1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22 /*
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2011, 2014 by Delphix. All rights reserved.
26 */
27
28 #include <sys/zfs_context.h>
29 #include <sys/fm/fs/zfs.h>
30 #include <sys/spa.h>
31 #include <sys/spa_impl.h>
32 #include <sys/dmu.h>
33 #include <sys/dmu_tx.h>
34 #include <sys/vdev_impl.h>
35 #include <sys/uberblock_impl.h>
36 #include <sys/metaslab.h>
37 #include <sys/metaslab_impl.h>
38 #include <sys/space_map.h>
39 #include <sys/space_reftree.h>
40 #include <sys/zio.h>
41 #include <sys/zap.h>
42 #include <sys/fs/zfs.h>
43 #include <sys/arc.h>
44 #include <sys/zil.h>
45 #include <sys/dsl_scan.h>
46
47 /*
48 * Virtual device management.
49 */
50
51 static vdev_ops_t *vdev_ops_table[] = {
52 &vdev_root_ops,
53 &vdev_raidz_ops,
54 &vdev_mirror_ops,
55 &vdev_replacing_ops,
56 &vdev_spare_ops,
57 &vdev_disk_ops,
58 &vdev_file_ops,
59 &vdev_missing_ops,
60 &vdev_hole_ops,
61 NULL
62 };
63
64 /* maximum scrub/resilver I/O queue per leaf vdev */
65 int zfs_scrub_limit = 10;
66
67 /*
68 * Given a vdev type, return the appropriate ops vector.
69 */
70 static vdev_ops_t *
71 vdev_getops(const char *type)
72 {
73 vdev_ops_t *ops, **opspp;
74
75 for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++)
76 if (strcmp(ops->vdev_op_type, type) == 0)
77 break;
78
79 return (ops);
80 }
81
82 /*
83 * Default asize function: return the MAX of psize with the asize of
84 * all children. This is what's used by anything other than RAID-Z.
85 */
86 uint64_t
87 vdev_default_asize(vdev_t *vd, uint64_t psize)
88 {
89 uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift);
90 uint64_t csize;
91
92 for (int c = 0; c < vd->vdev_children; c++) {
93 csize = vdev_psize_to_asize(vd->vdev_child[c], psize);
94 asize = MAX(asize, csize);
95 }
96
97 return (asize);
98 }
99
100 /*
101 * Get the minimum allocatable size. We define the allocatable size as
102 * the vdev's asize rounded to the nearest metaslab. This allows us to
103 * replace or attach devices which don't have the same physical size but
104 * can still satisfy the same number of allocations.
105 */
106 uint64_t
107 vdev_get_min_asize(vdev_t *vd)
108 {
109 vdev_t *pvd = vd->vdev_parent;
110
111 /*
112 * If our parent is NULL (inactive spare or cache) or is the root,
113 * just return our own asize.
114 */
115 if (pvd == NULL)
116 return (vd->vdev_asize);
117
118 /*
119 * The top-level vdev just returns the allocatable size rounded
120 * to the nearest metaslab.
121 */
122 if (vd == vd->vdev_top)
123 return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift));
124
125 /*
126 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
127 * so each child must provide at least 1/Nth of its asize.
128 */
129 if (pvd->vdev_ops == &vdev_raidz_ops)
130 return (pvd->vdev_min_asize / pvd->vdev_children);
131
132 return (pvd->vdev_min_asize);
133 }
134
135 void
136 vdev_set_min_asize(vdev_t *vd)
137 {
138 vd->vdev_min_asize = vdev_get_min_asize(vd);
139
140 for (int c = 0; c < vd->vdev_children; c++)
141 vdev_set_min_asize(vd->vdev_child[c]);
142 }
143
144 vdev_t *
145 vdev_lookup_top(spa_t *spa, uint64_t vdev)
146 {
147 vdev_t *rvd = spa->spa_root_vdev;
148
149 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
150
151 if (vdev < rvd->vdev_children) {
152 ASSERT(rvd->vdev_child[vdev] != NULL);
153 return (rvd->vdev_child[vdev]);
154 }
155
156 return (NULL);
157 }
158
159 vdev_t *
160 vdev_lookup_by_guid(vdev_t *vd, uint64_t guid)
161 {
162 vdev_t *mvd;
163
164 if (vd->vdev_guid == guid)
165 return (vd);
166
167 for (int c = 0; c < vd->vdev_children; c++)
168 if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) !=
169 NULL)
170 return (mvd);
171
172 return (NULL);
173 }
174
175 static int
176 vdev_count_leaves_impl(vdev_t *vd)
177 {
178 vdev_t *mvd;
179 int n = 0;
180
181 if (vd->vdev_children == 0)
182 return (1);
183
184 for (int c = 0; c < vd->vdev_children; c++)
185 n += vdev_count_leaves_impl(vd->vdev_child[c]);
186
187 return (n);
188 }
189
190 int
191 vdev_count_leaves(spa_t *spa)
192 {
193 return (vdev_count_leaves_impl(spa->spa_root_vdev));
194 }
195
196 void
197 vdev_add_child(vdev_t *pvd, vdev_t *cvd)
198 {
199 size_t oldsize, newsize;
200 uint64_t id = cvd->vdev_id;
201 vdev_t **newchild;
202
203 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
204 ASSERT(cvd->vdev_parent == NULL);
205
206 cvd->vdev_parent = pvd;
207
208 if (pvd == NULL)
209 return;
210
211 ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL);
212
213 oldsize = pvd->vdev_children * sizeof (vdev_t *);
214 pvd->vdev_children = MAX(pvd->vdev_children, id + 1);
215 newsize = pvd->vdev_children * sizeof (vdev_t *);
216
217 newchild = kmem_zalloc(newsize, KM_SLEEP);
218 if (pvd->vdev_child != NULL) {
219 bcopy(pvd->vdev_child, newchild, oldsize);
220 kmem_free(pvd->vdev_child, oldsize);
221 }
222
223 pvd->vdev_child = newchild;
224 pvd->vdev_child[id] = cvd;
225
226 cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd);
227 ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL);
228
229 /*
230 * Walk up all ancestors to update guid sum.
231 */
232 for (; pvd != NULL; pvd = pvd->vdev_parent)
233 pvd->vdev_guid_sum += cvd->vdev_guid_sum;
234 }
235
236 void
237 vdev_remove_child(vdev_t *pvd, vdev_t *cvd)
238 {
239 int c;
240 uint_t id = cvd->vdev_id;
241
242 ASSERT(cvd->vdev_parent == pvd);
243
244 if (pvd == NULL)
245 return;
246
247 ASSERT(id < pvd->vdev_children);
248 ASSERT(pvd->vdev_child[id] == cvd);
249
250 pvd->vdev_child[id] = NULL;
251 cvd->vdev_parent = NULL;
252
253 for (c = 0; c < pvd->vdev_children; c++)
254 if (pvd->vdev_child[c])
255 break;
256
257 if (c == pvd->vdev_children) {
258 kmem_free(pvd->vdev_child, c * sizeof (vdev_t *));
259 pvd->vdev_child = NULL;
260 pvd->vdev_children = 0;
261 }
262
263 /*
264 * Walk up all ancestors to update guid sum.
265 */
266 for (; pvd != NULL; pvd = pvd->vdev_parent)
267 pvd->vdev_guid_sum -= cvd->vdev_guid_sum;
268 }
269
270 /*
271 * Remove any holes in the child array.
272 */
273 void
274 vdev_compact_children(vdev_t *pvd)
275 {
276 vdev_t **newchild, *cvd;
277 int oldc = pvd->vdev_children;
278 int newc;
279
280 ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
281
282 for (int c = newc = 0; c < oldc; c++)
283 if (pvd->vdev_child[c])
284 newc++;
285
286 newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP);
287
288 for (int c = newc = 0; c < oldc; c++) {
289 if ((cvd = pvd->vdev_child[c]) != NULL) {
290 newchild[newc] = cvd;
291 cvd->vdev_id = newc++;
292 }
293 }
294
295 kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *));
296 pvd->vdev_child = newchild;
297 pvd->vdev_children = newc;
298 }
299
300 /*
301 * Allocate and minimally initialize a vdev_t.
302 */
303 vdev_t *
304 vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops)
305 {
306 vdev_t *vd;
307
308 vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP);
309
310 if (spa->spa_root_vdev == NULL) {
311 ASSERT(ops == &vdev_root_ops);
312 spa->spa_root_vdev = vd;
313 spa->spa_load_guid = spa_generate_guid(NULL);
314 }
315
316 if (guid == 0 && ops != &vdev_hole_ops) {
317 if (spa->spa_root_vdev == vd) {
318 /*
319 * The root vdev's guid will also be the pool guid,
320 * which must be unique among all pools.
321 */
322 guid = spa_generate_guid(NULL);
323 } else {
324 /*
325 * Any other vdev's guid must be unique within the pool.
326 */
327 guid = spa_generate_guid(spa);
328 }
329 ASSERT(!spa_guid_exists(spa_guid(spa), guid));
330 }
331
332 vd->vdev_spa = spa;
333 vd->vdev_id = id;
334 vd->vdev_guid = guid;
335 vd->vdev_guid_sum = guid;
336 vd->vdev_ops = ops;
337 vd->vdev_state = VDEV_STATE_CLOSED;
338 vd->vdev_ishole = (ops == &vdev_hole_ops);
339
340 mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL);
341 mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL);
342 mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL);
343 for (int t = 0; t < DTL_TYPES; t++) {
344 vd->vdev_dtl[t] = range_tree_create(NULL, NULL,
345 &vd->vdev_dtl_lock);
346 }
347 txg_list_create(&vd->vdev_ms_list,
348 offsetof(struct metaslab, ms_txg_node));
349 txg_list_create(&vd->vdev_dtl_list,
350 offsetof(struct vdev, vdev_dtl_node));
351 vd->vdev_stat.vs_timestamp = gethrtime();
352 vdev_queue_init(vd);
353 vdev_cache_init(vd);
354
355 return (vd);
356 }
357
358 /*
359 * Allocate a new vdev. The 'alloctype' is used to control whether we are
360 * creating a new vdev or loading an existing one - the behavior is slightly
361 * different for each case.
362 */
363 int
364 vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
365 int alloctype)
366 {
367 vdev_ops_t *ops;
368 char *type;
369 uint64_t guid = 0, islog, nparity;
370 vdev_t *vd;
371
372 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
373
374 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
375 return (SET_ERROR(EINVAL));
376
377 if ((ops = vdev_getops(type)) == NULL)
378 return (SET_ERROR(EINVAL));
379
380 /*
381 * If this is a load, get the vdev guid from the nvlist.
382 * Otherwise, vdev_alloc_common() will generate one for us.
383 */
384 if (alloctype == VDEV_ALLOC_LOAD) {
385 uint64_t label_id;
386
387 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) ||
388 label_id != id)
389 return (SET_ERROR(EINVAL));
390
391 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
392 return (SET_ERROR(EINVAL));
393 } else if (alloctype == VDEV_ALLOC_SPARE) {
394 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
395 return (SET_ERROR(EINVAL));
396 } else if (alloctype == VDEV_ALLOC_L2CACHE) {
397 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
398 return (SET_ERROR(EINVAL));
399 } else if (alloctype == VDEV_ALLOC_ROOTPOOL) {
400 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
401 return (SET_ERROR(EINVAL));
402 }
403
404 /*
405 * The first allocated vdev must be of type 'root'.
406 */
407 if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL)
408 return (SET_ERROR(EINVAL));
409
410 /*
411 * Determine whether we're a log vdev.
412 */
413 islog = 0;
414 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog);
415 if (islog && spa_version(spa) < SPA_VERSION_SLOGS)
416 return (SET_ERROR(ENOTSUP));
417
418 if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES)
419 return (SET_ERROR(ENOTSUP));
420
421 /*
422 * Set the nparity property for RAID-Z vdevs.
423 */
424 nparity = -1ULL;
425 if (ops == &vdev_raidz_ops) {
426 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
427 &nparity) == 0) {
428 if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY)
429 return (SET_ERROR(EINVAL));
430 /*
431 * Previous versions could only support 1 or 2 parity
432 * device.
433 */
434 if (nparity > 1 &&
435 spa_version(spa) < SPA_VERSION_RAIDZ2)
436 return (SET_ERROR(ENOTSUP));
437 if (nparity > 2 &&
438 spa_version(spa) < SPA_VERSION_RAIDZ3)
439 return (SET_ERROR(ENOTSUP));
440 } else {
441 /*
442 * We require the parity to be specified for SPAs that
443 * support multiple parity levels.
444 */
445 if (spa_version(spa) >= SPA_VERSION_RAIDZ2)
446 return (SET_ERROR(EINVAL));
447 /*
448 * Otherwise, we default to 1 parity device for RAID-Z.
449 */
450 nparity = 1;
451 }
452 } else {
453 nparity = 0;
454 }
455 ASSERT(nparity != -1ULL);
456
457 vd = vdev_alloc_common(spa, id, guid, ops);
458
459 vd->vdev_islog = islog;
460 vd->vdev_nparity = nparity;
461
462 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0)
463 vd->vdev_path = spa_strdup(vd->vdev_path);
464 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0)
465 vd->vdev_devid = spa_strdup(vd->vdev_devid);
466 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH,
467 &vd->vdev_physpath) == 0)
468 vd->vdev_physpath = spa_strdup(vd->vdev_physpath);
469 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0)
470 vd->vdev_fru = spa_strdup(vd->vdev_fru);
471
472 /*
473 * Set the whole_disk property. If it's not specified, leave the value
474 * as -1.
