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 https://opensource.org/licenses/CDDL-1.0.
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, 2021 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 * Copyright (c) 2021, Klara Inc.
32 * Copyright (c) 2021, 2023 Hewlett Packard Enterprise Development LP.
33 */
34
35 #include <sys/zfs_context.h>
36 #include <sys/fm/fs/zfs.h>
37 #include <sys/spa.h>
38 #include <sys/spa_impl.h>
39 #include <sys/bpobj.h>
40 #include <sys/dmu.h>
41 #include <sys/dmu_tx.h>
42 #include <sys/dsl_dir.h>
43 #include <sys/vdev_impl.h>
44 #include <sys/vdev_rebuild.h>
45 #include <sys/vdev_draid.h>
46 #include <sys/uberblock_impl.h>
47 #include <sys/metaslab.h>
48 #include <sys/metaslab_impl.h>
49 #include <sys/space_map.h>
50 #include <sys/space_reftree.h>
51 #include <sys/zio.h>
52 #include <sys/zap.h>
53 #include <sys/fs/zfs.h>
54 #include <sys/arc.h>
55 #include <sys/zil.h>
56 #include <sys/dsl_scan.h>
57 #include <sys/vdev_raidz.h>
58 #include <sys/abd.h>
59 #include <sys/vdev_initialize.h>
60 #include <sys/vdev_trim.h>
61 #include <sys/vdev_raidz.h>
62 #include <sys/zvol.h>
63 #include <sys/zfs_ratelimit.h>
64 #include "zfs_prop.h"
65
66 /*
67 * One metaslab from each (normal-class) vdev is used by the ZIL. These are
68 * called "embedded slog metaslabs", are referenced by vdev_log_mg, and are
69 * part of the spa_embedded_log_class. The metaslab with the most free space
70 * in each vdev is selected for this purpose when the pool is opened (or a
71 * vdev is added). See vdev_metaslab_init().
72 *
73 * Log blocks can be allocated from the following locations. Each one is tried
74 * in order until the allocation succeeds:
75 * 1. dedicated log vdevs, aka "slog" (spa_log_class)
76 * 2. embedded slog metaslabs (spa_embedded_log_class)
77 * 3. other metaslabs in normal vdevs (spa_normal_class)
78 *
79 * zfs_embedded_slog_min_ms disables the embedded slog if there are fewer
80 * than this number of metaslabs in the vdev. This ensures that we don't set
81 * aside an unreasonable amount of space for the ZIL. If set to less than
82 * 1 << (spa_slop_shift + 1), on small pools the usable space may be reduced
83 * (by more than 1<<spa_slop_shift) due to the embedded slog metaslab.
84 */
85 static uint_t zfs_embedded_slog_min_ms = 64;
86
87 /* default target for number of metaslabs per top-level vdev */
88 static uint_t zfs_vdev_default_ms_count = 200;
89
90 /* minimum number of metaslabs per top-level vdev */
91 static uint_t zfs_vdev_min_ms_count = 16;
92
93 /* practical upper limit of total metaslabs per top-level vdev */
94 static uint_t zfs_vdev_ms_count_limit = 1ULL << 17;
95
96 /* lower limit for metaslab size (512M) */
97 static uint_t zfs_vdev_default_ms_shift = 29;
98
99 /* upper limit for metaslab size (16G) */
100 static uint_t zfs_vdev_max_ms_shift = 34;
101
102 int vdev_validate_skip = B_FALSE;
103
104 /*
105 * Since the DTL space map of a vdev is not expected to have a lot of
106 * entries, we default its block size to 4K.
107 */
108 int zfs_vdev_dtl_sm_blksz = (1 << 12);
109
110 /*
111 * Rate limit slow IO (delay) events to this many per second.
112 */
113 static unsigned int zfs_slow_io_events_per_second = 20;
114
115 /*
116 * Rate limit checksum events after this many checksum errors per second.
117 */
118 static unsigned int zfs_checksum_events_per_second = 20;
119
120 /*
121 * Ignore errors during scrub/resilver. Allows to work around resilver
122 * upon import when there are pool errors.
123 */
124 static int zfs_scan_ignore_errors = 0;
125
126 /*
127 * vdev-wide space maps that have lots of entries written to them at
128 * the end of each transaction can benefit from a higher I/O bandwidth
129 * (e.g. vdev_obsolete_sm), thus we default their block size to 128K.
130 */
131 int zfs_vdev_standard_sm_blksz = (1 << 17);
132
133 /*
134 * Tunable parameter for debugging or performance analysis. Setting this
135 * will cause pool corruption on power loss if a volatile out-of-order
136 * write cache is enabled.
137 */
138 int zfs_nocacheflush = 0;
139
140 /*
141 * Maximum and minimum ashift values that can be automatically set based on
142 * vdev's physical ashift (disk's physical sector size). While ASHIFT_MAX
143 * is higher than the maximum value, it is intentionally limited here to not
144 * excessively impact pool space efficiency. Higher ashift values may still
145 * be forced by vdev logical ashift or by user via ashift property, but won't
146 * be set automatically as a performance optimization.
147 */
148 uint_t zfs_vdev_max_auto_ashift = 14;
149 uint_t zfs_vdev_min_auto_ashift = ASHIFT_MIN;
150
151 void
vdev_dbgmsg(vdev_t * vd,const char * fmt,...)152 vdev_dbgmsg(vdev_t *vd, const char *fmt, ...)
153 {
154 va_list adx;
155 char buf[256];
156
157 va_start(adx, fmt);
158 (void) vsnprintf(buf, sizeof (buf), fmt, adx);
159 va_end(adx);
160
161 if (vd->vdev_path != NULL) {
162 zfs_dbgmsg("%s vdev '%s': %s", vd->vdev_ops->vdev_op_type,
163 vd->vdev_path, buf);
164 } else {
165 zfs_dbgmsg("%s-%llu vdev (guid %llu): %s",
166 vd->vdev_ops->vdev_op_type,
167 (u_longlong_t)vd->vdev_id,
168 (u_longlong_t)vd->vdev_guid, buf);
169 }
170 }
171
172 void
vdev_dbgmsg_print_tree(vdev_t * vd,int indent)173 vdev_dbgmsg_print_tree(vdev_t *vd, int indent)
174 {
175 char state[20];
176
177 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) {
178 zfs_dbgmsg("%*svdev %llu: %s", indent, "",
179 (u_longlong_t)vd->vdev_id,
180 vd->vdev_ops->vdev_op_type);
181 return;
182 }
183
184 switch (vd->vdev_state) {
185 case VDEV_STATE_UNKNOWN:
186 (void) snprintf(state, sizeof (state), "unknown");
187 break;
188 case VDEV_STATE_CLOSED:
189 (void) snprintf(state, sizeof (state), "closed");
190 break;
191 case VDEV_STATE_OFFLINE:
192 (void) snprintf(state, sizeof (state), "offline");
193 break;
194 case VDEV_STATE_REMOVED:
195 (void) snprintf(state, sizeof (state), "removed");
196 break;
197 case VDEV_STATE_CANT_OPEN:
198 (void) snprintf(state, sizeof (state), "can't open");
199 break;
200 case VDEV_STATE_FAULTED:
201 (void) snprintf(state, sizeof (state), "faulted");
202 break;
203 case VDEV_STATE_DEGRADED:
204 (void) snprintf(state, sizeof (state), "degraded");
205 break;
206 case VDEV_STATE_HEALTHY:
207 (void) snprintf(state, sizeof (state), "healthy");
208 break;
209 default:
210 (void) snprintf(state, sizeof (state), "<state %u>",
211 (uint_t)vd->vdev_state);
212 }
213
214 zfs_dbgmsg("%*svdev %u: %s%s, guid: %llu, path: %s, %s", indent,
215 "", (int)vd->vdev_id, vd->vdev_ops->vdev_op_type,
216 vd->vdev_islog ? " (log)" : "",
217 (u_longlong_t)vd->vdev_guid,
218 vd->vdev_path ? vd->vdev_path : "N/A", state);
219
220 for (uint64_t i = 0; i < vd->vdev_children; i++)
221 vdev_dbgmsg_print_tree(vd->vdev_child[i], indent + 2);
222 }
223
224 /*
225 * Virtual device management.
226 */
227
228 static vdev_ops_t *const vdev_ops_table[] = {
229 &vdev_root_ops,
230 &vdev_raidz_ops,
231 &vdev_draid_ops,
232 &vdev_draid_spare_ops,
233 &vdev_mirror_ops,
234 &vdev_replacing_ops,
235 &vdev_spare_ops,
236 &vdev_disk_ops,
237 &vdev_file_ops,
238 &vdev_missing_ops,
239 &vdev_hole_ops,
240 &vdev_indirect_ops,
241 NULL
242 };
243
244 /*
245 * Given a vdev type, return the appropriate ops vector.
246 */
247 static vdev_ops_t *
vdev_getops(const char * type)248 vdev_getops(const char *type)
249 {
250 vdev_ops_t *ops, *const *opspp;
251
252 for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++)
253 if (strcmp(ops->vdev_op_type, type) == 0)
254 break;
255
256 return (ops);
257 }
258
259 /*
260 * Given a vdev and a metaslab class, find which metaslab group we're
261 * interested in. All vdevs may belong to two different metaslab classes.
262 * Dedicated slog devices use only the primary metaslab group, rather than a
263 * separate log group. For embedded slogs, the vdev_log_mg will be non-NULL.
264 */
265 metaslab_group_t *
vdev_get_mg(vdev_t * vd,metaslab_class_t * mc)266 vdev_get_mg(vdev_t *vd, metaslab_class_t *mc)
267 {
268 if (mc == spa_embedded_log_class(vd->vdev_spa) &&
269 vd->vdev_log_mg != NULL)
270 return (vd->vdev_log_mg);
271 else
272 return (vd->vdev_mg);
273 }
274
275 void
vdev_default_xlate(vdev_t * vd,const range_seg64_t * logical_rs,range_seg64_t * physical_rs,range_seg64_t * remain_rs)276 vdev_default_xlate(vdev_t *vd, const range_seg64_t *logical_rs,
277 range_seg64_t *physical_rs, range_seg64_t *remain_rs)
278 {
279 (void) vd, (void) remain_rs;
280
281 physical_rs->rs_start = logical_rs->rs_start;
282 physical_rs->rs_end = logical_rs->rs_end;
283 }
284
285 /*
286 * Derive the enumerated allocation bias from string input.
287 * String origin is either the per-vdev zap or zpool(8).
288 */
289 static vdev_alloc_bias_t
vdev_derive_alloc_bias(const char * bias)290 vdev_derive_alloc_bias(const char *bias)
291 {
292 vdev_alloc_bias_t alloc_bias = VDEV_BIAS_NONE;
293
294 if (strcmp(bias, VDEV_ALLOC_BIAS_LOG) == 0)
295 alloc_bias = VDEV_BIAS_LOG;
296 else if (strcmp(bias, VDEV_ALLOC_BIAS_SPECIAL) == 0)
297 alloc_bias = VDEV_BIAS_SPECIAL;
298 else if (strcmp(bias, VDEV_ALLOC_BIAS_DEDUP) == 0)
299 alloc_bias = VDEV_BIAS_DEDUP;
300
301 return (alloc_bias);
302 }
303
304 /*
305 * Default asize function: return the MAX of psize with the asize of
306 * all children. This is what's used by anything other than RAID-Z.
307 */
308 uint64_t
vdev_default_asize(vdev_t * vd,uint64_t psize,uint64_t txg)309 vdev_default_asize(vdev_t *vd, uint64_t psize, uint64_t txg)
310 {
311 uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift);
312 uint64_t csize;
313
314 for (int c = 0; c < vd->vdev_children; c++) {
315 csize = vdev_psize_to_asize_txg(vd->vdev_child[c], psize, txg);
316 asize = MAX(asize, csize);
317 }
318
319 return (asize);
320 }
321
322 uint64_t
vdev_default_min_asize(vdev_t * vd)323 vdev_default_min_asize(vdev_t *vd)
324 {
325 return (vd->vdev_min_asize);
326 }
327
328 /*
329 * Get the minimum allocatable size. We define the allocatable size as
330 * the vdev's asize rounded to the nearest metaslab. This allows us to
331 * replace or attach devices which don't have the same physical size but
332 * can still satisfy the same number of allocations.
333 */
334 uint64_t
vdev_get_min_asize(vdev_t * vd)335 vdev_get_min_asize(vdev_t *vd)
336 {
337 vdev_t *pvd = vd->vdev_parent;
338
339 /*
340 * If our parent is NULL (inactive spare or cache) or is the root,
341 * just return our own asize.
342 */
343 if (pvd == NULL)
344 return (vd->vdev_asize);
345
346 /*
347 * The top-level vdev just returns the allocatable size rounded
348 * to the nearest metaslab.
349 */
350 if (vd == vd->vdev_top)
351 return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift));
352
353 return (pvd->vdev_ops->vdev_op_min_asize(pvd));
354 }
355
356 void
vdev_set_min_asize(vdev_t * vd)357 vdev_set_min_asize(vdev_t *vd)
358 {
359 vd->vdev_min_asize = vdev_get_min_asize(vd);
360
361 for (int c = 0; c < vd->vdev_children; c++)
362 vdev_set_min_asize(vd->vdev_child[c]);
363 }
364
365 /*
366 * Get the minimal allocation size for the top-level vdev.
367 */
368 uint64_t
vdev_get_min_alloc(vdev_t * vd)369 vdev_get_min_alloc(vdev_t *vd)
370 {
371 uint64_t min_alloc = 1ULL << vd->vdev_ashift;
372
373 if (vd->vdev_ops->vdev_op_min_alloc != NULL)
374 min_alloc = vd->vdev_ops->vdev_op_min_alloc(vd);
375
376 return (min_alloc);
377 }
378
379 /*
380 * Get the parity level for a top-level vdev.
381 */
382 uint64_t
vdev_get_nparity(vdev_t * vd)383 vdev_get_nparity(vdev_t *vd)
384 {
385 uint64_t nparity = 0;
386
387 if (vd->vdev_ops->vdev_op_nparity != NULL)
388 nparity = vd->vdev_ops->vdev_op_nparity(vd);
389
390 return (nparity);
391 }
392
393 static int
vdev_prop_get_int(vdev_t * vd,vdev_prop_t prop,uint64_t * value)394 vdev_prop_get_int(vdev_t *vd, vdev_prop_t prop, uint64_t *value)
395 {
396 spa_t *spa = vd->vdev_spa;
397 objset_t *mos = spa->spa_meta_objset;
398 uint64_t objid;
399 int err;
400
401 if (vd->vdev_root_zap != 0) {
402 objid = vd->vdev_root_zap;
403 } else if (vd->vdev_top_zap != 0) {
404 objid = vd->vdev_top_zap;
405 } else if (vd->vdev_leaf_zap != 0) {
406 objid = vd->vdev_leaf_zap;
407 } else {
408 return (EINVAL);
409 }
410
411 err = zap_lookup(mos, objid, vdev_prop_to_name(prop),
412 sizeof (uint64_t), 1, value);
413
414 if (err == ENOENT)
415 *value = vdev_prop_default_numeric(prop);
416
417 return (err);
418 }
419
420 /*
421 * Get the number of data disks for a top-level vdev.
422 */
423 uint64_t
vdev_get_ndisks(vdev_t * vd)424 vdev_get_ndisks(vdev_t *vd)
425 {
426 uint64_t ndisks = 1;
427
428 if (vd->vdev_ops->vdev_op_ndisks != NULL)
429 ndisks = vd->vdev_ops->vdev_op_ndisks(vd);
430
431 return (ndisks);
432 }
433
434 vdev_t *
vdev_lookup_top(spa_t * spa,uint64_t vdev)435 vdev_lookup_top(spa_t *spa, uint64_t vdev)
436 {
437 vdev_t *rvd = spa->spa_root_vdev;
438
439 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
440
441 if (vdev < rvd->vdev_children) {
442 ASSERT(rvd->vdev_child[vdev] != NULL);
443 return (rvd->vdev_child[vdev]);
444 }
445
446 return (NULL);
447 }
448
449 vdev_t *
vdev_lookup_by_guid(vdev_t * vd,uint64_t guid)450 vdev_lookup_by_guid(vdev_t *vd, uint64_t guid)
451 {
452 vdev_t *mvd;
453
454 if (vd->vdev_guid == guid)
455 return (vd);
456
457 for (int c = 0; c < vd->vdev_children; c++)
458 if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) !=
459 NULL)
460 return (mvd);
461
462 return (NULL);
463 }
464
465 static int
vdev_count_leaves_impl(vdev_t * vd)466 vdev_count_leaves_impl(vdev_t *vd)
467 {
468 int n = 0;
469
470 if (vd->vdev_ops->vdev_op_leaf)
471 return (1);
472
473 for (int c = 0; c < vd->vdev_children; c++)
474 n += vdev_count_leaves_impl(vd->vdev_child[c]);
475
476 return (n);
477 }
478
479 int
vdev_count_leaves(spa_t * spa)480 vdev_count_leaves(spa_t *spa)
481 {
482 int rc;
483
484 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
485 rc = vdev_count_leaves_impl(spa->spa_root_vdev);
486 spa_config_exit(spa, SCL_VDEV, FTAG);
487
488 return (rc);
489 }
490
491 void
vdev_add_child(vdev_t * pvd,vdev_t * cvd)492 vdev_add_child(vdev_t *pvd, vdev_t *cvd)
493 {
494 size_t oldsize, newsize;
495 uint64_t id = cvd->vdev_id;
496 vdev_t **newchild;
497
498 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
499 ASSERT(cvd->vdev_parent == NULL);
500
501 cvd->vdev_parent = pvd;
502
503 if (pvd == NULL)
504 return;
505
506 ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL);
507
508 oldsize = pvd->vdev_children * sizeof (vdev_t *);
509 pvd->vdev_children = MAX(pvd->vdev_children, id + 1);
510 newsize = pvd->vdev_children * sizeof (vdev_t *);
511
512 newchild = kmem_alloc(newsize, KM_SLEEP);
513 if (pvd->vdev_child != NULL) {
514 memcpy(newchild, pvd->vdev_child, oldsize);
515 kmem_free(pvd->vdev_child, oldsize);
516 }
517
518 pvd->vdev_child = newchild;
519 pvd->vdev_child[id] = cvd;
520
521 cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd);
522 ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL);
523
524 /*
525 * Walk up all ancestors to update guid sum.
526 */
527 for (; pvd != NULL; pvd = pvd->vdev_parent)
528 pvd->vdev_guid_sum += cvd->vdev_guid_sum;
529
530 if (cvd->vdev_ops->vdev_op_leaf) {
531 list_insert_head(&cvd->vdev_spa->spa_leaf_list, cvd);
532 cvd->vdev_spa->spa_leaf_list_gen++;
533 }
534 }
535
536 void
vdev_remove_child(vdev_t * pvd,vdev_t * cvd)537 vdev_remove_child(vdev_t *pvd, vdev_t *cvd)
538 {
539 int c;
540 uint_t id = cvd->vdev_id;
541
542 ASSERT(cvd->vdev_parent == pvd);
543
544 if (pvd == NULL)
545 return;
546
547 ASSERT(id < pvd->vdev_children);
548 ASSERT(pvd->vdev_child[id] == cvd);
549
550 pvd->vdev_child[id] = NULL;
551 cvd->vdev_parent = NULL;
552
553 for (c = 0; c < pvd->vdev_children; c++)
554 if (pvd->vdev_child[c])
555 break;
556
557 if (c == pvd->vdev_children) {
558 kmem_free(pvd->vdev_child, c * sizeof (vdev_t *));
559 pvd->vdev_child = NULL;
560 pvd->vdev_children = 0;
561 }
562
563 if (cvd->vdev_ops->vdev_op_leaf) {
564 spa_t *spa = cvd->vdev_spa;
565 list_remove(&spa->spa_leaf_list, cvd);
566 spa->spa_leaf_list_gen++;
567 }
568
569 /*
570 * Walk up all ancestors to update guid sum.
571 */
572 for (; pvd != NULL; pvd = pvd->vdev_parent)
573 pvd->vdev_guid_sum -= cvd->vdev_guid_sum;
574 }
575
576 /*
577 * Remove any holes in the child array.
578 */
579 void
vdev_compact_children(vdev_t * pvd)580 vdev_compact_children(vdev_t *pvd)
581 {
582 vdev_t **newchild, *cvd;
583 int oldc = pvd->vdev_children;
584 int newc;
585
586 ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
587
588 if (oldc == 0)
589 return;
590
591 for (int c = newc = 0; c < oldc; c++)
592 if (pvd->vdev_child[c])
593 newc++;
594
595 if (newc > 0) {
596 newchild = kmem_zalloc(newc * sizeof (vdev_t *), KM_SLEEP);
597
598 for (int c = newc = 0; c < oldc; c++) {
599 if ((cvd = pvd->vdev_child[c]) != NULL) {
600 newchild[newc] = cvd;
601 cvd->vdev_id = newc++;
602 }
603 }
604 } else {
605 newchild = NULL;
606 }
607
608 kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *));
609 pvd->vdev_child = newchild;
610 pvd->vdev_children = newc;
611 }
612
613 /*
614 * Allocate and minimally initialize a vdev_t.
615 */
616 vdev_t *
vdev_alloc_common(spa_t * spa,uint_t id,uint64_t guid,vdev_ops_t * ops)617 vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops)
618 {
619 vdev_t *vd;
620 vdev_indirect_config_t *vic;
621
622 vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP);
623 vic = &vd->vdev_indirect_config;
624
625 if (spa->spa_root_vdev == NULL) {
626 ASSERT(ops == &vdev_root_ops);
627 spa->spa_root_vdev = vd;
628 spa->spa_load_guid = spa_generate_guid(NULL);
629 }
630
631 if (guid == 0 && ops != &vdev_hole_ops) {
632 if (spa->spa_root_vdev == vd) {
633 /*
634 * The root vdev's guid will also be the pool guid,
635 * which must be unique among all pools.
636 */
637 guid = spa_generate_guid(NULL);
638 } else {
639 /*
640 * Any other vdev's guid must be unique within the pool.
641 */
642 guid = spa_generate_guid(spa);
643 }
644 ASSERT(!spa_guid_exists(spa_guid(spa), guid));
645 }
646
647 vd->vdev_spa = spa;
648 vd->vdev_id = id;
649 vd->vdev_guid = guid;
650 vd->vdev_guid_sum = guid;
651 vd->vdev_ops = ops;
652 vd->vdev_state = VDEV_STATE_CLOSED;
653 vd->vdev_ishole = (ops == &vdev_hole_ops);
654 vic->vic_prev_indirect_vdev = UINT64_MAX;
655
656 rw_init(&vd->vdev_indirect_rwlock, NULL, RW_DEFAULT, NULL);
657 mutex_init(&vd->vdev_obsolete_lock, NULL, MUTEX_DEFAULT, NULL);
658 vd->vdev_obsolete_segments = range_tree_create(NULL, RANGE_SEG64, NULL,
659 0, 0);
660
661 /*
662 * Initialize rate limit structs for events. We rate limit ZIO delay
663 * and checksum events so that we don't overwhelm ZED with thousands
664 * of events when a disk is acting up.
665 */
666 zfs_ratelimit_init(&vd->vdev_delay_rl, &zfs_slow_io_events_per_second,
667 1);
668 zfs_ratelimit_init(&vd->vdev_deadman_rl, &zfs_slow_io_events_per_second,
669 1);
670 zfs_ratelimit_init(&vd->vdev_checksum_rl,
671 &zfs_checksum_events_per_second, 1);
672
673 /*
674 * Default Thresholds for tuning ZED
675 */
676 vd->vdev_checksum_n = vdev_prop_default_numeric(VDEV_PROP_CHECKSUM_N);
677 vd->vdev_checksum_t = vdev_prop_default_numeric(VDEV_PROP_CHECKSUM_T);
678 vd->vdev_io_n = vdev_prop_default_numeric(VDEV_PROP_IO_N);
679 vd->vdev_io_t = vdev_prop_default_numeric(VDEV_PROP_IO_T);
680 vd->vdev_slow_io_n = vdev_prop_default_numeric(VDEV_PROP_SLOW_IO_N);
681 vd->vdev_slow_io_t = vdev_prop_default_numeric(VDEV_PROP_SLOW_IO_T);
682
683 list_link_init(&vd->vdev_config_dirty_node);
684 list_link_init(&vd->vdev_state_dirty_node);
685 list_link_init(&vd->vdev_initialize_node);
686 list_link_init(&vd->vdev_leaf_node);
687 list_link_init(&vd->vdev_trim_node);
688
689 mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_NOLOCKDEP, NULL);
690 mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL);
691 mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL);
692 mutex_init(&vd->vdev_scan_io_queue_lock, NULL, MUTEX_DEFAULT, NULL);
693
694 mutex_init(&vd->vdev_initialize_lock, NULL, MUTEX_DEFAULT, NULL);
695 mutex_init(&vd->vdev_initialize_io_lock, NULL, MUTEX_DEFAULT, NULL);
696 cv_init(&vd->vdev_initialize_cv, NULL, CV_DEFAULT, NULL);
697 cv_init(&vd->vdev_initialize_io_cv, NULL, CV_DEFAULT, NULL);
698
699 mutex_init(&vd->vdev_trim_lock, NULL, MUTEX_DEFAULT, NULL);
700 mutex_init(&vd->vdev_autotrim_lock, NULL, MUTEX_DEFAULT, NULL);
701 mutex_init(&vd->vdev_trim_io_lock, NULL, MUTEX_DEFAULT, NULL);
702 cv_init(&vd->vdev_trim_cv, NULL, CV_DEFAULT, NULL);
703 cv_init(&vd->vdev_autotrim_cv, NULL, CV_DEFAULT, NULL);
704 cv_init(&vd->vdev_autotrim_kick_cv, NULL, CV_DEFAULT, NULL);
705 cv_init(&vd->vdev_trim_io_cv, NULL, CV_DEFAULT, NULL);
706
707 mutex_init(&vd->vdev_rebuild_lock, NULL, MUTEX_DEFAULT, NULL);
708 cv_init(&vd->vdev_rebuild_cv, NULL, CV_DEFAULT, NULL);
709
710 for (int t = 0; t < DTL_TYPES; t++) {
711 vd->vdev_dtl[t] = range_tree_create(NULL, RANGE_SEG64, NULL, 0,
712 0);
713 }
714
715 txg_list_create(&vd->vdev_ms_list, spa,
716 offsetof(struct metaslab, ms_txg_node));
717 txg_list_create(&vd->vdev_dtl_list, spa,
718 offsetof(struct vdev, vdev_dtl_node));
719 vd->vdev_stat.vs_timestamp = gethrtime();
720 vdev_queue_init(vd);
721
722 return (vd);
723 }
724
725 /*
726 * Allocate a new vdev. The 'alloctype' is used to control whether we are
727 * creating a new vdev or loading an existing one - the behavior is slightly
728 * different for each case.
729 */
730 int
vdev_alloc(spa_t * spa,vdev_t ** vdp,nvlist_t * nv,vdev_t * parent,uint_t id,int alloctype)731 vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
732 int alloctype)
733 {
734 vdev_ops_t *ops;
735 const char *type;
736 uint64_t guid = 0, islog;
737 vdev_t *vd;
738 vdev_indirect_config_t *vic;
739 const char *tmp = NULL;
740 int rc;
741 vdev_alloc_bias_t alloc_bias = VDEV_BIAS_NONE;
742 boolean_t top_level = (parent && !parent->vdev_parent);
743
744 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
745
746 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
747 return (SET_ERROR(EINVAL));
748
749 if ((ops = vdev_getops(type)) == NULL)
750 return (SET_ERROR(EINVAL));
751
752 /*
753 * If this is a load, get the vdev guid from the nvlist.
754 * Otherwise, vdev_alloc_common() will generate one for us.
755 */
756 if (alloctype == VDEV_ALLOC_LOAD) {
757 uint64_t label_id;
758
759 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) ||
760 label_id != id)
761 return (SET_ERROR(EINVAL));
762
763 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
764 return (SET_ERROR(EINVAL));
765 } else if (alloctype == VDEV_ALLOC_SPARE) {
766 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
767 return (SET_ERROR(EINVAL));
768 } else if (alloctype == VDEV_ALLOC_L2CACHE) {
769 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
770 return (SET_ERROR(EINVAL));
771 } else if (alloctype == VDEV_ALLOC_ROOTPOOL) {
772 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
773 return (SET_ERROR(EINVAL));
774 }
775
776 /*
777 * The first allocated vdev must be of type 'root'.
778 */
779 if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL)
780 return (SET_ERROR(EINVAL));
781
782 /*
783 * Determine whether we're a log vdev.
784 */
785 islog = 0;
786 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog);
787 if (islog && spa_version(spa) < SPA_VERSION_SLOGS)
788 return (SET_ERROR(ENOTSUP));
789
790 if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES)
791 return (SET_ERROR(ENOTSUP));
792
793 if (top_level && alloctype == VDEV_ALLOC_ADD) {
794 const char *bias;
795
796 /*
797 * If creating a top-level vdev, check for allocation
798 * classes input.
799 */
800 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_ALLOCATION_BIAS,
801 &bias) == 0) {
802 alloc_bias = vdev_derive_alloc_bias(bias);
803
804 /* spa_vdev_add() expects feature to be enabled */
805 if (spa->spa_load_state != SPA_LOAD_CREATE &&
806 !spa_feature_is_enabled(spa,
807 SPA_FEATURE_ALLOCATION_CLASSES)) {
808 return (SET_ERROR(ENOTSUP));
809 }
810 }
811
812 /* spa_vdev_add() expects feature to be enabled */
813 if (ops == &vdev_draid_ops &&
814 spa->spa_load_state != SPA_LOAD_CREATE &&
815 !spa_feature_is_enabled(spa, SPA_FEATURE_DRAID)) {
816 return (SET_ERROR(ENOTSUP));
817 }
818 }
819
820 /*
821 * Initialize the vdev specific data. This is done before calling
822 * vdev_alloc_common() since it may fail and this simplifies the
823 * error reporting and cleanup code paths.
824 */
825 void *tsd = NULL;
826 if (ops->vdev_op_init != NULL) {
827 rc = ops->vdev_op_init(spa, nv, &tsd);
828 if (rc != 0) {
829 return (rc);
830 }
831 }
832
833 vd = vdev_alloc_common(spa, id, guid, ops);
834 vd->vdev_tsd = tsd;
835 vd->vdev_islog = islog;
836
837 if (top_level && alloc_bias != VDEV_BIAS_NONE)
838 vd->vdev_alloc_bias = alloc_bias;
839
840 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &tmp) == 0)
841 vd->vdev_path = spa_strdup(tmp);
842
843 /*
844 * ZPOOL_CONFIG_AUX_STATE = "external" means we previously forced a
845 * fault on a vdev and want it to persist across imports (like with
846 * zpool offline -f).
847 */
848 rc = nvlist_lookup_string(nv, ZPOOL_CONFIG_AUX_STATE, &tmp);
849 if (rc == 0 && tmp != NULL && strcmp(tmp, "external") == 0) {
850 vd->vdev_stat.vs_aux = VDEV_AUX_EXTERNAL;
851 vd->vdev_faulted = 1;
852 vd->vdev_label_aux = VDEV_AUX_EXTERNAL;
853 }
854
855 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &tmp) == 0)
856 vd->vdev_devid = spa_strdup(tmp);
857 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH, &tmp) == 0)
858 vd->vdev_physpath = spa_strdup(tmp);
859
860 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH,
861 &tmp) == 0)
862 vd->vdev_enc_sysfs_path = spa_strdup(tmp);
863
864 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &tmp) == 0)
865 vd->vdev_fru = spa_strdup(tmp);
866
867 /*
868 * Set the whole_disk property. If it's not specified, leave the value
869 * as -1.
