xref: /freebsd/sys/cam/cam_periph.c (revision f4b37ed0f8b307b1f3f0f630ca725d68f1dff30d)
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 again:
112 	ndrivers = nperiph_drivers + 2;
113 	newdrivers = malloc(sizeof(*newdrivers) * ndrivers, M_CAMPERIPH,
114 			    M_WAITOK);
115 	xpt_lock_buses();
116 	if (ndrivers != nperiph_drivers + 2) {
117 		/*
118 		 * Lost race against itself; go around.
119 		 */
120 		xpt_unlock_buses();
121 		free(newdrivers, M_CAMPERIPH);
122 		goto again;
123 	}
124 	if (periph_drivers)
125 		bcopy(periph_drivers, newdrivers,
126 		      sizeof(*newdrivers) * nperiph_drivers);
127 	newdrivers[nperiph_drivers] = drv;
128 	newdrivers[nperiph_drivers + 1] = NULL;
129 	old = periph_drivers;
130 	periph_drivers = newdrivers;
131 	nperiph_drivers++;
132 	xpt_unlock_buses();
133 	if (old)
134 		free(old, M_CAMPERIPH);
135 	/* If driver marked as early or it is late now, initialize it. */
136 	if (((drv->flags & CAM_PERIPH_DRV_EARLY) != 0 && initialized > 0) ||
137 	    initialized > 1)
138 		(*drv->init)();
139 }
140 
141 void
142 periphdriver_init(int level)
143 {
144 	int	i, early;
145 
146 	initialized = max(initialized, level);
147 	for (i = 0; periph_drivers[i] != NULL; i++) {
148 		early = (periph_drivers[i]->flags & CAM_PERIPH_DRV_EARLY) ? 1 : 2;
149 		if (early == initialized)
150 			(*periph_drivers[i]->init)();
151 	}
152 }
153 
154 cam_status
155 cam_periph_alloc(periph_ctor_t *periph_ctor,
156 		 periph_oninv_t *periph_oninvalidate,
157 		 periph_dtor_t *periph_dtor, periph_start_t *periph_start,
158 		 char *name, cam_periph_type type, struct cam_path *path,
159 		 ac_callback_t *ac_callback, ac_code code, void *arg)
160 {
161 	struct		periph_driver **p_drv;
162 	struct		cam_sim *sim;
163 	struct		cam_periph *periph;
164 	struct		cam_periph *cur_periph;
165 	path_id_t	path_id;
166 	target_id_t	target_id;
167 	lun_id_t	lun_id;
168 	cam_status	status;
169 	u_int		init_level;
170 
171 	init_level = 0;
172 	/*
173 	 * Handle Hot-Plug scenarios.  If there is already a peripheral
174 	 * of our type assigned to this path, we are likely waiting for
175 	 * final close on an old, invalidated, peripheral.  If this is
176 	 * the case, queue up a deferred call to the peripheral's async
177 	 * handler.  If it looks like a mistaken re-allocation, complain.
178 	 */
179 	if ((periph = cam_periph_find(path, name)) != NULL) {
180 
181 		if ((periph->flags & CAM_PERIPH_INVALID) != 0
182 		 && (periph->flags & CAM_PERIPH_NEW_DEV_FOUND) == 0) {
183 			periph->flags |= CAM_PERIPH_NEW_DEV_FOUND;
184 			periph->deferred_callback = ac_callback;
185 			periph->deferred_ac = code;
186 			return (CAM_REQ_INPROG);
187 		} else {
188 			printf("cam_periph_alloc: attempt to re-allocate "
189 			       "valid device %s%d rejected flags %#x "
190 			       "refcount %d\n", periph->periph_name,
191 			       periph->unit_number, periph->flags,
192 			       periph->refcount);
193 		}
194 		return (CAM_REQ_INVALID);
195 	}
196 
197 	periph = (struct cam_periph *)malloc(sizeof(*periph), M_CAMPERIPH,
198 					     M_NOWAIT|M_ZERO);
199 
200 	if (periph == NULL)
201 		return (CAM_RESRC_UNAVAIL);
202 
203 	init_level++;
204 
205 
206 	sim = xpt_path_sim(path);
207 	path_id = xpt_path_path_id(path);
208 	target_id = xpt_path_target_id(path);
209 	lun_id = xpt_path_lun_id(path);
210 	periph->periph_start = periph_start;
211 	periph->periph_dtor = periph_dtor;
212 	periph->periph_oninval = periph_oninvalidate;
213 	periph->type = type;
214 	periph->periph_name = name;
215 	periph->scheduled_priority = CAM_PRIORITY_NONE;
216 	periph->immediate_priority = CAM_PRIORITY_NONE;
217 	periph->refcount = 1;		/* Dropped by invalidation. */
218 	periph->sim = sim;
219 	SLIST_INIT(&periph->ccb_list);
220 	status = xpt_create_path(&path, periph, path_id, target_id, lun_id);
221 	if (status != CAM_REQ_CMP)
222 		goto failure;
223 	periph->path = path;
224 
225 	xpt_lock_buses();
226 	for (p_drv = periph_drivers; *p_drv != NULL; p_drv++) {
227 		if (strcmp((*p_drv)->driver_name, name) == 0)
228 			break;
229 	}
230 	if (*p_drv == NULL) {
231 		printf("cam_periph_alloc: invalid periph name '%s'\n", name);
232 		xpt_unlock_buses();
233 		xpt_free_path(periph->path);
234 		free(periph, M_CAMPERIPH);
235 		return (CAM_REQ_INVALID);
236 	}
237 	periph->unit_number = camperiphunit(*p_drv, path_id, target_id, lun_id);
238 	cur_periph = TAILQ_FIRST(&(*p_drv)->units);
239 	while (cur_periph != NULL
240 	    && cur_periph->unit_number < periph->unit_number)
241 		cur_periph = TAILQ_NEXT(cur_periph, unit_links);
242 	if (cur_periph != NULL) {
243 		KASSERT(cur_periph->unit_number != periph->unit_number, ("duplicate units on periph list"));
244 		TAILQ_INSERT_BEFORE(cur_periph, periph, unit_links);
245 	} else {
246 		TAILQ_INSERT_TAIL(&(*p_drv)->units, periph, unit_links);
247 		(*p_drv)->generation++;
248 	}
249 	xpt_unlock_buses();
250 
251 	init_level++;
252 
253 	status = xpt_add_periph(periph);
254 	if (status != CAM_REQ_CMP)
255 		goto failure;
256 
257 	init_level++;
258 	CAM_DEBUG(periph->path, CAM_DEBUG_INFO, ("Periph created\n"));
259 
260 	status = periph_ctor(periph, arg);
261 
262 	if (status == CAM_REQ_CMP)
263 		init_level++;
264 
265 failure:
266 	switch (init_level) {
267 	case 4:
268 		/* Initialized successfully */
269 		break;
270 	case 3:
271 		CAM_DEBUG(periph->path, CAM_DEBUG_INFO, ("Periph destroyed\n"));
272 		xpt_remove_periph(periph);
273 		/* FALLTHROUGH */
274 	case 2:
275 		xpt_lock_buses();
276 		TAILQ_REMOVE(&(*p_drv)->units, periph, unit_links);
277 		xpt_unlock_buses();
278 		xpt_free_path(periph->path);
279 		/* FALLTHROUGH */
280 	case 1:
281 		free(periph, M_CAMPERIPH);
282 		/* FALLTHROUGH */
283 	case 0:
284 		/* No cleanup to perform. */
285 		break;
286 	default:
287 		panic("%s: Unknown init level", __func__);
288 	}
289 	return(status);
290 }
291 
292 /*
293  * Find a peripheral structure with the specified path, target, lun,
294  * and (optionally) type.  If the name is NULL, this function will return
295  * the first peripheral driver that matches the specified path.
296  */
297 struct cam_periph *
298 cam_periph_find(struct cam_path *path, char *name)
299 {
300 	struct periph_driver **p_drv;
301 	struct cam_periph *periph;
302 
303 	xpt_lock_buses();
304 	for (p_drv = periph_drivers; *p_drv != NULL; p_drv++) {
305 
306 		if (name != NULL && (strcmp((*p_drv)->driver_name, name) != 0))
307 			continue;
308 
309 		TAILQ_FOREACH(periph, &(*p_drv)->units, unit_links) {
310 			if (xpt_path_comp(periph->path, path) == 0) {
311 				xpt_unlock_buses();
312 				cam_periph_assert(periph, MA_OWNED);
313 				return(periph);
314 			}
315 		}
316 		if (name != NULL) {
317 			xpt_unlock_buses();
318 			return(NULL);
319 		}
320 	}
321 	xpt_unlock_buses();
322 	return(NULL);
323 }
324 
325 /*
326  * Find peripheral driver instances attached to the specified path.
