1/*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22/*
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
|
25 * Copyright (c) 2011 by Delphix. All rights reserved.
|
| 25 * Copyright (c) 2012 by Delphix. All rights reserved. |
26 */
27
28#include <sys/zfs_context.h>
29#include <sys/fm/fs/zfs.h>
30#include <sys/spa.h>
31#include <sys/spa_impl.h>
32#include <sys/dmu.h>
33#include <sys/dmu_tx.h>
34#include <sys/vdev_impl.h>
35#include <sys/uberblock_impl.h>
36#include <sys/metaslab.h>
37#include <sys/metaslab_impl.h>
38#include <sys/space_map.h>
39#include <sys/zio.h>
40#include <sys/zap.h>
41#include <sys/fs/zfs.h>
42#include <sys/arc.h>
43#include <sys/zil.h>
44#include <sys/dsl_scan.h>
45
46SYSCTL_DECL(_vfs_zfs);
47SYSCTL_NODE(_vfs_zfs, OID_AUTO, vdev, CTLFLAG_RW, 0, "ZFS VDEV");
48
49/*
50 * Virtual device management.
51 */
52
53static vdev_ops_t *vdev_ops_table[] = {
54 &vdev_root_ops,
55 &vdev_raidz_ops,
56 &vdev_mirror_ops,
57 &vdev_replacing_ops,
58 &vdev_spare_ops,
59#ifdef _KERNEL
60 &vdev_geom_ops,
61#else
62 &vdev_disk_ops,
63#endif
64 &vdev_file_ops,
65 &vdev_missing_ops,
66 &vdev_hole_ops,
67 NULL
68};
69
70/* maximum scrub/resilver I/O queue per leaf vdev */
71int zfs_scrub_limit = 10;
72
73TUNABLE_INT("vfs.zfs.scrub_limit", &zfs_scrub_limit);
74SYSCTL_INT(_vfs_zfs, OID_AUTO, scrub_limit, CTLFLAG_RDTUN, &zfs_scrub_limit, 0,
75 "Maximum scrub/resilver I/O queue");
76
77/*
78 * Given a vdev type, return the appropriate ops vector.
79 */
80static vdev_ops_t *
81vdev_getops(const char *type)
82{
83 vdev_ops_t *ops, **opspp;
84
85 for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++)
86 if (strcmp(ops->vdev_op_type, type) == 0)
87 break;
88
89 return (ops);
90}
91
92/*
93 * Default asize function: return the MAX of psize with the asize of
94 * all children. This is what's used by anything other than RAID-Z.
95 */
96uint64_t
97vdev_default_asize(vdev_t *vd, uint64_t psize)
98{
99 uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift);
100 uint64_t csize;
101
102 for (int c = 0; c < vd->vdev_children; c++) {
103 csize = vdev_psize_to_asize(vd->vdev_child[c], psize);
104 asize = MAX(asize, csize);
105 }
106
107 return (asize);
108}
109
110/*
111 * Get the minimum allocatable size. We define the allocatable size as
112 * the vdev's asize rounded to the nearest metaslab. This allows us to
113 * replace or attach devices which don't have the same physical size but
114 * can still satisfy the same number of allocations.
115 */
116uint64_t
117vdev_get_min_asize(vdev_t *vd)
118{
119 vdev_t *pvd = vd->vdev_parent;
120
121 /*
|
122 * The our parent is NULL (inactive spare or cache) or is the root,
|
| 122 * If our parent is NULL (inactive spare or cache) or is the root, |
123 * just return our own asize.
124 */
125 if (pvd == NULL)
126 return (vd->vdev_asize);
127
128 /*
129 * The top-level vdev just returns the allocatable size rounded
130 * to the nearest metaslab.
131 */
132 if (vd == vd->vdev_top)
133 return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift));
134
135 /*
136 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
137 * so each child must provide at least 1/Nth of its asize.
138 */
139 if (pvd->vdev_ops == &vdev_raidz_ops)
140 return (pvd->vdev_min_asize / pvd->vdev_children);
141
142 return (pvd->vdev_min_asize);
143}
144
145void
146vdev_set_min_asize(vdev_t *vd)
147{
148 vd->vdev_min_asize = vdev_get_min_asize(vd);
149
150 for (int c = 0; c < vd->vdev_children; c++)
151 vdev_set_min_asize(vd->vdev_child[c]);
152}
153
154vdev_t *
155vdev_lookup_top(spa_t *spa, uint64_t vdev)
156{
157 vdev_t *rvd = spa->spa_root_vdev;
158
159 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
160
161 if (vdev < rvd->vdev_children) {
162 ASSERT(rvd->vdev_child[vdev] != NULL);
163 return (rvd->vdev_child[vdev]);
164 }
165
166 return (NULL);
167}
168
169vdev_t *
170vdev_lookup_by_guid(vdev_t *vd, uint64_t guid)
171{
172 vdev_t *mvd;
173
174 if (vd->vdev_guid == guid)
175 return (vd);
176
177 for (int c = 0; c < vd->vdev_children; c++)
178 if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) !=
179 NULL)
180 return (mvd);
181
182 return (NULL);
183}
184
185void
186vdev_add_child(vdev_t *pvd, vdev_t *cvd)
187{
188 size_t oldsize, newsize;
189 uint64_t id = cvd->vdev_id;
190 vdev_t **newchild;
191
192 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
193 ASSERT(cvd->vdev_parent == NULL);
194
195 cvd->vdev_parent = pvd;
196
197 if (pvd == NULL)
198 return;
199
200 ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL);
201
202 oldsize = pvd->vdev_children * sizeof (vdev_t *);
203 pvd->vdev_children = MAX(pvd->vdev_children, id + 1);
204 newsize = pvd->vdev_children * sizeof (vdev_t *);
205
206 newchild = kmem_zalloc(newsize, KM_SLEEP);
207 if (pvd->vdev_child != NULL) {
208 bcopy(pvd->vdev_child, newchild, oldsize);
209 kmem_free(pvd->vdev_child, oldsize);
210 }
211
212 pvd->vdev_child = newchild;
213 pvd->vdev_child[id] = cvd;
214
215 cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd);
216 ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL);
217
218 /*
219 * Walk up all ancestors to update guid sum.
220 */
221 for (; pvd != NULL; pvd = pvd->vdev_parent)
222 pvd->vdev_guid_sum += cvd->vdev_guid_sum;
223}
224
225void
226vdev_remove_child(vdev_t *pvd, vdev_t *cvd)
227{
228 int c;
229 uint_t id = cvd->vdev_id;
230
231 ASSERT(cvd->vdev_parent == pvd);
232
233 if (pvd == NULL)
234 return;
235
236 ASSERT(id < pvd->vdev_children);
237 ASSERT(pvd->vdev_child[id] == cvd);
238
239 pvd->vdev_child[id] = NULL;
240 cvd->vdev_parent = NULL;
241
242 for (c = 0; c < pvd->vdev_children; c++)
243 if (pvd->vdev_child[c])
244 break;
245
246 if (c == pvd->vdev_children) {
247 kmem_free(pvd->vdev_child, c * sizeof (vdev_t *));
248 pvd->vdev_child = NULL;
249 pvd->vdev_children = 0;
250 }
251
252 /*
253 * Walk up all ancestors to update guid sum.
254 */
255 for (; pvd != NULL; pvd = pvd->vdev_parent)
256 pvd->vdev_guid_sum -= cvd->vdev_guid_sum;
257}
258
259/*
260 * Remove any holes in the child array.
261 */
262void
263vdev_compact_children(vdev_t *pvd)
264{
265 vdev_t **newchild, *cvd;
266 int oldc = pvd->vdev_children;
267 int newc;
268
269 ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
270
271 for (int c = newc = 0; c < oldc; c++)
272 if (pvd->vdev_child[c])
273 newc++;
274
275 newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP);
276
277 for (int c = newc = 0; c < oldc; c++) {
278 if ((cvd = pvd->vdev_child[c]) != NULL) {
279 newchild[newc] = cvd;
280 cvd->vdev_id = newc++;
281 }
282 }
283
284 kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *));
285 pvd->vdev_child = newchild;
286 pvd->vdev_children = newc;
287}
288
289/*
290 * Allocate and minimally initialize a vdev_t.
291 */
292vdev_t *
293vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops)
294{
295 vdev_t *vd;
296
297 vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP);
298
299 if (spa->spa_root_vdev == NULL) {
300 ASSERT(ops == &vdev_root_ops);
301 spa->spa_root_vdev = vd;
302 spa->spa_load_guid = spa_generate_guid(NULL);
303 }
304
305 if (guid == 0 && ops != &vdev_hole_ops) {
306 if (spa->spa_root_vdev == vd) {
307 /*
308 * The root vdev's guid will also be the pool guid,
309 * which must be unique among all pools.
310 */
311 guid = spa_generate_guid(NULL);
312 } else {
313 /*
314 * Any other vdev's guid must be unique within the pool.
315 */
316 guid = spa_generate_guid(spa);
317 }
318 ASSERT(!spa_guid_exists(spa_guid(spa), guid));
319 }
320
321 vd->vdev_spa = spa;
322 vd->vdev_id = id;
323 vd->vdev_guid = guid;
324 vd->vdev_guid_sum = guid;
325 vd->vdev_ops = ops;
326 vd->vdev_state = VDEV_STATE_CLOSED;
327 vd->vdev_ishole = (ops == &vdev_hole_ops);
328
329 mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL);
330 mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL);
331 mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL);
332 for (int t = 0; t < DTL_TYPES; t++) {
333 space_map_create(&vd->vdev_dtl[t], 0, -1ULL, 0,
334 &vd->vdev_dtl_lock);
335 }
336 txg_list_create(&vd->vdev_ms_list,
337 offsetof(struct metaslab, ms_txg_node));
338 txg_list_create(&vd->vdev_dtl_list,
339 offsetof(struct vdev, vdev_dtl_node));
340 vd->vdev_stat.vs_timestamp = gethrtime();
341 vdev_queue_init(vd);
342 vdev_cache_init(vd);
343
344 return (vd);
345}
346
347/*
348 * Allocate a new vdev. The 'alloctype' is used to control whether we are
349 * creating a new vdev or loading an existing one - the behavior is slightly
350 * different for each case.
351 */
352int
353vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
354 int alloctype)
355{
356 vdev_ops_t *ops;
357 char *type;
358 uint64_t guid = 0, islog, nparity;
359 vdev_t *vd;
360
361 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
362
363 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
364 return (EINVAL);
365
366 if ((ops = vdev_getops(type)) == NULL)
367 return (EINVAL);
368
369 /*
370 * If this is a load, get the vdev guid from the nvlist.
371 * Otherwise, vdev_alloc_common() will generate one for us.
372 */
373 if (alloctype == VDEV_ALLOC_LOAD) {
374 uint64_t label_id;
375
376 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) ||
377 label_id != id)
378 return (EINVAL);
379
380 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
381 return (EINVAL);
382 } else if (alloctype == VDEV_ALLOC_SPARE) {
383 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
384 return (EINVAL);
385 } else if (alloctype == VDEV_ALLOC_L2CACHE) {
386 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
387 return (EINVAL);
388 } else if (alloctype == VDEV_ALLOC_ROOTPOOL) {
389 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
390 return (EINVAL);
391 }
392
393 /*
394 * The first allocated vdev must be of type 'root'.
395 */
396 if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL)
397 return (EINVAL);
398
399 /*
400 * Determine whether we're a log vdev.
401 */
402 islog = 0;
403 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog);
404 if (islog && spa_version(spa) < SPA_VERSION_SLOGS)
405 return (ENOTSUP);
406
407 if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES)
408 return (ENOTSUP);
409
410 /*
411 * Set the nparity property for RAID-Z vdevs.
412 */
413 nparity = -1ULL;
414 if (ops == &vdev_raidz_ops) {
415 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
416 &nparity) == 0) {
417 if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY)
418 return (EINVAL);
419 /*
420 * Previous versions could only support 1 or 2 parity
421 * device.
422 */
423 if (nparity > 1 &&
424 spa_version(spa) < SPA_VERSION_RAIDZ2)
425 return (ENOTSUP);
426 if (nparity > 2 &&
427 spa_version(spa) < SPA_VERSION_RAIDZ3)
428 return (ENOTSUP);
429 } else {
430 /*
431 * We require the parity to be specified for SPAs that
432 * support multiple parity levels.
433 */
434 if (spa_version(spa) >= SPA_VERSION_RAIDZ2)
435 return (EINVAL);
436 /*
437 * Otherwise, we default to 1 parity device for RAID-Z.