870 */
871 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
872 &vd->vdev_wholedisk) != 0)
873 vd->vdev_wholedisk = -1ULL;
874
875 vic = &vd->vdev_indirect_config;
876
877 ASSERT0(vic->vic_mapping_object);
878 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_OBJECT,
879 &vic->vic_mapping_object);
880 ASSERT0(vic->vic_births_object);
881 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_BIRTHS,
882 &vic->vic_births_object);
883 ASSERT3U(vic->vic_prev_indirect_vdev, ==, UINT64_MAX);
884 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_PREV_INDIRECT_VDEV,
885 &vic->vic_prev_indirect_vdev);
886
887 /*
888 * Look for the 'not present' flag. This will only be set if the device
889 * was not present at the time of import.
890 */
891 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
892 &vd->vdev_not_present);
893
894 /*
895 * Get the alignment requirement. Ignore pool ashift for vdev
896 * attach case.
897 */
898 if (alloctype != VDEV_ALLOC_ATTACH) {
899 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT,
900 &vd->vdev_ashift);
901 } else {
902 vd->vdev_attaching = B_TRUE;
903 }
904
905 /*
906 * Retrieve the vdev creation time.
907 */
908 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG,
909 &vd->vdev_crtxg);
910
911 if (vd->vdev_ops == &vdev_root_ops &&
912 (alloctype == VDEV_ALLOC_LOAD ||
913 alloctype == VDEV_ALLOC_SPLIT ||
914 alloctype == VDEV_ALLOC_ROOTPOOL)) {
915 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_VDEV_ROOT_ZAP,
916 &vd->vdev_root_zap);
917 }
918
919 /*
920 * If we're a top-level vdev, try to load the allocation parameters.
921 */
922 if (top_level &&
923 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
924 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
925 &vd->vdev_ms_array);
926 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
927 &vd->vdev_ms_shift);
928 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE,
929 &vd->vdev_asize);
930 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NONALLOCATING,
931 &vd->vdev_noalloc);
932 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING,
933 &vd->vdev_removing);
934 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP,
935 &vd->vdev_top_zap);
936 vd->vdev_rz_expanding = nvlist_exists(nv,
937 ZPOOL_CONFIG_RAIDZ_EXPANDING);
938 } else {
939 ASSERT0(vd->vdev_top_zap);
940 }
941
942 if (top_level && alloctype != VDEV_ALLOC_ATTACH) {
943 ASSERT(alloctype == VDEV_ALLOC_LOAD ||
944 alloctype == VDEV_ALLOC_ADD ||
945 alloctype == VDEV_ALLOC_SPLIT ||
946 alloctype == VDEV_ALLOC_ROOTPOOL);
947 /* Note: metaslab_group_create() is now deferred */
948 }
949
950 if (vd->vdev_ops->vdev_op_leaf &&
951 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
952 (void) nvlist_lookup_uint64(nv,
953 ZPOOL_CONFIG_VDEV_LEAF_ZAP, &vd->vdev_leaf_zap);
954 } else {
955 ASSERT0(vd->vdev_leaf_zap);
956 }
957
958 /*
959 * If we're a leaf vdev, try to load the DTL object and other state.
960 */
961
962 if (vd->vdev_ops->vdev_op_leaf &&
963 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE ||
964 alloctype == VDEV_ALLOC_ROOTPOOL)) {
965 if (alloctype == VDEV_ALLOC_LOAD) {
966 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL,
967 &vd->vdev_dtl_object);
968 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE,
969 &vd->vdev_unspare);
970 }
971
972 if (alloctype == VDEV_ALLOC_ROOTPOOL) {
973 uint64_t spare = 0;
974
975 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
976 &spare) == 0 && spare)
977 spa_spare_add(vd);
978 }
979
980 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE,
981 &vd->vdev_offline);
982
983 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
984 &vd->vdev_resilver_txg);
985
986 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REBUILD_TXG,
987 &vd->vdev_rebuild_txg);
988
989 if (nvlist_exists(nv, ZPOOL_CONFIG_RESILVER_DEFER))
990 vdev_defer_resilver(vd);
991
992 /*
993 * In general, when importing a pool we want to ignore the
994 * persistent fault state, as the diagnosis made on another
995 * system may not be valid in the current context. The only
996 * exception is if we forced a vdev to a persistently faulted
997 * state with 'zpool offline -f'. The persistent fault will
998 * remain across imports until cleared.
999 *
1000 * Local vdevs will remain in the faulted state.
1001 */
1002 if (spa_load_state(spa) == SPA_LOAD_OPEN ||
1003 spa_load_state(spa) == SPA_LOAD_IMPORT) {
1004 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED,
1005 &vd->vdev_faulted);
1006 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED,
1007 &vd->vdev_degraded);
1008 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED,
1009 &vd->vdev_removed);
1010
1011 if (vd->vdev_faulted || vd->vdev_degraded) {
1012 const char *aux;
1013
1014 vd->vdev_label_aux =
1015 VDEV_AUX_ERR_EXCEEDED;
1016 if (nvlist_lookup_string(nv,
1017 ZPOOL_CONFIG_AUX_STATE, &aux) == 0 &&
1018 strcmp(aux, "external") == 0)
1019 vd->vdev_label_aux = VDEV_AUX_EXTERNAL;
1020 else
1021 vd->vdev_faulted = 0ULL;
1022 }
1023 }
1024 }
1025
1026 /*
1027 * Add ourselves to the parent's list of children.
1028 */
1029 vdev_add_child(parent, vd);
1030
1031 *vdp = vd;
1032
1033 return (0);
1034 }
1035
1036 void
vdev_free(vdev_t * vd)1037 vdev_free(vdev_t *vd)
1038 {
1039 spa_t *spa = vd->vdev_spa;
1040
1041 ASSERT3P(vd->vdev_initialize_thread, ==, NULL);
1042 ASSERT3P(vd->vdev_trim_thread, ==, NULL);
1043 ASSERT3P(vd->vdev_autotrim_thread, ==, NULL);
1044 ASSERT3P(vd->vdev_rebuild_thread, ==, NULL);
1045
1046 /*
1047 * Scan queues are normally destroyed at the end of a scan. If the
1048 * queue exists here, that implies the vdev is being removed while
1049 * the scan is still running.
1050 */
1051 if (vd->vdev_scan_io_queue != NULL) {
1052 mutex_enter(&vd->vdev_scan_io_queue_lock);
1053 dsl_scan_io_queue_destroy(vd->vdev_scan_io_queue);
1054 vd->vdev_scan_io_queue = NULL;
1055 mutex_exit(&vd->vdev_scan_io_queue_lock);
1056 }
1057
1058 /*
1059 * vdev_free() implies closing the vdev first. This is simpler than
1060 * trying to ensure complicated semantics for all callers.
1061 */
1062 vdev_close(vd);
1063
1064 ASSERT(!list_link_active(&vd->vdev_config_dirty_node));
1065 ASSERT(!list_link_active(&vd->vdev_state_dirty_node));
1066
1067 /*
1068 * Free all children.
1069 */
1070 for (int c = 0; c < vd->vdev_children; c++)
1071 vdev_free(vd->vdev_child[c]);
1072
1073 ASSERT(vd->vdev_child == NULL);
1074 ASSERT(vd->vdev_guid_sum == vd->vdev_guid);
1075
1076 if (vd->vdev_ops->vdev_op_fini != NULL)
1077 vd->vdev_ops->vdev_op_fini(vd);
1078
1079 /*
1080 * Discard allocation state.
1081 */
1082 if (vd->vdev_mg != NULL) {
1083 vdev_metaslab_fini(vd);
1084 metaslab_group_destroy(vd->vdev_mg);
1085 vd->vdev_mg = NULL;
1086 }
1087 if (vd->vdev_log_mg != NULL) {
1088 ASSERT0(vd->vdev_ms_count);
1089 metaslab_group_destroy(vd->vdev_log_mg);
1090 vd->vdev_log_mg = NULL;
1091 }
1092
1093 ASSERT0(vd->vdev_stat.vs_space);
1094 ASSERT0(vd->vdev_stat.vs_dspace);
1095 ASSERT0(vd->vdev_stat.vs_alloc);
1096
1097 /*
1098 * Remove this vdev from its parent's child list.
1099 */
1100 vdev_remove_child(vd->vdev_parent, vd);
1101
1102 ASSERT(vd->vdev_parent == NULL);
1103 ASSERT(!list_link_active(&vd->vdev_leaf_node));
1104
1105 /*
1106 * Clean up vdev structure.
1107 */
1108 vdev_queue_fini(vd);
1109
1110 if (vd->vdev_path)
1111 spa_strfree(vd->vdev_path);
1112 if (vd->vdev_devid)
1113 spa_strfree(vd->vdev_devid);
1114 if (vd->vdev_physpath)
1115 spa_strfree(vd->vdev_physpath);
1116
1117 if (vd->vdev_enc_sysfs_path)
1118 spa_strfree(vd->vdev_enc_sysfs_path);
1119
1120 if (vd->vdev_fru)
1121 spa_strfree(vd->vdev_fru);
1122
1123 if (vd->vdev_isspare)
1124 spa_spare_remove(vd);
1125 if (vd->vdev_isl2cache)
1126 spa_l2cache_remove(vd);
1127
1128 txg_list_destroy(&vd->vdev_ms_list);
1129 txg_list_destroy(&vd->vdev_dtl_list);
1130
1131 mutex_enter(&vd->vdev_dtl_lock);
1132 space_map_close(vd->vdev_dtl_sm);
1133 for (int t = 0; t < DTL_TYPES; t++) {
1134 range_tree_vacate(vd->vdev_dtl[t], NULL, NULL);
1135 range_tree_destroy(vd->vdev_dtl[t]);
1136 }
1137 mutex_exit(&vd->vdev_dtl_lock);
1138
1139 EQUIV(vd->vdev_indirect_births != NULL,
1140 vd->vdev_indirect_mapping != NULL);
1141 if (vd->vdev_indirect_births != NULL) {
1142 vdev_indirect_mapping_close(vd->vdev_indirect_mapping);
1143 vdev_indirect_births_close(vd->vdev_indirect_births);
1144 }
1145
1146 if (vd->vdev_obsolete_sm != NULL) {
1147 ASSERT(vd->vdev_removing ||
1148 vd->vdev_ops == &vdev_indirect_ops);
1149 space_map_close(vd->vdev_obsolete_sm);
1150 vd->vdev_obsolete_sm = NULL;
1151 }
1152 range_tree_destroy(vd->vdev_obsolete_segments);
1153 rw_destroy(&vd->vdev_indirect_rwlock);
1154 mutex_destroy(&vd->vdev_obsolete_lock);
1155
1156 mutex_destroy(&vd->vdev_dtl_lock);
1157 mutex_destroy(&vd->vdev_stat_lock);
1158 mutex_destroy(&vd->vdev_probe_lock);
1159 mutex_destroy(&vd->vdev_scan_io_queue_lock);
1160
1161 mutex_destroy(&vd->vdev_initialize_lock);
1162 mutex_destroy(&vd->vdev_initialize_io_lock);
1163 cv_destroy(&vd->vdev_initialize_io_cv);
1164 cv_destroy(&vd->vdev_initialize_cv);
1165
1166 mutex_destroy(&vd->vdev_trim_lock);
1167 mutex_destroy(&vd->vdev_autotrim_lock);
1168 mutex_destroy(&vd->vdev_trim_io_lock);
1169 cv_destroy(&vd->vdev_trim_cv);
1170 cv_destroy(&vd->vdev_autotrim_cv);
1171 cv_destroy(&vd->vdev_autotrim_kick_cv);
1172 cv_destroy(&vd->vdev_trim_io_cv);
1173
1174 mutex_destroy(&vd->vdev_rebuild_lock);
1175 cv_destroy(&vd->vdev_rebuild_cv);
1176
1177 zfs_ratelimit_fini(&vd->vdev_delay_rl);
1178 zfs_ratelimit_fini(&vd->vdev_deadman_rl);
1179 zfs_ratelimit_fini(&vd->vdev_checksum_rl);
1180
1181 if (vd == spa->spa_root_vdev)
1182 spa->spa_root_vdev = NULL;
1183
1184 kmem_free(vd, sizeof (vdev_t));
1185 }
1186
1187 /*
1188 * Transfer top-level vdev state from svd to tvd.
1189 */
1190 static void
vdev_top_transfer(vdev_t * svd,vdev_t * tvd)1191 vdev_top_transfer(vdev_t *svd, vdev_t *tvd)
1192 {
1193 spa_t *spa = svd->vdev_spa;
1194 metaslab_t *msp;
1195 vdev_t *vd;
1196 int t;
1197
1198 ASSERT(tvd == tvd->vdev_top);
1199
1200 tvd->vdev_ms_array = svd->vdev_ms_array;
1201 tvd->vdev_ms_shift = svd->vdev_ms_shift;
1202 tvd->vdev_ms_count = svd->vdev_ms_count;
1203 tvd->vdev_top_zap = svd->vdev_top_zap;
1204
1205 svd->vdev_ms_array = 0;
1206 svd->vdev_ms_shift = 0;
1207 svd->vdev_ms_count = 0;
1208 svd->vdev_top_zap = 0;
1209
1210 if (tvd->vdev_mg)
1211 ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg);
1212 if (tvd->vdev_log_mg)
1213 ASSERT3P(tvd->vdev_log_mg, ==, svd->vdev_log_mg);
1214 tvd->vdev_mg = svd->vdev_mg;
1215 tvd->vdev_log_mg = svd->vdev_log_mg;
1216 tvd->vdev_ms = svd->vdev_ms;
1217
1218 svd->vdev_mg = NULL;
1219 svd->vdev_log_mg = NULL;
1220 svd->vdev_ms = NULL;
1221
1222 if (tvd->vdev_mg != NULL)
1223 tvd->vdev_mg->mg_vd = tvd;
1224 if (tvd->vdev_log_mg != NULL)
1225 tvd->vdev_log_mg->mg_vd = tvd;
1226
1227 tvd->vdev_checkpoint_sm = svd->vdev_checkpoint_sm;
1228 svd->vdev_checkpoint_sm = NULL;
1229
1230 tvd->vdev_alloc_bias = svd->vdev_alloc_bias;
1231 svd->vdev_alloc_bias = VDEV_BIAS_NONE;
1232
1233 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc;
1234 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space;
1235 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace;
1236
1237 svd->vdev_stat.vs_alloc = 0;
1238 svd->vdev_stat.vs_space = 0;
1239 svd->vdev_stat.vs_dspace = 0;
1240
1241 /*
1242 * State which may be set on a top-level vdev that's in the
1243 * process of being removed.
1244 */
1245 ASSERT0(tvd->vdev_indirect_config.vic_births_object);
1246 ASSERT0(tvd->vdev_indirect_config.vic_mapping_object);
1247 ASSERT3U(tvd->vdev_indirect_config.vic_prev_indirect_vdev, ==, -1ULL);
1248 ASSERT3P(tvd->vdev_indirect_mapping, ==, NULL);
1249 ASSERT3P(tvd->vdev_indirect_births, ==, NULL);
1250 ASSERT3P(tvd->vdev_obsolete_sm, ==, NULL);
1251 ASSERT0(tvd->vdev_noalloc);
1252 ASSERT0(tvd->vdev_removing);
1253 ASSERT0(tvd->vdev_rebuilding);
1254 tvd->vdev_noalloc = svd->vdev_noalloc;
1255 tvd->vdev_removing = svd->vdev_removing;
1256 tvd->vdev_rebuilding = svd->vdev_rebuilding;
1257 tvd->vdev_rebuild_config = svd->vdev_rebuild_config;
1258 tvd->vdev_indirect_config = svd->vdev_indirect_config;
1259 tvd->vdev_indirect_mapping = svd->vdev_indirect_mapping;
1260 tvd->vdev_indirect_births = svd->vdev_indirect_births;
1261 range_tree_swap(&svd->vdev_obsolete_segments,
1262 &tvd->vdev_obsolete_segments);
1263 tvd->vdev_obsolete_sm = svd->vdev_obsolete_sm;
1264 svd->vdev_indirect_config.vic_mapping_object = 0;
1265 svd->vdev_indirect_config.vic_births_object = 0;
1266 svd->vdev_indirect_config.vic_prev_indirect_vdev = -1ULL;
1267 svd->vdev_indirect_mapping = NULL;
1268 svd->vdev_indirect_births = NULL;
1269 svd->vdev_obsolete_sm = NULL;
1270 svd->vdev_noalloc = 0;
1271 svd->vdev_removing = 0;
1272 svd->vdev_rebuilding = 0;
1273
1274 for (t = 0; t < TXG_SIZE; t++) {
1275 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL)
1276 (void) txg_list_add(&tvd->vdev_ms_list, msp, t);
1277 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL)
1278 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t);
1279 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t))
1280 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t);
1281 }
1282
1283 if (list_link_active(&svd->vdev_config_dirty_node)) {
1284 vdev_config_clean(svd);
1285 vdev_config_dirty(tvd);
1286 }
1287
1288 if (list_link_active(&svd->vdev_state_dirty_node)) {
1289 vdev_state_clean(svd);
1290 vdev_state_dirty(tvd);
1291 }
1292
1293 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio;
1294 svd->vdev_deflate_ratio = 0;
1295
1296 tvd->vdev_islog = svd->vdev_islog;
1297 svd->vdev_islog = 0;
1298
1299 dsl_scan_io_queue_vdev_xfer(svd, tvd);
1300 }
1301
1302 static void
vdev_top_update(vdev_t * tvd,vdev_t * vd)1303 vdev_top_update(vdev_t *tvd, vdev_t *vd)
1304 {
1305 if (vd == NULL)
1306 return;
1307
1308 vd->vdev_top = tvd;
1309
1310 for (int c = 0; c < vd->vdev_children; c++)
1311 vdev_top_update(tvd, vd->vdev_child[c]);
1312 }
1313
1314 /*
1315 * Add a mirror/replacing vdev above an existing vdev. There is no need to
1316 * call .vdev_op_init() since mirror/replacing vdevs do not have private state.
1317 */
1318 vdev_t *
vdev_add_parent(vdev_t * cvd,vdev_ops_t * ops)1319 vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops)
1320 {
1321 spa_t *spa = cvd->vdev_spa;
1322 vdev_t *pvd = cvd->vdev_parent;
1323 vdev_t *mvd;
1324
1325 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1326
1327 mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops);
1328
1329 mvd->vdev_asize = cvd->vdev_asize;
1330 mvd->vdev_min_asize = cvd->vdev_min_asize;
1331 mvd->vdev_max_asize = cvd->vdev_max_asize;
1332 mvd->vdev_psize = cvd->vdev_psize;
1333 mvd->vdev_ashift = cvd->vdev_ashift;
1334 mvd->vdev_logical_ashift = cvd->vdev_logical_ashift;
1335 mvd->vdev_physical_ashift = cvd->vdev_physical_ashift;
1336 mvd->vdev_state = cvd->vdev_state;
1337 mvd->vdev_crtxg = cvd->vdev_crtxg;
1338
1339 vdev_remove_child(pvd, cvd);
1340 vdev_add_child(pvd, mvd);
1341 cvd->vdev_id = mvd->vdev_children;
1342 vdev_add_child(mvd, cvd);
1343 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
1344
1345 if (mvd == mvd->vdev_top)
1346 vdev_top_transfer(cvd, mvd);
1347
1348 return (mvd);
1349 }
1350
1351 /*
1352 * Remove a 1-way mirror/replacing vdev from the tree.
1353 */
1354 void
vdev_remove_parent(vdev_t * cvd)1355 vdev_remove_parent(vdev_t *cvd)
1356 {
1357 vdev_t *mvd = cvd->vdev_parent;
1358 vdev_t *pvd = mvd->vdev_parent;
1359
1360 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1361
1362 ASSERT(mvd->vdev_children == 1);
1363 ASSERT(mvd->vdev_ops == &vdev_mirror_ops ||
1364 mvd->vdev_ops == &vdev_replacing_ops ||
1365 mvd->vdev_ops == &vdev_spare_ops);
1366 cvd->vdev_ashift = mvd->vdev_ashift;
1367 cvd->vdev_logical_ashift = mvd->vdev_logical_ashift;
1368 cvd->vdev_physical_ashift = mvd->vdev_physical_ashift;
1369 vdev_remove_child(mvd, cvd);
1370 vdev_remove_child(pvd, mvd);
1371
1372 /*
1373 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
1374 * Otherwise, we could have detached an offline device, and when we
1375 * go to import the pool we'll think we have two top-level vdevs,
1376 * instead of a different version of the same top-level vdev.
1377 */
1378 if (mvd->vdev_top == mvd) {
1379 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid;
1380 cvd->vdev_orig_guid = cvd->vdev_guid;
1381 cvd->vdev_guid += guid_delta;
1382 cvd->vdev_guid_sum += guid_delta;
1383
1384 /*
1385 * If pool not set for autoexpand, we need to also preserve
1386 * mvd's asize to prevent automatic expansion of cvd.
1387 * Otherwise if we are adjusting the mirror by attaching and
1388 * detaching children of non-uniform sizes, the mirror could
1389 * autoexpand, unexpectedly requiring larger devices to
1390 * re-establish the mirror.
1391 */
1392 if (!cvd->vdev_spa->spa_autoexpand)
1393 cvd->vdev_asize = mvd->vdev_asize;
1394 }
1395 cvd->vdev_id = mvd->vdev_id;
1396 vdev_add_child(pvd, cvd);
1397 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
1398
1399 if (cvd == cvd->vdev_top)
1400 vdev_top_transfer(mvd, cvd);
1401
1402 ASSERT(mvd->vdev_children == 0);
1403 vdev_free(mvd);
1404 }
1405
1406 /*
1407 * Choose GCD for spa_gcd_alloc.
1408 */
1409 static uint64_t
vdev_gcd(uint64_t a,uint64_t b)1410 vdev_gcd(uint64_t a, uint64_t b)
1411 {
1412 while (b != 0) {
1413 uint64_t t = b;
1414 b = a % b;
1415 a = t;
1416 }
1417 return (a);
1418 }
1419
1420 /*
1421 * Set spa_min_alloc and spa_gcd_alloc.
1422 */
1423 static void
vdev_spa_set_alloc(spa_t * spa,uint64_t min_alloc)1424 vdev_spa_set_alloc(spa_t *spa, uint64_t min_alloc)
1425 {
1426 if (min_alloc < spa->spa_min_alloc)
1427 spa->spa_min_alloc = min_alloc;
1428 if (spa->spa_gcd_alloc == INT_MAX) {
1429 spa->spa_gcd_alloc = min_alloc;
1430 } else {
1431 spa->spa_gcd_alloc = vdev_gcd(min_alloc,
1432 spa->spa_gcd_alloc);
1433 }
1434 }
1435
1436 void
vdev_metaslab_group_create(vdev_t * vd)1437 vdev_metaslab_group_create(vdev_t *vd)
1438 {
1439 spa_t *spa = vd->vdev_spa;
1440
1441 /*
1442 * metaslab_group_create was delayed until allocation bias was available
1443 */
1444 if (vd->vdev_mg == NULL) {
1445 metaslab_class_t *mc;
1446
1447 if (vd->vdev_islog && vd->vdev_alloc_bias == VDEV_BIAS_NONE)
1448 vd->vdev_alloc_bias = VDEV_BIAS_LOG;
1449
1450 ASSERT3U(vd->vdev_islog, ==,
1451 (vd->vdev_alloc_bias == VDEV_BIAS_LOG));
1452
1453 switch (vd->vdev_alloc_bias) {
1454 case VDEV_BIAS_LOG:
1455 mc = spa_log_class(spa);
1456 break;
1457 case VDEV_BIAS_SPECIAL:
1458 mc = spa_special_class(spa);
1459 break;
1460 case VDEV_BIAS_DEDUP:
1461 mc = spa_dedup_class(spa);
1462 break;
1463 default:
1464 mc = spa_normal_class(spa);
1465 }
1466
1467 vd->vdev_mg = metaslab_group_create(mc, vd,
1468 spa->spa_alloc_count);
1469
1470 if (!vd->vdev_islog) {
1471 vd->vdev_log_mg = metaslab_group_create(
1472 spa_embedded_log_class(spa), vd, 1);
1473 }
1474
1475 /*
1476 * The spa ashift min/max only apply for the normal metaslab
1477 * class. Class destination is late binding so ashift boundary
1478 * setting had to wait until now.
1479 */
1480 if (vd->vdev_top == vd && vd->vdev_ashift != 0 &&
1481 mc == spa_normal_class(spa) && vd->vdev_aux == NULL) {
1482 if (vd->vdev_ashift > spa->spa_max_ashift)
1483 spa->spa_max_ashift = vd->vdev_ashift;
1484 if (vd->vdev_ashift < spa->spa_min_ashift)
1485 spa->spa_min_ashift = vd->vdev_ashift;
1486
1487 uint64_t min_alloc = vdev_get_min_alloc(vd);
1488 vdev_spa_set_alloc(spa, min_alloc);
1489 }
1490 }
1491 }
1492
1493 int
vdev_metaslab_init(vdev_t * vd,uint64_t txg)1494 vdev_metaslab_init(vdev_t *vd, uint64_t txg)
1495 {
1496 spa_t *spa = vd->vdev_spa;
1497 uint64_t oldc = vd->vdev_ms_count;
1498 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift;
1499 metaslab_t **mspp;
1500 int error;
1501 boolean_t expanding = (oldc != 0);
1502
1503 ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1504
1505 /*
1506 * This vdev is not being allocated from yet or is a hole.
1507 */
1508 if (vd->vdev_ms_shift == 0)
1509 return (0);
1510
1511 ASSERT(!vd->vdev_ishole);
1512
1513 ASSERT(oldc <= newc);
1514
1515 mspp = vmem_zalloc(newc * sizeof (*mspp), KM_SLEEP);
1516
1517 if (expanding) {
1518 memcpy(mspp, vd->vdev_ms, oldc * sizeof (*mspp));
1519 vmem_free(vd->vdev_ms, oldc * sizeof (*mspp));
1520 }
1521
1522 vd->vdev_ms = mspp;
1523 vd->vdev_ms_count = newc;
1524
1525 for (uint64_t m = oldc; m < newc; m++) {
1526 uint64_t object = 0;
1527 /*
1528 * vdev_ms_array may be 0 if we are creating the "fake"
1529 * metaslabs for an indirect vdev for zdb's leak detection.
1530 * See zdb_leak_init().
1531 */
1532 if (txg == 0 && vd->vdev_ms_array != 0) {
1533 error = dmu_read(spa->spa_meta_objset,
1534 vd->vdev_ms_array,
1535 m * sizeof (uint64_t), sizeof (uint64_t), &object,
1536 DMU_READ_PREFETCH);
1537 if (error != 0) {
1538 vdev_dbgmsg(vd, "unable to read the metaslab "
1539 "array [error=%d]", error);
1540 return (error);
1541 }
1542 }
1543
1544 error = metaslab_init(vd->vdev_mg, m, object, txg,
1545 &(vd->vdev_ms[m]));
1546 if (error != 0) {
1547 vdev_dbgmsg(vd, "metaslab_init failed [error=%d]",
1548 error);
1549 return (error);
1550 }
1551 }
1552
1553 /*
1554 * Find the emptiest metaslab on the vdev and mark it for use for
1555 * embedded slog by moving it from the regular to the log metaslab
1556 * group.
1557 */
1558 if (vd->vdev_mg->mg_class == spa_normal_class(spa) &&
1559 vd->vdev_ms_count > zfs_embedded_slog_min_ms &&
1560 avl_is_empty(&vd->vdev_log_mg->mg_metaslab_tree)) {
1561 uint64_t slog_msid = 0;
1562 uint64_t smallest = UINT64_MAX;
1563
1564 /*
1565 * Note, we only search the new metaslabs, because the old
1566 * (pre-existing) ones may be active (e.g. have non-empty
1567 * range_tree's), and we don't move them to the new
1568 * metaslab_t.
1569 */
1570 for (uint64_t m = oldc; m < newc; m++) {
1571 uint64_t alloc =
1572 space_map_allocated(vd->vdev_ms[m]->ms_sm);
1573 if (alloc < smallest) {
1574 slog_msid = m;
1575 smallest = alloc;
1576 }
1577 }
1578 metaslab_t *slog_ms = vd->vdev_ms[slog_msid];
1579 /*
1580 * The metaslab was marked as dirty at the end of
1581 * metaslab_init(). Remove it from the dirty list so that we
1582 * can uninitialize and reinitialize it to the new class.
1583 */
1584 if (txg != 0) {
1585 (void) txg_list_remove_this(&vd->vdev_ms_list,
1586 slog_ms, txg);
1587 }
1588 uint64_t sm_obj = space_map_object(slog_ms->ms_sm);
1589 metaslab_fini(slog_ms);
1590 VERIFY0(metaslab_init(vd->vdev_log_mg, slog_msid, sm_obj, txg,
1591 &vd->vdev_ms[slog_msid]));
1592 }
1593
1594 if (txg == 0)
1595 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER);
1596
1597 /*
1598 * If the vdev is marked as non-allocating then don't
1599 * activate the metaslabs since we want to ensure that
1600 * no allocations are performed on this device.
1601 */
1602 if (vd->vdev_noalloc) {
1603 /* track non-allocating vdev space */
1604 spa->spa_nonallocating_dspace += spa_deflate(spa) ?
1605 vd->vdev_stat.vs_dspace : vd->vdev_stat.vs_space;
1606 } else if (!expanding) {
1607 metaslab_group_activate(vd->vdev_mg);
1608 if (vd->vdev_log_mg != NULL)
1609 metaslab_group_activate(vd->vdev_log_mg);
1610 }
1611
1612 if (txg == 0)
1613 spa_config_exit(spa, SCL_ALLOC, FTAG);
1614
1615 return (0);
1616 }
1617
1618 void
vdev_metaslab_fini(vdev_t * vd)1619 vdev_metaslab_fini(vdev_t *vd)
1620 {
1621 if (vd->vdev_checkpoint_sm != NULL) {
1622 ASSERT(spa_feature_is_active(vd->vdev_spa,
1623 SPA_FEATURE_POOL_CHECKPOINT));
1624 space_map_close(vd->vdev_checkpoint_sm);
1625 /*
1626 * Even though we close the space map, we need to set its
1627 * pointer to NULL. The reason is that vdev_metaslab_fini()
1628 * may be called multiple times for certain operations
1629 * (i.e. when destroying a pool) so we need to ensure that
1630 * this clause never executes twice. This logic is similar
1631 * to the one used for the vdev_ms clause below.
