1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22 /*
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2011, 2015 by Delphix. All rights reserved.
25 * Copyright 2015 Nexenta Systems, Inc. All rights reserved.
26 * Copyright (c) 2014 Integros [integros.com]
27 * Copyright 2016 Toomas Soome <tsoome@me.com>
28 */
29
30 #include <sys/zfs_context.h>
31 #include <sys/fm/fs/zfs.h>
32 #include <sys/spa.h>
33 #include <sys/spa_impl.h>
34 #include <sys/dmu.h>
35 #include <sys/dmu_tx.h>
36 #include <sys/vdev_impl.h>
37 #include <sys/uberblock_impl.h>
38 #include <sys/metaslab.h>
39 #include <sys/metaslab_impl.h>
40 #include <sys/space_map.h>
41 #include <sys/space_reftree.h>
42 #include <sys/zio.h>
43 #include <sys/zap.h>
44 #include <sys/fs/zfs.h>
45 #include <sys/arc.h>
46 #include <sys/zil.h>
47 #include <sys/dsl_scan.h>
48
49 /*
50 * Virtual device management.
51 */
52
53 static vdev_ops_t *vdev_ops_table[] = {
54 &vdev_root_ops,
55 &vdev_raidz_ops,
56 &vdev_mirror_ops,
57 &vdev_replacing_ops,
58 &vdev_spare_ops,
59 &vdev_disk_ops,
60 &vdev_file_ops,
61 &vdev_missing_ops,
62 &vdev_hole_ops,
63 NULL
64 };
65
66 /* maximum scrub/resilver I/O queue per leaf vdev */
67 int zfs_scrub_limit = 10;
68
69 /*
70 * When a vdev is added, it will be divided into approximately (but no
71 * more than) this number of metaslabs.
72 */
73 int metaslabs_per_vdev = 200;
74
75 /*
76 * Given a vdev type, return the appropriate ops vector.
77 */
78 static vdev_ops_t *
79 vdev_getops(const char *type)
80 {
81 vdev_ops_t *ops, **opspp;
82
83 for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++)
84 if (strcmp(ops->vdev_op_type, type) == 0)
85 break;
86
87 return (ops);
88 }
89
90 /*
91 * Default asize function: return the MAX of psize with the asize of
92 * all children. This is what's used by anything other than RAID-Z.
93 */
94 uint64_t
95 vdev_default_asize(vdev_t *vd, uint64_t psize)
96 {
97 uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift);
98 uint64_t csize;
99
100 for (int c = 0; c < vd->vdev_children; c++) {
101 csize = vdev_psize_to_asize(vd->vdev_child[c], psize);
102 asize = MAX(asize, csize);
103 }
104
105 return (asize);
106 }
107
108 /*
109 * Get the minimum allocatable size. We define the allocatable size as
110 * the vdev's asize rounded to the nearest metaslab. This allows us to
111 * replace or attach devices which don't have the same physical size but
112 * can still satisfy the same number of allocations.
113 */
114 uint64_t
115 vdev_get_min_asize(vdev_t *vd)
116 {
117 vdev_t *pvd = vd->vdev_parent;
118
119 /*
120 * If our parent is NULL (inactive spare or cache) or is the root,
121 * just return our own asize.
122 */
123 if (pvd == NULL)
124 return (vd->vdev_asize);
125
126 /*
127 * The top-level vdev just returns the allocatable size rounded
128 * to the nearest metaslab.
129 */
130 if (vd == vd->vdev_top)
131 return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift));
132
133 /*
134 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
135 * so each child must provide at least 1/Nth of its asize.
136 */
137 if (pvd->vdev_ops == &vdev_raidz_ops)
138 return (pvd->vdev_min_asize / pvd->vdev_children);
139
140 return (pvd->vdev_min_asize);
141 }
142
143 void
144 vdev_set_min_asize(vdev_t *vd)
145 {
146 vd->vdev_min_asize = vdev_get_min_asize(vd);
147
148 for (int c = 0; c < vd->vdev_children; c++)
149 vdev_set_min_asize(vd->vdev_child[c]);
150 }
151
152 vdev_t *
153 vdev_lookup_top(spa_t *spa, uint64_t vdev)
154 {
155 vdev_t *rvd = spa->spa_root_vdev;
156
157 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
158
159 if (vdev < rvd->vdev_children) {
160 ASSERT(rvd->vdev_child[vdev] != NULL);
161 return (rvd->vdev_child[vdev]);
162 }
163
164 return (NULL);
165 }
166
167 vdev_t *
168 vdev_lookup_by_guid(vdev_t *vd, uint64_t guid)
169 {
170 vdev_t *mvd;
171
172 if (vd->vdev_guid == guid)
173 return (vd);
174
175 for (int c = 0; c < vd->vdev_children; c++)
176 if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) !=
177 NULL)
178 return (mvd);
179
180 return (NULL);
181 }
182
183 static int
184 vdev_count_leaves_impl(vdev_t *vd)
185 {
186 int n = 0;
187
188 if (vd->vdev_ops->vdev_op_leaf)
189 return (1);
190
191 for (int c = 0; c < vd->vdev_children; c++)
192 n += vdev_count_leaves_impl(vd->vdev_child[c]);
193
194 return (n);
195 }
196
197 int
198 vdev_count_leaves(spa_t *spa)
199 {
200 return (vdev_count_leaves_impl(spa->spa_root_vdev));
201 }
202
203 void
204 vdev_add_child(vdev_t *pvd, vdev_t *cvd)
205 {
206 size_t oldsize, newsize;
207 uint64_t id = cvd->vdev_id;
208 vdev_t **newchild;
209 spa_t *spa = cvd->vdev_spa;
210
211 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
212 ASSERT(cvd->vdev_parent == NULL);
213
214 cvd->vdev_parent = pvd;
215
216 if (pvd == NULL)
217 return;
218
219 ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL);
220
221 oldsize = pvd->vdev_children * sizeof (vdev_t *);
222 pvd->vdev_children = MAX(pvd->vdev_children, id + 1);
223 newsize = pvd->vdev_children * sizeof (vdev_t *);
224
225 newchild = kmem_zalloc(newsize, KM_SLEEP);
226 if (pvd->vdev_child != NULL) {
227 bcopy(pvd->vdev_child, newchild, oldsize);
228 kmem_free(pvd->vdev_child, oldsize);
229 }
230
231 pvd->vdev_child = newchild;
232 pvd->vdev_child[id] = cvd;
233
234 cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd);
235 ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL);
236
237 /*
238 * Walk up all ancestors to update guid sum.
239 */
240 for (; pvd != NULL; pvd = pvd->vdev_parent)
241 pvd->vdev_guid_sum += cvd->vdev_guid_sum;
242 }
243
244 void
245 vdev_remove_child(vdev_t *pvd, vdev_t *cvd)
246 {
247 int c;
248 uint_t id = cvd->vdev_id;
249
250 ASSERT(cvd->vdev_parent == pvd);
251
252 if (pvd == NULL)
253 return;
254
255 ASSERT(id < pvd->vdev_children);
256 ASSERT(pvd->vdev_child[id] == cvd);
257
258 pvd->vdev_child[id] = NULL;
259 cvd->vdev_parent = NULL;
260
261 for (c = 0; c < pvd->vdev_children; c++)
262 if (pvd->vdev_child[c])
263 break;
264
265 if (c == pvd->vdev_children) {
266 kmem_free(pvd->vdev_child, c * sizeof (vdev_t *));
267 pvd->vdev_child = NULL;
268 pvd->vdev_children = 0;
269 }
270
271 /*
272 * Walk up all ancestors to update guid sum.
273 */
274 for (; pvd != NULL; pvd = pvd->vdev_parent)
275 pvd->vdev_guid_sum -= cvd->vdev_guid_sum;
276 }
277
278 /*
279 * Remove any holes in the child array.
280 */
281 void
282 vdev_compact_children(vdev_t *pvd)
283 {
284 vdev_t **newchild, *cvd;
285 int oldc = pvd->vdev_children;
286 int newc;
287
288 ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
289
290 for (int c = newc = 0; c < oldc; c++)
291 if (pvd->vdev_child[c])
292 newc++;
293
294 newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP);
295
296 for (int c = newc = 0; c < oldc; c++) {
297 if ((cvd = pvd->vdev_child[c]) != NULL) {
298 newchild[newc] = cvd;
299 cvd->vdev_id = newc++;
300 }
301 }
302
303 kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *));
304 pvd->vdev_child = newchild;
305 pvd->vdev_children = newc;
306 }
307
308 /*
309 * Allocate and minimally initialize a vdev_t.
310 */
311 vdev_t *
312 vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops)
313 {
314 vdev_t *vd;
315
316 vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP);
317
318 if (spa->spa_root_vdev == NULL) {
319 ASSERT(ops == &vdev_root_ops);
320 spa->spa_root_vdev = vd;
321 spa->spa_load_guid = spa_generate_guid(NULL);
322 }
323
324 if (guid == 0 && ops != &vdev_hole_ops) {
325 if (spa->spa_root_vdev == vd) {
326 /*
327 * The root vdev's guid will also be the pool guid,
328 * which must be unique among all pools.
329 */
330 guid = spa_generate_guid(NULL);
331 } else {
332 /*
333 * Any other vdev's guid must be unique within the pool.
334 */
335 guid = spa_generate_guid(spa);
336 }
337 ASSERT(!spa_guid_exists(spa_guid(spa), guid));
338 }
339
340 vd->vdev_spa = spa;
341 vd->vdev_id = id;
342 vd->vdev_guid = guid;
343 vd->vdev_guid_sum = guid;
344 vd->vdev_ops = ops;
345 vd->vdev_state = VDEV_STATE_CLOSED;
346 vd->vdev_ishole = (ops == &vdev_hole_ops);
347
348 mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL);
349 mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL);
350 mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL);
351 mutex_init(&vd->vdev_queue_lock, NULL, MUTEX_DEFAULT, NULL);
352 for (int t = 0; t < DTL_TYPES; t++) {
353 vd->vdev_dtl[t] = range_tree_create(NULL, NULL,
354 &vd->vdev_dtl_lock);
355 }
356 txg_list_create(&vd->vdev_ms_list,
357 offsetof(struct metaslab, ms_txg_node));
358 txg_list_create(&vd->vdev_dtl_list,
359 offsetof(struct vdev, vdev_dtl_node));
360 vd->vdev_stat.vs_timestamp = gethrtime();
361 vdev_queue_init(vd);
362 vdev_cache_init(vd);
363
364 return (vd);
365 }
366
367 /*
368 * Allocate a new vdev. The 'alloctype' is used to control whether we are
369 * creating a new vdev or loading an existing one - the behavior is slightly
370 * different for each case.
371 */
372 int
373 vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
374 int alloctype)
375 {
376 vdev_ops_t *ops;
377 char *type;
378 uint64_t guid = 0, islog, nparity;
379 vdev_t *vd;
380
381 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
382
383 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
384 return (SET_ERROR(EINVAL));
385
386 if ((ops = vdev_getops(type)) == NULL)
387 return (SET_ERROR(EINVAL));
388
389 /*
390 * If this is a load, get the vdev guid from the nvlist.
391 * Otherwise, vdev_alloc_common() will generate one for us.
392 */
393 if (alloctype == VDEV_ALLOC_LOAD) {
394 uint64_t label_id;
395
396 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) ||
397 label_id != id)
398 return (SET_ERROR(EINVAL));
399
400 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
401 return (SET_ERROR(EINVAL));
402 } else if (alloctype == VDEV_ALLOC_SPARE) {
403 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
404 return (SET_ERROR(EINVAL));
405 } else if (alloctype == VDEV_ALLOC_L2CACHE) {
406 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
407 return (SET_ERROR(EINVAL));
408 } else if (alloctype == VDEV_ALLOC_ROOTPOOL) {
409 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
410 return (SET_ERROR(EINVAL));
411 }
412
413 /*
414 * The first allocated vdev must be of type 'root'.
415 */
416 if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL)
417 return (SET_ERROR(EINVAL));
418
419 /*
420 * Determine whether we're a log vdev.
421 */
422 islog = 0;
423 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog);
424 if (islog && spa_version(spa) < SPA_VERSION_SLOGS)
425 return (SET_ERROR(ENOTSUP));
426
427 if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES)
428 return (SET_ERROR(ENOTSUP));
429
430 /*
431 * Set the nparity property for RAID-Z vdevs.
432 */
433 nparity = -1ULL;
434 if (ops == &vdev_raidz_ops) {
435 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
436 &nparity) == 0) {
437 if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY)
438 return (SET_ERROR(EINVAL));
439 /*
440 * Previous versions could only support 1 or 2 parity
441 * device.
442 */
443 if (nparity > 1 &&
444 spa_version(spa) < SPA_VERSION_RAIDZ2)
445 return (SET_ERROR(ENOTSUP));
446 if (nparity > 2 &&
447 spa_version(spa) < SPA_VERSION_RAIDZ3)
448 return (SET_ERROR(ENOTSUP));
449 } else {
450 /*
451 * We require the parity to be specified for SPAs that
452 * support multiple parity levels.
453 */
454 if (spa_version(spa) >= SPA_VERSION_RAIDZ2)
455 return (SET_ERROR(EINVAL));
456 /*
457 * Otherwise, we default to 1 parity device for RAID-Z.
458 */
459 nparity = 1;
460 }
461 } else {
462 nparity = 0;
463 }
464 ASSERT(nparity != -1ULL);
465
466 vd = vdev_alloc_common(spa, id, guid, ops);
467
468 vd->vdev_islog = islog;
469 vd->vdev_nparity = nparity;
470
471 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0)
472 vd->vdev_path = spa_strdup(vd->vdev_path);
473 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0)
474 vd->vdev_devid = spa_strdup(vd->vdev_devid);
475 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH,
476 &vd->vdev_physpath) == 0)
477 vd->vdev_physpath = spa_strdup(vd->vdev_physpath);
478 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0)
479 vd->vdev_fru = spa_strdup(vd->vdev_fru);
480
481 /*
482 * Set the whole_disk property. If it's not specified, leave the value
483 * as -1.
484 */
485 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
486 &vd->vdev_wholedisk) != 0)
487 vd->vdev_wholedisk = -1ULL;
488
489 /*
490 * Look for the 'not present' flag. This will only be set if the device
491 * was not present at the time of import.