438 */
439 nparity = 1;
440 }
441 } else {
442 nparity = 0;
443 }
444 ASSERT(nparity != -1ULL);
445
446 vd = vdev_alloc_common(spa, id, guid, ops);
447
448 vd->vdev_islog = islog;
449 vd->vdev_nparity = nparity;
450
451 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0)
452 vd->vdev_path = spa_strdup(vd->vdev_path);
453 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0)
454 vd->vdev_devid = spa_strdup(vd->vdev_devid);
455 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH,
456 &vd->vdev_physpath) == 0)
457 vd->vdev_physpath = spa_strdup(vd->vdev_physpath);
458 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0)
459 vd->vdev_fru = spa_strdup(vd->vdev_fru);
460
461 /*
462 * Set the whole_disk property. If it's not specified, leave the value
463 * as -1.
464 */
465 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
466 &vd->vdev_wholedisk) != 0)
467 vd->vdev_wholedisk = -1ULL;
468
469 /*
470 * Look for the 'not present' flag. This will only be set if the device
471 * was not present at the time of import.
472 */
473 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
474 &vd->vdev_not_present);
475
476 /*
477 * Get the alignment requirement.
478 */
479 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift);
480
481 /*
482 * Retrieve the vdev creation time.
483 */
484 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG,
485 &vd->vdev_crtxg);
486
487 /*
488 * If we're a top-level vdev, try to load the allocation parameters.
489 */
490 if (parent && !parent->vdev_parent &&
491 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
492 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
493 &vd->vdev_ms_array);
494 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
495 &vd->vdev_ms_shift);
496 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE,
497 &vd->vdev_asize);
498 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING,
499 &vd->vdev_removing);
500 }
501
502 if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) {
503 ASSERT(alloctype == VDEV_ALLOC_LOAD ||
504 alloctype == VDEV_ALLOC_ADD ||
505 alloctype == VDEV_ALLOC_SPLIT ||
506 alloctype == VDEV_ALLOC_ROOTPOOL);
507 vd->vdev_mg = metaslab_group_create(islog ?
508 spa_log_class(spa) : spa_normal_class(spa), vd);
509 }
510
511 /*
512 * If we're a leaf vdev, try to load the DTL object and other state.
513 */
514 if (vd->vdev_ops->vdev_op_leaf &&
515 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE ||
516 alloctype == VDEV_ALLOC_ROOTPOOL)) {
517 if (alloctype == VDEV_ALLOC_LOAD) {
518 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL,
519 &vd->vdev_dtl_smo.smo_object);
520 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE,
521 &vd->vdev_unspare);
522 }
523
524 if (alloctype == VDEV_ALLOC_ROOTPOOL) {
525 uint64_t spare = 0;
526
527 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
528 &spare) == 0 && spare)
529 spa_spare_add(vd);
530 }
531
532 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE,
533 &vd->vdev_offline);
534
535 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVERING,
536 &vd->vdev_resilvering);
537
538 /*
539 * When importing a pool, we want to ignore the persistent fault
540 * state, as the diagnosis made on another system may not be
541 * valid in the current context. Local vdevs will
542 * remain in the faulted state.
543 */
544 if (spa_load_state(spa) == SPA_LOAD_OPEN) {
545 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED,
546 &vd->vdev_faulted);
547 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED,
548 &vd->vdev_degraded);
549 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED,
550 &vd->vdev_removed);
551
552 if (vd->vdev_faulted || vd->vdev_degraded) {
553 char *aux;
554
555 vd->vdev_label_aux =
556 VDEV_AUX_ERR_EXCEEDED;
557 if (nvlist_lookup_string(nv,
558 ZPOOL_CONFIG_AUX_STATE, &aux) == 0 &&
559 strcmp(aux, "external") == 0)
560 vd->vdev_label_aux = VDEV_AUX_EXTERNAL;
561 }
562 }
563 }
564
565 /*
566 * Add ourselves to the parent's list of children.
567 */
568 vdev_add_child(parent, vd);
569
570 *vdp = vd;
571
572 return (0);
573}
574
575void
576vdev_free(vdev_t *vd)
577{
578 spa_t *spa = vd->vdev_spa;
579
580 /*
581 * vdev_free() implies closing the vdev first. This is simpler than
582 * trying to ensure complicated semantics for all callers.
583 */
584 vdev_close(vd);
585
586 ASSERT(!list_link_active(&vd->vdev_config_dirty_node));
587 ASSERT(!list_link_active(&vd->vdev_state_dirty_node));
588
589 /*
590 * Free all children.
591 */
592 for (int c = 0; c < vd->vdev_children; c++)
593 vdev_free(vd->vdev_child[c]);
594
595 ASSERT(vd->vdev_child == NULL);
596 ASSERT(vd->vdev_guid_sum == vd->vdev_guid);
597
598 /*
599 * Discard allocation state.
600 */
601 if (vd->vdev_mg != NULL) {
602 vdev_metaslab_fini(vd);
603 metaslab_group_destroy(vd->vdev_mg);
604 }
605
606 ASSERT3U(vd->vdev_stat.vs_space, ==, 0);
607 ASSERT3U(vd->vdev_stat.vs_dspace, ==, 0);
608 ASSERT3U(vd->vdev_stat.vs_alloc, ==, 0);
609
610 /*
611 * Remove this vdev from its parent's child list.
612 */
613 vdev_remove_child(vd->vdev_parent, vd);
614
615 ASSERT(vd->vdev_parent == NULL);
616
617 /*
618 * Clean up vdev structure.
619 */
620 vdev_queue_fini(vd);
621 vdev_cache_fini(vd);
622
623 if (vd->vdev_path)
624 spa_strfree(vd->vdev_path);
625 if (vd->vdev_devid)
626 spa_strfree(vd->vdev_devid);
627 if (vd->vdev_physpath)
628 spa_strfree(vd->vdev_physpath);
629 if (vd->vdev_fru)
630 spa_strfree(vd->vdev_fru);
631
632 if (vd->vdev_isspare)
633 spa_spare_remove(vd);
634 if (vd->vdev_isl2cache)
635 spa_l2cache_remove(vd);
636
637 txg_list_destroy(&vd->vdev_ms_list);
638 txg_list_destroy(&vd->vdev_dtl_list);
639
640 mutex_enter(&vd->vdev_dtl_lock);
641 for (int t = 0; t < DTL_TYPES; t++) {
642 space_map_unload(&vd->vdev_dtl[t]);
643 space_map_destroy(&vd->vdev_dtl[t]);
644 }
645 mutex_exit(&vd->vdev_dtl_lock);
646
647 mutex_destroy(&vd->vdev_dtl_lock);
648 mutex_destroy(&vd->vdev_stat_lock);
649 mutex_destroy(&vd->vdev_probe_lock);
650
651 if (vd == spa->spa_root_vdev)
652 spa->spa_root_vdev = NULL;
653
654 kmem_free(vd, sizeof (vdev_t));
655}
656
657/*
658 * Transfer top-level vdev state from svd to tvd.
659 */
660static void
661vdev_top_transfer(vdev_t *svd, vdev_t *tvd)
662{
663 spa_t *spa = svd->vdev_spa;
664 metaslab_t *msp;
665 vdev_t *vd;
666 int t;
667
668 ASSERT(tvd == tvd->vdev_top);
669
670 tvd->vdev_ms_array = svd->vdev_ms_array;
671 tvd->vdev_ms_shift = svd->vdev_ms_shift;
672 tvd->vdev_ms_count = svd->vdev_ms_count;
673
674 svd->vdev_ms_array = 0;
675 svd->vdev_ms_shift = 0;
676 svd->vdev_ms_count = 0;
677
678 if (tvd->vdev_mg)
679 ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg);
680 tvd->vdev_mg = svd->vdev_mg;
681 tvd->vdev_ms = svd->vdev_ms;
682
683 svd->vdev_mg = NULL;
684 svd->vdev_ms = NULL;
685
686 if (tvd->vdev_mg != NULL)
687 tvd->vdev_mg->mg_vd = tvd;
688
689 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc;
690 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space;
691 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace;
692
693 svd->vdev_stat.vs_alloc = 0;
694 svd->vdev_stat.vs_space = 0;
695 svd->vdev_stat.vs_dspace = 0;
696
697 for (t = 0; t < TXG_SIZE; t++) {
698 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL)
699 (void) txg_list_add(&tvd->vdev_ms_list, msp, t);
700 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL)
701 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t);
702 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t))
703 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t);
704 }
705
706 if (list_link_active(&svd->vdev_config_dirty_node)) {
707 vdev_config_clean(svd);
708 vdev_config_dirty(tvd);
709 }
710
711 if (list_link_active(&svd->vdev_state_dirty_node)) {
712 vdev_state_clean(svd);
713 vdev_state_dirty(tvd);
714 }
715
716 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio;
717 svd->vdev_deflate_ratio = 0;
718
719 tvd->vdev_islog = svd->vdev_islog;
720 svd->vdev_islog = 0;
721}
722
723static void
724vdev_top_update(vdev_t *tvd, vdev_t *vd)
725{
726 if (vd == NULL)
727 return;
728
729 vd->vdev_top = tvd;
730
731 for (int c = 0; c < vd->vdev_children; c++)
732 vdev_top_update(tvd, vd->vdev_child[c]);
733}
734
735/*
736 * Add a mirror/replacing vdev above an existing vdev.
737 */
738vdev_t *
739vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops)
740{
741 spa_t *spa = cvd->vdev_spa;
742 vdev_t *pvd = cvd->vdev_parent;
743 vdev_t *mvd;
744
745 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
746
747 mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops);
748
749 mvd->vdev_asize = cvd->vdev_asize;
750 mvd->vdev_min_asize = cvd->vdev_min_asize;
|
| 751 mvd->vdev_max_asize = cvd->vdev_max_asize; |
752 mvd->vdev_ashift = cvd->vdev_ashift;
753 mvd->vdev_state = cvd->vdev_state;
754 mvd->vdev_crtxg = cvd->vdev_crtxg;
755
756 vdev_remove_child(pvd, cvd);
757 vdev_add_child(pvd, mvd);
758 cvd->vdev_id = mvd->vdev_children;
759 vdev_add_child(mvd, cvd);
760 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
761
762 if (mvd == mvd->vdev_top)
763 vdev_top_transfer(cvd, mvd);
764
765 return (mvd);
766}
767
768/*
769 * Remove a 1-way mirror/replacing vdev from the tree.
770 */
771void
772vdev_remove_parent(vdev_t *cvd)
773{
774 vdev_t *mvd = cvd->vdev_parent;
775 vdev_t *pvd = mvd->vdev_parent;
776
777 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
778
779 ASSERT(mvd->vdev_children == 1);
780 ASSERT(mvd->vdev_ops == &vdev_mirror_ops ||
781 mvd->vdev_ops == &vdev_replacing_ops ||
782 mvd->vdev_ops == &vdev_spare_ops);
783 cvd->vdev_ashift = mvd->vdev_ashift;
784
785 vdev_remove_child(mvd, cvd);
786 vdev_remove_child(pvd, mvd);
787
788 /*
789 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
790 * Otherwise, we could have detached an offline device, and when we
791 * go to import the pool we'll think we have two top-level vdevs,
792 * instead of a different version of the same top-level vdev.
793 */
794 if (mvd->vdev_top == mvd) {
795 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid;
796 cvd->vdev_orig_guid = cvd->vdev_guid;
797 cvd->vdev_guid += guid_delta;
798 cvd->vdev_guid_sum += guid_delta;
799 }
800 cvd->vdev_id = mvd->vdev_id;
801 vdev_add_child(pvd, cvd);
802 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
803
804 if (cvd == cvd->vdev_top)
805 vdev_top_transfer(mvd, cvd);
806
807 ASSERT(mvd->vdev_children == 0);
808 vdev_free(mvd);
809}
810
811int
812vdev_metaslab_init(vdev_t *vd, uint64_t txg)
813{
814 spa_t *spa = vd->vdev_spa;
815 objset_t *mos = spa->spa_meta_objset;
816 uint64_t m;
817 uint64_t oldc = vd->vdev_ms_count;
818 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift;
819 metaslab_t **mspp;
820 int error;
821
822 ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER));
823
824 /*
825 * This vdev is not being allocated from yet or is a hole.
826 */
827 if (vd->vdev_ms_shift == 0)
828 return (0);
829
830 ASSERT(!vd->vdev_ishole);
831
832 /*
833 * Compute the raidz-deflation ratio. Note, we hard-code
834 * in 128k (1 << 17) because it is the current "typical" blocksize.
835 * Even if SPA_MAXBLOCKSIZE changes, this algorithm must never change,
836 * or we will inconsistently account for existing bp's.
