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 (c) 2011, 2014 by Delphix. All rights reserved.
25 * Copyright 2015 Nexenta Systems, Inc. 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 * When a vdev is added, it will be divided into approximately (but no
69 * more than) this number of metaslabs.
70 */
71 int metaslabs_per_vdev = 200;
72
73 /*
74 * Given a vdev type, return the appropriate ops vector.
75 */
76 static vdev_ops_t *
77 vdev_getops(const char *type)
78 {
79 vdev_ops_t *ops, **opspp;
80
81 for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++)
82 if (strcmp(ops->vdev_op_type, type) == 0)
83 break;
84
85 return (ops);
86 }
87
88 /*
89 * Default asize function: return the MAX of psize with the asize of
90 * all children. This is what's used by anything other than RAID-Z.
91 */
92 uint64_t
93 vdev_default_asize(vdev_t *vd, uint64_t psize)
94 {
95 uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift);
96 uint64_t csize;
97
98 for (int c = 0; c < vd->vdev_children; c++) {
99 csize = vdev_psize_to_asize(vd->vdev_child[c], psize);
100 asize = MAX(asize, csize);
101 }
102
103 return (asize);
104 }
105
106 /*
107 * Get the minimum allocatable size. We define the allocatable size as
108 * the vdev's asize rounded to the nearest metaslab. This allows us to
109 * replace or attach devices which don't have the same physical size but
110 * can still satisfy the same number of allocations.
111 */
112 uint64_t
113 vdev_get_min_asize(vdev_t *vd)
114 {
115 vdev_t *pvd = vd->vdev_parent;
116
117 /*
118 * If our parent is NULL (inactive spare or cache) or is the root,
119 * just return our own asize.
120 */
121 if (pvd == NULL)
122 return (vd->vdev_asize);
123
124 /*
125 * The top-level vdev just returns the allocatable size rounded
126 * to the nearest metaslab.
127 */
128 if (vd == vd->vdev_top)
129 return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift));
130
131 /*
132 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
133 * so each child must provide at least 1/Nth of its asize.
134 */
135 if (pvd->vdev_ops == &vdev_raidz_ops)
136 return (pvd->vdev_min_asize / pvd->vdev_children);
137
138 return (pvd->vdev_min_asize);
139 }
140
141 void
142 vdev_set_min_asize(vdev_t *vd)
143 {
144 vd->vdev_min_asize = vdev_get_min_asize(vd);
145
146 for (int c = 0; c < vd->vdev_children; c++)
147 vdev_set_min_asize(vd->vdev_child[c]);
148 }
149
150 vdev_t *
151 vdev_lookup_top(spa_t *spa, uint64_t vdev)
152 {
153 vdev_t *rvd = spa->spa_root_vdev;
154
155 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
156
157 if (vdev < rvd->vdev_children) {
158 ASSERT(rvd->vdev_child[vdev] != NULL);
159 return (rvd->vdev_child[vdev]);
160 }
161
162 return (NULL);
163 }
164
165 vdev_t *
166 vdev_lookup_by_guid(vdev_t *vd, uint64_t guid)
167 {
168 vdev_t *mvd;
169
170 if (vd->vdev_guid == guid)
171 return (vd);
172
173 for (int c = 0; c < vd->vdev_children; c++)
174 if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) !=
175 NULL)
176 return (mvd);
177
178 return (NULL);
179 }
180
181 static int
182 vdev_count_leaves_impl(vdev_t *vd)
183 {
184 int n = 0;
185
186 if (vd->vdev_ops->vdev_op_leaf)
187 return (1);
188
189 for (int c = 0; c < vd->vdev_children; c++)
190 n += vdev_count_leaves_impl(vd->vdev_child[c]);
191
192 return (n);
193 }
194
195 int
196 vdev_count_leaves(spa_t *spa)
197 {
198 return (vdev_count_leaves_impl(spa->spa_root_vdev));
199 }
200
201 void
202 vdev_add_child(vdev_t *pvd, vdev_t *cvd)
203 {
204 size_t oldsize, newsize;
205 uint64_t id = cvd->vdev_id;
206 vdev_t **newchild;
207
208 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
209 ASSERT(cvd->vdev_parent == NULL);
210
211 cvd->vdev_parent = pvd;
212
213 if (pvd == NULL)
214 return;
215
216 ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL);
217
218 oldsize = pvd->vdev_children * sizeof (vdev_t *);
219 pvd->vdev_children = MAX(pvd->vdev_children, id + 1);
220 newsize = pvd->vdev_children * sizeof (vdev_t *);
221
222 newchild = kmem_zalloc(newsize, KM_SLEEP);
223 if (pvd->vdev_child != NULL) {
224 bcopy(pvd->vdev_child, newchild, oldsize);
225 kmem_free(pvd->vdev_child, oldsize);
226 }
227
228 pvd->vdev_child = newchild;
229 pvd->vdev_child[id] = cvd;
230
231 cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd);
232 ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL);
233
234 /*
235 * Walk up all ancestors to update guid sum.
236 */
237 for (; pvd != NULL; pvd = pvd->vdev_parent)
238 pvd->vdev_guid_sum += cvd->vdev_guid_sum;
239 }
240
241 void
242 vdev_remove_child(vdev_t *pvd, vdev_t *cvd)
243 {
244 int c;
245 uint_t id = cvd->vdev_id;
246
247 ASSERT(cvd->vdev_parent == pvd);
248
249 if (pvd == NULL)
250 return;
251
252 ASSERT(id < pvd->vdev_children);
253 ASSERT(pvd->vdev_child[id] == cvd);
254
255 pvd->vdev_child[id] = NULL;
256 cvd->vdev_parent = NULL;
257
258 for (c = 0; c < pvd->vdev_children; c++)
259 if (pvd->vdev_child[c])
260 break;
261
262 if (c == pvd->vdev_children) {
263 kmem_free(pvd->vdev_child, c * sizeof (vdev_t *));
264 pvd->vdev_child = NULL;
265 pvd->vdev_children = 0;
266 }
267
268 /*
269 * Walk up all ancestors to update guid sum.
270 */
271 for (; pvd != NULL; pvd = pvd->vdev_parent)
272 pvd->vdev_guid_sum -= cvd->vdev_guid_sum;
273 }
274
275 /*
276 * Remove any holes in the child array.
277 */
278 void
279 vdev_compact_children(vdev_t *pvd)
280 {
281 vdev_t **newchild, *cvd;
282 int oldc = pvd->vdev_children;
283 int newc;
284
285 ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
286
287 for (int c = newc = 0; c < oldc; c++)
288 if (pvd->vdev_child[c])
289 newc++;
290
291 newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP);
292
293 for (int c = newc = 0; c < oldc; c++) {
294 if ((cvd = pvd->vdev_child[c]) != NULL) {
295 newchild[newc] = cvd;
296 cvd->vdev_id = newc++;
297 }
298 }
299
300 kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *));
301 pvd->vdev_child = newchild;
302 pvd->vdev_children = newc;
303 }
304
305 /*
306 * Allocate and minimally initialize a vdev_t.
307 */
308 vdev_t *
309 vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops)
310 {
311 vdev_t *vd;
312
313 vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP);
314
315 if (spa->spa_root_vdev == NULL) {
316 ASSERT(ops == &vdev_root_ops);
317 spa->spa_root_vdev = vd;
318 spa->spa_load_guid = spa_generate_guid(NULL);
319 }
320
321 if (guid == 0 && ops != &vdev_hole_ops) {
322 if (spa->spa_root_vdev == vd) {
323 /*
324 * The root vdev's guid will also be the pool guid,
325 * which must be unique among all pools.
326 */
327 guid = spa_generate_guid(NULL);
328 } else {
329 /*
330 * Any other vdev's guid must be unique within the pool.
331 */
332 guid = spa_generate_guid(spa);
333 }
334 ASSERT(!spa_guid_exists(spa_guid(spa), guid));
335 }
336
337 vd->vdev_spa = spa;
338 vd->vdev_id = id;
339 vd->vdev_guid = guid;
340 vd->vdev_guid_sum = guid;
341 vd->vdev_ops = ops;
342 vd->vdev_state = VDEV_STATE_CLOSED;
343 vd->vdev_ishole = (ops == &vdev_hole_ops);
344
345 mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL);
346 mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL);
347 mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL);
348 for (int t = 0; t < DTL_TYPES; t++) {
349 vd->vdev_dtl[t] = range_tree_create(NULL, NULL,
350 &vd->vdev_dtl_lock);
351 }
352 txg_list_create(&vd->vdev_ms_list,
353 offsetof(struct metaslab, ms_txg_node));
354 txg_list_create(&vd->vdev_dtl_list,
355 offsetof(struct vdev, vdev_dtl_node));
356 vd->vdev_stat.vs_timestamp = gethrtime();
357 vdev_queue_init(vd);
358 vdev_cache_init(vd);
359
360 return (vd);
361 }
362
363 /*
364 * Allocate a new vdev. The 'alloctype' is used to control whether we are
365 * creating a new vdev or loading an existing one - the behavior is slightly
366 * different for each case.
367 */
368 int
369 vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
370 int alloctype)
371 {
372 vdev_ops_t *ops;
373 char *type;
374 uint64_t guid = 0, islog, nparity;
375 vdev_t *vd;
376
377 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
378
379 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
380 return (SET_ERROR(EINVAL));
381
382 if ((ops = vdev_getops(type)) == NULL)
383 return (SET_ERROR(EINVAL));
384
385 /*
386 * If this is a load, get the vdev guid from the nvlist.
387 * Otherwise, vdev_alloc_common() will generate one for us.
388 */
389 if (alloctype == VDEV_ALLOC_LOAD) {
390 uint64_t label_id;
391
392 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) ||
393 label_id != id)
394 return (SET_ERROR(EINVAL));
395
396 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
397 return (SET_ERROR(EINVAL));
398 } else if (alloctype == VDEV_ALLOC_SPARE) {
399 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
400 return (SET_ERROR(EINVAL));
401 } else if (alloctype == VDEV_ALLOC_L2CACHE) {
402 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
403 return (SET_ERROR(EINVAL));
404 } else if (alloctype == VDEV_ALLOC_ROOTPOOL) {
405 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
406 return (SET_ERROR(EINVAL));
407 }
408
409 /*
410 * The first allocated vdev must be of type 'root'.
411 */
412 if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL)
413 return (SET_ERROR(EINVAL));
414
415 /*
416 * Determine whether we're a log vdev.
417 */
418 islog = 0;
419 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog);
420 if (islog && spa_version(spa) < SPA_VERSION_SLOGS)
421 return (SET_ERROR(ENOTSUP));
422
423 if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES)
424 return (SET_ERROR(ENOTSUP));
425
426 /*
427 * Set the nparity property for RAID-Z vdevs.
428 */
429 nparity = -1ULL;
430 if (ops == &vdev_raidz_ops) {
431 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
432 &nparity) == 0) {
433 if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY)
434 return (SET_ERROR(EINVAL));
435 /*
436 * Previous versions could only support 1 or 2 parity
437 * device.
438 */
439 if (nparity > 1 &&
440 spa_version(spa) < SPA_VERSION_RAIDZ2)
441 return (SET_ERROR(ENOTSUP));
442 if (nparity > 2 &&
443 spa_version(spa) < SPA_VERSION_RAIDZ3)
444 return (SET_ERROR(ENOTSUP));
445 } else {
446 /*
447 * We require the parity to be specified for SPAs that
448 * support multiple parity levels.
449 */
450 if (spa_version(spa) >= SPA_VERSION_RAIDZ2)
451 return (SET_ERROR(EINVAL));
452 /*
453 * Otherwise, we default to 1 parity device for RAID-Z.
454 */
455 nparity = 1;
456 }
457 } else {
458 nparity = 0;
459 }
460 ASSERT(nparity != -1ULL);
461
462 vd = vdev_alloc_common(spa, id, guid, ops);
463
464 vd->vdev_islog = islog;
465 vd->vdev_nparity = nparity;
466
467 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0)
468 vd->vdev_path = spa_strdup(vd->vdev_path);
469 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0)
470 vd->vdev_devid = spa_strdup(vd->vdev_devid);
471 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH,
472 &vd->vdev_physpath) == 0)
473 vd->vdev_physpath = spa_strdup(vd->vdev_physpath);
474 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0)
475 vd->vdev_fru = spa_strdup(vd->vdev_fru);
476
477 /*
478 * Set the whole_disk property. If it's not specified, leave the value
479 * as -1.
480 */
481 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
482 &vd->vdev_wholedisk) != 0)
483 vd->vdev_wholedisk = -1ULL;
484
485 /*
486 * Look for the 'not present' flag. This will only be set if the device
487 * was not present at the time of import.
488 */
489 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
490 &vd->vdev_not_present);
491
492 /*
493 * Get the alignment requirement.
494 */
495 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift);
496
497 /*
498 * Retrieve the vdev creation time.
499 */
500 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG,
501 &vd->vdev_crtxg);
502
503 /*
504 * If we're a top-level vdev, try to load the allocation parameters.
505 */
506 if (parent && !parent->vdev_parent &&
507 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
508 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
509 &vd->vdev_ms_array);
510 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
511 &vd->vdev_ms_shift);
512 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE,
513 &vd->vdev_asize);
514 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING,
515 &vd->vdev_removing);
516 }
517
518 if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) {
519 ASSERT(alloctype == VDEV_ALLOC_LOAD ||
520 alloctype == VDEV_ALLOC_ADD ||
521 alloctype == VDEV_ALLOC_SPLIT ||
522 alloctype == VDEV_ALLOC_ROOTPOOL);
523 vd->vdev_mg = metaslab_group_create(islog ?
