1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22 /*
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2011, 2018 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 */
31
32 #include <sys/zfs_context.h>
33 #include <sys/fm/fs/zfs.h>
34 #include <sys/spa.h>
35 #include <sys/spa_impl.h>
36 #include <sys/bpobj.h>
37 #include <sys/dmu.h>
38 #include <sys/dmu_tx.h>
39 #include <sys/dsl_dir.h>
40 #include <sys/vdev_impl.h>
41 #include <sys/uberblock_impl.h>
42 #include <sys/metaslab.h>
43 #include <sys/metaslab_impl.h>
44 #include <sys/space_map.h>
45 #include <sys/space_reftree.h>
46 #include <sys/zio.h>
47 #include <sys/zap.h>
48 #include <sys/fs/zfs.h>
49 #include <sys/arc.h>
50 #include <sys/zil.h>
51 #include <sys/dsl_scan.h>
52 #include <sys/abd.h>
53 #include <sys/vdev_initialize.h>
54
55 /*
56 * Virtual device management.
57 */
58
59 static vdev_ops_t *vdev_ops_table[] = {
60 &vdev_root_ops,
61 &vdev_raidz_ops,
62 &vdev_mirror_ops,
63 &vdev_replacing_ops,
64 &vdev_spare_ops,
65 &vdev_disk_ops,
66 &vdev_file_ops,
67 &vdev_missing_ops,
68 &vdev_hole_ops,
69 &vdev_indirect_ops,
70 NULL
71 };
72
73 /* maximum scrub/resilver I/O queue per leaf vdev */
74 int zfs_scrub_limit = 10;
75
76 /* default target for number of metaslabs per top-level vdev */
77 int zfs_vdev_default_ms_count = 200;
78
79 /* minimum number of metaslabs per top-level vdev */
80 int zfs_vdev_min_ms_count = 16;
81
82 /* practical upper limit of total metaslabs per top-level vdev */
83 int zfs_vdev_ms_count_limit = 1ULL << 17;
84
85 /* lower limit for metaslab size (512M) */
86 int zfs_vdev_default_ms_shift = 29;
87
88 /* upper limit for metaslab size (16G) */
89 int zfs_vdev_max_ms_shift = 34;
90
91 boolean_t vdev_validate_skip = B_FALSE;
92
93 /*
94 * Since the DTL space map of a vdev is not expected to have a lot of
95 * entries, we default its block size to 4K.
96 */
97 int vdev_dtl_sm_blksz = (1 << 12);
98
99 /*
100 * vdev-wide space maps that have lots of entries written to them at
101 * the end of each transaction can benefit from a higher I/O bandwidth
102 * (e.g. vdev_obsolete_sm), thus we default their block size to 128K.
103 */
104 int vdev_standard_sm_blksz = (1 << 17);
105
106 int zfs_ashift_min;
107
108 /*PRINTFLIKE2*/
109 void
110 vdev_dbgmsg(vdev_t *vd, const char *fmt, ...)
111 {
112 va_list adx;
113 char buf[256];
114
115 va_start(adx, fmt);
116 (void) vsnprintf(buf, sizeof (buf), fmt, adx);
117 va_end(adx);
118
119 if (vd->vdev_path != NULL) {
120 zfs_dbgmsg("%s vdev '%s': %s", vd->vdev_ops->vdev_op_type,
121 vd->vdev_path, buf);
122 } else {
123 zfs_dbgmsg("%s-%llu vdev (guid %llu): %s",
124 vd->vdev_ops->vdev_op_type,
125 (u_longlong_t)vd->vdev_id,
126 (u_longlong_t)vd->vdev_guid, buf);
127 }
128 }
129
130 void
131 vdev_dbgmsg_print_tree(vdev_t *vd, int indent)
132 {
133 char state[20];
134
135 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) {
136 zfs_dbgmsg("%*svdev %u: %s", indent, "", vd->vdev_id,
137 vd->vdev_ops->vdev_op_type);
138 return;
139 }
140
141 switch (vd->vdev_state) {
142 case VDEV_STATE_UNKNOWN:
143 (void) snprintf(state, sizeof (state), "unknown");
144 break;
145 case VDEV_STATE_CLOSED:
146 (void) snprintf(state, sizeof (state), "closed");
147 break;
148 case VDEV_STATE_OFFLINE:
149 (void) snprintf(state, sizeof (state), "offline");
150 break;
151 case VDEV_STATE_REMOVED:
152 (void) snprintf(state, sizeof (state), "removed");
153 break;
154 case VDEV_STATE_CANT_OPEN:
155 (void) snprintf(state, sizeof (state), "can't open");
156 break;
157 case VDEV_STATE_FAULTED:
158 (void) snprintf(state, sizeof (state), "faulted");
159 break;
160 case VDEV_STATE_DEGRADED:
161 (void) snprintf(state, sizeof (state), "degraded");
162 break;
163 case VDEV_STATE_HEALTHY:
164 (void) snprintf(state, sizeof (state), "healthy");
165 break;
166 default:
167 (void) snprintf(state, sizeof (state), "<state %u>",
168 (uint_t)vd->vdev_state);
169 }
170
171 zfs_dbgmsg("%*svdev %u: %s%s, guid: %llu, path: %s, %s", indent,
172 "", (int)vd->vdev_id, vd->vdev_ops->vdev_op_type,
173 vd->vdev_islog ? " (log)" : "",
174 (u_longlong_t)vd->vdev_guid,
175 vd->vdev_path ? vd->vdev_path : "N/A", state);
176
177 for (uint64_t i = 0; i < vd->vdev_children; i++)
178 vdev_dbgmsg_print_tree(vd->vdev_child[i], indent + 2);
179 }
180
181 /*
182 * Given a vdev type, return the appropriate ops vector.
183 */
184 static vdev_ops_t *
185 vdev_getops(const char *type)
186 {
187 vdev_ops_t *ops, **opspp;
188
189 for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++)
190 if (strcmp(ops->vdev_op_type, type) == 0)
191 break;
192
193 return (ops);
194 }
195
196 /*
197 * Derive the enumerated alloction bias from string input.
198 * String origin is either the per-vdev zap or zpool(1M).
199 */
200 static vdev_alloc_bias_t
201 vdev_derive_alloc_bias(const char *bias)
202 {
203 vdev_alloc_bias_t alloc_bias = VDEV_BIAS_NONE;
204
205 if (strcmp(bias, VDEV_ALLOC_BIAS_LOG) == 0)
206 alloc_bias = VDEV_BIAS_LOG;
207 else if (strcmp(bias, VDEV_ALLOC_BIAS_SPECIAL) == 0)
208 alloc_bias = VDEV_BIAS_SPECIAL;
209 else if (strcmp(bias, VDEV_ALLOC_BIAS_DEDUP) == 0)
210 alloc_bias = VDEV_BIAS_DEDUP;
211
212 return (alloc_bias);
213 }
214
215 /* ARGSUSED */
216 void
217 vdev_default_xlate(vdev_t *vd, const range_seg_t *in, range_seg_t *res)
218 {
219 res->rs_start = in->rs_start;
220 res->rs_end = in->rs_end;
221 }
222
223 /*
224 * Default asize function: return the MAX of psize with the asize of
225 * all children. This is what's used by anything other than RAID-Z.
226 */
227 uint64_t
228 vdev_default_asize(vdev_t *vd, uint64_t psize)
229 {
230 uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift);
231 uint64_t csize;
232
233 for (int c = 0; c < vd->vdev_children; c++) {
234 csize = vdev_psize_to_asize(vd->vdev_child[c], psize);
235 asize = MAX(asize, csize);
236 }
237
238 return (asize);
239 }
240
241 /*
242 * Get the minimum allocatable size. We define the allocatable size as
243 * the vdev's asize rounded to the nearest metaslab. This allows us to
244 * replace or attach devices which don't have the same physical size but
245 * can still satisfy the same number of allocations.
246 */
247 uint64_t
248 vdev_get_min_asize(vdev_t *vd)
249 {
250 vdev_t *pvd = vd->vdev_parent;
251
252 /*
253 * If our parent is NULL (inactive spare or cache) or is the root,
254 * just return our own asize.
255 */
256 if (pvd == NULL)
257 return (vd->vdev_asize);
258
259 /*
260 * The top-level vdev just returns the allocatable size rounded
261 * to the nearest metaslab.
262 */
263 if (vd == vd->vdev_top)
264 return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift));
265
266 /*
267 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
268 * so each child must provide at least 1/Nth of its asize.
269 */
270 if (pvd->vdev_ops == &vdev_raidz_ops)
271 return ((pvd->vdev_min_asize + pvd->vdev_children - 1) /
272 pvd->vdev_children);
273
274 return (pvd->vdev_min_asize);
275 }
276
277 void
278 vdev_set_min_asize(vdev_t *vd)
279 {
280 vd->vdev_min_asize = vdev_get_min_asize(vd);
281
282 for (int c = 0; c < vd->vdev_children; c++)
283 vdev_set_min_asize(vd->vdev_child[c]);
284 }
285
286 vdev_t *
287 vdev_lookup_top(spa_t *spa, uint64_t vdev)
288 {
289 vdev_t *rvd = spa->spa_root_vdev;
290
291 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
292
293 if (vdev < rvd->vdev_children) {
294 ASSERT(rvd->vdev_child[vdev] != NULL);
295 return (rvd->vdev_child[vdev]);
296 }
297
298 return (NULL);
299 }
300
301 vdev_t *
302 vdev_lookup_by_guid(vdev_t *vd, uint64_t guid)
303 {
304 vdev_t *mvd;
305
306 if (vd->vdev_guid == guid)
307 return (vd);
308
309 for (int c = 0; c < vd->vdev_children; c++)
310 if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) !=
311 NULL)
312 return (mvd);
313
314 return (NULL);
315 }
316
317 static int
318 vdev_count_leaves_impl(vdev_t *vd)
319 {
320 int n = 0;
321
322 if (vd->vdev_ops->vdev_op_leaf)
323 return (1);
324
325 for (int c = 0; c < vd->vdev_children; c++)
326 n += vdev_count_leaves_impl(vd->vdev_child[c]);
327
328 return (n);
329 }
330
331 int
332 vdev_count_leaves(spa_t *spa)
333 {
334 return (vdev_count_leaves_impl(spa->spa_root_vdev));
335 }
336
337 void
338 vdev_add_child(vdev_t *pvd, vdev_t *cvd)
339 {
340 size_t oldsize, newsize;
341 uint64_t id = cvd->vdev_id;
342 vdev_t **newchild;
343 spa_t *spa = cvd->vdev_spa;
344
345 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
346 ASSERT(cvd->vdev_parent == NULL);
347
348 cvd->vdev_parent = pvd;
349
350 if (pvd == NULL)
351 return;
352
353 ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL);
354
355 oldsize = pvd->vdev_children * sizeof (vdev_t *);
356 pvd->vdev_children = MAX(pvd->vdev_children, id + 1);
357 newsize = pvd->vdev_children * sizeof (vdev_t *);
358
359 newchild = kmem_zalloc(newsize, KM_SLEEP);
360 if (pvd->vdev_child != NULL) {
361 bcopy(pvd->vdev_child, newchild, oldsize);
362 kmem_free(pvd->vdev_child, oldsize);
363 }
364
365 pvd->vdev_child = newchild;
366 pvd->vdev_child[id] = cvd;
367
368 cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd);
369 ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL);
370
371 /*
372 * Walk up all ancestors to update guid sum.
373 */
374 for (; pvd != NULL; pvd = pvd->vdev_parent)
375 pvd->vdev_guid_sum += cvd->vdev_guid_sum;
376
377 if (cvd->vdev_ops->vdev_op_leaf) {
378 list_insert_head(&cvd->vdev_spa->spa_leaf_list, cvd);
379 cvd->vdev_spa->spa_leaf_list_gen++;
380 }
381 }
382
383 void
384 vdev_remove_child(vdev_t *pvd, vdev_t *cvd)
385 {
386 int c;
387 uint_t id = cvd->vdev_id;
388
389 ASSERT(cvd->vdev_parent == pvd);
390
391 if (pvd == NULL)
392 return;
393
394 ASSERT(id < pvd->vdev_children);
395 ASSERT(pvd->vdev_child[id] == cvd);
396
397 pvd->vdev_child[id] = NULL;
398 cvd->vdev_parent = NULL;
399
400 for (c = 0; c < pvd->vdev_children; c++)
401 if (pvd->vdev_child[c])
402 break;
403
404 if (c == pvd->vdev_children) {
405 kmem_free(pvd->vdev_child, c * sizeof (vdev_t *));
406 pvd->vdev_child = NULL;
407 pvd->vdev_children = 0;
408 }
409
410 if (cvd->vdev_ops->vdev_op_leaf) {
411 spa_t *spa = cvd->vdev_spa;
412 list_remove(&spa->spa_leaf_list, cvd);
413 spa->spa_leaf_list_gen++;
414 }
415
416 /*
417 * Walk up all ancestors to update guid sum.
418 */
419 for (; pvd != NULL; pvd = pvd->vdev_parent)
420 pvd->vdev_guid_sum -= cvd->vdev_guid_sum;
421 }
422
423 /*
424 * Remove any holes in the child array.
425 */
426 void
427 vdev_compact_children(vdev_t *pvd)
428 {
429 vdev_t **newchild, *cvd;
430 int oldc = pvd->vdev_children;
431 int newc;
432
433 ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
434
435 for (int c = newc = 0; c < oldc; c++)
436 if (pvd->vdev_child[c])
437 newc++;
438
439 newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP);
440
441 for (int c = newc = 0; c < oldc; c++) {
442 if ((cvd = pvd->vdev_child[c]) != NULL) {
443 newchild[newc] = cvd;
444 cvd->vdev_id = newc++;
445 }
446 }
447
448 kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *));
449 pvd->vdev_child = newchild;
450 pvd->vdev_children = newc;
451 }
452
453 /*
454 * Allocate and minimally initialize a vdev_t.
455 */
456 vdev_t *
457 vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops)
458 {
459 vdev_t *vd;
460 vdev_indirect_config_t *vic;
461
462 vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP);
463 vic = &vd->vdev_indirect_config;
464
465 if (spa->spa_root_vdev == NULL) {
466 ASSERT(ops == &vdev_root_ops);
467 spa->spa_root_vdev = vd;
468 spa->spa_load_guid = spa_generate_guid(NULL);
469 }
470
471 if (guid == 0 && ops != &vdev_hole_ops) {
472 if (spa->spa_root_vdev == vd) {
473 /*
474 * The root vdev's guid will also be the pool guid,
475 * which must be unique among all pools.
476 */
477 guid = spa_generate_guid(NULL);
478 } else {
479 /*
480 * Any other vdev's guid must be unique within the pool.
481 */
482 guid = spa_generate_guid(spa);
483 }
484 ASSERT(!spa_guid_exists(spa_guid(spa), guid));
485 }
486
487 vd->vdev_spa = spa;
488 vd->vdev_id = id;
489 vd->vdev_guid = guid;
490 vd->vdev_guid_sum = guid;
491 vd->vdev_ops = ops;
492 vd->vdev_state = VDEV_STATE_CLOSED;
493 vd->vdev_ishole = (ops == &vdev_hole_ops);
494 vic->vic_prev_indirect_vdev = UINT64_MAX;
495
496 rw_init(&vd->vdev_indirect_rwlock, NULL, RW_DEFAULT, NULL);
497 mutex_init(&vd->vdev_obsolete_lock, NULL, MUTEX_DEFAULT, NULL);
498 vd->vdev_obsolete_segments = range_tree_create(NULL, NULL);
499
500 list_link_init(&vd->vdev_leaf_node);
501 mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL);
502 mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL);
503 mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL);
504 mutex_init(&vd->vdev_queue_lock, NULL, MUTEX_DEFAULT, NULL);
505 mutex_init(&vd->vdev_initialize_lock, NULL, MUTEX_DEFAULT, NULL);
506 mutex_init(&vd->vdev_initialize_io_lock, NULL, MUTEX_DEFAULT, NULL);
507 cv_init(&vd->vdev_initialize_cv, NULL, CV_DEFAULT, NULL);
508 cv_init(&vd->vdev_initialize_io_cv, NULL, CV_DEFAULT, NULL);
509
510 for (int t = 0; t < DTL_TYPES; t++) {
511 vd->vdev_dtl[t] = range_tree_create(NULL, NULL);
512 }
513 txg_list_create(&vd->vdev_ms_list, spa,
514 offsetof(struct metaslab, ms_txg_node));
515 txg_list_create(&vd->vdev_dtl_list, spa,
516 offsetof(struct vdev, vdev_dtl_node));
517 vd->vdev_stat.vs_timestamp = gethrtime();
518 vdev_queue_init(vd);
519 vdev_cache_init(vd);
520
521 return (vd);
522 }
523
524 /*
525 * Allocate a new vdev. The 'alloctype' is used to control whether we are
526 * creating a new vdev or loading an existing one - the behavior is slightly
527 * different for each case.
528 */
529 int
530 vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
531 int alloctype)
532 {
533 vdev_ops_t *ops;
534 char *type;
535 uint64_t guid = 0, islog, nparity;
536 vdev_t *vd;
537 vdev_indirect_config_t *vic;
538 vdev_alloc_bias_t alloc_bias = VDEV_BIAS_NONE;
539 boolean_t top_level = (parent && !parent->vdev_parent);
540
541 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
542
543 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
544 return (SET_ERROR(EINVAL));
545
546 if ((ops = vdev_getops(type)) == NULL)
547 return (SET_ERROR(EINVAL));
548
549 /*
550 * If this is a load, get the vdev guid from the nvlist.
551 * Otherwise, vdev_alloc_common() will generate one for us.
552 */
553 if (alloctype == VDEV_ALLOC_LOAD) {
554 uint64_t label_id;
555
556 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) ||
557 label_id != id)
558 return (SET_ERROR(EINVAL));
559
560 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
561 return (SET_ERROR(EINVAL));
562 } else if (alloctype == VDEV_ALLOC_SPARE) {
563 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
564 return (SET_ERROR(EINVAL));
565 } else if (alloctype == VDEV_ALLOC_L2CACHE) {
566 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
567 return (SET_ERROR(EINVAL));
568 } else if (alloctype == VDEV_ALLOC_ROOTPOOL) {
569 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
570 return (SET_ERROR(EINVAL));
571 }
572
573 /*
574 * The first allocated vdev must be of type 'root'.
575 */
576 if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL)
577 return (SET_ERROR(EINVAL));
578
579 /*
580 * Determine whether we're a log vdev.
581 */
582 islog = 0;
583 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog);
584 if (islog && spa_version(spa) < SPA_VERSION_SLOGS)
585 return (SET_ERROR(ENOTSUP));
586
587 if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES)
588 return (SET_ERROR(ENOTSUP));
589
590 /*
591 * Set the nparity property for RAID-Z vdevs.
592 */
593 nparity = -1ULL;
594 if (ops == &vdev_raidz_ops) {
595 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
596 &nparity) == 0) {
597 if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY)
598 return (SET_ERROR(EINVAL));
599 /*
600 * Previous versions could only support 1 or 2 parity
601 * device.
