]>
Commit | Line | Data |
---|---|---|
34dc7c2f BB |
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 | /* | |
428870ff | 23 | * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. |
3541dc6d | 24 | * Copyright 2011 Nexenta Systems, Inc. All rights reserved. |
c3520e7f | 25 | * Copyright (c) 2011, 2015 by Delphix. All rights reserved. |
34dc7c2f BB |
26 | */ |
27 | ||
34dc7c2f BB |
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> | |
93cf2076 | 39 | #include <sys/space_reftree.h> |
34dc7c2f BB |
40 | #include <sys/zio.h> |
41 | #include <sys/zap.h> | |
42 | #include <sys/fs/zfs.h> | |
b128c09f | 43 | #include <sys/arc.h> |
9babb374 | 44 | #include <sys/zil.h> |
428870ff | 45 | #include <sys/dsl_scan.h> |
a6255b7f | 46 | #include <sys/abd.h> |
6c285672 | 47 | #include <sys/zvol.h> |
6078881a | 48 | #include <sys/zfs_ratelimit.h> |
34dc7c2f | 49 | |
b8bcca18 MA |
50 | /* |
51 | * When a vdev is added, it will be divided into approximately (but no | |
52 | * more than) this number of metaslabs. | |
53 | */ | |
54 | int metaslabs_per_vdev = 200; | |
55 | ||
34dc7c2f BB |
56 | /* |
57 | * Virtual device management. | |
58 | */ | |
59 | ||
60 | static vdev_ops_t *vdev_ops_table[] = { | |
61 | &vdev_root_ops, | |
62 | &vdev_raidz_ops, | |
63 | &vdev_mirror_ops, | |
64 | &vdev_replacing_ops, | |
65 | &vdev_spare_ops, | |
66 | &vdev_disk_ops, | |
67 | &vdev_file_ops, | |
68 | &vdev_missing_ops, | |
428870ff | 69 | &vdev_hole_ops, |
34dc7c2f BB |
70 | NULL |
71 | }; | |
72 | ||
34dc7c2f BB |
73 | /* |
74 | * Given a vdev type, return the appropriate ops vector. | |
75 | */ | |
76 | static vdev_ops_t * | |
77 | vdev_getops(const char *type) | |
78 | { | |
79 | vdev_ops_t *ops, **opspp; | |
80 | ||
81 | for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++) | |
82 | if (strcmp(ops->vdev_op_type, type) == 0) | |
83 | break; | |
84 | ||
85 | return (ops); | |
86 | } | |
87 | ||
88 | /* | |
89 | * Default asize function: return the MAX of psize with the asize of | |
90 | * all children. This is what's used by anything other than RAID-Z. | |
91 | */ | |
92 | uint64_t | |
93 | vdev_default_asize(vdev_t *vd, uint64_t psize) | |
94 | { | |
95 | uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift); | |
96 | uint64_t csize; | |
d6320ddb | 97 | int c; |
34dc7c2f | 98 | |
d6320ddb | 99 | for (c = 0; c < vd->vdev_children; c++) { |
34dc7c2f BB |
100 | csize = vdev_psize_to_asize(vd->vdev_child[c], psize); |
101 | asize = MAX(asize, csize); | |
102 | } | |
103 | ||
104 | return (asize); | |
105 | } | |
106 | ||
107 | /* | |
9babb374 BB |
108 | * Get the minimum allocatable size. We define the allocatable size as |
109 | * the vdev's asize rounded to the nearest metaslab. This allows us to | |
110 | * replace or attach devices which don't have the same physical size but | |
111 | * can still satisfy the same number of allocations. | |
34dc7c2f BB |
112 | */ |
113 | uint64_t | |
9babb374 | 114 | vdev_get_min_asize(vdev_t *vd) |
34dc7c2f | 115 | { |
9babb374 | 116 | vdev_t *pvd = vd->vdev_parent; |
34dc7c2f | 117 | |
9babb374 | 118 | /* |
1bd201e7 | 119 | * If our parent is NULL (inactive spare or cache) or is the root, |
9babb374 BB |
120 | * just return our own asize. |
121 | */ | |
122 | if (pvd == NULL) | |
123 | return (vd->vdev_asize); | |
34dc7c2f BB |
124 | |
125 | /* | |
9babb374 BB |
126 | * The top-level vdev just returns the allocatable size rounded |
127 | * to the nearest metaslab. | |
34dc7c2f | 128 | */ |
9babb374 BB |
129 | if (vd == vd->vdev_top) |
130 | return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift)); | |
34dc7c2f | 131 | |
9babb374 BB |
132 | /* |
133 | * The allocatable space for a raidz vdev is N * sizeof(smallest child), | |
134 | * so each child must provide at least 1/Nth of its asize. | |
135 | */ | |
136 | if (pvd->vdev_ops == &vdev_raidz_ops) | |
137 | return (pvd->vdev_min_asize / pvd->vdev_children); | |
34dc7c2f | 138 | |
9babb374 BB |
139 | return (pvd->vdev_min_asize); |
140 | } | |
141 | ||
142 | void | |
143 | vdev_set_min_asize(vdev_t *vd) | |
144 | { | |
d6320ddb | 145 | int c; |
9babb374 | 146 | vd->vdev_min_asize = vdev_get_min_asize(vd); |
34dc7c2f | 147 | |
d6320ddb | 148 | for (c = 0; c < vd->vdev_children; c++) |
9babb374 | 149 | vdev_set_min_asize(vd->vdev_child[c]); |
34dc7c2f BB |
150 | } |
151 | ||
152 | vdev_t * | |
153 | vdev_lookup_top(spa_t *spa, uint64_t vdev) | |
154 | { | |
155 | vdev_t *rvd = spa->spa_root_vdev; | |
156 | ||
b128c09f | 157 | ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); |
34dc7c2f | 158 | |
b128c09f BB |
159 | if (vdev < rvd->vdev_children) { |
160 | ASSERT(rvd->vdev_child[vdev] != NULL); | |
34dc7c2f | 161 | return (rvd->vdev_child[vdev]); |
b128c09f | 162 | } |
34dc7c2f BB |
163 | |
164 | return (NULL); | |
165 | } | |
166 | ||
167 | vdev_t * | |
168 | vdev_lookup_by_guid(vdev_t *vd, uint64_t guid) | |
169 | { | |
34dc7c2f | 170 | vdev_t *mvd; |
d6320ddb | 171 | int c; |
34dc7c2f BB |
172 | |
173 | if (vd->vdev_guid == guid) | |
174 | return (vd); | |
175 | ||
d6320ddb | 176 | for (c = 0; c < vd->vdev_children; c++) |
34dc7c2f BB |
177 | if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) != |
178 | NULL) | |
179 | return (mvd); | |
180 | ||
181 | return (NULL); | |
182 | } | |
183 | ||
9c43027b AJ |
184 | static int |
185 | vdev_count_leaves_impl(vdev_t *vd) | |
186 | { | |
187 | int n = 0; | |
188 | int c; | |
189 | ||
190 | if (vd->vdev_ops->vdev_op_leaf) | |
191 | return (1); | |
192 | ||
193 | for (c = 0; c < vd->vdev_children; c++) | |
194 | n += vdev_count_leaves_impl(vd->vdev_child[c]); | |
195 | ||
196 | return (n); | |
197 | } | |
198 | ||
199 | int | |
200 | vdev_count_leaves(spa_t *spa) | |
201 | { | |
202 | return (vdev_count_leaves_impl(spa->spa_root_vdev)); | |
203 | } | |
204 | ||
34dc7c2f BB |
205 | void |
206 | vdev_add_child(vdev_t *pvd, vdev_t *cvd) | |
207 | { | |
208 | size_t oldsize, newsize; | |
209 | uint64_t id = cvd->vdev_id; | |
210 | vdev_t **newchild; | |
211 | ||
44de2f02 | 212 | ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
34dc7c2f BB |
213 | ASSERT(cvd->vdev_parent == NULL); |
214 | ||
215 | cvd->vdev_parent = pvd; | |
216 | ||
217 | if (pvd == NULL) | |
218 | return; | |
219 | ||
220 | ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL); | |
221 | ||
222 | oldsize = pvd->vdev_children * sizeof (vdev_t *); | |
223 | pvd->vdev_children = MAX(pvd->vdev_children, id + 1); | |
224 | newsize = pvd->vdev_children * sizeof (vdev_t *); | |
225 | ||
79c76d5b | 226 | newchild = kmem_alloc(newsize, KM_SLEEP); |
34dc7c2f BB |
227 | if (pvd->vdev_child != NULL) { |
228 | bcopy(pvd->vdev_child, newchild, oldsize); | |
229 | kmem_free(pvd->vdev_child, oldsize); | |
230 | } | |
231 | ||
232 | pvd->vdev_child = newchild; | |
233 | pvd->vdev_child[id] = cvd; | |
234 | ||
235 | cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd); | |
236 | ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL); | |
237 | ||
238 | /* | |
239 | * Walk up all ancestors to update guid sum. | |
240 | */ | |
241 | for (; pvd != NULL; pvd = pvd->vdev_parent) | |
242 | pvd->vdev_guid_sum += cvd->vdev_guid_sum; | |
34dc7c2f BB |
243 | } |
244 | ||
245 | void | |
246 | vdev_remove_child(vdev_t *pvd, vdev_t *cvd) | |
247 | { | |
248 | int c; | |
249 | uint_t id = cvd->vdev_id; | |
250 | ||
251 | ASSERT(cvd->vdev_parent == pvd); | |
252 | ||
253 | if (pvd == NULL) | |
254 | return; | |
255 | ||
256 | ASSERT(id < pvd->vdev_children); | |
257 | ASSERT(pvd->vdev_child[id] == cvd); | |
258 | ||
259 | pvd->vdev_child[id] = NULL; | |
260 | cvd->vdev_parent = NULL; | |
261 | ||
262 | for (c = 0; c < pvd->vdev_children; c++) | |
263 | if (pvd->vdev_child[c]) | |
264 | break; | |
265 | ||
266 | if (c == pvd->vdev_children) { | |
267 | kmem_free(pvd->vdev_child, c * sizeof (vdev_t *)); | |
268 | pvd->vdev_child = NULL; | |
269 | pvd->vdev_children = 0; | |
270 | } | |
271 | ||
272 | /* | |
273 | * Walk up all ancestors to update guid sum. | |
274 | */ | |
275 | for (; pvd != NULL; pvd = pvd->vdev_parent) | |
276 | pvd->vdev_guid_sum -= cvd->vdev_guid_sum; | |
34dc7c2f BB |
277 | } |
278 | ||
279 | /* | |
280 | * Remove any holes in the child array. | |
281 | */ | |
282 | void | |
283 | vdev_compact_children(vdev_t *pvd) | |
284 | { | |
285 | vdev_t **newchild, *cvd; | |
286 | int oldc = pvd->vdev_children; | |
9babb374 | 287 | int newc; |
d6320ddb | 288 | int c; |
34dc7c2f | 289 | |
b128c09f | 290 | ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
34dc7c2f | 291 | |
d6320ddb | 292 | for (c = newc = 0; c < oldc; c++) |
34dc7c2f BB |
293 | if (pvd->vdev_child[c]) |
294 | newc++; | |
295 | ||
79c76d5b | 296 | newchild = kmem_zalloc(newc * sizeof (vdev_t *), KM_SLEEP); |
34dc7c2f | 297 | |
d6320ddb | 298 | for (c = newc = 0; c < oldc; c++) { |
34dc7c2f BB |
299 | if ((cvd = pvd->vdev_child[c]) != NULL) { |
300 | newchild[newc] = cvd; | |
301 | cvd->vdev_id = newc++; | |
302 | } | |
303 | } | |
304 | ||
305 | kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *)); | |
306 | pvd->vdev_child = newchild; | |
307 | pvd->vdev_children = newc; | |
308 | } | |
309 | ||
310 | /* | |
311 | * Allocate and minimally initialize a vdev_t. | |
312 | */ | |
428870ff | 313 | vdev_t * |
34dc7c2f BB |
314 | vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops) |
315 | { | |
316 | vdev_t *vd; | |
d6320ddb | 317 | int t; |
34dc7c2f | 318 | |
79c76d5b | 319 | vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP); |
34dc7c2f BB |
320 | |
321 | if (spa->spa_root_vdev == NULL) { | |
322 | ASSERT(ops == &vdev_root_ops); | |
323 | spa->spa_root_vdev = vd; | |
3541dc6d | 324 | spa->spa_load_guid = spa_generate_guid(NULL); |
34dc7c2f BB |
325 | } |
326 | ||
428870ff | 327 | if (guid == 0 && ops != &vdev_hole_ops) { |
34dc7c2f BB |
328 | if (spa->spa_root_vdev == vd) { |
329 | /* | |
330 | * The root vdev's guid will also be the pool guid, | |
331 | * which must be unique among all pools. | |
332 | */ | |
428870ff | 333 | guid = spa_generate_guid(NULL); |
34dc7c2f BB |
334 | } else { |
335 | /* | |
336 | * Any other vdev's guid must be unique within the pool. | |
337 | */ | |
428870ff | 338 | guid = spa_generate_guid(spa); |
34dc7c2f BB |
339 | } |
340 | ASSERT(!spa_guid_exists(spa_guid(spa), guid)); | |
341 | } | |
342 | ||
343 | vd->vdev_spa = spa; | |
344 | vd->vdev_id = id; | |
345 | vd->vdev_guid = guid; | |
346 | vd->vdev_guid_sum = guid; | |
347 | vd->vdev_ops = ops; | |
348 | vd->vdev_state = VDEV_STATE_CLOSED; | |
428870ff | 349 | vd->vdev_ishole = (ops == &vdev_hole_ops); |
34dc7c2f | 350 | |
6078881a TH |
351 | /* |
352 | * Initialize rate limit structs for events. We rate limit ZIO delay | |
353 | * and checksum events so that we don't overwhelm ZED with thousands | |
354 | * of events when a disk is acting up. | |
355 | */ | |
356 | zfs_ratelimit_init(&vd->vdev_delay_rl, DELAYS_PER_SECOND, 1); | |
357 | zfs_ratelimit_init(&vd->vdev_checksum_rl, CHECKSUMS_PER_SECOND, 1); | |
358 | ||
98f72a53 BB |
359 | list_link_init(&vd->vdev_config_dirty_node); |
360 | list_link_init(&vd->vdev_state_dirty_node); | |
448d7aaa | 361 | mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_NOLOCKDEP, NULL); |
34dc7c2f | 362 | mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL); |
b128c09f | 363 | mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL); |
3dfb57a3 | 364 | mutex_init(&vd->vdev_queue_lock, NULL, MUTEX_DEFAULT, NULL); |
6078881a | 365 | |
d6320ddb | 366 | for (t = 0; t < DTL_TYPES; t++) { |
93cf2076 | 367 | vd->vdev_dtl[t] = range_tree_create(NULL, NULL, |
fb5f0bc8 BB |
368 | &vd->vdev_dtl_lock); |
369 | } | |
34dc7c2f BB |
370 | txg_list_create(&vd->vdev_ms_list, |
371 | offsetof(struct metaslab, ms_txg_node)); | |
372 | txg_list_create(&vd->vdev_dtl_list, | |
373 | offsetof(struct vdev, vdev_dtl_node)); | |
374 | vd->vdev_stat.vs_timestamp = gethrtime(); | |
375 | vdev_queue_init(vd); | |
376 | vdev_cache_init(vd); | |
377 | ||
378 | return (vd); | |
379 | } | |
380 | ||
381 | /* | |
382 | * Allocate a new vdev. The 'alloctype' is used to control whether we are | |
383 | * creating a new vdev or loading an existing one - the behavior is slightly | |
384 | * different for each case. | |
385 | */ | |
386 | int | |
387 | vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id, | |
388 | int alloctype) | |
389 | { | |
390 | vdev_ops_t *ops; | |
391 | char *type; | |
392 | uint64_t guid = 0, islog, nparity; | |
393 | vdev_t *vd; | |
394 | ||
b128c09f | 395 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
34dc7c2f BB |
396 | |
397 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0) | |
2e528b49 | 398 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
399 | |
400 | if ((ops = vdev_getops(type)) == NULL) | |
2e528b49 | 401 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
402 | |
403 | /* | |
404 | * If this is a load, get the vdev guid from the nvlist. | |
405 | * Otherwise, vdev_alloc_common() will generate one for us. | |
406 | */ | |
407 | if (alloctype == VDEV_ALLOC_LOAD) { | |
408 | uint64_t label_id; | |
409 | ||
410 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) || | |
411 | label_id != id) | |
2e528b49 | 412 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
413 | |
414 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) | |
2e528b49 | 415 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
416 | } else if (alloctype == VDEV_ALLOC_SPARE) { |
417 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) | |
2e528b49 | 418 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
419 | } else if (alloctype == VDEV_ALLOC_L2CACHE) { |
420 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) | |
2e528b49 | 421 | return (SET_ERROR(EINVAL)); |
9babb374 BB |
422 | } else if (alloctype == VDEV_ALLOC_ROOTPOOL) { |
423 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) | |
2e528b49 | 424 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
425 | } |
426 | ||
427 | /* | |
428 | * The first allocated vdev must be of type 'root'. | |
429 | */ | |
430 | if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL) | |
2e528b49 | 431 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
432 | |
433 | /* | |
434 | * Determine whether we're a log vdev. | |
435 | */ | |
436 | islog = 0; | |
437 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog); | |
438 | if (islog && spa_version(spa) < SPA_VERSION_SLOGS) | |
2e528b49 | 439 | return (SET_ERROR(ENOTSUP)); |
34dc7c2f | 440 | |
428870ff | 441 | if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES) |
2e528b49 | 442 | return (SET_ERROR(ENOTSUP)); |
428870ff | 443 | |
34dc7c2f BB |
444 | /* |
445 | * Set the nparity property for RAID-Z vdevs. | |
446 | */ | |
447 | nparity = -1ULL; | |
448 | if (ops == &vdev_raidz_ops) { | |
449 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY, | |
450 | &nparity) == 0) { | |
428870ff | 451 | if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY) |
2e528b49 | 452 | return (SET_ERROR(EINVAL)); |
34dc7c2f | 453 | /* |
45d1cae3 BB |
454 | * Previous versions could only support 1 or 2 parity |
455 | * device. | |
34dc7c2f | 456 | */ |
45d1cae3 BB |
457 | if (nparity > 1 && |
458 | spa_version(spa) < SPA_VERSION_RAIDZ2) | |
2e528b49 | 459 | return (SET_ERROR(ENOTSUP)); |
45d1cae3 BB |
460 | if (nparity > 2 && |
461 | spa_version(spa) < SPA_VERSION_RAIDZ3) | |
2e528b49 | 462 | return (SET_ERROR(ENOTSUP)); |
34dc7c2f BB |
463 | } else { |
464 | /* | |
465 | * We require the parity to be specified for SPAs that | |
466 | * support multiple parity levels. | |
467 | */ | |
45d1cae3 | 468 | if (spa_version(spa) >= SPA_VERSION_RAIDZ2) |
2e528b49 | 469 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
470 | /* |
471 | * Otherwise, we default to 1 parity device for RAID-Z. | |
472 | */ | |
473 | nparity = 1; | |
474 | } | |
475 | } else { | |
476 | nparity = 0; | |
477 | } | |
478 | ASSERT(nparity != -1ULL); | |
479 | ||
480 | vd = vdev_alloc_common(spa, id, guid, ops); | |
481 | ||
482 | vd->vdev_islog = islog; | |
483 | vd->vdev_nparity = nparity; | |
484 | ||
485 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0) | |
486 | vd->vdev_path = spa_strdup(vd->vdev_path); | |
487 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0) | |
488 | vd->vdev_devid = spa_strdup(vd->vdev_devid); | |
489 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH, | |
490 | &vd->vdev_physpath) == 0) | |
491 | vd->vdev_physpath = spa_strdup(vd->vdev_physpath); | |
1bbd8770 TH |
492 | |
493 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH, | |
494 | &vd->vdev_enc_sysfs_path) == 0) | |
495 | vd->vdev_enc_sysfs_path = spa_strdup(vd->vdev_enc_sysfs_path); | |
496 | ||
9babb374 BB |
497 | if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0) |
498 | vd->vdev_fru = spa_strdup(vd->vdev_fru); | |
34dc7c2f BB |
499 | |
500 | /* | |
501 | * Set the whole_disk property. If it's not specified, leave the value | |
502 | * as -1. | |
503 | */ | |
504 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, | |
505 | &vd->vdev_wholedisk) != 0) | |
506 | vd->vdev_wholedisk = -1ULL; | |
507 | ||
508 | /* | |
509 | * Look for the 'not present' flag. This will only be set if the device | |
510 | * was not present at the time of import. | |
511 | */ | |
9babb374 BB |
512 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, |
513 | &vd->vdev_not_present); | |
34dc7c2f BB |
514 | |
515 | /* | |
516 | * Get the alignment requirement. | |
517 | */ | |
518 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift); | |
519 | ||
428870ff BB |
520 | /* |
521 | * Retrieve the vdev creation time. | |
522 | */ | |
523 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, | |
524 | &vd->vdev_crtxg); | |
525 | ||
34dc7c2f BB |
526 | /* |