475 */
476 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
477 &vd->vdev_wholedisk) != 0)
478 vd->vdev_wholedisk = -1ULL;
479
480 /*
481 * Look for the 'not present' flag. This will only be set if the device
482 * was not present at the time of import.
483 */
484 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
485 &vd->vdev_not_present);
486
487 /*
488 * Get the alignment requirement.
489 */
490 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift);
491
492 /*
493 * Retrieve the vdev creation time.
494 */
495 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG,
496 &vd->vdev_crtxg);
497
498 /*
499 * If we're a top-level vdev, try to load the allocation parameters.
500 */
501 if (parent && !parent->vdev_parent &&
502 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
503 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
504 &vd->vdev_ms_array);
505 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
506 &vd->vdev_ms_shift);
507 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE,
508 &vd->vdev_asize);
509 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING,
510 &vd->vdev_removing);
511 }
512
513 if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) {
514 ASSERT(alloctype == VDEV_ALLOC_LOAD ||
515 alloctype == VDEV_ALLOC_ADD ||
516 alloctype == VDEV_ALLOC_SPLIT ||
517 alloctype == VDEV_ALLOC_ROOTPOOL);
518 vd->vdev_mg = metaslab_group_create(islog ?
519 spa_log_class(spa) : spa_normal_class(spa), vd);
520 }
521
522 /*
523 * If we're a leaf vdev, try to load the DTL object and other state.
524 */
525 if (vd->vdev_ops->vdev_op_leaf &&
526 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE ||
527 alloctype == VDEV_ALLOC_ROOTPOOL)) {
528 if (alloctype == VDEV_ALLOC_LOAD) {
529 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL,
530 &vd->vdev_dtl_object);
531 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE,
532 &vd->vdev_unspare);
533 }
534
535 if (alloctype == VDEV_ALLOC_ROOTPOOL) {
536 uint64_t spare = 0;
537
538 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
539 &spare) == 0 && spare)
540 spa_spare_add(vd);
541 }
542
543 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE,
544 &vd->vdev_offline);
545
546 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
547 &vd->vdev_resilver_txg);
548
549 /*
550 * When importing a pool, we want to ignore the persistent fault
551 * state, as the diagnosis made on another system may not be
552 * valid in the current context. Local vdevs will
553 * remain in the faulted state.
554 */
555 if (spa_load_state(spa) == SPA_LOAD_OPEN) {
556 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED,
557 &vd->vdev_faulted);
558 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED,
559 &vd->vdev_degraded);
560 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED,
561 &vd->vdev_removed);
562
563 if (vd->vdev_faulted || vd->vdev_degraded) {
564 char *aux;
565
566 vd->vdev_label_aux =
567 VDEV_AUX_ERR_EXCEEDED;
568 if (nvlist_lookup_string(nv,
569 ZPOOL_CONFIG_AUX_STATE, &aux) == 0 &&
570 strcmp(aux, "external") == 0)
571 vd->vdev_label_aux = VDEV_AUX_EXTERNAL;
572 }
573 }
574 }
575
576 /*
577 * Add ourselves to the parent's list of children.
578 */
579 vdev_add_child(parent, vd);
580
581 *vdp = vd;
582
583 return (0);
584 }
585
586 void
587 vdev_free(vdev_t *vd)
588 {
589 spa_t *spa = vd->vdev_spa;
590
591 /*
592 * vdev_free() implies closing the vdev first. This is simpler than
593 * trying to ensure complicated semantics for all callers.
594 */
595 vdev_close(vd);
596
597 ASSERT(!list_link_active(&vd->vdev_config_dirty_node));
598 ASSERT(!list_link_active(&vd->vdev_state_dirty_node));
599
600 /*
601 * Free all children.
602 */
603 for (int c = 0; c < vd->vdev_children; c++)
604 vdev_free(vd->vdev_child[c]);
605
606 ASSERT(vd->vdev_child == NULL);
607 ASSERT(vd->vdev_guid_sum == vd->vdev_guid);
608
609 /*
610 * Discard allocation state.
611 */
612 if (vd->vdev_mg != NULL) {
613 vdev_metaslab_fini(vd);
614 metaslab_group_destroy(vd->vdev_mg);
615 }
616
617 ASSERT0(vd->vdev_stat.vs_space);
618 ASSERT0(vd->vdev_stat.vs_dspace);
619 ASSERT0(vd->vdev_stat.vs_alloc);
620
621 /*
622 * Remove this vdev from its parent's child list.
623 */
624 vdev_remove_child(vd->vdev_parent, vd);
625
626 ASSERT(vd->vdev_parent == NULL);
627
628 /*
629 * Clean up vdev structure.
630 */
631 vdev_queue_fini(vd);
632 vdev_cache_fini(vd);
633
634 if (vd->vdev_path)
635 spa_strfree(vd->vdev_path);
636 if (vd->vdev_devid)
637 spa_strfree(vd->vdev_devid);
638 if (vd->vdev_physpath)
639 spa_strfree(vd->vdev_physpath);
640 if (vd->vdev_fru)
641 spa_strfree(vd->vdev_fru);
642
643 if (vd->vdev_isspare)
644 spa_spare_remove(vd);
645 if (vd->vdev_isl2cache)
646 spa_l2cache_remove(vd);
647
648 txg_list_destroy(&vd->vdev_ms_list);
649 txg_list_destroy(&vd->vdev_dtl_list);
650
651 mutex_enter(&vd->vdev_dtl_lock);
652 space_map_close(vd->vdev_dtl_sm);
653 for (int t = 0; t < DTL_TYPES; t++) {
654 range_tree_vacate(vd->vdev_dtl[t], NULL, NULL);
655 range_tree_destroy(vd->vdev_dtl[t]);
656 }
657 mutex_exit(&vd->vdev_dtl_lock);
658
659 mutex_destroy(&vd->vdev_dtl_lock);
660 mutex_destroy(&vd->vdev_stat_lock);
661 mutex_destroy(&vd->vdev_probe_lock);
662
663 if (vd == spa->spa_root_vdev)
664 spa->spa_root_vdev = NULL;
665
666 kmem_free(vd, sizeof (vdev_t));
667 }
668
669 /*
670 * Transfer top-level vdev state from svd to tvd.
671 */
672 static void
673 vdev_top_transfer(vdev_t *svd, vdev_t *tvd)
674 {
675 spa_t *spa = svd->vdev_spa;
676 metaslab_t *msp;
677 vdev_t *vd;
678 int t;
679
680 ASSERT(tvd == tvd->vdev_top);
681
682 tvd->vdev_ms_array = svd->vdev_ms_array;
683 tvd->vdev_ms_shift = svd->vdev_ms_shift;
684 tvd->vdev_ms_count = svd->vdev_ms_count;
685
686 svd->vdev_ms_array = 0;
687 svd->vdev_ms_shift = 0;
688 svd->vdev_ms_count = 0;
689
690 if (tvd->vdev_mg)
691 ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg);
692 tvd->vdev_mg = svd->vdev_mg;
693 tvd->vdev_ms = svd->vdev_ms;
694
695 svd->vdev_mg = NULL;
696 svd->vdev_ms = NULL;
697
698 if (tvd->vdev_mg != NULL)
699 tvd->vdev_mg->mg_vd = tvd;
700
701 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc;
702 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space;
703 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace;
704
705 svd->vdev_stat.vs_alloc = 0;
706 svd->vdev_stat.vs_space = 0;
707 svd->vdev_stat.vs_dspace = 0;
708
709 for (t = 0; t < TXG_SIZE; t++) {
710 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL)
711 (void) txg_list_add(&tvd->vdev_ms_list, msp, t);
712 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL)
713 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t);
714 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t))
715 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t);
716 }
717
718 if (list_link_active(&svd->vdev_config_dirty_node)) {
719 vdev_config_clean(svd);
720 vdev_config_dirty(tvd);
721 }
722
723 if (list_link_active(&svd->vdev_state_dirty_node)) {
724 vdev_state_clean(svd);
725 vdev_state_dirty(tvd);
726 }
727
728 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio;
729 svd->vdev_deflate_ratio = 0;
730
731 tvd->vdev_islog = svd->vdev_islog;
732 svd->vdev_islog = 0;
733 }
734
735 static void
736 vdev_top_update(vdev_t *tvd, vdev_t *vd)
737 {
738 if (vd == NULL)
739 return;
740
741 vd->vdev_top = tvd;
742
743 for (int c = 0; c < vd->vdev_children; c++)
744 vdev_top_update(tvd, vd->vdev_child[c]);
745 }
746
747 /*
748 * Add a mirror/replacing vdev above an existing vdev.
749 */
750 vdev_t *
751 vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops)
752 {
753 spa_t *spa = cvd->vdev_spa;
754 vdev_t *pvd = cvd->vdev_parent;
755 vdev_t *mvd;
756
757 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
758
759 mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops);
760
761 mvd->vdev_asize = cvd->vdev_asize;
762 mvd->vdev_min_asize = cvd->vdev_min_asize;
763 mvd->vdev_max_asize = cvd->vdev_max_asize;
764 mvd->vdev_ashift = cvd->vdev_ashift;
765 mvd->vdev_state = cvd->vdev_state;
766 mvd->vdev_crtxg = cvd->vdev_crtxg;
767
768 vdev_remove_child(pvd, cvd);
769 vdev_add_child(pvd, mvd);
770 cvd->vdev_id = mvd->vdev_children;
771 vdev_add_child(mvd, cvd);
772 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
773
774 if (mvd == mvd->vdev_top)
775 vdev_top_transfer(cvd, mvd);
776
777 return (mvd);
778 }
779
780 /*
781 * Remove a 1-way mirror/replacing vdev from the tree.
782 */
783 void
784 vdev_remove_parent(vdev_t *cvd)
785 {
786 vdev_t *mvd = cvd->vdev_parent;
787 vdev_t *pvd = mvd->vdev_parent;
788
789 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
790
791 ASSERT(mvd->vdev_children == 1);
792 ASSERT(mvd->vdev_ops == &vdev_mirror_ops ||
793 mvd->vdev_ops == &vdev_replacing_ops ||
794 mvd->vdev_ops == &vdev_spare_ops);
795 cvd->vdev_ashift = mvd->vdev_ashift;
796
797 vdev_remove_child(mvd, cvd);
798 vdev_remove_child(pvd, mvd);
799
800 /*
801 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
802 * Otherwise, we could have detached an offline device, and when we
803 * go to import the pool we'll think we have two top-level vdevs,
804 * instead of a different version of the same top-level vdev.
805 */
806 if (mvd->vdev_top == mvd) {
807 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid;
808 cvd->vdev_orig_guid = cvd->vdev_guid;
809 cvd->vdev_guid += guid_delta;
810 cvd->vdev_guid_sum += guid_delta;
811 }
812 cvd->vdev_id = mvd->vdev_id;
813 vdev_add_child(pvd, cvd);
814 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
815
816 if (cvd == cvd->vdev_top)
817 vdev_top_transfer(mvd, cvd);
818
819 ASSERT(mvd->vdev_children == 0);
820 vdev_free(mvd);
821 }
822
823 int
824 vdev_metaslab_init(vdev_t *vd, uint64_t txg)
825 {
826 spa_t *spa = vd->vdev_spa;
827 objset_t *mos = spa->spa_meta_objset;
828 uint64_t m;
829 uint64_t oldc = vd->vdev_ms_count;
830 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift;
831 metaslab_t **mspp;
832 int error;
833
834 ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER));
835
836 /*
837 * This vdev is not being allocated from yet or is a hole.
838 */
839 if (vd->vdev_ms_shift == 0)
840 return (0);
841
842 ASSERT(!vd->vdev_ishole);
843
844 /*
845 * Compute the raidz-deflation ratio. Note, we hard-code
846 * in 128k (1 << 17) because it is the current "typical" blocksize.
847 * Even if SPA_MAXBLOCKSIZE changes, this algorithm must never change,
848 * or we will inconsistently account for existing bp's.
849 */
850 vd->vdev_deflate_ratio = (1 << 17) /
851 (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT);
852
853 ASSERT(oldc <= newc);
854
855 mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP);
856
857 if (oldc != 0) {
858 bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp));
859 kmem_free(vd->vdev_ms, oldc * sizeof (*mspp));
860 }
861
862 vd->vdev_ms = mspp;
863 vd->vdev_ms_count = newc;
864
865 for (m = oldc; m < newc; m++) {
866 uint64_t object = 0;
867
868 if (txg == 0) {
869 error = dmu_read(mos, vd->vdev_ms_array,
870 m * sizeof (uint64_t), sizeof (uint64_t), &object,
871 DMU_READ_PREFETCH);
872 if (error)
873 return (error);
874 }
875 vd->vdev_ms[m] = metaslab_init(vd->vdev_mg, m, object, txg);
876 }
877
878 if (txg == 0)
879 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER);
880
881 /*
882 * If the vdev is being removed we don't activate
883 * the metaslabs since we want to ensure that no new
884 * allocations are performed on this device.