524 spa_log_class(spa) : spa_normal_class(spa), vd);
525 }
526
527 /*
528 * If we're a leaf vdev, try to load the DTL object and other state.
529 */
530 if (vd->vdev_ops->vdev_op_leaf &&
531 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE ||
532 alloctype == VDEV_ALLOC_ROOTPOOL)) {
533 if (alloctype == VDEV_ALLOC_LOAD) {
534 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL,
535 &vd->vdev_dtl_object);
536 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE,
537 &vd->vdev_unspare);
538 }
539
540 if (alloctype == VDEV_ALLOC_ROOTPOOL) {
541 uint64_t spare = 0;
542
543 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
544 &spare) == 0 && spare)
545 spa_spare_add(vd);
546 }
547
548 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE,
549 &vd->vdev_offline);
550
551 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
552 &vd->vdev_resilver_txg);
553
554 /*
555 * When importing a pool, we want to ignore the persistent fault
556 * state, as the diagnosis made on another system may not be
557 * valid in the current context. Local vdevs will
558 * remain in the faulted state.
559 */
560 if (spa_load_state(spa) == SPA_LOAD_OPEN) {
561 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED,
562 &vd->vdev_faulted);
563 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED,
564 &vd->vdev_degraded);
565 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED,
566 &vd->vdev_removed);
567
568 if (vd->vdev_faulted || vd->vdev_degraded) {
569 char *aux;
570
571 vd->vdev_label_aux =
572 VDEV_AUX_ERR_EXCEEDED;
573 if (nvlist_lookup_string(nv,
574 ZPOOL_CONFIG_AUX_STATE, &aux) == 0 &&
575 strcmp(aux, "external") == 0)
576 vd->vdev_label_aux = VDEV_AUX_EXTERNAL;
577 }
578 }
579 }
580
581 /*
582 * Add ourselves to the parent's list of children.
583 */
584 vdev_add_child(parent, vd);
585
586 *vdp = vd;
587
588 return (0);
589 }
590
591 void
592 vdev_free(vdev_t *vd)
593 {
594 spa_t *spa = vd->vdev_spa;
595
596 /*
597 * vdev_free() implies closing the vdev first. This is simpler than
598 * trying to ensure complicated semantics for all callers.
599 */
600 vdev_close(vd);
601
602 ASSERT(!list_link_active(&vd->vdev_config_dirty_node));
603 ASSERT(!list_link_active(&vd->vdev_state_dirty_node));
604
605 /*
606 * Free all children.
607 */
608 for (int c = 0; c < vd->vdev_children; c++)
609 vdev_free(vd->vdev_child[c]);
610
611 ASSERT(vd->vdev_child == NULL);
612 ASSERT(vd->vdev_guid_sum == vd->vdev_guid);
613
614 /*
615 * Discard allocation state.
616 */
617 if (vd->vdev_mg != NULL) {
618 vdev_metaslab_fini(vd);
619 metaslab_group_destroy(vd->vdev_mg);
620 }
621
622 ASSERT0(vd->vdev_stat.vs_space);
623 ASSERT0(vd->vdev_stat.vs_dspace);
624 ASSERT0(vd->vdev_stat.vs_alloc);
625
626 /*
627 * Remove this vdev from its parent's child list.
628 */
629 vdev_remove_child(vd->vdev_parent, vd);
630
631 ASSERT(vd->vdev_parent == NULL);
632
633 /*
634 * Clean up vdev structure.
635 */
636 vdev_queue_fini(vd);
637 vdev_cache_fini(vd);
638
639 if (vd->vdev_path)
640 spa_strfree(vd->vdev_path);
641 if (vd->vdev_devid)
642 spa_strfree(vd->vdev_devid);
643 if (vd->vdev_physpath)
644 spa_strfree(vd->vdev_physpath);
645 if (vd->vdev_fru)
646 spa_strfree(vd->vdev_fru);
647
648 if (vd->vdev_isspare)
649 spa_spare_remove(vd);
650 if (vd->vdev_isl2cache)
651 spa_l2cache_remove(vd);
652
653 txg_list_destroy(&vd->vdev_ms_list);
654 txg_list_destroy(&vd->vdev_dtl_list);
655
656 mutex_enter(&vd->vdev_dtl_lock);
657 space_map_close(vd->vdev_dtl_sm);
658 for (int t = 0; t < DTL_TYPES; t++) {
659 range_tree_vacate(vd->vdev_dtl[t], NULL, NULL);
660 range_tree_destroy(vd->vdev_dtl[t]);
661 }
662 mutex_exit(&vd->vdev_dtl_lock);
663
664 mutex_destroy(&vd->vdev_dtl_lock);
665 mutex_destroy(&vd->vdev_stat_lock);
666 mutex_destroy(&vd->vdev_probe_lock);
667
668 if (vd == spa->spa_root_vdev)
669 spa->spa_root_vdev = NULL;
670
671 kmem_free(vd, sizeof (vdev_t));
672 }
673
674 /*
675 * Transfer top-level vdev state from svd to tvd.
676 */
677 static void
678 vdev_top_transfer(vdev_t *svd, vdev_t *tvd)
679 {
680 spa_t *spa = svd->vdev_spa;
681 metaslab_t *msp;
682 vdev_t *vd;
683 int t;
684
685 ASSERT(tvd == tvd->vdev_top);
686
687 tvd->vdev_ms_array = svd->vdev_ms_array;
688 tvd->vdev_ms_shift = svd->vdev_ms_shift;
689 tvd->vdev_ms_count = svd->vdev_ms_count;
690
691 svd->vdev_ms_array = 0;
692 svd->vdev_ms_shift = 0;
693 svd->vdev_ms_count = 0;
694
695 if (tvd->vdev_mg)
696 ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg);
697 tvd->vdev_mg = svd->vdev_mg;
698 tvd->vdev_ms = svd->vdev_ms;
699
700 svd->vdev_mg = NULL;
701 svd->vdev_ms = NULL;
702
703 if (tvd->vdev_mg != NULL)
704 tvd->vdev_mg->mg_vd = tvd;
705
706 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc;
707 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space;
708 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace;
709
710 svd->vdev_stat.vs_alloc = 0;
711 svd->vdev_stat.vs_space = 0;
712 svd->vdev_stat.vs_dspace = 0;
713
714 for (t = 0; t < TXG_SIZE; t++) {
715 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL)
716 (void) txg_list_add(&tvd->vdev_ms_list, msp, t);
717 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL)
718 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t);
719 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t))
720 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t);
721 }
722
723 if (list_link_active(&svd->vdev_config_dirty_node)) {
724 vdev_config_clean(svd);
725 vdev_config_dirty(tvd);
726 }
727
728 if (list_link_active(&svd->vdev_state_dirty_node)) {
729 vdev_state_clean(svd);
730 vdev_state_dirty(tvd);
731 }
732
733 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio;
734 svd->vdev_deflate_ratio = 0;
735
736 tvd->vdev_islog = svd->vdev_islog;
737 svd->vdev_islog = 0;
738 }
739
740 static void
741 vdev_top_update(vdev_t *tvd, vdev_t *vd)
742 {
743 if (vd == NULL)
744 return;
745
746 vd->vdev_top = tvd;
747
748 for (int c = 0; c < vd->vdev_children; c++)
749 vdev_top_update(tvd, vd->vdev_child[c]);
750 }
751
752 /*
753 * Add a mirror/replacing vdev above an existing vdev.
754 */
755 vdev_t *
756 vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops)
757 {
758 spa_t *spa = cvd->vdev_spa;
759 vdev_t *pvd = cvd->vdev_parent;
760 vdev_t *mvd;
761
762 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
763
764 mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops);
765
766 mvd->vdev_asize = cvd->vdev_asize;
767 mvd->vdev_min_asize = cvd->vdev_min_asize;
768 mvd->vdev_max_asize = cvd->vdev_max_asize;
769 mvd->vdev_ashift = cvd->vdev_ashift;
770 mvd->vdev_state = cvd->vdev_state;
771 mvd->vdev_crtxg = cvd->vdev_crtxg;
772
773 vdev_remove_child(pvd, cvd);
774 vdev_add_child(pvd, mvd);
775 cvd->vdev_id = mvd->vdev_children;
776 vdev_add_child(mvd, cvd);
777 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
778
779 if (mvd == mvd->vdev_top)
780 vdev_top_transfer(cvd, mvd);
781
782 return (mvd);
783 }
784
785 /*
786 * Remove a 1-way mirror/replacing vdev from the tree.
787 */
788 void
789 vdev_remove_parent(vdev_t *cvd)
790 {
791 vdev_t *mvd = cvd->vdev_parent;
792 vdev_t *pvd = mvd->vdev_parent;
793
794 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
795
796 ASSERT(mvd->vdev_children == 1);
797 ASSERT(mvd->vdev_ops == &vdev_mirror_ops ||
798 mvd->vdev_ops == &vdev_replacing_ops ||
799 mvd->vdev_ops == &vdev_spare_ops);
800 cvd->vdev_ashift = mvd->vdev_ashift;
801
802 vdev_remove_child(mvd, cvd);
803 vdev_remove_child(pvd, mvd);
804
805 /*
806 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
807 * Otherwise, we could have detached an offline device, and when we
808 * go to import the pool we'll think we have two top-level vdevs,
809 * instead of a different version of the same top-level vdev.
810 */
811 if (mvd->vdev_top == mvd) {
812 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid;
813 cvd->vdev_orig_guid = cvd->vdev_guid;
814 cvd->vdev_guid += guid_delta;
815 cvd->vdev_guid_sum += guid_delta;
816 }
817 cvd->vdev_id = mvd->vdev_id;
818 vdev_add_child(pvd, cvd);
819 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
820
821 if (cvd == cvd->vdev_top)
822 vdev_top_transfer(mvd, cvd);
823
824 ASSERT(mvd->vdev_children == 0);
825 vdev_free(mvd);
826 }
827
828 int
829 vdev_metaslab_init(vdev_t *vd, uint64_t txg)
830 {
831 spa_t *spa = vd->vdev_spa;
832 objset_t *mos = spa->spa_meta_objset;
833 uint64_t m;
834 uint64_t oldc = vd->vdev_ms_count;
835 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift;
836 metaslab_t **mspp;
837 int error;
838
839 ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER));
840
841 /*
842 * This vdev is not being allocated from yet or is a hole.
843 */
844 if (vd->vdev_ms_shift == 0)
845 return (0);
846
847 ASSERT(!vd->vdev_ishole);
848
849 /*
850 * Compute the raidz-deflation ratio. Note, we hard-code
851 * in 128k (1 << 17) because it is the "typical" blocksize.
852 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change,
853 * otherwise it would inconsistently account for existing bp's.
854 */
855 vd->vdev_deflate_ratio = (1 << 17) /
856 (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT);
857
858 ASSERT(oldc <= newc);
859
860 mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP);
861
862 if (oldc != 0) {
863 bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp));
864 kmem_free(vd->vdev_ms, oldc * sizeof (*mspp));
865 }
866
867 vd->vdev_ms = mspp;
868 vd->vdev_ms_count = newc;
869
870 for (m = oldc; m < newc; m++) {
871 uint64_t object = 0;
872
873 if (txg == 0) {
874 error = dmu_read(mos, vd->vdev_ms_array,
875 m * sizeof (uint64_t), sizeof (uint64_t), &object,
876 DMU_READ_PREFETCH);
877 if (error)
878 return (error);
879 }
880
881 error = metaslab_init(vd->vdev_mg, m, object, txg,
882 &(vd->vdev_ms[m]));
883 if (error)
884 return (error);
885 }
886
887 if (txg == 0)
888 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER);
889
890 /*
891 * If the vdev is being removed we don't activate
892 * the metaslabs since we want to ensure that no new
893 * allocations are performed on this device.
894 */
895 if (oldc == 0 && !vd->vdev_removing)
896 metaslab_group_activate(vd->vdev_mg);
897
898 if (txg == 0)
899 spa_config_exit(spa, SCL_ALLOC, FTAG);
900
901 return (0);
902 }
903
904 void
905 vdev_metaslab_fini(vdev_t *vd)
906 {
907 uint64_t m;
908 uint64_t count = vd->vdev_ms_count;
909
910 if (vd->vdev_ms != NULL) {
911 metaslab_group_passivate(vd->vdev_mg);
912 for (m = 0; m < count; m++) {
913 metaslab_t *msp = vd->vdev_ms[m];
914
915 if (msp != NULL)
916 metaslab_fini(msp);
917 }
918 kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *));
919 vd->vdev_ms = NULL;
920 }
921 }
922
923 typedef struct vdev_probe_stats {
924 boolean_t vps_readable;
925 boolean_t vps_writeable;
926 int vps_flags;
927 } vdev_probe_stats_t;
928
929 static void
930 vdev_probe_done(zio_t *zio)
931 {
932 spa_t *spa = zio->io_spa;
933 vdev_t *vd = zio->io_vd;
934 vdev_probe_stats_t *vps = zio->io_private;
935
936 ASSERT(vd->vdev_probe_zio != NULL);
937
938 if (zio->io_type == ZIO_TYPE_READ) {
939 if (zio->io_error == 0)
940 vps->vps_readable = 1;
941 if (zio->io_error == 0 && spa_writeable(spa)) {
942 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd,
943 zio->io_offset, zio->io_size, zio->io_data,
944 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
945 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE));
946 } else {
947 zio_buf_free(zio->io_data, zio->io_size);
948 }
949 } else if (zio->io_type == ZIO_TYPE_WRITE) {
950 if (zio->io_error == 0)
951 vps->vps_writeable = 1;
952 zio_buf_free(zio->io_data, zio->io_size);
953 } else if (zio->io_type == ZIO_TYPE_NULL) {
954 zio_t *pio;
955
956 vd->vdev_cant_read |= !vps->vps_readable;
957 vd->vdev_cant_write |= !vps->vps_writeable;
958
959 if (vdev_readable(vd) &&
960 (vdev_writeable(vd) || !spa_writeable(spa))) {
961 zio->io_error = 0;
962 } else {
963 ASSERT(zio->io_error != 0);
964 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE,
965 spa, vd, NULL, 0, 0);
966 zio->io_error = SET_ERROR(ENXIO);
967 }
968
969 mutex_enter(&vd->vdev_probe_lock);
970 ASSERT(vd->vdev_probe_zio == zio);
971 vd->vdev_probe_zio = NULL;
972 mutex_exit(&vd->vdev_probe_lock);
973
974 while ((pio = zio_walk_parents(zio)) != NULL)
975 if (!vdev_accessible(vd, pio))
976 pio->io_error = SET_ERROR(ENXIO);
977
978 kmem_free(vps, sizeof (*vps));
979 }
980 }
981
982 /*
983 * Determine whether this device is accessible.
