xref: /freebsd/sys/cam/cam_periph.c (revision c03c5b1c)
1 /*-
2  * Common functions for CAM "type" (peripheral) drivers.
3  *
4  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
5  *
6  * Copyright (c) 1997, 1998 Justin T. Gibbs.
7  * Copyright (c) 1997, 1998, 1999, 2000 Kenneth D. Merry.
8  * All rights reserved.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions, and the following disclaimer,
15  *    without modification, immediately at the beginning of the file.
16  * 2. The name of the author may not be used to endorse or promote products
17  *    derived from this software without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE FOR
23  * ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/types.h>
38 #include <sys/malloc.h>
39 #include <sys/kernel.h>
40 #include <sys/bio.h>
41 #include <sys/conf.h>
42 #include <sys/devctl.h>
43 #include <sys/lock.h>
44 #include <sys/mutex.h>
45 #include <sys/buf.h>
46 #include <sys/proc.h>
47 #include <sys/devicestat.h>
48 #include <sys/sbuf.h>
49 #include <sys/sysctl.h>
50 #include <vm/vm.h>
51 #include <vm/vm_extern.h>
52 
53 #include <cam/cam.h>
54 #include <cam/cam_ccb.h>
55 #include <cam/cam_queue.h>
56 #include <cam/cam_xpt_periph.h>
57 #include <cam/cam_xpt_internal.h>
58 #include <cam/cam_periph.h>
59 #include <cam/cam_debug.h>
60 #include <cam/cam_sim.h>
61 
62 #include <cam/scsi/scsi_all.h>
63 #include <cam/scsi/scsi_message.h>
64 #include <cam/scsi/scsi_pass.h>
65 
66 static	u_int		camperiphnextunit(struct periph_driver *p_drv,
67 					  u_int newunit, bool wired,
68 					  path_id_t pathid, target_id_t target,
69 					  lun_id_t lun);
70 static	u_int		camperiphunit(struct periph_driver *p_drv,
71 				      path_id_t pathid, target_id_t target,
72 				      lun_id_t lun,
73 				      const char *sn);
74 static	void		camperiphdone(struct cam_periph *periph,
75 					union ccb *done_ccb);
76 static  void		camperiphfree(struct cam_periph *periph);
77 static int		camperiphscsistatuserror(union ccb *ccb,
78 					        union ccb **orig_ccb,
79 						 cam_flags camflags,
80 						 u_int32_t sense_flags,
81 						 int *openings,
82 						 u_int32_t *relsim_flags,
83 						 u_int32_t *timeout,
84 						 u_int32_t  *action,
85 						 const char **action_string);
86 static	int		camperiphscsisenseerror(union ccb *ccb,
87 					        union ccb **orig_ccb,
88 					        cam_flags camflags,
89 					        u_int32_t sense_flags,
90 					        int *openings,
91 					        u_int32_t *relsim_flags,
92 					        u_int32_t *timeout,
93 					        u_int32_t *action,
94 					        const char **action_string);
95 static void		cam_periph_devctl_notify(union ccb *ccb);
96 
97 static int nperiph_drivers;
98 static int initialized = 0;
99 struct periph_driver **periph_drivers;
100 
101 static MALLOC_DEFINE(M_CAMPERIPH, "CAM periph", "CAM peripheral buffers");
102 
103 static int periph_selto_delay = 1000;
104 TUNABLE_INT("kern.cam.periph_selto_delay", &periph_selto_delay);
105 static int periph_noresrc_delay = 500;
106 TUNABLE_INT("kern.cam.periph_noresrc_delay", &periph_noresrc_delay);
107 static int periph_busy_delay = 500;
108 TUNABLE_INT("kern.cam.periph_busy_delay", &periph_busy_delay);
109 
110 static u_int periph_mapmem_thresh = 65536;
111 SYSCTL_UINT(_kern_cam, OID_AUTO, mapmem_thresh, CTLFLAG_RWTUN,
112     &periph_mapmem_thresh, 0, "Threshold for user-space buffer mapping");
113 
114 void
115 periphdriver_register(void *data)
116 {
117 	struct periph_driver *drv = (struct periph_driver *)data;
118 	struct periph_driver **newdrivers, **old;
119 	int ndrivers;
120 
121 again:
122 	ndrivers = nperiph_drivers + 2;
123 	newdrivers = malloc(sizeof(*newdrivers) * ndrivers, M_CAMPERIPH,
124 			    M_WAITOK);
125 	xpt_lock_buses();
126 	if (ndrivers != nperiph_drivers + 2) {
127 		/*
128 		 * Lost race against itself; go around.
129 		 */
130 		xpt_unlock_buses();
131 		free(newdrivers, M_CAMPERIPH);
132 		goto again;
133 	}
134 	if (periph_drivers)
135 		bcopy(periph_drivers, newdrivers,
136 		      sizeof(*newdrivers) * nperiph_drivers);
137 	newdrivers[nperiph_drivers] = drv;
138 	newdrivers[nperiph_drivers + 1] = NULL;
139 	old = periph_drivers;
140 	periph_drivers = newdrivers;
141 	nperiph_drivers++;
142 	xpt_unlock_buses();
143 	if (old)
144 		free(old, M_CAMPERIPH);
145 	/* If driver marked as early or it is late now, initialize it. */
146 	if (((drv->flags & CAM_PERIPH_DRV_EARLY) != 0 && initialized > 0) ||
147 	    initialized > 1)
148 		(*drv->init)();
149 }
150 
151 int
152 periphdriver_unregister(void *data)
153 {
154 	struct periph_driver *drv = (struct periph_driver *)data;
155 	int error, n;
156 
157 	/* If driver marked as early or it is late now, deinitialize it. */
158 	if (((drv->flags & CAM_PERIPH_DRV_EARLY) != 0 && initialized > 0) ||
159 	    initialized > 1) {
160 		if (drv->deinit == NULL) {
161 			printf("CAM periph driver '%s' doesn't have deinit.\n",
162 			    drv->driver_name);
163 			return (EOPNOTSUPP);
164 		}
165 		error = drv->deinit();
166 		if (error != 0)
167 			return (error);
168 	}
169 
170 	xpt_lock_buses();
171 	for (n = 0; n < nperiph_drivers && periph_drivers[n] != drv; n++)
172 		;
173 	KASSERT(n < nperiph_drivers,
174 	    ("Periph driver '%s' was not registered", drv->driver_name));
175 	for (; n + 1 < nperiph_drivers; n++)
176 		periph_drivers[n] = periph_drivers[n + 1];
177 	periph_drivers[n + 1] = NULL;
178 	nperiph_drivers--;
179 	xpt_unlock_buses();
180 	return (0);
181 }
182 
183 void
184 periphdriver_init(int level)
185 {
186 	int	i, early;
187 
188 	initialized = max(initialized, level);
189 	for (i = 0; periph_drivers[i] != NULL; i++) {
190 		early = (periph_drivers[i]->flags & CAM_PERIPH_DRV_EARLY) ? 1 : 2;
191 		if (early == initialized)
192 			(*periph_drivers[i]->init)();
193 	}
194 }
195 
196 cam_status
197 cam_periph_alloc(periph_ctor_t *periph_ctor,
198 		 periph_oninv_t *periph_oninvalidate,
199 		 periph_dtor_t *periph_dtor, periph_start_t *periph_start,
200 		 char *name, cam_periph_type type, struct cam_path *path,
201 		 ac_callback_t *ac_callback, ac_code code, void *arg)
202 {
203 	struct		periph_driver **p_drv;
204 	struct		cam_sim *sim;
205 	struct		cam_periph *periph;
206 	struct		cam_periph *cur_periph;
207 	path_id_t	path_id;
208 	target_id_t	target_id;
209 	lun_id_t	lun_id;
210 	cam_status	status;
211 	u_int		init_level;
212 
213 	init_level = 0;
214 	/*
215 	 * Handle Hot-Plug scenarios.  If there is already a peripheral
216 	 * of our type assigned to this path, we are likely waiting for
217 	 * final close on an old, invalidated, peripheral.  If this is
218 	 * the case, queue up a deferred call to the peripheral's async
219 	 * handler.  If it looks like a mistaken re-allocation, complain.
220 	 */
221 	if ((periph = cam_periph_find(path, name)) != NULL) {
222 		if ((periph->flags & CAM_PERIPH_INVALID) != 0
223 		 && (periph->flags & CAM_PERIPH_NEW_DEV_FOUND) == 0) {
224 			periph->flags |= CAM_PERIPH_NEW_DEV_FOUND;
225 			periph->deferred_callback = ac_callback;
226 			periph->deferred_ac = code;
227 			return (CAM_REQ_INPROG);
228 		} else {
229 			printf("cam_periph_alloc: attempt to re-allocate "
230 			       "valid device %s%d rejected flags %#x "
231 			       "refcount %d\n", periph->periph_name,
232 			       periph->unit_number, periph->flags,
233 			       periph->refcount);
234 		}
235 		return (CAM_REQ_INVALID);
236 	}
237 
238 	periph = (struct cam_periph *)malloc(sizeof(*periph), M_CAMPERIPH,
239 					     M_NOWAIT|M_ZERO);
240 
241 	if (periph == NULL)
242 		return (CAM_RESRC_UNAVAIL);
243 
244 	init_level++;
245 
246 	sim = xpt_path_sim(path);
247 	path_id = xpt_path_path_id(path);
248 	target_id = xpt_path_target_id(path);
249 	lun_id = xpt_path_lun_id(path);
250 	periph->periph_start = periph_start;
251 	periph->periph_dtor = periph_dtor;
252 	periph->periph_oninval = periph_oninvalidate;
253 	periph->type = type;
254 	periph->periph_name = name;
255 	periph->scheduled_priority = CAM_PRIORITY_NONE;
256 	periph->immediate_priority = CAM_PRIORITY_NONE;
257 	periph->refcount = 1;		/* Dropped by invalidation. */
258 	periph->sim = sim;
259 	SLIST_INIT(&periph->ccb_list);
260 	status = xpt_create_path(&path, periph, path_id, target_id, lun_id);
261 	if (status != CAM_REQ_CMP)
262 		goto failure;
263 	periph->path = path;
264 
265 	xpt_lock_buses();
266 	for (p_drv = periph_drivers; *p_drv != NULL; p_drv++) {
267 		if (strcmp((*p_drv)->driver_name, name) == 0)
268 			break;
269 	}
270 	if (*p_drv == NULL) {
271 		printf("cam_periph_alloc: invalid periph name '%s'\n", name);
272 		xpt_unlock_buses();
273 		xpt_free_path(periph->path);
274 		free(periph, M_CAMPERIPH);
275 		return (CAM_REQ_INVALID);
276 	}
277 	periph->unit_number = camperiphunit(*p_drv, path_id, target_id, lun_id,
278 	    path->device->serial_num);
279 	cur_periph = TAILQ_FIRST(&(*p_drv)->units);
280 	while (cur_periph != NULL
281 	    && cur_periph->unit_number < periph->unit_number)
282 		cur_periph = TAILQ_NEXT(cur_periph, unit_links);
283 	if (cur_periph != NULL) {
284 		KASSERT(cur_periph->unit_number != periph->unit_number,
285 		    ("duplicate units on periph list"));
286 		TAILQ_INSERT_BEFORE(cur_periph, periph, unit_links);
287 	} else {
288 		TAILQ_INSERT_TAIL(&(*p_drv)->units, periph, unit_links);
289 		(*p_drv)->generation++;
290 	}
291 	xpt_unlock_buses();
292 
293 	init_level++;
294 
295 	status = xpt_add_periph(periph);
296 	if (status != CAM_REQ_CMP)
297 		goto failure;
298 
299 	init_level++;
300 	CAM_DEBUG(periph->path, CAM_DEBUG_INFO, ("Periph created\n"));
301 
302 	status = periph_ctor(periph, arg);
303 
304 	if (status == CAM_REQ_CMP)
305 		init_level++;
306 
307 failure:
308 	switch (init_level) {
309 	case 4:
310 		/* Initialized successfully */
311 		break;
312 	case 3:
313 		CAM_DEBUG(periph->path, CAM_DEBUG_INFO, ("Periph destroyed\n"));
314 		xpt_remove_periph(periph);
315 		/* FALLTHROUGH */
316 	case 2:
317 		xpt_lock_buses();
318 		TAILQ_REMOVE(&(*p_drv)->units, periph, unit_links);
319 		xpt_unlock_buses();
320 		xpt_free_path(periph->path);
321 		/* FALLTHROUGH */
322 	case 1:
323 		free(periph, M_CAMPERIPH);
324 		/* FALLTHROUGH */
325 	case 0:
326 		/* No cleanup to perform. */
327 		break;
328 	default:
329 		panic("%s: Unknown init level", __func__);
330 	}
331 	return(status);
332 }
333 
334 /*
335  * Find a peripheral structure with the specified path, target, lun,
336  * and (optionally) type.  If the name is NULL, this function will return
337  * the first peripheral driver that matches the specified path.
