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