984 *
985 * Read and write to several known locations: the pad regions of each
986 * vdev label but the first, which we leave alone in case it contains
987 * a VTOC.
988 */
989 zio_t *
990 vdev_probe(vdev_t *vd, zio_t *zio)
991 {
992 spa_t *spa = vd->vdev_spa;
993 vdev_probe_stats_t *vps = NULL;
994 zio_t *pio;
995
996 ASSERT(vd->vdev_ops->vdev_op_leaf);
997
998 /*
999 * Don't probe the probe.
1000 */
1001 if (zio && (zio->io_flags & ZIO_FLAG_PROBE))
1002 return (NULL);
1003
1004 /*
1005 * To prevent 'probe storms' when a device fails, we create
1006 * just one probe i/o at a time. All zios that want to probe
1007 * this vdev will become parents of the probe io.
1008 */
1009 mutex_enter(&vd->vdev_probe_lock);
1010
1011 if ((pio = vd->vdev_probe_zio) == NULL) {
1012 vps = kmem_zalloc(sizeof (*vps), KM_SLEEP);
1013
1014 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE |
1015 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE |
1016 ZIO_FLAG_TRYHARD;
1017
1018 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) {
1019 /*
1020 * vdev_cant_read and vdev_cant_write can only
1021 * transition from TRUE to FALSE when we have the
1022 * SCL_ZIO lock as writer; otherwise they can only
1023 * transition from FALSE to TRUE. This ensures that
1024 * any zio looking at these values can assume that
1025 * failures persist for the life of the I/O. That's
1026 * important because when a device has intermittent
1027 * connectivity problems, we want to ensure that
1028 * they're ascribed to the device (ENXIO) and not
1029 * the zio (EIO).
1030 *
1031 * Since we hold SCL_ZIO as writer here, clear both
1032 * values so the probe can reevaluate from first
1033 * principles.
1034 */
1035 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER;
1036 vd->vdev_cant_read = B_FALSE;
1037 vd->vdev_cant_write = B_FALSE;
1038 }
1039
1040 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd,
1041 vdev_probe_done, vps,
1042 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE);
1043
1044 /*
1045 * We can't change the vdev state in this context, so we
1046 * kick off an async task to do it on our behalf.
1047 */
1048 if (zio != NULL) {
1049 vd->vdev_probe_wanted = B_TRUE;
1050 spa_async_request(spa, SPA_ASYNC_PROBE);
1051 }
1052 }
1053
1054 if (zio != NULL)
1055 zio_add_child(zio, pio);
1056
1057 mutex_exit(&vd->vdev_probe_lock);
1058
1059 if (vps == NULL) {
1060 ASSERT(zio != NULL);
1061 return (NULL);
1062 }
1063
1064 for (int l = 1; l < VDEV_LABELS; l++) {
1065 zio_nowait(zio_read_phys(pio, vd,
1066 vdev_label_offset(vd->vdev_psize, l,
1067 offsetof(vdev_label_t, vl_pad2)),
1068 VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE),
1069 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1070 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE));
1071 }
1072
1073 if (zio == NULL)
1074 return (pio);
1075
1076 zio_nowait(pio);
1077 return (NULL);
1078 }
1079
1080 static void
1081 vdev_open_child(void *arg)
1082 {
1083 vdev_t *vd = arg;
1084
1085 vd->vdev_open_thread = curthread;
1086 vd->vdev_open_error = vdev_open(vd);
1087 vd->vdev_open_thread = NULL;
1088 }
1089
1090 boolean_t
1091 vdev_uses_zvols(vdev_t *vd)
1092 {
1093 if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR,
1094 strlen(ZVOL_DIR)) == 0)
1095 return (B_TRUE);
1096 for (int c = 0; c < vd->vdev_children; c++)
1097 if (vdev_uses_zvols(vd->vdev_child[c]))
1098 return (B_TRUE);
1099 return (B_FALSE);
1100 }
1101
1102 void
1103 vdev_open_children(vdev_t *vd)
1104 {
1105 taskq_t *tq;
1106 int children = vd->vdev_children;
1107
1108 /*
1109 * in order to handle pools on top of zvols, do the opens
1110 * in a single thread so that the same thread holds the
1111 * spa_namespace_lock
1112 */
1113 if (vdev_uses_zvols(vd)) {
1114 for (int c = 0; c < children; c++)
1115 vd->vdev_child[c]->vdev_open_error =
1116 vdev_open(vd->vdev_child[c]);
1117 return;
1118 }
1119 tq = taskq_create("vdev_open", children, minclsyspri,
1120 children, children, TASKQ_PREPOPULATE);
1121
1122 for (int c = 0; c < children; c++)
1123 VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c],
1124 TQ_SLEEP) != NULL);
1125
1126 taskq_destroy(tq);
1127 }
1128
1129 /*
1130 * Prepare a virtual device for access.
1131 */
1132 int
1133 vdev_open(vdev_t *vd)
1134 {
1135 spa_t *spa = vd->vdev_spa;
1136 int error;
1137 uint64_t osize = 0;
1138 uint64_t max_osize = 0;
1139 uint64_t asize, max_asize, psize;
1140 uint64_t ashift = 0;
1141
1142 ASSERT(vd->vdev_open_thread == curthread ||
1143 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1144 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED ||
1145 vd->vdev_state == VDEV_STATE_CANT_OPEN ||
1146 vd->vdev_state == VDEV_STATE_OFFLINE);
1147
1148 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1149 vd->vdev_cant_read = B_FALSE;
1150 vd->vdev_cant_write = B_FALSE;
1151 vd->vdev_min_asize = vdev_get_min_asize(vd);
1152
1153 /*
1154 * If this vdev is not removed, check its fault status. If it's
1155 * faulted, bail out of the open.
1156 */
1157 if (!vd->vdev_removed && vd->vdev_faulted) {
1158 ASSERT(vd->vdev_children == 0);
1159 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1160 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1161 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1162 vd->vdev_label_aux);
1163 return (SET_ERROR(ENXIO));
1164 } else if (vd->vdev_offline) {
1165 ASSERT(vd->vdev_children == 0);
1166 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE);
1167 return (SET_ERROR(ENXIO));
1168 }
1169
1170 error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize, &ashift);
1171
1172 /*
1173 * Reset the vdev_reopening flag so that we actually close
1174 * the vdev on error.
1175 */
1176 vd->vdev_reopening = B_FALSE;
1177 if (zio_injection_enabled && error == 0)
1178 error = zio_handle_device_injection(vd, NULL, ENXIO);
1179
1180 if (error) {
1181 if (vd->vdev_removed &&
1182 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED)
1183 vd->vdev_removed = B_FALSE;
1184
1185 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1186 vd->vdev_stat.vs_aux);
1187 return (error);
1188 }
1189
1190 vd->vdev_removed = B_FALSE;
1191
1192 /*
1193 * Recheck the faulted flag now that we have confirmed that
1194 * the vdev is accessible. If we're faulted, bail.
1195 */
1196 if (vd->vdev_faulted) {
1197 ASSERT(vd->vdev_children == 0);
1198 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1199 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1200 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1201 vd->vdev_label_aux);
1202 return (SET_ERROR(ENXIO));
1203 }
1204
1205 if (vd->vdev_degraded) {
1206 ASSERT(vd->vdev_children == 0);
1207 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1208 VDEV_AUX_ERR_EXCEEDED);
1209 } else {
1210 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0);
1211 }
1212
1213 /*
1214 * For hole or missing vdevs we just return success.
1215 */
1216 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops)
1217 return (0);
1218
1219 for (int c = 0; c < vd->vdev_children; c++) {
1220 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) {
1221 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1222 VDEV_AUX_NONE);
1223 break;
1224 }
1225 }
1226
1227 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t));
1228 max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t));
1229
1230 if (vd->vdev_children == 0) {
1231 if (osize < SPA_MINDEVSIZE) {
1232 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1233 VDEV_AUX_TOO_SMALL);
1234 return (SET_ERROR(EOVERFLOW));
1235 }
1236 psize = osize;
1237 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE);
1238 max_asize = max_osize - (VDEV_LABEL_START_SIZE +
1239 VDEV_LABEL_END_SIZE);
1240 } else {
1241 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE -
1242 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) {
1243 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1244 VDEV_AUX_TOO_SMALL);
1245 return (SET_ERROR(EOVERFLOW));
1246 }
1247 psize = 0;
1248 asize = osize;
1249 max_asize = max_osize;
1250 }
1251
1252 vd->vdev_psize = psize;
1253
1254 /*
1255 * Make sure the allocatable size hasn't shrunk.
1256 */
1257 if (asize < vd->vdev_min_asize) {
1258 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1259 VDEV_AUX_BAD_LABEL);
1260 return (SET_ERROR(EINVAL));
1261 }
1262
1263 if (vd->vdev_asize == 0) {
1264 /*
1265 * This is the first-ever open, so use the computed values.
1266 * For testing purposes, a higher ashift can be requested.
1267 */
1268 vd->vdev_asize = asize;
1269 vd->vdev_max_asize = max_asize;
1270 vd->vdev_ashift = MAX(ashift, vd->vdev_ashift);
1271 } else {
1272 /*
1273 * Detect if the alignment requirement has increased.
1274 * We don't want to make the pool unavailable, just
1275 * issue a warning instead.
1276 */
1277 if (ashift > vd->vdev_top->vdev_ashift &&
1278 vd->vdev_ops->vdev_op_leaf) {
1279 cmn_err(CE_WARN,
1280 "Disk, '%s', has a block alignment that is "
1281 "larger than the pool's alignment\n",
1282 vd->vdev_path);
1283 }
1284 vd->vdev_max_asize = max_asize;
1285 }
1286
1287 /*
1288 * If all children are healthy and the asize has increased,
1289 * then we've experienced dynamic LUN growth. If automatic
1290 * expansion is enabled then use the additional space.
1291 */
1292 if (vd->vdev_state == VDEV_STATE_HEALTHY && asize > vd->vdev_asize &&
1293 (vd->vdev_expanding || spa->spa_autoexpand))
1294 vd->vdev_asize = asize;
1295
1296 vdev_set_min_asize(vd);
1297
1298 /*
1299 * Ensure we can issue some IO before declaring the
1300 * vdev open for business.
1301 */
1302 if (vd->vdev_ops->vdev_op_leaf &&
1303 (error = zio_wait(vdev_probe(vd, NULL))) != 0) {
1304 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1305 VDEV_AUX_ERR_EXCEEDED);
1306 return (error);
1307 }
1308
1309 /*
1310 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1311 * resilver. But don't do this if we are doing a reopen for a scrub,
1312 * since this would just restart the scrub we are already doing.
1313 */
1314 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen &&
1315 vdev_resilver_needed(vd, NULL, NULL))
1316 spa_async_request(spa, SPA_ASYNC_RESILVER);
1317
1318 return (0);
1319 }
1320
1321 /*
1322 * Called once the vdevs are all opened, this routine validates the label
1323 * contents. This needs to be done before vdev_load() so that we don't
1324 * inadvertently do repair I/Os to the wrong device.
1325 *
1326 * If 'strict' is false ignore the spa guid check. This is necessary because
1327 * if the machine crashed during a re-guid the new guid might have been written
1328 * to all of the vdev labels, but not the cached config. The strict check
1329 * will be performed when the pool is opened again using the mos config.
1330 *
1331 * This function will only return failure if one of the vdevs indicates that it
1332 * has since been destroyed or exported. This is only possible if
1333 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1334 * will be updated but the function will return 0.
1335 */
1336 int
1337 vdev_validate(vdev_t *vd, boolean_t strict)
1338 {
1339 spa_t *spa = vd->vdev_spa;
1340 nvlist_t *label;
1341 uint64_t guid = 0, top_guid;
1342 uint64_t state;
1343
1344 for (int c = 0; c < vd->vdev_children; c++)
1345 if (vdev_validate(vd->vdev_child[c], strict) != 0)
1346 return (SET_ERROR(EBADF));
1347
1348 /*
1349 * If the device has already failed, or was marked offline, don't do
1350 * any further validation. Otherwise, label I/O will fail and we will
1351 * overwrite the previous state.
