Print this page
4185 New hash algorithm support
| 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
|
↓ open down ↓ |
15 lines elided |
↑ open up ↑ |
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 + * Copyright 2013 Saso Kiselkov. All rights reserved.
26 27 */
27 28
28 29 /*
29 30 * SPA: Storage Pool Allocator
30 31 *
31 32 * This file contains all the routines used when modifying on-disk SPA state.
32 33 * This includes opening, importing, destroying, exporting a pool, and syncing a
33 34 * pool.
34 35 */
35 36
36 37 #include <sys/zfs_context.h>
37 38 #include <sys/fm/fs/zfs.h>
38 39 #include <sys/spa_impl.h>
39 40 #include <sys/zio.h>
40 41 #include <sys/zio_checksum.h>
41 42 #include <sys/dmu.h>
42 43 #include <sys/dmu_tx.h>
43 44 #include <sys/zap.h>
44 45 #include <sys/zil.h>
45 46 #include <sys/ddt.h>
46 47 #include <sys/vdev_impl.h>
47 48 #include <sys/metaslab.h>
48 49 #include <sys/metaslab_impl.h>
49 50 #include <sys/uberblock_impl.h>
50 51 #include <sys/txg.h>
51 52 #include <sys/avl.h>
52 53 #include <sys/dmu_traverse.h>
53 54 #include <sys/dmu_objset.h>
54 55 #include <sys/unique.h>
55 56 #include <sys/dsl_pool.h>
56 57 #include <sys/dsl_dataset.h>
57 58 #include <sys/dsl_dir.h>
58 59 #include <sys/dsl_prop.h>
59 60 #include <sys/dsl_synctask.h>
60 61 #include <sys/fs/zfs.h>
61 62 #include <sys/arc.h>
62 63 #include <sys/callb.h>
63 64 #include <sys/systeminfo.h>
64 65 #include <sys/spa_boot.h>
65 66 #include <sys/zfs_ioctl.h>
66 67 #include <sys/dsl_scan.h>
67 68 #include <sys/zfeature.h>
68 69 #include <sys/dsl_destroy.h>
69 70
70 71 #ifdef _KERNEL
71 72 #include <sys/bootprops.h>
72 73 #include <sys/callb.h>
73 74 #include <sys/cpupart.h>
74 75 #include <sys/pool.h>
75 76 #include <sys/sysdc.h>
76 77 #include <sys/zone.h>
77 78 #endif /* _KERNEL */
78 79
79 80 #include "zfs_prop.h"
80 81 #include "zfs_comutil.h"
81 82
82 83 /*
83 84 * The interval, in seconds, at which failed configuration cache file writes
84 85 * should be retried.
85 86 */
86 87 static int zfs_ccw_retry_interval = 300;
87 88
88 89 typedef enum zti_modes {
89 90 ZTI_MODE_FIXED, /* value is # of threads (min 1) */
90 91 ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */
91 92 ZTI_MODE_NULL, /* don't create a taskq */
92 93 ZTI_NMODES
93 94 } zti_modes_t;
94 95
95 96 #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) }
96 97 #define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 }
97 98 #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 }
98 99
99 100 #define ZTI_N(n) ZTI_P(n, 1)
100 101 #define ZTI_ONE ZTI_N(1)
101 102
102 103 typedef struct zio_taskq_info {
103 104 zti_modes_t zti_mode;
104 105 uint_t zti_value;
105 106 uint_t zti_count;
106 107 } zio_taskq_info_t;
107 108
108 109 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
109 110 "issue", "issue_high", "intr", "intr_high"
110 111 };
111 112
112 113 /*
113 114 * This table defines the taskq settings for each ZFS I/O type. When
114 115 * initializing a pool, we use this table to create an appropriately sized
115 116 * taskq. Some operations are low volume and therefore have a small, static
116 117 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
117 118 * macros. Other operations process a large amount of data; the ZTI_BATCH
118 119 * macro causes us to create a taskq oriented for throughput. Some operations
119 120 * are so high frequency and short-lived that the taskq itself can become a a
120 121 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
121 122 * additional degree of parallelism specified by the number of threads per-
122 123 * taskq and the number of taskqs; when dispatching an event in this case, the
123 124 * particular taskq is chosen at random.
124 125 *
125 126 * The different taskq priorities are to handle the different contexts (issue
126 127 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
127 128 * need to be handled with minimum delay.
128 129 */
129 130 const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
130 131 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
131 132 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */
132 133 { ZTI_N(8), ZTI_NULL, ZTI_BATCH, ZTI_NULL }, /* READ */
133 134 { ZTI_BATCH, ZTI_N(5), ZTI_N(8), ZTI_N(5) }, /* WRITE */
134 135 { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */
135 136 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */
136 137 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */
137 138 };
138 139
139 140 static void spa_sync_version(void *arg, dmu_tx_t *tx);
140 141 static void spa_sync_props(void *arg, dmu_tx_t *tx);
141 142 static boolean_t spa_has_active_shared_spare(spa_t *spa);
142 143 static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
143 144 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
144 145 char **ereport);
145 146 static void spa_vdev_resilver_done(spa_t *spa);
146 147
147 148 uint_t zio_taskq_batch_pct = 75; /* 1 thread per cpu in pset */
148 149 id_t zio_taskq_psrset_bind = PS_NONE;
149 150 boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
150 151 uint_t zio_taskq_basedc = 80; /* base duty cycle */
151 152
152 153 boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
153 154 extern int zfs_sync_pass_deferred_free;
154 155
155 156 /*
156 157 * This (illegal) pool name is used when temporarily importing a spa_t in order
157 158 * to get the vdev stats associated with the imported devices.
158 159 */
159 160 #define TRYIMPORT_NAME "$import"
160 161
161 162 /*
162 163 * ==========================================================================
163 164 * SPA properties routines
164 165 * ==========================================================================
165 166 */
166 167
167 168 /*
168 169 * Add a (source=src, propname=propval) list to an nvlist.
169 170 */
170 171 static void
171 172 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
172 173 uint64_t intval, zprop_source_t src)
173 174 {
174 175 const char *propname = zpool_prop_to_name(prop);
175 176 nvlist_t *propval;
176 177
177 178 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
178 179 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
179 180
180 181 if (strval != NULL)
181 182 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
182 183 else
183 184 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
184 185
185 186 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
186 187 nvlist_free(propval);
187 188 }
188 189
189 190 /*
190 191 * Get property values from the spa configuration.
191 192 */
192 193 static void
193 194 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
194 195 {
195 196 vdev_t *rvd = spa->spa_root_vdev;
196 197 dsl_pool_t *pool = spa->spa_dsl_pool;
197 198 uint64_t size;
198 199 uint64_t alloc;
199 200 uint64_t space;
200 201 uint64_t cap, version;
201 202 zprop_source_t src = ZPROP_SRC_NONE;
202 203 spa_config_dirent_t *dp;
203 204
204 205 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
205 206
206 207 if (rvd != NULL) {
207 208 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
208 209 size = metaslab_class_get_space(spa_normal_class(spa));
209 210 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
210 211 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
211 212 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
212 213 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
213 214 size - alloc, src);
214 215
215 216 space = 0;
216 217 for (int c = 0; c < rvd->vdev_children; c++) {
217 218 vdev_t *tvd = rvd->vdev_child[c];
218 219 space += tvd->vdev_max_asize - tvd->vdev_asize;
219 220 }
220 221 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, space,
221 222 src);
222 223
223 224 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
224 225 (spa_mode(spa) == FREAD), src);
225 226
226 227 cap = (size == 0) ? 0 : (alloc * 100 / size);
227 228 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
228 229
229 230 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
230 231 ddt_get_pool_dedup_ratio(spa), src);
231 232
232 233 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
233 234 rvd->vdev_state, src);
234 235
235 236 version = spa_version(spa);
236 237 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
237 238 src = ZPROP_SRC_DEFAULT;
238 239 else
239 240 src = ZPROP_SRC_LOCAL;
240 241 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
241 242 }
242 243
243 244 if (pool != NULL) {
244 245 dsl_dir_t *freedir = pool->dp_free_dir;
245 246
246 247 /*
247 248 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
248 249 * when opening pools before this version freedir will be NULL.
249 250 */
250 251 if (freedir != NULL) {
251 252 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
252 253 freedir->dd_phys->dd_used_bytes, src);
253 254 } else {
254 255 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
255 256 NULL, 0, src);
256 257 }
257 258 }
258 259
259 260 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
260 261
261 262 if (spa->spa_comment != NULL) {
262 263 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
263 264 0, ZPROP_SRC_LOCAL);
264 265 }
265 266
266 267 if (spa->spa_root != NULL)
267 268 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
268 269 0, ZPROP_SRC_LOCAL);
269 270
270 271 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
271 272 if (dp->scd_path == NULL) {
272 273 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
273 274 "none", 0, ZPROP_SRC_LOCAL);
274 275 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
275 276 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
276 277 dp->scd_path, 0, ZPROP_SRC_LOCAL);
277 278 }
278 279 }
279 280 }
280 281
281 282 /*
282 283 * Get zpool property values.
283 284 */
284 285 int
285 286 spa_prop_get(spa_t *spa, nvlist_t **nvp)
286 287 {
287 288 objset_t *mos = spa->spa_meta_objset;
288 289 zap_cursor_t zc;
289 290 zap_attribute_t za;
290 291 int err;
291 292
292 293 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
293 294
294 295 mutex_enter(&spa->spa_props_lock);
295 296
296 297 /*
297 298 * Get properties from the spa config.
298 299 */
299 300 spa_prop_get_config(spa, nvp);
300 301
301 302 /* If no pool property object, no more prop to get. */
302 303 if (mos == NULL || spa->spa_pool_props_object == 0) {
303 304 mutex_exit(&spa->spa_props_lock);
304 305 return (0);
305 306 }
306 307
307 308 /*
308 309 * Get properties from the MOS pool property object.
309 310 */
310 311 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
311 312 (err = zap_cursor_retrieve(&zc, &za)) == 0;
312 313 zap_cursor_advance(&zc)) {
313 314 uint64_t intval = 0;
314 315 char *strval = NULL;
315 316 zprop_source_t src = ZPROP_SRC_DEFAULT;
316 317 zpool_prop_t prop;
317 318
318 319 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
319 320 continue;
320 321
321 322 switch (za.za_integer_length) {
322 323 case 8:
323 324 /* integer property */
324 325 if (za.za_first_integer !=
325 326 zpool_prop_default_numeric(prop))
326 327 src = ZPROP_SRC_LOCAL;
327 328
328 329 if (prop == ZPOOL_PROP_BOOTFS) {
329 330 dsl_pool_t *dp;
330 331 dsl_dataset_t *ds = NULL;
331 332
332 333 dp = spa_get_dsl(spa);
333 334 dsl_pool_config_enter(dp, FTAG);
334 335 if (err = dsl_dataset_hold_obj(dp,
335 336 za.za_first_integer, FTAG, &ds)) {
336 337 dsl_pool_config_exit(dp, FTAG);
337 338 break;
338 339 }
339 340
340 341 strval = kmem_alloc(
341 342 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
342 343 KM_SLEEP);
343 344 dsl_dataset_name(ds, strval);
344 345 dsl_dataset_rele(ds, FTAG);
345 346 dsl_pool_config_exit(dp, FTAG);
346 347 } else {
347 348 strval = NULL;
348 349 intval = za.za_first_integer;
349 350 }
350 351
351 352 spa_prop_add_list(*nvp, prop, strval, intval, src);
352 353
353 354 if (strval != NULL)
354 355 kmem_free(strval,
355 356 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
356 357
357 358 break;
358 359
359 360 case 1:
360 361 /* string property */
361 362 strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
362 363 err = zap_lookup(mos, spa->spa_pool_props_object,
363 364 za.za_name, 1, za.za_num_integers, strval);
364 365 if (err) {
365 366 kmem_free(strval, za.za_num_integers);
366 367 break;
367 368 }
368 369 spa_prop_add_list(*nvp, prop, strval, 0, src);
369 370 kmem_free(strval, za.za_num_integers);
370 371 break;
371 372
372 373 default:
373 374 break;
374 375 }
375 376 }
376 377 zap_cursor_fini(&zc);
377 378 mutex_exit(&spa->spa_props_lock);
378 379 out:
379 380 if (err && err != ENOENT) {
380 381 nvlist_free(*nvp);
381 382 *nvp = NULL;
382 383 return (err);
383 384 }
384 385
385 386 return (0);
386 387 }
387 388
388 389 /*
389 390 * Validate the given pool properties nvlist and modify the list
390 391 * for the property values to be set.
391 392 */
392 393 static int
393 394 spa_prop_validate(spa_t *spa, nvlist_t *props)
394 395 {
395 396 nvpair_t *elem;
396 397 int error = 0, reset_bootfs = 0;
397 398 uint64_t objnum = 0;
398 399 boolean_t has_feature = B_FALSE;
399 400
400 401 elem = NULL;
401 402 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
402 403 uint64_t intval;
403 404 char *strval, *slash, *check, *fname;
404 405 const char *propname = nvpair_name(elem);
405 406 zpool_prop_t prop = zpool_name_to_prop(propname);
406 407
407 408 switch (prop) {
408 409 case ZPROP_INVAL:
409 410 if (!zpool_prop_feature(propname)) {
410 411 error = SET_ERROR(EINVAL);
411 412 break;
412 413 }
413 414
414 415 /*
415 416 * Sanitize the input.
416 417 */
417 418 if (nvpair_type(elem) != DATA_TYPE_UINT64) {
418 419 error = SET_ERROR(EINVAL);
419 420 break;
420 421 }
421 422
422 423 if (nvpair_value_uint64(elem, &intval) != 0) {
423 424 error = SET_ERROR(EINVAL);
424 425 break;
425 426 }
426 427
427 428 if (intval != 0) {
428 429 error = SET_ERROR(EINVAL);
429 430 break;
430 431 }
431 432
432 433 fname = strchr(propname, '@') + 1;
433 434 if (zfeature_lookup_name(fname, NULL) != 0) {
434 435 error = SET_ERROR(EINVAL);
435 436 break;
436 437 }
437 438
438 439 has_feature = B_TRUE;
439 440 break;
440 441
441 442 case ZPOOL_PROP_VERSION:
442 443 error = nvpair_value_uint64(elem, &intval);
443 444 if (!error &&
444 445 (intval < spa_version(spa) ||
445 446 intval > SPA_VERSION_BEFORE_FEATURES ||
446 447 has_feature))
447 448 error = SET_ERROR(EINVAL);
448 449 break;
449 450
450 451 case ZPOOL_PROP_DELEGATION:
451 452 case ZPOOL_PROP_AUTOREPLACE:
452 453 case ZPOOL_PROP_LISTSNAPS:
453 454 case ZPOOL_PROP_AUTOEXPAND:
454 455 error = nvpair_value_uint64(elem, &intval);
455 456 if (!error && intval > 1)
456 457 error = SET_ERROR(EINVAL);
457 458 break;
458 459
459 460 case ZPOOL_PROP_BOOTFS:
460 461 /*
461 462 * If the pool version is less than SPA_VERSION_BOOTFS,
462 463 * or the pool is still being created (version == 0),
463 464 * the bootfs property cannot be set.
464 465 */
465 466 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
466 467 error = SET_ERROR(ENOTSUP);
467 468 break;
468 469 }
469 470
470 471 /*
471 472 * Make sure the vdev config is bootable
472 473 */
473 474 if (!vdev_is_bootable(spa->spa_root_vdev)) {
474 475 error = SET_ERROR(ENOTSUP);
475 476 break;
476 477 }
477 478
478 479 reset_bootfs = 1;
479 480
480 481 error = nvpair_value_string(elem, &strval);
481 482
482 483 if (!error) {
483 484 objset_t *os;
484 485 uint64_t compress;
485 486
486 487 if (strval == NULL || strval[0] == '\0') {
487 488 objnum = zpool_prop_default_numeric(
488 489 ZPOOL_PROP_BOOTFS);
489 490 break;
490 491 }
491 492
492 493 if (error = dmu_objset_hold(strval, FTAG, &os))
493 494 break;
494 495
495 496 /* Must be ZPL and not gzip compressed. */
496 497
497 498 if (dmu_objset_type(os) != DMU_OST_ZFS) {
498 499 error = SET_ERROR(ENOTSUP);
499 500 } else if ((error =
500 501 dsl_prop_get_int_ds(dmu_objset_ds(os),
501 502 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
502 503 &compress)) == 0 &&
503 504 !BOOTFS_COMPRESS_VALID(compress)) {
504 505 error = SET_ERROR(ENOTSUP);
505 506 } else {
506 507 objnum = dmu_objset_id(os);
507 508 }
508 509 dmu_objset_rele(os, FTAG);
509 510 }
510 511 break;
511 512
512 513 case ZPOOL_PROP_FAILUREMODE:
513 514 error = nvpair_value_uint64(elem, &intval);
514 515 if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
515 516 intval > ZIO_FAILURE_MODE_PANIC))
516 517 error = SET_ERROR(EINVAL);
517 518
518 519 /*
519 520 * This is a special case which only occurs when
520 521 * the pool has completely failed. This allows
521 522 * the user to change the in-core failmode property
522 523 * without syncing it out to disk (I/Os might
523 524 * currently be blocked). We do this by returning
524 525 * EIO to the caller (spa_prop_set) to trick it
525 526 * into thinking we encountered a property validation
526 527 * error.
527 528 */
528 529 if (!error && spa_suspended(spa)) {
529 530 spa->spa_failmode = intval;
530 531 error = SET_ERROR(EIO);
531 532 }
532 533 break;
533 534
534 535 case ZPOOL_PROP_CACHEFILE:
535 536 if ((error = nvpair_value_string(elem, &strval)) != 0)
536 537 break;
537 538
538 539 if (strval[0] == '\0')
539 540 break;
540 541
541 542 if (strcmp(strval, "none") == 0)
542 543 break;
543 544
544 545 if (strval[0] != '/') {
545 546 error = SET_ERROR(EINVAL);
546 547 break;
547 548 }
548 549
549 550 slash = strrchr(strval, '/');
550 551 ASSERT(slash != NULL);
551 552
552 553 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
553 554 strcmp(slash, "/..") == 0)
554 555 error = SET_ERROR(EINVAL);
555 556 break;
556 557
557 558 case ZPOOL_PROP_COMMENT:
558 559 if ((error = nvpair_value_string(elem, &strval)) != 0)
559 560 break;
560 561 for (check = strval; *check != '\0'; check++) {
561 562 /*
562 563 * The kernel doesn't have an easy isprint()
563 564 * check. For this kernel check, we merely
564 565 * check ASCII apart from DEL. Fix this if
565 566 * there is an easy-to-use kernel isprint().
