xref: /freebsd/sys/cam/cam_periph.c (revision 01b792f1f535c12a1a14000cf3360ef6c36cee2d)
1 /*-
2  * Common functions for CAM "type" (peripheral) drivers.
3  *
4  * Copyright (c) 1997, 1998 Justin T. Gibbs.
5  * Copyright (c) 1997, 1998, 1999, 2000 Kenneth D. Merry.
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions, and the following disclaimer,
13  *    without modification, immediately at the beginning of the file.
14  * 2. The name of the author may not be used to endorse or promote products
15  *    derived from this software without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE FOR
21  * ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  */
29 
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD$");
32 
33 #include <sys/param.h>
34 #include <sys/systm.h>
35 #include <sys/types.h>
36 #include <sys/malloc.h>
37 #include <sys/kernel.h>
38 #include <sys/bio.h>
39 #include <sys/lock.h>
40 #include <sys/mutex.h>
41 #include <sys/buf.h>
42 #include <sys/proc.h>
43 #include <sys/devicestat.h>
44 #include <sys/bus.h>
45 #include <sys/sbuf.h>
46 #include <vm/vm.h>
47 #include <vm/vm_extern.h>
48 
49 #include <cam/cam.h>
50 #include <cam/cam_ccb.h>
51 #include <cam/cam_queue.h>
52 #include <cam/cam_xpt_periph.h>
53 #include <cam/cam_periph.h>
54 #include <cam/cam_debug.h>
55 #include <cam/cam_sim.h>
56 
57 #include <cam/scsi/scsi_all.h>
58 #include <cam/scsi/scsi_message.h>
59 #include <cam/scsi/scsi_pass.h>
60 
61 static	u_int		camperiphnextunit(struct periph_driver *p_drv,
62 					  u_int newunit, int wired,
63 					  path_id_t pathid, target_id_t target,
64 					  lun_id_t lun);
65 static	u_int		camperiphunit(struct periph_driver *p_drv,
66 				      path_id_t pathid, target_id_t target,
67 				      lun_id_t lun);
68 static	void		camperiphdone(struct cam_periph *periph,
69 					union ccb *done_ccb);
70 static  void		camperiphfree(struct cam_periph *periph);
71 static int		camperiphscsistatuserror(union ccb *ccb,
72 					        union ccb **orig_ccb,
73 						 cam_flags camflags,
74 						 u_int32_t sense_flags,
75 						 int *openings,
76 						 u_int32_t *relsim_flags,
77 						 u_int32_t *timeout,
78 						 u_int32_t  *action,
79 						 const char **action_string);
80 static	int		camperiphscsisenseerror(union ccb *ccb,
81 					        union ccb **orig_ccb,
82 					        cam_flags camflags,
83 					        u_int32_t sense_flags,
84 					        int *openings,
85 					        u_int32_t *relsim_flags,
86 					        u_int32_t *timeout,
87 					        u_int32_t *action,
88 					        const char **action_string);
89 
90 static int nperiph_drivers;
91 static int initialized = 0;
92 struct periph_driver **periph_drivers;
93 
94 static MALLOC_DEFINE(M_CAMPERIPH, "CAM periph", "CAM peripheral buffers");
95 
96 static int periph_selto_delay = 1000;
97 TUNABLE_INT("kern.cam.periph_selto_delay", &periph_selto_delay);
98 static int periph_noresrc_delay = 500;
99 TUNABLE_INT("kern.cam.periph_noresrc_delay", &periph_noresrc_delay);
100 static int periph_busy_delay = 500;
101 TUNABLE_INT("kern.cam.periph_busy_delay", &periph_busy_delay);
102 
103 
104 void
105 periphdriver_register(void *data)
106 {
107 	struct periph_driver *drv = (struct periph_driver *)data;
108 	struct periph_driver **newdrivers, **old;
109 	int ndrivers;
110 
111 	ndrivers = nperiph_drivers + 2;
112 	newdrivers = malloc(sizeof(*newdrivers) * ndrivers, M_CAMPERIPH,
113 			    M_WAITOK);
114 	if (periph_drivers)
115 		bcopy(periph_drivers, newdrivers,
116 		      sizeof(*newdrivers) * nperiph_drivers);
117 	newdrivers[nperiph_drivers] = drv;
118 	newdrivers[nperiph_drivers + 1] = NULL;
119 	old = periph_drivers;
120 	periph_drivers = newdrivers;
121 	if (old)
122 		free(old, M_CAMPERIPH);
123 	nperiph_drivers++;
124 	/* If driver marked as early or it is late now, initialize it. */
125 	if (((drv->flags & CAM_PERIPH_DRV_EARLY) != 0 && initialized > 0) ||
126 	    initialized > 1)
127 		(*drv->init)();
128 }
129 
130 void
131 periphdriver_init(int level)
132 {
133 	int	i, early;
134 
135 	initialized = max(initialized, level);
136 	for (i = 0; periph_drivers[i] != NULL; i++) {
137 		early = (periph_drivers[i]->flags & CAM_PERIPH_DRV_EARLY) ? 1 : 2;
138 		if (early == initialized)
139 			(*periph_drivers[i]->init)();
140 	}
141 }
142 
143 cam_status
144 cam_periph_alloc(periph_ctor_t *periph_ctor,
145 		 periph_oninv_t *periph_oninvalidate,
146 		 periph_dtor_t *periph_dtor, periph_start_t *periph_start,
147 		 char *name, cam_periph_type type, struct cam_path *path,
148 		 ac_callback_t *ac_callback, ac_code code, void *arg)
149 {
150 	struct		periph_driver **p_drv;
151 	struct		cam_sim *sim;
152 	struct		cam_periph *periph;
153 	struct		cam_periph *cur_periph;
154 	path_id_t	path_id;
155 	target_id_t	target_id;
156 	lun_id_t	lun_id;
157 	cam_status	status;
158 	u_int		init_level;
159 
160 	init_level = 0;
161 	/*
162 	 * Handle Hot-Plug scenarios.  If there is already a peripheral
163 	 * of our type assigned to this path, we are likely waiting for
164 	 * final close on an old, invalidated, peripheral.  If this is
165 	 * the case, queue up a deferred call to the peripheral's async
166 	 * handler.  If it looks like a mistaken re-allocation, complain.
167 	 */
168 	if ((periph = cam_periph_find(path, name)) != NULL) {
169 
170 		if ((periph->flags & CAM_PERIPH_INVALID) != 0
171 		 && (periph->flags & CAM_PERIPH_NEW_DEV_FOUND) == 0) {
172 			periph->flags |= CAM_PERIPH_NEW_DEV_FOUND;
173 			periph->deferred_callback = ac_callback;
174 			periph->deferred_ac = code;
175 			return (CAM_REQ_INPROG);
176 		} else {
177 			printf("cam_periph_alloc: attempt to re-allocate "
178 			       "valid device %s%d rejected flags %#x "
179 			       "refcount %d\n", periph->periph_name,
180 			       periph->unit_number, periph->flags,
181 			       periph->refcount);
182 		}
183 		return (CAM_REQ_INVALID);
184 	}
185 
186 	periph = (struct cam_periph *)malloc(sizeof(*periph), M_CAMPERIPH,
187 					     M_NOWAIT|M_ZERO);
188 
189 	if (periph == NULL)
190 		return (CAM_RESRC_UNAVAIL);
191 
192 	init_level++;
193 
194 
195 	sim = xpt_path_sim(path);
196 	path_id = xpt_path_path_id(path);
197 	target_id = xpt_path_target_id(path);
198 	lun_id = xpt_path_lun_id(path);
199 	periph->periph_start = periph_start;
200 	periph->periph_dtor = periph_dtor;
201 	periph->periph_oninval = periph_oninvalidate;
202 	periph->type = type;
203 	periph->periph_name = name;
204 	periph->scheduled_priority = CAM_PRIORITY_NONE;
205 	periph->immediate_priority = CAM_PRIORITY_NONE;
206 	periph->refcount = 1;		/* Dropped by invalidation. */
207 	periph->sim = sim;
208 	SLIST_INIT(&periph->ccb_list);
209 	status = xpt_create_path(&path, periph, path_id, target_id, lun_id);
210 	if (status != CAM_REQ_CMP)
211 		goto failure;
212 	periph->path = path;
213 
214 	xpt_lock_buses();
215 	for (p_drv = periph_drivers; *p_drv != NULL; p_drv++) {
216 		if (strcmp((*p_drv)->driver_name, name) == 0)
217 			break;
218 	}
219 	if (*p_drv == NULL) {
220 		printf("cam_periph_alloc: invalid periph name '%s'\n", name);
221 		xpt_unlock_buses();
222 		xpt_free_path(periph->path);
223 		free(periph, M_CAMPERIPH);
224 		return (CAM_REQ_INVALID);
225 	}
226 	periph->unit_number = camperiphunit(*p_drv, path_id, target_id, lun_id);
227 	cur_periph = TAILQ_FIRST(&(*p_drv)->units);
228 	while (cur_periph != NULL
229 	    && cur_periph->unit_number < periph->unit_number)
230 		cur_periph = TAILQ_NEXT(cur_periph, unit_links);
231 	if (cur_periph != NULL) {
232 		KASSERT(cur_periph->unit_number != periph->unit_number, ("duplicate units on periph list"));
233 		TAILQ_INSERT_BEFORE(cur_periph, periph, unit_links);
234 	} else {
235 		TAILQ_INSERT_TAIL(&(*p_drv)->units, periph, unit_links);
236 		(*p_drv)->generation++;
237 	}
238 	xpt_unlock_buses();
239 
240 	init_level++;
241 
242 	status = xpt_add_periph(periph);
243 	if (status != CAM_REQ_CMP)
244 		goto failure;
245 
246 	init_level++;
247 	CAM_DEBUG(periph->path, CAM_DEBUG_INFO, ("Periph created\n"));
248 
249 	status = periph_ctor(periph, arg);
250 
251 	if (status == CAM_REQ_CMP)
252 		init_level++;
253 
254 failure:
255 	switch (init_level) {
256 	case 4:
257 		/* Initialized successfully */
258 		break;
259 	case 3:
260 		CAM_DEBUG(periph->path, CAM_DEBUG_INFO, ("Periph destroyed\n"));
261 		xpt_remove_periph(periph);
262 		/* FALLTHROUGH */
263 	case 2:
264 		xpt_lock_buses();
265 		TAILQ_REMOVE(&(*p_drv)->units, periph, unit_links);
266 		xpt_unlock_buses();
267 		xpt_free_path(periph->path);
268 		/* FALLTHROUGH */
269 	case 1:
270 		free(periph, M_CAMPERIPH);
271 		/* FALLTHROUGH */
272 	case 0:
273 		/* No cleanup to perform. */
274 		break;
275 	default:
276 		panic("%s: Unknown init level", __func__);
277 	}
278 	return(status);
279 }
280 
281 /*
282  * Find a peripheral structure with the specified path, target, lun,
283  * and (optionally) type.  If the name is NULL, this function will return
284  * the first peripheral driver that matches the specified path.