527 | * If we're a top-level vdev, try to load the allocation parameters. | |
528 | */ | |
428870ff BB |
529 | if (parent && !parent->vdev_parent && |
530 | (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { | |
34dc7c2f BB |
531 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, |
532 | &vd->vdev_ms_array); | |
533 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, | |
534 | &vd->vdev_ms_shift); | |
535 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE, | |
536 | &vd->vdev_asize); | |
428870ff BB |
537 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING, |
538 | &vd->vdev_removing); | |
e0ab3ab5 JS |
539 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP, |
540 | &vd->vdev_top_zap); | |
541 | } else { | |
542 | ASSERT0(vd->vdev_top_zap); | |
428870ff BB |
543 | } |
544 | ||
5ffb9d1d | 545 | if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) { |
428870ff BB |
546 | ASSERT(alloctype == VDEV_ALLOC_LOAD || |
547 | alloctype == VDEV_ALLOC_ADD || | |
548 | alloctype == VDEV_ALLOC_SPLIT || | |
549 | alloctype == VDEV_ALLOC_ROOTPOOL); | |
550 | vd->vdev_mg = metaslab_group_create(islog ? | |
551 | spa_log_class(spa) : spa_normal_class(spa), vd); | |
34dc7c2f BB |
552 | } |
553 | ||
e0ab3ab5 JS |
554 | if (vd->vdev_ops->vdev_op_leaf && |
555 | (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { | |
556 | (void) nvlist_lookup_uint64(nv, | |
557 | ZPOOL_CONFIG_VDEV_LEAF_ZAP, &vd->vdev_leaf_zap); | |
558 | } else { | |
559 | ASSERT0(vd->vdev_leaf_zap); | |
560 | } | |
561 | ||
34dc7c2f BB |
562 | /* |
563 | * If we're a leaf vdev, try to load the DTL object and other state. | |
564 | */ | |
e0ab3ab5 | 565 | |
b128c09f | 566 | if (vd->vdev_ops->vdev_op_leaf && |
9babb374 BB |
567 | (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE || |
568 | alloctype == VDEV_ALLOC_ROOTPOOL)) { | |
b128c09f BB |
569 | if (alloctype == VDEV_ALLOC_LOAD) { |
570 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL, | |
93cf2076 | 571 | &vd->vdev_dtl_object); |
b128c09f BB |
572 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE, |
573 | &vd->vdev_unspare); | |
574 | } | |
9babb374 BB |
575 | |
576 | if (alloctype == VDEV_ALLOC_ROOTPOOL) { | |
577 | uint64_t spare = 0; | |
578 | ||
579 | if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE, | |
580 | &spare) == 0 && spare) | |
581 | spa_spare_add(vd); | |
582 | } | |
583 | ||
34dc7c2f BB |
584 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, |
585 | &vd->vdev_offline); | |
b128c09f | 586 | |
5d1f7fb6 GW |
587 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG, |
588 | &vd->vdev_resilver_txg); | |
572e2857 | 589 | |
34dc7c2f BB |
590 | /* |
591 | * When importing a pool, we want to ignore the persistent fault | |
592 | * state, as the diagnosis made on another system may not be | |
428870ff BB |
593 | * valid in the current context. Local vdevs will |
594 | * remain in the faulted state. | |
34dc7c2f | 595 | */ |
428870ff | 596 | if (spa_load_state(spa) == SPA_LOAD_OPEN) { |
34dc7c2f BB |
597 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, |
598 | &vd->vdev_faulted); | |
599 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED, | |
600 | &vd->vdev_degraded); | |
601 | (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, | |
602 | &vd->vdev_removed); | |
428870ff BB |
603 | |
604 | if (vd->vdev_faulted || vd->vdev_degraded) { | |
605 | char *aux; | |
606 | ||
607 | vd->vdev_label_aux = | |
608 | VDEV_AUX_ERR_EXCEEDED; | |
609 | if (nvlist_lookup_string(nv, | |
610 | ZPOOL_CONFIG_AUX_STATE, &aux) == 0 && | |
611 | strcmp(aux, "external") == 0) | |
612 | vd->vdev_label_aux = VDEV_AUX_EXTERNAL; | |
613 | } | |
34dc7c2f BB |
614 | } |
615 | } | |
616 | ||
617 | /* | |
618 | * Add ourselves to the parent's list of children. | |
619 | */ | |
620 | vdev_add_child(parent, vd); | |
621 | ||
622 | *vdp = vd; | |
623 | ||
624 | return (0); | |
625 | } | |
626 | ||
627 | void | |
628 | vdev_free(vdev_t *vd) | |
629 | { | |
d6320ddb | 630 | int c, t; |
34dc7c2f BB |
631 | spa_t *spa = vd->vdev_spa; |
632 | ||
633 | /* | |
634 | * vdev_free() implies closing the vdev first. This is simpler than | |
635 | * trying to ensure complicated semantics for all callers. | |
636 | */ | |
637 | vdev_close(vd); | |
638 | ||
b128c09f | 639 | ASSERT(!list_link_active(&vd->vdev_config_dirty_node)); |
428870ff | 640 | ASSERT(!list_link_active(&vd->vdev_state_dirty_node)); |
34dc7c2f BB |
641 | |
642 | /* | |
643 | * Free all children. | |
644 | */ | |
d6320ddb | 645 | for (c = 0; c < vd->vdev_children; c++) |
34dc7c2f BB |
646 | vdev_free(vd->vdev_child[c]); |
647 | ||
648 | ASSERT(vd->vdev_child == NULL); | |
649 | ASSERT(vd->vdev_guid_sum == vd->vdev_guid); | |
650 | ||
651 | /* | |
652 | * Discard allocation state. | |
653 | */ | |
428870ff | 654 | if (vd->vdev_mg != NULL) { |
34dc7c2f | 655 | vdev_metaslab_fini(vd); |
428870ff BB |
656 | metaslab_group_destroy(vd->vdev_mg); |
657 | } | |
34dc7c2f | 658 | |
c99c9001 MS |
659 | ASSERT0(vd->vdev_stat.vs_space); |
660 | ASSERT0(vd->vdev_stat.vs_dspace); | |
661 | ASSERT0(vd->vdev_stat.vs_alloc); | |
34dc7c2f BB |
662 | |
663 | /* | |
664 | * Remove this vdev from its parent's child list. | |
665 | */ | |
666 | vdev_remove_child(vd->vdev_parent, vd); | |
667 | ||
668 | ASSERT(vd->vdev_parent == NULL); | |
669 | ||
670 | /* | |
671 | * Clean up vdev structure. | |
672 | */ | |
673 | vdev_queue_fini(vd); | |
674 | vdev_cache_fini(vd); | |
675 | ||
676 | if (vd->vdev_path) | |
677 | spa_strfree(vd->vdev_path); | |
678 | if (vd->vdev_devid) | |
679 | spa_strfree(vd->vdev_devid); | |
680 | if (vd->vdev_physpath) | |
681 | spa_strfree(vd->vdev_physpath); | |
1bbd8770 TH |
682 | |
683 | if (vd->vdev_enc_sysfs_path) | |
684 | spa_strfree(vd->vdev_enc_sysfs_path); | |
685 | ||
9babb374 BB |
686 | if (vd->vdev_fru) |
687 | spa_strfree(vd->vdev_fru); | |
34dc7c2f BB |
688 | |
689 | if (vd->vdev_isspare) | |
690 | spa_spare_remove(vd); | |
691 | if (vd->vdev_isl2cache) | |
692 | spa_l2cache_remove(vd); | |
693 | ||
694 | txg_list_destroy(&vd->vdev_ms_list); | |
695 | txg_list_destroy(&vd->vdev_dtl_list); | |
fb5f0bc8 | 696 | |
34dc7c2f | 697 | mutex_enter(&vd->vdev_dtl_lock); |
93cf2076 | 698 | space_map_close(vd->vdev_dtl_sm); |
d6320ddb | 699 | for (t = 0; t < DTL_TYPES; t++) { |
93cf2076 GW |
700 | range_tree_vacate(vd->vdev_dtl[t], NULL, NULL); |
701 | range_tree_destroy(vd->vdev_dtl[t]); | |
fb5f0bc8 | 702 | } |
34dc7c2f | 703 | mutex_exit(&vd->vdev_dtl_lock); |
fb5f0bc8 | 704 | |
3dfb57a3 | 705 | mutex_destroy(&vd->vdev_queue_lock); |
34dc7c2f BB |
706 | mutex_destroy(&vd->vdev_dtl_lock); |
707 | mutex_destroy(&vd->vdev_stat_lock); | |
b128c09f | 708 | mutex_destroy(&vd->vdev_probe_lock); |
34dc7c2f BB |
709 | |
710 | if (vd == spa->spa_root_vdev) | |
711 | spa->spa_root_vdev = NULL; | |
712 | ||
713 | kmem_free(vd, sizeof (vdev_t)); | |
714 | } | |
715 | ||
716 | /* | |
717 | * Transfer top-level vdev state from svd to tvd. | |
718 | */ | |
719 | static void | |
720 | vdev_top_transfer(vdev_t *svd, vdev_t *tvd) | |
721 | { | |
722 | spa_t *spa = svd->vdev_spa; | |
723 | metaslab_t *msp; | |
724 | vdev_t *vd; | |
725 | int t; | |
726 | ||
727 | ASSERT(tvd == tvd->vdev_top); | |
728 | ||
77943bc1 | 729 | tvd->vdev_pending_fastwrite = svd->vdev_pending_fastwrite; |
34dc7c2f BB |
730 | tvd->vdev_ms_array = svd->vdev_ms_array; |
731 | tvd->vdev_ms_shift = svd->vdev_ms_shift; | |
732 | tvd->vdev_ms_count = svd->vdev_ms_count; | |
e0ab3ab5 | 733 | tvd->vdev_top_zap = svd->vdev_top_zap; |
34dc7c2f BB |
734 | |
735 | svd->vdev_ms_array = 0; | |
736 | svd->vdev_ms_shift = 0; | |
737 | svd->vdev_ms_count = 0; | |
e0ab3ab5 | 738 | svd->vdev_top_zap = 0; |
34dc7c2f | 739 | |
5ffb9d1d GW |
740 | if (tvd->vdev_mg) |
741 | ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg); | |
34dc7c2f BB |
742 | tvd->vdev_mg = svd->vdev_mg; |
743 | tvd->vdev_ms = svd->vdev_ms; | |
744 | ||
745 | svd->vdev_mg = NULL; | |
746 | svd->vdev_ms = NULL; | |
747 | ||
748 | if (tvd->vdev_mg != NULL) | |
749 | tvd->vdev_mg->mg_vd = tvd; | |
750 | ||
751 | tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc; | |
752 | tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space; | |
753 | tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace; | |
754 | ||
755 | svd->vdev_stat.vs_alloc = 0; | |
756 | svd->vdev_stat.vs_space = 0; | |
757 | svd->vdev_stat.vs_dspace = 0; | |
758 | ||
759 | for (t = 0; t < TXG_SIZE; t++) { | |
760 | while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL) | |
761 | (void) txg_list_add(&tvd->vdev_ms_list, msp, t); | |
762 | while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL) | |
763 | (void) txg_list_add(&tvd->vdev_dtl_list, vd, t); | |
764 | if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t)) | |
765 | (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t); | |
766 | } | |
767 | ||
b128c09f | 768 | if (list_link_active(&svd->vdev_config_dirty_node)) { |
34dc7c2f BB |
769 | vdev_config_clean(svd); |
770 | vdev_config_dirty(tvd); | |
771 | } | |
772 | ||
b128c09f BB |
773 | if (list_link_active(&svd->vdev_state_dirty_node)) { |
774 | vdev_state_clean(svd); | |
775 | vdev_state_dirty(tvd); | |
776 | } | |
777 | ||
34dc7c2f BB |
778 | tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio; |
779 | svd->vdev_deflate_ratio = 0; | |
780 | ||
781 | tvd->vdev_islog = svd->vdev_islog; | |
782 | svd->vdev_islog = 0; | |
783 | } | |
784 | ||
785 | static void | |
786 | vdev_top_update(vdev_t *tvd, vdev_t *vd) | |
787 | { | |
d6320ddb BB |
788 | int c; |
789 | ||
34dc7c2f BB |
790 | if (vd == NULL) |
791 | return; | |
792 | ||
793 | vd->vdev_top = tvd; | |
794 | ||
d6320ddb | 795 | for (c = 0; c < vd->vdev_children; c++) |
34dc7c2f BB |
796 | vdev_top_update(tvd, vd->vdev_child[c]); |
797 | } | |
798 | ||
799 | /* | |
800 | * Add a mirror/replacing vdev above an existing vdev. | |
801 | */ | |
802 | vdev_t * | |
803 | vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops) | |
804 | { | |
805 | spa_t *spa = cvd->vdev_spa; | |
806 | vdev_t *pvd = cvd->vdev_parent; | |
807 | vdev_t *mvd; | |
808 | ||
b128c09f | 809 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
34dc7c2f BB |
810 | |
811 | mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops); | |
812 | ||
813 | mvd->vdev_asize = cvd->vdev_asize; | |
9babb374 | 814 | mvd->vdev_min_asize = cvd->vdev_min_asize; |
1bd201e7 | 815 | mvd->vdev_max_asize = cvd->vdev_max_asize; |
34dc7c2f BB |
816 | mvd->vdev_ashift = cvd->vdev_ashift; |
817 | mvd->vdev_state = cvd->vdev_state; | |
428870ff | 818 | mvd->vdev_crtxg = cvd->vdev_crtxg; |
34dc7c2f BB |
819 | |
820 | vdev_remove_child(pvd, cvd); | |
821 | vdev_add_child(pvd, mvd); | |
822 | cvd->vdev_id = mvd->vdev_children; | |
823 | vdev_add_child(mvd, cvd); | |
824 | vdev_top_update(cvd->vdev_top, cvd->vdev_top); | |
825 | ||
826 | if (mvd == mvd->vdev_top) | |
827 | vdev_top_transfer(cvd, mvd); | |
828 | ||
829 | return (mvd); | |
830 | } | |
831 | ||
832 | /* | |
833 | * Remove a 1-way mirror/replacing vdev from the tree. | |
834 | */ | |
835 | void | |
836 | vdev_remove_parent(vdev_t *cvd) | |
837 | { | |
838 | vdev_t *mvd = cvd->vdev_parent; | |
839 | vdev_t *pvd = mvd->vdev_parent; | |
840 | ||
b128c09f | 841 | ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); |
34dc7c2f BB |
842 | |
843 | ASSERT(mvd->vdev_children == 1); | |
844 | ASSERT(mvd->vdev_ops == &vdev_mirror_ops || | |
845 | mvd->vdev_ops == &vdev_replacing_ops || | |
846 | mvd->vdev_ops == &vdev_spare_ops); | |
847 | cvd->vdev_ashift = mvd->vdev_ashift; | |
848 | ||
849 | vdev_remove_child(mvd, cvd); | |
850 | vdev_remove_child(pvd, mvd); | |
fb5f0bc8 | 851 | |
34dc7c2f | 852 | /* |
b128c09f BB |
853 | * If cvd will replace mvd as a top-level vdev, preserve mvd's guid. |
854 | * Otherwise, we could have detached an offline device, and when we | |
855 | * go to import the pool we'll think we have two top-level vdevs, | |
856 | * instead of a different version of the same top-level vdev. | |
34dc7c2f | 857 | */ |
fb5f0bc8 BB |
858 | if (mvd->vdev_top == mvd) { |
859 | uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid; | |
428870ff | 860 | cvd->vdev_orig_guid = cvd->vdev_guid; |
fb5f0bc8 BB |
861 | cvd->vdev_guid += guid_delta; |
862 | cvd->vdev_guid_sum += guid_delta; | |
61e99a73 AB |
863 | |
864 | /* | |
865 | * If pool not set for autoexpand, we need to also preserve | |
866 | * mvd's asize to prevent automatic expansion of cvd. | |
867 | * Otherwise if we are adjusting the mirror by attaching and | |
868 | * detaching children of non-uniform sizes, the mirror could | |
869 | * autoexpand, unexpectedly requiring larger devices to | |
870 | * re-establish the mirror. | |
871 | */ | |
872 | if (!cvd->vdev_spa->spa_autoexpand) | |
873 | cvd->vdev_asize = mvd->vdev_asize; | |
fb5f0bc8 | 874 | } |
b128c09f BB |
875 | cvd->vdev_id = mvd->vdev_id; |
876 | vdev_add_child(pvd, cvd); | |
34dc7c2f BB |
877 | vdev_top_update(cvd->vdev_top, cvd->vdev_top); |
878 | ||
879 | if (cvd == cvd->vdev_top) | |
880 | vdev_top_transfer(mvd, cvd); | |
881 | ||
882 | ASSERT(mvd->vdev_children == 0); | |
883 | vdev_free(mvd); | |
884 | } | |
885 | ||
886 | int | |
887 | vdev_metaslab_init(vdev_t *vd, uint64_t txg) | |
888 | { | |
889 | spa_t *spa = vd->vdev_spa; | |
890 | objset_t *mos = spa->spa_meta_objset; | |
34dc7c2f BB |
891 | uint64_t m; |
892 | uint64_t oldc = vd->vdev_ms_count; | |
893 | uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift; | |
894 | metaslab_t **mspp; | |
895 | int error; | |
896 | ||
428870ff BB |
897 | ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER)); |
898 | ||
899 | /* | |
900 | * This vdev is not being allocated from yet or is a hole. | |
901 | */ | |
902 | if (vd->vdev_ms_shift == 0) | |
34dc7c2f BB |
903 | return (0); |
904 | ||
428870ff BB |
905 | ASSERT(!vd->vdev_ishole); |
906 | ||
9babb374 BB |
907 | /* |
908 | * Compute the raidz-deflation ratio. Note, we hard-code | |
f1512ee6 MA |
909 | * in 128k (1 << 17) because it is the "typical" blocksize. |
910 | * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change, | |
911 | * otherwise it would inconsistently account for existing bp's. | |
9babb374 BB |
912 | */ |
913 | vd->vdev_deflate_ratio = (1 << 17) / | |
914 | (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT); | |
915 | ||
34dc7c2f BB |
916 | ASSERT(oldc <= newc); |
917 | ||
bffb68a2 | 918 | mspp = vmem_zalloc(newc * sizeof (*mspp), KM_SLEEP); |
34dc7c2f BB |
919 | |
920 | if (oldc != 0) { | |
921 | bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp)); | |
bffb68a2 | 922 | vmem_free(vd->vdev_ms, oldc * sizeof (*mspp)); |
34dc7c2f BB |
923 | } |
924 | ||
925 | vd->vdev_ms = mspp; | |
926 | vd->vdev_ms_count = newc; | |
927 | ||
928 | for (m = oldc; m < newc; m++) { | |
93cf2076 GW |
929 | uint64_t object = 0; |
930 | ||
34dc7c2f | 931 | if (txg == 0) { |
34dc7c2f | 932 | error = dmu_read(mos, vd->vdev_ms_array, |
9babb374 BB |
933 | m * sizeof (uint64_t), sizeof (uint64_t), &object, |
934 | DMU_READ_PREFETCH); | |
34dc7c2f BB |
935 | if (error) |
936 | return (error); | |
34dc7c2f | 937 | } |
fb42a493 PS |
938 | |
939 | error = metaslab_init(vd->vdev_mg, m, object, txg, | |
940 | &(vd->vdev_ms[m])); | |
941 | if (error) | |
942 | return (error); | |
34dc7c2f BB |
943 | } |
944 | ||
428870ff BB |
945 | if (txg == 0) |
946 | spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER); | |
947 | ||
948 | /* | |
949 | * If the vdev is being removed we don't activate | |
950 | * the metaslabs since we want to ensure that no new | |
951 | * allocations are performed on this device. | |
952 | */ | |
953 | if (oldc == 0 && !vd->vdev_removing) | |
954 | metaslab_group_activate(vd->vdev_mg); | |
955 | ||
956 | if (txg == 0) | |
957 | spa_config_exit(spa, SCL_ALLOC, FTAG); | |
958 | ||
34dc7c2f BB |
959 | return (0); |
960 | } | |
961 | ||
962 | void | |
963 | vdev_metaslab_fini(vdev_t *vd) | |
964 | { | |
965 | uint64_t m; | |
966 | uint64_t count = vd->vdev_ms_count; | |
967 | ||
968 | if (vd->vdev_ms != NULL) { | |
428870ff | 969 | metaslab_group_passivate(vd->vdev_mg); |
93cf2076 GW |
970 | for (m = 0; m < count; m++) { |
971 | metaslab_t *msp = vd->vdev_ms[m]; | |
972 | ||
973 | if (msp != NULL) | |
974 | metaslab_fini(msp); | |
975 | } | |
bffb68a2 | 976 | vmem_free(vd->vdev_ms, count * sizeof (metaslab_t *)); |
34dc7c2f BB |
977 | vd->vdev_ms = NULL; |
978 | } | |
920dd524 ED |
979 | |
980 | ASSERT3U(vd->vdev_pending_fastwrite, ==, 0); | |
34dc7c2f BB |
981 | } |
982 | ||
b128c09f BB |
983 | typedef struct vdev_probe_stats { |
984 | boolean_t vps_readable; | |
985 | boolean_t vps_writeable; | |
986 | int vps_flags; | |
b128c09f BB |
987 | } vdev_probe_stats_t; |
988 | ||
989 | static void | |
990 | vdev_probe_done(zio_t *zio) | |
34dc7c2f | 991 | { |
fb5f0bc8 | 992 | spa_t *spa = zio->io_spa; |
d164b209 | 993 | vdev_t *vd = zio->io_vd; |
b128c09f | 994 | vdev_probe_stats_t *vps = zio->io_private; |
d164b209 BB |
995 | |
996 | ASSERT(vd->vdev_probe_zio != NULL); | |
b128c09f BB |
997 | |
998 | if (zio->io_type == ZIO_TYPE_READ) { | |
b128c09f BB |
999 | if (zio->io_error == 0) |
1000 | vps->vps_readable = 1; | |
fb5f0bc8 | 1001 | if (zio->io_error == 0 && spa_writeable(spa)) { |
d164b209 | 1002 | zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd, |
a6255b7f | 1003 | zio->io_offset, zio->io_size, zio->io_abd, |
b128c09f BB |
1004 | ZIO_CHECKSUM_OFF, vdev_probe_done, vps, |
1005 | ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE)); | |
1006 | } else { | |
a6255b7f | 1007 | abd_free(zio->io_abd); |
b128c09f BB |
1008 | } |
1009 | } else if (zio->io_type == ZIO_TYPE_WRITE) { | |
b128c09f BB |
1010 | if (zio->io_error == 0) |
1011 | vps->vps_writeable = 1; | |