327  */
328 int
329 cam_periph_list(struct cam_path *path, struct sbuf *sb)
330 {
331 	struct sbuf local_sb;
332 	struct periph_driver **p_drv;
333 	struct cam_periph *periph;
334 	int count;
335 	int sbuf_alloc_len;
336 
337 	sbuf_alloc_len = 16;
338 retry:
339 	sbuf_new(&local_sb, NULL, sbuf_alloc_len, SBUF_FIXEDLEN);
340 	count = 0;
341 	xpt_lock_buses();
342 	for (p_drv = periph_drivers; *p_drv != NULL; p_drv++) {
343 
344 		TAILQ_FOREACH(periph, &(*p_drv)->units, unit_links) {
345 			if (xpt_path_comp(periph->path, path) != 0)
346 				continue;
347 
348 			if (sbuf_len(&local_sb) != 0)
349 				sbuf_cat(&local_sb, ",");
350 
351 			sbuf_printf(&local_sb, "%s%d", periph->periph_name,
352 				    periph->unit_number);
353 
354 			if (sbuf_error(&local_sb) == ENOMEM) {
355 				sbuf_alloc_len *= 2;
356 				xpt_unlock_buses();
357 				sbuf_delete(&local_sb);
358 				goto retry;
359 			}
360 			count++;
361 		}
362 	}
363 	xpt_unlock_buses();
364 	sbuf_finish(&local_sb);
365 	sbuf_cpy(sb, sbuf_data(&local_sb));
366 	sbuf_delete(&local_sb);
367 	return (count);
368 }
369 
370 cam_status
371 cam_periph_acquire(struct cam_periph *periph)
372 {
373 	cam_status status;
374 
375 	status = CAM_REQ_CMP_ERR;
376 	if (periph == NULL)
377 		return (status);
378 
379 	xpt_lock_buses();
380 	if ((periph->flags & CAM_PERIPH_INVALID) == 0) {
381 		periph->refcount++;
382 		status = CAM_REQ_CMP;
383 	}
384 	xpt_unlock_buses();
385 
386 	return (status);
387 }
388 
389 void
390 cam_periph_doacquire(struct cam_periph *periph)
391 {
392 
393 	xpt_lock_buses();
394 	KASSERT(periph->refcount >= 1,
395 	    ("cam_periph_doacquire() with refcount == %d", periph->refcount));
396 	periph->refcount++;
397 	xpt_unlock_buses();
398 }
399 
400 void
401 cam_periph_release_locked_buses(struct cam_periph *periph)
402 {
403 
404 	cam_periph_assert(periph, MA_OWNED);
405 	KASSERT(periph->refcount >= 1, ("periph->refcount >= 1"));
406 	if (--periph->refcount == 0)
407 		camperiphfree(periph);
408 }
409 
410 void
411 cam_periph_release_locked(struct cam_periph *periph)
412 {
413 
414 	if (periph == NULL)
415 		return;
416 
417 	xpt_lock_buses();
418 	cam_periph_release_locked_buses(periph);
419 	xpt_unlock_buses();
420 }
421 
422 void
423 cam_periph_release(struct cam_periph *periph)
424 {
425 	struct mtx *mtx;
426 
427 	if (periph == NULL)
428 		return;
429 
430 	cam_periph_assert(periph, MA_NOTOWNED);
431 	mtx = cam_periph_mtx(periph);
432 	mtx_lock(mtx);
433 	cam_periph_release_locked(periph);
434 	mtx_unlock(mtx);
435 }
436 
437 int
438 cam_periph_hold(struct cam_periph *periph, int priority)
439 {
440 	int error;
441 
442 	/*
443 	 * Increment the reference count on the peripheral
444 	 * while we wait for our lock attempt to succeed
445 	 * to ensure the peripheral doesn't disappear out
446 	 * from user us while we sleep.
447 	 */
448 
449 	if (cam_periph_acquire(periph) != CAM_REQ_CMP)
450 		return (ENXIO);
451 
452 	cam_periph_assert(periph, MA_OWNED);
453 	while ((periph->flags & CAM_PERIPH_LOCKED) != 0) {
454 		periph->flags |= CAM_PERIPH_LOCK_WANTED;
455 		if ((error = cam_periph_sleep(periph, periph, priority,
456 		    "caplck", 0)) != 0) {
457 			cam_periph_release_locked(periph);
458 			return (error);
459 		}
460 		if (periph->flags & CAM_PERIPH_INVALID) {
461 			cam_periph_release_locked(periph);
462 			return (ENXIO);
463 		}
464 	}
465 
466 	periph->flags |= CAM_PERIPH_LOCKED;
467 	return (0);
468 }
469 
470 void
471 cam_periph_unhold(struct cam_periph *periph)
472 {
473 
474 	cam_periph_assert(periph, MA_OWNED);
475 
476 	periph->flags &= ~CAM_PERIPH_LOCKED;
477 	if ((periph->flags & CAM_PERIPH_LOCK_WANTED) != 0) {
478 		periph->flags &= ~CAM_PERIPH_LOCK_WANTED;
479 		wakeup(periph);
480 	}
481 
482 	cam_periph_release_locked(periph);
483 }
484 
485 /*
486  * Look for the next unit number that is not currently in use for this
487  * peripheral type starting at "newunit".  Also exclude unit numbers that
488  * are reserved by for future "hardwiring" unless we already know that this
489  * is a potential wired device.  Only assume that the device is "wired" the
490  * first time through the loop since after that we'll be looking at unit
491  * numbers that did not match a wiring entry.
492  */
493 static u_int
494 camperiphnextunit(struct periph_driver *p_drv, u_int newunit, int wired,
495 		  path_id_t pathid, target_id_t target, lun_id_t lun)
496 {
497 	struct	cam_periph *periph;
498 	char	*periph_name;
499 	int	i, val, dunit, r;
500 	const char *dname, *strval;
501 
502 	periph_name = p_drv->driver_name;
503 	for (;;newunit++) {
504 
505 		for (periph = TAILQ_FIRST(&p_drv->units);
506 		     periph != NULL && periph->unit_number != newunit;
507 		     periph = TAILQ_NEXT(periph, unit_links))
508 			;
509 
510 		if (periph != NULL && periph->unit_number == newunit) {
511 			if (wired != 0) {
512 				xpt_print(periph->path, "Duplicate Wired "
513 				    "Device entry!\n");
514 				xpt_print(periph->path, "Second device (%s "
515 				    "device at scbus%d target %d lun %d) will "
516 				    "not be wired\n", periph_name, pathid,
517 				    target, lun);
518 				wired = 0;
519 			}
520 			continue;
521 		}
522 		if (wired)
523 			break;
524 
525 		/*
526 		 * Don't match entries like "da 4" as a wired down
527 		 * device, but do match entries like "da 4 target 5"
528 		 * or even "da 4 scbus 1".
529 		 */
530 		i = 0;
531 		dname = periph_name;
532 		for (;;) {
533 			r = resource_find_dev(&i, dname, &dunit, NULL, NULL);
534 			if (r != 0)
535 				break;
536 			/* if no "target" and no specific scbus, skip */
537 			if (resource_int_value(dname, dunit, "target", &val) &&
538 			    (resource_string_value(dname, dunit, "at",&strval)||
539 			     strcmp(strval, "scbus") == 0))
540 				continue;
541 			if (newunit == dunit)
542 				break;
543 		}
544 		if (r != 0)
545 			break;
546 	}
547 	return (newunit);
548 }
549 
550 static u_int
551 camperiphunit(struct periph_driver *p_drv, path_id_t pathid,
552 	      target_id_t target, lun_id_t lun)
553 {
554 	u_int	unit;
555 	int	wired, i, val, dunit;
556 	const char *dname, *strval;
557 	char	pathbuf[32], *periph_name;
558 
559 	periph_name = p_drv->driver_name;
560 	snprintf(pathbuf, sizeof(pathbuf), "scbus%d", pathid);
561 	unit = 0;
562 	i = 0;
563 	dname = periph_name;
564 	for (wired = 0; resource_find_dev(&i, dname, &dunit, NULL, NULL) == 0;
565 	     wired = 0) {
566 		if (resource_string_value(dname, dunit, "at", &strval) == 0) {
567 			if (strcmp(strval, pathbuf) != 0)
568 				continue;
569 			wired++;
570 		}
571 		if (resource_int_value(dname, dunit, "target", &val) == 0) {
572 			if (val != target)
573 				continue;
574 			wired++;
575 		}
576 		if (resource_int_value(dname, dunit, "lun", &val) == 0) {
577 			if (val != lun)
578 				continue;
579 			wired++;
580 		}
581 		if (wired != 0) {
582 			unit = dunit;
583 			break;
584 		}
585 	}
586 
587 	/*
588 	 * Either start from 0 looking for the next unit or from
589 	 * the unit number given in the resource config.  This way,
590 	 * if we have wildcard matches, we don't return the same
591 	 * unit number twice.