1352 */
1353 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1354 uint64_t aux_guid = 0;
1355 nvlist_t *nvl;
1356 uint64_t txg = spa_last_synced_txg(spa) != 0 ?
1357 spa_last_synced_txg(spa) : -1ULL;
1358
1359 if ((label = vdev_label_read_config(vd, txg)) == NULL) {
1360 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1361 VDEV_AUX_BAD_LABEL);
1362 return (0);
1363 }
1364
1365 /*
1366 * Determine if this vdev has been split off into another
1367 * pool. If so, then refuse to open it.
1368 */
1369 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID,
1370 &aux_guid) == 0 && aux_guid == spa_guid(spa)) {
1371 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1372 VDEV_AUX_SPLIT_POOL);
1373 nvlist_free(label);
1374 return (0);
1375 }
1376
1377 if (strict && (nvlist_lookup_uint64(label,
1378 ZPOOL_CONFIG_POOL_GUID, &guid) != 0 ||
1379 guid != spa_guid(spa))) {
1380 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1381 VDEV_AUX_CORRUPT_DATA);
1382 nvlist_free(label);
1383 return (0);
1384 }
1385
1386 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl)
1387 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID,
1388 &aux_guid) != 0)
1389 aux_guid = 0;
1390
1391 /*
1392 * If this vdev just became a top-level vdev because its
1393 * sibling was detached, it will have adopted the parent's
1394 * vdev guid -- but the label may or may not be on disk yet.
1395 * Fortunately, either version of the label will have the
1396 * same top guid, so if we're a top-level vdev, we can
1397 * safely compare to that instead.
1398 *
1399 * If we split this vdev off instead, then we also check the
1400 * original pool's guid. We don't want to consider the vdev
1401 * corrupt if it is partway through a split operation.
1402 */
1403 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
1404 &guid) != 0 ||
1405 nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID,
1406 &top_guid) != 0 ||
1407 ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) &&
1408 (vd->vdev_guid != top_guid || vd != vd->vdev_top))) {
1409 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1410 VDEV_AUX_CORRUPT_DATA);
1411 nvlist_free(label);
1412 return (0);
1413 }
1414
1415 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1416 &state) != 0) {
1417 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1418 VDEV_AUX_CORRUPT_DATA);
1419 nvlist_free(label);
1420 return (0);
1421 }
1422
1423 nvlist_free(label);
1424
1425 /*
1426 * If this is a verbatim import, no need to check the
1427 * state of the pool.
1428 */
1429 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) &&
1430 spa_load_state(spa) == SPA_LOAD_OPEN &&
1431 state != POOL_STATE_ACTIVE)
1432 return (SET_ERROR(EBADF));
1433
1434 /*
1435 * If we were able to open and validate a vdev that was
1436 * previously marked permanently unavailable, clear that state
1437 * now.
1438 */
1439 if (vd->vdev_not_present)
1440 vd->vdev_not_present = 0;
1441 }
1442
1443 return (0);
1444 }
1445
1446 /*
1447 * Close a virtual device.
1448 */
1449 void
1450 vdev_close(vdev_t *vd)
1451 {
1452 spa_t *spa = vd->vdev_spa;
1453 vdev_t *pvd = vd->vdev_parent;
1454
1455 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1456
1457 /*
1458 * If our parent is reopening, then we are as well, unless we are
1459 * going offline.
1460 */
1461 if (pvd != NULL && pvd->vdev_reopening)
1462 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline);
1463
1464 vd->vdev_ops->vdev_op_close(vd);
1465
1466 vdev_cache_purge(vd);
1467
1468 /*
1469 * We record the previous state before we close it, so that if we are
1470 * doing a reopen(), we don't generate FMA ereports if we notice that
1471 * it's still faulted.
1472 */
1473 vd->vdev_prevstate = vd->vdev_state;
1474
1475 if (vd->vdev_offline)
1476 vd->vdev_state = VDEV_STATE_OFFLINE;
1477 else
1478 vd->vdev_state = VDEV_STATE_CLOSED;
1479 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1480 }
1481
1482 void
1483 vdev_hold(vdev_t *vd)
1484 {
1485 spa_t *spa = vd->vdev_spa;
1486
1487 ASSERT(spa_is_root(spa));
1488 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1489 return;
1490
1491 for (int c = 0; c < vd->vdev_children; c++)
1492 vdev_hold(vd->vdev_child[c]);
1493
1494 if (vd->vdev_ops->vdev_op_leaf)
1495 vd->vdev_ops->vdev_op_hold(vd);
1496 }
1497
1498 void
1499 vdev_rele(vdev_t *vd)
1500 {
1501 spa_t *spa = vd->vdev_spa;
1502
1503 ASSERT(spa_is_root(spa));
1504 for (int c = 0; c < vd->vdev_children; c++)
1505 vdev_rele(vd->vdev_child[c]);
1506
1507 if (vd->vdev_ops->vdev_op_leaf)
1508 vd->vdev_ops->vdev_op_rele(vd);
1509 }
1510
1511 /*
1512 * Reopen all interior vdevs and any unopened leaves. We don't actually
1513 * reopen leaf vdevs which had previously been opened as they might deadlock
1514 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1515 * If the leaf has never been opened then open it, as usual.
1516 */
1517 void
1518 vdev_reopen(vdev_t *vd)
1519 {
1520 spa_t *spa = vd->vdev_spa;
1521
1522 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1523
1524 /* set the reopening flag unless we're taking the vdev offline */
1525 vd->vdev_reopening = !vd->vdev_offline;
1526 vdev_close(vd);
1527 (void) vdev_open(vd);
1528
1529 /*
1530 * Call vdev_validate() here to make sure we have the same device.
1531 * Otherwise, a device with an invalid label could be successfully
1532 * opened in response to vdev_reopen().
1533 */
1534 if (vd->vdev_aux) {
1535 (void) vdev_validate_aux(vd);
1536 if (vdev_readable(vd) && vdev_writeable(vd) &&
1537 vd->vdev_aux == &spa->spa_l2cache &&
1538 !l2arc_vdev_present(vd))
1539 l2arc_add_vdev(spa, vd);
1540 } else {
1541 (void) vdev_validate(vd, B_TRUE);
1542 }
1543
1544 /*
1545 * Reassess parent vdev's health.
1546 */
1547 vdev_propagate_state(vd);
1548 }
1549
1550 int
1551 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
1552 {
1553 int error;
1554
1555 /*
1556 * Normally, partial opens (e.g. of a mirror) are allowed.
1557 * For a create, however, we want to fail the request if
1558 * there are any components we can't open.
1559 */
1560 error = vdev_open(vd);
1561
1562 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) {
1563 vdev_close(vd);
1564 return (error ? error : ENXIO);
1565 }
1566
1567 /*
1568 * Recursively load DTLs and initialize all labels.
1569 */
1570 if ((error = vdev_dtl_load(vd)) != 0 ||
1571 (error = vdev_label_init(vd, txg, isreplacing ?
1572 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
1573 vdev_close(vd);
1574 return (error);
1575 }
1576
1577 return (0);
1578 }
1579
1580 void
1581 vdev_metaslab_set_size(vdev_t *vd)
1582 {
1583 /*
1584 * Aim for roughly metaslabs_per_vdev (default 200) metaslabs per vdev.
1585 */
1586 vd->vdev_ms_shift = highbit64(vd->vdev_asize / metaslabs_per_vdev);
1587 vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT);
1588 }
1589
1590 void
1591 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
1592 {
1593 ASSERT(vd == vd->vdev_top);
1594 ASSERT(!vd->vdev_ishole);
1595 ASSERT(ISP2(flags));
1596 ASSERT(spa_writeable(vd->vdev_spa));
1597
1598 if (flags & VDD_METASLAB)
1599 (void) txg_list_add(&vd->vdev_ms_list, arg, txg);
1600
1601 if (flags & VDD_DTL)
1602 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg);
1603
1604 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
1605 }
1606
1607 void
1608 vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg)
1609 {
1610 for (int c = 0; c < vd->vdev_children; c++)
1611 vdev_dirty_leaves(vd->vdev_child[c], flags, txg);
1612
1613 if (vd->vdev_ops->vdev_op_leaf)
1614 vdev_dirty(vd->vdev_top, flags, vd, txg);
1615 }
1616
1617 /*
1618 * DTLs.
1619 *
1620 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1621 * the vdev has less than perfect replication. There are four kinds of DTL:
1622 *
1623 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1624 *
1625 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1626 *
1627 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1628 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1629 * txgs that was scrubbed.
1630 *
1631 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1632 * persistent errors or just some device being offline.
1633 * Unlike the other three, the DTL_OUTAGE map is not generally
1634 * maintained; it's only computed when needed, typically to
1635 * determine whether a device can be detached.
1636 *
1637 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1638 * either has the data or it doesn't.
1639 *
1640 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1641 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1642 * if any child is less than fully replicated, then so is its parent.
1643 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1644 * comprising only those txgs which appear in 'maxfaults' or more children;
1645 * those are the txgs we don't have enough replication to read. For example,
1646 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1647 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1648 * two child DTL_MISSING maps.
1649 *
1650 * It should be clear from the above that to compute the DTLs and outage maps
1651 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1652 * Therefore, that is all we keep on disk. When loading the pool, or after
1653 * a configuration change, we generate all other DTLs from first principles.
1654 */
1655 void
1656 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1657 {
1658 range_tree_t *rt = vd->vdev_dtl[t];
1659
1660 ASSERT(t < DTL_TYPES);
1661 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1662 ASSERT(spa_writeable(vd->vdev_spa));
1663
1664 mutex_enter(rt->rt_lock);
1665 if (!range_tree_contains(rt, txg, size))
1666 range_tree_add(rt, txg, size);
1667 mutex_exit(rt->rt_lock);
1668 }
1669
1670 boolean_t
1671 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1672 {
1673 range_tree_t *rt = vd->vdev_dtl[t];
1674 boolean_t dirty = B_FALSE;
1675
1676 ASSERT(t < DTL_TYPES);
1677 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1678
1679 mutex_enter(rt->rt_lock);
1680 if (range_tree_space(rt) != 0)
1681 dirty = range_tree_contains(rt, txg, size);
1682 mutex_exit(rt->rt_lock);
1683
1684 return (dirty);
1685 }
1686
1687 boolean_t
1688 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t)
1689 {
1690 range_tree_t *rt = vd->vdev_dtl[t];
1691 boolean_t empty;
1692
1693 mutex_enter(rt->rt_lock);
1694 empty = (range_tree_space(rt) == 0);
1695 mutex_exit(rt->rt_lock);
1696
1697 return (empty);
1698 }
1699
1700 /*
1701 * Returns the lowest txg in the DTL range.
1702 */
1703 static uint64_t
1704 vdev_dtl_min(vdev_t *vd)
1705 {
1706 range_seg_t *rs;
1707
1708 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
1709 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
1710 ASSERT0(vd->vdev_children);
1711
1712 rs = avl_first(&vd->vdev_dtl[DTL_MISSING]->rt_root);
1713 return (rs->rs_start - 1);
1714 }
1715
1716 /*
1717 * Returns the highest txg in the DTL.
1718 */
1719 static uint64_t
1720 vdev_dtl_max(vdev_t *vd)
1721 {
1722 range_seg_t *rs;
1723
1724 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
1725 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
1726 ASSERT0(vd->vdev_children);
1727
1728 rs = avl_last(&vd->vdev_dtl[DTL_MISSING]->rt_root);
1729 return (rs->rs_end);
1730 }
1731
1732 /*
1733 * Determine if a resilvering vdev should remove any DTL entries from
1734 * its range. If the vdev was resilvering for the entire duration of the
1735 * scan then it should excise that range from its DTLs. Otherwise, this
1736 * vdev is considered partially resilvered and should leave its DTL
1737 * entries intact. The comment in vdev_dtl_reassess() describes how we
1738 * excise the DTLs.
1739 */
1740 static boolean_t
1741 vdev_dtl_should_excise(vdev_t *vd)
1742 {
1743 spa_t *spa = vd->vdev_spa;
1744 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
1745
1746 ASSERT0(scn->scn_phys.scn_errors);
1747 ASSERT0(vd->vdev_children);
1748
1749 if (vd->vdev_resilver_txg == 0 ||
1750 range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0)
1751 return (B_TRUE);
1752
1753 /*
1754 * When a resilver is initiated the scan will assign the scn_max_txg
1755 * value to the highest txg value that exists in all DTLs. If this
1756 * device's max DTL is not part of this scan (i.e. it is not in
1757 * the range (scn_min_txg, scn_max_txg] then it is not eligible
1758 * for excision.
1759 */
1760 if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) {
1761 ASSERT3U(scn->scn_phys.scn_min_txg, <=, vdev_dtl_min(vd));
1762 ASSERT3U(scn->scn_phys.scn_min_txg, <, vd->vdev_resilver_txg);
1763 ASSERT3U(vd->vdev_resilver_txg, <=, scn->scn_phys.scn_max_txg);
1764 return (B_TRUE);
1765 }
1766 return (B_FALSE);
1767 }
1768
1769 /*
1770 * Reassess DTLs after a config change or scrub completion.
1771 */
1772 void
1773 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
1774 {
1775 spa_t *spa = vd->vdev_spa;
1776 avl_tree_t reftree;
1777 int minref;
1778
1779 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1780
1781 for (int c = 0; c < vd->vdev_children; c++)
1782 vdev_dtl_reassess(vd->vdev_child[c], txg,
1783 scrub_txg, scrub_done);
1784
1785 if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux)
1786 return;
1787
1788 if (vd->vdev_ops->vdev_op_leaf) {
1789 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
1790
1791 mutex_enter(&vd->vdev_dtl_lock);
1792
1793 /*
1794 * If we've completed a scan cleanly then determine
1795 * if this vdev should remove any DTLs. We only want to
1796 * excise regions on vdevs that were available during
1797 * the entire duration of this scan.
1798 */
1799 if (scrub_txg != 0 &&
1800 (spa->spa_scrub_started ||
1801 (scn != NULL && scn->scn_phys.scn_errors == 0)) &&
1802 vdev_dtl_should_excise(vd)) {
1803 /*
1804 * We completed a scrub up to scrub_txg. If we
1805 * did it without rebooting, then the scrub dtl
1806 * will be valid, so excise the old region and
1807 * fold in the scrub dtl. Otherwise, leave the
1808 * dtl as-is if there was an error.
1809 *
1810 * There's little trick here: to excise the beginning
1811 * of the DTL_MISSING map, we put it into a reference
1812 * tree and then add a segment with refcnt -1 that
1813 * covers the range [0, scrub_txg). This means
1814 * that each txg in that range has refcnt -1 or 0.
1815 * We then add DTL_SCRUB with a refcnt of 2, so that
1816 * entries in the range [0, scrub_txg) will have a
1817 * positive refcnt -- either 1 or 2. We then convert
1818 * the reference tree into the new DTL_MISSING map.
1819 */
1820 space_reftree_create(&reftree);
1821 space_reftree_add_map(&reftree,
1822 vd->vdev_dtl[DTL_MISSING], 1);
1823 space_reftree_add_seg(&reftree, 0, scrub_txg, -1);
1824 space_reftree_add_map(&reftree,
1825 vd->vdev_dtl[DTL_SCRUB], 2);
1826 space_reftree_generate_map(&reftree,
1827 vd->vdev_dtl[DTL_MISSING], 1);
1828 space_reftree_destroy(&reftree);
1829 }
1830 range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
1831 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
1832 range_tree_add, vd->vdev_dtl[DTL_PARTIAL]);
1833 if (scrub_done)
1834 range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
1835 range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
1836 if (!vdev_readable(vd))
1837 range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
1838 else
1839 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
1840 range_tree_add, vd->vdev_dtl[DTL_OUTAGE]);
1841
1842 /*
1843 * If the vdev was resilvering and no longer has any
1844 * DTLs then reset its resilvering flag.
1845 */
1846 if (vd->vdev_resilver_txg != 0 &&
1847 range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0 &&
1848 range_tree_space(vd->vdev_dtl[DTL_OUTAGE]) == 0)
1849 vd->vdev_resilver_txg = 0;
1850
1851 mutex_exit(&vd->vdev_dtl_lock);
1852
1853 if (txg != 0)
1854 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
1855 return;
1856 }
1857
1858 mutex_enter(&vd->vdev_dtl_lock);
1859 for (int t = 0; t < DTL_TYPES; t++) {
1860 /* account for child's outage in parent's missing map */
1861 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t;
1862 if (t == DTL_SCRUB)
1863 continue; /* leaf vdevs only */
1864 if (t == DTL_PARTIAL)
1865 minref = 1; /* i.e. non-zero */
1866 else if (vd->vdev_nparity != 0)
1867 minref = vd->vdev_nparity + 1; /* RAID-Z */
1868 else
1869 minref = vd->vdev_children; /* any kind of mirror */
1870 space_reftree_create(&reftree);
1871 for (int c = 0; c < vd->vdev_children; c++) {
1872 vdev_t *cvd = vd->vdev_child[c];
1873 mutex_enter(&cvd->vdev_dtl_lock);
1874 space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1);
1875 mutex_exit(&cvd->vdev_dtl_lock);
1876 }
1877 space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref);
1878 space_reftree_destroy(&reftree);
1879 }
1880 mutex_exit(&vd->vdev_dtl_lock);
1881 }
1882
1883 int
1884 vdev_dtl_load(vdev_t *vd)
1885 {
1886 spa_t *spa = vd->vdev_spa;
1887 objset_t *mos = spa->spa_meta_objset;
1888 int error = 0;
1889
1890 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) {
1891 ASSERT(!vd->vdev_ishole);
1892
1893 error = space_map_open(&vd->vdev_dtl_sm, mos,
1894 vd->vdev_dtl_object, 0, -1ULL, 0, &vd->vdev_dtl_lock);
1895 if (error)
1896 return (error);
1897 ASSERT(vd->vdev_dtl_sm != NULL);
1898
1899 mutex_enter(&vd->vdev_dtl_lock);
1900
1901 /*
1902 * Now that we've opened the space_map we need to update
1903 * the in-core DTL.
1904 */
1905 space_map_update(vd->vdev_dtl_sm);
1906
1907 error = space_map_load(vd->vdev_dtl_sm,
1908 vd->vdev_dtl[DTL_MISSING], SM_ALLOC);
1909 mutex_exit(&vd->vdev_dtl_lock);
1910
1911 return (error);
1912 }
1913
1914 for (int c = 0; c < vd->vdev_children; c++) {
1915 error = vdev_dtl_load(vd->vdev_child[c]);
1916 if (error != 0)
1917 break;
1918 }
1919
1920 return (error);
1921 }
1922
1923 void
1924 vdev_dtl_sync(vdev_t *vd, uint64_t txg)
1925 {
1926 spa_t *spa = vd->vdev_spa;
1927 range_tree_t *rt = vd->vdev_dtl[DTL_MISSING];
1928 objset_t *mos = spa->spa_meta_objset;
1929 range_tree_t *rtsync;
1930 kmutex_t rtlock;
1931 dmu_tx_t *tx;
1932 uint64_t object = space_map_object(vd->vdev_dtl_sm);
1933
1934 ASSERT(!vd->vdev_ishole);
1935 ASSERT(vd->vdev_ops->vdev_op_leaf);
1936
1937 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
1938
1939 if (vd->vdev_detached || vd->vdev_top->vdev_removing) {
1940 mutex_enter(&vd->vdev_dtl_lock);
1941 space_map_free(vd->vdev_dtl_sm, tx);
1942 space_map_close(vd->vdev_dtl_sm);
1943 vd->vdev_dtl_sm = NULL;
1944 mutex_exit(&vd->vdev_dtl_lock);
1945 dmu_tx_commit(tx);
1946 return;
1947 }
1948
1949 if (vd->vdev_dtl_sm == NULL) {
1950 uint64_t new_object;
1951
1952 new_object = space_map_alloc(mos, tx);
1953 VERIFY3U(new_object, !=, 0);
1954
1955 VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object,
1956 0, -1ULL, 0, &vd->vdev_dtl_lock));
1957 ASSERT(vd->vdev_dtl_sm != NULL);
1958 }
1959
1960 mutex_init(&rtlock, NULL, MUTEX_DEFAULT, NULL);
1961
1962 rtsync = range_tree_create(NULL, NULL, &rtlock);
1963
1964 mutex_enter(&rtlock);
1965
1966 mutex_enter(&vd->vdev_dtl_lock);
1967 range_tree_walk(rt, range_tree_add, rtsync);
1968 mutex_exit(&vd->vdev_dtl_lock);
1969
1970 space_map_truncate(vd->vdev_dtl_sm, tx);
1971 space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, tx);
1972 range_tree_vacate(rtsync, NULL, NULL);
1973
1974 range_tree_destroy(rtsync);
1975
1976 mutex_exit(&rtlock);
1977 mutex_destroy(&rtlock);
1978
1979 /*
1980 * If the object for the space map has changed then dirty
1981 * the top level so that we update the config.
1982 */
1983 if (object != space_map_object(vd->vdev_dtl_sm)) {
1984 zfs_dbgmsg("txg %llu, spa %s, DTL old object %llu, "
1985 "new object %llu", txg, spa_name(spa), object,
1986 space_map_object(vd->vdev_dtl_sm));
1987 vdev_config_dirty(vd->vdev_top);
1988 }
1989
1990 dmu_tx_commit(tx);
1991
1992 mutex_enter(&vd->vdev_dtl_lock);
1993 space_map_update(vd->vdev_dtl_sm);
1994 mutex_exit(&vd->vdev_dtl_lock);
1995 }
1996
1997 /*
1998 * Determine whether the specified vdev can be offlined/detached/removed
1999 * without losing data.
2000 */
2001 boolean_t
2002 vdev_dtl_required(vdev_t *vd)
2003 {
2004 spa_t *spa = vd->vdev_spa;
2005 vdev_t *tvd = vd->vdev_top;
2006 uint8_t cant_read = vd->vdev_cant_read;
2007 boolean_t required;
2008
2009 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2010
2011 if (vd == spa->spa_root_vdev || vd == tvd)
2012 return (B_TRUE);
2013
2014 /*
2015 * Temporarily mark the device as unreadable, and then determine
2016 * whether this results in any DTL outages in the top-level vdev.
2017 * If not, we can safely offline/detach/remove the device.
2018 */
2019 vd->vdev_cant_read = B_TRUE;
2020 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
2021 required = !vdev_dtl_empty(tvd, DTL_OUTAGE);
2022 vd->vdev_cant_read = cant_read;
2023 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
2024
2025 if (!required && zio_injection_enabled)
2026 required = !!zio_handle_device_injection(vd, NULL, ECHILD);
2027
2028 return (required);
2029 }
2030
2031 /*
2032 * Determine if resilver is needed, and if so the txg range.
2033 */
2034 boolean_t
2035 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
2036 {
2037 boolean_t needed = B_FALSE;
2038 uint64_t thismin = UINT64_MAX;
2039 uint64_t thismax = 0;
2040
2041 if (vd->vdev_children == 0) {
2042 mutex_enter(&vd->vdev_dtl_lock);
2043 if (range_tree_space(vd->vdev_dtl[DTL_MISSING]) != 0 &&
2044 vdev_writeable(vd)) {
2045
2046 thismin = vdev_dtl_min(vd);
2047 thismax = vdev_dtl_max(vd);
2048 needed = B_TRUE;
2049 }
2050 mutex_exit(&vd->vdev_dtl_lock);
2051 } else {
2052 for (int c = 0; c < vd->vdev_children; c++) {
2053 vdev_t *cvd = vd->vdev_child[c];
2054 uint64_t cmin, cmax;
2055
2056 if (vdev_resilver_needed(cvd, &cmin, &cmax)) {
2057 thismin = MIN(thismin, cmin);
2058 thismax = MAX(thismax, cmax);
2059 needed = B_TRUE;
2060 }
2061 }
2062 }
2063
2064 if (needed && minp) {
2065 *minp = thismin;
2066 *maxp = thismax;
2067 }
2068 return (needed);
2069 }
2070
2071 void
2072 vdev_load(vdev_t *vd)
2073 {
2074 /*
2075 * Recursively load all children.
2076 */
2077 for (int c = 0; c < vd->vdev_children; c++)
2078 vdev_load(vd->vdev_child[c]);
2079
2080 /*
2081 * If this is a top-level vdev, initialize its metaslabs.
2082 */
2083 if (vd == vd->vdev_top && !vd->vdev_ishole &&
2084 (vd->vdev_ashift == 0 || vd->vdev_asize == 0 ||
2085 vdev_metaslab_init(vd, 0) != 0))
2086 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2087 VDEV_AUX_CORRUPT_DATA);
2088
2089 /*
2090 * If this is a leaf vdev, load its DTL.
2091 */
2092 if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0)
2093 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2094 VDEV_AUX_CORRUPT_DATA);
2095 }
2096
2097 /*
2098 * The special vdev case is used for hot spares and l2cache devices. Its
2099 * sole purpose it to set the vdev state for the associated vdev. To do this,
2100 * we make sure that we can open the underlying device, then try to read the
2101 * label, and make sure that the label is sane and that it hasn't been
2102 * repurposed to another pool.
2103 */
2104 int
2105 vdev_validate_aux(vdev_t *vd)
2106 {
2107 nvlist_t *label;
2108 uint64_t guid, version;
2109 uint64_t state;
2110
2111 if (!vdev_readable(vd))
2112 return (0);
2113
2114 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) {
2115 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2116 VDEV_AUX_CORRUPT_DATA);
2117 return (-1);
2118 }
2119
2120 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 ||
2121 !SPA_VERSION_IS_SUPPORTED(version) ||
2122 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 ||
2123 guid != vd->vdev_guid ||
2124 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) {
2125 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2126 VDEV_AUX_CORRUPT_DATA);
2127 nvlist_free(label);
2128 return (-1);
2129 }
2130
2131 /*
2132 * We don't actually check the pool state here. If it's in fact in
2133 * use by another pool, we update this fact on the fly when requested.
2134 */
2135 nvlist_free(label);
2136 return (0);
2137 }
2138
2139 void
2140 vdev_remove(vdev_t *vd, uint64_t txg)
2141 {
2142 spa_t *spa = vd->vdev_spa;
2143 objset_t *mos = spa->spa_meta_objset;
2144 dmu_tx_t *tx;
2145
2146 tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
2147
2148 if (vd->vdev_ms != NULL) {
2149 metaslab_group_t *mg = vd->vdev_mg;
2150
2151 metaslab_group_histogram_verify(mg);
2152 metaslab_class_histogram_verify(mg->mg_class);
2153
2154 for (int m = 0; m < vd->vdev_ms_count; m++) {
2155 metaslab_t *msp = vd->vdev_ms[m];
2156
2157 if (msp == NULL || msp->ms_sm == NULL)
2158 continue;
2159
2160 mutex_enter(&msp->ms_lock);
2161 /*
2162 * If the metaslab was not loaded when the vdev
2163 * was removed then the histogram accounting may
2164 * not be accurate. Update the histogram information
2165 * here so that we ensure that the metaslab group
2166 * and metaslab class are up-to-date.
