1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22 /*
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2013 by Delphix. All rights reserved.
26 */
27
28 #include <sys/zfs_context.h>
29 #include <sys/fm/fs/zfs.h>
30 #include <sys/spa.h>
31 #include <sys/spa_impl.h>
32 #include <sys/dmu.h>
33 #include <sys/dmu_tx.h>
34 #include <sys/vdev_impl.h>
35 #include <sys/uberblock_impl.h>
36 #include <sys/metaslab.h>
37 #include <sys/metaslab_impl.h>
38 #include <sys/space_map.h>
39 #include <sys/zio.h>
40 #include <sys/zap.h>
41 #include <sys/fs/zfs.h>
42 #include <sys/arc.h>
43 #include <sys/zil.h>
44 #include <sys/dsl_scan.h>
45
46 /*
47 * Virtual device management.
48 */
49
50 static vdev_ops_t *vdev_ops_table[] = {
51 &vdev_root_ops,
52 &vdev_raidz_ops,
53 &vdev_mirror_ops,
54 &vdev_replacing_ops,
55 &vdev_spare_ops,
56 &vdev_disk_ops,
57 &vdev_file_ops,
58 &vdev_missing_ops,
59 &vdev_hole_ops,
60 NULL
61 };
62
63 /* maximum scrub/resilver I/O queue per leaf vdev */
64 int zfs_scrub_limit = 10;
65
66 /*
67 * Given a vdev type, return the appropriate ops vector.
68 */
69 static vdev_ops_t *
70 vdev_getops(const char *type)
71 {
72 vdev_ops_t *ops, **opspp;
73
74 for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++)
75 if (strcmp(ops->vdev_op_type, type) == 0)
76 break;
77
78 return (ops);
79 }
80
81 /*
82 * Default asize function: return the MAX of psize with the asize of
83 * all children. This is what's used by anything other than RAID-Z.
84 */
85 uint64_t
86 vdev_default_asize(vdev_t *vd, uint64_t psize)
87 {
88 uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift);
89 uint64_t csize;
90
91 for (int c = 0; c < vd->vdev_children; c++) {
92 csize = vdev_psize_to_asize(vd->vdev_child[c], psize);
93 asize = MAX(asize, csize);
94 }
95
96 return (asize);
97 }
98
99 /*
100 * Get the minimum allocatable size. We define the allocatable size as
101 * the vdev's asize rounded to the nearest metaslab. This allows us to
102 * replace or attach devices which don't have the same physical size but
103 * can still satisfy the same number of allocations.
104 */
105 uint64_t
106 vdev_get_min_asize(vdev_t *vd)
107 {
108 vdev_t *pvd = vd->vdev_parent;
109
110 /*
111 * If our parent is NULL (inactive spare or cache) or is the root,
112 * just return our own asize.
113 */
114 if (pvd == NULL)
115 return (vd->vdev_asize);
116
117 /*
118 * The top-level vdev just returns the allocatable size rounded
119 * to the nearest metaslab.
120 */
121 if (vd == vd->vdev_top)
122 return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift));
123
124 /*
125 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
126 * so each child must provide at least 1/Nth of its asize.
127 */
128 if (pvd->vdev_ops == &vdev_raidz_ops)
129 return (pvd->vdev_min_asize / pvd->vdev_children);
130
131 return (pvd->vdev_min_asize);
132 }
133
134 void
135 vdev_set_min_asize(vdev_t *vd)
136 {
137 vd->vdev_min_asize = vdev_get_min_asize(vd);
138
139 for (int c = 0; c < vd->vdev_children; c++)
140 vdev_set_min_asize(vd->vdev_child[c]);
141 }
142
143 vdev_t *
144 vdev_lookup_top(spa_t *spa, uint64_t vdev)
145 {
146 vdev_t *rvd = spa->spa_root_vdev;
147
148 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
149
150 if (vdev < rvd->vdev_children) {
151 ASSERT(rvd->vdev_child[vdev] != NULL);
152 return (rvd->vdev_child[vdev]);
153 }
154
155 return (NULL);
156 }
157
158 vdev_t *
159 vdev_lookup_by_guid(vdev_t *vd, uint64_t guid)
160 {
161 vdev_t *mvd;
162
163 if (vd->vdev_guid == guid)
164 return (vd);
165
166 for (int c = 0; c < vd->vdev_children; c++)
167 if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) !=
168 NULL)
169 return (mvd);
170
171 return (NULL);
172 }
173
174 void
175 vdev_add_child(vdev_t *pvd, vdev_t *cvd)
176 {
177 size_t oldsize, newsize;
178 uint64_t id = cvd->vdev_id;
179 vdev_t **newchild;
180
181 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
182 ASSERT(cvd->vdev_parent == NULL);
183
184 cvd->vdev_parent = pvd;
185
186 if (pvd == NULL)
187 return;
188
189 ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL);
190
191 oldsize = pvd->vdev_children * sizeof (vdev_t *);
192 pvd->vdev_children = MAX(pvd->vdev_children, id + 1);
193 newsize = pvd->vdev_children * sizeof (vdev_t *);
194
195 newchild = kmem_zalloc(newsize, KM_SLEEP);
196 if (pvd->vdev_child != NULL) {
197 bcopy(pvd->vdev_child, newchild, oldsize);
198 kmem_free(pvd->vdev_child, oldsize);
199 }
200
201 pvd->vdev_child = newchild;
202 pvd->vdev_child[id] = cvd;
203
204 cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd);
205 ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL);
206
207 /*
208 * Walk up all ancestors to update guid sum.
209 */
210 for (; pvd != NULL; pvd = pvd->vdev_parent)
211 pvd->vdev_guid_sum += cvd->vdev_guid_sum;
212 }
213
214 void
215 vdev_remove_child(vdev_t *pvd, vdev_t *cvd)
216 {
217 int c;
218 uint_t id = cvd->vdev_id;
219
220 ASSERT(cvd->vdev_parent == pvd);
221
222 if (pvd == NULL)
223 return;
224
225 ASSERT(id < pvd->vdev_children);
226 ASSERT(pvd->vdev_child[id] == cvd);
227
228 pvd->vdev_child[id] = NULL;
229 cvd->vdev_parent = NULL;
230
231 for (c = 0; c < pvd->vdev_children; c++)
232 if (pvd->vdev_child[c])
233 break;
234
235 if (c == pvd->vdev_children) {
236 kmem_free(pvd->vdev_child, c * sizeof (vdev_t *));
237 pvd->vdev_child = NULL;
238 pvd->vdev_children = 0;
239 }
240
241 /*
242 * Walk up all ancestors to update guid sum.
243 */
244 for (; pvd != NULL; pvd = pvd->vdev_parent)
245 pvd->vdev_guid_sum -= cvd->vdev_guid_sum;
246 }
247
248 /*
249 * Remove any holes in the child array.
250 */
251 void
252 vdev_compact_children(vdev_t *pvd)
253 {
254 vdev_t **newchild, *cvd;
255 int oldc = pvd->vdev_children;
256 int newc;
257
258 ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
259
260 for (int c = newc = 0; c < oldc; c++)
261 if (pvd->vdev_child[c])
262 newc++;
263
264 newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP);
265
266 for (int c = newc = 0; c < oldc; c++) {
267 if ((cvd = pvd->vdev_child[c]) != NULL) {
268 newchild[newc] = cvd;
269 cvd->vdev_id = newc++;
270 }
271 }
272
273 kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *));
274 pvd->vdev_child = newchild;
275 pvd->vdev_children = newc;
276 }
277
278 /*
279 * Allocate and minimally initialize a vdev_t.
280 */
281 vdev_t *
282 vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops)
283 {
284 vdev_t *vd;
285
286 vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP);
287
288 if (spa->spa_root_vdev == NULL) {
289 ASSERT(ops == &vdev_root_ops);
290 spa->spa_root_vdev = vd;
291 spa->spa_load_guid = spa_generate_guid(NULL);
292 }
293
294 if (guid == 0 && ops != &vdev_hole_ops) {
295 if (spa->spa_root_vdev == vd) {
296 /*
297 * The root vdev's guid will also be the pool guid,
298 * which must be unique among all pools.
299 */
300 guid = spa_generate_guid(NULL);
301 } else {
302 /*
303 * Any other vdev's guid must be unique within the pool.
304 */
305 guid = spa_generate_guid(spa);
306 }
307 ASSERT(!spa_guid_exists(spa_guid(spa), guid));
308 }
309
310 vd->vdev_spa = spa;
311 vd->vdev_id = id;
312 vd->vdev_guid = guid;
313 vd->vdev_guid_sum = guid;
314 vd->vdev_ops = ops;
315 vd->vdev_state = VDEV_STATE_CLOSED;
316 vd->vdev_ishole = (ops == &vdev_hole_ops);
317
318 mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL);
319 mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL);
320 mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL);
321 for (int t = 0; t < DTL_TYPES; t++) {
322 space_map_create(&vd->vdev_dtl[t], 0, -1ULL, 0,
323 &vd->vdev_dtl_lock);
324 }
325 txg_list_create(&vd->vdev_ms_list,
326 offsetof(struct metaslab, ms_txg_node));
327 txg_list_create(&vd->vdev_dtl_list,
328 offsetof(struct vdev, vdev_dtl_node));
329 vd->vdev_stat.vs_timestamp = gethrtime();
330 vdev_queue_init(vd);
331 vdev_cache_init(vd);
332
333 return (vd);
334 }
335
336 /*
337 * Allocate a new vdev. The 'alloctype' is used to control whether we are
338 * creating a new vdev or loading an existing one - the behavior is slightly
339 * different for each case.
340 */
341 int
342 vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
343 int alloctype)
344 {
345 vdev_ops_t *ops;
346 char *type;
347 uint64_t guid = 0, islog, nparity;
348 vdev_t *vd;
349
350 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
351
352 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
353 return (SET_ERROR(EINVAL));
354
355 if ((ops = vdev_getops(type)) == NULL)
356 return (SET_ERROR(EINVAL));
357
358 /*
359 * If this is a load, get the vdev guid from the nvlist.
360 * Otherwise, vdev_alloc_common() will generate one for us.
361 */
362 if (alloctype == VDEV_ALLOC_LOAD) {
363 uint64_t label_id;
364
365 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) ||
366 label_id != id)
367 return (SET_ERROR(EINVAL));
368
369 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
370 return (SET_ERROR(EINVAL));
371 } else if (alloctype == VDEV_ALLOC_SPARE) {
372 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
373 return (SET_ERROR(EINVAL));
374 } else if (alloctype == VDEV_ALLOC_L2CACHE) {
375 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
376 return (SET_ERROR(EINVAL));
377 } else if (alloctype == VDEV_ALLOC_ROOTPOOL) {
378 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
379 return (SET_ERROR(EINVAL));
380 }
381
382 /*
383 * The first allocated vdev must be of type 'root'.
384 */
385 if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL)
386 return (SET_ERROR(EINVAL));
387
388 /*
389 * Determine whether we're a log vdev.
390 */
391 islog = 0;
392 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog);
393 if (islog && spa_version(spa) < SPA_VERSION_SLOGS)
394 return (SET_ERROR(ENOTSUP));
395
396 if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES)
397 return (SET_ERROR(ENOTSUP));
398
399 /*
400 * Set the nparity property for RAID-Z vdevs.
401 */
402 nparity = -1ULL;
403 if (ops == &vdev_raidz_ops) {
404 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
405 &nparity) == 0) {
406 if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY)
407 return (SET_ERROR(EINVAL));
408 /*
409 * Previous versions could only support 1 or 2 parity
410 * device.
411 */
412 if (nparity > 1 &&
413 spa_version(spa) < SPA_VERSION_RAIDZ2)
414 return (SET_ERROR(ENOTSUP));
415 if (nparity > 2 &&
416 spa_version(spa) < SPA_VERSION_RAIDZ3)
417 return (SET_ERROR(ENOTSUP));
418 } else {
419 /*
420 * We require the parity to be specified for SPAs that
421 * support multiple parity levels.
422 */
423 if (spa_version(spa) >= SPA_VERSION_RAIDZ2)
424 return (SET_ERROR(EINVAL));
425 /*
426 * Otherwise, we default to 1 parity device for RAID-Z.
427 */
428 nparity = 1;
429 }
430 } else {
431 nparity = 0;
432 }
433 ASSERT(nparity != -1ULL);
434
435 vd = vdev_alloc_common(spa, id, guid, ops);
436
437 vd->vdev_islog = islog;
438 vd->vdev_nparity = nparity;
439
440 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0)
441 vd->vdev_path = spa_strdup(vd->vdev_path);
442 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0)
443 vd->vdev_devid = spa_strdup(vd->vdev_devid);
444 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH,
445 &vd->vdev_physpath) == 0)
446 vd->vdev_physpath = spa_strdup(vd->vdev_physpath);
447 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0)
448 vd->vdev_fru = spa_strdup(vd->vdev_fru);
449
450 /*
451 * Set the whole_disk property. If it's not specified, leave the value
452 * as -1.
453 */
454 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
455 &vd->vdev_wholedisk) != 0)
456 vd->vdev_wholedisk = -1ULL;
457
458 /*
459 * Look for the 'not present' flag. This will only be set if the device
460 * was not present at the time of import.
