Print this page
3525 Persistent L2ARC
| 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_BATCH, /* cpu-intensive; value is ignored */
91 91 ZTI_MODE_NULL, /* don't create a taskq */
92 92 ZTI_NMODES
93 93 } zti_modes_t;
94 94
95 95 #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) }
96 96 #define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 }
97 97 #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 }
98 98
99 99 #define ZTI_N(n) ZTI_P(n, 1)
100 100 #define ZTI_ONE ZTI_N(1)
101 101
102 102 typedef struct zio_taskq_info {
103 103 zti_modes_t zti_mode;
104 104 uint_t zti_value;
105 105 uint_t zti_count;
106 106 } zio_taskq_info_t;
107 107
108 108 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
109 109 "issue", "issue_high", "intr", "intr_high"
110 110 };
111 111
112 112 /*
113 113 * This table defines the taskq settings for each ZFS I/O type. When
114 114 * initializing a pool, we use this table to create an appropriately sized
115 115 * taskq. Some operations are low volume and therefore have a small, static
116 116 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
117 117 * macros. Other operations process a large amount of data; the ZTI_BATCH
118 118 * macro causes us to create a taskq oriented for throughput. Some operations
119 119 * are so high frequency and short-lived that the taskq itself can become a a
120 120 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
121 121 * additional degree of parallelism specified by the number of threads per-
122 122 * taskq and the number of taskqs; when dispatching an event in this case, the
123 123 * particular taskq is chosen at random.
124 124 *
125 125 * The different taskq priorities are to handle the different contexts (issue
126 126 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
127 127 * need to be handled with minimum delay.
128 128 */
129 129 const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
130 130 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
131 131 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */
132 132 { ZTI_N(8), ZTI_NULL, ZTI_BATCH, ZTI_NULL }, /* READ */
133 133 { ZTI_BATCH, ZTI_N(5), ZTI_N(8), ZTI_N(5) }, /* WRITE */
134 134 { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */
135 135 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */
136 136 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */
137 137 };
138 138
139 139 static void spa_sync_version(void *arg, dmu_tx_t *tx);
140 140 static void spa_sync_props(void *arg, dmu_tx_t *tx);
141 141 static boolean_t spa_has_active_shared_spare(spa_t *spa);
142 142 static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
143 143 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
144 144 char **ereport);
145 145 static void spa_vdev_resilver_done(spa_t *spa);
146 146
147 147 uint_t zio_taskq_batch_pct = 75; /* 1 thread per cpu in pset */
148 148 id_t zio_taskq_psrset_bind = PS_NONE;
149 149 boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
150 150 uint_t zio_taskq_basedc = 80; /* base duty cycle */
151 151
152 152 boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
153 153 extern int zfs_sync_pass_deferred_free;
154 154
155 155 /*
156 156 * This (illegal) pool name is used when temporarily importing a spa_t in order
157 157 * to get the vdev stats associated with the imported devices.
158 158 */
159 159 #define TRYIMPORT_NAME "$import"
160 160
161 161 /*
162 162 * ==========================================================================
163 163 * SPA properties routines
164 164 * ==========================================================================
165 165 */
166 166
167 167 /*
168 168 * Add a (source=src, propname=propval) list to an nvlist.
169 169 */
170 170 static void
171 171 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
172 172 uint64_t intval, zprop_source_t src)
173 173 {
174 174 const char *propname = zpool_prop_to_name(prop);
175 175 nvlist_t *propval;
176 176
177 177 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
178 178 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
179 179
180 180 if (strval != NULL)
181 181 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
182 182 else
183 183 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
184 184
185 185 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
186 186 nvlist_free(propval);
187 187 }
188 188
189 189 /*
190 190 * Get property values from the spa configuration.
191 191 */
192 192 static void
193 193 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
194 194 {
195 195 vdev_t *rvd = spa->spa_root_vdev;
196 196 dsl_pool_t *pool = spa->spa_dsl_pool;
197 197 uint64_t size;
198 198 uint64_t alloc;
199 199 uint64_t space;
200 200 uint64_t cap, version;
201 201 zprop_source_t src = ZPROP_SRC_NONE;
202 202 spa_config_dirent_t *dp;
203 203
204 204 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
205 205
206 206 if (rvd != NULL) {
207 207 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
208 208 size = metaslab_class_get_space(spa_normal_class(spa));
209 209 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
210 210 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
211 211 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
212 212 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
213 213 size - alloc, src);
214 214
215 215 space = 0;
216 216 for (int c = 0; c < rvd->vdev_children; c++) {
217 217 vdev_t *tvd = rvd->vdev_child[c];
218 218 space += tvd->vdev_max_asize - tvd->vdev_asize;
219 219 }
220 220 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, space,
221 221 src);
222 222
223 223 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
224 224 (spa_mode(spa) == FREAD), src);
225 225
226 226 cap = (size == 0) ? 0 : (alloc * 100 / size);
227 227 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
228 228
229 229 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
230 230 ddt_get_pool_dedup_ratio(spa), src);
231 231
232 232 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
233 233 rvd->vdev_state, src);
234 234
235 235 version = spa_version(spa);
236 236 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
237 237 src = ZPROP_SRC_DEFAULT;
238 238 else
239 239 src = ZPROP_SRC_LOCAL;
240 240 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
241 241 }
242 242
243 243 if (pool != NULL) {
244 244 dsl_dir_t *freedir = pool->dp_free_dir;
245 245
246 246 /*
247 247 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
248 248 * when opening pools before this version freedir will be NULL.
249 249 */
250 250 if (freedir != NULL) {
251 251 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
252 252 freedir->dd_phys->dd_used_bytes, src);
253 253 } else {
254 254 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
255 255 NULL, 0, src);
256 256 }
257 257 }
258 258
259 259 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
260 260
261 261 if (spa->spa_comment != NULL) {
262 262 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
263 263 0, ZPROP_SRC_LOCAL);
264 264 }
265 265
266 266 if (spa->spa_root != NULL)
267 267 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
268 268 0, ZPROP_SRC_LOCAL);
269 269
270 270 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
271 271 if (dp->scd_path == NULL) {
272 272 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
273 273 "none", 0, ZPROP_SRC_LOCAL);
274 274 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
275 275 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
276 276 dp->scd_path, 0, ZPROP_SRC_LOCAL);
277 277 }
278 278 }
279 279 }
280 280
281 281 /*
282 282 * Get zpool property values.
283 283 */
284 284 int
285 285 spa_prop_get(spa_t *spa, nvlist_t **nvp)
286 286 {
287 287 objset_t *mos = spa->spa_meta_objset;
288 288 zap_cursor_t zc;
289 289 zap_attribute_t za;
290 290 int err;
291 291
292 292 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
293 293
294 294 mutex_enter(&spa->spa_props_lock);
295 295
296 296 /*
297 297 * Get properties from the spa config.
298 298 */
299 299 spa_prop_get_config(spa, nvp);
300 300
301 301 /* If no pool property object, no more prop to get. */
302 302 if (mos == NULL || spa->spa_pool_props_object == 0) {
303 303 mutex_exit(&spa->spa_props_lock);
304 304 return (0);
305 305 }
306 306
307 307 /*
308 308 * Get properties from the MOS pool property object.
309 309 */
310 310 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
311 311 (err = zap_cursor_retrieve(&zc, &za)) == 0;
312 312 zap_cursor_advance(&zc)) {
313 313 uint64_t intval = 0;
314 314 char *strval = NULL;
315 315 zprop_source_t src = ZPROP_SRC_DEFAULT;
316 316 zpool_prop_t prop;
317 317
318 318 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
319 319 continue;
320 320
321 321 switch (za.za_integer_length) {
322 322 case 8:
323 323 /* integer property */
324 324 if (za.za_first_integer !=
325 325 zpool_prop_default_numeric(prop))
326 326 src = ZPROP_SRC_LOCAL;
327 327
328 328 if (prop == ZPOOL_PROP_BOOTFS) {
329 329 dsl_pool_t *dp;
330 330 dsl_dataset_t *ds = NULL;
331 331
332 332 dp = spa_get_dsl(spa);
333 333 dsl_pool_config_enter(dp, FTAG);
334 334 if (err = dsl_dataset_hold_obj(dp,
335 335 za.za_first_integer, FTAG, &ds)) {
336 336 dsl_pool_config_exit(dp, FTAG);
337 337 break;
338 338 }
339 339
340 340 strval = kmem_alloc(
341 341 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
342 342 KM_SLEEP);
343 343 dsl_dataset_name(ds, strval);
344 344 dsl_dataset_rele(ds, FTAG);
345 345 dsl_pool_config_exit(dp, FTAG);
346 346 } else {
347 347 strval = NULL;
348 348 intval = za.za_first_integer;
349 349 }
350 350
351 351 spa_prop_add_list(*nvp, prop, strval, intval, src);
352 352
353 353 if (strval != NULL)
354 354 kmem_free(strval,
355 355 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
356 356
357 357 break;
358 358
359 359 case 1:
360 360 /* string property */
361 361 strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
362 362 err = zap_lookup(mos, spa->spa_pool_props_object,
363 363 za.za_name, 1, za.za_num_integers, strval);
364 364 if (err) {
365 365 kmem_free(strval, za.za_num_integers);
366 366 break;
367 367 }
368 368 spa_prop_add_list(*nvp, prop, strval, 0, src);
369 369 kmem_free(strval, za.za_num_integers);
370 370 break;
371 371
372 372 default:
373 373 break;
374 374 }
375 375 }
376 376 zap_cursor_fini(&zc);
377 377 mutex_exit(&spa->spa_props_lock);
378 378 out:
379 379 if (err && err != ENOENT) {
380 380 nvlist_free(*nvp);
381 381 *nvp = NULL;
382 382 return (err);
383 383 }
384 384
385 385 return (0);
386 386 }
387 387
388 388 /*
389 389 * Validate the given pool properties nvlist and modify the list
390 390 * for the property values to be set.
391 391 */
392 392 static int
393 393 spa_prop_validate(spa_t *spa, nvlist_t *props)
394 394 {
395 395 nvpair_t *elem;
396 396 int error = 0, reset_bootfs = 0;
397 397 uint64_t objnum = 0;
398 398 boolean_t has_feature = B_FALSE;
399 399
400 400 elem = NULL;
401 401 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
402 402 uint64_t intval;
403 403 char *strval, *slash, *check, *fname;
404 404 const char *propname = nvpair_name(elem);
405 405 zpool_prop_t prop = zpool_name_to_prop(propname);
406 406
407 407 switch (prop) {
408 408 case ZPROP_INVAL:
409 409 if (!zpool_prop_feature(propname)) {
410 410 error = SET_ERROR(EINVAL);
411 411 break;
412 412 }
413 413
414 414 /*
415 415 * Sanitize the input.
416 416 */
417 417 if (nvpair_type(elem) != DATA_TYPE_UINT64) {
418 418 error = SET_ERROR(EINVAL);
419 419 break;
420 420 }
421 421
422 422 if (nvpair_value_uint64(elem, &intval) != 0) {
423 423 error = SET_ERROR(EINVAL);
424 424 break;
425 425 }
426 426
427 427 if (intval != 0) {
428 428 error = SET_ERROR(EINVAL);
429 429 break;
430 430 }
431 431
432 432 fname = strchr(propname, '@') + 1;
433 433 if (zfeature_lookup_name(fname, NULL) != 0) {
434 434 error = SET_ERROR(EINVAL);
435 435 break;
436 436 }
437 437
438 438 has_feature = B_TRUE;
439 439 break;
440 440
441 441 case ZPOOL_PROP_VERSION:
442 442 error = nvpair_value_uint64(elem, &intval);
443 443 if (!error &&
444 444 (intval < spa_version(spa) ||
445 445 intval > SPA_VERSION_BEFORE_FEATURES ||
446 446 has_feature))
447 447 error = SET_ERROR(EINVAL);
448 448 break;
449 449
450 450 case ZPOOL_PROP_DELEGATION:
451 451 case ZPOOL_PROP_AUTOREPLACE:
452 452 case ZPOOL_PROP_LISTSNAPS:
453 453 case ZPOOL_PROP_AUTOEXPAND:
454 454 error = nvpair_value_uint64(elem, &intval);
455 455 if (!error && intval > 1)
456 456 error = SET_ERROR(EINVAL);
457 457 break;
458 458
459 459 case ZPOOL_PROP_BOOTFS:
460 460 /*
461 461 * If the pool version is less than SPA_VERSION_BOOTFS,
462 462 * or the pool is still being created (version == 0),
463 463 * the bootfs property cannot be set.
464 464 */
465 465 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
466 466 error = SET_ERROR(ENOTSUP);
467 467 break;
468 468 }
469 469
470 470 /*
471 471 * Make sure the vdev config is bootable
472 472 */
473 473 if (!vdev_is_bootable(spa->spa_root_vdev)) {
474 474 error = SET_ERROR(ENOTSUP);
475 475 break;
476 476 }
477 477
478 478 reset_bootfs = 1;
479 479
480 480 error = nvpair_value_string(elem, &strval);
481 481
482 482 if (!error) {
483 483 objset_t *os;
484 484 uint64_t compress;
485 485
486 486 if (strval == NULL || strval[0] == '\0') {
487 487 objnum = zpool_prop_default_numeric(
488 488 ZPOOL_PROP_BOOTFS);
489 489 break;
490 490 }
491 491
492 492 if (error = dmu_objset_hold(strval, FTAG, &os))
493 493 break;
494 494
495 495 /* Must be ZPL and not gzip compressed. */
496 496
497 497 if (dmu_objset_type(os) != DMU_OST_ZFS) {
498 498 error = SET_ERROR(ENOTSUP);
499 499 } else if ((error =
500 500 dsl_prop_get_int_ds(dmu_objset_ds(os),
501 501 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
502 502 &compress)) == 0 &&
503 503 !BOOTFS_COMPRESS_VALID(compress)) {
504 504 error = SET_ERROR(ENOTSUP);
505 505 } else {
506 506 objnum = dmu_objset_id(os);
507 507 }
508 508 dmu_objset_rele(os, FTAG);
509 509 }
510 510 break;
511 511
512 512 case ZPOOL_PROP_FAILUREMODE:
513 513 error = nvpair_value_uint64(elem, &intval);
514 514 if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
515 515 intval > ZIO_FAILURE_MODE_PANIC))
516 516 error = SET_ERROR(EINVAL);
517 517
518 518 /*
519 519 * This is a special case which only occurs when
520 520 * the pool has completely failed. This allows
521 521 * the user to change the in-core failmode property
522 522 * without syncing it out to disk (I/Os might
523 523 * currently be blocked). We do this by returning
524 524 * EIO to the caller (spa_prop_set) to trick it
525 525 * into thinking we encountered a property validation
526 526 * error.
527 527 */
528 528 if (!error && spa_suspended(spa)) {
529 529 spa->spa_failmode = intval;
530 530 error = SET_ERROR(EIO);
531 531 }
532 532 break;
533 533
534 534 case ZPOOL_PROP_CACHEFILE:
535 535 if ((error = nvpair_value_string(elem, &strval)) != 0)
536 536 break;
537 537
538 538 if (strval[0] == '\0')
539 539 break;
540 540
541 541 if (strcmp(strval, "none") == 0)
542 542 break;
543 543
544 544 if (strval[0] != '/') {
545 545 error = SET_ERROR(EINVAL);
546 546 break;
547 547 }
548 548
549 549 slash = strrchr(strval, '/');
550 550 ASSERT(slash != NULL);
551 551
552 552 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
553 553 strcmp(slash, "/..") == 0)
554 554 error = SET_ERROR(EINVAL);
555 555 break;
556 556
557 557 case ZPOOL_PROP_COMMENT:
558 558 if ((error = nvpair_value_string(elem, &strval)) != 0)
559 559 break;
560 560 for (check = strval; *check != '\0'; check++) {
561 561 /*
562 562 * The kernel doesn't have an easy isprint()
563 563 * check. For this kernel check, we merely
564 564 * check ASCII apart from DEL. Fix this if
565 565 * there is an easy-to-use kernel isprint().
566 566 */
567 567 if (*check >= 0x7f) {
568 568 error = SET_ERROR(EINVAL);
569 569 break;
570 570 }
571 571 check++;
572 572 }
573 573 if (strlen(strval) > ZPROP_MAX_COMMENT)
574 574 error = E2BIG;
575 575 break;
576 576
577 577 case ZPOOL_PROP_DEDUPDITTO:
578 578 if (spa_version(spa) < SPA_VERSION_DEDUP)
579 579 error = SET_ERROR(ENOTSUP);
580 580 else
581 581 error = nvpair_value_uint64(elem, &intval);
582 582 if (error == 0 &&
583 583 intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
584 584 error = SET_ERROR(EINVAL);
585 585 break;
586 586 }
587 587
588 588 if (error)
589 589 break;
590 590 }
591 591
592 592 if (!error && reset_bootfs) {
593 593 error = nvlist_remove(props,
594 594 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
595 595
596 596 if (!error) {
597 597 error = nvlist_add_uint64(props,
598 598 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
599 599 }
600 600 }
601 601
602 602 return (error);
603 603 }
604 604
605 605 void
606 606 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
607 607 {
608 608 char *cachefile;
609 609 spa_config_dirent_t *dp;
610 610
611 611 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
612 612 &cachefile) != 0)
613 613 return;
614 614
615 615 dp = kmem_alloc(sizeof (spa_config_dirent_t),
616 616 KM_SLEEP);
617 617
618 618 if (cachefile[0] == '\0')
619 619 dp->scd_path = spa_strdup(spa_config_path);
620 620 else if (strcmp(cachefile, "none") == 0)
621 621 dp->scd_path = NULL;
622 622 else
623 623 dp->scd_path = spa_strdup(cachefile);
624 624
625 625 list_insert_head(&spa->spa_config_list, dp);
626 626 if (need_sync)
627 627 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
628 628 }
629 629
630 630 int
631 631 spa_prop_set(spa_t *spa, nvlist_t *nvp)
632 632 {
633 633 int error;
634 634 nvpair_t *elem = NULL;
635 635 boolean_t need_sync = B_FALSE;
636 636
637 637 if ((error = spa_prop_validate(spa, nvp)) != 0)
638 638 return (error);
639 639
640 640 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
641 641 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
642 642
643 643 if (prop == ZPOOL_PROP_CACHEFILE ||
644 644 prop == ZPOOL_PROP_ALTROOT ||
645 645 prop == ZPOOL_PROP_READONLY)
646 646 continue;
647 647
648 648 if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) {
649 649 uint64_t ver;
650 650
651 651 if (prop == ZPOOL_PROP_VERSION) {
652 652 VERIFY(nvpair_value_uint64(elem, &ver) == 0);
653 653 } else {
654 654 ASSERT(zpool_prop_feature(nvpair_name(elem)));
655 655 ver = SPA_VERSION_FEATURES;
656 656 need_sync = B_TRUE;
657 657 }
658 658
659 659 /* Save time if the version is already set. */
660 660 if (ver == spa_version(spa))
661 661 continue;
662 662
663 663 /*
664 664 * In addition to the pool directory object, we might
665 665 * create the pool properties object, the features for
666 666 * read object, the features for write object, or the
667 667 * feature descriptions object.
668 668 */
669 669 error = dsl_sync_task(spa->spa_name, NULL,
670 670 spa_sync_version, &ver, 6);
671 671 if (error)
672 672 return (error);
673 673 continue;
674 674 }
675 675
676 676 need_sync = B_TRUE;
677 677 break;
678 678 }
679 679
680 680 if (need_sync) {
681 681 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props,
682 682 nvp, 6));
683 683 }
684 684
685 685 return (0);
686 686 }
687 687
688 688 /*
689 689 * If the bootfs property value is dsobj, clear it.
690 690 */
691 691 void
692 692 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
693 693 {
694 694 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
695 695 VERIFY(zap_remove(spa->spa_meta_objset,
696 696 spa->spa_pool_props_object,
697 697 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
698 698 spa->spa_bootfs = 0;
699 699 }
700 700 }
701 701
702 702 /*ARGSUSED*/
703 703 static int
704 704 spa_change_guid_check(void *arg, dmu_tx_t *tx)
705 705 {
706 706 uint64_t *newguid = arg;
707 707 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
708 708 vdev_t *rvd = spa->spa_root_vdev;
709 709 uint64_t vdev_state;
710 710
711 711 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
712 712 vdev_state = rvd->vdev_state;
713 713 spa_config_exit(spa, SCL_STATE, FTAG);
714 714
715 715 if (vdev_state != VDEV_STATE_HEALTHY)
716 716 return (SET_ERROR(ENXIO));
717 717
718 718 ASSERT3U(spa_guid(spa), !=, *newguid);
719 719
720 720 return (0);
721 721 }
722 722
723 723 static void
724 724 spa_change_guid_sync(void *arg, dmu_tx_t *tx)
725 725 {
726 726 uint64_t *newguid = arg;
727 727 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
728 728 uint64_t oldguid;
729 729 vdev_t *rvd = spa->spa_root_vdev;
730 730
731 731 oldguid = spa_guid(spa);
732 732
733 733 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
734 734 rvd->vdev_guid = *newguid;
735 735 rvd->vdev_guid_sum += (*newguid - oldguid);
736 736 vdev_config_dirty(rvd);
737 737 spa_config_exit(spa, SCL_STATE, FTAG);
738 738
739 739 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu",
740 740 oldguid, *newguid);
741 741 }
742 742
743 743 /*
744 744 * Change the GUID for the pool. This is done so that we can later
745 745 * re-import a pool built from a clone of our own vdevs. We will modify
746 746 * the root vdev's guid, our own pool guid, and then mark all of our
747 747 * vdevs dirty. Note that we must make sure that all our vdevs are
748 748 * online when we do this, or else any vdevs that weren't present
749 749 * would be orphaned from our pool. We are also going to issue a
750 750 * sysevent to update any watchers.
751 751 */
752 752 int
753 753 spa_change_guid(spa_t *spa)
754 754 {
755 755 int error;
756 756 uint64_t guid;
757 757
758 758 mutex_enter(&spa->spa_vdev_top_lock);
759 759 mutex_enter(&spa_namespace_lock);
760 760 guid = spa_generate_guid(NULL);
761 761
762 762 error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
763 763 spa_change_guid_sync, &guid, 5);
764 764
765 765 if (error == 0) {
766 766 spa_config_sync(spa, B_FALSE, B_TRUE);
767 767 spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
768 768 }
769 769
770 770 mutex_exit(&spa_namespace_lock);
771 771 mutex_exit(&spa->spa_vdev_top_lock);
772 772
773 773 return (error);
774 774 }
775 775
776 776 /*
777 777 * ==========================================================================
778 778 * SPA state manipulation (open/create/destroy/import/export)
779 779 * ==========================================================================
780 780 */
781 781
782 782 static int
783 783 spa_error_entry_compare(const void *a, const void *b)
784 784 {
785 785 spa_error_entry_t *sa = (spa_error_entry_t *)a;
786 786 spa_error_entry_t *sb = (spa_error_entry_t *)b;
787 787 int ret;
788 788
789 789 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
790 790 sizeof (zbookmark_t));
791 791
792 792 if (ret < 0)
793 793 return (-1);
794 794 else if (ret > 0)
795 795 return (1);
796 796 else
797 797 return (0);
798 798 }
799 799
800 800 /*
801 801 * Utility function which retrieves copies of the current logs and
802 802 * re-initializes them in the process.
803 803 */
804 804 void
805 805 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
806 806 {
807 807 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
808 808
809 809 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
810 810 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
811 811
812 812 avl_create(&spa->spa_errlist_scrub,
813 813 spa_error_entry_compare, sizeof (spa_error_entry_t),
814 814 offsetof(spa_error_entry_t, se_avl));
815 815 avl_create(&spa->spa_errlist_last,
816 816 spa_error_entry_compare, sizeof (spa_error_entry_t),
817 817 offsetof(spa_error_entry_t, se_avl));
818 818 }
819 819
820 820 static void
821 821 spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
822 822 {
823 823 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
824 824 enum zti_modes mode = ztip->zti_mode;
825 825 uint_t value = ztip->zti_value;
826 826 uint_t count = ztip->zti_count;
827 827 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
828 828 char name[32];
829 829 uint_t flags = 0;
830 830 boolean_t batch = B_FALSE;
831 831
832 832 if (mode == ZTI_MODE_NULL) {
833 833 tqs->stqs_count = 0;
834 834 tqs->stqs_taskq = NULL;
835 835 return;
836 836 }
837 837
838 838 ASSERT3U(count, >, 0);
839 839
840 840 tqs->stqs_count = count;
841 841 tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP);
842 842
843 843 switch (mode) {
844 844 case ZTI_MODE_FIXED:
845 845 ASSERT3U(value, >=, 1);
846 846 value = MAX(value, 1);
847 847 break;
848 848
849 849 case ZTI_MODE_BATCH:
850 850 batch = B_TRUE;
851 851 flags |= TASKQ_THREADS_CPU_PCT;
852 852 value = zio_taskq_batch_pct;
853 853 break;
854 854
855 855 default:
856 856 panic("unrecognized mode for %s_%s taskq (%u:%u) in "
857 857 "spa_activate()",
858 858 zio_type_name[t], zio_taskq_types[q], mode, value);
859 859 break;
860 860 }
861 861
862 862 for (uint_t i = 0; i < count; i++) {
863 863 taskq_t *tq;
864 864
865 865 if (count > 1) {
866 866 (void) snprintf(name, sizeof (name), "%s_%s_%u",
867 867 zio_type_name[t], zio_taskq_types[q], i);
868 868 } else {
869 869 (void) snprintf(name, sizeof (name), "%s_%s",
870 870 zio_type_name[t], zio_taskq_types[q]);
871 871 }
872 872
873 873 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
874 874 if (batch)
875 875 flags |= TASKQ_DC_BATCH;
876 876
877 877 tq = taskq_create_sysdc(name, value, 50, INT_MAX,
878 878 spa->spa_proc, zio_taskq_basedc, flags);
879 879 } else {
880 880 pri_t pri = maxclsyspri;
881 881 /*
882 882 * The write issue taskq can be extremely CPU
883 883 * intensive. Run it at slightly lower priority
884 884 * than the other taskqs.
885 885 */
886 886 if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE)
887 887 pri--;
888 888
889 889 tq = taskq_create_proc(name, value, pri, 50,
890 890 INT_MAX, spa->spa_proc, flags);
891 891 }
892 892
893 893 tqs->stqs_taskq[i] = tq;
894 894 }
895 895 }
896 896
897 897 static void
898 898 spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
899 899 {
900 900 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
901 901
902 902 if (tqs->stqs_taskq == NULL) {
903 903 ASSERT0(tqs->stqs_count);
904 904 return;
905 905 }
906 906
907 907 for (uint_t i = 0; i < tqs->stqs_count; i++) {
908 908 ASSERT3P(tqs->stqs_taskq[i], !=, NULL);
909 909 taskq_destroy(tqs->stqs_taskq[i]);
910 910 }
911 911
912 912 kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *));
913 913 tqs->stqs_taskq = NULL;
914 914 }
915 915
916 916 /*
917 917 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
918 918 * Note that a type may have multiple discrete taskqs to avoid lock contention
919 919 * on the taskq itself. In that case we choose which taskq at random by using
920 920 * the low bits of gethrtime().
921 921 */
922 922 void
923 923 spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
924 924 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent)
925 925 {
926 926 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
927 927 taskq_t *tq;
928 928
929 929 ASSERT3P(tqs->stqs_taskq, !=, NULL);
930 930 ASSERT3U(tqs->stqs_count, !=, 0);
931 931
932 932 if (tqs->stqs_count == 1) {
933 933 tq = tqs->stqs_taskq[0];
934 934 } else {
935 935 tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count];
936 936 }
937 937
938 938 taskq_dispatch_ent(tq, func, arg, flags, ent);
939 939 }
940 940
941 941 static void
942 942 spa_create_zio_taskqs(spa_t *spa)
943 943 {
944 944 for (int t = 0; t < ZIO_TYPES; t++) {
945 945 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
946 946 spa_taskqs_init(spa, t, q);
947 947 }
948 948 }
949 949 }
950 950
951 951 #ifdef _KERNEL
952 952 static void
953 953 spa_thread(void *arg)
954 954 {
955 955 callb_cpr_t cprinfo;
956 956
957 957 spa_t *spa = arg;
958 958 user_t *pu = PTOU(curproc);
959 959
960 960 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
961 961 spa->spa_name);
962 962
963 963 ASSERT(curproc != &p0);
964 964 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
965 965 "zpool-%s", spa->spa_name);
966 966 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
967 967
968 968 /* bind this thread to the requested psrset */
969 969 if (zio_taskq_psrset_bind != PS_NONE) {
970 970 pool_lock();
971 971 mutex_enter(&cpu_lock);
972 972 mutex_enter(&pidlock);
973 973 mutex_enter(&curproc->p_lock);
974 974
975 975 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
976 976 0, NULL, NULL) == 0) {
977 977 curthread->t_bind_pset = zio_taskq_psrset_bind;
978 978 } else {
979 979 cmn_err(CE_WARN,
980 980 "Couldn't bind process for zfs pool \"%s\" to "
981 981 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
982 982 }
983 983
984 984 mutex_exit(&curproc->p_lock);
985 985 mutex_exit(&pidlock);
986 986 mutex_exit(&cpu_lock);
987 987 pool_unlock();
988 988 }
989 989
990 990 if (zio_taskq_sysdc) {
991 991 sysdc_thread_enter(curthread, 100, 0);
992 992 }
993 993
994 994 spa->spa_proc = curproc;
995 995 spa->spa_did = curthread->t_did;
996 996
997 997 spa_create_zio_taskqs(spa);
998 998
999 999 mutex_enter(&spa->spa_proc_lock);
1000 1000 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
1001 1001
1002 1002 spa->spa_proc_state = SPA_PROC_ACTIVE;
1003 1003 cv_broadcast(&spa->spa_proc_cv);
1004 1004
1005 1005 CALLB_CPR_SAFE_BEGIN(&cprinfo);
1006 1006 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
1007 1007 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1008 1008 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
1009 1009
1010 1010 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
1011 1011 spa->spa_proc_state = SPA_PROC_GONE;
1012 1012 spa->spa_proc = &p0;
1013 1013 cv_broadcast(&spa->spa_proc_cv);
1014 1014 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
1015 1015
1016 1016 mutex_enter(&curproc->p_lock);
1017 1017 lwp_exit();
1018 1018 }
1019 1019 #endif
1020 1020
1021 1021 /*
1022 1022 * Activate an uninitialized pool.