566 567 */
567 568 if (*check >= 0x7f) {
568 569 error = SET_ERROR(EINVAL);
569 570 break;
570 571 }
571 572 check++;
572 573 }
573 574 if (strlen(strval) > ZPROP_MAX_COMMENT)
574 575 error = E2BIG;
575 576 break;
576 577
577 578 case ZPOOL_PROP_DEDUPDITTO:
578 579 if (spa_version(spa) < SPA_VERSION_DEDUP)
579 580 error = SET_ERROR(ENOTSUP);
580 581 else
581 582 error = nvpair_value_uint64(elem, &intval);
582 583 if (error == 0 &&
583 584 intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
584 585 error = SET_ERROR(EINVAL);
585 586 break;
586 587 }
587 588
588 589 if (error)
589 590 break;
590 591 }
591 592
592 593 if (!error && reset_bootfs) {
593 594 error = nvlist_remove(props,
594 595 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
595 596
596 597 if (!error) {
597 598 error = nvlist_add_uint64(props,
598 599 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
599 600 }
600 601 }
601 602
602 603 return (error);
603 604 }
604 605
605 606 void
606 607 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
607 608 {
608 609 char *cachefile;
609 610 spa_config_dirent_t *dp;
610 611
611 612 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
612 613 &cachefile) != 0)
613 614 return;
614 615
615 616 dp = kmem_alloc(sizeof (spa_config_dirent_t),
616 617 KM_SLEEP);
617 618
618 619 if (cachefile[0] == '\0')
619 620 dp->scd_path = spa_strdup(spa_config_path);
620 621 else if (strcmp(cachefile, "none") == 0)
621 622 dp->scd_path = NULL;
622 623 else
623 624 dp->scd_path = spa_strdup(cachefile);
624 625
625 626 list_insert_head(&spa->spa_config_list, dp);
626 627 if (need_sync)
627 628 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
628 629 }
629 630
630 631 int
631 632 spa_prop_set(spa_t *spa, nvlist_t *nvp)
632 633 {
633 634 int error;
634 635 nvpair_t *elem = NULL;
635 636 boolean_t need_sync = B_FALSE;
636 637
637 638 if ((error = spa_prop_validate(spa, nvp)) != 0)
638 639 return (error);
639 640
640 641 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
641 642 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
642 643
643 644 if (prop == ZPOOL_PROP_CACHEFILE ||
644 645 prop == ZPOOL_PROP_ALTROOT ||
645 646 prop == ZPOOL_PROP_READONLY)
646 647 continue;
647 648
648 649 if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) {
649 650 uint64_t ver;
650 651
651 652 if (prop == ZPOOL_PROP_VERSION) {
652 653 VERIFY(nvpair_value_uint64(elem, &ver) == 0);
653 654 } else {
654 655 ASSERT(zpool_prop_feature(nvpair_name(elem)));
655 656 ver = SPA_VERSION_FEATURES;
656 657 need_sync = B_TRUE;
657 658 }
658 659
659 660 /* Save time if the version is already set. */
660 661 if (ver == spa_version(spa))
661 662 continue;
662 663
663 664 /*
664 665 * In addition to the pool directory object, we might
665 666 * create the pool properties object, the features for
666 667 * read object, the features for write object, or the
667 668 * feature descriptions object.
668 669 */
669 670 error = dsl_sync_task(spa->spa_name, NULL,
670 671 spa_sync_version, &ver, 6);
671 672 if (error)
672 673 return (error);
673 674 continue;
674 675 }
675 676
676 677 need_sync = B_TRUE;
677 678 break;
678 679 }
679 680
680 681 if (need_sync) {
681 682 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props,
682 683 nvp, 6));
683 684 }
684 685
685 686 return (0);
686 687 }
687 688
688 689 /*
689 690 * If the bootfs property value is dsobj, clear it.
690 691 */
691 692 void
692 693 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
693 694 {
694 695 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
695 696 VERIFY(zap_remove(spa->spa_meta_objset,
696 697 spa->spa_pool_props_object,
697 698 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
698 699 spa->spa_bootfs = 0;
699 700 }
700 701 }
701 702
702 703 /*ARGSUSED*/
703 704 static int
704 705 spa_change_guid_check(void *arg, dmu_tx_t *tx)
705 706 {
706 707 uint64_t *newguid = arg;
707 708 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
708 709 vdev_t *rvd = spa->spa_root_vdev;
709 710 uint64_t vdev_state;
710 711
711 712 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
712 713 vdev_state = rvd->vdev_state;
713 714 spa_config_exit(spa, SCL_STATE, FTAG);
714 715
715 716 if (vdev_state != VDEV_STATE_HEALTHY)
716 717 return (SET_ERROR(ENXIO));
717 718
718 719 ASSERT3U(spa_guid(spa), !=, *newguid);
719 720
720 721 return (0);
721 722 }
722 723
723 724 static void
724 725 spa_change_guid_sync(void *arg, dmu_tx_t *tx)
725 726 {
726 727 uint64_t *newguid = arg;
727 728 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
728 729 uint64_t oldguid;
729 730 vdev_t *rvd = spa->spa_root_vdev;
730 731
731 732 oldguid = spa_guid(spa);
732 733
733 734 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
734 735 rvd->vdev_guid = *newguid;
735 736 rvd->vdev_guid_sum += (*newguid - oldguid);
736 737 vdev_config_dirty(rvd);
737 738 spa_config_exit(spa, SCL_STATE, FTAG);
738 739
739 740 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu",
740 741 oldguid, *newguid);
741 742 }
742 743
743 744 /*
744 745 * Change the GUID for the pool. This is done so that we can later
745 746 * re-import a pool built from a clone of our own vdevs. We will modify
746 747 * the root vdev's guid, our own pool guid, and then mark all of our
747 748 * vdevs dirty. Note that we must make sure that all our vdevs are
748 749 * online when we do this, or else any vdevs that weren't present
749 750 * would be orphaned from our pool. We are also going to issue a
750 751 * sysevent to update any watchers.
751 752 */
752 753 int
753 754 spa_change_guid(spa_t *spa)
754 755 {
755 756 int error;
756 757 uint64_t guid;
757 758
758 759 mutex_enter(&spa->spa_vdev_top_lock);
759 760 mutex_enter(&spa_namespace_lock);
760 761 guid = spa_generate_guid(NULL);
761 762
762 763 error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
763 764 spa_change_guid_sync, &guid, 5);
764 765
765 766 if (error == 0) {
766 767 spa_config_sync(spa, B_FALSE, B_TRUE);
767 768 spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
768 769 }
769 770
770 771 mutex_exit(&spa_namespace_lock);
771 772 mutex_exit(&spa->spa_vdev_top_lock);
772 773
773 774 return (error);
774 775 }
775 776
776 777 /*
777 778 * ==========================================================================
778 779 * SPA state manipulation (open/create/destroy/import/export)
779 780 * ==========================================================================
780 781 */
781 782
782 783 static int
783 784 spa_error_entry_compare(const void *a, const void *b)
784 785 {
785 786 spa_error_entry_t *sa = (spa_error_entry_t *)a;
786 787 spa_error_entry_t *sb = (spa_error_entry_t *)b;
787 788 int ret;
788 789
789 790 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
790 791 sizeof (zbookmark_t));
791 792
792 793 if (ret < 0)
793 794 return (-1);
794 795 else if (ret > 0)
795 796 return (1);
796 797 else
797 798 return (0);
798 799 }
799 800
800 801 /*
801 802 * Utility function which retrieves copies of the current logs and
802 803 * re-initializes them in the process.
803 804 */
804 805 void
805 806 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
806 807 {
807 808 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
808 809
809 810 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
810 811 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
811 812
812 813 avl_create(&spa->spa_errlist_scrub,
813 814 spa_error_entry_compare, sizeof (spa_error_entry_t),
814 815 offsetof(spa_error_entry_t, se_avl));
815 816 avl_create(&spa->spa_errlist_last,
816 817 spa_error_entry_compare, sizeof (spa_error_entry_t),
817 818 offsetof(spa_error_entry_t, se_avl));
818 819 }
819 820
820 821 static void
821 822 spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
822 823 {
823 824 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
824 825 enum zti_modes mode = ztip->zti_mode;
825 826 uint_t value = ztip->zti_value;
826 827 uint_t count = ztip->zti_count;
827 828 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
828 829 char name[32];
829 830 uint_t flags = 0;
830 831 boolean_t batch = B_FALSE;
831 832
832 833 if (mode == ZTI_MODE_NULL) {
833 834 tqs->stqs_count = 0;
834 835 tqs->stqs_taskq = NULL;
835 836 return;
836 837 }
837 838
838 839 ASSERT3U(count, >, 0);
839 840
840 841 tqs->stqs_count = count;
841 842 tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP);
842 843
843 844 switch (mode) {
844 845 case ZTI_MODE_FIXED:
845 846 ASSERT3U(value, >=, 1);
846 847 value = MAX(value, 1);
847 848 break;
848 849
849 850 case ZTI_MODE_BATCH:
850 851 batch = B_TRUE;
851 852 flags |= TASKQ_THREADS_CPU_PCT;
852 853 value = zio_taskq_batch_pct;
853 854 break;
854 855
855 856 default:
856 857 panic("unrecognized mode for %s_%s taskq (%u:%u) in "
857 858 "spa_activate()",
858 859 zio_type_name[t], zio_taskq_types[q], mode, value);
859 860 break;
860 861 }
861 862
862 863 for (uint_t i = 0; i < count; i++) {
863 864 taskq_t *tq;
864 865
865 866 if (count > 1) {
866 867 (void) snprintf(name, sizeof (name), "%s_%s_%u",
867 868 zio_type_name[t], zio_taskq_types[q], i);
868 869 } else {
869 870 (void) snprintf(name, sizeof (name), "%s_%s",
870 871 zio_type_name[t], zio_taskq_types[q]);
871 872 }
872 873
873 874 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
874 875 if (batch)
875 876 flags |= TASKQ_DC_BATCH;
876 877
877 878 tq = taskq_create_sysdc(name, value, 50, INT_MAX,
878 879 spa->spa_proc, zio_taskq_basedc, flags);
879 880 } else {
880 881 pri_t pri = maxclsyspri;
881 882 /*
882 883 * The write issue taskq can be extremely CPU
883 884 * intensive. Run it at slightly lower priority
884 885 * than the other taskqs.
885 886 */
886 887 if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE)
887 888 pri--;
888 889
889 890 tq = taskq_create_proc(name, value, pri, 50,
890 891 INT_MAX, spa->spa_proc, flags);
891 892 }
892 893
893 894 tqs->stqs_taskq[i] = tq;
894 895 }
895 896 }
896 897
897 898 static void
898 899 spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
899 900 {
900 901 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
901 902
902 903 if (tqs->stqs_taskq == NULL) {
903 904 ASSERT0(tqs->stqs_count);
904 905 return;
905 906 }
906 907
907 908 for (uint_t i = 0; i < tqs->stqs_count; i++) {
908 909 ASSERT3P(tqs->stqs_taskq[i], !=, NULL);
909 910 taskq_destroy(tqs->stqs_taskq[i]);
910 911 }
911 912
912 913 kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *));
913 914 tqs->stqs_taskq = NULL;
914 915 }
915 916
916 917 /*
917 918 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
918 919 * Note that a type may have multiple discrete taskqs to avoid lock contention
919 920 * on the taskq itself. In that case we choose which taskq at random by using
920 921 * the low bits of gethrtime().
921 922 */
922 923 void
923 924 spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
924 925 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent)
925 926 {
926 927 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
927 928 taskq_t *tq;
928 929
929 930 ASSERT3P(tqs->stqs_taskq, !=, NULL);
930 931 ASSERT3U(tqs->stqs_count, !=, 0);
931 932
932 933 if (tqs->stqs_count == 1) {
933 934 tq = tqs->stqs_taskq[0];
934 935 } else {
935 936 tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count];
936 937 }
937 938
938 939 taskq_dispatch_ent(tq, func, arg, flags, ent);
939 940 }
940 941
941 942 static void
942 943 spa_create_zio_taskqs(spa_t *spa)
943 944 {
944 945 for (int t = 0; t < ZIO_TYPES; t++) {
945 946 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
946 947 spa_taskqs_init(spa, t, q);
947 948 }
948 949 }
949 950 }
950 951
951 952 #ifdef _KERNEL
952 953 static void
953 954 spa_thread(void *arg)
954 955 {
955 956 callb_cpr_t cprinfo;
956 957
957 958 spa_t *spa = arg;
958 959 user_t *pu = PTOU(curproc);
959 960
960 961 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
961 962 spa->spa_name);
962 963
963 964 ASSERT(curproc != &p0);
964 965 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
965 966 "zpool-%s", spa->spa_name);
966 967 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
967 968
968 969 /* bind this thread to the requested psrset */
969 970 if (zio_taskq_psrset_bind != PS_NONE) {
970 971 pool_lock();
971 972 mutex_enter(&cpu_lock);
972 973 mutex_enter(&pidlock);
973 974 mutex_enter(&curproc->p_lock);
974 975
975 976 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
976 977 0, NULL, NULL) == 0) {
977 978 curthread->t_bind_pset = zio_taskq_psrset_bind;
978 979 } else {
979 980 cmn_err(CE_WARN,
980 981 "Couldn't bind process for zfs pool \"%s\" to "
981 982 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
982 983 }
983 984
984 985 mutex_exit(&curproc->p_lock);
985 986 mutex_exit(&pidlock);
986 987 mutex_exit(&cpu_lock);
987 988 pool_unlock();
988 989 }
989 990
990 991 if (zio_taskq_sysdc) {
991 992 sysdc_thread_enter(curthread, 100, 0);
992 993 }
993 994
994 995 spa->spa_proc = curproc;
995 996 spa->spa_did = curthread->t_did;
996 997
997 998 spa_create_zio_taskqs(spa);
998 999
999 1000 mutex_enter(&spa->spa_proc_lock);
1000 1001 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
1001 1002
1002 1003 spa->spa_proc_state = SPA_PROC_ACTIVE;
1003 1004 cv_broadcast(&spa->spa_proc_cv);
1004 1005
1005 1006 CALLB_CPR_SAFE_BEGIN(&cprinfo);
1006 1007 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
1007 1008 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1008 1009 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
1009 1010
1010 1011 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
1011 1012 spa->spa_proc_state = SPA_PROC_GONE;
1012 1013 spa->spa_proc = &p0;
1013 1014 cv_broadcast(&spa->spa_proc_cv);
1014 1015 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
1015 1016
1016 1017 mutex_enter(&curproc->p_lock);
1017 1018 lwp_exit();
1018 1019 }
1019 1020 #endif
1020 1021
1021 1022 /*
1022 1023 * Activate an uninitialized pool.
1023 1024 */
1024 1025 static void
1025 1026 spa_activate(spa_t *spa, int mode)
1026 1027 {
1027 1028 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
1028 1029
1029 1030 spa->spa_state = POOL_STATE_ACTIVE;
1030 1031 spa->spa_mode = mode;
1031 1032
1032 1033 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
1033 1034 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
1034 1035
1035 1036 /* Try to create a covering process */
1036 1037 mutex_enter(&spa->spa_proc_lock);
1037 1038 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
1038 1039 ASSERT(spa->spa_proc == &p0);
1039 1040 spa->spa_did = 0;
1040 1041
1041 1042 /* Only create a process if we're going to be around a while. */
1042 1043 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
1043 1044 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
1044 1045 NULL, 0) == 0) {
1045 1046 spa->spa_proc_state = SPA_PROC_CREATED;
1046 1047 while (spa->spa_proc_state == SPA_PROC_CREATED) {
1047 1048 cv_wait(&spa->spa_proc_cv,
1048 1049 &spa->spa_proc_lock);
1049 1050 }
1050 1051 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1051 1052 ASSERT(spa->spa_proc != &p0);
1052 1053 ASSERT(spa->spa_did != 0);
1053 1054 } else {
1054 1055 #ifdef _KERNEL
1055 1056 cmn_err(CE_WARN,
1056 1057 "Couldn't create process for zfs pool \"%s\"\n",
1057 1058 spa->spa_name);
1058 1059 #endif
1059 1060 }
1060 1061 }
1061 1062 mutex_exit(&spa->spa_proc_lock);
1062 1063
1063 1064 /* If we didn't create a process, we need to create our taskqs. */
1064 1065 if (spa->spa_proc == &p0) {
1065 1066 spa_create_zio_taskqs(spa);
1066 1067 }
1067 1068
1068 1069 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1069 1070 offsetof(vdev_t, vdev_config_dirty_node));
1070 1071 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1071 1072 offsetof(vdev_t, vdev_state_dirty_node));
1072 1073
1073 1074 txg_list_create(&spa->spa_vdev_txg_list,
1074 1075 offsetof(struct vdev, vdev_txg_node));
1075 1076
1076 1077 avl_create(&spa->spa_errlist_scrub,
1077 1078 spa_error_entry_compare, sizeof (spa_error_entry_t),
1078 1079 offsetof(spa_error_entry_t, se_avl));
1079 1080 avl_create(&spa->spa_errlist_last,
1080 1081 spa_error_entry_compare, sizeof (spa_error_entry_t),
1081 1082 offsetof(spa_error_entry_t, se_avl));
1082 1083 }
1083 1084
1084 1085 /*
1085 1086 * Opposite of spa_activate().
1086 1087 */
1087 1088 static void
1088 1089 spa_deactivate(spa_t *spa)
1089 1090 {
1090 1091 ASSERT(spa->spa_sync_on == B_FALSE);
1091 1092 ASSERT(spa->spa_dsl_pool == NULL);
1092 1093 ASSERT(spa->spa_root_vdev == NULL);
1093 1094 ASSERT(spa->spa_async_zio_root == NULL);
1094 1095 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1095 1096
1096 1097 txg_list_destroy(&spa->spa_vdev_txg_list);
1097 1098
1098 1099 list_destroy(&spa->spa_config_dirty_list);
1099 1100 list_destroy(&spa->spa_state_dirty_list);
1100 1101
1101 1102 for (int t = 0; t < ZIO_TYPES; t++) {
1102 1103 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1103 1104 spa_taskqs_fini(spa, t, q);
1104 1105 }
1105 1106 }
1106 1107
1107 1108 metaslab_class_destroy(spa->spa_normal_class);
1108 1109 spa->spa_normal_class = NULL;
1109 1110
1110 1111 metaslab_class_destroy(spa->spa_log_class);
1111 1112 spa->spa_log_class = NULL;
1112 1113
1113 1114 /*
1114 1115 * If this was part of an import or the open otherwise failed, we may
1115 1116 * still have errors left in the queues. Empty them just in case.