1632 */
1633 vd->vdev_checkpoint_sm = NULL;
1634 }
1635
1636 if (vd->vdev_ms != NULL) {
1637 metaslab_group_t *mg = vd->vdev_mg;
1638
1639 metaslab_group_passivate(mg);
1640 if (vd->vdev_log_mg != NULL) {
1641 ASSERT(!vd->vdev_islog);
1642 metaslab_group_passivate(vd->vdev_log_mg);
1643 }
1644
1645 uint64_t count = vd->vdev_ms_count;
1646 for (uint64_t m = 0; m < count; m++) {
1647 metaslab_t *msp = vd->vdev_ms[m];
1648 if (msp != NULL)
1649 metaslab_fini(msp);
1650 }
1651 vmem_free(vd->vdev_ms, count * sizeof (metaslab_t *));
1652 vd->vdev_ms = NULL;
1653 vd->vdev_ms_count = 0;
1654
1655 for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) {
1656 ASSERT0(mg->mg_histogram[i]);
1657 if (vd->vdev_log_mg != NULL)
1658 ASSERT0(vd->vdev_log_mg->mg_histogram[i]);
1659 }
1660 }
1661 ASSERT0(vd->vdev_ms_count);
1662 }
1663
1664 typedef struct vdev_probe_stats {
1665 boolean_t vps_readable;
1666 boolean_t vps_writeable;
1667 boolean_t vps_zio_done_probe;
1668 int vps_flags;
1669 } vdev_probe_stats_t;
1670
1671 static void
vdev_probe_done(zio_t * zio)1672 vdev_probe_done(zio_t *zio)
1673 {
1674 spa_t *spa = zio->io_spa;
1675 vdev_t *vd = zio->io_vd;
1676 vdev_probe_stats_t *vps = zio->io_private;
1677
1678 ASSERT(vd->vdev_probe_zio != NULL);
1679
1680 if (zio->io_type == ZIO_TYPE_READ) {
1681 if (zio->io_error == 0)
1682 vps->vps_readable = 1;
1683 if (zio->io_error == 0 && spa_writeable(spa)) {
1684 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd,
1685 zio->io_offset, zio->io_size, zio->io_abd,
1686 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1687 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE));
1688 } else {
1689 abd_free(zio->io_abd);
1690 }
1691 } else if (zio->io_type == ZIO_TYPE_WRITE) {
1692 if (zio->io_error == 0)
1693 vps->vps_writeable = 1;
1694 abd_free(zio->io_abd);
1695 } else if (zio->io_type == ZIO_TYPE_NULL) {
1696 zio_t *pio;
1697 zio_link_t *zl;
1698
1699 vd->vdev_cant_read |= !vps->vps_readable;
1700 vd->vdev_cant_write |= !vps->vps_writeable;
1701 vdev_dbgmsg(vd, "probe done, cant_read=%u cant_write=%u",
1702 vd->vdev_cant_read, vd->vdev_cant_write);
1703
1704 if (vdev_readable(vd) &&
1705 (vdev_writeable(vd) || !spa_writeable(spa))) {
1706 zio->io_error = 0;
1707 } else {
1708 ASSERT(zio->io_error != 0);
1709 vdev_dbgmsg(vd, "failed probe");
1710 (void) zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE,
1711 spa, vd, NULL, NULL, 0);
1712 zio->io_error = SET_ERROR(ENXIO);
1713
1714 /*
1715 * If this probe was initiated from zio pipeline, then
1716 * change the state in a spa_async_request. Probes that
1717 * were initiated from a vdev_open can change the state
1718 * as part of the open call.
1719 */
1720 if (vps->vps_zio_done_probe) {
1721 vd->vdev_fault_wanted = B_TRUE;
1722 spa_async_request(spa, SPA_ASYNC_FAULT_VDEV);
1723 }
1724 }
1725
1726 mutex_enter(&vd->vdev_probe_lock);
1727 ASSERT(vd->vdev_probe_zio == zio);
1728 vd->vdev_probe_zio = NULL;
1729 mutex_exit(&vd->vdev_probe_lock);
1730
1731 zl = NULL;
1732 while ((pio = zio_walk_parents(zio, &zl)) != NULL)
1733 if (!vdev_accessible(vd, pio))
1734 pio->io_error = SET_ERROR(ENXIO);
1735
1736 kmem_free(vps, sizeof (*vps));
1737 }
1738 }
1739
1740 /*
1741 * Determine whether this device is accessible.
1742 *
1743 * Read and write to several known locations: the pad regions of each
1744 * vdev label but the first, which we leave alone in case it contains
1745 * a VTOC.
1746 */
1747 zio_t *
vdev_probe(vdev_t * vd,zio_t * zio)1748 vdev_probe(vdev_t *vd, zio_t *zio)
1749 {
1750 spa_t *spa = vd->vdev_spa;
1751 vdev_probe_stats_t *vps = NULL;
1752 zio_t *pio;
1753
1754 ASSERT(vd->vdev_ops->vdev_op_leaf);
1755
1756 /*
1757 * Don't probe the probe.
1758 */
1759 if (zio && (zio->io_flags & ZIO_FLAG_PROBE))
1760 return (NULL);
1761
1762 /*
1763 * To prevent 'probe storms' when a device fails, we create
1764 * just one probe i/o at a time. All zios that want to probe
1765 * this vdev will become parents of the probe io.
1766 */
1767 mutex_enter(&vd->vdev_probe_lock);
1768
1769 if ((pio = vd->vdev_probe_zio) == NULL) {
1770 vps = kmem_zalloc(sizeof (*vps), KM_SLEEP);
1771
1772 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE |
1773 ZIO_FLAG_DONT_AGGREGATE | ZIO_FLAG_TRYHARD;
1774 vps->vps_zio_done_probe = (zio != NULL);
1775
1776 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) {
1777 /*
1778 * vdev_cant_read and vdev_cant_write can only
1779 * transition from TRUE to FALSE when we have the
1780 * SCL_ZIO lock as writer; otherwise they can only
1781 * transition from FALSE to TRUE. This ensures that
1782 * any zio looking at these values can assume that
1783 * failures persist for the life of the I/O. That's
1784 * important because when a device has intermittent
1785 * connectivity problems, we want to ensure that
1786 * they're ascribed to the device (ENXIO) and not
1787 * the zio (EIO).
1788 *
1789 * Since we hold SCL_ZIO as writer here, clear both
1790 * values so the probe can reevaluate from first
1791 * principles.
1792 */
1793 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER;
1794 vd->vdev_cant_read = B_FALSE;
1795 vd->vdev_cant_write = B_FALSE;
1796 }
1797
1798 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd,
1799 vdev_probe_done, vps,
1800 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE);
1801 }
1802
1803 if (zio != NULL)
1804 zio_add_child(zio, pio);
1805
1806 mutex_exit(&vd->vdev_probe_lock);
1807
1808 if (vps == NULL) {
1809 ASSERT(zio != NULL);
1810 return (NULL);
1811 }
1812
1813 for (int l = 1; l < VDEV_LABELS; l++) {
1814 zio_nowait(zio_read_phys(pio, vd,
1815 vdev_label_offset(vd->vdev_psize, l,
1816 offsetof(vdev_label_t, vl_be)), VDEV_PAD_SIZE,
1817 abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE),
1818 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1819 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE));
1820 }
1821
1822 if (zio == NULL)
1823 return (pio);
1824
1825 zio_nowait(pio);
1826 return (NULL);
1827 }
1828
1829 static void
vdev_load_child(void * arg)1830 vdev_load_child(void *arg)
1831 {
1832 vdev_t *vd = arg;
1833
1834 vd->vdev_load_error = vdev_load(vd);
1835 }
1836
1837 static void
vdev_open_child(void * arg)1838 vdev_open_child(void *arg)
1839 {
1840 vdev_t *vd = arg;
1841
1842 vd->vdev_open_thread = curthread;
1843 vd->vdev_open_error = vdev_open(vd);
1844 vd->vdev_open_thread = NULL;
1845 }
1846
1847 static boolean_t
vdev_uses_zvols(vdev_t * vd)1848 vdev_uses_zvols(vdev_t *vd)
1849 {
1850 #ifdef _KERNEL
1851 if (zvol_is_zvol(vd->vdev_path))
1852 return (B_TRUE);
1853 #endif
1854
1855 for (int c = 0; c < vd->vdev_children; c++)
1856 if (vdev_uses_zvols(vd->vdev_child[c]))
1857 return (B_TRUE);
1858
1859 return (B_FALSE);
1860 }
1861
1862 /*
1863 * Returns B_TRUE if the passed child should be opened.
1864 */
1865 static boolean_t
vdev_default_open_children_func(vdev_t * vd)1866 vdev_default_open_children_func(vdev_t *vd)
1867 {
1868 (void) vd;
1869 return (B_TRUE);
1870 }
1871
1872 /*
1873 * Open the requested child vdevs. If any of the leaf vdevs are using
1874 * a ZFS volume then do the opens in a single thread. This avoids a
1875 * deadlock when the current thread is holding the spa_namespace_lock.
1876 */
1877 static void
vdev_open_children_impl(vdev_t * vd,vdev_open_children_func_t * open_func)1878 vdev_open_children_impl(vdev_t *vd, vdev_open_children_func_t *open_func)
1879 {
1880 int children = vd->vdev_children;
1881
1882 taskq_t *tq = taskq_create("vdev_open", children, minclsyspri,
1883 children, children, TASKQ_PREPOPULATE);
1884 vd->vdev_nonrot = B_TRUE;
1885
1886 for (int c = 0; c < children; c++) {
1887 vdev_t *cvd = vd->vdev_child[c];
1888
1889 if (open_func(cvd) == B_FALSE)
1890 continue;
1891
1892 if (tq == NULL || vdev_uses_zvols(vd)) {
1893 cvd->vdev_open_error = vdev_open(cvd);
1894 } else {
1895 VERIFY(taskq_dispatch(tq, vdev_open_child,
1896 cvd, TQ_SLEEP) != TASKQID_INVALID);
1897 }
1898
1899 vd->vdev_nonrot &= cvd->vdev_nonrot;
1900 }
1901
1902 if (tq != NULL) {
1903 taskq_wait(tq);
1904 taskq_destroy(tq);
1905 }
1906 }
1907
1908 /*
1909 * Open all child vdevs.
1910 */
1911 void
vdev_open_children(vdev_t * vd)1912 vdev_open_children(vdev_t *vd)
1913 {
1914 vdev_open_children_impl(vd, vdev_default_open_children_func);
1915 }
1916
1917 /*
1918 * Conditionally open a subset of child vdevs.
1919 */
1920 void
vdev_open_children_subset(vdev_t * vd,vdev_open_children_func_t * open_func)1921 vdev_open_children_subset(vdev_t *vd, vdev_open_children_func_t *open_func)
1922 {
1923 vdev_open_children_impl(vd, open_func);
1924 }
1925
1926 /*
1927 * Compute the raidz-deflation ratio. Note, we hard-code 128k (1 << 17)
1928 * because it is the "typical" blocksize. Even though SPA_MAXBLOCKSIZE
1929 * changed, this algorithm can not change, otherwise it would inconsistently
1930 * account for existing bp's. We also hard-code txg 0 for the same reason
1931 * since expanded RAIDZ vdevs can use a different asize for different birth
1932 * txg's.
1933 */
1934 static void
vdev_set_deflate_ratio(vdev_t * vd)1935 vdev_set_deflate_ratio(vdev_t *vd)
1936 {
1937 if (vd == vd->vdev_top && !vd->vdev_ishole && vd->vdev_ashift != 0) {
1938 vd->vdev_deflate_ratio = (1 << 17) /
1939 (vdev_psize_to_asize_txg(vd, 1 << 17, 0) >>
1940 SPA_MINBLOCKSHIFT);
1941 }
1942 }
1943
1944 /*
1945 * Choose the best of two ashifts, preferring one between logical ashift
1946 * (absolute minimum) and administrator defined maximum, otherwise take
1947 * the biggest of the two.
1948 */
1949 uint64_t
vdev_best_ashift(uint64_t logical,uint64_t a,uint64_t b)1950 vdev_best_ashift(uint64_t logical, uint64_t a, uint64_t b)
1951 {
1952 if (a > logical && a <= zfs_vdev_max_auto_ashift) {
1953 if (b <= logical || b > zfs_vdev_max_auto_ashift)
1954 return (a);
1955 else
1956 return (MAX(a, b));
1957 } else if (b <= logical || b > zfs_vdev_max_auto_ashift)
1958 return (MAX(a, b));
1959 return (b);
1960 }
1961
1962 /*
1963 * Maximize performance by inflating the configured ashift for top level
1964 * vdevs to be as close to the physical ashift as possible while maintaining
1965 * administrator defined limits and ensuring it doesn't go below the
1966 * logical ashift.
1967 */
1968 static void
vdev_ashift_optimize(vdev_t * vd)1969 vdev_ashift_optimize(vdev_t *vd)
1970 {
1971 ASSERT(vd == vd->vdev_top);
1972
1973 if (vd->vdev_ashift < vd->vdev_physical_ashift &&
1974 vd->vdev_physical_ashift <= zfs_vdev_max_auto_ashift) {
1975 vd->vdev_ashift = MIN(
1976 MAX(zfs_vdev_max_auto_ashift, vd->vdev_ashift),
1977 MAX(zfs_vdev_min_auto_ashift,
1978 vd->vdev_physical_ashift));
1979 } else {
1980 /*
1981 * If the logical and physical ashifts are the same, then
1982 * we ensure that the top-level vdev's ashift is not smaller
1983 * than our minimum ashift value. For the unusual case
1984 * where logical ashift > physical ashift, we can't cap
1985 * the calculated ashift based on max ashift as that
1986 * would cause failures.
1987 * We still check if we need to increase it to match
1988 * the min ashift.
1989 */
1990 vd->vdev_ashift = MAX(zfs_vdev_min_auto_ashift,
1991 vd->vdev_ashift);
1992 }
1993 }
1994
1995 /*
1996 * Prepare a virtual device for access.
1997 */
1998 int
vdev_open(vdev_t * vd)1999 vdev_open(vdev_t *vd)
2000 {
2001 spa_t *spa = vd->vdev_spa;
2002 int error;
2003 uint64_t osize = 0;
2004 uint64_t max_osize = 0;
2005 uint64_t asize, max_asize, psize;
2006 uint64_t logical_ashift = 0;
2007 uint64_t physical_ashift = 0;
2008
2009 ASSERT(vd->vdev_open_thread == curthread ||
2010 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2011 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED ||
2012 vd->vdev_state == VDEV_STATE_CANT_OPEN ||
2013 vd->vdev_state == VDEV_STATE_OFFLINE);
2014
2015 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
2016 vd->vdev_cant_read = B_FALSE;
2017 vd->vdev_cant_write = B_FALSE;
2018 vd->vdev_min_asize = vdev_get_min_asize(vd);
2019
2020 /*
2021 * If this vdev is not removed, check its fault status. If it's
2022 * faulted, bail out of the open.
2023 */
2024 if (!vd->vdev_removed && vd->vdev_faulted) {
2025 ASSERT(vd->vdev_children == 0);
2026 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
2027 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
2028 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
2029 vd->vdev_label_aux);
2030 return (SET_ERROR(ENXIO));
2031 } else if (vd->vdev_offline) {
2032 ASSERT(vd->vdev_children == 0);
2033 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE);
2034 return (SET_ERROR(ENXIO));
2035 }
2036
2037 error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize,
2038 &logical_ashift, &physical_ashift);
2039
2040 /* Keep the device in removed state if unplugged */
2041 if (error == ENOENT && vd->vdev_removed) {
2042 vdev_set_state(vd, B_TRUE, VDEV_STATE_REMOVED,
2043 VDEV_AUX_NONE);
2044 return (error);
2045 }
2046
2047 /*
2048 * Physical volume size should never be larger than its max size, unless
2049 * the disk has shrunk while we were reading it or the device is buggy
2050 * or damaged: either way it's not safe for use, bail out of the open.
2051 */
2052 if (osize > max_osize) {
2053 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2054 VDEV_AUX_OPEN_FAILED);
2055 return (SET_ERROR(ENXIO));
2056 }
2057
2058 /*
2059 * Reset the vdev_reopening flag so that we actually close
2060 * the vdev on error.
2061 */
2062 vd->vdev_reopening = B_FALSE;
2063 if (zio_injection_enabled && error == 0)
2064 error = zio_handle_device_injection(vd, NULL, SET_ERROR(ENXIO));
2065
2066 if (error) {
2067 if (vd->vdev_removed &&
2068 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED)
2069 vd->vdev_removed = B_FALSE;
2070
2071 if (vd->vdev_stat.vs_aux == VDEV_AUX_CHILDREN_OFFLINE) {
2072 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE,
2073 vd->vdev_stat.vs_aux);
2074 } else {
2075 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2076 vd->vdev_stat.vs_aux);
2077 }
2078 return (error);
2079 }
2080
2081 vd->vdev_removed = B_FALSE;
2082
2083 /*
2084 * Recheck the faulted flag now that we have confirmed that
2085 * the vdev is accessible. If we're faulted, bail.
2086 */
2087 if (vd->vdev_faulted) {
2088 ASSERT(vd->vdev_children == 0);
2089 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
2090 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
2091 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
2092 vd->vdev_label_aux);
2093 return (SET_ERROR(ENXIO));
2094 }
2095
2096 if (vd->vdev_degraded) {
2097 ASSERT(vd->vdev_children == 0);
2098 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
2099 VDEV_AUX_ERR_EXCEEDED);
2100 } else {
2101 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0);
2102 }
2103
2104 /*
2105 * For hole or missing vdevs we just return success.
2106 */
2107 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops)
2108 return (0);
2109
2110 for (int c = 0; c < vd->vdev_children; c++) {
2111 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) {
2112 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
2113 VDEV_AUX_NONE);
2114 break;
2115 }
2116 }
2117
2118 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t));
2119 max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t));
2120
2121 if (vd->vdev_children == 0) {
2122 if (osize < SPA_MINDEVSIZE) {
2123 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2124 VDEV_AUX_TOO_SMALL);
2125 return (SET_ERROR(EOVERFLOW));
2126 }
2127 psize = osize;
2128 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE);
2129 max_asize = max_osize - (VDEV_LABEL_START_SIZE +
2130 VDEV_LABEL_END_SIZE);
2131 } else {
2132 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE -
2133 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) {
2134 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2135 VDEV_AUX_TOO_SMALL);
2136 return (SET_ERROR(EOVERFLOW));
2137 }
2138 psize = 0;
2139 asize = osize;
2140 max_asize = max_osize;
2141 }
2142
2143 /*
2144 * If the vdev was expanded, record this so that we can re-create the
2145 * uberblock rings in labels {2,3}, during the next sync.
2146 */
2147 if ((psize > vd->vdev_psize) && (vd->vdev_psize != 0))
2148 vd->vdev_copy_uberblocks = B_TRUE;
2149
2150 vd->vdev_psize = psize;
2151
2152 /*
2153 * Make sure the allocatable size hasn't shrunk too much.
2154 */
2155 if (asize < vd->vdev_min_asize) {
2156 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2157 VDEV_AUX_BAD_LABEL);
2158 return (SET_ERROR(EINVAL));
2159 }
2160
2161 /*
2162 * We can always set the logical/physical ashift members since
2163 * their values are only used to calculate the vdev_ashift when
2164 * the device is first added to the config. These values should
2165 * not be used for anything else since they may change whenever
2166 * the device is reopened and we don't store them in the label.
2167 */
2168 vd->vdev_physical_ashift =
2169 MAX(physical_ashift, vd->vdev_physical_ashift);
2170 vd->vdev_logical_ashift = MAX(logical_ashift,
2171 vd->vdev_logical_ashift);
2172
2173 if (vd->vdev_asize == 0) {
2174 /*
2175 * This is the first-ever open, so use the computed values.
2176 * For compatibility, a different ashift can be requested.
2177 */
2178 vd->vdev_asize = asize;
2179 vd->vdev_max_asize = max_asize;
2180
2181 /*
2182 * If the vdev_ashift was not overridden at creation time,
2183 * then set it the logical ashift and optimize the ashift.
2184 */
2185 if (vd->vdev_ashift == 0) {
2186 vd->vdev_ashift = vd->vdev_logical_ashift;
2187
2188 if (vd->vdev_logical_ashift > ASHIFT_MAX) {
2189 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2190 VDEV_AUX_ASHIFT_TOO_BIG);
2191 return (SET_ERROR(EDOM));
2192 }
2193
2194 if (vd->vdev_top == vd && vd->vdev_attaching == B_FALSE)
2195 vdev_ashift_optimize(vd);
2196 vd->vdev_attaching = B_FALSE;
2197 }
2198 if (vd->vdev_ashift != 0 && (vd->vdev_ashift < ASHIFT_MIN ||
2199 vd->vdev_ashift > ASHIFT_MAX)) {
2200 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2201 VDEV_AUX_BAD_ASHIFT);
2202 return (SET_ERROR(EDOM));
2203 }
2204 } else {
2205 /*
2206 * Make sure the alignment required hasn't increased.
2207 */
2208 if (vd->vdev_ashift > vd->vdev_top->vdev_ashift &&
2209 vd->vdev_ops->vdev_op_leaf) {
2210 (void) zfs_ereport_post(
2211 FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT,
2212 spa, vd, NULL, NULL, 0);
2213 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2214 VDEV_AUX_BAD_LABEL);
2215 return (SET_ERROR(EDOM));
2216 }
2217 vd->vdev_max_asize = max_asize;
2218 }
2219
2220 /*
2221 * If all children are healthy we update asize if either:
2222 * The asize has increased, due to a device expansion caused by dynamic
2223 * LUN growth or vdev replacement, and automatic expansion is enabled;
2224 * making the additional space available.
2225 *
2226 * The asize has decreased, due to a device shrink usually caused by a
2227 * vdev replace with a smaller device. This ensures that calculations
2228 * based of max_asize and asize e.g. esize are always valid. It's safe
2229 * to do this as we've already validated that asize is greater than
2230 * vdev_min_asize.
2231 */
2232 if (vd->vdev_state == VDEV_STATE_HEALTHY &&
2233 ((asize > vd->vdev_asize &&
2234 (vd->vdev_expanding || spa->spa_autoexpand)) ||
2235 (asize < vd->vdev_asize)))
2236 vd->vdev_asize = asize;
2237
2238 vdev_set_min_asize(vd);
2239
2240 /*
2241 * Ensure we can issue some IO before declaring the
2242 * vdev open for business.
2243 */
2244 if (vd->vdev_ops->vdev_op_leaf &&
2245 (error = zio_wait(vdev_probe(vd, NULL))) != 0) {
2246 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
2247 VDEV_AUX_ERR_EXCEEDED);
2248 return (error);
2249 }
2250
2251 /*
2252 * Track the minimum allocation size.
2253 */
2254 if (vd->vdev_top == vd && vd->vdev_ashift != 0 &&
2255 vd->vdev_islog == 0 && vd->vdev_aux == NULL) {
2256 uint64_t min_alloc = vdev_get_min_alloc(vd);
2257 vdev_spa_set_alloc(spa, min_alloc);
2258 }
2259
2260 /*
2261 * If this is a leaf vdev, assess whether a resilver is needed.
2262 * But don't do this if we are doing a reopen for a scrub, since
2263 * this would just restart the scrub we are already doing.
2264 */
2265 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen)
2266 dsl_scan_assess_vdev(spa->spa_dsl_pool, vd);
2267
2268 return (0);
2269 }
2270
2271 static void
vdev_validate_child(void * arg)2272 vdev_validate_child(void *arg)
2273 {
2274 vdev_t *vd = arg;
2275
2276 vd->vdev_validate_thread = curthread;
2277 vd->vdev_validate_error = vdev_validate(vd);
2278 vd->vdev_validate_thread = NULL;
2279 }
2280
2281 /*
2282 * Called once the vdevs are all opened, this routine validates the label
2283 * contents. This needs to be done before vdev_load() so that we don't
2284 * inadvertently do repair I/Os to the wrong device.
2285 *
2286 * This function will only return failure if one of the vdevs indicates that it
2287 * has since been destroyed or exported. This is only possible if
2288 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
2289 * will be updated but the function will return 0.
2290 */
2291 int
vdev_validate(vdev_t * vd)2292 vdev_validate(vdev_t *vd)
2293 {
2294 spa_t *spa = vd->vdev_spa;
2295 taskq_t *tq = NULL;
2296 nvlist_t *label;
2297 uint64_t guid = 0, aux_guid = 0, top_guid;
2298 uint64_t state;
2299 nvlist_t *nvl;
2300 uint64_t txg;
2301 int children = vd->vdev_children;
2302
2303 if (vdev_validate_skip)
2304 return (0);
2305
2306 if (children > 0) {
2307 tq = taskq_create("vdev_validate", children, minclsyspri,
2308 children, children, TASKQ_PREPOPULATE);
2309 }
2310
2311 for (uint64_t c = 0; c < children; c++) {
2312 vdev_t *cvd = vd->vdev_child[c];
2313
2314 if (tq == NULL || vdev_uses_zvols(cvd)) {
2315 vdev_validate_child(cvd);
2316 } else {
2317 VERIFY(taskq_dispatch(tq, vdev_validate_child, cvd,
2318 TQ_SLEEP) != TASKQID_INVALID);
2319 }
2320 }
2321 if (tq != NULL) {
2322 taskq_wait(tq);
2323 taskq_destroy(tq);
2324 }
2325 for (int c = 0; c < children; c++) {
2326 int error = vd->vdev_child[c]->vdev_validate_error;
2327
2328 if (error != 0)
2329 return (SET_ERROR(EBADF));
2330 }
2331
2332
2333 /*
2334 * If the device has already failed, or was marked offline, don't do
2335 * any further validation. Otherwise, label I/O will fail and we will
2336 * overwrite the previous state.
2337 */
2338 if (!vd->vdev_ops->vdev_op_leaf || !vdev_readable(vd))
2339 return (0);
2340
2341 /*
2342 * If we are performing an extreme rewind, we allow for a label that
2343 * was modified at a point after the current txg.
2344 * If config lock is not held do not check for the txg. spa_sync could
2345 * be updating the vdev's label before updating spa_last_synced_txg.
2346 */
2347 if (spa->spa_extreme_rewind || spa_last_synced_txg(spa) == 0 ||
2348 spa_config_held(spa, SCL_CONFIG, RW_WRITER) != SCL_CONFIG)
2349 txg = UINT64_MAX;
2350 else
2351 txg = spa_last_synced_txg(spa);
2352
2353 if ((label = vdev_label_read_config(vd, txg)) == NULL) {
2354 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2355 VDEV_AUX_BAD_LABEL);
2356 vdev_dbgmsg(vd, "vdev_validate: failed reading config for "
2357 "txg %llu", (u_longlong_t)txg);
2358 return (0);
2359 }
2360
2361 /*
2362 * Determine if this vdev has been split off into another
2363 * pool. If so, then refuse to open it.
2364 */
2365 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID,
2366 &aux_guid) == 0 && aux_guid == spa_guid(spa)) {
2367 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2368 VDEV_AUX_SPLIT_POOL);
2369 nvlist_free(label);
2370 vdev_dbgmsg(vd, "vdev_validate: vdev split into other pool");
2371 return (0);
2372 }
2373
2374 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, &guid) != 0) {
2375 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2376 VDEV_AUX_CORRUPT_DATA);
2377 nvlist_free(label);
2378 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
2379 ZPOOL_CONFIG_POOL_GUID);
2380 return (0);
2381 }
2382
2383 /*
2384 * If config is not trusted then ignore the spa guid check. This is
2385 * necessary because if the machine crashed during a re-guid the new
2386 * guid might have been written to all of the vdev labels, but not the
2387 * cached config. The check will be performed again once we have the
2388 * trusted config from the MOS.
2389 */
2390 if (spa->spa_trust_config && guid != spa_guid(spa)) {
2391 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2392 VDEV_AUX_CORRUPT_DATA);
2393 nvlist_free(label);
2394 vdev_dbgmsg(vd, "vdev_validate: vdev label pool_guid doesn't "
2395 "match config (%llu != %llu)", (u_longlong_t)guid,
2396 (u_longlong_t)spa_guid(spa));
2397 return (0);
2398 }
2399
2400 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl)
2401 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID,
2402 &aux_guid) != 0)
2403 aux_guid = 0;
2404
2405 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0) {
2406 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2407 VDEV_AUX_CORRUPT_DATA);
2408 nvlist_free(label);
2409 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
2410 ZPOOL_CONFIG_GUID);
2411 return (0);
2412 }
2413
2414 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, &top_guid)
2415 != 0) {
2416 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2417 VDEV_AUX_CORRUPT_DATA);
2418 nvlist_free(label);
2419 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
2420 ZPOOL_CONFIG_TOP_GUID);
2421 return (0);
2422 }
2423
2424 /*
2425 * If this vdev just became a top-level vdev because its sibling was
2426 * detached, it will have adopted the parent's vdev guid -- but the
2427 * label may or may not be on disk yet. Fortunately, either version
2428 * of the label will have the same top guid, so if we're a top-level
2429 * vdev, we can safely compare to that instead.
2430 * However, if the config comes from a cachefile that failed to update
2431 * after the detach, a top-level vdev will appear as a non top-level
2432 * vdev in the config. Also relax the constraints if we perform an
2433 * extreme rewind.
2434 *
2435 * If we split this vdev off instead, then we also check the
2436 * original pool's guid. We don't want to consider the vdev
2437 * corrupt if it is partway through a split operation.
2438 */
2439 if (vd->vdev_guid != guid && vd->vdev_guid != aux_guid) {
2440 boolean_t mismatch = B_FALSE;
2441 if (spa->spa_trust_config && !spa->spa_extreme_rewind) {
2442 if (vd != vd->vdev_top || vd->vdev_guid != top_guid)
2443 mismatch = B_TRUE;
2444 } else {
2445 if (vd->vdev_guid != top_guid &&
2446 vd->vdev_top->vdev_guid != guid)
2447 mismatch = B_TRUE;
2448 }
2449
2450 if (mismatch) {
2451 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2452 VDEV_AUX_CORRUPT_DATA);
2453 nvlist_free(label);
2454 vdev_dbgmsg(vd, "vdev_validate: config guid "
2455 "doesn't match label guid");
2456 vdev_dbgmsg(vd, "CONFIG: guid %llu, top_guid %llu",
2457 (u_longlong_t)vd->vdev_guid,
2458 (u_longlong_t)vd->vdev_top->vdev_guid);
2459 vdev_dbgmsg(vd, "LABEL: guid %llu, top_guid %llu, "
2460 "aux_guid %llu", (u_longlong_t)guid,
2461 (u_longlong_t)top_guid, (u_longlong_t)aux_guid);
2462 return (0);
2463 }
2464 }
2465
2466 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
2467 &state) != 0) {
2468 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2469 VDEV_AUX_CORRUPT_DATA);
2470 nvlist_free(label);
2471 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
2472 ZPOOL_CONFIG_POOL_STATE);
2473 return (0);
2474 }
2475
2476 nvlist_free(label);
2477
2478 /*
2479 * If this is a verbatim import, no need to check the
2480 * state of the pool.
2481 */
2482 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) &&
2483 spa_load_state(spa) == SPA_LOAD_OPEN &&
2484 state != POOL_STATE_ACTIVE) {
2485 vdev_dbgmsg(vd, "vdev_validate: invalid pool state (%llu) "
2486 "for spa %s", (u_longlong_t)state, spa->spa_name);
2487 return (SET_ERROR(EBADF));
2488 }
2489
2490 /*
2491 * If we were able to open and validate a vdev that was
2492 * previously marked permanently unavailable, clear that state
2493 * now.
2494 */
2495 if (vd->vdev_not_present)
2496 vd->vdev_not_present = 0;
2497
2498 return (0);
2499 }
2500
2501 static void
vdev_update_path(const char * prefix,char * svd,char ** dvd,uint64_t guid)2502 vdev_update_path(const char *prefix, char *svd, char **dvd, uint64_t guid)
2503 {
2504 if (svd != NULL && *dvd != NULL) {
2505 if (strcmp(svd, *dvd) != 0) {
2506 zfs_dbgmsg("vdev_copy_path: vdev %llu: %s changed "
2507 "from '%s' to '%s'", (u_longlong_t)guid, prefix,
2508 *dvd, svd);
2509 spa_strfree(*dvd);
2510 *dvd = spa_strdup(svd);
2511 }
2512 } else if (svd != NULL) {
2513 *dvd = spa_strdup(svd);
2514 zfs_dbgmsg("vdev_copy_path: vdev %llu: path set to '%s'",
2515 (u_longlong_t)guid, *dvd);
2516 }
2517 }
2518
2519 static void
vdev_copy_path_impl(vdev_t * svd,vdev_t * dvd)2520 vdev_copy_path_impl(vdev_t *svd, vdev_t *dvd)
2521 {
2522 char *old, *new;
2523
2524 vdev_update_path("vdev_path", svd->vdev_path, &dvd->vdev_path,
2525 dvd->vdev_guid);
2526
2527 vdev_update_path("vdev_devid", svd->vdev_devid, &dvd->vdev_devid,
2528 dvd->vdev_guid);
2529
2530 vdev_update_path("vdev_physpath", svd->vdev_physpath,
2531 &dvd->vdev_physpath, dvd->vdev_guid);
2532
2533 /*
2534 * Our enclosure sysfs path may have changed between imports
2535 */
2536 old = dvd->vdev_enc_sysfs_path;
2537 new = svd->vdev_enc_sysfs_path;
2538 if ((old != NULL && new == NULL) ||
2539 (old == NULL && new != NULL) ||
2540 ((old != NULL && new != NULL) && strcmp(new, old) != 0)) {
2541 zfs_dbgmsg("vdev_copy_path: vdev %llu: vdev_enc_sysfs_path "
2542 "changed from '%s' to '%s'", (u_longlong_t)dvd->vdev_guid,
2543 old, new);
2544
2545 if (dvd->vdev_enc_sysfs_path)
2546 spa_strfree(dvd->vdev_enc_sysfs_path);
2547
2548 if (svd->vdev_enc_sysfs_path) {
2549 dvd->vdev_enc_sysfs_path = spa_strdup(
2550 svd->vdev_enc_sysfs_path);
2551 } else {
2552 dvd->vdev_enc_sysfs_path = NULL;
2553 }
2554 }
2555 }
2556
2557 /*
2558 * Recursively copy vdev paths from one vdev to another. Source and destination
2559 * vdev trees must have same geometry otherwise return error. Intended to copy
2560 * paths from userland config into MOS config.