285  */
286 struct cam_periph *
287 cam_periph_find(struct cam_path *path, char *name)
288 {
289 	struct periph_driver **p_drv;
290 	struct cam_periph *periph;
291 
292 	xpt_lock_buses();
293 	for (p_drv = periph_drivers; *p_drv != NULL; p_drv++) {
294 
295 		if (name != NULL && (strcmp((*p_drv)->driver_name, name) != 0))
296 			continue;
297 
298 		TAILQ_FOREACH(periph, &(*p_drv)->units, unit_links) {
299 			if (xpt_path_comp(periph->path, path) == 0) {
300 				xpt_unlock_buses();
301 				cam_periph_assert(periph, MA_OWNED);
302 				return(periph);
303 			}
304 		}
305 		if (name != NULL) {
306 			xpt_unlock_buses();
307 			return(NULL);
308 		}
309 	}
310 	xpt_unlock_buses();
311 	return(NULL);
312 }
313 
314 /*
315  * Find peripheral driver instances attached to the specified path.
316  */
317 int
318 cam_periph_list(struct cam_path *path, struct sbuf *sb)
319 {
320 	struct sbuf local_sb;
321 	struct periph_driver **p_drv;
322 	struct cam_periph *periph;
323 	int count;
324 	int sbuf_alloc_len;
325 
326 	sbuf_alloc_len = 16;
327 retry:
328 	sbuf_new(&local_sb, NULL, sbuf_alloc_len, SBUF_FIXEDLEN);
329 	count = 0;
330 	xpt_lock_buses();
331 	for (p_drv = periph_drivers; *p_drv != NULL; p_drv++) {
332 
333 		TAILQ_FOREACH(periph, &(*p_drv)->units, unit_links) {
334 			if (xpt_path_comp(periph->path, path) != 0)
335 				continue;
336 
337 			if (sbuf_len(&local_sb) != 0)
338 				sbuf_cat(&local_sb, ",");
339 
340 			sbuf_printf(&local_sb, "%s%d", periph->periph_name,
341 				    periph->unit_number);
342 
343 			if (sbuf_error(&local_sb) == ENOMEM) {
344 				sbuf_alloc_len *= 2;
345 				xpt_unlock_buses();
346 				sbuf_delete(&local_sb);
347 				goto retry;
348 			}
349 			count++;
350 		}
351 	}
352 	xpt_unlock_buses();
353 	sbuf_finish(&local_sb);
354 	sbuf_cpy(sb, sbuf_data(&local_sb));
355 	sbuf_delete(&local_sb);
356 	return (count);
357 }
358 
359 cam_status
360 cam_periph_acquire(struct cam_periph *periph)
361 {
362 	cam_status status;
363 
364 	status = CAM_REQ_CMP_ERR;
365 	if (periph == NULL)
366 		return (status);
367 
368 	xpt_lock_buses();
369 	if ((periph->flags & CAM_PERIPH_INVALID) == 0) {
370 		periph->refcount++;
371 		status = CAM_REQ_CMP;
372 	}
373 	xpt_unlock_buses();
374 
375 	return (status);
376 }
377 
378 void
379 cam_periph_doacquire(struct cam_periph *periph)
380 {
381 
382 	xpt_lock_buses();
383 	KASSERT(periph->refcount >= 1,
384 	    ("cam_periph_doacquire() with refcount == %d", periph->refcount));
385 	periph->refcount++;
386 	xpt_unlock_buses();
387 }
388 
389 void
390 cam_periph_release_locked_buses(struct cam_periph *periph)
391 {
392 
393 	cam_periph_assert(periph, MA_OWNED);
394 	KASSERT(periph->refcount >= 1, ("periph->refcount >= 1"));
395 	if (--periph->refcount == 0)
396 		camperiphfree(periph);
397 }
398 
399 void
400 cam_periph_release_locked(struct cam_periph *periph)
401 {
402 
403 	if (periph == NULL)
404 		return;
405 
406 	xpt_lock_buses();
407 	cam_periph_release_locked_buses(periph);
408 	xpt_unlock_buses();
409 }
410 
411 void
412 cam_periph_release(struct cam_periph *periph)
413 {
414 	struct mtx *mtx;
415 
416 	if (periph == NULL)
417 		return;
418 
419 	cam_periph_assert(periph, MA_NOTOWNED);
420 	mtx = cam_periph_mtx(periph);
421 	mtx_lock(mtx);
422 	cam_periph_release_locked(periph);
423 	mtx_unlock(mtx);
424 }
425 
426 int
427 cam_periph_hold(struct cam_periph *periph, int priority)
428 {
429 	int error;
430 
431 	/*
432 	 * Increment the reference count on the peripheral
433 	 * while we wait for our lock attempt to succeed
434 	 * to ensure the peripheral doesn't disappear out
435 	 * from user us while we sleep.
436 	 */
437 
438 	if (cam_periph_acquire(periph) != CAM_REQ_CMP)
439 		return (ENXIO);
440 
441 	cam_periph_assert(periph, MA_OWNED);
442 	while ((periph->flags & CAM_PERIPH_LOCKED) != 0) {
443 		periph->flags |= CAM_PERIPH_LOCK_WANTED;
444 		if ((error = cam_periph_sleep(periph, periph, priority,
445 		    "caplck", 0)) != 0) {
446 			cam_periph_release_locked(periph);
447 			return (error);
448 		}
449 		if (periph->flags & CAM_PERIPH_INVALID) {
450 			cam_periph_release_locked(periph);
451 			return (ENXIO);
452 		}
453 	}
454 
455 	periph->flags |= CAM_PERIPH_LOCKED;
456 	return (0);
457 }
458 
459 void
460 cam_periph_unhold(struct cam_periph *periph)
461 {
462 
463 	cam_periph_assert(periph, MA_OWNED);
464 
465 	periph->flags &= ~CAM_PERIPH_LOCKED;
466 	if ((periph->flags & CAM_PERIPH_LOCK_WANTED) != 0) {
467 		periph->flags &= ~CAM_PERIPH_LOCK_WANTED;
468 		wakeup(periph);
469 	}
470 
471 	cam_periph_release_locked(periph);
472 }
473 
474 /*
475  * Look for the next unit number that is not currently in use for this
476  * peripheral type starting at "newunit".  Also exclude unit numbers that
477  * are reserved by for future "hardwiring" unless we already know that this
478  * is a potential wired device.  Only assume that the device is "wired" the
479  * first time through the loop since after that we'll be looking at unit
480  * numbers that did not match a wiring entry.
