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