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