xref: /freebsd/sys/cam/ctl/ctl.c (revision cc16dea626cf2fc80cde667ac4798065108e596c)
1 /*-
2  * Copyright (c) 2003-2009 Silicon Graphics International Corp.
3  * Copyright (c) 2012 The FreeBSD Foundation
4  * All rights reserved.
5  *
6  * Portions of this software were developed by Edward Tomasz Napierala
7  * under sponsorship from the FreeBSD Foundation.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions, and the following disclaimer,
14  *    without modification.
15  * 2. Redistributions in binary form must reproduce at minimum a disclaimer
16  *    substantially similar to the "NO WARRANTY" disclaimer below
17  *    ("Disclaimer") and any redistribution must be conditioned upon
18  *    including a substantially similar Disclaimer requirement for further
19  *    binary redistribution.
20  *
21  * NO WARRANTY
22  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
23  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
24  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR
25  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
26  * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
30  * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
31  * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
32  * POSSIBILITY OF SUCH DAMAGES.
33  *
34  * $Id: //depot/users/kenm/FreeBSD-test2/sys/cam/ctl/ctl.c#8 $
35  */
36 /*
37  * CAM Target Layer, a SCSI device emulation subsystem.
38  *
39  * Author: Ken Merry <ken@FreeBSD.org>
40  */
41 
42 #define _CTL_C
43 
44 #include <sys/cdefs.h>
45 __FBSDID("$FreeBSD$");
46 
47 #include <sys/param.h>
48 #include <sys/systm.h>
49 #include <sys/kernel.h>
50 #include <sys/types.h>
51 #include <sys/kthread.h>
52 #include <sys/bio.h>
53 #include <sys/fcntl.h>
54 #include <sys/lock.h>
55 #include <sys/module.h>
56 #include <sys/mutex.h>
57 #include <sys/condvar.h>
58 #include <sys/malloc.h>
59 #include <sys/conf.h>
60 #include <sys/ioccom.h>
61 #include <sys/queue.h>
62 #include <sys/sbuf.h>
63 #include <sys/endian.h>
64 #include <sys/sysctl.h>
65 
66 #include <cam/cam.h>
67 #include <cam/scsi/scsi_all.h>
68 #include <cam/scsi/scsi_da.h>
69 #include <cam/ctl/ctl_io.h>
70 #include <cam/ctl/ctl.h>
71 #include <cam/ctl/ctl_frontend.h>
72 #include <cam/ctl/ctl_frontend_internal.h>
73 #include <cam/ctl/ctl_util.h>
74 #include <cam/ctl/ctl_backend.h>
75 #include <cam/ctl/ctl_ioctl.h>
76 #include <cam/ctl/ctl_ha.h>
77 #include <cam/ctl/ctl_private.h>
78 #include <cam/ctl/ctl_debug.h>
79 #include <cam/ctl/ctl_scsi_all.h>
80 #include <cam/ctl/ctl_error.h>
81 
82 struct ctl_softc *control_softc = NULL;
83 
84 /*
85  * The default is to run with CTL_DONE_THREAD turned on.  Completed
86  * transactions are queued for processing by the CTL work thread.  When
87  * CTL_DONE_THREAD is not defined, completed transactions are processed in
88  * the caller's context.
89  */
90 #define CTL_DONE_THREAD
91 
92 /*
93  * Use the serial number and device ID provided by the backend, rather than
94  * making up our own.
95  */
96 #define CTL_USE_BACKEND_SN
97 
98 /*
99  * Size and alignment macros needed for Copan-specific HA hardware.  These
100  * can go away when the HA code is re-written, and uses busdma for any
101  * hardware.
102  */
103 #define	CTL_ALIGN_8B(target, source, type)				\
104 	if (((uint32_t)source & 0x7) != 0)				\
105 		target = (type)(source + (0x8 - ((uint32_t)source & 0x7)));\
106 	else								\
107 		target = (type)source;
108 
109 #define	CTL_SIZE_8B(target, size)					\
110 	if ((size & 0x7) != 0)						\
111 		target = size + (0x8 - (size & 0x7));			\
112 	else								\
113 		target = size;
114 
115 #define CTL_ALIGN_8B_MARGIN	16
116 
117 /*
118  * Template mode pages.
119  */
120 
121 /*
122  * Note that these are default values only.  The actual values will be
123  * filled in when the user does a mode sense.
124  */
125 static struct copan_power_subpage power_page_default = {
126 	/*page_code*/ PWR_PAGE_CODE | SMPH_SPF,
127 	/*subpage*/ PWR_SUBPAGE_CODE,
128 	/*page_length*/ {(sizeof(struct copan_power_subpage) - 4) & 0xff00,
129 			 (sizeof(struct copan_power_subpage) - 4) & 0x00ff},
130 	/*page_version*/ PWR_VERSION,
131 	/* total_luns */ 26,
132 	/* max_active_luns*/ PWR_DFLT_MAX_LUNS,
133 	/*reserved*/ {0, 0, 0, 0, 0, 0, 0, 0, 0,
134 		      0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
135 		      0, 0, 0, 0, 0, 0}
136 };
137 
138 static struct copan_power_subpage power_page_changeable = {
139 	/*page_code*/ PWR_PAGE_CODE | SMPH_SPF,
140 	/*subpage*/ PWR_SUBPAGE_CODE,
141 	/*page_length*/ {(sizeof(struct copan_power_subpage) - 4) & 0xff00,
142 			 (sizeof(struct copan_power_subpage) - 4) & 0x00ff},
143 	/*page_version*/ 0,
144 	/* total_luns */ 0,
145 	/* max_active_luns*/ 0,
146 	/*reserved*/ {0, 0, 0, 0, 0, 0, 0, 0, 0,
147 		      0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
148 		      0, 0, 0, 0, 0, 0}
149 };
150 
151 static struct copan_aps_subpage aps_page_default = {
152 	APS_PAGE_CODE | SMPH_SPF, //page_code
153 	APS_SUBPAGE_CODE, //subpage
154 	{(sizeof(struct copan_aps_subpage) - 4) & 0xff00,
155 	 (sizeof(struct copan_aps_subpage) - 4) & 0x00ff}, //page_length
156 	APS_VERSION, //page_version
157 	0, //lock_active
158 	{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
159 	0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
160 	0, 0, 0, 0, 0} //reserved
161 };
162 
163 static struct copan_aps_subpage aps_page_changeable = {
164 	APS_PAGE_CODE | SMPH_SPF, //page_code
165 	APS_SUBPAGE_CODE, //subpage
166 	{(sizeof(struct copan_aps_subpage) - 4) & 0xff00,
167 	 (sizeof(struct copan_aps_subpage) - 4) & 0x00ff}, //page_length
168 	0, //page_version
169 	0, //lock_active
170 	{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
171 	0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
172 	0, 0, 0, 0, 0} //reserved
173 };
174 
175 static struct copan_debugconf_subpage debugconf_page_default = {
176 	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
177 	DBGCNF_SUBPAGE_CODE,		/* subpage */
178 	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
179 	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
180 	DBGCNF_VERSION,			/* page_version */
181 	{CTL_TIME_IO_DEFAULT_SECS>>8,
182 	 CTL_TIME_IO_DEFAULT_SECS>>0},	/* ctl_time_io_secs */
183 };
184 
185 static struct copan_debugconf_subpage debugconf_page_changeable = {
186 	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
187 	DBGCNF_SUBPAGE_CODE,		/* subpage */
188 	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
189 	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
190 	0,				/* page_version */
191 	{0xff,0xff},			/* ctl_time_io_secs */
192 };
193 
194 static struct scsi_format_page format_page_default = {
195 	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
196 	/*page_length*/sizeof(struct scsi_format_page) - 2,
197 	/*tracks_per_zone*/ {0, 0},
198 	/*alt_sectors_per_zone*/ {0, 0},
199 	/*alt_tracks_per_zone*/ {0, 0},
200 	/*alt_tracks_per_lun*/ {0, 0},
201 	/*sectors_per_track*/ {(CTL_DEFAULT_SECTORS_PER_TRACK >> 8) & 0xff,
202 			        CTL_DEFAULT_SECTORS_PER_TRACK & 0xff},
203 	/*bytes_per_sector*/ {0, 0},
204 	/*interleave*/ {0, 0},
205 	/*track_skew*/ {0, 0},
206 	/*cylinder_skew*/ {0, 0},
207 	/*flags*/ SFP_HSEC,
208 	/*reserved*/ {0, 0, 0}
209 };
210 
211 static struct scsi_format_page format_page_changeable = {
212 	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
213 	/*page_length*/sizeof(struct scsi_format_page) - 2,
214 	/*tracks_per_zone*/ {0, 0},
215 	/*alt_sectors_per_zone*/ {0, 0},
216 	/*alt_tracks_per_zone*/ {0, 0},
217 	/*alt_tracks_per_lun*/ {0, 0},
218 	/*sectors_per_track*/ {0, 0},
219 	/*bytes_per_sector*/ {0, 0},
220 	/*interleave*/ {0, 0},
221 	/*track_skew*/ {0, 0},
222 	/*cylinder_skew*/ {0, 0},
223 	/*flags*/ 0,
224 	/*reserved*/ {0, 0, 0}
225 };
226 
227 static struct scsi_rigid_disk_page rigid_disk_page_default = {
228 	/*page_code*/SMS_RIGID_DISK_PAGE,
229 	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
230 	/*cylinders*/ {0, 0, 0},
231 	/*heads*/ CTL_DEFAULT_HEADS,
232 	/*start_write_precomp*/ {0, 0, 0},
233 	/*start_reduced_current*/ {0, 0, 0},
234 	/*step_rate*/ {0, 0},
235 	/*landing_zone_cylinder*/ {0, 0, 0},
236 	/*rpl*/ SRDP_RPL_DISABLED,
237 	/*rotational_offset*/ 0,
238 	/*reserved1*/ 0,
239 	/*rotation_rate*/ {(CTL_DEFAULT_ROTATION_RATE >> 8) & 0xff,
240 			   CTL_DEFAULT_ROTATION_RATE & 0xff},
241 	/*reserved2*/ {0, 0}
242 };
243 
244 static struct scsi_rigid_disk_page rigid_disk_page_changeable = {
245 	/*page_code*/SMS_RIGID_DISK_PAGE,
246 	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
247 	/*cylinders*/ {0, 0, 0},
248 	/*heads*/ 0,
249 	/*start_write_precomp*/ {0, 0, 0},
250 	/*start_reduced_current*/ {0, 0, 0},
251 	/*step_rate*/ {0, 0},
252 	/*landing_zone_cylinder*/ {0, 0, 0},
253 	/*rpl*/ 0,
254 	/*rotational_offset*/ 0,
255 	/*reserved1*/ 0,
256 	/*rotation_rate*/ {0, 0},
257 	/*reserved2*/ {0, 0}
258 };
259 
260 static struct scsi_caching_page caching_page_default = {
261 	/*page_code*/SMS_CACHING_PAGE,
262 	/*page_length*/sizeof(struct scsi_caching_page) - 2,
263 	/*flags1*/ SCP_DISC | SCP_WCE,
264 	/*ret_priority*/ 0,
265 	/*disable_pf_transfer_len*/ {0xff, 0xff},
266 	/*min_prefetch*/ {0, 0},
267 	/*max_prefetch*/ {0xff, 0xff},
268 	/*max_pf_ceiling*/ {0xff, 0xff},
269 	/*flags2*/ 0,
270 	/*cache_segments*/ 0,
271 	/*cache_seg_size*/ {0, 0},
272 	/*reserved*/ 0,
273 	/*non_cache_seg_size*/ {0, 0, 0}
274 };
275 
276 static struct scsi_caching_page caching_page_changeable = {
277 	/*page_code*/SMS_CACHING_PAGE,
278 	/*page_length*/sizeof(struct scsi_caching_page) - 2,
279 	/*flags1*/ 0,
280 	/*ret_priority*/ 0,
281 	/*disable_pf_transfer_len*/ {0, 0},
282 	/*min_prefetch*/ {0, 0},
283 	/*max_prefetch*/ {0, 0},
284 	/*max_pf_ceiling*/ {0, 0},
285 	/*flags2*/ 0,
286 	/*cache_segments*/ 0,
287 	/*cache_seg_size*/ {0, 0},
288 	/*reserved*/ 0,
289 	/*non_cache_seg_size*/ {0, 0, 0}
290 };
291 
292 static struct scsi_control_page control_page_default = {
293 	/*page_code*/SMS_CONTROL_MODE_PAGE,
294 	/*page_length*/sizeof(struct scsi_control_page) - 2,
295 	/*rlec*/0,
296 	/*queue_flags*/0,
297 	/*eca_and_aen*/0,
298 	/*reserved*/0,
299 	/*aen_holdoff_period*/{0, 0}
300 };
301 
302 static struct scsi_control_page control_page_changeable = {
303 	/*page_code*/SMS_CONTROL_MODE_PAGE,
304 	/*page_length*/sizeof(struct scsi_control_page) - 2,
305 	/*rlec*/SCP_DSENSE,
306 	/*queue_flags*/0,
307 	/*eca_and_aen*/0,
308 	/*reserved*/0,
309 	/*aen_holdoff_period*/{0, 0}
310 };
311 
312 
313 /*
314  * XXX KDM move these into the softc.
315  */
316 static int rcv_sync_msg;
317 static int persis_offset;
318 static uint8_t ctl_pause_rtr;
319 static int     ctl_is_single = 1;
320 static int     index_to_aps_page;
321 
322 SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD, 0, "CAM Target Layer");
323 
324 /*
325  * Serial number (0x80), device id (0x83), and supported pages (0x00)
326  */
327 #define SCSI_EVPD_NUM_SUPPORTED_PAGES	3
328 
329 static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
330 				  int param);
331 static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
332 static int ctl_init(void);
333 void ctl_shutdown(void);
334 static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
335 static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
336 static void ctl_ioctl_online(void *arg);
337 static void ctl_ioctl_offline(void *arg);
338 static int ctl_ioctl_targ_enable(void *arg, struct ctl_id targ_id);
339 static int ctl_ioctl_targ_disable(void *arg, struct ctl_id targ_id);
340 static int ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id);
341 static int ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id);
342 static int ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio);
343 static int ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio, int have_lock);
344 static int ctl_ioctl_submit_wait(union ctl_io *io);
345 static void ctl_ioctl_datamove(union ctl_io *io);
346 static void ctl_ioctl_done(union ctl_io *io);
347 static void ctl_ioctl_hard_startstop_callback(void *arg,
348 					      struct cfi_metatask *metatask);
349 static void ctl_ioctl_bbrread_callback(void *arg,struct cfi_metatask *metatask);
350 static int ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
351 			      struct ctl_ooa *ooa_hdr,
352 			      struct ctl_ooa_entry *kern_entries);
353 static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
354 		     struct thread *td);
355 uint32_t ctl_get_resindex(struct ctl_nexus *nexus);
356 uint32_t ctl_port_idx(int port_num);
357 #ifdef unused
358 static union ctl_io *ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port,
359 				   uint32_t targ_target, uint32_t targ_lun,
360 				   int can_wait);
361 static void ctl_kfree_io(union ctl_io *io);
362 #endif /* unused */
363 static void ctl_free_io_internal(union ctl_io *io, int have_lock);
364 static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
365 			 struct ctl_be_lun *be_lun, struct ctl_id target_id);
366 static int ctl_free_lun(struct ctl_lun *lun);
367 static void ctl_create_lun(struct ctl_be_lun *be_lun);
368 /**
369 static void ctl_failover_change_pages(struct ctl_softc *softc,
370 				      struct ctl_scsiio *ctsio, int master);
371 **/
372 
373 static int ctl_do_mode_select(union ctl_io *io);
374 static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
375 			   uint64_t res_key, uint64_t sa_res_key,
376 			   uint8_t type, uint32_t residx,
377 			   struct ctl_scsiio *ctsio,
378 			   struct scsi_per_res_out *cdb,
379 			   struct scsi_per_res_out_parms* param);
380 static void ctl_pro_preempt_other(struct ctl_lun *lun,
381 				  union ctl_ha_msg *msg);
382 static void ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg);
383 static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
384 static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
385 static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
386 static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
387 static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
388 static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint32_t *len);
389 static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2);
390 static ctl_action ctl_check_for_blockage(union ctl_io *pending_io,
391 					 union ctl_io *ooa_io);
392 static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
393 				union ctl_io *starting_io);
394 static int ctl_check_blocked(struct ctl_lun *lun);
395 static int ctl_scsiio_lun_check(struct ctl_softc *ctl_softc,
396 				struct ctl_lun *lun,
397 				struct ctl_cmd_entry *entry,
398 				struct ctl_scsiio *ctsio);
399 //static int ctl_check_rtr(union ctl_io *pending_io, struct ctl_softc *softc);
400 static void ctl_failover(void);
401 static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
402 			       struct ctl_scsiio *ctsio);
403 static int ctl_scsiio(struct ctl_scsiio *ctsio);
404 
405 static int ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
406 static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
407 			    ctl_ua_type ua_type);
408 static int ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io,
409 			 ctl_ua_type ua_type);
410 static int ctl_abort_task(union ctl_io *io);
411 static void ctl_run_task_queue(struct ctl_softc *ctl_softc);
412 #ifdef CTL_IO_DELAY
413 static void ctl_datamove_timer_wakeup(void *arg);
414 static void ctl_done_timer_wakeup(void *arg);
415 #endif /* CTL_IO_DELAY */
416 
417 static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
418 static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
419 static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
420 static void ctl_datamove_remote_write(union ctl_io *io);
421 static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
422 static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
423 static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
424 static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
425 				    ctl_ha_dt_cb callback);
426 static void ctl_datamove_remote_read(union ctl_io *io);
427 static void ctl_datamove_remote(union ctl_io *io);
428 static int ctl_process_done(union ctl_io *io, int have_lock);
429 static void ctl_work_thread(void *arg);
430 
431 /*
432  * Load the serialization table.  This isn't very pretty, but is probably
433  * the easiest way to do it.
434  */
435 #include "ctl_ser_table.c"
436 
437 /*
438  * We only need to define open, close and ioctl routines for this driver.
439  */
440 static struct cdevsw ctl_cdevsw = {
441 	.d_version =	D_VERSION,
442 	.d_flags =	0,
443 	.d_open =	ctl_open,
444 	.d_close =	ctl_close,
445 	.d_ioctl =	ctl_ioctl,
446 	.d_name =	"ctl",
447 };
448 
449 
450 MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
451 
452 static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
453 
454 static moduledata_t ctl_moduledata = {
455 	"ctl",
456 	ctl_module_event_handler,
457 	NULL
458 };
459 
460 DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
461 MODULE_VERSION(ctl, 1);
462 
463 static void
464 ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
465 			    union ctl_ha_msg *msg_info)
466 {
467 	struct ctl_scsiio *ctsio;
468 
469 	if (msg_info->hdr.original_sc == NULL) {
470 		printf("%s: original_sc == NULL!\n", __func__);
471 		/* XXX KDM now what? */
472 		return;
473 	}
474 
475 	ctsio = &msg_info->hdr.original_sc->scsiio;
476 	ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
477 	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
478 	ctsio->io_hdr.status = msg_info->hdr.status;
479 	ctsio->scsi_status = msg_info->scsi.scsi_status;
480 	ctsio->sense_len = msg_info->scsi.sense_len;
481 	ctsio->sense_residual = msg_info->scsi.sense_residual;
482 	ctsio->residual = msg_info->scsi.residual;
483 	memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
484 	       sizeof(ctsio->sense_data));
485 	memcpy(&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
486 	       &msg_info->scsi.lbalen, sizeof(msg_info->scsi.lbalen));
487 	STAILQ_INSERT_TAIL(&ctl_softc->isc_queue, &ctsio->io_hdr, links);
488 	ctl_wakeup_thread();
489 }
490 
491 static void
492 ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
493 				union ctl_ha_msg *msg_info)
494 {
495 	struct ctl_scsiio *ctsio;
496 
497 	if (msg_info->hdr.serializing_sc == NULL) {
498 		printf("%s: serializing_sc == NULL!\n", __func__);
499 		/* XXX KDM now what? */
500 		return;
501 	}
502 
503 	ctsio = &msg_info->hdr.serializing_sc->scsiio;
504 #if 0
505 	/*
506 	 * Attempt to catch the situation where an I/O has
507 	 * been freed, and we're using it again.
508 	 */
509 	if (ctsio->io_hdr.io_type == 0xff) {
510 		union ctl_io *tmp_io;
511 		tmp_io = (union ctl_io *)ctsio;
512 		printf("%s: %p use after free!\n", __func__,
513 		       ctsio);
514 		printf("%s: type %d msg %d cdb %x iptl: "
515 		       "%d:%d:%d:%d tag 0x%04x "
516 		       "flag %#x status %x\n",
517 			__func__,
518 			tmp_io->io_hdr.io_type,
519 			tmp_io->io_hdr.msg_type,
520 			tmp_io->scsiio.cdb[0],
521 			tmp_io->io_hdr.nexus.initid.id,
522 			tmp_io->io_hdr.nexus.targ_port,
523 			tmp_io->io_hdr.nexus.targ_target.id,
524 			tmp_io->io_hdr.nexus.targ_lun,
525 			(tmp_io->io_hdr.io_type ==
526 			CTL_IO_TASK) ?
527 			tmp_io->taskio.tag_num :
528 			tmp_io->scsiio.tag_num,
529 		        tmp_io->io_hdr.flags,
530 			tmp_io->io_hdr.status);
531 	}
532 #endif
533 	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
534 	STAILQ_INSERT_TAIL(&ctl_softc->isc_queue, &ctsio->io_hdr, links);
535 	ctl_wakeup_thread();
536 }
537 
538 /*
539  * ISC (Inter Shelf Communication) event handler.  Events from the HA
540  * subsystem come in here.
541  */
542 static void
543 ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
544 {
545 	struct ctl_softc *ctl_softc;
546 	union ctl_io *io;
547 	struct ctl_prio *presio;
548 	ctl_ha_status isc_status;
549 
550 	ctl_softc = control_softc;
551 	io = NULL;
552 
553 
554 #if 0
555 	printf("CTL: Isc Msg event %d\n", event);
556 #endif
557 	if (event == CTL_HA_EVT_MSG_RECV) {
558 		union ctl_ha_msg msg_info;
559 
560 		isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
561 					     sizeof(msg_info), /*wait*/ 0);
562 #if 0
563 		printf("CTL: msg_type %d\n", msg_info.msg_type);
564 #endif
565 		if (isc_status != 0) {
566 			printf("Error receiving message, status = %d\n",
567 			       isc_status);
568 			return;
569 		}
570 		mtx_lock(&ctl_softc->ctl_lock);
571 
572 		switch (msg_info.hdr.msg_type) {
573 		case CTL_MSG_SERIALIZE:
574 #if 0
575 			printf("Serialize\n");
576 #endif
577 			io = ctl_alloc_io((void *)ctl_softc->othersc_pool);
578 			if (io == NULL) {
579 				printf("ctl_isc_event_handler: can't allocate "
580 				       "ctl_io!\n");
581 				/* Bad Juju */
582 				/* Need to set busy and send msg back */
583 				mtx_unlock(&ctl_softc->ctl_lock);
584 				msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
585 				msg_info.hdr.status = CTL_SCSI_ERROR;
586 				msg_info.scsi.scsi_status = SCSI_STATUS_BUSY;
587 				msg_info.scsi.sense_len = 0;
588 			        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
589 				    sizeof(msg_info), 0) > CTL_HA_STATUS_SUCCESS){
590 				}
591 				goto bailout;
592 			}
593 			ctl_zero_io(io);
594 			// populate ctsio from msg_info
595 			io->io_hdr.io_type = CTL_IO_SCSI;
596 			io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
597 			io->io_hdr.original_sc = msg_info.hdr.original_sc;
598 #if 0
599 			printf("pOrig %x\n", (int)msg_info.original_sc);
600 #endif
601 			io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
602 					    CTL_FLAG_IO_ACTIVE;
603 			/*
604 			 * If we're in serialization-only mode, we don't
605 			 * want to go through full done processing.  Thus
606 			 * the COPY flag.
607 			 *
608 			 * XXX KDM add another flag that is more specific.
609 			 */
610 			if (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)
611 				io->io_hdr.flags |= CTL_FLAG_INT_COPY;
612 			io->io_hdr.nexus = msg_info.hdr.nexus;
613 #if 0
614 			printf("targ %d, port %d, iid %d, lun %d\n",
615 			       io->io_hdr.nexus.targ_target.id,
616 			       io->io_hdr.nexus.targ_port,
617 			       io->io_hdr.nexus.initid.id,
618 			       io->io_hdr.nexus.targ_lun);
619 #endif
620 			io->scsiio.tag_num = msg_info.scsi.tag_num;
621 			io->scsiio.tag_type = msg_info.scsi.tag_type;
622 			memcpy(io->scsiio.cdb, msg_info.scsi.cdb,
623 			       CTL_MAX_CDBLEN);
624 			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
625 				struct ctl_cmd_entry *entry;
626 				uint8_t opcode;
627 
628 				opcode = io->scsiio.cdb[0];
629 				entry = &ctl_cmd_table[opcode];
630 				io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
631 				io->io_hdr.flags |=
632 					entry->flags & CTL_FLAG_DATA_MASK;
633 			}
634 			STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
635 					   &io->io_hdr, links);
636 			ctl_wakeup_thread();
637 			break;
638 
639 		/* Performed on the Originating SC, XFER mode only */
640 		case CTL_MSG_DATAMOVE: {
641 			struct ctl_sg_entry *sgl;
642 			int i, j;
643 
644 			io = msg_info.hdr.original_sc;
645 			if (io == NULL) {
646 				printf("%s: original_sc == NULL!\n", __func__);
647 				/* XXX KDM do something here */
648 				break;
649 			}
650 			io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
651 			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
652 			/*
653 			 * Keep track of this, we need to send it back over
654 			 * when the datamove is complete.
655 			 */
656 			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
657 
658 			if (msg_info.dt.sg_sequence == 0) {
659 				/*
660 				 * XXX KDM we use the preallocated S/G list
661 				 * here, but we'll need to change this to
662 				 * dynamic allocation if we need larger S/G
663 				 * lists.
664 				 */
665 				if (msg_info.dt.kern_sg_entries >
666 				    sizeof(io->io_hdr.remote_sglist) /
667 				    sizeof(io->io_hdr.remote_sglist[0])) {
668 					printf("%s: number of S/G entries "
669 					    "needed %u > allocated num %zd\n",
670 					    __func__,
671 					    msg_info.dt.kern_sg_entries,
672 					    sizeof(io->io_hdr.remote_sglist)/
673 					    sizeof(io->io_hdr.remote_sglist[0]));
674 
675 					/*
676 					 * XXX KDM send a message back to
677 					 * the other side to shut down the
678 					 * DMA.  The error will come back
679 					 * through via the normal channel.
680 					 */
681 					break;
682 				}
683 				sgl = io->io_hdr.remote_sglist;
684 				memset(sgl, 0,
685 				       sizeof(io->io_hdr.remote_sglist));
686 
687 				io->scsiio.kern_data_ptr = (uint8_t *)sgl;
688 
689 				io->scsiio.kern_sg_entries =
690 					msg_info.dt.kern_sg_entries;
691 				io->scsiio.rem_sg_entries =
692 					msg_info.dt.kern_sg_entries;
693 				io->scsiio.kern_data_len =
694 					msg_info.dt.kern_data_len;
695 				io->scsiio.kern_total_len =
696 					msg_info.dt.kern_total_len;
697 				io->scsiio.kern_data_resid =
698 					msg_info.dt.kern_data_resid;
699 				io->scsiio.kern_rel_offset =
700 					msg_info.dt.kern_rel_offset;
701 				/*
702 				 * Clear out per-DMA flags.
703 				 */
704 				io->io_hdr.flags &= ~CTL_FLAG_RDMA_MASK;
705 				/*
706 				 * Add per-DMA flags that are set for this
707 				 * particular DMA request.
708 				 */
709 				io->io_hdr.flags |= msg_info.dt.flags &
710 						    CTL_FLAG_RDMA_MASK;
711 			} else
712 				sgl = (struct ctl_sg_entry *)
713 					io->scsiio.kern_data_ptr;
714 
715 			for (i = msg_info.dt.sent_sg_entries, j = 0;
716 			     i < (msg_info.dt.sent_sg_entries +
717 			     msg_info.dt.cur_sg_entries); i++, j++) {
718 				sgl[i].addr = msg_info.dt.sg_list[j].addr;
719 				sgl[i].len = msg_info.dt.sg_list[j].len;
720 
721 #if 0
722 				printf("%s: L: %p,%d -> %p,%d j=%d, i=%d\n",
723 				       __func__,
724 				       msg_info.dt.sg_list[j].addr,
725 				       msg_info.dt.sg_list[j].len,
726 				       sgl[i].addr, sgl[i].len, j, i);
727 #endif
728 			}
729 #if 0
730 			memcpy(&sgl[msg_info.dt.sent_sg_entries],
731 			       msg_info.dt.sg_list,
732 			       sizeof(*sgl) * msg_info.dt.cur_sg_entries);
733 #endif
734 
735 			/*
736 			 * If this is the last piece of the I/O, we've got
737 			 * the full S/G list.  Queue processing in the thread.
738 			 * Otherwise wait for the next piece.
739 			 */
740 			if (msg_info.dt.sg_last != 0) {
741 				STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
742 						   &io->io_hdr, links);
743 				ctl_wakeup_thread();
744 			}
745 			break;
746 		}
747 		/* Performed on the Serializing (primary) SC, XFER mode only */
748 		case CTL_MSG_DATAMOVE_DONE: {
749 			if (msg_info.hdr.serializing_sc == NULL) {
750 				printf("%s: serializing_sc == NULL!\n",
751 				       __func__);
752 				/* XXX KDM now what? */
753 				break;
754 			}
755 			/*
756 			 * We grab the sense information here in case
757 			 * there was a failure, so we can return status
758 			 * back to the initiator.
759 			 */
760 			io = msg_info.hdr.serializing_sc;
761 			io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
762 			io->io_hdr.status = msg_info.hdr.status;
763 			io->scsiio.scsi_status = msg_info.scsi.scsi_status;
764 			io->scsiio.sense_len = msg_info.scsi.sense_len;
765 			io->scsiio.sense_residual =msg_info.scsi.sense_residual;
766 			io->io_hdr.port_status = msg_info.scsi.fetd_status;
767 			io->scsiio.residual = msg_info.scsi.residual;
768 			memcpy(&io->scsiio.sense_data,&msg_info.scsi.sense_data,
769 			       sizeof(io->scsiio.sense_data));
770 
771 			STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
772 					   &io->io_hdr, links);
773 			ctl_wakeup_thread();
774 			break;
775 		}
776 
777 		/* Preformed on Originating SC, SER_ONLY mode */
778 		case CTL_MSG_R2R:
779 			io = msg_info.hdr.original_sc;
780 			if (io == NULL) {
781 				printf("%s: Major Bummer\n", __func__);
782 				mtx_unlock(&ctl_softc->ctl_lock);
783 				return;
784 			} else {
785 #if 0
786 				printf("pOrig %x\n",(int) ctsio);
787 #endif
788 			}
789 			io->io_hdr.msg_type = CTL_MSG_R2R;
790 			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
791 			STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
792 					   &io->io_hdr, links);
793 			ctl_wakeup_thread();
794 			break;
795 
796 		/*
797 		 * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
798 		 * mode.
799 		 * Performed on the Originating (i.e. secondary) SC in XFER
800 		 * mode
801 		 */
802 		case CTL_MSG_FINISH_IO:
803 			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER)
804 				ctl_isc_handler_finish_xfer(ctl_softc,
805 							    &msg_info);
806 			else
807 				ctl_isc_handler_finish_ser_only(ctl_softc,
808 								&msg_info);
809 			break;
810 
811 		/* Preformed on Originating SC */
812 		case CTL_MSG_BAD_JUJU:
813 			io = msg_info.hdr.original_sc;
814 			if (io == NULL) {
815 				printf("%s: Bad JUJU!, original_sc is NULL!\n",
816 				       __func__);
817 				break;
818 			}
819 			ctl_copy_sense_data(&msg_info, io);
820 			/*
821 			 * IO should have already been cleaned up on other
822 			 * SC so clear this flag so we won't send a message
823 			 * back to finish the IO there.
824 			 */
825 			io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
826 			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
827 
828 			/* io = msg_info.hdr.serializing_sc; */
829 			io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
830 		        STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
831 					   &io->io_hdr, links);
832 			ctl_wakeup_thread();
833 			break;
834 
835 		/* Handle resets sent from the other side */
836 		case CTL_MSG_MANAGE_TASKS: {
837 			struct ctl_taskio *taskio;
838 			taskio = (struct ctl_taskio *)ctl_alloc_io(
839 				(void *)ctl_softc->othersc_pool);
840 			if (taskio == NULL) {
841 				printf("ctl_isc_event_handler: can't allocate "
842 				       "ctl_io!\n");
843 				/* Bad Juju */
844 				/* should I just call the proper reset func
845 				   here??? */
846 				mtx_unlock(&ctl_softc->ctl_lock);
847 				goto bailout;
848 			}
849 			ctl_zero_io((union ctl_io *)taskio);
850 			taskio->io_hdr.io_type = CTL_IO_TASK;
851 			taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
852 			taskio->io_hdr.nexus = msg_info.hdr.nexus;
853 			taskio->task_action = msg_info.task.task_action;
854 			taskio->tag_num = msg_info.task.tag_num;
855 			taskio->tag_type = msg_info.task.tag_type;
856 #ifdef CTL_TIME_IO
857 			taskio->io_hdr.start_time = time_uptime;
858 			getbintime(&taskio->io_hdr.start_bt);
859 #if 0
860 			cs_prof_gettime(&taskio->io_hdr.start_ticks);
861 #endif
862 #endif /* CTL_TIME_IO */
863 		        STAILQ_INSERT_TAIL(&ctl_softc->task_queue,
864 					   &taskio->io_hdr, links);
865 			ctl_softc->flags |= CTL_FLAG_TASK_PENDING;
866 			ctl_wakeup_thread();
867 			break;
868 		}
869 		/* Persistent Reserve action which needs attention */
870 		case CTL_MSG_PERS_ACTION:
871 			presio = (struct ctl_prio *)ctl_alloc_io(
872 				(void *)ctl_softc->othersc_pool);
873 			if (presio == NULL) {
874 				printf("ctl_isc_event_handler: can't allocate "
875 				       "ctl_io!\n");
876 				/* Bad Juju */
877 				/* Need to set busy and send msg back */
878 				mtx_unlock(&ctl_softc->ctl_lock);
879 				goto bailout;
880 			}
881 			ctl_zero_io((union ctl_io *)presio);
882 			presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
883 			presio->pr_msg = msg_info.pr;
884 		        STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
885 					   &presio->io_hdr, links);
886 			ctl_wakeup_thread();
887 			break;
888 		case CTL_MSG_SYNC_FE:
889 			rcv_sync_msg = 1;
890 			break;
891 		case CTL_MSG_APS_LOCK: {
892 			// It's quicker to execute this then to
893 			// queue it.
894 			struct ctl_lun *lun;
895 			struct ctl_page_index *page_index;
896 			struct copan_aps_subpage *current_sp;
897 
898 			lun = ctl_softc->ctl_luns[msg_info.hdr.nexus.targ_lun];
899 			page_index = &lun->mode_pages.index[index_to_aps_page];
900 			current_sp = (struct copan_aps_subpage *)
901 				     (page_index->page_data +
902 				     (page_index->page_len * CTL_PAGE_CURRENT));
903 
904 			current_sp->lock_active = msg_info.aps.lock_flag;
905 		        break;
906 		}
907 		default:
908 		        printf("How did I get here?\n");
909 		}
910 		mtx_unlock(&ctl_softc->ctl_lock);
911 	} else if (event == CTL_HA_EVT_MSG_SENT) {
912 		if (param != CTL_HA_STATUS_SUCCESS) {
913 			printf("Bad status from ctl_ha_msg_send status %d\n",
914 			       param);
915 		}
916 		return;
917 	} else if (event == CTL_HA_EVT_DISCONNECT) {
918 		printf("CTL: Got a disconnect from Isc\n");
919 		return;
920 	} else {
921 		printf("ctl_isc_event_handler: Unknown event %d\n", event);
922 		return;
923 	}
924 
925 bailout:
926 	return;
927 }
928 
929 static void
930 ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
931 {
932 	struct scsi_sense_data *sense;
933 
934 	sense = &dest->scsiio.sense_data;
935 	bcopy(&src->scsi.sense_data, sense, sizeof(*sense));
936 	dest->scsiio.scsi_status = src->scsi.scsi_status;
937 	dest->scsiio.sense_len = src->scsi.sense_len;
938 	dest->io_hdr.status = src->hdr.status;
939 }
940 
941 static int
942 ctl_init(void)
943 {
944 	struct ctl_softc *softc;
945 	struct ctl_io_pool *internal_pool, *emergency_pool, *other_pool;
946 	struct ctl_frontend *fe;
947 	struct ctl_lun *lun;
948         uint8_t sc_id =0;
949 #if 0
950 	int i;
951 #endif
952 	int error, retval;
953 	//int isc_retval;
954 
955 	retval = 0;
956 	ctl_pause_rtr = 0;
957         rcv_sync_msg = 0;
958 
959 	control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
960 			       M_WAITOK | M_ZERO);
961 	softc = control_softc;
962 
963 	softc->dev = make_dev(&ctl_cdevsw, 0, UID_ROOT, GID_OPERATOR, 0600,
964 			      "cam/ctl");
965 
966 	softc->dev->si_drv1 = softc;
967 
968 	/*
969 	 * By default, return a "bad LUN" peripheral qualifier for unknown
970 	 * LUNs.  The user can override this default using the tunable or
971 	 * sysctl.  See the comment in ctl_inquiry_std() for more details.
972 	 */
973 	softc->inquiry_pq_no_lun = 1;
974 	TUNABLE_INT_FETCH("kern.cam.ctl.inquiry_pq_no_lun",
975 			  &softc->inquiry_pq_no_lun);
976 	sysctl_ctx_init(&softc->sysctl_ctx);
977 	softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
978 		SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
979 		CTLFLAG_RD, 0, "CAM Target Layer");
980 
981 	if (softc->sysctl_tree == NULL) {
982 		printf("%s: unable to allocate sysctl tree\n", __func__);
983 		destroy_dev(softc->dev);
984 		free(control_softc, M_DEVBUF);
985 		control_softc = NULL;
986 		return (ENOMEM);
987 	}
988 
989 	SYSCTL_ADD_INT(&softc->sysctl_ctx,
990 		       SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO,
991 		       "inquiry_pq_no_lun", CTLFLAG_RW,
992 		       &softc->inquiry_pq_no_lun, 0,
993 		       "Report no lun possible for invalid LUNs");
994 
995 	mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
996 	softc->open_count = 0;
997 
998 	/*
999 	 * Default to actually sending a SYNCHRONIZE CACHE command down to
1000 	 * the drive.
1001 	 */
1002 	softc->flags = CTL_FLAG_REAL_SYNC;
1003 
1004 	/*
1005 	 * In Copan's HA scheme, the "master" and "slave" roles are
1006 	 * figured out through the slot the controller is in.  Although it
1007 	 * is an active/active system, someone has to be in charge.
1008  	 */
1009 #ifdef NEEDTOPORT
1010         scmicro_rw(SCMICRO_GET_SHELF_ID, &sc_id);
1011 #endif
1012 
1013         if (sc_id == 0) {
1014 		softc->flags |= CTL_FLAG_MASTER_SHELF;
1015 		persis_offset = 0;
1016 	} else
1017 		persis_offset = CTL_MAX_INITIATORS;
1018 
1019 	/*
1020 	 * XXX KDM need to figure out where we want to get our target ID
1021 	 * and WWID.  Is it different on each port?
1022 	 */
1023 	softc->target.id = 0;
1024 	softc->target.wwid[0] = 0x12345678;
1025 	softc->target.wwid[1] = 0x87654321;
1026 	STAILQ_INIT(&softc->lun_list);
1027 	STAILQ_INIT(&softc->pending_lun_queue);
1028 	STAILQ_INIT(&softc->task_queue);
1029 	STAILQ_INIT(&softc->incoming_queue);
1030 	STAILQ_INIT(&softc->rtr_queue);
1031 	STAILQ_INIT(&softc->done_queue);
1032 	STAILQ_INIT(&softc->isc_queue);
1033 	STAILQ_INIT(&softc->fe_list);
1034 	STAILQ_INIT(&softc->be_list);
1035 	STAILQ_INIT(&softc->io_pools);
1036 
1037 	lun = &softc->lun;
1038 
1039 	/*
1040 	 * We don't bother calling these with ctl_lock held here, because,
1041 	 * in theory, no one else can try to do anything while we're in our
1042 	 * module init routine.
1043 	 */
1044 	if (ctl_pool_create(softc, CTL_POOL_INTERNAL, CTL_POOL_ENTRIES_INTERNAL,
1045 			    &internal_pool)!= 0){
1046 		printf("ctl: can't allocate %d entry internal pool, "
1047 		       "exiting\n", CTL_POOL_ENTRIES_INTERNAL);
1048 		return (ENOMEM);
1049 	}
1050 
1051 	if (ctl_pool_create(softc, CTL_POOL_EMERGENCY,
1052 			    CTL_POOL_ENTRIES_EMERGENCY, &emergency_pool) != 0) {
1053 		printf("ctl: can't allocate %d entry emergency pool, "
1054 		       "exiting\n", CTL_POOL_ENTRIES_EMERGENCY);
1055 		ctl_pool_free(softc, internal_pool);
1056 		return (ENOMEM);
1057 	}
1058 
1059 	if (ctl_pool_create(softc, CTL_POOL_4OTHERSC, CTL_POOL_ENTRIES_OTHER_SC,
1060 	                    &other_pool) != 0)
1061 	{
1062 		printf("ctl: can't allocate %d entry other SC pool, "
1063 		       "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1064 		ctl_pool_free(softc, internal_pool);
1065 		ctl_pool_free(softc, emergency_pool);
1066 		return (ENOMEM);
1067 	}
1068 
1069 	softc->internal_pool = internal_pool;
1070 	softc->emergency_pool = emergency_pool;
1071 	softc->othersc_pool = other_pool;
1072 
1073 	mtx_lock(&softc->ctl_lock);
1074 	ctl_pool_acquire(internal_pool);
1075 	ctl_pool_acquire(emergency_pool);
1076 	ctl_pool_acquire(other_pool);
1077 	mtx_unlock(&softc->ctl_lock);
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 	error = kproc_create(ctl_work_thread, softc, &softc->work_thread, 0, 0,
1090 			 "ctl_thrd");
1091 	if (error != 0) {
1092 		printf("error creating CTL work thread!\n");
1093 		mtx_lock(&softc->ctl_lock);
1094 		ctl_free_lun(lun);
1095 		ctl_pool_free(softc, internal_pool);
1096 		ctl_pool_free(softc, emergency_pool);
1097 		ctl_pool_free(softc, other_pool);
1098 		mtx_unlock(&softc->ctl_lock);
1099 		return (error);
1100 	}
1101 	printf("ctl: CAM Target Layer loaded\n");
1102 
1103 	/*
1104 	 * Initialize the initiator and portname mappings
1105 	 */
1106 	memset(softc->wwpn_iid, 0, sizeof(softc->wwpn_iid));
1107 
1108 	/*
1109 	 * Initialize the ioctl front end.
1110 	 */
1111 	fe = &softc->ioctl_info.fe;
1112 	sprintf(softc->ioctl_info.port_name, "CTL ioctl");
1113 	fe->port_type = CTL_PORT_IOCTL;
1114 	fe->num_requested_ctl_io = 100;
1115 	fe->port_name = softc->ioctl_info.port_name;
1116 	fe->port_online = ctl_ioctl_online;
1117 	fe->port_offline = ctl_ioctl_offline;
1118 	fe->onoff_arg = &softc->ioctl_info;
1119 	fe->targ_enable = ctl_ioctl_targ_enable;
1120 	fe->targ_disable = ctl_ioctl_targ_disable;
1121 	fe->lun_enable = ctl_ioctl_lun_enable;
1122 	fe->lun_disable = ctl_ioctl_lun_disable;
1123 	fe->targ_lun_arg = &softc->ioctl_info;
1124 	fe->fe_datamove = ctl_ioctl_datamove;
1125 	fe->fe_done = ctl_ioctl_done;
1126 	fe->max_targets = 15;
1127 	fe->max_target_id = 15;
1128 
1129 	if (ctl_frontend_register(&softc->ioctl_info.fe,
1130 	                  (softc->flags & CTL_FLAG_MASTER_SHELF)) != 0) {
1131 		printf("ctl: ioctl front end registration failed, will "
1132 		       "continue anyway\n");
1133 	}
1134 
1135 #ifdef CTL_IO_DELAY
1136 	if (sizeof(struct callout) > CTL_TIMER_BYTES) {
1137 		printf("sizeof(struct callout) %zd > CTL_TIMER_BYTES %zd\n",
1138 		       sizeof(struct callout), CTL_TIMER_BYTES);
1139 		return (EINVAL);
1140 	}
1141 #endif /* CTL_IO_DELAY */
1142 
1143 	return (0);
1144 }
1145 
1146 void
1147 ctl_shutdown(void)
1148 {
1149 	struct ctl_softc *softc;
1150 	struct ctl_lun *lun, *next_lun;
1151 	struct ctl_io_pool *pool, *next_pool;
1152 
1153 	softc = (struct ctl_softc *)control_softc;
1154 
1155 	if (ctl_frontend_deregister(&softc->ioctl_info.fe) != 0)
1156 		printf("ctl: ioctl front end deregistration failed\n");
1157 
1158 	mtx_lock(&softc->ctl_lock);
1159 
1160 	/*
1161 	 * Free up each LUN.
1162 	 */
1163 	for (lun = STAILQ_FIRST(&softc->lun_list); lun != NULL; lun = next_lun){
1164 		next_lun = STAILQ_NEXT(lun, links);
1165 		ctl_free_lun(lun);
1166 	}
1167 
1168 	/*
1169 	 * This will rip the rug out from under any FETDs or anyone else
1170 	 * that has a pool allocated.  Since we increment our module
1171 	 * refcount any time someone outside the main CTL module allocates
1172 	 * a pool, we shouldn't have any problems here.  The user won't be
1173 	 * able to unload the CTL module until client modules have
1174 	 * successfully unloaded.
1175 	 */
1176 	for (pool = STAILQ_FIRST(&softc->io_pools); pool != NULL;
1177 	     pool = next_pool) {
1178 		next_pool = STAILQ_NEXT(pool, links);
1179 		ctl_pool_free(softc, pool);
1180 	}
1181 
1182 	mtx_unlock(&softc->ctl_lock);
1183 
1184 #if 0
1185 	ctl_shutdown_thread(softc->work_thread);
1186 #endif
1187 
1188 	mtx_destroy(&softc->ctl_lock);
1189 
1190 	destroy_dev(softc->dev);
1191 
1192 	sysctl_ctx_free(&softc->sysctl_ctx);
1193 
1194 	free(control_softc, M_DEVBUF);
1195 	control_softc = NULL;
1196 
1197 	printf("ctl: CAM Target Layer unloaded\n");
1198 }
1199 
1200 static int
1201 ctl_module_event_handler(module_t mod, int what, void *arg)
1202 {
1203 
1204 	switch (what) {
1205 	case MOD_LOAD:
1206 		return (ctl_init());
1207 	case MOD_UNLOAD:
1208 		return (EBUSY);
1209 	default:
1210 		return (EOPNOTSUPP);
1211 	}
1212 }
1213 
1214 /*
1215  * XXX KDM should we do some access checks here?  Bump a reference count to
1216  * prevent a CTL module from being unloaded while someone has it open?
1217  */
1218 static int
1219 ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1220 {
1221 	return (0);
1222 }
1223 
1224 static int
1225 ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1226 {
1227 	return (0);
1228 }
1229 
1230 int
1231 ctl_port_enable(ctl_port_type port_type)
1232 {
1233 	struct ctl_softc *softc;
1234 	struct ctl_frontend *fe;
1235 
1236 	if (ctl_is_single == 0) {
1237 		union ctl_ha_msg msg_info;
1238 		int isc_retval;
1239 
1240 #if 0
1241 		printf("%s: HA mode, synchronizing frontend enable\n",
1242 		        __func__);
1243 #endif
1244 		msg_info.hdr.msg_type = CTL_MSG_SYNC_FE;
1245 	        if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1246 		        sizeof(msg_info), 1 )) > CTL_HA_STATUS_SUCCESS) {
1247 			printf("Sync msg send error retval %d\n", isc_retval);
1248 		}
1249 		if (!rcv_sync_msg) {
1250 			isc_retval=ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
1251 			        sizeof(msg_info), 1);
1252 		}
1253 #if 0
1254         	printf("CTL:Frontend Enable\n");
1255 	} else {
1256 		printf("%s: single mode, skipping frontend synchronization\n",
1257 		        __func__);
1258 #endif
1259 	}
1260 
1261 	softc = control_softc;
1262 
1263 	STAILQ_FOREACH(fe, &softc->fe_list, links) {
1264 		if (port_type & fe->port_type)
1265 		{
1266 #if 0
1267 			printf("port %d\n", fe->targ_port);
1268 #endif
1269 			ctl_frontend_online(fe);
1270 		}
1271 	}
1272 
1273 	return (0);
1274 }
1275 
1276 int
1277 ctl_port_disable(ctl_port_type port_type)
1278 {
1279 	struct ctl_softc *softc;
1280 	struct ctl_frontend *fe;
1281 
1282 	softc = control_softc;
1283 
1284 	STAILQ_FOREACH(fe, &softc->fe_list, links) {
1285 		if (port_type & fe->port_type)
1286 			ctl_frontend_offline(fe);
1287 	}
1288 
1289 	return (0);
1290 }
1291 
1292 /*
1293  * Returns 0 for success, 1 for failure.
1294  * Currently the only failure mode is if there aren't enough entries
1295  * allocated.  So, in case of a failure, look at num_entries_dropped,
1296  * reallocate and try again.
1297  */
1298 int
1299 ctl_port_list(struct ctl_port_entry *entries, int num_entries_alloced,
1300 	      int *num_entries_filled, int *num_entries_dropped,
1301 	      ctl_port_type port_type, int no_virtual)
1302 {
1303 	struct ctl_softc *softc;
1304 	struct ctl_frontend *fe;
1305 	int entries_dropped, entries_filled;
1306 	int retval;
1307 	int i;
1308 
1309 	softc = control_softc;
1310 
1311 	retval = 0;
1312 	entries_filled = 0;
1313 	entries_dropped = 0;
1314 
1315 	i = 0;
1316 	mtx_lock(&softc->ctl_lock);
1317 	STAILQ_FOREACH(fe, &softc->fe_list, links) {
1318 		struct ctl_port_entry *entry;
1319 
1320 		if ((fe->port_type & port_type) == 0)
1321 			continue;
1322 
1323 		if ((no_virtual != 0)
1324 		 && (fe->virtual_port != 0))
1325 			continue;
1326 
1327 		if (entries_filled >= num_entries_alloced) {
1328 			entries_dropped++;
1329 			continue;
1330 		}
1331 		entry = &entries[i];
1332 
1333 		entry->port_type = fe->port_type;
1334 		strlcpy(entry->port_name, fe->port_name,
1335 			sizeof(entry->port_name));
1336 		entry->physical_port = fe->physical_port;
1337 		entry->virtual_port = fe->virtual_port;
1338 		entry->wwnn = fe->wwnn;
1339 		entry->wwpn = fe->wwpn;
1340 
1341 		i++;
1342 		entries_filled++;
1343 	}
1344 
1345 	mtx_unlock(&softc->ctl_lock);
1346 
1347 	if (entries_dropped > 0)
1348 		retval = 1;
1349 
1350 	*num_entries_dropped = entries_dropped;
1351 	*num_entries_filled = entries_filled;
1352 
1353 	return (retval);
1354 }
1355 
1356 static void
1357 ctl_ioctl_online(void *arg)
1358 {
1359 	struct ctl_ioctl_info *ioctl_info;
1360 
1361 	ioctl_info = (struct ctl_ioctl_info *)arg;
1362 
1363 	ioctl_info->flags |= CTL_IOCTL_FLAG_ENABLED;
1364 }
1365 
1366 static void
1367 ctl_ioctl_offline(void *arg)
1368 {
1369 	struct ctl_ioctl_info *ioctl_info;
1370 
1371 	ioctl_info = (struct ctl_ioctl_info *)arg;
1372 
1373 	ioctl_info->flags &= ~CTL_IOCTL_FLAG_ENABLED;
1374 }
1375 
1376 /*
1377  * Remove an initiator by port number and initiator ID.
1378  * Returns 0 for success, 1 for failure.
1379  */
1380 int
1381 ctl_remove_initiator(int32_t targ_port, uint32_t iid)
1382 {
1383 	struct ctl_softc *softc;
1384 
1385 	softc = control_softc;
1386 
1387 	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1388 
1389 	if ((targ_port < 0)
1390 	 || (targ_port > CTL_MAX_PORTS)) {
1391 		printf("%s: invalid port number %d\n", __func__, targ_port);
1392 		return (1);
1393 	}
1394 	if (iid > CTL_MAX_INIT_PER_PORT) {
1395 		printf("%s: initiator ID %u > maximun %u!\n",
1396 		       __func__, iid, CTL_MAX_INIT_PER_PORT);
1397 		return (1);
1398 	}
1399 
1400 	mtx_lock(&softc->ctl_lock);
1401 
1402 	softc->wwpn_iid[targ_port][iid].in_use = 0;
1403 
1404 	mtx_unlock(&softc->ctl_lock);
1405 
1406 	return (0);
1407 }
1408 
1409 /*
1410  * Add an initiator to the initiator map.
1411  * Returns 0 for success, 1 for failure.
1412  */
1413 int
1414 ctl_add_initiator(uint64_t wwpn, int32_t targ_port, uint32_t iid)
1415 {
1416 	struct ctl_softc *softc;
1417 	int retval;
1418 
1419 	softc = control_softc;
1420 
1421 	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1422 
1423 	retval = 0;
1424 
1425 	if ((targ_port < 0)
1426 	 || (targ_port > CTL_MAX_PORTS)) {
1427 		printf("%s: invalid port number %d\n", __func__, targ_port);
1428 		return (1);
1429 	}
1430 	if (iid > CTL_MAX_INIT_PER_PORT) {
1431 		printf("%s: WWPN %#jx initiator ID %u > maximun %u!\n",
1432 		       __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
1433 		return (1);
1434 	}
1435 
1436 	mtx_lock(&softc->ctl_lock);
1437 
1438 	if (softc->wwpn_iid[targ_port][iid].in_use != 0) {
1439 		/*
1440 		 * We don't treat this as an error.
1441 		 */
1442 		if (softc->wwpn_iid[targ_port][iid].wwpn == wwpn) {
1443 			printf("%s: port %d iid %u WWPN %#jx arrived again?\n",
1444 			       __func__, targ_port, iid, (uintmax_t)wwpn);
1445 			goto bailout;
1446 		}
1447 
1448 		/*
1449 		 * This is an error, but what do we do about it?  The
1450 		 * driver is telling us we have a new WWPN for this
1451 		 * initiator ID, so we pretty much need to use it.
1452 		 */
1453 		printf("%s: port %d iid %u WWPN %#jx arrived, WWPN %#jx is "
1454 		       "still at that address\n", __func__, targ_port, iid,
1455 		       (uintmax_t)wwpn,
1456 		       (uintmax_t)softc->wwpn_iid[targ_port][iid].wwpn);
1457 
1458 		/*
1459 		 * XXX KDM clear have_ca and ua_pending on each LUN for
1460 		 * this initiator.
1461 		 */
1462 	}
1463 	softc->wwpn_iid[targ_port][iid].in_use = 1;
1464 	softc->wwpn_iid[targ_port][iid].iid = iid;
1465 	softc->wwpn_iid[targ_port][iid].wwpn = wwpn;
1466 	softc->wwpn_iid[targ_port][iid].port = targ_port;
1467 
1468 bailout:
1469 
1470 	mtx_unlock(&softc->ctl_lock);
1471 
1472 	return (retval);
1473 }
1474 
1475 /*
1476  * XXX KDM should we pretend to do something in the target/lun
1477  * enable/disable functions?
1478  */
1479 static int
1480 ctl_ioctl_targ_enable(void *arg, struct ctl_id targ_id)
1481 {
1482 	return (0);
1483 }
1484 
1485 static int
1486 ctl_ioctl_targ_disable(void *arg, struct ctl_id targ_id)
1487 {
1488 	return (0);
1489 }
1490 
1491 static int
1492 ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id)
1493 {
1494 	return (0);
1495 }
1496 
1497 static int
1498 ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id)
1499 {
1500 	return (0);
1501 }
1502 
1503 /*
1504  * Data movement routine for the CTL ioctl frontend port.
1505  */
1506 static int
1507 ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio)
1508 {
1509 	struct ctl_sg_entry *ext_sglist, *kern_sglist;
1510 	struct ctl_sg_entry ext_entry, kern_entry;
1511 	int ext_sglen, ext_sg_entries, kern_sg_entries;
1512 	int ext_sg_start, ext_offset;
1513 	int len_to_copy, len_copied;
1514 	int kern_watermark, ext_watermark;
1515 	int ext_sglist_malloced;
1516 	int i, j;
1517 
1518 	ext_sglist_malloced = 0;
1519 	ext_sg_start = 0;
1520 	ext_offset = 0;
1521 
1522 	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove\n"));
1523 
1524 	/*
1525 	 * If this flag is set, fake the data transfer.
1526 	 */
1527 	if (ctsio->io_hdr.flags & CTL_FLAG_NO_DATAMOVE) {
1528 		ctsio->ext_data_filled = ctsio->ext_data_len;
1529 		goto bailout;
1530 	}
1531 
1532 	/*
1533 	 * To simplify things here, if we have a single buffer, stick it in
1534 	 * a S/G entry and just make it a single entry S/G list.
1535 	 */
1536 	if (ctsio->io_hdr.flags & CTL_FLAG_EDPTR_SGLIST) {
1537 		int len_seen;
1538 
1539 		ext_sglen = ctsio->ext_sg_entries * sizeof(*ext_sglist);
1540 
1541 		ext_sglist = (struct ctl_sg_entry *)malloc(ext_sglen, M_CTL,
1542 							   M_WAITOK);
1543 		ext_sglist_malloced = 1;
1544 		if (copyin(ctsio->ext_data_ptr, ext_sglist,
1545 				   ext_sglen) != 0) {
1546 			ctl_set_internal_failure(ctsio,
1547 						 /*sks_valid*/ 0,
1548 						 /*retry_count*/ 0);
1549 			goto bailout;
1550 		}
1551 		ext_sg_entries = ctsio->ext_sg_entries;
1552 		len_seen = 0;
1553 		for (i = 0; i < ext_sg_entries; i++) {
1554 			if ((len_seen + ext_sglist[i].len) >=
1555 			     ctsio->ext_data_filled) {
1556 				ext_sg_start = i;
1557 				ext_offset = ctsio->ext_data_filled - len_seen;
1558 				break;
1559 			}
1560 			len_seen += ext_sglist[i].len;
1561 		}
1562 	} else {
1563 		ext_sglist = &ext_entry;
1564 		ext_sglist->addr = ctsio->ext_data_ptr;
1565 		ext_sglist->len = ctsio->ext_data_len;
1566 		ext_sg_entries = 1;
1567 		ext_sg_start = 0;
1568 		ext_offset = ctsio->ext_data_filled;
1569 	}
1570 
1571 	if (ctsio->kern_sg_entries > 0) {
1572 		kern_sglist = (struct ctl_sg_entry *)ctsio->kern_data_ptr;
1573 		kern_sg_entries = ctsio->kern_sg_entries;
1574 	} else {
1575 		kern_sglist = &kern_entry;
1576 		kern_sglist->addr = ctsio->kern_data_ptr;
1577 		kern_sglist->len = ctsio->kern_data_len;
1578 		kern_sg_entries = 1;
1579 	}
1580 
1581 
1582 	kern_watermark = 0;
1583 	ext_watermark = ext_offset;
1584 	len_copied = 0;
1585 	for (i = ext_sg_start, j = 0;
1586 	     i < ext_sg_entries && j < kern_sg_entries;) {
1587 		uint8_t *ext_ptr, *kern_ptr;
1588 
1589 		len_to_copy = ctl_min(ext_sglist[i].len - ext_watermark,
1590 				      kern_sglist[j].len - kern_watermark);
1591 
1592 		ext_ptr = (uint8_t *)ext_sglist[i].addr;
1593 		ext_ptr = ext_ptr + ext_watermark;
1594 		if (ctsio->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
1595 			/*
1596 			 * XXX KDM fix this!