602 */
603 if (nparity > 1 &&
604 spa_version(spa) < SPA_VERSION_RAIDZ2)
605 return (SET_ERROR(ENOTSUP));
606 if (nparity > 2 &&
607 spa_version(spa) < SPA_VERSION_RAIDZ3)
608 return (SET_ERROR(ENOTSUP));
609 } else {
610 /*
611 * We require the parity to be specified for SPAs that
612 * support multiple parity levels.
613 */
614 if (spa_version(spa) >= SPA_VERSION_RAIDZ2)
615 return (SET_ERROR(EINVAL));
616 /*
617 * Otherwise, we default to 1 parity device for RAID-Z.
618 */
619 nparity = 1;
620 }
621 } else {
622 nparity = 0;
623 }
624 ASSERT(nparity != -1ULL);
625
626 /*
627 * If creating a top-level vdev, check for allocation classes input
628 */
629 if (top_level && alloctype == VDEV_ALLOC_ADD) {
630 char *bias;
631
632 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_ALLOCATION_BIAS,
633 &bias) == 0) {
634 alloc_bias = vdev_derive_alloc_bias(bias);
635
636 /* spa_vdev_add() expects feature to be enabled */
637 if (spa->spa_load_state != SPA_LOAD_CREATE &&
638 !spa_feature_is_enabled(spa,
639 SPA_FEATURE_ALLOCATION_CLASSES)) {
640 return (SET_ERROR(ENOTSUP));
641 }
642 }
643 }
644
645 vd = vdev_alloc_common(spa, id, guid, ops);
646 vic = &vd->vdev_indirect_config;
647
648 vd->vdev_islog = islog;
649 vd->vdev_nparity = nparity;
650 if (top_level && alloc_bias != VDEV_BIAS_NONE)
651 vd->vdev_alloc_bias = alloc_bias;
652
653 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0)
654 vd->vdev_path = spa_strdup(vd->vdev_path);
655 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0)
656 vd->vdev_devid = spa_strdup(vd->vdev_devid);
657 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH,
658 &vd->vdev_physpath) == 0)
659 vd->vdev_physpath = spa_strdup(vd->vdev_physpath);
660 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0)
661 vd->vdev_fru = spa_strdup(vd->vdev_fru);
662
663 /*
664 * Set the whole_disk property. If it's not specified, leave the value
665 * as -1.
666 */
667 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
668 &vd->vdev_wholedisk) != 0)
669 vd->vdev_wholedisk = -1ULL;
670
671 ASSERT0(vic->vic_mapping_object);
672 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_OBJECT,
673 &vic->vic_mapping_object);
674 ASSERT0(vic->vic_births_object);
675 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_BIRTHS,
676 &vic->vic_births_object);
677 ASSERT3U(vic->vic_prev_indirect_vdev, ==, UINT64_MAX);
678 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_PREV_INDIRECT_VDEV,
679 &vic->vic_prev_indirect_vdev);
680
681 /*
682 * Look for the 'not present' flag. This will only be set if the device
683 * was not present at the time of import.
684 */
685 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
686 &vd->vdev_not_present);
687
688 /*
689 * Get the alignment requirement.
690 */
691 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift);
692
693 /*
694 * Retrieve the vdev creation time.
695 */
696 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG,
697 &vd->vdev_crtxg);
698
699 /*
700 * If we're a top-level vdev, try to load the allocation parameters.
701 */
702 if (top_level &&
703 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
704 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
705 &vd->vdev_ms_array);
706 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
707 &vd->vdev_ms_shift);
708 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE,
709 &vd->vdev_asize);
710 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING,
711 &vd->vdev_removing);
712 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP,
713 &vd->vdev_top_zap);
714 } else {
715 ASSERT0(vd->vdev_top_zap);
716 }
717
718 if (top_level && alloctype != VDEV_ALLOC_ATTACH) {
719 ASSERT(alloctype == VDEV_ALLOC_LOAD ||
720 alloctype == VDEV_ALLOC_ADD ||
721 alloctype == VDEV_ALLOC_SPLIT ||
722 alloctype == VDEV_ALLOC_ROOTPOOL);
723 /* Note: metaslab_group_create() is now deferred */
724 }
725
726 if (vd->vdev_ops->vdev_op_leaf &&
727 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
728 (void) nvlist_lookup_uint64(nv,
729 ZPOOL_CONFIG_VDEV_LEAF_ZAP, &vd->vdev_leaf_zap);
730 } else {
731 ASSERT0(vd->vdev_leaf_zap);
732 }
733
734 /*
735 * If we're a leaf vdev, try to load the DTL object and other state.
736 */
737
738 if (vd->vdev_ops->vdev_op_leaf &&
739 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE ||
740 alloctype == VDEV_ALLOC_ROOTPOOL)) {
741 if (alloctype == VDEV_ALLOC_LOAD) {
742 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL,
743 &vd->vdev_dtl_object);
744 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE,
745 &vd->vdev_unspare);
746 }
747
748 if (alloctype == VDEV_ALLOC_ROOTPOOL) {
749 uint64_t spare = 0;
750
751 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
752 &spare) == 0 && spare)
753 spa_spare_add(vd);
754 }
755
756 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE,
757 &vd->vdev_offline);
758
759 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
760 &vd->vdev_resilver_txg);
761
762 /*
763 * When importing a pool, we want to ignore the persistent fault
764 * state, as the diagnosis made on another system may not be
765 * valid in the current context. Local vdevs will
766 * remain in the faulted state.
767 */
768 if (spa_load_state(spa) == SPA_LOAD_OPEN) {
769 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED,
770 &vd->vdev_faulted);
771 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED,
772 &vd->vdev_degraded);
773 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED,
774 &vd->vdev_removed);
775
776 if (vd->vdev_faulted || vd->vdev_degraded) {
777 char *aux;
778
779 vd->vdev_label_aux =
780 VDEV_AUX_ERR_EXCEEDED;
781 if (nvlist_lookup_string(nv,
782 ZPOOL_CONFIG_AUX_STATE, &aux) == 0 &&
783 strcmp(aux, "external") == 0)
784 vd->vdev_label_aux = VDEV_AUX_EXTERNAL;
785 }
786 }
787 }
788
789 /*
790 * Add ourselves to the parent's list of children.
791 */
792 vdev_add_child(parent, vd);
793
794 *vdp = vd;
795
796 return (0);
797 }
798
799 void
800 vdev_free(vdev_t *vd)
801 {
802 spa_t *spa = vd->vdev_spa;
803 ASSERT3P(vd->vdev_initialize_thread, ==, NULL);
804
805 /*
806 * vdev_free() implies closing the vdev first. This is simpler than
807 * trying to ensure complicated semantics for all callers.
808 */
809 vdev_close(vd);
810
811 ASSERT(!list_link_active(&vd->vdev_config_dirty_node));
812 ASSERT(!list_link_active(&vd->vdev_state_dirty_node));
813
814 /*
815 * Free all children.
816 */
817 for (int c = 0; c < vd->vdev_children; c++)
818 vdev_free(vd->vdev_child[c]);
819
820 ASSERT(vd->vdev_child == NULL);
821 ASSERT(vd->vdev_guid_sum == vd->vdev_guid);
822 ASSERT(vd->vdev_initialize_thread == NULL);
823
824 /*
825 * Discard allocation state.
826 */
827 if (vd->vdev_mg != NULL) {
828 vdev_metaslab_fini(vd);
829 metaslab_group_destroy(vd->vdev_mg);
830 }
831
832 ASSERT0(vd->vdev_stat.vs_space);
833 ASSERT0(vd->vdev_stat.vs_dspace);
834 ASSERT0(vd->vdev_stat.vs_alloc);
835
836 /*
837 * Remove this vdev from its parent's child list.
838 */
839 vdev_remove_child(vd->vdev_parent, vd);
840
841 ASSERT(vd->vdev_parent == NULL);
842 ASSERT(!list_link_active(&vd->vdev_leaf_node));
843
844 /*
845 * Clean up vdev structure.
846 */
847 vdev_queue_fini(vd);
848 vdev_cache_fini(vd);
849
850 if (vd->vdev_path)
851 spa_strfree(vd->vdev_path);
852 if (vd->vdev_devid)
853 spa_strfree(vd->vdev_devid);
854 if (vd->vdev_physpath)
855 spa_strfree(vd->vdev_physpath);
856 if (vd->vdev_fru)
857 spa_strfree(vd->vdev_fru);
858
859 if (vd->vdev_isspare)
860 spa_spare_remove(vd);
861 if (vd->vdev_isl2cache)
862 spa_l2cache_remove(vd);
863
864 txg_list_destroy(&vd->vdev_ms_list);
865 txg_list_destroy(&vd->vdev_dtl_list);
866
867 mutex_enter(&vd->vdev_dtl_lock);
868 space_map_close(vd->vdev_dtl_sm);
869 for (int t = 0; t < DTL_TYPES; t++) {
870 range_tree_vacate(vd->vdev_dtl[t], NULL, NULL);
871 range_tree_destroy(vd->vdev_dtl[t]);
872 }
873 mutex_exit(&vd->vdev_dtl_lock);
874
875 EQUIV(vd->vdev_indirect_births != NULL,
876 vd->vdev_indirect_mapping != NULL);
877 if (vd->vdev_indirect_births != NULL) {
878 vdev_indirect_mapping_close(vd->vdev_indirect_mapping);
879 vdev_indirect_births_close(vd->vdev_indirect_births);
880 }
881
882 if (vd->vdev_obsolete_sm != NULL) {
883 ASSERT(vd->vdev_removing ||
884 vd->vdev_ops == &vdev_indirect_ops);
885 space_map_close(vd->vdev_obsolete_sm);
886 vd->vdev_obsolete_sm = NULL;
887 }
888 range_tree_destroy(vd->vdev_obsolete_segments);
889 rw_destroy(&vd->vdev_indirect_rwlock);
890 mutex_destroy(&vd->vdev_obsolete_lock);
891
892 mutex_destroy(&vd->vdev_queue_lock);
893 mutex_destroy(&vd->vdev_dtl_lock);
894 mutex_destroy(&vd->vdev_stat_lock);
895 mutex_destroy(&vd->vdev_probe_lock);
896 mutex_destroy(&vd->vdev_initialize_lock);
897 mutex_destroy(&vd->vdev_initialize_io_lock);
898 cv_destroy(&vd->vdev_initialize_io_cv);
899 cv_destroy(&vd->vdev_initialize_cv);
900
901 if (vd == spa->spa_root_vdev)
902 spa->spa_root_vdev = NULL;
903
904 kmem_free(vd, sizeof (vdev_t));
905 }
906
907 /*
908 * Transfer top-level vdev state from svd to tvd.
909 */
910 static void
911 vdev_top_transfer(vdev_t *svd, vdev_t *tvd)
912 {
913 spa_t *spa = svd->vdev_spa;
914 metaslab_t *msp;
915 vdev_t *vd;
916 int t;
917
918 ASSERT(tvd == tvd->vdev_top);
919
920 tvd->vdev_ms_array = svd->vdev_ms_array;
921 tvd->vdev_ms_shift = svd->vdev_ms_shift;
922 tvd->vdev_ms_count = svd->vdev_ms_count;
923 tvd->vdev_top_zap = svd->vdev_top_zap;
924
925 svd->vdev_ms_array = 0;
926 svd->vdev_ms_shift = 0;
927 svd->vdev_ms_count = 0;
928 svd->vdev_top_zap = 0;
929
930 if (tvd->vdev_mg)
931 ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg);
932 tvd->vdev_mg = svd->vdev_mg;
933 tvd->vdev_ms = svd->vdev_ms;
934
935 svd->vdev_mg = NULL;
936 svd->vdev_ms = NULL;
937
938 if (tvd->vdev_mg != NULL)
939 tvd->vdev_mg->mg_vd = tvd;
940
941 tvd->vdev_checkpoint_sm = svd->vdev_checkpoint_sm;
942 svd->vdev_checkpoint_sm = NULL;
943
944 tvd->vdev_alloc_bias = svd->vdev_alloc_bias;
945 svd->vdev_alloc_bias = VDEV_BIAS_NONE;
946
947 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc;
948 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space;
949 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace;
950
951 svd->vdev_stat.vs_alloc = 0;
952 svd->vdev_stat.vs_space = 0;
953 svd->vdev_stat.vs_dspace = 0;
954
955 /*
956 * State which may be set on a top-level vdev that's in the
957 * process of being removed.
958 */
959 ASSERT0(tvd->vdev_indirect_config.vic_births_object);
960 ASSERT0(tvd->vdev_indirect_config.vic_mapping_object);
961 ASSERT3U(tvd->vdev_indirect_config.vic_prev_indirect_vdev, ==, -1ULL);
962 ASSERT3P(tvd->vdev_indirect_mapping, ==, NULL);
963 ASSERT3P(tvd->vdev_indirect_births, ==, NULL);
964 ASSERT3P(tvd->vdev_obsolete_sm, ==, NULL);
965 ASSERT0(tvd->vdev_removing);
966 tvd->vdev_removing = svd->vdev_removing;
967 tvd->vdev_indirect_config = svd->vdev_indirect_config;
968 tvd->vdev_indirect_mapping = svd->vdev_indirect_mapping;
969 tvd->vdev_indirect_births = svd->vdev_indirect_births;
970 range_tree_swap(&svd->vdev_obsolete_segments,
971 &tvd->vdev_obsolete_segments);
972 tvd->vdev_obsolete_sm = svd->vdev_obsolete_sm;
973 svd->vdev_indirect_config.vic_mapping_object = 0;
974 svd->vdev_indirect_config.vic_births_object = 0;
975 svd->vdev_indirect_config.vic_prev_indirect_vdev = -1ULL;
976 svd->vdev_indirect_mapping = NULL;
977 svd->vdev_indirect_births = NULL;
978 svd->vdev_obsolete_sm = NULL;
979 svd->vdev_removing = 0;
980
981 for (t = 0; t < TXG_SIZE; t++) {
982 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL)
983 (void) txg_list_add(&tvd->vdev_ms_list, msp, t);
984 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL)
985 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t);
986 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t))
987 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t);
988 }
989
990 if (list_link_active(&svd->vdev_config_dirty_node)) {
991 vdev_config_clean(svd);
992 vdev_config_dirty(tvd);
993 }
994
995 if (list_link_active(&svd->vdev_state_dirty_node)) {
996 vdev_state_clean(svd);
997 vdev_state_dirty(tvd);
998 }
999
1000 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio;
1001 svd->vdev_deflate_ratio = 0;
1002
1003 tvd->vdev_islog = svd->vdev_islog;
1004 svd->vdev_islog = 0;
1005 }
1006
1007 static void
1008 vdev_top_update(vdev_t *tvd, vdev_t *vd)
1009 {
1010 if (vd == NULL)
1011 return;
1012
1013 vd->vdev_top = tvd;
1014
1015 for (int c = 0; c < vd->vdev_children; c++)
1016 vdev_top_update(tvd, vd->vdev_child[c]);
1017 }
1018
1019 /*
1020 * Add a mirror/replacing vdev above an existing vdev.
1021 */
1022 vdev_t *
1023 vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops)
1024 {
1025 spa_t *spa = cvd->vdev_spa;
1026 vdev_t *pvd = cvd->vdev_parent;
1027 vdev_t *mvd;
1028
1029 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1030
1031 mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops);
1032
1033 mvd->vdev_asize = cvd->vdev_asize;
1034 mvd->vdev_min_asize = cvd->vdev_min_asize;
1035 mvd->vdev_max_asize = cvd->vdev_max_asize;
1036 mvd->vdev_psize = cvd->vdev_psize;
1037 mvd->vdev_ashift = cvd->vdev_ashift;
1038 mvd->vdev_state = cvd->vdev_state;
1039 mvd->vdev_crtxg = cvd->vdev_crtxg;
1040
1041 vdev_remove_child(pvd, cvd);
1042 vdev_add_child(pvd, mvd);
1043 cvd->vdev_id = mvd->vdev_children;
1044 vdev_add_child(mvd, cvd);
1045 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
1046
1047 if (mvd == mvd->vdev_top)
1048 vdev_top_transfer(cvd, mvd);
1049
1050 return (mvd);
1051 }
1052
1053 /*
1054 * Remove a 1-way mirror/replacing vdev from the tree.
1055 */
1056 void
1057 vdev_remove_parent(vdev_t *cvd)
1058 {
1059 vdev_t *mvd = cvd->vdev_parent;
1060 vdev_t *pvd = mvd->vdev_parent;
1061
1062 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1063
1064 ASSERT(mvd->vdev_children == 1);
1065 ASSERT(mvd->vdev_ops == &vdev_mirror_ops ||
1066 mvd->vdev_ops == &vdev_replacing_ops ||
1067 mvd->vdev_ops == &vdev_spare_ops);
1068 cvd->vdev_ashift = mvd->vdev_ashift;
1069
1070 vdev_remove_child(mvd, cvd);
1071 vdev_remove_child(pvd, mvd);
1072
1073 /*
1074 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
1075 * Otherwise, we could have detached an offline device, and when we
1076 * go to import the pool we'll think we have two top-level vdevs,
1077 * instead of a different version of the same top-level vdev.
1078 */
1079 if (mvd->vdev_top == mvd) {
1080 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid;
1081 cvd->vdev_orig_guid = cvd->vdev_guid;
1082 cvd->vdev_guid += guid_delta;
1083 cvd->vdev_guid_sum += guid_delta;
1084 }
1085 cvd->vdev_id = mvd->vdev_id;
1086 vdev_add_child(pvd, cvd);
1087 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
1088
1089 if (cvd == cvd->vdev_top)
1090 vdev_top_transfer(mvd, cvd);
1091
1092 ASSERT(mvd->vdev_children == 0);
1093 vdev_free(mvd);
1094 }
1095
1096 static void
1097 vdev_metaslab_group_create(vdev_t *vd)
1098 {
1099 spa_t *spa = vd->vdev_spa;
1100
1101 /*
1102 * metaslab_group_create was delayed until allocation bias was available
1103 */
1104 if (vd->vdev_mg == NULL) {
1105 metaslab_class_t *mc;
1106
1107 if (vd->vdev_islog && vd->vdev_alloc_bias == VDEV_BIAS_NONE)
1108 vd->vdev_alloc_bias = VDEV_BIAS_LOG;
1109
1110 ASSERT3U(vd->vdev_islog, ==,
1111 (vd->vdev_alloc_bias == VDEV_BIAS_LOG));
1112
1113 switch (vd->vdev_alloc_bias) {
1114 case VDEV_BIAS_LOG:
1115 mc = spa_log_class(spa);
1116 break;
1117 case VDEV_BIAS_SPECIAL:
1118 mc = spa_special_class(spa);
1119 break;
1120 case VDEV_BIAS_DEDUP:
1121 mc = spa_dedup_class(spa);
1122 break;
1123 default:
1124 mc = spa_normal_class(spa);
1125 }
1126
1127 vd->vdev_mg = metaslab_group_create(mc, vd,
1128 spa->spa_alloc_count);
1129
1130 /*
1131 * The spa ashift values currently only reflect the
1132 * general vdev classes. Class destination is late
1133 * binding so ashift checking had to wait until now
1134 */
1135 if (vd->vdev_top == vd && vd->vdev_ashift != 0 &&
1136 mc == spa_normal_class(spa) && vd->vdev_aux == NULL) {
1137 if (vd->vdev_ashift > spa->spa_max_ashift)
1138 spa->spa_max_ashift = vd->vdev_ashift;
1139 if (vd->vdev_ashift < spa->spa_min_ashift)
1140 spa->spa_min_ashift = vd->vdev_ashift;
1141 }
1142 }
1143 }
1144
1145 int
1146 vdev_metaslab_init(vdev_t *vd, uint64_t txg)
1147 {
1148 spa_t *spa = vd->vdev_spa;
1149 objset_t *mos = spa->spa_meta_objset;
1150 uint64_t m;
1151 uint64_t oldc = vd->vdev_ms_count;
1152 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift;
1153 metaslab_t **mspp;
1154 int error;
1155 boolean_t expanding = (oldc != 0);
1156
1157 ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1158
1159 /*
1160 * This vdev is not being allocated from yet or is a hole.