338  */
339 struct cam_periph *
340 cam_periph_find(struct cam_path *path, char *name)
341 {
342 	struct periph_driver **p_drv;
343 	struct cam_periph *periph;
344 
345 	xpt_lock_buses();
346 	for (p_drv = periph_drivers; *p_drv != NULL; p_drv++) {
347 		if (name != NULL && (strcmp((*p_drv)->driver_name, name) != 0))
348 			continue;
349 
350 		TAILQ_FOREACH(periph, &(*p_drv)->units, unit_links) {
351 			if (xpt_path_comp(periph->path, path) == 0) {
352 				xpt_unlock_buses();
353 				cam_periph_assert(periph, MA_OWNED);
354 				return(periph);
355 			}
356 		}
357 		if (name != NULL) {
358 			xpt_unlock_buses();
359 			return(NULL);
360 		}
361 	}
362 	xpt_unlock_buses();
363 	return(NULL);
364 }
365 
366 /*
367  * Find peripheral driver instances attached to the specified path.
368  */
369 int
370 cam_periph_list(struct cam_path *path, struct sbuf *sb)
371 {
372 	struct sbuf local_sb;
373 	struct periph_driver **p_drv;
374 	struct cam_periph *periph;
375 	int count;
376 	int sbuf_alloc_len;
377 
378 	sbuf_alloc_len = 16;
379 retry:
380 	sbuf_new(&local_sb, NULL, sbuf_alloc_len, SBUF_FIXEDLEN);
381 	count = 0;
382 	xpt_lock_buses();
383 	for (p_drv = periph_drivers; *p_drv != NULL; p_drv++) {
384 		TAILQ_FOREACH(periph, &(*p_drv)->units, unit_links) {
385 			if (xpt_path_comp(periph->path, path) != 0)
386 				continue;
387 
388 			if (sbuf_len(&local_sb) != 0)
389 				sbuf_cat(&local_sb, ",");
390 
391 			sbuf_printf(&local_sb, "%s%d", periph->periph_name,
392 				    periph->unit_number);
393 
394 			if (sbuf_error(&local_sb) == ENOMEM) {
395 				sbuf_alloc_len *= 2;
396 				xpt_unlock_buses();
397 				sbuf_delete(&local_sb);
398 				goto retry;
399 			}
400 			count++;
401 		}
402 	}
403 	xpt_unlock_buses();
404 	sbuf_finish(&local_sb);
405 	if (sbuf_len(sb) != 0)
406 		sbuf_cat(sb, ",");
407 	sbuf_cat(sb, sbuf_data(&local_sb));
408 	sbuf_delete(&local_sb);
409 	return (count);
410 }
411 
412 int
413 cam_periph_acquire(struct cam_periph *periph)
414 {
415 	int status;
416 
417 	if (periph == NULL)
418 		return (EINVAL);
419 
420 	status = ENOENT;
421 	xpt_lock_buses();
422 	if ((periph->flags & CAM_PERIPH_INVALID) == 0) {
423 		periph->refcount++;
424 		status = 0;
425 	}
426 	xpt_unlock_buses();
427 
428 	return (status);
429 }
430 
431 void
432 cam_periph_doacquire(struct cam_periph *periph)
433 {
434 
435 	xpt_lock_buses();
436 	KASSERT(periph->refcount >= 1,
437 	    ("cam_periph_doacquire() with refcount == %d", periph->refcount));
438 	periph->refcount++;
439 	xpt_unlock_buses();
440 }
441 
442 void
443 cam_periph_release_locked_buses(struct cam_periph *periph)
444 {
445 
446 	cam_periph_assert(periph, MA_OWNED);
447 	KASSERT(periph->refcount >= 1, ("periph->refcount >= 1"));
448 	if (--periph->refcount == 0)
449 		camperiphfree(periph);
450 }
451 
452 void
453 cam_periph_release_locked(struct cam_periph *periph)
454 {
455 
456 	if (periph == NULL)
457 		return;
458 
459 	xpt_lock_buses();
460 	cam_periph_release_locked_buses(periph);
461 	xpt_unlock_buses();
462 }
463 
464 void
465 cam_periph_release(struct cam_periph *periph)
466 {
467 	struct mtx *mtx;
468 
469 	if (periph == NULL)
470 		return;
471 
472 	cam_periph_assert(periph, MA_NOTOWNED);
473 	mtx = cam_periph_mtx(periph);
474 	mtx_lock(mtx);
475 	cam_periph_release_locked(periph);
476 	mtx_unlock(mtx);
477 }
478 
479 /*
480  * hold/unhold act as mutual exclusion for sections of the code that
481  * need to sleep and want to make sure that other sections that
482  * will interfere are held off. This only protects exclusive sections
483  * from each other.
484  */
485 int
486 cam_periph_hold(struct cam_periph *periph, int priority)
487 {
488 	int error;
489 
490 	/*
491 	 * Increment the reference count on the peripheral
492 	 * while we wait for our lock attempt to succeed
493 	 * to ensure the peripheral doesn't disappear out
494 	 * from user us while we sleep.
495 	 */
496 
497 	if (cam_periph_acquire(periph) != 0)
498 		return (ENXIO);
499 
500 	cam_periph_assert(periph, MA_OWNED);
501 	while ((periph->flags & CAM_PERIPH_LOCKED) != 0) {
502 		periph->flags |= CAM_PERIPH_LOCK_WANTED;
503 		if ((error = cam_periph_sleep(periph, periph, priority,
504 		    "caplck", 0)) != 0) {
505 			cam_periph_release_locked(periph);
506 			return (error);
507 		}
508 		if (periph->flags & CAM_PERIPH_INVALID) {
509 			cam_periph_release_locked(periph);
510 			return (ENXIO);
511 		}
512 	}
513 
514 	periph->flags |= CAM_PERIPH_LOCKED;
515 	return (0);
516 }
517 
518 void
519 cam_periph_unhold(struct cam_periph *periph)
520 {
521 
522 	cam_periph_assert(periph, MA_OWNED);
523 
524 	periph->flags &= ~CAM_PERIPH_LOCKED;
525 	if ((periph->flags & CAM_PERIPH_LOCK_WANTED) != 0) {
526 		periph->flags &= ~CAM_PERIPH_LOCK_WANTED;
527 		wakeup(periph);
528 	}
529 
530 	cam_periph_release_locked(periph);
531 }
532 
533 /*
534  * Look for the next unit number that is not currently in use for this
535  * peripheral type starting at "newunit".  Also exclude unit numbers that
536  * are reserved by for future "hardwiring" unless we already know that this
537  * is a potential wired device.  Only assume that the device is "wired" the
538  * first time through the loop since after that we'll be looking at unit
539  * numbers that did not match a wiring entry.
540  */
541 static u_int
542 camperiphnextunit(struct periph_driver *p_drv, u_int newunit, bool wired,
543 		  path_id_t pathid, target_id_t target, lun_id_t lun)
544 {
545 	struct	cam_periph *periph;
546 	char	*periph_name;
547 	int	i, val, dunit, r;
548 	const char *dname, *strval;
549 
550 	periph_name = p_drv->driver_name;
551 	for (;;newunit++) {
552 		for (periph = TAILQ_FIRST(&p_drv->units);
553 		     periph != NULL && periph->unit_number != newunit;
554 		     periph = TAILQ_NEXT(periph, unit_links))
555 			;
556 
557 		if (periph != NULL && periph->unit_number == newunit) {
558 			if (wired) {
559 				xpt_print(periph->path, "Duplicate Wired "
560 				    "Device entry!\n");
561 				xpt_print(periph->path, "Second device (%s "
562 				    "device at scbus%d target %d lun %d) will "
563 				    "not be wired\n", periph_name, pathid,
564 				    target, lun);
565 				wired = false;
566 			}
567 			continue;
568 		}
569 		if (wired)
570 			break;
571 
572 		/*
573 		 * Don't allow the mere presence of any attributes of a device
574 		 * means that it is for a wired down entry. Instead, insist that
575 		 * one of the matching criteria from camperiphunit be present
576 		 * for the device.
577 		 */
578 		i = 0;
579 		dname = periph_name;
580 		for (;;) {
581 			r = resource_find_dev(&i, dname, &dunit, NULL, NULL);
582 			if (r != 0)
583 				break;
584 
585 			if (newunit != dunit)
586 				continue;
587 			if (resource_string_value(dname, dunit, "sn", &strval) == 0 ||
588 			    resource_int_value(dname, dunit, "lun", &val) == 0 ||
589 			    resource_int_value(dname, dunit, "target", &val) == 0 ||
590 			    resource_string_value(dname, dunit, "at", &strval) == 0)
591 				break;
592 		}
593 		if (r != 0)
594 			break;
595 	}
596 	return (newunit);
597 }
598 
599 static u_int
600 camperiphunit(struct periph_driver *p_drv, path_id_t pathid,
601     target_id_t target, lun_id_t lun, const char *sn)
602 {
603 	bool	wired = false;
604 	u_int	unit;
605 	int	i, val, dunit;
606 	const char *dname, *strval;
607 	char	pathbuf[32], *periph_name;
608 
609 	periph_name = p_drv->driver_name;
610 	snprintf(pathbuf, sizeof(pathbuf), "scbus%d", pathid);
611 	unit = 0;
612 	i = 0;
613 	dname = periph_name;
614 
615 	for (wired = false; resource_find_dev(&i, dname, &dunit, NULL, NULL) == 0;
616 	     wired = false) {
617 		if (resource_string_value(dname, dunit, "at", &strval) == 0) {
618 			if (strcmp(strval, pathbuf) != 0)
619 				continue;
620 			wired = true;
621 		}
622 		if (resource_int_value(dname, dunit, "target", &val) == 0) {
623 			if (val != target)
624 				continue;
625 			wired = true;
626 		}
627 		if (resource_int_value(dname, dunit, "lun", &val) == 0) {
628 			if (val != lun)
629 				continue;
630 			wired = true;
631 		}
632 		if (resource_string_value(dname, dunit, "sn", &strval) == 0) {
633 			if (sn == NULL || strcmp(strval, sn) != 0)
634 				continue;
635 			wired = true;
636 		}
637 		if (wired) {
638 			unit = dunit;
639 			break;
640 		}
641 	}
642 
643 	/*
644 	 * Either start from 0 looking for the next unit or from
645 	 * the unit number given in the resource config.  This way,
646 	 * if we have wildcard matches, we don't return the same
647 	 * unit number twice.