1597 			 */
1598 			panic("need to implement bus address support");
1599 #if 0
1600 			kern_ptr = bus_to_virt(kern_sglist[j].addr);
1601 #endif
1602 		} else
1603 			kern_ptr = (uint8_t *)kern_sglist[j].addr;
1604 		kern_ptr = kern_ptr + kern_watermark;
1605 
1606 		kern_watermark += len_to_copy;
1607 		ext_watermark += len_to_copy;
1608 
1609 		if ((ctsio->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
1610 		     CTL_FLAG_DATA_IN) {
1611 			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1612 					 "bytes to user\n", len_to_copy));
1613 			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1614 					 "to %p\n", kern_ptr, ext_ptr));
1615 			if (copyout(kern_ptr, ext_ptr, len_to_copy) != 0) {
1616 				ctl_set_internal_failure(ctsio,
1617 							 /*sks_valid*/ 0,
1618 							 /*retry_count*/ 0);
1619 				goto bailout;
1620 			}
1621 		} else {
1622 			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1623 					 "bytes from user\n", len_to_copy));
1624 			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1625 					 "to %p\n", ext_ptr, kern_ptr));
1626 			if (copyin(ext_ptr, kern_ptr, len_to_copy)!= 0){
1627 				ctl_set_internal_failure(ctsio,
1628 							 /*sks_valid*/ 0,
1629 							 /*retry_count*/0);
1630 				goto bailout;
1631 			}
1632 		}
1633 
1634 		len_copied += len_to_copy;
1635 
1636 		if (ext_sglist[i].len == ext_watermark) {
1637 			i++;
1638 			ext_watermark = 0;
1639 		}
1640 
1641 		if (kern_sglist[j].len == kern_watermark) {
1642 			j++;
1643 			kern_watermark = 0;
1644 		}
1645 	}
1646 
1647 	ctsio->ext_data_filled += len_copied;
1648 
1649 	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_sg_entries: %d, "
1650 			 "kern_sg_entries: %d\n", ext_sg_entries,
1651 			 kern_sg_entries));
1652 	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_data_len = %d, "
1653 			 "kern_data_len = %d\n", ctsio->ext_data_len,
1654 			 ctsio->kern_data_len));
1655 
1656 
1657 	/* XXX KDM set residual?? */
1658 bailout:
1659 
1660 	if (ext_sglist_malloced != 0)
1661 		free(ext_sglist, M_CTL);
1662 
1663 	return (CTL_RETVAL_COMPLETE);
1664 }
1665 
1666 /*
1667  * Serialize a command that went down the "wrong" side, and so was sent to
1668  * this controller for execution.  The logic is a little different than the
1669  * standard case in ctl_scsiio_precheck().  Errors in this case need to get
1670  * sent back to the other side, but in the success case, we execute the
1671  * command on this side (XFER mode) or tell the other side to execute it
1672  * (SER_ONLY mode).
1673  */
1674 static int
1675 ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio, int have_lock)
1676 {
1677 	struct ctl_softc *ctl_softc;
1678 	union ctl_ha_msg msg_info;
1679 	struct ctl_lun *lun;
1680 	int retval = 0;
1681 
1682 	ctl_softc = control_softc;
1683 	if (have_lock == 0)
1684 		mtx_lock(&ctl_softc->ctl_lock);
1685 
1686 	lun = ctl_softc->ctl_luns[ctsio->io_hdr.nexus.targ_lun];
1687 	if (lun==NULL)
1688 	{
1689 		/*
1690 		 * Why isn't LUN defined? The other side wouldn't
1691 		 * send a cmd if the LUN is undefined.
1692 		 */
1693 		printf("%s: Bad JUJU!, LUN is NULL!\n", __func__);
1694 
1695 		/* "Logical unit not supported" */
1696 		ctl_set_sense_data(&msg_info.scsi.sense_data,
1697 				   lun,
1698 				   /*sense_format*/SSD_TYPE_NONE,
1699 				   /*current_error*/ 1,
1700 				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1701 				   /*asc*/ 0x25,
1702 				   /*ascq*/ 0x00,
1703 				   SSD_ELEM_NONE);
1704 
1705 		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1706 		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1707 		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1708 		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1709 		msg_info.hdr.serializing_sc = NULL;
1710 		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1711 	        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1712 				sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1713 		}
1714 		if (have_lock == 0)
1715 			mtx_unlock(&ctl_softc->ctl_lock);
1716 		return(1);
1717 
1718 	}
1719 
1720     	TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1721 
1722 	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
1723 		(union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
1724 		 ooa_links))) {
1725 	case CTL_ACTION_BLOCK:
1726 		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
1727 		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
1728 				  blocked_links);
1729 		break;
1730 	case CTL_ACTION_PASS:
1731 	case CTL_ACTION_SKIP:
1732 		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
1733 			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
1734 			STAILQ_INSERT_TAIL(&ctl_softc->rtr_queue,
1735 					   &ctsio->io_hdr, links);
1736 		} else {
1737 
1738 			/* send msg back to other side */
1739 			msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1740 			msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
1741 			msg_info.hdr.msg_type = CTL_MSG_R2R;
1742 #if 0
1743 			printf("2. pOrig %x\n", (int)msg_info.hdr.original_sc);
1744 #endif
1745 		        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1746 			    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1747 			}
1748 		}
1749 		break;
1750 	case CTL_ACTION_OVERLAP:
1751 		/* OVERLAPPED COMMANDS ATTEMPTED */
1752 		ctl_set_sense_data(&msg_info.scsi.sense_data,
1753 				   lun,
1754 				   /*sense_format*/SSD_TYPE_NONE,
1755 				   /*current_error*/ 1,
1756 				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1757 				   /*asc*/ 0x4E,
1758 				   /*ascq*/ 0x00,
1759 				   SSD_ELEM_NONE);
1760 
1761 		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1762 		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1763 		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1764 		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1765 		msg_info.hdr.serializing_sc = NULL;
1766 		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1767 #if 0
1768 		printf("BAD JUJU:Major Bummer Overlap\n");
1769 #endif
1770 		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1771 		retval = 1;
1772 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1773 		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1774 		}
1775 		break;
1776 	case CTL_ACTION_OVERLAP_TAG:
1777 		/* TAGGED OVERLAPPED COMMANDS (NN = QUEUE TAG) */
1778 		ctl_set_sense_data(&msg_info.scsi.sense_data,
1779 				   lun,
1780 				   /*sense_format*/SSD_TYPE_NONE,
1781 				   /*current_error*/ 1,
1782 				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1783 				   /*asc*/ 0x4D,
1784 				   /*ascq*/ ctsio->tag_num & 0xff,
1785 				   SSD_ELEM_NONE);
1786 
1787 		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1788 		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1789 		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1790 		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1791 		msg_info.hdr.serializing_sc = NULL;
1792 		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1793 #if 0
1794 		printf("BAD JUJU:Major Bummer Overlap Tag\n");
1795 #endif
1796 		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1797 		retval = 1;
1798 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1799 		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1800 		}
1801 		break;
1802 	case CTL_ACTION_ERROR:
1803 	default:
1804 		/* "Internal target failure" */
1805 		ctl_set_sense_data(&msg_info.scsi.sense_data,
1806 				   lun,
1807 				   /*sense_format*/SSD_TYPE_NONE,
1808 				   /*current_error*/ 1,
1809 				   /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
1810 				   /*asc*/ 0x44,
1811 				   /*ascq*/ 0x00,
1812 				   SSD_ELEM_NONE);
1813 
1814 		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1815 		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1816 		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1817 		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1818 		msg_info.hdr.serializing_sc = NULL;
1819 		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1820 #if 0
1821 		printf("BAD JUJU:Major Bummer HW Error\n");
1822 #endif
1823 		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1824 		retval = 1;
1825 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1826 		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1827 		}
1828 		break;
1829 	}
1830 	if (have_lock == 0)
1831 		mtx_unlock(&ctl_softc->ctl_lock);
1832 	return (retval);
1833 }
1834 
1835 static int
1836 ctl_ioctl_submit_wait(union ctl_io *io)
1837 {
1838 	struct ctl_fe_ioctl_params params;
1839 	ctl_fe_ioctl_state last_state;
1840 	int done, retval;
1841 
1842 	retval = 0;
1843 
1844 	bzero(&params, sizeof(params));
1845 
1846 	mtx_init(&params.ioctl_mtx, "ctliocmtx", NULL, MTX_DEF);
1847 	cv_init(&params.sem, "ctlioccv");
1848 	params.state = CTL_IOCTL_INPROG;
1849 	last_state = params.state;
1850 
1851 	io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = &params;
1852 
1853 	CTL_DEBUG_PRINT(("ctl_ioctl_submit_wait\n"));
1854 
1855 	/* This shouldn't happen */
1856 	if ((retval = ctl_queue(io)) != CTL_RETVAL_COMPLETE)
1857 		return (retval);
1858 
1859 	done = 0;
1860 
1861 	do {
1862 		mtx_lock(&params.ioctl_mtx);
1863 		/*
1864 		 * Check the state here, and don't sleep if the state has
1865 		 * already changed (i.e. wakeup has already occured, but we
1866 		 * weren't waiting yet).
1867 		 */
1868 		if (params.state == last_state) {
1869 			/* XXX KDM cv_wait_sig instead? */
1870 			cv_wait(&params.sem, &params.ioctl_mtx);
1871 		}
1872 		last_state = params.state;
1873 
1874 		switch (params.state) {
1875 		case CTL_IOCTL_INPROG:
1876 			/* Why did we wake up? */
1877 			/* XXX KDM error here? */
1878 			mtx_unlock(&params.ioctl_mtx);
1879 			break;
1880 		case CTL_IOCTL_DATAMOVE:
1881 			CTL_DEBUG_PRINT(("got CTL_IOCTL_DATAMOVE\n"));
1882 
1883 			/*
1884 			 * change last_state back to INPROG to avoid
1885 			 * deadlock on subsequent data moves.
1886 			 */
1887 			params.state = last_state = CTL_IOCTL_INPROG;
1888 
1889 			mtx_unlock(&params.ioctl_mtx);
1890 			ctl_ioctl_do_datamove(&io->scsiio);
1891 			/*
1892 			 * Note that in some cases, most notably writes,
1893 			 * this will queue the I/O and call us back later.
1894 			 * In other cases, generally reads, this routine
1895 			 * will immediately call back and wake us up,
1896 			 * probably using our own context.
1897 			 */
1898 			io->scsiio.be_move_done(io);
1899 			break;
1900 		case CTL_IOCTL_DONE:
1901 			mtx_unlock(&params.ioctl_mtx);
1902 			CTL_DEBUG_PRINT(("got CTL_IOCTL_DONE\n"));
1903 			done = 1;
1904 			break;
1905 		default:
1906 			mtx_unlock(&params.ioctl_mtx);
1907 			/* XXX KDM error here? */
1908 			break;
1909 		}
1910 	} while (done == 0);
1911 
1912 	mtx_destroy(&params.ioctl_mtx);
1913 	cv_destroy(&params.sem);
1914 
1915 	return (CTL_RETVAL_COMPLETE);
1916 }
1917 
1918 static void
1919 ctl_ioctl_datamove(union ctl_io *io)
1920 {
1921 	struct ctl_fe_ioctl_params *params;
1922 
1923 	params = (struct ctl_fe_ioctl_params *)
1924 		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
1925 
1926 	mtx_lock(&params->ioctl_mtx);
1927 	params->state = CTL_IOCTL_DATAMOVE;
1928 	cv_broadcast(&params->sem);
1929 	mtx_unlock(&params->ioctl_mtx);
1930 }
1931 
1932 static void
1933 ctl_ioctl_done(union ctl_io *io)
1934 {
1935 	struct ctl_fe_ioctl_params *params;
1936 
1937 	params = (struct ctl_fe_ioctl_params *)
1938 		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
1939 
1940 	mtx_lock(&params->ioctl_mtx);
1941 	params->state = CTL_IOCTL_DONE;
1942 	cv_broadcast(&params->sem);
1943 	mtx_unlock(&params->ioctl_mtx);
1944 }
1945 
1946 static void
1947 ctl_ioctl_hard_startstop_callback(void *arg, struct cfi_metatask *metatask)
1948 {
1949 	struct ctl_fe_ioctl_startstop_info *sd_info;
1950 
1951 	sd_info = (struct ctl_fe_ioctl_startstop_info *)arg;
1952 
1953 	sd_info->hs_info.status = metatask->status;
1954 	sd_info->hs_info.total_luns = metatask->taskinfo.startstop.total_luns;
1955 	sd_info->hs_info.luns_complete =
1956 		metatask->taskinfo.startstop.luns_complete;
1957 	sd_info->hs_info.luns_failed = metatask->taskinfo.startstop.luns_failed;
1958 
1959 	cv_broadcast(&sd_info->sem);
1960 }
1961 
1962 static void
1963 ctl_ioctl_bbrread_callback(void *arg, struct cfi_metatask *metatask)
1964 {
1965 	struct ctl_fe_ioctl_bbrread_info *fe_bbr_info;
1966 
1967 	fe_bbr_info = (struct ctl_fe_ioctl_bbrread_info *)arg;
1968 
1969 	mtx_lock(fe_bbr_info->lock);
1970 	fe_bbr_info->bbr_info->status = metatask->status;
1971 	fe_bbr_info->bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
1972 	fe_bbr_info->wakeup_done = 1;
1973 	mtx_unlock(fe_bbr_info->lock);
1974 
1975 	cv_broadcast(&fe_bbr_info->sem);
1976 }
1977 
1978 /*
1979  * Returns 0 for success, errno for failure.
1980  */
1981 static int
1982 ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
1983 		   struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
1984 {
1985 	union ctl_io *io;
1986 	int retval;
1987 
1988 	retval = 0;
1989 
1990 	mtx_assert(&control_softc->ctl_lock, MA_OWNED);
1991 
1992 	for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
1993 	     (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
1994 	     ooa_links)) {
1995 		struct ctl_ooa_entry *entry;
1996 
1997 		/*
1998 		 * If we've got more than we can fit, just count the
1999 		 * remaining entries.
2000 		 */
2001 		if (*cur_fill_num >= ooa_hdr->alloc_num)
2002 			continue;
2003 
2004 		entry = &kern_entries[*cur_fill_num];
2005 
2006 		entry->tag_num = io->scsiio.tag_num;
2007 		entry->lun_num = lun->lun;
2008 #ifdef CTL_TIME_IO
2009 		entry->start_bt = io->io_hdr.start_bt;
2010 #endif
2011 		bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2012 		entry->cdb_len = io->scsiio.cdb_len;
2013 		if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
2014 			entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2015 
2016 		if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2017 			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2018 
2019 		if (io->io_hdr.flags & CTL_FLAG_ABORT)
2020 			entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2021 
2022 		if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2023 			entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2024 
2025 		if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2026 			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2027 	}
2028 
2029 	return (retval);
2030 }
2031 
2032 static void *
2033 ctl_copyin_alloc(void *user_addr, int len, char *error_str,
2034 		 size_t error_str_len)
2035 {
2036 	void *kptr;
2037 
2038 	kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2039 
2040 	if (copyin(user_addr, kptr, len) != 0) {
2041 		snprintf(error_str, error_str_len, "Error copying %d bytes "
2042 			 "from user address %p to kernel address %p", len,
2043 			 user_addr, kptr);
2044 		free(kptr, M_CTL);
2045 		return (NULL);
2046 	}
2047 
2048 	return (kptr);
2049 }
2050 
2051 static void
2052 ctl_free_args(int num_be_args, struct ctl_be_arg *be_args)
2053 {
2054 	int i;
2055 
2056 	if (be_args == NULL)
2057 		return;
2058 
2059 	for (i = 0; i < num_be_args; i++) {
2060 		free(be_args[i].kname, M_CTL);
2061 		free(be_args[i].kvalue, M_CTL);
2062 	}
2063 
2064 	free(be_args, M_CTL);
2065 }
2066 
2067 static struct ctl_be_arg *
2068 ctl_copyin_args(int num_be_args, struct ctl_be_arg *be_args,
2069 		char *error_str, size_t error_str_len)
2070 {
2071 	struct ctl_be_arg *args;
2072 	int i;
2073 
2074 	args = ctl_copyin_alloc(be_args, num_be_args * sizeof(*be_args),
2075 				error_str, error_str_len);
2076 
2077 	if (args == NULL)
2078 		goto bailout;
2079 
2080 	for (i = 0; i < num_be_args; i++) {
2081 		args[i].kname = NULL;
2082 		args[i].kvalue = NULL;
2083 	}
2084 
2085 	for (i = 0; i < num_be_args; i++) {
2086 		uint8_t *tmpptr;
2087 
2088 		args[i].kname = ctl_copyin_alloc(args[i].name,
2089 			args[i].namelen, error_str, error_str_len);
2090 		if (args[i].kname == NULL)
2091 			goto bailout;
2092 
2093 		if (args[i].kname[args[i].namelen - 1] != '\0') {
2094 			snprintf(error_str, error_str_len, "Argument %d "
2095 				 "name is not NUL-terminated", i);
2096 			goto bailout;
2097 		}
2098 
2099 		args[i].kvalue = NULL;
2100 
2101 		tmpptr = ctl_copyin_alloc(args[i].value,
2102 			args[i].vallen, error_str, error_str_len);
2103 		if (tmpptr == NULL)
2104 			goto bailout;
2105 
2106 		args[i].kvalue = tmpptr;
2107 
2108 		if ((args[i].flags & CTL_BEARG_ASCII)
2109 		 && (tmpptr[args[i].vallen - 1] != '\0')) {
2110 			snprintf(error_str, error_str_len, "Argument %d "
2111 				 "value is not NUL-terminated", i);
2112 			goto bailout;
2113 		}
2114 	}
2115 
2116 	return (args);
2117 bailout:
2118 
2119 	ctl_free_args(num_be_args, args);
2120 
2121 	return (NULL);
2122 }
2123 
2124 /*
2125  * Escape characters that are illegal or not recommended in XML.
2126  */
2127 int
2128 ctl_sbuf_printf_esc(struct sbuf *sb, char *str)
2129 {
2130 	int retval;
2131 
2132 	retval = 0;
2133 
2134 	for (; *str; str++) {
2135 		switch (*str) {
2136 		case '&':
2137 			retval = sbuf_printf(sb, "&amp;");
2138 			break;
2139 		case '>':
2140 			retval = sbuf_printf(sb, "&gt;");
2141 			break;
2142 		case '<':
2143 			retval = sbuf_printf(sb, "&lt;");
2144 			break;
2145 		default:
2146 			retval = sbuf_putc(sb, *str);
2147 			break;
2148 		}
2149 
2150 		if (retval != 0)
2151 			break;
2152 
2153 	}
2154 
2155 	return (retval);
2156 }
2157 
2158 static int
2159 ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2160 	  struct thread *td)
2161 {
2162 	struct ctl_softc *softc;
2163 	int retval;
2164 
2165 	softc = control_softc;
2166 
2167 	retval = 0;
2168 
2169 	switch (cmd) {
2170 	case CTL_IO: {
2171 		union ctl_io *io;
2172 		void *pool_tmp;
2173 
2174 		/*
2175 		 * If we haven't been "enabled", don't allow any SCSI I/O
2176 		 * to this FETD.
2177 		 */
2178 		if ((softc->ioctl_info.flags & CTL_IOCTL_FLAG_ENABLED) == 0) {
2179 			retval = -EPERM;
2180 			break;
2181 		}
2182 
2183 		io = ctl_alloc_io(softc->ioctl_info.fe.ctl_pool_ref);
2184 		if (io == NULL) {
2185 			printf("ctl_ioctl: can't allocate ctl_io!\n");
2186 			retval = -ENOSPC;
2187 			break;
2188 		}
2189 
2190 		/*
2191 		 * Need to save the pool reference so it doesn't get
2192 		 * spammed by the user's ctl_io.
2193 		 */
2194 		pool_tmp = io->io_hdr.pool;
2195 
2196 		memcpy(io, (void *)addr, sizeof(*io));
2197 
2198 		io->io_hdr.pool = pool_tmp;
2199 		/*
2200 		 * No status yet, so make sure the status is set properly.
2201 		 */
2202 		io->io_hdr.status = CTL_STATUS_NONE;
2203 
2204 		/*
2205 		 * The user sets the initiator ID, target and LUN IDs.
2206 		 */
2207 		io->io_hdr.nexus.targ_port = softc->ioctl_info.fe.targ_port;
2208 		io->io_hdr.flags |= CTL_FLAG_USER_REQ;
2209 		if ((io->io_hdr.io_type == CTL_IO_SCSI)
2210 		 && (io->scsiio.tag_type != CTL_TAG_UNTAGGED))
2211 			io->scsiio.tag_num = softc->ioctl_info.cur_tag_num++;
2212 
2213 		retval = ctl_ioctl_submit_wait(io);
2214 
2215 		if (retval != 0) {
2216 			ctl_free_io(io);
2217 			break;
2218 		}
2219 
2220 		memcpy((void *)addr, io, sizeof(*io));
2221 
2222 		/* return this to our pool */
2223 		ctl_free_io(io);
2224 
2225 		break;
2226 	}
2227 	case CTL_ENABLE_PORT:
2228 	case CTL_DISABLE_PORT:
2229 	case CTL_SET_PORT_WWNS: {
2230 		struct ctl_frontend *fe;
2231 		struct ctl_port_entry *entry;
2232 
2233 		entry = (struct ctl_port_entry *)addr;
2234 
2235 		mtx_lock(&softc->ctl_lock);
2236 		STAILQ_FOREACH(fe, &softc->fe_list, links) {
2237 			int action, done;
2238 
2239 			action = 0;
2240 			done = 0;
2241 
2242 			if ((entry->port_type == CTL_PORT_NONE)
2243 			 && (entry->targ_port == fe->targ_port)) {
2244 				/*
2245 				 * If the user only wants to enable or
2246 				 * disable or set WWNs on a specific port,
2247 				 * do the operation and we're done.
2248 				 */
2249 				action = 1;
2250 				done = 1;
2251 			} else if (entry->port_type & fe->port_type) {
2252 				/*
2253 				 * Compare the user's type mask with the
2254 				 * particular frontend type to see if we
2255 				 * have a match.
2256 				 */
2257 				action = 1;
2258 				done = 0;
2259 
2260 				/*
2261 				 * Make sure the user isn't trying to set
2262 				 * WWNs on multiple ports at the same time.
2263 				 */
2264 				if (cmd == CTL_SET_PORT_WWNS) {
2265 					printf("%s: Can't set WWNs on "
2266 					       "multiple ports\n", __func__);
2267 					retval = EINVAL;
2268 					break;
2269 				}
2270 			}
2271 			if (action != 0) {
2272 				/*
2273 				 * XXX KDM we have to drop the lock here,
2274 				 * because the online/offline operations
2275 				 * can potentially block.  We need to
2276 				 * reference count the frontends so they
2277 				 * can't go away,
2278 				 */
2279 				mtx_unlock(&softc->ctl_lock);
2280 
2281 				if (cmd == CTL_ENABLE_PORT) {
2282 					struct ctl_lun *lun;
2283 
2284 					STAILQ_FOREACH(lun, &softc->lun_list,
2285 						       links) {
2286 						fe->lun_enable(fe->targ_lun_arg,
2287 						    lun->target,
2288 						    lun->lun);
2289 					}
2290 
2291 					ctl_frontend_online(fe);
2292 				} else if (cmd == CTL_DISABLE_PORT) {
2293 					struct ctl_lun *lun;
2294 
2295 					ctl_frontend_offline(fe);
2296 
2297 					STAILQ_FOREACH(lun, &softc->lun_list,
2298 						       links) {
2299 						fe->lun_disable(
2300 						    fe->targ_lun_arg,
2301 						    lun->target,
2302 						    lun->lun);
2303 					}
2304 				}
2305 
2306 				mtx_lock(&softc->ctl_lock);
2307 
2308 				if (cmd == CTL_SET_PORT_WWNS)
2309 					ctl_frontend_set_wwns(fe,
2310 					    (entry->flags & CTL_PORT_WWNN_VALID) ?
2311 					    1 : 0, entry->wwnn,
2312 					    (entry->flags & CTL_PORT_WWPN_VALID) ?
2313 					    1 : 0, entry->wwpn);
2314 			}
2315 			if (done != 0)
2316 				break;
2317 		}
2318 		mtx_unlock(&softc->ctl_lock);
2319 		break;
2320 	}
2321 	case CTL_GET_PORT_LIST: {
2322 		struct ctl_frontend *fe;
2323 		struct ctl_port_list *list;
2324 		int i;
2325 
2326 		list = (struct ctl_port_list *)addr;
2327 
2328 		if (list->alloc_len != (list->alloc_num *
2329 		    sizeof(struct ctl_port_entry))) {
2330 			printf("%s: CTL_GET_PORT_LIST: alloc_len %u != "
2331 			       "alloc_num %u * sizeof(struct ctl_port_entry) "
2332 			       "%zu\n", __func__, list->alloc_len,
2333 			       list->alloc_num, sizeof(struct ctl_port_entry));
2334 			retval = EINVAL;
2335 			break;
2336 		}
2337 		list->fill_len = 0;
2338 		list->fill_num = 0;
2339 		list->dropped_num = 0;
2340 		i = 0;
2341 		mtx_lock(&softc->ctl_lock);
2342 		STAILQ_FOREACH(fe, &softc->fe_list, links) {
2343 			struct ctl_port_entry entry, *list_entry;
2344 
2345 			if (list->fill_num >= list->alloc_num) {
2346 				list->dropped_num++;
2347 				continue;
2348 			}
2349 
2350 			entry.port_type = fe->port_type;
2351 			strlcpy(entry.port_name, fe->port_name,
2352 				sizeof(entry.port_name));
2353 			entry.targ_port = fe->targ_port;
2354 			entry.physical_port = fe->physical_port;
2355 			entry.virtual_port = fe->virtual_port;
2356 			entry.wwnn = fe->wwnn;
2357 			entry.wwpn = fe->wwpn;
2358 			if (fe->status & CTL_PORT_STATUS_ONLINE)
2359 				entry.online = 1;
2360 			else
2361 				entry.online = 0;
2362 
2363 			list_entry = &list->entries[i];
2364 
2365 			retval = copyout(&entry, list_entry, sizeof(entry));
2366 			if (retval != 0) {
2367 				printf("%s: CTL_GET_PORT_LIST: copyout "
2368 				       "returned %d\n", __func__, retval);
2369 				break;
2370 			}
2371 			i++;
2372 			list->fill_num++;
2373 			list->fill_len += sizeof(entry);
2374 		}
2375 		mtx_unlock(&softc->ctl_lock);
2376 
2377 		/*
2378 		 * If this is non-zero, we had a copyout fault, so there's
2379 		 * probably no point in attempting to set the status inside
2380 		 * the structure.
2381 		 */
2382 		if (retval != 0)
2383 			break;
2384 
2385 		if (list->dropped_num > 0)
2386 			list->status = CTL_PORT_LIST_NEED_MORE_SPACE;
2387 		else
2388 			list->status = CTL_PORT_LIST_OK;
2389 		break;
2390 	}
2391 	case CTL_DUMP_OOA: {
2392 		struct ctl_lun *lun;
2393 		union ctl_io *io;
2394 		char printbuf[128];
2395 		struct sbuf sb;
2396 
2397 		mtx_lock(&softc->ctl_lock);
2398 		printf("Dumping OOA queues:\n");
2399 		STAILQ_FOREACH(lun, &softc->lun_list, links) {
2400 			for (io = (union ctl_io *)TAILQ_FIRST(
2401 			     &lun->ooa_queue); io != NULL;
2402 			     io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2403 			     ooa_links)) {
2404 				sbuf_new(&sb, printbuf, sizeof(printbuf),
2405 					 SBUF_FIXEDLEN);
2406 				sbuf_printf(&sb, "LUN %jd tag 0x%04x%s%s%s%s: ",
2407 					    (intmax_t)lun->lun,
2408 					    io->scsiio.tag_num,
2409 					    (io->io_hdr.flags &
2410 					    CTL_FLAG_BLOCKED) ? "" : " BLOCKED",
2411 					    (io->io_hdr.flags &
2412 					    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
2413 					    (io->io_hdr.flags &
2414 					    CTL_FLAG_ABORT) ? " ABORT" : "",
2415 			                    (io->io_hdr.flags &
2416 		                        CTL_FLAG_IS_WAS_ON_RTR) ? " RTR" : "");
2417 				ctl_scsi_command_string(&io->scsiio, NULL, &sb);
2418 				sbuf_finish(&sb);
2419 				printf("%s\n", sbuf_data(&sb));
2420 			}
2421 		}
2422 		printf("OOA queues dump done\n");
2423 		mtx_unlock(&softc->ctl_lock);
2424 		break;
2425 	}
2426 	case CTL_GET_OOA: {
2427 		struct ctl_lun *lun;
2428 		struct ctl_ooa *ooa_hdr;
2429 		struct ctl_ooa_entry *entries;
2430 		uint32_t cur_fill_num;
2431 
2432 		ooa_hdr = (struct ctl_ooa *)addr;
2433 
2434 		if ((ooa_hdr->alloc_len == 0)
2435 		 || (ooa_hdr->alloc_num == 0)) {
2436 			printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2437 			       "must be non-zero\n", __func__,
2438 			       ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2439 			retval = EINVAL;
2440 			break;
2441 		}
2442 
2443 		if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2444 		    sizeof(struct ctl_ooa_entry))) {
2445 			printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2446 			       "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2447 			       __func__, ooa_hdr->alloc_len,
2448 			       ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2449 			retval = EINVAL;
2450 			break;
2451 		}
2452 
2453 		entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2454 		if (entries == NULL) {
2455 			printf("%s: could not allocate %d bytes for OOA "
2456 			       "dump\n", __func__, ooa_hdr->alloc_len);
2457 			retval = ENOMEM;
2458 			break;
2459 		}
2460 
2461 		mtx_lock(&softc->ctl_lock);
2462 		if (((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0)
2463 		 && ((ooa_hdr->lun_num > CTL_MAX_LUNS)
2464 		  || (softc->ctl_luns[ooa_hdr->lun_num] == NULL))) {
2465 			mtx_unlock(&softc->ctl_lock);
2466 			free(entries, M_CTL);
2467 			printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2468 			       __func__, (uintmax_t)ooa_hdr->lun_num);
2469 			retval = EINVAL;
2470 			break;
2471 		}
2472 
2473 		cur_fill_num = 0;
2474 
2475 		if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2476 			STAILQ_FOREACH(lun, &softc->lun_list, links) {
2477 				retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2478 					ooa_hdr, entries);
2479 				if (retval != 0)
2480 					break;
2481 			}
2482 			if (retval != 0) {
2483 				mtx_unlock(&softc->ctl_lock);
2484 				free(entries, M_CTL);
2485 				break;
2486 			}
2487 		} else {
2488 			lun = softc->ctl_luns[ooa_hdr->lun_num];
2489 
2490 			retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,ooa_hdr,
2491 						    entries);
2492 		}
2493 		mtx_unlock(&softc->ctl_lock);
2494 
2495 		ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2496 		ooa_hdr->fill_len = ooa_hdr->fill_num *
2497 			sizeof(struct ctl_ooa_entry);
2498 		retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2499 		if (retval != 0) {
2500 			printf("%s: error copying out %d bytes for OOA dump\n",
2501 			       __func__, ooa_hdr->fill_len);
2502 		}
2503 
2504 		getbintime(&ooa_hdr->cur_bt);
2505 
2506 		if (cur_fill_num > ooa_hdr->alloc_num) {
2507 			ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2508 			ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2509 		} else {
2510 			ooa_hdr->dropped_num = 0;
2511 			ooa_hdr->status = CTL_OOA_OK;
2512 		}
2513 
2514 		free(entries, M_CTL);
2515 		break;
2516 	}
2517 	case CTL_CHECK_OOA: {
2518 		union ctl_io *io;
2519 		struct ctl_lun *lun;
2520 		struct ctl_ooa_info *ooa_info;
2521 
2522 
2523 		ooa_info = (struct ctl_ooa_info *)addr;
2524 
2525 		if (ooa_info->lun_id >= CTL_MAX_LUNS) {
2526 			ooa_info->status = CTL_OOA_INVALID_LUN;
2527 			break;
2528 		}
2529 		mtx_lock(&softc->ctl_lock);
2530 		lun = softc->ctl_luns[ooa_info->lun_id];
2531 		if (lun == NULL) {
2532 			mtx_unlock(&softc->ctl_lock);
2533 			ooa_info->status = CTL_OOA_INVALID_LUN;
2534 			break;
2535 		}
2536 
2537 		ooa_info->num_entries = 0;
2538 		for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
2539 		     io != NULL; io = (union ctl_io *)TAILQ_NEXT(
2540 		     &io->io_hdr, ooa_links)) {
2541 			ooa_info->num_entries++;
2542 		}
2543 
2544 		mtx_unlock(&softc->ctl_lock);
2545 		ooa_info->status = CTL_OOA_SUCCESS;
2546 
2547 		break;
2548 	}
2549 	case CTL_HARD_START:
2550 	case CTL_HARD_STOP: {
2551 		struct ctl_fe_ioctl_startstop_info ss_info;
2552 		struct cfi_metatask *metatask;
2553 		struct mtx hs_mtx;
2554 
2555 		mtx_init(&hs_mtx, "HS Mutex", NULL, MTX_DEF);
2556 
2557 		cv_init(&ss_info.sem, "hard start/stop cv" );
2558 
2559 		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2560 		if (metatask == NULL) {
2561 			retval = ENOMEM;
2562 			mtx_destroy(&hs_mtx);
2563 			break;
2564 		}
2565 
2566 		if (cmd == CTL_HARD_START)
2567 			metatask->tasktype = CFI_TASK_STARTUP;
2568 		else
2569 			metatask->tasktype = CFI_TASK_SHUTDOWN;
2570 
2571 		metatask->callback = ctl_ioctl_hard_startstop_callback;
2572 		metatask->callback_arg = &ss_info;
2573 
2574 		cfi_action(metatask);
2575 
2576 		/* Wait for the callback */
2577 		mtx_lock(&hs_mtx);
2578 		cv_wait_sig(&ss_info.sem, &hs_mtx);
2579 		mtx_unlock(&hs_mtx);
2580 
2581 		/*
2582 		 * All information has been copied from the metatask by the
2583 		 * time cv_broadcast() is called, so we free the metatask here.
2584 		 */
2585 		cfi_free_metatask(metatask);
2586 
2587 		memcpy((void *)addr, &ss_info.hs_info, sizeof(ss_info.hs_info));
2588 
2589 		mtx_destroy(&hs_mtx);
2590 		break;
2591 	}
2592 	case CTL_BBRREAD: {
2593 		struct ctl_bbrread_info *bbr_info;
2594 		struct ctl_fe_ioctl_bbrread_info fe_bbr_info;
2595 		struct mtx bbr_mtx;
2596 		struct cfi_metatask *metatask;
2597 
2598 		bbr_info = (struct ctl_bbrread_info *)addr;
2599 
2600 		bzero(&fe_bbr_info, sizeof(fe_bbr_info));
2601 
2602 		bzero(&bbr_mtx, sizeof(bbr_mtx));
2603 		mtx_init(&bbr_mtx, "BBR Mutex", NULL, MTX_DEF);
2604 
2605 		fe_bbr_info.bbr_info = bbr_info;
2606 		fe_bbr_info.lock = &bbr_mtx;
2607 
2608 		cv_init(&fe_bbr_info.sem, "BBR read cv");
2609 		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2610 
2611 		if (metatask == NULL) {
2612 			mtx_destroy(&bbr_mtx);
2613 			cv_destroy(&fe_bbr_info.sem);
2614 			retval = ENOMEM;
2615 			break;
2616 		}
2617 		metatask->tasktype = CFI_TASK_BBRREAD;
2618 		metatask->callback = ctl_ioctl_bbrread_callback;
2619 		metatask->callback_arg = &fe_bbr_info;
2620 		metatask->taskinfo.bbrread.lun_num = bbr_info->lun_num;
2621 		metatask->taskinfo.bbrread.lba = bbr_info->lba;
2622 		metatask->taskinfo.bbrread.len = bbr_info->len;
2623 
2624 		cfi_action(metatask);
2625 
2626 		mtx_lock(&bbr_mtx);
2627 		while (fe_bbr_info.wakeup_done == 0)
2628 			cv_wait_sig(&fe_bbr_info.sem, &bbr_mtx);
2629 		mtx_unlock(&bbr_mtx);
2630 
2631 		bbr_info->status = metatask->status;
2632 		bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2633 		bbr_info->scsi_status = metatask->taskinfo.bbrread.scsi_status;
2634 		memcpy(&bbr_info->sense_data,
2635 		       &metatask->taskinfo.bbrread.sense_data,
2636 		       ctl_min(sizeof(bbr_info->sense_data),
2637 			       sizeof(metatask->taskinfo.bbrread.sense_data)));
2638 
2639 		cfi_free_metatask(metatask);
2640 
2641 		mtx_destroy(&bbr_mtx);
2642 		cv_destroy(&fe_bbr_info.sem);
2643 
2644 		break;
2645 	}
2646 	case CTL_DELAY_IO: {
2647 		struct ctl_io_delay_info *delay_info;
2648 #ifdef CTL_IO_DELAY
2649 		struct ctl_lun *lun;
2650 #endif /* CTL_IO_DELAY */
2651 
2652 		delay_info = (struct ctl_io_delay_info *)addr;
2653 
2654 #ifdef CTL_IO_DELAY
2655 		mtx_lock(&softc->ctl_lock);
2656 
2657 		if ((delay_info->lun_id > CTL_MAX_LUNS)
2658 		 || (softc->ctl_luns[delay_info->lun_id] == NULL)) {
2659 			delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2660 		} else {
2661 			lun = softc->ctl_luns[delay_info->lun_id];
2662 
2663 			delay_info->status = CTL_DELAY_STATUS_OK;
2664 
2665 			switch (delay_info->delay_type) {
2666 			case CTL_DELAY_TYPE_CONT:
2667 				break;
2668 			case CTL_DELAY_TYPE_ONESHOT:
2669 				break;
2670 			default:
2671 				delay_info->status =
2672 					CTL_DELAY_STATUS_INVALID_TYPE;
2673 				break;
2674 			}
2675 
2676 			switch (delay_info->delay_loc) {
2677 			case CTL_DELAY_LOC_DATAMOVE:
2678 				lun->delay_info.datamove_type =
2679 					delay_info->delay_type;
2680 				lun->delay_info.datamove_delay =
2681 					delay_info->delay_secs;
2682 				break;
2683 			case CTL_DELAY_LOC_DONE:
2684 				lun->delay_info.done_type =
2685 					delay_info->delay_type;
2686 				lun->delay_info.done_delay =
2687 					delay_info->delay_secs;
2688 				break;
2689 			default:
2690 				delay_info->status =
2691 					CTL_DELAY_STATUS_INVALID_LOC;
2692 				break;
2693 			}
2694 		}
2695 
2696 		mtx_unlock(&softc->ctl_lock);
2697 #else
2698 		delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2699 #endif /* CTL_IO_DELAY */
2700 		break;
2701 	}
2702 	case CTL_REALSYNC_SET: {
2703 		int *syncstate;
2704 
2705 		syncstate = (int *)addr;
2706 
2707 		mtx_lock(&softc->ctl_lock);
2708 		switch (*syncstate) {
2709 		case 0:
2710 			softc->flags &= ~CTL_FLAG_REAL_SYNC;
2711 			break;
2712 		case 1:
2713 			softc->flags |= CTL_FLAG_REAL_SYNC;
2714 			break;
2715 		default:
2716 			retval = -EINVAL;
2717 			break;
2718 		}
2719 		mtx_unlock(&softc->ctl_lock);
2720 		break;
2721 	}
2722 	case CTL_REALSYNC_GET: {
2723 		int *syncstate;
2724 
2725 		syncstate = (int*)addr;
2726 
2727 		mtx_lock(&softc->ctl_lock);
2728 		if (softc->flags & CTL_FLAG_REAL_SYNC)
2729 			*syncstate = 1;
2730 		else
2731 			*syncstate = 0;
2732 		mtx_unlock(&softc->ctl_lock);
2733 
2734 		break;
2735 	}
2736 	case CTL_SETSYNC:
2737 	case CTL_GETSYNC: {
2738 		struct ctl_sync_info *sync_info;
2739 		struct ctl_lun *lun;
2740 
2741 		sync_info = (struct ctl_sync_info *)addr;
2742 
2743 		mtx_lock(&softc->ctl_lock);
2744 		lun = softc->ctl_luns[sync_info->lun_id];
2745 		if (lun == NULL) {
2746 			mtx_unlock(&softc->ctl_lock);
2747 			sync_info->status = CTL_GS_SYNC_NO_LUN;
2748 		}
2749 		/*
2750 		 * Get or set the sync interval.  We're not bounds checking
2751 		 * in the set case, hopefully the user won't do something
2752 		 * silly.
2753 		 */
2754 		if (cmd == CTL_GETSYNC)
2755 			sync_info->sync_interval = lun->sync_interval;
2756 		else
2757 			lun->sync_interval = sync_info->sync_interval;
2758 
2759 		mtx_unlock(&softc->ctl_lock);
2760 
2761 		sync_info->status = CTL_GS_SYNC_OK;
2762 
2763 		break;
2764 	}
2765 	case CTL_GETSTATS: {
2766 		struct ctl_stats *stats;
2767 		struct ctl_lun *lun;
2768 		int i;
2769 
2770 		stats = (struct ctl_stats *)addr;
2771 
2772 		if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) >
2773 		     stats->alloc_len) {
2774 			stats->status = CTL_SS_NEED_MORE_SPACE;
2775 			stats->num_luns = softc->num_luns;
2776 			break;
2777 		}
2778 		/*
2779 		 * XXX KDM no locking here.  If the LUN list changes,
2780 		 * things can blow up.
2781 		 */
2782 		for (i = 0, lun = STAILQ_FIRST(&softc->lun_list); lun != NULL;
2783 		     i++, lun = STAILQ_NEXT(lun, links)) {
2784 			retval = copyout(&lun->stats, &stats->lun_stats[i],
2785 					 sizeof(lun->stats));
2786 			if (retval != 0)
2787 				break;
2788 		}
2789 		stats->num_luns = softc->num_luns;
2790 		stats->fill_len = sizeof(struct ctl_lun_io_stats) *
2791 				 softc->num_luns;
2792 		stats->status = CTL_SS_OK;
2793 #ifdef CTL_TIME_IO
2794 		stats->flags = CTL_STATS_FLAG_TIME_VALID;
2795 #else
2796 		stats->flags = CTL_STATS_FLAG_NONE;
2797 #endif
2798 		getnanouptime(&stats->timestamp);
2799 		break;
2800 	}
2801 	case CTL_ERROR_INJECT: {
2802 		struct ctl_error_desc *err_desc, *new_err_desc;
2803 		struct ctl_lun *lun;
2804 
2805 		err_desc = (struct ctl_error_desc *)addr;
2806 
2807 		new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2808 				      M_WAITOK | M_ZERO);
2809 		bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2810 
2811 		mtx_lock(&softc->ctl_lock);
2812 		lun = softc->ctl_luns[err_desc->lun_id];
2813 		if (lun == NULL) {
2814 			mtx_unlock(&softc->ctl_lock);
2815 			printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2816 			       __func__, (uintmax_t)err_desc->lun_id);
2817 			retval = EINVAL;
2818 			break;
2819 		}
2820 
2821 		/*
2822 		 * We could do some checking here to verify the validity
2823 		 * of the request, but given the complexity of error
2824 		 * injection requests, the checking logic would be fairly
2825 		 * complex.
2826 		 *
2827 		 * For now, if the request is invalid, it just won't get
2828 		 * executed and might get deleted.
2829 		 */
2830 		STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2831 
2832 		/*
2833 		 * XXX KDM check to make sure the serial number is unique,
2834 		 * in case we somehow manage to wrap.  That shouldn't
2835 		 * happen for a very long time, but it's the right thing to
2836 		 * do.
2837 		 */
2838 		new_err_desc->serial = lun->error_serial;
2839 		err_desc->serial = lun->error_serial;
2840 		lun->error_serial++;
2841 
2842 		mtx_unlock(&softc->ctl_lock);
2843 		break;
2844 	}
2845 	case CTL_ERROR_INJECT_DELETE: {
2846 		struct ctl_error_desc *delete_desc, *desc, *desc2;
2847 		struct ctl_lun *lun;
2848 		int delete_done;
2849 
2850 		delete_desc = (struct ctl_error_desc *)addr;
2851 		delete_done = 0;
2852 
2853 		mtx_lock(&softc->ctl_lock);
2854 		lun = softc->ctl_luns[delete_desc->lun_id];
2855 		if (lun == NULL) {
2856 			mtx_unlock(&softc->ctl_lock);
2857 			printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2858 			       __func__, (uintmax_t)delete_desc->lun_id);
2859 			retval = EINVAL;
2860 			break;
2861 		}
2862 		STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2863 			if (desc->serial != delete_desc->serial)
2864 				continue;
2865 
2866 			STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2867 				      links);
2868 			free(desc, M_CTL);
2869 			delete_done = 1;
2870 		}
2871 		mtx_unlock(&softc->ctl_lock);
2872 		if (delete_done == 0) {
2873 			printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2874 			       "error serial %ju on LUN %u\n", __func__,
2875 			       delete_desc->serial, delete_desc->lun_id);
2876 			retval = EINVAL;
2877 			break;
2878 		}
2879 		break;
2880 	}
2881 	case CTL_DUMP_STRUCTS: {
2882 		int i, j, k;
2883 		struct ctl_frontend *fe;
2884 
2885 		printf("CTL IID to WWPN map start:\n");
2886 		for (i = 0; i < CTL_MAX_PORTS; i++) {
2887 			for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
2888 				if (softc->wwpn_iid[i][j].in_use == 0)
2889 					continue;
2890 
2891 				printf("port %d iid %u WWPN %#jx\n",
2892 				       softc->wwpn_iid[i][j].port,
2893 				       softc->wwpn_iid[i][j].iid,
2894 				       (uintmax_t)softc->wwpn_iid[i][j].wwpn);
2895 			}
2896 		}
2897 		printf("CTL IID to WWPN map end\n");
2898 		printf("CTL Persistent Reservation information start:\n");
2899 		for (i = 0; i < CTL_MAX_LUNS; i++) {
2900 			struct ctl_lun *lun;
2901 
2902 			lun = softc->ctl_luns[i];
2903 
2904 			if ((lun == NULL)
2905 			 || ((lun->flags & CTL_LUN_DISABLED) != 0))
2906 				continue;
2907 
2908 			for (j = 0; j < (CTL_MAX_PORTS * 2); j++) {
2909 				for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2910 					if (lun->per_res[j+k].registered == 0)
2911 						continue;
2912 					printf("LUN %d port %d iid %d key "
2913 					       "%#jx\n", i, j, k,
2914 					       (uintmax_t)scsi_8btou64(
2915 					       lun->per_res[j+k].res_key.key));
2916 				}
2917 			}
2918 		}
2919 		printf("CTL Persistent Reservation information end\n");
2920 		printf("CTL Frontends:\n");
2921 		/*
2922 		 * XXX KDM calling this without a lock.  We'd likely want
2923 		 * to drop the lock before calling the frontend's dump
2924 		 * routine anyway.
2925 		 */
2926 		STAILQ_FOREACH(fe, &softc->fe_list, links) {
2927 			printf("Frontend %s Type %u pport %d vport %d WWNN "
2928 			       "%#jx WWPN %#jx\n", fe->port_name, fe->port_type,
2929 			       fe->physical_port, fe->virtual_port,
2930 			       (uintmax_t)fe->wwnn, (uintmax_t)fe->wwpn);
2931 
2932 			/*
2933 			 * Frontends are not required to support the dump
2934 			 * routine.
2935 			 */
2936 			if (fe->fe_dump == NULL)
2937 				continue;
2938 
2939 			fe->fe_dump();
2940 		}
2941 		printf("CTL Frontend information end\n");
2942 		break;
2943 	}
2944 	case CTL_LUN_REQ: {
2945 		struct ctl_lun_req *lun_req;
2946 		struct ctl_backend_driver *backend;
2947 
2948 		lun_req = (struct ctl_lun_req *)addr;
2949 
2950 		backend = ctl_backend_find(lun_req->backend);
2951 		if (backend == NULL) {
2952 			lun_req->status = CTL_LUN_ERROR;
2953 			snprintf(lun_req->error_str,
2954 				 sizeof(lun_req->error_str),
2955 				 "Backend \"%s\" not found.",
2956 				 lun_req->backend);
2957 			break;
2958 		}
2959 		if (lun_req->num_be_args > 0) {
2960 			lun_req->kern_be_args = ctl_copyin_args(
2961 				lun_req->num_be_args,
2962 				lun_req->be_args,
2963 				lun_req->error_str,
2964 				sizeof(lun_req->error_str));
2965 			if (lun_req->kern_be_args == NULL) {
2966 				lun_req->status = CTL_LUN_ERROR;
2967 				break;
2968 			}
2969 		}
2970 
2971 		retval = backend->ioctl(dev, cmd, addr, flag, td);
2972 
2973 		if (lun_req->num_be_args > 0) {
2974 			ctl_free_args(lun_req->num_be_args,
2975 				      lun_req->kern_be_args);
2976 		}
2977 		break;
2978 	}
2979 	case CTL_LUN_LIST: {
2980 		struct sbuf *sb;
2981 		struct ctl_lun *lun;
2982 		struct ctl_lun_list *list;
2983 
2984 		list = (struct ctl_lun_list *)addr;
2985 
2986 		/*
2987 		 * Allocate a fixed length sbuf here, based on the length
2988 		 * of the user's buffer.  We could allocate an auto-extending
2989 		 * buffer, and then tell the user how much larger our
2990 		 * amount of data is than his buffer, but that presents
2991 		 * some problems:
2992 		 *
2993 		 * 1.  The sbuf(9) routines use a blocking malloc, and so
2994 		 *     we can't hold a lock while calling them with an
2995 		 *     auto-extending buffer.
2996  		 *
2997 		 * 2.  There is not currently a LUN reference counting
2998 		 *     mechanism, outside of outstanding transactions on
2999 		 *     the LUN's OOA queue.  So a LUN could go away on us
3000 		 *     while we're getting the LUN number, backend-specific
3001 		 *     information, etc.  Thus, given the way things
3002 		 *     currently work, we need to hold the CTL lock while
3003 		 *     grabbing LUN information.
3004 		 *
3005 		 * So, from the user's standpoint, the best thing to do is
3006 		 * allocate what he thinks is a reasonable buffer length,
3007 		 * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
3008 		 * double the buffer length and try again.  (And repeat
3009 		 * that until he succeeds.)
3010 		 */
3011 		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3012 		if (sb == NULL) {
3013 			list->status = CTL_LUN_LIST_ERROR;
3014 			snprintf(list->error_str, sizeof(list->error_str),
3015 				 "Unable to allocate %d bytes for LUN list",
3016 				 list->alloc_len);
3017 			break;
3018 		}
3019 
3020 		sbuf_printf(sb, "<ctllunlist>\n");
3021 
3022 		mtx_lock(&softc->ctl_lock);
3023 
3024 		STAILQ_FOREACH(lun, &softc->lun_list, links) {
3025 			retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3026 					     (uintmax_t)lun->lun);
3027 
3028 			/*
3029 			 * Bail out as soon as we see that we've overfilled
3030 			 * the buffer.
3031 			 */
3032 			if (retval != 0)
3033 				break;
3034 
3035 			retval = sbuf_printf(sb, "<backend_type>%s"
3036 					     "</backend_type>\n",
3037 					     (lun->backend == NULL) ?  "none" :
3038 					     lun->backend->name);
3039 
3040 			if (retval != 0)
3041 				break;
3042 
3043 			retval = sbuf_printf(sb, "<lun_type>%d</lun_type>\n",
3044 					     lun->be_lun->lun_type);
3045 
3046 			if (retval != 0)
3047 				break;
3048 
3049 			if (lun->backend == NULL) {
3050 				retval = sbuf_printf(sb, "</lun>\n");
3051 				if (retval != 0)
3052 					break;
3053 				continue;
3054 			}
3055 
3056 			retval = sbuf_printf(sb, "<size>%ju</size>\n",
3057 					     (lun->be_lun->maxlba > 0) ?
3058 					     lun->be_lun->maxlba + 1 : 0);
3059 
3060 			if (retval != 0)
3061 				break;
3062 
3063 			retval = sbuf_printf(sb, "<blocksize>%u</blocksize>\n",
3064 					     lun->be_lun->blocksize);
3065 
3066 			if (retval != 0)
3067 				break;
3068 
3069 			retval = sbuf_printf(sb, "<serial_number>");
3070 
3071 			if (retval != 0)
3072 				break;
3073 
3074 			retval = ctl_sbuf_printf_esc(sb,
3075 						     lun->be_lun->serial_num);
3076 
3077 			if (retval != 0)
3078 				break;
3079 
3080 			retval = sbuf_printf(sb, "</serial_number>\n");
3081 
3082 			if (retval != 0)
3083 				break;
3084 
3085 			retval = sbuf_printf(sb, "<device_id>");
3086 
3087 			if (retval != 0)
3088 				break;
3089 
3090 			retval = ctl_sbuf_printf_esc(sb,lun->be_lun->device_id);
3091 
3092 			if (retval != 0)
3093 				break;
3094 
3095 			retval = sbuf_printf(sb, "</device_id>\n");
3096 
3097 			if (retval != 0)
3098 				break;
3099 
3100 			if (lun->backend->lun_info == NULL) {
3101 				retval = sbuf_printf(sb, "</lun>\n");
3102 				if (retval != 0)
3103 					break;
3104 				continue;
3105 			}
3106 
3107 			retval =lun->backend->lun_info(lun->be_lun->be_lun, sb);
3108 
3109 			if (retval != 0)
3110 				break;
3111 
3112 			retval = sbuf_printf(sb, "</lun>\n");
3113 
3114 			if (retval != 0)
3115 				break;
3116 		}
3117 		mtx_unlock(&softc->ctl_lock);
3118 
3119 		if ((retval != 0)
3120 		 || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3121 			retval = 0;
3122 			sbuf_delete(sb);
3123 			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3124 			snprintf(list->error_str, sizeof(list->error_str),
3125 				 "Out of space, %d bytes is too small",
3126 				 list->alloc_len);
3127 			break;
3128 		}
3129 
3130 		sbuf_finish(sb);
3131 
3132 		retval = copyout(sbuf_data(sb), list->lun_xml,
3133 				 sbuf_len(sb) + 1);
3134 
3135 		list->fill_len = sbuf_len(sb) + 1;
3136 		list->status = CTL_LUN_LIST_OK;
3137 		sbuf_delete(sb);
3138 		break;
3139 	}
3140 	default: {
3141 		/* XXX KDM should we fix this? */
3142 #if 0
3143 		struct ctl_backend_driver *backend;
3144 		unsigned int type;
3145 		int found;
3146 
3147 		found = 0;
3148 
3149 		/*
3150 		 * We encode the backend type as the ioctl type for backend
3151 		 * ioctls.  So parse it out here, and then search for a
3152 		 * backend of this type.
3153 		 */
3154 		type = _IOC_TYPE(cmd);
3155 
3156 		STAILQ_FOREACH(backend, &softc->be_list, links) {
3157 			if (backend->type == type) {
3158 				found = 1;
3159 				break;
3160 			}
3161 		}
3162 		if (found == 0) {
3163 			printf("ctl: unknown ioctl command %#lx or backend "
3164 			       "%d\n", cmd, type);
3165 			retval = -EINVAL;
3166 			break;
3167 		}
3168 		retval = backend->ioctl(dev, cmd, addr, flag, td);
3169 #endif
3170 		retval = ENOTTY;
3171 		break;
3172 	}
3173 	}
3174 	return (retval);
3175 }
3176 
3177 uint32_t
3178 ctl_get_initindex(struct ctl_nexus *nexus)
3179 {
3180 	if (nexus->targ_port < CTL_MAX_PORTS)
3181 		return (nexus->initid.id +
3182 			(nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3183 	else
3184 		return (nexus->initid.id +
3185 		       ((nexus->targ_port - CTL_MAX_PORTS) *
3186 			CTL_MAX_INIT_PER_PORT));
3187 }
3188 
3189 uint32_t
3190 ctl_get_resindex(struct ctl_nexus *nexus)
3191 {
3192 	return (nexus->initid.id + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3193 }
3194 
3195 uint32_t
3196 ctl_port_idx(int port_num)
3197 {
3198 	if (port_num < CTL_MAX_PORTS)
3199 		return(port_num);
3200 	else
3201 		return(port_num - CTL_MAX_PORTS);
3202 }
3203 
3204 /*
3205  * Note:  This only works for bitmask sizes that are at least 32 bits, and
3206  * that are a power of 2.
3207  */
3208 int
3209 ctl_ffz(uint32_t *mask, uint32_t size)
3210 {
3211 	uint32_t num_chunks, num_pieces;
3212 	int i, j;
3213 
3214 	num_chunks = (size >> 5);
3215 	if (num_chunks == 0)
3216 		num_chunks++;
3217 	num_pieces = ctl_min((sizeof(uint32_t) * 8), size);
3218 
3219 	for (i = 0; i < num_chunks; i++) {
3220 		for (j = 0; j < num_pieces; j++) {
3221 			if ((mask[i] & (1 << j)) == 0)
3222 				return ((i << 5) + j);
3223 		}
3224 	}
3225 
3226 	return (-1);
3227 }
3228 
3229 int
3230 ctl_set_mask(uint32_t *mask, uint32_t bit)
3231 {
3232 	uint32_t chunk, piece;
3233 
3234 	chunk = bit >> 5;
3235 	piece = bit % (sizeof(uint32_t) * 8);
3236 
3237 	if ((mask[chunk] & (1 << piece)) != 0)
3238 		return (-1);
3239 	else
3240 		mask[chunk] |= (1 << piece);
3241 
3242 	return (0);
3243 }
3244 
3245 int
3246 ctl_clear_mask(uint32_t *mask, uint32_t bit)
3247 {
3248 	uint32_t chunk, piece;
3249 
3250 	chunk = bit >> 5;
3251 	piece = bit % (sizeof(uint32_t) * 8);
3252 
3253 	if ((mask[chunk] & (1 << piece)) == 0)
3254 		return (-1);
3255 	else
3256 		mask[chunk] &= ~(1 << piece);
3257 
3258 	return (0);
3259 }
3260 
3261 int
3262 ctl_is_set(uint32_t *mask, uint32_t bit)
3263 {
3264 	uint32_t chunk, piece;
3265 
3266 	chunk = bit >> 5;
3267 	piece = bit % (sizeof(uint32_t) * 8);
3268 
3269 	if ((mask[chunk] & (1 << piece)) == 0)
3270 		return (0);
3271 	else
3272 		return (1);
3273 }
3274 
3275 #ifdef unused
3276 /*
3277  * The bus, target and lun are optional, they can be filled in later.
3278  * can_wait is used to determine whether we can wait on the malloc or not.
3279  */
3280 union ctl_io*
3281 ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port, uint32_t targ_target,
3282 	      uint32_t targ_lun, int can_wait)
3283 {
3284 	union ctl_io *io;
3285 
3286 	if (can_wait)
3287 		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_WAITOK);
3288 	else
3289 		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_NOWAIT);
3290 
3291 	if (io != NULL) {
3292 		io->io_hdr.io_type = io_type;
3293 		io->io_hdr.targ_port = targ_port;
3294 		/*
3295 		 * XXX KDM this needs to change/go away.  We need to move
3296 		 * to a preallocated pool of ctl_scsiio structures.
3297 		 */
3298 		io->io_hdr.nexus.targ_target.id = targ_target;
3299 		io->io_hdr.nexus.targ_lun = targ_lun;
3300 	}
3301 
3302 	return (io);
3303 }
3304 
3305 void
3306 ctl_kfree_io(union ctl_io *io)
3307 {
3308 	free(io, M_CTL);
3309 }
3310 #endif /* unused */
3311 
3312 /*
3313  * ctl_softc, pool_type, total_ctl_io are passed in.
3314  * npool is passed out.
3315  */
3316 int
3317 ctl_pool_create(struct ctl_softc *ctl_softc, ctl_pool_type pool_type,
3318 		uint32_t total_ctl_io, struct ctl_io_pool **npool)
3319 {
3320 	uint32_t i;
3321 	union ctl_io *cur_io, *next_io;
3322 	struct ctl_io_pool *pool;
3323 	int retval;
3324 
3325 	retval = 0;
3326 
3327 	pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3328 					    M_NOWAIT | M_ZERO);
3329 	if (pool == NULL) {
3330 		retval = -ENOMEM;
3331 		goto bailout;
3332 	}
3333 
3334 	pool->type = pool_type;
3335 	pool->ctl_softc = ctl_softc;
3336 
3337 	mtx_lock(&ctl_softc->ctl_lock);
3338 	pool->id = ctl_softc->cur_pool_id++;
3339 	mtx_unlock(&ctl_softc->ctl_lock);
3340 
3341 	pool->flags = CTL_POOL_FLAG_NONE;
3342 	STAILQ_INIT(&pool->free_queue);
3343 
3344 	/*
3345 	 * XXX KDM other options here:
3346 	 * - allocate a page at a time
3347 	 * - allocate one big chunk of memory.