837 */
838 vd->vdev_deflate_ratio = (1 << 17) /
839 (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT);
840
841 ASSERT(oldc <= newc);
842
843 mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP);
844
845 if (oldc != 0) {
846 bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp));
847 kmem_free(vd->vdev_ms, oldc * sizeof (*mspp));
848 }
849
850 vd->vdev_ms = mspp;
851 vd->vdev_ms_count = newc;
852
853 for (m = oldc; m < newc; m++) {
854 space_map_obj_t smo = { 0, 0, 0 };
855 if (txg == 0) {
856 uint64_t object = 0;
857 error = dmu_read(mos, vd->vdev_ms_array,
858 m * sizeof (uint64_t), sizeof (uint64_t), &object,
859 DMU_READ_PREFETCH);
860 if (error)
861 return (error);
862 if (object != 0) {
863 dmu_buf_t *db;
864 error = dmu_bonus_hold(mos, object, FTAG, &db);
865 if (error)
866 return (error);
867 ASSERT3U(db->db_size, >=, sizeof (smo));
868 bcopy(db->db_data, &smo, sizeof (smo));
869 ASSERT3U(smo.smo_object, ==, object);
870 dmu_buf_rele(db, FTAG);
871 }
872 }
873 vd->vdev_ms[m] = metaslab_init(vd->vdev_mg, &smo,
874 m << vd->vdev_ms_shift, 1ULL << vd->vdev_ms_shift, txg);
875 }
876
877 if (txg == 0)
878 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER);
879
880 /*
881 * If the vdev is being removed we don't activate
882 * the metaslabs since we want to ensure that no new
883 * allocations are performed on this device.
884 */
885 if (oldc == 0 && !vd->vdev_removing)
886 metaslab_group_activate(vd->vdev_mg);
887
888 if (txg == 0)
889 spa_config_exit(spa, SCL_ALLOC, FTAG);
890
891 return (0);
892}
893
894void
895vdev_metaslab_fini(vdev_t *vd)
896{
897 uint64_t m;
898 uint64_t count = vd->vdev_ms_count;
899
900 if (vd->vdev_ms != NULL) {
901 metaslab_group_passivate(vd->vdev_mg);
902 for (m = 0; m < count; m++)
903 if (vd->vdev_ms[m] != NULL)
904 metaslab_fini(vd->vdev_ms[m]);
905 kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *));
906 vd->vdev_ms = NULL;
907 }
908}
909
910typedef struct vdev_probe_stats {
911 boolean_t vps_readable;
912 boolean_t vps_writeable;
913 int vps_flags;
914} vdev_probe_stats_t;
915
916static void
917vdev_probe_done(zio_t *zio)
918{
919 spa_t *spa = zio->io_spa;
920 vdev_t *vd = zio->io_vd;
921 vdev_probe_stats_t *vps = zio->io_private;
922
923 ASSERT(vd->vdev_probe_zio != NULL);
924
925 if (zio->io_type == ZIO_TYPE_READ) {
926 if (zio->io_error == 0)
927 vps->vps_readable = 1;
928 if (zio->io_error == 0 && spa_writeable(spa)) {
929 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd,
930 zio->io_offset, zio->io_size, zio->io_data,
931 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
932 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE));
933 } else {
934 zio_buf_free(zio->io_data, zio->io_size);
935 }
936 } else if (zio->io_type == ZIO_TYPE_WRITE) {
937 if (zio->io_error == 0)
938 vps->vps_writeable = 1;
939 zio_buf_free(zio->io_data, zio->io_size);
940 } else if (zio->io_type == ZIO_TYPE_NULL) {
941 zio_t *pio;
942
943 vd->vdev_cant_read |= !vps->vps_readable;
944 vd->vdev_cant_write |= !vps->vps_writeable;
945
946 if (vdev_readable(vd) &&
947 (vdev_writeable(vd) || !spa_writeable(spa))) {
948 zio->io_error = 0;
949 } else {
950 ASSERT(zio->io_error != 0);
951 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE,
952 spa, vd, NULL, 0, 0);
953 zio->io_error = ENXIO;
954 }
955
956 mutex_enter(&vd->vdev_probe_lock);
957 ASSERT(vd->vdev_probe_zio == zio);
958 vd->vdev_probe_zio = NULL;
959 mutex_exit(&vd->vdev_probe_lock);
960
961 while ((pio = zio_walk_parents(zio)) != NULL)
962 if (!vdev_accessible(vd, pio))
963 pio->io_error = ENXIO;
964
965 kmem_free(vps, sizeof (*vps));
966 }
967}
968
969/*
970 * Determine whether this device is accessible by reading and writing
971 * to several known locations: the pad regions of each vdev label
972 * but the first (which we leave alone in case it contains a VTOC).
973 */
974zio_t *
975vdev_probe(vdev_t *vd, zio_t *zio)
976{
977 spa_t *spa = vd->vdev_spa;
978 vdev_probe_stats_t *vps = NULL;
979 zio_t *pio;
980
981 ASSERT(vd->vdev_ops->vdev_op_leaf);
982
983 /*
984 * Don't probe the probe.
985 */
986 if (zio && (zio->io_flags & ZIO_FLAG_PROBE))
987 return (NULL);
988
989 /*
990 * To prevent 'probe storms' when a device fails, we create
991 * just one probe i/o at a time. All zios that want to probe
992 * this vdev will become parents of the probe io.
993 */
994 mutex_enter(&vd->vdev_probe_lock);
995
996 if ((pio = vd->vdev_probe_zio) == NULL) {
997 vps = kmem_zalloc(sizeof (*vps), KM_SLEEP);
998
999 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE |
1000 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE |
1001 ZIO_FLAG_TRYHARD;
1002
1003 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) {
1004 /*
1005 * vdev_cant_read and vdev_cant_write can only
1006 * transition from TRUE to FALSE when we have the
1007 * SCL_ZIO lock as writer; otherwise they can only
1008 * transition from FALSE to TRUE. This ensures that
1009 * any zio looking at these values can assume that
1010 * failures persist for the life of the I/O. That's
1011 * important because when a device has intermittent
1012 * connectivity problems, we want to ensure that
1013 * they're ascribed to the device (ENXIO) and not
1014 * the zio (EIO).
1015 *
1016 * Since we hold SCL_ZIO as writer here, clear both
1017 * values so the probe can reevaluate from first
1018 * principles.
1019 */
1020 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER;
1021 vd->vdev_cant_read = B_FALSE;
1022 vd->vdev_cant_write = B_FALSE;
1023 }
1024
1025 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd,
1026 vdev_probe_done, vps,
1027 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE);
1028
1029 /*
1030 * We can't change the vdev state in this context, so we
1031 * kick off an async task to do it on our behalf.
1032 */
1033 if (zio != NULL) {
1034 vd->vdev_probe_wanted = B_TRUE;
1035 spa_async_request(spa, SPA_ASYNC_PROBE);
1036 }
1037 }
1038
1039 if (zio != NULL)
1040 zio_add_child(zio, pio);
1041
1042 mutex_exit(&vd->vdev_probe_lock);
1043
1044 if (vps == NULL) {
1045 ASSERT(zio != NULL);
1046 return (NULL);
1047 }
1048
1049 for (int l = 1; l < VDEV_LABELS; l++) {
1050 zio_nowait(zio_read_phys(pio, vd,
1051 vdev_label_offset(vd->vdev_psize, l,
1052 offsetof(vdev_label_t, vl_pad2)),
1053 VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE),
1054 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1055 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE));
1056 }
1057
1058 if (zio == NULL)
1059 return (pio);
1060
1061 zio_nowait(pio);
1062 return (NULL);
1063}
1064
1065static void
1066vdev_open_child(void *arg)
1067{
1068 vdev_t *vd = arg;
1069
1070 vd->vdev_open_thread = curthread;
1071 vd->vdev_open_error = vdev_open(vd);
1072 vd->vdev_open_thread = NULL;
1073}
1074
1075boolean_t
1076vdev_uses_zvols(vdev_t *vd)
1077{
1078 if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR,
1079 strlen(ZVOL_DIR)) == 0)
1080 return (B_TRUE);
1081 for (int c = 0; c < vd->vdev_children; c++)
1082 if (vdev_uses_zvols(vd->vdev_child[c]))
1083 return (B_TRUE);
1084 return (B_FALSE);
1085}
1086
1087void
1088vdev_open_children(vdev_t *vd)
1089{
1090 taskq_t *tq;
1091 int children = vd->vdev_children;
1092
1093 /*
1094 * in order to handle pools on top of zvols, do the opens
1095 * in a single thread so that the same thread holds the
1096 * spa_namespace_lock
1097 */
1098 if (B_TRUE || vdev_uses_zvols(vd)) {
1099 for (int c = 0; c < children; c++)
1100 vd->vdev_child[c]->vdev_open_error =
1101 vdev_open(vd->vdev_child[c]);
1102 return;
1103 }
1104 tq = taskq_create("vdev_open", children, minclsyspri,
1105 children, children, TASKQ_PREPOPULATE);
1106
1107 for (int c = 0; c < children; c++)
1108 VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c],
1109 TQ_SLEEP) != 0);
1110
1111 taskq_destroy(tq);
1112}
1113
1114/*
1115 * Prepare a virtual device for access.
1116 */
1117int
1118vdev_open(vdev_t *vd)
1119{
1120 spa_t *spa = vd->vdev_spa;
1121 int error;
1122 uint64_t osize = 0;
|
1122 uint64_t asize, psize;
|
1123 uint64_t max_osize = 0;
1124 uint64_t asize, max_asize, psize; |
1125 uint64_t ashift = 0;
1126
1127 ASSERT(vd->vdev_open_thread == curthread ||
1128 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1129 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED ||
1130 vd->vdev_state == VDEV_STATE_CANT_OPEN ||
1131 vd->vdev_state == VDEV_STATE_OFFLINE);
1132
1133 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1134 vd->vdev_cant_read = B_FALSE;
1135 vd->vdev_cant_write = B_FALSE;
1136 vd->vdev_min_asize = vdev_get_min_asize(vd);
1137
1138 /*
1139 * If this vdev is not removed, check its fault status. If it's
1140 * faulted, bail out of the open.
1141 */
1142 if (!vd->vdev_removed && vd->vdev_faulted) {
1143 ASSERT(vd->vdev_children == 0);
1144 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1145 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1146 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1147 vd->vdev_label_aux);
1148 return (ENXIO);
1149 } else if (vd->vdev_offline) {
1150 ASSERT(vd->vdev_children == 0);
1151 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE);
1152 return (ENXIO);
1153 }
1154
|
1153 error = vd->vdev_ops->vdev_op_open(vd, &osize, &ashift);
|
| 1155 error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize, &ashift); |
1156
1157 /*
1158 * Reset the vdev_reopening flag so that we actually close
1159 * the vdev on error.
1160 */
1161 vd->vdev_reopening = B_FALSE;
1162 if (zio_injection_enabled && error == 0)
1163 error = zio_handle_device_injection(vd, NULL, ENXIO);
1164
1165 if (error) {
1166 if (vd->vdev_removed &&
1167 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED)
1168 vd->vdev_removed = B_FALSE;
1169
1170 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1171 vd->vdev_stat.vs_aux);
1172 return (error);
1173 }
1174
1175 vd->vdev_removed = B_FALSE;
1176
1177 /*
1178 * Recheck the faulted flag now that we have confirmed that
1179 * the vdev is accessible. If we're faulted, bail.
1180 */
1181 if (vd->vdev_faulted) {
1182 ASSERT(vd->vdev_children == 0);
1183 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1184 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1185 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1186 vd->vdev_label_aux);
1187 return (ENXIO);
1188 }
1189
1190 if (vd->vdev_degraded) {
1191 ASSERT(vd->vdev_children == 0);
1192 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1193 VDEV_AUX_ERR_EXCEEDED);
1194 } else {
1195 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0);
1196 }
1197
1198 /*
1199 * For hole or missing vdevs we just return success.
1200 */
1201 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops)
1202 return (0);
1203
1204 for (int c = 0; c < vd->vdev_children; c++) {
1205 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) {
1206 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1207 VDEV_AUX_NONE);
1208 break;
1209 }
1210 }
1211
1212 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t));
|
| 1213 max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t)); |
1214
1215 if (vd->vdev_children == 0) {
1216 if (osize < SPA_MINDEVSIZE) {
1217 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1218 VDEV_AUX_TOO_SMALL);
1219 return (EOVERFLOW);
1220 }
1221 psize = osize;
1222 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE);
|
1223 max_asize = max_osize - (VDEV_LABEL_START_SIZE +
1224 VDEV_LABEL_END_SIZE); |
1225 } else {
1226 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE -
1227 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) {
1228 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1229 VDEV_AUX_TOO_SMALL);
1230 return (EOVERFLOW);
1231 }
1232 psize = 0;
1233 asize = osize;
|
| 1234 max_asize = max_osize; |
1235 }
1236
1237 vd->vdev_psize = psize;
1238
1239 /*
1240 * Make sure the allocatable size hasn't shrunk.
1241 */
1242 if (asize < vd->vdev_min_asize) {
1243 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1244 VDEV_AUX_BAD_LABEL);
1245 return (EINVAL);
1246 }
1247
1248 if (vd->vdev_asize == 0) {
1249 /*
1250 * This is the first-ever open, so use the computed values.
1251 * For testing purposes, a higher ashift can be requested.