2167 */
2168 metaslab_group_histogram_remove(mg, msp);
2169
2170 VERIFY0(space_map_allocated(msp->ms_sm));
2171 space_map_free(msp->ms_sm, tx);
2172 space_map_close(msp->ms_sm);
2173 msp->ms_sm = NULL;
2174 mutex_exit(&msp->ms_lock);
2175 }
2176
2177 metaslab_group_histogram_verify(mg);
2178 metaslab_class_histogram_verify(mg->mg_class);
2179 for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++)
2180 ASSERT0(mg->mg_histogram[i]);
2181
2182 }
2183
2184 if (vd->vdev_ms_array) {
2185 (void) dmu_object_free(mos, vd->vdev_ms_array, tx);
2186 vd->vdev_ms_array = 0;
2187 }
2188 dmu_tx_commit(tx);
2189 }
2190
2191 void
2192 vdev_sync_done(vdev_t *vd, uint64_t txg)
2193 {
2194 metaslab_t *msp;
2195 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg));
2196
2197 ASSERT(!vd->vdev_ishole);
2198
2199 while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))
2200 metaslab_sync_done(msp, txg);
2201
2202 if (reassess)
2203 metaslab_sync_reassess(vd->vdev_mg);
2204 }
2205
2206 void
2207 vdev_sync(vdev_t *vd, uint64_t txg)
2208 {
2209 spa_t *spa = vd->vdev_spa;
2210 vdev_t *lvd;
2211 metaslab_t *msp;
2212 dmu_tx_t *tx;
2213
2214 ASSERT(!vd->vdev_ishole);
2215
2216 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) {
2217 ASSERT(vd == vd->vdev_top);
2218 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2219 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset,
2220 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx);
2221 ASSERT(vd->vdev_ms_array != 0);
2222 vdev_config_dirty(vd);
2223 dmu_tx_commit(tx);
2224 }
2225
2226 /*
2227 * Remove the metadata associated with this vdev once it's empty.
2228 */
2229 if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing)
2230 vdev_remove(vd, txg);
2231
2232 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) {
2233 metaslab_sync(msp, txg);
2234 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg));
2235 }
2236
2237 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL)
2238 vdev_dtl_sync(lvd, txg);
2239
2240 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg));
2241 }
2242
2243 uint64_t
2244 vdev_psize_to_asize(vdev_t *vd, uint64_t psize)
2245 {
2246 return (vd->vdev_ops->vdev_op_asize(vd, psize));
2247 }
2248
2249 /*
2250 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2251 * not be opened, and no I/O is attempted.
2252 */
2253 int
2254 vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2255 {
2256 vdev_t *vd, *tvd;
2257
2258 spa_vdev_state_enter(spa, SCL_NONE);
2259
2260 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2261 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2262
2263 if (!vd->vdev_ops->vdev_op_leaf)
2264 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2265
2266 tvd = vd->vdev_top;
2267
2268 /*
2269 * We don't directly use the aux state here, but if we do a
2270 * vdev_reopen(), we need this value to be present to remember why we
2271 * were faulted.
2272 */
2273 vd->vdev_label_aux = aux;
2274
2275 /*
2276 * Faulted state takes precedence over degraded.
2277 */
2278 vd->vdev_delayed_close = B_FALSE;
2279 vd->vdev_faulted = 1ULL;
2280 vd->vdev_degraded = 0ULL;
2281 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux);
2282
2283 /*
2284 * If this device has the only valid copy of the data, then
2285 * back off and simply mark the vdev as degraded instead.
2286 */
2287 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) {
2288 vd->vdev_degraded = 1ULL;
2289 vd->vdev_faulted = 0ULL;
2290
2291 /*
2292 * If we reopen the device and it's not dead, only then do we
2293 * mark it degraded.
2294 */
2295 vdev_reopen(tvd);
2296
2297 if (vdev_readable(vd))
2298 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux);
2299 }
2300
2301 return (spa_vdev_state_exit(spa, vd, 0));
2302 }
2303
2304 /*
2305 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2306 * user that something is wrong. The vdev continues to operate as normal as far
2307 * as I/O is concerned.
2308 */
2309 int
2310 vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2311 {
2312 vdev_t *vd;
2313
2314 spa_vdev_state_enter(spa, SCL_NONE);
2315
2316 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2317 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2318
2319 if (!vd->vdev_ops->vdev_op_leaf)
2320 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2321
2322 /*
2323 * If the vdev is already faulted, then don't do anything.
2324 */
2325 if (vd->vdev_faulted || vd->vdev_degraded)
2326 return (spa_vdev_state_exit(spa, NULL, 0));
2327
2328 vd->vdev_degraded = 1ULL;
2329 if (!vdev_is_dead(vd))
2330 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
2331 aux);
2332
2333 return (spa_vdev_state_exit(spa, vd, 0));
2334 }
2335
2336 /*
2337 * Online the given vdev.
2338 *
2339 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
2340 * spare device should be detached when the device finishes resilvering.
2341 * Second, the online should be treated like a 'test' online case, so no FMA
2342 * events are generated if the device fails to open.
2343 */
2344 int
2345 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
2346 {
2347 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev;
2348
2349 spa_vdev_state_enter(spa, SCL_NONE);
2350
2351 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2352 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2353
2354 if (!vd->vdev_ops->vdev_op_leaf)
2355 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2356
2357 tvd = vd->vdev_top;
2358 vd->vdev_offline = B_FALSE;
2359 vd->vdev_tmpoffline = B_FALSE;
2360 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE);
2361 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT);
2362
2363 /* XXX - L2ARC 1.0 does not support expansion */
2364 if (!vd->vdev_aux) {
2365 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2366 pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND);
2367 }
2368
2369 vdev_reopen(tvd);
2370 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE;
2371
2372 if (!vd->vdev_aux) {
2373 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2374 pvd->vdev_expanding = B_FALSE;
2375 }
2376
2377 if (newstate)
2378 *newstate = vd->vdev_state;
2379 if ((flags & ZFS_ONLINE_UNSPARE) &&
2380 !vdev_is_dead(vd) && vd->vdev_parent &&
2381 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2382 vd->vdev_parent->vdev_child[0] == vd)
2383 vd->vdev_unspare = B_TRUE;
2384
2385 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) {
2386
2387 /* XXX - L2ARC 1.0 does not support expansion */
2388 if (vd->vdev_aux)
2389 return (spa_vdev_state_exit(spa, vd, ENOTSUP));
2390 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2391 }
2392 return (spa_vdev_state_exit(spa, vd, 0));
2393 }
2394
2395 static int
2396 vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags)
2397 {
2398 vdev_t *vd, *tvd;
2399 int error = 0;
2400 uint64_t generation;
2401 metaslab_group_t *mg;
2402
2403 top:
2404 spa_vdev_state_enter(spa, SCL_ALLOC);
2405
2406 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2407 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2408
2409 if (!vd->vdev_ops->vdev_op_leaf)
2410 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2411
2412 tvd = vd->vdev_top;
2413 mg = tvd->vdev_mg;
2414 generation = spa->spa_config_generation + 1;
2415
2416 /*
2417 * If the device isn't already offline, try to offline it.
2418 */
2419 if (!vd->vdev_offline) {
2420 /*
2421 * If this device has the only valid copy of some data,
2422 * don't allow it to be offlined. Log devices are always
2423 * expendable.
2424 */
2425 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2426 vdev_dtl_required(vd))
2427 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2428
2429 /*
2430 * If the top-level is a slog and it has had allocations
2431 * then proceed. We check that the vdev's metaslab group
2432 * is not NULL since it's possible that we may have just
2433 * added this vdev but not yet initialized its metaslabs.
2434 */
2435 if (tvd->vdev_islog && mg != NULL) {
2436 /*
2437 * Prevent any future allocations.
2438 */
2439 metaslab_group_passivate(mg);
2440 (void) spa_vdev_state_exit(spa, vd, 0);
2441
2442 error = spa_offline_log(spa);
2443
2444 spa_vdev_state_enter(spa, SCL_ALLOC);
2445
2446 /*
2447 * Check to see if the config has changed.
2448 */
2449 if (error || generation != spa->spa_config_generation) {
2450 metaslab_group_activate(mg);
2451 if (error)
2452 return (spa_vdev_state_exit(spa,
2453 vd, error));
2454 (void) spa_vdev_state_exit(spa, vd, 0);
2455 goto top;
2456 }
2457 ASSERT0(tvd->vdev_stat.vs_alloc);
2458 }
2459
2460 /*
2461 * Offline this device and reopen its top-level vdev.
2462 * If the top-level vdev is a log device then just offline
2463 * it. Otherwise, if this action results in the top-level
2464 * vdev becoming unusable, undo it and fail the request.
2465 */
2466 vd->vdev_offline = B_TRUE;
2467 vdev_reopen(tvd);
2468
2469 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2470 vdev_is_dead(tvd)) {
2471 vd->vdev_offline = B_FALSE;
2472 vdev_reopen(tvd);
2473 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2474 }
2475
2476 /*
2477 * Add the device back into the metaslab rotor so that
2478 * once we online the device it's open for business.
2479 */
2480 if (tvd->vdev_islog && mg != NULL)
2481 metaslab_group_activate(mg);
2482 }
2483
2484 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY);
2485
2486 return (spa_vdev_state_exit(spa, vd, 0));
2487 }
2488
2489 int
2490 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags)
2491 {
2492 int error;
2493
2494 mutex_enter(&spa->spa_vdev_top_lock);
2495 error = vdev_offline_locked(spa, guid, flags);
2496 mutex_exit(&spa->spa_vdev_top_lock);
2497
2498 return (error);
2499 }
2500
2501 /*
2502 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2503 * vdev_offline(), we assume the spa config is locked. We also clear all
2504 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2505 */
2506 void
2507 vdev_clear(spa_t *spa, vdev_t *vd)
2508 {
2509 vdev_t *rvd = spa->spa_root_vdev;
2510
2511 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2512
2513 if (vd == NULL)
2514 vd = rvd;
2515
2516 vd->vdev_stat.vs_read_errors = 0;
2517 vd->vdev_stat.vs_write_errors = 0;
2518 vd->vdev_stat.vs_checksum_errors = 0;
2519
2520 for (int c = 0; c < vd->vdev_children; c++)
2521 vdev_clear(spa, vd->vdev_child[c]);
2522
2523 /*
2524 * If we're in the FAULTED state or have experienced failed I/O, then
2525 * clear the persistent state and attempt to reopen the device. We
2526 * also mark the vdev config dirty, so that the new faulted state is
2527 * written out to disk.
2528 */
2529 if (vd->vdev_faulted || vd->vdev_degraded ||
2530 !vdev_readable(vd) || !vdev_writeable(vd)) {
2531
2532 /*
2533 * When reopening in reponse to a clear event, it may be due to
2534 * a fmadm repair request. In this case, if the device is
2535 * still broken, we want to still post the ereport again.
2536 */
2537 vd->vdev_forcefault = B_TRUE;
2538
2539 vd->vdev_faulted = vd->vdev_degraded = 0ULL;
2540 vd->vdev_cant_read = B_FALSE;
2541 vd->vdev_cant_write = B_FALSE;
2542
2543 vdev_reopen(vd == rvd ? rvd : vd->vdev_top);
2544
2545 vd->vdev_forcefault = B_FALSE;
2546
2547 if (vd != rvd && vdev_writeable(vd->vdev_top))
2548 vdev_state_dirty(vd->vdev_top);
2549
2550 if (vd->vdev_aux == NULL && !vdev_is_dead(vd))
2551 spa_async_request(spa, SPA_ASYNC_RESILVER);
2552
2553 spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR);
2554 }
2555
2556 /*
2557 * When clearing a FMA-diagnosed fault, we always want to
2558 * unspare the device, as we assume that the original spare was
2559 * done in response to the FMA fault.
2560 */
2561 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL &&
2562 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2563 vd->vdev_parent->vdev_child[0] == vd)
2564 vd->vdev_unspare = B_TRUE;
2565 }
2566
2567 boolean_t
2568 vdev_is_dead(vdev_t *vd)
2569 {
2570 /*
2571 * Holes and missing devices are always considered "dead".
2572 * This simplifies the code since we don't have to check for
2573 * these types of devices in the various code paths.
2574 * Instead we rely on the fact that we skip over dead devices
2575 * before issuing I/O to them.
2576 */
2577 return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole ||
2578 vd->vdev_ops == &vdev_missing_ops);
2579 }
2580
2581 boolean_t
2582 vdev_readable(vdev_t *vd)
2583 {
2584 return (!vdev_is_dead(vd) && !vd->vdev_cant_read);
2585 }
2586
2587 boolean_t
2588 vdev_writeable(vdev_t *vd)
2589 {
2590 return (!vdev_is_dead(vd) && !vd->vdev_cant_write);
2591 }
2592
2593 boolean_t
2594 vdev_allocatable(vdev_t *vd)
2595 {
2596 uint64_t state = vd->vdev_state;
2597
2598 /*
2599 * We currently allow allocations from vdevs which may be in the
2600 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2601 * fails to reopen then we'll catch it later when we're holding
2602 * the proper locks. Note that we have to get the vdev state
2603 * in a local variable because although it changes atomically,
2604 * we're asking two separate questions about it.
