n", isc_retval);
7428 		}
7429 	} else {
7430 		/* Reserved but not all registrants */
7431 		/* sa_res_key is res holder */
7432 		if (memcmp(param->serv_act_res_key,
7433                    lun->per_res[lun->pr_res_idx].res_key.key,
7434                    sizeof(struct scsi_per_res_key)) == 0) {
7435 			/* validate scope and type */
7436 			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7437 			     SPR_LU_SCOPE) {
7438 				ctl_set_invalid_field(/*ctsio*/ ctsio,
7439 						      /*sks_valid*/ 1,
7440 						      /*command*/ 1,
7441 						      /*field*/ 2,
7442 						      /*bit_valid*/ 1,
7443 						      /*bit*/ 4);
7444 				ctl_done((union ctl_io *)ctsio);
7445 				return (1);
7446 			}
7447 
7448 			if (type>8 || type==2 || type==4 || type==0) {
7449 				ctl_set_invalid_field(/*ctsio*/ ctsio,
7450 						      /*sks_valid*/ 1,
7451 						      /*command*/ 1,
7452 						      /*field*/ 2,
7453 						      /*bit_valid*/ 1,
7454 						      /*bit*/ 0);
7455 				ctl_done((union ctl_io *)ctsio);
7456 				return (1);
7457 			}
7458 
7459 			/*
7460 			 * Do the following:
7461 			 * if sa_res_key != res_key remove all
7462 			 * registrants w/sa_res_key and generate UA
7463 			 * for these registrants(Registrations
7464 			 * Preempted) if it wasn't an exclusive
7465 			 * reservation generate UA(Reservations
7466 			 * Preempted) for all other registered nexuses
7467 			 * if the type has changed. Establish the new
7468 			 * reservation and holder. If res_key and
7469 			 * sa_res_key are the same do the above
7470 			 * except don't unregister the res holder.
7471 			 */
7472 
7473 			/*
7474 			 * Temporarily unregister so it won't get
7475 			 * removed or UA generated
7476 			 */
7477 			lun->per_res[residx].registered = 0;
7478 			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7479 				if (lun->per_res[i].registered == 0)
7480 					continue;
7481 
7482 				if (memcmp(param->serv_act_res_key,
7483 				    lun->per_res[i].res_key.key,
7484 				    sizeof(struct scsi_per_res_key)) == 0) {
7485 					lun->per_res[i].registered = 0;
7486 					memset(&lun->per_res[i].res_key,
7487 					       0,
7488 					       sizeof(struct scsi_per_res_key));
7489 					lun->pr_key_count--;
7490 
7491 					if (!persis_offset
7492 					 && i < CTL_MAX_INITIATORS)
7493 						lun->pending_sense[i
7494 							].ua_pending |=
7495 							CTL_UA_REG_PREEMPT;
7496 					else if (persis_offset
7497 					      && i >= persis_offset)
7498 						lun->pending_sense[
7499 						  i-persis_offset].ua_pending |=
7500 						  CTL_UA_REG_PREEMPT;
7501 				} else if (type != lun->res_type
7502 					&& (lun->res_type == SPR_TYPE_WR_EX_RO
7503 					 || lun->res_type ==SPR_TYPE_EX_AC_RO)){
7504 						if (!persis_offset
7505 						 && i < CTL_MAX_INITIATORS)
7506 							lun->pending_sense[i
7507 							].ua_pending |=
7508 							CTL_UA_RES_RELEASE;
7509 						else if (persis_offset
7510 						      && i >= persis_offset)
7511 							lun->pending_sense[
7512 							i-persis_offset
7513 							].ua_pending |=
7514 							CTL_UA_RES_RELEASE;
7515 				}
7516 			}
7517 			lun->per_res[residx].registered = 1;
7518 			lun->res_type = type;
7519 			if (lun->res_type != SPR_TYPE_WR_EX_AR
7520 			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7521 				lun->pr_res_idx = residx;
7522 			else
7523 				lun->pr_res_idx =
7524 					CTL_PR_ALL_REGISTRANTS;
7525 
7526 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7527 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7528 			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7529 			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7530 			persis_io.pr.pr_info.res_type = type;
7531 			memcpy(persis_io.pr.pr_info.sa_res_key,
7532 			       param->serv_act_res_key,
7533 			       sizeof(param->serv_act_res_key));
7534 			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7535 			     &persis_io, sizeof(persis_io), 0)) >
7536 			     CTL_HA_STATUS_SUCCESS) {
7537 				printf("CTL:Persis Out error returned "
7538 				       "from ctl_ha_msg_send %d\n",
7539 				       isc_retval);
7540 			}
7541 		} else {
7542 			/*
7543 			 * sa_res_key is not the res holder just
7544 			 * remove registrants
7545 			 */
7546 			int found=0;
7547 			mtx_lock(&softc->ctl_lock);
7548 
7549 			for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7550 				if (memcmp(param->serv_act_res_key,
7551 				    lun->per_res[i].res_key.key,
7552 				    sizeof(struct scsi_per_res_key)) != 0)
7553 					continue;
7554 
7555 				found = 1;
7556 				lun->per_res[i].registered = 0;
7557 				memset(&lun->per_res[i].res_key, 0,
7558 				       sizeof(struct scsi_per_res_key));
7559 				lun->pr_key_count--;
7560 
7561 				if (!persis_offset
7562 				 && i < CTL_MAX_INITIATORS)
7563 					lun->pending_sense[i].ua_pending |=
7564 						CTL_UA_REG_PREEMPT;
7565 				else if (persis_offset
7566 				      && i >= persis_offset)
7567 					lun->pending_sense[
7568 						i-persis_offset].ua_pending |=
7569 						CTL_UA_REG_PREEMPT;
7570 			}
7571 
7572 			if (!found) {
7573 				mtx_unlock(&softc->ctl_lock);
7574 				free(ctsio->kern_data_ptr, M_CTL);
7575 				ctl_set_reservation_conflict(ctsio);
7576 				ctl_done((union ctl_io *)ctsio);
7577 		        	return (1);
7578 			}
7579 			mtx_unlock(&softc->ctl_lock);
7580 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7581 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7582 			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7583 			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7584 			persis_io.pr.pr_info.res_type = type;
7585 			memcpy(persis_io.pr.pr_info.sa_res_key,
7586 			       param->serv_act_res_key,
7587 			       sizeof(param->serv_act_res_key));
7588 			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7589 			     &persis_io, sizeof(persis_io), 0)) >
7590 			     CTL_HA_STATUS_SUCCESS) {
7591 				printf("CTL:Persis Out error returned "
7592 				       "from ctl_ha_msg_send %d\n",
7593 				isc_retval);
7594 			}
7595 		}
7596 	}
7597 
7598 	lun->PRGeneration++;
7599 
7600 	return (retval);
7601 }
7602 
7603 static void
7604 ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
7605 {
7606 	int i;
7607 
7608 	if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7609 	 || lun->pr_res_idx == CTL_PR_NO_RESERVATION
7610 	 || memcmp(&lun->per_res[lun->pr_res_idx].res_key,
7611 		   msg->pr.pr_info.sa_res_key,
7612 		   sizeof(struct scsi_per_res_key)) != 0) {
7613 		uint64_t sa_res_key;
7614 		sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
7615 
7616 		if (sa_res_key == 0) {
7617 			/* temporarily unregister this nexus */
7618 			lun->per_res[msg->pr.pr_info.residx].registered = 0;
7619 
7620 			/*
7621 			 * Unregister everybody else and build UA for
7622 			 * them
7623 			 */
7624 			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7625 				if (lun->per_res[i].registered == 0)
7626 					continue;
7627 
7628 				if (!persis_offset
7629 				 && i < CTL_MAX_INITIATORS)
7630 					lun->pending_sense[i].ua_pending |=
7631 						CTL_UA_REG_PREEMPT;
7632 				else if (persis_offset && i >= persis_offset)
7633 					lun->pending_sense[i -
7634 						persis_offset].ua_pending |=
7635 						CTL_UA_REG_PREEMPT;
7636 				lun->per_res[i].registered = 0;
7637 				memset(&lun->per_res[i].res_key, 0,
7638 				       sizeof(struct scsi_per_res_key));
7639 			}
7640 
7641 			lun->per_res[msg->pr.pr_info.residx].registered = 1;
7642 			lun->pr_key_count = 1;
7643 			lun->res_type = msg->pr.pr_info.res_type;
7644 			if (lun->res_type != SPR_TYPE_WR_EX_AR
7645 			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7646 				lun->pr_res_idx = msg->pr.pr_info.residx;
7647 		} else {
7648 		        for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7649 				if (memcmp(msg->pr.pr_info.sa_res_key,
7650 		                   lun->per_res[i].res_key.key,
7651 		                   sizeof(struct scsi_per_res_key)) != 0)
7652 					continue;
7653 
7654 				lun->per_res[i].registered = 0;
7655 				memset(&lun->per_res[i].res_key, 0,
7656 				       sizeof(struct scsi_per_res_key));
7657 				lun->pr_key_count--;
7658 
7659 				if (!persis_offset
7660 				 && i < persis_offset)
7661 					lun->pending_sense[i].ua_pending |=
7662 						CTL_UA_REG_PREEMPT;
7663 				else if (persis_offset
7664 				      && i >= persis_offset)
7665 					lun->pending_sense[i -
7666 						persis_offset].ua_pending |=
7667 						CTL_UA_REG_PREEMPT;
7668 			}
7669 		}
7670 	} else {
7671 		/*
7672 		 * Temporarily unregister so it won't get removed
7673 		 * or UA generated
7674 		 */
7675 		lun->per_res[msg->pr.pr_info.residx].registered = 0;
7676 		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7677 			if (lun->per_res[i].registered == 0)
7678 				continue;
7679 
7680 			if (memcmp(msg->pr.pr_info.sa_res_key,
7681 	                   lun->per_res[i].res_key.key,
7682 	                   sizeof(struct scsi_per_res_key)) == 0) {
7683 				lun->per_res[i].registered = 0;
7684 				memset(&lun->per_res[i].res_key, 0,
7685 				       sizeof(struct scsi_per_res_key));
7686 				lun->pr_key_count--;
7687 				if (!persis_offset
7688 				 && i < CTL_MAX_INITIATORS)
7689 					lun->pending_sense[i].ua_pending |=
7690 						CTL_UA_REG_PREEMPT;
7691 				else if (persis_offset
7692 				      && i >= persis_offset)
7693 					lun->pending_sense[i -
7694 						persis_offset].ua_pending |=
7695 						CTL_UA_REG_PREEMPT;
7696 			} else if (msg->pr.pr_info.res_type != lun->res_type
7697 				&& (lun->res_type == SPR_TYPE_WR_EX_RO
7698 				 || lun->res_type == SPR_TYPE_EX_AC_RO)) {
7699 					if (!persis_offset
7700 					 && i < persis_offset)
7701 						lun->pending_sense[i
7702 							].ua_pending |=
7703 							CTL_UA_RES_RELEASE;
7704 					else if (persis_offset
7705 					      && i >= persis_offset)
7706 					lun->pending_sense[i -
7707 						persis_offset].ua_pending |=
7708 						CTL_UA_RES_RELEASE;
7709 			}
7710 		}
7711 		lun->per_res[msg->pr.pr_info.residx].registered = 1;
7712 		lun->res_type = msg->pr.pr_info.res_type;
7713 		if (lun->res_type != SPR_TYPE_WR_EX_AR
7714 		 && lun->res_type != SPR_TYPE_EX_AC_AR)
7715 			lun->pr_res_idx = msg->pr.pr_info.residx;
7716 		else
7717 			lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
7718 	}
7719 	lun->PRGeneration++;
7720 
7721 }
7722 
7723 
7724 int
7725 ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
7726 {
7727 	int retval;
7728 	int isc_retval;
7729 	u_int32_t param_len;
7730 	struct scsi_per_res_out *cdb;
7731 	struct ctl_lun *lun;
7732 	struct scsi_per_res_out_parms* param;
7733 	struct ctl_softc *softc;
7734 	uint32_t residx;
7735 	uint64_t res_key, sa_res_key;
7736 	uint8_t type;
7737 	union ctl_ha_msg persis_io;
7738 	int    i;
7739 
7740 	CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
7741 
7742 	retval = CTL_RETVAL_COMPLETE;
7743 
7744 	softc = control_softc;
7745 
7746 	cdb = (struct scsi_per_res_out *)ctsio->cdb;
7747 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7748 
7749 	/*
7750 	 * We only support whole-LUN scope.  The scope & type are ignored for
7751 	 * register, register and ignore existing key and clear.
7752 	 * We sometimes ignore scope and type on preempts too!!
7753 	 * Verify reservation type here as well.
7754 	 */
7755 	type = cdb->scope_type & SPR_TYPE_MASK;
7756 	if ((cdb->action == SPRO_RESERVE)
7757 	 || (cdb->action == SPRO_RELEASE)) {
7758 		if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
7759 			ctl_set_invalid_field(/*ctsio*/ ctsio,
7760 					      /*sks_valid*/ 1,
7761 					      /*command*/ 1,
7762 					      /*field*/ 2,
7763 					      /*bit_valid*/ 1,
7764 					      /*bit*/ 4);
7765 			ctl_done((union ctl_io *)ctsio);
7766 			return (CTL_RETVAL_COMPLETE);
7767 		}
7768 
7769 		if (type>8 || type==2 || type==4 || type==0) {
7770 			ctl_set_invalid_field(/*ctsio*/ ctsio,
7771 					      /*sks_valid*/ 1,
7772 					      /*command*/ 1,
7773 					      /*field*/ 2,
7774 					      /*bit_valid*/ 1,
7775 					      /*bit*/ 0);
7776 			ctl_done((union ctl_io *)ctsio);
7777 			return (CTL_RETVAL_COMPLETE);
7778 		}
7779 	}
7780 
7781 	switch (cdb->action & SPRO_ACTION_MASK) {
7782 	case SPRO_REGISTER:
7783 	case SPRO_RESERVE:
7784 	case SPRO_RELEASE:
7785 	case SPRO_CLEAR:
7786 	case SPRO_PREEMPT:
7787 	case SPRO_REG_IGNO:
7788 		break;
7789 	case SPRO_REG_MOVE:
7790 	case SPRO_PRE_ABO:
7791 	default:
7792 		ctl_set_invalid_field(/*ctsio*/ ctsio,
7793 				      /*sks_valid*/ 1,
7794 				      /*command*/ 1,
7795 				      /*field*/ 1,
7796 				      /*bit_valid*/ 1,
7797 				      /*bit*/ 0);
7798 		ctl_done((union ctl_io *)ctsio);
7799 		return (CTL_RETVAL_COMPLETE);
7800 		break; /* NOTREACHED */
7801 	}
7802 
7803 	param_len = scsi_4btoul(cdb->length);
7804 
7805 	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
7806 		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
7807 		ctsio->kern_data_len = param_len;
7808 		ctsio->kern_total_len = param_len;
7809 		ctsio->kern_data_resid = 0;
7810 		ctsio->kern_rel_offset = 0;
7811 		ctsio->kern_sg_entries = 0;
7812 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7813 		ctsio->be_move_done = ctl_config_move_done;
7814 		ctl_datamove((union ctl_io *)ctsio);
7815 
7816 		return (CTL_RETVAL_COMPLETE);
7817 	}
7818 
7819 	param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
7820 
7821 	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
7822 	res_key = scsi_8btou64(param->res_key.key);
7823 	sa_res_key = scsi_8btou64(param->serv_act_res_key);
7824 
7825 	/*
7826 	 * Validate the reservation key here except for SPRO_REG_IGNO
7827 	 * This must be done for all other service actions
7828 	 */
7829 	if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
7830 		mtx_lock(&softc->ctl_lock);
7831 		if (lun->per_res[residx].registered) {
7832 		    if (memcmp(param->res_key.key,
7833 			       lun->per_res[residx].res_key.key,
7834 			       ctl_min(sizeof(param->res_key),
7835 			       sizeof(lun->per_res[residx].res_key))) != 0) {
7836 				/*
7837 				 * The current key passed in doesn't match
7838 				 * the one the initiator previously
7839 				 * registered.
7840 				 */
7841 				mtx_unlock(&softc->ctl_lock);
7842 				free(ctsio->kern_data_ptr, M_CTL);
7843 				ctl_set_reservation_conflict(ctsio);
7844 				ctl_done((union ctl_io *)ctsio);
7845 				return (CTL_RETVAL_COMPLETE);
7846 			}
7847 		} else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
7848 		    /*
7849 			 * We are not registered
7850 			 */
7851 			mtx_unlock(&softc->ctl_lock);
7852 			free(ctsio->kern_data_ptr, M_CTL);
7853 			ctl_set_reservation_conflict(ctsio);
7854 			ctl_done((union ctl_io *)ctsio);
7855 			return (CTL_RETVAL_COMPLETE);
7856 		} else if (res_key != 0) {
7857 			/*
7858 			 * We are not registered and trying to register but
7859 			 * the register key isn't zero.
7860 			 */
7861 			mtx_unlock(&softc->ctl_lock);
7862 			free(ctsio->kern_data_ptr, M_CTL);
7863 			ctl_set_reservation_conflict(ctsio);
7864 			ctl_done((union ctl_io *)ctsio);
7865 			return (CTL_RETVAL_COMPLETE);
7866 		}
7867 		mtx_unlock(&softc->ctl_lock);
7868 	}
7869 
7870 	switch (cdb->action & SPRO_ACTION_MASK) {
7871 	case SPRO_REGISTER:
7872 	case SPRO_REG_IGNO: {
7873 
7874 #if 0
7875 		printf("Registration received\n");
7876 #endif
7877 
7878 		/*
7879 		 * We don't support any of these options, as we report in
7880 		 * the read capabilities request (see
7881 		 * ctl_persistent_reserve_in(), above).
7882 		 */
7883 		if ((param->flags & SPR_SPEC_I_PT)
7884 		 || (param->flags & SPR_ALL_TG_PT)
7885 		 || (param->flags & SPR_APTPL)) {
7886 			int bit_ptr;
7887 
7888 			if (param->flags & SPR_APTPL)
7889 				bit_ptr = 0;
7890 			else if (param->flags & SPR_ALL_TG_PT)
7891 				bit_ptr = 2;
7892 			else /* SPR_SPEC_I_PT */
7893 				bit_ptr = 3;
7894 
7895 			free(ctsio->kern_data_ptr, M_CTL);
7896 			ctl_set_invalid_field(ctsio,
7897 					      /*sks_valid*/ 1,
7898 					      /*command*/ 0,
7899 					      /*field*/ 20,
7900 					      /*bit_valid*/ 1,
7901 					      /*bit*/ bit_ptr);
7902 			ctl_done((union ctl_io *)ctsio);
7903 			return (CTL_RETVAL_COMPLETE);
7904 		}
7905 
7906 		mtx_lock(&softc->ctl_lock);
7907 
7908 		/*
7909 		 * The initiator wants to clear the
7910 		 * key/unregister.
7911 		 */
7912 		if (sa_res_key == 0) {
7913 			if ((res_key == 0
7914 			  && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
7915 			 || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
7916 			  && !lun->per_res[residx].registered)) {
7917 				mtx_unlock(&softc->ctl_lock);
7918 				goto done;
7919 			}
7920 
7921 			lun->per_res[residx].registered = 0;
7922 			memset(&lun->per_res[residx].res_key,
7923 			       0, sizeof(lun->per_res[residx].res_key));
7924 			lun->pr_key_count--;
7925 
7926 			if (residx == lun->pr_res_idx) {
7927 				lun->flags &= ~CTL_LUN_PR_RESERVED;
7928 				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
7929 
7930 				if ((lun->res_type == SPR_TYPE_WR_EX_RO
7931 				  || lun->res_type == SPR_TYPE_EX_AC_RO)
7932 				 && lun->pr_key_count) {
7933 					/*
7934 					 * If the reservation is a registrants
7935 					 * only type we need to generate a UA
7936 					 * for other registered inits.  The
7937 					 * sense code should be RESERVATIONS
7938 					 * RELEASED
7939 					 */
7940 
7941 					for (i = 0; i < CTL_MAX_INITIATORS;i++){
7942 						if (lun->per_res[
7943 						    i+persis_offset].registered
7944 						    == 0)
7945 							continue;
7946 						lun->pending_sense[i
7947 							].ua_pending |=
7948 							CTL_UA_RES_RELEASE;
7949 					}
7950 				}
7951 				lun->res_type = 0;
7952 			} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7953 				if (lun->pr_key_count==0) {
7954 					lun->flags &= ~CTL_LUN_PR_RESERVED;
7955 					lun->res_type = 0;
7956 					lun->pr_res_idx = CTL_PR_NO_RESERVATION;
7957 				}
7958 			}
7959 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7960 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7961 			persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
7962 			persis_io.pr.pr_info.residx = residx;
7963 			if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7964 			     &persis_io, sizeof(persis_io), 0 )) >
7965 			     CTL_HA_STATUS_SUCCESS) {
7966 				printf("CTL:Persis Out error returned from "
7967 				       "ctl_ha_msg_send %d\n", isc_retval);
7968 			}
7969 			mtx_unlock(&softc->ctl_lock);
7970 		} else /* sa_res_key != 0 */ {
7971 
7972 			/*
7973 			 * If we aren't registered currently then increment
7974 			 * the key count and set the registered flag.
7975 			 */
7976 			if (!lun->per_res[residx].registered) {
7977 				lun->pr_key_count++;
7978 				lun->per_res[residx].registered = 1;
7979 			}
7980 
7981 			memcpy(&lun->per_res[residx].res_key,
7982 			       param->serv_act_res_key,
7983 			       ctl_min(sizeof(param->serv_act_res_key),
7984 			       sizeof(lun->per_res[residx].res_key)));
7985 
7986 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7987 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7988 			persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
7989 			persis_io.pr.pr_info.residx = residx;
7990 			memcpy(persis_io.pr.pr_info.sa_res_key,
7991 			       param->serv_act_res_key,
7992 			       sizeof(param->serv_act_res_key));
7993 			mtx_unlock(&softc->ctl_lock);
7994 			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7995 			     &persis_io, sizeof(persis_io), 0)) >
7996 			     CTL_HA_STATUS_SUCCESS) {
7997 				printf("CTL:Persis Out error returned from "
7998 				       "ctl_ha_msg_send %d\n", isc_retval);
7999 			}
8000 		}
8001 		lun->PRGeneration++;
8002 
8003 		break;
8004 	}
8005 	case SPRO_RESERVE:
8006 #if 0
8007                 printf("Reserve executed type %d\n", type);
8008 #endif
8009 		mtx_lock(&softc->ctl_lock);
8010 		if (lun->flags & CTL_LUN_PR_RESERVED) {
8011 			/*
8012 			 * if this isn't the reservation holder and it's
8013 			 * not a "all registrants" type or if the type is
8014 			 * different then we have a conflict
8015 			 */
8016 			if ((lun->pr_res_idx != residx
8017 			  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8018 			 || lun->res_type != type) {
8019 				mtx_unlock(&softc->ctl_lock);
8020 				free(ctsio->kern_data_ptr, M_CTL);
8021 				ctl_set_reservation_conflict(ctsio);
8022 				ctl_done((union ctl_io *)ctsio);
8023 				return (CTL_RETVAL_COMPLETE);
8024 			}
8025 		} else /* create a reservation */ {
8026 			/*
8027 			 * If it's not an "all registrants" type record
8028 			 * reservation holder
8029 			 */
8030 			if (type != SPR_TYPE_WR_EX_AR
8031 			 && type != SPR_TYPE_EX_AC_AR)
8032 				lun->pr_res_idx = residx; /* Res holder */
8033 			else
8034 				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8035 
8036 			lun->flags |= CTL_LUN_PR_RESERVED;
8037 			lun->res_type = type;
8038 
8039 			mtx_unlock(&softc->ctl_lock);
8040 
8041 			/* send msg to other side */
8042 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8043 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8044 			persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8045 			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8046 			persis_io.pr.pr_info.res_type = type;
8047 			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8048 			     &persis_io, sizeof(persis_io), 0)) >
8049 			     CTL_HA_STATUS_SUCCESS) {
8050 				printf("CTL:Persis Out error returned from "
8051 				       "ctl_ha_msg_send %d\n", isc_retval);
8052 			}
8053 		}
8054 		break;
8055 
8056 	case SPRO_RELEASE:
8057 		mtx_lock(&softc->ctl_lock);
8058 		if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8059 			/* No reservation exists return good status */
8060 			mtx_unlock(&softc->ctl_lock);
8061 			goto done;
8062 		}
8063 		/*
8064 		 * Is this nexus a reservation holder?
8065 		 */
8066 		if (lun->pr_res_idx != residx
8067 		 && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8068 			/*
8069 			 * not a res holder return good status but
8070 			 * do nothing
8071 			 */
8072 			mtx_unlock(&softc->ctl_lock);
8073 			goto done;
8074 		}
8075 
8076 		if (lun->res_type != type) {
8077 			mtx_unlock(&softc->ctl_lock);
8078 			free(ctsio->kern_data_ptr, M_CTL);
8079 			ctl_set_illegal_pr_release(ctsio);
8080 			ctl_done((union ctl_io *)ctsio);
8081 			return (CTL_RETVAL_COMPLETE);
8082 		}
8083 
8084 		/* okay to release */
8085 		lun->flags &= ~CTL_LUN_PR_RESERVED;
8086 		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8087 		lun->res_type = 0;
8088 
8089 		/*
8090 		 * if this isn't an exclusive access
8091 		 * res generate UA for all other
8092 		 * registrants.
8093 		 */
8094 		if (type != SPR_TYPE_EX_AC
8095 		 && type != SPR_TYPE_WR_EX) {
8096 			/*
8097 			 * temporarily unregister so we don't generate UA
8098 			 */
8099 			lun->per_res[residx].registered = 0;
8100 
8101 			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8102 				if (lun->per_res[i+persis_offset].registered
8103 				    == 0)
8104 					continue;
8105 				lun->pending_sense[i].ua_pending |=
8106 					CTL_UA_RES_RELEASE;
8107 			}
8108 
8109 			lun->per_res[residx].registered = 1;
8110 		}
8111 		mtx_unlock(&softc->ctl_lock);
8112 		/* Send msg to other side */
8113 		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8114 		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8115 		persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8116 		if ((isc_retval=ctl_ha_msg_send( CTL_HA_CHAN_CTL, &persis_io,
8117 		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8118 			printf("CTL:Persis Out error returned from "
8119 			       "ctl_ha_msg_send %d\n", isc_retval);
8120 		}
8121 		break;
8122 
8123 	case SPRO_CLEAR:
8124 		/* send msg to other side */
8125 
8126 		mtx_lock(&softc->ctl_lock);
8127 		lun->flags &= ~CTL_LUN_PR_RESERVED;
8128 		lun->res_type = 0;
8129 		lun->pr_key_count = 0;
8130 		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8131 
8132 
8133 		memset(&lun->per_res[residx].res_key,
8134 		       0, sizeof(lun->per_res[residx].res_key));
8135 		lun->per_res[residx].registered = 0;
8136 
8137 		for (i=0; i < 2*CTL_MAX_INITIATORS; i++)
8138 			if (lun->per_res[i].registered) {
8139 				if (!persis_offset && i < CTL_MAX_INITIATORS)
8140 					lun->pending_sense[i].ua_pending |=
8141 						CTL_UA_RES_PREEMPT;
8142 				else if (persis_offset && i >= persis_offset)
8143 					lun->pending_sense[i-persis_offset
8144 					    ].ua_pending |= CTL_UA_RES_PREEMPT;
8145 
8146 				memset(&lun->per_res[i].res_key,
8147 				       0, sizeof(struct scsi_per_res_key));
8148 				lun->per_res[i].registered = 0;
8149 			}
8150 		lun->PRGeneration++;
8151 		mtx_unlock(&softc->ctl_lock);
8152 		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8153 		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8154 		persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8155 		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8156 		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8157 			printf("CTL:Persis Out error returned from "
8158 			       "ctl_ha_msg_send %d\n", isc_retval);
8159 		}
8160 		break;
8161 
8162 	case SPRO_PREEMPT: {
8163 		int nretval;
8164 
8165 		nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8166 					  residx, ctsio, cdb, param);
8167 		if (nretval != 0)
8168 			return (CTL_RETVAL_COMPLETE);
8169 		break;
8170 	}
8171 	case SPRO_REG_MOVE:
8172 	case SPRO_PRE_ABO:
8173 	default:
8174 		free(ctsio->kern_data_ptr, M_CTL);
8175 		ctl_set_invalid_field(/*ctsio*/ ctsio,
8176 				      /*sks_valid*/ 1,
8177 				      /*command*/ 1,
8178 				      /*field*/ 1,
8179 				      /*bit_valid*/ 1,
8180 				      /*bit*/ 0);
8181 		ctl_done((union ctl_io *)ctsio);
8182 		return (CTL_RETVAL_COMPLETE);
8183 		break; /* NOTREACHED */
8184 	}
8185 
8186 done:
8187 	free(ctsio->kern_data_ptr, M_CTL);
8188 	ctl_set_success(ctsio);
8189 	ctl_done((union ctl_io *)ctsio);
8190 
8191 	return (retval);
8192 }
8193 
8194 /*
8195  * This routine is for handling a message from the other SC pertaining to
8196  * persistent reserve out. All the error checking will have been done
8197  * so only perorming the action need be done here to keep the two
8198  * in sync.
8199  */
8200 static void
8201 ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8202 {
8203 	struct ctl_lun *lun;
8204 	struct ctl_softc *softc;
8205 	int i;
8206 
8207 	softc = control_softc;
8208 
8209 	mtx_lock(&softc->ctl_lock);
8210 
8211 	lun = softc->ctl_luns[msg->hdr.nexus.targ_lun];
8212 	switch(msg->pr.pr_info.action) {
8213 	case CTL_PR_REG_KEY:
8214 		if (!lun->per_res[msg->pr.pr_info.residx].registered) {
8215 			lun->per_res[msg->pr.pr_info.residx].registered = 1;
8216 			lun->pr_key_count++;
8217 		}
8218 		lun->PRGeneration++;
8219 		memcpy(&lun->per_res[msg->pr.pr_info.residx].res_key,
8220 		       msg->pr.pr_info.sa_res_key,
8221 		       sizeof(struct scsi_per_res_key));
8222 		break;
8223 
8224 	case CTL_PR_UNREG_KEY:
8225 		lun->per_res[msg->pr.pr_info.residx].registered = 0;
8226 		memset(&lun->per_res[msg->pr.pr_info.residx].res_key,
8227 		       0, sizeof(struct scsi_per_res_key));
8228 		lun->pr_key_count--;
8229 
8230 		/* XXX Need to see if the reservation has been released */
8231 		/* if so do we need to generate UA? */
8232 		if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8233 			lun->flags &= ~CTL_LUN_PR_RESERVED;
8234 			lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8235 
8236 			if ((lun->res_type == SPR_TYPE_WR_EX_RO
8237 			  || lun->res_type == SPR_TYPE_EX_AC_RO)
8238 			 && lun->pr_key_count) {
8239 				/*
8240 				 * If the reservation is a registrants
8241 				 * only type we need to generate a UA
8242 				 * for other registered inits.  The
8243 				 * sense code should be RESERVATIONS
8244 				 * RELEASED
8245 				 */
8246 
8247 				for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8248 					if (lun->per_res[i+
8249 					    persis_offset].registered == 0)
8250 						continue;
8251 
8252 					lun->pending_sense[i
8253 						].ua_pending |=
8254 						CTL_UA_RES_RELEASE;
8255 				}
8256 			}
8257 			lun->res_type = 0;
8258 		} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8259 			if (lun->pr_key_count==0) {
8260 				lun->flags &= ~CTL_LUN_PR_RESERVED;
8261 				lun->res_type = 0;
8262 				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8263 			}
8264 		}
8265 		lun->PRGeneration++;
8266 		break;
8267 
8268 	case CTL_PR_RESERVE:
8269 		lun->flags |= CTL_LUN_PR_RESERVED;
8270 		lun->res_type = msg->pr.pr_info.res_type;
8271 		lun->pr_res_idx = msg->pr.pr_info.residx;
8272 
8273 		break;
8274 
8275 	case CTL_PR_RELEASE:
8276 		/*
8277 		 * if this isn't an exclusive access res generate UA for all
8278 		 * other registrants.
8279 		 */
8280 		if (lun->res_type != SPR_TYPE_EX_AC
8281 		 && lun->res_type != SPR_TYPE_WR_EX) {
8282 			for (i = 0; i < CTL_MAX_INITIATORS; i++)
8283 				if (lun->per_res[i+persis_offset].registered)
8284 					lun->pending_sense[i].ua_pending |=
8285 						CTL_UA_RES_RELEASE;
8286 		}
8287 
8288 		lun->flags &= ~CTL_LUN_PR_RESERVED;
8289 		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8290 		lun->res_type = 0;
8291 		break;
8292 
8293 	case CTL_PR_PREEMPT:
8294 		ctl_pro_preempt_other(lun, msg);
8295 		break;
8296 	case CTL_PR_CLEAR:
8297 		lun->flags &= ~CTL_LUN_PR_RESERVED;
8298 		lun->res_type = 0;
8299 		lun->pr_key_count = 0;
8300 		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8301 
8302 		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8303 			if (lun->per_res[i].registered == 0)
8304 				continue;
8305 			if (!persis_offset
8306 			 && i < CTL_MAX_INITIATORS)
8307 				lun->pending_sense[i].ua_pending |=
8308 					CTL_UA_RES_PREEMPT;
8309 			else if (persis_offset
8310 			      && i >= persis_offset)
8311    				lun->pending_sense[i-persis_offset].ua_pending|=
8312 					CTL_UA_RES_PREEMPT;
8313 			memset(&lun->per_res[i].res_key, 0,
8314 			       sizeof(struct scsi_per_res_key));
8315 			lun->per_res[i].registered = 0;
8316 		}
8317 		lun->PRGeneration++;
8318 		break;
8319 	}
8320 
8321 	mtx_unlock(&softc->ctl_lock);
8322 }
8323 
8324 int
8325 ctl_read_write(struct ctl_scsiio *ctsio)
8326 {
8327 	struct ctl_lun *lun;
8328 	struct ctl_lba_len lbalen;
8329 	uint64_t lba;
8330 	uint32_t num_blocks;
8331 	int reladdr, fua, dpo, ebp;
8332 	int retval;
8333 	int isread;
8334 
8335 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8336 
8337 	CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8338 
8339 	reladdr = 0;
8340 	fua = 0;
8341 	dpo = 0;
8342 	ebp = 0;
8343 
8344 	retval = CTL_RETVAL_COMPLETE;
8345 
8346 	isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8347 	      || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8348 	if (lun->flags & CTL_LUN_PR_RESERVED && isread) {
8349 		uint32_t residx;
8350 
8351 		/*
8352 		 * XXX KDM need a lock here.
8353 		 */
8354 		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8355 		if ((lun->res_type == SPR_TYPE_EX_AC
8356 		  && residx != lun->pr_res_idx)
8357 		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
8358 		   || lun->res_type == SPR_TYPE_EX_AC_AR)
8359 		  && !lun->per_res[residx].registered)) {
8360 			ctl_set_reservation_conflict(ctsio);
8361 			ctl_done((union ctl_io *)ctsio);
8362 			return (CTL_RETVAL_COMPLETE);
8363 	        }
8364 	}
8365 
8366 	switch (ctsio->cdb[0]) {
8367 	case READ_6:
8368 	case WRITE_6: {
8369 		struct scsi_rw_6 *cdb;
8370 
8371 		cdb = (struct scsi_rw_6 *)ctsio->cdb;
8372 
8373 		lba = scsi_3btoul(cdb->addr);
8374 		/* only 5 bits are valid in the most significant address byte */
8375 		lba &= 0x1fffff;
8376 		num_blocks = cdb->length;
8377 		/*
8378 		 * This is correct according to SBC-2.
8379 		 */
8380 		if (num_blocks == 0)
8381 			num_blocks = 256;
8382 		break;
8383 	}
8384 	case READ_10:
8385 	case WRITE_10: {
8386 		struct scsi_rw_10 *cdb;
8387 
8388 		cdb = (struct scsi_rw_10 *)ctsio->cdb;
8389 
8390 		if (cdb->byte2 & SRW10_RELADDR)
8391 			reladdr = 1;
8392 		if (cdb->byte2 & SRW10_FUA)
8393 			fua = 1;
8394 		if (cdb->byte2 & SRW10_DPO)
8395 			dpo = 1;
8396 
8397 		if ((cdb->opcode == WRITE_10)
8398 		 && (cdb->byte2 & SRW10_EBP))
8399 			ebp = 1;
8400 
8401 		lba = scsi_4btoul(cdb->addr);
8402 		num_blocks = scsi_2btoul(cdb->length);
8403 		break;
8404 	}
8405 	case WRITE_VERIFY_10: {
8406 		struct scsi_write_verify_10 *cdb;
8407 
8408 		cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8409 
8410 		/*
8411 		 * XXX KDM we should do actual write verify support at some
8412 		 * point.  This is obviously fake, we're just translating
8413 		 * things to a write.  So we don't even bother checking the
8414 		 * BYTCHK field, since we don't do any verification.  If
8415 		 * the user asks for it, we'll just pretend we did it.
8416 		 */
8417 		if (cdb->byte2 & SWV_DPO)
8418 			dpo = 1;
8419 
8420 		lba = scsi_4btoul(cdb->addr);
8421 		num_blocks = scsi_2btoul(cdb->length);
8422 		break;
8423 	}
8424 	case READ_12:
8425 	case WRITE_12: {
8426 		struct scsi_rw_12 *cdb;
8427 
8428 		cdb = (struct scsi_rw_12 *)ctsio->cdb;
8429 
8430 		if (cdb->byte2 & SRW12_RELADDR)
8431 			reladdr = 1;
8432 		if (cdb->byte2 & SRW12_FUA)
8433 			fua = 1;
8434 		if (cdb->byte2 & SRW12_DPO)
8435 			dpo = 1;
8436 		lba = scsi_4btoul(cdb->addr);
8437 		num_blocks = scsi_4btoul(cdb->length);
8438 		break;
8439 	}
8440 	case WRITE_VERIFY_12: {
8441 		struct scsi_write_verify_12 *cdb;
8442 
8443 		cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8444 
8445 		if (cdb->byte2 & SWV_DPO)
8446 			dpo = 1;
8447 
8448 		lba = scsi_4btoul(cdb->addr);
8449 		num_blocks = scsi_4btoul(cdb->length);
8450 
8451 		break;
8452 	}
8453 	case READ_16:
8454 	case WRITE_16: {
8455 		struct scsi_rw_16 *cdb;
8456 
8457 		cdb = (struct scsi_rw_16 *)ctsio->cdb;
8458 
8459 		if (cdb->byte2 & SRW12_RELADDR)
8460 			reladdr = 1;
8461 		if (cdb->byte2 & SRW12_FUA)
8462 			fua = 1;
8463 		if (cdb->byte2 & SRW12_DPO)
8464 			dpo = 1;
8465 
8466 		lba = scsi_8btou64(cdb->addr);
8467 		num_blocks = scsi_4btoul(cdb->length);
8468 		break;
8469 	}
8470 	case WRITE_VERIFY_16: {
8471 		struct scsi_write_verify_16 *cdb;
8472 
8473 		cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
8474 
8475 		if (cdb->byte2 & SWV_DPO)
8476 			dpo = 1;
8477 
8478 		lba = scsi_8btou64(cdb->addr);
8479 		num_blocks = scsi_4btoul(cdb->length);
8480 		break;
8481 	}
8482 	default:
8483 		/*
8484 		 * We got a command we don't support.  This shouldn't
8485 		 * happen, commands should be filtered out above us.
8486 		 */
8487 		ctl_set_invalid_opcode(ctsio);
8488 		ctl_done((union ctl_io *)ctsio);
8489 
8490 		return (CTL_RETVAL_COMPLETE);
8491 		break; /* NOTREACHED */
8492 	}
8493 
8494 	/*
8495 	 * XXX KDM what do we do with the DPO and FUA bits?  FUA might be
8496 	 * interesting for us, but if RAIDCore is in write-back mode,
8497 	 * getting it to do write-through for a particular transaction may
8498 	 * not be possible.
8499 	 */
8500 	/*
8501 	 * We don't support relative addressing.  That also requires
8502 	 * supporting linked commands, which we don't do.
8503 	 */
8504 	if (reladdr != 0) {
8505 		ctl_set_invalid_field(ctsio,
8506 				      /*sks_valid*/ 1,
8507 				      /*command*/ 1,
8508 				      /*field*/ 1,
8509 				      /*bit_valid*/ 1,
8510 				      /*bit*/ 0);
8511 		ctl_done((union ctl_io *)ctsio);
8512 		return (CTL_RETVAL_COMPLETE);
8513 	}
8514 
8515 	/*
8516 	 * The first check is to make sure we're in bounds, the second
8517 	 * check is to catch wrap-around problems.  If the lba + num blocks
8518 	 * is less than the lba, then we've wrapped around and the block
8519 	 * range is invalid anyway.
8520 	 */
8521 	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8522 	 || ((lba + num_blocks) < lba)) {
8523 		ctl_set_lba_out_of_range(ctsio);
8524 		ctl_done((union ctl_io *)ctsio);
8525 		return (CTL_RETVAL_COMPLETE);
8526 	}
8527 
8528 	/*
8529 	 * According to SBC-3, a transfer length of 0 is not an error.
8530 	 * Note that this cannot happen with WRITE(6) or READ(6), since 0
8531 	 * translates to 256 blocks for those commands.
8532 	 */
8533 	if (num_blocks == 0) {
8534 		ctl_set_success(ctsio);
8535 		ctl_done((union ctl_io *)ctsio);
8536 		return (CTL_RETVAL_COMPLETE);
8537 	}
8538 
8539 	lbalen.lba = lba;
8540 	lbalen.len = num_blocks;
8541 	memcpy(ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes, &lbalen,
8542 	       sizeof(lbalen));
8543 
8544 	CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
8545 
8546 	retval = lun->backend->data_submit((union ctl_io *)ctsio);
8547 
8548 	return (retval);
8549 }
8550 
8551 int
8552 ctl_report_luns(struct ctl_scsiio *ctsio)
8553 {
8554 	struct scsi_report_luns *cdb;
8555 	struct scsi_report_luns_data *lun_data;
8556 	struct ctl_lun *lun, *request_lun;
8557 	int num_luns, retval;
8558 	uint32_t alloc_len, lun_datalen;
8559 	int num_filled, well_known;
8560 	uint32_t initidx;
8561 
8562 	retval = CTL_RETVAL_COMPLETE;
8563 	well_known = 0;
8564 
8565 	cdb = (struct scsi_report_luns *)ctsio->cdb;
8566 
8567 	CTL_DEBUG_PRINT(("ctl_report_luns\n"));
8568 
8569 	mtx_lock(&control_softc->ctl_lock);
8570 	num_luns = control_softc->num_luns;
8571 	mtx_unlock(&control_softc->ctl_lock);
8572 
8573 	switch (cdb->select_report) {
8574 	case RPL_REPORT_DEFAULT:
8575 	case RPL_REPORT_ALL:
8576 		break;
8577 	case RPL_REPORT_WELLKNOWN:
8578 		well_known = 1;
8579 		num_luns = 0;
8580 		break;
8581 	default:
8582 		ctl_set_invalid_field(ctsio,
8583 				      /*sks_valid*/ 1,
8584 				      /*command*/ 1,
8585 				      /*field*/ 2,
8586 				      /*bit_valid*/ 0,
8587 				      /*bit*/ 0);
8588 		ctl_done((union ctl_io *)ctsio);
8589 		return (retval);
8590 		break; /* NOTREACHED */
8591 	}
8592 
8593 	alloc_len = scsi_4btoul(cdb->length);
8594 	/*
8595 	 * The initiator has to allocate at least 16 bytes for this request,
8596 	 * so he can at least get the header and the first LUN.  Otherwise
8597 	 * we reject the request (per SPC-3 rev 14, section 6.21).
8598 	 */
8599 	if (alloc_len < (sizeof(struct scsi_report_luns_data) +
8600 	    sizeof(struct scsi_report_luns_lundata))) {
8601 		ctl_set_invalid_field(ctsio,
8602 				      /*sks_valid*/ 1,
8603 				      /*command*/ 1,
8604 				      /*field*/ 6,
8605 				      /*bit_valid*/ 0,
8606 				      /*bit*/ 0);
8607 		ctl_done((union ctl_io *)ctsio);
8608 		return (retval);
8609 	}
8610 
8611 	request_lun = (struct ctl_lun *)
8612 		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8613 
8614 	lun_datalen = sizeof(*lun_data) +
8615 		(num_luns * sizeof(struct scsi_report_luns_lundata));
8616 
8617 	ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
8618 	lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
8619 	ctsio->kern_sg_entries = 0;
8620 
8621 	if (lun_datalen < alloc_len) {
8622 		ctsio->residual = alloc_len - lun_datalen;
8623 		ctsio->kern_data_len = lun_datalen;
8624 		ctsio->kern_total_len = lun_datalen;
8625 	} else {
8626 		ctsio->residual = 0;
8627 		ctsio->kern_data_len = alloc_len;
8628 		ctsio->kern_total_len = alloc_len;
8629 	}
8630 	ctsio->kern_data_resid = 0;
8631 	ctsio->kern_rel_offset = 0;
8632 	ctsio->kern_sg_entries = 0;
8633 
8634 	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
8635 
8636 	/*
8637 	 * We set this to the actual data length, regardless of how much
8638 	 * space we actually have to return results.  If the user looks at
8639 	 * this value, he'll know whether or not he allocated enough space
8640 	 * and reissue the command if necessary.  We don't support well
8641 	 * known logical units, so if the user asks for that, return none.
8642 	 */
8643 	scsi_ulto4b(lun_datalen - 8, lun_data->length);
8644 
8645 	mtx_lock(&control_softc->ctl_lock);
8646 	for (num_filled = 0, lun = STAILQ_FIRST(&control_softc->lun_list);
8647 	     (lun != NULL) && (num_filled < num_luns);
8648 	     lun = STAILQ_NEXT(lun, links)) {
8649 
8650 		if (lun->lun <= 0xff) {
8651 			/*
8652 			 * Peripheral addressing method, bus number 0.
8653 			 */
8654 			lun_data->luns[num_filled].lundata[0] =
8655 				RPL_LUNDATA_ATYP_PERIPH;
8656 			lun_data->luns[num_filled].lundata[1] = lun->lun;
8657 			num_filled++;
8658 		} else if (lun->lun <= 0x3fff) {
8659 			/*
8660 			 * Flat addressing method.
8661 			 */
8662 			lun_data->luns[num_filled].lundata[0] =
8663 				RPL_LUNDATA_ATYP_FLAT |
8664 				(lun->lun & RPL_LUNDATA_FLAT_LUN_MASK);
8665 #ifdef OLDCTLHEADERS
8666 				(SRLD_ADDR_FLAT << SRLD_ADDR_SHIFT) |
8667 				(lun->lun & SRLD_BUS_LUN_MASK);
8668 #endif
8669 			lun_data->luns[num_filled].lundata[1] =
8670 #ifdef OLDCTLHEADERS
8671 				lun->lun >> SRLD_BUS_LUN_BITS;
8672 #endif
8673 				lun->lun >> RPL_LUNDATA_FLAT_LUN_BITS;
8674 			num_filled++;
8675 		} else {
8676 			printf("ctl_report_luns: bogus LUN number %jd, "
8677 			       "skipping\n", (intmax_t)lun->lun);
8678 		}
8679 		/*
8680 		 * According to SPC-3, rev 14 section 6.21:
8681 		 *
8682 		 * "The execution of a REPORT LUNS command to any valid and
8683 		 * installed logical unit shall clear the REPORTED LUNS DATA
8684 		 * HAS CHANGED unit attention condition for all logical
8685 		 * units of that target with respect to the requesting
8686 		 * initiator. A valid and installed logical unit is one
8687 		 * having a PERIPHERAL QUALIFIER of 000b in the standard
8688 		 * INQUIRY data (see 6.4.2)."
8689 		 *
8690 		 * If request_lun is NULL, the LUN this report luns command
8691 		 * was issued to is either disabled or doesn't exist. In that
8692 		 * case, we shouldn't clear any pending lun change unit
8693 		 * attention.
8694 		 */
8695 		if (request_lun != NULL)
8696 			lun->pending_sense[initidx].ua_pending &=
8697 				~CTL_UA_LUN_CHANGE;
8698 	}
8699 	mtx_unlock(&control_softc->ctl_lock);
8700 
8701 	/*
8702 	 * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
8703 	 * this request.
8704 	 */
8705 	ctsio->scsi_status = SCSI_STATUS_OK;
8706 
8707 	ctsio->be_move_done = ctl_config_move_done;
8708 	ctl_datamove((union ctl_io *)ctsio);
8709 
8710 	return (retval);
8711 }
8712 
8713 int
8714 ctl_request_sense(struct ctl_scsiio *ctsio)
8715 {
8716 	struct scsi_request_sense *cdb;
8717 	struct scsi_sense_data *sense_ptr;
8718 	struct ctl_lun *lun;
8719 	uint32_t initidx;
8720 	int have_error;
8721 	scsi_sense_data_type sense_format;
8722 
8723 	cdb = (struct scsi_request_sense *)ctsio->cdb;
8724 
8725 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8726 
8727 	CTL_DEBUG_PRINT(("ctl_request_sense\n"));
8728 
8729 	/*
8730 	 * Determine which sense format the user wants.
8731 	 */
8732 	if (cdb->byte2 & SRS_DESC)
8733 		sense_format = SSD_TYPE_DESC;
8734 	else
8735 		sense_format = SSD_TYPE_FIXED;
8736 
8737 	ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
8738 	sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
8739 	ctsio->kern_sg_entries = 0;
8740 
8741 	/*
8742 	 * struct scsi_sense_data, which is currently set to 256 bytes, is
8743 	 * larger than the largest allowed value for the length field in the
8744 	 * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
8745 	 */
8746 	ctsio->residual = 0;
8747 	ctsio->kern_data_len = cdb->length;
8748 	ctsio->kern_total_len = cdb->length;
8749 
8750 	ctsio->kern_data_resid = 0;
8751 	ctsio->kern_rel_offset = 0;
8752 	ctsio->kern_sg_entries = 0;
8753 
8754 	/*
8755 	 * If we don't have a LUN, we don't have any pending sense.
8756 	 */
8757 	if (lun == NULL)
8758 		goto no_sense;
8759 
8760 	have_error = 0;
8761 	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
8762 	/*
8763 	 * Check for pending sense, and then for pending unit attentions.
8764 	 * Pending sense gets returned first, then pending unit attentions.
8765 	 */
8766 	mtx_lock(&lun->ctl_softc->ctl_lock);
8767 	if (ctl_is_set(lun->have_ca, initidx)) {
8768 		scsi_sense_data_type stored_format;
8769 
8770 		/*
8771 		 * Check to see which sense format was used for the stored
8772 		 * sense data.
8773 		 */
8774 		stored_format = scsi_sense_type(
8775 		    &lun->pending_sense[initidx].sense);
8776 
8777 		/*
8778 		 * If the user requested a different sense format than the
8779 		 * one we stored, then we need to convert it to the other
8780 		 * format.  If we're going from descriptor to fixed format
8781 		 * sense data, we may lose things in translation, depending
8782 		 * on what options were used.
8783 		 *
8784 		 * If the stored format is SSD_TYPE_NONE (i.e. invalid),
8785 		 * for some reason we'll just copy it out as-is.
8786 		 */
8787 		if ((stored_format == SSD_TYPE_FIXED)
8788 		 && (sense_format == SSD_TYPE_DESC))
8789 			ctl_sense_to_desc((struct scsi_sense_data_fixed *)
8790 			    &lun->pending_sense[initidx].sense,
8791 			    (struct scsi_sense_data_desc *)sense_ptr);
8792 		else if ((stored_format == SSD_TYPE_DESC)
8793 		      && (sense_format == SSD_TYPE_FIXED))
8794 			ctl_sense_to_fixed((struct scsi_sense_data_desc *)
8795 			    &lun->pending_sense[initidx].sense,
8796 			    (struct scsi_sense_data_fixed *)sense_ptr);
8797 		else
8798 			memcpy(sense_ptr, &lun->pending_sense[initidx].sense,
8799 			       ctl_min(sizeof(*sense_ptr),
8800 			       sizeof(lun->pending_sense[initidx].sense)));
8801 
8802 		ctl_clear_mask(lun->have_ca, initidx);
8803 		have_error = 1;
8804 	} else if (lun->pending_sense[initidx].ua_pending != CTL_UA_NONE) {
8805 		ctl_ua_type ua_type;
8806 
8807 		ua_type = ctl_build_ua(lun->pending_sense[initidx].ua_pending,
8808 				       sense_ptr, sense_format);
8809 		if (ua_type != CTL_UA_NONE) {
8810 			have_error = 1;
8811 			/* We're reporting this UA, so clear it */
8812 			lun->pending_sense[initidx].ua_pending &= ~ua_type;
8813 		}
8814 	}
8815 	mtx_unlock(&lun->ctl_softc->ctl_lock);
8816 
8817 	/*
8818 	 * We already have a pending error, return it.
8819 	 */
8820 	if (have_error != 0) {
8821 		/*
8822 		 * We report the SCSI status as OK, since the status of the
8823 		 * request sense command itself is OK.
8824 		 */
8825 		ctsio->scsi_status = SCSI_STATUS_OK;
8826 
8827 		/*
8828 		 * We report 0 for the sense length, because we aren't doing
8829 		 * autosense in this case.  We're reporting sense as
8830 		 * parameter data.
8831 		 */
8832 		ctsio->sense_len = 0;
8833 
8834 		ctsio->be_move_done = ctl_config_move_done;
8835 		ctl_datamove((union ctl_io *)ctsio);
8836 
8837 		return (CTL_RETVAL_COMPLETE);
8838 	}
8839 
8840 no_sense:
8841 
8842 	/*
8843 	 * No sense information to report, so we report that everything is
8844 	 * okay.
8845 	 */
8846 	ctl_set_sense_data(sense_ptr,
8847 			   lun,
8848 			   sense_format,
8849 			   /*current_error*/ 1,
8850 			   /*sense_key*/ SSD_KEY_NO_SENSE,
8851 			   /*asc*/ 0x00,
8852 			   /*ascq*/ 0x00,
8853 			   SSD_ELEM_NONE);
8854 
8855 	ctsio->scsi_status = SCSI_STATUS_OK;
8856 
8857 	/*
8858 	 * We report 0 for the sense length, because we aren't doing
8859 	 * autosense in this case.  We're reporting sense as parameter data.
8860 	 */
8861 	ctsio->sense_len = 0;
8862 	ctsio->be_move_done = ctl_config_move_done;
8863 	ctl_datamove((union ctl_io *)ctsio);
8864 
8865 	return (CTL_RETVAL_COMPLETE);
8866 }
8867 
8868 int
8869 ctl_tur(struct ctl_scsiio *ctsio)
8870 {
8871 	struct ctl_lun *lun;
8872 
8873 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8874 
8875 	CTL_DEBUG_PRINT(("ctl_tur\n"));
8876 
8877 	if (lun == NULL)
8878 		return (-EINVAL);
8879 
8880 	ctsio->scsi_status = SCSI_STATUS_OK;
8881 	ctsio->io_hdr.status = CTL_SUCCESS;
8882 
8883 	ctl_done((union ctl_io *)ctsio);
8884 
8885 	return (CTL_RETVAL_COMPLETE);
8886 }
8887 
8888 #ifdef notyet
8889 static int
8890 ctl_cmddt_inquiry(struct ctl_scsiio *ctsio)
8891 {
8892 
8893 }
8894 #endif
8895 
8896 static int
8897 ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
8898 {
8899 	struct scsi_vpd_supported_pages *pages;
8900 	int sup_page_size;
8901 	struct ctl_lun *lun;
8902 
8903 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8904 
8905 	sup_page_size = sizeof(struct scsi_vpd_supported_pages) +
8906 		SCSI_EVPD_NUM_SUPPORTED_PAGES;
8907 	/*
8908 	 * XXX KDM GFP_???  We probably don't want to wait here,
8909 	 * unless we end up having a process/thread context.
8910 	 */
8911 	ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
8912 	if (ctsio->kern_data_ptr == NULL) {
8913 		ctsio->io_hdr.status = CTL_SCSI_ERROR;
8914 		ctsio->scsi_status = SCSI_STATUS_BUSY;
8915 		ctl_done((union ctl_io *)ctsio);
8916 		return (CTL_RETVAL_COMPLETE);
8917 	}
8918 	pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
8919 	ctsio->kern_sg_entries = 0;
8920 
8921 	if (sup_page_size < alloc_len) {
8922 		ctsio->residual = alloc_len - sup_page_size;
8923 		ctsio->kern_data_len = sup_page_size;
8924 		ctsio->kern_total_len = sup_page_size;
8925 	} else {
8926 		ctsio->residual = 0;
8927 		ctsio->kern_data_len = alloc_len;
8928 		ctsio->kern_total_len = alloc_len;
8929 	}
8930 	ctsio->kern_data_resid = 0;
8931 	ctsio->kern_rel_offset = 0;
8932 	ctsio->kern_sg_entries = 0;
8933 
8934 	/*
8935 	 * The control device is always connected.  The disk device, on the
8936 	 * other hand, may not be online all the time.  Need to change this
8937 	 * to figure out whether the disk device is actually online or not.
8938 	 */
8939 	if (lun != NULL)
8940 		pages->device = (SID_QUAL_LU_CONNECTED << 5) |
8941 				lun->be_lun->lun_type;
8942 	else
8943 		pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
8944 
8945 	pages->length = SCSI_EVPD_NUM_SUPPORTED_PAGES;
8946 	/* Supported VPD pages */
8947 	pages->page_list[0] = SVPD_SUPPORTED_PAGES;
8948 	/* Serial Number */
8949 	pages->page_list[1] = SVPD_UNIT_SERIAL_NUMBER;
8950 	/* Device Identification */
8951 	pages->page_list[2] = SVPD_DEVICE_ID;
8952 
8953 	ctsio->scsi_status = SCSI_STATUS_OK;
8954 
8955 	ctsio->be_move_done = ctl_config_move_done;
8956 	ctl_datamove((union ctl_io *)ctsio);
8957 
8958 	return (CTL_RETVAL_COMPLETE);
8959 }
8960 
8961 static int
8962 ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
8963 {
8964 	struct scsi_vpd_unit_serial_number *sn_ptr;
8965 	struct ctl_lun *lun;
8966 #ifndef CTL_USE_BACKEND_SN
8967 	char tmpstr[32];
8968 #endif
8969 
8970 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8971 
8972 	/* XXX KDM which malloc flags here?? */
8973 	ctsio->kern_data_ptr = malloc(sizeof(*sn_ptr), M_CTL, M_WAITOK | M_ZERO);
8974 	if (ctsio->kern_data_ptr == NULL) {
8975 		ctsio->io_hdr.status = CTL_SCSI_ERROR;
8976 		ctsio->scsi_status = SCSI_STATUS_BUSY;
8977 		ctl_done((union ctl_io *)ctsio);
8978 		return (CTL_RETVAL_COMPLETE);
8979 	}
8980 	sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
8981 	ctsio->kern_sg_entries = 0;
8982 
8983 	if (sizeof(*sn_ptr) < alloc_len) {
8984 		ctsio->residual = alloc_len - sizeof(*sn_ptr);
8985 		ctsio->kern_data_len = sizeof(*sn_ptr);
8986 		ctsio->kern_total_len = sizeof(*sn_ptr);
8987 	} else {
8988 		ctsio->residual = 0;
8989 		ctsio->kern_data_len = alloc_len;
8990 		ctsio->kern_total_len = alloc_len;
8991 	}
8992 	ctsio->kern_data_resid = 0;
8993 	ctsio->kern_rel_offset = 0;
8994 	ctsio->kern_sg_entries = 0;
8995 
8996 	/*
8997 	 * The control device is always connected.  The disk device, on the
8998 	 * other hand, may not be online all the time.  Need to change this
8999 	 * to figure out whether the disk device is actually online or not.
9000 	 */
9001 	if (lun != NULL)
9002 		sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9003 				  lun->be_lun->lun_type;
9004 	else
9005 		sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9006 
9007 	sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9008 	sn_ptr->length = ctl_min(sizeof(*sn_ptr) - 4, CTL_SN_LEN);
9009 #ifdef CTL_USE_BACKEND_SN
9010 	/*
9011 	 * If we don't have a LUN, we just leave the serial number as
9012 	 * all spaces.
9013 	 */
9014 	memset(sn_ptr->serial_num, 0x20, sizeof(sn_ptr->serial_num));
9015 	if (lun != NULL) {
9016 		strncpy((char *)sn_ptr->serial_num,
9017 			(char *)lun->be_lun->serial_num, CTL_SN_LEN);
9018 	}
9019 #else
9020 	/*
9021 	 * Note that we're using a non-unique serial number here,
9022 	 */
9023 	snprintf(tmpstr, sizeof(tmpstr), "MYSERIALNUMIS000");
9024 	memset(sn_ptr->serial_num, 0x20, sizeof(sn_ptr->serial_num));
9025 	strncpy(sn_ptr->serial_num, tmpstr, ctl_min(CTL_SN_LEN,
9026 		ctl_min(sizeof(tmpstr), sizeof(*sn_ptr) - 4)));
9027 #endif
9028 	ctsio->scsi_status = SCSI_STATUS_OK;
9029 
9030 	ctsio->be_move_done = ctl_config_move_done;
9031 	ctl_datamove((union ctl_io *)ctsio);
9032 
9033 	return (CTL_RETVAL_COMPLETE);
9034 }
9035 
9036 
9037 static int
9038 ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9039 {
9040 	struct scsi_vpd_device_id *devid_ptr;
9041 	struct scsi_vpd_id_descriptor *desc, *desc1;
9042 	struct scsi_vpd_id_descriptor *desc2, *desc3; /* for types 4h and 5h */
9043 	struct scsi_vpd_id_t10 *t10id;
9044 	struct ctl_softc *ctl_softc;
9045 	struct ctl_lun *lun;
9046 	struct ctl_frontend *fe;
9047 #ifndef CTL_USE_BACKEND_SN
9048 	char tmpstr[32];
9049 #endif /* CTL_USE_BACKEND_SN */
9050 	int devid_len;
9051 
9052 	ctl_softc = control_softc;
9053 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9054 
9055 	devid_len = sizeof(struct scsi_vpd_device_id) +
9056 		sizeof(struct scsi_vpd_id_descriptor) +
9057 		sizeof(struct scsi_vpd_id_t10) + CTL_DEVID_LEN +
9058 		sizeof(struct scsi_vpd_id_descriptor) + CTL_WWPN_LEN +
9059 		sizeof(struct scsi_vpd_id_descriptor) +
9060 		sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9061 		sizeof(struct scsi_vpd_id_descriptor) +
9062 		sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9063 
9064 	/* XXX KDM which malloc flags here ?? */
9065 	ctsio->kern_data_ptr = malloc(devid_len, M_CTL, M_WAITOK | M_ZERO);
9066 	if (ctsio->kern_data_ptr == NULL) {
9067 		ctsio->io_hdr.status = CTL_SCSI_ERROR;
9068 		ctsio->scsi_status = SCSI_STATUS_BUSY;
9069 		ctl_done((union ctl_io *)ctsio);
9070 		return (CTL_RETVAL_COMPLETE);
9071 	}
9072 	devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9073 	ctsio->kern_sg_entries = 0;
9074 
9075 	if (devid_len < alloc_len) {
9076 		ctsio->residual = alloc_len - devid_len;
9077 		ctsio->kern_data_len = devid_len;
9078 		ctsio->kern_total_len = devid_len;
9079 	} else {
9080 		ctsio->residual = 0;
9081 		ctsio->kern_data_len = alloc_len;
9082 		ctsio->kern_total_len = alloc_len;
9083 	}
9084 	ctsio->kern_data_resid = 0;
9085 	ctsio->kern_rel_offset = 0;
9086 	ctsio->kern_sg_entries = 0;
9087 
9088 	desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9089 	t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
9090 	desc1 = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9091 		sizeof(struct scsi_vpd_id_t10) + CTL_DEVID_LEN);
9092 	desc2 = (struct scsi_vpd_id_descriptor *)(&desc1->identifier[0] +
9093 	          CTL_WWPN_LEN);
9094 	desc3 = (struct scsi_vpd_id_descriptor *)(&desc2->identifier[0] +
9095 	         sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9096 
9097 	/*
9098 	 * The control device is always connected.  The disk device, on the
9099 	 * other hand, may not be online all the time.
9100 	 */
9101 	if (lun != NULL)
9102 		devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9103 				     lun->be_lun->lun_type;
9104 	else
9105 		devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9106 
9107 	devid_ptr->page_code = SVPD_DEVICE_ID;
9108 
9109 	scsi_ulto2b(devid_len - 4, devid_ptr->length);
9110 
9111 	mtx_lock(&ctl_softc->ctl_lock);
9112 
9113 	fe = ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)];
9114 
9115 	/*
9116 	 * For Fibre channel,
9117 	 */
9118 	if (fe->port_type == CTL_PORT_FC)
9119 	{
9120 		desc->proto_codeset = (SCSI_PROTO_FC << 4) |
9121 				      SVPD_ID_CODESET_ASCII;
9122         	desc1->proto_codeset = (SCSI_PROTO_FC << 4) |
9123 		              SVPD_ID_CODESET_BINARY;
9124 	}
9125 	else
9126 	{
9127 		desc->proto_codeset = (SCSI_PROTO_SPI << 4) |
9128 				      SVPD_ID_CODESET_ASCII;
9129         	desc1->proto_codeset = (SCSI_PROTO_SPI << 4) |
9130 		              SVPD_ID_CODESET_BINARY;
9131 	}
9132 	desc2->proto_codeset = desc3->proto_codeset = desc1->proto_codeset;
9133 	mtx_unlock(&ctl_softc->ctl_lock);
9134 
9135 	/*
9136 	 * We're using a LUN association here.  i.e., this device ID is a
9137 	 * per-LUN identifier.
9138 	 */
9139 	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
9140 	desc->length = sizeof(*t10id) + CTL_DEVID_LEN;
9141 	strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
9142 
9143 	/*
9144 	 * desc1 is for the WWPN which is a port asscociation.
9145 	 */
9146 	desc1->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT | SVPD_ID_TYPE_NAA;
9147 	desc1->length = CTL_WWPN_LEN;
9148 	/* XXX Call Reggie's get_WWNN func here then add port # to the end */
9149 	/* For testing just create the WWPN */
9150 #if 0
9151 	ddb_GetWWNN((char *)desc1->identifier);
9152 
9153 	/* NOTE: if the port is 0 or 8 we don't want to subtract 1 */
9154 	/* This is so Copancontrol will return something sane */
9155 	if (ctsio->io_hdr.nexus.targ_port!=0 &&
9156 	    ctsio->io_hdr.nexus.targ_port!=8)
9157 		desc1->identifier[7] += ctsio->io_hdr.nexus.targ_port-1;
9158 	else
9159 		desc1->identifier[7] += ctsio->io_hdr.nexus.targ_port;
9160 #endif
9161 
9162 	be64enc(desc1->identifier, fe->wwpn);
9163 
9164 	/*
9165 	 * desc2 is for the Relative Target Port(type 4h) identifier
9166 	 */
9167 	desc2->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT
9168 	                 | SVPD_ID_TYPE_RELTARG;
9169 	desc2->length = 4;
9170 //#if 0
9171 	/* NOTE: if the port is 0 or 8 we don't want to subtract 1 */
9172 	/* This is so Copancontrol will return something sane */
9173 	if (ctsio->io_hdr.nexus.targ_port!=0 &&
9174 	    ctsio->io_hdr.nexus.targ_port!=8)
9175 		desc2->identifier[3] = ctsio->io_hdr.nexus.targ_port - 1;
9176 	else
9177 	        desc2->identifier[3] = ctsio->io_hdr.nexus.targ_port;
9178 //#endif
9179 
9180 	/*
9181 	 * desc3 is for the Target Port Group(type 5h) identifier
9182 	 */
9183 	desc3->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT
9184 	                 | SVPD_ID_TYPE_TPORTGRP;
9185 	desc3->length = 4;
9186 	if (ctsio->io_hdr.nexus.targ_port < CTL_MAX_PORTS || ctl_is_single)
9187 		desc3->identifier[3] = 1;
9188 	else
9189 		desc3->identifier[3] = 2;
9190 
9191 #ifdef CTL_USE_BACKEND_SN
9192 	/*
9193 	 * If we've actually got a backend, copy the device id from the
9194 	 * per-LUN data.  Otherwise, set it to all spaces.
9195 	 */
9196 	if (lun != NULL) {
9197 		/*
9198 		 * Copy the backend's LUN ID.
9199 		 */
9200 		strncpy((char *)t10id->vendor_spec_id,
9201 			(char *)lun->be_lun->device_id, CTL_DEVID_LEN);
9202 	} else {
9203 		/*
9204 		 * No backend, set this to spaces.
9205 		 */
9206 		memset(t10id->vendor_spec_id, 0x20, CTL_DEVID_LEN);
9207 	}
9208 #else
9209 	snprintf(tmpstr, sizeof(tmpstr), "MYDEVICEIDIS%4d",
9210 		 (lun != NULL) ?  (int)lun->lun : 0);
9211 	strncpy(t10id->vendor_spec_id, tmpstr, ctl_min(CTL_DEVID_LEN,
9212 		sizeof(tmpstr)));
9213 #endif
9214 
9215 	ctsio->scsi_status = SCSI_STATUS_OK;
9216 
9217 	ctsio->be_move_done = ctl_config_move_done;
9218 	ctl_datamove((union ctl_io *)ctsio);
9219 
9220 	return (CTL_RETVAL_COMPLETE);
9221 }
9222 
9223 static int
9224 ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
9225 {
9226 	struct scsi_inquiry *cdb;
9227 	int alloc_len, retval;
9228 
9229 	cdb = (struct scsi_inquiry *)ctsio->cdb;
9230 
9231 	retval = CTL_RETVAL_COMPLETE;
9232 
9233 	alloc_len = scsi_2btoul(cdb->length);
9234 
9235 	switch (cdb->page_code) {
9236 	case SVPD_SUPPORTED_PAGES:
9237 		retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
9238 		break;
9239 	case SVPD_UNIT_SERIAL_NUMBER:
9240 		retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
9241 		break;
9242 	case SVPD_DEVICE_ID:
9243 		retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
9244 		break;
9245 	default:
9246 		ctl_set_invalid_field(ctsio,
9247 				      /*sks_valid*/ 1,
9248 				      /*command*/ 1,
9249 				      /*field*/ 2,
9250 				      /*bit_valid*/ 0,
9251 				      /*bit*/ 0);
9252 		ctl_done((union ctl_io *)ctsio);
9253 		retval = CTL_RETVAL_COMPLETE;
9254 		break;
9255 	}
9256 
9257 	return (retval);
9258 }
9259 
9260 static int
9261 ctl_inquiry_std(struct ctl_scsiio *ctsio)
9262 {
9263 	struct scsi_inquiry_data *inq_ptr;
9264 	struct scsi_inquiry *cdb;
9265 	struct ctl_softc *ctl_softc;
9266 	struct ctl_lun *lun;
9267 	uint32_t alloc_len;
9268 	int is_fc;
9269 
9270 	ctl_softc = control_softc;
9271 
9272 	/*
9273 	 * Figure out whether we're talking to a Fibre Channel port or not.
9274 	 * We treat the ioctl front end, and any SCSI adapters, as packetized
9275 	 * SCSI front ends.
9276 	 */
9277 	mtx_lock(&ctl_softc->ctl_lock);
9278 	if (ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)]->port_type !=
9279 	    CTL_PORT_FC)
9280 		is_fc = 0;
9281 	else
9282 		is_fc = 1;
9283 	mtx_unlock(&ctl_softc->ctl_lock);
9284 
9285 	lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9286 	cdb = (struct scsi_inquiry *)ctsio->cdb;
9287 	alloc_len = scsi_2btoul(cdb->length);
9288 
9289 	/*
9290 	 * We malloc the full inquiry data size here and fill it
9291 	 * in.  If the user only asks for less, we'll give him
9292 	 * that much.
9293 	 */
9294 	/* XXX KDM what malloc flags should we use here?? */
9295 	ctsio->kern_data_ptr = malloc(sizeof(*inq_ptr), M_CTL, M_WAITOK | M_ZERO);
9296 	if (ctsio->kern_data_ptr == NULL) {
9297 		ctsio->io_hdr.status = CTL_SCSI_ERROR;
9298 		ctsio->scsi_status = SCSI_STATUS_BUSY;
9299 		ctl_done((union ctl_io *)ctsio);
9300 		return (CTL_RETVAL_COMPLETE);
9301 	}
9302 	inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
9303 	ctsio->kern_sg_entries = 0;
9304 	ctsio->kern_data_resid = 0;
9305 	ctsio->kern_rel_offset = 0;
9306 
9307 	if (sizeof(*inq_ptr) < alloc_len) {
9308 		ctsio->residual = alloc_len - sizeof(*inq_ptr);
9309 		ctsio->kern_data_len = sizeof(*inq_ptr);
9310 		ctsio->kern_total_len = sizeof(*inq_ptr);
9311 	} else {
9312 		ctsio->residual = 0;
9313 		ctsio->kern_data_len = alloc_len;
9314 		ctsio->kern_total_len = alloc_len;
9315 	}
9316 
9317 	/*
9318 	 * If we have a LUN configured, report it as connected.  Otherwise,
9319 	 * report that it is offline or no device is supported, depending
9320 	 * on the value of inquiry_pq_no_lun.
9321 	 *
9322 	 * According to the spec (SPC-4 r34), the peripheral qualifier
9323 	 * SID_QUAL_LU_OFFLINE (001b) is used in the following scenario:
9324 	 *
9325 	 * "A peripheral device having the specified peripheral device type
9326 	 * is not connected to this logical unit. However, the device
9327 	 * server is capable of supporting the specified peripheral device
9328 	 * type on this logical unit."
9329 	 *
9330 	 * According to the same spec, the peripheral qualifier
9331 	 * SID_QUAL_BAD_LU (011b) is used in this scenario:
9332 	 *
9333 	 * "The device server is not capable of supporting a peripheral
9334 	 * device on this logical unit. For this peripheral qualifier the
9335 	 * peripheral device type shall be set to 1Fh. All other peripheral
9336 	 * device type values are reserved for this peripheral qualifier."
9337 	 *
9338 	 * Given the text, it would seem that we probably want to report that
9339 	 * the LUN is offline here.  There is no LUN connected, but we can
9340 	 * support a LUN at the given LUN number.
9341 	 *
9342 	 * In the real world, though, it sounds like things are a little
9343 	 * different:
9344 	 *
9345 	 * - Linux, when presented with a LUN with the offline peripheral
9346 	 *   qualifier, will create an sg driver instance for it.  So when
9347 	 *   you attach it to CTL, you wind up with a ton of sg driver
9348 	 *   instances.  (One for every LUN that Linux bothered to probe.)
9349 	 *   Linux does this despite the fact that it issues a REPORT LUNs
9350 	 *   to LUN 0 to get the inventory of supported LUNs.
9351 	 *
9352 	 * - There is other anecdotal evidence (from Emulex folks) about
9353 	 *   arrays that use the offline peripheral qualifier for LUNs that
9354 	 *   are on the "passive" path in an active/passive array.
9355 	 *
9356 	 * So the solution is provide a hopefully reasonable default
9357 	 * (return bad/no LUN) and allow the user to change the behavior
9358 	 * with a tunable/sysctl variable.
9359 	 */
9360 	if (lun != NULL)
9361 		inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9362 				  lun->be_lun->lun_type;
9363 	else if (ctl_softc->inquiry_pq_no_lun == 0)
9364 		inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9365 	else
9366 		inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
9367 
9368 	/* RMB in byte 2 is 0 */
9369 	inq_ptr->version = SCSI_REV_SPC3;
9370 
9371 	/*
9372 	 * According to SAM-3, even if a device only supports a single
9373 	 * level of LUN addressing, it should still set the HISUP bit:
9374 	 *
9375 	 * 4.9.1 Logical unit numbers overview
9376 	 *
9377 	 * All logical unit number formats described in this standard are
9378 	 * hierarchical in structure even when only a single level in that
9379 	 * hierarchy is used. The HISUP bit shall be set to one in the
9380 	 * standard INQUIRY data (see SPC-2) when any logical unit number
9381 	 * format described in this standard is used.  Non-hierarchical
9382 	 * formats are outside the scope of this standard.
9383 	 *
9384 	 * Therefore we set the HiSup bit here.
9385 	 *
9386 	 * The reponse format is 2, per SPC-3.
9387 	 */
9388 	inq_ptr->response_format = SID_HiSup | 2;
9389 
9390 	inq_ptr->additional_length = sizeof(*inq_ptr) - 4;
9391 	CTL_DEBUG_PRINT(("additional_length = %d\n",
9392 			 inq_ptr->additional_length));
9393 
9394 	inq_ptr->spc3_flags = SPC3_SID_TPGS_IMPLICIT;
9395 	/* 16 bit addressing */
9396 	if (is_fc == 0)
9397 		inq_ptr->spc2_flags = SPC2_SID_ADDR16;
9398 	/* XXX set the SID_MultiP bit here if we're actually going to
9399 	   respond on multiple ports */
9400 	inq_ptr->spc2_flags |= SPC2_SID_MultiP;
9401 
9402 	/* 16 bit data bus, synchronous transfers */
9403 	/* XXX these flags don't apply for FC */
9404 	if (is_fc == 0)
9405 		inq_ptr->flags = SID_WBus16 | SID_Sync;
9406 	/*
9407 	 * XXX KDM do we want to support tagged queueing on the control
9408 	 * device at all?
9409 	 */
9410 	if ((lun == NULL)
9411 	 || (lun->be_lun->lun_type != T_PROCESSOR))
9412 		inq_ptr->flags |= SID_CmdQue;
9413 	/*
9414 	 * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
9415 	 * We have 8 bytes for the vendor name, and 16 bytes for the device
9416 	 * name and 4 bytes for the revision.
9417 	 */
9418 	strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
9419 	if (lun == NULL) {
9420 		strcpy(inq_ptr->product, CTL_DIRECT_PRODUCT);
9421 	} else {
9422 		switch (lun->be_lun->lun_type) {
9423 		case T_DIRECT:
9424 			strcpy(inq_ptr->product, CTL_DIRECT_PRODUCT);
9425 			break;
9426 		case T_PROCESSOR:
9427 			strcpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT);
9428 			break;
9429 		default:
9430 			strcpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT);
9431 			break;
9432 		}
9433 	}
9434 
9435 	/*
9436 	 * XXX make this a macro somewhere so it automatically gets
9437 	 * incremented when we make changes.
9438 	 */
9439 	strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
9440 
9441 	/*
9442 	 * For parallel SCSI, we support double transition and single
9443 	 * transition clocking.  We also support QAS (Quick Arbitration
9444 	 * and Selection) and Information Unit transfers on both the
9445 	 * control and array devices.
9446 	 */
9447 	if (is_fc == 0)
9448 		inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
9449 				    SID_SPI_IUS;
9450 
9451 	/* SAM-3 */
9452 	scsi_ulto2b(0x0060, inq_ptr->version1);
9453 	/* SPC-3 (no version claimed) XXX should we claim a version? */
9454 	scsi_ulto2b(0x0300, inq_ptr->version2);
9455 	if (is_fc) {
9456 		/* FCP-2 ANSI INCITS.350:2003 */
9457 		scsi_ulto2b(0x0917, inq_ptr->version3);
9458 	} else {
9459 		/* SPI-4 ANSI INCITS.362:200x */
9460 		scsi_ulto2b(0x0B56, inq_ptr->version3);
9461 	}
9462 
9463 	if (lun == NULL) {
9464 		/* SBC-2 (no version claimed) XXX should we claim a version? */
9465 		scsi_ulto2b(0x0320, inq_ptr->version4);
9466 	} else {
9467 		switch (lun->be_lun->lun_type) {
9468 		case T_DIRECT:
9469 			/*
9470 			 * SBC-2 (no version claimed) XXX should we claim a
9471 			 * version?
9472 			 */
9473 			scsi_ulto2b(0x0320, inq_ptr->version4);
9474 			break;
9475 		case T_PROCESSOR:
9476 		default:
9477 			break;
9478 		}
9479 	}
9480 
9481 	ctsio->scsi_status = SCSI_STATUS_OK;
9482 	if (ctsio->kern_data_len > 0) {
9483 		ctsio->be_move_done = ctl_config_move_done;
9484 		ctl_datamove((union ctl_io *)ctsio);
9485 	} else {
9486 		ctsio->io_hdr.status = CTL_SUCCESS;
9487 		ctl_done((union ctl_io *)ctsio);
9488 	}
9489 
9490 	return (CTL_RETVAL_COMPLETE);
9491 }
9492 
9493 int
9494 ctl_inquiry(struct ctl_scsiio *ctsio)
9495 {
9496 	struct scsi_inquiry *cdb;
9497 	int retval;
9498 
9499 	cdb = (struct scsi_inquiry *)ctsio->cdb;
9500 
9501 	retval = 0;
9502 
9503 	CTL_DEBUG_PRINT(("ctl_inquiry\n"));
9504 
9505 	/*
9506 	 * Right now, we don't support the CmdDt inquiry information.
9507 	 * This would be nice to support in the future.  When we do
9508 	 * support it, we should change this test so that it checks to make
9509 	 * sure SI_EVPD and SI_CMDDT aren't both set at the same time.
9510 	 */
9511 #ifdef notyet
9512 	if (((cdb->byte2 & SI_EVPD)
9513 	 && (cdb->byte2 & SI_CMDDT)))
9514 #endif
9515 	if (cdb->byte2 & SI_CMDDT) {
9516 		/*
9517 		 * Point to the SI_CMDDT bit.  We might change this
9518 		 * when we support SI_CMDDT, but since both bits would be
9519 		 * "wrong", this should probably just stay as-is then.
9520 		 */
9521 		ctl_set_invalid_field(ctsio,
9522 				      /*sks_valid*/ 1,
9523 				      /*command*/ 1,
9524 				      /*field*/ 1,
9525 				      /*bit_valid*/ 1,
9526 				      /*bit*/ 1);
9527 		ctl_done((union ctl_io *)ctsio);
9528 		return (CTL_RETVAL_COMPLETE);
9529 	}
9530 	if (cdb->byte2 & SI_EVPD)
9531 		retval = ctl_inquiry_evpd(ctsio);
9532 #ifdef notyet
9533 	else if (cdb->byte2 & SI_CMDDT)
9534 		retval = ctl_inquiry_cmddt(ctsio);
9535 #endif
9536 	else
9537 		retval = ctl_inquiry_std(ctsio);
9538 
9539 	return (retval);
9540 }
9541 
9542 /*
9543  * For known CDB types, parse the LBA and length.
9544  */
9545 static int
9546 ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint32_t *len)
9547 {
9548 	if (io->io_hdr.io_type != CTL_IO_SCSI)
9549 		return (1);
9550 
9551 	switch (io->scsiio.cdb[0]) {
9552 	case READ_6:
9553 	case WRITE_6: {
9554 		struct scsi_rw_6 *cdb;
9555 
9556 		cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
9557 
9558 		*lba = scsi_3btoul(cdb->addr);
9559 		/* only 5 bits are valid in the most significant address byte */
9560 		*lba &= 0x1fffff;
9561 		*len = cdb->length;
9562 		break;
9563 	}
9564 	case READ_10:
9565 	case WRITE_10: {
9566 		struct scsi_rw_10 *cdb;
9567 
9568 		cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
9569 
9570 		*lba = scsi_4btoul(cdb->addr);
9571 		*len = scsi_2btoul(cdb->length);
9572 		break;
9573 	}
9574 	case WRITE_VERIFY_10: {
9575 		struct scsi_write_verify_10 *cdb;
9576 
9577 		cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
9578 
9579 		*lba = scsi_4btoul(cdb->addr);
9580 		*len = scsi_2btoul(cdb->length);
9581 		break;
9582 	}
9583 	case READ_12:
9584 	case WRITE_12: {
9585 		struct scsi_rw_12 *cdb;
9586 
9587 		cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
9588 
9589 		*lba = scsi_4btoul(cdb->addr);
9590 		*len = scsi_4btoul(cdb->length);
9591 		break;
9592 	}
9593 	case WRITE_VERIFY_12: {
9594 		struct scsi_write_verify_12 *cdb;
9595 
9596 		cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
9597 
9598 		*lba = scsi_4btoul(cdb->addr);
9599 		*len = scsi_4btoul(cdb->length);
9600 		break;
9601 	}
9602 	case READ_16:
9603 	case WRITE_16: {
9604 		struct scsi_rw_16 *cdb;
9605 
9606 		cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
9607 
9608 		*lba = scsi_8btou64(cdb->addr);
9609 		*len = scsi_4btoul(cdb->length);
9610 		break;
9611 	}
9612 	case WRITE_VERIFY_16: {
9613 		struct scsi_write_verify_16 *cdb;
9614 
9615 		cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
9616 
9617 
9618 		*lba = scsi_8btou64(cdb->addr);
9619 		*len = scsi_4btoul(cdb->length);
9620 		break;
9621 	}
9622 	default:
9623 		return (1);
9624 		break; /* NOTREACHED */
9625 	}
9626 
9627 	return (0);
9628 }
9629 
9630 static ctl_action
9631 ctl_extent_check_lba(uint64_t lba1, uint32_t len1, uint64_t lba2, uint32_t len2)
9632 {
9633 	uint64_t endlba1, endlba2;
9634 
9635 	endlba1 = lba1 + len1 - 1;
9636 	endlba2 = lba2 + len2 - 1;
9637 
9638 	if ((endlba1 < lba2)
9639 	 || (endlba2 < lba1))
9640 		return (CTL_ACTION_PASS);
9641 	else
9642 		return (CTL_ACTION_BLOCK);
9643 }
9644 
9645 static ctl_action
9646 ctl_extent_check(union ctl_io *io1, union ctl_io *io2)
9647 {
9648 	uint64_t lba1, lba2;
9649 	uint32_t len1, len2;
9650 	int retval;
9651 
9652 	retval = ctl_get_lba_len(io1, &lba1, &len1);
9653 	if (retval != 0)
9654 		return (CTL_ACTION_ERROR);
9655 
9656 	retval = ctl_get_lba_len(io2, &lba2, &len2);
9657 	if (retval != 0)
9658 		return (CTL_ACTION_ERROR);
9659 
9660 	return (ctl_extent_check_lba(lba1, len1, lba2, len2));
9661 }
9662 
9663 static ctl_action
9664 ctl_check_for_blockage(union ctl_io *pending_io, union ctl_io *ooa_io)
9665 {
9666 	struct ctl_cmd_entry *pending_entry, *ooa_entry;
9667 	ctl_serialize_action *serialize_row;
9668 
9669 	/*
9670 	 * The initiator attempted multiple untagged commands at the same
9671 	 * time.  Can't do that.
9672 	 */
9673 	if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
9674 	 && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
9675 	 && ((pending_io->io_hdr.nexus.targ_port ==
9676 	      ooa_io->io_hdr.nexus.targ_port)
9677 	  && (pending_io->io_hdr.nexus.initid.id ==
9678 	      ooa_io->io_hdr.nexus.initid.id))
9679 	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
9680 		return (CTL_ACTION_OVERLAP);
9681 
9682 	/*
9683 	 * The initiator attempted to send multiple tagged commands with
9684 	 * the same ID.  (It's fine if different initiators have the same
9685 	 * tag ID.)
9686 	 *
9687 	 * Even if all of those conditions are true, we don't kill the I/O
9688 	 * if the command ahead of us has been aborted.  We won't end up
9689 	 * sending it to the FETD, and it's perfectly legal to resend a
9690 	 * command with the same tag number as long as the previous
9691 	 * instance of this tag number has been aborted somehow.
9692 	 */
9693 	if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
9694 	 && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
9695 	 && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
9696 	 && ((pending_io->io_hdr.nexus.targ_port ==
9697 	      ooa_io->io_hdr.nexus.targ_port)
9698 	  && (pending_io->io_hdr.nexus.initid.id ==
9699 	      ooa_io->io_hdr.nexus.initid.id))
9700 	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
9701 		return (CTL_ACTION_OVERLAP_TAG);
9702 
9703 	/*
9704 	 * If we get a head of queue tag, SAM-3 says that we should
9705 	 * immediately execute it.
9706 	 *
9707 	 * What happens if this command would normally block for some other
9708 	 * reason?  e.g. a request sense with a head of queue tag
9709 	 * immediately after a write.  Normally that would block, but this
9710 	 * will result in its getting executed immediately...
9711 	 *
9712 	 * We currently return "pass" instead of "skip", so we'll end up
9713 	 * going through the rest of the queue to check for overlapped tags.
9714 	 *
9715 	 * XXX KDM check for other types of blockage first??
9716 	 */
9717 	if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
9718 		return (CTL_ACTION_PASS);
9719 
9720 	/*
9721 	 * Ordered tags have to block until all items ahead of them
9722 	 * have completed.  If we get called with an ordered tag, we always
9723 	 * block, if something else is ahead of us in the queue.
9724 	 */
9725 	if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
9726 		return (CTL_ACTION_BLOCK);
9727 
9728 	/*
9729 	 * Simple tags get blocked until all head of queue and ordered tags
9730 	 * ahead of them have completed.  I'm lumping untagged commands in
9731 	 * with simple tags here.  XXX KDM is that the right thing to do?
9732 	 */
9733 	if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
9734 	  || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
9735 	 && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
9736 	  || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
9737 		return (CTL_ACTION_BLOCK);
9738 
9739 	pending_entry = &ctl_cmd_table[pending_io->scsiio.cdb[0]];
9740 	ooa_entry = &ctl_cmd_table[ooa_io->scsiio.cdb[0]];
9741 
9742 	serialize_row = ctl_serialize_table[ooa_entry->seridx];
9743 
9744 	switch (serialize_row[pending_entry->seridx]) {
9745 	case CTL_SER_BLOCK:
9746 		return (CTL_ACTION_BLOCK);
9747 		break; /* NOTREACHED */
9748 	case CTL_SER_EXTENT:
9749 		return (ctl_extent_check(pending_io, ooa_io));
9750 		break; /* NOTREACHED */
9751 	case CTL_SER_PASS:
9752 		return (CTL_ACTION_PASS);
9753 		break; /* NOTREACHED */
9754 	case CTL_SER_SKIP:
9755 		return (CTL_ACTION_SKIP);
9756 		break;
9757 	default:
9758 		panic("invalid serialization value %d",
9759 		      serialize_row[pending_entry->seridx]);
9760 		break; /* NOTREACHED */
9761 	}
9762 
9763 	return (CTL_ACTION_ERROR);
9764 }
9765 
9766 /*
9767  * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
9768  * Assumptions:
9769  * - caller holds ctl_lock
9770  * - pending_io is generally either incoming, or on the blocked queue
9771  * - starting I/O is the I/O we want to start the check with.
9772  */
9773 static ctl_action
9774 ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
9775 	      union ctl_io *starting_io)
9776 {
9777 	union ctl_io *ooa_io;
9778 	ctl_action action;
9779 
9780 	/*
9781 	 * Run back along the OOA queue, starting with the current
9782 	 * blocked I/O and going through every I/O before it on the
9783 	 * queue.  If starting_io is NULL, we'll just end up returning
9784 	 * CTL_ACTION_PASS.
9785 	 */
9786 	for (ooa_io = starting_io; ooa_io != NULL;
9787 	     ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
9788 	     ooa_links)){
9789 
9790 		/*
9791 		 * This routine just checks to see whether
9792 		 * cur_blocked is blocked by ooa_io, which is ahead
9793 		 * of it in the queue.  It doesn't queue/dequeue
9794 		 * cur_blocked.
9795 		 */
9796 		action = ctl_check_for_blockage(pending_io, ooa_io);
9797 		switch (action) {
9798 		case CTL_ACTION_BLOCK:
9799 		case CTL_ACTION_OVERLAP:
9800 		case CTL_ACTION_OVERLAP_TAG:
9801 		case CTL_ACTION_SKIP:
9802 		case CTL_ACTION_ERROR:
9803 			return (action);
9804 			break; /* NOTREACHED */
9805 		case CTL_ACTION_PASS:
9806 			break;
9807 		default:
9808 			panic("invalid action %d", action);
9809 			break;  /* NOTREACHED */
9810 		}
9811 	}
9812 
9813 	return (CTL_ACTION_PASS);
9814 }
9815 
9816 /*
9817  * Assumptions:
9818  * - An I/O has just completed, and has been removed from the per-LUN OOA
9819  *   queue, so some items on the blocked queue may now be unblocked.
9820  * - The caller holds ctl_softc->ctl_lock
9821  */
9822 static int
9823 ctl_check_blocked(struct ctl_lun *lun)
9824 {
9825 	union ctl_io *cur_blocked, *next_blocked;
9826 
9827 	/*
9828 	 * Run forward from the head of the blocked queue, checking each
9829 	 * entry against the I/Os prior to it on the OOA queue to see if
9830 	 * there is still any blockage.
9831 	 *
9832 	 * We cannot use the TAILQ_FOREACH() macro, because it can't deal
9833 	 * with our removing a variable on it while it is traversing the
9834 	 * list.
9835 	 */
9836 	for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
9837 	     cur_blocked != NULL; cur_blocked = next_blocked) {
9838 		union ctl_io *prev_ooa;
9839 		ctl_action action;
9840 
9841 		next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
9842 							  blocked_links);
9843 
9844 		prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
9845 						      ctl_ooaq, ooa_links);
9846 
9847 		/*
9848 		 * If cur_blocked happens to be the first item in the OOA
9849 		 * queue now, prev_ooa will be NULL, and the action
9850 		 * returned will just be CTL_ACTION_PASS.
9851 		 */
9852 		action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
9853 
9854 		switch (action) {
9855 		case CTL_ACTION_BLOCK:
9856 			/* Nothing to do here, still blocked */
9857 			break;
9858 		case CTL_ACTION_OVERLAP:
9859 		case CTL_ACTION_OVERLAP_TAG:
9860 			/*
9861 			 * This shouldn't happen!  In theory we've already
9862 			 * checked this command for overlap...
9863 			 */
9864 			break;
9865 		case CTL_ACTION_PASS:
9866 		case CTL_ACTION_SKIP: {
9867 			struct ctl_softc *softc;
9868 			struct ctl_cmd_entry *entry;
9869 			uint32_t initidx;
9870 			uint8_t opcode;
9871 			int isc_retval;
9872 
9873 			/*
9874 			 * The skip case shouldn't happen, this transaction
9875 			 * should have never made it onto the blocked queue.
9876 			 */
9877 			/*
9878 			 * This I/O is no longer blocked, we can remove it
9879 			 * from the blocked queue.  Since this is a TAILQ
9880 			 * (doubly linked list), we can do O(1) removals
9881 			 * from any place on the list.
9882 			 */
9883 			TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
9884 				     blocked_links);
9885 			cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
9886 
9887 			if (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC){
9888 				/*
9889 				 * Need to send IO back to original side to
9890 				 * run
9891 				 */
9892 				union ctl_ha_msg msg_info;
9893 
9894 				msg_info.hdr.original_sc =
9895 					cur_blocked->io_hdr.original_sc;
9896 				msg_info.hdr.serializing_sc = cur_blocked;
9897 				msg_info.hdr.msg_type = CTL_MSG_R2R;
9898 				if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
9899 				     &msg_info, sizeof(msg_info), 0)) >
9900 				     CTL_HA_STATUS_SUCCESS) {
9901 					printf("CTL:Check Blocked error from "
9902 					       "ctl_ha_msg_send %d\n",
9903 					       isc_retval);
9904 				}
9905 				break;
9906 			}
9907 			opcode = cur_blocked->scsiio.cdb[0];
9908 			entry = &ctl_cmd_table[opcode];
9909 			softc = control_softc;
9910 
9911 			initidx = ctl_get_initindex(&cur_blocked->io_hdr.nexus);
9912 
9913 			/*
9914 			 * Check this I/O for LUN state changes that may
9915 			 * have happened while this command was blocked.
9916 			 * The LUN state may have been changed by a command
9917 			 * ahead of us in the queue, so we need to re-check
9918 			 * for any states that can be caused by SCSI
9919 			 * commands.
9920 			 */
9921 			if (ctl_scsiio_lun_check(softc, lun, entry,
9922 						 &cur_blocked->scsiio) == 0) {
9923 				cur_blocked->io_hdr.flags |=
9924 				                      CTL_FLAG_IS_WAS_ON_RTR;
9925 				STAILQ_INSERT_TAIL(&lun->ctl_softc->rtr_queue,
9926 						   &cur_blocked->io_hdr, links);
9927 				/*
9928 				 * In the non CTL_DONE_THREAD case, we need
9929 				 * to wake up the work thread here.  When
9930 				 * we're processing completed requests from
9931 				 * the work thread context, we'll pop back
9932 				 * around and end up pulling things off the
9933 				 * RtR queue.  When we aren't processing
9934 				 * things from the work thread context,
9935 				 * though, we won't ever check the RtR queue.
9936 				 * So we need to wake up the thread to clear
9937 				 * things off the queue.  Otherwise this
9938 				 * transaction will just sit on the RtR queue
9939 				 * until a new I/O comes in.  (Which may or
9940 				 * may not happen...)
9941 				 */
9942 #ifndef CTL_DONE_THREAD
9943 				ctl_wakeup_thread();
9944 #endif
9945 			} else
9946 				ctl_done_lock(cur_blocked, /*have_lock*/ 1);
9947 			break;
9948 		}
9949 		default:
9950 			/*
9951 			 * This probably shouldn't happen -- we shouldn't
9952 			 * get CTL_ACTION_ERROR, or anything else.
9953 			 */
9954 			break;
9955 		}
9956 	}
9957 
9958 	return (CTL_RETVAL_COMPLETE);
9959 }
9960 
9961 /*
9962  * This routine (with one exception) checks LUN flags that can be set by
9963  * commands ahead of us in the OOA queue.  These flags have to be checked
9964  * when a command initially comes in, and when we pull a command off the
9965  * blocked queue and are preparing to execute it.  The reason we have to
9966  * check these flags for commands on the blocked queue is that the LUN
9967  * state may have been changed by a command ahead of us while we're on the
9968  * blocked queue.
9969  *
9970  * Ordering is somewhat important with these checks, so please pay
9971  * careful attention to the placement of any new checks.
9972  */
9973 static int
9974 ctl_scsiio_lun_check(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
9975 		     struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
9976 {
9977 	int retval;
9978 
9979 	retval = 0;
9980 
9981 	/*
9982 	 * If this shelf is a secondary shelf controller, we have to reject
9983 	 * any media access commands.
9984 	 */
9985 #if 0
9986 	/* No longer needed for HA */
9987 	if (((ctl_softc->flags & CTL_FLAG_MASTER_SHELF) == 0)
9988 	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_SECONDARY) == 0)) {
9989 		ctl_set_lun_standby(ctsio);
9990 		retval = 1;
9991 		goto bailout;
9992 	}
9993 #endif
9994 
9995 	/*
9996 	 * Check for a reservation conflict.  If this command isn't allowed
9997 	 * even on reserved LUNs, and if this initiator isn't the one who
9998 	 * reserved us, reject the command with a reservation conflict.
9999 	 */
10000 	if ((lun->flags & CTL_LUN_RESERVED)
10001 	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
10002 		if ((ctsio->io_hdr.nexus.initid.id != lun->rsv_nexus.initid.id)
10003 		 || (ctsio->io_hdr.nexus.targ_port != lun->rsv_nexus.targ_port)
10004 		 || (ctsio->io_hdr.nexus.targ_target.id !=
10005 		     lun->rsv_nexus.targ_target.id)) {
10006 			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
10007 			ctsio->io_hdr.status = CTL_SCSI_ERROR;
10008 			retval = 1;
10009 			goto bailout;
10010 		}
10011 	}
10012 
10013 	if ( (lun->flags & CTL_LUN_PR_RESERVED)
10014 	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV) == 0)) {
10015 		uint32_t residx;
10016 
10017 		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
10018 		/*
10019 		 * if we aren't registered or it's a res holder type
10020 		 * reservation and this isn't the res holder then set a
10021 		 * conflict.
10022 		 * NOTE: Commands which might be allowed on write exclusive
10023 		 * type reservations are checked in the particular command
10024 		 * for a conflict. Read and SSU are the only ones.
10025 		 */
10026 		if (!lun->per_res[residx].registered
10027 		 || (residx != lun->pr_res_idx && lun->res_type < 4)) {
10028 			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
10029 			ctsio->io_hdr.status = CTL_SCSI_ERROR;
10030 			retval = 1;
10031 			goto bailout;
10032 		}
10033 
10034 	}
10035 
10036 	if ((lun->flags & CTL_LUN_OFFLINE)
10037 	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_OFFLINE) == 0)) {
10038 		ctl_set_lun_not_ready(ctsio);
10039 		retval = 1;
10040 		goto bailout;
10041 	}
10042 
10043 	/*
10044 	 * If the LUN is stopped, see if this particular command is allowed
10045 	 * for a stopped lun.  Otherwise, reject it with 0x04,0x02.
10046 	 */
10047 	if ((lun->flags & CTL_LUN_STOPPED)
10048 	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_STOPPED) == 0)) {
10049 		/* "Logical unit not ready, initializing cmd. required" */
10050 		ctl_set_lun_stopped(ctsio);
10051 		retval = 1;
10052 		goto bailout;
10053 	}
10054 
10055 	if ((lun->flags & CTL_LUN_INOPERABLE)
10056 	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_INOPERABLE) == 0)) {
10057 		/* "Medium format corrupted" */
10058 		ctl_set_medium_format_corrupted(ctsio);
10059 		retval = 1;
10060 		goto bailout;
10061 	}
10062 
10063 bailout:
10064 	return (retval);
10065 
10066 }
10067 
10068 static void
10069 ctl_failover_io(union ctl_io *io, int have_lock)
10070 {
10071 	ctl_set_busy(&io->scsiio);
10072 	ctl_done_lock(io, have_lock);
10073 }
10074 
10075 static void
10076 ctl_failover(void)
10077 {
10078 	struct ctl_lun *lun;
10079 	struct ctl_softc *ctl_softc;
10080 	union ctl_io *next_io, *pending_io;
10081 	union ctl_io *io;
10082 	int lun_idx;
10083 	int i;
10084 
10085 	ctl_softc = control_softc;
10086 
10087 	mtx_lock(&ctl_softc->ctl_lock);
10088 	/*
10089 	 * Remove any cmds from the other SC from the rtr queue.  These
10090 	 * will obviously only be for LUNs for which we're the primary.
10091 	 * We can't send status or get/send data for these commands.
10092 	 * Since they haven't been executed yet, we can just remove them.
10093 	 * We'll either abort them or delete them below, depending on
10094 	 * which HA mode we're in.
10095 	 */
10096 	for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->rtr_queue);
10097 	     io != NULL; io = next_io) {
10098 		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
10099 		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
10100 			STAILQ_REMOVE(&ctl_softc->rtr_queue, &io->io_hdr,
10101 				      ctl_io_hdr, links);
10102 	}
10103 
10104 	for (lun_idx=0; lun_idx < ctl_softc->num_luns; lun_idx++) {
10105 		lun = ctl_softc->ctl_luns[lun_idx];
10106 		if (lun==NULL)
10107 			continue;
10108 
10109 		/*
10110 		 * Processor LUNs are primary on both sides.
10111 		 * XXX will this always be true?
10112 		 */
10113 		if (lun->be_lun->lun_type == T_PROCESSOR)
10114 			continue;
10115 
10116 		if ((lun->flags & CTL_LUN_PRIMARY_SC)
10117 		 && (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
10118 			printf("FAILOVER: primary lun %d\n", lun_idx);
10119 		        /*
10120 			 * Remove all commands from the other SC. First from the
10121 			 * blocked queue then from the ooa queue. Once we have
10122 			 * removed them. Call ctl_check_blocked to see if there
10123 			 * is anything that can run.
10124 			 */
10125 			for (io = (union ctl_io *)TAILQ_FIRST(
10126 			     &lun->blocked_queue); io != NULL; io = next_io) {
10127 
10128 		        	next_io = (union ctl_io *)TAILQ_NEXT(
10129 				    &io->io_hdr, blocked_links);
10130 
10131 				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
10132 					TAILQ_REMOVE(&lun->blocked_queue,
10133 						     &io->io_hdr,blocked_links);
10134 					io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
10135 					TAILQ_REMOVE(&lun->ooa_queue,
10136 						     &io->io_hdr, ooa_links);
10137 
10138 					ctl_free_io_internal(io, 1);
10139 				}
10140 			}
10141 
10142 			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
10143 	     		     io != NULL; io = next_io) {
10144 
10145 		        	next_io = (union ctl_io *)TAILQ_NEXT(
10146 				    &io->io_hdr, ooa_links);
10147 
10148 				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
10149 
10150 					TAILQ_REMOVE(&lun->ooa_queue,
10151 						&io->io_hdr,
10152 					     	ooa_links);
10153 
10154 					ctl_free_io_internal(io, 1);
10155 				}
10156 			}
10157 			ctl_check_blocked(lun);
10158 		} else if ((lun->flags & CTL_LUN_PRIMARY_SC)
10159 			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
10160 
10161 			printf("FAILOVER: primary lun %d\n", lun_idx);
10162 			/*
10163 			 * Abort all commands from the other SC.  We can't
10164 			 * send status back for them now.  These should get
10165 			 * cleaned up when they are completed or come out
10166 			 * for a datamove operation.
10167 			 */
10168 			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
10169 	     		     io != NULL; io = next_io) {
10170 		        	next_io = (union ctl_io *)TAILQ_NEXT(
10171 					&io->io_hdr, ooa_links);
10172 
10173 				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
10174 					io->io_hdr.flags |= CTL_FLAG_ABORT;
10175 			}
10176 		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
10177 			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
10178 
10179 			printf("FAILOVER: secondary lun %d\n", lun_idx);
10180 
10181 			lun->flags |= CTL_LUN_PRIMARY_SC;
10182 
10183 			/*
10184 			 * We send all I/O that was sent to this controller
10185 			 * and redirected to the other side back with
10186 			 * busy status, and have the initiator retry it.
10187 			 * Figuring out how much data has been transferred,
10188 			 * etc. and picking up where we left off would be
10189 			 * very tricky.
10190 			 *
10191 			 * XXX KDM need to remove I/O from the blocked
10192 			 * queue as well!
10193 			 */
10194 			for (pending_io = (union ctl_io *)TAILQ_FIRST(
10195 			     &lun->ooa_queue); pending_io != NULL;
10196 			     pending_io = next_io) {
10197 
10198 				next_io =  (union ctl_io *)TAILQ_NEXT(
10199 					&pending_io->io_hdr, ooa_links);
10200 
10201 				pending_io->io_hdr.flags &=
10202 					~CTL_FLAG_SENT_2OTHER_SC;
10203 
10204 				if (pending_io->io_hdr.flags &
10205 				    CTL_FLAG_IO_ACTIVE) {
10206 					pending_io->io_hdr.flags |=
10207 						CTL_FLAG_FAILOVER;
10208 				} else {
10209 					ctl_set_busy(&pending_io->scsiio);
10210 					ctl_done_lock(pending_io,
10211 						      /*have_lock*/1);
10212 				}
10213 			}
10214 
10215 			/*
10216 			 * Build Unit Attention
10217 			 */
10218 			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
10219 				lun->pending_sense[i].ua_pending |=
10220 				                     CTL_UA_ASYM_ACC_CHANGE;
10221 			}
10222 		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
10223 			&& (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
10224 			printf("FAILOVER: secondary lun %d\n", lun_idx);
10225 			/*
10226 			 * if the first io on the OOA is not on the RtR queue
10227 			 * add it.
10228 			 */
10229 			lun->flags |= CTL_LUN_PRIMARY_SC;
10230 
10231 			pending_io = (union ctl_io *)TAILQ_FIRST(
10232 			    &lun->ooa_queue);
10233 			if (pending_io==NULL) {
10234 				printf("Nothing on OOA queue\n");
10235 				continue;
10236 			}
10237 
10238 			pending_io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
10239 			if ((pending_io->io_hdr.flags &
10240 			     CTL_FLAG_IS_WAS_ON_RTR) == 0) {
10241 				pending_io->io_hdr.flags |=
10242 				    CTL_FLAG_IS_WAS_ON_RTR;
10243 				STAILQ_INSERT_TAIL(&ctl_softc->rtr_queue,
10244 						   &pending_io->io_hdr, links);
10245 			}
10246 #if 0
10247 			else
10248 			{
10249 				printf("Tag 0x%04x is running\n",
10250 				      pending_io->scsiio.tag_num);
10251 			}
10252 #endif
10253 
10254 			next_io = (union ctl_io *)TAILQ_NEXT(
10255 			    &pending_io->io_hdr, ooa_links);
10256 			for (pending_io=next_io; pending_io != NULL;
10257 			     pending_io = next_io) {
10258 				pending_io->io_hdr.flags &=
10259 				    ~CTL_FLAG_SENT_2OTHER_SC;
10260 				next_io = (union ctl_io *)TAILQ_NEXT(
10261 					&pending_io->io_hdr, ooa_links);
10262 				if (pending_io->io_hdr.flags &
10263 				    CTL_FLAG_IS_WAS_ON_RTR) {
10264 #if 0
10265 				        printf("Tag 0x%04x is running\n",
10266 				      		pending_io->scsiio.tag_num);
10267 #endif
10268 					continue;
10269 				}
10270 
10271 				switch (ctl_check_ooa(lun, pending_io,
10272 			            (union ctl_io *)TAILQ_PREV(
10273 				    &pending_io->io_hdr, ctl_ooaq,
10274 				    ooa_links))) {
10275 
10276 				case CTL_ACTION_BLOCK:
10277 					TAILQ_INSERT_TAIL(&lun->blocked_queue,
10278 							  &pending_io->io_hdr,
10279 							  blocked_links);
10280 					pending_io->io_hdr.flags |=
10281 					    CTL_FLAG_BLOCKED;
10282 					break;
10283 				case CTL_ACTION_PASS:
10284 				case CTL_ACTION_SKIP:
10285 					pending_io->io_hdr.flags |=
10286 					    CTL_FLAG_IS_WAS_ON_RTR;
10287 					STAILQ_INSERT_TAIL(
10288 					    &ctl_softc->rtr_queue,
10289 					    &pending_io->io_hdr, links);
10290 					break;
10291 				case CTL_ACTION_OVERLAP:
10292 					ctl_set_overlapped_cmd(
10293 					    (struct ctl_scsiio *)pending_io);
10294 					ctl_done_lock(pending_io,
10295 						      /*have_lock*/ 1);
10296 					break;
10297 				case CTL_ACTION_OVERLAP_TAG:
10298 					ctl_set_overlapped_tag(
10299 					    (struct ctl_scsiio *)pending_io,
10300 					    pending_io->scsiio.tag_num & 0xff);
10301 					ctl_done_lock(pending_io,
10302 						      /*have_lock*/ 1);
10303 					break;
10304 				case CTL_ACTION_ERROR:
10305 				default:
10306 					ctl_set_internal_failure(
10307 						(struct ctl_scsiio *)pending_io,
10308 						0,  // sks_valid
10309 						0); //retry count
10310 					ctl_done_lock(pending_io,
10311 						      /*have_lock*/ 1);
10312 					break;
10313 				}
10314 			}
10315 
10316 			/*
10317 			 * Build Unit Attention
10318 			 */
10319 			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
10320 				lun->pending_sense[i].ua_pending |=
10321 				                     CTL_UA_ASYM_ACC_CHANGE;
10322 			}
10323 		} else {
10324 			panic("Unhandled HA mode failover, LUN flags = %#x, "
10325 			      "ha_mode = #%x", lun->flags, ctl_softc->ha_mode);
10326 		}
10327 	}
10328 	ctl_pause_rtr = 0;
10329 	mtx_unlock(&ctl_softc->ctl_lock);
10330 }
10331 
10332 static int
10333 ctl_scsiio_precheck(struct ctl_softc *ctl_softc, struct ctl_scsiio *ctsio)
10334 {
10335 	struct ctl_lun *lun;
10336 	struct ctl_cmd_entry *entry;
10337 	uint8_t opcode;
10338 	uint32_t initidx;
10339 	int retval;
10340 
10341 	retval = 0;
10342 
10343 	lun = NULL;
10344 
10345 	opcode = ctsio->cdb[0];
10346 
10347 	mtx_lock(&ctl_softc->ctl_lock);
10348 
10349 	if ((ctsio->io_hdr.nexus.targ_lun < CTL_MAX_LUNS)
10350 	 && (ctl_softc->ctl_luns[ctsio->io_hdr.nexus.targ_lun] != NULL)) {
10351 		lun = ctl_softc->ctl_luns[ctsio->io_hdr.nexus.targ_lun];
10352 		/*
10353 		 * If the LUN is invalid, pretend that it doesn't exist.
10354 		 * It will go away as soon as all pending I/O has been
10355 		 * completed.
10356 		 */
10357 		if (lun->flags & CTL_LUN_DISABLED) {
10358 			lun = NULL;
10359 		} else {
10360 			ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
10361 			ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
10362 				lun->be_lun;
10363 			if (lun->be_lun->lun_type == T_PROCESSOR) {
10364 				ctsio->io_hdr.flags |= CTL_FLAG_CONTROL_DEV;
10365 			}
10366 		}
10367 	} else {
10368 		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
10369 		ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
10370 	}
10371 
10372 	entry = &ctl_cmd_table[opcode];
10373 
10374 	ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
10375 	ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
10376 
10377 	/*
10378 	 * Check to see whether we can send this command to LUNs that don't
10379 	 * exist.  This should pretty much only be the case for inquiry
10380 	 * and request sense.  Further checks, below, really require having
10381 	 * a LUN, so we can't really check the command anymore.  Just put
10382 	 * it on the rtr queue.
10383 	 */
10384 	if (lun == NULL) {
10385 		if (entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS)
10386 			goto queue_rtr;
10387 
10388 		ctl_set_unsupported_lun(ctsio);
10389 		mtx_unlock(&ctl_softc->ctl_lock);
10390 		ctl_done((union ctl_io *)ctsio);
10391 		goto bailout;
10392 	} else {
10393 		/*
10394 		 * Every I/O goes into the OOA queue for a particular LUN, and
10395 		 * stays there until completion.
10396 		 */
10397 		TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
10398 
10399 		/*
10400 		 * Make sure we support this particular command on this LUN.
10401 		 * e.g., we don't support writes to the control LUN.
10402 		 */
10403 		switch (lun->be_lun->lun_type) {
10404 		case T_PROCESSOR:
10405 		 	if (((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0)
10406 			 && ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS)
10407 			      == 0)) {
10408 				ctl_set_invalid_opcode(ctsio);
10409 				mtx_unlock(&ctl_softc->ctl_lock);
10410 				ctl_done((union ctl_io *)ctsio);
10411 				goto bailout;
10412 			}
10413 			break;
10414 		case T_DIRECT:
10415 			if (((entry->flags & CTL_CMD_FLAG_OK_ON_SLUN) == 0)
10416 			 && ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS)
10417 			      == 0)){
10418 				ctl_set_invalid_opcode(ctsio);
10419 				mtx_unlock(&ctl_softc->ctl_lock);
10420 				ctl_done((union ctl_io *)ctsio);
10421 				goto bailout;
10422 			}
10423 			break;
10424 		default:
10425 			printf("Unsupported CTL LUN type %d\n",
10426 			       lun->be_lun->lun_type);
10427 			panic("Unsupported CTL LUN type %d\n",
10428 			      lun->be_lun->lun_type);
10429 			break; /* NOTREACHED */
10430 		}
10431 	}
10432 
10433 	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
10434 
10435 	/*
10436 	 * If we've got a request sense, it'll clear the contingent
10437 	 * allegiance condition.  Otherwise, if we have a CA condition for
10438 	 * this initiator, clear it, because it sent down a command other
10439 	 * than request sense.
10440 	 */
10441 	if ((opcode != REQUEST_SENSE)
10442 	 && (ctl_is_set(lun->have_ca, initidx)))
10443 		ctl_clear_mask(lun->have_ca, initidx);
10444 
10445 	/*
10446 	 * If the command has this flag set, it handles its own unit
10447 	 * attention reporting, we shouldn't do anything.  Otherwise we
10448 	 * check for any pending unit attentions, and send them back to the
10449 	 * initiator.  We only do this when a command initially comes in,
10450 	 * not when we pull it off the blocked queue.
10451 	 *
10452 	 * According to SAM-3, section 5.3.2, the order that things get
10453 	 * presented back to the host is basically unit attentions caused
10454 	 * by some sort of reset event, busy status, reservation conflicts
10455 	 * or task set full, and finally any other status.
10456 	 *
10457 	 * One issue here is that some of the unit attentions we report
10458 	 * don't fall into the "reset" category (e.g. "reported luns data
10459 	 * has changed").  So reporting it here, before the reservation
10460 	 * check, may be technically wrong.  I guess the only thing to do
10461 	 * would be to check for and report the reset events here, and then
10462 	 * check for the other unit attention types after we check for a
10463 	 * reservation conflict.
10464 	 *
10465 	 * XXX KDM need to fix this
10466 	 */
10467 	if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
10468 		ctl_ua_type ua_type;
10469 
10470 		ua_type = lun->pending_sense[initidx].ua_pending;
10471 		if (ua_type != CTL_UA_NONE) {
10472 			scsi_sense_data_type sense_format;
10473 
10474 			if (lun != NULL)
10475 				sense_format = (lun->flags &
10476 				    CTL_LUN_SENSE_DESC) ? SSD_TYPE_DESC :
10477 				    SSD_TYPE_FIXED;
10478 			else
10479 				sense_format = SSD_TYPE_FIXED;
10480 
10481 			ua_type = ctl_build_ua(ua_type, &ctsio->sense_data,
10482 					       sense_format);
10483 			if (ua_type != CTL_UA_NONE) {
10484 				ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
10485 				ctsio->io_hdr.status = CTL_SCSI_ERROR |
10486 						       CTL_AUTOSENSE;
10487 				ctsio->sense_len = SSD_FULL_SIZE;
10488 				lun->pending_sense[initidx].ua_pending &=
10489 					~ua_type;
10490 				mtx_unlock(&ctl_softc->ctl_lock);
10491 				ctl_done((union ctl_io *)ctsio);
10492 				goto bailout;
10493 			}
10494 		}
10495 	}
10496 
10497 
10498 	if (ctl_scsiio_lun_check(ctl_softc, lun, entry, ctsio) != 0) {
10499 		mtx_unlock(&ctl_softc->ctl_lock);
10500 		ctl_done((union ctl_io *)ctsio);
10501 		goto bailout;
10502 	}
10503 
10504 	/*
10505 	 * XXX CHD this is where we want to send IO to other side if
10506 	 * this LUN is secondary on this SC. We will need to make a copy
10507 	 * of the IO and flag the IO on this side as SENT_2OTHER and the flag
10508 	 * the copy we send as FROM_OTHER.
10509 	 * We also need to stuff the address of the original IO so we can
10510 	 * find it easily. Something similar will need be done on the other
10511 	 * side so when we are done we can find the copy.
10512 	 */
10513 	if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
10514 		union ctl_ha_msg msg_info;
10515 		int isc_retval;
10516 
10517 		ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
10518 
10519 		msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
10520 		msg_info.hdr.original_sc = (union ctl_io *)ctsio;
10521 #if 0
10522 		printf("1. ctsio %p\n", ctsio);
10523 #endif
10524 		msg_info.hdr.serializing_sc = NULL;
10525 		msg_info.hdr.nexus = ctsio->io_hdr.nexus;
10526 		msg_info.scsi.tag_num = ctsio->tag_num;
10527 		msg_info.scsi.tag_type = ctsio->tag_type;
10528 		memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
10529 
10530 		ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
10531 
10532 		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
10533 		    (void *)&msg_info, sizeof(msg_info), 0)) >
10534 		    CTL_HA_STATUS_SUCCESS) {
10535 			printf("CTL:precheck, ctl_ha_msg_send returned %d\n",
10536 			       isc_retval);
10537 			printf("CTL:opcode is %x\n",opcode);
10538 		} else {
10539 #if 0
10540 			printf("CTL:Precheck sent msg, opcode is %x\n",opcode);
10541 #endif
10542 		}
10543 
10544 		/*
10545 		 * XXX KDM this I/O is off the incoming queue, but hasn't
10546 		 * been inserted on any other queue.  We may need to come
10547 		 * up with a holding queue while we wait for serialization
10548 		 * so that we have an idea of what we're waiting for from
10549 		 * the other side.
10550 		 */
10551 		goto bailout_unlock;
10552 	}
10553 
10554 	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
10555 			      (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
10556 			      ctl_ooaq, ooa_links))) {
10557 	case CTL_ACTION_BLOCK:
10558 		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
10559 		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
10560 				  blocked_links);
10561 		goto bailout_unlock;
10562 		break; /* NOTREACHED */
10563 	case CTL_ACTION_PASS:
10564 	case CTL_ACTION_SKIP:
10565 		goto queue_rtr;
10566 		break; /* NOTREACHED */
10567 	case CTL_ACTION_OVERLAP:
10568 		ctl_set_overlapped_cmd(ctsio);
10569 		mtx_unlock(&ctl_softc->ctl_lock);
10570 		ctl_done((union ctl_io *)ctsio);
10571 		goto bailout;
10572 		break; /* NOTREACHED */
10573 	case CTL_ACTION_OVERLAP_TAG:
10574 		ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
10575 		mtx_unlock(&ctl_softc->ctl_lock);
10576 		ctl_done((union ctl_io *)ctsio);
10577 		goto bailout;
10578 		break; /* NOTREACHED */
10579 	case CTL_ACTION_ERROR:
10580 	default:
10581 		ctl_set_internal_failure(ctsio,
10582 					 /*sks_valid*/ 0,
10583 					 /*retry_count*/ 0);
10584 		mtx_unlock(&ctl_softc->ctl_lock);
10585 		ctl_done((union ctl_io *)ctsio);
10586 		goto bailout;
10587 		break; /* NOTREACHED */
10588 	}
10589 
10590 	goto bailout_unlock;
10591 
10592 queue_rtr:
10593 	ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
10594 	STAILQ_INSERT_TAIL(&ctl_softc->rtr_queue, &ctsio->io_hdr, links);
10595 
10596 bailout_unlock:
10597 	mtx_unlock(&ctl_softc->ctl_lock);
10598 
10599 bailout:
10600 	return (retval);
10601 }
10602 
10603 static int
10604 ctl_scsiio(struct ctl_scsiio *ctsio)
10605 {
10606 	int retval;
10607 	struct ctl_cmd_entry *entry;
10608 
10609 	retval = CTL_RETVAL_COMPLETE;
10610 
10611 	CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
10612 
10613 	entry = &ctl_cmd_table[ctsio->cdb[0]];
10614 
10615 	/*
10616 	 * If this I/O has been aborted, just send it straight to
10617 	 * ctl_done() without executing it.
10618 	 */
10619 	if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
10620 		ctl_done((union ctl_io *)ctsio);
10621 		goto bailout;
10622 	}
10623 
10624 	/*
10625 	 * All the checks should have been handled by ctl_scsiio_precheck().
10626 	 * We should be clear now to just execute the I/O.
10627 	 */
10628 	retval = entry->execute(ctsio);
10629 
10630 bailout:
10631 	return (retval);
10632 }
10633 
10634 /*
10635  * Since we only implement one target right now, a bus reset simply resets
10636  * our single target.
10637  */
10638 static int
10639 ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io)
10640 {
10641 	return(ctl_target_reset(ctl_softc, io, CTL_UA_BUS_RESET));
10642 }
10643 
10644 static int
10645 ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
10646 		 ctl_ua_type ua_type)
10647 {
10648 	struct ctl_lun *lun;
10649 	int retval;
10650 
10651 	if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
10652 		union ctl_ha_msg msg_info;
10653 
10654 		io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
10655 		msg_info.hdr.nexus = io->io_hdr.nexus;
10656 		if (ua_type==CTL_UA_TARG_RESET)
10657 			msg_info.task.task_action = CTL_TASK_TARGET_RESET;
10658 		else
10659 			msg_info.task.task_action = CTL_TASK_BUS_RESET;
10660 		msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
10661 		msg_info.hdr.original_sc = NULL;
10662 		msg_info.hdr.serializing_sc = NULL;
10663 		if (CTL_HA_STATUS_SUCCESS != ctl_ha_msg_send(CTL_HA_CHAN_CTL,
10664 		    (void *)&msg_info, sizeof(msg_info), 0)) {
10665 		}
10666 	}
10667 	retval = 0;
10668 
10669 	STAILQ_FOREACH(lun, &ctl_softc->lun_list, links)
10670 		retval += ctl_lun_reset(lun, io, ua_type);
10671 
10672 	return (retval);
10673 }
10674 
10675 /*
10676  * The LUN should always be set.  The I/O is optional, and is used to
10677  * distinguish between I/Os sent by this initiator, and by other
10678  * initiators.  We set unit attention for initiators other than this one.
10679  * SAM-3 is vague on this point.  It does say that a unit attention should
10680  * be established for other initiators when a LUN is reset (see section
10681  * 5.7.3), but it doesn't specifically say that the unit attention should
10682  * be established for this particular initiator when a LUN is reset.  Here
10683  * is the relevant text, from SAM-3 rev 8:
10684  *
10685  * 5.7.2 When a SCSI initiator port aborts its own tasks
10686  *
10687  * When a SCSI initiator port causes its own task(s) to be aborted, no
10688  * notification that the task(s) have been aborted shall be returned to
10689  * the SCSI initiator port other than the completion response for the
10690  * command or task management function action that caused the task(s) to
10691  * be aborted and notification(s) associated with related effects of the
10692  * action (e.g., a reset unit attention condition).
10693  *
10694  * XXX KDM for now, we're setting unit attention for all initiators.
10695  */
10696 static int
10697 ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
10698 {
10699 	union ctl_io *xio;
10700 #if 0
10701 	uint32_t initindex;
10702 #endif
10703 	int i;
10704 
10705 	/*
10706 	 * Run through the OOA queue and abort each I/O.
10707 	 */
10708 #if 0
10709 	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
10710 #endif
10711 	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
10712 	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
10713 		xio->io_hdr.flags |= CTL_FLAG_ABORT;
10714 	}
10715 
10716 	/*
10717 	 * This version sets unit attention for every
10718 	 */
10719 #if 0
10720 	initindex = ctl_get_initindex(&io->io_hdr.nexus);
10721 	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
10722 		if (initindex == i)
10723 			continue;
10724 		lun->pending_sense[i].ua_pending |= ua_type;
10725 	}
10726 #endif
10727 
10728 	/*
10729 	 * A reset (any kind, really) clears reservations established with
10730 	 * RESERVE/RELEASE.  It does not clear reservations established
10731 	 * with PERSISTENT RESERVE OUT, but we don't support that at the
10732 	 * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
10733 	 * reservations made with the RESERVE/RELEASE commands, because
10734 	 * those commands are obsolete in SPC-3.
10735 	 */
10736 	lun->flags &= ~CTL_LUN_RESERVED;
10737 
10738 	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
10739 		ctl_clear_mask(lun->have_ca, i);
10740 		lun->pending_sense[i].ua_pending |= ua_type;
10741 	}
10742 
10743 	return (0);
10744 }
10745 
10746 static int
10747 ctl_abort_task(union ctl_io *io)
10748 {
10749 	union ctl_io *xio;
10750 	struct ctl_lun *lun;
10751 	struct ctl_softc *ctl_softc;
10752 #if 0
10753 	struct sbuf sb;
10754 	char printbuf[128];
10755 #endif
10756 	int found;
10757 
10758 	ctl_softc = control_softc;
10759 	found = 0;
10760 
10761 	/*
10762 	 * Look up the LUN.
10763 	 */
10764 	if ((io->io_hdr.nexus.targ_lun < CTL_MAX_LUNS)
10765 	 && (ctl_softc->ctl_luns[io->io_hdr.nexus.targ_lun] != NULL))
10766 		lun = ctl_softc->ctl_luns[io->io_hdr.nexus.targ_lun];
10767 	else
10768 		goto bailout;
10769 
10770 #if 0
10771 	printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
10772 	       lun->lun, io->taskio.tag_num, io->taskio.tag_type);
10773 #endif
10774 
10775 	/*
10776 	 * Run through the OOA queue and attempt to find the given I/O.
10777 	 * The target port, initiator ID, tag type and tag number have to
10778 	 * match the values that we got from the initiator.  If we have an
10779 	 * untagged command to abort, simply abort the first untagged command
10780 	 * we come to.  We only allow one untagged command at a time of course.
10781 	 */
10782 #if 0
10783 	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
10784 #endif
10785 	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
10786 	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
10787 #if 0
10788 		sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
10789 
10790 		sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
10791 			    lun->lun, xio->scsiio.tag_num,
10792 			    xio->scsiio.tag_type,
10793 			    (xio->io_hdr.blocked_links.tqe_prev
10794 			    == NULL) ? "" : " BLOCKED",
10795 			    (xio->io_hdr.flags &
10796 			    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
10797 			    (xio->io_hdr.flags &
10798 			    CTL_FLAG_ABORT) ? " ABORT" : "",
10799 			    (xio->io_hdr.flags &
10800 			    CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
10801 		ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
10802 		sbuf_finish(&sb);
10803 		printf("%s\n", sbuf_data(&sb));
10804 #endif
10805 
10806 		if ((xio->io_hdr.nexus.targ_port == io->io_hdr.nexus.targ_port)
10807 		 && (xio->io_hdr.nexus.initid.id ==
10808 		     io->io_hdr.nexus.initid.id)) {
10809 			/*
10810 			 * If the abort says that the task is untagged, the
10811 			 * task in the queue must be untagged.  Otherwise,
10812 			 * we just check to see whether the tag numbers
10813 			 * match.  This is because the QLogic firmware
10814 			 * doesn't pass back the tag type in an abort
10815 			 * request.
10816 			 */
10817 #if 0
10818 			if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
10819 			  && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
10820 			 || (xio->scsiio.tag_num == io->taskio.tag_num)) {
10821 #endif
10822 			/*
10823 			 * XXX KDM we've got problems with FC, because it
10824 			 * doesn't send down a tag type with aborts.  So we
10825 			 * can only really go by the tag number...
10826 			 * This may cause problems with parallel SCSI.
10827 			 * Need to figure that out!!
10828 			 */
10829 			if (xio->scsiio.tag_num == io->taskio.tag_num) {
10830 				xio->io_hdr.flags |= CTL_FLAG_ABORT;
10831 				found = 1;
10832 				if ((io->io_hdr.flags &
10833 				     CTL_FLAG_FROM_OTHER_SC) == 0 &&
10834 				    !(lun->flags & CTL_LUN_PRIMARY_SC)) {
10835 					union ctl_ha_msg msg_info;
10836 
10837 					io->io_hdr.flags |=
10838 					                CTL_FLAG_SENT_2OTHER_SC;
10839 					msg_info.hdr.nexus = io->io_hdr.nexus;
10840 					msg_info.task.task_action =
10841 						CTL_TASK_ABORT_TASK;
10842 					msg_info.task.tag_num =
10843 						io->taskio.tag_num;
10844 					msg_info.task.tag_type =
10845 						io->taskio.tag_type;
10846 					msg_info.hdr.msg_type =
10847 						CTL_MSG_MANAGE_TASKS;
10848 					msg_info.hdr.original_sc = NULL;
10849 					msg_info.hdr.serializing_sc = NULL;
10850 #if 0
10851 					printf("Sent Abort to other side\n");
10852 #endif
10853 					if (CTL_HA_STATUS_SUCCESS !=
10854 					        ctl_ha_msg_send(CTL_HA_CHAN_CTL,
10855 		    				(void *)&msg_info,
10856 						sizeof(msg_info), 0)) {
10857 					}
10858 				}
10859 #if 0
10860 				printf("ctl_abort_task: found I/O to abort\n");
10861 #endif
10862 				break;
10863 			}
10864 		}
10865 	}
10866 
10867 bailout:
10868 
10869 	if (found == 0) {
10870 		/*
10871 		 * This isn't really an error.  It's entirely possible for
10872 		 * the abort and command completion to cross on the wire.
10873 		 * This is more of an informative/diagnostic error.
10874 		 */
10875 #if 0
10876 		printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
10877 		       "%d:%d:%d:%d tag %d type %d\n",
10878 		       io->io_hdr.nexus.initid.id,
10879 		       io->io_hdr.nexus.targ_port,
10880 		       io->io_hdr.nexus.targ_target.id,
10881 		       io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
10882 		       io->taskio.tag_type);
10883 #endif
10884 		return (1);
10885 	} else
10886 		return (0);
10887 }
10888 
10889 /*
10890  * Assumptions:  caller holds ctl_softc->ctl_lock
10891  *
10892  * This routine cannot block!  It must be callable from an interrupt
10893  * handler as well as from the work thread.