481  */
482 static u_int
483 camperiphnextunit(struct periph_driver *p_drv, u_int newunit, int wired,
484 		  path_id_t pathid, target_id_t target, lun_id_t lun)
485 {
486 	struct	cam_periph *periph;
487 	char	*periph_name;
488 	int	i, val, dunit, r;
489 	const char *dname, *strval;
490 
491 	periph_name = p_drv->driver_name;
492 	for (;;newunit++) {
493 
494 		for (periph = TAILQ_FIRST(&p_drv->units);
495 		     periph != NULL && periph->unit_number != newunit;
496 		     periph = TAILQ_NEXT(periph, unit_links))
497 			;
498 
499 		if (periph != NULL && periph->unit_number == newunit) {
500 			if (wired != 0) {
501 				xpt_print(periph->path, "Duplicate Wired "
502 				    "Device entry!\n");
503 				xpt_print(periph->path, "Second device (%s "
504 				    "device at scbus%d target %d lun %d) will "
505 				    "not be wired\n", periph_name, pathid,
506 				    target, lun);
507 				wired = 0;
508 			}
509 			continue;
510 		}
511 		if (wired)
512 			break;
513 
514 		/*
515 		 * Don't match entries like "da 4" as a wired down
516 		 * device, but do match entries like "da 4 target 5"
517 		 * or even "da 4 scbus 1".
518 		 */
519 		i = 0;
520 		dname = periph_name;
521 		for (;;) {
522 			r = resource_find_dev(&i, dname, &dunit, NULL, NULL);
523 			if (r != 0)
524 				break;
525 			/* if no "target" and no specific scbus, skip */
526 			if (resource_int_value(dname, dunit, "target", &val) &&
527 			    (resource_string_value(dname, dunit, "at",&strval)||
528 			     strcmp(strval, "scbus") == 0))
529 				continue;
530 			if (newunit == dunit)
531 				break;
532 		}
533 		if (r != 0)
534 			break;
535 	}
536 	return (newunit);
537 }
538 
539 static u_int
540 camperiphunit(struct periph_driver *p_drv, path_id_t pathid,
541 	      target_id_t target, lun_id_t lun)
542 {
543 	u_int	unit;
544 	int	wired, i, val, dunit;
545 	const char *dname, *strval;
546 	char	pathbuf[32], *periph_name;
547 
548 	periph_name = p_drv->driver_name;
549 	snprintf(pathbuf, sizeof(pathbuf), "scbus%d", pathid);
550 	unit = 0;
551 	i = 0;
552 	dname = periph_name;
553 	for (wired = 0; resource_find_dev(&i, dname, &dunit, NULL, NULL) == 0;
554 	     wired = 0) {
555 		if (resource_string_value(dname, dunit, "at", &strval) == 0) {
556 			if (strcmp(strval, pathbuf) != 0)
557 				continue;
558 			wired++;
559 		}
560 		if (resource_int_value(dname, dunit, "target", &val) == 0) {
561 			if (val != target)
562 				continue;
563 			wired++;
564 		}
565 		if (resource_int_value(dname, dunit, "lun", &val) == 0) {
566 			if (val != lun)
567 				continue;
568 			wired++;
569 		}
570 		if (wired != 0) {
571 			unit = dunit;
572 			break;
573 		}
574 	}
575 
576 	/*
577 	 * Either start from 0 looking for the next unit or from
578 	 * the unit number given in the resource config.  This way,
579 	 * if we have wildcard matches, we don't return the same
580 	 * unit number twice.
581 	 */
582 	unit = camperiphnextunit(p_drv, unit, wired, pathid, target, lun);
583 
584 	return (unit);
585 }
586 
587 void
588 cam_periph_invalidate(struct cam_periph *periph)
589 {
590 
591 	cam_periph_assert(periph, MA_OWNED);
592 	/*
593 	 * We only call this routine the first time a peripheral is
594 	 * invalidated.
595 	 */
596 	if ((periph->flags & CAM_PERIPH_INVALID) != 0)
597 		return;
598 
599 	CAM_DEBUG(periph->path, CAM_DEBUG_INFO, ("Periph invalidated\n"));
600 	if ((periph->flags & CAM_PERIPH_ANNOUNCED) && !rebooting)
601 		xpt_denounce_periph(periph);
602 	periph->flags |= CAM_PERIPH_INVALID;
603 	periph->flags &= ~CAM_PERIPH_NEW_DEV_FOUND;
604 	if (periph->periph_oninval != NULL)
605 		periph->periph_oninval(periph);
606 	cam_periph_release_locked(periph);
607 }
608 
609 static void
610 camperiphfree(struct cam_periph *periph)
611 {
612 	struct periph_driver **p_drv;
613 
614 	cam_periph_assert(periph, MA_OWNED);
615 	KASSERT(periph->periph_allocating == 0, ("%s%d: freed while allocating",
616 	    periph->periph_name, periph->unit_number));
617 	for (p_drv = periph_drivers; *p_drv != NULL; p_drv++) {
618 		if (strcmp((*p_drv)->driver_name, periph->periph_name) == 0)
619 			break;
620 	}
621 	if (*p_drv == NULL) {
622 		printf("camperiphfree: attempt to free non-existant periph\n");
623 		return;
624 	}
625 
626 	/*
627 	 * We need to set this flag before dropping the topology lock, to
628 	 * let anyone who is traversing the list that this peripheral is
629 	 * about to be freed, and there will be no more reference count
630 	 * checks.
631 	 */
632 	periph->flags |= CAM_PERIPH_FREE;
633 
634 	/*
635 	 * The peripheral destructor semantics dictate calling with only the
636 	 * SIM mutex held.  Since it might sleep, it should not be called
637 	 * with the topology lock held.
638 	 */
639 	xpt_unlock_buses();
640 
641 	/*
642 	 * We need to call the peripheral destructor prior to removing the
643 	 * peripheral from the list.  Otherwise, we risk running into a
644 	 * scenario where the peripheral unit number may get reused
645 	 * (because it has been removed from the list), but some resources
646 	 * used by the peripheral are still hanging around.  In particular,
647 	 * the devfs nodes used by some peripherals like the pass(4) driver
648 	 * aren't fully cleaned up until the destructor is run.  If the
649 	 * unit number is reused before the devfs instance is fully gone,
650 	 * devfs will panic.
651 	 */
652 	if (periph->periph_dtor != NULL)
653 		periph->periph_dtor(periph);
654 
655 	/*
656 	 * The peripheral list is protected by the topology lock.
657 	 */
658 	xpt_lock_buses();
659 
660 	TAILQ_REMOVE(&(*p_drv)->units, periph, unit_links);
661 	(*p_drv)->generation++;
662 
663 	xpt_remove_periph(periph);
664 
665 	xpt_unlock_buses();
666 	if ((periph->flags & CAM_PERIPH_ANNOUNCED) && !rebooting)
667 		xpt_print(periph->path, "Periph destroyed\n");
668 	else
669 		CAM_DEBUG(periph->path, CAM_DEBUG_INFO, ("Periph destroyed\n"));
670 
671 	if (periph->flags & CAM_PERIPH_NEW_DEV_FOUND) {
672 		union ccb ccb;
673 		void *arg;
674 
675 		switch (periph->deferred_ac) {
676 		case AC_FOUND_DEVICE:
677 			ccb.ccb_h.func_code = XPT_GDEV_TYPE;
678 			xpt_setup_ccb(&ccb.ccb_h, periph->path, CAM_PRIORITY_NORMAL);
679 			xpt_action(&ccb);
680 			arg = &ccb;
681 			break;
682 		case AC_PATH_REGISTERED:
683 			ccb.ccb_h.func_code = XPT_PATH_INQ;
684 			xpt_setup_ccb(&ccb.ccb_h, periph->path, CAM_PRIORITY_NORMAL);
685 			xpt_action(&ccb);
686 			arg = &ccb;
687 			break;
688 		default:
689 			arg = NULL;
690 			break;
691 		}
692 		periph->deferred_callback(NULL, periph->deferred_ac,
693 					  periph->path, arg);
694 	}
695 	xpt_free_path(periph->path);
696 	free(periph, M_CAMPERIPH);
697 	xpt_lock_buses();
698 }
699 
700 /*
701  * Map user virtual pointers into kernel virtual address space, so we can
702  * access the memory.  This is now a generic function that centralizes most
703  * of the sanity checks on the data flags, if any.
704  * This also only works for up to MAXPHYS memory.  Since we use
705  * buffers to map stuff in and out, we're limited to the buffer size.
706  */
707 int
708 cam_periph_mapmem(union ccb *ccb, struct cam_periph_map_info *mapinfo)
709 {
710 	int numbufs, i, j;
711 	int flags[CAM_PERIPH_MAXMAPS];
712 	u_int8_t **data_ptrs[CAM_PERIPH_MAXMAPS];
713 	u_int32_t lengths[CAM_PERIPH_MAXMAPS];
714 	u_int32_t dirs[CAM_PERIPH_MAXMAPS];
715 	/* Some controllers may not be able to handle more data. */
716 	size_t maxmap = DFLTPHYS;
717 
718 	switch(ccb->ccb_h.func_code) {
719 	case XPT_DEV_MATCH:
720 		if (ccb->cdm.match_buf_len == 0) {
721 			printf("cam_periph_mapmem: invalid match buffer "
722 			       "length 0\n");
723 			return(EINVAL);
724 		}
725 		if (ccb->cdm.pattern_buf_len > 0) {
726 			data_ptrs[0] = (u_int8_t **)&ccb->cdm.patterns;
727 			lengths[0] = ccb->cdm.pattern_buf_len;
728 			dirs[0] = CAM_DIR_OUT;
729 			data_ptrs[1] = (u_int8_t **)&ccb->cdm.matches;
730 			lengths[1] = ccb->cdm.match_buf_len;
731 			dirs[1] = CAM_DIR_IN;
732 			numbufs = 2;
733 		} else {
734 			data_ptrs[0] = (u_int8_t **)&ccb->cdm.matches;
735 			lengths[0] = ccb->cdm.match_buf_len;
736 			dirs[0] = CAM_DIR_IN;
737 			numbufs = 1;
738 		}
739 		/*
740 		 * This request will not go to the hardware, no reason
741 		 * to be so strict. vmapbuf() is able to map up to MAXPHYS.