1252 */
1253 vd->vdev_asize = asize;
|
| 1254 vd->vdev_max_asize = max_asize; |
1255 vd->vdev_ashift = MAX(ashift, vd->vdev_ashift);
1256 } else {
1257 /*
1258 * Make sure the alignment requirement hasn't increased.
1259 */
1260 if (ashift > vd->vdev_top->vdev_ashift) {
1261 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1262 VDEV_AUX_BAD_LABEL);
1263 return (EINVAL);
1264 }
|
| 1265 vd->vdev_max_asize = max_asize; |
1266 }
1267
1268 /*
1269 * If all children are healthy and the asize has increased,
1270 * then we've experienced dynamic LUN growth. If automatic
1271 * expansion is enabled then use the additional space.
1272 */
1273 if (vd->vdev_state == VDEV_STATE_HEALTHY && asize > vd->vdev_asize &&
1274 (vd->vdev_expanding || spa->spa_autoexpand))
1275 vd->vdev_asize = asize;
1276
1277 vdev_set_min_asize(vd);
1278
1279 /*
1280 * Ensure we can issue some IO before declaring the
1281 * vdev open for business.
1282 */
1283 if (vd->vdev_ops->vdev_op_leaf &&
1284 (error = zio_wait(vdev_probe(vd, NULL))) != 0) {
1285 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1286 VDEV_AUX_ERR_EXCEEDED);
1287 return (error);
1288 }
1289
1290 /*
1291 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1292 * resilver. But don't do this if we are doing a reopen for a scrub,
1293 * since this would just restart the scrub we are already doing.
1294 */
1295 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen &&
1296 vdev_resilver_needed(vd, NULL, NULL))
1297 spa_async_request(spa, SPA_ASYNC_RESILVER);
1298
1299 return (0);
1300}
1301
1302/*
1303 * Called once the vdevs are all opened, this routine validates the label
1304 * contents. This needs to be done before vdev_load() so that we don't
1305 * inadvertently do repair I/Os to the wrong device.
1306 *
1307 * If 'strict' is false ignore the spa guid check. This is necessary because
1308 * if the machine crashed during a re-guid the new guid might have been written
1309 * to all of the vdev labels, but not the cached config. The strict check
1310 * will be performed when the pool is opened again using the mos config.
1311 *
1312 * This function will only return failure if one of the vdevs indicates that it
1313 * has since been destroyed or exported. This is only possible if
1314 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1315 * will be updated but the function will return 0.
1316 */
1317int
1318vdev_validate(vdev_t *vd, boolean_t strict)
1319{
1320 spa_t *spa = vd->vdev_spa;
1321 nvlist_t *label;
1322 uint64_t guid = 0, top_guid;
1323 uint64_t state;
1324
1325 for (int c = 0; c < vd->vdev_children; c++)
1326 if (vdev_validate(vd->vdev_child[c], strict) != 0)
1327 return (EBADF);
1328
1329 /*
1330 * If the device has already failed, or was marked offline, don't do
1331 * any further validation. Otherwise, label I/O will fail and we will
1332 * overwrite the previous state.
1333 */
1334 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1335 uint64_t aux_guid = 0;
1336 nvlist_t *nvl;
1337
1338 if ((label = vdev_label_read_config(vd)) == NULL) {
1339 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1340 VDEV_AUX_BAD_LABEL);
1341 return (0);
1342 }
1343
1344 /*
1345 * Determine if this vdev has been split off into another
1346 * pool. If so, then refuse to open it.
1347 */
1348 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID,
1349 &aux_guid) == 0 && aux_guid == spa_guid(spa)) {
1350 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1351 VDEV_AUX_SPLIT_POOL);
1352 nvlist_free(label);
1353 return (0);
1354 }
1355
1356 if (strict && (nvlist_lookup_uint64(label,
1357 ZPOOL_CONFIG_POOL_GUID, &guid) != 0 ||
1358 guid != spa_guid(spa))) {
1359 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1360 VDEV_AUX_CORRUPT_DATA);
1361 nvlist_free(label);
1362 return (0);
1363 }
1364
1365 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl)
1366 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID,
1367 &aux_guid) != 0)
1368 aux_guid = 0;
1369
1370 /*
1371 * If this vdev just became a top-level vdev because its
1372 * sibling was detached, it will have adopted the parent's
1373 * vdev guid -- but the label may or may not be on disk yet.
1374 * Fortunately, either version of the label will have the
1375 * same top guid, so if we're a top-level vdev, we can
1376 * safely compare to that instead.
1377 *
1378 * If we split this vdev off instead, then we also check the
1379 * original pool's guid. We don't want to consider the vdev
1380 * corrupt if it is partway through a split operation.
1381 */
1382 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
1383 &guid) != 0 ||
1384 nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID,
1385 &top_guid) != 0 ||
1386 ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) &&
1387 (vd->vdev_guid != top_guid || vd != vd->vdev_top))) {
1388 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1389 VDEV_AUX_CORRUPT_DATA);
1390 nvlist_free(label);
1391 return (0);
1392 }
1393
1394 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1395 &state) != 0) {
1396 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1397 VDEV_AUX_CORRUPT_DATA);
1398 nvlist_free(label);
1399 return (0);
1400 }
1401
1402 nvlist_free(label);
1403
1404 /*
1405 * If this is a verbatim import, no need to check the
1406 * state of the pool.
1407 */
1408 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) &&
1409 spa_load_state(spa) == SPA_LOAD_OPEN &&
1410 state != POOL_STATE_ACTIVE)
1411 return (EBADF);
1412
1413 /*
1414 * If we were able to open and validate a vdev that was
1415 * previously marked permanently unavailable, clear that state
1416 * now.
1417 */
1418 if (vd->vdev_not_present)
1419 vd->vdev_not_present = 0;
1420 }
1421
1422 return (0);
1423}
1424
1425/*
1426 * Close a virtual device.
1427 */
1428void
1429vdev_close(vdev_t *vd)
1430{
1431 spa_t *spa = vd->vdev_spa;
1432 vdev_t *pvd = vd->vdev_parent;
1433
1434 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1435
1436 /*
1437 * If our parent is reopening, then we are as well, unless we are
1438 * going offline.
1439 */
1440 if (pvd != NULL && pvd->vdev_reopening)
1441 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline);
1442
1443 vd->vdev_ops->vdev_op_close(vd);
1444
1445 vdev_cache_purge(vd);
1446
1447 /*
1448 * We record the previous state before we close it, so that if we are
1449 * doing a reopen(), we don't generate FMA ereports if we notice that
1450 * it's still faulted.
1451 */
1452 vd->vdev_prevstate = vd->vdev_state;
1453
1454 if (vd->vdev_offline)
1455 vd->vdev_state = VDEV_STATE_OFFLINE;
1456 else
1457 vd->vdev_state = VDEV_STATE_CLOSED;
1458 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1459}
1460
1461void
1462vdev_hold(vdev_t *vd)
1463{
1464 spa_t *spa = vd->vdev_spa;
1465
1466 ASSERT(spa_is_root(spa));
1467 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1468 return;
1469
1470 for (int c = 0; c < vd->vdev_children; c++)
1471 vdev_hold(vd->vdev_child[c]);
1472
1473 if (vd->vdev_ops->vdev_op_leaf)
1474 vd->vdev_ops->vdev_op_hold(vd);
1475}
1476
1477void
1478vdev_rele(vdev_t *vd)
1479{
1480 spa_t *spa = vd->vdev_spa;
1481
1482 ASSERT(spa_is_root(spa));
1483 for (int c = 0; c < vd->vdev_children; c++)
1484 vdev_rele(vd->vdev_child[c]);
1485
1486 if (vd->vdev_ops->vdev_op_leaf)
1487 vd->vdev_ops->vdev_op_rele(vd);
1488}
1489
1490/*
1491 * Reopen all interior vdevs and any unopened leaves. We don't actually
1492 * reopen leaf vdevs which had previously been opened as they might deadlock
1493 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1494 * If the leaf has never been opened then open it, as usual.
1495 */
1496void
1497vdev_reopen(vdev_t *vd)
1498{
1499 spa_t *spa = vd->vdev_spa;
1500
1501 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1502
1503 /* set the reopening flag unless we're taking the vdev offline */
1504 vd->vdev_reopening = !vd->vdev_offline;
1505 vdev_close(vd);
1506 (void) vdev_open(vd);
1507
1508 /*
1509 * Call vdev_validate() here to make sure we have the same device.
1510 * Otherwise, a device with an invalid label could be successfully
1511 * opened in response to vdev_reopen().
1512 */
1513 if (vd->vdev_aux) {
1514 (void) vdev_validate_aux(vd);
1515 if (vdev_readable(vd) && vdev_writeable(vd) &&
1516 vd->vdev_aux == &spa->spa_l2cache &&
1517 !l2arc_vdev_present(vd))
1518 l2arc_add_vdev(spa, vd);
1519 } else {
1520 (void) vdev_validate(vd, B_TRUE);
1521 }
1522
1523 /*
1524 * Reassess parent vdev's health.
1525 */
1526 vdev_propagate_state(vd);
1527}
1528
1529int
1530vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
1531{
1532 int error;
1533
1534 /*
1535 * Normally, partial opens (e.g. of a mirror) are allowed.
1536 * For a create, however, we want to fail the request if
1537 * there are any components we can't open.
1538 */
1539 error = vdev_open(vd);
1540
1541 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) {
1542 vdev_close(vd);
1543 return (error ? error : ENXIO);
1544 }
1545
1546 /*
1547 * Recursively initialize all labels.
1548 */
1549 if ((error = vdev_label_init(vd, txg, isreplacing ?
1550 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
1551 vdev_close(vd);
1552 return (error);
1553 }
1554
1555 return (0);
1556}
1557
1558void
1559vdev_metaslab_set_size(vdev_t *vd)
1560{
1561 /*
1562 * Aim for roughly 200 metaslabs per vdev.
1563 */
1564 vd->vdev_ms_shift = highbit(vd->vdev_asize / 200);
1565 vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT);
1566}
1567
1568void
1569vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
1570{
1571 ASSERT(vd == vd->vdev_top);
1572 ASSERT(!vd->vdev_ishole);
1573 ASSERT(ISP2(flags));
1574 ASSERT(spa_writeable(vd->vdev_spa));
1575
1576 if (flags & VDD_METASLAB)
1577 (void) txg_list_add(&vd->vdev_ms_list, arg, txg);
1578
1579 if (flags & VDD_DTL)
1580 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg);
1581
1582 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
1583}
1584
1585/*
1586 * DTLs.
1587 *
1588 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1589 * the vdev has less than perfect replication. There are four kinds of DTL:
1590 *
1591 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1592 *
1593 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1594 *
1595 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1596 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1597 * txgs that was scrubbed.
1598 *
1599 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1600 * persistent errors or just some device being offline.
1601 * Unlike the other three, the DTL_OUTAGE map is not generally
1602 * maintained; it's only computed when needed, typically to
1603 * determine whether a device can be detached.
1604 *
1605 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1606 * either has the data or it doesn't.
1607 *
1608 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1609 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1610 * if any child is less than fully replicated, then so is its parent.
1611 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1612 * comprising only those txgs which appear in 'maxfaults' or more children;
1613 * those are the txgs we don't have enough replication to read. For example,
1614 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1615 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1616 * two child DTL_MISSING maps.
1617 *
1618 * It should be clear from the above that to compute the DTLs and outage maps
1619 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1620 * Therefore, that is all we keep on disk. When loading the pool, or after
1621 * a configuration change, we generate all other DTLs from first principles.
1622 */
1623void
1624vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1625{
1626 space_map_t *sm = &vd->vdev_dtl[t];
1627
1628 ASSERT(t < DTL_TYPES);
1629 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1630 ASSERT(spa_writeable(vd->vdev_spa));
1631
1632 mutex_enter(sm->sm_lock);
1633 if (!space_map_contains(sm, txg, size))
1634 space_map_add(sm, txg, size);
1635 mutex_exit(sm->sm_lock);
1636}
1637
1638boolean_t
1639vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1640{
1641 space_map_t *sm = &vd->vdev_dtl[t];
1642 boolean_t dirty = B_FALSE;
1643
1644 ASSERT(t < DTL_TYPES);
1645 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1646
1647 mutex_enter(sm->sm_lock);
1648 if (sm->sm_space != 0)
1649 dirty = space_map_contains(sm, txg, size);
1650 mutex_exit(sm->sm_lock);
1651
1652 return (dirty);
1653}
1654
1655boolean_t
1656vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t)
1657{
1658 space_map_t *sm = &vd->vdev_dtl[t];
1659 boolean_t empty;
1660
1661 mutex_enter(sm->sm_lock);
1662 empty = (sm->sm_space == 0);
1663 mutex_exit(sm->sm_lock);
1664
1665 return (empty);
1666}
1667
1668/*
1669 * Reassess DTLs after a config change or scrub completion.
1670 */
1671void
1672vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
1673{
1674 spa_t *spa = vd->vdev_spa;
1675 avl_tree_t reftree;
1676 int minref;
1677
1678 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1679
1680 for (int c = 0; c < vd->vdev_children; c++)
1681 vdev_dtl_reassess(vd->vdev_child[c], txg,
1682 scrub_txg, scrub_done);
1683
1684 if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux)
1685 return;
1686
1687 if (vd->vdev_ops->vdev_op_leaf) {
1688 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
1689
1690 mutex_enter(&vd->vdev_dtl_lock);
1691 if (scrub_txg != 0 &&
1692 (spa->spa_scrub_started ||
1693 (scn && scn->scn_phys.scn_errors == 0))) {
1694 /*
1695 * We completed a scrub up to scrub_txg. If we
1696 * did it without rebooting, then the scrub dtl
1697 * will be valid, so excise the old region and
1698 * fold in the scrub dtl. Otherwise, leave the
1699 * dtl as-is if there was an error.
1700 *
1701 * There's little trick here: to excise the beginning
1702 * of the DTL_MISSING map, we put it into a reference
1703 * tree and then add a segment with refcnt -1 that
1704 * covers the range [0, scrub_txg). This means
1705 * that each txg in that range has refcnt -1 or 0.
1706 * We then add DTL_SCRUB with a refcnt of 2, so that
1707 * entries in the range [0, scrub_txg) will have a
1708 * positive refcnt -- either 1 or 2. We then convert
1709 * the reference tree into the new DTL_MISSING map.
1710 */
1711 space_map_ref_create(&reftree);
1712 space_map_ref_add_map(&reftree,
1713 &vd->vdev_dtl[DTL_MISSING], 1);
1714 space_map_ref_add_seg(&reftree, 0, scrub_txg, -1);
1715 space_map_ref_add_map(&reftree,
1716 &vd->vdev_dtl[DTL_SCRUB], 2);
1717 space_map_ref_generate_map(&reftree,
1718 &vd->vdev_dtl[DTL_MISSING], 1);
1719 space_map_ref_destroy(&reftree);
1720 }
1721 space_map_vacate(&vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
1722 space_map_walk(&vd->vdev_dtl[DTL_MISSING],
1723 space_map_add, &vd->vdev_dtl[DTL_PARTIAL]);
1724 if (scrub_done)
1725 space_map_vacate(&vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
1726 space_map_vacate(&vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
1727 if (!vdev_readable(vd))
1728 space_map_add(&vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
1729 else
1730 space_map_walk(&vd->vdev_dtl[DTL_MISSING],
1731 space_map_add, &vd->vdev_dtl[DTL_OUTAGE]);
1732 mutex_exit(&vd->vdev_dtl_lock);
1733
1734 if (txg != 0)
1735 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
1736 return;
1737 }
1738
1739 mutex_enter(&vd->vdev_dtl_lock);
1740 for (int t = 0; t < DTL_TYPES; t++) {
1741 /* account for child's outage in parent's missing map */
1742 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t;
1743 if (t == DTL_SCRUB)
1744 continue; /* leaf vdevs only */
1745 if (t == DTL_PARTIAL)
1746 minref = 1; /* i.e. non-zero */
1747 else if (vd->vdev_nparity != 0)
1748 minref = vd->vdev_nparity + 1; /* RAID-Z */
1749 else
1750 minref = vd->vdev_children; /* any kind of mirror */
1751 space_map_ref_create(&reftree);
1752 for (int c = 0; c < vd->vdev_children; c++) {
1753 vdev_t *cvd = vd->vdev_child[c];
1754 mutex_enter(&cvd->vdev_dtl_lock);
1755 space_map_ref_add_map(&reftree, &cvd->vdev_dtl[s], 1);
1756 mutex_exit(&cvd->vdev_dtl_lock);
1757 }
1758 space_map_ref_generate_map(&reftree, &vd->vdev_dtl[t], minref);
1759 space_map_ref_destroy(&reftree);
1760 }
1761 mutex_exit(&vd->vdev_dtl_lock);
1762}
1763
1764static int
1765vdev_dtl_load(vdev_t *vd)
1766{
1767 spa_t *spa = vd->vdev_spa;
1768 space_map_obj_t *smo = &vd->vdev_dtl_smo;
1769 objset_t *mos = spa->spa_meta_objset;
1770 dmu_buf_t *db;
1771 int error;
1772
1773 ASSERT(vd->vdev_children == 0);
1774
1775 if (smo->smo_object == 0)
1776 return (0);
1777
1778 ASSERT(!vd->vdev_ishole);
1779
1780 if ((error = dmu_bonus_hold(mos, smo->smo_object, FTAG, &db)) != 0)
1781 return (error);
1782
1783 ASSERT3U(db->db_size, >=, sizeof (*smo));
1784 bcopy(db->db_data, smo, sizeof (*smo));
1785 dmu_buf_rele(db, FTAG);
1786
1787 mutex_enter(&vd->vdev_dtl_lock);
1788 error = space_map_load(&vd->vdev_dtl[DTL_MISSING],
1789 NULL, SM_ALLOC, smo, mos);
1790 mutex_exit(&vd->vdev_dtl_lock);
1791
1792 return (error);
1793}
1794
1795void
1796vdev_dtl_sync(vdev_t *vd, uint64_t txg)
1797{
1798 spa_t *spa = vd->vdev_spa;
1799 space_map_obj_t *smo = &vd->vdev_dtl_smo;
1800 space_map_t *sm = &vd->vdev_dtl[DTL_MISSING];
1801 objset_t *mos = spa->spa_meta_objset;
1802 space_map_t smsync;
1803 kmutex_t smlock;
1804 dmu_buf_t *db;
1805 dmu_tx_t *tx;
1806
1807 ASSERT(!vd->vdev_ishole);
1808
1809 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
1810
1811 if (vd->vdev_detached) {
1812 if (smo->smo_object != 0) {
1813 int err = dmu_object_free(mos, smo->smo_object, tx);
1814 ASSERT3U(err, ==, 0);
1815 smo->smo_object = 0;
1816 }
1817 dmu_tx_commit(tx);
1818 return;
1819 }
1820
1821 if (smo->smo_object == 0) {
1822 ASSERT(smo->smo_objsize == 0);
1823 ASSERT(smo->smo_alloc == 0);
1824 smo->smo_object = dmu_object_alloc(mos,
1825 DMU_OT_SPACE_MAP, 1 << SPACE_MAP_BLOCKSHIFT,
1826 DMU_OT_SPACE_MAP_HEADER, sizeof (*smo), tx);
1827 ASSERT(smo->smo_object != 0);
1828 vdev_config_dirty(vd->vdev_top);
1829 }
1830
1831 mutex_init(&smlock, NULL, MUTEX_DEFAULT, NULL);
1832
1833 space_map_create(&smsync, sm->sm_start, sm->sm_size, sm->sm_shift,
1834 &smlock);
1835
1836 mutex_enter(&smlock);
1837
1838 mutex_enter(&vd->vdev_dtl_lock);
1839 space_map_walk(sm, space_map_add, &smsync);
1840 mutex_exit(&vd->vdev_dtl_lock);
1841
1842 space_map_truncate(smo, mos, tx);
1843 space_map_sync(&smsync, SM_ALLOC, smo, mos, tx);
1844
1845 space_map_destroy(&smsync);
1846
1847 mutex_exit(&smlock);
1848 mutex_destroy(&smlock);
1849
1850 VERIFY(0 == dmu_bonus_hold(mos, smo->smo_object, FTAG, &db));
1851 dmu_buf_will_dirty(db, tx);
1852 ASSERT3U(db->db_size, >=, sizeof (*smo));
1853 bcopy(smo, db->db_data, sizeof (*smo));
1854 dmu_buf_rele(db, FTAG);
1855
1856 dmu_tx_commit(tx);
1857}
1858
1859/*
1860 * Determine whether the specified vdev can be offlined/detached/removed
1861 * without losing data.
1862 */
1863boolean_t
1864vdev_dtl_required(vdev_t *vd)
1865{
1866 spa_t *spa = vd->vdev_spa;
1867 vdev_t *tvd = vd->vdev_top;
1868 uint8_t cant_read = vd->vdev_cant_read;
1869 boolean_t required;
1870
1871 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1872
1873 if (vd == spa->spa_root_vdev || vd == tvd)
1874 return (B_TRUE);
1875
1876 /*
1877 * Temporarily mark the device as unreadable, and then determine
1878 * whether this results in any DTL outages in the top-level vdev.
1879 * If not, we can safely offline/detach/remove the device.
1880 */
1881 vd->vdev_cant_read = B_TRUE;
1882 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
1883 required = !vdev_dtl_empty(tvd, DTL_OUTAGE);
1884 vd->vdev_cant_read = cant_read;
1885 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
1886
1887 if (!required && zio_injection_enabled)
1888 required = !!zio_handle_device_injection(vd, NULL, ECHILD);
1889
1890 return (required);
1891}
1892
1893/*
1894 * Determine if resilver is needed, and if so the txg range.
1895 */
1896boolean_t
1897vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
1898{
1899 boolean_t needed = B_FALSE;
1900 uint64_t thismin = UINT64_MAX;
1901 uint64_t thismax = 0;
1902
1903 if (vd->vdev_children == 0) {
1904 mutex_enter(&vd->vdev_dtl_lock);
1905 if (vd->vdev_dtl[DTL_MISSING].sm_space != 0 &&
1906 vdev_writeable(vd)) {
1907 space_seg_t *ss;
1908
1909 ss = avl_first(&vd->vdev_dtl[DTL_MISSING].sm_root);
1910 thismin = ss->ss_start - 1;
1911 ss = avl_last(&vd->vdev_dtl[DTL_MISSING].sm_root);
1912 thismax = ss->ss_end;
1913 needed = B_TRUE;
1914 }
1915 mutex_exit(&vd->vdev_dtl_lock);
1916 } else {
1917 for (int c = 0; c < vd->vdev_children; c++) {
1918 vdev_t *cvd = vd->vdev_child[c];
1919 uint64_t cmin, cmax;
1920
1921 if (vdev_resilver_needed(cvd, &cmin, &cmax)) {
1922 thismin = MIN(thismin, cmin);
1923 thismax = MAX(thismax, cmax);
1924 needed = B_TRUE;
1925 }
1926 }
1927 }
1928
1929 if (needed && minp) {
1930 *minp = thismin;
1931 *maxp = thismax;
1932 }
1933 return (needed);
1934}
1935
1936void
1937vdev_load(vdev_t *vd)
1938{
1939 /*
1940 * Recursively load all children.
1941 */
1942 for (int c = 0; c < vd->vdev_children; c++)
1943 vdev_load(vd->vdev_child[c]);
1944
1945 /*
1946 * If this is a top-level vdev, initialize its metaslabs.
1947 */
1948 if (vd == vd->vdev_top && !vd->vdev_ishole &&
1949 (vd->vdev_ashift == 0 || vd->vdev_asize == 0 ||
1950 vdev_metaslab_init(vd, 0) != 0))
1951 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1952 VDEV_AUX_CORRUPT_DATA);
1953
1954 /*
1955 * If this is a leaf vdev, load its DTL.
1956 */
1957 if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0)
1958 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1959 VDEV_AUX_CORRUPT_DATA);
1960}
1961
1962/*
1963 * The special vdev case is used for hot spares and l2cache devices. Its
1964 * sole purpose it to set the vdev state for the associated vdev. To do this,
1965 * we make sure that we can open the underlying device, then try to read the
1966 * label, and make sure that the label is sane and that it hasn't been
1967 * repurposed to another pool.
1968 */
1969int
1970vdev_validate_aux(vdev_t *vd)
1971{
1972 nvlist_t *label;
1973 uint64_t guid, version;
1974 uint64_t state;
1975
1976 if (!vdev_readable(vd))
1977 return (0);
1978
1979 if ((label = vdev_label_read_config(vd)) == NULL) {
1980 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1981 VDEV_AUX_CORRUPT_DATA);
1982 return (-1);
1983 }
1984
1985 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 ||
1986 version > SPA_VERSION ||
1987 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 ||
1988 guid != vd->vdev_guid ||
1989 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) {
1990 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1991 VDEV_AUX_CORRUPT_DATA);
1992 nvlist_free(label);
1993 return (-1);
1994 }
1995
1996 /*
1997 * We don't actually check the pool state here. If it's in fact in
1998 * use by another pool, we update this fact on the fly when requested.
1999 */
2000 nvlist_free(label);
2001 return (0);
2002}
2003
2004void
2005vdev_remove(vdev_t *vd, uint64_t txg)
2006{
2007 spa_t *spa = vd->vdev_spa;
2008 objset_t *mos = spa->spa_meta_objset;
2009 dmu_tx_t *tx;
2010
2011 tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
2012
2013 if (vd->vdev_dtl_smo.smo_object) {
2014 ASSERT3U(vd->vdev_dtl_smo.smo_alloc, ==, 0);
2015 (void) dmu_object_free(mos, vd->vdev_dtl_smo.smo_object, tx);
2016 vd->vdev_dtl_smo.smo_object = 0;
2017 }
2018
2019 if (vd->vdev_ms != NULL) {
2020 for (int m = 0; m < vd->vdev_ms_count; m++) {
2021 metaslab_t *msp = vd->vdev_ms[m];
2022
2023 if (msp == NULL || msp->ms_smo.smo_object == 0)
2024 continue;
2025
2026 ASSERT3U(msp->ms_smo.smo_alloc, ==, 0);
2027 (void) dmu_object_free(mos, msp->ms_smo.smo_object, tx);
2028 msp->ms_smo.smo_object = 0;
2029 }
2030 }
2031
2032 if (vd->vdev_ms_array) {
2033 (void) dmu_object_free(mos, vd->vdev_ms_array, tx);
2034 vd->vdev_ms_array = 0;
2035 vd->vdev_ms_shift = 0;
2036 }
2037 dmu_tx_commit(tx);
2038}
2039
2040void
2041vdev_sync_done(vdev_t *vd, uint64_t txg)
2042{
2043 metaslab_t *msp;
2044 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg));
2045
2046 ASSERT(!vd->vdev_ishole);
2047
2048 while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))
2049 metaslab_sync_done(msp, txg);
2050
2051 if (reassess)
2052 metaslab_sync_reassess(vd->vdev_mg);
2053}
2054
2055void
2056vdev_sync(vdev_t *vd, uint64_t txg)
2057{
2058 spa_t *spa = vd->vdev_spa;
2059 vdev_t *lvd;
2060 metaslab_t *msp;
2061 dmu_tx_t *tx;
2062
2063 ASSERT(!vd->vdev_ishole);
2064
2065 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) {
2066 ASSERT(vd == vd->vdev_top);
2067 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2068 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset,
2069 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx);
2070 ASSERT(vd->vdev_ms_array != 0);
2071 vdev_config_dirty(vd);
2072 dmu_tx_commit(tx);
2073 }
2074
2075 /*
2076 * Remove the metadata associated with this vdev once it's empty.
2077 */
2078 if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing)
2079 vdev_remove(vd, txg);
2080
2081 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) {
2082 metaslab_sync(msp, txg);
2083 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg));
2084 }
2085
2086 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL)
2087 vdev_dtl_sync(lvd, txg);
2088
2089 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg));
2090}
2091
2092uint64_t
2093vdev_psize_to_asize(vdev_t *vd, uint64_t psize)
2094{
2095 return (vd->vdev_ops->vdev_op_asize(vd, psize));
2096}
2097
2098/*
2099 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2100 * not be opened, and no I/O is attempted.
2101 */
2102int
2103vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2104{
2105 vdev_t *vd, *tvd;
2106
2107 spa_vdev_state_enter(spa, SCL_NONE);
2108
2109 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2110 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2111
2112 if (!vd->vdev_ops->vdev_op_leaf)
2113 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2114
2115 tvd = vd->vdev_top;
2116
2117 /*
2118 * We don't directly use the aux state here, but if we do a
2119 * vdev_reopen(), we need this value to be present to remember why we
2120 * were faulted.
2121 */
2122 vd->vdev_label_aux = aux;
2123
2124 /*
2125 * Faulted state takes precedence over degraded.
2126 */
2127 vd->vdev_delayed_close = B_FALSE;
2128 vd->vdev_faulted = 1ULL;
2129 vd->vdev_degraded = 0ULL;
2130 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux);
2131
2132 /*
2133 * If this device has the only valid copy of the data, then
2134 * back off and simply mark the vdev as degraded instead.
2135 */
2136 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) {
2137 vd->vdev_degraded = 1ULL;
2138 vd->vdev_faulted = 0ULL;
2139
2140 /*
2141 * If we reopen the device and it's not dead, only then do we
2142 * mark it degraded.
2143 */
2144 vdev_reopen(tvd);
2145
2146 if (vdev_readable(vd))
2147 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux);
2148 }
2149
2150 return (spa_vdev_state_exit(spa, vd, 0));
2151}
2152
2153/*
2154 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2155 * user that something is wrong. The vdev continues to operate as normal as far
2156 * as I/O is concerned.
2157 */
2158int
2159vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2160{
2161 vdev_t *vd;
2162
2163 spa_vdev_state_enter(spa, SCL_NONE);
2164
2165 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2166 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2167
2168 if (!vd->vdev_ops->vdev_op_leaf)
2169 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2170
2171 /*
2172 * If the vdev is already faulted, then don't do anything.
2173 */
2174 if (vd->vdev_faulted || vd->vdev_degraded)
2175 return (spa_vdev_state_exit(spa, NULL, 0));
2176
2177 vd->vdev_degraded = 1ULL;
2178 if (!vdev_is_dead(vd))
2179 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
2180 aux);
2181
2182 return (spa_vdev_state_exit(spa, vd, 0));
2183}
2184
2185/*
2186 * Online the given vdev. If 'unspare' is set, it implies two things. First,
2187 * any attached spare device should be detached when the device finishes
2188 * resilvering. Second, the online should be treated like a 'test' online case,
2189 * so no FMA events are generated if the device fails to open.
2190 */
2191int
2192vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
2193{
2194 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev;
2195
2196 spa_vdev_state_enter(spa, SCL_NONE);
2197
2198 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2199 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2200
2201 if (!vd->vdev_ops->vdev_op_leaf)
2202 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2203
2204 tvd = vd->vdev_top;
2205 vd->vdev_offline = B_FALSE;
2206 vd->vdev_tmpoffline = B_FALSE;
2207 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE);
2208 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT);
2209
2210 /* XXX - L2ARC 1.0 does not support expansion */
2211 if (!vd->vdev_aux) {
2212 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2213 pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND);
2214 }
2215
2216 vdev_reopen(tvd);
2217 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE;
2218
2219 if (!vd->vdev_aux) {
2220 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2221 pvd->vdev_expanding = B_FALSE;
2222 }
2223
2224 if (newstate)
2225 *newstate = vd->vdev_state;
2226 if ((flags & ZFS_ONLINE_UNSPARE) &&
2227 !vdev_is_dead(vd) && vd->vdev_parent &&
2228 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2229 vd->vdev_parent->vdev_child[0] == vd)
2230 vd->vdev_unspare = B_TRUE;
2231
2232 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) {
2233
2234 /* XXX - L2ARC 1.0 does not support expansion */
2235 if (vd->vdev_aux)
2236 return (spa_vdev_state_exit(spa, vd, ENOTSUP));
2237 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2238 }
2239 return (spa_vdev_state_exit(spa, vd, 0));
2240}
2241
2242static int
2243vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags)
2244{
2245 vdev_t *vd, *tvd;
2246 int error = 0;
2247 uint64_t generation;
2248 metaslab_group_t *mg;
2249
2250top:
2251 spa_vdev_state_enter(spa, SCL_ALLOC);
2252
2253 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2254 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2255
2256 if (!vd->vdev_ops->vdev_op_leaf)
2257 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2258
2259 tvd = vd->vdev_top;
2260 mg = tvd->vdev_mg;
2261 generation = spa->spa_config_generation + 1;
2262
2263 /*
2264 * If the device isn't already offline, try to offline it.
2265 */
2266 if (!vd->vdev_offline) {
2267 /*
2268 * If this device has the only valid copy of some data,
2269 * don't allow it to be offlined. Log devices are always
2270 * expendable.
2271 */
2272 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2273 vdev_dtl_required(vd))
2274 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2275
2276 /*
2277 * If the top-level is a slog and it has had allocations
2278 * then proceed. We check that the vdev's metaslab group
2279 * is not NULL since it's possible that we may have just
2280 * added this vdev but not yet initialized its metaslabs.
2281 */
2282 if (tvd->vdev_islog && mg != NULL) {
2283 /*
2284 * Prevent any future allocations.
2285 */
2286 metaslab_group_passivate(mg);
2287 (void) spa_vdev_state_exit(spa, vd, 0);
2288
2289 error = spa_offline_log(spa);
2290
2291 spa_vdev_state_enter(spa, SCL_ALLOC);
2292
2293 /*
2294 * Check to see if the config has changed.
2295 */
2296 if (error || generation != spa->spa_config_generation) {
2297 metaslab_group_activate(mg);
2298 if (error)
2299 return (spa_vdev_state_exit(spa,
2300 vd, error));
2301 (void) spa_vdev_state_exit(spa, vd, 0);
2302 goto top;
2303 }
2304 ASSERT3U(tvd->vdev_stat.vs_alloc, ==, 0);
2305 }
2306
2307 /*
2308 * Offline this device and reopen its top-level vdev.
2309 * If the top-level vdev is a log device then just offline
2310 * it. Otherwise, if this action results in the top-level
2311 * vdev becoming unusable, undo it and fail the request.
2312 */
2313 vd->vdev_offline = B_TRUE;
2314 vdev_reopen(tvd);
2315
2316 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2317 vdev_is_dead(tvd)) {
2318 vd->vdev_offline = B_FALSE;
2319 vdev_reopen(tvd);
2320 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2321 }
2322
2323 /*
2324 * Add the device back into the metaslab rotor so that
2325 * once we online the device it's open for business.
2326 */
2327 if (tvd->vdev_islog && mg != NULL)
2328 metaslab_group_activate(mg);
2329 }
2330
2331 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY);
2332
2333 return (spa_vdev_state_exit(spa, vd, 0));
2334}
2335
2336int
2337vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags)
2338{
2339 int error;
2340
2341 mutex_enter(&spa->spa_vdev_top_lock);
2342 error = vdev_offline_locked(spa, guid, flags);
2343 mutex_exit(&spa->spa_vdev_top_lock);
2344
2345 return (error);
2346}
2347
2348/*
2349 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2350 * vdev_offline(), we assume the spa config is locked. We also clear all
2351 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2352 */
2353void
2354vdev_clear(spa_t *spa, vdev_t *vd)
2355{
2356 vdev_t *rvd = spa->spa_root_vdev;
2357
2358 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2359
2360 if (vd == NULL)
2361 vd = rvd;
2362
2363 vd->vdev_stat.vs_read_errors = 0;
2364 vd->vdev_stat.vs_write_errors = 0;
2365 vd->vdev_stat.vs_checksum_errors = 0;
2366
2367 for (int c = 0; c < vd->vdev_children; c++)
2368 vdev_clear(spa, vd->vdev_child[c]);
2369
2370 /*
2371 * If we're in the FAULTED state or have experienced failed I/O, then
2372 * clear the persistent state and attempt to reopen the device. We
2373 * also mark the vdev config dirty, so that the new faulted state is
2374 * written out to disk.
2375 */
2376 if (vd->vdev_faulted || vd->vdev_degraded ||
2377 !vdev_readable(vd) || !vdev_writeable(vd)) {
2378
2379 /*
2380 * When reopening in reponse to a clear event, it may be due to
2381 * a fmadm repair request. In this case, if the device is
2382 * still broken, we want to still post the ereport again.
2383 */
2384 vd->vdev_forcefault = B_TRUE;
2385
2386 vd->vdev_faulted = vd->vdev_degraded = 0ULL;
2387 vd->vdev_cant_read = B_FALSE;
2388 vd->vdev_cant_write = B_FALSE;
2389
2390 vdev_reopen(vd == rvd ? rvd : vd->vdev_top);
2391
2392 vd->vdev_forcefault = B_FALSE;
2393
2394 if (vd != rvd && vdev_writeable(vd->vdev_top))
2395 vdev_state_dirty(vd->vdev_top);
2396
2397 if (vd->vdev_aux == NULL && !vdev_is_dead(vd))
2398 spa_async_request(spa, SPA_ASYNC_RESILVER);
2399
2400 spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR);
2401 }
2402
2403 /*
2404 * When clearing a FMA-diagnosed fault, we always want to
2405 * unspare the device, as we assume that the original spare was
2406 * done in response to the FMA fault.
2407 */
2408 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL &&
2409 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2410 vd->vdev_parent->vdev_child[0] == vd)
2411 vd->vdev_unspare = B_TRUE;
2412}
2413
2414boolean_t
2415vdev_is_dead(vdev_t *vd)
2416{
2417 /*
2418 * Holes and missing devices are always considered "dead".
2419 * This simplifies the code since we don't have to check for
2420 * these types of devices in the various code paths.
2421 * Instead we rely on the fact that we skip over dead devices
2422 * before issuing I/O to them.
2423 */
2424 return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole ||
2425 vd->vdev_ops == &vdev_missing_ops);
2426}
2427
2428boolean_t
2429vdev_readable(vdev_t *vd)
2430{
2431 return (!vdev_is_dead(vd) && !vd->vdev_cant_read);
2432}
2433
2434boolean_t
2435vdev_writeable(vdev_t *vd)
2436{
2437 return (!vdev_is_dead(vd) && !vd->vdev_cant_write);
2438}
2439
2440boolean_t
2441vdev_allocatable(vdev_t *vd)
2442{
2443 uint64_t state = vd->vdev_state;
2444
2445 /*
2446 * We currently allow allocations from vdevs which may be in the
2447 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2448 * fails to reopen then we'll catch it later when we're holding
2449 * the proper locks. Note that we have to get the vdev state
2450 * in a local variable because although it changes atomically,
2451 * we're asking two separate questions about it.
2452 */
2453 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) &&
2454 !vd->vdev_cant_write && !vd->vdev_ishole);
2455}
2456
2457boolean_t
2458vdev_accessible(vdev_t *vd, zio_t *zio)
2459{
2460 ASSERT(zio->io_vd == vd);
2461
2462 if (vdev_is_dead(vd) || vd->vdev_remove_wanted)
2463 return (B_FALSE);
2464
2465 if (zio->io_type == ZIO_TYPE_READ)
2466 return (!vd->vdev_cant_read);
2467
2468 if (zio->io_type == ZIO_TYPE_WRITE)
2469 return (!vd->vdev_cant_write);
2470
2471 return (B_TRUE);
2472}
2473
2474/*
2475 * Get statistics for the given vdev.
2476 */
2477void
2478vdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
2479{
2480 vdev_t *rvd = vd->vdev_spa->spa_root_vdev;
2481
2482 mutex_enter(&vd->vdev_stat_lock);
2483 bcopy(&vd->vdev_stat, vs, sizeof (*vs));
2484 vs->vs_timestamp = gethrtime() - vs->vs_timestamp;
2485 vs->vs_state = vd->vdev_state;
2486 vs->vs_rsize = vdev_get_min_asize(vd);
2487 if (vd->vdev_ops->vdev_op_leaf)
2488 vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
|
| 2489 vs->vs_esize = vd->vdev_max_asize - vd->vdev_asize; |
2490 mutex_exit(&vd->vdev_stat_lock);
2491
2492 /*
2493 * If we're getting stats on the root vdev, aggregate the I/O counts
2494 * over all top-level vdevs (i.e. the direct children of the root).
2495 */
2496 if (vd == rvd) {
2497 for (int c = 0; c < rvd->vdev_children; c++) {
2498 vdev_t *cvd = rvd->vdev_child[c];
2499 vdev_stat_t *cvs = &cvd->vdev_stat;
2500
2501 mutex_enter(&vd->vdev_stat_lock);
2502 for (int t = 0; t < ZIO_TYPES; t++) {
2503 vs->vs_ops[t] += cvs->vs_ops[t];
2504 vs->vs_bytes[t] += cvs->vs_bytes[t];
2505 }
2506 cvs->vs_scan_removing = cvd->vdev_removing;
2507 mutex_exit(&vd->vdev_stat_lock);
2508 }
2509 }
2510}
2511
2512void
2513vdev_clear_stats(vdev_t *vd)
2514{
2515 mutex_enter(&vd->vdev_stat_lock);
2516 vd->vdev_stat.vs_space = 0;
2517 vd->vdev_stat.vs_dspace = 0;
2518 vd->vdev_stat.vs_alloc = 0;
2519 mutex_exit(&vd->vdev_stat_lock);
2520}
2521
2522void
2523vdev_scan_stat_init(vdev_t *vd)
2524{
2525 vdev_stat_t *vs = &vd->vdev_stat;
2526
2527 for (int c = 0; c < vd->vdev_children; c++)
2528 vdev_scan_stat_init(vd->vdev_child[c]);
2529
2530 mutex_enter(&vd->vdev_stat_lock);
2531 vs->vs_scan_processed = 0;
2532 mutex_exit(&vd->vdev_stat_lock);
2533}
2534
2535void
2536vdev_stat_update(zio_t *zio, uint64_t psize)
2537{
2538 spa_t *spa = zio->io_spa;
2539 vdev_t *rvd = spa->spa_root_vdev;
2540 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd;
2541 vdev_t *pvd;
2542 uint64_t txg = zio->io_txg;
2543 vdev_stat_t *vs = &vd->vdev_stat;
2544 zio_type_t type = zio->io_type;
2545 int flags = zio->io_flags;
2546
2547 /*
2548 * If this i/o is a gang leader, it didn't do any actual work.
2549 */
2550 if (zio->io_gang_tree)
2551 return;
2552
2553 if (zio->io_error == 0) {
2554 /*
2555 * If this is a root i/o, don't count it -- we've already
2556 * counted the top-level vdevs, and vdev_get_stats() will
2557 * aggregate them when asked. This reduces contention on
2558 * the root vdev_stat_lock and implicitly handles blocks
2559 * that compress away to holes, for which there is no i/o.
2560 * (Holes never create vdev children, so all the counters
2561 * remain zero, which is what we want.)
2562 *
2563 * Note: this only applies to successful i/o (io_error == 0)
2564 * because unlike i/o counts, errors are not additive.
2565 * When reading a ditto block, for example, failure of
2566 * one top-level vdev does not imply a root-level error.
2567 */
2568 if (vd == rvd)
2569 return;
2570
2571 ASSERT(vd == zio->io_vd);
2572
2573 if (flags & ZIO_FLAG_IO_BYPASS)
2574 return;
2575
2576 mutex_enter(&vd->vdev_stat_lock);
2577
2578 if (flags & ZIO_FLAG_IO_REPAIR) {
2579 if (flags & ZIO_FLAG_SCAN_THREAD) {
2580 dsl_scan_phys_t *scn_phys =
2581 &spa->spa_dsl_pool->dp_scan->scn_phys;
2582 uint64_t *processed = &scn_phys->scn_processed;
2583
2584 /* XXX cleanup? */
2585 if (vd->vdev_ops->vdev_op_leaf)
2586 atomic_add_64(processed, psize);
2587 vs->vs_scan_processed += psize;
2588 }
2589
2590 if (flags & ZIO_FLAG_SELF_HEAL)
2591 vs->vs_self_healed += psize;
2592 }
2593
2594 vs->vs_ops[type]++;
2595 vs->vs_bytes[type] += psize;
2596
2597 mutex_exit(&vd->vdev_stat_lock);
2598 return;
2599 }
2600
2601 if (flags & ZIO_FLAG_SPECULATIVE)
2602 return;
2603
2604 /*
2605 * If this is an I/O error that is going to be retried, then ignore the
2606 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
2607 * hard errors, when in reality they can happen for any number of
2608 * innocuous reasons (bus resets, MPxIO link failure, etc).
2609 */
2610 if (zio->io_error == EIO &&
2611 !(zio->io_flags & ZIO_FLAG_IO_RETRY))
2612 return;
2613
2614 /*
2615 * Intent logs writes won't propagate their error to the root
2616 * I/O so don't mark these types of failures as pool-level
2617 * errors.
2618 */
2619 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE))
2620 return;
2621
2622 mutex_enter(&vd->vdev_stat_lock);
2623 if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) {
2624 if (zio->io_error == ECKSUM)
2625 vs->vs_checksum_errors++;
2626 else
2627 vs->vs_read_errors++;
2628 }
2629 if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd))
2630 vs->vs_write_errors++;
2631 mutex_exit(&vd->vdev_stat_lock);
2632
2633 if (type == ZIO_TYPE_WRITE && txg != 0 &&
2634 (!(flags & ZIO_FLAG_IO_REPAIR) ||
2635 (flags & ZIO_FLAG_SCAN_THREAD) ||
2636 spa->spa_claiming)) {
2637 /*
2638 * This is either a normal write (not a repair), or it's
2639 * a repair induced by the scrub thread, or it's a repair
2640 * made by zil_claim() during spa_load() in the first txg.
2641 * In the normal case, we commit the DTL change in the same
2642 * txg as the block was born. In the scrub-induced repair
2643 * case, we know that scrubs run in first-pass syncing context,
2644 * so we commit the DTL change in spa_syncing_txg(spa).
2645 * In the zil_claim() case, we commit in spa_first_txg(spa).
2646 *
2647 * We currently do not make DTL entries for failed spontaneous
2648 * self-healing writes triggered by normal (non-scrubbing)
2649 * reads, because we have no transactional context in which to
2650 * do so -- and it's not clear that it'd be desirable anyway.
2651 */
2652 if (vd->vdev_ops->vdev_op_leaf) {
2653 uint64_t commit_txg = txg;
2654 if (flags & ZIO_FLAG_SCAN_THREAD) {
2655 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2656 ASSERT(spa_sync_pass(spa) == 1);
2657 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1);
2658 commit_txg = spa_syncing_txg(spa);
2659 } else if (spa->spa_claiming) {
2660 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2661 commit_txg = spa_first_txg(spa);
2662 }
2663 ASSERT(commit_txg >= spa_syncing_txg(spa));
2664 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1))
2665 return;
2666 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2667 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1);
2668 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg);
2669 }
2670 if (vd != rvd)
2671 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1);
2672 }
2673}
2674
2675/*
2676 * Update the in-core space usage stats for this vdev, its metaslab class,
2677 * and the root vdev.
2678 */
2679void
2680vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta,
2681 int64_t space_delta)
2682{
2683 int64_t dspace_delta = space_delta;
2684 spa_t *spa = vd->vdev_spa;
2685 vdev_t *rvd = spa->spa_root_vdev;
2686 metaslab_group_t *mg = vd->vdev_mg;
2687 metaslab_class_t *mc = mg ? mg->mg_class : NULL;
2688
2689 ASSERT(vd == vd->vdev_top);
2690
2691 /*
2692 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
2693 * factor. We must calculate this here and not at the root vdev
2694 * because the root vdev's psize-to-asize is simply the max of its
2695 * childrens', thus not accurate enough for us.
2696 */
2697 ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0);
2698 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache);
2699 dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) *
2700 vd->vdev_deflate_ratio;
2701
2702 mutex_enter(&vd->vdev_stat_lock);
2703 vd->vdev_stat.vs_alloc += alloc_delta;
2704 vd->vdev_stat.vs_space += space_delta;
2705 vd->vdev_stat.vs_dspace += dspace_delta;
2706 mutex_exit(&vd->vdev_stat_lock);
2707
2708 if (mc == spa_normal_class(spa)) {
2709 mutex_enter(&rvd->vdev_stat_lock);
2710 rvd->vdev_stat.vs_alloc += alloc_delta;
2711 rvd->vdev_stat.vs_space += space_delta;
2712 rvd->vdev_stat.vs_dspace += dspace_delta;
2713 mutex_exit(&rvd->vdev_stat_lock);
2714 }
2715
2716 if (mc != NULL) {
2717 ASSERT(rvd == vd->vdev_parent);
2718 ASSERT(vd->vdev_ms_count != 0);
2719
2720 metaslab_class_space_update(mc,
2721 alloc_delta, defer_delta, space_delta, dspace_delta);
2722 }
2723}
2724
2725/*
2726 * Mark a top-level vdev's config as dirty, placing it on the dirty list
2727 * so that it will be written out next time the vdev configuration is synced.
2728 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
2729 */
2730void
2731vdev_config_dirty(vdev_t *vd)
2732{
2733 spa_t *spa = vd->vdev_spa;
2734 vdev_t *rvd = spa->spa_root_vdev;
2735 int c;
2736
2737 ASSERT(spa_writeable(spa));
2738
2739 /*
2740 * If this is an aux vdev (as with l2cache and spare devices), then we
2741 * update the vdev config manually and set the sync flag.
2742 */
2743 if (vd->vdev_aux != NULL) {
2744 spa_aux_vdev_t *sav = vd->vdev_aux;
2745 nvlist_t **aux;
2746 uint_t naux;
2747
2748 for (c = 0; c < sav->sav_count; c++) {
2749 if (sav->sav_vdevs[c] == vd)
2750 break;
2751 }
2752
2753 if (c == sav->sav_count) {
2754 /*
2755 * We're being removed. There's nothing more to do.
2756 */
2757 ASSERT(sav->sav_sync == B_TRUE);
2758 return;
2759 }
2760
2761 sav->sav_sync = B_TRUE;
2762
2763 if (nvlist_lookup_nvlist_array(sav->sav_config,
2764 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) {
2765 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
2766 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0);
2767 }
2768
2769 ASSERT(c < naux);
2770
2771 /*
2772 * Setting the nvlist in the middle if the array is a little
2773 * sketchy, but it will work.
2774 */
2775 nvlist_free(aux[c]);
2776 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0);
2777
2778 return;
2779 }
2780
2781 /*
2782 * The dirty list is protected by the SCL_CONFIG lock. The caller
2783 * must either hold SCL_CONFIG as writer, or must be the sync thread
2784 * (which holds SCL_CONFIG as reader). There's only one sync thread,
2785 * so this is sufficient to ensure mutual exclusion.
2786 */
2787 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
2788 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2789 spa_config_held(spa, SCL_CONFIG, RW_READER)));
2790
2791 if (vd == rvd) {
2792 for (c = 0; c < rvd->vdev_children; c++)
2793 vdev_config_dirty(rvd->vdev_child[c]);
2794 } else {
2795 ASSERT(vd == vd->vdev_top);
2796
2797 if (!list_link_active(&vd->vdev_config_dirty_node) &&
2798 !vd->vdev_ishole)
2799 list_insert_head(&spa->spa_config_dirty_list, vd);
2800 }
2801}
2802
2803void
2804vdev_config_clean(vdev_t *vd)
2805{
2806 spa_t *spa = vd->vdev_spa;
2807
2808 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
2809 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2810 spa_config_held(spa, SCL_CONFIG, RW_READER)));
2811
2812 ASSERT(list_link_active(&vd->vdev_config_dirty_node));
2813 list_remove(&spa->spa_config_dirty_list, vd);
2814}
2815
2816/*
2817 * Mark a top-level vdev's state as dirty, so that the next pass of
2818 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
2819 * the state changes from larger config changes because they require
2820 * much less locking, and are often needed for administrative actions.
2821 */
2822void
2823vdev_state_dirty(vdev_t *vd)
2824{
2825 spa_t *spa = vd->vdev_spa;
2826
2827 ASSERT(spa_writeable(spa));
2828 ASSERT(vd == vd->vdev_top);
2829
2830 /*
2831 * The state list is protected by the SCL_STATE lock. The caller
2832 * must either hold SCL_STATE as writer, or must be the sync thread
2833 * (which holds SCL_STATE as reader). There's only one sync thread,
2834 * so this is sufficient to ensure mutual exclusion.
2835 */
2836 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
2837 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2838 spa_config_held(spa, SCL_STATE, RW_READER)));
2839
2840 if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole)
2841 list_insert_head(&spa->spa_state_dirty_list, vd);
2842}
2843
2844void
2845vdev_state_clean(vdev_t *vd)
2846{
2847 spa_t *spa = vd->vdev_spa;
2848
2849 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
2850 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2851 spa_config_held(spa, SCL_STATE, RW_READER)));
2852
2853 ASSERT(list_link_active(&vd->vdev_state_dirty_node));
2854 list_remove(&spa->spa_state_dirty_list, vd);
2855}
2856
2857/*
2858 * Propagate vdev state up from children to parent.
2859 */
2860void
2861vdev_propagate_state(vdev_t *vd)
2862{
2863 spa_t *spa = vd->vdev_spa;
2864 vdev_t *rvd = spa->spa_root_vdev;
2865 int degraded = 0, faulted = 0;
2866 int corrupted = 0;
2867 vdev_t *child;
2868
2869 if (vd->vdev_children > 0) {
2870 for (int c = 0; c < vd->vdev_children; c++) {
2871 child = vd->vdev_child[c];
2872
2873 /*
2874 * Don't factor holes into the decision.
2875 */
2876 if (child->vdev_ishole)
2877 continue;
2878
2879 if (!vdev_readable(child) ||
2880 (!vdev_writeable(child) && spa_writeable(spa))) {
2881 /*
2882 * Root special: if there is a top-level log
2883 * device, treat the root vdev as if it were
2884 * degraded.
2885 */
2886 if (child->vdev_islog && vd == rvd)
2887 degraded++;
2888 else
2889 faulted++;
2890 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) {
2891 degraded++;
2892 }
2893
2894 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA)
2895 corrupted++;
2896 }
2897
2898 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded);
2899
2900 /*
2901 * Root special: if there is a top-level vdev that cannot be
2902 * opened due to corrupted metadata, then propagate the root
2903 * vdev's aux state as 'corrupt' rather than 'insufficient
2904 * replicas'.
2905 */
2906 if (corrupted && vd == rvd &&
2907 rvd->vdev_state == VDEV_STATE_CANT_OPEN)
2908 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN,
2909 VDEV_AUX_CORRUPT_DATA);
2910 }
2911
2912 if (vd->vdev_parent)
2913 vdev_propagate_state(vd->vdev_parent);
2914}
2915
2916/*
2917 * Set a vdev's state. If this is during an open, we don't update the parent
2918 * state, because we're in the process of opening children depth-first.
2919 * Otherwise, we propagate the change to the parent.
2920 *
2921 * If this routine places a device in a faulted state, an appropriate ereport is
2922 * generated.
2923 */
2924void
2925vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
2926{
2927 uint64_t save_state;
2928 spa_t *spa = vd->vdev_spa;
2929
2930 if (state == vd->vdev_state) {
2931 vd->vdev_stat.vs_aux = aux;
2932 return;
2933 }
2934
2935 save_state = vd->vdev_state;
2936
2937 vd->vdev_state = state;
2938 vd->vdev_stat.vs_aux = aux;
2939
2940 /*
2941 * If we are setting the vdev state to anything but an open state, then
2942 * always close the underlying device unless the device has requested
2943 * a delayed close (i.e. we're about to remove or fault the device).
2944 * Otherwise, we keep accessible but invalid devices open forever.
2945 * We don't call vdev_close() itself, because that implies some extra
2946 * checks (offline, etc) that we don't want here. This is limited to
2947 * leaf devices, because otherwise closing the device will affect other
2948 * children.
2949 */
2950 if (!vd->vdev_delayed_close && vdev_is_dead(vd) &&
2951 vd->vdev_ops->vdev_op_leaf)
2952 vd->vdev_ops->vdev_op_close(vd);
2953
2954 /*
2955 * If we have brought this vdev back into service, we need
2956 * to notify fmd so that it can gracefully repair any outstanding
2957 * cases due to a missing device. We do this in all cases, even those
2958 * that probably don't correlate to a repaired fault. This is sure to
2959 * catch all cases, and we let the zfs-retire agent sort it out. If
2960 * this is a transient state it's OK, as the retire agent will
2961 * double-check the state of the vdev before repairing it.
2962 */
2963 if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf &&
2964 vd->vdev_prevstate != state)
2965 zfs_post_state_change(spa, vd);
2966
2967 if (vd->vdev_removed &&
2968 state == VDEV_STATE_CANT_OPEN &&
2969 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) {
2970 /*
2971 * If the previous state is set to VDEV_STATE_REMOVED, then this
2972 * device was previously marked removed and someone attempted to
2973 * reopen it. If this failed due to a nonexistent device, then
2974 * keep the device in the REMOVED state. We also let this be if
2975 * it is one of our special test online cases, which is only
2976 * attempting to online the device and shouldn't generate an FMA
2977 * fault.
2978 */
2979 vd->vdev_state = VDEV_STATE_REMOVED;
2980 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
2981 } else if (state == VDEV_STATE_REMOVED) {
2982 vd->vdev_removed = B_TRUE;
2983 } else if (state == VDEV_STATE_CANT_OPEN) {
2984 /*
2985 * If we fail to open a vdev during an import or recovery, we
2986 * mark it as "not available", which signifies that it was
2987 * never there to begin with. Failure to open such a device
2988 * is not considered an error.
2989 */
2990 if ((spa_load_state(spa) == SPA_LOAD_IMPORT ||
2991 spa_load_state(spa) == SPA_LOAD_RECOVER) &&
2992 vd->vdev_ops->vdev_op_leaf)
2993 vd->vdev_not_present = 1;
2994
2995 /*
2996 * Post the appropriate ereport. If the 'prevstate' field is
2997 * set to something other than VDEV_STATE_UNKNOWN, it indicates
2998 * that this is part of a vdev_reopen(). In this case, we don't
2999 * want to post the ereport if the device was already in the
3000 * CANT_OPEN state beforehand.
3001 *
3002 * If the 'checkremove' flag is set, then this is an attempt to
3003 * online the device in response to an insertion event. If we
3004 * hit this case, then we have detected an insertion event for a
3005 * faulted or offline device that wasn't in the removed state.
3006 * In this scenario, we don't post an ereport because we are
3007 * about to replace the device, or attempt an online with
3008 * vdev_forcefault, which will generate the fault for us.
3009 */
3010 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) &&
3011 !vd->vdev_not_present && !vd->vdev_checkremove &&
3012 vd != spa->spa_root_vdev) {
3013 const char *class;
3014
3015 switch (aux) {
3016 case VDEV_AUX_OPEN_FAILED:
3017 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED;
3018 break;
3019 case VDEV_AUX_CORRUPT_DATA:
3020 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA;
3021 break;
3022 case VDEV_AUX_NO_REPLICAS:
3023 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS;
3024 break;
3025 case VDEV_AUX_BAD_GUID_SUM:
3026 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM;
3027 break;
3028 case VDEV_AUX_TOO_SMALL:
3029 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL;
3030 break;
3031 case VDEV_AUX_BAD_LABEL:
3032 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL;
3033 break;
3034 default:
3035 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN;
3036 }
3037
3038 zfs_ereport_post(class, spa, vd, NULL, save_state, 0);
3039 }
3040
3041 /* Erase any notion of persistent removed state */
3042 vd->vdev_removed = B_FALSE;
3043 } else {
3044 vd->vdev_removed = B_FALSE;
3045 }
3046
3047 if (!isopen && vd->vdev_parent)
3048 vdev_propagate_state(vd->vdev_parent);
3049}
3050
3051/*
3052 * Check the vdev configuration to ensure that it's capable of supporting
3053 * a root pool.
3054 *
3055 * On Solaris, we do not support RAID-Z or partial configuration. In
3056 * addition, only a single top-level vdev is allowed and none of the
3057 * leaves can be wholedisks.
3058 *
3059 * For FreeBSD, we can boot from any configuration. There is a
3060 * limitation that the boot filesystem must be either uncompressed or
3061 * compresses with lzjb compression but I'm not sure how to enforce
3062 * that here.
3063 */
3064boolean_t
3065vdev_is_bootable(vdev_t *vd)
3066{
3067#ifdef sun
3068 if (!vd->vdev_ops->vdev_op_leaf) {
3069 char *vdev_type = vd->vdev_ops->vdev_op_type;
3070
3071 if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 &&
3072 vd->vdev_children > 1) {
3073 return (B_FALSE);
3074 } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 ||
3075 strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) {
3076 return (B_FALSE);
3077 }
3078 } else if (vd->vdev_wholedisk == 1) {
3079 return (B_FALSE);
3080 }
3081
3082 for (int c = 0; c < vd->vdev_children; c++) {
3083 if (!vdev_is_bootable(vd->vdev_child[c]))
3084 return (B_FALSE);
3085 }
3086#endif /* sun */
3087 return (B_TRUE);
3088}
3089
3090/*
3091 * Load the state from the original vdev tree (ovd) which
3092 * we've retrieved from the MOS config object. If the original
3093 * vdev was offline or faulted then we transfer that state to the
3094 * device in the current vdev tree (nvd).
3095 */
3096void
3097vdev_load_log_state(vdev_t *nvd, vdev_t *ovd)
3098{
3099 spa_t *spa = nvd->vdev_spa;
3100
3101 ASSERT(nvd->vdev_top->vdev_islog);
3102 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
3103 ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid);
3104
3105 for (int c = 0; c < nvd->vdev_children; c++)
3106 vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]);
3107
3108 if (nvd->vdev_ops->vdev_op_leaf) {
3109 /*
3110 * Restore the persistent vdev state
3111 */
3112 nvd->vdev_offline = ovd->vdev_offline;
3113 nvd->vdev_faulted = ovd->vdev_faulted;
3114 nvd->vdev_degraded = ovd->vdev_degraded;
3115 nvd->vdev_removed = ovd->vdev_removed;
3116 }
3117}
3118
3119/*
3120 * Determine if a log device has valid content. If the vdev was
3121 * removed or faulted in the MOS config then we know that
3122 * the content on the log device has already been written to the pool.
3123 */
3124boolean_t
3125vdev_log_state_valid(vdev_t *vd)
3126{
3127 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted &&
3128 !vd->vdev_removed)
3129 return (B_TRUE);
3130
3131 for (int c = 0; c < vd->vdev_children; c++)
3132 if (vdev_log_state_valid(vd->vdev_child[c]))
3133 return (B_TRUE);
3134
3135 return (B_FALSE);
3136}
3137
3138/*
3139 * Expand a vdev if possible.
3140 */
3141void
3142vdev_expand(vdev_t *vd, uint64_t txg)
3143{
3144 ASSERT(vd->vdev_top == vd);
3145 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3146
3147 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) {
3148 VERIFY(vdev_metaslab_init(vd, txg) == 0);
3149 vdev_config_dirty(vd);
3150 }
3151}
3152
3153/*
3154 * Split a vdev.
3155 */
3156void
3157vdev_split(vdev_t *vd)
3158{
3159 vdev_t *cvd, *pvd = vd->vdev_parent;
3160
3161 vdev_remove_child(pvd, vd);
3162 vdev_compact_children(pvd);
3163
3164 cvd = pvd->vdev_child[0];
3165 if (pvd->vdev_children == 1) {
3166 vdev_remove_parent(cvd);
3167 cvd->vdev_splitting = B_TRUE;
3168 }
3169 vdev_propagate_state(cvd);
3170}
|