2561 */
2562 int
vdev_copy_path_strict(vdev_t * svd,vdev_t * dvd)2563 vdev_copy_path_strict(vdev_t *svd, vdev_t *dvd)
2564 {
2565 if ((svd->vdev_ops == &vdev_missing_ops) ||
2566 (svd->vdev_ishole && dvd->vdev_ishole) ||
2567 (dvd->vdev_ops == &vdev_indirect_ops))
2568 return (0);
2569
2570 if (svd->vdev_ops != dvd->vdev_ops) {
2571 vdev_dbgmsg(svd, "vdev_copy_path: vdev type mismatch: %s != %s",
2572 svd->vdev_ops->vdev_op_type, dvd->vdev_ops->vdev_op_type);
2573 return (SET_ERROR(EINVAL));
2574 }
2575
2576 if (svd->vdev_guid != dvd->vdev_guid) {
2577 vdev_dbgmsg(svd, "vdev_copy_path: guids mismatch (%llu != "
2578 "%llu)", (u_longlong_t)svd->vdev_guid,
2579 (u_longlong_t)dvd->vdev_guid);
2580 return (SET_ERROR(EINVAL));
2581 }
2582
2583 if (svd->vdev_children != dvd->vdev_children) {
2584 vdev_dbgmsg(svd, "vdev_copy_path: children count mismatch: "
2585 "%llu != %llu", (u_longlong_t)svd->vdev_children,
2586 (u_longlong_t)dvd->vdev_children);
2587 return (SET_ERROR(EINVAL));
2588 }
2589
2590 for (uint64_t i = 0; i < svd->vdev_children; i++) {
2591 int error = vdev_copy_path_strict(svd->vdev_child[i],
2592 dvd->vdev_child[i]);
2593 if (error != 0)
2594 return (error);
2595 }
2596
2597 if (svd->vdev_ops->vdev_op_leaf)
2598 vdev_copy_path_impl(svd, dvd);
2599
2600 return (0);
2601 }
2602
2603 static void
vdev_copy_path_search(vdev_t * stvd,vdev_t * dvd)2604 vdev_copy_path_search(vdev_t *stvd, vdev_t *dvd)
2605 {
2606 ASSERT(stvd->vdev_top == stvd);
2607 ASSERT3U(stvd->vdev_id, ==, dvd->vdev_top->vdev_id);
2608
2609 for (uint64_t i = 0; i < dvd->vdev_children; i++) {
2610 vdev_copy_path_search(stvd, dvd->vdev_child[i]);
2611 }
2612
2613 if (!dvd->vdev_ops->vdev_op_leaf || !vdev_is_concrete(dvd))
2614 return;
2615
2616 /*
2617 * The idea here is that while a vdev can shift positions within
2618 * a top vdev (when replacing, attaching mirror, etc.) it cannot
2619 * step outside of it.
2620 */
2621 vdev_t *vd = vdev_lookup_by_guid(stvd, dvd->vdev_guid);
2622
2623 if (vd == NULL || vd->vdev_ops != dvd->vdev_ops)
2624 return;
2625
2626 ASSERT(vd->vdev_ops->vdev_op_leaf);
2627
2628 vdev_copy_path_impl(vd, dvd);
2629 }
2630
2631 /*
2632 * Recursively copy vdev paths from one root vdev to another. Source and
2633 * destination vdev trees may differ in geometry. For each destination leaf
2634 * vdev, search a vdev with the same guid and top vdev id in the source.
2635 * Intended to copy paths from userland config into MOS config.
2636 */
2637 void
vdev_copy_path_relaxed(vdev_t * srvd,vdev_t * drvd)2638 vdev_copy_path_relaxed(vdev_t *srvd, vdev_t *drvd)
2639 {
2640 uint64_t children = MIN(srvd->vdev_children, drvd->vdev_children);
2641 ASSERT(srvd->vdev_ops == &vdev_root_ops);
2642 ASSERT(drvd->vdev_ops == &vdev_root_ops);
2643
2644 for (uint64_t i = 0; i < children; i++) {
2645 vdev_copy_path_search(srvd->vdev_child[i],
2646 drvd->vdev_child[i]);
2647 }
2648 }
2649
2650 /*
2651 * Close a virtual device.
2652 */
2653 void
vdev_close(vdev_t * vd)2654 vdev_close(vdev_t *vd)
2655 {
2656 vdev_t *pvd = vd->vdev_parent;
2657 spa_t *spa __maybe_unused = vd->vdev_spa;
2658
2659 ASSERT(vd != NULL);
2660 ASSERT(vd->vdev_open_thread == curthread ||
2661 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2662
2663 /*
2664 * If our parent is reopening, then we are as well, unless we are
2665 * going offline.
2666 */
2667 if (pvd != NULL && pvd->vdev_reopening)
2668 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline);
2669
2670 vd->vdev_ops->vdev_op_close(vd);
2671
2672 /*
2673 * We record the previous state before we close it, so that if we are
2674 * doing a reopen(), we don't generate FMA ereports if we notice that
2675 * it's still faulted.
2676 */
2677 vd->vdev_prevstate = vd->vdev_state;
2678
2679 if (vd->vdev_offline)
2680 vd->vdev_state = VDEV_STATE_OFFLINE;
2681 else
2682 vd->vdev_state = VDEV_STATE_CLOSED;
2683 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
2684 }
2685
2686 void
vdev_hold(vdev_t * vd)2687 vdev_hold(vdev_t *vd)
2688 {
2689 spa_t *spa = vd->vdev_spa;
2690
2691 ASSERT(spa_is_root(spa));
2692 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
2693 return;
2694
2695 for (int c = 0; c < vd->vdev_children; c++)
2696 vdev_hold(vd->vdev_child[c]);
2697
2698 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_ops->vdev_op_hold != NULL)
2699 vd->vdev_ops->vdev_op_hold(vd);
2700 }
2701
2702 void
vdev_rele(vdev_t * vd)2703 vdev_rele(vdev_t *vd)
2704 {
2705 ASSERT(spa_is_root(vd->vdev_spa));
2706 for (int c = 0; c < vd->vdev_children; c++)
2707 vdev_rele(vd->vdev_child[c]);
2708
2709 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_ops->vdev_op_rele != NULL)
2710 vd->vdev_ops->vdev_op_rele(vd);
2711 }
2712
2713 /*
2714 * Reopen all interior vdevs and any unopened leaves. We don't actually
2715 * reopen leaf vdevs which had previously been opened as they might deadlock
2716 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
2717 * If the leaf has never been opened then open it, as usual.
2718 */
2719 void
vdev_reopen(vdev_t * vd)2720 vdev_reopen(vdev_t *vd)
2721 {
2722 spa_t *spa = vd->vdev_spa;
2723
2724 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2725
2726 /* set the reopening flag unless we're taking the vdev offline */
2727 vd->vdev_reopening = !vd->vdev_offline;
2728 vdev_close(vd);
2729 (void) vdev_open(vd);
2730
2731 /*
2732 * Call vdev_validate() here to make sure we have the same device.
2733 * Otherwise, a device with an invalid label could be successfully
2734 * opened in response to vdev_reopen().
2735 */
2736 if (vd->vdev_aux) {
2737 (void) vdev_validate_aux(vd);
2738 if (vdev_readable(vd) && vdev_writeable(vd) &&
2739 vd->vdev_aux == &spa->spa_l2cache) {
2740 /*
2741 * In case the vdev is present we should evict all ARC
2742 * buffers and pointers to log blocks and reclaim their
2743 * space before restoring its contents to L2ARC.
2744 */
2745 if (l2arc_vdev_present(vd)) {
2746 l2arc_rebuild_vdev(vd, B_TRUE);
2747 } else {
2748 l2arc_add_vdev(spa, vd);
2749 }
2750 spa_async_request(spa, SPA_ASYNC_L2CACHE_REBUILD);
2751 spa_async_request(spa, SPA_ASYNC_L2CACHE_TRIM);
2752 }
2753 } else {
2754 (void) vdev_validate(vd);
2755 }
2756
2757 /*
2758 * Recheck if resilver is still needed and cancel any
2759 * scheduled resilver if resilver is unneeded.
2760 */
2761 if (!vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL) &&
2762 spa->spa_async_tasks & SPA_ASYNC_RESILVER) {
2763 mutex_enter(&spa->spa_async_lock);
2764 spa->spa_async_tasks &= ~SPA_ASYNC_RESILVER;
2765 mutex_exit(&spa->spa_async_lock);
2766 }
2767
2768 /*
2769 * Reassess parent vdev's health.
2770 */
2771 vdev_propagate_state(vd);
2772 }
2773
2774 int
vdev_create(vdev_t * vd,uint64_t txg,boolean_t isreplacing)2775 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
2776 {
2777 int error;
2778
2779 /*
2780 * Normally, partial opens (e.g. of a mirror) are allowed.
2781 * For a create, however, we want to fail the request if
2782 * there are any components we can't open.
2783 */
2784 error = vdev_open(vd);
2785
2786 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) {
2787 vdev_close(vd);
2788 return (error ? error : SET_ERROR(ENXIO));
2789 }
2790
2791 /*
2792 * Recursively load DTLs and initialize all labels.
2793 */
2794 if ((error = vdev_dtl_load(vd)) != 0 ||
2795 (error = vdev_label_init(vd, txg, isreplacing ?
2796 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
2797 vdev_close(vd);
2798 return (error);
2799 }
2800
2801 return (0);
2802 }
2803
2804 void
vdev_metaslab_set_size(vdev_t * vd)2805 vdev_metaslab_set_size(vdev_t *vd)
2806 {
2807 uint64_t asize = vd->vdev_asize;
2808 uint64_t ms_count = asize >> zfs_vdev_default_ms_shift;
2809 uint64_t ms_shift;
2810
2811 /*
2812 * There are two dimensions to the metaslab sizing calculation:
2813 * the size of the metaslab and the count of metaslabs per vdev.
2814 *
2815 * The default values used below are a good balance between memory
2816 * usage (larger metaslab size means more memory needed for loaded
2817 * metaslabs; more metaslabs means more memory needed for the
2818 * metaslab_t structs), metaslab load time (larger metaslabs take
2819 * longer to load), and metaslab sync time (more metaslabs means
2820 * more time spent syncing all of them).
2821 *
2822 * In general, we aim for zfs_vdev_default_ms_count (200) metaslabs.
2823 * The range of the dimensions are as follows:
2824 *
2825 * 2^29 <= ms_size <= 2^34
2826 * 16 <= ms_count <= 131,072
2827 *
2828 * On the lower end of vdev sizes, we aim for metaslabs sizes of
2829 * at least 512MB (2^29) to minimize fragmentation effects when
2830 * testing with smaller devices. However, the count constraint
2831 * of at least 16 metaslabs will override this minimum size goal.
2832 *
2833 * On the upper end of vdev sizes, we aim for a maximum metaslab
2834 * size of 16GB. However, we will cap the total count to 2^17
2835 * metaslabs to keep our memory footprint in check and let the
2836 * metaslab size grow from there if that limit is hit.
2837 *
2838 * The net effect of applying above constrains is summarized below.
2839 *
2840 * vdev size metaslab count
2841 * --------------|-----------------
2842 * < 8GB ~16
2843 * 8GB - 100GB one per 512MB
2844 * 100GB - 3TB ~200
2845 * 3TB - 2PB one per 16GB
2846 * > 2PB ~131,072
2847 * --------------------------------
2848 *
2849 * Finally, note that all of the above calculate the initial
2850 * number of metaslabs. Expanding a top-level vdev will result
2851 * in additional metaslabs being allocated making it possible
2852 * to exceed the zfs_vdev_ms_count_limit.
2853 */
2854
2855 if (ms_count < zfs_vdev_min_ms_count)
2856 ms_shift = highbit64(asize / zfs_vdev_min_ms_count);
2857 else if (ms_count > zfs_vdev_default_ms_count)
2858 ms_shift = highbit64(asize / zfs_vdev_default_ms_count);
2859 else
2860 ms_shift = zfs_vdev_default_ms_shift;
2861
2862 if (ms_shift < SPA_MAXBLOCKSHIFT) {
2863 ms_shift = SPA_MAXBLOCKSHIFT;
2864 } else if (ms_shift > zfs_vdev_max_ms_shift) {
2865 ms_shift = zfs_vdev_max_ms_shift;
2866 /* cap the total count to constrain memory footprint */
2867 if ((asize >> ms_shift) > zfs_vdev_ms_count_limit)
2868 ms_shift = highbit64(asize / zfs_vdev_ms_count_limit);
2869 }
2870
2871 vd->vdev_ms_shift = ms_shift;
2872 ASSERT3U(vd->vdev_ms_shift, >=, SPA_MAXBLOCKSHIFT);
2873 }
2874
2875 void
vdev_dirty(vdev_t * vd,int flags,void * arg,uint64_t txg)2876 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
2877 {
2878 ASSERT(vd == vd->vdev_top);
2879 /* indirect vdevs don't have metaslabs or dtls */
2880 ASSERT(vdev_is_concrete(vd) || flags == 0);
2881 ASSERT(ISP2(flags));
2882 ASSERT(spa_writeable(vd->vdev_spa));
2883
2884 if (flags & VDD_METASLAB)
2885 (void) txg_list_add(&vd->vdev_ms_list, arg, txg);
2886
2887 if (flags & VDD_DTL)
2888 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg);
2889
2890 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
2891 }
2892
2893 void
vdev_dirty_leaves(vdev_t * vd,int flags,uint64_t txg)2894 vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg)
2895 {
2896 for (int c = 0; c < vd->vdev_children; c++)
2897 vdev_dirty_leaves(vd->vdev_child[c], flags, txg);
2898
2899 if (vd->vdev_ops->vdev_op_leaf)
2900 vdev_dirty(vd->vdev_top, flags, vd, txg);
2901 }
2902
2903 /*
2904 * DTLs.
2905 *
2906 * A vdev's DTL (dirty time log) is the set of transaction groups for which
2907 * the vdev has less than perfect replication. There are four kinds of DTL:
2908 *
2909 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
2910 *
2911 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
2912 *
2913 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
2914 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
2915 * txgs that was scrubbed.
2916 *
2917 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
2918 * persistent errors or just some device being offline.
2919 * Unlike the other three, the DTL_OUTAGE map is not generally
2920 * maintained; it's only computed when needed, typically to
2921 * determine whether a device can be detached.
2922 *
2923 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
2924 * either has the data or it doesn't.
2925 *
2926 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
2927 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
2928 * if any child is less than fully replicated, then so is its parent.
2929 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
2930 * comprising only those txgs which appear in 'maxfaults' or more children;
2931 * those are the txgs we don't have enough replication to read. For example,
2932 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
2933 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
2934 * two child DTL_MISSING maps.
2935 *
2936 * It should be clear from the above that to compute the DTLs and outage maps
2937 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
2938 * Therefore, that is all we keep on disk. When loading the pool, or after
2939 * a configuration change, we generate all other DTLs from first principles.
2940 */
2941 void
vdev_dtl_dirty(vdev_t * vd,vdev_dtl_type_t t,uint64_t txg,uint64_t size)2942 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
2943 {
2944 range_tree_t *rt = vd->vdev_dtl[t];
2945
2946 ASSERT(t < DTL_TYPES);
2947 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
2948 ASSERT(spa_writeable(vd->vdev_spa));
2949
2950 mutex_enter(&vd->vdev_dtl_lock);
2951 if (!range_tree_contains(rt, txg, size))
2952 range_tree_add(rt, txg, size);
2953 mutex_exit(&vd->vdev_dtl_lock);
2954 }
2955
2956 boolean_t
vdev_dtl_contains(vdev_t * vd,vdev_dtl_type_t t,uint64_t txg,uint64_t size)2957 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
2958 {
2959 range_tree_t *rt = vd->vdev_dtl[t];
2960 boolean_t dirty = B_FALSE;
2961
2962 ASSERT(t < DTL_TYPES);
2963 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
2964
2965 /*
2966 * While we are loading the pool, the DTLs have not been loaded yet.
2967 * This isn't a problem but it can result in devices being tried
2968 * which are known to not have the data. In which case, the import
2969 * is relying on the checksum to ensure that we get the right data.
2970 * Note that while importing we are only reading the MOS, which is
2971 * always checksummed.
2972 */
2973 mutex_enter(&vd->vdev_dtl_lock);
2974 if (!range_tree_is_empty(rt))
2975 dirty = range_tree_contains(rt, txg, size);
2976 mutex_exit(&vd->vdev_dtl_lock);
2977
2978 return (dirty);
2979 }
2980
2981 boolean_t
vdev_dtl_empty(vdev_t * vd,vdev_dtl_type_t t)2982 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t)
2983 {
2984 range_tree_t *rt = vd->vdev_dtl[t];
2985 boolean_t empty;
2986
2987 mutex_enter(&vd->vdev_dtl_lock);
2988 empty = range_tree_is_empty(rt);
2989 mutex_exit(&vd->vdev_dtl_lock);
2990
2991 return (empty);
2992 }
2993
2994 /*
2995 * Check if the txg falls within the range which must be
2996 * resilvered. DVAs outside this range can always be skipped.
2997 */
2998 boolean_t
vdev_default_need_resilver(vdev_t * vd,const dva_t * dva,size_t psize,uint64_t phys_birth)2999 vdev_default_need_resilver(vdev_t *vd, const dva_t *dva, size_t psize,
3000 uint64_t phys_birth)
3001 {
3002 (void) dva, (void) psize;
3003
3004 /* Set by sequential resilver. */
3005 if (phys_birth == TXG_UNKNOWN)
3006 return (B_TRUE);
3007
3008 return (vdev_dtl_contains(vd, DTL_PARTIAL, phys_birth, 1));
3009 }
3010
3011 /*
3012 * Returns B_TRUE if the vdev determines the DVA needs to be resilvered.
3013 */
3014 boolean_t
vdev_dtl_need_resilver(vdev_t * vd,const dva_t * dva,size_t psize,uint64_t phys_birth)3015 vdev_dtl_need_resilver(vdev_t *vd, const dva_t *dva, size_t psize,
3016 uint64_t phys_birth)
3017 {
3018 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
3019
3020 if (vd->vdev_ops->vdev_op_need_resilver == NULL ||
3021 vd->vdev_ops->vdev_op_leaf)
3022 return (B_TRUE);
3023
3024 return (vd->vdev_ops->vdev_op_need_resilver(vd, dva, psize,
3025 phys_birth));
3026 }
3027
3028 /*
3029 * Returns the lowest txg in the DTL range.
3030 */
3031 static uint64_t
vdev_dtl_min(vdev_t * vd)3032 vdev_dtl_min(vdev_t *vd)
3033 {
3034 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
3035 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
3036 ASSERT0(vd->vdev_children);
3037
3038 return (range_tree_min(vd->vdev_dtl[DTL_MISSING]) - 1);
3039 }
3040
3041 /*
3042 * Returns the highest txg in the DTL.
3043 */
3044 static uint64_t
vdev_dtl_max(vdev_t * vd)3045 vdev_dtl_max(vdev_t *vd)
3046 {
3047 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
3048 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
3049 ASSERT0(vd->vdev_children);
3050
3051 return (range_tree_max(vd->vdev_dtl[DTL_MISSING]));
3052 }
3053
3054 /*
3055 * Determine if a resilvering vdev should remove any DTL entries from
3056 * its range. If the vdev was resilvering for the entire duration of the
3057 * scan then it should excise that range from its DTLs. Otherwise, this
3058 * vdev is considered partially resilvered and should leave its DTL
3059 * entries intact. The comment in vdev_dtl_reassess() describes how we
3060 * excise the DTLs.
3061 */
3062 static boolean_t
vdev_dtl_should_excise(vdev_t * vd,boolean_t rebuild_done)3063 vdev_dtl_should_excise(vdev_t *vd, boolean_t rebuild_done)
3064 {
3065 ASSERT0(vd->vdev_children);
3066
3067 if (vd->vdev_state < VDEV_STATE_DEGRADED)
3068 return (B_FALSE);
3069
3070 if (vd->vdev_resilver_deferred)
3071 return (B_FALSE);
3072
3073 if (range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]))
3074 return (B_TRUE);
3075
3076 if (rebuild_done) {
3077 vdev_rebuild_t *vr = &vd->vdev_top->vdev_rebuild_config;
3078 vdev_rebuild_phys_t *vrp = &vr->vr_rebuild_phys;
3079
3080 /* Rebuild not initiated by attach */
3081 if (vd->vdev_rebuild_txg == 0)
3082 return (B_TRUE);
3083
3084 /*
3085 * When a rebuild completes without error then all missing data
3086 * up to the rebuild max txg has been reconstructed and the DTL
3087 * is eligible for excision.
3088 */
3089 if (vrp->vrp_rebuild_state == VDEV_REBUILD_COMPLETE &&
3090 vdev_dtl_max(vd) <= vrp->vrp_max_txg) {
3091 ASSERT3U(vrp->vrp_min_txg, <=, vdev_dtl_min(vd));
3092 ASSERT3U(vrp->vrp_min_txg, <, vd->vdev_rebuild_txg);
3093 ASSERT3U(vd->vdev_rebuild_txg, <=, vrp->vrp_max_txg);
3094 return (B_TRUE);
3095 }
3096 } else {
3097 dsl_scan_t *scn = vd->vdev_spa->spa_dsl_pool->dp_scan;
3098 dsl_scan_phys_t *scnp __maybe_unused = &scn->scn_phys;
3099
3100 /* Resilver not initiated by attach */
3101 if (vd->vdev_resilver_txg == 0)
3102 return (B_TRUE);
3103
3104 /*
3105 * When a resilver is initiated the scan will assign the
3106 * scn_max_txg value to the highest txg value that exists
3107 * in all DTLs. If this device's max DTL is not part of this
3108 * scan (i.e. it is not in the range (scn_min_txg, scn_max_txg]
3109 * then it is not eligible for excision.
3110 */
3111 if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) {
3112 ASSERT3U(scnp->scn_min_txg, <=, vdev_dtl_min(vd));
3113 ASSERT3U(scnp->scn_min_txg, <, vd->vdev_resilver_txg);
3114 ASSERT3U(vd->vdev_resilver_txg, <=, scnp->scn_max_txg);
3115 return (B_TRUE);
3116 }
3117 }
3118
3119 return (B_FALSE);
3120 }
3121
3122 /*
3123 * Reassess DTLs after a config change or scrub completion. If txg == 0 no
3124 * write operations will be issued to the pool.
3125 */
3126 void
vdev_dtl_reassess(vdev_t * vd,uint64_t txg,uint64_t scrub_txg,boolean_t scrub_done,boolean_t rebuild_done)3127 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg,
3128 boolean_t scrub_done, boolean_t rebuild_done)
3129 {
3130 spa_t *spa = vd->vdev_spa;
3131 avl_tree_t reftree;
3132 int minref;
3133
3134 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
3135
3136 for (int c = 0; c < vd->vdev_children; c++)
3137 vdev_dtl_reassess(vd->vdev_child[c], txg,
3138 scrub_txg, scrub_done, rebuild_done);
3139
3140 if (vd == spa->spa_root_vdev || !vdev_is_concrete(vd) || vd->vdev_aux)
3141 return;
3142
3143 if (vd->vdev_ops->vdev_op_leaf) {
3144 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
3145 vdev_rebuild_t *vr = &vd->vdev_top->vdev_rebuild_config;
3146 boolean_t check_excise = B_FALSE;
3147 boolean_t wasempty = B_TRUE;
3148
3149 mutex_enter(&vd->vdev_dtl_lock);
3150
3151 /*
3152 * If requested, pretend the scan or rebuild completed cleanly.
3153 */
3154 if (zfs_scan_ignore_errors) {
3155 if (scn != NULL)
3156 scn->scn_phys.scn_errors = 0;
3157 if (vr != NULL)
3158 vr->vr_rebuild_phys.vrp_errors = 0;
3159 }
3160
3161 if (scrub_txg != 0 &&
3162 !range_tree_is_empty(vd->vdev_dtl[DTL_MISSING])) {
3163 wasempty = B_FALSE;
3164 zfs_dbgmsg("guid:%llu txg:%llu scrub:%llu started:%d "
3165 "dtl:%llu/%llu errors:%llu",
3166 (u_longlong_t)vd->vdev_guid, (u_longlong_t)txg,
3167 (u_longlong_t)scrub_txg, spa->spa_scrub_started,
3168 (u_longlong_t)vdev_dtl_min(vd),
3169 (u_longlong_t)vdev_dtl_max(vd),
3170 (u_longlong_t)(scn ? scn->scn_phys.scn_errors : 0));
3171 }
3172
3173 /*
3174 * If we've completed a scrub/resilver or a rebuild cleanly
3175 * then determine if this vdev should remove any DTLs. We
3176 * only want to excise regions on vdevs that were available
3177 * during the entire duration of this scan.
3178 */
3179 if (rebuild_done &&
3180 vr != NULL && vr->vr_rebuild_phys.vrp_errors == 0) {
3181 check_excise = B_TRUE;
3182 } else {
3183 if (spa->spa_scrub_started ||
3184 (scn != NULL && scn->scn_phys.scn_errors == 0)) {
3185 check_excise = B_TRUE;
3186 }
3187 }
3188
3189 if (scrub_txg && check_excise &&
3190 vdev_dtl_should_excise(vd, rebuild_done)) {
3191 /*
3192 * We completed a scrub, resilver or rebuild up to
3193 * scrub_txg. If we did it without rebooting, then
3194 * the scrub dtl will be valid, so excise the old
3195 * region and fold in the scrub dtl. Otherwise,
3196 * leave the dtl as-is if there was an error.
3197 *
3198 * There's little trick here: to excise the beginning
3199 * of the DTL_MISSING map, we put it into a reference
3200 * tree and then add a segment with refcnt -1 that
3201 * covers the range [0, scrub_txg). This means
3202 * that each txg in that range has refcnt -1 or 0.
3203 * We then add DTL_SCRUB with a refcnt of 2, so that
3204 * entries in the range [0, scrub_txg) will have a
3205 * positive refcnt -- either 1 or 2. We then convert
3206 * the reference tree into the new DTL_MISSING map.
3207 */
3208 space_reftree_create(&reftree);
3209 space_reftree_add_map(&reftree,
3210 vd->vdev_dtl[DTL_MISSING], 1);
3211 space_reftree_add_seg(&reftree, 0, scrub_txg, -1);
3212 space_reftree_add_map(&reftree,
3213 vd->vdev_dtl[DTL_SCRUB], 2);
3214 space_reftree_generate_map(&reftree,
3215 vd->vdev_dtl[DTL_MISSING], 1);
3216 space_reftree_destroy(&reftree);
3217
3218 if (!range_tree_is_empty(vd->vdev_dtl[DTL_MISSING])) {
3219 zfs_dbgmsg("update DTL_MISSING:%llu/%llu",
3220 (u_longlong_t)vdev_dtl_min(vd),
3221 (u_longlong_t)vdev_dtl_max(vd));
3222 } else if (!wasempty) {
3223 zfs_dbgmsg("DTL_MISSING is now empty");
3224 }
3225 }
3226 range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
3227 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
3228 range_tree_add, vd->vdev_dtl[DTL_PARTIAL]);
3229 if (scrub_done)
3230 range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
3231 range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
3232 if (!vdev_readable(vd))
3233 range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
3234 else
3235 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
3236 range_tree_add, vd->vdev_dtl[DTL_OUTAGE]);
3237
3238 /*
3239 * If the vdev was resilvering or rebuilding and no longer
3240 * has any DTLs then reset the appropriate flag and dirty
3241 * the top level so that we persist the change.
3242 */
3243 if (txg != 0 &&
3244 range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]) &&
3245 range_tree_is_empty(vd->vdev_dtl[DTL_OUTAGE])) {
3246 if (vd->vdev_rebuild_txg != 0) {
3247 vd->vdev_rebuild_txg = 0;
3248 vdev_config_dirty(vd->vdev_top);
3249 } else if (vd->vdev_resilver_txg != 0) {
3250 vd->vdev_resilver_txg = 0;
3251 vdev_config_dirty(vd->vdev_top);
3252 }
3253 }
3254
3255 mutex_exit(&vd->vdev_dtl_lock);
3256
3257 if (txg != 0)
3258 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
3259 } else {
3260 mutex_enter(&vd->vdev_dtl_lock);
3261 for (int t = 0; t < DTL_TYPES; t++) {
3262 /* account for child's outage in parent's missing map */
3263 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t;
3264 if (t == DTL_SCRUB) {
3265 /* leaf vdevs only */
3266 continue;
3267 }
3268 if (t == DTL_PARTIAL) {
3269 /* i.e. non-zero */
3270 minref = 1;
3271 } else if (vdev_get_nparity(vd) != 0) {
3272 /* RAIDZ, DRAID */
3273 minref = vdev_get_nparity(vd) + 1;
3274 } else {
3275 /* any kind of mirror */
3276 minref = vd->vdev_children;
3277 }
3278 space_reftree_create(&reftree);
3279 for (int c = 0; c < vd->vdev_children; c++) {
3280 vdev_t *cvd = vd->vdev_child[c];
3281 mutex_enter(&cvd->vdev_dtl_lock);
3282 space_reftree_add_map(&reftree,
3283 cvd->vdev_dtl[s], 1);
3284 mutex_exit(&cvd->vdev_dtl_lock);
3285 }
3286 space_reftree_generate_map(&reftree,
3287 vd->vdev_dtl[t], minref);
3288 space_reftree_destroy(&reftree);
3289 }
3290 mutex_exit(&vd->vdev_dtl_lock);
3291 }
3292
3293 if (vd->vdev_top->vdev_ops == &vdev_raidz_ops) {
3294 raidz_dtl_reassessed(vd);
3295 }
3296 }
3297
3298 /*
3299 * Iterate over all the vdevs except spare, and post kobj events
3300 */
3301 void
vdev_post_kobj_evt(vdev_t * vd)3302 vdev_post_kobj_evt(vdev_t *vd)
3303 {
3304 if (vd->vdev_ops->vdev_op_kobj_evt_post &&
3305 vd->vdev_kobj_flag == B_FALSE) {
3306 vd->vdev_kobj_flag = B_TRUE;
3307 vd->vdev_ops->vdev_op_kobj_evt_post(vd);
3308 }
3309
3310 for (int c = 0; c < vd->vdev_children; c++)
3311 vdev_post_kobj_evt(vd->vdev_child[c]);
3312 }
3313
3314 /*
3315 * Iterate over all the vdevs except spare, and clear kobj events
3316 */
3317 void
vdev_clear_kobj_evt(vdev_t * vd)3318 vdev_clear_kobj_evt(vdev_t *vd)
3319 {
3320 vd->vdev_kobj_flag = B_FALSE;
3321
3322 for (int c = 0; c < vd->vdev_children; c++)
3323 vdev_clear_kobj_evt(vd->vdev_child[c]);
3324 }
3325
3326 int
vdev_dtl_load(vdev_t * vd)3327 vdev_dtl_load(vdev_t *vd)
3328 {
3329 spa_t *spa = vd->vdev_spa;
3330 objset_t *mos = spa->spa_meta_objset;
3331 range_tree_t *rt;
3332 int error = 0;
3333
3334 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) {
3335 ASSERT(vdev_is_concrete(vd));
3336
3337 /*
3338 * If the dtl cannot be sync'd there is no need to open it.