742 		 */
743 		maxmap = MAXPHYS;
744 		break;
745 	case XPT_SCSI_IO:
746 	case XPT_CONT_TARGET_IO:
747 		if ((ccb->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_NONE)
748 			return(0);
749 		if ((ccb->ccb_h.flags & CAM_DATA_MASK) != CAM_DATA_VADDR)
750 			return (EINVAL);
751 		data_ptrs[0] = &ccb->csio.data_ptr;
752 		lengths[0] = ccb->csio.dxfer_len;
753 		dirs[0] = ccb->ccb_h.flags & CAM_DIR_MASK;
754 		numbufs = 1;
755 		break;
756 	case XPT_ATA_IO:
757 		if ((ccb->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_NONE)
758 			return(0);
759 		if ((ccb->ccb_h.flags & CAM_DATA_MASK) != CAM_DATA_VADDR)
760 			return (EINVAL);
761 		data_ptrs[0] = &ccb->ataio.data_ptr;
762 		lengths[0] = ccb->ataio.dxfer_len;
763 		dirs[0] = ccb->ccb_h.flags & CAM_DIR_MASK;
764 		numbufs = 1;
765 		break;
766 	case XPT_SMP_IO:
767 		data_ptrs[0] = &ccb->smpio.smp_request;
768 		lengths[0] = ccb->smpio.smp_request_len;
769 		dirs[0] = CAM_DIR_OUT;
770 		data_ptrs[1] = &ccb->smpio.smp_response;
771 		lengths[1] = ccb->smpio.smp_response_len;
772 		dirs[1] = CAM_DIR_IN;
773 		numbufs = 2;
774 		break;
775 	case XPT_DEV_ADVINFO:
776 		if (ccb->cdai.bufsiz == 0)
777 			return (0);
778 
779 		data_ptrs[0] = (uint8_t **)&ccb->cdai.buf;
780 		lengths[0] = ccb->cdai.bufsiz;
781 		dirs[0] = CAM_DIR_IN;
782 		numbufs = 1;
783 
784 		/*
785 		 * This request will not go to the hardware, no reason
786 		 * to be so strict. vmapbuf() is able to map up to MAXPHYS.
787 		 */
788 		maxmap = MAXPHYS;
789 		break;
790 	default:
791 		return(EINVAL);
792 		break; /* NOTREACHED */
793 	}
794 
795 	/*
796 	 * Check the transfer length and permissions first, so we don't
797 	 * have to unmap any previously mapped buffers.
798 	 */
799 	for (i = 0; i < numbufs; i++) {
800 
801 		flags[i] = 0;
802 
803 		/*
804 		 * The userland data pointer passed in may not be page
805 		 * aligned.  vmapbuf() truncates the address to a page
806 		 * boundary, so if the address isn't page aligned, we'll
807 		 * need enough space for the given transfer length, plus
808 		 * whatever extra space is necessary to make it to the page
809 		 * boundary.
810 		 */
811 		if ((lengths[i] +
812 		    (((vm_offset_t)(*data_ptrs[i])) & PAGE_MASK)) > maxmap){
813 			printf("cam_periph_mapmem: attempt to map %lu bytes, "
814 			       "which is greater than %lu\n",
815 			       (long)(lengths[i] +
816 			       (((vm_offset_t)(*data_ptrs[i])) & PAGE_MASK)),
817 			       (u_long)maxmap);
818 			return(E2BIG);
819 		}
820 
821 		if (dirs[i] & CAM_DIR_OUT) {
822 			flags[i] = BIO_WRITE;
823 		}
824 
825 		if (dirs[i] & CAM_DIR_IN) {
826 			flags[i] = BIO_READ;
827 		}
828 
829 	}
830 
831 	/*
832 	 * This keeps the the kernel stack of current thread from getting
833 	 * swapped.  In low-memory situations where the kernel stack might
834 	 * otherwise get swapped out, this holds it and allows the thread
835 	 * to make progress and release the kernel mapped pages sooner.
836 	 *
837 	 * XXX KDM should I use P_NOSWAP instead?
838 	 */
839 	PHOLD(curproc);
840 
841 	for (i = 0; i < numbufs; i++) {
842 		/*
843 		 * Get the buffer.
844 		 */
845 		mapinfo->bp[i] = getpbuf(NULL);
846 
847 		/* save the buffer's data address */
848 		mapinfo->bp[i]->b_saveaddr = mapinfo->bp[i]->b_data;
849 
850 		/* put our pointer in the data slot */
851 		mapinfo->bp[i]->b_data = *data_ptrs[i];
852 
853 		/* set the transfer length, we know it's < MAXPHYS */
854 		mapinfo->bp[i]->b_bufsize = lengths[i];
855 
856 		/* set the direction */
857 		mapinfo->bp[i]->b_iocmd = flags[i];
858 
859 		/*
860 		 * Map the buffer into kernel memory.
861 		 *
862 		 * Note that useracc() alone is not a  sufficient test.
863 		 * vmapbuf() can still fail due to a smaller file mapped
864 		 * into a larger area of VM, or if userland races against
865 		 * vmapbuf() after the useracc() check.
866 		 */
867 		if (vmapbuf(mapinfo->bp[i], 1) < 0) {
868 			for (j = 0; j < i; ++j) {
869 				*data_ptrs[j] = mapinfo->bp[j]->b_saveaddr;
870 				vunmapbuf(mapinfo->bp[j]);
871 				relpbuf(mapinfo->bp[j], NULL);
872 			}
873 			relpbuf(mapinfo->bp[i], NULL);
874 			PRELE(curproc);
875 			return(EACCES);
876 		}
877 
878 		/* set our pointer to the new mapped area */
879 		*data_ptrs[i] = mapinfo->bp[i]->b_data;
880 
881 		mapinfo->num_bufs_used++;
882 	}
883 
884 	/*
885 	 * Now that we've gotten this far, change ownership to the kernel
886 	 * of the buffers so that we don't run afoul of returning to user
887 	 * space with locks (on the buffer) held.
888 	 */
889 	for (i = 0; i < numbufs; i++) {
890 		BUF_KERNPROC(mapinfo->bp[i]);
891 	}
892 
893 
894 	return(0);
895 }
896 
897 /*
898  * Unmap memory segments mapped into kernel virtual address space by
899  * cam_periph_mapmem().
900  */
901 void
902 cam_periph_unmapmem(union ccb *ccb, struct cam_periph_map_info *mapinfo)
903 {
904 	int numbufs, i;
905 	u_int8_t **data_ptrs[CAM_PERIPH_MAXMAPS];
906 
907 	if (mapinfo->num_bufs_used <= 0) {
908 		/* nothing to free and the process wasn't held. */
909 		return;
910 	}
911 
912 	switch (ccb->ccb_h.func_code) {
913 	case XPT_DEV_MATCH:
914 		numbufs = min(mapinfo->num_bufs_used, 2);
915 
916 		if (numbufs == 1) {
917 			data_ptrs[0] = (u_int8_t **)&ccb->cdm.matches;
918 		} else {
919 			data_ptrs[0] = (u_int8_t **)&ccb->cdm.patterns;
920 			data_ptrs[1] = (u_int8_t **)&ccb->cdm.matches;
921 		}
922 		break;
923 	case XPT_SCSI_IO:
924 	case XPT_CONT_TARGET_IO:
925 		data_ptrs[0] = &ccb->csio.data_ptr;
926 		numbufs = min(mapinfo->num_bufs_used, 1);
927 		break;
928 	case XPT_ATA_IO:
929 		data_ptrs[0] = &ccb->ataio.data_ptr;
930 		numbufs = min(mapinfo->num_bufs_used, 1);
931 		break;
932 	case XPT_SMP_IO:
933 		numbufs = min(mapinfo->num_bufs_used, 2);
934 		data_ptrs[0] = &ccb->smpio.smp_request;
935 		data_ptrs[1] = &ccb->smpio.smp_response;
936 		break;
937 	case XPT_DEV_ADVINFO:
938 		numbufs = min(mapinfo->num_bufs_used, 1);
939 		data_ptrs[0] = (uint8_t **)&ccb->cdai.buf;
940 		break;
941 	default:
942 		/* allow ourselves to be swapped once again */
943 		PRELE(curproc);
944 		return;
945 		break; /* NOTREACHED */
946 	}
947 
948 	for (i = 0; i < numbufs; i++) {
949 		/* Set the user's pointer back to the original value */
950 		*data_ptrs[i] = mapinfo->bp[i]->b_saveaddr;
951 
952 		/* unmap the buffer */
953 		vunmapbuf(mapinfo->bp[i]);
954 
955 		/* release the buffer */
956 		relpbuf(mapinfo->bp[i], NULL);
957 	}
958 
959 	/* allow ourselves to be swapped once again */
960 	PRELE(curproc);
961 }
962 
963 void
964 cam_periph_ccbwait(union ccb *ccb)
965 {
966 
967 	if ((ccb->ccb_h.pinfo.index != CAM_UNQUEUED_INDEX)
968 	 || ((ccb->ccb_h.status & CAM_STATUS_MASK) == CAM_REQ_INPROG))
969 		xpt_path_sleep(ccb->ccb_h.path, &ccb->ccb_h.cbfcnp, PRIBIO,
970 		    "cbwait", 0);
971 }
972 
973 int
974 cam_periph_ioctl(struct cam_periph *periph, u_long cmd, caddr_t addr,
975 		 int (*error_routine)(union ccb *ccb,
976 				      cam_flags camflags,
977 				      u_int32_t sense_flags))
978 {
979 	union ccb 	     *ccb;
980 	int 		     error;
981 	int		     found;
982 
983 	error = found = 0;
984 
985 	switch(cmd){
986 	case CAMGETPASSTHRU:
987 		ccb = cam_periph_getccb(periph, CAM_PRIORITY_NORMAL);
988 		xpt_setup_ccb(&ccb->ccb_h,
989 			      ccb->ccb_h.path,
990 			      CAM_PRIORITY_NORMAL);
991 		ccb->ccb_h.func_code = XPT_GDEVLIST;
992 
993 		/*
994 		 * Basically, the point of this is that we go through
995 		 * getting the list of devices, until we find a passthrough
996 		 * device.  In the current version of the CAM code, the
997 		 * only way to determine what type of device we're dealing
998 		 * with is by its name.
