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