]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * CDDL HEADER START | |
3 | * | |
4 | * The contents of this file are subject to the terms of the | |
5 | * Common Development and Distribution License (the "License"). | |
6 | * You may not use this file except in compliance with the License. | |
7 | * | |
8 | * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE | |
9 | * or http://www.opensolaris.org/os/licensing. | |
10 | * See the License for the specific language governing permissions | |
11 | * and limitations under the License. | |
12 | * | |
13 | * When distributing Covered Code, include this CDDL HEADER in each | |
14 | * file and include the License file at usr/src/OPENSOLARIS.LICENSE. | |
15 | * If applicable, add the following below this CDDL HEADER, with the | |
16 | * fields enclosed by brackets "[]" replaced with your own identifying | |
17 | * information: Portions Copyright [yyyy] [name of copyright owner] | |
18 | * | |
19 | * CDDL HEADER END | |
20 | */ | |
21 | ||
22 | /* | |
23 | * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. | |
24 | * Copyright (c) 2013 by Delphix. All rights reserved. | |
25 | * Copyright (c) 2015, Nexenta Systems, Inc. All rights reserved. | |
26 | * Copyright (c) 2013, 2014, Nexenta Systems, Inc. All rights reserved. | |
27 | * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. | |
28 | * Copyright 2013 Saso Kiselkov. All rights reserved. | |
29 | * Copyright (c) 2016 Actifio, Inc. All rights reserved. | |
30 | */ | |
31 | ||
32 | /* | |
33 | * SPA: Storage Pool Allocator | |
34 | * | |
35 | * This file contains all the routines used when modifying on-disk SPA state. | |
36 | * This includes opening, importing, destroying, exporting a pool, and syncing a | |
37 | * pool. | |
38 | */ | |
39 | ||
40 | #include <sys/zfs_context.h> | |
41 | #include <sys/fm/fs/zfs.h> | |
42 | #include <sys/spa_impl.h> | |
43 | #include <sys/zio.h> | |
44 | #include <sys/zio_checksum.h> | |
45 | #include <sys/dmu.h> | |
46 | #include <sys/dmu_tx.h> | |
47 | #include <sys/zap.h> | |
48 | #include <sys/zil.h> | |
49 | #include <sys/ddt.h> | |
50 | #include <sys/vdev_impl.h> | |
51 | #include <sys/vdev_disk.h> | |
52 | #include <sys/metaslab.h> | |
53 | #include <sys/metaslab_impl.h> | |
54 | #include <sys/uberblock_impl.h> | |
55 | #include <sys/txg.h> | |
56 | #include <sys/avl.h> | |
57 | #include <sys/dmu_traverse.h> | |
58 | #include <sys/dmu_objset.h> | |
59 | #include <sys/unique.h> | |
60 | #include <sys/dsl_pool.h> | |
61 | #include <sys/dsl_dataset.h> | |
62 | #include <sys/dsl_dir.h> | |
63 | #include <sys/dsl_prop.h> | |
64 | #include <sys/dsl_synctask.h> | |
65 | #include <sys/fs/zfs.h> | |
66 | #include <sys/arc.h> | |
67 | #include <sys/callb.h> | |
68 | #include <sys/systeminfo.h> | |
69 | #include <sys/spa_boot.h> | |
70 | #include <sys/zfs_ioctl.h> | |
71 | #include <sys/dsl_scan.h> | |
72 | #include <sys/zfeature.h> | |
73 | #include <sys/dsl_destroy.h> | |
74 | #include <sys/zvol.h> | |
75 | ||
76 | #ifdef _KERNEL | |
77 | #include <sys/bootprops.h> | |
78 | #include <sys/callb.h> | |
79 | #include <sys/cpupart.h> | |
80 | #include <sys/pool.h> | |
81 | #include <sys/sysdc.h> | |
82 | #include <sys/zone.h> | |
83 | #endif /* _KERNEL */ | |
84 | ||
85 | #include "zfs_prop.h" | |
86 | #include "zfs_comutil.h" | |
87 | ||
88 | /* | |
89 | * The interval, in seconds, at which failed configuration cache file writes | |
90 | * should be retried. | |
91 | */ | |
92 | static int zfs_ccw_retry_interval = 300; | |
93 | ||
94 | typedef enum zti_modes { | |
95 | ZTI_MODE_FIXED, /* value is # of threads (min 1) */ | |
96 | ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */ | |
97 | ZTI_MODE_NULL, /* don't create a taskq */ | |
98 | ZTI_NMODES | |
99 | } zti_modes_t; | |
100 | ||
101 | #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) } | |
102 | #define ZTI_PCT(n) { ZTI_MODE_ONLINE_PERCENT, (n), 1 } | |
103 | #define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 } | |
104 | #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 } | |
105 | ||
106 | #define ZTI_N(n) ZTI_P(n, 1) | |
107 | #define ZTI_ONE ZTI_N(1) | |
108 | ||
109 | typedef struct zio_taskq_info { | |
110 | zti_modes_t zti_mode; | |
111 | uint_t zti_value; | |
112 | uint_t zti_count; | |
113 | } zio_taskq_info_t; | |
114 | ||
115 | static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = { | |
116 | "iss", "iss_h", "int", "int_h" | |
117 | }; | |
118 | ||
119 | /* | |
120 | * This table defines the taskq settings for each ZFS I/O type. When | |
121 | * initializing a pool, we use this table to create an appropriately sized | |
122 | * taskq. Some operations are low volume and therefore have a small, static | |
123 | * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE | |
124 | * macros. Other operations process a large amount of data; the ZTI_BATCH | |
125 | * macro causes us to create a taskq oriented for throughput. Some operations | |
126 | * are so high frequency and short-lived that the taskq itself can become a a | |
127 | * point of lock contention. The ZTI_P(#, #) macro indicates that we need an | |
128 | * additional degree of parallelism specified by the number of threads per- | |
129 | * taskq and the number of taskqs; when dispatching an event in this case, the | |
130 | * particular taskq is chosen at random. | |
131 | * | |
132 | * The different taskq priorities are to handle the different contexts (issue | |
133 | * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that | |
134 | * need to be handled with minimum delay. | |
135 | */ | |
136 | const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = { | |
137 | /* ISSUE ISSUE_HIGH INTR INTR_HIGH */ | |
138 | { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */ | |
139 | { ZTI_N(8), ZTI_NULL, ZTI_P(12, 8), ZTI_NULL }, /* READ */ | |
140 | { ZTI_BATCH, ZTI_N(5), ZTI_P(12, 8), ZTI_N(5) }, /* WRITE */ | |
141 | { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */ | |
142 | { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */ | |
143 | { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */ | |
144 | }; | |
145 | ||
146 | static void spa_sync_version(void *arg, dmu_tx_t *tx); | |
147 | static void spa_sync_props(void *arg, dmu_tx_t *tx); | |
148 | static boolean_t spa_has_active_shared_spare(spa_t *spa); | |
149 | static inline int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config, | |
150 | spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig, | |
151 | char **ereport); | |
152 | static void spa_vdev_resilver_done(spa_t *spa); | |
153 | ||
154 | uint_t zio_taskq_batch_pct = 75; /* 1 thread per cpu in pset */ | |
155 | id_t zio_taskq_psrset_bind = PS_NONE; | |
156 | boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */ | |
157 | uint_t zio_taskq_basedc = 80; /* base duty cycle */ | |
158 | ||
159 | boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */ | |
160 | ||
161 | /* | |
162 | * This (illegal) pool name is used when temporarily importing a spa_t in order | |
163 | * to get the vdev stats associated with the imported devices. | |
164 | */ | |
165 | #define TRYIMPORT_NAME "$import" | |
166 | ||
167 | /* | |
168 | * ========================================================================== | |
169 | * SPA properties routines | |
170 | * ========================================================================== | |
171 | */ | |
172 | ||
173 | /* | |
174 | * Add a (source=src, propname=propval) list to an nvlist. | |
175 | */ | |
176 | static void | |
177 | spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval, | |
178 | uint64_t intval, zprop_source_t src) | |
179 | { | |
180 | const char *propname = zpool_prop_to_name(prop); | |
181 | nvlist_t *propval; | |
182 | ||
183 | VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0); | |
184 | VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0); | |
185 | ||
186 | if (strval != NULL) | |
187 | VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0); | |
188 | else | |
189 | VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0); | |
190 | ||
191 | VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0); | |
192 | nvlist_free(propval); | |
193 | } | |
194 | ||
195 | /* | |
196 | * Get property values from the spa configuration. | |
197 | */ | |
198 | static void | |
199 | spa_prop_get_config(spa_t *spa, nvlist_t **nvp) | |
200 | { | |
201 | vdev_t *rvd = spa->spa_root_vdev; | |
202 | dsl_pool_t *pool = spa->spa_dsl_pool; | |
203 | uint64_t size, alloc, cap, version; | |
204 | const zprop_source_t src = ZPROP_SRC_NONE; | |
205 | spa_config_dirent_t *dp; | |
206 | metaslab_class_t *mc = spa_normal_class(spa); | |
207 | ||
208 | ASSERT(MUTEX_HELD(&spa->spa_props_lock)); | |
209 | ||
210 | if (rvd != NULL) { | |
211 | alloc = metaslab_class_get_alloc(spa_normal_class(spa)); | |
212 | size = metaslab_class_get_space(spa_normal_class(spa)); | |
213 | spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src); | |
214 | spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src); | |
215 | spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src); | |
216 | spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL, | |
217 | size - alloc, src); | |
218 | ||
219 | spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL, | |
220 | metaslab_class_fragmentation(mc), src); | |
221 | spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, | |
222 | metaslab_class_expandable_space(mc), src); | |
223 | spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL, | |
224 | (spa_mode(spa) == FREAD), src); | |
225 | ||
226 | cap = (size == 0) ? 0 : (alloc * 100 / size); | |
227 | spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src); | |
228 | ||
229 | spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL, | |
230 | ddt_get_pool_dedup_ratio(spa), src); | |
231 | ||
232 | spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL, | |
233 | rvd->vdev_state, src); | |
234 | ||
235 | version = spa_version(spa); | |
236 | if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION)) { | |
237 | spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, | |
238 | version, ZPROP_SRC_DEFAULT); | |
239 | } else { | |
240 | spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, | |
241 | version, ZPROP_SRC_LOCAL); | |
242 | } | |
243 | } | |
244 | ||
245 | if (pool != NULL) { | |
246 | /* | |
247 | * The $FREE directory was introduced in SPA_VERSION_DEADLISTS, | |
248 | * when opening pools before this version freedir will be NULL. | |
249 | */ | |
250 | if (pool->dp_free_dir != NULL) { | |
251 | spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL, | |
252 | dsl_dir_phys(pool->dp_free_dir)->dd_used_bytes, | |
253 | src); | |
254 | } else { | |
255 | spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, | |
256 | NULL, 0, src); | |
257 | } | |
258 | ||
259 | if (pool->dp_leak_dir != NULL) { | |
260 | spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL, | |
261 | dsl_dir_phys(pool->dp_leak_dir)->dd_used_bytes, | |
262 | src); | |
263 | } else { | |
264 | spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, | |
265 | NULL, 0, src); | |
266 | } | |
267 | } | |
268 | ||
269 | spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src); | |
270 | ||
271 | if (spa->spa_comment != NULL) { | |
272 | spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment, | |
273 | 0, ZPROP_SRC_LOCAL); | |
274 | } | |
275 | ||
276 | if (spa->spa_root != NULL) | |
277 | spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root, | |
278 | 0, ZPROP_SRC_LOCAL); | |
279 | ||
280 | if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) { | |
281 | spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL, | |
282 | MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE); | |
283 | } else { | |
284 | spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL, | |
285 | SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE); | |
286 | } | |
287 | ||
288 | if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_DNODE)) { | |
289 | spa_prop_add_list(*nvp, ZPOOL_PROP_MAXDNODESIZE, NULL, | |
290 | DNODE_MAX_SIZE, ZPROP_SRC_NONE); | |
291 | } else { | |
292 | spa_prop_add_list(*nvp, ZPOOL_PROP_MAXDNODESIZE, NULL, | |
293 | DNODE_MIN_SIZE, ZPROP_SRC_NONE); | |
294 | } | |
295 | ||
296 | if ((dp = list_head(&spa->spa_config_list)) != NULL) { | |
297 | if (dp->scd_path == NULL) { | |
298 | spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, | |
299 | "none", 0, ZPROP_SRC_LOCAL); | |
300 | } else if (strcmp(dp->scd_path, spa_config_path) != 0) { | |
301 | spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, | |
302 | dp->scd_path, 0, ZPROP_SRC_LOCAL); | |
303 | } | |
304 | } | |
305 | } | |
306 | ||
307 | /* | |
308 | * Get zpool property values. | |
309 | */ | |
310 | int | |
311 | spa_prop_get(spa_t *spa, nvlist_t **nvp) | |
312 | { | |
313 | objset_t *mos = spa->spa_meta_objset; | |
314 | zap_cursor_t zc; | |
315 | zap_attribute_t za; | |
316 | int err; | |
317 | ||
318 | err = nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP); | |
319 | if (err) | |
320 | return (err); | |
321 | ||
322 | mutex_enter(&spa->spa_props_lock); | |
323 | ||
324 | /* | |
325 | * Get properties from the spa config. | |
326 | */ | |
327 | spa_prop_get_config(spa, nvp); | |
328 | ||
329 | /* If no pool property object, no more prop to get. */ | |
330 | if (mos == NULL || spa->spa_pool_props_object == 0) { | |
331 | mutex_exit(&spa->spa_props_lock); | |
332 | goto out; | |
333 | } | |
334 | ||
335 | /* | |
336 | * Get properties from the MOS pool property object. | |
337 | */ | |
338 | for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object); | |
339 | (err = zap_cursor_retrieve(&zc, &za)) == 0; | |
340 | zap_cursor_advance(&zc)) { | |
341 | uint64_t intval = 0; | |
342 | char *strval = NULL; | |
343 | zprop_source_t src = ZPROP_SRC_DEFAULT; | |
344 | zpool_prop_t prop; | |
345 | ||
346 | if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL) | |
347 | continue; | |
348 | ||
349 | switch (za.za_integer_length) { | |
350 | case 8: | |
351 | /* integer property */ | |
352 | if (za.za_first_integer != | |
353 | zpool_prop_default_numeric(prop)) | |
354 | src = ZPROP_SRC_LOCAL; | |
355 | ||
356 | if (prop == ZPOOL_PROP_BOOTFS) { | |
357 | dsl_pool_t *dp; | |
358 | dsl_dataset_t *ds = NULL; | |
359 | ||
360 | dp = spa_get_dsl(spa); | |
361 | dsl_pool_config_enter(dp, FTAG); | |
362 | if ((err = dsl_dataset_hold_obj(dp, | |
363 | za.za_first_integer, FTAG, &ds))) { | |
364 | dsl_pool_config_exit(dp, FTAG); | |
365 | break; | |
366 | } | |
367 | ||
368 | strval = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, | |
369 | KM_SLEEP); | |
370 | dsl_dataset_name(ds, strval); | |
371 | dsl_dataset_rele(ds, FTAG); | |
372 | dsl_pool_config_exit(dp, FTAG); | |
373 | } else { | |
374 | strval = NULL; | |
375 | intval = za.za_first_integer; | |
376 | } | |
377 | ||
378 | spa_prop_add_list(*nvp, prop, strval, intval, src); | |
379 | ||
380 | if (strval != NULL) | |
381 | kmem_free(strval, ZFS_MAX_DATASET_NAME_LEN); | |
382 | ||
383 | break; | |
384 | ||
385 | case 1: | |
386 | /* string property */ | |
387 | strval = kmem_alloc(za.za_num_integers, KM_SLEEP); | |
388 | err = zap_lookup(mos, spa->spa_pool_props_object, | |
389 | za.za_name, 1, za.za_num_integers, strval); | |
390 | if (err) { | |
391 | kmem_free(strval, za.za_num_integers); | |
392 | break; | |
393 | } | |
394 | spa_prop_add_list(*nvp, prop, strval, 0, src); | |
395 | kmem_free(strval, za.za_num_integers); | |
396 | break; | |
397 | ||
398 | default: | |
399 | break; | |
400 | } | |
401 | } | |
402 | zap_cursor_fini(&zc); | |
403 | mutex_exit(&spa->spa_props_lock); | |
404 | out: | |
405 | if (err && err != ENOENT) { | |
406 | nvlist_free(*nvp); | |
407 | *nvp = NULL; | |
408 | return (err); | |
409 | } | |
410 | ||
411 | return (0); | |
412 | } | |
413 | ||
414 | /* | |
415 | * Validate the given pool properties nvlist and modify the list | |
416 | * for the property values to be set. | |
417 | */ | |
418 | static int | |
419 | spa_prop_validate(spa_t *spa, nvlist_t *props) | |
420 | { | |
421 | nvpair_t *elem; | |
422 | int error = 0, reset_bootfs = 0; | |
423 | uint64_t objnum = 0; | |
424 | boolean_t has_feature = B_FALSE; | |
425 | ||
426 | elem = NULL; | |
427 | while ((elem = nvlist_next_nvpair(props, elem)) != NULL) { | |
428 | uint64_t intval; | |
429 | char *strval, *slash, *check, *fname; | |
430 | const char *propname = nvpair_name(elem); | |
431 | zpool_prop_t prop = zpool_name_to_prop(propname); | |
432 | ||
433 | switch ((int)prop) { | |
434 | case ZPROP_INVAL: | |
435 | if (!zpool_prop_feature(propname)) { | |
436 | error = SET_ERROR(EINVAL); | |
437 | break; | |
438 | } | |
439 | ||
440 | /* | |
441 | * Sanitize the input. | |
442 | */ | |
443 | if (nvpair_type(elem) != DATA_TYPE_UINT64) { | |
444 | error = SET_ERROR(EINVAL); | |
445 | break; | |
446 | } | |
447 | ||
448 | if (nvpair_value_uint64(elem, &intval) != 0) { | |
449 | error = SET_ERROR(EINVAL); | |
450 | break; | |
451 | } | |
452 | ||
453 | if (intval != 0) { | |
454 | error = SET_ERROR(EINVAL); | |
455 | break; | |
456 | } | |
457 | ||
458 | fname = strchr(propname, '@') + 1; | |
459 | if (zfeature_lookup_name(fname, NULL) != 0) { | |
460 | error = SET_ERROR(EINVAL); | |
461 | break; | |
462 | } | |
463 | ||
464 | has_feature = B_TRUE; | |
465 | break; | |
466 | ||
467 | case ZPOOL_PROP_VERSION: | |
468 | error = nvpair_value_uint64(elem, &intval); | |
469 | if (!error && | |
470 | (intval < spa_version(spa) || | |
471 | intval > SPA_VERSION_BEFORE_FEATURES || | |
472 | has_feature)) | |
473 | error = SET_ERROR(EINVAL); | |
474 | break; | |
475 | ||
476 | case ZPOOL_PROP_DELEGATION: | |
477 | case ZPOOL_PROP_AUTOREPLACE: | |
478 | case ZPOOL_PROP_LISTSNAPS: | |
479 | case ZPOOL_PROP_AUTOEXPAND: | |
480 | error = nvpair_value_uint64(elem, &intval); | |
481 | if (!error && intval > 1) | |
482 | error = SET_ERROR(EINVAL); | |
483 | break; | |
484 | ||
485 | case ZPOOL_PROP_BOOTFS: | |
486 | /* | |
487 | * If the pool version is less than SPA_VERSION_BOOTFS, | |
488 | * or the pool is still being created (version == 0), | |
489 | * the bootfs property cannot be set. | |
490 | */ | |
491 | if (spa_version(spa) < SPA_VERSION_BOOTFS) { | |
492 | error = SET_ERROR(ENOTSUP); | |
493 | break; | |
494 | } | |
495 | ||
496 | /* | |
497 | * Make sure the vdev config is bootable | |
498 | */ | |
499 | if (!vdev_is_bootable(spa->spa_root_vdev)) { | |
500 | error = SET_ERROR(ENOTSUP); | |
501 | break; | |
502 | } | |
503 | ||
504 | reset_bootfs = 1; | |
505 | ||
506 | error = nvpair_value_string(elem, &strval); | |
507 | ||
508 | if (!error) { | |
509 | objset_t *os; | |
510 | uint64_t propval; | |
511 | ||
512 | if (strval == NULL || strval[0] == '\0') { | |
513 | objnum = zpool_prop_default_numeric( | |
514 | ZPOOL_PROP_BOOTFS); | |
515 | break; | |
516 | } | |
517 | ||
518 | error = dmu_objset_hold(strval, FTAG, &os); | |
519 | if (error) | |
520 | break; | |
521 | ||
522 | /* | |
523 | * Must be ZPL, and its property settings | |
524 | * must be supported by GRUB (compression | |
525 | * is not gzip, and large blocks or large | |
526 | * dnodes are not used). | |
527 | */ | |
528 | ||
529 | if (dmu_objset_type(os) != DMU_OST_ZFS) { | |
530 | error = SET_ERROR(ENOTSUP); | |
531 | } else if ((error = | |
532 | dsl_prop_get_int_ds(dmu_objset_ds(os), | |
533 | zfs_prop_to_name(ZFS_PROP_COMPRESSION), | |
534 | &propval)) == 0 && | |
535 | !BOOTFS_COMPRESS_VALID(propval)) { | |
536 | error = SET_ERROR(ENOTSUP); | |
537 | } else if ((error = | |
538 | dsl_prop_get_int_ds(dmu_objset_ds(os), | |
539 | zfs_prop_to_name(ZFS_PROP_RECORDSIZE), | |
540 | &propval)) == 0 && | |
541 | propval > SPA_OLD_MAXBLOCKSIZE) { | |
542 | error = SET_ERROR(ENOTSUP); | |
543 | } else if ((error = | |
544 | dsl_prop_get_int_ds(dmu_objset_ds(os), | |
545 | zfs_prop_to_name(ZFS_PROP_DNODESIZE), | |
546 | &propval)) == 0 && | |
547 | propval != ZFS_DNSIZE_LEGACY) { | |
548 | error = SET_ERROR(ENOTSUP); | |
549 | } else { | |
550 | objnum = dmu_objset_id(os); | |
551 | } | |
552 | dmu_objset_rele(os, FTAG); | |
553 | } | |
554 | break; | |
555 | ||
556 | case ZPOOL_PROP_FAILUREMODE: | |
557 | error = nvpair_value_uint64(elem, &intval); | |
558 | if (!error && intval > ZIO_FAILURE_MODE_PANIC) | |
559 | error = SET_ERROR(EINVAL); | |
560 | ||
561 | /* | |
562 | * This is a special case which only occurs when | |
563 | * the pool has completely failed. This allows | |
564 | * the user to change the in-core failmode property | |
565 | * without syncing it out to disk (I/Os might | |
566 | * currently be blocked). We do this by returning | |
567 | * EIO to the caller (spa_prop_set) to trick it | |
568 | * into thinking we encountered a property validation | |
569 | * error. | |
570 | */ | |
571 | if (!error && spa_suspended(spa)) { | |
572 | spa->spa_failmode = intval; | |
573 | error = SET_ERROR(EIO); | |
574 | } | |
575 | break; | |
576 | ||
577 | case ZPOOL_PROP_CACHEFILE: | |
578 | if ((error = nvpair_value_string(elem, &strval)) != 0) | |
579 | break; | |
580 | ||
581 | if (strval[0] == '\0') | |
582 | break; | |
583 | ||
584 | if (strcmp(strval, "none") == 0) | |
585 | break; | |
586 | ||
587 | if (strval[0] != '/') { | |
588 | error = SET_ERROR(EINVAL); | |
589 | break; | |
590 | } | |
591 | ||
592 | slash = strrchr(strval, '/'); | |
593 | ASSERT(slash != NULL); | |
594 | ||
595 | if (slash[1] == '\0' || strcmp(slash, "/.") == 0 || | |
596 | strcmp(slash, "/..") == 0) | |
597 | error = SET_ERROR(EINVAL); | |
598 | break; | |
599 | ||
600 | case ZPOOL_PROP_COMMENT: | |
601 | if ((error = nvpair_value_string(elem, &strval)) != 0) | |
602 | break; | |
603 | for (check = strval; *check != '\0'; check++) { | |
604 | if (!isprint(*check)) { | |
605 | error = SET_ERROR(EINVAL); | |
606 | break; | |
607 | } | |
608 | } | |
609 | if (strlen(strval) > ZPROP_MAX_COMMENT) | |
610 | error = SET_ERROR(E2BIG); | |
611 | break; | |
612 | ||
613 | case ZPOOL_PROP_DEDUPDITTO: | |
614 | if (spa_version(spa) < SPA_VERSION_DEDUP) | |
615 | error = SET_ERROR(ENOTSUP); | |
616 | else | |
617 | error = nvpair_value_uint64(elem, &intval); | |
618 | if (error == 0 && | |
619 | intval != 0 && intval < ZIO_DEDUPDITTO_MIN) | |
620 | error = SET_ERROR(EINVAL); | |
621 | break; | |
622 | ||
623 | default: | |
624 | break; | |
625 | } | |
626 | ||
627 | if (error) | |
628 | break; | |
629 | } | |
630 | ||
631 | if (!error && reset_bootfs) { | |
632 | error = nvlist_remove(props, | |
633 | zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING); | |
634 | ||
635 | if (!error) { | |
636 | error = nvlist_add_uint64(props, | |
637 | zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum); | |
638 | } | |
639 | } | |
640 | ||
641 | return (error); | |
642 | } | |
643 | ||
644 | void | |
645 | spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync) | |
646 | { | |
647 | char *cachefile; | |
648 | spa_config_dirent_t *dp; | |
649 | ||
650 | if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), | |
651 | &cachefile) != 0) | |
652 | return; | |
653 | ||
654 | dp = kmem_alloc(sizeof (spa_config_dirent_t), | |
655 | KM_SLEEP); | |
656 | ||
657 | if (cachefile[0] == '\0') | |
658 | dp->scd_path = spa_strdup(spa_config_path); | |
659 | else if (strcmp(cachefile, "none") == 0) | |
660 | dp->scd_path = NULL; | |
661 | else | |
662 | dp->scd_path = spa_strdup(cachefile); | |
663 | ||
664 | list_insert_head(&spa->spa_config_list, dp); | |
665 | if (need_sync) | |
666 | spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); | |
667 | } | |
668 | ||
669 | int | |
670 | spa_prop_set(spa_t *spa, nvlist_t *nvp) | |
671 | { | |
672 | int error; | |
673 | nvpair_t *elem = NULL; | |
674 | boolean_t need_sync = B_FALSE; | |
675 | ||
676 | if ((error = spa_prop_validate(spa, nvp)) != 0) | |
677 | return (error); | |
678 | ||
679 | while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) { | |
680 | zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem)); | |
681 | ||
682 | if (prop == ZPOOL_PROP_CACHEFILE || | |
683 | prop == ZPOOL_PROP_ALTROOT || | |
684 | prop == ZPOOL_PROP_READONLY) | |
685 | continue; | |
686 | ||
687 | if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) { | |
688 | uint64_t ver; | |
689 | ||
690 | if (prop == ZPOOL_PROP_VERSION) { | |
691 | VERIFY(nvpair_value_uint64(elem, &ver) == 0); | |
692 | } else { | |
693 | ASSERT(zpool_prop_feature(nvpair_name(elem))); | |
694 | ver = SPA_VERSION_FEATURES; | |
695 | need_sync = B_TRUE; | |
696 | } | |
697 | ||
698 | /* Save time if the version is already set. */ | |
699 | if (ver == spa_version(spa)) | |
700 | continue; | |
701 | ||
702 | /* | |
703 | * In addition to the pool directory object, we might | |
704 | * create the pool properties object, the features for | |
705 | * read object, the features for write object, or the | |
706 | * feature descriptions object. | |
707 | */ | |
708 | error = dsl_sync_task(spa->spa_name, NULL, | |
709 | spa_sync_version, &ver, | |
710 | 6, ZFS_SPACE_CHECK_RESERVED); | |
711 | if (error) | |
712 | return (error); | |
713 | continue; | |
714 | } | |
715 | ||
716 | need_sync = B_TRUE; | |
717 | break; | |
718 | } | |
719 | ||
720 | if (need_sync) { | |
721 | return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props, | |
722 | nvp, 6, ZFS_SPACE_CHECK_RESERVED)); | |
723 | } | |
724 | ||
725 | return (0); | |
726 | } | |
727 | ||
728 | /* | |
729 | * If the bootfs property value is dsobj, clear it. | |
730 | */ | |
731 | void | |
732 | spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx) | |
733 | { | |
734 | if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) { | |
735 | VERIFY(zap_remove(spa->spa_meta_objset, | |
736 | spa->spa_pool_props_object, | |
737 | zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0); | |
738 | spa->spa_bootfs = 0; | |
739 | } | |
740 | } | |
741 | ||
742 | /*ARGSUSED*/ | |
743 | static int | |
744 | spa_change_guid_check(void *arg, dmu_tx_t *tx) | |
745 | { | |
746 | spa_t *spa = dmu_tx_pool(tx)->dp_spa; | |
747 | vdev_t *rvd = spa->spa_root_vdev; | |
748 | uint64_t vdev_state; | |
749 | ASSERTV(uint64_t *newguid = arg); | |
750 | ||
751 | spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); | |
752 | vdev_state = rvd->vdev_state; | |
753 | spa_config_exit(spa, SCL_STATE, FTAG); | |
754 | ||
755 | if (vdev_state != VDEV_STATE_HEALTHY) | |
756 | return (SET_ERROR(ENXIO)); | |
757 | ||
758 | ASSERT3U(spa_guid(spa), !=, *newguid); | |
759 | ||
760 | return (0); | |
761 | } | |
762 | ||
763 | static void | |
764 | spa_change_guid_sync(void *arg, dmu_tx_t *tx) | |
765 | { | |
766 | uint64_t *newguid = arg; | |
767 | spa_t *spa = dmu_tx_pool(tx)->dp_spa; | |
768 | uint64_t oldguid; | |
769 | vdev_t *rvd = spa->spa_root_vdev; | |
770 | ||
771 | oldguid = spa_guid(spa); | |
772 | ||
773 | spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); | |
774 | rvd->vdev_guid = *newguid; | |
775 | rvd->vdev_guid_sum += (*newguid - oldguid); | |
776 | vdev_config_dirty(rvd); | |
777 | spa_config_exit(spa, SCL_STATE, FTAG); | |
778 | ||
779 | spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu", | |
780 | oldguid, *newguid); | |
781 | } | |
782 | ||
783 | /* | |
784 | * Change the GUID for the pool. This is done so that we can later | |
785 | * re-import a pool built from a clone of our own vdevs. We will modify | |
786 | * the root vdev's guid, our own pool guid, and then mark all of our | |
787 | * vdevs dirty. Note that we must make sure that all our vdevs are | |
788 | * online when we do this, or else any vdevs that weren't present | |
789 | * would be orphaned from our pool. We are also going to issue a | |
790 | * sysevent to update any watchers. | |
791 | */ | |
792 | int | |
793 | spa_change_guid(spa_t *spa) | |
794 | { | |
795 | int error; | |
796 | uint64_t guid; | |
797 | ||
798 | mutex_enter(&spa->spa_vdev_top_lock); | |
799 | mutex_enter(&spa_namespace_lock); | |
800 | guid = spa_generate_guid(NULL); | |
801 | ||
802 | error = dsl_sync_task(spa->spa_name, spa_change_guid_check, | |
803 | spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED); | |
804 | ||
805 | if (error == 0) { | |
806 | spa_config_sync(spa, B_FALSE, B_TRUE); | |
807 | spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID); | |
808 | } | |
809 | ||
810 | mutex_exit(&spa_namespace_lock); | |
811 | mutex_exit(&spa->spa_vdev_top_lock); | |
812 | ||
813 | return (error); | |
814 | } | |
815 | ||
816 | /* | |
817 | * ========================================================================== | |
818 | * SPA state manipulation (open/create/destroy/import/export) | |
819 | * ========================================================================== | |
820 | */ | |
821 | ||
822 | static int | |
823 | spa_error_entry_compare(const void *a, const void *b) | |
824 | { | |
825 | const spa_error_entry_t *sa = (const spa_error_entry_t *)a; | |
826 | const spa_error_entry_t *sb = (const spa_error_entry_t *)b; | |
827 | int ret; | |
828 | ||
829 | ret = memcmp(&sa->se_bookmark, &sb->se_bookmark, | |
830 | sizeof (zbookmark_phys_t)); | |
831 | ||
832 | return (AVL_ISIGN(ret)); | |
833 | } | |
834 | ||
835 | /* | |
836 | * Utility function which retrieves copies of the current logs and | |
837 | * re-initializes them in the process. | |
838 | */ | |
839 | void | |
840 | spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub) | |
841 | { | |
842 | ASSERT(MUTEX_HELD(&spa->spa_errlist_lock)); | |
843 | ||
844 | bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t)); | |
845 | bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t)); | |
846 | ||
847 | avl_create(&spa->spa_errlist_scrub, | |
848 | spa_error_entry_compare, sizeof (spa_error_entry_t), | |
849 | offsetof(spa_error_entry_t, se_avl)); | |
850 | avl_create(&spa->spa_errlist_last, | |
851 | spa_error_entry_compare, sizeof (spa_error_entry_t), | |
852 | offsetof(spa_error_entry_t, se_avl)); | |
853 | } | |
854 | ||
855 | static void | |
856 | spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q) | |
857 | { | |
858 | const zio_taskq_info_t *ztip = &zio_taskqs[t][q]; | |
859 | enum zti_modes mode = ztip->zti_mode; | |
860 | uint_t value = ztip->zti_value; | |
861 | uint_t count = ztip->zti_count; | |
862 | spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; | |
863 | char name[32]; | |
864 | uint_t i, flags = 0; | |
865 | boolean_t batch = B_FALSE; | |
866 | ||
867 | if (mode == ZTI_MODE_NULL) { | |
868 | tqs->stqs_count = 0; | |
869 | tqs->stqs_taskq = NULL; | |
870 | return; | |
871 | } | |
872 | ||
873 | ASSERT3U(count, >, 0); | |
874 | ||
875 | tqs->stqs_count = count; | |
876 | tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP); | |
877 | ||
878 | switch (mode) { | |
879 | case ZTI_MODE_FIXED: | |
880 | ASSERT3U(value, >=, 1); | |
881 | value = MAX(value, 1); | |
882 | flags |= TASKQ_DYNAMIC; | |
883 | break; | |
884 | ||
885 | case ZTI_MODE_BATCH: | |
886 | batch = B_TRUE; | |
887 | flags |= TASKQ_THREADS_CPU_PCT; | |
888 | value = MIN(zio_taskq_batch_pct, 100); | |
889 | break; | |
890 | ||
891 | default: | |
892 | panic("unrecognized mode for %s_%s taskq (%u:%u) in " | |
893 | "spa_activate()", | |
894 | zio_type_name[t], zio_taskq_types[q], mode, value); | |
895 | break; | |
896 | } | |
897 | ||
898 | for (i = 0; i < count; i++) { | |
899 | taskq_t *tq; | |
900 | ||
901 | if (count > 1) { | |
902 | (void) snprintf(name, sizeof (name), "%s_%s_%u", | |
903 | zio_type_name[t], zio_taskq_types[q], i); | |
904 | } else { | |
905 | (void) snprintf(name, sizeof (name), "%s_%s", | |
906 | zio_type_name[t], zio_taskq_types[q]); | |
907 | } | |
908 | ||
909 | if (zio_taskq_sysdc && spa->spa_proc != &p0) { | |
910 | if (batch) | |
911 | flags |= TASKQ_DC_BATCH; | |
912 | ||