592 	 */
593 	unit = camperiphnextunit(p_drv, unit, wired, pathid, target, lun);
594 
595 	return (unit);
596 }
597 
598 void
599 cam_periph_invalidate(struct cam_periph *periph)
600 {
601 
602 	cam_periph_assert(periph, MA_OWNED);
603 	/*
604 	 * We only call this routine the first time a peripheral is
605 	 * invalidated.
606 	 */
607 	if ((periph->flags & CAM_PERIPH_INVALID) != 0)
608 		return;
609 
610 	CAM_DEBUG(periph->path, CAM_DEBUG_INFO, ("Periph invalidated\n"));
611 	if ((periph->flags & CAM_PERIPH_ANNOUNCED) && !rebooting)
612 		xpt_denounce_periph(periph);
613 	periph->flags |= CAM_PERIPH_INVALID;
614 	periph->flags &= ~CAM_PERIPH_NEW_DEV_FOUND;
615 	if (periph->periph_oninval != NULL)
616 		periph->periph_oninval(periph);
617 	cam_periph_release_locked(periph);
618 }
619 
620 static void
621 camperiphfree(struct cam_periph *periph)
622 {
623 	struct periph_driver **p_drv;
624 
625 	cam_periph_assert(periph, MA_OWNED);
626 	KASSERT(periph->periph_allocating == 0, ("%s%d: freed while allocating",
627 	    periph->periph_name, periph->unit_number));
628 	for (p_drv = periph_drivers; *p_drv != NULL; p_drv++) {
629 		if (strcmp((*p_drv)->driver_name, periph->periph_name) == 0)
630 			break;
631 	}
632 	if (*p_drv == NULL) {
633 		printf("camperiphfree: attempt to free non-existant periph\n");
634 		return;
635 	}
636 
637 	/*
638 	 * We need to set this flag before dropping the topology lock, to
639 	 * let anyone who is traversing the list that this peripheral is
640 	 * about to be freed, and there will be no more reference count
641 	 * checks.
642 	 */
643 	periph->flags |= CAM_PERIPH_FREE;
644 
645 	/*
646 	 * The peripheral destructor semantics dictate calling with only the
647 	 * SIM mutex held.  Since it might sleep, it should not be called
648 	 * with the topology lock held.
649 	 */
650 	xpt_unlock_buses();
651 
652 	/*
653 	 * We need to call the peripheral destructor prior to removing the
654 	 * peripheral from the list.  Otherwise, we risk running into a
655 	 * scenario where the peripheral unit number may get reused
656 	 * (because it has been removed from the list), but some resources
657 	 * used by the peripheral are still hanging around.  In particular,
658 	 * the devfs nodes used by some peripherals like the pass(4) driver
659 	 * aren't fully cleaned up until the destructor is run.  If the
660 	 * unit number is reused before the devfs instance is fully gone,
661 	 * devfs will panic.
662 	 */
663 	if (periph->periph_dtor != NULL)
664 		periph->periph_dtor(periph);
665 
666 	/*
667 	 * The peripheral list is protected by the topology lock.
668 	 */
669 	xpt_lock_buses();
670 
671 	TAILQ_REMOVE(&(*p_drv)->units, periph, unit_links);
672 	(*p_drv)->generation++;
673 
674 	xpt_remove_periph(periph);
675 
676 	xpt_unlock_buses();
677 	if ((periph->flags & CAM_PERIPH_ANNOUNCED) && !rebooting)
678 		xpt_print(periph->path, "Periph destroyed\n");
679 	else
680 		CAM_DEBUG(periph->path, CAM_DEBUG_INFO, ("Periph destroyed\n"));
681 
682 	if (periph->flags & CAM_PERIPH_NEW_DEV_FOUND) {
683 		union ccb ccb;
684 		void *arg;
685 
686 		switch (periph->deferred_ac) {
687 		case AC_FOUND_DEVICE:
688 			ccb.ccb_h.func_code = XPT_GDEV_TYPE;
689 			xpt_setup_ccb(&ccb.ccb_h, periph->path, CAM_PRIORITY_NORMAL);
690 			xpt_action(&ccb);
691 			arg = &ccb;
692 			break;
693 		case AC_PATH_REGISTERED:
694 			ccb.ccb_h.func_code = XPT_PATH_INQ;
695 			xpt_setup_ccb(&ccb.ccb_h, periph->path, CAM_PRIORITY_NORMAL);
696 			xpt_action(&ccb);
697 			arg = &ccb;
698 			break;
699 		default:
700 			arg = NULL;
701 			break;
702 		}
703 		periph->deferred_callback(NULL, periph->deferred_ac,
704 					  periph->path, arg);
705 	}
706 	xpt_free_path(periph->path);
707 	free(periph, M_CAMPERIPH);
708 	xpt_lock_buses();
709 }
710 
711 /*
712  * Map user virtual pointers into kernel virtual address space, so we can
713  * access the memory.  This is now a generic function that centralizes most
714  * of the sanity checks on the data flags, if any.
715  * This also only works for up to MAXPHYS memory.  Since we use
716  * buffers to map stuff in and out, we're limited to the buffer size.
717  */
718 int
719 cam_periph_mapmem(union ccb *ccb, struct cam_periph_map_info *mapinfo)
720 {
721 	int numbufs, i, j;
722 	int flags[CAM_PERIPH_MAXMAPS];
723 	u_int8_t **data_ptrs[CAM_PERIPH_MAXMAPS];
724 	u_int32_t lengths[CAM_PERIPH_MAXMAPS];
725 	u_int32_t dirs[CAM_PERIPH_MAXMAPS];
726 	/* Some controllers may not be able to handle more data. */
727 	size_t maxmap = DFLTPHYS;
728 
729 	switch(ccb->ccb_h.func_code) {
730 	case XPT_DEV_MATCH:
731 		if (ccb->cdm.match_buf_len == 0) {
732 			printf("cam_periph_mapmem: invalid match buffer "
733 			       "length 0\n");
734 			return(EINVAL);
735 		}
736 		if (ccb->cdm.pattern_buf_len > 0) {
737 			data_ptrs[0] = (u_int8_t **)&ccb->cdm.patterns;
738 			lengths[0] = ccb->cdm.pattern_buf_len;
739 			dirs[0] = CAM_DIR_OUT;
740 			data_ptrs[1] = (u_int8_t **)&ccb->cdm.matches;
741 			lengths[1] = ccb->cdm.match_buf_len;
742 			dirs[1] = CAM_DIR_IN;
743 			numbufs = 2;
744 		} else {
745 			data_ptrs[0] = (u_int8_t **)&ccb->cdm.matches;
746 			lengths[0] = ccb->cdm.match_buf_len;
747 			dirs[0] = CAM_DIR_IN;
748 			numbufs = 1;
749 		}
750 		/*
751 		 * This request will not go to the hardware, no reason
752 		 * to be so strict. vmapbuf() is able to map up to MAXPHYS.
753 		 */
754 		maxmap = MAXPHYS;
755 		break;
756 	case XPT_SCSI_IO:
757 	case XPT_CONT_TARGET_IO:
758 		if ((ccb->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_NONE)
759 			return(0);
760 		if ((ccb->ccb_h.flags & CAM_DATA_MASK) != CAM_DATA_VADDR)
761 			return (EINVAL);
762 		data_ptrs[0] = &ccb->csio.data_ptr;
763 		lengths[0] = ccb->csio.dxfer_len;
764 		dirs[0] = ccb->ccb_h.flags & CAM_DIR_MASK;
765 		numbufs = 1;
766 		break;
767 	case XPT_ATA_IO:
768 		if ((ccb->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_NONE)
769 			return(0);
770 		if ((ccb->ccb_h.flags & CAM_DATA_MASK) != CAM_DATA_VADDR)
771 			return (EINVAL);
772 		data_ptrs[0] = &ccb->ataio.data_ptr;
773 		lengths[0] = ccb->ataio.dxfer_len;
774 		dirs[0] = ccb->ccb_h.flags & CAM_DIR_MASK;
775 		numbufs = 1;
776 		break;
777 	case XPT_SMP_IO:
778 		data_ptrs[0] = &ccb->smpio.smp_request;
779 		lengths[0] = ccb->smpio.smp_request_len;
780 		dirs[0] = CAM_DIR_OUT;
781 		data_ptrs[1] = &ccb->smpio.smp_response;
782 		lengths[1] = ccb->smpio.smp_response_len;
783 		dirs[1] = CAM_DIR_IN;
784 		numbufs = 2;
785 		break;
786 	case XPT_DEV_ADVINFO:
787 		if (ccb->cdai.bufsiz == 0)
788 			return (0);
789 
790 		data_ptrs[0] = (uint8_t **)&ccb->cdai.buf;
791 		lengths[0] = ccb->cdai.bufsiz;
792 		dirs[0] = CAM_DIR_IN;
793 		numbufs = 1;
794 
795 		/*
796 		 * This request will not go to the hardware, no reason
797 		 * to be so strict. vmapbuf() is able to map up to MAXPHYS.