885 */
886 if (oldc == 0 && !vd->vdev_removing)
887 metaslab_group_activate(vd->vdev_mg);
888
889 if (txg == 0)
890 spa_config_exit(spa, SCL_ALLOC, FTAG);
891
892 return (0);
893 }
894
895 void
896 vdev_metaslab_fini(vdev_t *vd)
897 {
898 uint64_t m;
899 uint64_t count = vd->vdev_ms_count;
900
901 if (vd->vdev_ms != NULL) {
902 metaslab_group_passivate(vd->vdev_mg);
903 for (m = 0; m < count; m++) {
904 metaslab_t *msp = vd->vdev_ms[m];
905
906 if (msp != NULL)
907 metaslab_fini(msp);
908 }
909 kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *));
910 vd->vdev_ms = NULL;
911 }
912 }
913
914 typedef struct vdev_probe_stats {
915 boolean_t vps_readable;
916 boolean_t vps_writeable;
917 int vps_flags;
918 } vdev_probe_stats_t;
919
920 static void
921 vdev_probe_done(zio_t *zio)
922 {
923 spa_t *spa = zio->io_spa;
924 vdev_t *vd = zio->io_vd;
925 vdev_probe_stats_t *vps = zio->io_private;
926
927 ASSERT(vd->vdev_probe_zio != NULL);
928
929 if (zio->io_type == ZIO_TYPE_READ) {
930 if (zio->io_error == 0)
931 vps->vps_readable = 1;
932 if (zio->io_error == 0 && spa_writeable(spa)) {
933 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd,
934 zio->io_offset, zio->io_size, zio->io_data,
935 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
936 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE));
937 } else {
938 zio_buf_free(zio->io_data, zio->io_size);
939 }
940 } else if (zio->io_type == ZIO_TYPE_WRITE) {
941 if (zio->io_error == 0)
942 vps->vps_writeable = 1;
943 zio_buf_free(zio->io_data, zio->io_size);
944 } else if (zio->io_type == ZIO_TYPE_NULL) {
945 zio_t *pio;
946
947 vd->vdev_cant_read |= !vps->vps_readable;
948 vd->vdev_cant_write |= !vps->vps_writeable;
949
950 if (vdev_readable(vd) &&
951 (vdev_writeable(vd) || !spa_writeable(spa))) {
952 zio->io_error = 0;
953 } else {
954 ASSERT(zio->io_error != 0);
955 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE,
956 spa, vd, NULL, 0, 0);
957 zio->io_error = SET_ERROR(ENXIO);
958 }
959
960 mutex_enter(&vd->vdev_probe_lock);
961 ASSERT(vd->vdev_probe_zio == zio);
962 vd->vdev_probe_zio = NULL;
963 mutex_exit(&vd->vdev_probe_lock);
964
965 while ((pio = zio_walk_parents(zio)) != NULL)
966 if (!vdev_accessible(vd, pio))
967 pio->io_error = SET_ERROR(ENXIO);
968
969 kmem_free(vps, sizeof (*vps));
970 }
971 }
972
973 /*
974 * Determine whether this device is accessible.
975 *
976 * Read and write to several known locations: the pad regions of each
977 * vdev label but the first, which we leave alone in case it contains
978 * a VTOC.
979 */
980 zio_t *
981 vdev_probe(vdev_t *vd, zio_t *zio)
982 {
983 spa_t *spa = vd->vdev_spa;
984 vdev_probe_stats_t *vps = NULL;
985 zio_t *pio;
986
987 ASSERT(vd->vdev_ops->vdev_op_leaf);
988
989 /*
990 * Don't probe the probe.
991 */
992 if (zio && (zio->io_flags & ZIO_FLAG_PROBE))
993 return (NULL);
994
995 /*
996 * To prevent 'probe storms' when a device fails, we create
997 * just one probe i/o at a time. All zios that want to probe
998 * this vdev will become parents of the probe io.
999 */
1000 mutex_enter(&vd->vdev_probe_lock);
1001
1002 if ((pio = vd->vdev_probe_zio) == NULL) {
1003 vps = kmem_zalloc(sizeof (*vps), KM_SLEEP);
1004
1005 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE |
1006 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE |
1007 ZIO_FLAG_TRYHARD;
1008
1009 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) {
1010 /*
1011 * vdev_cant_read and vdev_cant_write can only
1012 * transition from TRUE to FALSE when we have the
1013 * SCL_ZIO lock as writer; otherwise they can only
1014 * transition from FALSE to TRUE. This ensures that
1015 * any zio looking at these values can assume that
1016 * failures persist for the life of the I/O. That's
1017 * important because when a device has intermittent
1018 * connectivity problems, we want to ensure that
1019 * they're ascribed to the device (ENXIO) and not
1020 * the zio (EIO).
1021 *
1022 * Since we hold SCL_ZIO as writer here, clear both
1023 * values so the probe can reevaluate from first
1024 * principles.
1025 */
1026 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER;
1027 vd->vdev_cant_read = B_FALSE;
1028 vd->vdev_cant_write = B_FALSE;
1029 }
1030
1031 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd,
1032 vdev_probe_done, vps,
1033 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE);
1034
1035 /*
1036 * We can't change the vdev state in this context, so we
1037 * kick off an async task to do it on our behalf.
1038 */
1039 if (zio != NULL) {
1040 vd->vdev_probe_wanted = B_TRUE;
1041 spa_async_request(spa, SPA_ASYNC_PROBE);
1042 }
1043 }
1044
1045 if (zio != NULL)
1046 zio_add_child(zio, pio);
1047
1048 mutex_exit(&vd->vdev_probe_lock);
1049
1050 if (vps == NULL) {
1051 ASSERT(zio != NULL);
1052 return (NULL);
1053 }
1054
1055 for (int l = 1; l < VDEV_LABELS; l++) {
1056 zio_nowait(zio_read_phys(pio, vd,
1057 vdev_label_offset(vd->vdev_psize, l,
1058 offsetof(vdev_label_t, vl_pad2)),
1059 VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE),
1060 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1061 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE));
1062 }
1063
1064 if (zio == NULL)
1065 return (pio);
1066
1067 zio_nowait(pio);
1068 return (NULL);
1069 }
1070
1071 static void
1072 vdev_open_child(void *arg)
1073 {
1074 vdev_t *vd = arg;
1075
1076 vd->vdev_open_thread = curthread;
1077 vd->vdev_open_error = vdev_open(vd);
1078 vd->vdev_open_thread = NULL;
1079 }
1080
1081 boolean_t
1082 vdev_uses_zvols(vdev_t *vd)
1083 {
1084 if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR,
1085 strlen(ZVOL_DIR)) == 0)
1086 return (B_TRUE);
1087 for (int c = 0; c < vd->vdev_children; c++)
1088 if (vdev_uses_zvols(vd->vdev_child[c]))
1089 return (B_TRUE);
1090 return (B_FALSE);
1091 }
1092
1093 void
1094 vdev_open_children(vdev_t *vd)
1095 {
1096 taskq_t *tq;
1097 int children = vd->vdev_children;
1098
1099 /*
1100 * in order to handle pools on top of zvols, do the opens
1101 * in a single thread so that the same thread holds the
1102 * spa_namespace_lock
1103 */
1104 if (vdev_uses_zvols(vd)) {
1105 for (int c = 0; c < children; c++)
1106 vd->vdev_child[c]->vdev_open_error =
1107 vdev_open(vd->vdev_child[c]);
1108 return;
1109 }
1110 tq = taskq_create("vdev_open", children, minclsyspri,
1111 children, children, TASKQ_PREPOPULATE);
1112
1113 for (int c = 0; c < children; c++)
1114 VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c],
1115 TQ_SLEEP) != NULL);
1116
1117 taskq_destroy(tq);
1118 }
1119
1120 /*
1121 * Prepare a virtual device for access.
1122 */
1123 int
1124 vdev_open(vdev_t *vd)
1125 {
1126 spa_t *spa = vd->vdev_spa;
1127 int error;
1128 uint64_t osize = 0;
1129 uint64_t max_osize = 0;
1130 uint64_t asize, max_asize, psize;
1131 uint64_t ashift = 0;
1132
1133 ASSERT(vd->vdev_open_thread == curthread ||
1134 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1135 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED ||
1136 vd->vdev_state == VDEV_STATE_CANT_OPEN ||
1137 vd->vdev_state == VDEV_STATE_OFFLINE);
1138
1139 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1140 vd->vdev_cant_read = B_FALSE;
1141 vd->vdev_cant_write = B_FALSE;
1142 vd->vdev_min_asize = vdev_get_min_asize(vd);
1143
1144 /*
1145 * If this vdev is not removed, check its fault status. If it's
1146 * faulted, bail out of the open.
1147 */
1148 if (!vd->vdev_removed && vd->vdev_faulted) {
1149 ASSERT(vd->vdev_children == 0);
1150 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1151 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1152 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1153 vd->vdev_label_aux);
1154 return (SET_ERROR(ENXIO));
1155 } else if (vd->vdev_offline) {
1156 ASSERT(vd->vdev_children == 0);
1157 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE);
1158 return (SET_ERROR(ENXIO));
1159 }
1160
1161 error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize, &ashift);
1162
1163 /*
1164 * Reset the vdev_reopening flag so that we actually close
1165 * the vdev on error.
1166 */
1167 vd->vdev_reopening = B_FALSE;
1168 if (zio_injection_enabled && error == 0)
1169 error = zio_handle_device_injection(vd, NULL, ENXIO);
1170
1171 if (error) {
1172 if (vd->vdev_removed &&
1173 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED)
1174 vd->vdev_removed = B_FALSE;
1175
1176 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1177 vd->vdev_stat.vs_aux);
1178 return (error);
1179 }
1180
1181 vd->vdev_removed = B_FALSE;
1182
1183 /*
1184 * Recheck the faulted flag now that we have confirmed that
1185 * the vdev is accessible. If we're faulted, bail.
1186 */
1187 if (vd->vdev_faulted) {
1188 ASSERT(vd->vdev_children == 0);
1189 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1190 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1191 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1192 vd->vdev_label_aux);
1193 return (SET_ERROR(ENXIO));
1194 }
1195
1196 if (vd->vdev_degraded) {
1197 ASSERT(vd->vdev_children == 0);
1198 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1199 VDEV_AUX_ERR_EXCEEDED);
1200 } else {
1201 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0);
1202 }
1203
1204 /*
1205 * For hole or missing vdevs we just return success.
1206 */
1207 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops)
1208 return (0);
1209
1210 for (int c = 0; c < vd->vdev_children; c++) {
1211 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) {
1212 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1213 VDEV_AUX_NONE);
1214 break;
1215 }
1216 }
1217
1218 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t));
1219 max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t));
1220
1221 if (vd->vdev_children == 0) {
1222 if (osize < SPA_MINDEVSIZE) {
1223 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1224 VDEV_AUX_TOO_SMALL);
1225 return (SET_ERROR(EOVERFLOW));
1226 }
1227 psize = osize;
1228 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE);
1229 max_asize = max_osize - (VDEV_LABEL_START_SIZE +
1230 VDEV_LABEL_END_SIZE);
1231 } else {
1232 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE -
1233 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) {
1234 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1235 VDEV_AUX_TOO_SMALL);
1236 return (SET_ERROR(EOVERFLOW));
1237 }
1238 psize = 0;
1239 asize = osize;
1240 max_asize = max_osize;
1241 }
1242
1243 vd->vdev_psize = psize;
1244
1245 /*
1246 * Make sure the allocatable size hasn't shrunk.
1247 */
1248 if (asize < vd->vdev_min_asize) {
1249 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1250 VDEV_AUX_BAD_LABEL);
1251 return (SET_ERROR(EINVAL));
1252 }
1253
1254 if (vd->vdev_asize == 0) {
1255 /*
1256 * This is the first-ever open, so use the computed values.
1257 * For testing purposes, a higher ashift can be requested.
1258 */
1259 vd->vdev_asize = asize;
1260 vd->vdev_max_asize = max_asize;
1261 vd->vdev_ashift = MAX(ashift, vd->vdev_ashift);
1262 } else {
1263 /*
1264 * Detect if the alignment requirement has increased.
1265 * We don't want to make the pool unavailable, just
1266 * issue a warning instead.
1267 */
1268 if (ashift > vd->vdev_top->vdev_ashift &&
1269 vd->vdev_ops->vdev_op_leaf) {
1270 cmn_err(CE_WARN,
1271 "Disk, '%s', has a block alignment that is "
1272 "larger than the pool's alignment\n",
1273 vd->vdev_path);
1274 }
1275 vd->vdev_max_asize = max_asize;
1276 }
1277
1278 /*
1279 * If all children are healthy and the asize has increased,
1280 * then we've experienced dynamic LUN growth. If automatic
1281 * expansion is enabled then use the additional space.
1282 */
1283 if (vd->vdev_state == VDEV_STATE_HEALTHY && asize > vd->vdev_asize &&
1284 (vd->vdev_expanding || spa->spa_autoexpand))
1285 vd->vdev_asize = asize;
1286
1287 vdev_set_min_asize(vd);
1288
1289 /*
1290 * Ensure we can issue some IO before declaring the
1291 * vdev open for business.
1292 */
1293 if (vd->vdev_ops->vdev_op_leaf &&
1294 (error = zio_wait(vdev_probe(vd, NULL))) != 0) {
1295 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1296 VDEV_AUX_ERR_EXCEEDED);
1297 return (error);
1298 }
1299
1300 /*
1301 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1302 * resilver. But don't do this if we are doing a reopen for a scrub,
1303 * since this would just restart the scrub we are already doing.
1304 */
1305 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen &&
1306 vdev_resilver_needed(vd, NULL, NULL))
1307 spa_async_request(spa, SPA_ASYNC_RESILVER);
1308
1309 return (0);
1310 }
1311
1312 /*
1313 * Called once the vdevs are all opened, this routine validates the label
1314 * contents. This needs to be done before vdev_load() so that we don't
1315 * inadvertently do repair I/Os to the wrong device.