2605 */
2606 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) &&
2607 !vd->vdev_cant_write && !vd->vdev_ishole);
2608 }
2609
2610 boolean_t
2611 vdev_accessible(vdev_t *vd, zio_t *zio)
2612 {
2613 ASSERT(zio->io_vd == vd);
2614
2615 if (vdev_is_dead(vd) || vd->vdev_remove_wanted)
2616 return (B_FALSE);
2617
2618 if (zio->io_type == ZIO_TYPE_READ)
2619 return (!vd->vdev_cant_read);
2620
2621 if (zio->io_type == ZIO_TYPE_WRITE)
2622 return (!vd->vdev_cant_write);
2623
2624 return (B_TRUE);
2625 }
2626
2627 /*
2628 * Get statistics for the given vdev.
2629 */
2630 void
2631 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
2632 {
2633 spa_t *spa = vd->vdev_spa;
2634 vdev_t *rvd = spa->spa_root_vdev;
2635
2636 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
2637
2638 mutex_enter(&vd->vdev_stat_lock);
2639 bcopy(&vd->vdev_stat, vs, sizeof (*vs));
2640 vs->vs_timestamp = gethrtime() - vs->vs_timestamp;
2641 vs->vs_state = vd->vdev_state;
2642 vs->vs_rsize = vdev_get_min_asize(vd);
2643 if (vd->vdev_ops->vdev_op_leaf)
2644 vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
2645 vs->vs_esize = vd->vdev_max_asize - vd->vdev_asize;
2646 if (vd->vdev_aux == NULL && vd == vd->vdev_top && !vd->vdev_ishole) {
2647 vs->vs_fragmentation = vd->vdev_mg->mg_fragmentation;
2648 }
2649
2650 /*
2651 * If we're getting stats on the root vdev, aggregate the I/O counts
2652 * over all top-level vdevs (i.e. the direct children of the root).
2653 */
2654 if (vd == rvd) {
2655 for (int c = 0; c < rvd->vdev_children; c++) {
2656 vdev_t *cvd = rvd->vdev_child[c];
2657 vdev_stat_t *cvs = &cvd->vdev_stat;
2658
2659 for (int t = 0; t < ZIO_TYPES; t++) {
2660 vs->vs_ops[t] += cvs->vs_ops[t];
2661 vs->vs_bytes[t] += cvs->vs_bytes[t];
2662 }
2663 cvs->vs_scan_removing = cvd->vdev_removing;
2664 }
2665 }
2666 mutex_exit(&vd->vdev_stat_lock);
2667 }
2668
2669 void
2670 vdev_clear_stats(vdev_t *vd)
2671 {
2672 mutex_enter(&vd->vdev_stat_lock);
2673 vd->vdev_stat.vs_space = 0;
2674 vd->vdev_stat.vs_dspace = 0;
2675 vd->vdev_stat.vs_alloc = 0;
2676 mutex_exit(&vd->vdev_stat_lock);
2677 }
2678
2679 void
2680 vdev_scan_stat_init(vdev_t *vd)
2681 {
2682 vdev_stat_t *vs = &vd->vdev_stat;
2683
2684 for (int c = 0; c < vd->vdev_children; c++)
2685 vdev_scan_stat_init(vd->vdev_child[c]);
2686
2687 mutex_enter(&vd->vdev_stat_lock);
2688 vs->vs_scan_processed = 0;
2689 mutex_exit(&vd->vdev_stat_lock);
2690 }
2691
2692 void
2693 vdev_stat_update(zio_t *zio, uint64_t psize)
2694 {
2695 spa_t *spa = zio->io_spa;
2696 vdev_t *rvd = spa->spa_root_vdev;
2697 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd;
2698 vdev_t *pvd;
2699 uint64_t txg = zio->io_txg;
2700 vdev_stat_t *vs = &vd->vdev_stat;
2701 zio_type_t type = zio->io_type;
2702 int flags = zio->io_flags;
2703
2704 /*
2705 * If this i/o is a gang leader, it didn't do any actual work.
2706 */
2707 if (zio->io_gang_tree)
2708 return;
2709
2710 if (zio->io_error == 0) {
2711 /*
2712 * If this is a root i/o, don't count it -- we've already
2713 * counted the top-level vdevs, and vdev_get_stats() will
2714 * aggregate them when asked. This reduces contention on
2715 * the root vdev_stat_lock and implicitly handles blocks
2716 * that compress away to holes, for which there is no i/o.
2717 * (Holes never create vdev children, so all the counters
2718 * remain zero, which is what we want.)
2719 *
2720 * Note: this only applies to successful i/o (io_error == 0)
2721 * because unlike i/o counts, errors are not additive.
2722 * When reading a ditto block, for example, failure of
2723 * one top-level vdev does not imply a root-level error.
2724 */
2725 if (vd == rvd)
2726 return;
2727
2728 ASSERT(vd == zio->io_vd);
2729
2730 if (flags & ZIO_FLAG_IO_BYPASS)
2731 return;
2732
2733 mutex_enter(&vd->vdev_stat_lock);
2734
2735 if (flags & ZIO_FLAG_IO_REPAIR) {
2736 if (flags & ZIO_FLAG_SCAN_THREAD) {
2737 dsl_scan_phys_t *scn_phys =
2738 &spa->spa_dsl_pool->dp_scan->scn_phys;
2739 uint64_t *processed = &scn_phys->scn_processed;
2740
2741 /* XXX cleanup? */
2742 if (vd->vdev_ops->vdev_op_leaf)
2743 atomic_add_64(processed, psize);
2744 vs->vs_scan_processed += psize;
2745 }
2746
2747 if (flags & ZIO_FLAG_SELF_HEAL)
2748 vs->vs_self_healed += psize;
2749 }
2750
2751 vs->vs_ops[type]++;
2752 vs->vs_bytes[type] += psize;
2753
2754 mutex_exit(&vd->vdev_stat_lock);
2755 return;
2756 }
2757
2758 if (flags & ZIO_FLAG_SPECULATIVE)
2759 return;
2760
2761 /*
2762 * If this is an I/O error that is going to be retried, then ignore the
2763 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
2764 * hard errors, when in reality they can happen for any number of
2765 * innocuous reasons (bus resets, MPxIO link failure, etc).
2766 */
2767 if (zio->io_error == EIO &&
2768 !(zio->io_flags & ZIO_FLAG_IO_RETRY))
2769 return;
2770
2771 /*
2772 * Intent logs writes won't propagate their error to the root
2773 * I/O so don't mark these types of failures as pool-level
2774 * errors.
2775 */
2776 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE))
2777 return;
2778
2779 mutex_enter(&vd->vdev_stat_lock);
2780 if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) {
2781 if (zio->io_error == ECKSUM)
2782 vs->vs_checksum_errors++;
2783 else
2784 vs->vs_read_errors++;
2785 }
2786 if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd))
2787 vs->vs_write_errors++;
2788 mutex_exit(&vd->vdev_stat_lock);
2789
2790 if (type == ZIO_TYPE_WRITE && txg != 0 &&
2791 (!(flags & ZIO_FLAG_IO_REPAIR) ||
2792 (flags & ZIO_FLAG_SCAN_THREAD) ||
2793 spa->spa_claiming)) {
2794 /*
2795 * This is either a normal write (not a repair), or it's
2796 * a repair induced by the scrub thread, or it's a repair
2797 * made by zil_claim() during spa_load() in the first txg.
2798 * In the normal case, we commit the DTL change in the same
2799 * txg as the block was born. In the scrub-induced repair
2800 * case, we know that scrubs run in first-pass syncing context,
2801 * so we commit the DTL change in spa_syncing_txg(spa).
2802 * In the zil_claim() case, we commit in spa_first_txg(spa).
2803 *
2804 * We currently do not make DTL entries for failed spontaneous
2805 * self-healing writes triggered by normal (non-scrubbing)
2806 * reads, because we have no transactional context in which to
2807 * do so -- and it's not clear that it'd be desirable anyway.
2808 */
2809 if (vd->vdev_ops->vdev_op_leaf) {
2810 uint64_t commit_txg = txg;
2811 if (flags & ZIO_FLAG_SCAN_THREAD) {
2812 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2813 ASSERT(spa_sync_pass(spa) == 1);
2814 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1);
2815 commit_txg = spa_syncing_txg(spa);
2816 } else if (spa->spa_claiming) {
2817 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2818 commit_txg = spa_first_txg(spa);
2819 }
2820 ASSERT(commit_txg >= spa_syncing_txg(spa));
2821 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1))
2822 return;
2823 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2824 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1);
2825 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg);
2826 }
2827 if (vd != rvd)
2828 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1);
2829 }
2830 }
2831
2832 /*
2833 * Update the in-core space usage stats for this vdev, its metaslab class,
2834 * and the root vdev.
2835 */
2836 void
2837 vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta,
2838 int64_t space_delta)
2839 {
2840 int64_t dspace_delta = space_delta;
2841 spa_t *spa = vd->vdev_spa;
2842 vdev_t *rvd = spa->spa_root_vdev;
2843 metaslab_group_t *mg = vd->vdev_mg;
2844 metaslab_class_t *mc = mg ? mg->mg_class : NULL;
2845
2846 ASSERT(vd == vd->vdev_top);
2847
2848 /*
2849 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
2850 * factor. We must calculate this here and not at the root vdev
2851 * because the root vdev's psize-to-asize is simply the max of its
2852 * childrens', thus not accurate enough for us.
2853 */
2854 ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0);
2855 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache);
2856 dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) *
2857 vd->vdev_deflate_ratio;
2858
2859 mutex_enter(&vd->vdev_stat_lock);
2860 vd->vdev_stat.vs_alloc += alloc_delta;
2861 vd->vdev_stat.vs_space += space_delta;
2862 vd->vdev_stat.vs_dspace += dspace_delta;
2863 mutex_exit(&vd->vdev_stat_lock);
2864
2865 if (mc == spa_normal_class(spa)) {
2866 mutex_enter(&rvd->vdev_stat_lock);
2867 rvd->vdev_stat.vs_alloc += alloc_delta;
2868 rvd->vdev_stat.vs_space += space_delta;
2869 rvd->vdev_stat.vs_dspace += dspace_delta;
2870 mutex_exit(&rvd->vdev_stat_lock);
2871 }
2872
2873 if (mc != NULL) {
2874 ASSERT(rvd == vd->vdev_parent);
2875 ASSERT(vd->vdev_ms_count != 0);
2876
2877 metaslab_class_space_update(mc,
2878 alloc_delta, defer_delta, space_delta, dspace_delta);
2879 }
2880 }
2881
2882 /*
2883 * Mark a top-level vdev's config as dirty, placing it on the dirty list
2884 * so that it will be written out next time the vdev configuration is synced.
2885 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
2886 */
2887 void
2888 vdev_config_dirty(vdev_t *vd)
2889 {
2890 spa_t *spa = vd->vdev_spa;
2891 vdev_t *rvd = spa->spa_root_vdev;
2892 int c;
2893
2894 ASSERT(spa_writeable(spa));
2895
2896 /*
2897 * If this is an aux vdev (as with l2cache and spare devices), then we
2898 * update the vdev config manually and set the sync flag.
2899 */
2900 if (vd->vdev_aux != NULL) {
2901 spa_aux_vdev_t *sav = vd->vdev_aux;
2902 nvlist_t **aux;
2903 uint_t naux;
2904
2905 for (c = 0; c < sav->sav_count; c++) {
2906 if (sav->sav_vdevs[c] == vd)
2907 break;
2908 }
2909
2910 if (c == sav->sav_count) {
2911 /*
2912 * We're being removed. There's nothing more to do.
2913 */
2914 ASSERT(sav->sav_sync == B_TRUE);
2915 return;
2916 }
2917
2918 sav->sav_sync = B_TRUE;
2919
2920 if (nvlist_lookup_nvlist_array(sav->sav_config,
2921 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) {
2922 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
2923 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0);
2924 }
2925
2926 ASSERT(c < naux);
2927
2928 /*
2929 * Setting the nvlist in the middle if the array is a little
2930 * sketchy, but it will work.
2931 */
2932 nvlist_free(aux[c]);
2933 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0);
2934
2935 return;
2936 }
2937
2938 /*
2939 * The dirty list is protected by the SCL_CONFIG lock. The caller
2940 * must either hold SCL_CONFIG as writer, or must be the sync thread
2941 * (which holds SCL_CONFIG as reader). There's only one sync thread,
2942 * so this is sufficient to ensure mutual exclusion.
2943 */
2944 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
2945 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2946 spa_config_held(spa, SCL_CONFIG, RW_READER)));
2947
2948 if (vd == rvd) {
2949 for (c = 0; c < rvd->vdev_children; c++)
2950 vdev_config_dirty(rvd->vdev_child[c]);
2951 } else {
2952 ASSERT(vd == vd->vdev_top);
2953
2954 if (!list_link_active(&vd->vdev_config_dirty_node) &&
2955 !vd->vdev_ishole)
2956 list_insert_head(&spa->spa_config_dirty_list, vd);
2957 }
2958 }
2959
2960 void
2961 vdev_config_clean(vdev_t *vd)
2962 {
2963 spa_t *spa = vd->vdev_spa;
2964
2965 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
2966 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2967 spa_config_held(spa, SCL_CONFIG, RW_READER)));
2968
2969 ASSERT(list_link_active(&vd->vdev_config_dirty_node));
2970 list_remove(&spa->spa_config_dirty_list, vd);
2971 }
2972
2973 /*
2974 * Mark a top-level vdev's state as dirty, so that the next pass of
2975 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
2976 * the state changes from larger config changes because they require
2977 * much less locking, and are often needed for administrative actions.