798 		 */
799 		maxmap = MAXPHYS;
800 		break;
801 	default:
802 		return(EINVAL);
803 		break; /* NOTREACHED */
804 	}
805 
806 	/*
807 	 * Check the transfer length and permissions first, so we don't
808 	 * have to unmap any previously mapped buffers.
809 	 */
810 	for (i = 0; i < numbufs; i++) {
811 
812 		flags[i] = 0;
813 
814 		/*
815 		 * The userland data pointer passed in may not be page
816 		 * aligned.  vmapbuf() truncates the address to a page
817 		 * boundary, so if the address isn't page aligned, we'll
818 		 * need enough space for the given transfer length, plus
819 		 * whatever extra space is necessary to make it to the page
820 		 * boundary.
821 		 */
822 		if ((lengths[i] +
823 		    (((vm_offset_t)(*data_ptrs[i])) & PAGE_MASK)) > maxmap){
824 			printf("cam_periph_mapmem: attempt to map %lu bytes, "
825 			       "which is greater than %lu\n",
826 			       (long)(lengths[i] +
827 			       (((vm_offset_t)(*data_ptrs[i])) & PAGE_MASK)),
828 			       (u_long)maxmap);
829 			return(E2BIG);
830 		}
831 
832 		if (dirs[i] & CAM_DIR_OUT) {
833 			flags[i] = BIO_WRITE;
834 		}
835 
836 		if (dirs[i] & CAM_DIR_IN) {
837 			flags[i] = BIO_READ;
838 		}
839 
840 	}
841 
842 	/*
843 	 * This keeps the the kernel stack of current thread from getting
844 	 * swapped.  In low-memory situations where the kernel stack might
845 	 * otherwise get swapped out, this holds it and allows the thread
846 	 * to make progress and release the kernel mapped pages sooner.
847 	 *
848 	 * XXX KDM should I use P_NOSWAP instead?
849 	 */
850 	PHOLD(curproc);
851 
852 	for (i = 0; i < numbufs; i++) {
853 		/*
854 		 * Get the buffer.
855 		 */
856 		mapinfo->bp[i] = getpbuf(NULL);
857 
858 		/* put our pointer in the data slot */
859 		mapinfo->bp[i]->b_data = *data_ptrs[i];
860 
861 		/* save the user's data address */
862 		mapinfo->bp[i]->b_caller1 = *data_ptrs[i];
863 
864 		/* set the transfer length, we know it's < MAXPHYS */
865 		mapinfo->bp[i]->b_bufsize = lengths[i];
866 
867 		/* set the direction */
868 		mapinfo->bp[i]->b_iocmd = flags[i];
869 
870 		/*
871 		 * Map the buffer into kernel memory.
872 		 *
873 		 * Note that useracc() alone is not a  sufficient test.
874 		 * vmapbuf() can still fail due to a smaller file mapped
875 		 * into a larger area of VM, or if userland races against
876 		 * vmapbuf() after the useracc() check.
877 		 */
878 		if (vmapbuf(mapinfo->bp[i], 1) < 0) {
879 			for (j = 0; j < i; ++j) {
880 				*data_ptrs[j] = mapinfo->bp[j]->b_caller1;
881 				vunmapbuf(mapinfo->bp[j]);
882 				relpbuf(mapinfo->bp[j], NULL);
883 			}
884 			relpbuf(mapinfo->bp[i], NULL);
885 			PRELE(curproc);
886 			return(EACCES);
887 		}
888 
889 		/* set our pointer to the new mapped area */
890 		*data_ptrs[i] = mapinfo->bp[i]->b_data;
891 
892 		mapinfo->num_bufs_used++;
893 	}
894 
895 	/*
896 	 * Now that we've gotten this far, change ownership to the kernel
897 	 * of the buffers so that we don't run afoul of returning to user
898 	 * space with locks (on the buffer) held.
899 	 */
900 	for (i = 0; i < numbufs; i++) {
901 		BUF_KERNPROC(mapinfo->bp[i]);
902 	}
903 
904 
905 	return(0);
906 }
907 
908 /*
909  * Unmap memory segments mapped into kernel virtual address space by
910  * cam_periph_mapmem().
911  */
912 void
913 cam_periph_unmapmem(union ccb *ccb, struct cam_periph_map_info *mapinfo)
914 {
915 	int numbufs, i;
916 	u_int8_t **data_ptrs[CAM_PERIPH_MAXMAPS];
917 
918 	if (mapinfo->num_bufs_used <= 0) {
919 		/* nothing to free and the process wasn't held. */
920 		return;
921 	}
922 
923 	switch (ccb->ccb_h.func_code) {
924 	case XPT_DEV_MATCH:
925 		numbufs = min(mapinfo->num_bufs_used, 2);
926 
927 		if (numbufs == 1) {
928 			data_ptrs[0] = (u_int8_t **)&ccb->cdm.matches;
929 		} else {
930 			data_ptrs[0] = (u_int8_t **)&ccb->cdm.patterns;
931 			data_ptrs[1] = (u_int8_t **)&ccb->cdm.matches;
932 		}
933 		break;
934 	case XPT_SCSI_IO:
935 	case XPT_CONT_TARGET_IO:
936 		data_ptrs[0] = &ccb->csio.data_ptr;
937 		numbufs = min(mapinfo->num_bufs_used, 1);
938 		break;
939 	case XPT_ATA_IO:
940 		data_ptrs[0] = &ccb->ataio.data_ptr;
941 		numbufs = min(mapinfo->num_bufs_used, 1);
942 		break;
943 	case XPT_SMP_IO:
944 		numbufs = min(mapinfo->num_bufs_used, 2);
945 		data_ptrs[0] = &ccb->smpio.smp_request;
946 		data_ptrs[1] = &ccb->smpio.smp_response;
947 		break;
948 	case XPT_DEV_ADVINFO:
949 		numbufs = min(mapinfo->num_bufs_used, 1);
950 		data_ptrs[0] = (uint8_t **)&ccb->cdai.buf;
951 		break;
952 	default:
953 		/* allow ourselves to be swapped once again */
954 		PRELE(curproc);
955 		return;
956 		break; /* NOTREACHED */
957 	}
958 
959 	for (i = 0; i < numbufs; i++) {
960 		/* Set the user's pointer back to the original value */
961 		*data_ptrs[i] = mapinfo->bp[i]->b_caller1;
962 
963 		/* unmap the buffer */
964 		vunmapbuf(mapinfo->bp[i]);
965 
966 		/* release the buffer */
967 		relpbuf(mapinfo->bp[i], NULL);
968 	}
969 
970 	/* allow ourselves to be swapped once again */
971 	PRELE(curproc);
972 }
973 
974 void
975 cam_periph_ccbwait(union ccb *ccb)
976 {
977 
978 	if ((ccb->ccb_h.pinfo.index != CAM_UNQUEUED_INDEX)
979 	 || ((ccb->ccb_h.status & CAM_STATUS_MASK) == CAM_REQ_INPROG))
980 		xpt_path_sleep(ccb->ccb_h.path, &ccb->ccb_h.cbfcnp, PRIBIO,
981 		    "cbwait", 0);
982 }
983 
984 int
985 cam_periph_ioctl(struct cam_periph *periph, u_long cmd, caddr_t addr,
986 		 int (*error_routine)(union ccb *ccb,
987 				      cam_flags camflags,
988 				      u_int32_t sense_flags))
989 {
990 	union ccb 	     *ccb;
991 	int 		     error;
992 	int		     found;
993 
994 	error = found = 0;
995 
996 	switch(cmd){
997 	case CAMGETPASSTHRU:
998 		ccb = cam_periph_getccb(periph, CAM_PRIORITY_NORMAL);
999 		xpt_setup_ccb(&ccb->ccb_h,
1000 			      ccb->ccb_h.path,
1001 			      CAM_PRIORITY_NORMAL);
1002 		ccb->ccb_h.func_code = XPT_GDEVLIST;
1003 
1004 		/*
1005 		 * Basically, the point of this is that we go through
1006 		 * getting the list of devices, until we find a passthrough
1007 		 * device.  In the current version of the CAM code, the
1008 		 * only way to determine what type of device we're dealing
1009 		 * with is by its name.
