Print this page
3949 ztest fault injection should avoid resilvering devices
3950 ztest: deadman fires when we're doing a scan
3951 ztest hang when running dedup test
3952 ztest: ztest_reguid test and ztest_fault_inject don't place nice together
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Reviewed by: Adam Leventhal <ahl@delphix.com>
Split |
Close |
Expand all |
Collapse all |
--- old/usr/src/uts/common/fs/zfs/spa.c
+++ new/usr/src/uts/common/fs/zfs/spa.c
1 1 /*
2 2 * CDDL HEADER START
3 3 *
4 4 * The contents of this file are subject to the terms of the
5 5 * Common Development and Distribution License (the "License").
6 6 * You may not use this file except in compliance with the License.
7 7 *
8 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 9 * or http://www.opensolaris.org/os/licensing.
10 10 * See the License for the specific language governing permissions
11 11 * and limitations under the License.
12 12 *
13 13 * When distributing Covered Code, include this CDDL HEADER in each
14 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 15 * If applicable, add the following below this CDDL HEADER, with the
16 16 * fields enclosed by brackets "[]" replaced with your own identifying
17 17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 18 *
19 19 * CDDL HEADER END
20 20 */
21 21
22 22 /*
23 23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 24 * Copyright (c) 2013 by Delphix. All rights reserved.
25 25 * Copyright 2013 Nexenta Systems, Inc. All rights reserved.
26 26 */
27 27
28 28 /*
29 29 * SPA: Storage Pool Allocator
30 30 *
31 31 * This file contains all the routines used when modifying on-disk SPA state.
32 32 * This includes opening, importing, destroying, exporting a pool, and syncing a
33 33 * pool.
34 34 */
35 35
36 36 #include <sys/zfs_context.h>
37 37 #include <sys/fm/fs/zfs.h>
38 38 #include <sys/spa_impl.h>
39 39 #include <sys/zio.h>
40 40 #include <sys/zio_checksum.h>
41 41 #include <sys/dmu.h>
42 42 #include <sys/dmu_tx.h>
43 43 #include <sys/zap.h>
44 44 #include <sys/zil.h>
45 45 #include <sys/ddt.h>
46 46 #include <sys/vdev_impl.h>
47 47 #include <sys/metaslab.h>
48 48 #include <sys/metaslab_impl.h>
49 49 #include <sys/uberblock_impl.h>
50 50 #include <sys/txg.h>
51 51 #include <sys/avl.h>
52 52 #include <sys/dmu_traverse.h>
53 53 #include <sys/dmu_objset.h>
54 54 #include <sys/unique.h>
55 55 #include <sys/dsl_pool.h>
56 56 #include <sys/dsl_dataset.h>
57 57 #include <sys/dsl_dir.h>
58 58 #include <sys/dsl_prop.h>
59 59 #include <sys/dsl_synctask.h>
60 60 #include <sys/fs/zfs.h>
61 61 #include <sys/arc.h>
62 62 #include <sys/callb.h>
63 63 #include <sys/systeminfo.h>
64 64 #include <sys/spa_boot.h>
65 65 #include <sys/zfs_ioctl.h>
66 66 #include <sys/dsl_scan.h>
67 67 #include <sys/zfeature.h>
68 68 #include <sys/dsl_destroy.h>
69 69
70 70 #ifdef _KERNEL
71 71 #include <sys/bootprops.h>
72 72 #include <sys/callb.h>
73 73 #include <sys/cpupart.h>
74 74 #include <sys/pool.h>
75 75 #include <sys/sysdc.h>
76 76 #include <sys/zone.h>
77 77 #endif /* _KERNEL */
78 78
79 79 #include "zfs_prop.h"
80 80 #include "zfs_comutil.h"
81 81
82 82 /*
83 83 * The interval, in seconds, at which failed configuration cache file writes
84 84 * should be retried.
85 85 */
86 86 static int zfs_ccw_retry_interval = 300;
87 87
88 88 typedef enum zti_modes {
89 89 ZTI_MODE_FIXED, /* value is # of threads (min 1) */
90 90 ZTI_MODE_ONLINE_PERCENT, /* value is % of online CPUs */
91 91 ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */
92 92 ZTI_MODE_NULL, /* don't create a taskq */
93 93 ZTI_NMODES
94 94 } zti_modes_t;
95 95
96 96 #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) }
97 97 #define ZTI_PCT(n) { ZTI_MODE_ONLINE_PERCENT, (n), 1 }
98 98 #define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 }
99 99 #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 }
100 100
101 101 #define ZTI_N(n) ZTI_P(n, 1)
102 102 #define ZTI_ONE ZTI_N(1)
103 103
104 104 typedef struct zio_taskq_info {
105 105 zti_modes_t zti_mode;
106 106 uint_t zti_value;
107 107 uint_t zti_count;
108 108 } zio_taskq_info_t;
109 109
110 110 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
111 111 "issue", "issue_high", "intr", "intr_high"
112 112 };
113 113
114 114 /*
115 115 * This table defines the taskq settings for each ZFS I/O type. When
116 116 * initializing a pool, we use this table to create an appropriately sized
117 117 * taskq. Some operations are low volume and therefore have a small, static
118 118 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
119 119 * macros. Other operations process a large amount of data; the ZTI_BATCH
120 120 * macro causes us to create a taskq oriented for throughput. Some operations
121 121 * are so high frequency and short-lived that the taskq itself can become a a
122 122 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
123 123 * additional degree of parallelism specified by the number of threads per-
124 124 * taskq and the number of taskqs; when dispatching an event in this case, the
125 125 * particular taskq is chosen at random.
126 126 *
127 127 * The different taskq priorities are to handle the different contexts (issue
128 128 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
129 129 * need to be handled with minimum delay.
130 130 */
131 131 const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
132 132 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
133 133 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */
134 134 { ZTI_N(8), ZTI_NULL, ZTI_BATCH, ZTI_NULL }, /* READ */
135 135 { ZTI_BATCH, ZTI_N(5), ZTI_N(8), ZTI_N(5) }, /* WRITE */
136 136 { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */
137 137 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */
138 138 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */
139 139 };
140 140
141 141 static void spa_sync_version(void *arg, dmu_tx_t *tx);
142 142 static void spa_sync_props(void *arg, dmu_tx_t *tx);
143 143 static boolean_t spa_has_active_shared_spare(spa_t *spa);
144 144 static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
145 145 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
146 146 char **ereport);
147 147 static void spa_vdev_resilver_done(spa_t *spa);
148 148
149 149 uint_t zio_taskq_batch_pct = 100; /* 1 thread per cpu in pset */
150 150 id_t zio_taskq_psrset_bind = PS_NONE;
151 151 boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
152 152 uint_t zio_taskq_basedc = 80; /* base duty cycle */
153 153
154 154 boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
155 155 extern int zfs_sync_pass_deferred_free;
156 156
157 157 /*
158 158 * This (illegal) pool name is used when temporarily importing a spa_t in order
159 159 * to get the vdev stats associated with the imported devices.
160 160 */
161 161 #define TRYIMPORT_NAME "$import"
162 162
163 163 /*
164 164 * ==========================================================================
165 165 * SPA properties routines
166 166 * ==========================================================================
167 167 */
168 168
169 169 /*
170 170 * Add a (source=src, propname=propval) list to an nvlist.
171 171 */
172 172 static void
173 173 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
174 174 uint64_t intval, zprop_source_t src)
175 175 {
176 176 const char *propname = zpool_prop_to_name(prop);
177 177 nvlist_t *propval;
178 178
179 179 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
180 180 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
181 181
182 182 if (strval != NULL)
183 183 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
184 184 else
185 185 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
186 186
187 187 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
188 188 nvlist_free(propval);
189 189 }
190 190
191 191 /*
192 192 * Get property values from the spa configuration.
193 193 */
194 194 static void
195 195 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
196 196 {
197 197 vdev_t *rvd = spa->spa_root_vdev;
198 198 dsl_pool_t *pool = spa->spa_dsl_pool;
199 199 uint64_t size;
200 200 uint64_t alloc;
201 201 uint64_t space;
202 202 uint64_t cap, version;
203 203 zprop_source_t src = ZPROP_SRC_NONE;
204 204 spa_config_dirent_t *dp;
205 205
206 206 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
207 207
208 208 if (rvd != NULL) {
209 209 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
210 210 size = metaslab_class_get_space(spa_normal_class(spa));
211 211 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
212 212 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
213 213 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
214 214 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
215 215 size - alloc, src);
216 216
217 217 space = 0;
218 218 for (int c = 0; c < rvd->vdev_children; c++) {
219 219 vdev_t *tvd = rvd->vdev_child[c];
220 220 space += tvd->vdev_max_asize - tvd->vdev_asize;
221 221 }
222 222 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, space,
223 223 src);
224 224
225 225 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
226 226 (spa_mode(spa) == FREAD), src);
227 227
228 228 cap = (size == 0) ? 0 : (alloc * 100 / size);
229 229 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
230 230
231 231 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
232 232 ddt_get_pool_dedup_ratio(spa), src);
233 233
234 234 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
235 235 rvd->vdev_state, src);
236 236
237 237 version = spa_version(spa);
238 238 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
239 239 src = ZPROP_SRC_DEFAULT;
240 240 else
241 241 src = ZPROP_SRC_LOCAL;
242 242 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
243 243 }
244 244
245 245 if (pool != NULL) {
246 246 dsl_dir_t *freedir = pool->dp_free_dir;
247 247
248 248 /*
249 249 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
250 250 * when opening pools before this version freedir will be NULL.
251 251 */
252 252 if (freedir != NULL) {
253 253 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
254 254 freedir->dd_phys->dd_used_bytes, src);
255 255 } else {
256 256 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
257 257 NULL, 0, src);
258 258 }
259 259 }
260 260
261 261 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
262 262
263 263 if (spa->spa_comment != NULL) {
264 264 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
265 265 0, ZPROP_SRC_LOCAL);
266 266 }
267 267
268 268 if (spa->spa_root != NULL)
269 269 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
270 270 0, ZPROP_SRC_LOCAL);
271 271
272 272 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
273 273 if (dp->scd_path == NULL) {
274 274 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
275 275 "none", 0, ZPROP_SRC_LOCAL);
276 276 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
277 277 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
278 278 dp->scd_path, 0, ZPROP_SRC_LOCAL);
279 279 }
280 280 }
281 281 }
282 282
283 283 /*
284 284 * Get zpool property values.
285 285 */
286 286 int
287 287 spa_prop_get(spa_t *spa, nvlist_t **nvp)
288 288 {
289 289 objset_t *mos = spa->spa_meta_objset;
290 290 zap_cursor_t zc;
291 291 zap_attribute_t za;
292 292 int err;
293 293
294 294 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
295 295
296 296 mutex_enter(&spa->spa_props_lock);
297 297
298 298 /*
299 299 * Get properties from the spa config.
300 300 */
301 301 spa_prop_get_config(spa, nvp);
302 302
303 303 /* If no pool property object, no more prop to get. */
304 304 if (mos == NULL || spa->spa_pool_props_object == 0) {
305 305 mutex_exit(&spa->spa_props_lock);
306 306 return (0);
307 307 }
308 308
309 309 /*
310 310 * Get properties from the MOS pool property object.
311 311 */
312 312 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
313 313 (err = zap_cursor_retrieve(&zc, &za)) == 0;
314 314 zap_cursor_advance(&zc)) {
315 315 uint64_t intval = 0;
316 316 char *strval = NULL;
317 317 zprop_source_t src = ZPROP_SRC_DEFAULT;
318 318 zpool_prop_t prop;
319 319
320 320 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
321 321 continue;
322 322
323 323 switch (za.za_integer_length) {
324 324 case 8:
325 325 /* integer property */
326 326 if (za.za_first_integer !=
327 327 zpool_prop_default_numeric(prop))
328 328 src = ZPROP_SRC_LOCAL;
329 329
330 330 if (prop == ZPOOL_PROP_BOOTFS) {
331 331 dsl_pool_t *dp;
332 332 dsl_dataset_t *ds = NULL;
333 333
334 334 dp = spa_get_dsl(spa);
335 335 dsl_pool_config_enter(dp, FTAG);
336 336 if (err = dsl_dataset_hold_obj(dp,
337 337 za.za_first_integer, FTAG, &ds)) {
338 338 dsl_pool_config_exit(dp, FTAG);
339 339 break;
340 340 }
341 341
342 342 strval = kmem_alloc(
343 343 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
344 344 KM_SLEEP);
345 345 dsl_dataset_name(ds, strval);
346 346 dsl_dataset_rele(ds, FTAG);
347 347 dsl_pool_config_exit(dp, FTAG);
348 348 } else {
349 349 strval = NULL;
350 350 intval = za.za_first_integer;
351 351 }
352 352
353 353 spa_prop_add_list(*nvp, prop, strval, intval, src);
354 354
355 355 if (strval != NULL)
356 356 kmem_free(strval,
357 357 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
358 358
359 359 break;
360 360
361 361 case 1:
362 362 /* string property */
363 363 strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
364 364 err = zap_lookup(mos, spa->spa_pool_props_object,
365 365 za.za_name, 1, za.za_num_integers, strval);
366 366 if (err) {
367 367 kmem_free(strval, za.za_num_integers);
368 368 break;
369 369 }
370 370 spa_prop_add_list(*nvp, prop, strval, 0, src);
371 371 kmem_free(strval, za.za_num_integers);
372 372 break;
373 373
374 374 default:
375 375 break;
376 376 }
377 377 }
378 378 zap_cursor_fini(&zc);
379 379 mutex_exit(&spa->spa_props_lock);
380 380 out:
381 381 if (err && err != ENOENT) {
382 382 nvlist_free(*nvp);
383 383 *nvp = NULL;
384 384 return (err);
385 385 }
386 386
387 387 return (0);
388 388 }
389 389
390 390 /*
391 391 * Validate the given pool properties nvlist and modify the list
392 392 * for the property values to be set.
393 393 */
394 394 static int
395 395 spa_prop_validate(spa_t *spa, nvlist_t *props)
396 396 {
397 397 nvpair_t *elem;
398 398 int error = 0, reset_bootfs = 0;
399 399 uint64_t objnum = 0;
400 400 boolean_t has_feature = B_FALSE;
401 401
402 402 elem = NULL;
403 403 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
404 404 uint64_t intval;
405 405 char *strval, *slash, *check, *fname;
406 406 const char *propname = nvpair_name(elem);
407 407 zpool_prop_t prop = zpool_name_to_prop(propname);
408 408
409 409 switch (prop) {
410 410 case ZPROP_INVAL:
411 411 if (!zpool_prop_feature(propname)) {
412 412 error = SET_ERROR(EINVAL);
413 413 break;
414 414 }
415 415
416 416 /*
417 417 * Sanitize the input.
418 418 */
419 419 if (nvpair_type(elem) != DATA_TYPE_UINT64) {
420 420 error = SET_ERROR(EINVAL);
421 421 break;
422 422 }
423 423
424 424 if (nvpair_value_uint64(elem, &intval) != 0) {
425 425 error = SET_ERROR(EINVAL);
426 426 break;
427 427 }
428 428
429 429 if (intval != 0) {
430 430 error = SET_ERROR(EINVAL);
431 431 break;
432 432 }
433 433
434 434 fname = strchr(propname, '@') + 1;
435 435 if (zfeature_lookup_name(fname, NULL) != 0) {
436 436 error = SET_ERROR(EINVAL);
437 437 break;
438 438 }
439 439
440 440 has_feature = B_TRUE;
441 441 break;
442 442
443 443 case ZPOOL_PROP_VERSION:
444 444 error = nvpair_value_uint64(elem, &intval);
445 445 if (!error &&
446 446 (intval < spa_version(spa) ||
447 447 intval > SPA_VERSION_BEFORE_FEATURES ||
448 448 has_feature))
449 449 error = SET_ERROR(EINVAL);
450 450 break;
451 451
452 452 case ZPOOL_PROP_DELEGATION:
453 453 case ZPOOL_PROP_AUTOREPLACE:
454 454 case ZPOOL_PROP_LISTSNAPS:
455 455 case ZPOOL_PROP_AUTOEXPAND:
456 456 error = nvpair_value_uint64(elem, &intval);
457 457 if (!error && intval > 1)
458 458 error = SET_ERROR(EINVAL);
459 459 break;
460 460
461 461 case ZPOOL_PROP_BOOTFS:
462 462 /*
463 463 * If the pool version is less than SPA_VERSION_BOOTFS,
464 464 * or the pool is still being created (version == 0),
465 465 * the bootfs property cannot be set.
466 466 */
467 467 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
468 468 error = SET_ERROR(ENOTSUP);
469 469 break;
470 470 }
471 471
472 472 /*
473 473 * Make sure the vdev config is bootable
474 474 */
475 475 if (!vdev_is_bootable(spa->spa_root_vdev)) {
476 476 error = SET_ERROR(ENOTSUP);
477 477 break;
478 478 }
479 479
480 480 reset_bootfs = 1;
481 481
482 482 error = nvpair_value_string(elem, &strval);
483 483
484 484 if (!error) {
485 485 objset_t *os;
486 486 uint64_t compress;
487 487
488 488 if (strval == NULL || strval[0] == '\0') {
489 489 objnum = zpool_prop_default_numeric(
490 490 ZPOOL_PROP_BOOTFS);
491 491 break;
492 492 }
493 493
494 494 if (error = dmu_objset_hold(strval, FTAG, &os))
495 495 break;
496 496
497 497 /* Must be ZPL and not gzip compressed. */
498 498
499 499 if (dmu_objset_type(os) != DMU_OST_ZFS) {
500 500 error = SET_ERROR(ENOTSUP);
501 501 } else if ((error =
502 502 dsl_prop_get_int_ds(dmu_objset_ds(os),
503 503 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
504 504 &compress)) == 0 &&
505 505 !BOOTFS_COMPRESS_VALID(compress)) {
506 506 error = SET_ERROR(ENOTSUP);
507 507 } else {
508 508 objnum = dmu_objset_id(os);
509 509 }
510 510 dmu_objset_rele(os, FTAG);
511 511 }
512 512 break;
513 513
514 514 case ZPOOL_PROP_FAILUREMODE:
515 515 error = nvpair_value_uint64(elem, &intval);
516 516 if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
517 517 intval > ZIO_FAILURE_MODE_PANIC))
518 518 error = SET_ERROR(EINVAL);
519 519
520 520 /*
521 521 * This is a special case which only occurs when
522 522 * the pool has completely failed. This allows
523 523 * the user to change the in-core failmode property
524 524 * without syncing it out to disk (I/Os might
525 525 * currently be blocked). We do this by returning
526 526 * EIO to the caller (spa_prop_set) to trick it
527 527 * into thinking we encountered a property validation
528 528 * error.
529 529 */
530 530 if (!error && spa_suspended(spa)) {
531 531 spa->spa_failmode = intval;
532 532 error = SET_ERROR(EIO);
533 533 }
534 534 break;
535 535
536 536 case ZPOOL_PROP_CACHEFILE:
537 537 if ((error = nvpair_value_string(elem, &strval)) != 0)
538 538 break;
539 539
540 540 if (strval[0] == '\0')
541 541 break;
542 542
543 543 if (strcmp(strval, "none") == 0)
544 544 break;
545 545
546 546 if (strval[0] != '/') {
547 547 error = SET_ERROR(EINVAL);
548 548 break;
549 549 }
550 550
551 551 slash = strrchr(strval, '/');
552 552 ASSERT(slash != NULL);
553 553
554 554 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
555 555 strcmp(slash, "/..") == 0)
556 556 error = SET_ERROR(EINVAL);
557 557 break;
558 558
559 559 case ZPOOL_PROP_COMMENT:
560 560 if ((error = nvpair_value_string(elem, &strval)) != 0)
561 561 break;
562 562 for (check = strval; *check != '\0'; check++) {
563 563 /*
564 564 * The kernel doesn't have an easy isprint()
565 565 * check. For this kernel check, we merely
566 566 * check ASCII apart from DEL. Fix this if
567 567 * there is an easy-to-use kernel isprint().
568 568 */
569 569 if (*check >= 0x7f) {
570 570 error = SET_ERROR(EINVAL);
571 571 break;
572 572 }
573 573 check++;
574 574 }
575 575 if (strlen(strval) > ZPROP_MAX_COMMENT)
576 576 error = E2BIG;
577 577 break;
578 578
579 579 case ZPOOL_PROP_DEDUPDITTO:
580 580 if (spa_version(spa) < SPA_VERSION_DEDUP)
581 581 error = SET_ERROR(ENOTSUP);
582 582 else
583 583 error = nvpair_value_uint64(elem, &intval);
584 584 if (error == 0 &&
585 585 intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
586 586 error = SET_ERROR(EINVAL);
587 587 break;
588 588 }
589 589
590 590 if (error)
591 591 break;
592 592 }
593 593
594 594 if (!error && reset_bootfs) {
595 595 error = nvlist_remove(props,
596 596 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
597 597
598 598 if (!error) {
599 599 error = nvlist_add_uint64(props,
600 600 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
601 601 }
602 602 }
603 603
604 604 return (error);
605 605 }
606 606
607 607 void
608 608 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
609 609 {
610 610 char *cachefile;
611 611 spa_config_dirent_t *dp;
612 612
613 613 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
614 614 &cachefile) != 0)
615 615 return;
616 616
617 617 dp = kmem_alloc(sizeof (spa_config_dirent_t),
618 618 KM_SLEEP);
619 619
620 620 if (cachefile[0] == '\0')
621 621 dp->scd_path = spa_strdup(spa_config_path);
622 622 else if (strcmp(cachefile, "none") == 0)
623 623 dp->scd_path = NULL;
624 624 else
625 625 dp->scd_path = spa_strdup(cachefile);
626 626
627 627 list_insert_head(&spa->spa_config_list, dp);
628 628 if (need_sync)
629 629 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
630 630 }
631 631
632 632 int
633 633 spa_prop_set(spa_t *spa, nvlist_t *nvp)
634 634 {
635 635 int error;
636 636 nvpair_t *elem = NULL;
637 637 boolean_t need_sync = B_FALSE;
638 638
639 639 if ((error = spa_prop_validate(spa, nvp)) != 0)
640 640 return (error);
641 641
642 642 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
643 643 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
644 644
645 645 if (prop == ZPOOL_PROP_CACHEFILE ||
646 646 prop == ZPOOL_PROP_ALTROOT ||
647 647 prop == ZPOOL_PROP_READONLY)
648 648 continue;
649 649
650 650 if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) {
651 651 uint64_t ver;
652 652
653 653 if (prop == ZPOOL_PROP_VERSION) {
654 654 VERIFY(nvpair_value_uint64(elem, &ver) == 0);
655 655 } else {
656 656 ASSERT(zpool_prop_feature(nvpair_name(elem)));
657 657 ver = SPA_VERSION_FEATURES;
658 658 need_sync = B_TRUE;
659 659 }
660 660
661 661 /* Save time if the version is already set. */
662 662 if (ver == spa_version(spa))
663 663 continue;
664 664
665 665 /*
666 666 * In addition to the pool directory object, we might
667 667 * create the pool properties object, the features for
668 668 * read object, the features for write object, or the
669 669 * feature descriptions object.
670 670 */
671 671 error = dsl_sync_task(spa->spa_name, NULL,
672 672 spa_sync_version, &ver, 6);
673 673 if (error)
674 674 return (error);
675 675 continue;
676 676 }
677 677
678 678 need_sync = B_TRUE;
679 679 break;
680 680 }
681 681
682 682 if (need_sync) {
683 683 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props,
684 684 nvp, 6));
685 685 }
686 686
687 687 return (0);
688 688 }
689 689
690 690 /*
691 691 * If the bootfs property value is dsobj, clear it.
692 692 */
693 693 void
694 694 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
695 695 {
696 696 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
697 697 VERIFY(zap_remove(spa->spa_meta_objset,
698 698 spa->spa_pool_props_object,
699 699 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
700 700 spa->spa_bootfs = 0;
701 701 }
702 702 }
703 703
704 704 /*ARGSUSED*/
705 705 static int
706 706 spa_change_guid_check(void *arg, dmu_tx_t *tx)
707 707 {
708 708 uint64_t *newguid = arg;
709 709 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
710 710 vdev_t *rvd = spa->spa_root_vdev;
711 711 uint64_t vdev_state;
712 712
713 713 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
714 714 vdev_state = rvd->vdev_state;
715 715 spa_config_exit(spa, SCL_STATE, FTAG);
716 716
717 717 if (vdev_state != VDEV_STATE_HEALTHY)
718 718 return (SET_ERROR(ENXIO));
719 719
720 720 ASSERT3U(spa_guid(spa), !=, *newguid);
721 721
722 722 return (0);
723 723 }
724 724
725 725 static void
726 726 spa_change_guid_sync(void *arg, dmu_tx_t *tx)
727 727 {
728 728 uint64_t *newguid = arg;
729 729 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
730 730 uint64_t oldguid;
731 731 vdev_t *rvd = spa->spa_root_vdev;
732 732
733 733 oldguid = spa_guid(spa);
734 734
735 735 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
736 736 rvd->vdev_guid = *newguid;
737 737 rvd->vdev_guid_sum += (*newguid - oldguid);
738 738 vdev_config_dirty(rvd);
739 739 spa_config_exit(spa, SCL_STATE, FTAG);
740 740
741 741 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu",
742 742 oldguid, *newguid);
743 743 }
744 744
745 745 /*
746 746 * Change the GUID for the pool. This is done so that we can later
747 747 * re-import a pool built from a clone of our own vdevs. We will modify
748 748 * the root vdev's guid, our own pool guid, and then mark all of our
749 749 * vdevs dirty. Note that we must make sure that all our vdevs are
↓ open down ↓ |
749 lines elided |
↑ open up ↑ |
750 750 * online when we do this, or else any vdevs that weren't present
751 751 * would be orphaned from our pool. We are also going to issue a
752 752 * sysevent to update any watchers.
753 753 */
754 754 int
755 755 spa_change_guid(spa_t *spa)
756 756 {
757 757 int error;
758 758 uint64_t guid;
759 759
760 + mutex_enter(&spa->spa_vdev_top_lock);
760 761 mutex_enter(&spa_namespace_lock);
761 762 guid = spa_generate_guid(NULL);
762 763
763 764 error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
764 765 spa_change_guid_sync, &guid, 5);
765 766
766 767 if (error == 0) {
767 768 spa_config_sync(spa, B_FALSE, B_TRUE);
768 769 spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
769 770 }
770 771
771 772 mutex_exit(&spa_namespace_lock);
773 + mutex_exit(&spa->spa_vdev_top_lock);
772 774
773 775 return (error);
774 776 }
775 777
776 778 /*
777 779 * ==========================================================================
778 780 * SPA state manipulation (open/create/destroy/import/export)
779 781 * ==========================================================================
780 782 */
781 783
782 784 static int
783 785 spa_error_entry_compare(const void *a, const void *b)
784 786 {
785 787 spa_error_entry_t *sa = (spa_error_entry_t *)a;
786 788 spa_error_entry_t *sb = (spa_error_entry_t *)b;
787 789 int ret;
788 790
789 791 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
790 792 sizeof (zbookmark_t));
791 793
792 794 if (ret < 0)
793 795 return (-1);
794 796 else if (ret > 0)
795 797 return (1);
796 798 else
797 799 return (0);
798 800 }
799 801
800 802 /*
801 803 * Utility function which retrieves copies of the current logs and
802 804 * re-initializes them in the process.
803 805 */
804 806 void
805 807 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
806 808 {
807 809 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
808 810
809 811 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
810 812 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
811 813
812 814 avl_create(&spa->spa_errlist_scrub,
813 815 spa_error_entry_compare, sizeof (spa_error_entry_t),
814 816 offsetof(spa_error_entry_t, se_avl));
815 817 avl_create(&spa->spa_errlist_last,
816 818 spa_error_entry_compare, sizeof (spa_error_entry_t),
817 819 offsetof(spa_error_entry_t, se_avl));
818 820 }
819 821
820 822 static void
821 823 spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
822 824 {
823 825 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
824 826 enum zti_modes mode = ztip->zti_mode;
825 827 uint_t value = ztip->zti_value;
826 828 uint_t count = ztip->zti_count;
827 829 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
828 830 char name[32];
829 831 uint_t flags = 0;
830 832 boolean_t batch = B_FALSE;
831 833
832 834 if (mode == ZTI_MODE_NULL) {
833 835 tqs->stqs_count = 0;
834 836 tqs->stqs_taskq = NULL;
835 837 return;
836 838 }
837 839
838 840 ASSERT3U(count, >, 0);
839 841
840 842 tqs->stqs_count = count;
841 843 tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP);
842 844
843 845 for (uint_t i = 0; i < count; i++) {
844 846 taskq_t *tq;
845 847
846 848 switch (mode) {
847 849 case ZTI_MODE_FIXED:
848 850 ASSERT3U(value, >=, 1);
849 851 value = MAX(value, 1);
850 852 break;
851 853
852 854 case ZTI_MODE_BATCH:
853 855 batch = B_TRUE;
854 856 flags |= TASKQ_THREADS_CPU_PCT;
855 857 value = zio_taskq_batch_pct;
856 858 break;
857 859
858 860 case ZTI_MODE_ONLINE_PERCENT:
859 861 flags |= TASKQ_THREADS_CPU_PCT;
860 862 break;
861 863
862 864 default:
863 865 panic("unrecognized mode for %s_%s taskq (%u:%u) in "
864 866 "spa_activate()",
865 867 zio_type_name[t], zio_taskq_types[q], mode, value);
866 868 break;
867 869 }
868 870
869 871 if (count > 1) {
870 872 (void) snprintf(name, sizeof (name), "%s_%s_%u",
871 873 zio_type_name[t], zio_taskq_types[q], i);
872 874 } else {
873 875 (void) snprintf(name, sizeof (name), "%s_%s",
874 876 zio_type_name[t], zio_taskq_types[q]);
875 877 }
876 878
877 879 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
878 880 if (batch)
879 881 flags |= TASKQ_DC_BATCH;
880 882
881 883 tq = taskq_create_sysdc(name, value, 50, INT_MAX,
882 884 spa->spa_proc, zio_taskq_basedc, flags);
883 885 } else {
884 886 tq = taskq_create_proc(name, value, maxclsyspri, 50,
885 887 INT_MAX, spa->spa_proc, flags);
886 888 }
887 889
888 890 tqs->stqs_taskq[i] = tq;
889 891 }
890 892 }
891 893
892 894 static void
893 895 spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
894 896 {
895 897 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
896 898
897 899 if (tqs->stqs_taskq == NULL) {
898 900 ASSERT0(tqs->stqs_count);
899 901 return;
900 902 }
901 903
902 904 for (uint_t i = 0; i < tqs->stqs_count; i++) {
903 905 ASSERT3P(tqs->stqs_taskq[i], !=, NULL);
904 906 taskq_destroy(tqs->stqs_taskq[i]);
905 907 }
906 908
907 909 kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *));
908 910 tqs->stqs_taskq = NULL;
909 911 }
910 912
911 913 /*
912 914 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
913 915 * Note that a type may have multiple discrete taskqs to avoid lock contention
914 916 * on the taskq itself. In that case we choose which taskq at random by using
915 917 * the low bits of gethrtime().