3348 	 * Page allocation might work well, but would take a little more
3349 	 * tracking.
3350 	 */
3351 	for (i = 0; i < total_ctl_io; i++) {
3352 		cur_io = (union ctl_io *)malloc(sizeof(*cur_io), M_CTL,
3353 						M_NOWAIT);
3354 		if (cur_io == NULL) {
3355 			retval = ENOMEM;
3356 			break;
3357 		}
3358 		cur_io->io_hdr.pool = pool;
3359 		STAILQ_INSERT_TAIL(&pool->free_queue, &cur_io->io_hdr, links);
3360 		pool->total_ctl_io++;
3361 		pool->free_ctl_io++;
3362 	}
3363 
3364 	if (retval != 0) {
3365 		for (cur_io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3366 		     cur_io != NULL; cur_io = next_io) {
3367 			next_io = (union ctl_io *)STAILQ_NEXT(&cur_io->io_hdr,
3368 							      links);
3369 			STAILQ_REMOVE(&pool->free_queue, &cur_io->io_hdr,
3370 				      ctl_io_hdr, links);
3371 			free(cur_io, M_CTL);
3372 		}
3373 
3374 		free(pool, M_CTL);
3375 		goto bailout;
3376 	}
3377 	mtx_lock(&ctl_softc->ctl_lock);
3378 	ctl_softc->num_pools++;
3379 	STAILQ_INSERT_TAIL(&ctl_softc->io_pools, pool, links);
3380 	/*
3381 	 * Increment our usage count if this is an external consumer, so we
3382 	 * can't get unloaded until the external consumer (most likely a
3383 	 * FETD) unloads and frees his pool.
3384 	 *
3385 	 * XXX KDM will this increment the caller's module use count, or
3386 	 * mine?
3387 	 */
3388 #if 0
3389 	if ((pool_type != CTL_POOL_EMERGENCY)
3390 	 && (pool_type != CTL_POOL_INTERNAL)
3391 	 && (pool_type != CTL_POOL_IOCTL)
3392 	 && (pool_type != CTL_POOL_4OTHERSC))
3393 		MOD_INC_USE_COUNT;
3394 #endif
3395 
3396 	mtx_unlock(&ctl_softc->ctl_lock);
3397 
3398 	*npool = pool;
3399 
3400 bailout:
3401 
3402 	return (retval);
3403 }
3404 
3405 int
3406 ctl_pool_acquire(struct ctl_io_pool *pool)
3407 {
3408 
3409 	mtx_assert(&control_softc->ctl_lock, MA_OWNED);
3410 
3411 	if (pool == NULL)
3412 		return (-EINVAL);
3413 
3414 	if (pool->flags & CTL_POOL_FLAG_INVALID)
3415 		return (-EINVAL);
3416 
3417 	pool->refcount++;
3418 
3419 	return (0);
3420 }
3421 
3422 int
3423 ctl_pool_invalidate(struct ctl_io_pool *pool)
3424 {
3425 
3426 	mtx_assert(&control_softc->ctl_lock, MA_OWNED);
3427 
3428 	if (pool == NULL)
3429 		return (-EINVAL);
3430 
3431 	pool->flags |= CTL_POOL_FLAG_INVALID;
3432 
3433 	return (0);
3434 }
3435 
3436 int
3437 ctl_pool_release(struct ctl_io_pool *pool)
3438 {
3439 
3440 	mtx_assert(&control_softc->ctl_lock, MA_OWNED);
3441 
3442 	if (pool == NULL)
3443 		return (-EINVAL);
3444 
3445 	if ((--pool->refcount == 0)
3446 	 && (pool->flags & CTL_POOL_FLAG_INVALID)) {
3447 		ctl_pool_free(pool->ctl_softc, pool);
3448 	}
3449 
3450 	return (0);
3451 }
3452 
3453 void
3454 ctl_pool_free(struct ctl_softc *ctl_softc, struct ctl_io_pool *pool)
3455 {
3456 	union ctl_io *cur_io, *next_io;
3457 
3458 	mtx_assert(&ctl_softc->ctl_lock, MA_OWNED);
3459 
3460 	for (cur_io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3461 	     cur_io != NULL; cur_io = next_io) {
3462 		next_io = (union ctl_io *)STAILQ_NEXT(&cur_io->io_hdr,
3463 						      links);
3464 		STAILQ_REMOVE(&pool->free_queue, &cur_io->io_hdr, ctl_io_hdr,
3465 			      links);
3466 		free(cur_io, M_CTL);
3467 	}
3468 
3469 	STAILQ_REMOVE(&ctl_softc->io_pools, pool, ctl_io_pool, links);
3470 	ctl_softc->num_pools--;
3471 
3472 	/*
3473 	 * XXX KDM will this decrement the caller's usage count or mine?
3474 	 */
3475 #if 0
3476 	if ((pool->type != CTL_POOL_EMERGENCY)
3477 	 && (pool->type != CTL_POOL_INTERNAL)
3478 	 && (pool->type != CTL_POOL_IOCTL))
3479 		MOD_DEC_USE_COUNT;
3480 #endif
3481 
3482 	free(pool, M_CTL);
3483 }
3484 
3485 /*
3486  * This routine does not block (except for spinlocks of course).
3487  * It tries to allocate a ctl_io union from the caller's pool as quickly as
3488  * possible.
3489  */
3490 union ctl_io *
3491 ctl_alloc_io(void *pool_ref)
3492 {
3493 	union ctl_io *io;
3494 	struct ctl_softc *ctl_softc;
3495 	struct ctl_io_pool *pool, *npool;
3496 	struct ctl_io_pool *emergency_pool;
3497 
3498 	pool = (struct ctl_io_pool *)pool_ref;
3499 
3500 	if (pool == NULL) {
3501 		printf("%s: pool is NULL\n", __func__);
3502 		return (NULL);
3503 	}
3504 
3505 	emergency_pool = NULL;
3506 
3507 	ctl_softc = pool->ctl_softc;
3508 
3509 	mtx_lock(&ctl_softc->ctl_lock);
3510 	/*
3511 	 * First, try to get the io structure from the user's pool.
3512 	 */
3513 	if (ctl_pool_acquire(pool) == 0) {
3514 		io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3515 		if (io != NULL) {
3516 			STAILQ_REMOVE_HEAD(&pool->free_queue, links);
3517 			pool->total_allocated++;
3518 			pool->free_ctl_io--;
3519 			mtx_unlock(&ctl_softc->ctl_lock);
3520 			return (io);
3521 		} else
3522 			ctl_pool_release(pool);
3523 	}
3524 	/*
3525 	 * If he doesn't have any io structures left, search for an
3526 	 * emergency pool and grab one from there.
3527 	 */
3528 	STAILQ_FOREACH(npool, &ctl_softc->io_pools, links) {
3529 		if (npool->type != CTL_POOL_EMERGENCY)
3530 			continue;
3531 
3532 		if (ctl_pool_acquire(npool) != 0)
3533 			continue;
3534 
3535 		emergency_pool = npool;
3536 
3537 		io = (union ctl_io *)STAILQ_FIRST(&npool->free_queue);
3538 		if (io != NULL) {
3539 			STAILQ_REMOVE_HEAD(&npool->free_queue, links);
3540 			npool->total_allocated++;
3541 			npool->free_ctl_io--;
3542 			mtx_unlock(&ctl_softc->ctl_lock);
3543 			return (io);
3544 		} else
3545 			ctl_pool_release(npool);
3546 	}
3547 
3548 	/* Drop the spinlock before we malloc */
3549 	mtx_unlock(&ctl_softc->ctl_lock);
3550 
3551 	/*
3552 	 * The emergency pool (if it exists) didn't have one, so try an
3553 	 * atomic (i.e. nonblocking) malloc and see if we get lucky.
3554 	 */
3555 	io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_NOWAIT);
3556 	if (io != NULL) {
3557 		/*
3558 		 * If the emergency pool exists but is empty, add this
3559 		 * ctl_io to its list when it gets freed.
3560 		 */
3561 		if (emergency_pool != NULL) {
3562 			mtx_lock(&ctl_softc->ctl_lock);
3563 			if (ctl_pool_acquire(emergency_pool) == 0) {
3564 				io->io_hdr.pool = emergency_pool;
3565 				emergency_pool->total_ctl_io++;
3566 				/*
3567 				 * Need to bump this, otherwise
3568 				 * total_allocated and total_freed won't
3569 				 * match when we no longer have anything
3570 				 * outstanding.
3571 				 */
3572 				emergency_pool->total_allocated++;
3573 			}
3574 			mtx_unlock(&ctl_softc->ctl_lock);
3575 		} else
3576 			io->io_hdr.pool = NULL;
3577 	}
3578 
3579 	return (io);
3580 }
3581 
3582 static void
3583 ctl_free_io_internal(union ctl_io *io, int have_lock)
3584 {
3585 	if (io == NULL)
3586 		return;
3587 
3588 	/*
3589 	 * If this ctl_io has a pool, return it to that pool.
3590 	 */
3591 	if (io->io_hdr.pool != NULL) {
3592 		struct ctl_io_pool *pool;
3593 #if 0
3594 		struct ctl_softc *ctl_softc;
3595 		union ctl_io *tmp_io;
3596 		unsigned long xflags;
3597 		int i;
3598 
3599 		ctl_softc = control_softc;
3600 #endif
3601 
3602 		pool = (struct ctl_io_pool *)io->io_hdr.pool;
3603 
3604 		if (have_lock == 0)
3605 			mtx_lock(&pool->ctl_softc->ctl_lock);
3606 #if 0
3607 		save_flags(xflags);
3608 
3609 		for (i = 0, tmp_io = (union ctl_io *)STAILQ_FIRST(
3610 		     &ctl_softc->task_queue); tmp_io != NULL; i++,
3611 		     tmp_io = (union ctl_io *)STAILQ_NEXT(&tmp_io->io_hdr,
3612 		     links)) {
3613 			if (tmp_io == io) {
3614 				printf("%s: %p is still on the task queue!\n",
3615 				       __func__, tmp_io);
3616 				printf("%s: (%d): type %d "
3617 				       "msg %d cdb %x iptl: "
3618 				       "%d:%d:%d:%d tag 0x%04x "
3619 				       "flg %#lx\n",
3620 					__func__, i,
3621 					tmp_io->io_hdr.io_type,
3622 					tmp_io->io_hdr.msg_type,
3623 					tmp_io->scsiio.cdb[0],
3624 					tmp_io->io_hdr.nexus.initid.id,
3625 					tmp_io->io_hdr.nexus.targ_port,
3626 					tmp_io->io_hdr.nexus.targ_target.id,
3627 					tmp_io->io_hdr.nexus.targ_lun,
3628 					(tmp_io->io_hdr.io_type ==
3629 					CTL_IO_TASK) ?
3630 					tmp_io->taskio.tag_num :
3631 					tmp_io->scsiio.tag_num,
3632 					xflags);
3633 				panic("I/O still on the task queue!");
3634 			}
3635 		}
3636 #endif
3637 		io->io_hdr.io_type = 0xff;
3638 		STAILQ_INSERT_TAIL(&pool->free_queue, &io->io_hdr, links);
3639 		pool->total_freed++;
3640 		pool->free_ctl_io++;
3641 		ctl_pool_release(pool);
3642 		if (have_lock == 0)
3643 			mtx_unlock(&pool->ctl_softc->ctl_lock);
3644 	} else {
3645 		/*
3646 		 * Otherwise, just free it.  We probably malloced it and
3647 		 * the emergency pool wasn't available.
3648 		 */
3649 		free(io, M_CTL);
3650 	}
3651 
3652 }
3653 
3654 void
3655 ctl_free_io(union ctl_io *io)
3656 {
3657 	ctl_free_io_internal(io, /*have_lock*/ 0);
3658 }
3659 
3660 void
3661 ctl_zero_io(union ctl_io *io)
3662 {
3663 	void *pool_ref;
3664 
3665 	if (io == NULL)
3666 		return;
3667 
3668 	/*
3669 	 * May need to preserve linked list pointers at some point too.
3670 	 */
3671 	pool_ref = io->io_hdr.pool;
3672 
3673 	memset(io, 0, sizeof(*io));
3674 
3675 	io->io_hdr.pool = pool_ref;
3676 }
3677 
3678 /*
3679  * This routine is currently used for internal copies of ctl_ios that need
3680  * to persist for some reason after we've already returned status to the
3681  * FETD.  (Thus the flag set.)
3682  *
3683  * XXX XXX
3684  * Note that this makes a blind copy of all fields in the ctl_io, except
3685  * for the pool reference.  This includes any memory that has been
3686  * allocated!  That memory will no longer be valid after done has been
3687  * called, so this would be VERY DANGEROUS for command that actually does
3688  * any reads or writes.  Right now (11/7/2005), this is only used for immediate
3689  * start and stop commands, which don't transfer any data, so this is not a
3690  * problem.  If it is used for anything else, the caller would also need to
3691  * allocate data buffer space and this routine would need to be modified to
3692  * copy the data buffer(s) as well.
3693  */
3694 void
3695 ctl_copy_io(union ctl_io *src, union ctl_io *dest)
3696 {
3697 	void *pool_ref;
3698 
3699 	if ((src == NULL)
3700 	 || (dest == NULL))
3701 		return;
3702 
3703 	/*
3704 	 * May need to preserve linked list pointers at some point too.
3705 	 */
3706 	pool_ref = dest->io_hdr.pool;
3707 
3708 	memcpy(dest, src, ctl_min(sizeof(*src), sizeof(*dest)));
3709 
3710 	dest->io_hdr.pool = pool_ref;
3711 	/*
3712 	 * We need to know that this is an internal copy, and doesn't need
3713 	 * to get passed back to the FETD that allocated it.
3714 	 */
3715 	dest->io_hdr.flags |= CTL_FLAG_INT_COPY;
3716 }
3717 
3718 #ifdef NEEDTOPORT
3719 static void
3720 ctl_update_power_subpage(struct copan_power_subpage *page)
3721 {
3722 	int num_luns, num_partitions, config_type;
3723 	struct ctl_softc *softc;
3724 	cs_BOOL_t aor_present, shelf_50pct_power;
3725 	cs_raidset_personality_t rs_type;
3726 	int max_active_luns;
3727 
3728 	softc = control_softc;
3729 
3730 	/* subtract out the processor LUN */
3731 	num_luns = softc->num_luns - 1;
3732 	/*
3733 	 * Default to 7 LUNs active, which was the only number we allowed
3734 	 * in the past.
3735 	 */
3736 	max_active_luns = 7;
3737 
3738 	num_partitions = config_GetRsPartitionInfo();
3739 	config_type = config_GetConfigType();
3740 	shelf_50pct_power = config_GetShelfPowerMode();
3741 	aor_present = config_IsAorRsPresent();
3742 
3743 	rs_type = ddb_GetRsRaidType(1);
3744 	if ((rs_type != CS_RAIDSET_PERSONALITY_RAID5)
3745 	 && (rs_type != CS_RAIDSET_PERSONALITY_RAID1)) {
3746 		EPRINT(0, "Unsupported RS type %d!", rs_type);
3747 	}
3748 
3749 
3750 	page->total_luns = num_luns;
3751 
3752 	switch (config_type) {
3753 	case 40:
3754 		/*
3755 		 * In a 40 drive configuration, it doesn't matter what DC
3756 		 * cards we have, whether we have AOR enabled or not,
3757 		 * partitioning or not, or what type of RAIDset we have.
3758 		 * In that scenario, we can power up every LUN we present
3759 		 * to the user.
3760 		 */
3761 		max_active_luns = num_luns;
3762 
3763 		break;
3764 	case 64:
3765 		if (shelf_50pct_power == CS_FALSE) {
3766 			/* 25% power */
3767 			if (aor_present == CS_TRUE) {
3768 				if (rs_type ==
3769 				     CS_RAIDSET_PERSONALITY_RAID5) {
3770 					max_active_luns = 7;
3771 				} else if (rs_type ==
3772 					 CS_RAIDSET_PERSONALITY_RAID1){
3773 					max_active_luns = 14;
3774 				} else {
3775 					/* XXX KDM now what?? */
3776 				}
3777 			} else {
3778 				if (rs_type ==
3779 				     CS_RAIDSET_PERSONALITY_RAID5) {
3780 					max_active_luns = 8;
3781 				} else if (rs_type ==
3782 					 CS_RAIDSET_PERSONALITY_RAID1){
3783 					max_active_luns = 16;
3784 				} else {
3785 					/* XXX KDM now what?? */
3786 				}
3787 			}
3788 		} else {
3789 			/* 50% power */
3790 			/*
3791 			 * With 50% power in a 64 drive configuration, we
3792 			 * can power all LUNs we present.
3793 			 */
3794 			max_active_luns = num_luns;
3795 		}
3796 		break;
3797 	case 112:
3798 		if (shelf_50pct_power == CS_FALSE) {
3799 			/* 25% power */
3800 			if (aor_present == CS_TRUE) {
3801 				if (rs_type ==
3802 				     CS_RAIDSET_PERSONALITY_RAID5) {
3803 					max_active_luns = 7;
3804 				} else if (rs_type ==
3805 					 CS_RAIDSET_PERSONALITY_RAID1){
3806 					max_active_luns = 14;
3807 				} else {
3808 					/* XXX KDM now what?? */
3809 				}
3810 			} else {
3811 				if (rs_type ==
3812 				     CS_RAIDSET_PERSONALITY_RAID5) {
3813 					max_active_luns = 8;
3814 				} else if (rs_type ==
3815 					 CS_RAIDSET_PERSONALITY_RAID1){
3816 					max_active_luns = 16;
3817 				} else {
3818 					/* XXX KDM now what?? */
3819 				}
3820 			}
3821 		} else {
3822 			/* 50% power */
3823 			if (aor_present == CS_TRUE) {
3824 				if (rs_type ==
3825 				     CS_RAIDSET_PERSONALITY_RAID5) {
3826 					max_active_luns = 14;
3827 				} else if (rs_type ==
3828 					 CS_RAIDSET_PERSONALITY_RAID1){
3829 					/*
3830 					 * We're assuming here that disk
3831 					 * caching is enabled, and so we're
3832 					 * able to power up half of each
3833 					 * LUN, and cache all writes.
3834 					 */
3835 					max_active_luns = num_luns;
3836 				} else {
3837 					/* XXX KDM now what?? */
3838 				}
3839 			} else {
3840 				if (rs_type ==
3841 				     CS_RAIDSET_PERSONALITY_RAID5) {
3842 					max_active_luns = 15;
3843 				} else if (rs_type ==
3844 					 CS_RAIDSET_PERSONALITY_RAID1){
3845 					max_active_luns = 30;
3846 				} else {
3847 					/* XXX KDM now what?? */
3848 				}
3849 			}
3850 		}
3851 		break;
3852 	default:
3853 		/*
3854 		 * In this case, we have an unknown configuration, so we
3855 		 * just use the default from above.
3856 		 */
3857 		break;
3858 	}
3859 
3860 	page->max_active_luns = max_active_luns;
3861 #if 0
3862 	printk("%s: total_luns = %d, max_active_luns = %d\n", __func__,
3863 	       page->total_luns, page->max_active_luns);
3864 #endif
3865 }
3866 #endif /* NEEDTOPORT */
3867 
3868 /*
3869  * This routine could be used in the future to load default and/or saved
3870  * mode page parameters for a particuar lun.
3871  */
3872 static int
3873 ctl_init_page_index(struct ctl_lun *lun)
3874 {
3875 	int i;
3876 	struct ctl_page_index *page_index;
3877 	struct ctl_softc *softc;
3878 
3879 	memcpy(&lun->mode_pages.index, page_index_template,
3880 	       sizeof(page_index_template));
3881 
3882 	softc = lun->ctl_softc;
3883 
3884 	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
3885 
3886 		page_index = &lun->mode_pages.index[i];
3887 		/*
3888 		 * If this is a disk-only mode page, there's no point in
3889 		 * setting it up.  For some pages, we have to have some
3890 		 * basic information about the disk in order to calculate the
3891 		 * mode page data.
3892 		 */
3893 		if ((lun->be_lun->lun_type != T_DIRECT)
3894 		 && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
3895 			continue;
3896 
3897 		switch (page_index->page_code & SMPH_PC_MASK) {
3898 		case SMS_FORMAT_DEVICE_PAGE: {
3899 			struct scsi_format_page *format_page;
3900 
3901 			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3902 				panic("subpage is incorrect!");
3903 
3904 			/*
3905 			 * Sectors per track are set above.  Bytes per
3906 			 * sector need to be set here on a per-LUN basis.
3907 			 */
3908 			memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
3909 			       &format_page_default,
3910 			       sizeof(format_page_default));
3911 			memcpy(&lun->mode_pages.format_page[
3912 			       CTL_PAGE_CHANGEABLE], &format_page_changeable,
3913 			       sizeof(format_page_changeable));
3914 			memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
3915 			       &format_page_default,
3916 			       sizeof(format_page_default));
3917 			memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
3918 			       &format_page_default,
3919 			       sizeof(format_page_default));
3920 
3921 			format_page = &lun->mode_pages.format_page[
3922 				CTL_PAGE_CURRENT];
3923 			scsi_ulto2b(lun->be_lun->blocksize,
3924 				    format_page->bytes_per_sector);
3925 
3926 			format_page = &lun->mode_pages.format_page[
3927 				CTL_PAGE_DEFAULT];
3928 			scsi_ulto2b(lun->be_lun->blocksize,
3929 				    format_page->bytes_per_sector);
3930 
3931 			format_page = &lun->mode_pages.format_page[
3932 				CTL_PAGE_SAVED];
3933 			scsi_ulto2b(lun->be_lun->blocksize,
3934 				    format_page->bytes_per_sector);
3935 
3936 			page_index->page_data =
3937 				(uint8_t *)lun->mode_pages.format_page;
3938 			break;
3939 		}
3940 		case SMS_RIGID_DISK_PAGE: {
3941 			struct scsi_rigid_disk_page *rigid_disk_page;
3942 			uint32_t sectors_per_cylinder;
3943 			uint64_t cylinders;
3944 #ifndef	__XSCALE__
3945 			int shift;
3946 #endif /* !__XSCALE__ */
3947 
3948 			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3949 				panic("invalid subpage value %d",
3950 				      page_index->subpage);
3951 
3952 			/*
3953 			 * Rotation rate and sectors per track are set
3954 			 * above.  We calculate the cylinders here based on
3955 			 * capacity.  Due to the number of heads and
3956 			 * sectors per track we're using, smaller arrays
3957 			 * may turn out to have 0 cylinders.  Linux and
3958 			 * FreeBSD don't pay attention to these mode pages
3959 			 * to figure out capacity, but Solaris does.  It
3960 			 * seems to deal with 0 cylinders just fine, and
3961 			 * works out a fake geometry based on the capacity.
3962 			 */
3963 			memcpy(&lun->mode_pages.rigid_disk_page[
3964 			       CTL_PAGE_CURRENT], &rigid_disk_page_default,
3965 			       sizeof(rigid_disk_page_default));
3966 			memcpy(&lun->mode_pages.rigid_disk_page[
3967 			       CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
3968 			       sizeof(rigid_disk_page_changeable));
3969 			memcpy(&lun->mode_pages.rigid_disk_page[
3970 			       CTL_PAGE_DEFAULT], &rigid_disk_page_default,
3971 			       sizeof(rigid_disk_page_default));
3972 			memcpy(&lun->mode_pages.rigid_disk_page[
3973 			       CTL_PAGE_SAVED], &rigid_disk_page_default,
3974 			       sizeof(rigid_disk_page_default));
3975 
3976 			sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
3977 				CTL_DEFAULT_HEADS;
3978 
3979 			/*
3980 			 * The divide method here will be more accurate,
3981 			 * probably, but results in floating point being
3982 			 * used in the kernel on i386 (__udivdi3()).  On the
3983 			 * XScale, though, __udivdi3() is implemented in
3984 			 * software.
3985 			 *
3986 			 * The shift method for cylinder calculation is
3987 			 * accurate if sectors_per_cylinder is a power of
3988 			 * 2.  Otherwise it might be slightly off -- you
3989 			 * might have a bit of a truncation problem.
3990 			 */
3991 #ifdef	__XSCALE__
3992 			cylinders = (lun->be_lun->maxlba + 1) /
3993 				sectors_per_cylinder;
3994 #else
3995 			for (shift = 31; shift > 0; shift--) {
3996 				if (sectors_per_cylinder & (1 << shift))
3997 					break;
3998 			}
3999 			cylinders = (lun->be_lun->maxlba + 1) >> shift;
4000 #endif
4001 
4002 			/*
4003 			 * We've basically got 3 bytes, or 24 bits for the
4004 			 * cylinder size in the mode page.  If we're over,
4005 			 * just round down to 2^24.
4006 			 */
4007 			if (cylinders > 0xffffff)
4008 				cylinders = 0xffffff;
4009 
4010 			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4011 				CTL_PAGE_CURRENT];
4012 			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4013 
4014 			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4015 				CTL_PAGE_DEFAULT];
4016 			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4017 
4018 			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4019 				CTL_PAGE_SAVED];
4020 			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4021 
4022 			page_index->page_data =
4023 				(uint8_t *)lun->mode_pages.rigid_disk_page;
4024 			break;
4025 		}
4026 		case SMS_CACHING_PAGE: {
4027 
4028 			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4029 				panic("invalid subpage value %d",
4030 				      page_index->subpage);
4031 			/*
4032 			 * Defaults should be okay here, no calculations
4033 			 * needed.
4034 			 */
4035 			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4036 			       &caching_page_default,
4037 			       sizeof(caching_page_default));
4038 			memcpy(&lun->mode_pages.caching_page[
4039 			       CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4040 			       sizeof(caching_page_changeable));
4041 			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4042 			       &caching_page_default,
4043 			       sizeof(caching_page_default));
4044 			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4045 			       &caching_page_default,
4046 			       sizeof(caching_page_default));
4047 			page_index->page_data =
4048 				(uint8_t *)lun->mode_pages.caching_page;
4049 			break;
4050 		}
4051 		case SMS_CONTROL_MODE_PAGE: {
4052 
4053 			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4054 				panic("invalid subpage value %d",
4055 				      page_index->subpage);
4056 
4057 			/*
4058 			 * Defaults should be okay here, no calculations
4059 			 * needed.
4060 			 */
4061 			memcpy(&lun->mode_pages.control_page[CTL_PAGE_CURRENT],
4062 			       &control_page_default,
4063 			       sizeof(control_page_default));
4064 			memcpy(&lun->mode_pages.control_page[
4065 			       CTL_PAGE_CHANGEABLE], &control_page_changeable,
4066 			       sizeof(control_page_changeable));
4067 			memcpy(&lun->mode_pages.control_page[CTL_PAGE_DEFAULT],
4068 			       &control_page_default,
4069 			       sizeof(control_page_default));
4070 			memcpy(&lun->mode_pages.control_page[CTL_PAGE_SAVED],
4071 			       &control_page_default,
4072 			       sizeof(control_page_default));
4073 			page_index->page_data =
4074 				(uint8_t *)lun->mode_pages.control_page;
4075 			break;
4076 
4077 		}
4078 		case SMS_VENDOR_SPECIFIC_PAGE:{
4079 			switch (page_index->subpage) {
4080 			case PWR_SUBPAGE_CODE: {
4081 				struct copan_power_subpage *current_page,
4082 							   *saved_page;
4083 
4084 				memcpy(&lun->mode_pages.power_subpage[
4085 				       CTL_PAGE_CURRENT],
4086 				       &power_page_default,
4087 				       sizeof(power_page_default));
4088 				memcpy(&lun->mode_pages.power_subpage[
4089 				       CTL_PAGE_CHANGEABLE],
4090 				       &power_page_changeable,
4091 				       sizeof(power_page_changeable));
4092 				memcpy(&lun->mode_pages.power_subpage[
4093 				       CTL_PAGE_DEFAULT],
4094 				       &power_page_default,
4095 				       sizeof(power_page_default));
4096 				memcpy(&lun->mode_pages.power_subpage[
4097 				       CTL_PAGE_SAVED],
4098 				       &power_page_default,
4099 				       sizeof(power_page_default));
4100 				page_index->page_data =
4101 				    (uint8_t *)lun->mode_pages.power_subpage;
4102 
4103 				current_page = (struct copan_power_subpage *)
4104 					(page_index->page_data +
4105 					 (page_index->page_len *
4106 					  CTL_PAGE_CURRENT));
4107 			        saved_page = (struct copan_power_subpage *)
4108 				        (page_index->page_data +
4109 					 (page_index->page_len *
4110 					  CTL_PAGE_SAVED));
4111 				break;
4112 			}
4113 			case APS_SUBPAGE_CODE: {
4114 				struct copan_aps_subpage *current_page,
4115 							 *saved_page;
4116 
4117 				// This gets set multiple times but
4118 				// it should always be the same. It's
4119 				// only done during init so who cares.
4120 				index_to_aps_page = i;
4121 
4122 				memcpy(&lun->mode_pages.aps_subpage[
4123 				       CTL_PAGE_CURRENT],
4124 				       &aps_page_default,
4125 				       sizeof(aps_page_default));
4126 				memcpy(&lun->mode_pages.aps_subpage[
4127 				       CTL_PAGE_CHANGEABLE],
4128 				       &aps_page_changeable,
4129 				       sizeof(aps_page_changeable));
4130 				memcpy(&lun->mode_pages.aps_subpage[
4131 				       CTL_PAGE_DEFAULT],
4132 				       &aps_page_default,
4133 				       sizeof(aps_page_default));
4134 				memcpy(&lun->mode_pages.aps_subpage[
4135 				       CTL_PAGE_SAVED],
4136 				       &aps_page_default,
4137 				       sizeof(aps_page_default));
4138 				page_index->page_data =
4139 					(uint8_t *)lun->mode_pages.aps_subpage;
4140 
4141 				current_page = (struct copan_aps_subpage *)
4142 					(page_index->page_data +
4143 					 (page_index->page_len *
4144 					  CTL_PAGE_CURRENT));
4145 				saved_page = (struct copan_aps_subpage *)
4146 					(page_index->page_data +
4147 					 (page_index->page_len *
4148 					  CTL_PAGE_SAVED));
4149 				break;
4150 			}
4151 			case DBGCNF_SUBPAGE_CODE: {
4152 				struct copan_debugconf_subpage *current_page,
4153 							       *saved_page;
4154 
4155 				memcpy(&lun->mode_pages.debugconf_subpage[
4156 				       CTL_PAGE_CURRENT],
4157 				       &debugconf_page_default,
4158 				       sizeof(debugconf_page_default));
4159 				memcpy(&lun->mode_pages.debugconf_subpage[
4160 				       CTL_PAGE_CHANGEABLE],
4161 				       &debugconf_page_changeable,
4162 				       sizeof(debugconf_page_changeable));
4163 				memcpy(&lun->mode_pages.debugconf_subpage[
4164 				       CTL_PAGE_DEFAULT],
4165 				       &debugconf_page_default,
4166 				       sizeof(debugconf_page_default));
4167 				memcpy(&lun->mode_pages.debugconf_subpage[
4168 				       CTL_PAGE_SAVED],
4169 				       &debugconf_page_default,
4170 				       sizeof(debugconf_page_default));
4171 				page_index->page_data =
4172 					(uint8_t *)lun->mode_pages.debugconf_subpage;
4173 
4174 				current_page = (struct copan_debugconf_subpage *)
4175 					(page_index->page_data +
4176 					 (page_index->page_len *
4177 					  CTL_PAGE_CURRENT));
4178 				saved_page = (struct copan_debugconf_subpage *)
4179 					(page_index->page_data +
4180 					 (page_index->page_len *
4181 					  CTL_PAGE_SAVED));
4182 				break;
4183 			}
4184 			default:
4185 				panic("invalid subpage value %d",
4186 				      page_index->subpage);
4187 				break;
4188 			}
4189    			break;
4190 		}
4191 		default:
4192 			panic("invalid page value %d",
4193 			      page_index->page_code & SMPH_PC_MASK);
4194 			break;
4195     	}
4196 	}
4197 
4198 	return (CTL_RETVAL_COMPLETE);
4199 }
4200 
4201 /*
4202  * LUN allocation.
4203  *
4204  * Requirements:
4205  * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4206  *   wants us to allocate the LUN and he can block.
4207  * - ctl_softc is always set
4208  * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4209  *
4210  * Returns 0 for success, non-zero (errno) for failure.
4211  */
4212 static int
4213 ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4214 	      struct ctl_be_lun *const be_lun, struct ctl_id target_id)
4215 {
4216 	struct ctl_lun *nlun, *lun;
4217 	struct ctl_frontend *fe;
4218 	int lun_number, i, lun_malloced;
4219 
4220 	if (be_lun == NULL)
4221 		return (EINVAL);
4222 
4223 	/*
4224 	 * We currently only support Direct Access or Processor LUN types.
4225 	 */
4226 	switch (be_lun->lun_type) {
4227 	case T_DIRECT:
4228 		break;
4229 	case T_PROCESSOR:
4230 		break;
4231 	case T_SEQUENTIAL:
4232 	case T_CHANGER:
4233 	default:
4234 		be_lun->lun_config_status(be_lun->be_lun,
4235 					  CTL_LUN_CONFIG_FAILURE);
4236 		break;
4237 	}
4238 	if (ctl_lun == NULL) {
4239 		lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4240 		lun_malloced = 1;
4241 	} else {
4242 		lun_malloced = 0;
4243 		lun = ctl_lun;
4244 	}
4245 
4246 	memset(lun, 0, sizeof(*lun));
4247 	if (lun_malloced)
4248 		lun->flags = CTL_LUN_MALLOCED;
4249 
4250 	mtx_lock(&ctl_softc->ctl_lock);
4251 	/*
4252 	 * See if the caller requested a particular LUN number.  If so, see
4253 	 * if it is available.  Otherwise, allocate the first available LUN.
4254 	 */
4255 	if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4256 		if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4257 		 || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4258 			mtx_unlock(&ctl_softc->ctl_lock);
4259 			if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4260 				printf("ctl: requested LUN ID %d is higher "
4261 				       "than CTL_MAX_LUNS - 1 (%d)\n",
4262 				       be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4263 			} else {
4264 				/*
4265 				 * XXX KDM return an error, or just assign
4266 				 * another LUN ID in this case??
4267 				 */
4268 				printf("ctl: requested LUN ID %d is already "
4269 				       "in use\n", be_lun->req_lun_id);
4270 			}
4271 			if (lun->flags & CTL_LUN_MALLOCED)
4272 				free(lun, M_CTL);
4273 			be_lun->lun_config_status(be_lun->be_lun,
4274 						  CTL_LUN_CONFIG_FAILURE);
4275 			return (ENOSPC);
4276 		}
4277 		lun_number = be_lun->req_lun_id;
4278 	} else {
4279 		lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, CTL_MAX_LUNS);
4280 		if (lun_number == -1) {
4281 			mtx_unlock(&ctl_softc->ctl_lock);
4282 			printf("ctl: can't allocate LUN on target %ju, out of "
4283 			       "LUNs\n", (uintmax_t)target_id.id);
4284 			if (lun->flags & CTL_LUN_MALLOCED)
4285 				free(lun, M_CTL);
4286 			be_lun->lun_config_status(be_lun->be_lun,
4287 						  CTL_LUN_CONFIG_FAILURE);
4288 			return (ENOSPC);
4289 		}
4290 	}
4291 	ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4292 
4293 	lun->target = target_id;
4294 	lun->lun = lun_number;
4295 	lun->be_lun = be_lun;
4296 	/*
4297 	 * The processor LUN is always enabled.  Disk LUNs come on line
4298 	 * disabled, and must be enabled by the backend.
4299 	 */
4300 	lun->flags |= CTL_LUN_DISABLED;
4301 	lun->backend = be_lun->be;
4302 	be_lun->ctl_lun = lun;
4303 	be_lun->lun_id = lun_number;
4304 	atomic_add_int(&be_lun->be->num_luns, 1);
4305 	if (be_lun->flags & CTL_LUN_FLAG_POWERED_OFF)
4306 		lun->flags |= CTL_LUN_STOPPED;
4307 
4308 	if (be_lun->flags & CTL_LUN_FLAG_INOPERABLE)
4309 		lun->flags |= CTL_LUN_INOPERABLE;
4310 
4311 	if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4312 		lun->flags |= CTL_LUN_PRIMARY_SC;
4313 
4314 	lun->ctl_softc = ctl_softc;
4315 	TAILQ_INIT(&lun->ooa_queue);
4316 	TAILQ_INIT(&lun->blocked_queue);
4317 	STAILQ_INIT(&lun->error_list);
4318 
4319 	/*
4320 	 * Initialize the mode page index.
4321 	 */
4322 	ctl_init_page_index(lun);
4323 
4324 	/*
4325 	 * Set the poweron UA for all initiators on this LUN only.
4326 	 */
4327 	for (i = 0; i < CTL_MAX_INITIATORS; i++)
4328 		lun->pending_sense[i].ua_pending = CTL_UA_POWERON;
4329 
4330 	/*
4331 	 * Now, before we insert this lun on the lun list, set the lun
4332 	 * inventory changed UA for all other luns.
4333 	 */
4334 	STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4335 		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4336 			nlun->pending_sense[i].ua_pending |= CTL_UA_LUN_CHANGE;
4337 		}
4338 	}
4339 
4340 	STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4341 
4342 	ctl_softc->ctl_luns[lun_number] = lun;
4343 
4344 	ctl_softc->num_luns++;
4345 
4346 	/* Setup statistics gathering */
4347 	lun->stats.device_type = be_lun->lun_type;
4348 	lun->stats.lun_number = lun_number;
4349 	if (lun->stats.device_type == T_DIRECT)
4350 		lun->stats.blocksize = be_lun->blocksize;
4351 	else
4352 		lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4353 	for (i = 0;i < CTL_MAX_PORTS;i++)
4354 		lun->stats.ports[i].targ_port = i;
4355 
4356 	mtx_unlock(&ctl_softc->ctl_lock);
4357 
4358 	lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4359 
4360 	/*
4361 	 * Run through each registered FETD and bring it online if it isn't
4362 	 * already.  Enable the target ID if it hasn't been enabled, and
4363 	 * enable this particular LUN.
4364 	 */
4365 	STAILQ_FOREACH(fe, &ctl_softc->fe_list, links) {
4366 		int retval;
4367 
4368 		/*
4369 		 * XXX KDM this only works for ONE TARGET ID.  We'll need
4370 		 * to do things differently if we go to a multiple target
4371 		 * ID scheme.
4372 		 */
4373 		if ((fe->status & CTL_PORT_STATUS_TARG_ONLINE) == 0) {
4374 
4375 			retval = fe->targ_enable(fe->targ_lun_arg, target_id);
4376 			if (retval != 0) {
4377 				printf("ctl_alloc_lun: FETD %s port %d "
4378 				       "returned error %d for targ_enable on "
4379 				       "target %ju\n", fe->port_name,
4380 				       fe->targ_port, retval,
4381 				       (uintmax_t)target_id.id);
4382 			} else
4383 				fe->status |= CTL_PORT_STATUS_TARG_ONLINE;
4384 		}
4385 
4386 		retval = fe->lun_enable(fe->targ_lun_arg, target_id,lun_number);
4387 		if (retval != 0) {
4388 			printf("ctl_alloc_lun: FETD %s port %d returned error "
4389 			       "%d for lun_enable on target %ju lun %d\n",
4390 			       fe->port_name, fe->targ_port, retval,
4391 			       (uintmax_t)target_id.id, lun_number);
4392 		} else
4393 			fe->status |= CTL_PORT_STATUS_LUN_ONLINE;
4394 	}
4395 	return (0);
4396 }
4397 
4398 /*
4399  * Delete a LUN.
4400  * Assumptions:
4401  * - LUN has already been marked invalid and any pending I/O has been taken
4402  *   care of.
4403  */
4404 static int
4405 ctl_free_lun(struct ctl_lun *lun)
4406 {
4407 	struct ctl_softc *softc;
4408 #if 0
4409 	struct ctl_frontend *fe;
4410 #endif
4411 	struct ctl_lun *nlun;
4412 	union ctl_io *io, *next_io;
4413 	int i;
4414 
4415 	softc = lun->ctl_softc;
4416 
4417 	mtx_assert(&softc->ctl_lock, MA_OWNED);
4418 
4419 	STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4420 
4421 	ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4422 
4423 	softc->ctl_luns[lun->lun] = NULL;
4424 
4425 	if (TAILQ_FIRST(&lun->ooa_queue) != NULL) {
4426 		printf("ctl_free_lun: aieee!! freeing a LUN with "
4427 		       "outstanding I/O!!\n");
4428 	}
4429 
4430 	/*
4431 	 * If we have anything pending on the RtR queue, remove it.
4432 	 */
4433 	for (io = (union ctl_io *)STAILQ_FIRST(&softc->rtr_queue); io != NULL;
4434 	     io = next_io) {
4435 		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
4436 		if ((io->io_hdr.nexus.targ_target.id == lun->target.id)
4437 		 && (io->io_hdr.nexus.targ_lun == lun->lun))
4438 			STAILQ_REMOVE(&softc->rtr_queue, &io->io_hdr,
4439 				      ctl_io_hdr, links);
4440 	}
4441 
4442 	/*
4443 	 * Then remove everything from the blocked queue.
4444 	 */
4445 	for (io = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue); io != NULL;
4446 	     io = next_io) {
4447 		next_io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,blocked_links);
4448 		TAILQ_REMOVE(&lun->blocked_queue, &io->io_hdr, blocked_links);
4449 		io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
4450 	}
4451 
4452 	/*
4453 	 * Now clear out the OOA queue, and free all the I/O.
4454 	 * XXX KDM should we notify the FETD here?  We probably need to
4455 	 * quiesce the LUN before deleting it.
4456 	 */
4457 	for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); io != NULL;
4458 	     io = next_io) {
4459 		next_io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr, ooa_links);
4460 		TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
4461 		ctl_free_io_internal(io, /*have_lock*/ 1);
4462 	}
4463 
4464 	softc->num_luns--;
4465 
4466 	/*
4467 	 * XXX KDM this scheme only works for a single target/multiple LUN
4468 	 * setup.  It needs to be revamped for a multiple target scheme.
4469 	 *
4470 	 * XXX KDM this results in fe->lun_disable() getting called twice,
4471 	 * once when ctl_disable_lun() is called, and a second time here.
4472 	 * We really need to re-think the LUN disable semantics.  There
4473 	 * should probably be several steps/levels to LUN removal:
4474 	 *  - disable
4475 	 *  - invalidate
4476 	 *  - free
4477  	 *
4478 	 * Right now we only have a disable method when communicating to
4479 	 * the front end ports, at least for individual LUNs.
4480 	 */
4481 #if 0
4482 	STAILQ_FOREACH(fe, &softc->fe_list, links) {
4483 		int retval;
4484 
4485 		retval = fe->lun_disable(fe->targ_lun_arg, lun->target,
4486 					 lun->lun);
4487 		if (retval != 0) {
4488 			printf("ctl_free_lun: FETD %s port %d returned error "
4489 			       "%d for lun_disable on target %ju lun %jd\n",
4490 			       fe->port_name, fe->targ_port, retval,
4491 			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4492 		}
4493 
4494 		if (STAILQ_FIRST(&softc->lun_list) == NULL) {
4495 			fe->status &= ~CTL_PORT_STATUS_LUN_ONLINE;
4496 
4497 			retval = fe->targ_disable(fe->targ_lun_arg,lun->target);
4498 			if (retval != 0) {
4499 				printf("ctl_free_lun: FETD %s port %d "
4500 				       "returned error %d for targ_disable on "
4501 				       "target %ju\n", fe->port_name,
4502 				       fe->targ_port, retval,
4503 				       (uintmax_t)lun->target.id);
4504 			} else
4505 				fe->status &= ~CTL_PORT_STATUS_TARG_ONLINE;
4506 
4507 			if ((fe->status & CTL_PORT_STATUS_TARG_ONLINE) != 0)
4508 				continue;
4509 
4510 #if 0
4511 			fe->port_offline(fe->onoff_arg);
4512 			fe->status &= ~CTL_PORT_STATUS_ONLINE;
4513 #endif
4514 		}
4515 	}
4516 #endif
4517 
4518 	/*
4519 	 * Tell the backend to free resources, if this LUN has a backend.
4520 	 */
4521 	atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4522 	lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4523 
4524 	if (lun->flags & CTL_LUN_MALLOCED)
4525 		free(lun, M_CTL);
4526 
4527 	STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4528 		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4529 			nlun->pending_sense[i].ua_pending |= CTL_UA_LUN_CHANGE;
4530 		}
4531 	}
4532 
4533 	return (0);
4534 }
4535 
4536 static void
4537 ctl_create_lun(struct ctl_be_lun *be_lun)
4538 {
4539 	struct ctl_softc *ctl_softc;
4540 
4541 	ctl_softc = control_softc;
4542 
4543 	/*
4544 	 * ctl_alloc_lun() should handle all potential failure cases.
4545 	 */
4546 	ctl_alloc_lun(ctl_softc, NULL, be_lun, ctl_softc->target);
4547 }
4548 
4549 int
4550 ctl_add_lun(struct ctl_be_lun *be_lun)
4551 {
4552 	struct ctl_softc *ctl_softc;
4553 
4554 	ctl_softc = control_softc;
4555 
4556 	mtx_lock(&ctl_softc->ctl_lock);
4557 	STAILQ_INSERT_TAIL(&ctl_softc->pending_lun_queue, be_lun, links);
4558 	mtx_unlock(&ctl_softc->ctl_lock);
4559 
4560 	ctl_wakeup_thread();
4561 
4562 	return (0);
4563 }
4564 
4565 int
4566 ctl_enable_lun(struct ctl_be_lun *be_lun)
4567 {
4568 	struct ctl_softc *ctl_softc;
4569 	struct ctl_frontend *fe, *nfe;
4570 	struct ctl_lun *lun;
4571 	int retval;
4572 
4573 	ctl_softc = control_softc;
4574 
4575 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4576 
4577 	mtx_lock(&ctl_softc->ctl_lock);
4578 	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4579 		/*
4580 		 * eh?  Why did we get called if the LUN is already
4581 		 * enabled?
4582 		 */
4583 		mtx_unlock(&ctl_softc->ctl_lock);
4584 		return (0);
4585 	}
4586 	lun->flags &= ~CTL_LUN_DISABLED;
4587 
4588 	for (fe = STAILQ_FIRST(&ctl_softc->fe_list); fe != NULL; fe = nfe) {
4589 		nfe = STAILQ_NEXT(fe, links);
4590 
4591 		/*
4592 		 * Drop the lock while we call the FETD's enable routine.
4593 		 * This can lead to a callback into CTL (at least in the
4594 		 * case of the internal initiator frontend.
4595 		 */
4596 		mtx_unlock(&ctl_softc->ctl_lock);
4597 		retval = fe->lun_enable(fe->targ_lun_arg, lun->target,lun->lun);
4598 		mtx_lock(&ctl_softc->ctl_lock);
4599 		if (retval != 0) {
4600 			printf("%s: FETD %s port %d returned error "
4601 			       "%d for lun_enable on target %ju lun %jd\n",
4602 			       __func__, fe->port_name, fe->targ_port, retval,
4603 			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4604 		}
4605 #if 0
4606 		 else {
4607             /* NOTE:  TODO:  why does lun enable affect port status? */
4608 			fe->status |= CTL_PORT_STATUS_LUN_ONLINE;
4609 		}
4610 #endif
4611 	}
4612 
4613 	mtx_unlock(&ctl_softc->ctl_lock);
4614 
4615 	return (0);
4616 }
4617 
4618 int
4619 ctl_disable_lun(struct ctl_be_lun *be_lun)
4620 {
4621 	struct ctl_softc *ctl_softc;
4622 	struct ctl_frontend *fe;
4623 	struct ctl_lun *lun;
4624 	int retval;
4625 
4626 	ctl_softc = control_softc;
4627 
4628 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4629 
4630 	mtx_lock(&ctl_softc->ctl_lock);
4631 
4632 	if (lun->flags & CTL_LUN_DISABLED) {
4633 		mtx_unlock(&ctl_softc->ctl_lock);
4634 		return (0);
4635 	}
4636 	lun->flags |= CTL_LUN_DISABLED;
4637 
4638 	STAILQ_FOREACH(fe, &ctl_softc->fe_list, links) {
4639 		mtx_unlock(&ctl_softc->ctl_lock);
4640 		/*
4641 		 * Drop the lock before we call the frontend's disable
4642 		 * routine, to avoid lock order reversals.
4643 		 *
4644 		 * XXX KDM what happens if the frontend list changes while
4645 		 * we're traversing it?  It's unlikely, but should be handled.
4646 		 */
4647 		retval = fe->lun_disable(fe->targ_lun_arg, lun->target,
4648 					 lun->lun);
4649 		mtx_lock(&ctl_softc->ctl_lock);
4650 		if (retval != 0) {
4651 			printf("ctl_alloc_lun: FETD %s port %d returned error "
4652 			       "%d for lun_disable on target %ju lun %jd\n",
4653 			       fe->port_name, fe->targ_port, retval,
4654 			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4655 		}
4656 	}
4657 
4658 	mtx_unlock(&ctl_softc->ctl_lock);
4659 
4660 	return (0);
4661 }
4662 
4663 int
4664 ctl_start_lun(struct ctl_be_lun *be_lun)
4665 {
4666 	struct ctl_softc *ctl_softc;
4667 	struct ctl_lun *lun;
4668 
4669 	ctl_softc = control_softc;
4670 
4671 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4672 
4673 	mtx_lock(&ctl_softc->ctl_lock);
4674 	lun->flags &= ~CTL_LUN_STOPPED;
4675 	mtx_unlock(&ctl_softc->ctl_lock);
4676 
4677 	return (0);
4678 }
4679 
4680 int
4681 ctl_stop_lun(struct ctl_be_lun *be_lun)
4682 {
4683 	struct ctl_softc *ctl_softc;
4684 	struct ctl_lun *lun;
4685 
4686 	ctl_softc = control_softc;
4687 
4688 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4689 
4690 	mtx_lock(&ctl_softc->ctl_lock);
4691 	lun->flags |= CTL_LUN_STOPPED;
4692 	mtx_unlock(&ctl_softc->ctl_lock);
4693 
4694 	return (0);
4695 }
4696 
4697 int
4698 ctl_lun_offline(struct ctl_be_lun *be_lun)
4699 {
4700 	struct ctl_softc *ctl_softc;
4701 	struct ctl_lun *lun;
4702 
4703 	ctl_softc = control_softc;
4704 
4705 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4706 
4707 	mtx_lock(&ctl_softc->ctl_lock);
4708 	lun->flags |= CTL_LUN_OFFLINE;
4709 	mtx_unlock(&ctl_softc->ctl_lock);
4710 
4711 	return (0);
4712 }
4713 
4714 int
4715 ctl_lun_online(struct ctl_be_lun *be_lun)
4716 {
4717 	struct ctl_softc *ctl_softc;
4718 	struct ctl_lun *lun;
4719 
4720 	ctl_softc = control_softc;
4721 
4722 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4723 
4724 	mtx_lock(&ctl_softc->ctl_lock);
4725 	lun->flags &= ~CTL_LUN_OFFLINE;
4726 	mtx_unlock(&ctl_softc->ctl_lock);
4727 
4728 	return (0);
4729 }
4730 
4731 int
4732 ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4733 {
4734 	struct ctl_softc *ctl_softc;
4735 	struct ctl_lun *lun;
4736 
4737 	ctl_softc = control_softc;
4738 
4739 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4740 
4741 	mtx_lock(&ctl_softc->ctl_lock);
4742 
4743 	/*
4744 	 * The LUN needs to be disabled before it can be marked invalid.
4745 	 */
4746 	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4747 		mtx_unlock(&ctl_softc->ctl_lock);
4748 		return (-1);
4749 	}
4750 	/*
4751 	 * Mark the LUN invalid.
4752 	 */
4753 	lun->flags |= CTL_LUN_INVALID;
4754 
4755 	/*
4756 	 * If there is nothing in the OOA queue, go ahead and free the LUN.
4757 	 * If we have something in the OOA queue, we'll free it when the
4758 	 * last I/O completes.
4759 	 */
4760 	if (TAILQ_FIRST(&lun->ooa_queue) == NULL)
4761 		ctl_free_lun(lun);
4762 	mtx_unlock(&ctl_softc->ctl_lock);
4763 
4764 	return (0);
4765 }
4766 
4767 int
4768 ctl_lun_inoperable(struct ctl_be_lun *be_lun)
4769 {
4770 	struct ctl_softc *ctl_softc;
4771 	struct ctl_lun *lun;
4772 
4773 	ctl_softc = control_softc;
4774 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4775 
4776 	mtx_lock(&ctl_softc->ctl_lock);
4777 	lun->flags |= CTL_LUN_INOPERABLE;
4778 	mtx_unlock(&ctl_softc->ctl_lock);
4779 
4780 	return (0);
4781 }
4782 
4783 int
4784 ctl_lun_operable(struct ctl_be_lun *be_lun)
4785 {
4786 	struct ctl_softc *ctl_softc;
4787 	struct ctl_lun *lun;
4788 
4789 	ctl_softc = control_softc;
4790 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4791 
4792 	mtx_lock(&ctl_softc->ctl_lock);
4793 	lun->flags &= ~CTL_LUN_INOPERABLE;
4794 	mtx_unlock(&ctl_softc->ctl_lock);
4795 
4796 	return (0);
4797 }
4798 
4799 int
4800 ctl_lun_power_lock(struct ctl_be_lun *be_lun, struct ctl_nexus *nexus,
4801 		   int lock)
4802 {
4803 	struct ctl_softc *softc;
4804 	struct ctl_lun *lun;
4805 	struct copan_aps_subpage *current_sp;
4806 	struct ctl_page_index *page_index;
4807 	int i;
4808 
4809 	softc = control_softc;
4810 
4811 	mtx_lock(&softc->ctl_lock);
4812 
4813 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4814 
4815 	page_index = NULL;
4816 	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
4817 		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
4818 		     APS_PAGE_CODE)
4819 			continue;
4820 
4821 		if (lun->mode_pages.index[i].subpage != APS_SUBPAGE_CODE)
4822 			continue;
4823 		page_index = &lun->mode_pages.index[i];
4824 	}
4825 
4826 	if (page_index == NULL) {
4827 		mtx_unlock(&softc->ctl_lock);
4828 		printf("%s: APS subpage not found for lun %ju!\n", __func__,
4829 		       (uintmax_t)lun->lun);
4830 		return (1);
4831 	}
4832 #if 0
4833 	if ((softc->aps_locked_lun != 0)
4834 	 && (softc->aps_locked_lun != lun->lun)) {
4835 		printf("%s: attempt to lock LUN %llu when %llu is already "
4836 		       "locked\n");
4837 		mtx_unlock(&softc->ctl_lock);
4838 		return (1);
4839 	}
4840 #endif
4841 
4842 	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
4843 		(page_index->page_len * CTL_PAGE_CURRENT));
4844 
4845 	if (lock != 0) {
4846 		current_sp->lock_active = APS_LOCK_ACTIVE;
4847 		softc->aps_locked_lun = lun->lun;
4848 	} else {
4849 		current_sp->lock_active = 0;
4850 		softc->aps_locked_lun = 0;
4851 	}
4852 
4853 
4854 	/*
4855 	 * If we're in HA mode, try to send the lock message to the other
4856 	 * side.
4857 	 */
4858 	if (ctl_is_single == 0) {
4859 		int isc_retval;
4860 		union ctl_ha_msg lock_msg;
4861 
4862 		lock_msg.hdr.nexus = *nexus;
4863 		lock_msg.hdr.msg_type = CTL_MSG_APS_LOCK;
4864 		if (lock != 0)
4865 			lock_msg.aps.lock_flag = 1;
4866 		else
4867 			lock_msg.aps.lock_flag = 0;
4868 		isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &lock_msg,
4869 					 sizeof(lock_msg), 0);
4870 		if (isc_retval > CTL_HA_STATUS_SUCCESS) {
4871 			printf("%s: APS (lock=%d) error returned from "
4872 			       "ctl_ha_msg_send: %d\n", __func__, lock, isc_retval);
4873 			mtx_unlock(&softc->ctl_lock);
4874 			return (1);
4875 		}
4876 	}
4877 
4878 	mtx_unlock(&softc->ctl_lock);
4879 
4880 	return (0);
4881 }
4882 
4883 void
4884 ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
4885 {
4886 	struct ctl_lun *lun;
4887 	struct ctl_softc *softc;
4888 	int i;
4889 
4890 	softc = control_softc;
4891 
4892 	mtx_lock(&softc->ctl_lock);
4893 
4894 	lun = (struct ctl_lun *)be_lun->ctl_lun;
4895 
4896 	for (i = 0; i < CTL_MAX_INITIATORS; i++)
4897 		lun->pending_sense[i].ua_pending |= CTL_UA_CAPACITY_CHANGED;
4898 
4899 	mtx_unlock(&softc->ctl_lock);
4900 }
4901 
4902 /*
4903  * Backend "memory move is complete" callback for requests that never
4904  * make it down to say RAIDCore's configuration code.
4905  */
4906 int
4907 ctl_config_move_done(union ctl_io *io)
4908 {
4909 	int retval;
4910 
4911 	retval = CTL_RETVAL_COMPLETE;
4912 
4913 
4914 	CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
4915 	/*
4916 	 * XXX KDM this shouldn't happen, but what if it does?
4917 	 */
4918 	if (io->io_hdr.io_type != CTL_IO_SCSI)
4919 		panic("I/O type isn't CTL_IO_SCSI!");
4920 
4921 	if ((io->io_hdr.port_status == 0)
4922 	 && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
4923 	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE))
4924 		io->io_hdr.status = CTL_SUCCESS;
4925 	else if ((io->io_hdr.port_status != 0)
4926 	      && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
4927 	      && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)){
4928 		/*
4929 		 * For hardware error sense keys, the sense key
4930 		 * specific value is defined to be a retry count,
4931 		 * but we use it to pass back an internal FETD
4932 		 * error code.  XXX KDM  Hopefully the FETD is only
4933 		 * using 16 bits for an error code, since that's
4934 		 * all the space we have in the sks field.
4935 		 */
4936 		ctl_set_internal_failure(&io->scsiio,
4937 					 /*sks_valid*/ 1,
4938 					 /*retry_count*/
4939 					 io->io_hdr.port_status);
4940 		free(io->scsiio.kern_data_ptr, M_CTL);
4941 		ctl_done(io);
4942 		goto bailout;
4943 	}
4944 
4945 	if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
4946 	 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
4947 	 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
4948 		/*
4949 		 * XXX KDM just assuming a single pointer here, and not a
4950 		 * S/G list.  If we start using S/G lists for config data,
4951 		 * we'll need to know how to clean them up here as well.
4952 		 */
4953 		free(io->scsiio.kern_data_ptr, M_CTL);
4954 		/* Hopefully the user has already set the status... */
4955 		ctl_done(io);
4956 	} else {
4957 		/*
4958 		 * XXX KDM now we need to continue data movement.  Some
4959 		 * options:
4960 		 * - call ctl_scsiio() again?  We don't do this for data
4961 		 *   writes, because for those at least we know ahead of
4962 		 *   time where the write will go and how long it is.  For
4963 		 *   config writes, though, that information is largely
4964 		 *   contained within the write itself, thus we need to
4965 		 *   parse out the data again.
4966 		 *
4967 		 * - Call some other function once the data is in?
4968 		 */
4969 
4970 		/*
4971 		 * XXX KDM call ctl_scsiio() again for now, and check flag
4972 		 * bits to see whether we're allocated or not.
4973 		 */
4974 		retval = ctl_scsiio(&io->scsiio);
4975 	}
4976 bailout:
4977 	return (retval);
4978 }
4979 
4980 /*
4981  * This gets called by a backend driver when it is done with a
4982  * configuration write.
4983  */
4984 void
4985 ctl_config_write_done(union ctl_io *io)
4986 {
4987 	/*
4988 	 * If the IO_CONT flag is set, we need to call the supplied
4989 	 * function to continue processing the I/O, instead of completing
4990 	 * the I/O just yet.
4991 	 *
4992 	 * If there is an error, though, we don't want to keep processing.
4993 	 * Instead, just send status back to the initiator.
4994 	 */
4995 	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT)
4996 	 && (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)
4997 	  || ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))) {
4998 		io->scsiio.io_cont(io);
4999 		return;
5000 	}
5001 	/*
5002 	 * Since a configuration write can be done for commands that actually
5003 	 * have data allocated, like write buffer, and commands that have
5004 	 * no data, like start/stop unit, we need to check here.