1010 		 */
1011 		while (found == 0) {
1012 			ccb->cgdl.index = 0;
1013 			ccb->cgdl.status = CAM_GDEVLIST_MORE_DEVS;
1014 			while (ccb->cgdl.status == CAM_GDEVLIST_MORE_DEVS) {
1015 
1016 				/* we want the next device in the list */
1017 				xpt_action(ccb);
1018 				if (strncmp(ccb->cgdl.periph_name,
1019 				    "pass", 4) == 0){
1020 					found = 1;
1021 					break;
1022 				}
1023 			}
1024 			if ((ccb->cgdl.status == CAM_GDEVLIST_LAST_DEVICE) &&
1025 			    (found == 0)) {
1026 				ccb->cgdl.periph_name[0] = '\0';
1027 				ccb->cgdl.unit_number = 0;
1028 				break;
1029 			}
1030 		}
1031 
1032 		/* copy the result back out */
1033 		bcopy(ccb, addr, sizeof(union ccb));
1034 
1035 		/* and release the ccb */
1036 		xpt_release_ccb(ccb);
1037 
1038 		break;
1039 	default:
1040 		error = ENOTTY;
1041 		break;
1042 	}
1043 	return(error);
1044 }
1045 
1046 static void
1047 cam_periph_done(struct cam_periph *periph, union ccb *done_ccb)
1048 {
1049 
1050 	/* Caller will release the CCB */
1051 	wakeup(&done_ccb->ccb_h.cbfcnp);
1052 }
1053 
1054 int
1055 cam_periph_runccb(union ccb *ccb,
1056 		  int (*error_routine)(union ccb *ccb,
1057 				       cam_flags camflags,
1058 				       u_int32_t sense_flags),
1059 		  cam_flags camflags, u_int32_t sense_flags,
1060 		  struct devstat *ds)
1061 {
1062 	struct bintime *starttime;
1063 	struct bintime ltime;
1064 	int error;
1065 
1066 	starttime = NULL;
1067 	xpt_path_assert(ccb->ccb_h.path, MA_OWNED);
1068 
1069 	/*
1070 	 * If the user has supplied a stats structure, and if we understand
1071 	 * this particular type of ccb, record the transaction start.
1072 	 */
1073 	if ((ds != NULL) && (ccb->ccb_h.func_code == XPT_SCSI_IO ||
1074 	    ccb->ccb_h.func_code == XPT_ATA_IO)) {
1075 		starttime = &ltime;
1076 		binuptime(starttime);
1077 		devstat_start_transaction(ds, starttime);
1078 	}
1079 
1080 	ccb->ccb_h.cbfcnp = cam_periph_done;
1081 	xpt_action(ccb);
1082 
1083 	do {
1084 		cam_periph_ccbwait(ccb);
1085 		if ((ccb->ccb_h.status & CAM_STATUS_MASK) == CAM_REQ_CMP)
1086 			error = 0;
1087 		else if (error_routine != NULL)
1088 			error = (*error_routine)(ccb, camflags, sense_flags);
1089 		else
1090 			error = 0;
1091 
1092 	} while (error == ERESTART);
1093 
1094 	if ((ccb->ccb_h.status & CAM_DEV_QFRZN) != 0) {
1095 		cam_release_devq(ccb->ccb_h.path,
1096 				 /* relsim_flags */0,
1097 				 /* openings */0,
1098 				 /* timeout */0,
1099 				 /* getcount_only */ FALSE);
1100 		ccb->ccb_h.status &= ~CAM_DEV_QFRZN;
1101 	}
1102 
1103 	if (ds != NULL) {
1104 		if (ccb->ccb_h.func_code == XPT_SCSI_IO) {
1105 			devstat_end_transaction(ds,
1106 					ccb->csio.dxfer_len - ccb->csio.resid,
1107 					ccb->csio.tag_action & 0x3,
1108 					((ccb->ccb_h.flags & CAM_DIR_MASK) ==
1109 					CAM_DIR_NONE) ?  DEVSTAT_NO_DATA :
1110 					(ccb->ccb_h.flags & CAM_DIR_OUT) ?
1111 					DEVSTAT_WRITE :
1112 					DEVSTAT_READ, NULL, starttime);
1113 		} else if (ccb->ccb_h.func_code == XPT_ATA_IO) {
1114 			devstat_end_transaction(ds,
1115 					ccb->ataio.dxfer_len - ccb->ataio.resid,
1116 					ccb->ataio.tag_action & 0x3,
1117 					((ccb->ccb_h.flags & CAM_DIR_MASK) ==
1118 					CAM_DIR_NONE) ?  DEVSTAT_NO_DATA :
1119 					(ccb->ccb_h.flags & CAM_DIR_OUT) ?
1120 					DEVSTAT_WRITE :
1121 					DEVSTAT_READ, NULL, starttime);
1122 		}
1123 	}
1124 
1125 	return(error);
1126 }
1127 
1128 void
1129 cam_freeze_devq(struct cam_path *path)
1130 {
1131 	struct ccb_hdr ccb_h;
1132 
1133 	CAM_DEBUG(path, CAM_DEBUG_TRACE, ("cam_freeze_devq\n"));
1134 	xpt_setup_ccb(&ccb_h, path, /*priority*/1);
1135 	ccb_h.func_code = XPT_NOOP;
1136 	ccb_h.flags = CAM_DEV_QFREEZE;
1137 	xpt_action((union ccb *)&ccb_h);
1138 }
1139 
1140 u_int32_t
1141 cam_release_devq(struct cam_path *path, u_int32_t relsim_flags,
1142 		 u_int32_t openings, u_int32_t arg,
1143 		 int getcount_only)
1144 {
1145 	struct ccb_relsim crs;
1146 
1147 	CAM_DEBUG(path, CAM_DEBUG_TRACE, ("cam_release_devq(%u, %u, %u, %d)\n",
1148 	    relsim_flags, openings, arg, getcount_only));
1149 	xpt_setup_ccb(&crs.ccb_h, path, CAM_PRIORITY_NORMAL);
1150 	crs.ccb_h.func_code = XPT_REL_SIMQ;
1151 	crs.ccb_h.flags = getcount_only ? CAM_DEV_QFREEZE : 0;
1152 	crs.release_flags = relsim_flags;
1153 	crs.openings = openings;
1154 	crs.release_timeout = arg;
1155 	xpt_action((union ccb *)&crs);
1156 	return (crs.qfrozen_cnt);
1157 }
1158 
1159 #define saved_ccb_ptr ppriv_ptr0
1160 static void
1161 camperiphdone(struct cam_periph *periph, union ccb *done_ccb)
1162 {
1163 	union ccb      *saved_ccb;
1164 	cam_status	status;
1165 	struct scsi_start_stop_unit *scsi_cmd;
1166 	int    error_code, sense_key, asc, ascq;
1167 
1168 	scsi_cmd = (struct scsi_start_stop_unit *)
1169 	    &done_ccb->csio.cdb_io.cdb_bytes;
1170 	status = done_ccb->ccb_h.status;
1171 
1172 	if ((status & CAM_STATUS_MASK) != CAM_REQ_CMP) {
1173 		if (scsi_extract_sense_ccb(done_ccb,
1174 		    &error_code, &sense_key, &asc, &ascq)) {
1175 			/*
1176 			 * If the error is "invalid field in CDB",
1177 			 * and the load/eject flag is set, turn the
1178 			 * flag off and try again.  This is just in
1179 			 * case the drive in question barfs on the
1180 			 * load eject flag.  The CAM code should set
1181 			 * the load/eject flag by default for
1182 			 * removable media.
1183 			 */
1184 			if ((scsi_cmd->opcode == START_STOP_UNIT) &&
1185 			    ((scsi_cmd->how & SSS_LOEJ) != 0) &&
1186 			     (asc == 0x24) && (ascq == 0x00)) {
1187 				scsi_cmd->how &= ~SSS_LOEJ;
1188 				if (status & CAM_DEV_QFRZN) {
1189 					cam_release_devq(done_ccb->ccb_h.path,
1190 					    0, 0, 0, 0);
1191 					done_ccb->ccb_h.status &=
1192 					    ~CAM_DEV_QFRZN;
1193 				}
1194 				xpt_action(done_ccb);
1195 				goto out;
1196 			}
1197 		}
1198 		if (cam_periph_error(done_ccb,
1199 		    0, SF_RETRY_UA | SF_NO_PRINT, NULL) == ERESTART)
1200 			goto out;
1201 		if (done_ccb->ccb_h.status & CAM_DEV_QFRZN) {
1202 			cam_release_devq(done_ccb->ccb_h.path, 0, 0, 0, 0);
1203 			done_ccb->ccb_h.status &= ~CAM_DEV_QFRZN;
1204 		}
1205 	} else {
1206 		/*
1207 		 * If we have successfully taken a device from the not
1208 		 * ready to ready state, re-scan the device and re-get
1209 		 * the inquiry information.  Many devices (mostly disks)
1210 		 * don't properly report their inquiry information unless
1211 		 * they are spun up.