916 918 */
917 919 void
918 920 spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
919 921 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent)
920 922 {
921 923 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
922 924 taskq_t *tq;
923 925
924 926 ASSERT3P(tqs->stqs_taskq, !=, NULL);
925 927 ASSERT3U(tqs->stqs_count, !=, 0);
926 928
927 929 if (tqs->stqs_count == 1) {
928 930 tq = tqs->stqs_taskq[0];
929 931 } else {
930 932 tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count];
931 933 }
932 934
933 935 taskq_dispatch_ent(tq, func, arg, flags, ent);
934 936 }
935 937
936 938 static void
937 939 spa_create_zio_taskqs(spa_t *spa)
938 940 {
939 941 for (int t = 0; t < ZIO_TYPES; t++) {
940 942 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
941 943 spa_taskqs_init(spa, t, q);
942 944 }
943 945 }
944 946 }
945 947
946 948 #ifdef _KERNEL
947 949 static void
948 950 spa_thread(void *arg)
949 951 {
950 952 callb_cpr_t cprinfo;
951 953
952 954 spa_t *spa = arg;
953 955 user_t *pu = PTOU(curproc);
954 956
955 957 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
956 958 spa->spa_name);
957 959
958 960 ASSERT(curproc != &p0);
959 961 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
960 962 "zpool-%s", spa->spa_name);
961 963 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
962 964
963 965 /* bind this thread to the requested psrset */
964 966 if (zio_taskq_psrset_bind != PS_NONE) {
965 967 pool_lock();
966 968 mutex_enter(&cpu_lock);
967 969 mutex_enter(&pidlock);
968 970 mutex_enter(&curproc->p_lock);
969 971
970 972 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
971 973 0, NULL, NULL) == 0) {
972 974 curthread->t_bind_pset = zio_taskq_psrset_bind;
973 975 } else {
974 976 cmn_err(CE_WARN,
975 977 "Couldn't bind process for zfs pool \"%s\" to "
976 978 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
977 979 }
978 980
979 981 mutex_exit(&curproc->p_lock);
980 982 mutex_exit(&pidlock);
981 983 mutex_exit(&cpu_lock);
982 984 pool_unlock();
983 985 }
984 986
985 987 if (zio_taskq_sysdc) {
986 988 sysdc_thread_enter(curthread, 100, 0);
987 989 }
988 990
989 991 spa->spa_proc = curproc;
990 992 spa->spa_did = curthread->t_did;
991 993
992 994 spa_create_zio_taskqs(spa);
993 995
994 996 mutex_enter(&spa->spa_proc_lock);
995 997 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
996 998
997 999 spa->spa_proc_state = SPA_PROC_ACTIVE;
998 1000 cv_broadcast(&spa->spa_proc_cv);
999 1001
1000 1002 CALLB_CPR_SAFE_BEGIN(&cprinfo);
1001 1003 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
1002 1004 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1003 1005 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
1004 1006
1005 1007 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
1006 1008 spa->spa_proc_state = SPA_PROC_GONE;
1007 1009 spa->spa_proc = &p0;
1008 1010 cv_broadcast(&spa->spa_proc_cv);
1009 1011 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
1010 1012
1011 1013 mutex_enter(&curproc->p_lock);
1012 1014 lwp_exit();
1013 1015 }
1014 1016 #endif
1015 1017
1016 1018 /*
1017 1019 * Activate an uninitialized pool.
1018 1020 */
1019 1021 static void
1020 1022 spa_activate(spa_t *spa, int mode)
1021 1023 {
1022 1024 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
1023 1025
1024 1026 spa->spa_state = POOL_STATE_ACTIVE;
1025 1027 spa->spa_mode = mode;
1026 1028
1027 1029 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
1028 1030 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
1029 1031
1030 1032 /* Try to create a covering process */
1031 1033 mutex_enter(&spa->spa_proc_lock);
1032 1034 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
1033 1035 ASSERT(spa->spa_proc == &p0);
1034 1036 spa->spa_did = 0;
1035 1037
1036 1038 /* Only create a process if we're going to be around a while. */
1037 1039 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
1038 1040 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
1039 1041 NULL, 0) == 0) {
1040 1042 spa->spa_proc_state = SPA_PROC_CREATED;
1041 1043 while (spa->spa_proc_state == SPA_PROC_CREATED) {
1042 1044 cv_wait(&spa->spa_proc_cv,
1043 1045 &spa->spa_proc_lock);
1044 1046 }
1045 1047 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1046 1048 ASSERT(spa->spa_proc != &p0);
1047 1049 ASSERT(spa->spa_did != 0);
1048 1050 } else {
1049 1051 #ifdef _KERNEL
1050 1052 cmn_err(CE_WARN,
1051 1053 "Couldn't create process for zfs pool \"%s\"\n",
1052 1054 spa->spa_name);
1053 1055 #endif
1054 1056 }
1055 1057 }
1056 1058 mutex_exit(&spa->spa_proc_lock);
1057 1059
1058 1060 /* If we didn't create a process, we need to create our taskqs. */
1059 1061 if (spa->spa_proc == &p0) {
1060 1062 spa_create_zio_taskqs(spa);
1061 1063 }
1062 1064
1063 1065 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1064 1066 offsetof(vdev_t, vdev_config_dirty_node));
1065 1067 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1066 1068 offsetof(vdev_t, vdev_state_dirty_node));
1067 1069
1068 1070 txg_list_create(&spa->spa_vdev_txg_list,
1069 1071 offsetof(struct vdev, vdev_txg_node));
1070 1072
1071 1073 avl_create(&spa->spa_errlist_scrub,
1072 1074 spa_error_entry_compare, sizeof (spa_error_entry_t),
1073 1075 offsetof(spa_error_entry_t, se_avl));
1074 1076 avl_create(&spa->spa_errlist_last,
1075 1077 spa_error_entry_compare, sizeof (spa_error_entry_t),
1076 1078 offsetof(spa_error_entry_t, se_avl));
1077 1079 }
1078 1080
1079 1081 /*
1080 1082 * Opposite of spa_activate().
1081 1083 */
1082 1084 static void
1083 1085 spa_deactivate(spa_t *spa)
1084 1086 {
1085 1087 ASSERT(spa->spa_sync_on == B_FALSE);
1086 1088 ASSERT(spa->spa_dsl_pool == NULL);
1087 1089 ASSERT(spa->spa_root_vdev == NULL);
1088 1090 ASSERT(spa->spa_async_zio_root == NULL);
1089 1091 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1090 1092
1091 1093 txg_list_destroy(&spa->spa_vdev_txg_list);
1092 1094
1093 1095 list_destroy(&spa->spa_config_dirty_list);
1094 1096 list_destroy(&spa->spa_state_dirty_list);
1095 1097
1096 1098 for (int t = 0; t < ZIO_TYPES; t++) {
1097 1099 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1098 1100 spa_taskqs_fini(spa, t, q);
1099 1101 }
1100 1102 }
1101 1103
1102 1104 metaslab_class_destroy(spa->spa_normal_class);
1103 1105 spa->spa_normal_class = NULL;
1104 1106
1105 1107 metaslab_class_destroy(spa->spa_log_class);
1106 1108 spa->spa_log_class = NULL;
1107 1109
1108 1110 /*
1109 1111 * If this was part of an import or the open otherwise failed, we may
1110 1112 * still have errors left in the queues. Empty them just in case.
1111 1113 */
1112 1114 spa_errlog_drain(spa);
1113 1115
1114 1116 avl_destroy(&spa->spa_errlist_scrub);
1115 1117 avl_destroy(&spa->spa_errlist_last);
1116 1118
1117 1119 spa->spa_state = POOL_STATE_UNINITIALIZED;
1118 1120
1119 1121 mutex_enter(&spa->spa_proc_lock);
1120 1122 if (spa->spa_proc_state != SPA_PROC_NONE) {
1121 1123 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1122 1124 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1123 1125 cv_broadcast(&spa->spa_proc_cv);
1124 1126 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1125 1127 ASSERT(spa->spa_proc != &p0);
1126 1128 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1127 1129 }
1128 1130 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1129 1131 spa->spa_proc_state = SPA_PROC_NONE;
1130 1132 }
1131 1133 ASSERT(spa->spa_proc == &p0);
1132 1134 mutex_exit(&spa->spa_proc_lock);
1133 1135
1134 1136 /*
1135 1137 * We want to make sure spa_thread() has actually exited the ZFS
1136 1138 * module, so that the module can't be unloaded out from underneath
1137 1139 * it.
1138 1140 */
1139 1141 if (spa->spa_did != 0) {
1140 1142 thread_join(spa->spa_did);
1141 1143 spa->spa_did = 0;
1142 1144 }
1143 1145 }
1144 1146
1145 1147 /*
1146 1148 * Verify a pool configuration, and construct the vdev tree appropriately. This
1147 1149 * will create all the necessary vdevs in the appropriate layout, with each vdev
1148 1150 * in the CLOSED state. This will prep the pool before open/creation/import.
1149 1151 * All vdev validation is done by the vdev_alloc() routine.
1150 1152 */
1151 1153 static int
1152 1154 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1153 1155 uint_t id, int atype)
1154 1156 {
1155 1157 nvlist_t **child;
1156 1158 uint_t children;
1157 1159 int error;
1158 1160
1159 1161 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1160 1162 return (error);
1161 1163
1162 1164 if ((*vdp)->vdev_ops->vdev_op_leaf)
1163 1165 return (0);
1164 1166
1165 1167 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1166 1168 &child, &children);
1167 1169
1168 1170 if (error == ENOENT)
1169 1171 return (0);
1170 1172
1171 1173 if (error) {
1172 1174 vdev_free(*vdp);
1173 1175 *vdp = NULL;
1174 1176 return (SET_ERROR(EINVAL));
1175 1177 }
1176 1178
1177 1179 for (int c = 0; c < children; c++) {
1178 1180 vdev_t *vd;
1179 1181 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1180 1182 atype)) != 0) {
1181 1183 vdev_free(*vdp);
1182 1184 *vdp = NULL;
1183 1185 return (error);
1184 1186 }
1185 1187 }
1186 1188
1187 1189 ASSERT(*vdp != NULL);
1188 1190
1189 1191 return (0);
1190 1192 }
1191 1193
1192 1194 /*
1193 1195 * Opposite of spa_load().
1194 1196 */
1195 1197 static void
1196 1198 spa_unload(spa_t *spa)
1197 1199 {
1198 1200 int i;
1199 1201
1200 1202 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1201 1203
1202 1204 /*
1203 1205 * Stop async tasks.
1204 1206 */
1205 1207 spa_async_suspend(spa);
1206 1208
1207 1209 /*
1208 1210 * Stop syncing.
1209 1211 */
1210 1212 if (spa->spa_sync_on) {
1211 1213 txg_sync_stop(spa->spa_dsl_pool);
1212 1214 spa->spa_sync_on = B_FALSE;
1213 1215 }
1214 1216
1215 1217 /*
1216 1218 * Wait for any outstanding async I/O to complete.
1217 1219 */
1218 1220 if (spa->spa_async_zio_root != NULL) {
1219 1221 (void) zio_wait(spa->spa_async_zio_root);
1220 1222 spa->spa_async_zio_root = NULL;
1221 1223 }
1222 1224
1223 1225 bpobj_close(&spa->spa_deferred_bpobj);
1224 1226
1225 1227 /*
1226 1228 * Close the dsl pool.
1227 1229 */
1228 1230 if (spa->spa_dsl_pool) {
1229 1231 dsl_pool_close(spa->spa_dsl_pool);
1230 1232 spa->spa_dsl_pool = NULL;
1231 1233 spa->spa_meta_objset = NULL;
1232 1234 }
1233 1235
1234 1236 ddt_unload(spa);
1235 1237
1236 1238 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1237 1239
1238 1240 /*
1239 1241 * Drop and purge level 2 cache
1240 1242 */
1241 1243 spa_l2cache_drop(spa);
1242 1244
1243 1245 /*
1244 1246 * Close all vdevs.
1245 1247 */
1246 1248 if (spa->spa_root_vdev)
1247 1249 vdev_free(spa->spa_root_vdev);
1248 1250 ASSERT(spa->spa_root_vdev == NULL);
1249 1251
1250 1252 for (i = 0; i < spa->spa_spares.sav_count; i++)
1251 1253 vdev_free(spa->spa_spares.sav_vdevs[i]);
1252 1254 if (spa->spa_spares.sav_vdevs) {
1253 1255 kmem_free(spa->spa_spares.sav_vdevs,
1254 1256 spa->spa_spares.sav_count * sizeof (void *));
1255 1257 spa->spa_spares.sav_vdevs = NULL;
1256 1258 }
1257 1259 if (spa->spa_spares.sav_config) {
1258 1260 nvlist_free(spa->spa_spares.sav_config);
1259 1261 spa->spa_spares.sav_config = NULL;
1260 1262 }
1261 1263 spa->spa_spares.sav_count = 0;
1262 1264
1263 1265 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1264 1266 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1265 1267 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1266 1268 }
1267 1269 if (spa->spa_l2cache.sav_vdevs) {
1268 1270 kmem_free(spa->spa_l2cache.sav_vdevs,
1269 1271 spa->spa_l2cache.sav_count * sizeof (void *));
1270 1272 spa->spa_l2cache.sav_vdevs = NULL;
1271 1273 }
1272 1274 if (spa->spa_l2cache.sav_config) {
1273 1275 nvlist_free(spa->spa_l2cache.sav_config);
1274 1276 spa->spa_l2cache.sav_config = NULL;
1275 1277 }
1276 1278 spa->spa_l2cache.sav_count = 0;
1277 1279
1278 1280 spa->spa_async_suspended = 0;
1279 1281
1280 1282 if (spa->spa_comment != NULL) {
1281 1283 spa_strfree(spa->spa_comment);
1282 1284 spa->spa_comment = NULL;
1283 1285 }
1284 1286
1285 1287 spa_config_exit(spa, SCL_ALL, FTAG);
1286 1288 }
1287 1289
1288 1290 /*
1289 1291 * Load (or re-load) the current list of vdevs describing the active spares for
1290 1292 * this pool. When this is called, we have some form of basic information in
1291 1293 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1292 1294 * then re-generate a more complete list including status information.
1293 1295 */
1294 1296 static void
1295 1297 spa_load_spares(spa_t *spa)
1296 1298 {
1297 1299 nvlist_t **spares;
1298 1300 uint_t nspares;
1299 1301 int i;
1300 1302 vdev_t *vd, *tvd;
1301 1303
1302 1304 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1303 1305
1304 1306 /*
1305 1307 * First, close and free any existing spare vdevs.
1306 1308 */
1307 1309 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1308 1310 vd = spa->spa_spares.sav_vdevs[i];
1309 1311
1310 1312 /* Undo the call to spa_activate() below */
1311 1313 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1312 1314 B_FALSE)) != NULL && tvd->vdev_isspare)
1313 1315 spa_spare_remove(tvd);
1314 1316 vdev_close(vd);
1315 1317 vdev_free(vd);
1316 1318 }
1317 1319
1318 1320 if (spa->spa_spares.sav_vdevs)
1319 1321 kmem_free(spa->spa_spares.sav_vdevs,
1320 1322 spa->spa_spares.sav_count * sizeof (void *));
1321 1323
1322 1324 if (spa->spa_spares.sav_config == NULL)
1323 1325 nspares = 0;
1324 1326 else
1325 1327 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1326 1328 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1327 1329
1328 1330 spa->spa_spares.sav_count = (int)nspares;
1329 1331 spa->spa_spares.sav_vdevs = NULL;
1330 1332
1331 1333 if (nspares == 0)
1332 1334 return;
1333 1335
1334 1336 /*
1335 1337 * Construct the array of vdevs, opening them to get status in the
1336 1338 * process. For each spare, there is potentially two different vdev_t
1337 1339 * structures associated with it: one in the list of spares (used only
1338 1340 * for basic validation purposes) and one in the active vdev
1339 1341 * configuration (if it's spared in). During this phase we open and
1340 1342 * validate each vdev on the spare list. If the vdev also exists in the
1341 1343 * active configuration, then we also mark this vdev as an active spare.
1342 1344 */
1343 1345 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1344 1346 KM_SLEEP);
1345 1347 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1346 1348 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1347 1349 VDEV_ALLOC_SPARE) == 0);
1348 1350 ASSERT(vd != NULL);
1349 1351
1350 1352 spa->spa_spares.sav_vdevs[i] = vd;
1351 1353
1352 1354 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1353 1355 B_FALSE)) != NULL) {
1354 1356 if (!tvd->vdev_isspare)
1355 1357 spa_spare_add(tvd);
1356 1358
1357 1359 /*
1358 1360 * We only mark the spare active if we were successfully
1359 1361 * able to load the vdev. Otherwise, importing a pool
1360 1362 * with a bad active spare would result in strange
1361 1363 * behavior, because multiple pool would think the spare
1362 1364 * is actively in use.
1363 1365 *
1364 1366 * There is a vulnerability here to an equally bizarre
1365 1367 * circumstance, where a dead active spare is later
1366 1368 * brought back to life (onlined or otherwise). Given
1367 1369 * the rarity of this scenario, and the extra complexity
1368 1370 * it adds, we ignore the possibility.
1369 1371 */
1370 1372 if (!vdev_is_dead(tvd))
1371 1373 spa_spare_activate(tvd);
1372 1374 }
1373 1375
1374 1376 vd->vdev_top = vd;
1375 1377 vd->vdev_aux = &spa->spa_spares;
1376 1378
1377 1379 if (vdev_open(vd) != 0)
1378 1380 continue;
1379 1381
1380 1382 if (vdev_validate_aux(vd) == 0)
1381 1383 spa_spare_add(vd);
1382 1384 }
1383 1385
1384 1386 /*
1385 1387 * Recompute the stashed list of spares, with status information
1386 1388 * this time.
1387 1389 */
1388 1390 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1389 1391 DATA_TYPE_NVLIST_ARRAY) == 0);
1390 1392
1391 1393 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1392 1394 KM_SLEEP);
1393 1395 for (i = 0; i < spa->spa_spares.sav_count; i++)
1394 1396 spares[i] = vdev_config_generate(spa,
1395 1397 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1396 1398 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1397 1399 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1398 1400 for (i = 0; i < spa->spa_spares.sav_count; i++)
1399 1401 nvlist_free(spares[i]);
1400 1402 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1401 1403 }
1402 1404
1403 1405 /*
1404 1406 * Load (or re-load) the current list of vdevs describing the active l2cache for
1405 1407 * this pool. When this is called, we have some form of basic information in
1406 1408 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1407 1409 * then re-generate a more complete list including status information.
1408 1410 * Devices which are already active have their details maintained, and are
1409 1411 * not re-opened.
1410 1412 */
1411 1413 static void
1412 1414 spa_load_l2cache(spa_t *spa)
1413 1415 {
1414 1416 nvlist_t **l2cache;
1415 1417 uint_t nl2cache;
1416 1418 int i, j, oldnvdevs;
1417 1419 uint64_t guid;
1418 1420 vdev_t *vd, **oldvdevs, **newvdevs;
1419 1421 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1420 1422
1421 1423 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1422 1424
1423 1425 if (sav->sav_config != NULL) {
1424 1426 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1425 1427 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1426 1428 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1427 1429 } else {
1428 1430 nl2cache = 0;
1429 1431 newvdevs = NULL;
1430 1432 }
1431 1433
1432 1434 oldvdevs = sav->sav_vdevs;
1433 1435 oldnvdevs = sav->sav_count;
1434 1436 sav->sav_vdevs = NULL;
1435 1437 sav->sav_count = 0;
1436 1438
1437 1439 /*
1438 1440 * Process new nvlist of vdevs.
1439 1441 */
1440 1442 for (i = 0; i < nl2cache; i++) {
1441 1443 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1442 1444 &guid) == 0);
1443 1445
1444 1446 newvdevs[i] = NULL;
1445 1447 for (j = 0; j < oldnvdevs; j++) {
1446 1448 vd = oldvdevs[j];
1447 1449 if (vd != NULL && guid == vd->vdev_guid) {
1448 1450 /*
1449 1451 * Retain previous vdev for add/remove ops.
1450 1452 */
1451 1453 newvdevs[i] = vd;
1452 1454 oldvdevs[j] = NULL;
1453 1455 break;
1454 1456 }
1455 1457 }
1456 1458
1457 1459 if (newvdevs[i] == NULL) {
1458 1460 /*
1459 1461 * Create new vdev
1460 1462 */
1461 1463 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1462 1464 VDEV_ALLOC_L2CACHE) == 0);
1463 1465 ASSERT(vd != NULL);
1464 1466 newvdevs[i] = vd;
1465 1467
1466 1468 /*
1467 1469 * Commit this vdev as an l2cache device,
1468 1470 * even if it fails to open.
1469 1471 */
1470 1472 spa_l2cache_add(vd);
1471 1473
1472 1474 vd->vdev_top = vd;
1473 1475 vd->vdev_aux = sav;
1474 1476
1475 1477 spa_l2cache_activate(vd);
1476 1478
1477 1479 if (vdev_open(vd) != 0)
1478 1480 continue;
1479 1481
1480 1482 (void) vdev_validate_aux(vd);
1481 1483
1482 1484 if (!vdev_is_dead(vd))
1483 1485 l2arc_add_vdev(spa, vd);
1484 1486 }
1485 1487 }
1486 1488
1487 1489 /*
1488 1490 * Purge vdevs that were dropped
1489 1491 */
1490 1492 for (i = 0; i < oldnvdevs; i++) {
1491 1493 uint64_t pool;
1492 1494
1493 1495 vd = oldvdevs[i];
1494 1496 if (vd != NULL) {
1495 1497 ASSERT(vd->vdev_isl2cache);
1496 1498
1497 1499 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1498 1500 pool != 0ULL && l2arc_vdev_present(vd))
1499 1501 l2arc_remove_vdev(vd);
1500 1502 vdev_clear_stats(vd);
1501 1503 vdev_free(vd);
1502 1504 }
1503 1505 }
1504 1506
1505 1507 if (oldvdevs)
1506 1508 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1507 1509
1508 1510 if (sav->sav_config == NULL)
1509 1511 goto out;
1510 1512
1511 1513 sav->sav_vdevs = newvdevs;
1512 1514 sav->sav_count = (int)nl2cache;
1513 1515
1514 1516 /*
1515 1517 * Recompute the stashed list of l2cache devices, with status
1516 1518 * information this time.
1517 1519 */
1518 1520 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1519 1521 DATA_TYPE_NVLIST_ARRAY) == 0);
1520 1522
1521 1523 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1522 1524 for (i = 0; i < sav->sav_count; i++)
1523 1525 l2cache[i] = vdev_config_generate(spa,
1524 1526 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1525 1527 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1526 1528 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1527 1529 out:
1528 1530 for (i = 0; i < sav->sav_count; i++)
1529 1531 nvlist_free(l2cache[i]);
1530 1532 if (sav->sav_count)
1531 1533 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1532 1534 }
1533 1535
1534 1536 static int
1535 1537 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1536 1538 {
1537 1539 dmu_buf_t *db;
1538 1540 char *packed = NULL;
1539 1541 size_t nvsize = 0;
1540 1542 int error;
1541 1543 *value = NULL;
1542 1544
1543 1545 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1544 1546 nvsize = *(uint64_t *)db->db_data;
1545 1547 dmu_buf_rele(db, FTAG);
1546 1548
1547 1549 packed = kmem_alloc(nvsize, KM_SLEEP);
1548 1550 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1549 1551 DMU_READ_PREFETCH);
1550 1552 if (error == 0)
1551 1553 error = nvlist_unpack(packed, nvsize, value, 0);
1552 1554 kmem_free(packed, nvsize);
1553 1555
1554 1556 return (error);
1555 1557 }
1556 1558
1557 1559 /*
1558 1560 * Checks to see if the given vdev could not be opened, in which case we post a
1559 1561 * sysevent to notify the autoreplace code that the device has been removed.
1560 1562 */
1561 1563 static void
1562 1564 spa_check_removed(vdev_t *vd)
1563 1565 {
1564 1566 for (int c = 0; c < vd->vdev_children; c++)
1565 1567 spa_check_removed(vd->vdev_child[c]);
1566 1568
1567 1569 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1568 1570 !vd->vdev_ishole) {
1569 1571 zfs_post_autoreplace(vd->vdev_spa, vd);
1570 1572 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1571 1573 }
1572 1574 }
1573 1575
1574 1576 /*
1575 1577 * Validate the current config against the MOS config
1576 1578 */
1577 1579 static boolean_t
1578 1580 spa_config_valid(spa_t *spa, nvlist_t *config)
1579 1581 {
1580 1582 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1581 1583 nvlist_t *nv;
1582 1584
1583 1585 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1584 1586
1585 1587 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1586 1588 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1587 1589
1588 1590 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1589 1591
1590 1592 /*
1591 1593 * If we're doing a normal import, then build up any additional
1592 1594 * diagnostic information about missing devices in this config.
1593 1595 * We'll pass this up to the user for further processing.
1594 1596 */
1595 1597 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1596 1598 nvlist_t **child, *nv;
1597 1599 uint64_t idx = 0;
1598 1600
1599 1601 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1600 1602 KM_SLEEP);
1601 1603 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1602 1604
1603 1605 for (int c = 0; c < rvd->vdev_children; c++) {
1604 1606 vdev_t *tvd = rvd->vdev_child[c];
1605 1607 vdev_t *mtvd = mrvd->vdev_child[c];
1606 1608
1607 1609 if (tvd->vdev_ops == &vdev_missing_ops &&
1608 1610 mtvd->vdev_ops != &vdev_missing_ops &&
1609 1611 mtvd->vdev_islog)
1610 1612 child[idx++] = vdev_config_generate(spa, mtvd,
1611 1613 B_FALSE, 0);
1612 1614 }
1613 1615
1614 1616 if (idx) {
1615 1617 VERIFY(nvlist_add_nvlist_array(nv,
1616 1618 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1617 1619 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1618 1620 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1619 1621
1620 1622 for (int i = 0; i < idx; i++)
1621 1623 nvlist_free(child[i]);
1622 1624 }
1623 1625 nvlist_free(nv);
1624 1626 kmem_free(child, rvd->vdev_children * sizeof (char **));
1625 1627 }
1626 1628
1627 1629 /*
1628 1630 * Compare the root vdev tree with the information we have
1629 1631 * from the MOS config (mrvd). Check each top-level vdev
1630 1632 * with the corresponding MOS config top-level (mtvd).
1631 1633 */
1632 1634 for (int c = 0; c < rvd->vdev_children; c++) {
1633 1635 vdev_t *tvd = rvd->vdev_child[c];
1634 1636 vdev_t *mtvd = mrvd->vdev_child[c];
1635 1637
1636 1638 /*
1637 1639 * Resolve any "missing" vdevs in the current configuration.
1638 1640 * If we find that the MOS config has more accurate information
1639 1641 * about the top-level vdev then use that vdev instead.
1640 1642 */
1641 1643 if (tvd->vdev_ops == &vdev_missing_ops &&
1642 1644 mtvd->vdev_ops != &vdev_missing_ops) {
1643 1645
1644 1646 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1645 1647 continue;
1646 1648
1647 1649 /*
1648 1650 * Device specific actions.
1649 1651 */
1650 1652 if (mtvd->vdev_islog) {
1651 1653 spa_set_log_state(spa, SPA_LOG_CLEAR);
1652 1654 } else {
1653 1655 /*
1654 1656 * XXX - once we have 'readonly' pool
1655 1657 * support we should be able to handle
1656 1658 * missing data devices by transitioning
1657 1659 * the pool to readonly.
1658 1660 */
1659 1661 continue;
1660 1662 }
1661 1663
1662 1664 /*
1663 1665 * Swap the missing vdev with the data we were
1664 1666 * able to obtain from the MOS config.
