]>
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) 2011, 2017 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) 2014 Integros [integros.com] | |
30 | * Copyright 2016 Toomas Soome <tsoome@me.com> | |
31 | * Copyright (c) 2016 Actifio, Inc. All rights reserved. | |
32 | * Copyright (c) 2017 Datto Inc. | |
33 | * Copyright 2017 Joyent, Inc. | |
34 | */ | |
35 | ||
36 | /* | |
37 | * SPA: Storage Pool Allocator | |
38 | * | |
39 | * This file contains all the routines used when modifying on-disk SPA state. | |
40 | * This includes opening, importing, destroying, exporting a pool, and syncing a | |
41 | * pool. | |
42 | */ | |
43 | ||
44 | #include <sys/zfs_context.h> | |
45 | #include <sys/fm/fs/zfs.h> | |
46 | #include <sys/spa_impl.h> | |
47 | #include <sys/zio.h> | |
48 | #include <sys/zio_checksum.h> | |
49 | #include <sys/dmu.h> | |
50 | #include <sys/dmu_tx.h> | |
51 | #include <sys/zap.h> | |
52 | #include <sys/zil.h> | |
53 | #include <sys/ddt.h> | |
54 | #include <sys/vdev_impl.h> | |
55 | #include <sys/vdev_disk.h> | |
56 | #include <sys/metaslab.h> | |
57 | #include <sys/metaslab_impl.h> | |
58 | #include <sys/mmp.h> | |
59 | #include <sys/uberblock_impl.h> | |
60 | #include <sys/txg.h> | |
61 | #include <sys/avl.h> | |
62 | #include <sys/dmu_traverse.h> | |
63 | #include <sys/dmu_objset.h> | |
64 | #include <sys/unique.h> | |
65 | #include <sys/dsl_pool.h> | |
66 | #include <sys/dsl_dataset.h> | |
67 | #include <sys/dsl_dir.h> | |
68 | #include <sys/dsl_prop.h> | |
69 | #include <sys/dsl_synctask.h> | |
70 | #include <sys/fs/zfs.h> | |
71 | #include <sys/arc.h> | |
72 | #include <sys/callb.h> | |
73 | #include <sys/systeminfo.h> | |
74 | #include <sys/spa_boot.h> | |
75 | #include <sys/zfs_ioctl.h> | |
76 | #include <sys/dsl_scan.h> | |
77 | #include <sys/zfeature.h> | |
78 | #include <sys/dsl_destroy.h> | |
79 | #include <sys/zvol.h> | |
80 | ||
81 | #ifdef _KERNEL | |
82 | #include <sys/fm/protocol.h> | |
83 | #include <sys/fm/util.h> | |
84 | #include <sys/bootprops.h> | |
85 | #include <sys/callb.h> | |
86 | #include <sys/cpupart.h> | |
87 | #include <sys/pool.h> | |
88 | #include <sys/sysdc.h> | |
89 | #include <sys/zone.h> | |
90 | #endif /* _KERNEL */ | |
91 | ||
92 | #include "zfs_prop.h" | |
93 | #include "zfs_comutil.h" | |
94 | ||
95 | /* | |
96 | * The interval, in seconds, at which failed configuration cache file writes | |
97 | * should be retried. | |
98 | */ | |
99 | static int zfs_ccw_retry_interval = 300; | |
100 | ||
101 | typedef enum zti_modes { | |
102 | ZTI_MODE_FIXED, /* value is # of threads (min 1) */ | |
103 | ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */ | |
104 | ZTI_MODE_NULL, /* don't create a taskq */ | |
105 | ZTI_NMODES | |
106 | } zti_modes_t; | |
107 | ||
108 | #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) } | |
109 | #define ZTI_PCT(n) { ZTI_MODE_ONLINE_PERCENT, (n), 1 } | |
110 | #define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 } | |
111 | #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 } | |
112 | ||
113 | #define ZTI_N(n) ZTI_P(n, 1) | |
114 | #define ZTI_ONE ZTI_N(1) | |
115 | ||
116 | typedef struct zio_taskq_info { | |
117 | zti_modes_t zti_mode; | |
118 | uint_t zti_value; | |
119 | uint_t zti_count; | |
120 | } zio_taskq_info_t; | |
121 | ||
122 | static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = { | |
123 | "iss", "iss_h", "int", "int_h" | |
124 | }; | |
125 | ||
126 | /* | |
127 | * This table defines the taskq settings for each ZFS I/O type. When | |
128 | * initializing a pool, we use this table to create an appropriately sized | |
129 | * taskq. Some operations are low volume and therefore have a small, static | |
130 | * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE | |
131 | * macros. Other operations process a large amount of data; the ZTI_BATCH | |
132 | * macro causes us to create a taskq oriented for throughput. Some operations | |
133 | * are so high frequency and short-lived that the taskq itself can become a a | |
134 | * point of lock contention. The ZTI_P(#, #) macro indicates that we need an | |
135 | * additional degree of parallelism specified by the number of threads per- | |
136 | * taskq and the number of taskqs; when dispatching an event in this case, the | |
137 | * particular taskq is chosen at random. | |
138 | * | |
139 | * The different taskq priorities are to handle the different contexts (issue | |
140 | * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that | |
141 | * need to be handled with minimum delay. | |
142 | */ | |
143 | const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = { | |
144 | /* ISSUE ISSUE_HIGH INTR INTR_HIGH */ | |
145 | { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */ | |
146 | { ZTI_N(8), ZTI_NULL, ZTI_P(12, 8), ZTI_NULL }, /* READ */ | |
147 | { ZTI_BATCH, ZTI_N(5), ZTI_P(12, 8), ZTI_N(5) }, /* WRITE */ | |
148 | { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */ | |
149 | { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */ | |
150 | { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */ | |
151 | }; | |
152 | ||
153 | static sysevent_t *spa_event_create(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, | |
154 | const char *name); | |
155 | static void spa_event_post(sysevent_t *ev); | |
156 | static void spa_sync_version(void *arg, dmu_tx_t *tx); | |
157 | static void spa_sync_props(void *arg, dmu_tx_t *tx); | |
158 | static boolean_t spa_has_active_shared_spare(spa_t *spa); | |
159 | static inline int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config, | |
160 | spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig, | |
161 | char **ereport); | |
162 | static void spa_vdev_resilver_done(spa_t *spa); | |
163 | ||
164 | uint_t zio_taskq_batch_pct = 75; /* 1 thread per cpu in pset */ | |
165 | id_t zio_taskq_psrset_bind = PS_NONE; | |
166 | boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */ | |
167 | uint_t zio_taskq_basedc = 80; /* base duty cycle */ | |
168 | ||
169 | boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */ | |
170 | ||
171 | /* | |
172 | * This (illegal) pool name is used when temporarily importing a spa_t in order | |
173 | * to get the vdev stats associated with the imported devices. | |
174 | */ | |
175 | #define TRYIMPORT_NAME "$import" | |
176 | ||
177 | /* | |
178 | * ========================================================================== | |
179 | * SPA properties routines | |
180 | * ========================================================================== | |
181 | */ | |
182 | ||
183 | /* | |
184 | * Add a (source=src, propname=propval) list to an nvlist. | |
185 | */ | |
186 | static void | |
187 | spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval, | |
188 | uint64_t intval, zprop_source_t src) | |
189 | { | |
190 | const char *propname = zpool_prop_to_name(prop); | |
191 | nvlist_t *propval; | |
192 | ||
193 | VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0); | |
194 | VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0); | |
195 | ||
196 | if (strval != NULL) | |
197 | VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0); | |
198 | else | |
199 | VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0); | |
200 | ||
201 | VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0); | |
202 | nvlist_free(propval); | |
203 | } | |
204 | ||
205 | /* | |
206 | * Get property values from the spa configuration. | |
207 | */ | |
208 | static void | |
209 | spa_prop_get_config(spa_t *spa, nvlist_t **nvp) | |
210 | { | |
211 | vdev_t *rvd = spa->spa_root_vdev; | |
212 | dsl_pool_t *pool = spa->spa_dsl_pool; | |
213 | uint64_t size, alloc, cap, version; | |
214 | const zprop_source_t src = ZPROP_SRC_NONE; | |
215 | spa_config_dirent_t *dp; | |
216 | metaslab_class_t *mc = spa_normal_class(spa); | |
217 | ||
218 | ASSERT(MUTEX_HELD(&spa->spa_props_lock)); | |
219 | ||
220 | if (rvd != NULL) { | |
221 | alloc = metaslab_class_get_alloc(spa_normal_class(spa)); | |
222 | size = metaslab_class_get_space(spa_normal_class(spa)); | |
223 | spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src); | |
224 | spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src); | |
225 | spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src); | |
226 | spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL, | |
227 | size - alloc, src); | |
228 | ||
229 | spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL, | |
230 | metaslab_class_fragmentation(mc), src); | |
231 | spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, | |
232 | metaslab_class_expandable_space(mc), src); | |
233 | spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL, | |
234 | (spa_mode(spa) == FREAD), src); | |
235 | ||
236 | cap = (size == 0) ? 0 : (alloc * 100 / size); | |
237 | spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src); | |
238 | ||
239 | spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL, | |
240 | ddt_get_pool_dedup_ratio(spa), src); | |
241 | ||
242 | spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL, | |
243 | rvd->vdev_state, src); | |
244 | ||
245 | version = spa_version(spa); | |
246 | if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION)) { | |
247 | spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, | |
248 | version, ZPROP_SRC_DEFAULT); | |
249 | } else { | |
250 | spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, | |
251 | version, ZPROP_SRC_LOCAL); | |
252 | } | |
253 | } | |
254 | ||
255 | if (pool != NULL) { | |
256 | /* | |
257 | * The $FREE directory was introduced in SPA_VERSION_DEADLISTS, | |
258 | * when opening pools before this version freedir will be NULL. | |
259 | */ | |
260 | if (pool->dp_free_dir != NULL) { | |
261 | spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL, | |
262 | dsl_dir_phys(pool->dp_free_dir)->dd_used_bytes, | |
263 | src); | |
264 | } else { | |
265 | spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, | |
266 | NULL, 0, src); | |
267 | } | |
268 | ||
269 | if (pool->dp_leak_dir != NULL) { | |
270 | spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL, | |
271 | dsl_dir_phys(pool->dp_leak_dir)->dd_used_bytes, | |
272 | src); | |
273 | } else { | |
274 | spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, | |
275 | NULL, 0, src); | |
276 | } | |
277 | } | |
278 | ||
279 | spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src); | |
280 | ||
281 | if (spa->spa_comment != NULL) { | |
282 | spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment, | |
283 | 0, ZPROP_SRC_LOCAL); | |
284 | } | |
285 | ||
286 | if (spa->spa_root != NULL) | |
287 | spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root, | |
288 | 0, ZPROP_SRC_LOCAL); | |
289 | ||
290 | if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) { | |
291 | spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL, | |
292 | MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE); | |
293 | } else { | |
294 | spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL, | |
295 | SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE); | |
296 | } | |
297 | ||
298 | if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_DNODE)) { | |
299 | spa_prop_add_list(*nvp, ZPOOL_PROP_MAXDNODESIZE, NULL, | |
300 | DNODE_MAX_SIZE, ZPROP_SRC_NONE); | |
301 | } else { | |
302 | spa_prop_add_list(*nvp, ZPOOL_PROP_MAXDNODESIZE, NULL, | |
303 | DNODE_MIN_SIZE, ZPROP_SRC_NONE); | |
304 | } | |
305 | ||
306 | if ((dp = list_head(&spa->spa_config_list)) != NULL) { | |
307 | if (dp->scd_path == NULL) { | |
308 | spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, | |
309 | "none", 0, ZPROP_SRC_LOCAL); | |
310 | } else if (strcmp(dp->scd_path, spa_config_path) != 0) { | |
311 | spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, | |
312 | dp->scd_path, 0, ZPROP_SRC_LOCAL); | |
313 | } | |
314 | } | |
315 | } | |
316 | ||
317 | /* | |
318 | * Get zpool property values. | |
319 | */ | |
320 | int | |
321 | spa_prop_get(spa_t *spa, nvlist_t **nvp) | |
322 | { | |
323 | objset_t *mos = spa->spa_meta_objset; | |
324 | zap_cursor_t zc; | |
325 | zap_attribute_t za; | |
326 | int err; | |
327 | ||
328 | err = nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP); | |
329 | if (err) | |
330 | return (err); | |
331 | ||
332 | mutex_enter(&spa->spa_props_lock); | |
333 | ||
334 | /* | |
335 | * Get properties from the spa config. | |
336 | */ | |
337 | spa_prop_get_config(spa, nvp); | |
338 | ||
339 | /* If no pool property object, no more prop to get. */ | |
340 | if (mos == NULL || spa->spa_pool_props_object == 0) { | |
341 | mutex_exit(&spa->spa_props_lock); | |
342 | goto out; | |
343 | } | |
344 | ||
345 | /* | |
346 | * Get properties from the MOS pool property object. | |
347 | */ | |
348 | for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object); | |
349 | (err = zap_cursor_retrieve(&zc, &za)) == 0; | |
350 | zap_cursor_advance(&zc)) { | |
351 | uint64_t intval = 0; | |
352 | char *strval = NULL; | |
353 | zprop_source_t src = ZPROP_SRC_DEFAULT; | |
354 | zpool_prop_t prop; | |
355 | ||
356 | if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL) | |
357 | continue; | |
358 | ||
359 | switch (za.za_integer_length) { | |
360 | case 8: | |
361 | /* integer property */ | |
362 | if (za.za_first_integer != | |
363 | zpool_prop_default_numeric(prop)) | |
364 | src = ZPROP_SRC_LOCAL; | |
365 | ||
366 | if (prop == ZPOOL_PROP_BOOTFS) { | |
367 | dsl_pool_t *dp; | |
368 | dsl_dataset_t *ds = NULL; | |
369 | ||
370 | dp = spa_get_dsl(spa); | |
371 | dsl_pool_config_enter(dp, FTAG); | |
372 | if ((err = dsl_dataset_hold_obj(dp, | |
373 | za.za_first_integer, FTAG, &ds))) { | |
374 | dsl_pool_config_exit(dp, FTAG); | |
375 | break; | |
376 | } | |
377 | ||
378 | strval = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, | |
379 | KM_SLEEP); | |
380 | dsl_dataset_name(ds, strval); | |
381 | dsl_dataset_rele(ds, FTAG); | |
382 | dsl_pool_config_exit(dp, FTAG); | |
383 | } else { | |
384 | strval = NULL; | |
385 | intval = za.za_first_integer; | |
386 | } | |
387 | ||
388 | spa_prop_add_list(*nvp, prop, strval, intval, src); | |
389 | ||
390 | if (strval != NULL) | |
391 | kmem_free(strval, ZFS_MAX_DATASET_NAME_LEN); | |
392 | ||
393 | break; | |
394 | ||
395 | case 1: | |
396 | /* string property */ | |
397 | strval = kmem_alloc(za.za_num_integers, KM_SLEEP); | |
398 | err = zap_lookup(mos, spa->spa_pool_props_object, | |
399 | za.za_name, 1, za.za_num_integers, strval); | |
400 | if (err) { | |
401 | kmem_free(strval, za.za_num_integers); | |
402 | break; | |
403 | } | |
404 | spa_prop_add_list(*nvp, prop, strval, 0, src); | |
405 | kmem_free(strval, za.za_num_integers); | |
406 | break; | |
407 | ||
408 | default: | |
409 | break; | |
410 | } | |
411 | } | |
412 | zap_cursor_fini(&zc); | |
413 | mutex_exit(&spa->spa_props_lock); | |
414 | out: | |
415 | if (err && err != ENOENT) { | |
416 | nvlist_free(*nvp); | |
417 | *nvp = NULL; | |
418 | return (err); | |
419 | } | |
420 | ||
421 | return (0); | |
422 | } | |
423 | ||
424 | /* | |
425 | * Validate the given pool properties nvlist and modify the list | |
426 | * for the property values to be set. | |
427 | */ | |
428 | static int | |
429 | spa_prop_validate(spa_t *spa, nvlist_t *props) | |
430 | { | |
431 | nvpair_t *elem; | |
432 | int error = 0, reset_bootfs = 0; | |
433 | uint64_t objnum = 0; | |
434 | boolean_t has_feature = B_FALSE; | |
435 | ||
436 | elem = NULL; | |
437 | while ((elem = nvlist_next_nvpair(props, elem)) != NULL) { | |
438 | uint64_t intval; | |
439 | char *strval, *slash, *check, *fname; | |
440 | const char *propname = nvpair_name(elem); | |
441 | zpool_prop_t prop = zpool_name_to_prop(propname); | |
442 | ||
443 | switch ((int)prop) { | |
444 | case ZPROP_INVAL: | |
445 | if (!zpool_prop_feature(propname)) { | |
446 | error = SET_ERROR(EINVAL); | |
447 | break; | |
448 | } | |
449 | ||
450 | /* | |
451 | * Sanitize the input. | |
452 | */ | |
453 | if (nvpair_type(elem) != DATA_TYPE_UINT64) { | |
454 | error = SET_ERROR(EINVAL); | |
455 | break; | |
456 | } | |
457 | ||
458 | if (nvpair_value_uint64(elem, &intval) != 0) { | |
459 | error = SET_ERROR(EINVAL); | |
460 | break; | |
461 | } | |
462 | ||
463 | if (intval != 0) { | |
464 | error = SET_ERROR(EINVAL); | |
465 | break; | |
466 | } | |
467 | ||
468 | fname = strchr(propname, '@') + 1; | |
469 | if (zfeature_lookup_name(fname, NULL) != 0) { | |
470 | error = SET_ERROR(EINVAL); | |
471 | break; | |
472 | } | |
473 | ||
474 | has_feature = B_TRUE; | |
475 | break; | |
476 | ||
477 | case ZPOOL_PROP_VERSION: | |
478 | error = nvpair_value_uint64(elem, &intval); | |
479 | if (!error && | |
480 | (intval < spa_version(spa) || | |
481 | intval > SPA_VERSION_BEFORE_FEATURES || | |
482 | has_feature)) | |
483 | error = SET_ERROR(EINVAL); | |
484 | break; | |
485 | ||
486 | case ZPOOL_PROP_DELEGATION: | |
487 | case ZPOOL_PROP_AUTOREPLACE: | |
488 | case ZPOOL_PROP_LISTSNAPS: | |
489 | case ZPOOL_PROP_AUTOEXPAND: | |
490 | error = nvpair_value_uint64(elem, &intval); | |
491 | if (!error && intval > 1) | |
492 | error = SET_ERROR(EINVAL); | |
493 | break; | |
494 | ||
495 | case ZPOOL_PROP_MULTIHOST: | |
496 | error = nvpair_value_uint64(elem, &intval); | |
497 | if (!error && intval > 1) | |
498 | error = SET_ERROR(EINVAL); | |
499 | ||
500 | if (!error && !spa_get_hostid()) | |
501 | error = SET_ERROR(ENOTSUP); | |
502 | ||
503 | break; | |
504 | ||
505 | case ZPOOL_PROP_BOOTFS: | |
506 | /* | |
507 | * If the pool version is less than SPA_VERSION_BOOTFS, | |
508 | * or the pool is still being created (version == 0), | |
509 | * the bootfs property cannot be set. | |
510 | */ | |
511 | if (spa_version(spa) < SPA_VERSION_BOOTFS) { | |
512 | error = SET_ERROR(ENOTSUP); | |
513 | break; | |
514 | } | |
515 | ||
516 | /* | |
517 | * Make sure the vdev config is bootable | |
518 | */ | |
519 | if (!vdev_is_bootable(spa->spa_root_vdev)) { | |
520 | error = SET_ERROR(ENOTSUP); | |
521 | break; | |
522 | } | |
523 | ||
524 | reset_bootfs = 1; | |
525 | ||
526 | error = nvpair_value_string(elem, &strval); | |
527 | ||
528 | if (!error) { | |
529 | objset_t *os; | |
530 | uint64_t propval; | |
531 | ||
532 | if (strval == NULL || strval[0] == '\0') { | |
533 | objnum = zpool_prop_default_numeric( | |
534 | ZPOOL_PROP_BOOTFS); | |
535 | break; | |
536 | } | |
537 | ||
538 | error = dmu_objset_hold(strval, FTAG, &os); | |
539 | if (error) | |
540 | break; | |
541 | ||
542 | /* | |
543 | * Must be ZPL, and its property settings | |
544 | * must be supported by GRUB (compression | |
545 | * is not gzip, and large blocks or large | |
546 | * dnodes are not used). | |
547 | */ | |
548 | ||
549 | if (dmu_objset_type(os) != DMU_OST_ZFS) { | |
550 | error = SET_ERROR(ENOTSUP); | |
551 | } else if ((error = | |
552 | dsl_prop_get_int_ds(dmu_objset_ds(os), | |
553 | zfs_prop_to_name(ZFS_PROP_COMPRESSION), | |
554 | &propval)) == 0 && | |
555 | !BOOTFS_COMPRESS_VALID(propval)) { | |
556 | error = SET_ERROR(ENOTSUP); | |
557 | } else if ((error = | |
558 | dsl_prop_get_int_ds(dmu_objset_ds(os), | |
559 | zfs_prop_to_name(ZFS_PROP_DNODESIZE), | |
560 | &propval)) == 0 && | |
561 | propval != ZFS_DNSIZE_LEGACY) { | |
562 | error = SET_ERROR(ENOTSUP); | |
563 | } else { | |
564 | objnum = dmu_objset_id(os); | |
565 | } | |
566 | dmu_objset_rele(os, FTAG); | |
567 | } | |
568 | break; | |
569 | ||
570 | case ZPOOL_PROP_FAILUREMODE: | |
571 | error = nvpair_value_uint64(elem, &intval); | |
572 | if (!error && intval > ZIO_FAILURE_MODE_PANIC) | |
573 | error = SET_ERROR(EINVAL); | |
574 | ||
575 | /* | |
576 | * This is a special case which only occurs when | |
577 | * the pool has completely failed. This allows | |
578 | * the user to change the in-core failmode property | |
579 | * without syncing it out to disk (I/Os might | |
580 | * currently be blocked). We do this by returning | |
581 | * EIO to the caller (spa_prop_set) to trick it | |
582 | * into thinking we encountered a property validation | |
583 | * error. | |
584 | */ | |
585 | if (!error && spa_suspended(spa)) { | |
586 | spa->spa_failmode = intval; | |
587 | error = SET_ERROR(EIO); | |
588 | } | |
589 | break; | |
590 | ||
591 | case ZPOOL_PROP_CACHEFILE: | |
592 | if ((error = nvpair_value_string(elem, &strval)) != 0) | |
593 | break; | |
594 | ||
595 | if (strval[0] == '\0') | |
596 | break; | |
597 | ||
598 | if (strcmp(strval, "none") == 0) | |
599 | break; | |
600 | ||
601 | if (strval[0] != '/') { | |
602 | error = SET_ERROR(EINVAL); | |
603 | break; | |
604 | } | |
605 | ||
606 | slash = strrchr(strval, '/'); | |
607 | ASSERT(slash != NULL); | |
608 | ||
609 | if (slash[1] == '\0' || strcmp(slash, "/.") == 0 || | |
610 | strcmp(slash, "/..") == 0) | |
611 | error = SET_ERROR(EINVAL); | |
612 | break; | |
613 | ||
614 | case ZPOOL_PROP_COMMENT: | |
615 | if ((error = nvpair_value_string(elem, &strval)) != 0) | |
616 | break; | |
617 | for (check = strval; *check != '\0'; check++) { | |
618 | if (!isprint(*check)) { | |
619 | error = SET_ERROR(EINVAL); | |
620 | break; | |
621 | } | |
622 | } | |
623 | if (strlen(strval) > ZPROP_MAX_COMMENT) | |
624 | error = SET_ERROR(E2BIG); | |
625 | break; | |
626 | ||
627 | case ZPOOL_PROP_DEDUPDITTO: | |
628 | if (spa_version(spa) < SPA_VERSION_DEDUP) | |
629 | error = SET_ERROR(ENOTSUP); | |
630 | else | |
631 | error = nvpair_value_uint64(elem, &intval); | |
632 | if (error == 0 && | |
633 | intval != 0 && intval < ZIO_DEDUPDITTO_MIN) | |
634 | error = SET_ERROR(EINVAL); | |
635 | break; | |
636 | ||
637 | default: | |
638 | break; | |
639 | } | |
640 | ||
641 | if (error) | |
642 | break; | |
643 | } | |
644 | ||
645 | if (!error && reset_bootfs) { | |
646 | error = nvlist_remove(props, | |
647 | zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING); | |
648 | ||
649 | if (!error) { | |
650 | error = nvlist_add_uint64(props, | |
651 | zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum); | |
652 | } | |
653 | } | |
654 | ||
655 | return (error); | |
656 | } | |
657 | ||
658 | void | |
659 | spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync) | |
660 | { | |
661 | char *cachefile; | |
662 | spa_config_dirent_t *dp; | |
663 | ||
664 | if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), | |
665 | &cachefile) != 0) | |
666 | return; | |
667 | ||
668 | dp = kmem_alloc(sizeof (spa_config_dirent_t), | |
669 | KM_SLEEP); | |
670 | ||
671 | if (cachefile[0] == '\0') | |
672 | dp->scd_path = spa_strdup(spa_config_path); | |
673 | else if (strcmp(cachefile, "none") == 0) | |
674 | dp->scd_path = NULL; | |
675 | else | |
676 | dp->scd_path = spa_strdup(cachefile); | |
677 | ||
678 | list_insert_head(&spa->spa_config_list, dp); | |
679 | if (need_sync) | |
680 | spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); | |
681 | } | |
682 | ||
683 | int | |
684 | spa_prop_set(spa_t *spa, nvlist_t *nvp) | |
685 | { | |
686 | int error; | |
687 | nvpair_t *elem = NULL; | |
688 | boolean_t need_sync = B_FALSE; | |
689 | ||
690 | if ((error = spa_prop_validate(spa, nvp)) != 0) | |
691 | return (error); | |
692 | ||
693 | while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) { | |
694 | zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem)); | |
695 | ||
696 | if (prop == ZPOOL_PROP_CACHEFILE || | |
697 | prop == ZPOOL_PROP_ALTROOT || | |
698 | prop == ZPOOL_PROP_READONLY) | |
699 | continue; | |
700 | ||
701 | if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) { | |
702 | uint64_t ver; | |
703 | ||
704 | if (prop == ZPOOL_PROP_VERSION) { | |
705 | VERIFY(nvpair_value_uint64(elem, &ver) == 0); | |
706 | } else { | |
707 | ASSERT(zpool_prop_feature(nvpair_name(elem))); | |
708 | ver = SPA_VERSION_FEATURES; | |
709 | need_sync = B_TRUE; | |
710 | } | |
711 | ||
712 | /* Save time if the version is already set. */ | |
713 | if (ver == spa_version(spa)) | |
714 | continue; | |
715 | ||
716 | /* | |
717 | * In addition to the pool directory object, we might | |
718 | * create the pool properties object, the features for | |
719 | * read object, the features for write object, or the | |
720 | * feature descriptions object. | |
721 | */ | |
722 | error = dsl_sync_task(spa->spa_name, NULL, | |
723 | spa_sync_version, &ver, | |
724 | 6, ZFS_SPACE_CHECK_RESERVED); | |
725 | if (error) | |
726 | return (error); | |
727 | continue; | |
728 | } | |
729 | ||
730 | need_sync = B_TRUE; | |
731 | break; | |
732 | } | |
733 | ||
734 | if (need_sync) { | |
735 | return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props, | |
736 | nvp, 6, ZFS_SPACE_CHECK_RESERVED)); | |
737 | } | |
738 | ||
739 | return (0); | |
740 | } | |
741 | ||
742 | /* | |
743 | * If the bootfs property value is dsobj, clear it. | |
744 | */ | |
745 | void | |
746 | spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx) | |
747 | { | |
748 | if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) { | |
749 | VERIFY(zap_remove(spa->spa_meta_objset, | |
750 | spa->spa_pool_props_object, | |
751 | zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0); | |
752 | spa->spa_bootfs = 0; | |
753 | } | |
754 | } | |
755 | ||
756 | /*ARGSUSED*/ | |
757 | static int | |
758 | spa_change_guid_check(void *arg, dmu_tx_t *tx) | |
759 | { | |
760 | ASSERTV(uint64_t *newguid = arg); | |
761 | spa_t *spa = dmu_tx_pool(tx)->dp_spa; | |
762 | vdev_t *rvd = spa->spa_root_vdev; | |
763 | uint64_t vdev_state; | |
764 | ||
765 | spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); | |
766 | vdev_state = rvd->vdev_state; | |
767 | spa_config_exit(spa, SCL_STATE, FTAG); | |
768 | ||
769 | if (vdev_state != VDEV_STATE_HEALTHY) | |
770 | return (SET_ERROR(ENXIO)); | |
771 | ||
772 | ASSERT3U(spa_guid(spa), !=, *newguid); | |
773 | ||
774 | return (0); | |
775 | } | |
776 | ||
777 | static void | |
778 | spa_change_guid_sync(void *arg, dmu_tx_t *tx) | |
779 | { | |
780 | uint64_t *newguid = arg; | |
781 | spa_t *spa = dmu_tx_pool(tx)->dp_spa; | |
782 | uint64_t oldguid; | |
783 | vdev_t *rvd = spa->spa_root_vdev; | |
784 | ||
785 | oldguid = spa_guid(spa); | |
786 | ||
787 | spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); | |
788 | rvd->vdev_guid = *newguid; | |
789 | rvd->vdev_guid_sum += (*newguid - oldguid); | |
790 | vdev_config_dirty(rvd); | |
791 | spa_config_exit(spa, SCL_STATE, FTAG); | |
792 | ||
793 | spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu", | |
794 | oldguid, *newguid); | |
795 | } | |
796 | ||
797 | /* | |
798 | * Change the GUID for the pool. This is done so that we can later | |
799 | * re-import a pool built from a clone of our own vdevs. We will modify | |
800 | * the root vdev's guid, our own pool guid, and then mark all of our | |
801 | * vdevs dirty. Note that we must make sure that all our vdevs are | |
802 | * online when we do this, or else any vdevs that weren't present | |
803 | * would be orphaned from our pool. We are also going to issue a | |
804 | * sysevent to update any watchers. | |
805 | */ | |
806 | int | |
807 | spa_change_guid(spa_t *spa) | |
808 | { | |
809 | int error; | |
810 | uint64_t guid; | |
811 | ||
812 | mutex_enter(&spa->spa_vdev_top_lock); | |
813 | mutex_enter(&spa_namespace_lock); | |
814 | guid = spa_generate_guid(NULL); | |
815 | ||
816 | error = dsl_sync_task(spa->spa_name, spa_change_guid_check, | |
817 | spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED); | |
818 | ||
819 | if (error == 0) { | |
820 | spa_config_sync(spa, B_FALSE, B_TRUE); | |
821 | spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_REGUID); | |
822 | } | |
823 | ||
824 | mutex_exit(&spa_namespace_lock); | |
825 | mutex_exit(&spa->spa_vdev_top_lock); | |
826 | ||
827 | return (error); | |
828 | } | |
829 | ||
830 | /* | |
831 | * ========================================================================== | |
832 | * SPA state manipulation (open/create/destroy/import/export) | |
833 | * ========================================================================== | |
834 | */ | |
835 | ||
836 | static int | |
837 | spa_error_entry_compare(const void *a, const void *b) | |
838 | { | |
839 | const spa_error_entry_t *sa = (const spa_error_entry_t *)a; | |
840 | const spa_error_entry_t *sb = (const spa_error_entry_t *)b; | |
841 | int ret; | |
842 | ||
843 | ret = memcmp(&sa->se_bookmark, &sb->se_bookmark, | |
844 | sizeof (zbookmark_phys_t)); | |
845 | ||
846 | return (AVL_ISIGN(ret)); | |
847 | } | |
848 | ||
849 | /* | |
850 | * Utility function which retrieves copies of the current logs and | |
851 | * re-initializes them in the process. | |
852 | */ | |
853 | void | |
854 | spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub) | |
855 | { | |
856 | ASSERT(MUTEX_HELD(&spa->spa_errlist_lock)); | |
857 | ||
858 | bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t)); | |
859 | bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t)); | |
860 | ||
861 | avl_create(&spa->spa_errlist_scrub, | |
862 | spa_error_entry_compare, sizeof (spa_error_entry_t), | |
863 | offsetof(spa_error_entry_t, se_avl)); | |
864 | avl_create(&spa->spa_errlist_last, | |
865 | spa_error_entry_compare, sizeof (spa_error_entry_t), | |
866 | offsetof(spa_error_entry_t, se_avl)); | |
867 | } | |
868 | ||
869 | static void | |
870 | spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q) | |
871 | { | |
872 | const zio_taskq_info_t *ztip = &zio_taskqs[t][q]; | |
873 | enum zti_modes mode = ztip->zti_mode; | |
874 | uint_t value = ztip->zti_value; | |
875 | uint_t count = ztip->zti_count; | |
876 | spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; | |
877 | char name[32]; | |
878 | uint_t flags = 0; | |
879 | boolean_t batch = B_FALSE; | |
880 | ||
881 | if (mode == ZTI_MODE_NULL) { | |
882 | tqs->stqs_count = 0; | |
883 | tqs->stqs_taskq = NULL; | |
884 | return; | |
885 | } | |
886 | ||
887 | ASSERT3U(count, >, 0); | |
888 | ||
889 | tqs->stqs_count = count; | |
890 | tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP); | |
891 | ||
892 | switch (mode) { | |
893 | case ZTI_MODE_FIXED: | |
894 | ASSERT3U(value, >=, 1); | |
895 | value = MAX(value, 1); | |
896 | flags |= TASKQ_DYNAMIC; | |
897 | break; | |
898 | ||
899 | case ZTI_MODE_BATCH: | |
900 | batch = B_TRUE; | |
901 | flags |= TASKQ_THREADS_CPU_PCT; | |
902 | value = MIN(zio_taskq_batch_pct, 100); | |
903 | break; | |
904 | ||
905 | default: | |
906 | panic("unrecognized mode for %s_%s taskq (%u:%u) in " | |
907 | "spa_activate()", | |
908 | zio_type_name[t], zio_taskq_types[q], mode, value); | |
909 | break; | |
910 | } | |
911 | ||
912 | for (uint_t i = 0; i < count; i++) { | |
913 | taskq_t *tq; | |
914 | ||
915 | if (count > 1) { | |
916 | (void) snprintf(name, sizeof (name), "%s_%s_%u", | |
917 | zio_type_name[t], zio_taskq_types[q], i); | |
918 | } else { | |
919 | (void) snprintf(name, sizeof (name), "%s_%s", | |
920 | zio_type_name[t], zio_taskq_types[q]); | |
921 | } | |
922 | ||
923 | if (zio_taskq_sysdc && spa->spa_proc != &p0) { | |
924 | if (batch) | |
925 | flags |= TASKQ_DC_BATCH; | |
926 | ||
927 | tq = taskq_create_sysdc(name, value, 50, INT_MAX, | |
928 | spa->spa_proc, zio_taskq_basedc, flags); | |
929 | } else { | |
930 | pri_t pri = maxclsyspri; | |
931 | /* | |
932 | * The write issue taskq can be extremely CPU | |
933 | * intensive. Run it at slightly less important | |
934 | * priority than the other taskqs. Under Linux this | |
935 | * means incrementing the priority value on platforms | |
936 | * like illumos it should be decremented. | |
937 | */ | |
938 | if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE) | |
939 | pri++; | |
940 | ||
941 | tq = taskq_create_proc(name, value, pri, 50, | |
942 | INT_MAX, spa->spa_proc, flags); | |
943 | } | |
944 | ||
945 | tqs->stqs_taskq[i] = tq; | |
946 | } | |
947 | } | |
948 | ||
949 | static void | |
950 | spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q) | |
951 | { | |
952 | spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; | |
953 | ||
954 | if (tqs->stqs_taskq == NULL) { | |
955 | ASSERT3U(tqs->stqs_count, ==, 0); | |
956 | return; | |
957 | } | |
958 | ||
959 | for (uint_t i = 0; i < tqs->stqs_count; i++) { | |
960 | ASSERT3P(tqs->stqs_taskq[i], !=, NULL); | |
961 | taskq_destroy(tqs->stqs_taskq[i]); | |
962 | } | |
963 | ||
964 | kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *)); | |
965 | tqs->stqs_taskq = NULL; | |
966 | } | |
967 | ||
968 | /* | |
969 | * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority. | |
970 | * Note that a type may have multiple discrete taskqs to avoid lock contention | |
971 | * on the taskq itself. In that case we choose which taskq at random by using | |
972 | * the low bits of gethrtime(). | |
973 | */ | |
974 | void | |
975 | spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q, | |
976 | task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent) | |
977 | { | |
978 | spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; | |
979 | taskq_t *tq; | |
980 | ||
981 | ASSERT3P(tqs->stqs_taskq, !=, NULL); | |
982 | ASSERT3U(tqs->stqs_count, !=, 0); | |
983 | ||
984 | if (tqs->stqs_count == 1) { | |
985 | tq = tqs->stqs_taskq[0]; | |
986 | } else { | |
987 | tq = tqs->stqs_taskq[((uint64_t)gethrtime()) % tqs->stqs_count]; | |