1116 1117 */
1117 1118 spa_errlog_drain(spa);
1118 1119
1119 1120 avl_destroy(&spa->spa_errlist_scrub);
1120 1121 avl_destroy(&spa->spa_errlist_last);
1121 1122
1122 1123 spa->spa_state = POOL_STATE_UNINITIALIZED;
1123 1124
1124 1125 mutex_enter(&spa->spa_proc_lock);
1125 1126 if (spa->spa_proc_state != SPA_PROC_NONE) {
1126 1127 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1127 1128 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1128 1129 cv_broadcast(&spa->spa_proc_cv);
1129 1130 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1130 1131 ASSERT(spa->spa_proc != &p0);
1131 1132 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1132 1133 }
1133 1134 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1134 1135 spa->spa_proc_state = SPA_PROC_NONE;
1135 1136 }
1136 1137 ASSERT(spa->spa_proc == &p0);
1137 1138 mutex_exit(&spa->spa_proc_lock);
1138 1139
1139 1140 /*
1140 1141 * We want to make sure spa_thread() has actually exited the ZFS
1141 1142 * module, so that the module can't be unloaded out from underneath
1142 1143 * it.
1143 1144 */
1144 1145 if (spa->spa_did != 0) {
1145 1146 thread_join(spa->spa_did);
1146 1147 spa->spa_did = 0;
1147 1148 }
1148 1149 }
1149 1150
1150 1151 /*
1151 1152 * Verify a pool configuration, and construct the vdev tree appropriately. This
1152 1153 * will create all the necessary vdevs in the appropriate layout, with each vdev
1153 1154 * in the CLOSED state. This will prep the pool before open/creation/import.
1154 1155 * All vdev validation is done by the vdev_alloc() routine.
1155 1156 */
1156 1157 static int
1157 1158 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1158 1159 uint_t id, int atype)
1159 1160 {
1160 1161 nvlist_t **child;
1161 1162 uint_t children;
1162 1163 int error;
1163 1164
1164 1165 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1165 1166 return (error);
1166 1167
1167 1168 if ((*vdp)->vdev_ops->vdev_op_leaf)
1168 1169 return (0);
1169 1170
1170 1171 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1171 1172 &child, &children);
1172 1173
1173 1174 if (error == ENOENT)
1174 1175 return (0);
1175 1176
1176 1177 if (error) {
1177 1178 vdev_free(*vdp);
1178 1179 *vdp = NULL;
1179 1180 return (SET_ERROR(EINVAL));
1180 1181 }
1181 1182
1182 1183 for (int c = 0; c < children; c++) {
1183 1184 vdev_t *vd;
1184 1185 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1185 1186 atype)) != 0) {
1186 1187 vdev_free(*vdp);
1187 1188 *vdp = NULL;
1188 1189 return (error);
1189 1190 }
1190 1191 }
1191 1192
1192 1193 ASSERT(*vdp != NULL);
1193 1194
1194 1195 return (0);
1195 1196 }
1196 1197
1197 1198 /*
1198 1199 * Opposite of spa_load().
1199 1200 */
1200 1201 static void
1201 1202 spa_unload(spa_t *spa)
1202 1203 {
1203 1204 int i;
1204 1205
1205 1206 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1206 1207
1207 1208 /*
1208 1209 * Stop async tasks.
1209 1210 */
1210 1211 spa_async_suspend(spa);
1211 1212
1212 1213 /*
1213 1214 * Stop syncing.
1214 1215 */
1215 1216 if (spa->spa_sync_on) {
1216 1217 txg_sync_stop(spa->spa_dsl_pool);
1217 1218 spa->spa_sync_on = B_FALSE;
1218 1219 }
1219 1220
1220 1221 /*
1221 1222 * Wait for any outstanding async I/O to complete.
1222 1223 */
1223 1224 if (spa->spa_async_zio_root != NULL) {
1224 1225 (void) zio_wait(spa->spa_async_zio_root);
1225 1226 spa->spa_async_zio_root = NULL;
1226 1227 }
1227 1228
1228 1229 bpobj_close(&spa->spa_deferred_bpobj);
1229 1230
1230 1231 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1231 1232
1232 1233 /*
1233 1234 * Close all vdevs.
1234 1235 */
1235 1236 if (spa->spa_root_vdev)
1236 1237 vdev_free(spa->spa_root_vdev);
1237 1238 ASSERT(spa->spa_root_vdev == NULL);
1238 1239
1239 1240 /*
1240 1241 * Close the dsl pool.
1241 1242 */
1242 1243 if (spa->spa_dsl_pool) {
1243 1244 dsl_pool_close(spa->spa_dsl_pool);
1244 1245 spa->spa_dsl_pool = NULL;
1245 1246 spa->spa_meta_objset = NULL;
1246 1247 }
1247 1248
1248 1249 ddt_unload(spa);
1249 1250
1250 1251
1251 1252 /*
1252 1253 * Drop and purge level 2 cache
1253 1254 */
1254 1255 spa_l2cache_drop(spa);
1255 1256
1256 1257 for (i = 0; i < spa->spa_spares.sav_count; i++)
1257 1258 vdev_free(spa->spa_spares.sav_vdevs[i]);
1258 1259 if (spa->spa_spares.sav_vdevs) {
1259 1260 kmem_free(spa->spa_spares.sav_vdevs,
1260 1261 spa->spa_spares.sav_count * sizeof (void *));
1261 1262 spa->spa_spares.sav_vdevs = NULL;
1262 1263 }
1263 1264 if (spa->spa_spares.sav_config) {
1264 1265 nvlist_free(spa->spa_spares.sav_config);
1265 1266 spa->spa_spares.sav_config = NULL;
1266 1267 }
1267 1268 spa->spa_spares.sav_count = 0;
1268 1269
1269 1270 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1270 1271 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1271 1272 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1272 1273 }
1273 1274 if (spa->spa_l2cache.sav_vdevs) {
1274 1275 kmem_free(spa->spa_l2cache.sav_vdevs,
1275 1276 spa->spa_l2cache.sav_count * sizeof (void *));
1276 1277 spa->spa_l2cache.sav_vdevs = NULL;
1277 1278 }
1278 1279 if (spa->spa_l2cache.sav_config) {
1279 1280 nvlist_free(spa->spa_l2cache.sav_config);
1280 1281 spa->spa_l2cache.sav_config = NULL;
1281 1282 }
1282 1283 spa->spa_l2cache.sav_count = 0;
1283 1284
1284 1285 spa->spa_async_suspended = 0;
1285 1286
1286 1287 if (spa->spa_comment != NULL) {
1287 1288 spa_strfree(spa->spa_comment);
1288 1289 spa->spa_comment = NULL;
1289 1290 }
1290 1291
1291 1292 spa_config_exit(spa, SCL_ALL, FTAG);
1292 1293 }
1293 1294
1294 1295 /*
1295 1296 * Load (or re-load) the current list of vdevs describing the active spares for
1296 1297 * this pool. When this is called, we have some form of basic information in
1297 1298 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1298 1299 * then re-generate a more complete list including status information.
1299 1300 */
1300 1301 static void
1301 1302 spa_load_spares(spa_t *spa)
1302 1303 {
1303 1304 nvlist_t **spares;
1304 1305 uint_t nspares;
1305 1306 int i;
1306 1307 vdev_t *vd, *tvd;
1307 1308
1308 1309 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1309 1310
1310 1311 /*
1311 1312 * First, close and free any existing spare vdevs.
1312 1313 */
1313 1314 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1314 1315 vd = spa->spa_spares.sav_vdevs[i];
1315 1316
1316 1317 /* Undo the call to spa_activate() below */
1317 1318 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1318 1319 B_FALSE)) != NULL && tvd->vdev_isspare)
1319 1320 spa_spare_remove(tvd);
1320 1321 vdev_close(vd);
1321 1322 vdev_free(vd);
1322 1323 }
1323 1324
1324 1325 if (spa->spa_spares.sav_vdevs)
1325 1326 kmem_free(spa->spa_spares.sav_vdevs,
1326 1327 spa->spa_spares.sav_count * sizeof (void *));
1327 1328
1328 1329 if (spa->spa_spares.sav_config == NULL)
1329 1330 nspares = 0;
1330 1331 else
1331 1332 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1332 1333 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1333 1334
1334 1335 spa->spa_spares.sav_count = (int)nspares;
1335 1336 spa->spa_spares.sav_vdevs = NULL;
1336 1337
1337 1338 if (nspares == 0)
1338 1339 return;
1339 1340
1340 1341 /*
1341 1342 * Construct the array of vdevs, opening them to get status in the
1342 1343 * process. For each spare, there is potentially two different vdev_t
1343 1344 * structures associated with it: one in the list of spares (used only
1344 1345 * for basic validation purposes) and one in the active vdev
1345 1346 * configuration (if it's spared in). During this phase we open and
1346 1347 * validate each vdev on the spare list. If the vdev also exists in the
1347 1348 * active configuration, then we also mark this vdev as an active spare.
1348 1349 */
1349 1350 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1350 1351 KM_SLEEP);
1351 1352 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1352 1353 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1353 1354 VDEV_ALLOC_SPARE) == 0);
1354 1355 ASSERT(vd != NULL);
1355 1356
1356 1357 spa->spa_spares.sav_vdevs[i] = vd;
1357 1358
1358 1359 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1359 1360 B_FALSE)) != NULL) {
1360 1361 if (!tvd->vdev_isspare)
1361 1362 spa_spare_add(tvd);
1362 1363
1363 1364 /*
1364 1365 * We only mark the spare active if we were successfully
1365 1366 * able to load the vdev. Otherwise, importing a pool
1366 1367 * with a bad active spare would result in strange
1367 1368 * behavior, because multiple pool would think the spare
1368 1369 * is actively in use.
1369 1370 *
1370 1371 * There is a vulnerability here to an equally bizarre
1371 1372 * circumstance, where a dead active spare is later
1372 1373 * brought back to life (onlined or otherwise). Given
1373 1374 * the rarity of this scenario, and the extra complexity
1374 1375 * it adds, we ignore the possibility.
1375 1376 */
1376 1377 if (!vdev_is_dead(tvd))
1377 1378 spa_spare_activate(tvd);
1378 1379 }
1379 1380
1380 1381 vd->vdev_top = vd;
1381 1382 vd->vdev_aux = &spa->spa_spares;
1382 1383
1383 1384 if (vdev_open(vd) != 0)
1384 1385 continue;
1385 1386
1386 1387 if (vdev_validate_aux(vd) == 0)
1387 1388 spa_spare_add(vd);
1388 1389 }
1389 1390
1390 1391 /*
1391 1392 * Recompute the stashed list of spares, with status information
1392 1393 * this time.
1393 1394 */
1394 1395 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1395 1396 DATA_TYPE_NVLIST_ARRAY) == 0);
1396 1397
1397 1398 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1398 1399 KM_SLEEP);
1399 1400 for (i = 0; i < spa->spa_spares.sav_count; i++)
1400 1401 spares[i] = vdev_config_generate(spa,
1401 1402 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1402 1403 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1403 1404 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1404 1405 for (i = 0; i < spa->spa_spares.sav_count; i++)
1405 1406 nvlist_free(spares[i]);
1406 1407 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1407 1408 }
1408 1409
1409 1410 /*
1410 1411 * Load (or re-load) the current list of vdevs describing the active l2cache for
1411 1412 * this pool. When this is called, we have some form of basic information in
1412 1413 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1413 1414 * then re-generate a more complete list including status information.
1414 1415 * Devices which are already active have their details maintained, and are
1415 1416 * not re-opened.
1416 1417 */
1417 1418 static void
1418 1419 spa_load_l2cache(spa_t *spa)
1419 1420 {
1420 1421 nvlist_t **l2cache;
1421 1422 uint_t nl2cache;
1422 1423 int i, j, oldnvdevs;
1423 1424 uint64_t guid;
1424 1425 vdev_t *vd, **oldvdevs, **newvdevs;
1425 1426 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1426 1427
1427 1428 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1428 1429
1429 1430 if (sav->sav_config != NULL) {
1430 1431 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1431 1432 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1432 1433 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1433 1434 } else {
1434 1435 nl2cache = 0;
1435 1436 newvdevs = NULL;
1436 1437 }
1437 1438
1438 1439 oldvdevs = sav->sav_vdevs;
1439 1440 oldnvdevs = sav->sav_count;
1440 1441 sav->sav_vdevs = NULL;
1441 1442 sav->sav_count = 0;
1442 1443
1443 1444 /*
1444 1445 * Process new nvlist of vdevs.
1445 1446 */
1446 1447 for (i = 0; i < nl2cache; i++) {
1447 1448 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1448 1449 &guid) == 0);
1449 1450
1450 1451 newvdevs[i] = NULL;
1451 1452 for (j = 0; j < oldnvdevs; j++) {
1452 1453 vd = oldvdevs[j];
1453 1454 if (vd != NULL && guid == vd->vdev_guid) {
1454 1455 /*
1455 1456 * Retain previous vdev for add/remove ops.
1456 1457 */
1457 1458 newvdevs[i] = vd;
1458 1459 oldvdevs[j] = NULL;
1459 1460 break;
1460 1461 }
1461 1462 }
1462 1463
1463 1464 if (newvdevs[i] == NULL) {
1464 1465 /*
1465 1466 * Create new vdev
1466 1467 */
1467 1468 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1468 1469 VDEV_ALLOC_L2CACHE) == 0);
1469 1470 ASSERT(vd != NULL);
1470 1471 newvdevs[i] = vd;
1471 1472
1472 1473 /*
1473 1474 * Commit this vdev as an l2cache device,
1474 1475 * even if it fails to open.
1475 1476 */
1476 1477 spa_l2cache_add(vd);
1477 1478
1478 1479 vd->vdev_top = vd;
1479 1480 vd->vdev_aux = sav;
1480 1481
1481 1482 spa_l2cache_activate(vd);
1482 1483
1483 1484 if (vdev_open(vd) != 0)
1484 1485 continue;
1485 1486
1486 1487 (void) vdev_validate_aux(vd);
1487 1488
1488 1489 if (!vdev_is_dead(vd))
1489 1490 l2arc_add_vdev(spa, vd);
1490 1491 }
1491 1492 }
1492 1493
1493 1494 /*
1494 1495 * Purge vdevs that were dropped
1495 1496 */
1496 1497 for (i = 0; i < oldnvdevs; i++) {
1497 1498 uint64_t pool;
1498 1499
1499 1500 vd = oldvdevs[i];
1500 1501 if (vd != NULL) {
1501 1502 ASSERT(vd->vdev_isl2cache);
1502 1503
1503 1504 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1504 1505 pool != 0ULL && l2arc_vdev_present(vd))
1505 1506 l2arc_remove_vdev(vd);
1506 1507 vdev_clear_stats(vd);
1507 1508 vdev_free(vd);
1508 1509 }
1509 1510 }
1510 1511
1511 1512 if (oldvdevs)
1512 1513 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1513 1514
1514 1515 if (sav->sav_config == NULL)
1515 1516 goto out;
1516 1517
1517 1518 sav->sav_vdevs = newvdevs;
1518 1519 sav->sav_count = (int)nl2cache;
1519 1520
1520 1521 /*
1521 1522 * Recompute the stashed list of l2cache devices, with status
1522 1523 * information this time.
1523 1524 */
1524 1525 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1525 1526 DATA_TYPE_NVLIST_ARRAY) == 0);
1526 1527
1527 1528 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1528 1529 for (i = 0; i < sav->sav_count; i++)
1529 1530 l2cache[i] = vdev_config_generate(spa,
1530 1531 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1531 1532 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1532 1533 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1533 1534 out:
1534 1535 for (i = 0; i < sav->sav_count; i++)
1535 1536 nvlist_free(l2cache[i]);
1536 1537 if (sav->sav_count)
1537 1538 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1538 1539 }
1539 1540
1540 1541 static int
1541 1542 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1542 1543 {
1543 1544 dmu_buf_t *db;
1544 1545 char *packed = NULL;
1545 1546 size_t nvsize = 0;
1546 1547 int error;
1547 1548 *value = NULL;
1548 1549
1549 1550 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1550 1551 nvsize = *(uint64_t *)db->db_data;
1551 1552 dmu_buf_rele(db, FTAG);
1552 1553
1553 1554 packed = kmem_alloc(nvsize, KM_SLEEP);
1554 1555 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1555 1556 DMU_READ_PREFETCH);
1556 1557 if (error == 0)
1557 1558 error = nvlist_unpack(packed, nvsize, value, 0);
1558 1559 kmem_free(packed, nvsize);
1559 1560
1560 1561 return (error);
1561 1562 }
1562 1563
1563 1564 /*
1564 1565 * Checks to see if the given vdev could not be opened, in which case we post a
1565 1566 * sysevent to notify the autoreplace code that the device has been removed.
1566 1567 */
1567 1568 static void
1568 1569 spa_check_removed(vdev_t *vd)
1569 1570 {
1570 1571 for (int c = 0; c < vd->vdev_children; c++)
1571 1572 spa_check_removed(vd->vdev_child[c]);
1572 1573
1573 1574 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1574 1575 !vd->vdev_ishole) {
1575 1576 zfs_post_autoreplace(vd->vdev_spa, vd);
1576 1577 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1577 1578 }
1578 1579 }
1579 1580
1580 1581 /*
1581 1582 * Validate the current config against the MOS config
1582 1583 */
1583 1584 static boolean_t
1584 1585 spa_config_valid(spa_t *spa, nvlist_t *config)
1585 1586 {
1586 1587 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1587 1588 nvlist_t *nv;
1588 1589
1589 1590 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1590 1591
1591 1592 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1592 1593 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1593 1594
1594 1595 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1595 1596
1596 1597 /*
1597 1598 * If we're doing a normal import, then build up any additional
1598 1599 * diagnostic information about missing devices in this config.
1599 1600 * We'll pass this up to the user for further processing.
1600 1601 */
1601 1602 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1602 1603 nvlist_t **child, *nv;
1603 1604 uint64_t idx = 0;
1604 1605
1605 1606 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1606 1607 KM_SLEEP);
1607 1608 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1608 1609
1609 1610 for (int c = 0; c < rvd->vdev_children; c++) {
1610 1611 vdev_t *tvd = rvd->vdev_child[c];
1611 1612 vdev_t *mtvd = mrvd->vdev_child[c];
1612 1613
1613 1614 if (tvd->vdev_ops == &vdev_missing_ops &&
1614 1615 mtvd->vdev_ops != &vdev_missing_ops &&
1615 1616 mtvd->vdev_islog)
1616 1617 child[idx++] = vdev_config_generate(spa, mtvd,
1617 1618 B_FALSE, 0);
1618 1619 }
1619 1620
1620 1621 if (idx) {
1621 1622 VERIFY(nvlist_add_nvlist_array(nv,
1622 1623 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1623 1624 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1624 1625 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1625 1626
1626 1627 for (int i = 0; i < idx; i++)
1627 1628 nvlist_free(child[i]);
1628 1629 }
1629 1630 nvlist_free(nv);
1630 1631 kmem_free(child, rvd->vdev_children * sizeof (char **));
1631 1632 }
1632 1633
1633 1634 /*
1634 1635 * Compare the root vdev tree with the information we have
1635 1636 * from the MOS config (mrvd). Check each top-level vdev
1636 1637 * with the corresponding MOS config top-level (mtvd).