1665 1667 */
1666 1668 vdev_remove_child(rvd, tvd);
1667 1669 vdev_remove_child(mrvd, mtvd);
1668 1670
1669 1671 vdev_add_child(rvd, mtvd);
1670 1672 vdev_add_child(mrvd, tvd);
1671 1673
1672 1674 spa_config_exit(spa, SCL_ALL, FTAG);
1673 1675 vdev_load(mtvd);
1674 1676 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1675 1677
1676 1678 vdev_reopen(rvd);
1677 1679 } else if (mtvd->vdev_islog) {
1678 1680 /*
1679 1681 * Load the slog device's state from the MOS config
1680 1682 * since it's possible that the label does not
1681 1683 * contain the most up-to-date information.
1682 1684 */
1683 1685 vdev_load_log_state(tvd, mtvd);
1684 1686 vdev_reopen(tvd);
1685 1687 }
1686 1688 }
1687 1689 vdev_free(mrvd);
1688 1690 spa_config_exit(spa, SCL_ALL, FTAG);
1689 1691
1690 1692 /*
1691 1693 * Ensure we were able to validate the config.
1692 1694 */
1693 1695 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1694 1696 }
1695 1697
1696 1698 /*
1697 1699 * Check for missing log devices
1698 1700 */
1699 1701 static boolean_t
1700 1702 spa_check_logs(spa_t *spa)
1701 1703 {
1702 1704 boolean_t rv = B_FALSE;
1703 1705
1704 1706 switch (spa->spa_log_state) {
1705 1707 case SPA_LOG_MISSING:
1706 1708 /* need to recheck in case slog has been restored */
1707 1709 case SPA_LOG_UNKNOWN:
1708 1710 rv = (dmu_objset_find(spa->spa_name, zil_check_log_chain,
1709 1711 NULL, DS_FIND_CHILDREN) != 0);
1710 1712 if (rv)
1711 1713 spa_set_log_state(spa, SPA_LOG_MISSING);
1712 1714 break;
1713 1715 }
1714 1716 return (rv);
1715 1717 }
1716 1718
1717 1719 static boolean_t
1718 1720 spa_passivate_log(spa_t *spa)
1719 1721 {
1720 1722 vdev_t *rvd = spa->spa_root_vdev;
1721 1723 boolean_t slog_found = B_FALSE;
1722 1724
1723 1725 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1724 1726
1725 1727 if (!spa_has_slogs(spa))
1726 1728 return (B_FALSE);
1727 1729
1728 1730 for (int c = 0; c < rvd->vdev_children; c++) {
1729 1731 vdev_t *tvd = rvd->vdev_child[c];
1730 1732 metaslab_group_t *mg = tvd->vdev_mg;
1731 1733
1732 1734 if (tvd->vdev_islog) {
1733 1735 metaslab_group_passivate(mg);
1734 1736 slog_found = B_TRUE;
1735 1737 }
1736 1738 }
1737 1739
1738 1740 return (slog_found);
1739 1741 }
1740 1742
1741 1743 static void
1742 1744 spa_activate_log(spa_t *spa)
1743 1745 {
1744 1746 vdev_t *rvd = spa->spa_root_vdev;
1745 1747
1746 1748 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1747 1749
1748 1750 for (int c = 0; c < rvd->vdev_children; c++) {
1749 1751 vdev_t *tvd = rvd->vdev_child[c];
1750 1752 metaslab_group_t *mg = tvd->vdev_mg;
1751 1753
1752 1754 if (tvd->vdev_islog)
1753 1755 metaslab_group_activate(mg);
1754 1756 }
1755 1757 }
1756 1758
1757 1759 int
1758 1760 spa_offline_log(spa_t *spa)
1759 1761 {
1760 1762 int error;
1761 1763
1762 1764 error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1763 1765 NULL, DS_FIND_CHILDREN);
1764 1766 if (error == 0) {
1765 1767 /*
1766 1768 * We successfully offlined the log device, sync out the
1767 1769 * current txg so that the "stubby" block can be removed
1768 1770 * by zil_sync().
1769 1771 */
1770 1772 txg_wait_synced(spa->spa_dsl_pool, 0);
1771 1773 }
1772 1774 return (error);
1773 1775 }
1774 1776
1775 1777 static void
1776 1778 spa_aux_check_removed(spa_aux_vdev_t *sav)
1777 1779 {
1778 1780 for (int i = 0; i < sav->sav_count; i++)
1779 1781 spa_check_removed(sav->sav_vdevs[i]);
1780 1782 }
1781 1783
1782 1784 void
1783 1785 spa_claim_notify(zio_t *zio)
1784 1786 {
1785 1787 spa_t *spa = zio->io_spa;
1786 1788
1787 1789 if (zio->io_error)
1788 1790 return;
1789 1791
1790 1792 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1791 1793 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1792 1794 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1793 1795 mutex_exit(&spa->spa_props_lock);
1794 1796 }
1795 1797
1796 1798 typedef struct spa_load_error {
1797 1799 uint64_t sle_meta_count;
1798 1800 uint64_t sle_data_count;
1799 1801 } spa_load_error_t;
1800 1802
1801 1803 static void
1802 1804 spa_load_verify_done(zio_t *zio)
1803 1805 {
1804 1806 blkptr_t *bp = zio->io_bp;
1805 1807 spa_load_error_t *sle = zio->io_private;
1806 1808 dmu_object_type_t type = BP_GET_TYPE(bp);
1807 1809 int error = zio->io_error;
1808 1810
1809 1811 if (error) {
1810 1812 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1811 1813 type != DMU_OT_INTENT_LOG)
1812 1814 atomic_add_64(&sle->sle_meta_count, 1);
1813 1815 else
1814 1816 atomic_add_64(&sle->sle_data_count, 1);
1815 1817 }
1816 1818 zio_data_buf_free(zio->io_data, zio->io_size);
1817 1819 }
1818 1820
1819 1821 /*ARGSUSED*/
1820 1822 static int
1821 1823 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1822 1824 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1823 1825 {
1824 1826 if (bp != NULL) {
1825 1827 zio_t *rio = arg;
1826 1828 size_t size = BP_GET_PSIZE(bp);
1827 1829 void *data = zio_data_buf_alloc(size);
1828 1830
1829 1831 zio_nowait(zio_read(rio, spa, bp, data, size,
1830 1832 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1831 1833 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1832 1834 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1833 1835 }
1834 1836 return (0);
1835 1837 }
1836 1838
1837 1839 static int
1838 1840 spa_load_verify(spa_t *spa)
1839 1841 {
1840 1842 zio_t *rio;
1841 1843 spa_load_error_t sle = { 0 };
1842 1844 zpool_rewind_policy_t policy;
1843 1845 boolean_t verify_ok = B_FALSE;
1844 1846 int error;
1845 1847
1846 1848 zpool_get_rewind_policy(spa->spa_config, &policy);
1847 1849
1848 1850 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1849 1851 return (0);
1850 1852
1851 1853 rio = zio_root(spa, NULL, &sle,
1852 1854 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1853 1855
1854 1856 error = traverse_pool(spa, spa->spa_verify_min_txg,
1855 1857 TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1856 1858
1857 1859 (void) zio_wait(rio);
1858 1860
1859 1861 spa->spa_load_meta_errors = sle.sle_meta_count;
1860 1862 spa->spa_load_data_errors = sle.sle_data_count;
1861 1863
1862 1864 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1863 1865 sle.sle_data_count <= policy.zrp_maxdata) {
1864 1866 int64_t loss = 0;
1865 1867
1866 1868 verify_ok = B_TRUE;
1867 1869 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1868 1870 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1869 1871
1870 1872 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1871 1873 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1872 1874 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1873 1875 VERIFY(nvlist_add_int64(spa->spa_load_info,
1874 1876 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1875 1877 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1876 1878 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1877 1879 } else {
1878 1880 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1879 1881 }
1880 1882
1881 1883 if (error) {
1882 1884 if (error != ENXIO && error != EIO)
1883 1885 error = SET_ERROR(EIO);
1884 1886 return (error);
1885 1887 }
1886 1888
1887 1889 return (verify_ok ? 0 : EIO);
1888 1890 }
1889 1891
1890 1892 /*
1891 1893 * Find a value in the pool props object.
1892 1894 */
1893 1895 static void
1894 1896 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1895 1897 {
1896 1898 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1897 1899 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1898 1900 }
1899 1901
1900 1902 /*
1901 1903 * Find a value in the pool directory object.
1902 1904 */
1903 1905 static int
1904 1906 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1905 1907 {
1906 1908 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1907 1909 name, sizeof (uint64_t), 1, val));
1908 1910 }
1909 1911
1910 1912 static int
1911 1913 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1912 1914 {
1913 1915 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1914 1916 return (err);
1915 1917 }
1916 1918
1917 1919 /*
1918 1920 * Fix up config after a partly-completed split. This is done with the
1919 1921 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
1920 1922 * pool have that entry in their config, but only the splitting one contains
1921 1923 * a list of all the guids of the vdevs that are being split off.
1922 1924 *
1923 1925 * This function determines what to do with that list: either rejoin
1924 1926 * all the disks to the pool, or complete the splitting process. To attempt
1925 1927 * the rejoin, each disk that is offlined is marked online again, and
1926 1928 * we do a reopen() call. If the vdev label for every disk that was
1927 1929 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1928 1930 * then we call vdev_split() on each disk, and complete the split.
1929 1931 *
1930 1932 * Otherwise we leave the config alone, with all the vdevs in place in
1931 1933 * the original pool.
1932 1934 */
1933 1935 static void
1934 1936 spa_try_repair(spa_t *spa, nvlist_t *config)
1935 1937 {
1936 1938 uint_t extracted;
1937 1939 uint64_t *glist;
1938 1940 uint_t i, gcount;
1939 1941 nvlist_t *nvl;
1940 1942 vdev_t **vd;
1941 1943 boolean_t attempt_reopen;
1942 1944
1943 1945 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1944 1946 return;
1945 1947
1946 1948 /* check that the config is complete */
1947 1949 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1948 1950 &glist, &gcount) != 0)
1949 1951 return;
1950 1952
1951 1953 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1952 1954
1953 1955 /* attempt to online all the vdevs & validate */
1954 1956 attempt_reopen = B_TRUE;
1955 1957 for (i = 0; i < gcount; i++) {
1956 1958 if (glist[i] == 0) /* vdev is hole */
1957 1959 continue;
1958 1960
1959 1961 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1960 1962 if (vd[i] == NULL) {
1961 1963 /*
1962 1964 * Don't bother attempting to reopen the disks;
1963 1965 * just do the split.
1964 1966 */
1965 1967 attempt_reopen = B_FALSE;
1966 1968 } else {
1967 1969 /* attempt to re-online it */
1968 1970 vd[i]->vdev_offline = B_FALSE;
1969 1971 }
1970 1972 }
1971 1973
1972 1974 if (attempt_reopen) {
1973 1975 vdev_reopen(spa->spa_root_vdev);
1974 1976
1975 1977 /* check each device to see what state it's in */
1976 1978 for (extracted = 0, i = 0; i < gcount; i++) {
1977 1979 if (vd[i] != NULL &&
1978 1980 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1979 1981 break;
1980 1982 ++extracted;
1981 1983 }
1982 1984 }
1983 1985
1984 1986 /*
1985 1987 * If every disk has been moved to the new pool, or if we never
1986 1988 * even attempted to look at them, then we split them off for
1987 1989 * good.
1988 1990 */
1989 1991 if (!attempt_reopen || gcount == extracted) {
1990 1992 for (i = 0; i < gcount; i++)
1991 1993 if (vd[i] != NULL)
1992 1994 vdev_split(vd[i]);
1993 1995 vdev_reopen(spa->spa_root_vdev);
1994 1996 }
1995 1997
1996 1998 kmem_free(vd, gcount * sizeof (vdev_t *));
1997 1999 }
1998 2000
1999 2001 static int
2000 2002 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
2001 2003 boolean_t mosconfig)
2002 2004 {
2003 2005 nvlist_t *config = spa->spa_config;
2004 2006 char *ereport = FM_EREPORT_ZFS_POOL;
2005 2007 char *comment;
2006 2008 int error;
2007 2009 uint64_t pool_guid;
2008 2010 nvlist_t *nvl;
2009 2011
2010 2012 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
2011 2013 return (SET_ERROR(EINVAL));
2012 2014
2013 2015 ASSERT(spa->spa_comment == NULL);
2014 2016 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2015 2017 spa->spa_comment = spa_strdup(comment);
2016 2018
2017 2019 /*
2018 2020 * Versioning wasn't explicitly added to the label until later, so if
2019 2021 * it's not present treat it as the initial version.
2020 2022 */
2021 2023 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2022 2024 &spa->spa_ubsync.ub_version) != 0)
2023 2025 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2024 2026
2025 2027 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2026 2028 &spa->spa_config_txg);
2027 2029
2028 2030 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
2029 2031 spa_guid_exists(pool_guid, 0)) {
2030 2032 error = SET_ERROR(EEXIST);
2031 2033 } else {
2032 2034 spa->spa_config_guid = pool_guid;
2033 2035
2034 2036 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
2035 2037 &nvl) == 0) {
2036 2038 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
2037 2039 KM_SLEEP) == 0);
2038 2040 }
2039 2041
2040 2042 nvlist_free(spa->spa_load_info);
2041 2043 spa->spa_load_info = fnvlist_alloc();
2042 2044
2043 2045 gethrestime(&spa->spa_loaded_ts);
2044 2046 error = spa_load_impl(spa, pool_guid, config, state, type,
2045 2047 mosconfig, &ereport);
2046 2048 }
2047 2049
2048 2050 spa->spa_minref = refcount_count(&spa->spa_refcount);
2049 2051 if (error) {
2050 2052 if (error != EEXIST) {
2051 2053 spa->spa_loaded_ts.tv_sec = 0;
2052 2054 spa->spa_loaded_ts.tv_nsec = 0;
2053 2055 }
2054 2056 if (error != EBADF) {
2055 2057 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2056 2058 }
2057 2059 }
2058 2060 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2059 2061 spa->spa_ena = 0;
2060 2062
2061 2063 return (error);
2062 2064 }
2063 2065
2064 2066 /*
2065 2067 * Load an existing storage pool, using the pool's builtin spa_config as a
2066 2068 * source of configuration information.
2067 2069 */
2068 2070 static int
2069 2071 spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2070 2072 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2071 2073 char **ereport)
2072 2074 {
2073 2075 int error = 0;
2074 2076 nvlist_t *nvroot = NULL;
2075 2077 nvlist_t *label;
2076 2078 vdev_t *rvd;
2077 2079 uberblock_t *ub = &spa->spa_uberblock;
2078 2080 uint64_t children, config_cache_txg = spa->spa_config_txg;
2079 2081 int orig_mode = spa->spa_mode;
2080 2082 int parse;
2081 2083 uint64_t obj;
2082 2084 boolean_t missing_feat_write = B_FALSE;
2083 2085
2084 2086 /*
2085 2087 * If this is an untrusted config, access the pool in read-only mode.
2086 2088 * This prevents things like resilvering recently removed devices.
2087 2089 */
2088 2090 if (!mosconfig)
2089 2091 spa->spa_mode = FREAD;
2090 2092
2091 2093 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2092 2094
2093 2095 spa->spa_load_state = state;
2094 2096
2095 2097 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2096 2098 return (SET_ERROR(EINVAL));
2097 2099
2098 2100 parse = (type == SPA_IMPORT_EXISTING ?
2099 2101 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2100 2102
2101 2103 /*
2102 2104 * Create "The Godfather" zio to hold all async IOs
2103 2105 */
2104 2106 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2105 2107 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2106 2108
2107 2109 /*
2108 2110 * Parse the configuration into a vdev tree. We explicitly set the
2109 2111 * value that will be returned by spa_version() since parsing the
2110 2112 * configuration requires knowing the version number.
2111 2113 */
2112 2114 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2113 2115 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2114 2116 spa_config_exit(spa, SCL_ALL, FTAG);
2115 2117
2116 2118 if (error != 0)
2117 2119 return (error);
2118 2120
2119 2121 ASSERT(spa->spa_root_vdev == rvd);
2120 2122
2121 2123 if (type != SPA_IMPORT_ASSEMBLE) {
2122 2124 ASSERT(spa_guid(spa) == pool_guid);
2123 2125 }
2124 2126
2125 2127 /*
2126 2128 * Try to open all vdevs, loading each label in the process.
2127 2129 */
2128 2130 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2129 2131 error = vdev_open(rvd);
2130 2132 spa_config_exit(spa, SCL_ALL, FTAG);
2131 2133 if (error != 0)
2132 2134 return (error);
2133 2135
2134 2136 /*
2135 2137 * We need to validate the vdev labels against the configuration that
2136 2138 * we have in hand, which is dependent on the setting of mosconfig. If
2137 2139 * mosconfig is true then we're validating the vdev labels based on
2138 2140 * that config. Otherwise, we're validating against the cached config
2139 2141 * (zpool.cache) that was read when we loaded the zfs module, and then
2140 2142 * later we will recursively call spa_load() and validate against
2141 2143 * the vdev config.
2142 2144 *
2143 2145 * If we're assembling a new pool that's been split off from an
2144 2146 * existing pool, the labels haven't yet been updated so we skip
2145 2147 * validation for now.
2146 2148 */
2147 2149 if (type != SPA_IMPORT_ASSEMBLE) {
2148 2150 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2149 2151 error = vdev_validate(rvd, mosconfig);
2150 2152 spa_config_exit(spa, SCL_ALL, FTAG);
2151 2153
2152 2154 if (error != 0)
2153 2155 return (error);
2154 2156
2155 2157 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2156 2158 return (SET_ERROR(ENXIO));
2157 2159 }
2158 2160
2159 2161 /*
2160 2162 * Find the best uberblock.
2161 2163 */
2162 2164 vdev_uberblock_load(rvd, ub, &label);
2163 2165
2164 2166 /*
2165 2167 * If we weren't able to find a single valid uberblock, return failure.
2166 2168 */
2167 2169 if (ub->ub_txg == 0) {
2168 2170 nvlist_free(label);
2169 2171 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2170 2172 }
2171 2173
2172 2174 /*
2173 2175 * If the pool has an unsupported version we can't open it.
2174 2176 */
2175 2177 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2176 2178 nvlist_free(label);
2177 2179 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2178 2180 }
2179 2181
2180 2182 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2181 2183 nvlist_t *features;
2182 2184
2183 2185 /*
2184 2186 * If we weren't able to find what's necessary for reading the
2185 2187 * MOS in the label, return failure.
2186 2188 */
2187 2189 if (label == NULL || nvlist_lookup_nvlist(label,
2188 2190 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2189 2191 nvlist_free(label);
2190 2192 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2191 2193 ENXIO));
2192 2194 }
2193 2195
2194 2196 /*
2195 2197 * Update our in-core representation with the definitive values
2196 2198 * from the label.
2197 2199 */
2198 2200 nvlist_free(spa->spa_label_features);
2199 2201 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2200 2202 }
2201 2203
2202 2204 nvlist_free(label);
2203 2205
2204 2206 /*
2205 2207 * Look through entries in the label nvlist's features_for_read. If
2206 2208 * there is a feature listed there which we don't understand then we
2207 2209 * cannot open a pool.
2208 2210 */
2209 2211 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2210 2212 nvlist_t *unsup_feat;
2211 2213
2212 2214 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2213 2215 0);
2214 2216
2215 2217 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2216 2218 NULL); nvp != NULL;
2217 2219 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2218 2220 if (!zfeature_is_supported(nvpair_name(nvp))) {
2219 2221 VERIFY(nvlist_add_string(unsup_feat,
2220 2222 nvpair_name(nvp), "") == 0);
2221 2223 }
2222 2224 }
2223 2225
2224 2226 if (!nvlist_empty(unsup_feat)) {
2225 2227 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2226 2228 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2227 2229 nvlist_free(unsup_feat);
2228 2230 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2229 2231 ENOTSUP));
2230 2232 }
2231 2233
2232 2234 nvlist_free(unsup_feat);
2233 2235 }
2234 2236
2235 2237 /*
2236 2238 * If the vdev guid sum doesn't match the uberblock, we have an
2237 2239 * incomplete configuration. We first check to see if the pool
2238 2240 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2239 2241 * If it is, defer the vdev_guid_sum check till later so we
2240 2242 * can handle missing vdevs.
2241 2243 */
2242 2244 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2243 2245 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2244 2246 rvd->vdev_guid_sum != ub->ub_guid_sum)
2245 2247 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2246 2248
2247 2249 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2248 2250 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2249 2251 spa_try_repair(spa, config);
2250 2252 spa_config_exit(spa, SCL_ALL, FTAG);
2251 2253 nvlist_free(spa->spa_config_splitting);
2252 2254 spa->spa_config_splitting = NULL;
2253 2255 }
2254 2256
2255 2257 /*
2256 2258 * Initialize internal SPA structures.
2257 2259 */
2258 2260 spa->spa_state = POOL_STATE_ACTIVE;
2259 2261 spa->spa_ubsync = spa->spa_uberblock;
2260 2262 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2261 2263 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2262 2264 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2263 2265 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2264 2266 spa->spa_claim_max_txg = spa->spa_first_txg;
2265 2267 spa->spa_prev_software_version = ub->ub_software_version;
2266 2268
2267 2269 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2268 2270 if (error)
2269 2271 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2270 2272 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2271 2273
2272 2274 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2273 2275 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2274 2276
2275 2277 if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2276 2278 boolean_t missing_feat_read = B_FALSE;
2277 2279 nvlist_t *unsup_feat, *enabled_feat;
2278 2280
2279 2281 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2280 2282 &spa->spa_feat_for_read_obj) != 0) {
2281 2283 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2282 2284 }
2283 2285
2284 2286 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2285 2287 &spa->spa_feat_for_write_obj) != 0) {
2286 2288 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2287 2289 }
2288 2290
2289 2291 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2290 2292 &spa->spa_feat_desc_obj) != 0) {
2291 2293 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2292 2294 }
2293 2295
2294 2296 enabled_feat = fnvlist_alloc();
2295 2297 unsup_feat = fnvlist_alloc();
2296 2298
2297 2299 if (!feature_is_supported(spa->spa_meta_objset,
2298 2300 spa->spa_feat_for_read_obj, spa->spa_feat_desc_obj,
2299 2301 unsup_feat, enabled_feat))
2300 2302 missing_feat_read = B_TRUE;
2301 2303
2302 2304 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2303 2305 if (!feature_is_supported(spa->spa_meta_objset,
2304 2306 spa->spa_feat_for_write_obj, spa->spa_feat_desc_obj,
2305 2307 unsup_feat, enabled_feat)) {
2306 2308 missing_feat_write = B_TRUE;
2307 2309 }
2308 2310 }
2309 2311
2310 2312 fnvlist_add_nvlist(spa->spa_load_info,
2311 2313 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2312 2314
2313 2315 if (!nvlist_empty(unsup_feat)) {
2314 2316 fnvlist_add_nvlist(spa->spa_load_info,
2315 2317 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2316 2318 }
2317 2319
2318 2320 fnvlist_free(enabled_feat);
2319 2321 fnvlist_free(unsup_feat);
2320 2322
2321 2323 if (!missing_feat_read) {
2322 2324 fnvlist_add_boolean(spa->spa_load_info,
2323 2325 ZPOOL_CONFIG_CAN_RDONLY);
2324 2326 }
2325 2327
2326 2328 /*
2327 2329 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2328 2330 * twofold: to determine whether the pool is available for
2329 2331 * import in read-write mode and (if it is not) whether the
2330 2332 * pool is available for import in read-only mode. If the pool
2331 2333 * is available for import in read-write mode, it is displayed
2332 2334 * as available in userland; if it is not available for import
2333 2335 * in read-only mode, it is displayed as unavailable in
2334 2336 * userland. If the pool is available for import in read-only
2335 2337 * mode but not read-write mode, it is displayed as unavailable
2336 2338 * in userland with a special note that the pool is actually
2337 2339 * available for open in read-only mode.
2338 2340 *
2339 2341 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2340 2342 * missing a feature for write, we must first determine whether
2341 2343 * the pool can be opened read-only before returning to
2342 2344 * userland in order to know whether to display the
2343 2345 * abovementioned note.
2344 2346 */
2345 2347 if (missing_feat_read || (missing_feat_write &&
2346 2348 spa_writeable(spa))) {
2347 2349 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2348 2350 ENOTSUP));
2349 2351 }
2350 2352 }
2351 2353
2352 2354 spa->spa_is_initializing = B_TRUE;
2353 2355 error = dsl_pool_open(spa->spa_dsl_pool);
2354 2356 spa->spa_is_initializing = B_FALSE;
2355 2357 if (error != 0)
2356 2358 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2357 2359
2358 2360 if (!mosconfig) {
2359 2361 uint64_t hostid;
2360 2362 nvlist_t *policy = NULL, *nvconfig;
2361 2363
2362 2364 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2363 2365 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2364 2366
2365 2367 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2366 2368 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2367 2369 char *hostname;
2368 2370 unsigned long myhostid = 0;
2369 2371
2370 2372 VERIFY(nvlist_lookup_string(nvconfig,
2371 2373 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2372 2374
2373 2375 #ifdef _KERNEL
2374 2376 myhostid = zone_get_hostid(NULL);
2375 2377 #else /* _KERNEL */
2376 2378 /*
2377 2379 * We're emulating the system's hostid in userland, so
2378 2380 * we can't use zone_get_hostid().
2379 2381 */
2380 2382 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2381 2383 #endif /* _KERNEL */
2382 2384 if (hostid != 0 && myhostid != 0 &&
2383 2385 hostid != myhostid) {
2384 2386 nvlist_free(nvconfig);
2385 2387 cmn_err(CE_WARN, "pool '%s' could not be "
2386 2388 "loaded as it was last accessed by "
2387 2389 "another system (host: %s hostid: 0x%lx). "
2388 2390 "See: http://illumos.org/msg/ZFS-8000-EY",
2389 2391 spa_name(spa), hostname,
2390 2392 (unsigned long)hostid);
2391 2393 return (SET_ERROR(EBADF));
2392 2394 }
2393 2395 }
2394 2396 if (nvlist_lookup_nvlist(spa->spa_config,
2395 2397 ZPOOL_REWIND_POLICY, &policy) == 0)
2396 2398 VERIFY(nvlist_add_nvlist(nvconfig,
2397 2399 ZPOOL_REWIND_POLICY, policy) == 0);
2398 2400
2399 2401 spa_config_set(spa, nvconfig);
2400 2402 spa_unload(spa);
2401 2403 spa_deactivate(spa);
2402 2404 spa_activate(spa, orig_mode);
2403 2405
2404 2406 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2405 2407 }
2406 2408
2407 2409 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2408 2410 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2409 2411 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2410 2412 if (error != 0)
2411 2413 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2412 2414
2413 2415 /*
2414 2416 * Load the bit that tells us to use the new accounting function
2415 2417 * (raid-z deflation). If we have an older pool, this will not
2416 2418 * be present.
2417 2419 */
2418 2420 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2419 2421 if (error != 0 && error != ENOENT)
2420 2422 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2421 2423
2422 2424 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2423 2425 &spa->spa_creation_version);
2424 2426 if (error != 0 && error != ENOENT)
2425 2427 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2426 2428
2427 2429 /*
2428 2430 * Load the persistent error log. If we have an older pool, this will
2429 2431 * not be present.
2430 2432 */
2431 2433 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2432 2434 if (error != 0 && error != ENOENT)
2433 2435 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2434 2436
2435 2437 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2436 2438 &spa->spa_errlog_scrub);
2437 2439 if (error != 0 && error != ENOENT)
2438 2440 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2439 2441
2440 2442 /*
2441 2443 * Load the history object. If we have an older pool, this
2442 2444 * will not be present.
2443 2445 */
2444 2446 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2445 2447 if (error != 0 && error != ENOENT)
2446 2448 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2447 2449
2448 2450 /*
2449 2451 * If we're assembling the pool from the split-off vdevs of
2450 2452 * an existing pool, we don't want to attach the spares & cache
2451 2453 * devices.
2452 2454 */
2453 2455
2454 2456 /*
2455 2457 * Load any hot spares for this pool.
2456 2458 */
2457 2459 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2458 2460 if (error != 0 && error != ENOENT)
2459 2461 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2460 2462 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2461 2463 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2462 2464 if (load_nvlist(spa, spa->spa_spares.sav_object,
2463 2465 &spa->spa_spares.sav_config) != 0)
2464 2466 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2465 2467
2466 2468 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2467 2469 spa_load_spares(spa);
2468 2470 spa_config_exit(spa, SCL_ALL, FTAG);
2469 2471 } else if (error == 0) {
2470 2472 spa->spa_spares.sav_sync = B_TRUE;
2471 2473 }
2472 2474
2473 2475 /*
2474 2476 * Load any level 2 ARC devices for this pool.
2475 2477 */
2476 2478 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2477 2479 &spa->spa_l2cache.sav_object);
2478 2480 if (error != 0 && error != ENOENT)
2479 2481 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2480 2482 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2481 2483 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2482 2484 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2483 2485 &spa->spa_l2cache.sav_config) != 0)
2484 2486 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2485 2487
2486 2488 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2487 2489 spa_load_l2cache(spa);
2488 2490 spa_config_exit(spa, SCL_ALL, FTAG);
2489 2491 } else if (error == 0) {
2490 2492 spa->spa_l2cache.sav_sync = B_TRUE;
2491 2493 }
2492 2494
2493 2495 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2494 2496
2495 2497 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2496 2498 if (error && error != ENOENT)
2497 2499 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2498 2500
2499 2501 if (error == 0) {
2500 2502 uint64_t autoreplace;
2501 2503
2502 2504 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2503 2505 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2504 2506 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2505 2507 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2506 2508 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2507 2509 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2508 2510 &spa->spa_dedup_ditto);
2509 2511
2510 2512 spa->spa_autoreplace = (autoreplace != 0);
2511 2513 }
2512 2514
2513 2515 /*
2514 2516 * If the 'autoreplace' property is set, then post a resource notifying
2515 2517 * the ZFS DE that it should not issue any faults for unopenable
2516 2518 * devices. We also iterate over the vdevs, and post a sysevent for any
2517 2519 * unopenable vdevs so that the normal autoreplace handler can take
2518 2520 * over.