1023 1023 */
1024 1024 static void
1025 1025 spa_activate(spa_t *spa, int mode)
1026 1026 {
1027 1027 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
1028 1028
1029 1029 spa->spa_state = POOL_STATE_ACTIVE;
1030 1030 spa->spa_mode = mode;
1031 1031
1032 1032 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
1033 1033 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
1034 1034
1035 1035 /* Try to create a covering process */
1036 1036 mutex_enter(&spa->spa_proc_lock);
1037 1037 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
1038 1038 ASSERT(spa->spa_proc == &p0);
1039 1039 spa->spa_did = 0;
1040 1040
1041 1041 /* Only create a process if we're going to be around a while. */
1042 1042 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
1043 1043 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
1044 1044 NULL, 0) == 0) {
1045 1045 spa->spa_proc_state = SPA_PROC_CREATED;
1046 1046 while (spa->spa_proc_state == SPA_PROC_CREATED) {
1047 1047 cv_wait(&spa->spa_proc_cv,
1048 1048 &spa->spa_proc_lock);
1049 1049 }
1050 1050 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1051 1051 ASSERT(spa->spa_proc != &p0);
1052 1052 ASSERT(spa->spa_did != 0);
1053 1053 } else {
1054 1054 #ifdef _KERNEL
1055 1055 cmn_err(CE_WARN,
1056 1056 "Couldn't create process for zfs pool \"%s\"\n",
1057 1057 spa->spa_name);
1058 1058 #endif
1059 1059 }
1060 1060 }
1061 1061 mutex_exit(&spa->spa_proc_lock);
1062 1062
1063 1063 /* If we didn't create a process, we need to create our taskqs. */
1064 1064 if (spa->spa_proc == &p0) {
1065 1065 spa_create_zio_taskqs(spa);
1066 1066 }
1067 1067
1068 1068 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1069 1069 offsetof(vdev_t, vdev_config_dirty_node));
1070 1070 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1071 1071 offsetof(vdev_t, vdev_state_dirty_node));
1072 1072
1073 1073 txg_list_create(&spa->spa_vdev_txg_list,
1074 1074 offsetof(struct vdev, vdev_txg_node));
1075 1075
1076 1076 avl_create(&spa->spa_errlist_scrub,
1077 1077 spa_error_entry_compare, sizeof (spa_error_entry_t),
1078 1078 offsetof(spa_error_entry_t, se_avl));
1079 1079 avl_create(&spa->spa_errlist_last,
1080 1080 spa_error_entry_compare, sizeof (spa_error_entry_t),
1081 1081 offsetof(spa_error_entry_t, se_avl));
1082 1082 }
1083 1083
1084 1084 /*
1085 1085 * Opposite of spa_activate().
1086 1086 */
1087 1087 static void
1088 1088 spa_deactivate(spa_t *spa)
1089 1089 {
1090 1090 ASSERT(spa->spa_sync_on == B_FALSE);
1091 1091 ASSERT(spa->spa_dsl_pool == NULL);
1092 1092 ASSERT(spa->spa_root_vdev == NULL);
1093 1093 ASSERT(spa->spa_async_zio_root == NULL);
1094 1094 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1095 1095
1096 1096 txg_list_destroy(&spa->spa_vdev_txg_list);
1097 1097
1098 1098 list_destroy(&spa->spa_config_dirty_list);
1099 1099 list_destroy(&spa->spa_state_dirty_list);
1100 1100
1101 1101 for (int t = 0; t < ZIO_TYPES; t++) {
1102 1102 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1103 1103 spa_taskqs_fini(spa, t, q);
1104 1104 }
1105 1105 }
1106 1106
1107 1107 metaslab_class_destroy(spa->spa_normal_class);
1108 1108 spa->spa_normal_class = NULL;
1109 1109
1110 1110 metaslab_class_destroy(spa->spa_log_class);
1111 1111 spa->spa_log_class = NULL;
1112 1112
1113 1113 /*
1114 1114 * If this was part of an import or the open otherwise failed, we may
1115 1115 * still have errors left in the queues. Empty them just in case.
1116 1116 */
1117 1117 spa_errlog_drain(spa);
1118 1118
1119 1119 avl_destroy(&spa->spa_errlist_scrub);
1120 1120 avl_destroy(&spa->spa_errlist_last);
1121 1121
1122 1122 spa->spa_state = POOL_STATE_UNINITIALIZED;
1123 1123
1124 1124 mutex_enter(&spa->spa_proc_lock);
1125 1125 if (spa->spa_proc_state != SPA_PROC_NONE) {
1126 1126 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1127 1127 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1128 1128 cv_broadcast(&spa->spa_proc_cv);
1129 1129 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1130 1130 ASSERT(spa->spa_proc != &p0);
1131 1131 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1132 1132 }
1133 1133 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1134 1134 spa->spa_proc_state = SPA_PROC_NONE;
1135 1135 }
1136 1136 ASSERT(spa->spa_proc == &p0);
1137 1137 mutex_exit(&spa->spa_proc_lock);
1138 1138
1139 1139 /*
1140 1140 * We want to make sure spa_thread() has actually exited the ZFS
1141 1141 * module, so that the module can't be unloaded out from underneath
1142 1142 * it.
1143 1143 */
1144 1144 if (spa->spa_did != 0) {
1145 1145 thread_join(spa->spa_did);
1146 1146 spa->spa_did = 0;
1147 1147 }
1148 1148 }
1149 1149
1150 1150 /*
1151 1151 * Verify a pool configuration, and construct the vdev tree appropriately. This
1152 1152 * will create all the necessary vdevs in the appropriate layout, with each vdev
1153 1153 * in the CLOSED state. This will prep the pool before open/creation/import.
1154 1154 * All vdev validation is done by the vdev_alloc() routine.
1155 1155 */
1156 1156 static int
1157 1157 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1158 1158 uint_t id, int atype)
1159 1159 {
1160 1160 nvlist_t **child;
1161 1161 uint_t children;
1162 1162 int error;
1163 1163
1164 1164 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1165 1165 return (error);
1166 1166
1167 1167 if ((*vdp)->vdev_ops->vdev_op_leaf)
1168 1168 return (0);
1169 1169
1170 1170 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1171 1171 &child, &children);
1172 1172
1173 1173 if (error == ENOENT)
1174 1174 return (0);
1175 1175
1176 1176 if (error) {
1177 1177 vdev_free(*vdp);
1178 1178 *vdp = NULL;
1179 1179 return (SET_ERROR(EINVAL));
1180 1180 }
1181 1181
1182 1182 for (int c = 0; c < children; c++) {
1183 1183 vdev_t *vd;
1184 1184 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1185 1185 atype)) != 0) {
1186 1186 vdev_free(*vdp);
1187 1187 *vdp = NULL;
1188 1188 return (error);
1189 1189 }
1190 1190 }
1191 1191
1192 1192 ASSERT(*vdp != NULL);
1193 1193
1194 1194 return (0);
1195 1195 }
1196 1196
1197 1197 /*
1198 1198 * Opposite of spa_load().
1199 1199 */
1200 1200 static void
1201 1201 spa_unload(spa_t *spa)
1202 1202 {
1203 1203 int i;
1204 1204
1205 1205 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1206 1206
1207 1207 /*
1208 1208 * Stop async tasks.
1209 1209 */
1210 1210 spa_async_suspend(spa);
1211 1211
1212 1212 /*
1213 1213 * Stop syncing.
1214 1214 */
1215 1215 if (spa->spa_sync_on) {
1216 1216 txg_sync_stop(spa->spa_dsl_pool);
1217 1217 spa->spa_sync_on = B_FALSE;
1218 1218 }
1219 1219
1220 1220 /*
1221 1221 * Wait for any outstanding async I/O to complete.
1222 1222 */
1223 1223 if (spa->spa_async_zio_root != NULL) {
1224 1224 (void) zio_wait(spa->spa_async_zio_root);
1225 1225 spa->spa_async_zio_root = NULL;
1226 1226 }
1227 1227
1228 1228 bpobj_close(&spa->spa_deferred_bpobj);
1229 1229
1230 1230 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1231 1231
1232 1232 /*
1233 1233 * Close all vdevs.
1234 1234 */
1235 1235 if (spa->spa_root_vdev)
1236 1236 vdev_free(spa->spa_root_vdev);
1237 1237 ASSERT(spa->spa_root_vdev == NULL);
1238 1238
1239 1239 /*
1240 1240 * Close the dsl pool.
1241 1241 */
1242 1242 if (spa->spa_dsl_pool) {
1243 1243 dsl_pool_close(spa->spa_dsl_pool);
1244 1244 spa->spa_dsl_pool = NULL;
1245 1245 spa->spa_meta_objset = NULL;
1246 1246 }
1247 1247
1248 1248 ddt_unload(spa);
1249 1249
1250 1250
1251 1251 /*
1252 1252 * Drop and purge level 2 cache
1253 1253 */
1254 1254 spa_l2cache_drop(spa);
1255 1255
1256 1256 for (i = 0; i < spa->spa_spares.sav_count; i++)
1257 1257 vdev_free(spa->spa_spares.sav_vdevs[i]);
1258 1258 if (spa->spa_spares.sav_vdevs) {
1259 1259 kmem_free(spa->spa_spares.sav_vdevs,
1260 1260 spa->spa_spares.sav_count * sizeof (void *));
1261 1261 spa->spa_spares.sav_vdevs = NULL;
1262 1262 }
1263 1263 if (spa->spa_spares.sav_config) {
1264 1264 nvlist_free(spa->spa_spares.sav_config);
1265 1265 spa->spa_spares.sav_config = NULL;
1266 1266 }
1267 1267 spa->spa_spares.sav_count = 0;
1268 1268
1269 1269 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1270 1270 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1271 1271 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1272 1272 }
1273 1273 if (spa->spa_l2cache.sav_vdevs) {
1274 1274 kmem_free(spa->spa_l2cache.sav_vdevs,
1275 1275 spa->spa_l2cache.sav_count * sizeof (void *));
1276 1276 spa->spa_l2cache.sav_vdevs = NULL;
1277 1277 }
1278 1278 if (spa->spa_l2cache.sav_config) {
1279 1279 nvlist_free(spa->spa_l2cache.sav_config);
1280 1280 spa->spa_l2cache.sav_config = NULL;
1281 1281 }
1282 1282 spa->spa_l2cache.sav_count = 0;
1283 1283
1284 1284 spa->spa_async_suspended = 0;
1285 1285
1286 1286 if (spa->spa_comment != NULL) {
1287 1287 spa_strfree(spa->spa_comment);
1288 1288 spa->spa_comment = NULL;
1289 1289 }
1290 1290
1291 1291 spa_config_exit(spa, SCL_ALL, FTAG);
1292 1292 }
1293 1293
1294 1294 /*
1295 1295 * Load (or re-load) the current list of vdevs describing the active spares for
1296 1296 * this pool. When this is called, we have some form of basic information in
1297 1297 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1298 1298 * then re-generate a more complete list including status information.
1299 1299 */
1300 1300 static void
1301 1301 spa_load_spares(spa_t *spa)
1302 1302 {
1303 1303 nvlist_t **spares;
1304 1304 uint_t nspares;
1305 1305 int i;
1306 1306 vdev_t *vd, *tvd;
1307 1307
1308 1308 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1309 1309
1310 1310 /*
1311 1311 * First, close and free any existing spare vdevs.
1312 1312 */
1313 1313 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1314 1314 vd = spa->spa_spares.sav_vdevs[i];
1315 1315
1316 1316 /* Undo the call to spa_activate() below */
1317 1317 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1318 1318 B_FALSE)) != NULL && tvd->vdev_isspare)
1319 1319 spa_spare_remove(tvd);
1320 1320 vdev_close(vd);
1321 1321 vdev_free(vd);
1322 1322 }
1323 1323
1324 1324 if (spa->spa_spares.sav_vdevs)
1325 1325 kmem_free(spa->spa_spares.sav_vdevs,
1326 1326 spa->spa_spares.sav_count * sizeof (void *));
1327 1327
1328 1328 if (spa->spa_spares.sav_config == NULL)
1329 1329 nspares = 0;
1330 1330 else
1331 1331 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1332 1332 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1333 1333
1334 1334 spa->spa_spares.sav_count = (int)nspares;
1335 1335 spa->spa_spares.sav_vdevs = NULL;
1336 1336
1337 1337 if (nspares == 0)
1338 1338 return;
1339 1339
1340 1340 /*
1341 1341 * Construct the array of vdevs, opening them to get status in the
1342 1342 * process. For each spare, there is potentially two different vdev_t
1343 1343 * structures associated with it: one in the list of spares (used only
1344 1344 * for basic validation purposes) and one in the active vdev
1345 1345 * configuration (if it's spared in). During this phase we open and
1346 1346 * validate each vdev on the spare list. If the vdev also exists in the
1347 1347 * active configuration, then we also mark this vdev as an active spare.
1348 1348 */
1349 1349 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1350 1350 KM_SLEEP);
1351 1351 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1352 1352 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1353 1353 VDEV_ALLOC_SPARE) == 0);
1354 1354 ASSERT(vd != NULL);
1355 1355
1356 1356 spa->spa_spares.sav_vdevs[i] = vd;
1357 1357
1358 1358 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1359 1359 B_FALSE)) != NULL) {
1360 1360 if (!tvd->vdev_isspare)
1361 1361 spa_spare_add(tvd);
1362 1362
1363 1363 /*
1364 1364 * We only mark the spare active if we were successfully
1365 1365 * able to load the vdev. Otherwise, importing a pool
1366 1366 * with a bad active spare would result in strange
1367 1367 * behavior, because multiple pool would think the spare
1368 1368 * is actively in use.
1369 1369 *
1370 1370 * There is a vulnerability here to an equally bizarre
1371 1371 * circumstance, where a dead active spare is later
1372 1372 * brought back to life (onlined or otherwise). Given
1373 1373 * the rarity of this scenario, and the extra complexity
1374 1374 * it adds, we ignore the possibility.
1375 1375 */
1376 1376 if (!vdev_is_dead(tvd))
1377 1377 spa_spare_activate(tvd);
1378 1378 }
1379 1379
1380 1380 vd->vdev_top = vd;
1381 1381 vd->vdev_aux = &spa->spa_spares;
1382 1382
1383 1383 if (vdev_open(vd) != 0)
1384 1384 continue;
1385 1385
1386 1386 if (vdev_validate_aux(vd) == 0)
1387 1387 spa_spare_add(vd);
1388 1388 }
1389 1389
1390 1390 /*
1391 1391 * Recompute the stashed list of spares, with status information
1392 1392 * this time.
1393 1393 */
1394 1394 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1395 1395 DATA_TYPE_NVLIST_ARRAY) == 0);
1396 1396
1397 1397 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1398 1398 KM_SLEEP);
1399 1399 for (i = 0; i < spa->spa_spares.sav_count; i++)
1400 1400 spares[i] = vdev_config_generate(spa,
1401 1401 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1402 1402 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1403 1403 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1404 1404 for (i = 0; i < spa->spa_spares.sav_count; i++)
1405 1405 nvlist_free(spares[i]);
1406 1406 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1407 1407 }
1408 1408
1409 1409 /*
1410 1410 * Load (or re-load) the current list of vdevs describing the active l2cache for
1411 1411 * this pool. When this is called, we have some form of basic information in
1412 1412 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1413 1413 * then re-generate a more complete list including status information.
1414 1414 * Devices which are already active have their details maintained, and are
1415 1415 * not re-opened.
1416 1416 */
1417 1417 static void
1418 1418 spa_load_l2cache(spa_t *spa)
1419 1419 {
1420 1420 nvlist_t **l2cache;
1421 1421 uint_t nl2cache;
1422 1422 int i, j, oldnvdevs;
1423 1423 uint64_t guid;
1424 1424 vdev_t *vd, **oldvdevs, **newvdevs;
1425 1425 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1426 1426
1427 1427 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1428 1428
1429 1429 if (sav->sav_config != NULL) {
1430 1430 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1431 1431 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1432 1432 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1433 1433 } else {
1434 1434 nl2cache = 0;
1435 1435 newvdevs = NULL;
1436 1436 }
1437 1437
1438 1438 oldvdevs = sav->sav_vdevs;
1439 1439 oldnvdevs = sav->sav_count;
1440 1440 sav->sav_vdevs = NULL;
1441 1441 sav->sav_count = 0;
1442 1442
1443 1443 /*
1444 1444 * Process new nvlist of vdevs.
1445 1445 */
1446 1446 for (i = 0; i < nl2cache; i++) {
1447 1447 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1448 1448 &guid) == 0);
1449 1449
1450 1450 newvdevs[i] = NULL;
1451 1451 for (j = 0; j < oldnvdevs; j++) {
1452 1452 vd = oldvdevs[j];
1453 1453 if (vd != NULL && guid == vd->vdev_guid) {
1454 1454 /*
1455 1455 * Retain previous vdev for add/remove ops.
1456 1456 */
1457 1457 newvdevs[i] = vd;
1458 1458 oldvdevs[j] = NULL;
1459 1459 break;
1460 1460 }
1461 1461 }
1462 1462
1463 1463 if (newvdevs[i] == NULL) {
1464 1464 /*
1465 1465 * Create new vdev
1466 1466 */
1467 1467 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1468 1468 VDEV_ALLOC_L2CACHE) == 0);
1469 1469 ASSERT(vd != NULL);
1470 1470 newvdevs[i] = vd;
1471 1471
1472 1472 /*
1473 1473 * Commit this vdev as an l2cache device,
1474 1474 * even if it fails to open.
1475 1475 */
1476 1476 spa_l2cache_add(vd);
1477 1477
|
↓ open down ↓ |
1477 lines elided |
↑ open up ↑ |
1478 1478 vd->vdev_top = vd;
1479 1479 vd->vdev_aux = sav;
1480 1480
1481 1481 spa_l2cache_activate(vd);
1482 1482
1483 1483 if (vdev_open(vd) != 0)
1484 1484 continue;
1485 1485
1486 1486 (void) vdev_validate_aux(vd);
1487 1487
1488 - if (!vdev_is_dead(vd))
1489 - l2arc_add_vdev(spa, vd);
1488 + if (!vdev_is_dead(vd)) {
1489 + boolean_t do_rebuild = B_FALSE;
1490 +
1491 + (void) nvlist_lookup_boolean_value(l2cache[i],
1492 + ZPOOL_CONFIG_L2CACHE_PERSISTENT,
1493 + &do_rebuild);
1494 + l2arc_add_vdev(spa, vd, do_rebuild);
1495 + }
1490 1496 }
1491 1497 }
1492 1498
1493 1499 /*
1494 1500 * Purge vdevs that were dropped
1495 1501 */
1496 1502 for (i = 0; i < oldnvdevs; i++) {
1497 1503 uint64_t pool;
1498 1504
1499 1505 vd = oldvdevs[i];
1500 1506 if (vd != NULL) {
1501 1507 ASSERT(vd->vdev_isl2cache);
1502 1508
1503 1509 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1504 1510 pool != 0ULL && l2arc_vdev_present(vd))
1505 1511 l2arc_remove_vdev(vd);
1506 1512 vdev_clear_stats(vd);
1507 1513 vdev_free(vd);
1508 1514 }
1509 1515 }
1510 1516
1511 1517 if (oldvdevs)
1512 1518 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1513 1519
1514 1520 if (sav->sav_config == NULL)
1515 1521 goto out;
1516 1522
1517 1523 sav->sav_vdevs = newvdevs;
1518 1524 sav->sav_count = (int)nl2cache;
1519 1525
1520 1526 /*
1521 1527 * Recompute the stashed list of l2cache devices, with status
1522 1528 * information this time.
1523 1529 */
1524 1530 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1525 1531 DATA_TYPE_NVLIST_ARRAY) == 0);
1526 1532
1527 1533 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1528 1534 for (i = 0; i < sav->sav_count; i++)
1529 1535 l2cache[i] = vdev_config_generate(spa,
1530 1536 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1531 1537 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1532 1538 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1533 1539 out:
1534 1540 for (i = 0; i < sav->sav_count; i++)
1535 1541 nvlist_free(l2cache[i]);
1536 1542 if (sav->sav_count)
1537 1543 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1538 1544 }
1539 1545
1540 1546 static int
1541 1547 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1542 1548 {
1543 1549 dmu_buf_t *db;
1544 1550 char *packed = NULL;
1545 1551 size_t nvsize = 0;
1546 1552 int error;
1547 1553 *value = NULL;
1548 1554
1549 1555 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1550 1556 nvsize = *(uint64_t *)db->db_data;
1551 1557 dmu_buf_rele(db, FTAG);
1552 1558
1553 1559 packed = kmem_alloc(nvsize, KM_SLEEP);
1554 1560 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1555 1561 DMU_READ_PREFETCH);
1556 1562 if (error == 0)
1557 1563 error = nvlist_unpack(packed, nvsize, value, 0);
1558 1564 kmem_free(packed, nvsize);
1559 1565
1560 1566 return (error);
1561 1567 }
1562 1568
1563 1569 /*
1564 1570 * Checks to see if the given vdev could not be opened, in which case we post a
1565 1571 * sysevent to notify the autoreplace code that the device has been removed.
1566 1572 */
1567 1573 static void
1568 1574 spa_check_removed(vdev_t *vd)
1569 1575 {
1570 1576 for (int c = 0; c < vd->vdev_children; c++)
1571 1577 spa_check_removed(vd->vdev_child[c]);
1572 1578
1573 1579 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1574 1580 !vd->vdev_ishole) {
1575 1581 zfs_post_autoreplace(vd->vdev_spa, vd);
1576 1582 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1577 1583 }
1578 1584 }
1579 1585
1580 1586 /*
1581 1587 * Validate the current config against the MOS config
1582 1588 */
1583 1589 static boolean_t
1584 1590 spa_config_valid(spa_t *spa, nvlist_t *config)
1585 1591 {
1586 1592 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1587 1593 nvlist_t *nv;
1588 1594
1589 1595 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1590 1596
1591 1597 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1592 1598 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1593 1599
1594 1600 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1595 1601
1596 1602 /*
1597 1603 * If we're doing a normal import, then build up any additional
1598 1604 * diagnostic information about missing devices in this config.
1599 1605 * We'll pass this up to the user for further processing.
1600 1606 */
1601 1607 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1602 1608 nvlist_t **child, *nv;
1603 1609 uint64_t idx = 0;
1604 1610
1605 1611 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1606 1612 KM_SLEEP);
1607 1613 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1608 1614
1609 1615 for (int c = 0; c < rvd->vdev_children; c++) {
1610 1616 vdev_t *tvd = rvd->vdev_child[c];
1611 1617 vdev_t *mtvd = mrvd->vdev_child[c];
1612 1618
1613 1619 if (tvd->vdev_ops == &vdev_missing_ops &&
1614 1620 mtvd->vdev_ops != &vdev_missing_ops &&
1615 1621 mtvd->vdev_islog)
1616 1622 child[idx++] = vdev_config_generate(spa, mtvd,
1617 1623 B_FALSE, 0);
1618 1624 }
1619 1625
1620 1626 if (idx) {
1621 1627 VERIFY(nvlist_add_nvlist_array(nv,
1622 1628 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1623 1629 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1624 1630 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1625 1631
1626 1632 for (int i = 0; i < idx; i++)
1627 1633 nvlist_free(child[i]);
1628 1634 }
1629 1635 nvlist_free(nv);
1630 1636 kmem_free(child, rvd->vdev_children * sizeof (char **));
1631 1637 }
1632 1638
1633 1639 /*
1634 1640 * Compare the root vdev tree with the information we have
1635 1641 * from the MOS config (mrvd). Check each top-level vdev
1636 1642 * with the corresponding MOS config top-level (mtvd).
1637 1643 */
1638 1644 for (int c = 0; c < rvd->vdev_children; c++) {
1639 1645 vdev_t *tvd = rvd->vdev_child[c];
1640 1646 vdev_t *mtvd = mrvd->vdev_child[c];
1641 1647
1642 1648 /*
1643 1649 * Resolve any "missing" vdevs in the current configuration.
1644 1650 * If we find that the MOS config has more accurate information
1645 1651 * about the top-level vdev then use that vdev instead.
1646 1652 */
1647 1653 if (tvd->vdev_ops == &vdev_missing_ops &&
1648 1654 mtvd->vdev_ops != &vdev_missing_ops) {
1649 1655
1650 1656 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1651 1657 continue;
1652 1658
1653 1659 /*
1654 1660 * Device specific actions.
1655 1661 */
1656 1662 if (mtvd->vdev_islog) {
1657 1663 spa_set_log_state(spa, SPA_LOG_CLEAR);
1658 1664 } else {
1659 1665 /*
1660 1666 * XXX - once we have 'readonly' pool
1661 1667 * support we should be able to handle
1662 1668 * missing data devices by transitioning
1663 1669 * the pool to readonly.
1664 1670 */
1665 1671 continue;
1666 1672 }
1667 1673
1668 1674 /*
1669 1675 * Swap the missing vdev with the data we were
1670 1676 * able to obtain from the MOS config.
1671 1677 */
1672 1678 vdev_remove_child(rvd, tvd);
1673 1679 vdev_remove_child(mrvd, mtvd);
1674 1680
1675 1681 vdev_add_child(rvd, mtvd);
1676 1682 vdev_add_child(mrvd, tvd);
1677 1683
1678 1684 spa_config_exit(spa, SCL_ALL, FTAG);
1679 1685 vdev_load(mtvd);
1680 1686 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1681 1687
1682 1688 vdev_reopen(rvd);
1683 1689 } else if (mtvd->vdev_islog) {
1684 1690 /*
1685 1691 * Load the slog device's state from the MOS config
1686 1692 * since it's possible that the label does not
1687 1693 * contain the most up-to-date information.
1688 1694 */
1689 1695 vdev_load_log_state(tvd, mtvd);
1690 1696 vdev_reopen(tvd);
1691 1697 }
1692 1698 }
1693 1699 vdev_free(mrvd);
1694 1700 spa_config_exit(spa, SCL_ALL, FTAG);
1695 1701
1696 1702 /*
1697 1703 * Ensure we were able to validate the config.
1698 1704 */
1699 1705 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1700 1706 }
1701 1707
1702 1708 /*
1703 1709 * Check for missing log devices
1704 1710 */
1705 1711 static boolean_t
1706 1712 spa_check_logs(spa_t *spa)
1707 1713 {
1708 1714 boolean_t rv = B_FALSE;
1709 1715
1710 1716 switch (spa->spa_log_state) {
1711 1717 case SPA_LOG_MISSING:
1712 1718 /* need to recheck in case slog has been restored */
1713 1719 case SPA_LOG_UNKNOWN:
1714 1720 rv = (dmu_objset_find(spa->spa_name, zil_check_log_chain,
1715 1721 NULL, DS_FIND_CHILDREN) != 0);
1716 1722 if (rv)
1717 1723 spa_set_log_state(spa, SPA_LOG_MISSING);
1718 1724 break;
1719 1725 }
1720 1726 return (rv);
1721 1727 }
1722 1728
1723 1729 static boolean_t
1724 1730 spa_passivate_log(spa_t *spa)
1725 1731 {
1726 1732 vdev_t *rvd = spa->spa_root_vdev;
1727 1733 boolean_t slog_found = B_FALSE;
1728 1734
1729 1735 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1730 1736
1731 1737 if (!spa_has_slogs(spa))
1732 1738 return (B_FALSE);
1733 1739
1734 1740 for (int c = 0; c < rvd->vdev_children; c++) {
1735 1741 vdev_t *tvd = rvd->vdev_child[c];
1736 1742 metaslab_group_t *mg = tvd->vdev_mg;
1737 1743
1738 1744 if (tvd->vdev_islog) {
1739 1745 metaslab_group_passivate(mg);
1740 1746 slog_found = B_TRUE;
1741 1747 }
1742 1748 }
1743 1749
1744 1750 return (slog_found);
1745 1751 }
1746 1752
1747 1753 static void
1748 1754 spa_activate_log(spa_t *spa)
1749 1755 {
1750 1756 vdev_t *rvd = spa->spa_root_vdev;
1751 1757
1752 1758 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1753 1759
1754 1760 for (int c = 0; c < rvd->vdev_children; c++) {
1755 1761 vdev_t *tvd = rvd->vdev_child[c];
1756 1762 metaslab_group_t *mg = tvd->vdev_mg;
1757 1763
1758 1764 if (tvd->vdev_islog)
1759 1765 metaslab_group_activate(mg);
1760 1766 }
1761 1767 }
1762 1768
1763 1769 int
1764 1770 spa_offline_log(spa_t *spa)
1765 1771 {
1766 1772 int error;
1767 1773
1768 1774 error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1769 1775 NULL, DS_FIND_CHILDREN);
1770 1776 if (error == 0) {
1771 1777 /*
1772 1778 * We successfully offlined the log device, sync out the
1773 1779 * current txg so that the "stubby" block can be removed
1774 1780 * by zil_sync().