1637 1638 */
1638 1639 for (int c = 0; c < rvd->vdev_children; c++) {
1639 1640 vdev_t *tvd = rvd->vdev_child[c];
1640 1641 vdev_t *mtvd = mrvd->vdev_child[c];
1641 1642
1642 1643 /*
1643 1644 * Resolve any "missing" vdevs in the current configuration.
1644 1645 * If we find that the MOS config has more accurate information
1645 1646 * about the top-level vdev then use that vdev instead.
1646 1647 */
1647 1648 if (tvd->vdev_ops == &vdev_missing_ops &&
1648 1649 mtvd->vdev_ops != &vdev_missing_ops) {
1649 1650
1650 1651 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1651 1652 continue;
1652 1653
1653 1654 /*
1654 1655 * Device specific actions.
1655 1656 */
1656 1657 if (mtvd->vdev_islog) {
1657 1658 spa_set_log_state(spa, SPA_LOG_CLEAR);
1658 1659 } else {
1659 1660 /*
1660 1661 * XXX - once we have 'readonly' pool
1661 1662 * support we should be able to handle
1662 1663 * missing data devices by transitioning
1663 1664 * the pool to readonly.
1664 1665 */
1665 1666 continue;
1666 1667 }
1667 1668
1668 1669 /*
1669 1670 * Swap the missing vdev with the data we were
1670 1671 * able to obtain from the MOS config.
1671 1672 */
1672 1673 vdev_remove_child(rvd, tvd);
1673 1674 vdev_remove_child(mrvd, mtvd);
1674 1675
1675 1676 vdev_add_child(rvd, mtvd);
1676 1677 vdev_add_child(mrvd, tvd);
1677 1678
1678 1679 spa_config_exit(spa, SCL_ALL, FTAG);
1679 1680 vdev_load(mtvd);
1680 1681 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1681 1682
1682 1683 vdev_reopen(rvd);
1683 1684 } else if (mtvd->vdev_islog) {
1684 1685 /*
1685 1686 * Load the slog device's state from the MOS config
1686 1687 * since it's possible that the label does not
1687 1688 * contain the most up-to-date information.
1688 1689 */
1689 1690 vdev_load_log_state(tvd, mtvd);
1690 1691 vdev_reopen(tvd);
1691 1692 }
1692 1693 }
1693 1694 vdev_free(mrvd);
1694 1695 spa_config_exit(spa, SCL_ALL, FTAG);
1695 1696
1696 1697 /*
1697 1698 * Ensure we were able to validate the config.
1698 1699 */
1699 1700 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1700 1701 }
1701 1702
1702 1703 /*
1703 1704 * Check for missing log devices
1704 1705 */
1705 1706 static boolean_t
1706 1707 spa_check_logs(spa_t *spa)
1707 1708 {
1708 1709 boolean_t rv = B_FALSE;
1709 1710
1710 1711 switch (spa->spa_log_state) {
1711 1712 case SPA_LOG_MISSING:
1712 1713 /* need to recheck in case slog has been restored */
1713 1714 case SPA_LOG_UNKNOWN:
1714 1715 rv = (dmu_objset_find(spa->spa_name, zil_check_log_chain,
1715 1716 NULL, DS_FIND_CHILDREN) != 0);
1716 1717 if (rv)
1717 1718 spa_set_log_state(spa, SPA_LOG_MISSING);
1718 1719 break;
1719 1720 }
1720 1721 return (rv);
1721 1722 }
1722 1723
1723 1724 static boolean_t
1724 1725 spa_passivate_log(spa_t *spa)
1725 1726 {
1726 1727 vdev_t *rvd = spa->spa_root_vdev;
1727 1728 boolean_t slog_found = B_FALSE;
1728 1729
1729 1730 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1730 1731
1731 1732 if (!spa_has_slogs(spa))
1732 1733 return (B_FALSE);
1733 1734
1734 1735 for (int c = 0; c < rvd->vdev_children; c++) {
1735 1736 vdev_t *tvd = rvd->vdev_child[c];
1736 1737 metaslab_group_t *mg = tvd->vdev_mg;
1737 1738
1738 1739 if (tvd->vdev_islog) {
1739 1740 metaslab_group_passivate(mg);
1740 1741 slog_found = B_TRUE;
1741 1742 }
1742 1743 }
1743 1744
1744 1745 return (slog_found);
1745 1746 }
1746 1747
1747 1748 static void
1748 1749 spa_activate_log(spa_t *spa)
1749 1750 {
1750 1751 vdev_t *rvd = spa->spa_root_vdev;
1751 1752
1752 1753 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1753 1754
1754 1755 for (int c = 0; c < rvd->vdev_children; c++) {
1755 1756 vdev_t *tvd = rvd->vdev_child[c];
1756 1757 metaslab_group_t *mg = tvd->vdev_mg;
1757 1758
1758 1759 if (tvd->vdev_islog)
1759 1760 metaslab_group_activate(mg);
1760 1761 }
1761 1762 }
1762 1763
1763 1764 int
1764 1765 spa_offline_log(spa_t *spa)
1765 1766 {
1766 1767 int error;
1767 1768
1768 1769 error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1769 1770 NULL, DS_FIND_CHILDREN);
1770 1771 if (error == 0) {
1771 1772 /*
1772 1773 * We successfully offlined the log device, sync out the
1773 1774 * current txg so that the "stubby" block can be removed
1774 1775 * by zil_sync().
1775 1776 */
1776 1777 txg_wait_synced(spa->spa_dsl_pool, 0);
1777 1778 }
1778 1779 return (error);
1779 1780 }
1780 1781
1781 1782 static void
1782 1783 spa_aux_check_removed(spa_aux_vdev_t *sav)
1783 1784 {
1784 1785 for (int i = 0; i < sav->sav_count; i++)
1785 1786 spa_check_removed(sav->sav_vdevs[i]);
1786 1787 }
1787 1788
1788 1789 void
1789 1790 spa_claim_notify(zio_t *zio)
1790 1791 {
1791 1792 spa_t *spa = zio->io_spa;
1792 1793
1793 1794 if (zio->io_error)
1794 1795 return;
1795 1796
1796 1797 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1797 1798 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1798 1799 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1799 1800 mutex_exit(&spa->spa_props_lock);
1800 1801 }
1801 1802
1802 1803 typedef struct spa_load_error {
1803 1804 uint64_t sle_meta_count;
1804 1805 uint64_t sle_data_count;
1805 1806 } spa_load_error_t;
1806 1807
1807 1808 static void
1808 1809 spa_load_verify_done(zio_t *zio)
1809 1810 {
1810 1811 blkptr_t *bp = zio->io_bp;
1811 1812 spa_load_error_t *sle = zio->io_private;
1812 1813 dmu_object_type_t type = BP_GET_TYPE(bp);
1813 1814 int error = zio->io_error;
1814 1815
1815 1816 if (error) {
1816 1817 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1817 1818 type != DMU_OT_INTENT_LOG)
1818 1819 atomic_add_64(&sle->sle_meta_count, 1);
1819 1820 else
1820 1821 atomic_add_64(&sle->sle_data_count, 1);
1821 1822 }
1822 1823 zio_data_buf_free(zio->io_data, zio->io_size);
1823 1824 }
1824 1825
1825 1826 /*ARGSUSED*/
1826 1827 static int
1827 1828 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1828 1829 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1829 1830 {
1830 1831 if (bp != NULL) {
1831 1832 zio_t *rio = arg;
1832 1833 size_t size = BP_GET_PSIZE(bp);
1833 1834 void *data = zio_data_buf_alloc(size);
1834 1835
1835 1836 zio_nowait(zio_read(rio, spa, bp, data, size,
1836 1837 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1837 1838 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1838 1839 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1839 1840 }
1840 1841 return (0);
1841 1842 }
1842 1843
1843 1844 static int
1844 1845 spa_load_verify(spa_t *spa)
1845 1846 {
1846 1847 zio_t *rio;
1847 1848 spa_load_error_t sle = { 0 };
1848 1849 zpool_rewind_policy_t policy;
1849 1850 boolean_t verify_ok = B_FALSE;
1850 1851 int error;
1851 1852
1852 1853 zpool_get_rewind_policy(spa->spa_config, &policy);
1853 1854
1854 1855 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1855 1856 return (0);
1856 1857
1857 1858 rio = zio_root(spa, NULL, &sle,
1858 1859 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1859 1860
1860 1861 error = traverse_pool(spa, spa->spa_verify_min_txg,
1861 1862 TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1862 1863
1863 1864 (void) zio_wait(rio);
1864 1865
1865 1866 spa->spa_load_meta_errors = sle.sle_meta_count;
1866 1867 spa->spa_load_data_errors = sle.sle_data_count;
1867 1868
1868 1869 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1869 1870 sle.sle_data_count <= policy.zrp_maxdata) {
1870 1871 int64_t loss = 0;
1871 1872
1872 1873 verify_ok = B_TRUE;
1873 1874 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1874 1875 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1875 1876
1876 1877 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1877 1878 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1878 1879 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1879 1880 VERIFY(nvlist_add_int64(spa->spa_load_info,
1880 1881 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1881 1882 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1882 1883 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1883 1884 } else {
1884 1885 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1885 1886 }
1886 1887
1887 1888 if (error) {
1888 1889 if (error != ENXIO && error != EIO)
1889 1890 error = SET_ERROR(EIO);
1890 1891 return (error);
1891 1892 }
1892 1893
1893 1894 return (verify_ok ? 0 : EIO);
1894 1895 }
1895 1896
1896 1897 /*
1897 1898 * Find a value in the pool props object.
1898 1899 */
1899 1900 static void
1900 1901 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1901 1902 {
1902 1903 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1903 1904 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1904 1905 }
1905 1906
1906 1907 /*
1907 1908 * Find a value in the pool directory object.
1908 1909 */
1909 1910 static int
1910 1911 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1911 1912 {
1912 1913 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1913 1914 name, sizeof (uint64_t), 1, val));
1914 1915 }
1915 1916
1916 1917 static int
1917 1918 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1918 1919 {
1919 1920 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1920 1921 return (err);
1921 1922 }
1922 1923
1923 1924 /*
1924 1925 * Fix up config after a partly-completed split. This is done with the
1925 1926 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
1926 1927 * pool have that entry in their config, but only the splitting one contains
1927 1928 * a list of all the guids of the vdevs that are being split off.
1928 1929 *
1929 1930 * This function determines what to do with that list: either rejoin
1930 1931 * all the disks to the pool, or complete the splitting process. To attempt
1931 1932 * the rejoin, each disk that is offlined is marked online again, and
1932 1933 * we do a reopen() call. If the vdev label for every disk that was
1933 1934 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1934 1935 * then we call vdev_split() on each disk, and complete the split.
1935 1936 *
1936 1937 * Otherwise we leave the config alone, with all the vdevs in place in
1937 1938 * the original pool.
1938 1939 */
1939 1940 static void
1940 1941 spa_try_repair(spa_t *spa, nvlist_t *config)
1941 1942 {
1942 1943 uint_t extracted;
1943 1944 uint64_t *glist;
1944 1945 uint_t i, gcount;
1945 1946 nvlist_t *nvl;
1946 1947 vdev_t **vd;
1947 1948 boolean_t attempt_reopen;
1948 1949
1949 1950 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1950 1951 return;
1951 1952
1952 1953 /* check that the config is complete */
1953 1954 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1954 1955 &glist, &gcount) != 0)
1955 1956 return;
1956 1957
1957 1958 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1958 1959
1959 1960 /* attempt to online all the vdevs & validate */
1960 1961 attempt_reopen = B_TRUE;
1961 1962 for (i = 0; i < gcount; i++) {
1962 1963 if (glist[i] == 0) /* vdev is hole */
1963 1964 continue;
1964 1965
1965 1966 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1966 1967 if (vd[i] == NULL) {
1967 1968 /*
1968 1969 * Don't bother attempting to reopen the disks;
1969 1970 * just do the split.
1970 1971 */
1971 1972 attempt_reopen = B_FALSE;
1972 1973 } else {
1973 1974 /* attempt to re-online it */
1974 1975 vd[i]->vdev_offline = B_FALSE;
1975 1976 }
1976 1977 }
1977 1978
1978 1979 if (attempt_reopen) {
1979 1980 vdev_reopen(spa->spa_root_vdev);
1980 1981
1981 1982 /* check each device to see what state it's in */
1982 1983 for (extracted = 0, i = 0; i < gcount; i++) {
1983 1984 if (vd[i] != NULL &&
1984 1985 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1985 1986 break;
1986 1987 ++extracted;
1987 1988 }
1988 1989 }
1989 1990
1990 1991 /*
1991 1992 * If every disk has been moved to the new pool, or if we never
1992 1993 * even attempted to look at them, then we split them off for
1993 1994 * good.
1994 1995 */
1995 1996 if (!attempt_reopen || gcount == extracted) {
1996 1997 for (i = 0; i < gcount; i++)
1997 1998 if (vd[i] != NULL)
1998 1999 vdev_split(vd[i]);
1999 2000 vdev_reopen(spa->spa_root_vdev);
2000 2001 }
2001 2002
2002 2003 kmem_free(vd, gcount * sizeof (vdev_t *));
2003 2004 }
2004 2005
2005 2006 static int
2006 2007 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
2007 2008 boolean_t mosconfig)
2008 2009 {
2009 2010 nvlist_t *config = spa->spa_config;
2010 2011 char *ereport = FM_EREPORT_ZFS_POOL;
2011 2012 char *comment;
2012 2013 int error;
2013 2014 uint64_t pool_guid;
2014 2015 nvlist_t *nvl;
2015 2016
2016 2017 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
2017 2018 return (SET_ERROR(EINVAL));
2018 2019
2019 2020 ASSERT(spa->spa_comment == NULL);
2020 2021 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2021 2022 spa->spa_comment = spa_strdup(comment);
2022 2023
2023 2024 /*
2024 2025 * Versioning wasn't explicitly added to the label until later, so if
2025 2026 * it's not present treat it as the initial version.
2026 2027 */
2027 2028 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2028 2029 &spa->spa_ubsync.ub_version) != 0)
2029 2030 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2030 2031
2031 2032 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2032 2033 &spa->spa_config_txg);
2033 2034
2034 2035 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
2035 2036 spa_guid_exists(pool_guid, 0)) {
2036 2037 error = SET_ERROR(EEXIST);
2037 2038 } else {
2038 2039 spa->spa_config_guid = pool_guid;
2039 2040
2040 2041 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
2041 2042 &nvl) == 0) {
2042 2043 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
2043 2044 KM_SLEEP) == 0);
2044 2045 }
2045 2046
2046 2047 nvlist_free(spa->spa_load_info);
2047 2048 spa->spa_load_info = fnvlist_alloc();
2048 2049
2049 2050 gethrestime(&spa->spa_loaded_ts);
2050 2051 error = spa_load_impl(spa, pool_guid, config, state, type,
2051 2052 mosconfig, &ereport);
2052 2053 }
2053 2054
2054 2055 spa->spa_minref = refcount_count(&spa->spa_refcount);
2055 2056 if (error) {
2056 2057 if (error != EEXIST) {
2057 2058 spa->spa_loaded_ts.tv_sec = 0;
2058 2059 spa->spa_loaded_ts.tv_nsec = 0;
2059 2060 }
2060 2061 if (error != EBADF) {
2061 2062 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2062 2063 }
2063 2064 }
2064 2065 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2065 2066 spa->spa_ena = 0;
2066 2067
2067 2068 return (error);
2068 2069 }
2069 2070
2070 2071 /*
2071 2072 * Load an existing storage pool, using the pool's builtin spa_config as a
2072 2073 * source of configuration information.
2073 2074 */
2074 2075 static int
2075 2076 spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2076 2077 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2077 2078 char **ereport)
2078 2079 {
2079 2080 int error = 0;
2080 2081 nvlist_t *nvroot = NULL;
2081 2082 nvlist_t *label;
2082 2083 vdev_t *rvd;
2083 2084 uberblock_t *ub = &spa->spa_uberblock;
2084 2085 uint64_t children, config_cache_txg = spa->spa_config_txg;
2085 2086 int orig_mode = spa->spa_mode;
2086 2087 int parse;
2087 2088 uint64_t obj;
2088 2089 boolean_t missing_feat_write = B_FALSE;
2089 2090
2090 2091 /*
2091 2092 * If this is an untrusted config, access the pool in read-only mode.
2092 2093 * This prevents things like resilvering recently removed devices.
2093 2094 */
2094 2095 if (!mosconfig)
2095 2096 spa->spa_mode = FREAD;
2096 2097
2097 2098 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2098 2099
2099 2100 spa->spa_load_state = state;
2100 2101
2101 2102 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2102 2103 return (SET_ERROR(EINVAL));
2103 2104
2104 2105 parse = (type == SPA_IMPORT_EXISTING ?
2105 2106 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2106 2107
2107 2108 /*
2108 2109 * Create "The Godfather" zio to hold all async IOs
2109 2110 */
2110 2111 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2111 2112 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2112 2113
2113 2114 /*
2114 2115 * Parse the configuration into a vdev tree. We explicitly set the
2115 2116 * value that will be returned by spa_version() since parsing the
2116 2117 * configuration requires knowing the version number.
2117 2118 */
2118 2119 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2119 2120 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2120 2121 spa_config_exit(spa, SCL_ALL, FTAG);
2121 2122
2122 2123 if (error != 0)
2123 2124 return (error);
2124 2125
2125 2126 ASSERT(spa->spa_root_vdev == rvd);
2126 2127
2127 2128 if (type != SPA_IMPORT_ASSEMBLE) {
2128 2129 ASSERT(spa_guid(spa) == pool_guid);
2129 2130 }
2130 2131
2131 2132 /*
2132 2133 * Try to open all vdevs, loading each label in the process.
2133 2134 */
2134 2135 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2135 2136 error = vdev_open(rvd);
2136 2137 spa_config_exit(spa, SCL_ALL, FTAG);
2137 2138 if (error != 0)
2138 2139 return (error);
2139 2140
2140 2141 /*
2141 2142 * We need to validate the vdev labels against the configuration that
2142 2143 * we have in hand, which is dependent on the setting of mosconfig. If
2143 2144 * mosconfig is true then we're validating the vdev labels based on
2144 2145 * that config. Otherwise, we're validating against the cached config
2145 2146 * (zpool.cache) that was read when we loaded the zfs module, and then
2146 2147 * later we will recursively call spa_load() and validate against
2147 2148 * the vdev config.
2148 2149 *
2149 2150 * If we're assembling a new pool that's been split off from an
2150 2151 * existing pool, the labels haven't yet been updated so we skip
2151 2152 * validation for now.