3339 */
3340 if (spa->spa_mode == SPA_MODE_READ && !spa->spa_read_spacemaps)
3341 return (0);
3342
3343 error = space_map_open(&vd->vdev_dtl_sm, mos,
3344 vd->vdev_dtl_object, 0, -1ULL, 0);
3345 if (error)
3346 return (error);
3347 ASSERT(vd->vdev_dtl_sm != NULL);
3348
3349 rt = range_tree_create(NULL, RANGE_SEG64, NULL, 0, 0);
3350 error = space_map_load(vd->vdev_dtl_sm, rt, SM_ALLOC);
3351 if (error == 0) {
3352 mutex_enter(&vd->vdev_dtl_lock);
3353 range_tree_walk(rt, range_tree_add,
3354 vd->vdev_dtl[DTL_MISSING]);
3355 mutex_exit(&vd->vdev_dtl_lock);
3356 }
3357
3358 range_tree_vacate(rt, NULL, NULL);
3359 range_tree_destroy(rt);
3360
3361 return (error);
3362 }
3363
3364 for (int c = 0; c < vd->vdev_children; c++) {
3365 error = vdev_dtl_load(vd->vdev_child[c]);
3366 if (error != 0)
3367 break;
3368 }
3369
3370 return (error);
3371 }
3372
3373 static void
vdev_zap_allocation_data(vdev_t * vd,dmu_tx_t * tx)3374 vdev_zap_allocation_data(vdev_t *vd, dmu_tx_t *tx)
3375 {
3376 spa_t *spa = vd->vdev_spa;
3377 objset_t *mos = spa->spa_meta_objset;
3378 vdev_alloc_bias_t alloc_bias = vd->vdev_alloc_bias;
3379 const char *string;
3380
3381 ASSERT(alloc_bias != VDEV_BIAS_NONE);
3382
3383 string =
3384 (alloc_bias == VDEV_BIAS_LOG) ? VDEV_ALLOC_BIAS_LOG :
3385 (alloc_bias == VDEV_BIAS_SPECIAL) ? VDEV_ALLOC_BIAS_SPECIAL :
3386 (alloc_bias == VDEV_BIAS_DEDUP) ? VDEV_ALLOC_BIAS_DEDUP : NULL;
3387
3388 ASSERT(string != NULL);
3389 VERIFY0(zap_add(mos, vd->vdev_top_zap, VDEV_TOP_ZAP_ALLOCATION_BIAS,
3390 1, strlen(string) + 1, string, tx));
3391
3392 if (alloc_bias == VDEV_BIAS_SPECIAL || alloc_bias == VDEV_BIAS_DEDUP) {
3393 spa_activate_allocation_classes(spa, tx);
3394 }
3395 }
3396
3397 void
vdev_destroy_unlink_zap(vdev_t * vd,uint64_t zapobj,dmu_tx_t * tx)3398 vdev_destroy_unlink_zap(vdev_t *vd, uint64_t zapobj, dmu_tx_t *tx)
3399 {
3400 spa_t *spa = vd->vdev_spa;
3401
3402 VERIFY0(zap_destroy(spa->spa_meta_objset, zapobj, tx));
3403 VERIFY0(zap_remove_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps,
3404 zapobj, tx));
3405 }
3406
3407 uint64_t
vdev_create_link_zap(vdev_t * vd,dmu_tx_t * tx)3408 vdev_create_link_zap(vdev_t *vd, dmu_tx_t *tx)
3409 {
3410 spa_t *spa = vd->vdev_spa;
3411 uint64_t zap = zap_create(spa->spa_meta_objset, DMU_OTN_ZAP_METADATA,
3412 DMU_OT_NONE, 0, tx);
3413
3414 ASSERT(zap != 0);
3415 VERIFY0(zap_add_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps,
3416 zap, tx));
3417
3418 return (zap);
3419 }
3420
3421 void
vdev_construct_zaps(vdev_t * vd,dmu_tx_t * tx)3422 vdev_construct_zaps(vdev_t *vd, dmu_tx_t *tx)
3423 {
3424 if (vd->vdev_ops != &vdev_hole_ops &&
3425 vd->vdev_ops != &vdev_missing_ops &&
3426 vd->vdev_ops != &vdev_root_ops &&
3427 !vd->vdev_top->vdev_removing) {
3428 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_leaf_zap == 0) {
3429 vd->vdev_leaf_zap = vdev_create_link_zap(vd, tx);
3430 }
3431 if (vd == vd->vdev_top && vd->vdev_top_zap == 0) {
3432 vd->vdev_top_zap = vdev_create_link_zap(vd, tx);
3433 if (vd->vdev_alloc_bias != VDEV_BIAS_NONE)
3434 vdev_zap_allocation_data(vd, tx);
3435 }
3436 }
3437 if (vd->vdev_ops == &vdev_root_ops && vd->vdev_root_zap == 0 &&
3438 spa_feature_is_enabled(vd->vdev_spa, SPA_FEATURE_AVZ_V2)) {
3439 if (!spa_feature_is_active(vd->vdev_spa, SPA_FEATURE_AVZ_V2))
3440 spa_feature_incr(vd->vdev_spa, SPA_FEATURE_AVZ_V2, tx);
3441 vd->vdev_root_zap = vdev_create_link_zap(vd, tx);
3442 }
3443
3444 for (uint64_t i = 0; i < vd->vdev_children; i++) {
3445 vdev_construct_zaps(vd->vdev_child[i], tx);
3446 }
3447 }
3448
3449 static void
vdev_dtl_sync(vdev_t * vd,uint64_t txg)3450 vdev_dtl_sync(vdev_t *vd, uint64_t txg)
3451 {
3452 spa_t *spa = vd->vdev_spa;
3453 range_tree_t *rt = vd->vdev_dtl[DTL_MISSING];
3454 objset_t *mos = spa->spa_meta_objset;
3455 range_tree_t *rtsync;
3456 dmu_tx_t *tx;
3457 uint64_t object = space_map_object(vd->vdev_dtl_sm);
3458
3459 ASSERT(vdev_is_concrete(vd));
3460 ASSERT(vd->vdev_ops->vdev_op_leaf);
3461
3462 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
3463
3464 if (vd->vdev_detached || vd->vdev_top->vdev_removing) {
3465 mutex_enter(&vd->vdev_dtl_lock);
3466 space_map_free(vd->vdev_dtl_sm, tx);
3467 space_map_close(vd->vdev_dtl_sm);
3468 vd->vdev_dtl_sm = NULL;
3469 mutex_exit(&vd->vdev_dtl_lock);
3470
3471 /*
3472 * We only destroy the leaf ZAP for detached leaves or for
3473 * removed log devices. Removed data devices handle leaf ZAP
3474 * cleanup later, once cancellation is no longer possible.
3475 */
3476 if (vd->vdev_leaf_zap != 0 && (vd->vdev_detached ||
3477 vd->vdev_top->vdev_islog)) {
3478 vdev_destroy_unlink_zap(vd, vd->vdev_leaf_zap, tx);
3479 vd->vdev_leaf_zap = 0;
3480 }
3481
3482 dmu_tx_commit(tx);
3483 return;
3484 }
3485
3486 if (vd->vdev_dtl_sm == NULL) {
3487 uint64_t new_object;
3488
3489 new_object = space_map_alloc(mos, zfs_vdev_dtl_sm_blksz, tx);
3490 VERIFY3U(new_object, !=, 0);
3491
3492 VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object,
3493 0, -1ULL, 0));
3494 ASSERT(vd->vdev_dtl_sm != NULL);
3495 }
3496
3497 rtsync = range_tree_create(NULL, RANGE_SEG64, NULL, 0, 0);
3498
3499 mutex_enter(&vd->vdev_dtl_lock);
3500 range_tree_walk(rt, range_tree_add, rtsync);
3501 mutex_exit(&vd->vdev_dtl_lock);
3502
3503 space_map_truncate(vd->vdev_dtl_sm, zfs_vdev_dtl_sm_blksz, tx);
3504 space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, SM_NO_VDEVID, tx);
3505 range_tree_vacate(rtsync, NULL, NULL);
3506
3507 range_tree_destroy(rtsync);
3508
3509 /*
3510 * If the object for the space map has changed then dirty
3511 * the top level so that we update the config.
3512 */
3513 if (object != space_map_object(vd->vdev_dtl_sm)) {
3514 vdev_dbgmsg(vd, "txg %llu, spa %s, DTL old object %llu, "
3515 "new object %llu", (u_longlong_t)txg, spa_name(spa),
3516 (u_longlong_t)object,
3517 (u_longlong_t)space_map_object(vd->vdev_dtl_sm));
3518 vdev_config_dirty(vd->vdev_top);
3519 }
3520
3521 dmu_tx_commit(tx);
3522 }
3523
3524 /*
3525 * Determine whether the specified vdev can be offlined/detached/removed
3526 * without losing data.
3527 */
3528 boolean_t
vdev_dtl_required(vdev_t * vd)3529 vdev_dtl_required(vdev_t *vd)
3530 {
3531 spa_t *spa = vd->vdev_spa;
3532 vdev_t *tvd = vd->vdev_top;
3533 uint8_t cant_read = vd->vdev_cant_read;
3534 boolean_t required;
3535
3536 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
3537
3538 if (vd == spa->spa_root_vdev || vd == tvd)
3539 return (B_TRUE);
3540
3541 /*
3542 * Temporarily mark the device as unreadable, and then determine
3543 * whether this results in any DTL outages in the top-level vdev.
3544 * If not, we can safely offline/detach/remove the device.
3545 */
3546 vd->vdev_cant_read = B_TRUE;
3547 vdev_dtl_reassess(tvd, 0, 0, B_FALSE, B_FALSE);
3548 required = !vdev_dtl_empty(tvd, DTL_OUTAGE);
3549 vd->vdev_cant_read = cant_read;
3550 vdev_dtl_reassess(tvd, 0, 0, B_FALSE, B_FALSE);
3551
3552 if (!required && zio_injection_enabled) {
3553 required = !!zio_handle_device_injection(vd, NULL,
3554 SET_ERROR(ECHILD));
3555 }
3556
3557 return (required);
3558 }
3559
3560 /*
3561 * Determine if resilver is needed, and if so the txg range.
3562 */
3563 boolean_t
vdev_resilver_needed(vdev_t * vd,uint64_t * minp,uint64_t * maxp)3564 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
3565 {
3566 boolean_t needed = B_FALSE;
3567 uint64_t thismin = UINT64_MAX;
3568 uint64_t thismax = 0;
3569
3570 if (vd->vdev_children == 0) {
3571 mutex_enter(&vd->vdev_dtl_lock);
3572 if (!range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]) &&
3573 vdev_writeable(vd)) {
3574
3575 thismin = vdev_dtl_min(vd);
3576 thismax = vdev_dtl_max(vd);
3577 needed = B_TRUE;
3578 }
3579 mutex_exit(&vd->vdev_dtl_lock);
3580 } else {
3581 for (int c = 0; c < vd->vdev_children; c++) {
3582 vdev_t *cvd = vd->vdev_child[c];
3583 uint64_t cmin, cmax;
3584
3585 if (vdev_resilver_needed(cvd, &cmin, &cmax)) {
3586 thismin = MIN(thismin, cmin);
3587 thismax = MAX(thismax, cmax);
3588 needed = B_TRUE;
3589 }
3590 }
3591 }
3592
3593 if (needed && minp) {
3594 *minp = thismin;
3595 *maxp = thismax;
3596 }
3597 return (needed);
3598 }
3599
3600 /*
3601 * Gets the checkpoint space map object from the vdev's ZAP. On success sm_obj
3602 * will contain either the checkpoint spacemap object or zero if none exists.
3603 * All other errors are returned to the caller.
3604 */
3605 int
vdev_checkpoint_sm_object(vdev_t * vd,uint64_t * sm_obj)3606 vdev_checkpoint_sm_object(vdev_t *vd, uint64_t *sm_obj)
3607 {
3608 ASSERT0(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER));
3609
3610 if (vd->vdev_top_zap == 0) {
3611 *sm_obj = 0;
3612 return (0);
3613 }
3614
3615 int error = zap_lookup(spa_meta_objset(vd->vdev_spa), vd->vdev_top_zap,
3616 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM, sizeof (uint64_t), 1, sm_obj);
3617 if (error == ENOENT) {
3618 *sm_obj = 0;
3619 error = 0;
3620 }
3621
3622 return (error);
3623 }
3624
3625 int
vdev_load(vdev_t * vd)3626 vdev_load(vdev_t *vd)
3627 {
3628 int children = vd->vdev_children;
3629 int error = 0;
3630 taskq_t *tq = NULL;
3631
3632 /*
3633 * It's only worthwhile to use the taskq for the root vdev, because the
3634 * slow part is metaslab_init, and that only happens for top-level
3635 * vdevs.
3636 */
3637 if (vd->vdev_ops == &vdev_root_ops && vd->vdev_children > 0) {
3638 tq = taskq_create("vdev_load", children, minclsyspri,
3639 children, children, TASKQ_PREPOPULATE);
3640 }
3641
3642 /*
3643 * Recursively load all children.
3644 */
3645 for (int c = 0; c < vd->vdev_children; c++) {
3646 vdev_t *cvd = vd->vdev_child[c];
3647
3648 if (tq == NULL || vdev_uses_zvols(cvd)) {
3649 cvd->vdev_load_error = vdev_load(cvd);
3650 } else {
3651 VERIFY(taskq_dispatch(tq, vdev_load_child,
3652 cvd, TQ_SLEEP) != TASKQID_INVALID);
3653 }
3654 }
3655
3656 if (tq != NULL) {
3657 taskq_wait(tq);
3658 taskq_destroy(tq);
3659 }
3660
3661 for (int c = 0; c < vd->vdev_children; c++) {
3662 int error = vd->vdev_child[c]->vdev_load_error;
3663
3664 if (error != 0)
3665 return (error);
3666 }
3667
3668 vdev_set_deflate_ratio(vd);
3669
3670 if (vd->vdev_ops == &vdev_raidz_ops) {
3671 error = vdev_raidz_load(vd);
3672 if (error != 0)
3673 return (error);
3674 }
3675
3676 /*
3677 * On spa_load path, grab the allocation bias from our zap
3678 */
3679 if (vd == vd->vdev_top && vd->vdev_top_zap != 0) {
3680 spa_t *spa = vd->vdev_spa;
3681 char bias_str[64];
3682
3683 error = zap_lookup(spa->spa_meta_objset, vd->vdev_top_zap,
3684 VDEV_TOP_ZAP_ALLOCATION_BIAS, 1, sizeof (bias_str),
3685 bias_str);
3686 if (error == 0) {
3687 ASSERT(vd->vdev_alloc_bias == VDEV_BIAS_NONE);
3688 vd->vdev_alloc_bias = vdev_derive_alloc_bias(bias_str);
3689 } else if (error != ENOENT) {
3690 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
3691 VDEV_AUX_CORRUPT_DATA);
3692 vdev_dbgmsg(vd, "vdev_load: zap_lookup(top_zap=%llu) "
3693 "failed [error=%d]",
3694 (u_longlong_t)vd->vdev_top_zap, error);
3695 return (error);
3696 }
3697 }
3698
3699 if (vd == vd->vdev_top && vd->vdev_top_zap != 0) {
3700 spa_t *spa = vd->vdev_spa;
3701 uint64_t failfast;
3702
3703 error = zap_lookup(spa->spa_meta_objset, vd->vdev_top_zap,
3704 vdev_prop_to_name(VDEV_PROP_FAILFAST), sizeof (failfast),
3705 1, &failfast);
3706 if (error == 0) {
3707 vd->vdev_failfast = failfast & 1;
3708 } else if (error == ENOENT) {
3709 vd->vdev_failfast = vdev_prop_default_numeric(
3710 VDEV_PROP_FAILFAST);
3711 } else {
3712 vdev_dbgmsg(vd,
3713 "vdev_load: zap_lookup(top_zap=%llu) "
3714 "failed [error=%d]",
3715 (u_longlong_t)vd->vdev_top_zap, error);
3716 }
3717 }
3718
3719 /*
3720 * Load any rebuild state from the top-level vdev zap.
3721 */
3722 if (vd == vd->vdev_top && vd->vdev_top_zap != 0) {
3723 error = vdev_rebuild_load(vd);
3724 if (error && error != ENOTSUP) {
3725 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
3726 VDEV_AUX_CORRUPT_DATA);
3727 vdev_dbgmsg(vd, "vdev_load: vdev_rebuild_load "
3728 "failed [error=%d]", error);
3729 return (error);
3730 }
3731 }
3732
3733 if (vd->vdev_top_zap != 0 || vd->vdev_leaf_zap != 0) {
3734 uint64_t zapobj;
3735
3736 if (vd->vdev_top_zap != 0)
3737 zapobj = vd->vdev_top_zap;
3738 else
3739 zapobj = vd->vdev_leaf_zap;
3740
3741 error = vdev_prop_get_int(vd, VDEV_PROP_CHECKSUM_N,
3742 &vd->vdev_checksum_n);
3743 if (error && error != ENOENT)
3744 vdev_dbgmsg(vd, "vdev_load: zap_lookup(zap=%llu) "
3745 "failed [error=%d]", (u_longlong_t)zapobj, error);
3746
3747 error = vdev_prop_get_int(vd, VDEV_PROP_CHECKSUM_T,
3748 &vd->vdev_checksum_t);
3749 if (error && error != ENOENT)
3750 vdev_dbgmsg(vd, "vdev_load: zap_lookup(zap=%llu) "
3751 "failed [error=%d]", (u_longlong_t)zapobj, error);
3752
3753 error = vdev_prop_get_int(vd, VDEV_PROP_IO_N,
3754 &vd->vdev_io_n);
3755 if (error && error != ENOENT)
3756 vdev_dbgmsg(vd, "vdev_load: zap_lookup(zap=%llu) "
3757 "failed [error=%d]", (u_longlong_t)zapobj, error);
3758
3759 error = vdev_prop_get_int(vd, VDEV_PROP_IO_T,
3760 &vd->vdev_io_t);
3761 if (error && error != ENOENT)
3762 vdev_dbgmsg(vd, "vdev_load: zap_lookup(zap=%llu) "
3763 "failed [error=%d]", (u_longlong_t)zapobj, error);
3764
3765 error = vdev_prop_get_int(vd, VDEV_PROP_SLOW_IO_N,
3766 &vd->vdev_slow_io_n);
3767 if (error && error != ENOENT)
3768 vdev_dbgmsg(vd, "vdev_load: zap_lookup(zap=%llu) "
3769 "failed [error=%d]", (u_longlong_t)zapobj, error);
3770
3771 error = vdev_prop_get_int(vd, VDEV_PROP_SLOW_IO_T,
3772 &vd->vdev_slow_io_t);
3773 if (error && error != ENOENT)
3774 vdev_dbgmsg(vd, "vdev_load: zap_lookup(zap=%llu) "
3775 "failed [error=%d]", (u_longlong_t)zapobj, error);
3776 }
3777
3778 /*
3779 * If this is a top-level vdev, initialize its metaslabs.
3780 */
3781 if (vd == vd->vdev_top && vdev_is_concrete(vd)) {
3782 vdev_metaslab_group_create(vd);
3783
3784 if (vd->vdev_ashift == 0 || vd->vdev_asize == 0) {
3785 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
3786 VDEV_AUX_CORRUPT_DATA);
3787 vdev_dbgmsg(vd, "vdev_load: invalid size. ashift=%llu, "
3788 "asize=%llu", (u_longlong_t)vd->vdev_ashift,
3789 (u_longlong_t)vd->vdev_asize);
3790 return (SET_ERROR(ENXIO));
3791 }
3792
3793 error = vdev_metaslab_init(vd, 0);
3794 if (error != 0) {
3795 vdev_dbgmsg(vd, "vdev_load: metaslab_init failed "
3796 "[error=%d]", error);
3797 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
3798 VDEV_AUX_CORRUPT_DATA);
3799 return (error);
3800 }
3801
3802 uint64_t checkpoint_sm_obj;
3803 error = vdev_checkpoint_sm_object(vd, &checkpoint_sm_obj);
3804 if (error == 0 && checkpoint_sm_obj != 0) {
3805 objset_t *mos = spa_meta_objset(vd->vdev_spa);
3806 ASSERT(vd->vdev_asize != 0);
3807 ASSERT3P(vd->vdev_checkpoint_sm, ==, NULL);
3808
3809 error = space_map_open(&vd->vdev_checkpoint_sm,
3810 mos, checkpoint_sm_obj, 0, vd->vdev_asize,
3811 vd->vdev_ashift);
3812 if (error != 0) {
3813 vdev_dbgmsg(vd, "vdev_load: space_map_open "
3814 "failed for checkpoint spacemap (obj %llu) "
3815 "[error=%d]",
3816 (u_longlong_t)checkpoint_sm_obj, error);
3817 return (error);
3818 }
3819 ASSERT3P(vd->vdev_checkpoint_sm, !=, NULL);
3820
3821 /*
3822 * Since the checkpoint_sm contains free entries
3823 * exclusively we can use space_map_allocated() to
3824 * indicate the cumulative checkpointed space that
3825 * has been freed.
3826 */
3827 vd->vdev_stat.vs_checkpoint_space =
3828 -space_map_allocated(vd->vdev_checkpoint_sm);
3829 vd->vdev_spa->spa_checkpoint_info.sci_dspace +=
3830 vd->vdev_stat.vs_checkpoint_space;
3831 } else if (error != 0) {
3832 vdev_dbgmsg(vd, "vdev_load: failed to retrieve "
3833 "checkpoint space map object from vdev ZAP "
3834 "[error=%d]", error);
3835 return (error);
3836 }
3837 }
3838
3839 /*
3840 * If this is a leaf vdev, load its DTL.
3841 */
3842 if (vd->vdev_ops->vdev_op_leaf && (error = vdev_dtl_load(vd)) != 0) {
3843 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
3844 VDEV_AUX_CORRUPT_DATA);
3845 vdev_dbgmsg(vd, "vdev_load: vdev_dtl_load failed "
3846 "[error=%d]", error);
3847 return (error);
3848 }
3849
3850 uint64_t obsolete_sm_object;
3851 error = vdev_obsolete_sm_object(vd, &obsolete_sm_object);
3852 if (error == 0 && obsolete_sm_object != 0) {
3853 objset_t *mos = vd->vdev_spa->spa_meta_objset;
3854 ASSERT(vd->vdev_asize != 0);
3855 ASSERT3P(vd->vdev_obsolete_sm, ==, NULL);
3856
3857 if ((error = space_map_open(&vd->vdev_obsolete_sm, mos,
3858 obsolete_sm_object, 0, vd->vdev_asize, 0))) {
3859 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
3860 VDEV_AUX_CORRUPT_DATA);
3861 vdev_dbgmsg(vd, "vdev_load: space_map_open failed for "
3862 "obsolete spacemap (obj %llu) [error=%d]",
3863 (u_longlong_t)obsolete_sm_object, error);
3864 return (error);
3865 }
3866 } else if (error != 0) {
3867 vdev_dbgmsg(vd, "vdev_load: failed to retrieve obsolete "
3868 "space map object from vdev ZAP [error=%d]", error);
3869 return (error);
3870 }
3871
3872 return (0);
3873 }
3874
3875 /*
3876 * The special vdev case is used for hot spares and l2cache devices. Its
3877 * sole purpose it to set the vdev state for the associated vdev. To do this,
3878 * we make sure that we can open the underlying device, then try to read the
3879 * label, and make sure that the label is sane and that it hasn't been
3880 * repurposed to another pool.
3881 */
3882 int
vdev_validate_aux(vdev_t * vd)3883 vdev_validate_aux(vdev_t *vd)
3884 {
3885 nvlist_t *label;
3886 uint64_t guid, version;
3887 uint64_t state;
3888
3889 if (!vdev_readable(vd))
3890 return (0);
3891
3892 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) {
3893 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
3894 VDEV_AUX_CORRUPT_DATA);
3895 return (-1);
3896 }
3897
3898 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 ||
3899 !SPA_VERSION_IS_SUPPORTED(version) ||
3900 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 ||
3901 guid != vd->vdev_guid ||
3902 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) {
3903 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
3904 VDEV_AUX_CORRUPT_DATA);
3905 nvlist_free(label);
3906 return (-1);
3907 }
3908
3909 /*
3910 * We don't actually check the pool state here. If it's in fact in
3911 * use by another pool, we update this fact on the fly when requested.
3912 */
3913 nvlist_free(label);
3914 return (0);
3915 }
3916
3917 static void
vdev_destroy_ms_flush_data(vdev_t * vd,dmu_tx_t * tx)3918 vdev_destroy_ms_flush_data(vdev_t *vd, dmu_tx_t *tx)
3919 {
3920 objset_t *mos = spa_meta_objset(vd->vdev_spa);
3921
3922 if (vd->vdev_top_zap == 0)
3923 return;
3924
3925 uint64_t object = 0;
3926 int err = zap_lookup(mos, vd->vdev_top_zap,
3927 VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS, sizeof (uint64_t), 1, &object);
3928 if (err == ENOENT)
3929 return;
3930 VERIFY0(err);
3931
3932 VERIFY0(dmu_object_free(mos, object, tx));
3933 VERIFY0(zap_remove(mos, vd->vdev_top_zap,
3934 VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS, tx));
3935 }
3936
3937 /*
3938 * Free the objects used to store this vdev's spacemaps, and the array
3939 * that points to them.
3940 */
3941 void
vdev_destroy_spacemaps(vdev_t * vd,dmu_tx_t * tx)3942 vdev_destroy_spacemaps(vdev_t *vd, dmu_tx_t *tx)
3943 {
3944 if (vd->vdev_ms_array == 0)
3945 return;
3946
3947 objset_t *mos = vd->vdev_spa->spa_meta_objset;
3948 uint64_t array_count = vd->vdev_asize >> vd->vdev_ms_shift;
3949 size_t array_bytes = array_count * sizeof (uint64_t);
3950 uint64_t *smobj_array = kmem_alloc(array_bytes, KM_SLEEP);
3951 VERIFY0(dmu_read(mos, vd->vdev_ms_array, 0,
3952 array_bytes, smobj_array, 0));
3953
3954 for (uint64_t i = 0; i < array_count; i++) {
3955 uint64_t smobj = smobj_array[i];
3956 if (smobj == 0)
3957 continue;
3958
3959 space_map_free_obj(mos, smobj, tx);
3960 }
3961
3962 kmem_free(smobj_array, array_bytes);
3963 VERIFY0(dmu_object_free(mos, vd->vdev_ms_array, tx));
3964 vdev_destroy_ms_flush_data(vd, tx);
3965 vd->vdev_ms_array = 0;
3966 }
3967
3968 static void
vdev_remove_empty_log(vdev_t * vd,uint64_t txg)3969 vdev_remove_empty_log(vdev_t *vd, uint64_t txg)
3970 {
3971 spa_t *spa = vd->vdev_spa;
3972
3973 ASSERT(vd->vdev_islog);
3974 ASSERT(vd == vd->vdev_top);
3975 ASSERT3U(txg, ==, spa_syncing_txg(spa));
3976
3977 dmu_tx_t *tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
3978
3979 vdev_destroy_spacemaps(vd, tx);
3980 if (vd->vdev_top_zap != 0) {
3981 vdev_destroy_unlink_zap(vd, vd->vdev_top_zap, tx);
3982 vd->vdev_top_zap = 0;
3983 }
3984
3985 dmu_tx_commit(tx);
3986 }
3987
3988 void
vdev_sync_done(vdev_t * vd,uint64_t txg)3989 vdev_sync_done(vdev_t *vd, uint64_t txg)
3990 {
3991 metaslab_t *msp;
3992 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg));
3993
3994 ASSERT(vdev_is_concrete(vd));
3995
3996 while ((msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))
3997 != NULL)
3998 metaslab_sync_done(msp, txg);
3999
4000 if (reassess) {
4001 metaslab_sync_reassess(vd->vdev_mg);
4002 if (vd->vdev_log_mg != NULL)
4003 metaslab_sync_reassess(vd->vdev_log_mg);
4004 }
4005 }
4006
4007 void
vdev_sync(vdev_t * vd,uint64_t txg)4008 vdev_sync(vdev_t *vd, uint64_t txg)
4009 {
4010 spa_t *spa = vd->vdev_spa;
4011 vdev_t *lvd;
4012 metaslab_t *msp;
4013
4014 ASSERT3U(txg, ==, spa->spa_syncing_txg);
4015 dmu_tx_t *tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
4016 if (range_tree_space(vd->vdev_obsolete_segments) > 0) {
4017 ASSERT(vd->vdev_removing ||
4018 vd->vdev_ops == &vdev_indirect_ops);
4019
4020 vdev_indirect_sync_obsolete(vd, tx);
4021
4022 /*
4023 * If the vdev is indirect, it can't have dirty
4024 * metaslabs or DTLs.
4025 */
4026 if (vd->vdev_ops == &vdev_indirect_ops) {
4027 ASSERT(txg_list_empty(&vd->vdev_ms_list, txg));
4028 ASSERT(txg_list_empty(&vd->vdev_dtl_list, txg));
4029 dmu_tx_commit(tx);
4030 return;
4031 }
4032 }
4033
4034 ASSERT(vdev_is_concrete(vd));
4035
4036 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0 &&
4037 !vd->vdev_removing) {
4038 ASSERT(vd == vd->vdev_top);
4039 ASSERT0(vd->vdev_indirect_config.vic_mapping_object);
4040 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset,
4041 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx);
4042 ASSERT(vd->vdev_ms_array != 0);
4043 vdev_config_dirty(vd);
4044 }
4045
4046 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) {
4047 metaslab_sync(msp, txg);
4048 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg));
4049 }
4050
4051 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL)
4052 vdev_dtl_sync(lvd, txg);
4053
4054 /*
4055 * If this is an empty log device being removed, destroy the
4056 * metadata associated with it.
4057 */
4058 if (vd->vdev_islog && vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing)
4059 vdev_remove_empty_log(vd, txg);
4060
4061 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg));
4062 dmu_tx_commit(tx);
4063 }
4064
4065 /*
4066 * Return the amount of space that should be (or was) allocated for the given
4067 * psize (compressed block size) in the given TXG. Note that for expanded
4068 * RAIDZ vdevs, the size allocated for older BP's may be larger. See
4069 * vdev_raidz_asize().
4070 */
4071 uint64_t
vdev_psize_to_asize_txg(vdev_t * vd,uint64_t psize,uint64_t txg)4072 vdev_psize_to_asize_txg(vdev_t *vd, uint64_t psize, uint64_t txg)
4073 {
4074 return (vd->vdev_ops->vdev_op_asize(vd, psize, txg));
4075 }
4076
4077 uint64_t
vdev_psize_to_asize(vdev_t * vd,uint64_t psize)4078 vdev_psize_to_asize(vdev_t *vd, uint64_t psize)
4079 {
4080 return (vdev_psize_to_asize_txg(vd, psize, 0));
4081 }
4082
4083 /*
4084 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
4085 * not be opened, and no I/O is attempted.
4086 */
4087 int
vdev_fault(spa_t * spa,uint64_t guid,vdev_aux_t aux)4088 vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux)
4089 {
4090 vdev_t *vd, *tvd;
4091
4092 spa_vdev_state_enter(spa, SCL_NONE);
4093
4094 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
4095 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENODEV)));
4096
4097 if (!vd->vdev_ops->vdev_op_leaf)
4098 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENOTSUP)));
4099
4100 tvd = vd->vdev_top;
4101
4102 /*
4103 * If user did a 'zpool offline -f' then make the fault persist across
4104 * reboots.