2519 2521 */
2520 2522 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2521 2523 spa_check_removed(spa->spa_root_vdev);
2522 2524 /*
2523 2525 * For the import case, this is done in spa_import(), because
2524 2526 * at this point we're using the spare definitions from
2525 2527 * the MOS config, not necessarily from the userland config.
2526 2528 */
2527 2529 if (state != SPA_LOAD_IMPORT) {
2528 2530 spa_aux_check_removed(&spa->spa_spares);
2529 2531 spa_aux_check_removed(&spa->spa_l2cache);
2530 2532 }
2531 2533 }
2532 2534
2533 2535 /*
2534 2536 * Load the vdev state for all toplevel vdevs.
2535 2537 */
2536 2538 vdev_load(rvd);
2537 2539
2538 2540 /*
2539 2541 * Propagate the leaf DTLs we just loaded all the way up the tree.
2540 2542 */
2541 2543 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2542 2544 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2543 2545 spa_config_exit(spa, SCL_ALL, FTAG);
2544 2546
2545 2547 /*
2546 2548 * Load the DDTs (dedup tables).
2547 2549 */
2548 2550 error = ddt_load(spa);
2549 2551 if (error != 0)
2550 2552 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2551 2553
2552 2554 spa_update_dspace(spa);
2553 2555
2554 2556 /*
2555 2557 * Validate the config, using the MOS config to fill in any
2556 2558 * information which might be missing. If we fail to validate
2557 2559 * the config then declare the pool unfit for use. If we're
2558 2560 * assembling a pool from a split, the log is not transferred
2559 2561 * over.
2560 2562 */
2561 2563 if (type != SPA_IMPORT_ASSEMBLE) {
2562 2564 nvlist_t *nvconfig;
2563 2565
2564 2566 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2565 2567 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2566 2568
2567 2569 if (!spa_config_valid(spa, nvconfig)) {
2568 2570 nvlist_free(nvconfig);
2569 2571 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2570 2572 ENXIO));
2571 2573 }
2572 2574 nvlist_free(nvconfig);
2573 2575
2574 2576 /*
2575 2577 * Now that we've validated the config, check the state of the
2576 2578 * root vdev. If it can't be opened, it indicates one or
2577 2579 * more toplevel vdevs are faulted.
2578 2580 */
2579 2581 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2580 2582 return (SET_ERROR(ENXIO));
2581 2583
2582 2584 if (spa_check_logs(spa)) {
2583 2585 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2584 2586 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2585 2587 }
2586 2588 }
2587 2589
2588 2590 if (missing_feat_write) {
2589 2591 ASSERT(state == SPA_LOAD_TRYIMPORT);
2590 2592
2591 2593 /*
2592 2594 * At this point, we know that we can open the pool in
2593 2595 * read-only mode but not read-write mode. We now have enough
2594 2596 * information and can return to userland.
2595 2597 */
2596 2598 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2597 2599 }
2598 2600
2599 2601 /*
2600 2602 * We've successfully opened the pool, verify that we're ready
2601 2603 * to start pushing transactions.
2602 2604 */
2603 2605 if (state != SPA_LOAD_TRYIMPORT) {
2604 2606 if (error = spa_load_verify(spa))
2605 2607 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2606 2608 error));
2607 2609 }
2608 2610
2609 2611 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2610 2612 spa->spa_load_max_txg == UINT64_MAX)) {
2611 2613 dmu_tx_t *tx;
2612 2614 int need_update = B_FALSE;
2613 2615
2614 2616 ASSERT(state != SPA_LOAD_TRYIMPORT);
2615 2617
2616 2618 /*
2617 2619 * Claim log blocks that haven't been committed yet.
2618 2620 * This must all happen in a single txg.
2619 2621 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2620 2622 * invoked from zil_claim_log_block()'s i/o done callback.
2621 2623 * Price of rollback is that we abandon the log.
2622 2624 */
2623 2625 spa->spa_claiming = B_TRUE;
2624 2626
2625 2627 tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2626 2628 spa_first_txg(spa));
2627 2629 (void) dmu_objset_find(spa_name(spa),
2628 2630 zil_claim, tx, DS_FIND_CHILDREN);
2629 2631 dmu_tx_commit(tx);
2630 2632
2631 2633 spa->spa_claiming = B_FALSE;
2632 2634
2633 2635 spa_set_log_state(spa, SPA_LOG_GOOD);
2634 2636 spa->spa_sync_on = B_TRUE;
2635 2637 txg_sync_start(spa->spa_dsl_pool);
2636 2638
2637 2639 /*
2638 2640 * Wait for all claims to sync. We sync up to the highest
2639 2641 * claimed log block birth time so that claimed log blocks
2640 2642 * don't appear to be from the future. spa_claim_max_txg
2641 2643 * will have been set for us by either zil_check_log_chain()
2642 2644 * (invoked from spa_check_logs()) or zil_claim() above.
2643 2645 */
2644 2646 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2645 2647
2646 2648 /*
2647 2649 * If the config cache is stale, or we have uninitialized
2648 2650 * metaslabs (see spa_vdev_add()), then update the config.
2649 2651 *
2650 2652 * If this is a verbatim import, trust the current
2651 2653 * in-core spa_config and update the disk labels.
2652 2654 */
2653 2655 if (config_cache_txg != spa->spa_config_txg ||
2654 2656 state == SPA_LOAD_IMPORT ||
2655 2657 state == SPA_LOAD_RECOVER ||
2656 2658 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2657 2659 need_update = B_TRUE;
2658 2660
2659 2661 for (int c = 0; c < rvd->vdev_children; c++)
2660 2662 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2661 2663 need_update = B_TRUE;
2662 2664
2663 2665 /*
2664 2666 * Update the config cache asychronously in case we're the
2665 2667 * root pool, in which case the config cache isn't writable yet.
2666 2668 */
2667 2669 if (need_update)
2668 2670 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2669 2671
2670 2672 /*
2671 2673 * Check all DTLs to see if anything needs resilvering.
2672 2674 */
2673 2675 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2674 2676 vdev_resilver_needed(rvd, NULL, NULL))
2675 2677 spa_async_request(spa, SPA_ASYNC_RESILVER);
2676 2678
2677 2679 /*
2678 2680 * Log the fact that we booted up (so that we can detect if
2679 2681 * we rebooted in the middle of an operation).
2680 2682 */
2681 2683 spa_history_log_version(spa, "open");
2682 2684
2683 2685 /*
2684 2686 * Delete any inconsistent datasets.
2685 2687 */
2686 2688 (void) dmu_objset_find(spa_name(spa),
2687 2689 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2688 2690
2689 2691 /*
2690 2692 * Clean up any stale temporary dataset userrefs.
2691 2693 */
2692 2694 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2693 2695 }
2694 2696
2695 2697 return (0);
2696 2698 }
2697 2699
2698 2700 static int
2699 2701 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2700 2702 {
2701 2703 int mode = spa->spa_mode;
2702 2704
2703 2705 spa_unload(spa);
2704 2706 spa_deactivate(spa);
2705 2707
2706 2708 spa->spa_load_max_txg--;
2707 2709
2708 2710 spa_activate(spa, mode);
2709 2711 spa_async_suspend(spa);
2710 2712
2711 2713 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2712 2714 }
2713 2715
2714 2716 /*
2715 2717 * If spa_load() fails this function will try loading prior txg's. If
2716 2718 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2717 2719 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2718 2720 * function will not rewind the pool and will return the same error as
2719 2721 * spa_load().
2720 2722 */
2721 2723 static int
2722 2724 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2723 2725 uint64_t max_request, int rewind_flags)
2724 2726 {
2725 2727 nvlist_t *loadinfo = NULL;
2726 2728 nvlist_t *config = NULL;
2727 2729 int load_error, rewind_error;
2728 2730 uint64_t safe_rewind_txg;
2729 2731 uint64_t min_txg;
2730 2732
2731 2733 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2732 2734 spa->spa_load_max_txg = spa->spa_load_txg;
2733 2735 spa_set_log_state(spa, SPA_LOG_CLEAR);
2734 2736 } else {
2735 2737 spa->spa_load_max_txg = max_request;
2736 2738 }
2737 2739
2738 2740 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2739 2741 mosconfig);
2740 2742 if (load_error == 0)
2741 2743 return (0);
2742 2744
2743 2745 if (spa->spa_root_vdev != NULL)
2744 2746 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2745 2747
2746 2748 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2747 2749 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2748 2750
2749 2751 if (rewind_flags & ZPOOL_NEVER_REWIND) {
2750 2752 nvlist_free(config);
2751 2753 return (load_error);
2752 2754 }
2753 2755
2754 2756 if (state == SPA_LOAD_RECOVER) {
2755 2757 /* Price of rolling back is discarding txgs, including log */
2756 2758 spa_set_log_state(spa, SPA_LOG_CLEAR);
2757 2759 } else {
2758 2760 /*
2759 2761 * If we aren't rolling back save the load info from our first
2760 2762 * import attempt so that we can restore it after attempting
2761 2763 * to rewind.
2762 2764 */
2763 2765 loadinfo = spa->spa_load_info;
2764 2766 spa->spa_load_info = fnvlist_alloc();
2765 2767 }
2766 2768
2767 2769 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2768 2770 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2769 2771 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2770 2772 TXG_INITIAL : safe_rewind_txg;
2771 2773
2772 2774 /*
2773 2775 * Continue as long as we're finding errors, we're still within
2774 2776 * the acceptable rewind range, and we're still finding uberblocks
2775 2777 */
2776 2778 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2777 2779 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2778 2780 if (spa->spa_load_max_txg < safe_rewind_txg)
2779 2781 spa->spa_extreme_rewind = B_TRUE;
2780 2782 rewind_error = spa_load_retry(spa, state, mosconfig);
2781 2783 }
2782 2784
2783 2785 spa->spa_extreme_rewind = B_FALSE;
2784 2786 spa->spa_load_max_txg = UINT64_MAX;
2785 2787
2786 2788 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2787 2789 spa_config_set(spa, config);
2788 2790
2789 2791 if (state == SPA_LOAD_RECOVER) {
2790 2792 ASSERT3P(loadinfo, ==, NULL);
2791 2793 return (rewind_error);
2792 2794 } else {
2793 2795 /* Store the rewind info as part of the initial load info */
2794 2796 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2795 2797 spa->spa_load_info);
2796 2798
2797 2799 /* Restore the initial load info */
2798 2800 fnvlist_free(spa->spa_load_info);
2799 2801 spa->spa_load_info = loadinfo;
2800 2802
2801 2803 return (load_error);
2802 2804 }
2803 2805 }
2804 2806
2805 2807 /*
2806 2808 * Pool Open/Import
2807 2809 *
2808 2810 * The import case is identical to an open except that the configuration is sent
2809 2811 * down from userland, instead of grabbed from the configuration cache. For the
2810 2812 * case of an open, the pool configuration will exist in the
2811 2813 * POOL_STATE_UNINITIALIZED state.
2812 2814 *
2813 2815 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2814 2816 * the same time open the pool, without having to keep around the spa_t in some
2815 2817 * ambiguous state.
2816 2818 */
2817 2819 static int
2818 2820 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2819 2821 nvlist_t **config)
2820 2822 {
2821 2823 spa_t *spa;
2822 2824 spa_load_state_t state = SPA_LOAD_OPEN;
2823 2825 int error;
2824 2826 int locked = B_FALSE;
2825 2827
2826 2828 *spapp = NULL;
2827 2829
2828 2830 /*
2829 2831 * As disgusting as this is, we need to support recursive calls to this
2830 2832 * function because dsl_dir_open() is called during spa_load(), and ends
2831 2833 * up calling spa_open() again. The real fix is to figure out how to
2832 2834 * avoid dsl_dir_open() calling this in the first place.
2833 2835 */
2834 2836 if (mutex_owner(&spa_namespace_lock) != curthread) {
2835 2837 mutex_enter(&spa_namespace_lock);
2836 2838 locked = B_TRUE;
2837 2839 }
2838 2840
2839 2841 if ((spa = spa_lookup(pool)) == NULL) {
2840 2842 if (locked)
2841 2843 mutex_exit(&spa_namespace_lock);
2842 2844 return (SET_ERROR(ENOENT));
2843 2845 }
2844 2846
2845 2847 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2846 2848 zpool_rewind_policy_t policy;
2847 2849
2848 2850 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2849 2851 &policy);
2850 2852 if (policy.zrp_request & ZPOOL_DO_REWIND)
2851 2853 state = SPA_LOAD_RECOVER;
2852 2854
2853 2855 spa_activate(spa, spa_mode_global);
2854 2856
2855 2857 if (state != SPA_LOAD_RECOVER)
2856 2858 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2857 2859
2858 2860 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2859 2861 policy.zrp_request);
2860 2862
2861 2863 if (error == EBADF) {
2862 2864 /*
2863 2865 * If vdev_validate() returns failure (indicated by
2864 2866 * EBADF), it indicates that one of the vdevs indicates
2865 2867 * that the pool has been exported or destroyed. If
2866 2868 * this is the case, the config cache is out of sync and
2867 2869 * we should remove the pool from the namespace.
2868 2870 */
2869 2871 spa_unload(spa);
2870 2872 spa_deactivate(spa);
2871 2873 spa_config_sync(spa, B_TRUE, B_TRUE);
2872 2874 spa_remove(spa);
2873 2875 if (locked)
2874 2876 mutex_exit(&spa_namespace_lock);
2875 2877 return (SET_ERROR(ENOENT));
2876 2878 }
2877 2879
2878 2880 if (error) {
2879 2881 /*
2880 2882 * We can't open the pool, but we still have useful
2881 2883 * information: the state of each vdev after the
2882 2884 * attempted vdev_open(). Return this to the user.
2883 2885 */
2884 2886 if (config != NULL && spa->spa_config) {
2885 2887 VERIFY(nvlist_dup(spa->spa_config, config,
2886 2888 KM_SLEEP) == 0);
2887 2889 VERIFY(nvlist_add_nvlist(*config,
2888 2890 ZPOOL_CONFIG_LOAD_INFO,
2889 2891 spa->spa_load_info) == 0);
2890 2892 }
2891 2893 spa_unload(spa);
2892 2894 spa_deactivate(spa);
2893 2895 spa->spa_last_open_failed = error;
2894 2896 if (locked)
2895 2897 mutex_exit(&spa_namespace_lock);
2896 2898 *spapp = NULL;
2897 2899 return (error);
2898 2900 }
2899 2901 }
2900 2902
2901 2903 spa_open_ref(spa, tag);
2902 2904
2903 2905 if (config != NULL)
2904 2906 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2905 2907
2906 2908 /*
2907 2909 * If we've recovered the pool, pass back any information we
2908 2910 * gathered while doing the load.
2909 2911 */
2910 2912 if (state == SPA_LOAD_RECOVER) {
2911 2913 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2912 2914 spa->spa_load_info) == 0);
2913 2915 }
2914 2916
2915 2917 if (locked) {
2916 2918 spa->spa_last_open_failed = 0;
2917 2919 spa->spa_last_ubsync_txg = 0;
2918 2920 spa->spa_load_txg = 0;
2919 2921 mutex_exit(&spa_namespace_lock);
2920 2922 }
2921 2923
2922 2924 *spapp = spa;
2923 2925
2924 2926 return (0);
2925 2927 }
2926 2928
2927 2929 int
2928 2930 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2929 2931 nvlist_t **config)
2930 2932 {
2931 2933 return (spa_open_common(name, spapp, tag, policy, config));
2932 2934 }
2933 2935
2934 2936 int
2935 2937 spa_open(const char *name, spa_t **spapp, void *tag)
2936 2938 {
2937 2939 return (spa_open_common(name, spapp, tag, NULL, NULL));
2938 2940 }
2939 2941
2940 2942 /*
2941 2943 * Lookup the given spa_t, incrementing the inject count in the process,
2942 2944 * preventing it from being exported or destroyed.
2943 2945 */
2944 2946 spa_t *
2945 2947 spa_inject_addref(char *name)
2946 2948 {
2947 2949 spa_t *spa;
2948 2950
2949 2951 mutex_enter(&spa_namespace_lock);
2950 2952 if ((spa = spa_lookup(name)) == NULL) {
2951 2953 mutex_exit(&spa_namespace_lock);
2952 2954 return (NULL);
2953 2955 }
2954 2956 spa->spa_inject_ref++;
2955 2957 mutex_exit(&spa_namespace_lock);
2956 2958
2957 2959 return (spa);
2958 2960 }
2959 2961
2960 2962 void
2961 2963 spa_inject_delref(spa_t *spa)
2962 2964 {
2963 2965 mutex_enter(&spa_namespace_lock);
2964 2966 spa->spa_inject_ref--;
2965 2967 mutex_exit(&spa_namespace_lock);
2966 2968 }
2967 2969
2968 2970 /*
2969 2971 * Add spares device information to the nvlist.
2970 2972 */
2971 2973 static void
2972 2974 spa_add_spares(spa_t *spa, nvlist_t *config)
2973 2975 {
2974 2976 nvlist_t **spares;
2975 2977 uint_t i, nspares;
2976 2978 nvlist_t *nvroot;
2977 2979 uint64_t guid;
2978 2980 vdev_stat_t *vs;
2979 2981 uint_t vsc;
2980 2982 uint64_t pool;
2981 2983
2982 2984 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2983 2985
2984 2986 if (spa->spa_spares.sav_count == 0)
2985 2987 return;
2986 2988
2987 2989 VERIFY(nvlist_lookup_nvlist(config,
2988 2990 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2989 2991 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2990 2992 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2991 2993 if (nspares != 0) {
2992 2994 VERIFY(nvlist_add_nvlist_array(nvroot,
2993 2995 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2994 2996 VERIFY(nvlist_lookup_nvlist_array(nvroot,
2995 2997 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2996 2998
2997 2999 /*
2998 3000 * Go through and find any spares which have since been
2999 3001 * repurposed as an active spare. If this is the case, update
3000 3002 * their status appropriately.
3001 3003 */
3002 3004 for (i = 0; i < nspares; i++) {
3003 3005 VERIFY(nvlist_lookup_uint64(spares[i],
3004 3006 ZPOOL_CONFIG_GUID, &guid) == 0);
3005 3007 if (spa_spare_exists(guid, &pool, NULL) &&
3006 3008 pool != 0ULL) {
3007 3009 VERIFY(nvlist_lookup_uint64_array(
3008 3010 spares[i], ZPOOL_CONFIG_VDEV_STATS,
3009 3011 (uint64_t **)&vs, &vsc) == 0);
3010 3012 vs->vs_state = VDEV_STATE_CANT_OPEN;
3011 3013 vs->vs_aux = VDEV_AUX_SPARED;
3012 3014 }
3013 3015 }
3014 3016 }
3015 3017 }
3016 3018
3017 3019 /*
3018 3020 * Add l2cache device information to the nvlist, including vdev stats.
3019 3021 */
3020 3022 static void
3021 3023 spa_add_l2cache(spa_t *spa, nvlist_t *config)
3022 3024 {
3023 3025 nvlist_t **l2cache;
3024 3026 uint_t i, j, nl2cache;
3025 3027 nvlist_t *nvroot;
3026 3028 uint64_t guid;
3027 3029 vdev_t *vd;
3028 3030 vdev_stat_t *vs;
3029 3031 uint_t vsc;
3030 3032
3031 3033 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3032 3034
3033 3035 if (spa->spa_l2cache.sav_count == 0)
3034 3036 return;
3035 3037
3036 3038 VERIFY(nvlist_lookup_nvlist(config,
3037 3039 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3038 3040 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3039 3041 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3040 3042 if (nl2cache != 0) {
3041 3043 VERIFY(nvlist_add_nvlist_array(nvroot,
3042 3044 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3043 3045 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3044 3046 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3045 3047
3046 3048 /*
3047 3049 * Update level 2 cache device stats.
3048 3050 */
3049 3051
3050 3052 for (i = 0; i < nl2cache; i++) {
3051 3053 VERIFY(nvlist_lookup_uint64(l2cache[i],
3052 3054 ZPOOL_CONFIG_GUID, &guid) == 0);
3053 3055
3054 3056 vd = NULL;
3055 3057 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3056 3058 if (guid ==
3057 3059 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3058 3060 vd = spa->spa_l2cache.sav_vdevs[j];
3059 3061 break;
3060 3062 }
3061 3063 }
3062 3064 ASSERT(vd != NULL);
3063 3065
3064 3066 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3065 3067 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3066 3068 == 0);
3067 3069 vdev_get_stats(vd, vs);
3068 3070 }
3069 3071 }
3070 3072 }
3071 3073
3072 3074 static void
3073 3075 spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3074 3076 {
3075 3077 nvlist_t *features;
3076 3078 zap_cursor_t zc;
3077 3079 zap_attribute_t za;
3078 3080
3079 3081 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3080 3082 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3081 3083
3082 3084 if (spa->spa_feat_for_read_obj != 0) {
3083 3085 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3084 3086 spa->spa_feat_for_read_obj);
3085 3087 zap_cursor_retrieve(&zc, &za) == 0;
3086 3088 zap_cursor_advance(&zc)) {
3087 3089 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3088 3090 za.za_num_integers == 1);
3089 3091 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3090 3092 za.za_first_integer));
3091 3093 }
3092 3094 zap_cursor_fini(&zc);
3093 3095 }
3094 3096
3095 3097 if (spa->spa_feat_for_write_obj != 0) {
3096 3098 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3097 3099 spa->spa_feat_for_write_obj);
3098 3100 zap_cursor_retrieve(&zc, &za) == 0;
3099 3101 zap_cursor_advance(&zc)) {
3100 3102 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3101 3103 za.za_num_integers == 1);
3102 3104 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3103 3105 za.za_first_integer));
3104 3106 }
3105 3107 zap_cursor_fini(&zc);
3106 3108 }
3107 3109
3108 3110 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3109 3111 features) == 0);
3110 3112 nvlist_free(features);
3111 3113 }
3112 3114
3113 3115 int
3114 3116 spa_get_stats(const char *name, nvlist_t **config,
3115 3117 char *altroot, size_t buflen)
3116 3118 {
3117 3119 int error;
3118 3120 spa_t *spa;
3119 3121
3120 3122 *config = NULL;
3121 3123 error = spa_open_common(name, &spa, FTAG, NULL, config);
3122 3124
3123 3125 if (spa != NULL) {
3124 3126 /*
3125 3127 * This still leaves a window of inconsistency where the spares
3126 3128 * or l2cache devices could change and the config would be
3127 3129 * self-inconsistent.
3128 3130 */
3129 3131 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3130 3132
3131 3133 if (*config != NULL) {
3132 3134 uint64_t loadtimes[2];
3133 3135
3134 3136 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3135 3137 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3136 3138 VERIFY(nvlist_add_uint64_array(*config,
3137 3139 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3138 3140
3139 3141 VERIFY(nvlist_add_uint64(*config,
3140 3142 ZPOOL_CONFIG_ERRCOUNT,
3141 3143 spa_get_errlog_size(spa)) == 0);
3142 3144
3143 3145 if (spa_suspended(spa))
3144 3146 VERIFY(nvlist_add_uint64(*config,
3145 3147 ZPOOL_CONFIG_SUSPENDED,
3146 3148 spa->spa_failmode) == 0);
3147 3149
3148 3150 spa_add_spares(spa, *config);
3149 3151 spa_add_l2cache(spa, *config);
3150 3152 spa_add_feature_stats(spa, *config);
3151 3153 }
3152 3154 }
3153 3155
3154 3156 /*
3155 3157 * We want to get the alternate root even for faulted pools, so we cheat
3156 3158 * and call spa_lookup() directly.
3157 3159 */
3158 3160 if (altroot) {
3159 3161 if (spa == NULL) {
3160 3162 mutex_enter(&spa_namespace_lock);
3161 3163 spa = spa_lookup(name);
3162 3164 if (spa)
3163 3165 spa_altroot(spa, altroot, buflen);
3164 3166 else
3165 3167 altroot[0] = '\0';
3166 3168 spa = NULL;
3167 3169 mutex_exit(&spa_namespace_lock);
3168 3170 } else {
3169 3171 spa_altroot(spa, altroot, buflen);
3170 3172 }
3171 3173 }
3172 3174
3173 3175 if (spa != NULL) {
3174 3176 spa_config_exit(spa, SCL_CONFIG, FTAG);
3175 3177 spa_close(spa, FTAG);
3176 3178 }
3177 3179
3178 3180 return (error);
3179 3181 }
3180 3182
3181 3183 /*
3182 3184 * Validate that the auxiliary device array is well formed. We must have an
3183 3185 * array of nvlists, each which describes a valid leaf vdev. If this is an
3184 3186 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3185 3187 * specified, as long as they are well-formed.
3186 3188 */
3187 3189 static int
3188 3190 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3189 3191 spa_aux_vdev_t *sav, const char *config, uint64_t version,
3190 3192 vdev_labeltype_t label)
3191 3193 {
3192 3194 nvlist_t **dev;
3193 3195 uint_t i, ndev;
3194 3196 vdev_t *vd;
3195 3197 int error;
3196 3198
3197 3199 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3198 3200
3199 3201 /*
3200 3202 * It's acceptable to have no devs specified.
3201 3203 */
3202 3204 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3203 3205 return (0);
3204 3206
3205 3207 if (ndev == 0)
3206 3208 return (SET_ERROR(EINVAL));
3207 3209
3208 3210 /*
3209 3211 * Make sure the pool is formatted with a version that supports this
3210 3212 * device type.
3211 3213 */
3212 3214 if (spa_version(spa) < version)
3213 3215 return (SET_ERROR(ENOTSUP));
3214 3216
3215 3217 /*
3216 3218 * Set the pending device list so we correctly handle device in-use
3217 3219 * checking.
3218 3220 */
3219 3221 sav->sav_pending = dev;
3220 3222 sav->sav_npending = ndev;
3221 3223
3222 3224 for (i = 0; i < ndev; i++) {
3223 3225 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3224 3226 mode)) != 0)
3225 3227 goto out;
3226 3228
3227 3229 if (!vd->vdev_ops->vdev_op_leaf) {
3228 3230 vdev_free(vd);
3229 3231 error = SET_ERROR(EINVAL);
3230 3232 goto out;
3231 3233 }
3232 3234
3233 3235 /*
3234 3236 * The L2ARC currently only supports disk devices in
3235 3237 * kernel context. For user-level testing, we allow it.
3236 3238 */
3237 3239 #ifdef _KERNEL
3238 3240 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3239 3241 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3240 3242 error = SET_ERROR(ENOTBLK);
3241 3243 vdev_free(vd);
3242 3244 goto out;
3243 3245 }
3244 3246 #endif
3245 3247 vd->vdev_top = vd;
3246 3248
3247 3249 if ((error = vdev_open(vd)) == 0 &&
3248 3250 (error = vdev_label_init(vd, crtxg, label)) == 0) {
3249 3251 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3250 3252 vd->vdev_guid) == 0);
3251 3253 }
3252 3254
3253 3255 vdev_free(vd);
3254 3256
3255 3257 if (error &&
3256 3258 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3257 3259 goto out;
3258 3260 else
3259 3261 error = 0;
3260 3262 }
3261 3263
3262 3264 out:
3263 3265 sav->sav_pending = NULL;
3264 3266 sav->sav_npending = 0;
3265 3267 return (error);
3266 3268 }
3267 3269
3268 3270 static int
3269 3271 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3270 3272 {
3271 3273 int error;
3272 3274
3273 3275 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3274 3276
3275 3277 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3276 3278 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3277 3279 VDEV_LABEL_SPARE)) != 0) {
3278 3280 return (error);
3279 3281 }
3280 3282
3281 3283 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3282 3284 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3283 3285 VDEV_LABEL_L2CACHE));
3284 3286 }
3285 3287
3286 3288 static void
3287 3289 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3288 3290 const char *config)
3289 3291 {
3290 3292 int i;
3291 3293
3292 3294 if (sav->sav_config != NULL) {
3293 3295 nvlist_t **olddevs;
3294 3296 uint_t oldndevs;
3295 3297 nvlist_t **newdevs;
3296 3298
3297 3299 /*
3298 3300 * Generate new dev list by concatentating with the
3299 3301 * current dev list.
3300 3302 */
3301 3303 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3302 3304 &olddevs, &oldndevs) == 0);
3303 3305
3304 3306 newdevs = kmem_alloc(sizeof (void *) *
3305 3307 (ndevs + oldndevs), KM_SLEEP);
3306 3308 for (i = 0; i < oldndevs; i++)
3307 3309 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3308 3310 KM_SLEEP) == 0);
3309 3311 for (i = 0; i < ndevs; i++)
3310 3312 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3311 3313 KM_SLEEP) == 0);
3312 3314
3313 3315 VERIFY(nvlist_remove(sav->sav_config, config,
3314 3316 DATA_TYPE_NVLIST_ARRAY) == 0);
3315 3317
3316 3318 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3317 3319 config, newdevs, ndevs + oldndevs) == 0);
3318 3320 for (i = 0; i < oldndevs + ndevs; i++)
3319 3321 nvlist_free(newdevs[i]);
3320 3322 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3321 3323 } else {
3322 3324 /*
3323 3325 * Generate a new dev list.