1316 *
1317 * If 'strict' is false ignore the spa guid check. This is necessary because
1318 * if the machine crashed during a re-guid the new guid might have been written
1319 * to all of the vdev labels, but not the cached config. The strict check
1320 * will be performed when the pool is opened again using the mos config.
1321 *
1322 * This function will only return failure if one of the vdevs indicates that it
1323 * has since been destroyed or exported. This is only possible if
1324 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1325 * will be updated but the function will return 0.
1326 */
1327 int
1328 vdev_validate(vdev_t *vd, boolean_t strict)
1329 {
1330 spa_t *spa = vd->vdev_spa;
1331 nvlist_t *label;
1332 uint64_t guid = 0, top_guid;
1333 uint64_t state;
1334
1335 for (int c = 0; c < vd->vdev_children; c++)
1336 if (vdev_validate(vd->vdev_child[c], strict) != 0)
1337 return (SET_ERROR(EBADF));
1338
1339 /*
1340 * If the device has already failed, or was marked offline, don't do
1341 * any further validation. Otherwise, label I/O will fail and we will
1342 * overwrite the previous state.
1343 */
1344 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1345 uint64_t aux_guid = 0;
1346 nvlist_t *nvl;
1347 uint64_t txg = spa_last_synced_txg(spa) != 0 ?
1348 spa_last_synced_txg(spa) : -1ULL;
1349
1350 if ((label = vdev_label_read_config(vd, txg)) == NULL) {
1351 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1352 VDEV_AUX_BAD_LABEL);
1353 return (0);
1354 }
1355
1356 /*
1357 * Determine if this vdev has been split off into another
1358 * pool. If so, then refuse to open it.
1359 */
1360 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID,
1361 &aux_guid) == 0 && aux_guid == spa_guid(spa)) {
1362 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1363 VDEV_AUX_SPLIT_POOL);
1364 nvlist_free(label);
1365 return (0);
1366 }
1367
1368 if (strict && (nvlist_lookup_uint64(label,
1369 ZPOOL_CONFIG_POOL_GUID, &guid) != 0 ||
1370 guid != spa_guid(spa))) {
1371 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1372 VDEV_AUX_CORRUPT_DATA);
1373 nvlist_free(label);
1374 return (0);
1375 }
1376
1377 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl)
1378 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID,
1379 &aux_guid) != 0)
1380 aux_guid = 0;
1381
1382 /*
1383 * If this vdev just became a top-level vdev because its
1384 * sibling was detached, it will have adopted the parent's
1385 * vdev guid -- but the label may or may not be on disk yet.
1386 * Fortunately, either version of the label will have the
1387 * same top guid, so if we're a top-level vdev, we can
1388 * safely compare to that instead.
1389 *
1390 * If we split this vdev off instead, then we also check the
1391 * original pool's guid. We don't want to consider the vdev
1392 * corrupt if it is partway through a split operation.
1393 */
1394 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
1395 &guid) != 0 ||
1396 nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID,
1397 &top_guid) != 0 ||
1398 ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) &&
1399 (vd->vdev_guid != top_guid || vd != vd->vdev_top))) {
1400 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1401 VDEV_AUX_CORRUPT_DATA);
1402 nvlist_free(label);
1403 return (0);
1404 }
1405
1406 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1407 &state) != 0) {
1408 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1409 VDEV_AUX_CORRUPT_DATA);
1410 nvlist_free(label);
1411 return (0);
1412 }
1413
1414 nvlist_free(label);
1415
1416 /*
1417 * If this is a verbatim import, no need to check the
1418 * state of the pool.
1419 */
1420 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) &&
1421 spa_load_state(spa) == SPA_LOAD_OPEN &&
1422 state != POOL_STATE_ACTIVE)
1423 return (SET_ERROR(EBADF));
1424
1425 /*
1426 * If we were able to open and validate a vdev that was
1427 * previously marked permanently unavailable, clear that state
1428 * now.
1429 */
1430 if (vd->vdev_not_present)
1431 vd->vdev_not_present = 0;
1432 }
1433
1434 return (0);
1435 }
1436
1437 /*
1438 * Close a virtual device.
1439 */
1440 void
1441 vdev_close(vdev_t *vd)
1442 {
1443 spa_t *spa = vd->vdev_spa;
1444 vdev_t *pvd = vd->vdev_parent;
1445
1446 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1447
1448 /*
1449 * If our parent is reopening, then we are as well, unless we are
1450 * going offline.
1451 */
1452 if (pvd != NULL && pvd->vdev_reopening)
1453 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline);
1454
1455 vd->vdev_ops->vdev_op_close(vd);
1456
1457 vdev_cache_purge(vd);
1458
1459 /*
1460 * We record the previous state before we close it, so that if we are
1461 * doing a reopen(), we don't generate FMA ereports if we notice that
1462 * it's still faulted.
1463 */
1464 vd->vdev_prevstate = vd->vdev_state;
1465
1466 if (vd->vdev_offline)
1467 vd->vdev_state = VDEV_STATE_OFFLINE;
1468 else
1469 vd->vdev_state = VDEV_STATE_CLOSED;
1470 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1471 }
1472
1473 void
1474 vdev_hold(vdev_t *vd)
1475 {
1476 spa_t *spa = vd->vdev_spa;
1477
1478 ASSERT(spa_is_root(spa));
1479 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1480 return;
1481
1482 for (int c = 0; c < vd->vdev_children; c++)
1483 vdev_hold(vd->vdev_child[c]);
1484
1485 if (vd->vdev_ops->vdev_op_leaf)
1486 vd->vdev_ops->vdev_op_hold(vd);
1487 }
1488
1489 void
1490 vdev_rele(vdev_t *vd)
1491 {
1492 spa_t *spa = vd->vdev_spa;
1493
1494 ASSERT(spa_is_root(spa));
1495 for (int c = 0; c < vd->vdev_children; c++)
1496 vdev_rele(vd->vdev_child[c]);
1497
1498 if (vd->vdev_ops->vdev_op_leaf)
1499 vd->vdev_ops->vdev_op_rele(vd);
1500 }
1501
1502 /*
1503 * Reopen all interior vdevs and any unopened leaves. We don't actually
1504 * reopen leaf vdevs which had previously been opened as they might deadlock
1505 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1506 * If the leaf has never been opened then open it, as usual.
1507 */
1508 void
1509 vdev_reopen(vdev_t *vd)
1510 {
1511 spa_t *spa = vd->vdev_spa;
1512
1513 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1514
1515 /* set the reopening flag unless we're taking the vdev offline */
1516 vd->vdev_reopening = !vd->vdev_offline;
1517 vdev_close(vd);
1518 (void) vdev_open(vd);
1519
1520 /*
1521 * Call vdev_validate() here to make sure we have the same device.
1522 * Otherwise, a device with an invalid label could be successfully
1523 * opened in response to vdev_reopen().
1524 */
1525 if (vd->vdev_aux) {
1526 (void) vdev_validate_aux(vd);
1527 if (vdev_readable(vd) && vdev_writeable(vd) &&
1528 vd->vdev_aux == &spa->spa_l2cache &&
1529 !l2arc_vdev_present(vd))
1530 l2arc_add_vdev(spa, vd);
1531 } else {
1532 (void) vdev_validate(vd, B_TRUE);
1533 }
1534
1535 /*
1536 * Reassess parent vdev's health.
1537 */
1538 vdev_propagate_state(vd);
1539 }
1540
1541 int
1542 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
1543 {
1544 int error;
1545
1546 /*
1547 * Normally, partial opens (e.g. of a mirror) are allowed.
1548 * For a create, however, we want to fail the request if
1549 * there are any components we can't open.
1550 */
1551 error = vdev_open(vd);
1552
1553 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) {
1554 vdev_close(vd);
1555 return (error ? error : ENXIO);
1556 }
1557
1558 /*
1559 * Recursively load DTLs and initialize all labels.
1560 */
1561 if ((error = vdev_dtl_load(vd)) != 0 ||
1562 (error = vdev_label_init(vd, txg, isreplacing ?
1563 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
1564 vdev_close(vd);
1565 return (error);
1566 }
1567
1568 return (0);
1569 }
1570
1571 void
1572 vdev_metaslab_set_size(vdev_t *vd)
1573 {
1574 /*
1575 * Aim for roughly 200 metaslabs per vdev.
1576 */
1577 vd->vdev_ms_shift = highbit64(vd->vdev_asize / 200);
1578 vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT);
1579 }
1580
1581 void
1582 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
1583 {
1584 ASSERT(vd == vd->vdev_top);
1585 ASSERT(!vd->vdev_ishole);
1586 ASSERT(ISP2(flags));
1587 ASSERT(spa_writeable(vd->vdev_spa));
1588
1589 if (flags & VDD_METASLAB)
1590 (void) txg_list_add(&vd->vdev_ms_list, arg, txg);
1591
1592 if (flags & VDD_DTL)
1593 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg);
1594
1595 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
1596 }
1597
1598 void
1599 vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg)
1600 {
1601 for (int c = 0; c < vd->vdev_children; c++)
1602 vdev_dirty_leaves(vd->vdev_child[c], flags, txg);
1603
1604 if (vd->vdev_ops->vdev_op_leaf)
1605 vdev_dirty(vd->vdev_top, flags, vd, txg);
1606 }
1607
1608 /*
1609 * DTLs.
1610 *
1611 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1612 * the vdev has less than perfect replication. There are four kinds of DTL:
1613 *
1614 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1615 *
1616 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1617 *
1618 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1619 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1620 * txgs that was scrubbed.
1621 *
1622 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1623 * persistent errors or just some device being offline.
1624 * Unlike the other three, the DTL_OUTAGE map is not generally
1625 * maintained; it's only computed when needed, typically to
1626 * determine whether a device can be detached.
1627 *
1628 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1629 * either has the data or it doesn't.
1630 *
1631 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1632 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1633 * if any child is less than fully replicated, then so is its parent.
1634 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1635 * comprising only those txgs which appear in 'maxfaults' or more children;
1636 * those are the txgs we don't have enough replication to read. For example,
1637 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1638 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1639 * two child DTL_MISSING maps.
1640 *
1641 * It should be clear from the above that to compute the DTLs and outage maps
1642 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1643 * Therefore, that is all we keep on disk. When loading the pool, or after
1644 * a configuration change, we generate all other DTLs from first principles.
1645 */
1646 void
1647 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1648 {
1649 range_tree_t *rt = vd->vdev_dtl[t];
1650
1651 ASSERT(t < DTL_TYPES);
1652 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1653 ASSERT(spa_writeable(vd->vdev_spa));
1654
1655 mutex_enter(rt->rt_lock);
1656 if (!range_tree_contains(rt, txg, size))
1657 range_tree_add(rt, txg, size);
1658 mutex_exit(rt->rt_lock);
1659 }
1660
1661 boolean_t
1662 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1663 {
1664 range_tree_t *rt = vd->vdev_dtl[t];
1665 boolean_t dirty = B_FALSE;
1666
1667 ASSERT(t < DTL_TYPES);
1668 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1669
1670 mutex_enter(rt->rt_lock);
1671 if (range_tree_space(rt) != 0)
1672 dirty = range_tree_contains(rt, txg, size);
1673 mutex_exit(rt->rt_lock);
1674
1675 return (dirty);
1676 }
1677
1678 boolean_t
1679 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t)
1680 {
1681 range_tree_t *rt = vd->vdev_dtl[t];
1682 boolean_t empty;
1683
1684 mutex_enter(rt->rt_lock);
1685 empty = (range_tree_space(rt) == 0);
1686 mutex_exit(rt->rt_lock);
1687
1688 return (empty);
1689 }
1690
1691 /*
1692 * Returns the lowest txg in the DTL range.
1693 */
1694 static uint64_t
1695 vdev_dtl_min(vdev_t *vd)
1696 {
1697 range_seg_t *rs;
1698
1699 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
1700 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
1701 ASSERT0(vd->vdev_children);
1702
1703 rs = avl_first(&vd->vdev_dtl[DTL_MISSING]->rt_root);
1704 return (rs->rs_start - 1);
1705 }
1706
1707 /*
1708 * Returns the highest txg in the DTL.
1709 */
1710 static uint64_t
1711 vdev_dtl_max(vdev_t *vd)
1712 {
1713 range_seg_t *rs;
1714
1715 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
1716 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
1717 ASSERT0(vd->vdev_children);
1718
1719 rs = avl_last(&vd->vdev_dtl[DTL_MISSING]->rt_root);
1720 return (rs->rs_end);
1721 }
1722
1723 /*
1724 * Determine if a resilvering vdev should remove any DTL entries from
1725 * its range. If the vdev was resilvering for the entire duration of the
1726 * scan then it should excise that range from its DTLs. Otherwise, this
1727 * vdev is considered partially resilvered and should leave its DTL
1728 * entries intact. The comment in vdev_dtl_reassess() describes how we
1729 * excise the DTLs.
1730 */
1731 static boolean_t
1732 vdev_dtl_should_excise(vdev_t *vd)
1733 {
1734 spa_t *spa = vd->vdev_spa;
1735 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
1736
1737 ASSERT0(scn->scn_phys.scn_errors);
1738 ASSERT0(vd->vdev_children);
1739
1740 if (vd->vdev_resilver_txg == 0 ||
1741 range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0)
1742 return (B_TRUE);
1743
1744 /*
1745 * When a resilver is initiated the scan will assign the scn_max_txg
1746 * value to the highest txg value that exists in all DTLs. If this
1747 * device's max DTL is not part of this scan (i.e. it is not in
1748 * the range (scn_min_txg, scn_max_txg] then it is not eligible
1749 * for excision.