4105 */
4106 if (aux == VDEV_AUX_EXTERNAL_PERSIST) {
4107 /*
4108 * There are two kinds of forced faults: temporary and
4109 * persistent. Temporary faults go away at pool import, while
4110 * persistent faults stay set. Both types of faults can be
4111 * cleared with a zpool clear.
4112 *
4113 * We tell if a vdev is persistently faulted by looking at the
4114 * ZPOOL_CONFIG_AUX_STATE nvpair. If it's set to "external" at
4115 * import then it's a persistent fault. Otherwise, it's
4116 * temporary. We get ZPOOL_CONFIG_AUX_STATE set to "external"
4117 * by setting vd.vdev_stat.vs_aux to VDEV_AUX_EXTERNAL. This
4118 * tells vdev_config_generate() (which gets run later) to set
4119 * ZPOOL_CONFIG_AUX_STATE to "external" in the nvlist.
4120 */
4121 vd->vdev_stat.vs_aux = VDEV_AUX_EXTERNAL;
4122 vd->vdev_tmpoffline = B_FALSE;
4123 aux = VDEV_AUX_EXTERNAL;
4124 } else {
4125 vd->vdev_tmpoffline = B_TRUE;
4126 }
4127
4128 /*
4129 * We don't directly use the aux state here, but if we do a
4130 * vdev_reopen(), we need this value to be present to remember why we
4131 * were faulted.
4132 */
4133 vd->vdev_label_aux = aux;
4134
4135 /*
4136 * Faulted state takes precedence over degraded.
4137 */
4138 vd->vdev_delayed_close = B_FALSE;
4139 vd->vdev_faulted = 1ULL;
4140 vd->vdev_degraded = 0ULL;
4141 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux);
4142
4143 /*
4144 * If this device has the only valid copy of the data, then
4145 * back off and simply mark the vdev as degraded instead.
4146 */
4147 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) {
4148 vd->vdev_degraded = 1ULL;
4149 vd->vdev_faulted = 0ULL;
4150
4151 /*
4152 * If we reopen the device and it's not dead, only then do we
4153 * mark it degraded.
4154 */
4155 vdev_reopen(tvd);
4156
4157 if (vdev_readable(vd))
4158 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux);
4159 }
4160
4161 return (spa_vdev_state_exit(spa, vd, 0));
4162 }
4163
4164 /*
4165 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
4166 * user that something is wrong. The vdev continues to operate as normal as far
4167 * as I/O is concerned.
4168 */
4169 int
vdev_degrade(spa_t * spa,uint64_t guid,vdev_aux_t aux)4170 vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux)
4171 {
4172 vdev_t *vd;
4173
4174 spa_vdev_state_enter(spa, SCL_NONE);
4175
4176 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
4177 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENODEV)));
4178
4179 if (!vd->vdev_ops->vdev_op_leaf)
4180 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENOTSUP)));
4181
4182 /*
4183 * If the vdev is already faulted, then don't do anything.
4184 */
4185 if (vd->vdev_faulted || vd->vdev_degraded)
4186 return (spa_vdev_state_exit(spa, NULL, 0));
4187
4188 vd->vdev_degraded = 1ULL;
4189 if (!vdev_is_dead(vd))
4190 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
4191 aux);
4192
4193 return (spa_vdev_state_exit(spa, vd, 0));
4194 }
4195
4196 int
vdev_remove_wanted(spa_t * spa,uint64_t guid)4197 vdev_remove_wanted(spa_t *spa, uint64_t guid)
4198 {
4199 vdev_t *vd;
4200
4201 spa_vdev_state_enter(spa, SCL_NONE);
4202
4203 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
4204 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENODEV)));
4205
4206 /*
4207 * If the vdev is already removed, or expanding which can trigger
4208 * repartition add/remove events, then don't do anything.
4209 */
4210 if (vd->vdev_removed || vd->vdev_expanding)
4211 return (spa_vdev_state_exit(spa, NULL, 0));
4212
4213 /*
4214 * Confirm the vdev has been removed, otherwise don't do anything.
4215 */
4216 if (vd->vdev_ops->vdev_op_leaf && !zio_wait(vdev_probe(vd, NULL)))
4217 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(EEXIST)));
4218
4219 vd->vdev_remove_wanted = B_TRUE;
4220 spa_async_request(spa, SPA_ASYNC_REMOVE);
4221
4222 return (spa_vdev_state_exit(spa, vd, 0));
4223 }
4224
4225
4226 /*
4227 * Online the given vdev.
4228 *
4229 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
4230 * spare device should be detached when the device finishes resilvering.
4231 * Second, the online should be treated like a 'test' online case, so no FMA
4232 * events are generated if the device fails to open.
4233 */
4234 int
vdev_online(spa_t * spa,uint64_t guid,uint64_t flags,vdev_state_t * newstate)4235 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
4236 {
4237 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev;
4238 boolean_t wasoffline;
4239 vdev_state_t oldstate;
4240
4241 spa_vdev_state_enter(spa, SCL_NONE);
4242
4243 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
4244 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENODEV)));
4245
4246 wasoffline = (vd->vdev_offline || vd->vdev_tmpoffline);
4247 oldstate = vd->vdev_state;
4248
4249 tvd = vd->vdev_top;
4250 vd->vdev_offline = B_FALSE;
4251 vd->vdev_tmpoffline = B_FALSE;
4252 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE);
4253 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT);
4254
4255 /* XXX - L2ARC 1.0 does not support expansion */
4256 if (!vd->vdev_aux) {
4257 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
4258 pvd->vdev_expanding = !!((flags & ZFS_ONLINE_EXPAND) ||
4259 spa->spa_autoexpand);
4260 vd->vdev_expansion_time = gethrestime_sec();
4261 }
4262
4263 vdev_reopen(tvd);
4264 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE;
4265
4266 if (!vd->vdev_aux) {
4267 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
4268 pvd->vdev_expanding = B_FALSE;
4269 }
4270
4271 if (newstate)
4272 *newstate = vd->vdev_state;
4273 if ((flags & ZFS_ONLINE_UNSPARE) &&
4274 !vdev_is_dead(vd) && vd->vdev_parent &&
4275 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
4276 vd->vdev_parent->vdev_child[0] == vd)
4277 vd->vdev_unspare = B_TRUE;
4278
4279 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) {
4280
4281 /* XXX - L2ARC 1.0 does not support expansion */
4282 if (vd->vdev_aux)
4283 return (spa_vdev_state_exit(spa, vd, ENOTSUP));
4284 spa->spa_ccw_fail_time = 0;
4285 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
4286 }
4287
4288 /* Restart initializing if necessary */
4289 mutex_enter(&vd->vdev_initialize_lock);
4290 if (vdev_writeable(vd) &&
4291 vd->vdev_initialize_thread == NULL &&
4292 vd->vdev_initialize_state == VDEV_INITIALIZE_ACTIVE) {
4293 (void) vdev_initialize(vd);
4294 }
4295 mutex_exit(&vd->vdev_initialize_lock);
4296
4297 /*
4298 * Restart trimming if necessary. We do not restart trimming for cache
4299 * devices here. This is triggered by l2arc_rebuild_vdev()
4300 * asynchronously for the whole device or in l2arc_evict() as it evicts
4301 * space for upcoming writes.
4302 */
4303 mutex_enter(&vd->vdev_trim_lock);
4304 if (vdev_writeable(vd) && !vd->vdev_isl2cache &&
4305 vd->vdev_trim_thread == NULL &&
4306 vd->vdev_trim_state == VDEV_TRIM_ACTIVE) {
4307 (void) vdev_trim(vd, vd->vdev_trim_rate, vd->vdev_trim_partial,
4308 vd->vdev_trim_secure);
4309 }
4310 mutex_exit(&vd->vdev_trim_lock);
4311
4312 if (wasoffline ||
4313 (oldstate < VDEV_STATE_DEGRADED &&
4314 vd->vdev_state >= VDEV_STATE_DEGRADED)) {
4315 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_ONLINE);
4316
4317 /*
4318 * Asynchronously detach spare vdev if resilver or
4319 * rebuild is not required
4320 */
4321 if (vd->vdev_unspare &&
4322 !dsl_scan_resilvering(spa->spa_dsl_pool) &&
4323 !dsl_scan_resilver_scheduled(spa->spa_dsl_pool) &&
4324 !vdev_rebuild_active(tvd))
4325 spa_async_request(spa, SPA_ASYNC_DETACH_SPARE);
4326 }
4327 return (spa_vdev_state_exit(spa, vd, 0));
4328 }
4329
4330 static int
vdev_offline_locked(spa_t * spa,uint64_t guid,uint64_t flags)4331 vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags)
4332 {
4333 vdev_t *vd, *tvd;
4334 int error = 0;
4335 uint64_t generation;
4336 metaslab_group_t *mg;
4337
4338 top:
4339 spa_vdev_state_enter(spa, SCL_ALLOC);
4340
4341 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
4342 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENODEV)));
4343
4344 if (!vd->vdev_ops->vdev_op_leaf)
4345 return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENOTSUP)));
4346
4347 if (vd->vdev_ops == &vdev_draid_spare_ops)
4348 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
4349
4350 tvd = vd->vdev_top;
4351 mg = tvd->vdev_mg;
4352 generation = spa->spa_config_generation + 1;
4353
4354 /*
4355 * If the device isn't already offline, try to offline it.
4356 */
4357 if (!vd->vdev_offline) {
4358 /*
4359 * If this device has the only valid copy of some data,
4360 * don't allow it to be offlined. Log devices are always
4361 * expendable.
4362 */
4363 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
4364 vdev_dtl_required(vd))
4365 return (spa_vdev_state_exit(spa, NULL,
4366 SET_ERROR(EBUSY)));
4367
4368 /*
4369 * If the top-level is a slog and it has had allocations
4370 * then proceed. We check that the vdev's metaslab group
4371 * is not NULL since it's possible that we may have just
4372 * added this vdev but not yet initialized its metaslabs.
4373 */
4374 if (tvd->vdev_islog && mg != NULL) {
4375 /*
4376 * Prevent any future allocations.
4377 */
4378 ASSERT3P(tvd->vdev_log_mg, ==, NULL);
4379 metaslab_group_passivate(mg);
4380 (void) spa_vdev_state_exit(spa, vd, 0);
4381
4382 error = spa_reset_logs(spa);
4383
4384 /*
4385 * If the log device was successfully reset but has
4386 * checkpointed data, do not offline it.
4387 */
4388 if (error == 0 &&
4389 tvd->vdev_checkpoint_sm != NULL) {
4390 ASSERT3U(space_map_allocated(
4391 tvd->vdev_checkpoint_sm), !=, 0);
4392 error = ZFS_ERR_CHECKPOINT_EXISTS;
4393 }
4394
4395 spa_vdev_state_enter(spa, SCL_ALLOC);
4396
4397 /*
4398 * Check to see if the config has changed.
4399 */
4400 if (error || generation != spa->spa_config_generation) {
4401 metaslab_group_activate(mg);
4402 if (error)
4403 return (spa_vdev_state_exit(spa,
4404 vd, error));
4405 (void) spa_vdev_state_exit(spa, vd, 0);
4406 goto top;
4407 }
4408 ASSERT0(tvd->vdev_stat.vs_alloc);
4409 }
4410
4411 /*
4412 * Offline this device and reopen its top-level vdev.
4413 * If the top-level vdev is a log device then just offline
4414 * it. Otherwise, if this action results in the top-level
4415 * vdev becoming unusable, undo it and fail the request.
4416 */
4417 vd->vdev_offline = B_TRUE;
4418 vdev_reopen(tvd);
4419
4420 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
4421 vdev_is_dead(tvd)) {
4422 vd->vdev_offline = B_FALSE;
4423 vdev_reopen(tvd);
4424 return (spa_vdev_state_exit(spa, NULL,
4425 SET_ERROR(EBUSY)));
4426 }
4427
4428 /*
4429 * Add the device back into the metaslab rotor so that
4430 * once we online the device it's open for business.
4431 */
4432 if (tvd->vdev_islog && mg != NULL)
4433 metaslab_group_activate(mg);
4434 }
4435
4436 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY);
4437
4438 return (spa_vdev_state_exit(spa, vd, 0));
4439 }
4440
4441 int
vdev_offline(spa_t * spa,uint64_t guid,uint64_t flags)4442 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags)
4443 {
4444 int error;
4445
4446 mutex_enter(&spa->spa_vdev_top_lock);
4447 error = vdev_offline_locked(spa, guid, flags);
4448 mutex_exit(&spa->spa_vdev_top_lock);
4449
4450 return (error);
4451 }
4452
4453 /*
4454 * Clear the error counts associated with this vdev. Unlike vdev_online() and
4455 * vdev_offline(), we assume the spa config is locked. We also clear all
4456 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
4457 */
4458 void
vdev_clear(spa_t * spa,vdev_t * vd)4459 vdev_clear(spa_t *spa, vdev_t *vd)
4460 {
4461 vdev_t *rvd = spa->spa_root_vdev;
4462
4463 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
4464
4465 if (vd == NULL)
4466 vd = rvd;
4467
4468 vd->vdev_stat.vs_read_errors = 0;
4469 vd->vdev_stat.vs_write_errors = 0;
4470 vd->vdev_stat.vs_checksum_errors = 0;
4471 vd->vdev_stat.vs_slow_ios = 0;
4472
4473 for (int c = 0; c < vd->vdev_children; c++)
4474 vdev_clear(spa, vd->vdev_child[c]);
4475
4476 /*
4477 * It makes no sense to "clear" an indirect or removed vdev.
4478 */
4479 if (!vdev_is_concrete(vd) || vd->vdev_removed)
4480 return;
4481
4482 /*
4483 * If we're in the FAULTED state or have experienced failed I/O, then
4484 * clear the persistent state and attempt to reopen the device. We
4485 * also mark the vdev config dirty, so that the new faulted state is
4486 * written out to disk.
4487 */
4488 if (vd->vdev_faulted || vd->vdev_degraded ||
4489 !vdev_readable(vd) || !vdev_writeable(vd)) {
4490 /*
4491 * When reopening in response to a clear event, it may be due to
4492 * a fmadm repair request. In this case, if the device is
4493 * still broken, we want to still post the ereport again.
4494 */
4495 vd->vdev_forcefault = B_TRUE;
4496
4497 vd->vdev_faulted = vd->vdev_degraded = 0ULL;
4498 vd->vdev_cant_read = B_FALSE;
4499 vd->vdev_cant_write = B_FALSE;
4500 vd->vdev_stat.vs_aux = 0;
4501
4502 vdev_reopen(vd == rvd ? rvd : vd->vdev_top);
4503
4504 vd->vdev_forcefault = B_FALSE;
4505
4506 if (vd != rvd && vdev_writeable(vd->vdev_top))
4507 vdev_state_dirty(vd->vdev_top);
4508
4509 /* If a resilver isn't required, check if vdevs can be culled */
4510 if (vd->vdev_aux == NULL && !vdev_is_dead(vd) &&
4511 !dsl_scan_resilvering(spa->spa_dsl_pool) &&
4512 !dsl_scan_resilver_scheduled(spa->spa_dsl_pool))
4513 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
4514
4515 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_CLEAR);
4516 }
4517
4518 /*
4519 * When clearing a FMA-diagnosed fault, we always want to
4520 * unspare the device, as we assume that the original spare was
4521 * done in response to the FMA fault.
4522 */
4523 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL &&
4524 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
4525 vd->vdev_parent->vdev_child[0] == vd)
4526 vd->vdev_unspare = B_TRUE;
4527
4528 /* Clear recent error events cache (i.e. duplicate events tracking) */
4529 zfs_ereport_clear(spa, vd);
4530 }
4531
4532 boolean_t
vdev_is_dead(vdev_t * vd)4533 vdev_is_dead(vdev_t *vd)
4534 {
4535 /*
4536 * Holes and missing devices are always considered "dead".
4537 * This simplifies the code since we don't have to check for
4538 * these types of devices in the various code paths.
4539 * Instead we rely on the fact that we skip over dead devices
4540 * before issuing I/O to them.
4541 */
4542 return (vd->vdev_state < VDEV_STATE_DEGRADED ||
4543 vd->vdev_ops == &vdev_hole_ops ||
4544 vd->vdev_ops == &vdev_missing_ops);
4545 }
4546
4547 boolean_t
vdev_readable(vdev_t * vd)4548 vdev_readable(vdev_t *vd)
4549 {
4550 return (!vdev_is_dead(vd) && !vd->vdev_cant_read);
4551 }
4552
4553 boolean_t
vdev_writeable(vdev_t * vd)4554 vdev_writeable(vdev_t *vd)
4555 {
4556 return (!vdev_is_dead(vd) && !vd->vdev_cant_write &&
4557 vdev_is_concrete(vd));
4558 }
4559
4560 boolean_t
vdev_allocatable(vdev_t * vd)4561 vdev_allocatable(vdev_t *vd)
4562 {
4563 uint64_t state = vd->vdev_state;
4564
4565 /*
4566 * We currently allow allocations from vdevs which may be in the
4567 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
4568 * fails to reopen then we'll catch it later when we're holding
4569 * the proper locks. Note that we have to get the vdev state
4570 * in a local variable because although it changes atomically,
4571 * we're asking two separate questions about it.
4572 */
4573 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) &&
4574 !vd->vdev_cant_write && vdev_is_concrete(vd) &&
4575 vd->vdev_mg->mg_initialized);
4576 }
4577
4578 boolean_t
vdev_accessible(vdev_t * vd,zio_t * zio)4579 vdev_accessible(vdev_t *vd, zio_t *zio)
4580 {
4581 ASSERT(zio->io_vd == vd);
4582
4583 if (vdev_is_dead(vd) || vd->vdev_remove_wanted)
4584 return (B_FALSE);
4585
4586 if (zio->io_type == ZIO_TYPE_READ)
4587 return (!vd->vdev_cant_read);
4588
4589 if (zio->io_type == ZIO_TYPE_WRITE)
4590 return (!vd->vdev_cant_write);
4591
4592 return (B_TRUE);
4593 }
4594
4595 static void
vdev_get_child_stat(vdev_t * cvd,vdev_stat_t * vs,vdev_stat_t * cvs)4596 vdev_get_child_stat(vdev_t *cvd, vdev_stat_t *vs, vdev_stat_t *cvs)
4597 {
4598 /*
4599 * Exclude the dRAID spare when aggregating to avoid double counting
4600 * the ops and bytes. These IOs are counted by the physical leaves.
4601 */
4602 if (cvd->vdev_ops == &vdev_draid_spare_ops)
4603 return;
4604
4605 for (int t = 0; t < VS_ZIO_TYPES; t++) {
4606 vs->vs_ops[t] += cvs->vs_ops[t];
4607 vs->vs_bytes[t] += cvs->vs_bytes[t];
4608 }
4609
4610 cvs->vs_scan_removing = cvd->vdev_removing;
4611 }
4612
4613 /*
4614 * Get extended stats
4615 */
4616 static void
vdev_get_child_stat_ex(vdev_t * cvd,vdev_stat_ex_t * vsx,vdev_stat_ex_t * cvsx)4617 vdev_get_child_stat_ex(vdev_t *cvd, vdev_stat_ex_t *vsx, vdev_stat_ex_t *cvsx)
4618 {
4619 (void) cvd;
4620
4621 int t, b;
4622 for (t = 0; t < ZIO_TYPES; t++) {
4623 for (b = 0; b < ARRAY_SIZE(vsx->vsx_disk_histo[0]); b++)
4624 vsx->vsx_disk_histo[t][b] += cvsx->vsx_disk_histo[t][b];
4625
4626 for (b = 0; b < ARRAY_SIZE(vsx->vsx_total_histo[0]); b++) {
4627 vsx->vsx_total_histo[t][b] +=
4628 cvsx->vsx_total_histo[t][b];
4629 }
4630 }
4631
4632 for (t = 0; t < ZIO_PRIORITY_NUM_QUEUEABLE; t++) {
4633 for (b = 0; b < ARRAY_SIZE(vsx->vsx_queue_histo[0]); b++) {
4634 vsx->vsx_queue_histo[t][b] +=
4635 cvsx->vsx_queue_histo[t][b];
4636 }
4637 vsx->vsx_active_queue[t] += cvsx->vsx_active_queue[t];
4638 vsx->vsx_pend_queue[t] += cvsx->vsx_pend_queue[t];
4639
4640 for (b = 0; b < ARRAY_SIZE(vsx->vsx_ind_histo[0]); b++)
4641 vsx->vsx_ind_histo[t][b] += cvsx->vsx_ind_histo[t][b];
4642
4643 for (b = 0; b < ARRAY_SIZE(vsx->vsx_agg_histo[0]); b++)
4644 vsx->vsx_agg_histo[t][b] += cvsx->vsx_agg_histo[t][b];
4645 }
4646
4647 }
4648
4649 boolean_t
vdev_is_spacemap_addressable(vdev_t * vd)4650 vdev_is_spacemap_addressable(vdev_t *vd)
4651 {
4652 if (spa_feature_is_active(vd->vdev_spa, SPA_FEATURE_SPACEMAP_V2))
4653 return (B_TRUE);
4654
4655 /*
4656 * If double-word space map entries are not enabled we assume
4657 * 47 bits of the space map entry are dedicated to the entry's
4658 * offset (see SM_OFFSET_BITS in space_map.h). We then use that
4659 * to calculate the maximum address that can be described by a
4660 * space map entry for the given device.
4661 */
4662 uint64_t shift = vd->vdev_ashift + SM_OFFSET_BITS;
4663
4664 if (shift >= 63) /* detect potential overflow */
4665 return (B_TRUE);
4666
4667 return (vd->vdev_asize < (1ULL << shift));
4668 }
4669
4670 /*
4671 * Get statistics for the given vdev.
4672 */
4673 static void
vdev_get_stats_ex_impl(vdev_t * vd,vdev_stat_t * vs,vdev_stat_ex_t * vsx)4674 vdev_get_stats_ex_impl(vdev_t *vd, vdev_stat_t *vs, vdev_stat_ex_t *vsx)
4675 {
4676 int t;
4677 /*
4678 * If we're getting stats on the root vdev, aggregate the I/O counts
4679 * over all top-level vdevs (i.e. the direct children of the root).
4680 */
4681 if (!vd->vdev_ops->vdev_op_leaf) {
4682 if (vs) {
4683 memset(vs->vs_ops, 0, sizeof (vs->vs_ops));
4684 memset(vs->vs_bytes, 0, sizeof (vs->vs_bytes));
4685 }
4686 if (vsx)
4687 memset(vsx, 0, sizeof (*vsx));
4688
4689 for (int c = 0; c < vd->vdev_children; c++) {
4690 vdev_t *cvd = vd->vdev_child[c];
4691 vdev_stat_t *cvs = &cvd->vdev_stat;
4692 vdev_stat_ex_t *cvsx = &cvd->vdev_stat_ex;
4693
4694 vdev_get_stats_ex_impl(cvd, cvs, cvsx);
4695 if (vs)
4696 vdev_get_child_stat(cvd, vs, cvs);
4697 if (vsx)
4698 vdev_get_child_stat_ex(cvd, vsx, cvsx);
4699 }
4700 } else {
4701 /*
4702 * We're a leaf. Just copy our ZIO active queue stats in. The
4703 * other leaf stats are updated in vdev_stat_update().
4704 */
4705 if (!vsx)
4706 return;
4707
4708 memcpy(vsx, &vd->vdev_stat_ex, sizeof (vd->vdev_stat_ex));
4709
4710 for (t = 0; t < ZIO_PRIORITY_NUM_QUEUEABLE; t++) {
4711 vsx->vsx_active_queue[t] = vd->vdev_queue.vq_cactive[t];
4712 vsx->vsx_pend_queue[t] = vdev_queue_class_length(vd, t);
4713 }
4714 }
4715 }
4716
4717 void
vdev_get_stats_ex(vdev_t * vd,vdev_stat_t * vs,vdev_stat_ex_t * vsx)4718 vdev_get_stats_ex(vdev_t *vd, vdev_stat_t *vs, vdev_stat_ex_t *vsx)
4719 {
4720 vdev_t *tvd = vd->vdev_top;
4721 mutex_enter(&vd->vdev_stat_lock);
4722 if (vs) {
4723 memcpy(vs, &vd->vdev_stat, sizeof (*vs));
4724 vs->vs_timestamp = gethrtime() - vs->vs_timestamp;
4725 vs->vs_state = vd->vdev_state;
4726 vs->vs_rsize = vdev_get_min_asize(vd);
4727
4728 if (vd->vdev_ops->vdev_op_leaf) {
4729 vs->vs_pspace = vd->vdev_psize;
4730 vs->vs_rsize += VDEV_LABEL_START_SIZE +
4731 VDEV_LABEL_END_SIZE;
4732 /*
4733 * Report initializing progress. Since we don't
4734 * have the initializing locks held, this is only
4735 * an estimate (although a fairly accurate one).
4736 */
4737 vs->vs_initialize_bytes_done =
4738 vd->vdev_initialize_bytes_done;
4739 vs->vs_initialize_bytes_est =
4740 vd->vdev_initialize_bytes_est;
4741 vs->vs_initialize_state = vd->vdev_initialize_state;
4742 vs->vs_initialize_action_time =
4743 vd->vdev_initialize_action_time;
4744
4745 /*
4746 * Report manual TRIM progress. Since we don't have
4747 * the manual TRIM locks held, this is only an
4748 * estimate (although fairly accurate one).
4749 */
4750 vs->vs_trim_notsup = !vd->vdev_has_trim;
4751 vs->vs_trim_bytes_done = vd->vdev_trim_bytes_done;
4752 vs->vs_trim_bytes_est = vd->vdev_trim_bytes_est;
4753 vs->vs_trim_state = vd->vdev_trim_state;
4754 vs->vs_trim_action_time = vd->vdev_trim_action_time;
4755
4756 /* Set when there is a deferred resilver. */
4757 vs->vs_resilver_deferred = vd->vdev_resilver_deferred;
4758 }
4759
4760 /*
4761 * Report expandable space on top-level, non-auxiliary devices
4762 * only. The expandable space is reported in terms of metaslab
4763 * sized units since that determines how much space the pool
4764 * can expand.
4765 */
4766 if (vd->vdev_aux == NULL && tvd != NULL) {
4767 vs->vs_esize = P2ALIGN(
4768 vd->vdev_max_asize - vd->vdev_asize,
4769 1ULL << tvd->vdev_ms_shift);
4770 }
4771
4772 vs->vs_configured_ashift = vd->vdev_top != NULL
4773 ? vd->vdev_top->vdev_ashift : vd->vdev_ashift;
4774 vs->vs_logical_ashift = vd->vdev_logical_ashift;
4775 if (vd->vdev_physical_ashift <= ASHIFT_MAX)
4776 vs->vs_physical_ashift = vd->vdev_physical_ashift;
4777 else
4778 vs->vs_physical_ashift = 0;
4779
4780 /*
4781 * Report fragmentation and rebuild progress for top-level,
4782 * non-auxiliary, concrete devices.
4783 */
4784 if (vd->vdev_aux == NULL && vd == vd->vdev_top &&
4785 vdev_is_concrete(vd)) {
4786 /*
4787 * The vdev fragmentation rating doesn't take into
4788 * account the embedded slog metaslab (vdev_log_mg).
4789 * Since it's only one metaslab, it would have a tiny
4790 * impact on the overall fragmentation.
4791 */
4792 vs->vs_fragmentation = (vd->vdev_mg != NULL) ?
4793 vd->vdev_mg->mg_fragmentation : 0;
4794 }
4795 vs->vs_noalloc = MAX(vd->vdev_noalloc,
4796 tvd ? tvd->vdev_noalloc : 0);
4797 }
4798
4799 vdev_get_stats_ex_impl(vd, vs, vsx);
4800 mutex_exit(&vd->vdev_stat_lock);
4801 }
4802
4803 void
vdev_get_stats(vdev_t * vd,vdev_stat_t * vs)4804 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
4805 {
4806 return (vdev_get_stats_ex(vd, vs, NULL));
4807 }
4808
4809 void
vdev_clear_stats(vdev_t * vd)4810 vdev_clear_stats(vdev_t *vd)
4811 {
4812 mutex_enter(&vd->vdev_stat_lock);
4813 vd->vdev_stat.vs_space = 0;
4814 vd->vdev_stat.vs_dspace = 0;
4815 vd->vdev_stat.vs_alloc = 0;
4816 mutex_exit(&vd->vdev_stat_lock);
4817 }
4818
4819 void
vdev_scan_stat_init(vdev_t * vd)4820 vdev_scan_stat_init(vdev_t *vd)
4821 {
4822 vdev_stat_t *vs = &vd->vdev_stat;
4823
4824 for (int c = 0; c < vd->vdev_children; c++)
4825 vdev_scan_stat_init(vd->vdev_child[c]);
4826
4827 mutex_enter(&vd->vdev_stat_lock);
4828 vs->vs_scan_processed = 0;
4829 mutex_exit(&vd->vdev_stat_lock);
4830 }
4831
4832 void
vdev_stat_update(zio_t * zio,uint64_t psize)4833 vdev_stat_update(zio_t *zio, uint64_t psize)
4834 {
4835 spa_t *spa = zio->io_spa;
4836 vdev_t *rvd = spa->spa_root_vdev;
4837 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd;
4838 vdev_t *pvd;
4839 uint64_t txg = zio->io_txg;
4840 /* Suppress ASAN false positive */
4841 #ifdef __SANITIZE_ADDRESS__
4842 vdev_stat_t *vs = vd ? &vd->vdev_stat : NULL;
4843 vdev_stat_ex_t *vsx = vd ? &vd->vdev_stat_ex : NULL;
4844 #else
4845 vdev_stat_t *vs = &vd->vdev_stat;
4846 vdev_stat_ex_t *vsx = &vd->vdev_stat_ex;
4847 #endif
4848 zio_type_t type = zio->io_type;
4849 int flags = zio->io_flags;
4850
4851 /*
4852 * If this i/o is a gang leader, it didn't do any actual work.
4853 */
4854 if (zio->io_gang_tree)
4855 return;
4856
4857 if (zio->io_error == 0) {
4858 /*
4859 * If this is a root i/o, don't count it -- we've already
4860 * counted the top-level vdevs, and vdev_get_stats() will
4861 * aggregate them when asked. This reduces contention on
4862 * the root vdev_stat_lock and implicitly handles blocks
4863 * that compress away to holes, for which there is no i/o.
4864 * (Holes never create vdev children, so all the counters
4865 * remain zero, which is what we want.)
4866 *
4867 * Note: this only applies to successful i/o (io_error == 0)
4868 * because unlike i/o counts, errors are not additive.
4869 * When reading a ditto block, for example, failure of
4870 * one top-level vdev does not imply a root-level error.
4871 */
4872 if (vd == rvd)
4873 return;
4874
4875 ASSERT(vd == zio->io_vd);
4876
4877 if (flags & ZIO_FLAG_IO_BYPASS)
4878 return;
4879
4880 mutex_enter(&vd->vdev_stat_lock);
4881
4882 if (flags & ZIO_FLAG_IO_REPAIR) {
4883 /*
4884 * Repair is the result of a resilver issued by the
4885 * scan thread (spa_sync).
4886 */
4887 if (flags & ZIO_FLAG_SCAN_THREAD) {
4888 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
4889 dsl_scan_phys_t *scn_phys = &scn->scn_phys;
4890 uint64_t *processed = &scn_phys->scn_processed;
4891
4892 if (vd->vdev_ops->vdev_op_leaf)
4893 atomic_add_64(processed, psize);
4894 vs->vs_scan_processed += psize;
4895 }
4896
4897 /*
4898 * Repair is the result of a rebuild issued by the
4899 * rebuild thread (vdev_rebuild_thread). To avoid
4900 * double counting repaired bytes the virtual dRAID
4901 * spare vdev is excluded from the processed bytes.