648 	 */
649 	unit = camperiphnextunit(p_drv, unit, wired, pathid, target, lun);
650 
651 	return (unit);
652 }
653 
654 void
655 cam_periph_invalidate(struct cam_periph *periph)
656 {
657 
658 	cam_periph_assert(periph, MA_OWNED);
659 	/*
660 	 * We only tear down the device the first time a peripheral is
661 	 * invalidated.
662 	 */
663 	if ((periph->flags & CAM_PERIPH_INVALID) != 0)
664 		return;
665 
666 	CAM_DEBUG(periph->path, CAM_DEBUG_INFO, ("Periph invalidated\n"));
667 	if ((periph->flags & CAM_PERIPH_ANNOUNCED) && !rebooting) {
668 		struct sbuf sb;
669 		char buffer[160];
670 
671 		sbuf_new(&sb, buffer, 160, SBUF_FIXEDLEN);
672 		xpt_denounce_periph_sbuf(periph, &sb);
673 		sbuf_finish(&sb);
674 		sbuf_putbuf(&sb);
675 	}
676 	periph->flags |= CAM_PERIPH_INVALID;
677 	periph->flags &= ~CAM_PERIPH_NEW_DEV_FOUND;
678 	if (periph->periph_oninval != NULL)
679 		periph->periph_oninval(periph);
680 	cam_periph_release_locked(periph);
681 }
682 
683 static void
684 camperiphfree(struct cam_periph *periph)
685 {
686 	struct periph_driver **p_drv;
687 	struct periph_driver *drv;
688 
689 	cam_periph_assert(periph, MA_OWNED);
690 	KASSERT(periph->periph_allocating == 0, ("%s%d: freed while allocating",
691 	    periph->periph_name, periph->unit_number));
692 	for (p_drv = periph_drivers; *p_drv != NULL; p_drv++) {
693 		if (strcmp((*p_drv)->driver_name, periph->periph_name) == 0)
694 			break;
695 	}
696 	if (*p_drv == NULL) {
697 		printf("camperiphfree: attempt to free non-existant periph\n");
698 		return;
699 	}
700 	/*
701 	 * Cache a pointer to the periph_driver structure.  If a
702 	 * periph_driver is added or removed from the array (see
703 	 * periphdriver_register()) while we drop the toplogy lock
704 	 * below, p_drv may change.  This doesn't protect against this
705 	 * particular periph_driver going away.  That will require full
706 	 * reference counting in the periph_driver infrastructure.
707 	 */
708 	drv = *p_drv;
709 
710 	/*
711 	 * We need to set this flag before dropping the topology lock, to
712 	 * let anyone who is traversing the list that this peripheral is
713 	 * about to be freed, and there will be no more reference count
714 	 * checks.
715 	 */
716 	periph->flags |= CAM_PERIPH_FREE;
717 
718 	/*
719 	 * The peripheral destructor semantics dictate calling with only the
720 	 * SIM mutex held.  Since it might sleep, it should not be called
721 	 * with the topology lock held.
722 	 */
723 	xpt_unlock_buses();
724 
725 	/*
726 	 * We need to call the peripheral destructor prior to removing the
727 	 * peripheral from the list.  Otherwise, we risk running into a
728 	 * scenario where the peripheral unit number may get reused
729 	 * (because it has been removed from the list), but some resources
730 	 * used by the peripheral are still hanging around.  In particular,
731 	 * the devfs nodes used by some peripherals like the pass(4) driver
732 	 * aren't fully cleaned up until the destructor is run.  If the
733 	 * unit number is reused before the devfs instance is fully gone,
734 	 * devfs will panic.
735 	 */
736 	if (periph->periph_dtor != NULL)
737 		periph->periph_dtor(periph);
738 
739 	/*
740 	 * The peripheral list is protected by the topology lock. We have to
741 	 * remove the periph from the drv list before we call deferred_ac. The
742 	 * AC_FOUND_DEVICE callback won't create a new periph if it's still there.
743 	 */
744 	xpt_lock_buses();
745 
746 	TAILQ_REMOVE(&drv->units, periph, unit_links);
747 	drv->generation++;
748 
749 	xpt_remove_periph(periph);
750 
751 	xpt_unlock_buses();
752 	if ((periph->flags & CAM_PERIPH_ANNOUNCED) && !rebooting)
753 		xpt_print(periph->path, "Periph destroyed\n");
754 	else
755 		CAM_DEBUG(periph->path, CAM_DEBUG_INFO, ("Periph destroyed\n"));
756 
757 	if (periph->flags & CAM_PERIPH_NEW_DEV_FOUND) {
758 		union ccb ccb;
759 		void *arg;
760 
761 		memset(&ccb, 0, sizeof(ccb));
762 		switch (periph->deferred_ac) {
763 		case AC_FOUND_DEVICE:
764 			ccb.ccb_h.func_code = XPT_GDEV_TYPE;
765 			xpt_setup_ccb(&ccb.ccb_h, periph->path, CAM_PRIORITY_NORMAL);
766 			xpt_action(&ccb);
767 			arg = &ccb;
768 			break;
769 		case AC_PATH_REGISTERED:
770 			xpt_path_inq(&ccb.cpi, periph->path);
771 			arg = &ccb;
772 			break;
773 		default:
774 			arg = NULL;
775 			break;
776 		}
777 		periph->deferred_callback(NULL, periph->deferred_ac,
778 					  periph->path, arg);
779 	}
780 	xpt_free_path(periph->path);
781 	free(periph, M_CAMPERIPH);
782 	xpt_lock_buses();
783 }
784 
785 /*
786  * Map user virtual pointers into kernel virtual address space, so we can
787  * access the memory.  This is now a generic function that centralizes most
788  * of the sanity checks on the data flags, if any.
789  * This also only works for up to maxphys memory.  Since we use
790  * buffers to map stuff in and out, we're limited to the buffer size.
791  */
792 int
793 cam_periph_mapmem(union ccb *ccb, struct cam_periph_map_info *mapinfo,
794     u_int maxmap)
795 {
796 	int numbufs, i;
797 	u_int8_t **data_ptrs[CAM_PERIPH_MAXMAPS];
798 	u_int32_t lengths[CAM_PERIPH_MAXMAPS];
799 	u_int32_t dirs[CAM_PERIPH_MAXMAPS];
800 
801 	bzero(mapinfo, sizeof(*mapinfo));
802 	if (maxmap == 0)
803 		maxmap = DFLTPHYS;	/* traditional default */
804 	else if (maxmap > maxphys)
805 		maxmap = maxphys;	/* for safety */
806 	switch(ccb->ccb_h.func_code) {
807 	case XPT_DEV_MATCH:
808 		if (ccb->cdm.match_buf_len == 0) {
809 			printf("cam_periph_mapmem: invalid match buffer "
810 			       "length 0\n");
811 			return(EINVAL);
812 		}
813 		if (ccb->cdm.pattern_buf_len > 0) {
814 			data_ptrs[0] = (u_int8_t **)&ccb->cdm.patterns;
815 			lengths[0] = ccb->cdm.pattern_buf_len;
816 			dirs[0] = CAM_DIR_OUT;
817 			data_ptrs[1] = (u_int8_t **)&ccb->cdm.matches;
818 			lengths[1] = ccb->cdm.match_buf_len;
819 			dirs[1] = CAM_DIR_IN;
820 			numbufs = 2;
821 		} else {
822 			data_ptrs[0] = (u_int8_t **)&ccb->cdm.matches;
823 			lengths[0] = ccb->cdm.match_buf_len;
824 			dirs[0] = CAM_DIR_IN;
825 			numbufs = 1;
826 		}
827 		/*
828 		 * This request will not go to the hardware, no reason
829 		 * to be so strict. vmapbuf() is able to map up to maxphys.
830 		 */
831 		maxmap = maxphys;
832 		break;
833 	case XPT_SCSI_IO:
834 	case XPT_CONT_TARGET_IO:
835 		if ((ccb->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_NONE)
836 			return(0);
837 		if ((ccb->ccb_h.flags & CAM_DATA_MASK) != CAM_DATA_VADDR)
838 			return (EINVAL);
839 		data_ptrs[0] = &ccb->csio.data_ptr;
840 		lengths[0] = ccb->csio.dxfer_len;
841 		dirs[0] = ccb->ccb_h.flags & CAM_DIR_MASK;
842 		numbufs = 1;
843 		break;
844 	case XPT_ATA_IO:
845 		if ((ccb->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_NONE)
846 			return(0);
847 		if ((ccb->ccb_h.flags & CAM_DATA_MASK) != CAM_DATA_VADDR)
848 			return (EINVAL);
849 		data_ptrs[0] = &ccb->ataio.data_ptr;
850 		lengths[0] = ccb->ataio.dxfer_len;
851 		dirs[0] = ccb->ccb_h.flags & CAM_DIR_MASK;
852 		numbufs = 1;
853 		break;
854 	case XPT_MMC_IO:
855 		if ((ccb->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_NONE)
856 			return(0);
857 		/* Two mappings: one for cmd->data and one for cmd->data->data */
858 		data_ptrs[0] = (unsigned char **)&ccb->mmcio.cmd.data;
859 		lengths[0] = sizeof(struct mmc_data *);
860 		dirs[0] = ccb->ccb_h.flags & CAM_DIR_MASK;
861 		data_ptrs[1] = (unsigned char **)&ccb->mmcio.cmd.data->data;
862 		lengths[1] = ccb->mmcio.cmd.data->len;
863 		dirs[1] = ccb->ccb_h.flags & CAM_DIR_MASK;
864 		numbufs = 2;
865 		break;
866 	case XPT_SMP_IO:
867 		data_ptrs[0] = &ccb->smpio.smp_request;
868 		lengths[0] = ccb->smpio.smp_request_len;
869 		dirs[0] = CAM_DIR_OUT;
870 		data_ptrs[1] = &ccb->smpio.smp_response;
871 		lengths[1] = ccb->smpio.smp_response_len;
872 		dirs[1] = CAM_DIR_IN;
873 		numbufs = 2;
874 		break;
875 	case XPT_NVME_IO:
876 	case XPT_NVME_ADMIN:
877 		if ((ccb->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_NONE)
878 			return (0);
879 		if ((ccb->ccb_h.flags & CAM_DATA_MASK) != CAM_DATA_VADDR)
880 			return (EINVAL);
881 		data_ptrs[0] = &ccb->nvmeio.data_ptr;
882 		lengths[0] = ccb->nvmeio.dxfer_len;
883 		dirs[0] = ccb->ccb_h.flags & CAM_DIR_MASK;
884 		numbufs = 1;
885 		break;
886 	case XPT_DEV_ADVINFO:
887 		if (ccb->cdai.bufsiz == 0)
888 			return (0);
889 
890 		data_ptrs[0] = (uint8_t **)&ccb->cdai.buf;
891 		lengths[0] = ccb->cdai.bufsiz;
892 		dirs[0] = CAM_DIR_IN;
893 		numbufs = 1;
894 
895 		/*
896 		 * This request will not go to the hardware, no reason
897 		 * to be so strict. vmapbuf() is able to map up to maxphys.