1750 */
1751 if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) {
1752 ASSERT3U(scn->scn_phys.scn_min_txg, <=, vdev_dtl_min(vd));
1753 ASSERT3U(scn->scn_phys.scn_min_txg, <, vd->vdev_resilver_txg);
1754 ASSERT3U(vd->vdev_resilver_txg, <=, scn->scn_phys.scn_max_txg);
1755 return (B_TRUE);
1756 }
1757 return (B_FALSE);
1758 }
1759
1760 /*
1761 * Reassess DTLs after a config change or scrub completion.
1762 */
1763 void
1764 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
1765 {
1766 spa_t *spa = vd->vdev_spa;
1767 avl_tree_t reftree;
1768 int minref;
1769
1770 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1771
1772 for (int c = 0; c < vd->vdev_children; c++)
1773 vdev_dtl_reassess(vd->vdev_child[c], txg,
1774 scrub_txg, scrub_done);
1775
1776 if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux)
1777 return;
1778
1779 if (vd->vdev_ops->vdev_op_leaf) {
1780 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
1781
1782 mutex_enter(&vd->vdev_dtl_lock);
1783
1784 /*
1785 * If we've completed a scan cleanly then determine
1786 * if this vdev should remove any DTLs. We only want to
1787 * excise regions on vdevs that were available during
1788 * the entire duration of this scan.
1789 */
1790 if (scrub_txg != 0 &&
1791 (spa->spa_scrub_started ||
1792 (scn != NULL && scn->scn_phys.scn_errors == 0)) &&
1793 vdev_dtl_should_excise(vd)) {
1794 /*
1795 * We completed a scrub up to scrub_txg. If we
1796 * did it without rebooting, then the scrub dtl
1797 * will be valid, so excise the old region and
1798 * fold in the scrub dtl. Otherwise, leave the
1799 * dtl as-is if there was an error.
1800 *
1801 * There's little trick here: to excise the beginning
1802 * of the DTL_MISSING map, we put it into a reference
1803 * tree and then add a segment with refcnt -1 that
1804 * covers the range [0, scrub_txg). This means
1805 * that each txg in that range has refcnt -1 or 0.
1806 * We then add DTL_SCRUB with a refcnt of 2, so that
1807 * entries in the range [0, scrub_txg) will have a
1808 * positive refcnt -- either 1 or 2. We then convert
1809 * the reference tree into the new DTL_MISSING map.
1810 */
1811 space_reftree_create(&reftree);
1812 space_reftree_add_map(&reftree,
1813 vd->vdev_dtl[DTL_MISSING], 1);
1814 space_reftree_add_seg(&reftree, 0, scrub_txg, -1);
1815 space_reftree_add_map(&reftree,
1816 vd->vdev_dtl[DTL_SCRUB], 2);
1817 space_reftree_generate_map(&reftree,
1818 vd->vdev_dtl[DTL_MISSING], 1);
1819 space_reftree_destroy(&reftree);
1820 }
1821 range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
1822 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
1823 range_tree_add, vd->vdev_dtl[DTL_PARTIAL]);
1824 if (scrub_done)
1825 range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
1826 range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
1827 if (!vdev_readable(vd))
1828 range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
1829 else
1830 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
1831 range_tree_add, vd->vdev_dtl[DTL_OUTAGE]);
1832
1833 /*
1834 * If the vdev was resilvering and no longer has any
1835 * DTLs then reset its resilvering flag.
1836 */
1837 if (vd->vdev_resilver_txg != 0 &&
1838 range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0 &&
1839 range_tree_space(vd->vdev_dtl[DTL_OUTAGE]) == 0)
1840 vd->vdev_resilver_txg = 0;
1841
1842 mutex_exit(&vd->vdev_dtl_lock);
1843
1844 if (txg != 0)
1845 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
1846 return;
1847 }
1848
1849 mutex_enter(&vd->vdev_dtl_lock);
1850 for (int t = 0; t < DTL_TYPES; t++) {
1851 /* account for child's outage in parent's missing map */
1852 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t;
1853 if (t == DTL_SCRUB)
1854 continue; /* leaf vdevs only */
1855 if (t == DTL_PARTIAL)
1856 minref = 1; /* i.e. non-zero */
1857 else if (vd->vdev_nparity != 0)
1858 minref = vd->vdev_nparity + 1; /* RAID-Z */
1859 else
1860 minref = vd->vdev_children; /* any kind of mirror */
1861 space_reftree_create(&reftree);
1862 for (int c = 0; c < vd->vdev_children; c++) {
1863 vdev_t *cvd = vd->vdev_child[c];
1864 mutex_enter(&cvd->vdev_dtl_lock);
1865 space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1);
1866 mutex_exit(&cvd->vdev_dtl_lock);
1867 }
1868 space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref);
1869 space_reftree_destroy(&reftree);
1870 }
1871 mutex_exit(&vd->vdev_dtl_lock);
1872 }
1873
1874 int
1875 vdev_dtl_load(vdev_t *vd)
1876 {
1877 spa_t *spa = vd->vdev_spa;
1878 objset_t *mos = spa->spa_meta_objset;
1879 int error = 0;
1880
1881 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) {
1882 ASSERT(!vd->vdev_ishole);
1883
1884 error = space_map_open(&vd->vdev_dtl_sm, mos,
1885 vd->vdev_dtl_object, 0, -1ULL, 0, &vd->vdev_dtl_lock);
1886 if (error)
1887 return (error);
1888 ASSERT(vd->vdev_dtl_sm != NULL);
1889
1890 mutex_enter(&vd->vdev_dtl_lock);
1891
1892 /*
1893 * Now that we've opened the space_map we need to update
1894 * the in-core DTL.
1895 */
1896 space_map_update(vd->vdev_dtl_sm);
1897
1898 error = space_map_load(vd->vdev_dtl_sm,
1899 vd->vdev_dtl[DTL_MISSING], SM_ALLOC);
1900 mutex_exit(&vd->vdev_dtl_lock);
1901
1902 return (error);
1903 }
1904
1905 for (int c = 0; c < vd->vdev_children; c++) {
1906 error = vdev_dtl_load(vd->vdev_child[c]);
1907 if (error != 0)
1908 break;
1909 }
1910
1911 return (error);
1912 }
1913
1914 void
1915 vdev_dtl_sync(vdev_t *vd, uint64_t txg)
1916 {
1917 spa_t *spa = vd->vdev_spa;
1918 range_tree_t *rt = vd->vdev_dtl[DTL_MISSING];
1919 objset_t *mos = spa->spa_meta_objset;
1920 range_tree_t *rtsync;
1921 kmutex_t rtlock;
1922 dmu_tx_t *tx;
1923 uint64_t object = space_map_object(vd->vdev_dtl_sm);
1924
1925 ASSERT(!vd->vdev_ishole);
1926 ASSERT(vd->vdev_ops->vdev_op_leaf);
1927
1928 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
1929
1930 if (vd->vdev_detached || vd->vdev_top->vdev_removing) {
1931 mutex_enter(&vd->vdev_dtl_lock);
1932 space_map_free(vd->vdev_dtl_sm, tx);
1933 space_map_close(vd->vdev_dtl_sm);
1934 vd->vdev_dtl_sm = NULL;
1935 mutex_exit(&vd->vdev_dtl_lock);
1936 dmu_tx_commit(tx);
1937 return;
1938 }
1939
1940 if (vd->vdev_dtl_sm == NULL) {
1941 uint64_t new_object;
1942
1943 new_object = space_map_alloc(mos, tx);
1944 VERIFY3U(new_object, !=, 0);
1945
1946 VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object,
1947 0, -1ULL, 0, &vd->vdev_dtl_lock));
1948 ASSERT(vd->vdev_dtl_sm != NULL);
1949 }
1950
1951 mutex_init(&rtlock, NULL, MUTEX_DEFAULT, NULL);
1952
1953 rtsync = range_tree_create(NULL, NULL, &rtlock);
1954
1955 mutex_enter(&rtlock);
1956
1957 mutex_enter(&vd->vdev_dtl_lock);
1958 range_tree_walk(rt, range_tree_add, rtsync);
1959 mutex_exit(&vd->vdev_dtl_lock);
1960
1961 space_map_truncate(vd->vdev_dtl_sm, tx);
1962 space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, tx);
1963 range_tree_vacate(rtsync, NULL, NULL);
1964
1965 range_tree_destroy(rtsync);
1966
1967 mutex_exit(&rtlock);
1968 mutex_destroy(&rtlock);
1969
1970 /*
1971 * If the object for the space map has changed then dirty
1972 * the top level so that we update the config.
1973 */
1974 if (object != space_map_object(vd->vdev_dtl_sm)) {
1975 zfs_dbgmsg("txg %llu, spa %s, DTL old object %llu, "
1976 "new object %llu", txg, spa_name(spa), object,
1977 space_map_object(vd->vdev_dtl_sm));
1978 vdev_config_dirty(vd->vdev_top);
1979 }
1980
1981 dmu_tx_commit(tx);
1982
1983 mutex_enter(&vd->vdev_dtl_lock);
1984 space_map_update(vd->vdev_dtl_sm);
1985 mutex_exit(&vd->vdev_dtl_lock);
1986 }
1987
1988 /*
1989 * Determine whether the specified vdev can be offlined/detached/removed
1990 * without losing data.
1991 */
1992 boolean_t
1993 vdev_dtl_required(vdev_t *vd)
1994 {
1995 spa_t *spa = vd->vdev_spa;
1996 vdev_t *tvd = vd->vdev_top;
1997 uint8_t cant_read = vd->vdev_cant_read;
1998 boolean_t required;
1999
2000 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2001
2002 if (vd == spa->spa_root_vdev || vd == tvd)
2003 return (B_TRUE);
2004
2005 /*
2006 * Temporarily mark the device as unreadable, and then determine
2007 * whether this results in any DTL outages in the top-level vdev.
2008 * If not, we can safely offline/detach/remove the device.
2009 */
2010 vd->vdev_cant_read = B_TRUE;
2011 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
2012 required = !vdev_dtl_empty(tvd, DTL_OUTAGE);
2013 vd->vdev_cant_read = cant_read;
2014 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
2015
2016 if (!required && zio_injection_enabled)
2017 required = !!zio_handle_device_injection(vd, NULL, ECHILD);
2018
2019 return (required);
2020 }
2021
2022 /*
2023 * Determine if resilver is needed, and if so the txg range.
2024 */
2025 boolean_t
2026 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
2027 {
2028 boolean_t needed = B_FALSE;
2029 uint64_t thismin = UINT64_MAX;
2030 uint64_t thismax = 0;
2031
2032 if (vd->vdev_children == 0) {
2033 mutex_enter(&vd->vdev_dtl_lock);
2034 if (range_tree_space(vd->vdev_dtl[DTL_MISSING]) != 0 &&
2035 vdev_writeable(vd)) {
2036
2037 thismin = vdev_dtl_min(vd);
2038 thismax = vdev_dtl_max(vd);
2039 needed = B_TRUE;
2040 }
2041 mutex_exit(&vd->vdev_dtl_lock);
2042 } else {
2043 for (int c = 0; c < vd->vdev_children; c++) {
2044 vdev_t *cvd = vd->vdev_child[c];
2045 uint64_t cmin, cmax;
2046
2047 if (vdev_resilver_needed(cvd, &cmin, &cmax)) {
2048 thismin = MIN(thismin, cmin);
2049 thismax = MAX(thismax, cmax);
2050 needed = B_TRUE;
2051 }
2052 }
2053 }
2054
2055 if (needed && minp) {
2056 *minp = thismin;
2057 *maxp = thismax;
2058 }
2059 return (needed);
2060 }
2061
2062 void
2063 vdev_load(vdev_t *vd)
2064 {
2065 /*
2066 * Recursively load all children.
2067 */
2068 for (int c = 0; c < vd->vdev_children; c++)
2069 vdev_load(vd->vdev_child[c]);
2070
2071 /*
2072 * If this is a top-level vdev, initialize its metaslabs.
2073 */
2074 if (vd == vd->vdev_top && !vd->vdev_ishole &&
2075 (vd->vdev_ashift == 0 || vd->vdev_asize == 0 ||
2076 vdev_metaslab_init(vd, 0) != 0))
2077 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2078 VDEV_AUX_CORRUPT_DATA);
2079
2080 /*
2081 * If this is a leaf vdev, load its DTL.
2082 */
2083 if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0)
2084 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2085 VDEV_AUX_CORRUPT_DATA);
2086 }
2087
2088 /*
2089 * The special vdev case is used for hot spares and l2cache devices. Its
2090 * sole purpose it to set the vdev state for the associated vdev. To do this,
2091 * we make sure that we can open the underlying device, then try to read the
2092 * label, and make sure that the label is sane and that it hasn't been
2093 * repurposed to another pool.
2094 */
2095 int
2096 vdev_validate_aux(vdev_t *vd)
2097 {
2098 nvlist_t *label;
2099 uint64_t guid, version;
2100 uint64_t state;
2101
2102 if (!vdev_readable(vd))
2103 return (0);
2104
2105 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) {
2106 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2107 VDEV_AUX_CORRUPT_DATA);
2108 return (-1);
2109 }
2110
2111 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 ||
2112 !SPA_VERSION_IS_SUPPORTED(version) ||
2113 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 ||
2114 guid != vd->vdev_guid ||
2115 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) {
2116 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2117 VDEV_AUX_CORRUPT_DATA);
2118 nvlist_free(label);
2119 return (-1);
2120 }
2121
2122 /*
2123 * We don't actually check the pool state here. If it's in fact in
2124 * use by another pool, we update this fact on the fly when requested.