988 | } | |
989 | ||
990 | taskq_dispatch_ent(tq, func, arg, flags, ent); | |
991 | } | |
992 | ||
993 | /* | |
994 | * Same as spa_taskq_dispatch_ent() but block on the task until completion. | |
995 | */ | |
996 | void | |
997 | spa_taskq_dispatch_sync(spa_t *spa, zio_type_t t, zio_taskq_type_t q, | |
998 | task_func_t *func, void *arg, uint_t flags) | |
999 | { | |
1000 | spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; | |
1001 | taskq_t *tq; | |
1002 | taskqid_t id; | |
1003 | ||
1004 | ASSERT3P(tqs->stqs_taskq, !=, NULL); | |
1005 | ASSERT3U(tqs->stqs_count, !=, 0); | |
1006 | ||
1007 | if (tqs->stqs_count == 1) { | |
1008 | tq = tqs->stqs_taskq[0]; | |
1009 | } else { | |
1010 | tq = tqs->stqs_taskq[((uint64_t)gethrtime()) % tqs->stqs_count]; | |
1011 | } | |
1012 | ||
1013 | id = taskq_dispatch(tq, func, arg, flags); | |
1014 | if (id) | |
1015 | taskq_wait_id(tq, id); | |
1016 | } | |
1017 | ||
1018 | static void | |
1019 | spa_create_zio_taskqs(spa_t *spa) | |
1020 | { | |
1021 | for (int t = 0; t < ZIO_TYPES; t++) { | |
1022 | for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { | |
1023 | spa_taskqs_init(spa, t, q); | |
1024 | } | |
1025 | } | |
1026 | } | |
1027 | ||
1028 | /* | |
1029 | * Disabled until spa_thread() can be adapted for Linux. | |
1030 | */ | |
1031 | #undef HAVE_SPA_THREAD | |
1032 | ||
1033 | #if defined(_KERNEL) && defined(HAVE_SPA_THREAD) | |
1034 | static void | |
1035 | spa_thread(void *arg) | |
1036 | { | |
1037 | callb_cpr_t cprinfo; | |
1038 | ||
1039 | spa_t *spa = arg; | |
1040 | user_t *pu = PTOU(curproc); | |
1041 | ||
1042 | CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr, | |
1043 | spa->spa_name); | |
1044 | ||
1045 | ASSERT(curproc != &p0); | |
1046 | (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs), | |
1047 | "zpool-%s", spa->spa_name); | |
1048 | (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm)); | |
1049 | ||
1050 | /* bind this thread to the requested psrset */ | |
1051 | if (zio_taskq_psrset_bind != PS_NONE) { | |
1052 | pool_lock(); | |
1053 | mutex_enter(&cpu_lock); | |
1054 | mutex_enter(&pidlock); | |
1055 | mutex_enter(&curproc->p_lock); | |
1056 | ||
1057 | if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind, | |
1058 | 0, NULL, NULL) == 0) { | |
1059 | curthread->t_bind_pset = zio_taskq_psrset_bind; | |
1060 | } else { | |
1061 | cmn_err(CE_WARN, | |
1062 | "Couldn't bind process for zfs pool \"%s\" to " | |
1063 | "pset %d\n", spa->spa_name, zio_taskq_psrset_bind); | |
1064 | } | |
1065 | ||
1066 | mutex_exit(&curproc->p_lock); | |
1067 | mutex_exit(&pidlock); | |
1068 | mutex_exit(&cpu_lock); | |
1069 | pool_unlock(); | |
1070 | } | |
1071 | ||
1072 | if (zio_taskq_sysdc) { | |
1073 | sysdc_thread_enter(curthread, 100, 0); | |
1074 | } | |
1075 | ||
1076 | spa->spa_proc = curproc; | |
1077 | spa->spa_did = curthread->t_did; | |
1078 | ||
1079 | spa_create_zio_taskqs(spa); | |
1080 | ||
1081 | mutex_enter(&spa->spa_proc_lock); | |
1082 | ASSERT(spa->spa_proc_state == SPA_PROC_CREATED); | |
1083 | ||
1084 | spa->spa_proc_state = SPA_PROC_ACTIVE; | |
1085 | cv_broadcast(&spa->spa_proc_cv); | |
1086 | ||
1087 | CALLB_CPR_SAFE_BEGIN(&cprinfo); | |
1088 | while (spa->spa_proc_state == SPA_PROC_ACTIVE) | |
1089 | cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); | |
1090 | CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock); | |
1091 | ||
1092 | ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE); | |
1093 | spa->spa_proc_state = SPA_PROC_GONE; | |
1094 | spa->spa_proc = &p0; | |
1095 | cv_broadcast(&spa->spa_proc_cv); | |
1096 | CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */ | |
1097 | ||
1098 | mutex_enter(&curproc->p_lock); | |
1099 | lwp_exit(); | |
1100 | } | |
1101 | #endif | |
1102 | ||
1103 | /* | |
1104 | * Activate an uninitialized pool. | |
1105 | */ | |
1106 | static void | |
1107 | spa_activate(spa_t *spa, int mode) | |
1108 | { | |
1109 | ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); | |
1110 | ||
1111 | spa->spa_state = POOL_STATE_ACTIVE; | |
1112 | spa->spa_mode = mode; | |
1113 | ||
1114 | spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops); | |
1115 | spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops); | |
1116 | ||
1117 | /* Try to create a covering process */ | |
1118 | mutex_enter(&spa->spa_proc_lock); | |
1119 | ASSERT(spa->spa_proc_state == SPA_PROC_NONE); | |
1120 | ASSERT(spa->spa_proc == &p0); | |
1121 | spa->spa_did = 0; | |
1122 | ||
1123 | #ifdef HAVE_SPA_THREAD | |
1124 | /* Only create a process if we're going to be around a while. */ | |
1125 | if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) { | |
1126 | if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri, | |
1127 | NULL, 0) == 0) { | |
1128 | spa->spa_proc_state = SPA_PROC_CREATED; | |
1129 | while (spa->spa_proc_state == SPA_PROC_CREATED) { | |
1130 | cv_wait(&spa->spa_proc_cv, | |
1131 | &spa->spa_proc_lock); | |
1132 | } | |
1133 | ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); | |
1134 | ASSERT(spa->spa_proc != &p0); | |
1135 | ASSERT(spa->spa_did != 0); | |
1136 | } else { | |
1137 | #ifdef _KERNEL | |
1138 | cmn_err(CE_WARN, | |
1139 | "Couldn't create process for zfs pool \"%s\"\n", | |
1140 | spa->spa_name); | |
1141 | #endif | |
1142 | } | |
1143 | } | |
1144 | #endif /* HAVE_SPA_THREAD */ | |
1145 | mutex_exit(&spa->spa_proc_lock); | |
1146 | ||
1147 | /* If we didn't create a process, we need to create our taskqs. */ | |
1148 | if (spa->spa_proc == &p0) { | |
1149 | spa_create_zio_taskqs(spa); | |
1150 | } | |
1151 | ||
1152 | list_create(&spa->spa_config_dirty_list, sizeof (vdev_t), | |
1153 | offsetof(vdev_t, vdev_config_dirty_node)); | |
1154 | list_create(&spa->spa_evicting_os_list, sizeof (objset_t), | |
1155 | offsetof(objset_t, os_evicting_node)); | |
1156 | list_create(&spa->spa_state_dirty_list, sizeof (vdev_t), | |
1157 | offsetof(vdev_t, vdev_state_dirty_node)); | |
1158 | ||
1159 | txg_list_create(&spa->spa_vdev_txg_list, spa, | |
1160 | offsetof(struct vdev, vdev_txg_node)); | |
1161 | ||
1162 | avl_create(&spa->spa_errlist_scrub, | |
1163 | spa_error_entry_compare, sizeof (spa_error_entry_t), | |
1164 | offsetof(spa_error_entry_t, se_avl)); | |
1165 | avl_create(&spa->spa_errlist_last, | |
1166 | spa_error_entry_compare, sizeof (spa_error_entry_t), | |
1167 | offsetof(spa_error_entry_t, se_avl)); | |
1168 | ||
1169 | spa_keystore_init(&spa->spa_keystore); | |
1170 | ||
1171 | /* | |
1172 | * This taskq is used to perform zvol-minor-related tasks | |
1173 | * asynchronously. This has several advantages, including easy | |
1174 | * resolution of various deadlocks (zfsonlinux bug #3681). | |
1175 | * | |
1176 | * The taskq must be single threaded to ensure tasks are always | |
1177 | * processed in the order in which they were dispatched. | |
1178 | * | |
1179 | * A taskq per pool allows one to keep the pools independent. | |
1180 | * This way if one pool is suspended, it will not impact another. | |
1181 | * | |
1182 | * The preferred location to dispatch a zvol minor task is a sync | |
1183 | * task. In this context, there is easy access to the spa_t and minimal | |
1184 | * error handling is required because the sync task must succeed. | |
1185 | */ | |
1186 | spa->spa_zvol_taskq = taskq_create("z_zvol", 1, defclsyspri, | |
1187 | 1, INT_MAX, 0); | |
1188 | ||
1189 | /* | |
1190 | * The taskq to upgrade datasets in this pool. Currently used by | |
1191 | * feature SPA_FEATURE_USEROBJ_ACCOUNTING. | |
1192 | */ | |
1193 | spa->spa_upgrade_taskq = taskq_create("z_upgrade", boot_ncpus, | |
1194 | defclsyspri, 1, INT_MAX, TASKQ_DYNAMIC); | |
1195 | } | |
1196 | ||
1197 | /* | |
1198 | * Opposite of spa_activate(). | |
1199 | */ | |
1200 | static void | |
1201 | spa_deactivate(spa_t *spa) | |
1202 | { | |
1203 | ASSERT(spa->spa_sync_on == B_FALSE); | |
1204 | ASSERT(spa->spa_dsl_pool == NULL); | |
1205 | ASSERT(spa->spa_root_vdev == NULL); | |
1206 | ASSERT(spa->spa_async_zio_root == NULL); | |
1207 | ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED); | |
1208 | ||
1209 | spa_evicting_os_wait(spa); | |
1210 | ||
1211 | if (spa->spa_zvol_taskq) { | |
1212 | taskq_destroy(spa->spa_zvol_taskq); | |
1213 | spa->spa_zvol_taskq = NULL; | |
1214 | } | |
1215 | ||
1216 | if (spa->spa_upgrade_taskq) { | |
1217 | taskq_destroy(spa->spa_upgrade_taskq); | |
1218 | spa->spa_upgrade_taskq = NULL; | |
1219 | } | |
1220 | ||
1221 | txg_list_destroy(&spa->spa_vdev_txg_list); | |
1222 | ||
1223 | list_destroy(&spa->spa_config_dirty_list); | |
1224 | list_destroy(&spa->spa_evicting_os_list); | |
1225 | list_destroy(&spa->spa_state_dirty_list); | |
1226 | ||
1227 | taskq_cancel_id(system_delay_taskq, spa->spa_deadman_tqid); | |
1228 | ||
1229 | for (int t = 0; t < ZIO_TYPES; t++) { | |
1230 | for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { | |
1231 | spa_taskqs_fini(spa, t, q); | |
1232 | } | |
1233 | } | |
1234 | ||
1235 | metaslab_class_destroy(spa->spa_normal_class); | |
1236 | spa->spa_normal_class = NULL; | |
1237 | ||
1238 | metaslab_class_destroy(spa->spa_log_class); | |
1239 | spa->spa_log_class = NULL; | |
1240 | ||
1241 | /* | |
1242 | * If this was part of an import or the open otherwise failed, we may | |
1243 | * still have errors left in the queues. Empty them just in case. | |
1244 | */ | |
1245 | spa_errlog_drain(spa); | |
1246 | avl_destroy(&spa->spa_errlist_scrub); | |
1247 | avl_destroy(&spa->spa_errlist_last); | |
1248 | ||
1249 | spa_keystore_fini(&spa->spa_keystore); | |
1250 | ||
1251 | spa->spa_state = POOL_STATE_UNINITIALIZED; | |
1252 | ||
1253 | mutex_enter(&spa->spa_proc_lock); | |
1254 | if (spa->spa_proc_state != SPA_PROC_NONE) { | |
1255 | ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); | |
1256 | spa->spa_proc_state = SPA_PROC_DEACTIVATE; | |
1257 | cv_broadcast(&spa->spa_proc_cv); | |
1258 | while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) { | |
1259 | ASSERT(spa->spa_proc != &p0); | |
1260 | cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); | |
1261 | } | |
1262 | ASSERT(spa->spa_proc_state == SPA_PROC_GONE); | |
1263 | spa->spa_proc_state = SPA_PROC_NONE; | |
1264 | } | |
1265 | ASSERT(spa->spa_proc == &p0); | |
1266 | mutex_exit(&spa->spa_proc_lock); | |
1267 | ||
1268 | /* | |
1269 | * We want to make sure spa_thread() has actually exited the ZFS | |
1270 | * module, so that the module can't be unloaded out from underneath | |
1271 | * it. | |
1272 | */ | |
1273 | if (spa->spa_did != 0) { | |
1274 | thread_join(spa->spa_did); | |
1275 | spa->spa_did = 0; | |
1276 | } | |
1277 | } | |
1278 | ||
1279 | /* | |
1280 | * Verify a pool configuration, and construct the vdev tree appropriately. This | |
1281 | * will create all the necessary vdevs in the appropriate layout, with each vdev | |
1282 | * in the CLOSED state. This will prep the pool before open/creation/import. | |
1283 | * All vdev validation is done by the vdev_alloc() routine. | |
1284 | */ | |
1285 | static int | |
1286 | spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, | |
1287 | uint_t id, int atype) | |
1288 | { | |
1289 | nvlist_t **child; | |
1290 | uint_t children; | |
1291 | int error; | |
1292 | ||
1293 | if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0) | |
1294 | return (error); | |
1295 | ||
1296 | if ((*vdp)->vdev_ops->vdev_op_leaf) | |
1297 | return (0); | |
1298 | ||
1299 | error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, | |
1300 | &child, &children); | |
1301 | ||
1302 | if (error == ENOENT) | |
1303 | return (0); | |
1304 | ||
1305 | if (error) { | |
1306 | vdev_free(*vdp); | |
1307 | *vdp = NULL; | |
1308 | return (SET_ERROR(EINVAL)); | |
1309 | } | |
1310 | ||
1311 | for (int c = 0; c < children; c++) { | |
1312 | vdev_t *vd; | |
1313 | if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c, | |
1314 | atype)) != 0) { | |
1315 | vdev_free(*vdp); | |
1316 | *vdp = NULL; | |
1317 | return (error); | |
1318 | } | |
1319 | } | |
1320 | ||
1321 | ASSERT(*vdp != NULL); | |
1322 | ||
1323 | return (0); | |
1324 | } | |
1325 | ||
1326 | /* | |
1327 | * Opposite of spa_load(). | |
1328 | */ | |
1329 | static void | |
1330 | spa_unload(spa_t *spa) | |
1331 | { | |
1332 | int i; | |
1333 | ||
1334 | ASSERT(MUTEX_HELD(&spa_namespace_lock)); | |
1335 | ||
1336 | /* | |
1337 | * Stop async tasks. | |
1338 | */ | |
1339 | spa_async_suspend(spa); | |
1340 | ||
1341 | /* | |
1342 | * Stop syncing. | |
1343 | */ | |
1344 | if (spa->spa_sync_on) { | |
1345 | txg_sync_stop(spa->spa_dsl_pool); | |
1346 | spa->spa_sync_on = B_FALSE; | |
1347 | } | |
1348 | ||
1349 | /* | |
1350 | * Even though vdev_free() also calls vdev_metaslab_fini, we need | |
1351 | * to call it earlier, before we wait for async i/o to complete. | |
1352 | * This ensures that there is no async metaslab prefetching, by | |
1353 | * calling taskq_wait(mg_taskq). | |
1354 | */ | |
1355 | if (spa->spa_root_vdev != NULL) { | |
1356 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
1357 | for (int c = 0; c < spa->spa_root_vdev->vdev_children; c++) | |
1358 | vdev_metaslab_fini(spa->spa_root_vdev->vdev_child[c]); | |
1359 | spa_config_exit(spa, SCL_ALL, FTAG); | |
1360 | } | |
1361 | ||
1362 | if (spa->spa_mmp.mmp_thread) | |
1363 | mmp_thread_stop(spa); | |
1364 | ||
1365 | /* | |
1366 | * Wait for any outstanding async I/O to complete. | |
1367 | */ | |
1368 | if (spa->spa_async_zio_root != NULL) { | |
1369 | for (int i = 0; i < max_ncpus; i++) | |
1370 | (void) zio_wait(spa->spa_async_zio_root[i]); | |
1371 | kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *)); | |
1372 | spa->spa_async_zio_root = NULL; | |
1373 | } | |
1374 | ||
1375 | bpobj_close(&spa->spa_deferred_bpobj); | |
1376 | ||
1377 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
1378 | ||
1379 | /* | |
1380 | * Close all vdevs. | |
1381 | */ | |
1382 | if (spa->spa_root_vdev) | |
1383 | vdev_free(spa->spa_root_vdev); | |
1384 | ASSERT(spa->spa_root_vdev == NULL); | |
1385 | ||
1386 | /* | |
1387 | * Close the dsl pool. | |
1388 | */ | |
1389 | if (spa->spa_dsl_pool) { | |
1390 | dsl_pool_close(spa->spa_dsl_pool); | |
1391 | spa->spa_dsl_pool = NULL; | |
1392 | spa->spa_meta_objset = NULL; | |
1393 | } | |
1394 | ||
1395 | ddt_unload(spa); | |
1396 | ||
1397 | /* | |
1398 | * Drop and purge level 2 cache | |
1399 | */ | |
1400 | spa_l2cache_drop(spa); | |
1401 | ||
1402 | for (i = 0; i < spa->spa_spares.sav_count; i++) | |
1403 | vdev_free(spa->spa_spares.sav_vdevs[i]); | |
1404 | if (spa->spa_spares.sav_vdevs) { | |
1405 | kmem_free(spa->spa_spares.sav_vdevs, | |
1406 | spa->spa_spares.sav_count * sizeof (void *)); | |
1407 | spa->spa_spares.sav_vdevs = NULL; | |
1408 | } | |
1409 | if (spa->spa_spares.sav_config) { | |
1410 | nvlist_free(spa->spa_spares.sav_config); | |
1411 | spa->spa_spares.sav_config = NULL; | |
1412 | } | |
1413 | spa->spa_spares.sav_count = 0; | |
1414 | ||
1415 | for (i = 0; i < spa->spa_l2cache.sav_count; i++) { | |
1416 | vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]); | |
1417 | vdev_free(spa->spa_l2cache.sav_vdevs[i]); | |
1418 | } | |
1419 | if (spa->spa_l2cache.sav_vdevs) { | |
1420 | kmem_free(spa->spa_l2cache.sav_vdevs, | |
1421 | spa->spa_l2cache.sav_count * sizeof (void *)); | |
1422 | spa->spa_l2cache.sav_vdevs = NULL; | |
1423 | } | |
1424 | if (spa->spa_l2cache.sav_config) { | |
1425 | nvlist_free(spa->spa_l2cache.sav_config); | |
1426 | spa->spa_l2cache.sav_config = NULL; | |
1427 | } | |
1428 | spa->spa_l2cache.sav_count = 0; | |
1429 | ||
1430 | spa->spa_async_suspended = 0; | |
1431 | ||
1432 | if (spa->spa_comment != NULL) { | |
1433 | spa_strfree(spa->spa_comment); | |
1434 | spa->spa_comment = NULL; | |
1435 | } | |
1436 | ||
1437 | spa_config_exit(spa, SCL_ALL, FTAG); | |
1438 | } | |
1439 | ||
1440 | /* | |
1441 | * Load (or re-load) the current list of vdevs describing the active spares for | |
1442 | * this pool. When this is called, we have some form of basic information in | |
1443 | * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and | |
1444 | * then re-generate a more complete list including status information. | |
1445 | */ | |
1446 | static void | |
1447 | spa_load_spares(spa_t *spa) | |
1448 | { | |
1449 | nvlist_t **spares; | |
1450 | uint_t nspares; | |
1451 | int i; | |
1452 | vdev_t *vd, *tvd; | |
1453 | ||
1454 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); | |
1455 | ||
1456 | /* | |
1457 | * First, close and free any existing spare vdevs. | |
1458 | */ | |
1459 | for (i = 0; i < spa->spa_spares.sav_count; i++) { | |
1460 | vd = spa->spa_spares.sav_vdevs[i]; | |
1461 | ||
1462 | /* Undo the call to spa_activate() below */ | |
1463 | if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, | |
1464 | B_FALSE)) != NULL && tvd->vdev_isspare) | |
1465 | spa_spare_remove(tvd); | |
1466 | vdev_close(vd); | |
1467 | vdev_free(vd); | |
1468 | } | |
1469 | ||
1470 | if (spa->spa_spares.sav_vdevs) | |
1471 | kmem_free(spa->spa_spares.sav_vdevs, | |
1472 | spa->spa_spares.sav_count * sizeof (void *)); | |
1473 | ||
1474 | if (spa->spa_spares.sav_config == NULL) | |
1475 | nspares = 0; | |
1476 | else | |
1477 | VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, | |
1478 | ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); | |
1479 | ||
1480 | spa->spa_spares.sav_count = (int)nspares; | |
1481 | spa->spa_spares.sav_vdevs = NULL; | |
1482 | ||
1483 | if (nspares == 0) | |
1484 | return; | |
1485 | ||
1486 | /* | |
1487 | * Construct the array of vdevs, opening them to get status in the | |
1488 | * process. For each spare, there is potentially two different vdev_t | |
1489 | * structures associated with it: one in the list of spares (used only | |
1490 | * for basic validation purposes) and one in the active vdev | |
1491 | * configuration (if it's spared in). During this phase we open and | |
1492 | * validate each vdev on the spare list. If the vdev also exists in the | |
1493 | * active configuration, then we also mark this vdev as an active spare. | |
1494 | */ | |
1495 | spa->spa_spares.sav_vdevs = kmem_zalloc(nspares * sizeof (void *), | |
1496 | KM_SLEEP); | |
1497 | for (i = 0; i < spa->spa_spares.sav_count; i++) { | |
1498 | VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0, | |
1499 | VDEV_ALLOC_SPARE) == 0); | |
1500 | ASSERT(vd != NULL); | |
1501 | ||
1502 | spa->spa_spares.sav_vdevs[i] = vd; | |
1503 | ||
1504 | if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, | |
1505 | B_FALSE)) != NULL) { | |
1506 | if (!tvd->vdev_isspare) | |
1507 | spa_spare_add(tvd); | |
1508 | ||
1509 | /* | |
1510 | * We only mark the spare active if we were successfully | |
1511 | * able to load the vdev. Otherwise, importing a pool | |
1512 | * with a bad active spare would result in strange | |
1513 | * behavior, because multiple pool would think the spare | |
1514 | * is actively in use. | |
1515 | * | |
1516 | * There is a vulnerability here to an equally bizarre | |
1517 | * circumstance, where a dead active spare is later | |
1518 | * brought back to life (onlined or otherwise). Given | |
1519 | * the rarity of this scenario, and the extra complexity | |
1520 | * it adds, we ignore the possibility. | |
1521 | */ | |
1522 | if (!vdev_is_dead(tvd)) | |
1523 | spa_spare_activate(tvd); | |
1524 | } | |
1525 | ||
1526 | vd->vdev_top = vd; | |
1527 | vd->vdev_aux = &spa->spa_spares; | |
1528 | ||
1529 | if (vdev_open(vd) != 0) | |
1530 | continue; | |
1531 | ||
1532 | if (vdev_validate_aux(vd) == 0) | |
1533 | spa_spare_add(vd); | |
1534 | } | |
1535 | ||
1536 | /* | |
1537 | * Recompute the stashed list of spares, with status information | |
1538 | * this time. | |
1539 | */ | |
1540 | VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, | |
1541 | DATA_TYPE_NVLIST_ARRAY) == 0); | |
1542 | ||
1543 | spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *), | |
1544 | KM_SLEEP); | |
1545 | for (i = 0; i < spa->spa_spares.sav_count; i++) | |
1546 | spares[i] = vdev_config_generate(spa, | |
1547 | spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE); | |
1548 | VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, | |
1549 | ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0); | |
1550 | for (i = 0; i < spa->spa_spares.sav_count; i++) | |
1551 | nvlist_free(spares[i]); | |
1552 | kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *)); | |
1553 | } | |
1554 | ||
1555 | /* | |
1556 | * Load (or re-load) the current list of vdevs describing the active l2cache for | |
1557 | * this pool. When this is called, we have some form of basic information in | |
1558 | * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and | |
1559 | * then re-generate a more complete list including status information. | |
1560 | * Devices which are already active have their details maintained, and are | |
1561 | * not re-opened. | |
1562 | */ | |
1563 | static void | |
1564 | spa_load_l2cache(spa_t *spa) | |
1565 | { | |
1566 | nvlist_t **l2cache = NULL; | |
1567 | uint_t nl2cache; | |
1568 | int i, j, oldnvdevs; | |
1569 | uint64_t guid; | |
1570 | vdev_t *vd, **oldvdevs, **newvdevs; | |
1571 | spa_aux_vdev_t *sav = &spa->spa_l2cache; | |
1572 | ||
1573 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); | |
1574 | ||
1575 | oldvdevs = sav->sav_vdevs; | |
1576 | oldnvdevs = sav->sav_count; | |
1577 | sav->sav_vdevs = NULL; | |
1578 | sav->sav_count = 0; | |
1579 | ||
1580 | if (sav->sav_config == NULL) { | |
1581 | nl2cache = 0; | |
1582 | newvdevs = NULL; | |
1583 | goto out; | |
1584 | } | |
1585 | ||
1586 | VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, | |
1587 | ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); | |
1588 | newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP); | |
1589 | ||
1590 | /* | |
1591 | * Process new nvlist of vdevs. | |
1592 | */ | |
1593 | for (i = 0; i < nl2cache; i++) { | |
1594 | VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID, | |
1595 | &guid) == 0); | |
1596 | ||
1597 | newvdevs[i] = NULL; | |
1598 | for (j = 0; j < oldnvdevs; j++) { | |
1599 | vd = oldvdevs[j]; | |
1600 | if (vd != NULL && guid == vd->vdev_guid) { | |
1601 | /* | |
1602 | * Retain previous vdev for add/remove ops. | |
1603 | */ | |
1604 | newvdevs[i] = vd; | |
1605 | oldvdevs[j] = NULL; | |
1606 | break; | |
1607 | } | |
1608 | } | |
1609 | ||
1610 | if (newvdevs[i] == NULL) { | |
1611 | /* | |
1612 | * Create new vdev | |
1613 | */ | |
1614 | VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0, | |
1615 | VDEV_ALLOC_L2CACHE) == 0); | |
1616 | ASSERT(vd != NULL); | |
1617 | newvdevs[i] = vd; | |
1618 | ||
1619 | /* | |
1620 | * Commit this vdev as an l2cache device, | |
1621 | * even if it fails to open. | |
1622 | */ | |
1623 | spa_l2cache_add(vd); | |
1624 | ||
1625 | vd->vdev_top = vd; | |
1626 | vd->vdev_aux = sav; | |
1627 | ||
1628 | spa_l2cache_activate(vd); | |
1629 | ||
1630 | if (vdev_open(vd) != 0) | |
1631 | continue; | |
1632 | ||
1633 | (void) vdev_validate_aux(vd); | |
1634 | ||
1635 | if (!vdev_is_dead(vd)) | |
1636 | l2arc_add_vdev(spa, vd); | |
1637 | } | |
1638 | } | |
1639 | ||
1640 | sav->sav_vdevs = newvdevs; | |
1641 | sav->sav_count = (int)nl2cache; | |
1642 | ||
1643 | /* | |
1644 | * Recompute the stashed list of l2cache devices, with status | |
1645 | * information this time. | |
1646 | */ | |
1647 | VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE, | |
1648 | DATA_TYPE_NVLIST_ARRAY) == 0); | |
1649 | ||
1650 | if (sav->sav_count > 0) | |
1651 | l2cache = kmem_alloc(sav->sav_count * sizeof (void *), | |
1652 | KM_SLEEP); | |
1653 | for (i = 0; i < sav->sav_count; i++) | |
1654 | l2cache[i] = vdev_config_generate(spa, | |
1655 | sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE); | |
1656 | VERIFY(nvlist_add_nvlist_array(sav->sav_config, | |
1657 | ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0); | |
1658 | ||
1659 | out: | |
1660 | /* | |
1661 | * Purge vdevs that were dropped | |
1662 | */ | |
1663 | for (i = 0; i < oldnvdevs; i++) { | |
1664 | uint64_t pool; | |
1665 | ||
1666 | vd = oldvdevs[i]; | |
1667 | if (vd != NULL) { | |
1668 | ASSERT(vd->vdev_isl2cache); | |
1669 | ||
1670 | if (spa_l2cache_exists(vd->vdev_guid, &pool) && | |
1671 | pool != 0ULL && l2arc_vdev_present(vd)) | |
1672 | l2arc_remove_vdev(vd); | |
1673 | vdev_clear_stats(vd); | |
1674 | vdev_free(vd); | |
1675 | } | |
1676 | } | |
1677 | ||
1678 | if (oldvdevs) | |
1679 | kmem_free(oldvdevs, oldnvdevs * sizeof (void *)); | |
1680 | ||
1681 | for (i = 0; i < sav->sav_count; i++) | |
1682 | nvlist_free(l2cache[i]); | |
1683 | if (sav->sav_count) | |
1684 | kmem_free(l2cache, sav->sav_count * sizeof (void *)); | |
1685 | } | |
1686 | ||
1687 | static int | |
1688 | load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value) | |
1689 | { | |
1690 | dmu_buf_t *db; | |
1691 | char *packed = NULL; | |
1692 | size_t nvsize = 0; | |
1693 | int error; | |
1694 | *value = NULL; | |
1695 | ||
1696 | error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db); | |
1697 | if (error) | |
1698 | return (error); | |
1699 | ||
1700 | nvsize = *(uint64_t *)db->db_data; | |
1701 | dmu_buf_rele(db, FTAG); | |
1702 | ||
1703 | packed = vmem_alloc(nvsize, KM_SLEEP); | |
1704 | error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed, | |
1705 | DMU_READ_PREFETCH); | |
1706 | if (error == 0) | |
1707 | error = nvlist_unpack(packed, nvsize, value, 0); | |
1708 | vmem_free(packed, nvsize); | |
1709 | ||
1710 | return (error); | |
1711 | } | |
1712 | ||
1713 | /* | |
1714 | * Checks to see if the given vdev could not be opened, in which case we post a | |
1715 | * sysevent to notify the autoreplace code that the device has been removed. | |
1716 | */ | |
1717 | static void | |
1718 | spa_check_removed(vdev_t *vd) | |
1719 | { | |
1720 | for (int c = 0; c < vd->vdev_children; c++) | |
1721 | spa_check_removed(vd->vdev_child[c]); | |
1722 | ||
1723 | if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) && | |
1724 | !vd->vdev_ishole) { | |
1725 | zfs_post_autoreplace(vd->vdev_spa, vd); | |
1726 | spa_event_notify(vd->vdev_spa, vd, NULL, ESC_ZFS_VDEV_CHECK); | |
1727 | } | |
1728 | } | |
1729 | ||
1730 | static void | |
1731 | spa_config_valid_zaps(vdev_t *vd, vdev_t *mvd) | |
1732 | { | |
1733 | ASSERT3U(vd->vdev_children, ==, mvd->vdev_children); | |
1734 | ||
1735 | vd->vdev_top_zap = mvd->vdev_top_zap; | |
1736 | vd->vdev_leaf_zap = mvd->vdev_leaf_zap; | |
1737 | ||
1738 | for (uint64_t i = 0; i < vd->vdev_children; i++) { | |
1739 | spa_config_valid_zaps(vd->vdev_child[i], mvd->vdev_child[i]); | |
1740 | } | |
1741 | } | |
1742 | ||
1743 | /* | |
1744 | * Validate the current config against the MOS config | |
1745 | */ | |
1746 | static boolean_t | |
1747 | spa_config_valid(spa_t *spa, nvlist_t *config) | |
1748 | { | |
1749 | vdev_t *mrvd, *rvd = spa->spa_root_vdev; | |
1750 | nvlist_t *nv; | |
1751 | ||
1752 | VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0); | |
1753 | ||
1754 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
1755 | VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0); | |
1756 | ||
1757 | ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children); | |
1758 | ||
1759 | /* | |
1760 | * If we're doing a normal import, then build up any additional | |
1761 | * diagnostic information about missing devices in this config. | |
1762 | * We'll pass this up to the user for further processing. | |
1763 | */ | |
1764 | if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) { | |
1765 | nvlist_t **child, *nv; | |
1766 | uint64_t idx = 0; | |
1767 | ||
1768 | child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t *), | |
1769 | KM_SLEEP); | |
1770 | VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); | |
1771 | ||
1772 | for (int c = 0; c < rvd->vdev_children; c++) { | |
1773 | vdev_t *tvd = rvd->vdev_child[c]; | |
1774 | vdev_t *mtvd = mrvd->vdev_child[c]; | |
1775 | ||
1776 | if (tvd->vdev_ops == &vdev_missing_ops && | |
1777 | mtvd->vdev_ops != &vdev_missing_ops && | |
1778 | mtvd->vdev_islog) | |
1779 | child[idx++] = vdev_config_generate(spa, mtvd, | |
1780 | B_FALSE, 0); | |
1781 | } | |
1782 | ||
1783 | if (idx) { | |
1784 | VERIFY(nvlist_add_nvlist_array(nv, | |
1785 | ZPOOL_CONFIG_CHILDREN, child, idx) == 0); | |
1786 | VERIFY(nvlist_add_nvlist(spa->spa_load_info, | |
1787 | ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0); | |
1788 | ||
1789 | for (int i = 0; i < idx; i++) | |
1790 | nvlist_free(child[i]); | |
1791 | } | |
1792 | nvlist_free(nv); | |
1793 | kmem_free(child, rvd->vdev_children * sizeof (char **)); | |
1794 | } | |
1795 | ||
1796 | /* | |
1797 | * Compare the root vdev tree with the information we have | |
1798 | * from the MOS config (mrvd). Check each top-level vdev | |
1799 | * with the corresponding MOS config top-level (mtvd). | |
1800 | */ | |
1801 | for (int c = 0; c < rvd->vdev_children; c++) { | |
1802 | vdev_t *tvd = rvd->vdev_child[c]; | |
1803 | vdev_t *mtvd = mrvd->vdev_child[c]; | |
1804 | ||
1805 | /* | |
1806 | * Resolve any "missing" vdevs in the current configuration. | |
1807 | * If we find that the MOS config has more accurate information | |
1808 | * about the top-level vdev then use that vdev instead. | |
1809 | */ | |
1810 | if (tvd->vdev_ops == &vdev_missing_ops && | |
1811 | mtvd->vdev_ops != &vdev_missing_ops) { | |
1812 | ||
1813 | if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) | |
1814 | continue; | |
1815 | ||
1816 | /* | |
1817 | * Device specific actions. | |
1818 | */ | |
1819 | if (mtvd->vdev_islog) { | |
1820 | spa_set_log_state(spa, SPA_LOG_CLEAR); | |
1821 | } else { | |
1822 | /* | |
1823 | * XXX - once we have 'readonly' pool | |
1824 | * support we should be able to handle | |
1825 | * missing data devices by transitioning | |
1826 | * the pool to readonly. | |
1827 | */ | |
1828 | continue; | |
1829 | } | |
1830 | ||
1831 | /* | |
1832 | * Swap the missing vdev with the data we were | |
1833 | * able to obtain from the MOS config. | |
1834 | */ | |
1835 | vdev_remove_child(rvd, tvd); | |
1836 | vdev_remove_child(mrvd, mtvd); | |
1837 | ||
1838 | vdev_add_child(rvd, mtvd); | |
1839 | vdev_add_child(mrvd, tvd); | |
1840 | ||
1841 | spa_config_exit(spa, SCL_ALL, FTAG); | |
1842 | vdev_load(mtvd); | |
1843 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
1844 | ||
1845 | vdev_reopen(rvd); | |
1846 | } else { | |
1847 | if (mtvd->vdev_islog) { | |
1848 | /* | |
1849 | * Load the slog device's state from the MOS | |
1850 | * config since it's possible that the label | |
1851 | * does not contain the most up-to-date | |
1852 | * information. | |
1853 | */ | |
1854 | vdev_load_log_state(tvd, mtvd); | |
1855 | vdev_reopen(tvd); | |
1856 | } | |
1857 | ||
1858 | /* | |
1859 | * Per-vdev ZAP info is stored exclusively in the MOS. | |
1860 | */ | |
1861 | spa_config_valid_zaps(tvd, mtvd); | |
1862 | } | |
1863 | } | |
1864 | ||
1865 | vdev_free(mrvd); | |
1866 | spa_config_exit(spa, SCL_ALL, FTAG); | |
1867 | ||
1868 | /* | |
1869 | * Ensure we were able to validate the config. | |
1870 | */ | |
1871 | return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum); | |
1872 | } | |
1873 | ||
1874 | /* | |
1875 | * Check for missing log devices | |
1876 | */ | |
1877 | static boolean_t | |
1878 | spa_check_logs(spa_t *spa) | |
1879 | { | |
1880 | boolean_t rv = B_FALSE; | |
1881 | dsl_pool_t *dp = spa_get_dsl(spa); | |
1882 | ||
1883 | switch (spa->spa_log_state) { | |
1884 | default: | |
1885 | break; | |
1886 | case SPA_LOG_MISSING: | |
1887 | /* need to recheck in case slog has been restored */ | |
1888 | case SPA_LOG_UNKNOWN: | |
1889 | rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj, | |
1890 | zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0); | |
1891 | if (rv) | |
1892 | spa_set_log_state(spa, SPA_LOG_MISSING); | |
1893 | break; | |
1894 | } | |
1895 | return (rv); | |
1896 | } | |
1897 | ||
1898 | static boolean_t | |
1899 | spa_passivate_log(spa_t *spa) | |
1900 | { | |
1901 | vdev_t *rvd = spa->spa_root_vdev; | |
1902 | boolean_t slog_found = B_FALSE; | |
1903 | ||
1904 | ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); | |
1905 | ||
1906 | if (!spa_has_slogs(spa)) | |
1907 | return (B_FALSE); | |
1908 | ||
1909 | for (int c = 0; c < rvd->vdev_children; c++) { | |
1910 | vdev_t *tvd = rvd->vdev_child[c]; | |
1911 | metaslab_group_t *mg = tvd->vdev_mg; | |
1912 | ||
1913 | if (tvd->vdev_islog) { | |
1914 | metaslab_group_passivate(mg); | |
1915 | slog_found = B_TRUE; | |
1916 | } | |
1917 | } | |
1918 | ||
1919 | return (slog_found); | |
1920 | } | |
1921 | ||
1922 | static void | |
1923 | spa_activate_log(spa_t *spa) | |
1924 | { | |
1925 | vdev_t *rvd = spa->spa_root_vdev; | |
1926 | ||
1927 | ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); | |
1928 | ||
1929 | for (int c = 0; c < rvd->vdev_children; c++) { | |
1930 | vdev_t *tvd = rvd->vdev_child[c]; | |
1931 | metaslab_group_t *mg = tvd->vdev_mg; | |
1932 | ||
1933 | if (tvd->vdev_islog) | |
1934 | metaslab_group_activate(mg); | |
1935 | } | |
1936 | } | |
1937 | ||
1938 | int | |
1939 | spa_offline_log(spa_t *spa) | |
1940 | { | |
1941 | int error; | |
1942 | ||