1161 */
1162 if (vd->vdev_ms_shift == 0)
1163 return (0);
1164
1165 ASSERT(!vd->vdev_ishole);
1166
1167 ASSERT(oldc <= newc);
1168
1169 mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP);
1170
1171 if (expanding) {
1172 bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp));
1173 kmem_free(vd->vdev_ms, oldc * sizeof (*mspp));
1174 }
1175
1176 vd->vdev_ms = mspp;
1177 vd->vdev_ms_count = newc;
1178 for (m = oldc; m < newc; m++) {
1179 uint64_t object = 0;
1180
1181 /*
1182 * vdev_ms_array may be 0 if we are creating the "fake"
1183 * metaslabs for an indirect vdev for zdb's leak detection.
1184 * See zdb_leak_init().
1185 */
1186 if (txg == 0 && vd->vdev_ms_array != 0) {
1187 error = dmu_read(mos, vd->vdev_ms_array,
1188 m * sizeof (uint64_t), sizeof (uint64_t), &object,
1189 DMU_READ_PREFETCH);
1190 if (error != 0) {
1191 vdev_dbgmsg(vd, "unable to read the metaslab "
1192 "array [error=%d]", error);
1193 return (error);
1194 }
1195 }
1196
1197 #ifndef _KERNEL
1198 /*
1199 * To accomodate zdb_leak_init() fake indirect
1200 * metaslabs, we allocate a metaslab group for
1201 * indirect vdevs which normally don't have one.
1202 */
1203 if (vd->vdev_mg == NULL) {
1204 ASSERT0(vdev_is_concrete(vd));
1205 vdev_metaslab_group_create(vd);
1206 }
1207 #endif
1208 error = metaslab_init(vd->vdev_mg, m, object, txg,
1209 &(vd->vdev_ms[m]));
1210 if (error != 0) {
1211 vdev_dbgmsg(vd, "metaslab_init failed [error=%d]",
1212 error);
1213 return (error);
1214 }
1215 }
1216
1217 if (txg == 0)
1218 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER);
1219
1220 /*
1221 * If the vdev is being removed we don't activate
1222 * the metaslabs since we want to ensure that no new
1223 * allocations are performed on this device.
1224 */
1225 if (!expanding && !vd->vdev_removing) {
1226 metaslab_group_activate(vd->vdev_mg);
1227 }
1228
1229 if (txg == 0)
1230 spa_config_exit(spa, SCL_ALLOC, FTAG);
1231
1232 return (0);
1233 }
1234
1235 void
1236 vdev_metaslab_fini(vdev_t *vd)
1237 {
1238 if (vd->vdev_checkpoint_sm != NULL) {
1239 ASSERT(spa_feature_is_active(vd->vdev_spa,
1240 SPA_FEATURE_POOL_CHECKPOINT));
1241 space_map_close(vd->vdev_checkpoint_sm);
1242 /*
1243 * Even though we close the space map, we need to set its
1244 * pointer to NULL. The reason is that vdev_metaslab_fini()
1245 * may be called multiple times for certain operations
1246 * (i.e. when destroying a pool) so we need to ensure that
1247 * this clause never executes twice. This logic is similar
1248 * to the one used for the vdev_ms clause below.
1249 */
1250 vd->vdev_checkpoint_sm = NULL;
1251 }
1252
1253 if (vd->vdev_ms != NULL) {
1254 uint64_t count = vd->vdev_ms_count;
1255
1256 metaslab_group_passivate(vd->vdev_mg);
1257 for (uint64_t m = 0; m < count; m++) {
1258 metaslab_t *msp = vd->vdev_ms[m];
1259
1260 if (msp != NULL)
1261 metaslab_fini(msp);
1262 }
1263 kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *));
1264 vd->vdev_ms = NULL;
1265
1266 vd->vdev_ms_count = 0;
1267 }
1268 ASSERT0(vd->vdev_ms_count);
1269 }
1270
1271 typedef struct vdev_probe_stats {
1272 boolean_t vps_readable;
1273 boolean_t vps_writeable;
1274 int vps_flags;
1275 } vdev_probe_stats_t;
1276
1277 static void
1278 vdev_probe_done(zio_t *zio)
1279 {
1280 spa_t *spa = zio->io_spa;
1281 vdev_t *vd = zio->io_vd;
1282 vdev_probe_stats_t *vps = zio->io_private;
1283
1284 ASSERT(vd->vdev_probe_zio != NULL);
1285
1286 if (zio->io_type == ZIO_TYPE_READ) {
1287 if (zio->io_error == 0)
1288 vps->vps_readable = 1;
1289 if (zio->io_error == 0 && spa_writeable(spa)) {
1290 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd,
1291 zio->io_offset, zio->io_size, zio->io_abd,
1292 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1293 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE));
1294 } else {
1295 abd_free(zio->io_abd);
1296 }
1297 } else if (zio->io_type == ZIO_TYPE_WRITE) {
1298 if (zio->io_error == 0)
1299 vps->vps_writeable = 1;
1300 abd_free(zio->io_abd);
1301 } else if (zio->io_type == ZIO_TYPE_NULL) {
1302 zio_t *pio;
1303
1304 vd->vdev_cant_read |= !vps->vps_readable;
1305 vd->vdev_cant_write |= !vps->vps_writeable;
1306
1307 if (vdev_readable(vd) &&
1308 (vdev_writeable(vd) || !spa_writeable(spa))) {
1309 zio->io_error = 0;
1310 } else {
1311 ASSERT(zio->io_error != 0);
1312 vdev_dbgmsg(vd, "failed probe");
1313 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE,
1314 spa, vd, NULL, 0, 0);
1315 zio->io_error = SET_ERROR(ENXIO);
1316 }
1317
1318 mutex_enter(&vd->vdev_probe_lock);
1319 ASSERT(vd->vdev_probe_zio == zio);
1320 vd->vdev_probe_zio = NULL;
1321 mutex_exit(&vd->vdev_probe_lock);
1322
1323 zio_link_t *zl = NULL;
1324 while ((pio = zio_walk_parents(zio, &zl)) != NULL)
1325 if (!vdev_accessible(vd, pio))
1326 pio->io_error = SET_ERROR(ENXIO);
1327
1328 kmem_free(vps, sizeof (*vps));
1329 }
1330 }
1331
1332 /*
1333 * Determine whether this device is accessible.
1334 *
1335 * Read and write to several known locations: the pad regions of each
1336 * vdev label but the first, which we leave alone in case it contains
1337 * a VTOC.
1338 */
1339 zio_t *
1340 vdev_probe(vdev_t *vd, zio_t *zio)
1341 {
1342 spa_t *spa = vd->vdev_spa;
1343 vdev_probe_stats_t *vps = NULL;
1344 zio_t *pio;
1345
1346 ASSERT(vd->vdev_ops->vdev_op_leaf);
1347
1348 /*
1349 * Don't probe the probe.
1350 */
1351 if (zio && (zio->io_flags & ZIO_FLAG_PROBE))
1352 return (NULL);
1353
1354 /*
1355 * To prevent 'probe storms' when a device fails, we create
1356 * just one probe i/o at a time. All zios that want to probe
1357 * this vdev will become parents of the probe io.
1358 */
1359 mutex_enter(&vd->vdev_probe_lock);
1360
1361 if ((pio = vd->vdev_probe_zio) == NULL) {
1362 vps = kmem_zalloc(sizeof (*vps), KM_SLEEP);
1363
1364 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE |
1365 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE |
1366 ZIO_FLAG_TRYHARD;
1367
1368 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) {
1369 /*
1370 * vdev_cant_read and vdev_cant_write can only
1371 * transition from TRUE to FALSE when we have the
1372 * SCL_ZIO lock as writer; otherwise they can only
1373 * transition from FALSE to TRUE. This ensures that
1374 * any zio looking at these values can assume that
1375 * failures persist for the life of the I/O. That's
1376 * important because when a device has intermittent
1377 * connectivity problems, we want to ensure that
1378 * they're ascribed to the device (ENXIO) and not
1379 * the zio (EIO).
1380 *
1381 * Since we hold SCL_ZIO as writer here, clear both
1382 * values so the probe can reevaluate from first
1383 * principles.
1384 */
1385 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER;
1386 vd->vdev_cant_read = B_FALSE;
1387 vd->vdev_cant_write = B_FALSE;
1388 }
1389
1390 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd,
1391 vdev_probe_done, vps,
1392 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE);
1393
1394 /*
1395 * We can't change the vdev state in this context, so we
1396 * kick off an async task to do it on our behalf.
1397 */
1398 if (zio != NULL) {
1399 vd->vdev_probe_wanted = B_TRUE;
1400 spa_async_request(spa, SPA_ASYNC_PROBE);
1401 }
1402 }
1403
1404 if (zio != NULL)
1405 zio_add_child(zio, pio);
1406
1407 mutex_exit(&vd->vdev_probe_lock);
1408
1409 if (vps == NULL) {
1410 ASSERT(zio != NULL);
1411 return (NULL);
1412 }
1413
1414 for (int l = 1; l < VDEV_LABELS; l++) {
1415 zio_nowait(zio_read_phys(pio, vd,
1416 vdev_label_offset(vd->vdev_psize, l,
1417 offsetof(vdev_label_t, vl_pad2)), VDEV_PAD_SIZE,
1418 abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE),
1419 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1420 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE));
1421 }
1422
1423 if (zio == NULL)
1424 return (pio);
1425
1426 zio_nowait(pio);
1427 return (NULL);
1428 }
1429
1430 static void
1431 vdev_open_child(void *arg)
1432 {
1433 vdev_t *vd = arg;
1434
1435 vd->vdev_open_thread = curthread;
1436 vd->vdev_open_error = vdev_open(vd);
1437 vd->vdev_open_thread = NULL;
1438 }
1439
1440 boolean_t
1441 vdev_uses_zvols(vdev_t *vd)
1442 {
1443 if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR,
1444 strlen(ZVOL_DIR)) == 0)
1445 return (B_TRUE);
1446 for (int c = 0; c < vd->vdev_children; c++)
1447 if (vdev_uses_zvols(vd->vdev_child[c]))
1448 return (B_TRUE);
1449 return (B_FALSE);
1450 }
1451
1452 void
1453 vdev_open_children(vdev_t *vd)
1454 {
1455 taskq_t *tq;
1456 int children = vd->vdev_children;
1457
1458 /*
1459 * in order to handle pools on top of zvols, do the opens
1460 * in a single thread so that the same thread holds the
1461 * spa_namespace_lock
1462 */
1463 if (vdev_uses_zvols(vd)) {
1464 for (int c = 0; c < children; c++)
1465 vd->vdev_child[c]->vdev_open_error =
1466 vdev_open(vd->vdev_child[c]);
1467 return;
1468 }
1469 tq = taskq_create("vdev_open", children, minclsyspri,
1470 children, children, TASKQ_PREPOPULATE);
1471
1472 for (int c = 0; c < children; c++)
1473 VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c],
1474 TQ_SLEEP) != TASKQID_INVALID);
1475
1476 taskq_destroy(tq);
1477 }
1478
1479 /*
1480 * Compute the raidz-deflation ratio. Note, we hard-code
1481 * in 128k (1 << 17) because it is the "typical" blocksize.
1482 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change,
1483 * otherwise it would inconsistently account for existing bp's.
1484 */
1485 static void
1486 vdev_set_deflate_ratio(vdev_t *vd)
1487 {
1488 if (vd == vd->vdev_top && !vd->vdev_ishole && vd->vdev_ashift != 0) {
1489 vd->vdev_deflate_ratio = (1 << 17) /
1490 (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT);
1491 }
1492 }
1493
1494 /*
1495 * Prepare a virtual device for access.
1496 */
1497 int
1498 vdev_open(vdev_t *vd)
1499 {
1500 spa_t *spa = vd->vdev_spa;
1501 int error;
1502 uint64_t osize = 0;
1503 uint64_t max_osize = 0;
1504 uint64_t asize, max_asize, psize;
1505 uint64_t ashift = 0;
1506
1507 ASSERT(vd->vdev_open_thread == curthread ||
1508 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1509 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED ||
1510 vd->vdev_state == VDEV_STATE_CANT_OPEN ||
1511 vd->vdev_state == VDEV_STATE_OFFLINE);
1512
1513 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1514 vd->vdev_cant_read = B_FALSE;
1515 vd->vdev_cant_write = B_FALSE;
1516 vd->vdev_min_asize = vdev_get_min_asize(vd);
1517
1518 /*
1519 * If this vdev is not removed, check its fault status. If it's
1520 * faulted, bail out of the open.
1521 */
1522 if (!vd->vdev_removed && vd->vdev_faulted) {
1523 ASSERT(vd->vdev_children == 0);
1524 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1525 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1526 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1527 vd->vdev_label_aux);
1528 return (SET_ERROR(ENXIO));
1529 } else if (vd->vdev_offline) {
1530 ASSERT(vd->vdev_children == 0);
1531 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE);
1532 return (SET_ERROR(ENXIO));
1533 }
1534
1535 error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize, &ashift);
1536
1537 /*
1538 * Reset the vdev_reopening flag so that we actually close
1539 * the vdev on error.
1540 */
1541 vd->vdev_reopening = B_FALSE;
1542 if (zio_injection_enabled && error == 0)
1543 error = zio_handle_device_injection(vd, NULL, ENXIO);
1544
1545 if (error) {
1546 if (vd->vdev_removed &&
1547 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED)
1548 vd->vdev_removed = B_FALSE;
1549
1550 if (vd->vdev_stat.vs_aux == VDEV_AUX_CHILDREN_OFFLINE) {
1551 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE,
1552 vd->vdev_stat.vs_aux);
1553 } else {
1554 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1555 vd->vdev_stat.vs_aux);
1556 }
1557 return (error);
1558 }
1559
1560 vd->vdev_removed = B_FALSE;
1561
1562 /*
1563 * Recheck the faulted flag now that we have confirmed that
1564 * the vdev is accessible. If we're faulted, bail.
1565 */
1566 if (vd->vdev_faulted) {
1567 ASSERT(vd->vdev_children == 0);
1568 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1569 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1570 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1571 vd->vdev_label_aux);
1572 return (SET_ERROR(ENXIO));
1573 }
1574
1575 if (vd->vdev_degraded) {
1576 ASSERT(vd->vdev_children == 0);
1577 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1578 VDEV_AUX_ERR_EXCEEDED);
1579 } else {
1580 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0);
1581 }
1582
1583 /*
1584 * For hole or missing vdevs we just return success.
1585 */
1586 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops)
1587 return (0);
1588
1589 for (int c = 0; c < vd->vdev_children; c++) {
1590 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) {
1591 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1592 VDEV_AUX_NONE);
1593 break;
1594 }
1595 }
1596
1597 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t));
1598 max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t));
1599
1600 if (vd->vdev_children == 0) {
1601 if (osize < SPA_MINDEVSIZE) {
1602 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1603 VDEV_AUX_TOO_SMALL);
1604 return (SET_ERROR(EOVERFLOW));
1605 }
1606 psize = osize;
1607 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE);
1608 max_asize = max_osize - (VDEV_LABEL_START_SIZE +
1609 VDEV_LABEL_END_SIZE);
1610 } else {
1611 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE -
1612 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) {
1613 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1614 VDEV_AUX_TOO_SMALL);
1615 return (SET_ERROR(EOVERFLOW));
1616 }
1617 psize = 0;
1618 asize = osize;
1619 max_asize = max_osize;
1620 }
1621
1622 vd->vdev_psize = psize;
1623
1624 /*
1625 * Make sure the allocatable size hasn't shrunk too much.
1626 */
1627 if (asize < vd->vdev_min_asize) {
1628 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1629 VDEV_AUX_BAD_LABEL);
1630 return (SET_ERROR(EINVAL));
1631 }
1632
1633 if (vd->vdev_asize == 0) {
1634 /*
1635 * This is the first-ever open, so use the computed values.
1636 * For testing purposes, a higher ashift can be requested.
1637 */
1638 vd->vdev_asize = asize;
1639 vd->vdev_max_asize = max_asize;
1640 vd->vdev_ashift = MAX(ashift, vd->vdev_ashift);
1641 vd->vdev_ashift = MAX(zfs_ashift_min, vd->vdev_ashift);
1642 } else {
1643 /*
1644 * Detect if the alignment requirement has increased.
1645 * We don't want to make the pool unavailable, just
1646 * issue a warning instead.
1647 */
1648 if (ashift > vd->vdev_top->vdev_ashift &&
1649 vd->vdev_ops->vdev_op_leaf) {
1650 cmn_err(CE_WARN,
1651 "Disk, '%s', has a block alignment that is "
1652 "larger than the pool's alignment\n",
1653 vd->vdev_path);
1654 }
1655 vd->vdev_max_asize = max_asize;
1656 }
1657
1658 /*
1659 * If all children are healthy we update asize if either:
1660 * The asize has increased, due to a device expansion caused by dynamic
1661 * LUN growth or vdev replacement, and automatic expansion is enabled;
1662 * making the additional space available.
1663 *
1664 * The asize has decreased, due to a device shrink usually caused by a
1665 * vdev replace with a smaller device. This ensures that calculations
1666 * based of max_asize and asize e.g. esize are always valid. It's safe
1667 * to do this as we've already validated that asize is greater than
1668 * vdev_min_asize.
1669 */
1670 if (vd->vdev_state == VDEV_STATE_HEALTHY &&
1671 ((asize > vd->vdev_asize &&
1672 (vd->vdev_expanding || spa->spa_autoexpand)) ||
1673 (asize < vd->vdev_asize)))
1674 vd->vdev_asize = asize;
1675
1676 vdev_set_min_asize(vd);
1677
1678 /*
1679 * Ensure we can issue some IO before declaring the
1680 * vdev open for business.
1681 */
1682 if (vd->vdev_ops->vdev_op_leaf &&
1683 (error = zio_wait(vdev_probe(vd, NULL))) != 0) {
1684 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1685 VDEV_AUX_ERR_EXCEEDED);
1686 return (error);
1687 }
1688
1689 /*
1690 * Track the min and max ashift values for normal data devices.