1212 		 */
1213 		if (scsi_cmd->opcode == START_STOP_UNIT)
1214 			xpt_async(AC_INQ_CHANGED, done_ccb->ccb_h.path, NULL);
1215 	}
1216 
1217 	/*
1218 	 * Perform the final retry with the original CCB so that final
1219 	 * error processing is performed by the owner of the CCB.
1220 	 */
1221 	saved_ccb = (union ccb *)done_ccb->ccb_h.saved_ccb_ptr;
1222 	bcopy(saved_ccb, done_ccb, sizeof(*done_ccb));
1223 	xpt_free_ccb(saved_ccb);
1224 	if (done_ccb->ccb_h.cbfcnp != camperiphdone)
1225 		periph->flags &= ~CAM_PERIPH_RECOVERY_INPROG;
1226 	xpt_action(done_ccb);
1227 
1228 out:
1229 	/* Drop freeze taken due to CAM_DEV_QFREEZE flag set. */
1230 	cam_release_devq(done_ccb->ccb_h.path, 0, 0, 0, 0);
1231 }
1232 
1233 /*
1234  * Generic Async Event handler.  Peripheral drivers usually
1235  * filter out the events that require personal attention,
1236  * and leave the rest to this function.
1237  */
1238 void
1239 cam_periph_async(struct cam_periph *periph, u_int32_t code,
1240 		 struct cam_path *path, void *arg)
1241 {
1242 	switch (code) {
1243 	case AC_LOST_DEVICE:
1244 		cam_periph_invalidate(periph);
1245 		break;
1246 	default:
1247 		break;
1248 	}
1249 }
1250 
1251 void
1252 cam_periph_bus_settle(struct cam_periph *periph, u_int bus_settle)
1253 {
1254 	struct ccb_getdevstats cgds;
1255 
1256 	xpt_setup_ccb(&cgds.ccb_h, periph->path, CAM_PRIORITY_NORMAL);
1257 	cgds.ccb_h.func_code = XPT_GDEV_STATS;
1258 	xpt_action((union ccb *)&cgds);
1259 	cam_periph_freeze_after_event(periph, &cgds.last_reset, bus_settle);
1260 }
1261 
1262 void
1263 cam_periph_freeze_after_event(struct cam_periph *periph,
1264 			      struct timeval* event_time, u_int duration_ms)
1265 {
1266 	struct timeval delta;
1267 	struct timeval duration_tv;
1268 
1269 	if (!timevalisset(event_time))
1270 		return;
1271 
1272 	microtime(&delta);
1273 	timevalsub(&delta, event_time);
1274 	duration_tv.tv_sec = duration_ms / 1000;
1275 	duration_tv.tv_usec = (duration_ms % 1000) * 1000;
1276 	if (timevalcmp(&delta, &duration_tv, <)) {
1277 		timevalsub(&duration_tv, &delta);
1278 
1279 		duration_ms = duration_tv.tv_sec * 1000;
1280 		duration_ms += duration_tv.tv_usec / 1000;
1281 		cam_freeze_devq(periph->path);
1282 		cam_release_devq(periph->path,
1283 				RELSIM_RELEASE_AFTER_TIMEOUT,
1284 				/*reduction*/0,
1285 				/*timeout*/duration_ms,
1286 				/*getcount_only*/0);
1287 	}
1288 
1289 }
1290 
1291 static int
1292 camperiphscsistatuserror(union ccb *ccb, union ccb **orig_ccb,
1293     cam_flags camflags, u_int32_t sense_flags,
1294     int *openings, u_int32_t *relsim_flags,
1295     u_int32_t *timeout, u_int32_t *action, const char **action_string)
1296 {
1297 	int error;
1298 
1299 	switch (ccb->csio.scsi_status) {
1300 	case SCSI_STATUS_OK:
1301 	case SCSI_STATUS_COND_MET:
1302 	case SCSI_STATUS_INTERMED:
1303 	case SCSI_STATUS_INTERMED_COND_MET:
1304 		error = 0;
1305 		break;
1306 	case SCSI_STATUS_CMD_TERMINATED:
1307 	case SCSI_STATUS_CHECK_COND:
1308 		error = camperiphscsisenseerror(ccb, orig_ccb,
1309 					        camflags,
1310 					        sense_flags,
1311 					        openings,
1312 					        relsim_flags,
1313 					        timeout,
1314 					        action,
1315 					        action_string);
1316 		break;
1317 	case SCSI_STATUS_QUEUE_FULL:
1318 	{
1319 		/* no decrement */
1320 		struct ccb_getdevstats cgds;
1321 
1322 		/*
1323 		 * First off, find out what the current
1324 		 * transaction counts are.
1325 		 */
1326 		xpt_setup_ccb(&cgds.ccb_h,
1327 			      ccb->ccb_h.path,
1328 			      CAM_PRIORITY_NORMAL);
1329 		cgds.ccb_h.func_code = XPT_GDEV_STATS;
1330 		xpt_action((union ccb *)&cgds);
1331 
1332 		/*
1333 		 * If we were the only transaction active, treat
1334 		 * the QUEUE FULL as if it were a BUSY condition.
1335 		 */
1336 		if (cgds.dev_active != 0) {
1337 			int total_openings;
1338 
1339 			/*
1340 		 	 * Reduce the number of openings to
1341 			 * be 1 less than the amount it took
1342 			 * to get a queue full bounded by the
1343 			 * minimum allowed tag count for this
1344 			 * device.
1345 		 	 */
1346 			total_openings = cgds.dev_active + cgds.dev_openings;
1347 			*openings = cgds.dev_active;
1348 			if (*openings < cgds.mintags)
1349 				*openings = cgds.mintags;
1350 			if (*openings < total_openings)
1351 				*relsim_flags = RELSIM_ADJUST_OPENINGS;
1352 			else {
1353 				/*
1354 				 * Some devices report queue full for
1355 				 * temporary resource shortages.  For
1356 				 * this reason, we allow a minimum
1357 				 * tag count to be entered via a
1358 				 * quirk entry to prevent the queue
1359 				 * count on these devices from falling
1360 				 * to a pessimisticly low value.  We
1361 				 * still wait for the next successful
1362 				 * completion, however, before queueing
1363 				 * more transactions to the device.
1364 				 */
1365 				*relsim_flags = RELSIM_RELEASE_AFTER_CMDCMPLT;
1366 			}
1367 			*timeout = 0;
1368 			error = ERESTART;
1369 			*action &= ~SSQ_PRINT_SENSE;
1370 			break;
1371 		}
1372 		/* FALLTHROUGH */
1373 	}
1374 	case SCSI_STATUS_BUSY:
1375 		/*
1376 		 * Restart the queue after either another
1377 		 * command completes or a 1 second timeout.
1378 		 */
1379 		if ((sense_flags & SF_RETRY_BUSY) != 0 ||
1380 		    (ccb->ccb_h.retry_count--) > 0) {
1381 			error = ERESTART;
1382 			*relsim_flags = RELSIM_RELEASE_AFTER_TIMEOUT
1383 				      | RELSIM_RELEASE_AFTER_CMDCMPLT;
1384 			*timeout = 1000;
1385 		} else {
1386 			error = EIO;
1387 		}
1388 		break;
1389 	case SCSI_STATUS_RESERV_CONFLICT:
1390 	default:
1391 		error = EIO;
1392 		break;
1393 	}
1394 	return (error);
1395 }
1396 
1397 static int
1398 camperiphscsisenseerror(union ccb *ccb, union ccb **orig,
1399     cam_flags camflags, u_int32_t sense_flags,
1400     int *openings, u_int32_t *relsim_flags,
1401     u_int32_t *timeout, u_int32_t *action, const char **action_string)
1402 {
1403 	struct cam_periph *periph;
1404 	union ccb *orig_ccb = ccb;
1405 	int error, recoveryccb;
1406 
1407 	periph = xpt_path_periph(ccb->ccb_h.path);
1408 	recoveryccb = (ccb->ccb_h.cbfcnp == camperiphdone);
1409 	if ((periph->flags & CAM_PERIPH_RECOVERY_INPROG) && !recoveryccb) {
1410 		/*
1411 		 * If error recovery is already in progress, don't attempt
1412 		 * to process this error, but requeue it unconditionally
1413 		 * and attempt to process it once error recovery has
1414 		 * completed.  This failed command is probably related to
1415 		 * the error that caused the currently active error recovery
1416 		 * action so our  current recovery efforts should also
1417 		 * address this command.  Be aware that the error recovery
1418 		 * code assumes that only one recovery action is in progress
1419 		 * on a particular peripheral instance at any given time
1420 		 * (e.g. only one saved CCB for error recovery) so it is
1421 		 * imperitive that we don't violate this assumption.