913 | tq = taskq_create_sysdc(name, value, 50, INT_MAX, | |
914 | spa->spa_proc, zio_taskq_basedc, flags); | |
915 | } else { | |
916 | pri_t pri = maxclsyspri; | |
917 | /* | |
918 | * The write issue taskq can be extremely CPU | |
919 | * intensive. Run it at slightly less important | |
920 | * priority than the other taskqs. Under Linux this | |
921 | * means incrementing the priority value on platforms | |
922 | * like illumos it should be decremented. | |
923 | */ | |
924 | if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE) | |
925 | pri++; | |
926 | ||
927 | tq = taskq_create_proc(name, value, pri, 50, | |
928 | INT_MAX, spa->spa_proc, flags); | |
929 | } | |
930 | ||
931 | tqs->stqs_taskq[i] = tq; | |
932 | } | |
933 | } | |
934 | ||
935 | static void | |
936 | spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q) | |
937 | { | |
938 | spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; | |
939 | uint_t i; | |
940 | ||
941 | if (tqs->stqs_taskq == NULL) { | |
942 | ASSERT3U(tqs->stqs_count, ==, 0); | |
943 | return; | |
944 | } | |
945 | ||
946 | for (i = 0; i < tqs->stqs_count; i++) { | |
947 | ASSERT3P(tqs->stqs_taskq[i], !=, NULL); | |
948 | taskq_destroy(tqs->stqs_taskq[i]); | |
949 | } | |
950 | ||
951 | kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *)); | |
952 | tqs->stqs_taskq = NULL; | |
953 | } | |
954 | ||
955 | /* | |
956 | * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority. | |
957 | * Note that a type may have multiple discrete taskqs to avoid lock contention | |
958 | * on the taskq itself. In that case we choose which taskq at random by using | |
959 | * the low bits of gethrtime(). | |
960 | */ | |
961 | void | |
962 | spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q, | |
963 | task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent) | |
964 | { | |
965 | spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; | |
966 | taskq_t *tq; | |
967 | ||
968 | ASSERT3P(tqs->stqs_taskq, !=, NULL); | |
969 | ASSERT3U(tqs->stqs_count, !=, 0); | |
970 | ||
971 | if (tqs->stqs_count == 1) { | |
972 | tq = tqs->stqs_taskq[0]; | |
973 | } else { | |
974 | tq = tqs->stqs_taskq[((uint64_t)gethrtime()) % tqs->stqs_count]; | |
975 | } | |
976 | ||
977 | taskq_dispatch_ent(tq, func, arg, flags, ent); | |
978 | } | |
979 | ||
980 | /* | |
981 | * Same as spa_taskq_dispatch_ent() but block on the task until completion. | |
982 | */ | |
983 | void | |
984 | spa_taskq_dispatch_sync(spa_t *spa, zio_type_t t, zio_taskq_type_t q, | |
985 | task_func_t *func, void *arg, uint_t flags) | |
986 | { | |
987 | spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; | |
988 | taskq_t *tq; | |
989 | taskqid_t id; | |
990 | ||
991 | ASSERT3P(tqs->stqs_taskq, !=, NULL); | |
992 | ASSERT3U(tqs->stqs_count, !=, 0); | |
993 | ||
994 | if (tqs->stqs_count == 1) { | |
995 | tq = tqs->stqs_taskq[0]; | |
996 | } else { | |
997 | tq = tqs->stqs_taskq[((uint64_t)gethrtime()) % tqs->stqs_count]; | |
998 | } | |
999 | ||
1000 | id = taskq_dispatch(tq, func, arg, flags); | |
1001 | if (id) | |
1002 | taskq_wait_id(tq, id); | |
1003 | } | |
1004 | ||
1005 | static void | |
1006 | spa_create_zio_taskqs(spa_t *spa) | |
1007 | { | |
1008 | int t, q; | |
1009 | ||
1010 | for (t = 0; t < ZIO_TYPES; t++) { | |
1011 | for (q = 0; q < ZIO_TASKQ_TYPES; q++) { | |
1012 | spa_taskqs_init(spa, t, q); | |
1013 | } | |
1014 | } | |
1015 | } | |
1016 | ||
1017 | #if defined(_KERNEL) && defined(HAVE_SPA_THREAD) | |
1018 | static void | |
1019 | spa_thread(void *arg) | |
1020 | { | |
1021 | callb_cpr_t cprinfo; | |
1022 | ||
1023 | spa_t *spa = arg; | |
1024 | user_t *pu = PTOU(curproc); | |
1025 | ||
1026 | CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr, | |
1027 | spa->spa_name); | |
1028 | ||
1029 | ASSERT(curproc != &p0); | |
1030 | (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs), | |
1031 | "zpool-%s", spa->spa_name); | |
1032 | (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm)); | |
1033 | ||
1034 | /* bind this thread to the requested psrset */ | |
1035 | if (zio_taskq_psrset_bind != PS_NONE) { | |
1036 | pool_lock(); | |
1037 | mutex_enter(&cpu_lock); | |
1038 | mutex_enter(&pidlock); | |
1039 | mutex_enter(&curproc->p_lock); | |
1040 | ||
1041 | if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind, | |
1042 | 0, NULL, NULL) == 0) { | |
1043 | curthread->t_bind_pset = zio_taskq_psrset_bind; | |
1044 | } else { | |
1045 | cmn_err(CE_WARN, | |
1046 | "Couldn't bind process for zfs pool \"%s\" to " | |
1047 | "pset %d\n", spa->spa_name, zio_taskq_psrset_bind); | |
1048 | } | |
1049 | ||
1050 | mutex_exit(&curproc->p_lock); | |
1051 | mutex_exit(&pidlock); | |
1052 | mutex_exit(&cpu_lock); | |
1053 | pool_unlock(); | |
1054 | } | |
1055 | ||
1056 | if (zio_taskq_sysdc) { | |
1057 | sysdc_thread_enter(curthread, 100, 0); | |
1058 | } | |
1059 | ||
1060 | spa->spa_proc = curproc; | |
1061 | spa->spa_did = curthread->t_did; | |
1062 | ||
1063 | spa_create_zio_taskqs(spa); | |
1064 | ||
1065 | mutex_enter(&spa->spa_proc_lock); | |
1066 | ASSERT(spa->spa_proc_state == SPA_PROC_CREATED); | |
1067 | ||
1068 | spa->spa_proc_state = SPA_PROC_ACTIVE; | |
1069 | cv_broadcast(&spa->spa_proc_cv); | |
1070 | ||
1071 | CALLB_CPR_SAFE_BEGIN(&cprinfo); | |
1072 | while (spa->spa_proc_state == SPA_PROC_ACTIVE) | |
1073 | cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); | |
1074 | CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock); | |
1075 | ||
1076 | ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE); | |
1077 | spa->spa_proc_state = SPA_PROC_GONE; | |
1078 | spa->spa_proc = &p0; | |
1079 | cv_broadcast(&spa->spa_proc_cv); | |
1080 | CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */ | |
1081 | ||
1082 | mutex_enter(&curproc->p_lock); | |
1083 | lwp_exit(); | |
1084 | } | |
1085 | #endif | |
1086 | ||
1087 | /* | |
1088 | * Activate an uninitialized pool. | |
1089 | */ | |
1090 | static void | |
1091 | spa_activate(spa_t *spa, int mode) | |
1092 | { | |
1093 | ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); | |
1094 | ||
1095 | spa->spa_state = POOL_STATE_ACTIVE; | |
1096 | spa->spa_mode = mode; | |
1097 | ||
1098 | spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops); | |
1099 | spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops); | |
1100 | ||
1101 | /* Try to create a covering process */ | |
1102 | mutex_enter(&spa->spa_proc_lock); | |
1103 | ASSERT(spa->spa_proc_state == SPA_PROC_NONE); | |
1104 | ASSERT(spa->spa_proc == &p0); | |
1105 | spa->spa_did = 0; | |
1106 | ||
1107 | #ifdef HAVE_SPA_THREAD | |
1108 | /* Only create a process if we're going to be around a while. */ | |
1109 | if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) { | |
1110 | if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri, | |
1111 | NULL, 0) == 0) { | |
1112 | spa->spa_proc_state = SPA_PROC_CREATED; | |
1113 | while (spa->spa_proc_state == SPA_PROC_CREATED) { | |
1114 | cv_wait(&spa->spa_proc_cv, | |
1115 | &spa->spa_proc_lock); | |
1116 | } | |
1117 | ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); | |
1118 | ASSERT(spa->spa_proc != &p0); | |
1119 | ASSERT(spa->spa_did != 0); | |
1120 | } else { | |
1121 | #ifdef _KERNEL | |
1122 | cmn_err(CE_WARN, | |
1123 | "Couldn't create process for zfs pool \"%s\"\n", | |
1124 | spa->spa_name); | |
1125 | #endif | |
1126 | } | |
1127 | } | |
1128 | #endif /* HAVE_SPA_THREAD */ | |
1129 | mutex_exit(&spa->spa_proc_lock); | |
1130 | ||
1131 | /* If we didn't create a process, we need to create our taskqs. */ | |
1132 | if (spa->spa_proc == &p0) { | |
1133 | spa_create_zio_taskqs(spa); | |
1134 | } | |
1135 | ||
1136 | list_create(&spa->spa_config_dirty_list, sizeof (vdev_t), | |
1137 | offsetof(vdev_t, vdev_config_dirty_node)); | |
1138 | list_create(&spa->spa_evicting_os_list, sizeof (objset_t), | |
1139 | offsetof(objset_t, os_evicting_node)); | |
1140 | list_create(&spa->spa_state_dirty_list, sizeof (vdev_t), | |
1141 | offsetof(vdev_t, vdev_state_dirty_node)); | |
1142 | ||
1143 | txg_list_create(&spa->spa_vdev_txg_list, | |
1144 | offsetof(struct vdev, vdev_txg_node)); | |
1145 | ||
1146 | avl_create(&spa->spa_errlist_scrub, | |
1147 | spa_error_entry_compare, sizeof (spa_error_entry_t), | |
1148 | offsetof(spa_error_entry_t, se_avl)); | |
1149 | avl_create(&spa->spa_errlist_last, | |
1150 | spa_error_entry_compare, sizeof (spa_error_entry_t), | |
1151 | offsetof(spa_error_entry_t, se_avl)); | |
1152 | ||
1153 | /* | |
1154 | * This taskq is used to perform zvol-minor-related tasks | |
1155 | * asynchronously. This has several advantages, including easy | |
1156 | * resolution of various deadlocks (zfsonlinux bug #3681). | |
1157 | * | |
1158 | * The taskq must be single threaded to ensure tasks are always | |
1159 | * processed in the order in which they were dispatched. | |
1160 | * | |
1161 | * A taskq per pool allows one to keep the pools independent. | |
1162 | * This way if one pool is suspended, it will not impact another. | |
1163 | * | |
1164 | * The preferred location to dispatch a zvol minor task is a sync | |
1165 | * task. In this context, there is easy access to the spa_t and minimal | |
1166 | * error handling is required because the sync task must succeed. | |
1167 | */ | |
1168 | spa->spa_zvol_taskq = taskq_create("z_zvol", 1, defclsyspri, | |
1169 | 1, INT_MAX, 0); | |
1170 | ||
1171 | /* | |
1172 | * The taskq to upgrade datasets in this pool. Currently used by | |
1173 | * feature SPA_FEATURE_USEROBJ_ACCOUNTING. | |
1174 | */ | |
1175 | spa->spa_upgrade_taskq = taskq_create("z_upgrade", boot_ncpus, | |
1176 | defclsyspri, 1, INT_MAX, TASKQ_DYNAMIC); | |
1177 | } | |
1178 | ||
1179 | /* | |
1180 | * Opposite of spa_activate(). | |
1181 | */ | |
1182 | static void | |
1183 | spa_deactivate(spa_t *spa) | |
1184 | { | |
1185 | int t, q; | |
1186 | ||
1187 | ASSERT(spa->spa_sync_on == B_FALSE); | |
1188 | ASSERT(spa->spa_dsl_pool == NULL); | |
1189 | ASSERT(spa->spa_root_vdev == NULL); | |
1190 | ASSERT(spa->spa_async_zio_root == NULL); | |
1191 | ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED); | |
1192 | ||
1193 | spa_evicting_os_wait(spa); | |
1194 | ||
1195 | if (spa->spa_zvol_taskq) { | |
1196 | taskq_destroy(spa->spa_zvol_taskq); | |
1197 | spa->spa_zvol_taskq = NULL; | |
1198 | } | |
1199 | ||
1200 | if (spa->spa_upgrade_taskq) { | |
1201 | taskq_destroy(spa->spa_upgrade_taskq); | |
1202 | spa->spa_upgrade_taskq = NULL; | |
1203 | } | |
1204 | ||
1205 | txg_list_destroy(&spa->spa_vdev_txg_list); | |
1206 | ||
1207 | list_destroy(&spa->spa_config_dirty_list); | |
1208 | list_destroy(&spa->spa_evicting_os_list); | |
1209 | list_destroy(&spa->spa_state_dirty_list); | |
1210 | ||
1211 | taskq_cancel_id(system_delay_taskq, spa->spa_deadman_tqid); | |
1212 | ||
1213 | for (t = 0; t < ZIO_TYPES; t++) { | |
1214 | for (q = 0; q < ZIO_TASKQ_TYPES; q++) { | |
1215 | spa_taskqs_fini(spa, t, q); | |
1216 | } | |
1217 | } | |
1218 | ||
1219 | metaslab_class_destroy(spa->spa_normal_class); | |
1220 | spa->spa_normal_class = NULL; | |
1221 | ||
1222 | metaslab_class_destroy(spa->spa_log_class); | |
1223 | spa->spa_log_class = NULL; | |
1224 | ||
1225 | /* | |
1226 | * If this was part of an import or the open otherwise failed, we may | |
1227 | * still have errors left in the queues. Empty them just in case. | |
1228 | */ | |
1229 | spa_errlog_drain(spa); | |
1230 | ||
1231 | avl_destroy(&spa->spa_errlist_scrub); | |
1232 | avl_destroy(&spa->spa_errlist_last); | |
1233 | ||
1234 | spa->spa_state = POOL_STATE_UNINITIALIZED; | |
1235 | ||
1236 | mutex_enter(&spa->spa_proc_lock); | |
1237 | if (spa->spa_proc_state != SPA_PROC_NONE) { | |
1238 | ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); | |
1239 | spa->spa_proc_state = SPA_PROC_DEACTIVATE; | |
1240 | cv_broadcast(&spa->spa_proc_cv); | |
1241 | while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) { | |
1242 | ASSERT(spa->spa_proc != &p0); | |
1243 | cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); | |
1244 | } | |
1245 | ASSERT(spa->spa_proc_state == SPA_PROC_GONE); | |
1246 | spa->spa_proc_state = SPA_PROC_NONE; | |
1247 | } | |
1248 | ASSERT(spa->spa_proc == &p0); | |
1249 | mutex_exit(&spa->spa_proc_lock); | |
1250 | ||
1251 | /* | |
1252 | * We want to make sure spa_thread() has actually exited the ZFS | |
1253 | * module, so that the module can't be unloaded out from underneath | |
1254 | * it. | |
1255 | */ | |
1256 | if (spa->spa_did != 0) { | |
1257 | thread_join(spa->spa_did); | |
1258 | spa->spa_did = 0; | |
1259 | } | |
1260 | } | |
1261 | ||
1262 | /* | |
1263 | * Verify a pool configuration, and construct the vdev tree appropriately. This | |
1264 | * will create all the necessary vdevs in the appropriate layout, with each vdev | |
1265 | * in the CLOSED state. This will prep the pool before open/creation/import. | |
1266 | * All vdev validation is done by the vdev_alloc() routine. | |
1267 | */ | |
1268 | static int | |
1269 | spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, | |
1270 | uint_t id, int atype) | |
1271 | { | |
1272 | nvlist_t **child; | |
1273 | uint_t children; | |
1274 | int error; | |
1275 | int c; | |
1276 | ||
1277 | if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0) | |
1278 | return (error); | |
1279 | ||
1280 | if ((*vdp)->vdev_ops->vdev_op_leaf) | |
1281 | return (0); | |
1282 | ||
1283 | error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, | |
1284 | &child, &children); | |
1285 | ||
1286 | if (error == ENOENT) | |
1287 | return (0); | |
1288 | ||
1289 | if (error) { | |
1290 | vdev_free(*vdp); | |
1291 | *vdp = NULL; | |
1292 | return (SET_ERROR(EINVAL)); | |
1293 | } | |
1294 | ||
1295 | for (c = 0; c < children; c++) { | |
1296 | vdev_t *vd; | |
1297 | if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c, | |
1298 | atype)) != 0) { | |
1299 | vdev_free(*vdp); | |
1300 | *vdp = NULL; | |
1301 | return (error); | |
1302 | } | |
1303 | } | |
1304 | ||
1305 | ASSERT(*vdp != NULL); | |
1306 | ||
1307 | return (0); | |
1308 | } | |
1309 | ||
1310 | /* | |
1311 | * Opposite of spa_load(). | |
1312 | */ | |
1313 | static void | |
1314 | spa_unload(spa_t *spa) | |
1315 | { | |
1316 | int i, c; | |
1317 | ||
1318 | ASSERT(MUTEX_HELD(&spa_namespace_lock)); | |
1319 | ||
1320 | /* | |
1321 | * Stop async tasks. | |
1322 | */ | |
1323 | spa_async_suspend(spa); | |
1324 | ||
1325 | /* | |
1326 | * Stop syncing. | |
1327 | */ | |
1328 | if (spa->spa_sync_on) { | |
1329 | txg_sync_stop(spa->spa_dsl_pool); | |
1330 | spa->spa_sync_on = B_FALSE; | |
1331 | } | |
1332 | ||
1333 | /* | |
1334 | * Even though vdev_free() also calls vdev_metaslab_fini, we need | |
1335 | * to call it earlier, before we wait for async i/o to complete. | |
1336 | * This ensures that there is no async metaslab prefetching, by | |
1337 | * calling taskq_wait(mg_taskq). | |
1338 | */ | |
1339 | if (spa->spa_root_vdev != NULL) { | |
1340 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
1341 | for (c = 0; c < spa->spa_root_vdev->vdev_children; c++) | |
1342 | vdev_metaslab_fini(spa->spa_root_vdev->vdev_child[c]); | |
1343 | spa_config_exit(spa, SCL_ALL, FTAG); | |
1344 | } | |
1345 | ||
1346 | /* | |
1347 | * Wait for any outstanding async I/O to complete. | |
1348 | */ | |
1349 | if (spa->spa_async_zio_root != NULL) { | |
1350 | for (i = 0; i < max_ncpus; i++) | |
1351 | (void) zio_wait(spa->spa_async_zio_root[i]); | |
1352 | kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *)); | |
1353 | spa->spa_async_zio_root = NULL; | |
1354 | } | |
1355 | ||
1356 | bpobj_close(&spa->spa_deferred_bpobj); | |
1357 | ||
1358 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
1359 | ||
1360 | /* | |
1361 | * Close all vdevs. | |
1362 | */ | |
1363 | if (spa->spa_root_vdev) | |
1364 | vdev_free(spa->spa_root_vdev); | |
1365 | ASSERT(spa->spa_root_vdev == NULL); | |
1366 | ||
1367 | /* | |
1368 | * Close the dsl pool. | |
1369 | */ | |
1370 | if (spa->spa_dsl_pool) { | |
1371 | dsl_pool_close(spa->spa_dsl_pool); | |
1372 | spa->spa_dsl_pool = NULL; | |
1373 | spa->spa_meta_objset = NULL; | |
1374 | } | |
1375 | ||
1376 | ddt_unload(spa); | |
1377 | ||
1378 | /* | |
1379 | * Drop and purge level 2 cache | |
1380 | */ | |
1381 | spa_l2cache_drop(spa); | |
1382 | ||
1383 | for (i = 0; i < spa->spa_spares.sav_count; i++) | |
1384 | vdev_free(spa->spa_spares.sav_vdevs[i]); | |
1385 | if (spa->spa_spares.sav_vdevs) { | |
1386 | kmem_free(spa->spa_spares.sav_vdevs, | |
1387 | spa->spa_spares.sav_count * sizeof (void *)); | |
1388 | spa->spa_spares.sav_vdevs = NULL; | |
1389 | } | |
1390 | if (spa->spa_spares.sav_config) { | |
1391 | nvlist_free(spa->spa_spares.sav_config); | |
1392 | spa->spa_spares.sav_config = NULL; | |
1393 | } | |
1394 | spa->spa_spares.sav_count = 0; | |
1395 | ||
1396 | for (i = 0; i < spa->spa_l2cache.sav_count; i++) { | |
1397 | vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]); | |
1398 | vdev_free(spa->spa_l2cache.sav_vdevs[i]); | |
1399 | } | |
1400 | if (spa->spa_l2cache.sav_vdevs) { | |
1401 | kmem_free(spa->spa_l2cache.sav_vdevs, | |
1402 | spa->spa_l2cache.sav_count * sizeof (void *)); | |
1403 | spa->spa_l2cache.sav_vdevs = NULL; | |
1404 | } | |
1405 | if (spa->spa_l2cache.sav_config) { | |
1406 | nvlist_free(spa->spa_l2cache.sav_config); | |
1407 | spa->spa_l2cache.sav_config = NULL; | |
1408 | } | |
1409 | spa->spa_l2cache.sav_count = 0; | |
1410 | ||
1411 | spa->spa_async_suspended = 0; | |
1412 | ||
1413 | if (spa->spa_comment != NULL) { | |
1414 | spa_strfree(spa->spa_comment); | |
1415 | spa->spa_comment = NULL; | |
1416 | } | |
1417 | ||
1418 | spa_config_exit(spa, SCL_ALL, FTAG); | |
1419 | } | |
1420 | ||
1421 | /* | |
1422 | * Load (or re-load) the current list of vdevs describing the active spares for | |
1423 | * this pool. When this is called, we have some form of basic information in | |
1424 | * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and | |
1425 | * then re-generate a more complete list including status information. | |
1426 | */ | |
1427 | static void | |
1428 | spa_load_spares(spa_t *spa) | |
1429 | { | |
1430 | nvlist_t **spares; | |
1431 | uint_t nspares; | |
1432 | int i; | |
1433 | vdev_t *vd, *tvd; | |
1434 | ||
1435 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); | |
1436 | ||
1437 | /* | |
1438 | * First, close and free any existing spare vdevs. | |
1439 | */ | |
1440 | for (i = 0; i < spa->spa_spares.sav_count; i++) { | |
1441 | vd = spa->spa_spares.sav_vdevs[i]; | |
1442 | ||
1443 | /* Undo the call to spa_activate() below */ | |
1444 | if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, | |
1445 | B_FALSE)) != NULL && tvd->vdev_isspare) | |
1446 | spa_spare_remove(tvd); | |
1447 | vdev_close(vd); | |
1448 | vdev_free(vd); | |
1449 | } | |
1450 | ||
1451 | if (spa->spa_spares.sav_vdevs) | |
1452 | kmem_free(spa->spa_spares.sav_vdevs, | |
1453 | spa->spa_spares.sav_count * sizeof (void *)); | |
1454 | ||
1455 | if (spa->spa_spares.sav_config == NULL) | |
1456 | nspares = 0; | |
1457 | else | |
1458 | VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, | |
1459 | ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); | |
1460 | ||
1461 | spa->spa_spares.sav_count = (int)nspares; | |
1462 | spa->spa_spares.sav_vdevs = NULL; | |
1463 | ||
1464 | if (nspares == 0) | |
1465 | return; | |
1466 | ||
1467 | /* | |
1468 | * Construct the array of vdevs, opening them to get status in the | |
1469 | * process. For each spare, there is potentially two different vdev_t | |
1470 | * structures associated with it: one in the list of spares (used only | |
1471 | * for basic validation purposes) and one in the active vdev | |
1472 | * configuration (if it's spared in). During this phase we open and | |
1473 | * validate each vdev on the spare list. If the vdev also exists in the | |
1474 | * active configuration, then we also mark this vdev as an active spare. | |
1475 | */ | |
1476 | spa->spa_spares.sav_vdevs = kmem_zalloc(nspares * sizeof (void *), | |
1477 | KM_SLEEP); | |
1478 | for (i = 0; i < spa->spa_spares.sav_count; i++) { | |
1479 | VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0, | |
1480 | VDEV_ALLOC_SPARE) == 0); | |
1481 | ASSERT(vd != NULL); | |
1482 | ||
1483 | spa->spa_spares.sav_vdevs[i] = vd; | |
1484 | ||
1485 | if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, | |
1486 | B_FALSE)) != NULL) { | |
1487 | if (!tvd->vdev_isspare) | |
1488 | spa_spare_add(tvd); | |
1489 | ||
1490 | /* | |
1491 | * We only mark the spare active if we were successfully | |
1492 | * able to load the vdev. Otherwise, importing a pool | |
1493 | * with a bad active spare would result in strange | |
1494 | * behavior, because multiple pool would think the spare | |
1495 | * is actively in use. | |
1496 | * | |
1497 | * There is a vulnerability here to an equally bizarre | |
1498 | * circumstance, where a dead active spare is later | |
1499 | * brought back to life (onlined or otherwise). Given | |
1500 | * the rarity of this scenario, and the extra complexity | |
1501 | * it adds, we ignore the possibility. | |
1502 | */ | |
1503 | if (!vdev_is_dead(tvd)) | |
1504 | spa_spare_activate(tvd); | |
1505 | } | |
1506 | ||
1507 | vd->vdev_top = vd; | |
1508 | vd->vdev_aux = &spa->spa_spares; | |
1509 | ||
1510 | if (vdev_open(vd) != 0) | |
1511 | continue; | |
1512 | ||
1513 | if (vdev_validate_aux(vd) == 0) | |
1514 | spa_spare_add(vd); | |
1515 | } | |
1516 | ||
1517 | /* | |
1518 | * Recompute the stashed list of spares, with status information | |
1519 | * this time. | |
1520 | */ | |
1521 | VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, | |
1522 | DATA_TYPE_NVLIST_ARRAY) == 0); | |
1523 | ||
1524 | spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *), | |
1525 | KM_SLEEP); | |
1526 | for (i = 0; i < spa->spa_spares.sav_count; i++) | |
1527 | spares[i] = vdev_config_generate(spa, | |
1528 | spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE); | |
1529 | VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, | |
1530 | ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0); | |
1531 | for (i = 0; i < spa->spa_spares.sav_count; i++) | |
1532 | nvlist_free(spares[i]); | |
1533 | kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *)); | |
1534 | } | |
1535 | ||
1536 | /* | |
1537 | * Load (or re-load) the current list of vdevs describing the active l2cache for | |
1538 | * this pool. When this is called, we have some form of basic information in | |
1539 | * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and | |
1540 | * then re-generate a more complete list including status information. | |
1541 | * Devices which are already active have their details maintained, and are | |
1542 | * not re-opened. | |
1543 | */ | |
1544 | static void | |
1545 | spa_load_l2cache(spa_t *spa) | |
1546 | { | |
1547 | nvlist_t **l2cache; | |
1548 | uint_t nl2cache; | |
1549 | int i, j, oldnvdevs; | |
1550 | uint64_t guid; | |
1551 | vdev_t *vd, **oldvdevs, **newvdevs; | |
1552 | spa_aux_vdev_t *sav = &spa->spa_l2cache; | |
1553 | ||
1554 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); | |
1555 | ||
1556 | oldvdevs = sav->sav_vdevs; | |
1557 | oldnvdevs = sav->sav_count; | |
1558 | sav->sav_vdevs = NULL; | |
1559 | sav->sav_count = 0; | |
1560 | ||
1561 | if (sav->sav_config == NULL) { | |
1562 | nl2cache = 0; | |
1563 | newvdevs = NULL; | |
1564 | goto out; | |
1565 | } | |
1566 | ||
1567 | VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, | |
1568 | ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); | |
1569 | newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP); | |
1570 | ||
1571 | /* | |
1572 | * Process new nvlist of vdevs. | |
1573 | */ | |
1574 | for (i = 0; i < nl2cache; i++) { | |
1575 | VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID, | |
1576 | &guid) == 0); | |
1577 | ||
1578 | newvdevs[i] = NULL; | |
1579 | for (j = 0; j < oldnvdevs; j++) { | |
1580 | vd = oldvdevs[j]; | |
1581 | if (vd != NULL && guid == vd->vdev_guid) { | |
1582 | /* | |
1583 | * Retain previous vdev for add/remove ops. | |
1584 | */ | |
1585 | newvdevs[i] = vd; | |
1586 | oldvdevs[j] = NULL; | |
1587 | break; | |
1588 | } | |
1589 | } | |
1590 | ||
1591 | if (newvdevs[i] == NULL) { | |
1592 | /* | |
1593 | * Create new vdev | |
1594 | */ | |
1595 | VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0, | |
1596 | VDEV_ALLOC_L2CACHE) == 0); | |
1597 | ASSERT(vd != NULL); | |
1598 | newvdevs[i] = vd; | |
1599 | ||
1600 | /* | |
1601 | * Commit this vdev as an l2cache device, | |
1602 | * even if it fails to open. | |
1603 | */ | |
1604 | spa_l2cache_add(vd); | |
1605 | ||
1606 | vd->vdev_top = vd; | |
1607 | vd->vdev_aux = sav; | |
1608 | ||
1609 | spa_l2cache_activate(vd); | |
1610 | ||
1611 | if (vdev_open(vd) != 0) | |
1612 | continue; | |
1613 | ||
1614 | (void) vdev_validate_aux(vd); | |
1615 | ||
1616 | if (!vdev_is_dead(vd)) | |
1617 | l2arc_add_vdev(spa, vd); | |
1618 | } | |
1619 | } | |
1620 | ||
1621 | sav->sav_vdevs = newvdevs; | |
1622 | sav->sav_count = (int)nl2cache; | |
1623 | ||
1624 | /* | |
1625 | * Recompute the stashed list of l2cache devices, with status | |
1626 | * information this time. | |
1627 | */ | |
1628 | VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE, | |
1629 | DATA_TYPE_NVLIST_ARRAY) == 0); | |
1630 | ||
1631 | l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); | |
1632 | for (i = 0; i < sav->sav_count; i++) | |
1633 | l2cache[i] = vdev_config_generate(spa, | |
1634 | sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE); | |
1635 | VERIFY(nvlist_add_nvlist_array(sav->sav_config, | |
1636 | ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0); | |
1637 | ||
1638 | out: | |
1639 | /* | |
1640 | * Purge vdevs that were dropped | |
1641 | */ | |
1642 | for (i = 0; i < oldnvdevs; i++) { | |
1643 | uint64_t pool; | |
1644 | ||
1645 | vd = oldvdevs[i]; | |
1646 | if (vd != NULL) { | |
1647 | ASSERT(vd->vdev_isl2cache); | |
1648 | ||
1649 | if (spa_l2cache_exists(vd->vdev_guid, &pool) && | |
1650 | pool != 0ULL && l2arc_vdev_present(vd)) | |
1651 | l2arc_remove_vdev(vd); | |
1652 | vdev_clear_stats(vd); | |
1653 | vdev_free(vd); | |
1654 | } | |
1655 | } | |
1656 | ||
1657 | if (oldvdevs) | |
1658 | kmem_free(oldvdevs, oldnvdevs * sizeof (void *)); | |
1659 | ||
1660 | for (i = 0; i < sav->sav_count; i++) | |
1661 | nvlist_free(l2cache[i]); | |
1662 | if (sav->sav_count) | |
1663 | kmem_free(l2cache, sav->sav_count * sizeof (void *)); | |
1664 | } | |
1665 | ||
1666 | static int | |
1667 | load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value) | |
1668 | { | |
1669 | dmu_buf_t *db; | |
1670 | char *packed = NULL; | |
1671 | size_t nvsize = 0; | |
1672 | int error; | |
1673 | *value = NULL; | |
1674 | ||
1675 | error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db); | |
1676 | if (error) | |
1677 | return (error); | |
1678 | ||
1679 | nvsize = *(uint64_t *)db->db_data; | |
1680 | dmu_buf_rele(db, FTAG); | |
1681 | ||
1682 | packed = vmem_alloc(nvsize, KM_SLEEP); | |
1683 | error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed, | |
1684 | DMU_READ_PREFETCH); | |
1685 | if (error == 0) | |
1686 | error = nvlist_unpack(packed, nvsize, value, 0); | |
1687 | vmem_free(packed, nvsize); | |
1688 | ||
1689 | return (error); | |
1690 | } | |
1691 | ||
1692 | /* | |
1693 | * Checks to see if the given vdev could not be opened, in which case we post a | |
1694 | * sysevent to notify the autoreplace code that the device has been removed. | |
1695 | */ | |
1696 | static void | |
1697 | spa_check_removed(vdev_t *vd) | |
1698 | { | |
1699 | int c; | |
1700 | ||
1701 | for (c = 0; c < vd->vdev_children; c++) | |
1702 | spa_check_removed(vd->vdev_child[c]); | |
1703 | ||
1704 | if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) && | |
1705 | !vd->vdev_ishole) { | |
1706 | zfs_post_autoreplace(vd->vdev_spa, vd); | |
1707 | spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK); | |
1708 | } | |
1709 | } | |
1710 | ||
1711 | static void | |
1712 | spa_config_valid_zaps(vdev_t *vd, vdev_t *mvd) | |
1713 | { | |
1714 | uint64_t i; | |
1715 | ||
1716 | ASSERT3U(vd->vdev_children, ==, mvd->vdev_children); | |
1717 | ||
1718 | vd->vdev_top_zap = mvd->vdev_top_zap; | |
1719 | vd->vdev_leaf_zap = mvd->vdev_leaf_zap; | |
1720 | ||
1721 | for (i = 0; i < vd->vdev_children; i++) { | |
1722 | spa_config_valid_zaps(vd->vdev_child[i], mvd->vdev_child[i]); | |
1723 | } | |
1724 | } | |
1725 | ||
1726 | /* | |
1727 | * Validate the current config against the MOS config | |
1728 | */ | |
1729 | static boolean_t | |
1730 | spa_config_valid(spa_t *spa, nvlist_t *config) | |
1731 | { | |
1732 | vdev_t *mrvd, *rvd = spa->spa_root_vdev; | |
1733 | nvlist_t *nv; | |
1734 | int c, i; | |
1735 | ||
1736 | VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0); | |
1737 | ||
1738 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
1739 | VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0); | |
1740 | ||
1741 | ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children); | |
1742 | ||
1743 | /* | |
1744 | * If we're doing a normal import, then build up any additional | |
1745 | * diagnostic information about missing devices in this config. | |
1746 | * We'll pass this up to the user for further processing. | |
1747 | */ | |
1748 | if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) { | |
1749 | nvlist_t **child, *nv; | |
1750 | uint64_t idx = 0; | |
1751 | ||
1752 | child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t *), | |
1753 | KM_SLEEP); | |
1754 | VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); | |
1755 | ||
1756 | for (c = 0; c < rvd->vdev_children; c++) { | |
1757 | vdev_t *tvd = rvd->vdev_child[c]; | |
1758 | vdev_t *mtvd = mrvd->vdev_child[c]; | |
1759 | ||
1760 | if (tvd->vdev_ops == &vdev_missing_ops && | |
1761 | mtvd->vdev_ops != &vdev_missing_ops && | |
1762 | mtvd->vdev_islog) | |
1763 | child[idx++] = vdev_config_generate(spa, mtvd, | |
1764 | B_FALSE, 0); | |
1765 | } | |
1766 | ||
1767 | if (idx) { | |
1768 | VERIFY(nvlist_add_nvlist_array(nv, | |
1769 | ZPOOL_CONFIG_CHILDREN, child, idx) == 0); | |
1770 | VERIFY(nvlist_add_nvlist(spa->spa_load_info, | |
1771 | ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0); | |
1772 | ||
1773 | for (i = 0; i < idx; i++) | |
1774 | nvlist_free(child[i]); | |
1775 | } | |
1776 | nvlist_free(nv); | |
1777 | kmem_free(child, rvd->vdev_children * sizeof (char **)); | |
1778 | } | |
1779 | ||
1780 | /* | |
1781 | * Compare the root vdev tree with the information we have | |
1782 | * from the MOS config (mrvd). Check each top-level vdev | |
1783 | * with the corresponding MOS config top-level (mtvd). | |
1784 | */ | |
1785 | for (c = 0; c < rvd->vdev_children; c++) { | |
1786 | vdev_t *tvd = rvd->vdev_child[c]; | |
1787 | vdev_t *mtvd = mrvd->vdev_child[c]; | |
1788 | ||
1789 | /* | |
1790 | * Resolve any "missing" vdevs in the current configuration. | |
1791 | * If we find that the MOS config has more accurate information | |
1792 | * about the top-level vdev then use that vdev instead. | |
1793 | */ | |
1794 | if (tvd->vdev_ops == &vdev_missing_ops && | |
1795 | mtvd->vdev_ops != &vdev_missing_ops) { | |
1796 | ||
1797 | if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) | |
1798 | continue; | |
1799 | ||
1800 | /* | |
1801 | * Device specific actions. | |
1802 | */ | |
1803 | if (mtvd->vdev_islog) { | |
1804 | spa_set_log_state(spa, SPA_LOG_CLEAR); | |
1805 | } else { | |
1806 | /* | |
1807 | * XXX - once we have 'readonly' pool | |
1808 | * support we should be able to handle | |
1809 | * missing data devices by transitioning | |
1810 | * the pool to readonly. | |
1811 | */ | |
1812 | continue; | |
1813 | } | |