4902 */
4903 if (zio->io_priority == ZIO_PRIORITY_REBUILD) {
4904 vdev_t *tvd = vd->vdev_top;
4905 vdev_rebuild_t *vr = &tvd->vdev_rebuild_config;
4906 vdev_rebuild_phys_t *vrp = &vr->vr_rebuild_phys;
4907 uint64_t *rebuilt = &vrp->vrp_bytes_rebuilt;
4908
4909 if (vd->vdev_ops->vdev_op_leaf &&
4910 vd->vdev_ops != &vdev_draid_spare_ops) {
4911 atomic_add_64(rebuilt, psize);
4912 }
4913 vs->vs_rebuild_processed += psize;
4914 }
4915
4916 if (flags & ZIO_FLAG_SELF_HEAL)
4917 vs->vs_self_healed += psize;
4918 }
4919
4920 /*
4921 * The bytes/ops/histograms are recorded at the leaf level and
4922 * aggregated into the higher level vdevs in vdev_get_stats().
4923 */
4924 if (vd->vdev_ops->vdev_op_leaf &&
4925 (zio->io_priority < ZIO_PRIORITY_NUM_QUEUEABLE)) {
4926 zio_type_t vs_type = type;
4927 zio_priority_t priority = zio->io_priority;
4928
4929 /*
4930 * TRIM ops and bytes are reported to user space as
4931 * ZIO_TYPE_FLUSH. This is done to preserve the
4932 * vdev_stat_t structure layout for user space.
4933 */
4934 if (type == ZIO_TYPE_TRIM)
4935 vs_type = ZIO_TYPE_FLUSH;
4936
4937 /*
4938 * Solely for the purposes of 'zpool iostat -lqrw'
4939 * reporting use the priority to categorize the IO.
4940 * Only the following are reported to user space:
4941 *
4942 * ZIO_PRIORITY_SYNC_READ,
4943 * ZIO_PRIORITY_SYNC_WRITE,
4944 * ZIO_PRIORITY_ASYNC_READ,
4945 * ZIO_PRIORITY_ASYNC_WRITE,
4946 * ZIO_PRIORITY_SCRUB,
4947 * ZIO_PRIORITY_TRIM,
4948 * ZIO_PRIORITY_REBUILD.
4949 */
4950 if (priority == ZIO_PRIORITY_INITIALIZING) {
4951 ASSERT3U(type, ==, ZIO_TYPE_WRITE);
4952 priority = ZIO_PRIORITY_ASYNC_WRITE;
4953 } else if (priority == ZIO_PRIORITY_REMOVAL) {
4954 priority = ((type == ZIO_TYPE_WRITE) ?
4955 ZIO_PRIORITY_ASYNC_WRITE :
4956 ZIO_PRIORITY_ASYNC_READ);
4957 }
4958
4959 vs->vs_ops[vs_type]++;
4960 vs->vs_bytes[vs_type] += psize;
4961
4962 if (flags & ZIO_FLAG_DELEGATED) {
4963 vsx->vsx_agg_histo[priority]
4964 [RQ_HISTO(zio->io_size)]++;
4965 } else {
4966 vsx->vsx_ind_histo[priority]
4967 [RQ_HISTO(zio->io_size)]++;
4968 }
4969
4970 if (zio->io_delta && zio->io_delay) {
4971 vsx->vsx_queue_histo[priority]
4972 [L_HISTO(zio->io_delta - zio->io_delay)]++;
4973 vsx->vsx_disk_histo[type]
4974 [L_HISTO(zio->io_delay)]++;
4975 vsx->vsx_total_histo[type]
4976 [L_HISTO(zio->io_delta)]++;
4977 }
4978 }
4979
4980 mutex_exit(&vd->vdev_stat_lock);
4981 return;
4982 }
4983
4984 if (flags & ZIO_FLAG_SPECULATIVE)
4985 return;
4986
4987 /*
4988 * If this is an I/O error that is going to be retried, then ignore the
4989 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
4990 * hard errors, when in reality they can happen for any number of
4991 * innocuous reasons (bus resets, MPxIO link failure, etc).
4992 */
4993 if (zio->io_error == EIO &&
4994 !(zio->io_flags & ZIO_FLAG_IO_RETRY))
4995 return;
4996
4997 /*
4998 * Intent logs writes won't propagate their error to the root
4999 * I/O so don't mark these types of failures as pool-level
5000 * errors.
5001 */
5002 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE))
5003 return;
5004
5005 if (type == ZIO_TYPE_WRITE && txg != 0 &&
5006 (!(flags & ZIO_FLAG_IO_REPAIR) ||
5007 (flags & ZIO_FLAG_SCAN_THREAD) ||
5008 spa->spa_claiming)) {
5009 /*
5010 * This is either a normal write (not a repair), or it's
5011 * a repair induced by the scrub thread, or it's a repair
5012 * made by zil_claim() during spa_load() in the first txg.
5013 * In the normal case, we commit the DTL change in the same
5014 * txg as the block was born. In the scrub-induced repair
5015 * case, we know that scrubs run in first-pass syncing context,
5016 * so we commit the DTL change in spa_syncing_txg(spa).
5017 * In the zil_claim() case, we commit in spa_first_txg(spa).
5018 *
5019 * We currently do not make DTL entries for failed spontaneous
5020 * self-healing writes triggered by normal (non-scrubbing)
5021 * reads, because we have no transactional context in which to
5022 * do so -- and it's not clear that it'd be desirable anyway.
5023 */
5024 if (vd->vdev_ops->vdev_op_leaf) {
5025 uint64_t commit_txg = txg;
5026 if (flags & ZIO_FLAG_SCAN_THREAD) {
5027 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
5028 ASSERT(spa_sync_pass(spa) == 1);
5029 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1);
5030 commit_txg = spa_syncing_txg(spa);
5031 } else if (spa->spa_claiming) {
5032 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
5033 commit_txg = spa_first_txg(spa);
5034 }
5035 ASSERT(commit_txg >= spa_syncing_txg(spa));
5036 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1))
5037 return;
5038 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
5039 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1);
5040 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg);
5041 }
5042 if (vd != rvd)
5043 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1);
5044 }
5045 }
5046
5047 int64_t
vdev_deflated_space(vdev_t * vd,int64_t space)5048 vdev_deflated_space(vdev_t *vd, int64_t space)
5049 {
5050 ASSERT((space & (SPA_MINBLOCKSIZE-1)) == 0);
5051 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache);
5052
5053 return ((space >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio);
5054 }
5055
5056 /*
5057 * Update the in-core space usage stats for this vdev, its metaslab class,
5058 * and the root vdev.
5059 */
5060 void
vdev_space_update(vdev_t * vd,int64_t alloc_delta,int64_t defer_delta,int64_t space_delta)5061 vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta,
5062 int64_t space_delta)
5063 {
5064 (void) defer_delta;
5065 int64_t dspace_delta;
5066 spa_t *spa = vd->vdev_spa;
5067 vdev_t *rvd = spa->spa_root_vdev;
5068
5069 ASSERT(vd == vd->vdev_top);
5070
5071 /*
5072 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
5073 * factor. We must calculate this here and not at the root vdev
5074 * because the root vdev's psize-to-asize is simply the max of its
5075 * children's, thus not accurate enough for us.
5076 */
5077 dspace_delta = vdev_deflated_space(vd, space_delta);
5078
5079 mutex_enter(&vd->vdev_stat_lock);
5080 /* ensure we won't underflow */
5081 if (alloc_delta < 0) {
5082 ASSERT3U(vd->vdev_stat.vs_alloc, >=, -alloc_delta);
5083 }
5084
5085 vd->vdev_stat.vs_alloc += alloc_delta;
5086 vd->vdev_stat.vs_space += space_delta;
5087 vd->vdev_stat.vs_dspace += dspace_delta;
5088 mutex_exit(&vd->vdev_stat_lock);
5089
5090 /* every class but log contributes to root space stats */
5091 if (vd->vdev_mg != NULL && !vd->vdev_islog) {
5092 ASSERT(!vd->vdev_isl2cache);
5093 mutex_enter(&rvd->vdev_stat_lock);
5094 rvd->vdev_stat.vs_alloc += alloc_delta;
5095 rvd->vdev_stat.vs_space += space_delta;
5096 rvd->vdev_stat.vs_dspace += dspace_delta;
5097 mutex_exit(&rvd->vdev_stat_lock);
5098 }
5099 /* Note: metaslab_class_space_update moved to metaslab_space_update */
5100 }
5101
5102 /*
5103 * Mark a top-level vdev's config as dirty, placing it on the dirty list
5104 * so that it will be written out next time the vdev configuration is synced.
5105 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
5106 */
5107 void
vdev_config_dirty(vdev_t * vd)5108 vdev_config_dirty(vdev_t *vd)
5109 {
5110 spa_t *spa = vd->vdev_spa;
5111 vdev_t *rvd = spa->spa_root_vdev;
5112 int c;
5113
5114 ASSERT(spa_writeable(spa));
5115
5116 /*
5117 * If this is an aux vdev (as with l2cache and spare devices), then we
5118 * update the vdev config manually and set the sync flag.
5119 */
5120 if (vd->vdev_aux != NULL) {
5121 spa_aux_vdev_t *sav = vd->vdev_aux;
5122 nvlist_t **aux;
5123 uint_t naux;
5124
5125 for (c = 0; c < sav->sav_count; c++) {
5126 if (sav->sav_vdevs[c] == vd)
5127 break;
5128 }
5129
5130 if (c == sav->sav_count) {
5131 /*
5132 * We're being removed. There's nothing more to do.
5133 */
5134 ASSERT(sav->sav_sync == B_TRUE);
5135 return;
5136 }
5137
5138 sav->sav_sync = B_TRUE;
5139
5140 if (nvlist_lookup_nvlist_array(sav->sav_config,
5141 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) {
5142 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
5143 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0);
5144 }
5145
5146 ASSERT(c < naux);
5147
5148 /*
5149 * Setting the nvlist in the middle if the array is a little
5150 * sketchy, but it will work.
5151 */
5152 nvlist_free(aux[c]);
5153 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0);
5154
5155 return;
5156 }
5157
5158 /*
5159 * The dirty list is protected by the SCL_CONFIG lock. The caller
5160 * must either hold SCL_CONFIG as writer, or must be the sync thread
5161 * (which holds SCL_CONFIG as reader). There's only one sync thread,
5162 * so this is sufficient to ensure mutual exclusion.
5163 */
5164 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
5165 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
5166 spa_config_held(spa, SCL_CONFIG, RW_READER)));
5167
5168 if (vd == rvd) {
5169 for (c = 0; c < rvd->vdev_children; c++)
5170 vdev_config_dirty(rvd->vdev_child[c]);
5171 } else {
5172 ASSERT(vd == vd->vdev_top);
5173
5174 if (!list_link_active(&vd->vdev_config_dirty_node) &&
5175 vdev_is_concrete(vd)) {
5176 list_insert_head(&spa->spa_config_dirty_list, vd);
5177 }
5178 }
5179 }
5180
5181 void
vdev_config_clean(vdev_t * vd)5182 vdev_config_clean(vdev_t *vd)
5183 {
5184 spa_t *spa = vd->vdev_spa;
5185
5186 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
5187 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
5188 spa_config_held(spa, SCL_CONFIG, RW_READER)));
5189
5190 ASSERT(list_link_active(&vd->vdev_config_dirty_node));
5191 list_remove(&spa->spa_config_dirty_list, vd);
5192 }
5193
5194 /*
5195 * Mark a top-level vdev's state as dirty, so that the next pass of
5196 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
5197 * the state changes from larger config changes because they require
5198 * much less locking, and are often needed for administrative actions.
5199 */
5200 void
vdev_state_dirty(vdev_t * vd)5201 vdev_state_dirty(vdev_t *vd)
5202 {
5203 spa_t *spa = vd->vdev_spa;
5204
5205 ASSERT(spa_writeable(spa));
5206 ASSERT(vd == vd->vdev_top);
5207
5208 /*
5209 * The state list is protected by the SCL_STATE lock. The caller
5210 * must either hold SCL_STATE as writer, or must be the sync thread
5211 * (which holds SCL_STATE as reader). There's only one sync thread,
5212 * so this is sufficient to ensure mutual exclusion.
5213 */
5214 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
5215 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
5216 spa_config_held(spa, SCL_STATE, RW_READER)));
5217
5218 if (!list_link_active(&vd->vdev_state_dirty_node) &&
5219 vdev_is_concrete(vd))
5220 list_insert_head(&spa->spa_state_dirty_list, vd);
5221 }
5222
5223 void
vdev_state_clean(vdev_t * vd)5224 vdev_state_clean(vdev_t *vd)
5225 {
5226 spa_t *spa = vd->vdev_spa;
5227
5228 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
5229 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
5230 spa_config_held(spa, SCL_STATE, RW_READER)));
5231
5232 ASSERT(list_link_active(&vd->vdev_state_dirty_node));
5233 list_remove(&spa->spa_state_dirty_list, vd);
5234 }
5235
5236 /*
5237 * Propagate vdev state up from children to parent.
5238 */
5239 void
vdev_propagate_state(vdev_t * vd)5240 vdev_propagate_state(vdev_t *vd)
5241 {
5242 spa_t *spa = vd->vdev_spa;
5243 vdev_t *rvd = spa->spa_root_vdev;
5244 int degraded = 0, faulted = 0;
5245 int corrupted = 0;
5246 vdev_t *child;
5247
5248 if (vd->vdev_children > 0) {
5249 for (int c = 0; c < vd->vdev_children; c++) {
5250 child = vd->vdev_child[c];
5251
5252 /*
5253 * Don't factor holes or indirect vdevs into the
5254 * decision.
5255 */
5256 if (!vdev_is_concrete(child))
5257 continue;
5258
5259 if (!vdev_readable(child) ||
5260 (!vdev_writeable(child) && spa_writeable(spa))) {
5261 /*
5262 * Root special: if there is a top-level log
5263 * device, treat the root vdev as if it were
5264 * degraded.
5265 */
5266 if (child->vdev_islog && vd == rvd)
5267 degraded++;
5268 else
5269 faulted++;
5270 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) {
5271 degraded++;
5272 }
5273
5274 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA)
5275 corrupted++;
5276 }
5277
5278 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded);
5279
5280 /*
5281 * Root special: if there is a top-level vdev that cannot be
5282 * opened due to corrupted metadata, then propagate the root
5283 * vdev's aux state as 'corrupt' rather than 'insufficient
5284 * replicas'.
5285 */
5286 if (corrupted && vd == rvd &&
5287 rvd->vdev_state == VDEV_STATE_CANT_OPEN)
5288 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN,
5289 VDEV_AUX_CORRUPT_DATA);
5290 }
5291
5292 if (vd->vdev_parent)
5293 vdev_propagate_state(vd->vdev_parent);
5294 }
5295
5296 /*
5297 * Set a vdev's state. If this is during an open, we don't update the parent
5298 * state, because we're in the process of opening children depth-first.
5299 * Otherwise, we propagate the change to the parent.
5300 *
5301 * If this routine places a device in a faulted state, an appropriate ereport is
5302 * generated.
5303 */
5304 void
vdev_set_state(vdev_t * vd,boolean_t isopen,vdev_state_t state,vdev_aux_t aux)5305 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
5306 {
5307 uint64_t save_state;
5308 spa_t *spa = vd->vdev_spa;
5309
5310 if (state == vd->vdev_state) {
5311 /*
5312 * Since vdev_offline() code path is already in an offline
5313 * state we can miss a statechange event to OFFLINE. Check
5314 * the previous state to catch this condition.
5315 */
5316 if (vd->vdev_ops->vdev_op_leaf &&
5317 (state == VDEV_STATE_OFFLINE) &&
5318 (vd->vdev_prevstate >= VDEV_STATE_FAULTED)) {
5319 /* post an offline state change */
5320 zfs_post_state_change(spa, vd, vd->vdev_prevstate);
5321 }
5322 vd->vdev_stat.vs_aux = aux;
5323 return;
5324 }
5325
5326 save_state = vd->vdev_state;
5327
5328 vd->vdev_state = state;
5329 vd->vdev_stat.vs_aux = aux;
5330
5331 /*
5332 * If we are setting the vdev state to anything but an open state, then
5333 * always close the underlying device unless the device has requested
5334 * a delayed close (i.e. we're about to remove or fault the device).
5335 * Otherwise, we keep accessible but invalid devices open forever.
5336 * We don't call vdev_close() itself, because that implies some extra
5337 * checks (offline, etc) that we don't want here. This is limited to
5338 * leaf devices, because otherwise closing the device will affect other
5339 * children.
5340 */
5341 if (!vd->vdev_delayed_close && vdev_is_dead(vd) &&
5342 vd->vdev_ops->vdev_op_leaf)
5343 vd->vdev_ops->vdev_op_close(vd);
5344
5345 if (vd->vdev_removed &&
5346 state == VDEV_STATE_CANT_OPEN &&
5347 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) {
5348 /*
5349 * If the previous state is set to VDEV_STATE_REMOVED, then this
5350 * device was previously marked removed and someone attempted to
5351 * reopen it. If this failed due to a nonexistent device, then
5352 * keep the device in the REMOVED state. We also let this be if
5353 * it is one of our special test online cases, which is only
5354 * attempting to online the device and shouldn't generate an FMA
5355 * fault.
5356 */
5357 vd->vdev_state = VDEV_STATE_REMOVED;
5358 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
5359 } else if (state == VDEV_STATE_REMOVED) {
5360 vd->vdev_removed = B_TRUE;
5361 } else if (state == VDEV_STATE_CANT_OPEN) {
5362 /*
5363 * If we fail to open a vdev during an import or recovery, we
5364 * mark it as "not available", which signifies that it was
5365 * never there to begin with. Failure to open such a device
5366 * is not considered an error.
5367 */
5368 if ((spa_load_state(spa) == SPA_LOAD_IMPORT ||
5369 spa_load_state(spa) == SPA_LOAD_RECOVER) &&
5370 vd->vdev_ops->vdev_op_leaf)
5371 vd->vdev_not_present = 1;
5372
5373 /*
5374 * Post the appropriate ereport. If the 'prevstate' field is
5375 * set to something other than VDEV_STATE_UNKNOWN, it indicates
5376 * that this is part of a vdev_reopen(). In this case, we don't
5377 * want to post the ereport if the device was already in the
5378 * CANT_OPEN state beforehand.
5379 *
5380 * If the 'checkremove' flag is set, then this is an attempt to
5381 * online the device in response to an insertion event. If we
5382 * hit this case, then we have detected an insertion event for a
5383 * faulted or offline device that wasn't in the removed state.
5384 * In this scenario, we don't post an ereport because we are
5385 * about to replace the device, or attempt an online with
5386 * vdev_forcefault, which will generate the fault for us.
5387 */
5388 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) &&
5389 !vd->vdev_not_present && !vd->vdev_checkremove &&
5390 vd != spa->spa_root_vdev) {
5391 const char *class;
5392
5393 switch (aux) {
5394 case VDEV_AUX_OPEN_FAILED:
5395 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED;
5396 break;
5397 case VDEV_AUX_CORRUPT_DATA:
5398 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA;
5399 break;
5400 case VDEV_AUX_NO_REPLICAS:
5401 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS;
5402 break;
5403 case VDEV_AUX_BAD_GUID_SUM:
5404 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM;
5405 break;
5406 case VDEV_AUX_TOO_SMALL:
5407 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL;
5408 break;
5409 case VDEV_AUX_BAD_LABEL:
5410 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL;
5411 break;
5412 case VDEV_AUX_BAD_ASHIFT:
5413 class = FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT;
5414 break;
5415 default:
5416 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN;
5417 }
5418
5419 (void) zfs_ereport_post(class, spa, vd, NULL, NULL,
5420 save_state);
5421 }
5422
5423 /* Erase any notion of persistent removed state */
5424 vd->vdev_removed = B_FALSE;
5425 } else {
5426 vd->vdev_removed = B_FALSE;
5427 }
5428
5429 /*
5430 * Notify ZED of any significant state-change on a leaf vdev.
5431 *
5432 */
5433 if (vd->vdev_ops->vdev_op_leaf) {
5434 /* preserve original state from a vdev_reopen() */
5435 if ((vd->vdev_prevstate != VDEV_STATE_UNKNOWN) &&
5436 (vd->vdev_prevstate != vd->vdev_state) &&
5437 (save_state <= VDEV_STATE_CLOSED))
5438 save_state = vd->vdev_prevstate;
5439
5440 /* filter out state change due to initial vdev_open */
5441 if (save_state > VDEV_STATE_CLOSED)
5442 zfs_post_state_change(spa, vd, save_state);
5443 }
5444
5445 if (!isopen && vd->vdev_parent)
5446 vdev_propagate_state(vd->vdev_parent);
5447 }
5448
5449 boolean_t
vdev_children_are_offline(vdev_t * vd)5450 vdev_children_are_offline(vdev_t *vd)
5451 {
5452 ASSERT(!vd->vdev_ops->vdev_op_leaf);
5453
5454 for (uint64_t i = 0; i < vd->vdev_children; i++) {
5455 if (vd->vdev_child[i]->vdev_state != VDEV_STATE_OFFLINE)
5456 return (B_FALSE);
5457 }
5458
5459 return (B_TRUE);
5460 }
5461
5462 /*
5463 * Check the vdev configuration to ensure that it's capable of supporting
5464 * a root pool. We do not support partial configuration.
5465 */
5466 boolean_t
vdev_is_bootable(vdev_t * vd)5467 vdev_is_bootable(vdev_t *vd)
5468 {
5469 if (!vd->vdev_ops->vdev_op_leaf) {
5470 const char *vdev_type = vd->vdev_ops->vdev_op_type;
5471
5472 if (strcmp(vdev_type, VDEV_TYPE_MISSING) == 0)
5473 return (B_FALSE);
5474 }
5475
5476 for (int c = 0; c < vd->vdev_children; c++) {
5477 if (!vdev_is_bootable(vd->vdev_child[c]))
5478 return (B_FALSE);
5479 }
5480 return (B_TRUE);
5481 }
5482
5483 boolean_t
vdev_is_concrete(vdev_t * vd)5484 vdev_is_concrete(vdev_t *vd)
5485 {
5486 vdev_ops_t *ops = vd->vdev_ops;
5487 if (ops == &vdev_indirect_ops || ops == &vdev_hole_ops ||
5488 ops == &vdev_missing_ops || ops == &vdev_root_ops) {
5489 return (B_FALSE);
5490 } else {
5491 return (B_TRUE);
5492 }
5493 }
5494
5495 /*
5496 * Determine if a log device has valid content. If the vdev was
5497 * removed or faulted in the MOS config then we know that
5498 * the content on the log device has already been written to the pool.
5499 */
5500 boolean_t
vdev_log_state_valid(vdev_t * vd)5501 vdev_log_state_valid(vdev_t *vd)
5502 {
5503 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted &&
5504 !vd->vdev_removed)
5505 return (B_TRUE);
5506
5507 for (int c = 0; c < vd->vdev_children; c++)
5508 if (vdev_log_state_valid(vd->vdev_child[c]))
5509 return (B_TRUE);
5510
5511 return (B_FALSE);
5512 }
5513
5514 /*
5515 * Expand a vdev if possible.
5516 */
5517 void
vdev_expand(vdev_t * vd,uint64_t txg)5518 vdev_expand(vdev_t *vd, uint64_t txg)
5519 {
5520 ASSERT(vd->vdev_top == vd);
5521 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5522 ASSERT(vdev_is_concrete(vd));
5523
5524 vdev_set_deflate_ratio(vd);
5525
5526 if ((vd->vdev_spa->spa_raidz_expand == NULL ||
5527 vd->vdev_spa->spa_raidz_expand->vre_vdev_id != vd->vdev_id) &&
5528 (vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count &&
5529 vdev_is_concrete(vd)) {
5530 vdev_metaslab_group_create(vd);
5531 VERIFY(vdev_metaslab_init(vd, txg) == 0);
5532 vdev_config_dirty(vd);
5533 }
5534 }
5535
5536 /*
5537 * Split a vdev.
5538 */
5539 void
vdev_split(vdev_t * vd)5540 vdev_split(vdev_t *vd)
5541 {
5542 vdev_t *cvd, *pvd = vd->vdev_parent;
5543
5544 VERIFY3U(pvd->vdev_children, >, 1);
5545
5546 vdev_remove_child(pvd, vd);
5547 vdev_compact_children(pvd);
5548
5549 ASSERT3P(pvd->vdev_child, !=, NULL);
5550
5551 cvd = pvd->vdev_child[0];
5552 if (pvd->vdev_children == 1) {
5553 vdev_remove_parent(cvd);
5554 cvd->vdev_splitting = B_TRUE;
5555 }
5556 vdev_propagate_state(cvd);
5557 }
5558
5559 void
vdev_deadman(vdev_t * vd,const char * tag)5560 vdev_deadman(vdev_t *vd, const char *tag)
5561 {
5562 for (int c = 0; c < vd->vdev_children; c++) {
5563 vdev_t *cvd = vd->vdev_child[c];
5564
5565 vdev_deadman(cvd, tag);
5566 }
5567
5568 if (vd->vdev_ops->vdev_op_leaf) {
5569 vdev_queue_t *vq = &vd->vdev_queue;
5570
5571 mutex_enter(&vq->vq_lock);
5572 if (vq->vq_active > 0) {
5573 spa_t *spa = vd->vdev_spa;
5574 zio_t *fio;
5575 uint64_t delta;
5576
5577 zfs_dbgmsg("slow vdev: %s has %u active IOs",
5578 vd->vdev_path, vq->vq_active);
5579
5580 /*
5581 * Look at the head of all the pending queues,
5582 * if any I/O has been outstanding for longer than
5583 * the spa_deadman_synctime invoke the deadman logic.
5584 */
5585 fio = list_head(&vq->vq_active_list);
5586 delta = gethrtime() - fio->io_timestamp;
5587 if (delta > spa_deadman_synctime(spa))
5588 zio_deadman(fio, tag);
5589 }
5590 mutex_exit(&vq->vq_lock);
5591 }
5592 }
5593
5594 void
vdev_defer_resilver(vdev_t * vd)5595 vdev_defer_resilver(vdev_t *vd)
5596 {
5597 ASSERT(vd->vdev_ops->vdev_op_leaf);
5598
5599 vd->vdev_resilver_deferred = B_TRUE;
5600 vd->vdev_spa->spa_resilver_deferred = B_TRUE;
5601 }
5602
5603 /*
5604 * Clears the resilver deferred flag on all leaf devs under vd. Returns
5605 * B_TRUE if we have devices that need to be resilvered and are available to
5606 * accept resilver I/Os.
5607 */
5608 boolean_t
vdev_clear_resilver_deferred(vdev_t * vd,dmu_tx_t * tx)5609 vdev_clear_resilver_deferred(vdev_t *vd, dmu_tx_t *tx)
5610 {
5611 boolean_t resilver_needed = B_FALSE;
5612 spa_t *spa = vd->vdev_spa;
5613
5614 for (int c = 0; c < vd->vdev_children; c++) {
5615 vdev_t *cvd = vd->vdev_child[c];
5616 resilver_needed |= vdev_clear_resilver_deferred(cvd, tx);
5617 }
5618
5619 if (vd == spa->spa_root_vdev &&
5620 spa_feature_is_active(spa, SPA_FEATURE_RESILVER_DEFER)) {
5621 spa_feature_decr(spa, SPA_FEATURE_RESILVER_DEFER, tx);
5622 vdev_config_dirty(vd);
5623 spa->spa_resilver_deferred = B_FALSE;
5624 return (resilver_needed);
5625 }
5626
5627 if (!vdev_is_concrete(vd) || vd->vdev_aux ||
5628 !vd->vdev_ops->vdev_op_leaf)
5629 return (resilver_needed);
5630
5631 vd->vdev_resilver_deferred = B_FALSE;
5632
5633 return (!vdev_is_dead(vd) && !vd->vdev_offline &&
5634 vdev_resilver_needed(vd, NULL, NULL));
5635 }
5636
5637 boolean_t
vdev_xlate_is_empty(range_seg64_t * rs)5638 vdev_xlate_is_empty(range_seg64_t *rs)
5639 {
5640 return (rs->rs_start == rs->rs_end);
5641 }
5642
5643 /*
5644 * Translate a logical range to the first contiguous physical range for the
5645 * specified vdev_t. This function is initially called with a leaf vdev and
5646 * will walk each parent vdev until it reaches a top-level vdev. Once the
5647 * top-level is reached the physical range is initialized and the recursive
5648 * function begins to unwind. As it unwinds it calls the parent's vdev
5649 * specific translation function to do the real conversion.
5650 */
5651 void
vdev_xlate(vdev_t * vd,const range_seg64_t * logical_rs,range_seg64_t * physical_rs,range_seg64_t * remain_rs)5652 vdev_xlate(vdev_t *vd, const range_seg64_t *logical_rs,
5653 range_seg64_t *physical_rs, range_seg64_t *remain_rs)
5654 {
5655 /*
5656 * Walk up the vdev tree
5657 */
5658 if (vd != vd->vdev_top) {
5659 vdev_xlate(vd->vdev_parent, logical_rs, physical_rs,
5660 remain_rs);
5661 } else {
5662 /*
5663 * We've reached the top-level vdev, initialize the physical
5664 * range to the logical range and set an empty remaining
5665 * range then start to unwind.
5666 */
5667 physical_rs->rs_start = logical_rs->rs_start;
5668 physical_rs->rs_end = logical_rs->rs_end;
5669
5670 remain_rs->rs_start = logical_rs->rs_start;
5671 remain_rs->rs_end = logical_rs->rs_start;
5672
5673 return;
5674 }
5675
5676 vdev_t *pvd = vd->vdev_parent;
5677 ASSERT3P(pvd, !=, NULL);
5678 ASSERT3P(pvd->vdev_ops->vdev_op_xlate, !=, NULL);
5679
5680 /*
5681 * As this recursive function unwinds, translate the logical
5682 * range into its physical and any remaining components by calling
5683 * the vdev specific translate function.
5684 */
5685 range_seg64_t intermediate = { 0 };
5686 pvd->vdev_ops->vdev_op_xlate(vd, physical_rs, &intermediate, remain_rs);
5687
5688 physical_rs->rs_start = intermediate.rs_start;
5689 physical_rs->rs_end = intermediate.rs_end;
5690 }
5691
5692 void
vdev_xlate_walk(vdev_t * vd,const range_seg64_t * logical_rs,vdev_xlate_func_t * func,void * arg)5693 vdev_xlate_walk(vdev_t *vd, const range_seg64_t *logical_rs,
5694 vdev_xlate_func_t *func, void *arg)
5695 {
5696 range_seg64_t iter_rs = *logical_rs;
5697 range_seg64_t physical_rs;
5698 range_seg64_t remain_rs;
5699
5700 while (!vdev_xlate_is_empty(&iter_rs)) {
5701
5702 vdev_xlate(vd, &iter_rs, &physical_rs, &remain_rs);
5703
5704 /*
5705 * With raidz and dRAID, it's possible that the logical range
5706 * does not live on this leaf vdev. Only when there is a non-
5707 * zero physical size call the provided function.
5708 */
5709 if (!vdev_xlate_is_empty(&physical_rs))
5710 func(arg, &physical_rs);
5711
5712 iter_rs = remain_rs;
5713 }
5714 }
5715
5716 static char *
vdev_name(vdev_t * vd,char * buf,int buflen)5717 vdev_name(vdev_t *vd, char *buf, int buflen)
5718 {
5719 if (vd->vdev_path == NULL) {
5720 if (strcmp(vd->vdev_ops->vdev_op_type, "root") == 0) {
5721 strlcpy(buf, vd->vdev_spa->spa_name, buflen);
5722 } else if (!vd->vdev_ops->vdev_op_leaf) {
5723 snprintf(buf, buflen, "%s-%llu",
5724 vd->vdev_ops->vdev_op_type,
5725 (u_longlong_t)vd->vdev_id);
5726 }
5727 } else {
5728 strlcpy(buf, vd->vdev_path, buflen);
5729 }
5730 return (buf);
5731 }
5732
5733 /*
5734 * Look at the vdev tree and determine whether any devices are currently being
5735 * replaced.