3324 3326 */
3325 3327 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3326 3328 KM_SLEEP) == 0);
3327 3329 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3328 3330 devs, ndevs) == 0);
3329 3331 }
3330 3332 }
3331 3333
3332 3334 /*
3333 3335 * Stop and drop level 2 ARC devices
3334 3336 */
3335 3337 void
3336 3338 spa_l2cache_drop(spa_t *spa)
3337 3339 {
3338 3340 vdev_t *vd;
3339 3341 int i;
3340 3342 spa_aux_vdev_t *sav = &spa->spa_l2cache;
3341 3343
3342 3344 for (i = 0; i < sav->sav_count; i++) {
3343 3345 uint64_t pool;
3344 3346
3345 3347 vd = sav->sav_vdevs[i];
3346 3348 ASSERT(vd != NULL);
3347 3349
3348 3350 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3349 3351 pool != 0ULL && l2arc_vdev_present(vd))
3350 3352 l2arc_remove_vdev(vd);
3351 3353 }
3352 3354 }
3353 3355
3354 3356 /*
3355 3357 * Pool Creation
3356 3358 */
3357 3359 int
3358 3360 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3359 3361 nvlist_t *zplprops)
3360 3362 {
3361 3363 spa_t *spa;
3362 3364 char *altroot = NULL;
3363 3365 vdev_t *rvd;
3364 3366 dsl_pool_t *dp;
3365 3367 dmu_tx_t *tx;
3366 3368 int error = 0;
3367 3369 uint64_t txg = TXG_INITIAL;
3368 3370 nvlist_t **spares, **l2cache;
3369 3371 uint_t nspares, nl2cache;
3370 3372 uint64_t version, obj;
3371 3373 boolean_t has_features;
3372 3374
3373 3375 /*
3374 3376 * If this pool already exists, return failure.
3375 3377 */
3376 3378 mutex_enter(&spa_namespace_lock);
3377 3379 if (spa_lookup(pool) != NULL) {
3378 3380 mutex_exit(&spa_namespace_lock);
3379 3381 return (SET_ERROR(EEXIST));
3380 3382 }
3381 3383
3382 3384 /*
3383 3385 * Allocate a new spa_t structure.
3384 3386 */
3385 3387 (void) nvlist_lookup_string(props,
3386 3388 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3387 3389 spa = spa_add(pool, NULL, altroot);
3388 3390 spa_activate(spa, spa_mode_global);
3389 3391
3390 3392 if (props && (error = spa_prop_validate(spa, props))) {
3391 3393 spa_deactivate(spa);
3392 3394 spa_remove(spa);
3393 3395 mutex_exit(&spa_namespace_lock);
3394 3396 return (error);
3395 3397 }
3396 3398
3397 3399 has_features = B_FALSE;
3398 3400 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3399 3401 elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3400 3402 if (zpool_prop_feature(nvpair_name(elem)))
3401 3403 has_features = B_TRUE;
3402 3404 }
3403 3405
3404 3406 if (has_features || nvlist_lookup_uint64(props,
3405 3407 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3406 3408 version = SPA_VERSION;
3407 3409 }
3408 3410 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3409 3411
3410 3412 spa->spa_first_txg = txg;
3411 3413 spa->spa_uberblock.ub_txg = txg - 1;
3412 3414 spa->spa_uberblock.ub_version = version;
3413 3415 spa->spa_ubsync = spa->spa_uberblock;
3414 3416
3415 3417 /*
3416 3418 * Create "The Godfather" zio to hold all async IOs
3417 3419 */
3418 3420 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
3419 3421 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
3420 3422
3421 3423 /*
3422 3424 * Create the root vdev.
3423 3425 */
3424 3426 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3425 3427
3426 3428 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3427 3429
3428 3430 ASSERT(error != 0 || rvd != NULL);
3429 3431 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3430 3432
3431 3433 if (error == 0 && !zfs_allocatable_devs(nvroot))
3432 3434 error = SET_ERROR(EINVAL);
3433 3435
3434 3436 if (error == 0 &&
3435 3437 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3436 3438 (error = spa_validate_aux(spa, nvroot, txg,
3437 3439 VDEV_ALLOC_ADD)) == 0) {
3438 3440 for (int c = 0; c < rvd->vdev_children; c++) {
3439 3441 vdev_metaslab_set_size(rvd->vdev_child[c]);
3440 3442 vdev_expand(rvd->vdev_child[c], txg);
3441 3443 }
3442 3444 }
3443 3445
3444 3446 spa_config_exit(spa, SCL_ALL, FTAG);
3445 3447
3446 3448 if (error != 0) {
3447 3449 spa_unload(spa);
3448 3450 spa_deactivate(spa);
3449 3451 spa_remove(spa);
3450 3452 mutex_exit(&spa_namespace_lock);
3451 3453 return (error);
3452 3454 }
3453 3455
3454 3456 /*
3455 3457 * Get the list of spares, if specified.
3456 3458 */
3457 3459 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3458 3460 &spares, &nspares) == 0) {
3459 3461 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3460 3462 KM_SLEEP) == 0);
3461 3463 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3462 3464 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3463 3465 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3464 3466 spa_load_spares(spa);
3465 3467 spa_config_exit(spa, SCL_ALL, FTAG);
3466 3468 spa->spa_spares.sav_sync = B_TRUE;
3467 3469 }
3468 3470
3469 3471 /*
3470 3472 * Get the list of level 2 cache devices, if specified.
3471 3473 */
3472 3474 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3473 3475 &l2cache, &nl2cache) == 0) {
3474 3476 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3475 3477 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3476 3478 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3477 3479 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3478 3480 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3479 3481 spa_load_l2cache(spa);
3480 3482 spa_config_exit(spa, SCL_ALL, FTAG);
3481 3483 spa->spa_l2cache.sav_sync = B_TRUE;
3482 3484 }
3483 3485
3484 3486 spa->spa_is_initializing = B_TRUE;
3485 3487 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3486 3488 spa->spa_meta_objset = dp->dp_meta_objset;
3487 3489 spa->spa_is_initializing = B_FALSE;
3488 3490
3489 3491 /*
3490 3492 * Create DDTs (dedup tables).
3491 3493 */
3492 3494 ddt_create(spa);
3493 3495
3494 3496 spa_update_dspace(spa);
3495 3497
3496 3498 tx = dmu_tx_create_assigned(dp, txg);
3497 3499
3498 3500 /*
3499 3501 * Create the pool config object.
3500 3502 */
3501 3503 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3502 3504 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3503 3505 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3504 3506
3505 3507 if (zap_add(spa->spa_meta_objset,
3506 3508 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3507 3509 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3508 3510 cmn_err(CE_PANIC, "failed to add pool config");
3509 3511 }
3510 3512
3511 3513 if (spa_version(spa) >= SPA_VERSION_FEATURES)
3512 3514 spa_feature_create_zap_objects(spa, tx);
3513 3515
3514 3516 if (zap_add(spa->spa_meta_objset,
3515 3517 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3516 3518 sizeof (uint64_t), 1, &version, tx) != 0) {
3517 3519 cmn_err(CE_PANIC, "failed to add pool version");
3518 3520 }
3519 3521
3520 3522 /* Newly created pools with the right version are always deflated. */
3521 3523 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3522 3524 spa->spa_deflate = TRUE;
3523 3525 if (zap_add(spa->spa_meta_objset,
3524 3526 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3525 3527 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3526 3528 cmn_err(CE_PANIC, "failed to add deflate");
3527 3529 }
3528 3530 }
3529 3531
3530 3532 /*
3531 3533 * Create the deferred-free bpobj. Turn off compression
3532 3534 * because sync-to-convergence takes longer if the blocksize
3533 3535 * keeps changing.
3534 3536 */
3535 3537 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3536 3538 dmu_object_set_compress(spa->spa_meta_objset, obj,
3537 3539 ZIO_COMPRESS_OFF, tx);
3538 3540 if (zap_add(spa->spa_meta_objset,
3539 3541 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3540 3542 sizeof (uint64_t), 1, &obj, tx) != 0) {
3541 3543 cmn_err(CE_PANIC, "failed to add bpobj");
3542 3544 }
3543 3545 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3544 3546 spa->spa_meta_objset, obj));
3545 3547
3546 3548 /*
3547 3549 * Create the pool's history object.
3548 3550 */
3549 3551 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3550 3552 spa_history_create_obj(spa, tx);
3551 3553
3552 3554 /*
3553 3555 * Set pool properties.
3554 3556 */
3555 3557 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3556 3558 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3557 3559 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3558 3560 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3559 3561
3560 3562 if (props != NULL) {
3561 3563 spa_configfile_set(spa, props, B_FALSE);
3562 3564 spa_sync_props(props, tx);
3563 3565 }
3564 3566
3565 3567 dmu_tx_commit(tx);
3566 3568
3567 3569 spa->spa_sync_on = B_TRUE;
3568 3570 txg_sync_start(spa->spa_dsl_pool);
3569 3571
3570 3572 /*
3571 3573 * We explicitly wait for the first transaction to complete so that our
3572 3574 * bean counters are appropriately updated.
3573 3575 */
3574 3576 txg_wait_synced(spa->spa_dsl_pool, txg);
3575 3577
3576 3578 spa_config_sync(spa, B_FALSE, B_TRUE);
3577 3579
3578 3580 spa_history_log_version(spa, "create");
3579 3581
3580 3582 spa->spa_minref = refcount_count(&spa->spa_refcount);
3581 3583
3582 3584 mutex_exit(&spa_namespace_lock);
3583 3585
3584 3586 return (0);
3585 3587 }
3586 3588
3587 3589 #ifdef _KERNEL
3588 3590 /*
3589 3591 * Get the root pool information from the root disk, then import the root pool
3590 3592 * during the system boot up time.
3591 3593 */
3592 3594 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3593 3595
3594 3596 static nvlist_t *
3595 3597 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3596 3598 {
3597 3599 nvlist_t *config;
3598 3600 nvlist_t *nvtop, *nvroot;
3599 3601 uint64_t pgid;
3600 3602
3601 3603 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3602 3604 return (NULL);
3603 3605
3604 3606 /*
3605 3607 * Add this top-level vdev to the child array.
3606 3608 */
3607 3609 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3608 3610 &nvtop) == 0);
3609 3611 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3610 3612 &pgid) == 0);
3611 3613 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3612 3614
3613 3615 /*
3614 3616 * Put this pool's top-level vdevs into a root vdev.
3615 3617 */
3616 3618 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3617 3619 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3618 3620 VDEV_TYPE_ROOT) == 0);
3619 3621 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3620 3622 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3621 3623 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3622 3624 &nvtop, 1) == 0);
3623 3625
3624 3626 /*
3625 3627 * Replace the existing vdev_tree with the new root vdev in
3626 3628 * this pool's configuration (remove the old, add the new).
3627 3629 */
3628 3630 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3629 3631 nvlist_free(nvroot);
3630 3632 return (config);
3631 3633 }
3632 3634
3633 3635 /*
3634 3636 * Walk the vdev tree and see if we can find a device with "better"
3635 3637 * configuration. A configuration is "better" if the label on that
3636 3638 * device has a more recent txg.
3637 3639 */
3638 3640 static void
3639 3641 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3640 3642 {
3641 3643 for (int c = 0; c < vd->vdev_children; c++)
3642 3644 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3643 3645
3644 3646 if (vd->vdev_ops->vdev_op_leaf) {
3645 3647 nvlist_t *label;
3646 3648 uint64_t label_txg;
3647 3649
3648 3650 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3649 3651 &label) != 0)
3650 3652 return;
3651 3653
3652 3654 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3653 3655 &label_txg) == 0);
3654 3656
3655 3657 /*
3656 3658 * Do we have a better boot device?
3657 3659 */
3658 3660 if (label_txg > *txg) {
3659 3661 *txg = label_txg;
3660 3662 *avd = vd;
3661 3663 }
3662 3664 nvlist_free(label);
3663 3665 }
3664 3666 }
3665 3667
3666 3668 /*
3667 3669 * Import a root pool.
3668 3670 *
3669 3671 * For x86. devpath_list will consist of devid and/or physpath name of
3670 3672 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3671 3673 * The GRUB "findroot" command will return the vdev we should boot.
3672 3674 *
3673 3675 * For Sparc, devpath_list consists the physpath name of the booting device
3674 3676 * no matter the rootpool is a single device pool or a mirrored pool.
3675 3677 * e.g.
3676 3678 * "/pci@1f,0/ide@d/disk@0,0:a"
3677 3679 */
3678 3680 int
3679 3681 spa_import_rootpool(char *devpath, char *devid)
3680 3682 {
3681 3683 spa_t *spa;
3682 3684 vdev_t *rvd, *bvd, *avd = NULL;
3683 3685 nvlist_t *config, *nvtop;
3684 3686 uint64_t guid, txg;
3685 3687 char *pname;
3686 3688 int error;
3687 3689
3688 3690 /*
3689 3691 * Read the label from the boot device and generate a configuration.
3690 3692 */
3691 3693 config = spa_generate_rootconf(devpath, devid, &guid);
3692 3694 #if defined(_OBP) && defined(_KERNEL)
3693 3695 if (config == NULL) {
3694 3696 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3695 3697 /* iscsi boot */
3696 3698 get_iscsi_bootpath_phy(devpath);
3697 3699 config = spa_generate_rootconf(devpath, devid, &guid);
3698 3700 }
3699 3701 }
3700 3702 #endif
3701 3703 if (config == NULL) {
3702 3704 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3703 3705 devpath);
3704 3706 return (SET_ERROR(EIO));
3705 3707 }
3706 3708
3707 3709 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3708 3710 &pname) == 0);
3709 3711 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3710 3712
3711 3713 mutex_enter(&spa_namespace_lock);
3712 3714 if ((spa = spa_lookup(pname)) != NULL) {
3713 3715 /*
3714 3716 * Remove the existing root pool from the namespace so that we
3715 3717 * can replace it with the correct config we just read in.
3716 3718 */
3717 3719 spa_remove(spa);
3718 3720 }
3719 3721
3720 3722 spa = spa_add(pname, config, NULL);
3721 3723 spa->spa_is_root = B_TRUE;
3722 3724 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3723 3725
3724 3726 /*
3725 3727 * Build up a vdev tree based on the boot device's label config.
3726 3728 */
3727 3729 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3728 3730 &nvtop) == 0);
3729 3731 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3730 3732 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3731 3733 VDEV_ALLOC_ROOTPOOL);
3732 3734 spa_config_exit(spa, SCL_ALL, FTAG);
3733 3735 if (error) {
3734 3736 mutex_exit(&spa_namespace_lock);
3735 3737 nvlist_free(config);
3736 3738 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3737 3739 pname);
3738 3740 return (error);
3739 3741 }
3740 3742
3741 3743 /*
3742 3744 * Get the boot vdev.
3743 3745 */
3744 3746 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3745 3747 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3746 3748 (u_longlong_t)guid);
3747 3749 error = SET_ERROR(ENOENT);
3748 3750 goto out;
3749 3751 }
3750 3752
3751 3753 /*
3752 3754 * Determine if there is a better boot device.
3753 3755 */
3754 3756 avd = bvd;
3755 3757 spa_alt_rootvdev(rvd, &avd, &txg);
3756 3758 if (avd != bvd) {
3757 3759 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3758 3760 "try booting from '%s'", avd->vdev_path);
3759 3761 error = SET_ERROR(EINVAL);
3760 3762 goto out;
3761 3763 }
3762 3764
3763 3765 /*
3764 3766 * If the boot device is part of a spare vdev then ensure that
3765 3767 * we're booting off the active spare.
3766 3768 */
3767 3769 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3768 3770 !bvd->vdev_isspare) {
3769 3771 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3770 3772 "try booting from '%s'",
3771 3773 bvd->vdev_parent->
3772 3774 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3773 3775 error = SET_ERROR(EINVAL);
3774 3776 goto out;
3775 3777 }
3776 3778
3777 3779 error = 0;
3778 3780 out:
3779 3781 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3780 3782 vdev_free(rvd);
3781 3783 spa_config_exit(spa, SCL_ALL, FTAG);
3782 3784 mutex_exit(&spa_namespace_lock);
3783 3785
3784 3786 nvlist_free(config);
3785 3787 return (error);
3786 3788 }
3787 3789
3788 3790 #endif
3789 3791
3790 3792 /*
3791 3793 * Import a non-root pool into the system.
3792 3794 */
3793 3795 int
3794 3796 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3795 3797 {
3796 3798 spa_t *spa;
3797 3799 char *altroot = NULL;
3798 3800 spa_load_state_t state = SPA_LOAD_IMPORT;
3799 3801 zpool_rewind_policy_t policy;
3800 3802 uint64_t mode = spa_mode_global;
3801 3803 uint64_t readonly = B_FALSE;
3802 3804 int error;
3803 3805 nvlist_t *nvroot;
3804 3806 nvlist_t **spares, **l2cache;
3805 3807 uint_t nspares, nl2cache;
3806 3808
3807 3809 /*
3808 3810 * If a pool with this name exists, return failure.
3809 3811 */
3810 3812 mutex_enter(&spa_namespace_lock);
3811 3813 if (spa_lookup(pool) != NULL) {
3812 3814 mutex_exit(&spa_namespace_lock);
3813 3815 return (SET_ERROR(EEXIST));
3814 3816 }
3815 3817
3816 3818 /*
3817 3819 * Create and initialize the spa structure.
3818 3820 */
3819 3821 (void) nvlist_lookup_string(props,
3820 3822 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3821 3823 (void) nvlist_lookup_uint64(props,
3822 3824 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3823 3825 if (readonly)
3824 3826 mode = FREAD;
3825 3827 spa = spa_add(pool, config, altroot);
3826 3828 spa->spa_import_flags = flags;
3827 3829
3828 3830 /*
3829 3831 * Verbatim import - Take a pool and insert it into the namespace
3830 3832 * as if it had been loaded at boot.
3831 3833 */
3832 3834 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3833 3835 if (props != NULL)
3834 3836 spa_configfile_set(spa, props, B_FALSE);
3835 3837
3836 3838 spa_config_sync(spa, B_FALSE, B_TRUE);
3837 3839
3838 3840 mutex_exit(&spa_namespace_lock);
3839 3841 spa_history_log_version(spa, "import");
3840 3842
3841 3843 return (0);
3842 3844 }
3843 3845
3844 3846 spa_activate(spa, mode);
3845 3847
3846 3848 /*
3847 3849 * Don't start async tasks until we know everything is healthy.
3848 3850 */
3849 3851 spa_async_suspend(spa);
3850 3852
3851 3853 zpool_get_rewind_policy(config, &policy);
3852 3854 if (policy.zrp_request & ZPOOL_DO_REWIND)
3853 3855 state = SPA_LOAD_RECOVER;
3854 3856
3855 3857 /*
3856 3858 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
3857 3859 * because the user-supplied config is actually the one to trust when
3858 3860 * doing an import.
3859 3861 */
3860 3862 if (state != SPA_LOAD_RECOVER)
3861 3863 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3862 3864
3863 3865 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3864 3866 policy.zrp_request);
3865 3867
3866 3868 /*
3867 3869 * Propagate anything learned while loading the pool and pass it
3868 3870 * back to caller (i.e. rewind info, missing devices, etc).
3869 3871 */
3870 3872 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3871 3873 spa->spa_load_info) == 0);
3872 3874
3873 3875 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3874 3876 /*
3875 3877 * Toss any existing sparelist, as it doesn't have any validity
3876 3878 * anymore, and conflicts with spa_has_spare().
3877 3879 */
3878 3880 if (spa->spa_spares.sav_config) {
3879 3881 nvlist_free(spa->spa_spares.sav_config);
3880 3882 spa->spa_spares.sav_config = NULL;
3881 3883 spa_load_spares(spa);
3882 3884 }
3883 3885 if (spa->spa_l2cache.sav_config) {
3884 3886 nvlist_free(spa->spa_l2cache.sav_config);
3885 3887 spa->spa_l2cache.sav_config = NULL;
3886 3888 spa_load_l2cache(spa);
3887 3889 }
3888 3890
3889 3891 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3890 3892 &nvroot) == 0);
3891 3893 if (error == 0)
3892 3894 error = spa_validate_aux(spa, nvroot, -1ULL,
3893 3895 VDEV_ALLOC_SPARE);
3894 3896 if (error == 0)
3895 3897 error = spa_validate_aux(spa, nvroot, -1ULL,
3896 3898 VDEV_ALLOC_L2CACHE);
3897 3899 spa_config_exit(spa, SCL_ALL, FTAG);
3898 3900
3899 3901 if (props != NULL)
3900 3902 spa_configfile_set(spa, props, B_FALSE);
3901 3903
3902 3904 if (error != 0 || (props && spa_writeable(spa) &&
3903 3905 (error = spa_prop_set(spa, props)))) {
3904 3906 spa_unload(spa);
3905 3907 spa_deactivate(spa);
3906 3908 spa_remove(spa);
3907 3909 mutex_exit(&spa_namespace_lock);
3908 3910 return (error);
3909 3911 }
3910 3912
3911 3913 spa_async_resume(spa);
3912 3914
3913 3915 /*
3914 3916 * Override any spares and level 2 cache devices as specified by
3915 3917 * the user, as these may have correct device names/devids, etc.
3916 3918 */
3917 3919 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3918 3920 &spares, &nspares) == 0) {
3919 3921 if (spa->spa_spares.sav_config)
3920 3922 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3921 3923 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3922 3924 else
3923 3925 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3924 3926 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3925 3927 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3926 3928 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3927 3929 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3928 3930 spa_load_spares(spa);
3929 3931 spa_config_exit(spa, SCL_ALL, FTAG);
3930 3932 spa->spa_spares.sav_sync = B_TRUE;
3931 3933 }
3932 3934 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3933 3935 &l2cache, &nl2cache) == 0) {
3934 3936 if (spa->spa_l2cache.sav_config)
3935 3937 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3936 3938 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3937 3939 else
3938 3940 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3939 3941 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3940 3942 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3941 3943 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3942 3944 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3943 3945 spa_load_l2cache(spa);
3944 3946 spa_config_exit(spa, SCL_ALL, FTAG);
3945 3947 spa->spa_l2cache.sav_sync = B_TRUE;
3946 3948 }
3947 3949
3948 3950 /*
3949 3951 * Check for any removed devices.
3950 3952 */
3951 3953 if (spa->spa_autoreplace) {
3952 3954 spa_aux_check_removed(&spa->spa_spares);
3953 3955 spa_aux_check_removed(&spa->spa_l2cache);
3954 3956 }
3955 3957
3956 3958 if (spa_writeable(spa)) {
3957 3959 /*
3958 3960 * Update the config cache to include the newly-imported pool.
3959 3961 */
3960 3962 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3961 3963 }
3962 3964
3963 3965 /*
3964 3966 * It's possible that the pool was expanded while it was exported.
3965 3967 * We kick off an async task to handle this for us.
3966 3968 */
3967 3969 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
3968 3970
3969 3971 mutex_exit(&spa_namespace_lock);
3970 3972 spa_history_log_version(spa, "import");
3971 3973
3972 3974 return (0);
3973 3975 }
3974 3976
3975 3977 nvlist_t *
3976 3978 spa_tryimport(nvlist_t *tryconfig)
3977 3979 {
3978 3980 nvlist_t *config = NULL;
3979 3981 char *poolname;
3980 3982 spa_t *spa;
3981 3983 uint64_t state;
3982 3984 int error;
3983 3985
3984 3986 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
3985 3987 return (NULL);
3986 3988
3987 3989 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
3988 3990 return (NULL);
3989 3991
3990 3992 /*
3991 3993 * Create and initialize the spa structure.
3992 3994 */
3993 3995 mutex_enter(&spa_namespace_lock);
3994 3996 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
3995 3997 spa_activate(spa, FREAD);
3996 3998
3997 3999 /*
3998 4000 * Pass off the heavy lifting to spa_load().
3999 4001 * Pass TRUE for mosconfig because the user-supplied config
4000 4002 * is actually the one to trust when doing an import.
4001 4003 */
4002 4004 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
4003 4005
4004 4006 /*
4005 4007 * If 'tryconfig' was at least parsable, return the current config.
4006 4008 */
4007 4009 if (spa->spa_root_vdev != NULL) {
4008 4010 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4009 4011 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4010 4012 poolname) == 0);
4011 4013 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4012 4014 state) == 0);
4013 4015 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4014 4016 spa->spa_uberblock.ub_timestamp) == 0);
4015 4017 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4016 4018 spa->spa_load_info) == 0);
4017 4019
4018 4020 /*
4019 4021 * If the bootfs property exists on this pool then we
4020 4022 * copy it out so that external consumers can tell which
4021 4023 * pools are bootable.
4022 4024 */
4023 4025 if ((!error || error == EEXIST) && spa->spa_bootfs) {
4024 4026 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4025 4027
4026 4028 /*
4027 4029 * We have to play games with the name since the
4028 4030 * pool was opened as TRYIMPORT_NAME.
4029 4031 */
4030 4032 if (dsl_dsobj_to_dsname(spa_name(spa),
4031 4033 spa->spa_bootfs, tmpname) == 0) {
4032 4034 char *cp;
4033 4035 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4034 4036
4035 4037 cp = strchr(tmpname, '/');
4036 4038 if (cp == NULL) {
4037 4039 (void) strlcpy(dsname, tmpname,
4038 4040 MAXPATHLEN);
4039 4041 } else {
4040 4042 (void) snprintf(dsname, MAXPATHLEN,
4041 4043 "%s/%s", poolname, ++cp);
4042 4044 }
4043 4045 VERIFY(nvlist_add_string(config,
4044 4046 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4045 4047 kmem_free(dsname, MAXPATHLEN);
4046 4048 }
4047 4049 kmem_free(tmpname, MAXPATHLEN);
4048 4050 }
4049 4051
4050 4052 /*
4051 4053 * Add the list of hot spares and level 2 cache devices.
4052 4054 */
4053 4055 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4054 4056 spa_add_spares(spa, config);
4055 4057 spa_add_l2cache(spa, config);
4056 4058 spa_config_exit(spa, SCL_CONFIG, FTAG);
4057 4059 }
4058 4060
4059 4061 spa_unload(spa);
4060 4062 spa_deactivate(spa);
4061 4063 spa_remove(spa);
4062 4064 mutex_exit(&spa_namespace_lock);
4063 4065
4064 4066 return (config);
4065 4067 }
4066 4068
4067 4069 /*
4068 4070 * Pool export/destroy
4069 4071 *
4070 4072 * The act of destroying or exporting a pool is very simple. We make sure there
4071 4073 * is no more pending I/O and any references to the pool are gone. Then, we
4072 4074 * update the pool state and sync all the labels to disk, removing the
4073 4075 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4074 4076 * we don't sync the labels or remove the configuration cache.
4075 4077 */
4076 4078 static int
4077 4079 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4078 4080 boolean_t force, boolean_t hardforce)
4079 4081 {
4080 4082 spa_t *spa;
4081 4083
4082 4084 if (oldconfig)
4083 4085 *oldconfig = NULL;
4084 4086
4085 4087 if (!(spa_mode_global & FWRITE))
4086 4088 return (SET_ERROR(EROFS));
4087 4089
4088 4090 mutex_enter(&spa_namespace_lock);
4089 4091 if ((spa = spa_lookup(pool)) == NULL) {
4090 4092 mutex_exit(&spa_namespace_lock);
4091 4093 return (SET_ERROR(ENOENT));
4092 4094 }
4093 4095
4094 4096 /*
4095 4097 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4096 4098 * reacquire the namespace lock, and see if we can export.
4097 4099 */
4098 4100 spa_open_ref(spa, FTAG);
4099 4101 mutex_exit(&spa_namespace_lock);
4100 4102 spa_async_suspend(spa);
4101 4103 mutex_enter(&spa_namespace_lock);
4102 4104 spa_close(spa, FTAG);
4103 4105
4104 4106 /*
4105 4107 * The pool will be in core if it's openable,
4106 4108 * in which case we can modify its state.
4107 4109 */
4108 4110 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4109 4111 /*
4110 4112 * Objsets may be open only because they're dirty, so we
4111 4113 * have to force it to sync before checking spa_refcnt.
4112 4114 */
4113 4115 txg_wait_synced(spa->spa_dsl_pool, 0);
4114 4116
4115 4117 /*
4116 4118 * A pool cannot be exported or destroyed if there are active
4117 4119 * references. If we are resetting a pool, allow references by
4118 4120 * fault injection handlers.
4119 4121 */
4120 4122 if (!spa_refcount_zero(spa) ||
4121 4123 (spa->spa_inject_ref != 0 &&
4122 4124 new_state != POOL_STATE_UNINITIALIZED)) {
4123 4125 spa_async_resume(spa);
4124 4126 mutex_exit(&spa_namespace_lock);
4125 4127 return (SET_ERROR(EBUSY));
4126 4128 }
4127 4129
4128 4130 /*
4129 4131 * A pool cannot be exported if it has an active shared spare.
4130 4132 * This is to prevent other pools stealing the active spare
4131 4133 * from an exported pool. At user's own will, such pool can
4132 4134 * be forcedly exported.
4133 4135 */
4134 4136 if (!force && new_state == POOL_STATE_EXPORTED &&
4135 4137 spa_has_active_shared_spare(spa)) {
4136 4138 spa_async_resume(spa);
4137 4139 mutex_exit(&spa_namespace_lock);
4138 4140 return (SET_ERROR(EXDEV));
4139 4141 }
4140 4142
4141 4143 /*
4142 4144 * We want this to be reflected on every label,
4143 4145 * so mark them all dirty. spa_unload() will do the
4144 4146 * final sync that pushes these changes out.