2125 */
2126 nvlist_free(label);
2127 return (0);
2128 }
2129
2130 void
2131 vdev_remove(vdev_t *vd, uint64_t txg)
2132 {
2133 spa_t *spa = vd->vdev_spa;
2134 objset_t *mos = spa->spa_meta_objset;
2135 dmu_tx_t *tx;
2136
2137 tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
2138
2139 if (vd->vdev_ms != NULL) {
2140 for (int m = 0; m < vd->vdev_ms_count; m++) {
2141 metaslab_t *msp = vd->vdev_ms[m];
2142
2143 if (msp == NULL || msp->ms_sm == NULL)
2144 continue;
2145
2146 mutex_enter(&msp->ms_lock);
2147 VERIFY0(space_map_allocated(msp->ms_sm));
2148 space_map_free(msp->ms_sm, tx);
2149 space_map_close(msp->ms_sm);
2150 msp->ms_sm = NULL;
2151 mutex_exit(&msp->ms_lock);
2152 }
2153 }
2154
2155 if (vd->vdev_ms_array) {
2156 (void) dmu_object_free(mos, vd->vdev_ms_array, tx);
2157 vd->vdev_ms_array = 0;
2158 }
2159 dmu_tx_commit(tx);
2160 }
2161
2162 void
2163 vdev_sync_done(vdev_t *vd, uint64_t txg)
2164 {
2165 metaslab_t *msp;
2166 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg));
2167
2168 ASSERT(!vd->vdev_ishole);
2169
2170 while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))
2171 metaslab_sync_done(msp, txg);
2172
2173 if (reassess)
2174 metaslab_sync_reassess(vd->vdev_mg);
2175 }
2176
2177 void
2178 vdev_sync(vdev_t *vd, uint64_t txg)
2179 {
2180 spa_t *spa = vd->vdev_spa;
2181 vdev_t *lvd;
2182 metaslab_t *msp;
2183 dmu_tx_t *tx;
2184
2185 ASSERT(!vd->vdev_ishole);
2186
2187 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) {
2188 ASSERT(vd == vd->vdev_top);
2189 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2190 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset,
2191 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx);
2192 ASSERT(vd->vdev_ms_array != 0);
2193 vdev_config_dirty(vd);
2194 dmu_tx_commit(tx);
2195 }
2196
2197 /*
2198 * Remove the metadata associated with this vdev once it's empty.
2199 */
2200 if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing)
2201 vdev_remove(vd, txg);
2202
2203 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) {
2204 metaslab_sync(msp, txg);
2205 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg));
2206 }
2207
2208 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL)
2209 vdev_dtl_sync(lvd, txg);
2210
2211 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg));
2212 }
2213
2214 uint64_t
2215 vdev_psize_to_asize(vdev_t *vd, uint64_t psize)
2216 {
2217 return (vd->vdev_ops->vdev_op_asize(vd, psize));
2218 }
2219
2220 /*
2221 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2222 * not be opened, and no I/O is attempted.
2223 */
2224 int
2225 vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2226 {
2227 vdev_t *vd, *tvd;
2228
2229 spa_vdev_state_enter(spa, SCL_NONE);
2230
2231 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2232 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2233
2234 if (!vd->vdev_ops->vdev_op_leaf)
2235 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2236
2237 tvd = vd->vdev_top;
2238
2239 /*
2240 * We don't directly use the aux state here, but if we do a
2241 * vdev_reopen(), we need this value to be present to remember why we
2242 * were faulted.
2243 */
2244 vd->vdev_label_aux = aux;
2245
2246 /*
2247 * Faulted state takes precedence over degraded.
2248 */
2249 vd->vdev_delayed_close = B_FALSE;
2250 vd->vdev_faulted = 1ULL;
2251 vd->vdev_degraded = 0ULL;
2252 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux);
2253
2254 /*
2255 * If this device has the only valid copy of the data, then
2256 * back off and simply mark the vdev as degraded instead.
2257 */
2258 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) {
2259 vd->vdev_degraded = 1ULL;
2260 vd->vdev_faulted = 0ULL;
2261
2262 /*
2263 * If we reopen the device and it's not dead, only then do we
2264 * mark it degraded.
2265 */
2266 vdev_reopen(tvd);
2267
2268 if (vdev_readable(vd))
2269 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux);
2270 }
2271
2272 return (spa_vdev_state_exit(spa, vd, 0));
2273 }
2274
2275 /*
2276 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2277 * user that something is wrong. The vdev continues to operate as normal as far
2278 * as I/O is concerned.
2279 */
2280 int
2281 vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2282 {
2283 vdev_t *vd;
2284
2285 spa_vdev_state_enter(spa, SCL_NONE);
2286
2287 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2288 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2289
2290 if (!vd->vdev_ops->vdev_op_leaf)
2291 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2292
2293 /*
2294 * If the vdev is already faulted, then don't do anything.
2295 */
2296 if (vd->vdev_faulted || vd->vdev_degraded)
2297 return (spa_vdev_state_exit(spa, NULL, 0));
2298
2299 vd->vdev_degraded = 1ULL;
2300 if (!vdev_is_dead(vd))
2301 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
2302 aux);
2303
2304 return (spa_vdev_state_exit(spa, vd, 0));
2305 }
2306
2307 /*
2308 * Online the given vdev.
2309 *
2310 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
2311 * spare device should be detached when the device finishes resilvering.
2312 * Second, the online should be treated like a 'test' online case, so no FMA
2313 * events are generated if the device fails to open.
2314 */
2315 int
2316 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
2317 {
2318 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev;
2319
2320 spa_vdev_state_enter(spa, SCL_NONE);
2321
2322 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2323 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2324
2325 if (!vd->vdev_ops->vdev_op_leaf)
2326 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2327
2328 tvd = vd->vdev_top;
2329 vd->vdev_offline = B_FALSE;
2330 vd->vdev_tmpoffline = B_FALSE;
2331 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE);
2332 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT);
2333
2334 /* XXX - L2ARC 1.0 does not support expansion */
2335 if (!vd->vdev_aux) {
2336 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2337 pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND);
2338 }
2339
2340 vdev_reopen(tvd);
2341 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE;
2342
2343 if (!vd->vdev_aux) {
2344 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2345 pvd->vdev_expanding = B_FALSE;
2346 }
2347
2348 if (newstate)
2349 *newstate = vd->vdev_state;
2350 if ((flags & ZFS_ONLINE_UNSPARE) &&
2351 !vdev_is_dead(vd) && vd->vdev_parent &&
2352 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2353 vd->vdev_parent->vdev_child[0] == vd)
2354 vd->vdev_unspare = B_TRUE;
2355
2356 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) {
2357
2358 /* XXX - L2ARC 1.0 does not support expansion */
2359 if (vd->vdev_aux)
2360 return (spa_vdev_state_exit(spa, vd, ENOTSUP));
2361 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2362 }
2363 return (spa_vdev_state_exit(spa, vd, 0));
2364 }
2365
2366 static int
2367 vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags)
2368 {
2369 vdev_t *vd, *tvd;
2370 int error = 0;
2371 uint64_t generation;
2372 metaslab_group_t *mg;
2373
2374 top:
2375 spa_vdev_state_enter(spa, SCL_ALLOC);
2376
2377 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2378 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2379
2380 if (!vd->vdev_ops->vdev_op_leaf)
2381 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2382
2383 tvd = vd->vdev_top;
2384 mg = tvd->vdev_mg;
2385 generation = spa->spa_config_generation + 1;
2386
2387 /*
2388 * If the device isn't already offline, try to offline it.
2389 */
2390 if (!vd->vdev_offline) {
2391 /*
2392 * If this device has the only valid copy of some data,
2393 * don't allow it to be offlined. Log devices are always
2394 * expendable.
2395 */
2396 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2397 vdev_dtl_required(vd))
2398 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2399
2400 /*
2401 * If the top-level is a slog and it has had allocations
2402 * then proceed. We check that the vdev's metaslab group
2403 * is not NULL since it's possible that we may have just
2404 * added this vdev but not yet initialized its metaslabs.
2405 */
2406 if (tvd->vdev_islog && mg != NULL) {
2407 /*
2408 * Prevent any future allocations.
2409 */
2410 metaslab_group_passivate(mg);
2411 (void) spa_vdev_state_exit(spa, vd, 0);
2412
2413 error = spa_offline_log(spa);
2414
2415 spa_vdev_state_enter(spa, SCL_ALLOC);
2416
2417 /*
2418 * Check to see if the config has changed.
2419 */
2420 if (error || generation != spa->spa_config_generation) {
2421 metaslab_group_activate(mg);
2422 if (error)
2423 return (spa_vdev_state_exit(spa,
2424 vd, error));
2425 (void) spa_vdev_state_exit(spa, vd, 0);
2426 goto top;
2427 }
2428 ASSERT0(tvd->vdev_stat.vs_alloc);
2429 }
2430
2431 /*
2432 * Offline this device and reopen its top-level vdev.
2433 * If the top-level vdev is a log device then just offline
2434 * it. Otherwise, if this action results in the top-level
2435 * vdev becoming unusable, undo it and fail the request.
2436 */
2437 vd->vdev_offline = B_TRUE;
2438 vdev_reopen(tvd);
2439
2440 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2441 vdev_is_dead(tvd)) {
2442 vd->vdev_offline = B_FALSE;
2443 vdev_reopen(tvd);
2444 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2445 }
2446
2447 /*
2448 * Add the device back into the metaslab rotor so that
2449 * once we online the device it's open for business.
2450 */
2451 if (tvd->vdev_islog && mg != NULL)
2452 metaslab_group_activate(mg);
2453 }
2454
2455 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY);
2456
2457 return (spa_vdev_state_exit(spa, vd, 0));
2458 }
2459
2460 int
2461 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags)
2462 {
2463 int error;
2464
2465 mutex_enter(&spa->spa_vdev_top_lock);
2466 error = vdev_offline_locked(spa, guid, flags);
2467 mutex_exit(&spa->spa_vdev_top_lock);
2468
2469 return (error);
2470 }
2471
2472 /*
2473 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2474 * vdev_offline(), we assume the spa config is locked. We also clear all
2475 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2476 */
2477 void
2478 vdev_clear(spa_t *spa, vdev_t *vd)
2479 {
2480 vdev_t *rvd = spa->spa_root_vdev;
2481
2482 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2483
2484 if (vd == NULL)
2485 vd = rvd;
2486
2487 vd->vdev_stat.vs_read_errors = 0;
2488 vd->vdev_stat.vs_write_errors = 0;
2489 vd->vdev_stat.vs_checksum_errors = 0;
2490
2491 for (int c = 0; c < vd->vdev_children; c++)
2492 vdev_clear(spa, vd->vdev_child[c]);
2493
2494 /*
2495 * If we're in the FAULTED state or have experienced failed I/O, then
2496 * clear the persistent state and attempt to reopen the device. We
2497 * also mark the vdev config dirty, so that the new faulted state is
2498 * written out to disk.
2499 */
2500 if (vd->vdev_faulted || vd->vdev_degraded ||
2501 !vdev_readable(vd) || !vdev_writeable(vd)) {
2502
2503 /*
2504 * When reopening in reponse to a clear event, it may be due to
2505 * a fmadm repair request. In this case, if the device is
2506 * still broken, we want to still post the ereport again.
2507 */
2508 vd->vdev_forcefault = B_TRUE;
2509
2510 vd->vdev_faulted = vd->vdev_degraded = 0ULL;
2511 vd->vdev_cant_read = B_FALSE;
2512 vd->vdev_cant_write = B_FALSE;
2513
2514 vdev_reopen(vd == rvd ? rvd : vd->vdev_top);
2515
2516 vd->vdev_forcefault = B_FALSE;
2517
2518 if (vd != rvd && vdev_writeable(vd->vdev_top))
2519 vdev_state_dirty(vd->vdev_top);
2520
2521 if (vd->vdev_aux == NULL && !vdev_is_dead(vd))
2522 spa_async_request(spa, SPA_ASYNC_RESILVER);
2523
2524 spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR);
2525 }
2526
2527 /*
2528 * When clearing a FMA-diagnosed fault, we always want to
2529 * unspare the device, as we assume that the original spare was
2530 * done in response to the FMA fault.
2531 */
2532 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL &&
2533 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2534 vd->vdev_parent->vdev_child[0] == vd)
2535 vd->vdev_unspare = B_TRUE;
2536 }
2537
2538 boolean_t
2539 vdev_is_dead(vdev_t *vd)
2540 {
2541 /*
2542 * Holes and missing devices are always considered "dead".
2543 * This simplifies the code since we don't have to check for
2544 * these types of devices in the various code paths.
2545 * Instead we rely on the fact that we skip over dead devices
2546 * before issuing I/O to them.
2547 */
2548 return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole ||
2549 vd->vdev_ops == &vdev_missing_ops);
2550 }
2551
2552 boolean_t
2553 vdev_readable(vdev_t *vd)
2554 {
2555 return (!vdev_is_dead(vd) && !vd->vdev_cant_read);
2556 }
2557
2558 boolean_t
2559 vdev_writeable(vdev_t *vd)
2560 {
2561 return (!vdev_is_dead(vd) && !vd->vdev_cant_write);
2562 }
2563
2564 boolean_t
2565 vdev_allocatable(vdev_t *vd)
2566 {
2567 uint64_t state = vd->vdev_state;
2568
2569 /*
2570 * We currently allow allocations from vdevs which may be in the
2571 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2572 * fails to reopen then we'll catch it later when we're holding
2573 * the proper locks. Note that we have to get the vdev state
2574 * in a local variable because although it changes atomically,
2575 * we're asking two separate questions about it.