1814 | ||
1815 | /* | |
1816 | * Swap the missing vdev with the data we were | |
1817 | * able to obtain from the MOS config. | |
1818 | */ | |
1819 | vdev_remove_child(rvd, tvd); | |
1820 | vdev_remove_child(mrvd, mtvd); | |
1821 | ||
1822 | vdev_add_child(rvd, mtvd); | |
1823 | vdev_add_child(mrvd, tvd); | |
1824 | ||
1825 | spa_config_exit(spa, SCL_ALL, FTAG); | |
1826 | vdev_load(mtvd); | |
1827 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
1828 | ||
1829 | vdev_reopen(rvd); | |
1830 | } else { | |
1831 | if (mtvd->vdev_islog) { | |
1832 | /* | |
1833 | * Load the slog device's state from the MOS | |
1834 | * config since it's possible that the label | |
1835 | * does not contain the most up-to-date | |
1836 | * information. | |
1837 | */ | |
1838 | vdev_load_log_state(tvd, mtvd); | |
1839 | vdev_reopen(tvd); | |
1840 | } | |
1841 | ||
1842 | /* | |
1843 | * Per-vdev ZAP info is stored exclusively in the MOS. | |
1844 | */ | |
1845 | spa_config_valid_zaps(tvd, mtvd); | |
1846 | } | |
1847 | } | |
1848 | ||
1849 | vdev_free(mrvd); | |
1850 | spa_config_exit(spa, SCL_ALL, FTAG); | |
1851 | ||
1852 | /* | |
1853 | * Ensure we were able to validate the config. | |
1854 | */ | |
1855 | return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum); | |
1856 | } | |
1857 | ||
1858 | /* | |
1859 | * Check for missing log devices | |
1860 | */ | |
1861 | static boolean_t | |
1862 | spa_check_logs(spa_t *spa) | |
1863 | { | |
1864 | boolean_t rv = B_FALSE; | |
1865 | dsl_pool_t *dp = spa_get_dsl(spa); | |
1866 | ||
1867 | switch (spa->spa_log_state) { | |
1868 | default: | |
1869 | break; | |
1870 | case SPA_LOG_MISSING: | |
1871 | /* need to recheck in case slog has been restored */ | |
1872 | case SPA_LOG_UNKNOWN: | |
1873 | rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj, | |
1874 | zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0); | |
1875 | if (rv) | |
1876 | spa_set_log_state(spa, SPA_LOG_MISSING); | |
1877 | break; | |
1878 | } | |
1879 | return (rv); | |
1880 | } | |
1881 | ||
1882 | static boolean_t | |
1883 | spa_passivate_log(spa_t *spa) | |
1884 | { | |
1885 | vdev_t *rvd = spa->spa_root_vdev; | |
1886 | boolean_t slog_found = B_FALSE; | |
1887 | int c; | |
1888 | ||
1889 | ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); | |
1890 | ||
1891 | if (!spa_has_slogs(spa)) | |
1892 | return (B_FALSE); | |
1893 | ||
1894 | for (c = 0; c < rvd->vdev_children; c++) { | |
1895 | vdev_t *tvd = rvd->vdev_child[c]; | |
1896 | metaslab_group_t *mg = tvd->vdev_mg; | |
1897 | ||
1898 | if (tvd->vdev_islog) { | |
1899 | metaslab_group_passivate(mg); | |
1900 | slog_found = B_TRUE; | |
1901 | } | |
1902 | } | |
1903 | ||
1904 | return (slog_found); | |
1905 | } | |
1906 | ||
1907 | static void | |
1908 | spa_activate_log(spa_t *spa) | |
1909 | { | |
1910 | vdev_t *rvd = spa->spa_root_vdev; | |
1911 | int c; | |
1912 | ||
1913 | ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); | |
1914 | ||
1915 | for (c = 0; c < rvd->vdev_children; c++) { | |
1916 | vdev_t *tvd = rvd->vdev_child[c]; | |
1917 | metaslab_group_t *mg = tvd->vdev_mg; | |
1918 | ||
1919 | if (tvd->vdev_islog) | |
1920 | metaslab_group_activate(mg); | |
1921 | } | |
1922 | } | |
1923 | ||
1924 | int | |
1925 | spa_offline_log(spa_t *spa) | |
1926 | { | |
1927 | int error; | |
1928 | ||
1929 | error = dmu_objset_find(spa_name(spa), zil_vdev_offline, | |
1930 | NULL, DS_FIND_CHILDREN); | |
1931 | if (error == 0) { | |
1932 | /* | |
1933 | * We successfully offlined the log device, sync out the | |
1934 | * current txg so that the "stubby" block can be removed | |
1935 | * by zil_sync(). | |
1936 | */ | |
1937 | txg_wait_synced(spa->spa_dsl_pool, 0); | |
1938 | } | |
1939 | return (error); | |
1940 | } | |
1941 | ||
1942 | static void | |
1943 | spa_aux_check_removed(spa_aux_vdev_t *sav) | |
1944 | { | |
1945 | int i; | |
1946 | ||
1947 | for (i = 0; i < sav->sav_count; i++) | |
1948 | spa_check_removed(sav->sav_vdevs[i]); | |
1949 | } | |
1950 | ||
1951 | void | |
1952 | spa_claim_notify(zio_t *zio) | |
1953 | { | |
1954 | spa_t *spa = zio->io_spa; | |
1955 | ||
1956 | if (zio->io_error) | |
1957 | return; | |
1958 | ||
1959 | mutex_enter(&spa->spa_props_lock); /* any mutex will do */ | |
1960 | if (spa->spa_claim_max_txg < zio->io_bp->blk_birth) | |
1961 | spa->spa_claim_max_txg = zio->io_bp->blk_birth; | |
1962 | mutex_exit(&spa->spa_props_lock); | |
1963 | } | |
1964 | ||
1965 | typedef struct spa_load_error { | |
1966 | uint64_t sle_meta_count; | |
1967 | uint64_t sle_data_count; | |
1968 | } spa_load_error_t; | |
1969 | ||
1970 | static void | |
1971 | spa_load_verify_done(zio_t *zio) | |
1972 | { | |
1973 | blkptr_t *bp = zio->io_bp; | |
1974 | spa_load_error_t *sle = zio->io_private; | |
1975 | dmu_object_type_t type = BP_GET_TYPE(bp); | |
1976 | int error = zio->io_error; | |
1977 | spa_t *spa = zio->io_spa; | |
1978 | ||
1979 | abd_free(zio->io_abd); | |
1980 | if (error) { | |
1981 | if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) && | |
1982 | type != DMU_OT_INTENT_LOG) | |
1983 | atomic_inc_64(&sle->sle_meta_count); | |
1984 | else | |
1985 | atomic_inc_64(&sle->sle_data_count); | |
1986 | } | |
1987 | ||
1988 | mutex_enter(&spa->spa_scrub_lock); | |
1989 | spa->spa_scrub_inflight--; | |
1990 | cv_broadcast(&spa->spa_scrub_io_cv); | |
1991 | mutex_exit(&spa->spa_scrub_lock); | |
1992 | } | |
1993 | ||
1994 | /* | |
1995 | * Maximum number of concurrent scrub i/os to create while verifying | |
1996 | * a pool while importing it. | |
1997 | */ | |
1998 | int spa_load_verify_maxinflight = 10000; | |
1999 | int spa_load_verify_metadata = B_TRUE; | |
2000 | int spa_load_verify_data = B_TRUE; | |
2001 | ||
2002 | /*ARGSUSED*/ | |
2003 | static int | |
2004 | spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, | |
2005 | const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg) | |
2006 | { | |
2007 | zio_t *rio; | |
2008 | size_t size; | |
2009 | ||
2010 | if (bp == NULL || BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp)) | |
2011 | return (0); | |
2012 | /* | |
2013 | * Note: normally this routine will not be called if | |
2014 | * spa_load_verify_metadata is not set. However, it may be useful | |
2015 | * to manually set the flag after the traversal has begun. | |
2016 | */ | |
2017 | if (!spa_load_verify_metadata) | |
2018 | return (0); | |
2019 | if (!BP_IS_METADATA(bp) && !spa_load_verify_data) | |
2020 | return (0); | |
2021 | ||
2022 | rio = arg; | |
2023 | size = BP_GET_PSIZE(bp); | |
2024 | ||
2025 | mutex_enter(&spa->spa_scrub_lock); | |
2026 | while (spa->spa_scrub_inflight >= spa_load_verify_maxinflight) | |
2027 | cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); | |
2028 | spa->spa_scrub_inflight++; | |
2029 | mutex_exit(&spa->spa_scrub_lock); | |
2030 | ||
2031 | zio_nowait(zio_read(rio, spa, bp, abd_alloc_for_io(size, B_FALSE), size, | |
2032 | spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB, | |
2033 | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL | | |
2034 | ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb)); | |
2035 | return (0); | |
2036 | } | |
2037 | ||
2038 | /* ARGSUSED */ | |
2039 | int | |
2040 | verify_dataset_name_len(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg) | |
2041 | { | |
2042 | if (dsl_dataset_namelen(ds) >= ZFS_MAX_DATASET_NAME_LEN) | |
2043 | return (SET_ERROR(ENAMETOOLONG)); | |
2044 | ||
2045 | return (0); | |
2046 | } | |
2047 | ||
2048 | static int | |
2049 | spa_load_verify(spa_t *spa) | |
2050 | { | |
2051 | zio_t *rio; | |
2052 | spa_load_error_t sle = { 0 }; | |
2053 | zpool_rewind_policy_t policy; | |
2054 | boolean_t verify_ok = B_FALSE; | |
2055 | int error = 0; | |
2056 | ||
2057 | zpool_get_rewind_policy(spa->spa_config, &policy); | |
2058 | ||
2059 | if (policy.zrp_request & ZPOOL_NEVER_REWIND) | |
2060 | return (0); | |
2061 | ||
2062 | dsl_pool_config_enter(spa->spa_dsl_pool, FTAG); | |
2063 | error = dmu_objset_find_dp(spa->spa_dsl_pool, | |
2064 | spa->spa_dsl_pool->dp_root_dir_obj, verify_dataset_name_len, NULL, | |
2065 | DS_FIND_CHILDREN); | |
2066 | dsl_pool_config_exit(spa->spa_dsl_pool, FTAG); | |
2067 | if (error != 0) | |
2068 | return (error); | |
2069 | ||
2070 | rio = zio_root(spa, NULL, &sle, | |
2071 | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE); | |
2072 | ||
2073 | if (spa_load_verify_metadata) { | |
2074 | error = traverse_pool(spa, spa->spa_verify_min_txg, | |
2075 | TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, | |
2076 | spa_load_verify_cb, rio); | |
2077 | } | |
2078 | ||
2079 | (void) zio_wait(rio); | |
2080 | ||
2081 | spa->spa_load_meta_errors = sle.sle_meta_count; | |
2082 | spa->spa_load_data_errors = sle.sle_data_count; | |
2083 | ||
2084 | if (!error && sle.sle_meta_count <= policy.zrp_maxmeta && | |
2085 | sle.sle_data_count <= policy.zrp_maxdata) { | |
2086 | int64_t loss = 0; | |
2087 | ||
2088 | verify_ok = B_TRUE; | |
2089 | spa->spa_load_txg = spa->spa_uberblock.ub_txg; | |
2090 | spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp; | |
2091 | ||
2092 | loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts; | |
2093 | VERIFY(nvlist_add_uint64(spa->spa_load_info, | |
2094 | ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0); | |
2095 | VERIFY(nvlist_add_int64(spa->spa_load_info, | |
2096 | ZPOOL_CONFIG_REWIND_TIME, loss) == 0); | |
2097 | VERIFY(nvlist_add_uint64(spa->spa_load_info, | |
2098 | ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0); | |
2099 | } else { | |
2100 | spa->spa_load_max_txg = spa->spa_uberblock.ub_txg; | |
2101 | } | |
2102 | ||
2103 | if (error) { | |
2104 | if (error != ENXIO && error != EIO) | |
2105 | error = SET_ERROR(EIO); | |
2106 | return (error); | |
2107 | } | |
2108 | ||
2109 | return (verify_ok ? 0 : EIO); | |
2110 | } | |
2111 | ||
2112 | /* | |
2113 | * Find a value in the pool props object. | |
2114 | */ | |
2115 | static void | |
2116 | spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val) | |
2117 | { | |
2118 | (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object, | |
2119 | zpool_prop_to_name(prop), sizeof (uint64_t), 1, val); | |
2120 | } | |
2121 | ||
2122 | /* | |
2123 | * Find a value in the pool directory object. | |
2124 | */ | |
2125 | static int | |
2126 | spa_dir_prop(spa_t *spa, const char *name, uint64_t *val) | |
2127 | { | |
2128 | return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, | |
2129 | name, sizeof (uint64_t), 1, val)); | |
2130 | } | |
2131 | ||
2132 | static int | |
2133 | spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err) | |
2134 | { | |
2135 | vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux); | |
2136 | return (err); | |
2137 | } | |
2138 | ||
2139 | /* | |
2140 | * Fix up config after a partly-completed split. This is done with the | |
2141 | * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off | |
2142 | * pool have that entry in their config, but only the splitting one contains | |
2143 | * a list of all the guids of the vdevs that are being split off. | |
2144 | * | |
2145 | * This function determines what to do with that list: either rejoin | |
2146 | * all the disks to the pool, or complete the splitting process. To attempt | |
2147 | * the rejoin, each disk that is offlined is marked online again, and | |
2148 | * we do a reopen() call. If the vdev label for every disk that was | |
2149 | * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL) | |
2150 | * then we call vdev_split() on each disk, and complete the split. | |
2151 | * | |
2152 | * Otherwise we leave the config alone, with all the vdevs in place in | |
2153 | * the original pool. | |
2154 | */ | |
2155 | static void | |
2156 | spa_try_repair(spa_t *spa, nvlist_t *config) | |
2157 | { | |
2158 | uint_t extracted; | |
2159 | uint64_t *glist; | |
2160 | uint_t i, gcount; | |
2161 | nvlist_t *nvl; | |
2162 | vdev_t **vd; | |
2163 | boolean_t attempt_reopen; | |
2164 | ||
2165 | if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0) | |
2166 | return; | |
2167 | ||
2168 | /* check that the config is complete */ | |
2169 | if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, | |
2170 | &glist, &gcount) != 0) | |
2171 | return; | |
2172 | ||
2173 | vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP); | |
2174 | ||
2175 | /* attempt to online all the vdevs & validate */ | |
2176 | attempt_reopen = B_TRUE; | |
2177 | for (i = 0; i < gcount; i++) { | |
2178 | if (glist[i] == 0) /* vdev is hole */ | |
2179 | continue; | |
2180 | ||
2181 | vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE); | |
2182 | if (vd[i] == NULL) { | |
2183 | /* | |
2184 | * Don't bother attempting to reopen the disks; | |
2185 | * just do the split. | |
2186 | */ | |
2187 | attempt_reopen = B_FALSE; | |
2188 | } else { | |
2189 | /* attempt to re-online it */ | |
2190 | vd[i]->vdev_offline = B_FALSE; | |
2191 | } | |
2192 | } | |
2193 | ||
2194 | if (attempt_reopen) { | |
2195 | vdev_reopen(spa->spa_root_vdev); | |
2196 | ||
2197 | /* check each device to see what state it's in */ | |
2198 | for (extracted = 0, i = 0; i < gcount; i++) { | |
2199 | if (vd[i] != NULL && | |
2200 | vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL) | |
2201 | break; | |
2202 | ++extracted; | |
2203 | } | |
2204 | } | |
2205 | ||
2206 | /* | |
2207 | * If every disk has been moved to the new pool, or if we never | |
2208 | * even attempted to look at them, then we split them off for | |
2209 | * good. | |
2210 | */ | |
2211 | if (!attempt_reopen || gcount == extracted) { | |
2212 | for (i = 0; i < gcount; i++) | |
2213 | if (vd[i] != NULL) | |
2214 | vdev_split(vd[i]); | |
2215 | vdev_reopen(spa->spa_root_vdev); | |
2216 | } | |
2217 | ||
2218 | kmem_free(vd, gcount * sizeof (vdev_t *)); | |
2219 | } | |
2220 | ||
2221 | static int | |
2222 | spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type, | |
2223 | boolean_t mosconfig) | |
2224 | { | |
2225 | nvlist_t *config = spa->spa_config; | |
2226 | char *ereport = FM_EREPORT_ZFS_POOL; | |
2227 | char *comment; | |
2228 | int error; | |
2229 | uint64_t pool_guid; | |
2230 | nvlist_t *nvl; | |
2231 | ||
2232 | if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) | |
2233 | return (SET_ERROR(EINVAL)); | |
2234 | ||
2235 | ASSERT(spa->spa_comment == NULL); | |
2236 | if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0) | |
2237 | spa->spa_comment = spa_strdup(comment); | |
2238 | ||
2239 | /* | |
2240 | * Versioning wasn't explicitly added to the label until later, so if | |
2241 | * it's not present treat it as the initial version. | |
2242 | */ | |
2243 | if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, | |
2244 | &spa->spa_ubsync.ub_version) != 0) | |
2245 | spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; | |
2246 | ||
2247 | (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, | |
2248 | &spa->spa_config_txg); | |
2249 | ||
2250 | if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) && | |
2251 | spa_guid_exists(pool_guid, 0)) { | |
2252 | error = SET_ERROR(EEXIST); | |
2253 | } else { | |
2254 | spa->spa_config_guid = pool_guid; | |
2255 | ||
2256 | if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, | |
2257 | &nvl) == 0) { | |
2258 | VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting, | |
2259 | KM_SLEEP) == 0); | |
2260 | } | |
2261 | ||
2262 | nvlist_free(spa->spa_load_info); | |
2263 | spa->spa_load_info = fnvlist_alloc(); | |
2264 | ||
2265 | gethrestime(&spa->spa_loaded_ts); | |
2266 | error = spa_load_impl(spa, pool_guid, config, state, type, | |
2267 | mosconfig, &ereport); | |
2268 | } | |
2269 | ||
2270 | /* | |
2271 | * Don't count references from objsets that are already closed | |
2272 | * and are making their way through the eviction process. | |
2273 | */ | |
2274 | spa_evicting_os_wait(spa); | |
2275 | spa->spa_minref = refcount_count(&spa->spa_refcount); | |
2276 | if (error) { | |
2277 | if (error != EEXIST) { | |
2278 | spa->spa_loaded_ts.tv_sec = 0; | |
2279 | spa->spa_loaded_ts.tv_nsec = 0; | |
2280 | } | |
2281 | if (error != EBADF) { | |
2282 | zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0); | |
2283 | } | |
2284 | } | |
2285 | spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE; | |
2286 | spa->spa_ena = 0; | |
2287 | ||
2288 | return (error); | |
2289 | } | |
2290 | ||
2291 | #ifdef ZFS_DEBUG | |
2292 | /* | |
2293 | * Count the number of per-vdev ZAPs associated with all of the vdevs in the | |
2294 | * vdev tree rooted in the given vd, and ensure that each ZAP is present in the | |
2295 | * spa's per-vdev ZAP list. | |
2296 | */ | |
2297 | static uint64_t | |
2298 | vdev_count_verify_zaps(vdev_t *vd) | |
2299 | { | |
2300 | spa_t *spa = vd->vdev_spa; | |
2301 | uint64_t total = 0; | |
2302 | uint64_t i; | |
2303 | ||
2304 | if (vd->vdev_top_zap != 0) { | |
2305 | total++; | |
2306 | ASSERT0(zap_lookup_int(spa->spa_meta_objset, | |
2307 | spa->spa_all_vdev_zaps, vd->vdev_top_zap)); | |
2308 | } | |
2309 | if (vd->vdev_leaf_zap != 0) { | |
2310 | total++; | |
2311 | ASSERT0(zap_lookup_int(spa->spa_meta_objset, | |
2312 | spa->spa_all_vdev_zaps, vd->vdev_leaf_zap)); | |
2313 | } | |
2314 | ||
2315 | for (i = 0; i < vd->vdev_children; i++) { | |
2316 | total += vdev_count_verify_zaps(vd->vdev_child[i]); | |
2317 | } | |
2318 | ||
2319 | return (total); | |
2320 | } | |
2321 | #endif | |
2322 | ||
2323 | /* | |
2324 | * Load an existing storage pool, using the pool's builtin spa_config as a | |
2325 | * source of configuration information. | |
2326 | */ | |
2327 | __attribute__((always_inline)) | |
2328 | static inline int | |
2329 | spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config, | |
2330 | spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig, | |
2331 | char **ereport) | |
2332 | { | |
2333 | int error = 0; | |
2334 | nvlist_t *nvroot = NULL; | |
2335 | nvlist_t *label; | |
2336 | vdev_t *rvd; | |
2337 | uberblock_t *ub = &spa->spa_uberblock; | |
2338 | uint64_t children, config_cache_txg = spa->spa_config_txg; | |
2339 | int orig_mode = spa->spa_mode; | |
2340 | int parse, i; | |
2341 | uint64_t obj; | |
2342 | boolean_t missing_feat_write = B_FALSE; | |
2343 | nvlist_t *mos_config; | |
2344 | ||
2345 | /* | |
2346 | * If this is an untrusted config, access the pool in read-only mode. | |
2347 | * This prevents things like resilvering recently removed devices. | |
2348 | */ | |
2349 | if (!mosconfig) | |
2350 | spa->spa_mode = FREAD; | |
2351 | ||
2352 | ASSERT(MUTEX_HELD(&spa_namespace_lock)); | |
2353 | ||
2354 | spa->spa_load_state = state; | |
2355 | ||
2356 | if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot)) | |
2357 | return (SET_ERROR(EINVAL)); | |
2358 | ||
2359 | parse = (type == SPA_IMPORT_EXISTING ? | |
2360 | VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT); | |
2361 | ||
2362 | /* | |
2363 | * Create "The Godfather" zio to hold all async IOs | |
2364 | */ | |
2365 | spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), | |
2366 | KM_SLEEP); | |
2367 | for (i = 0; i < max_ncpus; i++) { | |
2368 | spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, | |
2369 | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | | |
2370 | ZIO_FLAG_GODFATHER); | |
2371 | } | |
2372 | ||
2373 | /* | |
2374 | * Parse the configuration into a vdev tree. We explicitly set the | |
2375 | * value that will be returned by spa_version() since parsing the | |
2376 | * configuration requires knowing the version number. | |
2377 | */ | |
2378 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
2379 | error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse); | |
2380 | spa_config_exit(spa, SCL_ALL, FTAG); | |
2381 | ||
2382 | if (error != 0) | |
2383 | return (error); | |
2384 | ||
2385 | ASSERT(spa->spa_root_vdev == rvd); | |
2386 | ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT); | |
2387 | ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT); | |
2388 | ||
2389 | if (type != SPA_IMPORT_ASSEMBLE) { | |
2390 | ASSERT(spa_guid(spa) == pool_guid); | |
2391 | } | |
2392 | ||
2393 | /* | |
2394 | * Try to open all vdevs, loading each label in the process. | |
2395 | */ | |
2396 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
2397 | error = vdev_open(rvd); | |
2398 | spa_config_exit(spa, SCL_ALL, FTAG); | |
2399 | if (error != 0) | |
2400 | return (error); | |
2401 | ||
2402 | /* | |
2403 | * We need to validate the vdev labels against the configuration that | |
2404 | * we have in hand, which is dependent on the setting of mosconfig. If | |
2405 | * mosconfig is true then we're validating the vdev labels based on | |
2406 | * that config. Otherwise, we're validating against the cached config | |
2407 | * (zpool.cache) that was read when we loaded the zfs module, and then | |
2408 | * later we will recursively call spa_load() and validate against | |
2409 | * the vdev config. | |
2410 | * | |
2411 | * If we're assembling a new pool that's been split off from an | |
2412 | * existing pool, the labels haven't yet been updated so we skip | |
2413 | * validation for now. | |
2414 | */ | |
2415 | if (type != SPA_IMPORT_ASSEMBLE) { | |
2416 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
2417 | error = vdev_validate(rvd, mosconfig); | |
2418 | spa_config_exit(spa, SCL_ALL, FTAG); | |
2419 | ||
2420 | if (error != 0) | |
2421 | return (error); | |
2422 | ||
2423 | if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) | |
2424 | return (SET_ERROR(ENXIO)); | |
2425 | } | |
2426 | ||
2427 | /* | |
2428 | * Find the best uberblock. | |
2429 | */ | |
2430 | vdev_uberblock_load(rvd, ub, &label); | |
2431 | ||
2432 | /* | |
2433 | * If we weren't able to find a single valid uberblock, return failure. | |
2434 | */ | |
2435 | if (ub->ub_txg == 0) { | |
2436 | nvlist_free(label); | |
2437 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO)); | |
2438 | } | |
2439 | ||
2440 | /* | |
2441 | * If the pool has an unsupported version we can't open it. | |
2442 | */ | |
2443 | if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) { | |
2444 | nvlist_free(label); | |
2445 | return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP)); | |
2446 | } | |
2447 | ||
2448 | if (ub->ub_version >= SPA_VERSION_FEATURES) { | |
2449 | nvlist_t *features; | |
2450 | ||
2451 | /* | |
2452 | * If we weren't able to find what's necessary for reading the | |
2453 | * MOS in the label, return failure. | |
2454 | */ | |
2455 | if (label == NULL || nvlist_lookup_nvlist(label, | |
2456 | ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) { | |
2457 | nvlist_free(label); | |
2458 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, | |
2459 | ENXIO)); | |
2460 | } | |
2461 | ||
2462 | /* | |
2463 | * Update our in-core representation with the definitive values | |
2464 | * from the label. | |
2465 | */ | |
2466 | nvlist_free(spa->spa_label_features); | |
2467 | VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0); | |
2468 | } | |
2469 | ||
2470 | nvlist_free(label); | |
2471 | ||
2472 | /* | |
2473 | * Look through entries in the label nvlist's features_for_read. If | |
2474 | * there is a feature listed there which we don't understand then we | |
2475 | * cannot open a pool. | |
2476 | */ | |
2477 | if (ub->ub_version >= SPA_VERSION_FEATURES) { | |
2478 | nvlist_t *unsup_feat; | |
2479 | nvpair_t *nvp; | |
2480 | ||
2481 | VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) == | |
2482 | 0); | |
2483 | ||
2484 | for (nvp = nvlist_next_nvpair(spa->spa_label_features, NULL); | |
2485 | nvp != NULL; | |
2486 | nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) { | |
2487 | if (!zfeature_is_supported(nvpair_name(nvp))) { | |
2488 | VERIFY(nvlist_add_string(unsup_feat, | |
2489 | nvpair_name(nvp), "") == 0); | |
2490 | } | |
2491 | } | |
2492 | ||
2493 | if (!nvlist_empty(unsup_feat)) { | |
2494 | VERIFY(nvlist_add_nvlist(spa->spa_load_info, | |
2495 | ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0); | |
2496 | nvlist_free(unsup_feat); | |
2497 | return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, | |
2498 | ENOTSUP)); | |
2499 | } | |
2500 | ||
2501 | nvlist_free(unsup_feat); | |
2502 | } | |
2503 | ||
2504 | /* | |
2505 | * If the vdev guid sum doesn't match the uberblock, we have an | |
2506 | * incomplete configuration. We first check to see if the pool | |
2507 | * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN). | |
2508 | * If it is, defer the vdev_guid_sum check till later so we | |
2509 | * can handle missing vdevs. | |
2510 | */ | |
2511 | if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN, | |
2512 | &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE && | |
2513 | rvd->vdev_guid_sum != ub->ub_guid_sum) | |
2514 | return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO)); | |
2515 | ||
2516 | if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) { | |
2517 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
2518 | spa_try_repair(spa, config); | |
2519 | spa_config_exit(spa, SCL_ALL, FTAG); | |
2520 | nvlist_free(spa->spa_config_splitting); | |
2521 | spa->spa_config_splitting = NULL; | |
2522 | } | |
2523 | ||
2524 | /* | |
2525 | * Initialize internal SPA structures. | |
2526 | */ | |
2527 | spa->spa_state = POOL_STATE_ACTIVE; | |
2528 | spa->spa_ubsync = spa->spa_uberblock; | |
2529 | spa->spa_verify_min_txg = spa->spa_extreme_rewind ? | |
2530 | TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1; | |
2531 | spa->spa_first_txg = spa->spa_last_ubsync_txg ? | |
2532 | spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1; | |
2533 | spa->spa_claim_max_txg = spa->spa_first_txg; | |
2534 | spa->spa_prev_software_version = ub->ub_software_version; | |
2535 | ||
2536 | error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool); | |
2537 | if (error) | |
2538 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2539 | spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset; | |
2540 | ||
2541 | if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0) | |
2542 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2543 | ||
2544 | if (spa_version(spa) >= SPA_VERSION_FEATURES) { | |
2545 | boolean_t missing_feat_read = B_FALSE; | |
2546 | nvlist_t *unsup_feat, *enabled_feat; | |
2547 | spa_feature_t i; | |
2548 | ||
2549 | if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ, | |
2550 | &spa->spa_feat_for_read_obj) != 0) { | |
2551 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2552 | } | |
2553 | ||
2554 | if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE, | |
2555 | &spa->spa_feat_for_write_obj) != 0) { | |
2556 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2557 | } | |
2558 | ||
2559 | if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS, | |
2560 | &spa->spa_feat_desc_obj) != 0) { | |
2561 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2562 | } | |
2563 | ||
2564 | enabled_feat = fnvlist_alloc(); | |
2565 | unsup_feat = fnvlist_alloc(); | |
2566 | ||
2567 | if (!spa_features_check(spa, B_FALSE, | |
2568 | unsup_feat, enabled_feat)) | |
2569 | missing_feat_read = B_TRUE; | |
2570 | ||
2571 | if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) { | |
2572 | if (!spa_features_check(spa, B_TRUE, | |
2573 | unsup_feat, enabled_feat)) { | |
2574 | missing_feat_write = B_TRUE; | |
2575 | } | |
2576 | } | |
2577 | ||
2578 | fnvlist_add_nvlist(spa->spa_load_info, | |
2579 | ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat); | |
2580 | ||
2581 | if (!nvlist_empty(unsup_feat)) { | |
2582 | fnvlist_add_nvlist(spa->spa_load_info, | |
2583 | ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat); | |
2584 | } | |
2585 | ||
2586 | fnvlist_free(enabled_feat); | |
2587 | fnvlist_free(unsup_feat); | |
2588 | ||
2589 | if (!missing_feat_read) { | |
2590 | fnvlist_add_boolean(spa->spa_load_info, | |
2591 | ZPOOL_CONFIG_CAN_RDONLY); | |
2592 | } | |
2593 | ||
2594 | /* | |
2595 | * If the state is SPA_LOAD_TRYIMPORT, our objective is | |
2596 | * twofold: to determine whether the pool is available for | |
2597 | * import in read-write mode and (if it is not) whether the | |
2598 | * pool is available for import in read-only mode. If the pool | |
2599 | * is available for import in read-write mode, it is displayed | |
2600 | * as available in userland; if it is not available for import | |
2601 | * in read-only mode, it is displayed as unavailable in | |
2602 | * userland. If the pool is available for import in read-only | |
2603 | * mode but not read-write mode, it is displayed as unavailable | |
2604 | * in userland with a special note that the pool is actually | |
2605 | * available for open in read-only mode. | |
2606 | * | |
2607 | * As a result, if the state is SPA_LOAD_TRYIMPORT and we are | |
2608 | * missing a feature for write, we must first determine whether | |
2609 | * the pool can be opened read-only before returning to | |
2610 | * userland in order to know whether to display the | |
2611 | * abovementioned note. | |
2612 | */ | |
2613 | if (missing_feat_read || (missing_feat_write && | |
2614 | spa_writeable(spa))) { | |
2615 | return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, | |
2616 | ENOTSUP)); | |
2617 | } | |
2618 | ||
2619 | /* | |
2620 | * Load refcounts for ZFS features from disk into an in-memory | |
2621 | * cache during SPA initialization. | |
2622 | */ | |
2623 | for (i = 0; i < SPA_FEATURES; i++) { | |
2624 | uint64_t refcount; | |
2625 | ||
2626 | error = feature_get_refcount_from_disk(spa, | |
2627 | &spa_feature_table[i], &refcount); | |
2628 | if (error == 0) { | |
2629 | spa->spa_feat_refcount_cache[i] = refcount; | |
2630 | } else if (error == ENOTSUP) { | |
2631 | spa->spa_feat_refcount_cache[i] = | |
2632 | SPA_FEATURE_DISABLED; | |
2633 | } else { | |
2634 | return (spa_vdev_err(rvd, | |
2635 | VDEV_AUX_CORRUPT_DATA, EIO)); | |
2636 | } | |
2637 | } | |
2638 | } | |
2639 | ||
2640 | if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) { | |
2641 | if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG, | |
2642 | &spa->spa_feat_enabled_txg_obj) != 0) | |
2643 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2644 | } | |
2645 | ||
2646 | spa->spa_is_initializing = B_TRUE; | |
2647 | error = dsl_pool_open(spa->spa_dsl_pool); | |
2648 | spa->spa_is_initializing = B_FALSE; | |
2649 | if (error != 0) | |
2650 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2651 | ||
2652 | if (!mosconfig) { | |
2653 | uint64_t hostid; | |
2654 | nvlist_t *policy = NULL, *nvconfig; | |
2655 | ||
2656 | if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) | |
2657 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2658 | ||
2659 | if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig, | |
2660 | ZPOOL_CONFIG_HOSTID, &hostid) == 0) { | |
2661 | char *hostname; | |
2662 | unsigned long myhostid = 0; | |
2663 | ||
2664 | VERIFY(nvlist_lookup_string(nvconfig, | |
2665 | ZPOOL_CONFIG_HOSTNAME, &hostname) == 0); | |
2666 | ||
2667 | #ifdef _KERNEL | |
2668 | myhostid = zone_get_hostid(NULL); | |
2669 | #else /* _KERNEL */ | |
2670 | /* | |
2671 | * We're emulating the system's hostid in userland, so | |
2672 | * we can't use zone_get_hostid(). | |
2673 | */ | |
2674 | (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid); | |
2675 | #endif /* _KERNEL */ | |
2676 | if (hostid != 0 && myhostid != 0 && | |
2677 | hostid != myhostid) { | |
2678 | nvlist_free(nvconfig); | |
2679 | cmn_err(CE_WARN, "pool '%s' could not be " | |
2680 | "loaded as it was last accessed by another " | |
2681 | "system (host: %s hostid: 0x%lx). See: " | |
2682 | "http://zfsonlinux.org/msg/ZFS-8000-EY", | |
2683 | spa_name(spa), hostname, | |
2684 | (unsigned long)hostid); | |
2685 | return (SET_ERROR(EBADF)); | |
2686 | } | |
2687 | } | |