898 		 */
899 		maxmap = maxphys;
900 		break;
901 	default:
902 		return(EINVAL);
903 		break; /* NOTREACHED */
904 	}
905 
906 	/*
907 	 * Check the transfer length and permissions first, so we don't
908 	 * have to unmap any previously mapped buffers.
909 	 */
910 	for (i = 0; i < numbufs; i++) {
911 		if (lengths[i] > maxmap) {
912 			printf("cam_periph_mapmem: attempt to map %lu bytes, "
913 			       "which is greater than %lu\n",
914 			       (long)(lengths[i]), (u_long)maxmap);
915 			return (E2BIG);
916 		}
917 	}
918 
919 	/*
920 	 * This keeps the kernel stack of current thread from getting
921 	 * swapped.  In low-memory situations where the kernel stack might
922 	 * otherwise get swapped out, this holds it and allows the thread
923 	 * to make progress and release the kernel mapped pages sooner.
924 	 *
925 	 * XXX KDM should I use P_NOSWAP instead?
926 	 */
927 	PHOLD(curproc);
928 
929 	for (i = 0; i < numbufs; i++) {
930 		/* Save the user's data address. */
931 		mapinfo->orig[i] = *data_ptrs[i];
932 
933 		/*
934 		 * For small buffers use malloc+copyin/copyout instead of
935 		 * mapping to KVA to avoid expensive TLB shootdowns.  For
936 		 * small allocations malloc is backed by UMA, and so much
937 		 * cheaper on SMP systems.
938 		 */
939 		if (lengths[i] <= periph_mapmem_thresh &&
940 		    ccb->ccb_h.func_code != XPT_MMC_IO) {
941 			*data_ptrs[i] = malloc(lengths[i], M_CAMPERIPH,
942 			    M_WAITOK);
943 			if (dirs[i] != CAM_DIR_IN) {
944 				if (copyin(mapinfo->orig[i], *data_ptrs[i],
945 				    lengths[i]) != 0) {
946 					free(*data_ptrs[i], M_CAMPERIPH);
947 					*data_ptrs[i] = mapinfo->orig[i];
948 					goto fail;
949 				}
950 			} else
951 				bzero(*data_ptrs[i], lengths[i]);
952 			continue;
953 		}
954 
955 		/*
956 		 * Get the buffer.
957 		 */
958 		mapinfo->bp[i] = uma_zalloc(pbuf_zone, M_WAITOK);
959 
960 		/* set the direction */
961 		mapinfo->bp[i]->b_iocmd = (dirs[i] == CAM_DIR_OUT) ?
962 		    BIO_WRITE : BIO_READ;
963 
964 		/* Map the buffer into kernel memory. */
965 		if (vmapbuf(mapinfo->bp[i], *data_ptrs[i], lengths[i], 1) < 0) {
966 			uma_zfree(pbuf_zone, mapinfo->bp[i]);
967 			goto fail;
968 		}
969 
970 		/* set our pointer to the new mapped area */
971 		*data_ptrs[i] = mapinfo->bp[i]->b_data;
972 	}
973 
974 	/*
975 	 * Now that we've gotten this far, change ownership to the kernel
976 	 * of the buffers so that we don't run afoul of returning to user
977 	 * space with locks (on the buffer) held.
978 	 */
979 	for (i = 0; i < numbufs; i++) {
980 		if (mapinfo->bp[i])
981 			BUF_KERNPROC(mapinfo->bp[i]);
982 	}
983 
984 	mapinfo->num_bufs_used = numbufs;
985 	return(0);
986 
987 fail:
988 	for (i--; i >= 0; i--) {
989 		if (mapinfo->bp[i]) {
990 			vunmapbuf(mapinfo->bp[i]);
991 			uma_zfree(pbuf_zone, mapinfo->bp[i]);
992 		} else
993 			free(*data_ptrs[i], M_CAMPERIPH);
994 		*data_ptrs[i] = mapinfo->orig[i];
995 	}
996 	PRELE(curproc);
997 	return(EACCES);
998 }
999 
1000 /*
1001  * Unmap memory segments mapped into kernel virtual address space by
1002  * cam_periph_mapmem().
1003  */
1004 void
1005 cam_periph_unmapmem(union ccb *ccb, struct cam_periph_map_info *mapinfo)
1006 {
1007 	int numbufs, i;
1008 	u_int8_t **data_ptrs[CAM_PERIPH_MAXMAPS];
1009 	u_int32_t lengths[CAM_PERIPH_MAXMAPS];
1010 	u_int32_t dirs[CAM_PERIPH_MAXMAPS];
1011 
1012 	if (mapinfo->num_bufs_used <= 0) {
1013 		/* nothing to free and the process wasn't held. */
1014 		return;
1015 	}
1016 
1017 	switch (ccb->ccb_h.func_code) {
1018 	case XPT_DEV_MATCH:
1019 		if (ccb->cdm.pattern_buf_len > 0) {
1020 			data_ptrs[0] = (u_int8_t **)&ccb->cdm.patterns;
1021 			lengths[0] = ccb->cdm.pattern_buf_len;
1022 			dirs[0] = CAM_DIR_OUT;
1023 			data_ptrs[1] = (u_int8_t **)&ccb->cdm.matches;
1024 			lengths[1] = ccb->cdm.match_buf_len;
1025 			dirs[1] = CAM_DIR_IN;
1026 			numbufs = 2;
1027 		} else {
1028 			data_ptrs[0] = (u_int8_t **)&ccb->cdm.matches;
1029 			lengths[0] = ccb->cdm.match_buf_len;
1030 			dirs[0] = CAM_DIR_IN;
1031 			numbufs = 1;
1032 		}
1033 		break;
1034 	case XPT_SCSI_IO:
1035 	case XPT_CONT_TARGET_IO:
1036 		data_ptrs[0] = &ccb->csio.data_ptr;
1037 		lengths[0] = ccb->csio.dxfer_len;
1038 		dirs[0] = ccb->ccb_h.flags & CAM_DIR_MASK;
1039 		numbufs = 1;
1040 		break;
1041 	case XPT_ATA_IO:
1042 		data_ptrs[0] = &ccb->ataio.data_ptr;
1043 		lengths[0] = ccb->ataio.dxfer_len;
1044 		dirs[0] = ccb->ccb_h.flags & CAM_DIR_MASK;
1045 		numbufs = 1;
1046 		break;
1047 	case XPT_MMC_IO:
1048 		data_ptrs[0] = (u_int8_t **)&ccb->mmcio.cmd.data;
1049 		lengths[0] = sizeof(struct mmc_data *);
1050 		dirs[0] = ccb->ccb_h.flags & CAM_DIR_MASK;
1051 		data_ptrs[1] = (u_int8_t **)&ccb->mmcio.cmd.data->data;
1052 		lengths[1] = ccb->mmcio.cmd.data->len;
1053 		dirs[1] = ccb->ccb_h.flags & CAM_DIR_MASK;
1054 		numbufs = 2;
1055 		break;
1056 	case XPT_SMP_IO:
1057 		data_ptrs[0] = &ccb->smpio.smp_request;
1058 		lengths[0] = ccb->smpio.smp_request_len;
1059 		dirs[0] = CAM_DIR_OUT;
1060 		data_ptrs[1] = &ccb->smpio.smp_response;
1061 		lengths[1] = ccb->smpio.smp_response_len;
1062 		dirs[1] = CAM_DIR_IN;
1063 		numbufs = 2;
1064 		break;
1065 	case XPT_NVME_IO:
1066 	case XPT_NVME_ADMIN:
1067 		data_ptrs[0] = &ccb->nvmeio.data_ptr;
1068 		lengths[0] = ccb->nvmeio.dxfer_len;
1069 		dirs[0] = ccb->ccb_h.flags & CAM_DIR_MASK;
1070 		numbufs = 1;
1071 		break;
1072 	case XPT_DEV_ADVINFO:
1073 		data_ptrs[0] = (uint8_t **)&ccb->cdai.buf;
1074 		lengths[0] = ccb->cdai.bufsiz;
1075 		dirs[0] = CAM_DIR_IN;
1076 		numbufs = 1;
1077 		break;
1078 	default:
1079 		/* allow ourselves to be swapped once again */
1080 		PRELE(curproc);
1081 		return;
1082 		break; /* NOTREACHED */
1083 	}
1084 
1085 	for (i = 0; i < numbufs; i++) {
1086 		if (mapinfo->bp[i]) {
1087 			/* unmap the buffer */
1088 			vunmapbuf(mapinfo->bp[i]);
1089 
1090 			/* release the buffer */
1091 			uma_zfree(pbuf_zone, mapinfo->bp[i]);
1092 		} else {
1093 			if (dirs[i] != CAM_DIR_OUT) {
1094 				copyout(*data_ptrs[i], mapinfo->orig[i],
1095 				    lengths[i]);
1096 			}
1097 			free(*data_ptrs[i], M_CAMPERIPH);
1098 		}
1099 
1100 		/* Set the user's pointer back to the original value */
1101 		*data_ptrs[i] = mapinfo->orig[i];
1102 	}
1103 
1104 	/* allow ourselves to be swapped once again */
1105 	PRELE(curproc);
1106 }
1107 
1108 int
1109 cam_periph_ioctl(struct cam_periph *periph, u_long cmd, caddr_t addr,
1110 		 int (*error_routine)(union ccb *ccb,
1111 				      cam_flags camflags,
1112 				      u_int32_t sense_flags))
1113 {
1114 	union ccb 	     *ccb;
1115 	int 		     error;
1116 	int		     found;
1117 
1118 	error = found = 0;
1119 
1120 	switch(cmd){
1121 	case CAMGETPASSTHRU:
1122 		ccb = cam_periph_getccb(periph, CAM_PRIORITY_NORMAL);
1123 		xpt_setup_ccb(&ccb->ccb_h,
1124 			      ccb->ccb_h.path,
1125 			      CAM_PRIORITY_NORMAL);
1126 		ccb->ccb_h.func_code = XPT_GDEVLIST;
1127 
1128 		/*
1129 		 * Basically, the point of this is that we go through
1130 		 * getting the list of devices, until we find a passthrough
1131 		 * device.  In the current version of the CAM code, the
1132 		 * only way to determine what type of device we're dealing
1133 		 * with is by its name.