1943 | error = dmu_objset_find(spa_name(spa), zil_vdev_offline, | |
1944 | NULL, DS_FIND_CHILDREN); | |
1945 | if (error == 0) { | |
1946 | /* | |
1947 | * We successfully offlined the log device, sync out the | |
1948 | * current txg so that the "stubby" block can be removed | |
1949 | * by zil_sync(). | |
1950 | */ | |
1951 | txg_wait_synced(spa->spa_dsl_pool, 0); | |
1952 | } | |
1953 | return (error); | |
1954 | } | |
1955 | ||
1956 | static void | |
1957 | spa_aux_check_removed(spa_aux_vdev_t *sav) | |
1958 | { | |
1959 | for (int i = 0; i < sav->sav_count; i++) | |
1960 | spa_check_removed(sav->sav_vdevs[i]); | |
1961 | } | |
1962 | ||
1963 | void | |
1964 | spa_claim_notify(zio_t *zio) | |
1965 | { | |
1966 | spa_t *spa = zio->io_spa; | |
1967 | ||
1968 | if (zio->io_error) | |
1969 | return; | |
1970 | ||
1971 | mutex_enter(&spa->spa_props_lock); /* any mutex will do */ | |
1972 | if (spa->spa_claim_max_txg < zio->io_bp->blk_birth) | |
1973 | spa->spa_claim_max_txg = zio->io_bp->blk_birth; | |
1974 | mutex_exit(&spa->spa_props_lock); | |
1975 | } | |
1976 | ||
1977 | typedef struct spa_load_error { | |
1978 | uint64_t sle_meta_count; | |
1979 | uint64_t sle_data_count; | |
1980 | } spa_load_error_t; | |
1981 | ||
1982 | static void | |
1983 | spa_load_verify_done(zio_t *zio) | |
1984 | { | |
1985 | blkptr_t *bp = zio->io_bp; | |
1986 | spa_load_error_t *sle = zio->io_private; | |
1987 | dmu_object_type_t type = BP_GET_TYPE(bp); | |
1988 | int error = zio->io_error; | |
1989 | spa_t *spa = zio->io_spa; | |
1990 | ||
1991 | abd_free(zio->io_abd); | |
1992 | if (error) { | |
1993 | if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) && | |
1994 | type != DMU_OT_INTENT_LOG) | |
1995 | atomic_inc_64(&sle->sle_meta_count); | |
1996 | else | |
1997 | atomic_inc_64(&sle->sle_data_count); | |
1998 | } | |
1999 | ||
2000 | mutex_enter(&spa->spa_scrub_lock); | |
2001 | spa->spa_load_verify_ios--; | |
2002 | cv_broadcast(&spa->spa_scrub_io_cv); | |
2003 | mutex_exit(&spa->spa_scrub_lock); | |
2004 | } | |
2005 | ||
2006 | /* | |
2007 | * Maximum number of concurrent scrub i/os to create while verifying | |
2008 | * a pool while importing it. | |
2009 | */ | |
2010 | int spa_load_verify_maxinflight = 10000; | |
2011 | int spa_load_verify_metadata = B_TRUE; | |
2012 | int spa_load_verify_data = B_TRUE; | |
2013 | ||
2014 | /*ARGSUSED*/ | |
2015 | static int | |
2016 | spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, | |
2017 | const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg) | |
2018 | { | |
2019 | if (bp == NULL || BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp)) | |
2020 | return (0); | |
2021 | /* | |
2022 | * Note: normally this routine will not be called if | |
2023 | * spa_load_verify_metadata is not set. However, it may be useful | |
2024 | * to manually set the flag after the traversal has begun. | |
2025 | */ | |
2026 | if (!spa_load_verify_metadata) | |
2027 | return (0); | |
2028 | if (!BP_IS_METADATA(bp) && !spa_load_verify_data) | |
2029 | return (0); | |
2030 | ||
2031 | zio_t *rio = arg; | |
2032 | size_t size = BP_GET_PSIZE(bp); | |
2033 | ||
2034 | mutex_enter(&spa->spa_scrub_lock); | |
2035 | while (spa->spa_load_verify_ios >= spa_load_verify_maxinflight) | |
2036 | cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); | |
2037 | spa->spa_load_verify_ios++; | |
2038 | mutex_exit(&spa->spa_scrub_lock); | |
2039 | ||
2040 | zio_nowait(zio_read(rio, spa, bp, abd_alloc_for_io(size, B_FALSE), size, | |
2041 | spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB, | |
2042 | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL | | |
2043 | ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb)); | |
2044 | return (0); | |
2045 | } | |
2046 | ||
2047 | /* ARGSUSED */ | |
2048 | int | |
2049 | verify_dataset_name_len(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg) | |
2050 | { | |
2051 | if (dsl_dataset_namelen(ds) >= ZFS_MAX_DATASET_NAME_LEN) | |
2052 | return (SET_ERROR(ENAMETOOLONG)); | |
2053 | ||
2054 | return (0); | |
2055 | } | |
2056 | ||
2057 | static int | |
2058 | spa_load_verify(spa_t *spa) | |
2059 | { | |
2060 | zio_t *rio; | |
2061 | spa_load_error_t sle = { 0 }; | |
2062 | zpool_rewind_policy_t policy; | |
2063 | boolean_t verify_ok = B_FALSE; | |
2064 | int error = 0; | |
2065 | ||
2066 | zpool_get_rewind_policy(spa->spa_config, &policy); | |
2067 | ||
2068 | if (policy.zrp_request & ZPOOL_NEVER_REWIND) | |
2069 | return (0); | |
2070 | ||
2071 | dsl_pool_config_enter(spa->spa_dsl_pool, FTAG); | |
2072 | error = dmu_objset_find_dp(spa->spa_dsl_pool, | |
2073 | spa->spa_dsl_pool->dp_root_dir_obj, verify_dataset_name_len, NULL, | |
2074 | DS_FIND_CHILDREN); | |
2075 | dsl_pool_config_exit(spa->spa_dsl_pool, FTAG); | |
2076 | if (error != 0) | |
2077 | return (error); | |
2078 | ||
2079 | rio = zio_root(spa, NULL, &sle, | |
2080 | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE); | |
2081 | ||
2082 | if (spa_load_verify_metadata) { | |
2083 | error = traverse_pool(spa, spa->spa_verify_min_txg, | |
2084 | TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA | | |
2085 | TRAVERSE_NO_DECRYPT, spa_load_verify_cb, rio); | |
2086 | } | |
2087 | ||
2088 | (void) zio_wait(rio); | |
2089 | ||
2090 | spa->spa_load_meta_errors = sle.sle_meta_count; | |
2091 | spa->spa_load_data_errors = sle.sle_data_count; | |
2092 | ||
2093 | if (!error && sle.sle_meta_count <= policy.zrp_maxmeta && | |
2094 | sle.sle_data_count <= policy.zrp_maxdata) { | |
2095 | int64_t loss = 0; | |
2096 | ||
2097 | verify_ok = B_TRUE; | |
2098 | spa->spa_load_txg = spa->spa_uberblock.ub_txg; | |
2099 | spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp; | |
2100 | ||
2101 | loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts; | |
2102 | VERIFY(nvlist_add_uint64(spa->spa_load_info, | |
2103 | ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0); | |
2104 | VERIFY(nvlist_add_int64(spa->spa_load_info, | |
2105 | ZPOOL_CONFIG_REWIND_TIME, loss) == 0); | |
2106 | VERIFY(nvlist_add_uint64(spa->spa_load_info, | |
2107 | ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0); | |
2108 | } else { | |
2109 | spa->spa_load_max_txg = spa->spa_uberblock.ub_txg; | |
2110 | } | |
2111 | ||
2112 | if (error) { | |
2113 | if (error != ENXIO && error != EIO) | |
2114 | error = SET_ERROR(EIO); | |
2115 | return (error); | |
2116 | } | |
2117 | ||
2118 | return (verify_ok ? 0 : EIO); | |
2119 | } | |
2120 | ||
2121 | /* | |
2122 | * Find a value in the pool props object. | |
2123 | */ | |
2124 | static void | |
2125 | spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val) | |
2126 | { | |
2127 | (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object, | |
2128 | zpool_prop_to_name(prop), sizeof (uint64_t), 1, val); | |
2129 | } | |
2130 | ||
2131 | /* | |
2132 | * Find a value in the pool directory object. | |
2133 | */ | |
2134 | static int | |
2135 | spa_dir_prop(spa_t *spa, const char *name, uint64_t *val) | |
2136 | { | |
2137 | return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, | |
2138 | name, sizeof (uint64_t), 1, val)); | |
2139 | } | |
2140 | ||
2141 | static int | |
2142 | spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err) | |
2143 | { | |
2144 | vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux); | |
2145 | return (err); | |
2146 | } | |
2147 | ||
2148 | /* | |
2149 | * Fix up config after a partly-completed split. This is done with the | |
2150 | * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off | |
2151 | * pool have that entry in their config, but only the splitting one contains | |
2152 | * a list of all the guids of the vdevs that are being split off. | |
2153 | * | |
2154 | * This function determines what to do with that list: either rejoin | |
2155 | * all the disks to the pool, or complete the splitting process. To attempt | |
2156 | * the rejoin, each disk that is offlined is marked online again, and | |
2157 | * we do a reopen() call. If the vdev label for every disk that was | |
2158 | * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL) | |
2159 | * then we call vdev_split() on each disk, and complete the split. | |
2160 | * | |
2161 | * Otherwise we leave the config alone, with all the vdevs in place in | |
2162 | * the original pool. | |
2163 | */ | |
2164 | static void | |
2165 | spa_try_repair(spa_t *spa, nvlist_t *config) | |
2166 | { | |
2167 | uint_t extracted; | |
2168 | uint64_t *glist; | |
2169 | uint_t i, gcount; | |
2170 | nvlist_t *nvl; | |
2171 | vdev_t **vd; | |
2172 | boolean_t attempt_reopen; | |
2173 | ||
2174 | if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0) | |
2175 | return; | |
2176 | ||
2177 | /* check that the config is complete */ | |
2178 | if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, | |
2179 | &glist, &gcount) != 0) | |
2180 | return; | |
2181 | ||
2182 | vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP); | |
2183 | ||
2184 | /* attempt to online all the vdevs & validate */ | |
2185 | attempt_reopen = B_TRUE; | |
2186 | for (i = 0; i < gcount; i++) { | |
2187 | if (glist[i] == 0) /* vdev is hole */ | |
2188 | continue; | |
2189 | ||
2190 | vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE); | |
2191 | if (vd[i] == NULL) { | |
2192 | /* | |
2193 | * Don't bother attempting to reopen the disks; | |
2194 | * just do the split. | |
2195 | */ | |
2196 | attempt_reopen = B_FALSE; | |
2197 | } else { | |
2198 | /* attempt to re-online it */ | |
2199 | vd[i]->vdev_offline = B_FALSE; | |
2200 | } | |
2201 | } | |
2202 | ||
2203 | if (attempt_reopen) { | |
2204 | vdev_reopen(spa->spa_root_vdev); | |
2205 | ||
2206 | /* check each device to see what state it's in */ | |
2207 | for (extracted = 0, i = 0; i < gcount; i++) { | |
2208 | if (vd[i] != NULL && | |
2209 | vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL) | |
2210 | break; | |
2211 | ++extracted; | |
2212 | } | |
2213 | } | |
2214 | ||
2215 | /* | |
2216 | * If every disk has been moved to the new pool, or if we never | |
2217 | * even attempted to look at them, then we split them off for | |
2218 | * good. | |
2219 | */ | |
2220 | if (!attempt_reopen || gcount == extracted) { | |
2221 | for (i = 0; i < gcount; i++) | |
2222 | if (vd[i] != NULL) | |
2223 | vdev_split(vd[i]); | |
2224 | vdev_reopen(spa->spa_root_vdev); | |
2225 | } | |
2226 | ||
2227 | kmem_free(vd, gcount * sizeof (vdev_t *)); | |
2228 | } | |
2229 | ||
2230 | static int | |
2231 | spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type, | |
2232 | boolean_t mosconfig) | |
2233 | { | |
2234 | nvlist_t *config = spa->spa_config; | |
2235 | char *ereport = FM_EREPORT_ZFS_POOL; | |
2236 | char *comment; | |
2237 | int error; | |
2238 | uint64_t pool_guid; | |
2239 | nvlist_t *nvl; | |
2240 | ||
2241 | if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) | |
2242 | return (SET_ERROR(EINVAL)); | |
2243 | ||
2244 | ASSERT(spa->spa_comment == NULL); | |
2245 | if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0) | |
2246 | spa->spa_comment = spa_strdup(comment); | |
2247 | ||
2248 | /* | |
2249 | * Versioning wasn't explicitly added to the label until later, so if | |
2250 | * it's not present treat it as the initial version. | |
2251 | */ | |
2252 | if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, | |
2253 | &spa->spa_ubsync.ub_version) != 0) | |
2254 | spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; | |
2255 | ||
2256 | (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, | |
2257 | &spa->spa_config_txg); | |
2258 | ||
2259 | if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) && | |
2260 | spa_guid_exists(pool_guid, 0)) { | |
2261 | error = SET_ERROR(EEXIST); | |
2262 | } else { | |
2263 | spa->spa_config_guid = pool_guid; | |
2264 | ||
2265 | if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, | |
2266 | &nvl) == 0) { | |
2267 | VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting, | |
2268 | KM_SLEEP) == 0); | |
2269 | } | |
2270 | ||
2271 | nvlist_free(spa->spa_load_info); | |
2272 | spa->spa_load_info = fnvlist_alloc(); | |
2273 | ||
2274 | gethrestime(&spa->spa_loaded_ts); | |
2275 | error = spa_load_impl(spa, pool_guid, config, state, type, | |
2276 | mosconfig, &ereport); | |
2277 | } | |
2278 | ||
2279 | /* | |
2280 | * Don't count references from objsets that are already closed | |
2281 | * and are making their way through the eviction process. | |
2282 | */ | |
2283 | spa_evicting_os_wait(spa); | |
2284 | spa->spa_minref = refcount_count(&spa->spa_refcount); | |
2285 | if (error) { | |
2286 | if (error != EEXIST) { | |
2287 | spa->spa_loaded_ts.tv_sec = 0; | |
2288 | spa->spa_loaded_ts.tv_nsec = 0; | |
2289 | } | |
2290 | if (error != EBADF) { | |
2291 | zfs_ereport_post(ereport, spa, NULL, NULL, NULL, 0, 0); | |
2292 | } | |
2293 | } | |
2294 | spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE; | |
2295 | spa->spa_ena = 0; | |
2296 | ||
2297 | return (error); | |
2298 | } | |
2299 | ||
2300 | #ifdef ZFS_DEBUG | |
2301 | /* | |
2302 | * Count the number of per-vdev ZAPs associated with all of the vdevs in the | |
2303 | * vdev tree rooted in the given vd, and ensure that each ZAP is present in the | |
2304 | * spa's per-vdev ZAP list. | |
2305 | */ | |
2306 | static uint64_t | |
2307 | vdev_count_verify_zaps(vdev_t *vd) | |
2308 | { | |
2309 | spa_t *spa = vd->vdev_spa; | |
2310 | uint64_t total = 0; | |
2311 | ||
2312 | if (vd->vdev_top_zap != 0) { | |
2313 | total++; | |
2314 | ASSERT0(zap_lookup_int(spa->spa_meta_objset, | |
2315 | spa->spa_all_vdev_zaps, vd->vdev_top_zap)); | |
2316 | } | |
2317 | if (vd->vdev_leaf_zap != 0) { | |
2318 | total++; | |
2319 | ASSERT0(zap_lookup_int(spa->spa_meta_objset, | |
2320 | spa->spa_all_vdev_zaps, vd->vdev_leaf_zap)); | |
2321 | } | |
2322 | ||
2323 | for (uint64_t i = 0; i < vd->vdev_children; i++) { | |
2324 | total += vdev_count_verify_zaps(vd->vdev_child[i]); | |
2325 | } | |
2326 | ||
2327 | return (total); | |
2328 | } | |
2329 | #endif | |
2330 | ||
2331 | /* | |
2332 | * Determine whether the activity check is required. | |
2333 | */ | |
2334 | static boolean_t | |
2335 | spa_activity_check_required(spa_t *spa, uberblock_t *ub, nvlist_t *label, | |
2336 | nvlist_t *config) | |
2337 | { | |
2338 | uint64_t state = 0; | |
2339 | uint64_t hostid = 0; | |
2340 | uint64_t tryconfig_txg = 0; | |
2341 | uint64_t tryconfig_timestamp = 0; | |
2342 | nvlist_t *nvinfo; | |
2343 | ||
2344 | if (nvlist_exists(config, ZPOOL_CONFIG_LOAD_INFO)) { | |
2345 | nvinfo = fnvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO); | |
2346 | (void) nvlist_lookup_uint64(nvinfo, ZPOOL_CONFIG_MMP_TXG, | |
2347 | &tryconfig_txg); | |
2348 | (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_TIMESTAMP, | |
2349 | &tryconfig_timestamp); | |
2350 | } | |
2351 | ||
2352 | (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE, &state); | |
2353 | ||
2354 | /* | |
2355 | * Disable the MMP activity check - This is used by zdb which | |
2356 | * is intended to be used on potentially active pools. | |
2357 | */ | |
2358 | if (spa->spa_import_flags & ZFS_IMPORT_SKIP_MMP) | |
2359 | return (B_FALSE); | |
2360 | ||
2361 | /* | |
2362 | * Skip the activity check when the MMP feature is disabled. | |
2363 | */ | |
2364 | if (ub->ub_mmp_magic == MMP_MAGIC && ub->ub_mmp_delay == 0) | |
2365 | return (B_FALSE); | |
2366 | /* | |
2367 | * If the tryconfig_* values are nonzero, they are the results of an | |
2368 | * earlier tryimport. If they match the uberblock we just found, then | |
2369 | * the pool has not changed and we return false so we do not test a | |
2370 | * second time. | |
2371 | */ | |
2372 | if (tryconfig_txg && tryconfig_txg == ub->ub_txg && | |
2373 | tryconfig_timestamp && tryconfig_timestamp == ub->ub_timestamp) | |
2374 | return (B_FALSE); | |
2375 | ||
2376 | /* | |
2377 | * Allow the activity check to be skipped when importing the pool | |
2378 | * on the same host which last imported it. Since the hostid from | |
2379 | * configuration may be stale use the one read from the label. | |
2380 | */ | |
2381 | if (nvlist_exists(label, ZPOOL_CONFIG_HOSTID)) | |
2382 | hostid = fnvlist_lookup_uint64(label, ZPOOL_CONFIG_HOSTID); | |
2383 | ||
2384 | if (hostid == spa_get_hostid()) | |
2385 | return (B_FALSE); | |
2386 | ||
2387 | /* | |
2388 | * Skip the activity test when the pool was cleanly exported. | |
2389 | */ | |
2390 | if (state != POOL_STATE_ACTIVE) | |
2391 | return (B_FALSE); | |
2392 | ||
2393 | return (B_TRUE); | |
2394 | } | |
2395 | ||
2396 | /* | |
2397 | * Perform the import activity check. If the user canceled the import or | |
2398 | * we detected activity then fail. | |
2399 | */ | |
2400 | static int | |
2401 | spa_activity_check(spa_t *spa, uberblock_t *ub, nvlist_t *config) | |
2402 | { | |
2403 | uint64_t import_intervals = MAX(zfs_multihost_import_intervals, 1); | |
2404 | uint64_t txg = ub->ub_txg; | |
2405 | uint64_t timestamp = ub->ub_timestamp; | |
2406 | uint64_t import_delay = NANOSEC; | |
2407 | hrtime_t import_expire; | |
2408 | nvlist_t *mmp_label = NULL; | |
2409 | vdev_t *rvd = spa->spa_root_vdev; | |
2410 | kcondvar_t cv; | |
2411 | kmutex_t mtx; | |
2412 | int error = 0; | |
2413 | ||
2414 | cv_init(&cv, NULL, CV_DEFAULT, NULL); | |
2415 | mutex_init(&mtx, NULL, MUTEX_DEFAULT, NULL); | |
2416 | mutex_enter(&mtx); | |
2417 | ||
2418 | /* | |
2419 | * If ZPOOL_CONFIG_MMP_TXG is present an activity check was performed | |
2420 | * during the earlier tryimport. If the txg recorded there is 0 then | |
2421 | * the pool is known to be active on another host. | |
2422 | * | |
2423 | * Otherwise, the pool might be in use on another node. Check for | |
2424 | * changes in the uberblocks on disk if necessary. | |
2425 | */ | |
2426 | if (nvlist_exists(config, ZPOOL_CONFIG_LOAD_INFO)) { | |
2427 | nvlist_t *nvinfo = fnvlist_lookup_nvlist(config, | |
2428 | ZPOOL_CONFIG_LOAD_INFO); | |
2429 | ||
2430 | if (nvlist_exists(nvinfo, ZPOOL_CONFIG_MMP_TXG) && | |
2431 | fnvlist_lookup_uint64(nvinfo, ZPOOL_CONFIG_MMP_TXG) == 0) { | |
2432 | vdev_uberblock_load(rvd, ub, &mmp_label); | |
2433 | error = SET_ERROR(EREMOTEIO); | |
2434 | goto out; | |
2435 | } | |
2436 | } | |
2437 | ||
2438 | /* | |
2439 | * Preferentially use the zfs_multihost_interval from the node which | |
2440 | * last imported the pool. This value is stored in an MMP uberblock as. | |
2441 | * | |
2442 | * ub_mmp_delay * vdev_count_leaves() == zfs_multihost_interval | |
2443 | */ | |
2444 | if (ub->ub_mmp_magic == MMP_MAGIC && ub->ub_mmp_delay) | |
2445 | import_delay = MAX(import_delay, import_intervals * | |
2446 | ub->ub_mmp_delay * MAX(vdev_count_leaves(spa), 1)); | |
2447 | ||
2448 | /* Apply a floor using the local default values. */ | |
2449 | import_delay = MAX(import_delay, import_intervals * | |
2450 | MSEC2NSEC(MAX(zfs_multihost_interval, MMP_MIN_INTERVAL))); | |
2451 | ||
2452 | /* Add a small random factor in case of simultaneous imports (0-25%) */ | |
2453 | import_expire = gethrtime() + import_delay + | |
2454 | (import_delay * spa_get_random(250) / 1000); | |
2455 | ||
2456 | while (gethrtime() < import_expire) { | |
2457 | vdev_uberblock_load(rvd, ub, &mmp_label); | |
2458 | ||
2459 | if (txg != ub->ub_txg || timestamp != ub->ub_timestamp) { | |
2460 | error = SET_ERROR(EREMOTEIO); | |
2461 | break; | |
2462 | } | |
2463 | ||
2464 | if (mmp_label) { | |
2465 | nvlist_free(mmp_label); | |
2466 | mmp_label = NULL; | |
2467 | } | |
2468 | ||
2469 | error = cv_timedwait_sig(&cv, &mtx, ddi_get_lbolt() + hz); | |
2470 | if (error != -1) { | |
2471 | error = SET_ERROR(EINTR); | |
2472 | break; | |
2473 | } | |
2474 | error = 0; | |
2475 | } | |
2476 | ||
2477 | out: | |
2478 | mutex_exit(&mtx); | |
2479 | mutex_destroy(&mtx); | |
2480 | cv_destroy(&cv); | |
2481 | ||
2482 | /* | |
2483 | * If the pool is determined to be active store the status in the | |
2484 | * spa->spa_load_info nvlist. If the remote hostname or hostid are | |
2485 | * available from configuration read from disk store them as well. | |
2486 | * This allows 'zpool import' to generate a more useful message. | |
2487 | * | |
2488 | * ZPOOL_CONFIG_MMP_STATE - observed pool status (mandatory) | |
2489 | * ZPOOL_CONFIG_MMP_HOSTNAME - hostname from the active pool | |
2490 | * ZPOOL_CONFIG_MMP_HOSTID - hostid from the active pool | |
2491 | */ | |
2492 | if (error == EREMOTEIO) { | |
2493 | char *hostname = "<unknown>"; | |
2494 | uint64_t hostid = 0; | |
2495 | ||
2496 | if (mmp_label) { | |
2497 | if (nvlist_exists(mmp_label, ZPOOL_CONFIG_HOSTNAME)) { | |
2498 | hostname = fnvlist_lookup_string(mmp_label, | |
2499 | ZPOOL_CONFIG_HOSTNAME); | |
2500 | fnvlist_add_string(spa->spa_load_info, | |
2501 | ZPOOL_CONFIG_MMP_HOSTNAME, hostname); | |
2502 | } | |
2503 | ||
2504 | if (nvlist_exists(mmp_label, ZPOOL_CONFIG_HOSTID)) { | |
2505 | hostid = fnvlist_lookup_uint64(mmp_label, | |
2506 | ZPOOL_CONFIG_HOSTID); | |
2507 | fnvlist_add_uint64(spa->spa_load_info, | |
2508 | ZPOOL_CONFIG_MMP_HOSTID, hostid); | |
2509 | } | |
2510 | } | |
2511 | ||
2512 | fnvlist_add_uint64(spa->spa_load_info, | |
2513 | ZPOOL_CONFIG_MMP_STATE, MMP_STATE_ACTIVE); | |
2514 | fnvlist_add_uint64(spa->spa_load_info, | |
2515 | ZPOOL_CONFIG_MMP_TXG, 0); | |
2516 | ||
2517 | error = spa_vdev_err(rvd, VDEV_AUX_ACTIVE, EREMOTEIO); | |
2518 | } | |
2519 | ||
2520 | if (mmp_label) | |
2521 | nvlist_free(mmp_label); | |
2522 | ||
2523 | return (error); | |
2524 | } | |
2525 | ||
2526 | /* | |
2527 | * Load an existing storage pool, using the pool's builtin spa_config as a | |
2528 | * source of configuration information. | |
2529 | */ | |
2530 | __attribute__((always_inline)) | |
2531 | static inline int | |
2532 | spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config, | |
2533 | spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig, | |
2534 | char **ereport) | |
2535 | { | |
2536 | int error = 0; | |
2537 | nvlist_t *nvroot = NULL; | |
2538 | nvlist_t *label; | |
2539 | vdev_t *rvd; | |
2540 | uberblock_t *ub = &spa->spa_uberblock; | |
2541 | uint64_t children, config_cache_txg = spa->spa_config_txg; | |
2542 | int orig_mode = spa->spa_mode; | |
2543 | int parse; | |
2544 | uint64_t obj; | |
2545 | boolean_t missing_feat_write = B_FALSE; | |
2546 | boolean_t activity_check = B_FALSE; | |
2547 | ||
2548 | /* | |
2549 | * If this is an untrusted config, access the pool in read-only mode. | |
2550 | * This prevents things like resilvering recently removed devices. | |
2551 | */ | |
2552 | if (!mosconfig) | |
2553 | spa->spa_mode = FREAD; | |
2554 | ||
2555 | ASSERT(MUTEX_HELD(&spa_namespace_lock)); | |
2556 | ||
2557 | spa->spa_load_state = state; | |
2558 | ||
2559 | if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot)) | |
2560 | return (SET_ERROR(EINVAL)); | |
2561 | ||
2562 | parse = (type == SPA_IMPORT_EXISTING ? | |
2563 | VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT); | |
2564 | ||
2565 | /* | |
2566 | * Create "The Godfather" zio to hold all async IOs | |
2567 | */ | |
2568 | spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), | |
2569 | KM_SLEEP); | |
2570 | for (int i = 0; i < max_ncpus; i++) { | |
2571 | spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, | |
2572 | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | | |
2573 | ZIO_FLAG_GODFATHER); | |
2574 | } | |
2575 | ||
2576 | /* | |
2577 | * Parse the configuration into a vdev tree. We explicitly set the | |
2578 | * value that will be returned by spa_version() since parsing the | |
2579 | * configuration requires knowing the version number. | |
2580 | */ | |
2581 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
2582 | error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse); | |
2583 | spa_config_exit(spa, SCL_ALL, FTAG); | |
2584 | ||
2585 | if (error != 0) | |
2586 | return (error); | |
2587 | ||
2588 | ASSERT(spa->spa_root_vdev == rvd); | |
2589 | ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT); | |
2590 | ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT); | |
2591 | ||
2592 | if (type != SPA_IMPORT_ASSEMBLE) { | |
2593 | ASSERT(spa_guid(spa) == pool_guid); | |
2594 | } | |
2595 | ||
2596 | /* | |
2597 | * Try to open all vdevs, loading each label in the process. | |
2598 | */ | |
2599 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
2600 | error = vdev_open(rvd); | |
2601 | spa_config_exit(spa, SCL_ALL, FTAG); | |
2602 | if (error != 0) | |
2603 | return (error); | |
2604 | ||
2605 | /* | |
2606 | * We need to validate the vdev labels against the configuration that | |
2607 | * we have in hand, which is dependent on the setting of mosconfig. If | |
2608 | * mosconfig is true then we're validating the vdev labels based on | |
2609 | * that config. Otherwise, we're validating against the cached config | |
2610 | * (zpool.cache) that was read when we loaded the zfs module, and then | |
2611 | * later we will recursively call spa_load() and validate against | |
2612 | * the vdev config. | |
2613 | * | |
2614 | * If we're assembling a new pool that's been split off from an | |
2615 | * existing pool, the labels haven't yet been updated so we skip | |
2616 | * validation for now. | |
2617 | */ | |
2618 | if (type != SPA_IMPORT_ASSEMBLE) { | |
2619 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
2620 | error = vdev_validate(rvd, mosconfig); | |
2621 | spa_config_exit(spa, SCL_ALL, FTAG); | |
2622 | ||
2623 | if (error != 0) | |
2624 | return (error); | |
2625 | ||
2626 | if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) | |
2627 | return (SET_ERROR(ENXIO)); | |
2628 | } | |
2629 | ||
2630 | /* | |
2631 | * Find the best uberblock. | |
2632 | */ | |
2633 | vdev_uberblock_load(rvd, ub, &label); | |
2634 | ||
2635 | /* | |
2636 | * If we weren't able to find a single valid uberblock, return failure. | |
2637 | */ | |
2638 | if (ub->ub_txg == 0) { | |
2639 | nvlist_free(label); | |
2640 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO)); | |
2641 | } | |
2642 | ||
2643 | /* | |
2644 | * For pools which have the multihost property on determine if the | |
2645 | * pool is truly inactive and can be safely imported. Prevent | |
2646 | * hosts which don't have a hostid set from importing the pool. | |
2647 | */ | |
2648 | activity_check = spa_activity_check_required(spa, ub, label, config); | |
2649 | if (activity_check) { | |
2650 | if (ub->ub_mmp_magic == MMP_MAGIC && ub->ub_mmp_delay && | |
2651 | spa_get_hostid() == 0) { | |
2652 | nvlist_free(label); | |
2653 | fnvlist_add_uint64(spa->spa_load_info, | |
2654 | ZPOOL_CONFIG_MMP_STATE, MMP_STATE_NO_HOSTID); | |
2655 | return (spa_vdev_err(rvd, VDEV_AUX_ACTIVE, EREMOTEIO)); | |
2656 | } | |
2657 | ||
2658 | error = spa_activity_check(spa, ub, config); | |
2659 | if (error) { | |
2660 | nvlist_free(label); | |
2661 | return (error); | |
2662 | } | |
2663 | ||
2664 | fnvlist_add_uint64(spa->spa_load_info, | |
2665 | ZPOOL_CONFIG_MMP_STATE, MMP_STATE_INACTIVE); | |
2666 | fnvlist_add_uint64(spa->spa_load_info, | |
2667 | ZPOOL_CONFIG_MMP_TXG, ub->ub_txg); | |
2668 | } | |
2669 | ||
2670 | /* | |
2671 | * If the pool has an unsupported version we can't open it. | |
2672 | */ | |
2673 | if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) { | |
2674 | nvlist_free(label); | |
2675 | return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP)); | |
2676 | } | |
2677 | ||
2678 | if (ub->ub_version >= SPA_VERSION_FEATURES) { | |
2679 | nvlist_t *features; | |
2680 | ||
2681 | /* | |
2682 | * If we weren't able to find what's necessary for reading the | |
2683 | * MOS in the label, return failure. | |
2684 | */ | |
2685 | if (label == NULL || nvlist_lookup_nvlist(label, | |
2686 | ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) { | |
2687 | nvlist_free(label); | |
2688 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, | |
2689 | ENXIO)); | |
2690 | } | |
2691 | ||
2692 | /* | |
2693 | * Update our in-core representation with the definitive values | |
2694 | * from the label. | |
2695 | */ | |
2696 | nvlist_free(spa->spa_label_features); | |
2697 | VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0); | |
2698 | } | |
2699 | ||
2700 | nvlist_free(label); | |
2701 | ||
2702 | /* | |
2703 | * Look through entries in the label nvlist's features_for_read. If | |
2704 | * there is a feature listed there which we don't understand then we | |
2705 | * cannot open a pool. | |
2706 | */ | |
2707 | if (ub->ub_version >= SPA_VERSION_FEATURES) { | |
2708 | nvlist_t *unsup_feat; | |
2709 | ||
2710 | VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) == | |
2711 | 0); | |
2712 | ||
2713 | for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features, | |
2714 | NULL); nvp != NULL; | |
2715 | nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) { | |
2716 | if (!zfeature_is_supported(nvpair_name(nvp))) { | |
2717 | VERIFY(nvlist_add_string(unsup_feat, | |
2718 | nvpair_name(nvp), "") == 0); | |
2719 | } | |
2720 | } | |
2721 | ||
2722 | if (!nvlist_empty(unsup_feat)) { | |
2723 | VERIFY(nvlist_add_nvlist(spa->spa_load_info, | |
2724 | ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0); | |
2725 | nvlist_free(unsup_feat); | |
2726 | return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, | |
2727 | ENOTSUP)); | |
2728 | } | |
2729 | ||
2730 | nvlist_free(unsup_feat); | |
2731 | } | |
2732 | ||
2733 | /* | |
2734 | * If the vdev guid sum doesn't match the uberblock, we have an | |
2735 | * incomplete configuration. We first check to see if the pool | |
2736 | * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN). | |
2737 | * If it is, defer the vdev_guid_sum check till later so we | |
2738 | * can handle missing vdevs. | |
2739 | */ | |
2740 | if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN, | |
2741 | &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE && | |
2742 | rvd->vdev_guid_sum != ub->ub_guid_sum) | |
2743 | return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO)); | |
2744 | ||
2745 | if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) { | |
2746 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
2747 | spa_try_repair(spa, config); | |
2748 | spa_config_exit(spa, SCL_ALL, FTAG); | |
2749 | nvlist_free(spa->spa_config_splitting); | |
2750 | spa->spa_config_splitting = NULL; | |
2751 | } | |
2752 | ||
2753 | /* | |
2754 | * Initialize internal SPA structures. | |
2755 | */ | |
2756 | spa->spa_state = POOL_STATE_ACTIVE; | |
2757 | spa->spa_ubsync = spa->spa_uberblock; | |
2758 | spa->spa_verify_min_txg = spa->spa_extreme_rewind ? | |
2759 | TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1; | |
2760 | spa->spa_first_txg = spa->spa_last_ubsync_txg ? | |
2761 | spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1; | |
2762 | spa->spa_claim_max_txg = spa->spa_first_txg; | |
2763 | spa->spa_prev_software_version = ub->ub_software_version; | |
2764 | ||
2765 | error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool); | |
2766 | if (error) | |
2767 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2768 | spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset; | |
2769 | ||
2770 | if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0) | |
2771 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2772 | ||
2773 | if (spa_version(spa) >= SPA_VERSION_FEATURES) { | |
2774 | boolean_t missing_feat_read = B_FALSE; | |
2775 | nvlist_t *unsup_feat, *enabled_feat; | |
2776 | ||
2777 | if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ, | |
2778 | &spa->spa_feat_for_read_obj) != 0) { | |
2779 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2780 | } | |
2781 | ||
2782 | if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE, | |
2783 | &spa->spa_feat_for_write_obj) != 0) { | |
2784 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2785 | } | |
2786 | ||
2787 | if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS, | |
2788 | &spa->spa_feat_desc_obj) != 0) { | |
2789 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2790 | } | |
2791 | ||
2792 | enabled_feat = fnvlist_alloc(); | |
2793 | unsup_feat = fnvlist_alloc(); | |
2794 | ||
2795 | if (!spa_features_check(spa, B_FALSE, | |
2796 | unsup_feat, enabled_feat)) | |
2797 | missing_feat_read = B_TRUE; | |
2798 | ||
2799 | if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) { | |
2800 | if (!spa_features_check(spa, B_TRUE, | |
2801 | unsup_feat, enabled_feat)) { | |
2802 | missing_feat_write = B_TRUE; | |
2803 | } | |
2804 | } | |
2805 | ||
2806 | fnvlist_add_nvlist(spa->spa_load_info, | |
2807 | ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat); | |
2808 | ||
2809 | if (!nvlist_empty(unsup_feat)) { | |
2810 | fnvlist_add_nvlist(spa->spa_load_info, | |
2811 | ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat); | |
2812 | } | |
2813 | ||
2814 | fnvlist_free(enabled_feat); | |
2815 | fnvlist_free(unsup_feat); | |
2816 | ||
2817 | if (!missing_feat_read) { | |
2818 | fnvlist_add_boolean(spa->spa_load_info, | |
2819 | ZPOOL_CONFIG_CAN_RDONLY); | |
2820 | } | |
2821 | ||
2822 | /* | |
2823 | * If the state is SPA_LOAD_TRYIMPORT, our objective is | |
2824 | * twofold: to determine whether the pool is available for | |
2825 | * import in read-write mode and (if it is not) whether the | |
2826 | * pool is available for import in read-only mode. If the pool | |
2827 | * is available for import in read-write mode, it is displayed | |
2828 | * as available in userland; if it is not available for import | |
2829 | * in read-only mode, it is displayed as unavailable in | |
2830 | * userland. If the pool is available for import in read-only | |
2831 | * mode but not read-write mode, it is displayed as unavailable | |
2832 | * in userland with a special note that the pool is actually | |
2833 | * available for open in read-only mode. | |
2834 | * | |
2835 | * As a result, if the state is SPA_LOAD_TRYIMPORT and we are | |
2836 | * missing a feature for write, we must first determine whether | |
2837 | * the pool can be opened read-only before returning to | |
2838 | * userland in order to know whether to display the | |
2839 | * abovementioned note. | |
2840 | */ | |
2841 | if (missing_feat_read || (missing_feat_write && | |
2842 | spa_writeable(spa))) { | |
2843 | return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, | |
2844 | ENOTSUP)); | |
2845 | } | |
2846 | ||
2847 | /* | |
2848 | * Load refcounts for ZFS features from disk into an in-memory | |
2849 | * cache during SPA initialization. | |
2850 | */ | |
2851 | for (spa_feature_t i = 0; i < SPA_FEATURES; i++) { | |
2852 | uint64_t refcount; | |
2853 | ||
2854 | error = feature_get_refcount_from_disk(spa, | |
2855 | &spa_feature_table[i], &refcount); | |
2856 | if (error == 0) { | |
2857 | spa->spa_feat_refcount_cache[i] = refcount; | |
2858 | } else if (error == ENOTSUP) { | |
2859 | spa->spa_feat_refcount_cache[i] = | |
2860 | SPA_FEATURE_DISABLED; | |
2861 | } else { | |
2862 | return (spa_vdev_err(rvd, | |
2863 | VDEV_AUX_CORRUPT_DATA, EIO)); | |
2864 | } | |
2865 | } | |
2866 | } | |
2867 | ||
2868 | if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) { | |
2869 | if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG, | |
2870 | &spa->spa_feat_enabled_txg_obj) != 0) | |
2871 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2872 | } | |
2873 | ||
2874 | spa->spa_is_initializing = B_TRUE; | |
2875 | error = dsl_pool_open(spa->spa_dsl_pool); | |
2876 | spa->spa_is_initializing = B_FALSE; | |
2877 | if (error != 0) | |
2878 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2879 | ||
2880 | if (!mosconfig) { | |
2881 | uint64_t hostid; | |
2882 | nvlist_t *policy = NULL, *nvconfig; | |
2883 | ||
2884 | if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) | |
2885 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2886 | ||
2887 | if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig, | |
2888 | ZPOOL_CONFIG_HOSTID, &hostid) == 0) { | |
2889 | char *hostname; | |
2890 | unsigned long myhostid = 0; | |
2891 | ||
2892 | VERIFY(nvlist_lookup_string(nvconfig, | |
2893 | ZPOOL_CONFIG_HOSTNAME, &hostname) == 0); | |
2894 | ||
2895 | myhostid = spa_get_hostid(); | |
2896 | if (hostid && myhostid && hostid != myhostid) { | |
2897 | nvlist_free(nvconfig); | |
2898 | return (SET_ERROR(EBADF)); | |
2899 | } | |
2900 | } | |
2901 | if (nvlist_lookup_nvlist(spa->spa_config, | |
2902 | ZPOOL_REWIND_POLICY, &policy) == 0) | |
2903 | VERIFY(nvlist_add_nvlist(nvconfig, | |
2904 | ZPOOL_REWIND_POLICY, policy) == 0); | |
2905 | ||
2906 | spa_config_set(spa, nvconfig); | |
2907 | spa_unload(spa); | |
2908 | spa_deactivate(spa); | |
2909 | spa_activate(spa, orig_mode); | |
2910 | ||
2911 | return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE)); | |
2912 | } | |
2913 | ||
2914 | /* Grab the checksum salt from the MOS. */ | |
2915 | error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, | |
2916 | DMU_POOL_CHECKSUM_SALT, 1, | |
2917 | sizeof (spa->spa_cksum_salt.zcs_bytes), | |
2918 | spa->spa_cksum_salt.zcs_bytes); | |
2919 | if (error == ENOENT) { | |
2920 | /* Generate a new salt for subsequent use */ | |
2921 | (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes, | |
2922 | sizeof (spa->spa_cksum_salt.zcs_bytes)); | |
2923 | } else if (error != 0) { | |
2924 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2925 | } | |
2926 | ||
2927 | if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0) | |
2928 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2929 | error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj); | |
2930 | if (error != 0) | |
2931 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2932 | ||
2933 | /* | |
2934 | * Load the bit that tells us to use the new accounting function | |
2935 | * (raid-z deflation). If we have an older pool, this will not | |
2936 | * be present. | |
2937 | */ | |
2938 | error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate); | |
2939 | if (error != 0 && error != ENOENT) | |
2940 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2941 | ||
2942 | error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION, | |
2943 | &spa->spa_creation_version); | |
2944 | if (error != 0 && error != ENOENT) | |
2945 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2946 | ||
2947 | /* | |
2948 | * Load the persistent error log. If we have an older pool, this will | |
2949 | * not be present. | |
2950 | */ | |
2951 | error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last); | |
2952 | if (error != 0 && error != ENOENT) | |
2953 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2954 | ||
2955 | error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB, | |
2956 | &spa->spa_errlog_scrub); | |
2957 | if (error != 0 && error != ENOENT) | |
2958 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2959 | ||
2960 | /* | |
2961 | * Load the history object. If we have an older pool, this | |
2962 | * will not be present. | |
2963 | */ | |
2964 | error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history); | |
2965 | if (error != 0 && error != ENOENT) | |
2966 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2967 | ||
2968 | /* | |
2969 | * Load the per-vdev ZAP map. If we have an older pool, this will not | |
2970 | * be present; in this case, defer its creation to a later time to | |
2971 | * avoid dirtying the MOS this early / out of sync context. See | |
2972 | * spa_sync_config_object. | |
2973 | */ | |
2974 | ||
2975 | /* The sentinel is only available in the MOS config. */ | |
2976 | nvlist_t *mos_config; | |
2977 | if (load_nvlist(spa, spa->spa_config_object, &mos_config) != 0) | |
2978 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2979 | ||
2980 | error = spa_dir_prop(spa, DMU_POOL_VDEV_ZAP_MAP, | |
2981 | &spa->spa_all_vdev_zaps); | |
2982 | ||
2983 | if (error == ENOENT) { | |
2984 | VERIFY(!nvlist_exists(mos_config, | |
2985 | ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)); | |
2986 | spa->spa_avz_action = AVZ_ACTION_INITIALIZE; | |
2987 | ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev)); | |
2988 | } else if (error != 0) { | |
2989 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
2990 | } else if (!nvlist_exists(mos_config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)) { | |
2991 | /* | |
2992 | * An older version of ZFS overwrote the sentinel value, so | |
2993 | * we have orphaned per-vdev ZAPs in the MOS. Defer their | |
2994 | * destruction to later; see spa_sync_config_object. | |
2995 | */ | |
2996 | spa->spa_avz_action = AVZ_ACTION_DESTROY; | |
2997 | /* | |
2998 | * We're assuming that no vdevs have had their ZAPs created | |
2999 | * before this. Better be sure of it. | |
3000 | */ | |
3001 | ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev)); | |
3002 | } | |
3003 | nvlist_free(mos_config); | |
3004 | ||
3005 | /* | |
3006 | * If we're assembling the pool from the split-off vdevs of | |
3007 | * an existing pool, we don't want to attach the spares & cache | |
3008 | * devices. | |
3009 | */ | |
3010 | ||
3011 | /* | |
3012 | * Load any hot spares for this pool. | |
3013 | */ | |
3014 | error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object); | |
3015 | if (error != 0 && error != ENOENT) | |
3016 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
3017 | if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { | |
3018 | ASSERT(spa_version(spa) >= SPA_VERSION_SPARES); | |
3019 | if (load_nvlist(spa, spa->spa_spares.sav_object, | |
3020 | &spa->spa_spares.sav_config) != 0) | |
3021 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
3022 | ||
3023 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
3024 | spa_load_spares(spa); | |
3025 | spa_config_exit(spa, SCL_ALL, FTAG); | |
3026 | } else if (error == 0) { | |
3027 | spa->spa_spares.sav_sync = B_TRUE; | |
3028 | } | |
3029 | ||
3030 | /* | |
3031 | * Load any level 2 ARC devices for this pool. | |
3032 | */ | |
3033 | error = spa_dir_prop(spa, DMU_POOL_L2CACHE, | |
3034 | &spa->spa_l2cache.sav_object); | |
3035 | if (error != 0 && error != ENOENT) | |
3036 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
3037 | if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { | |
3038 | ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE); | |
3039 | if (load_nvlist(spa, spa->spa_l2cache.sav_object, | |
3040 | &spa->spa_l2cache.sav_config) != 0) | |
3041 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
3042 | ||
3043 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
3044 | spa_load_l2cache(spa); | |
3045 | spa_config_exit(spa, SCL_ALL, FTAG); | |
3046 | } else if (error == 0) { | |
3047 | spa->spa_l2cache.sav_sync = B_TRUE; | |
3048 | } | |
3049 | ||
3050 | spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); | |
3051 | ||
3052 | error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object); | |
3053 | if (error && error != ENOENT) | |
3054 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
3055 | ||
3056 | if (error == 0) { | |
3057 | uint64_t autoreplace = 0; | |
3058 | ||
3059 | spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs); | |
3060 | spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace); | |
3061 | spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation); | |
3062 | spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode); | |
3063 | spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand); | |
3064 | spa_prop_find(spa, ZPOOL_PROP_MULTIHOST, &spa->spa_multihost); | |
3065 | spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO, | |
3066 | &spa->spa_dedup_ditto); | |
3067 | ||
3068 | spa->spa_autoreplace = (autoreplace != 0); | |
3069 | } | |
3070 | ||
3071 | /* | |
3072 | * If the 'multihost' property is set, then never allow a pool to | |
3073 | * be imported when the system hostid is zero. The exception to | |
3074 | * this rule is zdb which is always allowed to access pools. | |
3075 | */ | |
3076 | if (spa_multihost(spa) && spa_get_hostid() == 0 && | |
3077 | (spa->spa_import_flags & ZFS_IMPORT_SKIP_MMP) == 0) { | |
3078 | fnvlist_add_uint64(spa->spa_load_info, | |
3079 | ZPOOL_CONFIG_MMP_STATE, MMP_STATE_NO_HOSTID); | |
3080 | return (spa_vdev_err(rvd, VDEV_AUX_ACTIVE, EREMOTEIO)); | |
3081 | } | |
3082 | ||
3083 | /* | |
3084 | * If the 'autoreplace' property is set, then post a resource notifying | |
3085 | * the ZFS DE that it should not issue any faults for unopenable | |
3086 | * devices. We also iterate over the vdevs, and post a sysevent for any | |
3087 | * unopenable vdevs so that the normal autoreplace handler can take | |
3088 | * over. | |
3089 | */ | |
3090 | if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) { | |
3091 | spa_check_removed(spa->spa_root_vdev); | |
3092 | /* | |
3093 | * For the import case, this is done in spa_import(), because | |
3094 | * at this point we're using the spare definitions from | |
3095 | * the MOS config, not necessarily from the userland config. | |
3096 | */ | |
3097 | if (state != SPA_LOAD_IMPORT) { | |
3098 | spa_aux_check_removed(&spa->spa_spares); | |
3099 | spa_aux_check_removed(&spa->spa_l2cache); | |
3100 | } | |
3101 | } | |
3102 | ||
3103 | /* | |
3104 | * Load the vdev state for all toplevel vdevs. | |
3105 | */ | |
3106 | vdev_load(rvd); | |
3107 | ||
3108 | /* | |
3109 | * Propagate the leaf DTLs we just loaded all the way up the tree. | |
3110 | */ | |
3111 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
3112 | vdev_dtl_reassess(rvd, 0, 0, B_FALSE); | |
3113 | spa_config_exit(spa, SCL_ALL, FTAG); | |
3114 | ||
3115 | /* | |
3116 | * Load the DDTs (dedup tables). | |
3117 | */ | |
3118 | error = ddt_load(spa); | |
3119 | if (error != 0) | |
3120 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
3121 | ||
3122 | spa_update_dspace(spa); | |
3123 | ||
3124 | /* | |
3125 | * Validate the config, using the MOS config to fill in any | |
3126 | * information which might be missing. If we fail to validate | |
3127 | * the config then declare the pool unfit for use. If we're | |
3128 | * assembling a pool from a split, the log is not transferred | |
3129 | * over. | |
3130 | */ | |
3131 | if (type != SPA_IMPORT_ASSEMBLE) { | |
3132 | nvlist_t *nvconfig; | |
3133 | ||
3134 | if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) | |
3135 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); | |
3136 | ||
3137 | if (!spa_config_valid(spa, nvconfig)) { | |
3138 | nvlist_free(nvconfig); | |
3139 | return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, | |
3140 | ENXIO)); | |
3141 | } | |
3142 | nvlist_free(nvconfig); | |
3143 | ||
3144 | /* | |
3145 | * Now that we've validated the config, check the state of the | |
3146 | * root vdev. If it can't be opened, it indicates one or | |
3147 | * more toplevel vdevs are faulted. | |
3148 | */ | |
3149 | if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) | |
3150 | return (SET_ERROR(ENXIO)); | |
3151 | ||
3152 | if (spa_writeable(spa) && spa_check_logs(spa)) { | |
3153 | *ereport = FM_EREPORT_ZFS_LOG_REPLAY; | |
3154 | return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO)); | |
3155 | } | |
3156 | } | |
3157 | ||
3158 | if (missing_feat_write) { | |
3159 | ASSERT(state == SPA_LOAD_TRYIMPORT); | |
3160 | ||
3161 | /* | |
3162 | * At this point, we know that we can open the pool in | |
3163 | * read-only mode but not read-write mode. We now have enough | |
3164 | * information and can return to userland. | |
3165 | */ | |
3166 | return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP)); | |
3167 | } | |
3168 | ||
3169 | /* | |
3170 | * We've successfully opened the pool, verify that we're ready | |
3171 | * to start pushing transactions. | |
3172 | */ | |
3173 | if (state != SPA_LOAD_TRYIMPORT) { | |
3174 | if ((error = spa_load_verify(spa))) | |
3175 | return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, | |
3176 | error)); | |
3177 | } | |
3178 | ||
3179 | if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER || | |
3180 | spa->spa_load_max_txg == UINT64_MAX)) { | |
3181 | dmu_tx_t *tx; | |
3182 | int need_update = B_FALSE; | |
3183 | dsl_pool_t *dp = spa_get_dsl(spa); | |
3184 | ||
3185 | ASSERT(state != SPA_LOAD_TRYIMPORT); | |
3186 | ||
3187 | /* | |
3188 | * Claim log blocks that haven't been committed yet. | |
3189 | * This must all happen in a single txg. | |
3190 | * Note: spa_claim_max_txg is updated by spa_claim_notify(), | |
3191 | * invoked from zil_claim_log_block()'s i/o done callback. | |
3192 | * Price of rollback is that we abandon the log. | |
3193 | */ | |
3194 | spa->spa_claiming = B_TRUE; | |
3195 | ||
3196 | tx = dmu_tx_create_assigned(dp, spa_first_txg(spa)); | |
3197 | (void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj, | |
3198 | zil_claim, tx, DS_FIND_CHILDREN); | |
3199 | dmu_tx_commit(tx); | |
3200 | ||
3201 | spa->spa_claiming = B_FALSE; | |
3202 | ||
3203 | spa_set_log_state(spa, SPA_LOG_GOOD); | |
3204 | spa->spa_sync_on = B_TRUE; | |
3205 | txg_sync_start(spa->spa_dsl_pool); | |
3206 | mmp_thread_start(spa); | |
3207 | ||
3208 | /* | |
3209 | * Wait for all claims to sync. We sync up to the highest | |
3210 | * claimed log block birth time so that claimed log blocks | |
3211 | * don't appear to be from the future. spa_claim_max_txg | |
3212 | * will have been set for us by either zil_check_log_chain() | |
3213 | * (invoked from spa_check_logs()) or zil_claim() above. | |
3214 | */ | |
3215 | txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg); | |
3216 | ||
3217 | /* | |
3218 | * If the config cache is stale, or we have uninitialized | |
3219 | * metaslabs (see spa_vdev_add()), then update the config. | |
3220 | * | |
3221 | * If this is a verbatim import, trust the current | |
3222 | * in-core spa_config and update the disk labels. | |
3223 | */ | |
3224 | if (config_cache_txg != spa->spa_config_txg || | |
3225 | state == SPA_LOAD_IMPORT || | |
3226 | state == SPA_LOAD_RECOVER || | |
3227 | (spa->spa_import_flags & ZFS_IMPORT_VERBATIM)) | |
3228 | need_update = B_TRUE; | |
3229 | ||
3230 | for (int c = 0; c < rvd->vdev_children; c++) | |
3231 | if (rvd->vdev_child[c]->vdev_ms_array == 0) | |
3232 | need_update = B_TRUE; | |
3233 | ||
3234 | /* | |
3235 | * Update the config cache asychronously in case we're the | |
3236 | * root pool, in which case the config cache isn't writable yet. | |
3237 | */ | |
3238 | if (need_update) | |
3239 | spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); | |
3240 | ||
3241 | /* | |
3242 | * Check all DTLs to see if anything needs resilvering. | |
3243 | */ | |
3244 | if (!dsl_scan_resilvering(spa->spa_dsl_pool) && | |
3245 | vdev_resilver_needed(rvd, NULL, NULL)) | |
3246 | spa_async_request(spa, SPA_ASYNC_RESILVER); | |
3247 | ||
3248 | /* | |
3249 | * Log the fact that we booted up (so that we can detect if | |
3250 | * we rebooted in the middle of an operation). | |
3251 | */ | |
3252 | spa_history_log_version(spa, "open", NULL); | |
3253 | ||
3254 | /* | |
3255 | * Delete any inconsistent datasets. | |
3256 | */ | |
3257 | (void) dmu_objset_find(spa_name(spa), | |
3258 | dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN); | |
3259 | ||
3260 | /* | |
3261 | * Clean up any stale temporary dataset userrefs. | |
3262 | */ | |
3263 | dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool); | |
3264 | } | |
3265 | ||
3266 | return (0); | |
3267 | } | |
3268 | ||
3269 | static int | |
3270 | spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig) | |
3271 | { | |
3272 | int mode = spa->spa_mode; | |
3273 | ||
3274 | spa_unload(spa); | |
3275 | spa_deactivate(spa); | |
3276 | ||
3277 | spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1; | |
3278 | ||
3279 | spa_activate(spa, mode); | |
3280 | spa_async_suspend(spa); | |
3281 | ||
3282 | return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig)); | |
3283 | } | |
3284 | ||
3285 | /* | |
3286 | * If spa_load() fails this function will try loading prior txg's. If | |
3287 | * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool | |
3288 | * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this | |
3289 | * function will not rewind the pool and will return the same error as | |
3290 | * spa_load(). | |
3291 | */ | |
3292 | static int | |
3293 | spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig, | |
3294 | uint64_t max_request, int rewind_flags) | |
3295 | { | |
3296 | nvlist_t *loadinfo = NULL; | |
3297 | nvlist_t *config = NULL; | |
3298 | int load_error, rewind_error; | |
3299 | uint64_t safe_rewind_txg; | |
3300 | uint64_t min_txg; | |
3301 | ||
3302 | if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) { | |
3303 | spa->spa_load_max_txg = spa->spa_load_txg; | |
3304 | spa_set_log_state(spa, SPA_LOG_CLEAR); | |
3305 | } else { | |
3306 | spa->spa_load_max_txg = max_request; | |
3307 | if (max_request != UINT64_MAX) | |
3308 | spa->spa_extreme_rewind = B_TRUE; | |
3309 | } | |
3310 | ||
3311 | load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING, | |
3312 | mosconfig); | |
3313 | if (load_error == 0) | |
3314 | return (0); | |
3315 | ||
3316 | if (spa->spa_root_vdev != NULL) | |
3317 | config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); | |
3318 | ||
3319 | spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg; | |
3320 | spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp; | |
3321 | ||
3322 | if (rewind_flags & ZPOOL_NEVER_REWIND) { | |
3323 | nvlist_free(config); | |
3324 | return (load_error); | |
3325 | } | |
3326 | ||
3327 | if (state == SPA_LOAD_RECOVER) { | |
3328 | /* Price of rolling back is discarding txgs, including log */ | |
3329 | spa_set_log_state(spa, SPA_LOG_CLEAR); | |
3330 | } else { | |
3331 | /* | |
3332 | * If we aren't rolling back save the load info from our first | |
3333 | * import attempt so that we can restore it after attempting | |
3334 | * to rewind. | |
3335 | */ | |
3336 | loadinfo = spa->spa_load_info; | |
3337 | spa->spa_load_info = fnvlist_alloc(); | |
3338 | } | |
3339 | ||
3340 | spa->spa_load_max_txg = spa->spa_last_ubsync_txg; | |
3341 | safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE; | |
3342 | min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ? | |
3343 | TXG_INITIAL : safe_rewind_txg; | |
3344 | ||
3345 | /* | |
3346 | * Continue as long as we're finding errors, we're still within | |
3347 | * the acceptable rewind range, and we're still finding uberblocks | |
3348 | */ | |
3349 | while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg && | |
3350 | spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) { | |
3351 | if (spa->spa_load_max_txg < safe_rewind_txg) | |
3352 | spa->spa_extreme_rewind = B_TRUE; | |
3353 | rewind_error = spa_load_retry(spa, state, mosconfig); | |
3354 | } | |
3355 | ||
3356 | spa->spa_extreme_rewind = B_FALSE; | |
3357 | spa->spa_load_max_txg = UINT64_MAX; | |
3358 | ||
3359 | if (config && (rewind_error || state != SPA_LOAD_RECOVER)) | |
3360 | spa_config_set(spa, config); | |
3361 | else | |
3362 | nvlist_free(config); | |
3363 | ||
3364 | if (state == SPA_LOAD_RECOVER) { | |
3365 | ASSERT3P(loadinfo, ==, NULL); | |
3366 | return (rewind_error); | |
3367 | } else { | |
3368 | /* Store the rewind info as part of the initial load info */ | |
3369 | fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO, | |
3370 | spa->spa_load_info); | |
3371 | ||
3372 | /* Restore the initial load info */ | |
3373 | fnvlist_free(spa->spa_load_info); | |
3374 | spa->spa_load_info = loadinfo; | |
3375 | ||
3376 | return (load_error); | |
3377 | } | |
3378 | } | |
3379 | ||
3380 | /* | |
3381 | * Pool Open/Import | |
3382 | * | |
3383 | * The import case is identical to an open except that the configuration is sent | |
3384 | * down from userland, instead of grabbed from the configuration cache. For the | |
3385 | * case of an open, the pool configuration will exist in the | |
3386 | * POOL_STATE_UNINITIALIZED state. | |
3387 | * | |
3388 | * The stats information (gen/count/ustats) is used to gather vdev statistics at | |
3389 | * the same time open the pool, without having to keep around the spa_t in some | |
3390 | * ambiguous state. | |
3391 | */ | |
3392 | static int | |
3393 | spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy, | |
3394 | nvlist_t **config) | |
3395 | { | |
3396 | spa_t *spa; | |
3397 | spa_load_state_t state = SPA_LOAD_OPEN; | |
3398 | int error; | |
3399 | int locked = B_FALSE; | |
3400 | int firstopen = B_FALSE; | |
3401 | ||
3402 | *spapp = NULL; | |
3403 | ||
3404 | /* | |
3405 | * As disgusting as this is, we need to support recursive calls to this | |
3406 | * function because dsl_dir_open() is called during spa_load(), and ends | |
3407 | * up calling spa_open() again. The real fix is to figure out how to | |
3408 | * avoid dsl_dir_open() calling this in the first place. | |
3409 | */ | |
3410 | if (MUTEX_NOT_HELD(&spa_namespace_lock)) { | |
3411 | mutex_enter(&spa_namespace_lock); | |
3412 | locked = B_TRUE; | |
3413 | } | |
3414 | ||
3415 | if ((spa = spa_lookup(pool)) == NULL) { | |
3416 | if (locked) | |
3417 | mutex_exit(&spa_namespace_lock); | |
3418 | return (SET_ERROR(ENOENT)); | |
3419 | } | |
3420 | ||
3421 | if (spa->spa_state == POOL_STATE_UNINITIALIZED) { | |
3422 | zpool_rewind_policy_t policy; | |
3423 | ||
3424 | firstopen = B_TRUE; | |
3425 | ||
3426 | zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config, | |
3427 | &policy); | |
3428 | if (policy.zrp_request & ZPOOL_DO_REWIND) | |
3429 | state = SPA_LOAD_RECOVER; | |
3430 | ||
3431 | spa_activate(spa, spa_mode_global); | |
3432 | ||
3433 | if (state != SPA_LOAD_RECOVER) | |
3434 | spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; | |
3435 | ||
3436 | error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg, | |
3437 | policy.zrp_request); | |
3438 | ||
3439 | if (error == EBADF) { | |
3440 | /* | |
3441 | * If vdev_validate() returns failure (indicated by | |
3442 | * EBADF), it indicates that one of the vdevs indicates | |
3443 | * that the pool has been exported or destroyed. If | |
3444 | * this is the case, the config cache is out of sync and | |
3445 | * we should remove the pool from the namespace. | |
3446 | */ | |
3447 | spa_unload(spa); | |
3448 | spa_deactivate(spa); | |
3449 | spa_config_sync(spa, B_TRUE, B_TRUE); | |
3450 | spa_remove(spa); | |
3451 | if (locked) | |
3452 | mutex_exit(&spa_namespace_lock); | |
3453 | return (SET_ERROR(ENOENT)); | |
3454 | } | |
3455 | ||
3456 | if (error) { | |
3457 | /* | |
3458 | * We can't open the pool, but we still have useful | |
3459 | * information: the state of each vdev after the | |
3460 | * attempted vdev_open(). Return this to the user. | |
3461 | */ | |
3462 | if (config != NULL && spa->spa_config) { | |
3463 | VERIFY(nvlist_dup(spa->spa_config, config, | |
3464 | KM_SLEEP) == 0); | |
3465 | VERIFY(nvlist_add_nvlist(*config, | |
3466 | ZPOOL_CONFIG_LOAD_INFO, | |
3467 | spa->spa_load_info) == 0); | |
3468 | } | |
3469 | spa_unload(spa); | |
3470 | spa_deactivate(spa); | |
3471 | spa->spa_last_open_failed = error; | |
3472 | if (locked) | |
3473 | mutex_exit(&spa_namespace_lock); | |
3474 | *spapp = NULL; | |
3475 | return (error); | |
3476 | } | |
3477 | } | |
3478 | ||
3479 | spa_open_ref(spa, tag); | |
3480 | ||
3481 | if (config != NULL) | |
3482 | *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); | |
3483 | ||
3484 | /* | |
3485 | * If we've recovered the pool, pass back any information we | |
3486 | * gathered while doing the load. | |
3487 | */ | |
3488 | if (state == SPA_LOAD_RECOVER) { | |
3489 | VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO, | |
3490 | spa->spa_load_info) == 0); | |
3491 | } | |
3492 | ||
3493 | if (locked) { | |
3494 | spa->spa_last_open_failed = 0; | |
3495 | spa->spa_last_ubsync_txg = 0; | |
3496 | spa->spa_load_txg = 0; | |
3497 | mutex_exit(&spa_namespace_lock); | |
3498 | } | |
3499 | ||
3500 | if (firstopen) | |
3501 | zvol_create_minors(spa, spa_name(spa), B_TRUE); | |
3502 | ||
3503 | *spapp = spa; | |
3504 | ||
3505 | return (0); | |
3506 | } | |
3507 | ||
3508 | int | |
3509 | spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy, | |
3510 | nvlist_t **config) | |
3511 | { | |
3512 | return (spa_open_common(name, spapp, tag, policy, config)); | |
3513 | } | |
3514 | ||
3515 | int | |
3516 | spa_open(const char *name, spa_t **spapp, void *tag) | |
3517 | { | |
3518 | return (spa_open_common(name, spapp, tag, NULL, NULL)); | |
3519 | } | |
3520 | ||
3521 | /* | |
3522 | * Lookup the given spa_t, incrementing the inject count in the process, | |
3523 | * preventing it from being exported or destroyed. | |
3524 | */ | |
3525 | spa_t * | |
3526 | spa_inject_addref(char *name) | |
3527 | { | |
3528 | spa_t *spa; | |
3529 | ||
3530 | mutex_enter(&spa_namespace_lock); | |
3531 | if ((spa = spa_lookup(name)) == NULL) { | |
3532 | mutex_exit(&spa_namespace_lock); | |
3533 | return (NULL); | |
3534 | } | |
3535 | spa->spa_inject_ref++; | |
3536 | mutex_exit(&spa_namespace_lock); | |
3537 | ||
3538 | return (spa); | |
3539 | } | |
3540 | ||
3541 | void | |
3542 | spa_inject_delref(spa_t *spa) | |
3543 | { | |
3544 | mutex_enter(&spa_namespace_lock); | |
3545 | spa->spa_inject_ref--; | |
3546 | mutex_exit(&spa_namespace_lock); | |
3547 | } | |
3548 | ||
3549 | /* | |
3550 | * Add spares device information to the nvlist. | |
3551 | */ | |
3552 | static void | |
3553 | spa_add_spares(spa_t *spa, nvlist_t *config) | |
3554 | { | |
3555 | nvlist_t **spares; | |
3556 | uint_t i, nspares; | |
3557 | nvlist_t *nvroot; | |
3558 | uint64_t guid; | |
3559 | vdev_stat_t *vs; | |
3560 | uint_t vsc; | |
3561 | uint64_t pool; | |
3562 | ||
3563 | ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); | |
3564 | ||
3565 | if (spa->spa_spares.sav_count == 0) | |
3566 | return; | |
3567 | ||
3568 | VERIFY(nvlist_lookup_nvlist(config, | |
3569 | ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); | |
3570 | VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, | |
3571 | ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); | |
3572 | if (nspares != 0) { | |
3573 | VERIFY(nvlist_add_nvlist_array(nvroot, | |
3574 | ZPOOL_CONFIG_SPARES, spares, nspares) == 0); | |
3575 | VERIFY(nvlist_lookup_nvlist_array(nvroot, | |
3576 | ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); | |
3577 | ||
3578 | /* | |
3579 | * Go through and find any spares which have since been | |
3580 | * repurposed as an active spare. If this is the case, update | |
3581 | * their status appropriately. | |
3582 | */ | |
3583 | for (i = 0; i < nspares; i++) { | |
3584 | VERIFY(nvlist_lookup_uint64(spares[i], | |
3585 | ZPOOL_CONFIG_GUID, &guid) == 0); | |
3586 | if (spa_spare_exists(guid, &pool, NULL) && | |
3587 | pool != 0ULL) { | |
3588 | VERIFY(nvlist_lookup_uint64_array( | |
3589 | spares[i], ZPOOL_CONFIG_VDEV_STATS, | |
3590 | (uint64_t **)&vs, &vsc) == 0); | |
3591 | vs->vs_state = VDEV_STATE_CANT_OPEN; | |
3592 | vs->vs_aux = VDEV_AUX_SPARED; | |
3593 | } | |
3594 | } | |
3595 | } | |
3596 | } | |
3597 | ||
3598 | /* | |
3599 | * Add l2cache device information to the nvlist, including vdev stats. | |
3600 | */ | |
3601 | static void | |
3602 | spa_add_l2cache(spa_t *spa, nvlist_t *config) | |
3603 | { | |
3604 | nvlist_t **l2cache; | |
3605 | uint_t i, j, nl2cache; | |
3606 | nvlist_t *nvroot; | |
3607 | uint64_t guid; | |
3608 | vdev_t *vd; | |
3609 | vdev_stat_t *vs; | |
3610 | uint_t vsc; | |
3611 | ||
3612 | ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); | |
3613 | ||
3614 | if (spa->spa_l2cache.sav_count == 0) | |
3615 | return; | |
3616 | ||
3617 | VERIFY(nvlist_lookup_nvlist(config, | |
3618 | ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); | |
3619 | VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, | |
3620 | ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); | |
3621 | if (nl2cache != 0) { | |
3622 | VERIFY(nvlist_add_nvlist_array(nvroot, | |
3623 | ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); | |
3624 | VERIFY(nvlist_lookup_nvlist_array(nvroot, | |
3625 | ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); | |
3626 | ||
3627 | /* | |
3628 | * Update level 2 cache device stats. | |
3629 | */ | |
3630 | ||
3631 | for (i = 0; i < nl2cache; i++) { | |
3632 | VERIFY(nvlist_lookup_uint64(l2cache[i], | |
3633 | ZPOOL_CONFIG_GUID, &guid) == 0); | |
3634 | ||
3635 | vd = NULL; | |
3636 | for (j = 0; j < spa->spa_l2cache.sav_count; j++) { | |
3637 | if (guid == | |
3638 | spa->spa_l2cache.sav_vdevs[j]->vdev_guid) { | |
3639 | vd = spa->spa_l2cache.sav_vdevs[j]; | |
3640 | break; | |
3641 | } | |
3642 | } | |
3643 | ASSERT(vd != NULL); | |
3644 | ||
3645 | VERIFY(nvlist_lookup_uint64_array(l2cache[i], | |
3646 | ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc) | |
3647 | == 0); | |
3648 | vdev_get_stats(vd, vs); | |
3649 | vdev_config_generate_stats(vd, l2cache[i]); | |
3650 | ||
3651 | } | |
3652 | } | |
3653 | } | |
3654 | ||
3655 | static void | |
3656 | spa_feature_stats_from_disk(spa_t *spa, nvlist_t *features) | |
3657 | { | |
3658 | zap_cursor_t zc; | |
3659 | zap_attribute_t za; | |
3660 | ||
3661 | if (spa->spa_feat_for_read_obj != 0) { | |
3662 | for (zap_cursor_init(&zc, spa->spa_meta_objset, | |
3663 | spa->spa_feat_for_read_obj); | |
3664 | zap_cursor_retrieve(&zc, &za) == 0; | |
3665 | zap_cursor_advance(&zc)) { | |
3666 | ASSERT(za.za_integer_length == sizeof (uint64_t) && | |
3667 | za.za_num_integers == 1); | |
3668 | VERIFY0(nvlist_add_uint64(features, za.za_name, | |
3669 | za.za_first_integer)); | |
3670 | } | |
3671 | zap_cursor_fini(&zc); | |
3672 | } | |
3673 | ||
3674 | if (spa->spa_feat_for_write_obj != 0) { | |
3675 | for (zap_cursor_init(&zc, spa->spa_meta_objset, | |
3676 | spa->spa_feat_for_write_obj); | |
3677 | zap_cursor_retrieve(&zc, &za) == 0; | |
3678 | zap_cursor_advance(&zc)) { | |
3679 | ASSERT(za.za_integer_length == sizeof (uint64_t) && | |
3680 | za.za_num_integers == 1); | |
3681 | VERIFY0(nvlist_add_uint64(features, za.za_name, | |
3682 | za.za_first_integer)); | |
3683 | } | |
3684 | zap_cursor_fini(&zc); | |
3685 | } | |
3686 | } | |
3687 | ||
3688 | static void | |
3689 | spa_feature_stats_from_cache(spa_t *spa, nvlist_t *features) | |
3690 | { | |
3691 | int i; | |
3692 | ||
3693 | for (i = 0; i < SPA_FEATURES; i++) { | |
3694 | zfeature_info_t feature = spa_feature_table[i]; | |
3695 | uint64_t refcount; | |
3696 | ||
3697 | if (feature_get_refcount(spa, &feature, &refcount) != 0) | |
3698 | continue; | |
3699 | ||
3700 | VERIFY0(nvlist_add_uint64(features, feature.fi_guid, refcount)); | |
3701 | } | |
3702 | } | |
3703 | ||
3704 | /* | |
3705 | * Store a list of pool features and their reference counts in the | |
3706 | * config. | |
3707 | * | |
3708 | * The first time this is called on a spa, allocate a new nvlist, fetch | |
3709 | * the pool features and reference counts from disk, then save the list | |
3710 | * in the spa. In subsequent calls on the same spa use the saved nvlist | |
3711 | * and refresh its values from the cached reference counts. This | |
3712 | * ensures we don't block here on I/O on a suspended pool so 'zpool | |
3713 | * clear' can resume the pool. | |
3714 | */ | |
3715 | static void | |
3716 | spa_add_feature_stats(spa_t *spa, nvlist_t *config) | |
3717 | { | |
3718 | nvlist_t *features; | |
3719 | ||
3720 | ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); | |
3721 | ||
3722 | mutex_enter(&spa->spa_feat_stats_lock); | |
3723 | features = spa->spa_feat_stats; | |
3724 | ||
3725 | if (features != NULL) { | |
3726 | spa_feature_stats_from_cache(spa, features); | |
3727 | } else { | |
3728 | VERIFY0(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP)); | |
3729 | spa->spa_feat_stats = features; | |
3730 | spa_feature_stats_from_disk(spa, features); | |
3731 | } | |
3732 | ||
3733 | VERIFY0(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS, | |
3734 | features)); | |
3735 | ||
3736 | mutex_exit(&spa->spa_feat_stats_lock); | |
3737 | } | |
3738 | ||
3739 | int | |
3740 | spa_get_stats(const char *name, nvlist_t **config, | |
3741 | char *altroot, size_t buflen) | |
3742 | { | |
3743 | int error; | |
3744 | spa_t *spa; | |
3745 | ||
3746 | *config = NULL; | |
3747 | error = spa_open_common(name, &spa, FTAG, NULL, config); | |
3748 | ||
3749 | if (spa != NULL) { | |
3750 | /* | |
3751 | * This still leaves a window of inconsistency where the spares | |
3752 | * or l2cache devices could change and the config would be | |
3753 | * self-inconsistent. | |
3754 | */ | |
3755 | spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); | |
3756 | ||
3757 | if (*config != NULL) { | |
3758 | uint64_t loadtimes[2]; | |
3759 | ||
3760 | loadtimes[0] = spa->spa_loaded_ts.tv_sec; | |
3761 | loadtimes[1] = spa->spa_loaded_ts.tv_nsec; | |
3762 | VERIFY(nvlist_add_uint64_array(*config, | |
3763 | ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0); | |
3764 | ||
3765 | VERIFY(nvlist_add_uint64(*config, | |
3766 | ZPOOL_CONFIG_ERRCOUNT, | |
3767 | spa_get_errlog_size(spa)) == 0); | |
3768 | ||
3769 | if (spa_suspended(spa)) | |
3770 | VERIFY(nvlist_add_uint64(*config, | |
3771 | ZPOOL_CONFIG_SUSPENDED, | |
3772 | spa->spa_failmode) == 0); | |
3773 | ||
3774 | spa_add_spares(spa, *config); | |
3775 | spa_add_l2cache(spa, *config); | |
3776 | spa_add_feature_stats(spa, *config); | |
3777 | } | |
3778 | } | |
3779 | ||
3780 | /* | |
3781 | * We want to get the alternate root even for faulted pools, so we cheat | |
3782 | * and call spa_lookup() directly. | |
3783 | */ | |
3784 | if (altroot) { | |
3785 | if (spa == NULL) { | |
3786 | mutex_enter(&spa_namespace_lock); | |
3787 | spa = spa_lookup(name); | |
3788 | if (spa) | |
3789 | spa_altroot(spa, altroot, buflen); | |
3790 | else | |
3791 | altroot[0] = '\0'; | |
3792 | spa = NULL; | |
3793 | mutex_exit(&spa_namespace_lock); | |
3794 | } else { | |
3795 | spa_altroot(spa, altroot, buflen); | |
3796 | } | |
3797 | } | |
3798 | ||
3799 | if (spa != NULL) { | |
3800 | spa_config_exit(spa, SCL_CONFIG, FTAG); | |
3801 | spa_close(spa, FTAG); | |
3802 | } | |
3803 | ||
3804 | return (error); | |
3805 | } | |
3806 | ||
3807 | /* | |
3808 | * Validate that the auxiliary device array is well formed. We must have an | |
3809 | * array of nvlists, each which describes a valid leaf vdev. If this is an | |
3810 | * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be | |
3811 | * specified, as long as they are well-formed. | |
3812 | */ | |
3813 | static int | |
3814 | spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode, | |
3815 | spa_aux_vdev_t *sav, const char *config, uint64_t version, | |
3816 | vdev_labeltype_t label) | |
3817 | { | |
3818 | nvlist_t **dev; | |
3819 | uint_t i, ndev; | |
3820 | vdev_t *vd; | |
3821 | int error; | |
3822 | ||
3823 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); | |
3824 | ||
3825 | /* | |
3826 | * It's acceptable to have no devs specified. | |
3827 | */ | |
3828 | if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0) | |
3829 | return (0); | |
3830 | ||
3831 | if (ndev == 0) | |
3832 | return (SET_ERROR(EINVAL)); | |
3833 | ||
3834 | /* | |
3835 | * Make sure the pool is formatted with a version that supports this | |
3836 | * device type. | |
3837 | */ | |
3838 | if (spa_version(spa) < version) | |
3839 | return (SET_ERROR(ENOTSUP)); | |
3840 | ||
3841 | /* | |
3842 | * Set the pending device list so we correctly handle device in-use | |
3843 | * checking. | |
3844 | */ | |
3845 | sav->sav_pending = dev; | |
3846 | sav->sav_npending = ndev; | |
3847 | ||
3848 | for (i = 0; i < ndev; i++) { | |
3849 | if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0, | |
3850 | mode)) != 0) | |
3851 | goto out; | |
3852 | ||
3853 | if (!vd->vdev_ops->vdev_op_leaf) { | |
3854 | vdev_free(vd); | |
3855 | error = SET_ERROR(EINVAL); | |
3856 | goto out; | |
3857 | } | |
3858 | ||
3859 | vd->vdev_top = vd; | |
3860 | ||
3861 | if ((error = vdev_open(vd)) == 0 && | |
3862 | (error = vdev_label_init(vd, crtxg, label)) == 0) { | |
3863 | VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID, | |
3864 | vd->vdev_guid) == 0); | |
3865 | } | |
3866 | ||
3867 | vdev_free(vd); | |
3868 | ||
3869 | if (error && | |
3870 | (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE)) | |
3871 | goto out; | |
3872 | else | |
3873 | error = 0; | |
3874 | } | |
3875 | ||
3876 | out: | |
3877 | sav->sav_pending = NULL; | |
3878 | sav->sav_npending = 0; | |
3879 | return (error); | |
3880 | } | |
3881 | ||
3882 | static int | |
3883 | spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode) | |
3884 | { | |
3885 | int error; | |
3886 | ||
3887 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); | |
3888 | ||
3889 | if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode, | |
3890 | &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES, | |
3891 | VDEV_LABEL_SPARE)) != 0) { | |
3892 | return (error); | |
3893 | } | |
3894 | ||
3895 | return (spa_validate_aux_devs(spa, nvroot, crtxg, mode, | |
3896 | &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE, | |
3897 | VDEV_LABEL_L2CACHE)); | |
3898 | } | |
3899 | ||
3900 | static void | |
3901 | spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs, | |
3902 | const char *config) | |
3903 | { | |
3904 | int i; | |
3905 | ||
3906 | if (sav->sav_config != NULL) { | |
3907 | nvlist_t **olddevs; | |
3908 | uint_t oldndevs; | |
3909 | nvlist_t **newdevs; | |
3910 | ||
3911 | /* | |
3912 | * Generate new dev list by concatenating with the | |
3913 | * current dev list. | |
3914 | */ | |
3915 | VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config, | |
3916 | &olddevs, &oldndevs) == 0); | |
3917 | ||
3918 | newdevs = kmem_alloc(sizeof (void *) * | |
3919 | (ndevs + oldndevs), KM_SLEEP); | |
3920 | for (i = 0; i < oldndevs; i++) | |
3921 | VERIFY(nvlist_dup(olddevs[i], &newdevs[i], | |
3922 | KM_SLEEP) == 0); | |
3923 | for (i = 0; i < ndevs; i++) | |
3924 | VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs], | |
3925 | KM_SLEEP) == 0); | |
3926 | ||
3927 | VERIFY(nvlist_remove(sav->sav_config, config, | |
3928 | DATA_TYPE_NVLIST_ARRAY) == 0); | |
3929 | ||
3930 | VERIFY(nvlist_add_nvlist_array(sav->sav_config, | |
3931 | config, newdevs, ndevs + oldndevs) == 0); | |
3932 | for (i = 0; i < oldndevs + ndevs; i++) | |
3933 | nvlist_free(newdevs[i]); | |
3934 | kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *)); | |
3935 | } else { | |
3936 | /* | |
3937 | * Generate a new dev list. | |
3938 | */ | |
3939 | VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME, | |
3940 | KM_SLEEP) == 0); | |
3941 | VERIFY(nvlist_add_nvlist_array(sav->sav_config, config, | |
3942 | devs, ndevs) == 0); | |
3943 | } | |
3944 | } | |
3945 | ||
3946 | /* | |
3947 | * Stop and drop level 2 ARC devices | |
3948 | */ | |
3949 | void | |
3950 | spa_l2cache_drop(spa_t *spa) | |
3951 | { | |
3952 | vdev_t *vd; | |
3953 | int i; | |
3954 | spa_aux_vdev_t *sav = &spa->spa_l2cache; | |
3955 | ||
3956 | for (i = 0; i < sav->sav_count; i++) { | |
3957 | uint64_t pool; | |
3958 | ||
3959 | vd = sav->sav_vdevs[i]; | |
3960 | ASSERT(vd != NULL); | |
3961 | ||
3962 | if (spa_l2cache_exists(vd->vdev_guid, &pool) && | |
3963 | pool != 0ULL && l2arc_vdev_present(vd)) | |
3964 | l2arc_remove_vdev(vd); | |
3965 | } | |
3966 | } | |
3967 | ||
3968 | /* | |
3969 | * Verify encryption parameters for spa creation. If we are encrypting, we must | |
3970 | * have the encryption feature flag enabled. | |
3971 | */ | |
3972 | static int | |
3973 | spa_create_check_encryption_params(dsl_crypto_params_t *dcp, | |
3974 | boolean_t has_encryption) | |
3975 | { | |
3976 | if (dcp->cp_crypt != ZIO_CRYPT_OFF && | |
3977 | dcp->cp_crypt != ZIO_CRYPT_INHERIT && | |
3978 | !has_encryption) | |
3979 | return (SET_ERROR(ENOTSUP)); | |
3980 | ||
3981 | return (dmu_objset_create_crypt_check(NULL, dcp)); | |
3982 | } | |
3983 | ||
3984 | /* | |
3985 | * Pool Creation | |
3986 | */ | |
3987 | int | |
3988 | spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props, | |
3989 | nvlist_t *zplprops, dsl_crypto_params_t *dcp) | |
3990 | { | |
3991 | spa_t *spa; | |
3992 | char *altroot = NULL; | |
3993 | vdev_t *rvd; | |
3994 | dsl_pool_t *dp; | |
3995 | dmu_tx_t *tx; | |
3996 | int error = 0; | |
3997 | uint64_t txg = TXG_INITIAL; | |
3998 | nvlist_t **spares, **l2cache; | |
3999 | uint_t nspares, nl2cache; | |
4000 | uint64_t version, obj, root_dsobj = 0; | |
4001 | boolean_t has_features; | |
4002 | boolean_t has_encryption; | |
4003 | spa_feature_t feat; | |
4004 | char *feat_name; | |
4005 | char *poolname; | |
4006 | nvlist_t *nvl; | |
4007 | ||
4008 | if (nvlist_lookup_string(props, "tname", &poolname) != 0) | |
4009 | poolname = (char *)pool; | |
4010 | ||
4011 | /* | |
4012 | * If this pool already exists, return failure. | |
4013 | */ | |
4014 | mutex_enter(&spa_namespace_lock); | |
4015 | if (spa_lookup(poolname) != NULL) { | |
4016 | mutex_exit(&spa_namespace_lock); | |
4017 | return (SET_ERROR(EEXIST)); | |
4018 | } | |
4019 | ||
4020 | /* | |
4021 | * Allocate a new spa_t structure. | |
4022 | */ | |
4023 | nvl = fnvlist_alloc(); | |
4024 | fnvlist_add_string(nvl, ZPOOL_CONFIG_POOL_NAME, pool); | |
4025 | (void) nvlist_lookup_string(props, | |
4026 | zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); | |
4027 | spa = spa_add(poolname, nvl, altroot); | |
4028 | fnvlist_free(nvl); | |
4029 | spa_activate(spa, spa_mode_global); | |
4030 | ||
4031 | if (props && (error = spa_prop_validate(spa, props))) { | |
4032 | spa_deactivate(spa); | |
4033 | spa_remove(spa); | |
4034 | mutex_exit(&spa_namespace_lock); | |
4035 | return (error); | |
4036 | } | |
4037 | ||
4038 | /* | |
4039 | * Temporary pool names should never be written to disk. | |
4040 | */ | |
4041 | if (poolname != pool) | |
4042 | spa->spa_import_flags |= ZFS_IMPORT_TEMP_NAME; | |
4043 | ||
4044 | has_features = B_FALSE; | |
4045 | has_encryption = B_FALSE; | |
4046 | for (nvpair_t *elem = nvlist_next_nvpair(props, NULL); | |
4047 | elem != NULL; elem = nvlist_next_nvpair(props, elem)) { | |
4048 | if (zpool_prop_feature(nvpair_name(elem))) { | |
4049 | has_features = B_TRUE; | |
4050 | ||
4051 | feat_name = strchr(nvpair_name(elem), '@') + 1; | |
4052 | VERIFY0(zfeature_lookup_name(feat_name, &feat)); | |
4053 | if (feat == SPA_FEATURE_ENCRYPTION) | |
4054 | has_encryption = B_TRUE; | |
4055 | } | |
4056 | } | |
4057 | ||
4058 | /* verify encryption params, if they were provided */ | |
4059 | if (dcp != NULL) { | |
4060 | error = spa_create_check_encryption_params(dcp, has_encryption); | |
4061 | if (error != 0) { | |
4062 | spa_deactivate(spa); | |
4063 | spa_remove(spa); | |
4064 | mutex_exit(&spa_namespace_lock); | |
4065 | return (error); | |
4066 | } | |
4067 | } | |
4068 | ||
4069 | if (has_features || nvlist_lookup_uint64(props, | |
4070 | zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) { | |
4071 | version = SPA_VERSION; | |
4072 | } | |
4073 | ASSERT(SPA_VERSION_IS_SUPPORTED(version)); | |
4074 | ||
4075 | spa->spa_first_txg = txg; | |
4076 | spa->spa_uberblock.ub_txg = txg - 1; | |
4077 | spa->spa_uberblock.ub_version = version; | |
4078 | spa->spa_ubsync = spa->spa_uberblock; | |
4079 | spa->spa_load_state = SPA_LOAD_CREATE; | |
4080 | ||
4081 | /* | |
4082 | * Create "The Godfather" zio to hold all async IOs | |
4083 | */ | |
4084 | spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), | |
4085 | KM_SLEEP); | |
4086 | for (int i = 0; i < max_ncpus; i++) { | |
4087 | spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, | |
4088 | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | | |
4089 | ZIO_FLAG_GODFATHER); | |
4090 | } | |
4091 | ||
4092 | /* | |
4093 | * Create the root vdev. | |
4094 | */ | |
4095 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
4096 | ||
4097 | error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD); | |
4098 | ||
4099 | ASSERT(error != 0 || rvd != NULL); | |
4100 | ASSERT(error != 0 || spa->spa_root_vdev == rvd); | |
4101 | ||
4102 | if (error == 0 && !zfs_allocatable_devs(nvroot)) | |
4103 | error = SET_ERROR(EINVAL); | |
4104 | ||
4105 | if (error == 0 && | |
4106 | (error = vdev_create(rvd, txg, B_FALSE)) == 0 && | |
4107 | (error = spa_validate_aux(spa, nvroot, txg, | |
4108 | VDEV_ALLOC_ADD)) == 0) { | |
4109 | for (int c = 0; c < rvd->vdev_children; c++) { | |
4110 | vdev_metaslab_set_size(rvd->vdev_child[c]); | |
4111 | vdev_expand(rvd->vdev_child[c], txg); | |
4112 | } | |
4113 | } | |
4114 | ||
4115 | spa_config_exit(spa, SCL_ALL, FTAG); | |
4116 | ||
4117 | if (error != 0) { | |
4118 | spa_unload(spa); | |
4119 | spa_deactivate(spa); | |
4120 | spa_remove(spa); | |
4121 | mutex_exit(&spa_namespace_lock); | |
4122 | return (error); | |
4123 | } | |
4124 | ||
4125 | /* | |
4126 | * Get the list of spares, if specified. | |
4127 | */ | |
4128 | if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, | |
4129 | &spares, &nspares) == 0) { | |
4130 | VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME, | |
4131 | KM_SLEEP) == 0); | |
4132 | VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, | |
4133 | ZPOOL_CONFIG_SPARES, spares, nspares) == 0); | |
4134 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
4135 | spa_load_spares(spa); | |
4136 | spa_config_exit(spa, SCL_ALL, FTAG); | |
4137 | spa->spa_spares.sav_sync = B_TRUE; | |
4138 | } | |
4139 | ||
4140 | /* | |
4141 | * Get the list of level 2 cache devices, if specified. | |
4142 | */ | |
4143 | if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, | |
4144 | &l2cache, &nl2cache) == 0) { | |
4145 | VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, | |
4146 | NV_UNIQUE_NAME, KM_SLEEP) == 0); | |
4147 | VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, | |
4148 | ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); | |
4149 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
4150 | spa_load_l2cache(spa); | |
4151 | spa_config_exit(spa, SCL_ALL, FTAG); | |
4152 | spa->spa_l2cache.sav_sync = B_TRUE; | |
4153 | } | |
4154 | ||
4155 | spa->spa_is_initializing = B_TRUE; | |
4156 | spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, dcp, txg); | |
4157 | spa->spa_is_initializing = B_FALSE; | |
4158 | ||
4159 | /* | |
4160 | * Create DDTs (dedup tables). | |
4161 | */ | |
4162 | ddt_create(spa); | |
4163 | ||
4164 | spa_update_dspace(spa); | |
4165 | ||
4166 | tx = dmu_tx_create_assigned(dp, txg); | |
4167 | ||
4168 | /* | |
4169 | * Create the pool's history object. | |
4170 | */ | |
4171 | if (version >= SPA_VERSION_ZPOOL_HISTORY && !spa->spa_history) | |
4172 | spa_history_create_obj(spa, tx); | |
4173 | ||
4174 | spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_CREATE); | |
4175 | spa_history_log_version(spa, "create", tx); | |
4176 | ||
4177 | /* | |
4178 | * Create the pool config object. | |
4179 | */ | |
4180 | spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset, | |
4181 | DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE, | |
4182 | DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx); | |
4183 | ||
4184 | if (zap_add(spa->spa_meta_objset, | |
4185 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG, | |
4186 | sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) { | |
4187 | cmn_err(CE_PANIC, "failed to add pool config"); | |
4188 | } | |
4189 | ||
4190 | if (zap_add(spa->spa_meta_objset, | |
4191 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION, | |
4192 | sizeof (uint64_t), 1, &version, tx) != 0) { | |
4193 | cmn_err(CE_PANIC, "failed to add pool version"); | |
4194 | } | |
4195 | ||
4196 | /* Newly created pools with the right version are always deflated. */ | |
4197 | if (version >= SPA_VERSION_RAIDZ_DEFLATE) { | |
4198 | spa->spa_deflate = TRUE; | |
4199 | if (zap_add(spa->spa_meta_objset, | |
4200 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, | |
4201 | sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) { | |
4202 | cmn_err(CE_PANIC, "failed to add deflate"); | |
4203 | } | |
4204 | } | |
4205 | ||
4206 | /* | |
4207 | * Create the deferred-free bpobj. Turn off compression | |
4208 | * because sync-to-convergence takes longer if the blocksize | |
4209 | * keeps changing. | |
4210 | */ | |
4211 | obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx); | |
4212 | dmu_object_set_compress(spa->spa_meta_objset, obj, | |
4213 | ZIO_COMPRESS_OFF, tx); | |
4214 | if (zap_add(spa->spa_meta_objset, | |
4215 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ, | |
4216 | sizeof (uint64_t), 1, &obj, tx) != 0) { | |
4217 | cmn_err(CE_PANIC, "failed to add bpobj"); | |
4218 | } | |
4219 | VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj, | |
4220 | spa->spa_meta_objset, obj)); | |
4221 | ||
4222 | /* | |
4223 | * Generate some random noise for salted checksums to operate on. | |
4224 | */ | |
4225 | (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes, | |
4226 | sizeof (spa->spa_cksum_salt.zcs_bytes)); | |
4227 | ||
4228 | /* | |
4229 | * Set pool properties. | |
4230 | */ | |
4231 | spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS); | |
4232 | spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); | |
4233 | spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE); | |
4234 | spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND); | |
4235 | spa->spa_multihost = zpool_prop_default_numeric(ZPOOL_PROP_MULTIHOST); | |
4236 | ||
4237 | if (props != NULL) { | |
4238 | spa_configfile_set(spa, props, B_FALSE); | |
4239 | spa_sync_props(props, tx); | |
4240 | } | |
4241 | ||
4242 | dmu_tx_commit(tx); | |
4243 | ||
4244 | /* | |
4245 | * If the root dataset is encrypted we will need to create key mappings | |
4246 | * for the zio layer before we start to write any data to disk and hold | |
4247 | * them until after the first txg has been synced. Waiting for the first | |
4248 | * transaction to complete also ensures that our bean counters are | |
4249 | * appropriately updated. | |
4250 | */ | |
4251 | if (dp->dp_root_dir->dd_crypto_obj != 0) { | |
4252 | root_dsobj = dsl_dir_phys(dp->dp_root_dir)->dd_head_dataset_obj; | |
4253 | VERIFY0(spa_keystore_create_mapping_impl(spa, root_dsobj, | |
4254 | dp->dp_root_dir, FTAG)); | |
4255 | } | |
4256 | ||
4257 | spa->spa_sync_on = B_TRUE; | |
4258 | txg_sync_start(dp); | |
4259 | mmp_thread_start(spa); | |
4260 | txg_wait_synced(dp, txg); | |
4261 | ||
4262 | if (dp->dp_root_dir->dd_crypto_obj != 0) | |
4263 | VERIFY0(spa_keystore_remove_mapping(spa, root_dsobj, FTAG)); | |
4264 | ||
4265 | spa_config_sync(spa, B_FALSE, B_TRUE); | |
4266 | ||
4267 | /* | |
4268 | * Don't count references from objsets that are already closed | |
4269 | * and are making their way through the eviction process. | |
4270 | */ | |
4271 | spa_evicting_os_wait(spa); | |
4272 | spa->spa_minref = refcount_count(&spa->spa_refcount); | |
4273 | spa->spa_load_state = SPA_LOAD_NONE; | |
4274 | ||
4275 | mutex_exit(&spa_namespace_lock); | |
4276 | ||
4277 | return (0); | |
4278 | } | |
4279 | ||
4280 | /* | |
4281 | * Import a non-root pool into the system. | |
4282 | */ | |
4283 | int | |
4284 | spa_import(char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags) | |
4285 | { | |
4286 | spa_t *spa; | |
4287 | char *altroot = NULL; | |
4288 | spa_load_state_t state = SPA_LOAD_IMPORT; | |
4289 | zpool_rewind_policy_t policy; | |
4290 | uint64_t mode = spa_mode_global; | |
4291 | uint64_t readonly = B_FALSE; | |
4292 | int error; | |
4293 | nvlist_t *nvroot; | |
4294 | nvlist_t **spares, **l2cache; | |
4295 | uint_t nspares, nl2cache; | |
4296 | ||
4297 | /* | |
4298 | * If a pool with this name exists, return failure. | |
4299 | */ | |
4300 | mutex_enter(&spa_namespace_lock); | |
4301 | if (spa_lookup(pool) != NULL) { | |
4302 | mutex_exit(&spa_namespace_lock); | |
4303 | return (SET_ERROR(EEXIST)); | |
4304 | } | |
4305 | ||
4306 | /* | |
4307 | * Create and initialize the spa structure. | |
4308 | */ | |
4309 | (void) nvlist_lookup_string(props, | |
4310 | zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); | |
4311 | (void) nvlist_lookup_uint64(props, | |
4312 | zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly); | |
4313 | if (readonly) | |
4314 | mode = FREAD; | |
4315 | spa = spa_add(pool, config, altroot); | |
4316 | spa->spa_import_flags = flags; | |
4317 | ||
4318 | /* | |
4319 | * Verbatim import - Take a pool and insert it into the namespace | |
4320 | * as if it had been loaded at boot. | |
4321 | */ | |
4322 | if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) { | |
4323 | if (props != NULL) | |
4324 | spa_configfile_set(spa, props, B_FALSE); | |
4325 | ||
4326 | spa_config_sync(spa, B_FALSE, B_TRUE); | |
4327 | spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT); | |
4328 | ||
4329 | mutex_exit(&spa_namespace_lock); | |
4330 | return (0); | |
4331 | } | |
4332 | ||
4333 | spa_activate(spa, mode); | |
4334 | ||
4335 | /* | |
4336 | * Don't start async tasks until we know everything is healthy. | |
4337 | */ | |
4338 | spa_async_suspend(spa); | |
4339 | ||
4340 | zpool_get_rewind_policy(config, &policy); | |
4341 | if (policy.zrp_request & ZPOOL_DO_REWIND) | |
4342 | state = SPA_LOAD_RECOVER; | |
4343 | ||
4344 | /* | |
4345 | * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig | |
4346 | * because the user-supplied config is actually the one to trust when | |
4347 | * doing an import. | |
4348 | */ | |
4349 | if (state != SPA_LOAD_RECOVER) | |
4350 | spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; | |
4351 | ||
4352 | error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg, | |
4353 | policy.zrp_request); | |
4354 | ||
4355 | /* | |
4356 | * Propagate anything learned while loading the pool and pass it | |
4357 | * back to caller (i.e. rewind info, missing devices, etc). | |
4358 | */ | |
4359 | VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, | |
4360 | spa->spa_load_info) == 0); | |
4361 | ||
4362 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
4363 | /* | |
4364 | * Toss any existing sparelist, as it doesn't have any validity | |
4365 | * anymore, and conflicts with spa_has_spare(). | |
4366 | */ | |
4367 | if (spa->spa_spares.sav_config) { | |
4368 | nvlist_free(spa->spa_spares.sav_config); | |
4369 | spa->spa_spares.sav_config = NULL; | |
4370 | spa_load_spares(spa); | |
4371 | } | |
4372 | if (spa->spa_l2cache.sav_config) { | |
4373 | nvlist_free(spa->spa_l2cache.sav_config); | |
4374 | spa->spa_l2cache.sav_config = NULL; | |
4375 | spa_load_l2cache(spa); | |
4376 | } | |
4377 | ||
4378 | VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, | |
4379 | &nvroot) == 0); | |
4380 | spa_config_exit(spa, SCL_ALL, FTAG); | |
4381 | ||
4382 | if (props != NULL) | |
4383 | spa_configfile_set(spa, props, B_FALSE); | |
4384 | ||
4385 | if (error != 0 || (props && spa_writeable(spa) && | |
4386 | (error = spa_prop_set(spa, props)))) { | |
4387 | spa_unload(spa); | |
4388 | spa_deactivate(spa); | |
4389 | spa_remove(spa); | |
4390 | mutex_exit(&spa_namespace_lock); | |
4391 | return (error); | |
4392 | } | |
4393 | ||
4394 | spa_async_resume(spa); | |
4395 | ||
4396 | /* | |
4397 | * Override any spares and level 2 cache devices as specified by | |
4398 | * the user, as these may have correct device names/devids, etc. | |
4399 | */ | |
4400 | if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, | |
4401 | &spares, &nspares) == 0) { | |
4402 | if (spa->spa_spares.sav_config) | |
4403 | VERIFY(nvlist_remove(spa->spa_spares.sav_config, | |
4404 | ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0); | |
4405 | else | |
4406 | VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, | |
4407 | NV_UNIQUE_NAME, KM_SLEEP) == 0); | |
4408 | VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, | |
4409 | ZPOOL_CONFIG_SPARES, spares, nspares) == 0); | |
4410 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
4411 | spa_load_spares(spa); | |
4412 | spa_config_exit(spa, SCL_ALL, FTAG); | |
4413 | spa->spa_spares.sav_sync = B_TRUE; | |
4414 | } | |
4415 | if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, | |
4416 | &l2cache, &nl2cache) == 0) { | |
4417 | if (spa->spa_l2cache.sav_config) | |
4418 | VERIFY(nvlist_remove(spa->spa_l2cache.sav_config, | |
4419 | ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0); | |
4420 | else | |
4421 | VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, | |
4422 | NV_UNIQUE_NAME, KM_SLEEP) == 0); | |
4423 | VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, | |
4424 | ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); | |
4425 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
4426 | spa_load_l2cache(spa); | |
4427 | spa_config_exit(spa, SCL_ALL, FTAG); | |
4428 | spa->spa_l2cache.sav_sync = B_TRUE; | |
4429 | } | |
4430 | ||
4431 | /* | |
4432 | * Check for any removed devices. | |
4433 | */ | |
4434 | if (spa->spa_autoreplace) { | |
4435 | spa_aux_check_removed(&spa->spa_spares); | |
4436 | spa_aux_check_removed(&spa->spa_l2cache); | |
4437 | } | |
4438 | ||
4439 | if (spa_writeable(spa)) { | |
4440 | /* | |
4441 | * Update the config cache to include the newly-imported pool. | |
4442 | */ | |
4443 | spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); | |
4444 | } | |
4445 | ||
4446 | /* | |
4447 | * It's possible that the pool was expanded while it was exported. | |
4448 | * We kick off an async task to handle this for us. | |
4449 | */ | |
4450 | spa_async_request(spa, SPA_ASYNC_AUTOEXPAND); | |
4451 | ||
4452 | spa_history_log_version(spa, "import", NULL); | |
4453 | ||
4454 | spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT); | |
4455 | ||
4456 | zvol_create_minors(spa, pool, B_TRUE); | |
4457 | ||
4458 | mutex_exit(&spa_namespace_lock); | |
4459 | ||
4460 | return (0); | |
4461 | } | |
4462 | ||
4463 | nvlist_t * | |
4464 | spa_tryimport(nvlist_t *tryconfig) | |
4465 | { | |
4466 | nvlist_t *config = NULL; | |
4467 | char *poolname; | |
4468 | spa_t *spa; | |
4469 | uint64_t state; | |
4470 | int error; | |
4471 | ||
4472 | if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname)) | |
4473 | return (NULL); | |
4474 | ||
4475 | if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state)) | |
4476 | return (NULL); | |
4477 | ||
4478 | /* | |
4479 | * Create and initialize the spa structure. | |
4480 | */ | |
4481 | mutex_enter(&spa_namespace_lock); | |
4482 | spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL); | |
4483 | spa_activate(spa, FREAD); | |
4484 | ||
4485 | /* | |
4486 | * Pass off the heavy lifting to spa_load(). | |
4487 | * Pass TRUE for mosconfig because the user-supplied config | |
4488 | * is actually the one to trust when doing an import. | |
4489 | */ | |
4490 | error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE); | |
4491 | ||
4492 | /* | |
4493 | * If 'tryconfig' was at least parsable, return the current config. | |
4494 | */ | |
4495 | if (spa->spa_root_vdev != NULL) { | |
4496 | config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); | |
4497 | VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, | |
4498 | poolname) == 0); | |
4499 | VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, | |
4500 | state) == 0); | |
4501 | VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP, | |
4502 | spa->spa_uberblock.ub_timestamp) == 0); | |
4503 | VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, | |
4504 | spa->spa_load_info) == 0); | |
4505 | VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_ERRATA, | |
4506 | spa->spa_errata) == 0); | |
4507 | ||
4508 | /* | |
4509 | * If the bootfs property exists on this pool then we | |
4510 | * copy it out so that external consumers can tell which | |
4511 | * pools are bootable. | |
4512 | */ | |
4513 | if ((!error || error == EEXIST) && spa->spa_bootfs) { | |
4514 | char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP); | |
4515 | ||
4516 | /* | |
4517 | * We have to play games with the name since the | |
4518 | * pool was opened as TRYIMPORT_NAME. | |
4519 | */ | |
4520 | if (dsl_dsobj_to_dsname(spa_name(spa), | |
4521 | spa->spa_bootfs, tmpname) == 0) { | |
4522 | char *cp; | |
4523 | char *dsname; | |
4524 | ||
4525 | dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP); | |
4526 | ||
4527 | cp = strchr(tmpname, '/'); | |
4528 | if (cp == NULL) { | |
4529 | (void) strlcpy(dsname, tmpname, | |
4530 | MAXPATHLEN); | |
4531 | } else { | |
4532 | (void) snprintf(dsname, MAXPATHLEN, | |
4533 | "%s/%s", poolname, ++cp); | |
4534 | } | |
4535 | VERIFY(nvlist_add_string(config, | |
4536 | ZPOOL_CONFIG_BOOTFS, dsname) == 0); | |
4537 | kmem_free(dsname, MAXPATHLEN); | |
4538 | } | |
4539 | kmem_free(tmpname, MAXPATHLEN); | |
4540 | } | |
4541 | ||
4542 | /* | |
4543 | * Add the list of hot spares and level 2 cache devices. | |
4544 | */ | |
4545 | spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); | |
4546 | spa_add_spares(spa, config); | |
4547 | spa_add_l2cache(spa, config); | |
4548 | spa_config_exit(spa, SCL_CONFIG, FTAG); | |
4549 | } | |
4550 | ||
4551 | spa_unload(spa); | |
4552 | spa_deactivate(spa); | |
4553 | spa_remove(spa); | |
4554 | mutex_exit(&spa_namespace_lock); | |
4555 | ||
4556 | return (config); | |
4557 | } | |
4558 | ||
4559 | /* | |
4560 | * Pool export/destroy | |
4561 | * | |
4562 | * The act of destroying or exporting a pool is very simple. We make sure there | |
4563 | * is no more pending I/O and any references to the pool are gone. Then, we | |
4564 | * update the pool state and sync all the labels to disk, removing the | |
4565 | * configuration from the cache afterwards. If the 'hardforce' flag is set, then | |
4566 | * we don't sync the labels or remove the configuration cache. | |
4567 | */ | |
4568 | static int | |
4569 | spa_export_common(char *pool, int new_state, nvlist_t **oldconfig, | |
4570 | boolean_t force, boolean_t hardforce) | |
4571 | { | |
4572 | spa_t *spa; | |
4573 | ||
4574 | if (oldconfig) | |
4575 | *oldconfig = NULL; | |
4576 | ||
4577 | if (!(spa_mode_global & FWRITE)) | |
4578 | return (SET_ERROR(EROFS)); | |
4579 | ||
4580 | mutex_enter(&spa_namespace_lock); | |
4581 | if ((spa = spa_lookup(pool)) == NULL) { | |
4582 | mutex_exit(&spa_namespace_lock); | |
4583 | return (SET_ERROR(ENOENT)); | |
4584 | } | |
4585 | ||
4586 | /* | |
4587 | * Put a hold on the pool, drop the namespace lock, stop async tasks, | |
4588 | * reacquire the namespace lock, and see if we can export. | |
4589 | */ | |
4590 | spa_open_ref(spa, FTAG); | |
4591 | mutex_exit(&spa_namespace_lock); | |
4592 | spa_async_suspend(spa); | |
4593 | if (spa->spa_zvol_taskq) { | |
4594 | zvol_remove_minors(spa, spa_name(spa), B_TRUE); | |
4595 | taskq_wait(spa->spa_zvol_taskq); | |
4596 | } | |
4597 | mutex_enter(&spa_namespace_lock); | |
4598 | spa_close(spa, FTAG); | |
4599 | ||
4600 | if (spa->spa_state == POOL_STATE_UNINITIALIZED) | |
4601 | goto export_spa; | |
4602 | /* | |
4603 | * The pool will be in core if it's openable, in which case we can | |
4604 | * modify its state. Objsets may be open only because they're dirty, | |
4605 | * so we have to force it to sync before checking spa_refcnt. | |
4606 | */ | |
4607 | if (spa->spa_sync_on) { | |
4608 | txg_wait_synced(spa->spa_dsl_pool, 0); | |
4609 | spa_evicting_os_wait(spa); | |
4610 | } | |
4611 | ||
4612 | /* | |
4613 | * A pool cannot be exported or destroyed if there are active | |
4614 | * references. If we are resetting a pool, allow references by | |
4615 | * fault injection handlers. | |
4616 | */ | |
4617 | if (!spa_refcount_zero(spa) || | |
4618 | (spa->spa_inject_ref != 0 && | |
4619 | new_state != POOL_STATE_UNINITIALIZED)) { | |
4620 | spa_async_resume(spa); | |
4621 | mutex_exit(&spa_namespace_lock); | |
4622 | return (SET_ERROR(EBUSY)); | |
4623 | } | |
4624 | ||
4625 | if (spa->spa_sync_on) { | |
4626 | /* | |
4627 | * A pool cannot be exported if it has an active shared spare. | |
4628 | * This is to prevent other pools stealing the active spare | |
4629 | * from an exported pool. At user's own will, such pool can | |
4630 | * be forcedly exported. | |
4631 | */ | |
4632 | if (!force && new_state == POOL_STATE_EXPORTED && | |
4633 | spa_has_active_shared_spare(spa)) { | |
4634 | spa_async_resume(spa); | |
4635 | mutex_exit(&spa_namespace_lock); | |
4636 | return (SET_ERROR(EXDEV)); | |
4637 | } | |
4638 | ||
4639 | /* | |
4640 | * We want this to be reflected on every label, | |
4641 | * so mark them all dirty. spa_unload() will do the | |
4642 | * final sync that pushes these changes out. | |
4643 | */ | |
4644 | if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) { | |
4645 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
4646 | spa->spa_state = new_state; | |
4647 | spa->spa_final_txg = spa_last_synced_txg(spa) + | |
4648 | TXG_DEFER_SIZE + 1; | |
4649 | vdev_config_dirty(spa->spa_root_vdev); | |
4650 | spa_config_exit(spa, SCL_ALL, FTAG); | |
4651 | } | |
4652 | } | |
4653 | ||
4654 | export_spa: | |
4655 | if (new_state == POOL_STATE_DESTROYED) | |
4656 | spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_DESTROY); | |
4657 | else if (new_state == POOL_STATE_EXPORTED) | |
4658 | spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_EXPORT); | |
4659 | ||
4660 | if (spa->spa_state != POOL_STATE_UNINITIALIZED) { | |
4661 | spa_unload(spa); | |
4662 | spa_deactivate(spa); | |
4663 | } | |
4664 | ||
4665 | if (oldconfig && spa->spa_config) | |
4666 | VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0); | |
4667 | ||
4668 | if (new_state != POOL_STATE_UNINITIALIZED) { | |
4669 | if (!hardforce) | |
4670 | spa_config_sync(spa, B_TRUE, B_TRUE); | |
4671 | spa_remove(spa); | |
4672 | } | |
4673 | mutex_exit(&spa_namespace_lock); | |
4674 | ||
4675 | return (0); | |
4676 | } | |
4677 | ||
4678 | /* | |
4679 | * Destroy a storage pool. | |
4680 | */ | |
4681 | int | |
4682 | spa_destroy(char *pool) | |
4683 | { | |
4684 | return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL, | |
4685 | B_FALSE, B_FALSE)); | |
4686 | } | |
4687 | ||
4688 | /* | |
4689 | * Export a storage pool. | |
4690 | */ | |
4691 | int | |
4692 | spa_export(char *pool, nvlist_t **oldconfig, boolean_t force, | |
4693 | boolean_t hardforce) | |
4694 | { | |
4695 | return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig, | |
4696 | force, hardforce)); | |
4697 | } | |
4698 | ||
4699 | /* | |
4700 | * Similar to spa_export(), this unloads the spa_t without actually removing it | |
4701 | * from the namespace in any way. | |
4702 | */ | |
4703 | int | |
4704 | spa_reset(char *pool) | |
4705 | { | |
4706 | return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL, | |
4707 | B_FALSE, B_FALSE)); | |
4708 | } | |
4709 | ||
4710 | /* | |
4711 | * ========================================================================== | |
4712 | * Device manipulation | |
4713 | * ========================================================================== | |
4714 | */ | |
4715 | ||
4716 | /* | |
4717 | * Add a device to a storage pool. | |
4718 | */ | |
4719 | int | |
4720 | spa_vdev_add(spa_t *spa, nvlist_t *nvroot) | |
4721 | { | |
4722 | uint64_t txg, id; | |
4723 | int error; | |
4724 | vdev_t *rvd = spa->spa_root_vdev; | |
4725 | vdev_t *vd, *tvd; | |
4726 | nvlist_t **spares, **l2cache; | |
4727 | uint_t nspares, nl2cache; | |
4728 | ||
4729 | ASSERT(spa_writeable(spa)); | |
4730 | ||
4731 | txg = spa_vdev_enter(spa); | |
4732 | ||
4733 | if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0, | |
4734 | VDEV_ALLOC_ADD)) != 0) | |
4735 | return (spa_vdev_exit(spa, NULL, txg, error)); | |
4736 | ||
4737 | spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */ | |
4738 | ||
4739 | if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares, | |
4740 | &nspares) != 0) | |
4741 | nspares = 0; | |
4742 | ||
4743 | if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache, | |
4744 | &nl2cache) != 0) | |
4745 | nl2cache = 0; | |
4746 | ||
4747 | if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0) | |
4748 | return (spa_vdev_exit(spa, vd, txg, EINVAL)); | |
4749 | ||
4750 | if (vd->vdev_children != 0 && | |
4751 | (error = vdev_create(vd, txg, B_FALSE)) != 0) | |
4752 | return (spa_vdev_exit(spa, vd, txg, error)); | |
4753 | ||
4754 | /* | |
4755 | * We must validate the spares and l2cache devices after checking the | |
4756 | * children. Otherwise, vdev_inuse() will blindly overwrite the spare. | |
4757 | */ | |
4758 | if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0) | |
4759 | return (spa_vdev_exit(spa, vd, txg, error)); | |
4760 | ||
4761 | /* | |
4762 | * Transfer each new top-level vdev from vd to rvd. | |
4763 | */ | |
4764 | for (int c = 0; c < vd->vdev_children; c++) { | |
4765 | ||
4766 | /* | |
4767 | * Set the vdev id to the first hole, if one exists. | |
4768 | */ | |
4769 | for (id = 0; id < rvd->vdev_children; id++) { | |
4770 | if (rvd->vdev_child[id]->vdev_ishole) { | |
4771 | vdev_free(rvd->vdev_child[id]); | |
4772 | break; | |
4773 | } | |
4774 | } | |
4775 | tvd = vd->vdev_child[c]; | |
4776 | vdev_remove_child(vd, tvd); | |
4777 | tvd->vdev_id = id; | |
4778 | vdev_add_child(rvd, tvd); | |
4779 | vdev_config_dirty(tvd); | |
4780 | } | |
4781 | ||
4782 | if (nspares != 0) { | |
4783 | spa_set_aux_vdevs(&spa->spa_spares, spares, nspares, | |
4784 | ZPOOL_CONFIG_SPARES); | |
4785 | spa_load_spares(spa); | |
4786 | spa->spa_spares.sav_sync = B_TRUE; | |
4787 | } | |
4788 | ||
4789 | if (nl2cache != 0) { | |
4790 | spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache, | |
4791 | ZPOOL_CONFIG_L2CACHE); | |
4792 | spa_load_l2cache(spa); | |
4793 | spa->spa_l2cache.sav_sync = B_TRUE; | |
4794 | } | |
4795 | ||
4796 | /* | |
4797 | * We have to be careful when adding new vdevs to an existing pool. | |
4798 | * If other threads start allocating from these vdevs before we | |
4799 | * sync the config cache, and we lose power, then upon reboot we may | |
4800 | * fail to open the pool because there are DVAs that the config cache | |
4801 | * can't translate. Therefore, we first add the vdevs without | |
4802 | * initializing metaslabs; sync the config cache (via spa_vdev_exit()); | |
4803 | * and then let spa_config_update() initialize the new metaslabs. | |
4804 | * | |
4805 | * spa_load() checks for added-but-not-initialized vdevs, so that | |
4806 | * if we lose power at any point in this sequence, the remaining | |
4807 | * steps will be completed the next time we load the pool. | |
4808 | */ | |
4809 | (void) spa_vdev_exit(spa, vd, txg, 0); | |
4810 | ||
4811 | mutex_enter(&spa_namespace_lock); | |
4812 | spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); | |
4813 | spa_event_notify(spa, NULL, NULL, ESC_ZFS_VDEV_ADD); | |
4814 | mutex_exit(&spa_namespace_lock); | |
4815 | ||
4816 | return (0); | |
4817 | } | |
4818 | ||
4819 | /* | |
4820 | * Attach a device to a mirror. The arguments are the path to any device | |
4821 | * in the mirror, and the nvroot for the new device. If the path specifies | |
4822 | * a device that is not mirrored, we automatically insert the mirror vdev. | |
4823 | * | |
4824 | * If 'replacing' is specified, the new device is intended to replace the | |
4825 | * existing device; in this case the two devices are made into their own | |
4826 | * mirror using the 'replacing' vdev, which is functionally identical to | |
4827 | * the mirror vdev (it actually reuses all the same ops) but has a few | |
4828 | * extra rules: you can't attach to it after it's been created, and upon | |
4829 | * completion of resilvering, the first disk (the one being replaced) | |
4830 | * is automatically detached. | |
4831 | */ | |
4832 | int | |
4833 | spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing) | |
4834 | { | |
4835 | uint64_t txg, dtl_max_txg; | |
4836 | ASSERTV(vdev_t *rvd = spa->spa_root_vdev); | |
4837 | vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd; | |
4838 | vdev_ops_t *pvops; | |
4839 | char *oldvdpath, *newvdpath; | |
4840 | int newvd_isspare; | |
4841 | int error; | |
4842 | ||
4843 | ASSERT(spa_writeable(spa)); | |
4844 | ||
4845 | txg = spa_vdev_enter(spa); | |
4846 | ||
4847 | oldvd = spa_lookup_by_guid(spa, guid, B_FALSE); | |
4848 | ||
4849 | if (oldvd == NULL) | |
4850 | return (spa_vdev_exit(spa, NULL, txg, ENODEV)); | |
4851 | ||
4852 | if (!oldvd->vdev_ops->vdev_op_leaf) | |
4853 | return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); | |
4854 | ||
4855 | pvd = oldvd->vdev_parent; | |
4856 | ||
4857 | if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0, | |
4858 | VDEV_ALLOC_ATTACH)) != 0) | |
4859 | return (spa_vdev_exit(spa, NULL, txg, EINVAL)); | |
4860 | ||
4861 | if (newrootvd->vdev_children != 1) | |
4862 | return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); | |
4863 | ||
4864 | newvd = newrootvd->vdev_child[0]; | |
4865 | ||
4866 | if (!newvd->vdev_ops->vdev_op_leaf) | |
4867 | return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); | |
4868 | ||
4869 | if ((error = vdev_create(newrootvd, txg, replacing)) != 0) | |
4870 | return (spa_vdev_exit(spa, newrootvd, txg, error)); | |
4871 | ||
4872 | /* | |
4873 | * Spares can't replace logs | |
4874 | */ | |
4875 | if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare) | |
4876 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); | |
4877 | ||
4878 | if (!replacing) { | |
4879 | /* | |
4880 | * For attach, the only allowable parent is a mirror or the root | |
4881 | * vdev. | |
4882 | */ | |
4883 | if (pvd->vdev_ops != &vdev_mirror_ops && | |
4884 | pvd->vdev_ops != &vdev_root_ops) | |
4885 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); | |
4886 | ||
4887 | pvops = &vdev_mirror_ops; | |
4888 | } else { | |
4889 | /* | |
4890 | * Active hot spares can only be replaced by inactive hot | |
4891 | * spares. | |
4892 | */ | |
4893 | if (pvd->vdev_ops == &vdev_spare_ops && | |
4894 | oldvd->vdev_isspare && | |
4895 | !spa_has_spare(spa, newvd->vdev_guid)) | |
4896 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); | |
4897 | ||
4898 | /* | |
4899 | * If the source is a hot spare, and the parent isn't already a | |
4900 | * spare, then we want to create a new hot spare. Otherwise, we | |
4901 | * want to create a replacing vdev. The user is not allowed to | |
4902 | * attach to a spared vdev child unless the 'isspare' state is | |
4903 | * the same (spare replaces spare, non-spare replaces | |
4904 | * non-spare). | |
4905 | */ | |
4906 | if (pvd->vdev_ops == &vdev_replacing_ops && | |
4907 | spa_version(spa) < SPA_VERSION_MULTI_REPLACE) { | |
4908 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); | |
4909 | } else if (pvd->vdev_ops == &vdev_spare_ops && | |
4910 | newvd->vdev_isspare != oldvd->vdev_isspare) { | |
4911 | return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); | |
4912 | } | |
4913 | ||
4914 | if (newvd->vdev_isspare) | |
4915 | pvops = &vdev_spare_ops; | |
4916 | else | |
4917 | pvops = &vdev_replacing_ops; | |
4918 | } | |
4919 | ||
4920 | /* | |
4921 | * Make sure the new device is big enough. | |
4922 | */ | |
4923 | if (newvd->vdev_asize < vdev_get_min_asize(oldvd)) | |
4924 | return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW)); | |
4925 | ||
4926 | /* | |
4927 | * The new device cannot have a higher alignment requirement | |
4928 | * than the top-level vdev. | |
4929 | */ | |
4930 | if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift) | |
4931 | return (spa_vdev_exit(spa, newrootvd, txg, EDOM)); | |
4932 | ||
4933 | /* | |
4934 | * If this is an in-place replacement, update oldvd's path and devid | |
4935 | * to make it distinguishable from newvd, and unopenable from now on. | |
4936 | */ | |
4937 | if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) { | |
4938 | spa_strfree(oldvd->vdev_path); | |
4939 | oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5, | |
4940 | KM_SLEEP); | |
4941 | (void) sprintf(oldvd->vdev_path, "%s/%s", | |
4942 | newvd->vdev_path, "old"); | |
4943 | if (oldvd->vdev_devid != NULL) { | |
4944 | spa_strfree(oldvd->vdev_devid); | |
4945 | oldvd->vdev_devid = NULL; | |
4946 | } | |
4947 | } | |
4948 | ||
4949 | /* mark the device being resilvered */ | |
4950 | newvd->vdev_resilver_txg = txg; | |
4951 | ||
4952 | /* | |
4953 | * If the parent is not a mirror, or if we're replacing, insert the new | |
4954 | * mirror/replacing/spare vdev above oldvd. | |
4955 | */ | |
4956 | if (pvd->vdev_ops != pvops) | |
4957 | pvd = vdev_add_parent(oldvd, pvops); | |
4958 | ||
4959 | ASSERT(pvd->vdev_top->vdev_parent == rvd); | |
4960 | ASSERT(pvd->vdev_ops == pvops); | |
4961 | ASSERT(oldvd->vdev_parent == pvd); | |
4962 | ||
4963 | /* | |
4964 | * Extract the new device from its root and add it to pvd. | |
4965 | */ | |
4966 | vdev_remove_child(newrootvd, newvd); | |
4967 | newvd->vdev_id = pvd->vdev_children; | |
4968 | newvd->vdev_crtxg = oldvd->vdev_crtxg; | |
4969 | vdev_add_child(pvd, newvd); | |
4970 | ||
4971 | /* | |
4972 | * Reevaluate the parent vdev state. | |
4973 | */ | |
4974 | vdev_propagate_state(pvd); | |
4975 | ||
4976 | tvd = newvd->vdev_top; | |
4977 | ASSERT(pvd->vdev_top == tvd); | |
4978 | ASSERT(tvd->vdev_parent == rvd); | |
4979 | ||
4980 | vdev_config_dirty(tvd); | |
4981 | ||
4982 | /* | |
4983 | * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account | |
4984 | * for any dmu_sync-ed blocks. It will propagate upward when | |
4985 | * spa_vdev_exit() calls vdev_dtl_reassess(). | |
4986 | */ | |
4987 | dtl_max_txg = txg + TXG_CONCURRENT_STATES; | |
4988 | ||
4989 | vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL, | |
4990 | dtl_max_txg - TXG_INITIAL); | |
4991 | ||
4992 | if (newvd->vdev_isspare) { | |
4993 | spa_spare_activate(newvd); | |
4994 | spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_SPARE); | |
4995 | } | |
4996 | ||
4997 | oldvdpath = spa_strdup(oldvd->vdev_path); | |
4998 | newvdpath = spa_strdup(newvd->vdev_path); | |
4999 | newvd_isspare = newvd->vdev_isspare; | |
5000 | ||
5001 | /* | |
5002 | * Mark newvd's DTL dirty in this txg. | |
5003 | */ | |
5004 | vdev_dirty(tvd, VDD_DTL, newvd, txg); | |
5005 | ||
5006 | /* | |
5007 | * Schedule the resilver to restart in the future. We do this to | |
5008 | * ensure that dmu_sync-ed blocks have been stitched into the | |
5009 | * respective datasets. | |
5010 | */ | |
5011 | dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg); | |
5012 | ||
5013 | if (spa->spa_bootfs) | |
5014 | spa_event_notify(spa, newvd, NULL, ESC_ZFS_BOOTFS_VDEV_ATTACH); | |
5015 | ||
5016 | spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_ATTACH); | |
5017 | ||
5018 | /* | |
5019 | * Commit the config | |
5020 | */ | |
5021 | (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0); | |
5022 | ||
5023 | spa_history_log_internal(spa, "vdev attach", NULL, | |
5024 | "%s vdev=%s %s vdev=%s", | |
5025 | replacing && newvd_isspare ? "spare in" : | |
5026 | replacing ? "replace" : "attach", newvdpath, | |
5027 | replacing ? "for" : "to", oldvdpath); | |
5028 | ||
5029 | spa_strfree(oldvdpath); | |
5030 | spa_strfree(newvdpath); | |
5031 | ||
5032 | return (0); | |
5033 | } | |
5034 | ||
5035 | /* | |
5036 | * Detach a device from a mirror or replacing vdev. | |
5037 | * | |
5038 | * If 'replace_done' is specified, only detach if the parent | |
5039 | * is a replacing vdev. | |
5040 | */ | |
5041 | int | |
5042 | spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done) | |
5043 | { | |
5044 | uint64_t txg; | |
5045 | int error; | |
5046 | ASSERTV(vdev_t *rvd = spa->spa_root_vdev); | |
5047 | vdev_t *vd, *pvd, *cvd, *tvd; | |
5048 | boolean_t unspare = B_FALSE; | |
5049 | uint64_t unspare_guid = 0; | |
5050 | char *vdpath; | |
5051 | ||
5052 | ASSERT(spa_writeable(spa)); | |
5053 | ||
5054 | txg = spa_vdev_enter(spa); | |
5055 | ||
5056 | vd = spa_lookup_by_guid(spa, guid, B_FALSE); | |
5057 | ||
5058 | if (vd == NULL) | |
5059 | return (spa_vdev_exit(spa, NULL, txg, ENODEV)); | |
5060 | ||
5061 | if (!vd->vdev_ops->vdev_op_leaf) | |
5062 | return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); | |
5063 | ||
5064 | pvd = vd->vdev_parent; | |
5065 | ||
5066 | /* | |
5067 | * If the parent/child relationship is not as expected, don't do it. | |
5068 | * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing | |
5069 | * vdev that's replacing B with C. The user's intent in replacing | |
5070 | * is to go from M(A,B) to M(A,C). If the user decides to cancel | |
5071 | * the replace by detaching C, the expected behavior is to end up | |
5072 | * M(A,B). But suppose that right after deciding to detach C, | |
5073 | * the replacement of B completes. We would have M(A,C), and then | |
5074 | * ask to detach C, which would leave us with just A -- not what | |
5075 | * the user wanted. To prevent this, we make sure that the | |
5076 | * parent/child relationship hasn't changed -- in this example, | |
5077 | * that C's parent is still the replacing vdev R. | |
5078 | */ | |
5079 | if (pvd->vdev_guid != pguid && pguid != 0) | |
5080 | return (spa_vdev_exit(spa, NULL, txg, EBUSY)); | |
5081 | ||
5082 | /* | |
5083 | * Only 'replacing' or 'spare' vdevs can be replaced. | |
5084 | */ | |
5085 | if (replace_done && pvd->vdev_ops != &vdev_replacing_ops && | |
5086 | pvd->vdev_ops != &vdev_spare_ops) | |
5087 | return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); | |
5088 | ||
5089 | ASSERT(pvd->vdev_ops != &vdev_spare_ops || | |
5090 | spa_version(spa) >= SPA_VERSION_SPARES); | |
5091 | ||
5092 | /* | |
5093 | * Only mirror, replacing, and spare vdevs support detach. | |
5094 | */ | |
5095 | if (pvd->vdev_ops != &vdev_replacing_ops && | |
5096 | pvd->vdev_ops != &vdev_mirror_ops && | |
5097 | pvd->vdev_ops != &vdev_spare_ops) | |
5098 | return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); | |
5099 | ||
5100 | /* | |
5101 | * If this device has the only valid copy of some data, | |
5102 | * we cannot safely detach it. | |
5103 | */ | |
5104 | if (vdev_dtl_required(vd)) | |
5105 | return (spa_vdev_exit(spa, NULL, txg, EBUSY)); | |
5106 | ||
5107 | ASSERT(pvd->vdev_children >= 2); | |
5108 | ||
5109 | /* | |
5110 | * If we are detaching the second disk from a replacing vdev, then | |
5111 | * check to see if we changed the original vdev's path to have "/old" | |
5112 | * at the end in spa_vdev_attach(). If so, undo that change now. | |
5113 | */ | |
5114 | if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 && | |
5115 | vd->vdev_path != NULL) { | |
5116 | size_t len = strlen(vd->vdev_path); | |
5117 | ||
5118 | for (int c = 0; c < pvd->vdev_children; c++) { | |
5119 | cvd = pvd->vdev_child[c]; | |
5120 | ||
5121 | if (cvd == vd || cvd->vdev_path == NULL) | |
5122 | continue; | |
5123 | ||
5124 | if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 && | |
5125 | strcmp(cvd->vdev_path + len, "/old") == 0) { | |
5126 | spa_strfree(cvd->vdev_path); | |
5127 | cvd->vdev_path = spa_strdup(vd->vdev_path); | |
5128 | break; | |
5129 | } | |
5130 | } | |
5131 | } | |
5132 | ||
5133 | /* | |
5134 | * If we are detaching the original disk from a spare, then it implies | |
5135 | * that the spare should become a real disk, and be removed from the | |
5136 | * active spare list for the pool. | |
5137 | */ | |
5138 | if (pvd->vdev_ops == &vdev_spare_ops && | |
5139 | vd->vdev_id == 0 && | |
5140 | pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare) | |
5141 | unspare = B_TRUE; | |
5142 | ||
5143 | /* | |
5144 | * Erase the disk labels so the disk can be used for other things. | |
5145 | * This must be done after all other error cases are handled, | |
5146 | * but before we disembowel vd (so we can still do I/O to it). | |
5147 | * But if we can't do it, don't treat the error as fatal -- | |
5148 | * it may be that the unwritability of the disk is the reason | |
5149 | * it's being detached! | |
5150 | */ | |
5151 | error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); | |
5152 | ||
5153 | /* | |
5154 | * Remove vd from its parent and compact the parent's children. | |
5155 | */ | |
5156 | vdev_remove_child(pvd, vd); | |
5157 | vdev_compact_children(pvd); | |
5158 | ||
5159 | /* | |
5160 | * Remember one of the remaining children so we can get tvd below. | |
5161 | */ | |
5162 | cvd = pvd->vdev_child[pvd->vdev_children - 1]; | |
5163 | ||
5164 | /* | |
5165 | * If we need to remove the remaining child from the list of hot spares, | |
5166 | * do it now, marking the vdev as no longer a spare in the process. | |
5167 | * We must do this before vdev_remove_parent(), because that can | |
5168 | * change the GUID if it creates a new toplevel GUID. For a similar | |
5169 | * reason, we must remove the spare now, in the same txg as the detach; | |
5170 | * otherwise someone could attach a new sibling, change the GUID, and | |
5171 | * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail. | |
5172 | */ | |
5173 | if (unspare) { | |
5174 | ASSERT(cvd->vdev_isspare); | |
5175 | spa_spare_remove(cvd); | |
5176 | unspare_guid = cvd->vdev_guid; | |
5177 | (void) spa_vdev_remove(spa, unspare_guid, B_TRUE); | |
5178 | cvd->vdev_unspare = B_TRUE; | |
5179 | } | |
5180 | ||
5181 | /* | |
5182 | * If the parent mirror/replacing vdev only has one child, | |
5183 | * the parent is no longer needed. Remove it from the tree. | |
5184 | */ | |
5185 | if (pvd->vdev_children == 1) { | |
5186 | if (pvd->vdev_ops == &vdev_spare_ops) | |
5187 | cvd->vdev_unspare = B_FALSE; | |
5188 | vdev_remove_parent(cvd); | |
5189 | } | |
5190 | ||
5191 | ||
5192 | /* | |
5193 | * We don't set tvd until now because the parent we just removed | |
5194 | * may have been the previous top-level vdev. | |
5195 | */ | |
5196 | tvd = cvd->vdev_top; | |
5197 | ASSERT(tvd->vdev_parent == rvd); | |
5198 | ||
5199 | /* | |
5200 | * Reevaluate the parent vdev state. | |
5201 | */ | |
5202 | vdev_propagate_state(cvd); | |
5203 | ||
5204 | /* | |
5205 | * If the 'autoexpand' property is set on the pool then automatically | |
5206 | * try to expand the size of the pool. For example if the device we | |
5207 | * just detached was smaller than the others, it may be possible to | |
5208 | * add metaslabs (i.e. grow the pool). We need to reopen the vdev | |
5209 | * first so that we can obtain the updated sizes of the leaf vdevs. | |
5210 | */ | |
5211 | if (spa->spa_autoexpand) { | |
5212 | vdev_reopen(tvd); | |
5213 | vdev_expand(tvd, txg); | |
5214 | } | |
5215 | ||
5216 | vdev_config_dirty(tvd); | |
5217 | ||
5218 | /* | |
5219 | * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that | |
5220 | * vd->vdev_detached is set and free vd's DTL object in syncing context. | |
5221 | * But first make sure we're not on any *other* txg's DTL list, to | |
5222 | * prevent vd from being accessed after it's freed. | |
5223 | */ | |
5224 | vdpath = spa_strdup(vd->vdev_path ? vd->vdev_path : "none"); | |
5225 | for (int t = 0; t < TXG_SIZE; t++) | |
5226 | (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t); | |
5227 | vd->vdev_detached = B_TRUE; | |
5228 | vdev_dirty(tvd, VDD_DTL, vd, txg); | |
5229 | ||
5230 | spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_REMOVE); | |
5231 | ||
5232 | /* hang on to the spa before we release the lock */ | |
5233 | spa_open_ref(spa, FTAG); | |
5234 | ||
5235 | error = spa_vdev_exit(spa, vd, txg, 0); | |
5236 | ||
5237 | spa_history_log_internal(spa, "detach", NULL, | |
5238 | "vdev=%s", vdpath); | |
5239 | spa_strfree(vdpath); | |
5240 | ||
5241 | /* | |
5242 | * If this was the removal of the original device in a hot spare vdev, | |
5243 | * then we want to go through and remove the device from the hot spare | |
5244 | * list of every other pool. | |
5245 | */ | |
5246 | if (unspare) { | |
5247 | spa_t *altspa = NULL; | |
5248 | ||
5249 | mutex_enter(&spa_namespace_lock); | |
5250 | while ((altspa = spa_next(altspa)) != NULL) { | |
5251 | if (altspa->spa_state != POOL_STATE_ACTIVE || | |
5252 | altspa == spa) | |
5253 | continue; | |
5254 | ||
5255 | spa_open_ref(altspa, FTAG); | |
5256 | mutex_exit(&spa_namespace_lock); | |
5257 | (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE); | |
5258 | mutex_enter(&spa_namespace_lock); | |
5259 | spa_close(altspa, FTAG); | |
5260 | } | |
5261 | mutex_exit(&spa_namespace_lock); | |
5262 | ||
5263 | /* search the rest of the vdevs for spares to remove */ | |
5264 | spa_vdev_resilver_done(spa); | |
5265 | } | |
5266 | ||
5267 | /* all done with the spa; OK to release */ | |
5268 | mutex_enter(&spa_namespace_lock); | |
5269 | spa_close(spa, FTAG); | |
5270 | mutex_exit(&spa_namespace_lock); | |
5271 | ||
5272 | return (error); | |
5273 | } | |
5274 | ||
5275 | /* | |
5276 | * Split a set of devices from their mirrors, and create a new pool from them. | |
5277 | */ | |
5278 | int | |
5279 | spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config, | |
5280 | nvlist_t *props, boolean_t exp) | |
5281 | { | |
5282 | int error = 0; | |
5283 | uint64_t txg, *glist; | |
5284 | spa_t *newspa; | |
5285 | uint_t c, children, lastlog; | |
5286 | nvlist_t **child, *nvl, *tmp; | |
5287 | dmu_tx_t *tx; | |
5288 | char *altroot = NULL; | |
5289 | vdev_t *rvd, **vml = NULL; /* vdev modify list */ | |
5290 | boolean_t activate_slog; | |
5291 | ||
5292 | ASSERT(spa_writeable(spa)); | |
5293 | ||
5294 | txg = spa_vdev_enter(spa); | |
5295 | ||
5296 | /* clear the log and flush everything up to now */ | |
5297 | activate_slog = spa_passivate_log(spa); | |
5298 | (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); | |
5299 | error = spa_offline_log(spa); | |
5300 | txg = spa_vdev_config_enter(spa); | |
5301 | ||
5302 | if (activate_slog) | |
5303 | spa_activate_log(spa); | |
5304 | ||
5305 | if (error != 0) | |
5306 | return (spa_vdev_exit(spa, NULL, txg, error)); | |
5307 | ||
5308 | /* check new spa name before going any further */ | |
5309 | if (spa_lookup(newname) != NULL) | |
5310 | return (spa_vdev_exit(spa, NULL, txg, EEXIST)); | |
5311 | ||
5312 | /* | |
5313 | * scan through all the children to ensure they're all mirrors | |
5314 | */ | |
5315 | if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 || | |
5316 | nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child, | |
5317 | &children) != 0) | |
5318 | return (spa_vdev_exit(spa, NULL, txg, EINVAL)); | |
5319 | ||
5320 | /* first, check to ensure we've got the right child count */ | |
5321 | rvd = spa->spa_root_vdev; | |
5322 | lastlog = 0; | |
5323 | for (c = 0; c < rvd->vdev_children; c++) { | |
5324 | vdev_t *vd = rvd->vdev_child[c]; | |
5325 | ||
5326 | /* don't count the holes & logs as children */ | |
5327 | if (vd->vdev_islog || vd->vdev_ishole) { | |
5328 | if (lastlog == 0) | |
5329 | lastlog = c; | |
5330 | continue; | |
5331 | } | |
5332 | ||
5333 | lastlog = 0; | |
5334 | } | |
5335 | if (children != (lastlog != 0 ? lastlog : rvd->vdev_children)) | |
5336 | return (spa_vdev_exit(spa, NULL, txg, EINVAL)); | |
5337 | ||
5338 | /* next, ensure no spare or cache devices are part of the split */ | |
5339 | if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 || | |
5340 | nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0) | |
5341 | return (spa_vdev_exit(spa, NULL, txg, EINVAL)); | |
5342 | ||
5343 | vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP); | |
5344 | glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP); | |
5345 | ||
5346 | /* then, loop over each vdev and validate it */ | |
5347 | for (c = 0; c < children; c++) { | |
5348 | uint64_t is_hole = 0; | |
5349 | ||
5350 | (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE, | |
5351 | &is_hole); | |
5352 | ||
5353 | if (is_hole != 0) { | |
5354 | if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole || | |
5355 | spa->spa_root_vdev->vdev_child[c]->vdev_islog) { | |
5356 | continue; | |
5357 | } else { | |
5358 | error = SET_ERROR(EINVAL); | |
5359 | break; | |
5360 | } | |
5361 | } | |
5362 | ||
5363 | /* which disk is going to be split? */ | |
5364 | if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID, | |
5365 | &glist[c]) != 0) { | |
5366 | error = SET_ERROR(EINVAL); | |
5367 | break; | |
5368 | } | |
5369 | ||
5370 | /* look it up in the spa */ | |
5371 | vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE); | |
5372 | if (vml[c] == NULL) { | |
5373 | error = SET_ERROR(ENODEV); | |
5374 | break; | |
5375 | } | |
5376 | ||
5377 | /* make sure there's nothing stopping the split */ | |
5378 | if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops || | |
5379 | vml[c]->vdev_islog || | |
5380 | vml[c]->vdev_ishole || | |
5381 | vml[c]->vdev_isspare || | |
5382 | vml[c]->vdev_isl2cache || | |
5383 | !vdev_writeable(vml[c]) || | |
5384 | vml[c]->vdev_children != 0 || | |
5385 | vml[c]->vdev_state != VDEV_STATE_HEALTHY || | |
5386 | c != spa->spa_root_vdev->vdev_child[c]->vdev_id) { | |
5387 | error = SET_ERROR(EINVAL); | |
5388 | break; | |
5389 | } | |
5390 | ||
5391 | if (vdev_dtl_required(vml[c])) { | |
5392 | error = SET_ERROR(EBUSY); | |
5393 | break; | |
5394 | } | |
5395 | ||
5396 | /* we need certain info from the top level */ | |
5397 | VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY, | |
5398 | vml[c]->vdev_top->vdev_ms_array) == 0); | |
5399 | VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT, | |
5400 | vml[c]->vdev_top->vdev_ms_shift) == 0); | |
5401 | VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE, | |
5402 | vml[c]->vdev_top->vdev_asize) == 0); | |
5403 | VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT, | |
5404 | vml[c]->vdev_top->vdev_ashift) == 0); | |
5405 | ||
5406 | /* transfer per-vdev ZAPs */ | |
5407 | ASSERT3U(vml[c]->vdev_leaf_zap, !=, 0); | |
5408 | VERIFY0(nvlist_add_uint64(child[c], | |
5409 | ZPOOL_CONFIG_VDEV_LEAF_ZAP, vml[c]->vdev_leaf_zap)); | |
5410 | ||
5411 | ASSERT3U(vml[c]->vdev_top->vdev_top_zap, !=, 0); | |
5412 | VERIFY0(nvlist_add_uint64(child[c], | |
5413 | ZPOOL_CONFIG_VDEV_TOP_ZAP, | |
5414 | vml[c]->vdev_parent->vdev_top_zap)); | |
5415 | } | |
5416 | ||
5417 | if (error != 0) { | |
5418 | kmem_free(vml, children * sizeof (vdev_t *)); | |
5419 | kmem_free(glist, children * sizeof (uint64_t)); | |
5420 | return (spa_vdev_exit(spa, NULL, txg, error)); | |
5421 | } | |
5422 | ||
5423 | /* stop writers from using the disks */ | |
5424 | for (c = 0; c < children; c++) { | |
5425 | if (vml[c] != NULL) | |
5426 | vml[c]->vdev_offline = B_TRUE; | |
5427 | } | |
5428 | vdev_reopen(spa->spa_root_vdev); | |
5429 | ||
5430 | /* | |
5431 | * Temporarily record the splitting vdevs in the spa config. This | |
5432 | * will disappear once the config is regenerated. | |
5433 | */ | |
5434 | VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0); | |
5435 | VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, | |
5436 | glist, children) == 0); | |
5437 | kmem_free(glist, children * sizeof (uint64_t)); | |
5438 | ||
5439 | mutex_enter(&spa->spa_props_lock); | |
5440 | VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT, | |
5441 | nvl) == 0); | |
5442 | mutex_exit(&spa->spa_props_lock); | |
5443 | spa->spa_config_splitting = nvl; | |
5444 | vdev_config_dirty(spa->spa_root_vdev); | |
5445 | ||
5446 | /* configure and create the new pool */ | |
5447 | VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0); | |
5448 | VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, | |
5449 | exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0); | |
5450 | VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, | |
5451 | spa_version(spa)) == 0); | |
5452 | VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, | |
5453 | spa->spa_config_txg) == 0); | |
5454 | VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, | |
5455 | spa_generate_guid(NULL)) == 0); | |
5456 | VERIFY0(nvlist_add_boolean(config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)); | |
5457 | (void) nvlist_lookup_string(props, | |
5458 | zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); | |
5459 | ||
5460 | /* add the new pool to the namespace */ | |
5461 | newspa = spa_add(newname, config, altroot); | |
5462 | newspa->spa_avz_action = AVZ_ACTION_REBUILD; | |
5463 | newspa->spa_config_txg = spa->spa_config_txg; | |
5464 | spa_set_log_state(newspa, SPA_LOG_CLEAR); | |
5465 | ||
5466 | /* release the spa config lock, retaining the namespace lock */ | |
5467 | spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); | |
5468 | ||
5469 | if (zio_injection_enabled) | |
5470 | zio_handle_panic_injection(spa, FTAG, 1); | |
5471 | ||
5472 | spa_activate(newspa, spa_mode_global); | |
5473 | spa_async_suspend(newspa); | |
5474 | ||
5475 | /* create the new pool from the disks of the original pool */ | |
5476 | error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE); | |
5477 | if (error) | |
5478 | goto out; | |
5479 | ||
5480 | /* if that worked, generate a real config for the new pool */ | |
5481 | if (newspa->spa_root_vdev != NULL) { | |
5482 | VERIFY(nvlist_alloc(&newspa->spa_config_splitting, | |
5483 | NV_UNIQUE_NAME, KM_SLEEP) == 0); | |
5484 | VERIFY(nvlist_add_uint64(newspa->spa_config_splitting, | |
5485 | ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0); | |
5486 | spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL, | |
5487 | B_TRUE)); | |
5488 | } | |
5489 | ||
5490 | /* set the props */ | |
5491 | if (props != NULL) { | |
5492 | spa_configfile_set(newspa, props, B_FALSE); | |
5493 | error = spa_prop_set(newspa, props); | |
5494 | if (error) | |
5495 | goto out; | |
5496 | } | |
5497 | ||
5498 | /* flush everything */ | |
5499 | txg = spa_vdev_config_enter(newspa); | |
5500 | vdev_config_dirty(newspa->spa_root_vdev); | |
5501 | (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG); | |
5502 | ||
5503 | if (zio_injection_enabled) | |
5504 | zio_handle_panic_injection(spa, FTAG, 2); | |
5505 | ||
5506 | spa_async_resume(newspa); | |
5507 | ||
5508 | /* finally, update the original pool's config */ | |
5509 | txg = spa_vdev_config_enter(spa); | |
5510 | tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir); | |
5511 | error = dmu_tx_assign(tx, TXG_WAIT); | |
5512 | if (error != 0) | |
5513 | dmu_tx_abort(tx); | |
5514 | for (c = 0; c < children; c++) { | |
5515 | if (vml[c] != NULL) { | |
5516 | vdev_split(vml[c]); | |
5517 | if (error == 0) | |
5518 | spa_history_log_internal(spa, "detach", tx, | |
5519 | "vdev=%s", vml[c]->vdev_path); | |
5520 | ||
5521 | vdev_free(vml[c]); | |
5522 | } | |
5523 | } | |
5524 | spa->spa_avz_action = AVZ_ACTION_REBUILD; | |
5525 | vdev_config_dirty(spa->spa_root_vdev); | |
5526 | spa->spa_config_splitting = NULL; | |
5527 | nvlist_free(nvl); | |
5528 | if (error == 0) | |
5529 | dmu_tx_commit(tx); | |
5530 | (void) spa_vdev_exit(spa, NULL, txg, 0); | |
5531 | ||
5532 | if (zio_injection_enabled) | |
5533 | zio_handle_panic_injection(spa, FTAG, 3); | |
5534 | ||
5535 | /* split is complete; log a history record */ | |
5536 | spa_history_log_internal(newspa, "split", NULL, | |
5537 | "from pool %s", spa_name(spa)); | |
5538 | ||
5539 | kmem_free(vml, children * sizeof (vdev_t *)); | |
5540 | ||
5541 | /* if we're not going to mount the filesystems in userland, export */ | |
5542 | if (exp) | |
5543 | error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL, | |
5544 | B_FALSE, B_FALSE); | |
5545 | ||
5546 | return (error); | |
5547 | ||
5548 | out: | |
5549 | spa_unload(newspa); | |
5550 | spa_deactivate(newspa); | |
5551 | spa_remove(newspa); | |
5552 | ||
5553 | txg = spa_vdev_config_enter(spa); | |
5554 | ||
5555 | /* re-online all offlined disks */ | |
5556 | for (c = 0; c < children; c++) { | |
5557 | if (vml[c] != NULL) | |
5558 | vml[c]->vdev_offline = B_FALSE; | |
5559 | } | |
5560 | vdev_reopen(spa->spa_root_vdev); | |
5561 | ||
5562 | nvlist_free(spa->spa_config_splitting); | |
5563 | spa->spa_config_splitting = NULL; | |
5564 | (void) spa_vdev_exit(spa, NULL, txg, error); | |
5565 | ||
5566 | kmem_free(vml, children * sizeof (vdev_t *)); | |
5567 | return (error); | |
5568 | } | |
5569 | ||
5570 | static nvlist_t * | |
5571 | spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid) | |
5572 | { | |
5573 | for (int i = 0; i < count; i++) { | |
5574 | uint64_t guid; | |
5575 | ||
5576 | VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID, | |
5577 | &guid) == 0); | |
5578 | ||
5579 | if (guid == target_guid) | |
5580 | return (nvpp[i]); | |
5581 | } | |
5582 | ||
5583 | return (NULL); | |
5584 | } | |
5585 | ||
5586 | static void | |
5587 | spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count, | |
5588 | nvlist_t *dev_to_remove) | |
5589 | { | |
5590 | nvlist_t **newdev = NULL; | |
5591 | ||
5592 | if (count > 1) | |
5593 | newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP); | |
5594 | ||
5595 | for (int i = 0, j = 0; i < count; i++) { | |
5596 | if (dev[i] == dev_to_remove) | |
5597 | continue; | |
5598 | VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0); | |
5599 | } | |
5600 | ||
5601 | VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0); | |
5602 | VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0); | |
5603 | ||
5604 | for (int i = 0; i < count - 1; i++) | |
5605 | nvlist_free(newdev[i]); | |
5606 | ||
5607 | if (count > 1) | |
5608 | kmem_free(newdev, (count - 1) * sizeof (void *)); | |
5609 | } | |
5610 | ||
5611 | /* | |
5612 | * Evacuate the device. | |
5613 | */ | |
5614 | static int | |
5615 | spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd) | |
5616 | { | |
5617 | uint64_t txg; | |
5618 | int error = 0; | |
5619 | ||
5620 | ASSERT(MUTEX_HELD(&spa_namespace_lock)); | |
5621 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); | |
5622 | ASSERT(vd == vd->vdev_top); | |
5623 | ||
5624 | /* | |
5625 | * Evacuate the device. We don't hold the config lock as writer | |
5626 | * since we need to do I/O but we do keep the | |
5627 | * spa_namespace_lock held. Once this completes the device | |
5628 | * should no longer have any blocks allocated on it. | |
5629 | */ | |
5630 | if (vd->vdev_islog) { | |
5631 | if (vd->vdev_stat.vs_alloc != 0) | |
5632 | error = spa_offline_log(spa); | |
5633 | } else { | |
5634 | error = SET_ERROR(ENOTSUP); | |
5635 | } | |
5636 | ||
5637 | if (error) | |
5638 | return (error); | |
5639 | ||
5640 | /* | |
5641 | * The evacuation succeeded. Remove any remaining MOS metadata | |
5642 | * associated with this vdev, and wait for these changes to sync. | |
5643 | */ | |
5644 | ASSERT0(vd->vdev_stat.vs_alloc); | |
5645 | txg = spa_vdev_config_enter(spa); | |
5646 | vd->vdev_removing = B_TRUE; | |
5647 | vdev_dirty_leaves(vd, VDD_DTL, txg); | |
5648 | vdev_config_dirty(vd); | |
5649 | spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); | |
5650 | ||
5651 | return (0); | |
5652 | } | |
5653 | ||
5654 | /* | |
5655 | * Complete the removal by cleaning up the namespace. | |
5656 | */ | |
5657 | static void | |
5658 | spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd) | |
5659 | { | |
5660 | vdev_t *rvd = spa->spa_root_vdev; | |
5661 | uint64_t id = vd->vdev_id; | |
5662 | boolean_t last_vdev = (id == (rvd->vdev_children - 1)); | |
5663 | ||
5664 | ASSERT(MUTEX_HELD(&spa_namespace_lock)); | |
5665 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); | |
5666 | ASSERT(vd == vd->vdev_top); | |
5667 | ||
5668 | /* | |
5669 | * Only remove any devices which are empty. | |
5670 | */ | |
5671 | if (vd->vdev_stat.vs_alloc != 0) | |
5672 | return; | |
5673 | ||
5674 | (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); | |
5675 | ||
5676 | if (list_link_active(&vd->vdev_state_dirty_node)) | |
5677 | vdev_state_clean(vd); | |
5678 | if (list_link_active(&vd->vdev_config_dirty_node)) | |
5679 | vdev_config_clean(vd); | |
5680 | ||
5681 | vdev_free(vd); | |
5682 | ||
5683 | if (last_vdev) { | |
5684 | vdev_compact_children(rvd); | |
5685 | } else { | |
5686 | vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops); | |
5687 | vdev_add_child(rvd, vd); | |
5688 | } | |
5689 | vdev_config_dirty(rvd); | |
5690 | ||
5691 | /* | |
5692 | * Reassess the health of our root vdev. | |
5693 | */ | |
5694 | vdev_reopen(rvd); | |
5695 | } | |
5696 | ||
5697 | /* | |
5698 | * Remove a device from the pool - | |
5699 | * | |
5700 | * Removing a device from the vdev namespace requires several steps | |
5701 | * and can take a significant amount of time. As a result we use | |
5702 | * the spa_vdev_config_[enter/exit] functions which allow us to | |
5703 | * grab and release the spa_config_lock while still holding the namespace | |
5704 | * lock. During each step the configuration is synced out. | |
5705 | * | |
5706 | * Currently, this supports removing only hot spares, slogs, and level 2 ARC | |
5707 | * devices. | |
5708 | */ | |
5709 | int | |
5710 | spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare) | |
5711 | { | |
5712 | vdev_t *vd; | |
5713 | sysevent_t *ev = NULL; | |
5714 | metaslab_group_t *mg; | |
5715 | nvlist_t **spares, **l2cache, *nv; | |
5716 | uint64_t txg = 0; | |
5717 | uint_t nspares, nl2cache; | |
5718 | int error = 0; | |
5719 | boolean_t locked = MUTEX_HELD(&spa_namespace_lock); | |
5720 | ||
5721 | ASSERT(spa_writeable(spa)); | |
5722 | ||
5723 | if (!locked) | |
5724 | txg = spa_vdev_enter(spa); | |
5725 | ||
5726 | vd = spa_lookup_by_guid(spa, guid, B_FALSE); | |
5727 | ||
5728 | if (spa->spa_spares.sav_vdevs != NULL && | |
5729 | nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, | |
5730 | ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 && | |
5731 | (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) { | |
5732 | /* | |
5733 | * Only remove the hot spare if it's not currently in use | |
5734 | * in this pool. | |
5735 | */ | |
5736 | if (vd == NULL || unspare) { | |
5737 | if (vd == NULL) | |
5738 | vd = spa_lookup_by_guid(spa, guid, B_TRUE); | |
5739 | ev = spa_event_create(spa, vd, NULL, | |
5740 | ESC_ZFS_VDEV_REMOVE_AUX); | |
5741 | spa_vdev_remove_aux(spa->spa_spares.sav_config, | |
5742 | ZPOOL_CONFIG_SPARES, spares, nspares, nv); | |
5743 | spa_load_spares(spa); | |
5744 | spa->spa_spares.sav_sync = B_TRUE; | |
5745 | } else { | |
5746 | error = SET_ERROR(EBUSY); | |
5747 | } | |
5748 | } else if (spa->spa_l2cache.sav_vdevs != NULL && | |
5749 | nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, | |
5750 | ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 && | |
5751 | (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) { | |
5752 | /* | |
5753 | * Cache devices can always be removed. | |
5754 | */ | |
5755 | vd = spa_lookup_by_guid(spa, guid, B_TRUE); | |
5756 | ev = spa_event_create(spa, vd, NULL, ESC_ZFS_VDEV_REMOVE_AUX); | |
5757 | spa_vdev_remove_aux(spa->spa_l2cache.sav_config, | |
5758 | ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv); | |
5759 | spa_load_l2cache(spa); | |
5760 | spa->spa_l2cache.sav_sync = B_TRUE; | |
5761 | } else if (vd != NULL && vd->vdev_islog) { | |
5762 | ASSERT(!locked); | |
5763 | ASSERT(vd == vd->vdev_top); | |
5764 | ||
5765 | mg = vd->vdev_mg; | |
5766 | ||
5767 | /* | |
5768 | * Stop allocating from this vdev. | |
5769 | */ | |
5770 | metaslab_group_passivate(mg); | |
5771 | ||
5772 | /* | |
5773 | * Wait for the youngest allocations and frees to sync, | |
5774 | * and then wait for the deferral of those frees to finish. | |
5775 | */ | |
5776 | spa_vdev_config_exit(spa, NULL, | |
5777 | txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG); | |
5778 | ||
5779 | /* | |
5780 | * Attempt to evacuate the vdev. | |
5781 | */ | |
5782 | error = spa_vdev_remove_evacuate(spa, vd); | |
5783 | ||
5784 | txg = spa_vdev_config_enter(spa); | |
5785 | ||
5786 | /* | |
5787 | * If we couldn't evacuate the vdev, unwind. | |
5788 | */ | |
5789 | if (error) { | |
5790 | metaslab_group_activate(mg); | |
5791 | return (spa_vdev_exit(spa, NULL, txg, error)); | |
5792 | } | |
5793 | ||
5794 | /* | |
5795 | * Clean up the vdev namespace. | |
5796 | */ | |
5797 | ev = spa_event_create(spa, vd, NULL, ESC_ZFS_VDEV_REMOVE_DEV); | |
5798 | spa_vdev_remove_from_namespace(spa, vd); | |
5799 | ||
5800 | } else if (vd != NULL) { | |
5801 | /* | |
5802 | * Normal vdevs cannot be removed (yet). | |
5803 | */ | |
5804 | error = SET_ERROR(ENOTSUP); | |
5805 | } else { | |
5806 | /* | |
5807 | * There is no vdev of any kind with the specified guid. | |
5808 | */ | |
5809 | error = SET_ERROR(ENOENT); | |
5810 | } | |
5811 | ||
5812 | if (!locked) | |
5813 | error = spa_vdev_exit(spa, NULL, txg, error); | |
5814 | ||
5815 | if (ev) | |
5816 | spa_event_post(ev); | |
5817 | ||
5818 | return (error); | |
5819 | } | |
5820 | ||
5821 | /* | |
5822 | * Find any device that's done replacing, or a vdev marked 'unspare' that's | |
5823 | * currently spared, so we can detach it. | |
5824 | */ | |
5825 | static vdev_t * | |
5826 | spa_vdev_resilver_done_hunt(vdev_t *vd) | |
5827 | { | |
5828 | vdev_t *newvd, *oldvd; | |
5829 | ||
5830 | for (int c = 0; c < vd->vdev_children; c++) { | |
5831 | oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]); | |
5832 | if (oldvd != NULL) | |
5833 | return (oldvd); | |
5834 | } | |
5835 | ||
5836 | /* | |
5837 | * Check for a completed replacement. We always consider the first | |
5838 | * vdev in the list to be the oldest vdev, and the last one to be | |
5839 | * the newest (see spa_vdev_attach() for how that works). In | |
5840 | * the case where the newest vdev is faulted, we will not automatically | |
5841 | * remove it after a resilver completes. This is OK as it will require | |
5842 | * user intervention to determine which disk the admin wishes to keep. | |
5843 | */ | |
5844 | if (vd->vdev_ops == &vdev_replacing_ops) { | |
5845 | ASSERT(vd->vdev_children > 1); | |
5846 | ||
5847 | newvd = vd->vdev_child[vd->vdev_children - 1]; | |
5848 | oldvd = vd->vdev_child[0]; | |
5849 | ||
5850 | if (vdev_dtl_empty(newvd, DTL_MISSING) && | |
5851 | vdev_dtl_empty(newvd, DTL_OUTAGE) && | |
5852 | !vdev_dtl_required(oldvd)) | |
5853 | return (oldvd); | |
5854 | } | |
5855 | ||
5856 | /* | |
5857 | * Check for a completed resilver with the 'unspare' flag set. | |
5858 | */ | |
5859 | if (vd->vdev_ops == &vdev_spare_ops) { | |
5860 | vdev_t *first = vd->vdev_child[0]; | |
5861 | vdev_t *last = vd->vdev_child[vd->vdev_children - 1]; | |
5862 | ||
5863 | if (last->vdev_unspare) { | |
5864 | oldvd = first; | |
5865 | newvd = last; | |
5866 | } else if (first->vdev_unspare) { | |
5867 | oldvd = last; | |
5868 | newvd = first; | |
5869 | } else { | |
5870 | oldvd = NULL; | |
5871 | } | |
5872 | ||
5873 | if (oldvd != NULL && | |
5874 | vdev_dtl_empty(newvd, DTL_MISSING) && | |
5875 | vdev_dtl_empty(newvd, DTL_OUTAGE) && | |
5876 | !vdev_dtl_required(oldvd)) | |
5877 | return (oldvd); | |
5878 | ||
5879 | /* | |
5880 | * If there are more than two spares attached to a disk, | |
5881 | * and those spares are not required, then we want to | |
5882 | * attempt to free them up now so that they can be used | |
5883 | * by other pools. Once we're back down to a single | |
5884 | * disk+spare, we stop removing them. | |
5885 | */ | |
5886 | if (vd->vdev_children > 2) { | |
5887 | newvd = vd->vdev_child[1]; | |
5888 | ||
5889 | if (newvd->vdev_isspare && last->vdev_isspare && | |
5890 | vdev_dtl_empty(last, DTL_MISSING) && | |
5891 | vdev_dtl_empty(last, DTL_OUTAGE) && | |
5892 | !vdev_dtl_required(newvd)) | |
5893 | return (newvd); | |
5894 | } | |
5895 | } | |
5896 | ||
5897 | return (NULL); | |
5898 | } | |
5899 | ||
5900 | static void | |
5901 | spa_vdev_resilver_done(spa_t *spa) | |
5902 | { | |
5903 | vdev_t *vd, *pvd, *ppvd; | |
5904 | uint64_t guid, sguid, pguid, ppguid; | |
5905 | ||
5906 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
5907 | ||
5908 | while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) { | |
5909 | pvd = vd->vdev_parent; | |
5910 | ppvd = pvd->vdev_parent; | |
5911 | guid = vd->vdev_guid; | |
5912 | pguid = pvd->vdev_guid; | |
5913 | ppguid = ppvd->vdev_guid; | |
5914 | sguid = 0; | |
5915 | /* | |
5916 | * If we have just finished replacing a hot spared device, then | |
5917 | * we need to detach the parent's first child (the original hot | |
5918 | * spare) as well. | |
5919 | */ | |
5920 | if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 && | |
5921 | ppvd->vdev_children == 2) { | |
5922 | ASSERT(pvd->vdev_ops == &vdev_replacing_ops); | |
5923 | sguid = ppvd->vdev_child[1]->vdev_guid; | |
5924 | } | |
5925 | ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd)); | |
5926 | ||
5927 | spa_config_exit(spa, SCL_ALL, FTAG); | |
5928 | if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0) | |
5929 | return; | |
5930 | if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0) | |
5931 | return; | |
5932 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
5933 | } | |
5934 | ||
5935 | spa_config_exit(spa, SCL_ALL, FTAG); | |
5936 | } | |
5937 | ||
5938 | /* | |
5939 | * Update the stored path or FRU for this vdev. | |
5940 | */ | |
5941 | int | |
5942 | spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value, | |
5943 | boolean_t ispath) | |
5944 | { | |
5945 | vdev_t *vd; | |
5946 | boolean_t sync = B_FALSE; | |
5947 | ||
5948 | ASSERT(spa_writeable(spa)); | |
5949 | ||
5950 | spa_vdev_state_enter(spa, SCL_ALL); | |
5951 | ||
5952 | if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) | |
5953 | return (spa_vdev_state_exit(spa, NULL, ENOENT)); | |
5954 | ||
5955 | if (!vd->vdev_ops->vdev_op_leaf) | |
5956 | return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); | |
5957 | ||
5958 | if (ispath) { | |
5959 | if (strcmp(value, vd->vdev_path) != 0) { | |
5960 | spa_strfree(vd->vdev_path); | |
5961 | vd->vdev_path = spa_strdup(value); | |
5962 | sync = B_TRUE; | |
5963 | } | |
5964 | } else { | |
5965 | if (vd->vdev_fru == NULL) { | |
5966 | vd->vdev_fru = spa_strdup(value); | |
5967 | sync = B_TRUE; | |
5968 | } else if (strcmp(value, vd->vdev_fru) != 0) { | |
5969 | spa_strfree(vd->vdev_fru); | |
5970 | vd->vdev_fru = spa_strdup(value); | |
5971 | sync = B_TRUE; | |
5972 | } | |
5973 | } | |
5974 | ||
5975 | return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0)); | |
5976 | } | |
5977 | ||
5978 | int | |
5979 | spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath) | |
5980 | { | |
5981 | return (spa_vdev_set_common(spa, guid, newpath, B_TRUE)); | |
5982 | } | |
5983 | ||
5984 | int | |
5985 | spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru) | |
5986 | { | |
5987 | return (spa_vdev_set_common(spa, guid, newfru, B_FALSE)); | |
5988 | } | |
5989 | ||
5990 | /* | |
5991 | * ========================================================================== | |
5992 | * SPA Scanning | |
5993 | * ========================================================================== | |
5994 | */ | |
5995 | int | |
5996 | spa_scrub_pause_resume(spa_t *spa, pool_scrub_cmd_t cmd) | |
5997 | { | |
5998 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); | |
5999 | ||
6000 | if (dsl_scan_resilvering(spa->spa_dsl_pool)) | |
6001 | return (SET_ERROR(EBUSY)); | |
6002 | ||
6003 | return (dsl_scrub_set_pause_resume(spa->spa_dsl_pool, cmd)); | |
6004 | } | |
6005 | ||
6006 | int | |
6007 | spa_scan_stop(spa_t *spa) | |
6008 | { | |
6009 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); | |
6010 | if (dsl_scan_resilvering(spa->spa_dsl_pool)) | |
6011 | return (SET_ERROR(EBUSY)); | |
6012 | return (dsl_scan_cancel(spa->spa_dsl_pool)); | |
6013 | } | |
6014 | ||
6015 | int | |
6016 | spa_scan(spa_t *spa, pool_scan_func_t func) | |
6017 | { | |
6018 | ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); | |
6019 | ||
6020 | if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE) | |
6021 | return (SET_ERROR(ENOTSUP)); | |
6022 | ||
6023 | /* | |
6024 | * If a resilver was requested, but there is no DTL on a | |
6025 | * writeable leaf device, we have nothing to do. | |
6026 | */ | |
6027 | if (func == POOL_SCAN_RESILVER && | |
6028 | !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) { | |
6029 | spa_async_request(spa, SPA_ASYNC_RESILVER_DONE); | |
6030 | return (0); | |
6031 | } | |
6032 | ||
6033 | return (dsl_scan(spa->spa_dsl_pool, func)); | |
6034 | } | |
6035 | ||
6036 | /* | |
6037 | * ========================================================================== | |
6038 | * SPA async task processing | |
6039 | * ========================================================================== | |
6040 | */ | |
6041 | ||
6042 | static void | |
6043 | spa_async_remove(spa_t *spa, vdev_t *vd) | |
6044 | { | |
6045 | if (vd->vdev_remove_wanted) { | |
6046 | vd->vdev_remove_wanted = B_FALSE; | |
6047 | vd->vdev_delayed_close = B_FALSE; | |
6048 | vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE); | |
6049 | ||
6050 | /* | |
6051 | * We want to clear the stats, but we don't want to do a full | |
6052 | * vdev_clear() as that will cause us to throw away | |
6053 | * degraded/faulted state as well as attempt to reopen the | |
6054 | * device, all of which is a waste. | |
6055 | */ | |
6056 | vd->vdev_stat.vs_read_errors = 0; | |
6057 | vd->vdev_stat.vs_write_errors = 0; | |
6058 | vd->vdev_stat.vs_checksum_errors = 0; | |
6059 | ||
6060 | vdev_state_dirty(vd->vdev_top); | |
6061 | } | |
6062 | ||
6063 | for (int c = 0; c < vd->vdev_children; c++) | |
6064 | spa_async_remove(spa, vd->vdev_child[c]); | |
6065 | } | |
6066 | ||
6067 | static void | |
6068 | spa_async_probe(spa_t *spa, vdev_t *vd) | |
6069 | { | |
6070 | if (vd->vdev_probe_wanted) { | |
6071 | vd->vdev_probe_wanted = B_FALSE; | |
6072 | vdev_reopen(vd); /* vdev_open() does the actual probe */ | |
6073 | } | |
6074 | ||
6075 | for (int c = 0; c < vd->vdev_children; c++) | |
6076 | spa_async_probe(spa, vd->vdev_child[c]); | |
6077 | } | |
6078 | ||
6079 | static void | |
6080 | spa_async_autoexpand(spa_t *spa, vdev_t *vd) | |
6081 | { | |
6082 | if (!spa->spa_autoexpand) | |
6083 | return; | |
6084 | ||
6085 | for (int c = 0; c < vd->vdev_children; c++) { | |
6086 | vdev_t *cvd = vd->vdev_child[c]; | |
6087 | spa_async_autoexpand(spa, cvd); | |
6088 | } | |
6089 | ||
6090 | if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL) | |
6091 | return; | |
6092 | ||
6093 | spa_event_notify(vd->vdev_spa, vd, NULL, ESC_ZFS_VDEV_AUTOEXPAND); | |
6094 | } | |
6095 | ||
6096 | static void | |
6097 | spa_async_thread(void *arg) | |
6098 | { | |
6099 | spa_t *spa = (spa_t *)arg; | |
6100 | int tasks; | |
6101 | ||
6102 | ASSERT(spa->spa_sync_on); | |
6103 | ||
6104 | mutex_enter(&spa->spa_async_lock); | |
6105 | tasks = spa->spa_async_tasks; | |
6106 | spa->spa_async_tasks = 0; | |
6107 | mutex_exit(&spa->spa_async_lock); | |
6108 | ||
6109 | /* | |
6110 | * See if the config needs to be updated. | |
6111 | */ | |
6112 | if (tasks & SPA_ASYNC_CONFIG_UPDATE) { | |
6113 | uint64_t old_space, new_space; | |
6114 | ||
6115 | mutex_enter(&spa_namespace_lock); | |
6116 | old_space = metaslab_class_get_space(spa_normal_class(spa)); | |
6117 | spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); | |
6118 | new_space = metaslab_class_get_space(spa_normal_class(spa)); | |
6119 | mutex_exit(&spa_namespace_lock); | |
6120 | ||
6121 | /* | |
6122 | * If the pool grew as a result of the config update, | |
6123 | * then log an internal history event. | |
6124 | */ | |
6125 | if (new_space != old_space) { | |
6126 | spa_history_log_internal(spa, "vdev online", NULL, | |
6127 | "pool '%s' size: %llu(+%llu)", | |
6128 | spa_name(spa), new_space, new_space - old_space); | |
6129 | } | |
6130 | } | |
6131 | ||
6132 | /* | |
6133 | * See if any devices need to be marked REMOVED. | |
6134 | */ | |
6135 | if (tasks & SPA_ASYNC_REMOVE) { | |
6136 | spa_vdev_state_enter(spa, SCL_NONE); | |
6137 | spa_async_remove(spa, spa->spa_root_vdev); | |
6138 | for (int i = 0; i < spa->spa_l2cache.sav_count; i++) | |
6139 | spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]); | |
6140 | for (int i = 0; i < spa->spa_spares.sav_count; i++) | |
6141 | spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]); | |
6142 | (void) spa_vdev_state_exit(spa, NULL, 0); | |
6143 | } | |
6144 | ||
6145 | if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) { | |
6146 | spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); | |
6147 | spa_async_autoexpand(spa, spa->spa_root_vdev); | |
6148 | spa_config_exit(spa, SCL_CONFIG, FTAG); | |
6149 | } | |
6150 | ||
6151 | /* | |
6152 | * See if any devices need to be probed. | |
6153 | */ | |
6154 | if (tasks & SPA_ASYNC_PROBE) { | |
6155 | spa_vdev_state_enter(spa, SCL_NONE); | |
6156 | spa_async_probe(spa, spa->spa_root_vdev); | |
6157 | (void) spa_vdev_state_exit(spa, NULL, 0); | |
6158 | } | |
6159 | ||
6160 | /* | |
6161 | * If any devices are done replacing, detach them. | |
6162 | */ | |
6163 | if (tasks & SPA_ASYNC_RESILVER_DONE) | |
6164 | spa_vdev_resilver_done(spa); | |
6165 | ||
6166 | /* | |
6167 | * Kick off a resilver. | |
6168 | */ | |
6169 | if (tasks & SPA_ASYNC_RESILVER) | |
6170 | dsl_resilver_restart(spa->spa_dsl_pool, 0); | |
6171 | ||
6172 | /* | |
6173 | * Let the world know that we're done. | |
6174 | */ | |
6175 | mutex_enter(&spa->spa_async_lock); | |
6176 | spa->spa_async_thread = NULL; | |
6177 | cv_broadcast(&spa->spa_async_cv); | |
6178 | mutex_exit(&spa->spa_async_lock); | |
6179 | thread_exit(); | |
6180 | } | |
6181 | ||
6182 | void | |
6183 | spa_async_suspend(spa_t *spa) | |
6184 | { | |
6185 | mutex_enter(&spa->spa_async_lock); | |
6186 | spa->spa_async_suspended++; | |
6187 | while (spa->spa_async_thread != NULL) | |
6188 | cv_wait(&spa->spa_async_cv, &spa->spa_async_lock); | |
6189 | mutex_exit(&spa->spa_async_lock); | |
6190 | } | |
6191 | ||
6192 | void | |
6193 | spa_async_resume(spa_t *spa) | |
6194 | { | |
6195 | mutex_enter(&spa->spa_async_lock); | |
6196 | ASSERT(spa->spa_async_suspended != 0); | |
6197 | spa->spa_async_suspended--; | |
6198 | mutex_exit(&spa->spa_async_lock); | |
6199 | } | |
6200 | ||
6201 | static boolean_t | |
6202 | spa_async_tasks_pending(spa_t *spa) | |
6203 | { | |
6204 | uint_t non_config_tasks; | |
6205 | uint_t config_task; | |
6206 | boolean_t config_task_suspended; | |
6207 | ||
6208 | non_config_tasks = spa->spa_async_tasks & ~SPA_ASYNC_CONFIG_UPDATE; | |
6209 | config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE; | |
6210 | if (spa->spa_ccw_fail_time == 0) { | |
6211 | config_task_suspended = B_FALSE; | |
6212 | } else { | |
6213 | config_task_suspended = | |
6214 | (gethrtime() - spa->spa_ccw_fail_time) < | |
6215 | ((hrtime_t)zfs_ccw_retry_interval * NANOSEC); | |
6216 | } | |
6217 | ||
6218 | return (non_config_tasks || (config_task && !config_task_suspended)); | |
6219 | } | |
6220 | ||
6221 | static void | |
6222 | spa_async_dispatch(spa_t *spa) | |
6223 | { | |
6224 | mutex_enter(&spa->spa_async_lock); | |
6225 | if (spa_async_tasks_pending(spa) && | |
6226 | !spa->spa_async_suspended && | |
6227 | spa->spa_async_thread == NULL && | |
6228 | rootdir != NULL) | |
6229 | spa->spa_async_thread = thread_create(NULL, 0, | |
6230 | spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri); | |
6231 | mutex_exit(&spa->spa_async_lock); | |
6232 | } | |
6233 | ||
6234 | void | |
6235 | spa_async_request(spa_t *spa, int task) | |
6236 | { | |
6237 | zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task); | |
6238 | mutex_enter(&spa->spa_async_lock); | |
6239 | spa->spa_async_tasks |= task; | |
6240 | mutex_exit(&spa->spa_async_lock); | |
6241 | } | |
6242 | ||
6243 | /* | |
6244 | * ========================================================================== | |
6245 | * SPA syncing routines | |
6246 | * ========================================================================== | |
6247 | */ | |
6248 | ||
6249 | static int | |
6250 | bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) | |
6251 | { | |
6252 | bpobj_t *bpo = arg; | |
6253 | bpobj_enqueue(bpo, bp, tx); | |
6254 | return (0); | |
6255 | } | |
6256 | ||
6257 | static int | |
6258 | spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) | |
6259 | { | |
6260 | zio_t *zio = arg; | |
6261 | ||
6262 | zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp, | |
6263 | zio->io_flags)); | |
6264 | return (0); | |
6265 | } | |
6266 | ||
6267 | /* | |
6268 | * Note: this simple function is not inlined to make it easier to dtrace the | |
6269 | * amount of time spent syncing frees. | |
6270 | */ | |
6271 | static void | |
6272 | spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx) | |
6273 | { | |
6274 | zio_t *zio = zio_root(spa, NULL, NULL, 0); | |
6275 | bplist_iterate(bpl, spa_free_sync_cb, zio, tx); | |
6276 | VERIFY(zio_wait(zio) == 0); | |
6277 | } | |
6278 | ||
6279 | /* | |
6280 | * Note: this simple function is not inlined to make it easier to dtrace the | |
6281 | * amount of time spent syncing deferred frees. | |
6282 | */ | |
6283 | static void | |
6284 | spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx) | |
6285 | { | |
6286 | zio_t *zio = zio_root(spa, NULL, NULL, 0); | |
6287 | VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj, | |
6288 | spa_free_sync_cb, zio, tx), ==, 0); | |
6289 | VERIFY0(zio_wait(zio)); | |
6290 | } | |
6291 | ||
6292 | static void | |
6293 | spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx) | |
6294 | { | |
6295 | char *packed = NULL; | |
6296 | size_t bufsize; | |
6297 | size_t nvsize = 0; | |
6298 | dmu_buf_t *db; | |
6299 | ||
6300 | VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0); | |
6301 | ||
6302 | /* | |
6303 | * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration | |
6304 | * information. This avoids the dmu_buf_will_dirty() path and | |
6305 | * saves us a pre-read to get data we don't actually care about. | |
6306 | */ | |
6307 | bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE); | |
6308 | packed = vmem_alloc(bufsize, KM_SLEEP); | |
6309 | ||
6310 | VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR, | |
6311 | KM_SLEEP) == 0); | |
6312 | bzero(packed + nvsize, bufsize - nvsize); | |
6313 | ||
6314 | dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx); | |
6315 | ||
6316 | vmem_free(packed, bufsize); | |
6317 | ||
6318 | VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); | |
6319 | dmu_buf_will_dirty(db, tx); | |
6320 | *(uint64_t *)db->db_data = nvsize; | |
6321 | dmu_buf_rele(db, FTAG); | |
6322 | } | |
6323 | ||
6324 | static void | |
6325 | spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx, | |
6326 | const char *config, const char *entry) | |
6327 | { | |
6328 | nvlist_t *nvroot; | |
6329 | nvlist_t **list; | |
6330 | int i; | |
6331 | ||
6332 | if (!sav->sav_sync) | |
6333 | return; | |
6334 | ||
6335 | /* | |
6336 | * Update the MOS nvlist describing the list of available devices. | |
6337 | * spa_validate_aux() will have already made sure this nvlist is | |
6338 | * valid and the vdevs are labeled appropriately. | |
6339 | */ | |
6340 | if (sav->sav_object == 0) { | |
6341 | sav->sav_object = dmu_object_alloc(spa->spa_meta_objset, | |
6342 | DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE, | |
6343 | sizeof (uint64_t), tx); | |
6344 | VERIFY(zap_update(spa->spa_meta_objset, | |
6345 | DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1, | |
6346 | &sav->sav_object, tx) == 0); | |
6347 | } | |
6348 | ||
6349 | VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); | |
6350 | if (sav->sav_count == 0) { | |
6351 | VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0); | |
6352 | } else { | |
6353 | list = kmem_alloc(sav->sav_count*sizeof (void *), KM_SLEEP); | |
6354 | for (i = 0; i < sav->sav_count; i++) | |
6355 | list[i] = vdev_config_generate(spa, sav->sav_vdevs[i], | |
6356 | B_FALSE, VDEV_CONFIG_L2CACHE); | |
6357 | VERIFY(nvlist_add_nvlist_array(nvroot, config, list, | |
6358 | sav->sav_count) == 0); | |
6359 | for (i = 0; i < sav->sav_count; i++) | |
6360 | nvlist_free(list[i]); | |
6361 | kmem_free(list, sav->sav_count * sizeof (void *)); | |
6362 | } | |
6363 | ||
6364 | spa_sync_nvlist(spa, sav->sav_object, nvroot, tx); | |
6365 | nvlist_free(nvroot); | |
6366 | ||
6367 | sav->sav_sync = B_FALSE; | |
6368 | } | |
6369 | ||
6370 | /* | |
6371 | * Rebuild spa's all-vdev ZAP from the vdev ZAPs indicated in each vdev_t. | |
6372 | * The all-vdev ZAP must be empty. | |
6373 | */ | |
6374 | static void | |
6375 | spa_avz_build(vdev_t *vd, uint64_t avz, dmu_tx_t *tx) | |
6376 | { | |
6377 | spa_t *spa = vd->vdev_spa; | |
6378 | ||
6379 | if (vd->vdev_top_zap != 0) { | |
6380 | VERIFY0(zap_add_int(spa->spa_meta_objset, avz, | |
6381 | vd->vdev_top_zap, tx)); | |
6382 | } | |
6383 | if (vd->vdev_leaf_zap != 0) { | |
6384 | VERIFY0(zap_add_int(spa->spa_meta_objset, avz, | |
6385 | vd->vdev_leaf_zap, tx)); | |
6386 | } | |
6387 | for (uint64_t i = 0; i < vd->vdev_children; i++) { | |
6388 | spa_avz_build(vd->vdev_child[i], avz, tx); | |
6389 | } | |
6390 | } | |
6391 | ||
6392 | static void | |
6393 | spa_sync_config_object(spa_t *spa, dmu_tx_t *tx) | |
6394 | { | |
6395 | nvlist_t *config; | |
6396 | ||
6397 | /* | |
6398 | * If the pool is being imported from a pre-per-vdev-ZAP version of ZFS, | |
6399 | * its config may not be dirty but we still need to build per-vdev ZAPs. | |
6400 | * Similarly, if the pool is being assembled (e.g. after a split), we | |
6401 | * need to rebuild the AVZ although the config may not be dirty. | |
6402 | */ | |
6403 | if (list_is_empty(&spa->spa_config_dirty_list) && | |
6404 | spa->spa_avz_action == AVZ_ACTION_NONE) | |
6405 | return; | |
6406 | ||
6407 | spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); | |
6408 | ||
6409 | ASSERT(spa->spa_avz_action == AVZ_ACTION_NONE || | |
6410 | spa->spa_avz_action == AVZ_ACTION_INITIALIZE || | |
6411 | spa->spa_all_vdev_zaps != 0); | |
6412 | ||
6413 | if (spa->spa_avz_action == AVZ_ACTION_REBUILD) { | |
6414 | /* Make and build the new AVZ */ | |
6415 | uint64_t new_avz = zap_create(spa->spa_meta_objset, | |
6416 | DMU_OTN_ZAP_METADATA, DMU_OT_NONE, 0, tx); | |
6417 | spa_avz_build(spa->spa_root_vdev, new_avz, tx); | |
6418 | ||
6419 | /* Diff old AVZ with new one */ | |
6420 | zap_cursor_t zc; | |
6421 | zap_attribute_t za; | |
6422 | ||
6423 | for (zap_cursor_init(&zc, spa->spa_meta_objset, | |
6424 | spa->spa_all_vdev_zaps); | |
6425 | zap_cursor_retrieve(&zc, &za) == 0; | |
6426 | zap_cursor_advance(&zc)) { | |
6427 | uint64_t vdzap = za.za_first_integer; | |
6428 | if (zap_lookup_int(spa->spa_meta_objset, new_avz, | |
6429 | vdzap) == ENOENT) { | |
6430 | /* | |
6431 | * ZAP is listed in old AVZ but not in new one; | |
6432 | * destroy it | |
6433 | */ | |
6434 | VERIFY0(zap_destroy(spa->spa_meta_objset, vdzap, | |
6435 | tx)); | |
6436 | } | |
6437 | } | |
6438 | ||
6439 | zap_cursor_fini(&zc); | |
6440 | ||
6441 | /* Destroy the old AVZ */ | |
6442 | VERIFY0(zap_destroy(spa->spa_meta_objset, | |
6443 | spa->spa_all_vdev_zaps, tx)); | |
6444 | ||
6445 | /* Replace the old AVZ in the dir obj with the new one */ | |
6446 | VERIFY0(zap_update(spa->spa_meta_objset, | |
6447 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP, | |
6448 | sizeof (new_avz), 1, &new_avz, tx)); | |
6449 | ||
6450 | spa->spa_all_vdev_zaps = new_avz; | |
6451 | } else if (spa->spa_avz_action == AVZ_ACTION_DESTROY) { | |
6452 | zap_cursor_t zc; | |
6453 | zap_attribute_t za; | |
6454 | ||
6455 | /* Walk through the AVZ and destroy all listed ZAPs */ | |
6456 | for (zap_cursor_init(&zc, spa->spa_meta_objset, | |
6457 | spa->spa_all_vdev_zaps); | |
6458 | zap_cursor_retrieve(&zc, &za) == 0; | |
6459 | zap_cursor_advance(&zc)) { | |
6460 | uint64_t zap = za.za_first_integer; | |
6461 | VERIFY0(zap_destroy(spa->spa_meta_objset, zap, tx)); | |
6462 | } | |
6463 | ||
6464 | zap_cursor_fini(&zc); | |
6465 | ||
6466 | /* Destroy and unlink the AVZ itself */ | |
6467 | VERIFY0(zap_destroy(spa->spa_meta_objset, | |
6468 | spa->spa_all_vdev_zaps, tx)); | |
6469 | VERIFY0(zap_remove(spa->spa_meta_objset, | |
6470 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP, tx)); | |
6471 | spa->spa_all_vdev_zaps = 0; | |
6472 | } | |
6473 | ||
6474 | if (spa->spa_all_vdev_zaps == 0) { | |
6475 | spa->spa_all_vdev_zaps = zap_create_link(spa->spa_meta_objset, | |
6476 | DMU_OTN_ZAP_METADATA, DMU_POOL_DIRECTORY_OBJECT, | |
6477 | DMU_POOL_VDEV_ZAP_MAP, tx); | |
6478 | } | |
6479 | spa->spa_avz_action = AVZ_ACTION_NONE; | |
6480 | ||
6481 | /* Create ZAPs for vdevs that don't have them. */ | |
6482 | vdev_construct_zaps(spa->spa_root_vdev, tx); | |
6483 | ||
6484 | config = spa_config_generate(spa, spa->spa_root_vdev, | |
6485 | dmu_tx_get_txg(tx), B_FALSE); | |
6486 | ||
6487 | /* | |
6488 | * If we're upgrading the spa version then make sure that | |
6489 | * the config object gets updated with the correct version. | |
6490 | */ | |
6491 | if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version) | |
6492 | fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, | |
6493 | spa->spa_uberblock.ub_version); | |
6494 | ||
6495 | spa_config_exit(spa, SCL_STATE, FTAG); | |
6496 | ||
6497 | nvlist_free(spa->spa_config_syncing); | |
6498 | spa->spa_config_syncing = config; | |
6499 | ||
6500 | spa_sync_nvlist(spa, spa->spa_config_object, config, tx); | |
6501 | } | |
6502 | ||
6503 | static void | |
6504 | spa_sync_version(void *arg, dmu_tx_t *tx) | |
6505 | { | |
6506 | uint64_t *versionp = arg; | |
6507 | uint64_t version = *versionp; | |
6508 | spa_t *spa = dmu_tx_pool(tx)->dp_spa; | |
6509 | ||
6510 | /* | |
6511 | * Setting the version is special cased when first creating the pool. | |
6512 | */ | |
6513 | ASSERT(tx->tx_txg != TXG_INITIAL); | |
6514 | ||
6515 | ASSERT(SPA_VERSION_IS_SUPPORTED(version)); | |
6516 | ASSERT(version >= spa_version(spa)); | |
6517 | ||
6518 | spa->spa_uberblock.ub_version = version; | |
6519 | vdev_config_dirty(spa->spa_root_vdev); | |
6520 | spa_history_log_internal(spa, "set", tx, "version=%lld", version); | |
6521 | } | |
6522 | ||
6523 | /* | |
6524 | * Set zpool properties. | |
6525 | */ | |
6526 | static void | |
6527 | spa_sync_props(void *arg, dmu_tx_t *tx) | |
6528 | { | |
6529 | nvlist_t *nvp = arg; | |
6530 | spa_t *spa = dmu_tx_pool(tx)->dp_spa; | |
6531 | objset_t *mos = spa->spa_meta_objset; | |
6532 | nvpair_t *elem = NULL; | |
6533 | ||
6534 | mutex_enter(&spa->spa_props_lock); | |
6535 | ||
6536 | while ((elem = nvlist_next_nvpair(nvp, elem))) { | |
6537 | uint64_t intval; | |
6538 | char *strval, *fname; | |
6539 | zpool_prop_t prop; | |
6540 | const char *propname; | |
6541 | zprop_type_t proptype; | |
6542 | spa_feature_t fid; | |
6543 | ||
6544 | prop = zpool_name_to_prop(nvpair_name(elem)); | |
6545 | switch ((int)prop) { | |
6546 | case ZPROP_INVAL: | |
6547 | /* | |
6548 | * We checked this earlier in spa_prop_validate(). | |
6549 | */ | |
6550 | ASSERT(zpool_prop_feature(nvpair_name(elem))); | |
6551 | ||
6552 | fname = strchr(nvpair_name(elem), '@') + 1; | |
6553 | VERIFY0(zfeature_lookup_name(fname, &fid)); | |
6554 | ||
6555 | spa_feature_enable(spa, fid, tx); | |
6556 | spa_history_log_internal(spa, "set", tx, | |
6557 | "%s=enabled", nvpair_name(elem)); | |
6558 | break; | |
6559 | ||
6560 | case ZPOOL_PROP_VERSION: | |
6561 | intval = fnvpair_value_uint64(elem); | |
6562 | /* | |
6563 | * The version is synced separately before other | |
6564 | * properties and should be correct by now. | |
6565 | */ | |
6566 | ASSERT3U(spa_version(spa), >=, intval); | |
6567 | break; | |
6568 | ||
6569 | case ZPOOL_PROP_ALTROOT: | |
6570 | /* | |
6571 | * 'altroot' is a non-persistent property. It should | |
6572 | * have been set temporarily at creation or import time. | |
6573 | */ | |
6574 | ASSERT(spa->spa_root != NULL); | |
6575 | break; | |
6576 | ||
6577 | case ZPOOL_PROP_READONLY: | |
6578 | case ZPOOL_PROP_CACHEFILE: | |
6579 | /* | |
6580 | * 'readonly' and 'cachefile' are also non-persisitent | |
6581 | * properties. | |
6582 | */ | |
6583 | break; | |
6584 | case ZPOOL_PROP_COMMENT: | |
6585 | strval = fnvpair_value_string(elem); | |
6586 | if (spa->spa_comment != NULL) | |
6587 | spa_strfree(spa->spa_comment); | |
6588 | spa->spa_comment = spa_strdup(strval); | |
6589 | /* | |
6590 | * We need to dirty the configuration on all the vdevs | |
6591 | * so that their labels get updated. It's unnecessary | |
6592 | * to do this for pool creation since the vdev's | |
6593 | * configuration has already been dirtied. | |
6594 | */ | |
6595 | if (tx->tx_txg != TXG_INITIAL) | |
6596 | vdev_config_dirty(spa->spa_root_vdev); | |
6597 | spa_history_log_internal(spa, "set", tx, | |
6598 | "%s=%s", nvpair_name(elem), strval); | |
6599 | break; | |
6600 | default: | |
6601 | /* | |
6602 | * Set pool property values in the poolprops mos object. | |
6603 | */ | |
6604 | if (spa->spa_pool_props_object == 0) { | |
6605 | spa->spa_pool_props_object = | |
6606 | zap_create_link(mos, DMU_OT_POOL_PROPS, | |
6607 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS, | |
6608 | tx); | |
6609 | } | |
6610 | ||
6611 | /* normalize the property name */ | |
6612 | propname = zpool_prop_to_name(prop); | |
6613 | proptype = zpool_prop_get_type(prop); | |
6614 | ||
6615 | if (nvpair_type(elem) == DATA_TYPE_STRING) { | |
6616 | ASSERT(proptype == PROP_TYPE_STRING); | |
6617 | strval = fnvpair_value_string(elem); | |
6618 | VERIFY0(zap_update(mos, | |
6619 | spa->spa_pool_props_object, propname, | |
6620 | 1, strlen(strval) + 1, strval, tx)); | |
6621 | spa_history_log_internal(spa, "set", tx, | |
6622 | "%s=%s", nvpair_name(elem), strval); | |
6623 | } else if (nvpair_type(elem) == DATA_TYPE_UINT64) { | |
6624 | intval = fnvpair_value_uint64(elem); | |
6625 | ||
6626 | if (proptype == PROP_TYPE_INDEX) { | |
6627 | const char *unused; | |
6628 | VERIFY0(zpool_prop_index_to_string( | |
6629 | prop, intval, &unused)); | |
6630 | } | |
6631 | VERIFY0(zap_update(mos, | |
6632 | spa->spa_pool_props_object, propname, | |
6633 | 8, 1, &intval, tx)); | |
6634 | spa_history_log_internal(spa, "set", tx, | |
6635 | "%s=%lld", nvpair_name(elem), intval); | |
6636 | } else { | |
6637 | ASSERT(0); /* not allowed */ | |
6638 | } | |
6639 | ||
6640 | switch (prop) { | |
6641 | case ZPOOL_PROP_DELEGATION: | |
6642 | spa->spa_delegation = intval; | |
6643 | break; | |
6644 | case ZPOOL_PROP_BOOTFS: | |
6645 | spa->spa_bootfs = intval; | |
6646 | break; | |
6647 | case ZPOOL_PROP_FAILUREMODE: | |
6648 | spa->spa_failmode = intval; | |
6649 | break; | |
6650 | case ZPOOL_PROP_AUTOEXPAND: | |
6651 | spa->spa_autoexpand = intval; | |
6652 | if (tx->tx_txg != TXG_INITIAL) | |
6653 | spa_async_request(spa, | |
6654 | SPA_ASYNC_AUTOEXPAND); | |
6655 | break; | |
6656 | case ZPOOL_PROP_MULTIHOST: | |
6657 | spa->spa_multihost = intval; | |
6658 | break; | |
6659 | case ZPOOL_PROP_DEDUPDITTO: | |
6660 | spa->spa_dedup_ditto = intval; | |
6661 | break; | |
6662 | default: | |
6663 | break; | |
6664 | } | |
6665 | } | |
6666 | ||
6667 | } | |
6668 | ||
6669 | mutex_exit(&spa->spa_props_lock); | |
6670 | } | |
6671 | ||
6672 | /* | |
6673 | * Perform one-time upgrade on-disk changes. spa_version() does not | |
6674 | * reflect the new version this txg, so there must be no changes this | |
6675 | * txg to anything that the upgrade code depends on after it executes. | |
6676 | * Therefore this must be called after dsl_pool_sync() does the sync | |
6677 | * tasks. | |
6678 | */ | |
6679 | static void | |
6680 | spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx) | |
6681 | { | |
6682 | dsl_pool_t *dp = spa->spa_dsl_pool; | |
6683 | ||
6684 | ASSERT(spa->spa_sync_pass == 1); | |
6685 | ||
6686 | rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); | |
6687 | ||
6688 | if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN && | |
6689 | spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) { | |
6690 | dsl_pool_create_origin(dp, tx); | |
6691 | ||
6692 | /* Keeping the origin open increases spa_minref */ | |
6693 | spa->spa_minref += 3; | |
6694 | } | |
6695 | ||
6696 | if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES && | |
6697 | spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) { | |
6698 | dsl_pool_upgrade_clones(dp, tx); | |
6699 | } | |
6700 | ||
6701 | if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES && | |
6702 | spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) { | |
6703 | dsl_pool_upgrade_dir_clones(dp, tx); | |
6704 | ||
6705 | /* Keeping the freedir open increases spa_minref */ | |
6706 | spa->spa_minref += 3; | |
6707 | } | |
6708 | ||
6709 | if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES && | |
6710 | spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { | |
6711 | spa_feature_create_zap_objects(spa, tx); | |
6712 | } | |
6713 | ||
6714 | /* | |
6715 | * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable | |
6716 | * when possibility to use lz4 compression for metadata was added | |
6717 | * Old pools that have this feature enabled must be upgraded to have | |
6718 | * this feature active | |
6719 | */ | |
6720 | if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { | |
6721 | boolean_t lz4_en = spa_feature_is_enabled(spa, | |
6722 | SPA_FEATURE_LZ4_COMPRESS); | |
6723 | boolean_t lz4_ac = spa_feature_is_active(spa, | |
6724 | SPA_FEATURE_LZ4_COMPRESS); | |
6725 | ||
6726 | if (lz4_en && !lz4_ac) | |
6727 | spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx); | |
6728 | } | |
6729 | ||
6730 | /* | |
6731 | * If we haven't written the salt, do so now. Note that the | |
6732 | * feature may not be activated yet, but that's fine since | |
6733 | * the presence of this ZAP entry is backwards compatible. | |
6734 | */ | |
6735 | if (zap_contains(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, | |
6736 | DMU_POOL_CHECKSUM_SALT) == ENOENT) { | |
6737 | VERIFY0(zap_add(spa->spa_meta_objset, | |
6738 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CHECKSUM_SALT, 1, | |
6739 | sizeof (spa->spa_cksum_salt.zcs_bytes), | |
6740 | spa->spa_cksum_salt.zcs_bytes, tx)); | |
6741 | } | |
6742 | ||
6743 | rrw_exit(&dp->dp_config_rwlock, FTAG); | |
6744 | } | |
6745 | ||
6746 | /* | |
6747 | * Sync the specified transaction group. New blocks may be dirtied as | |
6748 | * part of the process, so we iterate until it converges. | |
6749 | */ | |
6750 | void | |
6751 | spa_sync(spa_t *spa, uint64_t txg) | |
6752 | { | |
6753 | dsl_pool_t *dp = spa->spa_dsl_pool; | |
6754 | objset_t *mos = spa->spa_meta_objset; | |
6755 | bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK]; | |
6756 | vdev_t *rvd = spa->spa_root_vdev; | |
6757 | vdev_t *vd; | |
6758 | dmu_tx_t *tx; | |
6759 | int error; | |
6760 | uint32_t max_queue_depth = zfs_vdev_async_write_max_active * | |
6761 | zfs_vdev_queue_depth_pct / 100; | |
6762 | ||
6763 | VERIFY(spa_writeable(spa)); | |
6764 | ||
6765 | /* | |
6766 | * Lock out configuration changes. | |
6767 | */ | |
6768 | spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); | |
6769 | ||
6770 | spa->spa_syncing_txg = txg; | |
6771 | spa->spa_sync_pass = 0; | |
6772 | ||
6773 | mutex_enter(&spa->spa_alloc_lock); | |
6774 | VERIFY0(avl_numnodes(&spa->spa_alloc_tree)); | |
6775 | mutex_exit(&spa->spa_alloc_lock); | |
6776 | ||
6777 | /* | |
6778 | * If there are any pending vdev state changes, convert them | |
6779 | * into config changes that go out with this transaction group. | |
6780 | */ | |
6781 | spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); | |
6782 | while (list_head(&spa->spa_state_dirty_list) != NULL) { | |
6783 | /* | |
6784 | * We need the write lock here because, for aux vdevs, | |
6785 | * calling vdev_config_dirty() modifies sav_config. | |
6786 | * This is ugly and will become unnecessary when we | |
6787 | * eliminate the aux vdev wart by integrating all vdevs | |
6788 | * into the root vdev tree. | |
6789 | */ | |
6790 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
6791 | spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER); | |
6792 | while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) { | |
6793 | vdev_state_clean(vd); | |
6794 | vdev_config_dirty(vd); | |
6795 | } | |
6796 | spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); | |
6797 | spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); | |
6798 | } | |
6799 | spa_config_exit(spa, SCL_STATE, FTAG); | |
6800 | ||
6801 | tx = dmu_tx_create_assigned(dp, txg); | |
6802 | ||
6803 | spa->spa_sync_starttime = gethrtime(); | |
6804 | taskq_cancel_id(system_delay_taskq, spa->spa_deadman_tqid); | |
6805 | spa->spa_deadman_tqid = taskq_dispatch_delay(system_delay_taskq, | |
6806 | spa_deadman, spa, TQ_SLEEP, ddi_get_lbolt() + | |
6807 | NSEC_TO_TICK(spa->spa_deadman_synctime)); | |
6808 | ||
6809 | /* | |
6810 | * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg, | |
6811 | * set spa_deflate if we have no raid-z vdevs. | |
6812 | */ | |
6813 | if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE && | |
6814 | spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) { | |
6815 | int i; | |
6816 | ||
6817 | for (i = 0; i < rvd->vdev_children; i++) { | |
6818 | vd = rvd->vdev_child[i]; | |
6819 | if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE) | |
6820 | break; | |
6821 | } | |
6822 | if (i == rvd->vdev_children) { | |
6823 | spa->spa_deflate = TRUE; | |
6824 | VERIFY(0 == zap_add(spa->spa_meta_objset, | |
6825 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, | |
6826 | sizeof (uint64_t), 1, &spa->spa_deflate, tx)); | |
6827 | } | |
6828 | } | |
6829 | ||
6830 | /* | |
6831 | * Set the top-level vdev's max queue depth. Evaluate each | |
6832 | * top-level's async write queue depth in case it changed. | |
6833 | * The max queue depth will not change in the middle of syncing | |
6834 | * out this txg. | |
6835 | */ | |
6836 | uint64_t queue_depth_total = 0; | |
6837 | for (int c = 0; c < rvd->vdev_children; c++) { | |
6838 | vdev_t *tvd = rvd->vdev_child[c]; | |
6839 | metaslab_group_t *mg = tvd->vdev_mg; | |
6840 | ||
6841 | if (mg == NULL || mg->mg_class != spa_normal_class(spa) || | |
6842 | !metaslab_group_initialized(mg)) | |
6843 | continue; | |
6844 | ||
6845 | /* | |
6846 | * It is safe to do a lock-free check here because only async | |
6847 | * allocations look at mg_max_alloc_queue_depth, and async | |
6848 | * allocations all happen from spa_sync(). | |
6849 | */ | |
6850 | ASSERT0(refcount_count(&mg->mg_alloc_queue_depth)); | |
6851 | mg->mg_max_alloc_queue_depth = max_queue_depth; | |
6852 | queue_depth_total += mg->mg_max_alloc_queue_depth; | |
6853 | } | |
6854 | metaslab_class_t *mc = spa_normal_class(spa); | |
6855 | ASSERT0(refcount_count(&mc->mc_alloc_slots)); | |
6856 | mc->mc_alloc_max_slots = queue_depth_total; | |
6857 | mc->mc_alloc_throttle_enabled = zio_dva_throttle_enabled; | |
6858 | ||
6859 | ASSERT3U(mc->mc_alloc_max_slots, <=, | |
6860 | max_queue_depth * rvd->vdev_children); | |
6861 | ||
6862 | /* | |
6863 | * Iterate to convergence. | |
6864 | */ | |
6865 | do { | |
6866 | int pass = ++spa->spa_sync_pass; | |
6867 | ||
6868 | spa_sync_config_object(spa, tx); | |
6869 | spa_sync_aux_dev(spa, &spa->spa_spares, tx, | |
6870 | ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES); | |
6871 | spa_sync_aux_dev(spa, &spa->spa_l2cache, tx, | |
6872 | ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE); | |
6873 | spa_errlog_sync(spa, txg); | |
6874 | dsl_pool_sync(dp, txg); | |
6875 | ||
6876 | if (pass < zfs_sync_pass_deferred_free) { | |
6877 | spa_sync_frees(spa, free_bpl, tx); | |
6878 | } else { | |
6879 | /* | |
6880 | * We can not defer frees in pass 1, because | |
6881 | * we sync the deferred frees later in pass 1. | |
6882 | */ | |
6883 | ASSERT3U(pass, >, 1); | |
6884 | bplist_iterate(free_bpl, bpobj_enqueue_cb, | |
6885 | &spa->spa_deferred_bpobj, tx); | |
6886 | } | |
6887 | ||
6888 | ddt_sync(spa, txg); | |
6889 | dsl_scan_sync(dp, tx); | |
6890 | ||
6891 | while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))) | |
6892 | vdev_sync(vd, txg); | |
6893 | ||
6894 | if (pass == 1) { | |
6895 | spa_sync_upgrades(spa, tx); | |
6896 | ASSERT3U(txg, >=, | |
6897 | spa->spa_uberblock.ub_rootbp.blk_birth); | |
6898 | /* | |
6899 | * Note: We need to check if the MOS is dirty | |
6900 | * because we could have marked the MOS dirty | |
6901 | * without updating the uberblock (e.g. if we | |
6902 | * have sync tasks but no dirty user data). We | |
6903 | * need to check the uberblock's rootbp because | |
6904 | * it is updated if we have synced out dirty | |
6905 | * data (though in this case the MOS will most | |
6906 | * likely also be dirty due to second order | |
6907 | * effects, we don't want to rely on that here). | |
6908 | */ | |
6909 | if (spa->spa_uberblock.ub_rootbp.blk_birth < txg && | |
6910 | !dmu_objset_is_dirty(mos, txg)) { | |
6911 | /* | |
6912 | * Nothing changed on the first pass, | |
6913 | * therefore this TXG is a no-op. Avoid | |
6914 | * syncing deferred frees, so that we | |
6915 | * can keep this TXG as a no-op. | |
6916 | */ | |
6917 | ASSERT(txg_list_empty(&dp->dp_dirty_datasets, | |
6918 | txg)); | |
6919 | ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); | |
6920 | ASSERT(txg_list_empty(&dp->dp_sync_tasks, txg)); | |
6921 | break; | |
6922 | } | |
6923 | spa_sync_deferred_frees(spa, tx); | |
6924 | } | |
6925 | ||
6926 | } while (dmu_objset_is_dirty(mos, txg)); | |
6927 | ||
6928 | #ifdef ZFS_DEBUG | |
6929 | if (!list_is_empty(&spa->spa_config_dirty_list)) { | |
6930 | /* | |
6931 | * Make sure that the number of ZAPs for all the vdevs matches | |
6932 | * the number of ZAPs in the per-vdev ZAP list. This only gets | |
6933 | * called if the config is dirty; otherwise there may be | |
6934 | * outstanding AVZ operations that weren't completed in | |
6935 | * spa_sync_config_object. | |
6936 | */ | |
6937 | uint64_t all_vdev_zap_entry_count; | |
6938 | ASSERT0(zap_count(spa->spa_meta_objset, | |
6939 | spa->spa_all_vdev_zaps, &all_vdev_zap_entry_count)); | |
6940 | ASSERT3U(vdev_count_verify_zaps(spa->spa_root_vdev), ==, | |
6941 | all_vdev_zap_entry_count); | |
6942 | } | |
6943 | #endif | |
6944 | ||
6945 | /* | |
6946 | * Rewrite the vdev configuration (which includes the uberblock) | |
6947 | * to commit the transaction group. | |
6948 | * | |
6949 | * If there are no dirty vdevs, we sync the uberblock to a few | |
6950 | * random top-level vdevs that are known to be visible in the | |
6951 | * config cache (see spa_vdev_add() for a complete description). | |
6952 | * If there *are* dirty vdevs, sync the uberblock to all vdevs. | |
6953 | */ | |
6954 | for (;;) { | |
6955 | /* | |
6956 | * We hold SCL_STATE to prevent vdev open/close/etc. | |
6957 | * while we're attempting to write the vdev labels. | |
6958 | */ | |
6959 | spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); | |
6960 | ||
6961 | if (list_is_empty(&spa->spa_config_dirty_list)) { | |
6962 | vdev_t *svd[SPA_DVAS_PER_BP]; | |
6963 | int svdcount = 0; | |
6964 | int children = rvd->vdev_children; | |
6965 | int c0 = spa_get_random(children); | |
6966 | ||
6967 | for (int c = 0; c < children; c++) { | |
6968 | vd = rvd->vdev_child[(c0 + c) % children]; | |
6969 | if (vd->vdev_ms_array == 0 || vd->vdev_islog) | |
6970 | continue; | |
6971 | svd[svdcount++] = vd; | |
6972 | if (svdcount == SPA_DVAS_PER_BP) | |
6973 | break; | |
6974 | } | |
6975 | error = vdev_config_sync(svd, svdcount, txg); | |
6976 | } else { | |
6977 | error = vdev_config_sync(rvd->vdev_child, | |
6978 | rvd->vdev_children, txg); | |
6979 | } | |
6980 | ||
6981 | if (error == 0) | |
6982 | spa->spa_last_synced_guid = rvd->vdev_guid; | |
6983 | ||
6984 | spa_config_exit(spa, SCL_STATE, FTAG); | |
6985 | ||
6986 | if (error == 0) | |
6987 | break; | |
6988 | zio_suspend(spa, NULL); | |
6989 | zio_resume_wait(spa); | |
6990 | } | |
6991 | dmu_tx_commit(tx); | |
6992 | ||
6993 | taskq_cancel_id(system_delay_taskq, spa->spa_deadman_tqid); | |
6994 | spa->spa_deadman_tqid = 0; | |
6995 | ||
6996 | /* | |
6997 | * Clear the dirty config list. | |
6998 | */ | |
6999 | while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL) | |
7000 | vdev_config_clean(vd); | |
7001 | ||
7002 | /* | |
7003 | * Now that the new config has synced transactionally, | |
7004 | * let it become visible to the config cache. | |
7005 | */ | |
7006 | if (spa->spa_config_syncing != NULL) { | |
7007 | spa_config_set(spa, spa->spa_config_syncing); | |
7008 | spa->spa_config_txg = txg; | |
7009 | spa->spa_config_syncing = NULL; | |
7010 | } | |
7011 | ||
7012 | dsl_pool_sync_done(dp, txg); | |
7013 | ||
7014 | mutex_enter(&spa->spa_alloc_lock); | |
7015 | VERIFY0(avl_numnodes(&spa->spa_alloc_tree)); | |
7016 | mutex_exit(&spa->spa_alloc_lock); | |
7017 | ||
7018 | /* | |
7019 | * Update usable space statistics. | |
7020 | */ | |
7021 | while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))) | |
7022 | vdev_sync_done(vd, txg); | |
7023 | ||
7024 | spa_update_dspace(spa); | |
7025 | ||
7026 | /* | |
7027 | * It had better be the case that we didn't dirty anything | |
7028 | * since vdev_config_sync(). | |
7029 | */ | |
7030 | ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg)); | |
7031 | ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); | |
7032 | ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg)); | |
7033 | ||
7034 | spa->spa_sync_pass = 0; | |
7035 | ||
7036 | /* | |
7037 | * Update the last synced uberblock here. We want to do this at | |
7038 | * the end of spa_sync() so that consumers of spa_last_synced_txg() | |
7039 | * will be guaranteed that all the processing associated with | |
7040 | * that txg has been completed. | |
7041 | */ | |
7042 | spa->spa_ubsync = spa->spa_uberblock; | |
7043 | spa_config_exit(spa, SCL_CONFIG, FTAG); | |
7044 | ||
7045 | spa_handle_ignored_writes(spa); | |
7046 | ||
7047 | /* | |
7048 | * If any async tasks have been requested, kick them off. | |
7049 | */ | |
7050 | spa_async_dispatch(spa); | |
7051 | } | |
7052 | ||
7053 | /* | |
7054 | * Sync all pools. We don't want to hold the namespace lock across these | |
7055 | * operations, so we take a reference on the spa_t and drop the lock during the | |
7056 | * sync. | |
7057 | */ | |
7058 | void | |
7059 | spa_sync_allpools(void) | |
7060 | { | |
7061 | spa_t *spa = NULL; | |
7062 | mutex_enter(&spa_namespace_lock); | |
7063 | while ((spa = spa_next(spa)) != NULL) { | |
7064 | if (spa_state(spa) != POOL_STATE_ACTIVE || | |
7065 | !spa_writeable(spa) || spa_suspended(spa)) | |
7066 | continue; | |
7067 | spa_open_ref(spa, FTAG); | |
7068 | mutex_exit(&spa_namespace_lock); | |
7069 | txg_wait_synced(spa_get_dsl(spa), 0); | |
7070 | mutex_enter(&spa_namespace_lock); | |
7071 | spa_close(spa, FTAG); | |
7072 | } | |
7073 | mutex_exit(&spa_namespace_lock); | |
7074 | } | |
7075 | ||
7076 | /* | |
7077 | * ========================================================================== | |
7078 | * Miscellaneous routines | |
7079 | * ========================================================================== | |
7080 | */ | |
7081 | ||
7082 | /* | |
7083 | * Remove all pools in the system. | |
7084 | */ | |
7085 | void | |
7086 | spa_evict_all(void) | |
7087 | { | |
7088 | spa_t *spa; | |
7089 | ||
7090 | /* | |
7091 | * Remove all cached state. All pools should be closed now, | |
7092 | * so every spa in the AVL tree should be unreferenced. | |
7093 | */ | |
7094 | mutex_enter(&spa_namespace_lock); | |
7095 | while ((spa = spa_next(NULL)) != NULL) { | |
7096 | /* | |
7097 | * Stop async tasks. The async thread may need to detach | |
7098 | * a device that's been replaced, which requires grabbing | |
7099 | * spa_namespace_lock, so we must drop it here. | |
7100 | */ | |
7101 | spa_open_ref(spa, FTAG); | |
7102 | mutex_exit(&spa_namespace_lock); | |
7103 | spa_async_suspend(spa); | |
7104 | mutex_enter(&spa_namespace_lock); | |
7105 | spa_close(spa, FTAG); | |
7106 | ||
7107 | if (spa->spa_state != POOL_STATE_UNINITIALIZED) { | |
7108 | spa_unload(spa); | |
7109 | spa_deactivate(spa); | |
7110 | } | |
7111 | spa_remove(spa); | |
7112 | } | |
7113 | mutex_exit(&spa_namespace_lock); | |
7114 | } | |
7115 | ||
7116 | vdev_t * | |
7117 | spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux) | |
7118 | { | |
7119 | vdev_t *vd; | |
7120 | int i; | |
7121 | ||
7122 | if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL) | |
7123 | return (vd); | |
7124 | ||
7125 | if (aux) { | |
7126 | for (i = 0; i < spa->spa_l2cache.sav_count; i++) { | |
7127 | vd = spa->spa_l2cache.sav_vdevs[i]; | |
7128 | if (vd->vdev_guid == guid) | |
7129 | return (vd); | |
7130 | } | |
7131 | ||
7132 | for (i = 0; i < spa->spa_spares.sav_count; i++) { | |
7133 | vd = spa->spa_spares.sav_vdevs[i]; | |
7134 | if (vd->vdev_guid == guid) | |
7135 | return (vd); | |
7136 | } | |
7137 | } | |
7138 | ||
7139 | return (NULL); | |
7140 | } | |
7141 | ||
7142 | void | |
7143 | spa_upgrade(spa_t *spa, uint64_t version) | |
7144 | { | |
7145 | ASSERT(spa_writeable(spa)); | |
7146 | ||
7147 | spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); | |
7148 | ||
7149 | /* | |
7150 | * This should only be called for a non-faulted pool, and since a | |
7151 | * future version would result in an unopenable pool, this shouldn't be | |
7152 | * possible. | |
7153 | */ | |
7154 | ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version)); | |
7155 | ASSERT3U(version, >=, spa->spa_uberblock.ub_version); | |
7156 | ||
7157 | spa->spa_uberblock.ub_version = version; | |
7158 | vdev_config_dirty(spa->spa_root_vdev); | |
7159 | ||
7160 | spa_config_exit(spa, SCL_ALL, FTAG); | |
7161 | ||
7162 | txg_wait_synced(spa_get_dsl(spa), 0); | |
7163 | } | |
7164 | ||
7165 | boolean_t | |
7166 | spa_has_spare(spa_t *spa, uint64_t guid) | |
7167 | { | |
7168 | int i; | |
7169 | uint64_t spareguid; | |
7170 | spa_aux_vdev_t *sav = &spa->spa_spares; | |
7171 | ||
7172 | for (i = 0; i < sav->sav_count; i++) | |
7173 | if (sav->sav_vdevs[i]->vdev_guid == guid) | |
7174 | return (B_TRUE); | |
7175 | ||
7176 | for (i = 0; i < sav->sav_npending; i++) { | |
7177 | if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID, | |
7178 | &spareguid) == 0 && spareguid == guid) | |
7179 | return (B_TRUE); | |
7180 | } | |
7181 | ||
7182 | return (B_FALSE); | |
7183 | } | |
7184 | ||
7185 | /* | |
7186 | * Check if a pool has an active shared spare device. | |
7187 | * Note: reference count of an active spare is 2, as a spare and as a replace | |
7188 | */ | |
7189 | static boolean_t | |
7190 | spa_has_active_shared_spare(spa_t *spa) | |
7191 | { | |
7192 | int i, refcnt; | |
7193 | uint64_t pool; | |
7194 | spa_aux_vdev_t *sav = &spa->spa_spares; | |
7195 | ||
7196 | for (i = 0; i < sav->sav_count; i++) { | |
7197 | if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool, | |
7198 | &refcnt) && pool != 0ULL && pool == spa_guid(spa) && | |
7199 | refcnt > 2) | |
7200 | return (B_TRUE); | |
7201 | } | |
7202 | ||
7203 | return (B_FALSE); | |
7204 | } | |
7205 | ||
7206 | static sysevent_t * | |
7207 | spa_event_create(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, const char *name) | |
7208 | { | |
7209 | sysevent_t *ev = NULL; | |
7210 | #ifdef _KERNEL | |
7211 | nvlist_t *resource; | |
7212 | ||
7213 | resource = zfs_event_create(spa, vd, FM_SYSEVENT_CLASS, name, hist_nvl); | |
7214 | if (resource) { | |
7215 | ev = kmem_alloc(sizeof (sysevent_t), KM_SLEEP); | |
7216 | ev->resource = resource; | |
7217 | } | |
7218 | #endif | |
7219 | return (ev); | |
7220 | } | |
7221 | ||
7222 | static void | |
7223 | spa_event_post(sysevent_t *ev) | |
7224 | { | |
7225 | #ifdef _KERNEL | |
7226 | if (ev) { | |
7227 | zfs_zevent_post(ev->resource, NULL, zfs_zevent_post_cb); | |
7228 | kmem_free(ev, sizeof (*ev)); | |
7229 | } | |
7230 | #endif | |
7231 | } | |
7232 | ||
7233 | /* | |
7234 | * Post a zevent corresponding to the given sysevent. The 'name' must be one | |
7235 | * of the event definitions in sys/sysevent/eventdefs.h. The payload will be | |
7236 | * filled in from the spa and (optionally) the vdev. This doesn't do anything | |
7237 | * in the userland libzpool, as we don't want consumers to misinterpret ztest | |
7238 | * or zdb as real changes. | |
7239 | */ | |
7240 | void | |
7241 | spa_event_notify(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, const char *name) | |
7242 | { | |
7243 | spa_event_post(spa_event_create(spa, vd, hist_nvl, name)); | |
7244 | } | |
7245 | ||
7246 | #if defined(_KERNEL) && defined(HAVE_SPL) | |
7247 | /* state manipulation functions */ | |
7248 | EXPORT_SYMBOL(spa_open); | |
7249 | EXPORT_SYMBOL(spa_open_rewind); | |
7250 | EXPORT_SYMBOL(spa_get_stats); | |
7251 | EXPORT_SYMBOL(spa_create); | |
7252 | EXPORT_SYMBOL(spa_import); | |
7253 | EXPORT_SYMBOL(spa_tryimport); | |
7254 | EXPORT_SYMBOL(spa_destroy); | |
7255 | EXPORT_SYMBOL(spa_export); | |
7256 | EXPORT_SYMBOL(spa_reset); | |
7257 | EXPORT_SYMBOL(spa_async_request); | |
7258 | EXPORT_SYMBOL(spa_async_suspend); | |
7259 | EXPORT_SYMBOL(spa_async_resume); | |
7260 | EXPORT_SYMBOL(spa_inject_addref); | |
7261 | EXPORT_SYMBOL(spa_inject_delref); | |
7262 | EXPORT_SYMBOL(spa_scan_stat_init); | |
7263 | EXPORT_SYMBOL(spa_scan_get_stats); | |
7264 | ||
7265 | /* device maniion */ | |
7266 | EXPORT_SYMBOL(spa_vdev_add); | |
7267 | EXPORT_SYMBOL(spa_vdev_attach); | |
7268 | EXPORT_SYMBOL(spa_vdev_detach); | |
7269 | EXPORT_SYMBOL(spa_vdev_remove); | |
7270 | EXPORT_SYMBOL(spa_vdev_setpath); | |
7271 | EXPORT_SYMBOL(spa_vdev_setfru); | |
7272 | EXPORT_SYMBOL(spa_vdev_split_mirror); | |
7273 | ||
7274 | /* spare statech is global across all pools) */ | |
7275 | EXPORT_SYMBOL(spa_spare_add); | |
7276 | EXPORT_SYMBOL(spa_spare_remove); | |
7277 | EXPORT_SYMBOL(spa_spare_exists); | |
7278 | EXPORT_SYMBOL(spa_spare_activate); | |
7279 | ||
7280 | /* L2ARC statech is global across all pools) */ | |
7281 | EXPORT_SYMBOL(spa_l2cache_add); | |
7282 | EXPORT_SYMBOL(spa_l2cache_remove); | |
7283 | EXPORT_SYMBOL(spa_l2cache_exists); | |
7284 | EXPORT_SYMBOL(spa_l2cache_activate); | |
7285 | EXPORT_SYMBOL(spa_l2cache_drop); | |
7286 | ||
7287 | /* scanning */ | |
7288 | EXPORT_SYMBOL(spa_scan); | |
7289 | EXPORT_SYMBOL(spa_scan_stop); | |
7290 | ||
7291 | /* spa syncing */ | |
7292 | EXPORT_SYMBOL(spa_sync); /* only for DMU use */ | |
7293 | EXPORT_SYMBOL(spa_sync_allpools); | |
7294 | ||
7295 | /* properties */ | |
7296 | EXPORT_SYMBOL(spa_prop_set); | |
7297 | EXPORT_SYMBOL(spa_prop_get); | |
7298 | EXPORT_SYMBOL(spa_prop_clear_bootfs); | |
7299 | ||
7300 | /* asynchronous event notification */ | |
7301 | EXPORT_SYMBOL(spa_event_notify); | |
7302 | #endif | |
7303 | ||
7304 | #if defined(_KERNEL) && defined(HAVE_SPL) | |
7305 | module_param(spa_load_verify_maxinflight, int, 0644); | |
7306 | MODULE_PARM_DESC(spa_load_verify_maxinflight, | |
7307 | "Max concurrent traversal I/Os while verifying pool during import -X"); | |
7308 | ||
7309 | module_param(spa_load_verify_metadata, int, 0644); | |
7310 | MODULE_PARM_DESC(spa_load_verify_metadata, | |
7311 | "Set to traverse metadata on pool import"); | |
7312 | ||
7313 | module_param(spa_load_verify_data, int, 0644); | |
7314 | MODULE_PARM_DESC(spa_load_verify_data, | |
7315 | "Set to traverse data on pool import"); | |
7316 | ||
7317 | /* CSTYLED */ | |
7318 | module_param(zio_taskq_batch_pct, uint, 0444); | |
7319 | MODULE_PARM_DESC(zio_taskq_batch_pct, | |
7320 | "Percentage of CPUs to run an IO worker thread"); | |
7321 | ||
7322 | #endif |