2688 | if (nvlist_lookup_nvlist(spa->spa_config, | |
2689 | ZPOOL_REWIND_POLICY, &policy) == 0) | |
2690 | VERIFY(nvlist_add_nvlist(nvconfig, | |
2691 | ZPOOL_REWIND_POLICY, policy) == 0); | |
2692 | ||
2693 | spa_config_set(spa, nvconfig); | |
2694 | spa_unload(spa); | |
2695 | spa_deactivate(spa); | |
2696 | spa_activate(spa, orig_mode); | |
2697 | ||
2698 | return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE)); | |
2699 | } | |
2700 | ||
2701 | /* Grab the checksum salt from the MOS. */ | |
2702 | error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, | |
2703 | DMU_POOL_CHECKSUM_SALT, 1, | |
2704 | sizeof (spa->spa_cksum_salt.zcs_bytes), | |
2705 | spa->spa_cksum_salt.zcs_bytes); | |
2706 | if (error == ENOENT) { | |
2707 | /* Generate a new salt for subsequent use */ | |
2708 | (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes, | |
2709 | sizeof (spa->spa_cksum_salt.zcs_bytes)); | |
2710 | } else if (error != 0) { | |
2711 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2712 | } | |
2713 | ||
2714 | if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0) | |
2715 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2716 | error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj); | |
2717 | if (error != 0) | |
2718 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2719 | ||
2720 | /* | |
2721 | * Load the bit that tells us to use the new accounting function | |
2722 | * (raid-z deflation). If we have an older pool, this will not | |
2723 | * be present. | |
2724 | */ | |
2725 | error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate); | |
2726 | if (error != 0 && error != ENOENT) | |
2727 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2728 | ||
2729 | error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION, | |
2730 | &spa->spa_creation_version); | |
2731 | if (error != 0 && error != ENOENT) | |
2732 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2733 | ||
2734 | /* | |
2735 | * Load the persistent error log. If we have an older pool, this will | |
2736 | * not be present. | |
2737 | */ | |
2738 | error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last); | |
2739 | if (error != 0 && error != ENOENT) | |
2740 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2741 | ||
2742 | error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB, | |
2743 | &spa->spa_errlog_scrub); | |
2744 | if (error != 0 && error != ENOENT) | |
2745 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2746 | ||
2747 | /* | |
2748 | * Load the history object. If we have an older pool, this | |
2749 | * will not be present. | |
2750 | */ | |
2751 | error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history); | |
2752 | if (error != 0 && error != ENOENT) | |
2753 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2754 | ||
2755 | /* | |
2756 | * Load the per-vdev ZAP map. If we have an older pool, this will not | |
2757 | * be present; in this case, defer its creation to a later time to | |
2758 | * avoid dirtying the MOS this early / out of sync context. See | |
2759 | * spa_sync_config_object. | |
2760 | */ | |
2761 | ||
2762 | /* The sentinel is only available in the MOS config. */ | |
2763 | if (load_nvlist(spa, spa->spa_config_object, &mos_config) != 0) | |
2764 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2765 | ||
2766 | error = spa_dir_prop(spa, DMU_POOL_VDEV_ZAP_MAP, | |
2767 | &spa->spa_all_vdev_zaps); | |
2768 | ||
2769 | if (error == ENOENT) { | |
2770 | VERIFY(!nvlist_exists(mos_config, | |
2771 | ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)); | |
2772 | spa->spa_avz_action = AVZ_ACTION_INITIALIZE; | |
2773 | ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev)); | |
2774 | } else if (error != 0) { | |
2775 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2776 | } else if (!nvlist_exists(mos_config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)) { | |
2777 | /* | |
2778 | * An older version of ZFS overwrote the sentinel value, so | |
2779 | * we have orphaned per-vdev ZAPs in the MOS. Defer their | |
2780 | * destruction to later; see spa_sync_config_object. | |
2781 | */ | |
2782 | spa->spa_avz_action = AVZ_ACTION_DESTROY; | |
2783 | /* | |
2784 | * We're assuming that no vdevs have had their ZAPs created | |
2785 | * before this. Better be sure of it. | |
2786 | */ | |
2787 | ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev)); | |
2788 | } | |
2789 | nvlist_free(mos_config); | |
2790 | ||
2791 | /* | |
2792 | * If we're assembling the pool from the split-off vdevs of | |
2793 | * an existing pool, we don't want to attach the spares & cache | |
2794 | * devices. | |
2795 | */ | |
2796 | ||
2797 | /* | |
2798 | * Load any hot spares for this pool. | |
2799 | */ | |
2800 | error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object); | |
2801 | if (error != 0 && error != ENOENT) | |
2802 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2803 | if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { | |
2804 | ASSERT(spa_version(spa) >= SPA_VERSION_SPARES); | |
2805 | if (load_nvlist(spa, spa->spa_spares.sav_object, | |
2806 | &spa->spa_spares.sav_config) != 0) | |
2807 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2808 | ||
2809 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
2810 | spa_load_spares(spa); | |
2811 | spa_config_exit(spa, SCL_ALL, FTAG); | |
2812 | } else if (error == 0) { | |
2813 | spa->spa_spares.sav_sync = B_TRUE; | |
2814 | } | |
2815 | ||
2816 | /* | |
2817 | * Load any level 2 ARC devices for this pool. | |
2818 | */ | |
2819 | error = spa_dir_prop(spa, DMU_POOL_L2CACHE, | |
2820 | &spa->spa_l2cache.sav_object); | |
2821 | if (error != 0 && error != ENOENT) | |
2822 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2823 | if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { | |
2824 | ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE); | |
2825 | if (load_nvlist(spa, spa->spa_l2cache.sav_object, | |
2826 | &spa->spa_l2cache.sav_config) != 0) | |
2827 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2828 | ||
2829 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
2830 | spa_load_l2cache(spa); | |
2831 | spa_config_exit(spa, SCL_ALL, FTAG); | |
2832 | } else if (error == 0) { | |
2833 | spa->spa_l2cache.sav_sync = B_TRUE; | |
2834 | } | |
2835 | ||
2836 | spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); | |
2837 | ||
2838 | error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object); | |
2839 | if (error && error != ENOENT) | |
2840 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2841 | ||
2842 | if (error == 0) { | |
2843 | uint64_t autoreplace = 0; | |
2844 | ||
2845 | spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs); | |
2846 | spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace); | |
2847 | spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation); | |
2848 | spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode); | |
2849 | spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand); | |
2850 | spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO, | |
2851 | &spa->spa_dedup_ditto); | |
2852 | ||
2853 | spa->spa_autoreplace = (autoreplace != 0); | |
2854 | } | |
2855 | ||
2856 | /* | |
2857 | * If the 'autoreplace' property is set, then post a resource notifying | |
2858 | * the ZFS DE that it should not issue any faults for unopenable | |
2859 | * devices. We also iterate over the vdevs, and post a sysevent for any | |
2860 | * unopenable vdevs so that the normal autoreplace handler can take | |
2861 | * over. | |
2862 | */ | |
2863 | if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) { | |
2864 | spa_check_removed(spa->spa_root_vdev); | |
2865 | /* | |
2866 | * For the import case, this is done in spa_import(), because | |
2867 | * at this point we're using the spare definitions from | |
2868 | * the MOS config, not necessarily from the userland config. | |
2869 | */ | |
2870 | if (state != SPA_LOAD_IMPORT) { | |
2871 | spa_aux_check_removed(&spa->spa_spares); | |
2872 | spa_aux_check_removed(&spa->spa_l2cache); | |
2873 | } | |
2874 | } | |
2875 | ||
2876 | /* | |
2877 | * Load the vdev state for all toplevel vdevs. | |
2878 | */ | |
2879 | vdev_load(rvd); | |
2880 | ||
2881 | /* | |
2882 | * Propagate the leaf DTLs we just loaded all the way up the tree. | |
2883 | */ | |
2884 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
2885 | vdev_dtl_reassess(rvd, 0, 0, B_FALSE); | |
2886 | spa_config_exit(spa, SCL_ALL, FTAG); | |
2887 | ||
2888 | /* | |
2889 | * Load the DDTs (dedup tables). | |
2890 | */ | |
2891 | error = ddt_load(spa); | |
2892 | if (error != 0) | |
2893 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2894 | ||
2895 | spa_update_dspace(spa); | |
2896 | ||
2897 | /* | |
2898 | * Validate the config, using the MOS config to fill in any | |
2899 | * information which might be missing. If we fail to validate | |
2900 | * the config then declare the pool unfit for use. If we're | |
2901 | * assembling a pool from a split, the log is not transferred | |
2902 | * over. | |
2903 | */ | |
2904 | if (type != SPA_IMPORT_ASSEMBLE) { | |
2905 | nvlist_t *nvconfig; | |
2906 | ||
2907 | if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) | |
2908 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2909 | ||
2910 | if (!spa_config_valid(spa, nvconfig)) { | |
2911 | nvlist_free(nvconfig); | |
2912 | return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, | |
2913 | ENXIO)); | |
2914 | } | |
2915 | nvlist_free(nvconfig); | |
2916 | ||
2917 | /* | |
2918 | * Now that we've validated the config, check the state of the | |
2919 | * root vdev. If it can't be opened, it indicates one or | |
2920 | * more toplevel vdevs are faulted. | |
2921 | */ | |
2922 | if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) | |
2923 | return (SET_ERROR(ENXIO)); | |
2924 | ||
2925 | if (spa_writeable(spa) && spa_check_logs(spa)) { | |
2926 | *ereport = FM_EREPORT_ZFS_LOG_REPLAY; | |
2927 | return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO)); | |
2928 | } | |
2929 | } | |
2930 | ||
2931 | if (missing_feat_write) { | |
2932 | ASSERT(state == SPA_LOAD_TRYIMPORT); | |
2933 | ||
2934 | /* | |
2935 | * At this point, we know that we can open the pool in | |
2936 | * read-only mode but not read-write mode. We now have enough | |
2937 | * information and can return to userland. | |
2938 | */ | |
2939 | return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP)); | |
2940 | } | |
2941 | ||
2942 | /* | |
2943 | * We've successfully opened the pool, verify that we're ready | |
2944 | * to start pushing transactions. | |
2945 | */ | |
2946 | if (state != SPA_LOAD_TRYIMPORT) { | |
2947 | if ((error = spa_load_verify(spa))) | |
2948 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, | |
2949 | error)); | |
2950 | } | |
2951 | ||
2952 | if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER || | |
2953 | spa->spa_load_max_txg == UINT64_MAX)) { | |
2954 | dmu_tx_t *tx; | |
2955 | int need_update = B_FALSE; | |
2956 | dsl_pool_t *dp = spa_get_dsl(spa); | |
2957 | int c; | |
2958 | ||
2959 | ASSERT(state != SPA_LOAD_TRYIMPORT); | |
2960 | ||
2961 | /* | |
2962 | * Claim log blocks that haven't been committed yet. | |
2963 | * This must all happen in a single txg. | |
2964 | * Note: spa_claim_max_txg is updated by spa_claim_notify(), | |
2965 | * invoked from zil_claim_log_block()'s i/o done callback. | |
2966 | * Price of rollback is that we abandon the log. | |
2967 | */ | |
2968 | spa->spa_claiming = B_TRUE; | |
2969 | ||
2970 | tx = dmu_tx_create_assigned(dp, spa_first_txg(spa)); | |
2971 | (void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj, | |
2972 | zil_claim, tx, DS_FIND_CHILDREN); | |
2973 | dmu_tx_commit(tx); | |
2974 | ||
2975 | spa->spa_claiming = B_FALSE; | |
2976 | ||
2977 | spa_set_log_state(spa, SPA_LOG_GOOD); | |
2978 | spa->spa_sync_on = B_TRUE; | |
2979 | txg_sync_start(spa->spa_dsl_pool); | |
2980 | ||
2981 | /* | |
2982 | * Wait for all claims to sync. We sync up to the highest | |
2983 | * claimed log block birth time so that claimed log blocks | |
2984 | * don't appear to be from the future. spa_claim_max_txg | |
2985 | * will have been set for us by either zil_check_log_chain() | |
2986 | * (invoked from spa_check_logs()) or zil_claim() above. | |
2987 | */ | |
2988 | txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg); | |
2989 | ||
2990 | /* | |
2991 | * If the config cache is stale, or we have uninitialized | |
2992 | * metaslabs (see spa_vdev_add()), then update the config. | |
2993 | * | |
2994 | * If this is a verbatim import, trust the current | |
2995 | * in-core spa_config and update the disk labels. | |
2996 | */ | |
2997 | if (config_cache_txg != spa->spa_config_txg || | |
2998 | state == SPA_LOAD_IMPORT || | |
2999 | state == SPA_LOAD_RECOVER || | |
3000 | (spa->spa_import_flags & ZFS_IMPORT_VERBATIM)) | |
3001 | need_update = B_TRUE; | |
3002 | ||
3003 | for (c = 0; c < rvd->vdev_children; c++) | |
3004 | if (rvd->vdev_child[c]->vdev_ms_array == 0) | |
3005 | need_update = B_TRUE; | |
3006 | ||
3007 | /* | |
3008 | * Update the config cache asychronously in case we're the | |
3009 | * root pool, in which case the config cache isn't writable yet. | |
3010 | */ | |
3011 | if (need_update) | |
3012 | spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); | |
3013 | ||
3014 | /* | |
3015 | * Check all DTLs to see if anything needs resilvering. | |
3016 | */ | |
3017 | if (!dsl_scan_resilvering(spa->spa_dsl_pool) && | |
3018 | vdev_resilver_needed(rvd, NULL, NULL)) | |
3019 | spa_async_request(spa, SPA_ASYNC_RESILVER); | |
3020 | ||
3021 | /* | |
3022 | * Log the fact that we booted up (so that we can detect if | |
3023 | * we rebooted in the middle of an operation). | |
3024 | */ | |
3025 | spa_history_log_version(spa, "open"); | |
3026 | ||
3027 | /* | |
3028 | * Delete any inconsistent datasets. | |
3029 | */ | |
3030 | (void) dmu_objset_find(spa_name(spa), | |
3031 | dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN); | |
3032 | ||
3033 | /* | |
3034 | * Clean up any stale temporary dataset userrefs. | |
3035 | */ | |
3036 | dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool); | |
3037 | } | |
3038 | ||
3039 | return (0); | |
3040 | } | |
3041 | ||
3042 | static int | |
3043 | spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig) | |
3044 | { | |
3045 | int mode = spa->spa_mode; | |
3046 | ||
3047 | spa_unload(spa); | |
3048 | spa_deactivate(spa); | |
3049 | ||
3050 | spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1; | |
3051 | ||
3052 | spa_activate(spa, mode); | |
3053 | spa_async_suspend(spa); | |
3054 | ||
3055 | return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig)); | |
3056 | } | |
3057 | ||
3058 | /* | |
3059 | * If spa_load() fails this function will try loading prior txg's. If | |
3060 | * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool | |
3061 | * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this | |
3062 | * function will not rewind the pool and will return the same error as | |
3063 | * spa_load(). | |
3064 | */ | |
3065 | static int | |
3066 | spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig, | |
3067 | uint64_t max_request, int rewind_flags) | |
3068 | { | |
3069 | nvlist_t *loadinfo = NULL; | |
3070 | nvlist_t *config = NULL; | |
3071 | int load_error, rewind_error; | |
3072 | uint64_t safe_rewind_txg; | |
3073 | uint64_t min_txg; | |
3074 | ||
3075 | if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) { | |
3076 | spa->spa_load_max_txg = spa->spa_load_txg; | |
3077 | spa_set_log_state(spa, SPA_LOG_CLEAR); | |
3078 | } else { | |
3079 | spa->spa_load_max_txg = max_request; | |
3080 | if (max_request != UINT64_MAX) | |
3081 | spa->spa_extreme_rewind = B_TRUE; | |
3082 | } | |
3083 | ||
3084 | load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING, | |
3085 | mosconfig); | |
3086 | if (load_error == 0) | |
3087 | return (0); | |
3088 | ||
3089 | if (spa->spa_root_vdev != NULL) | |
3090 | config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); | |
3091 | ||
3092 | spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg; | |
3093 | spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp; | |
3094 | ||
3095 | if (rewind_flags & ZPOOL_NEVER_REWIND) { | |
3096 | nvlist_free(config); | |
3097 | return (load_error); | |
3098 | } | |
3099 | ||
3100 | if (state == SPA_LOAD_RECOVER) { | |
3101 | /* Price of rolling back is discarding txgs, including log */ | |
3102 | spa_set_log_state(spa, SPA_LOG_CLEAR); | |
3103 | } else { | |
3104 | /* | |
3105 | * If we aren't rolling back save the load info from our first | |
3106 | * import attempt so that we can restore it after attempting | |
3107 | * to rewind. | |
3108 | */ | |
3109 | loadinfo = spa->spa_load_info; | |
3110 | spa->spa_load_info = fnvlist_alloc(); | |
3111 | } | |
3112 | ||
3113 | spa->spa_load_max_txg = spa->spa_last_ubsync_txg; | |
3114 | safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE; | |
3115 | min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ? | |
3116 | TXG_INITIAL : safe_rewind_txg; | |
3117 | ||
3118 | /* | |
3119 | * Continue as long as we're finding errors, we're still within | |
3120 | * the acceptable rewind range, and we're still finding uberblocks | |
3121 | */ | |
3122 | while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg && | |
3123 | spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) { | |
3124 | if (spa->spa_load_max_txg < safe_rewind_txg) | |
3125 | spa->spa_extreme_rewind = B_TRUE; | |
3126 | rewind_error = spa_load_retry(spa, state, mosconfig); | |
3127 | } | |
3128 | ||
3129 | spa->spa_extreme_rewind = B_FALSE; | |
3130 | spa->spa_load_max_txg = UINT64_MAX; | |
3131 | ||
3132 | if (config && (rewind_error || state != SPA_LOAD_RECOVER)) | |
3133 | spa_config_set(spa, config); | |
3134 | else | |
3135 | nvlist_free(config); | |
3136 | ||
3137 | if (state == SPA_LOAD_RECOVER) { | |
3138 | ASSERT3P(loadinfo, ==, NULL); | |
3139 | return (rewind_error); | |
3140 | } else { | |
3141 | /* Store the rewind info as part of the initial load info */ | |
3142 | fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO, | |
3143 | spa->spa_load_info); | |
3144 | ||
3145 | /* Restore the initial load info */ | |
3146 | fnvlist_free(spa->spa_load_info); | |
3147 | spa->spa_load_info = loadinfo; | |
3148 | ||
3149 | return (load_error); | |
3150 | } | |
3151 | } | |
3152 | ||
3153 | /* | |
3154 | * Pool Open/Import | |
3155 | * | |
3156 | * The import case is identical to an open except that the configuration is sent | |
3157 | * down from userland, instead of grabbed from the configuration cache. For the | |
3158 | * case of an open, the pool configuration will exist in the | |
3159 | * POOL_STATE_UNINITIALIZED state. | |
3160 | * | |
3161 | * The stats information (gen/count/ustats) is used to gather vdev statistics at | |
3162 | * the same time open the pool, without having to keep around the spa_t in some | |
3163 | * ambiguous state. | |
3164 | */ | |
3165 | static int | |
3166 | spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy, | |
3167 | nvlist_t **config) | |
3168 | { | |
3169 | spa_t *spa; | |
3170 | spa_load_state_t state = SPA_LOAD_OPEN; | |
3171 | int error; | |
3172 | int locked = B_FALSE; | |
3173 | int firstopen = B_FALSE; | |
3174 | ||
3175 | *spapp = NULL; | |
3176 | ||
3177 | /* | |
3178 | * As disgusting as this is, we need to support recursive calls to this | |
3179 | * function because dsl_dir_open() is called during spa_load(), and ends | |
3180 | * up calling spa_open() again. The real fix is to figure out how to | |
3181 | * avoid dsl_dir_open() calling this in the first place. | |
3182 | */ | |
3183 | if (mutex_owner(&spa_namespace_lock) != curthread) { | |
3184 | mutex_enter(&spa_namespace_lock); | |
3185 | locked = B_TRUE; | |
3186 | } | |
3187 | ||
3188 | if ((spa = spa_lookup(pool)) == NULL) { | |
3189 | if (locked) | |
3190 | mutex_exit(&spa_namespace_lock); | |
3191 | return (SET_ERROR(ENOENT)); | |
3192 | } | |
3193 | ||
3194 | if (spa->spa_state == POOL_STATE_UNINITIALIZED) { | |
3195 | zpool_rewind_policy_t policy; | |
3196 | ||
3197 | firstopen = B_TRUE; | |
3198 | ||
3199 | zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config, | |
3200 | &policy); | |
3201 | if (policy.zrp_request & ZPOOL_DO_REWIND) | |
3202 | state = SPA_LOAD_RECOVER; | |
3203 | ||
3204 | spa_activate(spa, spa_mode_global); | |
3205 | ||
3206 | if (state != SPA_LOAD_RECOVER) | |
3207 | spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; | |
3208 | ||
3209 | error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg, | |
3210 | policy.zrp_request); | |
3211 | ||
3212 | if (error == EBADF) { | |
3213 | /* | |
3214 | * If vdev_validate() returns failure (indicated by | |
3215 | * EBADF), it indicates that one of the vdevs indicates | |
3216 | * that the pool has been exported or destroyed. If | |
3217 | * this is the case, the config cache is out of sync and | |
3218 | * we should remove the pool from the namespace. | |
3219 | */ | |
3220 | spa_unload(spa); | |
3221 | spa_deactivate(spa); | |
3222 | spa_config_sync(spa, B_TRUE, B_TRUE); | |
3223 | spa_remove(spa); | |
3224 | if (locked) | |
3225 | mutex_exit(&spa_namespace_lock); | |
3226 | return (SET_ERROR(ENOENT)); | |
3227 | } | |
3228 | ||
3229 | if (error) { | |
3230 | /* | |
3231 | * We can't open the pool, but we still have useful | |
3232 | * information: the state of each vdev after the | |
3233 | * attempted vdev_open(). Return this to the user. | |
3234 | */ | |
3235 | if (config != NULL && spa->spa_config) { | |
3236 | VERIFY(nvlist_dup(spa->spa_config, config, | |
3237 | KM_SLEEP) == 0); | |
3238 | VERIFY(nvlist_add_nvlist(*config, | |
3239 | ZPOOL_CONFIG_LOAD_INFO, | |
3240 | spa->spa_load_info) == 0); | |
3241 | } | |
3242 | spa_unload(spa); | |
3243 | spa_deactivate(spa); | |
3244 | spa->spa_last_open_failed = error; | |
3245 | if (locked) | |
3246 | mutex_exit(&spa_namespace_lock); | |
3247 | *spapp = NULL; | |
3248 | return (error); | |
3249 | } | |
3250 | } | |
3251 | ||
3252 | spa_open_ref(spa, tag); | |
3253 | ||
3254 | if (config != NULL) | |
3255 | *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); | |
3256 | ||
3257 | /* | |
3258 | * If we've recovered the pool, pass back any information we | |
3259 | * gathered while doing the load. | |
3260 | */ | |
3261 | if (state == SPA_LOAD_RECOVER) { | |
3262 | VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO, | |
3263 | spa->spa_load_info) == 0); | |
3264 | } | |
3265 | ||
3266 | if (locked) { | |
3267 | spa->spa_last_open_failed = 0; | |
3268 | spa->spa_last_ubsync_txg = 0; | |
3269 | spa->spa_load_txg = 0; | |
3270 | mutex_exit(&spa_namespace_lock); | |
3271 | } | |
3272 | ||
3273 | if (firstopen) | |
3274 | zvol_create_minors(spa, spa_name(spa), B_TRUE); | |
3275 | ||
3276 | *spapp = spa; | |
3277 | ||
3278 | return (0); | |
3279 | } | |
3280 | ||
3281 | int | |
3282 | spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy, | |
3283 | nvlist_t **config) | |
3284 | { | |
3285 | return (spa_open_common(name, spapp, tag, policy, config)); | |
3286 | } | |
3287 | ||
3288 | int | |
3289 | spa_open(const char *name, spa_t **spapp, void *tag) | |
3290 | { | |
3291 | return (spa_open_common(name, spapp, tag, NULL, NULL)); | |
3292 | } | |
3293 | ||
3294 | /* | |
3295 | * Lookup the given spa_t, incrementing the inject count in the process, | |
3296 | * preventing it from being exported or destroyed. | |
3297 | */ | |
3298 | spa_t * | |
3299 | spa_inject_addref(char *name) | |
3300 | { | |
3301 | spa_t *spa; | |
3302 | ||
3303 | mutex_enter(&spa_namespace_lock); | |
3304 | if ((spa = spa_lookup(name)) == NULL) { | |
3305 | mutex_exit(&spa_namespace_lock); | |
3306 | return (NULL); | |
3307 | } | |
3308 | spa->spa_inject_ref++; | |
3309 | mutex_exit(&spa_namespace_lock); | |
3310 | ||
3311 | return (spa); | |
3312 | } | |
3313 | ||
3314 | void | |
3315 | spa_inject_delref(spa_t *spa) | |
3316 | { | |
3317 | mutex_enter(&spa_namespace_lock); | |
3318 | spa->spa_inject_ref--; | |
3319 | mutex_exit(&spa_namespace_lock); | |
3320 | } | |
3321 | ||
3322 | /* | |
3323 | * Add spares device information to the nvlist. | |
3324 | */ | |
3325 | static void | |
3326 | spa_add_spares(spa_t *spa, nvlist_t *config) | |
3327 | { | |
3328 | nvlist_t **spares; | |
3329 | uint_t i, nspares; | |
3330 | nvlist_t *nvroot; | |
3331 | uint64_t guid; | |
3332 | vdev_stat_t *vs; | |
3333 | uint_t vsc; | |
3334 | uint64_t pool; | |
3335 | ||
3336 | ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); | |
3337 | ||
3338 | if (spa->spa_spares.sav_count == 0) | |
3339 | return; | |
3340 | ||
3341 | VERIFY(nvlist_lookup_nvlist(config, | |
3342 | ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); | |
3343 | VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, | |
3344 | ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); | |
3345 | if (nspares != 0) { | |
3346 | VERIFY(nvlist_add_nvlist_array(nvroot, | |
3347 | ZPOOL_CONFIG_SPARES, spares, nspares) == 0); | |
3348 | VERIFY(nvlist_lookup_nvlist_array(nvroot, | |
3349 | ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); | |
3350 | ||
3351 | /* | |
3352 | * Go through and find any spares which have since been | |
3353 | * repurposed as an active spare. If this is the case, update | |
3354 | * their status appropriately. | |
3355 | */ | |
3356 | for (i = 0; i < nspares; i++) { | |
3357 | VERIFY(nvlist_lookup_uint64(spares[i], | |
3358 | ZPOOL_CONFIG_GUID, &guid) == 0); | |
3359 | if (spa_spare_exists(guid, &pool, NULL) && | |
3360 | pool != 0ULL) { | |
3361 | VERIFY(nvlist_lookup_uint64_array( | |
3362 | spares[i], ZPOOL_CONFIG_VDEV_STATS, | |
3363 | (uint64_t **)&vs, &vsc) == 0); | |
3364 | vs->vs_state = VDEV_STATE_CANT_OPEN; | |
3365 | vs->vs_aux = VDEV_AUX_SPARED; | |
3366 | } | |
3367 | } | |
3368 | } | |
3369 | } | |
3370 | ||
3371 | /* | |
3372 | * Add l2cache device information to the nvlist, including vdev stats. | |
3373 | */ | |
3374 | static void | |
3375 | spa_add_l2cache(spa_t *spa, nvlist_t *config) | |
3376 | { | |
3377 | nvlist_t **l2cache; | |
3378 | uint_t i, j, nl2cache; | |
3379 | nvlist_t *nvroot; | |
3380 | uint64_t guid; | |
3381 | vdev_t *vd; | |
3382 | vdev_stat_t *vs; | |
3383 | uint_t vsc; | |
3384 | ||
3385 | ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); | |
3386 | ||
3387 | if (spa->spa_l2cache.sav_count == 0) | |
3388 | return; | |
3389 | ||
3390 | VERIFY(nvlist_lookup_nvlist(config, | |
3391 | ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); | |
3392 | VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, | |
3393 | ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); | |
3394 | if (nl2cache != 0) { | |
3395 | VERIFY(nvlist_add_nvlist_array(nvroot, | |
3396 | ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); | |
3397 | VERIFY(nvlist_lookup_nvlist_array(nvroot, | |
3398 | ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); | |
3399 | ||
3400 | /* | |
3401 | * Update level 2 cache device stats. | |
3402 | */ | |
3403 | ||
3404 | for (i = 0; i < nl2cache; i++) { | |
3405 | VERIFY(nvlist_lookup_uint64(l2cache[i], | |
3406 | ZPOOL_CONFIG_GUID, &guid) == 0); | |
3407 | ||
3408 | vd = NULL; | |
3409 | for (j = 0; j < spa->spa_l2cache.sav_count; j++) { | |
3410 | if (guid == | |
3411 | spa->spa_l2cache.sav_vdevs[j]->vdev_guid) { | |
3412 | vd = spa->spa_l2cache.sav_vdevs[j]; | |
3413 | break; | |
3414 | } | |
3415 | } | |
3416 | ASSERT(vd != NULL); | |
3417 | ||
3418 | VERIFY(nvlist_lookup_uint64_array(l2cache[i], | |
3419 | ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc) | |
3420 | == 0); | |
3421 | vdev_get_stats(vd, vs); | |
3422 | vdev_config_generate_stats(vd, l2cache[i]); | |
3423 | ||
3424 | } | |
3425 | } | |
3426 | } | |
3427 | ||
3428 | static void | |
3429 | spa_feature_stats_from_disk(spa_t *spa, nvlist_t *features) | |
3430 | { | |
3431 | zap_cursor_t zc; | |
3432 | zap_attribute_t za; | |
3433 | ||
3434 | if (spa->spa_feat_for_read_obj != 0) { | |
3435 | for (zap_cursor_init(&zc, spa->spa_meta_objset, | |
3436 | spa->spa_feat_for_read_obj); | |
3437 | zap_cursor_retrieve(&zc, &za) == 0; | |
3438 | zap_cursor_advance(&zc)) { | |
3439 | ASSERT(za.za_integer_length == sizeof (uint64_t) && | |
3440 | za.za_num_integers == 1); | |
3441 | VERIFY0(nvlist_add_uint64(features, za.za_name, | |
3442 | za.za_first_integer)); | |
3443 | } | |
3444 | zap_cursor_fini(&zc); | |
3445 | } | |
3446 | ||
3447 | if (spa->spa_feat_for_write_obj != 0) { | |
3448 | for (zap_cursor_init(&zc, spa->spa_meta_objset, | |
3449 | spa->spa_feat_for_write_obj); | |
3450 | zap_cursor_retrieve(&zc, &za) == 0; | |
3451 | zap_cursor_advance(&zc)) { | |
3452 | ASSERT(za.za_integer_length == sizeof (uint64_t) && | |
3453 | za.za_num_integers == 1); | |
3454 | VERIFY0(nvlist_add_uint64(features, za.za_name, | |
3455 | za.za_first_integer)); | |
3456 | } | |
3457 | zap_cursor_fini(&zc); | |
3458 | } | |
3459 | } | |
3460 | ||
3461 | static void | |
3462 | spa_feature_stats_from_cache(spa_t *spa, nvlist_t *features) | |
3463 | { | |
3464 | int i; | |
3465 | ||
3466 | for (i = 0; i < SPA_FEATURES; i++) { | |
3467 | zfeature_info_t feature = spa_feature_table[i]; | |
3468 | uint64_t refcount; | |
3469 | ||
3470 | if (feature_get_refcount(spa, &feature, &refcount) != 0) | |
3471 | continue; | |
3472 | ||
3473 | VERIFY0(nvlist_add_uint64(features, feature.fi_guid, refcount)); | |
3474 | } | |
3475 | } | |
3476 | ||
3477 | /* | |
3478 | * Store a list of pool features and their reference counts in the | |
3479 | * config. | |
3480 | * | |
3481 | * The first time this is called on a spa, allocate a new nvlist, fetch | |
3482 | * the pool features and reference counts from disk, then save the list | |
3483 | * in the spa. In subsequent calls on the same spa use the saved nvlist | |
3484 | * and refresh its values from the cached reference counts. This | |
3485 | * ensures we don't block here on I/O on a suspended pool so 'zpool | |
3486 | * clear' can resume the pool. | |
3487 | */ | |
3488 | static void | |
3489 | spa_add_feature_stats(spa_t *spa, nvlist_t *config) | |
3490 | { | |
3491 | nvlist_t *features; | |
3492 | ||
3493 | ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); | |
3494 | ||
3495 | mutex_enter(&spa->spa_feat_stats_lock); | |
3496 | features = spa->spa_feat_stats; | |
3497 | ||
3498 | if (features != NULL) { | |
3499 | spa_feature_stats_from_cache(spa, features); | |
3500 | } else { | |
3501 | VERIFY0(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP)); | |
3502 | spa->spa_feat_stats = features; | |
3503 | spa_feature_stats_from_disk(spa, features); | |
3504 | } | |
3505 | ||
3506 | VERIFY0(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS, | |
3507 | features)); | |
3508 | ||
3509 | mutex_exit(&spa->spa_feat_stats_lock); | |
3510 | } | |
3511 | ||
3512 | int | |
3513 | spa_get_stats(const char *name, nvlist_t **config, | |
3514 | char *altroot, size_t buflen) | |
3515 | { | |
3516 | int error; | |
3517 | spa_t *spa; | |
3518 | ||
3519 | *config = NULL; | |
3520 | error = spa_open_common(name, &spa, FTAG, NULL, config); | |
3521 | ||
3522 | if (spa != NULL) { | |
3523 | /* | |
3524 | * This still leaves a window of inconsistency where the spares | |
3525 | * or l2cache devices could change and the config would be | |
3526 | * self-inconsistent. | |
3527 | */ | |
3528 | spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); | |
3529 | ||
3530 | if (*config != NULL) { | |
3531 | uint64_t loadtimes[2]; | |
3532 | ||
3533 | loadtimes[0] = spa->spa_loaded_ts.tv_sec; | |
3534 | loadtimes[1] = spa->spa_loaded_ts.tv_nsec; | |
3535 | VERIFY(nvlist_add_uint64_array(*config, | |
3536 | ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0); | |
3537 | ||
3538 | VERIFY(nvlist_add_uint64(*config, | |
3539 | ZPOOL_CONFIG_ERRCOUNT, | |
3540 | spa_get_errlog_size(spa)) == 0); | |
3541 | ||
3542 | if (spa_suspended(spa)) | |
3543 | VERIFY(nvlist_add_uint64(*config, | |
3544 | ZPOOL_CONFIG_SUSPENDED, | |
3545 | spa->spa_failmode) == 0); | |
3546 | ||
3547 | spa_add_spares(spa, *config); | |
3548 | spa_add_l2cache(spa, *config); | |
3549 | spa_add_feature_stats(spa, *config); | |
3550 | } | |
3551 | } | |
3552 | ||
3553 | /* | |
3554 | * We want to get the alternate root even for faulted pools, so we cheat | |
3555 | * and call spa_lookup() directly. | |
3556 | */ | |
3557 | if (altroot) { | |
3558 | if (spa == NULL) { | |
3559 | mutex_enter(&spa_namespace_lock); | |
3560 | spa = spa_lookup(name); | |
3561 | if (spa) | |
3562 | spa_altroot(spa, altroot, buflen); | |
3563 | else | |
3564 | altroot[0] = '\0'; | |
3565 | spa = NULL; | |
3566 | mutex_exit(&spa_namespace_lock); | |
3567 | } else { | |
3568 | spa_altroot(spa, altroot, buflen); | |
3569 | } | |
3570 | } | |
3571 | ||
3572 | if (spa != NULL) { | |
3573 | spa_config_exit(spa, SCL_CONFIG, FTAG); | |
3574 | spa_close(spa, FTAG); | |
3575 | } | |
3576 | ||
3577 | return (error); | |
3578 | } | |
3579 | ||
3580 | /* | |
3581 | * Validate that the auxiliary device array is well formed. We must have an | |
3582 | * array of nvlists, each which describes a valid leaf vdev. If this is an | |
3583 | * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be | |
3584 | * specified, as long as they are well-formed. | |
3585 | */ | |
3586 | static int | |
3587 | spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode, | |
3588 | spa_aux_vdev_t *sav, const char *config, uint64_t version, | |
3589 | vdev_labeltype_t label) | |
3590 | { | |
3591 | nvlist_t **dev; | |
3592 | uint_t i, ndev; | |
3593 | vdev_t *vd; | |
3594 | int error; | |
3595 | ||
3596 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); | |
3597 | ||
3598 | /* | |
3599 | * It's acceptable to have no devs specified. | |
3600 | */ | |
3601 | if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0) | |
3602 | return (0); | |
3603 | ||
3604 | if (ndev == 0) | |
3605 | return (SET_ERROR(EINVAL)); | |
3606 | ||
3607 | /* | |
3608 | * Make sure the pool is formatted with a version that supports this | |
3609 | * device type. | |
3610 | */ | |
3611 | if (spa_version(spa) < version) | |
3612 | return (SET_ERROR(ENOTSUP)); | |
3613 | ||
3614 | /* | |
3615 | * Set the pending device list so we correctly handle device in-use | |
3616 | * checking. | |
3617 | */ | |
3618 | sav->sav_pending = dev; | |
3619 | sav->sav_npending = ndev; | |
3620 | ||
3621 | for (i = 0; i < ndev; i++) { | |
3622 | if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0, | |
3623 | mode)) != 0) | |
3624 | goto out; | |
3625 | ||
3626 | if (!vd->vdev_ops->vdev_op_leaf) { | |
3627 | vdev_free(vd); | |
3628 | error = SET_ERROR(EINVAL); | |
3629 | goto out; | |
3630 | } | |
3631 | ||
3632 | /* | |
3633 | * The L2ARC currently only supports disk devices in | |
3634 | * kernel context. For user-level testing, we allow it. | |
3635 | */ | |
3636 | #ifdef _KERNEL | |
3637 | if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) && | |
3638 | strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) { | |
3639 | error = SET_ERROR(ENOTBLK); | |
3640 | vdev_free(vd); | |
3641 | goto out; | |
3642 | } | |
3643 | #endif | |
3644 | vd->vdev_top = vd; | |
3645 | ||
3646 | if ((error = vdev_open(vd)) == 0 && | |
3647 | (error = vdev_label_init(vd, crtxg, label)) == 0) { | |
3648 | VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID, | |
3649 | vd->vdev_guid) == 0); | |
3650 | } | |
3651 | ||
3652 | vdev_free(vd); | |
3653 | ||
3654 | if (error && | |
3655 | (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE)) | |
3656 | goto out; | |
3657 | else | |
3658 | error = 0; | |
3659 | } | |
3660 | ||
3661 | out: | |
3662 | sav->sav_pending = NULL; | |
3663 | sav->sav_npending = 0; | |
3664 | return (error); | |
3665 | } | |
3666 | ||
3667 | static int | |
3668 | spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode) | |
3669 | { | |
3670 | int error; | |
3671 | ||
3672 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); | |
3673 | ||
3674 | if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode, | |
3675 | &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES, | |
3676 | VDEV_LABEL_SPARE)) != 0) { | |
3677 | return (error); | |
3678 | } | |
3679 | ||
3680 | return (spa_validate_aux_devs(spa, nvroot, crtxg, mode, | |
3681 | &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE, | |
3682 | VDEV_LABEL_L2CACHE)); | |
3683 | } | |
3684 | ||
3685 | static void | |
3686 | spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs, | |
3687 | const char *config) | |
3688 | { | |
3689 | int i; | |
3690 | ||
3691 | if (sav->sav_config != NULL) { | |
3692 | nvlist_t **olddevs; | |
3693 | uint_t oldndevs; | |
3694 | nvlist_t **newdevs; | |
3695 | ||
3696 | /* | |
3697 | * Generate new dev list by concatenating with the | |
3698 | * current dev list. | |
3699 | */ | |
3700 | VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config, | |
3701 | &olddevs, &oldndevs) == 0); | |
3702 | ||
3703 | newdevs = kmem_alloc(sizeof (void *) * | |
3704 | (ndevs + oldndevs), KM_SLEEP); | |
3705 | for (i = 0; i < oldndevs; i++) | |
3706 | VERIFY(nvlist_dup(olddevs[i], &newdevs[i], | |
3707 | KM_SLEEP) == 0); | |
3708 | for (i = 0; i < ndevs; i++) | |
3709 | VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs], | |
3710 | KM_SLEEP) == 0); | |
3711 | ||
3712 | VERIFY(nvlist_remove(sav->sav_config, config, | |
3713 | DATA_TYPE_NVLIST_ARRAY) == 0); | |
3714 | ||
3715 | VERIFY(nvlist_add_nvlist_array(sav->sav_config, | |
3716 | config, newdevs, ndevs + oldndevs) == 0); | |
3717 | for (i = 0; i < oldndevs + ndevs; i++) | |
3718 | nvlist_free(newdevs[i]); | |
3719 | kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *)); | |
3720 | } else { | |
3721 | /* | |
3722 | * Generate a new dev list. | |
3723 | */ | |
3724 | VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME, | |
3725 | KM_SLEEP) == 0); | |
3726 | VERIFY(nvlist_add_nvlist_array(sav->sav_config, config, | |
3727 | devs, ndevs) == 0); | |
3728 | } | |
3729 | } | |
3730 | ||
3731 | /* | |
3732 | * Stop and drop level 2 ARC devices | |
3733 | */ | |
3734 | void | |
3735 | spa_l2cache_drop(spa_t *spa) | |
3736 | { | |
3737 | vdev_t *vd; | |
3738 | int i; | |
3739 | spa_aux_vdev_t *sav = &spa->spa_l2cache; | |
3740 | ||
3741 | for (i = 0; i < sav->sav_count; i++) { | |
3742 | uint64_t pool; | |
3743 | ||
3744 | vd = sav->sav_vdevs[i]; | |
3745 | ASSERT(vd != NULL); | |
3746 | ||
3747 | if (spa_l2cache_exists(vd->vdev_guid, &pool) && | |
3748 | pool != 0ULL && l2arc_vdev_present(vd)) | |
3749 | l2arc_remove_vdev(vd); | |
3750 | } | |
3751 | } | |
3752 | ||
3753 | /* | |
3754 | * Pool Creation | |
3755 | */ | |
3756 | int | |
3757 | spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props, | |
3758 | nvlist_t *zplprops) | |
3759 | { | |
3760 | spa_t *spa; | |
3761 | char *altroot = NULL; | |
3762 | vdev_t *rvd; | |
3763 | dsl_pool_t *dp; | |
3764 | dmu_tx_t *tx; | |
3765 | int error = 0; | |
3766 | uint64_t txg = TXG_INITIAL; | |
3767 | nvlist_t **spares, **l2cache; | |
3768 | uint_t nspares, nl2cache; | |
3769 | uint64_t version, obj; | |
3770 | boolean_t has_features; | |
3771 | nvpair_t *elem; | |
3772 | int c, i; | |
3773 | char *poolname; | |
3774 | nvlist_t *nvl; | |
3775 | ||
3776 | if (nvlist_lookup_string(props, "tname", &poolname) != 0) | |
3777 | poolname = (char *)pool; | |
3778 | ||
3779 | /* | |
3780 | * If this pool already exists, return failure. | |
3781 | */ | |
3782 | mutex_enter(&spa_namespace_lock); | |
3783 | if (spa_lookup(poolname) != NULL) { | |
3784 | mutex_exit(&spa_namespace_lock); | |
3785 | return (SET_ERROR(EEXIST)); | |
3786 | } | |
3787 | ||
3788 | /* | |
3789 | * Allocate a new spa_t structure. | |
3790 | */ | |
3791 | nvl = fnvlist_alloc(); | |
3792 | fnvlist_add_string(nvl, ZPOOL_CONFIG_POOL_NAME, pool); | |
3793 | (void) nvlist_lookup_string(props, | |
3794 | zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); | |
3795 | spa = spa_add(poolname, nvl, altroot); | |
3796 | fnvlist_free(nvl); | |
3797 | spa_activate(spa, spa_mode_global); | |
3798 | ||
3799 | if (props && (error = spa_prop_validate(spa, props))) { | |
3800 | spa_deactivate(spa); | |
3801 | spa_remove(spa); | |
3802 | mutex_exit(&spa_namespace_lock); | |
3803 | return (error); | |
3804 | } | |
3805 | ||
3806 | /* | |
3807 | * Temporary pool names should never be written to disk. | |
3808 | */ | |
3809 | if (poolname != pool) | |
3810 | spa->spa_import_flags |= ZFS_IMPORT_TEMP_NAME; | |
3811 | ||
3812 | has_features = B_FALSE; | |
3813 | for (elem = nvlist_next_nvpair(props, NULL); | |
3814 | elem != NULL; elem = nvlist_next_nvpair(props, elem)) { | |
3815 | if (zpool_prop_feature(nvpair_name(elem))) | |
3816 | has_features = B_TRUE; | |
3817 | } | |
3818 | ||
3819 | if (has_features || nvlist_lookup_uint64(props, | |
3820 | zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) { | |
3821 | version = SPA_VERSION; | |
3822 | } | |
3823 | ASSERT(SPA_VERSION_IS_SUPPORTED(version)); | |
3824 | ||
3825 | spa->spa_first_txg = txg; | |
3826 | spa->spa_uberblock.ub_txg = txg - 1; | |
3827 | spa->spa_uberblock.ub_version = version; | |
3828 | spa->spa_ubsync = spa->spa_uberblock; | |
3829 | spa->spa_load_state = SPA_LOAD_CREATE; | |
3830 | ||
3831 | /* | |
3832 | * Create "The Godfather" zio to hold all async IOs | |
3833 | */ | |
3834 | spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), | |
3835 | KM_SLEEP); | |
3836 | for (i = 0; i < max_ncpus; i++) { | |
3837 | spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, | |
3838 | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | | |
3839 | ZIO_FLAG_GODFATHER); | |
3840 | } | |
3841 | ||
3842 | /* | |
3843 | * Create the root vdev. | |
3844 | */ | |
3845 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
3846 | ||
3847 | error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD); | |
3848 | ||
3849 | ASSERT(error != 0 || rvd != NULL); | |
3850 | ASSERT(error != 0 || spa->spa_root_vdev == rvd); | |
3851 | ||
3852 | if (error == 0 && !zfs_allocatable_devs(nvroot)) | |
3853 | error = SET_ERROR(EINVAL); | |
3854 | ||
3855 | if (error == 0 && | |
3856 | (error = vdev_create(rvd, txg, B_FALSE)) == 0 && | |
3857 | (error = spa_validate_aux(spa, nvroot, txg, | |
3858 | VDEV_ALLOC_ADD)) == 0) { | |
3859 | for (c = 0; c < rvd->vdev_children; c++) { | |
3860 | vdev_metaslab_set_size(rvd->vdev_child[c]); | |
3861 | vdev_expand(rvd->vdev_child[c], txg); | |
3862 | } | |
3863 | } | |
3864 | ||
3865 | spa_config_exit(spa, SCL_ALL, FTAG); | |
3866 | ||
3867 | if (error != 0) { | |
3868 | spa_unload(spa); | |
3869 | spa_deactivate(spa); | |
3870 | spa_remove(spa); | |
3871 | mutex_exit(&spa_namespace_lock); | |
3872 | return (error); | |
3873 | } | |
3874 | ||
3875 | /* | |
3876 | * Get the list of spares, if specified. | |
3877 | */ | |
3878 | if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, | |
3879 | &spares, &nspares) == 0) { | |
3880 | VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME, | |
3881 | KM_SLEEP) == 0); | |
3882 | VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, | |
3883 | ZPOOL_CONFIG_SPARES, spares, nspares) == 0); | |
3884 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
3885 | spa_load_spares(spa); | |
3886 | spa_config_exit(spa, SCL_ALL, FTAG); | |
3887 | spa->spa_spares.sav_sync = B_TRUE; | |
3888 | } | |
3889 | ||
3890 | /* | |
3891 | * Get the list of level 2 cache devices, if specified. | |
3892 | */ | |
3893 | if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, | |
3894 | &l2cache, &nl2cache) == 0) { | |
3895 | VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, | |
3896 | NV_UNIQUE_NAME, KM_SLEEP) == 0); | |
3897 | VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, | |
3898 | ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); | |
3899 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
3900 | spa_load_l2cache(spa); | |
3901 | spa_config_exit(spa, SCL_ALL, FTAG); | |
3902 | spa->spa_l2cache.sav_sync = B_TRUE; | |
3903 | } | |
3904 | ||
3905 | spa->spa_is_initializing = B_TRUE; | |
3906 | spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg); | |
3907 | spa->spa_meta_objset = dp->dp_meta_objset; | |
3908 | spa->spa_is_initializing = B_FALSE; | |
3909 | ||
3910 | /* | |
3911 | * Create DDTs (dedup tables). | |
3912 | */ | |
3913 | ddt_create(spa); | |
3914 | ||
3915 | spa_update_dspace(spa); | |
3916 | ||
3917 | tx = dmu_tx_create_assigned(dp, txg); | |
3918 | ||
3919 | /* | |
3920 | * Create the pool config object. | |
3921 | */ | |
3922 | spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset, | |
3923 | DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE, | |
3924 | DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx); | |
3925 | ||
3926 | if (zap_add(spa->spa_meta_objset, | |
3927 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG, | |
3928 | sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) { | |
3929 | cmn_err(CE_PANIC, "failed to add pool config"); | |
3930 | } | |
3931 | ||
3932 | if (spa_version(spa) >= SPA_VERSION_FEATURES) | |
3933 | spa_feature_create_zap_objects(spa, tx); | |
3934 | ||
3935 | if (zap_add(spa->spa_meta_objset, | |
3936 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION, | |
3937 | sizeof (uint64_t), 1, &version, tx) != 0) { | |
3938 | cmn_err(CE_PANIC, "failed to add pool version"); | |
3939 | } | |
3940 | ||
3941 | /* Newly created pools with the right version are always deflated. */ | |
3942 | if (version >= SPA_VERSION_RAIDZ_DEFLATE) { | |
3943 | spa->spa_deflate = TRUE; | |
3944 | if (zap_add(spa->spa_meta_objset, | |
3945 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, | |
3946 | sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) { | |
3947 | cmn_err(CE_PANIC, "failed to add deflate"); | |
3948 | } | |
3949 | } | |
3950 | ||
3951 | /* | |
3952 | * Create the deferred-free bpobj. Turn off compression | |
3953 | * because sync-to-convergence takes longer if the blocksize | |
3954 | * keeps changing. | |
3955 | */ | |
3956 | obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx); | |
3957 | dmu_object_set_compress(spa->spa_meta_objset, obj, | |
3958 | ZIO_COMPRESS_OFF, tx); | |
3959 | if (zap_add(spa->spa_meta_objset, | |
3960 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ, | |
3961 | sizeof (uint64_t), 1, &obj, tx) != 0) { | |
3962 | cmn_err(CE_PANIC, "failed to add bpobj"); | |
3963 | } | |
3964 | VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj, | |
3965 | spa->spa_meta_objset, obj)); | |
3966 | ||
3967 | /* | |
3968 | * Create the pool's history object. | |
3969 | */ | |
3970 | if (version >= SPA_VERSION_ZPOOL_HISTORY) | |
3971 | spa_history_create_obj(spa, tx); | |
3972 | ||
3973 | /* | |
3974 | * Generate some random noise for salted checksums to operate on. | |
3975 | */ | |
3976 | (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes, | |
3977 | sizeof (spa->spa_cksum_salt.zcs_bytes)); | |
3978 | ||
3979 | /* | |
3980 | * Set pool properties. | |
3981 | */ | |
3982 | spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS); | |
3983 | spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); | |
3984 | spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE); | |
3985 | spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND); | |
3986 | ||
3987 | if (props != NULL) { | |
3988 | spa_configfile_set(spa, props, B_FALSE); | |
3989 | spa_sync_props(props, tx); | |
3990 | } | |
3991 | ||
3992 | dmu_tx_commit(tx); | |
3993 | ||
3994 | spa->spa_sync_on = B_TRUE; | |
3995 | txg_sync_start(spa->spa_dsl_pool); | |
3996 | ||
3997 | /* | |
3998 | * We explicitly wait for the first transaction to complete so that our | |
3999 | * bean counters are appropriately updated. | |
4000 | */ | |
4001 | txg_wait_synced(spa->spa_dsl_pool, txg); | |
4002 | ||
4003 | spa_config_sync(spa, B_FALSE, B_TRUE); | |
4004 | spa_event_notify(spa, NULL, ESC_ZFS_POOL_CREATE); | |
4005 | ||
4006 | spa_history_log_version(spa, "create"); | |
4007 | ||
4008 | /* | |
4009 | * Don't count references from objsets that are already closed | |
4010 | * and are making their way through the eviction process. | |
4011 | */ | |
4012 | spa_evicting_os_wait(spa); | |
4013 | spa->spa_minref = refcount_count(&spa->spa_refcount); | |
4014 | spa->spa_load_state = SPA_LOAD_NONE; | |
4015 | ||
4016 | mutex_exit(&spa_namespace_lock); | |
4017 | ||
4018 | return (0); | |
4019 | } | |
4020 | ||
4021 | /* | |
4022 | * Import a non-root pool into the system. | |
4023 | */ | |
4024 | int | |
4025 | spa_import(char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags) | |
4026 | { | |
4027 | spa_t *spa; | |
4028 | char *altroot = NULL; | |
4029 | spa_load_state_t state = SPA_LOAD_IMPORT; | |
4030 | zpool_rewind_policy_t policy; | |
4031 | uint64_t mode = spa_mode_global; | |
4032 | uint64_t readonly = B_FALSE; | |
4033 | int error; | |
4034 | nvlist_t *nvroot; | |
4035 | nvlist_t **spares, **l2cache; | |
4036 | uint_t nspares, nl2cache; | |
4037 | ||
4038 | /* | |
4039 | * If a pool with this name exists, return failure. | |
4040 | */ | |
4041 | mutex_enter(&spa_namespace_lock); | |
4042 | if (spa_lookup(pool) != NULL) { | |
4043 | mutex_exit(&spa_namespace_lock); | |
4044 | return (SET_ERROR(EEXIST)); | |
4045 | } | |
4046 | ||
4047 | /* | |
4048 | * Create and initialize the spa structure. | |
4049 | */ | |
4050 | (void) nvlist_lookup_string(props, | |
4051 | zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); | |
4052 | (void) nvlist_lookup_uint64(props, | |
4053 | zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly); | |
4054 | if (readonly) | |
4055 | mode = FREAD; | |
4056 | spa = spa_add(pool, config, altroot); | |
4057 | spa->spa_import_flags = flags; | |
4058 | ||
4059 | /* | |
4060 | * Verbatim import - Take a pool and insert it into the namespace | |
4061 | * as if it had been loaded at boot. | |
4062 | */ | |
4063 | if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) { | |
4064 | if (props != NULL) | |
4065 | spa_configfile_set(spa, props, B_FALSE); | |
4066 | ||
4067 | spa_config_sync(spa, B_FALSE, B_TRUE); | |
4068 | spa_event_notify(spa, NULL, ESC_ZFS_POOL_IMPORT); | |
4069 | ||
4070 | mutex_exit(&spa_namespace_lock); | |
4071 | return (0); | |
4072 | } | |
4073 | ||
4074 | spa_activate(spa, mode); | |
4075 | ||
4076 | /* | |
4077 | * Don't start async tasks until we know everything is healthy. | |
4078 | */ | |
4079 | spa_async_suspend(spa); | |
4080 | ||
4081 | zpool_get_rewind_policy(config, &policy); | |
4082 | if (policy.zrp_request & ZPOOL_DO_REWIND) | |
4083 | state = SPA_LOAD_RECOVER; | |
4084 | ||
4085 | /* | |
4086 | * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig | |
4087 | * because the user-supplied config is actually the one to trust when | |
4088 | * doing an import. | |
4089 | */ | |
4090 | if (state != SPA_LOAD_RECOVER) | |
4091 | spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; | |
4092 | ||
4093 | error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg, | |
4094 | policy.zrp_request); | |
4095 | ||
4096 | /* | |
4097 | * Propagate anything learned while loading the pool and pass it | |
4098 | * back to caller (i.e. rewind info, missing devices, etc). | |
4099 | */ | |
4100 | VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, | |
4101 | spa->spa_load_info) == 0); | |
4102 | ||
4103 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
4104 | /* | |
4105 | * Toss any existing sparelist, as it doesn't have any validity | |
4106 | * anymore, and conflicts with spa_has_spare(). | |
4107 | */ | |
4108 | if (spa->spa_spares.sav_config) { | |
4109 | nvlist_free(spa->spa_spares.sav_config); | |
4110 | spa->spa_spares.sav_config = NULL; | |
4111 | spa_load_spares(spa); | |
4112 | } | |
4113 | if (spa->spa_l2cache.sav_config) { | |
4114 | nvlist_free(spa->spa_l2cache.sav_config); | |
4115 | spa->spa_l2cache.sav_config = NULL; | |
4116 | spa_load_l2cache(spa); | |
4117 | } | |
4118 | ||
4119 | VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, | |
4120 | &nvroot) == 0); | |
4121 | if (error == 0) | |
4122 | error = spa_validate_aux(spa, nvroot, -1ULL, | |
4123 | VDEV_ALLOC_SPARE); | |
4124 | if (error == 0) | |
4125 | error = spa_validate_aux(spa, nvroot, -1ULL, | |
4126 | VDEV_ALLOC_L2CACHE); | |
4127 | spa_config_exit(spa, SCL_ALL, FTAG); | |
4128 | ||
4129 | if (props != NULL) | |
4130 | spa_configfile_set(spa, props, B_FALSE); | |
4131 | ||
4132 | if (error != 0 || (props && spa_writeable(spa) && | |
4133 | (error = spa_prop_set(spa, props)))) { | |
4134 | spa_unload(spa); | |
4135 | spa_deactivate(spa); | |
4136 | spa_remove(spa); | |
4137 | mutex_exit(&spa_namespace_lock); | |
4138 | return (error); | |
4139 | } | |
4140 | ||
4141 | spa_async_resume(spa); | |
4142 | ||
4143 | /* | |
4144 | * Override any spares and level 2 cache devices as specified by | |
4145 | * the user, as these may have correct device names/devids, etc. | |
4146 | */ | |
4147 | if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, | |
4148 | &spares, &nspares) == 0) { | |
4149 | if (spa->spa_spares.sav_config) | |
4150 | VERIFY(nvlist_remove(spa->spa_spares.sav_config, | |
4151 | ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0); | |
4152 | else | |
4153 | VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, | |
4154 | NV_UNIQUE_NAME, KM_SLEEP) == 0); | |
4155 | VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, | |
4156 | ZPOOL_CONFIG_SPARES, spares, nspares) == 0); | |
4157 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
4158 | spa_load_spares(spa); | |
4159 | spa_config_exit(spa, SCL_ALL, FTAG); | |
4160 | spa->spa_spares.sav_sync = B_TRUE; | |
4161 | } | |
4162 | if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, | |
4163 | &l2cache, &nl2cache) == 0) { | |
4164 | if (spa->spa_l2cache.sav_config) | |
4165 | VERIFY(nvlist_remove(spa->spa_l2cache.sav_config, | |
4166 | ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0); | |
4167 | else | |
4168 | VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, | |
4169 | NV_UNIQUE_NAME, KM_SLEEP) == 0); | |
4170 | VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, | |
4171 | ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); | |
4172 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
4173 | spa_load_l2cache(spa); | |
4174 | spa_config_exit(spa, SCL_ALL, FTAG); | |
4175 | spa->spa_l2cache.sav_sync = B_TRUE; | |
4176 | } | |
4177 | ||
4178 | /* | |
4179 | * Check for any removed devices. | |
4180 | */ | |
4181 | if (spa->spa_autoreplace) { | |
4182 | spa_aux_check_removed(&spa->spa_spares); | |
4183 | spa_aux_check_removed(&spa->spa_l2cache); | |
4184 | } | |
4185 | ||
4186 | if (spa_writeable(spa)) { | |
4187 | /* | |
4188 | * Update the config cache to include the newly-imported pool. | |
4189 | */ | |
4190 | spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); | |
4191 | } | |
4192 | ||
4193 | /* | |
4194 | * It's possible that the pool was expanded while it was exported. | |
4195 | * We kick off an async task to handle this for us. | |
4196 | */ | |
4197 | spa_async_request(spa, SPA_ASYNC_AUTOEXPAND); | |
4198 | ||
4199 | spa_history_log_version(spa, "import"); | |
4200 | ||
4201 | spa_event_notify(spa, NULL, ESC_ZFS_POOL_IMPORT); | |
4202 | ||
4203 | zvol_create_minors(spa, pool, B_TRUE); | |
4204 | ||
4205 | mutex_exit(&spa_namespace_lock); | |
4206 | ||
4207 | return (0); | |
4208 | } | |
4209 | ||
4210 | nvlist_t * | |
4211 | spa_tryimport(nvlist_t *tryconfig) | |
4212 | { | |
4213 | nvlist_t *config = NULL; | |
4214 | char *poolname; | |
4215 | spa_t *spa; | |
4216 | uint64_t state; | |
4217 | int error; | |
4218 | ||
4219 | if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname)) | |
4220 | return (NULL); | |
4221 | ||
4222 | if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state)) | |
4223 | return (NULL); | |
4224 | ||
4225 | /* | |
4226 | * Create and initialize the spa structure. | |
4227 | */ | |
4228 | mutex_enter(&spa_namespace_lock); | |
4229 | spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL); | |
4230 | spa_activate(spa, FREAD); | |
4231 | ||
4232 | /* | |
4233 | * Pass off the heavy lifting to spa_load(). | |
4234 | * Pass TRUE for mosconfig because the user-supplied config | |
4235 | * is actually the one to trust when doing an import. | |
4236 | */ | |
4237 | error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE); | |
4238 | ||
4239 | /* | |
4240 | * If 'tryconfig' was at least parsable, return the current config. | |
4241 | */ | |
4242 | if (spa->spa_root_vdev != NULL) { | |
4243 | config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); | |
4244 | VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, | |
4245 | poolname) == 0); | |
4246 | VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, | |
4247 | state) == 0); | |
4248 | VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP, | |
4249 | spa->spa_uberblock.ub_timestamp) == 0); | |
4250 | VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, | |
4251 | spa->spa_load_info) == 0); | |
4252 | VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_ERRATA, | |
4253 | spa->spa_errata) == 0); | |
4254 | ||
4255 | /* | |
4256 | * If the bootfs property exists on this pool then we | |
4257 | * copy it out so that external consumers can tell which | |
4258 | * pools are bootable. | |
4259 | */ | |
4260 | if ((!error || error == EEXIST) && spa->spa_bootfs) { | |
4261 | char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP); | |
4262 | ||
4263 | /* | |
4264 | * We have to play games with the name since the | |
4265 | * pool was opened as TRYIMPORT_NAME. | |
4266 | */ | |
4267 | if (dsl_dsobj_to_dsname(spa_name(spa), | |
4268 | spa->spa_bootfs, tmpname) == 0) { | |
4269 | char *cp; | |
4270 | char *dsname; | |
4271 | ||
4272 | dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP); | |
4273 | ||
4274 | cp = strchr(tmpname, '/'); | |
4275 | if (cp == NULL) { | |
4276 | (void) strlcpy(dsname, tmpname, | |
4277 | MAXPATHLEN); | |
4278 | } else { | |
4279 | (void) snprintf(dsname, MAXPATHLEN, | |
4280 | "%s/%s", poolname, ++cp); | |
4281 | } | |
4282 | VERIFY(nvlist_add_string(config, | |
4283 | ZPOOL_CONFIG_BOOTFS, dsname) == 0); | |
4284 | kmem_free(dsname, MAXPATHLEN); | |
4285 | } | |
4286 | kmem_free(tmpname, MAXPATHLEN); | |
4287 | } | |
4288 | ||
4289 | /* | |
4290 | * Add the list of hot spares and level 2 cache devices. | |
4291 | */ | |
4292 | spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); | |
4293 | spa_add_spares(spa, config); | |
4294 | spa_add_l2cache(spa, config); | |
4295 | spa_config_exit(spa, SCL_CONFIG, FTAG); | |
4296 | } | |
4297 | ||
4298 | spa_unload(spa); | |
4299 | spa_deactivate(spa); | |
4300 | spa_remove(spa); | |
4301 | mutex_exit(&spa_namespace_lock); | |
4302 | ||
4303 | return (config); | |
4304 | } | |
4305 | ||
4306 | /* | |
4307 | * Pool export/destroy | |
4308 | * | |
4309 | * The act of destroying or exporting a pool is very simple. We make sure there | |
4310 | * is no more pending I/O and any references to the pool are gone. Then, we | |
4311 | * update the pool state and sync all the labels to disk, removing the | |
4312 | * configuration from the cache afterwards. If the 'hardforce' flag is set, then | |
4313 | * we don't sync the labels or remove the configuration cache. | |
4314 | */ | |
4315 | static int | |
4316 | spa_export_common(char *pool, int new_state, nvlist_t **oldconfig, | |
4317 | boolean_t force, boolean_t hardforce) | |
4318 | { | |
4319 | spa_t *spa; | |
4320 | ||
4321 | if (oldconfig) | |
4322 | *oldconfig = NULL; | |
4323 | ||
4324 | if (!(spa_mode_global & FWRITE)) | |
4325 | return (SET_ERROR(EROFS)); | |
4326 | ||
4327 | mutex_enter(&spa_namespace_lock); | |
4328 | if ((spa = spa_lookup(pool)) == NULL) { | |
4329 | mutex_exit(&spa_namespace_lock); | |
4330 | return (SET_ERROR(ENOENT)); | |
4331 | } | |
4332 | ||
4333 | /* | |
4334 | * Put a hold on the pool, drop the namespace lock, stop async tasks, | |
4335 | * reacquire the namespace lock, and see if we can export. | |
4336 | */ | |
4337 | spa_open_ref(spa, FTAG); | |
4338 | mutex_exit(&spa_namespace_lock); | |
4339 | spa_async_suspend(spa); | |
4340 | if (spa->spa_zvol_taskq) { | |
4341 | zvol_remove_minors(spa, spa_name(spa), B_TRUE); | |
4342 | taskq_wait(spa->spa_zvol_taskq); | |
4343 | } | |
4344 | mutex_enter(&spa_namespace_lock); | |
4345 | spa_close(spa, FTAG); | |
4346 | ||
4347 | if (spa->spa_state == POOL_STATE_UNINITIALIZED) | |
4348 | goto export_spa; | |
4349 | /* | |
4350 | * The pool will be in core if it's openable, in which case we can | |
4351 | * modify its state. Objsets may be open only because they're dirty, | |
4352 | * so we have to force it to sync before checking spa_refcnt. | |
4353 | */ | |
4354 | if (spa->spa_sync_on) { | |
4355 | txg_wait_synced(spa->spa_dsl_pool, 0); | |
4356 | spa_evicting_os_wait(spa); | |
4357 | } | |
4358 | ||
4359 | /* | |
4360 | * A pool cannot be exported or destroyed if there are active | |
4361 | * references. If we are resetting a pool, allow references by | |
4362 | * fault injection handlers. | |
4363 | */ | |
4364 | if (!spa_refcount_zero(spa) || | |
4365 | (spa->spa_inject_ref != 0 && | |
4366 | new_state != POOL_STATE_UNINITIALIZED)) { | |
4367 | spa_async_resume(spa); | |
4368 | mutex_exit(&spa_namespace_lock); | |
4369 | return (SET_ERROR(EBUSY)); | |
4370 | } | |
4371 | ||
4372 | if (spa->spa_sync_on) { | |
4373 | /* | |
4374 | * A pool cannot be exported if it has an active shared spare. | |
4375 | * This is to prevent other pools stealing the active spare | |
4376 | * from an exported pool. At user's own will, such pool can | |
4377 | * be forcedly exported. | |
4378 | */ | |
4379 | if (!force && new_state == POOL_STATE_EXPORTED && | |
4380 | spa_has_active_shared_spare(spa)) { | |
4381 | spa_async_resume(spa); | |
4382 | mutex_exit(&spa_namespace_lock); | |
4383 | return (SET_ERROR(EXDEV)); | |
4384 | } | |
4385 | ||
4386 | /* | |
4387 | * We want this to be reflected on every label, | |
4388 | * so mark them all dirty. spa_unload() will do the | |
4389 | * final sync that pushes these changes out. | |
4390 | */ | |
4391 | if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) { | |
4392 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
4393 | spa->spa_state = new_state; | |
4394 | spa->spa_final_txg = spa_last_synced_txg(spa) + | |
4395 | TXG_DEFER_SIZE + 1; | |
4396 | vdev_config_dirty(spa->spa_root_vdev); | |
4397 | spa_config_exit(spa, SCL_ALL, FTAG); | |
4398 | } | |
4399 | } | |
4400 | ||
4401 | export_spa: | |
4402 | spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY); | |
4403 | ||
4404 | if (spa->spa_state != POOL_STATE_UNINITIALIZED) { | |
4405 | spa_unload(spa); | |
4406 | spa_deactivate(spa); | |
4407 | } | |
4408 | ||
4409 | if (oldconfig && spa->spa_config) | |
4410 | VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0); | |
4411 | ||
4412 | if (new_state != POOL_STATE_UNINITIALIZED) { | |
4413 | if (!hardforce) | |
4414 | spa_config_sync(spa, B_TRUE, B_TRUE); | |
4415 | spa_remove(spa); | |
4416 | } | |
4417 | mutex_exit(&spa_namespace_lock); | |
4418 | ||
4419 | return (0); | |
4420 | } | |
4421 | ||
4422 | /* | |
4423 | * Destroy a storage pool. | |
4424 | */ | |
4425 | int | |
4426 | spa_destroy(char *pool) | |
4427 | { | |
4428 | return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL, | |
4429 | B_FALSE, B_FALSE)); | |
4430 | } | |
4431 | ||
4432 | /* | |
4433 | * Export a storage pool. | |
4434 | */ | |
4435 | int | |
4436 | spa_export(char *pool, nvlist_t **oldconfig, boolean_t force, | |
4437 | boolean_t hardforce) | |
4438 | { | |
4439 | return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig, | |
4440 | force, hardforce)); | |
4441 | } | |
4442 | ||
4443 | /* | |
4444 | * Similar to spa_export(), this unloads the spa_t without actually removing it | |
4445 | * from the namespace in any way. | |
4446 | */ | |
4447 | int | |
4448 | spa_reset(char *pool) | |
4449 | { | |
4450 | return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL, | |
4451 | B_FALSE, B_FALSE)); | |
4452 | } | |
4453 | ||
4454 | /* | |
4455 | * ========================================================================== | |
4456 | * Device manipulation | |
4457 | * ========================================================================== | |
4458 | */ | |
4459 | ||
4460 | /* | |
4461 | * Add a device to a storage pool. | |
4462 | */ | |
4463 | int | |
4464 | spa_vdev_add(spa_t *spa, nvlist_t *nvroot) | |
4465 | { | |
4466 | uint64_t txg, id; | |
4467 | int error; | |
4468 | vdev_t *rvd = spa->spa_root_vdev; | |
4469 | vdev_t *vd, *tvd; | |
4470 | nvlist_t **spares, **l2cache; | |
4471 | uint_t nspares, nl2cache; | |
4472 | int c; | |
4473 | ||
4474 | ASSERT(spa_writeable(spa)); | |
4475 | ||
4476 | txg = spa_vdev_enter(spa); | |
4477 | ||
4478 | if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0, | |
4479 | VDEV_ALLOC_ADD)) != 0) | |
4480 | return (spa_vdev_exit(spa, NULL, txg, error)); | |
4481 | ||
4482 | spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */ | |
4483 | ||
4484 | if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares, | |
4485 | &nspares) != 0) | |
4486 | nspares = 0; | |
4487 | ||
4488 | if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache, | |
4489 | &nl2cache) != 0) | |
4490 | nl2cache = 0; | |
4491 | ||
4492 | if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0) | |
4493 | return (spa_vdev_exit(spa, vd, txg, EINVAL)); | |
4494 | ||
4495 | if (vd->vdev_children != 0 && | |
4496 | (error = vdev_create(vd, txg, B_FALSE)) != 0) | |
4497 | return (spa_vdev_exit(spa, vd, txg, error)); | |
4498 | ||
4499 | /* | |
4500 | * We must validate the spares and l2cache devices after checking the | |
4501 | * children. Otherwise, vdev_inuse() will blindly overwrite the spare. | |
4502 | */ | |
4503 | if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0) | |
4504 | return (spa_vdev_exit(spa, vd, txg, error)); | |
4505 | ||
4506 | /* | |
4507 | * Transfer each new top-level vdev from vd to rvd. | |
4508 | */ | |
4509 | for (c = 0; c < vd->vdev_children; c++) { | |
4510 | ||
4511 | /* | |
4512 | * Set the vdev id to the first hole, if one exists. | |
4513 | */ | |
4514 | for (id = 0; id < rvd->vdev_children; id++) { | |
4515 | if (rvd->vdev_child[id]->vdev_ishole) { | |
4516 | vdev_free(rvd->vdev_child[id]); | |
4517 | break; | |
4518 | } | |
4519 | } | |
4520 | tvd = vd->vdev_child[c]; | |
4521 | vdev_remove_child(vd, tvd); | |
4522 | tvd->vdev_id = id; | |
4523 | vdev_add_child(rvd, tvd); | |
4524 | vdev_config_dirty(tvd); | |
4525 | } | |
4526 | ||
4527 | if (nspares != 0) { | |
4528 | spa_set_aux_vdevs(&spa->spa_spares, spares, nspares, | |
4529 | ZPOOL_CONFIG_SPARES); | |
4530 | spa_load_spares(spa); | |
4531 | spa->spa_spares.sav_sync = B_TRUE; | |
4532 | } | |
4533 | ||
4534 | if (nl2cache != 0) { | |
4535 | spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache, | |
4536 | ZPOOL_CONFIG_L2CACHE); | |
4537 | spa_load_l2cache(spa); | |
4538 | spa->spa_l2cache.sav_sync = B_TRUE; | |
4539 | } | |
4540 | ||
4541 | /* | |
4542 | * We have to be careful when adding new vdevs to an existing pool. | |
4543 | * If other threads start allocating from these vdevs before we | |
4544 | * sync the config cache, and we lose power, then upon reboot we may | |
4545 | * fail to open the pool because there are DVAs that the config cache | |
4546 | * can't translate. Therefore, we first add the vdevs without | |
4547 | * initializing metaslabs; sync the config cache (via spa_vdev_exit()); | |
4548 | * and then let spa_config_update() initialize the new metaslabs. | |
4549 | * | |
4550 | * spa_load() checks for added-but-not-initialized vdevs, so that | |
4551 | * if we lose power at any point in this sequence, the remaining | |
4552 | * steps will be completed the next time we load the pool. | |
4553 | */ | |
4554 | (void) spa_vdev_exit(spa, vd, txg, 0); | |
4555 | ||
4556 | mutex_enter(&spa_namespace_lock); | |
4557 | spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); | |
4558 | spa_event_notify(spa, NULL, ESC_ZFS_VDEV_ADD); | |
4559 | mutex_exit(&spa_namespace_lock); | |
4560 | ||
4561 | return (0); | |
4562 | } | |
4563 | ||
4564 | /* | |
4565 | * Attach a device to a mirror. The arguments are the path to any device | |
4566 | * in the mirror, and the nvroot for the new device. If the path specifies | |
4567 | * a device that is not mirrored, we automatically insert the mirror vdev. | |
4568 | * | |
4569 | * If 'replacing' is specified, the new device is intended to replace the | |
4570 | * existing device; in this case the two devices are made into their own | |
4571 | * mirror using the 'replacing' vdev, which is functionally identical to | |
4572 | * the mirror vdev (it actually reuses all the same ops) but has a few | |
4573 | * extra rules: you can't attach to it after it's been created, and upon | |
4574 | * completion of resilvering, the first disk (the one being replaced) | |
4575 | * is automatically detached. | |
4576 | */ | |
4577 | int | |
4578 | spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing) | |
4579 | { | |
4580 | uint64_t txg, dtl_max_txg; | |
4581 | vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd; | |
4582 | vdev_ops_t *pvops; | |
4583 | char *oldvdpath, *newvdpath; | |
4584 | int newvd_isspare; | |
4585 | int error; | |
4586 | ASSERTV(vdev_t *rvd = spa->spa_root_vdev); | |
4587 | ||
4588 | ASSERT(spa_writeable(spa)); | |
4589 | ||
4590 | txg = spa_vdev_enter(spa); | |
4591 | ||
4592 | oldvd = spa_lookup_by_guid(spa, guid, B_FALSE); | |
4593 | ||
4594 | if (oldvd == NULL) | |
4595 | return (spa_vdev_exit(spa, NULL, txg, ENODEV)); | |
4596 | ||
4597 | if (!oldvd->vdev_ops->vdev_op_leaf) | |
4598 | return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); | |
4599 | ||
4600 | pvd = oldvd->vdev_parent; | |
4601 | ||
4602 | if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0, | |
4603 | VDEV_ALLOC_ATTACH)) != 0) | |
4604 | return (spa_vdev_exit(spa, NULL, txg, EINVAL)); | |
4605 | ||
4606 | if (newrootvd->vdev_children != 1) | |
4607 | return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); | |
4608 | ||
4609 | newvd = newrootvd->vdev_child[0]; | |
4610 | ||
4611 | if (!newvd->vdev_ops->vdev_op_leaf) | |
4612 | return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); | |
4613 | ||
4614 | if ((error = vdev_create(newrootvd, txg, replacing)) != 0) | |
4615 | return (spa_vdev_exit(spa, newrootvd, txg, error)); | |
4616 | ||
4617 | /* | |
4618 | * Spares can't replace logs | |
4619 | */ | |
4620 | if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare) | |
4621 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); | |
4622 | ||
4623 | if (!replacing) { | |
4624 | /* | |
4625 | * For attach, the only allowable parent is a mirror or the root | |
4626 | * vdev. | |
4627 | */ | |
4628 | if (pvd->vdev_ops != &vdev_mirror_ops && | |
4629 | pvd->vdev_ops != &vdev_root_ops) | |
4630 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); | |
4631 | ||
4632 | pvops = &vdev_mirror_ops; | |
4633 | } else { | |
4634 | /* | |
4635 | * Active hot spares can only be replaced by inactive hot | |
4636 | * spares. | |
4637 | */ | |
4638 | if (pvd->vdev_ops == &vdev_spare_ops && | |
4639 | oldvd->vdev_isspare && | |
4640 | !spa_has_spare(spa, newvd->vdev_guid)) | |
4641 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); | |
4642 | ||
4643 | /* | |
4644 | * If the source is a hot spare, and the parent isn't already a | |
4645 | * spare, then we want to create a new hot spare. Otherwise, we | |
4646 | * want to create a replacing vdev. The user is not allowed to | |
4647 | * attach to a spared vdev child unless the 'isspare' state is | |
4648 | * the same (spare replaces spare, non-spare replaces | |
4649 | * non-spare). | |
4650 | */ | |
4651 | if (pvd->vdev_ops == &vdev_replacing_ops && | |
4652 | spa_version(spa) < SPA_VERSION_MULTI_REPLACE) { | |
4653 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); | |
4654 | } else if (pvd->vdev_ops == &vdev_spare_ops && | |
4655 | newvd->vdev_isspare != oldvd->vdev_isspare) { | |
4656 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); | |
4657 | } | |
4658 | ||
4659 | if (newvd->vdev_isspare) | |
4660 | pvops = &vdev_spare_ops; | |
4661 | else | |
4662 | pvops = &vdev_replacing_ops; | |
4663 | } | |
4664 | ||
4665 | /* | |
4666 | * Make sure the new device is big enough. | |
4667 | */ | |
4668 | if (newvd->vdev_asize < vdev_get_min_asize(oldvd)) | |
4669 | return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW)); | |
4670 | ||
4671 | /* | |
4672 | * The new device cannot have a higher alignment requirement | |
4673 | * than the top-level vdev. | |
4674 | */ | |
4675 | if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift) | |
4676 | return (spa_vdev_exit(spa, newrootvd, txg, EDOM)); | |
4677 | ||
4678 | /* | |
4679 | * If this is an in-place replacement, update oldvd's path and devid | |
4680 | * to make it distinguishable from newvd, and unopenable from now on. | |
4681 | */ | |
4682 | if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) { | |
4683 | spa_strfree(oldvd->vdev_path); | |
4684 | oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5, | |
4685 | KM_SLEEP); | |
4686 | (void) sprintf(oldvd->vdev_path, "%s/%s", | |
4687 | newvd->vdev_path, "old"); | |
4688 | if (oldvd->vdev_devid != NULL) { | |
4689 | spa_strfree(oldvd->vdev_devid); | |
4690 | oldvd->vdev_devid = NULL; | |
4691 | } | |
4692 | } | |
4693 | ||
4694 | /* mark the device being resilvered */ | |
4695 | newvd->vdev_resilver_txg = txg; | |
4696 | ||
4697 | /* | |
4698 | * If the parent is not a mirror, or if we're replacing, insert the new | |
4699 | * mirror/replacing/spare vdev above oldvd. | |
4700 | */ | |
4701 | if (pvd->vdev_ops != pvops) | |
4702 | pvd = vdev_add_parent(oldvd, pvops); | |
4703 | ||
4704 | ASSERT(pvd->vdev_top->vdev_parent == rvd); | |
4705 | ASSERT(pvd->vdev_ops == pvops); | |
4706 | ASSERT(oldvd->vdev_parent == pvd); | |
4707 | ||
4708 | /* | |
4709 | * Extract the new device from its root and add it to pvd. | |
4710 | */ | |
4711 | vdev_remove_child(newrootvd, newvd); | |
4712 | newvd->vdev_id = pvd->vdev_children; | |
4713 | newvd->vdev_crtxg = oldvd->vdev_crtxg; | |
4714 | vdev_add_child(pvd, newvd); | |
4715 | ||
4716 | tvd = newvd->vdev_top; | |
4717 | ASSERT(pvd->vdev_top == tvd); | |
4718 | ASSERT(tvd->vdev_parent == rvd); | |
4719 | ||
4720 | vdev_config_dirty(tvd); | |
4721 | ||
4722 | /* | |
4723 | * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account | |
4724 | * for any dmu_sync-ed blocks. It will propagate upward when | |
4725 | * spa_vdev_exit() calls vdev_dtl_reassess(). | |
4726 | */ | |
4727 | dtl_max_txg = txg + TXG_CONCURRENT_STATES; | |
4728 | ||
4729 | vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL, | |
4730 | dtl_max_txg - TXG_INITIAL); | |
4731 | ||
4732 | if (newvd->vdev_isspare) { | |
4733 | spa_spare_activate(newvd); | |
4734 | spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE); | |
4735 | } | |
4736 | ||
4737 | oldvdpath = spa_strdup(oldvd->vdev_path); | |
4738 | newvdpath = spa_strdup(newvd->vdev_path); | |
4739 | newvd_isspare = newvd->vdev_isspare; | |
4740 | ||
4741 | /* | |
4742 | * Mark newvd's DTL dirty in this txg. | |
4743 | */ | |
4744 | vdev_dirty(tvd, VDD_DTL, newvd, txg); | |
4745 | ||
4746 | /* | |
4747 | * Schedule the resilver to restart in the future. We do this to | |
4748 | * ensure that dmu_sync-ed blocks have been stitched into the | |
4749 | * respective datasets. | |
4750 | */ | |
4751 | dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg); | |
4752 | ||
4753 | if (spa->spa_bootfs) | |
4754 | spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH); | |
4755 | ||
4756 | spa_event_notify(spa, newvd, ESC_ZFS_VDEV_ATTACH); | |
4757 | ||
4758 | /* | |
4759 | * Commit the config | |
4760 | */ | |
4761 | (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0); | |
4762 | ||
4763 | spa_history_log_internal(spa, "vdev attach", NULL, | |
4764 | "%s vdev=%s %s vdev=%s", | |
4765 | replacing && newvd_isspare ? "spare in" : | |
4766 | replacing ? "replace" : "attach", newvdpath, | |
4767 | replacing ? "for" : "to", oldvdpath); | |
4768 | ||
4769 | spa_strfree(oldvdpath); | |
4770 | spa_strfree(newvdpath); | |
4771 | ||
4772 | return (0); | |
4773 | } | |
4774 | ||
4775 | /* | |
4776 | * Detach a device from a mirror or replacing vdev. | |
4777 | * | |
4778 | * If 'replace_done' is specified, only detach if the parent | |
4779 | * is a replacing vdev. | |
4780 | */ | |
4781 | int | |
4782 | spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done) | |
4783 | { | |
4784 | uint64_t txg; | |
4785 | int error; | |
4786 | vdev_t *vd, *pvd, *cvd, *tvd; | |
4787 | boolean_t unspare = B_FALSE; | |
4788 | uint64_t unspare_guid = 0; | |
4789 | char *vdpath; | |
4790 | int c, t; | |
4791 | ASSERTV(vdev_t *rvd = spa->spa_root_vdev); | |
4792 | ASSERT(spa_writeable(spa)); | |
4793 | ||
4794 | txg = spa_vdev_enter(spa); | |
4795 | ||
4796 | vd = spa_lookup_by_guid(spa, guid, B_FALSE); | |
4797 | ||
4798 | if (vd == NULL) | |
4799 | return (spa_vdev_exit(spa, NULL, txg, ENODEV)); | |
4800 | ||
4801 | if (!vd->vdev_ops->vdev_op_leaf) | |
4802 | return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); | |
4803 | ||
4804 | pvd = vd->vdev_parent; | |
4805 | ||
4806 | /* | |
4807 | * If the parent/child relationship is not as expected, don't do it. | |
4808 | * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing | |
4809 | * vdev that's replacing B with C. The user's intent in replacing | |
4810 | * is to go from M(A,B) to M(A,C). If the user decides to cancel | |
4811 | * the replace by detaching C, the expected behavior is to end up | |
4812 | * M(A,B). But suppose that right after deciding to detach C, | |
4813 | * the replacement of B completes. We would have M(A,C), and then | |
4814 | * ask to detach C, which would leave us with just A -- not what | |
4815 | * the user wanted. To prevent this, we make sure that the | |
4816 | * parent/child relationship hasn't changed -- in this example, | |
4817 | * that C's parent is still the replacing vdev R. | |
4818 | */ | |
4819 | if (pvd->vdev_guid != pguid && pguid != 0) | |
4820 | return (spa_vdev_exit(spa, NULL, txg, EBUSY)); | |
4821 | ||
4822 | /* | |
4823 | * Only 'replacing' or 'spare' vdevs can be replaced. | |
4824 | */ | |
4825 | if (replace_done && pvd->vdev_ops != &vdev_replacing_ops && | |
4826 | pvd->vdev_ops != &vdev_spare_ops) | |
4827 | return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); | |
4828 | ||
4829 | ASSERT(pvd->vdev_ops != &vdev_spare_ops || | |
4830 | spa_version(spa) >= SPA_VERSION_SPARES); | |
4831 | ||
4832 | /* | |
4833 | * Only mirror, replacing, and spare vdevs support detach. | |
4834 | */ | |
4835 | if (pvd->vdev_ops != &vdev_replacing_ops && | |
4836 | pvd->vdev_ops != &vdev_mirror_ops && | |
4837 | pvd->vdev_ops != &vdev_spare_ops) | |
4838 | return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); | |
4839 | ||
4840 | /* | |
4841 | * If this device has the only valid copy of some data, | |
4842 | * we cannot safely detach it. | |
4843 | */ | |
4844 | if (vdev_dtl_required(vd)) | |
4845 | return (spa_vdev_exit(spa, NULL, txg, EBUSY)); | |
4846 | ||
4847 | ASSERT(pvd->vdev_children >= 2); | |
4848 | ||
4849 | /* | |
4850 | * If we are detaching the second disk from a replacing vdev, then | |
4851 | * check to see if we changed the original vdev's path to have "/old" | |
4852 | * at the end in spa_vdev_attach(). If so, undo that change now. | |
4853 | */ | |
4854 | if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 && | |
4855 | vd->vdev_path != NULL) { | |
4856 | size_t len = strlen(vd->vdev_path); | |
4857 | ||
4858 | for (c = 0; c < pvd->vdev_children; c++) { | |
4859 | cvd = pvd->vdev_child[c]; | |
4860 | ||
4861 | if (cvd == vd || cvd->vdev_path == NULL) | |
4862 | continue; | |
4863 | ||
4864 | if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 && | |
4865 | strcmp(cvd->vdev_path + len, "/old") == 0) { | |
4866 | spa_strfree(cvd->vdev_path); | |
4867 | cvd->vdev_path = spa_strdup(vd->vdev_path); | |
4868 | break; | |
4869 | } | |
4870 | } | |
4871 | } | |
4872 | ||
4873 | /* | |
4874 | * If we are detaching the original disk from a spare, then it implies | |
4875 | * that the spare should become a real disk, and be removed from the | |
4876 | * active spare list for the pool. | |
4877 | */ | |
4878 | if (pvd->vdev_ops == &vdev_spare_ops && | |
4879 | vd->vdev_id == 0 && | |
4880 | pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare) | |
4881 | unspare = B_TRUE; | |
4882 | ||
4883 | /* | |
4884 | * Erase the disk labels so the disk can be used for other things. | |
4885 | * This must be done after all other error cases are handled, | |
4886 | * but before we disembowel vd (so we can still do I/O to it). | |
4887 | * But if we can't do it, don't treat the error as fatal -- | |
4888 | * it may be that the unwritability of the disk is the reason | |
4889 | * it's being detached! | |
4890 | */ | |
4891 | error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); | |
4892 | ||
4893 | /* | |
4894 | * Remove vd from its parent and compact the parent's children. | |
4895 | */ | |
4896 | vdev_remove_child(pvd, vd); | |
4897 | vdev_compact_children(pvd); | |
4898 | ||
4899 | /* | |
4900 | * Remember one of the remaining children so we can get tvd below. | |
4901 | */ | |
4902 | cvd = pvd->vdev_child[pvd->vdev_children - 1]; | |
4903 | ||
4904 | /* | |
4905 | * If we need to remove the remaining child from the list of hot spares, | |
4906 | * do it now, marking the vdev as no longer a spare in the process. | |
4907 | * We must do this before vdev_remove_parent(), because that can | |
4908 | * change the GUID if it creates a new toplevel GUID. For a similar | |
4909 | * reason, we must remove the spare now, in the same txg as the detach; | |
4910 | * otherwise someone could attach a new sibling, change the GUID, and | |
4911 | * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail. | |
4912 | */ | |
4913 | if (unspare) { | |
4914 | ASSERT(cvd->vdev_isspare); | |
4915 | spa_spare_remove(cvd); | |
4916 | unspare_guid = cvd->vdev_guid; | |
4917 | (void) spa_vdev_remove(spa, unspare_guid, B_TRUE); | |
4918 | cvd->vdev_unspare = B_TRUE; | |
4919 | } | |
4920 | ||
4921 | /* | |
4922 | * If the parent mirror/replacing vdev only has one child, | |
4923 | * the parent is no longer needed. Remove it from the tree. | |
4924 | */ | |
4925 | if (pvd->vdev_children == 1) { | |
4926 | if (pvd->vdev_ops == &vdev_spare_ops) | |
4927 | cvd->vdev_unspare = B_FALSE; | |
4928 | vdev_remove_parent(cvd); | |
4929 | } | |
4930 | ||
4931 | ||
4932 | /* | |
4933 | * We don't set tvd until now because the parent we just removed | |
4934 | * may have been the previous top-level vdev. | |
4935 | */ | |
4936 | tvd = cvd->vdev_top; | |
4937 | ASSERT(tvd->vdev_parent == rvd); | |
4938 | ||
4939 | /* | |
4940 | * Reevaluate the parent vdev state. | |
4941 | */ | |
4942 | vdev_propagate_state(cvd); | |
4943 | ||
4944 | /* | |
4945 | * If the 'autoexpand' property is set on the pool then automatically | |
4946 | * try to expand the size of the pool. For example if the device we | |
4947 | * just detached was smaller than the others, it may be possible to | |
4948 | * add metaslabs (i.e. grow the pool). We need to reopen the vdev | |
4949 | * first so that we can obtain the updated sizes of the leaf vdevs. | |
4950 | */ | |
4951 | if (spa->spa_autoexpand) { | |
4952 | vdev_reopen(tvd); | |
4953 | vdev_expand(tvd, txg); | |
4954 | } | |
4955 | ||
4956 | vdev_config_dirty(tvd); | |
4957 | ||
4958 | /* | |
4959 | * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that | |
4960 | * vd->vdev_detached is set and free vd's DTL object in syncing context. | |
4961 | * But first make sure we're not on any *other* txg's DTL list, to | |
4962 | * prevent vd from being accessed after it's freed. | |
4963 | */ | |
4964 | vdpath = spa_strdup(vd->vdev_path ? vd->vdev_path : "none"); | |
4965 | for (t = 0; t < TXG_SIZE; t++) | |
4966 | (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t); | |
4967 | vd->vdev_detached = B_TRUE; | |
4968 | vdev_dirty(tvd, VDD_DTL, vd, txg); | |
4969 | ||
4970 | spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE); | |
4971 | ||
4972 | /* hang on to the spa before we release the lock */ | |
4973 | spa_open_ref(spa, FTAG); | |
4974 | ||
4975 | error = spa_vdev_exit(spa, vd, txg, 0); | |
4976 | ||
4977 | spa_history_log_internal(spa, "detach", NULL, | |
4978 | "vdev=%s", vdpath); | |
4979 | spa_strfree(vdpath); | |
4980 | ||
4981 | /* | |
4982 | * If this was the removal of the original device in a hot spare vdev, | |
4983 | * then we want to go through and remove the device from the hot spare | |
4984 | * list of every other pool. | |
4985 | */ | |
4986 | if (unspare) { | |
4987 | spa_t *altspa = NULL; | |
4988 | ||
4989 | mutex_enter(&spa_namespace_lock); | |
4990 | while ((altspa = spa_next(altspa)) != NULL) { | |
4991 | if (altspa->spa_state != POOL_STATE_ACTIVE || | |
4992 | altspa == spa) | |
4993 | continue; | |
4994 | ||
4995 | spa_open_ref(altspa, FTAG); | |
4996 | mutex_exit(&spa_namespace_lock); | |
4997 | (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE); | |
4998 | mutex_enter(&spa_namespace_lock); | |
4999 | spa_close(altspa, FTAG); | |
5000 | } | |
5001 | mutex_exit(&spa_namespace_lock); | |
5002 | ||
5003 | /* search the rest of the vdevs for spares to remove */ | |
5004 | spa_vdev_resilver_done(spa); | |
5005 | } | |
5006 | ||
5007 | /* all done with the spa; OK to release */ | |
5008 | mutex_enter(&spa_namespace_lock); | |
5009 | spa_close(spa, FTAG); | |
5010 | mutex_exit(&spa_namespace_lock); | |
5011 | ||
5012 | return (error); | |
5013 | } | |
5014 | ||
5015 | /* | |
5016 | * Split a set of devices from their mirrors, and create a new pool from them. | |
5017 | */ | |
5018 | int | |
5019 | spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config, | |
5020 | nvlist_t *props, boolean_t exp) | |
5021 | { | |
5022 | int error = 0; | |
5023 | uint64_t txg, *glist; | |
5024 | spa_t *newspa; | |
5025 | uint_t c, children, lastlog; | |
5026 | nvlist_t **child, *nvl, *tmp; | |
5027 | dmu_tx_t *tx; | |
5028 | char *altroot = NULL; | |
5029 | vdev_t *rvd, **vml = NULL; /* vdev modify list */ | |
5030 | boolean_t activate_slog; | |
5031 | ||
5032 | ASSERT(spa_writeable(spa)); | |
5033 | ||
5034 | txg = spa_vdev_enter(spa); | |
5035 | ||
5036 | /* clear the log and flush everything up to now */ | |
5037 | activate_slog = spa_passivate_log(spa); | |
5038 | (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); | |
5039 | error = spa_offline_log(spa); | |
5040 | txg = spa_vdev_config_enter(spa); | |
5041 | ||
5042 | if (activate_slog) | |
5043 | spa_activate_log(spa); | |
5044 | ||
5045 | if (error != 0) | |
5046 | return (spa_vdev_exit(spa, NULL, txg, error)); | |
5047 | ||
5048 | /* check new spa name before going any further */ | |
5049 | if (spa_lookup(newname) != NULL) | |
5050 | return (spa_vdev_exit(spa, NULL, txg, EEXIST)); | |
5051 | ||
5052 | /* | |
5053 | * scan through all the children to ensure they're all mirrors | |
5054 | */ | |
5055 | if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 || | |
5056 | nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child, | |
5057 | &children) != 0) | |
5058 | return (spa_vdev_exit(spa, NULL, txg, EINVAL)); | |
5059 | ||
5060 | /* first, check to ensure we've got the right child count */ | |
5061 | rvd = spa->spa_root_vdev; | |
5062 | lastlog = 0; | |
5063 | for (c = 0; c < rvd->vdev_children; c++) { | |
5064 | vdev_t *vd = rvd->vdev_child[c]; | |
5065 | ||
5066 | /* don't count the holes & logs as children */ | |
5067 | if (vd->vdev_islog || vd->vdev_ishole) { | |
5068 | if (lastlog == 0) | |
5069 | lastlog = c; | |
5070 | continue; | |
5071 | } | |
5072 | ||
5073 | lastlog = 0; | |
5074 | } | |
5075 | if (children != (lastlog != 0 ? lastlog : rvd->vdev_children)) | |
5076 | return (spa_vdev_exit(spa, NULL, txg, EINVAL)); | |
5077 | ||
5078 | /* next, ensure no spare or cache devices are part of the split */ | |
5079 | if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 || | |
5080 | nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0) | |
5081 | return (spa_vdev_exit(spa, NULL, txg, EINVAL)); | |
5082 | ||
5083 | vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP); | |
5084 | glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP); | |
5085 | ||
5086 | /* then, loop over each vdev and validate it */ | |
5087 | for (c = 0; c < children; c++) { | |
5088 | uint64_t is_hole = 0; | |
5089 | ||
5090 | (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE, | |
5091 | &is_hole); | |
5092 | ||
5093 | if (is_hole != 0) { | |
5094 | if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole || | |
5095 | spa->spa_root_vdev->vdev_child[c]->vdev_islog) { | |
5096 | continue; | |
5097 | } else { | |
5098 | error = SET_ERROR(EINVAL); | |
5099 | break; | |
5100 | } | |
5101 | } | |
5102 | ||
5103 | /* which disk is going to be split? */ | |
5104 | if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID, | |
5105 | &glist[c]) != 0) { | |
5106 | error = SET_ERROR(EINVAL); | |
5107 | break; | |
5108 | } | |
5109 | ||
5110 | /* look it up in the spa */ | |
5111 | vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE); | |
5112 | if (vml[c] == NULL) { | |
5113 | error = SET_ERROR(ENODEV); | |
5114 | break; | |
5115 | } | |
5116 | ||
5117 | /* make sure there's nothing stopping the split */ | |
5118 | if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops || | |
5119 | vml[c]->vdev_islog || | |
5120 | vml[c]->vdev_ishole || | |
5121 | vml[c]->vdev_isspare || | |
5122 | vml[c]->vdev_isl2cache || | |
5123 | !vdev_writeable(vml[c]) || | |
5124 | vml[c]->vdev_children != 0 || | |
5125 | vml[c]->vdev_state != VDEV_STATE_HEALTHY || | |
5126 | c != spa->spa_root_vdev->vdev_child[c]->vdev_id) { | |
5127 | error = SET_ERROR(EINVAL); | |
5128 | break; | |
5129 | } | |
5130 | ||
5131 | if (vdev_dtl_required(vml[c])) { | |
5132 | error = SET_ERROR(EBUSY); | |
5133 | break; | |
5134 | } | |
5135 | ||
5136 | /* we need certain info from the top level */ | |
5137 | VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY, | |
5138 | vml[c]->vdev_top->vdev_ms_array) == 0); | |
5139 | VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT, | |
5140 | vml[c]->vdev_top->vdev_ms_shift) == 0); | |
5141 | VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE, | |
5142 | vml[c]->vdev_top->vdev_asize) == 0); | |
5143 | VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT, | |
5144 | vml[c]->vdev_top->vdev_ashift) == 0); | |
5145 | ||
5146 | /* transfer per-vdev ZAPs */ | |
5147 | ASSERT3U(vml[c]->vdev_leaf_zap, !=, 0); | |
5148 | VERIFY0(nvlist_add_uint64(child[c], | |
5149 | ZPOOL_CONFIG_VDEV_LEAF_ZAP, vml[c]->vdev_leaf_zap)); | |
5150 | ||
5151 | ASSERT3U(vml[c]->vdev_top->vdev_top_zap, !=, 0); | |
5152 | VERIFY0(nvlist_add_uint64(child[c], | |
5153 | ZPOOL_CONFIG_VDEV_TOP_ZAP, | |
5154 | vml[c]->vdev_parent->vdev_top_zap)); | |
5155 | } | |
5156 | ||
5157 | if (error != 0) { | |
5158 | kmem_free(vml, children * sizeof (vdev_t *)); | |
5159 | kmem_free(glist, children * sizeof (uint64_t)); | |
5160 | return (spa_vdev_exit(spa, NULL, txg, error)); | |
5161 | } | |
5162 | ||
5163 | /* stop writers from using the disks */ | |
5164 | for (c = 0; c < children; c++) { | |
5165 | if (vml[c] != NULL) | |
5166 | vml[c]->vdev_offline = B_TRUE; | |
5167 | } | |
5168 | vdev_reopen(spa->spa_root_vdev); | |
5169 | ||
5170 | /* | |
5171 | * Temporarily record the splitting vdevs in the spa config. This | |
5172 | * will disappear once the config is regenerated. | |
5173 | */ | |
5174 | VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0); | |
5175 | VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, | |
5176 | glist, children) == 0); | |
5177 | kmem_free(glist, children * sizeof (uint64_t)); | |
5178 | ||
5179 | mutex_enter(&spa->spa_props_lock); | |
5180 | VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT, | |
5181 | nvl) == 0); | |
5182 | mutex_exit(&spa->spa_props_lock); | |
5183 | spa->spa_config_splitting = nvl; | |
5184 | vdev_config_dirty(spa->spa_root_vdev); | |
5185 | ||
5186 | /* configure and create the new pool */ | |
5187 | VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0); | |
5188 | VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, | |
5189 | exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0); | |
5190 | VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, | |
5191 | spa_version(spa)) == 0); | |
5192 | VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, | |
5193 | spa->spa_config_txg) == 0); | |
5194 | VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, | |
5195 | spa_generate_guid(NULL)) == 0); | |
5196 | VERIFY0(nvlist_add_boolean(config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)); | |
5197 | (void) nvlist_lookup_string(props, | |
5198 | zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); | |
5199 | ||
5200 | /* add the new pool to the namespace */ | |
5201 | newspa = spa_add(newname, config, altroot); | |
5202 | newspa->spa_avz_action = AVZ_ACTION_REBUILD; | |
5203 | newspa->spa_config_txg = spa->spa_config_txg; | |
5204 | spa_set_log_state(newspa, SPA_LOG_CLEAR); | |
5205 | ||
5206 | /* release the spa config lock, retaining the namespace lock */ | |
5207 | spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); | |
5208 | ||
5209 | if (zio_injection_enabled) | |
5210 | zio_handle_panic_injection(spa, FTAG, 1); | |
5211 | ||
5212 | spa_activate(newspa, spa_mode_global); | |
5213 | spa_async_suspend(newspa); | |
5214 | ||
5215 | /* create the new pool from the disks of the original pool */ | |
5216 | error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE); | |
5217 | if (error) | |
5218 | goto out; | |
5219 | ||
5220 | /* if that worked, generate a real config for the new pool */ | |
5221 | if (newspa->spa_root_vdev != NULL) { | |
5222 | VERIFY(nvlist_alloc(&newspa->spa_config_splitting, | |
5223 | NV_UNIQUE_NAME, KM_SLEEP) == 0); | |
5224 | VERIFY(nvlist_add_uint64(newspa->spa_config_splitting, | |
5225 | ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0); | |
5226 | spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL, | |
5227 | B_TRUE)); | |
5228 | } | |
5229 | ||
5230 | /* set the props */ | |
5231 | if (props != NULL) { | |
5232 | spa_configfile_set(newspa, props, B_FALSE); | |
5233 | error = spa_prop_set(newspa, props); | |
5234 | if (error) | |
5235 | goto out; | |
5236 | } | |
5237 | ||
5238 | /* flush everything */ | |
5239 | txg = spa_vdev_config_enter(newspa); | |
5240 | vdev_config_dirty(newspa->spa_root_vdev); | |
5241 | (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG); | |
5242 | ||
5243 | if (zio_injection_enabled) | |
5244 | zio_handle_panic_injection(spa, FTAG, 2); | |
5245 | ||
5246 | spa_async_resume(newspa); | |
5247 | ||
5248 | /* finally, update the original pool's config */ | |
5249 | txg = spa_vdev_config_enter(spa); | |
5250 | tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir); | |
5251 | error = dmu_tx_assign(tx, TXG_WAIT); | |
5252 | if (error != 0) | |
5253 | dmu_tx_abort(tx); | |
5254 | for (c = 0; c < children; c++) { | |
5255 | if (vml[c] != NULL) { | |
5256 | vdev_split(vml[c]); | |
5257 | if (error == 0) | |
5258 | spa_history_log_internal(spa, "detach", tx, | |
5259 | "vdev=%s", vml[c]->vdev_path); | |
5260 | ||
5261 | vdev_free(vml[c]); | |
5262 | } | |
5263 | } | |
5264 | spa->spa_avz_action = AVZ_ACTION_REBUILD; | |
5265 | vdev_config_dirty(spa->spa_root_vdev); | |
5266 | spa->spa_config_splitting = NULL; | |
5267 | nvlist_free(nvl); | |
5268 | if (error == 0) | |
5269 | dmu_tx_commit(tx); | |
5270 | (void) spa_vdev_exit(spa, NULL, txg, 0); | |
5271 | ||
5272 | if (zio_injection_enabled) | |
5273 | zio_handle_panic_injection(spa, FTAG, 3); | |
5274 | ||
5275 | /* split is complete; log a history record */ | |
5276 | spa_history_log_internal(newspa, "split", NULL, | |
5277 | "from pool %s", spa_name(spa)); | |
5278 | ||
5279 | kmem_free(vml, children * sizeof (vdev_t *)); | |
5280 | ||
5281 | /* if we're not going to mount the filesystems in userland, export */ | |
5282 | if (exp) | |
5283 | error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL, | |
5284 | B_FALSE, B_FALSE); | |
5285 | ||
5286 | return (error); | |
5287 | ||
5288 | out: | |
5289 | spa_unload(newspa); | |
5290 | spa_deactivate(newspa); | |
5291 | spa_remove(newspa); | |
5292 | ||
5293 | txg = spa_vdev_config_enter(spa); | |
5294 | ||
5295 | /* re-online all offlined disks */ | |
5296 | for (c = 0; c < children; c++) { | |
5297 | if (vml[c] != NULL) | |
5298 | vml[c]->vdev_offline = B_FALSE; | |
5299 | } | |
5300 | vdev_reopen(spa->spa_root_vdev); | |
5301 | ||
5302 | nvlist_free(spa->spa_config_splitting); | |
5303 | spa->spa_config_splitting = NULL; | |
5304 | (void) spa_vdev_exit(spa, NULL, txg, error); | |
5305 | ||
5306 | kmem_free(vml, children * sizeof (vdev_t *)); | |
5307 | return (error); | |
5308 | } | |
5309 | ||
5310 | static nvlist_t * | |
5311 | spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid) | |
5312 | { | |
5313 | int i; | |
5314 | ||
5315 | for (i = 0; i < count; i++) { | |
5316 | uint64_t guid; | |
5317 | ||
5318 | VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID, | |
5319 | &guid) == 0); | |
5320 | ||
5321 | if (guid == target_guid) | |
5322 | return (nvpp[i]); | |
5323 | } | |
5324 | ||
5325 | return (NULL); | |
5326 | } | |
5327 | ||
5328 | static void | |
5329 | spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count, | |
5330 | nvlist_t *dev_to_remove) | |
5331 | { | |
5332 | nvlist_t **newdev = NULL; | |
5333 | int i, j; | |
5334 | ||
5335 | if (count > 1) | |
5336 | newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP); | |
5337 | ||
5338 | for (i = 0, j = 0; i < count; i++) { | |
5339 | if (dev[i] == dev_to_remove) | |
5340 | continue; | |
5341 | VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0); | |
5342 | } | |
5343 | ||
5344 | VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0); | |
5345 | VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0); | |
5346 | ||
5347 | for (i = 0; i < count - 1; i++) | |
5348 | nvlist_free(newdev[i]); | |
5349 | ||
5350 | if (count > 1) | |
5351 | kmem_free(newdev, (count - 1) * sizeof (void *)); | |
5352 | } | |
5353 | ||
5354 | /* | |
5355 | * Evacuate the device. | |
5356 | */ | |
5357 | static int | |
5358 | spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd) | |
5359 | { | |
5360 | uint64_t txg; | |
5361 | int error = 0; | |
5362 | ||
5363 | ASSERT(MUTEX_HELD(&spa_namespace_lock)); | |
5364 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); | |
5365 | ASSERT(vd == vd->vdev_top); | |
5366 | ||
5367 | /* | |
5368 | * Evacuate the device. We don't hold the config lock as writer | |
5369 | * since we need to do I/O but we do keep the | |
5370 | * spa_namespace_lock held. Once this completes the device | |
5371 | * should no longer have any blocks allocated on it. | |
5372 | */ | |
5373 | if (vd->vdev_islog) { | |
5374 | if (vd->vdev_stat.vs_alloc != 0) | |
5375 | error = spa_offline_log(spa); | |
5376 | } else { | |
5377 | error = SET_ERROR(ENOTSUP); | |
5378 | } | |
5379 | ||
5380 | if (error) | |
5381 | return (error); | |
5382 | ||
5383 | /* | |
5384 | * The evacuation succeeded. Remove any remaining MOS metadata | |
5385 | * associated with this vdev, and wait for these changes to sync. | |
5386 | */ | |
5387 | ASSERT0(vd->vdev_stat.vs_alloc); | |
5388 | txg = spa_vdev_config_enter(spa); | |
5389 | vd->vdev_removing = B_TRUE; | |
5390 | vdev_dirty_leaves(vd, VDD_DTL, txg); | |
5391 | vdev_config_dirty(vd); | |
5392 | spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); | |
5393 | ||
5394 | return (0); | |
5395 | } | |
5396 | ||
5397 | /* | |
5398 | * Complete the removal by cleaning up the namespace. | |
5399 | */ | |
5400 | static void | |
5401 | spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd) | |
5402 | { | |
5403 | vdev_t *rvd = spa->spa_root_vdev; | |
5404 | uint64_t id = vd->vdev_id; | |
5405 | boolean_t last_vdev = (id == (rvd->vdev_children - 1)); | |
5406 | ||
5407 | ASSERT(MUTEX_HELD(&spa_namespace_lock)); | |
5408 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); | |
5409 | ASSERT(vd == vd->vdev_top); | |
5410 | ||
5411 | /* | |
5412 | * Only remove any devices which are empty. | |
5413 | */ | |
5414 | if (vd->vdev_stat.vs_alloc != 0) | |
5415 | return; | |
5416 | ||
5417 | (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); | |
5418 | ||
5419 | if (list_link_active(&vd->vdev_state_dirty_node)) | |
5420 | vdev_state_clean(vd); | |
5421 | if (list_link_active(&vd->vdev_config_dirty_node)) | |
5422 | vdev_config_clean(vd); | |
5423 | ||
5424 | vdev_free(vd); | |
5425 | ||
5426 | if (last_vdev) { | |
5427 | vdev_compact_children(rvd); | |
5428 | } else { | |
5429 | vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops); | |
5430 | vdev_add_child(rvd, vd); | |
5431 | } | |
5432 | vdev_config_dirty(rvd); | |
5433 | ||
5434 | /* | |
5435 | * Reassess the health of our root vdev. | |
5436 | */ | |
5437 | vdev_reopen(rvd); | |
5438 | } | |
5439 | ||
5440 | /* | |
5441 | * Remove a device from the pool - | |
5442 | * | |
5443 | * Removing a device from the vdev namespace requires several steps | |
5444 | * and can take a significant amount of time. As a result we use | |
5445 | * the spa_vdev_config_[enter/exit] functions which allow us to | |
5446 | * grab and release the spa_config_lock while still holding the namespace | |
5447 | * lock. During each step the configuration is synced out. | |
5448 | * | |
5449 | * Currently, this supports removing only hot spares, slogs, and level 2 ARC | |
5450 | * devices. | |
5451 | */ | |
5452 | int | |
5453 | spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare) | |
5454 | { | |
5455 | vdev_t *vd; | |
5456 | metaslab_group_t *mg; | |
5457 | nvlist_t **spares, **l2cache, *nv; | |
5458 | uint64_t txg = 0; | |
5459 | uint_t nspares, nl2cache; | |
5460 | int error = 0; | |
5461 | boolean_t locked = MUTEX_HELD(&spa_namespace_lock); | |
5462 | ||
5463 | ASSERT(spa_writeable(spa)); | |
5464 | ||
5465 | if (!locked) | |
5466 | txg = spa_vdev_enter(spa); | |
5467 | ||
5468 | vd = spa_lookup_by_guid(spa, guid, B_FALSE); | |
5469 | ||
5470 | if (spa->spa_spares.sav_vdevs != NULL && | |
5471 | nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, | |
5472 | ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 && | |
5473 | (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) { | |
5474 | /* | |
5475 | * Only remove the hot spare if it's not currently in use | |
5476 | * in this pool. | |
5477 | */ | |
5478 | if (vd == NULL || unspare) { | |
5479 | spa_vdev_remove_aux(spa->spa_spares.sav_config, | |
5480 | ZPOOL_CONFIG_SPARES, spares, nspares, nv); | |
5481 | spa_load_spares(spa); | |
5482 | spa->spa_spares.sav_sync = B_TRUE; | |
5483 | } else { | |
5484 | error = SET_ERROR(EBUSY); | |
5485 | } | |
5486 | spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE_AUX); | |
5487 | } else if (spa->spa_l2cache.sav_vdevs != NULL && | |
5488 | nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, | |
5489 | ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 && | |
5490 | (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) { | |
5491 | /* | |
5492 | * Cache devices can always be removed. | |
5493 | */ | |
5494 | spa_vdev_remove_aux(spa->spa_l2cache.sav_config, | |
5495 | ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv); | |
5496 | spa_load_l2cache(spa); | |
5497 | spa->spa_l2cache.sav_sync = B_TRUE; | |
5498 | spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE_AUX); | |
5499 | } else if (vd != NULL && vd->vdev_islog) { | |
5500 | ASSERT(!locked); | |
5501 | ASSERT(vd == vd->vdev_top); | |
5502 | ||
5503 | mg = vd->vdev_mg; | |
5504 | ||
5505 | /* | |
5506 | * Stop allocating from this vdev. | |
5507 | */ | |
5508 | metaslab_group_passivate(mg); | |
5509 | ||
5510 | /* | |
5511 | * Wait for the youngest allocations and frees to sync, | |
5512 | * and then wait for the deferral of those frees to finish. | |
5513 | */ | |
5514 | spa_vdev_config_exit(spa, NULL, | |
5515 | txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG); | |
5516 | ||
5517 | /* | |
5518 | * Attempt to evacuate the vdev. | |
5519 | */ | |
5520 | error = spa_vdev_remove_evacuate(spa, vd); | |
5521 | ||
5522 | txg = spa_vdev_config_enter(spa); | |
5523 | ||
5524 | /* | |
5525 | * If we couldn't evacuate the vdev, unwind. | |
5526 | */ | |
5527 | if (error) { | |
5528 | metaslab_group_activate(mg); | |
5529 | return (spa_vdev_exit(spa, NULL, txg, error)); | |
5530 | } | |
5531 | ||
5532 | /* | |
5533 | * Clean up the vdev namespace. | |
5534 | */ | |
5535 | spa_vdev_remove_from_namespace(spa, vd); | |
5536 | ||
5537 | spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE_DEV); | |
5538 | } else if (vd != NULL) { | |
5539 | /* | |
5540 | * Normal vdevs cannot be removed (yet). | |
5541 | */ | |
5542 | error = SET_ERROR(ENOTSUP); | |
5543 | } else { | |
5544 | /* | |
5545 | * There is no vdev of any kind with the specified guid. | |
5546 | */ | |
5547 | error = SET_ERROR(ENOENT); | |
5548 | } | |
5549 | ||
5550 | if (!locked) | |
5551 | error = spa_vdev_exit(spa, NULL, txg, error); | |
5552 | ||
5553 | return (error); | |
5554 | } | |
5555 | ||
5556 | /* | |
5557 | * Find any device that's done replacing, or a vdev marked 'unspare' that's | |
5558 | * currently spared, so we can detach it. | |
5559 | */ | |
5560 | static vdev_t * | |
5561 | spa_vdev_resilver_done_hunt(vdev_t *vd) | |
5562 | { | |
5563 | vdev_t *newvd, *oldvd; | |
5564 | int c; | |
5565 | ||
5566 | for (c = 0; c < vd->vdev_children; c++) { | |
5567 | oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]); | |
5568 | if (oldvd != NULL) | |
5569 | return (oldvd); | |
5570 | } | |
5571 | ||
5572 | /* | |
5573 | * Check for a completed replacement. We always consider the first | |
5574 | * vdev in the list to be the oldest vdev, and the last one to be | |
5575 | * the newest (see spa_vdev_attach() for how that works). In | |
5576 | * the case where the newest vdev is faulted, we will not automatically | |
5577 | * remove it after a resilver completes. This is OK as it will require | |
5578 | * user intervention to determine which disk the admin wishes to keep. | |
5579 | */ | |
5580 | if (vd->vdev_ops == &vdev_replacing_ops) { | |
5581 | ASSERT(vd->vdev_children > 1); | |
5582 | ||
5583 | newvd = vd->vdev_child[vd->vdev_children - 1]; | |
5584 | oldvd = vd->vdev_child[0]; | |
5585 | ||
5586 | if (vdev_dtl_empty(newvd, DTL_MISSING) && | |
5587 | vdev_dtl_empty(newvd, DTL_OUTAGE) && | |
5588 | !vdev_dtl_required(oldvd)) | |
5589 | return (oldvd); | |
5590 | } | |
5591 | ||
5592 | /* | |
5593 | * Check for a completed resilver with the 'unspare' flag set. | |
5594 | */ | |
5595 | if (vd->vdev_ops == &vdev_spare_ops) { | |
5596 | vdev_t *first = vd->vdev_child[0]; | |
5597 | vdev_t *last = vd->vdev_child[vd->vdev_children - 1]; | |
5598 | ||
5599 | if (last->vdev_unspare) { | |
5600 | oldvd = first; | |
5601 | newvd = last; | |
5602 | } else if (first->vdev_unspare) { | |
5603 | oldvd = last; | |
5604 | newvd = first; | |
5605 | } else { | |
5606 | oldvd = NULL; | |
5607 | } | |
5608 | ||
5609 | if (oldvd != NULL && | |
5610 | vdev_dtl_empty(newvd, DTL_MISSING) && | |
5611 | vdev_dtl_empty(newvd, DTL_OUTAGE) && | |
5612 | !vdev_dtl_required(oldvd)) | |
5613 | return (oldvd); | |
5614 | ||
5615 | /* | |
5616 | * If there are more than two spares attached to a disk, | |
5617 | * and those spares are not required, then we want to | |
5618 | * attempt to free them up now so that they can be used | |
5619 | * by other pools. Once we're back down to a single | |
5620 | * disk+spare, we stop removing them. | |
5621 | */ | |
5622 | if (vd->vdev_children > 2) { | |
5623 | newvd = vd->vdev_child[1]; | |
5624 | ||
5625 | if (newvd->vdev_isspare && last->vdev_isspare && | |
5626 | vdev_dtl_empty(last, DTL_MISSING) && | |
5627 | vdev_dtl_empty(last, DTL_OUTAGE) && | |
5628 | !vdev_dtl_required(newvd)) | |
5629 | return (newvd); | |
5630 | } | |
5631 | } | |
5632 | ||
5633 | return (NULL); | |
5634 | } | |
5635 | ||
5636 | static void | |
5637 | spa_vdev_resilver_done(spa_t *spa) | |
5638 | { | |
5639 | vdev_t *vd, *pvd, *ppvd; | |
5640 | uint64_t guid, sguid, pguid, ppguid; | |
5641 | ||
5642 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
5643 | ||
5644 | while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) { | |
5645 | pvd = vd->vdev_parent; | |
5646 | ppvd = pvd->vdev_parent; | |
5647 | guid = vd->vdev_guid; | |
5648 | pguid = pvd->vdev_guid; | |
5649 | ppguid = ppvd->vdev_guid; | |
5650 | sguid = 0; | |
5651 | /* | |
5652 | * If we have just finished replacing a hot spared device, then | |
5653 | * we need to detach the parent's first child (the original hot | |
5654 | * spare) as well. | |
5655 | */ | |
5656 | if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 && | |
5657 | ppvd->vdev_children == 2) { | |
5658 | ASSERT(pvd->vdev_ops == &vdev_replacing_ops); | |
5659 | sguid = ppvd->vdev_child[1]->vdev_guid; | |
5660 | } | |
5661 | ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd)); | |
5662 | ||
5663 | spa_config_exit(spa, SCL_ALL, FTAG); | |
5664 | if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0) | |
5665 | return; | |
5666 | if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0) | |
5667 | return; | |
5668 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
5669 | } | |
5670 | ||
5671 | spa_config_exit(spa, SCL_ALL, FTAG); | |
5672 | } | |
5673 | ||
5674 | /* | |
5675 | * Update the stored path or FRU for this vdev. | |
5676 | */ | |
5677 | int | |
5678 | spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value, | |
5679 | boolean_t ispath) | |
5680 | { | |
5681 | vdev_t *vd; | |
5682 | boolean_t sync = B_FALSE; | |
5683 | ||
5684 | ASSERT(spa_writeable(spa)); | |
5685 | ||
5686 | spa_vdev_state_enter(spa, SCL_ALL); | |
5687 | ||
5688 | if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) | |
5689 | return (spa_vdev_state_exit(spa, NULL, ENOENT)); | |
5690 | ||
5691 | if (!vd->vdev_ops->vdev_op_leaf) | |
5692 | return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); | |
5693 | ||
5694 | if (ispath) { | |
5695 | if (strcmp(value, vd->vdev_path) != 0) { | |
5696 | spa_strfree(vd->vdev_path); | |
5697 | vd->vdev_path = spa_strdup(value); | |
5698 | sync = B_TRUE; | |
5699 | } | |
5700 | } else { | |
5701 | if (vd->vdev_fru == NULL) { | |
5702 | vd->vdev_fru = spa_strdup(value); | |
5703 | sync = B_TRUE; | |
5704 | } else if (strcmp(value, vd->vdev_fru) != 0) { | |
5705 | spa_strfree(vd->vdev_fru); | |
5706 | vd->vdev_fru = spa_strdup(value); | |
5707 | sync = B_TRUE; | |
5708 | } | |
5709 | } | |
5710 | ||
5711 | return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0)); | |
5712 | } | |
5713 | ||
5714 | int | |
5715 | spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath) | |
5716 | { | |
5717 | return (spa_vdev_set_common(spa, guid, newpath, B_TRUE)); | |
5718 | } | |
5719 | ||
5720 | int | |
5721 | spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru) | |
5722 | { | |
5723 | return (spa_vdev_set_common(spa, guid, newfru, B_FALSE)); | |
5724 | } | |
5725 | ||
5726 | /* | |
5727 | * ========================================================================== | |
5728 | * SPA Scanning | |
5729 | * ========================================================================== | |
5730 | */ | |
5731 | ||
5732 | int | |
5733 | spa_scan_stop(spa_t *spa) | |
5734 | { | |
5735 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); | |
5736 | if (dsl_scan_resilvering(spa->spa_dsl_pool)) | |
5737 | return (SET_ERROR(EBUSY)); | |
5738 | return (dsl_scan_cancel(spa->spa_dsl_pool)); | |
5739 | } | |
5740 | ||
5741 | int | |
5742 | spa_scan(spa_t *spa, pool_scan_func_t func) | |
5743 | { | |
5744 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); | |
5745 | ||
5746 | if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE) | |
5747 | return (SET_ERROR(ENOTSUP)); | |
5748 | ||
5749 | /* | |
5750 | * If a resilver was requested, but there is no DTL on a | |
5751 | * writeable leaf device, we have nothing to do. | |
5752 | */ | |
5753 | if (func == POOL_SCAN_RESILVER && | |
5754 | !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) { | |
5755 | spa_async_request(spa, SPA_ASYNC_RESILVER_DONE); | |
5756 | return (0); | |
5757 | } | |
5758 | ||
5759 | return (dsl_scan(spa->spa_dsl_pool, func)); | |
5760 | } | |
5761 | ||
5762 | /* | |
5763 | * ========================================================================== | |
5764 | * SPA async task processing | |
5765 | * ========================================================================== | |
5766 | */ | |
5767 | ||
5768 | static void | |
5769 | spa_async_remove(spa_t *spa, vdev_t *vd) | |
5770 | { | |
5771 | int c; | |
5772 | ||
5773 | if (vd->vdev_remove_wanted) { | |
5774 | vd->vdev_remove_wanted = B_FALSE; | |
5775 | vd->vdev_delayed_close = B_FALSE; | |
5776 | vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE); | |
5777 | ||
5778 | /* | |
5779 | * We want to clear the stats, but we don't want to do a full | |
5780 | * vdev_clear() as that will cause us to throw away | |
5781 | * degraded/faulted state as well as attempt to reopen the | |
5782 | * device, all of which is a waste. | |
5783 | */ | |
5784 | vd->vdev_stat.vs_read_errors = 0; | |
5785 | vd->vdev_stat.vs_write_errors = 0; | |
5786 | vd->vdev_stat.vs_checksum_errors = 0; | |
5787 | ||
5788 | vdev_state_dirty(vd->vdev_top); | |
5789 | } | |
5790 | ||
5791 | for (c = 0; c < vd->vdev_children; c++) | |
5792 | spa_async_remove(spa, vd->vdev_child[c]); | |
5793 | } | |
5794 | ||
5795 | static void | |
5796 | spa_async_probe(spa_t *spa, vdev_t *vd) | |
5797 | { | |
5798 | int c; | |
5799 | ||
5800 | if (vd->vdev_probe_wanted) { | |
5801 | vd->vdev_probe_wanted = B_FALSE; | |
5802 | vdev_reopen(vd); /* vdev_open() does the actual probe */ | |
5803 | } | |
5804 | ||
5805 | for (c = 0; c < vd->vdev_children; c++) | |
5806 | spa_async_probe(spa, vd->vdev_child[c]); | |
5807 | } | |
5808 | ||
5809 | static void | |
5810 | spa_async_autoexpand(spa_t *spa, vdev_t *vd) | |
5811 | { | |
5812 | int c; | |
5813 | ||
5814 | if (!spa->spa_autoexpand) | |
5815 | return; | |
5816 | ||
5817 | for (c = 0; c < vd->vdev_children; c++) { | |
5818 | vdev_t *cvd = vd->vdev_child[c]; | |
5819 | spa_async_autoexpand(spa, cvd); | |
5820 | } | |
5821 | ||
5822 | if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL) | |
5823 | return; | |
5824 | ||
5825 | spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_AUTOEXPAND); | |
5826 | } | |
5827 | ||
5828 | static void | |
5829 | spa_async_thread(spa_t *spa) | |
5830 | { | |
5831 | int tasks, i; | |
5832 | ||
5833 | ASSERT(spa->spa_sync_on); | |
5834 | ||
5835 | mutex_enter(&spa->spa_async_lock); | |
5836 | tasks = spa->spa_async_tasks; | |
5837 | spa->spa_async_tasks = 0; | |
5838 | mutex_exit(&spa->spa_async_lock); | |
5839 | ||
5840 | /* | |
5841 | * See if the config needs to be updated. | |
5842 | */ | |
5843 | if (tasks & SPA_ASYNC_CONFIG_UPDATE) { | |
5844 | uint64_t old_space, new_space; | |
5845 | ||
5846 | mutex_enter(&spa_namespace_lock); | |
5847 | old_space = metaslab_class_get_space(spa_normal_class(spa)); | |
5848 | spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); | |
5849 | new_space = metaslab_class_get_space(spa_normal_class(spa)); | |
5850 | mutex_exit(&spa_namespace_lock); | |
5851 | ||
5852 | /* | |
5853 | * If the pool grew as a result of the config update, | |
5854 | * then log an internal history event. | |
5855 | */ | |
5856 | if (new_space != old_space) { | |
5857 | spa_history_log_internal(spa, "vdev online", NULL, | |
5858 | "pool '%s' size: %llu(+%llu)", | |
5859 | spa_name(spa), new_space, new_space - old_space); | |
5860 | } | |
5861 | } | |
5862 | ||
5863 | /* | |
5864 | * See if any devices need to be marked REMOVED. | |
5865 | */ | |
5866 | if (tasks & SPA_ASYNC_REMOVE) { | |
5867 | spa_vdev_state_enter(spa, SCL_NONE); | |
5868 | spa_async_remove(spa, spa->spa_root_vdev); | |
5869 | for (i = 0; i < spa->spa_l2cache.sav_count; i++) | |
5870 | spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]); | |
5871 | for (i = 0; i < spa->spa_spares.sav_count; i++) | |
5872 | spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]); | |
5873 | (void) spa_vdev_state_exit(spa, NULL, 0); | |
5874 | } | |
5875 | ||
5876 | if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) { | |
5877 | spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); | |
5878 | spa_async_autoexpand(spa, spa->spa_root_vdev); | |
5879 | spa_config_exit(spa, SCL_CONFIG, FTAG); | |
5880 | } | |
5881 | ||
5882 | /* | |
5883 | * See if any devices need to be probed. | |
5884 | */ | |
5885 | if (tasks & SPA_ASYNC_PROBE) { | |
5886 | spa_vdev_state_enter(spa, SCL_NONE); | |
5887 | spa_async_probe(spa, spa->spa_root_vdev); | |
5888 | (void) spa_vdev_state_exit(spa, NULL, 0); | |
5889 | } | |
5890 | ||
5891 | /* | |
5892 | * If any devices are done replacing, detach them. | |
5893 | */ | |
5894 | if (tasks & SPA_ASYNC_RESILVER_DONE) | |
5895 | spa_vdev_resilver_done(spa); | |
5896 | ||
5897 | /* | |
5898 | * Kick off a resilver. | |
5899 | */ | |
5900 | if (tasks & SPA_ASYNC_RESILVER) | |
5901 | dsl_resilver_restart(spa->spa_dsl_pool, 0); | |
5902 | ||
5903 | /* | |
5904 | * Let the world know that we're done. | |
5905 | */ | |
5906 | mutex_enter(&spa->spa_async_lock); | |
5907 | spa->spa_async_thread = NULL; | |
5908 | cv_broadcast(&spa->spa_async_cv); | |
5909 | mutex_exit(&spa->spa_async_lock); | |
5910 | thread_exit(); | |
5911 | } | |
5912 | ||
5913 | void | |
5914 | spa_async_suspend(spa_t *spa) | |
5915 | { | |
5916 | mutex_enter(&spa->spa_async_lock); | |
5917 | spa->spa_async_suspended++; | |
5918 | while (spa->spa_async_thread != NULL) | |
5919 | cv_wait(&spa->spa_async_cv, &spa->spa_async_lock); | |
5920 | mutex_exit(&spa->spa_async_lock); | |
5921 | } | |
5922 | ||
5923 | void | |
5924 | spa_async_resume(spa_t *spa) | |
5925 | { | |
5926 | mutex_enter(&spa->spa_async_lock); | |
5927 | ASSERT(spa->spa_async_suspended != 0); | |
5928 | spa->spa_async_suspended--; | |
5929 | mutex_exit(&spa->spa_async_lock); | |
5930 | } | |
5931 | ||
5932 | static boolean_t | |
5933 | spa_async_tasks_pending(spa_t *spa) | |
5934 | { | |
5935 | uint_t non_config_tasks; | |
5936 | uint_t config_task; | |
5937 | boolean_t config_task_suspended; | |
5938 | ||
5939 | non_config_tasks = spa->spa_async_tasks & ~SPA_ASYNC_CONFIG_UPDATE; | |
5940 | config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE; | |
5941 | if (spa->spa_ccw_fail_time == 0) { | |
5942 | config_task_suspended = B_FALSE; | |
5943 | } else { | |
5944 | config_task_suspended = | |
5945 | (gethrtime() - spa->spa_ccw_fail_time) < | |
5946 | ((hrtime_t)zfs_ccw_retry_interval * NANOSEC); | |
5947 | } | |
5948 | ||
5949 | return (non_config_tasks || (config_task && !config_task_suspended)); | |
5950 | } | |
5951 | ||
5952 | static void | |
5953 | spa_async_dispatch(spa_t *spa) | |
5954 | { | |
5955 | mutex_enter(&spa->spa_async_lock); | |
5956 | if (spa_async_tasks_pending(spa) && | |
5957 | !spa->spa_async_suspended && | |
5958 | spa->spa_async_thread == NULL && | |
5959 | rootdir != NULL) | |
5960 | spa->spa_async_thread = thread_create(NULL, 0, | |
5961 | spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri); | |
5962 | mutex_exit(&spa->spa_async_lock); | |
5963 | } | |
5964 | ||
5965 | void | |
5966 | spa_async_request(spa_t *spa, int task) | |
5967 | { | |
5968 | zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task); | |
5969 | mutex_enter(&spa->spa_async_lock); | |
5970 | spa->spa_async_tasks |= task; | |
5971 | mutex_exit(&spa->spa_async_lock); | |
5972 | } | |
5973 | ||
5974 | /* | |
5975 | * ========================================================================== | |
5976 | * SPA syncing routines | |
5977 | * ========================================================================== | |
5978 | */ | |
5979 | ||
5980 | static int | |
5981 | bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) | |
5982 | { | |
5983 | bpobj_t *bpo = arg; | |
5984 | bpobj_enqueue(bpo, bp, tx); | |
5985 | return (0); | |
5986 | } | |
5987 | ||
5988 | static int | |
5989 | spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) | |
5990 | { | |
5991 | zio_t *zio = arg; | |
5992 | ||
5993 | zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp, | |
5994 | zio->io_flags)); | |
5995 | return (0); | |
5996 | } | |
5997 | ||
5998 | /* | |
5999 | * Note: this simple function is not inlined to make it easier to dtrace the | |
6000 | * amount of time spent syncing frees. | |
6001 | */ | |
6002 | static void | |
6003 | spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx) | |
6004 | { | |
6005 | zio_t *zio = zio_root(spa, NULL, NULL, 0); | |
6006 | bplist_iterate(bpl, spa_free_sync_cb, zio, tx); | |
6007 | VERIFY(zio_wait(zio) == 0); | |
6008 | } | |
6009 | ||
6010 | /* | |
6011 | * Note: this simple function is not inlined to make it easier to dtrace the | |
6012 | * amount of time spent syncing deferred frees. | |
6013 | */ | |
6014 | static void | |
6015 | spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx) | |
6016 | { | |
6017 | zio_t *zio = zio_root(spa, NULL, NULL, 0); | |
6018 | VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj, | |
6019 | spa_free_sync_cb, zio, tx), ==, 0); | |
6020 | VERIFY0(zio_wait(zio)); | |
6021 | } | |
6022 | ||
6023 | static void | |
6024 | spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx) | |
6025 | { | |
6026 | char *packed = NULL; | |
6027 | size_t bufsize; | |
6028 | size_t nvsize = 0; | |
6029 | dmu_buf_t *db; | |
6030 | ||
6031 | VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0); | |
6032 | ||
6033 | /* | |
6034 | * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration | |
6035 | * information. This avoids the dmu_buf_will_dirty() path and | |
6036 | * saves us a pre-read to get data we don't actually care about. | |
6037 | */ | |
6038 | bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE); | |
6039 | packed = vmem_alloc(bufsize, KM_SLEEP); | |
6040 | ||
6041 | VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR, | |
6042 | KM_SLEEP) == 0); | |
6043 | bzero(packed + nvsize, bufsize - nvsize); | |
6044 | ||
6045 | dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx); | |
6046 | ||
6047 | vmem_free(packed, bufsize); | |
6048 | ||
6049 | VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); | |
6050 | dmu_buf_will_dirty(db, tx); | |
6051 | *(uint64_t *)db->db_data = nvsize; | |
6052 | dmu_buf_rele(db, FTAG); | |
6053 | } | |
6054 | ||
6055 | static void | |
6056 | spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx, | |
6057 | const char *config, const char *entry) | |
6058 | { | |
6059 | nvlist_t *nvroot; | |
6060 | nvlist_t **list; | |
6061 | int i; | |
6062 | ||
6063 | if (!sav->sav_sync) | |
6064 | return; | |
6065 | ||
6066 | /* | |
6067 | * Update the MOS nvlist describing the list of available devices. | |
6068 | * spa_validate_aux() will have already made sure this nvlist is | |
6069 | * valid and the vdevs are labeled appropriately. | |
6070 | */ | |
6071 | if (sav->sav_object == 0) { | |
6072 | sav->sav_object = dmu_object_alloc(spa->spa_meta_objset, | |
6073 | DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE, | |
6074 | sizeof (uint64_t), tx); | |
6075 | VERIFY(zap_update(spa->spa_meta_objset, | |
6076 | DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1, | |
6077 | &sav->sav_object, tx) == 0); | |
6078 | } | |
6079 | ||
6080 | VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); | |
6081 | if (sav->sav_count == 0) { | |
6082 | VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0); | |
6083 | } else { | |
6084 | list = kmem_alloc(sav->sav_count*sizeof (void *), KM_SLEEP); | |
6085 | for (i = 0; i < sav->sav_count; i++) | |
6086 | list[i] = vdev_config_generate(spa, sav->sav_vdevs[i], | |
6087 | B_FALSE, VDEV_CONFIG_L2CACHE); | |
6088 | VERIFY(nvlist_add_nvlist_array(nvroot, config, list, | |
6089 | sav->sav_count) == 0); | |
6090 | for (i = 0; i < sav->sav_count; i++) | |
6091 | nvlist_free(list[i]); | |
6092 | kmem_free(list, sav->sav_count * sizeof (void *)); | |
6093 | } | |
6094 | ||
6095 | spa_sync_nvlist(spa, sav->sav_object, nvroot, tx); | |
6096 | nvlist_free(nvroot); | |
6097 | ||
6098 | sav->sav_sync = B_FALSE; | |
6099 | } | |
6100 | ||
6101 | /* | |
6102 | * Rebuild spa's all-vdev ZAP from the vdev ZAPs indicated in each vdev_t. | |
6103 | * The all-vdev ZAP must be empty. | |
6104 | */ | |
6105 | static void | |
6106 | spa_avz_build(vdev_t *vd, uint64_t avz, dmu_tx_t *tx) | |
6107 | { | |
6108 | spa_t *spa = vd->vdev_spa; | |
6109 | uint64_t i; | |
6110 | ||
6111 | if (vd->vdev_top_zap != 0) { | |
6112 | VERIFY0(zap_add_int(spa->spa_meta_objset, avz, | |
6113 | vd->vdev_top_zap, tx)); | |
6114 | } | |
6115 | if (vd->vdev_leaf_zap != 0) { | |
6116 | VERIFY0(zap_add_int(spa->spa_meta_objset, avz, | |
6117 | vd->vdev_leaf_zap, tx)); | |
6118 | } | |
6119 | for (i = 0; i < vd->vdev_children; i++) { | |
6120 | spa_avz_build(vd->vdev_child[i], avz, tx); | |
6121 | } | |
6122 | } | |
6123 | ||
6124 | static void | |
6125 | spa_sync_config_object(spa_t *spa, dmu_tx_t *tx) | |
6126 | { | |
6127 | nvlist_t *config; | |
6128 | ||
6129 | /* | |
6130 | * If the pool is being imported from a pre-per-vdev-ZAP version of ZFS, | |
6131 | * its config may not be dirty but we still need to build per-vdev ZAPs. | |
6132 | * Similarly, if the pool is being assembled (e.g. after a split), we | |
6133 | * need to rebuild the AVZ although the config may not be dirty. | |
6134 | */ | |
6135 | if (list_is_empty(&spa->spa_config_dirty_list) && | |
6136 | spa->spa_avz_action == AVZ_ACTION_NONE) | |
6137 | return; | |
6138 | ||
6139 | spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); | |
6140 | ||
6141 | ASSERT(spa->spa_avz_action == AVZ_ACTION_NONE || | |
6142 | spa->spa_avz_action == AVZ_ACTION_INITIALIZE || | |
6143 | spa->spa_all_vdev_zaps != 0); | |
6144 | ||
6145 | if (spa->spa_avz_action == AVZ_ACTION_REBUILD) { | |
6146 | zap_cursor_t zc; | |
6147 | zap_attribute_t za; | |
6148 | ||
6149 | /* Make and build the new AVZ */ | |
6150 | uint64_t new_avz = zap_create(spa->spa_meta_objset, | |
6151 | DMU_OTN_ZAP_METADATA, DMU_OT_NONE, 0, tx); | |
6152 | spa_avz_build(spa->spa_root_vdev, new_avz, tx); | |
6153 | ||
6154 | /* Diff old AVZ with new one */ | |
6155 | for (zap_cursor_init(&zc, spa->spa_meta_objset, | |
6156 | spa->spa_all_vdev_zaps); | |
6157 | zap_cursor_retrieve(&zc, &za) == 0; | |
6158 | zap_cursor_advance(&zc)) { | |
6159 | uint64_t vdzap = za.za_first_integer; | |
6160 | if (zap_lookup_int(spa->spa_meta_objset, new_avz, | |
6161 | vdzap) == ENOENT) { | |
6162 | /* | |
6163 | * ZAP is listed in old AVZ but not in new one; | |
6164 | * destroy it | |
6165 | */ | |
6166 | VERIFY0(zap_destroy(spa->spa_meta_objset, vdzap, | |
6167 | tx)); | |
6168 | } | |
6169 | } | |
6170 | ||
6171 | zap_cursor_fini(&zc); | |
6172 | ||
6173 | /* Destroy the old AVZ */ | |
6174 | VERIFY0(zap_destroy(spa->spa_meta_objset, | |
6175 | spa->spa_all_vdev_zaps, tx)); | |
6176 | ||
6177 | /* Replace the old AVZ in the dir obj with the new one */ | |
6178 | VERIFY0(zap_update(spa->spa_meta_objset, | |
6179 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP, | |
6180 | sizeof (new_avz), 1, &new_avz, tx)); | |
6181 | ||
6182 | spa->spa_all_vdev_zaps = new_avz; | |
6183 | } else if (spa->spa_avz_action == AVZ_ACTION_DESTROY) { | |
6184 | zap_cursor_t zc; | |
6185 | zap_attribute_t za; | |
6186 | ||
6187 | /* Walk through the AVZ and destroy all listed ZAPs */ | |
6188 | for (zap_cursor_init(&zc, spa->spa_meta_objset, | |
6189 | spa->spa_all_vdev_zaps); | |
6190 | zap_cursor_retrieve(&zc, &za) == 0; | |
6191 | zap_cursor_advance(&zc)) { | |
6192 | uint64_t zap = za.za_first_integer; | |
6193 | VERIFY0(zap_destroy(spa->spa_meta_objset, zap, tx)); | |
6194 | } | |
6195 | ||
6196 | zap_cursor_fini(&zc); | |
6197 | ||
6198 | /* Destroy and unlink the AVZ itself */ | |
6199 | VERIFY0(zap_destroy(spa->spa_meta_objset, | |
6200 | spa->spa_all_vdev_zaps, tx)); | |
6201 | VERIFY0(zap_remove(spa->spa_meta_objset, | |
6202 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP, tx)); | |
6203 | spa->spa_all_vdev_zaps = 0; | |
6204 | } | |
6205 | ||
6206 | if (spa->spa_all_vdev_zaps == 0) { | |
6207 | spa->spa_all_vdev_zaps = zap_create_link(spa->spa_meta_objset, | |
6208 | DMU_OTN_ZAP_METADATA, DMU_POOL_DIRECTORY_OBJECT, | |
6209 | DMU_POOL_VDEV_ZAP_MAP, tx); | |
6210 | } | |
6211 | spa->spa_avz_action = AVZ_ACTION_NONE; | |
6212 | ||
6213 | /* Create ZAPs for vdevs that don't have them. */ | |
6214 | vdev_construct_zaps(spa->spa_root_vdev, tx); | |
6215 | ||
6216 | config = spa_config_generate(spa, spa->spa_root_vdev, | |
6217 | dmu_tx_get_txg(tx), B_FALSE); | |
6218 | ||
6219 | /* | |
6220 | * If we're upgrading the spa version then make sure that | |
6221 | * the config object gets updated with the correct version. | |
6222 | */ | |
6223 | if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version) | |
6224 | fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, | |
6225 | spa->spa_uberblock.ub_version); | |
6226 | ||
6227 | spa_config_exit(spa, SCL_STATE, FTAG); | |
6228 | ||
6229 | nvlist_free(spa->spa_config_syncing); | |
6230 | spa->spa_config_syncing = config; | |
6231 | ||
6232 | spa_sync_nvlist(spa, spa->spa_config_object, config, tx); | |
6233 | } | |
6234 | ||
6235 | static void | |
6236 | spa_sync_version(void *arg, dmu_tx_t *tx) | |
6237 | { | |
6238 | uint64_t *versionp = arg; | |
6239 | uint64_t version = *versionp; | |
6240 | spa_t *spa = dmu_tx_pool(tx)->dp_spa; | |
6241 | ||
6242 | /* | |
6243 | * Setting the version is special cased when first creating the pool. | |
6244 | */ | |
6245 | ASSERT(tx->tx_txg != TXG_INITIAL); | |
6246 | ||
6247 | ASSERT(SPA_VERSION_IS_SUPPORTED(version)); | |
6248 | ASSERT(version >= spa_version(spa)); | |
6249 | ||
6250 | spa->spa_uberblock.ub_version = version; | |
6251 | vdev_config_dirty(spa->spa_root_vdev); | |
6252 | spa_history_log_internal(spa, "set", tx, "version=%lld", version); | |
6253 | } | |
6254 | ||
6255 | /* | |
6256 | * Set zpool properties. | |
6257 | */ | |
6258 | static void | |
6259 | spa_sync_props(void *arg, dmu_tx_t *tx) | |
6260 | { | |
6261 | nvlist_t *nvp = arg; | |
6262 | spa_t *spa = dmu_tx_pool(tx)->dp_spa; | |
6263 | objset_t *mos = spa->spa_meta_objset; | |
6264 | nvpair_t *elem = NULL; | |
6265 | ||
6266 | mutex_enter(&spa->spa_props_lock); | |
6267 | ||
6268 | while ((elem = nvlist_next_nvpair(nvp, elem))) { | |
6269 | uint64_t intval; | |
6270 | char *strval, *fname; | |
6271 | zpool_prop_t prop; | |
6272 | const char *propname; | |
6273 | zprop_type_t proptype; | |
6274 | spa_feature_t fid; | |
6275 | ||
6276 | prop = zpool_name_to_prop(nvpair_name(elem)); | |
6277 | switch ((int)prop) { | |
6278 | case ZPROP_INVAL: | |
6279 | /* | |
6280 | * We checked this earlier in spa_prop_validate(). | |
6281 | */ | |
6282 | ASSERT(zpool_prop_feature(nvpair_name(elem))); | |
6283 | ||
6284 | fname = strchr(nvpair_name(elem), '@') + 1; | |
6285 | VERIFY0(zfeature_lookup_name(fname, &fid)); | |
6286 | ||
6287 | spa_feature_enable(spa, fid, tx); | |
6288 | spa_history_log_internal(spa, "set", tx, | |
6289 | "%s=enabled", nvpair_name(elem)); | |
6290 | break; | |
6291 | ||
6292 | case ZPOOL_PROP_VERSION: | |
6293 | intval = fnvpair_value_uint64(elem); | |
6294 | /* | |
6295 | * The version is synced separately before other | |
6296 | * properties and should be correct by now. | |
6297 | */ | |
6298 | ASSERT3U(spa_version(spa), >=, intval); | |
6299 | break; | |
6300 | ||
6301 | case ZPOOL_PROP_ALTROOT: | |
6302 | /* | |
6303 | * 'altroot' is a non-persistent property. It should | |
6304 | * have been set temporarily at creation or import time. | |
6305 | */ | |
6306 | ASSERT(spa->spa_root != NULL); | |
6307 | break; | |
6308 | ||
6309 | case ZPOOL_PROP_READONLY: | |
6310 | case ZPOOL_PROP_CACHEFILE: | |
6311 | /* | |
6312 | * 'readonly' and 'cachefile' are also non-persisitent | |
6313 | * properties. | |
6314 | */ | |
6315 | break; | |
6316 | case ZPOOL_PROP_COMMENT: | |
6317 | strval = fnvpair_value_string(elem); | |
6318 | if (spa->spa_comment != NULL) | |
6319 | spa_strfree(spa->spa_comment); | |
6320 | spa->spa_comment = spa_strdup(strval); | |
6321 | /* | |
6322 | * We need to dirty the configuration on all the vdevs | |
6323 | * so that their labels get updated. It's unnecessary | |
6324 | * to do this for pool creation since the vdev's | |
6325 | * configuration has already been dirtied. | |
6326 | */ | |
6327 | if (tx->tx_txg != TXG_INITIAL) | |
6328 | vdev_config_dirty(spa->spa_root_vdev); | |
6329 | spa_history_log_internal(spa, "set", tx, | |
6330 | "%s=%s", nvpair_name(elem), strval); | |
6331 | break; | |
6332 | default: | |
6333 | /* | |
6334 | * Set pool property values in the poolprops mos object. | |
6335 | */ | |
6336 | if (spa->spa_pool_props_object == 0) { | |
6337 | spa->spa_pool_props_object = | |
6338 | zap_create_link(mos, DMU_OT_POOL_PROPS, | |
6339 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS, | |
6340 | tx); | |
6341 | } | |
6342 | ||
6343 | /* normalize the property name */ | |
6344 | propname = zpool_prop_to_name(prop); | |
6345 | proptype = zpool_prop_get_type(prop); | |
6346 | ||
6347 | if (nvpair_type(elem) == DATA_TYPE_STRING) { | |
6348 | ASSERT(proptype == PROP_TYPE_STRING); | |
6349 | strval = fnvpair_value_string(elem); | |
6350 | VERIFY0(zap_update(mos, | |
6351 | spa->spa_pool_props_object, propname, | |
6352 | 1, strlen(strval) + 1, strval, tx)); | |
6353 | spa_history_log_internal(spa, "set", tx, | |
6354 | "%s=%s", nvpair_name(elem), strval); | |
6355 | } else if (nvpair_type(elem) == DATA_TYPE_UINT64) { | |
6356 | intval = fnvpair_value_uint64(elem); | |
6357 | ||
6358 | if (proptype == PROP_TYPE_INDEX) { | |
6359 | const char *unused; | |
6360 | VERIFY0(zpool_prop_index_to_string( | |
6361 | prop, intval, &unused)); | |
6362 | } | |
6363 | VERIFY0(zap_update(mos, | |
6364 | spa->spa_pool_props_object, propname, | |
6365 | 8, 1, &intval, tx)); | |
6366 | spa_history_log_internal(spa, "set", tx, | |
6367 | "%s=%lld", nvpair_name(elem), intval); | |
6368 | } else { | |
6369 | ASSERT(0); /* not allowed */ | |
6370 | } | |
6371 | ||
6372 | switch (prop) { | |
6373 | case ZPOOL_PROP_DELEGATION: | |
6374 | spa->spa_delegation = intval; | |
6375 | break; | |
6376 | case ZPOOL_PROP_BOOTFS: | |
6377 | spa->spa_bootfs = intval; | |
6378 | break; | |
6379 | case ZPOOL_PROP_FAILUREMODE: | |
6380 | spa->spa_failmode = intval; | |
6381 | break; | |
6382 | case ZPOOL_PROP_AUTOEXPAND: | |
6383 | spa->spa_autoexpand = intval; | |
6384 | if (tx->tx_txg != TXG_INITIAL) | |
6385 | spa_async_request(spa, | |
6386 | SPA_ASYNC_AUTOEXPAND); | |
6387 | break; | |
6388 | case ZPOOL_PROP_DEDUPDITTO: | |
6389 | spa->spa_dedup_ditto = intval; | |
6390 | break; | |
6391 | default: | |
6392 | break; | |
6393 | } | |
6394 | } | |
6395 | ||
6396 | } | |
6397 | ||
6398 | mutex_exit(&spa->spa_props_lock); | |
6399 | } | |
6400 | ||
6401 | /* | |
6402 | * Perform one-time upgrade on-disk changes. spa_version() does not | |
6403 | * reflect the new version this txg, so there must be no changes this | |
6404 | * txg to anything that the upgrade code depends on after it executes. | |
6405 | * Therefore this must be called after dsl_pool_sync() does the sync | |
6406 | * tasks. | |
6407 | */ | |
6408 | static void | |
6409 | spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx) | |
6410 | { | |
6411 | dsl_pool_t *dp = spa->spa_dsl_pool; | |
6412 | ||
6413 | ASSERT(spa->spa_sync_pass == 1); | |
6414 | ||
6415 | rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); | |
6416 | ||
6417 | if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN && | |
6418 | spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) { | |
6419 | dsl_pool_create_origin(dp, tx); | |
6420 | ||
6421 | /* Keeping the origin open increases spa_minref */ | |
6422 | spa->spa_minref += 3; | |
6423 | } | |
6424 | ||
6425 | if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES && | |
6426 | spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) { | |
6427 | dsl_pool_upgrade_clones(dp, tx); | |
6428 | } | |
6429 | ||
6430 | if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES && | |
6431 | spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) { | |
6432 | dsl_pool_upgrade_dir_clones(dp, tx); | |
6433 | ||
6434 | /* Keeping the freedir open increases spa_minref */ | |
6435 | spa->spa_minref += 3; | |
6436 | } | |
6437 | ||
6438 | if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES && | |
6439 | spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { | |
6440 | spa_feature_create_zap_objects(spa, tx); | |
6441 | } | |
6442 | ||
6443 | /* | |
6444 | * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable | |
6445 | * when possibility to use lz4 compression for metadata was added | |
6446 | * Old pools that have this feature enabled must be upgraded to have | |
6447 | * this feature active | |
6448 | */ | |
6449 | if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { | |
6450 | boolean_t lz4_en = spa_feature_is_enabled(spa, | |
6451 | SPA_FEATURE_LZ4_COMPRESS); | |
6452 | boolean_t lz4_ac = spa_feature_is_active(spa, | |
6453 | SPA_FEATURE_LZ4_COMPRESS); | |
6454 | ||
6455 | if (lz4_en && !lz4_ac) | |
6456 | spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx); | |
6457 | } | |
6458 | ||
6459 | /* | |
6460 | * If we haven't written the salt, do so now. Note that the | |
6461 | * feature may not be activated yet, but that's fine since | |
6462 | * the presence of this ZAP entry is backwards compatible. | |
6463 | */ | |
6464 | if (zap_contains(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, | |
6465 | DMU_POOL_CHECKSUM_SALT) == ENOENT) { | |
6466 | VERIFY0(zap_add(spa->spa_meta_objset, | |
6467 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CHECKSUM_SALT, 1, | |
6468 | sizeof (spa->spa_cksum_salt.zcs_bytes), | |
6469 | spa->spa_cksum_salt.zcs_bytes, tx)); | |
6470 | } | |
6471 | ||
6472 | rrw_exit(&dp->dp_config_rwlock, FTAG); | |
6473 | } | |
6474 | ||
6475 | /* | |
6476 | * Sync the specified transaction group. New blocks may be dirtied as | |
6477 | * part of the process, so we iterate until it converges. | |
6478 | */ | |
6479 | void | |
6480 | spa_sync(spa_t *spa, uint64_t txg) | |
6481 | { | |
6482 | dsl_pool_t *dp = spa->spa_dsl_pool; | |
6483 | objset_t *mos = spa->spa_meta_objset; | |
6484 | bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK]; | |
6485 | metaslab_class_t *mc; | |
6486 | vdev_t *rvd = spa->spa_root_vdev; | |
6487 | vdev_t *vd; | |
6488 | dmu_tx_t *tx; | |
6489 | int error; | |
6490 | uint32_t max_queue_depth = zfs_vdev_async_write_max_active * | |
6491 | zfs_vdev_queue_depth_pct / 100; | |
6492 | uint64_t queue_depth_total; | |
6493 | int c; | |
6494 | ||
6495 | VERIFY(spa_writeable(spa)); | |
6496 | ||
6497 | /* | |
6498 | * Lock out configuration changes. | |
6499 | */ | |
6500 | spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); | |
6501 | ||
6502 | spa->spa_syncing_txg = txg; | |
6503 | spa->spa_sync_pass = 0; | |
6504 | ||
6505 | mutex_enter(&spa->spa_alloc_lock); | |
6506 | VERIFY0(avl_numnodes(&spa->spa_alloc_tree)); | |
6507 | mutex_exit(&spa->spa_alloc_lock); | |
6508 | ||
6509 | /* | |
6510 | * If there are any pending vdev state changes, convert them | |
6511 | * into config changes that go out with this transaction group. | |
6512 | */ | |
6513 | spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); | |
6514 | while (list_head(&spa->spa_state_dirty_list) != NULL) { | |
6515 | /* | |
6516 | * We need the write lock here because, for aux vdevs, | |
6517 | * calling vdev_config_dirty() modifies sav_config. | |
6518 | * This is ugly and will become unnecessary when we | |
6519 | * eliminate the aux vdev wart by integrating all vdevs | |
6520 | * into the root vdev tree. | |
6521 | */ | |
6522 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
6523 | spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER); | |
6524 | while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) { | |
6525 | vdev_state_clean(vd); | |
6526 | vdev_config_dirty(vd); | |
6527 | } | |
6528 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
6529 | spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); | |
6530 | } | |
6531 | spa_config_exit(spa, SCL_STATE, FTAG); | |
6532 | ||
6533 | tx = dmu_tx_create_assigned(dp, txg); | |
6534 | ||
6535 | spa->spa_sync_starttime = gethrtime(); | |
6536 | taskq_cancel_id(system_delay_taskq, spa->spa_deadman_tqid); | |
6537 | spa->spa_deadman_tqid = taskq_dispatch_delay(system_delay_taskq, | |
6538 | spa_deadman, spa, TQ_SLEEP, ddi_get_lbolt() + | |
6539 | NSEC_TO_TICK(spa->spa_deadman_synctime)); | |
6540 | ||
6541 | /* | |
6542 | * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg, | |
6543 | * set spa_deflate if we have no raid-z vdevs. | |
6544 | */ | |
6545 | if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE && | |
6546 | spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) { | |
6547 | int i; | |
6548 | ||
6549 | for (i = 0; i < rvd->vdev_children; i++) { | |
6550 | vd = rvd->vdev_child[i]; | |
6551 | if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE) | |
6552 | break; | |
6553 | } | |
6554 | if (i == rvd->vdev_children) { | |
6555 | spa->spa_deflate = TRUE; | |
6556 | VERIFY(0 == zap_add(spa->spa_meta_objset, | |
6557 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, | |
6558 | sizeof (uint64_t), 1, &spa->spa_deflate, tx)); | |
6559 | } | |
6560 | } | |
6561 | ||
6562 | /* | |
6563 | * Set the top-level vdev's max queue depth. Evaluate each | |
6564 | * top-level's async write queue depth in case it changed. | |
6565 | * The max queue depth will not change in the middle of syncing | |
6566 | * out this txg. | |
6567 | */ | |
6568 | queue_depth_total = 0; | |
6569 | for (c = 0; c < rvd->vdev_children; c++) { | |
6570 | vdev_t *tvd = rvd->vdev_child[c]; | |
6571 | metaslab_group_t *mg = tvd->vdev_mg; | |
6572 | ||
6573 | if (mg == NULL || mg->mg_class != spa_normal_class(spa) || | |
6574 | !metaslab_group_initialized(mg)) | |
6575 | continue; | |
6576 | ||
6577 | /* | |
6578 | * It is safe to do a lock-free check here because only async | |
6579 | * allocations look at mg_max_alloc_queue_depth, and async | |
6580 | * allocations all happen from spa_sync(). | |
6581 | */ | |
6582 | ASSERT0(refcount_count(&mg->mg_alloc_queue_depth)); | |
6583 | mg->mg_max_alloc_queue_depth = max_queue_depth; | |
6584 | queue_depth_total += mg->mg_max_alloc_queue_depth; | |
6585 | } | |
6586 | mc = spa_normal_class(spa); | |
6587 | ASSERT0(refcount_count(&mc->mc_alloc_slots)); | |
6588 | mc->mc_alloc_max_slots = queue_depth_total; | |
6589 | mc->mc_alloc_throttle_enabled = zio_dva_throttle_enabled; | |
6590 | ||
6591 | ASSERT3U(mc->mc_alloc_max_slots, <=, | |
6592 | max_queue_depth * rvd->vdev_children); | |
6593 | ||
6594 | /* | |
6595 | * Iterate to convergence. | |
6596 | */ | |
6597 | do { | |
6598 | int pass = ++spa->spa_sync_pass; | |
6599 | ||
6600 | spa_sync_config_object(spa, tx); | |
6601 | spa_sync_aux_dev(spa, &spa->spa_spares, tx, | |
6602 | ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES); | |
6603 | spa_sync_aux_dev(spa, &spa->spa_l2cache, tx, | |
6604 | ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE); | |
6605 | spa_errlog_sync(spa, txg); | |
6606 | dsl_pool_sync(dp, txg); | |
6607 | ||
6608 | if (pass < zfs_sync_pass_deferred_free) { | |
6609 | spa_sync_frees(spa, free_bpl, tx); | |
6610 | } else { | |
6611 | /* | |
6612 | * We can not defer frees in pass 1, because | |
6613 | * we sync the deferred frees later in pass 1. | |
6614 | */ | |
6615 | ASSERT3U(pass, >, 1); | |
6616 | bplist_iterate(free_bpl, bpobj_enqueue_cb, | |
6617 | &spa->spa_deferred_bpobj, tx); | |
6618 | } | |
6619 | ||
6620 | ddt_sync(spa, txg); | |
6621 | dsl_scan_sync(dp, tx); | |
6622 | ||
6623 | while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))) | |
6624 | vdev_sync(vd, txg); | |
6625 | ||
6626 | if (pass == 1) { | |
6627 | spa_sync_upgrades(spa, tx); | |
6628 | ASSERT3U(txg, >=, | |
6629 | spa->spa_uberblock.ub_rootbp.blk_birth); | |
6630 | /* | |
6631 | * Note: We need to check if the MOS is dirty | |
6632 | * because we could have marked the MOS dirty | |
6633 | * without updating the uberblock (e.g. if we | |
6634 | * have sync tasks but no dirty user data). We | |
6635 | * need to check the uberblock's rootbp because | |
6636 | * it is updated if we have synced out dirty | |
6637 | * data (though in this case the MOS will most | |
6638 | * likely also be dirty due to second order | |
6639 | * effects, we don't want to rely on that here). | |
6640 | */ | |
6641 | if (spa->spa_uberblock.ub_rootbp.blk_birth < txg && | |
6642 | !dmu_objset_is_dirty(mos, txg)) { | |
6643 | /* | |
6644 | * Nothing changed on the first pass, | |
6645 | * therefore this TXG is a no-op. Avoid | |
6646 | * syncing deferred frees, so that we | |
6647 | * can keep this TXG as a no-op. | |
6648 | */ | |
6649 | ASSERT(txg_list_empty(&dp->dp_dirty_datasets, | |
6650 | txg)); | |
6651 | ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); | |
6652 | ASSERT(txg_list_empty(&dp->dp_sync_tasks, txg)); | |
6653 | break; | |
6654 | } | |
6655 | spa_sync_deferred_frees(spa, tx); | |
6656 | } | |
6657 | ||
6658 | } while (dmu_objset_is_dirty(mos, txg)); | |
6659 | ||
6660 | #ifdef ZFS_DEBUG | |
6661 | if (!list_is_empty(&spa->spa_config_dirty_list)) { | |
6662 | /* | |
6663 | * Make sure that the number of ZAPs for all the vdevs matches | |
6664 | * the number of ZAPs in the per-vdev ZAP list. This only gets | |
6665 | * called if the config is dirty; otherwise there may be | |
6666 | * outstanding AVZ operations that weren't completed in | |
6667 | * spa_sync_config_object. | |
6668 | */ | |
6669 | uint64_t all_vdev_zap_entry_count; | |
6670 | ASSERT0(zap_count(spa->spa_meta_objset, | |
6671 | spa->spa_all_vdev_zaps, &all_vdev_zap_entry_count)); | |
6672 | ASSERT3U(vdev_count_verify_zaps(spa->spa_root_vdev), ==, | |
6673 | all_vdev_zap_entry_count); | |
6674 | } | |
6675 | #endif | |
6676 | ||
6677 | /* | |
6678 | * Rewrite the vdev configuration (which includes the uberblock) | |
6679 | * to commit the transaction group. | |
6680 | * | |
6681 | * If there are no dirty vdevs, we sync the uberblock to a few | |
6682 | * random top-level vdevs that are known to be visible in the | |
6683 | * config cache (see spa_vdev_add() for a complete description). | |
6684 | * If there *are* dirty vdevs, sync the uberblock to all vdevs. | |
6685 | */ | |
6686 | for (;;) { | |
6687 | /* | |
6688 | * We hold SCL_STATE to prevent vdev open/close/etc. | |
6689 | * while we're attempting to write the vdev labels. | |
6690 | */ | |
6691 | spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); | |
6692 | ||
6693 | if (list_is_empty(&spa->spa_config_dirty_list)) { | |
6694 | vdev_t *svd[SPA_DVAS_PER_BP]; | |
6695 | int svdcount = 0; | |
6696 | int children = rvd->vdev_children; | |
6697 | int c0 = spa_get_random(children); | |
6698 | ||
6699 | for (c = 0; c < children; c++) { | |
6700 | vd = rvd->vdev_child[(c0 + c) % children]; | |
6701 | if (vd->vdev_ms_array == 0 || vd->vdev_islog) | |
6702 | continue; | |
6703 | svd[svdcount++] = vd; | |
6704 | if (svdcount == SPA_DVAS_PER_BP) | |
6705 | break; | |
6706 | } | |
6707 | error = vdev_config_sync(svd, svdcount, txg); | |
6708 | } else { | |
6709 | error = vdev_config_sync(rvd->vdev_child, | |
6710 | rvd->vdev_children, txg); | |
6711 | } | |
6712 | ||
6713 | if (error == 0) | |
6714 | spa->spa_last_synced_guid = rvd->vdev_guid; | |
6715 | ||
6716 | spa_config_exit(spa, SCL_STATE, FTAG); | |
6717 | ||
6718 | if (error == 0) | |
6719 | break; | |
6720 | zio_suspend(spa, NULL); | |
6721 | zio_resume_wait(spa); | |
6722 | } | |
6723 | dmu_tx_commit(tx); | |
6724 | ||
6725 | taskq_cancel_id(system_delay_taskq, spa->spa_deadman_tqid); | |
6726 | spa->spa_deadman_tqid = 0; | |
6727 | ||
6728 | /* | |
6729 | * Clear the dirty config list. | |
6730 | */ | |
6731 | while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL) | |
6732 | vdev_config_clean(vd); | |
6733 | ||
6734 | /* | |
6735 | * Now that the new config has synced transactionally, | |
6736 | * let it become visible to the config cache. | |
6737 | */ | |
6738 | if (spa->spa_config_syncing != NULL) { | |
6739 | spa_config_set(spa, spa->spa_config_syncing); | |
6740 | spa->spa_config_txg = txg; | |
6741 | spa->spa_config_syncing = NULL; | |
6742 | } | |
6743 | ||
6744 | spa->spa_ubsync = spa->spa_uberblock; | |
6745 | ||
6746 | dsl_pool_sync_done(dp, txg); | |
6747 | ||
6748 | mutex_enter(&spa->spa_alloc_lock); | |
6749 | VERIFY0(avl_numnodes(&spa->spa_alloc_tree)); | |
6750 | mutex_exit(&spa->spa_alloc_lock); | |
6751 | ||
6752 | /* | |
6753 | * Update usable space statistics. | |
6754 | */ | |
6755 | while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))) | |
6756 | vdev_sync_done(vd, txg); | |
6757 | ||
6758 | spa_update_dspace(spa); | |
6759 | ||
6760 | /* | |
6761 | * It had better be the case that we didn't dirty anything | |
6762 | * since vdev_config_sync(). | |
6763 | */ | |
6764 | ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg)); | |
6765 | ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); | |
6766 | ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg)); | |
6767 | ||
6768 | spa->spa_sync_pass = 0; | |
6769 | ||
6770 | spa_config_exit(spa, SCL_CONFIG, FTAG); | |
6771 | ||
6772 | spa_handle_ignored_writes(spa); | |
6773 | ||
6774 | /* | |
6775 | * If any async tasks have been requested, kick them off. | |
6776 | */ | |
6777 | spa_async_dispatch(spa); | |
6778 | } | |
6779 | ||
6780 | /* | |
6781 | * Sync all pools. We don't want to hold the namespace lock across these | |
6782 | * operations, so we take a reference on the spa_t and drop the lock during the | |
6783 | * sync. | |
6784 | */ | |
6785 | void | |
6786 | spa_sync_allpools(void) | |
6787 | { | |
6788 | spa_t *spa = NULL; | |
6789 | mutex_enter(&spa_namespace_lock); | |
6790 | while ((spa = spa_next(spa)) != NULL) { | |
6791 | if (spa_state(spa) != POOL_STATE_ACTIVE || | |
6792 | !spa_writeable(spa) || spa_suspended(spa)) | |
6793 | continue; | |
6794 | spa_open_ref(spa, FTAG); | |
6795 | mutex_exit(&spa_namespace_lock); | |
6796 | txg_wait_synced(spa_get_dsl(spa), 0); | |
6797 | mutex_enter(&spa_namespace_lock); | |
6798 | spa_close(spa, FTAG); | |
6799 | } | |
6800 | mutex_exit(&spa_namespace_lock); | |
6801 | } | |
6802 | ||
6803 | /* | |
6804 | * ========================================================================== | |
6805 | * Miscellaneous routines | |
6806 | * ========================================================================== | |
6807 | */ | |
6808 | ||
6809 | /* | |
6810 | * Remove all pools in the system. | |
6811 | */ | |
6812 | void | |
6813 | spa_evict_all(void) | |
6814 | { | |
6815 | spa_t *spa; | |
6816 | ||
6817 | /* | |
6818 | * Remove all cached state. All pools should be closed now, | |
6819 | * so every spa in the AVL tree should be unreferenced. | |
6820 | */ | |
6821 | mutex_enter(&spa_namespace_lock); | |
6822 | while ((spa = spa_next(NULL)) != NULL) { | |
6823 | /* | |
6824 | * Stop async tasks. The async thread may need to detach | |
6825 | * a device that's been replaced, which requires grabbing | |
6826 | * spa_namespace_lock, so we must drop it here. | |
6827 | */ | |
6828 | spa_open_ref(spa, FTAG); | |
6829 | mutex_exit(&spa_namespace_lock); | |
6830 | spa_async_suspend(spa); | |
6831 | mutex_enter(&spa_namespace_lock); | |
6832 | spa_close(spa, FTAG); | |
6833 | ||
6834 | if (spa->spa_state != POOL_STATE_UNINITIALIZED) { | |
6835 | spa_unload(spa); | |
6836 | spa_deactivate(spa); | |
6837 | } | |
6838 | spa_remove(spa); | |
6839 | } | |
6840 | mutex_exit(&spa_namespace_lock); | |
6841 | } | |
6842 | ||
6843 | vdev_t * | |
6844 | spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux) | |
6845 | { | |
6846 | vdev_t *vd; | |
6847 | int i; | |
6848 | ||
6849 | if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL) | |
6850 | return (vd); | |
6851 | ||
6852 | if (aux) { | |
6853 | for (i = 0; i < spa->spa_l2cache.sav_count; i++) { | |
6854 | vd = spa->spa_l2cache.sav_vdevs[i]; | |
6855 | if (vd->vdev_guid == guid) | |
6856 | return (vd); | |
6857 | } | |
6858 | ||
6859 | for (i = 0; i < spa->spa_spares.sav_count; i++) { | |
6860 | vd = spa->spa_spares.sav_vdevs[i]; | |
6861 | if (vd->vdev_guid == guid) | |
6862 | return (vd); | |
6863 | } | |
6864 | } | |
6865 | ||
6866 | return (NULL); | |
6867 | } | |
6868 | ||
6869 | void | |
6870 | spa_upgrade(spa_t *spa, uint64_t version) | |
6871 | { | |
6872 | ASSERT(spa_writeable(spa)); | |
6873 | ||
6874 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
6875 | ||
6876 | /* | |
6877 | * This should only be called for a non-faulted pool, and since a | |
6878 | * future version would result in an unopenable pool, this shouldn't be | |
6879 | * possible. | |
6880 | */ | |
6881 | ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version)); | |
6882 | ASSERT3U(version, >=, spa->spa_uberblock.ub_version); | |
6883 | ||
6884 | spa->spa_uberblock.ub_version = version; | |
6885 | vdev_config_dirty(spa->spa_root_vdev); | |
6886 | ||
6887 | spa_config_exit(spa, SCL_ALL, FTAG); | |
6888 | ||
6889 | txg_wait_synced(spa_get_dsl(spa), 0); | |
6890 | } | |
6891 | ||
6892 | boolean_t | |
6893 | spa_has_spare(spa_t *spa, uint64_t guid) | |
6894 | { | |
6895 | int i; | |
6896 | uint64_t spareguid; | |
6897 | spa_aux_vdev_t *sav = &spa->spa_spares; | |
6898 | ||
6899 | for (i = 0; i < sav->sav_count; i++) | |
6900 | if (sav->sav_vdevs[i]->vdev_guid == guid) | |
6901 | return (B_TRUE); | |
6902 | ||
6903 | for (i = 0; i < sav->sav_npending; i++) { | |
6904 | if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID, | |
6905 | &spareguid) == 0 && spareguid == guid) | |
6906 | return (B_TRUE); | |
6907 | } | |
6908 | ||
6909 | return (B_FALSE); | |
6910 | } | |
6911 | ||
6912 | /* | |
6913 | * Check if a pool has an active shared spare device. | |
6914 | * Note: reference count of an active spare is 2, as a spare and as a replace | |
6915 | */ | |
6916 | static boolean_t | |
6917 | spa_has_active_shared_spare(spa_t *spa) | |
6918 | { | |
6919 | int i, refcnt; | |
6920 | uint64_t pool; | |
6921 | spa_aux_vdev_t *sav = &spa->spa_spares; | |
6922 | ||
6923 | for (i = 0; i < sav->sav_count; i++) { | |
6924 | if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool, | |
6925 | &refcnt) && pool != 0ULL && pool == spa_guid(spa) && | |
6926 | refcnt > 2) | |
6927 | return (B_TRUE); | |
6928 | } | |
6929 | ||
6930 | return (B_FALSE); | |
6931 | } | |
6932 | ||
6933 | /* | |
6934 | * Post a zevent corresponding to the given sysevent. The 'name' must be one | |
6935 | * of the event definitions in sys/sysevent/eventdefs.h. The payload will be | |
6936 | * filled in from the spa and (optionally) the vdev. This doesn't do anything | |
6937 | * in the userland libzpool, as we don't want consumers to misinterpret ztest | |
6938 | * or zdb as real changes. | |
6939 | */ | |
6940 | void | |
6941 | spa_event_notify(spa_t *spa, vdev_t *vd, const char *name) | |
6942 | { | |
6943 | zfs_post_sysevent(spa, vd, name); | |
6944 | } | |
6945 | ||
6946 | #if defined(_KERNEL) && defined(HAVE_SPL) | |
6947 | /* state manipulation functions */ | |
6948 | EXPORT_SYMBOL(spa_open); | |
6949 | EXPORT_SYMBOL(spa_open_rewind); | |
6950 | EXPORT_SYMBOL(spa_get_stats); | |
6951 | EXPORT_SYMBOL(spa_create); | |
6952 | EXPORT_SYMBOL(spa_import); | |
6953 | EXPORT_SYMBOL(spa_tryimport); | |
6954 | EXPORT_SYMBOL(spa_destroy); | |
6955 | EXPORT_SYMBOL(spa_export); | |
6956 | EXPORT_SYMBOL(spa_reset); | |
6957 | EXPORT_SYMBOL(spa_async_request); | |
6958 | EXPORT_SYMBOL(spa_async_suspend); | |
6959 | EXPORT_SYMBOL(spa_async_resume); | |
6960 | EXPORT_SYMBOL(spa_inject_addref); | |
6961 | EXPORT_SYMBOL(spa_inject_delref); | |
6962 | EXPORT_SYMBOL(spa_scan_stat_init); | |
6963 | EXPORT_SYMBOL(spa_scan_get_stats); | |
6964 | ||
6965 | /* device maniion */ | |
6966 | EXPORT_SYMBOL(spa_vdev_add); | |
6967 | EXPORT_SYMBOL(spa_vdev_attach); | |
6968 | EXPORT_SYMBOL(spa_vdev_detach); | |
6969 | EXPORT_SYMBOL(spa_vdev_remove); | |
6970 | EXPORT_SYMBOL(spa_vdev_setpath); | |
6971 | EXPORT_SYMBOL(spa_vdev_setfru); | |
6972 | EXPORT_SYMBOL(spa_vdev_split_mirror); | |
6973 | ||
6974 | /* spare statech is global across all pools) */ | |
6975 | EXPORT_SYMBOL(spa_spare_add); | |
6976 | EXPORT_SYMBOL(spa_spare_remove); | |
6977 | EXPORT_SYMBOL(spa_spare_exists); | |
6978 | EXPORT_SYMBOL(spa_spare_activate); | |
6979 | ||
6980 | /* L2ARC statech is global across all pools) */ | |
6981 | EXPORT_SYMBOL(spa_l2cache_add); | |
6982 | EXPORT_SYMBOL(spa_l2cache_remove); | |
6983 | EXPORT_SYMBOL(spa_l2cache_exists); | |
6984 | EXPORT_SYMBOL(spa_l2cache_activate); | |
6985 | EXPORT_SYMBOL(spa_l2cache_drop); | |
6986 | ||
6987 | /* scanning */ | |
6988 | EXPORT_SYMBOL(spa_scan); | |
6989 | EXPORT_SYMBOL(spa_scan_stop); | |
6990 | ||
6991 | /* spa syncing */ | |
6992 | EXPORT_SYMBOL(spa_sync); /* only for DMU use */ | |
6993 | EXPORT_SYMBOL(spa_sync_allpools); | |
6994 | ||
6995 | /* properties */ | |
6996 | EXPORT_SYMBOL(spa_prop_set); | |
6997 | EXPORT_SYMBOL(spa_prop_get); | |
6998 | EXPORT_SYMBOL(spa_prop_clear_bootfs); | |
6999 | ||
7000 | /* asynchronous event notification */ | |
7001 | EXPORT_SYMBOL(spa_event_notify); | |
7002 | #endif | |
7003 | ||
7004 | #if defined(_KERNEL) && defined(HAVE_SPL) | |
7005 | module_param(spa_load_verify_maxinflight, int, 0644); | |
7006 | MODULE_PARM_DESC(spa_load_verify_maxinflight, | |
7007 | "Max concurrent traversal I/Os while verifying pool during import -X"); | |
7008 | ||
7009 | module_param(spa_load_verify_metadata, int, 0644); | |
7010 | MODULE_PARM_DESC(spa_load_verify_metadata, | |
7011 | "Set to traverse metadata on pool import"); | |
7012 | ||
7013 | module_param(spa_load_verify_data, int, 0644); | |
7014 | MODULE_PARM_DESC(spa_load_verify_data, | |
7015 | "Set to traverse data on pool import"); | |
7016 | ||
7017 | /* CSTYLED */ | |
7018 | module_param(zio_taskq_batch_pct, uint, 0444); | |
7019 | MODULE_PARM_DESC(zio_taskq_batch_pct, | |
7020 | "Percentage of CPUs to run an IO worker thread"); | |
7021 | ||
7022 | #endif |