4145 4147 */
4146 4148 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4147 4149 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4148 4150 spa->spa_state = new_state;
4149 4151 spa->spa_final_txg = spa_last_synced_txg(spa) +
4150 4152 TXG_DEFER_SIZE + 1;
4151 4153 vdev_config_dirty(spa->spa_root_vdev);
4152 4154 spa_config_exit(spa, SCL_ALL, FTAG);
4153 4155 }
4154 4156 }
4155 4157
4156 4158 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4157 4159
4158 4160 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4159 4161 spa_unload(spa);
4160 4162 spa_deactivate(spa);
4161 4163 }
4162 4164
4163 4165 if (oldconfig && spa->spa_config)
4164 4166 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4165 4167
4166 4168 if (new_state != POOL_STATE_UNINITIALIZED) {
4167 4169 if (!hardforce)
4168 4170 spa_config_sync(spa, B_TRUE, B_TRUE);
4169 4171 spa_remove(spa);
4170 4172 }
4171 4173 mutex_exit(&spa_namespace_lock);
4172 4174
4173 4175 return (0);
4174 4176 }
4175 4177
4176 4178 /*
4177 4179 * Destroy a storage pool.
4178 4180 */
4179 4181 int
4180 4182 spa_destroy(char *pool)
4181 4183 {
4182 4184 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4183 4185 B_FALSE, B_FALSE));
4184 4186 }
4185 4187
4186 4188 /*
4187 4189 * Export a storage pool.
4188 4190 */
4189 4191 int
4190 4192 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4191 4193 boolean_t hardforce)
4192 4194 {
4193 4195 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4194 4196 force, hardforce));
4195 4197 }
4196 4198
4197 4199 /*
4198 4200 * Similar to spa_export(), this unloads the spa_t without actually removing it
4199 4201 * from the namespace in any way.
4200 4202 */
4201 4203 int
4202 4204 spa_reset(char *pool)
4203 4205 {
4204 4206 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4205 4207 B_FALSE, B_FALSE));
4206 4208 }
4207 4209
4208 4210 /*
4209 4211 * ==========================================================================
4210 4212 * Device manipulation
4211 4213 * ==========================================================================
4212 4214 */
4213 4215
4214 4216 /*
4215 4217 * Add a device to a storage pool.
4216 4218 */
4217 4219 int
4218 4220 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4219 4221 {
4220 4222 uint64_t txg, id;
4221 4223 int error;
4222 4224 vdev_t *rvd = spa->spa_root_vdev;
4223 4225 vdev_t *vd, *tvd;
4224 4226 nvlist_t **spares, **l2cache;
4225 4227 uint_t nspares, nl2cache;
4226 4228
4227 4229 ASSERT(spa_writeable(spa));
4228 4230
4229 4231 txg = spa_vdev_enter(spa);
4230 4232
4231 4233 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4232 4234 VDEV_ALLOC_ADD)) != 0)
4233 4235 return (spa_vdev_exit(spa, NULL, txg, error));
4234 4236
4235 4237 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
4236 4238
4237 4239 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4238 4240 &nspares) != 0)
4239 4241 nspares = 0;
4240 4242
4241 4243 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4242 4244 &nl2cache) != 0)
4243 4245 nl2cache = 0;
4244 4246
4245 4247 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4246 4248 return (spa_vdev_exit(spa, vd, txg, EINVAL));
4247 4249
4248 4250 if (vd->vdev_children != 0 &&
4249 4251 (error = vdev_create(vd, txg, B_FALSE)) != 0)
4250 4252 return (spa_vdev_exit(spa, vd, txg, error));
4251 4253
4252 4254 /*
4253 4255 * We must validate the spares and l2cache devices after checking the
4254 4256 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4255 4257 */
4256 4258 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4257 4259 return (spa_vdev_exit(spa, vd, txg, error));
4258 4260
4259 4261 /*
4260 4262 * Transfer each new top-level vdev from vd to rvd.
4261 4263 */
4262 4264 for (int c = 0; c < vd->vdev_children; c++) {
4263 4265
4264 4266 /*
4265 4267 * Set the vdev id to the first hole, if one exists.
4266 4268 */
4267 4269 for (id = 0; id < rvd->vdev_children; id++) {
4268 4270 if (rvd->vdev_child[id]->vdev_ishole) {
4269 4271 vdev_free(rvd->vdev_child[id]);
4270 4272 break;
4271 4273 }
4272 4274 }
4273 4275 tvd = vd->vdev_child[c];
4274 4276 vdev_remove_child(vd, tvd);
4275 4277 tvd->vdev_id = id;
4276 4278 vdev_add_child(rvd, tvd);
4277 4279 vdev_config_dirty(tvd);
4278 4280 }
4279 4281
4280 4282 if (nspares != 0) {
4281 4283 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4282 4284 ZPOOL_CONFIG_SPARES);
4283 4285 spa_load_spares(spa);
4284 4286 spa->spa_spares.sav_sync = B_TRUE;
4285 4287 }
4286 4288
4287 4289 if (nl2cache != 0) {
4288 4290 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4289 4291 ZPOOL_CONFIG_L2CACHE);
4290 4292 spa_load_l2cache(spa);
4291 4293 spa->spa_l2cache.sav_sync = B_TRUE;
4292 4294 }
4293 4295
4294 4296 /*
4295 4297 * We have to be careful when adding new vdevs to an existing pool.
4296 4298 * If other threads start allocating from these vdevs before we
4297 4299 * sync the config cache, and we lose power, then upon reboot we may
4298 4300 * fail to open the pool because there are DVAs that the config cache
4299 4301 * can't translate. Therefore, we first add the vdevs without
4300 4302 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4301 4303 * and then let spa_config_update() initialize the new metaslabs.
4302 4304 *
4303 4305 * spa_load() checks for added-but-not-initialized vdevs, so that
4304 4306 * if we lose power at any point in this sequence, the remaining
4305 4307 * steps will be completed the next time we load the pool.
4306 4308 */
4307 4309 (void) spa_vdev_exit(spa, vd, txg, 0);
4308 4310
4309 4311 mutex_enter(&spa_namespace_lock);
4310 4312 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4311 4313 mutex_exit(&spa_namespace_lock);
4312 4314
4313 4315 return (0);
4314 4316 }
4315 4317
4316 4318 /*
4317 4319 * Attach a device to a mirror. The arguments are the path to any device
4318 4320 * in the mirror, and the nvroot for the new device. If the path specifies
4319 4321 * a device that is not mirrored, we automatically insert the mirror vdev.
4320 4322 *
4321 4323 * If 'replacing' is specified, the new device is intended to replace the
4322 4324 * existing device; in this case the two devices are made into their own
4323 4325 * mirror using the 'replacing' vdev, which is functionally identical to
4324 4326 * the mirror vdev (it actually reuses all the same ops) but has a few
4325 4327 * extra rules: you can't attach to it after it's been created, and upon
4326 4328 * completion of resilvering, the first disk (the one being replaced)
4327 4329 * is automatically detached.
4328 4330 */
4329 4331 int
4330 4332 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4331 4333 {
4332 4334 uint64_t txg, dtl_max_txg;
4333 4335 vdev_t *rvd = spa->spa_root_vdev;
4334 4336 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4335 4337 vdev_ops_t *pvops;
4336 4338 char *oldvdpath, *newvdpath;
4337 4339 int newvd_isspare;
4338 4340 int error;
4339 4341
4340 4342 ASSERT(spa_writeable(spa));
4341 4343
4342 4344 txg = spa_vdev_enter(spa);
4343 4345
4344 4346 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4345 4347
4346 4348 if (oldvd == NULL)
4347 4349 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4348 4350
4349 4351 if (!oldvd->vdev_ops->vdev_op_leaf)
4350 4352 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4351 4353
4352 4354 pvd = oldvd->vdev_parent;
4353 4355
4354 4356 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4355 4357 VDEV_ALLOC_ATTACH)) != 0)
4356 4358 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4357 4359
4358 4360 if (newrootvd->vdev_children != 1)
4359 4361 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4360 4362
4361 4363 newvd = newrootvd->vdev_child[0];
4362 4364
4363 4365 if (!newvd->vdev_ops->vdev_op_leaf)
4364 4366 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4365 4367
4366 4368 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4367 4369 return (spa_vdev_exit(spa, newrootvd, txg, error));
4368 4370
4369 4371 /*
4370 4372 * Spares can't replace logs
4371 4373 */
4372 4374 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4373 4375 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4374 4376
4375 4377 if (!replacing) {
4376 4378 /*
4377 4379 * For attach, the only allowable parent is a mirror or the root
4378 4380 * vdev.
4379 4381 */
4380 4382 if (pvd->vdev_ops != &vdev_mirror_ops &&
4381 4383 pvd->vdev_ops != &vdev_root_ops)
4382 4384 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4383 4385
4384 4386 pvops = &vdev_mirror_ops;
4385 4387 } else {
4386 4388 /*
4387 4389 * Active hot spares can only be replaced by inactive hot
4388 4390 * spares.
4389 4391 */
4390 4392 if (pvd->vdev_ops == &vdev_spare_ops &&
4391 4393 oldvd->vdev_isspare &&
4392 4394 !spa_has_spare(spa, newvd->vdev_guid))
4393 4395 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4394 4396
4395 4397 /*
4396 4398 * If the source is a hot spare, and the parent isn't already a
4397 4399 * spare, then we want to create a new hot spare. Otherwise, we
4398 4400 * want to create a replacing vdev. The user is not allowed to
4399 4401 * attach to a spared vdev child unless the 'isspare' state is
4400 4402 * the same (spare replaces spare, non-spare replaces
4401 4403 * non-spare).
4402 4404 */
4403 4405 if (pvd->vdev_ops == &vdev_replacing_ops &&
4404 4406 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4405 4407 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4406 4408 } else if (pvd->vdev_ops == &vdev_spare_ops &&
4407 4409 newvd->vdev_isspare != oldvd->vdev_isspare) {
4408 4410 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4409 4411 }
4410 4412
4411 4413 if (newvd->vdev_isspare)
4412 4414 pvops = &vdev_spare_ops;
4413 4415 else
4414 4416 pvops = &vdev_replacing_ops;
4415 4417 }
4416 4418
4417 4419 /*
4418 4420 * Make sure the new device is big enough.
4419 4421 */
4420 4422 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4421 4423 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4422 4424
4423 4425 /*
4424 4426 * The new device cannot have a higher alignment requirement
4425 4427 * than the top-level vdev.
4426 4428 */
4427 4429 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4428 4430 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4429 4431
4430 4432 /*
4431 4433 * If this is an in-place replacement, update oldvd's path and devid
4432 4434 * to make it distinguishable from newvd, and unopenable from now on.
4433 4435 */
4434 4436 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4435 4437 spa_strfree(oldvd->vdev_path);
4436 4438 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4437 4439 KM_SLEEP);
4438 4440 (void) sprintf(oldvd->vdev_path, "%s/%s",
4439 4441 newvd->vdev_path, "old");
4440 4442 if (oldvd->vdev_devid != NULL) {
4441 4443 spa_strfree(oldvd->vdev_devid);
4442 4444 oldvd->vdev_devid = NULL;
4443 4445 }
4444 4446 }
4445 4447
4446 4448 /* mark the device being resilvered */
4447 4449 newvd->vdev_resilvering = B_TRUE;
4448 4450
4449 4451 /*
4450 4452 * If the parent is not a mirror, or if we're replacing, insert the new
4451 4453 * mirror/replacing/spare vdev above oldvd.
4452 4454 */
4453 4455 if (pvd->vdev_ops != pvops)
4454 4456 pvd = vdev_add_parent(oldvd, pvops);
4455 4457
4456 4458 ASSERT(pvd->vdev_top->vdev_parent == rvd);
4457 4459 ASSERT(pvd->vdev_ops == pvops);
4458 4460 ASSERT(oldvd->vdev_parent == pvd);
4459 4461
4460 4462 /*
4461 4463 * Extract the new device from its root and add it to pvd.
4462 4464 */
4463 4465 vdev_remove_child(newrootvd, newvd);
4464 4466 newvd->vdev_id = pvd->vdev_children;
4465 4467 newvd->vdev_crtxg = oldvd->vdev_crtxg;
4466 4468 vdev_add_child(pvd, newvd);
4467 4469
4468 4470 tvd = newvd->vdev_top;
4469 4471 ASSERT(pvd->vdev_top == tvd);
4470 4472 ASSERT(tvd->vdev_parent == rvd);
4471 4473
4472 4474 vdev_config_dirty(tvd);
4473 4475
4474 4476 /*
4475 4477 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4476 4478 * for any dmu_sync-ed blocks. It will propagate upward when
4477 4479 * spa_vdev_exit() calls vdev_dtl_reassess().
4478 4480 */
4479 4481 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4480 4482
4481 4483 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4482 4484 dtl_max_txg - TXG_INITIAL);
4483 4485
4484 4486 if (newvd->vdev_isspare) {
4485 4487 spa_spare_activate(newvd);
4486 4488 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4487 4489 }
4488 4490
4489 4491 oldvdpath = spa_strdup(oldvd->vdev_path);
4490 4492 newvdpath = spa_strdup(newvd->vdev_path);
4491 4493 newvd_isspare = newvd->vdev_isspare;
4492 4494
4493 4495 /*
4494 4496 * Mark newvd's DTL dirty in this txg.
4495 4497 */
4496 4498 vdev_dirty(tvd, VDD_DTL, newvd, txg);
4497 4499
4498 4500 /*
4499 4501 * Restart the resilver
4500 4502 */
4501 4503 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4502 4504
4503 4505 /*
4504 4506 * Commit the config
4505 4507 */
4506 4508 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4507 4509
4508 4510 spa_history_log_internal(spa, "vdev attach", NULL,
4509 4511 "%s vdev=%s %s vdev=%s",
4510 4512 replacing && newvd_isspare ? "spare in" :
4511 4513 replacing ? "replace" : "attach", newvdpath,
4512 4514 replacing ? "for" : "to", oldvdpath);
4513 4515
4514 4516 spa_strfree(oldvdpath);
4515 4517 spa_strfree(newvdpath);
4516 4518
4517 4519 if (spa->spa_bootfs)
4518 4520 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4519 4521
4520 4522 return (0);
4521 4523 }
4522 4524
4523 4525 /*
4524 4526 * Detach a device from a mirror or replacing vdev.
4525 4527 *
4526 4528 * If 'replace_done' is specified, only detach if the parent
4527 4529 * is a replacing vdev.
4528 4530 */
4529 4531 int
4530 4532 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4531 4533 {
4532 4534 uint64_t txg;
4533 4535 int error;
4534 4536 vdev_t *rvd = spa->spa_root_vdev;
4535 4537 vdev_t *vd, *pvd, *cvd, *tvd;
4536 4538 boolean_t unspare = B_FALSE;
4537 4539 uint64_t unspare_guid = 0;
4538 4540 char *vdpath;
4539 4541
4540 4542 ASSERT(spa_writeable(spa));
4541 4543
4542 4544 txg = spa_vdev_enter(spa);
4543 4545
4544 4546 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4545 4547
4546 4548 if (vd == NULL)
4547 4549 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4548 4550
4549 4551 if (!vd->vdev_ops->vdev_op_leaf)
4550 4552 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4551 4553
4552 4554 pvd = vd->vdev_parent;
4553 4555
4554 4556 /*
4555 4557 * If the parent/child relationship is not as expected, don't do it.
4556 4558 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4557 4559 * vdev that's replacing B with C. The user's intent in replacing
4558 4560 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4559 4561 * the replace by detaching C, the expected behavior is to end up
4560 4562 * M(A,B). But suppose that right after deciding to detach C,
4561 4563 * the replacement of B completes. We would have M(A,C), and then
4562 4564 * ask to detach C, which would leave us with just A -- not what
4563 4565 * the user wanted. To prevent this, we make sure that the
4564 4566 * parent/child relationship hasn't changed -- in this example,
4565 4567 * that C's parent is still the replacing vdev R.
4566 4568 */
4567 4569 if (pvd->vdev_guid != pguid && pguid != 0)
4568 4570 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4569 4571
4570 4572 /*
4571 4573 * Only 'replacing' or 'spare' vdevs can be replaced.
4572 4574 */
4573 4575 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4574 4576 pvd->vdev_ops != &vdev_spare_ops)
4575 4577 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4576 4578
4577 4579 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4578 4580 spa_version(spa) >= SPA_VERSION_SPARES);
4579 4581
4580 4582 /*
4581 4583 * Only mirror, replacing, and spare vdevs support detach.
4582 4584 */
4583 4585 if (pvd->vdev_ops != &vdev_replacing_ops &&
4584 4586 pvd->vdev_ops != &vdev_mirror_ops &&
4585 4587 pvd->vdev_ops != &vdev_spare_ops)
4586 4588 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4587 4589
4588 4590 /*
4589 4591 * If this device has the only valid copy of some data,
4590 4592 * we cannot safely detach it.
4591 4593 */
4592 4594 if (vdev_dtl_required(vd))
4593 4595 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4594 4596
4595 4597 ASSERT(pvd->vdev_children >= 2);
4596 4598
4597 4599 /*
4598 4600 * If we are detaching the second disk from a replacing vdev, then
4599 4601 * check to see if we changed the original vdev's path to have "/old"
4600 4602 * at the end in spa_vdev_attach(). If so, undo that change now.
4601 4603 */
4602 4604 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4603 4605 vd->vdev_path != NULL) {
4604 4606 size_t len = strlen(vd->vdev_path);
4605 4607
4606 4608 for (int c = 0; c < pvd->vdev_children; c++) {
4607 4609 cvd = pvd->vdev_child[c];
4608 4610
4609 4611 if (cvd == vd || cvd->vdev_path == NULL)
4610 4612 continue;
4611 4613
4612 4614 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4613 4615 strcmp(cvd->vdev_path + len, "/old") == 0) {
4614 4616 spa_strfree(cvd->vdev_path);
4615 4617 cvd->vdev_path = spa_strdup(vd->vdev_path);
4616 4618 break;
4617 4619 }
4618 4620 }
4619 4621 }
4620 4622
4621 4623 /*
4622 4624 * If we are detaching the original disk from a spare, then it implies
4623 4625 * that the spare should become a real disk, and be removed from the
4624 4626 * active spare list for the pool.
4625 4627 */
4626 4628 if (pvd->vdev_ops == &vdev_spare_ops &&
4627 4629 vd->vdev_id == 0 &&
4628 4630 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4629 4631 unspare = B_TRUE;
4630 4632
4631 4633 /*
4632 4634 * Erase the disk labels so the disk can be used for other things.
4633 4635 * This must be done after all other error cases are handled,
4634 4636 * but before we disembowel vd (so we can still do I/O to it).
4635 4637 * But if we can't do it, don't treat the error as fatal --
4636 4638 * it may be that the unwritability of the disk is the reason
4637 4639 * it's being detached!
4638 4640 */
4639 4641 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4640 4642
4641 4643 /*
4642 4644 * Remove vd from its parent and compact the parent's children.
4643 4645 */
4644 4646 vdev_remove_child(pvd, vd);
4645 4647 vdev_compact_children(pvd);
4646 4648
4647 4649 /*
4648 4650 * Remember one of the remaining children so we can get tvd below.
4649 4651 */
4650 4652 cvd = pvd->vdev_child[pvd->vdev_children - 1];
4651 4653
4652 4654 /*
4653 4655 * If we need to remove the remaining child from the list of hot spares,
4654 4656 * do it now, marking the vdev as no longer a spare in the process.
4655 4657 * We must do this before vdev_remove_parent(), because that can
4656 4658 * change the GUID if it creates a new toplevel GUID. For a similar
4657 4659 * reason, we must remove the spare now, in the same txg as the detach;
4658 4660 * otherwise someone could attach a new sibling, change the GUID, and
4659 4661 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4660 4662 */
4661 4663 if (unspare) {
4662 4664 ASSERT(cvd->vdev_isspare);
4663 4665 spa_spare_remove(cvd);
4664 4666 unspare_guid = cvd->vdev_guid;
4665 4667 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4666 4668 cvd->vdev_unspare = B_TRUE;
↓ open down ↓ |
3885 lines elided |
↑ open up ↑ |
4667 4669 }
4668 4670
4669 4671 /*
4670 4672 * If the parent mirror/replacing vdev only has one child,
4671 4673 * the parent is no longer needed. Remove it from the tree.
4672 4674 */
4673 4675 if (pvd->vdev_children == 1) {
4674 4676 if (pvd->vdev_ops == &vdev_spare_ops)
4675 4677 cvd->vdev_unspare = B_FALSE;
4676 4678 vdev_remove_parent(cvd);
4677 - cvd->vdev_resilvering = B_FALSE;
4678 4679 }
4679 4680
4680 4681
4681 4682 /*
4682 4683 * We don't set tvd until now because the parent we just removed
4683 4684 * may have been the previous top-level vdev.
4684 4685 */
4685 4686 tvd = cvd->vdev_top;
4686 4687 ASSERT(tvd->vdev_parent == rvd);
4687 4688
4688 4689 /*
4689 4690 * Reevaluate the parent vdev state.
4690 4691 */
4691 4692 vdev_propagate_state(cvd);
4692 4693
4693 4694 /*
4694 4695 * If the 'autoexpand' property is set on the pool then automatically
4695 4696 * try to expand the size of the pool. For example if the device we
4696 4697 * just detached was smaller than the others, it may be possible to
4697 4698 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4698 4699 * first so that we can obtain the updated sizes of the leaf vdevs.
4699 4700 */
4700 4701 if (spa->spa_autoexpand) {
4701 4702 vdev_reopen(tvd);
4702 4703 vdev_expand(tvd, txg);
4703 4704 }
4704 4705
4705 4706 vdev_config_dirty(tvd);
4706 4707
4707 4708 /*
4708 4709 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
4709 4710 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4710 4711 * But first make sure we're not on any *other* txg's DTL list, to
4711 4712 * prevent vd from being accessed after it's freed.
4712 4713 */
4713 4714 vdpath = spa_strdup(vd->vdev_path);
4714 4715 for (int t = 0; t < TXG_SIZE; t++)
4715 4716 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4716 4717 vd->vdev_detached = B_TRUE;
4717 4718 vdev_dirty(tvd, VDD_DTL, vd, txg);
4718 4719
4719 4720 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4720 4721
4721 4722 /* hang on to the spa before we release the lock */
4722 4723 spa_open_ref(spa, FTAG);
4723 4724
4724 4725 error = spa_vdev_exit(spa, vd, txg, 0);
4725 4726
4726 4727 spa_history_log_internal(spa, "detach", NULL,
4727 4728 "vdev=%s", vdpath);
4728 4729 spa_strfree(vdpath);
4729 4730
4730 4731 /*
4731 4732 * If this was the removal of the original device in a hot spare vdev,
4732 4733 * then we want to go through and remove the device from the hot spare
4733 4734 * list of every other pool.
4734 4735 */
4735 4736 if (unspare) {
4736 4737 spa_t *altspa = NULL;
4737 4738
4738 4739 mutex_enter(&spa_namespace_lock);
4739 4740 while ((altspa = spa_next(altspa)) != NULL) {
4740 4741 if (altspa->spa_state != POOL_STATE_ACTIVE ||
4741 4742 altspa == spa)
4742 4743 continue;
4743 4744
4744 4745 spa_open_ref(altspa, FTAG);
4745 4746 mutex_exit(&spa_namespace_lock);
4746 4747 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4747 4748 mutex_enter(&spa_namespace_lock);
4748 4749 spa_close(altspa, FTAG);
4749 4750 }
4750 4751 mutex_exit(&spa_namespace_lock);
4751 4752
4752 4753 /* search the rest of the vdevs for spares to remove */
4753 4754 spa_vdev_resilver_done(spa);
4754 4755 }
4755 4756
4756 4757 /* all done with the spa; OK to release */
4757 4758 mutex_enter(&spa_namespace_lock);
4758 4759 spa_close(spa, FTAG);
4759 4760 mutex_exit(&spa_namespace_lock);
4760 4761
4761 4762 return (error);
4762 4763 }
4763 4764
4764 4765 /*
4765 4766 * Split a set of devices from their mirrors, and create a new pool from them.
4766 4767 */
4767 4768 int
4768 4769 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4769 4770 nvlist_t *props, boolean_t exp)
4770 4771 {
4771 4772 int error = 0;
4772 4773 uint64_t txg, *glist;
4773 4774 spa_t *newspa;
4774 4775 uint_t c, children, lastlog;
4775 4776 nvlist_t **child, *nvl, *tmp;
4776 4777 dmu_tx_t *tx;
4777 4778 char *altroot = NULL;
4778 4779 vdev_t *rvd, **vml = NULL; /* vdev modify list */
4779 4780 boolean_t activate_slog;
4780 4781
4781 4782 ASSERT(spa_writeable(spa));
4782 4783
4783 4784 txg = spa_vdev_enter(spa);
4784 4785
4785 4786 /* clear the log and flush everything up to now */
4786 4787 activate_slog = spa_passivate_log(spa);
4787 4788 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4788 4789 error = spa_offline_log(spa);
4789 4790 txg = spa_vdev_config_enter(spa);
4790 4791
4791 4792 if (activate_slog)
4792 4793 spa_activate_log(spa);
4793 4794
4794 4795 if (error != 0)
4795 4796 return (spa_vdev_exit(spa, NULL, txg, error));
4796 4797
4797 4798 /* check new spa name before going any further */
4798 4799 if (spa_lookup(newname) != NULL)
4799 4800 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4800 4801
4801 4802 /*
4802 4803 * scan through all the children to ensure they're all mirrors
4803 4804 */
4804 4805 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4805 4806 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4806 4807 &children) != 0)
4807 4808 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4808 4809
4809 4810 /* first, check to ensure we've got the right child count */
4810 4811 rvd = spa->spa_root_vdev;
4811 4812 lastlog = 0;
4812 4813 for (c = 0; c < rvd->vdev_children; c++) {
4813 4814 vdev_t *vd = rvd->vdev_child[c];
4814 4815
4815 4816 /* don't count the holes & logs as children */
4816 4817 if (vd->vdev_islog || vd->vdev_ishole) {
4817 4818 if (lastlog == 0)
4818 4819 lastlog = c;
4819 4820 continue;
4820 4821 }
4821 4822
4822 4823 lastlog = 0;
4823 4824 }
4824 4825 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4825 4826 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4826 4827
4827 4828 /* next, ensure no spare or cache devices are part of the split */
4828 4829 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4829 4830 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4830 4831 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4831 4832
4832 4833 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4833 4834 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4834 4835
4835 4836 /* then, loop over each vdev and validate it */
4836 4837 for (c = 0; c < children; c++) {
4837 4838 uint64_t is_hole = 0;
4838 4839
4839 4840 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4840 4841 &is_hole);
4841 4842
4842 4843 if (is_hole != 0) {
4843 4844 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4844 4845 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4845 4846 continue;
4846 4847 } else {
4847 4848 error = SET_ERROR(EINVAL);
4848 4849 break;
4849 4850 }
4850 4851 }
4851 4852
4852 4853 /* which disk is going to be split? */
4853 4854 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4854 4855 &glist[c]) != 0) {
4855 4856 error = SET_ERROR(EINVAL);
4856 4857 break;
4857 4858 }
4858 4859
4859 4860 /* look it up in the spa */
4860 4861 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4861 4862 if (vml[c] == NULL) {
4862 4863 error = SET_ERROR(ENODEV);
4863 4864 break;
4864 4865 }
4865 4866
4866 4867 /* make sure there's nothing stopping the split */
4867 4868 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4868 4869 vml[c]->vdev_islog ||
4869 4870 vml[c]->vdev_ishole ||
4870 4871 vml[c]->vdev_isspare ||
4871 4872 vml[c]->vdev_isl2cache ||
4872 4873 !vdev_writeable(vml[c]) ||
4873 4874 vml[c]->vdev_children != 0 ||
4874 4875 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4875 4876 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4876 4877 error = SET_ERROR(EINVAL);
4877 4878 break;
4878 4879 }
4879 4880
4880 4881 if (vdev_dtl_required(vml[c])) {
4881 4882 error = SET_ERROR(EBUSY);
4882 4883 break;
4883 4884 }
4884 4885
4885 4886 /* we need certain info from the top level */
4886 4887 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4887 4888 vml[c]->vdev_top->vdev_ms_array) == 0);
4888 4889 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4889 4890 vml[c]->vdev_top->vdev_ms_shift) == 0);
4890 4891 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4891 4892 vml[c]->vdev_top->vdev_asize) == 0);
4892 4893 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4893 4894 vml[c]->vdev_top->vdev_ashift) == 0);
4894 4895 }
4895 4896
4896 4897 if (error != 0) {
4897 4898 kmem_free(vml, children * sizeof (vdev_t *));
4898 4899 kmem_free(glist, children * sizeof (uint64_t));
4899 4900 return (spa_vdev_exit(spa, NULL, txg, error));
4900 4901 }
4901 4902
4902 4903 /* stop writers from using the disks */
4903 4904 for (c = 0; c < children; c++) {
4904 4905 if (vml[c] != NULL)
4905 4906 vml[c]->vdev_offline = B_TRUE;
4906 4907 }
4907 4908 vdev_reopen(spa->spa_root_vdev);
4908 4909
4909 4910 /*
4910 4911 * Temporarily record the splitting vdevs in the spa config. This
4911 4912 * will disappear once the config is regenerated.