2576 */
2577 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) &&
2578 !vd->vdev_cant_write && !vd->vdev_ishole);
2579 }
2580
2581 boolean_t
2582 vdev_accessible(vdev_t *vd, zio_t *zio)
2583 {
2584 ASSERT(zio->io_vd == vd);
2585
2586 if (vdev_is_dead(vd) || vd->vdev_remove_wanted)
2587 return (B_FALSE);
2588
2589 if (zio->io_type == ZIO_TYPE_READ)
2590 return (!vd->vdev_cant_read);
2591
2592 if (zio->io_type == ZIO_TYPE_WRITE)
2593 return (!vd->vdev_cant_write);
2594
2595 return (B_TRUE);
2596 }
2597
2598 /*
2599 * Get statistics for the given vdev.
2600 */
2601 void
2602 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
2603 {
2604 vdev_t *rvd = vd->vdev_spa->spa_root_vdev;
2605
2606 mutex_enter(&vd->vdev_stat_lock);
2607 bcopy(&vd->vdev_stat, vs, sizeof (*vs));
2608 vs->vs_timestamp = gethrtime() - vs->vs_timestamp;
2609 vs->vs_state = vd->vdev_state;
2610 vs->vs_rsize = vdev_get_min_asize(vd);
2611 if (vd->vdev_ops->vdev_op_leaf)
2612 vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
2613 vs->vs_esize = vd->vdev_max_asize - vd->vdev_asize;
2614 mutex_exit(&vd->vdev_stat_lock);
2615
2616 /*
2617 * If we're getting stats on the root vdev, aggregate the I/O counts
2618 * over all top-level vdevs (i.e. the direct children of the root).
2619 */
2620 if (vd == rvd) {
2621 for (int c = 0; c < rvd->vdev_children; c++) {
2622 vdev_t *cvd = rvd->vdev_child[c];
2623 vdev_stat_t *cvs = &cvd->vdev_stat;
2624
2625 mutex_enter(&vd->vdev_stat_lock);
2626 for (int t = 0; t < ZIO_TYPES; t++) {
2627 vs->vs_ops[t] += cvs->vs_ops[t];
2628 vs->vs_bytes[t] += cvs->vs_bytes[t];
2629 }
2630 cvs->vs_scan_removing = cvd->vdev_removing;
2631 mutex_exit(&vd->vdev_stat_lock);
2632 }
2633 }
2634 }
2635
2636 void
2637 vdev_clear_stats(vdev_t *vd)
2638 {
2639 mutex_enter(&vd->vdev_stat_lock);
2640 vd->vdev_stat.vs_space = 0;
2641 vd->vdev_stat.vs_dspace = 0;
2642 vd->vdev_stat.vs_alloc = 0;
2643 mutex_exit(&vd->vdev_stat_lock);
2644 }
2645
2646 void
2647 vdev_scan_stat_init(vdev_t *vd)
2648 {
2649 vdev_stat_t *vs = &vd->vdev_stat;
2650
2651 for (int c = 0; c < vd->vdev_children; c++)
2652 vdev_scan_stat_init(vd->vdev_child[c]);
2653
2654 mutex_enter(&vd->vdev_stat_lock);
2655 vs->vs_scan_processed = 0;
2656 mutex_exit(&vd->vdev_stat_lock);
2657 }
2658
2659 void
2660 vdev_stat_update(zio_t *zio, uint64_t psize)
2661 {
2662 spa_t *spa = zio->io_spa;
2663 vdev_t *rvd = spa->spa_root_vdev;
2664 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd;
2665 vdev_t *pvd;
2666 uint64_t txg = zio->io_txg;
2667 vdev_stat_t *vs = &vd->vdev_stat;
2668 zio_type_t type = zio->io_type;
2669 int flags = zio->io_flags;
2670
2671 /*
2672 * If this i/o is a gang leader, it didn't do any actual work.
2673 */
2674 if (zio->io_gang_tree)
2675 return;
2676
2677 if (zio->io_error == 0) {
2678 /*
2679 * If this is a root i/o, don't count it -- we've already
2680 * counted the top-level vdevs, and vdev_get_stats() will
2681 * aggregate them when asked. This reduces contention on
2682 * the root vdev_stat_lock and implicitly handles blocks
2683 * that compress away to holes, for which there is no i/o.
2684 * (Holes never create vdev children, so all the counters
2685 * remain zero, which is what we want.)
2686 *
2687 * Note: this only applies to successful i/o (io_error == 0)
2688 * because unlike i/o counts, errors are not additive.
2689 * When reading a ditto block, for example, failure of
2690 * one top-level vdev does not imply a root-level error.
2691 */
2692 if (vd == rvd)
2693 return;
2694
2695 ASSERT(vd == zio->io_vd);
2696
2697 if (flags & ZIO_FLAG_IO_BYPASS)
2698 return;
2699
2700 mutex_enter(&vd->vdev_stat_lock);
2701
2702 if (flags & ZIO_FLAG_IO_REPAIR) {
2703 if (flags & ZIO_FLAG_SCAN_THREAD) {
2704 dsl_scan_phys_t *scn_phys =
2705 &spa->spa_dsl_pool->dp_scan->scn_phys;
2706 uint64_t *processed = &scn_phys->scn_processed;
2707
2708 /* XXX cleanup? */
2709 if (vd->vdev_ops->vdev_op_leaf)
2710 atomic_add_64(processed, psize);
2711 vs->vs_scan_processed += psize;
2712 }
2713
2714 if (flags & ZIO_FLAG_SELF_HEAL)
2715 vs->vs_self_healed += psize;
2716 }
2717
2718 vs->vs_ops[type]++;
2719 vs->vs_bytes[type] += psize;
2720
2721 mutex_exit(&vd->vdev_stat_lock);
2722 return;
2723 }
2724
2725 if (flags & ZIO_FLAG_SPECULATIVE)
2726 return;
2727
2728 /*
2729 * If this is an I/O error that is going to be retried, then ignore the
2730 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
2731 * hard errors, when in reality they can happen for any number of
2732 * innocuous reasons (bus resets, MPxIO link failure, etc).
2733 */
2734 if (zio->io_error == EIO &&
2735 !(zio->io_flags & ZIO_FLAG_IO_RETRY))
2736 return;
2737
2738 /*
2739 * Intent logs writes won't propagate their error to the root
2740 * I/O so don't mark these types of failures as pool-level
2741 * errors.
2742 */
2743 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE))
2744 return;
2745
2746 mutex_enter(&vd->vdev_stat_lock);
2747 if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) {
2748 if (zio->io_error == ECKSUM)
2749 vs->vs_checksum_errors++;
2750 else
2751 vs->vs_read_errors++;
2752 }
2753 if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd))
2754 vs->vs_write_errors++;
2755 mutex_exit(&vd->vdev_stat_lock);
2756
2757 if (type == ZIO_TYPE_WRITE && txg != 0 &&
2758 (!(flags & ZIO_FLAG_IO_REPAIR) ||
2759 (flags & ZIO_FLAG_SCAN_THREAD) ||
2760 spa->spa_claiming)) {
2761 /*
2762 * This is either a normal write (not a repair), or it's
2763 * a repair induced by the scrub thread, or it's a repair
2764 * made by zil_claim() during spa_load() in the first txg.
2765 * In the normal case, we commit the DTL change in the same
2766 * txg as the block was born. In the scrub-induced repair
2767 * case, we know that scrubs run in first-pass syncing context,
2768 * so we commit the DTL change in spa_syncing_txg(spa).
2769 * In the zil_claim() case, we commit in spa_first_txg(spa).
2770 *
2771 * We currently do not make DTL entries for failed spontaneous
2772 * self-healing writes triggered by normal (non-scrubbing)
2773 * reads, because we have no transactional context in which to
2774 * do so -- and it's not clear that it'd be desirable anyway.
2775 */
2776 if (vd->vdev_ops->vdev_op_leaf) {
2777 uint64_t commit_txg = txg;
2778 if (flags & ZIO_FLAG_SCAN_THREAD) {
2779 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2780 ASSERT(spa_sync_pass(spa) == 1);
2781 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1);
2782 commit_txg = spa_syncing_txg(spa);
2783 } else if (spa->spa_claiming) {
2784 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2785 commit_txg = spa_first_txg(spa);
2786 }
2787 ASSERT(commit_txg >= spa_syncing_txg(spa));
2788 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1))
2789 return;
2790 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2791 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1);
2792 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg);
2793 }
2794 if (vd != rvd)
2795 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1);
2796 }
2797 }
2798
2799 /*
2800 * Update the in-core space usage stats for this vdev, its metaslab class,
2801 * and the root vdev.
2802 */
2803 void
2804 vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta,
2805 int64_t space_delta)
2806 {
2807 int64_t dspace_delta = space_delta;
2808 spa_t *spa = vd->vdev_spa;
2809 vdev_t *rvd = spa->spa_root_vdev;
2810 metaslab_group_t *mg = vd->vdev_mg;
2811 metaslab_class_t *mc = mg ? mg->mg_class : NULL;
2812
2813 ASSERT(vd == vd->vdev_top);
2814
2815 /*
2816 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
2817 * factor. We must calculate this here and not at the root vdev
2818 * because the root vdev's psize-to-asize is simply the max of its
2819 * childrens', thus not accurate enough for us.
2820 */
2821 ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0);
2822 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache);
2823 dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) *
2824 vd->vdev_deflate_ratio;
2825
2826 mutex_enter(&vd->vdev_stat_lock);
2827 vd->vdev_stat.vs_alloc += alloc_delta;
2828 vd->vdev_stat.vs_space += space_delta;
2829 vd->vdev_stat.vs_dspace += dspace_delta;
2830 mutex_exit(&vd->vdev_stat_lock);
2831
2832 if (mc == spa_normal_class(spa)) {
2833 mutex_enter(&rvd->vdev_stat_lock);
2834 rvd->vdev_stat.vs_alloc += alloc_delta;
2835 rvd->vdev_stat.vs_space += space_delta;
2836 rvd->vdev_stat.vs_dspace += dspace_delta;
2837 mutex_exit(&rvd->vdev_stat_lock);
2838 }
2839
2840 if (mc != NULL) {
2841 ASSERT(rvd == vd->vdev_parent);
2842 ASSERT(vd->vdev_ms_count != 0);
2843
2844 metaslab_class_space_update(mc,
2845 alloc_delta, defer_delta, space_delta, dspace_delta);
2846 }
2847 }
2848
2849 /*
2850 * Mark a top-level vdev's config as dirty, placing it on the dirty list
2851 * so that it will be written out next time the vdev configuration is synced.
2852 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
2853 */
2854 void
2855 vdev_config_dirty(vdev_t *vd)
2856 {
2857 spa_t *spa = vd->vdev_spa;
2858 vdev_t *rvd = spa->spa_root_vdev;
2859 int c;
2860
2861 ASSERT(spa_writeable(spa));
2862
2863 /*
2864 * If this is an aux vdev (as with l2cache and spare devices), then we
2865 * update the vdev config manually and set the sync flag.
2866 */
2867 if (vd->vdev_aux != NULL) {
2868 spa_aux_vdev_t *sav = vd->vdev_aux;
2869 nvlist_t **aux;
2870 uint_t naux;
2871
2872 for (c = 0; c < sav->sav_count; c++) {
2873 if (sav->sav_vdevs[c] == vd)
2874 break;
2875 }
2876
2877 if (c == sav->sav_count) {
2878 /*
2879 * We're being removed. There's nothing more to do.
2880 */
2881 ASSERT(sav->sav_sync == B_TRUE);
2882 return;
2883 }
2884
2885 sav->sav_sync = B_TRUE;
2886
2887 if (nvlist_lookup_nvlist_array(sav->sav_config,
2888 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) {
2889 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
2890 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0);
2891 }
2892
2893 ASSERT(c < naux);
2894
2895 /*
2896 * Setting the nvlist in the middle if the array is a little
2897 * sketchy, but it will work.
2898 */
2899 nvlist_free(aux[c]);
2900 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0);
2901
2902 return;
2903 }
2904
2905 /*
2906 * The dirty list is protected by the SCL_CONFIG lock. The caller
2907 * must either hold SCL_CONFIG as writer, or must be the sync thread
2908 * (which holds SCL_CONFIG as reader). There's only one sync thread,
2909 * so this is sufficient to ensure mutual exclusion.
2910 */
2911 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
2912 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2913 spa_config_held(spa, SCL_CONFIG, RW_READER)));
2914
2915 if (vd == rvd) {
2916 for (c = 0; c < rvd->vdev_children; c++)
2917 vdev_config_dirty(rvd->vdev_child[c]);
2918 } else {
2919 ASSERT(vd == vd->vdev_top);
2920
2921 if (!list_link_active(&vd->vdev_config_dirty_node) &&
2922 !vd->vdev_ishole)
2923 list_insert_head(&spa->spa_config_dirty_list, vd);
2924 }
2925 }
2926
2927 void
2928 vdev_config_clean(vdev_t *vd)
2929 {
2930 spa_t *spa = vd->vdev_spa;
2931
2932 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
2933 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2934 spa_config_held(spa, SCL_CONFIG, RW_READER)));
2935
2936 ASSERT(list_link_active(&vd->vdev_config_dirty_node));
2937 list_remove(&spa->spa_config_dirty_list, vd);
2938 }
2939
2940 /*
2941 * Mark a top-level vdev's state as dirty, so that the next pass of
2942 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
2943 * the state changes from larger config changes because they require
2944 * much less locking, and are often needed for administrative actions.