5005 	 */
5006 	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT)
5007 		free(io->scsiio.kern_data_ptr, M_CTL);
5008 	ctl_done(io);
5009 }
5010 
5011 /*
5012  * SCSI release command.
5013  */
5014 int
5015 ctl_scsi_release(struct ctl_scsiio *ctsio)
5016 {
5017 	int length, longid, thirdparty_id, resv_id;
5018 	struct ctl_softc *ctl_softc;
5019 	struct ctl_lun *lun;
5020 
5021 	length = 0;
5022 	resv_id = 0;
5023 
5024 	CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5025 
5026 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5027 	ctl_softc = control_softc;
5028 
5029 	switch (ctsio->cdb[0]) {
5030 	case RELEASE: {
5031 		struct scsi_release *cdb;
5032 
5033 		cdb = (struct scsi_release *)ctsio->cdb;
5034 		if ((cdb->byte2 & 0x1f) != 0) {
5035 			ctl_set_invalid_field(ctsio,
5036 					      /*sks_valid*/ 1,
5037 					      /*command*/ 1,
5038 					      /*field*/ 1,
5039 					      /*bit_valid*/ 0,
5040 					      /*bit*/ 0);
5041 			ctl_done((union ctl_io *)ctsio);
5042 			return (CTL_RETVAL_COMPLETE);
5043 		}
5044 		break;
5045 	}
5046 	case RELEASE_10: {
5047 		struct scsi_release_10 *cdb;
5048 
5049 		cdb = (struct scsi_release_10 *)ctsio->cdb;
5050 
5051 		if ((cdb->byte2 & SR10_EXTENT) != 0) {
5052 			ctl_set_invalid_field(ctsio,
5053 					      /*sks_valid*/ 1,
5054 					      /*command*/ 1,
5055 					      /*field*/ 1,
5056 					      /*bit_valid*/ 1,
5057 					      /*bit*/ 0);
5058 			ctl_done((union ctl_io *)ctsio);
5059 			return (CTL_RETVAL_COMPLETE);
5060 
5061 		}
5062 
5063 		if ((cdb->byte2 & SR10_3RDPTY) != 0) {
5064 			ctl_set_invalid_field(ctsio,
5065 					      /*sks_valid*/ 1,
5066 					      /*command*/ 1,
5067 					      /*field*/ 1,
5068 					      /*bit_valid*/ 1,
5069 					      /*bit*/ 4);
5070 			ctl_done((union ctl_io *)ctsio);
5071 			return (CTL_RETVAL_COMPLETE);
5072 		}
5073 
5074 		if (cdb->byte2 & SR10_LONGID)
5075 			longid = 1;
5076 		else
5077 			thirdparty_id = cdb->thirdparty_id;
5078 
5079 		resv_id = cdb->resv_id;
5080 		length = scsi_2btoul(cdb->length);
5081 		break;
5082 	}
5083 	}
5084 
5085 
5086 	/*
5087 	 * XXX KDM right now, we only support LUN reservation.  We don't
5088 	 * support 3rd party reservations, or extent reservations, which
5089 	 * might actually need the parameter list.  If we've gotten this
5090 	 * far, we've got a LUN reservation.  Anything else got kicked out
5091 	 * above.  So, according to SPC, ignore the length.
5092 	 */
5093 	length = 0;
5094 
5095 	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5096 	 && (length > 0)) {
5097 		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5098 		ctsio->kern_data_len = length;
5099 		ctsio->kern_total_len = length;
5100 		ctsio->kern_data_resid = 0;
5101 		ctsio->kern_rel_offset = 0;
5102 		ctsio->kern_sg_entries = 0;
5103 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5104 		ctsio->be_move_done = ctl_config_move_done;
5105 		ctl_datamove((union ctl_io *)ctsio);
5106 
5107 		return (CTL_RETVAL_COMPLETE);
5108 	}
5109 
5110 	if (length > 0)
5111 		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5112 
5113 	mtx_lock(&ctl_softc->ctl_lock);
5114 
5115 	/*
5116 	 * According to SPC, it is not an error for an intiator to attempt
5117 	 * to release a reservation on a LUN that isn't reserved, or that
5118 	 * is reserved by another initiator.  The reservation can only be
5119 	 * released, though, by the initiator who made it or by one of
5120 	 * several reset type events.
5121 	 */
5122 	if (lun->flags & CTL_LUN_RESERVED) {
5123 		if ((ctsio->io_hdr.nexus.initid.id == lun->rsv_nexus.initid.id)
5124 		 && (ctsio->io_hdr.nexus.targ_port == lun->rsv_nexus.targ_port)
5125 		 && (ctsio->io_hdr.nexus.targ_target.id ==
5126 		     lun->rsv_nexus.targ_target.id)) {
5127 			lun->flags &= ~CTL_LUN_RESERVED;
5128 		}
5129 	}
5130 
5131 	ctsio->scsi_status = SCSI_STATUS_OK;
5132 	ctsio->io_hdr.status = CTL_SUCCESS;
5133 
5134 	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5135 		free(ctsio->kern_data_ptr, M_CTL);
5136 		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5137 	}
5138 
5139 	mtx_unlock(&ctl_softc->ctl_lock);
5140 
5141 	ctl_done((union ctl_io *)ctsio);
5142 	return (CTL_RETVAL_COMPLETE);
5143 }
5144 
5145 int
5146 ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5147 {
5148 	int extent, thirdparty, longid;
5149 	int resv_id, length;
5150 	uint64_t thirdparty_id;
5151 	struct ctl_softc *ctl_softc;
5152 	struct ctl_lun *lun;
5153 
5154 	extent = 0;
5155 	thirdparty = 0;
5156 	longid = 0;
5157 	resv_id = 0;
5158 	length = 0;
5159 	thirdparty_id = 0;
5160 
5161 	CTL_DEBUG_PRINT(("ctl_reserve\n"));
5162 
5163 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5164 	ctl_softc = control_softc;
5165 
5166 	switch (ctsio->cdb[0]) {
5167 	case RESERVE: {
5168 		struct scsi_reserve *cdb;
5169 
5170 		cdb = (struct scsi_reserve *)ctsio->cdb;
5171 		if ((cdb->byte2 & 0x1f) != 0) {
5172 			ctl_set_invalid_field(ctsio,
5173 					      /*sks_valid*/ 1,
5174 					      /*command*/ 1,
5175 					      /*field*/ 1,
5176 					      /*bit_valid*/ 0,
5177 					      /*bit*/ 0);
5178 			ctl_done((union ctl_io *)ctsio);
5179 			return (CTL_RETVAL_COMPLETE);
5180 		}
5181 		resv_id = cdb->resv_id;
5182 		length = scsi_2btoul(cdb->length);
5183 		break;
5184 	}
5185 	case RESERVE_10: {
5186 		struct scsi_reserve_10 *cdb;
5187 
5188 		cdb = (struct scsi_reserve_10 *)ctsio->cdb;
5189 
5190 		if ((cdb->byte2 & SR10_EXTENT) != 0) {
5191 			ctl_set_invalid_field(ctsio,
5192 					      /*sks_valid*/ 1,
5193 					      /*command*/ 1,
5194 					      /*field*/ 1,
5195 					      /*bit_valid*/ 1,
5196 					      /*bit*/ 0);
5197 			ctl_done((union ctl_io *)ctsio);
5198 			return (CTL_RETVAL_COMPLETE);
5199 		}
5200 		if ((cdb->byte2 & SR10_3RDPTY) != 0) {
5201 			ctl_set_invalid_field(ctsio,
5202 					      /*sks_valid*/ 1,
5203 					      /*command*/ 1,
5204 					      /*field*/ 1,
5205 					      /*bit_valid*/ 1,
5206 					      /*bit*/ 4);
5207 			ctl_done((union ctl_io *)ctsio);
5208 			return (CTL_RETVAL_COMPLETE);
5209 		}
5210 		if (cdb->byte2 & SR10_LONGID)
5211 			longid = 1;
5212 		else
5213 			thirdparty_id = cdb->thirdparty_id;
5214 
5215 		resv_id = cdb->resv_id;
5216 		length = scsi_2btoul(cdb->length);
5217 		break;
5218 	}
5219 	}
5220 
5221 	/*
5222 	 * XXX KDM right now, we only support LUN reservation.  We don't
5223 	 * support 3rd party reservations, or extent reservations, which
5224 	 * might actually need the parameter list.  If we've gotten this
5225 	 * far, we've got a LUN reservation.  Anything else got kicked out
5226 	 * above.  So, according to SPC, ignore the length.
5227 	 */
5228 	length = 0;
5229 
5230 	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5231 	 && (length > 0)) {
5232 		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5233 		ctsio->kern_data_len = length;
5234 		ctsio->kern_total_len = length;
5235 		ctsio->kern_data_resid = 0;
5236 		ctsio->kern_rel_offset = 0;
5237 		ctsio->kern_sg_entries = 0;
5238 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5239 		ctsio->be_move_done = ctl_config_move_done;
5240 		ctl_datamove((union ctl_io *)ctsio);
5241 
5242 		return (CTL_RETVAL_COMPLETE);
5243 	}
5244 
5245 	if (length > 0)
5246 		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5247 
5248 	mtx_lock(&ctl_softc->ctl_lock);
5249 	if (lun->flags & CTL_LUN_RESERVED) {
5250 		if ((ctsio->io_hdr.nexus.initid.id != lun->rsv_nexus.initid.id)
5251 		 || (ctsio->io_hdr.nexus.targ_port != lun->rsv_nexus.targ_port)
5252 		 || (ctsio->io_hdr.nexus.targ_target.id !=
5253 		     lun->rsv_nexus.targ_target.id)) {
5254 			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
5255 			ctsio->io_hdr.status = CTL_SCSI_ERROR;
5256 			goto bailout;
5257 		}
5258 	}
5259 
5260 	lun->flags |= CTL_LUN_RESERVED;
5261 	lun->rsv_nexus = ctsio->io_hdr.nexus;
5262 
5263 	ctsio->scsi_status = SCSI_STATUS_OK;
5264 	ctsio->io_hdr.status = CTL_SUCCESS;
5265 
5266 bailout:
5267 	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5268 		free(ctsio->kern_data_ptr, M_CTL);
5269 		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5270 	}
5271 
5272 	mtx_unlock(&ctl_softc->ctl_lock);
5273 
5274 	ctl_done((union ctl_io *)ctsio);
5275 	return (CTL_RETVAL_COMPLETE);
5276 }
5277 
5278 int
5279 ctl_start_stop(struct ctl_scsiio *ctsio)
5280 {
5281 	struct scsi_start_stop_unit *cdb;
5282 	struct ctl_lun *lun;
5283 	struct ctl_softc *ctl_softc;
5284 	int retval;
5285 
5286 	CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5287 
5288 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5289 	ctl_softc = control_softc;
5290 	retval = 0;
5291 
5292 	cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5293 
5294 	/*
5295 	 * XXX KDM
5296 	 * We don't support the immediate bit on a stop unit.  In order to
5297 	 * do that, we would need to code up a way to know that a stop is
5298 	 * pending, and hold off any new commands until it completes, one
5299 	 * way or another.  Then we could accept or reject those commands
5300 	 * depending on its status.  We would almost need to do the reverse
5301 	 * of what we do below for an immediate start -- return the copy of
5302 	 * the ctl_io to the FETD with status to send to the host (and to
5303 	 * free the copy!) and then free the original I/O once the stop
5304 	 * actually completes.  That way, the OOA queue mechanism can work
5305 	 * to block commands that shouldn't proceed.  Another alternative
5306 	 * would be to put the copy in the queue in place of the original,
5307 	 * and return the original back to the caller.  That could be
5308 	 * slightly safer..
5309 	 */
5310 	if ((cdb->byte2 & SSS_IMMED)
5311 	 && ((cdb->how & SSS_START) == 0)) {
5312 		ctl_set_invalid_field(ctsio,
5313 				      /*sks_valid*/ 1,
5314 				      /*command*/ 1,
5315 				      /*field*/ 1,
5316 				      /*bit_valid*/ 1,
5317 				      /*bit*/ 0);
5318 		ctl_done((union ctl_io *)ctsio);
5319 		return (CTL_RETVAL_COMPLETE);
5320 	}
5321 
5322 	/*
5323 	 * We don't support the power conditions field.  We need to check
5324 	 * this prior to checking the load/eject and start/stop bits.
5325 	 */
5326 	if ((cdb->how & SSS_PC_MASK) != SSS_PC_START_VALID) {
5327 		ctl_set_invalid_field(ctsio,
5328 				      /*sks_valid*/ 1,
5329 				      /*command*/ 1,
5330 				      /*field*/ 4,
5331 				      /*bit_valid*/ 1,
5332 				      /*bit*/ 4);
5333 		ctl_done((union ctl_io *)ctsio);
5334 		return (CTL_RETVAL_COMPLETE);
5335 	}
5336 
5337 	/*
5338 	 * Media isn't removable, so we can't load or eject it.
5339 	 */
5340 	if ((cdb->how & SSS_LOEJ) != 0) {
5341 		ctl_set_invalid_field(ctsio,
5342 				      /*sks_valid*/ 1,
5343 				      /*command*/ 1,
5344 				      /*field*/ 4,
5345 				      /*bit_valid*/ 1,
5346 				      /*bit*/ 1);
5347 		ctl_done((union ctl_io *)ctsio);
5348 		return (CTL_RETVAL_COMPLETE);
5349 	}
5350 
5351 	if ((lun->flags & CTL_LUN_PR_RESERVED)
5352 	 && ((cdb->how & SSS_START)==0)) {
5353 		uint32_t residx;
5354 
5355 		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5356 		if (!lun->per_res[residx].registered
5357 		 || (lun->pr_res_idx!=residx && lun->res_type < 4)) {
5358 
5359 			ctl_set_reservation_conflict(ctsio);
5360 			ctl_done((union ctl_io *)ctsio);
5361 			return (CTL_RETVAL_COMPLETE);
5362 		}
5363 	}
5364 
5365 	/*
5366 	 * If there is no backend on this device, we can't start or stop
5367 	 * it.  In theory we shouldn't get any start/stop commands in the
5368 	 * first place at this level if the LUN doesn't have a backend.
5369 	 * That should get stopped by the command decode code.
5370 	 */
5371 	if (lun->backend == NULL) {
5372 		ctl_set_invalid_opcode(ctsio);
5373 		ctl_done((union ctl_io *)ctsio);
5374 		return (CTL_RETVAL_COMPLETE);
5375 	}
5376 
5377 	/*
5378 	 * XXX KDM Copan-specific offline behavior.
5379 	 * Figure out a reasonable way to port this?
5380 	 */
5381 #ifdef NEEDTOPORT
5382 	mtx_lock(&ctl_softc->ctl_lock);
5383 
5384 	if (((cdb->byte2 & SSS_ONOFFLINE) == 0)
5385 	 && (lun->flags & CTL_LUN_OFFLINE)) {
5386 		/*
5387 		 * If the LUN is offline, and the on/offline bit isn't set,
5388 		 * reject the start or stop.  Otherwise, let it through.
5389 		 */
5390 		mtx_unlock(&ctl_softc->ctl_lock);
5391 		ctl_set_lun_not_ready(ctsio);
5392 		ctl_done((union ctl_io *)ctsio);
5393 	} else {
5394 		mtx_unlock(&ctl_softc->ctl_lock);
5395 #endif /* NEEDTOPORT */
5396 		/*
5397 		 * This could be a start or a stop when we're online,
5398 		 * or a stop/offline or start/online.  A start or stop when
5399 		 * we're offline is covered in the case above.
5400 		 */
5401 		/*
5402 		 * In the non-immediate case, we send the request to
5403 		 * the backend and return status to the user when
5404 		 * it is done.
5405 		 *
5406 		 * In the immediate case, we allocate a new ctl_io
5407 		 * to hold a copy of the request, and send that to
5408 		 * the backend.  We then set good status on the
5409 		 * user's request and return it immediately.
5410 		 */
5411 		if (cdb->byte2 & SSS_IMMED) {
5412 			union ctl_io *new_io;
5413 
5414 			new_io = ctl_alloc_io(ctsio->io_hdr.pool);
5415 			if (new_io == NULL) {
5416 				ctl_set_busy(ctsio);
5417 				ctl_done((union ctl_io *)ctsio);
5418 			} else {
5419 				ctl_copy_io((union ctl_io *)ctsio,
5420 					    new_io);
5421 				retval = lun->backend->config_write(new_io);
5422 				ctl_set_success(ctsio);
5423 				ctl_done((union ctl_io *)ctsio);
5424 			}
5425 		} else {
5426 			retval = lun->backend->config_write(
5427 				(union ctl_io *)ctsio);
5428 		}
5429 #ifdef NEEDTOPORT
5430 	}
5431 #endif
5432 	return (retval);
5433 }
5434 
5435 /*
5436  * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5437  * we don't really do anything with the LBA and length fields if the user
5438  * passes them in.  Instead we'll just flush out the cache for the entire
5439  * LUN.
5440  */
5441 int
5442 ctl_sync_cache(struct ctl_scsiio *ctsio)
5443 {
5444 	struct ctl_lun *lun;
5445 	struct ctl_softc *ctl_softc;
5446 	uint64_t starting_lba;
5447 	uint32_t block_count;
5448 	int reladr, immed;
5449 	int retval;
5450 
5451 	CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5452 
5453 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5454 	ctl_softc = control_softc;
5455 	retval = 0;
5456 	reladr = 0;
5457 	immed = 0;
5458 
5459 	switch (ctsio->cdb[0]) {
5460 	case SYNCHRONIZE_CACHE: {
5461 		struct scsi_sync_cache *cdb;
5462 		cdb = (struct scsi_sync_cache *)ctsio->cdb;
5463 
5464 		if (cdb->byte2 & SSC_RELADR)
5465 			reladr = 1;
5466 
5467 		if (cdb->byte2 & SSC_IMMED)
5468 			immed = 1;
5469 
5470 		starting_lba = scsi_4btoul(cdb->begin_lba);
5471 		block_count = scsi_2btoul(cdb->lb_count);
5472 		break;
5473 	}
5474 	case SYNCHRONIZE_CACHE_16: {
5475 		struct scsi_sync_cache_16 *cdb;
5476 		cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5477 
5478 		if (cdb->byte2 & SSC_RELADR)
5479 			reladr = 1;
5480 
5481 		if (cdb->byte2 & SSC_IMMED)
5482 			immed = 1;
5483 
5484 		starting_lba = scsi_8btou64(cdb->begin_lba);
5485 		block_count = scsi_4btoul(cdb->lb_count);
5486 		break;
5487 	}
5488 	default:
5489 		ctl_set_invalid_opcode(ctsio);
5490 		ctl_done((union ctl_io *)ctsio);
5491 		goto bailout;
5492 		break; /* NOTREACHED */
5493 	}
5494 
5495 	if (immed) {
5496 		/*
5497 		 * We don't support the immediate bit.  Since it's in the
5498 		 * same place for the 10 and 16 byte SYNCHRONIZE CACHE
5499 		 * commands, we can just return the same error in either
5500 		 * case.
5501 		 */
5502 		ctl_set_invalid_field(ctsio,
5503 				      /*sks_valid*/ 1,
5504 				      /*command*/ 1,
5505 				      /*field*/ 1,
5506 				      /*bit_valid*/ 1,
5507 				      /*bit*/ 1);
5508 		ctl_done((union ctl_io *)ctsio);
5509 		goto bailout;
5510 	}
5511 
5512 	if (reladr) {
5513 		/*
5514 		 * We don't support the reladr bit either.  It can only be
5515 		 * used with linked commands, and we don't support linked
5516 		 * commands.  Since the bit is in the same place for the
5517 		 * 10 and 16 byte SYNCHRONIZE CACHE * commands, we can
5518 		 * just return the same error in either case.
5519 		 */
5520 		ctl_set_invalid_field(ctsio,
5521 				      /*sks_valid*/ 1,
5522 				      /*command*/ 1,
5523 				      /*field*/ 1,
5524 				      /*bit_valid*/ 1,
5525 				      /*bit*/ 0);
5526 		ctl_done((union ctl_io *)ctsio);
5527 		goto bailout;
5528 	}
5529 
5530 	/*
5531 	 * We check the LBA and length, but don't do anything with them.
5532 	 * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5533 	 * get flushed.  This check will just help satisfy anyone who wants
5534 	 * to see an error for an out of range LBA.
5535 	 */
5536 	if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5537 		ctl_set_lba_out_of_range(ctsio);
5538 		ctl_done((union ctl_io *)ctsio);
5539 		goto bailout;
5540 	}
5541 
5542 	/*
5543 	 * If this LUN has no backend, we can't flush the cache anyway.
5544 	 */
5545 	if (lun->backend == NULL) {
5546 		ctl_set_invalid_opcode(ctsio);
5547 		ctl_done((union ctl_io *)ctsio);
5548 		goto bailout;
5549 	}
5550 
5551 	/*
5552 	 * Check to see whether we're configured to send the SYNCHRONIZE
5553 	 * CACHE command directly to the back end.
5554 	 */
5555 	mtx_lock(&ctl_softc->ctl_lock);
5556 	if ((ctl_softc->flags & CTL_FLAG_REAL_SYNC)
5557 	 && (++(lun->sync_count) >= lun->sync_interval)) {
5558 		lun->sync_count = 0;
5559 		mtx_unlock(&ctl_softc->ctl_lock);
5560 		retval = lun->backend->config_write((union ctl_io *)ctsio);
5561 	} else {
5562 		mtx_unlock(&ctl_softc->ctl_lock);
5563 		ctl_set_success(ctsio);
5564 		ctl_done((union ctl_io *)ctsio);
5565 	}
5566 
5567 bailout:
5568 
5569 	return (retval);
5570 }
5571 
5572 int
5573 ctl_format(struct ctl_scsiio *ctsio)
5574 {
5575 	struct scsi_format *cdb;
5576 	struct ctl_lun *lun;
5577 	struct ctl_softc *ctl_softc;
5578 	int length, defect_list_len;
5579 
5580 	CTL_DEBUG_PRINT(("ctl_format\n"));
5581 
5582 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5583 	ctl_softc = control_softc;
5584 
5585 	cdb = (struct scsi_format *)ctsio->cdb;
5586 
5587 	length = 0;
5588 	if (cdb->byte2 & SF_FMTDATA) {
5589 		if (cdb->byte2 & SF_LONGLIST)
5590 			length = sizeof(struct scsi_format_header_long);
5591 		else
5592 			length = sizeof(struct scsi_format_header_short);
5593 	}
5594 
5595 	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5596 	 && (length > 0)) {
5597 		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5598 		ctsio->kern_data_len = length;
5599 		ctsio->kern_total_len = length;
5600 		ctsio->kern_data_resid = 0;
5601 		ctsio->kern_rel_offset = 0;
5602 		ctsio->kern_sg_entries = 0;
5603 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5604 		ctsio->be_move_done = ctl_config_move_done;
5605 		ctl_datamove((union ctl_io *)ctsio);
5606 
5607 		return (CTL_RETVAL_COMPLETE);
5608 	}
5609 
5610 	defect_list_len = 0;
5611 
5612 	if (cdb->byte2 & SF_FMTDATA) {
5613 		if (cdb->byte2 & SF_LONGLIST) {
5614 			struct scsi_format_header_long *header;
5615 
5616 			header = (struct scsi_format_header_long *)
5617 				ctsio->kern_data_ptr;
5618 
5619 			defect_list_len = scsi_4btoul(header->defect_list_len);
5620 			if (defect_list_len != 0) {
5621 				ctl_set_invalid_field(ctsio,
5622 						      /*sks_valid*/ 1,
5623 						      /*command*/ 0,
5624 						      /*field*/ 2,
5625 						      /*bit_valid*/ 0,
5626 						      /*bit*/ 0);
5627 				goto bailout;
5628 			}
5629 		} else {
5630 			struct scsi_format_header_short *header;
5631 
5632 			header = (struct scsi_format_header_short *)
5633 				ctsio->kern_data_ptr;
5634 
5635 			defect_list_len = scsi_2btoul(header->defect_list_len);
5636 			if (defect_list_len != 0) {
5637 				ctl_set_invalid_field(ctsio,
5638 						      /*sks_valid*/ 1,
5639 						      /*command*/ 0,
5640 						      /*field*/ 2,
5641 						      /*bit_valid*/ 0,
5642 						      /*bit*/ 0);
5643 				goto bailout;
5644 			}
5645 		}
5646 	}
5647 
5648 	/*
5649 	 * The format command will clear out the "Medium format corrupted"
5650 	 * status if set by the configuration code.  That status is really
5651 	 * just a way to notify the host that we have lost the media, and
5652 	 * get them to issue a command that will basically make them think
5653 	 * they're blowing away the media.
5654 	 */
5655 	mtx_lock(&ctl_softc->ctl_lock);
5656 	lun->flags &= ~CTL_LUN_INOPERABLE;
5657 	mtx_unlock(&ctl_softc->ctl_lock);
5658 
5659 	ctsio->scsi_status = SCSI_STATUS_OK;
5660 	ctsio->io_hdr.status = CTL_SUCCESS;
5661 bailout:
5662 
5663 	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5664 		free(ctsio->kern_data_ptr, M_CTL);
5665 		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5666 	}
5667 
5668 	ctl_done((union ctl_io *)ctsio);
5669 	return (CTL_RETVAL_COMPLETE);
5670 }
5671 
5672 int
5673 ctl_write_buffer(struct ctl_scsiio *ctsio)
5674 {
5675 	struct scsi_write_buffer *cdb;
5676 	struct copan_page_header *header;
5677 	struct ctl_lun *lun;
5678 	struct ctl_softc *ctl_softc;
5679 	int buffer_offset, len;
5680 	int retval;
5681 
5682 	header = NULL;
5683 
5684 	retval = CTL_RETVAL_COMPLETE;
5685 
5686 	CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5687 
5688 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5689 	ctl_softc = control_softc;
5690 	cdb = (struct scsi_write_buffer *)ctsio->cdb;
5691 
5692 	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5693 		ctl_set_invalid_field(ctsio,
5694 				      /*sks_valid*/ 1,
5695 				      /*command*/ 1,
5696 				      /*field*/ 1,
5697 				      /*bit_valid*/ 1,
5698 				      /*bit*/ 4);
5699 		ctl_done((union ctl_io *)ctsio);
5700 		return (CTL_RETVAL_COMPLETE);
5701 	}
5702 	if (cdb->buffer_id != 0) {
5703 		ctl_set_invalid_field(ctsio,
5704 				      /*sks_valid*/ 1,
5705 				      /*command*/ 1,
5706 				      /*field*/ 2,
5707 				      /*bit_valid*/ 0,
5708 				      /*bit*/ 0);
5709 		ctl_done((union ctl_io *)ctsio);
5710 		return (CTL_RETVAL_COMPLETE);
5711 	}
5712 
5713 	len = scsi_3btoul(cdb->length);
5714 	buffer_offset = scsi_3btoul(cdb->offset);
5715 
5716 	if (len > sizeof(lun->write_buffer)) {
5717 		ctl_set_invalid_field(ctsio,
5718 				      /*sks_valid*/ 1,
5719 				      /*command*/ 1,
5720 				      /*field*/ 6,
5721 				      /*bit_valid*/ 0,
5722 				      /*bit*/ 0);
5723 		ctl_done((union ctl_io *)ctsio);
5724 		return (CTL_RETVAL_COMPLETE);
5725 	}
5726 
5727 	if (buffer_offset != 0) {
5728 		ctl_set_invalid_field(ctsio,
5729 				      /*sks_valid*/ 1,
5730 				      /*command*/ 1,
5731 				      /*field*/ 3,
5732 				      /*bit_valid*/ 0,
5733 				      /*bit*/ 0);
5734 		ctl_done((union ctl_io *)ctsio);
5735 		return (CTL_RETVAL_COMPLETE);
5736 	}
5737 
5738 	/*
5739 	 * If we've got a kernel request that hasn't been malloced yet,
5740 	 * malloc it and tell the caller the data buffer is here.
5741 	 */
5742 	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5743 		ctsio->kern_data_ptr = lun->write_buffer;
5744 		ctsio->kern_data_len = len;
5745 		ctsio->kern_total_len = len;
5746 		ctsio->kern_data_resid = 0;
5747 		ctsio->kern_rel_offset = 0;
5748 		ctsio->kern_sg_entries = 0;
5749 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5750 		ctsio->be_move_done = ctl_config_move_done;
5751 		ctl_datamove((union ctl_io *)ctsio);
5752 
5753 		return (CTL_RETVAL_COMPLETE);
5754 	}
5755 
5756 	ctl_done((union ctl_io *)ctsio);
5757 
5758 	return (CTL_RETVAL_COMPLETE);
5759 }
5760 
5761 /*
5762  * Note that this function currently doesn't actually do anything inside
5763  * CTL to enforce things if the DQue bit is turned on.
5764  *
5765  * Also note that this function can't be used in the default case, because
5766  * the DQue bit isn't set in the changeable mask for the control mode page
5767  * anyway.  This is just here as an example for how to implement a page
5768  * handler, and a placeholder in case we want to allow the user to turn
5769  * tagged queueing on and off.
5770  *
5771  * The D_SENSE bit handling is functional, however, and will turn
5772  * descriptor sense on and off for a given LUN.
5773  */
5774 int
5775 ctl_control_page_handler(struct ctl_scsiio *ctsio,
5776 			 struct ctl_page_index *page_index, uint8_t *page_ptr)
5777 {
5778 	struct scsi_control_page *current_cp, *saved_cp, *user_cp;
5779 	struct ctl_lun *lun;
5780 	struct ctl_softc *softc;
5781 	int set_ua;
5782 	uint32_t initidx;
5783 
5784 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5785 	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5786 	set_ua = 0;
5787 
5788 	user_cp = (struct scsi_control_page *)page_ptr;
5789 	current_cp = (struct scsi_control_page *)
5790 		(page_index->page_data + (page_index->page_len *
5791 		CTL_PAGE_CURRENT));
5792 	saved_cp = (struct scsi_control_page *)
5793 		(page_index->page_data + (page_index->page_len *
5794 		CTL_PAGE_SAVED));
5795 
5796 	softc = control_softc;
5797 
5798 	mtx_lock(&softc->ctl_lock);
5799 	if (((current_cp->rlec & SCP_DSENSE) == 0)
5800 	 && ((user_cp->rlec & SCP_DSENSE) != 0)) {
5801 		/*
5802 		 * Descriptor sense is currently turned off and the user
5803 		 * wants to turn it on.
5804 		 */
5805 		current_cp->rlec |= SCP_DSENSE;
5806 		saved_cp->rlec |= SCP_DSENSE;
5807 		lun->flags |= CTL_LUN_SENSE_DESC;
5808 		set_ua = 1;
5809 	} else if (((current_cp->rlec & SCP_DSENSE) != 0)
5810 		&& ((user_cp->rlec & SCP_DSENSE) == 0)) {
5811 		/*
5812 		 * Descriptor sense is currently turned on, and the user
5813 		 * wants to turn it off.
5814 		 */
5815 		current_cp->rlec &= ~SCP_DSENSE;
5816 		saved_cp->rlec &= ~SCP_DSENSE;
5817 		lun->flags &= ~CTL_LUN_SENSE_DESC;
5818 		set_ua = 1;
5819 	}
5820 	if (current_cp->queue_flags & SCP_QUEUE_DQUE) {
5821 		if (user_cp->queue_flags & SCP_QUEUE_DQUE) {
5822 #ifdef NEEDTOPORT
5823 			csevent_log(CSC_CTL | CSC_SHELF_SW |
5824 				    CTL_UNTAG_TO_UNTAG,
5825 				    csevent_LogType_Trace,
5826 				    csevent_Severity_Information,
5827 				    csevent_AlertLevel_Green,
5828 				    csevent_FRU_Firmware,
5829 				    csevent_FRU_Unknown,
5830 				    "Received untagged to untagged transition");
5831 #endif /* NEEDTOPORT */
5832 		} else {
5833 #ifdef NEEDTOPORT
5834 			csevent_log(CSC_CTL | CSC_SHELF_SW |
5835 				    CTL_UNTAG_TO_TAG,
5836 				    csevent_LogType_ConfigChange,
5837 				    csevent_Severity_Information,
5838 				    csevent_AlertLevel_Green,
5839 				    csevent_FRU_Firmware,
5840 				    csevent_FRU_Unknown,
5841 				    "Received untagged to tagged "
5842 				    "queueing transition");
5843 #endif /* NEEDTOPORT */
5844 
5845 			current_cp->queue_flags &= ~SCP_QUEUE_DQUE;
5846 			saved_cp->queue_flags &= ~SCP_QUEUE_DQUE;
5847 			set_ua = 1;
5848 		}
5849 	} else {
5850 		if (user_cp->queue_flags & SCP_QUEUE_DQUE) {
5851 #ifdef NEEDTOPORT
5852 			csevent_log(CSC_CTL | CSC_SHELF_SW |
5853 				    CTL_TAG_TO_UNTAG,
5854 				    csevent_LogType_ConfigChange,
5855 				    csevent_Severity_Warning,
5856 				    csevent_AlertLevel_Yellow,
5857 				    csevent_FRU_Firmware,
5858 				    csevent_FRU_Unknown,
5859 				    "Received tagged queueing to untagged "
5860 				    "transition");
5861 #endif /* NEEDTOPORT */
5862 
5863 			current_cp->queue_flags |= SCP_QUEUE_DQUE;
5864 			saved_cp->queue_flags |= SCP_QUEUE_DQUE;
5865 			set_ua = 1;
5866 		} else {
5867 #ifdef NEEDTOPORT
5868 			csevent_log(CSC_CTL | CSC_SHELF_SW |
5869 				    CTL_TAG_TO_TAG,
5870 				    csevent_LogType_Trace,
5871 				    csevent_Severity_Information,
5872 				    csevent_AlertLevel_Green,
5873 				    csevent_FRU_Firmware,
5874 				    csevent_FRU_Unknown,
5875 				    "Received tagged queueing to tagged "
5876 				    "queueing transition");
5877 #endif /* NEEDTOPORT */
5878 		}
5879 	}
5880 	if (set_ua != 0) {
5881 		int i;
5882 		/*
5883 		 * Let other initiators know that the mode
5884 		 * parameters for this LUN have changed.
5885 		 */
5886 		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
5887 			if (i == initidx)
5888 				continue;
5889 
5890 			lun->pending_sense[i].ua_pending |=
5891 				CTL_UA_MODE_CHANGE;
5892 		}
5893 	}
5894 	mtx_unlock(&softc->ctl_lock);
5895 
5896 	return (0);
5897 }
5898 
5899 int
5900 ctl_power_sp_handler(struct ctl_scsiio *ctsio,
5901 		     struct ctl_page_index *page_index, uint8_t *page_ptr)
5902 {
5903 	return (0);
5904 }
5905 
5906 int
5907 ctl_power_sp_sense_handler(struct ctl_scsiio *ctsio,
5908 			   struct ctl_page_index *page_index, int pc)
5909 {
5910 	struct copan_power_subpage *page;
5911 
5912 	page = (struct copan_power_subpage *)page_index->page_data +
5913 		(page_index->page_len * pc);
5914 
5915 	switch (pc) {
5916 	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
5917 		/*
5918 		 * We don't update the changable bits for this page.
5919 		 */
5920 		break;
5921 	case SMS_PAGE_CTRL_CURRENT >> 6:
5922 	case SMS_PAGE_CTRL_DEFAULT >> 6:
5923 	case SMS_PAGE_CTRL_SAVED >> 6:
5924 #ifdef NEEDTOPORT
5925 		ctl_update_power_subpage(page);
5926 #endif
5927 		break;
5928 	default:
5929 #ifdef NEEDTOPORT
5930 		EPRINT(0, "Invalid PC %d!!", pc);
5931 #endif
5932 		break;
5933 	}
5934 	return (0);
5935 }
5936 
5937 
5938 int
5939 ctl_aps_sp_handler(struct ctl_scsiio *ctsio,
5940 		   struct ctl_page_index *page_index, uint8_t *page_ptr)
5941 {
5942 	struct copan_aps_subpage *user_sp;
5943 	struct copan_aps_subpage *current_sp;
5944 	union ctl_modepage_info *modepage_info;
5945 	struct ctl_softc *softc;
5946 	struct ctl_lun *lun;
5947 	int retval;
5948 
5949 	retval = CTL_RETVAL_COMPLETE;
5950 	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
5951 		     (page_index->page_len * CTL_PAGE_CURRENT));
5952 	softc = control_softc;
5953 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5954 
5955 	user_sp = (struct copan_aps_subpage *)page_ptr;
5956 
5957 	modepage_info = (union ctl_modepage_info *)
5958 		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
5959 
5960 	modepage_info->header.page_code = page_index->page_code & SMPH_PC_MASK;
5961 	modepage_info->header.subpage = page_index->subpage;
5962 	modepage_info->aps.lock_active = user_sp->lock_active;
5963 
5964 	mtx_lock(&softc->ctl_lock);
5965 
5966 	/*
5967 	 * If there is a request to lock the LUN and another LUN is locked
5968 	 * this is an error. If the requested LUN is already locked ignore
5969 	 * the request. If no LUN is locked attempt to lock it.
5970 	 * if there is a request to unlock the LUN and the LUN is currently
5971 	 * locked attempt to unlock it. Otherwise ignore the request. i.e.
5972 	 * if another LUN is locked or no LUN is locked.
5973 	 */
5974 	if (user_sp->lock_active & APS_LOCK_ACTIVE) {
5975 		if (softc->aps_locked_lun == lun->lun) {
5976 			/*
5977 			 * This LUN is already locked, so we're done.
5978 			 */
5979 			retval = CTL_RETVAL_COMPLETE;
5980 		} else if (softc->aps_locked_lun == 0) {
5981 			/*
5982 			 * No one has the lock, pass the request to the
5983 			 * backend.
5984 			 */
5985 			retval = lun->backend->config_write(
5986 				(union ctl_io *)ctsio);
5987 		} else {
5988 			/*
5989 			 * Someone else has the lock, throw out the request.
5990 			 */
5991 			ctl_set_already_locked(ctsio);
5992 			free(ctsio->kern_data_ptr, M_CTL);
5993 			ctl_done((union ctl_io *)ctsio);
5994 
5995 			/*
5996 			 * Set the return value so that ctl_do_mode_select()
5997 			 * won't try to complete the command.  We already
5998 			 * completed it here.
5999 			 */
6000 			retval = CTL_RETVAL_ERROR;
6001 		}
6002 	} else if (softc->aps_locked_lun == lun->lun) {
6003 		/*
6004 		 * This LUN is locked, so pass the unlock request to the
6005 		 * backend.
6006 		 */
6007 		retval = lun->backend->config_write((union ctl_io *)ctsio);
6008 	}
6009 	mtx_unlock(&softc->ctl_lock);
6010 
6011 	return (retval);
6012 }
6013 
6014 int
6015 ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio,
6016 				struct ctl_page_index *page_index,
6017 				uint8_t *page_ptr)
6018 {
6019 	uint8_t *c;
6020 	int i;
6021 
6022 	c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs;
6023 	ctl_time_io_secs =
6024 		(c[0] << 8) |
6025 		(c[1] << 0) |
6026 		0;
6027 	CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs));
6028 	printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs);
6029 	printf("page data:");
6030 	for (i=0; i<8; i++)
6031 		printf(" %.2x",page_ptr[i]);
6032 	printf("\n");
6033 	return (0);
6034 }
6035 
6036 int
6037 ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio,
6038 			       struct ctl_page_index *page_index,
6039 			       int pc)
6040 {
6041 	struct copan_debugconf_subpage *page;
6042 
6043 	page = (struct copan_debugconf_subpage *)page_index->page_data +
6044 		(page_index->page_len * pc);
6045 
6046 	switch (pc) {
6047 	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6048 	case SMS_PAGE_CTRL_DEFAULT >> 6:
6049 	case SMS_PAGE_CTRL_SAVED >> 6:
6050 		/*
6051 		 * We don't update the changable or default bits for this page.
6052 		 */
6053 		break;
6054 	case SMS_PAGE_CTRL_CURRENT >> 6:
6055 		page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8;
6056 		page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0;
6057 		break;
6058 	default:
6059 #ifdef NEEDTOPORT
6060 		EPRINT(0, "Invalid PC %d!!", pc);
6061 #endif /* NEEDTOPORT */
6062 		break;
6063 	}
6064 	return (0);
6065 }
6066 
6067 
6068 static int
6069 ctl_do_mode_select(union ctl_io *io)
6070 {
6071 	struct scsi_mode_page_header *page_header;
6072 	struct ctl_page_index *page_index;
6073 	struct ctl_scsiio *ctsio;
6074 	int control_dev, page_len;
6075 	int page_len_offset, page_len_size;
6076 	union ctl_modepage_info *modepage_info;
6077 	struct ctl_lun *lun;
6078 	int *len_left, *len_used;
6079 	int retval, i;
6080 
6081 	ctsio = &io->scsiio;
6082 	page_index = NULL;
6083 	page_len = 0;
6084 	retval = CTL_RETVAL_COMPLETE;
6085 
6086 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6087 
6088 	if (lun->be_lun->lun_type != T_DIRECT)
6089 		control_dev = 1;
6090 	else
6091 		control_dev = 0;
6092 
6093 	modepage_info = (union ctl_modepage_info *)
6094 		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6095 	len_left = &modepage_info->header.len_left;
6096 	len_used = &modepage_info->header.len_used;
6097 
6098 do_next_page:
6099 
6100 	page_header = (struct scsi_mode_page_header *)
6101 		(ctsio->kern_data_ptr + *len_used);
6102 
6103 	if (*len_left == 0) {
6104 		free(ctsio->kern_data_ptr, M_CTL);
6105 		ctl_set_success(ctsio);
6106 		ctl_done((union ctl_io *)ctsio);
6107 		return (CTL_RETVAL_COMPLETE);
6108 	} else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6109 
6110 		free(ctsio->kern_data_ptr, M_CTL);
6111 		ctl_set_param_len_error(ctsio);
6112 		ctl_done((union ctl_io *)ctsio);
6113 		return (CTL_RETVAL_COMPLETE);
6114 
6115 	} else if ((page_header->page_code & SMPH_SPF)
6116 		&& (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6117 
6118 		free(ctsio->kern_data_ptr, M_CTL);
6119 		ctl_set_param_len_error(ctsio);
6120 		ctl_done((union ctl_io *)ctsio);
6121 		return (CTL_RETVAL_COMPLETE);
6122 	}
6123 
6124 
6125 	/*
6126 	 * XXX KDM should we do something with the block descriptor?
6127 	 */
6128 	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6129 
6130 		if ((control_dev != 0)
6131 		 && (lun->mode_pages.index[i].page_flags &
6132 		     CTL_PAGE_FLAG_DISK_ONLY))
6133 			continue;
6134 
6135 		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
6136 		    (page_header->page_code & SMPH_PC_MASK))
6137 			continue;
6138 
6139 		/*
6140 		 * If neither page has a subpage code, then we've got a
6141 		 * match.
6142 		 */
6143 		if (((lun->mode_pages.index[i].page_code & SMPH_SPF) == 0)
6144 		 && ((page_header->page_code & SMPH_SPF) == 0)) {
6145 			page_index = &lun->mode_pages.index[i];
6146 			page_len = page_header->page_length;
6147 			break;
6148 		}
6149 
6150 		/*
6151 		 * If both pages have subpages, then the subpage numbers
6152 		 * have to match.
6153 		 */
6154 		if ((lun->mode_pages.index[i].page_code & SMPH_SPF)
6155 		  && (page_header->page_code & SMPH_SPF)) {
6156 			struct scsi_mode_page_header_sp *sph;
6157 
6158 			sph = (struct scsi_mode_page_header_sp *)page_header;
6159 
6160 			if (lun->mode_pages.index[i].subpage ==
6161 			    sph->subpage) {
6162 				page_index = &lun->mode_pages.index[i];
6163 				page_len = scsi_2btoul(sph->page_length);
6164 				break;
6165 			}
6166 		}
6167 	}
6168 
6169 	/*
6170 	 * If we couldn't find the page, or if we don't have a mode select
6171 	 * handler for it, send back an error to the user.
6172 	 */
6173 	if ((page_index == NULL)
6174 	 || (page_index->select_handler == NULL)) {
6175 		ctl_set_invalid_field(ctsio,
6176 				      /*sks_valid*/ 1,
6177 				      /*command*/ 0,
6178 				      /*field*/ *len_used,
6179 				      /*bit_valid*/ 0,
6180 				      /*bit*/ 0);
6181 		free(ctsio->kern_data_ptr, M_CTL);
6182 		ctl_done((union ctl_io *)ctsio);
6183 		return (CTL_RETVAL_COMPLETE);
6184 	}
6185 
6186 	if (page_index->page_code & SMPH_SPF) {
6187 		page_len_offset = 2;
6188 		page_len_size = 2;
6189 	} else {
6190 		page_len_size = 1;
6191 		page_len_offset = 1;
6192 	}
6193 
6194 	/*
6195 	 * If the length the initiator gives us isn't the one we specify in
6196 	 * the mode page header, or if they didn't specify enough data in
6197 	 * the CDB to avoid truncating this page, kick out the request.
6198 	 */
6199 	if ((page_len != (page_index->page_len - page_len_offset -
6200 			  page_len_size))
6201 	 || (*len_left < page_index->page_len)) {
6202 
6203 
6204 		ctl_set_invalid_field(ctsio,
6205 				      /*sks_valid*/ 1,
6206 				      /*command*/ 0,
6207 				      /*field*/ *len_used + page_len_offset,
6208 				      /*bit_valid*/ 0,
6209 				      /*bit*/ 0);
6210 		free(ctsio->kern_data_ptr, M_CTL);
6211 		ctl_done((union ctl_io *)ctsio);
6212 		return (CTL_RETVAL_COMPLETE);
6213 	}
6214 
6215 	/*
6216 	 * Run through the mode page, checking to make sure that the bits
6217 	 * the user changed are actually legal for him to change.
6218 	 */
6219 	for (i = 0; i < page_index->page_len; i++) {
6220 		uint8_t *user_byte, *change_mask, *current_byte;
6221 		int bad_bit;
6222 		int j;
6223 
6224 		user_byte = (uint8_t *)page_header + i;
6225 		change_mask = page_index->page_data +
6226 			      (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6227 		current_byte = page_index->page_data +
6228 			       (page_index->page_len * CTL_PAGE_CURRENT) + i;
6229 
6230 		/*
6231 		 * Check to see whether the user set any bits in this byte
6232 		 * that he is not allowed to set.
6233 		 */
6234 		if ((*user_byte & ~(*change_mask)) ==
6235 		    (*current_byte & ~(*change_mask)))
6236 			continue;
6237 
6238 		/*
6239 		 * Go through bit by bit to determine which one is illegal.
6240 		 */
6241 		bad_bit = 0;
6242 		for (j = 7; j >= 0; j--) {
6243 			if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6244 			    (((1 << i) & ~(*change_mask)) & *current_byte)) {
6245 				bad_bit = i;
6246 				break;
6247 			}
6248 		}
6249 		ctl_set_invalid_field(ctsio,
6250 				      /*sks_valid*/ 1,
6251 				      /*command*/ 0,
6252 				      /*field*/ *len_used + i,
6253 				      /*bit_valid*/ 1,
6254 				      /*bit*/ bad_bit);
6255 		free(ctsio->kern_data_ptr, M_CTL);
6256 		ctl_done((union ctl_io *)ctsio);
6257 		return (CTL_RETVAL_COMPLETE);
6258 	}
6259 
6260 	/*
6261 	 * Decrement these before we call the page handler, since we may
6262 	 * end up getting called back one way or another before the handler
6263 	 * returns to this context.
6264 	 */
6265 	*len_left -= page_index->page_len;
6266 	*len_used += page_index->page_len;
6267 
6268 	retval = page_index->select_handler(ctsio, page_index,
6269 					    (uint8_t *)page_header);
6270 
6271 	/*
6272 	 * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6273 	 * wait until this queued command completes to finish processing
6274 	 * the mode page.  If it returns anything other than
6275 	 * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6276 	 * already set the sense information, freed the data pointer, and
6277 	 * completed the io for us.
6278 	 */
6279 	if (retval != CTL_RETVAL_COMPLETE)
6280 		goto bailout_no_done;
6281 
6282 	/*
6283 	 * If the initiator sent us more than one page, parse the next one.
6284 	 */
6285 	if (*len_left > 0)
6286 		goto do_next_page;
6287 
6288 	ctl_set_success(ctsio);
6289 	free(ctsio->kern_data_ptr, M_CTL);
6290 	ctl_done((union ctl_io *)ctsio);
6291 
6292 bailout_no_done:
6293 
6294 	return (CTL_RETVAL_COMPLETE);
6295 
6296 }
6297 
6298 int
6299 ctl_mode_select(struct ctl_scsiio *ctsio)
6300 {
6301 	int param_len, pf, sp;
6302 	int header_size, bd_len;
6303 	int len_left, len_used;
6304 	struct ctl_page_index *page_index;
6305 	struct ctl_lun *lun;
6306 	int control_dev, page_len;
6307 	union ctl_modepage_info *modepage_info;
6308 	int retval;
6309 
6310 	pf = 0;
6311 	sp = 0;
6312 	page_len = 0;
6313 	len_used = 0;
6314 	len_left = 0;
6315 	retval = 0;
6316 	bd_len = 0;
6317 	page_index = NULL;
6318 
6319 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6320 
6321 	if (lun->be_lun->lun_type != T_DIRECT)
6322 		control_dev = 1;
6323 	else
6324 		control_dev = 0;
6325 
6326 	switch (ctsio->cdb[0]) {
6327 	case MODE_SELECT_6: {
6328 		struct scsi_mode_select_6 *cdb;
6329 
6330 		cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6331 
6332 		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6333 		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6334 
6335 		param_len = cdb->length;
6336 		header_size = sizeof(struct scsi_mode_header_6);
6337 		break;
6338 	}
6339 	case MODE_SELECT_10: {
6340 		struct scsi_mode_select_10 *cdb;
6341 
6342 		cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6343 
6344 		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6345 		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6346 
6347 		param_len = scsi_2btoul(cdb->length);
6348 		header_size = sizeof(struct scsi_mode_header_10);
6349 		break;
6350 	}
6351 	default:
6352 		ctl_set_invalid_opcode(ctsio);
6353 		ctl_done((union ctl_io *)ctsio);
6354 		return (CTL_RETVAL_COMPLETE);
6355 		break; /* NOTREACHED */
6356 	}
6357 
6358 	/*
6359 	 * From SPC-3:
6360 	 * "A parameter list length of zero indicates that the Data-Out Buffer
6361 	 * shall be empty. This condition shall not be considered as an error."
6362 	 */
6363 	if (param_len == 0) {
6364 		ctl_set_success(ctsio);
6365 		ctl_done((union ctl_io *)ctsio);
6366 		return (CTL_RETVAL_COMPLETE);
6367 	}
6368 
6369 	/*
6370 	 * Since we'll hit this the first time through, prior to
6371 	 * allocation, we don't need to free a data buffer here.
6372 	 */
6373 	if (param_len < header_size) {
6374 		ctl_set_param_len_error(ctsio);
6375 		ctl_done((union ctl_io *)ctsio);
6376 		return (CTL_RETVAL_COMPLETE);
6377 	}
6378 
6379 	/*
6380 	 * Allocate the data buffer and grab the user's data.  In theory,
6381 	 * we shouldn't have to sanity check the parameter list length here
6382 	 * because the maximum size is 64K.  We should be able to malloc
6383 	 * that much without too many problems.
6384 	 */
6385 	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6386 		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6387 		ctsio->kern_data_len = param_len;
6388 		ctsio->kern_total_len = param_len;
6389 		ctsio->kern_data_resid = 0;
6390 		ctsio->kern_rel_offset = 0;
6391 		ctsio->kern_sg_entries = 0;
6392 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6393 		ctsio->be_move_done = ctl_config_move_done;
6394 		ctl_datamove((union ctl_io *)ctsio);
6395 
6396 		return (CTL_RETVAL_COMPLETE);
6397 	}
6398 
6399 	switch (ctsio->cdb[0]) {
6400 	case MODE_SELECT_6: {
6401 		struct scsi_mode_header_6 *mh6;
6402 
6403 		mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6404 		bd_len = mh6->blk_desc_len;
6405 		break;
6406 	}
6407 	case MODE_SELECT_10: {
6408 		struct scsi_mode_header_10 *mh10;
6409 
6410 		mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6411 		bd_len = scsi_2btoul(mh10->blk_desc_len);
6412 		break;
6413 	}
6414 	default:
6415 		panic("Invalid CDB type %#x", ctsio->cdb[0]);
6416 		break;
6417 	}
6418 
6419 	if (param_len < (header_size + bd_len)) {
6420 		free(ctsio->kern_data_ptr, M_CTL);
6421 		ctl_set_param_len_error(ctsio);
6422 		ctl_done((union ctl_io *)ctsio);
6423 		return (CTL_RETVAL_COMPLETE);
6424 	}
6425 
6426 	/*
6427 	 * Set the IO_CONT flag, so that if this I/O gets passed to
6428 	 * ctl_config_write_done(), it'll get passed back to
6429 	 * ctl_do_mode_select() for further processing, or completion if
6430 	 * we're all done.
6431 	 */
6432 	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6433 	ctsio->io_cont = ctl_do_mode_select;
6434 
6435 	modepage_info = (union ctl_modepage_info *)
6436 		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6437 
6438 	memset(modepage_info, 0, sizeof(*modepage_info));
6439 
6440 	len_left = param_len - header_size - bd_len;
6441 	len_used = header_size + bd_len;
6442 
6443 	modepage_info->header.len_left = len_left;
6444 	modepage_info->header.len_used = len_used;
6445 
6446 	return (ctl_do_mode_select((union ctl_io *)ctsio));
6447 }
6448 
6449 int
6450 ctl_mode_sense(struct ctl_scsiio *ctsio)
6451 {
6452 	struct ctl_lun *lun;
6453 	int pc, page_code, dbd, llba, subpage;
6454 	int alloc_len, page_len, header_len, total_len;
6455 	struct scsi_mode_block_descr *block_desc;
6456 	struct ctl_page_index *page_index;
6457 	int control_dev;
6458 
6459 	dbd = 0;
6460 	llba = 0;
6461 	block_desc = NULL;
6462 	page_index = NULL;
6463 
6464 	CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6465 
6466 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6467 
6468 	if (lun->be_lun->lun_type != T_DIRECT)
6469 		control_dev = 1;
6470 	else
6471 		control_dev = 0;
6472 
6473 	switch (ctsio->cdb[0]) {
6474 	case MODE_SENSE_6: {
6475 		struct scsi_mode_sense_6 *cdb;
6476 
6477 		cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6478 
6479 		header_len = sizeof(struct scsi_mode_hdr_6);
6480 		if (cdb->byte2 & SMS_DBD)
6481 			dbd = 1;
6482 		else
6483 			header_len += sizeof(struct scsi_mode_block_descr);
6484 
6485 		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6486 		page_code = cdb->page & SMS_PAGE_CODE;
6487 		subpage = cdb->subpage;
6488 		alloc_len = cdb->length;
6489 		break;
6490 	}
6491 	case MODE_SENSE_10: {
6492 		struct scsi_mode_sense_10 *cdb;
6493 
6494 		cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6495 
6496 		header_len = sizeof(struct scsi_mode_hdr_10);
6497 
6498 		if (cdb->byte2 & SMS_DBD)
6499 			dbd = 1;
6500 		else
6501 			header_len += sizeof(struct scsi_mode_block_descr);
6502 		if (cdb->byte2 & SMS10_LLBAA)
6503 			llba = 1;
6504 		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6505 		page_code = cdb->page & SMS_PAGE_CODE;
6506 		subpage = cdb->subpage;
6507 		alloc_len = scsi_2btoul(cdb->length);
6508 		break;
6509 	}
6510 	default:
6511 		ctl_set_invalid_opcode(ctsio);
6512 		ctl_done((union ctl_io *)ctsio);
6513 		return (CTL_RETVAL_COMPLETE);
6514 		break; /* NOTREACHED */
6515 	}
6516 
6517 	/*
6518 	 * We have to make a first pass through to calculate the size of
6519 	 * the pages that match the user's query.  Then we allocate enough
6520 	 * memory to hold it, and actually copy the data into the buffer.
6521 	 */
6522 	switch (page_code) {
6523 	case SMS_ALL_PAGES_PAGE: {
6524 		int i;
6525 
6526 		page_len = 0;
6527 
6528 		/*
6529 		 * At the moment, values other than 0 and 0xff here are
6530 		 * reserved according to SPC-3.
6531 		 */
6532 		if ((subpage != SMS_SUBPAGE_PAGE_0)
6533 		 && (subpage != SMS_SUBPAGE_ALL)) {
6534 			ctl_set_invalid_field(ctsio,
6535 					      /*sks_valid*/ 1,
6536 					      /*command*/ 1,
6537 					      /*field*/ 3,
6538 					      /*bit_valid*/ 0,
6539 					      /*bit*/ 0);
6540 			ctl_done((union ctl_io *)ctsio);
6541 			return (CTL_RETVAL_COMPLETE);
6542 		}
6543 
6544 		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6545 			if ((control_dev != 0)
6546 			 && (lun->mode_pages.index[i].page_flags &
6547 			     CTL_PAGE_FLAG_DISK_ONLY))
6548 				continue;
6549 
6550 			/*
6551 			 * We don't use this subpage if the user didn't
6552 			 * request all subpages.
6553 			 */
6554 			if ((lun->mode_pages.index[i].subpage != 0)
6555 			 && (subpage == SMS_SUBPAGE_PAGE_0))
6556 				continue;
6557 
6558 #if 0
6559 			printf("found page %#x len %d\n",
6560 			       lun->mode_pages.index[i].page_code &
6561 			       SMPH_PC_MASK,
6562 			       lun->mode_pages.index[i].page_len);
6563 #endif
6564 			page_len += lun->mode_pages.index[i].page_len;
6565 		}
6566 		break;
6567 	}
6568 	default: {
6569 		int i;
6570 
6571 		page_len = 0;
6572 
6573 		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6574 			/* Look for the right page code */
6575 			if ((lun->mode_pages.index[i].page_code &
6576 			     SMPH_PC_MASK) != page_code)
6577 				continue;
6578 
6579 			/* Look for the right subpage or the subpage wildcard*/
6580 			if ((lun->mode_pages.index[i].subpage != subpage)
6581 			 && (subpage != SMS_SUBPAGE_ALL))
6582 				continue;
6583 
6584 			/* Make sure the page is supported for this dev type */
6585 			if ((control_dev != 0)
6586 			 && (lun->mode_pages.index[i].page_flags &
6587 			     CTL_PAGE_FLAG_DISK_ONLY))
6588 				continue;
6589 
6590 #if 0
6591 			printf("found page %#x len %d\n",
6592 			       lun->mode_pages.index[i].page_code &
6593 			       SMPH_PC_MASK,
6594 			       lun->mode_pages.index[i].page_len);
6595 #endif
6596 
6597 			page_len += lun->mode_pages.index[i].page_len;
6598 		}
6599 
6600 		if (page_len == 0) {
6601 			ctl_set_invalid_field(ctsio,
6602 					      /*sks_valid*/ 1,
6603 					      /*command*/ 1,
6604 					      /*field*/ 2,
6605 					      /*bit_valid*/ 1,
6606 					      /*bit*/ 5);
6607 			ctl_done((union ctl_io *)ctsio);
6608 			return (CTL_RETVAL_COMPLETE);
6609 		}
6610 		break;
6611 	}
6612 	}
6613 
6614 	total_len = header_len + page_len;
6615 #if 0
6616 	printf("header_len = %d, page_len = %d, total_len = %d\n",
6617 	       header_len, page_len, total_len);
6618 #endif
6619 
6620 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6621 	ctsio->kern_sg_entries = 0;
6622 	ctsio->kern_data_resid = 0;
6623 	ctsio->kern_rel_offset = 0;
6624 	if (total_len < alloc_len) {
6625 		ctsio->residual = alloc_len - total_len;
6626 		ctsio->kern_data_len = total_len;
6627 		ctsio->kern_total_len = total_len;
6628 	} else {
6629 		ctsio->residual = 0;
6630 		ctsio->kern_data_len = alloc_len;
6631 		ctsio->kern_total_len = alloc_len;
6632 	}
6633 
6634 	switch (ctsio->cdb[0]) {
6635 	case MODE_SENSE_6: {
6636 		struct scsi_mode_hdr_6 *header;
6637 
6638 		header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6639 
6640 		header->datalen = ctl_min(total_len - 1, 254);
6641 
6642 		if (dbd)
6643 			header->block_descr_len = 0;
6644 		else
6645 			header->block_descr_len =
6646 				sizeof(struct scsi_mode_block_descr);
6647 		block_desc = (struct scsi_mode_block_descr *)&header[1];
6648 		break;
6649 	}
6650 	case MODE_SENSE_10: {
6651 		struct scsi_mode_hdr_10 *header;
6652 		int datalen;
6653 
6654 		header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6655 
6656 		datalen = ctl_min(total_len - 2, 65533);
6657 		scsi_ulto2b(datalen, header->datalen);
6658 		if (dbd)
6659 			scsi_ulto2b(0, header->block_descr_len);
6660 		else
6661 			scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
6662 				    header->block_descr_len);
6663 		block_desc = (struct scsi_mode_block_descr *)&header[1];
6664 		break;
6665 	}
6666 	default:
6667 		panic("invalid CDB type %#x", ctsio->cdb[0]);
6668 		break; /* NOTREACHED */
6669 	}
6670 
6671 	/*
6672 	 * If we've got a disk, use its blocksize in the block
6673 	 * descriptor.  Otherwise, just set it to 0.