1691 *
1692 * DJB - TBD these should perhaps be tracked per allocation class
1693 * (e.g. spa_min_ashift is used to round up post compression buffers)
1694 */
1695 if (vd->vdev_top == vd && vd->vdev_ashift != 0 &&
1696 vd->vdev_alloc_bias == VDEV_BIAS_NONE &&
1697 vd->vdev_aux == NULL) {
1698 if (vd->vdev_ashift > spa->spa_max_ashift)
1699 spa->spa_max_ashift = vd->vdev_ashift;
1700 if (vd->vdev_ashift < spa->spa_min_ashift)
1701 spa->spa_min_ashift = vd->vdev_ashift;
1702 }
1703
1704 /*
1705 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1706 * resilver. But don't do this if we are doing a reopen for a scrub,
1707 * since this would just restart the scrub we are already doing.
1708 */
1709 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen &&
1710 vdev_resilver_needed(vd, NULL, NULL))
1711 spa_async_request(spa, SPA_ASYNC_RESILVER);
1712
1713 return (0);
1714 }
1715
1716 /*
1717 * Called once the vdevs are all opened, this routine validates the label
1718 * contents. This needs to be done before vdev_load() so that we don't
1719 * inadvertently do repair I/Os to the wrong device.
1720 *
1721 * This function will only return failure if one of the vdevs indicates that it
1722 * has since been destroyed or exported. This is only possible if
1723 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1724 * will be updated but the function will return 0.
1725 */
1726 int
1727 vdev_validate(vdev_t *vd)
1728 {
1729 spa_t *spa = vd->vdev_spa;
1730 nvlist_t *label;
1731 uint64_t guid = 0, aux_guid = 0, top_guid;
1732 uint64_t state;
1733 nvlist_t *nvl;
1734 uint64_t txg;
1735
1736 if (vdev_validate_skip)
1737 return (0);
1738
1739 for (uint64_t c = 0; c < vd->vdev_children; c++)
1740 if (vdev_validate(vd->vdev_child[c]) != 0)
1741 return (SET_ERROR(EBADF));
1742
1743 /*
1744 * If the device has already failed, or was marked offline, don't do
1745 * any further validation. Otherwise, label I/O will fail and we will
1746 * overwrite the previous state.
1747 */
1748 if (!vd->vdev_ops->vdev_op_leaf || !vdev_readable(vd))
1749 return (0);
1750
1751 /*
1752 * If we are performing an extreme rewind, we allow for a label that
1753 * was modified at a point after the current txg.
1754 * If config lock is not held do not check for the txg. spa_sync could
1755 * be updating the vdev's label before updating spa_last_synced_txg.
1756 */
1757 if (spa->spa_extreme_rewind || spa_last_synced_txg(spa) == 0 ||
1758 spa_config_held(spa, SCL_CONFIG, RW_WRITER) != SCL_CONFIG)
1759 txg = UINT64_MAX;
1760 else
1761 txg = spa_last_synced_txg(spa);
1762
1763 if ((label = vdev_label_read_config(vd, txg)) == NULL) {
1764 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1765 VDEV_AUX_BAD_LABEL);
1766 vdev_dbgmsg(vd, "vdev_validate: failed reading config for "
1767 "txg %llu", (u_longlong_t)txg);
1768 return (0);
1769 }
1770
1771 /*
1772 * Determine if this vdev has been split off into another
1773 * pool. If so, then refuse to open it.
1774 */
1775 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID,
1776 &aux_guid) == 0 && aux_guid == spa_guid(spa)) {
1777 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1778 VDEV_AUX_SPLIT_POOL);
1779 nvlist_free(label);
1780 vdev_dbgmsg(vd, "vdev_validate: vdev split into other pool");
1781 return (0);
1782 }
1783
1784 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, &guid) != 0) {
1785 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1786 VDEV_AUX_CORRUPT_DATA);
1787 nvlist_free(label);
1788 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
1789 ZPOOL_CONFIG_POOL_GUID);
1790 return (0);
1791 }
1792
1793 /*
1794 * If config is not trusted then ignore the spa guid check. This is
1795 * necessary because if the machine crashed during a re-guid the new
1796 * guid might have been written to all of the vdev labels, but not the
1797 * cached config. The check will be performed again once we have the
1798 * trusted config from the MOS.
1799 */
1800 if (spa->spa_trust_config && guid != spa_guid(spa)) {
1801 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1802 VDEV_AUX_CORRUPT_DATA);
1803 nvlist_free(label);
1804 vdev_dbgmsg(vd, "vdev_validate: vdev label pool_guid doesn't "
1805 "match config (%llu != %llu)", (u_longlong_t)guid,
1806 (u_longlong_t)spa_guid(spa));
1807 return (0);
1808 }
1809
1810 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl)
1811 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID,
1812 &aux_guid) != 0)
1813 aux_guid = 0;
1814
1815 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0) {
1816 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1817 VDEV_AUX_CORRUPT_DATA);
1818 nvlist_free(label);
1819 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
1820 ZPOOL_CONFIG_GUID);
1821 return (0);
1822 }
1823
1824 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, &top_guid)
1825 != 0) {
1826 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1827 VDEV_AUX_CORRUPT_DATA);
1828 nvlist_free(label);
1829 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
1830 ZPOOL_CONFIG_TOP_GUID);
1831 return (0);
1832 }
1833
1834 /*
1835 * If this vdev just became a top-level vdev because its sibling was
1836 * detached, it will have adopted the parent's vdev guid -- but the
1837 * label may or may not be on disk yet. Fortunately, either version
1838 * of the label will have the same top guid, so if we're a top-level
1839 * vdev, we can safely compare to that instead.
1840 * However, if the config comes from a cachefile that failed to update
1841 * after the detach, a top-level vdev will appear as a non top-level
1842 * vdev in the config. Also relax the constraints if we perform an
1843 * extreme rewind.
1844 *
1845 * If we split this vdev off instead, then we also check the
1846 * original pool's guid. We don't want to consider the vdev
1847 * corrupt if it is partway through a split operation.
1848 */
1849 if (vd->vdev_guid != guid && vd->vdev_guid != aux_guid) {
1850 boolean_t mismatch = B_FALSE;
1851 if (spa->spa_trust_config && !spa->spa_extreme_rewind) {
1852 if (vd != vd->vdev_top || vd->vdev_guid != top_guid)
1853 mismatch = B_TRUE;
1854 } else {
1855 if (vd->vdev_guid != top_guid &&
1856 vd->vdev_top->vdev_guid != guid)
1857 mismatch = B_TRUE;
1858 }
1859
1860 if (mismatch) {
1861 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1862 VDEV_AUX_CORRUPT_DATA);
1863 nvlist_free(label);
1864 vdev_dbgmsg(vd, "vdev_validate: config guid "
1865 "doesn't match label guid");
1866 vdev_dbgmsg(vd, "CONFIG: guid %llu, top_guid %llu",
1867 (u_longlong_t)vd->vdev_guid,
1868 (u_longlong_t)vd->vdev_top->vdev_guid);
1869 vdev_dbgmsg(vd, "LABEL: guid %llu, top_guid %llu, "
1870 "aux_guid %llu", (u_longlong_t)guid,
1871 (u_longlong_t)top_guid, (u_longlong_t)aux_guid);
1872 return (0);
1873 }
1874 }
1875
1876 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1877 &state) != 0) {
1878 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1879 VDEV_AUX_CORRUPT_DATA);
1880 nvlist_free(label);
1881 vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
1882 ZPOOL_CONFIG_POOL_STATE);
1883 return (0);
1884 }
1885
1886 nvlist_free(label);
1887
1888 /*
1889 * If this is a verbatim import, no need to check the
1890 * state of the pool.
1891 */
1892 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) &&
1893 spa_load_state(spa) == SPA_LOAD_OPEN &&
1894 state != POOL_STATE_ACTIVE) {
1895 vdev_dbgmsg(vd, "vdev_validate: invalid pool state (%llu) "
1896 "for spa %s", (u_longlong_t)state, spa->spa_name);
1897 return (SET_ERROR(EBADF));
1898 }
1899
1900 /*
1901 * If we were able to open and validate a vdev that was
1902 * previously marked permanently unavailable, clear that state
1903 * now.
1904 */
1905 if (vd->vdev_not_present)
1906 vd->vdev_not_present = 0;
1907
1908 return (0);
1909 }
1910
1911 static void
1912 vdev_copy_path_impl(vdev_t *svd, vdev_t *dvd)
1913 {
1914 if (svd->vdev_path != NULL && dvd->vdev_path != NULL) {
1915 if (strcmp(svd->vdev_path, dvd->vdev_path) != 0) {
1916 zfs_dbgmsg("vdev_copy_path: vdev %llu: path changed "
1917 "from '%s' to '%s'", (u_longlong_t)dvd->vdev_guid,
1918 dvd->vdev_path, svd->vdev_path);
1919 spa_strfree(dvd->vdev_path);
1920 dvd->vdev_path = spa_strdup(svd->vdev_path);
1921 }
1922 } else if (svd->vdev_path != NULL) {
1923 dvd->vdev_path = spa_strdup(svd->vdev_path);
1924 zfs_dbgmsg("vdev_copy_path: vdev %llu: path set to '%s'",
1925 (u_longlong_t)dvd->vdev_guid, dvd->vdev_path);
1926 }
1927 }
1928
1929 /*
1930 * Recursively copy vdev paths from one vdev to another. Source and destination
1931 * vdev trees must have same geometry otherwise return error. Intended to copy
1932 * paths from userland config into MOS config.
1933 */
1934 int
1935 vdev_copy_path_strict(vdev_t *svd, vdev_t *dvd)
1936 {
1937 if ((svd->vdev_ops == &vdev_missing_ops) ||
1938 (svd->vdev_ishole && dvd->vdev_ishole) ||
1939 (dvd->vdev_ops == &vdev_indirect_ops))
1940 return (0);
1941
1942 if (svd->vdev_ops != dvd->vdev_ops) {
1943 vdev_dbgmsg(svd, "vdev_copy_path: vdev type mismatch: %s != %s",
1944 svd->vdev_ops->vdev_op_type, dvd->vdev_ops->vdev_op_type);
1945 return (SET_ERROR(EINVAL));
1946 }
1947
1948 if (svd->vdev_guid != dvd->vdev_guid) {
1949 vdev_dbgmsg(svd, "vdev_copy_path: guids mismatch (%llu != "
1950 "%llu)", (u_longlong_t)svd->vdev_guid,
1951 (u_longlong_t)dvd->vdev_guid);
1952 return (SET_ERROR(EINVAL));
1953 }
1954
1955 if (svd->vdev_children != dvd->vdev_children) {
1956 vdev_dbgmsg(svd, "vdev_copy_path: children count mismatch: "
1957 "%llu != %llu", (u_longlong_t)svd->vdev_children,
1958 (u_longlong_t)dvd->vdev_children);
1959 return (SET_ERROR(EINVAL));
1960 }
1961
1962 for (uint64_t i = 0; i < svd->vdev_children; i++) {
1963 int error = vdev_copy_path_strict(svd->vdev_child[i],
1964 dvd->vdev_child[i]);
1965 if (error != 0)
1966 return (error);
1967 }
1968
1969 if (svd->vdev_ops->vdev_op_leaf)
1970 vdev_copy_path_impl(svd, dvd);
1971
1972 return (0);
1973 }
1974
1975 static void
1976 vdev_copy_path_search(vdev_t *stvd, vdev_t *dvd)
1977 {
1978 ASSERT(stvd->vdev_top == stvd);
1979 ASSERT3U(stvd->vdev_id, ==, dvd->vdev_top->vdev_id);
1980
1981 for (uint64_t i = 0; i < dvd->vdev_children; i++) {
1982 vdev_copy_path_search(stvd, dvd->vdev_child[i]);
1983 }
1984
1985 if (!dvd->vdev_ops->vdev_op_leaf || !vdev_is_concrete(dvd))
1986 return;
1987
1988 /*
1989 * The idea here is that while a vdev can shift positions within
1990 * a top vdev (when replacing, attaching mirror, etc.) it cannot
1991 * step outside of it.
1992 */
1993 vdev_t *vd = vdev_lookup_by_guid(stvd, dvd->vdev_guid);
1994
1995 if (vd == NULL || vd->vdev_ops != dvd->vdev_ops)
1996 return;
1997
1998 ASSERT(vd->vdev_ops->vdev_op_leaf);
1999
2000 vdev_copy_path_impl(vd, dvd);
2001 }
2002
2003 /*
2004 * Recursively copy vdev paths from one root vdev to another. Source and
2005 * destination vdev trees may differ in geometry. For each destination leaf
2006 * vdev, search a vdev with the same guid and top vdev id in the source.
2007 * Intended to copy paths from userland config into MOS config.
2008 */
2009 void
2010 vdev_copy_path_relaxed(vdev_t *srvd, vdev_t *drvd)
2011 {
2012 uint64_t children = MIN(srvd->vdev_children, drvd->vdev_children);
2013 ASSERT(srvd->vdev_ops == &vdev_root_ops);
2014 ASSERT(drvd->vdev_ops == &vdev_root_ops);
2015
2016 for (uint64_t i = 0; i < children; i++) {
2017 vdev_copy_path_search(srvd->vdev_child[i],
2018 drvd->vdev_child[i]);
2019 }
2020 }
2021
2022 /*
2023 * Close a virtual device.
2024 */
2025 void
2026 vdev_close(vdev_t *vd)
2027 {
2028 spa_t *spa = vd->vdev_spa;
2029 vdev_t *pvd = vd->vdev_parent;
2030
2031 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2032
2033 /*
2034 * If our parent is reopening, then we are as well, unless we are
2035 * going offline.
2036 */
2037 if (pvd != NULL && pvd->vdev_reopening)
2038 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline);
2039
2040 vd->vdev_ops->vdev_op_close(vd);
2041
2042 vdev_cache_purge(vd);
2043
2044 /*
2045 * We record the previous state before we close it, so that if we are
2046 * doing a reopen(), we don't generate FMA ereports if we notice that
2047 * it's still faulted.
2048 */
2049 vd->vdev_prevstate = vd->vdev_state;
2050
2051 if (vd->vdev_offline)
2052 vd->vdev_state = VDEV_STATE_OFFLINE;
2053 else
2054 vd->vdev_state = VDEV_STATE_CLOSED;
2055 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
2056 }
2057
2058 void
2059 vdev_hold(vdev_t *vd)
2060 {
2061 spa_t *spa = vd->vdev_spa;
2062
2063 ASSERT(spa_is_root(spa));
2064 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
2065 return;
2066
2067 for (int c = 0; c < vd->vdev_children; c++)
2068 vdev_hold(vd->vdev_child[c]);
2069
2070 if (vd->vdev_ops->vdev_op_leaf)
2071 vd->vdev_ops->vdev_op_hold(vd);
2072 }
2073
2074 void
2075 vdev_rele(vdev_t *vd)
2076 {
2077 spa_t *spa = vd->vdev_spa;
2078
2079 ASSERT(spa_is_root(spa));
2080 for (int c = 0; c < vd->vdev_children; c++)
2081 vdev_rele(vd->vdev_child[c]);
2082
2083 if (vd->vdev_ops->vdev_op_leaf)
2084 vd->vdev_ops->vdev_op_rele(vd);
2085 }
2086
2087 /*
2088 * Reopen all interior vdevs and any unopened leaves. We don't actually
2089 * reopen leaf vdevs which had previously been opened as they might deadlock
2090 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
2091 * If the leaf has never been opened then open it, as usual.
2092 */
2093 void
2094 vdev_reopen(vdev_t *vd)
2095 {
2096 spa_t *spa = vd->vdev_spa;
2097
2098 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2099
2100 /* set the reopening flag unless we're taking the vdev offline */
2101 vd->vdev_reopening = !vd->vdev_offline;
2102 vdev_close(vd);
2103 (void) vdev_open(vd);
2104
2105 /*
2106 * Call vdev_validate() here to make sure we have the same device.
2107 * Otherwise, a device with an invalid label could be successfully
2108 * opened in response to vdev_reopen().
2109 */
2110 if (vd->vdev_aux) {
2111 (void) vdev_validate_aux(vd);
2112 if (vdev_readable(vd) && vdev_writeable(vd) &&
2113 vd->vdev_aux == &spa->spa_l2cache &&
2114 !l2arc_vdev_present(vd))
2115 l2arc_add_vdev(spa, vd);
2116 } else {
2117 (void) vdev_validate(vd);
2118 }
2119
2120 /*
2121 * Reassess parent vdev's health.
2122 */
2123 vdev_propagate_state(vd);
2124 }
2125
2126 int
2127 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
2128 {
2129 int error;
2130
2131 /*
2132 * Normally, partial opens (e.g. of a mirror) are allowed.
2133 * For a create, however, we want to fail the request if
2134 * there are any components we can't open.
2135 */
2136 error = vdev_open(vd);
2137
2138 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) {
2139 vdev_close(vd);
2140 return (error ? error : ENXIO);
2141 }
2142
2143 /*
2144 * Recursively load DTLs and initialize all labels.
2145 */
2146 if ((error = vdev_dtl_load(vd)) != 0 ||
2147 (error = vdev_label_init(vd, txg, isreplacing ?
2148 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
2149 vdev_close(vd);
2150 return (error);
2151 }
2152
2153 return (0);
2154 }
2155
2156 void
2157 vdev_metaslab_set_size(vdev_t *vd)
2158 {
2159 uint64_t asize = vd->vdev_asize;
2160 uint64_t ms_count = asize >> zfs_vdev_default_ms_shift;
2161 uint64_t ms_shift;
2162
2163 /*
2164 * There are two dimensions to the metaslab sizing calculation:
2165 * the size of the metaslab and the count of metaslabs per vdev.
2166 *
2167 * The default values used below are a good balance between memory
2168 * usage (larger metaslab size means more memory needed for loaded
2169 * metaslabs; more metaslabs means more memory needed for the
2170 * metaslab_t structs), metaslab load time (larger metaslabs take
2171 * longer to load), and metaslab sync time (more metaslabs means
2172 * more time spent syncing all of them).
2173 *
2174 * In general, we aim for zfs_vdev_default_ms_count (200) metaslabs.
2175 * The range of the dimensions are as follows:
2176 *
2177 * 2^29 <= ms_size <= 2^34
2178 * 16 <= ms_count <= 131,072
2179 *
2180 * On the lower end of vdev sizes, we aim for metaslabs sizes of
2181 * at least 512MB (2^29) to minimize fragmentation effects when
2182 * testing with smaller devices. However, the count constraint
2183 * of at least 16 metaslabs will override this minimum size goal.
2184 *
2185 * On the upper end of vdev sizes, we aim for a maximum metaslab
2186 * size of 16GB. However, we will cap the total count to 2^17
2187 * metaslabs to keep our memory footprint in check and let the
2188 * metaslab size grow from there if that limit is hit.
2189 *
2190 * The net effect of applying above constrains is summarized below.
2191 *
2192 * vdev size metaslab count
2193 * --------------|-----------------
2194 * < 8GB ~16
2195 * 8GB - 100GB one per 512MB
2196 * 100GB - 3TB ~200
2197 * 3TB - 2PB one per 16GB
2198 * > 2PB ~131,072
2199 * --------------------------------
2200 *
2201 * Finally, note that all of the above calculate the initial
2202 * number of metaslabs. Expanding a top-level vdev will result
2203 * in additional metaslabs being allocated making it possible
2204 * to exceed the zfs_vdev_ms_count_limit.