492 */
493 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
494 &vd->vdev_not_present);
495
496 /*
497 * Get the alignment requirement.
498 */
499 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift);
500
501 /*
502 * Retrieve the vdev creation time.
503 */
504 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG,
505 &vd->vdev_crtxg);
506
507 /*
508 * If we're a top-level vdev, try to load the allocation parameters.
509 */
510 if (parent && !parent->vdev_parent &&
511 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
512 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
513 &vd->vdev_ms_array);
514 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
515 &vd->vdev_ms_shift);
516 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE,
517 &vd->vdev_asize);
518 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING,
519 &vd->vdev_removing);
520 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP,
521 &vd->vdev_top_zap);
522 } else {
523 ASSERT0(vd->vdev_top_zap);
524 }
525
526 if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) {
527 ASSERT(alloctype == VDEV_ALLOC_LOAD ||
528 alloctype == VDEV_ALLOC_ADD ||
529 alloctype == VDEV_ALLOC_SPLIT ||
530 alloctype == VDEV_ALLOC_ROOTPOOL);
531 vd->vdev_mg = metaslab_group_create(islog ?
532 spa_log_class(spa) : spa_normal_class(spa), vd);
533 }
534
535 if (vd->vdev_ops->vdev_op_leaf &&
536 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
537 (void) nvlist_lookup_uint64(nv,
538 ZPOOL_CONFIG_VDEV_LEAF_ZAP, &vd->vdev_leaf_zap);
539 } else {
540 ASSERT0(vd->vdev_leaf_zap);
541 }
542
543 /*
544 * If we're a leaf vdev, try to load the DTL object and other state.
545 */
546
547 if (vd->vdev_ops->vdev_op_leaf &&
548 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE ||
549 alloctype == VDEV_ALLOC_ROOTPOOL)) {
550 if (alloctype == VDEV_ALLOC_LOAD) {
551 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL,
552 &vd->vdev_dtl_object);
553 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE,
554 &vd->vdev_unspare);
555 }
556
557 if (alloctype == VDEV_ALLOC_ROOTPOOL) {
558 uint64_t spare = 0;
559
560 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
561 &spare) == 0 && spare)
562 spa_spare_add(vd);
563 }
564
565 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE,
566 &vd->vdev_offline);
567
568 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
569 &vd->vdev_resilver_txg);
570
571 /*
572 * When importing a pool, we want to ignore the persistent fault
573 * state, as the diagnosis made on another system may not be
574 * valid in the current context. Local vdevs will
575 * remain in the faulted state.
576 */
577 if (spa_load_state(spa) == SPA_LOAD_OPEN) {
578 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED,
579 &vd->vdev_faulted);
580 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED,
581 &vd->vdev_degraded);
582 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED,
583 &vd->vdev_removed);
584
585 if (vd->vdev_faulted || vd->vdev_degraded) {
586 char *aux;
587
588 vd->vdev_label_aux =
589 VDEV_AUX_ERR_EXCEEDED;
590 if (nvlist_lookup_string(nv,
591 ZPOOL_CONFIG_AUX_STATE, &aux) == 0 &&
592 strcmp(aux, "external") == 0)
593 vd->vdev_label_aux = VDEV_AUX_EXTERNAL;
594 }
595 }
596 }
597
598 /*
599 * Add ourselves to the parent's list of children.
600 */
601 vdev_add_child(parent, vd);
602
603 *vdp = vd;
604
605 return (0);
606 }
607
608 void
609 vdev_free(vdev_t *vd)
610 {
611 spa_t *spa = vd->vdev_spa;
612
613 /*
614 * vdev_free() implies closing the vdev first. This is simpler than
615 * trying to ensure complicated semantics for all callers.
616 */
617 vdev_close(vd);
618
619 ASSERT(!list_link_active(&vd->vdev_config_dirty_node));
620 ASSERT(!list_link_active(&vd->vdev_state_dirty_node));
621
622 /*
623 * Free all children.
624 */
625 for (int c = 0; c < vd->vdev_children; c++)
626 vdev_free(vd->vdev_child[c]);
627
628 ASSERT(vd->vdev_child == NULL);
629 ASSERT(vd->vdev_guid_sum == vd->vdev_guid);
630
631 /*
632 * Discard allocation state.
633 */
634 if (vd->vdev_mg != NULL) {
635 vdev_metaslab_fini(vd);
636 metaslab_group_destroy(vd->vdev_mg);
637 }
638
639 ASSERT0(vd->vdev_stat.vs_space);
640 ASSERT0(vd->vdev_stat.vs_dspace);
641 ASSERT0(vd->vdev_stat.vs_alloc);
642
643 /*
644 * Remove this vdev from its parent's child list.
645 */
646 vdev_remove_child(vd->vdev_parent, vd);
647
648 ASSERT(vd->vdev_parent == NULL);
649
650 /*
651 * Clean up vdev structure.
652 */
653 vdev_queue_fini(vd);
654 vdev_cache_fini(vd);
655
656 if (vd->vdev_path)
657 spa_strfree(vd->vdev_path);
658 if (vd->vdev_devid)
659 spa_strfree(vd->vdev_devid);
660 if (vd->vdev_physpath)
661 spa_strfree(vd->vdev_physpath);
662 if (vd->vdev_fru)
663 spa_strfree(vd->vdev_fru);
664
665 if (vd->vdev_isspare)
666 spa_spare_remove(vd);
667 if (vd->vdev_isl2cache)
668 spa_l2cache_remove(vd);
669
670 txg_list_destroy(&vd->vdev_ms_list);
671 txg_list_destroy(&vd->vdev_dtl_list);
672
673 mutex_enter(&vd->vdev_dtl_lock);
674 space_map_close(vd->vdev_dtl_sm);
675 for (int t = 0; t < DTL_TYPES; t++) {
676 range_tree_vacate(vd->vdev_dtl[t], NULL, NULL);
677 range_tree_destroy(vd->vdev_dtl[t]);
678 }
679 mutex_exit(&vd->vdev_dtl_lock);
680
681 mutex_destroy(&vd->vdev_queue_lock);
682 mutex_destroy(&vd->vdev_dtl_lock);
683 mutex_destroy(&vd->vdev_stat_lock);
684 mutex_destroy(&vd->vdev_probe_lock);
685
686 if (vd == spa->spa_root_vdev)
687 spa->spa_root_vdev = NULL;
688
689 kmem_free(vd, sizeof (vdev_t));
690 }
691
692 /*
693 * Transfer top-level vdev state from svd to tvd.
694 */
695 static void
696 vdev_top_transfer(vdev_t *svd, vdev_t *tvd)
697 {
698 spa_t *spa = svd->vdev_spa;
699 metaslab_t *msp;
700 vdev_t *vd;
701 int t;
702
703 ASSERT(tvd == tvd->vdev_top);
704
705 tvd->vdev_ms_array = svd->vdev_ms_array;
706 tvd->vdev_ms_shift = svd->vdev_ms_shift;
707 tvd->vdev_ms_count = svd->vdev_ms_count;
708 tvd->vdev_top_zap = svd->vdev_top_zap;
709
710 svd->vdev_ms_array = 0;
711 svd->vdev_ms_shift = 0;
712 svd->vdev_ms_count = 0;
713 svd->vdev_top_zap = 0;
714
715 if (tvd->vdev_mg)
716 ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg);
717 tvd->vdev_mg = svd->vdev_mg;
718 tvd->vdev_ms = svd->vdev_ms;
719
720 svd->vdev_mg = NULL;
721 svd->vdev_ms = NULL;
722
723 if (tvd->vdev_mg != NULL)
724 tvd->vdev_mg->mg_vd = tvd;
725
726 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc;
727 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space;
728 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace;
729
730 svd->vdev_stat.vs_alloc = 0;
731 svd->vdev_stat.vs_space = 0;
732 svd->vdev_stat.vs_dspace = 0;
733
734 for (t = 0; t < TXG_SIZE; t++) {
735 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL)
736 (void) txg_list_add(&tvd->vdev_ms_list, msp, t);
737 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL)
738 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t);
739 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t))
740 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t);
741 }
742
743 if (list_link_active(&svd->vdev_config_dirty_node)) {
744 vdev_config_clean(svd);
745 vdev_config_dirty(tvd);
746 }
747
748 if (list_link_active(&svd->vdev_state_dirty_node)) {
749 vdev_state_clean(svd);
750 vdev_state_dirty(tvd);
751 }
752
753 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio;
754 svd->vdev_deflate_ratio = 0;
755
756 tvd->vdev_islog = svd->vdev_islog;
757 svd->vdev_islog = 0;
758 }
759
760 static void
761 vdev_top_update(vdev_t *tvd, vdev_t *vd)
762 {
763 if (vd == NULL)
764 return;
765
766 vd->vdev_top = tvd;
767
768 for (int c = 0; c < vd->vdev_children; c++)
769 vdev_top_update(tvd, vd->vdev_child[c]);
770 }
771
772 /*
773 * Add a mirror/replacing vdev above an existing vdev.
774 */
775 vdev_t *
776 vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops)
777 {
778 spa_t *spa = cvd->vdev_spa;
779 vdev_t *pvd = cvd->vdev_parent;
780 vdev_t *mvd;
781
782 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
783
784 mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops);
785
786 mvd->vdev_asize = cvd->vdev_asize;
787 mvd->vdev_min_asize = cvd->vdev_min_asize;
788 mvd->vdev_max_asize = cvd->vdev_max_asize;
789 mvd->vdev_ashift = cvd->vdev_ashift;
790 mvd->vdev_state = cvd->vdev_state;
791 mvd->vdev_crtxg = cvd->vdev_crtxg;
792
793 vdev_remove_child(pvd, cvd);
794 vdev_add_child(pvd, mvd);
795 cvd->vdev_id = mvd->vdev_children;
796 vdev_add_child(mvd, cvd);
797 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
798
799 if (mvd == mvd->vdev_top)
800 vdev_top_transfer(cvd, mvd);
801
802 return (mvd);
803 }
804
805 /*
806 * Remove a 1-way mirror/replacing vdev from the tree.
807 */
808 void
809 vdev_remove_parent(vdev_t *cvd)
810 {
811 vdev_t *mvd = cvd->vdev_parent;
812 vdev_t *pvd = mvd->vdev_parent;
813
814 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
815
816 ASSERT(mvd->vdev_children == 1);
817 ASSERT(mvd->vdev_ops == &vdev_mirror_ops ||
818 mvd->vdev_ops == &vdev_replacing_ops ||
819 mvd->vdev_ops == &vdev_spare_ops);
820 cvd->vdev_ashift = mvd->vdev_ashift;
821
822 vdev_remove_child(mvd, cvd);
823 vdev_remove_child(pvd, mvd);
824
825 /*
826 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
827 * Otherwise, we could have detached an offline device, and when we
828 * go to import the pool we'll think we have two top-level vdevs,
829 * instead of a different version of the same top-level vdev.
830 */
831 if (mvd->vdev_top == mvd) {
832 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid;
833 cvd->vdev_orig_guid = cvd->vdev_guid;
834 cvd->vdev_guid += guid_delta;
835 cvd->vdev_guid_sum += guid_delta;
836 }
837 cvd->vdev_id = mvd->vdev_id;
838 vdev_add_child(pvd, cvd);
839 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
840
841 if (cvd == cvd->vdev_top)
842 vdev_top_transfer(mvd, cvd);
843
844 ASSERT(mvd->vdev_children == 0);
845 vdev_free(mvd);
846 }
847
848 int
849 vdev_metaslab_init(vdev_t *vd, uint64_t txg)
850 {
851 spa_t *spa = vd->vdev_spa;
852 objset_t *mos = spa->spa_meta_objset;
853 uint64_t m;
854 uint64_t oldc = vd->vdev_ms_count;
855 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift;
856 metaslab_t **mspp;
857 int error;
858
859 ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER));
860
861 /*
862 * This vdev is not being allocated from yet or is a hole.
863 */
864 if (vd->vdev_ms_shift == 0)
865 return (0);
866
867 ASSERT(!vd->vdev_ishole);
868
869 /*
870 * Compute the raidz-deflation ratio. Note, we hard-code
871 * in 128k (1 << 17) because it is the "typical" blocksize.
872 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change,
873 * otherwise it would inconsistently account for existing bp's.
874 */
875 vd->vdev_deflate_ratio = (1 << 17) /
876 (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT);
877
878 ASSERT(oldc <= newc);
879
880 mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP);
881
882 if (oldc != 0) {
883 bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp));
884 kmem_free(vd->vdev_ms, oldc * sizeof (*mspp));
885 }
886
887 vd->vdev_ms = mspp;
888 vd->vdev_ms_count = newc;
889
890 for (m = oldc; m < newc; m++) {
891 uint64_t object = 0;
892
893 if (txg == 0) {
894 error = dmu_read(mos, vd->vdev_ms_array,
895 m * sizeof (uint64_t), sizeof (uint64_t), &object,
896 DMU_READ_PREFETCH);
897 if (error)
898 return (error);
899 }
900
901 error = metaslab_init(vd->vdev_mg, m, object, txg,
902 &(vd->vdev_ms[m]));
903 if (error)
904 return (error);
905 }
906
907 if (txg == 0)
908 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER);
909
910 /*
911 * If the vdev is being removed we don't activate
912 * the metaslabs since we want to ensure that no new
913 * allocations are performed on this device.