1134 		 */
1135 		while (found == 0) {
1136 			ccb->cgdl.index = 0;
1137 			ccb->cgdl.status = CAM_GDEVLIST_MORE_DEVS;
1138 			while (ccb->cgdl.status == CAM_GDEVLIST_MORE_DEVS) {
1139 				/* we want the next device in the list */
1140 				xpt_action(ccb);
1141 				if (strncmp(ccb->cgdl.periph_name,
1142 				    "pass", 4) == 0){
1143 					found = 1;
1144 					break;
1145 				}
1146 			}
1147 			if ((ccb->cgdl.status == CAM_GDEVLIST_LAST_DEVICE) &&
1148 			    (found == 0)) {
1149 				ccb->cgdl.periph_name[0] = '\0';
1150 				ccb->cgdl.unit_number = 0;
1151 				break;
1152 			}
1153 		}
1154 
1155 		/* copy the result back out */
1156 		bcopy(ccb, addr, sizeof(union ccb));
1157 
1158 		/* and release the ccb */
1159 		xpt_release_ccb(ccb);
1160 
1161 		break;
1162 	default:
1163 		error = ENOTTY;
1164 		break;
1165 	}
1166 	return(error);
1167 }
1168 
1169 static void
1170 cam_periph_done_panic(struct cam_periph *periph, union ccb *done_ccb)
1171 {
1172 
1173 	panic("%s: already done with ccb %p", __func__, done_ccb);
1174 }
1175 
1176 static void
1177 cam_periph_done(struct cam_periph *periph, union ccb *done_ccb)
1178 {
1179 
1180 	/* Caller will release the CCB */
1181 	xpt_path_assert(done_ccb->ccb_h.path, MA_OWNED);
1182 	done_ccb->ccb_h.cbfcnp = cam_periph_done_panic;
1183 	wakeup(&done_ccb->ccb_h.cbfcnp);
1184 }
1185 
1186 static void
1187 cam_periph_ccbwait(union ccb *ccb)
1188 {
1189 
1190 	if ((ccb->ccb_h.func_code & XPT_FC_QUEUED) != 0) {
1191 		while (ccb->ccb_h.cbfcnp != cam_periph_done_panic)
1192 			xpt_path_sleep(ccb->ccb_h.path, &ccb->ccb_h.cbfcnp,
1193 			    PRIBIO, "cbwait", 0);
1194 	}
1195 	KASSERT(ccb->ccb_h.pinfo.index == CAM_UNQUEUED_INDEX &&
1196 	    (ccb->ccb_h.status & CAM_STATUS_MASK) != CAM_REQ_INPROG,
1197 	    ("%s: proceeding with incomplete ccb: ccb=%p, func_code=%#x, "
1198 	     "status=%#x, index=%d", __func__, ccb, ccb->ccb_h.func_code,
1199 	     ccb->ccb_h.status, ccb->ccb_h.pinfo.index));
1200 }
1201 
1202 /*
1203  * Dispatch a CCB and wait for it to complete.  If the CCB has set a
1204  * callback function (ccb->ccb_h.cbfcnp), it will be overwritten and lost.
1205  */
1206 int
1207 cam_periph_runccb(union ccb *ccb,
1208 		  int (*error_routine)(union ccb *ccb,
1209 				       cam_flags camflags,
1210 				       u_int32_t sense_flags),
1211 		  cam_flags camflags, u_int32_t sense_flags,
1212 		  struct devstat *ds)
1213 {
1214 	struct bintime *starttime;
1215 	struct bintime ltime;
1216 	int error;
1217 	bool must_poll;
1218 	uint32_t timeout = 1;
1219 
1220 	starttime = NULL;
1221 	xpt_path_assert(ccb->ccb_h.path, MA_OWNED);
1222 	KASSERT((ccb->ccb_h.flags & CAM_UNLOCKED) == 0,
1223 	    ("%s: ccb=%p, func_code=%#x, flags=%#x", __func__, ccb,
1224 	     ccb->ccb_h.func_code, ccb->ccb_h.flags));
1225 
1226 	/*
1227 	 * If the user has supplied a stats structure, and if we understand
1228 	 * this particular type of ccb, record the transaction start.
1229 	 */
1230 	if (ds != NULL &&
1231 	    (ccb->ccb_h.func_code == XPT_SCSI_IO ||
1232 	    ccb->ccb_h.func_code == XPT_ATA_IO ||
1233 	    ccb->ccb_h.func_code == XPT_NVME_IO)) {
1234 		starttime = &ltime;
1235 		binuptime(starttime);
1236 		devstat_start_transaction(ds, starttime);
1237 	}
1238 
1239 	/*
1240 	 * We must poll the I/O while we're dumping. The scheduler is normally
1241 	 * stopped for dumping, except when we call doadump from ddb. While the
1242 	 * scheduler is running in this case, we still need to poll the I/O to
1243 	 * avoid sleeping waiting for the ccb to complete.
1244 	 *
1245 	 * A panic triggered dump stops the scheduler, any callback from the
1246 	 * shutdown_post_sync event will run with the scheduler stopped, but
1247 	 * before we're officially dumping. To avoid hanging in adashutdown
1248 	 * initiated commands (or other similar situations), we have to test for
1249 	 * either SCHEDULER_STOPPED() here as well.
1250 	 *
1251 	 * To avoid locking problems, dumping/polling callers must call
1252 	 * without a periph lock held.
1253 	 */
1254 	must_poll = dumping || SCHEDULER_STOPPED();
1255 	ccb->ccb_h.cbfcnp = cam_periph_done;
1256 
1257 	/*
1258 	 * If we're polling, then we need to ensure that we have ample resources
1259 	 * in the periph.  cam_periph_error can reschedule the ccb by calling
1260 	 * xpt_action and returning ERESTART, so we have to effect the polling
1261 	 * in the do loop below.
1262 	 */
1263 	if (must_poll) {
1264 		if (cam_sim_pollable(ccb->ccb_h.path->bus->sim))
1265 			timeout = xpt_poll_setup(ccb);
1266 		else
1267 			timeout = 0;
1268 	}
1269 
1270 	if (timeout == 0) {
1271 		ccb->ccb_h.status = CAM_RESRC_UNAVAIL;
1272 		error = EBUSY;
1273 	} else {
1274 		xpt_action(ccb);
1275 		do {
1276 			if (must_poll) {
1277 				xpt_pollwait(ccb, timeout);
1278 				timeout = ccb->ccb_h.timeout * 10;
1279 			} else {
1280 				cam_periph_ccbwait(ccb);
1281 			}
1282 			if ((ccb->ccb_h.status & CAM_STATUS_MASK) == CAM_REQ_CMP)
1283 				error = 0;
1284 			else if (error_routine != NULL) {
1285 				ccb->ccb_h.cbfcnp = cam_periph_done;
1286 				error = (*error_routine)(ccb, camflags, sense_flags);
1287 			} else
1288 				error = 0;
1289 		} while (error == ERESTART);
1290 	}
1291 
1292 	if ((ccb->ccb_h.status & CAM_DEV_QFRZN) != 0) {
1293 		cam_release_devq(ccb->ccb_h.path,
1294 				 /* relsim_flags */0,
1295 				 /* openings */0,
1296 				 /* timeout */0,
1297 				 /* getcount_only */ FALSE);
1298 		ccb->ccb_h.status &= ~CAM_DEV_QFRZN;
1299 	}
1300 
1301 	if (ds != NULL) {
1302 		uint32_t bytes;
1303 		devstat_tag_type tag;
1304 		bool valid = true;
1305 
1306 		if (ccb->ccb_h.func_code == XPT_SCSI_IO) {
1307 			bytes = ccb->csio.dxfer_len - ccb->csio.resid;
1308 			tag = (devstat_tag_type)(ccb->csio.tag_action & 0x3);
1309 		} else if (ccb->ccb_h.func_code == XPT_ATA_IO) {
1310 			bytes = ccb->ataio.dxfer_len - ccb->ataio.resid;
1311 			tag = (devstat_tag_type)0;
1312 		} else if (ccb->ccb_h.func_code == XPT_NVME_IO) {
1313 			bytes = ccb->nvmeio.dxfer_len; /* NB: resid no possible */
1314 			tag = (devstat_tag_type)0;
1315 		} else {
1316 			valid = false;
1317 		}
1318 		if (valid)
1319 			devstat_end_transaction(ds, bytes, tag,
1320 			    ((ccb->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_NONE) ?
1321 			    DEVSTAT_NO_DATA : (ccb->ccb_h.flags & CAM_DIR_OUT) ?
1322 			    DEVSTAT_WRITE : DEVSTAT_READ, NULL, starttime);
1323 	}
1324 
1325 	return(error);
1326 }
1327 
1328 void
1329 cam_freeze_devq(struct cam_path *path)
1330 {
1331 	struct ccb_hdr ccb_h;
1332 
1333 	CAM_DEBUG(path, CAM_DEBUG_TRACE, ("cam_freeze_devq\n"));
1334 	memset(&ccb_h, 0, sizeof(ccb_h));
1335 	xpt_setup_ccb(&ccb_h, path, /*priority*/1);
1336 	ccb_h.func_code = XPT_NOOP;
1337 	ccb_h.flags = CAM_DEV_QFREEZE;
1338 	xpt_action((union ccb *)&ccb_h);
1339 }
1340 
1341 u_int32_t
1342 cam_release_devq(struct cam_path *path, u_int32_t relsim_flags,
1343 		 u_int32_t openings, u_int32_t arg,
1344 		 int getcount_only)
1345 {
1346 	struct ccb_relsim crs;
1347 
1348 	CAM_DEBUG(path, CAM_DEBUG_TRACE, ("cam_release_devq(%u, %u, %u, %d)\n",
1349 	    relsim_flags, openings, arg, getcount_only));
1350 	memset(&crs, 0, sizeof(crs));
1351 	xpt_setup_ccb(&crs.ccb_h, path, CAM_PRIORITY_NORMAL);
1352 	crs.ccb_h.func_code = XPT_REL_SIMQ;
1353 	crs.ccb_h.flags = getcount_only ? CAM_DEV_QFREEZE : 0;
1354 	crs.release_flags = relsim_flags;
1355 	crs.openings = openings;
1356 	crs.release_timeout = arg;
1357 	xpt_action((union ccb *)&crs);
1358 	return (crs.qfrozen_cnt);
1359 }
1360 
1361 #define saved_ccb_ptr ppriv_ptr0
1362 static void
1363 camperiphdone(struct cam_periph *periph, union ccb *done_ccb)
1364 {
1365 	union ccb      *saved_ccb;
1366 	cam_status	status;
1367 	struct scsi_start_stop_unit *scsi_cmd;
1368 	int		error = 0, error_code, sense_key, asc, ascq;
1369 	u_int16_t	done_flags;
1370 
1371 	scsi_cmd = (struct scsi_start_stop_unit *)
1372 	    &done_ccb->csio.cdb_io.cdb_bytes;
1373 	status = done_ccb->ccb_h.status;
1374 
1375 	if ((status & CAM_STATUS_MASK) != CAM_REQ_CMP) {
1376 		if (scsi_extract_sense_ccb(done_ccb,
1377 		    &error_code, &sense_key, &asc, &ascq)) {
1378 			/*
1379 			 * If the error is "invalid field in CDB",
1380 			 * and the load/eject flag is set, turn the
1381 			 * flag off and try again.  This is just in
1382 			 * case the drive in question barfs on the
1383 			 * load eject flag.  The CAM code should set
1384 			 * the load/eject flag by default for
1385 			 * removable media.
1386 			 */
1387 			if ((scsi_cmd->opcode == START_STOP_UNIT) &&
1388 			    ((scsi_cmd->how & SSS_LOEJ) != 0) &&
1389 			     (asc == 0x24) && (ascq == 0x00)) {
1390 				scsi_cmd->how &= ~SSS_LOEJ;
1391 				if (status & CAM_DEV_QFRZN) {
1392 					cam_release_devq(done_ccb->ccb_h.path,
1393 					    0, 0, 0, 0);
1394 					done_ccb->ccb_h.status &=
1395 					    ~CAM_DEV_QFRZN;
1396 				}
1397 				xpt_action(done_ccb);
1398 				goto out;
1399 			}
1400 		}
1401 		error = cam_periph_error(done_ccb, 0,
1402 		    SF_RETRY_UA | SF_NO_PRINT);
1403 		if (error == ERESTART)
1404 			goto out;
1405 		if (done_ccb->ccb_h.status & CAM_DEV_QFRZN) {
1406 			cam_release_devq(done_ccb->ccb_h.path, 0, 0, 0, 0);
1407 			done_ccb->ccb_h.status &= ~CAM_DEV_QFRZN;
1408 		}
1409 	} else {
1410 		/*
1411 		 * If we have successfully taken a device from the not
1412 		 * ready to ready state, re-scan the device and re-get
1413 		 * the inquiry information.  Many devices (mostly disks)
1414 		 * don't properly report their inquiry information unless
1415 		 * they are spun up.
1416 		 */
1417 		if (scsi_cmd->opcode == START_STOP_UNIT)
1418 			xpt_async(AC_INQ_CHANGED, done_ccb->ccb_h.path, NULL);
1419 	}
1420 
1421 	/* If we tried long wait and still failed, remember that. */
1422 	if ((periph->flags & CAM_PERIPH_RECOVERY_WAIT) &&
1423 	    (done_ccb->csio.cdb_io.cdb_bytes[0] == TEST_UNIT_READY)) {
1424 		periph->flags &= ~CAM_PERIPH_RECOVERY_WAIT;
1425 		if (error != 0 && done_ccb->ccb_h.retry_count == 0)
1426 			periph->flags |= CAM_PERIPH_RECOVERY_WAIT_FAILED;
1427 	}
1428 
1429 	/*
1430 	 * After recovery action(s) completed, return to the original CCB.
1431 	 * If the recovery CCB has failed, considering its own possible
1432 	 * retries and recovery, assume we are back in state where we have
1433 	 * been originally, but without recovery hopes left.  In such case,
1434 	 * after the final attempt below, we cancel any further retries,
1435 	 * blocking by that also any new recovery attempts for this CCB,
1436 	 * and the result will be the final one returned to the CCB owher.
1437 	 */
1438 
1439 	/*
1440 	 * Copy the CCB back, preserving the alloc_flags field.  Things
1441 	 * will crash horribly if the CCBs are not of the same size.
1442 	 */
1443 	saved_ccb = (union ccb *)done_ccb->ccb_h.saved_ccb_ptr;
1444 	KASSERT(saved_ccb->ccb_h.func_code == XPT_SCSI_IO,
1445 	    ("%s: saved_ccb func_code %#x != XPT_SCSI_IO",
1446 	     __func__, saved_ccb->ccb_h.func_code));
1447 	KASSERT(done_ccb->ccb_h.func_code == XPT_SCSI_IO,
1448 	    ("%s: done_ccb func_code %#x != XPT_SCSI_IO",
1449 	     __func__, done_ccb->ccb_h.func_code));
1450 	done_flags = done_ccb->ccb_h.alloc_flags;
1451 	bcopy(saved_ccb, done_ccb, sizeof(struct ccb_scsiio));
1452 	done_ccb->ccb_h.alloc_flags = done_flags;
1453 	xpt_free_ccb(saved_ccb);
1454 	if (done_ccb->ccb_h.cbfcnp != camperiphdone)
1455 		periph->flags &= ~CAM_PERIPH_RECOVERY_INPROG;
1456 	if (error != 0)
1457 		done_ccb->ccb_h.retry_count = 0;
1458 	xpt_action(done_ccb);
1459 
1460 out:
1461 	/* Drop freeze taken due to CAM_DEV_QFREEZE flag set. */
1462 	cam_release_devq(done_ccb->ccb_h.path, 0, 0, 0, 0);
1463 }
1464 
1465 /*
1466  * Generic Async Event handler.  Peripheral drivers usually
1467  * filter out the events that require personal attention,
1468  * and leave the rest to this function.
1469  */
1470 void
1471 cam_periph_async(struct cam_periph *periph, u_int32_t code,
1472 		 struct cam_path *path, void *arg)
1473 {
1474 	switch (code) {
1475 	case AC_LOST_DEVICE:
1476 		cam_periph_invalidate(periph);
1477 		break;
1478 	default:
1479 		break;
1480 	}
1481 }
1482 
1483 void
1484 cam_periph_bus_settle(struct cam_periph *periph, u_int bus_settle)
1485 {
1486 	struct ccb_getdevstats cgds;
1487 
1488 	memset(&cgds, 0, sizeof(cgds));
1489 	xpt_setup_ccb(&cgds.ccb_h, periph->path, CAM_PRIORITY_NORMAL);
1490 	cgds.ccb_h.func_code = XPT_GDEV_STATS;
1491 	xpt_action((union ccb *)&cgds);
1492 	cam_periph_freeze_after_event(periph, &cgds.last_reset, bus_settle);
1493 }
1494 
1495 void
1496 cam_periph_freeze_after_event(struct cam_periph *periph,
1497 			      struct timeval* event_time, u_int duration_ms)
1498 {
1499 	struct timeval delta;
1500 	struct timeval duration_tv;
1501 
1502 	if (!timevalisset(event_time))
1503 		return;
1504 
1505 	microtime(&delta);
1506 	timevalsub(&delta, event_time);
1507 	duration_tv.tv_sec = duration_ms / 1000;
1508 	duration_tv.tv_usec = (duration_ms % 1000) * 1000;
1509 	if (timevalcmp(&delta, &duration_tv, <)) {
1510 		timevalsub(&duration_tv, &delta);
1511 
1512 		duration_ms = duration_tv.tv_sec * 1000;
1513 		duration_ms += duration_tv.tv_usec / 1000;
1514 		cam_freeze_devq(periph->path);
1515 		cam_release_devq(periph->path,
1516 				RELSIM_RELEASE_AFTER_TIMEOUT,
1517 				/*reduction*/0,
1518 				/*timeout*/duration_ms,
1519 				/*getcount_only*/0);
1520 	}
1521 
1522 }
1523 
1524 static int
1525 camperiphscsistatuserror(union ccb *ccb, union ccb **orig_ccb,
1526     cam_flags camflags, u_int32_t sense_flags,
1527     int *openings, u_int32_t *relsim_flags,
1528     u_int32_t *timeout, u_int32_t *action, const char **action_string)
1529 {
1530 	struct cam_periph *periph;
1531 	int error;
1532 
1533 	switch (ccb->csio.scsi_status) {
1534 	case SCSI_STATUS_OK:
1535 	case SCSI_STATUS_COND_MET:
1536 	case SCSI_STATUS_INTERMED:
1537 	case SCSI_STATUS_INTERMED_COND_MET:
1538 		error = 0;
1539 		break;
1540 	case SCSI_STATUS_CMD_TERMINATED:
1541 	case SCSI_STATUS_CHECK_COND:
1542 		error = camperiphscsisenseerror(ccb, orig_ccb,
1543 					        camflags,
1544 					        sense_flags,
1545 					        openings,
1546 					        relsim_flags,
1547 					        timeout,
1548 					        action,
1549 					        action_string);
1550 		break;
1551 	case SCSI_STATUS_QUEUE_FULL:
1552 	{
1553 		/* no decrement */
1554 		struct ccb_getdevstats cgds;
1555 
1556 		/*
1557 		 * First off, find out what the current
1558 		 * transaction counts are.
1559 		 */
1560 		memset(&cgds, 0, sizeof(cgds));
1561 		xpt_setup_ccb(&cgds.ccb_h,
1562 			      ccb->ccb_h.path,
1563 			      CAM_PRIORITY_NORMAL);
1564 		cgds.ccb_h.func_code = XPT_GDEV_STATS;
1565 		xpt_action((union ccb *)&cgds);
1566 
1567 		/*
1568 		 * If we were the only transaction active, treat
1569 		 * the QUEUE FULL as if it were a BUSY condition.
1570 		 */
1571 		if (cgds.dev_active != 0) {
1572 			int total_openings;
1573 
1574 			/*
1575 		 	 * Reduce the number of openings to
1576 			 * be 1 less than the amount it took
1577 			 * to get a queue full bounded by the
1578 			 * minimum allowed tag count for this
1579 			 * device.
1580 		 	 */
1581 			total_openings = cgds.dev_active + cgds.dev_openings;
1582 			*openings = cgds.dev_active;
1583 			if (*openings < cgds.mintags)
1584 				*openings = cgds.mintags;
1585 			if (*openings < total_openings)
1586 				*relsim_flags = RELSIM_ADJUST_OPENINGS;
1587 			else {
1588 				/*
1589 				 * Some devices report queue full for
1590 				 * temporary resource shortages.  For
1591 				 * this reason, we allow a minimum
1592 				 * tag count to be entered via a
1593 				 * quirk entry to prevent the queue
1594 				 * count on these devices from falling
1595 				 * to a pessimisticly low value.  We
1596 				 * still wait for the next successful
1597 				 * completion, however, before queueing
1598 				 * more transactions to the device.
1599 				 */
1600 				*relsim_flags = RELSIM_RELEASE_AFTER_CMDCMPLT;
1601 			}
1602 			*timeout = 0;
1603 			error = ERESTART;
1604 			*action &= ~SSQ_PRINT_SENSE;
1605 			break;
1606 		}
1607 		/* FALLTHROUGH */
1608 	}
1609 	case SCSI_STATUS_BUSY:
1610 		/*
1611 		 * Restart the queue after either another
1612 		 * command completes or a 1 second timeout.
1613 		 */
1614 		periph = xpt_path_periph(ccb->ccb_h.path);
1615 		if (periph->flags & CAM_PERIPH_INVALID) {
1616 			error = EIO;
1617 			*action_string = "Periph was invalidated";
1618 		} else if ((sense_flags & SF_RETRY_BUSY) != 0 ||
1619 		    ccb->ccb_h.retry_count > 0) {
1620 			if ((sense_flags & SF_RETRY_BUSY) == 0)
1621 				ccb->ccb_h.retry_count--;
1622 			error = ERESTART;
1623 			*relsim_flags = RELSIM_RELEASE_AFTER_TIMEOUT
1624 				      | RELSIM_RELEASE_AFTER_CMDCMPLT;
1625 			*timeout = 1000;
1626 		} else {
1627 			error = EIO;
1628 			*action_string = "Retries exhausted";
1629 		}
1630 		break;
1631 	case SCSI_STATUS_RESERV_CONFLICT:
1632 	default:
1633 		error = EIO;
1634 		break;
1635 	}
1636 	return (error);
1637 }
1638 
1639 static int
1640 camperiphscsisenseerror(union ccb *ccb, union ccb **orig,
1641     cam_flags camflags, u_int32_t sense_flags,
1642     int *openings, u_int32_t *relsim_flags,
1643     u_int32_t *timeout, u_int32_t *action, const char **action_string)
1644 {
1645 	struct cam_periph *periph;
1646 	union ccb *orig_ccb = ccb;
1647 	int error, recoveryccb;
1648 	u_int16_t flags;
1649 
1650 #if defined(BUF_TRACKING) || defined(FULL_BUF_TRACKING)
1651 	if (ccb->ccb_h.func_code == XPT_SCSI_IO && ccb->csio.bio != NULL)
1652 		biotrack(ccb->csio.bio, __func__);
1653 #endif
1654 
1655 	periph = xpt_path_periph(ccb->ccb_h.path);
1656 	recoveryccb = (ccb->ccb_h.cbfcnp == camperiphdone);
1657 	if ((periph->flags & CAM_PERIPH_RECOVERY_INPROG) && !recoveryccb) {
1658 		/*
1659 		 * If error recovery is already in progress, don't attempt
1660 		 * to process this error, but requeue it unconditionally
1661 		 * and attempt to process it once error recovery has
1662 		 * completed.  This failed command is probably related to
1663 		 * the error that caused the currently active error recovery
1664 		 * action so our  current recovery efforts should also
1665 		 * address this command.  Be aware that the error recovery
1666 		 * code assumes that only one recovery action is in progress
1667 		 * on a particular peripheral instance at any given time
1668 		 * (e.g. only one saved CCB for error recovery) so it is
1669 		 * imperitive that we don't violate this assumption.