6674 	 */
6675 	if (dbd == 0) {
6676 		if (control_dev != 0)
6677 			scsi_ulto3b(lun->be_lun->blocksize,
6678 				    block_desc->block_len);
6679 		else
6680 			scsi_ulto3b(0, block_desc->block_len);
6681 	}
6682 
6683 	switch (page_code) {
6684 	case SMS_ALL_PAGES_PAGE: {
6685 		int i, data_used;
6686 
6687 		data_used = header_len;
6688 		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6689 			struct ctl_page_index *page_index;
6690 
6691 			page_index = &lun->mode_pages.index[i];
6692 
6693 			if ((control_dev != 0)
6694 			 && (page_index->page_flags &
6695 			    CTL_PAGE_FLAG_DISK_ONLY))
6696 				continue;
6697 
6698 			/*
6699 			 * We don't use this subpage if the user didn't
6700 			 * request all subpages.  We already checked (above)
6701 			 * to make sure the user only specified a subpage
6702 			 * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
6703 			 */
6704 			if ((page_index->subpage != 0)
6705 			 && (subpage == SMS_SUBPAGE_PAGE_0))
6706 				continue;
6707 
6708 			/*
6709 			 * Call the handler, if it exists, to update the
6710 			 * page to the latest values.
6711 			 */
6712 			if (page_index->sense_handler != NULL)
6713 				page_index->sense_handler(ctsio, page_index,pc);
6714 
6715 			memcpy(ctsio->kern_data_ptr + data_used,
6716 			       page_index->page_data +
6717 			       (page_index->page_len * pc),
6718 			       page_index->page_len);
6719 			data_used += page_index->page_len;
6720 		}
6721 		break;
6722 	}
6723 	default: {
6724 		int i, data_used;
6725 
6726 		data_used = header_len;
6727 
6728 		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6729 			struct ctl_page_index *page_index;
6730 
6731 			page_index = &lun->mode_pages.index[i];
6732 
6733 			/* Look for the right page code */
6734 			if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6735 				continue;
6736 
6737 			/* Look for the right subpage or the subpage wildcard*/
6738 			if ((page_index->subpage != subpage)
6739 			 && (subpage != SMS_SUBPAGE_ALL))
6740 				continue;
6741 
6742 			/* Make sure the page is supported for this dev type */
6743 			if ((control_dev != 0)
6744 			 && (page_index->page_flags &
6745 			     CTL_PAGE_FLAG_DISK_ONLY))
6746 				continue;
6747 
6748 			/*
6749 			 * Call the handler, if it exists, to update the
6750 			 * page to the latest values.
6751 			 */
6752 			if (page_index->sense_handler != NULL)
6753 				page_index->sense_handler(ctsio, page_index,pc);
6754 
6755 			memcpy(ctsio->kern_data_ptr + data_used,
6756 			       page_index->page_data +
6757 			       (page_index->page_len * pc),
6758 			       page_index->page_len);
6759 			data_used += page_index->page_len;
6760 		}
6761 		break;
6762 	}
6763 	}
6764 
6765 	ctsio->scsi_status = SCSI_STATUS_OK;
6766 
6767 	ctsio->be_move_done = ctl_config_move_done;
6768 	ctl_datamove((union ctl_io *)ctsio);
6769 
6770 	return (CTL_RETVAL_COMPLETE);
6771 }
6772 
6773 int
6774 ctl_read_capacity(struct ctl_scsiio *ctsio)
6775 {
6776 	struct scsi_read_capacity *cdb;
6777 	struct scsi_read_capacity_data *data;
6778 	struct ctl_lun *lun;
6779 	uint32_t lba;
6780 
6781 	CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
6782 
6783 	cdb = (struct scsi_read_capacity *)ctsio->cdb;
6784 
6785 	lba = scsi_4btoul(cdb->addr);
6786 	if (((cdb->pmi & SRC_PMI) == 0)
6787 	 && (lba != 0)) {
6788 		ctl_set_invalid_field(/*ctsio*/ ctsio,
6789 				      /*sks_valid*/ 1,
6790 				      /*command*/ 1,
6791 				      /*field*/ 2,
6792 				      /*bit_valid*/ 0,
6793 				      /*bit*/ 0);
6794 		ctl_done((union ctl_io *)ctsio);
6795 		return (CTL_RETVAL_COMPLETE);
6796 	}
6797 
6798 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6799 
6800 	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6801 	data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
6802 	ctsio->residual = 0;
6803 	ctsio->kern_data_len = sizeof(*data);
6804 	ctsio->kern_total_len = sizeof(*data);
6805 	ctsio->kern_data_resid = 0;
6806 	ctsio->kern_rel_offset = 0;
6807 	ctsio->kern_sg_entries = 0;
6808 
6809 	/*
6810 	 * If the maximum LBA is greater than 0xfffffffe, the user must
6811 	 * issue a SERVICE ACTION IN (16) command, with the read capacity
6812 	 * serivce action set.
6813 	 */
6814 	if (lun->be_lun->maxlba > 0xfffffffe)
6815 		scsi_ulto4b(0xffffffff, data->addr);
6816 	else
6817 		scsi_ulto4b(lun->be_lun->maxlba, data->addr);
6818 
6819 	/*
6820 	 * XXX KDM this may not be 512 bytes...
6821 	 */
6822 	scsi_ulto4b(lun->be_lun->blocksize, data->length);
6823 
6824 	ctsio->scsi_status = SCSI_STATUS_OK;
6825 
6826 	ctsio->be_move_done = ctl_config_move_done;
6827 	ctl_datamove((union ctl_io *)ctsio);
6828 
6829 	return (CTL_RETVAL_COMPLETE);
6830 }
6831 
6832 static int
6833 ctl_read_capacity_16(struct ctl_scsiio *ctsio)
6834 {
6835 	struct scsi_read_capacity_16 *cdb;
6836 	struct scsi_read_capacity_data_long *data;
6837 	struct ctl_lun *lun;
6838 	uint64_t lba;
6839 	uint32_t alloc_len;
6840 
6841 	CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
6842 
6843 	cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
6844 
6845 	alloc_len = scsi_4btoul(cdb->alloc_len);
6846 	lba = scsi_8btou64(cdb->addr);
6847 
6848 	if ((cdb->reladr & SRC16_PMI)
6849 	 && (lba != 0)) {
6850 		ctl_set_invalid_field(/*ctsio*/ ctsio,
6851 				      /*sks_valid*/ 1,
6852 				      /*command*/ 1,
6853 				      /*field*/ 2,
6854 				      /*bit_valid*/ 0,
6855 				      /*bit*/ 0);
6856 		ctl_done((union ctl_io *)ctsio);
6857 		return (CTL_RETVAL_COMPLETE);
6858 	}
6859 
6860 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6861 
6862 	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6863 	data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
6864 
6865 	if (sizeof(*data) < alloc_len) {
6866 		ctsio->residual = alloc_len - sizeof(*data);
6867 		ctsio->kern_data_len = sizeof(*data);
6868 		ctsio->kern_total_len = sizeof(*data);
6869 	} else {
6870 		ctsio->residual = 0;
6871 		ctsio->kern_data_len = alloc_len;
6872 		ctsio->kern_total_len = alloc_len;
6873 	}
6874 	ctsio->kern_data_resid = 0;
6875 	ctsio->kern_rel_offset = 0;
6876 	ctsio->kern_sg_entries = 0;
6877 
6878 	scsi_u64to8b(lun->be_lun->maxlba, data->addr);
6879 	/* XXX KDM this may not be 512 bytes... */
6880 	scsi_ulto4b(lun->be_lun->blocksize, data->length);
6881 
6882 	ctsio->scsi_status = SCSI_STATUS_OK;
6883 
6884 	ctsio->be_move_done = ctl_config_move_done;
6885 	ctl_datamove((union ctl_io *)ctsio);
6886 
6887 	return (CTL_RETVAL_COMPLETE);
6888 }
6889 
6890 int
6891 ctl_service_action_in(struct ctl_scsiio *ctsio)
6892 {
6893 	struct scsi_service_action_in *cdb;
6894 	int retval;
6895 
6896 	CTL_DEBUG_PRINT(("ctl_service_action_in\n"));
6897 
6898 	cdb = (struct scsi_service_action_in *)ctsio->cdb;
6899 
6900 	retval = CTL_RETVAL_COMPLETE;
6901 
6902 	switch (cdb->service_action) {
6903 	case SRC16_SERVICE_ACTION:
6904 		retval = ctl_read_capacity_16(ctsio);
6905 		break;
6906 	default:
6907 		ctl_set_invalid_field(/*ctsio*/ ctsio,
6908 				      /*sks_valid*/ 1,
6909 				      /*command*/ 1,
6910 				      /*field*/ 1,
6911 				      /*bit_valid*/ 1,
6912 				      /*bit*/ 4);
6913 		ctl_done((union ctl_io *)ctsio);
6914 		break;
6915 	}
6916 
6917 	return (retval);
6918 }
6919 
6920 int
6921 ctl_maintenance_in(struct ctl_scsiio *ctsio)
6922 {
6923 	struct scsi_maintenance_in *cdb;
6924 	int retval;
6925 	int alloc_len, total_len = 0;
6926 	int num_target_port_groups, single;
6927 	struct ctl_lun *lun;
6928 	struct ctl_softc *softc;
6929 	struct scsi_target_group_data *rtg_ptr;
6930 	struct scsi_target_port_group_descriptor *tpg_desc_ptr1, *tpg_desc_ptr2;
6931 	struct scsi_target_port_descriptor  *tp_desc_ptr1_1, *tp_desc_ptr1_2,
6932 	                                    *tp_desc_ptr2_1, *tp_desc_ptr2_2;
6933 
6934 	CTL_DEBUG_PRINT(("ctl_maintenance_in\n"));
6935 
6936 	cdb = (struct scsi_maintenance_in *)ctsio->cdb;
6937 	softc = control_softc;
6938 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6939 
6940 	retval = CTL_RETVAL_COMPLETE;
6941 
6942 	if ((cdb->byte2 & SERVICE_ACTION_MASK) != SA_RPRT_TRGT_GRP) {
6943 		ctl_set_invalid_field(/*ctsio*/ ctsio,
6944 				      /*sks_valid*/ 1,
6945 				      /*command*/ 1,
6946 				      /*field*/ 1,
6947 				      /*bit_valid*/ 1,
6948 				      /*bit*/ 4);
6949 		ctl_done((union ctl_io *)ctsio);
6950 		return(retval);
6951 	}
6952 
6953 	mtx_lock(&softc->ctl_lock);
6954 	single = ctl_is_single;
6955 	mtx_unlock(&softc->ctl_lock);
6956 
6957 	if (single)
6958         	num_target_port_groups = NUM_TARGET_PORT_GROUPS - 1;
6959 	else
6960         	num_target_port_groups = NUM_TARGET_PORT_GROUPS;
6961 
6962 	total_len = sizeof(struct scsi_target_group_data) +
6963 		sizeof(struct scsi_target_port_group_descriptor) *
6964 		num_target_port_groups +
6965 		sizeof(struct scsi_target_port_descriptor) *
6966 		NUM_PORTS_PER_GRP * num_target_port_groups;
6967 
6968 	alloc_len = scsi_4btoul(cdb->length);
6969 
6970 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6971 
6972 	ctsio->kern_sg_entries = 0;
6973 
6974 	if (total_len < alloc_len) {
6975 		ctsio->residual = alloc_len - total_len;
6976 		ctsio->kern_data_len = total_len;
6977 		ctsio->kern_total_len = total_len;
6978 	} else {
6979 		ctsio->residual = 0;
6980 		ctsio->kern_data_len = alloc_len;
6981 		ctsio->kern_total_len = alloc_len;
6982 	}
6983 	ctsio->kern_data_resid = 0;
6984 	ctsio->kern_rel_offset = 0;
6985 
6986 	rtg_ptr = (struct scsi_target_group_data *)ctsio->kern_data_ptr;
6987 
6988 	tpg_desc_ptr1 = &rtg_ptr->groups[0];
6989 	tp_desc_ptr1_1 = &tpg_desc_ptr1->descriptors[0];
6990 	tp_desc_ptr1_2 = (struct scsi_target_port_descriptor *)
6991 	        &tp_desc_ptr1_1->desc_list[0];
6992 
6993 	if (single == 0) {
6994 		tpg_desc_ptr2 = (struct scsi_target_port_group_descriptor *)
6995 	                &tp_desc_ptr1_2->desc_list[0];
6996 		tp_desc_ptr2_1 = &tpg_desc_ptr2->descriptors[0];
6997 		tp_desc_ptr2_2 = (struct scsi_target_port_descriptor *)
6998 	        	&tp_desc_ptr2_1->desc_list[0];
6999         } else {
7000 		tpg_desc_ptr2 = NULL;
7001 		tp_desc_ptr2_1 = NULL;
7002 		tp_desc_ptr2_2 = NULL;
7003 	}
7004 
7005 	scsi_ulto4b(total_len - 4, rtg_ptr->length);
7006 	if (single == 0) {
7007         	if (ctsio->io_hdr.nexus.targ_port < CTL_MAX_PORTS) {
7008 			if (lun->flags & CTL_LUN_PRIMARY_SC) {
7009 				tpg_desc_ptr1->pref_state = TPG_PRIMARY;
7010 				tpg_desc_ptr2->pref_state =
7011 					TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7012 			} else {
7013 				tpg_desc_ptr1->pref_state =
7014 					TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7015 				tpg_desc_ptr2->pref_state = TPG_PRIMARY;
7016 			}
7017 		} else {
7018 			if (lun->flags & CTL_LUN_PRIMARY_SC) {
7019 				tpg_desc_ptr1->pref_state =
7020 					TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7021 				tpg_desc_ptr2->pref_state = TPG_PRIMARY;
7022 			} else {
7023 				tpg_desc_ptr1->pref_state = TPG_PRIMARY;
7024 				tpg_desc_ptr2->pref_state =
7025 					TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7026 			}
7027 		}
7028 	} else {
7029 		tpg_desc_ptr1->pref_state = TPG_PRIMARY;
7030 	}
7031 	tpg_desc_ptr1->support = 0;
7032 	tpg_desc_ptr1->target_port_group[1] = 1;
7033 	tpg_desc_ptr1->status = TPG_IMPLICIT;
7034 	tpg_desc_ptr1->target_port_count= NUM_PORTS_PER_GRP;
7035 
7036 	if (single == 0) {
7037 		tpg_desc_ptr2->support = 0;
7038 		tpg_desc_ptr2->target_port_group[1] = 2;
7039 		tpg_desc_ptr2->status = TPG_IMPLICIT;
7040 		tpg_desc_ptr2->target_port_count = NUM_PORTS_PER_GRP;
7041 
7042 		tp_desc_ptr1_1->relative_target_port_identifier[1] = 1;
7043 		tp_desc_ptr1_2->relative_target_port_identifier[1] = 2;
7044 
7045 		tp_desc_ptr2_1->relative_target_port_identifier[1] = 9;
7046 		tp_desc_ptr2_2->relative_target_port_identifier[1] = 10;
7047 	} else {
7048         	if (ctsio->io_hdr.nexus.targ_port < CTL_MAX_PORTS) {
7049 			tp_desc_ptr1_1->relative_target_port_identifier[1] = 1;
7050 			tp_desc_ptr1_2->relative_target_port_identifier[1] = 2;
7051 		} else {
7052 			tp_desc_ptr1_1->relative_target_port_identifier[1] = 9;
7053 			tp_desc_ptr1_2->relative_target_port_identifier[1] = 10;
7054 		}
7055 	}
7056 
7057 	ctsio->be_move_done = ctl_config_move_done;
7058 
7059 	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7060 			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7061 			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7062 			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7063 			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7064 
7065 	ctl_datamove((union ctl_io *)ctsio);
7066 	return(retval);
7067 }
7068 
7069 int
7070 ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7071 {
7072 	struct scsi_per_res_in *cdb;
7073 	int alloc_len, total_len = 0;
7074 	/* struct scsi_per_res_in_rsrv in_data; */
7075 	struct ctl_lun *lun;
7076 	struct ctl_softc *softc;
7077 
7078 	CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7079 
7080 	softc = control_softc;
7081 
7082 	cdb = (struct scsi_per_res_in *)ctsio->cdb;
7083 
7084 	alloc_len = scsi_2btoul(cdb->length);
7085 
7086 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7087 
7088 retry:
7089 	mtx_lock(&softc->ctl_lock);
7090 	switch (cdb->action) {
7091 	case SPRI_RK: /* read keys */
7092 		total_len = sizeof(struct scsi_per_res_in_keys) +
7093 			lun->pr_key_count *
7094 			sizeof(struct scsi_per_res_key);
7095 		break;
7096 	case SPRI_RR: /* read reservation */
7097 		if (lun->flags & CTL_LUN_PR_RESERVED)
7098 			total_len = sizeof(struct scsi_per_res_in_rsrv);
7099 		else
7100 			total_len = sizeof(struct scsi_per_res_in_header);
7101 		break;
7102 	case SPRI_RC: /* report capabilities */
7103 		total_len = sizeof(struct scsi_per_res_cap);
7104 		break;
7105 	case SPRI_RS: /* read full status */
7106 	default:
7107 		mtx_unlock(&softc->ctl_lock);
7108 		ctl_set_invalid_field(ctsio,
7109 				      /*sks_valid*/ 1,
7110 				      /*command*/ 1,
7111 				      /*field*/ 1,
7112 				      /*bit_valid*/ 1,
7113 				      /*bit*/ 0);
7114 		ctl_done((union ctl_io *)ctsio);
7115 		return (CTL_RETVAL_COMPLETE);
7116 		break; /* NOTREACHED */
7117 	}
7118 	mtx_unlock(&softc->ctl_lock);
7119 
7120 	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7121 
7122 	if (total_len < alloc_len) {
7123 		ctsio->residual = alloc_len - total_len;
7124 		ctsio->kern_data_len = total_len;
7125 		ctsio->kern_total_len = total_len;
7126 	} else {
7127 		ctsio->residual = 0;
7128 		ctsio->kern_data_len = alloc_len;
7129 		ctsio->kern_total_len = alloc_len;
7130 	}
7131 
7132 	ctsio->kern_data_resid = 0;
7133 	ctsio->kern_rel_offset = 0;
7134 	ctsio->kern_sg_entries = 0;
7135 
7136 	mtx_lock(&softc->ctl_lock);
7137 	switch (cdb->action) {
7138 	case SPRI_RK: { // read keys
7139         struct scsi_per_res_in_keys *res_keys;
7140 		int i, key_count;
7141 
7142 		res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7143 
7144 		/*
7145 		 * We had to drop the lock to allocate our buffer, which
7146 		 * leaves time for someone to come in with another
7147 		 * persistent reservation.  (That is unlikely, though,
7148 		 * since this should be the only persistent reservation
7149 		 * command active right now.)
7150 		 */
7151 		if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7152 		    (lun->pr_key_count *
7153 		     sizeof(struct scsi_per_res_key)))){
7154 			mtx_unlock(&softc->ctl_lock);
7155 			free(ctsio->kern_data_ptr, M_CTL);
7156 			printf("%s: reservation length changed, retrying\n",
7157 			       __func__);
7158 			goto retry;
7159 		}
7160 
7161 		scsi_ulto4b(lun->PRGeneration, res_keys->header.generation);
7162 
7163 		scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7164 			     lun->pr_key_count, res_keys->header.length);
7165 
7166 		for (i = 0, key_count = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7167 			if (!lun->per_res[i].registered)
7168 				continue;
7169 
7170 			/*
7171 			 * We used lun->pr_key_count to calculate the
7172 			 * size to allocate.  If it turns out the number of
7173 			 * initiators with the registered flag set is
7174 			 * larger than that (i.e. they haven't been kept in
7175 			 * sync), we've got a problem.
7176 			 */
7177 			if (key_count >= lun->pr_key_count) {
7178 #ifdef NEEDTOPORT
7179 				csevent_log(CSC_CTL | CSC_SHELF_SW |
7180 					    CTL_PR_ERROR,
7181 					    csevent_LogType_Fault,
7182 					    csevent_AlertLevel_Yellow,
7183 					    csevent_FRU_ShelfController,
7184 					    csevent_FRU_Firmware,
7185 				        csevent_FRU_Unknown,
7186 					    "registered keys %d >= key "
7187 					    "count %d", key_count,
7188 					    lun->pr_key_count);
7189 #endif
7190 				key_count++;
7191 				continue;
7192 			}
7193 			memcpy(res_keys->keys[key_count].key,
7194 			       lun->per_res[i].res_key.key,
7195 			       ctl_min(sizeof(res_keys->keys[key_count].key),
7196 			       sizeof(lun->per_res[i].res_key)));
7197 			key_count++;
7198 		}
7199 		break;
7200 	}
7201 	case SPRI_RR: { // read reservation
7202 		struct scsi_per_res_in_rsrv *res;
7203 		int tmp_len, header_only;
7204 
7205 		res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7206 
7207 		scsi_ulto4b(lun->PRGeneration, res->header.generation);
7208 
7209 		if (lun->flags & CTL_LUN_PR_RESERVED)
7210 		{
7211 			tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7212 			scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7213 				    res->header.length);
7214 			header_only = 0;
7215 		} else {
7216 			tmp_len = sizeof(struct scsi_per_res_in_header);
7217 			scsi_ulto4b(0, res->header.length);
7218 			header_only = 1;
7219 		}
7220 
7221 		/*
7222 		 * We had to drop the lock to allocate our buffer, which
7223 		 * leaves time for someone to come in with another
7224 		 * persistent reservation.  (That is unlikely, though,
7225 		 * since this should be the only persistent reservation
7226 		 * command active right now.)
7227 		 */
7228 		if (tmp_len != total_len) {
7229 			mtx_unlock(&softc->ctl_lock);
7230 			free(ctsio->kern_data_ptr, M_CTL);
7231 			printf("%s: reservation status changed, retrying\n",
7232 			       __func__);
7233 			goto retry;
7234 		}
7235 
7236 		/*
7237 		 * No reservation held, so we're done.
7238 		 */
7239 		if (header_only != 0)
7240 			break;
7241 
7242 		/*
7243 		 * If the registration is an All Registrants type, the key
7244 		 * is 0, since it doesn't really matter.
7245 		 */
7246 		if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7247 			memcpy(res->data.reservation,
7248 			       &lun->per_res[lun->pr_res_idx].res_key,
7249 			       sizeof(struct scsi_per_res_key));
7250 		}
7251 		res->data.scopetype = lun->res_type;
7252 		break;
7253 	}
7254 	case SPRI_RC:     //report capabilities
7255 	{
7256 		struct scsi_per_res_cap *res_cap;
7257 		uint16_t type_mask;
7258 
7259 		res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7260 		scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7261 		res_cap->flags2 |= SPRI_TMV;
7262 		type_mask = SPRI_TM_WR_EX_AR |
7263 			    SPRI_TM_EX_AC_RO |
7264 			    SPRI_TM_WR_EX_RO |
7265 			    SPRI_TM_EX_AC |
7266 			    SPRI_TM_WR_EX |
7267 			    SPRI_TM_EX_AC_AR;
7268 		scsi_ulto2b(type_mask, res_cap->type_mask);
7269 		break;
7270 	}
7271 	case SPRI_RS: //read full status
7272 	default:
7273 		/*
7274 		 * This is a bug, because we just checked for this above,
7275 		 * and should have returned an error.
7276 		 */
7277 		panic("Invalid PR type %x", cdb->action);
7278 		break; /* NOTREACHED */
7279 	}
7280 	mtx_unlock(&softc->ctl_lock);
7281 
7282 	ctsio->be_move_done = ctl_config_move_done;
7283 
7284 	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7285 			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7286 			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7287 			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7288 			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7289 
7290 	ctl_datamove((union ctl_io *)ctsio);
7291 
7292 	return (CTL_RETVAL_COMPLETE);
7293 }
7294 
7295 /*
7296  * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7297  * it should return.
7298  */
7299 static int
7300 ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7301 		uint64_t sa_res_key, uint8_t type, uint32_t residx,
7302 		struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7303 		struct scsi_per_res_out_parms* param)
7304 {
7305 	union ctl_ha_msg persis_io;
7306 	int retval, i;
7307 	int isc_retval;
7308 
7309 	retval = 0;
7310 
7311 	if (sa_res_key == 0) {
7312 		mtx_lock(&softc->ctl_lock);
7313 		if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7314 			/* validate scope and type */
7315 			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7316 			     SPR_LU_SCOPE) {
7317 				mtx_unlock(&softc->ctl_lock);
7318 				ctl_set_invalid_field(/*ctsio*/ ctsio,
7319 						      /*sks_valid*/ 1,
7320 						      /*command*/ 1,
7321 						      /*field*/ 2,
7322 						      /*bit_valid*/ 1,
7323 						      /*bit*/ 4);
7324 				ctl_done((union ctl_io *)ctsio);
7325 				return (1);
7326 			}
7327 
7328 		        if (type>8 || type==2 || type==4 || type==0) {
7329 				mtx_unlock(&softc->ctl_lock);
7330 				ctl_set_invalid_field(/*ctsio*/ ctsio,
7331        	           				      /*sks_valid*/ 1,
7332 						      /*command*/ 1,
7333 						      /*field*/ 2,
7334 						      /*bit_valid*/ 1,
7335 						      /*bit*/ 0);
7336 				ctl_done((union ctl_io *)ctsio);
7337 				return (1);
7338 		        }
7339 
7340 			/* temporarily unregister this nexus */
7341 			lun->per_res[residx].registered = 0;
7342 
7343 			/*
7344 			 * Unregister everybody else and build UA for
7345 			 * them
7346 			 */
7347 			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7348 				if (lun->per_res[i].registered == 0)
7349 					continue;
7350 
7351 				if (!persis_offset
7352 				 && i <CTL_MAX_INITIATORS)
7353 					lun->pending_sense[i].ua_pending |=
7354 						CTL_UA_REG_PREEMPT;
7355 				else if (persis_offset
7356 				      && i >= persis_offset)
7357 					lun->pending_sense[i-persis_offset
7358 						].ua_pending |=
7359 						CTL_UA_REG_PREEMPT;
7360 				lun->per_res[i].registered = 0;
7361 				memset(&lun->per_res[i].res_key, 0,
7362 				       sizeof(struct scsi_per_res_key));
7363 			}
7364 			lun->per_res[residx].registered = 1;
7365 			lun->pr_key_count = 1;
7366 			lun->res_type = type;
7367 			if (lun->res_type != SPR_TYPE_WR_EX_AR
7368 			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7369 				lun->pr_res_idx = residx;
7370 
7371 			mtx_unlock(&softc->ctl_lock);
7372 			/* send msg to other side */
7373 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7374 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7375 			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7376 			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7377 			persis_io.pr.pr_info.res_type = type;
7378 			memcpy(persis_io.pr.pr_info.sa_res_key,
7379 			       param->serv_act_res_key,
7380 			       sizeof(param->serv_act_res_key));
7381 			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7382 			     &persis_io, sizeof(persis_io), 0)) >
7383 			     CTL_HA_STATUS_SUCCESS) {
7384 				printf("CTL:Persis Out error returned "
7385 				       "from ctl_ha_msg_send %d\n",
7386 				       isc_retval);
7387 			}
7388 		} else {
7389 			/* not all registrants */
7390 			mtx_unlock(&softc->ctl_lock);
7391 			free(ctsio->kern_data_ptr, M_CTL);
7392 			ctl_set_invalid_field(ctsio,
7393 					      /*sks_valid*/ 1,
7394 					      /*command*/ 0,
7395 					      /*field*/ 8,
7396 					      /*bit_valid*/ 0,
7397 					      /*bit*/ 0);
7398 			ctl_done((union ctl_io *)ctsio);
7399 			return (1);
7400 		}
7401 	} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7402 		|| !(lun->flags & CTL_LUN_PR_RESERVED)) {
7403 		int found = 0;
7404 
7405 		mtx_lock(&softc->ctl_lock);
7406 		if (res_key == sa_res_key) {
7407 			/* special case */
7408 			/*
7409 			 * The spec implies this is not good but doesn't
7410 			 * say what to do. There are two choices either
7411 			 * generate a res conflict or check condition
7412 			 * with illegal field in parameter data. Since
7413 			 * that is what is done when the sa_res_key is
7414 			 * zero I'll take that approach since this has
7415 			 * to do with the sa_res_key.
7416 			 */
7417 			mtx_unlock(&softc->ctl_lock);
7418 			free(ctsio->kern_data_ptr, M_CTL);
7419 			ctl_set_invalid_field(ctsio,
7420 					      /*sks_valid*/ 1,
7421 					      /*command*/ 0,
7422 					      /*field*/ 8,
7423 					      /*bit_valid*/ 0,
7424 					      /*bit*/ 0);
7425 			ctl_done((union ctl_io *)ctsio);
7426 			return (1);
7427 		}
7428 
7429 		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7430 			if (lun->per_res[i].registered
7431 			 && memcmp(param->serv_act_res_key,
7432 			    lun->per_res[i].res_key.key,
7433 			    sizeof(struct scsi_per_res_key)) != 0)
7434 				continue;
7435 
7436 			found = 1;
7437 			lun->per_res[i].registered = 0;
7438 			memset(&lun->per_res[i].res_key, 0,
7439 			       sizeof(struct scsi_per_res_key));
7440 			lun->pr_key_count--;
7441 
7442 			if (!persis_offset
7443 			 && i < CTL_MAX_INITIATORS)
7444 				lun->pending_sense[i].ua_pending |=
7445 					CTL_UA_REG_PREEMPT;
7446 			else if (persis_offset
7447 			      && i >= persis_offset)
7448 				lun->pending_sense[i-persis_offset].ua_pending|=
7449 					CTL_UA_REG_PREEMPT;
7450 		}
7451 		mtx_unlock(&softc->ctl_lock);
7452 		if (!found) {
7453 			free(ctsio->kern_data_ptr, M_CTL);
7454 			ctl_set_reservation_conflict(ctsio);
7455 			ctl_done((union ctl_io *)ctsio);
7456 			return (CTL_RETVAL_COMPLETE);
7457 		}
7458 		/* send msg to other side */
7459 		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7460 		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7461 		persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7462 		persis_io.pr.pr_info.residx = lun->pr_res_idx;
7463 		persis_io.pr.pr_info.res_type = type;
7464 		memcpy(persis_io.pr.pr_info.sa_res_key,
7465 		       param->serv_act_res_key,
7466 		       sizeof(param->serv_act_res_key));
7467 		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7468 		     &persis_io, sizeof(persis_io), 0)) >
7469 		     CTL_HA_STATUS_SUCCESS) {
7470 			printf("CTL:Persis Out error returned from "
7471 			       "ctl_ha_msg_send %d\n", isc_retval);
7472 		}
7473 	} else {
7474 		/* Reserved but not all registrants */
7475 		/* sa_res_key is res holder */
7476 		if (memcmp(param->serv_act_res_key,
7477                    lun->per_res[lun->pr_res_idx].res_key.key,
7478                    sizeof(struct scsi_per_res_key)) == 0) {
7479 			/* validate scope and type */
7480 			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7481 			     SPR_LU_SCOPE) {
7482 				ctl_set_invalid_field(/*ctsio*/ ctsio,
7483 						      /*sks_valid*/ 1,
7484 						      /*command*/ 1,
7485 						      /*field*/ 2,
7486 						      /*bit_valid*/ 1,
7487 						      /*bit*/ 4);
7488 				ctl_done((union ctl_io *)ctsio);
7489 				return (1);
7490 			}
7491 
7492 			if (type>8 || type==2 || type==4 || type==0) {
7493 				ctl_set_invalid_field(/*ctsio*/ ctsio,
7494 						      /*sks_valid*/ 1,
7495 						      /*command*/ 1,
7496 						      /*field*/ 2,
7497 						      /*bit_valid*/ 1,
7498 						      /*bit*/ 0);
7499 				ctl_done((union ctl_io *)ctsio);
7500 				return (1);
7501 			}
7502 
7503 			/*
7504 			 * Do the following:
7505 			 * if sa_res_key != res_key remove all
7506 			 * registrants w/sa_res_key and generate UA
7507 			 * for these registrants(Registrations
7508 			 * Preempted) if it wasn't an exclusive
7509 			 * reservation generate UA(Reservations
7510 			 * Preempted) for all other registered nexuses
7511 			 * if the type has changed. Establish the new
7512 			 * reservation and holder. If res_key and
7513 			 * sa_res_key are the same do the above
7514 			 * except don't unregister the res holder.
7515 			 */
7516 
7517 			/*
7518 			 * Temporarily unregister so it won't get
7519 			 * removed or UA generated
7520 			 */
7521 			lun->per_res[residx].registered = 0;
7522 			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7523 				if (lun->per_res[i].registered == 0)
7524 					continue;
7525 
7526 				if (memcmp(param->serv_act_res_key,
7527 				    lun->per_res[i].res_key.key,
7528 				    sizeof(struct scsi_per_res_key)) == 0) {
7529 					lun->per_res[i].registered = 0;
7530 					memset(&lun->per_res[i].res_key,
7531 					       0,
7532 					       sizeof(struct scsi_per_res_key));
7533 					lun->pr_key_count--;
7534 
7535 					if (!persis_offset
7536 					 && i < CTL_MAX_INITIATORS)
7537 						lun->pending_sense[i
7538 							].ua_pending |=
7539 							CTL_UA_REG_PREEMPT;
7540 					else if (persis_offset
7541 					      && i >= persis_offset)
7542 						lun->pending_sense[
7543 						  i-persis_offset].ua_pending |=
7544 						  CTL_UA_REG_PREEMPT;
7545 				} else if (type != lun->res_type
7546 					&& (lun->res_type == SPR_TYPE_WR_EX_RO
7547 					 || lun->res_type ==SPR_TYPE_EX_AC_RO)){
7548 						if (!persis_offset
7549 						 && i < CTL_MAX_INITIATORS)
7550 							lun->pending_sense[i
7551 							].ua_pending |=
7552 							CTL_UA_RES_RELEASE;
7553 						else if (persis_offset
7554 						      && i >= persis_offset)
7555 							lun->pending_sense[
7556 							i-persis_offset
7557 							].ua_pending |=
7558 							CTL_UA_RES_RELEASE;
7559 				}
7560 			}
7561 			lun->per_res[residx].registered = 1;
7562 			lun->res_type = type;
7563 			if (lun->res_type != SPR_TYPE_WR_EX_AR
7564 			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7565 				lun->pr_res_idx = residx;
7566 			else
7567 				lun->pr_res_idx =
7568 					CTL_PR_ALL_REGISTRANTS;
7569 
7570 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7571 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7572 			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7573 			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7574 			persis_io.pr.pr_info.res_type = type;
7575 			memcpy(persis_io.pr.pr_info.sa_res_key,
7576 			       param->serv_act_res_key,
7577 			       sizeof(param->serv_act_res_key));
7578 			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7579 			     &persis_io, sizeof(persis_io), 0)) >
7580 			     CTL_HA_STATUS_SUCCESS) {
7581 				printf("CTL:Persis Out error returned "
7582 				       "from ctl_ha_msg_send %d\n",
7583 				       isc_retval);
7584 			}
7585 		} else {
7586 			/*
7587 			 * sa_res_key is not the res holder just
7588 			 * remove registrants
7589 			 */
7590 			int found=0;
7591 			mtx_lock(&softc->ctl_lock);
7592 
7593 			for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7594 				if (memcmp(param->serv_act_res_key,
7595 				    lun->per_res[i].res_key.key,
7596 				    sizeof(struct scsi_per_res_key)) != 0)
7597 					continue;
7598 
7599 				found = 1;
7600 				lun->per_res[i].registered = 0;
7601 				memset(&lun->per_res[i].res_key, 0,
7602 				       sizeof(struct scsi_per_res_key));
7603 				lun->pr_key_count--;
7604 
7605 				if (!persis_offset
7606 				 && i < CTL_MAX_INITIATORS)
7607 					lun->pending_sense[i].ua_pending |=
7608 						CTL_UA_REG_PREEMPT;
7609 				else if (persis_offset
7610 				      && i >= persis_offset)
7611 					lun->pending_sense[
7612 						i-persis_offset].ua_pending |=
7613 						CTL_UA_REG_PREEMPT;
7614 			}
7615 
7616 			if (!found) {
7617 				mtx_unlock(&softc->ctl_lock);
7618 				free(ctsio->kern_data_ptr, M_CTL);
7619 				ctl_set_reservation_conflict(ctsio);
7620 				ctl_done((union ctl_io *)ctsio);
7621 		        	return (1);
7622 			}
7623 			mtx_unlock(&softc->ctl_lock);
7624 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7625 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7626 			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7627 			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7628 			persis_io.pr.pr_info.res_type = type;
7629 			memcpy(persis_io.pr.pr_info.sa_res_key,
7630 			       param->serv_act_res_key,
7631 			       sizeof(param->serv_act_res_key));
7632 			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7633 			     &persis_io, sizeof(persis_io), 0)) >
7634 			     CTL_HA_STATUS_SUCCESS) {
7635 				printf("CTL:Persis Out error returned "
7636 				       "from ctl_ha_msg_send %d\n",
7637 				isc_retval);
7638 			}
7639 		}
7640 	}
7641 
7642 	lun->PRGeneration++;
7643 
7644 	return (retval);
7645 }
7646 
7647 static void
7648 ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
7649 {
7650 	int i;
7651 
7652 	if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7653 	 || lun->pr_res_idx == CTL_PR_NO_RESERVATION
7654 	 || memcmp(&lun->per_res[lun->pr_res_idx].res_key,
7655 		   msg->pr.pr_info.sa_res_key,
7656 		   sizeof(struct scsi_per_res_key)) != 0) {
7657 		uint64_t sa_res_key;
7658 		sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
7659 
7660 		if (sa_res_key == 0) {
7661 			/* temporarily unregister this nexus */
7662 			lun->per_res[msg->pr.pr_info.residx].registered = 0;
7663 
7664 			/*
7665 			 * Unregister everybody else and build UA for
7666 			 * them
7667 			 */
7668 			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7669 				if (lun->per_res[i].registered == 0)
7670 					continue;
7671 
7672 				if (!persis_offset
7673 				 && i < CTL_MAX_INITIATORS)
7674 					lun->pending_sense[i].ua_pending |=
7675 						CTL_UA_REG_PREEMPT;
7676 				else if (persis_offset && i >= persis_offset)
7677 					lun->pending_sense[i -
7678 						persis_offset].ua_pending |=
7679 						CTL_UA_REG_PREEMPT;
7680 				lun->per_res[i].registered = 0;
7681 				memset(&lun->per_res[i].res_key, 0,
7682 				       sizeof(struct scsi_per_res_key));
7683 			}
7684 
7685 			lun->per_res[msg->pr.pr_info.residx].registered = 1;
7686 			lun->pr_key_count = 1;
7687 			lun->res_type = msg->pr.pr_info.res_type;
7688 			if (lun->res_type != SPR_TYPE_WR_EX_AR
7689 			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7690 				lun->pr_res_idx = msg->pr.pr_info.residx;
7691 		} else {
7692 		        for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7693 				if (memcmp(msg->pr.pr_info.sa_res_key,
7694 		                   lun->per_res[i].res_key.key,
7695 		                   sizeof(struct scsi_per_res_key)) != 0)
7696 					continue;
7697 
7698 				lun->per_res[i].registered = 0;
7699 				memset(&lun->per_res[i].res_key, 0,
7700 				       sizeof(struct scsi_per_res_key));
7701 				lun->pr_key_count--;
7702 
7703 				if (!persis_offset
7704 				 && i < persis_offset)
7705 					lun->pending_sense[i].ua_pending |=
7706 						CTL_UA_REG_PREEMPT;
7707 				else if (persis_offset
7708 				      && i >= persis_offset)
7709 					lun->pending_sense[i -
7710 						persis_offset].ua_pending |=
7711 						CTL_UA_REG_PREEMPT;
7712 			}
7713 		}
7714 	} else {
7715 		/*
7716 		 * Temporarily unregister so it won't get removed
7717 		 * or UA generated
7718 		 */
7719 		lun->per_res[msg->pr.pr_info.residx].registered = 0;
7720 		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7721 			if (lun->per_res[i].registered == 0)
7722 				continue;
7723 
7724 			if (memcmp(msg->pr.pr_info.sa_res_key,
7725 	                   lun->per_res[i].res_key.key,
7726 	                   sizeof(struct scsi_per_res_key)) == 0) {
7727 				lun->per_res[i].registered = 0;
7728 				memset(&lun->per_res[i].res_key, 0,
7729 				       sizeof(struct scsi_per_res_key));
7730 				lun->pr_key_count--;
7731 				if (!persis_offset
7732 				 && i < CTL_MAX_INITIATORS)
7733 					lun->pending_sense[i].ua_pending |=
7734 						CTL_UA_REG_PREEMPT;
7735 				else if (persis_offset
7736 				      && i >= persis_offset)
7737 					lun->pending_sense[i -
7738 						persis_offset].ua_pending |=
7739 						CTL_UA_REG_PREEMPT;
7740 			} else if (msg->pr.pr_info.res_type != lun->res_type
7741 				&& (lun->res_type == SPR_TYPE_WR_EX_RO
7742 				 || lun->res_type == SPR_TYPE_EX_AC_RO)) {
7743 					if (!persis_offset
7744 					 && i < persis_offset)
7745 						lun->pending_sense[i
7746 							].ua_pending |=
7747 							CTL_UA_RES_RELEASE;
7748 					else if (persis_offset
7749 					      && i >= persis_offset)
7750 					lun->pending_sense[i -
7751 						persis_offset].ua_pending |=
7752 						CTL_UA_RES_RELEASE;
7753 			}
7754 		}
7755 		lun->per_res[msg->pr.pr_info.residx].registered = 1;
7756 		lun->res_type = msg->pr.pr_info.res_type;
7757 		if (lun->res_type != SPR_TYPE_WR_EX_AR
7758 		 && lun->res_type != SPR_TYPE_EX_AC_AR)
7759 			lun->pr_res_idx = msg->pr.pr_info.residx;
7760 		else
7761 			lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
7762 	}
7763 	lun->PRGeneration++;
7764 
7765 }
7766 
7767 
7768 int
7769 ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
7770 {
7771 	int retval;
7772 	int isc_retval;
7773 	u_int32_t param_len;
7774 	struct scsi_per_res_out *cdb;
7775 	struct ctl_lun *lun;
7776 	struct scsi_per_res_out_parms* param;
7777 	struct ctl_softc *softc;
7778 	uint32_t residx;
7779 	uint64_t res_key, sa_res_key;
7780 	uint8_t type;
7781 	union ctl_ha_msg persis_io;
7782 	int    i;
7783 
7784 	CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
7785 
7786 	retval = CTL_RETVAL_COMPLETE;
7787 
7788 	softc = control_softc;
7789 
7790 	cdb = (struct scsi_per_res_out *)ctsio->cdb;
7791 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7792 
7793 	/*
7794 	 * We only support whole-LUN scope.  The scope & type are ignored for
7795 	 * register, register and ignore existing key and clear.
7796 	 * We sometimes ignore scope and type on preempts too!!
7797 	 * Verify reservation type here as well.
7798 	 */
7799 	type = cdb->scope_type & SPR_TYPE_MASK;
7800 	if ((cdb->action == SPRO_RESERVE)
7801 	 || (cdb->action == SPRO_RELEASE)) {
7802 		if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
7803 			ctl_set_invalid_field(/*ctsio*/ ctsio,
7804 					      /*sks_valid*/ 1,
7805 					      /*command*/ 1,
7806 					      /*field*/ 2,
7807 					      /*bit_valid*/ 1,
7808 					      /*bit*/ 4);
7809 			ctl_done((union ctl_io *)ctsio);
7810 			return (CTL_RETVAL_COMPLETE);
7811 		}
7812 
7813 		if (type>8 || type==2 || type==4 || type==0) {
7814 			ctl_set_invalid_field(/*ctsio*/ ctsio,
7815 					      /*sks_valid*/ 1,
7816 					      /*command*/ 1,
7817 					      /*field*/ 2,
7818 					      /*bit_valid*/ 1,
7819 					      /*bit*/ 0);
7820 			ctl_done((union ctl_io *)ctsio);
7821 			return (CTL_RETVAL_COMPLETE);
7822 		}
7823 	}
7824 
7825 	switch (cdb->action & SPRO_ACTION_MASK) {
7826 	case SPRO_REGISTER:
7827 	case SPRO_RESERVE:
7828 	case SPRO_RELEASE:
7829 	case SPRO_CLEAR:
7830 	case SPRO_PREEMPT:
7831 	case SPRO_REG_IGNO:
7832 		break;
7833 	case SPRO_REG_MOVE:
7834 	case SPRO_PRE_ABO:
7835 	default:
7836 		ctl_set_invalid_field(/*ctsio*/ ctsio,
7837 				      /*sks_valid*/ 1,
7838 				      /*command*/ 1,
7839 				      /*field*/ 1,
7840 				      /*bit_valid*/ 1,
7841 				      /*bit*/ 0);
7842 		ctl_done((union ctl_io *)ctsio);
7843 		return (CTL_RETVAL_COMPLETE);
7844 		break; /* NOTREACHED */
7845 	}
7846 
7847 	param_len = scsi_4btoul(cdb->length);
7848 
7849 	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
7850 		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
7851 		ctsio->kern_data_len = param_len;
7852 		ctsio->kern_total_len = param_len;
7853 		ctsio->kern_data_resid = 0;
7854 		ctsio->kern_rel_offset = 0;
7855 		ctsio->kern_sg_entries = 0;
7856 		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7857 		ctsio->be_move_done = ctl_config_move_done;
7858 		ctl_datamove((union ctl_io *)ctsio);
7859 
7860 		return (CTL_RETVAL_COMPLETE);
7861 	}
7862 
7863 	param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
7864 
7865 	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
7866 	res_key = scsi_8btou64(param->res_key.key);
7867 	sa_res_key = scsi_8btou64(param->serv_act_res_key);
7868 
7869 	/*
7870 	 * Validate the reservation key here except for SPRO_REG_IGNO
7871 	 * This must be done for all other service actions
7872 	 */
7873 	if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
7874 		mtx_lock(&softc->ctl_lock);
7875 		if (lun->per_res[residx].registered) {
7876 		    if (memcmp(param->res_key.key,
7877 			       lun->per_res[residx].res_key.key,
7878 			       ctl_min(sizeof(param->res_key),
7879 			       sizeof(lun->per_res[residx].res_key))) != 0) {
7880 				/*
7881 				 * The current key passed in doesn't match
7882 				 * the one the initiator previously
7883 				 * registered.
7884 				 */
7885 				mtx_unlock(&softc->ctl_lock);
7886 				free(ctsio->kern_data_ptr, M_CTL);
7887 				ctl_set_reservation_conflict(ctsio);
7888 				ctl_done((union ctl_io *)ctsio);
7889 				return (CTL_RETVAL_COMPLETE);
7890 			}
7891 		} else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
7892 			/*
7893 			 * We are not registered
7894 			 */
7895 			mtx_unlock(&softc->ctl_lock);
7896 			free(ctsio->kern_data_ptr, M_CTL);
7897 			ctl_set_reservation_conflict(ctsio);
7898 			ctl_done((union ctl_io *)ctsio);
7899 			return (CTL_RETVAL_COMPLETE);
7900 		} else if (res_key != 0) {
7901 			/*
7902 			 * We are not registered and trying to register but
7903 			 * the register key isn't zero.
7904 			 */
7905 			mtx_unlock(&softc->ctl_lock);
7906 			free(ctsio->kern_data_ptr, M_CTL);
7907 			ctl_set_reservation_conflict(ctsio);
7908 			ctl_done((union ctl_io *)ctsio);
7909 			return (CTL_RETVAL_COMPLETE);
7910 		}
7911 		mtx_unlock(&softc->ctl_lock);
7912 	}
7913 
7914 	switch (cdb->action & SPRO_ACTION_MASK) {
7915 	case SPRO_REGISTER:
7916 	case SPRO_REG_IGNO: {
7917 
7918 #if 0
7919 		printf("Registration received\n");
7920 #endif
7921 
7922 		/*
7923 		 * We don't support any of these options, as we report in
7924 		 * the read capabilities request (see
7925 		 * ctl_persistent_reserve_in(), above).
7926 		 */
7927 		if ((param->flags & SPR_SPEC_I_PT)
7928 		 || (param->flags & SPR_ALL_TG_PT)
7929 		 || (param->flags & SPR_APTPL)) {
7930 			int bit_ptr;
7931 
7932 			if (param->flags & SPR_APTPL)
7933 				bit_ptr = 0;
7934 			else if (param->flags & SPR_ALL_TG_PT)
7935 				bit_ptr = 2;
7936 			else /* SPR_SPEC_I_PT */
7937 				bit_ptr = 3;
7938 
7939 			free(ctsio->kern_data_ptr, M_CTL);
7940 			ctl_set_invalid_field(ctsio,
7941 					      /*sks_valid*/ 1,
7942 					      /*command*/ 0,
7943 					      /*field*/ 20,
7944 					      /*bit_valid*/ 1,
7945 					      /*bit*/ bit_ptr);
7946 			ctl_done((union ctl_io *)ctsio);
7947 			return (CTL_RETVAL_COMPLETE);
7948 		}
7949 
7950 		mtx_lock(&softc->ctl_lock);
7951 
7952 		/*
7953 		 * The initiator wants to clear the
7954 		 * key/unregister.
7955 		 */
7956 		if (sa_res_key == 0) {
7957 			if ((res_key == 0
7958 			  && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
7959 			 || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
7960 			  && !lun->per_res[residx].registered)) {
7961 				mtx_unlock(&softc->ctl_lock);
7962 				goto done;
7963 			}
7964 
7965 			lun->per_res[residx].registered = 0;
7966 			memset(&lun->per_res[residx].res_key,
7967 			       0, sizeof(lun->per_res[residx].res_key));
7968 			lun->pr_key_count--;
7969 
7970 			if (residx == lun->pr_res_idx) {
7971 				lun->flags &= ~CTL_LUN_PR_RESERVED;
7972 				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
7973 
7974 				if ((lun->res_type == SPR_TYPE_WR_EX_RO
7975 				  || lun->res_type == SPR_TYPE_EX_AC_RO)
7976 				 && lun->pr_key_count) {
7977 					/*
7978 					 * If the reservation is a registrants
7979 					 * only type we need to generate a UA
7980 					 * for other registered inits.  The
7981 					 * sense code should be RESERVATIONS
7982 					 * RELEASED
7983 					 */
7984 
7985 					for (i = 0; i < CTL_MAX_INITIATORS;i++){
7986 						if (lun->per_res[
7987 						    i+persis_offset].registered
7988 						    == 0)
7989 							continue;
7990 						lun->pending_sense[i
7991 							].ua_pending |=
7992 							CTL_UA_RES_RELEASE;
7993 					}
7994 				}
7995 				lun->res_type = 0;
7996 			} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7997 				if (lun->pr_key_count==0) {
7998 					lun->flags &= ~CTL_LUN_PR_RESERVED;
7999 					lun->res_type = 0;
8000 					lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8001 				}
8002 			}
8003 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8004 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8005 			persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8006 			persis_io.pr.pr_info.residx = residx;
8007 			if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8008 			     &persis_io, sizeof(persis_io), 0 )) >
8009 			     CTL_HA_STATUS_SUCCESS) {
8010 				printf("CTL:Persis Out error returned from "
8011 				       "ctl_ha_msg_send %d\n", isc_retval);
8012 			}
8013 			mtx_unlock(&softc->ctl_lock);
8014 		} else /* sa_res_key != 0 */ {
8015 
8016 			/*
8017 			 * If we aren't registered currently then increment
8018 			 * the key count and set the registered flag.
8019 			 */
8020 			if (!lun->per_res[residx].registered) {
8021 				lun->pr_key_count++;
8022 				lun->per_res[residx].registered = 1;
8023 			}
8024 
8025 			memcpy(&lun->per_res[residx].res_key,
8026 			       param->serv_act_res_key,
8027 			       ctl_min(sizeof(param->serv_act_res_key),
8028 			       sizeof(lun->per_res[residx].res_key)));
8029 
8030 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8031 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8032 			persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8033 			persis_io.pr.pr_info.residx = residx;
8034 			memcpy(persis_io.pr.pr_info.sa_res_key,
8035 			       param->serv_act_res_key,
8036 			       sizeof(param->serv_act_res_key));
8037 			mtx_unlock(&softc->ctl_lock);
8038 			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8039 			     &persis_io, sizeof(persis_io), 0)) >
8040 			     CTL_HA_STATUS_SUCCESS) {
8041 				printf("CTL:Persis Out error returned from "
8042 				       "ctl_ha_msg_send %d\n", isc_retval);
8043 			}
8044 		}
8045 		lun->PRGeneration++;
8046 
8047 		break;
8048 	}
8049 	case SPRO_RESERVE:
8050 #if 0
8051                 printf("Reserve executed type %d\n", type);
8052 #endif
8053 		mtx_lock(&softc->ctl_lock);
8054 		if (lun->flags & CTL_LUN_PR_RESERVED) {
8055 			/*
8056 			 * if this isn't the reservation holder and it's
8057 			 * not a "all registrants" type or if the type is
8058 			 * different then we have a conflict
8059 			 */
8060 			if ((lun->pr_res_idx != residx
8061 			  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8062 			 || lun->res_type != type) {
8063 				mtx_unlock(&softc->ctl_lock);
8064 				free(ctsio->kern_data_ptr, M_CTL);
8065 				ctl_set_reservation_conflict(ctsio);
8066 				ctl_done((union ctl_io *)ctsio);
8067 				return (CTL_RETVAL_COMPLETE);
8068 			}
8069 		} else /* create a reservation */ {
8070 			/*
8071 			 * If it's not an "all registrants" type record
8072 			 * reservation holder
8073 			 */
8074 			if (type != SPR_TYPE_WR_EX_AR
8075 			 && type != SPR_TYPE_EX_AC_AR)
8076 				lun->pr_res_idx = residx; /* Res holder */
8077 			else
8078 				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8079 
8080 			lun->flags |= CTL_LUN_PR_RESERVED;
8081 			lun->res_type = type;
8082 
8083 			mtx_unlock(&softc->ctl_lock);
8084 
8085 			/* send msg to other side */
8086 			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8087 			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8088 			persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8089 			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8090 			persis_io.pr.pr_info.res_type = type;
8091 			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8092 			     &persis_io, sizeof(persis_io), 0)) >
8093 			     CTL_HA_STATUS_SUCCESS) {
8094 				printf("CTL:Persis Out error returned from "
8095 				       "ctl_ha_msg_send %d\n", isc_retval);
8096 			}
8097 		}
8098 		break;
8099 
8100 	case SPRO_RELEASE:
8101 		mtx_lock(&softc->ctl_lock);
8102 		if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8103 			/* No reservation exists return good status */
8104 			mtx_unlock(&softc->ctl_lock);
8105 			goto done;
8106 		}
8107 		/*
8108 		 * Is this nexus a reservation holder?
8109 		 */
8110 		if (lun->pr_res_idx != residx
8111 		 && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8112 			/*
8113 			 * not a res holder return good status but
8114 			 * do nothing
8115 			 */
8116 			mtx_unlock(&softc->ctl_lock);
8117 			goto done;
8118 		}
8119 
8120 		if (lun->res_type != type) {
8121 			mtx_unlock(&softc->ctl_lock);
8122 			free(ctsio->kern_data_ptr, M_CTL);
8123 			ctl_set_illegal_pr_release(ctsio);
8124 			ctl_done((union ctl_io *)ctsio);
8125 			return (CTL_RETVAL_COMPLETE);
8126 		}
8127 
8128 		/* okay to release */
8129 		lun->flags &= ~CTL_LUN_PR_RESERVED;
8130 		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8131 		lun->res_type = 0;
8132 
8133 		/*
8134 		 * if this isn't an exclusive access
8135 		 * res generate UA for all other
8136 		 * registrants.
8137 		 */
8138 		if (type != SPR_TYPE_EX_AC
8139 		 && type != SPR_TYPE_WR_EX) {
8140 			/*
8141 			 * temporarily unregister so we don't generate UA
8142 			 */
8143 			lun->per_res[residx].registered = 0;
8144 
8145 			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8146 				if (lun->per_res[i+persis_offset].registered
8147 				    == 0)
8148 					continue;
8149 				lun->pending_sense[i].ua_pending |=
8150 					CTL_UA_RES_RELEASE;
8151 			}
8152 
8153 			lun->per_res[residx].registered = 1;
8154 		}
8155 		mtx_unlock(&softc->ctl_lock);
8156 		/* Send msg to other side */
8157 		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8158 		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8159 		persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8160 		if ((isc_retval=ctl_ha_msg_send( CTL_HA_CHAN_CTL, &persis_io,
8161 		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8162 			printf("CTL:Persis Out error returned from "
8163 			       "ctl_ha_msg_send %d\n", isc_retval);
8164 		}
8165 		break;
8166 
8167 	case SPRO_CLEAR:
8168 		/* send msg to other side */
8169 
8170 		mtx_lock(&softc->ctl_lock);
8171 		lun->flags &= ~CTL_LUN_PR_RESERVED;
8172 		lun->res_type = 0;
8173 		lun->pr_key_count = 0;
8174 		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8175 
8176 
8177 		memset(&lun->per_res[residx].res_key,
8178 		       0, sizeof(lun->per_res[residx].res_key));
8179 		lun->per_res[residx].registered = 0;
8180 
8181 		for (i=0; i < 2*CTL_MAX_INITIATORS; i++)
8182 			if (lun->per_res[i].registered) {
8183 				if (!persis_offset && i < CTL_MAX_INITIATORS)
8184 					lun->pending_sense[i].ua_pending |=
8185 						CTL_UA_RES_PREEMPT;
8186 				else if (persis_offset && i >= persis_offset)
8187 					lun->pending_sense[i-persis_offset
8188 					    ].ua_pending |= CTL_UA_RES_PREEMPT;
8189 
8190 				memset(&lun->per_res[i].res_key,
8191 				       0, sizeof(struct scsi_per_res_key));
8192 				lun->per_res[i].registered = 0;
8193 			}
8194 		lun->PRGeneration++;
8195 		mtx_unlock(&softc->ctl_lock);
8196 		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8197 		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8198 		persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8199 		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8200 		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8201 			printf("CTL:Persis Out error returned from "
8202 			       "ctl_ha_msg_send %d\n", isc_retval);
8203 		}
8204 		break;
8205 
8206 	case SPRO_PREEMPT: {
8207 		int nretval;
8208 
8209 		nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8210 					  residx, ctsio, cdb, param);
8211 		if (nretval != 0)
8212 			return (CTL_RETVAL_COMPLETE);
8213 		break;
8214 	}
8215 	case SPRO_REG_MOVE:
8216 	case SPRO_PRE_ABO:
8217 	default:
8218 		free(ctsio->kern_data_ptr, M_CTL);
8219 		ctl_set_invalid_field(/*ctsio*/ ctsio,
8220 				      /*sks_valid*/ 1,
8221 				      /*command*/ 1,
8222 				      /*field*/ 1,
8223 				      /*bit_valid*/ 1,
8224 				      /*bit*/ 0);
8225 		ctl_done((union ctl_io *)ctsio);
8226 		return (CTL_RETVAL_COMPLETE);
8227 		break; /* NOTREACHED */
8228 	}
8229 
8230 done:
8231 	free(ctsio->kern_data_ptr, M_CTL);
8232 	ctl_set_success(ctsio);
8233 	ctl_done((union ctl_io *)ctsio);
8234 
8235 	return (retval);
8236 }
8237 
8238 /*
8239  * This routine is for handling a message from the other SC pertaining to
8240  * persistent reserve out. All the error checking will have been done
8241  * so only perorming the action need be done here to keep the two
8242  * in sync.
8243  */
8244 static void
8245 ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8246 {
8247 	struct ctl_lun *lun;
8248 	struct ctl_softc *softc;
8249 	int i;
8250 
8251 	softc = control_softc;
8252 
8253 	mtx_lock(&softc->ctl_lock);
8254 
8255 	lun = softc->ctl_luns[msg->hdr.nexus.targ_lun];
8256 	switch(msg->pr.pr_info.action) {
8257 	case CTL_PR_REG_KEY:
8258 		if (!lun->per_res[msg->pr.pr_info.residx].registered) {
8259 			lun->per_res[msg->pr.pr_info.residx].registered = 1;
8260 			lun->pr_key_count++;
8261 		}
8262 		lun->PRGeneration++;
8263 		memcpy(&lun->per_res[msg->pr.pr_info.residx].res_key,
8264 		       msg->pr.pr_info.sa_res_key,
8265 		       sizeof(struct scsi_per_res_key));
8266 		break;
8267 
8268 	case CTL_PR_UNREG_KEY:
8269 		lun->per_res[msg->pr.pr_info.residx].registered = 0;
8270 		memset(&lun->per_res[msg->pr.pr_info.residx].res_key,
8271 		       0, sizeof(struct scsi_per_res_key));
8272 		lun->pr_key_count--;
8273 
8274 		/* XXX Need to see if the reservation has been released */
8275 		/* if so do we need to generate UA? */
8276 		if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8277 			lun->flags &= ~CTL_LUN_PR_RESERVED;
8278 			lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8279 
8280 			if ((lun->res_type == SPR_TYPE_WR_EX_RO
8281 			  || lun->res_type == SPR_TYPE_EX_AC_RO)
8282 			 && lun->pr_key_count) {
8283 				/*
8284 				 * If the reservation is a registrants
8285 				 * only type we need to generate a UA
8286 				 * for other registered inits.  The
8287 				 * sense code should be RESERVATIONS
8288 				 * RELEASED
8289 				 */
8290 
8291 				for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8292 					if (lun->per_res[i+
8293 					    persis_offset].registered == 0)
8294 						continue;
8295 
8296 					lun->pending_sense[i
8297 						].ua_pending |=
8298 						CTL_UA_RES_RELEASE;
8299 				}
8300 			}
8301 			lun->res_type = 0;
8302 		} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8303 			if (lun->pr_key_count==0) {
8304 				lun->flags &= ~CTL_LUN_PR_RESERVED;
8305 				lun->res_type = 0;
8306 				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8307 			}
8308 		}
8309 		lun->PRGeneration++;
8310 		break;
8311 
8312 	case CTL_PR_RESERVE:
8313 		lun->flags |= CTL_LUN_PR_RESERVED;
8314 		lun->res_type = msg->pr.pr_info.res_type;
8315 		lun->pr_res_idx = msg->pr.pr_info.residx;
8316 
8317 		break;
8318 
8319 	case CTL_PR_RELEASE:
8320 		/*
8321 		 * if this isn't an exclusive access res generate UA for all
8322 		 * other registrants.
8323 		 */
8324 		if (lun->res_type != SPR_TYPE_EX_AC
8325 		 && lun->res_type != SPR_TYPE_WR_EX) {
8326 			for (i = 0; i < CTL_MAX_INITIATORS; i++)
8327 				if (lun->per_res[i+persis_offset].registered)
8328 					lun->pending_sense[i].ua_pending |=
8329 						CTL_UA_RES_RELEASE;
8330 		}
8331 
8332 		lun->flags &= ~CTL_LUN_PR_RESERVED;
8333 		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8334 		lun->res_type = 0;
8335 		break;
8336 
8337 	case CTL_PR_PREEMPT:
8338 		ctl_pro_preempt_other(lun, msg);
8339 		break;
8340 	case CTL_PR_CLEAR:
8341 		lun->flags &= ~CTL_LUN_PR_RESERVED;
8342 		lun->res_type = 0;
8343 		lun->pr_key_count = 0;
8344 		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8345 
8346 		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8347 			if (lun->per_res[i].registered == 0)
8348 				continue;
8349 			if (!persis_offset
8350 			 && i < CTL_MAX_INITIATORS)
8351 				lun->pending_sense[i].ua_pending |=
8352 					CTL_UA_RES_PREEMPT;
8353 			else if (persis_offset
8354 			      && i >= persis_offset)
8355    				lun->pending_sense[i-persis_offset].ua_pending|=
8356 					CTL_UA_RES_PREEMPT;
8357 			memset(&lun->per_res[i].res_key, 0,
8358 			       sizeof(struct scsi_per_res_key));
8359 			lun->per_res[i].registered = 0;
8360 		}
8361 		lun->PRGeneration++;
8362 		break;
8363 	}
8364 
8365 	mtx_unlock(&softc->ctl_lock);
8366 }
8367 
8368 int
8369 ctl_read_write(struct ctl_scsiio *ctsio)
8370 {
8371 	struct ctl_lun *lun;
8372 	struct ctl_lba_len lbalen;
8373 	uint64_t lba;
8374 	uint32_t num_blocks;
8375 	int reladdr, fua, dpo, ebp;
8376 	int retval;
8377 	int isread;
8378 
8379 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8380 
8381 	CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8382 
8383 	reladdr = 0;
8384 	fua = 0;
8385 	dpo = 0;
8386 	ebp = 0;
8387 
8388 	retval = CTL_RETVAL_COMPLETE;
8389 
8390 	isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8391 	      || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8392 	if (lun->flags & CTL_LUN_PR_RESERVED && isread) {
8393 		uint32_t residx;
8394 
8395 		/*
8396 		 * XXX KDM need a lock here.
8397 		 */
8398 		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8399 		if ((lun->res_type == SPR_TYPE_EX_AC
8400 		  && residx != lun->pr_res_idx)
8401 		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
8402 		   || lun->res_type == SPR_TYPE_EX_AC_AR)
8403 		  && !lun->per_res[residx].registered)) {
8404 			ctl_set_reservation_conflict(ctsio);
8405 			ctl_done((union ctl_io *)ctsio);
8406 			return (CTL_RETVAL_COMPLETE);
8407 	        }
8408 	}
8409 
8410 	switch (ctsio->cdb[0]) {
8411 	case READ_6:
8412 	case WRITE_6: {
8413 		struct scsi_rw_6 *cdb;
8414 
8415 		cdb = (struct scsi_rw_6 *)ctsio->cdb;
8416 
8417 		lba = scsi_3btoul(cdb->addr);
8418 		/* only 5 bits are valid in the most significant address byte */
8419 		lba &= 0x1fffff;
8420 		num_blocks = cdb->length;
8421 		/*
8422 		 * This is correct according to SBC-2.
8423 		 */
8424 		if (num_blocks == 0)
8425 			num_blocks = 256;
8426 		break;
8427 	}
8428 	case READ_10:
8429 	case WRITE_10: {
8430 		struct scsi_rw_10 *cdb;
8431 
8432 		cdb = (struct scsi_rw_10 *)ctsio->cdb;
8433 
8434 		if (cdb->byte2 & SRW10_RELADDR)
8435 			reladdr = 1;
8436 		if (cdb->byte2 & SRW10_FUA)
8437 			fua = 1;
8438 		if (cdb->byte2 & SRW10_DPO)
8439 			dpo = 1;
8440 
8441 		if ((cdb->opcode == WRITE_10)
8442 		 && (cdb->byte2 & SRW10_EBP))
8443 			ebp = 1;
8444 
8445 		lba = scsi_4btoul(cdb->addr);
8446 		num_blocks = scsi_2btoul(cdb->length);
8447 		break;
8448 	}
8449 	case WRITE_VERIFY_10: {
8450 		struct scsi_write_verify_10 *cdb;
8451 
8452 		cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8453 
8454 		/*
8455 		 * XXX KDM we should do actual write verify support at some
8456 		 * point.  This is obviously fake, we're just translating
8457 		 * things to a write.  So we don't even bother checking the
8458 		 * BYTCHK field, since we don't do any verification.  If
8459 		 * the user asks for it, we'll just pretend we did it.
8460 		 */
8461 		if (cdb->byte2 & SWV_DPO)
8462 			dpo = 1;
8463 
8464 		lba = scsi_4btoul(cdb->addr);
8465 		num_blocks = scsi_2btoul(cdb->length);
8466 		break;
8467 	}
8468 	case READ_12:
8469 	case WRITE_12: {
8470 		struct scsi_rw_12 *cdb;
8471 
8472 		cdb = (struct scsi_rw_12 *)ctsio->cdb;
8473 
8474 		if (cdb->byte2 & SRW12_RELADDR)
8475 			reladdr = 1;
8476 		if (cdb->byte2 & SRW12_FUA)
8477 			fua = 1;
8478 		if (cdb->byte2 & SRW12_DPO)
8479 			dpo = 1;
8480 		lba = scsi_4btoul(cdb->addr);
8481 		num_blocks = scsi_4btoul(cdb->length);
8482 		break;
8483 	}
8484 	case WRITE_VERIFY_12: {
8485 		struct scsi_write_verify_12 *cdb;
8486 
8487 		cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8488 
8489 		if (cdb->byte2 & SWV_DPO)
8490 			dpo = 1;
8491 
8492 		lba = scsi_4btoul(cdb->addr);
8493 		num_blocks = scsi_4btoul(cdb->length);
8494 
8495 		break;
8496 	}
8497 	case READ_16:
8498 	case WRITE_16: {
8499 		struct scsi_rw_16 *cdb;
8500 
8501 		cdb = (struct scsi_rw_16 *)ctsio->cdb;
8502 
8503 		if (cdb->byte2 & SRW12_RELADDR)
8504 			reladdr = 1;
8505 		if (cdb->byte2 & SRW12_FUA)
8506 			fua = 1;
8507 		if (cdb->byte2 & SRW12_DPO)
8508 			dpo = 1;
8509 
8510 		lba = scsi_8btou64(cdb->addr);
8511 		num_blocks = scsi_4btoul(cdb->length);
8512 		break;
8513 	}
8514 	case WRITE_VERIFY_16: {
8515 		struct scsi_write_verify_16 *cdb;
8516 
8517 		cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
8518 
8519 		if (cdb->byte2 & SWV_DPO)
8520 			dpo = 1;
8521 
8522 		lba = scsi_8btou64(cdb->addr);
8523 		num_blocks = scsi_4btoul(cdb->length);
8524 		break;
8525 	}
8526 	default:
8527 		/*
8528 		 * We got a command we don't support.  This shouldn't
8529 		 * happen, commands should be filtered out above us.
8530 		 */
8531 		ctl_set_invalid_opcode(ctsio);
8532 		ctl_done((union ctl_io *)ctsio);
8533 
8534 		return (CTL_RETVAL_COMPLETE);
8535 		break; /* NOTREACHED */
8536 	}
8537 
8538 	/*
8539 	 * XXX KDM what do we do with the DPO and FUA bits?  FUA might be
8540 	 * interesting for us, but if RAIDCore is in write-back mode,
8541 	 * getting it to do write-through for a particular transaction may
8542 	 * not be possible.
8543 	 */
8544 	/*
8545 	 * We don't support relative addressing.  That also requires
8546 	 * supporting linked commands, which we don't do.
8547 	 */
8548 	if (reladdr != 0) {
8549 		ctl_set_invalid_field(ctsio,
8550 				      /*sks_valid*/ 1,
8551 				      /*command*/ 1,
8552 				      /*field*/ 1,
8553 				      /*bit_valid*/ 1,
8554 				      /*bit*/ 0);
8555 		ctl_done((union ctl_io *)ctsio);
8556 		return (CTL_RETVAL_COMPLETE);
8557 	}
8558 
8559 	/*
8560 	 * The first check is to make sure we're in bounds, the second
8561 	 * check is to catch wrap-around problems.  If the lba + num blocks
8562 	 * is less than the lba, then we've wrapped around and the block
8563 	 * range is invalid anyway.
8564 	 */
8565 	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8566 	 || ((lba + num_blocks) < lba)) {
8567 		ctl_set_lba_out_of_range(ctsio);
8568 		ctl_done((union ctl_io *)ctsio);
8569 		return (CTL_RETVAL_COMPLETE);
8570 	}
8571 
8572 	/*
8573 	 * According to SBC-3, a transfer length of 0 is not an error.
8574 	 * Note that this cannot happen with WRITE(6) or READ(6), since 0
8575 	 * translates to 256 blocks for those commands.
8576 	 */
8577 	if (num_blocks == 0) {
8578 		ctl_set_success(ctsio);
8579 		ctl_done((union ctl_io *)ctsio);
8580 		return (CTL_RETVAL_COMPLETE);
8581 	}
8582 
8583 	lbalen.lba = lba;
8584 	lbalen.len = num_blocks;
8585 	memcpy(ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes, &lbalen,
8586 	       sizeof(lbalen));
8587 
8588 	CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
8589 
8590 	retval = lun->backend->data_submit((union ctl_io *)ctsio);
8591 
8592 	return (retval);
8593 }
8594 
8595 int
8596 ctl_report_luns(struct ctl_scsiio *ctsio)
8597 {
8598 	struct scsi_report_luns *cdb;
8599 	struct scsi_report_luns_data *lun_data;
8600 	struct ctl_lun *lun, *request_lun;
8601 	int num_luns, retval;
8602 	uint32_t alloc_len, lun_datalen;
8603 	int num_filled, well_known;
8604 	uint32_t initidx;
8605 
8606 	retval = CTL_RETVAL_COMPLETE;
8607 	well_known = 0;
8608 
8609 	cdb = (struct scsi_report_luns *)ctsio->cdb;
8610 
8611 	CTL_DEBUG_PRINT(("ctl_report_luns\n"));
8612 
8613 	mtx_lock(&control_softc->ctl_lock);
8614 	num_luns = control_softc->num_luns;
8615 	mtx_unlock(&control_softc->ctl_lock);
8616 
8617 	switch (cdb->select_report) {
8618 	case RPL_REPORT_DEFAULT:
8619 	case RPL_REPORT_ALL:
8620 		break;
8621 	case RPL_REPORT_WELLKNOWN:
8622 		well_known = 1;
8623 		num_luns = 0;
8624 		break;
8625 	default:
8626 		ctl_set_invalid_field(ctsio,
8627 				      /*sks_valid*/ 1,
8628 				      /*command*/ 1,
8629 				      /*field*/ 2,
8630 				      /*bit_valid*/ 0,
8631 				      /*bit*/ 0);
8632 		ctl_done((union ctl_io *)ctsio);
8633 		return (retval);
8634 		break; /* NOTREACHED */
8635 	}
8636 
8637 	alloc_len = scsi_4btoul(cdb->length);
8638 	/*
8639 	 * The initiator has to allocate at least 16 bytes for this request,
8640 	 * so he can at least get the header and the first LUN.  Otherwise
8641 	 * we reject the request (per SPC-3 rev 14, section 6.21).
8642 	 */
8643 	if (alloc_len < (sizeof(struct scsi_report_luns_data) +
8644 	    sizeof(struct scsi_report_luns_lundata))) {
8645 		ctl_set_invalid_field(ctsio,
8646 				      /*sks_valid*/ 1,
8647 				      /*command*/ 1,
8648 				      /*field*/ 6,
8649 				      /*bit_valid*/ 0,
8650 				      /*bit*/ 0);
8651 		ctl_done((union ctl_io *)ctsio);
8652 		return (retval);
8653 	}
8654 
8655 	request_lun = (struct ctl_lun *)
8656 		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8657 
8658 	lun_datalen = sizeof(*lun_data) +
8659 		(num_luns * sizeof(struct scsi_report_luns_lundata));
8660 
8661 	ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
8662 	lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
8663 	ctsio->kern_sg_entries = 0;
8664 
8665 	if (lun_datalen < alloc_len) {
8666 		ctsio->residual = alloc_len - lun_datalen;
8667 		ctsio->kern_data_len = lun_datalen;
8668 		ctsio->kern_total_len = lun_datalen;
8669 	} else {
8670 		ctsio->residual = 0;
8671 		ctsio->kern_data_len = alloc_len;
8672 		ctsio->kern_total_len = alloc_len;
8673 	}
8674 	ctsio->kern_data_resid = 0;
8675 	ctsio->kern_rel_offset = 0;
8676 	ctsio->kern_sg_entries = 0;
8677 
8678 	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
8679 
8680 	/*
8681 	 * We set this to the actual data length, regardless of how much
8682 	 * space we actually have to return results.  If the user looks at
8683 	 * this value, he'll know whether or not he allocated enough space
8684 	 * and reissue the command if necessary.  We don't support well
8685 	 * known logical units, so if the user asks for that, return none.
8686 	 */
8687 	scsi_ulto4b(lun_datalen - 8, lun_data->length);
8688 
8689 	mtx_lock(&control_softc->ctl_lock);
8690 	for (num_filled = 0, lun = STAILQ_FIRST(&control_softc->lun_list);
8691 	     (lun != NULL) && (num_filled < num_luns);
8692 	     lun = STAILQ_NEXT(lun, links)) {
8693 
8694 		if (lun->lun <= 0xff) {
8695 			/*
8696 			 * Peripheral addressing method, bus number 0.
8697 			 */
8698 			lun_data->luns[num_filled].lundata[0] =
8699 				RPL_LUNDATA_ATYP_PERIPH;
8700 			lun_data->luns[num_filled].lundata[1] = lun->lun;
8701 			num_filled++;
8702 		} else if (lun->lun <= 0x3fff) {
8703 			/*
8704 			 * Flat addressing method.
8705 			 */
8706 			lun_data->luns[num_filled].lundata[0] =
8707 				RPL_LUNDATA_ATYP_FLAT |
8708 				(lun->lun & RPL_LUNDATA_FLAT_LUN_MASK);
8709 #ifdef OLDCTLHEADERS
8710 				(SRLD_ADDR_FLAT << SRLD_ADDR_SHIFT) |
8711 				(lun->lun & SRLD_BUS_LUN_MASK);
8712 #endif
8713 			lun_data->luns[num_filled].lundata[1] =
8714 #ifdef OLDCTLHEADERS
8715 				lun->lun >> SRLD_BUS_LUN_BITS;
8716 #endif
8717 				lun->lun >> RPL_LUNDATA_FLAT_LUN_BITS;
8718 			num_filled++;
8719 		} else {
8720 			printf("ctl_report_luns: bogus LUN number %jd, "
8721 			       "skipping\n", (intmax_t)lun->lun);
8722 		}
8723 		/*
8724 		 * According to SPC-3, rev 14 section 6.21:
8725 		 *
8726 		 * "The execution of a REPORT LUNS command to any valid and
8727 		 * installed logical unit shall clear the REPORTED LUNS DATA
8728 		 * HAS CHANGED unit attention condition for all logical
8729 		 * units of that target with respect to the requesting
8730 		 * initiator. A valid and installed logical unit is one
8731 		 * having a PERIPHERAL QUALIFIER of 000b in the standard
8732 		 * INQUIRY data (see 6.4.2)."
8733 		 *
8734 		 * If request_lun is NULL, the LUN this report luns command
8735 		 * was issued to is either disabled or doesn't exist. In that
8736 		 * case, we shouldn't clear any pending lun change unit
8737 		 * attention.
8738 		 */
8739 		if (request_lun != NULL)
8740 			lun->pending_sense[initidx].ua_pending &=
8741 				~CTL_UA_LUN_CHANGE;
8742 	}
8743 	mtx_unlock(&control_softc->ctl_lock);
8744 
8745 	/*
8746 	 * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
8747 	 * this request.
8748 	 */
8749 	ctsio->scsi_status = SCSI_STATUS_OK;
8750 
8751 	ctsio->be_move_done = ctl_config_move_done;
8752 	ctl_datamove((union ctl_io *)ctsio);
8753 
8754 	return (retval);
8755 }
8756 
8757 int
8758 ctl_request_sense(struct ctl_scsiio *ctsio)
8759 {
8760 	struct scsi_request_sense *cdb;
8761 	struct scsi_sense_data *sense_ptr;
8762 	struct ctl_lun *lun;
8763 	uint32_t initidx;
8764 	int have_error;
8765 	scsi_sense_data_type sense_format;
8766 
8767 	cdb = (struct scsi_request_sense *)ctsio->cdb;
8768 
8769 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8770 
8771 	CTL_DEBUG_PRINT(("ctl_request_sense\n"));
8772 
8773 	/*
8774 	 * Determine which sense format the user wants.
8775 	 */
8776 	if (cdb->byte2 & SRS_DESC)
8777 		sense_format = SSD_TYPE_DESC;
8778 	else
8779 		sense_format = SSD_TYPE_FIXED;
8780 
8781 	ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
8782 	sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
8783 	ctsio->kern_sg_entries = 0;
8784 
8785 	/*
8786 	 * struct scsi_sense_data, which is currently set to 256 bytes, is
8787 	 * larger than the largest allowed value for the length field in the
8788 	 * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
8789 	 */
8790 	ctsio->residual = 0;
8791 	ctsio->kern_data_len = cdb->length;
8792 	ctsio->kern_total_len = cdb->length;
8793 
8794 	ctsio->kern_data_resid = 0;
8795 	ctsio->kern_rel_offset = 0;
8796 	ctsio->kern_sg_entries = 0;
8797 
8798 	/*
8799 	 * If we don't have a LUN, we don't have any pending sense.
8800 	 */
8801 	if (lun == NULL)
8802 		goto no_sense;
8803 
8804 	have_error = 0;
8805 	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
8806 	/*
8807 	 * Check for pending sense, and then for pending unit attentions.
8808 	 * Pending sense gets returned first, then pending unit attentions.
8809 	 */
8810 	mtx_lock(&lun->ctl_softc->ctl_lock);
8811 	if (ctl_is_set(lun->have_ca, initidx)) {
8812 		scsi_sense_data_type stored_format;
8813 
8814 		/*
8815 		 * Check to see which sense format was used for the stored
8816 		 * sense data.
8817 		 */
8818 		stored_format = scsi_sense_type(
8819 		    &lun->pending_sense[initidx].sense);
8820 
8821 		/*
8822 		 * If the user requested a different sense format than the
8823 		 * one we stored, then we need to convert it to the other
8824 		 * format.  If we're going from descriptor to fixed format
8825 		 * sense data, we may lose things in translation, depending
8826 		 * on what options were used.
8827 		 *
8828 		 * If the stored format is SSD_TYPE_NONE (i.e. invalid),
8829 		 * for some reason we'll just copy it out as-is.
8830 		 */
8831 		if ((stored_format == SSD_TYPE_FIXED)
8832 		 && (sense_format == SSD_TYPE_DESC))
8833 			ctl_sense_to_desc((struct scsi_sense_data_fixed *)
8834 			    &lun->pending_sense[initidx].sense,
8835 			    (struct scsi_sense_data_desc *)sense_ptr);
8836 		else if ((stored_format == SSD_TYPE_DESC)
8837 		      && (sense_format == SSD_TYPE_FIXED))
8838 			ctl_sense_to_fixed((struct scsi_sense_data_desc *)
8839 			    &lun->pending_sense[initidx].sense,
8840 			    (struct scsi_sense_data_fixed *)sense_ptr);
8841 		else
8842 			memcpy(sense_ptr, &lun->pending_sense[initidx].sense,
8843 			       ctl_min(sizeof(*sense_ptr),
8844 			       sizeof(lun->pending_sense[initidx].sense)));
8845 
8846 		ctl_clear_mask(lun->have_ca, initidx);
8847 		have_error = 1;
8848 	} else if (lun->pending_sense[initidx].ua_pending != CTL_UA_NONE) {
8849 		ctl_ua_type ua_type;
8850 
8851 		ua_type = ctl_build_ua(lun->pending_sense[initidx].ua_pending,
8852 				       sense_ptr, sense_format);
8853 		if (ua_type != CTL_UA_NONE) {
8854 			have_error = 1;
8855 			/* We're reporting this UA, so clear it */
8856 			lun->pending_sense[initidx].ua_pending &= ~ua_type;
8857 		}
8858 	}
8859 	mtx_unlock(&lun->ctl_softc->ctl_lock);
8860 
8861 	/*
8862 	 * We already have a pending error, return it.
8863 	 */
8864 	if (have_error != 0) {
8865 		/*
8866 		 * We report the SCSI status as OK, since the status of the
8867 		 * request sense command itself is OK.
8868 		 */
8869 		ctsio->scsi_status = SCSI_STATUS_OK;
8870 
8871 		/*
8872 		 * We report 0 for the sense length, because we aren't doing
8873 		 * autosense in this case.  We're reporting sense as
8874 		 * parameter data.
8875 		 */
8876 		ctsio->sense_len = 0;
8877 
8878 		ctsio->be_move_done = ctl_config_move_done;
8879 		ctl_datamove((union ctl_io *)ctsio);
8880 
8881 		return (CTL_RETVAL_COMPLETE);
8882 	}
8883 
8884 no_sense:
8885 
8886 	/*
8887 	 * No sense information to report, so we report that everything is
8888 	 * okay.
8889 	 */
8890 	ctl_set_sense_data(sense_ptr,
8891 			   lun,
8892 			   sense_format,
8893 			   /*current_error*/ 1,
8894 			   /*sense_key*/ SSD_KEY_NO_SENSE,
8895 			   /*asc*/ 0x00,
8896 			   /*ascq*/ 0x00,
8897 			   SSD_ELEM_NONE);
8898 
8899 	ctsio->scsi_status = SCSI_STATUS_OK;
8900 
8901 	/*
8902 	 * We report 0 for the sense length, because we aren't doing
8903 	 * autosense in this case.  We're reporting sense as parameter data.
8904 	 */
8905 	ctsio->sense_len = 0;
8906 	ctsio->be_move_done = ctl_config_move_done;
8907 	ctl_datamove((union ctl_io *)ctsio);
8908 
8909 	return (CTL_RETVAL_COMPLETE);
8910 }
8911 
8912 int
8913 ctl_tur(struct ctl_scsiio *ctsio)
8914 {
8915 	struct ctl_lun *lun;
8916 
8917 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8918 
8919 	CTL_DEBUG_PRINT(("ctl_tur\n"));
8920 
8921 	if (lun == NULL)
8922 		return (-EINVAL);
8923 
8924 	ctsio->scsi_status = SCSI_STATUS_OK;
8925 	ctsio->io_hdr.status = CTL_SUCCESS;
8926 
8927 	ctl_done((union ctl_io *)ctsio);
8928 
8929 	return (CTL_RETVAL_COMPLETE);
8930 }
8931 
8932 #ifdef notyet
8933 static int
8934 ctl_cmddt_inquiry(struct ctl_scsiio *ctsio)
8935 {
8936 
8937 }
8938 #endif
8939 
8940 static int
8941 ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
8942 {
8943 	struct scsi_vpd_supported_pages *pages;
8944 	int sup_page_size;
8945 	struct ctl_lun *lun;
8946 
8947 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8948 
8949 	sup_page_size = sizeof(struct scsi_vpd_supported_pages) +
8950 		SCSI_EVPD_NUM_SUPPORTED_PAGES;
8951 	ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
8952 	pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
8953 	ctsio->kern_sg_entries = 0;
8954 
8955 	if (sup_page_size < alloc_len) {
8956 		ctsio->residual = alloc_len - sup_page_size;
8957 		ctsio->kern_data_len = sup_page_size;
8958 		ctsio->kern_total_len = sup_page_size;
8959 	} else {
8960 		ctsio->residual = 0;
8961 		ctsio->kern_data_len = alloc_len;
8962 		ctsio->kern_total_len = alloc_len;
8963 	}
8964 	ctsio->kern_data_resid = 0;
8965 	ctsio->kern_rel_offset = 0;
8966 	ctsio->kern_sg_entries = 0;
8967 
8968 	/*
8969 	 * The control device is always connected.  The disk device, on the
8970 	 * other hand, may not be online all the time.  Need to change this
8971 	 * to figure out whether the disk device is actually online or not.
8972 	 */
8973 	if (lun != NULL)
8974 		pages->device = (SID_QUAL_LU_CONNECTED << 5) |
8975 				lun->be_lun->lun_type;
8976 	else
8977 		pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
8978 
8979 	pages->length = SCSI_EVPD_NUM_SUPPORTED_PAGES;
8980 	/* Supported VPD pages */
8981 	pages->page_list[0] = SVPD_SUPPORTED_PAGES;
8982 	/* Serial Number */
8983 	pages->page_list[1] = SVPD_UNIT_SERIAL_NUMBER;
8984 	/* Device Identification */
8985 	pages->page_list[2] = SVPD_DEVICE_ID;
8986 
8987 	ctsio->scsi_status = SCSI_STATUS_OK;
8988 
8989 	ctsio->be_move_done = ctl_config_move_done;
8990 	ctl_datamove((union ctl_io *)ctsio);
8991 
8992 	return (CTL_RETVAL_COMPLETE);
8993 }
8994 
8995 static int
8996 ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
8997 {
8998 	struct scsi_vpd_unit_serial_number *sn_ptr;
8999 	struct ctl_lun *lun;
9000 #ifndef CTL_USE_BACKEND_SN
9001 	char tmpstr[32];
9002 #endif
9003 
9004 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9005 
9006 	ctsio->kern_data_ptr = malloc(sizeof(*sn_ptr), M_CTL, M_WAITOK | M_ZERO);
9007 	sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9008 	ctsio->kern_sg_entries = 0;
9009 
9010 	if (sizeof(*sn_ptr) < alloc_len) {
9011 		ctsio->residual = alloc_len - sizeof(*sn_ptr);
9012 		ctsio->kern_data_len = sizeof(*sn_ptr);
9013 		ctsio->kern_total_len = sizeof(*sn_ptr);
9014 	} else {
9015 		ctsio->residual = 0;
9016 		ctsio->kern_data_len = alloc_len;
9017 		ctsio->kern_total_len = alloc_len;
9018 	}
9019 	ctsio->kern_data_resid = 0;
9020 	ctsio->kern_rel_offset = 0;
9021 	ctsio->kern_sg_entries = 0;
9022 
9023 	/*
9024 	 * The control device is always connected.  The disk device, on the
9025 	 * other hand, may not be online all the time.  Need to change this
9026 	 * to figure out whether the disk device is actually online or not.
9027 	 */
9028 	if (lun != NULL)
9029 		sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9030 				  lun->be_lun->lun_type;
9031 	else
9032 		sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9033 
9034 	sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9035 	sn_ptr->length = ctl_min(sizeof(*sn_ptr) - 4, CTL_SN_LEN);
9036 #ifdef CTL_USE_BACKEND_SN
9037 	/*
9038 	 * If we don't have a LUN, we just leave the serial number as
9039 	 * all spaces.
9040 	 */
9041 	memset(sn_ptr->serial_num, 0x20, sizeof(sn_ptr->serial_num));
9042 	if (lun != NULL) {
9043 		strncpy((char *)sn_ptr->serial_num,
9044 			(char *)lun->be_lun->serial_num, CTL_SN_LEN);
9045 	}
9046 #else
9047 	/*
9048 	 * Note that we're using a non-unique serial number here,
9049 	 */
9050 	snprintf(tmpstr, sizeof(tmpstr), "MYSERIALNUMIS000");
9051 	memset(sn_ptr->serial_num, 0x20, sizeof(sn_ptr->serial_num));
9052 	strncpy(sn_ptr->serial_num, tmpstr, ctl_min(CTL_SN_LEN,
9053 		ctl_min(sizeof(tmpstr), sizeof(*sn_ptr) - 4)));
9054 #endif
9055 	ctsio->scsi_status = SCSI_STATUS_OK;
9056 
9057 	ctsio->be_move_done = ctl_config_move_done;
9058 	ctl_datamove((union ctl_io *)ctsio);
9059 
9060 	return (CTL_RETVAL_COMPLETE);
9061 }
9062 
9063 
9064 static int
9065 ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9066 {
9067 	struct scsi_vpd_device_id *devid_ptr;
9068 	struct scsi_vpd_id_descriptor *desc, *desc1;
9069 	struct scsi_vpd_id_descriptor *desc2, *desc3; /* for types 4h and 5h */
9070 	struct scsi_vpd_id_t10 *t10id;
9071 	struct ctl_softc *ctl_softc;
9072 	struct ctl_lun *lun;
9073 	struct ctl_frontend *fe;
9074 #ifndef CTL_USE_BACKEND_SN
9075 	char tmpstr[32];
9076 #endif /* CTL_USE_BACKEND_SN */
9077 	int devid_len;
9078 
9079 	ctl_softc = control_softc;
9080 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9081 
9082 	devid_len = sizeof(struct scsi_vpd_device_id) +
9083 		sizeof(struct scsi_vpd_id_descriptor) +
9084 		sizeof(struct scsi_vpd_id_t10) + CTL_DEVID_LEN +
9085 		sizeof(struct scsi_vpd_id_descriptor) + CTL_WWPN_LEN +
9086 		sizeof(struct scsi_vpd_id_descriptor) +
9087 		sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9088 		sizeof(struct scsi_vpd_id_descriptor) +
9089 		sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9090 
9091 	ctsio->kern_data_ptr = malloc(devid_len, M_CTL, M_WAITOK | M_ZERO);
9092 	devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9093 	ctsio->kern_sg_entries = 0;
9094 
9095 	if (devid_len < alloc_len) {
9096 		ctsio->residual = alloc_len - devid_len;
9097 		ctsio->kern_data_len = devid_len;
9098 		ctsio->kern_total_len = devid_len;
9099 	} else {
9100 		ctsio->residual = 0;
9101 		ctsio->kern_data_len = alloc_len;
9102 		ctsio->kern_total_len = alloc_len;
9103 	}
9104 	ctsio->kern_data_resid = 0;
9105 	ctsio->kern_rel_offset = 0;
9106 	ctsio->kern_sg_entries = 0;
9107 
9108 	desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9109 	t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
9110 	desc1 = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9111 		sizeof(struct scsi_vpd_id_t10) + CTL_DEVID_LEN);
9112 	desc2 = (struct scsi_vpd_id_descriptor *)(&desc1->identifier[0] +
9113 	          CTL_WWPN_LEN);
9114 	desc3 = (struct scsi_vpd_id_descriptor *)(&desc2->identifier[0] +
9115 	         sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9116 
9117 	/*
9118 	 * The control device is always connected.  The disk device, on the
9119 	 * other hand, may not be online all the time.
9120 	 */
9121 	if (lun != NULL)
9122 		devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9123 				     lun->be_lun->lun_type;
9124 	else
9125 		devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9126 
9127 	devid_ptr->page_code = SVPD_DEVICE_ID;
9128 
9129 	scsi_ulto2b(devid_len - 4, devid_ptr->length);
9130 
9131 	mtx_lock(&ctl_softc->ctl_lock);
9132 
9133 	fe = ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)];
9134 
9135 	/*
9136 	 * For Fibre channel,
9137 	 */
9138 	if (fe->port_type == CTL_PORT_FC)
9139 	{
9140 		desc->proto_codeset = (SCSI_PROTO_FC << 4) |
9141 				      SVPD_ID_CODESET_ASCII;
9142         	desc1->proto_codeset = (SCSI_PROTO_FC << 4) |
9143 		              SVPD_ID_CODESET_BINARY;
9144 	}
9145 	else
9146 	{
9147 		desc->proto_codeset = (SCSI_PROTO_SPI << 4) |
9148 				      SVPD_ID_CODESET_ASCII;
9149         	desc1->proto_codeset = (SCSI_PROTO_SPI << 4) |
9150 		              SVPD_ID_CODESET_BINARY;
9151 	}
9152 	desc2->proto_codeset = desc3->proto_codeset = desc1->proto_codeset;
9153 	mtx_unlock(&ctl_softc->ctl_lock);
9154 
9155 	/*
9156 	 * We're using a LUN association here.  i.e., this device ID is a
9157 	 * per-LUN identifier.
9158 	 */
9159 	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
9160 	desc->length = sizeof(*t10id) + CTL_DEVID_LEN;
9161 	strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
9162 
9163 	/*
9164 	 * desc1 is for the WWPN which is a port asscociation.
9165 	 */
9166 	desc1->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT | SVPD_ID_TYPE_NAA;
9167 	desc1->length = CTL_WWPN_LEN;
9168 	/* XXX Call Reggie's get_WWNN func here then add port # to the end */
9169 	/* For testing just create the WWPN */
9170 #if 0
9171 	ddb_GetWWNN((char *)desc1->identifier);
9172 
9173 	/* NOTE: if the port is 0 or 8 we don't want to subtract 1 */
9174 	/* This is so Copancontrol will return something sane */
9175 	if (ctsio->io_hdr.nexus.targ_port!=0 &&
9176 	    ctsio->io_hdr.nexus.targ_port!=8)
9177 		desc1->identifier[7] += ctsio->io_hdr.nexus.targ_port-1;
9178 	else
9179 		desc1->identifier[7] += ctsio->io_hdr.nexus.targ_port;
9180 #endif
9181 
9182 	be64enc(desc1->identifier, fe->wwpn);
9183 
9184 	/*
9185 	 * desc2 is for the Relative Target Port(type 4h) identifier
9186 	 */
9187 	desc2->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT
9188 	                 | SVPD_ID_TYPE_RELTARG;
9189 	desc2->length = 4;
9190 //#if 0
9191 	/* NOTE: if the port is 0 or 8 we don't want to subtract 1 */
9192 	/* This is so Copancontrol will return something sane */
9193 	if (ctsio->io_hdr.nexus.targ_port!=0 &&
9194 	    ctsio->io_hdr.nexus.targ_port!=8)
9195 		desc2->identifier[3] = ctsio->io_hdr.nexus.targ_port - 1;
9196 	else
9197 	        desc2->identifier[3] = ctsio->io_hdr.nexus.targ_port;
9198 //#endif
9199 
9200 	/*
9201 	 * desc3 is for the Target Port Group(type 5h) identifier
9202 	 */
9203 	desc3->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT
9204 	                 | SVPD_ID_TYPE_TPORTGRP;
9205 	desc3->length = 4;
9206 	if (ctsio->io_hdr.nexus.targ_port < CTL_MAX_PORTS || ctl_is_single)
9207 		desc3->identifier[3] = 1;
9208 	else
9209 		desc3->identifier[3] = 2;
9210 
9211 #ifdef CTL_USE_BACKEND_SN
9212 	/*
9213 	 * If we've actually got a backend, copy the device id from the
9214 	 * per-LUN data.  Otherwise, set it to all spaces.
9215 	 */
9216 	if (lun != NULL) {
9217 		/*
9218 		 * Copy the backend's LUN ID.
9219 		 */
9220 		strncpy((char *)t10id->vendor_spec_id,
9221 			(char *)lun->be_lun->device_id, CTL_DEVID_LEN);
9222 	} else {
9223 		/*
9224 		 * No backend, set this to spaces.
9225 		 */
9226 		memset(t10id->vendor_spec_id, 0x20, CTL_DEVID_LEN);
9227 	}
9228 #else
9229 	snprintf(tmpstr, sizeof(tmpstr), "MYDEVICEIDIS%4d",
9230 		 (lun != NULL) ?  (int)lun->lun : 0);
9231 	strncpy(t10id->vendor_spec_id, tmpstr, ctl_min(CTL_DEVID_LEN,
9232 		sizeof(tmpstr)));
9233 #endif
9234 
9235 	ctsio->scsi_status = SCSI_STATUS_OK;
9236 
9237 	ctsio->be_move_done = ctl_config_move_done;
9238 	ctl_datamove((union ctl_io *)ctsio);
9239 
9240 	return (CTL_RETVAL_COMPLETE);
9241 }
9242 
9243 static int
9244 ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
9245 {
9246 	struct scsi_inquiry *cdb;
9247 	int alloc_len, retval;
9248 
9249 	cdb = (struct scsi_inquiry *)ctsio->cdb;
9250 
9251 	retval = CTL_RETVAL_COMPLETE;
9252 
9253 	alloc_len = scsi_2btoul(cdb->length);
9254 
9255 	switch (cdb->page_code) {
9256 	case SVPD_SUPPORTED_PAGES:
9257 		retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
9258 		break;
9259 	case SVPD_UNIT_SERIAL_NUMBER:
9260 		retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
9261 		break;
9262 	case SVPD_DEVICE_ID:
9263 		retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
9264 		break;
9265 	default:
9266 		ctl_set_invalid_field(ctsio,
9267 				      /*sks_valid*/ 1,
9268 				      /*command*/ 1,
9269 				      /*field*/ 2,
9270 				      /*bit_valid*/ 0,
9271 				      /*bit*/ 0);
9272 		ctl_done((union ctl_io *)ctsio);
9273 		retval = CTL_RETVAL_COMPLETE;
9274 		break;
9275 	}
9276 
9277 	return (retval);
9278 }
9279 
9280 static int
9281 ctl_inquiry_std(struct ctl_scsiio *ctsio)
9282 {
9283 	struct scsi_inquiry_data *inq_ptr;
9284 	struct scsi_inquiry *cdb;
9285 	struct ctl_softc *ctl_softc;
9286 	struct ctl_lun *lun;
9287 	uint32_t alloc_len;
9288 	int is_fc;
9289 
9290 	ctl_softc = control_softc;
9291 
9292 	/*
9293 	 * Figure out whether we're talking to a Fibre Channel port or not.
9294 	 * We treat the ioctl front end, and any SCSI adapters, as packetized
9295 	 * SCSI front ends.
9296 	 */
9297 	mtx_lock(&ctl_softc->ctl_lock);
9298 	if (ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)]->port_type !=
9299 	    CTL_PORT_FC)
9300 		is_fc = 0;
9301 	else
9302 		is_fc = 1;
9303 	mtx_unlock(&ctl_softc->ctl_lock);
9304 
9305 	lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9306 	cdb = (struct scsi_inquiry *)ctsio->cdb;
9307 	alloc_len = scsi_2btoul(cdb->length);
9308 
9309 	/*
9310 	 * We malloc the full inquiry data size here and fill it
9311 	 * in.  If the user only asks for less, we'll give him
9312 	 * that much.
9313 	 */
9314 	ctsio->kern_data_ptr = malloc(sizeof(*inq_ptr), M_CTL, M_WAITOK | M_ZERO);
9315 	inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
9316 	ctsio->kern_sg_entries = 0;
9317 	ctsio->kern_data_resid = 0;
9318 	ctsio->kern_rel_offset = 0;
9319 
9320 	if (sizeof(*inq_ptr) < alloc_len) {
9321 		ctsio->residual = alloc_len - sizeof(*inq_ptr);
9322 		ctsio->kern_data_len = sizeof(*inq_ptr);
9323 		ctsio->kern_total_len = sizeof(*inq_ptr);
9324 	} else {
9325 		ctsio->residual = 0;
9326 		ctsio->kern_data_len = alloc_len;
9327 		ctsio->kern_total_len = alloc_len;
9328 	}
9329 
9330 	/*
9331 	 * If we have a LUN configured, report it as connected.  Otherwise,
9332 	 * report that it is offline or no device is supported, depending
9333 	 * on the value of inquiry_pq_no_lun.
9334 	 *
9335 	 * According to the spec (SPC-4 r34), the peripheral qualifier
9336 	 * SID_QUAL_LU_OFFLINE (001b) is used in the following scenario:
9337 	 *
9338 	 * "A peripheral device having the specified peripheral device type
9339 	 * is not connected to this logical unit. However, the device
9340 	 * server is capable of supporting the specified peripheral device
9341 	 * type on this logical unit."
9342 	 *
9343 	 * According to the same spec, the peripheral qualifier
9344 	 * SID_QUAL_BAD_LU (011b) is used in this scenario:
9345 	 *
9346 	 * "The device server is not capable of supporting a peripheral
9347 	 * device on this logical unit. For this peripheral qualifier the
9348 	 * peripheral device type shall be set to 1Fh. All other peripheral
9349 	 * device type values are reserved for this peripheral qualifier."
9350 	 *
9351 	 * Given the text, it would seem that we probably want to report that
9352 	 * the LUN is offline here.  There is no LUN connected, but we can
9353 	 * support a LUN at the given LUN number.
9354 	 *
9355 	 * In the real world, though, it sounds like things are a little
9356 	 * different:
9357 	 *
9358 	 * - Linux, when presented with a LUN with the offline peripheral
9359 	 *   qualifier, will create an sg driver instance for it.  So when
9360 	 *   you attach it to CTL, you wind up with a ton of sg driver
9361 	 *   instances.  (One for every LUN that Linux bothered to probe.)
9362 	 *   Linux does this despite the fact that it issues a REPORT LUNs
9363 	 *   to LUN 0 to get the inventory of supported LUNs.
9364 	 *
9365 	 * - There is other anecdotal evidence (from Emulex folks) about
9366 	 *   arrays that use the offline peripheral qualifier for LUNs that
9367 	 *   are on the "passive" path in an active/passive array.
9368 	 *
9369 	 * So the solution is provide a hopefully reasonable default
9370 	 * (return bad/no LUN) and allow the user to change the behavior
9371 	 * with a tunable/sysctl variable.
9372 	 */
9373 	if (lun != NULL)
9374 		inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9375 				  lun->be_lun->lun_type;
9376 	else if (ctl_softc->inquiry_pq_no_lun == 0)
9377 		inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9378 	else
9379 		inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
9380 
9381 	/* RMB in byte 2 is 0 */
9382 	inq_ptr->version = SCSI_REV_SPC3;
9383 
9384 	/*
9385 	 * According to SAM-3, even if a device only supports a single
9386 	 * level of LUN addressing, it should still set the HISUP bit:
9387 	 *
9388 	 * 4.9.1 Logical unit numbers overview
9389 	 *
9390 	 * All logical unit number formats described in this standard are
9391 	 * hierarchical in structure even when only a single level in that
9392 	 * hierarchy is used. The HISUP bit shall be set to one in the
9393 	 * standard INQUIRY data (see SPC-2) when any logical unit number
9394 	 * format described in this standard is used.  Non-hierarchical
9395 	 * formats are outside the scope of this standard.
9396 	 *
9397 	 * Therefore we set the HiSup bit here.
9398 	 *
9399 	 * The reponse format is 2, per SPC-3.
9400 	 */
9401 	inq_ptr->response_format = SID_HiSup | 2;
9402 
9403 	inq_ptr->additional_length = sizeof(*inq_ptr) - 4;
9404 	CTL_DEBUG_PRINT(("additional_length = %d\n",
9405 			 inq_ptr->additional_length));
9406 
9407 	inq_ptr->spc3_flags = SPC3_SID_TPGS_IMPLICIT;
9408 	/* 16 bit addressing */
9409 	if (is_fc == 0)
9410 		inq_ptr->spc2_flags = SPC2_SID_ADDR16;
9411 	/* XXX set the SID_MultiP bit here if we're actually going to
9412 	   respond on multiple ports */
9413 	inq_ptr->spc2_flags |= SPC2_SID_MultiP;
9414 
9415 	/* 16 bit data bus, synchronous transfers */
9416 	/* XXX these flags don't apply for FC */
9417 	if (is_fc == 0)
9418 		inq_ptr->flags = SID_WBus16 | SID_Sync;
9419 	/*
9420 	 * XXX KDM do we want to support tagged queueing on the control
9421 	 * device at all?
9422 	 */
9423 	if ((lun == NULL)
9424 	 || (lun->be_lun->lun_type != T_PROCESSOR))
9425 		inq_ptr->flags |= SID_CmdQue;
9426 	/*
9427 	 * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
9428 	 * We have 8 bytes for the vendor name, and 16 bytes for the device
9429 	 * name and 4 bytes for the revision.
9430 	 */
9431 	strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
9432 	if (lun == NULL) {
9433 		strcpy(inq_ptr->product, CTL_DIRECT_PRODUCT);
9434 	} else {
9435 		switch (lun->be_lun->lun_type) {
9436 		case T_DIRECT:
9437 			strcpy(inq_ptr->product, CTL_DIRECT_PRODUCT);
9438 			break;
9439 		case T_PROCESSOR:
9440 			strcpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT);
9441 			break;
9442 		default:
9443 			strcpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT);
9444 			break;
9445 		}
9446 	}
9447 
9448 	/*
9449 	 * XXX make this a macro somewhere so it automatically gets
9450 	 * incremented when we make changes.
9451 	 */
9452 	strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
9453 
9454 	/*
9455 	 * For parallel SCSI, we support double transition and single
9456 	 * transition clocking.  We also support QAS (Quick Arbitration
9457 	 * and Selection) and Information Unit transfers on both the
9458 	 * control and array devices.
9459 	 */
9460 	if (is_fc == 0)
9461 		inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
9462 				    SID_SPI_IUS;
9463 
9464 	/* SAM-3 */
9465 	scsi_ulto2b(0x0060, inq_ptr->version1);
9466 	/* SPC-3 (no version claimed) XXX should we claim a version? */
9467 	scsi_ulto2b(0x0300, inq_ptr->version2);
9468 	if (is_fc) {
9469 		/* FCP-2 ANSI INCITS.350:2003 */
9470 		scsi_ulto2b(0x0917, inq_ptr->version3);
9471 	} else {
9472 		/* SPI-4 ANSI INCITS.362:200x */
9473 		scsi_ulto2b(0x0B56, inq_ptr->version3);
9474 	}
9475 
9476 	if (lun == NULL) {
9477 		/* SBC-2 (no version claimed) XXX should we claim a version? */
9478 		scsi_ulto2b(0x0320, inq_ptr->version4);
9479 	} else {
9480 		switch (lun->be_lun->lun_type) {
9481 		case T_DIRECT:
9482 			/*
9483 			 * SBC-2 (no version claimed) XXX should we claim a
9484 			 * version?
9485 			 */
9486 			scsi_ulto2b(0x0320, inq_ptr->version4);
9487 			break;
9488 		case T_PROCESSOR:
9489 		default:
9490 			break;
9491 		}
9492 	}
9493 
9494 	ctsio->scsi_status = SCSI_STATUS_OK;
9495 	if (ctsio->kern_data_len > 0) {
9496 		ctsio->be_move_done = ctl_config_move_done;
9497 		ctl_datamove((union ctl_io *)ctsio);
9498 	} else {
9499 		ctsio->io_hdr.status = CTL_SUCCESS;
9500 		ctl_done((union ctl_io *)ctsio);
9501 	}
9502 
9503 	return (CTL_RETVAL_COMPLETE);
9504 }
9505 
9506 int
9507 ctl_inquiry(struct ctl_scsiio *ctsio)
9508 {
9509 	struct scsi_inquiry *cdb;
9510 	int retval;
9511 
9512 	cdb = (struct scsi_inquiry *)ctsio->cdb;
9513 
9514 	retval = 0;
9515 
9516 	CTL_DEBUG_PRINT(("ctl_inquiry\n"));
9517 
9518 	/*
9519 	 * Right now, we don't support the CmdDt inquiry information.
9520 	 * This would be nice to support in the future.  When we do
9521 	 * support it, we should change this test so that it checks to make
9522 	 * sure SI_EVPD and SI_CMDDT aren't both set at the same time.
9523 	 */
9524 #ifdef notyet
9525 	if (((cdb->byte2 & SI_EVPD)
9526 	 && (cdb->byte2 & SI_CMDDT)))
9527 #endif
9528 	if (cdb->byte2 & SI_CMDDT) {
9529 		/*
9530 		 * Point to the SI_CMDDT bit.  We might change this
9531 		 * when we support SI_CMDDT, but since both bits would be
9532 		 * "wrong", this should probably just stay as-is then.
9533 		 */
9534 		ctl_set_invalid_field(ctsio,
9535 				      /*sks_valid*/ 1,
9536 				      /*command*/ 1,
9537 				      /*field*/ 1,
9538 				      /*bit_valid*/ 1,
9539 				      /*bit*/ 1);
9540 		ctl_done((union ctl_io *)ctsio);
9541 		return (CTL_RETVAL_COMPLETE);
9542 	}
9543 	if (cdb->byte2 & SI_EVPD)
9544 		retval = ctl_inquiry_evpd(ctsio);
9545 #ifdef notyet
9546 	else if (cdb->byte2 & SI_CMDDT)
9547 		retval = ctl_inquiry_cmddt(ctsio);
9548 #endif
9549 	else
9550 		retval = ctl_inquiry_std(ctsio);
9551 
9552 	return (retval);
9553 }
9554 
9555 /*
9556  * For known CDB types, parse the LBA and length.
9557  */
9558 static int
9559 ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint32_t *len)
9560 {
9561 	if (io->io_hdr.io_type != CTL_IO_SCSI)
9562 		return (1);
9563 
9564 	switch (io->scsiio.cdb[0]) {
9565 	case READ_6:
9566 	case WRITE_6: {
9567 		struct scsi_rw_6 *cdb;
9568 
9569 		cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
9570 
9571 		*lba = scsi_3btoul(cdb->addr);
9572 		/* only 5 bits are valid in the most significant address byte */
9573 		*lba &= 0x1fffff;
9574 		*len = cdb->length;
9575 		break;
9576 	}
9577 	case READ_10:
9578 	case WRITE_10: {
9579 		struct scsi_rw_10 *cdb;
9580 
9581 		cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
9582 
9583 		*lba = scsi_4btoul(cdb->addr);
9584 		*len = scsi_2btoul(cdb->length);
9585 		break;
9586 	}
9587 	case WRITE_VERIFY_10: {
9588 		struct scsi_write_verify_10 *cdb;
9589 
9590 		cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
9591 
9592 		*lba = scsi_4btoul(cdb->addr);
9593 		*len = scsi_2btoul(cdb->length);
9594 		break;
9595 	}
9596 	case READ_12:
9597 	case WRITE_12: {
9598 		struct scsi_rw_12 *cdb;
9599 
9600 		cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
9601 
9602 		*lba = scsi_4btoul(cdb->addr);
9603 		*len = scsi_4btoul(cdb->length);
9604 		break;
9605 	}
9606 	case WRITE_VERIFY_12: {
9607 		struct scsi_write_verify_12 *cdb;
9608 
9609 		cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
9610 
9611 		*lba = scsi_4btoul(cdb->addr);
9612 		*len = scsi_4btoul(cdb->length);
9613 		break;
9614 	}
9615 	case READ_16:
9616 	case WRITE_16: {
9617 		struct scsi_rw_16 *cdb;
9618 
9619 		cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
9620 
9621 		*lba = scsi_8btou64(cdb->addr);
9622 		*len = scsi_4btoul(cdb->length);
9623 		break;
9624 	}
9625 	case WRITE_VERIFY_16: {
9626 		struct scsi_write_verify_16 *cdb;
9627 
9628 		cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
9629 
9630 
9631 		*lba = scsi_8btou64(cdb->addr);
9632 		*len = scsi_4btoul(cdb->length);
9633 		break;
9634 	}
9635 	default:
9636 		return (1);
9637 		break; /* NOTREACHED */
9638 	}
9639 
9640 	return (0);
9641 }
9642 
9643 static ctl_action
9644 ctl_extent_check_lba(uint64_t lba1, uint32_t len1, uint64_t lba2, uint32_t len2)
9645 {
9646 	uint64_t endlba1, endlba2;
9647 
9648 	endlba1 = lba1 + len1 - 1;
9649 	endlba2 = lba2 + len2 - 1;
9650 
9651 	if ((endlba1 < lba2)
9652 	 || (endlba2 < lba1))
9653 		return (CTL_ACTION_PASS);
9654 	else
9655 		return (CTL_ACTION_BLOCK);
9656 }
9657 
9658 static ctl_action
9659 ctl_extent_check(union ctl_io *io1, union ctl_io *io2)
9660 {
9661 	uint64_t lba1, lba2;
9662 	uint32_t len1, len2;
9663 	int retval;
9664 
9665 	retval = ctl_get_lba_len(io1, &lba1, &len1);
9666 	if (retval != 0)
9667 		return (CTL_ACTION_ERROR);
9668 
9669 	retval = ctl_get_lba_len(io2, &lba2, &len2);
9670 	if (retval != 0)
9671 		return (CTL_ACTION_ERROR);
9672 
9673 	return (ctl_extent_check_lba(lba1, len1, lba2, len2));
9674 }
9675 
9676 static ctl_action
9677 ctl_check_for_blockage(union ctl_io *pending_io, union ctl_io *ooa_io)
9678 {
9679 	struct ctl_cmd_entry *pending_entry, *ooa_entry;
9680 	ctl_serialize_action *serialize_row;
9681 
9682 	/*
9683 	 * The initiator attempted multiple untagged commands at the same
9684 	 * time.  Can't do that.
9685 	 */
9686 	if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
9687 	 && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
9688 	 && ((pending_io->io_hdr.nexus.targ_port ==
9689 	      ooa_io->io_hdr.nexus.targ_port)
9690 	  && (pending_io->io_hdr.nexus.initid.id ==
9691 	      ooa_io->io_hdr.nexus.initid.id))
9692 	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
9693 		return (CTL_ACTION_OVERLAP);
9694 
9695 	/*
9696 	 * The initiator attempted to send multiple tagged commands with
9697 	 * the same ID.  (It's fine if different initiators have the same
9698 	 * tag ID.)
9699 	 *
9700 	 * Even if all of those conditions are true, we don't kill the I/O
9701 	 * if the command ahead of us has been aborted.  We won't end up
9702 	 * sending it to the FETD, and it's perfectly legal to resend a
9703 	 * command with the same tag number as long as the previous
9704 	 * instance of this tag number has been aborted somehow.
9705 	 */
9706 	if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
9707 	 && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
9708 	 && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
9709 	 && ((pending_io->io_hdr.nexus.targ_port ==
9710 	      ooa_io->io_hdr.nexus.targ_port)
9711 	  && (pending_io->io_hdr.nexus.initid.id ==
9712 	      ooa_io->io_hdr.nexus.initid.id))
9713 	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
9714 		return (CTL_ACTION_OVERLAP_TAG);
9715 
9716 	/*
9717 	 * If we get a head of queue tag, SAM-3 says that we should
9718 	 * immediately execute it.
9719 	 *
9720 	 * What happens if this command would normally block for some other
9721 	 * reason?  e.g. a request sense with a head of queue tag
9722 	 * immediately after a write.  Normally that would block, but this
9723 	 * will result in its getting executed immediately...
9724 	 *
9725 	 * We currently return "pass" instead of "skip", so we'll end up
9726 	 * going through the rest of the queue to check for overlapped tags.
9727 	 *
9728 	 * XXX KDM check for other types of blockage first??
9729 	 */
9730 	if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
9731 		return (CTL_ACTION_PASS);
9732 
9733 	/*
9734 	 * Ordered tags have to block until all items ahead of them
9735 	 * have completed.  If we get called with an ordered tag, we always
9736 	 * block, if something else is ahead of us in the queue.
9737 	 */
9738 	if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
9739 		return (CTL_ACTION_BLOCK);
9740 
9741 	/*
9742 	 * Simple tags get blocked until all head of queue and ordered tags
9743 	 * ahead of them have completed.  I'm lumping untagged commands in
9744 	 * with simple tags here.  XXX KDM is that the right thing to do?
9745 	 */
9746 	if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
9747 	  || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
9748 	 && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
9749 	  || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
9750 		return (CTL_ACTION_BLOCK);
9751 
9752 	pending_entry = &ctl_cmd_table[pending_io->scsiio.cdb[0]];
9753 	ooa_entry = &ctl_cmd_table[ooa_io->scsiio.cdb[0]];
9754 
9755 	serialize_row = ctl_serialize_table[ooa_entry->seridx];
9756 
9757 	switch (serialize_row[pending_entry->seridx]) {
9758 	case CTL_SER_BLOCK:
9759 		return (CTL_ACTION_BLOCK);
9760 		break; /* NOTREACHED */
9761 	case CTL_SER_EXTENT:
9762 		return (ctl_extent_check(pending_io, ooa_io));
9763 		break; /* NOTREACHED */
9764 	case CTL_SER_PASS:
9765 		return (CTL_ACTION_PASS);
9766 		break; /* NOTREACHED */
9767 	case CTL_SER_SKIP:
9768 		return (CTL_ACTION_SKIP);
9769 		break;
9770 	default:
9771 		panic("invalid serialization value %d",
9772 		      serialize_row[pending_entry->seridx]);
9773 		break; /* NOTREACHED */
9774 	}
9775 
9776 	return (CTL_ACTION_ERROR);
9777 }
9778 
9779 /*
9780  * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
9781  * Assumptions:
9782  * - pending_io is generally either incoming, or on the blocked queue
9783  * - starting I/O is the I/O we want to start the check with.
9784  */
9785 static ctl_action
9786 ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
9787 	      union ctl_io *starting_io)
9788 {
9789 	union ctl_io *ooa_io;
9790 	ctl_action action;
9791 
9792 	mtx_assert(&control_softc->ctl_lock, MA_OWNED);
9793 
9794 	/*
9795 	 * Run back along the OOA queue, starting with the current
9796 	 * blocked I/O and going through every I/O before it on the
9797 	 * queue.  If starting_io is NULL, we'll just end up returning
9798 	 * CTL_ACTION_PASS.
9799 	 */
9800 	for (ooa_io = starting_io; ooa_io != NULL;
9801 	     ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
9802 	     ooa_links)){
9803 
9804 		/*
9805 		 * This routine just checks to see whether
9806 		 * cur_blocked is blocked by ooa_io, which is ahead
9807 		 * of it in the queue.  It doesn't queue/dequeue
9808 		 * cur_blocked.