5736 */
5737 boolean_t
vdev_replace_in_progress(vdev_t * vdev)5738 vdev_replace_in_progress(vdev_t *vdev)
5739 {
5740 ASSERT(spa_config_held(vdev->vdev_spa, SCL_ALL, RW_READER) != 0);
5741
5742 if (vdev->vdev_ops == &vdev_replacing_ops)
5743 return (B_TRUE);
5744
5745 /*
5746 * A 'spare' vdev indicates that we have a replace in progress, unless
5747 * it has exactly two children, and the second, the hot spare, has
5748 * finished being resilvered.
5749 */
5750 if (vdev->vdev_ops == &vdev_spare_ops && (vdev->vdev_children > 2 ||
5751 !vdev_dtl_empty(vdev->vdev_child[1], DTL_MISSING)))
5752 return (B_TRUE);
5753
5754 for (int i = 0; i < vdev->vdev_children; i++) {
5755 if (vdev_replace_in_progress(vdev->vdev_child[i]))
5756 return (B_TRUE);
5757 }
5758
5759 return (B_FALSE);
5760 }
5761
5762 /*
5763 * Add a (source=src, propname=propval) list to an nvlist.
5764 */
5765 static void
vdev_prop_add_list(nvlist_t * nvl,const char * propname,const char * strval,uint64_t intval,zprop_source_t src)5766 vdev_prop_add_list(nvlist_t *nvl, const char *propname, const char *strval,
5767 uint64_t intval, zprop_source_t src)
5768 {
5769 nvlist_t *propval;
5770
5771 propval = fnvlist_alloc();
5772 fnvlist_add_uint64(propval, ZPROP_SOURCE, src);
5773
5774 if (strval != NULL)
5775 fnvlist_add_string(propval, ZPROP_VALUE, strval);
5776 else
5777 fnvlist_add_uint64(propval, ZPROP_VALUE, intval);
5778
5779 fnvlist_add_nvlist(nvl, propname, propval);
5780 nvlist_free(propval);
5781 }
5782
5783 static void
vdev_props_set_sync(void * arg,dmu_tx_t * tx)5784 vdev_props_set_sync(void *arg, dmu_tx_t *tx)
5785 {
5786 vdev_t *vd;
5787 nvlist_t *nvp = arg;
5788 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
5789 objset_t *mos = spa->spa_meta_objset;
5790 nvpair_t *elem = NULL;
5791 uint64_t vdev_guid;
5792 uint64_t objid;
5793 nvlist_t *nvprops;
5794
5795 vdev_guid = fnvlist_lookup_uint64(nvp, ZPOOL_VDEV_PROPS_SET_VDEV);
5796 nvprops = fnvlist_lookup_nvlist(nvp, ZPOOL_VDEV_PROPS_SET_PROPS);
5797 vd = spa_lookup_by_guid(spa, vdev_guid, B_TRUE);
5798
5799 /* this vdev could get removed while waiting for this sync task */
5800 if (vd == NULL)
5801 return;
5802
5803 /*
5804 * Set vdev property values in the vdev props mos object.
5805 */
5806 if (vd->vdev_root_zap != 0) {
5807 objid = vd->vdev_root_zap;
5808 } else if (vd->vdev_top_zap != 0) {
5809 objid = vd->vdev_top_zap;
5810 } else if (vd->vdev_leaf_zap != 0) {
5811 objid = vd->vdev_leaf_zap;
5812 } else {
5813 panic("unexpected vdev type");
5814 }
5815
5816 mutex_enter(&spa->spa_props_lock);
5817
5818 while ((elem = nvlist_next_nvpair(nvprops, elem)) != NULL) {
5819 uint64_t intval;
5820 const char *strval;
5821 vdev_prop_t prop;
5822 const char *propname = nvpair_name(elem);
5823 zprop_type_t proptype;
5824
5825 switch (prop = vdev_name_to_prop(propname)) {
5826 case VDEV_PROP_USERPROP:
5827 if (vdev_prop_user(propname)) {
5828 strval = fnvpair_value_string(elem);
5829 if (strlen(strval) == 0) {
5830 /* remove the property if value == "" */
5831 (void) zap_remove(mos, objid, propname,
5832 tx);
5833 } else {
5834 VERIFY0(zap_update(mos, objid, propname,
5835 1, strlen(strval) + 1, strval, tx));
5836 }
5837 spa_history_log_internal(spa, "vdev set", tx,
5838 "vdev_guid=%llu: %s=%s",
5839 (u_longlong_t)vdev_guid, nvpair_name(elem),
5840 strval);
5841 }
5842 break;
5843 default:
5844 /* normalize the property name */
5845 propname = vdev_prop_to_name(prop);
5846 proptype = vdev_prop_get_type(prop);
5847
5848 if (nvpair_type(elem) == DATA_TYPE_STRING) {
5849 ASSERT(proptype == PROP_TYPE_STRING);
5850 strval = fnvpair_value_string(elem);
5851 VERIFY0(zap_update(mos, objid, propname,
5852 1, strlen(strval) + 1, strval, tx));
5853 spa_history_log_internal(spa, "vdev set", tx,
5854 "vdev_guid=%llu: %s=%s",
5855 (u_longlong_t)vdev_guid, nvpair_name(elem),
5856 strval);
5857 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
5858 intval = fnvpair_value_uint64(elem);
5859
5860 if (proptype == PROP_TYPE_INDEX) {
5861 const char *unused;
5862 VERIFY0(vdev_prop_index_to_string(
5863 prop, intval, &unused));
5864 }
5865 VERIFY0(zap_update(mos, objid, propname,
5866 sizeof (uint64_t), 1, &intval, tx));
5867 spa_history_log_internal(spa, "vdev set", tx,
5868 "vdev_guid=%llu: %s=%lld",
5869 (u_longlong_t)vdev_guid,
5870 nvpair_name(elem), (longlong_t)intval);
5871 } else {
5872 panic("invalid vdev property type %u",
5873 nvpair_type(elem));
5874 }
5875 }
5876
5877 }
5878
5879 mutex_exit(&spa->spa_props_lock);
5880 }
5881
5882 int
vdev_prop_set(vdev_t * vd,nvlist_t * innvl,nvlist_t * outnvl)5883 vdev_prop_set(vdev_t *vd, nvlist_t *innvl, nvlist_t *outnvl)
5884 {
5885 spa_t *spa = vd->vdev_spa;
5886 nvpair_t *elem = NULL;
5887 uint64_t vdev_guid;
5888 nvlist_t *nvprops;
5889 int error = 0;
5890
5891 ASSERT(vd != NULL);
5892
5893 /* Check that vdev has a zap we can use */
5894 if (vd->vdev_root_zap == 0 &&
5895 vd->vdev_top_zap == 0 &&
5896 vd->vdev_leaf_zap == 0)
5897 return (SET_ERROR(EINVAL));
5898
5899 if (nvlist_lookup_uint64(innvl, ZPOOL_VDEV_PROPS_SET_VDEV,
5900 &vdev_guid) != 0)
5901 return (SET_ERROR(EINVAL));
5902
5903 if (nvlist_lookup_nvlist(innvl, ZPOOL_VDEV_PROPS_SET_PROPS,
5904 &nvprops) != 0)
5905 return (SET_ERROR(EINVAL));
5906
5907 if ((vd = spa_lookup_by_guid(spa, vdev_guid, B_TRUE)) == NULL)
5908 return (SET_ERROR(EINVAL));
5909
5910 while ((elem = nvlist_next_nvpair(nvprops, elem)) != NULL) {
5911 const char *propname = nvpair_name(elem);
5912 vdev_prop_t prop = vdev_name_to_prop(propname);
5913 uint64_t intval = 0;
5914 const char *strval = NULL;
5915
5916 if (prop == VDEV_PROP_USERPROP && !vdev_prop_user(propname)) {
5917 error = EINVAL;
5918 goto end;
5919 }
5920
5921 if (vdev_prop_readonly(prop)) {
5922 error = EROFS;
5923 goto end;
5924 }
5925
5926 /* Special Processing */
5927 switch (prop) {
5928 case VDEV_PROP_PATH:
5929 if (vd->vdev_path == NULL) {
5930 error = EROFS;
5931 break;
5932 }
5933 if (nvpair_value_string(elem, &strval) != 0) {
5934 error = EINVAL;
5935 break;
5936 }
5937 /* New path must start with /dev/ */
5938 if (strncmp(strval, "/dev/", 5)) {
5939 error = EINVAL;
5940 break;
5941 }
5942 error = spa_vdev_setpath(spa, vdev_guid, strval);
5943 break;
5944 case VDEV_PROP_ALLOCATING:
5945 if (nvpair_value_uint64(elem, &intval) != 0) {
5946 error = EINVAL;
5947 break;
5948 }
5949 if (intval != vd->vdev_noalloc)
5950 break;
5951 if (intval == 0)
5952 error = spa_vdev_noalloc(spa, vdev_guid);
5953 else
5954 error = spa_vdev_alloc(spa, vdev_guid);
5955 break;
5956 case VDEV_PROP_FAILFAST:
5957 if (nvpair_value_uint64(elem, &intval) != 0) {
5958 error = EINVAL;
5959 break;
5960 }
5961 vd->vdev_failfast = intval & 1;
5962 break;
5963 case VDEV_PROP_CHECKSUM_N:
5964 if (nvpair_value_uint64(elem, &intval) != 0) {
5965 error = EINVAL;
5966 break;
5967 }
5968 vd->vdev_checksum_n = intval;
5969 break;
5970 case VDEV_PROP_CHECKSUM_T:
5971 if (nvpair_value_uint64(elem, &intval) != 0) {
5972 error = EINVAL;
5973 break;
5974 }
5975 vd->vdev_checksum_t = intval;
5976 break;
5977 case VDEV_PROP_IO_N:
5978 if (nvpair_value_uint64(elem, &intval) != 0) {
5979 error = EINVAL;
5980 break;
5981 }
5982 vd->vdev_io_n = intval;
5983 break;
5984 case VDEV_PROP_IO_T:
5985 if (nvpair_value_uint64(elem, &intval) != 0) {
5986 error = EINVAL;
5987 break;
5988 }
5989 vd->vdev_io_t = intval;
5990 break;
5991 case VDEV_PROP_SLOW_IO_N:
5992 if (nvpair_value_uint64(elem, &intval) != 0) {
5993 error = EINVAL;
5994 break;
5995 }
5996 vd->vdev_slow_io_n = intval;
5997 break;
5998 case VDEV_PROP_SLOW_IO_T:
5999 if (nvpair_value_uint64(elem, &intval) != 0) {
6000 error = EINVAL;
6001 break;
6002 }
6003 vd->vdev_slow_io_t = intval;
6004 break;
6005 default:
6006 /* Most processing is done in vdev_props_set_sync */
6007 break;
6008 }
6009 end:
6010 if (error != 0) {
6011 intval = error;
6012 vdev_prop_add_list(outnvl, propname, strval, intval, 0);
6013 return (error);
6014 }
6015 }
6016
6017 return (dsl_sync_task(spa->spa_name, NULL, vdev_props_set_sync,
6018 innvl, 6, ZFS_SPACE_CHECK_EXTRA_RESERVED));
6019 }
6020
6021 int
vdev_prop_get(vdev_t * vd,nvlist_t * innvl,nvlist_t * outnvl)6022 vdev_prop_get(vdev_t *vd, nvlist_t *innvl, nvlist_t *outnvl)
6023 {
6024 spa_t *spa = vd->vdev_spa;
6025 objset_t *mos = spa->spa_meta_objset;
6026 int err = 0;
6027 uint64_t objid;
6028 uint64_t vdev_guid;
6029 nvpair_t *elem = NULL;
6030 nvlist_t *nvprops = NULL;
6031 uint64_t intval = 0;
6032 char *strval = NULL;
6033 const char *propname = NULL;
6034 vdev_prop_t prop;
6035
6036 ASSERT(vd != NULL);
6037 ASSERT(mos != NULL);
6038
6039 if (nvlist_lookup_uint64(innvl, ZPOOL_VDEV_PROPS_GET_VDEV,
6040 &vdev_guid) != 0)
6041 return (SET_ERROR(EINVAL));
6042
6043 nvlist_lookup_nvlist(innvl, ZPOOL_VDEV_PROPS_GET_PROPS, &nvprops);
6044
6045 if (vd->vdev_root_zap != 0) {
6046 objid = vd->vdev_root_zap;
6047 } else if (vd->vdev_top_zap != 0) {
6048 objid = vd->vdev_top_zap;
6049 } else if (vd->vdev_leaf_zap != 0) {
6050 objid = vd->vdev_leaf_zap;
6051 } else {
6052 return (SET_ERROR(EINVAL));
6053 }
6054 ASSERT(objid != 0);
6055
6056 mutex_enter(&spa->spa_props_lock);
6057
6058 if (nvprops != NULL) {
6059 char namebuf[64] = { 0 };
6060
6061 while ((elem = nvlist_next_nvpair(nvprops, elem)) != NULL) {
6062 intval = 0;
6063 strval = NULL;
6064 propname = nvpair_name(elem);
6065 prop = vdev_name_to_prop(propname);
6066 zprop_source_t src = ZPROP_SRC_DEFAULT;
6067 uint64_t integer_size, num_integers;
6068
6069 switch (prop) {
6070 /* Special Read-only Properties */
6071 case VDEV_PROP_NAME:
6072 strval = vdev_name(vd, namebuf,
6073 sizeof (namebuf));
6074 if (strval == NULL)
6075 continue;
6076 vdev_prop_add_list(outnvl, propname, strval, 0,
6077 ZPROP_SRC_NONE);
6078 continue;
6079 case VDEV_PROP_CAPACITY:
6080 /* percent used */
6081 intval = (vd->vdev_stat.vs_dspace == 0) ? 0 :
6082 (vd->vdev_stat.vs_alloc * 100 /
6083 vd->vdev_stat.vs_dspace);
6084 vdev_prop_add_list(outnvl, propname, NULL,
6085 intval, ZPROP_SRC_NONE);
6086 continue;
6087 case VDEV_PROP_STATE:
6088 vdev_prop_add_list(outnvl, propname, NULL,
6089 vd->vdev_state, ZPROP_SRC_NONE);
6090 continue;
6091 case VDEV_PROP_GUID:
6092 vdev_prop_add_list(outnvl, propname, NULL,
6093 vd->vdev_guid, ZPROP_SRC_NONE);
6094 continue;
6095 case VDEV_PROP_ASIZE:
6096 vdev_prop_add_list(outnvl, propname, NULL,
6097 vd->vdev_asize, ZPROP_SRC_NONE);
6098 continue;
6099 case VDEV_PROP_PSIZE:
6100 vdev_prop_add_list(outnvl, propname, NULL,
6101 vd->vdev_psize, ZPROP_SRC_NONE);
6102 continue;
6103 case VDEV_PROP_ASHIFT:
6104 vdev_prop_add_list(outnvl, propname, NULL,
6105 vd->vdev_ashift, ZPROP_SRC_NONE);
6106 continue;
6107 case VDEV_PROP_SIZE:
6108 vdev_prop_add_list(outnvl, propname, NULL,
6109 vd->vdev_stat.vs_dspace, ZPROP_SRC_NONE);
6110 continue;
6111 case VDEV_PROP_FREE:
6112 vdev_prop_add_list(outnvl, propname, NULL,
6113 vd->vdev_stat.vs_dspace -
6114 vd->vdev_stat.vs_alloc, ZPROP_SRC_NONE);
6115 continue;
6116 case VDEV_PROP_ALLOCATED:
6117 vdev_prop_add_list(outnvl, propname, NULL,
6118 vd->vdev_stat.vs_alloc, ZPROP_SRC_NONE);
6119 continue;
6120 case VDEV_PROP_EXPANDSZ:
6121 vdev_prop_add_list(outnvl, propname, NULL,
6122 vd->vdev_stat.vs_esize, ZPROP_SRC_NONE);
6123 continue;
6124 case VDEV_PROP_FRAGMENTATION:
6125 vdev_prop_add_list(outnvl, propname, NULL,
6126 vd->vdev_stat.vs_fragmentation,
6127 ZPROP_SRC_NONE);
6128 continue;
6129 case VDEV_PROP_PARITY:
6130 vdev_prop_add_list(outnvl, propname, NULL,
6131 vdev_get_nparity(vd), ZPROP_SRC_NONE);
6132 continue;
6133 case VDEV_PROP_PATH:
6134 if (vd->vdev_path == NULL)
6135 continue;
6136 vdev_prop_add_list(outnvl, propname,
6137 vd->vdev_path, 0, ZPROP_SRC_NONE);
6138 continue;
6139 case VDEV_PROP_DEVID:
6140 if (vd->vdev_devid == NULL)
6141 continue;
6142 vdev_prop_add_list(outnvl, propname,
6143 vd->vdev_devid, 0, ZPROP_SRC_NONE);
6144 continue;
6145 case VDEV_PROP_PHYS_PATH:
6146 if (vd->vdev_physpath == NULL)
6147 continue;
6148 vdev_prop_add_list(outnvl, propname,
6149 vd->vdev_physpath, 0, ZPROP_SRC_NONE);
6150 continue;
6151 case VDEV_PROP_ENC_PATH:
6152 if (vd->vdev_enc_sysfs_path == NULL)
6153 continue;
6154 vdev_prop_add_list(outnvl, propname,
6155 vd->vdev_enc_sysfs_path, 0, ZPROP_SRC_NONE);
6156 continue;
6157 case VDEV_PROP_FRU:
6158 if (vd->vdev_fru == NULL)
6159 continue;
6160 vdev_prop_add_list(outnvl, propname,
6161 vd->vdev_fru, 0, ZPROP_SRC_NONE);
6162 continue;
6163 case VDEV_PROP_PARENT:
6164 if (vd->vdev_parent != NULL) {
6165 strval = vdev_name(vd->vdev_parent,
6166 namebuf, sizeof (namebuf));
6167 vdev_prop_add_list(outnvl, propname,
6168 strval, 0, ZPROP_SRC_NONE);
6169 }
6170 continue;
6171 case VDEV_PROP_CHILDREN:
6172 if (vd->vdev_children > 0)
6173 strval = kmem_zalloc(ZAP_MAXVALUELEN,
6174 KM_SLEEP);
6175 for (uint64_t i = 0; i < vd->vdev_children;
6176 i++) {
6177 const char *vname;
6178
6179 vname = vdev_name(vd->vdev_child[i],
6180 namebuf, sizeof (namebuf));
6181 if (vname == NULL)
6182 vname = "(unknown)";
6183 if (strlen(strval) > 0)
6184 strlcat(strval, ",",
6185 ZAP_MAXVALUELEN);
6186 strlcat(strval, vname, ZAP_MAXVALUELEN);
6187 }
6188 if (strval != NULL) {
6189 vdev_prop_add_list(outnvl, propname,
6190 strval, 0, ZPROP_SRC_NONE);
6191 kmem_free(strval, ZAP_MAXVALUELEN);
6192 }
6193 continue;
6194 case VDEV_PROP_NUMCHILDREN:
6195 vdev_prop_add_list(outnvl, propname, NULL,
6196 vd->vdev_children, ZPROP_SRC_NONE);
6197 continue;
6198 case VDEV_PROP_READ_ERRORS:
6199 vdev_prop_add_list(outnvl, propname, NULL,
6200 vd->vdev_stat.vs_read_errors,
6201 ZPROP_SRC_NONE);
6202 continue;
6203 case VDEV_PROP_WRITE_ERRORS:
6204 vdev_prop_add_list(outnvl, propname, NULL,
6205 vd->vdev_stat.vs_write_errors,
6206 ZPROP_SRC_NONE);
6207 continue;
6208 case VDEV_PROP_CHECKSUM_ERRORS:
6209 vdev_prop_add_list(outnvl, propname, NULL,
6210 vd->vdev_stat.vs_checksum_errors,
6211 ZPROP_SRC_NONE);
6212 continue;
6213 case VDEV_PROP_INITIALIZE_ERRORS:
6214 vdev_prop_add_list(outnvl, propname, NULL,
6215 vd->vdev_stat.vs_initialize_errors,
6216 ZPROP_SRC_NONE);
6217 continue;
6218 case VDEV_PROP_OPS_NULL:
6219 vdev_prop_add_list(outnvl, propname, NULL,
6220 vd->vdev_stat.vs_ops[ZIO_TYPE_NULL],
6221 ZPROP_SRC_NONE);
6222 continue;
6223 case VDEV_PROP_OPS_READ:
6224 vdev_prop_add_list(outnvl, propname, NULL,
6225 vd->vdev_stat.vs_ops[ZIO_TYPE_READ],
6226 ZPROP_SRC_NONE);
6227 continue;
6228 case VDEV_PROP_OPS_WRITE:
6229 vdev_prop_add_list(outnvl, propname, NULL,
6230 vd->vdev_stat.vs_ops[ZIO_TYPE_WRITE],
6231 ZPROP_SRC_NONE);
6232 continue;
6233 case VDEV_PROP_OPS_FREE:
6234 vdev_prop_add_list(outnvl, propname, NULL,
6235 vd->vdev_stat.vs_ops[ZIO_TYPE_FREE],
6236 ZPROP_SRC_NONE);
6237 continue;
6238 case VDEV_PROP_OPS_CLAIM:
6239 vdev_prop_add_list(outnvl, propname, NULL,
6240 vd->vdev_stat.vs_ops[ZIO_TYPE_CLAIM],
6241 ZPROP_SRC_NONE);
6242 continue;
6243 case VDEV_PROP_OPS_TRIM:
6244 /*
6245 * TRIM ops and bytes are reported to user
6246 * space as ZIO_TYPE_FLUSH. This is done to
6247 * preserve the vdev_stat_t structure layout
6248 * for user space.
6249 */
6250 vdev_prop_add_list(outnvl, propname, NULL,
6251 vd->vdev_stat.vs_ops[ZIO_TYPE_FLUSH],
6252 ZPROP_SRC_NONE);
6253 continue;
6254 case VDEV_PROP_BYTES_NULL:
6255 vdev_prop_add_list(outnvl, propname, NULL,
6256 vd->vdev_stat.vs_bytes[ZIO_TYPE_NULL],
6257 ZPROP_SRC_NONE);
6258 continue;
6259 case VDEV_PROP_BYTES_READ:
6260 vdev_prop_add_list(outnvl, propname, NULL,
6261 vd->vdev_stat.vs_bytes[ZIO_TYPE_READ],
6262 ZPROP_SRC_NONE);
6263 continue;
6264 case VDEV_PROP_BYTES_WRITE:
6265 vdev_prop_add_list(outnvl, propname, NULL,
6266 vd->vdev_stat.vs_bytes[ZIO_TYPE_WRITE],
6267 ZPROP_SRC_NONE);
6268 continue;
6269 case VDEV_PROP_BYTES_FREE:
6270 vdev_prop_add_list(outnvl, propname, NULL,
6271 vd->vdev_stat.vs_bytes[ZIO_TYPE_FREE],
6272 ZPROP_SRC_NONE);
6273 continue;
6274 case VDEV_PROP_BYTES_CLAIM:
6275 vdev_prop_add_list(outnvl, propname, NULL,
6276 vd->vdev_stat.vs_bytes[ZIO_TYPE_CLAIM],
6277 ZPROP_SRC_NONE);
6278 continue;
6279 case VDEV_PROP_BYTES_TRIM:
6280 /*
6281 * TRIM ops and bytes are reported to user
6282 * space as ZIO_TYPE_FLUSH. This is done to
6283 * preserve the vdev_stat_t structure layout
6284 * for user space.
6285 */
6286 vdev_prop_add_list(outnvl, propname, NULL,
6287 vd->vdev_stat.vs_bytes[ZIO_TYPE_FLUSH],
6288 ZPROP_SRC_NONE);
6289 continue;
6290 case VDEV_PROP_REMOVING:
6291 vdev_prop_add_list(outnvl, propname, NULL,
6292 vd->vdev_removing, ZPROP_SRC_NONE);
6293 continue;
6294 case VDEV_PROP_RAIDZ_EXPANDING:
6295 /* Only expose this for raidz */
6296 if (vd->vdev_ops == &vdev_raidz_ops) {
6297 vdev_prop_add_list(outnvl, propname,
6298 NULL, vd->vdev_rz_expanding,
6299 ZPROP_SRC_NONE);
6300 }
6301 continue;
6302 /* Numeric Properites */
6303 case VDEV_PROP_ALLOCATING:
6304 /* Leaf vdevs cannot have this property */
6305 if (vd->vdev_mg == NULL &&
6306 vd->vdev_top != NULL) {
6307 src = ZPROP_SRC_NONE;
6308 intval = ZPROP_BOOLEAN_NA;
6309 } else {
6310 err = vdev_prop_get_int(vd, prop,
6311 &intval);
6312 if (err && err != ENOENT)
6313 break;
6314
6315 if (intval ==
6316 vdev_prop_default_numeric(prop))
6317 src = ZPROP_SRC_DEFAULT;
6318 else
6319 src = ZPROP_SRC_LOCAL;
6320 }
6321
6322 vdev_prop_add_list(outnvl, propname, NULL,
6323 intval, src);
6324 break;
6325 case VDEV_PROP_FAILFAST:
6326 src = ZPROP_SRC_LOCAL;
6327 strval = NULL;
6328
6329 err = zap_lookup(mos, objid, nvpair_name(elem),
6330 sizeof (uint64_t), 1, &intval);
6331 if (err == ENOENT) {
6332 intval = vdev_prop_default_numeric(
6333 prop);
6334 err = 0;
6335 } else if (err) {
6336 break;
6337 }
6338 if (intval == vdev_prop_default_numeric(prop))
6339 src = ZPROP_SRC_DEFAULT;
6340
6341 vdev_prop_add_list(outnvl, propname, strval,
6342 intval, src);
6343 break;
6344 case VDEV_PROP_CHECKSUM_N:
6345 case VDEV_PROP_CHECKSUM_T:
6346 case VDEV_PROP_IO_N:
6347 case VDEV_PROP_IO_T:
6348 case VDEV_PROP_SLOW_IO_N:
6349 case VDEV_PROP_SLOW_IO_T:
6350 err = vdev_prop_get_int(vd, prop, &intval);
6351 if (err && err != ENOENT)
6352 break;
6353
6354 if (intval == vdev_prop_default_numeric(prop))
6355 src = ZPROP_SRC_DEFAULT;
6356 else
6357 src = ZPROP_SRC_LOCAL;
6358
6359 vdev_prop_add_list(outnvl, propname, NULL,
6360 intval, src);
6361 break;
6362 /* Text Properties */
6363 case VDEV_PROP_COMMENT:
6364 /* Exists in the ZAP below */
6365 /* FALLTHRU */
6366 case VDEV_PROP_USERPROP:
6367 /* User Properites */
6368 src = ZPROP_SRC_LOCAL;
6369
6370 err = zap_length(mos, objid, nvpair_name(elem),
6371 &integer_size, &num_integers);
6372 if (err)
6373 break;
6374
6375 switch (integer_size) {
6376 case 8:
6377 /* User properties cannot be integers */
6378 err = EINVAL;
6379 break;
6380 case 1:
6381 /* string property */
6382 strval = kmem_alloc(num_integers,
6383 KM_SLEEP);
6384 err = zap_lookup(mos, objid,
6385 nvpair_name(elem), 1,
6386 num_integers, strval);
6387 if (err) {
6388 kmem_free(strval,
6389 num_integers);
6390 break;
6391 }
6392 vdev_prop_add_list(outnvl, propname,
6393 strval, 0, src);
6394 kmem_free(strval, num_integers);
6395 break;
6396 }
6397 break;
6398 default:
6399 err = ENOENT;
6400 break;
6401 }
6402 if (err)
6403 break;
6404 }
6405 } else {
6406 /*
6407 * Get all properties from the MOS vdev property object.
6408 */
6409 zap_cursor_t zc;
6410 zap_attribute_t za;
6411 for (zap_cursor_init(&zc, mos, objid);
6412 (err = zap_cursor_retrieve(&zc, &za)) == 0;
6413 zap_cursor_advance(&zc)) {
6414 intval = 0;
6415 strval = NULL;
6416 zprop_source_t src = ZPROP_SRC_DEFAULT;
6417 propname = za.za_name;
6418
6419 switch (za.za_integer_length) {
6420 case 8:
6421 /* We do not allow integer user properties */
6422 /* This is likely an internal value */
6423 break;
6424 case 1:
6425 /* string property */
6426 strval = kmem_alloc(za.za_num_integers,
6427 KM_SLEEP);
6428 err = zap_lookup(mos, objid, za.za_name, 1,
6429 za.za_num_integers, strval);
6430 if (err) {
6431 kmem_free(strval, za.za_num_integers);
6432 break;
6433 }
6434 vdev_prop_add_list(outnvl, propname, strval, 0,
6435 src);
6436 kmem_free(strval, za.za_num_integers);
6437 break;
6438
6439 default:
6440 break;
6441 }
6442 }
6443 zap_cursor_fini(&zc);
6444 }
6445
6446 mutex_exit(&spa->spa_props_lock);
6447 if (err && err != ENOENT) {
6448 return (err);
6449 }
6450
6451 return (0);
6452 }
6453
6454 EXPORT_SYMBOL(vdev_fault);
6455 EXPORT_SYMBOL(vdev_degrade);
6456 EXPORT_SYMBOL(vdev_online);
6457 EXPORT_SYMBOL(vdev_offline);
6458 EXPORT_SYMBOL(vdev_clear);
6459
6460 ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, default_ms_count, UINT, ZMOD_RW,
6461 "Target number of metaslabs per top-level vdev");
6462
6463 ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, default_ms_shift, UINT, ZMOD_RW,
6464 "Default lower limit for metaslab size");
6465
6466 ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, max_ms_shift, UINT, ZMOD_RW,
6467 "Default upper limit for metaslab size");
6468
6469 ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, min_ms_count, UINT, ZMOD_RW,
6470 "Minimum number of metaslabs per top-level vdev");
6471
6472 ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, ms_count_limit, UINT, ZMOD_RW,
6473 "Practical upper limit of total metaslabs per top-level vdev");
6474
6475 ZFS_MODULE_PARAM(zfs, zfs_, slow_io_events_per_second, UINT, ZMOD_RW,
6476 "Rate limit slow IO (delay) events to this many per second");
6477
6478 /* BEGIN CSTYLED */
6479 ZFS_MODULE_PARAM(zfs, zfs_, checksum_events_per_second, UINT, ZMOD_RW,
6480 "Rate limit checksum events to this many checksum errors per second "
6481 "(do not set below ZED threshold).");
6482 /* END CSTYLED */
6483
6484 ZFS_MODULE_PARAM(zfs, zfs_, scan_ignore_errors, INT, ZMOD_RW,
6485 "Ignore errors during resilver/scrub");
6486
6487 ZFS_MODULE_PARAM(zfs_vdev, vdev_, validate_skip, INT, ZMOD_RW,
6488 "Bypass vdev_validate()");
6489
6490 ZFS_MODULE_PARAM(zfs, zfs_, nocacheflush, INT, ZMOD_RW,
6491 "Disable cache flushes");
6492
6493 ZFS_MODULE_PARAM(zfs, zfs_, embedded_slog_min_ms, UINT, ZMOD_RW,
6494 "Minimum number of metaslabs required to dedicate one for log blocks");
6495
6496 /* BEGIN CSTYLED */
6497 ZFS_MODULE_PARAM_CALL(zfs_vdev, zfs_vdev_, min_auto_ashift,
6498 param_set_min_auto_ashift, param_get_uint, ZMOD_RW,
6499 "Minimum ashift used when creating new top-level vdevs");
6500
6501 ZFS_MODULE_PARAM_CALL(zfs_vdev, zfs_vdev_, max_auto_ashift,
6502 param_set_max_auto_ashift, param_get_uint, ZMOD_RW,
6503 "Maximum ashift used when optimizing for logical -> physical sector "
6504 "size on new top-level vdevs");
6505 /* END CSTYLED */
6506