914 */
915 if (oldc == 0 && !vd->vdev_removing)
916 metaslab_group_activate(vd->vdev_mg);
917
918 if (txg == 0)
919 spa_config_exit(spa, SCL_ALLOC, FTAG);
920
921 return (0);
922 }
923
924 void
925 vdev_metaslab_fini(vdev_t *vd)
926 {
927 uint64_t m;
928 uint64_t count = vd->vdev_ms_count;
929
930 if (vd->vdev_ms != NULL) {
931 metaslab_group_passivate(vd->vdev_mg);
932 for (m = 0; m < count; m++) {
933 metaslab_t *msp = vd->vdev_ms[m];
934
935 if (msp != NULL)
936 metaslab_fini(msp);
937 }
938 kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *));
939 vd->vdev_ms = NULL;
940 }
941 }
942
943 typedef struct vdev_probe_stats {
944 boolean_t vps_readable;
945 boolean_t vps_writeable;
946 int vps_flags;
947 } vdev_probe_stats_t;
948
949 static void
950 vdev_probe_done(zio_t *zio)
951 {
952 spa_t *spa = zio->io_spa;
953 vdev_t *vd = zio->io_vd;
954 vdev_probe_stats_t *vps = zio->io_private;
955
956 ASSERT(vd->vdev_probe_zio != NULL);
957
958 if (zio->io_type == ZIO_TYPE_READ) {
959 if (zio->io_error == 0)
960 vps->vps_readable = 1;
961 if (zio->io_error == 0 && spa_writeable(spa)) {
962 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd,
963 zio->io_offset, zio->io_size, zio->io_data,
964 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
965 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE));
966 } else {
967 zio_buf_free(zio->io_data, zio->io_size);
968 }
969 } else if (zio->io_type == ZIO_TYPE_WRITE) {
970 if (zio->io_error == 0)
971 vps->vps_writeable = 1;
972 zio_buf_free(zio->io_data, zio->io_size);
973 } else if (zio->io_type == ZIO_TYPE_NULL) {
974 zio_t *pio;
975
976 vd->vdev_cant_read |= !vps->vps_readable;
977 vd->vdev_cant_write |= !vps->vps_writeable;
978
979 if (vdev_readable(vd) &&
980 (vdev_writeable(vd) || !spa_writeable(spa))) {
981 zio->io_error = 0;
982 } else {
983 ASSERT(zio->io_error != 0);
984 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE,
985 spa, vd, NULL, 0, 0);
986 zio->io_error = SET_ERROR(ENXIO);
987 }
988
989 mutex_enter(&vd->vdev_probe_lock);
990 ASSERT(vd->vdev_probe_zio == zio);
991 vd->vdev_probe_zio = NULL;
992 mutex_exit(&vd->vdev_probe_lock);
993
994 zio_link_t *zl = NULL;
995 while ((pio = zio_walk_parents(zio, &zl)) != NULL)
996 if (!vdev_accessible(vd, pio))
997 pio->io_error = SET_ERROR(ENXIO);
998
999 kmem_free(vps, sizeof (*vps));
1000 }
1001 }
1002
1003 /*
1004 * Determine whether this device is accessible.
1005 *
1006 * Read and write to several known locations: the pad regions of each
1007 * vdev label but the first, which we leave alone in case it contains
1008 * a VTOC.
1009 */
1010 zio_t *
1011 vdev_probe(vdev_t *vd, zio_t *zio)
1012 {
1013 spa_t *spa = vd->vdev_spa;
1014 vdev_probe_stats_t *vps = NULL;
1015 zio_t *pio;
1016
1017 ASSERT(vd->vdev_ops->vdev_op_leaf);
1018
1019 /*
1020 * Don't probe the probe.
1021 */
1022 if (zio && (zio->io_flags & ZIO_FLAG_PROBE))
1023 return (NULL);
1024
1025 /*
1026 * To prevent 'probe storms' when a device fails, we create
1027 * just one probe i/o at a time. All zios that want to probe
1028 * this vdev will become parents of the probe io.
1029 */
1030 mutex_enter(&vd->vdev_probe_lock);
1031
1032 if ((pio = vd->vdev_probe_zio) == NULL) {
1033 vps = kmem_zalloc(sizeof (*vps), KM_SLEEP);
1034
1035 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE |
1036 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE |
1037 ZIO_FLAG_TRYHARD;
1038
1039 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) {
1040 /*
1041 * vdev_cant_read and vdev_cant_write can only
1042 * transition from TRUE to FALSE when we have the
1043 * SCL_ZIO lock as writer; otherwise they can only
1044 * transition from FALSE to TRUE. This ensures that
1045 * any zio looking at these values can assume that
1046 * failures persist for the life of the I/O. That's
1047 * important because when a device has intermittent
1048 * connectivity problems, we want to ensure that
1049 * they're ascribed to the device (ENXIO) and not
1050 * the zio (EIO).
1051 *
1052 * Since we hold SCL_ZIO as writer here, clear both
1053 * values so the probe can reevaluate from first
1054 * principles.
1055 */
1056 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER;
1057 vd->vdev_cant_read = B_FALSE;
1058 vd->vdev_cant_write = B_FALSE;
1059 }
1060
1061 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd,
1062 vdev_probe_done, vps,
1063 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE);
1064
1065 /*
1066 * We can't change the vdev state in this context, so we
1067 * kick off an async task to do it on our behalf.
1068 */
1069 if (zio != NULL) {
1070 vd->vdev_probe_wanted = B_TRUE;
1071 spa_async_request(spa, SPA_ASYNC_PROBE);
1072 }
1073 }
1074
1075 if (zio != NULL)
1076 zio_add_child(zio, pio);
1077
1078 mutex_exit(&vd->vdev_probe_lock);
1079
1080 if (vps == NULL) {
1081 ASSERT(zio != NULL);
1082 return (NULL);
1083 }
1084
1085 for (int l = 1; l < VDEV_LABELS; l++) {
1086 zio_nowait(zio_read_phys(pio, vd,
1087 vdev_label_offset(vd->vdev_psize, l,
1088 offsetof(vdev_label_t, vl_pad2)),
1089 VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE),
1090 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1091 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE));
1092 }
1093
1094 if (zio == NULL)
1095 return (pio);
1096
1097 zio_nowait(pio);
1098 return (NULL);
1099 }
1100
1101 static void
1102 vdev_open_child(void *arg)
1103 {
1104 vdev_t *vd = arg;
1105
1106 vd->vdev_open_thread = curthread;
1107 vd->vdev_open_error = vdev_open(vd);
1108 vd->vdev_open_thread = NULL;
1109 vd->vdev_parent->vdev_nonrot &= vd->vdev_nonrot;
1110 }
1111
1112 boolean_t
1113 vdev_uses_zvols(vdev_t *vd)
1114 {
1115 if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR,
1116 strlen(ZVOL_DIR)) == 0)
1117 return (B_TRUE);
1118 for (int c = 0; c < vd->vdev_children; c++)
1119 if (vdev_uses_zvols(vd->vdev_child[c]))
1120 return (B_TRUE);
1121 return (B_FALSE);
1122 }
1123
1124 void
1125 vdev_open_children(vdev_t *vd)
1126 {
1127 taskq_t *tq;
1128 int children = vd->vdev_children;
1129
1130 vd->vdev_nonrot = B_TRUE;
1131
1132 /*
1133 * in order to handle pools on top of zvols, do the opens
1134 * in a single thread so that the same thread holds the
1135 * spa_namespace_lock
1136 */
1137 if (vdev_uses_zvols(vd)) {
1138 for (int c = 0; c < children; c++) {
1139 vd->vdev_child[c]->vdev_open_error =
1140 vdev_open(vd->vdev_child[c]);
1141 vd->vdev_nonrot &= vd->vdev_child[c]->vdev_nonrot;
1142 }
1143 return;
1144 }
1145 tq = taskq_create("vdev_open", children, minclsyspri,
1146 children, children, TASKQ_PREPOPULATE);
1147
1148 for (int c = 0; c < children; c++)
1149 VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c],
1150 TQ_SLEEP) != NULL);
1151
1152 taskq_destroy(tq);
1153
1154 for (int c = 0; c < children; c++)
1155 vd->vdev_nonrot &= vd->vdev_child[c]->vdev_nonrot;
1156 }
1157
1158 /*
1159 * Prepare a virtual device for access.
1160 */
1161 int
1162 vdev_open(vdev_t *vd)
1163 {
1164 spa_t *spa = vd->vdev_spa;
1165 int error;
1166 uint64_t osize = 0;
1167 uint64_t max_osize = 0;
1168 uint64_t asize, max_asize, psize;
1169 uint64_t ashift = 0;
1170
1171 ASSERT(vd->vdev_open_thread == curthread ||
1172 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1173 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED ||
1174 vd->vdev_state == VDEV_STATE_CANT_OPEN ||
1175 vd->vdev_state == VDEV_STATE_OFFLINE);
1176
1177 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1178 vd->vdev_cant_read = B_FALSE;
1179 vd->vdev_cant_write = B_FALSE;
1180 vd->vdev_min_asize = vdev_get_min_asize(vd);
1181
1182 /*
1183 * If this vdev is not removed, check its fault status. If it's
1184 * faulted, bail out of the open.
1185 */
1186 if (!vd->vdev_removed && vd->vdev_faulted) {
1187 ASSERT(vd->vdev_children == 0);
1188 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1189 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1190 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1191 vd->vdev_label_aux);
1192 return (SET_ERROR(ENXIO));
1193 } else if (vd->vdev_offline) {
1194 ASSERT(vd->vdev_children == 0);
1195 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE);
1196 return (SET_ERROR(ENXIO));
1197 }
1198
1199 error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize, &ashift);
1200
1201 /*
1202 * Reset the vdev_reopening flag so that we actually close
1203 * the vdev on error.
1204 */
1205 vd->vdev_reopening = B_FALSE;
1206 if (zio_injection_enabled && error == 0)
1207 error = zio_handle_device_injection(vd, NULL, ENXIO);
1208
1209 if (error) {
1210 if (vd->vdev_removed &&
1211 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED)
1212 vd->vdev_removed = B_FALSE;
1213
1214 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1215 vd->vdev_stat.vs_aux);
1216 return (error);
1217 }
1218
1219 vd->vdev_removed = B_FALSE;
1220
1221 /*
1222 * Recheck the faulted flag now that we have confirmed that
1223 * the vdev is accessible. If we're faulted, bail.
1224 */
1225 if (vd->vdev_faulted) {
1226 ASSERT(vd->vdev_children == 0);
1227 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1228 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1229 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1230 vd->vdev_label_aux);
1231 return (SET_ERROR(ENXIO));
1232 }
1233
1234 if (vd->vdev_degraded) {
1235 ASSERT(vd->vdev_children == 0);
1236 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1237 VDEV_AUX_ERR_EXCEEDED);
1238 } else {
1239 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0);
1240 }
1241
1242 /*
1243 * For hole or missing vdevs we just return success.
1244 */
1245 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops)
1246 return (0);
1247
1248 for (int c = 0; c < vd->vdev_children; c++) {
1249 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) {
1250 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1251 VDEV_AUX_NONE);
1252 break;
1253 }
1254 }
1255
1256 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t));
1257 max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t));
1258
1259 if (vd->vdev_children == 0) {
1260 if (osize < SPA_MINDEVSIZE) {
1261 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1262 VDEV_AUX_TOO_SMALL);
1263 return (SET_ERROR(EOVERFLOW));
1264 }
1265 psize = osize;
1266 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE);
1267 max_asize = max_osize - (VDEV_LABEL_START_SIZE +
1268 VDEV_LABEL_END_SIZE);
1269 } else {
1270 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE -
1271 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) {
1272 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1273 VDEV_AUX_TOO_SMALL);
1274 return (SET_ERROR(EOVERFLOW));
1275 }
1276 psize = 0;
1277 asize = osize;
1278 max_asize = max_osize;
1279 }
1280
1281 vd->vdev_psize = psize;
1282
1283 /*
1284 * Make sure the allocatable size hasn't shrunk.
1285 */
1286 if (asize < vd->vdev_min_asize) {
1287 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1288 VDEV_AUX_BAD_LABEL);
1289 return (SET_ERROR(EINVAL));
1290 }
1291
1292 if (vd->vdev_asize == 0) {
1293 /*
1294 * This is the first-ever open, so use the computed values.
1295 * For testing purposes, a higher ashift can be requested.
1296 */
1297 vd->vdev_asize = asize;
1298 vd->vdev_max_asize = max_asize;
1299 vd->vdev_ashift = MAX(ashift, vd->vdev_ashift);
1300 } else {
1301 /*
1302 * Detect if the alignment requirement has increased.
1303 * We don't want to make the pool unavailable, just
1304 * issue a warning instead.
1305 */
1306 if (ashift > vd->vdev_top->vdev_ashift &&
1307 vd->vdev_ops->vdev_op_leaf) {
1308 cmn_err(CE_WARN,
1309 "Disk, '%s', has a block alignment that is "
1310 "larger than the pool's alignment\n",
1311 vd->vdev_path);
1312 }
1313 vd->vdev_max_asize = max_asize;
1314 }
1315
1316 /*
1317 * If all children are healthy and the asize has increased,
1318 * then we've experienced dynamic LUN growth. If automatic
1319 * expansion is enabled then use the additional space.
1320 */
1321 if (vd->vdev_state == VDEV_STATE_HEALTHY && asize > vd->vdev_asize &&
1322 (vd->vdev_expanding || spa->spa_autoexpand))
1323 vd->vdev_asize = asize;
1324
1325 vdev_set_min_asize(vd);
1326
1327 /*
1328 * Ensure we can issue some IO before declaring the
1329 * vdev open for business.
1330 */
1331 if (vd->vdev_ops->vdev_op_leaf &&
1332 (error = zio_wait(vdev_probe(vd, NULL))) != 0) {
1333 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1334 VDEV_AUX_ERR_EXCEEDED);
1335 return (error);
1336 }
1337
1338 /*
1339 * Track the min and max ashift values for normal data devices.
1340 */
1341 if (vd->vdev_top == vd && vd->vdev_ashift != 0 &&
1342 !vd->vdev_islog && vd->vdev_aux == NULL) {
1343 if (vd->vdev_ashift > spa->spa_max_ashift)
1344 spa->spa_max_ashift = vd->vdev_ashift;
1345 if (vd->vdev_ashift < spa->spa_min_ashift)
1346 spa->spa_min_ashift = vd->vdev_ashift;
1347 }
1348
1349 /*
1350 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1351 * resilver. But don't do this if we are doing a reopen for a scrub,
1352 * since this would just restart the scrub we are already doing.
1353 */
1354 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen &&
1355 vdev_resilver_needed(vd, NULL, NULL))
1356 spa_async_request(spa, SPA_ASYNC_RESILVER);
1357
1358 return (0);
1359 }
1360
1361 /*
1362 * Called once the vdevs are all opened, this routine validates the label
1363 * contents. This needs to be done before vdev_load() so that we don't
1364 * inadvertently do repair I/Os to the wrong device.
1365 *
1366 * If 'strict' is false ignore the spa guid check. This is necessary because
1367 * if the machine crashed during a re-guid the new guid might have been written
1368 * to all of the vdev labels, but not the cached config. The strict check
1369 * will be performed when the pool is opened again using the mos config.
1370 *
1371 * This function will only return failure if one of the vdevs indicates that it
1372 * has since been destroyed or exported. This is only possible if
1373 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1374 * will be updated but the function will return 0.
1375 */
1376 int
1377 vdev_validate(vdev_t *vd, boolean_t strict)
1378 {
1379 spa_t *spa = vd->vdev_spa;
1380 nvlist_t *label;
1381 uint64_t guid = 0, top_guid;
1382 uint64_t state;
1383
1384 for (int c = 0; c < vd->vdev_children; c++)
1385 if (vdev_validate(vd->vdev_child[c], strict) != 0)
1386 return (SET_ERROR(EBADF));
1387
1388 /*
1389 * If the device has already failed, or was marked offline, don't do
1390 * any further validation. Otherwise, label I/O will fail and we will
1391 * overwrite the previous state.
1392 */
1393 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1394 uint64_t aux_guid = 0;
1395 nvlist_t *nvl;
1396 uint64_t txg = spa_last_synced_txg(spa) != 0 ?
1397 spa_last_synced_txg(spa) : -1ULL;
1398
1399 if ((label = vdev_label_read_config(vd, txg)) == NULL) {
1400 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1401 VDEV_AUX_BAD_LABEL);
1402 return (0);
1403 }
1404
1405 /*
1406 * Determine if this vdev has been split off into another
1407 * pool. If so, then refuse to open it.
1408 */
1409 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID,
1410 &aux_guid) == 0 && aux_guid == spa_guid(spa)) {
1411 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1412 VDEV_AUX_SPLIT_POOL);
1413 nvlist_free(label);
1414 return (0);
1415 }
1416
1417 if (strict && (nvlist_lookup_uint64(label,
1418 ZPOOL_CONFIG_POOL_GUID, &guid) != 0 ||
1419 guid != spa_guid(spa))) {
1420 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1421 VDEV_AUX_CORRUPT_DATA);
1422 nvlist_free(label);
1423 return (0);
1424 }
1425
1426 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl)
1427 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID,
1428 &aux_guid) != 0)
1429 aux_guid = 0;
1430
1431 /*
1432 * If this vdev just became a top-level vdev because its
1433 * sibling was detached, it will have adopted the parent's
1434 * vdev guid -- but the label may or may not be on disk yet.
1435 * Fortunately, either version of the label will have the
1436 * same top guid, so if we're a top-level vdev, we can
1437 * safely compare to that instead.
1438 *
1439 * If we split this vdev off instead, then we also check the
1440 * original pool's guid. We don't want to consider the vdev
1441 * corrupt if it is partway through a split operation.
1442 */
1443 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
1444 &guid) != 0 ||
1445 nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID,
1446 &top_guid) != 0 ||
1447 ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) &&
1448 (vd->vdev_guid != top_guid || vd != vd->vdev_top))) {
1449 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1450 VDEV_AUX_CORRUPT_DATA);
1451 nvlist_free(label);
1452 return (0);
1453 }
1454
1455 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1456 &state) != 0) {
1457 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1458 VDEV_AUX_CORRUPT_DATA);
1459 nvlist_free(label);
1460 return (0);
1461 }
1462
1463 nvlist_free(label);
1464
1465 /*
1466 * If this is a verbatim import, no need to check the
1467 * state of the pool.
1468 */
1469 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) &&
1470 spa_load_state(spa) == SPA_LOAD_OPEN &&
1471 state != POOL_STATE_ACTIVE)
1472 return (SET_ERROR(EBADF));
1473
1474 /*
1475 * If we were able to open and validate a vdev that was
1476 * previously marked permanently unavailable, clear that state
1477 * now.
1478 */
1479 if (vd->vdev_not_present)
1480 vd->vdev_not_present = 0;
1481 }
1482
1483 return (0);
1484 }
1485
1486 /*
1487 * Close a virtual device.
1488 */
1489 void
1490 vdev_close(vdev_t *vd)
1491 {
1492 spa_t *spa = vd->vdev_spa;
1493 vdev_t *pvd = vd->vdev_parent;
1494
1495 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1496
1497 /*
1498 * If our parent is reopening, then we are as well, unless we are
1499 * going offline.
1500 */
1501 if (pvd != NULL && pvd->vdev_reopening)
1502 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline);
1503
1504 vd->vdev_ops->vdev_op_close(vd);
1505
1506 vdev_cache_purge(vd);
1507
1508 /*
1509 * We record the previous state before we close it, so that if we are
1510 * doing a reopen(), we don't generate FMA ereports if we notice that
1511 * it's still faulted.
1512 */
1513 vd->vdev_prevstate = vd->vdev_state;
1514
1515 if (vd->vdev_offline)
1516 vd->vdev_state = VDEV_STATE_OFFLINE;
1517 else
1518 vd->vdev_state = VDEV_STATE_CLOSED;
1519 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1520 }
1521
1522 void
1523 vdev_hold(vdev_t *vd)
1524 {
1525 spa_t *spa = vd->vdev_spa;
1526
1527 ASSERT(spa_is_root(spa));
1528 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1529 return;
1530
1531 for (int c = 0; c < vd->vdev_children; c++)
1532 vdev_hold(vd->vdev_child[c]);
1533
1534 if (vd->vdev_ops->vdev_op_leaf)
1535 vd->vdev_ops->vdev_op_hold(vd);
1536 }
1537
1538 void
1539 vdev_rele(vdev_t *vd)
1540 {
1541 spa_t *spa = vd->vdev_spa;
1542
1543 ASSERT(spa_is_root(spa));
1544 for (int c = 0; c < vd->vdev_children; c++)
1545 vdev_rele(vd->vdev_child[c]);
1546
1547 if (vd->vdev_ops->vdev_op_leaf)
1548 vd->vdev_ops->vdev_op_rele(vd);
1549 }
1550
1551 /*
1552 * Reopen all interior vdevs and any unopened leaves. We don't actually
1553 * reopen leaf vdevs which had previously been opened as they might deadlock
1554 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1555 * If the leaf has never been opened then open it, as usual.
1556 */
1557 void
1558 vdev_reopen(vdev_t *vd)
1559 {
1560 spa_t *spa = vd->vdev_spa;
1561
1562 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1563
1564 /* set the reopening flag unless we're taking the vdev offline */
1565 vd->vdev_reopening = !vd->vdev_offline;
1566 vdev_close(vd);
1567 (void) vdev_open(vd);
1568
1569 /*
1570 * Call vdev_validate() here to make sure we have the same device.
1571 * Otherwise, a device with an invalid label could be successfully
1572 * opened in response to vdev_reopen().
1573 */
1574 if (vd->vdev_aux) {
1575 (void) vdev_validate_aux(vd);
1576 if (vdev_readable(vd) && vdev_writeable(vd) &&
1577 vd->vdev_aux == &spa->spa_l2cache &&
1578 !l2arc_vdev_present(vd))
1579 l2arc_add_vdev(spa, vd);
1580 } else {
1581 (void) vdev_validate(vd, B_TRUE);
1582 }
1583
1584 /*
1585 * Reassess parent vdev's health.
1586 */
1587 vdev_propagate_state(vd);
1588 }
1589
1590 int
1591 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
1592 {
1593 int error;
1594
1595 /*
1596 * Normally, partial opens (e.g. of a mirror) are allowed.
1597 * For a create, however, we want to fail the request if
1598 * there are any components we can't open.
1599 */
1600 error = vdev_open(vd);
1601
1602 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) {
1603 vdev_close(vd);
1604 return (error ? error : ENXIO);
1605 }
1606
1607 /*
1608 * Recursively load DTLs and initialize all labels.
1609 */
1610 if ((error = vdev_dtl_load(vd)) != 0 ||
1611 (error = vdev_label_init(vd, txg, isreplacing ?
1612 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
1613 vdev_close(vd);
1614 return (error);
1615 }
1616
1617 return (0);
1618 }
1619
1620 void
1621 vdev_metaslab_set_size(vdev_t *vd)
1622 {
1623 /*
1624 * Aim for roughly metaslabs_per_vdev (default 200) metaslabs per vdev.
1625 */
1626 vd->vdev_ms_shift = highbit64(vd->vdev_asize / metaslabs_per_vdev);
1627 vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT);
1628 }
1629
1630 void
1631 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
1632 {
1633 ASSERT(vd == vd->vdev_top);
1634 ASSERT(!vd->vdev_ishole);
1635 ASSERT(ISP2(flags));
1636 ASSERT(spa_writeable(vd->vdev_spa));
1637
1638 if (flags & VDD_METASLAB)
1639 (void) txg_list_add(&vd->vdev_ms_list, arg, txg);
1640
1641 if (flags & VDD_DTL)
1642 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg);
1643
1644 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
1645 }
1646
1647 void
1648 vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg)
1649 {
1650 for (int c = 0; c < vd->vdev_children; c++)
1651 vdev_dirty_leaves(vd->vdev_child[c], flags, txg);
1652
1653 if (vd->vdev_ops->vdev_op_leaf)
1654 vdev_dirty(vd->vdev_top, flags, vd, txg);
1655 }
1656
1657 /*
1658 * DTLs.
1659 *
1660 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1661 * the vdev has less than perfect replication. There are four kinds of DTL:
1662 *
1663 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1664 *
1665 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1666 *
1667 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1668 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1669 * txgs that was scrubbed.
1670 *
1671 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1672 * persistent errors or just some device being offline.
1673 * Unlike the other three, the DTL_OUTAGE map is not generally
1674 * maintained; it's only computed when needed, typically to
1675 * determine whether a device can be detached.
1676 *
1677 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1678 * either has the data or it doesn't.
1679 *
1680 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1681 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1682 * if any child is less than fully replicated, then so is its parent.
1683 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1684 * comprising only those txgs which appear in 'maxfaults' or more children;
1685 * those are the txgs we don't have enough replication to read. For example,
1686 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1687 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1688 * two child DTL_MISSING maps.
1689 *
1690 * It should be clear from the above that to compute the DTLs and outage maps
1691 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1692 * Therefore, that is all we keep on disk. When loading the pool, or after
1693 * a configuration change, we generate all other DTLs from first principles.
1694 */
1695 void
1696 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1697 {
1698 range_tree_t *rt = vd->vdev_dtl[t];
1699
1700 ASSERT(t < DTL_TYPES);
1701 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1702 ASSERT(spa_writeable(vd->vdev_spa));
1703
1704 mutex_enter(rt->rt_lock);
1705 if (!range_tree_contains(rt, txg, size))
1706 range_tree_add(rt, txg, size);
1707 mutex_exit(rt->rt_lock);
1708 }
1709
1710 boolean_t
1711 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1712 {
1713 range_tree_t *rt = vd->vdev_dtl[t];
1714 boolean_t dirty = B_FALSE;
1715
1716 ASSERT(t < DTL_TYPES);
1717 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1718
1719 mutex_enter(rt->rt_lock);
1720 if (range_tree_space(rt) != 0)
1721 dirty = range_tree_contains(rt, txg, size);
1722 mutex_exit(rt->rt_lock);
1723
1724 return (dirty);
1725 }
1726
1727 boolean_t
1728 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t)
1729 {
1730 range_tree_t *rt = vd->vdev_dtl[t];
1731 boolean_t empty;
1732
1733 mutex_enter(rt->rt_lock);
1734 empty = (range_tree_space(rt) == 0);
1735 mutex_exit(rt->rt_lock);
1736
1737 return (empty);
1738 }
1739
1740 /*
1741 * Returns the lowest txg in the DTL range.
1742 */
1743 static uint64_t
1744 vdev_dtl_min(vdev_t *vd)
1745 {
1746 range_seg_t *rs;
1747
1748 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
1749 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
1750 ASSERT0(vd->vdev_children);
1751
1752 rs = avl_first(&vd->vdev_dtl[DTL_MISSING]->rt_root);
1753 return (rs->rs_start - 1);
1754 }
1755
1756 /*
1757 * Returns the highest txg in the DTL.
1758 */
1759 static uint64_t
1760 vdev_dtl_max(vdev_t *vd)
1761 {
1762 range_seg_t *rs;
1763
1764 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
1765 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
1766 ASSERT0(vd->vdev_children);
1767
1768 rs = avl_last(&vd->vdev_dtl[DTL_MISSING]->rt_root);
1769 return (rs->rs_end);
1770 }
1771
1772 /*
1773 * Determine if a resilvering vdev should remove any DTL entries from
1774 * its range. If the vdev was resilvering for the entire duration of the
1775 * scan then it should excise that range from its DTLs. Otherwise, this
1776 * vdev is considered partially resilvered and should leave its DTL
1777 * entries intact. The comment in vdev_dtl_reassess() describes how we
1778 * excise the DTLs.
1779 */
1780 static boolean_t
1781 vdev_dtl_should_excise(vdev_t *vd)
1782 {
1783 spa_t *spa = vd->vdev_spa;
1784 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
1785
1786 ASSERT0(scn->scn_phys.scn_errors);
1787 ASSERT0(vd->vdev_children);
1788
1789 if (vd->vdev_resilver_txg == 0 ||
1790 range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0)
1791 return (B_TRUE);
1792
1793 /*
1794 * When a resilver is initiated the scan will assign the scn_max_txg
1795 * value to the highest txg value that exists in all DTLs. If this
1796 * device's max DTL is not part of this scan (i.e. it is not in
1797 * the range (scn_min_txg, scn_max_txg] then it is not eligible
1798 * for excision.
1799 */
1800 if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) {
1801 ASSERT3U(scn->scn_phys.scn_min_txg, <=, vdev_dtl_min(vd));
1802 ASSERT3U(scn->scn_phys.scn_min_txg, <, vd->vdev_resilver_txg);
1803 ASSERT3U(vd->vdev_resilver_txg, <=, scn->scn_phys.scn_max_txg);
1804 return (B_TRUE);
1805 }
1806 return (B_FALSE);
1807 }
1808
1809 /*
1810 * Reassess DTLs after a config change or scrub completion.
1811 */
1812 void
1813 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
1814 {
1815 spa_t *spa = vd->vdev_spa;
1816 avl_tree_t reftree;
1817 int minref;
1818
1819 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1820
1821 for (int c = 0; c < vd->vdev_children; c++)
1822 vdev_dtl_reassess(vd->vdev_child[c], txg,
1823 scrub_txg, scrub_done);
1824
1825 if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux)
1826 return;
1827
1828 if (vd->vdev_ops->vdev_op_leaf) {
1829 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
1830
1831 mutex_enter(&vd->vdev_dtl_lock);
1832
1833 /*
1834 * If we've completed a scan cleanly then determine
1835 * if this vdev should remove any DTLs. We only want to
1836 * excise regions on vdevs that were available during
1837 * the entire duration of this scan.
1838 */
1839 if (scrub_txg != 0 &&
1840 (spa->spa_scrub_started ||
1841 (scn != NULL && scn->scn_phys.scn_errors == 0)) &&
1842 vdev_dtl_should_excise(vd)) {
1843 /*
1844 * We completed a scrub up to scrub_txg. If we
1845 * did it without rebooting, then the scrub dtl
1846 * will be valid, so excise the old region and
1847 * fold in the scrub dtl. Otherwise, leave the
1848 * dtl as-is if there was an error.
1849 *
1850 * There's little trick here: to excise the beginning
1851 * of the DTL_MISSING map, we put it into a reference
1852 * tree and then add a segment with refcnt -1 that
1853 * covers the range [0, scrub_txg). This means
1854 * that each txg in that range has refcnt -1 or 0.
1855 * We then add DTL_SCRUB with a refcnt of 2, so that
1856 * entries in the range [0, scrub_txg) will have a
1857 * positive refcnt -- either 1 or 2. We then convert
1858 * the reference tree into the new DTL_MISSING map.
1859 */
1860 space_reftree_create(&reftree);
1861 space_reftree_add_map(&reftree,
1862 vd->vdev_dtl[DTL_MISSING], 1);
1863 space_reftree_add_seg(&reftree, 0, scrub_txg, -1);
1864 space_reftree_add_map(&reftree,
1865 vd->vdev_dtl[DTL_SCRUB], 2);
1866 space_reftree_generate_map(&reftree,
1867 vd->vdev_dtl[DTL_MISSING], 1);
1868 space_reftree_destroy(&reftree);
1869 }
1870 range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
1871 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
1872 range_tree_add, vd->vdev_dtl[DTL_PARTIAL]);
1873 if (scrub_done)
1874 range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
1875 range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
1876 if (!vdev_readable(vd))
1877 range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
1878 else
1879 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
1880 range_tree_add, vd->vdev_dtl[DTL_OUTAGE]);
1881
1882 /*
1883 * If the vdev was resilvering and no longer has any
1884 * DTLs then reset its resilvering flag.
1885 */
1886 if (vd->vdev_resilver_txg != 0 &&
1887 range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0 &&
1888 range_tree_space(vd->vdev_dtl[DTL_OUTAGE]) == 0)
1889 vd->vdev_resilver_txg = 0;
1890
1891 mutex_exit(&vd->vdev_dtl_lock);
1892
1893 if (txg != 0)
1894 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
1895 return;
1896 }
1897
1898 mutex_enter(&vd->vdev_dtl_lock);
1899 for (int t = 0; t < DTL_TYPES; t++) {
1900 /* account for child's outage in parent's missing map */
1901 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t;
1902 if (t == DTL_SCRUB)
1903 continue; /* leaf vdevs only */
1904 if (t == DTL_PARTIAL)
1905 minref = 1; /* i.e. non-zero */
1906 else if (vd->vdev_nparity != 0)
1907 minref = vd->vdev_nparity + 1; /* RAID-Z */
1908 else
1909 minref = vd->vdev_children; /* any kind of mirror */
1910 space_reftree_create(&reftree);
1911 for (int c = 0; c < vd->vdev_children; c++) {
1912 vdev_t *cvd = vd->vdev_child[c];
1913 mutex_enter(&cvd->vdev_dtl_lock);
1914 space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1);
1915 mutex_exit(&cvd->vdev_dtl_lock);
1916 }
1917 space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref);
1918 space_reftree_destroy(&reftree);
1919 }
1920 mutex_exit(&vd->vdev_dtl_lock);
1921 }
1922
1923 int
1924 vdev_dtl_load(vdev_t *vd)
1925 {
1926 spa_t *spa = vd->vdev_spa;
1927 objset_t *mos = spa->spa_meta_objset;
1928 int error = 0;
1929
1930 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) {
1931 ASSERT(!vd->vdev_ishole);
1932
1933 error = space_map_open(&vd->vdev_dtl_sm, mos,
1934 vd->vdev_dtl_object, 0, -1ULL, 0, &vd->vdev_dtl_lock);
1935 if (error)
1936 return (error);
1937 ASSERT(vd->vdev_dtl_sm != NULL);
1938
1939 mutex_enter(&vd->vdev_dtl_lock);
1940
1941 /*
1942 * Now that we've opened the space_map we need to update
1943 * the in-core DTL.
1944 */
1945 space_map_update(vd->vdev_dtl_sm);
1946
1947 error = space_map_load(vd->vdev_dtl_sm,
1948 vd->vdev_dtl[DTL_MISSING], SM_ALLOC);
1949 mutex_exit(&vd->vdev_dtl_lock);
1950
1951 return (error);
1952 }
1953
1954 for (int c = 0; c < vd->vdev_children; c++) {
1955 error = vdev_dtl_load(vd->vdev_child[c]);
1956 if (error != 0)
1957 break;
1958 }
1959
1960 return (error);
1961 }
1962
1963 void
1964 vdev_destroy_unlink_zap(vdev_t *vd, uint64_t zapobj, dmu_tx_t *tx)
1965 {
1966 spa_t *spa = vd->vdev_spa;
1967
1968 VERIFY0(zap_destroy(spa->spa_meta_objset, zapobj, tx));
1969 VERIFY0(zap_remove_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps,
1970 zapobj, tx));
1971 }
1972
1973 uint64_t
1974 vdev_create_link_zap(vdev_t *vd, dmu_tx_t *tx)
1975 {
1976 spa_t *spa = vd->vdev_spa;
1977 uint64_t zap = zap_create(spa->spa_meta_objset, DMU_OTN_ZAP_METADATA,
1978 DMU_OT_NONE, 0, tx);
1979
1980 ASSERT(zap != 0);
1981 VERIFY0(zap_add_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps,
1982 zap, tx));
1983
1984 return (zap);
1985 }
1986
1987 void
1988 vdev_construct_zaps(vdev_t *vd, dmu_tx_t *tx)
1989 {
1990 if (vd->vdev_ops != &vdev_hole_ops &&
1991 vd->vdev_ops != &vdev_missing_ops &&
1992 vd->vdev_ops != &vdev_root_ops &&
1993 !vd->vdev_top->vdev_removing) {
1994 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_leaf_zap == 0) {
1995 vd->vdev_leaf_zap = vdev_create_link_zap(vd, tx);
1996 }
1997 if (vd == vd->vdev_top && vd->vdev_top_zap == 0) {
1998 vd->vdev_top_zap = vdev_create_link_zap(vd, tx);
1999 }
2000 }
2001 for (uint64_t i = 0; i < vd->vdev_children; i++) {
2002 vdev_construct_zaps(vd->vdev_child[i], tx);
2003 }
2004 }
2005
2006 void
2007 vdev_dtl_sync(vdev_t *vd, uint64_t txg)
2008 {
2009 spa_t *spa = vd->vdev_spa;
2010 range_tree_t *rt = vd->vdev_dtl[DTL_MISSING];
2011 objset_t *mos = spa->spa_meta_objset;
2012 range_tree_t *rtsync;
2013 kmutex_t rtlock;
2014 dmu_tx_t *tx;
2015 uint64_t object = space_map_object(vd->vdev_dtl_sm);
2016
2017 ASSERT(!vd->vdev_ishole);
2018 ASSERT(vd->vdev_ops->vdev_op_leaf);
2019
2020 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2021
2022 if (vd->vdev_detached || vd->vdev_top->vdev_removing) {
2023 mutex_enter(&vd->vdev_dtl_lock);
2024 space_map_free(vd->vdev_dtl_sm, tx);
2025 space_map_close(vd->vdev_dtl_sm);
2026 vd->vdev_dtl_sm = NULL;
2027 mutex_exit(&vd->vdev_dtl_lock);
2028
2029 /*
2030 * We only destroy the leaf ZAP for detached leaves or for
2031 * removed log devices. Removed data devices handle leaf ZAP
2032 * cleanup later, once cancellation is no longer possible.
2033 */
2034 if (vd->vdev_leaf_zap != 0 && (vd->vdev_detached ||
2035 vd->vdev_top->vdev_islog)) {
2036 vdev_destroy_unlink_zap(vd, vd->vdev_leaf_zap, tx);
2037 vd->vdev_leaf_zap = 0;
2038 }
2039
2040 dmu_tx_commit(tx);
2041 return;
2042 }
2043
2044 if (vd->vdev_dtl_sm == NULL) {
2045 uint64_t new_object;
2046
2047 new_object = space_map_alloc(mos, tx);
2048 VERIFY3U(new_object, !=, 0);
2049
2050 VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object,
2051 0, -1ULL, 0, &vd->vdev_dtl_lock));
2052 ASSERT(vd->vdev_dtl_sm != NULL);
2053 }
2054
2055 mutex_init(&rtlock, NULL, MUTEX_DEFAULT, NULL);
2056
2057 rtsync = range_tree_create(NULL, NULL, &rtlock);
2058
2059 mutex_enter(&rtlock);
2060
2061 mutex_enter(&vd->vdev_dtl_lock);
2062 range_tree_walk(rt, range_tree_add, rtsync);
2063 mutex_exit(&vd->vdev_dtl_lock);
2064
2065 space_map_truncate(vd->vdev_dtl_sm, tx);
2066 space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, tx);
2067 range_tree_vacate(rtsync, NULL, NULL);
2068
2069 range_tree_destroy(rtsync);
2070
2071 mutex_exit(&rtlock);
2072 mutex_destroy(&rtlock);
2073
2074 /*
2075 * If the object for the space map has changed then dirty
2076 * the top level so that we update the config.
2077 */
2078 if (object != space_map_object(vd->vdev_dtl_sm)) {
2079 zfs_dbgmsg("txg %llu, spa %s, DTL old object %llu, "
2080 "new object %llu", txg, spa_name(spa), object,
2081 space_map_object(vd->vdev_dtl_sm));
2082 vdev_config_dirty(vd->vdev_top);
2083 }
2084
2085 dmu_tx_commit(tx);
2086
2087 mutex_enter(&vd->vdev_dtl_lock);
2088 space_map_update(vd->vdev_dtl_sm);
2089 mutex_exit(&vd->vdev_dtl_lock);
2090 }
2091
2092 /*
2093 * Determine whether the specified vdev can be offlined/detached/removed
2094 * without losing data.
2095 */
2096 boolean_t
2097 vdev_dtl_required(vdev_t *vd)
2098 {
2099 spa_t *spa = vd->vdev_spa;
2100 vdev_t *tvd = vd->vdev_top;
2101 uint8_t cant_read = vd->vdev_cant_read;
2102 boolean_t required;
2103
2104 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2105
2106 if (vd == spa->spa_root_vdev || vd == tvd)
2107 return (B_TRUE);
2108
2109 /*
2110 * Temporarily mark the device as unreadable, and then determine
2111 * whether this results in any DTL outages in the top-level vdev.
2112 * If not, we can safely offline/detach/remove the device.
2113 */
2114 vd->vdev_cant_read = B_TRUE;
2115 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
2116 required = !vdev_dtl_empty(tvd, DTL_OUTAGE);
2117 vd->vdev_cant_read = cant_read;
2118 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
2119
2120 if (!required && zio_injection_enabled)
2121 required = !!zio_handle_device_injection(vd, NULL, ECHILD);
2122
2123 return (required);
2124 }
2125
2126 /*
2127 * Determine if resilver is needed, and if so the txg range.
2128 */
2129 boolean_t
2130 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
2131 {
2132 boolean_t needed = B_FALSE;
2133 uint64_t thismin = UINT64_MAX;
2134 uint64_t thismax = 0;
2135
2136 if (vd->vdev_children == 0) {
2137 mutex_enter(&vd->vdev_dtl_lock);
2138 if (range_tree_space(vd->vdev_dtl[DTL_MISSING]) != 0 &&
2139 vdev_writeable(vd)) {
2140
2141 thismin = vdev_dtl_min(vd);
2142 thismax = vdev_dtl_max(vd);
2143 needed = B_TRUE;
2144 }
2145 mutex_exit(&vd->vdev_dtl_lock);
2146 } else {
2147 for (int c = 0; c < vd->vdev_children; c++) {
2148 vdev_t *cvd = vd->vdev_child[c];
2149 uint64_t cmin, cmax;
2150
2151 if (vdev_resilver_needed(cvd, &cmin, &cmax)) {
2152 thismin = MIN(thismin, cmin);
2153 thismax = MAX(thismax, cmax);
2154 needed = B_TRUE;
2155 }
2156 }
2157 }
2158
2159 if (needed && minp) {
2160 *minp = thismin;
2161 *maxp = thismax;
2162 }
2163 return (needed);
2164 }
2165
2166 void
2167 vdev_load(vdev_t *vd)
2168 {
2169 /*
2170 * Recursively load all children.
2171 */
2172 for (int c = 0; c < vd->vdev_children; c++)
2173 vdev_load(vd->vdev_child[c]);
2174
2175 /*
2176 * If this is a top-level vdev, initialize its metaslabs.
2177 */
2178 if (vd == vd->vdev_top && !vd->vdev_ishole &&
2179 (vd->vdev_ashift == 0 || vd->vdev_asize == 0 ||
2180 vdev_metaslab_init(vd, 0) != 0))
2181 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2182 VDEV_AUX_CORRUPT_DATA);
2183
2184 /*
2185 * If this is a leaf vdev, load its DTL.
2186 */
2187 if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0)
2188 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2189 VDEV_AUX_CORRUPT_DATA);
2190 }
2191
2192 /*
2193 * The special vdev case is used for hot spares and l2cache devices. Its
2194 * sole purpose it to set the vdev state for the associated vdev. To do this,
2195 * we make sure that we can open the underlying device, then try to read the
2196 * label, and make sure that the label is sane and that it hasn't been
2197 * repurposed to another pool.