4912 4913 */
4913 4914 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4914 4915 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4915 4916 glist, children) == 0);
4916 4917 kmem_free(glist, children * sizeof (uint64_t));
4917 4918
4918 4919 mutex_enter(&spa->spa_props_lock);
4919 4920 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4920 4921 nvl) == 0);
4921 4922 mutex_exit(&spa->spa_props_lock);
4922 4923 spa->spa_config_splitting = nvl;
4923 4924 vdev_config_dirty(spa->spa_root_vdev);
4924 4925
4925 4926 /* configure and create the new pool */
4926 4927 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4927 4928 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4928 4929 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4929 4930 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4930 4931 spa_version(spa)) == 0);
4931 4932 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4932 4933 spa->spa_config_txg) == 0);
4933 4934 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4934 4935 spa_generate_guid(NULL)) == 0);
4935 4936 (void) nvlist_lookup_string(props,
4936 4937 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4937 4938
4938 4939 /* add the new pool to the namespace */
4939 4940 newspa = spa_add(newname, config, altroot);
4940 4941 newspa->spa_config_txg = spa->spa_config_txg;
4941 4942 spa_set_log_state(newspa, SPA_LOG_CLEAR);
4942 4943
4943 4944 /* release the spa config lock, retaining the namespace lock */
4944 4945 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4945 4946
4946 4947 if (zio_injection_enabled)
4947 4948 zio_handle_panic_injection(spa, FTAG, 1);
4948 4949
4949 4950 spa_activate(newspa, spa_mode_global);
4950 4951 spa_async_suspend(newspa);
4951 4952
4952 4953 /* create the new pool from the disks of the original pool */
4953 4954 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
4954 4955 if (error)
4955 4956 goto out;
4956 4957
4957 4958 /* if that worked, generate a real config for the new pool */
4958 4959 if (newspa->spa_root_vdev != NULL) {
4959 4960 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
4960 4961 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4961 4962 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
4962 4963 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
4963 4964 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
4964 4965 B_TRUE));
4965 4966 }
4966 4967
4967 4968 /* set the props */
4968 4969 if (props != NULL) {
4969 4970 spa_configfile_set(newspa, props, B_FALSE);
4970 4971 error = spa_prop_set(newspa, props);
4971 4972 if (error)
4972 4973 goto out;
4973 4974 }
4974 4975
4975 4976 /* flush everything */
4976 4977 txg = spa_vdev_config_enter(newspa);
4977 4978 vdev_config_dirty(newspa->spa_root_vdev);
4978 4979 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
4979 4980
4980 4981 if (zio_injection_enabled)
4981 4982 zio_handle_panic_injection(spa, FTAG, 2);
4982 4983
4983 4984 spa_async_resume(newspa);
4984 4985
4985 4986 /* finally, update the original pool's config */
4986 4987 txg = spa_vdev_config_enter(spa);
4987 4988 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
4988 4989 error = dmu_tx_assign(tx, TXG_WAIT);
4989 4990 if (error != 0)
4990 4991 dmu_tx_abort(tx);
4991 4992 for (c = 0; c < children; c++) {
4992 4993 if (vml[c] != NULL) {
4993 4994 vdev_split(vml[c]);
4994 4995 if (error == 0)
4995 4996 spa_history_log_internal(spa, "detach", tx,
4996 4997 "vdev=%s", vml[c]->vdev_path);
4997 4998 vdev_free(vml[c]);
4998 4999 }
4999 5000 }
5000 5001 vdev_config_dirty(spa->spa_root_vdev);
5001 5002 spa->spa_config_splitting = NULL;
5002 5003 nvlist_free(nvl);
5003 5004 if (error == 0)
5004 5005 dmu_tx_commit(tx);
5005 5006 (void) spa_vdev_exit(spa, NULL, txg, 0);
5006 5007
5007 5008 if (zio_injection_enabled)
5008 5009 zio_handle_panic_injection(spa, FTAG, 3);
5009 5010
5010 5011 /* split is complete; log a history record */
5011 5012 spa_history_log_internal(newspa, "split", NULL,
5012 5013 "from pool %s", spa_name(spa));
5013 5014
5014 5015 kmem_free(vml, children * sizeof (vdev_t *));
5015 5016
5016 5017 /* if we're not going to mount the filesystems in userland, export */
5017 5018 if (exp)
5018 5019 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5019 5020 B_FALSE, B_FALSE);
5020 5021
5021 5022 return (error);
5022 5023
5023 5024 out:
5024 5025 spa_unload(newspa);
5025 5026 spa_deactivate(newspa);
5026 5027 spa_remove(newspa);
5027 5028
5028 5029 txg = spa_vdev_config_enter(spa);
5029 5030
5030 5031 /* re-online all offlined disks */
5031 5032 for (c = 0; c < children; c++) {
5032 5033 if (vml[c] != NULL)
5033 5034 vml[c]->vdev_offline = B_FALSE;
5034 5035 }
5035 5036 vdev_reopen(spa->spa_root_vdev);
5036 5037
5037 5038 nvlist_free(spa->spa_config_splitting);
5038 5039 spa->spa_config_splitting = NULL;
5039 5040 (void) spa_vdev_exit(spa, NULL, txg, error);
5040 5041
5041 5042 kmem_free(vml, children * sizeof (vdev_t *));
5042 5043 return (error);
5043 5044 }
5044 5045
5045 5046 static nvlist_t *
5046 5047 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5047 5048 {
5048 5049 for (int i = 0; i < count; i++) {
5049 5050 uint64_t guid;
5050 5051
5051 5052 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5052 5053 &guid) == 0);
5053 5054
5054 5055 if (guid == target_guid)
5055 5056 return (nvpp[i]);
5056 5057 }
5057 5058
5058 5059 return (NULL);
5059 5060 }
5060 5061
5061 5062 static void
5062 5063 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5063 5064 nvlist_t *dev_to_remove)
5064 5065 {
5065 5066 nvlist_t **newdev = NULL;
5066 5067
5067 5068 if (count > 1)
5068 5069 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
5069 5070
5070 5071 for (int i = 0, j = 0; i < count; i++) {
5071 5072 if (dev[i] == dev_to_remove)
5072 5073 continue;
5073 5074 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
5074 5075 }
5075 5076
5076 5077 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5077 5078 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5078 5079
5079 5080 for (int i = 0; i < count - 1; i++)
5080 5081 nvlist_free(newdev[i]);
5081 5082
5082 5083 if (count > 1)
5083 5084 kmem_free(newdev, (count - 1) * sizeof (void *));
5084 5085 }
5085 5086
5086 5087 /*
5087 5088 * Evacuate the device.
5088 5089 */
5089 5090 static int
5090 5091 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5091 5092 {
5092 5093 uint64_t txg;
5093 5094 int error = 0;
5094 5095
5095 5096 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5096 5097 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5097 5098 ASSERT(vd == vd->vdev_top);
5098 5099
5099 5100 /*
5100 5101 * Evacuate the device. We don't hold the config lock as writer
5101 5102 * since we need to do I/O but we do keep the
5102 5103 * spa_namespace_lock held. Once this completes the device
5103 5104 * should no longer have any blocks allocated on it.
5104 5105 */
5105 5106 if (vd->vdev_islog) {
5106 5107 if (vd->vdev_stat.vs_alloc != 0)
5107 5108 error = spa_offline_log(spa);
5108 5109 } else {
5109 5110 error = SET_ERROR(ENOTSUP);
5110 5111 }
5111 5112
5112 5113 if (error)
5113 5114 return (error);
5114 5115
5115 5116 /*
5116 5117 * The evacuation succeeded. Remove any remaining MOS metadata
5117 5118 * associated with this vdev, and wait for these changes to sync.
5118 5119 */
5119 5120 ASSERT0(vd->vdev_stat.vs_alloc);
5120 5121 txg = spa_vdev_config_enter(spa);
5121 5122 vd->vdev_removing = B_TRUE;
5122 5123 vdev_dirty(vd, 0, NULL, txg);
5123 5124 vdev_config_dirty(vd);
5124 5125 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5125 5126
5126 5127 return (0);
5127 5128 }
5128 5129
5129 5130 /*
5130 5131 * Complete the removal by cleaning up the namespace.
5131 5132 */
5132 5133 static void
5133 5134 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5134 5135 {
5135 5136 vdev_t *rvd = spa->spa_root_vdev;
5136 5137 uint64_t id = vd->vdev_id;
5137 5138 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5138 5139
5139 5140 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5140 5141 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5141 5142 ASSERT(vd == vd->vdev_top);
5142 5143
5143 5144 /*
5144 5145 * Only remove any devices which are empty.
5145 5146 */
5146 5147 if (vd->vdev_stat.vs_alloc != 0)
5147 5148 return;
5148 5149
5149 5150 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5150 5151
5151 5152 if (list_link_active(&vd->vdev_state_dirty_node))
5152 5153 vdev_state_clean(vd);
5153 5154 if (list_link_active(&vd->vdev_config_dirty_node))
5154 5155 vdev_config_clean(vd);
5155 5156
5156 5157 vdev_free(vd);
5157 5158
5158 5159 if (last_vdev) {
5159 5160 vdev_compact_children(rvd);
5160 5161 } else {
5161 5162 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5162 5163 vdev_add_child(rvd, vd);
5163 5164 }
5164 5165 vdev_config_dirty(rvd);
5165 5166
5166 5167 /*
5167 5168 * Reassess the health of our root vdev.
5168 5169 */
5169 5170 vdev_reopen(rvd);
5170 5171 }
5171 5172
5172 5173 /*
5173 5174 * Remove a device from the pool -
5174 5175 *
5175 5176 * Removing a device from the vdev namespace requires several steps
5176 5177 * and can take a significant amount of time. As a result we use
5177 5178 * the spa_vdev_config_[enter/exit] functions which allow us to
5178 5179 * grab and release the spa_config_lock while still holding the namespace
5179 5180 * lock. During each step the configuration is synced out.
5180 5181 *
5181 5182 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5182 5183 * devices.
5183 5184 */
5184 5185 int
5185 5186 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5186 5187 {
5187 5188 vdev_t *vd;
5188 5189 metaslab_group_t *mg;
5189 5190 nvlist_t **spares, **l2cache, *nv;
5190 5191 uint64_t txg = 0;
5191 5192 uint_t nspares, nl2cache;
5192 5193 int error = 0;
5193 5194 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5194 5195
5195 5196 ASSERT(spa_writeable(spa));
5196 5197
5197 5198 if (!locked)
5198 5199 txg = spa_vdev_enter(spa);
5199 5200
5200 5201 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5201 5202
5202 5203 if (spa->spa_spares.sav_vdevs != NULL &&
5203 5204 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5204 5205 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5205 5206 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5206 5207 /*
5207 5208 * Only remove the hot spare if it's not currently in use
5208 5209 * in this pool.
5209 5210 */
5210 5211 if (vd == NULL || unspare) {
5211 5212 spa_vdev_remove_aux(spa->spa_spares.sav_config,
5212 5213 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5213 5214 spa_load_spares(spa);
5214 5215 spa->spa_spares.sav_sync = B_TRUE;
5215 5216 } else {
5216 5217 error = SET_ERROR(EBUSY);
5217 5218 }
5218 5219 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
5219 5220 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5220 5221 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5221 5222 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5222 5223 /*
5223 5224 * Cache devices can always be removed.
5224 5225 */
5225 5226 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5226 5227 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5227 5228 spa_load_l2cache(spa);
5228 5229 spa->spa_l2cache.sav_sync = B_TRUE;
5229 5230 } else if (vd != NULL && vd->vdev_islog) {
5230 5231 ASSERT(!locked);
5231 5232 ASSERT(vd == vd->vdev_top);
5232 5233
5233 5234 /*
5234 5235 * XXX - Once we have bp-rewrite this should
5235 5236 * become the common case.
5236 5237 */
5237 5238
5238 5239 mg = vd->vdev_mg;
5239 5240
5240 5241 /*
5241 5242 * Stop allocating from this vdev.
5242 5243 */
5243 5244 metaslab_group_passivate(mg);
5244 5245
5245 5246 /*
5246 5247 * Wait for the youngest allocations and frees to sync,
5247 5248 * and then wait for the deferral of those frees to finish.
5248 5249 */
5249 5250 spa_vdev_config_exit(spa, NULL,
5250 5251 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5251 5252
5252 5253 /*
5253 5254 * Attempt to evacuate the vdev.
5254 5255 */
5255 5256 error = spa_vdev_remove_evacuate(spa, vd);
5256 5257
5257 5258 txg = spa_vdev_config_enter(spa);
5258 5259
5259 5260 /*
5260 5261 * If we couldn't evacuate the vdev, unwind.
5261 5262 */
5262 5263 if (error) {
5263 5264 metaslab_group_activate(mg);
5264 5265 return (spa_vdev_exit(spa, NULL, txg, error));
5265 5266 }
5266 5267
5267 5268 /*
5268 5269 * Clean up the vdev namespace.
5269 5270 */
5270 5271 spa_vdev_remove_from_namespace(spa, vd);
5271 5272
5272 5273 } else if (vd != NULL) {
5273 5274 /*
5274 5275 * Normal vdevs cannot be removed (yet).
5275 5276 */
5276 5277 error = SET_ERROR(ENOTSUP);
5277 5278 } else {
5278 5279 /*
5279 5280 * There is no vdev of any kind with the specified guid.
5280 5281 */
5281 5282 error = SET_ERROR(ENOENT);
5282 5283 }
5283 5284
5284 5285 if (!locked)
5285 5286 return (spa_vdev_exit(spa, NULL, txg, error));
5286 5287
5287 5288 return (error);
5288 5289 }
5289 5290
5290 5291 /*
5291 5292 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5292 5293 * currently spared, so we can detach it.
5293 5294 */
5294 5295 static vdev_t *
↓ open down ↓ |
607 lines elided |
↑ open up ↑ |
5295 5296 spa_vdev_resilver_done_hunt(vdev_t *vd)
5296 5297 {
5297 5298 vdev_t *newvd, *oldvd;
5298 5299
5299 5300 for (int c = 0; c < vd->vdev_children; c++) {
5300 5301 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5301 5302 if (oldvd != NULL)
5302 5303 return (oldvd);
5303 5304 }
5304 5305
5306 + if (vd->vdev_resilvering && vdev_dtl_empty(vd, DTL_MISSING) &&
5307 + vdev_dtl_empty(vd, DTL_OUTAGE)) {
5308 + ASSERT(vd->vdev_ops->vdev_op_leaf);
5309 + vd->vdev_resilvering = B_FALSE;
5310 + vdev_config_dirty(vd->vdev_top);
5311 + }
5312 +
5305 5313 /*
5306 5314 * Check for a completed replacement. We always consider the first
5307 5315 * vdev in the list to be the oldest vdev, and the last one to be
5308 5316 * the newest (see spa_vdev_attach() for how that works). In
5309 5317 * the case where the newest vdev is faulted, we will not automatically
5310 5318 * remove it after a resilver completes. This is OK as it will require
5311 5319 * user intervention to determine which disk the admin wishes to keep.
5312 5320 */
5313 5321 if (vd->vdev_ops == &vdev_replacing_ops) {
5314 5322 ASSERT(vd->vdev_children > 1);
5315 5323
5316 5324 newvd = vd->vdev_child[vd->vdev_children - 1];
5317 5325 oldvd = vd->vdev_child[0];
5318 5326
5319 5327 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5320 5328 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5321 5329 !vdev_dtl_required(oldvd))
5322 5330 return (oldvd);
5323 5331 }
5324 5332
5325 5333 /*
5326 5334 * Check for a completed resilver with the 'unspare' flag set.
5327 5335 */
5328 5336 if (vd->vdev_ops == &vdev_spare_ops) {
5329 5337 vdev_t *first = vd->vdev_child[0];
5330 5338 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5331 5339
5332 5340 if (last->vdev_unspare) {
5333 5341 oldvd = first;
5334 5342 newvd = last;
5335 5343 } else if (first->vdev_unspare) {
5336 5344 oldvd = last;
5337 5345 newvd = first;
5338 5346 } else {
5339 5347 oldvd = NULL;
5340 5348 }
5341 5349
5342 5350 if (oldvd != NULL &&
5343 5351 vdev_dtl_empty(newvd, DTL_MISSING) &&
5344 5352 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5345 5353 !vdev_dtl_required(oldvd))
5346 5354 return (oldvd);
5347 5355
5348 5356 /*
5349 5357 * If there are more than two spares attached to a disk,
5350 5358 * and those spares are not required, then we want to
5351 5359 * attempt to free them up now so that they can be used
5352 5360 * by other pools. Once we're back down to a single
5353 5361 * disk+spare, we stop removing them.
5354 5362 */
5355 5363 if (vd->vdev_children > 2) {
5356 5364 newvd = vd->vdev_child[1];
5357 5365
5358 5366 if (newvd->vdev_isspare && last->vdev_isspare &&
5359 5367 vdev_dtl_empty(last, DTL_MISSING) &&
5360 5368 vdev_dtl_empty(last, DTL_OUTAGE) &&
5361 5369 !vdev_dtl_required(newvd))
5362 5370 return (newvd);
5363 5371 }
5364 5372 }
5365 5373
5366 5374 return (NULL);
5367 5375 }
5368 5376
5369 5377 static void
5370 5378 spa_vdev_resilver_done(spa_t *spa)
5371 5379 {
5372 5380 vdev_t *vd, *pvd, *ppvd;
5373 5381 uint64_t guid, sguid, pguid, ppguid;
5374 5382
5375 5383 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5376 5384
5377 5385 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5378 5386 pvd = vd->vdev_parent;
5379 5387 ppvd = pvd->vdev_parent;
5380 5388 guid = vd->vdev_guid;
5381 5389 pguid = pvd->vdev_guid;
5382 5390 ppguid = ppvd->vdev_guid;
5383 5391 sguid = 0;
5384 5392 /*
5385 5393 * If we have just finished replacing a hot spared device, then
5386 5394 * we need to detach the parent's first child (the original hot
5387 5395 * spare) as well.
5388 5396 */
5389 5397 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5390 5398 ppvd->vdev_children == 2) {
5391 5399 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5392 5400 sguid = ppvd->vdev_child[1]->vdev_guid;
5393 5401 }
5394 5402 spa_config_exit(spa, SCL_ALL, FTAG);
5395 5403 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5396 5404 return;
5397 5405 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5398 5406 return;
5399 5407 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5400 5408 }
5401 5409
5402 5410 spa_config_exit(spa, SCL_ALL, FTAG);
5403 5411 }
5404 5412
5405 5413 /*
5406 5414 * Update the stored path or FRU for this vdev.
5407 5415 */
5408 5416 int
5409 5417 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5410 5418 boolean_t ispath)
5411 5419 {
5412 5420 vdev_t *vd;
5413 5421 boolean_t sync = B_FALSE;
5414 5422
5415 5423 ASSERT(spa_writeable(spa));
5416 5424
5417 5425 spa_vdev_state_enter(spa, SCL_ALL);
5418 5426
5419 5427 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5420 5428 return (spa_vdev_state_exit(spa, NULL, ENOENT));
5421 5429
5422 5430 if (!vd->vdev_ops->vdev_op_leaf)
5423 5431 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5424 5432
5425 5433 if (ispath) {
5426 5434 if (strcmp(value, vd->vdev_path) != 0) {
5427 5435 spa_strfree(vd->vdev_path);
5428 5436 vd->vdev_path = spa_strdup(value);
5429 5437 sync = B_TRUE;
5430 5438 }
5431 5439 } else {
5432 5440 if (vd->vdev_fru == NULL) {
5433 5441 vd->vdev_fru = spa_strdup(value);
5434 5442 sync = B_TRUE;
5435 5443 } else if (strcmp(value, vd->vdev_fru) != 0) {
5436 5444 spa_strfree(vd->vdev_fru);
5437 5445 vd->vdev_fru = spa_strdup(value);
5438 5446 sync = B_TRUE;
5439 5447 }
5440 5448 }
5441 5449
5442 5450 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5443 5451 }
5444 5452
5445 5453 int
5446 5454 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5447 5455 {
5448 5456 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5449 5457 }
5450 5458
5451 5459 int
5452 5460 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5453 5461 {
5454 5462 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5455 5463 }
5456 5464
5457 5465 /*
5458 5466 * ==========================================================================
5459 5467 * SPA Scanning
5460 5468 * ==========================================================================
5461 5469 */
5462 5470
5463 5471 int
5464 5472 spa_scan_stop(spa_t *spa)
5465 5473 {
5466 5474 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5467 5475 if (dsl_scan_resilvering(spa->spa_dsl_pool))
5468 5476 return (SET_ERROR(EBUSY));
5469 5477 return (dsl_scan_cancel(spa->spa_dsl_pool));
5470 5478 }
5471 5479
5472 5480 int
5473 5481 spa_scan(spa_t *spa, pool_scan_func_t func)
5474 5482 {
5475 5483 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5476 5484
5477 5485 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5478 5486 return (SET_ERROR(ENOTSUP));
5479 5487
5480 5488 /*
5481 5489 * If a resilver was requested, but there is no DTL on a
5482 5490 * writeable leaf device, we have nothing to do.
5483 5491 */
5484 5492 if (func == POOL_SCAN_RESILVER &&
5485 5493 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5486 5494 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5487 5495 return (0);
5488 5496 }
5489 5497
5490 5498 return (dsl_scan(spa->spa_dsl_pool, func));
5491 5499 }
5492 5500
5493 5501 /*
5494 5502 * ==========================================================================
5495 5503 * SPA async task processing
5496 5504 * ==========================================================================
5497 5505 */
5498 5506
5499 5507 static void
5500 5508 spa_async_remove(spa_t *spa, vdev_t *vd)
5501 5509 {
5502 5510 if (vd->vdev_remove_wanted) {
5503 5511 vd->vdev_remove_wanted = B_FALSE;
5504 5512 vd->vdev_delayed_close = B_FALSE;
5505 5513 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5506 5514
5507 5515 /*
5508 5516 * We want to clear the stats, but we don't want to do a full
5509 5517 * vdev_clear() as that will cause us to throw away
5510 5518 * degraded/faulted state as well as attempt to reopen the
5511 5519 * device, all of which is a waste.
5512 5520 */
5513 5521 vd->vdev_stat.vs_read_errors = 0;
5514 5522 vd->vdev_stat.vs_write_errors = 0;
5515 5523 vd->vdev_stat.vs_checksum_errors = 0;
5516 5524
5517 5525 vdev_state_dirty(vd->vdev_top);
5518 5526 }
5519 5527
5520 5528 for (int c = 0; c < vd->vdev_children; c++)
5521 5529 spa_async_remove(spa, vd->vdev_child[c]);
5522 5530 }
5523 5531
5524 5532 static void
5525 5533 spa_async_probe(spa_t *spa, vdev_t *vd)
5526 5534 {
5527 5535 if (vd->vdev_probe_wanted) {
5528 5536 vd->vdev_probe_wanted = B_FALSE;
5529 5537 vdev_reopen(vd); /* vdev_open() does the actual probe */
5530 5538 }
5531 5539
5532 5540 for (int c = 0; c < vd->vdev_children; c++)
5533 5541 spa_async_probe(spa, vd->vdev_child[c]);
5534 5542 }
5535 5543
5536 5544 static void
5537 5545 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5538 5546 {
5539 5547 sysevent_id_t eid;
5540 5548 nvlist_t *attr;
5541 5549 char *physpath;
5542 5550
5543 5551 if (!spa->spa_autoexpand)
5544 5552 return;
5545 5553
5546 5554 for (int c = 0; c < vd->vdev_children; c++) {
5547 5555 vdev_t *cvd = vd->vdev_child[c];
5548 5556 spa_async_autoexpand(spa, cvd);
5549 5557 }
5550 5558
5551 5559 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5552 5560 return;
5553 5561
5554 5562 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5555 5563 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5556 5564
5557 5565 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5558 5566 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5559 5567
5560 5568 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5561 5569 ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
5562 5570
5563 5571 nvlist_free(attr);
5564 5572 kmem_free(physpath, MAXPATHLEN);
5565 5573 }
5566 5574
5567 5575 static void
5568 5576 spa_async_thread(spa_t *spa)
5569 5577 {
5570 5578 int tasks;
5571 5579
5572 5580 ASSERT(spa->spa_sync_on);
5573 5581
5574 5582 mutex_enter(&spa->spa_async_lock);
5575 5583 tasks = spa->spa_async_tasks;
5576 5584 spa->spa_async_tasks = 0;
5577 5585 mutex_exit(&spa->spa_async_lock);
5578 5586
5579 5587 /*
5580 5588 * See if the config needs to be updated.
5581 5589 */
5582 5590 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5583 5591 uint64_t old_space, new_space;
5584 5592
5585 5593 mutex_enter(&spa_namespace_lock);
5586 5594 old_space = metaslab_class_get_space(spa_normal_class(spa));
5587 5595 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5588 5596 new_space = metaslab_class_get_space(spa_normal_class(spa));
5589 5597 mutex_exit(&spa_namespace_lock);
5590 5598
5591 5599 /*
5592 5600 * If the pool grew as a result of the config update,
5593 5601 * then log an internal history event.
5594 5602 */
5595 5603 if (new_space != old_space) {
5596 5604 spa_history_log_internal(spa, "vdev online", NULL,
5597 5605 "pool '%s' size: %llu(+%llu)",
5598 5606 spa_name(spa), new_space, new_space - old_space);
5599 5607 }
5600 5608 }
5601 5609
5602 5610 /*
5603 5611 * See if any devices need to be marked REMOVED.
5604 5612 */
5605 5613 if (tasks & SPA_ASYNC_REMOVE) {
5606 5614 spa_vdev_state_enter(spa, SCL_NONE);
5607 5615 spa_async_remove(spa, spa->spa_root_vdev);
5608 5616 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5609 5617 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5610 5618 for (int i = 0; i < spa->spa_spares.sav_count; i++)
5611 5619 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5612 5620 (void) spa_vdev_state_exit(spa, NULL, 0);
5613 5621 }
5614 5622
5615 5623 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5616 5624 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5617 5625 spa_async_autoexpand(spa, spa->spa_root_vdev);
5618 5626 spa_config_exit(spa, SCL_CONFIG, FTAG);
5619 5627 }
5620 5628
5621 5629 /*
5622 5630 * See if any devices need to be probed.
5623 5631 */
5624 5632 if (tasks & SPA_ASYNC_PROBE) {
5625 5633 spa_vdev_state_enter(spa, SCL_NONE);
5626 5634 spa_async_probe(spa, spa->spa_root_vdev);
5627 5635 (void) spa_vdev_state_exit(spa, NULL, 0);
5628 5636 }
5629 5637
5630 5638 /*
5631 5639 * If any devices are done replacing, detach them.
5632 5640 */
5633 5641 if (tasks & SPA_ASYNC_RESILVER_DONE)
5634 5642 spa_vdev_resilver_done(spa);
5635 5643
5636 5644 /*
5637 5645 * Kick off a resilver.
5638 5646 */
5639 5647 if (tasks & SPA_ASYNC_RESILVER)
5640 5648 dsl_resilver_restart(spa->spa_dsl_pool, 0);
5641 5649
5642 5650 /*
5643 5651 * Let the world know that we're done.
5644 5652 */
5645 5653 mutex_enter(&spa->spa_async_lock);
5646 5654 spa->spa_async_thread = NULL;
5647 5655 cv_broadcast(&spa->spa_async_cv);
5648 5656 mutex_exit(&spa->spa_async_lock);
5649 5657 thread_exit();
5650 5658 }
5651 5659
5652 5660 void
5653 5661 spa_async_suspend(spa_t *spa)
5654 5662 {
5655 5663 mutex_enter(&spa->spa_async_lock);
5656 5664 spa->spa_async_suspended++;
5657 5665 while (spa->spa_async_thread != NULL)
5658 5666 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5659 5667 mutex_exit(&spa->spa_async_lock);
5660 5668 }
5661 5669
5662 5670 void
5663 5671 spa_async_resume(spa_t *spa)
5664 5672 {
5665 5673 mutex_enter(&spa->spa_async_lock);
5666 5674 ASSERT(spa->spa_async_suspended != 0);
5667 5675 spa->spa_async_suspended--;
5668 5676 mutex_exit(&spa->spa_async_lock);
5669 5677 }
5670 5678
5671 5679 static boolean_t
5672 5680 spa_async_tasks_pending(spa_t *spa)
5673 5681 {
5674 5682 uint_t non_config_tasks;
5675 5683 uint_t config_task;
5676 5684 boolean_t config_task_suspended;
5677 5685
5678 5686 non_config_tasks = spa->spa_async_tasks & ~SPA_ASYNC_CONFIG_UPDATE;
5679 5687 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
5680 5688 if (spa->spa_ccw_fail_time == 0) {
5681 5689 config_task_suspended = B_FALSE;
5682 5690 } else {
5683 5691 config_task_suspended =
5684 5692 (gethrtime() - spa->spa_ccw_fail_time) <
5685 5693 (zfs_ccw_retry_interval * NANOSEC);
5686 5694 }
5687 5695
5688 5696 return (non_config_tasks || (config_task && !config_task_suspended));
5689 5697 }
5690 5698
5691 5699 static void
5692 5700 spa_async_dispatch(spa_t *spa)
5693 5701 {
5694 5702 mutex_enter(&spa->spa_async_lock);
5695 5703 if (spa_async_tasks_pending(spa) &&
5696 5704 !spa->spa_async_suspended &&
5697 5705 spa->spa_async_thread == NULL &&
5698 5706 rootdir != NULL)
5699 5707 spa->spa_async_thread = thread_create(NULL, 0,
5700 5708 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5701 5709 mutex_exit(&spa->spa_async_lock);
5702 5710 }
5703 5711
5704 5712 void
5705 5713 spa_async_request(spa_t *spa, int task)
5706 5714 {
5707 5715 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5708 5716 mutex_enter(&spa->spa_async_lock);
5709 5717 spa->spa_async_tasks |= task;
5710 5718 mutex_exit(&spa->spa_async_lock);
5711 5719 }
5712 5720
5713 5721 /*
5714 5722 * ==========================================================================
5715 5723 * SPA syncing routines
5716 5724 * ==========================================================================
5717 5725 */
5718 5726
5719 5727 static int
5720 5728 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5721 5729 {
5722 5730 bpobj_t *bpo = arg;
5723 5731 bpobj_enqueue(bpo, bp, tx);
5724 5732 return (0);
5725 5733 }
5726 5734
5727 5735 static int
5728 5736 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5729 5737 {
5730 5738 zio_t *zio = arg;
5731 5739
5732 5740 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5733 5741 zio->io_flags));
5734 5742 return (0);
5735 5743 }
5736 5744
5737 5745 static void
5738 5746 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5739 5747 {
5740 5748 char *packed = NULL;
5741 5749 size_t bufsize;
5742 5750 size_t nvsize = 0;
5743 5751 dmu_buf_t *db;
5744 5752
5745 5753 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5746 5754
5747 5755 /*
5748 5756 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5749 5757 * information. This avoids the dbuf_will_dirty() path and
5750 5758 * saves us a pre-read to get data we don't actually care about.