999 		 */
1000 		while (found == 0) {
1001 			ccb->cgdl.index = 0;
1002 			ccb->cgdl.status = CAM_GDEVLIST_MORE_DEVS;
1003 			while (ccb->cgdl.status == CAM_GDEVLIST_MORE_DEVS) {
1004 
1005 				/* we want the next device in the list */
1006 				xpt_action(ccb);
1007 				if (strncmp(ccb->cgdl.periph_name,
1008 				    "pass", 4) == 0){
1009 					found = 1;
1010 					break;
1011 				}
1012 			}
1013 			if ((ccb->cgdl.status == CAM_GDEVLIST_LAST_DEVICE) &&
1014 			    (found == 0)) {
1015 				ccb->cgdl.periph_name[0] = '\0';
1016 				ccb->cgdl.unit_number = 0;
1017 				break;
1018 			}
1019 		}
1020 
1021 		/* copy the result back out */
1022 		bcopy(ccb, addr, sizeof(union ccb));
1023 
1024 		/* and release the ccb */
1025 		xpt_release_ccb(ccb);
1026 
1027 		break;
1028 	default:
1029 		error = ENOTTY;
1030 		break;
1031 	}
1032 	return(error);
1033 }
1034 
1035 static void
1036 cam_periph_done(struct cam_periph *periph, union ccb *done_ccb)
1037 {
1038 
1039 	/* Caller will release the CCB */
1040 	wakeup(&done_ccb->ccb_h.cbfcnp);
1041 }
1042 
1043 int
1044 cam_periph_runccb(union ccb *ccb,
1045 		  int (*error_routine)(union ccb *ccb,
1046 				       cam_flags camflags,
1047 				       u_int32_t sense_flags),
1048 		  cam_flags camflags, u_int32_t sense_flags,
1049 		  struct devstat *ds)
1050 {
1051 	struct bintime *starttime;
1052 	struct bintime ltime;
1053 	int error;
1054 
1055 	starttime = NULL;
1056 	xpt_path_assert(ccb->ccb_h.path, MA_OWNED);
1057 
1058 	/*
1059 	 * If the user has supplied a stats structure, and if we understand
1060 	 * this particular type of ccb, record the transaction start.
1061 	 */
1062 	if ((ds != NULL) && (ccb->ccb_h.func_code == XPT_SCSI_IO ||
1063 	    ccb->ccb_h.func_code == XPT_ATA_IO)) {
1064 		starttime = &ltime;
1065 		binuptime(starttime);
1066 		devstat_start_transaction(ds, starttime);
1067 	}
1068 
1069 	ccb->ccb_h.cbfcnp = cam_periph_done;
1070 	xpt_action(ccb);
1071 
1072 	do {
1073 		cam_periph_ccbwait(ccb);
1074 		if ((ccb->ccb_h.status & CAM_STATUS_MASK) == CAM_REQ_CMP)
1075 			error = 0;
1076 		else if (error_routine != NULL)
1077 			error = (*error_routine)(ccb, camflags, sense_flags);
1078 		else
1079 			error = 0;
1080 
1081 	} while (error == ERESTART);
1082 
1083 	if ((ccb->ccb_h.status & CAM_DEV_QFRZN) != 0) {
1084 		cam_release_devq(ccb->ccb_h.path,
1085 				 /* relsim_flags */0,
1086 				 /* openings */0,
1087 				 /* timeout */0,
1088 				 /* getcount_only */ FALSE);
1089 		ccb->ccb_h.status &= ~CAM_DEV_QFRZN;
1090 	}
1091 
1092 	if (ds != NULL) {
1093 		if (ccb->ccb_h.func_code == XPT_SCSI_IO) {
1094 			devstat_end_transaction(ds,
1095 					ccb->csio.dxfer_len - ccb->csio.resid,
1096 					ccb->csio.tag_action & 0x3,
1097 					((ccb->ccb_h.flags & CAM_DIR_MASK) ==
1098 					CAM_DIR_NONE) ?  DEVSTAT_NO_DATA :
1099 					(ccb->ccb_h.flags & CAM_DIR_OUT) ?
1100 					DEVSTAT_WRITE :
1101 					DEVSTAT_READ, NULL, starttime);
1102 		} else if (ccb->ccb_h.func_code == XPT_ATA_IO) {
1103 			devstat_end_transaction(ds,
1104 					ccb->ataio.dxfer_len - ccb->ataio.resid,
1105 					ccb->ataio.tag_action & 0x3,
1106 					((ccb->ccb_h.flags & CAM_DIR_MASK) ==
1107 					CAM_DIR_NONE) ?  DEVSTAT_NO_DATA :
1108 					(ccb->ccb_h.flags & CAM_DIR_OUT) ?
1109 					DEVSTAT_WRITE :
1110 					DEVSTAT_READ, NULL, starttime);
1111 		}
1112 	}
1113 
1114 	return(error);
1115 }
1116 
1117 void
1118 cam_freeze_devq(struct cam_path *path)
1119 {
1120 	struct ccb_hdr ccb_h;
1121 
1122 	CAM_DEBUG(path, CAM_DEBUG_TRACE, ("cam_freeze_devq\n"));
1123 	xpt_setup_ccb(&ccb_h, path, /*priority*/1);
1124 	ccb_h.func_code = XPT_NOOP;
1125 	ccb_h.flags = CAM_DEV_QFREEZE;
1126 	xpt_action((union ccb *)&ccb_h);
1127 }
1128 
1129 u_int32_t
1130 cam_release_devq(struct cam_path *path, u_int32_t relsim_flags,
1131 		 u_int32_t openings, u_int32_t arg,
1132 		 int getcount_only)
1133 {
1134 	struct ccb_relsim crs;
1135 
1136 	CAM_DEBUG(path, CAM_DEBUG_TRACE, ("cam_release_devq(%u, %u, %u, %d)\n",
1137 	    relsim_flags, openings, arg, getcount_only));
1138 	xpt_setup_ccb(&crs.ccb_h, path, CAM_PRIORITY_NORMAL);
1139 	crs.ccb_h.func_code = XPT_REL_SIMQ;
1140 	crs.ccb_h.flags = getcount_only ? CAM_DEV_QFREEZE : 0;
1141 	crs.release_flags = relsim_flags;
1142 	crs.openings = openings;
1143 	crs.release_timeout = arg;
1144 	xpt_action((union ccb *)&crs);
1145 	return (crs.qfrozen_cnt);
1146 }
1147 
1148 #define saved_ccb_ptr ppriv_ptr0
1149 static void
1150 camperiphdone(struct cam_periph *periph, union ccb *done_ccb)
1151 {
1152 	union ccb      *saved_ccb;
1153 	cam_status	status;
1154 	struct scsi_start_stop_unit *scsi_cmd;
1155 	int    error_code, sense_key, asc, ascq;
1156 
1157 	scsi_cmd = (struct scsi_start_stop_unit *)
1158 	    &done_ccb->csio.cdb_io.cdb_bytes;
1159 	status = done_ccb->ccb_h.status;
1160 
1161 	if ((status & CAM_STATUS_MASK) != CAM_REQ_CMP) {
1162 		if (scsi_extract_sense_ccb(done_ccb,
1163 		    &error_code, &sense_key, &asc, &ascq)) {
1164 			/*
1165 			 * If the error is "invalid field in CDB",
1166 			 * and the load/eject flag is set, turn the
1167 			 * flag off and try again.  This is just in
1168 			 * case the drive in question barfs on the
1169 			 * load eject flag.  The CAM code should set
1170 			 * the load/eject flag by default for
1171 			 * removable media.