2198 */
2199 int
2200 vdev_validate_aux(vdev_t *vd)
2201 {
2202 nvlist_t *label;
2203 uint64_t guid, version;
2204 uint64_t state;
2205
2206 if (!vdev_readable(vd))
2207 return (0);
2208
2209 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) {
2210 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2211 VDEV_AUX_CORRUPT_DATA);
2212 return (-1);
2213 }
2214
2215 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 ||
2216 !SPA_VERSION_IS_SUPPORTED(version) ||
2217 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 ||
2218 guid != vd->vdev_guid ||
2219 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) {
2220 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2221 VDEV_AUX_CORRUPT_DATA);
2222 nvlist_free(label);
2223 return (-1);
2224 }
2225
2226 /*
2227 * We don't actually check the pool state here. If it's in fact in
2228 * use by another pool, we update this fact on the fly when requested.
2229 */
2230 nvlist_free(label);
2231 return (0);
2232 }
2233
2234 void
2235 vdev_remove(vdev_t *vd, uint64_t txg)
2236 {
2237 spa_t *spa = vd->vdev_spa;
2238 objset_t *mos = spa->spa_meta_objset;
2239 dmu_tx_t *tx;
2240
2241 tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
2242 ASSERT(vd == vd->vdev_top);
2243 ASSERT3U(txg, ==, spa_syncing_txg(spa));
2244
2245 if (vd->vdev_ms != NULL) {
2246 metaslab_group_t *mg = vd->vdev_mg;
2247
2248 metaslab_group_histogram_verify(mg);
2249 metaslab_class_histogram_verify(mg->mg_class);
2250
2251 for (int m = 0; m < vd->vdev_ms_count; m++) {
2252 metaslab_t *msp = vd->vdev_ms[m];
2253
2254 if (msp == NULL || msp->ms_sm == NULL)
2255 continue;
2256
2257 mutex_enter(&msp->ms_lock);
2258 /*
2259 * If the metaslab was not loaded when the vdev
2260 * was removed then the histogram accounting may
2261 * not be accurate. Update the histogram information
2262 * here so that we ensure that the metaslab group
2263 * and metaslab class are up-to-date.
2264 */
2265 metaslab_group_histogram_remove(mg, msp);
2266
2267 VERIFY0(space_map_allocated(msp->ms_sm));
2268 space_map_free(msp->ms_sm, tx);
2269 space_map_close(msp->ms_sm);
2270 msp->ms_sm = NULL;
2271 mutex_exit(&msp->ms_lock);
2272 }
2273
2274 metaslab_group_histogram_verify(mg);
2275 metaslab_class_histogram_verify(mg->mg_class);
2276 for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++)
2277 ASSERT0(mg->mg_histogram[i]);
2278
2279 }
2280
2281 if (vd->vdev_ms_array) {
2282 (void) dmu_object_free(mos, vd->vdev_ms_array, tx);
2283 vd->vdev_ms_array = 0;
2284 }
2285
2286 if (vd->vdev_islog && vd->vdev_top_zap != 0) {
2287 vdev_destroy_unlink_zap(vd, vd->vdev_top_zap, tx);
2288 vd->vdev_top_zap = 0;
2289 }
2290 dmu_tx_commit(tx);
2291 }
2292
2293 void
2294 vdev_sync_done(vdev_t *vd, uint64_t txg)
2295 {
2296 metaslab_t *msp;
2297 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg));
2298
2299 ASSERT(!vd->vdev_ishole);
2300
2301 while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))
2302 metaslab_sync_done(msp, txg);
2303
2304 if (reassess)
2305 metaslab_sync_reassess(vd->vdev_mg);
2306 }
2307
2308 void
2309 vdev_sync(vdev_t *vd, uint64_t txg)
2310 {
2311 spa_t *spa = vd->vdev_spa;
2312 vdev_t *lvd;
2313 metaslab_t *msp;
2314 dmu_tx_t *tx;
2315
2316 ASSERT(!vd->vdev_ishole);
2317
2318 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) {
2319 ASSERT(vd == vd->vdev_top);
2320 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2321 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset,
2322 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx);
2323 ASSERT(vd->vdev_ms_array != 0);
2324 vdev_config_dirty(vd);
2325 dmu_tx_commit(tx);
2326 }
2327
2328 /*
2329 * Remove the metadata associated with this vdev once it's empty.
2330 */
2331 if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing)
2332 vdev_remove(vd, txg);
2333
2334 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) {
2335 metaslab_sync(msp, txg);
2336 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg));
2337 }
2338
2339 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL)
2340 vdev_dtl_sync(lvd, txg);
2341
2342 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg));
2343 }
2344
2345 uint64_t
2346 vdev_psize_to_asize(vdev_t *vd, uint64_t psize)
2347 {
2348 return (vd->vdev_ops->vdev_op_asize(vd, psize));
2349 }
2350
2351 /*
2352 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2353 * not be opened, and no I/O is attempted.
2354 */
2355 int
2356 vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2357 {
2358 vdev_t *vd, *tvd;
2359
2360 spa_vdev_state_enter(spa, SCL_NONE);
2361
2362 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2363 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2364
2365 if (!vd->vdev_ops->vdev_op_leaf)
2366 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2367
2368 tvd = vd->vdev_top;
2369
2370 /*
2371 * We don't directly use the aux state here, but if we do a
2372 * vdev_reopen(), we need this value to be present to remember why we
2373 * were faulted.
2374 */
2375 vd->vdev_label_aux = aux;
2376
2377 /*
2378 * Faulted state takes precedence over degraded.
2379 */
2380 vd->vdev_delayed_close = B_FALSE;
2381 vd->vdev_faulted = 1ULL;
2382 vd->vdev_degraded = 0ULL;
2383 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux);
2384
2385 /*
2386 * If this device has the only valid copy of the data, then
2387 * back off and simply mark the vdev as degraded instead.
2388 */
2389 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) {
2390 vd->vdev_degraded = 1ULL;
2391 vd->vdev_faulted = 0ULL;
2392
2393 /*
2394 * If we reopen the device and it's not dead, only then do we
2395 * mark it degraded.
2396 */
2397 vdev_reopen(tvd);
2398
2399 if (vdev_readable(vd))
2400 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux);
2401 }
2402
2403 return (spa_vdev_state_exit(spa, vd, 0));
2404 }
2405
2406 /*
2407 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2408 * user that something is wrong. The vdev continues to operate as normal as far
2409 * as I/O is concerned.
2410 */
2411 int
2412 vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2413 {
2414 vdev_t *vd;
2415
2416 spa_vdev_state_enter(spa, SCL_NONE);
2417
2418 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2419 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2420
2421 if (!vd->vdev_ops->vdev_op_leaf)
2422 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2423
2424 /*
2425 * If the vdev is already faulted, then don't do anything.
2426 */
2427 if (vd->vdev_faulted || vd->vdev_degraded)
2428 return (spa_vdev_state_exit(spa, NULL, 0));
2429
2430 vd->vdev_degraded = 1ULL;
2431 if (!vdev_is_dead(vd))
2432 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
2433 aux);
2434
2435 return (spa_vdev_state_exit(spa, vd, 0));
2436 }
2437
2438 /*
2439 * Online the given vdev.
2440 *
2441 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
2442 * spare device should be detached when the device finishes resilvering.
2443 * Second, the online should be treated like a 'test' online case, so no FMA
2444 * events are generated if the device fails to open.
2445 */
2446 int
2447 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
2448 {
2449 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev;
2450 boolean_t postevent = B_FALSE;
2451
2452 spa_vdev_state_enter(spa, SCL_NONE);
2453
2454 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2455 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2456
2457 if (!vd->vdev_ops->vdev_op_leaf)
2458 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2459
2460 postevent =
2461 (vd->vdev_offline == B_TRUE || vd->vdev_tmpoffline == B_TRUE) ?
2462 B_TRUE : B_FALSE;
2463
2464 tvd = vd->vdev_top;
2465 vd->vdev_offline = B_FALSE;
2466 vd->vdev_tmpoffline = B_FALSE;
2467 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE);
2468 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT);
2469
2470 /* XXX - L2ARC 1.0 does not support expansion */
2471 if (!vd->vdev_aux) {
2472 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2473 pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND);
2474 }
2475
2476 vdev_reopen(tvd);
2477 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE;
2478
2479 if (!vd->vdev_aux) {
2480 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2481 pvd->vdev_expanding = B_FALSE;
2482 }
2483
2484 if (newstate)
2485 *newstate = vd->vdev_state;
2486 if ((flags & ZFS_ONLINE_UNSPARE) &&
2487 !vdev_is_dead(vd) && vd->vdev_parent &&
2488 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2489 vd->vdev_parent->vdev_child[0] == vd)
2490 vd->vdev_unspare = B_TRUE;
2491
2492 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) {
2493
2494 /* XXX - L2ARC 1.0 does not support expansion */
2495 if (vd->vdev_aux)
2496 return (spa_vdev_state_exit(spa, vd, ENOTSUP));
2497 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2498 }
2499
2500 if (postevent)
2501 spa_event_notify(spa, vd, ESC_ZFS_VDEV_ONLINE);
2502
2503 return (spa_vdev_state_exit(spa, vd, 0));
2504 }
2505
2506 static int
2507 vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags)
2508 {
2509 vdev_t *vd, *tvd;
2510 int error = 0;
2511 uint64_t generation;
2512 metaslab_group_t *mg;
2513
2514 top:
2515 spa_vdev_state_enter(spa, SCL_ALLOC);
2516
2517 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2518 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2519
2520 if (!vd->vdev_ops->vdev_op_leaf)
2521 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2522
2523 tvd = vd->vdev_top;
2524 mg = tvd->vdev_mg;
2525 generation = spa->spa_config_generation + 1;
2526
2527 /*
2528 * If the device isn't already offline, try to offline it.
2529 */
2530 if (!vd->vdev_offline) {
2531 /*
2532 * If this device has the only valid copy of some data,
2533 * don't allow it to be offlined. Log devices are always
2534 * expendable.
2535 */
2536 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2537 vdev_dtl_required(vd))
2538 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2539
2540 /*
2541 * If the top-level is a slog and it has had allocations
2542 * then proceed. We check that the vdev's metaslab group
2543 * is not NULL since it's possible that we may have just
2544 * added this vdev but not yet initialized its metaslabs.
2545 */
2546 if (tvd->vdev_islog && mg != NULL) {
2547 /*
2548 * Prevent any future allocations.
2549 */
2550 metaslab_group_passivate(mg);
2551 (void) spa_vdev_state_exit(spa, vd, 0);
2552
2553 error = spa_offline_log(spa);
2554
2555 spa_vdev_state_enter(spa, SCL_ALLOC);
2556
2557 /*
2558 * Check to see if the config has changed.
2559 */
2560 if (error || generation != spa->spa_config_generation) {
2561 metaslab_group_activate(mg);
2562 if (error)
2563 return (spa_vdev_state_exit(spa,
2564 vd, error));
2565 (void) spa_vdev_state_exit(spa, vd, 0);
2566 goto top;
2567 }
2568 ASSERT0(tvd->vdev_stat.vs_alloc);
2569 }
2570
2571 /*
2572 * Offline this device and reopen its top-level vdev.
2573 * If the top-level vdev is a log device then just offline
2574 * it. Otherwise, if this action results in the top-level
2575 * vdev becoming unusable, undo it and fail the request.
2576 */
2577 vd->vdev_offline = B_TRUE;
2578 vdev_reopen(tvd);
2579
2580 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2581 vdev_is_dead(tvd)) {
2582 vd->vdev_offline = B_FALSE;
2583 vdev_reopen(tvd);
2584 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2585 }
2586
2587 /*
2588 * Add the device back into the metaslab rotor so that
2589 * once we online the device it's open for business.
2590 */
2591 if (tvd->vdev_islog && mg != NULL)
2592 metaslab_group_activate(mg);
2593 }
2594
2595 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY);
2596
2597 return (spa_vdev_state_exit(spa, vd, 0));
2598 }
2599
2600 int
2601 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags)
2602 {
2603 int error;
2604
2605 mutex_enter(&spa->spa_vdev_top_lock);
2606 error = vdev_offline_locked(spa, guid, flags);
2607 mutex_exit(&spa->spa_vdev_top_lock);
2608
2609 return (error);
2610 }
2611
2612 /*
2613 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2614 * vdev_offline(), we assume the spa config is locked. We also clear all
2615 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2616 */
2617 void
2618 vdev_clear(spa_t *spa, vdev_t *vd)
2619 {
2620 vdev_t *rvd = spa->spa_root_vdev;
2621
2622 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2623
2624 if (vd == NULL)
2625 vd = rvd;
2626
2627 vd->vdev_stat.vs_read_errors = 0;
2628 vd->vdev_stat.vs_write_errors = 0;
2629 vd->vdev_stat.vs_checksum_errors = 0;
2630
2631 for (int c = 0; c < vd->vdev_children; c++)
2632 vdev_clear(spa, vd->vdev_child[c]);
2633
2634 /*
2635 * If we're in the FAULTED state or have experienced failed I/O, then
2636 * clear the persistent state and attempt to reopen the device. We
2637 * also mark the vdev config dirty, so that the new faulted state is
2638 * written out to disk.
2639 */
2640 if (vd->vdev_faulted || vd->vdev_degraded ||
2641 !vdev_readable(vd) || !vdev_writeable(vd)) {
2642
2643 /*
2644 * When reopening in reponse to a clear event, it may be due to
2645 * a fmadm repair request. In this case, if the device is
2646 * still broken, we want to still post the ereport again.
2647 */
2648 vd->vdev_forcefault = B_TRUE;
2649
2650 vd->vdev_faulted = vd->vdev_degraded = 0ULL;
2651 vd->vdev_cant_read = B_FALSE;
2652 vd->vdev_cant_write = B_FALSE;
2653
2654 vdev_reopen(vd == rvd ? rvd : vd->vdev_top);
2655
2656 vd->vdev_forcefault = B_FALSE;
2657
2658 if (vd != rvd && vdev_writeable(vd->vdev_top))
2659 vdev_state_dirty(vd->vdev_top);
2660
2661 if (vd->vdev_aux == NULL && !vdev_is_dead(vd))
2662 spa_async_request(spa, SPA_ASYNC_RESILVER);
2663
2664 spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR);
2665 }
2666
2667 /*
2668 * When clearing a FMA-diagnosed fault, we always want to
2669 * unspare the device, as we assume that the original spare was
2670 * done in response to the FMA fault.