2945 */
2946 void
2947 vdev_state_dirty(vdev_t *vd)
2948 {
2949 spa_t *spa = vd->vdev_spa;
2950
2951 ASSERT(spa_writeable(spa));
2952 ASSERT(vd == vd->vdev_top);
2953
2954 /*
2955 * The state list is protected by the SCL_STATE lock. The caller
2956 * must either hold SCL_STATE as writer, or must be the sync thread
2957 * (which holds SCL_STATE as reader). There's only one sync thread,
2958 * so this is sufficient to ensure mutual exclusion.
2959 */
2960 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
2961 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2962 spa_config_held(spa, SCL_STATE, RW_READER)));
2963
2964 if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole)
2965 list_insert_head(&spa->spa_state_dirty_list, vd);
2966 }
2967
2968 void
2969 vdev_state_clean(vdev_t *vd)
2970 {
2971 spa_t *spa = vd->vdev_spa;
2972
2973 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
2974 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2975 spa_config_held(spa, SCL_STATE, RW_READER)));
2976
2977 ASSERT(list_link_active(&vd->vdev_state_dirty_node));
2978 list_remove(&spa->spa_state_dirty_list, vd);
2979 }
2980
2981 /*
2982 * Propagate vdev state up from children to parent.
2983 */
2984 void
2985 vdev_propagate_state(vdev_t *vd)
2986 {
2987 spa_t *spa = vd->vdev_spa;
2988 vdev_t *rvd = spa->spa_root_vdev;
2989 int degraded = 0, faulted = 0;
2990 int corrupted = 0;
2991 vdev_t *child;
2992
2993 if (vd->vdev_children > 0) {
2994 for (int c = 0; c < vd->vdev_children; c++) {
2995 child = vd->vdev_child[c];
2996
2997 /*
2998 * Don't factor holes into the decision.
2999 */
3000 if (child->vdev_ishole)
3001 continue;
3002
3003 if (!vdev_readable(child) ||
3004 (!vdev_writeable(child) && spa_writeable(spa))) {
3005 /*
3006 * Root special: if there is a top-level log
3007 * device, treat the root vdev as if it were
3008 * degraded.
3009 */
3010 if (child->vdev_islog && vd == rvd)
3011 degraded++;
3012 else
3013 faulted++;
3014 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) {
3015 degraded++;
3016 }
3017
3018 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA)
3019 corrupted++;
3020 }
3021
3022 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded);
3023
3024 /*
3025 * Root special: if there is a top-level vdev that cannot be
3026 * opened due to corrupted metadata, then propagate the root
3027 * vdev's aux state as 'corrupt' rather than 'insufficient
3028 * replicas'.
3029 */
3030 if (corrupted && vd == rvd &&
3031 rvd->vdev_state == VDEV_STATE_CANT_OPEN)
3032 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN,
3033 VDEV_AUX_CORRUPT_DATA);
3034 }
3035
3036 if (vd->vdev_parent)
3037 vdev_propagate_state(vd->vdev_parent);
3038 }
3039
3040 /*
3041 * Set a vdev's state. If this is during an open, we don't update the parent
3042 * state, because we're in the process of opening children depth-first.
3043 * Otherwise, we propagate the change to the parent.
3044 *
3045 * If this routine places a device in a faulted state, an appropriate ereport is
3046 * generated.
3047 */
3048 void
3049 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
3050 {
3051 uint64_t save_state;
3052 spa_t *spa = vd->vdev_spa;
3053
3054 if (state == vd->vdev_state) {
3055 vd->vdev_stat.vs_aux = aux;
3056 return;
3057 }
3058
3059 save_state = vd->vdev_state;
3060
3061 vd->vdev_state = state;
3062 vd->vdev_stat.vs_aux = aux;
3063
3064 /*
3065 * If we are setting the vdev state to anything but an open state, then
3066 * always close the underlying device unless the device has requested
3067 * a delayed close (i.e. we're about to remove or fault the device).
3068 * Otherwise, we keep accessible but invalid devices open forever.
3069 * We don't call vdev_close() itself, because that implies some extra
3070 * checks (offline, etc) that we don't want here. This is limited to
3071 * leaf devices, because otherwise closing the device will affect other
3072 * children.
3073 */
3074 if (!vd->vdev_delayed_close && vdev_is_dead(vd) &&
3075 vd->vdev_ops->vdev_op_leaf)
3076 vd->vdev_ops->vdev_op_close(vd);
3077
3078 /*
3079 * If we have brought this vdev back into service, we need
3080 * to notify fmd so that it can gracefully repair any outstanding
3081 * cases due to a missing device. We do this in all cases, even those
3082 * that probably don't correlate to a repaired fault. This is sure to
3083 * catch all cases, and we let the zfs-retire agent sort it out. If
3084 * this is a transient state it's OK, as the retire agent will
3085 * double-check the state of the vdev before repairing it.
3086 */
3087 if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf &&
3088 vd->vdev_prevstate != state)
3089 zfs_post_state_change(spa, vd);
3090
3091 if (vd->vdev_removed &&
3092 state == VDEV_STATE_CANT_OPEN &&
3093 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) {
3094 /*
3095 * If the previous state is set to VDEV_STATE_REMOVED, then this
3096 * device was previously marked removed and someone attempted to
3097 * reopen it. If this failed due to a nonexistent device, then
3098 * keep the device in the REMOVED state. We also let this be if
3099 * it is one of our special test online cases, which is only
3100 * attempting to online the device and shouldn't generate an FMA
3101 * fault.
3102 */
3103 vd->vdev_state = VDEV_STATE_REMOVED;
3104 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
3105 } else if (state == VDEV_STATE_REMOVED) {
3106 vd->vdev_removed = B_TRUE;
3107 } else if (state == VDEV_STATE_CANT_OPEN) {
3108 /*
3109 * If we fail to open a vdev during an import or recovery, we
3110 * mark it as "not available", which signifies that it was
3111 * never there to begin with. Failure to open such a device
3112 * is not considered an error.
3113 */
3114 if ((spa_load_state(spa) == SPA_LOAD_IMPORT ||
3115 spa_load_state(spa) == SPA_LOAD_RECOVER) &&
3116 vd->vdev_ops->vdev_op_leaf)
3117 vd->vdev_not_present = 1;
3118
3119 /*
3120 * Post the appropriate ereport. If the 'prevstate' field is
3121 * set to something other than VDEV_STATE_UNKNOWN, it indicates
3122 * that this is part of a vdev_reopen(). In this case, we don't
3123 * want to post the ereport if the device was already in the
3124 * CANT_OPEN state beforehand.
3125 *
3126 * If the 'checkremove' flag is set, then this is an attempt to
3127 * online the device in response to an insertion event. If we
3128 * hit this case, then we have detected an insertion event for a
3129 * faulted or offline device that wasn't in the removed state.
3130 * In this scenario, we don't post an ereport because we are
3131 * about to replace the device, or attempt an online with
3132 * vdev_forcefault, which will generate the fault for us.
3133 */
3134 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) &&
3135 !vd->vdev_not_present && !vd->vdev_checkremove &&
3136 vd != spa->spa_root_vdev) {
3137 const char *class;
3138
3139 switch (aux) {
3140 case VDEV_AUX_OPEN_FAILED:
3141 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED;
3142 break;
3143 case VDEV_AUX_CORRUPT_DATA:
3144 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA;
3145 break;
3146 case VDEV_AUX_NO_REPLICAS:
3147 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS;
3148 break;
3149 case VDEV_AUX_BAD_GUID_SUM:
3150 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM;
3151 break;
3152 case VDEV_AUX_TOO_SMALL:
3153 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL;
3154 break;
3155 case VDEV_AUX_BAD_LABEL:
3156 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL;
3157 break;
3158 default:
3159 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN;
3160 }
3161
3162 zfs_ereport_post(class, spa, vd, NULL, save_state, 0);
3163 }
3164
3165 /* Erase any notion of persistent removed state */
3166 vd->vdev_removed = B_FALSE;
3167 } else {
3168 vd->vdev_removed = B_FALSE;
3169 }
3170
3171 if (!isopen && vd->vdev_parent)
3172 vdev_propagate_state(vd->vdev_parent);
3173 }
3174
3175 /*
3176 * Check the vdev configuration to ensure that it's capable of supporting
3177 * a root pool. Currently, we do not support RAID-Z or partial configuration.
3178 * In addition, only a single top-level vdev is allowed and none of the leaves
3179 * can be wholedisks.
3180 */
3181 boolean_t
3182 vdev_is_bootable(vdev_t *vd)
3183 {
3184 if (!vd->vdev_ops->vdev_op_leaf) {
3185 char *vdev_type = vd->vdev_ops->vdev_op_type;
3186
3187 if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 &&
3188 vd->vdev_children > 1) {
3189 return (B_FALSE);
3190 } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 ||
3191 strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) {
3192 return (B_FALSE);
3193 }
3194 } else if (vd->vdev_wholedisk == 1) {
3195 return (B_FALSE);
3196 }
3197
3198 for (int c = 0; c < vd->vdev_children; c++) {
3199 if (!vdev_is_bootable(vd->vdev_child[c]))
3200 return (B_FALSE);
3201 }
3202 return (B_TRUE);
3203 }
3204
3205 /*
3206 * Load the state from the original vdev tree (ovd) which
3207 * we've retrieved from the MOS config object. If the original
3208 * vdev was offline or faulted then we transfer that state to the
3209 * device in the current vdev tree (nvd).
3210 */
3211 void
3212 vdev_load_log_state(vdev_t *nvd, vdev_t *ovd)
3213 {
3214 spa_t *spa = nvd->vdev_spa;
3215
3216 ASSERT(nvd->vdev_top->vdev_islog);
3217 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
3218 ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid);
3219
3220 for (int c = 0; c < nvd->vdev_children; c++)
3221 vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]);
3222
3223 if (nvd->vdev_ops->vdev_op_leaf) {
3224 /*
3225 * Restore the persistent vdev state
3226 */
3227 nvd->vdev_offline = ovd->vdev_offline;
3228 nvd->vdev_faulted = ovd->vdev_faulted;
3229 nvd->vdev_degraded = ovd->vdev_degraded;
3230 nvd->vdev_removed = ovd->vdev_removed;
3231 }
3232 }
3233
3234 /*
3235 * Determine if a log device has valid content. If the vdev was
3236 * removed or faulted in the MOS config then we know that
3237 * the content on the log device has already been written to the pool.
3238 */
3239 boolean_t
3240 vdev_log_state_valid(vdev_t *vd)
3241 {
3242 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted &&
3243 !vd->vdev_removed)
3244 return (B_TRUE);
3245
3246 for (int c = 0; c < vd->vdev_children; c++)
3247 if (vdev_log_state_valid(vd->vdev_child[c]))
3248 return (B_TRUE);
3249
3250 return (B_FALSE);
3251 }
3252
3253 /*
3254 * Expand a vdev if possible.
3255 */
3256 void
3257 vdev_expand(vdev_t *vd, uint64_t txg)
3258 {
3259 ASSERT(vd->vdev_top == vd);
3260 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3261
3262 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) {
3263 VERIFY(vdev_metaslab_init(vd, txg) == 0);
3264 vdev_config_dirty(vd);
3265 }
3266 }
3267
3268 /*
3269 * Split a vdev.
3270 */
3271 void
3272 vdev_split(vdev_t *vd)
3273 {
3274 vdev_t *cvd, *pvd = vd->vdev_parent;
3275
3276 vdev_remove_child(pvd, vd);
3277 vdev_compact_children(pvd);
3278
3279 cvd = pvd->vdev_child[0];
3280 if (pvd->vdev_children == 1) {
3281 vdev_remove_parent(cvd);
3282 cvd->vdev_splitting = B_TRUE;
3283 }
3284 vdev_propagate_state(cvd);
3285 }
3286
3287 void
3288 vdev_deadman(vdev_t *vd)
3289 {
3290 for (int c = 0; c < vd->vdev_children; c++) {
3291 vdev_t *cvd = vd->vdev_child[c];
3292
3293 vdev_deadman(cvd);
3294 }
3295
3296 if (vd->vdev_ops->vdev_op_leaf) {
3297 vdev_queue_t *vq = &vd->vdev_queue;
3298
3299 mutex_enter(&vq->vq_lock);
3300 if (avl_numnodes(&vq->vq_active_tree) > 0) {
3301 spa_t *spa = vd->vdev_spa;
3302 zio_t *fio;
3303 uint64_t delta;
3304
3305 /*
3306 * Look at the head of all the pending queues,
3307 * if any I/O has been outstanding for longer than
3308 * the spa_deadman_synctime we panic the system.
3309 */
3310 fio = avl_first(&vq->vq_active_tree);
3311 delta = gethrtime() - fio->io_timestamp;
3312 if (delta > spa_deadman_synctime(spa)) {
3313 zfs_dbgmsg("SLOW IO: zio timestamp %lluns, "
3314 "delta %lluns, last io %lluns",
3315 fio->io_timestamp, delta,
3316 vq->vq_io_complete_ts);
3317 fm_panic("I/O to pool '%s' appears to be "
3318 "hung.", spa_name(spa));
3319 }
3320 }
3321 mutex_exit(&vq->vq_lock);
3322 }
3323 }