2152 2153 */
2153 2154 if (type != SPA_IMPORT_ASSEMBLE) {
2154 2155 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2155 2156 error = vdev_validate(rvd, mosconfig);
2156 2157 spa_config_exit(spa, SCL_ALL, FTAG);
2157 2158
2158 2159 if (error != 0)
2159 2160 return (error);
2160 2161
2161 2162 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2162 2163 return (SET_ERROR(ENXIO));
2163 2164 }
2164 2165
2165 2166 /*
2166 2167 * Find the best uberblock.
2167 2168 */
2168 2169 vdev_uberblock_load(rvd, ub, &label);
2169 2170
2170 2171 /*
2171 2172 * If we weren't able to find a single valid uberblock, return failure.
2172 2173 */
2173 2174 if (ub->ub_txg == 0) {
2174 2175 nvlist_free(label);
2175 2176 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2176 2177 }
2177 2178
2178 2179 /*
2179 2180 * If the pool has an unsupported version we can't open it.
2180 2181 */
2181 2182 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2182 2183 nvlist_free(label);
2183 2184 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2184 2185 }
2185 2186
2186 2187 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2187 2188 nvlist_t *features;
2188 2189
2189 2190 /*
2190 2191 * If we weren't able to find what's necessary for reading the
2191 2192 * MOS in the label, return failure.
2192 2193 */
2193 2194 if (label == NULL || nvlist_lookup_nvlist(label,
2194 2195 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2195 2196 nvlist_free(label);
2196 2197 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2197 2198 ENXIO));
2198 2199 }
2199 2200
2200 2201 /*
2201 2202 * Update our in-core representation with the definitive values
2202 2203 * from the label.
2203 2204 */
2204 2205 nvlist_free(spa->spa_label_features);
2205 2206 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2206 2207 }
2207 2208
2208 2209 nvlist_free(label);
2209 2210
2210 2211 /*
2211 2212 * Look through entries in the label nvlist's features_for_read. If
2212 2213 * there is a feature listed there which we don't understand then we
2213 2214 * cannot open a pool.
2214 2215 */
2215 2216 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2216 2217 nvlist_t *unsup_feat;
2217 2218
2218 2219 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2219 2220 0);
2220 2221
2221 2222 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2222 2223 NULL); nvp != NULL;
2223 2224 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2224 2225 if (!zfeature_is_supported(nvpair_name(nvp))) {
2225 2226 VERIFY(nvlist_add_string(unsup_feat,
2226 2227 nvpair_name(nvp), "") == 0);
2227 2228 }
2228 2229 }
2229 2230
2230 2231 if (!nvlist_empty(unsup_feat)) {
2231 2232 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2232 2233 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2233 2234 nvlist_free(unsup_feat);
2234 2235 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2235 2236 ENOTSUP));
2236 2237 }
2237 2238
2238 2239 nvlist_free(unsup_feat);
2239 2240 }
2240 2241
2241 2242 /*
2242 2243 * If the vdev guid sum doesn't match the uberblock, we have an
2243 2244 * incomplete configuration. We first check to see if the pool
2244 2245 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2245 2246 * If it is, defer the vdev_guid_sum check till later so we
2246 2247 * can handle missing vdevs.
2247 2248 */
2248 2249 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2249 2250 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2250 2251 rvd->vdev_guid_sum != ub->ub_guid_sum)
2251 2252 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2252 2253
2253 2254 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2254 2255 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2255 2256 spa_try_repair(spa, config);
2256 2257 spa_config_exit(spa, SCL_ALL, FTAG);
2257 2258 nvlist_free(spa->spa_config_splitting);
2258 2259 spa->spa_config_splitting = NULL;
2259 2260 }
2260 2261
2261 2262 /*
2262 2263 * Initialize internal SPA structures.
2263 2264 */
2264 2265 spa->spa_state = POOL_STATE_ACTIVE;
2265 2266 spa->spa_ubsync = spa->spa_uberblock;
2266 2267 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2267 2268 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2268 2269 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2269 2270 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2270 2271 spa->spa_claim_max_txg = spa->spa_first_txg;
2271 2272 spa->spa_prev_software_version = ub->ub_software_version;
2272 2273
2273 2274 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2274 2275 if (error)
2275 2276 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2276 2277 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2277 2278
2278 2279 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2279 2280 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2280 2281
2281 2282 if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2282 2283 boolean_t missing_feat_read = B_FALSE;
2283 2284 nvlist_t *unsup_feat, *enabled_feat;
2284 2285
2285 2286 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2286 2287 &spa->spa_feat_for_read_obj) != 0) {
2287 2288 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2288 2289 }
2289 2290
2290 2291 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2291 2292 &spa->spa_feat_for_write_obj) != 0) {
2292 2293 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2293 2294 }
2294 2295
2295 2296 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2296 2297 &spa->spa_feat_desc_obj) != 0) {
2297 2298 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2298 2299 }
2299 2300
2300 2301 enabled_feat = fnvlist_alloc();
2301 2302 unsup_feat = fnvlist_alloc();
2302 2303
2303 2304 if (!feature_is_supported(spa->spa_meta_objset,
2304 2305 spa->spa_feat_for_read_obj, spa->spa_feat_desc_obj,
2305 2306 unsup_feat, enabled_feat))
2306 2307 missing_feat_read = B_TRUE;
2307 2308
2308 2309 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2309 2310 if (!feature_is_supported(spa->spa_meta_objset,
2310 2311 spa->spa_feat_for_write_obj, spa->spa_feat_desc_obj,
2311 2312 unsup_feat, enabled_feat)) {
2312 2313 missing_feat_write = B_TRUE;
2313 2314 }
2314 2315 }
2315 2316
2316 2317 fnvlist_add_nvlist(spa->spa_load_info,
2317 2318 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2318 2319
2319 2320 if (!nvlist_empty(unsup_feat)) {
2320 2321 fnvlist_add_nvlist(spa->spa_load_info,
2321 2322 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2322 2323 }
2323 2324
2324 2325 fnvlist_free(enabled_feat);
2325 2326 fnvlist_free(unsup_feat);
2326 2327
2327 2328 if (!missing_feat_read) {
2328 2329 fnvlist_add_boolean(spa->spa_load_info,
2329 2330 ZPOOL_CONFIG_CAN_RDONLY);
2330 2331 }
2331 2332
2332 2333 /*
2333 2334 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2334 2335 * twofold: to determine whether the pool is available for
2335 2336 * import in read-write mode and (if it is not) whether the
2336 2337 * pool is available for import in read-only mode. If the pool
2337 2338 * is available for import in read-write mode, it is displayed
2338 2339 * as available in userland; if it is not available for import
2339 2340 * in read-only mode, it is displayed as unavailable in
2340 2341 * userland. If the pool is available for import in read-only
2341 2342 * mode but not read-write mode, it is displayed as unavailable
2342 2343 * in userland with a special note that the pool is actually
2343 2344 * available for open in read-only mode.
2344 2345 *
2345 2346 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2346 2347 * missing a feature for write, we must first determine whether
2347 2348 * the pool can be opened read-only before returning to
2348 2349 * userland in order to know whether to display the
2349 2350 * abovementioned note.
2350 2351 */
2351 2352 if (missing_feat_read || (missing_feat_write &&
2352 2353 spa_writeable(spa))) {
2353 2354 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2354 2355 ENOTSUP));
2355 2356 }
2356 2357 }
2357 2358
2358 2359 spa->spa_is_initializing = B_TRUE;
2359 2360 error = dsl_pool_open(spa->spa_dsl_pool);
2360 2361 spa->spa_is_initializing = B_FALSE;
2361 2362 if (error != 0)
2362 2363 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2363 2364
2364 2365 if (!mosconfig) {
2365 2366 uint64_t hostid;
2366 2367 nvlist_t *policy = NULL, *nvconfig;
2367 2368
2368 2369 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2369 2370 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2370 2371
2371 2372 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2372 2373 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2373 2374 char *hostname;
2374 2375 unsigned long myhostid = 0;
2375 2376
2376 2377 VERIFY(nvlist_lookup_string(nvconfig,
2377 2378 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2378 2379
2379 2380 #ifdef _KERNEL
2380 2381 myhostid = zone_get_hostid(NULL);
2381 2382 #else /* _KERNEL */
2382 2383 /*
2383 2384 * We're emulating the system's hostid in userland, so
2384 2385 * we can't use zone_get_hostid().
2385 2386 */
2386 2387 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2387 2388 #endif /* _KERNEL */
2388 2389 if (hostid != 0 && myhostid != 0 &&
2389 2390 hostid != myhostid) {
2390 2391 nvlist_free(nvconfig);
2391 2392 cmn_err(CE_WARN, "pool '%s' could not be "
2392 2393 "loaded as it was last accessed by "
2393 2394 "another system (host: %s hostid: 0x%lx). "
2394 2395 "See: http://illumos.org/msg/ZFS-8000-EY",
2395 2396 spa_name(spa), hostname,
2396 2397 (unsigned long)hostid);
2397 2398 return (SET_ERROR(EBADF));
2398 2399 }
2399 2400 }
2400 2401 if (nvlist_lookup_nvlist(spa->spa_config,
2401 2402 ZPOOL_REWIND_POLICY, &policy) == 0)
2402 2403 VERIFY(nvlist_add_nvlist(nvconfig,
|
↓ open down ↓ |
2367 lines elided |
↑ open up ↑ |
2403 2404 ZPOOL_REWIND_POLICY, policy) == 0);
2404 2405
2405 2406 spa_config_set(spa, nvconfig);
2406 2407 spa_unload(spa);
2407 2408 spa_deactivate(spa);
2408 2409 spa_activate(spa, orig_mode);
2409 2410
2410 2411 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2411 2412 }
2412 2413
2414 + /* Grab the secret checksum salt from the MOS. */
2415 + if (spa_dir_prop(spa, DMU_POOL_CHECKSUM_SALT,
2416 + &spa->spa_cksum_salt_obj) == 0) {
2417 + if (zap_lookup(spa->spa_meta_objset, spa->spa_cksum_salt_obj,
2418 + DMU_POOL_CHECKSUM_SALT, 1,
2419 + sizeof (spa->spa_cksum_salt.zcs_bytes),
2420 + spa->spa_cksum_salt.zcs_bytes) != 0) {
2421 + /*
2422 + * MOS format is broken, the salt object is there but
2423 + * is missing the actual salt value.
2424 + */
2425 + return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2426 + }
2427 + } else {
2428 + /* Generate a new salt for subsequent use */
2429 + (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
2430 + sizeof (spa->spa_cksum_salt.zcs_bytes));
2431 + }
2432 +
2413 2433 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2414 2434 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2415 2435 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2416 2436 if (error != 0)
2417 2437 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2418 2438
2419 2439 /*
2420 2440 * Load the bit that tells us to use the new accounting function
2421 2441 * (raid-z deflation). If we have an older pool, this will not
2422 2442 * be present.
2423 2443 */
2424 2444 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2425 2445 if (error != 0 && error != ENOENT)
2426 2446 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2427 2447
2428 2448 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2429 2449 &spa->spa_creation_version);
2430 2450 if (error != 0 && error != ENOENT)
2431 2451 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2432 2452
2433 2453 /*
2434 2454 * Load the persistent error log. If we have an older pool, this will
2435 2455 * not be present.
2436 2456 */
2437 2457 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2438 2458 if (error != 0 && error != ENOENT)
2439 2459 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2440 2460
2441 2461 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2442 2462 &spa->spa_errlog_scrub);
2443 2463 if (error != 0 && error != ENOENT)
2444 2464 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2445 2465
2446 2466 /*
2447 2467 * Load the history object. If we have an older pool, this
2448 2468 * will not be present.
2449 2469 */
2450 2470 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2451 2471 if (error != 0 && error != ENOENT)
2452 2472 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2453 2473
2454 2474 /*
2455 2475 * If we're assembling the pool from the split-off vdevs of
2456 2476 * an existing pool, we don't want to attach the spares & cache
2457 2477 * devices.
2458 2478 */
2459 2479
2460 2480 /*
2461 2481 * Load any hot spares for this pool.
2462 2482 */
2463 2483 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2464 2484 if (error != 0 && error != ENOENT)
2465 2485 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2466 2486 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2467 2487 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2468 2488 if (load_nvlist(spa, spa->spa_spares.sav_object,
2469 2489 &spa->spa_spares.sav_config) != 0)
2470 2490 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2471 2491
2472 2492 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2473 2493 spa_load_spares(spa);
2474 2494 spa_config_exit(spa, SCL_ALL, FTAG);
2475 2495 } else if (error == 0) {
2476 2496 spa->spa_spares.sav_sync = B_TRUE;
2477 2497 }
2478 2498
2479 2499 /*
2480 2500 * Load any level 2 ARC devices for this pool.
2481 2501 */
2482 2502 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2483 2503 &spa->spa_l2cache.sav_object);
2484 2504 if (error != 0 && error != ENOENT)
2485 2505 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2486 2506 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2487 2507 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2488 2508 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2489 2509 &spa->spa_l2cache.sav_config) != 0)
2490 2510 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2491 2511
2492 2512 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2493 2513 spa_load_l2cache(spa);
2494 2514 spa_config_exit(spa, SCL_ALL, FTAG);
2495 2515 } else if (error == 0) {
2496 2516 spa->spa_l2cache.sav_sync = B_TRUE;
2497 2517 }
2498 2518
2499 2519 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2500 2520
2501 2521 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2502 2522 if (error && error != ENOENT)
2503 2523 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2504 2524
2505 2525 if (error == 0) {
2506 2526 uint64_t autoreplace;
2507 2527
2508 2528 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2509 2529 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2510 2530 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2511 2531 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2512 2532 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2513 2533 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2514 2534 &spa->spa_dedup_ditto);
2515 2535
2516 2536 spa->spa_autoreplace = (autoreplace != 0);
2517 2537 }
2518 2538
2519 2539 /*
2520 2540 * If the 'autoreplace' property is set, then post a resource notifying
2521 2541 * the ZFS DE that it should not issue any faults for unopenable
2522 2542 * devices. We also iterate over the vdevs, and post a sysevent for any
2523 2543 * unopenable vdevs so that the normal autoreplace handler can take
2524 2544 * over.
2525 2545 */
2526 2546 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2527 2547 spa_check_removed(spa->spa_root_vdev);
2528 2548 /*
2529 2549 * For the import case, this is done in spa_import(), because
2530 2550 * at this point we're using the spare definitions from
2531 2551 * the MOS config, not necessarily from the userland config.
2532 2552 */
2533 2553 if (state != SPA_LOAD_IMPORT) {
2534 2554 spa_aux_check_removed(&spa->spa_spares);
2535 2555 spa_aux_check_removed(&spa->spa_l2cache);
2536 2556 }
2537 2557 }
2538 2558
2539 2559 /*
2540 2560 * Load the vdev state for all toplevel vdevs.
2541 2561 */
2542 2562 vdev_load(rvd);
2543 2563
2544 2564 /*
2545 2565 * Propagate the leaf DTLs we just loaded all the way up the tree.
2546 2566 */
2547 2567 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2548 2568 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2549 2569 spa_config_exit(spa, SCL_ALL, FTAG);
2550 2570
2551 2571 /*
2552 2572 * Load the DDTs (dedup tables).
2553 2573 */
2554 2574 error = ddt_load(spa);
2555 2575 if (error != 0)
2556 2576 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2557 2577
2558 2578 spa_update_dspace(spa);
2559 2579
2560 2580 /*
2561 2581 * Validate the config, using the MOS config to fill in any
2562 2582 * information which might be missing. If we fail to validate
2563 2583 * the config then declare the pool unfit for use. If we're
2564 2584 * assembling a pool from a split, the log is not transferred
2565 2585 * over.
2566 2586 */
2567 2587 if (type != SPA_IMPORT_ASSEMBLE) {
2568 2588 nvlist_t *nvconfig;
2569 2589
2570 2590 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2571 2591 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2572 2592
2573 2593 if (!spa_config_valid(spa, nvconfig)) {
2574 2594 nvlist_free(nvconfig);
2575 2595 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2576 2596 ENXIO));
2577 2597 }
2578 2598 nvlist_free(nvconfig);
2579 2599
2580 2600 /*
2581 2601 * Now that we've validated the config, check the state of the
2582 2602 * root vdev. If it can't be opened, it indicates one or
2583 2603 * more toplevel vdevs are faulted.
2584 2604 */
2585 2605 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2586 2606 return (SET_ERROR(ENXIO));
2587 2607
2588 2608 if (spa_check_logs(spa)) {
2589 2609 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2590 2610 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2591 2611 }
2592 2612 }
2593 2613
2594 2614 if (missing_feat_write) {
2595 2615 ASSERT(state == SPA_LOAD_TRYIMPORT);
2596 2616
2597 2617 /*
2598 2618 * At this point, we know that we can open the pool in
2599 2619 * read-only mode but not read-write mode. We now have enough
2600 2620 * information and can return to userland.
2601 2621 */
2602 2622 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2603 2623 }
2604 2624
2605 2625 /*
2606 2626 * We've successfully opened the pool, verify that we're ready
2607 2627 * to start pushing transactions.
2608 2628 */
2609 2629 if (state != SPA_LOAD_TRYIMPORT) {
2610 2630 if (error = spa_load_verify(spa))
2611 2631 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2612 2632 error));
2613 2633 }
2614 2634
2615 2635 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2616 2636 spa->spa_load_max_txg == UINT64_MAX)) {
2617 2637 dmu_tx_t *tx;
2618 2638 int need_update = B_FALSE;
2619 2639
2620 2640 ASSERT(state != SPA_LOAD_TRYIMPORT);
2621 2641
2622 2642 /*
2623 2643 * Claim log blocks that haven't been committed yet.
2624 2644 * This must all happen in a single txg.
2625 2645 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2626 2646 * invoked from zil_claim_log_block()'s i/o done callback.
2627 2647 * Price of rollback is that we abandon the log.
2628 2648 */
2629 2649 spa->spa_claiming = B_TRUE;
2630 2650
2631 2651 tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2632 2652 spa_first_txg(spa));
2633 2653 (void) dmu_objset_find(spa_name(spa),
2634 2654 zil_claim, tx, DS_FIND_CHILDREN);
2635 2655 dmu_tx_commit(tx);
2636 2656
2637 2657 spa->spa_claiming = B_FALSE;
2638 2658
2639 2659 spa_set_log_state(spa, SPA_LOG_GOOD);
2640 2660 spa->spa_sync_on = B_TRUE;
2641 2661 txg_sync_start(spa->spa_dsl_pool);
2642 2662
2643 2663 /*
2644 2664 * Wait for all claims to sync. We sync up to the highest
2645 2665 * claimed log block birth time so that claimed log blocks
2646 2666 * don't appear to be from the future. spa_claim_max_txg
2647 2667 * will have been set for us by either zil_check_log_chain()
2648 2668 * (invoked from spa_check_logs()) or zil_claim() above.