2671 */
2672 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL &&
2673 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2674 vd->vdev_parent->vdev_child[0] == vd)
2675 vd->vdev_unspare = B_TRUE;
2676 }
2677
2678 boolean_t
2679 vdev_is_dead(vdev_t *vd)
2680 {
2681 /*
2682 * Holes and missing devices are always considered "dead".
2683 * This simplifies the code since we don't have to check for
2684 * these types of devices in the various code paths.
2685 * Instead we rely on the fact that we skip over dead devices
2686 * before issuing I/O to them.
2687 */
2688 return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole ||
2689 vd->vdev_ops == &vdev_missing_ops);
2690 }
2691
2692 boolean_t
2693 vdev_readable(vdev_t *vd)
2694 {
2695 return (!vdev_is_dead(vd) && !vd->vdev_cant_read);
2696 }
2697
2698 boolean_t
2699 vdev_writeable(vdev_t *vd)
2700 {
2701 return (!vdev_is_dead(vd) && !vd->vdev_cant_write);
2702 }
2703
2704 boolean_t
2705 vdev_allocatable(vdev_t *vd)
2706 {
2707 uint64_t state = vd->vdev_state;
2708
2709 /*
2710 * We currently allow allocations from vdevs which may be in the
2711 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2712 * fails to reopen then we'll catch it later when we're holding
2713 * the proper locks. Note that we have to get the vdev state
2714 * in a local variable because although it changes atomically,
2715 * we're asking two separate questions about it.
2716 */
2717 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) &&
2718 !vd->vdev_cant_write && !vd->vdev_ishole &&
2719 vd->vdev_mg->mg_initialized);
2720 }
2721
2722 boolean_t
2723 vdev_accessible(vdev_t *vd, zio_t *zio)
2724 {
2725 ASSERT(zio->io_vd == vd);
2726
2727 if (vdev_is_dead(vd) || vd->vdev_remove_wanted)
2728 return (B_FALSE);
2729
2730 if (zio->io_type == ZIO_TYPE_READ)
2731 return (!vd->vdev_cant_read);
2732
2733 if (zio->io_type == ZIO_TYPE_WRITE)
2734 return (!vd->vdev_cant_write);
2735
2736 return (B_TRUE);
2737 }
2738
2739 /*
2740 * Get statistics for the given vdev.
2741 */
2742 void
2743 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
2744 {
2745 spa_t *spa = vd->vdev_spa;
2746 vdev_t *rvd = spa->spa_root_vdev;
2747 vdev_t *tvd = vd->vdev_top;
2748
2749 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
2750
2751 mutex_enter(&vd->vdev_stat_lock);
2752 bcopy(&vd->vdev_stat, vs, sizeof (*vs));
2753 vs->vs_timestamp = gethrtime() - vs->vs_timestamp;
2754 vs->vs_state = vd->vdev_state;
2755 vs->vs_rsize = vdev_get_min_asize(vd);
2756 if (vd->vdev_ops->vdev_op_leaf)
2757 vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
2758 /*
2759 * Report expandable space on top-level, non-auxillary devices only.
2760 * The expandable space is reported in terms of metaslab sized units
2761 * since that determines how much space the pool can expand.
2762 */
2763 if (vd->vdev_aux == NULL && tvd != NULL) {
2764 vs->vs_esize = P2ALIGN(vd->vdev_max_asize - vd->vdev_asize,
2765 1ULL << tvd->vdev_ms_shift);
2766 }
2767 if (vd->vdev_aux == NULL && vd == vd->vdev_top && !vd->vdev_ishole) {
2768 vs->vs_fragmentation = vd->vdev_mg->mg_fragmentation;
2769 }
2770
2771 /*
2772 * If we're getting stats on the root vdev, aggregate the I/O counts
2773 * over all top-level vdevs (i.e. the direct children of the root).
2774 */
2775 if (vd == rvd) {
2776 for (int c = 0; c < rvd->vdev_children; c++) {
2777 vdev_t *cvd = rvd->vdev_child[c];
2778 vdev_stat_t *cvs = &cvd->vdev_stat;
2779
2780 for (int t = 0; t < ZIO_TYPES; t++) {
2781 vs->vs_ops[t] += cvs->vs_ops[t];
2782 vs->vs_bytes[t] += cvs->vs_bytes[t];
2783 }
2784 cvs->vs_scan_removing = cvd->vdev_removing;
2785 }
2786 }
2787 mutex_exit(&vd->vdev_stat_lock);
2788 }
2789
2790 void
2791 vdev_clear_stats(vdev_t *vd)
2792 {
2793 mutex_enter(&vd->vdev_stat_lock);
2794 vd->vdev_stat.vs_space = 0;
2795 vd->vdev_stat.vs_dspace = 0;
2796 vd->vdev_stat.vs_alloc = 0;
2797 mutex_exit(&vd->vdev_stat_lock);
2798 }
2799
2800 void
2801 vdev_scan_stat_init(vdev_t *vd)
2802 {
2803 vdev_stat_t *vs = &vd->vdev_stat;
2804
2805 for (int c = 0; c < vd->vdev_children; c++)
2806 vdev_scan_stat_init(vd->vdev_child[c]);
2807
2808 mutex_enter(&vd->vdev_stat_lock);
2809 vs->vs_scan_processed = 0;
2810 mutex_exit(&vd->vdev_stat_lock);
2811 }
2812
2813 void
2814 vdev_stat_update(zio_t *zio, uint64_t psize)
2815 {
2816 spa_t *spa = zio->io_spa;
2817 vdev_t *rvd = spa->spa_root_vdev;
2818 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd;
2819 vdev_t *pvd;
2820 uint64_t txg = zio->io_txg;
2821 vdev_stat_t *vs = &vd->vdev_stat;
2822 zio_type_t type = zio->io_type;
2823 int flags = zio->io_flags;
2824
2825 /*
2826 * If this i/o is a gang leader, it didn't do any actual work.
2827 */
2828 if (zio->io_gang_tree)
2829 return;
2830
2831 if (zio->io_error == 0) {
2832 /*
2833 * If this is a root i/o, don't count it -- we've already
2834 * counted the top-level vdevs, and vdev_get_stats() will
2835 * aggregate them when asked. This reduces contention on
2836 * the root vdev_stat_lock and implicitly handles blocks
2837 * that compress away to holes, for which there is no i/o.
2838 * (Holes never create vdev children, so all the counters
2839 * remain zero, which is what we want.)
2840 *
2841 * Note: this only applies to successful i/o (io_error == 0)
2842 * because unlike i/o counts, errors are not additive.
2843 * When reading a ditto block, for example, failure of
2844 * one top-level vdev does not imply a root-level error.
2845 */
2846 if (vd == rvd)
2847 return;
2848
2849 ASSERT(vd == zio->io_vd);
2850
2851 if (flags & ZIO_FLAG_IO_BYPASS)
2852 return;
2853
2854 mutex_enter(&vd->vdev_stat_lock);
2855
2856 if (flags & ZIO_FLAG_IO_REPAIR) {
2857 if (flags & ZIO_FLAG_SCAN_THREAD) {
2858 dsl_scan_phys_t *scn_phys =
2859 &spa->spa_dsl_pool->dp_scan->scn_phys;
2860 uint64_t *processed = &scn_phys->scn_processed;
2861
2862 /* XXX cleanup? */
2863 if (vd->vdev_ops->vdev_op_leaf)
2864 atomic_add_64(processed, psize);
2865 vs->vs_scan_processed += psize;
2866 }
2867
2868 if (flags & ZIO_FLAG_SELF_HEAL)
2869 vs->vs_self_healed += psize;
2870 }
2871
2872 vs->vs_ops[type]++;
2873 vs->vs_bytes[type] += psize;
2874
2875 mutex_exit(&vd->vdev_stat_lock);
2876 return;
2877 }
2878
2879 if (flags & ZIO_FLAG_SPECULATIVE)
2880 return;
2881
2882 /*
2883 * If this is an I/O error that is going to be retried, then ignore the
2884 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
2885 * hard errors, when in reality they can happen for any number of
2886 * innocuous reasons (bus resets, MPxIO link failure, etc).
2887 */
2888 if (zio->io_error == EIO &&
2889 !(zio->io_flags & ZIO_FLAG_IO_RETRY))
2890 return;
2891
2892 /*
2893 * Intent logs writes won't propagate their error to the root
2894 * I/O so don't mark these types of failures as pool-level
2895 * errors.
2896 */
2897 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE))
2898 return;
2899
2900 mutex_enter(&vd->vdev_stat_lock);
2901 if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) {
2902 if (zio->io_error == ECKSUM)
2903 vs->vs_checksum_errors++;
2904 else
2905 vs->vs_read_errors++;
2906 }
2907 if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd))
2908 vs->vs_write_errors++;
2909 mutex_exit(&vd->vdev_stat_lock);
2910
2911 if (type == ZIO_TYPE_WRITE && txg != 0 &&
2912 (!(flags & ZIO_FLAG_IO_REPAIR) ||
2913 (flags & ZIO_FLAG_SCAN_THREAD) ||
2914 spa->spa_claiming)) {
2915 /*
2916 * This is either a normal write (not a repair), or it's
2917 * a repair induced by the scrub thread, or it's a repair
2918 * made by zil_claim() during spa_load() in the first txg.
2919 * In the normal case, we commit the DTL change in the same
2920 * txg as the block was born. In the scrub-induced repair
2921 * case, we know that scrubs run in first-pass syncing context,
2922 * so we commit the DTL change in spa_syncing_txg(spa).
2923 * In the zil_claim() case, we commit in spa_first_txg(spa).
2924 *
2925 * We currently do not make DTL entries for failed spontaneous
2926 * self-healing writes triggered by normal (non-scrubbing)
2927 * reads, because we have no transactional context in which to
2928 * do so -- and it's not clear that it'd be desirable anyway.
2929 */
2930 if (vd->vdev_ops->vdev_op_leaf) {
2931 uint64_t commit_txg = txg;
2932 if (flags & ZIO_FLAG_SCAN_THREAD) {
2933 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2934 ASSERT(spa_sync_pass(spa) == 1);
2935 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1);
2936 commit_txg = spa_syncing_txg(spa);
2937 } else if (spa->spa_claiming) {
2938 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2939 commit_txg = spa_first_txg(spa);
2940 }
2941 ASSERT(commit_txg >= spa_syncing_txg(spa));
2942 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1))
2943 return;
2944 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2945 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1);
2946 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg);
2947 }
2948 if (vd != rvd)
2949 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1);
2950 }
2951 }
2952
2953 /*
2954 * Update the in-core space usage stats for this vdev, its metaslab class,
2955 * and the root vdev.
2956 */
2957 void
2958 vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta,
2959 int64_t space_delta)
2960 {
2961 int64_t dspace_delta = space_delta;
2962 spa_t *spa = vd->vdev_spa;
2963 vdev_t *rvd = spa->spa_root_vdev;
2964 metaslab_group_t *mg = vd->vdev_mg;
2965 metaslab_class_t *mc = mg ? mg->mg_class : NULL;
2966
2967 ASSERT(vd == vd->vdev_top);
2968
2969 /*
2970 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
2971 * factor. We must calculate this here and not at the root vdev
2972 * because the root vdev's psize-to-asize is simply the max of its
2973 * childrens', thus not accurate enough for us.
2974 */
2975 ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0);
2976 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache);
2977 dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) *
2978 vd->vdev_deflate_ratio;
2979
2980 mutex_enter(&vd->vdev_stat_lock);
2981 vd->vdev_stat.vs_alloc += alloc_delta;
2982 vd->vdev_stat.vs_space += space_delta;
2983 vd->vdev_stat.vs_dspace += dspace_delta;
2984 mutex_exit(&vd->vdev_stat_lock);
2985
2986 if (mc == spa_normal_class(spa)) {
2987 mutex_enter(&rvd->vdev_stat_lock);
2988 rvd->vdev_stat.vs_alloc += alloc_delta;
2989 rvd->vdev_stat.vs_space += space_delta;
2990 rvd->vdev_stat.vs_dspace += dspace_delta;
2991 mutex_exit(&rvd->vdev_stat_lock);
2992 }
2993
2994 if (mc != NULL) {
2995 ASSERT(rvd == vd->vdev_parent);
2996 ASSERT(vd->vdev_ms_count != 0);
2997
2998 metaslab_class_space_update(mc,
2999 alloc_delta, defer_delta, space_delta, dspace_delta);
3000 }
3001 }
3002
3003 /*
3004 * Mark a top-level vdev's config as dirty, placing it on the dirty list
3005 * so that it will be written out next time the vdev configuration is synced.
3006 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
3007 */
3008 void
3009 vdev_config_dirty(vdev_t *vd)
3010 {
3011 spa_t *spa = vd->vdev_spa;
3012 vdev_t *rvd = spa->spa_root_vdev;
3013 int c;
3014
3015 ASSERT(spa_writeable(spa));
3016
3017 /*
3018 * If this is an aux vdev (as with l2cache and spare devices), then we
3019 * update the vdev config manually and set the sync flag.
3020 */
3021 if (vd->vdev_aux != NULL) {
3022 spa_aux_vdev_t *sav = vd->vdev_aux;
3023 nvlist_t **aux;
3024 uint_t naux;
3025
3026 for (c = 0; c < sav->sav_count; c++) {
3027 if (sav->sav_vdevs[c] == vd)
3028 break;
3029 }
3030
3031 if (c == sav->sav_count) {
3032 /*
3033 * We're being removed. There's nothing more to do.
3034 */
3035 ASSERT(sav->sav_sync == B_TRUE);
3036 return;
3037 }
3038
3039 sav->sav_sync = B_TRUE;
3040
3041 if (nvlist_lookup_nvlist_array(sav->sav_config,
3042 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) {
3043 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
3044 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0);
3045 }
3046
3047 ASSERT(c < naux);
3048
3049 /*
3050 * Setting the nvlist in the middle if the array is a little
3051 * sketchy, but it will work.
3052 */
3053 nvlist_free(aux[c]);
3054 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0);
3055
3056 return;
3057 }
3058
3059 /*
3060 * The dirty list is protected by the SCL_CONFIG lock. The caller
3061 * must either hold SCL_CONFIG as writer, or must be the sync thread
3062 * (which holds SCL_CONFIG as reader). There's only one sync thread,
3063 * so this is sufficient to ensure mutual exclusion.
3064 */
3065 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
3066 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3067 spa_config_held(spa, SCL_CONFIG, RW_READER)));
3068
3069 if (vd == rvd) {
3070 for (c = 0; c < rvd->vdev_children; c++)
3071 vdev_config_dirty(rvd->vdev_child[c]);
3072 } else {
3073 ASSERT(vd == vd->vdev_top);
3074
3075 if (!list_link_active(&vd->vdev_config_dirty_node) &&
3076 !vd->vdev_ishole)
3077 list_insert_head(&spa->spa_config_dirty_list, vd);
3078 }
3079 }
3080
3081 void
3082 vdev_config_clean(vdev_t *vd)
3083 {
3084 spa_t *spa = vd->vdev_spa;
3085
3086 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
3087 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3088 spa_config_held(spa, SCL_CONFIG, RW_READER)));
3089
3090 ASSERT(list_link_active(&vd->vdev_config_dirty_node));
3091 list_remove(&spa->spa_config_dirty_list, vd);
3092 }
3093
3094 /*
3095 * Mark a top-level vdev's state as dirty, so that the next pass of
3096 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
3097 * the state changes from larger config changes because they require
3098 * much less locking, and are often needed for administrative actions.
3099 */
3100 void
3101 vdev_state_dirty(vdev_t *vd)
3102 {
3103 spa_t *spa = vd->vdev_spa;
3104
3105 ASSERT(spa_writeable(spa));
3106 ASSERT(vd == vd->vdev_top);
3107
3108 /*
3109 * The state list is protected by the SCL_STATE lock. The caller
3110 * must either hold SCL_STATE as writer, or must be the sync thread
3111 * (which holds SCL_STATE as reader). There's only one sync thread,
3112 * so this is sufficient to ensure mutual exclusion.
3113 */
3114 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
3115 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3116 spa_config_held(spa, SCL_STATE, RW_READER)));
3117
3118 if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole)
3119 list_insert_head(&spa->spa_state_dirty_list, vd);
3120 }
3121
3122 void
3123 vdev_state_clean(vdev_t *vd)
3124 {
3125 spa_t *spa = vd->vdev_spa;
3126
3127 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
3128 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3129 spa_config_held(spa, SCL_STATE, RW_READER)));
3130
3131 ASSERT(list_link_active(&vd->vdev_state_dirty_node));
3132 list_remove(&spa->spa_state_dirty_list, vd);
3133 }
3134
3135 /*
3136 * Propagate vdev state up from children to parent.
3137 */
3138 void
3139 vdev_propagate_state(vdev_t *vd)
3140 {
3141 spa_t *spa = vd->vdev_spa;
3142 vdev_t *rvd = spa->spa_root_vdev;
3143 int degraded = 0, faulted = 0;
3144 int corrupted = 0;
3145 vdev_t *child;
3146
3147 if (vd->vdev_children > 0) {
3148 for (int c = 0; c < vd->vdev_children; c++) {
3149 child = vd->vdev_child[c];
3150
3151 /*
3152 * Don't factor holes into the decision.
3153 */
3154 if (child->vdev_ishole)
3155 continue;
3156
3157 if (!vdev_readable(child) ||
3158 (!vdev_writeable(child) && spa_writeable(spa))) {
3159 /*
3160 * Root special: if there is a top-level log
3161 * device, treat the root vdev as if it were
3162 * degraded.
3163 */
3164 if (child->vdev_islog && vd == rvd)
3165 degraded++;
3166 else
3167 faulted++;
3168 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) {
3169 degraded++;
3170 }
3171
3172 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA)
3173 corrupted++;
3174 }
3175
3176 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded);
3177
3178 /*
3179 * Root special: if there is a top-level vdev that cannot be
3180 * opened due to corrupted metadata, then propagate the root
3181 * vdev's aux state as 'corrupt' rather than 'insufficient
3182 * replicas'.
3183 */
3184 if (corrupted && vd == rvd &&
3185 rvd->vdev_state == VDEV_STATE_CANT_OPEN)
3186 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN,
3187 VDEV_AUX_CORRUPT_DATA);
3188 }
3189
3190 if (vd->vdev_parent)
3191 vdev_propagate_state(vd->vdev_parent);
3192 }
3193
3194 /*
3195 * Set a vdev's state. If this is during an open, we don't update the parent
3196 * state, because we're in the process of opening children depth-first.
3197 * Otherwise, we propagate the change to the parent.
3198 *
3199 * If this routine places a device in a faulted state, an appropriate ereport is
3200 * generated.
3201 */
3202 void
3203 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
3204 {
3205 uint64_t save_state;
3206 spa_t *spa = vd->vdev_spa;
3207
3208 if (state == vd->vdev_state) {
3209 vd->vdev_stat.vs_aux = aux;
3210 return;
3211 }
3212
3213 save_state = vd->vdev_state;
3214
3215 vd->vdev_state = state;
3216 vd->vdev_stat.vs_aux = aux;
3217
3218 /*
3219 * If we are setting the vdev state to anything but an open state, then
3220 * always close the underlying device unless the device has requested
3221 * a delayed close (i.e. we're about to remove or fault the device).
3222 * Otherwise, we keep accessible but invalid devices open forever.
3223 * We don't call vdev_close() itself, because that implies some extra
3224 * checks (offline, etc) that we don't want here. This is limited to
3225 * leaf devices, because otherwise closing the device will affect other
3226 * children.
3227 */
3228 if (!vd->vdev_delayed_close && vdev_is_dead(vd) &&
3229 vd->vdev_ops->vdev_op_leaf)
3230 vd->vdev_ops->vdev_op_close(vd);
3231
3232 /*
3233 * If we have brought this vdev back into service, we need
3234 * to notify fmd so that it can gracefully repair any outstanding
3235 * cases due to a missing device. We do this in all cases, even those
3236 * that probably don't correlate to a repaired fault. This is sure to
3237 * catch all cases, and we let the zfs-retire agent sort it out. If
3238 * this is a transient state it's OK, as the retire agent will
3239 * double-check the state of the vdev before repairing it.
3240 */
3241 if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf &&
3242 vd->vdev_prevstate != state)
3243 zfs_post_state_change(spa, vd);
3244
3245 if (vd->vdev_removed &&
3246 state == VDEV_STATE_CANT_OPEN &&
3247 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) {
3248 /*
3249 * If the previous state is set to VDEV_STATE_REMOVED, then this
3250 * device was previously marked removed and someone attempted to
3251 * reopen it. If this failed due to a nonexistent device, then
3252 * keep the device in the REMOVED state. We also let this be if
3253 * it is one of our special test online cases, which is only
3254 * attempting to online the device and shouldn't generate an FMA
3255 * fault.
3256 */
3257 vd->vdev_state = VDEV_STATE_REMOVED;
3258 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
3259 } else if (state == VDEV_STATE_REMOVED) {
3260 vd->vdev_removed = B_TRUE;
3261 } else if (state == VDEV_STATE_CANT_OPEN) {
3262 /*
3263 * If we fail to open a vdev during an import or recovery, we
3264 * mark it as "not available", which signifies that it was
3265 * never there to begin with. Failure to open such a device
3266 * is not considered an error.
3267 */
3268 if ((spa_load_state(spa) == SPA_LOAD_IMPORT ||
3269 spa_load_state(spa) == SPA_LOAD_RECOVER) &&
3270 vd->vdev_ops->vdev_op_leaf)
3271 vd->vdev_not_present = 1;
3272
3273 /*
3274 * Post the appropriate ereport. If the 'prevstate' field is
3275 * set to something other than VDEV_STATE_UNKNOWN, it indicates
3276 * that this is part of a vdev_reopen(). In this case, we don't
3277 * want to post the ereport if the device was already in the
3278 * CANT_OPEN state beforehand.
3279 *
3280 * If the 'checkremove' flag is set, then this is an attempt to
3281 * online the device in response to an insertion event. If we
3282 * hit this case, then we have detected an insertion event for a
3283 * faulted or offline device that wasn't in the removed state.
3284 * In this scenario, we don't post an ereport because we are
3285 * about to replace the device, or attempt an online with
3286 * vdev_forcefault, which will generate the fault for us.
3287 */
3288 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) &&
3289 !vd->vdev_not_present && !vd->vdev_checkremove &&
3290 vd != spa->spa_root_vdev) {
3291 const char *class;
3292
3293 switch (aux) {
3294 case VDEV_AUX_OPEN_FAILED:
3295 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED;
3296 break;
3297 case VDEV_AUX_CORRUPT_DATA:
3298 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA;
3299 break;
3300 case VDEV_AUX_NO_REPLICAS:
3301 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS;
3302 break;
3303 case VDEV_AUX_BAD_GUID_SUM:
3304 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM;
3305 break;
3306 case VDEV_AUX_TOO_SMALL:
3307 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL;
3308 break;
3309 case VDEV_AUX_BAD_LABEL:
3310 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL;
3311 break;
3312 default:
3313 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN;
3314 }
3315
3316 zfs_ereport_post(class, spa, vd, NULL, save_state, 0);
3317 }
3318
3319 /* Erase any notion of persistent removed state */
3320 vd->vdev_removed = B_FALSE;
3321 } else {
3322 vd->vdev_removed = B_FALSE;
3323 }
3324
3325 if (!isopen && vd->vdev_parent)
3326 vdev_propagate_state(vd->vdev_parent);
3327 }
3328
3329 /*
3330 * Check the vdev configuration to ensure that it's capable of supporting
3331 * a root pool. We do not support partial configuration.
3332 * In addition, only a single top-level vdev is allowed.
3333 */
3334 boolean_t
3335 vdev_is_bootable(vdev_t *vd)
3336 {
3337 if (!vd->vdev_ops->vdev_op_leaf) {
3338 char *vdev_type = vd->vdev_ops->vdev_op_type;
3339
3340 if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 &&
3341 vd->vdev_children > 1) {
3342 return (B_FALSE);
3343 } else if (strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) {
3344 return (B_FALSE);
3345 }
3346 }
3347
3348 for (int c = 0; c < vd->vdev_children; c++) {
3349 if (!vdev_is_bootable(vd->vdev_child[c]))
3350 return (B_FALSE);
3351 }
3352 return (B_TRUE);
3353 }
3354
3355 /*
3356 * Load the state from the original vdev tree (ovd) which
3357 * we've retrieved from the MOS config object. If the original
3358 * vdev was offline or faulted then we transfer that state to the
3359 * device in the current vdev tree (nvd).
3360 */
3361 void
3362 vdev_load_log_state(vdev_t *nvd, vdev_t *ovd)
3363 {
3364 spa_t *spa = nvd->vdev_spa;
3365
3366 ASSERT(nvd->vdev_top->vdev_islog);
3367 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
3368 ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid);
3369
3370 for (int c = 0; c < nvd->vdev_children; c++)
3371 vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]);
3372
3373 if (nvd->vdev_ops->vdev_op_leaf) {
3374 /*
3375 * Restore the persistent vdev state
3376 */
3377 nvd->vdev_offline = ovd->vdev_offline;
3378 nvd->vdev_faulted = ovd->vdev_faulted;
3379 nvd->vdev_degraded = ovd->vdev_degraded;
3380 nvd->vdev_removed = ovd->vdev_removed;
3381 }
3382 }
3383
3384 /*
3385 * Determine if a log device has valid content. If the vdev was
3386 * removed or faulted in the MOS config then we know that
3387 * the content on the log device has already been written to the pool.
3388 */
3389 boolean_t
3390 vdev_log_state_valid(vdev_t *vd)
3391 {
3392 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted &&
3393 !vd->vdev_removed)
3394 return (B_TRUE);
3395
3396 for (int c = 0; c < vd->vdev_children; c++)
3397 if (vdev_log_state_valid(vd->vdev_child[c]))
3398 return (B_TRUE);
3399
3400 return (B_FALSE);
3401 }
3402
3403 /*
3404 * Expand a vdev if possible.
3405 */
3406 void
3407 vdev_expand(vdev_t *vd, uint64_t txg)
3408 {
3409 ASSERT(vd->vdev_top == vd);
3410 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3411
3412 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) {
3413 VERIFY(vdev_metaslab_init(vd, txg) == 0);
3414 vdev_config_dirty(vd);
3415 }
3416 }
3417
3418 /*
3419 * Split a vdev.
3420 */
3421 void
3422 vdev_split(vdev_t *vd)
3423 {
3424 vdev_t *cvd, *pvd = vd->vdev_parent;
3425
3426 vdev_remove_child(pvd, vd);
3427 vdev_compact_children(pvd);
3428
3429 cvd = pvd->vdev_child[0];
3430 if (pvd->vdev_children == 1) {
3431 vdev_remove_parent(cvd);
3432 cvd->vdev_splitting = B_TRUE;
3433 }
3434 vdev_propagate_state(cvd);
3435 }
3436
3437 void
3438 vdev_deadman(vdev_t *vd)
3439 {
3440 for (int c = 0; c < vd->vdev_children; c++) {
3441 vdev_t *cvd = vd->vdev_child[c];
3442
3443 vdev_deadman(cvd);
3444 }
3445
3446 if (vd->vdev_ops->vdev_op_leaf) {
3447 vdev_queue_t *vq = &vd->vdev_queue;
3448
3449 mutex_enter(&vq->vq_lock);
3450 if (avl_numnodes(&vq->vq_active_tree) > 0) {
3451 spa_t *spa = vd->vdev_spa;
3452 zio_t *fio;
3453 uint64_t delta;
3454
3455 /*
3456 * Look at the head of all the pending queues,
3457 * if any I/O has been outstanding for longer than
3458 * the spa_deadman_synctime we panic the system.
3459 */
3460 fio = avl_first(&vq->vq_active_tree);
3461 delta = gethrtime() - fio->io_timestamp;
3462 if (delta > spa_deadman_synctime(spa)) {
3463 zfs_dbgmsg("SLOW IO: zio timestamp %lluns, "
3464 "delta %lluns, last io %lluns",
3465 fio->io_timestamp, delta,
3466 vq->vq_io_complete_ts);
3467 fm_panic("I/O to pool '%s' appears to be "
3468 "hung.", spa_name(spa));
3469 }
3470 }
3471 mutex_exit(&vq->vq_lock);
3472 }
3473 }