1172 			 */
1173 			if ((scsi_cmd->opcode == START_STOP_UNIT) &&
1174 			    ((scsi_cmd->how & SSS_LOEJ) != 0) &&
1175 			     (asc == 0x24) && (ascq == 0x00)) {
1176 				scsi_cmd->how &= ~SSS_LOEJ;
1177 				if (status & CAM_DEV_QFRZN) {
1178 					cam_release_devq(done_ccb->ccb_h.path,
1179 					    0, 0, 0, 0);
1180 					done_ccb->ccb_h.status &=
1181 					    ~CAM_DEV_QFRZN;
1182 				}
1183 				xpt_action(done_ccb);
1184 				goto out;
1185 			}
1186 		}
1187 		if (cam_periph_error(done_ccb,
1188 		    0, SF_RETRY_UA | SF_NO_PRINT, NULL) == ERESTART)
1189 			goto out;
1190 		if (done_ccb->ccb_h.status & CAM_DEV_QFRZN) {
1191 			cam_release_devq(done_ccb->ccb_h.path, 0, 0, 0, 0);
1192 			done_ccb->ccb_h.status &= ~CAM_DEV_QFRZN;
1193 		}
1194 	} else {
1195 		/*
1196 		 * If we have successfully taken a device from the not
1197 		 * ready to ready state, re-scan the device and re-get
1198 		 * the inquiry information.  Many devices (mostly disks)
1199 		 * don't properly report their inquiry information unless
1200 		 * they are spun up.
1201 		 */
1202 		if (scsi_cmd->opcode == START_STOP_UNIT)
1203 			xpt_async(AC_INQ_CHANGED, done_ccb->ccb_h.path, NULL);
1204 	}
1205 
1206 	/*
1207 	 * Perform the final retry with the original CCB so that final
1208 	 * error processing is performed by the owner of the CCB.
1209 	 */
1210 	saved_ccb = (union ccb *)done_ccb->ccb_h.saved_ccb_ptr;
1211 	bcopy(saved_ccb, done_ccb, sizeof(*done_ccb));
1212 	xpt_free_ccb(saved_ccb);
1213 	if (done_ccb->ccb_h.cbfcnp != camperiphdone)
1214 		periph->flags &= ~CAM_PERIPH_RECOVERY_INPROG;
1215 	xpt_action(done_ccb);
1216 
1217 out:
1218 	/* Drop freeze taken due to CAM_DEV_QFREEZE flag set. */
1219 	cam_release_devq(done_ccb->ccb_h.path, 0, 0, 0, 0);
1220 }
1221 
1222 /*
1223  * Generic Async Event handler.  Peripheral drivers usually
1224  * filter out the events that require personal attention,
1225  * and leave the rest to this function.
1226  */
1227 void
1228 cam_periph_async(struct cam_periph *periph, u_int32_t code,
1229 		 struct cam_path *path, void *arg)
1230 {
1231 	switch (code) {
1232 	case AC_LOST_DEVICE:
1233 		cam_periph_invalidate(periph);
1234 		break;
1235 	default:
1236 		break;
1237 	}
1238 }
1239 
1240 void
1241 cam_periph_bus_settle(struct cam_periph *periph, u_int bus_settle)
1242 {
1243 	struct ccb_getdevstats cgds;
1244 
1245 	xpt_setup_ccb(&cgds.ccb_h, periph->path, CAM_PRIORITY_NORMAL);
1246 	cgds.ccb_h.func_code = XPT_GDEV_STATS;
1247 	xpt_action((union ccb *)&cgds);
1248 	cam_periph_freeze_after_event(periph, &cgds.last_reset, bus_settle);
1249 }
1250 
1251 void
1252 cam_periph_freeze_after_event(struct cam_periph *periph,
1253 			      struct timeval* event_time, u_int duration_ms)
1254 {
1255 	struct timeval delta;
1256 	struct timeval duration_tv;
1257 
1258 	if (!timevalisset(event_time))
1259 		return;
1260 
1261 	microtime(&delta);
1262 	timevalsub(&delta, event_time);
1263 	duration_tv.tv_sec = duration_ms / 1000;
1264 	duration_tv.tv_usec = (duration_ms % 1000) * 1000;
1265 	if (timevalcmp(&delta, &duration_tv, <)) {
1266 		timevalsub(&duration_tv, &delta);
1267 
1268 		duration_ms = duration_tv.tv_sec * 1000;
1269 		duration_ms += duration_tv.tv_usec / 1000;
1270 		cam_freeze_devq(periph->path);
1271 		cam_release_devq(periph->path,
1272 				RELSIM_RELEASE_AFTER_TIMEOUT,
1273 				/*reduction*/0,
1274 				/*timeout*/duration_ms,
1275 				/*getcount_only*/0);
1276 	}
1277 
1278 }
1279 
1280 static int
1281 camperiphscsistatuserror(union ccb *ccb, union ccb **orig_ccb,
1282     cam_flags camflags, u_int32_t sense_flags,
1283     int *openings, u_int32_t *relsim_flags,
1284     u_int32_t *timeout, u_int32_t *action, const char **action_string)
1285 {
1286 	int error;
1287 
1288 	switch (ccb->csio.scsi_status) {
1289 	case SCSI_STATUS_OK:
1290 	case SCSI_STATUS_COND_MET:
1291 	case SCSI_STATUS_INTERMED:
1292 	case SCSI_STATUS_INTERMED_COND_MET:
1293 		error = 0;
1294 		break;
1295 	case SCSI_STATUS_CMD_TERMINATED:
1296 	case SCSI_STATUS_CHECK_COND:
1297 		error = camperiphscsisenseerror(ccb, orig_ccb,
1298 					        camflags,
1299 					        sense_flags,
1300 					        openings,
1301 					        relsim_flags,
1302 					        timeout,
1303 					        action,
1304 					        action_string);
1305 		break;
1306 	case SCSI_STATUS_QUEUE_FULL:
1307 	{
1308 		/* no decrement */
1309 		struct ccb_getdevstats cgds;
1310 
1311 		/*
1312 		 * First off, find out what the current
1313 		 * transaction counts are.
1314 		 */
1315 		xpt_setup_ccb(&cgds.ccb_h,
1316 			      ccb->ccb_h.path,
1317 			      CAM_PRIORITY_NORMAL);
1318 		cgds.ccb_h.func_code = XPT_GDEV_STATS;
1319 		xpt_action((union ccb *)&cgds);
1320 
1321 		/*
1322 		 * If we were the only transaction active, treat
1323 		 * the QUEUE FULL as if it were a BUSY condition.
1324 		 */
1325 		if (cgds.dev_active != 0) {
1326 			int total_openings;
1327 
1328 			/*
1329 		 	 * Reduce the number of openings to
1330 			 * be 1 less than the amount it took
1331 			 * to get a queue full bounded by the
1332 			 * minimum allowed tag count for this
1333 			 * device.
1334 		 	 */
1335 			total_openings = cgds.dev_active + cgds.dev_openings;
1336 			*openings = cgds.dev_active;
1337 			if (*openings < cgds.mintags)
1338 				*openings = cgds.mintags;
1339 			if (*openings < total_openings)
1340 				*relsim_flags = RELSIM_ADJUST_OPENINGS;
1341 			else {
1342 				/*
1343 				 * Some devices report queue full for
1344 				 * temporary resource shortages.  For
1345 				 * this reason, we allow a minimum
1346 				 * tag count to be entered via a
1347 				 * quirk entry to prevent the queue
1348 				 * count on these devices from falling
1349 				 * to a pessimisticly low value.  We
1350 				 * still wait for the next successful
1351 				 * completion, however, before queueing
1352 				 * more transactions to the device.
1353 				 */
1354 				*relsim_flags = RELSIM_RELEASE_AFTER_CMDCMPLT;
1355 			}
1356 			*timeout = 0;
1357 			error = ERESTART;
1358 			*action &= ~SSQ_PRINT_SENSE;
1359 			break;
1360 		}
1361 		/* FALLTHROUGH */
1362 	}
1363 	case SCSI_STATUS_BUSY:
1364 		/*
1365 		 * Restart the queue after either another
1366 		 * command completes or a 1 second timeout.