2649 2669 */
2650 2670 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2651 2671
2652 2672 /*
2653 2673 * If the config cache is stale, or we have uninitialized
2654 2674 * metaslabs (see spa_vdev_add()), then update the config.
2655 2675 *
2656 2676 * If this is a verbatim import, trust the current
2657 2677 * in-core spa_config and update the disk labels.
2658 2678 */
2659 2679 if (config_cache_txg != spa->spa_config_txg ||
2660 2680 state == SPA_LOAD_IMPORT ||
2661 2681 state == SPA_LOAD_RECOVER ||
2662 2682 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2663 2683 need_update = B_TRUE;
2664 2684
2665 2685 for (int c = 0; c < rvd->vdev_children; c++)
2666 2686 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2667 2687 need_update = B_TRUE;
2668 2688
2669 2689 /*
2670 2690 * Update the config cache asychronously in case we're the
2671 2691 * root pool, in which case the config cache isn't writable yet.
2672 2692 */
2673 2693 if (need_update)
2674 2694 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2675 2695
2676 2696 /*
2677 2697 * Check all DTLs to see if anything needs resilvering.
2678 2698 */
2679 2699 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2680 2700 vdev_resilver_needed(rvd, NULL, NULL))
2681 2701 spa_async_request(spa, SPA_ASYNC_RESILVER);
2682 2702
2683 2703 /*
2684 2704 * Log the fact that we booted up (so that we can detect if
2685 2705 * we rebooted in the middle of an operation).
2686 2706 */
2687 2707 spa_history_log_version(spa, "open");
2688 2708
2689 2709 /*
2690 2710 * Delete any inconsistent datasets.
2691 2711 */
2692 2712 (void) dmu_objset_find(spa_name(spa),
2693 2713 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2694 2714
2695 2715 /*
2696 2716 * Clean up any stale temporary dataset userrefs.
2697 2717 */
2698 2718 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2699 2719 }
2700 2720
2701 2721 return (0);
2702 2722 }
2703 2723
2704 2724 static int
2705 2725 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2706 2726 {
2707 2727 int mode = spa->spa_mode;
2708 2728
2709 2729 spa_unload(spa);
2710 2730 spa_deactivate(spa);
2711 2731
2712 2732 spa->spa_load_max_txg--;
2713 2733
2714 2734 spa_activate(spa, mode);
2715 2735 spa_async_suspend(spa);
2716 2736
2717 2737 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2718 2738 }
2719 2739
2720 2740 /*
2721 2741 * If spa_load() fails this function will try loading prior txg's. If
2722 2742 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2723 2743 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2724 2744 * function will not rewind the pool and will return the same error as
2725 2745 * spa_load().
2726 2746 */
2727 2747 static int
2728 2748 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2729 2749 uint64_t max_request, int rewind_flags)
2730 2750 {
2731 2751 nvlist_t *loadinfo = NULL;
2732 2752 nvlist_t *config = NULL;
2733 2753 int load_error, rewind_error;
2734 2754 uint64_t safe_rewind_txg;
2735 2755 uint64_t min_txg;
2736 2756
2737 2757 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2738 2758 spa->spa_load_max_txg = spa->spa_load_txg;
2739 2759 spa_set_log_state(spa, SPA_LOG_CLEAR);
2740 2760 } else {
2741 2761 spa->spa_load_max_txg = max_request;
2742 2762 }
2743 2763
2744 2764 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2745 2765 mosconfig);
2746 2766 if (load_error == 0)
2747 2767 return (0);
2748 2768
2749 2769 if (spa->spa_root_vdev != NULL)
2750 2770 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2751 2771
2752 2772 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2753 2773 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2754 2774
2755 2775 if (rewind_flags & ZPOOL_NEVER_REWIND) {
2756 2776 nvlist_free(config);
2757 2777 return (load_error);
2758 2778 }
2759 2779
2760 2780 if (state == SPA_LOAD_RECOVER) {
2761 2781 /* Price of rolling back is discarding txgs, including log */
2762 2782 spa_set_log_state(spa, SPA_LOG_CLEAR);
2763 2783 } else {
2764 2784 /*
2765 2785 * If we aren't rolling back save the load info from our first
2766 2786 * import attempt so that we can restore it after attempting
2767 2787 * to rewind.
2768 2788 */
2769 2789 loadinfo = spa->spa_load_info;
2770 2790 spa->spa_load_info = fnvlist_alloc();
2771 2791 }
2772 2792
2773 2793 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2774 2794 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2775 2795 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2776 2796 TXG_INITIAL : safe_rewind_txg;
2777 2797
2778 2798 /*
2779 2799 * Continue as long as we're finding errors, we're still within
2780 2800 * the acceptable rewind range, and we're still finding uberblocks
2781 2801 */
2782 2802 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2783 2803 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2784 2804 if (spa->spa_load_max_txg < safe_rewind_txg)
2785 2805 spa->spa_extreme_rewind = B_TRUE;
2786 2806 rewind_error = spa_load_retry(spa, state, mosconfig);
2787 2807 }
2788 2808
2789 2809 spa->spa_extreme_rewind = B_FALSE;
2790 2810 spa->spa_load_max_txg = UINT64_MAX;
2791 2811
2792 2812 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2793 2813 spa_config_set(spa, config);
2794 2814
2795 2815 if (state == SPA_LOAD_RECOVER) {
2796 2816 ASSERT3P(loadinfo, ==, NULL);
2797 2817 return (rewind_error);
2798 2818 } else {
2799 2819 /* Store the rewind info as part of the initial load info */
2800 2820 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2801 2821 spa->spa_load_info);
2802 2822
2803 2823 /* Restore the initial load info */
2804 2824 fnvlist_free(spa->spa_load_info);
2805 2825 spa->spa_load_info = loadinfo;
2806 2826
2807 2827 return (load_error);
2808 2828 }
2809 2829 }
2810 2830
2811 2831 /*
2812 2832 * Pool Open/Import
2813 2833 *
2814 2834 * The import case is identical to an open except that the configuration is sent
2815 2835 * down from userland, instead of grabbed from the configuration cache. For the
2816 2836 * case of an open, the pool configuration will exist in the
2817 2837 * POOL_STATE_UNINITIALIZED state.
2818 2838 *
2819 2839 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2820 2840 * the same time open the pool, without having to keep around the spa_t in some
2821 2841 * ambiguous state.
2822 2842 */
2823 2843 static int
2824 2844 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2825 2845 nvlist_t **config)
2826 2846 {
2827 2847 spa_t *spa;
2828 2848 spa_load_state_t state = SPA_LOAD_OPEN;
2829 2849 int error;
2830 2850 int locked = B_FALSE;
2831 2851
2832 2852 *spapp = NULL;
2833 2853
2834 2854 /*
2835 2855 * As disgusting as this is, we need to support recursive calls to this
2836 2856 * function because dsl_dir_open() is called during spa_load(), and ends
2837 2857 * up calling spa_open() again. The real fix is to figure out how to
2838 2858 * avoid dsl_dir_open() calling this in the first place.
2839 2859 */
2840 2860 if (mutex_owner(&spa_namespace_lock) != curthread) {
2841 2861 mutex_enter(&spa_namespace_lock);
2842 2862 locked = B_TRUE;
2843 2863 }
2844 2864
2845 2865 if ((spa = spa_lookup(pool)) == NULL) {
2846 2866 if (locked)
2847 2867 mutex_exit(&spa_namespace_lock);
2848 2868 return (SET_ERROR(ENOENT));
2849 2869 }
2850 2870
2851 2871 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2852 2872 zpool_rewind_policy_t policy;
2853 2873
2854 2874 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2855 2875 &policy);
2856 2876 if (policy.zrp_request & ZPOOL_DO_REWIND)
2857 2877 state = SPA_LOAD_RECOVER;
2858 2878
2859 2879 spa_activate(spa, spa_mode_global);
2860 2880
2861 2881 if (state != SPA_LOAD_RECOVER)
2862 2882 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2863 2883
2864 2884 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2865 2885 policy.zrp_request);
2866 2886
2867 2887 if (error == EBADF) {
2868 2888 /*
2869 2889 * If vdev_validate() returns failure (indicated by
2870 2890 * EBADF), it indicates that one of the vdevs indicates
2871 2891 * that the pool has been exported or destroyed. If
2872 2892 * this is the case, the config cache is out of sync and
2873 2893 * we should remove the pool from the namespace.
2874 2894 */
2875 2895 spa_unload(spa);
2876 2896 spa_deactivate(spa);
2877 2897 spa_config_sync(spa, B_TRUE, B_TRUE);
2878 2898 spa_remove(spa);
2879 2899 if (locked)
2880 2900 mutex_exit(&spa_namespace_lock);
2881 2901 return (SET_ERROR(ENOENT));
2882 2902 }
2883 2903
2884 2904 if (error) {
2885 2905 /*
2886 2906 * We can't open the pool, but we still have useful
2887 2907 * information: the state of each vdev after the
2888 2908 * attempted vdev_open(). Return this to the user.
2889 2909 */
2890 2910 if (config != NULL && spa->spa_config) {
2891 2911 VERIFY(nvlist_dup(spa->spa_config, config,
2892 2912 KM_SLEEP) == 0);
2893 2913 VERIFY(nvlist_add_nvlist(*config,
2894 2914 ZPOOL_CONFIG_LOAD_INFO,
2895 2915 spa->spa_load_info) == 0);
2896 2916 }
2897 2917 spa_unload(spa);
2898 2918 spa_deactivate(spa);
2899 2919 spa->spa_last_open_failed = error;
2900 2920 if (locked)
2901 2921 mutex_exit(&spa_namespace_lock);
2902 2922 *spapp = NULL;
2903 2923 return (error);
2904 2924 }
2905 2925 }
2906 2926
2907 2927 spa_open_ref(spa, tag);
2908 2928
2909 2929 if (config != NULL)
2910 2930 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2911 2931
2912 2932 /*
2913 2933 * If we've recovered the pool, pass back any information we
2914 2934 * gathered while doing the load.
2915 2935 */
2916 2936 if (state == SPA_LOAD_RECOVER) {
2917 2937 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2918 2938 spa->spa_load_info) == 0);
2919 2939 }
2920 2940
2921 2941 if (locked) {
2922 2942 spa->spa_last_open_failed = 0;
2923 2943 spa->spa_last_ubsync_txg = 0;
2924 2944 spa->spa_load_txg = 0;
2925 2945 mutex_exit(&spa_namespace_lock);
2926 2946 }
2927 2947
2928 2948 *spapp = spa;
2929 2949
2930 2950 return (0);
2931 2951 }
2932 2952
2933 2953 int
2934 2954 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2935 2955 nvlist_t **config)
2936 2956 {
2937 2957 return (spa_open_common(name, spapp, tag, policy, config));
2938 2958 }
2939 2959
2940 2960 int
2941 2961 spa_open(const char *name, spa_t **spapp, void *tag)
2942 2962 {
2943 2963 return (spa_open_common(name, spapp, tag, NULL, NULL));
2944 2964 }
2945 2965
2946 2966 /*
2947 2967 * Lookup the given spa_t, incrementing the inject count in the process,
2948 2968 * preventing it from being exported or destroyed.
2949 2969 */
2950 2970 spa_t *
2951 2971 spa_inject_addref(char *name)
2952 2972 {
2953 2973 spa_t *spa;
2954 2974
2955 2975 mutex_enter(&spa_namespace_lock);
2956 2976 if ((spa = spa_lookup(name)) == NULL) {
2957 2977 mutex_exit(&spa_namespace_lock);
2958 2978 return (NULL);
2959 2979 }
2960 2980 spa->spa_inject_ref++;
2961 2981 mutex_exit(&spa_namespace_lock);
2962 2982
2963 2983 return (spa);
2964 2984 }
2965 2985
2966 2986 void
2967 2987 spa_inject_delref(spa_t *spa)
2968 2988 {
2969 2989 mutex_enter(&spa_namespace_lock);
2970 2990 spa->spa_inject_ref--;
2971 2991 mutex_exit(&spa_namespace_lock);
2972 2992 }
2973 2993
2974 2994 /*
2975 2995 * Add spares device information to the nvlist.
2976 2996 */
2977 2997 static void
2978 2998 spa_add_spares(spa_t *spa, nvlist_t *config)
2979 2999 {
2980 3000 nvlist_t **spares;
2981 3001 uint_t i, nspares;
2982 3002 nvlist_t *nvroot;
2983 3003 uint64_t guid;
2984 3004 vdev_stat_t *vs;
2985 3005 uint_t vsc;
2986 3006 uint64_t pool;
2987 3007
2988 3008 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2989 3009
2990 3010 if (spa->spa_spares.sav_count == 0)
2991 3011 return;
2992 3012
2993 3013 VERIFY(nvlist_lookup_nvlist(config,
2994 3014 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2995 3015 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2996 3016 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2997 3017 if (nspares != 0) {
2998 3018 VERIFY(nvlist_add_nvlist_array(nvroot,
2999 3019 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3000 3020 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3001 3021 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3002 3022
3003 3023 /*
3004 3024 * Go through and find any spares which have since been
3005 3025 * repurposed as an active spare. If this is the case, update
3006 3026 * their status appropriately.
3007 3027 */
3008 3028 for (i = 0; i < nspares; i++) {
3009 3029 VERIFY(nvlist_lookup_uint64(spares[i],
3010 3030 ZPOOL_CONFIG_GUID, &guid) == 0);
3011 3031 if (spa_spare_exists(guid, &pool, NULL) &&
3012 3032 pool != 0ULL) {
3013 3033 VERIFY(nvlist_lookup_uint64_array(
3014 3034 spares[i], ZPOOL_CONFIG_VDEV_STATS,
3015 3035 (uint64_t **)&vs, &vsc) == 0);
3016 3036 vs->vs_state = VDEV_STATE_CANT_OPEN;
3017 3037 vs->vs_aux = VDEV_AUX_SPARED;
3018 3038 }
3019 3039 }
3020 3040 }
3021 3041 }
3022 3042
3023 3043 /*
3024 3044 * Add l2cache device information to the nvlist, including vdev stats.
3025 3045 */
3026 3046 static void
3027 3047 spa_add_l2cache(spa_t *spa, nvlist_t *config)
3028 3048 {
3029 3049 nvlist_t **l2cache;
3030 3050 uint_t i, j, nl2cache;
3031 3051 nvlist_t *nvroot;
3032 3052 uint64_t guid;
3033 3053 vdev_t *vd;
3034 3054 vdev_stat_t *vs;
3035 3055 uint_t vsc;
3036 3056
3037 3057 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3038 3058
3039 3059 if (spa->spa_l2cache.sav_count == 0)
3040 3060 return;
3041 3061
3042 3062 VERIFY(nvlist_lookup_nvlist(config,
3043 3063 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3044 3064 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3045 3065 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3046 3066 if (nl2cache != 0) {
3047 3067 VERIFY(nvlist_add_nvlist_array(nvroot,
3048 3068 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3049 3069 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3050 3070 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3051 3071
3052 3072 /*
3053 3073 * Update level 2 cache device stats.
3054 3074 */
3055 3075
3056 3076 for (i = 0; i < nl2cache; i++) {
3057 3077 VERIFY(nvlist_lookup_uint64(l2cache[i],
3058 3078 ZPOOL_CONFIG_GUID, &guid) == 0);
3059 3079
3060 3080 vd = NULL;
3061 3081 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3062 3082 if (guid ==
3063 3083 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3064 3084 vd = spa->spa_l2cache.sav_vdevs[j];
3065 3085 break;
3066 3086 }
3067 3087 }
3068 3088 ASSERT(vd != NULL);
3069 3089
3070 3090 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3071 3091 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3072 3092 == 0);
3073 3093 vdev_get_stats(vd, vs);
3074 3094 }
3075 3095 }
3076 3096 }
3077 3097
3078 3098 static void
3079 3099 spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3080 3100 {
3081 3101 nvlist_t *features;
3082 3102 zap_cursor_t zc;
3083 3103 zap_attribute_t za;
3084 3104
3085 3105 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3086 3106 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3087 3107
3088 3108 if (spa->spa_feat_for_read_obj != 0) {
3089 3109 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3090 3110 spa->spa_feat_for_read_obj);
3091 3111 zap_cursor_retrieve(&zc, &za) == 0;
3092 3112 zap_cursor_advance(&zc)) {
3093 3113 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3094 3114 za.za_num_integers == 1);
3095 3115 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3096 3116 za.za_first_integer));
3097 3117 }
3098 3118 zap_cursor_fini(&zc);
3099 3119 }
3100 3120
3101 3121 if (spa->spa_feat_for_write_obj != 0) {
3102 3122 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3103 3123 spa->spa_feat_for_write_obj);
3104 3124 zap_cursor_retrieve(&zc, &za) == 0;
3105 3125 zap_cursor_advance(&zc)) {
3106 3126 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3107 3127 za.za_num_integers == 1);
3108 3128 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3109 3129 za.za_first_integer));
3110 3130 }
3111 3131 zap_cursor_fini(&zc);
3112 3132 }
3113 3133
3114 3134 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3115 3135 features) == 0);
3116 3136 nvlist_free(features);
3117 3137 }
3118 3138
3119 3139 int
3120 3140 spa_get_stats(const char *name, nvlist_t **config,
3121 3141 char *altroot, size_t buflen)
3122 3142 {
3123 3143 int error;
3124 3144 spa_t *spa;
3125 3145
3126 3146 *config = NULL;
3127 3147 error = spa_open_common(name, &spa, FTAG, NULL, config);
3128 3148
3129 3149 if (spa != NULL) {
3130 3150 /*
3131 3151 * This still leaves a window of inconsistency where the spares
3132 3152 * or l2cache devices could change and the config would be
3133 3153 * self-inconsistent.
3134 3154 */
3135 3155 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3136 3156
3137 3157 if (*config != NULL) {
3138 3158 uint64_t loadtimes[2];
3139 3159
3140 3160 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3141 3161 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3142 3162 VERIFY(nvlist_add_uint64_array(*config,
3143 3163 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3144 3164
3145 3165 VERIFY(nvlist_add_uint64(*config,
3146 3166 ZPOOL_CONFIG_ERRCOUNT,
3147 3167 spa_get_errlog_size(spa)) == 0);
3148 3168
3149 3169 if (spa_suspended(spa))
3150 3170 VERIFY(nvlist_add_uint64(*config,
3151 3171 ZPOOL_CONFIG_SUSPENDED,
3152 3172 spa->spa_failmode) == 0);
3153 3173
3154 3174 spa_add_spares(spa, *config);
3155 3175 spa_add_l2cache(spa, *config);
3156 3176 spa_add_feature_stats(spa, *config);
3157 3177 }
3158 3178 }
3159 3179
3160 3180 /*
3161 3181 * We want to get the alternate root even for faulted pools, so we cheat
3162 3182 * and call spa_lookup() directly.
3163 3183 */
3164 3184 if (altroot) {
3165 3185 if (spa == NULL) {
3166 3186 mutex_enter(&spa_namespace_lock);
3167 3187 spa = spa_lookup(name);
3168 3188 if (spa)
3169 3189 spa_altroot(spa, altroot, buflen);
3170 3190 else
3171 3191 altroot[0] = '\0';
3172 3192 spa = NULL;
3173 3193 mutex_exit(&spa_namespace_lock);
3174 3194 } else {
3175 3195 spa_altroot(spa, altroot, buflen);
3176 3196 }
3177 3197 }
3178 3198
3179 3199 if (spa != NULL) {
3180 3200 spa_config_exit(spa, SCL_CONFIG, FTAG);
3181 3201 spa_close(spa, FTAG);
3182 3202 }
3183 3203
3184 3204 return (error);
3185 3205 }
3186 3206
3187 3207 /*
3188 3208 * Validate that the auxiliary device array is well formed. We must have an
3189 3209 * array of nvlists, each which describes a valid leaf vdev. If this is an
3190 3210 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3191 3211 * specified, as long as they are well-formed.
3192 3212 */
3193 3213 static int
3194 3214 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3195 3215 spa_aux_vdev_t *sav, const char *config, uint64_t version,
3196 3216 vdev_labeltype_t label)
3197 3217 {
3198 3218 nvlist_t **dev;
3199 3219 uint_t i, ndev;
3200 3220 vdev_t *vd;
3201 3221 int error;
3202 3222
3203 3223 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3204 3224
3205 3225 /*
3206 3226 * It's acceptable to have no devs specified.
3207 3227 */
3208 3228 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3209 3229 return (0);
3210 3230
3211 3231 if (ndev == 0)
3212 3232 return (SET_ERROR(EINVAL));
3213 3233
3214 3234 /*
3215 3235 * Make sure the pool is formatted with a version that supports this
3216 3236 * device type.
3217 3237 */
3218 3238 if (spa_version(spa) < version)
3219 3239 return (SET_ERROR(ENOTSUP));
3220 3240
3221 3241 /*
3222 3242 * Set the pending device list so we correctly handle device in-use
3223 3243 * checking.
3224 3244 */
3225 3245 sav->sav_pending = dev;
3226 3246 sav->sav_npending = ndev;
3227 3247
3228 3248 for (i = 0; i < ndev; i++) {
3229 3249 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3230 3250 mode)) != 0)
3231 3251 goto out;
3232 3252
3233 3253 if (!vd->vdev_ops->vdev_op_leaf) {
3234 3254 vdev_free(vd);
3235 3255 error = SET_ERROR(EINVAL);
3236 3256 goto out;
3237 3257 }
3238 3258
3239 3259 /*
3240 3260 * The L2ARC currently only supports disk devices in
3241 3261 * kernel context. For user-level testing, we allow it.
3242 3262 */
3243 3263 #ifdef _KERNEL
3244 3264 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3245 3265 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3246 3266 error = SET_ERROR(ENOTBLK);
3247 3267 vdev_free(vd);
3248 3268 goto out;
3249 3269 }
3250 3270 #endif
3251 3271 vd->vdev_top = vd;
3252 3272
3253 3273 if ((error = vdev_open(vd)) == 0 &&
3254 3274 (error = vdev_label_init(vd, crtxg, label)) == 0) {
3255 3275 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3256 3276 vd->vdev_guid) == 0);
3257 3277 }
3258 3278
3259 3279 vdev_free(vd);
3260 3280
3261 3281 if (error &&
3262 3282 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3263 3283 goto out;
3264 3284 else
3265 3285 error = 0;
3266 3286 }
3267 3287
3268 3288 out:
3269 3289 sav->sav_pending = NULL;
3270 3290 sav->sav_npending = 0;
3271 3291 return (error);
3272 3292 }
3273 3293
3274 3294 static int
3275 3295 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3276 3296 {
3277 3297 int error;
3278 3298
3279 3299 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3280 3300
3281 3301 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3282 3302 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3283 3303 VDEV_LABEL_SPARE)) != 0) {
3284 3304 return (error);
3285 3305 }
3286 3306
3287 3307 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3288 3308 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3289 3309 VDEV_LABEL_L2CACHE));
3290 3310 }
3291 3311
3292 3312 static void
3293 3313 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3294 3314 const char *config)
3295 3315 {
3296 3316 int i;
3297 3317
3298 3318 if (sav->sav_config != NULL) {
3299 3319 nvlist_t **olddevs;
3300 3320 uint_t oldndevs;
3301 3321 nvlist_t **newdevs;
3302 3322
3303 3323 /*
3304 3324 * Generate new dev list by concatentating with the
3305 3325 * current dev list.
3306 3326 */
3307 3327 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3308 3328 &olddevs, &oldndevs) == 0);
3309 3329
3310 3330 newdevs = kmem_alloc(sizeof (void *) *
3311 3331 (ndevs + oldndevs), KM_SLEEP);
3312 3332 for (i = 0; i < oldndevs; i++)
3313 3333 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3314 3334 KM_SLEEP) == 0);
3315 3335 for (i = 0; i < ndevs; i++)
3316 3336 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3317 3337 KM_SLEEP) == 0);
3318 3338
3319 3339 VERIFY(nvlist_remove(sav->sav_config, config,
3320 3340 DATA_TYPE_NVLIST_ARRAY) == 0);
3321 3341
3322 3342 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3323 3343 config, newdevs, ndevs + oldndevs) == 0);
3324 3344 for (i = 0; i < oldndevs + ndevs; i++)
3325 3345 nvlist_free(newdevs[i]);
3326 3346 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3327 3347 } else {
3328 3348 /*
3329 3349 * Generate a new dev list.
3330 3350 */
3331 3351 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3332 3352 KM_SLEEP) == 0);
3333 3353 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3334 3354 devs, ndevs) == 0);
3335 3355 }
3336 3356 }
3337 3357
3338 3358 /*
3339 3359 * Stop and drop level 2 ARC devices
3340 3360 */
3341 3361 void
3342 3362 spa_l2cache_drop(spa_t *spa)
3343 3363 {
3344 3364 vdev_t *vd;
3345 3365 int i;
3346 3366 spa_aux_vdev_t *sav = &spa->spa_l2cache;
3347 3367
3348 3368 for (i = 0; i < sav->sav_count; i++) {
3349 3369 uint64_t pool;
3350 3370
3351 3371 vd = sav->sav_vdevs[i];
3352 3372 ASSERT(vd != NULL);
3353 3373
3354 3374 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3355 3375 pool != 0ULL && l2arc_vdev_present(vd))
3356 3376 l2arc_remove_vdev(vd);
3357 3377 }
3358 3378 }
3359 3379
3360 3380 /*
3361 3381 * Pool Creation
3362 3382 */
3363 3383 int
3364 3384 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3365 3385 nvlist_t *zplprops)
3366 3386 {
3367 3387 spa_t *spa;
3368 3388 char *altroot = NULL;
3369 3389 vdev_t *rvd;
3370 3390 dsl_pool_t *dp;
3371 3391 dmu_tx_t *tx;
3372 3392 int error = 0;
3373 3393 uint64_t txg = TXG_INITIAL;
3374 3394 nvlist_t **spares, **l2cache;
3375 3395 uint_t nspares, nl2cache;
3376 3396 uint64_t version, obj;
3377 3397 boolean_t has_features;
3378 3398
3379 3399 /*
3380 3400 * If this pool already exists, return failure.
3381 3401 */
3382 3402 mutex_enter(&spa_namespace_lock);
3383 3403 if (spa_lookup(pool) != NULL) {
3384 3404 mutex_exit(&spa_namespace_lock);
3385 3405 return (SET_ERROR(EEXIST));
3386 3406 }
3387 3407
3388 3408 /*
3389 3409 * Allocate a new spa_t structure.
3390 3410 */
3391 3411 (void) nvlist_lookup_string(props,
3392 3412 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3393 3413 spa = spa_add(pool, NULL, altroot);
3394 3414 spa_activate(spa, spa_mode_global);
3395 3415
3396 3416 if (props && (error = spa_prop_validate(spa, props))) {
3397 3417 spa_deactivate(spa);
3398 3418 spa_remove(spa);
3399 3419 mutex_exit(&spa_namespace_lock);
3400 3420 return (error);
3401 3421 }
3402 3422
3403 3423 has_features = B_FALSE;
3404 3424 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3405 3425 elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3406 3426 if (zpool_prop_feature(nvpair_name(elem)))
3407 3427 has_features = B_TRUE;
3408 3428 }
3409 3429
3410 3430 if (has_features || nvlist_lookup_uint64(props,
3411 3431 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3412 3432 version = SPA_VERSION;
3413 3433 }
3414 3434 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3415 3435
3416 3436 spa->spa_first_txg = txg;
3417 3437 spa->spa_uberblock.ub_txg = txg - 1;
3418 3438 spa->spa_uberblock.ub_version = version;
3419 3439 spa->spa_ubsync = spa->spa_uberblock;
3420 3440
3421 3441 /*
3422 3442 * Create "The Godfather" zio to hold all async IOs
3423 3443 */
3424 3444 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
3425 3445 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
3426 3446
3427 3447 /*
3428 3448 * Create the root vdev.
3429 3449 */
3430 3450 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3431 3451
3432 3452 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3433 3453
3434 3454 ASSERT(error != 0 || rvd != NULL);
3435 3455 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3436 3456
3437 3457 if (error == 0 && !zfs_allocatable_devs(nvroot))
3438 3458 error = SET_ERROR(EINVAL);
3439 3459
3440 3460 if (error == 0 &&
3441 3461 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3442 3462 (error = spa_validate_aux(spa, nvroot, txg,
3443 3463 VDEV_ALLOC_ADD)) == 0) {
3444 3464 for (int c = 0; c < rvd->vdev_children; c++) {
3445 3465 vdev_metaslab_set_size(rvd->vdev_child[c]);
3446 3466 vdev_expand(rvd->vdev_child[c], txg);
3447 3467 }
3448 3468 }
3449 3469
3450 3470 spa_config_exit(spa, SCL_ALL, FTAG);
3451 3471
3452 3472 if (error != 0) {
3453 3473 spa_unload(spa);
3454 3474 spa_deactivate(spa);
3455 3475 spa_remove(spa);
3456 3476 mutex_exit(&spa_namespace_lock);
3457 3477 return (error);
3458 3478 }
3459 3479
3460 3480 /*
3461 3481 * Get the list of spares, if specified.
3462 3482 */
3463 3483 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3464 3484 &spares, &nspares) == 0) {
3465 3485 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3466 3486 KM_SLEEP) == 0);
3467 3487 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3468 3488 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3469 3489 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3470 3490 spa_load_spares(spa);
3471 3491 spa_config_exit(spa, SCL_ALL, FTAG);
3472 3492 spa->spa_spares.sav_sync = B_TRUE;
3473 3493 }
3474 3494
3475 3495 /*
3476 3496 * Get the list of level 2 cache devices, if specified.
3477 3497 */
3478 3498 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3479 3499 &l2cache, &nl2cache) == 0) {
3480 3500 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3481 3501 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3482 3502 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3483 3503 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3484 3504 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3485 3505 spa_load_l2cache(spa);
3486 3506 spa_config_exit(spa, SCL_ALL, FTAG);
3487 3507 spa->spa_l2cache.sav_sync = B_TRUE;
3488 3508 }
3489 3509
3490 3510 spa->spa_is_initializing = B_TRUE;
3491 3511 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3492 3512 spa->spa_meta_objset = dp->dp_meta_objset;
3493 3513 spa->spa_is_initializing = B_FALSE;
3494 3514
3495 3515 /*
3496 3516 * Create DDTs (dedup tables).
3497 3517 */
3498 3518 ddt_create(spa);
3499 3519
3500 3520 spa_update_dspace(spa);
3501 3521
3502 3522 tx = dmu_tx_create_assigned(dp, txg);
3503 3523
3504 3524 /*
3505 3525 * Create the pool config object.
3506 3526 */
3507 3527 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3508 3528 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3509 3529 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3510 3530
3511 3531 if (zap_add(spa->spa_meta_objset,
3512 3532 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3513 3533 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3514 3534 cmn_err(CE_PANIC, "failed to add pool config");
3515 3535 }
3516 3536
3517 3537 if (spa_version(spa) >= SPA_VERSION_FEATURES)
3518 3538 spa_feature_create_zap_objects(spa, tx);
3519 3539
3520 3540 if (zap_add(spa->spa_meta_objset,
3521 3541 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3522 3542 sizeof (uint64_t), 1, &version, tx) != 0) {
3523 3543 cmn_err(CE_PANIC, "failed to add pool version");
3524 3544 }
3525 3545
3526 3546 /* Newly created pools with the right version are always deflated. */
3527 3547 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3528 3548 spa->spa_deflate = TRUE;
3529 3549 if (zap_add(spa->spa_meta_objset,
3530 3550 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3531 3551 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3532 3552 cmn_err(CE_PANIC, "failed to add deflate");
3533 3553 }
3534 3554 }
3535 3555
3536 3556 /*
3537 3557 * Create the deferred-free bpobj. Turn off compression
3538 3558 * because sync-to-convergence takes longer if the blocksize
3539 3559 * keeps changing.
3540 3560 */
3541 3561 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3542 3562 dmu_object_set_compress(spa->spa_meta_objset, obj,
3543 3563 ZIO_COMPRESS_OFF, tx);
3544 3564 if (zap_add(spa->spa_meta_objset,
3545 3565 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3546 3566 sizeof (uint64_t), 1, &obj, tx) != 0) {
3547 3567 cmn_err(CE_PANIC, "failed to add bpobj");
3548 3568 }
|
↓ open down ↓ |
1126 lines elided |
↑ open up ↑ |
3549 3569 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3550 3570 spa->spa_meta_objset, obj));
3551 3571
3552 3572 /*
3553 3573 * Create the pool's history object.
3554 3574 */
3555 3575 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3556 3576 spa_history_create_obj(spa, tx);
3557 3577
3558 3578 /*
3579 + * Generate some random noise for salted checksums to operate on. As
3580 + * soon as a salted checksum is used for the first time we will
3581 + * generate the persistent MOS object to hold the salt (see
3582 + * spa_activate_salted_cksum).
3583 + */
3584 + (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
3585 + sizeof (spa->spa_cksum_salt.zcs_bytes));
3586 +
3587 + /*
3559 3588 * Set pool properties.
3560 3589 */
3561 3590 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3562 3591 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3563 3592 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3564 3593 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3565 3594
3566 3595 if (props != NULL) {
3567 3596 spa_configfile_set(spa, props, B_FALSE);
3568 3597 spa_sync_props(props, tx);
3569 3598 }
3570 3599
3571 3600 dmu_tx_commit(tx);
3572 3601
3573 3602 spa->spa_sync_on = B_TRUE;
3574 3603 txg_sync_start(spa->spa_dsl_pool);
3575 3604
3576 3605 /*
3577 3606 * We explicitly wait for the first transaction to complete so that our
3578 3607 * bean counters are appropriately updated.
3579 3608 */
3580 3609 txg_wait_synced(spa->spa_dsl_pool, txg);
3581 3610
3582 3611 spa_config_sync(spa, B_FALSE, B_TRUE);
3583 3612
3584 3613 spa_history_log_version(spa, "create");
3585 3614
3586 3615 spa->spa_minref = refcount_count(&spa->spa_refcount);
3587 3616
3588 3617 mutex_exit(&spa_namespace_lock);
3589 3618
3590 3619 return (0);
3591 3620 }
3592 3621
3593 3622 #ifdef _KERNEL
3594 3623 /*
3595 3624 * Get the root pool information from the root disk, then import the root pool
3596 3625 * during the system boot up time.
3597 3626 */
3598 3627 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3599 3628
3600 3629 static nvlist_t *
3601 3630 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3602 3631 {
3603 3632 nvlist_t *config;
3604 3633 nvlist_t *nvtop, *nvroot;
3605 3634 uint64_t pgid;
3606 3635
3607 3636 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3608 3637 return (NULL);
3609 3638
3610 3639 /*
3611 3640 * Add this top-level vdev to the child array.
3612 3641 */
3613 3642 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3614 3643 &nvtop) == 0);
3615 3644 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3616 3645 &pgid) == 0);
3617 3646 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3618 3647
3619 3648 /*
3620 3649 * Put this pool's top-level vdevs into a root vdev.
3621 3650 */
3622 3651 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3623 3652 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3624 3653 VDEV_TYPE_ROOT) == 0);
3625 3654 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3626 3655 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3627 3656 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3628 3657 &nvtop, 1) == 0);
3629 3658
3630 3659 /*
3631 3660 * Replace the existing vdev_tree with the new root vdev in
3632 3661 * this pool's configuration (remove the old, add the new).
3633 3662 */
3634 3663 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3635 3664 nvlist_free(nvroot);
3636 3665 return (config);
3637 3666 }
3638 3667
3639 3668 /*
3640 3669 * Walk the vdev tree and see if we can find a device with "better"
3641 3670 * configuration. A configuration is "better" if the label on that
3642 3671 * device has a more recent txg.
3643 3672 */
3644 3673 static void
3645 3674 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3646 3675 {
3647 3676 for (int c = 0; c < vd->vdev_children; c++)
3648 3677 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3649 3678
3650 3679 if (vd->vdev_ops->vdev_op_leaf) {
3651 3680 nvlist_t *label;
3652 3681 uint64_t label_txg;
3653 3682
3654 3683 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3655 3684 &label) != 0)
3656 3685 return;
3657 3686
3658 3687 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3659 3688 &label_txg) == 0);
3660 3689
3661 3690 /*
3662 3691 * Do we have a better boot device?
3663 3692 */
3664 3693 if (label_txg > *txg) {
3665 3694 *txg = label_txg;
3666 3695 *avd = vd;
3667 3696 }
3668 3697 nvlist_free(label);
3669 3698 }
3670 3699 }
3671 3700
3672 3701 /*
3673 3702 * Import a root pool.
3674 3703 *
3675 3704 * For x86. devpath_list will consist of devid and/or physpath name of
3676 3705 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3677 3706 * The GRUB "findroot" command will return the vdev we should boot.
3678 3707 *
3679 3708 * For Sparc, devpath_list consists the physpath name of the booting device
3680 3709 * no matter the rootpool is a single device pool or a mirrored pool.
3681 3710 * e.g.
3682 3711 * "/pci@1f,0/ide@d/disk@0,0:a"
3683 3712 */
3684 3713 int
3685 3714 spa_import_rootpool(char *devpath, char *devid)
3686 3715 {
3687 3716 spa_t *spa;
3688 3717 vdev_t *rvd, *bvd, *avd = NULL;
3689 3718 nvlist_t *config, *nvtop;
3690 3719 uint64_t guid, txg;
3691 3720 char *pname;
3692 3721 int error;
3693 3722
3694 3723 /*
3695 3724 * Read the label from the boot device and generate a configuration.