9809 		 */
9810 		action = ctl_check_for_blockage(pending_io, ooa_io);
9811 		switch (action) {
9812 		case CTL_ACTION_BLOCK:
9813 		case CTL_ACTION_OVERLAP:
9814 		case CTL_ACTION_OVERLAP_TAG:
9815 		case CTL_ACTION_SKIP:
9816 		case CTL_ACTION_ERROR:
9817 			return (action);
9818 			break; /* NOTREACHED */
9819 		case CTL_ACTION_PASS:
9820 			break;
9821 		default:
9822 			panic("invalid action %d", action);
9823 			break;  /* NOTREACHED */
9824 		}
9825 	}
9826 
9827 	return (CTL_ACTION_PASS);
9828 }
9829 
9830 /*
9831  * Assumptions:
9832  * - An I/O has just completed, and has been removed from the per-LUN OOA
9833  *   queue, so some items on the blocked queue may now be unblocked.
9834  */
9835 static int
9836 ctl_check_blocked(struct ctl_lun *lun)
9837 {
9838 	union ctl_io *cur_blocked, *next_blocked;
9839 
9840 	mtx_assert(&control_softc->ctl_lock, MA_OWNED);
9841 
9842 	/*
9843 	 * Run forward from the head of the blocked queue, checking each
9844 	 * entry against the I/Os prior to it on the OOA queue to see if
9845 	 * there is still any blockage.
9846 	 *
9847 	 * We cannot use the TAILQ_FOREACH() macro, because it can't deal
9848 	 * with our removing a variable on it while it is traversing the
9849 	 * list.
9850 	 */
9851 	for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
9852 	     cur_blocked != NULL; cur_blocked = next_blocked) {
9853 		union ctl_io *prev_ooa;
9854 		ctl_action action;
9855 
9856 		next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
9857 							  blocked_links);
9858 
9859 		prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
9860 						      ctl_ooaq, ooa_links);
9861 
9862 		/*
9863 		 * If cur_blocked happens to be the first item in the OOA
9864 		 * queue now, prev_ooa will be NULL, and the action
9865 		 * returned will just be CTL_ACTION_PASS.
9866 		 */
9867 		action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
9868 
9869 		switch (action) {
9870 		case CTL_ACTION_BLOCK:
9871 			/* Nothing to do here, still blocked */
9872 			break;
9873 		case CTL_ACTION_OVERLAP:
9874 		case CTL_ACTION_OVERLAP_TAG:
9875 			/*
9876 			 * This shouldn't happen!  In theory we've already
9877 			 * checked this command for overlap...
9878 			 */
9879 			break;
9880 		case CTL_ACTION_PASS:
9881 		case CTL_ACTION_SKIP: {
9882 			struct ctl_softc *softc;
9883 			struct ctl_cmd_entry *entry;
9884 			uint32_t initidx;
9885 			uint8_t opcode;
9886 			int isc_retval;
9887 
9888 			/*
9889 			 * The skip case shouldn't happen, this transaction
9890 			 * should have never made it onto the blocked queue.
9891 			 */
9892 			/*
9893 			 * This I/O is no longer blocked, we can remove it
9894 			 * from the blocked queue.  Since this is a TAILQ
9895 			 * (doubly linked list), we can do O(1) removals
9896 			 * from any place on the list.
9897 			 */
9898 			TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
9899 				     blocked_links);
9900 			cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
9901 
9902 			if (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC){
9903 				/*
9904 				 * Need to send IO back to original side to
9905 				 * run
9906 				 */
9907 				union ctl_ha_msg msg_info;
9908 
9909 				msg_info.hdr.original_sc =
9910 					cur_blocked->io_hdr.original_sc;
9911 				msg_info.hdr.serializing_sc = cur_blocked;
9912 				msg_info.hdr.msg_type = CTL_MSG_R2R;
9913 				if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
9914 				     &msg_info, sizeof(msg_info), 0)) >
9915 				     CTL_HA_STATUS_SUCCESS) {
9916 					printf("CTL:Check Blocked error from "
9917 					       "ctl_ha_msg_send %d\n",
9918 					       isc_retval);
9919 				}
9920 				break;
9921 			}
9922 			opcode = cur_blocked->scsiio.cdb[0];
9923 			entry = &ctl_cmd_table[opcode];
9924 			softc = control_softc;
9925 
9926 			initidx = ctl_get_initindex(&cur_blocked->io_hdr.nexus);
9927 
9928 			/*
9929 			 * Check this I/O for LUN state changes that may
9930 			 * have happened while this command was blocked.
9931 			 * The LUN state may have been changed by a command
9932 			 * ahead of us in the queue, so we need to re-check
9933 			 * for any states that can be caused by SCSI
9934 			 * commands.
9935 			 */
9936 			if (ctl_scsiio_lun_check(softc, lun, entry,
9937 						 &cur_blocked->scsiio) == 0) {
9938 				cur_blocked->io_hdr.flags |=
9939 				                      CTL_FLAG_IS_WAS_ON_RTR;
9940 				STAILQ_INSERT_TAIL(&lun->ctl_softc->rtr_queue,
9941 						   &cur_blocked->io_hdr, links);
9942 				/*
9943 				 * In the non CTL_DONE_THREAD case, we need
9944 				 * to wake up the work thread here.  When
9945 				 * we're processing completed requests from
9946 				 * the work thread context, we'll pop back
9947 				 * around and end up pulling things off the
9948 				 * RtR queue.  When we aren't processing
9949 				 * things from the work thread context,
9950 				 * though, we won't ever check the RtR queue.
9951 				 * So we need to wake up the thread to clear
9952 				 * things off the queue.  Otherwise this
9953 				 * transaction will just sit on the RtR queue
9954 				 * until a new I/O comes in.  (Which may or
9955 				 * may not happen...)
9956 				 */
9957 #ifndef CTL_DONE_THREAD
9958 				ctl_wakeup_thread();
9959 #endif
9960 			} else
9961 				ctl_done_lock(cur_blocked, /*have_lock*/ 1);
9962 			break;
9963 		}
9964 		default:
9965 			/*
9966 			 * This probably shouldn't happen -- we shouldn't
9967 			 * get CTL_ACTION_ERROR, or anything else.
9968 			 */
9969 			break;
9970 		}
9971 	}
9972 
9973 	return (CTL_RETVAL_COMPLETE);
9974 }
9975 
9976 /*
9977  * This routine (with one exception) checks LUN flags that can be set by
9978  * commands ahead of us in the OOA queue.  These flags have to be checked
9979  * when a command initially comes in, and when we pull a command off the
9980  * blocked queue and are preparing to execute it.  The reason we have to
9981  * check these flags for commands on the blocked queue is that the LUN
9982  * state may have been changed by a command ahead of us while we're on the
9983  * blocked queue.
9984  *
9985  * Ordering is somewhat important with these checks, so please pay
9986  * careful attention to the placement of any new checks.
9987  */
9988 static int
9989 ctl_scsiio_lun_check(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
9990 		     struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
9991 {
9992 	int retval;
9993 
9994 	retval = 0;
9995 
9996 	/*
9997 	 * If this shelf is a secondary shelf controller, we have to reject
9998 	 * any media access commands.
9999 	 */
10000 #if 0
10001 	/* No longer needed for HA */
10002 	if (((ctl_softc->flags & CTL_FLAG_MASTER_SHELF) == 0)
10003 	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_SECONDARY) == 0)) {
10004 		ctl_set_lun_standby(ctsio);
10005 		retval = 1;
10006 		goto bailout;
10007 	}
10008 #endif
10009 
10010 	/*
10011 	 * Check for a reservation conflict.  If this command isn't allowed
10012 	 * even on reserved LUNs, and if this initiator isn't the one who
10013 	 * reserved us, reject the command with a reservation conflict.
10014 	 */
10015 	if ((lun->flags & CTL_LUN_RESERVED)
10016 	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
10017 		if ((ctsio->io_hdr.nexus.initid.id != lun->rsv_nexus.initid.id)
10018 		 || (ctsio->io_hdr.nexus.targ_port != lun->rsv_nexus.targ_port)
10019 		 || (ctsio->io_hdr.nexus.targ_target.id !=
10020 		     lun->rsv_nexus.targ_target.id)) {
10021 			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
10022 			ctsio->io_hdr.status = CTL_SCSI_ERROR;
10023 			retval = 1;
10024 			goto bailout;
10025 		}
10026 	}
10027 
10028 	if ( (lun->flags & CTL_LUN_PR_RESERVED)
10029 	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV) == 0)) {
10030 		uint32_t residx;
10031 
10032 		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
10033 		/*
10034 		 * if we aren't registered or it's a res holder type
10035 		 * reservation and this isn't the res holder then set a
10036 		 * conflict.
10037 		 * NOTE: Commands which might be allowed on write exclusive
10038 		 * type reservations are checked in the particular command
10039 		 * for a conflict. Read and SSU are the only ones.
10040 		 */
10041 		if (!lun->per_res[residx].registered
10042 		 || (residx != lun->pr_res_idx && lun->res_type < 4)) {
10043 			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
10044 			ctsio->io_hdr.status = CTL_SCSI_ERROR;
10045 			retval = 1;
10046 			goto bailout;
10047 		}
10048 
10049 	}
10050 
10051 	if ((lun->flags & CTL_LUN_OFFLINE)
10052 	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_OFFLINE) == 0)) {
10053 		ctl_set_lun_not_ready(ctsio);
10054 		retval = 1;
10055 		goto bailout;
10056 	}
10057 
10058 	/*
10059 	 * If the LUN is stopped, see if this particular command is allowed
10060 	 * for a stopped lun.  Otherwise, reject it with 0x04,0x02.
10061 	 */
10062 	if ((lun->flags & CTL_LUN_STOPPED)
10063 	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_STOPPED) == 0)) {
10064 		/* "Logical unit not ready, initializing cmd. required" */
10065 		ctl_set_lun_stopped(ctsio);
10066 		retval = 1;
10067 		goto bailout;
10068 	}
10069 
10070 	if ((lun->flags & CTL_LUN_INOPERABLE)
10071 	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_INOPERABLE) == 0)) {
10072 		/* "Medium format corrupted" */
10073 		ctl_set_medium_format_corrupted(ctsio);
10074 		retval = 1;
10075 		goto bailout;
10076 	}
10077 
10078 bailout:
10079 	return (retval);
10080 
10081 }
10082 
10083 static void
10084 ctl_failover_io(union ctl_io *io, int have_lock)
10085 {
10086 	ctl_set_busy(&io->scsiio);
10087 	ctl_done_lock(io, have_lock);
10088 }
10089 
10090 static void
10091 ctl_failover(void)
10092 {
10093 	struct ctl_lun *lun;
10094 	struct ctl_softc *ctl_softc;
10095 	union ctl_io *next_io, *pending_io;
10096 	union ctl_io *io;
10097 	int lun_idx;
10098 	int i;
10099 
10100 	ctl_softc = control_softc;
10101 
10102 	mtx_lock(&ctl_softc->ctl_lock);
10103 	/*
10104 	 * Remove any cmds from the other SC from the rtr queue.  These
10105 	 * will obviously only be for LUNs for which we're the primary.
10106 	 * We can't send status or get/send data for these commands.
10107 	 * Since they haven't been executed yet, we can just remove them.
10108 	 * We'll either abort them or delete them below, depending on
10109 	 * which HA mode we're in.
10110 	 */
10111 	for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->rtr_queue);
10112 	     io != NULL; io = next_io) {
10113 		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
10114 		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
10115 			STAILQ_REMOVE(&ctl_softc->rtr_queue, &io->io_hdr,
10116 				      ctl_io_hdr, links);
10117 	}
10118 
10119 	for (lun_idx=0; lun_idx < ctl_softc->num_luns; lun_idx++) {
10120 		lun = ctl_softc->ctl_luns[lun_idx];
10121 		if (lun==NULL)
10122 			continue;
10123 
10124 		/*
10125 		 * Processor LUNs are primary on both sides.
10126 		 * XXX will this always be true?
10127 		 */
10128 		if (lun->be_lun->lun_type == T_PROCESSOR)
10129 			continue;
10130 
10131 		if ((lun->flags & CTL_LUN_PRIMARY_SC)
10132 		 && (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
10133 			printf("FAILOVER: primary lun %d\n", lun_idx);
10134 		        /*
10135 			 * Remove all commands from the other SC. First from the
10136 			 * blocked queue then from the ooa queue. Once we have
10137 			 * removed them. Call ctl_check_blocked to see if there
10138 			 * is anything that can run.
10139 			 */
10140 			for (io = (union ctl_io *)TAILQ_FIRST(
10141 			     &lun->blocked_queue); io != NULL; io = next_io) {
10142 
10143 		        	next_io = (union ctl_io *)TAILQ_NEXT(
10144 				    &io->io_hdr, blocked_links);
10145 
10146 				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
10147 					TAILQ_REMOVE(&lun->blocked_queue,
10148 						     &io->io_hdr,blocked_links);
10149 					io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
10150 					TAILQ_REMOVE(&lun->ooa_queue,
10151 						     &io->io_hdr, ooa_links);
10152 
10153 					ctl_free_io_internal(io, 1);
10154 				}
10155 			}
10156 
10157 			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
10158 	     		     io != NULL; io = next_io) {
10159 
10160 		        	next_io = (union ctl_io *)TAILQ_NEXT(
10161 				    &io->io_hdr, ooa_links);
10162 
10163 				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
10164 
10165 					TAILQ_REMOVE(&lun->ooa_queue,
10166 						&io->io_hdr,
10167 					     	ooa_links);
10168 
10169 					ctl_free_io_internal(io, 1);
10170 				}
10171 			}
10172 			ctl_check_blocked(lun);
10173 		} else if ((lun->flags & CTL_LUN_PRIMARY_SC)
10174 			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
10175 
10176 			printf("FAILOVER: primary lun %d\n", lun_idx);
10177 			/*
10178 			 * Abort all commands from the other SC.  We can't
10179 			 * send status back for them now.  These should get
10180 			 * cleaned up when they are completed or come out
10181 			 * for a datamove operation.
10182 			 */
10183 			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
10184 	     		     io != NULL; io = next_io) {
10185 		        	next_io = (union ctl_io *)TAILQ_NEXT(
10186 					&io->io_hdr, ooa_links);
10187 
10188 				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
10189 					io->io_hdr.flags |= CTL_FLAG_ABORT;
10190 			}
10191 		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
10192 			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
10193 
10194 			printf("FAILOVER: secondary lun %d\n", lun_idx);
10195 
10196 			lun->flags |= CTL_LUN_PRIMARY_SC;
10197 
10198 			/*
10199 			 * We send all I/O that was sent to this controller
10200 			 * and redirected to the other side back with
10201 			 * busy status, and have the initiator retry it.
10202 			 * Figuring out how much data has been transferred,
10203 			 * etc. and picking up where we left off would be
10204 			 * very tricky.
10205 			 *
10206 			 * XXX KDM need to remove I/O from the blocked
10207 			 * queue as well!
10208 			 */
10209 			for (pending_io = (union ctl_io *)TAILQ_FIRST(
10210 			     &lun->ooa_queue); pending_io != NULL;
10211 			     pending_io = next_io) {
10212 
10213 				next_io =  (union ctl_io *)TAILQ_NEXT(
10214 					&pending_io->io_hdr, ooa_links);
10215 
10216 				pending_io->io_hdr.flags &=
10217 					~CTL_FLAG_SENT_2OTHER_SC;
10218 
10219 				if (pending_io->io_hdr.flags &
10220 				    CTL_FLAG_IO_ACTIVE) {
10221 					pending_io->io_hdr.flags |=
10222 						CTL_FLAG_FAILOVER;
10223 				} else {
10224 					ctl_set_busy(&pending_io->scsiio);
10225 					ctl_done_lock(pending_io,
10226 						      /*have_lock*/1);
10227 				}
10228 			}
10229 
10230 			/*
10231 			 * Build Unit Attention
10232 			 */
10233 			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
10234 				lun->pending_sense[i].ua_pending |=
10235 				                     CTL_UA_ASYM_ACC_CHANGE;
10236 			}
10237 		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
10238 			&& (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
10239 			printf("FAILOVER: secondary lun %d\n", lun_idx);
10240 			/*
10241 			 * if the first io on the OOA is not on the RtR queue
10242 			 * add it.
10243 			 */
10244 			lun->flags |= CTL_LUN_PRIMARY_SC;
10245 
10246 			pending_io = (union ctl_io *)TAILQ_FIRST(
10247 			    &lun->ooa_queue);
10248 			if (pending_io==NULL) {
10249 				printf("Nothing on OOA queue\n");
10250 				continue;
10251 			}
10252 
10253 			pending_io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
10254 			if ((pending_io->io_hdr.flags &
10255 			     CTL_FLAG_IS_WAS_ON_RTR) == 0) {
10256 				pending_io->io_hdr.flags |=
10257 				    CTL_FLAG_IS_WAS_ON_RTR;
10258 				STAILQ_INSERT_TAIL(&ctl_softc->rtr_queue,
10259 						   &pending_io->io_hdr, links);
10260 			}
10261 #if 0
10262 			else
10263 			{
10264 				printf("Tag 0x%04x is running\n",
10265 				      pending_io->scsiio.tag_num);
10266 			}
10267 #endif
10268 
10269 			next_io = (union ctl_io *)TAILQ_NEXT(
10270 			    &pending_io->io_hdr, ooa_links);
10271 			for (pending_io=next_io; pending_io != NULL;
10272 			     pending_io = next_io) {
10273 				pending_io->io_hdr.flags &=
10274 				    ~CTL_FLAG_SENT_2OTHER_SC;
10275 				next_io = (union ctl_io *)TAILQ_NEXT(
10276 					&pending_io->io_hdr, ooa_links);
10277 				if (pending_io->io_hdr.flags &
10278 				    CTL_FLAG_IS_WAS_ON_RTR) {
10279 #if 0
10280 				        printf("Tag 0x%04x is running\n",
10281 				      		pending_io->scsiio.tag_num);
10282 #endif
10283 					continue;
10284 				}
10285 
10286 				switch (ctl_check_ooa(lun, pending_io,
10287 			            (union ctl_io *)TAILQ_PREV(
10288 				    &pending_io->io_hdr, ctl_ooaq,
10289 				    ooa_links))) {
10290 
10291 				case CTL_ACTION_BLOCK:
10292 					TAILQ_INSERT_TAIL(&lun->blocked_queue,
10293 							  &pending_io->io_hdr,
10294 							  blocked_links);
10295 					pending_io->io_hdr.flags |=
10296 					    CTL_FLAG_BLOCKED;
10297 					break;
10298 				case CTL_ACTION_PASS:
10299 				case CTL_ACTION_SKIP:
10300 					pending_io->io_hdr.flags |=
10301 					    CTL_FLAG_IS_WAS_ON_RTR;
10302 					STAILQ_INSERT_TAIL(
10303 					    &ctl_softc->rtr_queue,
10304 					    &pending_io->io_hdr, links);
10305 					break;
10306 				case CTL_ACTION_OVERLAP:
10307 					ctl_set_overlapped_cmd(
10308 					    (struct ctl_scsiio *)pending_io);
10309 					ctl_done_lock(pending_io,
10310 						      /*have_lock*/ 1);
10311 					break;
10312 				case CTL_ACTION_OVERLAP_TAG:
10313 					ctl_set_overlapped_tag(
10314 					    (struct ctl_scsiio *)pending_io,
10315 					    pending_io->scsiio.tag_num & 0xff);
10316 					ctl_done_lock(pending_io,
10317 						      /*have_lock*/ 1);
10318 					break;
10319 				case CTL_ACTION_ERROR:
10320 				default:
10321 					ctl_set_internal_failure(
10322 						(struct ctl_scsiio *)pending_io,
10323 						0,  // sks_valid
10324 						0); //retry count
10325 					ctl_done_lock(pending_io,
10326 						      /*have_lock*/ 1);
10327 					break;
10328 				}
10329 			}
10330 
10331 			/*
10332 			 * Build Unit Attention
10333 			 */
10334 			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
10335 				lun->pending_sense[i].ua_pending |=
10336 				                     CTL_UA_ASYM_ACC_CHANGE;
10337 			}
10338 		} else {
10339 			panic("Unhandled HA mode failover, LUN flags = %#x, "
10340 			      "ha_mode = #%x", lun->flags, ctl_softc->ha_mode);
10341 		}
10342 	}
10343 	ctl_pause_rtr = 0;
10344 	mtx_unlock(&ctl_softc->ctl_lock);
10345 }
10346 
10347 static int
10348 ctl_scsiio_precheck(struct ctl_softc *ctl_softc, struct ctl_scsiio *ctsio)
10349 {
10350 	struct ctl_lun *lun;
10351 	struct ctl_cmd_entry *entry;
10352 	uint8_t opcode;
10353 	uint32_t initidx;
10354 	int retval;
10355 
10356 	retval = 0;
10357 
10358 	lun = NULL;
10359 
10360 	opcode = ctsio->cdb[0];
10361 
10362 	mtx_lock(&ctl_softc->ctl_lock);
10363 
10364 	if ((ctsio->io_hdr.nexus.targ_lun < CTL_MAX_LUNS)
10365 	 && (ctl_softc->ctl_luns[ctsio->io_hdr.nexus.targ_lun] != NULL)) {
10366 		lun = ctl_softc->ctl_luns[ctsio->io_hdr.nexus.targ_lun];
10367 		/*
10368 		 * If the LUN is invalid, pretend that it doesn't exist.
10369 		 * It will go away as soon as all pending I/O has been
10370 		 * completed.
10371 		 */
10372 		if (lun->flags & CTL_LUN_DISABLED) {
10373 			lun = NULL;
10374 		} else {
10375 			ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
10376 			ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
10377 				lun->be_lun;
10378 			if (lun->be_lun->lun_type == T_PROCESSOR) {
10379 				ctsio->io_hdr.flags |= CTL_FLAG_CONTROL_DEV;
10380 			}
10381 		}
10382 	} else {
10383 		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
10384 		ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
10385 	}
10386 
10387 	entry = &ctl_cmd_table[opcode];
10388 
10389 	ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
10390 	ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
10391 
10392 	/*
10393 	 * Check to see whether we can send this command to LUNs that don't
10394 	 * exist.  This should pretty much only be the case for inquiry
10395 	 * and request sense.  Further checks, below, really require having
10396 	 * a LUN, so we can't really check the command anymore.  Just put
10397 	 * it on the rtr queue.
10398 	 */
10399 	if (lun == NULL) {
10400 		if (entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS)
10401 			goto queue_rtr;
10402 
10403 		ctl_set_unsupported_lun(ctsio);
10404 		mtx_unlock(&ctl_softc->ctl_lock);
10405 		ctl_done((union ctl_io *)ctsio);
10406 		goto bailout;
10407 	} else {
10408 		/*
10409 		 * Every I/O goes into the OOA queue for a particular LUN, and
10410 		 * stays there until completion.
10411 		 */
10412 		TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
10413 
10414 		/*
10415 		 * Make sure we support this particular command on this LUN.
10416 		 * e.g., we don't support writes to the control LUN.
10417 		 */
10418 		switch (lun->be_lun->lun_type) {
10419 		case T_PROCESSOR:
10420 		 	if (((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0)
10421 			 && ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS)
10422 			      == 0)) {
10423 				ctl_set_invalid_opcode(ctsio);
10424 				mtx_unlock(&ctl_softc->ctl_lock);
10425 				ctl_done((union ctl_io *)ctsio);
10426 				goto bailout;
10427 			}
10428 			break;
10429 		case T_DIRECT:
10430 			if (((entry->flags & CTL_CMD_FLAG_OK_ON_SLUN) == 0)
10431 			 && ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS)
10432 			      == 0)){
10433 				ctl_set_invalid_opcode(ctsio);
10434 				mtx_unlock(&ctl_softc->ctl_lock);
10435 				ctl_done((union ctl_io *)ctsio);
10436 				goto bailout;
10437 			}
10438 			break;
10439 		default:
10440 			printf("Unsupported CTL LUN type %d\n",
10441 			       lun->be_lun->lun_type);
10442 			panic("Unsupported CTL LUN type %d\n",
10443 			      lun->be_lun->lun_type);
10444 			break; /* NOTREACHED */
10445 		}
10446 	}
10447 
10448 	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
10449 
10450 	/*
10451 	 * If we've got a request sense, it'll clear the contingent
10452 	 * allegiance condition.  Otherwise, if we have a CA condition for
10453 	 * this initiator, clear it, because it sent down a command other
10454 	 * than request sense.
10455 	 */
10456 	if ((opcode != REQUEST_SENSE)
10457 	 && (ctl_is_set(lun->have_ca, initidx)))
10458 		ctl_clear_mask(lun->have_ca, initidx);
10459 
10460 	/*
10461 	 * If the command has this flag set, it handles its own unit
10462 	 * attention reporting, we shouldn't do anything.  Otherwise we
10463 	 * check for any pending unit attentions, and send them back to the
10464 	 * initiator.  We only do this when a command initially comes in,
10465 	 * not when we pull it off the blocked queue.
10466 	 *
10467 	 * According to SAM-3, section 5.3.2, the order that things get
10468 	 * presented back to the host is basically unit attentions caused
10469 	 * by some sort of reset event, busy status, reservation conflicts
10470 	 * or task set full, and finally any other status.
10471 	 *
10472 	 * One issue here is that some of the unit attentions we report
10473 	 * don't fall into the "reset" category (e.g. "reported luns data
10474 	 * has changed").  So reporting it here, before the reservation
10475 	 * check, may be technically wrong.  I guess the only thing to do
10476 	 * would be to check for and report the reset events here, and then
10477 	 * check for the other unit attention types after we check for a
10478 	 * reservation conflict.
10479 	 *
10480 	 * XXX KDM need to fix this
10481 	 */
10482 	if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
10483 		ctl_ua_type ua_type;
10484 
10485 		ua_type = lun->pending_sense[initidx].ua_pending;
10486 		if (ua_type != CTL_UA_NONE) {
10487 			scsi_sense_data_type sense_format;
10488 
10489 			if (lun != NULL)
10490 				sense_format = (lun->flags &
10491 				    CTL_LUN_SENSE_DESC) ? SSD_TYPE_DESC :
10492 				    SSD_TYPE_FIXED;
10493 			else
10494 				sense_format = SSD_TYPE_FIXED;
10495 
10496 			ua_type = ctl_build_ua(ua_type, &ctsio->sense_data,
10497 					       sense_format);
10498 			if (ua_type != CTL_UA_NONE) {
10499 				ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
10500 				ctsio->io_hdr.status = CTL_SCSI_ERROR |
10501 						       CTL_AUTOSENSE;
10502 				ctsio->sense_len = SSD_FULL_SIZE;
10503 				lun->pending_sense[initidx].ua_pending &=
10504 					~ua_type;
10505 				mtx_unlock(&ctl_softc->ctl_lock);
10506 				ctl_done((union ctl_io *)ctsio);
10507 				goto bailout;
10508 			}
10509 		}
10510 	}
10511 
10512 
10513 	if (ctl_scsiio_lun_check(ctl_softc, lun, entry, ctsio) != 0) {
10514 		mtx_unlock(&ctl_softc->ctl_lock);
10515 		ctl_done((union ctl_io *)ctsio);
10516 		goto bailout;
10517 	}
10518 
10519 	/*
10520 	 * XXX CHD this is where we want to send IO to other side if
10521 	 * this LUN is secondary on this SC. We will need to make a copy
10522 	 * of the IO and flag the IO on this side as SENT_2OTHER and the flag
10523 	 * the copy we send as FROM_OTHER.
10524 	 * We also need to stuff the address of the original IO so we can
10525 	 * find it easily. Something similar will need be done on the other
10526 	 * side so when we are done we can find the copy.
10527 	 */
10528 	if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
10529 		union ctl_ha_msg msg_info;
10530 		int isc_retval;
10531 
10532 		ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
10533 
10534 		msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
10535 		msg_info.hdr.original_sc = (union ctl_io *)ctsio;
10536 #if 0
10537 		printf("1. ctsio %p\n", ctsio);
10538 #endif
10539 		msg_info.hdr.serializing_sc = NULL;
10540 		msg_info.hdr.nexus = ctsio->io_hdr.nexus;
10541 		msg_info.scsi.tag_num = ctsio->tag_num;
10542 		msg_info.scsi.tag_type = ctsio->tag_type;
10543 		memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
10544 
10545 		ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
10546 
10547 		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
10548 		    (void *)&msg_info, sizeof(msg_info), 0)) >
10549 		    CTL_HA_STATUS_SUCCESS) {
10550 			printf("CTL:precheck, ctl_ha_msg_send returned %d\n",
10551 			       isc_retval);
10552 			printf("CTL:opcode is %x\n",opcode);
10553 		} else {
10554 #if 0
10555 			printf("CTL:Precheck sent msg, opcode is %x\n",opcode);
10556 #endif
10557 		}
10558 
10559 		/*
10560 		 * XXX KDM this I/O is off the incoming queue, but hasn't
10561 		 * been inserted on any other queue.  We may need to come
10562 		 * up with a holding queue while we wait for serialization
10563 		 * so that we have an idea of what we're waiting for from
10564 		 * the other side.
10565 		 */
10566 		goto bailout_unlock;
10567 	}
10568 
10569 	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
10570 			      (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
10571 			      ctl_ooaq, ooa_links))) {
10572 	case CTL_ACTION_BLOCK:
10573 		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
10574 		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
10575 				  blocked_links);
10576 		goto bailout_unlock;
10577 		break; /* NOTREACHED */
10578 	case CTL_ACTION_PASS:
10579 	case CTL_ACTION_SKIP:
10580 		goto queue_rtr;
10581 		break; /* NOTREACHED */
10582 	case CTL_ACTION_OVERLAP:
10583 		ctl_set_overlapped_cmd(ctsio);
10584 		mtx_unlock(&ctl_softc->ctl_lock);
10585 		ctl_done((union ctl_io *)ctsio);
10586 		goto bailout;
10587 		break; /* NOTREACHED */
10588 	case CTL_ACTION_OVERLAP_TAG:
10589 		ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
10590 		mtx_unlock(&ctl_softc->ctl_lock);
10591 		ctl_done((union ctl_io *)ctsio);
10592 		goto bailout;
10593 		break; /* NOTREACHED */
10594 	case CTL_ACTION_ERROR:
10595 	default:
10596 		ctl_set_internal_failure(ctsio,
10597 					 /*sks_valid*/ 0,
10598 					 /*retry_count*/ 0);
10599 		mtx_unlock(&ctl_softc->ctl_lock);
10600 		ctl_done((union ctl_io *)ctsio);
10601 		goto bailout;
10602 		break; /* NOTREACHED */
10603 	}
10604 
10605 	goto bailout_unlock;
10606 
10607 queue_rtr:
10608 	ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
10609 	STAILQ_INSERT_TAIL(&ctl_softc->rtr_queue, &ctsio->io_hdr, links);
10610 
10611 bailout_unlock:
10612 	mtx_unlock(&ctl_softc->ctl_lock);
10613 
10614 bailout:
10615 	return (retval);
10616 }
10617 
10618 static int
10619 ctl_scsiio(struct ctl_scsiio *ctsio)
10620 {
10621 	int retval;
10622 	struct ctl_cmd_entry *entry;
10623 
10624 	retval = CTL_RETVAL_COMPLETE;
10625 
10626 	CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
10627 
10628 	entry = &ctl_cmd_table[ctsio->cdb[0]];
10629 
10630 	/*
10631 	 * If this I/O has been aborted, just send it straight to
10632 	 * ctl_done() without executing it.
10633 	 */
10634 	if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
10635 		ctl_done((union ctl_io *)ctsio);
10636 		goto bailout;
10637 	}
10638 
10639 	/*
10640 	 * All the checks should have been handled by ctl_scsiio_precheck().
10641 	 * We should be clear now to just execute the I/O.
10642 	 */
10643 	retval = entry->execute(ctsio);
10644 
10645 bailout:
10646 	return (retval);
10647 }
10648 
10649 /*
10650  * Since we only implement one target right now, a bus reset simply resets
10651  * our single target.
10652  */
10653 static int
10654 ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io)
10655 {
10656 	return(ctl_target_reset(ctl_softc, io, CTL_UA_BUS_RESET));
10657 }
10658 
10659 static int
10660 ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
10661 		 ctl_ua_type ua_type)
10662 {
10663 	struct ctl_lun *lun;
10664 	int retval;
10665 
10666 	if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
10667 		union ctl_ha_msg msg_info;
10668 
10669 		io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
10670 		msg_info.hdr.nexus = io->io_hdr.nexus;
10671 		if (ua_type==CTL_UA_TARG_RESET)
10672 			msg_info.task.task_action = CTL_TASK_TARGET_RESET;
10673 		else
10674 			msg_info.task.task_action = CTL_TASK_BUS_RESET;
10675 		msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
10676 		msg_info.hdr.original_sc = NULL;
10677 		msg_info.hdr.serializing_sc = NULL;
10678 		if (CTL_HA_STATUS_SUCCESS != ctl_ha_msg_send(CTL_HA_CHAN_CTL,
10679 		    (void *)&msg_info, sizeof(msg_info), 0)) {
10680 		}
10681 	}
10682 	retval = 0;
10683 
10684 	STAILQ_FOREACH(lun, &ctl_softc->lun_list, links)
10685 		retval += ctl_lun_reset(lun, io, ua_type);
10686 
10687 	return (retval);
10688 }
10689 
10690 /*
10691  * The LUN should always be set.  The I/O is optional, and is used to
10692  * distinguish between I/Os sent by this initiator, and by other
10693  * initiators.  We set unit attention for initiators other than this one.
10694  * SAM-3 is vague on this point.  It does say that a unit attention should
10695  * be established for other initiators when a LUN is reset (see section
10696  * 5.7.3), but it doesn't specifically say that the unit attention should
10697  * be established for this particular initiator when a LUN is reset.  Here
10698  * is the relevant text, from SAM-3 rev 8:
10699  *
10700  * 5.7.2 When a SCSI initiator port aborts its own tasks
10701  *
10702  * When a SCSI initiator port causes its own task(s) to be aborted, no
10703  * notification that the task(s) have been aborted shall be returned to
10704  * the SCSI initiator port other than the completion response for the
10705  * command or task management function action that caused the task(s) to
10706  * be aborted and notification(s) associated with related effects of the
10707  * action (e.g., a reset unit attention condition).
10708  *
10709  * XXX KDM for now, we're setting unit attention for all initiators.
10710  */
10711 static int
10712 ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
10713 {
10714 	union ctl_io *xio;
10715 #if 0
10716 	uint32_t initindex;
10717 #endif
10718 	int i;
10719 
10720 	/*
10721 	 * Run through the OOA queue and abort each I/O.
10722 	 */
10723 #if 0
10724 	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
10725 #endif
10726 	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
10727 	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
10728 		xio->io_hdr.flags |= CTL_FLAG_ABORT;
10729 	}
10730 
10731 	/*
10732 	 * This version sets unit attention for every
10733 	 */
10734 #if 0
10735 	initindex = ctl_get_initindex(&io->io_hdr.nexus);
10736 	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
10737 		if (initindex == i)
10738 			continue;
10739 		lun->pending_sense[i].ua_pending |= ua_type;
10740 	}
10741 #endif
10742 
10743 	/*
10744 	 * A reset (any kind, really) clears reservations established with
10745 	 * RESERVE/RELEASE.  It does not clear reservations established
10746 	 * with PERSISTENT RESERVE OUT, but we don't support that at the
10747 	 * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
10748 	 * reservations made with the RESERVE/RELEASE commands, because
10749 	 * those commands are obsolete in SPC-3.
10750 	 */
10751 	lun->flags &= ~CTL_LUN_RESERVED;
10752 
10753 	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
10754 		ctl_clear_mask(lun->have_ca, i);
10755 		lun->pending_sense[i].ua_pending |= ua_type;
10756 	}
10757 
10758 	return (0);
10759 }
10760 
10761 static int
10762 ctl_abort_task(union ctl_io *io)
10763 {
10764 	union ctl_io *xio;
10765 	struct ctl_lun *lun;
10766 	struct ctl_softc *ctl_softc;
10767 #if 0
10768 	struct sbuf sb;
10769 	char printbuf[128];
10770 #endif
10771 	int found;
10772 
10773 	ctl_softc = control_softc;
10774 	found = 0;
10775 
10776 	/*
10777 	 * Look up the LUN.
10778 	 */
10779 	if ((io->io_hdr.nexus.targ_lun < CTL_MAX_LUNS)
10780 	 && (ctl_softc->ctl_luns[io->io_hdr.nexus.targ_lun] != NULL))
10781 		lun = ctl_softc->ctl_luns[io->io_hdr.nexus.targ_lun];
10782 	else
10783 		goto bailout;
10784 
10785 #if 0
10786 	printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
10787 	       lun->lun, io->taskio.tag_num, io->taskio.tag_type);
10788 #endif
10789 
10790 	/*
10791 	 * Run through the OOA queue and attempt to find the given I/O.
10792 	 * The target port, initiator ID, tag type and tag number have to
10793 	 * match the values that we got from the initiator.  If we have an
10794 	 * untagged command to abort, simply abort the first untagged command
10795 	 * we come to.  We only allow one untagged command at a time of course.
10796 	 */
10797 #if 0
10798 	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
10799 #endif
10800 	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
10801 	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
10802 #if 0
10803 		sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
10804 
10805 		sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
10806 			    lun->lun, xio->scsiio.tag_num,
10807 			    xio->scsiio.tag_type,
10808 			    (xio->io_hdr.blocked_links.tqe_prev
10809 			    == NULL) ? "" : " BLOCKED",
10810 			    (xio->io_hdr.flags &
10811 			    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
10812 			    (xio->io_hdr.flags &
10813 			    CTL_FLAG_ABORT) ? " ABORT" : "",
10814 			    (xio->io_hdr.flags &
10815 			    CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
10816 		ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
10817 		sbuf_finish(&sb);
10818 		printf("%s\n", sbuf_data(&sb));
10819 #endif
10820 
10821 		if ((xio->io_hdr.nexus.targ_port == io->io_hdr.nexus.targ_port)
10822 		 && (xio->io_hdr.nexus.initid.id ==
10823 		     io->io_hdr.nexus.initid.id)) {
10824 			/*
10825 			 * If the abort says that the task is untagged, the
10826 			 * task in the queue must be untagged.  Otherwise,
10827 			 * we just check to see whether the tag numbers
10828 			 * match.  This is because the QLogic firmware
10829 			 * doesn't pass back the tag type in an abort
10830 			 * request.
10831 			 */
10832 #if 0
10833 			if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
10834 			  && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
10835 			 || (xio->scsiio.tag_num == io->taskio.tag_num)) {
10836 #endif
10837 			/*
10838 			 * XXX KDM we've got problems with FC, because it
10839 			 * doesn't send down a tag type with aborts.  So we
10840 			 * can only really go by the tag number...
10841 			 * This may cause problems with parallel SCSI.
10842 			 * Need to figure that out!!
10843 			 */
10844 			if (xio->scsiio.tag_num == io->taskio.tag_num) {
10845 				xio->io_hdr.flags |= CTL_FLAG_ABORT;
10846 				found = 1;
10847 				if ((io->io_hdr.flags &
10848 				     CTL_FLAG_FROM_OTHER_SC) == 0 &&
10849 				    !(lun->flags & CTL_LUN_PRIMARY_SC)) {
10850 					union ctl_ha_msg msg_info;
10851 
10852 					io->io_hdr.flags |=
10853 					                CTL_FLAG_SENT_2OTHER_SC;
10854 					msg_info.hdr.nexus = io->io_hdr.nexus;
10855 					msg_info.task.task_action =
10856 						CTL_TASK_ABORT_TASK;
10857 					msg_info.task.tag_num =
10858 						io->taskio.tag_num;
10859 					msg_info.task.tag_type =
10860 						io->taskio.tag_type;
10861 					msg_info.hdr.msg_type =
10862 						CTL_MSG_MANAGE_TASKS;
10863 					msg_info.hdr.original_sc = NULL;
10864 					msg_info.hdr.serializing_sc = NULL;
10865 #if 0
10866 					printf("Sent Abort to other side\n");
10867 #endif
10868 					if (CTL_HA_STATUS_SUCCESS !=
10869 					        ctl_ha_msg_send(CTL_HA_CHAN_CTL,
10870 		    				(void *)&msg_info,
10871 						sizeof(msg_info), 0)) {
10872 					}
10873 				}
10874 #if 0
10875 				printf("ctl_abort_task: found I/O to abort\n");
10876 #endif
10877 				break;
10878 			}
10879 		}
10880 	}
10881 
10882 bailout:
10883 
10884 	if (found == 0) {
10885 		/*
10886 		 * This isn't really an error.  It's entirely possible for
10887 		 * the abort and command completion to cross on the wire.
10888 		 * This is more of an informative/diagnostic error.
10889 		 */
10890 #if 0
10891 		printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
10892 		       "%d:%d:%d:%d tag %d type %d\n",
10893 		       io->io_hdr.nexus.initid.id,
10894 		       io->io_hdr.nexus.targ_port,
10895 		       io->io_hdr.nexus.targ_target.id,
10896 		       io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
10897 		       io->taskio.tag_type);
10898 #endif
10899 		return (1);
10900 	} else
10901 		return (0);
10902 }
10903 
10904 /*
10905  * This routine cannot block!  It must be callable from an interrupt
10906  * handler as well as from the work thread.
10907  */
10908 static void
10909 ctl_run_task_queue(struct ctl_softc *ctl_softc)
10910 {
10911 	union ctl_io *io, *next_io;
10912 
10913 	mtx_assert(&ctl_softc->ctl_lock, MA_OWNED);
10914 
10915 	CTL_DEBUG_PRINT(("ctl_run_task_queue\n"));
10916 
10917 	for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->task_queue);
10918 	     io != NULL; io = next_io) {
10919 		int retval;
10920 		const char *task_desc;
10921 
10922 		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
10923 
10924 		retval = 0;
10925 
10926 		switch (io->io_hdr.io_type) {
10927 		case CTL_IO_TASK: {
10928 			task_desc = ctl_scsi_task_string(&io->taskio);
10929 			if (task_desc != NULL) {
10930 #ifdef NEEDTOPORT
10931 				csevent_log(CSC_CTL | CSC_SHELF_SW |
10932 					    CTL_TASK_REPORT,
10933 					    csevent_LogType_Trace,
10934 					    csevent_Severity_Information,
10935 					    csevent_AlertLevel_Green,
10936 					    csevent_FRU_Firmware,
10937 					    csevent_FRU_Unknown,
10938 					    "CTL: received task: %s",task_desc);
10939 #endif
10940 			} else {
10941 #ifdef NEEDTOPORT
10942 				csevent_log(CSC_CTL | CSC_SHELF_SW |
10943 					    CTL_TASK_REPORT,
10944 					    csevent_LogType_Trace,
10945 					    csevent_Severity_Information,
10946 					    csevent_AlertLevel_Green,
10947 					    csevent_FRU_Firmware,
10948 					    csevent_FRU_Unknown,
10949 					    "CTL: received unknown task "
10950 					    "type: %d (%#x)",
10951 					    io->taskio.task_action,
10952 					    io->taskio.task_action);
10953 #endif
10954 			}
10955 			switch (io->taskio.task_action) {
10956 			case CTL_TASK_ABORT_TASK:
10957 				retval = ctl_abort_task(io);
10958 				break;
10959 			case CTL_TASK_ABORT_TASK_SET:
10960 				break;
10961 			case CTL_TASK_CLEAR_ACA:
10962 				break;
10963 			case CTL_TASK_CLEAR_TASK_SET:
10964 				break;
10965 			case CTL_TASK_LUN_RESET: {
10966 				struct ctl_lun *lun;
10967 				uint32_t targ_lun;
10968 				int retval;
10969 
10970 				targ_lun = io->io_hdr.nexus.targ_lun;
10971 
10972 				if ((targ_lun < CTL_MAX_LUNS)
10973 				 && (ctl_softc->ctl_luns[targ_lun] != NULL))
10974 					lun = ctl_softc->ctl_luns[targ_lun];
10975 				else {
10976 					retval = 1;
10977 					break;
10978 				}
10979 
10980 				if (!(io->io_hdr.flags &
10981 				    CTL_FLAG_FROM_OTHER_SC)) {
10982 					union ctl_ha_msg msg_info;
10983 
10984 					io->io_hdr.flags |=
10985 						CTL_FLAG_SENT_2OTHER_SC;
10986 					msg_info.hdr.msg_type =
10987 						CTL_MSG_MANAGE_TASKS;
10988 					msg_info.hdr.nexus = io->io_hdr.nexus;
10989 					msg_info.task.task_action =
10990 						CTL_TASK_LUN_RESET;
10991 					msg_info.hdr.original_sc = NULL;
10992 					msg_info.hdr.serializing_sc = NULL;
10993 					if (CTL_HA_STATUS_SUCCESS !=
10994 					    ctl_ha_msg_send(CTL_HA_CHAN_CTL,
10995 					    (void *)&msg_info,
10996 					    sizeof(msg_info), 0)) {
10997 					}
10998 				}
10999 
11000 				retval = ctl_lun_reset(lun, io,
11001 						       CTL_UA_LUN_RESET);
11002 				break;
11003 			}
11004 			case CTL_TASK_TARGET_RESET:
11005 				retval = ctl_target_reset(ctl_softc, io,
11006 							  CTL_UA_TARG_RESET);
11007 				break;
11008 			case CTL_TASK_BUS_RESET:
11009 				retval = ctl_bus_reset(ctl_softc, io);
11010 				break;
11011 			case CTL_TASK_PORT_LOGIN:
11012 				break;
11013 			case CTL_TASK_PORT_LOGOUT:
11014 				break;
11015 			default:
11016 				printf("ctl_run_task_queue: got unknown task "
11017 				       "management event %d\n",
11018 				       io->taskio.task_action);
11019 				break;
11020 			}
11021 			if (retval == 0)
11022 				io->io_hdr.status = CTL_SUCCESS;
11023 			else
11024 				io->io_hdr.status = CTL_ERROR;
11025 
11026 			STAILQ_REMOVE(&ctl_softc->task_queue, &io->io_hdr,
11027 				      ctl_io_hdr, links);
11028 			/*
11029 			 * This will queue this I/O to the done queue, but the
11030 			 * work thread won't be able to process it until we
11031 			 * return and the lock is released.
11032 			 */
11033 			ctl_done_lock(io, /*have_lock*/ 1);
11034 			break;
11035 		}
11036 		default: {
11037 
11038 			printf("%s: invalid I/O type %d msg %d cdb %x"
11039 			       " iptl: %ju:%d:%ju:%d tag 0x%04x\n",
11040 			       __func__, io->io_hdr.io_type,
11041 			       io->io_hdr.msg_type, io->scsiio.cdb[0],
11042 			       (uintmax_t)io->io_hdr.nexus.initid.id,
11043 			       io->io_hdr.nexus.targ_port,
11044 			       (uintmax_t)io->io_hdr.nexus.targ_target.id,
11045 			       io->io_hdr.nexus.targ_lun,
11046 			       (io->io_hdr.io_type == CTL_IO_TASK) ?
11047 			       io->taskio.tag_num : io->scsiio.tag_num);
11048 			STAILQ_REMOVE(&ctl_softc->task_queue, &io->io_hdr,
11049 				      ctl_io_hdr, links);
11050 			ctl_free_io_internal(io, 1);
11051 			break;
11052 		}
11053 		}
11054 	}
11055 
11056 	ctl_softc->flags &= ~CTL_FLAG_TASK_PENDING;
11057 }
11058 
11059 /*
11060  * For HA operation.  Handle commands that come in from the other
11061  * controller.
11062  */
11063 static void
11064 ctl_handle_isc(union ctl_io *io)
11065 {
11066 	int free_io;
11067 	struct ctl_lun *lun;
11068 	struct ctl_softc *ctl_softc;
11069 
11070 	ctl_softc = control_softc;
11071 
11072 	lun = ctl_softc->ctl_luns[io->io_hdr.nexus.targ_lun];
11073 
11074 	switch (io->io_hdr.msg_type) {
11075 	case CTL_MSG_SERIALIZE:
11076 		free_io = ctl_serialize_other_sc_cmd(&io->scsiio,
11077 						     /*have_lock*/ 0);
11078 		break;
11079 	case CTL_MSG_R2R: {
11080 		uint8_t opcode;
11081 		struct ctl_cmd_entry *entry;
11082 
11083 		/*
11084 		 * This is only used in SER_ONLY mode.
11085 		 */
11086 		free_io = 0;
11087 		opcode = io->scsiio.cdb[0];
11088 		entry = &ctl_cmd_table[opcode];
11089 		mtx_lock(&ctl_softc->ctl_lock);
11090 		if (ctl_scsiio_lun_check(ctl_softc, lun,
11091 		    entry, (struct ctl_scsiio *)io) != 0) {
11092 			ctl_done_lock(io, /*have_lock*/ 1);
11093 			mtx_unlock(&ctl_softc->ctl_lock);
11094 			break;
11095 		}
11096 		io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11097 		STAILQ_INSERT_TAIL(&ctl_softc->rtr_queue,
11098 				   &io->io_hdr, links);
11099 		mtx_unlock(&ctl_softc->ctl_lock);
11100 		break;
11101 	}
11102 	case CTL_MSG_FINISH_IO:
11103 		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
11104 			free_io = 0;
11105 			ctl_done_lock(io, /*have_lock*/ 0);
11106 		} else {
11107 			free_io = 1;
11108 			mtx_lock(&ctl_softc->ctl_lock);
11109 			TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
11110 				     ooa_links);
11111 			STAILQ_REMOVE(&ctl_softc->task_queue,
11112 				      &io->io_hdr, ctl_io_hdr, links);
11113 			ctl_check_blocked(lun);
11114 			mtx_unlock(&ctl_softc->ctl_lock);
11115 		}
11116 		break;
11117 	case CTL_MSG_PERS_ACTION:
11118 		ctl_hndl_per_res_out_on_other_sc(
11119 			(union ctl_ha_msg *)&io->presio.pr_msg);
11120 		free_io = 1;
11121 		break;
11122 	case CTL_MSG_BAD_JUJU:
11123 		free_io = 0;
11124 		ctl_done_lock(io, /*have_lock*/ 0);
11125 		break;
11126 	case CTL_MSG_DATAMOVE:
11127 		/* Only used in XFER mode */
11128 		free_io = 0;
11129 		ctl_datamove_remote(io);
11130 		break;
11131 	case CTL_MSG_DATAMOVE_DONE:
11132 		/* Only used in XFER mode */
11133 		free_io = 0;
11134 		io->scsiio.be_move_done(io);
11135 		break;
11136 	default:
11137 		free_io = 1;
11138 		printf("%s: Invalid message type %d\n",
11139 		       __func__, io->io_hdr.msg_type);
11140 		break;
11141 	}
11142 	if (free_io)
11143 		ctl_free_io_internal(io, 0);
11144 
11145 }
11146 
11147 
11148 /*
11149  * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
11150  * there is no match.
11151  */
11152 static ctl_lun_error_pattern
11153 ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
11154 {
11155 	struct ctl_cmd_entry *entry;
11156 	ctl_lun_error_pattern filtered_pattern, pattern;
11157 	uint8_t opcode;
11158 
11159 	pattern = desc->error_pattern;
11160 
11161 	/*
11162 	 * XXX KDM we need more data passed into this function to match a
11163 	 * custom pattern, and we actually need to implement custom pattern
11164 	 * matching.
11165 	 */
11166 	if (pattern & CTL_LUN_PAT_CMD)
11167 		return (CTL_LUN_PAT_CMD);
11168 
11169 	if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
11170 		return (CTL_LUN_PAT_ANY);
11171 
11172 	opcode = ctsio->cdb[0];
11173 	entry = &ctl_cmd_table[opcode];
11174 
11175 	filtered_pattern = entry->pattern & pattern;
11176 
11177 	/*
11178 	 * If the user requested specific flags in the pattern (e.g.
11179 	 * CTL_LUN_PAT_RANGE), make sure the command supports all of those
11180 	 * flags.
11181 	 *
11182 	 * If the user did not specify any flags, it doesn't matter whether
11183 	 * or not the command supports the flags.
11184 	 */
11185 	if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
11186 	     (pattern & ~CTL_LUN_PAT_MASK))
11187 		return (CTL_LUN_PAT_NONE);
11188 
11189 	/*
11190 	 * If the user asked for a range check, see if the requested LBA
11191 	 * range overlaps with this command's LBA range.
11192 	 */
11193 	if (filtered_pattern & CTL_LUN_PAT_RANGE) {
11194 		uint64_t lba1;
11195 		uint32_t len1;
11196 		ctl_action action;
11197 		int retval;
11198 
11199 		retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
11200 		if (retval != 0)
11201 			return (CTL_LUN_PAT_NONE);
11202 
11203 		action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
11204 					      desc->lba_range.len);
11205 		/*
11206 		 * A "pass" means that the LBA ranges don't overlap, so
11207 		 * this doesn't match the user's range criteria.
11208 		 */
11209 		if (action == CTL_ACTION_PASS)
11210 			return (CTL_LUN_PAT_NONE);
11211 	}
11212 
11213 	return (filtered_pattern);
11214 }
11215 
11216 static void
11217 ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
11218 {
11219 	struct ctl_error_desc *desc, *desc2;
11220 
11221 	mtx_assert(&control_softc->ctl_lock, MA_OWNED);
11222 
11223 	STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
11224 		ctl_lun_error_pattern pattern;
11225 		/*
11226 		 * Check to see whether this particular command matches
11227 		 * the pattern in the descriptor.
11228 		 */
11229 		pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
11230 		if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
11231 			continue;
11232 
11233 		switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
11234 		case CTL_LUN_INJ_ABORTED:
11235 			ctl_set_aborted(&io->scsiio);
11236 			break;
11237 		case CTL_LUN_INJ_MEDIUM_ERR:
11238 			ctl_set_medium_error(&io->scsiio);
11239 			break;
11240 		case CTL_LUN_INJ_UA:
11241 			/* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
11242 			 * OCCURRED */
11243 			ctl_set_ua(&io->scsiio, 0x29, 0x00);
11244 			break;
11245 		case CTL_LUN_INJ_CUSTOM:
11246 			/*
11247 			 * We're assuming the user knows what he is doing.
11248 			 * Just copy the sense information without doing
11249 			 * checks.
11250 			 */
11251 			bcopy(&desc->custom_sense, &io->scsiio.sense_data,
11252 			      ctl_min(sizeof(desc->custom_sense),
11253 				      sizeof(io->scsiio.sense_data)));
11254 			io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
11255 			io->scsiio.sense_len = SSD_FULL_SIZE;
11256 			io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
11257 			break;
11258 		case CTL_LUN_INJ_NONE:
11259 		default:
11260 			/*
11261 			 * If this is an error injection type we don't know
11262 			 * about, clear the continuous flag (if it is set)
11263 			 * so it will get deleted below.
11264 			 */
11265 			desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
11266 			break;
11267 		}
11268 		/*
11269 		 * By default, each error injection action is a one-shot
11270 		 */
11271 		if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
11272 			continue;
11273 
11274 		STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
11275 
11276 		free(desc, M_CTL);
11277 	}
11278 }
11279 
11280 #ifdef CTL_IO_DELAY
11281 static void
11282 ctl_datamove_timer_wakeup(void *arg)
11283 {
11284 	union ctl_io *io;
11285 
11286 	io = (union ctl_io *)arg;
11287 
11288 	ctl_datamove(io);
11289 }
11290 #endif /* CTL_IO_DELAY */
11291 
11292 void
11293 ctl_datamove(union ctl_io *io)
11294 {
11295 	void (*fe_datamove)(union ctl_io *io);
11296 
11297 	mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
11298 
11299 	CTL_DEBUG_PRINT(("ctl_datamove\n"));
11300 
11301 #ifdef CTL_TIME_IO
11302 	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
11303 		char str[256];
11304 		char path_str[64];
11305 		struct sbuf sb;
11306 
11307 		ctl_scsi_path_string(io, path_str, sizeof(path_str));
11308 		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
11309 
11310 		sbuf_cat(&sb, path_str);
11311 		switch (io->io_hdr.io_type) {
11312 		case CTL_IO_SCSI:
11313 			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
11314 			sbuf_printf(&sb, "\n");
11315 			sbuf_cat(&sb, path_str);
11316 			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
11317 				    io->scsiio.tag_num, io->scsiio.tag_type);
11318 			break;
11319 		case CTL_IO_TASK:
11320 			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
11321 				    "Tag Type: %d\n", io->taskio.task_action,
11322 				    io->taskio.tag_num, io->taskio.tag_type);
11323 			break;
11324 		default:
11325 			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
11326 			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
11327 			break;
11328 		}
11329 		sbuf_cat(&sb, path_str);
11330 		sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
11331 			    (intmax_t)time_uptime - io->io_hdr.start_time);
11332 		sbuf_finish(&sb);
11333 		printf("%s", sbuf_data(&sb));
11334 	}
11335 #endif /* CTL_TIME_IO */
11336 
11337 	mtx_lock(&control_softc->ctl_lock);
11338 #ifdef CTL_IO_DELAY
11339 	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
11340 		struct ctl_lun *lun;
11341 
11342 		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
11343 
11344 		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
11345 	} else {
11346 		struct ctl_lun *lun;
11347 
11348 		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
11349 		if ((lun != NULL)
11350 		 && (lun->delay_info.datamove_delay > 0)) {
11351 			struct callout *callout;
11352 
11353 			callout = (struct callout *)&io->io_hdr.timer_bytes;
11354 			callout_init(callout, /*mpsafe*/ 1);
11355 			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
11356 			callout_reset(callout,
11357 				      lun->delay_info.datamove_delay * hz,
11358 				      ctl_datamove_timer_wakeup, io);
11359 			if (lun->delay_info.datamove_type ==
11360 			    CTL_DELAY_TYPE_ONESHOT)
11361 				lun->delay_info.datamove_delay = 0;
11362 			mtx_unlock(&control_softc->ctl_lock);
11363 			return;
11364 		}
11365 	}
11366 #endif
11367 	/*
11368 	 * If we have any pending task management commands, process them
11369 	 * first.  This is necessary to eliminate a race condition with the
11370 	 * FETD:
11371 	 *
11372 	 * - FETD submits a task management command, like an abort.
11373 	 * - Back end calls fe_datamove() to move the data for the aborted
11374 	 *   command.  The FETD can't really accept it, but if it did, it
11375 	 *   would end up transmitting data for a command that the initiator
11376 	 *   told us to abort.
11377 	 *
11378 	 * We close the race by processing all pending task management
11379 	 * commands here (we can't block!), and then check this I/O to see
11380 	 * if it has been aborted.  If so, return it to the back end with
11381 	 * bad status, so the back end can say return an error to the back end
11382 	 * and then when the back end returns an error, we can return the
11383 	 * aborted command to the FETD, so it can clean up its resources.
11384 	 */
11385 	if (control_softc->flags & CTL_FLAG_TASK_PENDING)
11386 		ctl_run_task_queue(control_softc);
11387 
11388 	/*
11389 	 * This command has been aborted.  Set the port status, so we fail
11390 	 * the data move.
11391 	 */
11392 	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
11393 		printf("ctl_datamove: tag 0x%04x on (%ju:%d:%ju:%d) aborted\n",
11394 		       io->scsiio.tag_num,(uintmax_t)io->io_hdr.nexus.initid.id,
11395 		       io->io_hdr.nexus.targ_port,
11396 		       (uintmax_t)io->io_hdr.nexus.targ_target.id,
11397 		       io->io_hdr.nexus.targ_lun);
11398 		io->io_hdr.status = CTL_CMD_ABORTED;
11399 		io->io_hdr.port_status = 31337;
11400 		mtx_unlock(&control_softc->ctl_lock);
11401 		/*
11402 		 * Note that the backend, in this case, will get the
11403 		 * callback in its context.  In other cases it may get
11404 		 * called in the frontend's interrupt thread context.
11405 		 */
11406 		io->scsiio.be_move_done(io);
11407 		return;
11408 	}
11409 
11410 	/*
11411 	 * If we're in XFER mode and this I/O is from the other shelf
11412 	 * controller, we need to send the DMA to the other side to
11413 	 * actually transfer the data to/from the host.  In serialize only
11414 	 * mode the transfer happens below CTL and ctl_datamove() is only
11415 	 * called on the machine that originally received the I/O.
11416 	 */
11417 	if ((control_softc->ha_mode == CTL_HA_MODE_XFER)
11418 	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11419 		union ctl_ha_msg msg;
11420 		uint32_t sg_entries_sent;
11421 		int do_sg_copy;
11422 		int i;
11423 
11424 		memset(&msg, 0, sizeof(msg));
11425 		msg.hdr.msg_type = CTL_MSG_DATAMOVE;
11426 		msg.hdr.original_sc = io->io_hdr.original_sc;
11427 		msg.hdr.serializing_sc = io;
11428 		msg.hdr.nexus = io->io_hdr.nexus;
11429 		msg.dt.flags = io->io_hdr.flags;
11430 		/*
11431 		 * We convert everything into a S/G list here.  We can't
11432 		 * pass by reference, only by value between controllers.
11433 		 * So we can't pass a pointer to the S/G list, only as many
11434 		 * S/G entries as we can fit in here.  If it's possible for
11435 		 * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
11436 		 * then we need to break this up into multiple transfers.
11437 		 */
11438 		if (io->scsiio.kern_sg_entries == 0) {
11439 			msg.dt.kern_sg_entries = 1;
11440 			/*
11441 			 * If this is in cached memory, flush the cache
11442 			 * before we send the DMA request to the other
11443 			 * controller.  We want to do this in either the
11444 			 * read or the write case.  The read case is
11445 			 * straightforward.  In the write case, we want to
11446 			 * make sure nothing is in the local cache that
11447 			 * could overwrite the DMAed data.
11448 			 */
11449 			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
11450 				/*
11451 				 * XXX KDM use bus_dmamap_sync() here.
11452 				 */
11453 			}
11454 
11455 			/*
11456 			 * Convert to a physical address if this is a
11457 			 * virtual address.
11458 			 */
11459 			if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
11460 				msg.dt.sg_list[0].addr =
11461 					io->scsiio.kern_data_ptr;
11462 			} else {
11463 				/*
11464 				 * XXX KDM use busdma here!
11465 				 */
11466 #if 0
11467 				msg.dt.sg_list[0].addr = (void *)
11468 					vtophys(io->scsiio.kern_data_ptr);
11469 #endif
11470 			}
11471 
11472 			msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
11473 			do_sg_copy = 0;
11474 		} else {
11475 			struct ctl_sg_entry *sgl;
11476 
11477 			do_sg_copy = 1;
11478 			msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
11479 			sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
11480 			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
11481 				/*
11482 				 * XXX KDM use bus_dmamap_sync() here.
11483 				 */
11484 			}
11485 		}
11486 
11487 		msg.dt.kern_data_len = io->scsiio.kern_data_len;
11488 		msg.dt.kern_total_len = io->scsiio.kern_total_len;
11489 		msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
11490 		msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
11491 		msg.dt.sg_sequence = 0;
11492 
11493 		/*
11494 		 * Loop until we've sent all of the S/G entries.  On the
11495 		 * other end, we'll recompose these S/G entries into one
11496 		 * contiguous list before passing it to the
11497 		 */
11498 		for (sg_entries_sent = 0; sg_entries_sent <
11499 		     msg.dt.kern_sg_entries; msg.dt.sg_sequence++) {
11500 			msg.dt.cur_sg_entries = ctl_min((sizeof(msg.dt.sg_list)/
11501 				sizeof(msg.dt.sg_list[0])),
11502 				msg.dt.kern_sg_entries - sg_entries_sent);
11503 
11504 			if (do_sg_copy != 0) {
11505 				struct ctl_sg_entry *sgl;
11506 				int j;
11507 
11508 				sgl = (struct ctl_sg_entry *)
11509 					io->scsiio.kern_data_ptr;
11510 				/*
11511 				 * If this is in cached memory, flush the cache
11512 				 * before we send the DMA request to the other
11513 				 * controller.  We want to do this in either
11514 				 * the * read or the write case.  The read
11515 				 * case is straightforward.  In the write
11516 				 * case, we want to make sure nothing is
11517 				 * in the local cache that could overwrite
11518 				 * the DMAed data.
11519 				 */
11520 
11521 				for (i = sg_entries_sent, j = 0;
11522 				     i < msg.dt.cur_sg_entries; i++, j++) {
11523 					if ((io->io_hdr.flags &
11524 					     CTL_FLAG_NO_DATASYNC) == 0) {
11525 						/*
11526 						 * XXX KDM use bus_dmamap_sync()
11527 						 */
11528 					}
11529 					if ((io->io_hdr.flags &
11530 					     CTL_FLAG_BUS_ADDR) == 0) {
11531 						/*
11532 						 * XXX KDM use busdma.
11533 						 */
11534 #if 0
11535 						msg.dt.sg_list[j].addr =(void *)
11536 						       vtophys(sgl[i].addr);
11537 #endif
11538 					} else {
11539 						msg.dt.sg_list[j].addr =
11540 							sgl[i].addr;
11541 					}
11542 					msg.dt.sg_list[j].len = sgl[i].len;
11543 				}
11544 			}
11545 
11546 			sg_entries_sent += msg.dt.cur_sg_entries;
11547 			if (sg_entries_sent >= msg.dt.kern_sg_entries)
11548 				msg.dt.sg_last = 1;
11549 			else
11550 				msg.dt.sg_last = 0;
11551 
11552 			/*
11553 			 * XXX KDM drop and reacquire the lock here?
11554 			 */
11555 			if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
11556 			    sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
11557 				/*
11558 				 * XXX do something here.
11559 				 */
11560 			}
11561 
11562 			msg.dt.sent_sg_entries = sg_entries_sent;
11563 		}
11564 		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11565 		if (io->io_hdr.flags & CTL_FLAG_FAILOVER)
11566 			ctl_failover_io(io, /*have_lock*/ 1);
11567 
11568 	} else {
11569 
11570 		/*
11571 		 * Lookup the fe_datamove() function for this particular
11572 		 * front end.
11573 		 */
11574 		fe_datamove =
11575 		    control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
11576 		mtx_unlock(&control_softc->ctl_lock);
11577 
11578 		fe_datamove(io);
11579 	}
11580 }
11581 
11582 static void
11583 ctl_send_datamove_done(union ctl_io *io, int have_lock)
11584 {
11585 	union ctl_ha_msg msg;
11586 	int isc_status;
11587 
11588 	memset(&msg, 0, sizeof(msg));
11589 
11590 	msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
11591 	msg.hdr.original_sc = io;
11592 	msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
11593 	msg.hdr.nexus = io->io_hdr.nexus;
11594 	msg.hdr.status = io->io_hdr.status;
11595 	msg.scsi.tag_num = io->scsiio.tag_num;
11596 	msg.scsi.tag_type = io->scsiio.tag_type;
11597 	msg.scsi.scsi_status = io->scsiio.scsi_status;
11598 	memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
11599 	       sizeof(io->scsiio.sense_data));
11600 	msg.scsi.sense_len = io->scsiio.sense_len;
11601 	msg.scsi.sense_residual = io->scsiio.sense_residual;
11602 	msg.scsi.fetd_status = io->io_hdr.port_status;
11603 	msg.scsi.residual = io->scsiio.residual;
11604 	io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11605 
11606 	if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
11607 		ctl_failover_io(io, /*have_lock*/ have_lock);
11608 		return;
11609 	}
11610 
11611 	isc_status = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0);
11612 	if (isc_status > CTL_HA_STATUS_SUCCESS) {
11613 		/* XXX do something if this fails */
11614 	}
11615 
11616 }
11617 
11618 /*
11619  * The DMA to the remote side is done, now we need to tell the other side
11620  * we're done so it can continue with its data movement.
11621  */
11622 static void
11623 ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
11624 {
11625 	union ctl_io *io;
11626 
11627 	io = rq->context;
11628 
11629 	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
11630 		printf("%s: ISC DMA write failed with error %d", __func__,
11631 		       rq->ret);
11632 		ctl_set_internal_failure(&io->scsiio,
11633 					 /*sks_valid*/ 1,
11634 					 /*retry_count*/ rq->ret);
11635 	}
11636 
11637 	ctl_dt_req_free(rq);
11638 
11639 	/*
11640 	 * In this case, we had to malloc the memory locally.  Free it.
11641 	 */
11642 	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
11643 		int i;
11644 		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
11645 			free(io->io_hdr.local_sglist[i].addr, M_CTL);
11646 	}
11647 	/*
11648 	 * The data is in local and remote memory, so now we need to send
11649 	 * status (good or back) back to the other side.
11650 	 */
11651 	ctl_send_datamove_done(io, /*have_lock*/ 0);
11652 }
11653 
11654 /*
11655  * We've moved the data from the host/controller into local memory.  Now we
11656  * need to push it over to the remote controller's memory.
11657  */
11658 static int
11659 ctl_datamove_remote_dm_write_cb(union ctl_io *io)
11660 {
11661 	int retval;
11662 
11663 	retval = 0;
11664 
11665 	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
11666 					  ctl_datamove_remote_write_cb);
11667 
11668 	return (retval);
11669 }
11670 
11671 static void
11672 ctl_datamove_remote_write(union ctl_io *io)
11673 {
11674 	int retval;
11675 	void (*fe_datamove)(union ctl_io *io);
11676 
11677 	/*
11678 	 * - Get the data from the host/HBA into local memory.
11679 	 * - DMA memory from the local controller to the remote controller.
11680 	 * - Send status back to the remote controller.
11681 	 */
11682 
11683 	retval = ctl_datamove_remote_sgl_setup(io);
11684 	if (retval != 0)
11685 		return;
11686 
11687 	/* Switch the pointer over so the FETD knows what to do */
11688 	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
11689 
11690 	/*
11691 	 * Use a custom move done callback, since we need to send completion
11692 	 * back to the other controller, not to the backend on this side.
11693 	 */
11694 	io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
11695 
11696 	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
11697 
11698 	fe_datamove(io);
11699 
11700 	return;
11701 
11702 }
11703 
11704 static int
11705 ctl_datamove_remote_dm_read_cb(union ctl_io *io)
11706 {
11707 #if 0
11708 	char str[256];
11709 	char path_str[64];
11710 	struct sbuf sb;
11711 #endif
11712 
11713 	/*
11714 	 * In this case, we had to malloc the memory locally.  Free it.
11715 	 */
11716 	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
11717 		int i;
11718 		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
11719 			free(io->io_hdr.local_sglist[i].addr, M_CTL);
11720 	}
11721 
11722 #if 0
11723 	scsi_path_string(io, path_str, sizeof(path_str));
11724 	sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
11725 	sbuf_cat(&sb, path_str);
11726 	scsi_command_string(&io->scsiio, NULL, &sb);
11727 	sbuf_printf(&sb, "\n");
11728 	sbuf_cat(&sb, path_str);
11729 	sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
11730 		    io->scsiio.tag_num, io->scsiio.tag_type);
11731 	sbuf_cat(&sb, path_str);
11732 	sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
11733 		    io->io_hdr.flags, io->io_hdr.status);
11734 	sbuf_finish(&sb);
11735 	printk("%s", sbuf_data(&sb));
11736 #endif
11737 
11738 
11739 	/*
11740 	 * The read is done, now we need to send status (good or bad) back
11741 	 * to the other side.
11742 	 */
11743 	ctl_send_datamove_done(io, /*have_lock*/ 0);
11744 
11745 	return (0);
11746 }
11747 
11748 static void
11749 ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
11750 {
11751 	union ctl_io *io;
11752 	void (*fe_datamove)(union ctl_io *io);
11753 
11754 	io = rq->context;
11755 
11756 	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
11757 		printf("%s: ISC DMA read failed with error %d", __func__,
11758 		       rq->ret);
11759 		ctl_set_internal_failure(&io->scsiio,
11760 					 /*sks_valid*/ 1,
11761 					 /*retry_count*/ rq->ret);
11762 	}
11763 
11764 	ctl_dt_req_free(rq);
11765 
11766 	/* Switch the pointer over so the FETD knows what to do */
11767 	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
11768 
11769 	/*
11770 	 * Use a custom move done callback, since we need to send completion
11771 	 * back to the other controller, not to the backend on this side.
11772 	 */
11773 	io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
11774 
11775 	/* XXX KDM add checks like the ones in ctl_datamove? */
11776 
11777 	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
11778 
11779 	fe_datamove(io);
11780 }
11781 
11782 static int
11783 ctl_datamove_remote_sgl_setup(union ctl_io *io)
11784 {
11785 	struct ctl_sg_entry *local_sglist, *remote_sglist;
11786 	struct ctl_sg_entry *local_dma_sglist, *remote_dma_sglist;
11787 	struct ctl_softc *softc;
11788 	int retval;
11789 	int i;
11790 
11791 	retval = 0;
11792 	softc = control_softc;
11793 
11794 	local_sglist = io->io_hdr.local_sglist;
11795 	local_dma_sglist = io->io_hdr.local_dma_sglist;
11796 	remote_sglist = io->io_hdr.remote_sglist;
11797 	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
11798 
11799 	if (io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) {
11800 		for (i = 0; i < io->scsiio.kern_sg_entries; i++) {
11801 			local_sglist[i].len = remote_sglist[i].len;
11802 
11803 			/*
11804 			 * XXX Detect the situation where the RS-level I/O
11805 			 * redirector on the other side has already read the
11806 			 * data off of the AOR RS on this side, and
11807 			 * transferred it to remote (mirror) memory on the
11808 			 * other side.  Since we already have the data in
11809 			 * memory here, we just need to use it.
11810 			 *
11811 			 * XXX KDM this can probably be removed once we
11812 			 * get the cache device code in and take the
11813 			 * current AOR implementation out.
11814 			 */
11815 #ifdef NEEDTOPORT
11816 			if ((remote_sglist[i].addr >=
11817 			     (void *)vtophys(softc->mirr->addr))
11818 			 && (remote_sglist[i].addr <
11819 			     ((void *)vtophys(softc->mirr->addr) +
11820 			     CacheMirrorOffset))) {
11821 				local_sglist[i].addr = remote_sglist[i].addr -
11822 					CacheMirrorOffset;
11823 				if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
11824 				     CTL_FLAG_DATA_IN)
11825 					io->io_hdr.flags |= CTL_FLAG_REDIR_DONE;
11826 			} else {
11827 				local_sglist[i].addr = remote_sglist[i].addr +
11828 					CacheMirrorOffset;
11829 			}
11830 #endif
11831 #if 0
11832 			printf("%s: local %p, remote %p, len %d\n",
11833 			       __func__, local_sglist[i].addr,
11834 			       remote_sglist[i].addr, local_sglist[i].len);
11835 #endif
11836 		}
11837 	} else {
11838 		uint32_t len_to_go;
11839 
11840 		/*
11841 		 * In this case, we don't have automatically allocated
11842 		 * memory for this I/O on this controller.  This typically
11843 		 * happens with internal CTL I/O -- e.g. inquiry, mode
11844 		 * sense, etc.  Anything coming from RAIDCore will have
11845 		 * a mirror area available.
11846 		 */
11847 		len_to_go = io->scsiio.kern_data_len;
11848 
11849 		/*
11850 		 * Clear the no datasync flag, we have to use malloced
11851 		 * buffers.
11852 		 */
11853 		io->io_hdr.flags &= ~CTL_FLAG_NO_DATASYNC;
11854 
11855 		/*
11856 		 * The difficult thing here is that the size of the various
11857 		 * S/G segments may be different than the size from the
11858 		 * remote controller.  That'll make it harder when DMAing
11859 		 * the data back to the other side.
11860 		 */
11861 		for (i = 0; (i < sizeof(io->io_hdr.remote_sglist) /
11862 		     sizeof(io->io_hdr.remote_sglist[0])) &&
11863 		     (len_to_go > 0); i++) {
11864 			local_sglist[i].len = ctl_min(len_to_go, 131072);
11865 			CTL_SIZE_8B(local_dma_sglist[i].len,
11866 				    local_sglist[i].len);
11867 			local_sglist[i].addr =
11868 				malloc(local_dma_sglist[i].len, M_CTL,M_WAITOK);
11869 
11870 			local_dma_sglist[i].addr = local_sglist[i].addr;
11871 
11872 			if (local_sglist[i].addr == NULL) {
11873 				int j;
11874 
11875 				printf("malloc failed for %zd bytes!",
11876 				       local_dma_sglist[i].len);
11877 				for (j = 0; j < i; j++) {
11878 					free(local_sglist[j].addr, M_CTL);
11879 				}
11880 				ctl_set_internal_failure(&io->scsiio,
11881 							 /*sks_valid*/ 1,
11882 							 /*retry_count*/ 4857);
11883 				retval = 1;
11884 				goto bailout_error;
11885 
11886 			}
11887 			/* XXX KDM do we need a sync here? */
11888 
11889 			len_to_go -= local_sglist[i].len;
11890 		}
11891 		/*
11892 		 * Reset the number of S/G entries accordingly.  The
11893 		 * original number of S/G entries is available in
11894 		 * rem_sg_entries.
11895 		 */
11896 		io->scsiio.kern_sg_entries = i;
11897 
11898 #if 0
11899 		printf("%s: kern_sg_entries = %d\n", __func__,
11900 		       io->scsiio.kern_sg_entries);
11901 		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
11902 			printf("%s: sg[%d] = %p, %d (DMA: %d)\n", __func__, i,
11903 			       local_sglist[i].addr, local_sglist[i].len,
11904 			       local_dma_sglist[i].len);
11905 #endif
11906 	}
11907 
11908 
11909 	return (retval);
11910 
11911 bailout_error:
11912 
11913 	ctl_send_datamove_done(io, /*have_lock*/ 0);
11914 
11915 	return (retval);
11916 }
11917 
11918 static int
11919 ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
11920 			 ctl_ha_dt_cb callback)
11921 {
11922 	struct ctl_ha_dt_req *rq;
11923 	struct ctl_sg_entry *remote_sglist, *local_sglist;
11924 	struct ctl_sg_entry *remote_dma_sglist, *local_dma_sglist;
11925 	uint32_t local_used, remote_used, total_used;
11926 	int retval;
11927 	int i, j;
11928 
11929 	retval = 0;
11930 
11931 	rq = ctl_dt_req_alloc();
11932 
11933 	/*
11934 	 * If we failed to allocate the request, and if the DMA didn't fail
11935 	 * anyway, set busy status.  This is just a resource allocation
11936 	 * failure.
11937 	 */
11938 	if ((rq == NULL)
11939 	 && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE))
11940 		ctl_set_busy(&io->scsiio);
11941 
11942 	if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) {
11943 
11944 		if (rq != NULL)
11945 			ctl_dt_req_free(rq);
11946 
11947 		/*
11948 		 * The data move failed.  We need to return status back
11949 		 * to the other controller.  No point in trying to DMA
11950 		 * data to the remote controller.
11951 		 */
11952 
11953 		ctl_send_datamove_done(io, /*have_lock*/ 0);
11954 
11955 		retval = 1;
11956 
11957 		goto bailout;
11958 	}
11959 
11960 	local_sglist = io->io_hdr.local_sglist;
11961 	local_dma_sglist = io->io_hdr.local_dma_sglist;
11962 	remote_sglist = io->io_hdr.remote_sglist;
11963 	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
11964 	local_used = 0;
11965 	remote_used = 0;
11966 	total_used = 0;
11967 
11968 	if (io->io_hdr.flags & CTL_FLAG_REDIR_DONE) {
11969 		rq->ret = CTL_HA_STATUS_SUCCESS;
11970 		rq->context = io;
11971 		callback(rq);
11972 		goto bailout;
11973 	}
11974 
11975 	/*
11976 	 * Pull/push the data over the wire from/to the other controller.
11977 	 * This takes into account the possibility that the local and
11978 	 * remote sglists may not be identical in terms of the size of
11979 	 * the elements and the number of elements.
11980 	 *
11981 	 * One fundamental assumption here is that the length allocated for
11982 	 * both the local and remote sglists is identical.  Otherwise, we've
11983 	 * essentially got a coding error of some sort.
11984 	 */
11985 	for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
11986 		int isc_ret;
11987 		uint32_t cur_len, dma_length;
11988 		uint8_t *tmp_ptr;
11989 
11990 		rq->id = CTL_HA_DATA_CTL;
11991 		rq->command = command;
11992 		rq->context = io;
11993 
11994 		/*
11995 		 * Both pointers should be aligned.  But it is possible
11996 		 * that the allocation length is not.  They should both
11997 		 * also have enough slack left over at the end, though,
11998 		 * to round up to the next 8 byte boundary.
11999 		 */
12000 		cur_len = ctl_min(local_sglist[i].len - local_used,
12001 				  remote_sglist[j].len - remote_used);
12002 
12003 		/*
12004 		 * In this case, we have a size issue and need to decrease
12005 		 * the size, except in the case where we actually have less
12006 		 * than 8 bytes left.  In that case, we need to increase
12007 		 * the DMA length to get the last bit.
12008 		 */
12009 		if ((cur_len & 0x7) != 0) {
12010 			if (cur_len > 0x7) {
12011 				cur_len = cur_len - (cur_len & 0x7);
12012 				dma_length = cur_len;
12013 			} else {
12014 				CTL_SIZE_8B(dma_length, cur_len);
12015 			}
12016 
12017 		} else
12018 			dma_length = cur_len;
12019 
12020 		/*
12021 		 * If we had to allocate memory for this I/O, instead of using
12022 		 * the non-cached mirror memory, we'll need to flush the cache
12023 		 * before trying to DMA to the other controller.
12024 		 *
12025 		 * We could end up doing this multiple times for the same
12026 		 * segment if we have a larger local segment than remote
12027 		 * segment.  That shouldn't be an issue.
12028 		 */
12029 		if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12030 			/*
12031 			 * XXX KDM use bus_dmamap_sync() here.
12032 			 */
12033 		}
12034 
12035 		rq->size = dma_length;
12036 
12037 		tmp_ptr = (uint8_t *)local_sglist[i].addr;
12038 		tmp_ptr += local_used;
12039 
12040 		/* Use physical addresses when talking to ISC hardware */
12041 		if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
12042 			/* XXX KDM use busdma */
12043 #if 0
12044 			rq->local = vtophys(tmp_ptr);
12045 #endif
12046 		} else
12047 			rq->local = tmp_ptr;
12048 
12049 		tmp_ptr = (uint8_t *)remote_sglist[j].addr;
12050 		tmp_ptr += remote_used;
12051 		rq->remote = tmp_ptr;
12052 
12053 		rq->callback = NULL;
12054 
12055 		local_used += cur_len;
12056 		if (local_used >= local_sglist[i].len) {
12057 			i++;
12058 			local_used = 0;
12059 		}
12060 
12061 		remote_used += cur_len;
12062 		if (remote_used >= remote_sglist[j].len) {
12063 			j++;
12064 			remote_used = 0;
12065 		}
12066 		total_used += cur_len;
12067 
12068 		if (total_used >= io->scsiio.kern_data_len)
12069 			rq->callback = callback;
12070 
12071 		if ((rq->size & 0x7) != 0) {
12072 			printf("%s: warning: size %d is not on 8b boundary\n",
12073 			       __func__, rq->size);
12074 		}
12075 		if (((uintptr_t)rq->local & 0x7) != 0) {
12076 			printf("%s: warning: local %p not on 8b boundary\n",
12077 			       __func__, rq->local);
12078 		}
12079 		if (((uintptr_t)rq->remote & 0x7) != 0) {
12080 			printf("%s: warning: remote %p not on 8b boundary\n",
12081 			       __func__, rq->local);
12082 		}
12083 #if 0
12084 		printf("%s: %s: local %#x remote %#x size %d\n", __func__,
12085 		       (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
12086 		       rq->local, rq->remote, rq->size);
12087 #endif
12088 
12089 		isc_ret = ctl_dt_single(rq);
12090 		if (isc_ret == CTL_HA_STATUS_WAIT)
12091 			continue;
12092 
12093 		if (isc_ret == CTL_HA_STATUS_DISCONNECT) {
12094 			rq->ret = CTL_HA_STATUS_SUCCESS;
12095 		} else {
12096 			rq->ret = isc_ret;
12097 		}
12098 		callback(rq);
12099 		goto bailout;
12100 	}
12101 
12102 bailout:
12103 	return (retval);
12104 
12105 }
12106 
12107 static void
12108 ctl_datamove_remote_read(union ctl_io *io)
12109 {
12110 	int retval;
12111 	int i;
12112 
12113 	/*
12114 	 * This will send an error to the other controller in the case of a
12115 	 * failure.
12116 	 */
12117 	retval = ctl_datamove_remote_sgl_setup(io);
12118 	if (retval != 0)
12119 		return;
12120 
12121 	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
12122 					  ctl_datamove_remote_read_cb);
12123 	if ((retval != 0)
12124 	 && ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0)) {
12125 		/*
12126 		 * Make sure we free memory if there was an error..  The
12127 		 * ctl_datamove_remote_xfer() function will send the
12128 		 * datamove done message, or call the callback with an
12129 		 * error if there is a problem.
12130 		 */
12131 		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12132 			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12133 	}
12134 
12135 	return;
12136 }
12137 
12138 /*
12139  * Process a datamove request from the other controller.  This is used for
12140  * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
12141  * first.  Once that is complete, the data gets DMAed into the remote
12142  * controller's memory.  For reads, we DMA from the remote controller's
12143  * memory into our memory first, and then move it out to the FETD.
12144  */
12145 static void
12146 ctl_datamove_remote(union ctl_io *io)
12147 {
12148 	struct ctl_softc *softc;
12149 
12150 	softc = control_softc;
12151 
12152 	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
12153 
12154 	/*
12155 	 * Note that we look for an aborted I/O here, but don't do some of
12156 	 * the other checks that ctl_datamove() normally does.  We don't
12157 	 * need to run the task queue, because this I/O is on the ISC
12158 	 * queue, which is executed by the work thread after the task queue.
12159 	 * We don't need to run the datamove delay code, since that should
12160 	 * have been done if need be on the other controller.
12161 	 */
12162 	mtx_lock(&softc->ctl_lock);
12163 
12164 	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12165 
12166 		printf("%s: tag 0x%04x on (%d:%d:%d:%d) aborted\n", __func__,
12167 		       io->scsiio.tag_num, io->io_hdr.nexus.initid.id,
12168 		       io->io_hdr.nexus.targ_port,
12169 		       io->io_hdr.nexus.targ_target.id,
12170 		       io->io_hdr.nexus.targ_lun);
12171 		io->io_hdr.status = CTL_CMD_ABORTED;
12172 		io->io_hdr.port_status = 31338;
12173 
12174 		mtx_unlock(&softc->ctl_lock);
12175 
12176 		ctl_send_datamove_done(io, /*have_lock*/ 0);
12177 
12178 		return;
12179 	}
12180 
12181 	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT) {
12182 		mtx_unlock(&softc->ctl_lock);
12183 		ctl_datamove_remote_write(io);
12184 	} else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN){
12185 		mtx_unlock(&softc->ctl_lock);
12186 		ctl_datamove_remote_read(io);
12187 	} else {
12188 		union ctl_ha_msg msg;
12189 		struct scsi_sense_data *sense;
12190 		uint8_t sks[3];
12191 		int retry_count;
12192 
12193 		memset(&msg, 0, sizeof(msg));
12194 
12195 		msg.hdr.msg_type = CTL_MSG_BAD_JUJU;
12196 		msg.hdr.status = CTL_SCSI_ERROR;
12197 		msg.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
12198 
12199 		retry_count = 4243;
12200 
12201 		sense = &msg.scsi.sense_data;
12202 		sks[0] = SSD_SCS_VALID;
12203 		sks[1] = (retry_count >> 8) & 0xff;
12204 		sks[2] = retry_count & 0xff;
12205 
12206 		/* "Internal target failure" */
12207 		scsi_set_sense_data(sense,
12208 				    /*sense_format*/ SSD_TYPE_NONE,
12209 				    /*current_error*/ 1,
12210 				    /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
12211 				    /*asc*/ 0x44,
12212 				    /*ascq*/ 0x00,
12213 				    /*type*/ SSD_ELEM_SKS,
12214 				    /*size*/ sizeof(sks),
12215 				    /*data*/ sks,
12216 				    SSD_ELEM_NONE);
12217 
12218 		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12219 		if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12220 			ctl_failover_io(io, /*have_lock*/ 1);
12221 			mtx_unlock(&softc->ctl_lock);
12222 			return;
12223 		}
12224 
12225 		mtx_unlock(&softc->ctl_lock);
12226 
12227 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0) >
12228 		    CTL_HA_STATUS_SUCCESS) {
12229 			/* XXX KDM what to do if this fails? */
12230 		}
12231 		return;
12232 	}
12233 
12234 }
12235 
12236 static int
12237 ctl_process_done(union ctl_io *io, int have_lock)
12238 {
12239 	struct ctl_lun *lun;
12240 	struct ctl_softc *ctl_softc;
12241 	void (*fe_done)(union ctl_io *io);
12242 	uint32_t targ_port = ctl_port_idx(io->io_hdr.nexus.targ_port);
12243 
12244 	CTL_DEBUG_PRINT(("ctl_process_done\n"));
12245 
12246 	fe_done =
12247 	    control_softc->ctl_ports[targ_port]->fe_done;
12248 
12249 #ifdef CTL_TIME_IO
12250 	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12251 		char str[256];
12252 		char path_str[64];
12253 		struct sbuf sb;
12254 
12255 		ctl_scsi_path_string(io, path_str, sizeof(path_str));
12256 		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12257 
12258 		sbuf_cat(&sb, path_str);
12259 		switch (io->io_hdr.io_type) {
12260 		case CTL_IO_SCSI:
12261 			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12262 			sbuf_printf(&sb, "\n");
12263 			sbuf_cat(&sb, path_str);
12264 			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12265 				    io->scsiio.tag_num, io->scsiio.tag_type);
12266 			break;
12267 		case CTL_IO_TASK:
12268 			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12269 				    "Tag Type: %d\n", io->taskio.task_action,
12270 				    io->taskio.tag_num, io->taskio.tag_type);
12271 			break;
12272 		default:
12273 			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12274 			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12275 			break;
12276 		}
12277 		sbuf_cat(&sb, path_str);
12278 		sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
12279 			    (intmax_t)time_uptime - io->io_hdr.start_time);
12280 		sbuf_finish(&sb);
12281 		printf("%s", sbuf_data(&sb));
12282 	}
12283 #endif /* CTL_TIME_IO */
12284 
12285 	switch (io->io_hdr.io_type) {
12286 	case CTL_IO_SCSI:
12287 		break;
12288 	case CTL_IO_TASK:
12289 		ctl_io_error_print(io, NULL);
12290 		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
12291 			ctl_free_io_internal(io, /*have_lock*/ 0);
12292 		else
12293 			fe_done(io);
12294 		return (CTL_RETVAL_COMPLETE);
12295 		break;
12296 	default:
12297 		printf("ctl_process_done: invalid io type %d\n",
12298 		       io->io_hdr.io_type);
12299 		panic("ctl_process_done: invalid io type %d\n",
12300 		      io->io_hdr.io_type);
12301 		break; /* NOTREACHED */
12302 	}
12303 
12304 	lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12305 	if (lun == NULL) {
12306 		CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
12307 				 io->io_hdr.nexus.targ_lun));
12308 		fe_done(io);
12309 		goto bailout;
12310 	}
12311 	ctl_softc = lun->ctl_softc;
12312 
12313 	/*
12314 	 * Remove this from the OOA queue.
12315 	 */
12316 	if (have_lock == 0)
12317 		mtx_lock(&ctl_softc->ctl_lock);
12318 
12319 	/*
12320 	 * Check to see if we have any errors to inject here.  We only
12321 	 * inject errors for commands that don't already have errors set.
12322 	 */
12323 	if ((STAILQ_FIRST(&lun->error_list) != NULL)
12324 	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))
12325 		ctl_inject_error(lun, io);
12326 
12327 	/*
12328 	 * XXX KDM how do we treat commands that aren't completed
12329 	 * successfully?
12330 	 *
12331 	 * XXX KDM should we also track I/O latency?
12332 	 */
12333 	if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) {
12334 		uint32_t blocksize;
12335 #ifdef CTL_TIME_IO
12336 		struct bintime cur_bt;
12337 #endif
12338 
12339 		if ((lun->be_lun != NULL)
12340 		 && (lun->be_lun->blocksize != 0))
12341 			blocksize = lun->be_lun->blocksize;
12342 		else
12343 			blocksize = 512;
12344 
12345 		switch (io->io_hdr.io_type) {
12346 		case CTL_IO_SCSI: {
12347 			int isread;
12348 			struct ctl_lba_len lbalen;
12349 
12350 			isread = 0;
12351 			switch (io->scsiio.cdb[0]) {
12352 			case READ_6:
12353 			case READ_10:
12354 			case READ_12:
12355 			case READ_16:
12356 				isread = 1;
12357 				/* FALLTHROUGH */
12358 			case WRITE_6:
12359 			case WRITE_10:
12360 			case WRITE_12:
12361 			case WRITE_16:
12362 			case WRITE_VERIFY_10:
12363 			case WRITE_VERIFY_12:
12364 			case WRITE_VERIFY_16:
12365 				memcpy(&lbalen, io->io_hdr.ctl_private[
12366 				       CTL_PRIV_LBA_LEN].bytes, sizeof(lbalen));
12367 
12368 				if (isread) {
12369 					lun->stats.ports[targ_port].bytes[CTL_STATS_READ] +=
12370 						lbalen.len * blocksize;
12371 					lun->stats.ports[targ_port].operations[CTL_STATS_READ]++;
12372 
12373 #ifdef CTL_TIME_IO
12374 					bintime_add(
12375 					   &lun->stats.ports[targ_port].dma_time[CTL_STATS_READ],
12376 					   &io->io_hdr.dma_bt);
12377 					lun->stats.ports[targ_port].num_dmas[CTL_STATS_READ] +=
12378 						io->io_hdr.num_dmas;
12379 					getbintime(&cur_bt);
12380 					bintime_sub(&cur_bt,
12381 						    &io->io_hdr.start_bt);
12382 
12383 					bintime_add(
12384 					    &lun->stats.ports[targ_port].time[CTL_STATS_READ],
12385 					    &cur_bt);
12386 
12387 #if 0
12388 					cs_prof_gettime(&cur_ticks);
12389 					lun->stats.time[CTL_STATS_READ] +=
12390 						cur_ticks -
12391 						io->io_hdr.start_ticks;
12392 #endif
12393 #if 0
12394 					lun->stats.time[CTL_STATS_READ] +=
12395 						jiffies - io->io_hdr.start_time;
12396 #endif
12397 #endif /* CTL_TIME_IO */
12398 				} else {
12399 					lun->stats.ports[targ_port].bytes[CTL_STATS_WRITE] +=
12400 						lbalen.len * blocksize;
12401 					lun->stats.ports[targ_port].operations[
12402 						CTL_STATS_WRITE]++;
12403 
12404 #ifdef CTL_TIME_IO
12405 					bintime_add(
12406 					  &lun->stats.ports[targ_port].dma_time[CTL_STATS_WRITE],
12407 					  &io->io_hdr.dma_bt);
12408 					lun->stats.ports[targ_port].num_dmas[CTL_STATS_WRITE] +=
12409 						io->io_hdr.num_dmas;
12410 					getbintime(&cur_bt);
12411 					bintime_sub(&cur_bt,
12412 						    &io->io_hdr.start_bt);
12413 
12414 					bintime_add(
12415 					    &lun->stats.ports[targ_port].time[CTL_STATS_WRITE],
12416 					    &cur_bt);
12417 #if 0
12418 					cs_prof_gettime(&cur_ticks);
12419 					lun->stats.ports[targ_port].time[CTL_STATS_WRITE] +=
12420 						cur_ticks -
12421 						io->io_hdr.start_ticks;
12422 					lun->stats.ports[targ_port].time[CTL_STATS_WRITE] +=
12423 						jiffies - io->io_hdr.start_time;
12424 #endif
12425 #endif /* CTL_TIME_IO */
12426 				}
12427 				break;
12428 			default:
12429 				lun->stats.ports[targ_port].operations[CTL_STATS_NO_IO]++;
12430 
12431 #ifdef CTL_TIME_IO
12432 				bintime_add(
12433 				  &lun->stats.ports[targ_port].dma_time[CTL_STATS_NO_IO],
12434 				  &io->io_hdr.dma_bt);
12435 				lun->stats.ports[targ_port].num_dmas[CTL_STATS_NO_IO] +=
12436 					io->io_hdr.num_dmas;
12437 				getbintime(&cur_bt);
12438 				bintime_sub(&cur_bt, &io->io_hdr.start_bt);
12439 
12440 				bintime_add(&lun->stats.ports[targ_port].time[CTL_STATS_NO_IO],
12441 					    &cur_bt);
12442 
12443 #if 0
12444 				cs_prof_gettime(&cur_ticks);
12445 				lun->stats.ports[targ_port].time[CTL_STATS_NO_IO] +=
12446 					cur_ticks -
12447 					io->io_hdr.start_ticks;
12448 				lun->stats.ports[targ_port].time[CTL_STATS_NO_IO] +=
12449 					jiffies - io->io_hdr.start_time;
12450 #endif
12451 #endif /* CTL_TIME_IO */
12452 				break;
12453 			}
12454 			break;
12455 		}
12456 		default:
12457 			break;
12458 		}
12459 	}
12460 
12461 	TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
12462 
12463 	/*
12464 	 * Run through the blocked queue on this LUN and see if anything
12465 	 * has become unblocked, now that this transaction is done.
12466 	 */
12467 	ctl_check_blocked(lun);
12468 
12469 	/*
12470 	 * If the LUN has been invalidated, free it if there is nothing
12471 	 * left on its OOA queue.
12472 	 */
12473 	if ((lun->flags & CTL_LUN_INVALID)
12474 	 && (TAILQ_FIRST(&lun->ooa_queue) == NULL))
12475 		ctl_free_lun(lun);
12476 
12477 	/*
12478 	 * If this command has been aborted, make sure we set the status
12479 	 * properly.  The FETD is responsible for freeing the I/O and doing
12480 	 * whatever it needs to do to clean up its state.
12481 	 */
12482 	if (io->io_hdr.flags & CTL_FLAG_ABORT)
12483 		io->io_hdr.status = CTL_CMD_ABORTED;
12484 
12485 	/*
12486 	 * We print out status for every task management command.  For SCSI
12487 	 * commands, we filter out any unit attention errors; they happen
12488 	 * on every boot, and would clutter up the log.  Note:  task
12489 	 * management commands aren't printed here, they are printed above,
12490 	 * since they should never even make it down here.
12491 	 */
12492 	switch (io->io_hdr.io_type) {
12493 	case CTL_IO_SCSI: {
12494 		int error_code, sense_key, asc, ascq;
12495 
12496 		sense_key = 0;
12497 
12498 		if (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SCSI_ERROR)
12499 		 && (io->scsiio.scsi_status == SCSI_STATUS_CHECK_COND)) {
12500 			/*
12501 			 * Since this is just for printing, no need to
12502 			 * show errors here.
12503 			 */
12504 			scsi_extract_sense_len(&io->scsiio.sense_data,
12505 					       io->scsiio.sense_len,
12506 					       &error_code,
12507 					       &sense_key,
12508 					       &asc,
12509 					       &ascq,
12510 					       /*show_errors*/ 0);
12511 		}
12512 
12513 		if (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
12514 		 && (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SCSI_ERROR)
12515 		  || (io->scsiio.scsi_status != SCSI_STATUS_CHECK_COND)
12516 		  || (sense_key != SSD_KEY_UNIT_ATTENTION))) {
12517 
12518 			if ((time_uptime - ctl_softc->last_print_jiffies) <= 0){
12519 				ctl_softc->skipped_prints++;
12520 				if (have_lock == 0)
12521 					mtx_unlock(&ctl_softc->ctl_lock);
12522 			} else {
12523 				uint32_t skipped_prints;
12524 
12525 				skipped_prints = ctl_softc->skipped_prints;
12526 
12527 				ctl_softc->skipped_prints = 0;
12528 				ctl_softc->last_print_jiffies = time_uptime;
12529 
12530 				if (have_lock == 0)
12531 					mtx_unlock(&ctl_softc->ctl_lock);
12532 				if (skipped_prints > 0) {
12533 #ifdef NEEDTOPORT
12534 					csevent_log(CSC_CTL | CSC_SHELF_SW |
12535 					    CTL_ERROR_REPORT,
12536 					    csevent_LogType_Trace,
12537 					    csevent_Severity_Information,
12538 					    csevent_AlertLevel_Green,
12539 					    csevent_FRU_Firmware,
12540 					    csevent_FRU_Unknown,
12541 					    "High CTL error volume, %d prints "
12542 					    "skipped", skipped_prints);
12543 #endif
12544 				}
12545 				ctl_io_error_print(io, NULL);
12546 			}
12547 		} else {
12548 			if (have_lock == 0)
12549 				mtx_unlock(&ctl_softc->ctl_lock);
12550 		}
12551 		break;
12552 	}
12553 	case CTL_IO_TASK:
12554 		if (have_lock == 0)
12555 			mtx_unlock(&ctl_softc->ctl_lock);
12556 		ctl_io_error_print(io, NULL);
12557 		break;
12558 	default:
12559 		if (have_lock == 0)
12560 			mtx_unlock(&ctl_softc->ctl_lock);
12561 		break;
12562 	}
12563 
12564 	/*
12565 	 * Tell the FETD or the other shelf controller we're done with this
12566 	 * command.  Note that only SCSI commands get to this point.  Task
12567 	 * management commands are completed above.
12568 	 *
12569 	 * We only send status to the other controller if we're in XFER
12570 	 * mode.  In SER_ONLY mode, the I/O is done on the controller that
12571 	 * received the I/O (from CTL's perspective), and so the status is
12572 	 * generated there.
12573 	 *
12574 	 * XXX KDM if we hold the lock here, we could cause a deadlock
12575 	 * if the frontend comes back in in this context to queue
12576 	 * something.
12577 	 */
12578 	if ((ctl_softc->ha_mode == CTL_HA_MODE_XFER)
12579 	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12580 		union ctl_ha_msg msg;
12581 
12582 		memset(&msg, 0, sizeof(msg));
12583 		msg.hdr.msg_type = CTL_MSG_FINISH_IO;
12584 		msg.hdr.original_sc = io->io_hdr.original_sc;
12585 		msg.hdr.nexus = io->io_hdr.nexus;
12586 		msg.hdr.status = io->io_hdr.status;
12587 		msg.scsi.scsi_status = io->scsiio.scsi_status;
12588 		msg.scsi.tag_num = io->scsiio.tag_num;
12589 		msg.scsi.tag_type = io->scsiio.tag_type;
12590 		msg.scsi.sense_len = io->scsiio.sense_len;
12591 		msg.scsi.sense_residual = io->scsiio.sense_residual;
12592 		msg.scsi.residual = io->scsiio.residual;
12593 		memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12594 		       sizeof(io->scsiio.sense_data));
12595 		/*
12596 		 * We copy this whether or not this is an I/O-related
12597 		 * command.  Otherwise, we'd have to go and check to see
12598 		 * whether it's a read/write command, and it really isn't
12599 		 * worth it.
12600 		 */
12601 		memcpy(&msg.scsi.lbalen,
12602 		       &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
12603 		       sizeof(msg.scsi.lbalen));
12604 
12605 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12606 				sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
12607 			/* XXX do something here */
12608 		}
12609 
12610 		ctl_free_io_internal(io, /*have_lock*/ 0);
12611 	} else
12612 		fe_done(io);
12613 
12614 bailout:
12615 
12616 	return (CTL_RETVAL_COMPLETE);
12617 }
12618 
12619 /*
12620  * Front end should call this if it doesn't do autosense.  When the request
12621  * sense comes back in from the initiator, we'll dequeue this and send it.
12622  */
12623 int
12624 ctl_queue_sense(union ctl_io *io)
12625 {
12626 	struct ctl_lun *lun;
12627 	struct ctl_softc *ctl_softc;
12628 	uint32_t initidx;
12629 
12630 	ctl_softc = control_softc;
12631 
12632 	CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
12633 
12634 	/*
12635 	 * LUN lookup will likely move to the ctl_work_thread() once we
12636 	 * have our new queueing infrastructure (that doesn't put things on
12637 	 * a per-LUN queue initially).  That is so that we can handle
12638 	 * things like an INQUIRY to a LUN that we don't have enabled.  We
12639 	 * can't deal with that right now.
12640 	 */
12641 	mtx_lock(&ctl_softc->ctl_lock);
12642 
12643 	/*
12644 	 * If we don't have a LUN for this, just toss the sense
12645 	 * information.
12646 	 */
12647 	if ((io->io_hdr.nexus.targ_lun < CTL_MAX_LUNS)
12648 	 && (ctl_softc->ctl_luns[io->io_hdr.nexus.targ_lun] != NULL))
12649 		lun = ctl_softc->ctl_luns[io->io_hdr.nexus.targ_lun];
12650 	else
12651 		goto bailout;
12652 
12653 	initidx = ctl_get_initindex(&io->io_hdr.nexus);
12654 
12655 	/*
12656 	 * Already have CA set for this LUN...toss the sense information.
12657 	 */
12658 	if (ctl_is_set(lun->have_ca, initidx))
12659 		goto bailout;
12660 
12661 	memcpy(&lun->pending_sense[initidx].sense, &io->scsiio.sense_data,
12662 	       ctl_min(sizeof(lun->pending_sense[initidx].sense),
12663 	       sizeof(io->scsiio.sense_data)));
12664 	ctl_set_mask(lun->have_ca, initidx);
12665 
12666 bailout:
12667 	mtx_unlock(&ctl_softc->ctl_lock);
12668 
12669 	ctl_free_io(io);
12670 
12671 	return (CTL_RETVAL_COMPLETE);
12672 }
12673 
12674 /*
12675  * Primary command inlet from frontend ports.  All SCSI and task I/O
12676  * requests must go through this function.
12677  */
12678 int
12679 ctl_queue(union ctl_io *io)
12680 {
12681 	struct ctl_softc *ctl_softc;
12682 
12683 	CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
12684 
12685 	ctl_softc = control_softc;
12686 
12687 #ifdef CTL_TIME_IO
12688 	io->io_hdr.start_time = time_uptime;
12689 	getbintime(&io->io_hdr.start_bt);
12690 #endif /* CTL_TIME_IO */
12691 
12692 	mtx_lock(&ctl_softc->ctl_lock);
12693 
12694 	switch (io->io_hdr.io_type) {
12695 	case CTL_IO_SCSI:
12696 		STAILQ_INSERT_TAIL(&ctl_softc->incoming_queue, &io->io_hdr,
12697 				   links);
12698 		break;
12699 	case CTL_IO_TASK:
12700 		STAILQ_INSERT_TAIL(&ctl_softc->task_queue, &io->io_hdr, links);
12701 		/*
12702 		 * Set the task pending flag.  This is necessary to close a
12703 		 * race condition with the FETD:
12704 		 *
12705 		 * - FETD submits a task management command, like an abort.
12706 		 * - Back end calls fe_datamove() to move the data for the
12707 		 *   aborted command.  The FETD can't really accept it, but
12708 		 *   if it did, it would end up transmitting data for a
12709 		 *   command that the initiator told us to abort.
12710 		 *
12711 		 * We close the race condition by setting the flag here,
12712 		 * and checking it in ctl_datamove(), before calling the
12713 		 * FETD's fe_datamove routine.  If we've got a task
12714 		 * pending, we run the task queue and then check to see
12715 		 * whether our particular I/O has been aborted.
12716 		 */
12717 		ctl_softc->flags |= CTL_FLAG_TASK_PENDING;
12718 		break;
12719 	default:
12720 		mtx_unlock(&ctl_softc->ctl_lock);
12721 		printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
12722 		return (-EINVAL);
12723 		break; /* NOTREACHED */
12724 	}
12725 	mtx_unlock(&ctl_softc->ctl_lock);
12726 
12727 	ctl_wakeup_thread();
12728 
12729 	return (CTL_RETVAL_COMPLETE);
12730 }
12731 
12732 #ifdef CTL_IO_DELAY
12733 static void
12734 ctl_done_timer_wakeup(void *arg)
12735 {
12736 	union ctl_io *io;
12737 
12738 	io = (union ctl_io *)arg;
12739 	ctl_done_lock(io, /*have_lock*/ 0);
12740 }
12741 #endif /* CTL_IO_DELAY */
12742 
12743 void
12744 ctl_done_lock(union ctl_io *io, int have_lock)
12745 {
12746 	struct ctl_softc *ctl_softc;
12747 #ifndef CTL_DONE_THREAD
12748 	union ctl_io *xio;
12749 #endif /* !CTL_DONE_THREAD */
12750 
12751 	ctl_softc = control_softc;
12752 
12753 	if (have_lock == 0)
12754 		mtx_lock(&ctl_softc->ctl_lock);
12755 
12756 	/*
12757 	 * Enable this to catch duplicate completion issues.
12758 	 */
12759 #if 0
12760 	if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
12761 		printf("%s: type %d msg %d cdb %x iptl: "
12762 		       "%d:%d:%d:%d tag 0x%04x "
12763 		       "flag %#x status %x\n",
12764 			__func__,
12765 			io->io_hdr.io_type,
12766 			io->io_hdr.msg_type,
12767 			io->scsiio.cdb[0],
12768 			io->io_hdr.nexus.initid.id,
12769 			io->io_hdr.nexus.targ_port,
12770 			io->io_hdr.nexus.targ_target.id,
12771 			io->io_hdr.nexus.targ_lun,
12772 			(io->io_hdr.io_type ==
12773 			CTL_IO_TASK) ?
12774 			io->taskio.tag_num :
12775 			io->scsiio.tag_num,
12776 		        io->io_hdr.flags,
12777 			io->io_hdr.status);
12778 	} else
12779 		io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
12780 #endif
12781 
12782 	/*
12783 	 * This is an internal copy of an I/O, and should not go through
12784 	 * the normal done processing logic.
12785 	 */
12786 	if (io->io_hdr.flags & CTL_FLAG_INT_COPY) {
12787 		if (have_lock == 0)
12788 			mtx_unlock(&ctl_softc->ctl_lock);
12789 		return;
12790 	}
12791 
12792 	/*
12793 	 * We need to send a msg to the serializing shelf to finish the IO
12794 	 * as well.  We don't send a finish message to the other shelf if
12795 	 * this is a task management command.  Task management commands
12796 	 * aren't serialized in the OOA queue, but rather just executed on
12797 	 * both shelf controllers for commands that originated on that
12798 	 * controller.
12799 	 */
12800 	if ((io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)
12801 	 && (io->io_hdr.io_type != CTL_IO_TASK)) {
12802 		union ctl_ha_msg msg_io;
12803 
12804 		msg_io.hdr.msg_type = CTL_MSG_FINISH_IO;
12805 		msg_io.hdr.serializing_sc = io->io_hdr.serializing_sc;
12806 		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_io,
12807 		    sizeof(msg_io), 0 ) != CTL_HA_STATUS_SUCCESS) {
12808 		}
12809 		/* continue on to finish IO */
12810 	}
12811 #ifdef CTL_IO_DELAY
12812 	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12813 		struct ctl_lun *lun;
12814 
12815 		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12816 
12817 		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12818 	} else {
12819 		struct ctl_lun *lun;
12820 
12821 		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12822 
12823 		if ((lun != NULL)
12824 		 && (lun->delay_info.done_delay > 0)) {
12825 			struct callout *callout;
12826 
12827 			callout = (struct callout *)&io->io_hdr.timer_bytes;
12828 			callout_init(callout, /*mpsafe*/ 1);
12829 			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12830 			callout_reset(callout,
12831 				      lun->delay_info.done_delay * hz,
12832 				      ctl_done_timer_wakeup, io);
12833 			if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
12834 				lun->delay_info.done_delay = 0;
12835 			if (have_lock == 0)
12836 				mtx_unlock(&ctl_softc->ctl_lock);
12837 			return;
12838 		}
12839 	}
12840 #endif /* CTL_IO_DELAY */
12841 
12842 	STAILQ_INSERT_TAIL(&ctl_softc->done_queue, &io->io_hdr, links);
12843 
12844 #ifdef CTL_DONE_THREAD
12845 	if (have_lock == 0)
12846 		mtx_unlock(&ctl_softc->ctl_lock);
12847 
12848 	ctl_wakeup_thread();
12849 #else /* CTL_DONE_THREAD */
12850 	for (xio = (union ctl_io *)STAILQ_FIRST(&ctl_softc->done_queue);
12851 	     xio != NULL;
12852 	     xio =(union ctl_io *)STAILQ_FIRST(&ctl_softc->done_queue)) {
12853 
12854 		STAILQ_REMOVE_HEAD(&ctl_softc->done_queue, links);
12855 
12856 		ctl_process_done(xio, /*have_lock*/ 1);
12857 	}
12858 	if (have_lock == 0)
12859 		mtx_unlock(&ctl_softc->ctl_lock);
12860 #endif /* CTL_DONE_THREAD */
12861 }
12862 
12863 void
12864 ctl_done(union ctl_io *io)
12865 {
12866 	ctl_done_lock(io, /*have_lock*/ 0);
12867 }
12868 
12869 int
12870 ctl_isc(struct ctl_scsiio *ctsio)
12871 {
12872 	struct ctl_lun *lun;
12873 	int retval;
12874 
12875 	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12876 
12877 	CTL_DEBUG_PRINT(("ctl_isc: command: %02x\n", ctsio->cdb[0]));
12878 
12879 	CTL_DEBUG_PRINT(("ctl_isc: calling data_submit()\n"));
12880 
12881 	retval = lun->backend->data_submit((union ctl_io *)ctsio);
12882 
12883 	return (retval);
12884 }
12885 
12886 
12887 static void
12888 ctl_work_thread(void *arg)
12889 {
12890 	struct ctl_softc *softc;
12891 	union ctl_io *io;
12892 	struct ctl_be_lun *be_lun;
12893 	int retval;
12894 
12895 	CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
12896 
12897 	softc = (struct ctl_softc *)arg;
12898 	if (softc == NULL)
12899 		return;
12900 
12901 	mtx_lock(&softc->ctl_lock);
12902 	for (;;) {
12903 		retval = 0;
12904 
12905 		/*
12906 		 * We handle the queues in this order:
12907 		 * - task management
12908 		 * - ISC
12909 		 * - done queue (to free up resources, unblock other commands)
12910 		 * - RtR queue
12911 		 * - incoming queue
12912 		 *
12913 		 * If those queues are empty, we break out of the loop and
12914 		 * go to sleep.
12915 		 */
12916 		io = (union ctl_io *)STAILQ_FIRST(&softc->task_queue);
12917 		if (io != NULL) {
12918 			ctl_run_task_queue(softc);
12919 			continue;
12920 		}
12921 		io = (union ctl_io *)STAILQ_FIRST(&softc->isc_queue);
12922 		if (io != NULL) {
12923 			STAILQ_REMOVE_HEAD(&softc->isc_queue, links);
12924 			ctl_handle_isc(io);
12925 			continue;
12926 		}
12927 		io = (union ctl_io *)STAILQ_FIRST(&softc->done_queue);
12928 		if (io != NULL) {
12929 			STAILQ_REMOVE_HEAD(&softc->done_queue, links);
12930 			/* clear any blocked commands, call fe_done */
12931 			mtx_unlock(&softc->ctl_lock);
12932 			/*
12933 			 * XXX KDM
12934 			 * Call this without a lock for now.  This will
12935 			 * depend on whether there is any way the FETD can
12936 			 * sleep or deadlock if called with the CTL lock
12937 			 * held.
12938 			 */
12939 			retval = ctl_process_done(io, /*have_lock*/ 0);
12940 			mtx_lock(&softc->ctl_lock);
12941 			continue;
12942 		}
12943 		if (!ctl_pause_rtr) {
12944 			io = (union ctl_io *)STAILQ_FIRST(&softc->rtr_queue);
12945 			if (io != NULL) {
12946 				STAILQ_REMOVE_HEAD(&softc->rtr_queue, links);
12947 				mtx_unlock(&softc->ctl_lock);
12948 				goto execute;
12949 			}
12950 		}
12951 		io = (union ctl_io *)STAILQ_FIRST(&softc->incoming_queue);
12952 		if (io != NULL) {
12953 			STAILQ_REMOVE_HEAD(&softc->incoming_queue, links);
12954 			mtx_unlock(&softc->ctl_lock);
12955 			ctl_scsiio_precheck(softc, &io->scsiio);
12956 			mtx_lock(&softc->ctl_lock);
12957 			continue;
12958 		}
12959 		/*
12960 		 * We might want to move this to a separate thread, so that
12961 		 * configuration requests (in this case LUN creations)
12962 		 * won't impact the I/O path.
12963 		 */
12964 		be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
12965 		if (be_lun != NULL) {
12966 			STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
12967 			mtx_unlock(&softc->ctl_lock);
12968 			ctl_create_lun(be_lun);
12969 			mtx_lock(&softc->ctl_lock);
12970 			continue;
12971 		}
12972 
12973 		/* XXX KDM use the PDROP flag?? */
12974 		/* Sleep until we have something to do. */
12975 		mtx_sleep(softc, &softc->ctl_lock, PRIBIO, "ctl_work", 0);
12976 
12977 		/* Back to the top of the loop to see what woke us up. */
12978 		continue;
12979 
12980 execute:
12981 		retval = ctl_scsiio(&io->scsiio);
12982 		switch (retval) {
12983 		case CTL_RETVAL_COMPLETE:
12984 			break;
12985 		default:
12986 			/*
12987 			 * Probably need to make sure this doesn't happen.
12988 			 */
12989 			break;
12990 		}
12991 		mtx_lock(&softc->ctl_lock);
12992 	}
12993 }
12994 
12995 void
12996 ctl_wakeup_thread()
12997 {
12998 	struct ctl_softc *softc;
12999 
13000 	softc = control_softc;
13001 
13002 	wakeup(softc);
13003 }
13004 
13005 /* Initialization and failover */
13006 
13007 void
13008 ctl_init_isc_msg(void)
13009 {
13010 	printf("CTL: Still calling this thing\n");
13011 }
13012 
13013 /*
13014  * Init component
13015  * 	Initializes component into configuration defined by bootMode
13016  *	(see hasc-sv.c)
13017  *  	returns hasc_Status:
13018  * 		OK
13019  *		ERROR - fatal error
13020  */
13021 static ctl_ha_comp_status
13022 ctl_isc_init(struct ctl_ha_component *c)
13023 {
13024 	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
13025 
13026 	c->status = ret;
13027 	return ret;
13028 }
13029 
13030 /* Start component
13031  * 	Starts component in state requested. If component starts successfully,
13032  *	it must set its own state to the requestrd state
13033  *	When requested state is HASC_STATE_HA, the component may refine it
13034  * 	by adding _SLAVE or _MASTER flags.
13035  *	Currently allowed state transitions are:
13036  *	UNKNOWN->HA		- initial startup
13037  *	UNKNOWN->SINGLE - initial startup when no parter detected
13038  *	HA->SINGLE		- failover
13039  * returns ctl_ha_comp_status:
13040  * 		OK	- component successfully started in requested state
13041  *		FAILED  - could not start the requested state, failover may
13042  * 			  be possible
13043  *		ERROR	- fatal error detected, no future startup possible
13044  */
13045 static ctl_ha_comp_status
13046 ctl_isc_start(struct ctl_ha_component *c, ctl_ha_state state)
13047 {
13048 	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
13049 
13050 	// UNKNOWN->HA or UNKNOWN->SINGLE (bootstrap)
13051 	if (c->state == CTL_HA_STATE_UNKNOWN ) {
13052 		ctl_is_single = 0;
13053 		if (ctl_ha_msg_create(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
13054 		    != CTL_HA_STATUS_SUCCESS) {
13055 			printf("ctl_isc_start: ctl_ha_msg_create failed.\n");
13056 			ret = CTL_HA_COMP_STATUS_ERROR;
13057 		}
13058 	} else if (CTL_HA_STATE_IS_HA(c->state)
13059 		&& CTL_HA_STATE_IS_SINGLE(state)){
13060 		// HA->SINGLE transition
13061 	        ctl_failover();
13062 		ctl_is_single = 1;
13063 	} else {
13064 		printf("ctl_isc_start:Invalid state transition %X->%X\n",
13065 		       c->state, state);
13066 		ret = CTL_HA_COMP_STATUS_ERROR;
13067 	}
13068 	if (CTL_HA_STATE_IS_SINGLE(state))
13069 		ctl_is_single = 1;
13070 
13071 	c->state = state;
13072 	c->status = ret;
13073 	return ret;
13074 }
13075 
13076 /*
13077  * Quiesce component
13078  * The component must clear any error conditions (set status to OK) and
13079  * prepare itself to another Start call
13080  * returns ctl_ha_comp_status:
13081  * 	OK
13082  *	ERROR
13083  */
13084 static ctl_ha_comp_status
13085 ctl_isc_quiesce(struct ctl_ha_component *c)
13086 {
13087 	int ret = CTL_HA_COMP_STATUS_OK;
13088 
13089 	ctl_pause_rtr = 1;
13090 	c->status = ret;
13091 	return ret;
13092 }
13093 
13094 struct ctl_ha_component ctl_ha_component_ctlisc =
13095 {
13096 	.name = "CTL ISC",
13097 	.state = CTL_HA_STATE_UNKNOWN,
13098 	.init = ctl_isc_init,
13099 	.start = ctl_isc_start,
13100 	.quiesce = ctl_isc_quiesce
13101 };
13102 
13103 /*
13104  *  vim: ts=8
13105  */
13106