1367 		 */
1368 		if ((sense_flags & SF_RETRY_BUSY) != 0 ||
1369 		    (ccb->ccb_h.retry_count--) > 0) {
1370 			error = ERESTART;
1371 			*relsim_flags = RELSIM_RELEASE_AFTER_TIMEOUT
1372 				      | RELSIM_RELEASE_AFTER_CMDCMPLT;
1373 			*timeout = 1000;
1374 		} else {
1375 			error = EIO;
1376 		}
1377 		break;
1378 	case SCSI_STATUS_RESERV_CONFLICT:
1379 	default:
1380 		error = EIO;
1381 		break;
1382 	}
1383 	return (error);
1384 }
1385 
1386 static int
1387 camperiphscsisenseerror(union ccb *ccb, union ccb **orig,
1388     cam_flags camflags, u_int32_t sense_flags,
1389     int *openings, u_int32_t *relsim_flags,
1390     u_int32_t *timeout, u_int32_t *action, const char **action_string)
1391 {
1392 	struct cam_periph *periph;
1393 	union ccb *orig_ccb = ccb;
1394 	int error, recoveryccb;
1395 
1396 	periph = xpt_path_periph(ccb->ccb_h.path);
1397 	recoveryccb = (ccb->ccb_h.cbfcnp == camperiphdone);
1398 	if ((periph->flags & CAM_PERIPH_RECOVERY_INPROG) && !recoveryccb) {
1399 		/*
1400 		 * If error recovery is already in progress, don't attempt
1401 		 * to process this error, but requeue it unconditionally
1402 		 * and attempt to process it once error recovery has
1403 		 * completed.  This failed command is probably related to
1404 		 * the error that caused the currently active error recovery
1405 		 * action so our  current recovery efforts should also
1406 		 * address this command.  Be aware that the error recovery
1407 		 * code assumes that only one recovery action is in progress
1408 		 * on a particular peripheral instance at any given time
1409 		 * (e.g. only one saved CCB for error recovery) so it is
1410 		 * imperitive that we don't violate this assumption.
1411 		 */
1412 		error = ERESTART;
1413 		*action &= ~SSQ_PRINT_SENSE;
1414 	} else {
1415 		scsi_sense_action err_action;
1416 		struct ccb_getdev cgd;
1417 
1418 		/*
1419 		 * Grab the inquiry data for this device.
1420 		 */
1421 		xpt_setup_ccb(&cgd.ccb_h, ccb->ccb_h.path, CAM_PRIORITY_NORMAL);
1422 		cgd.ccb_h.func_code = XPT_GDEV_TYPE;
1423 		xpt_action((union ccb *)&cgd);
1424 
1425 		err_action = scsi_error_action(&ccb->csio, &cgd.inq_data,
1426 		    sense_flags);
1427 		error = err_action & SS_ERRMASK;
1428 
1429 		/*
1430 		 * Do not autostart sequential access devices
1431 		 * to avoid unexpected tape loading.
1432 		 */
1433 		if ((err_action & SS_MASK) == SS_START &&
1434 		    SID_TYPE(&cgd.inq_data) == T_SEQUENTIAL) {
1435 			*action_string = "Will not autostart a "
1436 			    "sequential access device";
1437 			goto sense_error_done;
1438 		}
1439 
1440 		/*
1441 		 * Avoid recovery recursion if recovery action is the same.
1442 		 */
1443 		if ((err_action & SS_MASK) >= SS_START && recoveryccb) {
1444 			if (((err_action & SS_MASK) == SS_START &&
1445 			     ccb->csio.cdb_io.cdb_bytes[0] == START_STOP_UNIT) ||
1446 			    ((err_action & SS_MASK) == SS_TUR &&
1447 			     (ccb->csio.cdb_io.cdb_bytes[0] == TEST_UNIT_READY))) {
1448 				err_action = SS_RETRY|SSQ_DECREMENT_COUNT|EIO;
1449 				*relsim_flags = RELSIM_RELEASE_AFTER_TIMEOUT;
1450 				*timeout = 500;
1451 			}
1452 		}
1453 
1454 		/*
1455 		 * If the recovery action will consume a retry,
1456 		 * make sure we actually have retries available.
1457 		 */
1458 		if ((err_action & SSQ_DECREMENT_COUNT) != 0) {
1459 		 	if (ccb->ccb_h.retry_count > 0 &&
1460 			    (periph->flags & CAM_PERIPH_INVALID) == 0)
1461 		 		ccb->ccb_h.retry_count--;
1462 			else {
1463 				*action_string = "Retries exhausted";
1464 				goto sense_error_done;
1465 			}
1466 		}
1467 
1468 		if ((err_action & SS_MASK) >= SS_START) {
1469 			/*
1470 			 * Do common portions of commands that
1471 			 * use recovery CCBs.
1472 			 */
1473 			orig_ccb = xpt_alloc_ccb_nowait();
1474 			if (orig_ccb == NULL) {
1475 				*action_string = "Can't allocate recovery CCB";
1476 				goto sense_error_done;
1477 			}
1478 			/*
1479 			 * Clear freeze flag for original request here, as
1480 			 * this freeze will be dropped as part of ERESTART.
1481 			 */
1482 			ccb->ccb_h.status &= ~CAM_DEV_QFRZN;
1483 			bcopy(ccb, orig_ccb, sizeof(*orig_ccb));
1484 		}
1485 
1486 		switch (err_action & SS_MASK) {
1487 		case SS_NOP:
1488 			*action_string = "No recovery action needed";
1489 			error = 0;
1490 			break;
1491 		case SS_RETRY:
1492 			*action_string = "Retrying command (per sense data)";
1493 			error = ERESTART;
1494 			break;
1495 		case SS_FAIL:
1496 			*action_string = "Unretryable error";
1497 			break;
1498 		case SS_START:
1499 		{
1500 			int le;
1501 
1502 			/*
1503 			 * Send a start unit command to the device, and
1504 			 * then retry the command.
1505 			 */
1506 			*action_string = "Attempting to start unit";
1507 			periph->flags |= CAM_PERIPH_RECOVERY_INPROG;
1508 
1509 			/*
1510 			 * Check for removable media and set
1511 			 * load/eject flag appropriately.
1512 			 */
1513 			if (SID_IS_REMOVABLE(&cgd.inq_data))
1514 				le = TRUE;
1515 			else
1516 				le = FALSE;
1517 
1518 			scsi_start_stop(&ccb->csio,
1519 					/*retries*/1,
1520 					camperiphdone,
1521 					MSG_SIMPLE_Q_TAG,
1522 					/*start*/TRUE,
1523 					/*load/eject*/le,
1524 					/*immediate*/FALSE,
1525 					SSD_FULL_SIZE,
1526 					/*timeout*/50000);
1527 			break;
1528 		}
1529 		case SS_TUR:
1530 		{
1531 			/*
1532 			 * Send a Test Unit Ready to the device.
1533 			 * If the 'many' flag is set, we send 120
1534 			 * test unit ready commands, one every half
1535 			 * second.  Otherwise, we just send one TUR.
1536 			 * We only want to do this if the retry
1537 			 * count has not been exhausted.
1538 			 */
1539 			int retries;
1540 
1541 			if ((err_action & SSQ_MANY) != 0) {
1542 				*action_string = "Polling device for readiness";
1543 				retries = 120;
1544 			} else {
1545 				*action_string = "Testing device for readiness";
1546 				retries = 1;
1547 			}
1548 			periph->flags |= CAM_PERIPH_RECOVERY_INPROG;
1549 			scsi_test_unit_ready(&ccb->csio,
1550 					     retries,
1551 					     camperiphdone,
1552 					     MSG_SIMPLE_Q_TAG,
1553 					     SSD_FULL_SIZE,
1554 					     /*timeout*/5000);
1555 
1556 			/*
1557 			 * Accomplish our 500ms delay by deferring
1558 			 * the release of our device queue appropriately.
1559 			 */
1560 			*relsim_flags = RELSIM_RELEASE_AFTER_TIMEOUT;
1561 			*timeout = 500;
1562 			break;
1563 		}
1564 		default:
1565 			panic("Unhandled error action %x", err_action);
1566 		}
1567 
1568 		if ((err_action & SS_MASK) >= SS_START) {
1569 			/*
1570 			 * Drop the priority, so that the recovery
1571 			 * CCB is the first to execute.  Freeze the queue
1572 			 * after this command is sent so that we can
1573 			 * restore the old csio and have it queued in
1574 			 * the proper order before we release normal
1575 			 * transactions to the device.
1576 			 */
1577 			ccb->ccb_h.pinfo.priority--;
1578 			ccb->ccb_h.flags |= CAM_DEV_QFREEZE;
1579 			ccb->ccb_h.saved_ccb_ptr = orig_ccb;
1580 			error = ERESTART;
1581 			*orig = orig_ccb;
1582 		}
1583 
1584 sense_error_done:
1585 		*action = err_action;
1586 	}
1587 	return (error);
1588 }
1589 
1590 /*
1591  * Generic error handler.  Peripheral drivers usually filter
1592  * out the errors that they handle in a unique mannor, then
1593  * call this function.