1670 		 */
1671 		error = ERESTART;
1672 		*action &= ~SSQ_PRINT_SENSE;
1673 	} else {
1674 		scsi_sense_action err_action;
1675 		struct ccb_getdev cgd;
1676 
1677 		/*
1678 		 * Grab the inquiry data for this device.
1679 		 */
1680 		memset(&cgd, 0, sizeof(cgd));
1681 		xpt_setup_ccb(&cgd.ccb_h, ccb->ccb_h.path, CAM_PRIORITY_NORMAL);
1682 		cgd.ccb_h.func_code = XPT_GDEV_TYPE;
1683 		xpt_action((union ccb *)&cgd);
1684 
1685 		err_action = scsi_error_action(&ccb->csio, &cgd.inq_data,
1686 		    sense_flags);
1687 		error = err_action & SS_ERRMASK;
1688 
1689 		/*
1690 		 * Do not autostart sequential access devices
1691 		 * to avoid unexpected tape loading.
1692 		 */
1693 		if ((err_action & SS_MASK) == SS_START &&
1694 		    SID_TYPE(&cgd.inq_data) == T_SEQUENTIAL) {
1695 			*action_string = "Will not autostart a "
1696 			    "sequential access device";
1697 			goto sense_error_done;
1698 		}
1699 
1700 		/*
1701 		 * Avoid recovery recursion if recovery action is the same.
1702 		 */
1703 		if ((err_action & SS_MASK) >= SS_START && recoveryccb) {
1704 			if (((err_action & SS_MASK) == SS_START &&
1705 			     ccb->csio.cdb_io.cdb_bytes[0] == START_STOP_UNIT) ||
1706 			    ((err_action & SS_MASK) == SS_TUR &&
1707 			     (ccb->csio.cdb_io.cdb_bytes[0] == TEST_UNIT_READY))) {
1708 				err_action = SS_RETRY|SSQ_DECREMENT_COUNT|EIO;
1709 				*relsim_flags = RELSIM_RELEASE_AFTER_TIMEOUT;
1710 				*timeout = 500;
1711 			}
1712 		}
1713 
1714 		/*
1715 		 * If the recovery action will consume a retry,
1716 		 * make sure we actually have retries available.
1717 		 */
1718 		if ((err_action & SSQ_DECREMENT_COUNT) != 0) {
1719 		 	if (ccb->ccb_h.retry_count > 0 &&
1720 			    (periph->flags & CAM_PERIPH_INVALID) == 0)
1721 		 		ccb->ccb_h.retry_count--;
1722 			else {
1723 				*action_string = "Retries exhausted";
1724 				goto sense_error_done;
1725 			}
1726 		}
1727 
1728 		if ((err_action & SS_MASK) >= SS_START) {
1729 			/*
1730 			 * Do common portions of commands that
1731 			 * use recovery CCBs.
1732 			 */
1733 			orig_ccb = xpt_alloc_ccb_nowait();
1734 			if (orig_ccb == NULL) {
1735 				*action_string = "Can't allocate recovery CCB";
1736 				goto sense_error_done;
1737 			}
1738 			/*
1739 			 * Clear freeze flag for original request here, as
1740 			 * this freeze will be dropped as part of ERESTART.
1741 			 */
1742 			ccb->ccb_h.status &= ~CAM_DEV_QFRZN;
1743 
1744 			KASSERT(ccb->ccb_h.func_code == XPT_SCSI_IO,
1745 			    ("%s: ccb func_code %#x != XPT_SCSI_IO",
1746 			     __func__, ccb->ccb_h.func_code));
1747 			flags = orig_ccb->ccb_h.alloc_flags;
1748 			bcopy(ccb, orig_ccb, sizeof(struct ccb_scsiio));
1749 			orig_ccb->ccb_h.alloc_flags = flags;
1750 		}
1751 
1752 		switch (err_action & SS_MASK) {
1753 		case SS_NOP:
1754 			*action_string = "No recovery action needed";
1755 			error = 0;
1756 			break;
1757 		case SS_RETRY:
1758 			*action_string = "Retrying command (per sense data)";
1759 			error = ERESTART;
1760 			break;
1761 		case SS_FAIL:
1762 			*action_string = "Unretryable error";
1763 			break;
1764 		case SS_START:
1765 		{
1766 			int le;
1767 
1768 			/*
1769 			 * Send a start unit command to the device, and
1770 			 * then retry the command.
1771 			 */
1772 			*action_string = "Attempting to start unit";
1773 			periph->flags |= CAM_PERIPH_RECOVERY_INPROG;
1774 
1775 			/*
1776 			 * Check for removable media and set
1777 			 * load/eject flag appropriately.
1778 			 */
1779 			if (SID_IS_REMOVABLE(&cgd.inq_data))
1780 				le = TRUE;
1781 			else
1782 				le = FALSE;
1783 
1784 			scsi_start_stop(&ccb->csio,
1785 					/*retries*/1,
1786 					camperiphdone,
1787 					MSG_SIMPLE_Q_TAG,
1788 					/*start*/TRUE,
1789 					/*load/eject*/le,
1790 					/*immediate*/FALSE,
1791 					SSD_FULL_SIZE,
1792 					/*timeout*/50000);
1793 			break;
1794 		}
1795 		case SS_TUR:
1796 		{
1797 			/*
1798 			 * Send a Test Unit Ready to the device.
1799 			 * If the 'many' flag is set, we send 120
1800 			 * test unit ready commands, one every half
1801 			 * second.  Otherwise, we just send one TUR.
1802 			 * We only want to do this if the retry
1803 			 * count has not been exhausted.
1804 			 */
1805 			int retries;
1806 
1807 			if ((err_action & SSQ_MANY) != 0 && (periph->flags &
1808 			     CAM_PERIPH_RECOVERY_WAIT_FAILED) == 0) {
1809 				periph->flags |= CAM_PERIPH_RECOVERY_WAIT;
1810 				*action_string = "Polling device for readiness";
1811 				retries = 120;
1812 			} else {
1813 				*action_string = "Testing device for readiness";
1814 				retries = 1;
1815 			}
1816 			periph->flags |= CAM_PERIPH_RECOVERY_INPROG;
1817 			scsi_test_unit_ready(&ccb->csio,
1818 					     retries,
1819 					     camperiphdone,
1820 					     MSG_SIMPLE_Q_TAG,
1821 					     SSD_FULL_SIZE,
1822 					     /*timeout*/5000);
1823 
1824 			/*
1825 			 * Accomplish our 500ms delay by deferring
1826 			 * the release of our device queue appropriately.
1827 			 */
1828 			*relsim_flags = RELSIM_RELEASE_AFTER_TIMEOUT;
1829 			*timeout = 500;
1830 			break;
1831 		}
1832 		default:
1833 			panic("Unhandled error action %x", err_action);
1834 		}
1835 
1836 		if ((err_action & SS_MASK) >= SS_START) {
1837 			/*
1838 			 * Drop the priority, so that the recovery
1839 			 * CCB is the first to execute.  Freeze the queue
1840 			 * after this command is sent so that we can
1841 			 * restore the old csio and have it queued in
1842 			 * the proper order before we release normal
1843 			 * transactions to the device.
1844 			 */
1845 			ccb->ccb_h.pinfo.priority--;
1846 			ccb->ccb_h.flags |= CAM_DEV_QFREEZE;
1847 			ccb->ccb_h.saved_ccb_ptr = orig_ccb;
1848 			error = ERESTART;
1849 			*orig = orig_ccb;
1850 		}
1851 
1852 sense_error_done:
1853 		*action = err_action;
1854 	}
1855 	return (error);
1856 }
1857 
1858 /*
1859  * Generic error handler.  Peripheral drivers usually filter
1860  * out the errors that they handle in a unique manner, then
1861  * call this function.
1862  */
1863 int
1864 cam_periph_error(union ccb *ccb, cam_flags camflags,
1865 		 u_int32_t sense_flags)
1866 {
1867 	struct cam_path *newpath;
1868 	union ccb  *orig_ccb, *scan_ccb;
1869 	struct cam_periph *periph;
1870 	const char *action_string;
1871 	cam_status  status;
1872 	int	    frozen, error, openings, devctl_err;
1873 	u_int32_t   action, relsim_flags, timeout;
1874 
1875 	action = SSQ_PRINT_SENSE;
1876 	periph = xpt_path_periph(ccb->ccb_h.path);
1877 	action_string = NULL;
1878 	status = ccb->ccb_h.status;
1879 	frozen = (status & CAM_DEV_QFRZN) != 0;
1880 	status &= CAM_STATUS_MASK;
1881 	devctl_err = openings = relsim_flags = timeout = 0;
1882 	orig_ccb = ccb;
1883 
1884 	/* Filter the errors that should be reported via devctl */
1885 	switch (ccb->ccb_h.status & CAM_STATUS_MASK) {
1886 	case CAM_CMD_TIMEOUT:
1887 	case CAM_REQ_ABORTED:
1888 	case CAM_REQ_CMP_ERR:
1889 	case CAM_REQ_TERMIO:
1890 	case CAM_UNREC_HBA_ERROR:
1891 	case CAM_DATA_RUN_ERR:
1892 	case CAM_SCSI_STATUS_ERROR:
1893 	case CAM_ATA_STATUS_ERROR:
1894 	case CAM_SMP_STATUS_ERROR:
1895 		devctl_err++;
1896 		break;
1897 	default:
1898 		break;
1899 	}
1900 
1901 	switch (status) {
1902 	case CAM_REQ_CMP:
1903 		error = 0;
1904 		action &= ~SSQ_PRINT_SENSE;
1905 		break;
1906 	case CAM_SCSI_STATUS_ERROR:
1907 		error = camperiphscsistatuserror(ccb, &orig_ccb,
1908 		    camflags, sense_flags, &openings, &relsim_flags,
1909 		    &timeout, &action, &action_string);
1910 		break;
1911 	case CAM_AUTOSENSE_FAIL:
1912 		error = EIO;	/* we have to kill the command */
1913 		break;
1914 	case CAM_UA_ABORT:
1915 	case CAM_UA_TERMIO:
1916 	case CAM_MSG_REJECT_REC:
1917 		/* XXX Don't know that these are correct */
1918 		error = EIO;
1919 		break;
1920 	case CAM_SEL_TIMEOUT:
1921 		if ((camflags & CAM_RETRY_SELTO) != 0) {
1922 			if (ccb->ccb_h.retry_count > 0 &&
1923 			    (periph->flags & CAM_PERIPH_INVALID) == 0) {
1924 				ccb->ccb_h.retry_count--;
1925 				error = ERESTART;
1926 
1927 				/*
1928 				 * Wait a bit to give the device
1929 				 * time to recover before we try again.
1930 				 */
1931 				relsim_flags = RELSIM_RELEASE_AFTER_TIMEOUT;
1932 				timeout = periph_selto_delay;
1933 				break;
1934 			}
1935 			action_string = "Retries exhausted";
1936 		}
1937 		/* FALLTHROUGH */
1938 	case CAM_DEV_NOT_THERE:
1939 		error = ENXIO;
1940 		action = SSQ_LOST;
1941 		break;
1942 	case CAM_REQ_INVALID:
1943 	case CAM_PATH_INVALID:
1944 	case CAM_NO_HBA:
1945 	case CAM_PROVIDE_FAIL:
1946 	case CAM_REQ_TOO_BIG:
1947 	case CAM_LUN_INVALID:
1948 	case CAM_TID_INVALID:
1949 	case CAM_FUNC_NOTAVAIL:
1950 		error = EINVAL;
1951 		break;
1952 	case CAM_SCSI_BUS_RESET:
1953 	case CAM_BDR_SENT:
1954 		/*
1955 		 * Commands that repeatedly timeout and cause these
1956 		 * kinds of error recovery actions, should return
1957 		 * CAM_CMD_TIMEOUT, which allows us to safely assume
1958 		 * that this command was an innocent bystander to
1959 		 * these events and should be unconditionally
1960 		 * retried.