2978 */
2979 void
2980 vdev_state_dirty(vdev_t *vd)
2981 {
2982 spa_t *spa = vd->vdev_spa;
2983
2984 ASSERT(spa_writeable(spa));
2985 ASSERT(vd == vd->vdev_top);
2986
2987 /*
2988 * The state list is protected by the SCL_STATE lock. The caller
2989 * must either hold SCL_STATE as writer, or must be the sync thread
2990 * (which holds SCL_STATE as reader). There's only one sync thread,
2991 * so this is sufficient to ensure mutual exclusion.
2992 */
2993 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
2994 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2995 spa_config_held(spa, SCL_STATE, RW_READER)));
2996
2997 if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole)
2998 list_insert_head(&spa->spa_state_dirty_list, vd);
2999 }
3000
3001 void
3002 vdev_state_clean(vdev_t *vd)
3003 {
3004 spa_t *spa = vd->vdev_spa;
3005
3006 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
3007 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3008 spa_config_held(spa, SCL_STATE, RW_READER)));
3009
3010 ASSERT(list_link_active(&vd->vdev_state_dirty_node));
3011 list_remove(&spa->spa_state_dirty_list, vd);
3012 }
3013
3014 /*
3015 * Propagate vdev state up from children to parent.
3016 */
3017 void
3018 vdev_propagate_state(vdev_t *vd)
3019 {
3020 spa_t *spa = vd->vdev_spa;
3021 vdev_t *rvd = spa->spa_root_vdev;
3022 int degraded = 0, faulted = 0;
3023 int corrupted = 0;
3024 vdev_t *child;
3025
3026 if (vd->vdev_children > 0) {
3027 for (int c = 0; c < vd->vdev_children; c++) {
3028 child = vd->vdev_child[c];
3029
3030 /*
3031 * Don't factor holes into the decision.
3032 */
3033 if (child->vdev_ishole)
3034 continue;
3035
3036 if (!vdev_readable(child) ||
3037 (!vdev_writeable(child) && spa_writeable(spa))) {
3038 /*
3039 * Root special: if there is a top-level log
3040 * device, treat the root vdev as if it were
3041 * degraded.
3042 */
3043 if (child->vdev_islog && vd == rvd)
3044 degraded++;
3045 else
3046 faulted++;
3047 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) {
3048 degraded++;
3049 }
3050
3051 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA)
3052 corrupted++;
3053 }
3054
3055 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded);
3056
3057 /*
3058 * Root special: if there is a top-level vdev that cannot be
3059 * opened due to corrupted metadata, then propagate the root
3060 * vdev's aux state as 'corrupt' rather than 'insufficient
3061 * replicas'.
3062 */
3063 if (corrupted && vd == rvd &&
3064 rvd->vdev_state == VDEV_STATE_CANT_OPEN)
3065 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN,
3066 VDEV_AUX_CORRUPT_DATA);
3067 }
3068
3069 if (vd->vdev_parent)
3070 vdev_propagate_state(vd->vdev_parent);
3071 }
3072
3073 /*
3074 * Set a vdev's state. If this is during an open, we don't update the parent
3075 * state, because we're in the process of opening children depth-first.
3076 * Otherwise, we propagate the change to the parent.
3077 *
3078 * If this routine places a device in a faulted state, an appropriate ereport is
3079 * generated.
3080 */
3081 void
3082 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
3083 {
3084 uint64_t save_state;
3085 spa_t *spa = vd->vdev_spa;
3086
3087 if (state == vd->vdev_state) {
3088 vd->vdev_stat.vs_aux = aux;
3089 return;
3090 }
3091
3092 save_state = vd->vdev_state;
3093
3094 vd->vdev_state = state;
3095 vd->vdev_stat.vs_aux = aux;
3096
3097 /*
3098 * If we are setting the vdev state to anything but an open state, then
3099 * always close the underlying device unless the device has requested
3100 * a delayed close (i.e. we're about to remove or fault the device).
3101 * Otherwise, we keep accessible but invalid devices open forever.
3102 * We don't call vdev_close() itself, because that implies some extra
3103 * checks (offline, etc) that we don't want here. This is limited to
3104 * leaf devices, because otherwise closing the device will affect other
3105 * children.
3106 */
3107 if (!vd->vdev_delayed_close && vdev_is_dead(vd) &&
3108 vd->vdev_ops->vdev_op_leaf)
3109 vd->vdev_ops->vdev_op_close(vd);
3110
3111 /*
3112 * If we have brought this vdev back into service, we need
3113 * to notify fmd so that it can gracefully repair any outstanding
3114 * cases due to a missing device. We do this in all cases, even those
3115 * that probably don't correlate to a repaired fault. This is sure to
3116 * catch all cases, and we let the zfs-retire agent sort it out. If
3117 * this is a transient state it's OK, as the retire agent will
3118 * double-check the state of the vdev before repairing it.
3119 */
3120 if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf &&
3121 vd->vdev_prevstate != state)
3122 zfs_post_state_change(spa, vd);
3123
3124 if (vd->vdev_removed &&
3125 state == VDEV_STATE_CANT_OPEN &&
3126 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) {
3127 /*
3128 * If the previous state is set to VDEV_STATE_REMOVED, then this
3129 * device was previously marked removed and someone attempted to
3130 * reopen it. If this failed due to a nonexistent device, then
3131 * keep the device in the REMOVED state. We also let this be if
3132 * it is one of our special test online cases, which is only
3133 * attempting to online the device and shouldn't generate an FMA
3134 * fault.
3135 */
3136 vd->vdev_state = VDEV_STATE_REMOVED;
3137 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
3138 } else if (state == VDEV_STATE_REMOVED) {
3139 vd->vdev_removed = B_TRUE;
3140 } else if (state == VDEV_STATE_CANT_OPEN) {
3141 /*
3142 * If we fail to open a vdev during an import or recovery, we
3143 * mark it as "not available", which signifies that it was
3144 * never there to begin with. Failure to open such a device
3145 * is not considered an error.
3146 */
3147 if ((spa_load_state(spa) == SPA_LOAD_IMPORT ||
3148 spa_load_state(spa) == SPA_LOAD_RECOVER) &&
3149 vd->vdev_ops->vdev_op_leaf)
3150 vd->vdev_not_present = 1;
3151
3152 /*
3153 * Post the appropriate ereport. If the 'prevstate' field is
3154 * set to something other than VDEV_STATE_UNKNOWN, it indicates
3155 * that this is part of a vdev_reopen(). In this case, we don't
3156 * want to post the ereport if the device was already in the
3157 * CANT_OPEN state beforehand.
3158 *
3159 * If the 'checkremove' flag is set, then this is an attempt to
3160 * online the device in response to an insertion event. If we
3161 * hit this case, then we have detected an insertion event for a
3162 * faulted or offline device that wasn't in the removed state.
3163 * In this scenario, we don't post an ereport because we are
3164 * about to replace the device, or attempt an online with
3165 * vdev_forcefault, which will generate the fault for us.
3166 */
3167 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) &&
3168 !vd->vdev_not_present && !vd->vdev_checkremove &&
3169 vd != spa->spa_root_vdev) {
3170 const char *class;
3171
3172 switch (aux) {
3173 case VDEV_AUX_OPEN_FAILED:
3174 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED;
3175 break;
3176 case VDEV_AUX_CORRUPT_DATA:
3177 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA;
3178 break;
3179 case VDEV_AUX_NO_REPLICAS:
3180 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS;
3181 break;
3182 case VDEV_AUX_BAD_GUID_SUM:
3183 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM;
3184 break;
3185 case VDEV_AUX_TOO_SMALL:
3186 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL;
3187 break;
3188 case VDEV_AUX_BAD_LABEL:
3189 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL;
3190 break;
3191 default:
3192 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN;
3193 }
3194
3195 zfs_ereport_post(class, spa, vd, NULL, save_state, 0);
3196 }
3197
3198 /* Erase any notion of persistent removed state */
3199 vd->vdev_removed = B_FALSE;
3200 } else {
3201 vd->vdev_removed = B_FALSE;
3202 }
3203
3204 if (!isopen && vd->vdev_parent)
3205 vdev_propagate_state(vd->vdev_parent);
3206 }
3207
3208 /*
3209 * Check the vdev configuration to ensure that it's capable of supporting
3210 * a root pool. Currently, we do not support RAID-Z or partial configuration.
3211 * In addition, only a single top-level vdev is allowed and none of the leaves
3212 * can be wholedisks.
3213 */
3214 boolean_t
3215 vdev_is_bootable(vdev_t *vd)
3216 {
3217 if (!vd->vdev_ops->vdev_op_leaf) {
3218 char *vdev_type = vd->vdev_ops->vdev_op_type;
3219
3220 if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 &&
3221 vd->vdev_children > 1) {
3222 return (B_FALSE);
3223 } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 ||
3224 strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) {
3225 return (B_FALSE);
3226 }
3227 }
3228
3229 for (int c = 0; c < vd->vdev_children; c++) {
3230 if (!vdev_is_bootable(vd->vdev_child[c]))
3231 return (B_FALSE);
3232 }
3233 return (B_TRUE);
3234 }
3235
3236 /*
3237 * Load the state from the original vdev tree (ovd) which
3238 * we've retrieved from the MOS config object. If the original
3239 * vdev was offline or faulted then we transfer that state to the
3240 * device in the current vdev tree (nvd).
3241 */
3242 void
3243 vdev_load_log_state(vdev_t *nvd, vdev_t *ovd)
3244 {
3245 spa_t *spa = nvd->vdev_spa;
3246
3247 ASSERT(nvd->vdev_top->vdev_islog);
3248 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
3249 ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid);
3250
3251 for (int c = 0; c < nvd->vdev_children; c++)
3252 vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]);
3253
3254 if (nvd->vdev_ops->vdev_op_leaf) {
3255 /*
3256 * Restore the persistent vdev state
3257 */
3258 nvd->vdev_offline = ovd->vdev_offline;
3259 nvd->vdev_faulted = ovd->vdev_faulted;
3260 nvd->vdev_degraded = ovd->vdev_degraded;
3261 nvd->vdev_removed = ovd->vdev_removed;
3262 }
3263 }
3264
3265 /*
3266 * Determine if a log device has valid content. If the vdev was
3267 * removed or faulted in the MOS config then we know that
3268 * the content on the log device has already been written to the pool.
3269 */
3270 boolean_t
3271 vdev_log_state_valid(vdev_t *vd)
3272 {
3273 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted &&
3274 !vd->vdev_removed)
3275 return (B_TRUE);
3276
3277 for (int c = 0; c < vd->vdev_children; c++)
3278 if (vdev_log_state_valid(vd->vdev_child[c]))
3279 return (B_TRUE);
3280
3281 return (B_FALSE);
3282 }
3283
3284 /*
3285 * Expand a vdev if possible.
3286 */
3287 void
3288 vdev_expand(vdev_t *vd, uint64_t txg)
3289 {
3290 ASSERT(vd->vdev_top == vd);
3291 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3292
3293 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) {
3294 VERIFY(vdev_metaslab_init(vd, txg) == 0);
3295 vdev_config_dirty(vd);
3296 }
3297 }
3298
3299 /*
3300 * Split a vdev.
3301 */
3302 void
3303 vdev_split(vdev_t *vd)
3304 {
3305 vdev_t *cvd, *pvd = vd->vdev_parent;
3306
3307 vdev_remove_child(pvd, vd);
3308 vdev_compact_children(pvd);
3309
3310 cvd = pvd->vdev_child[0];
3311 if (pvd->vdev_children == 1) {
3312 vdev_remove_parent(cvd);
3313 cvd->vdev_splitting = B_TRUE;
3314 }
3315 vdev_propagate_state(cvd);
3316 }
3317
3318 void
3319 vdev_deadman(vdev_t *vd)
3320 {
3321 for (int c = 0; c < vd->vdev_children; c++) {
3322 vdev_t *cvd = vd->vdev_child[c];
3323
3324 vdev_deadman(cvd);
3325 }
3326
3327 if (vd->vdev_ops->vdev_op_leaf) {
3328 vdev_queue_t *vq = &vd->vdev_queue;
3329
3330 mutex_enter(&vq->vq_lock);
3331 if (avl_numnodes(&vq->vq_active_tree) > 0) {
3332 spa_t *spa = vd->vdev_spa;
3333 zio_t *fio;
3334 uint64_t delta;
3335
3336 /*
3337 * Look at the head of all the pending queues,
3338 * if any I/O has been outstanding for longer than
3339 * the spa_deadman_synctime we panic the system.
3340 */
3341 fio = avl_first(&vq->vq_active_tree);
3342 delta = gethrtime() - fio->io_timestamp;
3343 if (delta > spa_deadman_synctime(spa)) {
3344 zfs_dbgmsg("SLOW IO: zio timestamp %lluns, "
3345 "delta %lluns, last io %lluns",
3346 fio->io_timestamp, delta,
3347 vq->vq_io_complete_ts);
3348 fm_panic("I/O to pool '%s' appears to be "
3349 "hung.", spa_name(spa));
3350 }
3351 }
3352 mutex_exit(&vq->vq_lock);
3353 }
3354 }