1775 1781 */
1776 1782 txg_wait_synced(spa->spa_dsl_pool, 0);
1777 1783 }
1778 1784 return (error);
1779 1785 }
1780 1786
1781 1787 static void
1782 1788 spa_aux_check_removed(spa_aux_vdev_t *sav)
1783 1789 {
1784 1790 for (int i = 0; i < sav->sav_count; i++)
1785 1791 spa_check_removed(sav->sav_vdevs[i]);
1786 1792 }
1787 1793
1788 1794 void
1789 1795 spa_claim_notify(zio_t *zio)
1790 1796 {
1791 1797 spa_t *spa = zio->io_spa;
1792 1798
1793 1799 if (zio->io_error)
1794 1800 return;
1795 1801
1796 1802 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1797 1803 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1798 1804 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1799 1805 mutex_exit(&spa->spa_props_lock);
1800 1806 }
1801 1807
1802 1808 typedef struct spa_load_error {
1803 1809 uint64_t sle_meta_count;
1804 1810 uint64_t sle_data_count;
1805 1811 } spa_load_error_t;
1806 1812
1807 1813 static void
1808 1814 spa_load_verify_done(zio_t *zio)
1809 1815 {
1810 1816 blkptr_t *bp = zio->io_bp;
1811 1817 spa_load_error_t *sle = zio->io_private;
1812 1818 dmu_object_type_t type = BP_GET_TYPE(bp);
1813 1819 int error = zio->io_error;
1814 1820
1815 1821 if (error) {
1816 1822 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1817 1823 type != DMU_OT_INTENT_LOG)
1818 1824 atomic_add_64(&sle->sle_meta_count, 1);
1819 1825 else
1820 1826 atomic_add_64(&sle->sle_data_count, 1);
1821 1827 }
1822 1828 zio_data_buf_free(zio->io_data, zio->io_size);
1823 1829 }
1824 1830
1825 1831 /*ARGSUSED*/
1826 1832 static int
1827 1833 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1828 1834 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1829 1835 {
1830 1836 if (bp != NULL) {
1831 1837 zio_t *rio = arg;
1832 1838 size_t size = BP_GET_PSIZE(bp);
1833 1839 void *data = zio_data_buf_alloc(size);
1834 1840
1835 1841 zio_nowait(zio_read(rio, spa, bp, data, size,
1836 1842 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1837 1843 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1838 1844 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1839 1845 }
1840 1846 return (0);
1841 1847 }
1842 1848
1843 1849 static int
1844 1850 spa_load_verify(spa_t *spa)
1845 1851 {
1846 1852 zio_t *rio;
1847 1853 spa_load_error_t sle = { 0 };
1848 1854 zpool_rewind_policy_t policy;
1849 1855 boolean_t verify_ok = B_FALSE;
1850 1856 int error;
1851 1857
1852 1858 zpool_get_rewind_policy(spa->spa_config, &policy);
1853 1859
1854 1860 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1855 1861 return (0);
1856 1862
1857 1863 rio = zio_root(spa, NULL, &sle,
1858 1864 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1859 1865
1860 1866 error = traverse_pool(spa, spa->spa_verify_min_txg,
1861 1867 TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1862 1868
1863 1869 (void) zio_wait(rio);
1864 1870
1865 1871 spa->spa_load_meta_errors = sle.sle_meta_count;
1866 1872 spa->spa_load_data_errors = sle.sle_data_count;
1867 1873
1868 1874 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1869 1875 sle.sle_data_count <= policy.zrp_maxdata) {
1870 1876 int64_t loss = 0;
1871 1877
1872 1878 verify_ok = B_TRUE;
1873 1879 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1874 1880 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1875 1881
1876 1882 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1877 1883 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1878 1884 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1879 1885 VERIFY(nvlist_add_int64(spa->spa_load_info,
1880 1886 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1881 1887 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1882 1888 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1883 1889 } else {
1884 1890 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1885 1891 }
1886 1892
1887 1893 if (error) {
1888 1894 if (error != ENXIO && error != EIO)
1889 1895 error = SET_ERROR(EIO);
1890 1896 return (error);
1891 1897 }
1892 1898
1893 1899 return (verify_ok ? 0 : EIO);
1894 1900 }
1895 1901
1896 1902 /*
1897 1903 * Find a value in the pool props object.
1898 1904 */
1899 1905 static void
1900 1906 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1901 1907 {
1902 1908 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1903 1909 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1904 1910 }
1905 1911
1906 1912 /*
1907 1913 * Find a value in the pool directory object.
1908 1914 */
1909 1915 static int
1910 1916 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1911 1917 {
1912 1918 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1913 1919 name, sizeof (uint64_t), 1, val));
1914 1920 }
1915 1921
1916 1922 static int
1917 1923 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1918 1924 {
1919 1925 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1920 1926 return (err);
1921 1927 }
1922 1928
1923 1929 /*
1924 1930 * Fix up config after a partly-completed split. This is done with the
1925 1931 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
1926 1932 * pool have that entry in their config, but only the splitting one contains
1927 1933 * a list of all the guids of the vdevs that are being split off.
1928 1934 *
1929 1935 * This function determines what to do with that list: either rejoin
1930 1936 * all the disks to the pool, or complete the splitting process. To attempt
1931 1937 * the rejoin, each disk that is offlined is marked online again, and
1932 1938 * we do a reopen() call. If the vdev label for every disk that was
1933 1939 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1934 1940 * then we call vdev_split() on each disk, and complete the split.
1935 1941 *
1936 1942 * Otherwise we leave the config alone, with all the vdevs in place in
1937 1943 * the original pool.
1938 1944 */
1939 1945 static void
1940 1946 spa_try_repair(spa_t *spa, nvlist_t *config)
1941 1947 {
1942 1948 uint_t extracted;
1943 1949 uint64_t *glist;
1944 1950 uint_t i, gcount;
1945 1951 nvlist_t *nvl;
1946 1952 vdev_t **vd;
1947 1953 boolean_t attempt_reopen;
1948 1954
1949 1955 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1950 1956 return;
1951 1957
1952 1958 /* check that the config is complete */
1953 1959 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1954 1960 &glist, &gcount) != 0)
1955 1961 return;
1956 1962
1957 1963 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1958 1964
1959 1965 /* attempt to online all the vdevs & validate */
1960 1966 attempt_reopen = B_TRUE;
1961 1967 for (i = 0; i < gcount; i++) {
1962 1968 if (glist[i] == 0) /* vdev is hole */
1963 1969 continue;
1964 1970
1965 1971 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1966 1972 if (vd[i] == NULL) {
1967 1973 /*
1968 1974 * Don't bother attempting to reopen the disks;
1969 1975 * just do the split.
1970 1976 */
1971 1977 attempt_reopen = B_FALSE;
1972 1978 } else {
1973 1979 /* attempt to re-online it */
1974 1980 vd[i]->vdev_offline = B_FALSE;
1975 1981 }
1976 1982 }
1977 1983
1978 1984 if (attempt_reopen) {
1979 1985 vdev_reopen(spa->spa_root_vdev);
1980 1986
1981 1987 /* check each device to see what state it's in */
1982 1988 for (extracted = 0, i = 0; i < gcount; i++) {
1983 1989 if (vd[i] != NULL &&
1984 1990 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1985 1991 break;
1986 1992 ++extracted;
1987 1993 }
1988 1994 }
1989 1995
1990 1996 /*
1991 1997 * If every disk has been moved to the new pool, or if we never
1992 1998 * even attempted to look at them, then we split them off for
1993 1999 * good.
1994 2000 */
1995 2001 if (!attempt_reopen || gcount == extracted) {
1996 2002 for (i = 0; i < gcount; i++)
1997 2003 if (vd[i] != NULL)
1998 2004 vdev_split(vd[i]);
1999 2005 vdev_reopen(spa->spa_root_vdev);
2000 2006 }
2001 2007
2002 2008 kmem_free(vd, gcount * sizeof (vdev_t *));
2003 2009 }
2004 2010
2005 2011 static int
2006 2012 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
2007 2013 boolean_t mosconfig)
2008 2014 {
2009 2015 nvlist_t *config = spa->spa_config;
2010 2016 char *ereport = FM_EREPORT_ZFS_POOL;
2011 2017 char *comment;
2012 2018 int error;
2013 2019 uint64_t pool_guid;
2014 2020 nvlist_t *nvl;
2015 2021
2016 2022 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
2017 2023 return (SET_ERROR(EINVAL));
2018 2024
2019 2025 ASSERT(spa->spa_comment == NULL);
2020 2026 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2021 2027 spa->spa_comment = spa_strdup(comment);
2022 2028
2023 2029 /*
2024 2030 * Versioning wasn't explicitly added to the label until later, so if
2025 2031 * it's not present treat it as the initial version.
2026 2032 */
2027 2033 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2028 2034 &spa->spa_ubsync.ub_version) != 0)
2029 2035 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2030 2036
2031 2037 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2032 2038 &spa->spa_config_txg);
2033 2039
2034 2040 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
2035 2041 spa_guid_exists(pool_guid, 0)) {
2036 2042 error = SET_ERROR(EEXIST);
2037 2043 } else {
2038 2044 spa->spa_config_guid = pool_guid;
2039 2045
2040 2046 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
2041 2047 &nvl) == 0) {
2042 2048 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
2043 2049 KM_SLEEP) == 0);
2044 2050 }
2045 2051
2046 2052 nvlist_free(spa->spa_load_info);
2047 2053 spa->spa_load_info = fnvlist_alloc();
2048 2054
2049 2055 gethrestime(&spa->spa_loaded_ts);
2050 2056 error = spa_load_impl(spa, pool_guid, config, state, type,
2051 2057 mosconfig, &ereport);
2052 2058 }
2053 2059
2054 2060 spa->spa_minref = refcount_count(&spa->spa_refcount);
2055 2061 if (error) {
2056 2062 if (error != EEXIST) {
2057 2063 spa->spa_loaded_ts.tv_sec = 0;
2058 2064 spa->spa_loaded_ts.tv_nsec = 0;
2059 2065 }
2060 2066 if (error != EBADF) {
2061 2067 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2062 2068 }
2063 2069 }
2064 2070 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2065 2071 spa->spa_ena = 0;
2066 2072
2067 2073 return (error);
2068 2074 }
2069 2075
2070 2076 /*
2071 2077 * Load an existing storage pool, using the pool's builtin spa_config as a
2072 2078 * source of configuration information.
2073 2079 */
2074 2080 static int
2075 2081 spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2076 2082 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2077 2083 char **ereport)
2078 2084 {
2079 2085 int error = 0;
2080 2086 nvlist_t *nvroot = NULL;
2081 2087 nvlist_t *label;
2082 2088 vdev_t *rvd;
2083 2089 uberblock_t *ub = &spa->spa_uberblock;
2084 2090 uint64_t children, config_cache_txg = spa->spa_config_txg;
2085 2091 int orig_mode = spa->spa_mode;
2086 2092 int parse;
2087 2093 uint64_t obj;
2088 2094 boolean_t missing_feat_write = B_FALSE;
2089 2095
2090 2096 /*
2091 2097 * If this is an untrusted config, access the pool in read-only mode.
2092 2098 * This prevents things like resilvering recently removed devices.
2093 2099 */
2094 2100 if (!mosconfig)
2095 2101 spa->spa_mode = FREAD;
2096 2102
2097 2103 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2098 2104
2099 2105 spa->spa_load_state = state;
2100 2106
2101 2107 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2102 2108 return (SET_ERROR(EINVAL));
2103 2109
2104 2110 parse = (type == SPA_IMPORT_EXISTING ?
2105 2111 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2106 2112
2107 2113 /*
2108 2114 * Create "The Godfather" zio to hold all async IOs
2109 2115 */
2110 2116 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2111 2117 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2112 2118
2113 2119 /*
2114 2120 * Parse the configuration into a vdev tree. We explicitly set the
2115 2121 * value that will be returned by spa_version() since parsing the
2116 2122 * configuration requires knowing the version number.
2117 2123 */
2118 2124 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2119 2125 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2120 2126 spa_config_exit(spa, SCL_ALL, FTAG);
2121 2127
2122 2128 if (error != 0)
2123 2129 return (error);
2124 2130
2125 2131 ASSERT(spa->spa_root_vdev == rvd);
2126 2132
2127 2133 if (type != SPA_IMPORT_ASSEMBLE) {
2128 2134 ASSERT(spa_guid(spa) == pool_guid);
2129 2135 }
2130 2136
2131 2137 /*
2132 2138 * Try to open all vdevs, loading each label in the process.
2133 2139 */
2134 2140 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2135 2141 error = vdev_open(rvd);
2136 2142 spa_config_exit(spa, SCL_ALL, FTAG);
2137 2143 if (error != 0)
2138 2144 return (error);
2139 2145
2140 2146 /*
2141 2147 * We need to validate the vdev labels against the configuration that
2142 2148 * we have in hand, which is dependent on the setting of mosconfig. If
2143 2149 * mosconfig is true then we're validating the vdev labels based on
2144 2150 * that config. Otherwise, we're validating against the cached config
2145 2151 * (zpool.cache) that was read when we loaded the zfs module, and then
2146 2152 * later we will recursively call spa_load() and validate against
2147 2153 * the vdev config.
2148 2154 *
2149 2155 * If we're assembling a new pool that's been split off from an
2150 2156 * existing pool, the labels haven't yet been updated so we skip
2151 2157 * validation for now.
2152 2158 */
2153 2159 if (type != SPA_IMPORT_ASSEMBLE) {
2154 2160 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2155 2161 error = vdev_validate(rvd, mosconfig);
2156 2162 spa_config_exit(spa, SCL_ALL, FTAG);
2157 2163
2158 2164 if (error != 0)
2159 2165 return (error);
2160 2166
2161 2167 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2162 2168 return (SET_ERROR(ENXIO));
2163 2169 }
2164 2170
2165 2171 /*
2166 2172 * Find the best uberblock.
2167 2173 */
2168 2174 vdev_uberblock_load(rvd, ub, &label);
2169 2175
2170 2176 /*
2171 2177 * If we weren't able to find a single valid uberblock, return failure.
2172 2178 */
2173 2179 if (ub->ub_txg == 0) {
2174 2180 nvlist_free(label);
2175 2181 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2176 2182 }
2177 2183
2178 2184 /*
2179 2185 * If the pool has an unsupported version we can't open it.
2180 2186 */
2181 2187 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2182 2188 nvlist_free(label);
2183 2189 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2184 2190 }
2185 2191
2186 2192 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2187 2193 nvlist_t *features;
2188 2194
2189 2195 /*
2190 2196 * If we weren't able to find what's necessary for reading the
2191 2197 * MOS in the label, return failure.
2192 2198 */
2193 2199 if (label == NULL || nvlist_lookup_nvlist(label,
2194 2200 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2195 2201 nvlist_free(label);
2196 2202 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2197 2203 ENXIO));
2198 2204 }
2199 2205
2200 2206 /*
2201 2207 * Update our in-core representation with the definitive values
2202 2208 * from the label.
2203 2209 */
2204 2210 nvlist_free(spa->spa_label_features);
2205 2211 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2206 2212 }
2207 2213
2208 2214 nvlist_free(label);
2209 2215
2210 2216 /*
2211 2217 * Look through entries in the label nvlist's features_for_read. If
2212 2218 * there is a feature listed there which we don't understand then we
2213 2219 * cannot open a pool.
2214 2220 */
2215 2221 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2216 2222 nvlist_t *unsup_feat;
2217 2223
2218 2224 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2219 2225 0);
2220 2226
2221 2227 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2222 2228 NULL); nvp != NULL;
2223 2229 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2224 2230 if (!zfeature_is_supported(nvpair_name(nvp))) {
2225 2231 VERIFY(nvlist_add_string(unsup_feat,
2226 2232 nvpair_name(nvp), "") == 0);
2227 2233 }
2228 2234 }
2229 2235
2230 2236 if (!nvlist_empty(unsup_feat)) {
2231 2237 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2232 2238 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2233 2239 nvlist_free(unsup_feat);
2234 2240 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2235 2241 ENOTSUP));
2236 2242 }
2237 2243
2238 2244 nvlist_free(unsup_feat);
2239 2245 }
2240 2246
2241 2247 /*
2242 2248 * If the vdev guid sum doesn't match the uberblock, we have an
2243 2249 * incomplete configuration. We first check to see if the pool
2244 2250 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2245 2251 * If it is, defer the vdev_guid_sum check till later so we
2246 2252 * can handle missing vdevs.
2247 2253 */
2248 2254 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2249 2255 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2250 2256 rvd->vdev_guid_sum != ub->ub_guid_sum)
2251 2257 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2252 2258
2253 2259 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2254 2260 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2255 2261 spa_try_repair(spa, config);
2256 2262 spa_config_exit(spa, SCL_ALL, FTAG);
2257 2263 nvlist_free(spa->spa_config_splitting);
2258 2264 spa->spa_config_splitting = NULL;
2259 2265 }
2260 2266
2261 2267 /*
2262 2268 * Initialize internal SPA structures.
2263 2269 */
2264 2270 spa->spa_state = POOL_STATE_ACTIVE;
2265 2271 spa->spa_ubsync = spa->spa_uberblock;
2266 2272 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2267 2273 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2268 2274 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2269 2275 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2270 2276 spa->spa_claim_max_txg = spa->spa_first_txg;
2271 2277 spa->spa_prev_software_version = ub->ub_software_version;
2272 2278
2273 2279 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2274 2280 if (error)
2275 2281 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2276 2282 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2277 2283
2278 2284 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2279 2285 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2280 2286
2281 2287 if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2282 2288 boolean_t missing_feat_read = B_FALSE;
2283 2289 nvlist_t *unsup_feat, *enabled_feat;
2284 2290
2285 2291 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2286 2292 &spa->spa_feat_for_read_obj) != 0) {
2287 2293 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2288 2294 }
2289 2295
2290 2296 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2291 2297 &spa->spa_feat_for_write_obj) != 0) {
2292 2298 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2293 2299 }
2294 2300
2295 2301 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2296 2302 &spa->spa_feat_desc_obj) != 0) {
2297 2303 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2298 2304 }
2299 2305
2300 2306 enabled_feat = fnvlist_alloc();
2301 2307 unsup_feat = fnvlist_alloc();
2302 2308
2303 2309 if (!spa_features_check(spa, B_FALSE,
2304 2310 unsup_feat, enabled_feat))
2305 2311 missing_feat_read = B_TRUE;
2306 2312
2307 2313 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2308 2314 if (!spa_features_check(spa, B_TRUE,
2309 2315 unsup_feat, enabled_feat)) {
2310 2316 missing_feat_write = B_TRUE;
2311 2317 }
2312 2318 }
2313 2319
2314 2320 fnvlist_add_nvlist(spa->spa_load_info,
2315 2321 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2316 2322
2317 2323 if (!nvlist_empty(unsup_feat)) {
2318 2324 fnvlist_add_nvlist(spa->spa_load_info,
2319 2325 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2320 2326 }
2321 2327
2322 2328 fnvlist_free(enabled_feat);
2323 2329 fnvlist_free(unsup_feat);
2324 2330
2325 2331 if (!missing_feat_read) {
2326 2332 fnvlist_add_boolean(spa->spa_load_info,
2327 2333 ZPOOL_CONFIG_CAN_RDONLY);
2328 2334 }
2329 2335
2330 2336 /*
2331 2337 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2332 2338 * twofold: to determine whether the pool is available for
2333 2339 * import in read-write mode and (if it is not) whether the
2334 2340 * pool is available for import in read-only mode. If the pool
2335 2341 * is available for import in read-write mode, it is displayed
2336 2342 * as available in userland; if it is not available for import
2337 2343 * in read-only mode, it is displayed as unavailable in
2338 2344 * userland. If the pool is available for import in read-only
2339 2345 * mode but not read-write mode, it is displayed as unavailable
2340 2346 * in userland with a special note that the pool is actually
2341 2347 * available for open in read-only mode.
2342 2348 *
2343 2349 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2344 2350 * missing a feature for write, we must first determine whether
2345 2351 * the pool can be opened read-only before returning to
2346 2352 * userland in order to know whether to display the
2347 2353 * abovementioned note.
2348 2354 */
2349 2355 if (missing_feat_read || (missing_feat_write &&
2350 2356 spa_writeable(spa))) {
2351 2357 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2352 2358 ENOTSUP));
2353 2359 }
2354 2360 }
2355 2361
2356 2362 spa->spa_is_initializing = B_TRUE;
2357 2363 error = dsl_pool_open(spa->spa_dsl_pool);
2358 2364 spa->spa_is_initializing = B_FALSE;
2359 2365 if (error != 0)
2360 2366 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2361 2367
2362 2368 if (!mosconfig) {
2363 2369 uint64_t hostid;
2364 2370 nvlist_t *policy = NULL, *nvconfig;
2365 2371
2366 2372 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2367 2373 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2368 2374
2369 2375 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2370 2376 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2371 2377 char *hostname;
2372 2378 unsigned long myhostid = 0;
2373 2379
2374 2380 VERIFY(nvlist_lookup_string(nvconfig,
2375 2381 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2376 2382
2377 2383 #ifdef _KERNEL
2378 2384 myhostid = zone_get_hostid(NULL);
2379 2385 #else /* _KERNEL */
2380 2386 /*
2381 2387 * We're emulating the system's hostid in userland, so
2382 2388 * we can't use zone_get_hostid().
2383 2389 */
2384 2390 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2385 2391 #endif /* _KERNEL */
2386 2392 if (hostid != 0 && myhostid != 0 &&
2387 2393 hostid != myhostid) {
2388 2394 nvlist_free(nvconfig);
2389 2395 cmn_err(CE_WARN, "pool '%s' could not be "
2390 2396 "loaded as it was last accessed by "
2391 2397 "another system (host: %s hostid: 0x%lx). "
2392 2398 "See: http://illumos.org/msg/ZFS-8000-EY",
2393 2399 spa_name(spa), hostname,
2394 2400 (unsigned long)hostid);
2395 2401 return (SET_ERROR(EBADF));
2396 2402 }
2397 2403 }
2398 2404 if (nvlist_lookup_nvlist(spa->spa_config,
2399 2405 ZPOOL_REWIND_POLICY, &policy) == 0)
2400 2406 VERIFY(nvlist_add_nvlist(nvconfig,
2401 2407 ZPOOL_REWIND_POLICY, policy) == 0);
2402 2408
2403 2409 spa_config_set(spa, nvconfig);
2404 2410 spa_unload(spa);
2405 2411 spa_deactivate(spa);
2406 2412 spa_activate(spa, orig_mode);
2407 2413
2408 2414 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2409 2415 }
2410 2416
2411 2417 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2412 2418 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2413 2419 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2414 2420 if (error != 0)
2415 2421 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2416 2422
2417 2423 /*
2418 2424 * Load the bit that tells us to use the new accounting function
2419 2425 * (raid-z deflation). If we have an older pool, this will not
2420 2426 * be present.
2421 2427 */
2422 2428 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2423 2429 if (error != 0 && error != ENOENT)
2424 2430 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2425 2431
2426 2432 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2427 2433 &spa->spa_creation_version);
2428 2434 if (error != 0 && error != ENOENT)
2429 2435 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2430 2436
2431 2437 /*
2432 2438 * Load the persistent error log. If we have an older pool, this will
2433 2439 * not be present.
2434 2440 */
2435 2441 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2436 2442 if (error != 0 && error != ENOENT)
2437 2443 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2438 2444
2439 2445 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2440 2446 &spa->spa_errlog_scrub);
2441 2447 if (error != 0 && error != ENOENT)
2442 2448 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2443 2449
2444 2450 /*
2445 2451 * Load the history object. If we have an older pool, this
2446 2452 * will not be present.
2447 2453 */
2448 2454 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2449 2455 if (error != 0 && error != ENOENT)
2450 2456 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2451 2457
2452 2458 /*
2453 2459 * If we're assembling the pool from the split-off vdevs of
2454 2460 * an existing pool, we don't want to attach the spares & cache
2455 2461 * devices.
2456 2462 */
2457 2463
2458 2464 /*
2459 2465 * Load any hot spares for this pool.
2460 2466 */
2461 2467 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2462 2468 if (error != 0 && error != ENOENT)
2463 2469 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2464 2470 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2465 2471 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2466 2472 if (load_nvlist(spa, spa->spa_spares.sav_object,
2467 2473 &spa->spa_spares.sav_config) != 0)
2468 2474 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2469 2475
2470 2476 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2471 2477 spa_load_spares(spa);
2472 2478 spa_config_exit(spa, SCL_ALL, FTAG);
2473 2479 } else if (error == 0) {
2474 2480 spa->spa_spares.sav_sync = B_TRUE;
2475 2481 }
2476 2482
2477 2483 /*
2478 2484 * Load any level 2 ARC devices for this pool.
2479 2485 */
2480 2486 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2481 2487 &spa->spa_l2cache.sav_object);
2482 2488 if (error != 0 && error != ENOENT)
2483 2489 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2484 2490 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2485 2491 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2486 2492 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2487 2493 &spa->spa_l2cache.sav_config) != 0)
2488 2494 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2489 2495
2490 2496 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2491 2497 spa_load_l2cache(spa);
2492 2498 spa_config_exit(spa, SCL_ALL, FTAG);
2493 2499 } else if (error == 0) {
2494 2500 spa->spa_l2cache.sav_sync = B_TRUE;
2495 2501 }
2496 2502
2497 2503 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2498 2504
2499 2505 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2500 2506 if (error && error != ENOENT)
2501 2507 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2502 2508
2503 2509 if (error == 0) {
2504 2510 uint64_t autoreplace;
2505 2511
2506 2512 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2507 2513 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2508 2514 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2509 2515 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2510 2516 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2511 2517 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2512 2518 &spa->spa_dedup_ditto);
2513 2519
2514 2520 spa->spa_autoreplace = (autoreplace != 0);
2515 2521 }
2516 2522
2517 2523 /*
2518 2524 * If the 'autoreplace' property is set, then post a resource notifying
2519 2525 * the ZFS DE that it should not issue any faults for unopenable
2520 2526 * devices. We also iterate over the vdevs, and post a sysevent for any
2521 2527 * unopenable vdevs so that the normal autoreplace handler can take
2522 2528 * over.
2523 2529 */
2524 2530 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2525 2531 spa_check_removed(spa->spa_root_vdev);
2526 2532 /*
2527 2533 * For the import case, this is done in spa_import(), because
2528 2534 * at this point we're using the spare definitions from
2529 2535 * the MOS config, not necessarily from the userland config.
2530 2536 */
2531 2537 if (state != SPA_LOAD_IMPORT) {
2532 2538 spa_aux_check_removed(&spa->spa_spares);
2533 2539 spa_aux_check_removed(&spa->spa_l2cache);
2534 2540 }
2535 2541 }
2536 2542
2537 2543 /*
2538 2544 * Load the vdev state for all toplevel vdevs.
2539 2545 */
2540 2546 vdev_load(rvd);
2541 2547
2542 2548 /*
2543 2549 * Propagate the leaf DTLs we just loaded all the way up the tree.
2544 2550 */
2545 2551 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2546 2552 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2547 2553 spa_config_exit(spa, SCL_ALL, FTAG);
2548 2554
2549 2555 /*
2550 2556 * Load the DDTs (dedup tables).
2551 2557 */
2552 2558 error = ddt_load(spa);
2553 2559 if (error != 0)
2554 2560 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2555 2561
2556 2562 spa_update_dspace(spa);
2557 2563
2558 2564 /*
2559 2565 * Validate the config, using the MOS config to fill in any
2560 2566 * information which might be missing. If we fail to validate
2561 2567 * the config then declare the pool unfit for use. If we're
2562 2568 * assembling a pool from a split, the log is not transferred
2563 2569 * over.
2564 2570 */
2565 2571 if (type != SPA_IMPORT_ASSEMBLE) {
2566 2572 nvlist_t *nvconfig;
2567 2573
2568 2574 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2569 2575 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2570 2576
2571 2577 if (!spa_config_valid(spa, nvconfig)) {
2572 2578 nvlist_free(nvconfig);
2573 2579 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2574 2580 ENXIO));
2575 2581 }
2576 2582 nvlist_free(nvconfig);
2577 2583
2578 2584 /*
2579 2585 * Now that we've validated the config, check the state of the
2580 2586 * root vdev. If it can't be opened, it indicates one or
2581 2587 * more toplevel vdevs are faulted.