1594  */
1595 int
1596 cam_periph_error(union ccb *ccb, cam_flags camflags,
1597 		 u_int32_t sense_flags, union ccb *save_ccb)
1598 {
1599 	struct cam_path *newpath;
1600 	union ccb  *orig_ccb, *scan_ccb;
1601 	struct cam_periph *periph;
1602 	const char *action_string;
1603 	cam_status  status;
1604 	int	    frozen, error, openings;
1605 	u_int32_t   action, relsim_flags, timeout;
1606 
1607 	action = SSQ_PRINT_SENSE;
1608 	periph = xpt_path_periph(ccb->ccb_h.path);
1609 	action_string = NULL;
1610 	status = ccb->ccb_h.status;
1611 	frozen = (status & CAM_DEV_QFRZN) != 0;
1612 	status &= CAM_STATUS_MASK;
1613 	openings = relsim_flags = timeout = 0;
1614 	orig_ccb = ccb;
1615 
1616 	switch (status) {
1617 	case CAM_REQ_CMP:
1618 		error = 0;
1619 		action &= ~SSQ_PRINT_SENSE;
1620 		break;
1621 	case CAM_SCSI_STATUS_ERROR:
1622 		error = camperiphscsistatuserror(ccb, &orig_ccb,
1623 		    camflags, sense_flags, &openings, &relsim_flags,
1624 		    &timeout, &action, &action_string);
1625 		break;
1626 	case CAM_AUTOSENSE_FAIL:
1627 		error = EIO;	/* we have to kill the command */
1628 		break;
1629 	case CAM_UA_ABORT:
1630 	case CAM_UA_TERMIO:
1631 	case CAM_MSG_REJECT_REC:
1632 		/* XXX Don't know that these are correct */
1633 		error = EIO;
1634 		break;
1635 	case CAM_SEL_TIMEOUT:
1636 		if ((camflags & CAM_RETRY_SELTO) != 0) {
1637 			if (ccb->ccb_h.retry_count > 0 &&
1638 			    (periph->flags & CAM_PERIPH_INVALID) == 0) {
1639 				ccb->ccb_h.retry_count--;
1640 				error = ERESTART;
1641 
1642 				/*
1643 				 * Wait a bit to give the device
1644 				 * time to recover before we try again.
1645 				 */
1646 				relsim_flags = RELSIM_RELEASE_AFTER_TIMEOUT;
1647 				timeout = periph_selto_delay;
1648 				break;
1649 			}
1650 			action_string = "Retries exhausted";
1651 		}
1652 		/* FALLTHROUGH */
1653 	case CAM_DEV_NOT_THERE:
1654 		error = ENXIO;
1655 		action = SSQ_LOST;
1656 		break;
1657 	case CAM_REQ_INVALID:
1658 	case CAM_PATH_INVALID:
1659 	case CAM_NO_HBA:
1660 	case CAM_PROVIDE_FAIL:
1661 	case CAM_REQ_TOO_BIG:
1662 	case CAM_LUN_INVALID:
1663 	case CAM_TID_INVALID:
1664 	case CAM_FUNC_NOTAVAIL:
1665 		error = EINVAL;
1666 		break;
1667 	case CAM_SCSI_BUS_RESET:
1668 	case CAM_BDR_SENT:
1669 		/*
1670 		 * Commands that repeatedly timeout and cause these
1671 		 * kinds of error recovery actions, should return
1672 		 * CAM_CMD_TIMEOUT, which allows us to safely assume
1673 		 * that this command was an innocent bystander to
1674 		 * these events and should be unconditionally
1675 		 * retried.
1676 		 */
1677 	case CAM_REQUEUE_REQ:
1678 		/* Unconditional requeue if device is still there */
1679 		if (periph->flags & CAM_PERIPH_INVALID) {
1680 			action_string = "Periph was invalidated";
1681 			error = EIO;
1682 		} else if (sense_flags & SF_NO_RETRY) {
1683 			error = EIO;
1684 			action_string = "Retry was blocked";
1685 		} else {
1686 			error = ERESTART;
1687 			action &= ~SSQ_PRINT_SENSE;
1688 		}
1689 		break;
1690 	case CAM_RESRC_UNAVAIL:
1691 		/* Wait a bit for the resource shortage to abate. */
1692 		timeout = periph_noresrc_delay;
1693 		/* FALLTHROUGH */
1694 	case CAM_BUSY:
1695 		if (timeout == 0) {
1696 			/* Wait a bit for the busy condition to abate. */
1697 			timeout = periph_busy_delay;
1698 		}
1699 		relsim_flags = RELSIM_RELEASE_AFTER_TIMEOUT;
1700 		/* FALLTHROUGH */
1701 	case CAM_ATA_STATUS_ERROR:
1702 	case CAM_REQ_CMP_ERR:
1703 	case CAM_CMD_TIMEOUT:
1704 	case CAM_UNEXP_BUSFREE:
1705 	case CAM_UNCOR_PARITY:
1706 	case CAM_DATA_RUN_ERR:
1707 	default:
1708 		if (periph->flags & CAM_PERIPH_INVALID) {
1709 			error = EIO;
1710 			action_string = "Periph was invalidated";
1711 		} else if (ccb->ccb_h.retry_count == 0) {
1712 			error = EIO;
1713 			action_string = "Retries exhausted";
1714 		} else if (sense_flags & SF_NO_RETRY) {
1715 			error = EIO;
1716 			action_string = "Retry was blocked";
1717 		} else {
1718 			ccb->ccb_h.retry_count--;
1719 			error = ERESTART;
1720 		}
1721 		break;
1722 	}
1723 
1724 	if ((sense_flags & SF_PRINT_ALWAYS) ||
1725 	    CAM_DEBUGGED(ccb->ccb_h.path, CAM_DEBUG_INFO))
1726 		action |= SSQ_PRINT_SENSE;
1727 	else if (sense_flags & SF_NO_PRINT)
1728 		action &= ~SSQ_PRINT_SENSE;
1729 	if ((action & SSQ_PRINT_SENSE) != 0)
1730 		cam_error_print(orig_ccb, CAM_ESF_ALL, CAM_EPF_ALL);
1731 	if (error != 0 && (action & SSQ_PRINT_SENSE) != 0) {
1732 		if (error != ERESTART) {
1733 			if (action_string == NULL)
1734 				action_string = "Unretryable error";
1735 			xpt_print(ccb->ccb_h.path, "Error %d, %s\n",
1736 			    error, action_string);
1737 		} else if (action_string != NULL)
1738 			xpt_print(ccb->ccb_h.path, "%s\n", action_string);
1739 		else
1740 			xpt_print(ccb->ccb_h.path, "Retrying command\n");
1741 	}
1742 
1743 	if ((action & SSQ_LOST) != 0) {
1744 		lun_id_t lun_id;
1745 
1746 		/*
1747 		 * For a selection timeout, we consider all of the LUNs on
1748 		 * the target to be gone.  If the status is CAM_DEV_NOT_THERE,
1749 		 * then we only get rid of the device(s) specified by the
1750 		 * path in the original CCB.
1751 		 */
1752 		if (status == CAM_SEL_TIMEOUT)
1753 			lun_id = CAM_LUN_WILDCARD;
1754 		else
1755 			lun_id = xpt_path_lun_id(ccb->ccb_h.path);
1756 
1757 		/* Should we do more if we can't create the path?? */
1758 		if (xpt_create_path(&newpath, periph,
1759 				    xpt_path_path_id(ccb->ccb_h.path),
1760 				    xpt_path_target_id(ccb->ccb_h.path),
1761 				    lun_id) == CAM_REQ_CMP) {
1762 
1763 			/*
1764 			 * Let peripheral drivers know that this
1765 			 * device has gone away.
1766 			 */
1767 			xpt_async(AC_LOST_DEVICE, newpath, NULL);
1768 			xpt_free_path(newpath);
1769 		}
1770 	}
1771 
1772 	/* Broadcast UNIT ATTENTIONs to all periphs. */
1773 	if ((action & SSQ_UA) != 0)
1774 		xpt_async(AC_UNIT_ATTENTION, orig_ccb->ccb_h.path, orig_ccb);
1775 
1776 	/* Rescan target on "Reported LUNs data has changed" */
1777 	if ((action & SSQ_RESCAN) != 0) {
1778 		if (xpt_create_path(&newpath, NULL,
1779 				    xpt_path_path_id(ccb->ccb_h.path),
1780 				    xpt_path_target_id(ccb->ccb_h.path),
1781 				    CAM_LUN_WILDCARD) == CAM_REQ_CMP) {
1782 
1783 			scan_ccb = xpt_alloc_ccb_nowait();
1784 			if (scan_ccb != NULL) {
1785 				scan_ccb->ccb_h.path = newpath;
1786 				scan_ccb->ccb_h.func_code = XPT_SCAN_TGT;
1787 				scan_ccb->crcn.flags = 0;
1788 				xpt_rescan(scan_ccb);
1789 			} else {
1790 				xpt_print(newpath,
1791 				    "Can't allocate CCB to rescan target\n");
1792 				xpt_free_path(newpath);
1793 			}
1794 		}
1795 	}
1796 
1797 	/* Attempt a retry */
1798 	if (error == ERESTART || error == 0) {
1799 		if (frozen != 0)
1800 			ccb->ccb_h.status &= ~CAM_DEV_QFRZN;
1801 		if (error == ERESTART)
1802 			xpt_action(ccb);
1803 		if (frozen != 0)
1804 			cam_release_devq(ccb->ccb_h.path,
1805 					 relsim_flags,
1806 					 openings,
1807 					 timeout,
1808 					 /*getcount_only*/0);
1809 	}
1810 
1811 	return (error);
1812 }
1813