2205 */
2206
2207 if (ms_count < zfs_vdev_min_ms_count)
2208 ms_shift = highbit64(asize / zfs_vdev_min_ms_count);
2209 else if (ms_count > zfs_vdev_default_ms_count)
2210 ms_shift = highbit64(asize / zfs_vdev_default_ms_count);
2211 else
2212 ms_shift = zfs_vdev_default_ms_shift;
2213
2214 if (ms_shift < SPA_MAXBLOCKSHIFT) {
2215 ms_shift = SPA_MAXBLOCKSHIFT;
2216 } else if (ms_shift > zfs_vdev_max_ms_shift) {
2217 ms_shift = zfs_vdev_max_ms_shift;
2218 /* cap the total count to constrain memory footprint */
2219 if ((asize >> ms_shift) > zfs_vdev_ms_count_limit)
2220 ms_shift = highbit64(asize / zfs_vdev_ms_count_limit);
2221 }
2222
2223 vd->vdev_ms_shift = ms_shift;
2224 ASSERT3U(vd->vdev_ms_shift, >=, SPA_MAXBLOCKSHIFT);
2225 }
2226
2227 void
2228 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
2229 {
2230 ASSERT(vd == vd->vdev_top);
2231 /* indirect vdevs don't have metaslabs or dtls */
2232 ASSERT(vdev_is_concrete(vd) || flags == 0);
2233 ASSERT(ISP2(flags));
2234 ASSERT(spa_writeable(vd->vdev_spa));
2235
2236 if (flags & VDD_METASLAB)
2237 (void) txg_list_add(&vd->vdev_ms_list, arg, txg);
2238
2239 if (flags & VDD_DTL)
2240 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg);
2241
2242 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
2243 }
2244
2245 void
2246 vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg)
2247 {
2248 for (int c = 0; c < vd->vdev_children; c++)
2249 vdev_dirty_leaves(vd->vdev_child[c], flags, txg);
2250
2251 if (vd->vdev_ops->vdev_op_leaf)
2252 vdev_dirty(vd->vdev_top, flags, vd, txg);
2253 }
2254
2255 /*
2256 * DTLs.
2257 *
2258 * A vdev's DTL (dirty time log) is the set of transaction groups for which
2259 * the vdev has less than perfect replication. There are four kinds of DTL:
2260 *
2261 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
2262 *
2263 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
2264 *
2265 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
2266 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
2267 * txgs that was scrubbed.
2268 *
2269 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
2270 * persistent errors or just some device being offline.
2271 * Unlike the other three, the DTL_OUTAGE map is not generally
2272 * maintained; it's only computed when needed, typically to
2273 * determine whether a device can be detached.
2274 *
2275 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
2276 * either has the data or it doesn't.
2277 *
2278 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
2279 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
2280 * if any child is less than fully replicated, then so is its parent.
2281 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
2282 * comprising only those txgs which appear in 'maxfaults' or more children;
2283 * those are the txgs we don't have enough replication to read. For example,
2284 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
2285 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
2286 * two child DTL_MISSING maps.
2287 *
2288 * It should be clear from the above that to compute the DTLs and outage maps
2289 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
2290 * Therefore, that is all we keep on disk. When loading the pool, or after
2291 * a configuration change, we generate all other DTLs from first principles.
2292 */
2293 void
2294 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
2295 {
2296 range_tree_t *rt = vd->vdev_dtl[t];
2297
2298 ASSERT(t < DTL_TYPES);
2299 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
2300 ASSERT(spa_writeable(vd->vdev_spa));
2301
2302 mutex_enter(&vd->vdev_dtl_lock);
2303 if (!range_tree_contains(rt, txg, size))
2304 range_tree_add(rt, txg, size);
2305 mutex_exit(&vd->vdev_dtl_lock);
2306 }
2307
2308 boolean_t
2309 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
2310 {
2311 range_tree_t *rt = vd->vdev_dtl[t];
2312 boolean_t dirty = B_FALSE;
2313
2314 ASSERT(t < DTL_TYPES);
2315 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
2316
2317 /*
2318 * While we are loading the pool, the DTLs have not been loaded yet.
2319 * Ignore the DTLs and try all devices. This avoids a recursive
2320 * mutex enter on the vdev_dtl_lock, and also makes us try hard
2321 * when loading the pool (relying on the checksum to ensure that
2322 * we get the right data -- note that we while loading, we are
2323 * only reading the MOS, which is always checksummed).
2324 */
2325 if (vd->vdev_spa->spa_load_state != SPA_LOAD_NONE)
2326 return (B_FALSE);
2327
2328 mutex_enter(&vd->vdev_dtl_lock);
2329 if (!range_tree_is_empty(rt))
2330 dirty = range_tree_contains(rt, txg, size);
2331 mutex_exit(&vd->vdev_dtl_lock);
2332
2333 return (dirty);
2334 }
2335
2336 boolean_t
2337 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t)
2338 {
2339 range_tree_t *rt = vd->vdev_dtl[t];
2340 boolean_t empty;
2341
2342 mutex_enter(&vd->vdev_dtl_lock);
2343 empty = range_tree_is_empty(rt);
2344 mutex_exit(&vd->vdev_dtl_lock);
2345
2346 return (empty);
2347 }
2348
2349 /*
2350 * Returns the lowest txg in the DTL range.
2351 */
2352 static uint64_t
2353 vdev_dtl_min(vdev_t *vd)
2354 {
2355 range_seg_t *rs;
2356
2357 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
2358 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
2359 ASSERT0(vd->vdev_children);
2360
2361 rs = avl_first(&vd->vdev_dtl[DTL_MISSING]->rt_root);
2362 return (rs->rs_start - 1);
2363 }
2364
2365 /*
2366 * Returns the highest txg in the DTL.
2367 */
2368 static uint64_t
2369 vdev_dtl_max(vdev_t *vd)
2370 {
2371 range_seg_t *rs;
2372
2373 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
2374 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
2375 ASSERT0(vd->vdev_children);
2376
2377 rs = avl_last(&vd->vdev_dtl[DTL_MISSING]->rt_root);
2378 return (rs->rs_end);
2379 }
2380
2381 /*
2382 * Determine if a resilvering vdev should remove any DTL entries from
2383 * its range. If the vdev was resilvering for the entire duration of the
2384 * scan then it should excise that range from its DTLs. Otherwise, this
2385 * vdev is considered partially resilvered and should leave its DTL
2386 * entries intact. The comment in vdev_dtl_reassess() describes how we
2387 * excise the DTLs.
2388 */
2389 static boolean_t
2390 vdev_dtl_should_excise(vdev_t *vd)
2391 {
2392 spa_t *spa = vd->vdev_spa;
2393 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
2394
2395 ASSERT0(scn->scn_phys.scn_errors);
2396 ASSERT0(vd->vdev_children);
2397
2398 if (vd->vdev_state < VDEV_STATE_DEGRADED)
2399 return (B_FALSE);
2400
2401 if (vd->vdev_resilver_txg == 0 ||
2402 range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]))
2403 return (B_TRUE);
2404
2405 /*
2406 * When a resilver is initiated the scan will assign the scn_max_txg
2407 * value to the highest txg value that exists in all DTLs. If this
2408 * device's max DTL is not part of this scan (i.e. it is not in
2409 * the range (scn_min_txg, scn_max_txg] then it is not eligible
2410 * for excision.
2411 */
2412 if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) {
2413 ASSERT3U(scn->scn_phys.scn_min_txg, <=, vdev_dtl_min(vd));
2414 ASSERT3U(scn->scn_phys.scn_min_txg, <, vd->vdev_resilver_txg);
2415 ASSERT3U(vd->vdev_resilver_txg, <=, scn->scn_phys.scn_max_txg);
2416 return (B_TRUE);
2417 }
2418 return (B_FALSE);
2419 }
2420
2421 /*
2422 * Reassess DTLs after a config change or scrub completion.
2423 */
2424 void
2425 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
2426 {
2427 spa_t *spa = vd->vdev_spa;
2428 avl_tree_t reftree;
2429 int minref;
2430
2431 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
2432
2433 for (int c = 0; c < vd->vdev_children; c++)
2434 vdev_dtl_reassess(vd->vdev_child[c], txg,
2435 scrub_txg, scrub_done);
2436
2437 if (vd == spa->spa_root_vdev || !vdev_is_concrete(vd) || vd->vdev_aux)
2438 return;
2439
2440 if (vd->vdev_ops->vdev_op_leaf) {
2441 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
2442
2443 mutex_enter(&vd->vdev_dtl_lock);
2444
2445 /*
2446 * If we've completed a scan cleanly then determine
2447 * if this vdev should remove any DTLs. We only want to
2448 * excise regions on vdevs that were available during
2449 * the entire duration of this scan.
2450 */
2451 if (scrub_txg != 0 &&
2452 (spa->spa_scrub_started ||
2453 (scn != NULL && scn->scn_phys.scn_errors == 0)) &&
2454 vdev_dtl_should_excise(vd)) {
2455 /*
2456 * We completed a scrub up to scrub_txg. If we
2457 * did it without rebooting, then the scrub dtl
2458 * will be valid, so excise the old region and
2459 * fold in the scrub dtl. Otherwise, leave the
2460 * dtl as-is if there was an error.
2461 *
2462 * There's little trick here: to excise the beginning
2463 * of the DTL_MISSING map, we put it into a reference
2464 * tree and then add a segment with refcnt -1 that
2465 * covers the range [0, scrub_txg). This means
2466 * that each txg in that range has refcnt -1 or 0.
2467 * We then add DTL_SCRUB with a refcnt of 2, so that
2468 * entries in the range [0, scrub_txg) will have a
2469 * positive refcnt -- either 1 or 2. We then convert
2470 * the reference tree into the new DTL_MISSING map.
2471 */
2472 space_reftree_create(&reftree);
2473 space_reftree_add_map(&reftree,
2474 vd->vdev_dtl[DTL_MISSING], 1);
2475 space_reftree_add_seg(&reftree, 0, scrub_txg, -1);
2476 space_reftree_add_map(&reftree,
2477 vd->vdev_dtl[DTL_SCRUB], 2);
2478 space_reftree_generate_map(&reftree,
2479 vd->vdev_dtl[DTL_MISSING], 1);
2480 space_reftree_destroy(&reftree);
2481 }
2482 range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
2483 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
2484 range_tree_add, vd->vdev_dtl[DTL_PARTIAL]);
2485 if (scrub_done)
2486 range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
2487 range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
2488 if (!vdev_readable(vd))
2489 range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
2490 else
2491 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
2492 range_tree_add, vd->vdev_dtl[DTL_OUTAGE]);
2493
2494 /*
2495 * If the vdev was resilvering and no longer has any
2496 * DTLs then reset its resilvering flag.
2497 */
2498 if (vd->vdev_resilver_txg != 0 &&
2499 range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]) &&
2500 range_tree_is_empty(vd->vdev_dtl[DTL_OUTAGE]))
2501 vd->vdev_resilver_txg = 0;
2502
2503 mutex_exit(&vd->vdev_dtl_lock);
2504
2505 if (txg != 0)
2506 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
2507 return;
2508 }
2509
2510 mutex_enter(&vd->vdev_dtl_lock);
2511 for (int t = 0; t < DTL_TYPES; t++) {
2512 /* account for child's outage in parent's missing map */
2513 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t;
2514 if (t == DTL_SCRUB)
2515 continue; /* leaf vdevs only */
2516 if (t == DTL_PARTIAL)
2517 minref = 1; /* i.e. non-zero */
2518 else if (vd->vdev_nparity != 0)
2519 minref = vd->vdev_nparity + 1; /* RAID-Z */
2520 else
2521 minref = vd->vdev_children; /* any kind of mirror */
2522 space_reftree_create(&reftree);
2523 for (int c = 0; c < vd->vdev_children; c++) {
2524 vdev_t *cvd = vd->vdev_child[c];
2525 mutex_enter(&cvd->vdev_dtl_lock);
2526 space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1);
2527 mutex_exit(&cvd->vdev_dtl_lock);
2528 }
2529 space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref);
2530 space_reftree_destroy(&reftree);
2531 }
2532 mutex_exit(&vd->vdev_dtl_lock);
2533 }
2534
2535 int
2536 vdev_dtl_load(vdev_t *vd)
2537 {
2538 spa_t *spa = vd->vdev_spa;
2539 objset_t *mos = spa->spa_meta_objset;
2540 int error = 0;
2541
2542 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) {
2543 ASSERT(vdev_is_concrete(vd));
2544
2545 error = space_map_open(&vd->vdev_dtl_sm, mos,
2546 vd->vdev_dtl_object, 0, -1ULL, 0);
2547 if (error)
2548 return (error);
2549 ASSERT(vd->vdev_dtl_sm != NULL);
2550
2551 mutex_enter(&vd->vdev_dtl_lock);
2552
2553 /*
2554 * Now that we've opened the space_map we need to update
2555 * the in-core DTL.
2556 */
2557 space_map_update(vd->vdev_dtl_sm);
2558
2559 error = space_map_load(vd->vdev_dtl_sm,
2560 vd->vdev_dtl[DTL_MISSING], SM_ALLOC);
2561 mutex_exit(&vd->vdev_dtl_lock);
2562
2563 return (error);
2564 }
2565
2566 for (int c = 0; c < vd->vdev_children; c++) {
2567 error = vdev_dtl_load(vd->vdev_child[c]);
2568 if (error != 0)
2569 break;
2570 }
2571
2572 return (error);
2573 }
2574
2575 static void
2576 vdev_zap_allocation_data(vdev_t *vd, dmu_tx_t *tx)
2577 {
2578 spa_t *spa = vd->vdev_spa;
2579 objset_t *mos = spa->spa_meta_objset;
2580 vdev_alloc_bias_t alloc_bias = vd->vdev_alloc_bias;
2581 const char *string;
2582
2583 ASSERT(alloc_bias != VDEV_BIAS_NONE);
2584
2585 string =
2586 (alloc_bias == VDEV_BIAS_LOG) ? VDEV_ALLOC_BIAS_LOG :
2587 (alloc_bias == VDEV_BIAS_SPECIAL) ? VDEV_ALLOC_BIAS_SPECIAL :
2588 (alloc_bias == VDEV_BIAS_DEDUP) ? VDEV_ALLOC_BIAS_DEDUP : NULL;
2589
2590 ASSERT(string != NULL);
2591 VERIFY0(zap_add(mos, vd->vdev_top_zap, VDEV_TOP_ZAP_ALLOCATION_BIAS,
2592 1, strlen(string) + 1, string, tx));
2593
2594 if (alloc_bias == VDEV_BIAS_SPECIAL || alloc_bias == VDEV_BIAS_DEDUP) {
2595 spa_activate_allocation_classes(spa, tx);
2596 }
2597 }
2598
2599 void
2600 vdev_destroy_unlink_zap(vdev_t *vd, uint64_t zapobj, dmu_tx_t *tx)
2601 {
2602 spa_t *spa = vd->vdev_spa;
2603
2604 VERIFY0(zap_destroy(spa->spa_meta_objset, zapobj, tx));
2605 VERIFY0(zap_remove_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps,
2606 zapobj, tx));
2607 }
2608
2609 uint64_t
2610 vdev_create_link_zap(vdev_t *vd, dmu_tx_t *tx)
2611 {
2612 spa_t *spa = vd->vdev_spa;
2613 uint64_t zap = zap_create(spa->spa_meta_objset, DMU_OTN_ZAP_METADATA,
2614 DMU_OT_NONE, 0, tx);
2615
2616 ASSERT(zap != 0);
2617 VERIFY0(zap_add_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps,
2618 zap, tx));
2619
2620 return (zap);
2621 }
2622
2623 void
2624 vdev_construct_zaps(vdev_t *vd, dmu_tx_t *tx)
2625 {
2626 if (vd->vdev_ops != &vdev_hole_ops &&
2627 vd->vdev_ops != &vdev_missing_ops &&
2628 vd->vdev_ops != &vdev_root_ops &&
2629 !vd->vdev_top->vdev_removing) {
2630 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_leaf_zap == 0) {
2631 vd->vdev_leaf_zap = vdev_create_link_zap(vd, tx);
2632 }
2633 if (vd == vd->vdev_top && vd->vdev_top_zap == 0) {
2634 vd->vdev_top_zap = vdev_create_link_zap(vd, tx);
2635 if (vd->vdev_alloc_bias != VDEV_BIAS_NONE)
2636 vdev_zap_allocation_data(vd, tx);
2637 }
2638 }
2639
2640 for (uint64_t i = 0; i < vd->vdev_children; i++) {
2641 vdev_construct_zaps(vd->vdev_child[i], tx);
2642 }
2643 }
2644
2645 void
2646 vdev_dtl_sync(vdev_t *vd, uint64_t txg)
2647 {
2648 spa_t *spa = vd->vdev_spa;
2649 range_tree_t *rt = vd->vdev_dtl[DTL_MISSING];
2650 objset_t *mos = spa->spa_meta_objset;
2651 range_tree_t *rtsync;
2652 dmu_tx_t *tx;
2653 uint64_t object = space_map_object(vd->vdev_dtl_sm);
2654
2655 ASSERT(vdev_is_concrete(vd));
2656 ASSERT(vd->vdev_ops->vdev_op_leaf);
2657
2658 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2659
2660 if (vd->vdev_detached || vd->vdev_top->vdev_removing) {
2661 mutex_enter(&vd->vdev_dtl_lock);
2662 space_map_free(vd->vdev_dtl_sm, tx);
2663 space_map_close(vd->vdev_dtl_sm);
2664 vd->vdev_dtl_sm = NULL;
2665 mutex_exit(&vd->vdev_dtl_lock);
2666
2667 /*
2668 * We only destroy the leaf ZAP for detached leaves or for
2669 * removed log devices. Removed data devices handle leaf ZAP
2670 * cleanup later, once cancellation is no longer possible.
2671 */
2672 if (vd->vdev_leaf_zap != 0 && (vd->vdev_detached ||
2673 vd->vdev_top->vdev_islog)) {
2674 vdev_destroy_unlink_zap(vd, vd->vdev_leaf_zap, tx);
2675 vd->vdev_leaf_zap = 0;
2676 }
2677
2678 dmu_tx_commit(tx);
2679 return;
2680 }
2681
2682 if (vd->vdev_dtl_sm == NULL) {
2683 uint64_t new_object;
2684
2685 new_object = space_map_alloc(mos, vdev_dtl_sm_blksz, tx);
2686 VERIFY3U(new_object, !=, 0);
2687
2688 VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object,
2689 0, -1ULL, 0));
2690 ASSERT(vd->vdev_dtl_sm != NULL);
2691 }
2692
2693 rtsync = range_tree_create(NULL, NULL);
2694
2695 mutex_enter(&vd->vdev_dtl_lock);
2696 range_tree_walk(rt, range_tree_add, rtsync);
2697 mutex_exit(&vd->vdev_dtl_lock);
2698
2699 space_map_truncate(vd->vdev_dtl_sm, vdev_dtl_sm_blksz, tx);
2700 space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, SM_NO_VDEVID, tx);
2701 range_tree_vacate(rtsync, NULL, NULL);
2702
2703 range_tree_destroy(rtsync);
2704
2705 /*
2706 * If the object for the space map has changed then dirty
2707 * the top level so that we update the config.
2708 */
2709 if (object != space_map_object(vd->vdev_dtl_sm)) {
2710 vdev_dbgmsg(vd, "txg %llu, spa %s, DTL old object %llu, "
2711 "new object %llu", (u_longlong_t)txg, spa_name(spa),
2712 (u_longlong_t)object,
2713 (u_longlong_t)space_map_object(vd->vdev_dtl_sm));
2714 vdev_config_dirty(vd->vdev_top);
2715 }
2716
2717 dmu_tx_commit(tx);
2718
2719 mutex_enter(&vd->vdev_dtl_lock);
2720 space_map_update(vd->vdev_dtl_sm);
2721 mutex_exit(&vd->vdev_dtl_lock);
2722 }
2723
2724 /*
2725 * Determine whether the specified vdev can be offlined/detached/removed
2726 * without losing data.
2727 */
2728 boolean_t
2729 vdev_dtl_required(vdev_t *vd)
2730 {
2731 spa_t *spa = vd->vdev_spa;
2732 vdev_t *tvd = vd->vdev_top;
2733 uint8_t cant_read = vd->vdev_cant_read;
2734 boolean_t required;
2735
2736 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2737
2738 if (vd == spa->spa_root_vdev || vd == tvd)
2739 return (B_TRUE);
2740
2741 /*
2742 * Temporarily mark the device as unreadable, and then determine
2743 * whether this results in any DTL outages in the top-level vdev.
2744 * If not, we can safely offline/detach/remove the device.
2745 */
2746 vd->vdev_cant_read = B_TRUE;
2747 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
2748 required = !vdev_dtl_empty(tvd, DTL_OUTAGE);
2749 vd->vdev_cant_read = cant_read;
2750 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
2751
2752 if (!required && zio_injection_enabled)
2753 required = !!zio_handle_device_injection(vd, NULL, ECHILD);
2754
2755 return (required);
2756 }
2757
2758 /*
2759 * Determine if resilver is needed, and if so the txg range.
2760 */
2761 boolean_t
2762 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
2763 {
2764 boolean_t needed = B_FALSE;
2765 uint64_t thismin = UINT64_MAX;
2766 uint64_t thismax = 0;
2767
2768 if (vd->vdev_children == 0) {
2769 mutex_enter(&vd->vdev_dtl_lock);
2770 if (!range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]) &&
2771 vdev_writeable(vd)) {
2772
2773 thismin = vdev_dtl_min(vd);
2774 thismax = vdev_dtl_max(vd);
2775 needed = B_TRUE;
2776 }
2777 mutex_exit(&vd->vdev_dtl_lock);
2778 } else {
2779 for (int c = 0; c < vd->vdev_children; c++) {
2780 vdev_t *cvd = vd->vdev_child[c];
2781 uint64_t cmin, cmax;
2782
2783 if (vdev_resilver_needed(cvd, &cmin, &cmax)) {
2784 thismin = MIN(thismin, cmin);
2785 thismax = MAX(thismax, cmax);
2786 needed = B_TRUE;
2787 }
2788 }
2789 }
2790
2791 if (needed && minp) {
2792 *minp = thismin;
2793 *maxp = thismax;
2794 }
2795 return (needed);
2796 }
2797
2798 /*
2799 * Gets the checkpoint space map object from the vdev's ZAP.
2800 * Returns the spacemap object, or 0 if it wasn't in the ZAP
2801 * or the ZAP doesn't exist yet.
2802 */
2803 int
2804 vdev_checkpoint_sm_object(vdev_t *vd)
2805 {
2806 ASSERT0(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER));
2807 if (vd->vdev_top_zap == 0) {
2808 return (0);
2809 }
2810
2811 uint64_t sm_obj = 0;
2812 int err = zap_lookup(spa_meta_objset(vd->vdev_spa), vd->vdev_top_zap,
2813 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM, sizeof (uint64_t), 1, &sm_obj);
2814
2815 ASSERT(err == 0 || err == ENOENT);
2816
2817 return (sm_obj);
2818 }
2819
2820 int
2821 vdev_load(vdev_t *vd)
2822 {
2823 int error = 0;
2824 /*
2825 * Recursively load all children.
2826 */
2827 for (int c = 0; c < vd->vdev_children; c++) {
2828 error = vdev_load(vd->vdev_child[c]);
2829 if (error != 0) {
2830 return (error);
2831 }
2832 }
2833
2834 vdev_set_deflate_ratio(vd);
2835
2836 /*
2837 * On spa_load path, grab the allocation bias from our zap
2838 */
2839 if (vd == vd->vdev_top && vd->vdev_top_zap != 0) {
2840 spa_t *spa = vd->vdev_spa;
2841 char bias_str[64];
2842
2843 if (zap_lookup(spa->spa_meta_objset, vd->vdev_top_zap,
2844 VDEV_TOP_ZAP_ALLOCATION_BIAS, 1, sizeof (bias_str),
2845 bias_str) == 0) {
2846 ASSERT(vd->vdev_alloc_bias == VDEV_BIAS_NONE);
2847 vd->vdev_alloc_bias = vdev_derive_alloc_bias(bias_str);
2848 }
2849 }
2850
2851 /*
2852 * If this is a top-level vdev, initialize its metaslabs.
2853 */
2854 if (vd == vd->vdev_top && vdev_is_concrete(vd)) {
2855 vdev_metaslab_group_create(vd);
2856
2857 if (vd->vdev_ashift == 0 || vd->vdev_asize == 0) {
2858 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2859 VDEV_AUX_CORRUPT_DATA);
2860 vdev_dbgmsg(vd, "vdev_load: invalid size. ashift=%llu, "
2861 "asize=%llu", (u_longlong_t)vd->vdev_ashift,
2862 (u_longlong_t)vd->vdev_asize);
2863 return (SET_ERROR(ENXIO));
2864 } else if ((error = vdev_metaslab_init(vd, 0)) != 0) {
2865 vdev_dbgmsg(vd, "vdev_load: metaslab_init failed "
2866 "[error=%d]", error);
2867 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2868 VDEV_AUX_CORRUPT_DATA);
2869 return (error);
2870 }
2871
2872 uint64_t checkpoint_sm_obj = vdev_checkpoint_sm_object(vd);
2873 if (checkpoint_sm_obj != 0) {
2874 objset_t *mos = spa_meta_objset(vd->vdev_spa);
2875 ASSERT(vd->vdev_asize != 0);
2876 ASSERT3P(vd->vdev_checkpoint_sm, ==, NULL);
2877
2878 if ((error = space_map_open(&vd->vdev_checkpoint_sm,
2879 mos, checkpoint_sm_obj, 0, vd->vdev_asize,
2880 vd->vdev_ashift))) {
2881 vdev_dbgmsg(vd, "vdev_load: space_map_open "
2882 "failed for checkpoint spacemap (obj %llu) "
2883 "[error=%d]",
2884 (u_longlong_t)checkpoint_sm_obj, error);
2885 return (error);
2886 }
2887 ASSERT3P(vd->vdev_checkpoint_sm, !=, NULL);
2888 space_map_update(vd->vdev_checkpoint_sm);
2889
2890 /*
2891 * Since the checkpoint_sm contains free entries
2892 * exclusively we can use sm_alloc to indicate the
2893 * culmulative checkpointed space that has been freed.
2894 */
2895 vd->vdev_stat.vs_checkpoint_space =
2896 -vd->vdev_checkpoint_sm->sm_alloc;
2897 vd->vdev_spa->spa_checkpoint_info.sci_dspace +=
2898 vd->vdev_stat.vs_checkpoint_space;
2899 }
2900 }
2901
2902 /*
2903 * If this is a leaf vdev, load its DTL.
2904 */
2905 if (vd->vdev_ops->vdev_op_leaf && (error = vdev_dtl_load(vd)) != 0) {
2906 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2907 VDEV_AUX_CORRUPT_DATA);
2908 vdev_dbgmsg(vd, "vdev_load: vdev_dtl_load failed "
2909 "[error=%d]", error);
2910 return (error);
2911 }
2912
2913 uint64_t obsolete_sm_object = vdev_obsolete_sm_object(vd);
2914 if (obsolete_sm_object != 0) {
2915 objset_t *mos = vd->vdev_spa->spa_meta_objset;
2916 ASSERT(vd->vdev_asize != 0);
2917 ASSERT3P(vd->vdev_obsolete_sm, ==, NULL);
2918
2919 if ((error = space_map_open(&vd->vdev_obsolete_sm, mos,
2920 obsolete_sm_object, 0, vd->vdev_asize, 0))) {
2921 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2922 VDEV_AUX_CORRUPT_DATA);
2923 vdev_dbgmsg(vd, "vdev_load: space_map_open failed for "
2924 "obsolete spacemap (obj %llu) [error=%d]",
2925 (u_longlong_t)obsolete_sm_object, error);
2926 return (error);
2927 }
2928 space_map_update(vd->vdev_obsolete_sm);
2929 }
2930
2931 return (0);
2932 }
2933
2934 /*
2935 * The special vdev case is used for hot spares and l2cache devices. Its
2936 * sole purpose it to set the vdev state for the associated vdev. To do this,
2937 * we make sure that we can open the underlying device, then try to read the
2938 * label, and make sure that the label is sane and that it hasn't been
2939 * repurposed to another pool.
2940 */
2941 int
2942 vdev_validate_aux(vdev_t *vd)
2943 {
2944 nvlist_t *label;
2945 uint64_t guid, version;
2946 uint64_t state;
2947
2948 if (!vdev_readable(vd))
2949 return (0);
2950
2951 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) {
2952 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2953 VDEV_AUX_CORRUPT_DATA);
2954 return (-1);
2955 }
2956
2957 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 ||
2958 !SPA_VERSION_IS_SUPPORTED(version) ||
2959 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 ||
2960 guid != vd->vdev_guid ||
2961 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) {
2962 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2963 VDEV_AUX_CORRUPT_DATA);
2964 nvlist_free(label);
2965 return (-1);
2966 }
2967
2968 /*
2969 * We don't actually check the pool state here. If it's in fact in
2970 * use by another pool, we update this fact on the fly when requested.
2971 */
2972 nvlist_free(label);
2973 return (0);
2974 }
2975
2976 /*
2977 * Free the objects used to store this vdev's spacemaps, and the array
2978 * that points to them.
2979 */
2980 void
2981 vdev_destroy_spacemaps(vdev_t *vd, dmu_tx_t *tx)
2982 {
2983 if (vd->vdev_ms_array == 0)
2984 return;
2985
2986 objset_t *mos = vd->vdev_spa->spa_meta_objset;
2987 uint64_t array_count = vd->vdev_asize >> vd->vdev_ms_shift;
2988 size_t array_bytes = array_count * sizeof (uint64_t);
2989 uint64_t *smobj_array = kmem_alloc(array_bytes, KM_SLEEP);
2990 VERIFY0(dmu_read(mos, vd->vdev_ms_array, 0,
2991 array_bytes, smobj_array, 0));
2992
2993 for (uint64_t i = 0; i < array_count; i++) {
2994 uint64_t smobj = smobj_array[i];
2995 if (smobj == 0)
2996 continue;
2997
2998 space_map_free_obj(mos, smobj, tx);
2999 }
3000
3001 kmem_free(smobj_array, array_bytes);
3002 VERIFY0(dmu_object_free(mos, vd->vdev_ms_array, tx));
3003 vd->vdev_ms_array = 0;
3004 }
3005
3006 static void
3007 vdev_remove_empty_log(vdev_t *vd, uint64_t txg)
3008 {
3009 spa_t *spa = vd->vdev_spa;
3010
3011 ASSERT(vd->vdev_islog);
3012 ASSERT(vd == vd->vdev_top);
3013 ASSERT3U(txg, ==, spa_syncing_txg(spa));
3014
3015 if (vd->vdev_ms != NULL) {
3016 metaslab_group_t *mg = vd->vdev_mg;
3017
3018 metaslab_group_histogram_verify(mg);
3019 metaslab_class_histogram_verify(mg->mg_class);
3020
3021 for (int m = 0; m < vd->vdev_ms_count; m++) {
3022 metaslab_t *msp = vd->vdev_ms[m];
3023
3024 if (msp == NULL || msp->ms_sm == NULL)
3025 continue;
3026
3027 mutex_enter(&msp->ms_lock);
3028 /*
3029 * If the metaslab was not loaded when the vdev
3030 * was removed then the histogram accounting may
3031 * not be accurate. Update the histogram information
3032 * here so that we ensure that the metaslab group
3033 * and metaslab class are up-to-date.
3034 */
3035 metaslab_group_histogram_remove(mg, msp);
3036
3037 VERIFY0(space_map_allocated(msp->ms_sm));
3038 space_map_close(msp->ms_sm);
3039 msp->ms_sm = NULL;
3040 mutex_exit(&msp->ms_lock);
3041 }
3042
3043 if (vd->vdev_checkpoint_sm != NULL) {
3044 ASSERT(spa_has_checkpoint(spa));
3045 space_map_close(vd->vdev_checkpoint_sm);
3046 vd->vdev_checkpoint_sm = NULL;
3047 }
3048
3049 metaslab_group_histogram_verify(mg);
3050 metaslab_class_histogram_verify(mg->mg_class);
3051
3052 for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++)
3053 ASSERT0(mg->mg_histogram[i]);
3054 }
3055
3056 dmu_tx_t *tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
3057
3058 vdev_destroy_spacemaps(vd, tx);
3059 if (vd->vdev_top_zap != 0) {
3060 vdev_destroy_unlink_zap(vd, vd->vdev_top_zap, tx);
3061 vd->vdev_top_zap = 0;
3062 }
3063
3064 dmu_tx_commit(tx);
3065 }
3066
3067 void
3068 vdev_sync_done(vdev_t *vd, uint64_t txg)
3069 {
3070 metaslab_t *msp;
3071 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg));
3072
3073 ASSERT(vdev_is_concrete(vd));
3074
3075 while ((msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))
3076 != NULL)
3077 metaslab_sync_done(msp, txg);
3078
3079 if (reassess)
3080 metaslab_sync_reassess(vd->vdev_mg);
3081 }
3082
3083 void
3084 vdev_sync(vdev_t *vd, uint64_t txg)
3085 {
3086 spa_t *spa = vd->vdev_spa;
3087 vdev_t *lvd;
3088 metaslab_t *msp;
3089 dmu_tx_t *tx;
3090
3091 if (range_tree_space(vd->vdev_obsolete_segments) > 0) {
3092 dmu_tx_t *tx;
3093
3094 ASSERT(vd->vdev_removing ||
3095 vd->vdev_ops == &vdev_indirect_ops);
3096
3097 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
3098 vdev_indirect_sync_obsolete(vd, tx);
3099 dmu_tx_commit(tx);
3100
3101 /*
3102 * If the vdev is indirect, it can't have dirty
3103 * metaslabs or DTLs.
3104 */
3105 if (vd->vdev_ops == &vdev_indirect_ops) {
3106 ASSERT(txg_list_empty(&vd->vdev_ms_list, txg));
3107 ASSERT(txg_list_empty(&vd->vdev_dtl_list, txg));
3108 return;
3109 }
3110 }
3111
3112 ASSERT(vdev_is_concrete(vd));
3113
3114 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0 &&
3115 !vd->vdev_removing) {
3116 ASSERT(vd == vd->vdev_top);
3117 ASSERT0(vd->vdev_indirect_config.vic_mapping_object);
3118 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
3119 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset,
3120 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx);
3121 ASSERT(vd->vdev_ms_array != 0);
3122 vdev_config_dirty(vd);
3123 dmu_tx_commit(tx);
3124 }
3125
3126 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) {
3127 metaslab_sync(msp, txg);
3128 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg));
3129 }
3130
3131 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL)
3132 vdev_dtl_sync(lvd, txg);
3133
3134 /*
3135 * If this is an empty log device being removed, destroy the
3136 * metadata associated with it.
3137 */
3138 if (vd->vdev_islog && vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing)
3139 vdev_remove_empty_log(vd, txg);
3140
3141 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg));
3142 }
3143
3144 uint64_t
3145 vdev_psize_to_asize(vdev_t *vd, uint64_t psize)
3146 {
3147 return (vd->vdev_ops->vdev_op_asize(vd, psize));
3148 }
3149
3150 /*
3151 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
3152 * not be opened, and no I/O is attempted.
3153 */
3154 int
3155 vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux)
3156 {
3157 vdev_t *vd, *tvd;
3158
3159 spa_vdev_state_enter(spa, SCL_NONE);
3160
3161 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
3162 return (spa_vdev_state_exit(spa, NULL, ENODEV));
3163
3164 if (!vd->vdev_ops->vdev_op_leaf)
3165 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
3166
3167 tvd = vd->vdev_top;
3168
3169 /*
3170 * We don't directly use the aux state here, but if we do a
3171 * vdev_reopen(), we need this value to be present to remember why we
3172 * were faulted.
3173 */
3174 vd->vdev_label_aux = aux;
3175
3176 /*
3177 * Faulted state takes precedence over degraded.
3178 */
3179 vd->vdev_delayed_close = B_FALSE;
3180 vd->vdev_faulted = 1ULL;
3181 vd->vdev_degraded = 0ULL;
3182 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux);
3183
3184 /*
3185 * If this device has the only valid copy of the data, then
3186 * back off and simply mark the vdev as degraded instead.
3187 */
3188 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) {
3189 vd->vdev_degraded = 1ULL;
3190 vd->vdev_faulted = 0ULL;
3191
3192 /*
3193 * If we reopen the device and it's not dead, only then do we
3194 * mark it degraded.
3195 */
3196 vdev_reopen(tvd);
3197
3198 if (vdev_readable(vd))
3199 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux);
3200 }
3201
3202 return (spa_vdev_state_exit(spa, vd, 0));
3203 }
3204
3205 /*
3206 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
3207 * user that something is wrong. The vdev continues to operate as normal as far
3208 * as I/O is concerned.
3209 */
3210 int
3211 vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux)
3212 {
3213 vdev_t *vd;
3214
3215 spa_vdev_state_enter(spa, SCL_NONE);
3216
3217 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
3218 return (spa_vdev_state_exit(spa, NULL, ENODEV));
3219
3220 if (!vd->vdev_ops->vdev_op_leaf)
3221 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
3222
3223 /*
3224 * If the vdev is already faulted, then don't do anything.
3225 */
3226 if (vd->vdev_faulted || vd->vdev_degraded)
3227 return (spa_vdev_state_exit(spa, NULL, 0));
3228
3229 vd->vdev_degraded = 1ULL;
3230 if (!vdev_is_dead(vd))
3231 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
3232 aux);
3233
3234 return (spa_vdev_state_exit(spa, vd, 0));
3235 }
3236
3237 /*
3238 * Online the given vdev.
3239 *
3240 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
3241 * spare device should be detached when the device finishes resilvering.
3242 * Second, the online should be treated like a 'test' online case, so no FMA
3243 * events are generated if the device fails to open.
3244 */
3245 int
3246 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
3247 {
3248 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev;
3249 boolean_t wasoffline;
3250 vdev_state_t oldstate;
3251
3252 spa_vdev_state_enter(spa, SCL_NONE);
3253
3254 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
3255 return (spa_vdev_state_exit(spa, NULL, ENODEV));
3256
3257 if (!vd->vdev_ops->vdev_op_leaf)
3258 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
3259
3260 wasoffline = (vd->vdev_offline || vd->vdev_tmpoffline);
3261 oldstate = vd->vdev_state;
3262
3263 tvd = vd->vdev_top;
3264 vd->vdev_offline = B_FALSE;
3265 vd->vdev_tmpoffline = B_FALSE;
3266 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE);
3267 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT);
3268
3269 /* XXX - L2ARC 1.0 does not support expansion */
3270 if (!vd->vdev_aux) {
3271 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
3272 pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND);
3273 }
3274
3275 vdev_reopen(tvd);
3276 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE;
3277
3278 if (!vd->vdev_aux) {
3279 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
3280 pvd->vdev_expanding = B_FALSE;
3281 }
3282
3283 if (newstate)
3284 *newstate = vd->vdev_state;
3285 if ((flags & ZFS_ONLINE_UNSPARE) &&
3286 !vdev_is_dead(vd) && vd->vdev_parent &&
3287 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3288 vd->vdev_parent->vdev_child[0] == vd)
3289 vd->vdev_unspare = B_TRUE;
3290
3291 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) {
3292
3293 /* XXX - L2ARC 1.0 does not support expansion */
3294 if (vd->vdev_aux)
3295 return (spa_vdev_state_exit(spa, vd, ENOTSUP));
3296 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
3297 }
3298
3299 /* Restart initializing if necessary */
3300 mutex_enter(&vd->vdev_initialize_lock);
3301 if (vdev_writeable(vd) &&
3302 vd->vdev_initialize_thread == NULL &&
3303 vd->vdev_initialize_state == VDEV_INITIALIZE_ACTIVE) {
3304 (void) vdev_initialize(vd);
3305 }
3306 mutex_exit(&vd->vdev_initialize_lock);
3307
3308 if (wasoffline ||
3309 (oldstate < VDEV_STATE_DEGRADED &&
3310 vd->vdev_state >= VDEV_STATE_DEGRADED))
3311 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_ONLINE);
3312
3313 return (spa_vdev_state_exit(spa, vd, 0));
3314 }
3315
3316 static int
3317 vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags)
3318 {
3319 vdev_t *vd, *tvd;
3320 int error = 0;
3321 uint64_t generation;
3322 metaslab_group_t *mg;
3323
3324 top:
3325 spa_vdev_state_enter(spa, SCL_ALLOC);
3326
3327 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
3328 return (spa_vdev_state_exit(spa, NULL, ENODEV));
3329
3330 if (!vd->vdev_ops->vdev_op_leaf)
3331 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
3332
3333 tvd = vd->vdev_top;
3334 mg = tvd->vdev_mg;
3335 generation = spa->spa_config_generation + 1;
3336
3337 /*
3338 * If the device isn't already offline, try to offline it.
3339 */
3340 if (!vd->vdev_offline) {
3341 /*
3342 * If this device has the only valid copy of some data,
3343 * don't allow it to be offlined. Log devices are always
3344 * expendable.
3345 */
3346 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
3347 vdev_dtl_required(vd))
3348 return (spa_vdev_state_exit(spa, NULL, EBUSY));
3349
3350 /*
3351 * If the top-level is a slog and it has had allocations
3352 * then proceed. We check that the vdev's metaslab group
3353 * is not NULL since it's possible that we may have just
3354 * added this vdev but not yet initialized its metaslabs.
3355 */
3356 if (tvd->vdev_islog && mg != NULL) {
3357 /*
3358 * Prevent any future allocations.
3359 */
3360 metaslab_group_passivate(mg);
3361 (void) spa_vdev_state_exit(spa, vd, 0);
3362
3363 error = spa_reset_logs(spa);
3364
3365 /*
3366 * If the log device was successfully reset but has
3367 * checkpointed data, do not offline it.
3368 */
3369 if (error == 0 &&
3370 tvd->vdev_checkpoint_sm != NULL) {
3371 ASSERT3U(tvd->vdev_checkpoint_sm->sm_alloc,
3372 !=, 0);
3373 error = ZFS_ERR_CHECKPOINT_EXISTS;
3374 }
3375
3376 spa_vdev_state_enter(spa, SCL_ALLOC);
3377
3378 /*
3379 * Check to see if the config has changed.
3380 */
3381 if (error || generation != spa->spa_config_generation) {
3382 metaslab_group_activate(mg);
3383 if (error)
3384 return (spa_vdev_state_exit(spa,
3385 vd, error));
3386 (void) spa_vdev_state_exit(spa, vd, 0);
3387 goto top;
3388 }
3389 ASSERT0(tvd->vdev_stat.vs_alloc);
3390 }
3391
3392 /*
3393 * Offline this device and reopen its top-level vdev.
3394 * If the top-level vdev is a log device then just offline
3395 * it. Otherwise, if this action results in the top-level
3396 * vdev becoming unusable, undo it and fail the request.
3397 */
3398 vd->vdev_offline = B_TRUE;
3399 vdev_reopen(tvd);
3400
3401 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
3402 vdev_is_dead(tvd)) {
3403 vd->vdev_offline = B_FALSE;
3404 vdev_reopen(tvd);
3405 return (spa_vdev_state_exit(spa, NULL, EBUSY));
3406 }
3407
3408 /*
3409 * Add the device back into the metaslab rotor so that
3410 * once we online the device it's open for business.
3411 */
3412 if (tvd->vdev_islog && mg != NULL)
3413 metaslab_group_activate(mg);
3414 }
3415
3416 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY);
3417
3418 return (spa_vdev_state_exit(spa, vd, 0));
3419 }
3420
3421 int
3422 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags)
3423 {
3424 int error;
3425
3426 mutex_enter(&spa->spa_vdev_top_lock);
3427 error = vdev_offline_locked(spa, guid, flags);
3428 mutex_exit(&spa->spa_vdev_top_lock);
3429
3430 return (error);
3431 }
3432
3433 /*
3434 * Clear the error counts associated with this vdev. Unlike vdev_online() and
3435 * vdev_offline(), we assume the spa config is locked. We also clear all
3436 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
3437 */
3438 void
3439 vdev_clear(spa_t *spa, vdev_t *vd)
3440 {
3441 vdev_t *rvd = spa->spa_root_vdev;
3442
3443 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
3444
3445 if (vd == NULL)
3446 vd = rvd;
3447
3448 vd->vdev_stat.vs_read_errors = 0;
3449 vd->vdev_stat.vs_write_errors = 0;
3450 vd->vdev_stat.vs_checksum_errors = 0;
3451
3452 for (int c = 0; c < vd->vdev_children; c++)
3453 vdev_clear(spa, vd->vdev_child[c]);
3454
3455 /*
3456 * It makes no sense to "clear" an indirect vdev.
3457 */
3458 if (!vdev_is_concrete(vd))
3459 return;
3460
3461 /*
3462 * If we're in the FAULTED state or have experienced failed I/O, then
3463 * clear the persistent state and attempt to reopen the device. We
3464 * also mark the vdev config dirty, so that the new faulted state is
3465 * written out to disk.
3466 */
3467 if (vd->vdev_faulted || vd->vdev_degraded ||
3468 !vdev_readable(vd) || !vdev_writeable(vd)) {
3469
3470 /*
3471 * When reopening in reponse to a clear event, it may be due to
3472 * a fmadm repair request. In this case, if the device is
3473 * still broken, we want to still post the ereport again.
3474 */
3475 vd->vdev_forcefault = B_TRUE;
3476
3477 vd->vdev_faulted = vd->vdev_degraded = 0ULL;
3478 vd->vdev_cant_read = B_FALSE;
3479 vd->vdev_cant_write = B_FALSE;
3480
3481 vdev_reopen(vd == rvd ? rvd : vd->vdev_top);
3482
3483 vd->vdev_forcefault = B_FALSE;
3484
3485 if (vd != rvd && vdev_writeable(vd->vdev_top))
3486 vdev_state_dirty(vd->vdev_top);
3487
3488 if (vd->vdev_aux == NULL && !vdev_is_dead(vd))
3489 spa_async_request(spa, SPA_ASYNC_RESILVER);
3490
3491 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_CLEAR);
3492 }
3493
3494 /*
3495 * When clearing a FMA-diagnosed fault, we always want to
3496 * unspare the device, as we assume that the original spare was
3497 * done in response to the FMA fault.
3498 */
3499 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL &&
3500 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3501 vd->vdev_parent->vdev_child[0] == vd)
3502 vd->vdev_unspare = B_TRUE;
3503 }
3504
3505 boolean_t
3506 vdev_is_dead(vdev_t *vd)
3507 {
3508 /*
3509 * Holes and missing devices are always considered "dead".
3510 * This simplifies the code since we don't have to check for
3511 * these types of devices in the various code paths.
3512 * Instead we rely on the fact that we skip over dead devices
3513 * before issuing I/O to them.
3514 */
3515 return (vd->vdev_state < VDEV_STATE_DEGRADED ||
3516 vd->vdev_ops == &vdev_hole_ops ||
3517 vd->vdev_ops == &vdev_missing_ops);
3518 }
3519
3520 boolean_t
3521 vdev_readable(vdev_t *vd)
3522 {
3523 return (!vdev_is_dead(vd) && !vd->vdev_cant_read);
3524 }
3525
3526 boolean_t
3527 vdev_writeable(vdev_t *vd)
3528 {
3529 return (!vdev_is_dead(vd) && !vd->vdev_cant_write &&
3530 vdev_is_concrete(vd));
3531 }
3532
3533 boolean_t
3534 vdev_allocatable(vdev_t *vd)
3535 {
3536 uint64_t state = vd->vdev_state;
3537
3538 /*
3539 * We currently allow allocations from vdevs which may be in the
3540 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
3541 * fails to reopen then we'll catch it later when we're holding
3542 * the proper locks. Note that we have to get the vdev state
3543 * in a local variable because although it changes atomically,
3544 * we're asking two separate questions about it.
3545 */
3546 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) &&
3547 !vd->vdev_cant_write && vdev_is_concrete(vd) &&
3548 vd->vdev_mg->mg_initialized);
3549 }
3550
3551 boolean_t
3552 vdev_accessible(vdev_t *vd, zio_t *zio)
3553 {
3554 ASSERT(zio->io_vd == vd);
3555
3556 if (vdev_is_dead(vd) || vd->vdev_remove_wanted)
3557 return (B_FALSE);
3558
3559 if (zio->io_type == ZIO_TYPE_READ)
3560 return (!vd->vdev_cant_read);
3561
3562 if (zio->io_type == ZIO_TYPE_WRITE)
3563 return (!vd->vdev_cant_write);
3564
3565 return (B_TRUE);
3566 }
3567
3568 boolean_t
3569 vdev_is_spacemap_addressable(vdev_t *vd)
3570 {
3571 if (spa_feature_is_active(vd->vdev_spa, SPA_FEATURE_SPACEMAP_V2))
3572 return (B_TRUE);
3573
3574 /*
3575 * If double-word space map entries are not enabled we assume
3576 * 47 bits of the space map entry are dedicated to the entry's
3577 * offset (see SM_OFFSET_BITS in space_map.h). We then use that
3578 * to calculate the maximum address that can be described by a
3579 * space map entry for the given device.
3580 */
3581 uint64_t shift = vd->vdev_ashift + SM_OFFSET_BITS;
3582
3583 if (shift >= 63) /* detect potential overflow */
3584 return (B_TRUE);
3585
3586 return (vd->vdev_asize < (1ULL << shift));
3587 }
3588
3589 /*
3590 * Get statistics for the given vdev.
3591 */
3592 void
3593 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
3594 {
3595 spa_t *spa = vd->vdev_spa;
3596 vdev_t *rvd = spa->spa_root_vdev;
3597 vdev_t *tvd = vd->vdev_top;
3598
3599 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
3600
3601 mutex_enter(&vd->vdev_stat_lock);
3602 bcopy(&vd->vdev_stat, vs, sizeof (*vs));
3603 vs->vs_timestamp = gethrtime() - vs->vs_timestamp;
3604 vs->vs_state = vd->vdev_state;
3605 vs->vs_rsize = vdev_get_min_asize(vd);
3606 if (vd->vdev_ops->vdev_op_leaf) {
3607 vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
3608 /*
3609 * Report intializing progress. Since we don't have the
3610 * initializing locks held, this is only an estimate (although a
3611 * fairly accurate one).
3612 */
3613 vs->vs_initialize_bytes_done = vd->vdev_initialize_bytes_done;
3614 vs->vs_initialize_bytes_est = vd->vdev_initialize_bytes_est;
3615 vs->vs_initialize_state = vd->vdev_initialize_state;
3616 vs->vs_initialize_action_time = vd->vdev_initialize_action_time;
3617 }
3618 /*
3619 * Report expandable space on top-level, non-auxillary devices only.
3620 * The expandable space is reported in terms of metaslab sized units
3621 * since that determines how much space the pool can expand.
3622 */
3623 if (vd->vdev_aux == NULL && tvd != NULL) {
3624 vs->vs_esize = P2ALIGN(vd->vdev_max_asize - vd->vdev_asize -
3625 spa->spa_bootsize, 1ULL << tvd->vdev_ms_shift);
3626 }
3627 if (vd->vdev_aux == NULL && vd == vd->vdev_top &&
3628 vdev_is_concrete(vd)) {
3629 vs->vs_fragmentation = (vd->vdev_mg != NULL) ?
3630 vd->vdev_mg->mg_fragmentation : 0;
3631 }
3632
3633 /*
3634 * If we're getting stats on the root vdev, aggregate the I/O counts
3635 * over all top-level vdevs (i.e. the direct children of the root).
3636 */
3637 if (vd == rvd) {
3638 for (int c = 0; c < rvd->vdev_children; c++) {
3639 vdev_t *cvd = rvd->vdev_child[c];
3640 vdev_stat_t *cvs = &cvd->vdev_stat;
3641
3642 for (int t = 0; t < ZIO_TYPES; t++) {
3643 vs->vs_ops[t] += cvs->vs_ops[t];
3644 vs->vs_bytes[t] += cvs->vs_bytes[t];
3645 }
3646 cvs->vs_scan_removing = cvd->vdev_removing;
3647 }
3648 }
3649 mutex_exit(&vd->vdev_stat_lock);
3650 }
3651
3652 void
3653 vdev_clear_stats(vdev_t *vd)
3654 {
3655 mutex_enter(&vd->vdev_stat_lock);
3656 vd->vdev_stat.vs_space = 0;
3657 vd->vdev_stat.vs_dspace = 0;
3658 vd->vdev_stat.vs_alloc = 0;
3659 mutex_exit(&vd->vdev_stat_lock);
3660 }
3661
3662 void
3663 vdev_scan_stat_init(vdev_t *vd)
3664 {
3665 vdev_stat_t *vs = &vd->vdev_stat;
3666
3667 for (int c = 0; c < vd->vdev_children; c++)
3668 vdev_scan_stat_init(vd->vdev_child[c]);
3669
3670 mutex_enter(&vd->vdev_stat_lock);
3671 vs->vs_scan_processed = 0;
3672 mutex_exit(&vd->vdev_stat_lock);
3673 }
3674
3675 void
3676 vdev_stat_update(zio_t *zio, uint64_t psize)
3677 {
3678 spa_t *spa = zio->io_spa;
3679 vdev_t *rvd = spa->spa_root_vdev;
3680 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd;
3681 vdev_t *pvd;
3682 uint64_t txg = zio->io_txg;
3683 vdev_stat_t *vs = &vd->vdev_stat;
3684 zio_type_t type = zio->io_type;
3685 int flags = zio->io_flags;
3686
3687 /*
3688 * If this i/o is a gang leader, it didn't do any actual work.
3689 */
3690 if (zio->io_gang_tree)
3691 return;
3692
3693 if (zio->io_error == 0) {
3694 /*
3695 * If this is a root i/o, don't count it -- we've already
3696 * counted the top-level vdevs, and vdev_get_stats() will
3697 * aggregate them when asked. This reduces contention on
3698 * the root vdev_stat_lock and implicitly handles blocks
3699 * that compress away to holes, for which there is no i/o.
3700 * (Holes never create vdev children, so all the counters
3701 * remain zero, which is what we want.)
3702 *
3703 * Note: this only applies to successful i/o (io_error == 0)
3704 * because unlike i/o counts, errors are not additive.
3705 * When reading a ditto block, for example, failure of
3706 * one top-level vdev does not imply a root-level error.
3707 */
3708 if (vd == rvd)
3709 return;
3710
3711 ASSERT(vd == zio->io_vd);
3712
3713 if (flags & ZIO_FLAG_IO_BYPASS)
3714 return;
3715
3716 mutex_enter(&vd->vdev_stat_lock);
3717
3718 if (flags & ZIO_FLAG_IO_REPAIR) {
3719 if (flags & ZIO_FLAG_SCAN_THREAD) {
3720 dsl_scan_phys_t *scn_phys =
3721 &spa->spa_dsl_pool->dp_scan->scn_phys;
3722 uint64_t *processed = &scn_phys->scn_processed;
3723
3724 /* XXX cleanup? */
3725 if (vd->vdev_ops->vdev_op_leaf)
3726 atomic_add_64(processed, psize);
3727 vs->vs_scan_processed += psize;
3728 }
3729
3730 if (flags & ZIO_FLAG_SELF_HEAL)
3731 vs->vs_self_healed += psize;
3732 }
3733
3734 vs->vs_ops[type]++;
3735 vs->vs_bytes[type] += psize;
3736
3737 mutex_exit(&vd->vdev_stat_lock);
3738 return;
3739 }
3740
3741 if (flags & ZIO_FLAG_SPECULATIVE)
3742 return;
3743
3744 /*
3745 * If this is an I/O error that is going to be retried, then ignore the
3746 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
3747 * hard errors, when in reality they can happen for any number of
3748 * innocuous reasons (bus resets, MPxIO link failure, etc).
3749 */
3750 if (zio->io_error == EIO &&
3751 !(zio->io_flags & ZIO_FLAG_IO_RETRY))
3752 return;
3753
3754 /*
3755 * Intent logs writes won't propagate their error to the root
3756 * I/O so don't mark these types of failures as pool-level
3757 * errors.
3758 */
3759 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE))
3760 return;
3761
3762 mutex_enter(&vd->vdev_stat_lock);
3763 if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) {
3764 if (zio->io_error == ECKSUM)
3765 vs->vs_checksum_errors++;
3766 else
3767 vs->vs_read_errors++;
3768 }
3769 if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd))
3770 vs->vs_write_errors++;
3771 mutex_exit(&vd->vdev_stat_lock);
3772
3773 if (spa->spa_load_state == SPA_LOAD_NONE &&
3774 type == ZIO_TYPE_WRITE && txg != 0 &&
3775 (!(flags & ZIO_FLAG_IO_REPAIR) ||
3776 (flags & ZIO_FLAG_SCAN_THREAD) ||
3777 spa->spa_claiming)) {
3778 /*
3779 * This is either a normal write (not a repair), or it's
3780 * a repair induced by the scrub thread, or it's a repair
3781 * made by zil_claim() during spa_load() in the first txg.
3782 * In the normal case, we commit the DTL change in the same
3783 * txg as the block was born. In the scrub-induced repair
3784 * case, we know that scrubs run in first-pass syncing context,
3785 * so we commit the DTL change in spa_syncing_txg(spa).
3786 * In the zil_claim() case, we commit in spa_first_txg(spa).
3787 *
3788 * We currently do not make DTL entries for failed spontaneous
3789 * self-healing writes triggered by normal (non-scrubbing)
3790 * reads, because we have no transactional context in which to
3791 * do so -- and it's not clear that it'd be desirable anyway.
3792 */
3793 if (vd->vdev_ops->vdev_op_leaf) {
3794 uint64_t commit_txg = txg;
3795 if (flags & ZIO_FLAG_SCAN_THREAD) {
3796 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
3797 ASSERT(spa_sync_pass(spa) == 1);
3798 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1);
3799 commit_txg = spa_syncing_txg(spa);
3800 } else if (spa->spa_claiming) {
3801 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
3802 commit_txg = spa_first_txg(spa);
3803 }
3804 ASSERT(commit_txg >= spa_syncing_txg(spa));
3805 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1))
3806 return;
3807 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
3808 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1);
3809 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg);
3810 }
3811 if (vd != rvd)
3812 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1);
3813 }
3814 }
3815
3816 int64_t
3817 vdev_deflated_space(vdev_t *vd, int64_t space)
3818 {
3819 ASSERT((space & (SPA_MINBLOCKSIZE-1)) == 0);
3820 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache);
3821
3822 return ((space >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio);
3823 }
3824
3825 /*
3826 * Update the in-core space usage stats for this vdev and the root vdev.
3827 */
3828 void
3829 vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta,
3830 int64_t space_delta)
3831 {
3832 int64_t dspace_delta;
3833 spa_t *spa = vd->vdev_spa;
3834 vdev_t *rvd = spa->spa_root_vdev;
3835
3836 ASSERT(vd == vd->vdev_top);
3837
3838 /*
3839 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
3840 * factor. We must calculate this here and not at the root vdev
3841 * because the root vdev's psize-to-asize is simply the max of its
3842 * childrens', thus not accurate enough for us.
3843 */
3844 dspace_delta = vdev_deflated_space(vd, space_delta);
3845
3846 mutex_enter(&vd->vdev_stat_lock);
3847 vd->vdev_stat.vs_alloc += alloc_delta;
3848 vd->vdev_stat.vs_space += space_delta;
3849 vd->vdev_stat.vs_dspace += dspace_delta;
3850 mutex_exit(&vd->vdev_stat_lock);
3851
3852 /* every class but log contributes to root space stats */
3853 if (vd->vdev_mg != NULL && !vd->vdev_islog) {
3854 mutex_enter(&rvd->vdev_stat_lock);
3855 rvd->vdev_stat.vs_alloc += alloc_delta;
3856 rvd->vdev_stat.vs_space += space_delta;
3857 rvd->vdev_stat.vs_dspace += dspace_delta;
3858 mutex_exit(&rvd->vdev_stat_lock);
3859 }
3860 /* Note: metaslab_class_space_update moved to metaslab_space_update */
3861 }
3862
3863 /*
3864 * Mark a top-level vdev's config as dirty, placing it on the dirty list
3865 * so that it will be written out next time the vdev configuration is synced.
3866 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
3867 */
3868 void
3869 vdev_config_dirty(vdev_t *vd)
3870 {
3871 spa_t *spa = vd->vdev_spa;
3872 vdev_t *rvd = spa->spa_root_vdev;
3873 int c;
3874
3875 ASSERT(spa_writeable(spa));
3876
3877 /*
3878 * If this is an aux vdev (as with l2cache and spare devices), then we
3879 * update the vdev config manually and set the sync flag.
3880 */
3881 if (vd->vdev_aux != NULL) {
3882 spa_aux_vdev_t *sav = vd->vdev_aux;
3883 nvlist_t **aux;
3884 uint_t naux;
3885
3886 for (c = 0; c < sav->sav_count; c++) {
3887 if (sav->sav_vdevs[c] == vd)
3888 break;
3889 }
3890
3891 if (c == sav->sav_count) {
3892 /*
3893 * We're being removed. There's nothing more to do.
3894 */
3895 ASSERT(sav->sav_sync == B_TRUE);
3896 return;
3897 }
3898
3899 sav->sav_sync = B_TRUE;
3900
3901 if (nvlist_lookup_nvlist_array(sav->sav_config,
3902 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) {
3903 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
3904 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0);
3905 }
3906
3907 ASSERT(c < naux);
3908
3909 /*
3910 * Setting the nvlist in the middle if the array is a little
3911 * sketchy, but it will work.
3912 */
3913 nvlist_free(aux[c]);
3914 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0);
3915
3916 return;
3917 }
3918
3919 /*
3920 * The dirty list is protected by the SCL_CONFIG lock. The caller
3921 * must either hold SCL_CONFIG as writer, or must be the sync thread
3922 * (which holds SCL_CONFIG as reader). There's only one sync thread,
3923 * so this is sufficient to ensure mutual exclusion.
3924 */
3925 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
3926 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3927 spa_config_held(spa, SCL_CONFIG, RW_READER)));
3928
3929 if (vd == rvd) {
3930 for (c = 0; c < rvd->vdev_children; c++)
3931 vdev_config_dirty(rvd->vdev_child[c]);
3932 } else {
3933 ASSERT(vd == vd->vdev_top);
3934
3935 if (!list_link_active(&vd->vdev_config_dirty_node) &&
3936 vdev_is_concrete(vd)) {
3937 list_insert_head(&spa->spa_config_dirty_list, vd);
3938 }
3939 }
3940 }
3941
3942 void
3943 vdev_config_clean(vdev_t *vd)
3944 {
3945 spa_t *spa = vd->vdev_spa;
3946
3947 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
3948 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3949 spa_config_held(spa, SCL_CONFIG, RW_READER)));
3950
3951 ASSERT(list_link_active(&vd->vdev_config_dirty_node));
3952 list_remove(&spa->spa_config_dirty_list, vd);
3953 }
3954
3955 /*
3956 * Mark a top-level vdev's state as dirty, so that the next pass of
3957 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
3958 * the state changes from larger config changes because they require
3959 * much less locking, and are often needed for administrative actions.
3960 */
3961 void
3962 vdev_state_dirty(vdev_t *vd)
3963 {
3964 spa_t *spa = vd->vdev_spa;
3965
3966 ASSERT(spa_writeable(spa));
3967 ASSERT(vd == vd->vdev_top);
3968
3969 /*
3970 * The state list is protected by the SCL_STATE lock. The caller
3971 * must either hold SCL_STATE as writer, or must be the sync thread
3972 * (which holds SCL_STATE as reader). There's only one sync thread,
3973 * so this is sufficient to ensure mutual exclusion.
3974 */
3975 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
3976 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3977 spa_config_held(spa, SCL_STATE, RW_READER)));
3978
3979 if (!list_link_active(&vd->vdev_state_dirty_node) &&
3980 vdev_is_concrete(vd))
3981 list_insert_head(&spa->spa_state_dirty_list, vd);
3982 }
3983
3984 void
3985 vdev_state_clean(vdev_t *vd)
3986 {
3987 spa_t *spa = vd->vdev_spa;
3988
3989 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
3990 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3991 spa_config_held(spa, SCL_STATE, RW_READER)));
3992
3993 ASSERT(list_link_active(&vd->vdev_state_dirty_node));
3994 list_remove(&spa->spa_state_dirty_list, vd);
3995 }
3996
3997 /*
3998 * Propagate vdev state up from children to parent.
3999 */
4000 void
4001 vdev_propagate_state(vdev_t *vd)
4002 {
4003 spa_t *spa = vd->vdev_spa;
4004 vdev_t *rvd = spa->spa_root_vdev;
4005 int degraded = 0, faulted = 0;
4006 int corrupted = 0;
4007 vdev_t *child;
4008
4009 if (vd->vdev_children > 0) {
4010 for (int c = 0; c < vd->vdev_children; c++) {
4011 child = vd->vdev_child[c];
4012
4013 /*
4014 * Don't factor holes or indirect vdevs into the
4015 * decision.
4016 */
4017 if (!vdev_is_concrete(child))
4018 continue;
4019
4020 if (!vdev_readable(child) ||
4021 (!vdev_writeable(child) && spa_writeable(spa))) {
4022 /*
4023 * Root special: if there is a top-level log
4024 * device, treat the root vdev as if it were
4025 * degraded.
4026 */
4027 if (child->vdev_islog && vd == rvd)
4028 degraded++;
4029 else
4030 faulted++;
4031 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) {
4032 degraded++;
4033 }
4034
4035 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA)
4036 corrupted++;
4037 }
4038
4039 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded);
4040
4041 /*
4042 * Root special: if there is a top-level vdev that cannot be
4043 * opened due to corrupted metadata, then propagate the root
4044 * vdev's aux state as 'corrupt' rather than 'insufficient
4045 * replicas'.
4046 */
4047 if (corrupted && vd == rvd &&
4048 rvd->vdev_state == VDEV_STATE_CANT_OPEN)
4049 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN,
4050 VDEV_AUX_CORRUPT_DATA);
4051 }
4052
4053 if (vd->vdev_parent)
4054 vdev_propagate_state(vd->vdev_parent);
4055 }
4056
4057 /*
4058 * Set a vdev's state. If this is during an open, we don't update the parent
4059 * state, because we're in the process of opening children depth-first.
4060 * Otherwise, we propagate the change to the parent.
4061 *
4062 * If this routine places a device in a faulted state, an appropriate ereport is
4063 * generated.
4064 */
4065 void
4066 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
4067 {
4068 uint64_t save_state;
4069 spa_t *spa = vd->vdev_spa;
4070
4071 if (state == vd->vdev_state) {
4072 vd->vdev_stat.vs_aux = aux;
4073 return;
4074 }
4075
4076 save_state = vd->vdev_state;
4077
4078 vd->vdev_state = state;
4079 vd->vdev_stat.vs_aux = aux;
4080
4081 /*
4082 * If we are setting the vdev state to anything but an open state, then
4083 * always close the underlying device unless the device has requested
4084 * a delayed close (i.e. we're about to remove or fault the device).
4085 * Otherwise, we keep accessible but invalid devices open forever.
4086 * We don't call vdev_close() itself, because that implies some extra
4087 * checks (offline, etc) that we don't want here. This is limited to
4088 * leaf devices, because otherwise closing the device will affect other
4089 * children.
4090 */
4091 if (!vd->vdev_delayed_close && vdev_is_dead(vd) &&
4092 vd->vdev_ops->vdev_op_leaf)
4093 vd->vdev_ops->vdev_op_close(vd);
4094
4095 /*
4096 * If we have brought this vdev back into service, we need
4097 * to notify fmd so that it can gracefully repair any outstanding
4098 * cases due to a missing device. We do this in all cases, even those
4099 * that probably don't correlate to a repaired fault. This is sure to
4100 * catch all cases, and we let the zfs-retire agent sort it out. If
4101 * this is a transient state it's OK, as the retire agent will
4102 * double-check the state of the vdev before repairing it.
4103 */
4104 if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf &&
4105 vd->vdev_prevstate != state)
4106 zfs_post_state_change(spa, vd);
4107
4108 if (vd->vdev_removed &&
4109 state == VDEV_STATE_CANT_OPEN &&
4110 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) {
4111 /*
4112 * If the previous state is set to VDEV_STATE_REMOVED, then this
4113 * device was previously marked removed and someone attempted to
4114 * reopen it. If this failed due to a nonexistent device, then
4115 * keep the device in the REMOVED state. We also let this be if
4116 * it is one of our special test online cases, which is only
4117 * attempting to online the device and shouldn't generate an FMA
4118 * fault.
4119 */
4120 vd->vdev_state = VDEV_STATE_REMOVED;
4121 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
4122 } else if (state == VDEV_STATE_REMOVED) {
4123 vd->vdev_removed = B_TRUE;
4124 } else if (state == VDEV_STATE_CANT_OPEN) {
4125 /*
4126 * If we fail to open a vdev during an import or recovery, we
4127 * mark it as "not available", which signifies that it was
4128 * never there to begin with. Failure to open such a device
4129 * is not considered an error.
4130 */
4131 if ((spa_load_state(spa) == SPA_LOAD_IMPORT ||
4132 spa_load_state(spa) == SPA_LOAD_RECOVER) &&
4133 vd->vdev_ops->vdev_op_leaf)
4134 vd->vdev_not_present = 1;
4135
4136 /*
4137 * Post the appropriate ereport. If the 'prevstate' field is
4138 * set to something other than VDEV_STATE_UNKNOWN, it indicates
4139 * that this is part of a vdev_reopen(). In this case, we don't
4140 * want to post the ereport if the device was already in the
4141 * CANT_OPEN state beforehand.
4142 *
4143 * If the 'checkremove' flag is set, then this is an attempt to
4144 * online the device in response to an insertion event. If we
4145 * hit this case, then we have detected an insertion event for a
4146 * faulted or offline device that wasn't in the removed state.
4147 * In this scenario, we don't post an ereport because we are
4148 * about to replace the device, or attempt an online with
4149 * vdev_forcefault, which will generate the fault for us.
4150 */
4151 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) &&
4152 !vd->vdev_not_present && !vd->vdev_checkremove &&
4153 vd != spa->spa_root_vdev) {
4154 const char *class;
4155
4156 switch (aux) {
4157 case VDEV_AUX_OPEN_FAILED:
4158 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED;
4159 break;
4160 case VDEV_AUX_CORRUPT_DATA:
4161 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA;
4162 break;
4163 case VDEV_AUX_NO_REPLICAS:
4164 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS;
4165 break;
4166 case VDEV_AUX_BAD_GUID_SUM:
4167 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM;
4168 break;
4169 case VDEV_AUX_TOO_SMALL:
4170 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL;
4171 break;
4172 case VDEV_AUX_BAD_LABEL:
4173 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL;
4174 break;
4175 default:
4176 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN;
4177 }
4178
4179 zfs_ereport_post(class, spa, vd, NULL, save_state, 0);
4180 }
4181
4182 /* Erase any notion of persistent removed state */
4183 vd->vdev_removed = B_FALSE;
4184 } else {
4185 vd->vdev_removed = B_FALSE;
4186 }
4187
4188 if (!isopen && vd->vdev_parent)
4189 vdev_propagate_state(vd->vdev_parent);
4190 }
4191
4192 boolean_t
4193 vdev_children_are_offline(vdev_t *vd)
4194 {
4195 ASSERT(!vd->vdev_ops->vdev_op_leaf);
4196
4197 for (uint64_t i = 0; i < vd->vdev_children; i++) {
4198 if (vd->vdev_child[i]->vdev_state != VDEV_STATE_OFFLINE)
4199 return (B_FALSE);
4200 }
4201
4202 return (B_TRUE);
4203 }
4204
4205 /*
4206 * Check the vdev configuration to ensure that it's capable of supporting
4207 * a root pool. We do not support partial configuration.
4208 * In addition, only a single top-level vdev is allowed.
4209 */
4210 boolean_t
4211 vdev_is_bootable(vdev_t *vd)
4212 {
4213 if (!vd->vdev_ops->vdev_op_leaf) {
4214 char *vdev_type = vd->vdev_ops->vdev_op_type;
4215
4216 if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 &&
4217 vd->vdev_children > 1) {
4218 return (B_FALSE);
4219 } else if (strcmp(vdev_type, VDEV_TYPE_MISSING) == 0 ||
4220 strcmp(vdev_type, VDEV_TYPE_INDIRECT) == 0) {
4221 return (B_FALSE);
4222 }
4223 }
4224
4225 for (int c = 0; c < vd->vdev_children; c++) {
4226 if (!vdev_is_bootable(vd->vdev_child[c]))
4227 return (B_FALSE);
4228 }
4229 return (B_TRUE);
4230 }
4231
4232 boolean_t
4233 vdev_is_concrete(vdev_t *vd)
4234 {
4235 vdev_ops_t *ops = vd->vdev_ops;
4236 if (ops == &vdev_indirect_ops || ops == &vdev_hole_ops ||
4237 ops == &vdev_missing_ops || ops == &vdev_root_ops) {
4238 return (B_FALSE);
4239 } else {
4240 return (B_TRUE);
4241 }
4242 }
4243
4244 /*
4245 * Determine if a log device has valid content. If the vdev was
4246 * removed or faulted in the MOS config then we know that
4247 * the content on the log device has already been written to the pool.
4248 */
4249 boolean_t
4250 vdev_log_state_valid(vdev_t *vd)
4251 {
4252 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted &&
4253 !vd->vdev_removed)
4254 return (B_TRUE);
4255
4256 for (int c = 0; c < vd->vdev_children; c++)
4257 if (vdev_log_state_valid(vd->vdev_child[c]))
4258 return (B_TRUE);
4259
4260 return (B_FALSE);
4261 }
4262
4263 /*
4264 * Expand a vdev if possible.
4265 */
4266 void
4267 vdev_expand(vdev_t *vd, uint64_t txg)
4268 {
4269 ASSERT(vd->vdev_top == vd);
4270 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
4271 ASSERT(vdev_is_concrete(vd));
4272
4273 vdev_set_deflate_ratio(vd);
4274
4275 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count &&
4276 vdev_is_concrete(vd)) {
4277 vdev_metaslab_group_create(vd);
4278 VERIFY(vdev_metaslab_init(vd, txg) == 0);
4279 vdev_config_dirty(vd);
4280 }
4281 }
4282
4283 /*
4284 * Split a vdev.
4285 */
4286 void
4287 vdev_split(vdev_t *vd)
4288 {
4289 vdev_t *cvd, *pvd = vd->vdev_parent;
4290
4291 vdev_remove_child(pvd, vd);
4292 vdev_compact_children(pvd);
4293
4294 cvd = pvd->vdev_child[0];
4295 if (pvd->vdev_children == 1) {
4296 vdev_remove_parent(cvd);
4297 cvd->vdev_splitting = B_TRUE;
4298 }
4299 vdev_propagate_state(cvd);
4300 }
4301
4302 void
4303 vdev_deadman(vdev_t *vd)
4304 {
4305 for (int c = 0; c < vd->vdev_children; c++) {
4306 vdev_t *cvd = vd->vdev_child[c];
4307
4308 vdev_deadman(cvd);
4309 }
4310
4311 if (vd->vdev_ops->vdev_op_leaf) {
4312 vdev_queue_t *vq = &vd->vdev_queue;
4313
4314 mutex_enter(&vq->vq_lock);
4315 if (avl_numnodes(&vq->vq_active_tree) > 0) {
4316 spa_t *spa = vd->vdev_spa;
4317 zio_t *fio;
4318 uint64_t delta;
4319
4320 /*
4321 * Look at the head of all the pending queues,
4322 * if any I/O has been outstanding for longer than
4323 * the spa_deadman_synctime we panic the system.
4324 */
4325 fio = avl_first(&vq->vq_active_tree);
4326 delta = gethrtime() - fio->io_timestamp;
4327 if (delta > spa_deadman_synctime(spa)) {
4328 vdev_dbgmsg(vd, "SLOW IO: zio timestamp "
4329 "%lluns, delta %lluns, last io %lluns",
4330 fio->io_timestamp, (u_longlong_t)delta,
4331 vq->vq_io_complete_ts);
4332 fm_panic("I/O to pool '%s' appears to be "
4333 "hung.", spa_name(spa));
4334 }
4335 }
4336 mutex_exit(&vq->vq_lock);
4337 }
4338 }