2582 2588 */
2583 2589 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2584 2590 return (SET_ERROR(ENXIO));
2585 2591
2586 2592 if (spa_check_logs(spa)) {
2587 2593 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2588 2594 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2589 2595 }
2590 2596 }
2591 2597
2592 2598 if (missing_feat_write) {
2593 2599 ASSERT(state == SPA_LOAD_TRYIMPORT);
2594 2600
2595 2601 /*
2596 2602 * At this point, we know that we can open the pool in
2597 2603 * read-only mode but not read-write mode. We now have enough
2598 2604 * information and can return to userland.
2599 2605 */
2600 2606 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2601 2607 }
2602 2608
2603 2609 /*
2604 2610 * We've successfully opened the pool, verify that we're ready
2605 2611 * to start pushing transactions.
2606 2612 */
2607 2613 if (state != SPA_LOAD_TRYIMPORT) {
2608 2614 if (error = spa_load_verify(spa))
2609 2615 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2610 2616 error));
2611 2617 }
2612 2618
2613 2619 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2614 2620 spa->spa_load_max_txg == UINT64_MAX)) {
2615 2621 dmu_tx_t *tx;
2616 2622 int need_update = B_FALSE;
2617 2623
2618 2624 ASSERT(state != SPA_LOAD_TRYIMPORT);
2619 2625
2620 2626 /*
2621 2627 * Claim log blocks that haven't been committed yet.
2622 2628 * This must all happen in a single txg.
2623 2629 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2624 2630 * invoked from zil_claim_log_block()'s i/o done callback.
2625 2631 * Price of rollback is that we abandon the log.
2626 2632 */
2627 2633 spa->spa_claiming = B_TRUE;
2628 2634
2629 2635 tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2630 2636 spa_first_txg(spa));
2631 2637 (void) dmu_objset_find(spa_name(spa),
2632 2638 zil_claim, tx, DS_FIND_CHILDREN);
2633 2639 dmu_tx_commit(tx);
2634 2640
2635 2641 spa->spa_claiming = B_FALSE;
2636 2642
2637 2643 spa_set_log_state(spa, SPA_LOG_GOOD);
2638 2644 spa->spa_sync_on = B_TRUE;
2639 2645 txg_sync_start(spa->spa_dsl_pool);
2640 2646
2641 2647 /*
2642 2648 * Wait for all claims to sync. We sync up to the highest
2643 2649 * claimed log block birth time so that claimed log blocks
2644 2650 * don't appear to be from the future. spa_claim_max_txg
2645 2651 * will have been set for us by either zil_check_log_chain()
2646 2652 * (invoked from spa_check_logs()) or zil_claim() above.
2647 2653 */
2648 2654 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2649 2655
2650 2656 /*
2651 2657 * If the config cache is stale, or we have uninitialized
2652 2658 * metaslabs (see spa_vdev_add()), then update the config.
2653 2659 *
2654 2660 * If this is a verbatim import, trust the current
2655 2661 * in-core spa_config and update the disk labels.
2656 2662 */
2657 2663 if (config_cache_txg != spa->spa_config_txg ||
2658 2664 state == SPA_LOAD_IMPORT ||
2659 2665 state == SPA_LOAD_RECOVER ||
2660 2666 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2661 2667 need_update = B_TRUE;
2662 2668
2663 2669 for (int c = 0; c < rvd->vdev_children; c++)
2664 2670 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2665 2671 need_update = B_TRUE;
2666 2672
2667 2673 /*
2668 2674 * Update the config cache asychronously in case we're the
2669 2675 * root pool, in which case the config cache isn't writable yet.
2670 2676 */
2671 2677 if (need_update)
2672 2678 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2673 2679
2674 2680 /*
2675 2681 * Check all DTLs to see if anything needs resilvering.
2676 2682 */
2677 2683 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2678 2684 vdev_resilver_needed(rvd, NULL, NULL))
2679 2685 spa_async_request(spa, SPA_ASYNC_RESILVER);
2680 2686
2681 2687 /*
2682 2688 * Log the fact that we booted up (so that we can detect if
2683 2689 * we rebooted in the middle of an operation).
2684 2690 */
2685 2691 spa_history_log_version(spa, "open");
2686 2692
2687 2693 /*
2688 2694 * Delete any inconsistent datasets.
|
↓ open down ↓ |
1189 lines elided |
↑ open up ↑ |
2689 2695 */
2690 2696 (void) dmu_objset_find(spa_name(spa),
2691 2697 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2692 2698
2693 2699 /*
2694 2700 * Clean up any stale temporary dataset userrefs.
2695 2701 */
2696 2702 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2697 2703 }
2698 2704
2705 + spa_async_request(spa, SPA_ASYNC_L2CACHE_REBUILD);
2706 +
2699 2707 return (0);
2700 2708 }
2701 2709
2702 2710 static int
2703 2711 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2704 2712 {
2705 2713 int mode = spa->spa_mode;
2706 2714
2707 2715 spa_unload(spa);
2708 2716 spa_deactivate(spa);
2709 2717
2710 2718 spa->spa_load_max_txg--;
2711 2719
2712 2720 spa_activate(spa, mode);
2713 2721 spa_async_suspend(spa);
2714 2722
2715 2723 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2716 2724 }
2717 2725
2718 2726 /*
2719 2727 * If spa_load() fails this function will try loading prior txg's. If
2720 2728 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2721 2729 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2722 2730 * function will not rewind the pool and will return the same error as
2723 2731 * spa_load().
2724 2732 */
2725 2733 static int
2726 2734 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2727 2735 uint64_t max_request, int rewind_flags)
2728 2736 {
2729 2737 nvlist_t *loadinfo = NULL;
2730 2738 nvlist_t *config = NULL;
2731 2739 int load_error, rewind_error;
2732 2740 uint64_t safe_rewind_txg;
2733 2741 uint64_t min_txg;
2734 2742
2735 2743 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2736 2744 spa->spa_load_max_txg = spa->spa_load_txg;
2737 2745 spa_set_log_state(spa, SPA_LOG_CLEAR);
2738 2746 } else {
2739 2747 spa->spa_load_max_txg = max_request;
2740 2748 }
2741 2749
2742 2750 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2743 2751 mosconfig);
2744 2752 if (load_error == 0)
2745 2753 return (0);
2746 2754
2747 2755 if (spa->spa_root_vdev != NULL)
2748 2756 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2749 2757
2750 2758 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2751 2759 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2752 2760
2753 2761 if (rewind_flags & ZPOOL_NEVER_REWIND) {
2754 2762 nvlist_free(config);
2755 2763 return (load_error);
2756 2764 }
2757 2765
2758 2766 if (state == SPA_LOAD_RECOVER) {
2759 2767 /* Price of rolling back is discarding txgs, including log */
2760 2768 spa_set_log_state(spa, SPA_LOG_CLEAR);
2761 2769 } else {
2762 2770 /*
2763 2771 * If we aren't rolling back save the load info from our first
2764 2772 * import attempt so that we can restore it after attempting
2765 2773 * to rewind.
2766 2774 */
2767 2775 loadinfo = spa->spa_load_info;
2768 2776 spa->spa_load_info = fnvlist_alloc();
2769 2777 }
2770 2778
2771 2779 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2772 2780 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2773 2781 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2774 2782 TXG_INITIAL : safe_rewind_txg;
2775 2783
2776 2784 /*
2777 2785 * Continue as long as we're finding errors, we're still within
2778 2786 * the acceptable rewind range, and we're still finding uberblocks
2779 2787 */
2780 2788 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2781 2789 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2782 2790 if (spa->spa_load_max_txg < safe_rewind_txg)
2783 2791 spa->spa_extreme_rewind = B_TRUE;
2784 2792 rewind_error = spa_load_retry(spa, state, mosconfig);
2785 2793 }
2786 2794
2787 2795 spa->spa_extreme_rewind = B_FALSE;
2788 2796 spa->spa_load_max_txg = UINT64_MAX;
2789 2797
2790 2798 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2791 2799 spa_config_set(spa, config);
2792 2800
2793 2801 if (state == SPA_LOAD_RECOVER) {
2794 2802 ASSERT3P(loadinfo, ==, NULL);
2795 2803 return (rewind_error);
2796 2804 } else {
2797 2805 /* Store the rewind info as part of the initial load info */
2798 2806 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2799 2807 spa->spa_load_info);
2800 2808
2801 2809 /* Restore the initial load info */
2802 2810 fnvlist_free(spa->spa_load_info);
2803 2811 spa->spa_load_info = loadinfo;
2804 2812
2805 2813 return (load_error);
2806 2814 }
2807 2815 }
2808 2816
2809 2817 /*
2810 2818 * Pool Open/Import
2811 2819 *
2812 2820 * The import case is identical to an open except that the configuration is sent
2813 2821 * down from userland, instead of grabbed from the configuration cache. For the
2814 2822 * case of an open, the pool configuration will exist in the
2815 2823 * POOL_STATE_UNINITIALIZED state.
2816 2824 *
2817 2825 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2818 2826 * the same time open the pool, without having to keep around the spa_t in some
2819 2827 * ambiguous state.
2820 2828 */
2821 2829 static int
2822 2830 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2823 2831 nvlist_t **config)
2824 2832 {
2825 2833 spa_t *spa;
2826 2834 spa_load_state_t state = SPA_LOAD_OPEN;
2827 2835 int error;
2828 2836 int locked = B_FALSE;
2829 2837
2830 2838 *spapp = NULL;
2831 2839
2832 2840 /*
2833 2841 * As disgusting as this is, we need to support recursive calls to this
2834 2842 * function because dsl_dir_open() is called during spa_load(), and ends
2835 2843 * up calling spa_open() again. The real fix is to figure out how to
2836 2844 * avoid dsl_dir_open() calling this in the first place.
2837 2845 */
2838 2846 if (mutex_owner(&spa_namespace_lock) != curthread) {
2839 2847 mutex_enter(&spa_namespace_lock);
2840 2848 locked = B_TRUE;
2841 2849 }
2842 2850
2843 2851 if ((spa = spa_lookup(pool)) == NULL) {
2844 2852 if (locked)
2845 2853 mutex_exit(&spa_namespace_lock);
2846 2854 return (SET_ERROR(ENOENT));
2847 2855 }
2848 2856
2849 2857 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2850 2858 zpool_rewind_policy_t policy;
2851 2859
2852 2860 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2853 2861 &policy);
2854 2862 if (policy.zrp_request & ZPOOL_DO_REWIND)
2855 2863 state = SPA_LOAD_RECOVER;
2856 2864
2857 2865 spa_activate(spa, spa_mode_global);
2858 2866
2859 2867 if (state != SPA_LOAD_RECOVER)
2860 2868 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2861 2869
2862 2870 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2863 2871 policy.zrp_request);
2864 2872
2865 2873 if (error == EBADF) {
2866 2874 /*
2867 2875 * If vdev_validate() returns failure (indicated by
2868 2876 * EBADF), it indicates that one of the vdevs indicates
2869 2877 * that the pool has been exported or destroyed. If
2870 2878 * this is the case, the config cache is out of sync and
2871 2879 * we should remove the pool from the namespace.
2872 2880 */
2873 2881 spa_unload(spa);
2874 2882 spa_deactivate(spa);
2875 2883 spa_config_sync(spa, B_TRUE, B_TRUE);
2876 2884 spa_remove(spa);
2877 2885 if (locked)
2878 2886 mutex_exit(&spa_namespace_lock);
2879 2887 return (SET_ERROR(ENOENT));
2880 2888 }
2881 2889
2882 2890 if (error) {
2883 2891 /*
2884 2892 * We can't open the pool, but we still have useful
2885 2893 * information: the state of each vdev after the
2886 2894 * attempted vdev_open(). Return this to the user.
2887 2895 */
2888 2896 if (config != NULL && spa->spa_config) {
2889 2897 VERIFY(nvlist_dup(spa->spa_config, config,
2890 2898 KM_SLEEP) == 0);
2891 2899 VERIFY(nvlist_add_nvlist(*config,
2892 2900 ZPOOL_CONFIG_LOAD_INFO,
2893 2901 spa->spa_load_info) == 0);
2894 2902 }
2895 2903 spa_unload(spa);
2896 2904 spa_deactivate(spa);
2897 2905 spa->spa_last_open_failed = error;
2898 2906 if (locked)
2899 2907 mutex_exit(&spa_namespace_lock);
2900 2908 *spapp = NULL;
2901 2909 return (error);
2902 2910 }
2903 2911 }
2904 2912
2905 2913 spa_open_ref(spa, tag);
2906 2914
2907 2915 if (config != NULL)
2908 2916 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2909 2917
2910 2918 /*
2911 2919 * If we've recovered the pool, pass back any information we
2912 2920 * gathered while doing the load.
2913 2921 */
2914 2922 if (state == SPA_LOAD_RECOVER) {
2915 2923 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2916 2924 spa->spa_load_info) == 0);
2917 2925 }
2918 2926
2919 2927 if (locked) {
2920 2928 spa->spa_last_open_failed = 0;
2921 2929 spa->spa_last_ubsync_txg = 0;
2922 2930 spa->spa_load_txg = 0;
2923 2931 mutex_exit(&spa_namespace_lock);
2924 2932 }
2925 2933
2926 2934 *spapp = spa;
2927 2935
2928 2936 return (0);
2929 2937 }
2930 2938
2931 2939 int
2932 2940 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2933 2941 nvlist_t **config)
2934 2942 {
2935 2943 return (spa_open_common(name, spapp, tag, policy, config));
2936 2944 }
2937 2945
2938 2946 int
2939 2947 spa_open(const char *name, spa_t **spapp, void *tag)
2940 2948 {
2941 2949 return (spa_open_common(name, spapp, tag, NULL, NULL));
2942 2950 }
2943 2951
2944 2952 /*
2945 2953 * Lookup the given spa_t, incrementing the inject count in the process,
2946 2954 * preventing it from being exported or destroyed.
2947 2955 */
2948 2956 spa_t *
2949 2957 spa_inject_addref(char *name)
2950 2958 {
2951 2959 spa_t *spa;
2952 2960
2953 2961 mutex_enter(&spa_namespace_lock);
2954 2962 if ((spa = spa_lookup(name)) == NULL) {
2955 2963 mutex_exit(&spa_namespace_lock);
2956 2964 return (NULL);
2957 2965 }
2958 2966 spa->spa_inject_ref++;
2959 2967 mutex_exit(&spa_namespace_lock);
2960 2968
2961 2969 return (spa);
2962 2970 }
2963 2971
2964 2972 void
2965 2973 spa_inject_delref(spa_t *spa)
2966 2974 {
2967 2975 mutex_enter(&spa_namespace_lock);
2968 2976 spa->spa_inject_ref--;
2969 2977 mutex_exit(&spa_namespace_lock);
2970 2978 }
2971 2979
2972 2980 /*
2973 2981 * Add spares device information to the nvlist.
2974 2982 */
2975 2983 static void
2976 2984 spa_add_spares(spa_t *spa, nvlist_t *config)
2977 2985 {
2978 2986 nvlist_t **spares;
2979 2987 uint_t i, nspares;
2980 2988 nvlist_t *nvroot;
2981 2989 uint64_t guid;
2982 2990 vdev_stat_t *vs;
2983 2991 uint_t vsc;
2984 2992 uint64_t pool;
2985 2993
2986 2994 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2987 2995
2988 2996 if (spa->spa_spares.sav_count == 0)
2989 2997 return;
2990 2998
2991 2999 VERIFY(nvlist_lookup_nvlist(config,
2992 3000 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2993 3001 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2994 3002 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2995 3003 if (nspares != 0) {
2996 3004 VERIFY(nvlist_add_nvlist_array(nvroot,
2997 3005 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2998 3006 VERIFY(nvlist_lookup_nvlist_array(nvroot,
2999 3007 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3000 3008
3001 3009 /*
3002 3010 * Go through and find any spares which have since been
3003 3011 * repurposed as an active spare. If this is the case, update
3004 3012 * their status appropriately.
3005 3013 */
3006 3014 for (i = 0; i < nspares; i++) {
3007 3015 VERIFY(nvlist_lookup_uint64(spares[i],
3008 3016 ZPOOL_CONFIG_GUID, &guid) == 0);
3009 3017 if (spa_spare_exists(guid, &pool, NULL) &&
3010 3018 pool != 0ULL) {
3011 3019 VERIFY(nvlist_lookup_uint64_array(
3012 3020 spares[i], ZPOOL_CONFIG_VDEV_STATS,
3013 3021 (uint64_t **)&vs, &vsc) == 0);
3014 3022 vs->vs_state = VDEV_STATE_CANT_OPEN;
3015 3023 vs->vs_aux = VDEV_AUX_SPARED;
3016 3024 }
3017 3025 }
3018 3026 }
3019 3027 }
3020 3028
3021 3029 /*
3022 3030 * Add l2cache device information to the nvlist, including vdev stats.
3023 3031 */
3024 3032 static void
3025 3033 spa_add_l2cache(spa_t *spa, nvlist_t *config)
3026 3034 {
3027 3035 nvlist_t **l2cache;
3028 3036 uint_t i, j, nl2cache;
3029 3037 nvlist_t *nvroot;
3030 3038 uint64_t guid;
3031 3039 vdev_t *vd;
3032 3040 vdev_stat_t *vs;
3033 3041 uint_t vsc;
3034 3042
3035 3043 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3036 3044
3037 3045 if (spa->spa_l2cache.sav_count == 0)
3038 3046 return;
3039 3047
3040 3048 VERIFY(nvlist_lookup_nvlist(config,
3041 3049 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3042 3050 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3043 3051 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3044 3052 if (nl2cache != 0) {
3045 3053 VERIFY(nvlist_add_nvlist_array(nvroot,
3046 3054 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3047 3055 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3048 3056 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3049 3057
3050 3058 /*
3051 3059 * Update level 2 cache device stats.
3052 3060 */
3053 3061
3054 3062 for (i = 0; i < nl2cache; i++) {
3055 3063 VERIFY(nvlist_lookup_uint64(l2cache[i],
3056 3064 ZPOOL_CONFIG_GUID, &guid) == 0);
3057 3065
3058 3066 vd = NULL;
3059 3067 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3060 3068 if (guid ==
3061 3069 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3062 3070 vd = spa->spa_l2cache.sav_vdevs[j];
3063 3071 break;
3064 3072 }
3065 3073 }
3066 3074 ASSERT(vd != NULL);
3067 3075
3068 3076 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3069 3077 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3070 3078 == 0);
3071 3079 vdev_get_stats(vd, vs);
3072 3080 }
3073 3081 }
3074 3082 }
3075 3083
3076 3084 static void
3077 3085 spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3078 3086 {
3079 3087 nvlist_t *features;
3080 3088 zap_cursor_t zc;
3081 3089 zap_attribute_t za;
3082 3090
3083 3091 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3084 3092 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3085 3093
3086 3094 if (spa->spa_feat_for_read_obj != 0) {
3087 3095 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3088 3096 spa->spa_feat_for_read_obj);
3089 3097 zap_cursor_retrieve(&zc, &za) == 0;
3090 3098 zap_cursor_advance(&zc)) {
3091 3099 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3092 3100 za.za_num_integers == 1);
3093 3101 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3094 3102 za.za_first_integer));
3095 3103 }
3096 3104 zap_cursor_fini(&zc);
3097 3105 }
3098 3106
3099 3107 if (spa->spa_feat_for_write_obj != 0) {
3100 3108 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3101 3109 spa->spa_feat_for_write_obj);
3102 3110 zap_cursor_retrieve(&zc, &za) == 0;
3103 3111 zap_cursor_advance(&zc)) {
3104 3112 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3105 3113 za.za_num_integers == 1);
3106 3114 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3107 3115 za.za_first_integer));
3108 3116 }
3109 3117 zap_cursor_fini(&zc);
3110 3118 }
3111 3119
3112 3120 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3113 3121 features) == 0);
3114 3122 nvlist_free(features);
3115 3123 }
3116 3124
3117 3125 int
3118 3126 spa_get_stats(const char *name, nvlist_t **config,
3119 3127 char *altroot, size_t buflen)
3120 3128 {
3121 3129 int error;
3122 3130 spa_t *spa;
3123 3131
3124 3132 *config = NULL;
3125 3133 error = spa_open_common(name, &spa, FTAG, NULL, config);
3126 3134
3127 3135 if (spa != NULL) {
3128 3136 /*
3129 3137 * This still leaves a window of inconsistency where the spares
3130 3138 * or l2cache devices could change and the config would be
3131 3139 * self-inconsistent.
3132 3140 */
3133 3141 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3134 3142
3135 3143 if (*config != NULL) {
3136 3144 uint64_t loadtimes[2];
3137 3145
3138 3146 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3139 3147 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3140 3148 VERIFY(nvlist_add_uint64_array(*config,
3141 3149 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3142 3150
3143 3151 VERIFY(nvlist_add_uint64(*config,
3144 3152 ZPOOL_CONFIG_ERRCOUNT,
3145 3153 spa_get_errlog_size(spa)) == 0);
3146 3154
3147 3155 if (spa_suspended(spa))
3148 3156 VERIFY(nvlist_add_uint64(*config,
3149 3157 ZPOOL_CONFIG_SUSPENDED,
3150 3158 spa->spa_failmode) == 0);
3151 3159
3152 3160 spa_add_spares(spa, *config);
3153 3161 spa_add_l2cache(spa, *config);
3154 3162 spa_add_feature_stats(spa, *config);
3155 3163 }
3156 3164 }
3157 3165
3158 3166 /*
3159 3167 * We want to get the alternate root even for faulted pools, so we cheat
3160 3168 * and call spa_lookup() directly.
3161 3169 */
3162 3170 if (altroot) {
3163 3171 if (spa == NULL) {
3164 3172 mutex_enter(&spa_namespace_lock);
3165 3173 spa = spa_lookup(name);
3166 3174 if (spa)
3167 3175 spa_altroot(spa, altroot, buflen);
3168 3176 else
3169 3177 altroot[0] = '\0';
3170 3178 spa = NULL;
3171 3179 mutex_exit(&spa_namespace_lock);
3172 3180 } else {
3173 3181 spa_altroot(spa, altroot, buflen);
3174 3182 }
3175 3183 }
3176 3184
3177 3185 if (spa != NULL) {
3178 3186 spa_config_exit(spa, SCL_CONFIG, FTAG);
3179 3187 spa_close(spa, FTAG);
3180 3188 }
3181 3189
3182 3190 return (error);
3183 3191 }
3184 3192
3185 3193 /*
3186 3194 * Validate that the auxiliary device array is well formed. We must have an
3187 3195 * array of nvlists, each which describes a valid leaf vdev. If this is an
3188 3196 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3189 3197 * specified, as long as they are well-formed.
3190 3198 */
3191 3199 static int
3192 3200 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3193 3201 spa_aux_vdev_t *sav, const char *config, uint64_t version,
3194 3202 vdev_labeltype_t label)
3195 3203 {
3196 3204 nvlist_t **dev;
3197 3205 uint_t i, ndev;
3198 3206 vdev_t *vd;
3199 3207 int error;
3200 3208
3201 3209 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3202 3210
3203 3211 /*
3204 3212 * It's acceptable to have no devs specified.
3205 3213 */
3206 3214 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3207 3215 return (0);
3208 3216
3209 3217 if (ndev == 0)
3210 3218 return (SET_ERROR(EINVAL));
3211 3219
3212 3220 /*
3213 3221 * Make sure the pool is formatted with a version that supports this
3214 3222 * device type.
3215 3223 */
3216 3224 if (spa_version(spa) < version)
3217 3225 return (SET_ERROR(ENOTSUP));
3218 3226
3219 3227 /*
3220 3228 * Set the pending device list so we correctly handle device in-use
3221 3229 * checking.
3222 3230 */
3223 3231 sav->sav_pending = dev;
3224 3232 sav->sav_npending = ndev;
3225 3233
3226 3234 for (i = 0; i < ndev; i++) {
3227 3235 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3228 3236 mode)) != 0)
3229 3237 goto out;
3230 3238
3231 3239 if (!vd->vdev_ops->vdev_op_leaf) {
3232 3240 vdev_free(vd);
3233 3241 error = SET_ERROR(EINVAL);
3234 3242 goto out;
3235 3243 }
3236 3244
3237 3245 /*
3238 3246 * The L2ARC currently only supports disk devices in
3239 3247 * kernel context. For user-level testing, we allow it.
3240 3248 */
3241 3249 #ifdef _KERNEL
3242 3250 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3243 3251 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3244 3252 error = SET_ERROR(ENOTBLK);
3245 3253 vdev_free(vd);
3246 3254 goto out;
3247 3255 }
3248 3256 #endif
3249 3257 vd->vdev_top = vd;
3250 3258
3251 3259 if ((error = vdev_open(vd)) == 0 &&
3252 3260 (error = vdev_label_init(vd, crtxg, label)) == 0) {
3253 3261 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3254 3262 vd->vdev_guid) == 0);
3255 3263 }
3256 3264
3257 3265 vdev_free(vd);
3258 3266
3259 3267 if (error &&
3260 3268 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3261 3269 goto out;
3262 3270 else
3263 3271 error = 0;
3264 3272 }
3265 3273
3266 3274 out:
3267 3275 sav->sav_pending = NULL;
3268 3276 sav->sav_npending = 0;
3269 3277 return (error);
3270 3278 }
3271 3279
3272 3280 static int
3273 3281 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3274 3282 {
3275 3283 int error;
3276 3284
3277 3285 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3278 3286
3279 3287 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3280 3288 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3281 3289 VDEV_LABEL_SPARE)) != 0) {
3282 3290 return (error);
3283 3291 }
3284 3292
3285 3293 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3286 3294 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3287 3295 VDEV_LABEL_L2CACHE));
3288 3296 }
3289 3297
3290 3298 static void
3291 3299 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3292 3300 const char *config)
3293 3301 {
3294 3302 int i;
3295 3303
3296 3304 if (sav->sav_config != NULL) {
3297 3305 nvlist_t **olddevs;
3298 3306 uint_t oldndevs;
3299 3307 nvlist_t **newdevs;
3300 3308
3301 3309 /*
3302 3310 * Generate new dev list by concatentating with the
3303 3311 * current dev list.
3304 3312 */
3305 3313 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3306 3314 &olddevs, &oldndevs) == 0);
3307 3315
3308 3316 newdevs = kmem_alloc(sizeof (void *) *
3309 3317 (ndevs + oldndevs), KM_SLEEP);
3310 3318 for (i = 0; i < oldndevs; i++)
3311 3319 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3312 3320 KM_SLEEP) == 0);
3313 3321 for (i = 0; i < ndevs; i++)
3314 3322 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3315 3323 KM_SLEEP) == 0);
3316 3324
3317 3325 VERIFY(nvlist_remove(sav->sav_config, config,
3318 3326 DATA_TYPE_NVLIST_ARRAY) == 0);
3319 3327
3320 3328 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3321 3329 config, newdevs, ndevs + oldndevs) == 0);
3322 3330 for (i = 0; i < oldndevs + ndevs; i++)
3323 3331 nvlist_free(newdevs[i]);
3324 3332 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3325 3333 } else {
3326 3334 /*
3327 3335 * Generate a new dev list.
3328 3336 */
3329 3337 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3330 3338 KM_SLEEP) == 0);
3331 3339 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3332 3340 devs, ndevs) == 0);
3333 3341 }
3334 3342 }
3335 3343
3336 3344 /*
3337 3345 * Stop and drop level 2 ARC devices
3338 3346 */
3339 3347 void
3340 3348 spa_l2cache_drop(spa_t *spa)
3341 3349 {
3342 3350 vdev_t *vd;
3343 3351 int i;
3344 3352 spa_aux_vdev_t *sav = &spa->spa_l2cache;
3345 3353
3346 3354 for (i = 0; i < sav->sav_count; i++) {
3347 3355 uint64_t pool;
3348 3356
3349 3357 vd = sav->sav_vdevs[i];
3350 3358 ASSERT(vd != NULL);
3351 3359
3352 3360 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3353 3361 pool != 0ULL && l2arc_vdev_present(vd))
3354 3362 l2arc_remove_vdev(vd);
3355 3363 }
3356 3364 }
3357 3365
3358 3366 /*
3359 3367 * Pool Creation
3360 3368 */
3361 3369 int
3362 3370 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3363 3371 nvlist_t *zplprops)
3364 3372 {
3365 3373 spa_t *spa;
3366 3374 char *altroot = NULL;
3367 3375 vdev_t *rvd;
3368 3376 dsl_pool_t *dp;
3369 3377 dmu_tx_t *tx;
3370 3378 int error = 0;
3371 3379 uint64_t txg = TXG_INITIAL;
3372 3380 nvlist_t **spares, **l2cache;
3373 3381 uint_t nspares, nl2cache;
3374 3382 uint64_t version, obj;
3375 3383 boolean_t has_features;
3376 3384
3377 3385 /*
3378 3386 * If this pool already exists, return failure.
3379 3387 */
3380 3388 mutex_enter(&spa_namespace_lock);
3381 3389 if (spa_lookup(pool) != NULL) {
3382 3390 mutex_exit(&spa_namespace_lock);
3383 3391 return (SET_ERROR(EEXIST));
3384 3392 }
3385 3393
3386 3394 /*
3387 3395 * Allocate a new spa_t structure.
3388 3396 */
3389 3397 (void) nvlist_lookup_string(props,
3390 3398 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3391 3399 spa = spa_add(pool, NULL, altroot);
3392 3400 spa_activate(spa, spa_mode_global);
3393 3401
3394 3402 if (props && (error = spa_prop_validate(spa, props))) {
3395 3403 spa_deactivate(spa);
3396 3404 spa_remove(spa);
3397 3405 mutex_exit(&spa_namespace_lock);
3398 3406 return (error);
3399 3407 }
3400 3408
3401 3409 has_features = B_FALSE;
3402 3410 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3403 3411 elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3404 3412 if (zpool_prop_feature(nvpair_name(elem)))
3405 3413 has_features = B_TRUE;
3406 3414 }
3407 3415
3408 3416 if (has_features || nvlist_lookup_uint64(props,
3409 3417 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3410 3418 version = SPA_VERSION;
3411 3419 }
3412 3420 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3413 3421
3414 3422 spa->spa_first_txg = txg;
3415 3423 spa->spa_uberblock.ub_txg = txg - 1;
3416 3424 spa->spa_uberblock.ub_version = version;
3417 3425 spa->spa_ubsync = spa->spa_uberblock;
3418 3426
3419 3427 /*
3420 3428 * Create "The Godfather" zio to hold all async IOs
3421 3429 */
3422 3430 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
3423 3431 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
3424 3432
3425 3433 /*
3426 3434 * Create the root vdev.
3427 3435 */
3428 3436 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3429 3437
3430 3438 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3431 3439
3432 3440 ASSERT(error != 0 || rvd != NULL);
3433 3441 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3434 3442
3435 3443 if (error == 0 && !zfs_allocatable_devs(nvroot))
3436 3444 error = SET_ERROR(EINVAL);
3437 3445
3438 3446 if (error == 0 &&
3439 3447 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3440 3448 (error = spa_validate_aux(spa, nvroot, txg,
3441 3449 VDEV_ALLOC_ADD)) == 0) {
3442 3450 for (int c = 0; c < rvd->vdev_children; c++) {
3443 3451 vdev_metaslab_set_size(rvd->vdev_child[c]);
3444 3452 vdev_expand(rvd->vdev_child[c], txg);
3445 3453 }
3446 3454 }
3447 3455
3448 3456 spa_config_exit(spa, SCL_ALL, FTAG);
3449 3457
3450 3458 if (error != 0) {
3451 3459 spa_unload(spa);
3452 3460 spa_deactivate(spa);
3453 3461 spa_remove(spa);
3454 3462 mutex_exit(&spa_namespace_lock);
3455 3463 return (error);
3456 3464 }
3457 3465
3458 3466 /*
3459 3467 * Get the list of spares, if specified.
3460 3468 */
3461 3469 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3462 3470 &spares, &nspares) == 0) {
3463 3471 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3464 3472 KM_SLEEP) == 0);
3465 3473 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3466 3474 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3467 3475 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3468 3476 spa_load_spares(spa);
3469 3477 spa_config_exit(spa, SCL_ALL, FTAG);
3470 3478 spa->spa_spares.sav_sync = B_TRUE;
3471 3479 }
3472 3480
3473 3481 /*
3474 3482 * Get the list of level 2 cache devices, if specified.
3475 3483 */
3476 3484 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3477 3485 &l2cache, &nl2cache) == 0) {
3478 3486 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3479 3487 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3480 3488 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3481 3489 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3482 3490 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3483 3491 spa_load_l2cache(spa);
3484 3492 spa_config_exit(spa, SCL_ALL, FTAG);
3485 3493 spa->spa_l2cache.sav_sync = B_TRUE;
3486 3494 }
3487 3495
3488 3496 spa->spa_is_initializing = B_TRUE;
3489 3497 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3490 3498 spa->spa_meta_objset = dp->dp_meta_objset;
3491 3499 spa->spa_is_initializing = B_FALSE;
3492 3500
3493 3501 /*
3494 3502 * Create DDTs (dedup tables).
3495 3503 */
3496 3504 ddt_create(spa);
3497 3505
3498 3506 spa_update_dspace(spa);
3499 3507
3500 3508 tx = dmu_tx_create_assigned(dp, txg);
3501 3509
3502 3510 /*
3503 3511 * Create the pool config object.
3504 3512 */
3505 3513 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3506 3514 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3507 3515 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3508 3516
3509 3517 if (zap_add(spa->spa_meta_objset,
3510 3518 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3511 3519 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3512 3520 cmn_err(CE_PANIC, "failed to add pool config");
3513 3521 }
3514 3522
3515 3523 if (spa_version(spa) >= SPA_VERSION_FEATURES)
3516 3524 spa_feature_create_zap_objects(spa, tx);
3517 3525
3518 3526 if (zap_add(spa->spa_meta_objset,
3519 3527 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3520 3528 sizeof (uint64_t), 1, &version, tx) != 0) {
3521 3529 cmn_err(CE_PANIC, "failed to add pool version");
3522 3530 }
3523 3531
3524 3532 /* Newly created pools with the right version are always deflated. */
3525 3533 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3526 3534 spa->spa_deflate = TRUE;
3527 3535 if (zap_add(spa->spa_meta_objset,
3528 3536 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3529 3537 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3530 3538 cmn_err(CE_PANIC, "failed to add deflate");
3531 3539 }
3532 3540 }
3533 3541
3534 3542 /*
3535 3543 * Create the deferred-free bpobj. Turn off compression
3536 3544 * because sync-to-convergence takes longer if the blocksize
3537 3545 * keeps changing.
3538 3546 */
3539 3547 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3540 3548 dmu_object_set_compress(spa->spa_meta_objset, obj,
3541 3549 ZIO_COMPRESS_OFF, tx);
3542 3550 if (zap_add(spa->spa_meta_objset,
3543 3551 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3544 3552 sizeof (uint64_t), 1, &obj, tx) != 0) {
3545 3553 cmn_err(CE_PANIC, "failed to add bpobj");
3546 3554 }
3547 3555 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3548 3556 spa->spa_meta_objset, obj));
3549 3557
3550 3558 /*
3551 3559 * Create the pool's history object.
3552 3560 */
3553 3561 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3554 3562 spa_history_create_obj(spa, tx);
3555 3563
3556 3564 /*
3557 3565 * Set pool properties.
3558 3566 */
3559 3567 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3560 3568 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3561 3569 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3562 3570 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3563 3571
3564 3572 if (props != NULL) {
3565 3573 spa_configfile_set(spa, props, B_FALSE);
3566 3574 spa_sync_props(props, tx);
3567 3575 }
3568 3576
3569 3577 dmu_tx_commit(tx);
3570 3578
3571 3579 spa->spa_sync_on = B_TRUE;
3572 3580 txg_sync_start(spa->spa_dsl_pool);
3573 3581
3574 3582 /*
3575 3583 * We explicitly wait for the first transaction to complete so that our
3576 3584 * bean counters are appropriately updated.
3577 3585 */
3578 3586 txg_wait_synced(spa->spa_dsl_pool, txg);
3579 3587
3580 3588 spa_config_sync(spa, B_FALSE, B_TRUE);
3581 3589
3582 3590 spa_history_log_version(spa, "create");
3583 3591
3584 3592 spa->spa_minref = refcount_count(&spa->spa_refcount);
3585 3593
3586 3594 mutex_exit(&spa_namespace_lock);
3587 3595
3588 3596 return (0);
3589 3597 }
3590 3598
3591 3599 #ifdef _KERNEL
3592 3600 /*
3593 3601 * Get the root pool information from the root disk, then import the root pool
3594 3602 * during the system boot up time.
3595 3603 */
3596 3604 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3597 3605
3598 3606 static nvlist_t *
3599 3607 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3600 3608 {
3601 3609 nvlist_t *config;
3602 3610 nvlist_t *nvtop, *nvroot;
3603 3611 uint64_t pgid;
3604 3612
3605 3613 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3606 3614 return (NULL);
3607 3615
3608 3616 /*
3609 3617 * Add this top-level vdev to the child array.
3610 3618 */
3611 3619 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3612 3620 &nvtop) == 0);
3613 3621 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3614 3622 &pgid) == 0);
3615 3623 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3616 3624
3617 3625 /*
3618 3626 * Put this pool's top-level vdevs into a root vdev.
3619 3627 */
3620 3628 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3621 3629 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3622 3630 VDEV_TYPE_ROOT) == 0);
3623 3631 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3624 3632 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3625 3633 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3626 3634 &nvtop, 1) == 0);
3627 3635
3628 3636 /*
3629 3637 * Replace the existing vdev_tree with the new root vdev in
3630 3638 * this pool's configuration (remove the old, add the new).
3631 3639 */
3632 3640 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3633 3641 nvlist_free(nvroot);
3634 3642 return (config);
3635 3643 }
3636 3644
3637 3645 /*
3638 3646 * Walk the vdev tree and see if we can find a device with "better"
3639 3647 * configuration. A configuration is "better" if the label on that
3640 3648 * device has a more recent txg.
3641 3649 */
3642 3650 static void
3643 3651 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3644 3652 {
3645 3653 for (int c = 0; c < vd->vdev_children; c++)
3646 3654 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3647 3655
3648 3656 if (vd->vdev_ops->vdev_op_leaf) {
3649 3657 nvlist_t *label;
3650 3658 uint64_t label_txg;
3651 3659
3652 3660 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3653 3661 &label) != 0)
3654 3662 return;
3655 3663
3656 3664 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3657 3665 &label_txg) == 0);
3658 3666
3659 3667 /*
3660 3668 * Do we have a better boot device?
3661 3669 */
3662 3670 if (label_txg > *txg) {
3663 3671 *txg = label_txg;
3664 3672 *avd = vd;
3665 3673 }
3666 3674 nvlist_free(label);
3667 3675 }
3668 3676 }
3669 3677
3670 3678 /*
3671 3679 * Import a root pool.
3672 3680 *
3673 3681 * For x86. devpath_list will consist of devid and/or physpath name of
3674 3682 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3675 3683 * The GRUB "findroot" command will return the vdev we should boot.
3676 3684 *
3677 3685 * For Sparc, devpath_list consists the physpath name of the booting device
3678 3686 * no matter the rootpool is a single device pool or a mirrored pool.
3679 3687 * e.g.
3680 3688 * "/pci@1f,0/ide@d/disk@0,0:a"
3681 3689 */
3682 3690 int
3683 3691 spa_import_rootpool(char *devpath, char *devid)
3684 3692 {
3685 3693 spa_t *spa;
3686 3694 vdev_t *rvd, *bvd, *avd = NULL;
3687 3695 nvlist_t *config, *nvtop;
3688 3696 uint64_t guid, txg;
3689 3697 char *pname;
3690 3698 int error;
3691 3699
3692 3700 /*
3693 3701 * Read the label from the boot device and generate a configuration.
3694 3702 */
3695 3703 config = spa_generate_rootconf(devpath, devid, &guid);
3696 3704 #if defined(_OBP) && defined(_KERNEL)
3697 3705 if (config == NULL) {
3698 3706 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3699 3707 /* iscsi boot */
3700 3708 get_iscsi_bootpath_phy(devpath);
3701 3709 config = spa_generate_rootconf(devpath, devid, &guid);
3702 3710 }
3703 3711 }
3704 3712 #endif
3705 3713 if (config == NULL) {
3706 3714 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3707 3715 devpath);
3708 3716 return (SET_ERROR(EIO));
3709 3717 }
3710 3718
3711 3719 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3712 3720 &pname) == 0);
3713 3721 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3714 3722
3715 3723 mutex_enter(&spa_namespace_lock);
3716 3724 if ((spa = spa_lookup(pname)) != NULL) {
3717 3725 /*
3718 3726 * Remove the existing root pool from the namespace so that we
3719 3727 * can replace it with the correct config we just read in.
3720 3728 */
3721 3729 spa_remove(spa);
3722 3730 }
3723 3731
3724 3732 spa = spa_add(pname, config, NULL);
3725 3733 spa->spa_is_root = B_TRUE;
3726 3734 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3727 3735
3728 3736 /*
3729 3737 * Build up a vdev tree based on the boot device's label config.
3730 3738 */
3731 3739 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3732 3740 &nvtop) == 0);
3733 3741 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3734 3742 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3735 3743 VDEV_ALLOC_ROOTPOOL);
3736 3744 spa_config_exit(spa, SCL_ALL, FTAG);
3737 3745 if (error) {
3738 3746 mutex_exit(&spa_namespace_lock);
3739 3747 nvlist_free(config);
3740 3748 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3741 3749 pname);
3742 3750 return (error);
3743 3751 }
3744 3752
3745 3753 /*
3746 3754 * Get the boot vdev.
3747 3755 */
3748 3756 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3749 3757 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3750 3758 (u_longlong_t)guid);
3751 3759 error = SET_ERROR(ENOENT);
3752 3760 goto out;
3753 3761 }
3754 3762
3755 3763 /*
3756 3764 * Determine if there is a better boot device.
3757 3765 */
3758 3766 avd = bvd;
3759 3767 spa_alt_rootvdev(rvd, &avd, &txg);
3760 3768 if (avd != bvd) {
3761 3769 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3762 3770 "try booting from '%s'", avd->vdev_path);
3763 3771 error = SET_ERROR(EINVAL);
3764 3772 goto out;
3765 3773 }
3766 3774
3767 3775 /*
3768 3776 * If the boot device is part of a spare vdev then ensure that
3769 3777 * we're booting off the active spare.
3770 3778 */
3771 3779 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3772 3780 !bvd->vdev_isspare) {
3773 3781 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3774 3782 "try booting from '%s'",
3775 3783 bvd->vdev_parent->
3776 3784 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3777 3785 error = SET_ERROR(EINVAL);
3778 3786 goto out;
3779 3787 }
3780 3788
3781 3789 error = 0;
3782 3790 out:
3783 3791 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3784 3792 vdev_free(rvd);
3785 3793 spa_config_exit(spa, SCL_ALL, FTAG);
3786 3794 mutex_exit(&spa_namespace_lock);
3787 3795
3788 3796 nvlist_free(config);
3789 3797 return (error);
3790 3798 }
3791 3799
3792 3800 #endif
3793 3801
3794 3802 /*
3795 3803 * Import a non-root pool into the system.
3796 3804 */
3797 3805 int
3798 3806 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3799 3807 {
3800 3808 spa_t *spa;
3801 3809 char *altroot = NULL;
3802 3810 spa_load_state_t state = SPA_LOAD_IMPORT;
3803 3811 zpool_rewind_policy_t policy;
3804 3812 uint64_t mode = spa_mode_global;
3805 3813 uint64_t readonly = B_FALSE;
3806 3814 int error;
3807 3815 nvlist_t *nvroot;
3808 3816 nvlist_t **spares, **l2cache;
3809 3817 uint_t nspares, nl2cache;
3810 3818
3811 3819 /*
3812 3820 * If a pool with this name exists, return failure.
3813 3821 */
3814 3822 mutex_enter(&spa_namespace_lock);
3815 3823 if (spa_lookup(pool) != NULL) {
3816 3824 mutex_exit(&spa_namespace_lock);
3817 3825 return (SET_ERROR(EEXIST));
3818 3826 }
3819 3827
3820 3828 /*
3821 3829 * Create and initialize the spa structure.
3822 3830 */
3823 3831 (void) nvlist_lookup_string(props,
3824 3832 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3825 3833 (void) nvlist_lookup_uint64(props,
3826 3834 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3827 3835 if (readonly)
3828 3836 mode = FREAD;
3829 3837 spa = spa_add(pool, config, altroot);
3830 3838 spa->spa_import_flags = flags;
3831 3839
3832 3840 /*
3833 3841 * Verbatim import - Take a pool and insert it into the namespace
3834 3842 * as if it had been loaded at boot.
3835 3843 */
3836 3844 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3837 3845 if (props != NULL)
3838 3846 spa_configfile_set(spa, props, B_FALSE);
3839 3847
3840 3848 spa_config_sync(spa, B_FALSE, B_TRUE);
3841 3849
3842 3850 mutex_exit(&spa_namespace_lock);
3843 3851 return (0);
3844 3852 }
3845 3853
3846 3854 spa_activate(spa, mode);
3847 3855
3848 3856 /*
3849 3857 * Don't start async tasks until we know everything is healthy.
3850 3858 */
3851 3859 spa_async_suspend(spa);
3852 3860
3853 3861 zpool_get_rewind_policy(config, &policy);
3854 3862 if (policy.zrp_request & ZPOOL_DO_REWIND)
3855 3863 state = SPA_LOAD_RECOVER;
3856 3864
3857 3865 /*
3858 3866 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
3859 3867 * because the user-supplied config is actually the one to trust when
3860 3868 * doing an import.
3861 3869 */
3862 3870 if (state != SPA_LOAD_RECOVER)
3863 3871 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3864 3872
3865 3873 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3866 3874 policy.zrp_request);
3867 3875
3868 3876 /*
3869 3877 * Propagate anything learned while loading the pool and pass it
3870 3878 * back to caller (i.e. rewind info, missing devices, etc).
3871 3879 */
3872 3880 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3873 3881 spa->spa_load_info) == 0);
3874 3882
3875 3883 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3876 3884 /*
3877 3885 * Toss any existing sparelist, as it doesn't have any validity
3878 3886 * anymore, and conflicts with spa_has_spare().
3879 3887 */
3880 3888 if (spa->spa_spares.sav_config) {
3881 3889 nvlist_free(spa->spa_spares.sav_config);
3882 3890 spa->spa_spares.sav_config = NULL;
3883 3891 spa_load_spares(spa);
3884 3892 }
3885 3893 if (spa->spa_l2cache.sav_config) {
3886 3894 nvlist_free(spa->spa_l2cache.sav_config);
3887 3895 spa->spa_l2cache.sav_config = NULL;
3888 3896 spa_load_l2cache(spa);
3889 3897 }
3890 3898
3891 3899 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3892 3900 &nvroot) == 0);
3893 3901 if (error == 0)
3894 3902 error = spa_validate_aux(spa, nvroot, -1ULL,
3895 3903 VDEV_ALLOC_SPARE);
3896 3904 if (error == 0)
3897 3905 error = spa_validate_aux(spa, nvroot, -1ULL,
3898 3906 VDEV_ALLOC_L2CACHE);
3899 3907 spa_config_exit(spa, SCL_ALL, FTAG);
3900 3908
3901 3909 if (props != NULL)
3902 3910 spa_configfile_set(spa, props, B_FALSE);
3903 3911
3904 3912 if (error != 0 || (props && spa_writeable(spa) &&
3905 3913 (error = spa_prop_set(spa, props)))) {
3906 3914 spa_unload(spa);
3907 3915 spa_deactivate(spa);
3908 3916 spa_remove(spa);
3909 3917 mutex_exit(&spa_namespace_lock);
3910 3918 return (error);
3911 3919 }
3912 3920
3913 3921 spa_async_resume(spa);
3914 3922
3915 3923 /*
3916 3924 * Override any spares and level 2 cache devices as specified by
3917 3925 * the user, as these may have correct device names/devids, etc.
3918 3926 */
3919 3927 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3920 3928 &spares, &nspares) == 0) {
3921 3929 if (spa->spa_spares.sav_config)
3922 3930 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3923 3931 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3924 3932 else
3925 3933 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3926 3934 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3927 3935 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3928 3936 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3929 3937 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3930 3938 spa_load_spares(spa);
3931 3939 spa_config_exit(spa, SCL_ALL, FTAG);
3932 3940 spa->spa_spares.sav_sync = B_TRUE;
3933 3941 }
3934 3942 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3935 3943 &l2cache, &nl2cache) == 0) {
3936 3944 if (spa->spa_l2cache.sav_config)
3937 3945 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3938 3946 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3939 3947 else
3940 3948 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3941 3949 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3942 3950 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3943 3951 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3944 3952 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3945 3953 spa_load_l2cache(spa);
3946 3954 spa_config_exit(spa, SCL_ALL, FTAG);
3947 3955 spa->spa_l2cache.sav_sync = B_TRUE;
3948 3956 }
3949 3957
3950 3958 /*
3951 3959 * Check for any removed devices.
3952 3960 */
3953 3961 if (spa->spa_autoreplace) {
3954 3962 spa_aux_check_removed(&spa->spa_spares);
3955 3963 spa_aux_check_removed(&spa->spa_l2cache);
3956 3964 }
3957 3965
3958 3966 if (spa_writeable(spa)) {
3959 3967 /*
3960 3968 * Update the config cache to include the newly-imported pool.
3961 3969 */
3962 3970 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3963 3971 }
3964 3972
3965 3973 /*
3966 3974 * It's possible that the pool was expanded while it was exported.
3967 3975 * We kick off an async task to handle this for us.
3968 3976 */
3969 3977 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
3970 3978
3971 3979 mutex_exit(&spa_namespace_lock);
3972 3980 spa_history_log_version(spa, "import");
3973 3981
3974 3982 return (0);
3975 3983 }
3976 3984
3977 3985 nvlist_t *
3978 3986 spa_tryimport(nvlist_t *tryconfig)
3979 3987 {
3980 3988 nvlist_t *config = NULL;
3981 3989 char *poolname;
3982 3990 spa_t *spa;
3983 3991 uint64_t state;
3984 3992 int error;
3985 3993
3986 3994 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
3987 3995 return (NULL);
3988 3996
3989 3997 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
3990 3998 return (NULL);
3991 3999
3992 4000 /*
3993 4001 * Create and initialize the spa structure.
3994 4002 */
3995 4003 mutex_enter(&spa_namespace_lock);
3996 4004 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
3997 4005 spa_activate(spa, FREAD);
3998 4006
3999 4007 /*
4000 4008 * Pass off the heavy lifting to spa_load().
4001 4009 * Pass TRUE for mosconfig because the user-supplied config
4002 4010 * is actually the one to trust when doing an import.
4003 4011 */
4004 4012 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
4005 4013
4006 4014 /*
4007 4015 * If 'tryconfig' was at least parsable, return the current config.
4008 4016 */
4009 4017 if (spa->spa_root_vdev != NULL) {
4010 4018 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4011 4019 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4012 4020 poolname) == 0);
4013 4021 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4014 4022 state) == 0);
4015 4023 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4016 4024 spa->spa_uberblock.ub_timestamp) == 0);
4017 4025 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4018 4026 spa->spa_load_info) == 0);
4019 4027
4020 4028 /*
4021 4029 * If the bootfs property exists on this pool then we
4022 4030 * copy it out so that external consumers can tell which
4023 4031 * pools are bootable.
4024 4032 */
4025 4033 if ((!error || error == EEXIST) && spa->spa_bootfs) {
4026 4034 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4027 4035
4028 4036 /*
4029 4037 * We have to play games with the name since the
4030 4038 * pool was opened as TRYIMPORT_NAME.
4031 4039 */
4032 4040 if (dsl_dsobj_to_dsname(spa_name(spa),
4033 4041 spa->spa_bootfs, tmpname) == 0) {
4034 4042 char *cp;
4035 4043 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4036 4044
4037 4045 cp = strchr(tmpname, '/');
4038 4046 if (cp == NULL) {
4039 4047 (void) strlcpy(dsname, tmpname,
4040 4048 MAXPATHLEN);
4041 4049 } else {
4042 4050 (void) snprintf(dsname, MAXPATHLEN,
4043 4051 "%s/%s", poolname, ++cp);
4044 4052 }
4045 4053 VERIFY(nvlist_add_string(config,
4046 4054 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4047 4055 kmem_free(dsname, MAXPATHLEN);
4048 4056 }
4049 4057 kmem_free(tmpname, MAXPATHLEN);
4050 4058 }
4051 4059
4052 4060 /*
4053 4061 * Add the list of hot spares and level 2 cache devices.
4054 4062 */
4055 4063 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4056 4064 spa_add_spares(spa, config);
4057 4065 spa_add_l2cache(spa, config);
4058 4066 spa_config_exit(spa, SCL_CONFIG, FTAG);
4059 4067 }
4060 4068
4061 4069 spa_unload(spa);
4062 4070 spa_deactivate(spa);
4063 4071 spa_remove(spa);
4064 4072 mutex_exit(&spa_namespace_lock);
4065 4073
4066 4074 return (config);
4067 4075 }
4068 4076
4069 4077 /*
4070 4078 * Pool export/destroy
4071 4079 *
4072 4080 * The act of destroying or exporting a pool is very simple. We make sure there
4073 4081 * is no more pending I/O and any references to the pool are gone. Then, we
4074 4082 * update the pool state and sync all the labels to disk, removing the
4075 4083 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4076 4084 * we don't sync the labels or remove the configuration cache.
4077 4085 */
4078 4086 static int
4079 4087 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4080 4088 boolean_t force, boolean_t hardforce)
4081 4089 {
4082 4090 spa_t *spa;
4083 4091
4084 4092 if (oldconfig)
4085 4093 *oldconfig = NULL;
4086 4094
4087 4095 if (!(spa_mode_global & FWRITE))
4088 4096 return (SET_ERROR(EROFS));
4089 4097
4090 4098 mutex_enter(&spa_namespace_lock);
4091 4099 if ((spa = spa_lookup(pool)) == NULL) {
4092 4100 mutex_exit(&spa_namespace_lock);
4093 4101 return (SET_ERROR(ENOENT));
4094 4102 }
4095 4103
4096 4104 /*
4097 4105 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4098 4106 * reacquire the namespace lock, and see if we can export.
4099 4107 */
4100 4108 spa_open_ref(spa, FTAG);
4101 4109 mutex_exit(&spa_namespace_lock);
4102 4110 spa_async_suspend(spa);
4103 4111 mutex_enter(&spa_namespace_lock);
4104 4112 spa_close(spa, FTAG);
4105 4113
4106 4114 /*
4107 4115 * The pool will be in core if it's openable,
4108 4116 * in which case we can modify its state.
4109 4117 */
4110 4118 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4111 4119 /*
4112 4120 * Objsets may be open only because they're dirty, so we
4113 4121 * have to force it to sync before checking spa_refcnt.
4114 4122 */
4115 4123 txg_wait_synced(spa->spa_dsl_pool, 0);
4116 4124
4117 4125 /*
4118 4126 * A pool cannot be exported or destroyed if there are active
4119 4127 * references. If we are resetting a pool, allow references by
4120 4128 * fault injection handlers.
4121 4129 */
4122 4130 if (!spa_refcount_zero(spa) ||
4123 4131 (spa->spa_inject_ref != 0 &&
4124 4132 new_state != POOL_STATE_UNINITIALIZED)) {
4125 4133 spa_async_resume(spa);
4126 4134 mutex_exit(&spa_namespace_lock);
4127 4135 return (SET_ERROR(EBUSY));
4128 4136 }
4129 4137
4130 4138 /*
4131 4139 * A pool cannot be exported if it has an active shared spare.
4132 4140 * This is to prevent other pools stealing the active spare
4133 4141 * from an exported pool. At user's own will, such pool can
4134 4142 * be forcedly exported.
4135 4143 */
4136 4144 if (!force && new_state == POOL_STATE_EXPORTED &&
4137 4145 spa_has_active_shared_spare(spa)) {
4138 4146 spa_async_resume(spa);
4139 4147 mutex_exit(&spa_namespace_lock);
4140 4148 return (SET_ERROR(EXDEV));
4141 4149 }
4142 4150
4143 4151 /*
4144 4152 * We want this to be reflected on every label,
4145 4153 * so mark them all dirty. spa_unload() will do the
4146 4154 * final sync that pushes these changes out.
4147 4155 */
4148 4156 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4149 4157 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4150 4158 spa->spa_state = new_state;
4151 4159 spa->spa_final_txg = spa_last_synced_txg(spa) +
4152 4160 TXG_DEFER_SIZE + 1;
4153 4161 vdev_config_dirty(spa->spa_root_vdev);
4154 4162 spa_config_exit(spa, SCL_ALL, FTAG);
4155 4163 }
4156 4164 }
4157 4165
4158 4166 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4159 4167
4160 4168 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4161 4169 spa_unload(spa);
4162 4170 spa_deactivate(spa);
4163 4171 }
4164 4172
4165 4173 if (oldconfig && spa->spa_config)
4166 4174 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4167 4175
4168 4176 if (new_state != POOL_STATE_UNINITIALIZED) {
4169 4177 if (!hardforce)
4170 4178 spa_config_sync(spa, B_TRUE, B_TRUE);
4171 4179 spa_remove(spa);
4172 4180 }
4173 4181 mutex_exit(&spa_namespace_lock);
4174 4182
4175 4183 return (0);
4176 4184 }
4177 4185
4178 4186 /*
4179 4187 * Destroy a storage pool.
4180 4188 */
4181 4189 int
4182 4190 spa_destroy(char *pool)
4183 4191 {
4184 4192 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4185 4193 B_FALSE, B_FALSE));
4186 4194 }
4187 4195
4188 4196 /*
4189 4197 * Export a storage pool.
4190 4198 */
4191 4199 int
4192 4200 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4193 4201 boolean_t hardforce)
4194 4202 {
4195 4203 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4196 4204 force, hardforce));
4197 4205 }
4198 4206
4199 4207 /*
4200 4208 * Similar to spa_export(), this unloads the spa_t without actually removing it
4201 4209 * from the namespace in any way.
4202 4210 */
4203 4211 int
4204 4212 spa_reset(char *pool)
4205 4213 {
4206 4214 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4207 4215 B_FALSE, B_FALSE));
4208 4216 }
4209 4217
4210 4218 /*
4211 4219 * ==========================================================================
4212 4220 * Device manipulation
4213 4221 * ==========================================================================
4214 4222 */
4215 4223
4216 4224 /*
4217 4225 * Add a device to a storage pool.
4218 4226 */
4219 4227 int
4220 4228 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4221 4229 {
4222 4230 uint64_t txg, id;
4223 4231 int error;
4224 4232 vdev_t *rvd = spa->spa_root_vdev;
4225 4233 vdev_t *vd, *tvd;
4226 4234 nvlist_t **spares, **l2cache;
4227 4235 uint_t nspares, nl2cache;
4228 4236
4229 4237 ASSERT(spa_writeable(spa));
4230 4238
4231 4239 txg = spa_vdev_enter(spa);
4232 4240
4233 4241 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4234 4242 VDEV_ALLOC_ADD)) != 0)
4235 4243 return (spa_vdev_exit(spa, NULL, txg, error));
4236 4244
4237 4245 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
4238 4246
4239 4247 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4240 4248 &nspares) != 0)
4241 4249 nspares = 0;
4242 4250
4243 4251 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4244 4252 &nl2cache) != 0)
4245 4253 nl2cache = 0;
4246 4254
4247 4255 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4248 4256 return (spa_vdev_exit(spa, vd, txg, EINVAL));
4249 4257
4250 4258 if (vd->vdev_children != 0 &&
4251 4259 (error = vdev_create(vd, txg, B_FALSE)) != 0)
4252 4260 return (spa_vdev_exit(spa, vd, txg, error));
4253 4261
4254 4262 /*
4255 4263 * We must validate the spares and l2cache devices after checking the
4256 4264 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4257 4265 */
4258 4266 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4259 4267 return (spa_vdev_exit(spa, vd, txg, error));
4260 4268
4261 4269 /*
4262 4270 * Transfer each new top-level vdev from vd to rvd.
4263 4271 */
4264 4272 for (int c = 0; c < vd->vdev_children; c++) {
4265 4273
4266 4274 /*
4267 4275 * Set the vdev id to the first hole, if one exists.
4268 4276 */
4269 4277 for (id = 0; id < rvd->vdev_children; id++) {
4270 4278 if (rvd->vdev_child[id]->vdev_ishole) {
4271 4279 vdev_free(rvd->vdev_child[id]);
4272 4280 break;
4273 4281 }
4274 4282 }
4275 4283 tvd = vd->vdev_child[c];
4276 4284 vdev_remove_child(vd, tvd);
4277 4285 tvd->vdev_id = id;
4278 4286 vdev_add_child(rvd, tvd);
4279 4287 vdev_config_dirty(tvd);
4280 4288 }
4281 4289
4282 4290 if (nspares != 0) {
4283 4291 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4284 4292 ZPOOL_CONFIG_SPARES);
4285 4293 spa_load_spares(spa);
4286 4294 spa->spa_spares.sav_sync = B_TRUE;
4287 4295 }
4288 4296
4289 4297 if (nl2cache != 0) {
4290 4298 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4291 4299 ZPOOL_CONFIG_L2CACHE);
4292 4300 spa_load_l2cache(spa);
4293 4301 spa->spa_l2cache.sav_sync = B_TRUE;
4294 4302 }
4295 4303
4296 4304 /*
4297 4305 * We have to be careful when adding new vdevs to an existing pool.
4298 4306 * If other threads start allocating from these vdevs before we
4299 4307 * sync the config cache, and we lose power, then upon reboot we may
4300 4308 * fail to open the pool because there are DVAs that the config cache
4301 4309 * can't translate. Therefore, we first add the vdevs without
4302 4310 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4303 4311 * and then let spa_config_update() initialize the new metaslabs.
4304 4312 *
4305 4313 * spa_load() checks for added-but-not-initialized vdevs, so that
4306 4314 * if we lose power at any point in this sequence, the remaining
4307 4315 * steps will be completed the next time we load the pool.
4308 4316 */
4309 4317 (void) spa_vdev_exit(spa, vd, txg, 0);
4310 4318
4311 4319 mutex_enter(&spa_namespace_lock);
4312 4320 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4313 4321 mutex_exit(&spa_namespace_lock);
4314 4322
4315 4323 return (0);
4316 4324 }
4317 4325
4318 4326 /*
4319 4327 * Attach a device to a mirror. The arguments are the path to any device
4320 4328 * in the mirror, and the nvroot for the new device. If the path specifies
4321 4329 * a device that is not mirrored, we automatically insert the mirror vdev.
4322 4330 *
4323 4331 * If 'replacing' is specified, the new device is intended to replace the
4324 4332 * existing device; in this case the two devices are made into their own
4325 4333 * mirror using the 'replacing' vdev, which is functionally identical to
4326 4334 * the mirror vdev (it actually reuses all the same ops) but has a few
4327 4335 * extra rules: you can't attach to it after it's been created, and upon
4328 4336 * completion of resilvering, the first disk (the one being replaced)
4329 4337 * is automatically detached.
4330 4338 */
4331 4339 int
4332 4340 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4333 4341 {
4334 4342 uint64_t txg, dtl_max_txg;
4335 4343 vdev_t *rvd = spa->spa_root_vdev;
4336 4344 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4337 4345 vdev_ops_t *pvops;
4338 4346 char *oldvdpath, *newvdpath;
4339 4347 int newvd_isspare;
4340 4348 int error;
4341 4349
4342 4350 ASSERT(spa_writeable(spa));
4343 4351
4344 4352 txg = spa_vdev_enter(spa);
4345 4353
4346 4354 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4347 4355
4348 4356 if (oldvd == NULL)
4349 4357 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4350 4358
4351 4359 if (!oldvd->vdev_ops->vdev_op_leaf)
4352 4360 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4353 4361
4354 4362 pvd = oldvd->vdev_parent;
4355 4363
4356 4364 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4357 4365 VDEV_ALLOC_ATTACH)) != 0)
4358 4366 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4359 4367
4360 4368 if (newrootvd->vdev_children != 1)
4361 4369 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4362 4370
4363 4371 newvd = newrootvd->vdev_child[0];
4364 4372
4365 4373 if (!newvd->vdev_ops->vdev_op_leaf)
4366 4374 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4367 4375
4368 4376 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4369 4377 return (spa_vdev_exit(spa, newrootvd, txg, error));
4370 4378
4371 4379 /*
4372 4380 * Spares can't replace logs
4373 4381 */
4374 4382 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4375 4383 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4376 4384
4377 4385 if (!replacing) {
4378 4386 /*
4379 4387 * For attach, the only allowable parent is a mirror or the root
4380 4388 * vdev.
4381 4389 */
4382 4390 if (pvd->vdev_ops != &vdev_mirror_ops &&
4383 4391 pvd->vdev_ops != &vdev_root_ops)
4384 4392 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4385 4393
4386 4394 pvops = &vdev_mirror_ops;
4387 4395 } else {
4388 4396 /*
4389 4397 * Active hot spares can only be replaced by inactive hot
4390 4398 * spares.
4391 4399 */
4392 4400 if (pvd->vdev_ops == &vdev_spare_ops &&
4393 4401 oldvd->vdev_isspare &&
4394 4402 !spa_has_spare(spa, newvd->vdev_guid))
4395 4403 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4396 4404
4397 4405 /*
4398 4406 * If the source is a hot spare, and the parent isn't already a
4399 4407 * spare, then we want to create a new hot spare. Otherwise, we
4400 4408 * want to create a replacing vdev. The user is not allowed to
4401 4409 * attach to a spared vdev child unless the 'isspare' state is
4402 4410 * the same (spare replaces spare, non-spare replaces
4403 4411 * non-spare).
4404 4412 */
4405 4413 if (pvd->vdev_ops == &vdev_replacing_ops &&
4406 4414 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4407 4415 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4408 4416 } else if (pvd->vdev_ops == &vdev_spare_ops &&
4409 4417 newvd->vdev_isspare != oldvd->vdev_isspare) {
4410 4418 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4411 4419 }
4412 4420
4413 4421 if (newvd->vdev_isspare)
4414 4422 pvops = &vdev_spare_ops;
4415 4423 else
4416 4424 pvops = &vdev_replacing_ops;
4417 4425 }
4418 4426
4419 4427 /*
4420 4428 * Make sure the new device is big enough.
4421 4429 */
4422 4430 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4423 4431 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4424 4432
4425 4433 /*
4426 4434 * The new device cannot have a higher alignment requirement
4427 4435 * than the top-level vdev.
4428 4436 */
4429 4437 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4430 4438 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4431 4439
4432 4440 /*
4433 4441 * If this is an in-place replacement, update oldvd's path and devid
4434 4442 * to make it distinguishable from newvd, and unopenable from now on.
4435 4443 */
4436 4444 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4437 4445 spa_strfree(oldvd->vdev_path);
4438 4446 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4439 4447 KM_SLEEP);
4440 4448 (void) sprintf(oldvd->vdev_path, "%s/%s",
4441 4449 newvd->vdev_path, "old");
4442 4450 if (oldvd->vdev_devid != NULL) {
4443 4451 spa_strfree(oldvd->vdev_devid);
4444 4452 oldvd->vdev_devid = NULL;
4445 4453 }
4446 4454 }
4447 4455
4448 4456 /* mark the device being resilvered */
4449 4457 newvd->vdev_resilver_txg = txg;
4450 4458
4451 4459 /*
4452 4460 * If the parent is not a mirror, or if we're replacing, insert the new
4453 4461 * mirror/replacing/spare vdev above oldvd.
4454 4462 */
4455 4463 if (pvd->vdev_ops != pvops)
4456 4464 pvd = vdev_add_parent(oldvd, pvops);
4457 4465
4458 4466 ASSERT(pvd->vdev_top->vdev_parent == rvd);
4459 4467 ASSERT(pvd->vdev_ops == pvops);
4460 4468 ASSERT(oldvd->vdev_parent == pvd);
4461 4469
4462 4470 /*
4463 4471 * Extract the new device from its root and add it to pvd.
4464 4472 */
4465 4473 vdev_remove_child(newrootvd, newvd);
4466 4474 newvd->vdev_id = pvd->vdev_children;
4467 4475 newvd->vdev_crtxg = oldvd->vdev_crtxg;
4468 4476 vdev_add_child(pvd, newvd);
4469 4477
4470 4478 tvd = newvd->vdev_top;
4471 4479 ASSERT(pvd->vdev_top == tvd);
4472 4480 ASSERT(tvd->vdev_parent == rvd);
4473 4481
4474 4482 vdev_config_dirty(tvd);
4475 4483
4476 4484 /*
4477 4485 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4478 4486 * for any dmu_sync-ed blocks. It will propagate upward when
4479 4487 * spa_vdev_exit() calls vdev_dtl_reassess().
4480 4488 */
4481 4489 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4482 4490
4483 4491 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4484 4492 dtl_max_txg - TXG_INITIAL);
4485 4493
4486 4494 if (newvd->vdev_isspare) {
4487 4495 spa_spare_activate(newvd);
4488 4496 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4489 4497 }
4490 4498
4491 4499 oldvdpath = spa_strdup(oldvd->vdev_path);
4492 4500 newvdpath = spa_strdup(newvd->vdev_path);
4493 4501 newvd_isspare = newvd->vdev_isspare;
4494 4502
4495 4503 /*
4496 4504 * Mark newvd's DTL dirty in this txg.
4497 4505 */
4498 4506 vdev_dirty(tvd, VDD_DTL, newvd, txg);
4499 4507
4500 4508 /*
4501 4509 * Schedule the resilver to restart in the future. We do this to
4502 4510 * ensure that dmu_sync-ed blocks have been stitched into the
4503 4511 * respective datasets.
4504 4512 */
4505 4513 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4506 4514
4507 4515 /*
4508 4516 * Commit the config
4509 4517 */
4510 4518 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4511 4519
4512 4520 spa_history_log_internal(spa, "vdev attach", NULL,
4513 4521 "%s vdev=%s %s vdev=%s",
4514 4522 replacing && newvd_isspare ? "spare in" :
4515 4523 replacing ? "replace" : "attach", newvdpath,
4516 4524 replacing ? "for" : "to", oldvdpath);
4517 4525
4518 4526 spa_strfree(oldvdpath);
4519 4527 spa_strfree(newvdpath);
4520 4528
4521 4529 if (spa->spa_bootfs)
4522 4530 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4523 4531
4524 4532 return (0);
4525 4533 }
4526 4534
4527 4535 /*
4528 4536 * Detach a device from a mirror or replacing vdev.
4529 4537 *
4530 4538 * If 'replace_done' is specified, only detach if the parent
4531 4539 * is a replacing vdev.
4532 4540 */
4533 4541 int
4534 4542 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4535 4543 {
4536 4544 uint64_t txg;
4537 4545 int error;
4538 4546 vdev_t *rvd = spa->spa_root_vdev;
4539 4547 vdev_t *vd, *pvd, *cvd, *tvd;
4540 4548 boolean_t unspare = B_FALSE;
4541 4549 uint64_t unspare_guid = 0;
4542 4550 char *vdpath;
4543 4551
4544 4552 ASSERT(spa_writeable(spa));
4545 4553
4546 4554 txg = spa_vdev_enter(spa);
4547 4555
4548 4556 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4549 4557
4550 4558 if (vd == NULL)
4551 4559 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4552 4560
4553 4561 if (!vd->vdev_ops->vdev_op_leaf)
4554 4562 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4555 4563
4556 4564 pvd = vd->vdev_parent;
4557 4565
4558 4566 /*
4559 4567 * If the parent/child relationship is not as expected, don't do it.
4560 4568 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4561 4569 * vdev that's replacing B with C. The user's intent in replacing
4562 4570 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4563 4571 * the replace by detaching C, the expected behavior is to end up
4564 4572 * M(A,B). But suppose that right after deciding to detach C,
4565 4573 * the replacement of B completes. We would have M(A,C), and then
4566 4574 * ask to detach C, which would leave us with just A -- not what
4567 4575 * the user wanted. To prevent this, we make sure that the
4568 4576 * parent/child relationship hasn't changed -- in this example,
4569 4577 * that C's parent is still the replacing vdev R.
4570 4578 */
4571 4579 if (pvd->vdev_guid != pguid && pguid != 0)
4572 4580 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4573 4581
4574 4582 /*
4575 4583 * Only 'replacing' or 'spare' vdevs can be replaced.
4576 4584 */
4577 4585 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4578 4586 pvd->vdev_ops != &vdev_spare_ops)
4579 4587 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4580 4588
4581 4589 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4582 4590 spa_version(spa) >= SPA_VERSION_SPARES);
4583 4591
4584 4592 /*
4585 4593 * Only mirror, replacing, and spare vdevs support detach.
4586 4594 */
4587 4595 if (pvd->vdev_ops != &vdev_replacing_ops &&
4588 4596 pvd->vdev_ops != &vdev_mirror_ops &&
4589 4597 pvd->vdev_ops != &vdev_spare_ops)
4590 4598 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4591 4599
4592 4600 /*
4593 4601 * If this device has the only valid copy of some data,
4594 4602 * we cannot safely detach it.
4595 4603 */
4596 4604 if (vdev_dtl_required(vd))
4597 4605 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4598 4606
4599 4607 ASSERT(pvd->vdev_children >= 2);
4600 4608
4601 4609 /*
4602 4610 * If we are detaching the second disk from a replacing vdev, then
4603 4611 * check to see if we changed the original vdev's path to have "/old"
4604 4612 * at the end in spa_vdev_attach(). If so, undo that change now.
4605 4613 */
4606 4614 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4607 4615 vd->vdev_path != NULL) {
4608 4616 size_t len = strlen(vd->vdev_path);
4609 4617
4610 4618 for (int c = 0; c < pvd->vdev_children; c++) {
4611 4619 cvd = pvd->vdev_child[c];
4612 4620
4613 4621 if (cvd == vd || cvd->vdev_path == NULL)
4614 4622 continue;
4615 4623
4616 4624 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4617 4625 strcmp(cvd->vdev_path + len, "/old") == 0) {
4618 4626 spa_strfree(cvd->vdev_path);
4619 4627 cvd->vdev_path = spa_strdup(vd->vdev_path);
4620 4628 break;
4621 4629 }
4622 4630 }
4623 4631 }
4624 4632
4625 4633 /*
4626 4634 * If we are detaching the original disk from a spare, then it implies
4627 4635 * that the spare should become a real disk, and be removed from the
4628 4636 * active spare list for the pool.
4629 4637 */
4630 4638 if (pvd->vdev_ops == &vdev_spare_ops &&
4631 4639 vd->vdev_id == 0 &&
4632 4640 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4633 4641 unspare = B_TRUE;
4634 4642
4635 4643 /*
4636 4644 * Erase the disk labels so the disk can be used for other things.
4637 4645 * This must be done after all other error cases are handled,
4638 4646 * but before we disembowel vd (so we can still do I/O to it).
4639 4647 * But if we can't do it, don't treat the error as fatal --
4640 4648 * it may be that the unwritability of the disk is the reason
4641 4649 * it's being detached!
4642 4650 */
4643 4651 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4644 4652
4645 4653 /*
4646 4654 * Remove vd from its parent and compact the parent's children.
4647 4655 */
4648 4656 vdev_remove_child(pvd, vd);
4649 4657 vdev_compact_children(pvd);
4650 4658
4651 4659 /*
4652 4660 * Remember one of the remaining children so we can get tvd below.
4653 4661 */
4654 4662 cvd = pvd->vdev_child[pvd->vdev_children - 1];
4655 4663
4656 4664 /*
4657 4665 * If we need to remove the remaining child from the list of hot spares,
4658 4666 * do it now, marking the vdev as no longer a spare in the process.
4659 4667 * We must do this before vdev_remove_parent(), because that can
4660 4668 * change the GUID if it creates a new toplevel GUID. For a similar
4661 4669 * reason, we must remove the spare now, in the same txg as the detach;
4662 4670 * otherwise someone could attach a new sibling, change the GUID, and
4663 4671 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4664 4672 */
4665 4673 if (unspare) {
4666 4674 ASSERT(cvd->vdev_isspare);
4667 4675 spa_spare_remove(cvd);
4668 4676 unspare_guid = cvd->vdev_guid;
4669 4677 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4670 4678 cvd->vdev_unspare = B_TRUE;
4671 4679 }
4672 4680
4673 4681 /*
4674 4682 * If the parent mirror/replacing vdev only has one child,
4675 4683 * the parent is no longer needed. Remove it from the tree.
4676 4684 */
4677 4685 if (pvd->vdev_children == 1) {
4678 4686 if (pvd->vdev_ops == &vdev_spare_ops)
4679 4687 cvd->vdev_unspare = B_FALSE;
4680 4688 vdev_remove_parent(cvd);
4681 4689 }
4682 4690
4683 4691
4684 4692 /*
4685 4693 * We don't set tvd until now because the parent we just removed
4686 4694 * may have been the previous top-level vdev.
4687 4695 */
4688 4696 tvd = cvd->vdev_top;
4689 4697 ASSERT(tvd->vdev_parent == rvd);
4690 4698
4691 4699 /*
4692 4700 * Reevaluate the parent vdev state.
4693 4701 */
4694 4702 vdev_propagate_state(cvd);
4695 4703
4696 4704 /*
4697 4705 * If the 'autoexpand' property is set on the pool then automatically
4698 4706 * try to expand the size of the pool. For example if the device we
4699 4707 * just detached was smaller than the others, it may be possible to
4700 4708 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4701 4709 * first so that we can obtain the updated sizes of the leaf vdevs.
4702 4710 */
4703 4711 if (spa->spa_autoexpand) {
4704 4712 vdev_reopen(tvd);
4705 4713 vdev_expand(tvd, txg);
4706 4714 }
4707 4715
4708 4716 vdev_config_dirty(tvd);
4709 4717
4710 4718 /*
4711 4719 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
4712 4720 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4713 4721 * But first make sure we're not on any *other* txg's DTL list, to
4714 4722 * prevent vd from being accessed after it's freed.
4715 4723 */
4716 4724 vdpath = spa_strdup(vd->vdev_path);
4717 4725 for (int t = 0; t < TXG_SIZE; t++)
4718 4726 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4719 4727 vd->vdev_detached = B_TRUE;
4720 4728 vdev_dirty(tvd, VDD_DTL, vd, txg);
4721 4729
4722 4730 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4723 4731
4724 4732 /* hang on to the spa before we release the lock */
4725 4733 spa_open_ref(spa, FTAG);
4726 4734
4727 4735 error = spa_vdev_exit(spa, vd, txg, 0);
4728 4736
4729 4737 spa_history_log_internal(spa, "detach", NULL,
4730 4738 "vdev=%s", vdpath);
4731 4739 spa_strfree(vdpath);
4732 4740
4733 4741 /*
4734 4742 * If this was the removal of the original device in a hot spare vdev,
4735 4743 * then we want to go through and remove the device from the hot spare
4736 4744 * list of every other pool.
4737 4745 */
4738 4746 if (unspare) {
4739 4747 spa_t *altspa = NULL;
4740 4748
4741 4749 mutex_enter(&spa_namespace_lock);
4742 4750 while ((altspa = spa_next(altspa)) != NULL) {
4743 4751 if (altspa->spa_state != POOL_STATE_ACTIVE ||
4744 4752 altspa == spa)
4745 4753 continue;
4746 4754
4747 4755 spa_open_ref(altspa, FTAG);
4748 4756 mutex_exit(&spa_namespace_lock);
4749 4757 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4750 4758 mutex_enter(&spa_namespace_lock);
4751 4759 spa_close(altspa, FTAG);
4752 4760 }
4753 4761 mutex_exit(&spa_namespace_lock);
4754 4762
4755 4763 /* search the rest of the vdevs for spares to remove */
4756 4764 spa_vdev_resilver_done(spa);
4757 4765 }
4758 4766
4759 4767 /* all done with the spa; OK to release */
4760 4768 mutex_enter(&spa_namespace_lock);
4761 4769 spa_close(spa, FTAG);
4762 4770 mutex_exit(&spa_namespace_lock);
4763 4771
4764 4772 return (error);
4765 4773 }
4766 4774
4767 4775 /*
4768 4776 * Split a set of devices from their mirrors, and create a new pool from them.
4769 4777 */
4770 4778 int
4771 4779 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4772 4780 nvlist_t *props, boolean_t exp)
4773 4781 {
4774 4782 int error = 0;
4775 4783 uint64_t txg, *glist;
4776 4784 spa_t *newspa;
4777 4785 uint_t c, children, lastlog;
4778 4786 nvlist_t **child, *nvl, *tmp;
4779 4787 dmu_tx_t *tx;
4780 4788 char *altroot = NULL;
4781 4789 vdev_t *rvd, **vml = NULL; /* vdev modify list */
4782 4790 boolean_t activate_slog;
4783 4791
4784 4792 ASSERT(spa_writeable(spa));
4785 4793
4786 4794 txg = spa_vdev_enter(spa);
4787 4795
4788 4796 /* clear the log and flush everything up to now */
4789 4797 activate_slog = spa_passivate_log(spa);
4790 4798 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4791 4799 error = spa_offline_log(spa);
4792 4800 txg = spa_vdev_config_enter(spa);
4793 4801
4794 4802 if (activate_slog)
4795 4803 spa_activate_log(spa);
4796 4804
4797 4805 if (error != 0)
4798 4806 return (spa_vdev_exit(spa, NULL, txg, error));
4799 4807
4800 4808 /* check new spa name before going any further */
4801 4809 if (spa_lookup(newname) != NULL)
4802 4810 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4803 4811
4804 4812 /*
4805 4813 * scan through all the children to ensure they're all mirrors
4806 4814 */
4807 4815 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4808 4816 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4809 4817 &children) != 0)
4810 4818 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4811 4819
4812 4820 /* first, check to ensure we've got the right child count */
4813 4821 rvd = spa->spa_root_vdev;
4814 4822 lastlog = 0;
4815 4823 for (c = 0; c < rvd->vdev_children; c++) {
4816 4824 vdev_t *vd = rvd->vdev_child[c];
4817 4825
4818 4826 /* don't count the holes & logs as children */
4819 4827 if (vd->vdev_islog || vd->vdev_ishole) {
4820 4828 if (lastlog == 0)
4821 4829 lastlog = c;
4822 4830 continue;
4823 4831 }
4824 4832
4825 4833 lastlog = 0;
4826 4834 }
4827 4835 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4828 4836 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4829 4837
4830 4838 /* next, ensure no spare or cache devices are part of the split */
4831 4839 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4832 4840 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4833 4841 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4834 4842
4835 4843 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4836 4844 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4837 4845
4838 4846 /* then, loop over each vdev and validate it */
4839 4847 for (c = 0; c < children; c++) {
4840 4848 uint64_t is_hole = 0;
4841 4849
4842 4850 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4843 4851 &is_hole);
4844 4852
4845 4853 if (is_hole != 0) {
4846 4854 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4847 4855 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4848 4856 continue;
4849 4857 } else {
4850 4858 error = SET_ERROR(EINVAL);
4851 4859 break;
4852 4860 }
4853 4861 }
4854 4862
4855 4863 /* which disk is going to be split? */
4856 4864 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4857 4865 &glist[c]) != 0) {
4858 4866 error = SET_ERROR(EINVAL);
4859 4867 break;
4860 4868 }
4861 4869
4862 4870 /* look it up in the spa */
4863 4871 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4864 4872 if (vml[c] == NULL) {
4865 4873 error = SET_ERROR(ENODEV);
4866 4874 break;
4867 4875 }
4868 4876
4869 4877 /* make sure there's nothing stopping the split */
4870 4878 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4871 4879 vml[c]->vdev_islog ||
4872 4880 vml[c]->vdev_ishole ||
4873 4881 vml[c]->vdev_isspare ||
4874 4882 vml[c]->vdev_isl2cache ||
4875 4883 !vdev_writeable(vml[c]) ||
4876 4884 vml[c]->vdev_children != 0 ||
4877 4885 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4878 4886 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4879 4887 error = SET_ERROR(EINVAL);
4880 4888 break;
4881 4889 }
4882 4890
4883 4891 if (vdev_dtl_required(vml[c])) {
4884 4892 error = SET_ERROR(EBUSY);
4885 4893 break;
4886 4894 }
4887 4895
4888 4896 /* we need certain info from the top level */
4889 4897 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4890 4898 vml[c]->vdev_top->vdev_ms_array) == 0);
4891 4899 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4892 4900 vml[c]->vdev_top->vdev_ms_shift) == 0);
4893 4901 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4894 4902 vml[c]->vdev_top->vdev_asize) == 0);
4895 4903 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4896 4904 vml[c]->vdev_top->vdev_ashift) == 0);
4897 4905 }
4898 4906
4899 4907 if (error != 0) {
4900 4908 kmem_free(vml, children * sizeof (vdev_t *));
4901 4909 kmem_free(glist, children * sizeof (uint64_t));
4902 4910 return (spa_vdev_exit(spa, NULL, txg, error));
4903 4911 }
4904 4912
4905 4913 /* stop writers from using the disks */
4906 4914 for (c = 0; c < children; c++) {
4907 4915 if (vml[c] != NULL)
4908 4916 vml[c]->vdev_offline = B_TRUE;
4909 4917 }
4910 4918 vdev_reopen(spa->spa_root_vdev);
4911 4919
4912 4920 /*
4913 4921 * Temporarily record the splitting vdevs in the spa config. This
4914 4922 * will disappear once the config is regenerated.
4915 4923 */
4916 4924 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4917 4925 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4918 4926 glist, children) == 0);
4919 4927 kmem_free(glist, children * sizeof (uint64_t));
4920 4928
4921 4929 mutex_enter(&spa->spa_props_lock);
4922 4930 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4923 4931 nvl) == 0);
4924 4932 mutex_exit(&spa->spa_props_lock);
4925 4933 spa->spa_config_splitting = nvl;
4926 4934 vdev_config_dirty(spa->spa_root_vdev);
4927 4935
4928 4936 /* configure and create the new pool */
4929 4937 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4930 4938 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4931 4939 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4932 4940 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4933 4941 spa_version(spa)) == 0);
4934 4942 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4935 4943 spa->spa_config_txg) == 0);
4936 4944 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4937 4945 spa_generate_guid(NULL)) == 0);
4938 4946 (void) nvlist_lookup_string(props,
4939 4947 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4940 4948
4941 4949 /* add the new pool to the namespace */
4942 4950 newspa = spa_add(newname, config, altroot);
4943 4951 newspa->spa_config_txg = spa->spa_config_txg;
4944 4952 spa_set_log_state(newspa, SPA_LOG_CLEAR);
4945 4953
4946 4954 /* release the spa config lock, retaining the namespace lock */
4947 4955 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4948 4956
4949 4957 if (zio_injection_enabled)
4950 4958 zio_handle_panic_injection(spa, FTAG, 1);
4951 4959
4952 4960 spa_activate(newspa, spa_mode_global);
4953 4961 spa_async_suspend(newspa);
4954 4962
4955 4963 /* create the new pool from the disks of the original pool */
4956 4964 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
4957 4965 if (error)
4958 4966 goto out;
4959 4967
4960 4968 /* if that worked, generate a real config for the new pool */
4961 4969 if (newspa->spa_root_vdev != NULL) {
4962 4970 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
4963 4971 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4964 4972 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
4965 4973 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
4966 4974 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
4967 4975 B_TRUE));
4968 4976 }
4969 4977
4970 4978 /* set the props */
4971 4979 if (props != NULL) {
4972 4980 spa_configfile_set(newspa, props, B_FALSE);
4973 4981 error = spa_prop_set(newspa, props);
4974 4982 if (error)
4975 4983 goto out;
4976 4984 }
4977 4985
4978 4986 /* flush everything */
4979 4987 txg = spa_vdev_config_enter(newspa);
4980 4988 vdev_config_dirty(newspa->spa_root_vdev);
4981 4989 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
4982 4990
4983 4991 if (zio_injection_enabled)
4984 4992 zio_handle_panic_injection(spa, FTAG, 2);
4985 4993
4986 4994 spa_async_resume(newspa);
4987 4995
4988 4996 /* finally, update the original pool's config */
4989 4997 txg = spa_vdev_config_enter(spa);
4990 4998 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
4991 4999 error = dmu_tx_assign(tx, TXG_WAIT);
4992 5000 if (error != 0)
4993 5001 dmu_tx_abort(tx);
4994 5002 for (c = 0; c < children; c++) {
4995 5003 if (vml[c] != NULL) {
4996 5004 vdev_split(vml[c]);
4997 5005 if (error == 0)
4998 5006 spa_history_log_internal(spa, "detach", tx,
4999 5007 "vdev=%s", vml[c]->vdev_path);
5000 5008 vdev_free(vml[c]);
5001 5009 }
5002 5010 }
5003 5011 vdev_config_dirty(spa->spa_root_vdev);
5004 5012 spa->spa_config_splitting = NULL;
5005 5013 nvlist_free(nvl);
5006 5014 if (error == 0)
5007 5015 dmu_tx_commit(tx);
5008 5016 (void) spa_vdev_exit(spa, NULL, txg, 0);
5009 5017
5010 5018 if (zio_injection_enabled)
5011 5019 zio_handle_panic_injection(spa, FTAG, 3);
5012 5020
5013 5021 /* split is complete; log a history record */
5014 5022 spa_history_log_internal(newspa, "split", NULL,
5015 5023 "from pool %s", spa_name(spa));
5016 5024
5017 5025 kmem_free(vml, children * sizeof (vdev_t *));
5018 5026
5019 5027 /* if we're not going to mount the filesystems in userland, export */
5020 5028 if (exp)
5021 5029 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5022 5030 B_FALSE, B_FALSE);
5023 5031
5024 5032 return (error);
5025 5033
5026 5034 out:
5027 5035 spa_unload(newspa);
5028 5036 spa_deactivate(newspa);
5029 5037 spa_remove(newspa);
5030 5038
5031 5039 txg = spa_vdev_config_enter(spa);
5032 5040
5033 5041 /* re-online all offlined disks */
5034 5042 for (c = 0; c < children; c++) {
5035 5043 if (vml[c] != NULL)
5036 5044 vml[c]->vdev_offline = B_FALSE;
5037 5045 }
5038 5046 vdev_reopen(spa->spa_root_vdev);
5039 5047
5040 5048 nvlist_free(spa->spa_config_splitting);
5041 5049 spa->spa_config_splitting = NULL;
5042 5050 (void) spa_vdev_exit(spa, NULL, txg, error);
5043 5051
5044 5052 kmem_free(vml, children * sizeof (vdev_t *));
5045 5053 return (error);
5046 5054 }
5047 5055
5048 5056 static nvlist_t *
5049 5057 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5050 5058 {
5051 5059 for (int i = 0; i < count; i++) {
5052 5060 uint64_t guid;
5053 5061
5054 5062 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5055 5063 &guid) == 0);
5056 5064
5057 5065 if (guid == target_guid)
5058 5066 return (nvpp[i]);
5059 5067 }
5060 5068
5061 5069 return (NULL);
5062 5070 }
5063 5071
5064 5072 static void
5065 5073 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5066 5074 nvlist_t *dev_to_remove)
5067 5075 {
5068 5076 nvlist_t **newdev = NULL;
5069 5077
5070 5078 if (count > 1)
5071 5079 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
5072 5080
5073 5081 for (int i = 0, j = 0; i < count; i++) {
5074 5082 if (dev[i] == dev_to_remove)
5075 5083 continue;
5076 5084 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
5077 5085 }
5078 5086
5079 5087 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5080 5088 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5081 5089
5082 5090 for (int i = 0; i < count - 1; i++)
5083 5091 nvlist_free(newdev[i]);
5084 5092
5085 5093 if (count > 1)
5086 5094 kmem_free(newdev, (count - 1) * sizeof (void *));
5087 5095 }
5088 5096
5089 5097 /*
5090 5098 * Evacuate the device.
5091 5099 */
5092 5100 static int
5093 5101 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5094 5102 {
5095 5103 uint64_t txg;
5096 5104 int error = 0;
5097 5105
5098 5106 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5099 5107 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5100 5108 ASSERT(vd == vd->vdev_top);
5101 5109
5102 5110 /*
5103 5111 * Evacuate the device. We don't hold the config lock as writer
5104 5112 * since we need to do I/O but we do keep the
5105 5113 * spa_namespace_lock held. Once this completes the device
5106 5114 * should no longer have any blocks allocated on it.
5107 5115 */
5108 5116 if (vd->vdev_islog) {
5109 5117 if (vd->vdev_stat.vs_alloc != 0)
5110 5118 error = spa_offline_log(spa);
5111 5119 } else {
5112 5120 error = SET_ERROR(ENOTSUP);
5113 5121 }
5114 5122
5115 5123 if (error)
5116 5124 return (error);
5117 5125
5118 5126 /*
5119 5127 * The evacuation succeeded. Remove any remaining MOS metadata
5120 5128 * associated with this vdev, and wait for these changes to sync.
5121 5129 */
5122 5130 ASSERT0(vd->vdev_stat.vs_alloc);
5123 5131 txg = spa_vdev_config_enter(spa);
5124 5132 vd->vdev_removing = B_TRUE;
5125 5133 vdev_dirty_leaves(vd, VDD_DTL, txg);
5126 5134 vdev_config_dirty(vd);
5127 5135 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5128 5136
5129 5137 return (0);
5130 5138 }
5131 5139
5132 5140 /*
5133 5141 * Complete the removal by cleaning up the namespace.
5134 5142 */
5135 5143 static void
5136 5144 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5137 5145 {
5138 5146 vdev_t *rvd = spa->spa_root_vdev;
5139 5147 uint64_t id = vd->vdev_id;
5140 5148 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5141 5149
5142 5150 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5143 5151 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5144 5152 ASSERT(vd == vd->vdev_top);
5145 5153
5146 5154 /*
5147 5155 * Only remove any devices which are empty.
5148 5156 */
5149 5157 if (vd->vdev_stat.vs_alloc != 0)
5150 5158 return;
5151 5159
5152 5160 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5153 5161
5154 5162 if (list_link_active(&vd->vdev_state_dirty_node))
5155 5163 vdev_state_clean(vd);
5156 5164 if (list_link_active(&vd->vdev_config_dirty_node))
5157 5165 vdev_config_clean(vd);
5158 5166
5159 5167 vdev_free(vd);
5160 5168
5161 5169 if (last_vdev) {
5162 5170 vdev_compact_children(rvd);
5163 5171 } else {
5164 5172 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5165 5173 vdev_add_child(rvd, vd);
5166 5174 }
5167 5175 vdev_config_dirty(rvd);
5168 5176
5169 5177 /*
5170 5178 * Reassess the health of our root vdev.
5171 5179 */
5172 5180 vdev_reopen(rvd);
5173 5181 }
5174 5182
5175 5183 /*
5176 5184 * Remove a device from the pool -
5177 5185 *
5178 5186 * Removing a device from the vdev namespace requires several steps
5179 5187 * and can take a significant amount of time. As a result we use
5180 5188 * the spa_vdev_config_[enter/exit] functions which allow us to
5181 5189 * grab and release the spa_config_lock while still holding the namespace
5182 5190 * lock. During each step the configuration is synced out.
5183 5191 *
5184 5192 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5185 5193 * devices.
5186 5194 */
5187 5195 int
5188 5196 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5189 5197 {
5190 5198 vdev_t *vd;
5191 5199 metaslab_group_t *mg;
5192 5200 nvlist_t **spares, **l2cache, *nv;
5193 5201 uint64_t txg = 0;
5194 5202 uint_t nspares, nl2cache;
5195 5203 int error = 0;
5196 5204 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5197 5205
5198 5206 ASSERT(spa_writeable(spa));
5199 5207
5200 5208 if (!locked)
5201 5209 txg = spa_vdev_enter(spa);
5202 5210
5203 5211 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5204 5212
5205 5213 if (spa->spa_spares.sav_vdevs != NULL &&
5206 5214 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5207 5215 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5208 5216 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5209 5217 /*
5210 5218 * Only remove the hot spare if it's not currently in use
5211 5219 * in this pool.
5212 5220 */
5213 5221 if (vd == NULL || unspare) {
5214 5222 spa_vdev_remove_aux(spa->spa_spares.sav_config,
5215 5223 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5216 5224 spa_load_spares(spa);
5217 5225 spa->spa_spares.sav_sync = B_TRUE;
5218 5226 } else {
5219 5227 error = SET_ERROR(EBUSY);
5220 5228 }
5221 5229 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
5222 5230 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5223 5231 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5224 5232 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5225 5233 /*
5226 5234 * Cache devices can always be removed.
5227 5235 */
5228 5236 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5229 5237 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5230 5238 spa_load_l2cache(spa);
5231 5239 spa->spa_l2cache.sav_sync = B_TRUE;
5232 5240 } else if (vd != NULL && vd->vdev_islog) {
5233 5241 ASSERT(!locked);
5234 5242 ASSERT(vd == vd->vdev_top);
5235 5243
5236 5244 /*
5237 5245 * XXX - Once we have bp-rewrite this should
5238 5246 * become the common case.
5239 5247 */
5240 5248
5241 5249 mg = vd->vdev_mg;
5242 5250
5243 5251 /*
5244 5252 * Stop allocating from this vdev.
5245 5253 */
5246 5254 metaslab_group_passivate(mg);
5247 5255
5248 5256 /*
5249 5257 * Wait for the youngest allocations and frees to sync,
5250 5258 * and then wait for the deferral of those frees to finish.
5251 5259 */
5252 5260 spa_vdev_config_exit(spa, NULL,
5253 5261 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5254 5262
5255 5263 /*
5256 5264 * Attempt to evacuate the vdev.
5257 5265 */
5258 5266 error = spa_vdev_remove_evacuate(spa, vd);
5259 5267
5260 5268 txg = spa_vdev_config_enter(spa);
5261 5269
5262 5270 /*
5263 5271 * If we couldn't evacuate the vdev, unwind.
5264 5272 */
5265 5273 if (error) {
5266 5274 metaslab_group_activate(mg);
5267 5275 return (spa_vdev_exit(spa, NULL, txg, error));
5268 5276 }
5269 5277
5270 5278 /*
5271 5279 * Clean up the vdev namespace.
5272 5280 */
5273 5281 spa_vdev_remove_from_namespace(spa, vd);
5274 5282
5275 5283 } else if (vd != NULL) {
5276 5284 /*
5277 5285 * Normal vdevs cannot be removed (yet).
5278 5286 */
5279 5287 error = SET_ERROR(ENOTSUP);
5280 5288 } else {
5281 5289 /*
5282 5290 * There is no vdev of any kind with the specified guid.
5283 5291 */
5284 5292 error = SET_ERROR(ENOENT);
5285 5293 }
5286 5294
5287 5295 if (!locked)
5288 5296 return (spa_vdev_exit(spa, NULL, txg, error));
5289 5297
5290 5298 return (error);
5291 5299 }
5292 5300
5293 5301 /*
5294 5302 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5295 5303 * currently spared, so we can detach it.
5296 5304 */
5297 5305 static vdev_t *
5298 5306 spa_vdev_resilver_done_hunt(vdev_t *vd)
5299 5307 {
5300 5308 vdev_t *newvd, *oldvd;
5301 5309
5302 5310 for (int c = 0; c < vd->vdev_children; c++) {
5303 5311 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5304 5312 if (oldvd != NULL)
5305 5313 return (oldvd);
5306 5314 }
5307 5315
5308 5316 /*
5309 5317 * Check for a completed replacement. We always consider the first
5310 5318 * vdev in the list to be the oldest vdev, and the last one to be
5311 5319 * the newest (see spa_vdev_attach() for how that works). In
5312 5320 * the case where the newest vdev is faulted, we will not automatically
5313 5321 * remove it after a resilver completes. This is OK as it will require
5314 5322 * user intervention to determine which disk the admin wishes to keep.
5315 5323 */
5316 5324 if (vd->vdev_ops == &vdev_replacing_ops) {
5317 5325 ASSERT(vd->vdev_children > 1);
5318 5326
5319 5327 newvd = vd->vdev_child[vd->vdev_children - 1];
5320 5328 oldvd = vd->vdev_child[0];
5321 5329
5322 5330 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5323 5331 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5324 5332 !vdev_dtl_required(oldvd))
5325 5333 return (oldvd);
5326 5334 }
5327 5335
5328 5336 /*
5329 5337 * Check for a completed resilver with the 'unspare' flag set.
5330 5338 */
5331 5339 if (vd->vdev_ops == &vdev_spare_ops) {
5332 5340 vdev_t *first = vd->vdev_child[0];
5333 5341 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5334 5342
5335 5343 if (last->vdev_unspare) {
5336 5344 oldvd = first;
5337 5345 newvd = last;
5338 5346 } else if (first->vdev_unspare) {
5339 5347 oldvd = last;
5340 5348 newvd = first;
5341 5349 } else {
5342 5350 oldvd = NULL;
5343 5351 }
5344 5352
5345 5353 if (oldvd != NULL &&
5346 5354 vdev_dtl_empty(newvd, DTL_MISSING) &&
5347 5355 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5348 5356 !vdev_dtl_required(oldvd))
5349 5357 return (oldvd);
5350 5358
5351 5359 /*
5352 5360 * If there are more than two spares attached to a disk,
5353 5361 * and those spares are not required, then we want to
5354 5362 * attempt to free them up now so that they can be used
5355 5363 * by other pools. Once we're back down to a single
5356 5364 * disk+spare, we stop removing them.
5357 5365 */
5358 5366 if (vd->vdev_children > 2) {
5359 5367 newvd = vd->vdev_child[1];
5360 5368
5361 5369 if (newvd->vdev_isspare && last->vdev_isspare &&
5362 5370 vdev_dtl_empty(last, DTL_MISSING) &&
5363 5371 vdev_dtl_empty(last, DTL_OUTAGE) &&
5364 5372 !vdev_dtl_required(newvd))
5365 5373 return (newvd);
5366 5374 }
5367 5375 }
5368 5376
5369 5377 return (NULL);
5370 5378 }
5371 5379
5372 5380 static void
5373 5381 spa_vdev_resilver_done(spa_t *spa)
5374 5382 {
5375 5383 vdev_t *vd, *pvd, *ppvd;
5376 5384 uint64_t guid, sguid, pguid, ppguid;
5377 5385
5378 5386 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5379 5387
5380 5388 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5381 5389 pvd = vd->vdev_parent;
5382 5390 ppvd = pvd->vdev_parent;
5383 5391 guid = vd->vdev_guid;
5384 5392 pguid = pvd->vdev_guid;
5385 5393 ppguid = ppvd->vdev_guid;
5386 5394 sguid = 0;
5387 5395 /*
5388 5396 * If we have just finished replacing a hot spared device, then
5389 5397 * we need to detach the parent's first child (the original hot
5390 5398 * spare) as well.
5391 5399 */
5392 5400 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5393 5401 ppvd->vdev_children == 2) {
5394 5402 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5395 5403 sguid = ppvd->vdev_child[1]->vdev_guid;
5396 5404 }
5397 5405 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd));
5398 5406
5399 5407 spa_config_exit(spa, SCL_ALL, FTAG);
5400 5408 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5401 5409 return;
5402 5410 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5403 5411 return;
5404 5412 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5405 5413 }
5406 5414
5407 5415 spa_config_exit(spa, SCL_ALL, FTAG);
5408 5416 }
5409 5417
5410 5418 /*
5411 5419 * Update the stored path or FRU for this vdev.
5412 5420 */
5413 5421 int
5414 5422 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5415 5423 boolean_t ispath)
5416 5424 {
5417 5425 vdev_t *vd;
5418 5426 boolean_t sync = B_FALSE;
5419 5427
5420 5428 ASSERT(spa_writeable(spa));
5421 5429
5422 5430 spa_vdev_state_enter(spa, SCL_ALL);
5423 5431
5424 5432 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5425 5433 return (spa_vdev_state_exit(spa, NULL, ENOENT));
5426 5434
5427 5435 if (!vd->vdev_ops->vdev_op_leaf)
5428 5436 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5429 5437
5430 5438 if (ispath) {
5431 5439 if (strcmp(value, vd->vdev_path) != 0) {
5432 5440 spa_strfree(vd->vdev_path);
5433 5441 vd->vdev_path = spa_strdup(value);
5434 5442 sync = B_TRUE;
5435 5443 }
5436 5444 } else {
5437 5445 if (vd->vdev_fru == NULL) {
5438 5446 vd->vdev_fru = spa_strdup(value);
5439 5447 sync = B_TRUE;
5440 5448 } else if (strcmp(value, vd->vdev_fru) != 0) {
5441 5449 spa_strfree(vd->vdev_fru);
5442 5450 vd->vdev_fru = spa_strdup(value);
5443 5451 sync = B_TRUE;
5444 5452 }
5445 5453 }
5446 5454
5447 5455 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5448 5456 }
5449 5457
5450 5458 int
5451 5459 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5452 5460 {
5453 5461 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5454 5462 }
5455 5463
5456 5464 int
5457 5465 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5458 5466 {
5459 5467 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5460 5468 }
5461 5469
5462 5470 /*
5463 5471 * ==========================================================================
5464 5472 * SPA Scanning
5465 5473 * ==========================================================================
5466 5474 */
5467 5475
5468 5476 int
5469 5477 spa_scan_stop(spa_t *spa)
5470 5478 {
5471 5479 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5472 5480 if (dsl_scan_resilvering(spa->spa_dsl_pool))
5473 5481 return (SET_ERROR(EBUSY));
5474 5482 return (dsl_scan_cancel(spa->spa_dsl_pool));
5475 5483 }
5476 5484
5477 5485 int
5478 5486 spa_scan(spa_t *spa, pool_scan_func_t func)
5479 5487 {
5480 5488 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5481 5489
5482 5490 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5483 5491 return (SET_ERROR(ENOTSUP));
5484 5492
5485 5493 /*
5486 5494 * If a resilver was requested, but there is no DTL on a
5487 5495 * writeable leaf device, we have nothing to do.
5488 5496 */
5489 5497 if (func == POOL_SCAN_RESILVER &&
5490 5498 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5491 5499 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5492 5500 return (0);
5493 5501 }
5494 5502
5495 5503 return (dsl_scan(spa->spa_dsl_pool, func));
5496 5504 }
5497 5505
5498 5506 /*
5499 5507 * ==========================================================================
5500 5508 * SPA async task processing
5501 5509 * ==========================================================================
5502 5510 */
5503 5511
5504 5512 static void
5505 5513 spa_async_remove(spa_t *spa, vdev_t *vd)
5506 5514 {
5507 5515 if (vd->vdev_remove_wanted) {
5508 5516 vd->vdev_remove_wanted = B_FALSE;
5509 5517 vd->vdev_delayed_close = B_FALSE;
5510 5518 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5511 5519
5512 5520 /*
5513 5521 * We want to clear the stats, but we don't want to do a full
5514 5522 * vdev_clear() as that will cause us to throw away
5515 5523 * degraded/faulted state as well as attempt to reopen the
5516 5524 * device, all of which is a waste.
5517 5525 */
5518 5526 vd->vdev_stat.vs_read_errors = 0;
5519 5527 vd->vdev_stat.vs_write_errors = 0;
5520 5528 vd->vdev_stat.vs_checksum_errors = 0;
5521 5529
5522 5530 vdev_state_dirty(vd->vdev_top);
5523 5531 }
5524 5532
5525 5533 for (int c = 0; c < vd->vdev_children; c++)
5526 5534 spa_async_remove(spa, vd->vdev_child[c]);
5527 5535 }
5528 5536
5529 5537 static void
5530 5538 spa_async_probe(spa_t *spa, vdev_t *vd)
5531 5539 {
5532 5540 if (vd->vdev_probe_wanted) {
5533 5541 vd->vdev_probe_wanted = B_FALSE;
5534 5542 vdev_reopen(vd); /* vdev_open() does the actual probe */
5535 5543 }
5536 5544
5537 5545 for (int c = 0; c < vd->vdev_children; c++)
5538 5546 spa_async_probe(spa, vd->vdev_child[c]);
5539 5547 }
5540 5548
5541 5549 static void
5542 5550 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5543 5551 {
5544 5552 sysevent_id_t eid;
5545 5553 nvlist_t *attr;
5546 5554 char *physpath;
5547 5555
5548 5556 if (!spa->spa_autoexpand)
5549 5557 return;
5550 5558
5551 5559 for (int c = 0; c < vd->vdev_children; c++) {
5552 5560 vdev_t *cvd = vd->vdev_child[c];
5553 5561 spa_async_autoexpand(spa, cvd);
5554 5562 }
5555 5563
5556 5564 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5557 5565 return;
5558 5566
5559 5567 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5560 5568 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5561 5569
5562 5570 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5563 5571 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5564 5572
5565 5573 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5566 5574 ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
5567 5575
5568 5576 nvlist_free(attr);
5569 5577 kmem_free(physpath, MAXPATHLEN);
5570 5578 }
5571 5579
5572 5580 static void
5573 5581 spa_async_thread(spa_t *spa)
5574 5582 {
5575 5583 int tasks;
5576 5584
5577 5585 ASSERT(spa->spa_sync_on);
5578 5586
5579 5587 mutex_enter(&spa->spa_async_lock);
5580 5588 tasks = spa->spa_async_tasks;
5581 5589 spa->spa_async_tasks = 0;
5582 5590 mutex_exit(&spa->spa_async_lock);
5583 5591
5584 5592 /*
5585 5593 * See if the config needs to be updated.
5586 5594 */
5587 5595 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5588 5596 uint64_t old_space, new_space;
5589 5597
5590 5598 mutex_enter(&spa_namespace_lock);
5591 5599 old_space = metaslab_class_get_space(spa_normal_class(spa));
5592 5600 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5593 5601 new_space = metaslab_class_get_space(spa_normal_class(spa));
5594 5602 mutex_exit(&spa_namespace_lock);
5595 5603
5596 5604 /*
5597 5605 * If the pool grew as a result of the config update,
5598 5606 * then log an internal history event.
5599 5607 */
5600 5608 if (new_space != old_space) {
5601 5609 spa_history_log_internal(spa, "vdev online", NULL,
5602 5610 "pool '%s' size: %llu(+%llu)",
5603 5611 spa_name(spa), new_space, new_space - old_space);
5604 5612 }
5605 5613 }
5606 5614
5607 5615 /*
5608 5616 * See if any devices need to be marked REMOVED.
5609 5617 */
5610 5618 if (tasks & SPA_ASYNC_REMOVE) {
5611 5619 spa_vdev_state_enter(spa, SCL_NONE);
5612 5620 spa_async_remove(spa, spa->spa_root_vdev);
5613 5621 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5614 5622 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5615 5623 for (int i = 0; i < spa->spa_spares.sav_count; i++)
5616 5624 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5617 5625 (void) spa_vdev_state_exit(spa, NULL, 0);
5618 5626 }
5619 5627
5620 5628 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5621 5629 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5622 5630 spa_async_autoexpand(spa, spa->spa_root_vdev);
5623 5631 spa_config_exit(spa, SCL_CONFIG, FTAG);
5624 5632 }
5625 5633
5626 5634 /*
5627 5635 * See if any devices need to be probed.
5628 5636 */
5629 5637 if (tasks & SPA_ASYNC_PROBE) {
5630 5638 spa_vdev_state_enter(spa, SCL_NONE);
5631 5639 spa_async_probe(spa, spa->spa_root_vdev);
5632 5640 (void) spa_vdev_state_exit(spa, NULL, 0);
5633 5641 }
5634 5642
5635 5643 /*
5636 5644 * If any devices are done replacing, detach them.
5637 5645 */
|
↓ open down ↓ |
2929 lines elided |
↑ open up ↑ |
5638 5646 if (tasks & SPA_ASYNC_RESILVER_DONE)
5639 5647 spa_vdev_resilver_done(spa);
5640 5648
5641 5649 /*
5642 5650 * Kick off a resilver.
5643 5651 */
5644 5652 if (tasks & SPA_ASYNC_RESILVER)
5645 5653 dsl_resilver_restart(spa->spa_dsl_pool, 0);
5646 5654
5647 5655 /*
5656 + * Kick off L2 cache rebuilding.
5657 + */
5658 + if (tasks & SPA_ASYNC_L2CACHE_REBUILD)
5659 + l2arc_spa_rebuild_start(spa);
5660 +
5661 + /*
5648 5662 * Let the world know that we're done.
5649 5663 */
5650 5664 mutex_enter(&spa->spa_async_lock);
5651 5665 spa->spa_async_thread = NULL;
5652 5666 cv_broadcast(&spa->spa_async_cv);
5653 5667 mutex_exit(&spa->spa_async_lock);
5654 5668 thread_exit();
5655 5669 }
5656 5670
5657 5671 void
5658 5672 spa_async_suspend(spa_t *spa)
5659 5673 {
5660 5674 mutex_enter(&spa->spa_async_lock);
5661 5675 spa->spa_async_suspended++;
5662 5676 while (spa->spa_async_thread != NULL)
5663 5677 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5664 5678 mutex_exit(&spa->spa_async_lock);
5665 5679 }
5666 5680
5667 5681 void
5668 5682 spa_async_resume(spa_t *spa)
5669 5683 {
5670 5684 mutex_enter(&spa->spa_async_lock);
5671 5685 ASSERT(spa->spa_async_suspended != 0);
5672 5686 spa->spa_async_suspended--;
5673 5687 mutex_exit(&spa->spa_async_lock);
5674 5688 }
5675 5689
5676 5690 static boolean_t
5677 5691 spa_async_tasks_pending(spa_t *spa)
5678 5692 {
5679 5693 uint_t non_config_tasks;
5680 5694 uint_t config_task;
5681 5695 boolean_t config_task_suspended;
5682 5696
5683 5697 non_config_tasks = spa->spa_async_tasks & ~SPA_ASYNC_CONFIG_UPDATE;
5684 5698 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
5685 5699 if (spa->spa_ccw_fail_time == 0) {
5686 5700 config_task_suspended = B_FALSE;
5687 5701 } else {
5688 5702 config_task_suspended =
5689 5703 (gethrtime() - spa->spa_ccw_fail_time) <
5690 5704 (zfs_ccw_retry_interval * NANOSEC);
5691 5705 }
5692 5706
5693 5707 return (non_config_tasks || (config_task && !config_task_suspended));
5694 5708 }
5695 5709
5696 5710 static void
5697 5711 spa_async_dispatch(spa_t *spa)
5698 5712 {
5699 5713 mutex_enter(&spa->spa_async_lock);
5700 5714 if (spa_async_tasks_pending(spa) &&
5701 5715 !spa->spa_async_suspended &&
5702 5716 spa->spa_async_thread == NULL &&
5703 5717 rootdir != NULL)
5704 5718 spa->spa_async_thread = thread_create(NULL, 0,
5705 5719 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5706 5720 mutex_exit(&spa->spa_async_lock);
5707 5721 }
5708 5722
5709 5723 void
5710 5724 spa_async_request(spa_t *spa, int task)
5711 5725 {
5712 5726 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5713 5727 mutex_enter(&spa->spa_async_lock);
5714 5728 spa->spa_async_tasks |= task;
5715 5729 mutex_exit(&spa->spa_async_lock);
5716 5730 }
5717 5731
5718 5732 /*
5719 5733 * ==========================================================================
5720 5734 * SPA syncing routines
5721 5735 * ==========================================================================
5722 5736 */
5723 5737
5724 5738 static int
5725 5739 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5726 5740 {
5727 5741 bpobj_t *bpo = arg;
5728 5742 bpobj_enqueue(bpo, bp, tx);
5729 5743 return (0);
5730 5744 }
5731 5745
5732 5746 static int
5733 5747 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5734 5748 {
5735 5749 zio_t *zio = arg;
5736 5750
5737 5751 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5738 5752 zio->io_flags));
5739 5753 return (0);
5740 5754 }
5741 5755
5742 5756 /*
5743 5757 * Note: this simple function is not inlined to make it easier to dtrace the
5744 5758 * amount of time spent syncing frees.
5745 5759 */
5746 5760 static void
5747 5761 spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx)
5748 5762 {
5749 5763 zio_t *zio = zio_root(spa, NULL, NULL, 0);
5750 5764 bplist_iterate(bpl, spa_free_sync_cb, zio, tx);
5751 5765 VERIFY(zio_wait(zio) == 0);
5752 5766 }
5753 5767
5754 5768 /*
5755 5769 * Note: this simple function is not inlined to make it easier to dtrace the
5756 5770 * amount of time spent syncing deferred frees.
5757 5771 */
5758 5772 static void
5759 5773 spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx)
5760 5774 {
5761 5775 zio_t *zio = zio_root(spa, NULL, NULL, 0);
5762 5776 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj,
5763 5777 spa_free_sync_cb, zio, tx), ==, 0);
5764 5778 VERIFY0(zio_wait(zio));
5765 5779 }
5766 5780
5767 5781
5768 5782 static void
5769 5783 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5770 5784 {
5771 5785 char *packed = NULL;
5772 5786 size_t bufsize;
5773 5787 size_t nvsize = 0;
5774 5788 dmu_buf_t *db;
5775 5789
5776 5790 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5777 5791
5778 5792 /*
5779 5793 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5780 5794 * information. This avoids the dbuf_will_dirty() path and
5781 5795 * saves us a pre-read to get data we don't actually care about.
5782 5796 */
5783 5797 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
5784 5798 packed = kmem_alloc(bufsize, KM_SLEEP);
5785 5799
5786 5800 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5787 5801 KM_SLEEP) == 0);
5788 5802 bzero(packed + nvsize, bufsize - nvsize);
5789 5803
5790 5804 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5791 5805
5792 5806 kmem_free(packed, bufsize);
5793 5807
5794 5808 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5795 5809 dmu_buf_will_dirty(db, tx);
5796 5810 *(uint64_t *)db->db_data = nvsize;
5797 5811 dmu_buf_rele(db, FTAG);
5798 5812 }
5799 5813
5800 5814 static void
5801 5815 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5802 5816 const char *config, const char *entry)
5803 5817 {
5804 5818 nvlist_t *nvroot;
5805 5819 nvlist_t **list;
5806 5820 int i;
5807 5821
5808 5822 if (!sav->sav_sync)
5809 5823 return;
5810 5824
5811 5825 /*
5812 5826 * Update the MOS nvlist describing the list of available devices.
5813 5827 * spa_validate_aux() will have already made sure this nvlist is
5814 5828 * valid and the vdevs are labeled appropriately.
5815 5829 */
5816 5830 if (sav->sav_object == 0) {
5817 5831 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5818 5832 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5819 5833 sizeof (uint64_t), tx);
5820 5834 VERIFY(zap_update(spa->spa_meta_objset,
5821 5835 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5822 5836 &sav->sav_object, tx) == 0);
5823 5837 }
5824 5838
5825 5839 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5826 5840 if (sav->sav_count == 0) {
5827 5841 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5828 5842 } else {
5829 5843 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5830 5844 for (i = 0; i < sav->sav_count; i++)
5831 5845 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5832 5846 B_FALSE, VDEV_CONFIG_L2CACHE);
5833 5847 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5834 5848 sav->sav_count) == 0);
5835 5849 for (i = 0; i < sav->sav_count; i++)
5836 5850 nvlist_free(list[i]);
5837 5851 kmem_free(list, sav->sav_count * sizeof (void *));
5838 5852 }
5839 5853
5840 5854 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5841 5855 nvlist_free(nvroot);
5842 5856
5843 5857 sav->sav_sync = B_FALSE;
5844 5858 }
5845 5859
5846 5860 static void
5847 5861 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5848 5862 {
5849 5863 nvlist_t *config;
5850 5864
5851 5865 if (list_is_empty(&spa->spa_config_dirty_list))
5852 5866 return;
5853 5867
5854 5868 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5855 5869
5856 5870 config = spa_config_generate(spa, spa->spa_root_vdev,
5857 5871 dmu_tx_get_txg(tx), B_FALSE);
5858 5872
5859 5873 /*
5860 5874 * If we're upgrading the spa version then make sure that
5861 5875 * the config object gets updated with the correct version.
5862 5876 */
5863 5877 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
5864 5878 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5865 5879 spa->spa_uberblock.ub_version);
5866 5880
5867 5881 spa_config_exit(spa, SCL_STATE, FTAG);
5868 5882
5869 5883 if (spa->spa_config_syncing)
5870 5884 nvlist_free(spa->spa_config_syncing);
5871 5885 spa->spa_config_syncing = config;
5872 5886
5873 5887 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5874 5888 }
5875 5889
5876 5890 static void
5877 5891 spa_sync_version(void *arg, dmu_tx_t *tx)
5878 5892 {
5879 5893 uint64_t *versionp = arg;
5880 5894 uint64_t version = *versionp;
5881 5895 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
5882 5896
5883 5897 /*
5884 5898 * Setting the version is special cased when first creating the pool.
5885 5899 */
5886 5900 ASSERT(tx->tx_txg != TXG_INITIAL);
5887 5901
5888 5902 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
5889 5903 ASSERT(version >= spa_version(spa));
5890 5904
5891 5905 spa->spa_uberblock.ub_version = version;
5892 5906 vdev_config_dirty(spa->spa_root_vdev);
5893 5907 spa_history_log_internal(spa, "set", tx, "version=%lld", version);
5894 5908 }
5895 5909
5896 5910 /*
5897 5911 * Set zpool properties.
5898 5912 */
5899 5913 static void
5900 5914 spa_sync_props(void *arg, dmu_tx_t *tx)
5901 5915 {
5902 5916 nvlist_t *nvp = arg;
5903 5917 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
5904 5918 objset_t *mos = spa->spa_meta_objset;
5905 5919 nvpair_t *elem = NULL;
5906 5920
5907 5921 mutex_enter(&spa->spa_props_lock);
5908 5922
5909 5923 while ((elem = nvlist_next_nvpair(nvp, elem))) {
5910 5924 uint64_t intval;
5911 5925 char *strval, *fname;
5912 5926 zpool_prop_t prop;
5913 5927 const char *propname;
5914 5928 zprop_type_t proptype;
5915 5929 spa_feature_t fid;
5916 5930
5917 5931 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
5918 5932 case ZPROP_INVAL:
5919 5933 /*
5920 5934 * We checked this earlier in spa_prop_validate().
5921 5935 */
5922 5936 ASSERT(zpool_prop_feature(nvpair_name(elem)));
5923 5937
5924 5938 fname = strchr(nvpair_name(elem), '@') + 1;
5925 5939 VERIFY0(zfeature_lookup_name(fname, &fid));
5926 5940
5927 5941 spa_feature_enable(spa, fid, tx);
5928 5942 spa_history_log_internal(spa, "set", tx,
5929 5943 "%s=enabled", nvpair_name(elem));
5930 5944 break;
5931 5945
5932 5946 case ZPOOL_PROP_VERSION:
5933 5947 intval = fnvpair_value_uint64(elem);
5934 5948 /*
5935 5949 * The version is synced seperatly before other
5936 5950 * properties and should be correct by now.
5937 5951 */
5938 5952 ASSERT3U(spa_version(spa), >=, intval);
5939 5953 break;
5940 5954
5941 5955 case ZPOOL_PROP_ALTROOT:
5942 5956 /*
5943 5957 * 'altroot' is a non-persistent property. It should
5944 5958 * have been set temporarily at creation or import time.
5945 5959 */
5946 5960 ASSERT(spa->spa_root != NULL);
5947 5961 break;
5948 5962
5949 5963 case ZPOOL_PROP_READONLY:
5950 5964 case ZPOOL_PROP_CACHEFILE:
5951 5965 /*
5952 5966 * 'readonly' and 'cachefile' are also non-persisitent
5953 5967 * properties.
5954 5968 */
5955 5969 break;
5956 5970 case ZPOOL_PROP_COMMENT:
5957 5971 strval = fnvpair_value_string(elem);
5958 5972 if (spa->spa_comment != NULL)
5959 5973 spa_strfree(spa->spa_comment);
5960 5974 spa->spa_comment = spa_strdup(strval);
5961 5975 /*
5962 5976 * We need to dirty the configuration on all the vdevs
5963 5977 * so that their labels get updated. It's unnecessary
5964 5978 * to do this for pool creation since the vdev's
5965 5979 * configuratoin has already been dirtied.
5966 5980 */
5967 5981 if (tx->tx_txg != TXG_INITIAL)
5968 5982 vdev_config_dirty(spa->spa_root_vdev);
5969 5983 spa_history_log_internal(spa, "set", tx,
5970 5984 "%s=%s", nvpair_name(elem), strval);
5971 5985 break;
5972 5986 default:
5973 5987 /*
5974 5988 * Set pool property values in the poolprops mos object.
5975 5989 */
5976 5990 if (spa->spa_pool_props_object == 0) {
5977 5991 spa->spa_pool_props_object =
5978 5992 zap_create_link(mos, DMU_OT_POOL_PROPS,
5979 5993 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
5980 5994 tx);
5981 5995 }
5982 5996
5983 5997 /* normalize the property name */
5984 5998 propname = zpool_prop_to_name(prop);
5985 5999 proptype = zpool_prop_get_type(prop);
5986 6000
5987 6001 if (nvpair_type(elem) == DATA_TYPE_STRING) {
5988 6002 ASSERT(proptype == PROP_TYPE_STRING);
5989 6003 strval = fnvpair_value_string(elem);
5990 6004 VERIFY0(zap_update(mos,
5991 6005 spa->spa_pool_props_object, propname,
5992 6006 1, strlen(strval) + 1, strval, tx));
5993 6007 spa_history_log_internal(spa, "set", tx,
5994 6008 "%s=%s", nvpair_name(elem), strval);
5995 6009 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
5996 6010 intval = fnvpair_value_uint64(elem);
5997 6011
5998 6012 if (proptype == PROP_TYPE_INDEX) {
5999 6013 const char *unused;
6000 6014 VERIFY0(zpool_prop_index_to_string(
6001 6015 prop, intval, &unused));
6002 6016 }
6003 6017 VERIFY0(zap_update(mos,
6004 6018 spa->spa_pool_props_object, propname,
6005 6019 8, 1, &intval, tx));
6006 6020 spa_history_log_internal(spa, "set", tx,
6007 6021 "%s=%lld", nvpair_name(elem), intval);
6008 6022 } else {
6009 6023 ASSERT(0); /* not allowed */
6010 6024 }
6011 6025
6012 6026 switch (prop) {
6013 6027 case ZPOOL_PROP_DELEGATION:
6014 6028 spa->spa_delegation = intval;
6015 6029 break;
6016 6030 case ZPOOL_PROP_BOOTFS:
6017 6031 spa->spa_bootfs = intval;
6018 6032 break;
6019 6033 case ZPOOL_PROP_FAILUREMODE:
6020 6034 spa->spa_failmode = intval;
6021 6035 break;
6022 6036 case ZPOOL_PROP_AUTOEXPAND:
6023 6037 spa->spa_autoexpand = intval;
6024 6038 if (tx->tx_txg != TXG_INITIAL)
6025 6039 spa_async_request(spa,
6026 6040 SPA_ASYNC_AUTOEXPAND);
6027 6041 break;
6028 6042 case ZPOOL_PROP_DEDUPDITTO:
6029 6043 spa->spa_dedup_ditto = intval;
6030 6044 break;
6031 6045 default:
6032 6046 break;
6033 6047 }
6034 6048 }
6035 6049
6036 6050 }
6037 6051
6038 6052 mutex_exit(&spa->spa_props_lock);
6039 6053 }
6040 6054
6041 6055 /*
6042 6056 * Perform one-time upgrade on-disk changes. spa_version() does not
6043 6057 * reflect the new version this txg, so there must be no changes this
6044 6058 * txg to anything that the upgrade code depends on after it executes.
6045 6059 * Therefore this must be called after dsl_pool_sync() does the sync
6046 6060 * tasks.
6047 6061 */
6048 6062 static void
6049 6063 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
6050 6064 {
6051 6065 dsl_pool_t *dp = spa->spa_dsl_pool;
6052 6066
6053 6067 ASSERT(spa->spa_sync_pass == 1);
6054 6068
6055 6069 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
6056 6070
6057 6071 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
6058 6072 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
6059 6073 dsl_pool_create_origin(dp, tx);
6060 6074
6061 6075 /* Keeping the origin open increases spa_minref */
6062 6076 spa->spa_minref += 3;
6063 6077 }
6064 6078
6065 6079 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
6066 6080 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
6067 6081 dsl_pool_upgrade_clones(dp, tx);
6068 6082 }
6069 6083
6070 6084 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
6071 6085 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
6072 6086 dsl_pool_upgrade_dir_clones(dp, tx);
6073 6087
6074 6088 /* Keeping the freedir open increases spa_minref */
6075 6089 spa->spa_minref += 3;
6076 6090 }
6077 6091
6078 6092 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
6079 6093 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6080 6094 spa_feature_create_zap_objects(spa, tx);
6081 6095 }
6082 6096 rrw_exit(&dp->dp_config_rwlock, FTAG);
6083 6097 }
6084 6098
6085 6099 /*
6086 6100 * Sync the specified transaction group. New blocks may be dirtied as
6087 6101 * part of the process, so we iterate until it converges.
6088 6102 */
6089 6103 void
6090 6104 spa_sync(spa_t *spa, uint64_t txg)
6091 6105 {
6092 6106 dsl_pool_t *dp = spa->spa_dsl_pool;
6093 6107 objset_t *mos = spa->spa_meta_objset;
6094 6108 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
6095 6109 vdev_t *rvd = spa->spa_root_vdev;
6096 6110 vdev_t *vd;
6097 6111 dmu_tx_t *tx;
6098 6112 int error;
6099 6113
6100 6114 VERIFY(spa_writeable(spa));
6101 6115
6102 6116 /*
6103 6117 * Lock out configuration changes.
6104 6118 */
6105 6119 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6106 6120
6107 6121 spa->spa_syncing_txg = txg;
6108 6122 spa->spa_sync_pass = 0;
6109 6123
6110 6124 /*
6111 6125 * If there are any pending vdev state changes, convert them
6112 6126 * into config changes that go out with this transaction group.
6113 6127 */
6114 6128 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6115 6129 while (list_head(&spa->spa_state_dirty_list) != NULL) {
6116 6130 /*
6117 6131 * We need the write lock here because, for aux vdevs,
6118 6132 * calling vdev_config_dirty() modifies sav_config.
6119 6133 * This is ugly and will become unnecessary when we
6120 6134 * eliminate the aux vdev wart by integrating all vdevs
6121 6135 * into the root vdev tree.
6122 6136 */
6123 6137 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6124 6138 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
6125 6139 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
6126 6140 vdev_state_clean(vd);
6127 6141 vdev_config_dirty(vd);
6128 6142 }
6129 6143 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6130 6144 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
6131 6145 }
6132 6146 spa_config_exit(spa, SCL_STATE, FTAG);
6133 6147
6134 6148 tx = dmu_tx_create_assigned(dp, txg);
6135 6149
6136 6150 spa->spa_sync_starttime = gethrtime();
6137 6151 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid,
6138 6152 spa->spa_sync_starttime + spa->spa_deadman_synctime));
6139 6153
6140 6154 /*
6141 6155 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6142 6156 * set spa_deflate if we have no raid-z vdevs.
6143 6157 */
6144 6158 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
6145 6159 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
6146 6160 int i;
6147 6161
6148 6162 for (i = 0; i < rvd->vdev_children; i++) {
6149 6163 vd = rvd->vdev_child[i];
6150 6164 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
6151 6165 break;
6152 6166 }
6153 6167 if (i == rvd->vdev_children) {
6154 6168 spa->spa_deflate = TRUE;
6155 6169 VERIFY(0 == zap_add(spa->spa_meta_objset,
6156 6170 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
6157 6171 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
6158 6172 }
6159 6173 }
6160 6174
6161 6175 /*
6162 6176 * If anything has changed in this txg, or if someone is waiting
6163 6177 * for this txg to sync (eg, spa_vdev_remove()), push the
6164 6178 * deferred frees from the previous txg. If not, leave them
6165 6179 * alone so that we don't generate work on an otherwise idle
6166 6180 * system.
6167 6181 */
6168 6182 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
6169 6183 !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
6170 6184 !txg_list_empty(&dp->dp_sync_tasks, txg) ||
6171 6185 ((dsl_scan_active(dp->dp_scan) ||
6172 6186 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
6173 6187 spa_sync_deferred_frees(spa, tx);
6174 6188 }
6175 6189
6176 6190 /*
6177 6191 * Iterate to convergence.
6178 6192 */
6179 6193 do {
6180 6194 int pass = ++spa->spa_sync_pass;
6181 6195
6182 6196 spa_sync_config_object(spa, tx);
6183 6197 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
6184 6198 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
6185 6199 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
6186 6200 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
6187 6201 spa_errlog_sync(spa, txg);
6188 6202 dsl_pool_sync(dp, txg);
6189 6203
6190 6204 if (pass < zfs_sync_pass_deferred_free) {
6191 6205 spa_sync_frees(spa, free_bpl, tx);
6192 6206 } else {
6193 6207 bplist_iterate(free_bpl, bpobj_enqueue_cb,
6194 6208 &spa->spa_deferred_bpobj, tx);
6195 6209 }
6196 6210
6197 6211 ddt_sync(spa, txg);
6198 6212 dsl_scan_sync(dp, tx);
6199 6213
6200 6214 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
6201 6215 vdev_sync(vd, txg);
6202 6216
6203 6217 if (pass == 1)
6204 6218 spa_sync_upgrades(spa, tx);
6205 6219
6206 6220 } while (dmu_objset_is_dirty(mos, txg));
6207 6221
6208 6222 /*
6209 6223 * Rewrite the vdev configuration (which includes the uberblock)
6210 6224 * to commit the transaction group.
6211 6225 *
6212 6226 * If there are no dirty vdevs, we sync the uberblock to a few
6213 6227 * random top-level vdevs that are known to be visible in the
6214 6228 * config cache (see spa_vdev_add() for a complete description).
6215 6229 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6216 6230 */
6217 6231 for (;;) {
6218 6232 /*
6219 6233 * We hold SCL_STATE to prevent vdev open/close/etc.
6220 6234 * while we're attempting to write the vdev labels.
6221 6235 */
6222 6236 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6223 6237
6224 6238 if (list_is_empty(&spa->spa_config_dirty_list)) {
6225 6239 vdev_t *svd[SPA_DVAS_PER_BP];
6226 6240 int svdcount = 0;
6227 6241 int children = rvd->vdev_children;
6228 6242 int c0 = spa_get_random(children);
6229 6243
6230 6244 for (int c = 0; c < children; c++) {
6231 6245 vd = rvd->vdev_child[(c0 + c) % children];
6232 6246 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
6233 6247 continue;
6234 6248 svd[svdcount++] = vd;
6235 6249 if (svdcount == SPA_DVAS_PER_BP)
6236 6250 break;
6237 6251 }
6238 6252 error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
6239 6253 if (error != 0)
6240 6254 error = vdev_config_sync(svd, svdcount, txg,
6241 6255 B_TRUE);
6242 6256 } else {
6243 6257 error = vdev_config_sync(rvd->vdev_child,
6244 6258 rvd->vdev_children, txg, B_FALSE);
6245 6259 if (error != 0)
6246 6260 error = vdev_config_sync(rvd->vdev_child,
6247 6261 rvd->vdev_children, txg, B_TRUE);
6248 6262 }
6249 6263
6250 6264 if (error == 0)
6251 6265 spa->spa_last_synced_guid = rvd->vdev_guid;
6252 6266
6253 6267 spa_config_exit(spa, SCL_STATE, FTAG);
6254 6268
6255 6269 if (error == 0)
6256 6270 break;
6257 6271 zio_suspend(spa, NULL);
6258 6272 zio_resume_wait(spa);
6259 6273 }
6260 6274 dmu_tx_commit(tx);
6261 6275
6262 6276 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
6263 6277
6264 6278 /*
6265 6279 * Clear the dirty config list.
6266 6280 */
6267 6281 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
6268 6282 vdev_config_clean(vd);
6269 6283
6270 6284 /*
6271 6285 * Now that the new config has synced transactionally,
6272 6286 * let it become visible to the config cache.
6273 6287 */
6274 6288 if (spa->spa_config_syncing != NULL) {
6275 6289 spa_config_set(spa, spa->spa_config_syncing);
6276 6290 spa->spa_config_txg = txg;
6277 6291 spa->spa_config_syncing = NULL;
6278 6292 }
6279 6293
6280 6294 spa->spa_ubsync = spa->spa_uberblock;
6281 6295
6282 6296 dsl_pool_sync_done(dp, txg);
6283 6297
6284 6298 /*
6285 6299 * Update usable space statistics.
6286 6300 */
6287 6301 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
6288 6302 vdev_sync_done(vd, txg);
6289 6303
6290 6304 spa_update_dspace(spa);
6291 6305
6292 6306 /*
6293 6307 * It had better be the case that we didn't dirty anything
6294 6308 * since vdev_config_sync().
6295 6309 */
6296 6310 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
6297 6311 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
6298 6312 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
6299 6313
6300 6314 spa->spa_sync_pass = 0;
6301 6315
6302 6316 spa_config_exit(spa, SCL_CONFIG, FTAG);
6303 6317
6304 6318 spa_handle_ignored_writes(spa);
6305 6319
6306 6320 /*
6307 6321 * If any async tasks have been requested, kick them off.
6308 6322 */
6309 6323 spa_async_dispatch(spa);
6310 6324 }
6311 6325
6312 6326 /*
6313 6327 * Sync all pools. We don't want to hold the namespace lock across these
6314 6328 * operations, so we take a reference on the spa_t and drop the lock during the
6315 6329 * sync.
6316 6330 */
6317 6331 void
6318 6332 spa_sync_allpools(void)
6319 6333 {
6320 6334 spa_t *spa = NULL;
6321 6335 mutex_enter(&spa_namespace_lock);
6322 6336 while ((spa = spa_next(spa)) != NULL) {
6323 6337 if (spa_state(spa) != POOL_STATE_ACTIVE ||
6324 6338 !spa_writeable(spa) || spa_suspended(spa))
6325 6339 continue;
6326 6340 spa_open_ref(spa, FTAG);
6327 6341 mutex_exit(&spa_namespace_lock);
6328 6342 txg_wait_synced(spa_get_dsl(spa), 0);
6329 6343 mutex_enter(&spa_namespace_lock);
6330 6344 spa_close(spa, FTAG);
6331 6345 }
6332 6346 mutex_exit(&spa_namespace_lock);
6333 6347 }
6334 6348
6335 6349 /*
6336 6350 * ==========================================================================
6337 6351 * Miscellaneous routines
6338 6352 * ==========================================================================
6339 6353 */
6340 6354
6341 6355 /*
6342 6356 * Remove all pools in the system.
6343 6357 */
6344 6358 void
6345 6359 spa_evict_all(void)
6346 6360 {
6347 6361 spa_t *spa;
6348 6362
6349 6363 /*
6350 6364 * Remove all cached state. All pools should be closed now,
6351 6365 * so every spa in the AVL tree should be unreferenced.
6352 6366 */
6353 6367 mutex_enter(&spa_namespace_lock);
6354 6368 while ((spa = spa_next(NULL)) != NULL) {
6355 6369 /*
6356 6370 * Stop async tasks. The async thread may need to detach
6357 6371 * a device that's been replaced, which requires grabbing
6358 6372 * spa_namespace_lock, so we must drop it here.
6359 6373 */
6360 6374 spa_open_ref(spa, FTAG);
6361 6375 mutex_exit(&spa_namespace_lock);
6362 6376 spa_async_suspend(spa);
6363 6377 mutex_enter(&spa_namespace_lock);
6364 6378 spa_close(spa, FTAG);
6365 6379
6366 6380 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
6367 6381 spa_unload(spa);
6368 6382 spa_deactivate(spa);
6369 6383 }
6370 6384 spa_remove(spa);
6371 6385 }
6372 6386 mutex_exit(&spa_namespace_lock);
6373 6387 }
6374 6388
6375 6389 vdev_t *
6376 6390 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
6377 6391 {
6378 6392 vdev_t *vd;
6379 6393 int i;
6380 6394
6381 6395 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
6382 6396 return (vd);
6383 6397
6384 6398 if (aux) {
6385 6399 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
6386 6400 vd = spa->spa_l2cache.sav_vdevs[i];
6387 6401 if (vd->vdev_guid == guid)
6388 6402 return (vd);
6389 6403 }
6390 6404
6391 6405 for (i = 0; i < spa->spa_spares.sav_count; i++) {
6392 6406 vd = spa->spa_spares.sav_vdevs[i];
6393 6407 if (vd->vdev_guid == guid)
6394 6408 return (vd);
6395 6409 }
6396 6410 }
6397 6411
6398 6412 return (NULL);
6399 6413 }
6400 6414
6401 6415 void
6402 6416 spa_upgrade(spa_t *spa, uint64_t version)
6403 6417 {
6404 6418 ASSERT(spa_writeable(spa));
6405 6419
6406 6420 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6407 6421
6408 6422 /*
6409 6423 * This should only be called for a non-faulted pool, and since a
6410 6424 * future version would result in an unopenable pool, this shouldn't be
6411 6425 * possible.
6412 6426 */
6413 6427 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version));
6414 6428 ASSERT(version >= spa->spa_uberblock.ub_version);
6415 6429
6416 6430 spa->spa_uberblock.ub_version = version;
6417 6431 vdev_config_dirty(spa->spa_root_vdev);
6418 6432
6419 6433 spa_config_exit(spa, SCL_ALL, FTAG);
6420 6434
6421 6435 txg_wait_synced(spa_get_dsl(spa), 0);
6422 6436 }
6423 6437
6424 6438 boolean_t
6425 6439 spa_has_spare(spa_t *spa, uint64_t guid)
6426 6440 {
6427 6441 int i;
6428 6442 uint64_t spareguid;
6429 6443 spa_aux_vdev_t *sav = &spa->spa_spares;
6430 6444
6431 6445 for (i = 0; i < sav->sav_count; i++)
6432 6446 if (sav->sav_vdevs[i]->vdev_guid == guid)
6433 6447 return (B_TRUE);
6434 6448
6435 6449 for (i = 0; i < sav->sav_npending; i++) {
6436 6450 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
6437 6451 &spareguid) == 0 && spareguid == guid)
6438 6452 return (B_TRUE);
6439 6453 }
6440 6454
6441 6455 return (B_FALSE);
6442 6456 }
6443 6457
6444 6458 /*
6445 6459 * Check if a pool has an active shared spare device.
6446 6460 * Note: reference count of an active spare is 2, as a spare and as a replace
6447 6461 */
6448 6462 static boolean_t
6449 6463 spa_has_active_shared_spare(spa_t *spa)
6450 6464 {
6451 6465 int i, refcnt;
6452 6466 uint64_t pool;
6453 6467 spa_aux_vdev_t *sav = &spa->spa_spares;
6454 6468
6455 6469 for (i = 0; i < sav->sav_count; i++) {
6456 6470 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
6457 6471 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
6458 6472 refcnt > 2)
6459 6473 return (B_TRUE);
6460 6474 }
6461 6475
6462 6476 return (B_FALSE);
6463 6477 }
6464 6478
6465 6479 /*
6466 6480 * Post a sysevent corresponding to the given event. The 'name' must be one of
6467 6481 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
6468 6482 * filled in from the spa and (optionally) the vdev. This doesn't do anything
6469 6483 * in the userland libzpool, as we don't want consumers to misinterpret ztest
6470 6484 * or zdb as real changes.
6471 6485 */
6472 6486 void
6473 6487 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
6474 6488 {
6475 6489 #ifdef _KERNEL
6476 6490 sysevent_t *ev;
6477 6491 sysevent_attr_list_t *attr = NULL;
6478 6492 sysevent_value_t value;
6479 6493 sysevent_id_t eid;
6480 6494
6481 6495 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
6482 6496 SE_SLEEP);
6483 6497
6484 6498 value.value_type = SE_DATA_TYPE_STRING;
6485 6499 value.value.sv_string = spa_name(spa);
6486 6500 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
6487 6501 goto done;
6488 6502
6489 6503 value.value_type = SE_DATA_TYPE_UINT64;
6490 6504 value.value.sv_uint64 = spa_guid(spa);
6491 6505 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
6492 6506 goto done;
6493 6507
6494 6508 if (vd) {
6495 6509 value.value_type = SE_DATA_TYPE_UINT64;
6496 6510 value.value.sv_uint64 = vd->vdev_guid;
6497 6511 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
6498 6512 SE_SLEEP) != 0)
6499 6513 goto done;
6500 6514
6501 6515 if (vd->vdev_path) {
6502 6516 value.value_type = SE_DATA_TYPE_STRING;
6503 6517 value.value.sv_string = vd->vdev_path;
6504 6518 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
6505 6519 &value, SE_SLEEP) != 0)
6506 6520 goto done;
6507 6521 }
6508 6522 }
6509 6523
6510 6524 if (sysevent_attach_attributes(ev, attr) != 0)
6511 6525 goto done;
6512 6526 attr = NULL;
6513 6527
6514 6528 (void) log_sysevent(ev, SE_SLEEP, &eid);
6515 6529
6516 6530 done:
6517 6531 if (attr)
6518 6532 sysevent_free_attr(attr);
6519 6533 sysevent_free(ev);
6520 6534 #endif
6521 6535 }
|
↓ open down ↓ |
864 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX