]> git.proxmox.com Git - mirror_zfs.git/blob - module/zfs/vdev.c
Update references to nonexistent man pages in code
[mirror_zfs.git] / module / zfs / vdev.c
1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22 /*
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2011, 2020 by Delphix. All rights reserved.
25 * Copyright 2017 Nexenta Systems, Inc.
26 * Copyright (c) 2014 Integros [integros.com]
27 * Copyright 2016 Toomas Soome <tsoome@me.com>
28 * Copyright 2017 Joyent, Inc.
29 * Copyright (c) 2017, Intel Corporation.
30 * Copyright (c) 2019, Datto Inc. All rights reserved.
31 */
32
33 #include <sys/zfs_context.h>
34 #include <sys/fm/fs/zfs.h>
35 #include <sys/spa.h>
36 #include <sys/spa_impl.h>
37 #include <sys/bpobj.h>
38 #include <sys/dmu.h>
39 #include <sys/dmu_tx.h>
40 #include <sys/dsl_dir.h>
41 #include <sys/vdev_impl.h>
42 #include <sys/vdev_rebuild.h>
43 #include <sys/uberblock_impl.h>
44 #include <sys/metaslab.h>
45 #include <sys/metaslab_impl.h>
46 #include <sys/space_map.h>
47 #include <sys/space_reftree.h>
48 #include <sys/zio.h>
49 #include <sys/zap.h>
50 #include <sys/fs/zfs.h>
51 #include <sys/arc.h>
52 #include <sys/zil.h>
53 #include <sys/dsl_scan.h>
54 #include <sys/abd.h>
55 #include <sys/vdev_initialize.h>
56 #include <sys/vdev_trim.h>
57 #include <sys/zvol.h>
58 #include <sys/zfs_ratelimit.h>
59
60 /* default target for number of metaslabs per top-level vdev */
61 int zfs_vdev_default_ms_count = 200;
62
63 /* minimum number of metaslabs per top-level vdev */
64 int zfs_vdev_min_ms_count = 16;
65
66 /* practical upper limit of total metaslabs per top-level vdev */
67 int zfs_vdev_ms_count_limit = 1ULL << 17;
68
69 /* lower limit for metaslab size (512M) */
70 int zfs_vdev_default_ms_shift = 29;
71
72 /* upper limit for metaslab size (16G) */
73 int zfs_vdev_max_ms_shift = 34;
74
75 int vdev_validate_skip = B_FALSE;
76
77 /*
78 * Since the DTL space map of a vdev is not expected to have a lot of
79 * entries, we default its block size to 4K.
80 */
81 int zfs_vdev_dtl_sm_blksz = (1 << 12);
82
83 /*
84 * Rate limit slow IO (delay) events to this many per second.
85 */
86 unsigned int zfs_slow_io_events_per_second = 20;
87
88 /*
89 * Rate limit checksum events after this many checksum errors per second.
90 */
91 unsigned int zfs_checksum_events_per_second = 20;
92
93 /*
94 * Ignore errors during scrub/resilver. Allows to work around resilver
95 * upon import when there are pool errors.
96 */
97 int zfs_scan_ignore_errors = 0;
98
99 /*
100 * vdev-wide space maps that have lots of entries written to them at
101 * the end of each transaction can benefit from a higher I/O bandwidth
102 * (e.g. vdev_obsolete_sm), thus we default their block size to 128K.
103 */
104 int zfs_vdev_standard_sm_blksz = (1 << 17);
105
106 /*
107 * Tunable parameter for debugging or performance analysis. Setting this
108 * will cause pool corruption on power loss if a volatile out-of-order
109 * write cache is enabled.
110 */
111 int zfs_nocacheflush = 0;
112
113 uint64_t zfs_vdev_max_auto_ashift = ASHIFT_MAX;
114 uint64_t zfs_vdev_min_auto_ashift = ASHIFT_MIN;
115
116 /*PRINTFLIKE2*/
117 void
118 vdev_dbgmsg(vdev_t *vd, const char *fmt, ...)
119 {
120 va_list adx;
121 char buf[256];
122
123 va_start(adx, fmt);
124 (void) vsnprintf(buf, sizeof (buf), fmt, adx);
125 va_end(adx);
126
127 if (vd->vdev_path != NULL) {
128 zfs_dbgmsg("%s vdev '%s': %s", vd->vdev_ops->vdev_op_type,
129 vd->vdev_path, buf);
130 } else {
131 zfs_dbgmsg("%s-%llu vdev (guid %llu): %s",
132 vd->vdev_ops->vdev_op_type,
133 (u_longlong_t)vd->vdev_id,
134 (u_longlong_t)vd->vdev_guid, buf);
135 }
136 }
137
138 void
139 vdev_dbgmsg_print_tree(vdev_t *vd, int indent)
140 {
141 char state[20];
142
143 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) {
144 zfs_dbgmsg("%*svdev %u: %s", indent, "", vd->vdev_id,
145 vd->vdev_ops->vdev_op_type);
146 return;
147 }
148
149 switch (vd->vdev_state) {
150 case VDEV_STATE_UNKNOWN:
151 (void) snprintf(state, sizeof (state), "unknown");
152 break;
153 case VDEV_STATE_CLOSED:
154 (void) snprintf(state, sizeof (state), "closed");
155 break;
156 case VDEV_STATE_OFFLINE:
157 (void) snprintf(state, sizeof (state), "offline");
158 break;
159 case VDEV_STATE_REMOVED:
160 (void) snprintf(state, sizeof (state), "removed");
161 break;
162 case VDEV_STATE_CANT_OPEN:
163 (void) snprintf(state, sizeof (state), "can't open");
164 break;
165 case VDEV_STATE_FAULTED:
166 (void) snprintf(state, sizeof (state), "faulted");
167 break;
168 case VDEV_STATE_DEGRADED:
169 (void) snprintf(state, sizeof (state), "degraded");
170 break;
171 case VDEV_STATE_HEALTHY:
172 (void) snprintf(state, sizeof (state), "healthy");
173 break;
174 default:
175 (void) snprintf(state, sizeof (state), "<state %u>",
176 (uint_t)vd->vdev_state);
177 }
178
179 zfs_dbgmsg("%*svdev %u: %s%s, guid: %llu, path: %s, %s", indent,
180 "", (int)vd->vdev_id, vd->vdev_ops->vdev_op_type,
181 vd->vdev_islog ? " (log)" : "",
182 (u_longlong_t)vd->vdev_guid,
183 vd->vdev_path ? vd->vdev_path : "N/A", state);
184
185 for (uint64_t i = 0; i < vd->vdev_children; i++)
186 vdev_dbgmsg_print_tree(vd->vdev_child[i], indent + 2);
187 }
188
189 /*
190 * Virtual device management.
191 */
192
193 static vdev_ops_t *vdev_ops_table[] = {
194 &vdev_root_ops,
195 &vdev_raidz_ops,
196 &vdev_mirror_ops,
197 &vdev_replacing_ops,
198 &vdev_spare_ops,
199 &vdev_disk_ops,
200 &vdev_file_ops,
201 &vdev_missing_ops,
202 &vdev_hole_ops,
203 &vdev_indirect_ops,
204 NULL
205 };
206
207 /*
208 * Given a vdev type, return the appropriate ops vector.
209 */
210 static vdev_ops_t *
211 vdev_getops(const char *type)
212 {
213 vdev_ops_t *ops, **opspp;
214
215 for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++)
216 if (strcmp(ops->vdev_op_type, type) == 0)
217 break;
218
219 return (ops);
220 }
221
222 /* ARGSUSED */
223 void
224 vdev_default_xlate(vdev_t *vd, const range_seg64_t *in, range_seg64_t *res)
225 {
226 res->rs_start = in->rs_start;
227 res->rs_end = in->rs_end;
228 }
229
230 /*
231 * Derive the enumerated allocation bias from string input.
232 * String origin is either the per-vdev zap or zpool(8).
233 */
234 static vdev_alloc_bias_t
235 vdev_derive_alloc_bias(const char *bias)
236 {
237 vdev_alloc_bias_t alloc_bias = VDEV_BIAS_NONE;
238
239 if (strcmp(bias, VDEV_ALLOC_BIAS_LOG) == 0)
240 alloc_bias = VDEV_BIAS_LOG;
241 else if (strcmp(bias, VDEV_ALLOC_BIAS_SPECIAL) == 0)
242 alloc_bias = VDEV_BIAS_SPECIAL;
243 else if (strcmp(bias, VDEV_ALLOC_BIAS_DEDUP) == 0)
244 alloc_bias = VDEV_BIAS_DEDUP;
245
246 return (alloc_bias);
247 }
248
249 /*
250 * Default asize function: return the MAX of psize with the asize of
251 * all children. This is what's used by anything other than RAID-Z.
252 */
253 uint64_t
254 vdev_default_asize(vdev_t *vd, uint64_t psize)
255 {
256 uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift);
257 uint64_t csize;
258
259 for (int c = 0; c < vd->vdev_children; c++) {
260 csize = vdev_psize_to_asize(vd->vdev_child[c], psize);
261 asize = MAX(asize, csize);
262 }
263
264 return (asize);
265 }
266
267 /*
268 * Get the minimum allocatable size. We define the allocatable size as
269 * the vdev's asize rounded to the nearest metaslab. This allows us to
270 * replace or attach devices which don't have the same physical size but
271 * can still satisfy the same number of allocations.
272 */
273 uint64_t
274 vdev_get_min_asize(vdev_t *vd)
275 {
276 vdev_t *pvd = vd->vdev_parent;
277
278 /*
279 * If our parent is NULL (inactive spare or cache) or is the root,
280 * just return our own asize.
281 */
282 if (pvd == NULL)
283 return (vd->vdev_asize);
284
285 /*
286 * The top-level vdev just returns the allocatable size rounded
287 * to the nearest metaslab.
288 */
289 if (vd == vd->vdev_top)
290 return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift));
291
292 /*
293 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
294 * so each child must provide at least 1/Nth of its asize.
295 */
296 if (pvd->vdev_ops == &vdev_raidz_ops)
297 return ((pvd->vdev_min_asize + pvd->vdev_children - 1) /
298 pvd->vdev_children);
299
300 return (pvd->vdev_min_asize);
301 }
302
303 void
304 vdev_set_min_asize(vdev_t *vd)
305 {
306 vd->vdev_min_asize = vdev_get_min_asize(vd);
307
308 for (int c = 0; c < vd->vdev_children; c++)
309 vdev_set_min_asize(vd->vdev_child[c]);
310 }
311
312 vdev_t *
313 vdev_lookup_top(spa_t *spa, uint64_t vdev)
314 {
315 vdev_t *rvd = spa->spa_root_vdev;
316
317 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
318
319 if (vdev < rvd->vdev_children) {
320 ASSERT(rvd->vdev_child[vdev] != NULL);
321 return (rvd->vdev_child[vdev]);
322 }
323
324 return (NULL);
325 }
326
327 vdev_t *
328 vdev_lookup_by_guid(vdev_t *vd, uint64_t guid)
329 {
330 vdev_t *mvd;
331
332 if (vd->vdev_guid == guid)
333 return (vd);
334
335 for (int c = 0; c < vd->vdev_children; c++)
336 if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) !=
337 NULL)
338 return (mvd);
339
340 return (NULL);
341 }
342
343 static int
344 vdev_count_leaves_impl(vdev_t *vd)
345 {
346 int n = 0;
347
348 if (vd->vdev_ops->vdev_op_leaf)
349 return (1);
350
351 for (int c = 0; c < vd->vdev_children; c++)
352 n += vdev_count_leaves_impl(vd->vdev_child[c]);
353
354 return (n);
355 }
356
357 int
358 vdev_count_leaves(spa_t *spa)
359 {
360 int rc;
361
362 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
363 rc = vdev_count_leaves_impl(spa->spa_root_vdev);
364 spa_config_exit(spa, SCL_VDEV, FTAG);
365
366 return (rc);
367 }
368
369 void
370 vdev_add_child(vdev_t *pvd, vdev_t *cvd)
371 {
372 size_t oldsize, newsize;
373 uint64_t id = cvd->vdev_id;
374 vdev_t **newchild;
375
376 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
377 ASSERT(cvd->vdev_parent == NULL);
378
379 cvd->vdev_parent = pvd;
380
381 if (pvd == NULL)
382 return;
383
384 ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL);
385
386 oldsize = pvd->vdev_children * sizeof (vdev_t *);
387 pvd->vdev_children = MAX(pvd->vdev_children, id + 1);
388 newsize = pvd->vdev_children * sizeof (vdev_t *);
389
390 newchild = kmem_alloc(newsize, KM_SLEEP);
391 if (pvd->vdev_child != NULL) {
392 bcopy(pvd->vdev_child, newchild, oldsize);
393 kmem_free(pvd->vdev_child, oldsize);
394 }
395
396 pvd->vdev_child = newchild;
397 pvd->vdev_child[id] = cvd;
398
399 cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd);
400 ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL);
401
402 /*
403 * Walk up all ancestors to update guid sum.
404 */
405 for (; pvd != NULL; pvd = pvd->vdev_parent)
406 pvd->vdev_guid_sum += cvd->vdev_guid_sum;
407
408 if (cvd->vdev_ops->vdev_op_leaf) {
409 list_insert_head(&cvd->vdev_spa->spa_leaf_list, cvd);
410 cvd->vdev_spa->spa_leaf_list_gen++;
411 }
412 }
413
414 void
415 vdev_remove_child(vdev_t *pvd, vdev_t *cvd)
416 {
417 int c;
418 uint_t id = cvd->vdev_id;
419
420 ASSERT(cvd->vdev_parent == pvd);
421
422 if (pvd == NULL)
423 return;
424
425 ASSERT(id < pvd->vdev_children);
426 ASSERT(pvd->vdev_child[id] == cvd);
427
428 pvd->vdev_child[id] = NULL;
429 cvd->vdev_parent = NULL;
430
431 for (c = 0; c < pvd->vdev_children; c++)
432 if (pvd->vdev_child[c])
433 break;
434
435 if (c == pvd->vdev_children) {
436 kmem_free(pvd->vdev_child, c * sizeof (vdev_t *));
437 pvd->vdev_child = NULL;
438 pvd->vdev_children = 0;
439 }
440
441 if (cvd->vdev_ops->vdev_op_leaf) {
442 spa_t *spa = cvd->vdev_spa;
443 list_remove(&spa->spa_leaf_list, cvd);
444 spa->spa_leaf_list_gen++;
445 }
446
447 /*
448 * Walk up all ancestors to update guid sum.
449 */
450 for (; pvd != NULL; pvd = pvd->vdev_parent)
451 pvd->vdev_guid_sum -= cvd->vdev_guid_sum;
452 }
453
454 /*
455 * Remove any holes in the child array.
456 */
457 void
458 vdev_compact_children(vdev_t *pvd)
459 {
460 vdev_t **newchild, *cvd;
461 int oldc = pvd->vdev_children;
462 int newc;
463
464 ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
465
466 if (oldc == 0)
467 return;
468
469 for (int c = newc = 0; c < oldc; c++)
470 if (pvd->vdev_child[c])
471 newc++;
472
473 if (newc > 0) {
474 newchild = kmem_zalloc(newc * sizeof (vdev_t *), KM_SLEEP);
475
476 for (int c = newc = 0; c < oldc; c++) {
477 if ((cvd = pvd->vdev_child[c]) != NULL) {
478 newchild[newc] = cvd;
479 cvd->vdev_id = newc++;
480 }
481 }
482 } else {
483 newchild = NULL;
484 }
485
486 kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *));
487 pvd->vdev_child = newchild;
488 pvd->vdev_children = newc;
489 }
490
491 /*
492 * Allocate and minimally initialize a vdev_t.
493 */
494 vdev_t *
495 vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops)
496 {
497 vdev_t *vd;
498 vdev_indirect_config_t *vic;
499
500 vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP);
501 vic = &vd->vdev_indirect_config;
502
503 if (spa->spa_root_vdev == NULL) {
504 ASSERT(ops == &vdev_root_ops);
505 spa->spa_root_vdev = vd;
506 spa->spa_load_guid = spa_generate_guid(NULL);
507 }
508
509 if (guid == 0 && ops != &vdev_hole_ops) {
510 if (spa->spa_root_vdev == vd) {
511 /*
512 * The root vdev's guid will also be the pool guid,
513 * which must be unique among all pools.
514 */
515 guid = spa_generate_guid(NULL);
516 } else {
517 /*
518 * Any other vdev's guid must be unique within the pool.
519 */
520 guid = spa_generate_guid(spa);
521 }
522 ASSERT(!spa_guid_exists(spa_guid(spa), guid));
523 }
524
525 vd->vdev_spa = spa;
526 vd->vdev_id = id;
527 vd->vdev_guid = guid;
528 vd->vdev_guid_sum = guid;
529 vd->vdev_ops = ops;
530 vd->vdev_state = VDEV_STATE_CLOSED;
531 vd->vdev_ishole = (ops == &vdev_hole_ops);
532 vic->vic_prev_indirect_vdev = UINT64_MAX;
533
534 rw_init(&vd->vdev_indirect_rwlock, NULL, RW_DEFAULT, NULL);
535 mutex_init(&vd->vdev_obsolete_lock, NULL, MUTEX_DEFAULT, NULL);
536 vd->vdev_obsolete_segments = range_tree_create(NULL, RANGE_SEG64, NULL,
537 0, 0);
538
539 /*
540 * Initialize rate limit structs for events. We rate limit ZIO delay
541 * and checksum events so that we don't overwhelm ZED with thousands
542 * of events when a disk is acting up.
543 */
544 zfs_ratelimit_init(&vd->vdev_delay_rl, &zfs_slow_io_events_per_second,
545 1);
546 zfs_ratelimit_init(&vd->vdev_checksum_rl,
547 &zfs_checksum_events_per_second, 1);
548
549 list_link_init(&vd->vdev_config_dirty_node);
550 list_link_init(&vd->vdev_state_dirty_node);
551 list_link_init(&vd->vdev_initialize_node);
552 list_link_init(&vd->vdev_leaf_node);
553 list_link_init(&vd->vdev_trim_node);
554 mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_NOLOCKDEP, NULL);
555 mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL);
556 mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL);
557 mutex_init(&vd->vdev_scan_io_queue_lock, NULL, MUTEX_DEFAULT, NULL);
558
559 mutex_init(&vd->vdev_initialize_lock, NULL, MUTEX_DEFAULT, NULL);
560 mutex_init(&vd->vdev_initialize_io_lock, NULL, MUTEX_DEFAULT, NULL);
561 cv_init(&vd->vdev_initialize_cv, NULL, CV_DEFAULT, NULL);
562 cv_init(&vd->vdev_initialize_io_cv, NULL, CV_DEFAULT, NULL);
563
564 mutex_init(&vd->vdev_trim_lock, NULL, MUTEX_DEFAULT, NULL);
565 mutex_init(&vd->vdev_autotrim_lock, NULL, MUTEX_DEFAULT, NULL);
566 mutex_init(&vd->vdev_trim_io_lock, NULL, MUTEX_DEFAULT, NULL);
567 cv_init(&vd->vdev_trim_cv, NULL, CV_DEFAULT, NULL);
568 cv_init(&vd->vdev_autotrim_cv, NULL, CV_DEFAULT, NULL);
569 cv_init(&vd->vdev_trim_io_cv, NULL, CV_DEFAULT, NULL);
570
571 mutex_init(&vd->vdev_rebuild_lock, NULL, MUTEX_DEFAULT, NULL);
572 mutex_init(&vd->vdev_rebuild_io_lock, NULL, MUTEX_DEFAULT, NULL);
573 cv_init(&vd->vdev_rebuild_cv, NULL, CV_DEFAULT, NULL);
574 cv_init(&vd->vdev_rebuild_io_cv, NULL, CV_DEFAULT, NULL);
575
576 for (int t = 0; t < DTL_TYPES; t++) {
577 vd->vdev_dtl[t] = range_tree_create(NULL, RANGE_SEG64, NULL, 0,
578 0);
579 }
580
581 txg_list_create(&vd->vdev_ms_list, spa,
582 offsetof(struct metaslab, ms_txg_node));
583 txg_list_create(&vd->vdev_dtl_list, spa,
584 offsetof(struct vdev, vdev_dtl_node));
585 vd->vdev_stat.vs_timestamp = gethrtime();
586 vdev_queue_init(vd);
587 vdev_cache_init(vd);
588
589 return (vd);
590 }
591
592 /*
593 * Allocate a new vdev. The 'alloctype' is used to control whether we are
594 * creating a new vdev or loading an existing one - the behavior is slightly
595 * different for each case.
596 */
597 int
598 vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
599 int alloctype)
600 {
601 vdev_ops_t *ops;
602 char *type;
603 uint64_t guid = 0, islog, nparity;
604 vdev_t *vd;
605 vdev_indirect_config_t *vic;
606 char *tmp = NULL;
607 int rc;
608 vdev_alloc_bias_t alloc_bias = VDEV_BIAS_NONE;
609 boolean_t top_level = (parent && !parent->vdev_parent);
610
611 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
612
613 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
614 return (SET_ERROR(EINVAL));
615
616 if ((ops = vdev_getops(type)) == NULL)
617 return (SET_ERROR(EINVAL));
618
619 /*
620 * If this is a load, get the vdev guid from the nvlist.
621 * Otherwise, vdev_alloc_common() will generate one for us.
622 */
623 if (alloctype == VDEV_ALLOC_LOAD) {
624 uint64_t label_id;
625
626 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) ||
627 label_id != id)
628 return (SET_ERROR(EINVAL));
629
630 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
631 return (SET_ERROR(EINVAL));
632 } else if (alloctype == VDEV_ALLOC_SPARE) {
633 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
634 return (SET_ERROR(EINVAL));
635 } else if (alloctype == VDEV_ALLOC_L2CACHE) {
636 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
637 return (SET_ERROR(EINVAL));
638 } else if (alloctype == VDEV_ALLOC_ROOTPOOL) {
639 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
640 return (SET_ERROR(EINVAL));
641 }
642
643 /*
644 * The first allocated vdev must be of type 'root'.
645 */
646 if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL)
647 return (SET_ERROR(EINVAL));
648
649 /*
650 * Determine whether we're a log vdev.
651 */
652 islog = 0;
653 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog);
654 if (islog && spa_version(spa) < SPA_VERSION_SLOGS)
655 return (SET_ERROR(ENOTSUP));
656
657 if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES)
658 return (SET_ERROR(ENOTSUP));
659
660 /*
661 * Set the nparity property for RAID-Z vdevs.
662 */
663 nparity = -1ULL;
664 if (ops == &vdev_raidz_ops) {
665 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
666 &nparity) == 0) {
667 if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY)
668 return (SET_ERROR(EINVAL));
669 /*
670 * Previous versions could only support 1 or 2 parity
671 * device.
672 */
673 if (nparity > 1 &&
674 spa_version(spa) < SPA_VERSION_RAIDZ2)
675 return (SET_ERROR(ENOTSUP));
676 if (nparity > 2 &&
677 spa_version(spa) < SPA_VERSION_RAIDZ3)
678 return (SET_ERROR(ENOTSUP));
679 } else {
680 /*
681 * We require the parity to be specified for SPAs that
682 * support multiple parity levels.
683 */
684 if (spa_version(spa) >= SPA_VERSION_RAIDZ2)
685 return (SET_ERROR(EINVAL));
686 /*
687 * Otherwise, we default to 1 parity device for RAID-Z.