5751 5759 */
5752 5760 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
5753 5761 packed = kmem_alloc(bufsize, KM_SLEEP);
5754 5762
5755 5763 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5756 5764 KM_SLEEP) == 0);
5757 5765 bzero(packed + nvsize, bufsize - nvsize);
5758 5766
5759 5767 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5760 5768
5761 5769 kmem_free(packed, bufsize);
5762 5770
5763 5771 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5764 5772 dmu_buf_will_dirty(db, tx);
5765 5773 *(uint64_t *)db->db_data = nvsize;
5766 5774 dmu_buf_rele(db, FTAG);
5767 5775 }
5768 5776
5769 5777 static void
5770 5778 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5771 5779 const char *config, const char *entry)
5772 5780 {
5773 5781 nvlist_t *nvroot;
5774 5782 nvlist_t **list;
5775 5783 int i;
5776 5784
5777 5785 if (!sav->sav_sync)
5778 5786 return;
5779 5787
5780 5788 /*
5781 5789 * Update the MOS nvlist describing the list of available devices.
5782 5790 * spa_validate_aux() will have already made sure this nvlist is
5783 5791 * valid and the vdevs are labeled appropriately.
5784 5792 */
5785 5793 if (sav->sav_object == 0) {
5786 5794 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5787 5795 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5788 5796 sizeof (uint64_t), tx);
5789 5797 VERIFY(zap_update(spa->spa_meta_objset,
5790 5798 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5791 5799 &sav->sav_object, tx) == 0);
5792 5800 }
5793 5801
5794 5802 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5795 5803 if (sav->sav_count == 0) {
5796 5804 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5797 5805 } else {
5798 5806 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5799 5807 for (i = 0; i < sav->sav_count; i++)
5800 5808 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5801 5809 B_FALSE, VDEV_CONFIG_L2CACHE);
5802 5810 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5803 5811 sav->sav_count) == 0);
5804 5812 for (i = 0; i < sav->sav_count; i++)
5805 5813 nvlist_free(list[i]);
5806 5814 kmem_free(list, sav->sav_count * sizeof (void *));
5807 5815 }
5808 5816
5809 5817 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5810 5818 nvlist_free(nvroot);
5811 5819
5812 5820 sav->sav_sync = B_FALSE;
5813 5821 }
5814 5822
5815 5823 static void
5816 5824 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5817 5825 {
5818 5826 nvlist_t *config;
5819 5827
5820 5828 if (list_is_empty(&spa->spa_config_dirty_list))
5821 5829 return;
5822 5830
5823 5831 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5824 5832
5825 5833 config = spa_config_generate(spa, spa->spa_root_vdev,
5826 5834 dmu_tx_get_txg(tx), B_FALSE);
5827 5835
5828 5836 /*
5829 5837 * If we're upgrading the spa version then make sure that
5830 5838 * the config object gets updated with the correct version.
5831 5839 */
5832 5840 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
5833 5841 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5834 5842 spa->spa_uberblock.ub_version);
5835 5843
5836 5844 spa_config_exit(spa, SCL_STATE, FTAG);
5837 5845
5838 5846 if (spa->spa_config_syncing)
5839 5847 nvlist_free(spa->spa_config_syncing);
5840 5848 spa->spa_config_syncing = config;
5841 5849
5842 5850 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5843 5851 }
5844 5852
5845 5853 static void
5846 5854 spa_sync_version(void *arg, dmu_tx_t *tx)
5847 5855 {
5848 5856 uint64_t *versionp = arg;
5849 5857 uint64_t version = *versionp;
5850 5858 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
5851 5859
5852 5860 /*
5853 5861 * Setting the version is special cased when first creating the pool.
5854 5862 */
5855 5863 ASSERT(tx->tx_txg != TXG_INITIAL);
5856 5864
5857 5865 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
5858 5866 ASSERT(version >= spa_version(spa));
5859 5867
5860 5868 spa->spa_uberblock.ub_version = version;
5861 5869 vdev_config_dirty(spa->spa_root_vdev);
5862 5870 spa_history_log_internal(spa, "set", tx, "version=%lld", version);
5863 5871 }
5864 5872
5865 5873 /*
5866 5874 * Set zpool properties.
5867 5875 */
5868 5876 static void
5869 5877 spa_sync_props(void *arg, dmu_tx_t *tx)
5870 5878 {
5871 5879 nvlist_t *nvp = arg;
5872 5880 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
5873 5881 objset_t *mos = spa->spa_meta_objset;
5874 5882 nvpair_t *elem = NULL;
5875 5883
5876 5884 mutex_enter(&spa->spa_props_lock);
5877 5885
5878 5886 while ((elem = nvlist_next_nvpair(nvp, elem))) {
5879 5887 uint64_t intval;
5880 5888 char *strval, *fname;
5881 5889 zpool_prop_t prop;
5882 5890 const char *propname;
5883 5891 zprop_type_t proptype;
5884 5892 zfeature_info_t *feature;
5885 5893
5886 5894 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
5887 5895 case ZPROP_INVAL:
5888 5896 /*
5889 5897 * We checked this earlier in spa_prop_validate().
5890 5898 */
5891 5899 ASSERT(zpool_prop_feature(nvpair_name(elem)));
5892 5900
5893 5901 fname = strchr(nvpair_name(elem), '@') + 1;
5894 5902 VERIFY3U(0, ==, zfeature_lookup_name(fname, &feature));
5895 5903
5896 5904 spa_feature_enable(spa, feature, tx);
5897 5905 spa_history_log_internal(spa, "set", tx,
5898 5906 "%s=enabled", nvpair_name(elem));
5899 5907 break;
5900 5908
5901 5909 case ZPOOL_PROP_VERSION:
5902 5910 VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5903 5911 /*
5904 5912 * The version is synced seperatly before other
5905 5913 * properties and should be correct by now.
5906 5914 */
5907 5915 ASSERT3U(spa_version(spa), >=, intval);
5908 5916 break;
5909 5917
5910 5918 case ZPOOL_PROP_ALTROOT:
5911 5919 /*
5912 5920 * 'altroot' is a non-persistent property. It should
5913 5921 * have been set temporarily at creation or import time.
5914 5922 */
5915 5923 ASSERT(spa->spa_root != NULL);
5916 5924 break;
5917 5925
5918 5926 case ZPOOL_PROP_READONLY:
5919 5927 case ZPOOL_PROP_CACHEFILE:
5920 5928 /*
5921 5929 * 'readonly' and 'cachefile' are also non-persisitent
5922 5930 * properties.
5923 5931 */
5924 5932 break;
5925 5933 case ZPOOL_PROP_COMMENT:
5926 5934 VERIFY(nvpair_value_string(elem, &strval) == 0);
5927 5935 if (spa->spa_comment != NULL)
5928 5936 spa_strfree(spa->spa_comment);
5929 5937 spa->spa_comment = spa_strdup(strval);
5930 5938 /*
5931 5939 * We need to dirty the configuration on all the vdevs
5932 5940 * so that their labels get updated. It's unnecessary
5933 5941 * to do this for pool creation since the vdev's
5934 5942 * configuratoin has already been dirtied.
5935 5943 */
5936 5944 if (tx->tx_txg != TXG_INITIAL)
5937 5945 vdev_config_dirty(spa->spa_root_vdev);
5938 5946 spa_history_log_internal(spa, "set", tx,
5939 5947 "%s=%s", nvpair_name(elem), strval);
5940 5948 break;
5941 5949 default:
5942 5950 /*
5943 5951 * Set pool property values in the poolprops mos object.
5944 5952 */
5945 5953 if (spa->spa_pool_props_object == 0) {
5946 5954 spa->spa_pool_props_object =
5947 5955 zap_create_link(mos, DMU_OT_POOL_PROPS,
5948 5956 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
5949 5957 tx);
5950 5958 }
5951 5959
5952 5960 /* normalize the property name */
5953 5961 propname = zpool_prop_to_name(prop);
5954 5962 proptype = zpool_prop_get_type(prop);
5955 5963
5956 5964 if (nvpair_type(elem) == DATA_TYPE_STRING) {
5957 5965 ASSERT(proptype == PROP_TYPE_STRING);
5958 5966 VERIFY(nvpair_value_string(elem, &strval) == 0);
5959 5967 VERIFY(zap_update(mos,
5960 5968 spa->spa_pool_props_object, propname,
5961 5969 1, strlen(strval) + 1, strval, tx) == 0);
5962 5970 spa_history_log_internal(spa, "set", tx,
5963 5971 "%s=%s", nvpair_name(elem), strval);
5964 5972 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
5965 5973 VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5966 5974
5967 5975 if (proptype == PROP_TYPE_INDEX) {
5968 5976 const char *unused;
5969 5977 VERIFY(zpool_prop_index_to_string(
5970 5978 prop, intval, &unused) == 0);
5971 5979 }
5972 5980 VERIFY(zap_update(mos,
5973 5981 spa->spa_pool_props_object, propname,
5974 5982 8, 1, &intval, tx) == 0);
5975 5983 spa_history_log_internal(spa, "set", tx,
5976 5984 "%s=%lld", nvpair_name(elem), intval);
5977 5985 } else {
5978 5986 ASSERT(0); /* not allowed */
5979 5987 }
5980 5988
5981 5989 switch (prop) {
5982 5990 case ZPOOL_PROP_DELEGATION:
5983 5991 spa->spa_delegation = intval;
5984 5992 break;
5985 5993 case ZPOOL_PROP_BOOTFS:
5986 5994 spa->spa_bootfs = intval;
5987 5995 break;
5988 5996 case ZPOOL_PROP_FAILUREMODE:
5989 5997 spa->spa_failmode = intval;
5990 5998 break;
5991 5999 case ZPOOL_PROP_AUTOEXPAND:
5992 6000 spa->spa_autoexpand = intval;
5993 6001 if (tx->tx_txg != TXG_INITIAL)
5994 6002 spa_async_request(spa,
5995 6003 SPA_ASYNC_AUTOEXPAND);
5996 6004 break;
5997 6005 case ZPOOL_PROP_DEDUPDITTO:
5998 6006 spa->spa_dedup_ditto = intval;
5999 6007 break;
6000 6008 default:
6001 6009 break;
6002 6010 }
6003 6011 }
6004 6012
6005 6013 }
6006 6014
6007 6015 mutex_exit(&spa->spa_props_lock);
6008 6016 }
6009 6017
6010 6018 /*
6011 6019 * Perform one-time upgrade on-disk changes. spa_version() does not
6012 6020 * reflect the new version this txg, so there must be no changes this
6013 6021 * txg to anything that the upgrade code depends on after it executes.
6014 6022 * Therefore this must be called after dsl_pool_sync() does the sync
6015 6023 * tasks.
6016 6024 */
6017 6025 static void
6018 6026 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
6019 6027 {
6020 6028 dsl_pool_t *dp = spa->spa_dsl_pool;
6021 6029
6022 6030 ASSERT(spa->spa_sync_pass == 1);
6023 6031
6024 6032 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
6025 6033
6026 6034 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
6027 6035 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
6028 6036 dsl_pool_create_origin(dp, tx);
6029 6037
6030 6038 /* Keeping the origin open increases spa_minref */
6031 6039 spa->spa_minref += 3;
6032 6040 }
6033 6041
6034 6042 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
6035 6043 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
6036 6044 dsl_pool_upgrade_clones(dp, tx);
6037 6045 }
6038 6046
6039 6047 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
6040 6048 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
6041 6049 dsl_pool_upgrade_dir_clones(dp, tx);
6042 6050
6043 6051 /* Keeping the freedir open increases spa_minref */
6044 6052 spa->spa_minref += 3;
6045 6053 }
6046 6054
6047 6055 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
6048 6056 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6049 6057 spa_feature_create_zap_objects(spa, tx);
6050 6058 }
6051 6059 rrw_exit(&dp->dp_config_rwlock, FTAG);
6052 6060 }
6053 6061
6054 6062 /*
6055 6063 * Sync the specified transaction group. New blocks may be dirtied as
6056 6064 * part of the process, so we iterate until it converges.
6057 6065 */
6058 6066 void
6059 6067 spa_sync(spa_t *spa, uint64_t txg)
6060 6068 {
6061 6069 dsl_pool_t *dp = spa->spa_dsl_pool;
6062 6070 objset_t *mos = spa->spa_meta_objset;
6063 6071 bpobj_t *defer_bpo = &spa->spa_deferred_bpobj;
6064 6072 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
6065 6073 vdev_t *rvd = spa->spa_root_vdev;
6066 6074 vdev_t *vd;
6067 6075 dmu_tx_t *tx;
6068 6076 int error;
6069 6077
6070 6078 VERIFY(spa_writeable(spa));
6071 6079
6072 6080 /*
6073 6081 * Lock out configuration changes.
6074 6082 */
6075 6083 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6076 6084
6077 6085 spa->spa_syncing_txg = txg;
6078 6086 spa->spa_sync_pass = 0;
6079 6087
6080 6088 /*
6081 6089 * If there are any pending vdev state changes, convert them
6082 6090 * into config changes that go out with this transaction group.
6083 6091 */
6084 6092 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6085 6093 while (list_head(&spa->spa_state_dirty_list) != NULL) {
6086 6094 /*
6087 6095 * We need the write lock here because, for aux vdevs,
6088 6096 * calling vdev_config_dirty() modifies sav_config.
6089 6097 * This is ugly and will become unnecessary when we
6090 6098 * eliminate the aux vdev wart by integrating all vdevs
6091 6099 * into the root vdev tree.
6092 6100 */
6093 6101 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6094 6102 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
6095 6103 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
6096 6104 vdev_state_clean(vd);
6097 6105 vdev_config_dirty(vd);
6098 6106 }
6099 6107 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6100 6108 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
6101 6109 }
6102 6110 spa_config_exit(spa, SCL_STATE, FTAG);
6103 6111
6104 6112 tx = dmu_tx_create_assigned(dp, txg);
6105 6113
6106 6114 spa->spa_sync_starttime = gethrtime();
6107 6115 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid,
6108 6116 spa->spa_sync_starttime + spa->spa_deadman_synctime));
6109 6117
6110 6118 /*
6111 6119 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6112 6120 * set spa_deflate if we have no raid-z vdevs.
6113 6121 */
6114 6122 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
6115 6123 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
6116 6124 int i;
6117 6125
6118 6126 for (i = 0; i < rvd->vdev_children; i++) {
6119 6127 vd = rvd->vdev_child[i];
6120 6128 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
6121 6129 break;
6122 6130 }
6123 6131 if (i == rvd->vdev_children) {
6124 6132 spa->spa_deflate = TRUE;
6125 6133 VERIFY(0 == zap_add(spa->spa_meta_objset,
6126 6134 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
6127 6135 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
6128 6136 }
6129 6137 }
6130 6138
6131 6139 /*
6132 6140 * If anything has changed in this txg, or if someone is waiting
6133 6141 * for this txg to sync (eg, spa_vdev_remove()), push the
6134 6142 * deferred frees from the previous txg. If not, leave them
6135 6143 * alone so that we don't generate work on an otherwise idle
6136 6144 * system.
6137 6145 */
6138 6146 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
6139 6147 !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
6140 6148 !txg_list_empty(&dp->dp_sync_tasks, txg) ||
6141 6149 ((dsl_scan_active(dp->dp_scan) ||
6142 6150 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
6143 6151 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6144 6152 VERIFY3U(bpobj_iterate(defer_bpo,
6145 6153 spa_free_sync_cb, zio, tx), ==, 0);
6146 6154 VERIFY0(zio_wait(zio));
6147 6155 }
6148 6156
6149 6157 /*
6150 6158 * Iterate to convergence.
6151 6159 */
6152 6160 do {
6153 6161 int pass = ++spa->spa_sync_pass;
6154 6162
6155 6163 spa_sync_config_object(spa, tx);
6156 6164 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
6157 6165 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
6158 6166 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
6159 6167 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
6160 6168 spa_errlog_sync(spa, txg);
6161 6169 dsl_pool_sync(dp, txg);
6162 6170
6163 6171 if (pass < zfs_sync_pass_deferred_free) {
6164 6172 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6165 6173 bplist_iterate(free_bpl, spa_free_sync_cb,
6166 6174 zio, tx);
6167 6175 VERIFY(zio_wait(zio) == 0);
6168 6176 } else {
6169 6177 bplist_iterate(free_bpl, bpobj_enqueue_cb,
6170 6178 defer_bpo, tx);
6171 6179 }
6172 6180
6173 6181 ddt_sync(spa, txg);
6174 6182 dsl_scan_sync(dp, tx);
6175 6183
6176 6184 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
6177 6185 vdev_sync(vd, txg);
6178 6186
6179 6187 if (pass == 1)
6180 6188 spa_sync_upgrades(spa, tx);
6181 6189
6182 6190 } while (dmu_objset_is_dirty(mos, txg));
6183 6191
6184 6192 /*
6185 6193 * Rewrite the vdev configuration (which includes the uberblock)
6186 6194 * to commit the transaction group.
6187 6195 *
6188 6196 * If there are no dirty vdevs, we sync the uberblock to a few
6189 6197 * random top-level vdevs that are known to be visible in the
6190 6198 * config cache (see spa_vdev_add() for a complete description).
6191 6199 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6192 6200 */
6193 6201 for (;;) {
6194 6202 /*
6195 6203 * We hold SCL_STATE to prevent vdev open/close/etc.
6196 6204 * while we're attempting to write the vdev labels.
6197 6205 */
6198 6206 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6199 6207
6200 6208 if (list_is_empty(&spa->spa_config_dirty_list)) {
6201 6209 vdev_t *svd[SPA_DVAS_PER_BP];
6202 6210 int svdcount = 0;
6203 6211 int children = rvd->vdev_children;
6204 6212 int c0 = spa_get_random(children);
6205 6213
6206 6214 for (int c = 0; c < children; c++) {
6207 6215 vd = rvd->vdev_child[(c0 + c) % children];
6208 6216 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
6209 6217 continue;
6210 6218 svd[svdcount++] = vd;
6211 6219 if (svdcount == SPA_DVAS_PER_BP)
6212 6220 break;
6213 6221 }
6214 6222 error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
6215 6223 if (error != 0)
6216 6224 error = vdev_config_sync(svd, svdcount, txg,
6217 6225 B_TRUE);
6218 6226 } else {
6219 6227 error = vdev_config_sync(rvd->vdev_child,
6220 6228 rvd->vdev_children, txg, B_FALSE);
6221 6229 if (error != 0)
6222 6230 error = vdev_config_sync(rvd->vdev_child,
6223 6231 rvd->vdev_children, txg, B_TRUE);
6224 6232 }
6225 6233
6226 6234 if (error == 0)
6227 6235 spa->spa_last_synced_guid = rvd->vdev_guid;
6228 6236
6229 6237 spa_config_exit(spa, SCL_STATE, FTAG);
6230 6238
6231 6239 if (error == 0)
6232 6240 break;
6233 6241 zio_suspend(spa, NULL);
6234 6242 zio_resume_wait(spa);
6235 6243 }
6236 6244 dmu_tx_commit(tx);
6237 6245
6238 6246 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
6239 6247
6240 6248 /*
6241 6249 * Clear the dirty config list.
6242 6250 */
6243 6251 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
6244 6252 vdev_config_clean(vd);
6245 6253
6246 6254 /*
6247 6255 * Now that the new config has synced transactionally,
6248 6256 * let it become visible to the config cache.
6249 6257 */
6250 6258 if (spa->spa_config_syncing != NULL) {
6251 6259 spa_config_set(spa, spa->spa_config_syncing);
6252 6260 spa->spa_config_txg = txg;
6253 6261 spa->spa_config_syncing = NULL;
6254 6262 }
6255 6263
6256 6264 spa->spa_ubsync = spa->spa_uberblock;
6257 6265
6258 6266 dsl_pool_sync_done(dp, txg);
6259 6267
6260 6268 /*
6261 6269 * Update usable space statistics.
6262 6270 */
6263 6271 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
6264 6272 vdev_sync_done(vd, txg);
6265 6273
6266 6274 spa_update_dspace(spa);
6267 6275
6268 6276 /*
6269 6277 * It had better be the case that we didn't dirty anything
6270 6278 * since vdev_config_sync().
6271 6279 */
6272 6280 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
6273 6281 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
6274 6282 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
6275 6283
6276 6284 spa->spa_sync_pass = 0;
6277 6285
6278 6286 spa_config_exit(spa, SCL_CONFIG, FTAG);
6279 6287
6280 6288 spa_handle_ignored_writes(spa);
6281 6289
6282 6290 /*
6283 6291 * If any async tasks have been requested, kick them off.
6284 6292 */
6285 6293 spa_async_dispatch(spa);
6286 6294 }
6287 6295
6288 6296 /*
6289 6297 * Sync all pools. We don't want to hold the namespace lock across these
6290 6298 * operations, so we take a reference on the spa_t and drop the lock during the
6291 6299 * sync.
6292 6300 */
6293 6301 void
6294 6302 spa_sync_allpools(void)
6295 6303 {
6296 6304 spa_t *spa = NULL;
6297 6305 mutex_enter(&spa_namespace_lock);
6298 6306 while ((spa = spa_next(spa)) != NULL) {
6299 6307 if (spa_state(spa) != POOL_STATE_ACTIVE ||
6300 6308 !spa_writeable(spa) || spa_suspended(spa))
6301 6309 continue;
6302 6310 spa_open_ref(spa, FTAG);
6303 6311 mutex_exit(&spa_namespace_lock);
6304 6312 txg_wait_synced(spa_get_dsl(spa), 0);
6305 6313 mutex_enter(&spa_namespace_lock);
6306 6314 spa_close(spa, FTAG);
6307 6315 }
6308 6316 mutex_exit(&spa_namespace_lock);
6309 6317 }
6310 6318
6311 6319 /*
6312 6320 * ==========================================================================
6313 6321 * Miscellaneous routines
6314 6322 * ==========================================================================
6315 6323 */
6316 6324
6317 6325 /*
6318 6326 * Remove all pools in the system.
6319 6327 */
6320 6328 void
6321 6329 spa_evict_all(void)
6322 6330 {
6323 6331 spa_t *spa;
6324 6332
6325 6333 /*
6326 6334 * Remove all cached state. All pools should be closed now,
6327 6335 * so every spa in the AVL tree should be unreferenced.
6328 6336 */
6329 6337 mutex_enter(&spa_namespace_lock);
6330 6338 while ((spa = spa_next(NULL)) != NULL) {
6331 6339 /*
6332 6340 * Stop async tasks. The async thread may need to detach
6333 6341 * a device that's been replaced, which requires grabbing
6334 6342 * spa_namespace_lock, so we must drop it here.
6335 6343 */
6336 6344 spa_open_ref(spa, FTAG);
6337 6345 mutex_exit(&spa_namespace_lock);
6338 6346 spa_async_suspend(spa);
6339 6347 mutex_enter(&spa_namespace_lock);
6340 6348 spa_close(spa, FTAG);
6341 6349
6342 6350 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
6343 6351 spa_unload(spa);
6344 6352 spa_deactivate(spa);
6345 6353 }
6346 6354 spa_remove(spa);
6347 6355 }
6348 6356 mutex_exit(&spa_namespace_lock);
6349 6357 }
6350 6358
6351 6359 vdev_t *
6352 6360 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
6353 6361 {
6354 6362 vdev_t *vd;
6355 6363 int i;
6356 6364
6357 6365 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
6358 6366 return (vd);
6359 6367
6360 6368 if (aux) {
6361 6369 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
6362 6370 vd = spa->spa_l2cache.sav_vdevs[i];
6363 6371 if (vd->vdev_guid == guid)
6364 6372 return (vd);
6365 6373 }
6366 6374
6367 6375 for (i = 0; i < spa->spa_spares.sav_count; i++) {
6368 6376 vd = spa->spa_spares.sav_vdevs[i];
6369 6377 if (vd->vdev_guid == guid)
6370 6378 return (vd);
6371 6379 }
6372 6380 }
6373 6381
6374 6382 return (NULL);
6375 6383 }
6376 6384
6377 6385 void
6378 6386 spa_upgrade(spa_t *spa, uint64_t version)
6379 6387 {
6380 6388 ASSERT(spa_writeable(spa));
6381 6389
6382 6390 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6383 6391
6384 6392 /*
6385 6393 * This should only be called for a non-faulted pool, and since a
6386 6394 * future version would result in an unopenable pool, this shouldn't be
6387 6395 * possible.
6388 6396 */
6389 6397 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version));
6390 6398 ASSERT(version >= spa->spa_uberblock.ub_version);
6391 6399
6392 6400 spa->spa_uberblock.ub_version = version;
6393 6401 vdev_config_dirty(spa->spa_root_vdev);
6394 6402
6395 6403 spa_config_exit(spa, SCL_ALL, FTAG);
6396 6404
6397 6405 txg_wait_synced(spa_get_dsl(spa), 0);
6398 6406 }
6399 6407
6400 6408 boolean_t
6401 6409 spa_has_spare(spa_t *spa, uint64_t guid)
6402 6410 {
6403 6411 int i;
6404 6412 uint64_t spareguid;
6405 6413 spa_aux_vdev_t *sav = &spa->spa_spares;
6406 6414
6407 6415 for (i = 0; i < sav->sav_count; i++)
6408 6416 if (sav->sav_vdevs[i]->vdev_guid == guid)
6409 6417 return (B_TRUE);
6410 6418
6411 6419 for (i = 0; i < sav->sav_npending; i++) {
6412 6420 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
6413 6421 &spareguid) == 0 && spareguid == guid)
6414 6422 return (B_TRUE);
6415 6423 }
6416 6424
6417 6425 return (B_FALSE);
6418 6426 }
6419 6427
6420 6428 /*
6421 6429 * Check if a pool has an active shared spare device.
6422 6430 * Note: reference count of an active spare is 2, as a spare and as a replace
6423 6431 */
6424 6432 static boolean_t
6425 6433 spa_has_active_shared_spare(spa_t *spa)
6426 6434 {
6427 6435 int i, refcnt;
6428 6436 uint64_t pool;
6429 6437 spa_aux_vdev_t *sav = &spa->spa_spares;
6430 6438
6431 6439 for (i = 0; i < sav->sav_count; i++) {
6432 6440 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
6433 6441 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
6434 6442 refcnt > 2)
6435 6443 return (B_TRUE);
6436 6444 }
6437 6445
6438 6446 return (B_FALSE);
6439 6447 }
6440 6448
6441 6449 /*
6442 6450 * Post a sysevent corresponding to the given event. The 'name' must be one of
6443 6451 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
6444 6452 * filled in from the spa and (optionally) the vdev. This doesn't do anything
6445 6453 * in the userland libzpool, as we don't want consumers to misinterpret ztest
6446 6454 * or zdb as real changes.
6447 6455 */
6448 6456 void
6449 6457 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
6450 6458 {
6451 6459 #ifdef _KERNEL
6452 6460 sysevent_t *ev;
6453 6461 sysevent_attr_list_t *attr = NULL;
6454 6462 sysevent_value_t value;
6455 6463 sysevent_id_t eid;
6456 6464
6457 6465 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
6458 6466 SE_SLEEP);
6459 6467
6460 6468 value.value_type = SE_DATA_TYPE_STRING;
6461 6469 value.value.sv_string = spa_name(spa);
6462 6470 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
6463 6471 goto done;
6464 6472
6465 6473 value.value_type = SE_DATA_TYPE_UINT64;
6466 6474 value.value.sv_uint64 = spa_guid(spa);
6467 6475 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
6468 6476 goto done;
6469 6477
6470 6478 if (vd) {
6471 6479 value.value_type = SE_DATA_TYPE_UINT64;
6472 6480 value.value.sv_uint64 = vd->vdev_guid;
6473 6481 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
6474 6482 SE_SLEEP) != 0)
6475 6483 goto done;
6476 6484
6477 6485 if (vd->vdev_path) {
6478 6486 value.value_type = SE_DATA_TYPE_STRING;
6479 6487 value.value.sv_string = vd->vdev_path;
6480 6488 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
6481 6489 &value, SE_SLEEP) != 0)
6482 6490 goto done;
6483 6491 }
6484 6492 }
6485 6493
6486 6494 if (sysevent_attach_attributes(ev, attr) != 0)
6487 6495 goto done;
6488 6496 attr = NULL;
6489 6497
6490 6498 (void) log_sysevent(ev, SE_SLEEP, &eid);
6491 6499
6492 6500 done:
6493 6501 if (attr)
6494 6502 sysevent_free_attr(attr);
6495 6503 sysevent_free(ev);
6496 6504 #endif
6497 6505 }
↓ open down ↓ |
1183 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX