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