688 */
689 nparity = 1;
690 }
691 } else {
692 nparity = 0;
693 }
694 ASSERT(nparity != -1ULL);
695
696 /*
697 * If creating a top-level vdev, check for allocation classes input
698 */
699 if (top_level && alloctype == VDEV_ALLOC_ADD) {
700 char *bias;
701
702 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_ALLOCATION_BIAS,
703 &bias) == 0) {
704 alloc_bias = vdev_derive_alloc_bias(bias);
705
706 /* spa_vdev_add() expects feature to be enabled */
707 if (spa->spa_load_state != SPA_LOAD_CREATE &&
708 !spa_feature_is_enabled(spa,
709 SPA_FEATURE_ALLOCATION_CLASSES)) {
710 return (SET_ERROR(ENOTSUP));
711 }
712 }
713 }
714
715 vd = vdev_alloc_common(spa, id, guid, ops);
716 vic = &vd->vdev_indirect_config;
717
718 vd->vdev_islog = islog;
719 vd->vdev_nparity = nparity;
720 if (top_level && alloc_bias != VDEV_BIAS_NONE)
721 vd->vdev_alloc_bias = alloc_bias;
722
723 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0)
724 vd->vdev_path = spa_strdup(vd->vdev_path);
725
726 /*
727 * ZPOOL_CONFIG_AUX_STATE = "external" means we previously forced a
728 * fault on a vdev and want it to persist across imports (like with
729 * zpool offline -f).
730 */
731 rc = nvlist_lookup_string(nv, ZPOOL_CONFIG_AUX_STATE, &tmp);
732 if (rc == 0 && tmp != NULL && strcmp(tmp, "external") == 0) {
733 vd->vdev_stat.vs_aux = VDEV_AUX_EXTERNAL;
734 vd->vdev_faulted = 1;
735 vd->vdev_label_aux = VDEV_AUX_EXTERNAL;
736 }
737
738 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0)
739 vd->vdev_devid = spa_strdup(vd->vdev_devid);
740 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH,
741 &vd->vdev_physpath) == 0)
742 vd->vdev_physpath = spa_strdup(vd->vdev_physpath);
743
744 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH,
745 &vd->vdev_enc_sysfs_path) == 0)
746 vd->vdev_enc_sysfs_path = spa_strdup(vd->vdev_enc_sysfs_path);
747
748 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0)
749 vd->vdev_fru = spa_strdup(vd->vdev_fru);
750
751 /*
752 * Set the whole_disk property. If it's not specified, leave the value
753 * as -1.
754 */
755 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
756 &vd->vdev_wholedisk) != 0)
757 vd->vdev_wholedisk = -1ULL;
758
759 ASSERT0(vic->vic_mapping_object);
760 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_OBJECT,
761 &vic->vic_mapping_object);
762 ASSERT0(vic->vic_births_object);
763 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_BIRTHS,
764 &vic->vic_births_object);
765 ASSERT3U(vic->vic_prev_indirect_vdev, ==, UINT64_MAX);
766 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_PREV_INDIRECT_VDEV,
767 &vic->vic_prev_indirect_vdev);
768
769 /*
770 * Look for the 'not present' flag. This will only be set if the device
771 * was not present at the time of import.
772 */
773 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
774 &vd->vdev_not_present);
775
776 /*
777 * Get the alignment requirement.
778 */
779 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift);
780
781 /*
782 * Retrieve the vdev creation time.
783 */
784 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG,
785 &vd->vdev_crtxg);
786
787 /*
788 * If we're a top-level vdev, try to load the allocation parameters.
789 */
790 if (top_level &&
791 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
792 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
793 &vd->vdev_ms_array);
794 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
795 &vd->vdev_ms_shift);
796 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE,
797 &vd->vdev_asize);
798 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING,
799 &vd->vdev_removing);
800 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP,
801 &vd->vdev_top_zap);
802 } else {
803 ASSERT0(vd->vdev_top_zap);
804 }
805
806 if (top_level && alloctype != VDEV_ALLOC_ATTACH) {
807 ASSERT(alloctype == VDEV_ALLOC_LOAD ||
808 alloctype == VDEV_ALLOC_ADD ||
809 alloctype == VDEV_ALLOC_SPLIT ||
810 alloctype == VDEV_ALLOC_ROOTPOOL);
811 /* Note: metaslab_group_create() is now deferred */
812 }
813
814 if (vd->vdev_ops->vdev_op_leaf &&
815 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
816 (void) nvlist_lookup_uint64(nv,
817 ZPOOL_CONFIG_VDEV_LEAF_ZAP, &vd->vdev_leaf_zap);
818 } else {
819 ASSERT0(vd->vdev_leaf_zap);
820 }
821
822 /*
823 * If we're a leaf vdev, try to load the DTL object and other state.
824 */
825
826 if (vd->vdev_ops->vdev_op_leaf &&
827 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE ||
828 alloctype == VDEV_ALLOC_ROOTPOOL)) {
829 if (alloctype == VDEV_ALLOC_LOAD) {
830 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL,
831 &vd->vdev_dtl_object);
832 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE,
833 &vd->vdev_unspare);
834 }
835
836 if (alloctype == VDEV_ALLOC_ROOTPOOL) {
837 uint64_t spare = 0;
838
839 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
840 &spare) == 0 && spare)
841 spa_spare_add(vd);
842 }
843
844 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE,
845 &vd->vdev_offline);
846
847 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
848 &vd->vdev_resilver_txg);
849
850 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REBUILD_TXG,
851 &vd->vdev_rebuild_txg);
852
853 if (nvlist_exists(nv, ZPOOL_CONFIG_RESILVER_DEFER))
854 vdev_defer_resilver(vd);
855
856 /*
857 * In general, when importing a pool we want to ignore the
858 * persistent fault state, as the diagnosis made on another
859 * system may not be valid in the current context. The only
860 * exception is if we forced a vdev to a persistently faulted
861 * state with 'zpool offline -f'. The persistent fault will
862 * remain across imports until cleared.
863 *
864 * Local vdevs will remain in the faulted state.
865 */
866 if (spa_load_state(spa) == SPA_LOAD_OPEN ||
867 spa_load_state(spa) == SPA_LOAD_IMPORT) {
868 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED,
869 &vd->vdev_faulted);
870 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED,
871 &vd->vdev_degraded);
872 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED,
873 &vd->vdev_removed);
874
875 if (vd->vdev_faulted || vd->vdev_degraded) {
876 char *aux;
877
878 vd->vdev_label_aux =
879 VDEV_AUX_ERR_EXCEEDED;
880 if (nvlist_lookup_string(nv,
881 ZPOOL_CONFIG_AUX_STATE, &aux) == 0 &&
882 strcmp(aux, "external") == 0)
883 vd->vdev_label_aux = VDEV_AUX_EXTERNAL;
884 else
885 vd->vdev_faulted = 0ULL;
886 }
887 }
888 }
889
890 /*
891 * Add ourselves to the parent's list of children.
892 */
893 vdev_add_child(parent, vd);
894
895 *vdp = vd;
896
897 return (0);
898 }
899
900 void
901 vdev_free(vdev_t *vd)
902 {
903 spa_t *spa = vd->vdev_spa;
904
905 ASSERT3P(vd->vdev_initialize_thread, ==, NULL);
906 ASSERT3P(vd->vdev_trim_thread, ==, NULL);
907 ASSERT3P(vd->vdev_autotrim_thread, ==, NULL);
908 ASSERT3P(vd->vdev_rebuild_thread, ==, NULL);
909
910 /*
911 * Scan queues are normally destroyed at the end of a scan. If the
912 * queue exists here, that implies the vdev is being removed while
913 * the scan is still running.
914 */
915 if (vd->vdev_scan_io_queue != NULL) {
916 mutex_enter(&vd->vdev_scan_io_queue_lock);
917 dsl_scan_io_queue_destroy(vd->vdev_scan_io_queue);
918 vd->vdev_scan_io_queue = NULL;
919 mutex_exit(&vd->vdev_scan_io_queue_lock);
920 }
921
922 /*
923 * vdev_free() implies closing the vdev first. This is simpler than
924 * trying to ensure complicated semantics for all callers.
925 */
926 vdev_close(vd);
927
928 ASSERT(!list_link_active(&vd->vdev_config_dirty_node));
929 ASSERT(!list_link_active(&vd->vdev_state_dirty_node));
930
931 /*
932 * Free all children.
933 */
934 for (int c = 0; c < vd->vdev_children; c++)
935 vdev_free(vd->vdev_child[c]);
936
937 ASSERT(vd->vdev_child == NULL);
938 ASSERT(vd->vdev_guid_sum == vd->vdev_guid);
939
940 /*
941 * Discard allocation state.
942 */
943 if (vd->vdev_mg != NULL) {
944 vdev_metaslab_fini(vd);
945 metaslab_group_destroy(vd->vdev_mg);
946 vd->vdev_mg = NULL;
947 }
948
949 ASSERT0(vd->vdev_stat.vs_space);
950 ASSERT0(vd->vdev_stat.vs_dspace);
951 ASSERT0(vd->vdev_stat.vs_alloc);
952
953 /*
954 * Remove this vdev from its parent's child list.
955 */
956 vdev_remove_child(vd->vdev_parent, vd);
957
958 ASSERT(vd->vdev_parent == NULL);
959 ASSERT(!list_link_active(&vd->vdev_leaf_node));
960
961 /*
962 * Clean up vdev structure.
963 */
964 vdev_queue_fini(vd);
965 vdev_cache_fini(vd);
966
967 if (vd->vdev_path)
968 spa_strfree(vd->vdev_path);
969 if (vd->vdev_devid)
970 spa_strfree(vd->vdev_devid);
971 if (vd->vdev_physpath)
972 spa_strfree(vd->vdev_physpath);
973
974 if (vd->vdev_enc_sysfs_path)
975 spa_strfree(vd->vdev_enc_sysfs_path);
976
977 if (vd->vdev_fru)
978 spa_strfree(vd->vdev_fru);
979
980 if (vd->vdev_isspare)
981 spa_spare_remove(vd);
982 if (vd->vdev_isl2cache)
983 spa_l2cache_remove(vd);
984
985 txg_list_destroy(&vd->vdev_ms_list);
986 txg_list_destroy(&vd->vdev_dtl_list);
987
988 mutex_enter(&vd->vdev_dtl_lock);
989 space_map_close(vd->vdev_dtl_sm);
990 for (int t = 0; t < DTL_TYPES; t++) {
991 range_tree_vacate(vd->vdev_dtl[t], NULL, NULL);
992 range_tree_destroy(vd->vdev_dtl[t]);
993 }
994 mutex_exit(&vd->vdev_dtl_lock);
995
996 EQUIV(vd->vdev_indirect_births != NULL,
997 vd->vdev_indirect_mapping != NULL);
998 if (vd->vdev_indirect_births != NULL) {
999 vdev_indirect_mapping_close(vd->vdev_indirect_mapping);
1000 vdev_indirect_births_close(vd->vdev_indirect_births);
1001 }
1002
1003 if (vd->vdev_obsolete_sm != NULL) {
1004 ASSERT(vd->vdev_removing ||
1005 vd->vdev_ops == &vdev_indirect_ops);
1006 space_map_close(vd->vdev_obsolete_sm);
1007 vd->vdev_obsolete_sm = NULL;
1008 }
1009 range_tree_destroy(vd->vdev_obsolete_segments);
1010 rw_destroy(&vd->vdev_indirect_rwlock);
1011 mutex_destroy(&vd->vdev_obsolete_lock);
1012
1013 mutex_destroy(&vd->vdev_dtl_lock);
1014 mutex_destroy(&vd->vdev_stat_lock);
1015 mutex_destroy(&vd->vdev_probe_lock);
1016 mutex_destroy(&vd->vdev_scan_io_queue_lock);
1017
1018 mutex_destroy(&vd->vdev_initialize_lock);
1019 mutex_destroy(&vd->vdev_initialize_io_lock);
1020 cv_destroy(&vd->vdev_initialize_io_cv);
1021 cv_destroy(&vd->vdev_initialize_cv);
1022
1023 mutex_destroy(&vd->vdev_trim_lock);
1024 mutex_destroy(&vd->vdev_autotrim_lock);
1025 mutex_destroy(&vd->vdev_trim_io_lock);
1026 cv_destroy(&vd->vdev_trim_cv);
1027 cv_destroy(&vd->vdev_autotrim_cv);
1028 cv_destroy(&vd->vdev_trim_io_cv);
1029
1030 mutex_destroy(&vd->vdev_rebuild_lock);
1031 mutex_destroy(&vd->vdev_rebuild_io_lock);
1032 cv_destroy(&vd->vdev_rebuild_cv);
1033 cv_destroy(&vd->vdev_rebuild_io_cv);
1034
1035 zfs_ratelimit_fini(&vd->vdev_delay_rl);
1036 zfs_ratelimit_fini(&vd->vdev_checksum_rl);
1037
1038 if (vd == spa->spa_root_vdev)
1039 spa->spa_root_vdev = NULL;
1040
1041 kmem_free(vd, sizeof (vdev_t));
1042 }
1043
1044 /*
1045 * Transfer top-level vdev state from svd to tvd.
1046 */
1047 static void
1048 vdev_top_transfer(vdev_t *svd, vdev_t *tvd)
1049 {
1050 spa_t *spa = svd->vdev_spa;
1051 metaslab_t *msp;
1052 vdev_t *vd;
1053 int t;
1054
1055 ASSERT(tvd == tvd->vdev_top);
1056
1057 tvd->vdev_pending_fastwrite = svd->vdev_pending_fastwrite;
1058 tvd->vdev_ms_array = svd->vdev_ms_array;
1059 tvd->vdev_ms_shift = svd->vdev_ms_shift;
1060 tvd->vdev_ms_count = svd->vdev_ms_count;
1061 tvd->vdev_top_zap = svd->vdev_top_zap;
1062
1063 svd->vdev_ms_array = 0;
1064 svd->vdev_ms_shift = 0;
1065 svd->vdev_ms_count = 0;
1066 svd->vdev_top_zap = 0;
1067
1068 if (tvd->vdev_mg)
1069 ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg);
1070 tvd->vdev_mg = svd->vdev_mg;
1071 tvd->vdev_ms = svd->vdev_ms;
1072
1073 svd->vdev_mg = NULL;
1074 svd->vdev_ms = NULL;
1075
1076 if (tvd->vdev_mg != NULL)
1077 tvd->vdev_mg->mg_vd = tvd;
1078
1079 tvd->vdev_checkpoint_sm = svd->vdev_checkpoint_sm;
1080 svd->vdev_checkpoint_sm = NULL;
1081
1082 tvd->vdev_alloc_bias = svd->vdev_alloc_bias;
1083 svd->vdev_alloc_bias = VDEV_BIAS_NONE;
1084
1085 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc;
1086 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space;
1087 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace;
1088
1089 svd->vdev_stat.vs_alloc = 0;
1090 svd->vdev_stat.vs_space = 0;
1091 svd->vdev_stat.vs_dspace = 0;
1092
1093 /*
1094 * State which may be set on a top-level vdev that's in the
1095 * process of being removed.
1096 */
1097 ASSERT0(tvd->vdev_indirect_config.vic_births_object);
1098 ASSERT0(tvd->vdev_indirect_config.vic_mapping_object);
1099 ASSERT3U(tvd->vdev_indirect_config.vic_prev_indirect_vdev, ==, -1ULL);
1100 ASSERT3P(tvd->vdev_indirect_mapping, ==, NULL);
1101 ASSERT3P(tvd->vdev_indirect_births, ==, NULL);
1102 ASSERT3P(tvd->vdev_obsolete_sm, ==, NULL);
1103 ASSERT0(tvd->vdev_removing);
1104 ASSERT0(tvd->vdev_rebuilding);
1105 tvd->vdev_removing = svd->vdev_removing;
1106 tvd->vdev_rebuilding = svd->vdev_rebuilding;
1107 tvd->vdev_rebuild_config = svd->vdev_rebuild_config;
1108 tvd->vdev_indirect_config = svd->vdev_indirect_config;
1109 tvd->vdev_indirect_mapping = svd->vdev_indirect_mapping;
1110 tvd->vdev_indirect_births = svd->vdev_indirect_births;
1111 range_tree_swap(&svd->vdev_obsolete_segments,
1112 &tvd->vdev_obsolete_segments);
1113 tvd->vdev_obsolete_sm = svd->vdev_obsolete_sm;
1114 svd->vdev_indirect_config.vic_mapping_object = 0;
1115 svd->vdev_indirect_config.vic_births_object = 0;
1116 svd->vdev_indirect_config.vic_prev_indirect_vdev = -1ULL;
1117 svd->vdev_indirect_mapping = NULL;
1118 svd->vdev_indirect_births = NULL;
1119 svd->vdev_obsolete_sm = NULL;
1120 svd->vdev_removing = 0;
1121 svd->vdev_rebuilding = 0;
1122
1123 for (t = 0; t < TXG_SIZE; t++) {
1124 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL)
1125 (void) txg_list_add(&tvd->vdev_ms_list, msp, t);
1126 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL)
1127 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t);
1128 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t))
1129 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t);
1130 }
1131
1132 if (list_link_active(&svd->vdev_config_dirty_node)) {
1133 vdev_config_clean(svd);
1134 vdev_config_dirty(tvd);
1135 }
1136
1137 if (list_link_active(&svd->vdev_state_dirty_node)) {
1138 vdev_state_clean(svd);
1139 vdev_state_dirty(tvd);
1140 }
1141
1142 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio;
1143 svd->vdev_deflate_ratio = 0;
1144
1145 tvd->vdev_islog = svd->vdev_islog;
1146 svd->vdev_islog = 0;
1147
1148 dsl_scan_io_queue_vdev_xfer(svd, tvd);
1149 }
1150
1151 static void
1152 vdev_top_update(vdev_t *tvd, vdev_t *vd)
1153 {
1154 if (vd == NULL)
1155 return;
1156
1157 vd->vdev_top = tvd;
1158
1159 for (int c = 0; c < vd->vdev_children; c++)
1160 vdev_top_update(tvd, vd->vdev_child[c]);
1161 }
1162
1163 /*
1164 * Add a mirror/replacing vdev above an existing vdev.
1165 */
1166 vdev_t *
1167 vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops)
1168 {
1169 spa_t *spa = cvd->vdev_spa;
1170 vdev_t *pvd = cvd->vdev_parent;
1171 vdev_t *mvd;
1172
1173 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1174
1175 mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops);
1176
1177 mvd->vdev_asize = cvd->vdev_asize;
1178 mvd->vdev_min_asize = cvd->vdev_min_asize;
1179 mvd->vdev_max_asize = cvd->vdev_max_asize;
1180 mvd->vdev_psize = cvd->vdev_psize;
1181 mvd->vdev_ashift = cvd->vdev_ashift;
1182 mvd->vdev_logical_ashift = cvd->vdev_logical_ashift;
1183 mvd->vdev_physical_ashift = cvd->vdev_physical_ashift;
1184 mvd->vdev_state = cvd->vdev_state;
1185 mvd->vdev_crtxg = cvd->vdev_crtxg;
1186
1187 vdev_remove_child(pvd, cvd);
1188 vdev_add_child(pvd, mvd);
1189 cvd->vdev_id = mvd->vdev_children;
1190 vdev_add_child(mvd, cvd);
1191 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
1192
1193 if (mvd == mvd->vdev_top)
1194 vdev_top_transfer(cvd, mvd);
1195
1196 return (mvd);
1197 }
1198
1199 /*
1200 * Remove a 1-way mirror/replacing vdev from the tree.
1201 */
1202 void
1203 vdev_remove_parent(vdev_t *cvd)
1204 {
1205 vdev_t *mvd = cvd->vdev_parent;
1206 vdev_t *pvd = mvd->vdev_parent;
1207
1208 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1209
1210 ASSERT(mvd->vdev_children == 1);
1211 ASSERT(mvd->vdev_ops == &vdev_mirror_ops ||
1212 mvd->vdev_ops == &vdev_replacing_ops ||
1213 mvd->vdev_ops == &vdev_spare_ops);
1214 cvd->vdev_ashift = mvd->vdev_ashift;
1215 cvd->vdev_logical_ashift = mvd->vdev_logical_ashift;
1216 cvd->vdev_physical_ashift = mvd->vdev_physical_ashift;
1217 vdev_remove_child(mvd, cvd);
1218 vdev_remove_child(pvd, mvd);
1219
1220 /*
1221 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
1222 * Otherwise, we could have detached an offline device, and when we
1223 * go to import the pool we'll think we have two top-level vdevs,
1224 * instead of a different version of the same top-level vdev.
1225 */
1226 if (mvd->vdev_top == mvd) {
1227 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid;
1228 cvd->vdev_orig_guid = cvd->vdev_guid;
1229 cvd->vdev_guid += guid_delta;
1230 cvd->vdev_guid_sum += guid_delta;
1231
1232 /*
1233 * If pool not set for autoexpand, we need to also preserve
1234 * mvd's asize to prevent automatic expansion of cvd.
1235 * Otherwise if we are adjusting the mirror by attaching and
1236 * detaching children of non-uniform sizes, the mirror could
1237 * autoexpand, unexpectedly requiring larger devices to
1238 * re-establish the mirror.
1239 */
1240 if (!cvd->vdev_spa->spa_autoexpand)
1241 cvd->vdev_asize = mvd->vdev_asize;
1242 }
1243 cvd->vdev_id = mvd->vdev_id;
1244 vdev_add_child(pvd, cvd);
1245 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
1246
1247 if (cvd == cvd->vdev_top)
1248 vdev_top_transfer(mvd, cvd);
1249
1250 ASSERT(mvd->vdev_children == 0);
1251 vdev_free(mvd);
1252 }
1253
1254 static void
1255 vdev_metaslab_group_create(vdev_t *vd)
1256 {
1257 spa_t *spa = vd->vdev_spa;
1258
1259 /*
1260 * metaslab_group_create was delayed until allocation bias was available
1261 */
1262 if (vd->vdev_mg == NULL) {
1263 metaslab_class_t *mc;
1264
1265 if (vd->vdev_islog && vd->vdev_alloc_bias == VDEV_BIAS_NONE)
1266 vd->vdev_alloc_bias = VDEV_BIAS_LOG;
1267
1268 ASSERT3U(vd->vdev_islog, ==,
1269 (vd->vdev_alloc_bias == VDEV_BIAS_LOG));
1270
1271 switch (vd->vdev_alloc_bias) {
1272 case VDEV_BIAS_LOG:
1273 mc = spa_log_class(spa);
1274 break;
1275 case VDEV_BIAS_SPECIAL:
1276 mc = spa_special_class(spa);
1277 break;
1278 case VDEV_BIAS_DEDUP:
1279 mc = spa_dedup_class(spa);
1280 break;
1281 default:
1282 mc = spa_normal_class(spa);
1283 }
1284
1285 vd->vdev_mg = metaslab_group_create(mc, vd,
1286 spa->spa_alloc_count);
1287
1288 /*
1289 * The spa ashift min/max only apply for the normal metaslab
1290 * class. Class destination is late binding so ashift boundry
1291 * setting had to wait until now.
1292 */
1293 if (vd->vdev_top == vd && vd->vdev_ashift != 0 &&
1294 mc == spa_normal_class(spa) && vd->vdev_aux == NULL) {
1295 if (vd->vdev_ashift > spa->spa_max_ashift)
1296 spa->spa_max_ashift = vd->vdev_ashift;
1297 if (vd->vdev_ashift < spa->spa_min_ashift)
1298 spa->spa_min_ashift = vd->vdev_ashift;
1299 }
1300 }
1301 }
1302
1303 int
1304 vdev_metaslab_init(vdev_t *vd, uint64_t txg)
1305 {
1306 spa_t *spa = vd->vdev_spa;
1307 objset_t *mos = spa->spa_meta_objset;
1308 uint64_t m;
1309 uint64_t oldc = vd->vdev_ms_count;
1310 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift;
1311 metaslab_t **mspp;
1312 int error;
1313 boolean_t expanding = (oldc != 0);
1314
1315 ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1316
1317 /*
1318 * This vdev is not being allocated from yet or is a hole.
1319 */
1320 if (vd->vdev_ms_shift == 0)
1321 return (0);
1322
1323 ASSERT(!vd->vdev_ishole);
1324
1325 ASSERT(oldc <= newc);
1326
1327 mspp = vmem_zalloc(newc * sizeof (*mspp), KM_SLEEP);
1328
1329 if (expanding) {
1330 bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp));
1331 vmem_free(vd->vdev_ms, oldc * sizeof (*mspp));
1332 }
1333
1334 vd->vdev_ms = mspp;
1335 vd->vdev_ms_count = newc;
1336 for (m = oldc; m < newc; m++) {
1337 uint64_t object = 0;
1338
1339 /*
1340 * vdev_ms_array may be 0 if we are creating the "fake"
1341 * metaslabs for an indirect vdev for zdb's leak detection.
1342 * See zdb_leak_init().
1343 */
1344 if (txg == 0 && vd->vdev_ms_array != 0) {
1345 error = dmu_read(mos, vd->vdev_ms_array,
1346 m * sizeof (uint64_t), sizeof (uint64_t), &object,
1347 DMU_READ_PREFETCH);
1348 if (error != 0) {
1349 vdev_dbgmsg(vd, "unable to read the metaslab "
1350 "array [error=%d]", error);
1351 return (error);
1352 }
1353 }
1354
1355 #ifndef _KERNEL
1356 /*
1357 * To accommodate zdb_leak_init() fake indirect
1358 * metaslabs, we allocate a metaslab group for
1359 * indirect vdevs which normally don't have one.
1360 */
1361 if (vd->vdev_mg == NULL) {
1362 ASSERT0(vdev_is_concrete(vd));
1363 vdev_metaslab_group_create(vd);
1364 }
1365 #endif
1366 error = metaslab_init(vd->vdev_mg, m, object, txg,
1367 &(vd->vdev_ms[m]));
1368 if (error != 0) {
1369 vdev_dbgmsg(vd, "metaslab_init failed [error=%d]",
1370 error);
1371 return (error);
1372 }
1373 }
1374
1375 if (txg == 0)
1376 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER);
1377
1378 /*
1379 * If the vdev is being removed we don't activate
1380 * the metaslabs since we want to ensure that no new
1381 * allocations are performed on this device.
1382 */
1383 if (!expanding && !vd->vdev_removing) {
1384 metaslab_group_activate(vd->vdev_mg);
1385 }
1386
1387 if (txg == 0)
1388 spa_config_exit(spa, SCL_ALLOC, FTAG);
1389
1390 /*
1391 * Regardless whether this vdev was just added or it is being
1392 * expanded, the metaslab count has changed. Recalculate the
1393 * block limit.
1394 */
1395 spa_log_sm_set_blocklimit(spa);
1396
1397 return (0);
1398 }
1399
1400 void
1401 vdev_metaslab_fini(vdev_t *vd)
1402 {
1403 if (vd->vdev_checkpoint_sm != NULL) {
1404 ASSERT(spa_feature_is_active(vd->vdev_spa,
1405 SPA_FEATURE_POOL_CHECKPOINT));
1406 space_map_close(vd->vdev_checkpoint_sm);
1407 /*
1408 * Even though we close the space map, we need to set its
1409 * pointer to NULL. The reason is that vdev_metaslab_fini()
1410 * may be called multiple times for certain operations
1411 * (i.e. when destroying a pool) so we need to ensure that
1412 * this clause never executes twice. This logic is similar
1413 * to the one used for the vdev_ms clause below.
1414 */
1415 vd->vdev_checkpoint_sm = NULL;
1416 }
1417
1418 if (vd->vdev_ms != NULL) {
1419 metaslab_group_t *mg = vd->vdev_mg;
1420 metaslab_group_passivate(mg);
1421
1422 uint64_t count = vd->vdev_ms_count;
1423 for (uint64_t m = 0; m < count; m++) {
1424 metaslab_t *msp = vd->vdev_ms[m];
1425 if (msp != NULL)
1426 metaslab_fini(msp);
1427 }
1428 vmem_free(vd->vdev_ms, count * sizeof (metaslab_t *));
1429 vd->vdev_ms = NULL;
1430
1431 vd->vdev_ms_count = 0;
1432
1433 for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++)
1434 ASSERT0(mg->mg_histogram[i]);
1435 }
1436 ASSERT0(vd->vdev_ms_count);
1437 ASSERT3U(vd->vdev_pending_fastwrite, ==, 0);
1438 }
1439
1440 typedef struct vdev_probe_stats {
1441 boolean_t vps_readable;
1442 boolean_t vps_writeable;
1443 int vps_flags;
1444 } vdev_probe_stats_t;
1445
1446 static void
1447 vdev_probe_done(zio_t *zio)
1448 {
1449 spa_t *spa = zio->io_spa;
1450 vdev_t *vd = zio->io_vd;
1451 vdev_probe_stats_t *vps = zio->io_private;
1452
1453 ASSERT(vd->vdev_probe_zio != NULL);
1454
1455 if (zio->io_type == ZIO_TYPE_READ) {
1456 if (zio->io_error == 0)
1457 vps->vps_readable = 1;
1458 if (zio->io_error == 0 && spa_writeable(spa)) {
1459 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd,
1460 zio->io_offset, zio->io_size, zio->io_abd,
1461 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1462 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE));
1463 } else {
1464 abd_free(zio->io_abd);
1465 }
1466 } else if (zio->io_type == ZIO_TYPE_WRITE) {
1467 if (zio->io_error == 0)
1468 vps->vps_writeable = 1;
1469 abd_free(zio->io_abd);
1470 } else if (zio->io_type == ZIO_TYPE_NULL) {
1471 zio_t *pio;
1472 zio_link_t *zl;
1473
1474 vd->vdev_cant_read |= !vps->vps_readable;
1475 vd->vdev_cant_write |= !vps->vps_writeable;
1476
1477 if (vdev_readable(vd) &&
1478 (vdev_writeable(vd) || !spa_writeable(spa))) {
1479 zio->io_error = 0;
1480 } else {
1481 ASSERT(zio->io_error != 0);
1482 vdev_dbgmsg(vd, "failed probe");
1483 (void) zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE,
1484 spa, vd, NULL, NULL, 0);
1485 zio->io_error = SET_ERROR(ENXIO);
1486 }
1487
1488 mutex_enter(&vd->vdev_probe_lock);
1489 ASSERT(vd->vdev_probe_zio == zio);
1490 vd->vdev_probe_zio = NULL;
1491 mutex_exit(&vd->vdev_probe_lock);
1492
1493 zl = NULL;
1494 while ((pio = zio_walk_parents(zio, &zl)) != NULL)
1495 if (!vdev_accessible(vd, pio))
1496 pio->io_error = SET_ERROR(ENXIO);
1497
1498 kmem_free(vps, sizeof (*vps));
1499 }
1500 }
1501
1502 /*
1503 * Determine whether this device is accessible.
1504 *
1505 * Read and write to several known locations: the pad regions of each
1506 * vdev label but the first, which we leave alone in case it contains
1507 * a VTOC.
1508 */
1509 zio_t *
1510 vdev_probe(vdev_t *vd, zio_t *zio)
1511 {
1512 spa_t *spa = vd->vdev_spa;
1513 vdev_probe_stats_t *vps = NULL;
1514 zio_t *pio;
1515
1516 ASSERT(vd->vdev_ops->vdev_op_leaf);
1517
1518 /*
1519 * Don't probe the probe.
1520 */
1521 if (zio && (zio->io_flags & ZIO_FLAG_PROBE))
1522 return (NULL);
1523
1524 /*
1525 * To prevent 'probe storms' when a device fails, we create
1526 * just one probe i/o at a time. All zios that want to probe
1527 * this vdev will become parents of the probe io.
1528 */
1529 mutex_enter(&vd->vdev_probe_lock);
1530
1531 if ((pio = vd->vdev_probe_zio) == NULL) {
1532 vps = kmem_zalloc(sizeof (*vps), KM_SLEEP);
1533
1534 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE |
1535 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE |
1536 ZIO_FLAG_TRYHARD;
1537
1538 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) {
1539 /*
1540 * vdev_cant_read and vdev_cant_write can only
1541 * transition from TRUE to FALSE when we have the
1542 * SCL_ZIO lock as writer; otherwise they can only
1543 * transition from FALSE to TRUE. This ensures that
1544 * any zio looking at these values can assume that
1545 * failures persist for the life of the I/O. That's
1546 * important because when a device has intermittent
1547 * connectivity problems, we want to ensure that
1548 * they're ascribed to the device (ENXIO) and not
1549 * the zio (EIO).
1550 *
1551 * Since we hold SCL_ZIO as writer here, clear both
1552 * values so the probe can reevaluate from first
1553 * principles.
1554 */
1555 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER;
1556 vd->vdev_cant_read = B_FALSE;
1557 vd->vdev_cant_write = B_FALSE;
1558 }
1559
1560 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd,
1561 vdev_probe_done, vps,
1562 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE);
1563
1564 /*
1565 * We can't change the vdev state in this context, so we
1566 * kick off an async task to do it on our behalf.
1567 */
1568 if (zio != NULL) {
1569 vd->vdev_probe_wanted = B_TRUE;
1570 spa_async_request(spa, SPA_ASYNC_PROBE);
1571 }
1572 }
1573
1574 if (zio != NULL)
1575 zio_add_child(zio, pio);
1576
1577 mutex_exit(&vd->vdev_probe_lock);
1578
1579 if (vps == NULL) {
1580 ASSERT(zio != NULL);
1581 return (NULL);
1582 }
1583
1584 for (int l = 1; l < VDEV_LABELS; l++) {
1585 zio_nowait(zio_read_phys(pio, vd,
1586 vdev_label_offset(vd->vdev_psize, l,
1587 offsetof(vdev_label_t, vl_be)), VDEV_PAD_SIZE,
1588 abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE),
1589 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1590 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE));
1591 }
1592
1593 if (zio == NULL)
1594 return (pio);
1595
1596 zio_nowait(pio);
1597 return (NULL);
1598 }
1599
1600 static void
1601 vdev_open_child(void *arg)
1602 {
1603 vdev_t *vd = arg;
1604
1605 vd->vdev_open_thread = curthread;
1606 vd->vdev_open_error = vdev_open(vd);
1607 vd->vdev_open_thread = NULL;
1608 }
1609
1610 static boolean_t
1611 vdev_uses_zvols(vdev_t *vd)
1612 {
1613 #ifdef _KERNEL
1614 if (zvol_is_zvol(vd->vdev_path))
1615 return (B_TRUE);
1616 #endif
1617
1618 for (int c = 0; c < vd->vdev_children; c++)
1619 if (vdev_uses_zvols(vd->vdev_child[c]))
1620 return (B_TRUE);
1621
1622 return (B_FALSE);
1623 }
1624
1625 void
1626 vdev_open_children(vdev_t *vd)
1627 {
1628 taskq_t *tq;
1629 int children = vd->vdev_children;
1630
1631 /*
1632 * in order to handle pools on top of zvols, do the opens
1633 * in a single thread so that the same thread holds the
1634 * spa_namespace_lock
1635 */
1636 if (vdev_uses_zvols(vd)) {
1637 retry_sync:
1638 for (int c = 0; c < children; c++)
1639 vd->vdev_child[c]->vdev_open_error =
1640 vdev_open(vd->vdev_child[c]);
1641 } else {
1642 tq = taskq_create("vdev_open", children, minclsyspri,
1643 children, children, TASKQ_PREPOPULATE);
1644 if (tq == NULL)
1645 goto retry_sync;
1646
1647 for (int c = 0; c < children; c++)
1648 VERIFY(taskq_dispatch(tq, vdev_open_child,
1649 vd->vdev_child[c], TQ_SLEEP) != TASKQID_INVALID);
1650
1651 taskq_destroy(tq);
1652 }
1653
1654 vd->vdev_nonrot = B_TRUE;
1655
1656 for (int c = 0; c < children; c++)
1657 vd->vdev_nonrot &= vd->vdev_child[c]->vdev_nonrot;
1658 }
1659
1660 /*
1661 * Compute the raidz-deflation ratio. Note, we hard-code
1662 * in 128k (1 << 17) because it is the "typical" blocksize.
1663 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change,
1664 * otherwise it would inconsistently account for existing bp's.
1665 */
1666 static void
1667 vdev_set_deflate_ratio(vdev_t *vd)
1668 {
1669 if (vd == vd->vdev_top && !vd->vdev_ishole && vd->vdev_ashift != 0) {
1670 vd->vdev_deflate_ratio = (1 << 17) /
1671 (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT);
1672 }
1673 }
1674
1675 /*
1676 * Maximize performance by inflating the configured ashift for top level
1677 * vdevs to be as close to the physical ashift as possible while maintaining
1678 * administrator defined limits and ensuring it doesn't go below the
1679 * logical ashift.
1680 */
1681 static void
1682 vdev_ashift_optimize(vdev_t *vd)
1683 {
1684 ASSERT(vd == vd->vdev_top);
1685
1686 if (vd->vdev_ashift < vd->vdev_physical_ashift) {
1687 vd->vdev_ashift = MIN(
1688 MAX(zfs_vdev_max_auto_ashift, vd->vdev_ashift),
1689 MAX(zfs_vdev_min_auto_ashift,
1690 vd->vdev_physical_ashift));
1691 } else {
1692 /*
1693 * If the logical and physical ashifts are the same, then
1694 * we ensure that the top-level vdev's ashift is not smaller
1695 * than our minimum ashift value. For the unusual case
1696 * where logical ashift > physical ashift, we can't cap
1697 * the calculated ashift based on max ashift as that
1698 * would cause failures.
1699 * We still check if we need to increase it to match
1700 * the min ashift.
1701 */
1702 vd->vdev_ashift = MAX(zfs_vdev_min_auto_ashift,
1703 vd->vdev_ashift);
1704 }
1705 }
1706
1707 /*
1708 * Prepare a virtual device for access.
1709 */
1710 int
1711 vdev_open(vdev_t *vd)
1712 {
1713 spa_t *spa = vd->vdev_spa;
1714 int error;
1715 uint64_t osize = 0;
1716 uint64_t max_osize = 0;
1717 uint64_t asize, max_asize, psize;
1718 uint64_t logical_ashift = 0;
1719 uint64_t physical_ashift = 0;
1720
1721 ASSERT(vd->vdev_open_thread == curthread ||
1722 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1723 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED ||
1724 vd->vdev_state == VDEV_STATE_CANT_OPEN ||
1725 vd->vdev_state == VDEV_STATE_OFFLINE);
1726
1727 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1728 vd->vdev_cant_read = B_FALSE;
1729 vd->vdev_cant_write = B_FALSE;
1730 vd->vdev_min_asize = vdev_get_min_asize(vd);
1731
1732 /*
1733 * If this vdev is not removed, check its fault status. If it's
1734 * faulted, bail out of the open.
1735 */
1736 if (!vd->vdev_removed && vd->vdev_faulted) {
1737 ASSERT(vd->vdev_children == 0);
1738 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1739 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1740 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1741 vd->vdev_label_aux);
1742 return (SET_ERROR(ENXIO));
1743 } else if (vd->vdev_offline) {
1744 ASSERT(vd->vdev_children == 0);
1745 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE);
1746 return (SET_ERROR(ENXIO));
1747 }
1748
1749 error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize,
1750 &logical_ashift, &physical_ashift);
1751 /*
1752 * Physical volume size should never be larger than its max size, unless
1753 * the disk has shrunk while we were reading it or the device is buggy
1754 * or damaged: either way it's not safe for use, bail out of the open.
1755 */
1756 if (osize > max_osize) {
1757 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1758 VDEV_AUX_OPEN_FAILED);
1759 return (SET_ERROR(ENXIO));
1760 }
1761
1762 /*
1763 * Reset the vdev_reopening flag so that we actually close
1764 * the vdev on error.
1765 */
1766 vd->vdev_reopening = B_FALSE;
1767 if (zio_injection_enabled && error == 0)
1768 error = zio_handle_device_injection(vd, NULL, SET_ERROR(ENXIO));
1769
1770 if (error) {
1771 if (vd->vdev_removed &&
1772 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED)
1773 vd->vdev_removed = B_FALSE;
1774
1775 if (vd->vdev_stat.vs_aux == VDEV_AUX_CHILDREN_OFFLINE) {
1776 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE,
1777 vd->vdev_stat.vs_aux);
1778 } else {
1779 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1780 vd->vdev_stat.vs_aux);
1781 }
1782 return (error);
1783 }
1784
1785 vd->vdev_removed = B_FALSE;
1786
1787 /*
1788 * Recheck the faulted flag now that we have confirmed that
1789 * the vdev is accessible. If we're faulted, bail.
1790 */
1791 if (vd->vdev_faulted) {
1792 ASSERT(vd->vdev_children == 0);
1793 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1794 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1795 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1796 vd->vdev_label_aux);
1797 return (SET_ERROR(ENXIO));
1798 }
1799
1800 if (vd->vdev_degraded) {
1801 ASSERT(vd->vdev_children == 0);
1802 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1803 VDEV_AUX_ERR_EXCEEDED);
1804 } else {
1805 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0);
1806 }
1807
1808 /*
1809 * For hole or missing vdevs we just return success.
1810 */
1811 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops)
1812 return (0);
1813
1814 for (int c = 0; c < vd->vdev_children; c++) {
1815 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) {
1816 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1817 VDEV_AUX_NONE);
1818 break;
1819 }
1820 }
1821
1822 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t));
1823 max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t));
1824
1825 if (vd->vdev_children == 0) {
1826 if (osize < SPA_MINDEVSIZE) {
1827 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1828 VDEV_AUX_TOO_SMALL);
1829 return (SET_ERROR(EOVERFLOW));
1830 }
1831 psize = osize;
1832 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE);
1833 max_asize = max_osize - (VDEV_LABEL_START_SIZE +
1834 VDEV_LABEL_END_SIZE);
1835 } else {
1836 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE -
1837 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) {
1838 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1839 VDEV_AUX_TOO_SMALL);
1840 return (SET_ERROR(EOVERFLOW));
1841 }
1842 psize = 0;
1843 asize = osize;
1844 max_asize = max_osize;
1845 }
1846
1847 /*
1848 * If the vdev was expanded, record this so that we can re-create the
1849 * uberblock rings in labels {2,3}, during the next sync.
1850 */
1851 if ((psize > vd->vdev_psize) && (vd->vdev_psize != 0))
1852 vd->vdev_copy_uberblocks = B_TRUE;
1853
1854 vd->vdev_psize = psize;
1855
1856 /*
1857 * Make sure the allocatable size hasn't shrunk too much.
1858 */
1859 if (asize < vd->vdev_min_asize) {
1860 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1861 VDEV_AUX_BAD_LABEL);
1862 return (SET_ERROR(EINVAL));
1863 }
1864
1865 /*
1866 * We can always set the logical/physical ashift members since
1867 * their values are only used to calculate the vdev_ashift when
1868 * the device is first added to the config. These values should
1869 * not be used for anything else since they may change whenever
1870 * the device is reopened and we don't store them in the label.
1871 */
1872 vd->vdev_physical_ashift =
1873 MAX(physical_ashift, vd->vdev_physical_ashift);
1874 vd->vdev_logical_ashift = MAX(logical_ashift,
1875 vd->vdev_logical_ashift);
1876
1877 if (vd->vdev_asize == 0) {
1878 /*
1879 * This is the first-ever open, so use the computed values.
1880 * For compatibility, a different ashift can be requested.
1881 */
1882 vd->vdev_asize = asize;
1883 vd->vdev_max_asize = max_asize;
1884
1885 /*
1886 * If the vdev_ashift was not overriden at creation time,
1887 * then set it the logical ashift and optimize the ashift.
1888 */
1889 if (vd->vdev_ashift == 0) {
1890 vd->vdev_ashift = vd->vdev_logical_ashift;
1891
1892 if (vd->vdev_logical_ashift > ASHIFT_MAX) {
1893 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1894 VDEV_AUX_ASHIFT_TOO_BIG);
1895 return (SET_ERROR(EDOM));
1896 }
1897
1898 if (vd->vdev_top == vd) {
1899 vdev_ashift_optimize(vd);
1900 }
1901 }
1902 if (vd->vdev_ashift != 0 && (vd->vdev_ashift < ASHIFT_MIN ||
1903 vd->vdev_ashift > ASHIFT_MAX)) {
1904 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1905 VDEV_AUX_BAD_ASHIFT);
1906 return (SET_ERROR(EDOM));
1907 }
1908 } else {
1909 /*
1910 * Make sure the alignment required hasn't increased.
1911 */
1912 if (vd->vdev_ashift > vd->vdev_top->vdev_ashift &&
1913 vd->vdev_ops->vdev_op_leaf) {
1914 (void) zfs_ereport_post(
1915 FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT,
1916 spa, vd, NULL, NULL, 0);
1917 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1918 VDEV_AUX_BAD_LABEL);
1919 return (SET_ERROR(EDOM));
1920 }
1921 vd->vdev_max_asize = max_asize;
1922 }
1923
1924 /*
1925 * If all children are healthy we update asize if either:
1926 * The asize has increased, due to a device expansion caused by dynamic
1927 * LUN growth or vdev replacement, and automatic expansion is enabled;
1928 * making the additional space available.
1929 *
1930 * The asize has decreased, due to a device shrink usually caused by a
1931 * vdev replace with a smaller device. This ensures that calculations
1932 * based of max_asize and asize e.g. esize are always valid. It's safe
1933 * to do this as we've already validated that asize is greater than
1934 * vdev_min_asize.
1935 */
1936 if (vd->vdev_state == VDEV_STATE_HEALTHY &&
1937 ((asize > vd->vdev_asize &&
1938 (vd->vdev_expanding || spa->spa_autoexpand)) ||
1939 (asize < vd->vdev_asize)))
1940 vd->vdev_asize = asize;
1941
1942 vdev_set_min_asize(vd);
1943
1944 /*
1945 * Ensure we can issue some IO before declaring the
1946 * vdev open for business.
1947 */
1948 if (vd->vdev_ops->vdev_op_leaf &&
1949 (error = zio_wait(vdev_probe(vd, NULL))) != 0) {
1950 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1951 VDEV_AUX_ERR_EXCEEDED);
1952 return (error);
1953 }
1954
1955 /*
1956 * If this is a leaf vdev, assess whether a resilver is needed.
1957 * But don't do this if we are doing a reopen for a scrub, since
1958 * this would just restart the scrub we are already doing.
1959 */
1960 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen)
1961 dsl_scan_assess_vdev(spa->spa_dsl_pool, vd);
1962
1963 return (0);
1964 }
1965
1966 /*
1967 * Called once the vdevs are all opened, this routine validates the label
1968 * contents. This needs to be done before vdev_load() so that we don't
1969 * inadvertently do repair I/Os to the wrong device.
1970 *
1971 * This function will only return failure if one of the vdevs indicates that it
1972 * has since been destroyed or exported. This is only possible if
1973 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1974 * will be updated but the function will return 0.
1975 */
1976 int
1977 vdev_validate(vdev_t *vd)
1978 {
1979 spa_t *spa = vd->vdev_spa;
1980 nvlist_t *label;
1981 uint64_t guid = 0, aux_guid = 0, top_guid;
1982 uint64_t state;
1983 nvlist_t *nvl;
1984 uint64_t txg;
1985
1986 if (vdev_validate_skip)
1987 return (0);
1988
1989 for (uint64_t c = 0; c < vd->vdev_children; c++)
1990 if (vdev_validate(vd->vdev_child[c]) != 0)
1991 return (SET_ERROR(EBADF));
1992
1993 /*
1994 * If the device has already failed, or was marked offline, don't do
1995 * any further validation. Otherwise, label I/O will fail and we will
1996 * overwrite the previous state.
1997 */
1998 if (!vd->vdev_ops->vdev_op_leaf || !vdev_readable(vd))
1999 return (0);
2000
2001 /*
2002 * If we are performing an extreme rewind, we allow for a label that
2003 * was modified at a point after the current txg.
2004 * If config lock is not held do not check for the txg. spa_sync could
2005 * be updating the vdev's label before updating spa_last_synced_txg.
2006 */
2007 if (spa->spa_extreme_rewind || spa_last_synced_txg(spa) == 0 ||
2008 spa_config_held(spa, SCL_CONFIG, RW_WRITER) != SCL_CONFIG)
2009 txg = UINT64_MAX;
2010 else
2011 txg = spa_last_synced_txg(spa);
2012
2013 if ((label = vdev_label_read_config(vd, txg)) == NULL) {
2014 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2015 VDEV_AUX_BAD_LABEL);
2016 vdev_dbgmsg(vd, "vdev_validate: failed reading config for "
2017 "txg %llu", (u_longlong_t)txg);
2018 return (0);
2019 }
2020
2021 /*
2022 * Determine if this vdev has been split off into another
2023 * pool. If so, then refuse to open it.
2024 */
2025 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID,
2026 &aux_guid) == 0 && aux_guid == spa_guid(spa)) {
2027 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2028 VDEV_AUX_SPLIT_POOL);
2029 nvlist_free(label);
2030 vdev_dbgmsg(vd, "vdev_validate: vdev split into other pool");
2031 return (0);
2032 }
2033
2034 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, &guid) != 0) {
2035 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2036 VDEV_AUX_CORRUPT_DATA);
2037 nvlist_free(label);
2038 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
2039 ZPOOL_CONFIG_POOL_GUID);
2040 return (0);
2041 }
2042
2043 /*
2044 * If config is not trusted then ignore the spa guid check. This is
2045 * necessary because if the machine crashed during a re-guid the new
2046 * guid might have been written to all of the vdev labels, but not the
2047 * cached config. The check will be performed again once we have the
2048 * trusted config from the MOS.
2049 */
2050 if (spa->spa_trust_config && guid != spa_guid(spa)) {
2051 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2052 VDEV_AUX_CORRUPT_DATA);
2053 nvlist_free(label);
2054 vdev_dbgmsg(vd, "vdev_validate: vdev label pool_guid doesn't "
2055 "match config (%llu != %llu)", (u_longlong_t)guid,
2056 (u_longlong_t)spa_guid(spa));
2057 return (0);
2058 }
2059
2060 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl)
2061 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID,
2062 &aux_guid) != 0)
2063 aux_guid = 0;
2064
2065 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0) {
2066 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2067 VDEV_AUX_CORRUPT_DATA);
2068 nvlist_free(label);
2069 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
2070 ZPOOL_CONFIG_GUID);
2071 return (0);
2072 }
2073
2074 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, &top_guid)
2075 != 0) {
2076 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2077 VDEV_AUX_CORRUPT_DATA);
2078 nvlist_free(label);
2079 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
2080 ZPOOL_CONFIG_TOP_GUID);
2081 return (0);
2082 }
2083
2084 /*
2085 * If this vdev just became a top-level vdev because its sibling was
2086 * detached, it will have adopted the parent's vdev guid -- but the
2087 * label may or may not be on disk yet. Fortunately, either version
2088 * of the label will have the same top guid, so if we're a top-level
2089 * vdev, we can safely compare to that instead.
2090 * However, if the config comes from a cachefile that failed to update
2091 * after the detach, a top-level vdev will appear as a non top-level
2092 * vdev in the config. Also relax the constraints if we perform an
2093 * extreme rewind.
2094 *
2095 * If we split this vdev off instead, then we also check the
2096 * original pool's guid. We don't want to consider the vdev
2097 * corrupt if it is partway through a split operation.
2098 */
2099 if (vd->vdev_guid != guid && vd->vdev_guid != aux_guid) {
2100 boolean_t mismatch = B_FALSE;
2101 if (spa->spa_trust_config && !spa->spa_extreme_rewind) {
2102 if (vd != vd->vdev_top || vd->vdev_guid != top_guid)
2103 mismatch = B_TRUE;
2104 } else {
2105 if (vd->vdev_guid != top_guid &&
2106 vd->vdev_top->vdev_guid != guid)
2107 mismatch = B_TRUE;
2108 }
2109
2110 if (mismatch) {
2111 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2112 VDEV_AUX_CORRUPT_DATA);
2113 nvlist_free(label);
2114 vdev_dbgmsg(vd, "vdev_validate: config guid "
2115 "doesn't match label guid");
2116 vdev_dbgmsg(vd, "CONFIG: guid %llu, top_guid %llu",
2117 (u_longlong_t)vd->vdev_guid,
2118 (u_longlong_t)vd->vdev_top->vdev_guid);
2119 vdev_dbgmsg(vd, "LABEL: guid %llu, top_guid %llu, "
2120 "aux_guid %llu", (u_longlong_t)guid,
2121 (u_longlong_t)top_guid, (u_longlong_t)aux_guid);
2122 return (0);
2123 }
2124 }
2125
2126 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
2127 &state) != 0) {
2128 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2129 VDEV_AUX_CORRUPT_DATA);
2130 nvlist_free(label);
2131 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
2132 ZPOOL_CONFIG_POOL_STATE);
2133 return (0);
2134 }
2135
2136 nvlist_free(label);
2137
2138 /*
2139 * If this is a verbatim import, no need to check the
2140 * state of the pool.
2141 */
2142 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) &&
2143 spa_load_state(spa) == SPA_LOAD_OPEN &&
2144 state != POOL_STATE_ACTIVE) {
2145 vdev_dbgmsg(vd, "vdev_validate: invalid pool state (%llu) "
2146 "for spa %s", (u_longlong_t)state, spa->spa_name);
2147 return (SET_ERROR(EBADF));
2148 }
2149
2150 /*
2151 * If we were able to open and validate a vdev that was
2152 * previously marked permanently unavailable, clear that state
2153 * now.
2154 */
2155 if (vd->vdev_not_present)
2156 vd->vdev_not_present = 0;
2157
2158 return (0);
2159 }
2160
2161 static void
2162 vdev_copy_path_impl(vdev_t *svd, vdev_t *dvd)
2163 {
2164 if (svd->vdev_path != NULL && dvd->vdev_path != NULL) {
2165 if (strcmp(svd->vdev_path, dvd->vdev_path) != 0) {
2166 zfs_dbgmsg("vdev_copy_path: vdev %llu: path changed "
2167 "from '%s' to '%s'", (u_longlong_t)dvd->vdev_guid,
2168 dvd->vdev_path, svd->vdev_path);
2169 spa_strfree(dvd->vdev_path);
2170 dvd->vdev_path = spa_strdup(svd->vdev_path);
2171 }
2172 } else if (svd->vdev_path != NULL) {
2173 dvd->vdev_path = spa_strdup(svd->vdev_path);
2174 zfs_dbgmsg("vdev_copy_path: vdev %llu: path set to '%s'",
2175 (u_longlong_t)dvd->vdev_guid, dvd->vdev_path);
2176 }
2177 }
2178
2179 /*
2180 * Recursively copy vdev paths from one vdev to another. Source and destination
2181 * vdev trees must have same geometry otherwise return error. Intended to copy
2182 * paths from userland config into MOS config.
2183 */
2184 int
2185 vdev_copy_path_strict(vdev_t *svd, vdev_t *dvd)
2186 {
2187 if ((svd->vdev_ops == &vdev_missing_ops) ||
2188 (svd->vdev_ishole && dvd->vdev_ishole) ||
2189 (dvd->vdev_ops == &vdev_indirect_ops))
2190 return (0);
2191
2192 if (svd->vdev_ops != dvd->vdev_ops) {
2193 vdev_dbgmsg(svd, "vdev_copy_path: vdev type mismatch: %s != %s",
2194 svd->vdev_ops->vdev_op_type, dvd->vdev_ops->vdev_op_type);
2195 return (SET_ERROR(EINVAL));
2196 }
2197
2198 if (svd->vdev_guid != dvd->vdev_guid) {
2199 vdev_dbgmsg(svd, "vdev_copy_path: guids mismatch (%llu != "
2200 "%llu)", (u_longlong_t)svd->vdev_guid,
2201 (u_longlong_t)dvd->vdev_guid);
2202 return (SET_ERROR(EINVAL));
2203 }
2204
2205 if (svd->vdev_children != dvd->vdev_children) {
2206 vdev_dbgmsg(svd, "vdev_copy_path: children count mismatch: "
2207 "%llu != %llu", (u_longlong_t)svd->vdev_children,
2208 (u_longlong_t)dvd->vdev_children);
2209 return (SET_ERROR(EINVAL));
2210 }
2211
2212 for (uint64_t i = 0; i < svd->vdev_children; i++) {
2213 int error = vdev_copy_path_strict(svd->vdev_child[i],
2214 dvd->vdev_child[i]);
2215 if (error != 0)
2216 return (error);
2217 }
2218
2219 if (svd->vdev_ops->vdev_op_leaf)
2220 vdev_copy_path_impl(svd, dvd);
2221
2222 return (0);
2223 }
2224
2225 static void
2226 vdev_copy_path_search(vdev_t *stvd, vdev_t *dvd)
2227 {
2228 ASSERT(stvd->vdev_top == stvd);
2229 ASSERT3U(stvd->vdev_id, ==, dvd->vdev_top->vdev_id);
2230
2231 for (uint64_t i = 0; i < dvd->vdev_children; i++) {
2232 vdev_copy_path_search(stvd, dvd->vdev_child[i]);
2233 }
2234
2235 if (!dvd->vdev_ops->vdev_op_leaf || !vdev_is_concrete(dvd))
2236 return;
2237
2238 /*
2239 * The idea here is that while a vdev can shift positions within
2240 * a top vdev (when replacing, attaching mirror, etc.) it cannot
2241 * step outside of it.
2242 */
2243 vdev_t *vd = vdev_lookup_by_guid(stvd, dvd->vdev_guid);
2244
2245 if (vd == NULL || vd->vdev_ops != dvd->vdev_ops)
2246 return;
2247
2248 ASSERT(vd->vdev_ops->vdev_op_leaf);
2249
2250 vdev_copy_path_impl(vd, dvd);
2251 }
2252
2253 /*
2254 * Recursively copy vdev paths from one root vdev to another. Source and
2255 * destination vdev trees may differ in geometry. For each destination leaf
2256 * vdev, search a vdev with the same guid and top vdev id in the source.
2257 * Intended to copy paths from userland config into MOS config.
2258 */
2259 void
2260 vdev_copy_path_relaxed(vdev_t *srvd, vdev_t *drvd)
2261 {
2262 uint64_t children = MIN(srvd->vdev_children, drvd->vdev_children);
2263 ASSERT(srvd->vdev_ops == &vdev_root_ops);
2264 ASSERT(drvd->vdev_ops == &vdev_root_ops);
2265
2266 for (uint64_t i = 0; i < children; i++) {
2267 vdev_copy_path_search(srvd->vdev_child[i],
2268 drvd->vdev_child[i]);
2269 }
2270 }
2271
2272 /*
2273 * Close a virtual device.
2274 */
2275 void
2276 vdev_close(vdev_t *vd)
2277 {
2278 vdev_t *pvd = vd->vdev_parent;
2279 spa_t *spa __maybe_unused = vd->vdev_spa;
2280
2281 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2282
2283 /*
2284 * If our parent is reopening, then we are as well, unless we are
2285 * going offline.
2286 */
2287 if (pvd != NULL && pvd->vdev_reopening)
2288 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline);
2289
2290 vd->vdev_ops->vdev_op_close(vd);
2291
2292 vdev_cache_purge(vd);
2293
2294 /*
2295 * We record the previous state before we close it, so that if we are
2296 * doing a reopen(), we don't generate FMA ereports if we notice that
2297 * it's still faulted.
2298 */
2299 vd->vdev_prevstate = vd->vdev_state;
2300
2301 if (vd->vdev_offline)
2302 vd->vdev_state = VDEV_STATE_OFFLINE;
2303 else
2304 vd->vdev_state = VDEV_STATE_CLOSED;
2305 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
2306 }
2307
2308 void
2309 vdev_hold(vdev_t *vd)
2310 {
2311 spa_t *spa = vd->vdev_spa;
2312
2313 ASSERT(spa_is_root(spa));
2314 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
2315 return;
2316
2317 for (int c = 0; c < vd->vdev_children; c++)
2318 vdev_hold(vd->vdev_child[c]);
2319
2320 if (vd->vdev_ops->vdev_op_leaf)
2321 vd->vdev_ops->vdev_op_hold(vd);
2322 }
2323
2324 void
2325 vdev_rele(vdev_t *vd)
2326 {
2327 ASSERT(spa_is_root(vd->vdev_spa));
2328 for (int c = 0; c < vd->vdev_children; c++)
2329 vdev_rele(vd->vdev_child[c]);
2330
2331 if (vd->vdev_ops->vdev_op_leaf)
2332 vd->vdev_ops->vdev_op_rele(vd);
2333 }
2334
2335 /*
2336 * Reopen all interior vdevs and any unopened leaves. We don't actually
2337 * reopen leaf vdevs which had previously been opened as they might deadlock
2338 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
2339 * If the leaf has never been opened then open it, as usual.
2340 */
2341 void
2342 vdev_reopen(vdev_t *vd)
2343 {
2344 spa_t *spa = vd->vdev_spa;
2345
2346 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2347
2348 /* set the reopening flag unless we're taking the vdev offline */
2349 vd->vdev_reopening = !vd->vdev_offline;
2350 vdev_close(vd);
2351 (void) vdev_open(vd);
2352
2353 /*
2354 * Call vdev_validate() here to make sure we have the same device.
2355 * Otherwise, a device with an invalid label could be successfully
2356 * opened in response to vdev_reopen().
2357 */
2358 if (vd->vdev_aux) {
2359 (void) vdev_validate_aux(vd);
2360 if (vdev_readable(vd) && vdev_writeable(vd) &&
2361 vd->vdev_aux == &spa->spa_l2cache) {
2362 /*
2363 * In case the vdev is present we should evict all ARC
2364 * buffers and pointers to log blocks and reclaim their
2365 * space before restoring its contents to L2ARC.
2366 */
2367 if (l2arc_vdev_present(vd)) {
2368 l2arc_rebuild_vdev(vd, B_TRUE);
2369 } else {
2370 l2arc_add_vdev(spa, vd);
2371 }
2372 spa_async_request(spa, SPA_ASYNC_L2CACHE_REBUILD);
2373 spa_async_request(spa, SPA_ASYNC_L2CACHE_TRIM);
2374 }
2375 } else {
2376 (void) vdev_validate(vd);
2377 }
2378
2379 /*
2380 * Reassess parent vdev's health.
2381 */
2382 vdev_propagate_state(vd);
2383 }
2384
2385 int
2386 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
2387 {
2388 int error;
2389
2390 /*
2391 * Normally, partial opens (e.g. of a mirror) are allowed.
2392 * For a create, however, we want to fail the request if
2393 * there are any components we can't open.
2394 */
2395 error = vdev_open(vd);
2396
2397 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) {
2398 vdev_close(vd);
2399 return (error ? error : SET_ERROR(ENXIO));
2400 }
2401
2402 /*
2403 * Recursively load DTLs and initialize all labels.
2404 */
2405 if ((error = vdev_dtl_load(vd)) != 0 ||
2406 (error = vdev_label_init(vd, txg, isreplacing ?
2407 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
2408 vdev_close(vd);
2409 return (error);
2410 }
2411
2412 return (0);
2413 }
2414
2415 void
2416 vdev_metaslab_set_size(vdev_t *vd)
2417 {
2418 uint64_t asize = vd->vdev_asize;
2419 uint64_t ms_count = asize >> zfs_vdev_default_ms_shift;
2420 uint64_t ms_shift;
2421
2422 /*
2423 * There are two dimensions to the metaslab sizing calculation:
2424 * the size of the metaslab and the count of metaslabs per vdev.
2425 *
2426 * The default values used below are a good balance between memory
2427 * usage (larger metaslab size means more memory needed for loaded
2428 * metaslabs; more metaslabs means more memory needed for the
2429 * metaslab_t structs), metaslab load time (larger metaslabs take
2430 * longer to load), and metaslab sync time (more metaslabs means
2431 * more time spent syncing all of them).
2432 *
2433 * In general, we aim for zfs_vdev_default_ms_count (200) metaslabs.
2434 * The range of the dimensions are as follows:
2435 *
2436 * 2^29 <= ms_size <= 2^34
2437 * 16 <= ms_count <= 131,072
2438 *
2439 * On the lower end of vdev sizes, we aim for metaslabs sizes of
2440 * at least 512MB (2^29) to minimize fragmentation effects when
2441 * testing with smaller devices. However, the count constraint
2442 * of at least 16 metaslabs will override this minimum size goal.
2443 *
2444 * On the upper end of vdev sizes, we aim for a maximum metaslab
2445 * size of 16GB. However, we will cap the total count to 2^17
2446 * metaslabs to keep our memory footprint in check and let the
2447 * metaslab size grow from there if that limit is hit.
2448 *
2449 * The net effect of applying above constrains is summarized below.
2450 *
2451 * vdev size metaslab count
2452 * --------------|-----------------
2453 * < 8GB ~16
2454 * 8GB - 100GB one per 512MB
2455 * 100GB - 3TB ~200
2456 * 3TB - 2PB one per 16GB
2457 * > 2PB ~131,072
2458 * --------------------------------
2459 *
2460 * Finally, note that all of the above calculate the initial
2461 * number of metaslabs. Expanding a top-level vdev will result
2462 * in additional metaslabs being allocated making it possible
2463 * to exceed the zfs_vdev_ms_count_limit.
2464 */
2465
2466 if (ms_count < zfs_vdev_min_ms_count)
2467 ms_shift = highbit64(asize / zfs_vdev_min_ms_count);
2468 else if (ms_count > zfs_vdev_default_ms_count)
2469 ms_shift = highbit64(asize / zfs_vdev_default_ms_count);
2470 else
2471 ms_shift = zfs_vdev_default_ms_shift;
2472
2473 if (ms_shift < SPA_MAXBLOCKSHIFT) {
2474 ms_shift = SPA_MAXBLOCKSHIFT;
2475 } else if (ms_shift > zfs_vdev_max_ms_shift) {
2476 ms_shift = zfs_vdev_max_ms_shift;
2477 /* cap the total count to constrain memory footprint */
2478 if ((asize >> ms_shift) > zfs_vdev_ms_count_limit)
2479 ms_shift = highbit64(asize / zfs_vdev_ms_count_limit);
2480 }
2481
2482 vd->vdev_ms_shift = ms_shift;
2483 ASSERT3U(vd->vdev_ms_shift, >=, SPA_MAXBLOCKSHIFT);
2484 }
2485
2486 void
2487 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
2488 {
2489 ASSERT(vd == vd->vdev_top);
2490 /* indirect vdevs don't have metaslabs or dtls */
2491 ASSERT(vdev_is_concrete(vd) || flags == 0);
2492 ASSERT(ISP2(flags));
2493 ASSERT(spa_writeable(vd->vdev_spa));
2494
2495 if (flags & VDD_METASLAB)
2496 (void) txg_list_add(&vd->vdev_ms_list, arg, txg);
2497
2498 if (flags & VDD_DTL)
2499 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg);
2500
2501 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
2502 }
2503
2504 void
2505 vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg)
2506 {
2507 for (int c = 0; c < vd->vdev_children; c++)
2508 vdev_dirty_leaves(vd->vdev_child[c], flags, txg);
2509
2510 if (vd->vdev_ops->vdev_op_leaf)
2511 vdev_dirty(vd->vdev_top, flags, vd, txg);
2512 }
2513
2514 /*
2515 * DTLs.
2516 *
2517 * A vdev's DTL (dirty time log) is the set of transaction groups for which
2518 * the vdev has less than perfect replication. There are four kinds of DTL:
2519 *
2520 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
2521 *
2522 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
2523 *
2524 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
2525 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
2526 * txgs that was scrubbed.
2527 *
2528 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
2529 * persistent errors or just some device being offline.
2530 * Unlike the other three, the DTL_OUTAGE map is not generally
2531 * maintained; it's only computed when needed, typically to
2532 * determine whether a device can be detached.
2533 *
2534 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
2535 * either has the data or it doesn't.
2536 *
2537 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
2538 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
2539 * if any child is less than fully replicated, then so is its parent.
2540 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
2541 * comprising only those txgs which appear in 'maxfaults' or more children;
2542 * those are the txgs we don't have enough replication to read. For example,
2543 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
2544 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
2545 * two child DTL_MISSING maps.
2546 *
2547 * It should be clear from the above that to compute the DTLs and outage maps
2548 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
2549 * Therefore, that is all we keep on disk. When loading the pool, or after
2550 * a configuration change, we generate all other DTLs from first principles.
2551 */
2552 void
2553 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
2554 {
2555 range_tree_t *rt = vd->vdev_dtl[t];
2556
2557 ASSERT(t < DTL_TYPES);
2558 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
2559 ASSERT(spa_writeable(vd->vdev_spa));
2560
2561 mutex_enter(&vd->vdev_dtl_lock);
2562 if (!range_tree_contains(rt, txg, size))
2563 range_tree_add(rt, txg, size);
2564 mutex_exit(&vd->vdev_dtl_lock);
2565 }
2566
2567 boolean_t
2568 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
2569 {
2570 range_tree_t *rt = vd->vdev_dtl[t];
2571 boolean_t dirty = B_FALSE;
2572
2573 ASSERT(t < DTL_TYPES);
2574 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
2575
2576 /*
2577 * While we are loading the pool, the DTLs have not been loaded yet.
2578 * Ignore the DTLs and try all devices. This avoids a recursive
2579 * mutex enter on the vdev_dtl_lock, and also makes us try hard
2580 * when loading the pool (relying on the checksum to ensure that
2581 * we get the right data -- note that we while loading, we are
2582 * only reading the MOS, which is always checksummed).
2583 */
2584 if (vd->vdev_spa->spa_load_state != SPA_LOAD_NONE)
2585 return (B_FALSE);
2586
2587 mutex_enter(&vd->vdev_dtl_lock);
2588 if (!range_tree_is_empty(rt))
2589 dirty = range_tree_contains(rt, txg, size);
2590 mutex_exit(&vd->vdev_dtl_lock);
2591
2592 return (dirty);
2593 }
2594
2595 boolean_t
2596 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t)
2597 {
2598 range_tree_t *rt = vd->vdev_dtl[t];
2599 boolean_t empty;
2600
2601 mutex_enter(&vd->vdev_dtl_lock);
2602 empty = range_tree_is_empty(rt);
2603 mutex_exit(&vd->vdev_dtl_lock);
2604
2605 return (empty);
2606 }
2607
2608 /*
2609 * Returns B_TRUE if vdev determines offset needs to be resilvered.
2610 */
2611 boolean_t
2612 vdev_dtl_need_resilver(vdev_t *vd, uint64_t offset, size_t psize)
2613 {
2614 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
2615
2616 if (vd->vdev_ops->vdev_op_need_resilver == NULL ||
2617 vd->vdev_ops->vdev_op_leaf)
2618 return (B_TRUE);
2619
2620 return (vd->vdev_ops->vdev_op_need_resilver(vd, offset, psize));
2621 }
2622
2623 /*
2624 * Returns the lowest txg in the DTL range.
2625 */
2626 static uint64_t
2627 vdev_dtl_min(vdev_t *vd)
2628 {
2629 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
2630 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
2631 ASSERT0(vd->vdev_children);
2632
2633 return (range_tree_min(vd->vdev_dtl[DTL_MISSING]) - 1);
2634 }
2635
2636 /*
2637 * Returns the highest txg in the DTL.
2638 */
2639 static uint64_t
2640 vdev_dtl_max(vdev_t *vd)
2641 {
2642 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
2643 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
2644 ASSERT0(vd->vdev_children);
2645
2646 return (range_tree_max(vd->vdev_dtl[DTL_MISSING]));
2647 }
2648
2649 /*
2650 * Determine if a resilvering vdev should remove any DTL entries from
2651 * its range. If the vdev was resilvering for the entire duration of the
2652 * scan then it should excise that range from its DTLs. Otherwise, this
2653 * vdev is considered partially resilvered and should leave its DTL
2654 * entries intact. The comment in vdev_dtl_reassess() describes how we
2655 * excise the DTLs.
2656 */
2657 static boolean_t
2658 vdev_dtl_should_excise(vdev_t *vd, boolean_t rebuild_done)
2659 {
2660 ASSERT0(vd->vdev_children);
2661
2662 if (vd->vdev_state < VDEV_STATE_DEGRADED)
2663 return (B_FALSE);
2664
2665 if (vd->vdev_resilver_deferred)
2666 return (B_FALSE);
2667
2668 if (range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]))
2669 return (B_TRUE);
2670
2671 if (rebuild_done) {
2672 vdev_rebuild_t *vr = &vd->vdev_top->vdev_rebuild_config;
2673 vdev_rebuild_phys_t *vrp = &vr->vr_rebuild_phys;
2674
2675 /* Rebuild not initiated by attach */
2676 if (vd->vdev_rebuild_txg == 0)
2677 return (B_TRUE);
2678
2679 /*
2680 * When a rebuild completes without error then all missing data
2681 * up to the rebuild max txg has been reconstructed and the DTL
2682 * is eligible for excision.
2683 */
2684 if (vrp->vrp_rebuild_state == VDEV_REBUILD_COMPLETE &&
2685 vdev_dtl_max(vd) <= vrp->vrp_max_txg) {
2686 ASSERT3U(vrp->vrp_min_txg, <=, vdev_dtl_min(vd));
2687 ASSERT3U(vrp->vrp_min_txg, <, vd->vdev_rebuild_txg);
2688 ASSERT3U(vd->vdev_rebuild_txg, <=, vrp->vrp_max_txg);
2689 return (B_TRUE);
2690 }
2691 } else {
2692 dsl_scan_t *scn = vd->vdev_spa->spa_dsl_pool->dp_scan;
2693 dsl_scan_phys_t *scnp __maybe_unused = &scn->scn_phys;
2694
2695 /* Resilver not initiated by attach */
2696 if (vd->vdev_resilver_txg == 0)
2697 return (B_TRUE);
2698
2699 /*
2700 * When a resilver is initiated the scan will assign the
2701 * scn_max_txg value to the highest txg value that exists
2702 * in all DTLs. If this device's max DTL is not part of this
2703 * scan (i.e. it is not in the range (scn_min_txg, scn_max_txg]
2704 * then it is not eligible for excision.
2705 */
2706 if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) {
2707 ASSERT3U(scnp->scn_min_txg, <=, vdev_dtl_min(vd));
2708 ASSERT3U(scnp->scn_min_txg, <, vd->vdev_resilver_txg);
2709 ASSERT3U(vd->vdev_resilver_txg, <=, scnp->scn_max_txg);
2710 return (B_TRUE);
2711 }
2712 }
2713
2714 return (B_FALSE);
2715 }
2716
2717 /*
2718 * Reassess DTLs after a config change or scrub completion. If txg == 0 no
2719 * write operations will be issued to the pool.
2720 */
2721 void
2722 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg,
2723 boolean_t scrub_done, boolean_t rebuild_done)
2724 {
2725 spa_t *spa = vd->vdev_spa;
2726 avl_tree_t reftree;
2727 int minref;
2728
2729 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
2730
2731 for (int c = 0; c < vd->vdev_children; c++)
2732 vdev_dtl_reassess(vd->vdev_child[c], txg,
2733 scrub_txg, scrub_done, rebuild_done);
2734
2735 if (vd == spa->spa_root_vdev || !vdev_is_concrete(vd) || vd->vdev_aux)
2736 return;
2737
2738 if (vd->vdev_ops->vdev_op_leaf) {
2739 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
2740 vdev_rebuild_t *vr = &vd->vdev_top->vdev_rebuild_config;
2741 boolean_t check_excise = B_FALSE;
2742 boolean_t wasempty = B_TRUE;
2743
2744 mutex_enter(&vd->vdev_dtl_lock);
2745
2746 /*
2747 * If requested, pretend the scan or rebuild completed cleanly.
2748 */
2749 if (zfs_scan_ignore_errors) {
2750 if (scn != NULL)
2751 scn->scn_phys.scn_errors = 0;
2752 if (vr != NULL)
2753 vr->vr_rebuild_phys.vrp_errors = 0;
2754 }
2755
2756 if (scrub_txg != 0 &&
2757 !range_tree_is_empty(vd->vdev_dtl[DTL_MISSING])) {
2758 wasempty = B_FALSE;
2759 zfs_dbgmsg("guid:%llu txg:%llu scrub:%llu started:%d "
2760 "dtl:%llu/%llu errors:%llu",
2761 (u_longlong_t)vd->vdev_guid, (u_longlong_t)txg,
2762 (u_longlong_t)scrub_txg, spa->spa_scrub_started,
2763 (u_longlong_t)vdev_dtl_min(vd),
2764 (u_longlong_t)vdev_dtl_max(vd),
2765 (u_longlong_t)(scn ? scn->scn_phys.scn_errors : 0));
2766 }
2767
2768 /*
2769 * If we've completed a scrub/resilver or a rebuild cleanly
2770 * then determine if this vdev should remove any DTLs. We
2771 * only want to excise regions on vdevs that were available
2772 * during the entire duration of this scan.
2773 */
2774 if (rebuild_done &&
2775 vr != NULL && vr->vr_rebuild_phys.vrp_errors == 0) {
2776 check_excise = B_TRUE;
2777 } else {
2778 if (spa->spa_scrub_started ||
2779 (scn != NULL && scn->scn_phys.scn_errors == 0)) {
2780 check_excise = B_TRUE;
2781 }
2782 }
2783
2784 if (scrub_txg && check_excise &&
2785 vdev_dtl_should_excise(vd, rebuild_done)) {
2786 /*
2787 * We completed a scrub, resilver or rebuild up to
2788 * scrub_txg. If we did it without rebooting, then
2789 * the scrub dtl will be valid, so excise the old
2790 * region and fold in the scrub dtl. Otherwise,
2791 * leave the dtl as-is if there was an error.
2792 *
2793 * There's little trick here: to excise the beginning
2794 * of the DTL_MISSING map, we put it into a reference
2795 * tree and then add a segment with refcnt -1 that
2796 * covers the range [0, scrub_txg). This means
2797 * that each txg in that range has refcnt -1 or 0.
2798 * We then add DTL_SCRUB with a refcnt of 2, so that
2799 * entries in the range [0, scrub_txg) will have a
2800 * positive refcnt -- either 1 or 2. We then convert
2801 * the reference tree into the new DTL_MISSING map.
2802 */
2803 space_reftree_create(&reftree);
2804 space_reftree_add_map(&reftree,
2805 vd->vdev_dtl[DTL_MISSING], 1);
2806 space_reftree_add_seg(&reftree, 0, scrub_txg, -1);
2807 space_reftree_add_map(&reftree,
2808 vd->vdev_dtl[DTL_SCRUB], 2);
2809 space_reftree_generate_map(&reftree,
2810 vd->vdev_dtl[DTL_MISSING], 1);
2811 space_reftree_destroy(&reftree);
2812
2813 if (!range_tree_is_empty(vd->vdev_dtl[DTL_MISSING])) {
2814 zfs_dbgmsg("update DTL_MISSING:%llu/%llu",
2815 (u_longlong_t)vdev_dtl_min(vd),
2816 (u_longlong_t)vdev_dtl_max(vd));
2817 } else if (!wasempty) {
2818 zfs_dbgmsg("DTL_MISSING is now empty");
2819 }
2820 }
2821 range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
2822 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
2823 range_tree_add, vd->vdev_dtl[DTL_PARTIAL]);
2824 if (scrub_done)
2825 range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
2826 range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
2827 if (!vdev_readable(vd))
2828 range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
2829 else
2830 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
2831 range_tree_add, vd->vdev_dtl[DTL_OUTAGE]);
2832
2833 /*
2834 * If the vdev was resilvering or rebuilding and no longer
2835 * has any DTLs then reset the appropriate flag and dirty
2836 * the top level so that we persist the change.
2837 */
2838 if (txg != 0 &&
2839 range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]) &&
2840 range_tree_is_empty(vd->vdev_dtl[DTL_OUTAGE])) {
2841 if (vd->vdev_rebuild_txg != 0) {
2842 vd->vdev_rebuild_txg = 0;
2843 vdev_config_dirty(vd->vdev_top);
2844 } else if (vd->vdev_resilver_txg != 0) {
2845 vd->vdev_resilver_txg = 0;
2846 vdev_config_dirty(vd->vdev_top);
2847 }
2848 }
2849
2850 mutex_exit(&vd->vdev_dtl_lock);
2851
2852 if (txg != 0)
2853 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
2854 return;
2855 }
2856
2857 mutex_enter(&vd->vdev_dtl_lock);
2858 for (int t = 0; t < DTL_TYPES; t++) {
2859 /* account for child's outage in parent's missing map */
2860 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t;
2861 if (t == DTL_SCRUB)
2862 continue; /* leaf vdevs only */
2863 if (t == DTL_PARTIAL)
2864 minref = 1; /* i.e. non-zero */
2865 else if (vd->vdev_nparity != 0)
2866 minref = vd->vdev_nparity + 1; /* RAID-Z */
2867 else
2868 minref = vd->vdev_children; /* any kind of mirror */
2869 space_reftree_create(&reftree);
2870 for (int c = 0; c < vd->vdev_children; c++) {
2871 vdev_t *cvd = vd->vdev_child[c];
2872 mutex_enter(&cvd->vdev_dtl_lock);
2873 space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1);
2874 mutex_exit(&cvd->vdev_dtl_lock);
2875 }
2876 space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref);
2877 space_reftree_destroy(&reftree);
2878 }
2879 mutex_exit(&vd->vdev_dtl_lock);
2880 }
2881
2882 int
2883 vdev_dtl_load(vdev_t *vd)
2884 {
2885 spa_t *spa = vd->vdev_spa;
2886 objset_t *mos = spa->spa_meta_objset;
2887 int error = 0;
2888
2889 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) {
2890 ASSERT(vdev_is_concrete(vd));
2891
2892 error = space_map_open(&vd->vdev_dtl_sm, mos,
2893 vd->vdev_dtl_object, 0, -1ULL, 0);
2894 if (error)
2895 return (error);
2896 ASSERT(vd->vdev_dtl_sm != NULL);
2897
2898 mutex_enter(&vd->vdev_dtl_lock);
2899 error = space_map_load(vd->vdev_dtl_sm,
2900 vd->vdev_dtl[DTL_MISSING], SM_ALLOC);
2901 mutex_exit(&vd->vdev_dtl_lock);
2902
2903 return (error);
2904 }
2905
2906 for (int c = 0; c < vd->vdev_children; c++) {
2907 error = vdev_dtl_load(vd->vdev_child[c]);
2908 if (error != 0)
2909 break;
2910 }
2911
2912 return (error);
2913 }
2914
2915 static void
2916 vdev_zap_allocation_data(vdev_t *vd, dmu_tx_t *tx)
2917 {
2918 spa_t *spa = vd->vdev_spa;
2919 objset_t *mos = spa->spa_meta_objset;
2920 vdev_alloc_bias_t alloc_bias = vd->vdev_alloc_bias;
2921 const char *string;
2922
2923 ASSERT(alloc_bias != VDEV_BIAS_NONE);
2924
2925 string =
2926 (alloc_bias == VDEV_BIAS_LOG) ? VDEV_ALLOC_BIAS_LOG :
2927 (alloc_bias == VDEV_BIAS_SPECIAL) ? VDEV_ALLOC_BIAS_SPECIAL :
2928 (alloc_bias == VDEV_BIAS_DEDUP) ? VDEV_ALLOC_BIAS_DEDUP : NULL;
2929
2930 ASSERT(string != NULL);
2931 VERIFY0(zap_add(mos, vd->vdev_top_zap, VDEV_TOP_ZAP_ALLOCATION_BIAS,
2932 1, strlen(string) + 1, string, tx));
2933
2934 if (alloc_bias == VDEV_BIAS_SPECIAL || alloc_bias == VDEV_BIAS_DEDUP) {
2935 spa_activate_allocation_classes(spa, tx);
2936 }
2937 }
2938
2939 void
2940 vdev_destroy_unlink_zap(vdev_t *vd, uint64_t zapobj, dmu_tx_t *tx)
2941 {
2942 spa_t *spa = vd->vdev_spa;
2943
2944 VERIFY0(zap_destroy(spa->spa_meta_objset, zapobj, tx));
2945 VERIFY0(zap_remove_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps,
2946 zapobj, tx));
2947 }
2948
2949 uint64_t
2950 vdev_create_link_zap(vdev_t *vd, dmu_tx_t *tx)
2951 {
2952 spa_t *spa = vd->vdev_spa;
2953 uint64_t zap = zap_create(spa->spa_meta_objset, DMU_OTN_ZAP_METADATA,
2954 DMU_OT_NONE, 0, tx);
2955
2956 ASSERT(zap != 0);
2957 VERIFY0(zap_add_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps,
2958 zap, tx));
2959
2960 return (zap);
2961 }
2962
2963 void
2964 vdev_construct_zaps(vdev_t *vd, dmu_tx_t *tx)
2965 {
2966 if (vd->vdev_ops != &vdev_hole_ops &&
2967 vd->vdev_ops != &vdev_missing_ops &&
2968 vd->vdev_ops != &vdev_root_ops &&
2969 !vd->vdev_top->vdev_removing) {
2970 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_leaf_zap == 0) {
2971 vd->vdev_leaf_zap = vdev_create_link_zap(vd, tx);
2972 }
2973 if (vd == vd->vdev_top && vd->vdev_top_zap == 0) {
2974 vd->vdev_top_zap = vdev_create_link_zap(vd, tx);
2975 if (vd->vdev_alloc_bias != VDEV_BIAS_NONE)
2976 vdev_zap_allocation_data(vd, tx);
2977 }
2978 }
2979
2980 for (uint64_t i = 0; i < vd->vdev_children; i++) {
2981 vdev_construct_zaps(vd->vdev_child[i], tx);
2982 }
2983 }
2984
2985 static void
2986 vdev_dtl_sync(vdev_t *vd, uint64_t txg)
2987 {
2988 spa_t *spa = vd->vdev_spa;
2989 range_tree_t *rt = vd->vdev_dtl[DTL_MISSING];
2990 objset_t *mos = spa->spa_meta_objset;
2991 range_tree_t *rtsync;
2992 dmu_tx_t *tx;
2993 uint64_t object = space_map_object(vd->vdev_dtl_sm);
2994
2995 ASSERT(vdev_is_concrete(vd));
2996 ASSERT(vd->vdev_ops->vdev_op_leaf);
2997
2998 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2999
3000 if (vd->vdev_detached || vd->vdev_top->vdev_removing) {
3001 mutex_enter(&vd->vdev_dtl_lock);
3002 space_map_free(vd->vdev_dtl_sm, tx);
3003 space_map_close(vd->vdev_dtl_sm);
3004 vd->vdev_dtl_sm = NULL;
3005 mutex_exit(&vd->vdev_dtl_lock);
3006
3007 /*
3008 * We only destroy the leaf ZAP for detached leaves or for
3009 * removed log devices. Removed data devices handle leaf ZAP
3010 * cleanup later, once cancellation is no longer possible.
3011 */
3012 if (vd->vdev_leaf_zap != 0 && (vd->vdev_detached ||
3013 vd->vdev_top->vdev_islog)) {
3014 vdev_destroy_unlink_zap(vd, vd->vdev_leaf_zap, tx);
3015 vd->vdev_leaf_zap = 0;
3016 }
3017
3018 dmu_tx_commit(tx);
3019 return;
3020 }
3021
3022 if (vd->vdev_dtl_sm == NULL) {
3023 uint64_t new_object;
3024
3025 new_object = space_map_alloc(mos, zfs_vdev_dtl_sm_blksz, tx);
3026 VERIFY3U(new_object, !=, 0);
3027
3028 VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object,
3029 0, -1ULL, 0));
3030 ASSERT(vd->vdev_dtl_sm != NULL);
3031 }
3032
3033 rtsync = range_tree_create(NULL, RANGE_SEG64, NULL, 0, 0);
3034
3035 mutex_enter(&vd->vdev_dtl_lock);
3036 range_tree_walk(rt, range_tree_add, rtsync);
3037 mutex_exit(&vd->vdev_dtl_lock);
3038
3039 space_map_truncate(vd->vdev_dtl_sm, zfs_vdev_dtl_sm_blksz, tx);
3040 space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, SM_NO_VDEVID, tx);
3041 range_tree_vacate(rtsync, NULL, NULL);
3042
3043 range_tree_destroy(rtsync);
3044
3045 /*
3046 * If the object for the space map has changed then dirty
3047 * the top level so that we update the config.
3048 */
3049 if (object != space_map_object(vd->vdev_dtl_sm)) {
3050 vdev_dbgmsg(vd, "txg %llu, spa %s, DTL old object %llu, "
3051 "new object %llu", (u_longlong_t)txg, spa_name(spa),
3052 (u_longlong_t)object,
3053 (u_longlong_t)space_map_object(vd->vdev_dtl_sm));
3054 vdev_config_dirty(vd->vdev_top);
3055 }
3056
3057 dmu_tx_commit(tx);
3058 }
3059
3060 /*
3061 * Determine whether the specified vdev can be offlined/detached/removed
3062 * without losing data.
3063 */
3064 boolean_t
3065 vdev_dtl_required(vdev_t *vd)
3066 {
3067 spa_t *spa = vd->vdev_spa;
3068 vdev_t *tvd = vd->vdev_top;
3069 uint8_t cant_read = vd->vdev_cant_read;
3070 boolean_t required;
3071
3072 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
3073
3074 if (vd == spa->spa_root_vdev || vd == tvd)
3075 return (B_TRUE);
3076
3077 /*
3078 * Temporarily mark the device as unreadable, and then determine
3079 * whether this results in any DTL outages in the top-level vdev.
3080 * If not, we can safely offline/detach/remove the device.
3081 */
3082 vd->vdev_cant_read = B_TRUE;
3083 vdev_dtl_reassess(tvd, 0, 0, B_FALSE, B_FALSE);
3084 required = !vdev_dtl_empty(tvd, DTL_OUTAGE);
3085 vd->vdev_cant_read = cant_read;
3086 vdev_dtl_reassess(tvd, 0, 0, B_FALSE, B_FALSE);
3087
3088 if (!required && zio_injection_enabled) {
3089 required = !!zio_handle_device_injection(vd, NULL,
3090 SET_ERROR(ECHILD));
3091 }
3092
3093 return (required);
3094 }
3095
3096 /*
3097 * Determine if resilver is needed, and if so the txg range.
3098 */
3099 boolean_t
3100 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
3101 {
3102 boolean_t needed = B_FALSE;
3103 uint64_t thismin = UINT64_MAX;
3104 uint64_t thismax = 0;
3105
3106 if (vd->vdev_children == 0) {
3107 mutex_enter(&vd->vdev_dtl_lock);
3108 if (!range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]) &&
3109 vdev_writeable(vd)) {
3110
3111 thismin = vdev_dtl_min(vd);
3112 thismax = vdev_dtl_max(vd);
3113 needed = B_TRUE;
3114 }
3115 mutex_exit(&vd->vdev_dtl_lock);
3116 } else {
3117 for (int c = 0; c < vd->vdev_children; c++) {
3118 vdev_t *cvd = vd->vdev_child[c];
3119 uint64_t cmin, cmax;
3120
3121 if (vdev_resilver_needed(cvd, &cmin, &cmax)) {
3122 thismin = MIN(thismin, cmin);
3123 thismax = MAX(thismax, cmax);
3124 needed = B_TRUE;
3125 }
3126 }
3127 }
3128
3129 if (needed && minp) {
3130 *minp = thismin;
3131 *maxp = thismax;
3132 }
3133 return (needed);
3134 }
3135
3136 /*
3137 * Gets the checkpoint space map object from the vdev's ZAP. On success sm_obj
3138 * will contain either the checkpoint spacemap object or zero if none exists.
3139 * All other errors are returned to the caller.
3140 */
3141 int
3142 vdev_checkpoint_sm_object(vdev_t *vd, uint64_t *sm_obj)
3143 {
3144 ASSERT0(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER));
3145
3146 if (vd->vdev_top_zap == 0) {
3147 *sm_obj = 0;
3148 return (0);
3149 }
3150
3151 int error = zap_lookup(spa_meta_objset(vd->vdev_spa), vd->vdev_top_zap,
3152 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM, sizeof (uint64_t), 1, sm_obj);
3153 if (error == ENOENT) {
3154 *sm_obj = 0;
3155 error = 0;
3156 }
3157
3158 return (error);
3159 }
3160
3161 int
3162 vdev_load(vdev_t *vd)
3163 {
3164 int error = 0;
3165
3166 /*
3167 * Recursively load all children.
3168 */
3169 for (int c = 0; c < vd->vdev_children; c++) {
3170 error = vdev_load(vd->vdev_child[c]);
3171 if (error != 0) {
3172 return (error);
3173 }
3174 }
3175
3176 vdev_set_deflate_ratio(vd);
3177
3178 /*
3179 * On spa_load path, grab the allocation bias from our zap
3180 */
3181 if (vd == vd->vdev_top && vd->vdev_top_zap != 0) {
3182 spa_t *spa = vd->vdev_spa;
3183 char bias_str[64];
3184
3185 error = zap_lookup(spa->spa_meta_objset, vd->vdev_top_zap,
3186 VDEV_TOP_ZAP_ALLOCATION_BIAS, 1, sizeof (bias_str),
3187 bias_str);
3188 if (error == 0) {
3189 ASSERT(vd->vdev_alloc_bias == VDEV_BIAS_NONE);
3190 vd->vdev_alloc_bias = vdev_derive_alloc_bias(bias_str);
3191 } else if (error != ENOENT) {
3192 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
3193 VDEV_AUX_CORRUPT_DATA);
3194 vdev_dbgmsg(vd, "vdev_load: zap_lookup(top_zap=%llu) "
3195 "failed [error=%d]", vd->vdev_top_zap, error);
3196 return (error);
3197 }
3198 }
3199
3200 /*
3201 * Load any rebuild state from the top-level vdev zap.
3202 */
3203 if (vd == vd->vdev_top && vd->vdev_top_zap != 0) {
3204 error = vdev_rebuild_load(vd);
3205 if (error && error != ENOTSUP) {
3206 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
3207 VDEV_AUX_CORRUPT_DATA);
3208 vdev_dbgmsg(vd, "vdev_load: vdev_rebuild_load "
3209 "failed [error=%d]", error);
3210 return (error);
3211 }
3212 }
3213
3214 /*
3215 * If this is a top-level vdev, initialize its metaslabs.
3216 */
3217 if (vd == vd->vdev_top && vdev_is_concrete(vd)) {
3218 vdev_metaslab_group_create(vd);
3219
3220 if (vd->vdev_ashift == 0 || vd->vdev_asize == 0) {
3221 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
3222 VDEV_AUX_CORRUPT_DATA);
3223 vdev_dbgmsg(vd, "vdev_load: invalid size. ashift=%llu, "
3224 "asize=%llu", (u_longlong_t)vd->vdev_ashift,
3225 (u_longlong_t)vd->vdev_asize);
3226 return (SET_ERROR(ENXIO));
3227 }
3228
3229 error = vdev_metaslab_init(vd, 0);
3230 if (error != 0) {
3231 vdev_dbgmsg(vd, "vdev_load: metaslab_init failed "
3232 "[error=%d]", error);
3233 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
3234 VDEV_AUX_CORRUPT_DATA);
3235 return (error);
3236 }
3237
3238 uint64_t checkpoint_sm_obj;
3239 error = vdev_checkpoint_sm_object(vd, &checkpoint_sm_obj);
3240 if (error == 0 && checkpoint_sm_obj != 0) {
3241 objset_t *mos = spa_meta_objset(vd->vdev_spa);
3242 ASSERT(vd->vdev_asize != 0);
3243 ASSERT3P(vd->vdev_checkpoint_sm, ==, NULL);
3244
3245 error = space_map_open(&vd->vdev_checkpoint_sm,
3246 mos, checkpoint_sm_obj, 0, vd->vdev_asize,
3247 vd->vdev_ashift);
3248 if (error != 0) {
3249 vdev_dbgmsg(vd, "vdev_load: space_map_open "
3250 "failed for checkpoint spacemap (obj %llu) "
3251 "[error=%d]",
3252 (u_longlong_t)checkpoint_sm_obj, error);
3253 return (error);
3254 }
3255 ASSERT3P(vd->vdev_checkpoint_sm, !=, NULL);
3256
3257 /*
3258 * Since the checkpoint_sm contains free entries
3259 * exclusively we can use space_map_allocated() to
3260 * indicate the cumulative checkpointed space that
3261 * has been freed.
3262 */
3263 vd->vdev_stat.vs_checkpoint_space =
3264 -space_map_allocated(vd->vdev_checkpoint_sm);
3265 vd->vdev_spa->spa_checkpoint_info.sci_dspace +=
3266 vd->vdev_stat.vs_checkpoint_space;
3267 } else if (error != 0) {
3268 vdev_dbgmsg(vd, "vdev_load: failed to retrieve "
3269 "checkpoint space map object from vdev ZAP "
3270 "[error=%d]", error);
3271 return (error);
3272 }
3273 }
3274
3275 /*
3276 * If this is a leaf vdev, load its DTL.
3277 */
3278 if (vd->vdev_ops->vdev_op_leaf && (error = vdev_dtl_load(vd)) != 0) {
3279 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
3280 VDEV_AUX_CORRUPT_DATA);
3281 vdev_dbgmsg(vd, "vdev_load: vdev_dtl_load failed "
3282 "[error=%d]", error);
3283 return (error);
3284 }
3285
3286 uint64_t obsolete_sm_object;
3287 error = vdev_obsolete_sm_object(vd, &obsolete_sm_object);
3288 if (error == 0 && obsolete_sm_object != 0) {
3289 objset_t *mos = vd->vdev_spa->spa_meta_objset;
3290 ASSERT(vd->vdev_asize != 0);
3291 ASSERT3P(vd->vdev_obsolete_sm, ==, NULL);
3292
3293 if ((error = space_map_open(&vd->vdev_obsolete_sm, mos,
3294 obsolete_sm_object, 0, vd->vdev_asize, 0))) {
3295 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
3296 VDEV_AUX_CORRUPT_DATA);
3297 vdev_dbgmsg(vd, "vdev_load: space_map_open failed for "
3298 "obsolete spacemap (obj %llu) [error=%d]",
3299 (u_longlong_t)obsolete_sm_object, error);
3300 return (error);
3301 }
3302 } else if (error != 0) {
3303 vdev_dbgmsg(vd, "vdev_load: failed to retrieve obsolete "
3304 "space map object from vdev ZAP [error=%d]", error);
3305 return (error);
3306 }
3307
3308 return (0);
3309 }
3310
3311 /*
3312 * The special vdev case is used for hot spares and l2cache devices. Its
3313 * sole purpose it to set the vdev state for the associated vdev. To do this,
3314 * we make sure that we can open the underlying device, then try to read the
3315 * label, and make sure that the label is sane and that it hasn't been
3316 * repurposed to another pool.
3317 */
3318 int
3319 vdev_validate_aux(vdev_t *vd)
3320 {
3321 nvlist_t *label;
3322 uint64_t guid, version;
3323 uint64_t state;
3324
3325 if (!vdev_readable(vd))
3326 return (0);
3327
3328 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) {
3329 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
3330 VDEV_AUX_CORRUPT_DATA);
3331 return (-1);
3332 }
3333
3334 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 ||
3335 !SPA_VERSION_IS_SUPPORTED(version) ||
3336 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 ||
3337 guid != vd->vdev_guid ||
3338 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) {
3339 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
3340 VDEV_AUX_CORRUPT_DATA);
3341 nvlist_free(label);
3342 return (-1);
3343 }
3344
3345 /*
3346 * We don't actually check the pool state here. If it's in fact in
3347 * use by another pool, we update this fact on the fly when requested.
3348 */
3349 nvlist_free(label);
3350 return (0);
3351 }
3352
3353 static void
3354 vdev_destroy_ms_flush_data(vdev_t *vd, dmu_tx_t *tx)
3355 {
3356 objset_t *mos = spa_meta_objset(vd->vdev_spa);
3357
3358 if (vd->vdev_top_zap == 0)
3359 return;
3360
3361 uint64_t object = 0;
3362 int err = zap_lookup(mos, vd->vdev_top_zap,
3363 VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS, sizeof (uint64_t), 1, &object);
3364 if (err == ENOENT)
3365 return;
3366 VERIFY0(err);
3367
3368 VERIFY0(dmu_object_free(mos, object, tx));
3369 VERIFY0(zap_remove(mos, vd->vdev_top_zap,
3370 VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS, tx));
3371 }
3372
3373 /*
3374 * Free the objects used to store this vdev's spacemaps, and the array
3375 * that points to them.
3376 */
3377 void
3378 vdev_destroy_spacemaps(vdev_t *vd, dmu_tx_t *tx)
3379 {
3380 if (vd->vdev_ms_array == 0)
3381 return;
3382
3383 objset_t *mos = vd->vdev_spa->spa_meta_objset;
3384 uint64_t array_count = vd->vdev_asize >> vd->vdev_ms_shift;
3385 size_t array_bytes = array_count * sizeof (uint64_t);
3386 uint64_t *smobj_array = kmem_alloc(array_bytes, KM_SLEEP);
3387 VERIFY0(dmu_read(mos, vd->vdev_ms_array, 0,
3388 array_bytes, smobj_array, 0));
3389
3390 for (uint64_t i = 0; i < array_count; i++) {
3391 uint64_t smobj = smobj_array[i];
3392 if (smobj == 0)
3393 continue;
3394
3395 space_map_free_obj(mos, smobj, tx);
3396 }
3397
3398 kmem_free(smobj_array, array_bytes);
3399 VERIFY0(dmu_object_free(mos, vd->vdev_ms_array, tx));
3400 vdev_destroy_ms_flush_data(vd, tx);
3401 vd->vdev_ms_array = 0;
3402 }
3403
3404 static void
3405 vdev_remove_empty_log(vdev_t *vd, uint64_t txg)
3406 {
3407 spa_t *spa = vd->vdev_spa;
3408
3409 ASSERT(vd->vdev_islog);
3410 ASSERT(vd == vd->vdev_top);
3411 ASSERT3U(txg, ==, spa_syncing_txg(spa));
3412
3413 dmu_tx_t *tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
3414
3415 vdev_destroy_spacemaps(vd, tx);
3416 if (vd->vdev_top_zap != 0) {
3417 vdev_destroy_unlink_zap(vd, vd->vdev_top_zap, tx);
3418 vd->vdev_top_zap = 0;
3419 }
3420
3421 dmu_tx_commit(tx);
3422 }
3423
3424 void
3425 vdev_sync_done(vdev_t *vd, uint64_t txg)
3426 {
3427 metaslab_t *msp;
3428 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg));
3429
3430 ASSERT(vdev_is_concrete(vd));
3431
3432 while ((msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))
3433 != NULL)
3434 metaslab_sync_done(msp, txg);
3435
3436 if (reassess)
3437 metaslab_sync_reassess(vd->vdev_mg);
3438 }
3439
3440 void
3441 vdev_sync(vdev_t *vd, uint64_t txg)
3442 {
3443 spa_t *spa = vd->vdev_spa;
3444 vdev_t *lvd;
3445 metaslab_t *msp;
3446
3447 ASSERT3U(txg, ==, spa->spa_syncing_txg);
3448 dmu_tx_t *tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
3449 if (range_tree_space(vd->vdev_obsolete_segments) > 0) {
3450 ASSERT(vd->vdev_removing ||
3451 vd->vdev_ops == &vdev_indirect_ops);
3452
3453 vdev_indirect_sync_obsolete(vd, tx);
3454
3455 /*
3456 * If the vdev is indirect, it can't have dirty
3457 * metaslabs or DTLs.
3458 */
3459 if (vd->vdev_ops == &vdev_indirect_ops) {
3460 ASSERT(txg_list_empty(&vd->vdev_ms_list, txg));
3461 ASSERT(txg_list_empty(&vd->vdev_dtl_list, txg));
3462 dmu_tx_commit(tx);
3463 return;
3464 }
3465 }
3466
3467 ASSERT(vdev_is_concrete(vd));
3468
3469 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0 &&
3470 !vd->vdev_removing) {
3471 ASSERT(vd == vd->vdev_top);
3472 ASSERT0(vd->vdev_indirect_config.vic_mapping_object);
3473 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset,
3474 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx);
3475 ASSERT(vd->vdev_ms_array != 0);
3476 vdev_config_dirty(vd);
3477 }
3478
3479 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) {
3480 metaslab_sync(msp, txg);
3481 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg));
3482 }
3483
3484 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL)
3485 vdev_dtl_sync(lvd, txg);
3486
3487 /*
3488 * If this is an empty log device being removed, destroy the
3489 * metadata associated with it.
3490 */
3491 if (vd->vdev_islog && vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing)
3492 vdev_remove_empty_log(vd, txg);
3493
3494 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg));
3495 dmu_tx_commit(tx);
3496 }
3497
3498 uint64_t
3499 vdev_psize_to_asize(vdev_t *vd, uint64_t psize)
3500 {
3501 return (vd->vdev_ops->vdev_op_asize(vd, psize));
3502 }
3503
3504 /*
3505 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
3506 * not be opened, and no I/O is attempted.
3507 */
3508 int
3509 vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux)
3510 {
3511 vdev_t *vd, *tvd;
3512
3513 spa_vdev_state_enter(spa, SCL_NONE);
3514
3515 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
3516 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENODEV)));
3517
3518 if (!vd->vdev_ops->vdev_op_leaf)
3519 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENOTSUP)));
3520
3521 tvd = vd->vdev_top;
3522
3523 /*
3524 * If user did a 'zpool offline -f' then make the fault persist across
3525 * reboots.
3526 */
3527 if (aux == VDEV_AUX_EXTERNAL_PERSIST) {
3528 /*
3529 * There are two kinds of forced faults: temporary and
3530 * persistent. Temporary faults go away at pool import, while
3531 * persistent faults stay set. Both types of faults can be
3532 * cleared with a zpool clear.
3533 *
3534 * We tell if a vdev is persistently faulted by looking at the
3535 * ZPOOL_CONFIG_AUX_STATE nvpair. If it's set to "external" at
3536 * import then it's a persistent fault. Otherwise, it's
3537 * temporary. We get ZPOOL_CONFIG_AUX_STATE set to "external"
3538 * by setting vd.vdev_stat.vs_aux to VDEV_AUX_EXTERNAL. This
3539 * tells vdev_config_generate() (which gets run later) to set
3540 * ZPOOL_CONFIG_AUX_STATE to "external" in the nvlist.
3541 */
3542 vd->vdev_stat.vs_aux = VDEV_AUX_EXTERNAL;
3543 vd->vdev_tmpoffline = B_FALSE;
3544 aux = VDEV_AUX_EXTERNAL;
3545 } else {
3546 vd->vdev_tmpoffline = B_TRUE;
3547 }
3548
3549 /*
3550 * We don't directly use the aux state here, but if we do a
3551 * vdev_reopen(), we need this value to be present to remember why we
3552 * were faulted.
3553 */
3554 vd->vdev_label_aux = aux;
3555
3556 /*
3557 * Faulted state takes precedence over degraded.
3558 */
3559 vd->vdev_delayed_close = B_FALSE;
3560 vd->vdev_faulted = 1ULL;
3561 vd->vdev_degraded = 0ULL;
3562 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux);
3563
3564 /*
3565 * If this device has the only valid copy of the data, then
3566 * back off and simply mark the vdev as degraded instead.
3567 */
3568 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) {
3569 vd->vdev_degraded = 1ULL;
3570 vd->vdev_faulted = 0ULL;
3571
3572 /*
3573 * If we reopen the device and it's not dead, only then do we
3574 * mark it degraded.
3575 */
3576 vdev_reopen(tvd);
3577
3578 if (vdev_readable(vd))
3579 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux);
3580 }
3581
3582 return (spa_vdev_state_exit(spa, vd, 0));
3583 }
3584
3585 /*
3586 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
3587 * user that something is wrong. The vdev continues to operate as normal as far
3588 * as I/O is concerned.
3589 */
3590 int
3591 vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux)
3592 {
3593 vdev_t *vd;
3594
3595 spa_vdev_state_enter(spa, SCL_NONE);
3596
3597 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
3598 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENODEV)));
3599
3600 if (!vd->vdev_ops->vdev_op_leaf)
3601 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENOTSUP)));
3602
3603 /*
3604 * If the vdev is already faulted, then don't do anything.
3605 */
3606 if (vd->vdev_faulted || vd->vdev_degraded)
3607 return (spa_vdev_state_exit(spa, NULL, 0));
3608
3609 vd->vdev_degraded = 1ULL;
3610 if (!vdev_is_dead(vd))
3611 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
3612 aux);
3613
3614 return (spa_vdev_state_exit(spa, vd, 0));
3615 }
3616
3617 /*
3618 * Online the given vdev.
3619 *
3620 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
3621 * spare device should be detached when the device finishes resilvering.
3622 * Second, the online should be treated like a 'test' online case, so no FMA
3623 * events are generated if the device fails to open.
3624 */
3625 int
3626 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
3627 {
3628 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev;
3629 boolean_t wasoffline;
3630 vdev_state_t oldstate;
3631
3632 spa_vdev_state_enter(spa, SCL_NONE);
3633
3634 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
3635 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENODEV)));
3636
3637 if (!vd->vdev_ops->vdev_op_leaf)
3638 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENOTSUP)));
3639
3640 wasoffline = (vd->vdev_offline || vd->vdev_tmpoffline);
3641 oldstate = vd->vdev_state;
3642
3643 tvd = vd->vdev_top;
3644 vd->vdev_offline = B_FALSE;
3645 vd->vdev_tmpoffline = B_FALSE;
3646 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE);
3647 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT);
3648
3649 /* XXX - L2ARC 1.0 does not support expansion */
3650 if (!vd->vdev_aux) {
3651 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
3652 pvd->vdev_expanding = !!((flags & ZFS_ONLINE_EXPAND) ||
3653 spa->spa_autoexpand);
3654 vd->vdev_expansion_time = gethrestime_sec();
3655 }
3656
3657 vdev_reopen(tvd);
3658 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE;
3659
3660 if (!vd->vdev_aux) {
3661 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
3662 pvd->vdev_expanding = B_FALSE;
3663 }
3664
3665 if (newstate)
3666 *newstate = vd->vdev_state;
3667 if ((flags & ZFS_ONLINE_UNSPARE) &&
3668 !vdev_is_dead(vd) && vd->vdev_parent &&
3669 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3670 vd->vdev_parent->vdev_child[0] == vd)
3671 vd->vdev_unspare = B_TRUE;
3672
3673 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) {
3674
3675 /* XXX - L2ARC 1.0 does not support expansion */
3676 if (vd->vdev_aux)
3677 return (spa_vdev_state_exit(spa, vd, ENOTSUP));
3678 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
3679 }
3680
3681 /* Restart initializing if necessary */
3682 mutex_enter(&vd->vdev_initialize_lock);
3683 if (vdev_writeable(vd) &&
3684 vd->vdev_initialize_thread == NULL &&
3685 vd->vdev_initialize_state == VDEV_INITIALIZE_ACTIVE) {
3686 (void) vdev_initialize(vd);
3687 }
3688 mutex_exit(&vd->vdev_initialize_lock);
3689
3690 /*
3691 * Restart trimming if necessary. We do not restart trimming for cache
3692 * devices here. This is triggered by l2arc_rebuild_vdev()
3693 * asynchronously for the whole device or in l2arc_evict() as it evicts
3694 * space for upcoming writes.
3695 */
3696 mutex_enter(&vd->vdev_trim_lock);
3697 if (vdev_writeable(vd) && !vd->vdev_isl2cache &&
3698 vd->vdev_trim_thread == NULL &&
3699 vd->vdev_trim_state == VDEV_TRIM_ACTIVE) {
3700 (void) vdev_trim(vd, vd->vdev_trim_rate, vd->vdev_trim_partial,
3701 vd->vdev_trim_secure);
3702 }
3703 mutex_exit(&vd->vdev_trim_lock);
3704
3705 if (wasoffline ||
3706 (oldstate < VDEV_STATE_DEGRADED &&
3707 vd->vdev_state >= VDEV_STATE_DEGRADED))
3708 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_ONLINE);
3709
3710 return (spa_vdev_state_exit(spa, vd, 0));
3711 }
3712
3713 static int
3714 vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags)
3715 {
3716 vdev_t *vd, *tvd;
3717 int error = 0;
3718 uint64_t generation;
3719 metaslab_group_t *mg;
3720
3721 top:
3722 spa_vdev_state_enter(spa, SCL_ALLOC);
3723
3724 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
3725 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENODEV)));
3726
3727 if (!vd->vdev_ops->vdev_op_leaf)
3728 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENOTSUP)));
3729
3730 tvd = vd->vdev_top;
3731 mg = tvd->vdev_mg;
3732 generation = spa->spa_config_generation + 1;
3733
3734 /*
3735 * If the device isn't already offline, try to offline it.
3736 */
3737 if (!vd->vdev_offline) {
3738 /*
3739 * If this device has the only valid copy of some data,
3740 * don't allow it to be offlined. Log devices are always
3741 * expendable.
3742 */
3743 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
3744 vdev_dtl_required(vd))
3745 return (spa_vdev_state_exit(spa, NULL,
3746 SET_ERROR(EBUSY)));
3747
3748 /*
3749 * If the top-level is a slog and it has had allocations
3750 * then proceed. We check that the vdev's metaslab group
3751 * is not NULL since it's possible that we may have just
3752 * added this vdev but not yet initialized its metaslabs.
3753 */
3754 if (tvd->vdev_islog && mg != NULL) {
3755 /*
3756 * Prevent any future allocations.
3757 */
3758 metaslab_group_passivate(mg);
3759 (void) spa_vdev_state_exit(spa, vd, 0);
3760
3761 error = spa_reset_logs(spa);
3762
3763 /*
3764 * If the log device was successfully reset but has
3765 * checkpointed data, do not offline it.
3766 */
3767 if (error == 0 &&
3768 tvd->vdev_checkpoint_sm != NULL) {
3769 ASSERT3U(space_map_allocated(
3770 tvd->vdev_checkpoint_sm), !=, 0);
3771 error = ZFS_ERR_CHECKPOINT_EXISTS;
3772 }
3773
3774 spa_vdev_state_enter(spa, SCL_ALLOC);
3775
3776 /*
3777 * Check to see if the config has changed.
3778 */
3779 if (error || generation != spa->spa_config_generation) {
3780 metaslab_group_activate(mg);
3781 if (error)
3782 return (spa_vdev_state_exit(spa,
3783 vd, error));
3784 (void) spa_vdev_state_exit(spa, vd, 0);
3785 goto top;
3786 }
3787 ASSERT0(tvd->vdev_stat.vs_alloc);
3788 }
3789
3790 /*
3791 * Offline this device and reopen its top-level vdev.
3792 * If the top-level vdev is a log device then just offline
3793 * it. Otherwise, if this action results in the top-level
3794 * vdev becoming unusable, undo it and fail the request.
3795 */
3796 vd->vdev_offline = B_TRUE;
3797 vdev_reopen(tvd);
3798
3799 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
3800 vdev_is_dead(tvd)) {
3801 vd->vdev_offline = B_FALSE;
3802 vdev_reopen(tvd);
3803 return (spa_vdev_state_exit(spa, NULL,
3804 SET_ERROR(EBUSY)));
3805 }
3806
3807 /*
3808 * Add the device back into the metaslab rotor so that
3809 * once we online the device it's open for business.
3810 */
3811 if (tvd->vdev_islog && mg != NULL)
3812 metaslab_group_activate(mg);
3813 }
3814
3815 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY);
3816
3817 return (spa_vdev_state_exit(spa, vd, 0));
3818 }
3819
3820 int
3821 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags)
3822 {
3823 int error;
3824
3825 mutex_enter(&spa->spa_vdev_top_lock);
3826 error = vdev_offline_locked(spa, guid, flags);
3827 mutex_exit(&spa->spa_vdev_top_lock);
3828
3829 return (error);
3830 }
3831
3832 /*
3833 * Clear the error counts associated with this vdev. Unlike vdev_online() and
3834 * vdev_offline(), we assume the spa config is locked. We also clear all
3835 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
3836 */
3837 void
3838 vdev_clear(spa_t *spa, vdev_t *vd)
3839 {
3840 vdev_t *rvd = spa->spa_root_vdev;
3841
3842 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
3843
3844 if (vd == NULL)
3845 vd = rvd;
3846
3847 vd->vdev_stat.vs_read_errors = 0;
3848 vd->vdev_stat.vs_write_errors = 0;
3849 vd->vdev_stat.vs_checksum_errors = 0;
3850 vd->vdev_stat.vs_slow_ios = 0;
3851
3852 for (int c = 0; c < vd->vdev_children; c++)
3853 vdev_clear(spa, vd->vdev_child[c]);
3854
3855 /*
3856 * It makes no sense to "clear" an indirect vdev.
3857 */
3858 if (!vdev_is_concrete(vd))
3859 return;
3860
3861 /*
3862 * If we're in the FAULTED state or have experienced failed I/O, then
3863 * clear the persistent state and attempt to reopen the device. We
3864 * also mark the vdev config dirty, so that the new faulted state is
3865 * written out to disk.
3866 */
3867 if (vd->vdev_faulted || vd->vdev_degraded ||
3868 !vdev_readable(vd) || !vdev_writeable(vd)) {
3869 /*
3870 * When reopening in response to a clear event, it may be due to
3871 * a fmadm repair request. In this case, if the device is
3872 * still broken, we want to still post the ereport again.
3873 */
3874 vd->vdev_forcefault = B_TRUE;
3875
3876 vd->vdev_faulted = vd->vdev_degraded = 0ULL;
3877 vd->vdev_cant_read = B_FALSE;
3878 vd->vdev_cant_write = B_FALSE;
3879 vd->vdev_stat.vs_aux = 0;
3880
3881 vdev_reopen(vd == rvd ? rvd : vd->vdev_top);
3882
3883 vd->vdev_forcefault = B_FALSE;
3884
3885 if (vd != rvd && vdev_writeable(vd->vdev_top))
3886 vdev_state_dirty(vd->vdev_top);
3887
3888 /* If a resilver isn't required, check if vdevs can be culled */
3889 if (vd->vdev_aux == NULL && !vdev_is_dead(vd) &&
3890 !dsl_scan_resilvering(spa->spa_dsl_pool) &&
3891 !dsl_scan_resilver_scheduled(spa->spa_dsl_pool))
3892 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
3893
3894 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_CLEAR);
3895 }
3896
3897 /*
3898 * When clearing a FMA-diagnosed fault, we always want to
3899 * unspare the device, as we assume that the original spare was
3900 * done in response to the FMA fault.
3901 */
3902 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL &&
3903 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3904 vd->vdev_parent->vdev_child[0] == vd)
3905 vd->vdev_unspare = B_TRUE;
3906 }
3907
3908 boolean_t
3909 vdev_is_dead(vdev_t *vd)
3910 {
3911 /*
3912 * Holes and missing devices are always considered "dead".
3913 * This simplifies the code since we don't have to check for
3914 * these types of devices in the various code paths.
3915 * Instead we rely on the fact that we skip over dead devices
3916 * before issuing I/O to them.
3917 */
3918 return (vd->vdev_state < VDEV_STATE_DEGRADED ||
3919 vd->vdev_ops == &vdev_hole_ops ||
3920 vd->vdev_ops == &vdev_missing_ops);
3921 }
3922
3923 boolean_t
3924 vdev_readable(vdev_t *vd)
3925 {
3926 return (!vdev_is_dead(vd) && !vd->vdev_cant_read);
3927 }
3928
3929 boolean_t
3930 vdev_writeable(vdev_t *vd)
3931 {
3932 return (!vdev_is_dead(vd) && !vd->vdev_cant_write &&
3933 vdev_is_concrete(vd));
3934 }
3935
3936 boolean_t
3937 vdev_allocatable(vdev_t *vd)
3938 {
3939 uint64_t state = vd->vdev_state;
3940
3941 /*
3942 * We currently allow allocations from vdevs which may be in the
3943 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
3944 * fails to reopen then we'll catch it later when we're holding
3945 * the proper locks. Note that we have to get the vdev state
3946 * in a local variable because although it changes atomically,
3947 * we're asking two separate questions about it.
3948 */
3949 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) &&
3950 !vd->vdev_cant_write && vdev_is_concrete(vd) &&
3951 vd->vdev_mg->mg_initialized);
3952 }
3953
3954 boolean_t
3955 vdev_accessible(vdev_t *vd, zio_t *zio)
3956 {
3957 ASSERT(zio->io_vd == vd);
3958
3959 if (vdev_is_dead(vd) || vd->vdev_remove_wanted)
3960 return (B_FALSE);
3961
3962 if (zio->io_type == ZIO_TYPE_READ)
3963 return (!vd->vdev_cant_read);
3964
3965 if (zio->io_type == ZIO_TYPE_WRITE)
3966 return (!vd->vdev_cant_write);
3967
3968 return (B_TRUE);
3969 }
3970
3971 static void
3972 vdev_get_child_stat(vdev_t *cvd, vdev_stat_t *vs, vdev_stat_t *cvs)
3973 {
3974 for (int t = 0; t < VS_ZIO_TYPES; t++) {
3975 vs->vs_ops[t] += cvs->vs_ops[t];
3976 vs->vs_bytes[t] += cvs->vs_bytes[t];
3977 }
3978
3979 cvs->vs_scan_removing = cvd->vdev_removing;
3980 }
3981
3982 /*
3983 * Get extended stats
3984 */
3985 static void
3986 vdev_get_child_stat_ex(vdev_t *cvd, vdev_stat_ex_t *vsx, vdev_stat_ex_t *cvsx)
3987 {
3988 int t, b;
3989 for (t = 0; t < ZIO_TYPES; t++) {
3990 for (b = 0; b < ARRAY_SIZE(vsx->vsx_disk_histo[0]); b++)
3991 vsx->vsx_disk_histo[t][b] += cvsx->vsx_disk_histo[t][b];
3992
3993 for (b = 0; b < ARRAY_SIZE(vsx->vsx_total_histo[0]); b++) {
3994 vsx->vsx_total_histo[t][b] +=
3995 cvsx->vsx_total_histo[t][b];
3996 }
3997 }
3998
3999 for (t = 0; t < ZIO_PRIORITY_NUM_QUEUEABLE; t++) {
4000 for (b = 0; b < ARRAY_SIZE(vsx->vsx_queue_histo[0]); b++) {
4001 vsx->vsx_queue_histo[t][b] +=
4002 cvsx->vsx_queue_histo[t][b];
4003 }
4004 vsx->vsx_active_queue[t] += cvsx->vsx_active_queue[t];
4005 vsx->vsx_pend_queue[t] += cvsx->vsx_pend_queue[t];
4006
4007 for (b = 0; b < ARRAY_SIZE(vsx->vsx_ind_histo[0]); b++)
4008 vsx->vsx_ind_histo[t][b] += cvsx->vsx_ind_histo[t][b];
4009
4010 for (b = 0; b < ARRAY_SIZE(vsx->vsx_agg_histo[0]); b++)
4011 vsx->vsx_agg_histo[t][b] += cvsx->vsx_agg_histo[t][b];
4012 }
4013
4014 }
4015
4016 boolean_t
4017 vdev_is_spacemap_addressable(vdev_t *vd)
4018 {
4019 if (spa_feature_is_active(vd->vdev_spa, SPA_FEATURE_SPACEMAP_V2))
4020 return (B_TRUE);
4021
4022 /*
4023 * If double-word space map entries are not enabled we assume
4024 * 47 bits of the space map entry are dedicated to the entry's
4025 * offset (see SM_OFFSET_BITS in space_map.h). We then use that
4026 * to calculate the maximum address that can be described by a
4027 * space map entry for the given device.
4028 */
4029 uint64_t shift = vd->vdev_ashift + SM_OFFSET_BITS;
4030
4031 if (shift >= 63) /* detect potential overflow */
4032 return (B_TRUE);
4033
4034 return (vd->vdev_asize < (1ULL << shift));
4035 }
4036
4037 /*
4038 * Get statistics for the given vdev.
4039 */
4040 static void
4041 vdev_get_stats_ex_impl(vdev_t *vd, vdev_stat_t *vs, vdev_stat_ex_t *vsx)
4042 {
4043 int t;
4044 /*
4045 * If we're getting stats on the root vdev, aggregate the I/O counts
4046 * over all top-level vdevs (i.e. the direct children of the root).
4047 */
4048 if (!vd->vdev_ops->vdev_op_leaf) {
4049 if (vs) {
4050 memset(vs->vs_ops, 0, sizeof (vs->vs_ops));
4051 memset(vs->vs_bytes, 0, sizeof (vs->vs_bytes));
4052 }
4053 if (vsx)
4054 memset(vsx, 0, sizeof (*vsx));
4055
4056 for (int c = 0; c < vd->vdev_children; c++) {
4057 vdev_t *cvd = vd->vdev_child[c];
4058 vdev_stat_t *cvs = &cvd->vdev_stat;
4059 vdev_stat_ex_t *cvsx = &cvd->vdev_stat_ex;
4060
4061 vdev_get_stats_ex_impl(cvd, cvs, cvsx);
4062 if (vs)
4063 vdev_get_child_stat(cvd, vs, cvs);
4064 if (vsx)
4065 vdev_get_child_stat_ex(cvd, vsx, cvsx);
4066
4067 }
4068 } else {
4069 /*
4070 * We're a leaf. Just copy our ZIO active queue stats in. The
4071 * other leaf stats are updated in vdev_stat_update().
4072 */
4073 if (!vsx)
4074 return;
4075
4076 memcpy(vsx, &vd->vdev_stat_ex, sizeof (vd->vdev_stat_ex));
4077
4078 for (t = 0; t < ARRAY_SIZE(vd->vdev_queue.vq_class); t++) {
4079 vsx->vsx_active_queue[t] =
4080 vd->vdev_queue.vq_class[t].vqc_active;
4081 vsx->vsx_pend_queue[t] = avl_numnodes(
4082 &vd->vdev_queue.vq_class[t].vqc_queued_tree);
4083 }
4084 }
4085 }
4086
4087 void
4088 vdev_get_stats_ex(vdev_t *vd, vdev_stat_t *vs, vdev_stat_ex_t *vsx)
4089 {
4090 vdev_t *tvd = vd->vdev_top;
4091 mutex_enter(&vd->vdev_stat_lock);
4092 if (vs) {
4093 bcopy(&vd->vdev_stat, vs, sizeof (*vs));
4094 vs->vs_timestamp = gethrtime() - vs->vs_timestamp;
4095 vs->vs_state = vd->vdev_state;
4096 vs->vs_rsize = vdev_get_min_asize(vd);
4097
4098 if (vd->vdev_ops->vdev_op_leaf) {
4099 vs->vs_rsize += VDEV_LABEL_START_SIZE +
4100 VDEV_LABEL_END_SIZE;
4101 /*
4102 * Report initializing progress. Since we don't
4103 * have the initializing locks held, this is only
4104 * an estimate (although a fairly accurate one).
4105 */
4106 vs->vs_initialize_bytes_done =
4107 vd->vdev_initialize_bytes_done;
4108 vs->vs_initialize_bytes_est =
4109 vd->vdev_initialize_bytes_est;
4110 vs->vs_initialize_state = vd->vdev_initialize_state;
4111 vs->vs_initialize_action_time =
4112 vd->vdev_initialize_action_time;
4113
4114 /*
4115 * Report manual TRIM progress. Since we don't have
4116 * the manual TRIM locks held, this is only an
4117 * estimate (although fairly accurate one).
4118 */
4119 vs->vs_trim_notsup = !vd->vdev_has_trim;
4120 vs->vs_trim_bytes_done = vd->vdev_trim_bytes_done;
4121 vs->vs_trim_bytes_est = vd->vdev_trim_bytes_est;
4122 vs->vs_trim_state = vd->vdev_trim_state;
4123 vs->vs_trim_action_time = vd->vdev_trim_action_time;
4124
4125 /* Set when there is a deferred resilver. */
4126 vs->vs_resilver_deferred = vd->vdev_resilver_deferred;
4127 }
4128
4129 /*
4130 * Report expandable space on top-level, non-auxiliary devices
4131 * only. The expandable space is reported in terms of metaslab
4132 * sized units since that determines how much space the pool
4133 * can expand.
4134 */
4135 if (vd->vdev_aux == NULL && tvd != NULL) {
4136 vs->vs_esize = P2ALIGN(
4137 vd->vdev_max_asize - vd->vdev_asize,
4138 1ULL << tvd->vdev_ms_shift);
4139 }
4140
4141 vs->vs_configured_ashift = vd->vdev_top != NULL
4142 ? vd->vdev_top->vdev_ashift : vd->vdev_ashift;
4143 vs->vs_logical_ashift = vd->vdev_logical_ashift;
4144 vs->vs_physical_ashift = vd->vdev_physical_ashift;
4145
4146 /*
4147 * Report fragmentation and rebuild progress for top-level,
4148 * non-auxiliary, concrete devices.
4149 */
4150 if (vd->vdev_aux == NULL && vd == vd->vdev_top &&
4151 vdev_is_concrete(vd)) {
4152 vs->vs_fragmentation = (vd->vdev_mg != NULL) ?
4153 vd->vdev_mg->mg_fragmentation : 0;
4154 }
4155 }
4156
4157 vdev_get_stats_ex_impl(vd, vs, vsx);
4158 mutex_exit(&vd->vdev_stat_lock);
4159 }
4160
4161 void
4162 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
4163 {
4164 return (vdev_get_stats_ex(vd, vs, NULL));
4165 }
4166
4167 void
4168 vdev_clear_stats(vdev_t *vd)
4169 {
4170 mutex_enter(&vd->vdev_stat_lock);
4171 vd->vdev_stat.vs_space = 0;
4172 vd->vdev_stat.vs_dspace = 0;
4173 vd->vdev_stat.vs_alloc = 0;
4174 mutex_exit(&vd->vdev_stat_lock);
4175 }
4176
4177 void
4178 vdev_scan_stat_init(vdev_t *vd)
4179 {
4180 vdev_stat_t *vs = &vd->vdev_stat;
4181
4182 for (int c = 0; c < vd->vdev_children; c++)
4183 vdev_scan_stat_init(vd->vdev_child[c]);
4184
4185 mutex_enter(&vd->vdev_stat_lock);
4186 vs->vs_scan_processed = 0;
4187 mutex_exit(&vd->vdev_stat_lock);
4188 }
4189
4190 void
4191 vdev_stat_update(zio_t *zio, uint64_t psize)
4192 {
4193 spa_t *spa = zio->io_spa;
4194 vdev_t *rvd = spa->spa_root_vdev;
4195 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd;
4196 vdev_t *pvd;
4197 uint64_t txg = zio->io_txg;
4198 vdev_stat_t *vs = &vd->vdev_stat;
4199 vdev_stat_ex_t *vsx = &vd->vdev_stat_ex;
4200 zio_type_t type = zio->io_type;
4201 int flags = zio->io_flags;
4202
4203 /*
4204 * If this i/o is a gang leader, it didn't do any actual work.
4205 */
4206 if (zio->io_gang_tree)
4207 return;
4208
4209 if (zio->io_error == 0) {
4210 /*
4211 * If this is a root i/o, don't count it -- we've already
4212 * counted the top-level vdevs, and vdev_get_stats() will
4213 * aggregate them when asked. This reduces contention on
4214 * the root vdev_stat_lock and implicitly handles blocks
4215 * that compress away to holes, for which there is no i/o.
4216 * (Holes never create vdev children, so all the counters
4217 * remain zero, which is what we want.)
4218 *
4219 * Note: this only applies to successful i/o (io_error == 0)
4220 * because unlike i/o counts, errors are not additive.
4221 * When reading a ditto block, for example, failure of
4222 * one top-level vdev does not imply a root-level error.
4223 */
4224 if (vd == rvd)
4225 return;
4226
4227 ASSERT(vd == zio->io_vd);
4228
4229 if (flags & ZIO_FLAG_IO_BYPASS)
4230 return;
4231
4232 mutex_enter(&vd->vdev_stat_lock);
4233
4234 if (flags & ZIO_FLAG_IO_REPAIR) {
4235 /*
4236 * Repair is the result of a resilver issued by the
4237 * scan thread (spa_sync).
4238 */
4239 if (flags & ZIO_FLAG_SCAN_THREAD) {
4240 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
4241 dsl_scan_phys_t *scn_phys = &scn->scn_phys;
4242 uint64_t *processed = &scn_phys->scn_processed;
4243
4244 if (vd->vdev_ops->vdev_op_leaf)
4245 atomic_add_64(processed, psize);
4246 vs->vs_scan_processed += psize;
4247 }
4248
4249 /*
4250 * Repair is the result of a rebuild issued by the
4251 * rebuild thread (vdev_rebuild_thread).
4252 */
4253 if (zio->io_priority == ZIO_PRIORITY_REBUILD) {
4254 vdev_t *tvd = vd->vdev_top;
4255 vdev_rebuild_t *vr = &tvd->vdev_rebuild_config;
4256 vdev_rebuild_phys_t *vrp = &vr->vr_rebuild_phys;
4257 uint64_t *rebuilt = &vrp->vrp_bytes_rebuilt;
4258
4259 if (vd->vdev_ops->vdev_op_leaf)
4260 atomic_add_64(rebuilt, psize);
4261 vs->vs_rebuild_processed += psize;
4262 }
4263
4264 if (flags & ZIO_FLAG_SELF_HEAL)
4265 vs->vs_self_healed += psize;
4266 }
4267
4268 /*
4269 * The bytes/ops/histograms are recorded at the leaf level and
4270 * aggregated into the higher level vdevs in vdev_get_stats().
4271 */
4272 if (vd->vdev_ops->vdev_op_leaf &&
4273 (zio->io_priority < ZIO_PRIORITY_NUM_QUEUEABLE)) {
4274 zio_type_t vs_type = type;
4275 zio_priority_t priority = zio->io_priority;
4276
4277 /*
4278 * TRIM ops and bytes are reported to user space as
4279 * ZIO_TYPE_IOCTL. This is done to preserve the
4280 * vdev_stat_t structure layout for user space.
4281 */
4282 if (type == ZIO_TYPE_TRIM)
4283 vs_type = ZIO_TYPE_IOCTL;
4284
4285 /*
4286 * Solely for the purposes of 'zpool iostat -lqrw'
4287 * reporting use the priority to catagorize the IO.
4288 * Only the following are reported to user space:
4289 *
4290 * ZIO_PRIORITY_SYNC_READ,
4291 * ZIO_PRIORITY_SYNC_WRITE,
4292 * ZIO_PRIORITY_ASYNC_READ,
4293 * ZIO_PRIORITY_ASYNC_WRITE,
4294 * ZIO_PRIORITY_SCRUB,
4295 * ZIO_PRIORITY_TRIM.
4296 */
4297 if (priority == ZIO_PRIORITY_REBUILD) {
4298 priority = ((type == ZIO_TYPE_WRITE) ?
4299 ZIO_PRIORITY_ASYNC_WRITE :
4300 ZIO_PRIORITY_SCRUB);
4301 } else if (priority == ZIO_PRIORITY_INITIALIZING) {
4302 ASSERT3U(type, ==, ZIO_TYPE_WRITE);
4303 priority = ZIO_PRIORITY_ASYNC_WRITE;
4304 } else if (priority == ZIO_PRIORITY_REMOVAL) {
4305 priority = ((type == ZIO_TYPE_WRITE) ?
4306 ZIO_PRIORITY_ASYNC_WRITE :
4307 ZIO_PRIORITY_ASYNC_READ);
4308 }
4309
4310 vs->vs_ops[vs_type]++;
4311 vs->vs_bytes[vs_type] += psize;
4312
4313 if (flags & ZIO_FLAG_DELEGATED) {
4314 vsx->vsx_agg_histo[priority]
4315 [RQ_HISTO(zio->io_size)]++;
4316 } else {
4317 vsx->vsx_ind_histo[priority]
4318 [RQ_HISTO(zio->io_size)]++;
4319 }
4320
4321 if (zio->io_delta && zio->io_delay) {
4322 vsx->vsx_queue_histo[priority]
4323 [L_HISTO(zio->io_delta - zio->io_delay)]++;
4324 vsx->vsx_disk_histo[type]
4325 [L_HISTO(zio->io_delay)]++;
4326 vsx->vsx_total_histo[type]
4327 [L_HISTO(zio->io_delta)]++;
4328 }
4329 }
4330
4331 mutex_exit(&vd->vdev_stat_lock);
4332 return;
4333 }
4334
4335 if (flags & ZIO_FLAG_SPECULATIVE)
4336 return;
4337
4338 /*
4339 * If this is an I/O error that is going to be retried, then ignore the
4340 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
4341 * hard errors, when in reality they can happen for any number of
4342 * innocuous reasons (bus resets, MPxIO link failure, etc).
4343 */
4344 if (zio->io_error == EIO &&
4345 !(zio->io_flags & ZIO_FLAG_IO_RETRY))
4346 return;
4347
4348 /*
4349 * Intent logs writes won't propagate their error to the root
4350 * I/O so don't mark these types of failures as pool-level
4351 * errors.
4352 */
4353 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE))
4354 return;
4355
4356 if (spa->spa_load_state == SPA_LOAD_NONE &&
4357 type == ZIO_TYPE_WRITE && txg != 0 &&
4358 (!(flags & ZIO_FLAG_IO_REPAIR) ||
4359 (flags & ZIO_FLAG_SCAN_THREAD) ||
4360 spa->spa_claiming)) {
4361 /*
4362 * This is either a normal write (not a repair), or it's
4363 * a repair induced by the scrub thread, or it's a repair
4364 * made by zil_claim() during spa_load() in the first txg.
4365 * In the normal case, we commit the DTL change in the same
4366 * txg as the block was born. In the scrub-induced repair
4367 * case, we know that scrubs run in first-pass syncing context,
4368 * so we commit the DTL change in spa_syncing_txg(spa).
4369 * In the zil_claim() case, we commit in spa_first_txg(spa).
4370 *
4371 * We currently do not make DTL entries for failed spontaneous
4372 * self-healing writes triggered by normal (non-scrubbing)
4373 * reads, because we have no transactional context in which to
4374 * do so -- and it's not clear that it'd be desirable anyway.
4375 */
4376 if (vd->vdev_ops->vdev_op_leaf) {
4377 uint64_t commit_txg = txg;
4378 if (flags & ZIO_FLAG_SCAN_THREAD) {
4379 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
4380 ASSERT(spa_sync_pass(spa) == 1);
4381 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1);
4382 commit_txg = spa_syncing_txg(spa);
4383 } else if (spa->spa_claiming) {
4384 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
4385 commit_txg = spa_first_txg(spa);
4386 }
4387 ASSERT(commit_txg >= spa_syncing_txg(spa));
4388 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1))
4389 return;
4390 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
4391 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1);
4392 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg);
4393 }
4394 if (vd != rvd)
4395 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1);
4396 }
4397 }
4398
4399 int64_t
4400 vdev_deflated_space(vdev_t *vd, int64_t space)
4401 {
4402 ASSERT((space & (SPA_MINBLOCKSIZE-1)) == 0);
4403 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache);
4404
4405 return ((space >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio);
4406 }
4407
4408 /*
4409 * Update the in-core space usage stats for this vdev, its metaslab class,
4410 * and the root vdev.
4411 */
4412 void
4413 vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta,
4414 int64_t space_delta)
4415 {
4416 int64_t dspace_delta;
4417 spa_t *spa = vd->vdev_spa;
4418 vdev_t *rvd = spa->spa_root_vdev;
4419
4420 ASSERT(vd == vd->vdev_top);
4421
4422 /*
4423 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
4424 * factor. We must calculate this here and not at the root vdev
4425 * because the root vdev's psize-to-asize is simply the max of its
4426 * children's, thus not accurate enough for us.
4427 */
4428 dspace_delta = vdev_deflated_space(vd, space_delta);
4429
4430 mutex_enter(&vd->vdev_stat_lock);
4431 /* ensure we won't underflow */
4432 if (alloc_delta < 0) {
4433 ASSERT3U(vd->vdev_stat.vs_alloc, >=, -alloc_delta);
4434 }
4435
4436 vd->vdev_stat.vs_alloc += alloc_delta;
4437 vd->vdev_stat.vs_space += space_delta;
4438 vd->vdev_stat.vs_dspace += dspace_delta;
4439 mutex_exit(&vd->vdev_stat_lock);
4440
4441 /* every class but log contributes to root space stats */
4442 if (vd->vdev_mg != NULL && !vd->vdev_islog) {
4443 ASSERT(!vd->vdev_isl2cache);
4444 mutex_enter(&rvd->vdev_stat_lock);
4445 rvd->vdev_stat.vs_alloc += alloc_delta;
4446 rvd->vdev_stat.vs_space += space_delta;
4447 rvd->vdev_stat.vs_dspace += dspace_delta;
4448 mutex_exit(&rvd->vdev_stat_lock);
4449 }
4450 /* Note: metaslab_class_space_update moved to metaslab_space_update */
4451 }
4452
4453 /*
4454 * Mark a top-level vdev's config as dirty, placing it on the dirty list
4455 * so that it will be written out next time the vdev configuration is synced.
4456 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
4457 */
4458 void
4459 vdev_config_dirty(vdev_t *vd)
4460 {
4461 spa_t *spa = vd->vdev_spa;
4462 vdev_t *rvd = spa->spa_root_vdev;
4463 int c;
4464
4465 ASSERT(spa_writeable(spa));
4466
4467 /*
4468 * If this is an aux vdev (as with l2cache and spare devices), then we
4469 * update the vdev config manually and set the sync flag.
4470 */
4471 if (vd->vdev_aux != NULL) {
4472 spa_aux_vdev_t *sav = vd->vdev_aux;
4473 nvlist_t **aux;
4474 uint_t naux;
4475
4476 for (c = 0; c < sav->sav_count; c++) {
4477 if (sav->sav_vdevs[c] == vd)
4478 break;
4479 }
4480
4481 if (c == sav->sav_count) {
4482 /*
4483 * We're being removed. There's nothing more to do.
4484 */
4485 ASSERT(sav->sav_sync == B_TRUE);
4486 return;
4487 }
4488
4489 sav->sav_sync = B_TRUE;
4490
4491 if (nvlist_lookup_nvlist_array(sav->sav_config,
4492 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) {
4493 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
4494 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0);
4495 }
4496
4497 ASSERT(c < naux);
4498
4499 /*
4500 * Setting the nvlist in the middle if the array is a little
4501 * sketchy, but it will work.
4502 */
4503 nvlist_free(aux[c]);
4504 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0);
4505
4506 return;
4507 }
4508
4509 /*
4510 * The dirty list is protected by the SCL_CONFIG lock. The caller
4511 * must either hold SCL_CONFIG as writer, or must be the sync thread
4512 * (which holds SCL_CONFIG as reader). There's only one sync thread,
4513 * so this is sufficient to ensure mutual exclusion.
4514 */
4515 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
4516 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
4517 spa_config_held(spa, SCL_CONFIG, RW_READER)));
4518
4519 if (vd == rvd) {
4520 for (c = 0; c < rvd->vdev_children; c++)
4521 vdev_config_dirty(rvd->vdev_child[c]);
4522 } else {
4523 ASSERT(vd == vd->vdev_top);
4524
4525 if (!list_link_active(&vd->vdev_config_dirty_node) &&
4526 vdev_is_concrete(vd)) {
4527 list_insert_head(&spa->spa_config_dirty_list, vd);
4528 }
4529 }
4530 }
4531
4532 void
4533 vdev_config_clean(vdev_t *vd)
4534 {
4535 spa_t *spa = vd->vdev_spa;
4536
4537 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
4538 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
4539 spa_config_held(spa, SCL_CONFIG, RW_READER)));
4540
4541 ASSERT(list_link_active(&vd->vdev_config_dirty_node));
4542 list_remove(&spa->spa_config_dirty_list, vd);
4543 }
4544
4545 /*
4546 * Mark a top-level vdev's state as dirty, so that the next pass of
4547 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
4548 * the state changes from larger config changes because they require
4549 * much less locking, and are often needed for administrative actions.
4550 */
4551 void
4552 vdev_state_dirty(vdev_t *vd)
4553 {
4554 spa_t *spa = vd->vdev_spa;
4555
4556 ASSERT(spa_writeable(spa));
4557 ASSERT(vd == vd->vdev_top);
4558
4559 /*
4560 * The state list is protected by the SCL_STATE lock. The caller
4561 * must either hold SCL_STATE as writer, or must be the sync thread
4562 * (which holds SCL_STATE as reader). There's only one sync thread,
4563 * so this is sufficient to ensure mutual exclusion.
4564 */
4565 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
4566 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
4567 spa_config_held(spa, SCL_STATE, RW_READER)));
4568
4569 if (!list_link_active(&vd->vdev_state_dirty_node) &&
4570 vdev_is_concrete(vd))
4571 list_insert_head(&spa->spa_state_dirty_list, vd);
4572 }
4573
4574 void
4575 vdev_state_clean(vdev_t *vd)
4576 {
4577 spa_t *spa = vd->vdev_spa;
4578
4579 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
4580 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
4581 spa_config_held(spa, SCL_STATE, RW_READER)));
4582
4583 ASSERT(list_link_active(&vd->vdev_state_dirty_node));
4584 list_remove(&spa->spa_state_dirty_list, vd);
4585 }
4586
4587 /*
4588 * Propagate vdev state up from children to parent.
4589 */
4590 void
4591 vdev_propagate_state(vdev_t *vd)
4592 {
4593 spa_t *spa = vd->vdev_spa;
4594 vdev_t *rvd = spa->spa_root_vdev;
4595 int degraded = 0, faulted = 0;
4596 int corrupted = 0;
4597 vdev_t *child;
4598
4599 if (vd->vdev_children > 0) {
4600 for (int c = 0; c < vd->vdev_children; c++) {
4601 child = vd->vdev_child[c];
4602
4603 /*
4604 * Don't factor holes or indirect vdevs into the
4605 * decision.
4606 */
4607 if (!vdev_is_concrete(child))
4608 continue;
4609
4610 if (!vdev_readable(child) ||
4611 (!vdev_writeable(child) && spa_writeable(spa))) {
4612 /*
4613 * Root special: if there is a top-level log
4614 * device, treat the root vdev as if it were
4615 * degraded.
4616 */
4617 if (child->vdev_islog && vd == rvd)
4618 degraded++;
4619 else
4620 faulted++;
4621 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) {
4622 degraded++;
4623 }
4624
4625 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA)
4626 corrupted++;
4627 }
4628
4629 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded);
4630
4631 /*
4632 * Root special: if there is a top-level vdev that cannot be
4633 * opened due to corrupted metadata, then propagate the root
4634 * vdev's aux state as 'corrupt' rather than 'insufficient
4635 * replicas'.
4636 */
4637 if (corrupted && vd == rvd &&
4638 rvd->vdev_state == VDEV_STATE_CANT_OPEN)
4639 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN,
4640 VDEV_AUX_CORRUPT_DATA);
4641 }
4642
4643 if (vd->vdev_parent)
4644 vdev_propagate_state(vd->vdev_parent);
4645 }
4646
4647 /*
4648 * Set a vdev's state. If this is during an open, we don't update the parent
4649 * state, because we're in the process of opening children depth-first.
4650 * Otherwise, we propagate the change to the parent.
4651 *
4652 * If this routine places a device in a faulted state, an appropriate ereport is
4653 * generated.
4654 */
4655 void
4656 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
4657 {
4658 uint64_t save_state;
4659 spa_t *spa = vd->vdev_spa;
4660
4661 if (state == vd->vdev_state) {
4662 /*
4663 * Since vdev_offline() code path is already in an offline
4664 * state we can miss a statechange event to OFFLINE. Check
4665 * the previous state to catch this condition.
4666 */
4667 if (vd->vdev_ops->vdev_op_leaf &&
4668 (state == VDEV_STATE_OFFLINE) &&
4669 (vd->vdev_prevstate >= VDEV_STATE_FAULTED)) {
4670 /* post an offline state change */
4671 zfs_post_state_change(spa, vd, vd->vdev_prevstate);
4672 }
4673 vd->vdev_stat.vs_aux = aux;
4674 return;
4675 }
4676
4677 save_state = vd->vdev_state;
4678
4679 vd->vdev_state = state;
4680 vd->vdev_stat.vs_aux = aux;
4681
4682 /*
4683 * If we are setting the vdev state to anything but an open state, then
4684 * always close the underlying device unless the device has requested
4685 * a delayed close (i.e. we're about to remove or fault the device).
4686 * Otherwise, we keep accessible but invalid devices open forever.
4687 * We don't call vdev_close() itself, because that implies some extra
4688 * checks (offline, etc) that we don't want here. This is limited to
4689 * leaf devices, because otherwise closing the device will affect other
4690 * children.
4691 */
4692 if (!vd->vdev_delayed_close && vdev_is_dead(vd) &&
4693 vd->vdev_ops->vdev_op_leaf)
4694 vd->vdev_ops->vdev_op_close(vd);
4695
4696 if (vd->vdev_removed &&
4697 state == VDEV_STATE_CANT_OPEN &&
4698 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) {
4699 /*
4700 * If the previous state is set to VDEV_STATE_REMOVED, then this
4701 * device was previously marked removed and someone attempted to
4702 * reopen it. If this failed due to a nonexistent device, then
4703 * keep the device in the REMOVED state. We also let this be if
4704 * it is one of our special test online cases, which is only
4705 * attempting to online the device and shouldn't generate an FMA
4706 * fault.
4707 */
4708 vd->vdev_state = VDEV_STATE_REMOVED;
4709 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
4710 } else if (state == VDEV_STATE_REMOVED) {
4711 vd->vdev_removed = B_TRUE;
4712 } else if (state == VDEV_STATE_CANT_OPEN) {
4713 /*
4714 * If we fail to open a vdev during an import or recovery, we
4715 * mark it as "not available", which signifies that it was
4716 * never there to begin with. Failure to open such a device
4717 * is not considered an error.
4718 */
4719 if ((spa_load_state(spa) == SPA_LOAD_IMPORT ||
4720 spa_load_state(spa) == SPA_LOAD_RECOVER) &&
4721 vd->vdev_ops->vdev_op_leaf)
4722 vd->vdev_not_present = 1;
4723
4724 /*
4725 * Post the appropriate ereport. If the 'prevstate' field is
4726 * set to something other than VDEV_STATE_UNKNOWN, it indicates
4727 * that this is part of a vdev_reopen(). In this case, we don't
4728 * want to post the ereport if the device was already in the
4729 * CANT_OPEN state beforehand.
4730 *
4731 * If the 'checkremove' flag is set, then this is an attempt to
4732 * online the device in response to an insertion event. If we
4733 * hit this case, then we have detected an insertion event for a
4734 * faulted or offline device that wasn't in the removed state.
4735 * In this scenario, we don't post an ereport because we are
4736 * about to replace the device, or attempt an online with
4737 * vdev_forcefault, which will generate the fault for us.
4738 */
4739 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) &&
4740 !vd->vdev_not_present && !vd->vdev_checkremove &&
4741 vd != spa->spa_root_vdev) {
4742 const char *class;
4743
4744 switch (aux) {
4745 case VDEV_AUX_OPEN_FAILED:
4746 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED;
4747 break;
4748 case VDEV_AUX_CORRUPT_DATA:
4749 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA;
4750 break;
4751 case VDEV_AUX_NO_REPLICAS:
4752 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS;
4753 break;
4754 case VDEV_AUX_BAD_GUID_SUM:
4755 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM;
4756 break;
4757 case VDEV_AUX_TOO_SMALL:
4758 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL;
4759 break;
4760 case VDEV_AUX_BAD_LABEL:
4761 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL;
4762 break;
4763 case VDEV_AUX_BAD_ASHIFT:
4764 class = FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT;
4765 break;
4766 default:
4767 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN;
4768 }
4769
4770 (void) zfs_ereport_post(class, spa, vd, NULL, NULL,
4771 save_state);
4772 }
4773
4774 /* Erase any notion of persistent removed state */
4775 vd->vdev_removed = B_FALSE;
4776 } else {
4777 vd->vdev_removed = B_FALSE;
4778 }
4779
4780 /*
4781 * Notify ZED of any significant state-change on a leaf vdev.
4782 *
4783 */
4784 if (vd->vdev_ops->vdev_op_leaf) {
4785 /* preserve original state from a vdev_reopen() */
4786 if ((vd->vdev_prevstate != VDEV_STATE_UNKNOWN) &&
4787 (vd->vdev_prevstate != vd->vdev_state) &&
4788 (save_state <= VDEV_STATE_CLOSED))
4789 save_state = vd->vdev_prevstate;
4790
4791 /* filter out state change due to initial vdev_open */
4792 if (save_state > VDEV_STATE_CLOSED)
4793 zfs_post_state_change(spa, vd, save_state);
4794 }
4795
4796 if (!isopen && vd->vdev_parent)
4797 vdev_propagate_state(vd->vdev_parent);
4798 }
4799
4800 boolean_t
4801 vdev_children_are_offline(vdev_t *vd)
4802 {
4803 ASSERT(!vd->vdev_ops->vdev_op_leaf);
4804
4805 for (uint64_t i = 0; i < vd->vdev_children; i++) {
4806 if (vd->vdev_child[i]->vdev_state != VDEV_STATE_OFFLINE)
4807 return (B_FALSE);
4808 }
4809
4810 return (B_TRUE);
4811 }
4812
4813 /*
4814 * Check the vdev configuration to ensure that it's capable of supporting
4815 * a root pool. We do not support partial configuration.
4816 */
4817 boolean_t
4818 vdev_is_bootable(vdev_t *vd)
4819 {
4820 if (!vd->vdev_ops->vdev_op_leaf) {
4821 const char *vdev_type = vd->vdev_ops->vdev_op_type;
4822
4823 if (strcmp(vdev_type, VDEV_TYPE_MISSING) == 0 ||
4824 strcmp(vdev_type, VDEV_TYPE_INDIRECT) == 0) {
4825 return (B_FALSE);
4826 }
4827 }
4828
4829 for (int c = 0; c < vd->vdev_children; c++) {
4830 if (!vdev_is_bootable(vd->vdev_child[c]))
4831 return (B_FALSE);
4832 }
4833 return (B_TRUE);
4834 }
4835
4836 boolean_t
4837 vdev_is_concrete(vdev_t *vd)
4838 {
4839 vdev_ops_t *ops = vd->vdev_ops;
4840 if (ops == &vdev_indirect_ops || ops == &vdev_hole_ops ||
4841 ops == &vdev_missing_ops || ops == &vdev_root_ops) {
4842 return (B_FALSE);
4843 } else {
4844 return (B_TRUE);
4845 }
4846 }
4847
4848 /*
4849 * Determine if a log device has valid content. If the vdev was
4850 * removed or faulted in the MOS config then we know that
4851 * the content on the log device has already been written to the pool.
4852 */
4853 boolean_t
4854 vdev_log_state_valid(vdev_t *vd)
4855 {
4856 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted &&
4857 !vd->vdev_removed)
4858 return (B_TRUE);
4859
4860 for (int c = 0; c < vd->vdev_children; c++)
4861 if (vdev_log_state_valid(vd->vdev_child[c]))
4862 return (B_TRUE);
4863
4864 return (B_FALSE);
4865 }
4866
4867 /*
4868 * Expand a vdev if possible.
4869 */
4870 void
4871 vdev_expand(vdev_t *vd, uint64_t txg)
4872 {
4873 ASSERT(vd->vdev_top == vd);
4874 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
4875 ASSERT(vdev_is_concrete(vd));
4876
4877 vdev_set_deflate_ratio(vd);
4878
4879 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count &&
4880 vdev_is_concrete(vd)) {
4881 vdev_metaslab_group_create(vd);
4882 VERIFY(vdev_metaslab_init(vd, txg) == 0);
4883 vdev_config_dirty(vd);
4884 }
4885 }
4886
4887 /*
4888 * Split a vdev.
4889 */
4890 void
4891 vdev_split(vdev_t *vd)
4892 {
4893 vdev_t *cvd, *pvd = vd->vdev_parent;
4894
4895 vdev_remove_child(pvd, vd);
4896 vdev_compact_children(pvd);
4897
4898 cvd = pvd->vdev_child[0];
4899 if (pvd->vdev_children == 1) {
4900 vdev_remove_parent(cvd);
4901 cvd->vdev_splitting = B_TRUE;
4902 }
4903 vdev_propagate_state(cvd);
4904 }
4905
4906 void
4907 vdev_deadman(vdev_t *vd, char *tag)
4908 {
4909 for (int c = 0; c < vd->vdev_children; c++) {
4910 vdev_t *cvd = vd->vdev_child[c];
4911
4912 vdev_deadman(cvd, tag);
4913 }
4914
4915 if (vd->vdev_ops->vdev_op_leaf) {
4916 vdev_queue_t *vq = &vd->vdev_queue;
4917
4918 mutex_enter(&vq->vq_lock);
4919 if (avl_numnodes(&vq->vq_active_tree) > 0) {
4920 spa_t *spa = vd->vdev_spa;
4921 zio_t *fio;
4922 uint64_t delta;
4923
4924 zfs_dbgmsg("slow vdev: %s has %d active IOs",
4925 vd->vdev_path, avl_numnodes(&vq->vq_active_tree));
4926
4927 /*
4928 * Look at the head of all the pending queues,
4929 * if any I/O has been outstanding for longer than
4930 * the spa_deadman_synctime invoke the deadman logic.
4931 */
4932 fio = avl_first(&vq->vq_active_tree);
4933 delta = gethrtime() - fio->io_timestamp;
4934 if (delta > spa_deadman_synctime(spa))
4935 zio_deadman(fio, tag);
4936 }
4937 mutex_exit(&vq->vq_lock);
4938 }
4939 }
4940
4941 void
4942 vdev_defer_resilver(vdev_t *vd)
4943 {
4944 ASSERT(vd->vdev_ops->vdev_op_leaf);
4945
4946 vd->vdev_resilver_deferred = B_TRUE;
4947 vd->vdev_spa->spa_resilver_deferred = B_TRUE;
4948 }
4949
4950 /*
4951 * Clears the resilver deferred flag on all leaf devs under vd. Returns
4952 * B_TRUE if we have devices that need to be resilvered and are available to
4953 * accept resilver I/Os.
4954 */
4955 boolean_t
4956 vdev_clear_resilver_deferred(vdev_t *vd, dmu_tx_t *tx)
4957 {
4958 boolean_t resilver_needed = B_FALSE;
4959 spa_t *spa = vd->vdev_spa;
4960
4961 for (int c = 0; c < vd->vdev_children; c++) {
4962 vdev_t *cvd = vd->vdev_child[c];
4963 resilver_needed |= vdev_clear_resilver_deferred(cvd, tx);
4964 }
4965
4966 if (vd == spa->spa_root_vdev &&
4967 spa_feature_is_active(spa, SPA_FEATURE_RESILVER_DEFER)) {
4968 spa_feature_decr(spa, SPA_FEATURE_RESILVER_DEFER, tx);
4969 vdev_config_dirty(vd);
4970 spa->spa_resilver_deferred = B_FALSE;
4971 return (resilver_needed);
4972 }
4973
4974 if (!vdev_is_concrete(vd) || vd->vdev_aux ||
4975 !vd->vdev_ops->vdev_op_leaf)
4976 return (resilver_needed);
4977
4978 vd->vdev_resilver_deferred = B_FALSE;
4979
4980 return (!vdev_is_dead(vd) && !vd->vdev_offline &&
4981 vdev_resilver_needed(vd, NULL, NULL));
4982 }
4983
4984 /*
4985 * Translate a logical range to the physical range for the specified vdev_t.
4986 * This function is initially called with a leaf vdev and will walk each
4987 * parent vdev until it reaches a top-level vdev. Once the top-level is
4988 * reached the physical range is initialized and the recursive function
4989 * begins to unwind. As it unwinds it calls the parent's vdev specific
4990 * translation function to do the real conversion.
4991 */
4992 void
4993 vdev_xlate(vdev_t *vd, const range_seg64_t *logical_rs,
4994 range_seg64_t *physical_rs)
4995 {
4996 /*
4997 * Walk up the vdev tree
4998 */
4999 if (vd != vd->vdev_top) {
5000 vdev_xlate(vd->vdev_parent, logical_rs, physical_rs);
5001 } else {
5002 /*
5003 * We've reached the top-level vdev, initialize the
5004 * physical range to the logical range and start to
5005 * unwind.
5006 */
5007 physical_rs->rs_start = logical_rs->rs_start;
5008 physical_rs->rs_end = logical_rs->rs_end;
5009 return;
5010 }
5011
5012 vdev_t *pvd = vd->vdev_parent;
5013 ASSERT3P(pvd, !=, NULL);
5014 ASSERT3P(pvd->vdev_ops->vdev_op_xlate, !=, NULL);
5015
5016 /*
5017 * As this recursive function unwinds, translate the logical
5018 * range into its physical components by calling the
5019 * vdev specific translate function.
5020 */
5021 range_seg64_t intermediate = { 0 };
5022 pvd->vdev_ops->vdev_op_xlate(vd, physical_rs, &intermediate);
5023
5024 physical_rs->rs_start = intermediate.rs_start;
5025 physical_rs->rs_end = intermediate.rs_end;
5026 }
5027
5028 /*
5029 * Look at the vdev tree and determine whether any devices are currently being
5030 * replaced.
5031 */
5032 boolean_t
5033 vdev_replace_in_progress(vdev_t *vdev)
5034 {
5035 ASSERT(spa_config_held(vdev->vdev_spa, SCL_ALL, RW_READER) != 0);
5036
5037 if (vdev->vdev_ops == &vdev_replacing_ops)
5038 return (B_TRUE);
5039
5040 /*
5041 * A 'spare' vdev indicates that we have a replace in progress, unless
5042 * it has exactly two children, and the second, the hot spare, has
5043 * finished being resilvered.
5044 */
5045 if (vdev->vdev_ops == &vdev_spare_ops && (vdev->vdev_children > 2 ||
5046 !vdev_dtl_empty(vdev->vdev_child[1], DTL_MISSING)))
5047 return (B_TRUE);
5048
5049 for (int i = 0; i < vdev->vdev_children; i++) {
5050 if (vdev_replace_in_progress(vdev->vdev_child[i]))
5051 return (B_TRUE);
5052 }
5053
5054 return (B_FALSE);
5055 }
5056
5057 EXPORT_SYMBOL(vdev_fault);
5058 EXPORT_SYMBOL(vdev_degrade);
5059 EXPORT_SYMBOL(vdev_online);
5060 EXPORT_SYMBOL(vdev_offline);
5061 EXPORT_SYMBOL(vdev_clear);
5062
5063 /* BEGIN CSTYLED */
5064 ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, default_ms_count, INT, ZMOD_RW,
5065 "Target number of metaslabs per top-level vdev");
5066
5067 ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, default_ms_shift, INT, ZMOD_RW,
5068 "Default limit for metaslab size");
5069
5070 ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, min_ms_count, INT, ZMOD_RW,
5071 "Minimum number of metaslabs per top-level vdev");
5072
5073 ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, ms_count_limit, INT, ZMOD_RW,
5074 "Practical upper limit of total metaslabs per top-level vdev");
5075
5076 ZFS_MODULE_PARAM(zfs, zfs_, slow_io_events_per_second, UINT, ZMOD_RW,
5077 "Rate limit slow IO (delay) events to this many per second");
5078
5079 ZFS_MODULE_PARAM(zfs, zfs_, checksum_events_per_second, UINT, ZMOD_RW,
5080 "Rate limit checksum events to this many checksum errors per second "
5081 "(do not set below zed threshold).");
5082
5083 ZFS_MODULE_PARAM(zfs, zfs_, scan_ignore_errors, INT, ZMOD_RW,
5084 "Ignore errors during resilver/scrub");
5085
5086 ZFS_MODULE_PARAM(zfs_vdev, vdev_, validate_skip, INT, ZMOD_RW,
5087 "Bypass vdev_validate()");
5088
5089 ZFS_MODULE_PARAM(zfs, zfs_, nocacheflush, INT, ZMOD_RW,
5090 "Disable cache flushes");
5091
5092 ZFS_MODULE_PARAM_CALL(zfs_vdev, zfs_vdev_, min_auto_ashift,
5093 param_set_min_auto_ashift, param_get_ulong, ZMOD_RW,
5094 "Minimum ashift used when creating new top-level vdevs");
5095
5096 ZFS_MODULE_PARAM_CALL(zfs_vdev, zfs_vdev_, max_auto_ashift,
5097 param_set_max_auto_ashift, param_get_ulong, ZMOD_RW,
5098 "Maximum ashift used when optimizing for logical -> physical sector "
5099 "size on new top-level vdevs");
5100 /* END CSTYLED */