461 */
462 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
463 &vd->vdev_not_present);
464
465 /*
466 * Get the alignment requirement.
467 */
468 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift);
469
470 /*
471 * Retrieve the vdev creation time.
472 */
473 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG,
474 &vd->vdev_crtxg);
475
476 /*
477 * If we're a top-level vdev, try to load the allocation parameters.
478 */
479 if (parent && !parent->vdev_parent &&
480 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
481 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
482 &vd->vdev_ms_array);
483 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
484 &vd->vdev_ms_shift);
485 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE,
486 &vd->vdev_asize);
487 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING,
488 &vd->vdev_removing);
489 }
490
491 if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) {
492 ASSERT(alloctype == VDEV_ALLOC_LOAD ||
493 alloctype == VDEV_ALLOC_ADD ||
494 alloctype == VDEV_ALLOC_SPLIT ||
495 alloctype == VDEV_ALLOC_ROOTPOOL);
496 vd->vdev_mg = metaslab_group_create(islog ?
497 spa_log_class(spa) : spa_normal_class(spa), vd);
498 }
499
500 /*
501 * If we're a leaf vdev, try to load the DTL object and other state.
502 */
503 if (vd->vdev_ops->vdev_op_leaf &&
504 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE ||
505 alloctype == VDEV_ALLOC_ROOTPOOL)) {
506 if (alloctype == VDEV_ALLOC_LOAD) {
507 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL,
508 &vd->vdev_dtl_smo.smo_object);
509 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE,
510 &vd->vdev_unspare);
511 }
512
513 if (alloctype == VDEV_ALLOC_ROOTPOOL) {
514 uint64_t spare = 0;
515
516 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
517 &spare) == 0 && spare)
518 spa_spare_add(vd);
519 }
520
521 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE,
522 &vd->vdev_offline);
523
524 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVERING,
525 &vd->vdev_resilvering);
526
527 /*
528 * When importing a pool, we want to ignore the persistent fault
529 * state, as the diagnosis made on another system may not be
530 * valid in the current context. Local vdevs will
531 * remain in the faulted state.
532 */
533 if (spa_load_state(spa) == SPA_LOAD_OPEN) {
534 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED,
535 &vd->vdev_faulted);
536 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED,
537 &vd->vdev_degraded);
538 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED,
539 &vd->vdev_removed);
540
541 if (vd->vdev_faulted || vd->vdev_degraded) {
542 char *aux;
543
544 vd->vdev_label_aux =
545 VDEV_AUX_ERR_EXCEEDED;
546 if (nvlist_lookup_string(nv,
547 ZPOOL_CONFIG_AUX_STATE, &aux) == 0 &&
548 strcmp(aux, "external") == 0)
549 vd->vdev_label_aux = VDEV_AUX_EXTERNAL;
550 }
551 }
552 }
553
554 /*
555 * Add ourselves to the parent's list of children.
556 */
557 vdev_add_child(parent, vd);
558
559 *vdp = vd;
560
561 return (0);
562 }
563
564 void
565 vdev_free(vdev_t *vd)
566 {
567 spa_t *spa = vd->vdev_spa;
568
569 /*
570 * vdev_free() implies closing the vdev first. This is simpler than
571 * trying to ensure complicated semantics for all callers.
572 */
573 vdev_close(vd);
574
575 ASSERT(!list_link_active(&vd->vdev_config_dirty_node));
576 ASSERT(!list_link_active(&vd->vdev_state_dirty_node));
577
578 /*
579 * Free all children.
580 */
581 for (int c = 0; c < vd->vdev_children; c++)
582 vdev_free(vd->vdev_child[c]);
583
584 ASSERT(vd->vdev_child == NULL);
585 ASSERT(vd->vdev_guid_sum == vd->vdev_guid);
586
587 /*
588 * Discard allocation state.
589 */
590 if (vd->vdev_mg != NULL) {
591 vdev_metaslab_fini(vd);
592 metaslab_group_destroy(vd->vdev_mg);
593 }
594
595 ASSERT0(vd->vdev_stat.vs_space);
596 ASSERT0(vd->vdev_stat.vs_dspace);
597 ASSERT0(vd->vdev_stat.vs_alloc);
598
599 /*
600 * Remove this vdev from its parent's child list.
601 */
602 vdev_remove_child(vd->vdev_parent, vd);
603
604 ASSERT(vd->vdev_parent == NULL);
605
606 /*
607 * Clean up vdev structure.
608 */
609 vdev_queue_fini(vd);
610 vdev_cache_fini(vd);
611
612 if (vd->vdev_path)
613 spa_strfree(vd->vdev_path);
614 if (vd->vdev_devid)
615 spa_strfree(vd->vdev_devid);
616 if (vd->vdev_physpath)
617 spa_strfree(vd->vdev_physpath);
618 if (vd->vdev_fru)
619 spa_strfree(vd->vdev_fru);
620
621 if (vd->vdev_isspare)
622 spa_spare_remove(vd);
623 if (vd->vdev_isl2cache)
624 spa_l2cache_remove(vd);
625
626 txg_list_destroy(&vd->vdev_ms_list);
627 txg_list_destroy(&vd->vdev_dtl_list);
628
629 mutex_enter(&vd->vdev_dtl_lock);
630 for (int t = 0; t < DTL_TYPES; t++) {
631 space_map_unload(&vd->vdev_dtl[t]);
632 space_map_destroy(&vd->vdev_dtl[t]);
633 }
634 mutex_exit(&vd->vdev_dtl_lock);
635
636 mutex_destroy(&vd->vdev_dtl_lock);
637 mutex_destroy(&vd->vdev_stat_lock);
638 mutex_destroy(&vd->vdev_probe_lock);
639
640 if (vd == spa->spa_root_vdev)
641 spa->spa_root_vdev = NULL;
642
643 kmem_free(vd, sizeof (vdev_t));
644 }
645
646 /*
647 * Transfer top-level vdev state from svd to tvd.
648 */
649 static void
650 vdev_top_transfer(vdev_t *svd, vdev_t *tvd)
651 {
652 spa_t *spa = svd->vdev_spa;
653 metaslab_t *msp;
654 vdev_t *vd;
655 int t;
656
657 ASSERT(tvd == tvd->vdev_top);
658
659 tvd->vdev_ms_array = svd->vdev_ms_array;
660 tvd->vdev_ms_shift = svd->vdev_ms_shift;
661 tvd->vdev_ms_count = svd->vdev_ms_count;
662
663 svd->vdev_ms_array = 0;
664 svd->vdev_ms_shift = 0;
665 svd->vdev_ms_count = 0;
666
667 if (tvd->vdev_mg)
668 ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg);
669 tvd->vdev_mg = svd->vdev_mg;
670 tvd->vdev_ms = svd->vdev_ms;
671
672 svd->vdev_mg = NULL;
673 svd->vdev_ms = NULL;
674
675 if (tvd->vdev_mg != NULL)
676 tvd->vdev_mg->mg_vd = tvd;
677
678 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc;
679 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space;
680 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace;
681
682 svd->vdev_stat.vs_alloc = 0;
683 svd->vdev_stat.vs_space = 0;
684 svd->vdev_stat.vs_dspace = 0;
685
686 for (t = 0; t < TXG_SIZE; t++) {
687 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL)
688 (void) txg_list_add(&tvd->vdev_ms_list, msp, t);
689 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL)
690 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t);
691 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t))
692 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t);
693 }
694
695 if (list_link_active(&svd->vdev_config_dirty_node)) {
696 vdev_config_clean(svd);
697 vdev_config_dirty(tvd);
698 }
699
700 if (list_link_active(&svd->vdev_state_dirty_node)) {
701 vdev_state_clean(svd);
702 vdev_state_dirty(tvd);
703 }
704
705 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio;
706 svd->vdev_deflate_ratio = 0;
707
708 tvd->vdev_islog = svd->vdev_islog;
709 svd->vdev_islog = 0;
710 }
711
712 static void
713 vdev_top_update(vdev_t *tvd, vdev_t *vd)
714 {
715 if (vd == NULL)
716 return;
717
718 vd->vdev_top = tvd;
719
720 for (int c = 0; c < vd->vdev_children; c++)
721 vdev_top_update(tvd, vd->vdev_child[c]);
722 }
723
724 /*
725 * Add a mirror/replacing vdev above an existing vdev.
726 */
727 vdev_t *
728 vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops)
729 {
730 spa_t *spa = cvd->vdev_spa;
731 vdev_t *pvd = cvd->vdev_parent;
732 vdev_t *mvd;
733
734 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
735
736 mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops);
737
738 mvd->vdev_asize = cvd->vdev_asize;
739 mvd->vdev_min_asize = cvd->vdev_min_asize;
740 mvd->vdev_max_asize = cvd->vdev_max_asize;
741 mvd->vdev_ashift = cvd->vdev_ashift;
742 mvd->vdev_state = cvd->vdev_state;
743 mvd->vdev_crtxg = cvd->vdev_crtxg;
744
745 vdev_remove_child(pvd, cvd);
746 vdev_add_child(pvd, mvd);
747 cvd->vdev_id = mvd->vdev_children;
748 vdev_add_child(mvd, cvd);
749 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
750
751 if (mvd == mvd->vdev_top)
752 vdev_top_transfer(cvd, mvd);
753
754 return (mvd);
755 }
756
757 /*
758 * Remove a 1-way mirror/replacing vdev from the tree.
759 */
760 void
761 vdev_remove_parent(vdev_t *cvd)
762 {
763 vdev_t *mvd = cvd->vdev_parent;
764 vdev_t *pvd = mvd->vdev_parent;
765
766 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
767
768 ASSERT(mvd->vdev_children == 1);
769 ASSERT(mvd->vdev_ops == &vdev_mirror_ops ||
770 mvd->vdev_ops == &vdev_replacing_ops ||
771 mvd->vdev_ops == &vdev_spare_ops);
772 cvd->vdev_ashift = mvd->vdev_ashift;
773
774 vdev_remove_child(mvd, cvd);
775 vdev_remove_child(pvd, mvd);
776
777 /*
778 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
779 * Otherwise, we could have detached an offline device, and when we
780 * go to import the pool we'll think we have two top-level vdevs,
781 * instead of a different version of the same top-level vdev.
782 */
783 if (mvd->vdev_top == mvd) {
784 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid;
785 cvd->vdev_orig_guid = cvd->vdev_guid;
786 cvd->vdev_guid += guid_delta;
787 cvd->vdev_guid_sum += guid_delta;
788 }
789 cvd->vdev_id = mvd->vdev_id;
790 vdev_add_child(pvd, cvd);
791 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
792
793 if (cvd == cvd->vdev_top)
794 vdev_top_transfer(mvd, cvd);
795
796 ASSERT(mvd->vdev_children == 0);
797 vdev_free(mvd);
798 }
799
800 int
801 vdev_metaslab_init(vdev_t *vd, uint64_t txg)
802 {
803 spa_t *spa = vd->vdev_spa;
804 objset_t *mos = spa->spa_meta_objset;
805 uint64_t m;
806 uint64_t oldc = vd->vdev_ms_count;
807 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift;
808 metaslab_t **mspp;
809 int error;
810
811 ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER));
812
813 /*
814 * This vdev is not being allocated from yet or is a hole.
815 */
816 if (vd->vdev_ms_shift == 0)
817 return (0);
818
819 ASSERT(!vd->vdev_ishole);
820
821 /*
822 * Compute the raidz-deflation ratio. Note, we hard-code
823 * in 128k (1 << 17) because it is the current "typical" blocksize.
824 * Even if SPA_MAXBLOCKSIZE changes, this algorithm must never change,
825 * or we will inconsistently account for existing bp's.
826 */
827 vd->vdev_deflate_ratio = (1 << 17) /
828 (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT);
829
830 ASSERT(oldc <= newc);
831
832 mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP);
833
834 if (oldc != 0) {
835 bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp));
836 kmem_free(vd->vdev_ms, oldc * sizeof (*mspp));
837 }
838
839 vd->vdev_ms = mspp;
840 vd->vdev_ms_count = newc;
841
842 for (m = oldc; m < newc; m++) {
843 space_map_obj_t smo = { 0, 0, 0 };
844 if (txg == 0) {
845 uint64_t object = 0;
846 error = dmu_read(mos, vd->vdev_ms_array,
847 m * sizeof (uint64_t), sizeof (uint64_t), &object,
848 DMU_READ_PREFETCH);
849 if (error)
850 return (error);
851 if (object != 0) {
852 dmu_buf_t *db;
853 error = dmu_bonus_hold(mos, object, FTAG, &db);
854 if (error)
855 return (error);
856 ASSERT3U(db->db_size, >=, sizeof (smo));
857 bcopy(db->db_data, &smo, sizeof (smo));
858 ASSERT3U(smo.smo_object, ==, object);
859 dmu_buf_rele(db, FTAG);
860 }
861 }
862 vd->vdev_ms[m] = metaslab_init(vd->vdev_mg, &smo,
863 m << vd->vdev_ms_shift, 1ULL << vd->vdev_ms_shift, txg);
864 }
865
866 if (txg == 0)
867 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER);
868
869 /*
870 * If the vdev is being removed we don't activate
871 * the metaslabs since we want to ensure that no new
872 * allocations are performed on this device.
873 */
874 if (oldc == 0 && !vd->vdev_removing)
875 metaslab_group_activate(vd->vdev_mg);
876
877 if (txg == 0)
878 spa_config_exit(spa, SCL_ALLOC, FTAG);
879
880 return (0);
881 }
882
883 void
884 vdev_metaslab_fini(vdev_t *vd)
885 {
886 uint64_t m;
887 uint64_t count = vd->vdev_ms_count;
888
889 if (vd->vdev_ms != NULL) {
890 metaslab_group_passivate(vd->vdev_mg);
891 for (m = 0; m < count; m++)
892 if (vd->vdev_ms[m] != NULL)
893 metaslab_fini(vd->vdev_ms[m]);
894 kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *));
895 vd->vdev_ms = NULL;
896 }
897 }
898
899 typedef struct vdev_probe_stats {
900 boolean_t vps_readable;
901 boolean_t vps_writeable;
902 int vps_flags;
903 } vdev_probe_stats_t;
904
905 static void
906 vdev_probe_done(zio_t *zio)
907 {
908 spa_t *spa = zio->io_spa;
909 vdev_t *vd = zio->io_vd;
910 vdev_probe_stats_t *vps = zio->io_private;
911
912 ASSERT(vd->vdev_probe_zio != NULL);
913
914 if (zio->io_type == ZIO_TYPE_READ) {
915 if (zio->io_error == 0)
916 vps->vps_readable = 1;
917 if (zio->io_error == 0 && spa_writeable(spa)) {
918 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd,
919 zio->io_offset, zio->io_size, zio->io_data,
920 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
921 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE));
922 } else {
923 zio_buf_free(zio->io_data, zio->io_size);
924 }
925 } else if (zio->io_type == ZIO_TYPE_WRITE) {
926 if (zio->io_error == 0)
927 vps->vps_writeable = 1;
928 zio_buf_free(zio->io_data, zio->io_size);
929 } else if (zio->io_type == ZIO_TYPE_NULL) {
930 zio_t *pio;
931
932 vd->vdev_cant_read |= !vps->vps_readable;
933 vd->vdev_cant_write |= !vps->vps_writeable;
934
935 if (vdev_readable(vd) &&
936 (vdev_writeable(vd) || !spa_writeable(spa))) {
937 zio->io_error = 0;
938 } else {
939 ASSERT(zio->io_error != 0);
940 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE,
941 spa, vd, NULL, 0, 0);
942 zio->io_error = SET_ERROR(ENXIO);
943 }
944
945 mutex_enter(&vd->vdev_probe_lock);
946 ASSERT(vd->vdev_probe_zio == zio);
947 vd->vdev_probe_zio = NULL;
948 mutex_exit(&vd->vdev_probe_lock);
949
950 while ((pio = zio_walk_parents(zio)) != NULL)
951 if (!vdev_accessible(vd, pio))
952 pio->io_error = SET_ERROR(ENXIO);
953
954 kmem_free(vps, sizeof (*vps));
955 }
956 }
957
958 /*
959 * Determine whether this device is accessible.
960 *
961 * Read and write to several known locations: the pad regions of each
962 * vdev label but the first, which we leave alone in case it contains
963 * a VTOC.
964 */
965 zio_t *
966 vdev_probe(vdev_t *vd, zio_t *zio)
967 {
968 spa_t *spa = vd->vdev_spa;
969 vdev_probe_stats_t *vps = NULL;
970 zio_t *pio;
971
972 ASSERT(vd->vdev_ops->vdev_op_leaf);
973
974 /*
975 * Don't probe the probe.
976 */
977 if (zio && (zio->io_flags & ZIO_FLAG_PROBE))
978 return (NULL);
979
980 /*
981 * To prevent 'probe storms' when a device fails, we create
982 * just one probe i/o at a time. All zios that want to probe
983 * this vdev will become parents of the probe io.
984 */
985 mutex_enter(&vd->vdev_probe_lock);
986
987 if ((pio = vd->vdev_probe_zio) == NULL) {
988 vps = kmem_zalloc(sizeof (*vps), KM_SLEEP);
989
990 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE |
991 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE |
992 ZIO_FLAG_TRYHARD;
993
994 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) {
995 /*
996 * vdev_cant_read and vdev_cant_write can only
997 * transition from TRUE to FALSE when we have the
998 * SCL_ZIO lock as writer; otherwise they can only
999 * transition from FALSE to TRUE. This ensures that
1000 * any zio looking at these values can assume that
1001 * failures persist for the life of the I/O. That's
1002 * important because when a device has intermittent
1003 * connectivity problems, we want to ensure that
1004 * they're ascribed to the device (ENXIO) and not
1005 * the zio (EIO).
1006 *
1007 * Since we hold SCL_ZIO as writer here, clear both
1008 * values so the probe can reevaluate from first
1009 * principles.
1010 */
1011 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER;
1012 vd->vdev_cant_read = B_FALSE;
1013 vd->vdev_cant_write = B_FALSE;
1014 }
1015
1016 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd,
1017 vdev_probe_done, vps,
1018 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE);
1019
1020 /*
1021 * We can't change the vdev state in this context, so we
1022 * kick off an async task to do it on our behalf.
1023 */
1024 if (zio != NULL) {
1025 vd->vdev_probe_wanted = B_TRUE;
1026 spa_async_request(spa, SPA_ASYNC_PROBE);
1027 }
1028 }
1029
1030 if (zio != NULL)
1031 zio_add_child(zio, pio);
1032
1033 mutex_exit(&vd->vdev_probe_lock);
1034
1035 if (vps == NULL) {
1036 ASSERT(zio != NULL);
1037 return (NULL);
1038 }
1039
1040 for (int l = 1; l < VDEV_LABELS; l++) {
1041 zio_nowait(zio_read_phys(pio, vd,
1042 vdev_label_offset(vd->vdev_psize, l,
1043 offsetof(vdev_label_t, vl_pad2)),
1044 VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE),
1045 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1046 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE));
1047 }
1048
1049 if (zio == NULL)
1050 return (pio);
1051
1052 zio_nowait(pio);
1053 return (NULL);
1054 }
1055
1056 static void
1057 vdev_open_child(void *arg)
1058 {
1059 vdev_t *vd = arg;
1060
1061 vd->vdev_open_thread = curthread;
1062 vd->vdev_open_error = vdev_open(vd);
1063 vd->vdev_open_thread = NULL;
1064 }
1065
1066 boolean_t
1067 vdev_uses_zvols(vdev_t *vd)
1068 {
1069 if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR,
1070 strlen(ZVOL_DIR)) == 0)
1071 return (B_TRUE);
1072 for (int c = 0; c < vd->vdev_children; c++)
1073 if (vdev_uses_zvols(vd->vdev_child[c]))
1074 return (B_TRUE);
1075 return (B_FALSE);
1076 }
1077
1078 void
1079 vdev_open_children(vdev_t *vd)
1080 {
1081 taskq_t *tq;
1082 int children = vd->vdev_children;
1083
1084 /*
1085 * in order to handle pools on top of zvols, do the opens
1086 * in a single thread so that the same thread holds the
1087 * spa_namespace_lock
1088 */
1089 if (vdev_uses_zvols(vd)) {
1090 for (int c = 0; c < children; c++)
1091 vd->vdev_child[c]->vdev_open_error =
1092 vdev_open(vd->vdev_child[c]);
1093 return;
1094 }
1095 tq = taskq_create("vdev_open", children, minclsyspri,
1096 children, children, TASKQ_PREPOPULATE);
1097
1098 for (int c = 0; c < children; c++)
1099 VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c],
1100 TQ_SLEEP) != NULL);
1101
1102 taskq_destroy(tq);
1103 }
1104
1105 /*
1106 * Prepare a virtual device for access.
1107 */
1108 int
1109 vdev_open(vdev_t *vd)
1110 {
1111 spa_t *spa = vd->vdev_spa;
1112 int error;
1113 uint64_t osize = 0;
1114 uint64_t max_osize = 0;
1115 uint64_t asize, max_asize, psize;
1116 uint64_t ashift = 0;
1117
1118 ASSERT(vd->vdev_open_thread == curthread ||
1119 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1120 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED ||
1121 vd->vdev_state == VDEV_STATE_CANT_OPEN ||
1122 vd->vdev_state == VDEV_STATE_OFFLINE);
1123
1124 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1125 vd->vdev_cant_read = B_FALSE;
1126 vd->vdev_cant_write = B_FALSE;
1127 vd->vdev_min_asize = vdev_get_min_asize(vd);
1128
1129 /*
1130 * If this vdev is not removed, check its fault status. If it's
1131 * faulted, bail out of the open.
1132 */
1133 if (!vd->vdev_removed && vd->vdev_faulted) {
1134 ASSERT(vd->vdev_children == 0);
1135 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1136 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1137 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1138 vd->vdev_label_aux);
1139 return (SET_ERROR(ENXIO));
1140 } else if (vd->vdev_offline) {
1141 ASSERT(vd->vdev_children == 0);
1142 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE);
1143 return (SET_ERROR(ENXIO));
1144 }
1145
1146 error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize, &ashift);
1147
1148 /*
1149 * Reset the vdev_reopening flag so that we actually close
1150 * the vdev on error.
1151 */
1152 vd->vdev_reopening = B_FALSE;
1153 if (zio_injection_enabled && error == 0)
1154 error = zio_handle_device_injection(vd, NULL, ENXIO);
1155
1156 if (error) {
1157 if (vd->vdev_removed &&
1158 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED)
1159 vd->vdev_removed = B_FALSE;
1160
1161 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1162 vd->vdev_stat.vs_aux);
1163 return (error);
1164 }
1165
1166 vd->vdev_removed = B_FALSE;
1167
1168 /*
1169 * Recheck the faulted flag now that we have confirmed that
1170 * the vdev is accessible. If we're faulted, bail.
1171 */
1172 if (vd->vdev_faulted) {
1173 ASSERT(vd->vdev_children == 0);
1174 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1175 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1176 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1177 vd->vdev_label_aux);
1178 return (SET_ERROR(ENXIO));
1179 }
1180
1181 if (vd->vdev_degraded) {
1182 ASSERT(vd->vdev_children == 0);
1183 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1184 VDEV_AUX_ERR_EXCEEDED);
1185 } else {
1186 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0);
1187 }
1188
1189 /*
1190 * For hole or missing vdevs we just return success.
1191 */
1192 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops)
1193 return (0);
1194
1195 for (int c = 0; c < vd->vdev_children; c++) {
1196 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) {
1197 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1198 VDEV_AUX_NONE);
1199 break;
1200 }
1201 }
1202
1203 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t));
1204 max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t));
1205
1206 if (vd->vdev_children == 0) {
1207 if (osize < SPA_MINDEVSIZE) {
1208 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1209 VDEV_AUX_TOO_SMALL);
1210 return (SET_ERROR(EOVERFLOW));
1211 }
1212 psize = osize;
1213 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE);
1214 max_asize = max_osize - (VDEV_LABEL_START_SIZE +
1215 VDEV_LABEL_END_SIZE);
1216 } else {
1217 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE -
1218 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) {
1219 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1220 VDEV_AUX_TOO_SMALL);
1221 return (SET_ERROR(EOVERFLOW));
1222 }
1223 psize = 0;
1224 asize = osize;
1225 max_asize = max_osize;
1226 }
1227
1228 vd->vdev_psize = psize;
1229
1230 /*
1231 * Make sure the allocatable size hasn't shrunk.
1232 */
1233 if (asize < vd->vdev_min_asize) {
1234 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1235 VDEV_AUX_BAD_LABEL);
1236 return (SET_ERROR(EINVAL));
1237 }
1238
1239 if (vd->vdev_asize == 0) {
1240 /*
1241 * This is the first-ever open, so use the computed values.
1242 * For testing purposes, a higher ashift can be requested.
1243 */
1244 vd->vdev_asize = asize;
1245 vd->vdev_max_asize = max_asize;
1246 vd->vdev_ashift = MAX(ashift, vd->vdev_ashift);
1247 } else {
1248 /*
1249 * Detect if the alignment requirement has increased.
1250 * We don't want to make the pool unavailable, just
1251 * issue a warning instead.
1252 */
1253 if (ashift > vd->vdev_top->vdev_ashift &&
1254 vd->vdev_ops->vdev_op_leaf) {
1255 cmn_err(CE_WARN,
1256 "Disk, '%s', has a block alignment that is "
1257 "larger than the pool's alignment\n",
1258 vd->vdev_path);
1259 }
1260 vd->vdev_max_asize = max_asize;
1261 }
1262
1263 /*
1264 * If all children are healthy and the asize has increased,
1265 * then we've experienced dynamic LUN growth. If automatic
1266 * expansion is enabled then use the additional space.
1267 */
1268 if (vd->vdev_state == VDEV_STATE_HEALTHY && asize > vd->vdev_asize &&
1269 (vd->vdev_expanding || spa->spa_autoexpand))
1270 vd->vdev_asize = asize;
1271
1272 vdev_set_min_asize(vd);
1273
1274 /*
1275 * Ensure we can issue some IO before declaring the
1276 * vdev open for business.
1277 */
1278 if (vd->vdev_ops->vdev_op_leaf &&
1279 (error = zio_wait(vdev_probe(vd, NULL))) != 0) {
1280 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1281 VDEV_AUX_ERR_EXCEEDED);
1282 return (error);
1283 }
1284
1285 /*
1286 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1287 * resilver. But don't do this if we are doing a reopen for a scrub,
1288 * since this would just restart the scrub we are already doing.
1289 */
1290 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen &&
1291 vdev_resilver_needed(vd, NULL, NULL))
1292 spa_async_request(spa, SPA_ASYNC_RESILVER);
1293
1294 return (0);
1295 }
1296
1297 /*
1298 * Called once the vdevs are all opened, this routine validates the label
1299 * contents. This needs to be done before vdev_load() so that we don't
1300 * inadvertently do repair I/Os to the wrong device.
1301 *
1302 * If 'strict' is false ignore the spa guid check. This is necessary because
1303 * if the machine crashed during a re-guid the new guid might have been written
1304 * to all of the vdev labels, but not the cached config. The strict check
1305 * will be performed when the pool is opened again using the mos config.
1306 *
1307 * This function will only return failure if one of the vdevs indicates that it
1308 * has since been destroyed or exported. This is only possible if
1309 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1310 * will be updated but the function will return 0.
1311 */
1312 int
1313 vdev_validate(vdev_t *vd, boolean_t strict)
1314 {
1315 spa_t *spa = vd->vdev_spa;
1316 nvlist_t *label;
1317 uint64_t guid = 0, top_guid;
1318 uint64_t state;
1319
1320 for (int c = 0; c < vd->vdev_children; c++)
1321 if (vdev_validate(vd->vdev_child[c], strict) != 0)
1322 return (SET_ERROR(EBADF));
1323
1324 /*
1325 * If the device has already failed, or was marked offline, don't do
1326 * any further validation. Otherwise, label I/O will fail and we will
1327 * overwrite the previous state.
1328 */
1329 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1330 uint64_t aux_guid = 0;
1331 nvlist_t *nvl;
1332 uint64_t txg = spa_last_synced_txg(spa) != 0 ?
1333 spa_last_synced_txg(spa) : -1ULL;
1334
1335 if ((label = vdev_label_read_config(vd, txg)) == NULL) {
1336 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1337 VDEV_AUX_BAD_LABEL);
1338 return (0);
1339 }
1340
1341 /*
1342 * Determine if this vdev has been split off into another
1343 * pool. If so, then refuse to open it.
1344 */
1345 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID,
1346 &aux_guid) == 0 && aux_guid == spa_guid(spa)) {
1347 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1348 VDEV_AUX_SPLIT_POOL);
1349 nvlist_free(label);
1350 return (0);
1351 }
1352
1353 if (strict && (nvlist_lookup_uint64(label,
1354 ZPOOL_CONFIG_POOL_GUID, &guid) != 0 ||
1355 guid != spa_guid(spa))) {
1356 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1357 VDEV_AUX_CORRUPT_DATA);
1358 nvlist_free(label);
1359 return (0);
1360 }
1361
1362 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl)
1363 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID,
1364 &aux_guid) != 0)
1365 aux_guid = 0;
1366
1367 /*
1368 * If this vdev just became a top-level vdev because its
1369 * sibling was detached, it will have adopted the parent's
1370 * vdev guid -- but the label may or may not be on disk yet.
1371 * Fortunately, either version of the label will have the
1372 * same top guid, so if we're a top-level vdev, we can
1373 * safely compare to that instead.
1374 *
1375 * If we split this vdev off instead, then we also check the
1376 * original pool's guid. We don't want to consider the vdev
1377 * corrupt if it is partway through a split operation.
1378 */
1379 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
1380 &guid) != 0 ||
1381 nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID,
1382 &top_guid) != 0 ||
1383 ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) &&
1384 (vd->vdev_guid != top_guid || vd != vd->vdev_top))) {
1385 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1386 VDEV_AUX_CORRUPT_DATA);
1387 nvlist_free(label);
1388 return (0);
1389 }
1390
1391 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1392 &state) != 0) {
1393 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1394 VDEV_AUX_CORRUPT_DATA);
1395 nvlist_free(label);
1396 return (0);
1397 }
1398
1399 nvlist_free(label);
1400
1401 /*
1402 * If this is a verbatim import, no need to check the
1403 * state of the pool.
1404 */
1405 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) &&
1406 spa_load_state(spa) == SPA_LOAD_OPEN &&
1407 state != POOL_STATE_ACTIVE)
1408 return (SET_ERROR(EBADF));
1409
1410 /*
1411 * If we were able to open and validate a vdev that was
1412 * previously marked permanently unavailable, clear that state
1413 * now.
1414 */
1415 if (vd->vdev_not_present)
1416 vd->vdev_not_present = 0;
1417 }
1418
1419 return (0);
1420 }
1421
1422 /*
1423 * Close a virtual device.
1424 */
1425 void
1426 vdev_close(vdev_t *vd)
1427 {
1428 spa_t *spa = vd->vdev_spa;
1429 vdev_t *pvd = vd->vdev_parent;
1430
1431 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1432
1433 /*
1434 * If our parent is reopening, then we are as well, unless we are
1435 * going offline.
1436 */
1437 if (pvd != NULL && pvd->vdev_reopening)
1438 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline);
1439
1440 vd->vdev_ops->vdev_op_close(vd);
1441
1442 vdev_cache_purge(vd);
1443
1444 /*
1445 * We record the previous state before we close it, so that if we are
1446 * doing a reopen(), we don't generate FMA ereports if we notice that
1447 * it's still faulted.
1448 */
1449 vd->vdev_prevstate = vd->vdev_state;
1450
1451 if (vd->vdev_offline)
1452 vd->vdev_state = VDEV_STATE_OFFLINE;
1453 else
1454 vd->vdev_state = VDEV_STATE_CLOSED;
1455 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1456 }
1457
1458 void
1459 vdev_hold(vdev_t *vd)
1460 {
1461 spa_t *spa = vd->vdev_spa;
1462
1463 ASSERT(spa_is_root(spa));
1464 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1465 return;
1466
1467 for (int c = 0; c < vd->vdev_children; c++)
1468 vdev_hold(vd->vdev_child[c]);
1469
1470 if (vd->vdev_ops->vdev_op_leaf)
1471 vd->vdev_ops->vdev_op_hold(vd);
1472 }
1473
1474 void
1475 vdev_rele(vdev_t *vd)
1476 {
1477 spa_t *spa = vd->vdev_spa;
1478
1479 ASSERT(spa_is_root(spa));
1480 for (int c = 0; c < vd->vdev_children; c++)
1481 vdev_rele(vd->vdev_child[c]);
1482
1483 if (vd->vdev_ops->vdev_op_leaf)
1484 vd->vdev_ops->vdev_op_rele(vd);
1485 }
1486
1487 /*
1488 * Reopen all interior vdevs and any unopened leaves. We don't actually
1489 * reopen leaf vdevs which had previously been opened as they might deadlock
1490 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1491 * If the leaf has never been opened then open it, as usual.
1492 */
1493 void
1494 vdev_reopen(vdev_t *vd)
1495 {
1496 spa_t *spa = vd->vdev_spa;
1497
1498 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1499
1500 /* set the reopening flag unless we're taking the vdev offline */
1501 vd->vdev_reopening = !vd->vdev_offline;
1502 vdev_close(vd);
1503 (void) vdev_open(vd);
1504
1505 /*
1506 * Call vdev_validate() here to make sure we have the same device.
1507 * Otherwise, a device with an invalid label could be successfully
1508 * opened in response to vdev_reopen().
1509 */
1510 if (vd->vdev_aux) {
1511 (void) vdev_validate_aux(vd);
1512 if (vdev_readable(vd) && vdev_writeable(vd) &&
1513 vd->vdev_aux == &spa->spa_l2cache &&
1514 !l2arc_vdev_present(vd)) {
1515 /*
1516 * When reopening we can assume persistent L2ARC is
1517 * supported, since we've already opened the device
1518 * in the past and prepended an L2 uberblock.
1519 */
1520 l2arc_add_vdev(spa, vd, B_TRUE);
1521 }
1522 } else {
1523 (void) vdev_validate(vd, B_TRUE);
1524 }
1525
1526 /*
1527 * Reassess parent vdev's health.
1528 */
1529 vdev_propagate_state(vd);
1530 }
1531
1532 int
1533 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
1534 {
1535 int error;
1536
1537 /*
1538 * Normally, partial opens (e.g. of a mirror) are allowed.
1539 * For a create, however, we want to fail the request if
1540 * there are any components we can't open.
1541 */
1542 error = vdev_open(vd);
1543
1544 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) {
1545 vdev_close(vd);
1546 return (error ? error : ENXIO);
1547 }
1548
1549 /*
1550 * Recursively initialize all labels.
1551 */
1552 if ((error = vdev_label_init(vd, txg, isreplacing ?
1553 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
1554 vdev_close(vd);
1555 return (error);
1556 }
1557
1558 return (0);
1559 }
1560
1561 void
1562 vdev_metaslab_set_size(vdev_t *vd)
1563 {
1564 /*
1565 * Aim for roughly 200 metaslabs per vdev.
1566 */
1567 vd->vdev_ms_shift = highbit(vd->vdev_asize / 200);
1568 vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT);
1569 }
1570
1571 void
1572 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
1573 {
1574 ASSERT(vd == vd->vdev_top);
1575 ASSERT(!vd->vdev_ishole);
1576 ASSERT(ISP2(flags));
1577 ASSERT(spa_writeable(vd->vdev_spa));
1578
1579 if (flags & VDD_METASLAB)
1580 (void) txg_list_add(&vd->vdev_ms_list, arg, txg);
1581
1582 if (flags & VDD_DTL)
1583 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg);
1584
1585 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
1586 }
1587
1588 /*
1589 * DTLs.
1590 *
1591 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1592 * the vdev has less than perfect replication. There are four kinds of DTL:
1593 *
1594 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1595 *
1596 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1597 *
1598 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1599 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1600 * txgs that was scrubbed.
1601 *
1602 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1603 * persistent errors or just some device being offline.
1604 * Unlike the other three, the DTL_OUTAGE map is not generally
1605 * maintained; it's only computed when needed, typically to
1606 * determine whether a device can be detached.
1607 *
1608 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1609 * either has the data or it doesn't.
1610 *
1611 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1612 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1613 * if any child is less than fully replicated, then so is its parent.
1614 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1615 * comprising only those txgs which appear in 'maxfaults' or more children;
1616 * those are the txgs we don't have enough replication to read. For example,
1617 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1618 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1619 * two child DTL_MISSING maps.
1620 *
1621 * It should be clear from the above that to compute the DTLs and outage maps
1622 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1623 * Therefore, that is all we keep on disk. When loading the pool, or after
1624 * a configuration change, we generate all other DTLs from first principles.
1625 */
1626 void
1627 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1628 {
1629 space_map_t *sm = &vd->vdev_dtl[t];
1630
1631 ASSERT(t < DTL_TYPES);
1632 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1633 ASSERT(spa_writeable(vd->vdev_spa));
1634
1635 mutex_enter(sm->sm_lock);
1636 if (!space_map_contains(sm, txg, size))
1637 space_map_add(sm, txg, size);
1638 mutex_exit(sm->sm_lock);
1639 }
1640
1641 boolean_t
1642 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1643 {
1644 space_map_t *sm = &vd->vdev_dtl[t];
1645 boolean_t dirty = B_FALSE;
1646
1647 ASSERT(t < DTL_TYPES);
1648 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1649
1650 mutex_enter(sm->sm_lock);
1651 if (sm->sm_space != 0)
1652 dirty = space_map_contains(sm, txg, size);
1653 mutex_exit(sm->sm_lock);
1654
1655 return (dirty);
1656 }
1657
1658 boolean_t
1659 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t)
1660 {
1661 space_map_t *sm = &vd->vdev_dtl[t];
1662 boolean_t empty;
1663
1664 mutex_enter(sm->sm_lock);
1665 empty = (sm->sm_space == 0);
1666 mutex_exit(sm->sm_lock);
1667
1668 return (empty);
1669 }
1670
1671 /*
1672 * Reassess DTLs after a config change or scrub completion.
1673 */
1674 void
1675 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
1676 {
1677 spa_t *spa = vd->vdev_spa;
1678 avl_tree_t reftree;
1679 int minref;
1680
1681 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1682
1683 for (int c = 0; c < vd->vdev_children; c++)
1684 vdev_dtl_reassess(vd->vdev_child[c], txg,
1685 scrub_txg, scrub_done);
1686
1687 if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux)
1688 return;
1689
1690 if (vd->vdev_ops->vdev_op_leaf) {
1691 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
1692
1693 mutex_enter(&vd->vdev_dtl_lock);
1694 if (scrub_txg != 0 &&
1695 (spa->spa_scrub_started ||
1696 (scn && scn->scn_phys.scn_errors == 0))) {
1697 /*
1698 * We completed a scrub up to scrub_txg. If we
1699 * did it without rebooting, then the scrub dtl
1700 * will be valid, so excise the old region and
1701 * fold in the scrub dtl. Otherwise, leave the
1702 * dtl as-is if there was an error.
1703 *
1704 * There's little trick here: to excise the beginning
1705 * of the DTL_MISSING map, we put it into a reference
1706 * tree and then add a segment with refcnt -1 that
1707 * covers the range [0, scrub_txg). This means
1708 * that each txg in that range has refcnt -1 or 0.
1709 * We then add DTL_SCRUB with a refcnt of 2, so that
1710 * entries in the range [0, scrub_txg) will have a
1711 * positive refcnt -- either 1 or 2. We then convert
1712 * the reference tree into the new DTL_MISSING map.
1713 */
1714 space_map_ref_create(&reftree);
1715 space_map_ref_add_map(&reftree,
1716 &vd->vdev_dtl[DTL_MISSING], 1);
1717 space_map_ref_add_seg(&reftree, 0, scrub_txg, -1);
1718 space_map_ref_add_map(&reftree,
1719 &vd->vdev_dtl[DTL_SCRUB], 2);
1720 space_map_ref_generate_map(&reftree,
1721 &vd->vdev_dtl[DTL_MISSING], 1);
1722 space_map_ref_destroy(&reftree);
1723 }
1724 space_map_vacate(&vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
1725 space_map_walk(&vd->vdev_dtl[DTL_MISSING],
1726 space_map_add, &vd->vdev_dtl[DTL_PARTIAL]);
1727 if (scrub_done)
1728 space_map_vacate(&vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
1729 space_map_vacate(&vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
1730 if (!vdev_readable(vd))
1731 space_map_add(&vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
1732 else
1733 space_map_walk(&vd->vdev_dtl[DTL_MISSING],
1734 space_map_add, &vd->vdev_dtl[DTL_OUTAGE]);
1735 mutex_exit(&vd->vdev_dtl_lock);
1736
1737 if (txg != 0)
1738 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
1739 return;
1740 }
1741
1742 mutex_enter(&vd->vdev_dtl_lock);
1743 for (int t = 0; t < DTL_TYPES; t++) {
1744 /* account for child's outage in parent's missing map */
1745 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t;
1746 if (t == DTL_SCRUB)
1747 continue; /* leaf vdevs only */
1748 if (t == DTL_PARTIAL)
1749 minref = 1; /* i.e. non-zero */
1750 else if (vd->vdev_nparity != 0)
1751 minref = vd->vdev_nparity + 1; /* RAID-Z */
1752 else
1753 minref = vd->vdev_children; /* any kind of mirror */
1754 space_map_ref_create(&reftree);
1755 for (int c = 0; c < vd->vdev_children; c++) {
1756 vdev_t *cvd = vd->vdev_child[c];
1757 mutex_enter(&cvd->vdev_dtl_lock);
1758 space_map_ref_add_map(&reftree, &cvd->vdev_dtl[s], 1);
1759 mutex_exit(&cvd->vdev_dtl_lock);
1760 }
1761 space_map_ref_generate_map(&reftree, &vd->vdev_dtl[t], minref);
1762 space_map_ref_destroy(&reftree);
1763 }
1764 mutex_exit(&vd->vdev_dtl_lock);
1765 }
1766
1767 static int
1768 vdev_dtl_load(vdev_t *vd)
1769 {
1770 spa_t *spa = vd->vdev_spa;
1771 space_map_obj_t *smo = &vd->vdev_dtl_smo;
1772 objset_t *mos = spa->spa_meta_objset;
1773 dmu_buf_t *db;
1774 int error;
1775
1776 ASSERT(vd->vdev_children == 0);
1777
1778 if (smo->smo_object == 0)
1779 return (0);
1780
1781 ASSERT(!vd->vdev_ishole);
1782
1783 if ((error = dmu_bonus_hold(mos, smo->smo_object, FTAG, &db)) != 0)
1784 return (error);
1785
1786 ASSERT3U(db->db_size, >=, sizeof (*smo));
1787 bcopy(db->db_data, smo, sizeof (*smo));
1788 dmu_buf_rele(db, FTAG);
1789
1790 mutex_enter(&vd->vdev_dtl_lock);
1791 error = space_map_load(&vd->vdev_dtl[DTL_MISSING],
1792 NULL, SM_ALLOC, smo, mos);
1793 mutex_exit(&vd->vdev_dtl_lock);
1794
1795 return (error);
1796 }
1797
1798 void
1799 vdev_dtl_sync(vdev_t *vd, uint64_t txg)
1800 {
1801 spa_t *spa = vd->vdev_spa;
1802 space_map_obj_t *smo = &vd->vdev_dtl_smo;
1803 space_map_t *sm = &vd->vdev_dtl[DTL_MISSING];
1804 objset_t *mos = spa->spa_meta_objset;
1805 space_map_t smsync;
1806 kmutex_t smlock;
1807 dmu_buf_t *db;
1808 dmu_tx_t *tx;
1809
1810 ASSERT(!vd->vdev_ishole);
1811
1812 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
1813
1814 if (vd->vdev_detached) {
1815 if (smo->smo_object != 0) {
1816 int err = dmu_object_free(mos, smo->smo_object, tx);
1817 ASSERT0(err);
1818 smo->smo_object = 0;
1819 }
1820 dmu_tx_commit(tx);
1821 return;
1822 }
1823
1824 if (smo->smo_object == 0) {
1825 ASSERT(smo->smo_objsize == 0);
1826 ASSERT(smo->smo_alloc == 0);
1827 smo->smo_object = dmu_object_alloc(mos,
1828 DMU_OT_SPACE_MAP, 1 << SPACE_MAP_BLOCKSHIFT,
1829 DMU_OT_SPACE_MAP_HEADER, sizeof (*smo), tx);
1830 ASSERT(smo->smo_object != 0);
1831 vdev_config_dirty(vd->vdev_top);
1832 }
1833
1834 mutex_init(&smlock, NULL, MUTEX_DEFAULT, NULL);
1835
1836 space_map_create(&smsync, sm->sm_start, sm->sm_size, sm->sm_shift,
1837 &smlock);
1838
1839 mutex_enter(&smlock);
1840
1841 mutex_enter(&vd->vdev_dtl_lock);
1842 space_map_walk(sm, space_map_add, &smsync);
1843 mutex_exit(&vd->vdev_dtl_lock);
1844
1845 space_map_truncate(smo, mos, tx);
1846 space_map_sync(&smsync, SM_ALLOC, smo, mos, tx);
1847 space_map_vacate(&smsync, NULL, NULL);
1848
1849 space_map_destroy(&smsync);
1850
1851 mutex_exit(&smlock);
1852 mutex_destroy(&smlock);
1853
1854 VERIFY(0 == dmu_bonus_hold(mos, smo->smo_object, FTAG, &db));
1855 dmu_buf_will_dirty(db, tx);
1856 ASSERT3U(db->db_size, >=, sizeof (*smo));
1857 bcopy(smo, db->db_data, sizeof (*smo));
1858 dmu_buf_rele(db, FTAG);
1859
1860 dmu_tx_commit(tx);
1861 }
1862
1863 /*
1864 * Determine whether the specified vdev can be offlined/detached/removed
1865 * without losing data.
1866 */
1867 boolean_t
1868 vdev_dtl_required(vdev_t *vd)
1869 {
1870 spa_t *spa = vd->vdev_spa;
1871 vdev_t *tvd = vd->vdev_top;
1872 uint8_t cant_read = vd->vdev_cant_read;
1873 boolean_t required;
1874
1875 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1876
1877 if (vd == spa->spa_root_vdev || vd == tvd)
1878 return (B_TRUE);
1879
1880 /*
1881 * Temporarily mark the device as unreadable, and then determine
1882 * whether this results in any DTL outages in the top-level vdev.
1883 * If not, we can safely offline/detach/remove the device.
1884 */
1885 vd->vdev_cant_read = B_TRUE;
1886 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
1887 required = !vdev_dtl_empty(tvd, DTL_OUTAGE);
1888 vd->vdev_cant_read = cant_read;
1889 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
1890
1891 if (!required && zio_injection_enabled)
1892 required = !!zio_handle_device_injection(vd, NULL, ECHILD);
1893
1894 return (required);
1895 }
1896
1897 /*
1898 * Determine if resilver is needed, and if so the txg range.
1899 */
1900 boolean_t
1901 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
1902 {
1903 boolean_t needed = B_FALSE;
1904 uint64_t thismin = UINT64_MAX;
1905 uint64_t thismax = 0;
1906
1907 if (vd->vdev_children == 0) {
1908 mutex_enter(&vd->vdev_dtl_lock);
1909 if (vd->vdev_dtl[DTL_MISSING].sm_space != 0 &&
1910 vdev_writeable(vd)) {
1911 space_seg_t *ss;
1912
1913 ss = avl_first(&vd->vdev_dtl[DTL_MISSING].sm_root);
1914 thismin = ss->ss_start - 1;
1915 ss = avl_last(&vd->vdev_dtl[DTL_MISSING].sm_root);
1916 thismax = ss->ss_end;
1917 needed = B_TRUE;
1918 }
1919 mutex_exit(&vd->vdev_dtl_lock);
1920 } else {
1921 for (int c = 0; c < vd->vdev_children; c++) {
1922 vdev_t *cvd = vd->vdev_child[c];
1923 uint64_t cmin, cmax;
1924
1925 if (vdev_resilver_needed(cvd, &cmin, &cmax)) {
1926 thismin = MIN(thismin, cmin);
1927 thismax = MAX(thismax, cmax);
1928 needed = B_TRUE;
1929 }
1930 }
1931 }
1932
1933 if (needed && minp) {
1934 *minp = thismin;
1935 *maxp = thismax;
1936 }
1937 return (needed);
1938 }
1939
1940 void
1941 vdev_load(vdev_t *vd)
1942 {
1943 /*
1944 * Recursively load all children.
1945 */
1946 for (int c = 0; c < vd->vdev_children; c++)
1947 vdev_load(vd->vdev_child[c]);
1948
1949 /*
1950 * If this is a top-level vdev, initialize its metaslabs.
1951 */
1952 if (vd == vd->vdev_top && !vd->vdev_ishole &&
1953 (vd->vdev_ashift == 0 || vd->vdev_asize == 0 ||
1954 vdev_metaslab_init(vd, 0) != 0))
1955 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1956 VDEV_AUX_CORRUPT_DATA);
1957
1958 /*
1959 * If this is a leaf vdev, load its DTL.
1960 */
1961 if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0)
1962 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1963 VDEV_AUX_CORRUPT_DATA);
1964 }
1965
1966 /*
1967 * The special vdev case is used for hot spares and l2cache devices. Its
1968 * sole purpose it to set the vdev state for the associated vdev. To do this,
1969 * we make sure that we can open the underlying device, then try to read the
1970 * label, and make sure that the label is sane and that it hasn't been
1971 * repurposed to another pool.
1972 */
1973 int
1974 vdev_validate_aux(vdev_t *vd)
1975 {
1976 nvlist_t *label;
1977 uint64_t guid, version;
1978 uint64_t state;
1979
1980 if (!vdev_readable(vd))
1981 return (0);
1982
1983 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) {
1984 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1985 VDEV_AUX_CORRUPT_DATA);
1986 return (-1);
1987 }
1988
1989 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 ||
1990 !SPA_VERSION_IS_SUPPORTED(version) ||
1991 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 ||
1992 guid != vd->vdev_guid ||
1993 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) {
1994 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1995 VDEV_AUX_CORRUPT_DATA);
1996 nvlist_free(label);
1997 return (-1);
1998 }
1999
2000 /*
2001 * We don't actually check the pool state here. If it's in fact in
2002 * use by another pool, we update this fact on the fly when requested.
2003 */
2004 nvlist_free(label);
2005 return (0);
2006 }
2007
2008 void
2009 vdev_remove(vdev_t *vd, uint64_t txg)
2010 {
2011 spa_t *spa = vd->vdev_spa;
2012 objset_t *mos = spa->spa_meta_objset;
2013 dmu_tx_t *tx;
2014
2015 tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
2016
2017 if (vd->vdev_dtl_smo.smo_object) {
2018 ASSERT0(vd->vdev_dtl_smo.smo_alloc);
2019 (void) dmu_object_free(mos, vd->vdev_dtl_smo.smo_object, tx);
2020 vd->vdev_dtl_smo.smo_object = 0;
2021 }
2022
2023 if (vd->vdev_ms != NULL) {
2024 for (int m = 0; m < vd->vdev_ms_count; m++) {
2025 metaslab_t *msp = vd->vdev_ms[m];
2026
2027 if (msp == NULL || msp->ms_smo.smo_object == 0)
2028 continue;
2029
2030 ASSERT0(msp->ms_smo.smo_alloc);
2031 (void) dmu_object_free(mos, msp->ms_smo.smo_object, tx);
2032 msp->ms_smo.smo_object = 0;
2033 }
2034 }
2035
2036 if (vd->vdev_ms_array) {
2037 (void) dmu_object_free(mos, vd->vdev_ms_array, tx);
2038 vd->vdev_ms_array = 0;
2039 vd->vdev_ms_shift = 0;
2040 }
2041 dmu_tx_commit(tx);
2042 }
2043
2044 void
2045 vdev_sync_done(vdev_t *vd, uint64_t txg)
2046 {
2047 metaslab_t *msp;
2048 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg));
2049
2050 ASSERT(!vd->vdev_ishole);
2051
2052 while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))
2053 metaslab_sync_done(msp, txg);
2054
2055 if (reassess)
2056 metaslab_sync_reassess(vd->vdev_mg);
2057 }
2058
2059 void
2060 vdev_sync(vdev_t *vd, uint64_t txg)
2061 {
2062 spa_t *spa = vd->vdev_spa;
2063 vdev_t *lvd;
2064 metaslab_t *msp;
2065 dmu_tx_t *tx;
2066
2067 ASSERT(!vd->vdev_ishole);
2068
2069 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) {
2070 ASSERT(vd == vd->vdev_top);
2071 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2072 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset,
2073 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx);
2074 ASSERT(vd->vdev_ms_array != 0);
2075 vdev_config_dirty(vd);
2076 dmu_tx_commit(tx);
2077 }
2078
2079 /*
2080 * Remove the metadata associated with this vdev once it's empty.
2081 */
2082 if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing)
2083 vdev_remove(vd, txg);
2084
2085 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) {
2086 metaslab_sync(msp, txg);
2087 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg));
2088 }
2089
2090 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL)
2091 vdev_dtl_sync(lvd, txg);
2092
2093 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg));
2094 }
2095
2096 uint64_t
2097 vdev_psize_to_asize(vdev_t *vd, uint64_t psize)
2098 {
2099 return (vd->vdev_ops->vdev_op_asize(vd, psize));
2100 }
2101
2102 /*
2103 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2104 * not be opened, and no I/O is attempted.
2105 */
2106 int
2107 vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2108 {
2109 vdev_t *vd, *tvd;
2110
2111 spa_vdev_state_enter(spa, SCL_NONE);
2112
2113 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2114 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2115
2116 if (!vd->vdev_ops->vdev_op_leaf)
2117 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2118
2119 tvd = vd->vdev_top;
2120
2121 /*
2122 * We don't directly use the aux state here, but if we do a
2123 * vdev_reopen(), we need this value to be present to remember why we
2124 * were faulted.
2125 */
2126 vd->vdev_label_aux = aux;
2127
2128 /*
2129 * Faulted state takes precedence over degraded.
2130 */
2131 vd->vdev_delayed_close = B_FALSE;
2132 vd->vdev_faulted = 1ULL;
2133 vd->vdev_degraded = 0ULL;
2134 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux);
2135
2136 /*
2137 * If this device has the only valid copy of the data, then
2138 * back off and simply mark the vdev as degraded instead.
2139 */
2140 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) {
2141 vd->vdev_degraded = 1ULL;
2142 vd->vdev_faulted = 0ULL;
2143
2144 /*
2145 * If we reopen the device and it's not dead, only then do we
2146 * mark it degraded.
2147 */
2148 vdev_reopen(tvd);
2149
2150 if (vdev_readable(vd))
2151 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux);
2152 }
2153
2154 return (spa_vdev_state_exit(spa, vd, 0));
2155 }
2156
2157 /*
2158 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2159 * user that something is wrong. The vdev continues to operate as normal as far
2160 * as I/O is concerned.
2161 */
2162 int
2163 vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2164 {
2165 vdev_t *vd;
2166
2167 spa_vdev_state_enter(spa, SCL_NONE);
2168
2169 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2170 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2171
2172 if (!vd->vdev_ops->vdev_op_leaf)
2173 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2174
2175 /*
2176 * If the vdev is already faulted, then don't do anything.
2177 */
2178 if (vd->vdev_faulted || vd->vdev_degraded)
2179 return (spa_vdev_state_exit(spa, NULL, 0));
2180
2181 vd->vdev_degraded = 1ULL;
2182 if (!vdev_is_dead(vd))
2183 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
2184 aux);
2185
2186 return (spa_vdev_state_exit(spa, vd, 0));
2187 }
2188
2189 /*
2190 * Online the given vdev.
2191 *
2192 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
2193 * spare device should be detached when the device finishes resilvering.
2194 * Second, the online should be treated like a 'test' online case, so no FMA
2195 * events are generated if the device fails to open.
2196 */
2197 int
2198 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
2199 {
2200 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev;
2201
2202 spa_vdev_state_enter(spa, SCL_NONE);
2203
2204 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2205 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2206
2207 if (!vd->vdev_ops->vdev_op_leaf)
2208 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2209
2210 tvd = vd->vdev_top;
2211 vd->vdev_offline = B_FALSE;
2212 vd->vdev_tmpoffline = B_FALSE;
2213 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE);
2214 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT);
2215
2216 /* XXX - L2ARC 1.0 does not support expansion */
2217 if (!vd->vdev_aux) {
2218 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2219 pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND);
2220 }
2221
2222 vdev_reopen(tvd);
2223 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE;
2224
2225 if (!vd->vdev_aux) {
2226 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2227 pvd->vdev_expanding = B_FALSE;
2228 }
2229
2230 if (newstate)
2231 *newstate = vd->vdev_state;
2232 if ((flags & ZFS_ONLINE_UNSPARE) &&
2233 !vdev_is_dead(vd) && vd->vdev_parent &&
2234 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2235 vd->vdev_parent->vdev_child[0] == vd)
2236 vd->vdev_unspare = B_TRUE;
2237
2238 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) {
2239
2240 /* XXX - L2ARC 1.0 does not support expansion */
2241 if (vd->vdev_aux)
2242 return (spa_vdev_state_exit(spa, vd, ENOTSUP));
2243 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2244 }
2245 return (spa_vdev_state_exit(spa, vd, 0));
2246 }
2247
2248 static int
2249 vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags)
2250 {
2251 vdev_t *vd, *tvd;
2252 int error = 0;
2253 uint64_t generation;
2254 metaslab_group_t *mg;
2255
2256 top:
2257 spa_vdev_state_enter(spa, SCL_ALLOC);
2258
2259 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2260 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2261
2262 if (!vd->vdev_ops->vdev_op_leaf)
2263 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2264
2265 tvd = vd->vdev_top;
2266 mg = tvd->vdev_mg;
2267 generation = spa->spa_config_generation + 1;
2268
2269 /*
2270 * If the device isn't already offline, try to offline it.
2271 */
2272 if (!vd->vdev_offline) {
2273 /*
2274 * If this device has the only valid copy of some data,
2275 * don't allow it to be offlined. Log devices are always
2276 * expendable.
2277 */
2278 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2279 vdev_dtl_required(vd))
2280 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2281
2282 /*
2283 * If the top-level is a slog and it has had allocations
2284 * then proceed. We check that the vdev's metaslab group
2285 * is not NULL since it's possible that we may have just
2286 * added this vdev but not yet initialized its metaslabs.
2287 */
2288 if (tvd->vdev_islog && mg != NULL) {
2289 /*
2290 * Prevent any future allocations.
2291 */
2292 metaslab_group_passivate(mg);
2293 (void) spa_vdev_state_exit(spa, vd, 0);
2294
2295 error = spa_offline_log(spa);
2296
2297 spa_vdev_state_enter(spa, SCL_ALLOC);
2298
2299 /*
2300 * Check to see if the config has changed.
2301 */
2302 if (error || generation != spa->spa_config_generation) {
2303 metaslab_group_activate(mg);
2304 if (error)
2305 return (spa_vdev_state_exit(spa,
2306 vd, error));
2307 (void) spa_vdev_state_exit(spa, vd, 0);
2308 goto top;
2309 }
2310 ASSERT0(tvd->vdev_stat.vs_alloc);
2311 }
2312
2313 /*
2314 * Offline this device and reopen its top-level vdev.
2315 * If the top-level vdev is a log device then just offline
2316 * it. Otherwise, if this action results in the top-level
2317 * vdev becoming unusable, undo it and fail the request.
2318 */
2319 vd->vdev_offline = B_TRUE;
2320 vdev_reopen(tvd);
2321
2322 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2323 vdev_is_dead(tvd)) {
2324 vd->vdev_offline = B_FALSE;
2325 vdev_reopen(tvd);
2326 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2327 }
2328
2329 /*
2330 * Add the device back into the metaslab rotor so that
2331 * once we online the device it's open for business.
2332 */
2333 if (tvd->vdev_islog && mg != NULL)
2334 metaslab_group_activate(mg);
2335 }
2336
2337 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY);
2338
2339 return (spa_vdev_state_exit(spa, vd, 0));
2340 }
2341
2342 int
2343 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags)
2344 {
2345 int error;
2346
2347 mutex_enter(&spa->spa_vdev_top_lock);
2348 error = vdev_offline_locked(spa, guid, flags);
2349 mutex_exit(&spa->spa_vdev_top_lock);
2350
2351 return (error);
2352 }
2353
2354 /*
2355 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2356 * vdev_offline(), we assume the spa config is locked. We also clear all
2357 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2358 */
2359 void
2360 vdev_clear(spa_t *spa, vdev_t *vd)
2361 {
2362 vdev_t *rvd = spa->spa_root_vdev;
2363
2364 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2365
2366 if (vd == NULL)
2367 vd = rvd;
2368
2369 vd->vdev_stat.vs_read_errors = 0;
2370 vd->vdev_stat.vs_write_errors = 0;
2371 vd->vdev_stat.vs_checksum_errors = 0;
2372
2373 for (int c = 0; c < vd->vdev_children; c++)
2374 vdev_clear(spa, vd->vdev_child[c]);
2375
2376 /*
2377 * If we're in the FAULTED state or have experienced failed I/O, then
2378 * clear the persistent state and attempt to reopen the device. We
2379 * also mark the vdev config dirty, so that the new faulted state is
2380 * written out to disk.
2381 */
2382 if (vd->vdev_faulted || vd->vdev_degraded ||
2383 !vdev_readable(vd) || !vdev_writeable(vd)) {
2384
2385 /*
2386 * When reopening in reponse to a clear event, it may be due to
2387 * a fmadm repair request. In this case, if the device is
2388 * still broken, we want to still post the ereport again.
2389 */
2390 vd->vdev_forcefault = B_TRUE;
2391
2392 vd->vdev_faulted = vd->vdev_degraded = 0ULL;
2393 vd->vdev_cant_read = B_FALSE;
2394 vd->vdev_cant_write = B_FALSE;
2395
2396 vdev_reopen(vd == rvd ? rvd : vd->vdev_top);
2397
2398 vd->vdev_forcefault = B_FALSE;
2399
2400 if (vd != rvd && vdev_writeable(vd->vdev_top))
2401 vdev_state_dirty(vd->vdev_top);
2402
2403 if (vd->vdev_aux == NULL && !vdev_is_dead(vd))
2404 spa_async_request(spa, SPA_ASYNC_RESILVER);
2405
2406 spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR);
2407 }
2408
2409 /*
2410 * When clearing a FMA-diagnosed fault, we always want to
2411 * unspare the device, as we assume that the original spare was
2412 * done in response to the FMA fault.
2413 */
2414 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL &&
2415 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2416 vd->vdev_parent->vdev_child[0] == vd)
2417 vd->vdev_unspare = B_TRUE;
2418 }
2419
2420 boolean_t
2421 vdev_is_dead(vdev_t *vd)
2422 {
2423 /*
2424 * Holes and missing devices are always considered "dead".
2425 * This simplifies the code since we don't have to check for
2426 * these types of devices in the various code paths.
2427 * Instead we rely on the fact that we skip over dead devices
2428 * before issuing I/O to them.
2429 */
2430 return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole ||
2431 vd->vdev_ops == &vdev_missing_ops);
2432 }
2433
2434 boolean_t
2435 vdev_readable(vdev_t *vd)
2436 {
2437 return (!vdev_is_dead(vd) && !vd->vdev_cant_read);
2438 }
2439
2440 boolean_t
2441 vdev_writeable(vdev_t *vd)
2442 {
2443 return (!vdev_is_dead(vd) && !vd->vdev_cant_write);
2444 }
2445
2446 boolean_t
2447 vdev_allocatable(vdev_t *vd)
2448 {
2449 uint64_t state = vd->vdev_state;
2450
2451 /*
2452 * We currently allow allocations from vdevs which may be in the
2453 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2454 * fails to reopen then we'll catch it later when we're holding
2455 * the proper locks. Note that we have to get the vdev state
2456 * in a local variable because although it changes atomically,
2457 * we're asking two separate questions about it.
2458 */
2459 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) &&
2460 !vd->vdev_cant_write && !vd->vdev_ishole);
2461 }
2462
2463 boolean_t
2464 vdev_accessible(vdev_t *vd, zio_t *zio)
2465 {
2466 ASSERT(zio->io_vd == vd);
2467
2468 if (vdev_is_dead(vd) || vd->vdev_remove_wanted)
2469 return (B_FALSE);
2470
2471 if (zio->io_type == ZIO_TYPE_READ)
2472 return (!vd->vdev_cant_read);
2473
2474 if (zio->io_type == ZIO_TYPE_WRITE)
2475 return (!vd->vdev_cant_write);
2476
2477 return (B_TRUE);
2478 }
2479
2480 /*
2481 * Get statistics for the given vdev.
2482 */
2483 void
2484 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
2485 {
2486 vdev_t *rvd = vd->vdev_spa->spa_root_vdev;
2487
2488 mutex_enter(&vd->vdev_stat_lock);
2489 bcopy(&vd->vdev_stat, vs, sizeof (*vs));
2490 vs->vs_timestamp = gethrtime() - vs->vs_timestamp;
2491 vs->vs_state = vd->vdev_state;
2492 vs->vs_rsize = vdev_get_min_asize(vd);
2493 if (vd->vdev_ops->vdev_op_leaf)
2494 vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
2495 vs->vs_esize = vd->vdev_max_asize - vd->vdev_asize;
2496 mutex_exit(&vd->vdev_stat_lock);
2497
2498 /*
2499 * If we're getting stats on the root vdev, aggregate the I/O counts
2500 * over all top-level vdevs (i.e. the direct children of the root).
2501 */
2502 if (vd == rvd) {
2503 for (int c = 0; c < rvd->vdev_children; c++) {
2504 vdev_t *cvd = rvd->vdev_child[c];
2505 vdev_stat_t *cvs = &cvd->vdev_stat;
2506
2507 mutex_enter(&vd->vdev_stat_lock);
2508 for (int t = 0; t < ZIO_TYPES; t++) {
2509 vs->vs_ops[t] += cvs->vs_ops[t];
2510 vs->vs_bytes[t] += cvs->vs_bytes[t];
2511 }
2512 cvs->vs_scan_removing = cvd->vdev_removing;
2513 mutex_exit(&vd->vdev_stat_lock);
2514 }
2515 }
2516 }
2517
2518 void
2519 vdev_clear_stats(vdev_t *vd)
2520 {
2521 mutex_enter(&vd->vdev_stat_lock);
2522 vd->vdev_stat.vs_space = 0;
2523 vd->vdev_stat.vs_dspace = 0;
2524 vd->vdev_stat.vs_alloc = 0;
2525 mutex_exit(&vd->vdev_stat_lock);
2526 }
2527
2528 void
2529 vdev_scan_stat_init(vdev_t *vd)
2530 {
2531 vdev_stat_t *vs = &vd->vdev_stat;
2532
2533 for (int c = 0; c < vd->vdev_children; c++)
2534 vdev_scan_stat_init(vd->vdev_child[c]);
2535
2536 mutex_enter(&vd->vdev_stat_lock);
2537 vs->vs_scan_processed = 0;
2538 mutex_exit(&vd->vdev_stat_lock);
2539 }
2540
2541 void
2542 vdev_stat_update(zio_t *zio, uint64_t psize)
2543 {
2544 spa_t *spa = zio->io_spa;
2545 vdev_t *rvd = spa->spa_root_vdev;
2546 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd;
2547 vdev_t *pvd;
2548 uint64_t txg = zio->io_txg;
2549 vdev_stat_t *vs = &vd->vdev_stat;
2550 zio_type_t type = zio->io_type;
2551 int flags = zio->io_flags;
2552
2553 /*
2554 * If this i/o is a gang leader, it didn't do any actual work.
2555 */
2556 if (zio->io_gang_tree)
2557 return;
2558
2559 if (zio->io_error == 0) {
2560 /*
2561 * If this is a root i/o, don't count it -- we've already
2562 * counted the top-level vdevs, and vdev_get_stats() will
2563 * aggregate them when asked. This reduces contention on
2564 * the root vdev_stat_lock and implicitly handles blocks
2565 * that compress away to holes, for which there is no i/o.
2566 * (Holes never create vdev children, so all the counters
2567 * remain zero, which is what we want.)
2568 *
2569 * Note: this only applies to successful i/o (io_error == 0)
2570 * because unlike i/o counts, errors are not additive.
2571 * When reading a ditto block, for example, failure of
2572 * one top-level vdev does not imply a root-level error.
2573 */
2574 if (vd == rvd)
2575 return;
2576
2577 ASSERT(vd == zio->io_vd);
2578
2579 if (flags & ZIO_FLAG_IO_BYPASS)
2580 return;
2581
2582 mutex_enter(&vd->vdev_stat_lock);
2583
2584 if (flags & ZIO_FLAG_IO_REPAIR) {
2585 if (flags & ZIO_FLAG_SCAN_THREAD) {
2586 dsl_scan_phys_t *scn_phys =
2587 &spa->spa_dsl_pool->dp_scan->scn_phys;
2588 uint64_t *processed = &scn_phys->scn_processed;
2589
2590 /* XXX cleanup? */
2591 if (vd->vdev_ops->vdev_op_leaf)
2592 atomic_add_64(processed, psize);
2593 vs->vs_scan_processed += psize;
2594 }
2595
2596 if (flags & ZIO_FLAG_SELF_HEAL)
2597 vs->vs_self_healed += psize;
2598 }
2599
2600 vs->vs_ops[type]++;
2601 vs->vs_bytes[type] += psize;
2602
2603 mutex_exit(&vd->vdev_stat_lock);
2604 return;
2605 }
2606
2607 if (flags & ZIO_FLAG_SPECULATIVE)
2608 return;
2609
2610 /*
2611 * If this is an I/O error that is going to be retried, then ignore the
2612 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
2613 * hard errors, when in reality they can happen for any number of
2614 * innocuous reasons (bus resets, MPxIO link failure, etc).
2615 */
2616 if (zio->io_error == EIO &&
2617 !(zio->io_flags & ZIO_FLAG_IO_RETRY))
2618 return;
2619
2620 /*
2621 * Intent logs writes won't propagate their error to the root
2622 * I/O so don't mark these types of failures as pool-level
2623 * errors.
2624 */
2625 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE))
2626 return;
2627
2628 mutex_enter(&vd->vdev_stat_lock);
2629 if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) {
2630 if (zio->io_error == ECKSUM)
2631 vs->vs_checksum_errors++;
2632 else
2633 vs->vs_read_errors++;
2634 }
2635 if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd))
2636 vs->vs_write_errors++;
2637 mutex_exit(&vd->vdev_stat_lock);
2638
2639 if (type == ZIO_TYPE_WRITE && txg != 0 &&
2640 (!(flags & ZIO_FLAG_IO_REPAIR) ||
2641 (flags & ZIO_FLAG_SCAN_THREAD) ||
2642 spa->spa_claiming)) {
2643 /*
2644 * This is either a normal write (not a repair), or it's
2645 * a repair induced by the scrub thread, or it's a repair
2646 * made by zil_claim() during spa_load() in the first txg.
2647 * In the normal case, we commit the DTL change in the same
2648 * txg as the block was born. In the scrub-induced repair
2649 * case, we know that scrubs run in first-pass syncing context,
2650 * so we commit the DTL change in spa_syncing_txg(spa).
2651 * In the zil_claim() case, we commit in spa_first_txg(spa).
2652 *
2653 * We currently do not make DTL entries for failed spontaneous
2654 * self-healing writes triggered by normal (non-scrubbing)
2655 * reads, because we have no transactional context in which to
2656 * do so -- and it's not clear that it'd be desirable anyway.
2657 */
2658 if (vd->vdev_ops->vdev_op_leaf) {
2659 uint64_t commit_txg = txg;
2660 if (flags & ZIO_FLAG_SCAN_THREAD) {
2661 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2662 ASSERT(spa_sync_pass(spa) == 1);
2663 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1);
2664 commit_txg = spa_syncing_txg(spa);
2665 } else if (spa->spa_claiming) {
2666 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2667 commit_txg = spa_first_txg(spa);
2668 }
2669 ASSERT(commit_txg >= spa_syncing_txg(spa));
2670 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1))
2671 return;
2672 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2673 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1);
2674 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg);
2675 }
2676 if (vd != rvd)
2677 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1);
2678 }
2679 }
2680
2681 /*
2682 * Update the in-core space usage stats for this vdev, its metaslab class,
2683 * and the root vdev.
2684 */
2685 void
2686 vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta,
2687 int64_t space_delta)
2688 {
2689 int64_t dspace_delta = space_delta;
2690 spa_t *spa = vd->vdev_spa;
2691 vdev_t *rvd = spa->spa_root_vdev;
2692 metaslab_group_t *mg = vd->vdev_mg;
2693 metaslab_class_t *mc = mg ? mg->mg_class : NULL;
2694
2695 ASSERT(vd == vd->vdev_top);
2696
2697 /*
2698 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
2699 * factor. We must calculate this here and not at the root vdev
2700 * because the root vdev's psize-to-asize is simply the max of its
2701 * childrens', thus not accurate enough for us.
2702 */
2703 ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0);
2704 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache);
2705 dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) *
2706 vd->vdev_deflate_ratio;
2707
2708 mutex_enter(&vd->vdev_stat_lock);
2709 vd->vdev_stat.vs_alloc += alloc_delta;
2710 vd->vdev_stat.vs_space += space_delta;
2711 vd->vdev_stat.vs_dspace += dspace_delta;
2712 mutex_exit(&vd->vdev_stat_lock);
2713
2714 if (mc == spa_normal_class(spa)) {
2715 mutex_enter(&rvd->vdev_stat_lock);
2716 rvd->vdev_stat.vs_alloc += alloc_delta;
2717 rvd->vdev_stat.vs_space += space_delta;
2718 rvd->vdev_stat.vs_dspace += dspace_delta;
2719 mutex_exit(&rvd->vdev_stat_lock);
2720 }
2721
2722 if (mc != NULL) {
2723 ASSERT(rvd == vd->vdev_parent);
2724 ASSERT(vd->vdev_ms_count != 0);
2725
2726 metaslab_class_space_update(mc,
2727 alloc_delta, defer_delta, space_delta, dspace_delta);
2728 }
2729 }
2730
2731 /*
2732 * Mark a top-level vdev's config as dirty, placing it on the dirty list
2733 * so that it will be written out next time the vdev configuration is synced.
2734 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
2735 */
2736 void
2737 vdev_config_dirty(vdev_t *vd)
2738 {
2739 spa_t *spa = vd->vdev_spa;
2740 vdev_t *rvd = spa->spa_root_vdev;
2741 int c;
2742
2743 ASSERT(spa_writeable(spa));
2744
2745 /*
2746 * If this is an aux vdev (as with l2cache and spare devices), then we
2747 * update the vdev config manually and set the sync flag.
2748 */
2749 if (vd->vdev_aux != NULL) {
2750 spa_aux_vdev_t *sav = vd->vdev_aux;
2751 nvlist_t **aux;
2752 uint_t naux;
2753
2754 for (c = 0; c < sav->sav_count; c++) {
2755 if (sav->sav_vdevs[c] == vd)
2756 break;
2757 }
2758
2759 if (c == sav->sav_count) {
2760 /*
2761 * We're being removed. There's nothing more to do.
2762 */
2763 ASSERT(sav->sav_sync == B_TRUE);
2764 return;
2765 }
2766
2767 sav->sav_sync = B_TRUE;
2768
2769 if (nvlist_lookup_nvlist_array(sav->sav_config,
2770 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) {
2771 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
2772 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0);
2773 }
2774
2775 ASSERT(c < naux);
2776
2777 /*
2778 * Setting the nvlist in the middle if the array is a little
2779 * sketchy, but it will work.
2780 */
2781 nvlist_free(aux[c]);
2782 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0);
2783
2784 return;
2785 }
2786
2787 /*
2788 * The dirty list is protected by the SCL_CONFIG lock. The caller
2789 * must either hold SCL_CONFIG as writer, or must be the sync thread
2790 * (which holds SCL_CONFIG as reader). There's only one sync thread,
2791 * so this is sufficient to ensure mutual exclusion.
2792 */
2793 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
2794 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2795 spa_config_held(spa, SCL_CONFIG, RW_READER)));
2796
2797 if (vd == rvd) {
2798 for (c = 0; c < rvd->vdev_children; c++)
2799 vdev_config_dirty(rvd->vdev_child[c]);
2800 } else {
2801 ASSERT(vd == vd->vdev_top);
2802
2803 if (!list_link_active(&vd->vdev_config_dirty_node) &&
2804 !vd->vdev_ishole)
2805 list_insert_head(&spa->spa_config_dirty_list, vd);
2806 }
2807 }
2808
2809 void
2810 vdev_config_clean(vdev_t *vd)
2811 {
2812 spa_t *spa = vd->vdev_spa;
2813
2814 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
2815 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2816 spa_config_held(spa, SCL_CONFIG, RW_READER)));
2817
2818 ASSERT(list_link_active(&vd->vdev_config_dirty_node));
2819 list_remove(&spa->spa_config_dirty_list, vd);
2820 }
2821
2822 /*
2823 * Mark a top-level vdev's state as dirty, so that the next pass of
2824 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
2825 * the state changes from larger config changes because they require
2826 * much less locking, and are often needed for administrative actions.
2827 */
2828 void
2829 vdev_state_dirty(vdev_t *vd)
2830 {
2831 spa_t *spa = vd->vdev_spa;
2832
2833 ASSERT(spa_writeable(spa));
2834 ASSERT(vd == vd->vdev_top);
2835
2836 /*
2837 * The state list is protected by the SCL_STATE lock. The caller
2838 * must either hold SCL_STATE as writer, or must be the sync thread
2839 * (which holds SCL_STATE as reader). There's only one sync thread,
2840 * so this is sufficient to ensure mutual exclusion.
2841 */
2842 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
2843 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2844 spa_config_held(spa, SCL_STATE, RW_READER)));
2845
2846 if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole)
2847 list_insert_head(&spa->spa_state_dirty_list, vd);
2848 }
2849
2850 void
2851 vdev_state_clean(vdev_t *vd)
2852 {
2853 spa_t *spa = vd->vdev_spa;
2854
2855 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
2856 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2857 spa_config_held(spa, SCL_STATE, RW_READER)));
2858
2859 ASSERT(list_link_active(&vd->vdev_state_dirty_node));
2860 list_remove(&spa->spa_state_dirty_list, vd);
2861 }
2862
2863 /*
2864 * Propagate vdev state up from children to parent.
2865 */
2866 void
2867 vdev_propagate_state(vdev_t *vd)
2868 {
2869 spa_t *spa = vd->vdev_spa;
2870 vdev_t *rvd = spa->spa_root_vdev;
2871 int degraded = 0, faulted = 0;
2872 int corrupted = 0;
2873 vdev_t *child;
2874
2875 if (vd->vdev_children > 0) {
2876 for (int c = 0; c < vd->vdev_children; c++) {
2877 child = vd->vdev_child[c];
2878
2879 /*
2880 * Don't factor holes into the decision.
2881 */
2882 if (child->vdev_ishole)
2883 continue;
2884
2885 if (!vdev_readable(child) ||
2886 (!vdev_writeable(child) && spa_writeable(spa))) {
2887 /*
2888 * Root special: if there is a top-level log
2889 * device, treat the root vdev as if it were
2890 * degraded.
2891 */
2892 if (child->vdev_islog && vd == rvd)
2893 degraded++;
2894 else
2895 faulted++;
2896 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) {
2897 degraded++;
2898 }
2899
2900 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA)
2901 corrupted++;
2902 }
2903
2904 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded);
2905
2906 /*
2907 * Root special: if there is a top-level vdev that cannot be
2908 * opened due to corrupted metadata, then propagate the root
2909 * vdev's aux state as 'corrupt' rather than 'insufficient
2910 * replicas'.
2911 */
2912 if (corrupted && vd == rvd &&
2913 rvd->vdev_state == VDEV_STATE_CANT_OPEN)
2914 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN,
2915 VDEV_AUX_CORRUPT_DATA);
2916 }
2917
2918 if (vd->vdev_parent)
2919 vdev_propagate_state(vd->vdev_parent);
2920 }
2921
2922 /*
2923 * Set a vdev's state. If this is during an open, we don't update the parent
2924 * state, because we're in the process of opening children depth-first.
2925 * Otherwise, we propagate the change to the parent.
2926 *
2927 * If this routine places a device in a faulted state, an appropriate ereport is
2928 * generated.
2929 */
2930 void
2931 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
2932 {
2933 uint64_t save_state;
2934 spa_t *spa = vd->vdev_spa;
2935
2936 if (state == vd->vdev_state) {
2937 vd->vdev_stat.vs_aux = aux;
2938 return;
2939 }
2940
2941 save_state = vd->vdev_state;
2942
2943 vd->vdev_state = state;
2944 vd->vdev_stat.vs_aux = aux;
2945
2946 /*
2947 * If we are setting the vdev state to anything but an open state, then
2948 * always close the underlying device unless the device has requested
2949 * a delayed close (i.e. we're about to remove or fault the device).
2950 * Otherwise, we keep accessible but invalid devices open forever.
2951 * We don't call vdev_close() itself, because that implies some extra
2952 * checks (offline, etc) that we don't want here. This is limited to
2953 * leaf devices, because otherwise closing the device will affect other
2954 * children.
2955 */
2956 if (!vd->vdev_delayed_close && vdev_is_dead(vd) &&
2957 vd->vdev_ops->vdev_op_leaf)
2958 vd->vdev_ops->vdev_op_close(vd);
2959
2960 /*
2961 * If we have brought this vdev back into service, we need
2962 * to notify fmd so that it can gracefully repair any outstanding
2963 * cases due to a missing device. We do this in all cases, even those
2964 * that probably don't correlate to a repaired fault. This is sure to
2965 * catch all cases, and we let the zfs-retire agent sort it out. If
2966 * this is a transient state it's OK, as the retire agent will
2967 * double-check the state of the vdev before repairing it.
2968 */
2969 if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf &&
2970 vd->vdev_prevstate != state)
2971 zfs_post_state_change(spa, vd);
2972
2973 if (vd->vdev_removed &&
2974 state == VDEV_STATE_CANT_OPEN &&
2975 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) {
2976 /*
2977 * If the previous state is set to VDEV_STATE_REMOVED, then this
2978 * device was previously marked removed and someone attempted to
2979 * reopen it. If this failed due to a nonexistent device, then
2980 * keep the device in the REMOVED state. We also let this be if
2981 * it is one of our special test online cases, which is only
2982 * attempting to online the device and shouldn't generate an FMA
2983 * fault.
2984 */
2985 vd->vdev_state = VDEV_STATE_REMOVED;
2986 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
2987 } else if (state == VDEV_STATE_REMOVED) {
2988 vd->vdev_removed = B_TRUE;
2989 } else if (state == VDEV_STATE_CANT_OPEN) {
2990 /*
2991 * If we fail to open a vdev during an import or recovery, we
2992 * mark it as "not available", which signifies that it was
2993 * never there to begin with. Failure to open such a device
2994 * is not considered an error.
2995 */
2996 if ((spa_load_state(spa) == SPA_LOAD_IMPORT ||
2997 spa_load_state(spa) == SPA_LOAD_RECOVER) &&
2998 vd->vdev_ops->vdev_op_leaf)
2999 vd->vdev_not_present = 1;
3000
3001 /*
3002 * Post the appropriate ereport. If the 'prevstate' field is
3003 * set to something other than VDEV_STATE_UNKNOWN, it indicates
3004 * that this is part of a vdev_reopen(). In this case, we don't
3005 * want to post the ereport if the device was already in the
3006 * CANT_OPEN state beforehand.
3007 *
3008 * If the 'checkremove' flag is set, then this is an attempt to
3009 * online the device in response to an insertion event. If we
3010 * hit this case, then we have detected an insertion event for a
3011 * faulted or offline device that wasn't in the removed state.
3012 * In this scenario, we don't post an ereport because we are
3013 * about to replace the device, or attempt an online with
3014 * vdev_forcefault, which will generate the fault for us.
3015 */
3016 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) &&
3017 !vd->vdev_not_present && !vd->vdev_checkremove &&
3018 vd != spa->spa_root_vdev) {
3019 const char *class;
3020
3021 switch (aux) {
3022 case VDEV_AUX_OPEN_FAILED:
3023 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED;
3024 break;
3025 case VDEV_AUX_CORRUPT_DATA:
3026 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA;
3027 break;
3028 case VDEV_AUX_NO_REPLICAS:
3029 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS;
3030 break;
3031 case VDEV_AUX_BAD_GUID_SUM:
3032 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM;
3033 break;
3034 case VDEV_AUX_TOO_SMALL:
3035 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL;
3036 break;
3037 case VDEV_AUX_BAD_LABEL:
3038 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL;
3039 break;
3040 default:
3041 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN;
3042 }
3043
3044 zfs_ereport_post(class, spa, vd, NULL, save_state, 0);
3045 }
3046
3047 /* Erase any notion of persistent removed state */
3048 vd->vdev_removed = B_FALSE;
3049 } else {
3050 vd->vdev_removed = B_FALSE;
3051 }
3052
3053 if (!isopen && vd->vdev_parent)
3054 vdev_propagate_state(vd->vdev_parent);
3055 }
3056
3057 /*
3058 * Check the vdev configuration to ensure that it's capable of supporting
3059 * a root pool. Currently, we do not support RAID-Z or partial configuration.
3060 * In addition, only a single top-level vdev is allowed and none of the leaves
3061 * can be wholedisks.
3062 */
3063 boolean_t
3064 vdev_is_bootable(vdev_t *vd)
3065 {
3066 if (!vd->vdev_ops->vdev_op_leaf) {
3067 char *vdev_type = vd->vdev_ops->vdev_op_type;
3068
3069 if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 &&
3070 vd->vdev_children > 1) {
3071 return (B_FALSE);
3072 } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 ||
3073 strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) {
3074 return (B_FALSE);
3075 }
3076 } else if (vd->vdev_wholedisk == 1) {
3077 return (B_FALSE);
3078 }
3079
3080 for (int c = 0; c < vd->vdev_children; c++) {
3081 if (!vdev_is_bootable(vd->vdev_child[c]))
3082 return (B_FALSE);
3083 }
3084 return (B_TRUE);
3085 }
3086
3087 /*
3088 * Load the state from the original vdev tree (ovd) which
3089 * we've retrieved from the MOS config object. If the original
3090 * vdev was offline or faulted then we transfer that state to the
3091 * device in the current vdev tree (nvd).
3092 */
3093 void
3094 vdev_load_log_state(vdev_t *nvd, vdev_t *ovd)
3095 {
3096 spa_t *spa = nvd->vdev_spa;
3097
3098 ASSERT(nvd->vdev_top->vdev_islog);
3099 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
3100 ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid);
3101
3102 for (int c = 0; c < nvd->vdev_children; c++)
3103 vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]);
3104
3105 if (nvd->vdev_ops->vdev_op_leaf) {
3106 /*
3107 * Restore the persistent vdev state
3108 */
3109 nvd->vdev_offline = ovd->vdev_offline;
3110 nvd->vdev_faulted = ovd->vdev_faulted;
3111 nvd->vdev_degraded = ovd->vdev_degraded;
3112 nvd->vdev_removed = ovd->vdev_removed;
3113 }
3114 }
3115
3116 /*
3117 * Determine if a log device has valid content. If the vdev was
3118 * removed or faulted in the MOS config then we know that
3119 * the content on the log device has already been written to the pool.
3120 */
3121 boolean_t
3122 vdev_log_state_valid(vdev_t *vd)
3123 {
3124 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted &&
3125 !vd->vdev_removed)
3126 return (B_TRUE);
3127
3128 for (int c = 0; c < vd->vdev_children; c++)
3129 if (vdev_log_state_valid(vd->vdev_child[c]))
3130 return (B_TRUE);
3131
3132 return (B_FALSE);
3133 }
3134
3135 /*
3136 * Expand a vdev if possible.
3137 */
3138 void
3139 vdev_expand(vdev_t *vd, uint64_t txg)
3140 {
3141 ASSERT(vd->vdev_top == vd);
3142 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3143
3144 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) {
3145 VERIFY(vdev_metaslab_init(vd, txg) == 0);
3146 vdev_config_dirty(vd);
3147 }
3148 }
3149
3150 /*
3151 * Split a vdev.
3152 */
3153 void
3154 vdev_split(vdev_t *vd)
3155 {
3156 vdev_t *cvd, *pvd = vd->vdev_parent;
3157
3158 vdev_remove_child(pvd, vd);
3159 vdev_compact_children(pvd);
3160
3161 cvd = pvd->vdev_child[0];
3162 if (pvd->vdev_children == 1) {
3163 vdev_remove_parent(cvd);
3164 cvd->vdev_splitting = B_TRUE;
3165 }
3166 vdev_propagate_state(cvd);
3167 }
3168
3169 void
3170 vdev_deadman(vdev_t *vd)
3171 {
3172 for (int c = 0; c < vd->vdev_children; c++) {
3173 vdev_t *cvd = vd->vdev_child[c];
3174
3175 vdev_deadman(cvd);
3176 }
3177
3178 if (vd->vdev_ops->vdev_op_leaf) {
3179 vdev_queue_t *vq = &vd->vdev_queue;
3180
3181 mutex_enter(&vq->vq_lock);
3182 if (avl_numnodes(&vq->vq_pending_tree) > 0) {
3183 spa_t *spa = vd->vdev_spa;
3184 zio_t *fio;
3185 uint64_t delta;
3186
3187 /*
3188 * Look at the head of all the pending queues,
3189 * if any I/O has been outstanding for longer than
3190 * the spa_deadman_synctime we panic the system.
3191 */
3192 fio = avl_first(&vq->vq_pending_tree);
3193 delta = gethrtime() - fio->io_timestamp;
3194 if (delta > spa_deadman_synctime(spa)) {
3195 zfs_dbgmsg("SLOW IO: zio timestamp %lluns, "
3196 "delta %lluns, last io %lluns",
3197 fio->io_timestamp, delta,
3198 vq->vq_io_complete_ts);
3199 fm_panic("I/O to pool '%s' appears to be "
3200 "hung.", spa_name(spa));
3201 }
3202 }
3203 mutex_exit(&vq->vq_lock);
3204 }
3205 }