a6255b7f | 1012 | abd_free(zio->io_abd); |
b128c09f | 1013 | } else if (zio->io_type == ZIO_TYPE_NULL) { |
d164b209 | 1014 | zio_t *pio; |
3dfb57a3 | 1015 | zio_link_t *zl; |
b128c09f BB |
1016 | |
1017 | vd->vdev_cant_read |= !vps->vps_readable; | |
1018 | vd->vdev_cant_write |= !vps->vps_writeable; | |
1019 | ||
1020 | if (vdev_readable(vd) && | |
fb5f0bc8 | 1021 | (vdev_writeable(vd) || !spa_writeable(spa))) { |
b128c09f BB |
1022 | zio->io_error = 0; |
1023 | } else { | |
1024 | ASSERT(zio->io_error != 0); | |
1025 | zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE, | |
fb5f0bc8 | 1026 | spa, vd, NULL, 0, 0); |
2e528b49 | 1027 | zio->io_error = SET_ERROR(ENXIO); |
b128c09f | 1028 | } |
d164b209 BB |
1029 | |
1030 | mutex_enter(&vd->vdev_probe_lock); | |
1031 | ASSERT(vd->vdev_probe_zio == zio); | |
1032 | vd->vdev_probe_zio = NULL; | |
1033 | mutex_exit(&vd->vdev_probe_lock); | |
1034 | ||
3dfb57a3 DB |
1035 | zl = NULL; |
1036 | while ((pio = zio_walk_parents(zio, &zl)) != NULL) | |
d164b209 | 1037 | if (!vdev_accessible(vd, pio)) |
2e528b49 | 1038 | pio->io_error = SET_ERROR(ENXIO); |
d164b209 | 1039 | |
b128c09f BB |
1040 | kmem_free(vps, sizeof (*vps)); |
1041 | } | |
1042 | } | |
34dc7c2f | 1043 | |
b128c09f | 1044 | /* |
d3cc8b15 WA |
1045 | * Determine whether this device is accessible. |
1046 | * | |
1047 | * Read and write to several known locations: the pad regions of each | |
1048 | * vdev label but the first, which we leave alone in case it contains | |
1049 | * a VTOC. | |
b128c09f BB |
1050 | */ |
1051 | zio_t * | |
d164b209 | 1052 | vdev_probe(vdev_t *vd, zio_t *zio) |
b128c09f BB |
1053 | { |
1054 | spa_t *spa = vd->vdev_spa; | |
d164b209 BB |
1055 | vdev_probe_stats_t *vps = NULL; |
1056 | zio_t *pio; | |
d6320ddb | 1057 | int l; |
d164b209 BB |
1058 | |
1059 | ASSERT(vd->vdev_ops->vdev_op_leaf); | |
34dc7c2f | 1060 | |
d164b209 BB |
1061 | /* |
1062 | * Don't probe the probe. | |
1063 | */ | |
1064 | if (zio && (zio->io_flags & ZIO_FLAG_PROBE)) | |
1065 | return (NULL); | |
b128c09f | 1066 | |
d164b209 BB |
1067 | /* |
1068 | * To prevent 'probe storms' when a device fails, we create | |
1069 | * just one probe i/o at a time. All zios that want to probe | |
1070 | * this vdev will become parents of the probe io. | |
1071 | */ | |
1072 | mutex_enter(&vd->vdev_probe_lock); | |
b128c09f | 1073 | |
d164b209 | 1074 | if ((pio = vd->vdev_probe_zio) == NULL) { |
79c76d5b | 1075 | vps = kmem_zalloc(sizeof (*vps), KM_SLEEP); |
d164b209 BB |
1076 | |
1077 | vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE | | |
1078 | ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE | | |
9babb374 | 1079 | ZIO_FLAG_TRYHARD; |
d164b209 BB |
1080 | |
1081 | if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) { | |
1082 | /* | |
1083 | * vdev_cant_read and vdev_cant_write can only | |
1084 | * transition from TRUE to FALSE when we have the | |
1085 | * SCL_ZIO lock as writer; otherwise they can only | |
1086 | * transition from FALSE to TRUE. This ensures that | |
1087 | * any zio looking at these values can assume that | |
1088 | * failures persist for the life of the I/O. That's | |
1089 | * important because when a device has intermittent | |
1090 | * connectivity problems, we want to ensure that | |
1091 | * they're ascribed to the device (ENXIO) and not | |
1092 | * the zio (EIO). | |
1093 | * | |
1094 | * Since we hold SCL_ZIO as writer here, clear both | |
1095 | * values so the probe can reevaluate from first | |
1096 | * principles. | |
1097 | */ | |
1098 | vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER; | |
1099 | vd->vdev_cant_read = B_FALSE; | |
1100 | vd->vdev_cant_write = B_FALSE; | |
1101 | } | |
1102 | ||
1103 | vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd, | |
1104 | vdev_probe_done, vps, | |
1105 | vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE); | |
1106 | ||
428870ff BB |
1107 | /* |
1108 | * We can't change the vdev state in this context, so we | |
1109 | * kick off an async task to do it on our behalf. | |
1110 | */ | |
d164b209 BB |
1111 | if (zio != NULL) { |
1112 | vd->vdev_probe_wanted = B_TRUE; | |
1113 | spa_async_request(spa, SPA_ASYNC_PROBE); | |
1114 | } | |
b128c09f BB |
1115 | } |
1116 | ||
d164b209 BB |
1117 | if (zio != NULL) |
1118 | zio_add_child(zio, pio); | |
b128c09f | 1119 | |
d164b209 | 1120 | mutex_exit(&vd->vdev_probe_lock); |
b128c09f | 1121 | |
d164b209 BB |
1122 | if (vps == NULL) { |
1123 | ASSERT(zio != NULL); | |
1124 | return (NULL); | |
1125 | } | |
b128c09f | 1126 | |
d6320ddb | 1127 | for (l = 1; l < VDEV_LABELS; l++) { |
d164b209 | 1128 | zio_nowait(zio_read_phys(pio, vd, |
b128c09f | 1129 | vdev_label_offset(vd->vdev_psize, l, |
a6255b7f DQ |
1130 | offsetof(vdev_label_t, vl_pad2)), VDEV_PAD_SIZE, |
1131 | abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE), | |
b128c09f BB |
1132 | ZIO_CHECKSUM_OFF, vdev_probe_done, vps, |
1133 | ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE)); | |
1134 | } | |
1135 | ||
d164b209 BB |
1136 | if (zio == NULL) |
1137 | return (pio); | |
1138 | ||
1139 | zio_nowait(pio); | |
1140 | return (NULL); | |
34dc7c2f BB |
1141 | } |
1142 | ||
45d1cae3 BB |
1143 | static void |
1144 | vdev_open_child(void *arg) | |
1145 | { | |
1146 | vdev_t *vd = arg; | |
1147 | ||
1148 | vd->vdev_open_thread = curthread; | |
1149 | vd->vdev_open_error = vdev_open(vd); | |
1150 | vd->vdev_open_thread = NULL; | |
1151 | } | |
1152 | ||
6c285672 | 1153 | static boolean_t |
428870ff BB |
1154 | vdev_uses_zvols(vdev_t *vd) |
1155 | { | |
d6320ddb BB |
1156 | int c; |
1157 | ||
6c285672 JL |
1158 | #ifdef _KERNEL |
1159 | if (zvol_is_zvol(vd->vdev_path)) | |
428870ff | 1160 | return (B_TRUE); |
6c285672 JL |
1161 | #endif |
1162 | ||
d6320ddb | 1163 | for (c = 0; c < vd->vdev_children; c++) |
428870ff BB |
1164 | if (vdev_uses_zvols(vd->vdev_child[c])) |
1165 | return (B_TRUE); | |
6c285672 | 1166 | |
428870ff BB |
1167 | return (B_FALSE); |
1168 | } | |
1169 | ||
45d1cae3 BB |
1170 | void |
1171 | vdev_open_children(vdev_t *vd) | |
1172 | { | |
1173 | taskq_t *tq; | |
1174 | int children = vd->vdev_children; | |
d6320ddb | 1175 | int c; |
45d1cae3 | 1176 | |
428870ff BB |
1177 | /* |
1178 | * in order to handle pools on top of zvols, do the opens | |
1179 | * in a single thread so that the same thread holds the | |
1180 | * spa_namespace_lock | |
1181 | */ | |
1182 | if (vdev_uses_zvols(vd)) { | |
13d9a004 | 1183 | retry_sync: |
4770aa06 | 1184 | for (c = 0; c < children; c++) |
428870ff BB |
1185 | vd->vdev_child[c]->vdev_open_error = |
1186 | vdev_open(vd->vdev_child[c]); | |
4770aa06 HJ |
1187 | } else { |
1188 | tq = taskq_create("vdev_open", children, minclsyspri, | |
1189 | children, children, TASKQ_PREPOPULATE); | |
13d9a004 BB |
1190 | if (tq == NULL) |
1191 | goto retry_sync; | |
45d1cae3 | 1192 | |
4770aa06 HJ |
1193 | for (c = 0; c < children; c++) |
1194 | VERIFY(taskq_dispatch(tq, vdev_open_child, | |
48d3eb40 | 1195 | vd->vdev_child[c], TQ_SLEEP) != TASKQID_INVALID); |
45d1cae3 | 1196 | |
4770aa06 HJ |
1197 | taskq_destroy(tq); |
1198 | } | |
1199 | ||
1200 | vd->vdev_nonrot = B_TRUE; | |
fb40095f RY |
1201 | |
1202 | for (c = 0; c < children; c++) | |
1203 | vd->vdev_nonrot &= vd->vdev_child[c]->vdev_nonrot; | |
45d1cae3 BB |
1204 | } |
1205 | ||
34dc7c2f BB |
1206 | /* |
1207 | * Prepare a virtual device for access. | |
1208 | */ | |
1209 | int | |
1210 | vdev_open(vdev_t *vd) | |
1211 | { | |
fb5f0bc8 | 1212 | spa_t *spa = vd->vdev_spa; |
34dc7c2f | 1213 | int error; |
34dc7c2f | 1214 | uint64_t osize = 0; |
1bd201e7 CS |
1215 | uint64_t max_osize = 0; |
1216 | uint64_t asize, max_asize, psize; | |
34dc7c2f | 1217 | uint64_t ashift = 0; |
d6320ddb | 1218 | int c; |
34dc7c2f | 1219 | |
45d1cae3 BB |
1220 | ASSERT(vd->vdev_open_thread == curthread || |
1221 | spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
34dc7c2f BB |
1222 | ASSERT(vd->vdev_state == VDEV_STATE_CLOSED || |
1223 | vd->vdev_state == VDEV_STATE_CANT_OPEN || | |
1224 | vd->vdev_state == VDEV_STATE_OFFLINE); | |
1225 | ||
34dc7c2f | 1226 | vd->vdev_stat.vs_aux = VDEV_AUX_NONE; |
9babb374 BB |
1227 | vd->vdev_cant_read = B_FALSE; |
1228 | vd->vdev_cant_write = B_FALSE; | |
1229 | vd->vdev_min_asize = vdev_get_min_asize(vd); | |
34dc7c2f | 1230 | |
428870ff BB |
1231 | /* |
1232 | * If this vdev is not removed, check its fault status. If it's | |
1233 | * faulted, bail out of the open. | |
1234 | */ | |
34dc7c2f BB |
1235 | if (!vd->vdev_removed && vd->vdev_faulted) { |
1236 | ASSERT(vd->vdev_children == 0); | |
428870ff BB |
1237 | ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || |
1238 | vd->vdev_label_aux == VDEV_AUX_EXTERNAL); | |
34dc7c2f | 1239 | vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, |
428870ff | 1240 | vd->vdev_label_aux); |
2e528b49 | 1241 | return (SET_ERROR(ENXIO)); |
34dc7c2f BB |
1242 | } else if (vd->vdev_offline) { |
1243 | ASSERT(vd->vdev_children == 0); | |
1244 | vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE); | |
2e528b49 | 1245 | return (SET_ERROR(ENXIO)); |
34dc7c2f BB |
1246 | } |
1247 | ||
1bd201e7 | 1248 | error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize, &ashift); |
34dc7c2f | 1249 | |
428870ff BB |
1250 | /* |
1251 | * Reset the vdev_reopening flag so that we actually close | |
1252 | * the vdev on error. | |
1253 | */ | |
1254 | vd->vdev_reopening = B_FALSE; | |
34dc7c2f | 1255 | if (zio_injection_enabled && error == 0) |
9babb374 | 1256 | error = zio_handle_device_injection(vd, NULL, ENXIO); |
34dc7c2f BB |
1257 | |
1258 | if (error) { | |
1259 | if (vd->vdev_removed && | |
1260 | vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED) | |
1261 | vd->vdev_removed = B_FALSE; | |
1262 | ||
1263 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1264 | vd->vdev_stat.vs_aux); | |
1265 | return (error); | |
1266 | } | |
1267 | ||
1268 | vd->vdev_removed = B_FALSE; | |
1269 | ||
428870ff BB |
1270 | /* |
1271 | * Recheck the faulted flag now that we have confirmed that | |
1272 | * the vdev is accessible. If we're faulted, bail. | |
1273 | */ | |
1274 | if (vd->vdev_faulted) { | |
1275 | ASSERT(vd->vdev_children == 0); | |
1276 | ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || | |
1277 | vd->vdev_label_aux == VDEV_AUX_EXTERNAL); | |
1278 | vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, | |
1279 | vd->vdev_label_aux); | |
2e528b49 | 1280 | return (SET_ERROR(ENXIO)); |
428870ff BB |
1281 | } |
1282 | ||
34dc7c2f BB |
1283 | if (vd->vdev_degraded) { |
1284 | ASSERT(vd->vdev_children == 0); | |
1285 | vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, | |
1286 | VDEV_AUX_ERR_EXCEEDED); | |
1287 | } else { | |
428870ff | 1288 | vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0); |
34dc7c2f BB |
1289 | } |
1290 | ||
428870ff BB |
1291 | /* |
1292 | * For hole or missing vdevs we just return success. | |
1293 | */ | |
1294 | if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) | |
1295 | return (0); | |
1296 | ||
d6320ddb | 1297 | for (c = 0; c < vd->vdev_children; c++) { |
34dc7c2f BB |
1298 | if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) { |
1299 | vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, | |
1300 | VDEV_AUX_NONE); | |
1301 | break; | |
1302 | } | |
9babb374 | 1303 | } |
34dc7c2f BB |
1304 | |
1305 | osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t)); | |
1bd201e7 | 1306 | max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t)); |
34dc7c2f BB |
1307 | |
1308 | if (vd->vdev_children == 0) { | |
1309 | if (osize < SPA_MINDEVSIZE) { | |
1310 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1311 | VDEV_AUX_TOO_SMALL); | |
2e528b49 | 1312 | return (SET_ERROR(EOVERFLOW)); |
34dc7c2f BB |
1313 | } |
1314 | psize = osize; | |
1315 | asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE); | |
1bd201e7 CS |
1316 | max_asize = max_osize - (VDEV_LABEL_START_SIZE + |
1317 | VDEV_LABEL_END_SIZE); | |
34dc7c2f BB |
1318 | } else { |
1319 | if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE - | |
1320 | (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) { | |
1321 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1322 | VDEV_AUX_TOO_SMALL); | |
2e528b49 | 1323 | return (SET_ERROR(EOVERFLOW)); |
34dc7c2f BB |
1324 | } |
1325 | psize = 0; | |
1326 | asize = osize; | |
1bd201e7 | 1327 | max_asize = max_osize; |
34dc7c2f BB |
1328 | } |
1329 | ||
1330 | vd->vdev_psize = psize; | |
1331 | ||
9babb374 BB |
1332 | /* |
1333 | * Make sure the allocatable size hasn't shrunk. | |
1334 | */ | |
1335 | if (asize < vd->vdev_min_asize) { | |
1336 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
1337 | VDEV_AUX_BAD_LABEL); | |
2e528b49 | 1338 | return (SET_ERROR(EINVAL)); |
9babb374 BB |
1339 | } |
1340 | ||
34dc7c2f BB |
1341 | if (vd->vdev_asize == 0) { |
1342 | /* | |
1343 | * This is the first-ever open, so use the computed values. | |
b28e57cb | 1344 | * For compatibility, a different ashift can be requested. |
34dc7c2f BB |
1345 | */ |
1346 | vd->vdev_asize = asize; | |
1bd201e7 | 1347 | vd->vdev_max_asize = max_asize; |
b28e57cb MM |
1348 | if (vd->vdev_ashift == 0) |
1349 | vd->vdev_ashift = ashift; | |
34dc7c2f BB |
1350 | } else { |
1351 | /* | |
32a9872b GW |
1352 | * Detect if the alignment requirement has increased. |
1353 | * We don't want to make the pool unavailable, just | |
1354 | * post an event instead. | |
34dc7c2f | 1355 | */ |
32a9872b GW |
1356 | if (ashift > vd->vdev_top->vdev_ashift && |
1357 | vd->vdev_ops->vdev_op_leaf) { | |
1358 | zfs_ereport_post(FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT, | |
1359 | spa, vd, NULL, 0, 0); | |
34dc7c2f | 1360 | } |
32a9872b | 1361 | |
1bd201e7 | 1362 | vd->vdev_max_asize = max_asize; |
9babb374 | 1363 | } |
34dc7c2f | 1364 | |
9babb374 BB |
1365 | /* |
1366 | * If all children are healthy and the asize has increased, | |
1367 | * then we've experienced dynamic LUN growth. If automatic | |
1368 | * expansion is enabled then use the additional space. | |
1369 | */ | |
1370 | if (vd->vdev_state == VDEV_STATE_HEALTHY && asize > vd->vdev_asize && | |
1371 | (vd->vdev_expanding || spa->spa_autoexpand)) | |
1372 | vd->vdev_asize = asize; | |
34dc7c2f | 1373 | |
9babb374 | 1374 | vdev_set_min_asize(vd); |
34dc7c2f BB |
1375 | |
1376 | /* | |
1377 | * Ensure we can issue some IO before declaring the | |
1378 | * vdev open for business. | |
1379 | */ | |
b128c09f BB |
1380 | if (vd->vdev_ops->vdev_op_leaf && |
1381 | (error = zio_wait(vdev_probe(vd, NULL))) != 0) { | |
428870ff BB |
1382 | vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, |
1383 | VDEV_AUX_ERR_EXCEEDED); | |
34dc7c2f BB |
1384 | return (error); |
1385 | } | |
1386 | ||
c3520e7f MA |
1387 | /* |
1388 | * Track the min and max ashift values for normal data devices. | |
1389 | */ | |
1390 | if (vd->vdev_top == vd && vd->vdev_ashift != 0 && | |
1391 | !vd->vdev_islog && vd->vdev_aux == NULL) { | |
1392 | if (vd->vdev_ashift > spa->spa_max_ashift) | |
1393 | spa->spa_max_ashift = vd->vdev_ashift; | |
1394 | if (vd->vdev_ashift < spa->spa_min_ashift) | |
1395 | spa->spa_min_ashift = vd->vdev_ashift; | |
1396 | } | |
1397 | ||
34dc7c2f | 1398 | /* |
b128c09f | 1399 | * If a leaf vdev has a DTL, and seems healthy, then kick off a |
fb5f0bc8 BB |
1400 | * resilver. But don't do this if we are doing a reopen for a scrub, |
1401 | * since this would just restart the scrub we are already doing. | |
34dc7c2f | 1402 | */ |
fb5f0bc8 BB |
1403 | if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen && |
1404 | vdev_resilver_needed(vd, NULL, NULL)) | |
1405 | spa_async_request(spa, SPA_ASYNC_RESILVER); | |
34dc7c2f BB |
1406 | |
1407 | return (0); | |
1408 | } | |
1409 | ||
1410 | /* | |
1411 | * Called once the vdevs are all opened, this routine validates the label | |
1412 | * contents. This needs to be done before vdev_load() so that we don't | |
1413 | * inadvertently do repair I/Os to the wrong device. | |
1414 | * | |
c7f2d69d GW |
1415 | * If 'strict' is false ignore the spa guid check. This is necessary because |
1416 | * if the machine crashed during a re-guid the new guid might have been written | |
1417 | * to all of the vdev labels, but not the cached config. The strict check | |
1418 | * will be performed when the pool is opened again using the mos config. | |
1419 | * | |
34dc7c2f BB |
1420 | * This function will only return failure if one of the vdevs indicates that it |
1421 | * has since been destroyed or exported. This is only possible if | |
1422 | * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state | |
1423 | * will be updated but the function will return 0. | |
1424 | */ | |
1425 | int | |
c7f2d69d | 1426 | vdev_validate(vdev_t *vd, boolean_t strict) |
34dc7c2f BB |
1427 | { |
1428 | spa_t *spa = vd->vdev_spa; | |
34dc7c2f | 1429 | nvlist_t *label; |
428870ff | 1430 | uint64_t guid = 0, top_guid; |
34dc7c2f | 1431 | uint64_t state; |
d6320ddb | 1432 | int c; |
34dc7c2f | 1433 | |
d6320ddb | 1434 | for (c = 0; c < vd->vdev_children; c++) |
c7f2d69d | 1435 | if (vdev_validate(vd->vdev_child[c], strict) != 0) |
2e528b49 | 1436 | return (SET_ERROR(EBADF)); |
34dc7c2f BB |
1437 | |
1438 | /* | |
1439 | * If the device has already failed, or was marked offline, don't do | |
1440 | * any further validation. Otherwise, label I/O will fail and we will | |
1441 | * overwrite the previous state. | |
1442 | */ | |
b128c09f | 1443 | if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) { |
428870ff BB |
1444 | uint64_t aux_guid = 0; |
1445 | nvlist_t *nvl; | |
295304be GW |
1446 | uint64_t txg = spa_last_synced_txg(spa) != 0 ? |
1447 | spa_last_synced_txg(spa) : -1ULL; | |
34dc7c2f | 1448 | |
3bc7e0fb | 1449 | if ((label = vdev_label_read_config(vd, txg)) == NULL) { |
e35c5a82 | 1450 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, |
34dc7c2f BB |
1451 | VDEV_AUX_BAD_LABEL); |
1452 | return (0); | |
1453 | } | |
1454 | ||
428870ff BB |
1455 | /* |
1456 | * Determine if this vdev has been split off into another | |
1457 | * pool. If so, then refuse to open it. | |
1458 | */ | |
1459 | if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID, | |
1460 | &aux_guid) == 0 && aux_guid == spa_guid(spa)) { | |
1461 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
1462 | VDEV_AUX_SPLIT_POOL); | |
1463 | nvlist_free(label); | |
1464 | return (0); | |
1465 | } | |
1466 | ||
c7f2d69d GW |
1467 | if (strict && (nvlist_lookup_uint64(label, |
1468 | ZPOOL_CONFIG_POOL_GUID, &guid) != 0 || | |
1469 | guid != spa_guid(spa))) { | |
34dc7c2f BB |
1470 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, |
1471 | VDEV_AUX_CORRUPT_DATA); | |
1472 | nvlist_free(label); | |
1473 | return (0); | |
1474 | } | |
1475 | ||
428870ff BB |
1476 | if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl) |
1477 | != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID, | |
1478 | &aux_guid) != 0) | |
1479 | aux_guid = 0; | |
1480 | ||
b128c09f BB |
1481 | /* |
1482 | * If this vdev just became a top-level vdev because its | |
1483 | * sibling was detached, it will have adopted the parent's | |
1484 | * vdev guid -- but the label may or may not be on disk yet. | |
1485 | * Fortunately, either version of the label will have the | |
1486 | * same top guid, so if we're a top-level vdev, we can | |
1487 | * safely compare to that instead. | |
428870ff BB |
1488 | * |
1489 | * If we split this vdev off instead, then we also check the | |
1490 | * original pool's guid. We don't want to consider the vdev | |
1491 | * corrupt if it is partway through a split operation. | |
b128c09f | 1492 | */ |
34dc7c2f | 1493 | if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, |
b128c09f BB |
1494 | &guid) != 0 || |
1495 | nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, | |
1496 | &top_guid) != 0 || | |
428870ff | 1497 | ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) && |
b128c09f | 1498 | (vd->vdev_guid != top_guid || vd != vd->vdev_top))) { |
34dc7c2f BB |
1499 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, |
1500 | VDEV_AUX_CORRUPT_DATA); | |
1501 | nvlist_free(label); | |
1502 | return (0); | |
1503 | } | |
1504 | ||
1505 | if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, | |
1506 | &state) != 0) { | |
1507 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
1508 | VDEV_AUX_CORRUPT_DATA); | |
1509 | nvlist_free(label); | |
1510 | return (0); | |
1511 | } | |
1512 | ||
1513 | nvlist_free(label); | |
1514 | ||
45d1cae3 | 1515 | /* |
572e2857 | 1516 | * If this is a verbatim import, no need to check the |
45d1cae3 BB |
1517 | * state of the pool. |
1518 | */ | |
572e2857 | 1519 | if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) && |
428870ff | 1520 | spa_load_state(spa) == SPA_LOAD_OPEN && |
34dc7c2f | 1521 | state != POOL_STATE_ACTIVE) |
2e528b49 | 1522 | return (SET_ERROR(EBADF)); |
34dc7c2f | 1523 | |
b128c09f BB |
1524 | /* |
1525 | * If we were able to open and validate a vdev that was | |
1526 | * previously marked permanently unavailable, clear that state | |
1527 | * now. | |
1528 | */ | |
1529 | if (vd->vdev_not_present) | |
1530 | vd->vdev_not_present = 0; | |
1531 | } | |
34dc7c2f BB |
1532 | |
1533 | return (0); | |
1534 | } | |
1535 | ||
1536 | /* | |
1537 | * Close a virtual device. | |
1538 | */ | |
1539 | void | |
1540 | vdev_close(vdev_t *vd) | |
1541 | { | |
428870ff | 1542 | vdev_t *pvd = vd->vdev_parent; |
1fde1e37 | 1543 | ASSERTV(spa_t *spa = vd->vdev_spa); |
fb5f0bc8 BB |
1544 | |
1545 | ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
1546 | ||
428870ff BB |
1547 | /* |
1548 | * If our parent is reopening, then we are as well, unless we are | |
1549 | * going offline. | |
1550 | */ | |
1551 | if (pvd != NULL && pvd->vdev_reopening) | |
1552 | vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline); | |
1553 | ||
34dc7c2f BB |
1554 | vd->vdev_ops->vdev_op_close(vd); |
1555 | ||
1556 | vdev_cache_purge(vd); | |
1557 | ||
1558 | /* | |
9babb374 | 1559 | * We record the previous state before we close it, so that if we are |
34dc7c2f BB |
1560 | * doing a reopen(), we don't generate FMA ereports if we notice that |
1561 | * it's still faulted. | |
1562 | */ | |
1563 | vd->vdev_prevstate = vd->vdev_state; | |
1564 | ||
1565 | if (vd->vdev_offline) | |
1566 | vd->vdev_state = VDEV_STATE_OFFLINE; | |
1567 | else | |
1568 | vd->vdev_state = VDEV_STATE_CLOSED; | |
1569 | vd->vdev_stat.vs_aux = VDEV_AUX_NONE; | |
1570 | } | |
1571 | ||
428870ff BB |
1572 | void |
1573 | vdev_hold(vdev_t *vd) | |
1574 | { | |
1575 | spa_t *spa = vd->vdev_spa; | |
d6320ddb | 1576 | int c; |
428870ff BB |
1577 | |
1578 | ASSERT(spa_is_root(spa)); | |
1579 | if (spa->spa_state == POOL_STATE_UNINITIALIZED) | |
1580 | return; | |
1581 | ||
d6320ddb | 1582 | for (c = 0; c < vd->vdev_children; c++) |
428870ff BB |
1583 | vdev_hold(vd->vdev_child[c]); |
1584 | ||
1585 | if (vd->vdev_ops->vdev_op_leaf) | |
1586 | vd->vdev_ops->vdev_op_hold(vd); | |
1587 | } | |
1588 | ||
1589 | void | |
1590 | vdev_rele(vdev_t *vd) | |
1591 | { | |
d6320ddb | 1592 | int c; |
428870ff | 1593 | |
d6320ddb BB |
1594 | ASSERT(spa_is_root(vd->vdev_spa)); |
1595 | for (c = 0; c < vd->vdev_children; c++) | |
428870ff BB |
1596 | vdev_rele(vd->vdev_child[c]); |
1597 | ||
1598 | if (vd->vdev_ops->vdev_op_leaf) | |
1599 | vd->vdev_ops->vdev_op_rele(vd); | |
1600 | } | |
1601 | ||
1602 | /* | |
1603 | * Reopen all interior vdevs and any unopened leaves. We don't actually | |
1604 | * reopen leaf vdevs which had previously been opened as they might deadlock | |
1605 | * on the spa_config_lock. Instead we only obtain the leaf's physical size. | |
1606 | * If the leaf has never been opened then open it, as usual. | |
1607 | */ | |
34dc7c2f BB |
1608 | void |
1609 | vdev_reopen(vdev_t *vd) | |
1610 | { | |
1611 | spa_t *spa = vd->vdev_spa; | |
1612 | ||
b128c09f | 1613 | ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); |
34dc7c2f | 1614 | |
428870ff BB |
1615 | /* set the reopening flag unless we're taking the vdev offline */ |
1616 | vd->vdev_reopening = !vd->vdev_offline; | |
34dc7c2f BB |
1617 | vdev_close(vd); |
1618 | (void) vdev_open(vd); | |
1619 | ||
1620 | /* | |
1621 | * Call vdev_validate() here to make sure we have the same device. | |
1622 | * Otherwise, a device with an invalid label could be successfully | |
1623 | * opened in response to vdev_reopen(). | |
1624 | */ | |
b128c09f BB |
1625 | if (vd->vdev_aux) { |
1626 | (void) vdev_validate_aux(vd); | |
1627 | if (vdev_readable(vd) && vdev_writeable(vd) && | |
9babb374 BB |
1628 | vd->vdev_aux == &spa->spa_l2cache && |
1629 | !l2arc_vdev_present(vd)) | |
1630 | l2arc_add_vdev(spa, vd); | |
b128c09f | 1631 | } else { |
295304be | 1632 | (void) vdev_validate(vd, B_TRUE); |
b128c09f | 1633 | } |
34dc7c2f BB |
1634 | |
1635 | /* | |
1636 | * Reassess parent vdev's health. | |
1637 | */ | |
1638 | vdev_propagate_state(vd); | |
1639 | } | |
1640 | ||
1641 | int | |
1642 | vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing) | |
1643 | { | |
1644 | int error; | |
1645 | ||
1646 | /* | |
1647 | * Normally, partial opens (e.g. of a mirror) are allowed. | |
1648 | * For a create, however, we want to fail the request if | |
1649 | * there are any components we can't open. | |
1650 | */ | |
1651 | error = vdev_open(vd); | |
1652 | ||
1653 | if (error || vd->vdev_state != VDEV_STATE_HEALTHY) { | |
1654 | vdev_close(vd); | |
1655 | return (error ? error : ENXIO); | |
1656 | } | |
1657 | ||
1658 | /* | |
93cf2076 | 1659 | * Recursively load DTLs and initialize all labels. |
34dc7c2f | 1660 | */ |
93cf2076 GW |
1661 | if ((error = vdev_dtl_load(vd)) != 0 || |
1662 | (error = vdev_label_init(vd, txg, isreplacing ? | |
34dc7c2f BB |
1663 | VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) { |
1664 | vdev_close(vd); | |
1665 | return (error); | |
1666 | } | |
1667 | ||
1668 | return (0); | |
1669 | } | |
1670 | ||
34dc7c2f | 1671 | void |
9babb374 | 1672 | vdev_metaslab_set_size(vdev_t *vd) |
34dc7c2f BB |
1673 | { |
1674 | /* | |
b8bcca18 | 1675 | * Aim for roughly metaslabs_per_vdev (default 200) metaslabs per vdev. |
34dc7c2f | 1676 | */ |
b8bcca18 | 1677 | vd->vdev_ms_shift = highbit64(vd->vdev_asize / metaslabs_per_vdev); |
34dc7c2f | 1678 | vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT); |
34dc7c2f BB |
1679 | } |
1680 | ||
1681 | void | |
1682 | vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg) | |
1683 | { | |
1684 | ASSERT(vd == vd->vdev_top); | |
428870ff | 1685 | ASSERT(!vd->vdev_ishole); |
34dc7c2f | 1686 | ASSERT(ISP2(flags)); |
572e2857 | 1687 | ASSERT(spa_writeable(vd->vdev_spa)); |
34dc7c2f BB |
1688 | |
1689 | if (flags & VDD_METASLAB) | |
1690 | (void) txg_list_add(&vd->vdev_ms_list, arg, txg); | |
1691 | ||
1692 | if (flags & VDD_DTL) | |
1693 | (void) txg_list_add(&vd->vdev_dtl_list, arg, txg); | |
1694 | ||
1695 | (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg); | |
1696 | } | |
1697 | ||
93cf2076 GW |
1698 | void |
1699 | vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg) | |
1700 | { | |
1701 | int c; | |
1702 | ||
1703 | for (c = 0; c < vd->vdev_children; c++) | |
1704 | vdev_dirty_leaves(vd->vdev_child[c], flags, txg); | |
1705 | ||
1706 | if (vd->vdev_ops->vdev_op_leaf) | |
1707 | vdev_dirty(vd->vdev_top, flags, vd, txg); | |
1708 | } | |
1709 | ||
fb5f0bc8 BB |
1710 | /* |
1711 | * DTLs. | |
1712 | * | |
1713 | * A vdev's DTL (dirty time log) is the set of transaction groups for which | |
428870ff | 1714 | * the vdev has less than perfect replication. There are four kinds of DTL: |
fb5f0bc8 BB |
1715 | * |
1716 | * DTL_MISSING: txgs for which the vdev has no valid copies of the data | |
1717 | * | |
1718 | * DTL_PARTIAL: txgs for which data is available, but not fully replicated | |
1719 | * | |
1720 | * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon | |
1721 | * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of | |
1722 | * txgs that was scrubbed. | |
1723 | * | |
1724 | * DTL_OUTAGE: txgs which cannot currently be read, whether due to | |
1725 | * persistent errors or just some device being offline. | |
1726 | * Unlike the other three, the DTL_OUTAGE map is not generally | |
1727 | * maintained; it's only computed when needed, typically to | |
1728 | * determine whether a device can be detached. | |
1729 | * | |
1730 | * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device | |
1731 | * either has the data or it doesn't. | |
1732 | * | |
1733 | * For interior vdevs such as mirror and RAID-Z the picture is more complex. | |
1734 | * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because | |
1735 | * if any child is less than fully replicated, then so is its parent. | |
1736 | * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs, | |
1737 | * comprising only those txgs which appear in 'maxfaults' or more children; | |
1738 | * those are the txgs we don't have enough replication to read. For example, | |
1739 | * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2); | |
1740 | * thus, its DTL_MISSING consists of the set of txgs that appear in more than | |
1741 | * two child DTL_MISSING maps. | |
1742 | * | |
1743 | * It should be clear from the above that to compute the DTLs and outage maps | |
1744 | * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps. | |
1745 | * Therefore, that is all we keep on disk. When loading the pool, or after | |
1746 | * a configuration change, we generate all other DTLs from first principles. | |
1747 | */ | |
34dc7c2f | 1748 | void |
fb5f0bc8 | 1749 | vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) |
34dc7c2f | 1750 | { |
93cf2076 | 1751 | range_tree_t *rt = vd->vdev_dtl[t]; |
fb5f0bc8 BB |
1752 | |
1753 | ASSERT(t < DTL_TYPES); | |
1754 | ASSERT(vd != vd->vdev_spa->spa_root_vdev); | |
572e2857 | 1755 | ASSERT(spa_writeable(vd->vdev_spa)); |
fb5f0bc8 | 1756 | |
93cf2076 GW |
1757 | mutex_enter(rt->rt_lock); |
1758 | if (!range_tree_contains(rt, txg, size)) | |
1759 | range_tree_add(rt, txg, size); | |
1760 | mutex_exit(rt->rt_lock); | |
34dc7c2f BB |
1761 | } |
1762 | ||
fb5f0bc8 BB |
1763 | boolean_t |
1764 | vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) | |
34dc7c2f | 1765 | { |
93cf2076 | 1766 | range_tree_t *rt = vd->vdev_dtl[t]; |
fb5f0bc8 | 1767 | boolean_t dirty = B_FALSE; |
34dc7c2f | 1768 | |
fb5f0bc8 BB |
1769 | ASSERT(t < DTL_TYPES); |
1770 | ASSERT(vd != vd->vdev_spa->spa_root_vdev); | |
34dc7c2f | 1771 | |
93cf2076 GW |
1772 | mutex_enter(rt->rt_lock); |
1773 | if (range_tree_space(rt) != 0) | |
1774 | dirty = range_tree_contains(rt, txg, size); | |
1775 | mutex_exit(rt->rt_lock); | |
34dc7c2f BB |
1776 | |
1777 | return (dirty); | |
1778 | } | |
1779 | ||
fb5f0bc8 BB |
1780 | boolean_t |
1781 | vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t) | |
1782 | { | |
93cf2076 | 1783 | range_tree_t *rt = vd->vdev_dtl[t]; |
fb5f0bc8 BB |
1784 | boolean_t empty; |
1785 | ||
93cf2076 GW |
1786 | mutex_enter(rt->rt_lock); |
1787 | empty = (range_tree_space(rt) == 0); | |
1788 | mutex_exit(rt->rt_lock); | |
fb5f0bc8 BB |
1789 | |
1790 | return (empty); | |
1791 | } | |
1792 | ||
5d1f7fb6 GW |
1793 | /* |
1794 | * Returns the lowest txg in the DTL range. | |
1795 | */ | |
1796 | static uint64_t | |
1797 | vdev_dtl_min(vdev_t *vd) | |
1798 | { | |
93cf2076 | 1799 | range_seg_t *rs; |
5d1f7fb6 GW |
1800 | |
1801 | ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock)); | |
93cf2076 | 1802 | ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0); |
5d1f7fb6 GW |
1803 | ASSERT0(vd->vdev_children); |
1804 | ||
93cf2076 GW |
1805 | rs = avl_first(&vd->vdev_dtl[DTL_MISSING]->rt_root); |
1806 | return (rs->rs_start - 1); | |
5d1f7fb6 GW |
1807 | } |
1808 | ||
1809 | /* | |
1810 | * Returns the highest txg in the DTL. | |
1811 | */ | |
1812 | static uint64_t | |
1813 | vdev_dtl_max(vdev_t *vd) | |
1814 | { | |
93cf2076 | 1815 | range_seg_t *rs; |
5d1f7fb6 GW |
1816 | |
1817 | ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock)); | |
93cf2076 | 1818 | ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0); |
5d1f7fb6 GW |
1819 | ASSERT0(vd->vdev_children); |
1820 | ||
93cf2076 GW |
1821 | rs = avl_last(&vd->vdev_dtl[DTL_MISSING]->rt_root); |
1822 | return (rs->rs_end); | |
5d1f7fb6 GW |
1823 | } |
1824 | ||
1825 | /* | |
1826 | * Determine if a resilvering vdev should remove any DTL entries from | |
1827 | * its range. If the vdev was resilvering for the entire duration of the | |
1828 | * scan then it should excise that range from its DTLs. Otherwise, this | |
1829 | * vdev is considered partially resilvered and should leave its DTL | |
1830 | * entries intact. The comment in vdev_dtl_reassess() describes how we | |
1831 | * excise the DTLs. | |
1832 | */ | |
1833 | static boolean_t | |
1834 | vdev_dtl_should_excise(vdev_t *vd) | |
1835 | { | |
1836 | spa_t *spa = vd->vdev_spa; | |
1837 | dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; | |
1838 | ||
1839 | ASSERT0(scn->scn_phys.scn_errors); | |
1840 | ASSERT0(vd->vdev_children); | |
1841 | ||
1842 | if (vd->vdev_resilver_txg == 0 || | |
93cf2076 | 1843 | range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0) |
5d1f7fb6 GW |
1844 | return (B_TRUE); |
1845 | ||
1846 | /* | |
1847 | * When a resilver is initiated the scan will assign the scn_max_txg | |
1848 | * value to the highest txg value that exists in all DTLs. If this | |
1849 | * device's max DTL is not part of this scan (i.e. it is not in | |
1850 | * the range (scn_min_txg, scn_max_txg] then it is not eligible | |
1851 | * for excision. | |
1852 | */ | |
1853 | if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) { | |
1854 | ASSERT3U(scn->scn_phys.scn_min_txg, <=, vdev_dtl_min(vd)); | |
1855 | ASSERT3U(scn->scn_phys.scn_min_txg, <, vd->vdev_resilver_txg); | |
1856 | ASSERT3U(vd->vdev_resilver_txg, <=, scn->scn_phys.scn_max_txg); | |
1857 | return (B_TRUE); | |
1858 | } | |
1859 | return (B_FALSE); | |
1860 | } | |
1861 | ||
34dc7c2f BB |
1862 | /* |
1863 | * Reassess DTLs after a config change or scrub completion. | |
1864 | */ | |
1865 | void | |
1866 | vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done) | |
1867 | { | |
1868 | spa_t *spa = vd->vdev_spa; | |
fb5f0bc8 | 1869 | avl_tree_t reftree; |
d6320ddb | 1870 | int c, t, minref; |
34dc7c2f | 1871 | |
fb5f0bc8 | 1872 | ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); |
34dc7c2f | 1873 | |
d6320ddb | 1874 | for (c = 0; c < vd->vdev_children; c++) |
fb5f0bc8 BB |
1875 | vdev_dtl_reassess(vd->vdev_child[c], txg, |
1876 | scrub_txg, scrub_done); | |
1877 | ||
428870ff | 1878 | if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux) |
fb5f0bc8 BB |
1879 | return; |
1880 | ||
1881 | if (vd->vdev_ops->vdev_op_leaf) { | |
428870ff BB |
1882 | dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; |
1883 | ||
34dc7c2f | 1884 | mutex_enter(&vd->vdev_dtl_lock); |
5d1f7fb6 GW |
1885 | |
1886 | /* | |
1887 | * If we've completed a scan cleanly then determine | |
1888 | * if this vdev should remove any DTLs. We only want to | |
1889 | * excise regions on vdevs that were available during | |
1890 | * the entire duration of this scan. | |
1891 | */ | |
b128c09f | 1892 | if (scrub_txg != 0 && |
428870ff | 1893 | (spa->spa_scrub_started || |
5d1f7fb6 GW |
1894 | (scn != NULL && scn->scn_phys.scn_errors == 0)) && |
1895 | vdev_dtl_should_excise(vd)) { | |
b128c09f BB |
1896 | /* |
1897 | * We completed a scrub up to scrub_txg. If we | |
1898 | * did it without rebooting, then the scrub dtl | |
1899 | * will be valid, so excise the old region and | |
1900 | * fold in the scrub dtl. Otherwise, leave the | |
1901 | * dtl as-is if there was an error. | |
fb5f0bc8 BB |
1902 | * |
1903 | * There's little trick here: to excise the beginning | |
1904 | * of the DTL_MISSING map, we put it into a reference | |
1905 | * tree and then add a segment with refcnt -1 that | |
1906 | * covers the range [0, scrub_txg). This means | |
1907 | * that each txg in that range has refcnt -1 or 0. | |
1908 | * We then add DTL_SCRUB with a refcnt of 2, so that | |
1909 | * entries in the range [0, scrub_txg) will have a | |
1910 | * positive refcnt -- either 1 or 2. We then convert | |
1911 | * the reference tree into the new DTL_MISSING map. | |
b128c09f | 1912 | */ |
93cf2076 GW |
1913 | space_reftree_create(&reftree); |
1914 | space_reftree_add_map(&reftree, | |
1915 | vd->vdev_dtl[DTL_MISSING], 1); | |
1916 | space_reftree_add_seg(&reftree, 0, scrub_txg, -1); | |
1917 | space_reftree_add_map(&reftree, | |
1918 | vd->vdev_dtl[DTL_SCRUB], 2); | |
1919 | space_reftree_generate_map(&reftree, | |
1920 | vd->vdev_dtl[DTL_MISSING], 1); | |
1921 | space_reftree_destroy(&reftree); | |
34dc7c2f | 1922 | } |
93cf2076 GW |
1923 | range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL); |
1924 | range_tree_walk(vd->vdev_dtl[DTL_MISSING], | |
1925 | range_tree_add, vd->vdev_dtl[DTL_PARTIAL]); | |
34dc7c2f | 1926 | if (scrub_done) |
93cf2076 GW |
1927 | range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL); |
1928 | range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL); | |
fb5f0bc8 | 1929 | if (!vdev_readable(vd)) |
93cf2076 | 1930 | range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL); |
fb5f0bc8 | 1931 | else |
93cf2076 GW |
1932 | range_tree_walk(vd->vdev_dtl[DTL_MISSING], |
1933 | range_tree_add, vd->vdev_dtl[DTL_OUTAGE]); | |
5d1f7fb6 GW |
1934 | |
1935 | /* | |
1936 | * If the vdev was resilvering and no longer has any | |
d14fa5db | 1937 | * DTLs then reset its resilvering flag and dirty |
1938 | * the top level so that we persist the change. | |
5d1f7fb6 GW |
1939 | */ |
1940 | if (vd->vdev_resilver_txg != 0 && | |
93cf2076 | 1941 | range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0 && |
d14fa5db | 1942 | range_tree_space(vd->vdev_dtl[DTL_OUTAGE]) == 0) { |
5d1f7fb6 | 1943 | vd->vdev_resilver_txg = 0; |
d14fa5db | 1944 | vdev_config_dirty(vd->vdev_top); |
1945 | } | |
5d1f7fb6 | 1946 | |
34dc7c2f | 1947 | mutex_exit(&vd->vdev_dtl_lock); |
b128c09f | 1948 | |
34dc7c2f BB |
1949 | if (txg != 0) |
1950 | vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg); | |
1951 | return; | |
1952 | } | |
1953 | ||
34dc7c2f | 1954 | mutex_enter(&vd->vdev_dtl_lock); |
d6320ddb | 1955 | for (t = 0; t < DTL_TYPES; t++) { |
93cf2076 GW |
1956 | int c; |
1957 | ||
428870ff BB |
1958 | /* account for child's outage in parent's missing map */ |
1959 | int s = (t == DTL_MISSING) ? DTL_OUTAGE: t; | |
fb5f0bc8 BB |
1960 | if (t == DTL_SCRUB) |
1961 | continue; /* leaf vdevs only */ | |
1962 | if (t == DTL_PARTIAL) | |
1963 | minref = 1; /* i.e. non-zero */ | |
1964 | else if (vd->vdev_nparity != 0) | |
1965 | minref = vd->vdev_nparity + 1; /* RAID-Z */ | |
1966 | else | |
1967 | minref = vd->vdev_children; /* any kind of mirror */ | |
93cf2076 | 1968 | space_reftree_create(&reftree); |
d6320ddb | 1969 | for (c = 0; c < vd->vdev_children; c++) { |
fb5f0bc8 BB |
1970 | vdev_t *cvd = vd->vdev_child[c]; |
1971 | mutex_enter(&cvd->vdev_dtl_lock); | |
93cf2076 | 1972 | space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1); |
fb5f0bc8 BB |
1973 | mutex_exit(&cvd->vdev_dtl_lock); |
1974 | } | |
93cf2076 GW |
1975 | space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref); |
1976 | space_reftree_destroy(&reftree); | |
34dc7c2f | 1977 | } |
fb5f0bc8 | 1978 | mutex_exit(&vd->vdev_dtl_lock); |
34dc7c2f BB |
1979 | } |
1980 | ||
93cf2076 | 1981 | int |
34dc7c2f BB |
1982 | vdev_dtl_load(vdev_t *vd) |
1983 | { | |
1984 | spa_t *spa = vd->vdev_spa; | |
34dc7c2f | 1985 | objset_t *mos = spa->spa_meta_objset; |
93cf2076 GW |
1986 | int error = 0; |
1987 | int c; | |
34dc7c2f | 1988 | |
93cf2076 GW |
1989 | if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) { |
1990 | ASSERT(!vd->vdev_ishole); | |
34dc7c2f | 1991 | |
93cf2076 GW |
1992 | error = space_map_open(&vd->vdev_dtl_sm, mos, |
1993 | vd->vdev_dtl_object, 0, -1ULL, 0, &vd->vdev_dtl_lock); | |
1994 | if (error) | |
1995 | return (error); | |
1996 | ASSERT(vd->vdev_dtl_sm != NULL); | |
34dc7c2f | 1997 | |
93cf2076 | 1998 | mutex_enter(&vd->vdev_dtl_lock); |
428870ff | 1999 | |
93cf2076 GW |
2000 | /* |
2001 | * Now that we've opened the space_map we need to update | |
2002 | * the in-core DTL. | |
2003 | */ | |
2004 | space_map_update(vd->vdev_dtl_sm); | |
34dc7c2f | 2005 | |
93cf2076 GW |
2006 | error = space_map_load(vd->vdev_dtl_sm, |
2007 | vd->vdev_dtl[DTL_MISSING], SM_ALLOC); | |
2008 | mutex_exit(&vd->vdev_dtl_lock); | |
34dc7c2f | 2009 | |
93cf2076 GW |
2010 | return (error); |
2011 | } | |
2012 | ||
2013 | for (c = 0; c < vd->vdev_children; c++) { | |
2014 | error = vdev_dtl_load(vd->vdev_child[c]); | |
2015 | if (error != 0) | |
2016 | break; | |
2017 | } | |
34dc7c2f BB |
2018 | |
2019 | return (error); | |
2020 | } | |
2021 | ||
e0ab3ab5 JS |
2022 | void |
2023 | vdev_destroy_unlink_zap(vdev_t *vd, uint64_t zapobj, dmu_tx_t *tx) | |
2024 | { | |
2025 | spa_t *spa = vd->vdev_spa; | |
2026 | ||
2027 | VERIFY0(zap_destroy(spa->spa_meta_objset, zapobj, tx)); | |
2028 | VERIFY0(zap_remove_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps, | |
2029 | zapobj, tx)); | |
2030 | } | |
2031 | ||
2032 | uint64_t | |
2033 | vdev_create_link_zap(vdev_t *vd, dmu_tx_t *tx) | |
2034 | { | |
2035 | spa_t *spa = vd->vdev_spa; | |
2036 | uint64_t zap = zap_create(spa->spa_meta_objset, DMU_OTN_ZAP_METADATA, | |
2037 | DMU_OT_NONE, 0, tx); | |
2038 | ||
2039 | ASSERT(zap != 0); | |
2040 | VERIFY0(zap_add_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps, | |
2041 | zap, tx)); | |
2042 | ||
2043 | return (zap); | |
2044 | } | |
2045 | ||
2046 | void | |
2047 | vdev_construct_zaps(vdev_t *vd, dmu_tx_t *tx) | |
2048 | { | |
2049 | uint64_t i; | |
2050 | ||
2051 | if (vd->vdev_ops != &vdev_hole_ops && | |
2052 | vd->vdev_ops != &vdev_missing_ops && | |
2053 | vd->vdev_ops != &vdev_root_ops && | |
2054 | !vd->vdev_top->vdev_removing) { | |
2055 | if (vd->vdev_ops->vdev_op_leaf && vd->vdev_leaf_zap == 0) { | |
2056 | vd->vdev_leaf_zap = vdev_create_link_zap(vd, tx); | |
2057 | } | |
2058 | if (vd == vd->vdev_top && vd->vdev_top_zap == 0) { | |
2059 | vd->vdev_top_zap = vdev_create_link_zap(vd, tx); | |
2060 | } | |
2061 | } | |
2062 | for (i = 0; i < vd->vdev_children; i++) { | |
2063 | vdev_construct_zaps(vd->vdev_child[i], tx); | |
2064 | } | |
2065 | } | |
2066 | ||
34dc7c2f BB |
2067 | void |
2068 | vdev_dtl_sync(vdev_t *vd, uint64_t txg) | |
2069 | { | |
2070 | spa_t *spa = vd->vdev_spa; | |
93cf2076 | 2071 | range_tree_t *rt = vd->vdev_dtl[DTL_MISSING]; |
34dc7c2f | 2072 | objset_t *mos = spa->spa_meta_objset; |
93cf2076 GW |
2073 | range_tree_t *rtsync; |
2074 | kmutex_t rtlock; | |
34dc7c2f | 2075 | dmu_tx_t *tx; |
93cf2076 | 2076 | uint64_t object = space_map_object(vd->vdev_dtl_sm); |
34dc7c2f | 2077 | |
428870ff | 2078 | ASSERT(!vd->vdev_ishole); |
93cf2076 | 2079 | ASSERT(vd->vdev_ops->vdev_op_leaf); |
428870ff | 2080 | |
34dc7c2f BB |
2081 | tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); |
2082 | ||
93cf2076 GW |
2083 | if (vd->vdev_detached || vd->vdev_top->vdev_removing) { |
2084 | mutex_enter(&vd->vdev_dtl_lock); | |
2085 | space_map_free(vd->vdev_dtl_sm, tx); | |
2086 | space_map_close(vd->vdev_dtl_sm); | |
2087 | vd->vdev_dtl_sm = NULL; | |
2088 | mutex_exit(&vd->vdev_dtl_lock); | |
e0ab3ab5 JS |
2089 | |
2090 | /* | |
2091 | * We only destroy the leaf ZAP for detached leaves or for | |
2092 | * removed log devices. Removed data devices handle leaf ZAP | |
2093 | * cleanup later, once cancellation is no longer possible. | |
2094 | */ | |
2095 | if (vd->vdev_leaf_zap != 0 && (vd->vdev_detached || | |
2096 | vd->vdev_top->vdev_islog)) { | |
2097 | vdev_destroy_unlink_zap(vd, vd->vdev_leaf_zap, tx); | |
2098 | vd->vdev_leaf_zap = 0; | |
2099 | } | |
2100 | ||
34dc7c2f | 2101 | dmu_tx_commit(tx); |
34dc7c2f BB |
2102 | return; |
2103 | } | |
2104 | ||
93cf2076 GW |
2105 | if (vd->vdev_dtl_sm == NULL) { |
2106 | uint64_t new_object; | |
2107 | ||
2108 | new_object = space_map_alloc(mos, tx); | |
2109 | VERIFY3U(new_object, !=, 0); | |
2110 | ||
2111 | VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object, | |
2112 | 0, -1ULL, 0, &vd->vdev_dtl_lock)); | |
2113 | ASSERT(vd->vdev_dtl_sm != NULL); | |
34dc7c2f BB |
2114 | } |
2115 | ||
93cf2076 | 2116 | mutex_init(&rtlock, NULL, MUTEX_DEFAULT, NULL); |
34dc7c2f | 2117 | |
93cf2076 | 2118 | rtsync = range_tree_create(NULL, NULL, &rtlock); |
34dc7c2f | 2119 | |
93cf2076 | 2120 | mutex_enter(&rtlock); |
34dc7c2f BB |
2121 | |
2122 | mutex_enter(&vd->vdev_dtl_lock); | |
93cf2076 | 2123 | range_tree_walk(rt, range_tree_add, rtsync); |
34dc7c2f BB |
2124 | mutex_exit(&vd->vdev_dtl_lock); |
2125 | ||
93cf2076 GW |
2126 | space_map_truncate(vd->vdev_dtl_sm, tx); |
2127 | space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, tx); | |
2128 | range_tree_vacate(rtsync, NULL, NULL); | |
34dc7c2f | 2129 | |
93cf2076 | 2130 | range_tree_destroy(rtsync); |
34dc7c2f | 2131 | |
93cf2076 GW |
2132 | mutex_exit(&rtlock); |
2133 | mutex_destroy(&rtlock); | |
34dc7c2f | 2134 | |
93cf2076 GW |
2135 | /* |
2136 | * If the object for the space map has changed then dirty | |
2137 | * the top level so that we update the config. | |
2138 | */ | |
2139 | if (object != space_map_object(vd->vdev_dtl_sm)) { | |
2140 | zfs_dbgmsg("txg %llu, spa %s, DTL old object %llu, " | |
2141 | "new object %llu", txg, spa_name(spa), object, | |
2142 | space_map_object(vd->vdev_dtl_sm)); | |
2143 | vdev_config_dirty(vd->vdev_top); | |
2144 | } | |
34dc7c2f BB |
2145 | |
2146 | dmu_tx_commit(tx); | |
93cf2076 GW |
2147 | |
2148 | mutex_enter(&vd->vdev_dtl_lock); | |
2149 | space_map_update(vd->vdev_dtl_sm); | |
2150 | mutex_exit(&vd->vdev_dtl_lock); | |
34dc7c2f BB |
2151 | } |
2152 | ||
fb5f0bc8 BB |
2153 | /* |
2154 | * Determine whether the specified vdev can be offlined/detached/removed | |
2155 | * without losing data. | |
2156 | */ | |
2157 | boolean_t | |
2158 | vdev_dtl_required(vdev_t *vd) | |
2159 | { | |
2160 | spa_t *spa = vd->vdev_spa; | |
2161 | vdev_t *tvd = vd->vdev_top; | |
2162 | uint8_t cant_read = vd->vdev_cant_read; | |
2163 | boolean_t required; | |
2164 | ||
2165 | ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
2166 | ||
2167 | if (vd == spa->spa_root_vdev || vd == tvd) | |
2168 | return (B_TRUE); | |
2169 | ||
2170 | /* | |
2171 | * Temporarily mark the device as unreadable, and then determine | |
2172 | * whether this results in any DTL outages in the top-level vdev. | |
2173 | * If not, we can safely offline/detach/remove the device. | |
2174 | */ | |
2175 | vd->vdev_cant_read = B_TRUE; | |
2176 | vdev_dtl_reassess(tvd, 0, 0, B_FALSE); | |
2177 | required = !vdev_dtl_empty(tvd, DTL_OUTAGE); | |
2178 | vd->vdev_cant_read = cant_read; | |
2179 | vdev_dtl_reassess(tvd, 0, 0, B_FALSE); | |
2180 | ||
572e2857 BB |
2181 | if (!required && zio_injection_enabled) |
2182 | required = !!zio_handle_device_injection(vd, NULL, ECHILD); | |
2183 | ||
fb5f0bc8 BB |
2184 | return (required); |
2185 | } | |
2186 | ||
b128c09f BB |
2187 | /* |
2188 | * Determine if resilver is needed, and if so the txg range. | |
2189 | */ | |
2190 | boolean_t | |
2191 | vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp) | |
2192 | { | |
2193 | boolean_t needed = B_FALSE; | |
2194 | uint64_t thismin = UINT64_MAX; | |
2195 | uint64_t thismax = 0; | |
d6320ddb | 2196 | int c; |
b128c09f BB |
2197 | |
2198 | if (vd->vdev_children == 0) { | |
2199 | mutex_enter(&vd->vdev_dtl_lock); | |
93cf2076 | 2200 | if (range_tree_space(vd->vdev_dtl[DTL_MISSING]) != 0 && |
fb5f0bc8 | 2201 | vdev_writeable(vd)) { |
b128c09f | 2202 | |
5d1f7fb6 GW |
2203 | thismin = vdev_dtl_min(vd); |
2204 | thismax = vdev_dtl_max(vd); | |
b128c09f BB |
2205 | needed = B_TRUE; |
2206 | } | |
2207 | mutex_exit(&vd->vdev_dtl_lock); | |
2208 | } else { | |
d6320ddb | 2209 | for (c = 0; c < vd->vdev_children; c++) { |
b128c09f BB |
2210 | vdev_t *cvd = vd->vdev_child[c]; |
2211 | uint64_t cmin, cmax; | |
2212 | ||
2213 | if (vdev_resilver_needed(cvd, &cmin, &cmax)) { | |
2214 | thismin = MIN(thismin, cmin); | |
2215 | thismax = MAX(thismax, cmax); | |
2216 | needed = B_TRUE; | |
2217 | } | |
2218 | } | |
2219 | } | |
2220 | ||
2221 | if (needed && minp) { | |
2222 | *minp = thismin; | |
2223 | *maxp = thismax; | |
2224 | } | |
2225 | return (needed); | |
2226 | } | |
2227 | ||
34dc7c2f BB |
2228 | void |
2229 | vdev_load(vdev_t *vd) | |
2230 | { | |
d6320ddb BB |
2231 | int c; |
2232 | ||
34dc7c2f BB |
2233 | /* |
2234 | * Recursively load all children. | |
2235 | */ | |
d6320ddb | 2236 | for (c = 0; c < vd->vdev_children; c++) |
34dc7c2f BB |
2237 | vdev_load(vd->vdev_child[c]); |
2238 | ||
2239 | /* | |
2240 | * If this is a top-level vdev, initialize its metaslabs. | |
2241 | */ | |
428870ff | 2242 | if (vd == vd->vdev_top && !vd->vdev_ishole && |
34dc7c2f BB |
2243 | (vd->vdev_ashift == 0 || vd->vdev_asize == 0 || |
2244 | vdev_metaslab_init(vd, 0) != 0)) | |
2245 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
2246 | VDEV_AUX_CORRUPT_DATA); | |
34dc7c2f BB |
2247 | /* |
2248 | * If this is a leaf vdev, load its DTL. | |
2249 | */ | |
2250 | if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0) | |
2251 | vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
2252 | VDEV_AUX_CORRUPT_DATA); | |
2253 | } | |
2254 | ||
2255 | /* | |
2256 | * The special vdev case is used for hot spares and l2cache devices. Its | |
2257 | * sole purpose it to set the vdev state for the associated vdev. To do this, | |
2258 | * we make sure that we can open the underlying device, then try to read the | |
2259 | * label, and make sure that the label is sane and that it hasn't been | |
2260 | * repurposed to another pool. | |
2261 | */ | |
2262 | int | |
2263 | vdev_validate_aux(vdev_t *vd) | |
2264 | { | |
2265 | nvlist_t *label; | |
2266 | uint64_t guid, version; | |
2267 | uint64_t state; | |
2268 | ||
b128c09f BB |
2269 | if (!vdev_readable(vd)) |
2270 | return (0); | |
2271 | ||
3bc7e0fb | 2272 | if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) { |
34dc7c2f BB |
2273 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, |
2274 | VDEV_AUX_CORRUPT_DATA); | |
2275 | return (-1); | |
2276 | } | |
2277 | ||
2278 | if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 || | |
9ae529ec | 2279 | !SPA_VERSION_IS_SUPPORTED(version) || |
34dc7c2f BB |
2280 | nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 || |
2281 | guid != vd->vdev_guid || | |
2282 | nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) { | |
2283 | vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
2284 | VDEV_AUX_CORRUPT_DATA); | |
2285 | nvlist_free(label); | |
2286 | return (-1); | |
2287 | } | |
2288 | ||
2289 | /* | |
2290 | * We don't actually check the pool state here. If it's in fact in | |
2291 | * use by another pool, we update this fact on the fly when requested. | |
2292 | */ | |
2293 | nvlist_free(label); | |
2294 | return (0); | |
2295 | } | |
2296 | ||
428870ff BB |
2297 | void |
2298 | vdev_remove(vdev_t *vd, uint64_t txg) | |
2299 | { | |
2300 | spa_t *spa = vd->vdev_spa; | |
2301 | objset_t *mos = spa->spa_meta_objset; | |
2302 | dmu_tx_t *tx; | |
f3a7f661 | 2303 | int m, i; |
428870ff BB |
2304 | |
2305 | tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg); | |
e0ab3ab5 JS |
2306 | ASSERT(vd == vd->vdev_top); |
2307 | ASSERT3U(txg, ==, spa_syncing_txg(spa)); | |
428870ff | 2308 | |
428870ff | 2309 | if (vd->vdev_ms != NULL) { |
f3a7f661 GW |
2310 | metaslab_group_t *mg = vd->vdev_mg; |
2311 | ||
2312 | metaslab_group_histogram_verify(mg); | |
2313 | metaslab_class_histogram_verify(mg->mg_class); | |
2314 | ||
d6320ddb | 2315 | for (m = 0; m < vd->vdev_ms_count; m++) { |
428870ff BB |
2316 | metaslab_t *msp = vd->vdev_ms[m]; |
2317 | ||
93cf2076 | 2318 | if (msp == NULL || msp->ms_sm == NULL) |
428870ff BB |
2319 | continue; |
2320 | ||
93cf2076 | 2321 | mutex_enter(&msp->ms_lock); |
f3a7f661 GW |
2322 | /* |
2323 | * If the metaslab was not loaded when the vdev | |
2324 | * was removed then the histogram accounting may | |
2325 | * not be accurate. Update the histogram information | |
2326 | * here so that we ensure that the metaslab group | |
2327 | * and metaslab class are up-to-date. | |
2328 | */ | |
2329 | metaslab_group_histogram_remove(mg, msp); | |
2330 | ||
93cf2076 GW |
2331 | VERIFY0(space_map_allocated(msp->ms_sm)); |
2332 | space_map_free(msp->ms_sm, tx); | |
2333 | space_map_close(msp->ms_sm); | |
2334 | msp->ms_sm = NULL; | |
2335 | mutex_exit(&msp->ms_lock); | |
428870ff | 2336 | } |
f3a7f661 GW |
2337 | |
2338 | metaslab_group_histogram_verify(mg); | |
2339 | metaslab_class_histogram_verify(mg->mg_class); | |
2340 | for (i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) | |
2341 | ASSERT0(mg->mg_histogram[i]); | |
2342 | ||
428870ff BB |
2343 | } |
2344 | ||
2345 | if (vd->vdev_ms_array) { | |
2346 | (void) dmu_object_free(mos, vd->vdev_ms_array, tx); | |
2347 | vd->vdev_ms_array = 0; | |
428870ff | 2348 | } |
e0ab3ab5 JS |
2349 | |
2350 | if (vd->vdev_islog && vd->vdev_top_zap != 0) { | |
2351 | vdev_destroy_unlink_zap(vd, vd->vdev_top_zap, tx); | |
2352 | vd->vdev_top_zap = 0; | |
2353 | } | |
428870ff BB |
2354 | dmu_tx_commit(tx); |
2355 | } | |
2356 | ||
34dc7c2f BB |
2357 | void |
2358 | vdev_sync_done(vdev_t *vd, uint64_t txg) | |
2359 | { | |
2360 | metaslab_t *msp; | |
428870ff BB |
2361 | boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg)); |
2362 | ||
2363 | ASSERT(!vd->vdev_ishole); | |
34dc7c2f | 2364 | |
c65aa5b2 | 2365 | while ((msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))) |
34dc7c2f | 2366 | metaslab_sync_done(msp, txg); |
428870ff BB |
2367 | |
2368 | if (reassess) | |
2369 | metaslab_sync_reassess(vd->vdev_mg); | |
34dc7c2f BB |
2370 | } |
2371 | ||
2372 | void | |
2373 | vdev_sync(vdev_t *vd, uint64_t txg) | |
2374 | { | |
2375 | spa_t *spa = vd->vdev_spa; | |
2376 | vdev_t *lvd; | |
2377 | metaslab_t *msp; | |
2378 | dmu_tx_t *tx; | |
2379 | ||
428870ff BB |
2380 | ASSERT(!vd->vdev_ishole); |
2381 | ||
34dc7c2f BB |
2382 | if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) { |
2383 | ASSERT(vd == vd->vdev_top); | |
2384 | tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); | |
2385 | vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset, | |
2386 | DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx); | |
2387 | ASSERT(vd->vdev_ms_array != 0); | |
2388 | vdev_config_dirty(vd); | |
2389 | dmu_tx_commit(tx); | |
2390 | } | |
2391 | ||
428870ff BB |
2392 | /* |
2393 | * Remove the metadata associated with this vdev once it's empty. | |
2394 | */ | |
2395 | if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing) | |
2396 | vdev_remove(vd, txg); | |
2397 | ||
34dc7c2f BB |
2398 | while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) { |
2399 | metaslab_sync(msp, txg); | |
2400 | (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg)); | |
2401 | } | |
2402 | ||
2403 | while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL) | |
2404 | vdev_dtl_sync(lvd, txg); | |
2405 | ||
2406 | (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)); | |
2407 | } | |
2408 | ||
2409 | uint64_t | |
2410 | vdev_psize_to_asize(vdev_t *vd, uint64_t psize) | |
2411 | { | |
2412 | return (vd->vdev_ops->vdev_op_asize(vd, psize)); | |
2413 | } | |
2414 | ||
34dc7c2f BB |
2415 | /* |
2416 | * Mark the given vdev faulted. A faulted vdev behaves as if the device could | |
2417 | * not be opened, and no I/O is attempted. | |
2418 | */ | |
2419 | int | |
428870ff | 2420 | vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux) |
34dc7c2f | 2421 | { |
572e2857 | 2422 | vdev_t *vd, *tvd; |
34dc7c2f | 2423 | |
428870ff | 2424 | spa_vdev_state_enter(spa, SCL_NONE); |
34dc7c2f | 2425 | |
b128c09f BB |
2426 | if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) |
2427 | return (spa_vdev_state_exit(spa, NULL, ENODEV)); | |
34dc7c2f | 2428 | |
34dc7c2f | 2429 | if (!vd->vdev_ops->vdev_op_leaf) |
b128c09f | 2430 | return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); |
34dc7c2f | 2431 | |
572e2857 BB |
2432 | tvd = vd->vdev_top; |
2433 | ||
428870ff BB |
2434 | /* |
2435 | * We don't directly use the aux state here, but if we do a | |
2436 | * vdev_reopen(), we need this value to be present to remember why we | |
2437 | * were faulted. | |
2438 | */ | |
2439 | vd->vdev_label_aux = aux; | |
2440 | ||
34dc7c2f BB |
2441 | /* |
2442 | * Faulted state takes precedence over degraded. | |
2443 | */ | |
428870ff | 2444 | vd->vdev_delayed_close = B_FALSE; |
34dc7c2f BB |
2445 | vd->vdev_faulted = 1ULL; |
2446 | vd->vdev_degraded = 0ULL; | |
428870ff | 2447 | vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux); |
34dc7c2f BB |
2448 | |
2449 | /* | |
428870ff BB |
2450 | * If this device has the only valid copy of the data, then |
2451 | * back off and simply mark the vdev as degraded instead. | |
34dc7c2f | 2452 | */ |
572e2857 | 2453 | if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) { |
34dc7c2f BB |
2454 | vd->vdev_degraded = 1ULL; |
2455 | vd->vdev_faulted = 0ULL; | |
2456 | ||
2457 | /* | |
2458 | * If we reopen the device and it's not dead, only then do we | |
2459 | * mark it degraded. | |
2460 | */ | |
572e2857 | 2461 | vdev_reopen(tvd); |
34dc7c2f | 2462 | |
428870ff BB |
2463 | if (vdev_readable(vd)) |
2464 | vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux); | |
34dc7c2f BB |
2465 | } |
2466 | ||
b128c09f | 2467 | return (spa_vdev_state_exit(spa, vd, 0)); |
34dc7c2f BB |
2468 | } |
2469 | ||
2470 | /* | |
2471 | * Mark the given vdev degraded. A degraded vdev is purely an indication to the | |
2472 | * user that something is wrong. The vdev continues to operate as normal as far | |
2473 | * as I/O is concerned. | |
2474 | */ | |
2475 | int | |
428870ff | 2476 | vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux) |
34dc7c2f | 2477 | { |
b128c09f | 2478 | vdev_t *vd; |
34dc7c2f | 2479 | |
428870ff | 2480 | spa_vdev_state_enter(spa, SCL_NONE); |
34dc7c2f | 2481 | |
b128c09f BB |
2482 | if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) |
2483 | return (spa_vdev_state_exit(spa, NULL, ENODEV)); | |
34dc7c2f | 2484 | |
34dc7c2f | 2485 | if (!vd->vdev_ops->vdev_op_leaf) |
b128c09f | 2486 | return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); |
34dc7c2f BB |
2487 | |
2488 | /* | |
2489 | * If the vdev is already faulted, then don't do anything. | |
2490 | */ | |
b128c09f BB |
2491 | if (vd->vdev_faulted || vd->vdev_degraded) |
2492 | return (spa_vdev_state_exit(spa, NULL, 0)); | |
34dc7c2f BB |
2493 | |
2494 | vd->vdev_degraded = 1ULL; | |
2495 | if (!vdev_is_dead(vd)) | |
2496 | vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, | |
428870ff | 2497 | aux); |
34dc7c2f | 2498 | |
b128c09f | 2499 | return (spa_vdev_state_exit(spa, vd, 0)); |
34dc7c2f BB |
2500 | } |
2501 | ||
2502 | /* | |
d3cc8b15 WA |
2503 | * Online the given vdev. |
2504 | * | |
2505 | * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached | |
2506 | * spare device should be detached when the device finishes resilvering. | |
2507 | * Second, the online should be treated like a 'test' online case, so no FMA | |
2508 | * events are generated if the device fails to open. | |
34dc7c2f BB |
2509 | */ |
2510 | int | |
b128c09f | 2511 | vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate) |
34dc7c2f | 2512 | { |
9babb374 | 2513 | vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev; |
fb390aaf | 2514 | boolean_t postevent = B_FALSE; |
34dc7c2f | 2515 | |
428870ff | 2516 | spa_vdev_state_enter(spa, SCL_NONE); |
34dc7c2f | 2517 | |
b128c09f BB |
2518 | if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) |
2519 | return (spa_vdev_state_exit(spa, NULL, ENODEV)); | |
34dc7c2f BB |
2520 | |
2521 | if (!vd->vdev_ops->vdev_op_leaf) | |
b128c09f | 2522 | return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); |
34dc7c2f | 2523 | |
fb390aaf HR |
2524 | postevent = |
2525 | (vd->vdev_offline == B_TRUE || vd->vdev_tmpoffline == B_TRUE) ? | |
2526 | B_TRUE : B_FALSE; | |
2527 | ||
9babb374 | 2528 | tvd = vd->vdev_top; |
34dc7c2f BB |
2529 | vd->vdev_offline = B_FALSE; |
2530 | vd->vdev_tmpoffline = B_FALSE; | |
b128c09f BB |
2531 | vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE); |
2532 | vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT); | |
9babb374 BB |
2533 | |
2534 | /* XXX - L2ARC 1.0 does not support expansion */ | |
2535 | if (!vd->vdev_aux) { | |
2536 | for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) | |
2537 | pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND); | |
2538 | } | |
2539 | ||
2540 | vdev_reopen(tvd); | |
34dc7c2f BB |
2541 | vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE; |
2542 | ||
9babb374 BB |
2543 | if (!vd->vdev_aux) { |
2544 | for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) | |
2545 | pvd->vdev_expanding = B_FALSE; | |
2546 | } | |
2547 | ||
34dc7c2f BB |
2548 | if (newstate) |
2549 | *newstate = vd->vdev_state; | |
2550 | if ((flags & ZFS_ONLINE_UNSPARE) && | |
2551 | !vdev_is_dead(vd) && vd->vdev_parent && | |
2552 | vd->vdev_parent->vdev_ops == &vdev_spare_ops && | |
2553 | vd->vdev_parent->vdev_child[0] == vd) | |
2554 | vd->vdev_unspare = B_TRUE; | |
2555 | ||
9babb374 BB |
2556 | if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) { |
2557 | ||
2558 | /* XXX - L2ARC 1.0 does not support expansion */ | |
2559 | if (vd->vdev_aux) | |
2560 | return (spa_vdev_state_exit(spa, vd, ENOTSUP)); | |
2561 | spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); | |
2562 | } | |
fb390aaf HR |
2563 | |
2564 | if (postevent) | |
2565 | spa_event_notify(spa, vd, ESC_ZFS_VDEV_ONLINE); | |
2566 | ||
fb5f0bc8 | 2567 | return (spa_vdev_state_exit(spa, vd, 0)); |
34dc7c2f BB |
2568 | } |
2569 | ||
428870ff BB |
2570 | static int |
2571 | vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags) | |
34dc7c2f | 2572 | { |
9babb374 | 2573 | vdev_t *vd, *tvd; |
428870ff BB |
2574 | int error = 0; |
2575 | uint64_t generation; | |
2576 | metaslab_group_t *mg; | |
34dc7c2f | 2577 | |
428870ff BB |
2578 | top: |
2579 | spa_vdev_state_enter(spa, SCL_ALLOC); | |
34dc7c2f | 2580 | |
b128c09f BB |
2581 | if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) |
2582 | return (spa_vdev_state_exit(spa, NULL, ENODEV)); | |
34dc7c2f BB |
2583 | |
2584 | if (!vd->vdev_ops->vdev_op_leaf) | |
b128c09f | 2585 | return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); |
34dc7c2f | 2586 | |
9babb374 | 2587 | tvd = vd->vdev_top; |
428870ff BB |
2588 | mg = tvd->vdev_mg; |
2589 | generation = spa->spa_config_generation + 1; | |
9babb374 | 2590 | |
34dc7c2f BB |
2591 | /* |
2592 | * If the device isn't already offline, try to offline it. | |
2593 | */ | |
2594 | if (!vd->vdev_offline) { | |
2595 | /* | |
fb5f0bc8 | 2596 | * If this device has the only valid copy of some data, |
9babb374 BB |
2597 | * don't allow it to be offlined. Log devices are always |
2598 | * expendable. | |
34dc7c2f | 2599 | */ |
9babb374 BB |
2600 | if (!tvd->vdev_islog && vd->vdev_aux == NULL && |
2601 | vdev_dtl_required(vd)) | |
b128c09f | 2602 | return (spa_vdev_state_exit(spa, NULL, EBUSY)); |
34dc7c2f | 2603 | |
428870ff BB |
2604 | /* |
2605 | * If the top-level is a slog and it has had allocations | |
2606 | * then proceed. We check that the vdev's metaslab group | |
2607 | * is not NULL since it's possible that we may have just | |
2608 | * added this vdev but not yet initialized its metaslabs. | |
2609 | */ | |
2610 | if (tvd->vdev_islog && mg != NULL) { | |
2611 | /* | |
2612 | * Prevent any future allocations. | |
2613 | */ | |
2614 | metaslab_group_passivate(mg); | |
2615 | (void) spa_vdev_state_exit(spa, vd, 0); | |
2616 | ||
2617 | error = spa_offline_log(spa); | |
2618 | ||
2619 | spa_vdev_state_enter(spa, SCL_ALLOC); | |
2620 | ||
2621 | /* | |
2622 | * Check to see if the config has changed. | |
2623 | */ | |
2624 | if (error || generation != spa->spa_config_generation) { | |
2625 | metaslab_group_activate(mg); | |
2626 | if (error) | |
2627 | return (spa_vdev_state_exit(spa, | |
2628 | vd, error)); | |
2629 | (void) spa_vdev_state_exit(spa, vd, 0); | |
2630 | goto top; | |
2631 | } | |
c99c9001 | 2632 | ASSERT0(tvd->vdev_stat.vs_alloc); |
428870ff BB |
2633 | } |
2634 | ||
34dc7c2f BB |
2635 | /* |
2636 | * Offline this device and reopen its top-level vdev. | |
9babb374 BB |
2637 | * If the top-level vdev is a log device then just offline |
2638 | * it. Otherwise, if this action results in the top-level | |
2639 | * vdev becoming unusable, undo it and fail the request. | |
34dc7c2f BB |
2640 | */ |
2641 | vd->vdev_offline = B_TRUE; | |
9babb374 BB |
2642 | vdev_reopen(tvd); |
2643 | ||
2644 | if (!tvd->vdev_islog && vd->vdev_aux == NULL && | |
2645 | vdev_is_dead(tvd)) { | |
34dc7c2f | 2646 | vd->vdev_offline = B_FALSE; |
9babb374 | 2647 | vdev_reopen(tvd); |
b128c09f | 2648 | return (spa_vdev_state_exit(spa, NULL, EBUSY)); |
34dc7c2f | 2649 | } |
428870ff BB |
2650 | |
2651 | /* | |
2652 | * Add the device back into the metaslab rotor so that | |
2653 | * once we online the device it's open for business. | |
2654 | */ | |
2655 | if (tvd->vdev_islog && mg != NULL) | |
2656 | metaslab_group_activate(mg); | |
34dc7c2f BB |
2657 | } |
2658 | ||
b128c09f | 2659 | vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY); |
34dc7c2f | 2660 | |
428870ff BB |
2661 | return (spa_vdev_state_exit(spa, vd, 0)); |
2662 | } | |
9babb374 | 2663 | |
428870ff BB |
2664 | int |
2665 | vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags) | |
2666 | { | |
2667 | int error; | |
9babb374 | 2668 | |
428870ff BB |
2669 | mutex_enter(&spa->spa_vdev_top_lock); |
2670 | error = vdev_offline_locked(spa, guid, flags); | |
2671 | mutex_exit(&spa->spa_vdev_top_lock); | |
2672 | ||
2673 | return (error); | |
34dc7c2f BB |
2674 | } |
2675 | ||
2676 | /* | |
2677 | * Clear the error counts associated with this vdev. Unlike vdev_online() and | |
2678 | * vdev_offline(), we assume the spa config is locked. We also clear all | |
2679 | * children. If 'vd' is NULL, then the user wants to clear all vdevs. | |
34dc7c2f BB |
2680 | */ |
2681 | void | |
b128c09f | 2682 | vdev_clear(spa_t *spa, vdev_t *vd) |
34dc7c2f | 2683 | { |
b128c09f | 2684 | vdev_t *rvd = spa->spa_root_vdev; |
d6320ddb | 2685 | int c; |
b128c09f BB |
2686 | |
2687 | ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
34dc7c2f BB |
2688 | |
2689 | if (vd == NULL) | |
b128c09f | 2690 | vd = rvd; |
34dc7c2f BB |
2691 | |
2692 | vd->vdev_stat.vs_read_errors = 0; | |
2693 | vd->vdev_stat.vs_write_errors = 0; | |
2694 | vd->vdev_stat.vs_checksum_errors = 0; | |
34dc7c2f | 2695 | |
d6320ddb | 2696 | for (c = 0; c < vd->vdev_children; c++) |
b128c09f | 2697 | vdev_clear(spa, vd->vdev_child[c]); |
34dc7c2f BB |
2698 | |
2699 | /* | |
b128c09f BB |
2700 | * If we're in the FAULTED state or have experienced failed I/O, then |
2701 | * clear the persistent state and attempt to reopen the device. We | |
2702 | * also mark the vdev config dirty, so that the new faulted state is | |
2703 | * written out to disk. | |
34dc7c2f | 2704 | */ |
b128c09f BB |
2705 | if (vd->vdev_faulted || vd->vdev_degraded || |
2706 | !vdev_readable(vd) || !vdev_writeable(vd)) { | |
2707 | ||
428870ff BB |
2708 | /* |
2709 | * When reopening in reponse to a clear event, it may be due to | |
2710 | * a fmadm repair request. In this case, if the device is | |
2711 | * still broken, we want to still post the ereport again. | |
2712 | */ | |
2713 | vd->vdev_forcefault = B_TRUE; | |
2714 | ||
572e2857 | 2715 | vd->vdev_faulted = vd->vdev_degraded = 0ULL; |
b128c09f BB |
2716 | vd->vdev_cant_read = B_FALSE; |
2717 | vd->vdev_cant_write = B_FALSE; | |
2718 | ||
572e2857 | 2719 | vdev_reopen(vd == rvd ? rvd : vd->vdev_top); |
34dc7c2f | 2720 | |
428870ff BB |
2721 | vd->vdev_forcefault = B_FALSE; |
2722 | ||
572e2857 | 2723 | if (vd != rvd && vdev_writeable(vd->vdev_top)) |
b128c09f BB |
2724 | vdev_state_dirty(vd->vdev_top); |
2725 | ||
2726 | if (vd->vdev_aux == NULL && !vdev_is_dead(vd)) | |
34dc7c2f BB |
2727 | spa_async_request(spa, SPA_ASYNC_RESILVER); |
2728 | ||
fb390aaf | 2729 | spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR); |
34dc7c2f | 2730 | } |
428870ff BB |
2731 | |
2732 | /* | |
2733 | * When clearing a FMA-diagnosed fault, we always want to | |
2734 | * unspare the device, as we assume that the original spare was | |
2735 | * done in response to the FMA fault. | |
2736 | */ | |
2737 | if (!vdev_is_dead(vd) && vd->vdev_parent != NULL && | |
2738 | vd->vdev_parent->vdev_ops == &vdev_spare_ops && | |
2739 | vd->vdev_parent->vdev_child[0] == vd) | |
2740 | vd->vdev_unspare = B_TRUE; | |
34dc7c2f BB |
2741 | } |
2742 | ||
b128c09f BB |
2743 | boolean_t |
2744 | vdev_is_dead(vdev_t *vd) | |
2745 | { | |
428870ff BB |
2746 | /* |
2747 | * Holes and missing devices are always considered "dead". | |
2748 | * This simplifies the code since we don't have to check for | |
2749 | * these types of devices in the various code paths. | |
2750 | * Instead we rely on the fact that we skip over dead devices | |
2751 | * before issuing I/O to them. | |
2752 | */ | |
2753 | return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole || | |
2754 | vd->vdev_ops == &vdev_missing_ops); | |
b128c09f BB |
2755 | } |
2756 | ||
2757 | boolean_t | |
34dc7c2f BB |
2758 | vdev_readable(vdev_t *vd) |
2759 | { | |
b128c09f | 2760 | return (!vdev_is_dead(vd) && !vd->vdev_cant_read); |
34dc7c2f BB |
2761 | } |
2762 | ||
b128c09f | 2763 | boolean_t |
34dc7c2f BB |
2764 | vdev_writeable(vdev_t *vd) |
2765 | { | |
b128c09f | 2766 | return (!vdev_is_dead(vd) && !vd->vdev_cant_write); |
34dc7c2f BB |
2767 | } |
2768 | ||
b128c09f BB |
2769 | boolean_t |
2770 | vdev_allocatable(vdev_t *vd) | |
34dc7c2f | 2771 | { |
fb5f0bc8 BB |
2772 | uint64_t state = vd->vdev_state; |
2773 | ||
b128c09f | 2774 | /* |
fb5f0bc8 | 2775 | * We currently allow allocations from vdevs which may be in the |
b128c09f BB |
2776 | * process of reopening (i.e. VDEV_STATE_CLOSED). If the device |
2777 | * fails to reopen then we'll catch it later when we're holding | |
fb5f0bc8 BB |
2778 | * the proper locks. Note that we have to get the vdev state |
2779 | * in a local variable because although it changes atomically, | |
2780 | * we're asking two separate questions about it. | |
b128c09f | 2781 | */ |
fb5f0bc8 | 2782 | return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) && |
3dfb57a3 DB |
2783 | !vd->vdev_cant_write && !vd->vdev_ishole && |
2784 | vd->vdev_mg->mg_initialized); | |
34dc7c2f BB |
2785 | } |
2786 | ||
b128c09f BB |
2787 | boolean_t |
2788 | vdev_accessible(vdev_t *vd, zio_t *zio) | |
34dc7c2f | 2789 | { |
b128c09f | 2790 | ASSERT(zio->io_vd == vd); |
34dc7c2f | 2791 | |
b128c09f BB |
2792 | if (vdev_is_dead(vd) || vd->vdev_remove_wanted) |
2793 | return (B_FALSE); | |
34dc7c2f | 2794 | |
b128c09f BB |
2795 | if (zio->io_type == ZIO_TYPE_READ) |
2796 | return (!vd->vdev_cant_read); | |
34dc7c2f | 2797 | |
b128c09f BB |
2798 | if (zio->io_type == ZIO_TYPE_WRITE) |
2799 | return (!vd->vdev_cant_write); | |
34dc7c2f | 2800 | |
b128c09f | 2801 | return (B_TRUE); |
34dc7c2f BB |
2802 | } |
2803 | ||
193a37cb TH |
2804 | static void |
2805 | vdev_get_child_stat(vdev_t *cvd, vdev_stat_t *vs, vdev_stat_t *cvs) | |
34dc7c2f | 2806 | { |
193a37cb TH |
2807 | int t; |
2808 | for (t = 0; t < ZIO_TYPES; t++) { | |
2809 | vs->vs_ops[t] += cvs->vs_ops[t]; | |
2810 | vs->vs_bytes[t] += cvs->vs_bytes[t]; | |
2811 | } | |
34dc7c2f | 2812 | |
193a37cb TH |
2813 | cvs->vs_scan_removing = cvd->vdev_removing; |
2814 | } | |
f3a7f661 | 2815 | |
193a37cb TH |
2816 | /* |
2817 | * Get extended stats | |
2818 | */ | |
2819 | static void | |
2820 | vdev_get_child_stat_ex(vdev_t *cvd, vdev_stat_ex_t *vsx, vdev_stat_ex_t *cvsx) | |
2821 | { | |
2822 | int t, b; | |
2823 | for (t = 0; t < ZIO_TYPES; t++) { | |
7e945072 | 2824 | for (b = 0; b < ARRAY_SIZE(vsx->vsx_disk_histo[0]); b++) |
193a37cb | 2825 | vsx->vsx_disk_histo[t][b] += cvsx->vsx_disk_histo[t][b]; |
7e945072 TH |
2826 | |
2827 | for (b = 0; b < ARRAY_SIZE(vsx->vsx_total_histo[0]); b++) { | |
193a37cb TH |
2828 | vsx->vsx_total_histo[t][b] += |
2829 | cvsx->vsx_total_histo[t][b]; | |
2830 | } | |
f38dfec3 | 2831 | } |
34dc7c2f | 2832 | |
193a37cb | 2833 | for (t = 0; t < ZIO_PRIORITY_NUM_QUEUEABLE; t++) { |
7e945072 | 2834 | for (b = 0; b < ARRAY_SIZE(vsx->vsx_queue_histo[0]); b++) { |
193a37cb TH |
2835 | vsx->vsx_queue_histo[t][b] += |
2836 | cvsx->vsx_queue_histo[t][b]; | |
2837 | } | |
2838 | vsx->vsx_active_queue[t] += cvsx->vsx_active_queue[t]; | |
2839 | vsx->vsx_pend_queue[t] += cvsx->vsx_pend_queue[t]; | |
7e945072 TH |
2840 | |
2841 | for (b = 0; b < ARRAY_SIZE(vsx->vsx_ind_histo[0]); b++) | |
2842 | vsx->vsx_ind_histo[t][b] += cvsx->vsx_ind_histo[t][b]; | |
2843 | ||
2844 | for (b = 0; b < ARRAY_SIZE(vsx->vsx_agg_histo[0]); b++) | |
2845 | vsx->vsx_agg_histo[t][b] += cvsx->vsx_agg_histo[t][b]; | |
193a37cb | 2846 | } |
7e945072 | 2847 | |
193a37cb TH |
2848 | } |
2849 | ||
2850 | /* | |
2851 | * Get statistics for the given vdev. | |
2852 | */ | |
2853 | static void | |
2854 | vdev_get_stats_ex_impl(vdev_t *vd, vdev_stat_t *vs, vdev_stat_ex_t *vsx) | |
2855 | { | |
2856 | int c, t; | |
34dc7c2f BB |
2857 | /* |
2858 | * If we're getting stats on the root vdev, aggregate the I/O counts | |
2859 | * over all top-level vdevs (i.e. the direct children of the root). | |
2860 | */ | |
193a37cb TH |
2861 | if (!vd->vdev_ops->vdev_op_leaf) { |
2862 | if (vs) { | |
2863 | memset(vs->vs_ops, 0, sizeof (vs->vs_ops)); | |
2864 | memset(vs->vs_bytes, 0, sizeof (vs->vs_bytes)); | |
2865 | } | |
2866 | if (vsx) | |
2867 | memset(vsx, 0, sizeof (*vsx)); | |
2868 | ||
2869 | for (c = 0; c < vd->vdev_children; c++) { | |
2870 | vdev_t *cvd = vd->vdev_child[c]; | |
34dc7c2f | 2871 | vdev_stat_t *cvs = &cvd->vdev_stat; |
193a37cb TH |
2872 | vdev_stat_ex_t *cvsx = &cvd->vdev_stat_ex; |
2873 | ||
2874 | vdev_get_stats_ex_impl(cvd, cvs, cvsx); | |
2875 | if (vs) | |
2876 | vdev_get_child_stat(cvd, vs, cvs); | |
2877 | if (vsx) | |
2878 | vdev_get_child_stat_ex(cvd, vsx, cvsx); | |
34dc7c2f | 2879 | |
193a37cb TH |
2880 | } |
2881 | } else { | |
2882 | /* | |
2883 | * We're a leaf. Just copy our ZIO active queue stats in. The | |
2884 | * other leaf stats are updated in vdev_stat_update(). | |
2885 | */ | |
2886 | if (!vsx) | |
2887 | return; | |
2888 | ||
2889 | memcpy(vsx, &vd->vdev_stat_ex, sizeof (vd->vdev_stat_ex)); | |
2890 | ||
2891 | for (t = 0; t < ARRAY_SIZE(vd->vdev_queue.vq_class); t++) { | |
2892 | vsx->vsx_active_queue[t] = | |
2893 | vd->vdev_queue.vq_class[t].vqc_active; | |
2894 | vsx->vsx_pend_queue[t] = avl_numnodes( | |
2895 | &vd->vdev_queue.vq_class[t].vqc_queued_tree); | |
34dc7c2f BB |
2896 | } |
2897 | } | |
193a37cb TH |
2898 | } |
2899 | ||
2900 | void | |
2901 | vdev_get_stats_ex(vdev_t *vd, vdev_stat_t *vs, vdev_stat_ex_t *vsx) | |
2902 | { | |
2903 | mutex_enter(&vd->vdev_stat_lock); | |
2904 | if (vs) { | |
2905 | bcopy(&vd->vdev_stat, vs, sizeof (*vs)); | |
2906 | vs->vs_timestamp = gethrtime() - vs->vs_timestamp; | |
2907 | vs->vs_state = vd->vdev_state; | |
2908 | vs->vs_rsize = vdev_get_min_asize(vd); | |
2909 | if (vd->vdev_ops->vdev_op_leaf) | |
2910 | vs->vs_rsize += VDEV_LABEL_START_SIZE + | |
2911 | VDEV_LABEL_END_SIZE; | |
2912 | vs->vs_esize = vd->vdev_max_asize - vd->vdev_asize; | |
2913 | if (vd->vdev_aux == NULL && vd == vd->vdev_top && | |
2914 | !vd->vdev_ishole) { | |
2915 | vs->vs_fragmentation = vd->vdev_mg->mg_fragmentation; | |
2916 | } | |
2917 | } | |
2918 | ||
2919 | ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_READER) != 0); | |
2920 | vdev_get_stats_ex_impl(vd, vs, vsx); | |
f3a7f661 | 2921 | mutex_exit(&vd->vdev_stat_lock); |
34dc7c2f BB |
2922 | } |
2923 | ||
193a37cb TH |
2924 | void |
2925 | vdev_get_stats(vdev_t *vd, vdev_stat_t *vs) | |
2926 | { | |
2927 | return (vdev_get_stats_ex(vd, vs, NULL)); | |
2928 | } | |
2929 | ||
34dc7c2f BB |
2930 | void |
2931 | vdev_clear_stats(vdev_t *vd) | |
2932 | { | |
2933 | mutex_enter(&vd->vdev_stat_lock); | |
2934 | vd->vdev_stat.vs_space = 0; | |
2935 | vd->vdev_stat.vs_dspace = 0; | |
2936 | vd->vdev_stat.vs_alloc = 0; | |
2937 | mutex_exit(&vd->vdev_stat_lock); | |
2938 | } | |
2939 | ||
428870ff BB |
2940 | void |
2941 | vdev_scan_stat_init(vdev_t *vd) | |
2942 | { | |
2943 | vdev_stat_t *vs = &vd->vdev_stat; | |
d6320ddb | 2944 | int c; |
428870ff | 2945 | |
d6320ddb | 2946 | for (c = 0; c < vd->vdev_children; c++) |
428870ff BB |
2947 | vdev_scan_stat_init(vd->vdev_child[c]); |
2948 | ||
2949 | mutex_enter(&vd->vdev_stat_lock); | |
2950 | vs->vs_scan_processed = 0; | |
2951 | mutex_exit(&vd->vdev_stat_lock); | |
2952 | } | |
2953 | ||
34dc7c2f | 2954 | void |
b128c09f | 2955 | vdev_stat_update(zio_t *zio, uint64_t psize) |
34dc7c2f | 2956 | { |
fb5f0bc8 BB |
2957 | spa_t *spa = zio->io_spa; |
2958 | vdev_t *rvd = spa->spa_root_vdev; | |
b128c09f | 2959 | vdev_t *vd = zio->io_vd ? zio->io_vd : rvd; |
34dc7c2f BB |
2960 | vdev_t *pvd; |
2961 | uint64_t txg = zio->io_txg; | |
2962 | vdev_stat_t *vs = &vd->vdev_stat; | |
193a37cb | 2963 | vdev_stat_ex_t *vsx = &vd->vdev_stat_ex; |
34dc7c2f BB |
2964 | zio_type_t type = zio->io_type; |
2965 | int flags = zio->io_flags; | |
2966 | ||
b128c09f BB |
2967 | /* |
2968 | * If this i/o is a gang leader, it didn't do any actual work. | |
2969 | */ | |
2970 | if (zio->io_gang_tree) | |
2971 | return; | |
2972 | ||
34dc7c2f | 2973 | if (zio->io_error == 0) { |
b128c09f BB |
2974 | /* |
2975 | * If this is a root i/o, don't count it -- we've already | |
2976 | * counted the top-level vdevs, and vdev_get_stats() will | |
2977 | * aggregate them when asked. This reduces contention on | |
2978 | * the root vdev_stat_lock and implicitly handles blocks | |
2979 | * that compress away to holes, for which there is no i/o. | |
2980 | * (Holes never create vdev children, so all the counters | |
2981 | * remain zero, which is what we want.) | |
2982 | * | |
2983 | * Note: this only applies to successful i/o (io_error == 0) | |
2984 | * because unlike i/o counts, errors are not additive. | |
2985 | * When reading a ditto block, for example, failure of | |
2986 | * one top-level vdev does not imply a root-level error. | |
2987 | */ | |
2988 | if (vd == rvd) | |
2989 | return; | |
2990 | ||
2991 | ASSERT(vd == zio->io_vd); | |
fb5f0bc8 BB |
2992 | |
2993 | if (flags & ZIO_FLAG_IO_BYPASS) | |
2994 | return; | |
2995 | ||
2996 | mutex_enter(&vd->vdev_stat_lock); | |
2997 | ||
b128c09f | 2998 | if (flags & ZIO_FLAG_IO_REPAIR) { |
572e2857 | 2999 | if (flags & ZIO_FLAG_SCAN_THREAD) { |
428870ff BB |
3000 | dsl_scan_phys_t *scn_phys = |
3001 | &spa->spa_dsl_pool->dp_scan->scn_phys; | |
3002 | uint64_t *processed = &scn_phys->scn_processed; | |
3003 | ||
3004 | /* XXX cleanup? */ | |
3005 | if (vd->vdev_ops->vdev_op_leaf) | |
3006 | atomic_add_64(processed, psize); | |
3007 | vs->vs_scan_processed += psize; | |
3008 | } | |
3009 | ||
fb5f0bc8 | 3010 | if (flags & ZIO_FLAG_SELF_HEAL) |
b128c09f | 3011 | vs->vs_self_healed += psize; |
34dc7c2f | 3012 | } |
fb5f0bc8 | 3013 | |
193a37cb TH |
3014 | /* |
3015 | * The bytes/ops/histograms are recorded at the leaf level and | |
3016 | * aggregated into the higher level vdevs in vdev_get_stats(). | |
3017 | */ | |
4eb0db42 TH |
3018 | if (vd->vdev_ops->vdev_op_leaf && |
3019 | (zio->io_priority < ZIO_PRIORITY_NUM_QUEUEABLE)) { | |
193a37cb TH |
3020 | |
3021 | vs->vs_ops[type]++; | |
3022 | vs->vs_bytes[type] += psize; | |
3023 | ||
7e945072 TH |
3024 | if (flags & ZIO_FLAG_DELEGATED) { |
3025 | vsx->vsx_agg_histo[zio->io_priority] | |
3026 | [RQ_HISTO(zio->io_size)]++; | |
3027 | } else { | |
3028 | vsx->vsx_ind_histo[zio->io_priority] | |
3029 | [RQ_HISTO(zio->io_size)]++; | |
3030 | } | |
3031 | ||
193a37cb TH |
3032 | if (zio->io_delta && zio->io_delay) { |
3033 | vsx->vsx_queue_histo[zio->io_priority] | |
7e945072 | 3034 | [L_HISTO(zio->io_delta - zio->io_delay)]++; |
193a37cb | 3035 | vsx->vsx_disk_histo[type] |
7e945072 | 3036 | [L_HISTO(zio->io_delay)]++; |
193a37cb | 3037 | vsx->vsx_total_histo[type] |
7e945072 | 3038 | [L_HISTO(zio->io_delta)]++; |
193a37cb TH |
3039 | } |
3040 | } | |
fb5f0bc8 BB |
3041 | |
3042 | mutex_exit(&vd->vdev_stat_lock); | |
34dc7c2f BB |
3043 | return; |
3044 | } | |
3045 | ||
3046 | if (flags & ZIO_FLAG_SPECULATIVE) | |
3047 | return; | |
3048 | ||
9babb374 BB |
3049 | /* |
3050 | * If this is an I/O error that is going to be retried, then ignore the | |
3051 | * error. Otherwise, the user may interpret B_FAILFAST I/O errors as | |
3052 | * hard errors, when in reality they can happen for any number of | |
3053 | * innocuous reasons (bus resets, MPxIO link failure, etc). | |
3054 | */ | |
3055 | if (zio->io_error == EIO && | |
3056 | !(zio->io_flags & ZIO_FLAG_IO_RETRY)) | |
3057 | return; | |
3058 | ||
428870ff BB |
3059 | /* |
3060 | * Intent logs writes won't propagate their error to the root | |
3061 | * I/O so don't mark these types of failures as pool-level | |
3062 | * errors. | |
3063 | */ | |
3064 | if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE)) | |
3065 | return; | |
3066 | ||
b128c09f | 3067 | mutex_enter(&vd->vdev_stat_lock); |
9babb374 | 3068 | if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) { |
b128c09f BB |
3069 | if (zio->io_error == ECKSUM) |
3070 | vs->vs_checksum_errors++; | |
3071 | else | |
3072 | vs->vs_read_errors++; | |
34dc7c2f | 3073 | } |
9babb374 | 3074 | if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd)) |
b128c09f BB |
3075 | vs->vs_write_errors++; |
3076 | mutex_exit(&vd->vdev_stat_lock); | |
34dc7c2f | 3077 | |
fb5f0bc8 BB |
3078 | if (type == ZIO_TYPE_WRITE && txg != 0 && |
3079 | (!(flags & ZIO_FLAG_IO_REPAIR) || | |
572e2857 | 3080 | (flags & ZIO_FLAG_SCAN_THREAD) || |
428870ff | 3081 | spa->spa_claiming)) { |
fb5f0bc8 | 3082 | /* |
428870ff BB |
3083 | * This is either a normal write (not a repair), or it's |
3084 | * a repair induced by the scrub thread, or it's a repair | |
3085 | * made by zil_claim() during spa_load() in the first txg. | |
3086 | * In the normal case, we commit the DTL change in the same | |
3087 | * txg as the block was born. In the scrub-induced repair | |
3088 | * case, we know that scrubs run in first-pass syncing context, | |
3089 | * so we commit the DTL change in spa_syncing_txg(spa). | |
3090 | * In the zil_claim() case, we commit in spa_first_txg(spa). | |
fb5f0bc8 BB |
3091 | * |
3092 | * We currently do not make DTL entries for failed spontaneous | |
3093 | * self-healing writes triggered by normal (non-scrubbing) | |
3094 | * reads, because we have no transactional context in which to | |
3095 | * do so -- and it's not clear that it'd be desirable anyway. | |
3096 | */ | |
3097 | if (vd->vdev_ops->vdev_op_leaf) { | |
3098 | uint64_t commit_txg = txg; | |
572e2857 | 3099 | if (flags & ZIO_FLAG_SCAN_THREAD) { |
fb5f0bc8 BB |
3100 | ASSERT(flags & ZIO_FLAG_IO_REPAIR); |
3101 | ASSERT(spa_sync_pass(spa) == 1); | |
3102 | vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1); | |
428870ff BB |
3103 | commit_txg = spa_syncing_txg(spa); |
3104 | } else if (spa->spa_claiming) { | |
3105 | ASSERT(flags & ZIO_FLAG_IO_REPAIR); | |
3106 | commit_txg = spa_first_txg(spa); | |
fb5f0bc8 | 3107 | } |
428870ff | 3108 | ASSERT(commit_txg >= spa_syncing_txg(spa)); |
fb5f0bc8 | 3109 | if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1)) |
34dc7c2f | 3110 | return; |
fb5f0bc8 BB |
3111 | for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) |
3112 | vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1); | |
3113 | vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg); | |
34dc7c2f | 3114 | } |
fb5f0bc8 BB |
3115 | if (vd != rvd) |
3116 | vdev_dtl_dirty(vd, DTL_MISSING, txg, 1); | |
34dc7c2f BB |
3117 | } |
3118 | } | |
3119 | ||
34dc7c2f | 3120 | /* |
428870ff BB |
3121 | * Update the in-core space usage stats for this vdev, its metaslab class, |
3122 | * and the root vdev. | |
34dc7c2f BB |
3123 | */ |
3124 | void | |
428870ff BB |
3125 | vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta, |
3126 | int64_t space_delta) | |
34dc7c2f BB |
3127 | { |
3128 | int64_t dspace_delta = space_delta; | |
3129 | spa_t *spa = vd->vdev_spa; | |
3130 | vdev_t *rvd = spa->spa_root_vdev; | |
428870ff BB |
3131 | metaslab_group_t *mg = vd->vdev_mg; |
3132 | metaslab_class_t *mc = mg ? mg->mg_class : NULL; | |
34dc7c2f BB |
3133 | |
3134 | ASSERT(vd == vd->vdev_top); | |
3135 | ||
3136 | /* | |
3137 | * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion | |
3138 | * factor. We must calculate this here and not at the root vdev | |
3139 | * because the root vdev's psize-to-asize is simply the max of its | |
3140 | * childrens', thus not accurate enough for us. | |
3141 | */ | |
3142 | ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0); | |
9babb374 | 3143 | ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache); |
34dc7c2f BB |
3144 | dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) * |
3145 | vd->vdev_deflate_ratio; | |
3146 | ||
3147 | mutex_enter(&vd->vdev_stat_lock); | |
34dc7c2f | 3148 | vd->vdev_stat.vs_alloc += alloc_delta; |
428870ff | 3149 | vd->vdev_stat.vs_space += space_delta; |
34dc7c2f BB |
3150 | vd->vdev_stat.vs_dspace += dspace_delta; |
3151 | mutex_exit(&vd->vdev_stat_lock); | |
3152 | ||
428870ff | 3153 | if (mc == spa_normal_class(spa)) { |
34dc7c2f | 3154 | mutex_enter(&rvd->vdev_stat_lock); |
34dc7c2f | 3155 | rvd->vdev_stat.vs_alloc += alloc_delta; |
428870ff | 3156 | rvd->vdev_stat.vs_space += space_delta; |
34dc7c2f BB |
3157 | rvd->vdev_stat.vs_dspace += dspace_delta; |
3158 | mutex_exit(&rvd->vdev_stat_lock); | |
3159 | } | |
428870ff BB |
3160 | |
3161 | if (mc != NULL) { | |
3162 | ASSERT(rvd == vd->vdev_parent); | |
3163 | ASSERT(vd->vdev_ms_count != 0); | |
3164 | ||
3165 | metaslab_class_space_update(mc, | |
3166 | alloc_delta, defer_delta, space_delta, dspace_delta); | |
3167 | } | |
34dc7c2f BB |
3168 | } |
3169 | ||
3170 | /* | |
3171 | * Mark a top-level vdev's config as dirty, placing it on the dirty list | |
3172 | * so that it will be written out next time the vdev configuration is synced. | |
3173 | * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs. | |
3174 | */ | |
3175 | void | |
3176 | vdev_config_dirty(vdev_t *vd) | |
3177 | { | |
3178 | spa_t *spa = vd->vdev_spa; | |
3179 | vdev_t *rvd = spa->spa_root_vdev; | |
3180 | int c; | |
3181 | ||
572e2857 BB |
3182 | ASSERT(spa_writeable(spa)); |
3183 | ||
34dc7c2f | 3184 | /* |
9babb374 BB |
3185 | * If this is an aux vdev (as with l2cache and spare devices), then we |
3186 | * update the vdev config manually and set the sync flag. | |
b128c09f BB |
3187 | */ |
3188 | if (vd->vdev_aux != NULL) { | |
3189 | spa_aux_vdev_t *sav = vd->vdev_aux; | |
3190 | nvlist_t **aux; | |
3191 | uint_t naux; | |
3192 | ||
3193 | for (c = 0; c < sav->sav_count; c++) { | |
3194 | if (sav->sav_vdevs[c] == vd) | |
3195 | break; | |
3196 | } | |
3197 | ||
3198 | if (c == sav->sav_count) { | |
3199 | /* | |
3200 | * We're being removed. There's nothing more to do. | |
3201 | */ | |
3202 | ASSERT(sav->sav_sync == B_TRUE); | |
3203 | return; | |
3204 | } | |
3205 | ||
3206 | sav->sav_sync = B_TRUE; | |
3207 | ||
9babb374 BB |
3208 | if (nvlist_lookup_nvlist_array(sav->sav_config, |
3209 | ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) { | |
3210 | VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, | |
3211 | ZPOOL_CONFIG_SPARES, &aux, &naux) == 0); | |
3212 | } | |
b128c09f BB |
3213 | |
3214 | ASSERT(c < naux); | |
3215 | ||
3216 | /* | |
3217 | * Setting the nvlist in the middle if the array is a little | |
3218 | * sketchy, but it will work. | |
3219 | */ | |
3220 | nvlist_free(aux[c]); | |
428870ff | 3221 | aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0); |
b128c09f BB |
3222 | |
3223 | return; | |
3224 | } | |
3225 | ||
3226 | /* | |
3227 | * The dirty list is protected by the SCL_CONFIG lock. The caller | |
3228 | * must either hold SCL_CONFIG as writer, or must be the sync thread | |
3229 | * (which holds SCL_CONFIG as reader). There's only one sync thread, | |
34dc7c2f BB |
3230 | * so this is sufficient to ensure mutual exclusion. |
3231 | */ | |
b128c09f BB |
3232 | ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || |
3233 | (dsl_pool_sync_context(spa_get_dsl(spa)) && | |
3234 | spa_config_held(spa, SCL_CONFIG, RW_READER))); | |
34dc7c2f BB |
3235 | |
3236 | if (vd == rvd) { | |
3237 | for (c = 0; c < rvd->vdev_children; c++) | |
3238 | vdev_config_dirty(rvd->vdev_child[c]); | |
3239 | } else { | |
3240 | ASSERT(vd == vd->vdev_top); | |
3241 | ||
428870ff BB |
3242 | if (!list_link_active(&vd->vdev_config_dirty_node) && |
3243 | !vd->vdev_ishole) | |
b128c09f | 3244 | list_insert_head(&spa->spa_config_dirty_list, vd); |
34dc7c2f BB |
3245 | } |
3246 | } | |
3247 | ||
3248 | void | |
3249 | vdev_config_clean(vdev_t *vd) | |
3250 | { | |
3251 | spa_t *spa = vd->vdev_spa; | |
3252 | ||
b128c09f BB |
3253 | ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || |
3254 | (dsl_pool_sync_context(spa_get_dsl(spa)) && | |
3255 | spa_config_held(spa, SCL_CONFIG, RW_READER))); | |
34dc7c2f | 3256 | |
b128c09f BB |
3257 | ASSERT(list_link_active(&vd->vdev_config_dirty_node)); |
3258 | list_remove(&spa->spa_config_dirty_list, vd); | |
34dc7c2f BB |
3259 | } |
3260 | ||
b128c09f BB |
3261 | /* |
3262 | * Mark a top-level vdev's state as dirty, so that the next pass of | |
3263 | * spa_sync() can convert this into vdev_config_dirty(). We distinguish | |
3264 | * the state changes from larger config changes because they require | |
3265 | * much less locking, and are often needed for administrative actions. | |
3266 | */ | |
3267 | void | |
3268 | vdev_state_dirty(vdev_t *vd) | |
3269 | { | |
3270 | spa_t *spa = vd->vdev_spa; | |
3271 | ||
572e2857 | 3272 | ASSERT(spa_writeable(spa)); |
b128c09f BB |
3273 | ASSERT(vd == vd->vdev_top); |
3274 | ||
3275 | /* | |
3276 | * The state list is protected by the SCL_STATE lock. The caller | |
3277 | * must either hold SCL_STATE as writer, or must be the sync thread | |
3278 | * (which holds SCL_STATE as reader). There's only one sync thread, | |
3279 | * so this is sufficient to ensure mutual exclusion. | |
3280 | */ | |
3281 | ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || | |
3282 | (dsl_pool_sync_context(spa_get_dsl(spa)) && | |
3283 | spa_config_held(spa, SCL_STATE, RW_READER))); | |
3284 | ||
428870ff | 3285 | if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole) |
b128c09f BB |
3286 | list_insert_head(&spa->spa_state_dirty_list, vd); |
3287 | } | |
3288 | ||
3289 | void | |
3290 | vdev_state_clean(vdev_t *vd) | |
3291 | { | |
3292 | spa_t *spa = vd->vdev_spa; | |
3293 | ||
3294 | ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || | |
3295 | (dsl_pool_sync_context(spa_get_dsl(spa)) && | |
3296 | spa_config_held(spa, SCL_STATE, RW_READER))); | |
3297 | ||
3298 | ASSERT(list_link_active(&vd->vdev_state_dirty_node)); | |
3299 | list_remove(&spa->spa_state_dirty_list, vd); | |
3300 | } | |
3301 | ||
3302 | /* | |
3303 | * Propagate vdev state up from children to parent. | |
3304 | */ | |
34dc7c2f BB |
3305 | void |
3306 | vdev_propagate_state(vdev_t *vd) | |
3307 | { | |
fb5f0bc8 BB |
3308 | spa_t *spa = vd->vdev_spa; |
3309 | vdev_t *rvd = spa->spa_root_vdev; | |
34dc7c2f BB |
3310 | int degraded = 0, faulted = 0; |
3311 | int corrupted = 0; | |
34dc7c2f | 3312 | vdev_t *child; |
d6320ddb | 3313 | int c; |
34dc7c2f BB |
3314 | |
3315 | if (vd->vdev_children > 0) { | |
d6320ddb | 3316 | for (c = 0; c < vd->vdev_children; c++) { |
34dc7c2f | 3317 | child = vd->vdev_child[c]; |
b128c09f | 3318 | |
428870ff BB |
3319 | /* |
3320 | * Don't factor holes into the decision. | |
3321 | */ | |
3322 | if (child->vdev_ishole) | |
3323 | continue; | |
3324 | ||
b128c09f | 3325 | if (!vdev_readable(child) || |
fb5f0bc8 | 3326 | (!vdev_writeable(child) && spa_writeable(spa))) { |
b128c09f BB |
3327 | /* |
3328 | * Root special: if there is a top-level log | |
3329 | * device, treat the root vdev as if it were | |
3330 | * degraded. | |
3331 | */ | |
3332 | if (child->vdev_islog && vd == rvd) | |
3333 | degraded++; | |
3334 | else | |
3335 | faulted++; | |
3336 | } else if (child->vdev_state <= VDEV_STATE_DEGRADED) { | |
34dc7c2f | 3337 | degraded++; |
b128c09f | 3338 | } |
34dc7c2f BB |
3339 | |
3340 | if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA) | |
3341 | corrupted++; | |
3342 | } | |
3343 | ||
3344 | vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded); | |
3345 | ||
3346 | /* | |
b128c09f | 3347 | * Root special: if there is a top-level vdev that cannot be |
34dc7c2f BB |
3348 | * opened due to corrupted metadata, then propagate the root |
3349 | * vdev's aux state as 'corrupt' rather than 'insufficient | |
3350 | * replicas'. | |
3351 | */ | |
3352 | if (corrupted && vd == rvd && | |
3353 | rvd->vdev_state == VDEV_STATE_CANT_OPEN) | |
3354 | vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN, | |
3355 | VDEV_AUX_CORRUPT_DATA); | |
3356 | } | |
3357 | ||
b128c09f | 3358 | if (vd->vdev_parent) |
34dc7c2f BB |
3359 | vdev_propagate_state(vd->vdev_parent); |
3360 | } | |
3361 | ||
3362 | /* | |
3363 | * Set a vdev's state. If this is during an open, we don't update the parent | |
3364 | * state, because we're in the process of opening children depth-first. | |
3365 | * Otherwise, we propagate the change to the parent. | |
3366 | * | |
3367 | * If this routine places a device in a faulted state, an appropriate ereport is | |
3368 | * generated. | |
3369 | */ | |
3370 | void | |
3371 | vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux) | |
3372 | { | |
3373 | uint64_t save_state; | |
b128c09f | 3374 | spa_t *spa = vd->vdev_spa; |
34dc7c2f BB |
3375 | |
3376 | if (state == vd->vdev_state) { | |
976246fa DB |
3377 | /* |
3378 | * Since vdev_offline() code path is already in an offline | |
3379 | * state we can miss a statechange event to OFFLINE. Check | |
3380 | * the previous state to catch this condition. | |
3381 | */ | |
3382 | if (vd->vdev_ops->vdev_op_leaf && | |
3383 | (state == VDEV_STATE_OFFLINE) && | |
3384 | (vd->vdev_prevstate >= VDEV_STATE_FAULTED)) { | |
3385 | /* post an offline state change */ | |
3386 | zfs_post_state_change(spa, vd, vd->vdev_prevstate); | |
3387 | } | |
34dc7c2f BB |
3388 | vd->vdev_stat.vs_aux = aux; |
3389 | return; | |
3390 | } | |
3391 | ||
3392 | save_state = vd->vdev_state; | |
3393 | ||
3394 | vd->vdev_state = state; | |
3395 | vd->vdev_stat.vs_aux = aux; | |
3396 | ||
3397 | /* | |
3398 | * If we are setting the vdev state to anything but an open state, then | |
428870ff BB |
3399 | * always close the underlying device unless the device has requested |
3400 | * a delayed close (i.e. we're about to remove or fault the device). | |
3401 | * Otherwise, we keep accessible but invalid devices open forever. | |
3402 | * We don't call vdev_close() itself, because that implies some extra | |
3403 | * checks (offline, etc) that we don't want here. This is limited to | |
3404 | * leaf devices, because otherwise closing the device will affect other | |
3405 | * children. | |
34dc7c2f | 3406 | */ |
428870ff BB |
3407 | if (!vd->vdev_delayed_close && vdev_is_dead(vd) && |
3408 | vd->vdev_ops->vdev_op_leaf) | |
34dc7c2f BB |
3409 | vd->vdev_ops->vdev_op_close(vd); |
3410 | ||
3411 | if (vd->vdev_removed && | |
3412 | state == VDEV_STATE_CANT_OPEN && | |
3413 | (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) { | |
3414 | /* | |
3415 | * If the previous state is set to VDEV_STATE_REMOVED, then this | |
3416 | * device was previously marked removed and someone attempted to | |
3417 | * reopen it. If this failed due to a nonexistent device, then | |
3418 | * keep the device in the REMOVED state. We also let this be if | |
3419 | * it is one of our special test online cases, which is only | |
3420 | * attempting to online the device and shouldn't generate an FMA | |
3421 | * fault. | |
3422 | */ | |
3423 | vd->vdev_state = VDEV_STATE_REMOVED; | |
3424 | vd->vdev_stat.vs_aux = VDEV_AUX_NONE; | |
3425 | } else if (state == VDEV_STATE_REMOVED) { | |
34dc7c2f BB |
3426 | vd->vdev_removed = B_TRUE; |
3427 | } else if (state == VDEV_STATE_CANT_OPEN) { | |
3428 | /* | |
572e2857 BB |
3429 | * If we fail to open a vdev during an import or recovery, we |
3430 | * mark it as "not available", which signifies that it was | |
3431 | * never there to begin with. Failure to open such a device | |
3432 | * is not considered an error. | |
34dc7c2f | 3433 | */ |
572e2857 BB |
3434 | if ((spa_load_state(spa) == SPA_LOAD_IMPORT || |
3435 | spa_load_state(spa) == SPA_LOAD_RECOVER) && | |
34dc7c2f BB |
3436 | vd->vdev_ops->vdev_op_leaf) |
3437 | vd->vdev_not_present = 1; | |
3438 | ||
3439 | /* | |
3440 | * Post the appropriate ereport. If the 'prevstate' field is | |
3441 | * set to something other than VDEV_STATE_UNKNOWN, it indicates | |
3442 | * that this is part of a vdev_reopen(). In this case, we don't | |
3443 | * want to post the ereport if the device was already in the | |
3444 | * CANT_OPEN state beforehand. | |
3445 | * | |
3446 | * If the 'checkremove' flag is set, then this is an attempt to | |
3447 | * online the device in response to an insertion event. If we | |
3448 | * hit this case, then we have detected an insertion event for a | |
3449 | * faulted or offline device that wasn't in the removed state. | |
3450 | * In this scenario, we don't post an ereport because we are | |
3451 | * about to replace the device, or attempt an online with | |
3452 | * vdev_forcefault, which will generate the fault for us. | |
3453 | */ | |
3454 | if ((vd->vdev_prevstate != state || vd->vdev_forcefault) && | |
3455 | !vd->vdev_not_present && !vd->vdev_checkremove && | |
b128c09f | 3456 | vd != spa->spa_root_vdev) { |
34dc7c2f BB |
3457 | const char *class; |
3458 | ||
3459 | switch (aux) { | |
3460 | case VDEV_AUX_OPEN_FAILED: | |
3461 | class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED; | |
3462 | break; | |
3463 | case VDEV_AUX_CORRUPT_DATA: | |
3464 | class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA; | |
3465 | break; | |
3466 | case VDEV_AUX_NO_REPLICAS: | |
3467 | class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS; | |
3468 | break; | |
3469 | case VDEV_AUX_BAD_GUID_SUM: | |
3470 | class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM; | |
3471 | break; | |
3472 | case VDEV_AUX_TOO_SMALL: | |
3473 | class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL; | |
3474 | break; | |
3475 | case VDEV_AUX_BAD_LABEL: | |
3476 | class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL; | |
3477 | break; | |
3478 | default: | |
3479 | class = FM_EREPORT_ZFS_DEVICE_UNKNOWN; | |
3480 | } | |
3481 | ||
b128c09f | 3482 | zfs_ereport_post(class, spa, vd, NULL, save_state, 0); |
34dc7c2f BB |
3483 | } |
3484 | ||
3485 | /* Erase any notion of persistent removed state */ | |
3486 | vd->vdev_removed = B_FALSE; | |
3487 | } else { | |
3488 | vd->vdev_removed = B_FALSE; | |
3489 | } | |
3490 | ||
d02ca379 DB |
3491 | /* |
3492 | * Notify ZED of any significant state-change on a leaf vdev. | |
3493 | * | |
d02ca379 | 3494 | */ |
6078881a TH |
3495 | if (vd->vdev_ops->vdev_op_leaf) { |
3496 | /* preserve original state from a vdev_reopen() */ | |
3497 | if ((vd->vdev_prevstate != VDEV_STATE_UNKNOWN) && | |
3498 | (vd->vdev_prevstate != vd->vdev_state) && | |
3499 | (save_state <= VDEV_STATE_CLOSED)) | |
3500 | save_state = vd->vdev_prevstate; | |
3501 | ||
3502 | /* filter out state change due to initial vdev_open */ | |
3503 | if (save_state > VDEV_STATE_CLOSED) | |
3504 | zfs_post_state_change(spa, vd, save_state); | |
d02ca379 DB |
3505 | } |
3506 | ||
9babb374 BB |
3507 | if (!isopen && vd->vdev_parent) |
3508 | vdev_propagate_state(vd->vdev_parent); | |
34dc7c2f | 3509 | } |
b128c09f BB |
3510 | |
3511 | /* | |
3512 | * Check the vdev configuration to ensure that it's capable of supporting | |
57a4eddc | 3513 | * a root pool. |
b128c09f BB |
3514 | */ |
3515 | boolean_t | |
3516 | vdev_is_bootable(vdev_t *vd) | |
3517 | { | |
57a4eddc RL |
3518 | #if defined(__sun__) || defined(__sun) |
3519 | /* | |
3520 | * Currently, we do not support RAID-Z or partial configuration. | |
3521 | * In addition, only a single top-level vdev is allowed and none of the | |
3522 | * leaves can be wholedisks. | |
3523 | */ | |
d6320ddb BB |
3524 | int c; |
3525 | ||
b128c09f BB |
3526 | if (!vd->vdev_ops->vdev_op_leaf) { |
3527 | char *vdev_type = vd->vdev_ops->vdev_op_type; | |
3528 | ||
3529 | if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 && | |
3530 | vd->vdev_children > 1) { | |
3531 | return (B_FALSE); | |
3532 | } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 || | |
3533 | strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) { | |
3534 | return (B_FALSE); | |
3535 | } | |
3536 | } else if (vd->vdev_wholedisk == 1) { | |
3537 | return (B_FALSE); | |
3538 | } | |
3539 | ||
d6320ddb | 3540 | for (c = 0; c < vd->vdev_children; c++) { |
b128c09f BB |
3541 | if (!vdev_is_bootable(vd->vdev_child[c])) |
3542 | return (B_FALSE); | |
3543 | } | |
57a4eddc | 3544 | #endif /* __sun__ || __sun */ |
b128c09f BB |
3545 | return (B_TRUE); |
3546 | } | |
9babb374 | 3547 | |
428870ff BB |
3548 | /* |
3549 | * Load the state from the original vdev tree (ovd) which | |
3550 | * we've retrieved from the MOS config object. If the original | |
572e2857 BB |
3551 | * vdev was offline or faulted then we transfer that state to the |
3552 | * device in the current vdev tree (nvd). | |
428870ff | 3553 | */ |
9babb374 | 3554 | void |
428870ff | 3555 | vdev_load_log_state(vdev_t *nvd, vdev_t *ovd) |
9babb374 | 3556 | { |
d6320ddb | 3557 | int c; |
9babb374 | 3558 | |
572e2857 | 3559 | ASSERT(nvd->vdev_top->vdev_islog); |
1fde1e37 BB |
3560 | ASSERT(spa_config_held(nvd->vdev_spa, |
3561 | SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); | |
428870ff | 3562 | ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid); |
9babb374 | 3563 | |
d6320ddb | 3564 | for (c = 0; c < nvd->vdev_children; c++) |
428870ff | 3565 | vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]); |
9babb374 | 3566 | |
572e2857 | 3567 | if (nvd->vdev_ops->vdev_op_leaf) { |
9babb374 | 3568 | /* |
572e2857 | 3569 | * Restore the persistent vdev state |
9babb374 | 3570 | */ |
428870ff | 3571 | nvd->vdev_offline = ovd->vdev_offline; |
572e2857 BB |
3572 | nvd->vdev_faulted = ovd->vdev_faulted; |
3573 | nvd->vdev_degraded = ovd->vdev_degraded; | |
3574 | nvd->vdev_removed = ovd->vdev_removed; | |
9babb374 BB |
3575 | } |
3576 | } | |
3577 | ||
572e2857 BB |
3578 | /* |
3579 | * Determine if a log device has valid content. If the vdev was | |
3580 | * removed or faulted in the MOS config then we know that | |
3581 | * the content on the log device has already been written to the pool. | |
3582 | */ | |
3583 | boolean_t | |
3584 | vdev_log_state_valid(vdev_t *vd) | |
3585 | { | |
d6320ddb BB |
3586 | int c; |
3587 | ||
572e2857 BB |
3588 | if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted && |
3589 | !vd->vdev_removed) | |
3590 | return (B_TRUE); | |
3591 | ||
d6320ddb | 3592 | for (c = 0; c < vd->vdev_children; c++) |
572e2857 BB |
3593 | if (vdev_log_state_valid(vd->vdev_child[c])) |
3594 | return (B_TRUE); | |
3595 | ||
3596 | return (B_FALSE); | |
3597 | } | |
3598 | ||
9babb374 BB |
3599 | /* |
3600 | * Expand a vdev if possible. | |
3601 | */ | |
3602 | void | |
3603 | vdev_expand(vdev_t *vd, uint64_t txg) | |
3604 | { | |
3605 | ASSERT(vd->vdev_top == vd); | |
3606 | ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); | |
3607 | ||
3608 | if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) { | |
3609 | VERIFY(vdev_metaslab_init(vd, txg) == 0); | |
3610 | vdev_config_dirty(vd); | |
3611 | } | |
3612 | } | |
428870ff BB |
3613 | |
3614 | /* | |
3615 | * Split a vdev. | |
3616 | */ | |
3617 | void | |
3618 | vdev_split(vdev_t *vd) | |
3619 | { | |
3620 | vdev_t *cvd, *pvd = vd->vdev_parent; | |
3621 | ||
3622 | vdev_remove_child(pvd, vd); | |
3623 | vdev_compact_children(pvd); | |
3624 | ||
3625 | cvd = pvd->vdev_child[0]; | |
3626 | if (pvd->vdev_children == 1) { | |
3627 | vdev_remove_parent(cvd); | |
3628 | cvd->vdev_splitting = B_TRUE; | |
3629 | } | |
3630 | vdev_propagate_state(cvd); | |
3631 | } | |
c28b2279 | 3632 | |
cc92e9d0 GW |
3633 | void |
3634 | vdev_deadman(vdev_t *vd) | |
3635 | { | |
3636 | int c; | |
3637 | ||
3638 | for (c = 0; c < vd->vdev_children; c++) { | |
3639 | vdev_t *cvd = vd->vdev_child[c]; | |
3640 | ||
3641 | vdev_deadman(cvd); | |
3642 | } | |
3643 | ||
3644 | if (vd->vdev_ops->vdev_op_leaf) { | |
3645 | vdev_queue_t *vq = &vd->vdev_queue; | |
3646 | ||
3647 | mutex_enter(&vq->vq_lock); | |
e8b96c60 | 3648 | if (avl_numnodes(&vq->vq_active_tree) > 0) { |
cc92e9d0 GW |
3649 | spa_t *spa = vd->vdev_spa; |
3650 | zio_t *fio; | |
3651 | uint64_t delta; | |
3652 | ||
3653 | /* | |
3654 | * Look at the head of all the pending queues, | |
3655 | * if any I/O has been outstanding for longer than | |
3656 | * the spa_deadman_synctime we log a zevent. | |
3657 | */ | |
e8b96c60 | 3658 | fio = avl_first(&vq->vq_active_tree); |
cb682a17 MA |
3659 | delta = gethrtime() - fio->io_timestamp; |
3660 | if (delta > spa_deadman_synctime(spa)) { | |
3661 | zfs_dbgmsg("SLOW IO: zio timestamp %lluns, " | |
3662 | "delta %lluns, last io %lluns", | |
cc92e9d0 GW |
3663 | fio->io_timestamp, delta, |
3664 | vq->vq_io_complete_ts); | |
3665 | zfs_ereport_post(FM_EREPORT_ZFS_DELAY, | |
3666 | spa, vd, fio, 0, 0); | |
3667 | } | |
3668 | } | |
3669 | mutex_exit(&vq->vq_lock); | |
3670 | } | |
3671 | } | |
3672 | ||
c28b2279 BB |
3673 | #if defined(_KERNEL) && defined(HAVE_SPL) |
3674 | EXPORT_SYMBOL(vdev_fault); | |
3675 | EXPORT_SYMBOL(vdev_degrade); | |
3676 | EXPORT_SYMBOL(vdev_online); | |
3677 | EXPORT_SYMBOL(vdev_offline); | |
3678 | EXPORT_SYMBOL(vdev_clear); | |
b8bcca18 MA |
3679 | |
3680 | module_param(metaslabs_per_vdev, int, 0644); | |
3681 | MODULE_PARM_DESC(metaslabs_per_vdev, | |
3682 | "Divide added vdev into approximately (but no more than) this number " | |
3683 | "of metaslabs"); | |
c28b2279 | 3684 | #endif |