1422 		 */
1423 		error = ERESTART;
1424 		*action &= ~SSQ_PRINT_SENSE;
1425 	} else {
1426 		scsi_sense_action err_action;
1427 		struct ccb_getdev cgd;
1428 
1429 		/*
1430 		 * Grab the inquiry data for this device.
1431 		 */
1432 		xpt_setup_ccb(&cgd.ccb_h, ccb->ccb_h.path, CAM_PRIORITY_NORMAL);
1433 		cgd.ccb_h.func_code = XPT_GDEV_TYPE;
1434 		xpt_action((union ccb *)&cgd);
1435 
1436 		err_action = scsi_error_action(&ccb->csio, &cgd.inq_data,
1437 		    sense_flags);
1438 		error = err_action & SS_ERRMASK;
1439 
1440 		/*
1441 		 * Do not autostart sequential access devices
1442 		 * to avoid unexpected tape loading.
1443 		 */
1444 		if ((err_action & SS_MASK) == SS_START &&
1445 		    SID_TYPE(&cgd.inq_data) == T_SEQUENTIAL) {
1446 			*action_string = "Will not autostart a "
1447 			    "sequential access device";
1448 			goto sense_error_done;
1449 		}
1450 
1451 		/*
1452 		 * Avoid recovery recursion if recovery action is the same.
1453 		 */
1454 		if ((err_action & SS_MASK) >= SS_START && recoveryccb) {
1455 			if (((err_action & SS_MASK) == SS_START &&
1456 			     ccb->csio.cdb_io.cdb_bytes[0] == START_STOP_UNIT) ||
1457 			    ((err_action & SS_MASK) == SS_TUR &&
1458 			     (ccb->csio.cdb_io.cdb_bytes[0] == TEST_UNIT_READY))) {
1459 				err_action = SS_RETRY|SSQ_DECREMENT_COUNT|EIO;
1460 				*relsim_flags = RELSIM_RELEASE_AFTER_TIMEOUT;
1461 				*timeout = 500;
1462 			}
1463 		}
1464 
1465 		/*
1466 		 * If the recovery action will consume a retry,
1467 		 * make sure we actually have retries available.
1468 		 */
1469 		if ((err_action & SSQ_DECREMENT_COUNT) != 0) {
1470 		 	if (ccb->ccb_h.retry_count > 0 &&
1471 			    (periph->flags & CAM_PERIPH_INVALID) == 0)
1472 		 		ccb->ccb_h.retry_count--;
1473 			else {
1474 				*action_string = "Retries exhausted";
1475 				goto sense_error_done;
1476 			}
1477 		}
1478 
1479 		if ((err_action & SS_MASK) >= SS_START) {
1480 			/*
1481 			 * Do common portions of commands that
1482 			 * use recovery CCBs.
1483 			 */
1484 			orig_ccb = xpt_alloc_ccb_nowait();
1485 			if (orig_ccb == NULL) {
1486 				*action_string = "Can't allocate recovery CCB";
1487 				goto sense_error_done;
1488 			}
1489 			/*
1490 			 * Clear freeze flag for original request here, as
1491 			 * this freeze will be dropped as part of ERESTART.
1492 			 */
1493 			ccb->ccb_h.status &= ~CAM_DEV_QFRZN;
1494 			bcopy(ccb, orig_ccb, sizeof(*orig_ccb));
1495 		}
1496 
1497 		switch (err_action & SS_MASK) {
1498 		case SS_NOP:
1499 			*action_string = "No recovery action needed";
1500 			error = 0;
1501 			break;
1502 		case SS_RETRY:
1503 			*action_string = "Retrying command (per sense data)";
1504 			error = ERESTART;
1505 			break;
1506 		case SS_FAIL:
1507 			*action_string = "Unretryable error";
1508 			break;
1509 		case SS_START:
1510 		{
1511 			int le;
1512 
1513 			/*
1514 			 * Send a start unit command to the device, and
1515 			 * then retry the command.
1516 			 */
1517 			*action_string = "Attempting to start unit";
1518 			periph->flags |= CAM_PERIPH_RECOVERY_INPROG;
1519 
1520 			/*
1521 			 * Check for removable media and set
1522 			 * load/eject flag appropriately.
1523 			 */
1524 			if (SID_IS_REMOVABLE(&cgd.inq_data))
1525 				le = TRUE;
1526 			else
1527 				le = FALSE;
1528 
1529 			scsi_start_stop(&ccb->csio,
1530 					/*retries*/1,
1531 					camperiphdone,
1532 					MSG_SIMPLE_Q_TAG,
1533 					/*start*/TRUE,
1534 					/*load/eject*/le,
1535 					/*immediate*/FALSE,
1536 					SSD_FULL_SIZE,
1537 					/*timeout*/50000);
1538 			break;
1539 		}
1540 		case SS_TUR:
1541 		{
1542 			/*
1543 			 * Send a Test Unit Ready to the device.
1544 			 * If the 'many' flag is set, we send 120
1545 			 * test unit ready commands, one every half
1546 			 * second.  Otherwise, we just send one TUR.
1547 			 * We only want to do this if the retry
1548 			 * count has not been exhausted.
1549 			 */
1550 			int retries;
1551 
1552 			if ((err_action & SSQ_MANY) != 0) {
1553 				*action_string = "Polling device for readiness";
1554 				retries = 120;
1555 			} else {
1556 				*action_string = "Testing device for readiness";
1557 				retries = 1;
1558 			}
1559 			periph->flags |= CAM_PERIPH_RECOVERY_INPROG;
1560 			scsi_test_unit_ready(&ccb->csio,
1561 					     retries,
1562 					     camperiphdone,
1563 					     MSG_SIMPLE_Q_TAG,
1564 					     SSD_FULL_SIZE,
1565 					     /*timeout*/5000);
1566 
1567 			/*
1568 			 * Accomplish our 500ms delay by deferring
1569 			 * the release of our device queue appropriately.
1570 			 */
1571 			*relsim_flags = RELSIM_RELEASE_AFTER_TIMEOUT;
1572 			*timeout = 500;
1573 			break;
1574 		}
1575 		default:
1576 			panic("Unhandled error action %x", err_action);
1577 		}
1578 
1579 		if ((err_action & SS_MASK) >= SS_START) {
1580 			/*
1581 			 * Drop the priority, so that the recovery
1582 			 * CCB is the first to execute.  Freeze the queue
1583 			 * after this command is sent so that we can
1584 			 * restore the old csio and have it queued in
1585 			 * the proper order before we release normal
1586 			 * transactions to the device.
1587 			 */
1588 			ccb->ccb_h.pinfo.priority--;
1589 			ccb->ccb_h.flags |= CAM_DEV_QFREEZE;
1590 			ccb->ccb_h.saved_ccb_ptr = orig_ccb;
1591 			error = ERESTART;
1592 			*orig = orig_ccb;
1593 		}
1594 
1595 sense_error_done:
1596 		*action = err_action;
1597 	}
1598 	return (error);
1599 }
1600 
1601 /*
1602  * Generic error handler.  Peripheral drivers usually filter
1603  * out the errors that they handle in a unique mannor, then
1604  * call this function.
1605  */
1606 int
1607 cam_periph_error(union ccb *ccb, cam_flags camflags,
1608 		 u_int32_t sense_flags, union ccb *save_ccb)
1609 {
1610 	struct cam_path *newpath;
1611 	union ccb  *orig_ccb, *scan_ccb;
1612 	struct cam_periph *periph;
1613 	const char *action_string;
1614 	cam_status  status;
1615 	int	    frozen, error, openings;
1616 	u_int32_t   action, relsim_flags, timeout;
1617 
1618 	action = SSQ_PRINT_SENSE;
1619 	periph = xpt_path_periph(ccb->ccb_h.path);
1620 	action_string = NULL;
1621 	status = ccb->ccb_h.status;
1622 	frozen = (status & CAM_DEV_QFRZN) != 0;
1623 	status &= CAM_STATUS_MASK;
1624 	openings = relsim_flags = timeout = 0;
1625 	orig_ccb = ccb;
1626 
1627 	switch (status) {
1628 	case CAM_REQ_CMP:
1629 		error = 0;
1630 		action &= ~SSQ_PRINT_SENSE;
1631 		break;
1632 	case CAM_SCSI_STATUS_ERROR:
1633 		error = camperiphscsistatuserror(ccb, &orig_ccb,
1634 		    camflags, sense_flags, &openings, &relsim_flags,
1635 		    &timeout, &action, &action_string);
1636 		break;
1637 	case CAM_AUTOSENSE_FAIL:
1638 		error = EIO;	/* we have to kill the command */
1639 		break;
1640 	case CAM_UA_ABORT:
1641 	case CAM_UA_TERMIO:
1642 	case CAM_MSG_REJECT_REC:
1643 		/* XXX Don't know that these are correct */
1644 		error = EIO;
1645 		break;
1646 	case CAM_SEL_TIMEOUT:
1647 		if ((camflags & CAM_RETRY_SELTO) != 0) {
1648 			if (ccb->ccb_h.retry_count > 0 &&
1649 			    (periph->flags & CAM_PERIPH_INVALID) == 0) {
1650 				ccb->ccb_h.retry_count--;
1651 				error = ERESTART;
1652 
1653 				/*
1654 				 * Wait a bit to give the device
1655 				 * time to recover before we try again.
1656 				 */
1657 				relsim_flags = RELSIM_RELEASE_AFTER_TIMEOUT;
1658 				timeout = periph_selto_delay;
1659 				break;
1660 			}
1661 			action_string = "Retries exhausted";
1662 		}
1663 		/* FALLTHROUGH */
1664 	case CAM_DEV_NOT_THERE:
1665 		error = ENXIO;
1666 		action = SSQ_LOST;
1667 		break;
1668 	case CAM_REQ_INVALID:
1669 	case CAM_PATH_INVALID:
1670 	case CAM_NO_HBA:
1671 	case CAM_PROVIDE_FAIL:
1672 	case CAM_REQ_TOO_BIG:
1673 	case CAM_LUN_INVALID:
1674 	case CAM_TID_INVALID:
1675 	case CAM_FUNC_NOTAVAIL:
1676 		error = EINVAL;
1677 		break;
1678 	case CAM_SCSI_BUS_RESET:
1679 	case CAM_BDR_SENT:
1680 		/*
1681 		 * Commands that repeatedly timeout and cause these
1682 		 * kinds of error recovery actions, should return
1683 		 * CAM_CMD_TIMEOUT, which allows us to safely assume
1684 		 * that this command was an innocent bystander to
1685 		 * these events and should be unconditionally
1686 		 * retried.
1687 		 */
1688 	case CAM_REQUEUE_REQ:
1689 		/* Unconditional requeue if device is still there */
1690 		if (periph->flags & CAM_PERIPH_INVALID) {
1691 			action_string = "Periph was invalidated";
1692 			error = EIO;
1693 		} else if (sense_flags & SF_NO_RETRY) {
1694 			error = EIO;
1695 			action_string = "Retry was blocked";
1696 		} else {
1697 			error = ERESTART;
1698 			action &= ~SSQ_PRINT_SENSE;
1699 		}
1700 		break;
1701 	case CAM_RESRC_UNAVAIL:
1702 		/* Wait a bit for the resource shortage to abate. */
1703 		timeout = periph_noresrc_delay;
1704 		/* FALLTHROUGH */
1705 	case CAM_BUSY:
1706 		if (timeout == 0) {
1707 			/* Wait a bit for the busy condition to abate. */
1708 			timeout = periph_busy_delay;
1709 		}
1710 		relsim_flags = RELSIM_RELEASE_AFTER_TIMEOUT;
1711 		/* FALLTHROUGH */
1712 	case CAM_ATA_STATUS_ERROR:
1713 	case CAM_REQ_CMP_ERR:
1714 	case CAM_CMD_TIMEOUT:
1715 	case CAM_UNEXP_BUSFREE:
1716 	case CAM_UNCOR_PARITY:
1717 	case CAM_DATA_RUN_ERR:
1718 	default:
1719 		if (periph->flags & CAM_PERIPH_INVALID) {
1720 			error = EIO;
1721 			action_string = "Periph was invalidated";
1722 		} else if (ccb->ccb_h.retry_count == 0) {
1723 			error = EIO;
1724 			action_string = "Retries exhausted";
1725 		} else if (sense_flags & SF_NO_RETRY) {
1726 			error = EIO;
1727 			action_string = "Retry was blocked";
1728 		} else {
1729 			ccb->ccb_h.retry_count--;
1730 			error = ERESTART;
1731 		}
1732 		break;
1733 	}
1734 
1735 	if ((sense_flags & SF_PRINT_ALWAYS) ||
1736 	    CAM_DEBUGGED(ccb->ccb_h.path, CAM_DEBUG_INFO))
1737 		action |= SSQ_PRINT_SENSE;
1738 	else if (sense_flags & SF_NO_PRINT)
1739 		action &= ~SSQ_PRINT_SENSE;
1740 	if ((action & SSQ_PRINT_SENSE) != 0)
1741 		cam_error_print(orig_ccb, CAM_ESF_ALL, CAM_EPF_ALL);
1742 	if (error != 0 && (action & SSQ_PRINT_SENSE) != 0) {
1743 		if (error != ERESTART) {
1744 			if (action_string == NULL)
1745 				action_string = "Unretryable error";
1746 			xpt_print(ccb->ccb_h.path, "Error %d, %s\n",
1747 			    error, action_string);
1748 		} else if (action_string != NULL)
1749 			xpt_print(ccb->ccb_h.path, "%s\n", action_string);
1750 		else
1751 			xpt_print(ccb->ccb_h.path, "Retrying command\n");
1752 	}
1753 
1754 	if ((action & SSQ_LOST) != 0) {
1755 		lun_id_t lun_id;
1756 
1757 		/*
1758 		 * For a selection timeout, we consider all of the LUNs on
1759 		 * the target to be gone.  If the status is CAM_DEV_NOT_THERE,
1760 		 * then we only get rid of the device(s) specified by the
1761 		 * path in the original CCB.
1762 		 */
1763 		if (status == CAM_SEL_TIMEOUT)
1764 			lun_id = CAM_LUN_WILDCARD;
1765 		else
1766 			lun_id = xpt_path_lun_id(ccb->ccb_h.path);
1767 
1768 		/* Should we do more if we can't create the path?? */
1769 		if (xpt_create_path(&newpath, periph,
1770 				    xpt_path_path_id(ccb->ccb_h.path),
1771 				    xpt_path_target_id(ccb->ccb_h.path),
1772 				    lun_id) == CAM_REQ_CMP) {
1773 
1774 			/*
1775 			 * Let peripheral drivers know that this
1776 			 * device has gone away.
1777 			 */
1778 			xpt_async(AC_LOST_DEVICE, newpath, NULL);
1779 			xpt_free_path(newpath);
1780 		}
1781 	}
1782 
1783 	/* Broadcast UNIT ATTENTIONs to all periphs. */
1784 	if ((action & SSQ_UA) != 0)
1785 		xpt_async(AC_UNIT_ATTENTION, orig_ccb->ccb_h.path, orig_ccb);
1786 
1787 	/* Rescan target on "Reported LUNs data has changed" */
1788 	if ((action & SSQ_RESCAN) != 0) {
1789 		if (xpt_create_path(&newpath, NULL,
1790 				    xpt_path_path_id(ccb->ccb_h.path),
1791 				    xpt_path_target_id(ccb->ccb_h.path),
1792 				    CAM_LUN_WILDCARD) == CAM_REQ_CMP) {
1793 
1794 			scan_ccb = xpt_alloc_ccb_nowait();
1795 			if (scan_ccb != NULL) {
1796 				scan_ccb->ccb_h.path = newpath;
1797 				scan_ccb->ccb_h.func_code = XPT_SCAN_TGT;
1798 				scan_ccb->crcn.flags = 0;
1799 				xpt_rescan(scan_ccb);
1800 			} else {
1801 				xpt_print(newpath,
1802 				    "Can't allocate CCB to rescan target\n");
1803 				xpt_free_path(newpath);
1804 			}
1805 		}
1806 	}
1807 
1808 	/* Attempt a retry */
1809 	if (error == ERESTART || error == 0) {
1810 		if (frozen != 0)
1811 			ccb->ccb_h.status &= ~CAM_DEV_QFRZN;
1812 		if (error == ERESTART)
1813 			xpt_action(ccb);
1814 		if (frozen != 0)
1815 			cam_release_devq(ccb->ccb_h.path,
1816 					 relsim_flags,
1817 					 openings,
1818 					 timeout,
1819 					 /*getcount_only*/0);
1820 	}
1821 
1822 	return (error);
1823 }
1824