1961 		 */
1962 	case CAM_REQUEUE_REQ:
1963 		/* Unconditional requeue if device is still there */
1964 		if (periph->flags & CAM_PERIPH_INVALID) {
1965 			action_string = "Periph was invalidated";
1966 			error = EIO;
1967 		} else if (sense_flags & SF_NO_RETRY) {
1968 			error = EIO;
1969 			action_string = "Retry was blocked";
1970 		} else {
1971 			error = ERESTART;
1972 			action &= ~SSQ_PRINT_SENSE;
1973 		}
1974 		break;
1975 	case CAM_RESRC_UNAVAIL:
1976 		/* Wait a bit for the resource shortage to abate. */
1977 		timeout = periph_noresrc_delay;
1978 		/* FALLTHROUGH */
1979 	case CAM_BUSY:
1980 		if (timeout == 0) {
1981 			/* Wait a bit for the busy condition to abate. */
1982 			timeout = periph_busy_delay;
1983 		}
1984 		relsim_flags = RELSIM_RELEASE_AFTER_TIMEOUT;
1985 		/* FALLTHROUGH */
1986 	case CAM_ATA_STATUS_ERROR:
1987 	case CAM_REQ_CMP_ERR:
1988 	case CAM_CMD_TIMEOUT:
1989 	case CAM_UNEXP_BUSFREE:
1990 	case CAM_UNCOR_PARITY:
1991 	case CAM_DATA_RUN_ERR:
1992 	default:
1993 		if (periph->flags & CAM_PERIPH_INVALID) {
1994 			error = EIO;
1995 			action_string = "Periph was invalidated";
1996 		} else if (ccb->ccb_h.retry_count == 0) {
1997 			error = EIO;
1998 			action_string = "Retries exhausted";
1999 		} else if (sense_flags & SF_NO_RETRY) {
2000 			error = EIO;
2001 			action_string = "Retry was blocked";
2002 		} else {
2003 			ccb->ccb_h.retry_count--;
2004 			error = ERESTART;
2005 		}
2006 		break;
2007 	}
2008 
2009 	if ((sense_flags & SF_PRINT_ALWAYS) ||
2010 	    CAM_DEBUGGED(ccb->ccb_h.path, CAM_DEBUG_INFO))
2011 		action |= SSQ_PRINT_SENSE;
2012 	else if (sense_flags & SF_NO_PRINT)
2013 		action &= ~SSQ_PRINT_SENSE;
2014 	if ((action & SSQ_PRINT_SENSE) != 0)
2015 		cam_error_print(orig_ccb, CAM_ESF_ALL, CAM_EPF_ALL);
2016 	if (error != 0 && (action & SSQ_PRINT_SENSE) != 0) {
2017 		if (error != ERESTART) {
2018 			if (action_string == NULL)
2019 				action_string = "Unretryable error";
2020 			xpt_print(ccb->ccb_h.path, "Error %d, %s\n",
2021 			    error, action_string);
2022 		} else if (action_string != NULL)
2023 			xpt_print(ccb->ccb_h.path, "%s\n", action_string);
2024 		else {
2025 			xpt_print(ccb->ccb_h.path,
2026 			    "Retrying command, %d more tries remain\n",
2027 			    ccb->ccb_h.retry_count);
2028 		}
2029 	}
2030 
2031 	if (devctl_err && (error != 0 || (action & SSQ_PRINT_SENSE) != 0))
2032 		cam_periph_devctl_notify(orig_ccb);
2033 
2034 	if ((action & SSQ_LOST) != 0) {
2035 		lun_id_t lun_id;
2036 
2037 		/*
2038 		 * For a selection timeout, we consider all of the LUNs on
2039 		 * the target to be gone.  If the status is CAM_DEV_NOT_THERE,
2040 		 * then we only get rid of the device(s) specified by the
2041 		 * path in the original CCB.
2042 		 */
2043 		if (status == CAM_SEL_TIMEOUT)
2044 			lun_id = CAM_LUN_WILDCARD;
2045 		else
2046 			lun_id = xpt_path_lun_id(ccb->ccb_h.path);
2047 
2048 		/* Should we do more if we can't create the path?? */
2049 		if (xpt_create_path(&newpath, periph,
2050 				    xpt_path_path_id(ccb->ccb_h.path),
2051 				    xpt_path_target_id(ccb->ccb_h.path),
2052 				    lun_id) == CAM_REQ_CMP) {
2053 			/*
2054 			 * Let peripheral drivers know that this
2055 			 * device has gone away.
2056 			 */
2057 			xpt_async(AC_LOST_DEVICE, newpath, NULL);
2058 			xpt_free_path(newpath);
2059 		}
2060 	}
2061 
2062 	/* Broadcast UNIT ATTENTIONs to all periphs. */
2063 	if ((action & SSQ_UA) != 0)
2064 		xpt_async(AC_UNIT_ATTENTION, orig_ccb->ccb_h.path, orig_ccb);
2065 
2066 	/* Rescan target on "Reported LUNs data has changed" */
2067 	if ((action & SSQ_RESCAN) != 0) {
2068 		if (xpt_create_path(&newpath, NULL,
2069 				    xpt_path_path_id(ccb->ccb_h.path),
2070 				    xpt_path_target_id(ccb->ccb_h.path),
2071 				    CAM_LUN_WILDCARD) == CAM_REQ_CMP) {
2072 			scan_ccb = xpt_alloc_ccb_nowait();
2073 			if (scan_ccb != NULL) {
2074 				scan_ccb->ccb_h.path = newpath;
2075 				scan_ccb->ccb_h.func_code = XPT_SCAN_TGT;
2076 				scan_ccb->crcn.flags = 0;
2077 				xpt_rescan(scan_ccb);
2078 			} else {
2079 				xpt_print(newpath,
2080 				    "Can't allocate CCB to rescan target\n");
2081 				xpt_free_path(newpath);
2082 			}
2083 		}
2084 	}
2085 
2086 	/* Attempt a retry */
2087 	if (error == ERESTART || error == 0) {
2088 		if (frozen != 0)
2089 			ccb->ccb_h.status &= ~CAM_DEV_QFRZN;
2090 		if (error == ERESTART)
2091 			xpt_action(ccb);
2092 		if (frozen != 0)
2093 			cam_release_devq(ccb->ccb_h.path,
2094 					 relsim_flags,
2095 					 openings,
2096 					 timeout,
2097 					 /*getcount_only*/0);
2098 	}
2099 
2100 	return (error);
2101 }
2102 
2103 #define CAM_PERIPH_DEVD_MSG_SIZE	256
2104 
2105 static void
2106 cam_periph_devctl_notify(union ccb *ccb)
2107 {
2108 	struct cam_periph *periph;
2109 	struct ccb_getdev *cgd;
2110 	struct sbuf sb;
2111 	int serr, sk, asc, ascq;
2112 	char *sbmsg, *type;
2113 
2114 	sbmsg = malloc(CAM_PERIPH_DEVD_MSG_SIZE, M_CAMPERIPH, M_NOWAIT);
2115 	if (sbmsg == NULL)
2116 		return;
2117 
2118 	sbuf_new(&sb, sbmsg, CAM_PERIPH_DEVD_MSG_SIZE, SBUF_FIXEDLEN);
2119 
2120 	periph = xpt_path_periph(ccb->ccb_h.path);
2121 	sbuf_printf(&sb, "device=%s%d ", periph->periph_name,
2122 	    periph->unit_number);
2123 
2124 	sbuf_printf(&sb, "serial=\"");
2125 	if ((cgd = (struct ccb_getdev *)xpt_alloc_ccb_nowait()) != NULL) {
2126 		xpt_setup_ccb(&cgd->ccb_h, ccb->ccb_h.path,
2127 		    CAM_PRIORITY_NORMAL);
2128 		cgd->ccb_h.func_code = XPT_GDEV_TYPE;
2129 		xpt_action((union ccb *)cgd);
2130 
2131 		if (cgd->ccb_h.status == CAM_REQ_CMP)
2132 			sbuf_bcat(&sb, cgd->serial_num, cgd->serial_num_len);
2133 		xpt_free_ccb((union ccb *)cgd);
2134 	}
2135 	sbuf_printf(&sb, "\" ");
2136 	sbuf_printf(&sb, "cam_status=\"0x%x\" ", ccb->ccb_h.status);
2137 
2138 	switch (ccb->ccb_h.status & CAM_STATUS_MASK) {
2139 	case CAM_CMD_TIMEOUT:
2140 		sbuf_printf(&sb, "timeout=%d ", ccb->ccb_h.timeout);
2141 		type = "timeout";
2142 		break;
2143 	case CAM_SCSI_STATUS_ERROR:
2144 		sbuf_printf(&sb, "scsi_status=%d ", ccb->csio.scsi_status);
2145 		if (scsi_extract_sense_ccb(ccb, &serr, &sk, &asc, &ascq))
2146 			sbuf_printf(&sb, "scsi_sense=\"%02x %02x %02x %02x\" ",
2147 			    serr, sk, asc, ascq);
2148 		type = "error";
2149 		break;
2150 	case CAM_ATA_STATUS_ERROR:
2151 		sbuf_printf(&sb, "RES=\"");
2152 		ata_res_sbuf(&ccb->ataio.res, &sb);
2153 		sbuf_printf(&sb, "\" ");
2154 		type = "error";
2155 		break;
2156 	default:
2157 		type = "error";
2158 		break;
2159 	}
2160 
2161 	if (ccb->ccb_h.func_code == XPT_SCSI_IO) {
2162 		sbuf_printf(&sb, "CDB=\"");
2163 		scsi_cdb_sbuf(scsiio_cdb_ptr(&ccb->csio), &sb);
2164 		sbuf_printf(&sb, "\" ");
2165 	} else if (ccb->ccb_h.func_code == XPT_ATA_IO) {
2166 		sbuf_printf(&sb, "ACB=\"");
2167 		ata_cmd_sbuf(&ccb->ataio.cmd, &sb);
2168 		sbuf_printf(&sb, "\" ");
2169 	}
2170 
2171 	if (sbuf_finish(&sb) == 0)
2172 		devctl_notify("CAM", "periph", type, sbuf_data(&sb));
2173 	sbuf_delete(&sb);
2174 	free(sbmsg, M_CAMPERIPH);
2175 }
2176 
2177 /*
2178  * Sysctl to force an invalidation of the drive right now. Can be
2179  * called with CTLFLAG_MPSAFE since we take periph lock.
2180  */
2181 int
2182 cam_periph_invalidate_sysctl(SYSCTL_HANDLER_ARGS)
2183 {
2184 	struct cam_periph *periph;
2185 	int error, value;
2186 
2187 	periph = arg1;
2188 	value = 0;
2189 	error = sysctl_handle_int(oidp, &value, 0, req);
2190 	if (error != 0 || req->newptr == NULL || value != 1)
2191 		return (error);
2192 
2193 	cam_periph_lock(periph);
2194 	cam_periph_invalidate(periph);
2195 	cam_periph_unlock(periph);
2196 
2197 	return (0);
2198 }
2199