10894  */
10895 static void
10896 ctl_run_task_queue(struct ctl_softc *ctl_softc)
10897 {
10898 	union ctl_io *io, *next_io;
10899 
10900 	CTL_DEBUG_PRINT(("ctl_run_task_queue\n"));
10901 
10902 	for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->task_queue);
10903 	     io != NULL; io = next_io) {
10904 		int retval;
10905 		const char *task_desc;
10906 
10907 		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
10908 
10909 		retval = 0;
10910 
10911 		switch (io->io_hdr.io_type) {
10912 		case CTL_IO_TASK: {
10913 			task_desc = ctl_scsi_task_string(&io->taskio);
10914 			if (task_desc != NULL) {
10915 #ifdef NEEDTOPORT
10916 				csevent_log(CSC_CTL | CSC_SHELF_SW |
10917 					    CTL_TASK_REPORT,
10918 					    csevent_LogType_Trace,
10919 					    csevent_Severity_Information,
10920 					    csevent_AlertLevel_Green,
10921 					    csevent_FRU_Firmware,
10922 					    csevent_FRU_Unknown,
10923 					    "CTL: received task: %s",task_desc);
10924 #endif
10925 			} else {
10926 #ifdef NEEDTOPORT
10927 				csevent_log(CSC_CTL | CSC_SHELF_SW |
10928 					    CTL_TASK_REPORT,
10929 					    csevent_LogType_Trace,
10930 					    csevent_Severity_Information,
10931 					    csevent_AlertLevel_Green,
10932 					    csevent_FRU_Firmware,
10933 					    csevent_FRU_Unknown,
10934 					    "CTL: received unknown task "
10935 					    "type: %d (%#x)",
10936 					    io->taskio.task_action,
10937 					    io->taskio.task_action);
10938 #endif
10939 			}
10940 			switch (io->taskio.task_action) {
10941 			case CTL_TASK_ABORT_TASK:
10942 				retval = ctl_abort_task(io);
10943 				break;
10944 			case CTL_TASK_ABORT_TASK_SET:
10945 				break;
10946 			case CTL_TASK_CLEAR_ACA:
10947 				break;
10948 			case CTL_TASK_CLEAR_TASK_SET:
10949 				break;
10950 			case CTL_TASK_LUN_RESET: {
10951 				struct ctl_lun *lun;
10952 				uint32_t targ_lun;
10953 				int retval;
10954 
10955 				targ_lun = io->io_hdr.nexus.targ_lun;
10956 
10957 				if ((targ_lun < CTL_MAX_LUNS)
10958 				 && (ctl_softc->ctl_luns[targ_lun] != NULL))
10959 					lun = ctl_softc->ctl_luns[targ_lun];
10960 				else {
10961 					retval = 1;
10962 					break;
10963 				}
10964 
10965 				if (!(io->io_hdr.flags &
10966 				    CTL_FLAG_FROM_OTHER_SC)) {
10967 					union ctl_ha_msg msg_info;
10968 
10969 					io->io_hdr.flags |=
10970 						CTL_FLAG_SENT_2OTHER_SC;
10971 					msg_info.hdr.msg_type =
10972 						CTL_MSG_MANAGE_TASKS;
10973 					msg_info.hdr.nexus = io->io_hdr.nexus;
10974 					msg_info.task.task_action =
10975 						CTL_TASK_LUN_RESET;
10976 					msg_info.hdr.original_sc = NULL;
10977 					msg_info.hdr.serializing_sc = NULL;
10978 					if (CTL_HA_STATUS_SUCCESS !=
10979 					    ctl_ha_msg_send(CTL_HA_CHAN_CTL,
10980 					    (void *)&msg_info,
10981 					    sizeof(msg_info), 0)) {
10982 					}
10983 				}
10984 
10985 				retval = ctl_lun_reset(lun, io,
10986 						       CTL_UA_LUN_RESET);
10987 				break;
10988 			}
10989 			case CTL_TASK_TARGET_RESET:
10990 				retval = ctl_target_reset(ctl_softc, io,
10991 							  CTL_UA_TARG_RESET);
10992 				break;
10993 			case CTL_TASK_BUS_RESET:
10994 				retval = ctl_bus_reset(ctl_softc, io);
10995 				break;
10996 			case CTL_TASK_PORT_LOGIN:
10997 				break;
10998 			case CTL_TASK_PORT_LOGOUT:
10999 				break;
11000 			default:
11001 				printf("ctl_run_task_queue: got unknown task "
11002 				       "management event %d\n",
11003 				       io->taskio.task_action);
11004 				break;
11005 			}
11006 			if (retval == 0)
11007 				io->io_hdr.status = CTL_SUCCESS;
11008 			else
11009 				io->io_hdr.status = CTL_ERROR;
11010 
11011 			STAILQ_REMOVE(&ctl_softc->task_queue, &io->io_hdr,
11012 				      ctl_io_hdr, links);
11013 			/*
11014 			 * This will queue this I/O to the done queue, but the
11015 			 * work thread won't be able to process it until we
11016 			 * return and the lock is released.
11017 			 */
11018 			ctl_done_lock(io, /*have_lock*/ 1);
11019 			break;
11020 		}
11021 		default: {
11022 
11023 			printf("%s: invalid I/O type %d msg %d cdb %x"
11024 			       " iptl: %ju:%d:%ju:%d tag 0x%04x\n",
11025 			       __func__, io->io_hdr.io_type,
11026 			       io->io_hdr.msg_type, io->scsiio.cdb[0],
11027 			       (uintmax_t)io->io_hdr.nexus.initid.id,
11028 			       io->io_hdr.nexus.targ_port,
11029 			       (uintmax_t)io->io_hdr.nexus.targ_target.id,
11030 			       io->io_hdr.nexus.targ_lun,
11031 			       (io->io_hdr.io_type == CTL_IO_TASK) ?
11032 			       io->taskio.tag_num : io->scsiio.tag_num);
11033 			STAILQ_REMOVE(&ctl_softc->task_queue, &io->io_hdr,
11034 				      ctl_io_hdr, links);
11035 			ctl_free_io_internal(io, 1);
11036 			break;
11037 		}
11038 		}
11039 	}
11040 
11041 	ctl_softc->flags &= ~CTL_FLAG_TASK_PENDING;
11042 }
11043 
11044 /*
11045  * For HA operation.  Handle commands that come in from the other
11046  * controller.
11047  */
11048 static void
11049 ctl_handle_isc(union ctl_io *io)
11050 {
11051 	int free_io;
11052 	struct ctl_lun *lun;
11053 	struct ctl_softc *ctl_softc;
11054 
11055 	ctl_softc = control_softc;
11056 
11057 	lun = ctl_softc->ctl_luns[io->io_hdr.nexus.targ_lun];
11058 
11059 	switch (io->io_hdr.msg_type) {
11060 	case CTL_MSG_SERIALIZE:
11061 		free_io = ctl_serialize_other_sc_cmd(&io->scsiio,
11062 						     /*have_lock*/ 0);
11063 		break;
11064 	case CTL_MSG_R2R: {
11065 		uint8_t opcode;
11066 		struct ctl_cmd_entry *entry;
11067 
11068 		/*
11069 		 * This is only used in SER_ONLY mode.
11070 		 */
11071 		free_io = 0;
11072 		opcode = io->scsiio.cdb[0];
11073 		entry = &ctl_cmd_table[opcode];
11074 		mtx_lock(&ctl_softc->ctl_lock);
11075 		if (ctl_scsiio_lun_check(ctl_softc, lun,
11076 		    entry, (struct ctl_scsiio *)io) != 0) {
11077 			ctl_done_lock(io, /*have_lock*/ 1);
11078 			mtx_unlock(&ctl_softc->ctl_lock);
11079 			break;
11080 		}
11081 		io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11082 		STAILQ_INSERT_TAIL(&ctl_softc->rtr_queue,
11083 				   &io->io_hdr, links);
11084 		mtx_unlock(&ctl_softc->ctl_lock);
11085 		break;
11086 	}
11087 	case CTL_MSG_FINISH_IO:
11088 		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
11089 			free_io = 0;
11090 			ctl_done_lock(io, /*have_lock*/ 0);
11091 		} else {
11092 			free_io = 1;
11093 			mtx_lock(&ctl_softc->ctl_lock);
11094 			TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
11095 				     ooa_links);
11096 			STAILQ_REMOVE(&ctl_softc->task_queue,
11097 				      &io->io_hdr, ctl_io_hdr, links);
11098 			ctl_check_blocked(lun);
11099 			mtx_unlock(&ctl_softc->ctl_lock);
11100 		}
11101 		break;
11102 	case CTL_MSG_PERS_ACTION:
11103 		ctl_hndl_per_res_out_on_other_sc(
11104 			(union ctl_ha_msg *)&io->presio.pr_msg);
11105 		free_io = 1;
11106 		break;
11107 	case CTL_MSG_BAD_JUJU:
11108 		free_io = 0;
11109 		ctl_done_lock(io, /*have_lock*/ 0);
11110 		break;
11111 	case CTL_MSG_DATAMOVE:
11112 		/* Only used in XFER mode */
11113 		free_io = 0;
11114 		ctl_datamove_remote(io);
11115 		break;
11116 	case CTL_MSG_DATAMOVE_DONE:
11117 		/* Only used in XFER mode */
11118 		free_io = 0;
11119 		io->scsiio.be_move_done(io);
11120 		break;
11121 	default:
11122 		free_io = 1;
11123 		printf("%s: Invalid message type %d\n",
11124 		       __func__, io->io_hdr.msg_type);
11125 		break;
11126 	}
11127 	if (free_io)
11128 		ctl_free_io_internal(io, 0);
11129 
11130 }
11131 
11132 
11133 /*
11134  * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
11135  * there is no match.
11136  */
11137 static ctl_lun_error_pattern
11138 ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
11139 {
11140 	struct ctl_cmd_entry *entry;
11141 	ctl_lun_error_pattern filtered_pattern, pattern;
11142 	uint8_t opcode;
11143 
11144 	pattern = desc->error_pattern;
11145 
11146 	/*
11147 	 * XXX KDM we need more data passed into this function to match a
11148 	 * custom pattern, and we actually need to implement custom pattern
11149 	 * matching.
11150 	 */
11151 	if (pattern & CTL_LUN_PAT_CMD)
11152 		return (CTL_LUN_PAT_CMD);
11153 
11154 	if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
11155 		return (CTL_LUN_PAT_ANY);
11156 
11157 	opcode = ctsio->cdb[0];
11158 	entry = &ctl_cmd_table[opcode];
11159 
11160 	filtered_pattern = entry->pattern & pattern;
11161 
11162 	/*
11163 	 * If the user requested specific flags in the pattern (e.g.
11164 	 * CTL_LUN_PAT_RANGE), make sure the command supports all of those
11165 	 * flags.
11166 	 *
11167 	 * If the user did not specify any flags, it doesn't matter whether
11168 	 * or not the command supports the flags.
11169 	 */
11170 	if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
11171 	     (pattern & ~CTL_LUN_PAT_MASK))
11172 		return (CTL_LUN_PAT_NONE);
11173 
11174 	/*
11175 	 * If the user asked for a range check, see if the requested LBA
11176 	 * range overlaps with this command's LBA range.
11177 	 */
11178 	if (filtered_pattern & CTL_LUN_PAT_RANGE) {
11179 		uint64_t lba1;
11180 		uint32_t len1;
11181 		ctl_action action;
11182 		int retval;
11183 
11184 		retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
11185 		if (retval != 0)
11186 			return (CTL_LUN_PAT_NONE);
11187 
11188 		action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
11189 					      desc->lba_range.len);
11190 		/*
11191 		 * A "pass" means that the LBA ranges don't overlap, so
11192 		 * this doesn't match the user's range criteria.
11193 		 */
11194 		if (action == CTL_ACTION_PASS)
11195 			return (CTL_LUN_PAT_NONE);
11196 	}
11197 
11198 	return (filtered_pattern);
11199 }
11200 
11201 /*
11202  * Called with the CTL lock held.
11203  */
11204 static void
11205 ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
11206 {
11207 	struct ctl_error_desc *desc, *desc2;
11208 
11209 	STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
11210 		ctl_lun_error_pattern pattern;
11211 		/*
11212 		 * Check to see whether this particular command matches
11213 		 * the pattern in the descriptor.
11214 		 */
11215 		pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
11216 		if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
11217 			continue;
11218 
11219 		switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
11220 		case CTL_LUN_INJ_ABORTED:
11221 			ctl_set_aborted(&io->scsiio);
11222 			break;
11223 		case CTL_LUN_INJ_MEDIUM_ERR:
11224 			ctl_set_medium_error(&io->scsiio);
11225 			break;
11226 		case CTL_LUN_INJ_UA:
11227 			/* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
11228 			 * OCCURRED */
11229 			ctl_set_ua(&io->scsiio, 0x29, 0x00);
11230 			break;
11231 		case CTL_LUN_INJ_CUSTOM:
11232 			/*
11233 			 * We're assuming the user knows what he is doing.
11234 			 * Just copy the sense information without doing
11235 			 * checks.
11236 			 */
11237 			bcopy(&desc->custom_sense, &io->scsiio.sense_data,
11238 			      ctl_min(sizeof(desc->custom_sense),
11239 				      sizeof(io->scsiio.sense_data)));
11240 			io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
11241 			io->scsiio.sense_len = SSD_FULL_SIZE;
11242 			io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
11243 			break;
11244 		case CTL_LUN_INJ_NONE:
11245 		default:
11246 			/*
11247 			 * If this is an error injection type we don't know
11248 			 * about, clear the continuous flag (if it is set)
11249 			 * so it will get deleted below.
11250 			 */
11251 			desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
11252 			break;
11253 		}
11254 		/*
11255 		 * By default, each error injection action is a one-shot
11256 		 */
11257 		if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
11258 			continue;
11259 
11260 		STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
11261 
11262 		free(desc, M_CTL);
11263 	}
11264 }
11265 
11266 #ifdef CTL_IO_DELAY
11267 static void
11268 ctl_datamove_timer_wakeup(void *arg)
11269 {
11270 	union ctl_io *io;
11271 
11272 	io = (union ctl_io *)arg;
11273 
11274 	ctl_datamove(io);
11275 }
11276 #endif /* CTL_IO_DELAY */
11277 
11278 /*
11279  * Assumption:  caller does NOT hold ctl_lock
11280  */
11281 void
11282 ctl_datamove(union ctl_io *io)
11283 {
11284 	void (*fe_datamove)(union ctl_io *io);
11285 
11286 	CTL_DEBUG_PRINT(("ctl_datamove\n"));
11287 
11288 #ifdef CTL_TIME_IO
11289 	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
11290 		char str[256];
11291 		char path_str[64];
11292 		struct sbuf sb;
11293 
11294 		ctl_scsi_path_string(io, path_str, sizeof(path_str));
11295 		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
11296 
11297 		sbuf_cat(&sb, path_str);
11298 		switch (io->io_hdr.io_type) {
11299 		case CTL_IO_SCSI:
11300 			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
11301 			sbuf_printf(&sb, "\n");
11302 			sbuf_cat(&sb, path_str);
11303 			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
11304 				    io->scsiio.tag_num, io->scsiio.tag_type);
11305 			break;
11306 		case CTL_IO_TASK:
11307 			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
11308 				    "Tag Type: %d\n", io->taskio.task_action,
11309 				    io->taskio.tag_num, io->taskio.tag_type);
11310 			break;
11311 		default:
11312 			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
11313 			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
11314 			break;
11315 		}
11316 		sbuf_cat(&sb, path_str);
11317 		sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
11318 			    (intmax_t)time_uptime - io->io_hdr.start_time);
11319 		sbuf_finish(&sb);
11320 		printf("%s", sbuf_data(&sb));
11321 	}
11322 #endif /* CTL_TIME_IO */
11323 
11324 	mtx_lock(&control_softc->ctl_lock);
11325 #ifdef CTL_IO_DELAY
11326 	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
11327 		struct ctl_lun *lun;
11328 
11329 		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
11330 
11331 		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
11332 	} else {
11333 		struct ctl_lun *lun;
11334 
11335 		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
11336 		if ((lun != NULL)
11337 		 && (lun->delay_info.datamove_delay > 0)) {
11338 			struct callout *callout;
11339 
11340 			callout = (struct callout *)&io->io_hdr.timer_bytes;
11341 			callout_init(callout, /*mpsafe*/ 1);
11342 			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
11343 			callout_reset(callout,
11344 				      lun->delay_info.datamove_delay * hz,
11345 				      ctl_datamove_timer_wakeup, io);
11346 			if (lun->delay_info.datamove_type ==
11347 			    CTL_DELAY_TYPE_ONESHOT)
11348 				lun->delay_info.datamove_delay = 0;
11349 			mtx_unlock(&control_softc->ctl_lock);
11350 			return;
11351 		}
11352 	}
11353 #endif
11354 	/*
11355 	 * If we have any pending task management commands, process them
11356 	 * first.  This is necessary to eliminate a race condition with the
11357 	 * FETD:
11358 	 *
11359 	 * - FETD submits a task management command, like an abort.
11360 	 * - Back end calls fe_datamove() to move the data for the aborted
11361 	 *   command.  The FETD can't really accept it, but if it did, it
11362 	 *   would end up transmitting data for a command that the initiator
11363 	 *   told us to abort.
11364 	 *
11365 	 * We close the race by processing all pending task management
11366 	 * commands here (we can't block!), and then check this I/O to see
11367 	 * if it has been aborted.  If so, return it to the back end with
11368 	 * bad status, so the back end can say return an error to the back end
11369 	 * and then when the back end returns an error, we can return the
11370 	 * aborted command to the FETD, so it can clean up its resources.
11371 	 */
11372 	if (control_softc->flags & CTL_FLAG_TASK_PENDING)
11373 		ctl_run_task_queue(control_softc);
11374 
11375 	/*
11376 	 * This command has been aborted.  Set the port status, so we fail
11377 	 * the data move.
11378 	 */
11379 	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
11380 		printf("ctl_datamove: tag 0x%04x on (%ju:%d:%ju:%d) aborted\n",
11381 		       io->scsiio.tag_num,(uintmax_t)io->io_hdr.nexus.initid.id,
11382 		       io->io_hdr.nexus.targ_port,
11383 		       (uintmax_t)io->io_hdr.nexus.targ_target.id,
11384 		       io->io_hdr.nexus.targ_lun);
11385 		io->io_hdr.status = CTL_CMD_ABORTED;
11386 		io->io_hdr.port_status = 31337;
11387 		mtx_unlock(&control_softc->ctl_lock);
11388 		/*
11389 		 * Note that the backend, in this case, will get the
11390 		 * callback in its context.  In other cases it may get
11391 		 * called in the frontend's interrupt thread context.
11392 		 */
11393 		io->scsiio.be_move_done(io);
11394 		return;
11395 	}
11396 
11397 	/*
11398 	 * If we're in XFER mode and this I/O is from the other shelf
11399 	 * controller, we need to send the DMA to the other side to
11400 	 * actually transfer the data to/from the host.  In serialize only
11401 	 * mode the transfer happens below CTL and ctl_datamove() is only
11402 	 * called on the machine that originally received the I/O.
11403 	 */
11404 	if ((control_softc->ha_mode == CTL_HA_MODE_XFER)
11405 	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11406 		union ctl_ha_msg msg;
11407 		uint32_t sg_entries_sent;
11408 		int do_sg_copy;
11409 		int i;
11410 
11411 		memset(&msg, 0, sizeof(msg));
11412 		msg.hdr.msg_type = CTL_MSG_DATAMOVE;
11413 		msg.hdr.original_sc = io->io_hdr.original_sc;
11414 		msg.hdr.serializing_sc = io;
11415 		msg.hdr.nexus = io->io_hdr.nexus;
11416 		msg.dt.flags = io->io_hdr.flags;
11417 		/*
11418 		 * We convert everything into a S/G list here.  We can't
11419 		 * pass by reference, only by value between controllers.
11420 		 * So we can't pass a pointer to the S/G list, only as many
11421 		 * S/G entries as we can fit in here.  If it's possible for
11422 		 * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
11423 		 * then we need to break this up into multiple transfers.
11424 		 */
11425 		if (io->scsiio.kern_sg_entries == 0) {
11426 			msg.dt.kern_sg_entries = 1;
11427 			/*
11428 			 * If this is in cached memory, flush the cache
11429 			 * before we send the DMA request to the other
11430 			 * controller.  We want to do this in either the
11431 			 * read or the write case.  The read case is
11432 			 * straightforward.  In the write case, we want to
11433 			 * make sure nothing is in the local cache that
11434 			 * could overwrite the DMAed data.
11435 			 */
11436 			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
11437 				/*
11438 				 * XXX KDM use bus_dmamap_sync() here.
11439 				 */
11440 			}
11441 
11442 			/*
11443 			 * Convert to a physical address if this is a
11444 			 * virtual address.
11445 			 */
11446 			if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
11447 				msg.dt.sg_list[0].addr =
11448 					io->scsiio.kern_data_ptr;
11449 			} else {
11450 				/*
11451 				 * XXX KDM use busdma here!
11452 				 */
11453 #if 0
11454 				msg.dt.sg_list[0].addr = (void *)
11455 					vtophys(io->scsiio.kern_data_ptr);
11456 #endif
11457 			}
11458 
11459 			msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
11460 			do_sg_copy = 0;
11461 		} else {
11462 			struct ctl_sg_entry *sgl;
11463 
11464 			do_sg_copy = 1;
11465 			msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
11466 			sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
11467 			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
11468 				/*
11469 				 * XXX KDM use bus_dmamap_sync() here.
11470 				 */
11471 			}
11472 		}
11473 
11474 		msg.dt.kern_data_len = io->scsiio.kern_data_len;
11475 		msg.dt.kern_total_len = io->scsiio.kern_total_len;
11476 		msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
11477 		msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
11478 		msg.dt.sg_sequence = 0;
11479 
11480 		/*
11481 		 * Loop until we've sent all of the S/G entries.  On the
11482 		 * other end, we'll recompose these S/G entries into one
11483 		 * contiguous list before passing it to the
11484 		 */
11485 		for (sg_entries_sent = 0; sg_entries_sent <
11486 		     msg.dt.kern_sg_entries; msg.dt.sg_sequence++) {
11487 			msg.dt.cur_sg_entries = ctl_min((sizeof(msg.dt.sg_list)/
11488 				sizeof(msg.dt.sg_list[0])),
11489 				msg.dt.kern_sg_entries - sg_entries_sent);
11490 
11491 			if (do_sg_copy != 0) {
11492 				struct ctl_sg_entry *sgl;
11493 				int j;
11494 
11495 				sgl = (struct ctl_sg_entry *)
11496 					io->scsiio.kern_data_ptr;
11497 				/*
11498 				 * If this is in cached memory, flush the cache
11499 				 * before we send the DMA request to the other
11500 				 * controller.  We want to do this in either
11501 				 * the * read or the write case.  The read
11502 				 * case is straightforward.  In the write
11503 				 * case, we want to make sure nothing is
11504 				 * in the local cache that could overwrite
11505 				 * the DMAed data.
11506 				 */
11507 
11508 				for (i = sg_entries_sent, j = 0;
11509 				     i < msg.dt.cur_sg_entries; i++, j++) {
11510 					if ((io->io_hdr.flags &
11511 					     CTL_FLAG_NO_DATASYNC) == 0) {
11512 						/*
11513 						 * XXX KDM use bus_dmamap_sync()
11514 						 */
11515 					}
11516 					if ((io->io_hdr.flags &
11517 					     CTL_FLAG_BUS_ADDR) == 0) {
11518 						/*
11519 						 * XXX KDM use busdma.
11520 						 */
11521 #if 0
11522 						msg.dt.sg_list[j].addr =(void *)
11523 						       vtophys(sgl[i].addr);
11524 #endif
11525 					} else {
11526 						msg.dt.sg_list[j].addr =
11527 							sgl[i].addr;
11528 					}
11529 					msg.dt.sg_list[j].len = sgl[i].len;
11530 				}
11531 			}
11532 
11533 			sg_entries_sent += msg.dt.cur_sg_entries;
11534 			if (sg_entries_sent >= msg.dt.kern_sg_entries)
11535 				msg.dt.sg_last = 1;
11536 			else
11537 				msg.dt.sg_last = 0;
11538 
11539 			/*
11540 			 * XXX KDM drop and reacquire the lock here?
11541 			 */
11542 			if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
11543 			    sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
11544 				/*
11545 				 * XXX do something here.
11546 				 */
11547 			}
11548 
11549 			msg.dt.sent_sg_entries = sg_entries_sent;
11550 		}
11551 		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11552 		if (io->io_hdr.flags & CTL_FLAG_FAILOVER)
11553 			ctl_failover_io(io, /*have_lock*/ 1);
11554 
11555 	} else {
11556 
11557 		/*
11558 		 * Lookup the fe_datamove() function for this particular
11559 		 * front end.
11560 		 */
11561 		fe_datamove =
11562 		    control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
11563 		mtx_unlock(&control_softc->ctl_lock);
11564 
11565 		fe_datamove(io);
11566 	}
11567 }
11568 
11569 static void
11570 ctl_send_datamove_done(union ctl_io *io, int have_lock)
11571 {
11572 	union ctl_ha_msg msg;
11573 	int isc_status;
11574 
11575 	memset(&msg, 0, sizeof(msg));
11576 
11577 	msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
11578 	msg.hdr.original_sc = io;
11579 	msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
11580 	msg.hdr.nexus = io->io_hdr.nexus;
11581 	msg.hdr.status = io->io_hdr.status;
11582 	msg.scsi.tag_num = io->scsiio.tag_num;
11583 	msg.scsi.tag_type = io->scsiio.tag_type;
11584 	msg.scsi.scsi_status = io->scsiio.scsi_status;
11585 	memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
11586 	       sizeof(io->scsiio.sense_data));
11587 	msg.scsi.sense_len = io->scsiio.sense_len;
11588 	msg.scsi.sense_residual = io->scsiio.sense_residual;
11589 	msg.scsi.fetd_status = io->io_hdr.port_status;
11590 	msg.scsi.residual = io->scsiio.residual;
11591 	io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11592 
11593 	if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
11594 		ctl_failover_io(io, /*have_lock*/ have_lock);
11595 		return;
11596 	}
11597 
11598 	isc_status = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0);
11599 	if (isc_status > CTL_HA_STATUS_SUCCESS) {
11600 		/* XXX do something if this fails */
11601 	}
11602 
11603 }
11604 
11605 /*
11606  * The DMA to the remote side is done, now we need to tell the other side
11607  * we're done so it can continue with its data movement.
11608  */
11609 static void
11610 ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
11611 {
11612 	union ctl_io *io;
11613 
11614 	io = rq->context;
11615 
11616 	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
11617 		printf("%s: ISC DMA write failed with error %d", __func__,
11618 		       rq->ret);
11619 		ctl_set_internal_failure(&io->scsiio,
11620 					 /*sks_valid*/ 1,
11621 					 /*retry_count*/ rq->ret);
11622 	}
11623 
11624 	ctl_dt_req_free(rq);
11625 
11626 	/*
11627 	 * In this case, we had to malloc the memory locally.  Free it.
11628 	 */
11629 	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
11630 		int i;
11631 		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
11632 			free(io->io_hdr.local_sglist[i].addr, M_CTL);
11633 	}
11634 	/*
11635 	 * The data is in local and remote memory, so now we need to send
11636 	 * status (good or back) back to the other side.
11637 	 */
11638 	ctl_send_datamove_done(io, /*have_lock*/ 0);
11639 }
11640 
11641 /*
11642  * We've moved the data from the host/controller into local memory.  Now we
11643  * need to push it over to the remote controller's memory.
11644  */
11645 static int
11646 ctl_datamove_remote_dm_write_cb(union ctl_io *io)
11647 {
11648 	int retval;
11649 
11650 	retval = 0;
11651 
11652 	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
11653 					  ctl_datamove_remote_write_cb);
11654 
11655 	return (retval);
11656 }
11657 
11658 static void
11659 ctl_datamove_remote_write(union ctl_io *io)
11660 {
11661 	int retval;
11662 	void (*fe_datamove)(union ctl_io *io);
11663 
11664 	/*
11665 	 * - Get the data from the host/HBA into local memory.
11666 	 * - DMA memory from the local controller to the remote controller.
11667 	 * - Send status back to the remote controller.
11668 	 */
11669 
11670 	retval = ctl_datamove_remote_sgl_setup(io);
11671 	if (retval != 0)
11672 		return;
11673 
11674 	/* Switch the pointer over so the FETD knows what to do */
11675 	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
11676 
11677 	/*
11678 	 * Use a custom move done callback, since we need to send completion
11679 	 * back to the other controller, not to the backend on this side.
11680 	 */
11681 	io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
11682 
11683 	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
11684 
11685 	fe_datamove(io);
11686 
11687 	return;
11688 
11689 }
11690 
11691 static int
11692 ctl_datamove_remote_dm_read_cb(union ctl_io *io)
11693 {
11694 #if 0
11695 	char str[256];
11696 	char path_str[64];
11697 	struct sbuf sb;
11698 #endif
11699 
11700 	/*
11701 	 * In this case, we had to malloc the memory locally.  Free it.
11702 	 */
11703 	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
11704 		int i;
11705 		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
11706 			free(io->io_hdr.local_sglist[i].addr, M_CTL);
11707 	}
11708 
11709 #if 0
11710 	scsi_path_string(io, path_str, sizeof(path_str));
11711 	sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
11712 	sbuf_cat(&sb, path_str);
11713 	scsi_command_string(&io->scsiio, NULL, &sb);
11714 	sbuf_printf(&sb, "\n");
11715 	sbuf_cat(&sb, path_str);
11716 	sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
11717 		    io->scsiio.tag_num, io->scsiio.tag_type);
11718 	sbuf_cat(&sb, path_str);
11719 	sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
11720 		    io->io_hdr.flags, io->io_hdr.status);
11721 	sbuf_finish(&sb);
11722 	printk("%s", sbuf_data(&sb));
11723 #endif
11724 
11725 
11726 	/*
11727 	 * The read is done, now we need to send status (good or bad) back
11728 	 * to the other side.
11729 	 */
11730 	ctl_send_datamove_done(io, /*have_lock*/ 0);
11731 
11732 	return (0);
11733 }
11734 
11735 static void
11736 ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
11737 {
11738 	union ctl_io *io;
11739 	void (*fe_datamove)(union ctl_io *io);
11740 
11741 	io = rq->context;
11742 
11743 	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
11744 		printf("%s: ISC DMA read failed with error %d", __func__,
11745 		       rq->ret);
11746 		ctl_set_internal_failure(&io->scsiio,
11747 					 /*sks_valid*/ 1,
11748 					 /*retry_count*/ rq->ret);
11749 	}
11750 
11751 	ctl_dt_req_free(rq);
11752 
11753 	/* Switch the pointer over so the FETD knows what to do */
11754 	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
11755 
11756 	/*
11757 	 * Use a custom move done callback, since we need to send completion
11758 	 * back to the other controller, not to the backend on this side.
11759 	 */
11760 	io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
11761 
11762 	/* XXX KDM add checks like the ones in ctl_datamove? */
11763 
11764 	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
11765 
11766 	fe_datamove(io);
11767 }
11768 
11769 static int
11770 ctl_datamove_remote_sgl_setup(union ctl_io *io)
11771 {
11772 	struct ctl_sg_entry *local_sglist, *remote_sglist;
11773 	struct ctl_sg_entry *local_dma_sglist, *remote_dma_sglist;
11774 	struct ctl_softc *softc;
11775 	int retval;
11776 	int i;
11777 
11778 	retval = 0;
11779 	softc = control_softc;
11780 
11781 	local_sglist = io->io_hdr.local_sglist;
11782 	local_dma_sglist = io->io_hdr.local_dma_sglist;
11783 	remote_sglist = io->io_hdr.remote_sglist;
11784 	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
11785 
11786 	if (io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) {
11787 		for (i = 0; i < io->scsiio.kern_sg_entries; i++) {
11788 			local_sglist[i].len = remote_sglist[i].len;
11789 
11790 			/*
11791 			 * XXX Detect the situation where the RS-level I/O
11792 			 * redirector on the other side has already read the
11793 			 * data off of the AOR RS on this side, and
11794 			 * transferred it to remote (mirror) memory on the
11795 			 * other side.  Since we already have the data in
11796 			 * memory here, we just need to use it.
11797 			 *
11798 			 * XXX KDM this can probably be removed once we
11799 			 * get the cache device code in and take the
11800 			 * current AOR implementation out.
11801 			 */
11802 #ifdef NEEDTOPORT
11803 			if ((remote_sglist[i].addr >=
11804 			     (void *)vtophys(softc->mirr->addr))
11805 			 && (remote_sglist[i].addr <
11806 			     ((void *)vtophys(softc->mirr->addr) +
11807 			     CacheMirrorOffset))) {
11808 				local_sglist[i].addr = remote_sglist[i].addr -
11809 					CacheMirrorOffset;
11810 				if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
11811 				     CTL_FLAG_DATA_IN)
11812 					io->io_hdr.flags |= CTL_FLAG_REDIR_DONE;
11813 			} else {
11814 				local_sglist[i].addr = remote_sglist[i].addr +
11815 					CacheMirrorOffset;
11816 			}
11817 #endif
11818 #if 0
11819 			printf("%s: local %p, remote %p, len %d\n",
11820 			       __func__, local_sglist[i].addr,
11821 			       remote_sglist[i].addr, local_sglist[i].len);
11822 #endif
11823 		}
11824 	} else {
11825 		uint32_t len_to_go;
11826 
11827 		/*
11828 		 * In this case, we don't have automatically allocated
11829 		 * memory for this I/O on this controller.  This typically
11830 		 * happens with internal CTL I/O -- e.g. inquiry, mode
11831 		 * sense, etc.  Anything coming from RAIDCore will have
11832 		 * a mirror area available.
11833 		 */
11834 		len_to_go = io->scsiio.kern_data_len;
11835 
11836 		/*
11837 		 * Clear the no datasync flag, we have to use malloced
11838 		 * buffers.
11839 		 */
11840 		io->io_hdr.flags &= ~CTL_FLAG_NO_DATASYNC;
11841 
11842 		/*
11843 		 * The difficult thing here is that the size of the various
11844 		 * S/G segments may be different than the size from the
11845 		 * remote controller.  That'll make it harder when DMAing
11846 		 * the data back to the other side.
11847 		 */
11848 		for (i = 0; (i < sizeof(io->io_hdr.remote_sglist) /
11849 		     sizeof(io->io_hdr.remote_sglist[0])) &&
11850 		     (len_to_go > 0); i++) {
11851 			local_sglist[i].len = ctl_min(len_to_go, 131072);
11852 			CTL_SIZE_8B(local_dma_sglist[i].len,
11853 				    local_sglist[i].len);
11854 			local_sglist[i].addr =
11855 				malloc(local_dma_sglist[i].len, M_CTL,M_WAITOK);
11856 
11857 			local_dma_sglist[i].addr = local_sglist[i].addr;
11858 
11859 			if (local_sglist[i].addr == NULL) {
11860 				int j;
11861 
11862 				printf("malloc failed for %zd bytes!",
11863 				       local_dma_sglist[i].len);
11864 				for (j = 0; j < i; j++) {
11865 					free(local_sglist[j].addr, M_CTL);
11866 				}
11867 				ctl_set_internal_failure(&io->scsiio,
11868 							 /*sks_valid*/ 1,
11869 							 /*retry_count*/ 4857);
11870 				retval = 1;
11871 				goto bailout_error;
11872 
11873 			}
11874 			/* XXX KDM do we need a sync here? */
11875 
11876 			len_to_go -= local_sglist[i].len;
11877 		}
11878 		/*
11879 		 * Reset the number of S/G entries accordingly.  The
11880 		 * original number of S/G entries is available in
11881 		 * rem_sg_entries.
11882 		 */
11883 		io->scsiio.kern_sg_entries = i;
11884 
11885 #if 0
11886 		printf("%s: kern_sg_entries = %d\n", __func__,
11887 		       io->scsiio.kern_sg_entries);
11888 		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
11889 			printf("%s: sg[%d] = %p, %d (DMA: %d)\n", __func__, i,
11890 			       local_sglist[i].addr, local_sglist[i].len,
11891 			       local_dma_sglist[i].len);
11892 #endif
11893 	}
11894 
11895 
11896 	return (retval);
11897 
11898 bailout_error:
11899 
11900 	ctl_send_datamove_done(io, /*have_lock*/ 0);
11901 
11902 	return (retval);
11903 }
11904 
11905 static int
11906 ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
11907 			 ctl_ha_dt_cb callback)
11908 {
11909 	struct ctl_ha_dt_req *rq;
11910 	struct ctl_sg_entry *remote_sglist, *local_sglist;
11911 	struct ctl_sg_entry *remote_dma_sglist, *local_dma_sglist;
11912 	uint32_t local_used, remote_used, total_used;
11913 	int retval;
11914 	int i, j;
11915 
11916 	retval = 0;
11917 
11918 	rq = ctl_dt_req_alloc();
11919 
11920 	/*
11921 	 * If we failed to allocate the request, and if the DMA didn't fail
11922 	 * anyway, set busy status.  This is just a resource allocation
11923 	 * failure.
11924 	 */
11925 	if ((rq == NULL)
11926 	 && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE))
11927 		ctl_set_busy(&io->scsiio);
11928 
11929 	if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) {
11930 
11931 		if (rq != NULL)
11932 			ctl_dt_req_free(rq);
11933 
11934 		/*
11935 		 * The data move failed.  We need to return status back
11936 		 * to the other controller.  No point in trying to DMA
11937 		 * data to the remote controller.
11938 		 */
11939 
11940 		ctl_send_datamove_done(io, /*have_lock*/ 0);
11941 
11942 		retval = 1;
11943 
11944 		goto bailout;
11945 	}
11946 
11947 	local_sglist = io->io_hdr.local_sglist;
11948 	local_dma_sglist = io->io_hdr.local_dma_sglist;
11949 	remote_sglist = io->io_hdr.remote_sglist;
11950 	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
11951 	local_used = 0;
11952 	remote_used = 0;
11953 	total_used = 0;
11954 
11955 	if (io->io_hdr.flags & CTL_FLAG_REDIR_DONE) {
11956 		rq->ret = CTL_HA_STATUS_SUCCESS;
11957 		rq->context = io;
11958 		callback(rq);
11959 		goto bailout;
11960 	}
11961 
11962 	/*
11963 	 * Pull/push the data over the wire from/to the other controller.
11964 	 * This takes into account the possibility that the local and
11965 	 * remote sglists may not be identical in terms of the size of
11966 	 * the elements and the number of elements.
11967 	 *
11968 	 * One fundamental assumption here is that the length allocated for
11969 	 * both the local and remote sglists is identical.  Otherwise, we've
11970 	 * essentially got a coding error of some sort.
11971 	 */
11972 	for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
11973 		int isc_ret;
11974 		uint32_t cur_len, dma_length;
11975 		uint8_t *tmp_ptr;
11976 
11977 		rq->id = CTL_HA_DATA_CTL;
11978 		rq->command = command;
11979 		rq->context = io;
11980 
11981 		/*
11982 		 * Both pointers should be aligned.  But it is possible
11983 		 * that the allocation length is not.  They should both
11984 		 * also have enough slack left over at the end, though,
11985 		 * to round up to the next 8 byte boundary.
11986 		 */
11987 		cur_len = ctl_min(local_sglist[i].len - local_used,
11988 				  remote_sglist[j].len - remote_used);
11989 
11990 		/*
11991 		 * In this case, we have a size issue and need to decrease
11992 		 * the size, except in the case where we actually have less
11993 		 * than 8 bytes left.  In that case, we need to increase
11994 		 * the DMA length to get the last bit.
11995 		 */
11996 		if ((cur_len & 0x7) != 0) {
11997 			if (cur_len > 0x7) {
11998 				cur_len = cur_len - (cur_len & 0x7);
11999 				dma_length = cur_len;
12000 			} else {
12001 				CTL_SIZE_8B(dma_length, cur_len);
12002 			}
12003 
12004 		} else
12005 			dma_length = cur_len;
12006 
12007 		/*
12008 		 * If we had to allocate memory for this I/O, instead of using
12009 		 * the non-cached mirror memory, we'll need to flush the cache
12010 		 * before trying to DMA to the other controller.
12011 		 *
12012 		 * We could end up doing this multiple times for the same
12013 		 * segment if we have a larger local segment than remote
12014 		 * segment.  That shouldn't be an issue.
12015 		 */
12016 		if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12017 			/*
12018 			 * XXX KDM use bus_dmamap_sync() here.
12019 			 */
12020 		}
12021 
12022 		rq->size = dma_length;
12023 
12024 		tmp_ptr = (uint8_t *)local_sglist[i].addr;
12025 		tmp_ptr += local_used;
12026 
12027 		/* Use physical addresses when talking to ISC hardware */
12028 		if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
12029 			/* XXX KDM use busdma */
12030 #if 0
12031 			rq->local = vtophys(tmp_ptr);
12032 #endif
12033 		} else
12034 			rq->local = tmp_ptr;
12035 
12036 		tmp_ptr = (uint8_t *)remote_sglist[j].addr;
12037 		tmp_ptr += remote_used;
12038 		rq->remote = tmp_ptr;
12039 
12040 		rq->callback = NULL;
12041 
12042 		local_used += cur_len;
12043 		if (local_used >= local_sglist[i].len) {
12044 			i++;
12045 			local_used = 0;
12046 		}
12047 
12048 		remote_used += cur_len;
12049 		if (remote_used >= remote_sglist[j].len) {
12050 			j++;
12051 			remote_used = 0;
12052 		}
12053 		total_used += cur_len;
12054 
12055 		if (total_used >= io->scsiio.kern_data_len)
12056 			rq->callback = callback;
12057 
12058 		if ((rq->size & 0x7) != 0) {
12059 			printf("%s: warning: size %d is not on 8b boundary\n",
12060 			       __func__, rq->size);
12061 		}
12062 		if (((uintptr_t)rq->local & 0x7) != 0) {
12063 			printf("%s: warning: local %p not on 8b boundary\n",
12064 			       __func__, rq->local);
12065 		}
12066 		if (((uintptr_t)rq->remote & 0x7) != 0) {
12067 			printf("%s: warning: remote %p not on 8b boundary\n",
12068 			       __func__, rq->local);
12069 		}
12070 #if 0
12071 		printf("%s: %s: local %#x remote %#x size %d\n", __func__,
12072 		       (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
12073 		       rq->local, rq->remote, rq->size);
12074 #endif
12075 
12076 		isc_ret = ctl_dt_single(rq);
12077 		if (isc_ret == CTL_HA_STATUS_WAIT)
12078 			continue;
12079 
12080 		if (isc_ret == CTL_HA_STATUS_DISCONNECT) {
12081 			rq->ret = CTL_HA_STATUS_SUCCESS;
12082 		} else {
12083 			rq->ret = isc_ret;
12084 		}
12085 		callback(rq);
12086 		goto bailout;
12087 	}
12088 
12089 bailout:
12090 	return (retval);
12091 
12092 }
12093 
12094 static void
12095 ctl_datamove_remote_read(union ctl_io *io)
12096 {
12097 	int retval;
12098 	int i;
12099 
12100 	/*
12101 	 * This will send an error to the other controller in the case of a
12102 	 * failure.
12103 	 */
12104 	retval = ctl_datamove_remote_sgl_setup(io);
12105 	if (retval != 0)
12106 		return;
12107 
12108 	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
12109 					  ctl_datamove_remote_read_cb);
12110 	if ((retval != 0)
12111 	 && ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0)) {
12112 		/*
12113 		 * Make sure we free memory if there was an error..  The
12114 		 * ctl_datamove_remote_xfer() function will send the
12115 		 * datamove done message, or call the callback with an
12116 		 * error if there is a problem.
12117 		 */
12118 		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12119 			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12120 	}
12121 
12122 	return;
12123 }
12124 
12125 /*
12126  * Process a datamove request from the other controller.  This is used for
12127  * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
12128  * first.  Once that is complete, the data gets DMAed into the remote
12129  * controller's memory.  For reads, we DMA from the remote controller's
12130  * memory into our memory first, and then move it out to the FETD.
12131  *
12132  * Should be called without the ctl_lock held.
12133  */
12134 static void
12135 ctl_datamove_remote(union ctl_io *io)
12136 {
12137 	struct ctl_softc *softc;
12138 
12139 	softc = control_softc;
12140 
12141 	/*
12142 	 * Note that we look for an aborted I/O here, but don't do some of
12143 	 * the other checks that ctl_datamove() normally does.  We don't
12144 	 * need to run the task queue, because this I/O is on the ISC
12145 	 * queue, which is executed by the work thread after the task queue.
12146 	 * We don't need to run the datamove delay code, since that should
12147 	 * have been done if need be on the other controller.
12148 	 */
12149 	mtx_lock(&softc->ctl_lock);
12150 
12151 	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12152 
12153 		printf("%s: tag 0x%04x on (%d:%d:%d:%d) aborted\n", __func__,
12154 		       io->scsiio.tag_num, io->io_hdr.nexus.initid.id,
12155 		       io->io_hdr.nexus.targ_port,
12156 		       io->io_hdr.nexus.targ_target.id,
12157 		       io->io_hdr.nexus.targ_lun);
12158 		io->io_hdr.status = CTL_CMD_ABORTED;
12159 		io->io_hdr.port_status = 31338;
12160 
12161 		mtx_unlock(&softc->ctl_lock);
12162 
12163 		ctl_send_datamove_done(io, /*have_lock*/ 0);
12164 
12165 		return;
12166 	}
12167 
12168 	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT) {
12169 		mtx_unlock(&softc->ctl_lock);
12170 		ctl_datamove_remote_write(io);
12171 	} else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN){
12172 		mtx_unlock(&softc->ctl_lock);
12173 		ctl_datamove_remote_read(io);
12174 	} else {
12175 		union ctl_ha_msg msg;
12176 		struct scsi_sense_data *sense;
12177 		uint8_t sks[3];
12178 		int retry_count;
12179 
12180 		memset(&msg, 0, sizeof(msg));
12181 
12182 		msg.hdr.msg_type = CTL_MSG_BAD_JUJU;
12183 		msg.hdr.status = CTL_SCSI_ERROR;
12184 		msg.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
12185 
12186 		retry_count = 4243;
12187 
12188 		sense = &msg.scsi.sense_data;
12189 		sks[0] = SSD_SCS_VALID;
12190 		sks[1] = (retry_count >> 8) & 0xff;
12191 		sks[2] = retry_count & 0xff;
12192 
12193 		/* "Internal target failure" */
12194 		scsi_set_sense_data(sense,
12195 				    /*sense_format*/ SSD_TYPE_NONE,
12196 				    /*current_error*/ 1,
12197 				    /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
12198 				    /*asc*/ 0x44,
12199 				    /*ascq*/ 0x00,
12200 				    /*type*/ SSD_ELEM_SKS,
12201 				    /*size*/ sizeof(sks),
12202 				    /*data*/ sks,
12203 				    SSD_ELEM_NONE);
12204 
12205 		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12206 		if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12207 			ctl_failover_io(io, /*have_lock*/ 1);
12208 			mtx_unlock(&softc->ctl_lock);
12209 			return;
12210 		}
12211 
12212 		mtx_unlock(&softc->ctl_lock);
12213 
12214 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0) >
12215 		    CTL_HA_STATUS_SUCCESS) {
12216 			/* XXX KDM what to do if this fails? */
12217 		}
12218 		return;
12219 	}
12220 
12221 }
12222 
12223 static int
12224 ctl_process_done(union ctl_io *io, int have_lock)
12225 {
12226 	struct ctl_lun *lun;
12227 	struct ctl_softc *ctl_softc;
12228 	void (*fe_done)(union ctl_io *io);
12229 	uint32_t targ_port = ctl_port_idx(io->io_hdr.nexus.targ_port);
12230 
12231 	CTL_DEBUG_PRINT(("ctl_process_done\n"));
12232 
12233 	fe_done =
12234 	    control_softc->ctl_ports[targ_port]->fe_done;
12235 
12236 #ifdef CTL_TIME_IO
12237 	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12238 		char str[256];
12239 		char path_str[64];
12240 		struct sbuf sb;
12241 
12242 		ctl_scsi_path_string(io, path_str, sizeof(path_str));
12243 		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12244 
12245 		sbuf_cat(&sb, path_str);
12246 		switch (io->io_hdr.io_type) {
12247 		case CTL_IO_SCSI:
12248 			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12249 			sbuf_printf(&sb, "\n");
12250 			sbuf_cat(&sb, path_str);
12251 			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12252 				    io->scsiio.tag_num, io->scsiio.tag_type);
12253 			break;
12254 		case CTL_IO_TASK:
12255 			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12256 				    "Tag Type: %d\n", io->taskio.task_action,
12257 				    io->taskio.tag_num, io->taskio.tag_type);
12258 			break;
12259 		default:
12260 			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12261 			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12262 			break;
12263 		}
12264 		sbuf_cat(&sb, path_str);
12265 		sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
12266 			    (intmax_t)time_uptime - io->io_hdr.start_time);
12267 		sbuf_finish(&sb);
12268 		printf("%s", sbuf_data(&sb));
12269 	}
12270 #endif /* CTL_TIME_IO */
12271 
12272 	switch (io->io_hdr.io_type) {
12273 	case CTL_IO_SCSI:
12274 		break;
12275 	case CTL_IO_TASK:
12276 		ctl_io_error_print(io, NULL);
12277 		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
12278 			ctl_free_io_internal(io, /*have_lock*/ 0);
12279 		else
12280 			fe_done(io);
12281 		return (CTL_RETVAL_COMPLETE);
12282 		break;
12283 	default:
12284 		printf("ctl_process_done: invalid io type %d\n",
12285 		       io->io_hdr.io_type);
12286 		panic("ctl_process_done: invalid io type %d\n",
12287 		      io->io_hdr.io_type);
12288 		break; /* NOTREACHED */
12289 	}
12290 
12291 	lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12292 	if (lun == NULL) {
12293 		CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
12294 				 io->io_hdr.nexus.targ_lun));
12295 		fe_done(io);
12296 		goto bailout;
12297 	}
12298 	ctl_softc = lun->ctl_softc;
12299 
12300 	/*
12301 	 * Remove this from the OOA queue.
12302 	 */
12303 	if (have_lock == 0)
12304 		mtx_lock(&ctl_softc->ctl_lock);
12305 
12306 	/*
12307 	 * Check to see if we have any errors to inject here.  We only
12308 	 * inject errors for commands that don't already have errors set.
12309 	 */
12310 	if ((STAILQ_FIRST(&lun->error_list) != NULL)
12311 	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))
12312 		ctl_inject_error(lun, io);
12313 
12314 	/*
12315 	 * XXX KDM how do we treat commands that aren't completed
12316 	 * successfully?
12317 	 *
12318 	 * XXX KDM should we also track I/O latency?
12319 	 */
12320 	if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) {
12321 		uint32_t blocksize;
12322 #ifdef CTL_TIME_IO
12323 		struct bintime cur_bt;
12324 #endif
12325 
12326 		if ((lun->be_lun != NULL)
12327 		 && (lun->be_lun->blocksize != 0))
12328 			blocksize = lun->be_lun->blocksize;
12329 		else
12330 			blocksize = 512;
12331 
12332 		switch (io->io_hdr.io_type) {
12333 		case CTL_IO_SCSI: {
12334 			int isread;
12335 			struct ctl_lba_len lbalen;
12336 
12337 			isread = 0;
12338 			switch (io->scsiio.cdb[0]) {
12339 			case READ_6:
12340 			case READ_10:
12341 			case READ_12:
12342 			case READ_16:
12343 				isread = 1;
12344 				/* FALLTHROUGH */
12345 			case WRITE_6:
12346 			case WRITE_10:
12347 			case WRITE_12:
12348 			case WRITE_16:
12349 			case WRITE_VERIFY_10:
12350 			case WRITE_VERIFY_12:
12351 			case WRITE_VERIFY_16:
12352 				memcpy(&lbalen, io->io_hdr.ctl_private[
12353 				       CTL_PRIV_LBA_LEN].bytes, sizeof(lbalen));
12354 
12355 				if (isread) {
12356 					lun->stats.ports[targ_port].bytes[CTL_STATS_READ] +=
12357 						lbalen.len * blocksize;
12358 					lun->stats.ports[targ_port].operations[CTL_STATS_READ]++;
12359 
12360 #ifdef CTL_TIME_IO
12361 					bintime_add(
12362 					   &lun->stats.ports[targ_port].dma_time[CTL_STATS_READ],
12363 					   &io->io_hdr.dma_bt);
12364 					lun->stats.ports[targ_port].num_dmas[CTL_STATS_READ] +=
12365 						io->io_hdr.num_dmas;
12366 					getbintime(&cur_bt);
12367 					bintime_sub(&cur_bt,
12368 						    &io->io_hdr.start_bt);
12369 
12370 					bintime_add(
12371 					    &lun->stats.ports[targ_port].time[CTL_STATS_READ],
12372 					    &cur_bt);
12373 
12374 #if 0
12375 					cs_prof_gettime(&cur_ticks);
12376 					lun->stats.time[CTL_STATS_READ] +=
12377 						cur_ticks -
12378 						io->io_hdr.start_ticks;
12379 #endif
12380 #if 0
12381 					lun->stats.time[CTL_STATS_READ] +=
12382 						jiffies - io->io_hdr.start_time;
12383 #endif
12384 #endif /* CTL_TIME_IO */
12385 				} else {
12386 					lun->stats.ports[targ_port].bytes[CTL_STATS_WRITE] +=
12387 						lbalen.len * blocksize;
12388 					lun->stats.ports[targ_port].operations[
12389 						CTL_STATS_WRITE]++;
12390 
12391 #ifdef CTL_TIME_IO
12392 					bintime_add(
12393 					  &lun->stats.ports[targ_port].dma_time[CTL_STATS_WRITE],
12394 					  &io->io_hdr.dma_bt);
12395 					lun->stats.ports[targ_port].num_dmas[CTL_STATS_WRITE] +=
12396 						io->io_hdr.num_dmas;
12397 					getbintime(&cur_bt);
12398 					bintime_sub(&cur_bt,
12399 						    &io->io_hdr.start_bt);
12400 
12401 					bintime_add(
12402 					    &lun->stats.ports[targ_port].time[CTL_STATS_WRITE],
12403 					    &cur_bt);
12404 #if 0
12405 					cs_prof_gettime(&cur_ticks);
12406 					lun->stats.ports[targ_port].time[CTL_STATS_WRITE] +=
12407 						cur_ticks -
12408 						io->io_hdr.start_ticks;
12409 					lun->stats.ports[targ_port].time[CTL_STATS_WRITE] +=
12410 						jiffies - io->io_hdr.start_time;
12411 #endif
12412 #endif /* CTL_TIME_IO */
12413 				}
12414 				break;
12415 			default:
12416 				lun->stats.ports[targ_port].operations[CTL_STATS_NO_IO]++;
12417 
12418 #ifdef CTL_TIME_IO
12419 				bintime_add(
12420 				  &lun->stats.ports[targ_port].dma_time[CTL_STATS_NO_IO],
12421 				  &io->io_hdr.dma_bt);
12422 				lun->stats.ports[targ_port].num_dmas[CTL_STATS_NO_IO] +=
12423 					io->io_hdr.num_dmas;
12424 				getbintime(&cur_bt);
12425 				bintime_sub(&cur_bt, &io->io_hdr.start_bt);
12426 
12427 				bintime_add(&lun->stats.ports[targ_port].time[CTL_STATS_NO_IO],
12428 					    &cur_bt);
12429 
12430 #if 0
12431 				cs_prof_gettime(&cur_ticks);
12432 				lun->stats.ports[targ_port].time[CTL_STATS_NO_IO] +=
12433 					cur_ticks -
12434 					io->io_hdr.start_ticks;
12435 				lun->stats.ports[targ_port].time[CTL_STATS_NO_IO] +=
12436 					jiffies - io->io_hdr.start_time;
12437 #endif
12438 #endif /* CTL_TIME_IO */
12439 				break;
12440 			}
12441 			break;
12442 		}
12443 		default:
12444 			break;
12445 		}
12446 	}
12447 
12448 	TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
12449 
12450 	/*
12451 	 * Run through the blocked queue on this LUN and see if anything
12452 	 * has become unblocked, now that this transaction is done.
12453 	 */
12454 	ctl_check_blocked(lun);
12455 
12456 	/*
12457 	 * If the LUN has been invalidated, free it if there is nothing
12458 	 * left on its OOA queue.
12459 	 */
12460 	if ((lun->flags & CTL_LUN_INVALID)
12461 	 && (TAILQ_FIRST(&lun->ooa_queue) == NULL))
12462 		ctl_free_lun(lun);
12463 
12464 	/*
12465 	 * If this command has been aborted, make sure we set the status
12466 	 * properly.  The FETD is responsible for freeing the I/O and doing
12467 	 * whatever it needs to do to clean up its state.
12468 	 */
12469 	if (io->io_hdr.flags & CTL_FLAG_ABORT)
12470 		io->io_hdr.status = CTL_CMD_ABORTED;
12471 
12472 	/*
12473 	 * We print out status for every task management command.  For SCSI
12474 	 * commands, we filter out any unit attention errors; they happen
12475 	 * on every boot, and would clutter up the log.  Note:  task
12476 	 * management commands aren't printed here, they are printed above,
12477 	 * since they should never even make it down here.
12478 	 */
12479 	switch (io->io_hdr.io_type) {
12480 	case CTL_IO_SCSI: {
12481 		int error_code, sense_key, asc, ascq;
12482 
12483 		sense_key = 0;
12484 
12485 		if (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SCSI_ERROR)
12486 		 && (io->scsiio.scsi_status == SCSI_STATUS_CHECK_COND)) {
12487 			/*
12488 			 * Since this is just for printing, no need to
12489 			 * show errors here.
12490 			 */
12491 			scsi_extract_sense_len(&io->scsiio.sense_data,
12492 					       io->scsiio.sense_len,
12493 					       &error_code,
12494 					       &sense_key,
12495 					       &asc,
12496 					       &ascq,
12497 					       /*show_errors*/ 0);
12498 		}
12499 
12500 		if (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
12501 		 && (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SCSI_ERROR)
12502 		  || (io->scsiio.scsi_status != SCSI_STATUS_CHECK_COND)
12503 		  || (sense_key != SSD_KEY_UNIT_ATTENTION))) {
12504 
12505 			if ((time_uptime - ctl_softc->last_print_jiffies) <= 0){
12506 				ctl_softc->skipped_prints++;
12507 				if (have_lock == 0)
12508 					mtx_unlock(&ctl_softc->ctl_lock);
12509 			} else {
12510 				uint32_t skipped_prints;
12511 
12512 				skipped_prints = ctl_softc->skipped_prints;
12513 
12514 				ctl_softc->skipped_prints = 0;
12515 				ctl_softc->last_print_jiffies = time_uptime;
12516 
12517 				if (have_lock == 0)
12518 					mtx_unlock(&ctl_softc->ctl_lock);
12519 				if (skipped_prints > 0) {
12520 #ifdef NEEDTOPORT
12521 					csevent_log(CSC_CTL | CSC_SHELF_SW |
12522 					    CTL_ERROR_REPORT,
12523 					    csevent_LogType_Trace,
12524 					    csevent_Severity_Information,
12525 					    csevent_AlertLevel_Green,
12526 					    csevent_FRU_Firmware,
12527 					    csevent_FRU_Unknown,
12528 					    "High CTL error volume, %d prints "
12529 					    "skipped", skipped_prints);
12530 #endif
12531 				}
12532 				ctl_io_error_print(io, NULL);
12533 			}
12534 		} else {
12535 			if (have_lock == 0)
12536 				mtx_unlock(&ctl_softc->ctl_lock);
12537 		}
12538 		break;
12539 	}
12540 	case CTL_IO_TASK:
12541 		if (have_lock == 0)
12542 			mtx_unlock(&ctl_softc->ctl_lock);
12543 		ctl_io_error_print(io, NULL);
12544 		break;
12545 	default:
12546 		if (have_lock == 0)
12547 			mtx_unlock(&ctl_softc->ctl_lock);
12548 		break;
12549 	}
12550 
12551 	/*
12552 	 * Tell the FETD or the other shelf controller we're done with this
12553 	 * command.  Note that only SCSI commands get to this point.  Task
12554 	 * management commands are completed above.
12555 	 *
12556 	 * We only send status to the other controller if we're in XFER
12557 	 * mode.  In SER_ONLY mode, the I/O is done on the controller that
12558 	 * received the I/O (from CTL's perspective), and so the status is
12559 	 * generated there.
12560 	 *
12561 	 * XXX KDM if we hold the lock here, we could cause a deadlock
12562 	 * if the frontend comes back in in this context to queue
12563 	 * something.
12564 	 */
12565 	if ((ctl_softc->ha_mode == CTL_HA_MODE_XFER)
12566 	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12567 		union ctl_ha_msg msg;
12568 
12569 		memset(&msg, 0, sizeof(msg));
12570 		msg.hdr.msg_type = CTL_MSG_FINISH_IO;
12571 		msg.hdr.original_sc = io->io_hdr.original_sc;
12572 		msg.hdr.nexus = io->io_hdr.nexus;
12573 		msg.hdr.status = io->io_hdr.status;
12574 		msg.scsi.scsi_status = io->scsiio.scsi_status;
12575 		msg.scsi.tag_num = io->scsiio.tag_num;
12576 		msg.scsi.tag_type = io->scsiio.tag_type;
12577 		msg.scsi.sense_len = io->scsiio.sense_len;
12578 		msg.scsi.sense_residual = io->scsiio.sense_residual;
12579 		msg.scsi.residual = io->scsiio.residual;
12580 		memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12581 		       sizeof(io->scsiio.sense_data));
12582 		/*
12583 		 * We copy this whether or not this is an I/O-related
12584 		 * command.  Otherwise, we'd have to go and check to see
12585 		 * whether it's a read/write command, and it really isn't
12586 		 * worth it.
12587 		 */
12588 		memcpy(&msg.scsi.lbalen,
12589 		       &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
12590 		       sizeof(msg.scsi.lbalen));
12591 
12592 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12593 				sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
12594 			/* XXX do something here */
12595 		}
12596 
12597 		ctl_free_io_internal(io, /*have_lock*/ 0);
12598 	} else
12599 		fe_done(io);
12600 
12601 bailout:
12602 
12603 	return (CTL_RETVAL_COMPLETE);
12604 }
12605 
12606 /*
12607  * Front end should call this if it doesn't do autosense.  When the request
12608  * sense comes back in from the initiator, we'll dequeue this and send it.
12609  */
12610 int
12611 ctl_queue_sense(union ctl_io *io)
12612 {
12613 	struct ctl_lun *lun;
12614 	struct ctl_softc *ctl_softc;
12615 	uint32_t initidx;
12616 
12617 	ctl_softc = control_softc;
12618 
12619 	CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
12620 
12621 	/*
12622 	 * LUN lookup will likely move to the ctl_work_thread() once we
12623 	 * have our new queueing infrastructure (that doesn't put things on
12624 	 * a per-LUN queue initially).  That is so that we can handle
12625 	 * things like an INQUIRY to a LUN that we don't have enabled.  We
12626 	 * can't deal with that right now.
12627 	 */
12628 	mtx_lock(&ctl_softc->ctl_lock);
12629 
12630 	/*
12631 	 * If we don't have a LUN for this, just toss the sense
12632 	 * information.
12633 	 */
12634 	if ((io->io_hdr.nexus.targ_lun < CTL_MAX_LUNS)
12635 	 && (ctl_softc->ctl_luns[io->io_hdr.nexus.targ_lun] != NULL))
12636 		lun = ctl_softc->ctl_luns[io->io_hdr.nexus.targ_lun];
12637 	else
12638 		goto bailout;
12639 
12640 	initidx = ctl_get_initindex(&io->io_hdr.nexus);
12641 
12642 	/*
12643 	 * Already have CA set for this LUN...toss the sense information.
12644 	 */
12645 	if (ctl_is_set(lun->have_ca, initidx))
12646 		goto bailout;
12647 
12648 	memcpy(&lun->pending_sense[initidx].sense, &io->scsiio.sense_data,
12649 	       ctl_min(sizeof(lun->pending_sense[initidx].sense),
12650 	       sizeof(io->scsiio.sense_data)));
12651 	ctl_set_mask(lun->have_ca, initidx);
12652 
12653 bailout:
12654 	mtx_unlock(&ctl_softc->ctl_lock);
12655 
12656 	ctl_free_io(io);
12657 
12658 	return (CTL_RETVAL_COMPLETE);
12659 }
12660 
12661 /*
12662  * Primary command inlet from frontend ports.  All SCSI and task I/O
12663  * requests must go through this function.
12664  */
12665 int
12666 ctl_queue(union ctl_io *io)
12667 {
12668 	struct ctl_softc *ctl_softc;
12669 
12670 	CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
12671 
12672 	ctl_softc = control_softc;
12673 
12674 #ifdef CTL_TIME_IO
12675 	io->io_hdr.start_time = time_uptime;
12676 	getbintime(&io->io_hdr.start_bt);
12677 #endif /* CTL_TIME_IO */
12678 
12679 	mtx_lock(&ctl_softc->ctl_lock);
12680 
12681 	switch (io->io_hdr.io_type) {
12682 	case CTL_IO_SCSI:
12683 		STAILQ_INSERT_TAIL(&ctl_softc->incoming_queue, &io->io_hdr,
12684 				   links);
12685 		break;
12686 	case CTL_IO_TASK:
12687 		STAILQ_INSERT_TAIL(&ctl_softc->task_queue, &io->io_hdr, links);
12688 		/*
12689 		 * Set the task pending flag.  This is necessary to close a
12690 		 * race condition with the FETD:
12691 		 *
12692 		 * - FETD submits a task management command, like an abort.
12693 		 * - Back end calls fe_datamove() to move the data for the
12694 		 *   aborted command.  The FETD can't really accept it, but
12695 		 *   if it did, it would end up transmitting data for a
12696 		 *   command that the initiator told us to abort.
12697 		 *
12698 		 * We close the race condition by setting the flag here,
12699 		 * and checking it in ctl_datamove(), before calling the
12700 		 * FETD's fe_datamove routine.  If we've got a task
12701 		 * pending, we run the task queue and then check to see
12702 		 * whether our particular I/O has been aborted.
12703 		 */
12704 		ctl_softc->flags |= CTL_FLAG_TASK_PENDING;
12705 		break;
12706 	default:
12707 		mtx_unlock(&ctl_softc->ctl_lock);
12708 		printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
12709 		return (-EINVAL);
12710 		break; /* NOTREACHED */
12711 	}
12712 	mtx_unlock(&ctl_softc->ctl_lock);
12713 
12714 	ctl_wakeup_thread();
12715 
12716 	return (CTL_RETVAL_COMPLETE);
12717 }
12718 
12719 #ifdef CTL_IO_DELAY
12720 static void
12721 ctl_done_timer_wakeup(void *arg)
12722 {
12723 	union ctl_io *io;
12724 
12725 	io = (union ctl_io *)arg;
12726 	ctl_done_lock(io, /*have_lock*/ 0);
12727 }
12728 #endif /* CTL_IO_DELAY */
12729 
12730 void
12731 ctl_done_lock(union ctl_io *io, int have_lock)
12732 {
12733 	struct ctl_softc *ctl_softc;
12734 #ifndef CTL_DONE_THREAD
12735 	union ctl_io *xio;
12736 #endif /* !CTL_DONE_THREAD */
12737 
12738 	ctl_softc = control_softc;
12739 
12740 	if (have_lock == 0)
12741 		mtx_lock(&ctl_softc->ctl_lock);
12742 
12743 	/*
12744 	 * Enable this to catch duplicate completion issues.
12745 	 */
12746 #if 0
12747 	if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
12748 		printf("%s: type %d msg %d cdb %x iptl: "
12749 		       "%d:%d:%d:%d tag 0x%04x "
12750 		       "flag %#x status %x\n",
12751 			__func__,
12752 			io->io_hdr.io_type,
12753 			io->io_hdr.msg_type,
12754 			io->scsiio.cdb[0],
12755 			io->io_hdr.nexus.initid.id,
12756 			io->io_hdr.nexus.targ_port,
12757 			io->io_hdr.nexus.targ_target.id,
12758 			io->io_hdr.nexus.targ_lun,
12759 			(io->io_hdr.io_type ==
12760 			CTL_IO_TASK) ?
12761 			io->taskio.tag_num :
12762 			io->scsiio.tag_num,
12763 		        io->io_hdr.flags,
12764 			io->io_hdr.status);
12765 	} else
12766 		io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
12767 #endif
12768 
12769 	/*
12770 	 * This is an internal copy of an I/O, and should not go through
12771 	 * the normal done processing logic.
12772 	 */
12773 	if (io->io_hdr.flags & CTL_FLAG_INT_COPY) {
12774 		if (have_lock == 0)
12775 			mtx_unlock(&ctl_softc->ctl_lock);
12776 		return;
12777 	}
12778 
12779 	/*
12780 	 * We need to send a msg to the serializing shelf to finish the IO
12781 	 * as well.  We don't send a finish message to the other shelf if
12782 	 * this is a task management command.  Task management commands
12783 	 * aren't serialized in the OOA queue, but rather just executed on
12784 	 * both shelf controllers for commands that originated on that
12785 	 * controller.
12786 	 */
12787 	if ((io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)
12788 	 && (io->io_hdr.io_type != CTL_IO_TASK)) {
12789 		union ctl_ha_msg msg_io;
12790 
12791 		msg_io.hdr.msg_type = CTL_MSG_FINISH_IO;
12792 		msg_io.hdr.serializing_sc = io->io_hdr.serializing_sc;
12793 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_io,
12794 		    sizeof(msg_io), 0 ) != CTL_HA_STATUS_SUCCESS) {
12795 		}
12796 		/* continue on to finish IO */
12797 	}
12798 #ifdef CTL_IO_DELAY
12799 	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12800 		struct ctl_lun *lun;
12801 
12802 		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12803 
12804 		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12805 	} else {
12806 		struct ctl_lun *lun;
12807 
12808 		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12809 
12810 		if ((lun != NULL)
12811 		 && (lun->delay_info.done_delay > 0)) {
12812 			struct callout *callout;
12813 
12814 			callout = (struct callout *)&io->io_hdr.timer_bytes;
12815 			callout_init(callout, /*mpsafe*/ 1);
12816 			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12817 			callout_reset(callout,
12818 				      lun->delay_info.done_delay * hz,
12819 				      ctl_done_timer_wakeup, io);
12820 			if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
12821 				lun->delay_info.done_delay = 0;
12822 			if (have_lock == 0)
12823 				mtx_unlock(&ctl_softc->ctl_lock);
12824 			return;
12825 		}
12826 	}
12827 #endif /* CTL_IO_DELAY */
12828 
12829 	STAILQ_INSERT_TAIL(&ctl_softc->done_queue, &io->io_hdr, links);
12830 
12831 #ifdef CTL_DONE_THREAD
12832 	if (have_lock == 0)
12833 		mtx_unlock(&ctl_softc->ctl_lock);
12834 
12835 	ctl_wakeup_thread();
12836 #else /* CTL_DONE_THREAD */
12837 	for (xio = (union ctl_io *)STAILQ_FIRST(&ctl_softc->done_queue);
12838 	     xio != NULL;
12839 	     xio =(union ctl_io *)STAILQ_FIRST(&ctl_softc->done_queue)) {
12840 
12841 		STAILQ_REMOVE_HEAD(&ctl_softc->done_queue, links);
12842 
12843 		ctl_process_done(xio, /*have_lock*/ 1);
12844 	}
12845 	if (have_lock == 0)
12846 		mtx_unlock(&ctl_softc->ctl_lock);
12847 #endif /* CTL_DONE_THREAD */
12848 }
12849 
12850 void
12851 ctl_done(union ctl_io *io)
12852 {
12853 	ctl_done_lock(io, /*have_lock*/ 0);
12854 }
12855 
12856 int
12857 ctl_isc(struct ctl_scsiio *ctsio)
12858 {
12859 	struct ctl_lun *lun;
12860 	int retval;
12861 
12862 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12863 
12864 	CTL_DEBUG_PRINT(("ctl_isc: command: %02x\n", ctsio->cdb[0]));
12865 
12866 	CTL_DEBUG_PRINT(("ctl_isc: calling data_submit()\n"));
12867 
12868 	retval = lun->backend->data_submit((union ctl_io *)ctsio);
12869 
12870 	return (retval);
12871 }
12872 
12873 
12874 static void
12875 ctl_work_thread(void *arg)
12876 {
12877 	struct ctl_softc *softc;
12878 	union ctl_io *io;
12879 	struct ctl_be_lun *be_lun;
12880 	int retval;
12881 
12882 	CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
12883 
12884 	softc = (struct ctl_softc *)arg;
12885 	if (softc == NULL)
12886 		return;
12887 
12888 	mtx_lock(&softc->ctl_lock);
12889 	for (;;) {
12890 		retval = 0;
12891 
12892 		/*
12893 		 * We handle the queues in this order:
12894 		 * - task management
12895 		 * - ISC
12896 		 * - done queue (to free up resources, unblock other commands)
12897 		 * - RtR queue
12898 		 * - incoming queue
12899 		 *
12900 		 * If those queues are empty, we break out of the loop and
12901 		 * go to sleep.
12902 		 */
12903 		io = (union ctl_io *)STAILQ_FIRST(&softc->task_queue);
12904 		if (io != NULL) {
12905 			ctl_run_task_queue(softc);
12906 			continue;
12907 		}
12908 		io = (union ctl_io *)STAILQ_FIRST(&softc->isc_queue);
12909 		if (io != NULL) {
12910 			STAILQ_REMOVE_HEAD(&softc->isc_queue, links);
12911 			ctl_handle_isc(io);
12912 			continue;
12913 		}
12914 		io = (union ctl_io *)STAILQ_FIRST(&softc->done_queue);
12915 		if (io != NULL) {
12916 			STAILQ_REMOVE_HEAD(&softc->done_queue, links);
12917 			/* clear any blocked commands, call fe_done */
12918 			mtx_unlock(&softc->ctl_lock);
12919 			/*
12920 			 * XXX KDM
12921 			 * Call this without a lock for now.  This will
12922 			 * depend on whether there is any way the FETD can
12923 			 * sleep or deadlock if called with the CTL lock
12924 			 * held.
12925 			 */
12926 			retval = ctl_process_done(io, /*have_lock*/ 0);
12927 			mtx_lock(&softc->ctl_lock);
12928 			continue;
12929 		}
12930 		if (!ctl_pause_rtr) {
12931 			io = (union ctl_io *)STAILQ_FIRST(&softc->rtr_queue);
12932 			if (io != NULL) {
12933 				STAILQ_REMOVE_HEAD(&softc->rtr_queue, links);
12934 				mtx_unlock(&softc->ctl_lock);
12935 				goto execute;
12936 			}
12937 		}
12938 		io = (union ctl_io *)STAILQ_FIRST(&softc->incoming_queue);
12939 		if (io != NULL) {
12940 			STAILQ_REMOVE_HEAD(&softc->incoming_queue, links);
12941 			mtx_unlock(&softc->ctl_lock);
12942 			ctl_scsiio_precheck(softc, &io->scsiio);
12943 			mtx_lock(&softc->ctl_lock);
12944 			continue;
12945 		}
12946 		/*
12947 		 * We might want to move this to a separate thread, so that
12948 		 * configuration requests (in this case LUN creations)
12949 		 * won't impact the I/O path.
12950 		 */
12951 		be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
12952 		if (be_lun != NULL) {
12953 			STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
12954 			mtx_unlock(&softc->ctl_lock);
12955 			ctl_create_lun(be_lun);
12956 			mtx_lock(&softc->ctl_lock);
12957 			continue;
12958 		}
12959 
12960 		/* XXX KDM use the PDROP flag?? */
12961 		/* Sleep until we have something to do. */
12962 		mtx_sleep(softc, &softc->ctl_lock, PRIBIO, "ctl_work", 0);
12963 
12964 		/* Back to the top of the loop to see what woke us up. */
12965 		continue;
12966 
12967 execute:
12968 		retval = ctl_scsiio(&io->scsiio);
12969 		switch (retval) {
12970 		case CTL_RETVAL_COMPLETE:
12971 			break;
12972 		default:
12973 			/*
12974 			 * Probably need to make sure this doesn't happen.
12975 			 */
12976 			break;
12977 		}
12978 		mtx_lock(&softc->ctl_lock);
12979 	}
12980 }
12981 
12982 void
12983 ctl_wakeup_thread()
12984 {
12985 	struct ctl_softc *softc;
12986 
12987 	softc = control_softc;
12988 
12989 	wakeup(softc);
12990 }
12991 
12992 /* Initialization and failover */
12993 
12994 void
12995 ctl_init_isc_msg(void)
12996 {
12997 	printf("CTL: Still calling this thing\n");
12998 }
12999 
13000 /*
13001  * Init component
13002  * 	Initializes component into configuration defined by bootMode
13003  *	(see hasc-sv.c)
13004  *  	returns hasc_Status:
13005  * 		OK
13006  *		ERROR - fatal error
13007  */
13008 static ctl_ha_comp_status
13009 ctl_isc_init(struct ctl_ha_component *c)
13010 {
13011 	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
13012 
13013 	c->status = ret;
13014 	return ret;
13015 }
13016 
13017 /* Start component
13018  * 	Starts component in state requested. If component starts successfully,
13019  *	it must set its own state to the requestrd state
13020  *	When requested state is HASC_STATE_HA, the component may refine it
13021  * 	by adding _SLAVE or _MASTER flags.
13022  *	Currently allowed state transitions are:
13023  *	UNKNOWN->HA		- initial startup
13024  *	UNKNOWN->SINGLE - initial startup when no parter detected
13025  *	HA->SINGLE		- failover
13026  * returns ctl_ha_comp_status:
13027  * 		OK	- component successfully started in requested state
13028  *		FAILED  - could not start the requested state, failover may
13029  * 			  be possible
13030  *		ERROR	- fatal error detected, no future startup possible
13031  */
13032 static ctl_ha_comp_status
13033 ctl_isc_start(struct ctl_ha_component *c, ctl_ha_state state)
13034 {
13035 	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
13036 
13037 	// UNKNOWN->HA or UNKNOWN->SINGLE (bootstrap)
13038 	if (c->state == CTL_HA_STATE_UNKNOWN ) {
13039 		ctl_is_single = 0;
13040 		if (ctl_ha_msg_create(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
13041 		    != CTL_HA_STATUS_SUCCESS) {
13042 			printf("ctl_isc_start: ctl_ha_msg_create failed.\n");
13043 			ret = CTL_HA_COMP_STATUS_ERROR;
13044 		}
13045 	} else if (CTL_HA_STATE_IS_HA(c->state)
13046 		&& CTL_HA_STATE_IS_SINGLE(state)){
13047 		// HA->SINGLE transition
13048 	        ctl_failover();
13049 		ctl_is_single = 1;
13050 	} else {
13051 		printf("ctl_isc_start:Invalid state transition %X->%X\n",
13052 		       c->state, state);
13053 		ret = CTL_HA_COMP_STATUS_ERROR;
13054 	}
13055 	if (CTL_HA_STATE_IS_SINGLE(state))
13056 		ctl_is_single = 1;
13057 
13058 	c->state = state;
13059 	c->status = ret;
13060 	return ret;
13061 }
13062 
13063 /*
13064  * Quiesce component
13065  * The component must clear any error conditions (set status to OK) and
13066  * prepare itself to another Start call
13067  * returns ctl_ha_comp_status:
13068  * 	OK
13069  *	ERROR
13070  */
13071 static ctl_ha_comp_status
13072 ctl_isc_quiesce(struct ctl_ha_component *c)
13073 {
13074 	int ret = CTL_HA_COMP_STATUS_OK;
13075 
13076 	ctl_pause_rtr = 1;
13077 	c->status = ret;
13078 	return ret;
13079 }
13080 
13081 struct ctl_ha_component ctl_ha_component_ctlisc =
13082 {
13083 	.name = "CTL ISC",
13084 	.state = CTL_HA_STATE_UNKNOWN,
13085 	.init = ctl_isc_init,
13086 	.start = ctl_isc_start,
13087 	.quiesce = ctl_isc_quiesce
13088 };
13089 
13090 /*
13091  *  vim: ts=8
13092  */
13093