3696 3725 */
3697 3726 config = spa_generate_rootconf(devpath, devid, &guid);
3698 3727 #if defined(_OBP) && defined(_KERNEL)
3699 3728 if (config == NULL) {
3700 3729 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3701 3730 /* iscsi boot */
3702 3731 get_iscsi_bootpath_phy(devpath);
3703 3732 config = spa_generate_rootconf(devpath, devid, &guid);
3704 3733 }
3705 3734 }
3706 3735 #endif
3707 3736 if (config == NULL) {
3708 3737 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3709 3738 devpath);
3710 3739 return (SET_ERROR(EIO));
3711 3740 }
3712 3741
3713 3742 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3714 3743 &pname) == 0);
3715 3744 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3716 3745
3717 3746 mutex_enter(&spa_namespace_lock);
3718 3747 if ((spa = spa_lookup(pname)) != NULL) {
3719 3748 /*
3720 3749 * Remove the existing root pool from the namespace so that we
3721 3750 * can replace it with the correct config we just read in.
3722 3751 */
3723 3752 spa_remove(spa);
3724 3753 }
3725 3754
3726 3755 spa = spa_add(pname, config, NULL);
3727 3756 spa->spa_is_root = B_TRUE;
3728 3757 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3729 3758
3730 3759 /*
3731 3760 * Build up a vdev tree based on the boot device's label config.
3732 3761 */
3733 3762 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3734 3763 &nvtop) == 0);
3735 3764 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3736 3765 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3737 3766 VDEV_ALLOC_ROOTPOOL);
3738 3767 spa_config_exit(spa, SCL_ALL, FTAG);
3739 3768 if (error) {
3740 3769 mutex_exit(&spa_namespace_lock);
3741 3770 nvlist_free(config);
3742 3771 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3743 3772 pname);
3744 3773 return (error);
3745 3774 }
3746 3775
3747 3776 /*
3748 3777 * Get the boot vdev.
3749 3778 */
3750 3779 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3751 3780 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3752 3781 (u_longlong_t)guid);
3753 3782 error = SET_ERROR(ENOENT);
3754 3783 goto out;
3755 3784 }
3756 3785
3757 3786 /*
3758 3787 * Determine if there is a better boot device.
3759 3788 */
3760 3789 avd = bvd;
3761 3790 spa_alt_rootvdev(rvd, &avd, &txg);
3762 3791 if (avd != bvd) {
3763 3792 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3764 3793 "try booting from '%s'", avd->vdev_path);
3765 3794 error = SET_ERROR(EINVAL);
3766 3795 goto out;
3767 3796 }
3768 3797
3769 3798 /*
3770 3799 * If the boot device is part of a spare vdev then ensure that
3771 3800 * we're booting off the active spare.
3772 3801 */
3773 3802 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3774 3803 !bvd->vdev_isspare) {
3775 3804 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3776 3805 "try booting from '%s'",
3777 3806 bvd->vdev_parent->
3778 3807 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3779 3808 error = SET_ERROR(EINVAL);
3780 3809 goto out;
3781 3810 }
3782 3811
3783 3812 error = 0;
3784 3813 out:
3785 3814 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3786 3815 vdev_free(rvd);
3787 3816 spa_config_exit(spa, SCL_ALL, FTAG);
3788 3817 mutex_exit(&spa_namespace_lock);
3789 3818
3790 3819 nvlist_free(config);
3791 3820 return (error);
3792 3821 }
3793 3822
3794 3823 #endif
3795 3824
3796 3825 /*
3797 3826 * Import a non-root pool into the system.
3798 3827 */
3799 3828 int
3800 3829 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3801 3830 {
3802 3831 spa_t *spa;
3803 3832 char *altroot = NULL;
3804 3833 spa_load_state_t state = SPA_LOAD_IMPORT;
3805 3834 zpool_rewind_policy_t policy;
3806 3835 uint64_t mode = spa_mode_global;
3807 3836 uint64_t readonly = B_FALSE;
3808 3837 int error;
3809 3838 nvlist_t *nvroot;
3810 3839 nvlist_t **spares, **l2cache;
3811 3840 uint_t nspares, nl2cache;
3812 3841
3813 3842 /*
3814 3843 * If a pool with this name exists, return failure.
3815 3844 */
3816 3845 mutex_enter(&spa_namespace_lock);
3817 3846 if (spa_lookup(pool) != NULL) {
3818 3847 mutex_exit(&spa_namespace_lock);
3819 3848 return (SET_ERROR(EEXIST));
3820 3849 }
3821 3850
3822 3851 /*
3823 3852 * Create and initialize the spa structure.
3824 3853 */
3825 3854 (void) nvlist_lookup_string(props,
3826 3855 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3827 3856 (void) nvlist_lookup_uint64(props,
3828 3857 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3829 3858 if (readonly)
3830 3859 mode = FREAD;
3831 3860 spa = spa_add(pool, config, altroot);
3832 3861 spa->spa_import_flags = flags;
3833 3862
3834 3863 /*
3835 3864 * Verbatim import - Take a pool and insert it into the namespace
3836 3865 * as if it had been loaded at boot.
3837 3866 */
3838 3867 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3839 3868 if (props != NULL)
3840 3869 spa_configfile_set(spa, props, B_FALSE);
3841 3870
3842 3871 spa_config_sync(spa, B_FALSE, B_TRUE);
3843 3872
3844 3873 mutex_exit(&spa_namespace_lock);
3845 3874 return (0);
3846 3875 }
3847 3876
3848 3877 spa_activate(spa, mode);
3849 3878
3850 3879 /*
3851 3880 * Don't start async tasks until we know everything is healthy.
3852 3881 */
3853 3882 spa_async_suspend(spa);
3854 3883
3855 3884 zpool_get_rewind_policy(config, &policy);
3856 3885 if (policy.zrp_request & ZPOOL_DO_REWIND)
3857 3886 state = SPA_LOAD_RECOVER;
3858 3887
3859 3888 /*
3860 3889 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
3861 3890 * because the user-supplied config is actually the one to trust when
3862 3891 * doing an import.
3863 3892 */
3864 3893 if (state != SPA_LOAD_RECOVER)
3865 3894 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3866 3895
3867 3896 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3868 3897 policy.zrp_request);
3869 3898
3870 3899 /*
3871 3900 * Propagate anything learned while loading the pool and pass it
3872 3901 * back to caller (i.e. rewind info, missing devices, etc).
3873 3902 */
3874 3903 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3875 3904 spa->spa_load_info) == 0);
3876 3905
3877 3906 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3878 3907 /*
3879 3908 * Toss any existing sparelist, as it doesn't have any validity
3880 3909 * anymore, and conflicts with spa_has_spare().
3881 3910 */
3882 3911 if (spa->spa_spares.sav_config) {
3883 3912 nvlist_free(spa->spa_spares.sav_config);
3884 3913 spa->spa_spares.sav_config = NULL;
3885 3914 spa_load_spares(spa);
3886 3915 }
3887 3916 if (spa->spa_l2cache.sav_config) {
3888 3917 nvlist_free(spa->spa_l2cache.sav_config);
3889 3918 spa->spa_l2cache.sav_config = NULL;
3890 3919 spa_load_l2cache(spa);
3891 3920 }
3892 3921
3893 3922 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3894 3923 &nvroot) == 0);
3895 3924 if (error == 0)
3896 3925 error = spa_validate_aux(spa, nvroot, -1ULL,
3897 3926 VDEV_ALLOC_SPARE);
3898 3927 if (error == 0)
3899 3928 error = spa_validate_aux(spa, nvroot, -1ULL,
3900 3929 VDEV_ALLOC_L2CACHE);
3901 3930 spa_config_exit(spa, SCL_ALL, FTAG);
3902 3931
3903 3932 if (props != NULL)
3904 3933 spa_configfile_set(spa, props, B_FALSE);
3905 3934
3906 3935 if (error != 0 || (props && spa_writeable(spa) &&
3907 3936 (error = spa_prop_set(spa, props)))) {
3908 3937 spa_unload(spa);
3909 3938 spa_deactivate(spa);
3910 3939 spa_remove(spa);
3911 3940 mutex_exit(&spa_namespace_lock);
3912 3941 return (error);
3913 3942 }
3914 3943
3915 3944 spa_async_resume(spa);
3916 3945
3917 3946 /*
3918 3947 * Override any spares and level 2 cache devices as specified by
3919 3948 * the user, as these may have correct device names/devids, etc.
3920 3949 */
3921 3950 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3922 3951 &spares, &nspares) == 0) {
3923 3952 if (spa->spa_spares.sav_config)
3924 3953 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3925 3954 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3926 3955 else
3927 3956 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3928 3957 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3929 3958 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3930 3959 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3931 3960 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3932 3961 spa_load_spares(spa);
3933 3962 spa_config_exit(spa, SCL_ALL, FTAG);
3934 3963 spa->spa_spares.sav_sync = B_TRUE;
3935 3964 }
3936 3965 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3937 3966 &l2cache, &nl2cache) == 0) {
3938 3967 if (spa->spa_l2cache.sav_config)
3939 3968 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3940 3969 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3941 3970 else
3942 3971 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3943 3972 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3944 3973 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3945 3974 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3946 3975 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3947 3976 spa_load_l2cache(spa);
3948 3977 spa_config_exit(spa, SCL_ALL, FTAG);
3949 3978 spa->spa_l2cache.sav_sync = B_TRUE;
3950 3979 }
3951 3980
3952 3981 /*
3953 3982 * Check for any removed devices.
3954 3983 */
3955 3984 if (spa->spa_autoreplace) {
3956 3985 spa_aux_check_removed(&spa->spa_spares);
3957 3986 spa_aux_check_removed(&spa->spa_l2cache);
3958 3987 }
3959 3988
3960 3989 if (spa_writeable(spa)) {
3961 3990 /*
3962 3991 * Update the config cache to include the newly-imported pool.
3963 3992 */
3964 3993 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3965 3994 }
3966 3995
3967 3996 /*
3968 3997 * It's possible that the pool was expanded while it was exported.
3969 3998 * We kick off an async task to handle this for us.
3970 3999 */
3971 4000 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
3972 4001
3973 4002 mutex_exit(&spa_namespace_lock);
3974 4003 spa_history_log_version(spa, "import");
3975 4004
3976 4005 return (0);
3977 4006 }
3978 4007
3979 4008 nvlist_t *
3980 4009 spa_tryimport(nvlist_t *tryconfig)
3981 4010 {
3982 4011 nvlist_t *config = NULL;
3983 4012 char *poolname;
3984 4013 spa_t *spa;
3985 4014 uint64_t state;
3986 4015 int error;
3987 4016
3988 4017 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
3989 4018 return (NULL);
3990 4019
3991 4020 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
3992 4021 return (NULL);
3993 4022
3994 4023 /*
3995 4024 * Create and initialize the spa structure.
3996 4025 */
3997 4026 mutex_enter(&spa_namespace_lock);
3998 4027 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
3999 4028 spa_activate(spa, FREAD);
4000 4029
4001 4030 /*
4002 4031 * Pass off the heavy lifting to spa_load().
4003 4032 * Pass TRUE for mosconfig because the user-supplied config
4004 4033 * is actually the one to trust when doing an import.
4005 4034 */
4006 4035 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
4007 4036
4008 4037 /*
4009 4038 * If 'tryconfig' was at least parsable, return the current config.
4010 4039 */
4011 4040 if (spa->spa_root_vdev != NULL) {
4012 4041 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4013 4042 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4014 4043 poolname) == 0);
4015 4044 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4016 4045 state) == 0);
4017 4046 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4018 4047 spa->spa_uberblock.ub_timestamp) == 0);
4019 4048 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4020 4049 spa->spa_load_info) == 0);
4021 4050
4022 4051 /*
4023 4052 * If the bootfs property exists on this pool then we
4024 4053 * copy it out so that external consumers can tell which
4025 4054 * pools are bootable.
4026 4055 */
4027 4056 if ((!error || error == EEXIST) && spa->spa_bootfs) {
4028 4057 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4029 4058
4030 4059 /*
4031 4060 * We have to play games with the name since the
4032 4061 * pool was opened as TRYIMPORT_NAME.
4033 4062 */
4034 4063 if (dsl_dsobj_to_dsname(spa_name(spa),
4035 4064 spa->spa_bootfs, tmpname) == 0) {
4036 4065 char *cp;
4037 4066 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4038 4067
4039 4068 cp = strchr(tmpname, '/');
4040 4069 if (cp == NULL) {
4041 4070 (void) strlcpy(dsname, tmpname,
4042 4071 MAXPATHLEN);
4043 4072 } else {
4044 4073 (void) snprintf(dsname, MAXPATHLEN,
4045 4074 "%s/%s", poolname, ++cp);
4046 4075 }
4047 4076 VERIFY(nvlist_add_string(config,
4048 4077 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4049 4078 kmem_free(dsname, MAXPATHLEN);
4050 4079 }
4051 4080 kmem_free(tmpname, MAXPATHLEN);
4052 4081 }
4053 4082
4054 4083 /*
4055 4084 * Add the list of hot spares and level 2 cache devices.
4056 4085 */
4057 4086 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4058 4087 spa_add_spares(spa, config);
4059 4088 spa_add_l2cache(spa, config);
4060 4089 spa_config_exit(spa, SCL_CONFIG, FTAG);
4061 4090 }
4062 4091
4063 4092 spa_unload(spa);
4064 4093 spa_deactivate(spa);
4065 4094 spa_remove(spa);
4066 4095 mutex_exit(&spa_namespace_lock);
4067 4096
4068 4097 return (config);
4069 4098 }
4070 4099
4071 4100 /*
4072 4101 * Pool export/destroy
4073 4102 *
4074 4103 * The act of destroying or exporting a pool is very simple. We make sure there
4075 4104 * is no more pending I/O and any references to the pool are gone. Then, we
4076 4105 * update the pool state and sync all the labels to disk, removing the
4077 4106 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4078 4107 * we don't sync the labels or remove the configuration cache.
4079 4108 */
4080 4109 static int
4081 4110 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4082 4111 boolean_t force, boolean_t hardforce)
4083 4112 {
4084 4113 spa_t *spa;
4085 4114
4086 4115 if (oldconfig)
4087 4116 *oldconfig = NULL;
4088 4117
4089 4118 if (!(spa_mode_global & FWRITE))
4090 4119 return (SET_ERROR(EROFS));
4091 4120
4092 4121 mutex_enter(&spa_namespace_lock);
4093 4122 if ((spa = spa_lookup(pool)) == NULL) {
4094 4123 mutex_exit(&spa_namespace_lock);
4095 4124 return (SET_ERROR(ENOENT));
4096 4125 }
4097 4126
4098 4127 /*
4099 4128 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4100 4129 * reacquire the namespace lock, and see if we can export.
4101 4130 */
4102 4131 spa_open_ref(spa, FTAG);
4103 4132 mutex_exit(&spa_namespace_lock);
4104 4133 spa_async_suspend(spa);
4105 4134 mutex_enter(&spa_namespace_lock);
4106 4135 spa_close(spa, FTAG);
4107 4136
4108 4137 /*
4109 4138 * The pool will be in core if it's openable,
4110 4139 * in which case we can modify its state.
4111 4140 */
4112 4141 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4113 4142 /*
4114 4143 * Objsets may be open only because they're dirty, so we
4115 4144 * have to force it to sync before checking spa_refcnt.
4116 4145 */
4117 4146 txg_wait_synced(spa->spa_dsl_pool, 0);
4118 4147
4119 4148 /*
4120 4149 * A pool cannot be exported or destroyed if there are active
4121 4150 * references. If we are resetting a pool, allow references by
4122 4151 * fault injection handlers.
4123 4152 */
4124 4153 if (!spa_refcount_zero(spa) ||
4125 4154 (spa->spa_inject_ref != 0 &&
4126 4155 new_state != POOL_STATE_UNINITIALIZED)) {
4127 4156 spa_async_resume(spa);
4128 4157 mutex_exit(&spa_namespace_lock);
4129 4158 return (SET_ERROR(EBUSY));
4130 4159 }
4131 4160
4132 4161 /*
4133 4162 * A pool cannot be exported if it has an active shared spare.
4134 4163 * This is to prevent other pools stealing the active spare
4135 4164 * from an exported pool. At user's own will, such pool can
4136 4165 * be forcedly exported.
4137 4166 */
4138 4167 if (!force && new_state == POOL_STATE_EXPORTED &&
4139 4168 spa_has_active_shared_spare(spa)) {
4140 4169 spa_async_resume(spa);
4141 4170 mutex_exit(&spa_namespace_lock);
4142 4171 return (SET_ERROR(EXDEV));
4143 4172 }
4144 4173
4145 4174 /*
4146 4175 * We want this to be reflected on every label,
4147 4176 * so mark them all dirty. spa_unload() will do the
4148 4177 * final sync that pushes these changes out.
4149 4178 */
4150 4179 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4151 4180 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4152 4181 spa->spa_state = new_state;
4153 4182 spa->spa_final_txg = spa_last_synced_txg(spa) +
4154 4183 TXG_DEFER_SIZE + 1;
4155 4184 vdev_config_dirty(spa->spa_root_vdev);
4156 4185 spa_config_exit(spa, SCL_ALL, FTAG);
4157 4186 }
4158 4187 }
4159 4188
4160 4189 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4161 4190
4162 4191 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4163 4192 spa_unload(spa);
4164 4193 spa_deactivate(spa);
4165 4194 }
4166 4195
4167 4196 if (oldconfig && spa->spa_config)
4168 4197 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4169 4198
4170 4199 if (new_state != POOL_STATE_UNINITIALIZED) {
4171 4200 if (!hardforce)
4172 4201 spa_config_sync(spa, B_TRUE, B_TRUE);
4173 4202 spa_remove(spa);
4174 4203 }
4175 4204 mutex_exit(&spa_namespace_lock);
4176 4205
4177 4206 return (0);
4178 4207 }
4179 4208
4180 4209 /*
4181 4210 * Destroy a storage pool.
4182 4211 */
4183 4212 int
4184 4213 spa_destroy(char *pool)
4185 4214 {
4186 4215 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4187 4216 B_FALSE, B_FALSE));
4188 4217 }
4189 4218
4190 4219 /*
4191 4220 * Export a storage pool.
4192 4221 */
4193 4222 int
4194 4223 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4195 4224 boolean_t hardforce)
4196 4225 {
4197 4226 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4198 4227 force, hardforce));
4199 4228 }
4200 4229
4201 4230 /*
4202 4231 * Similar to spa_export(), this unloads the spa_t without actually removing it
4203 4232 * from the namespace in any way.
4204 4233 */
4205 4234 int
4206 4235 spa_reset(char *pool)
4207 4236 {
4208 4237 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4209 4238 B_FALSE, B_FALSE));
4210 4239 }
4211 4240
4212 4241 /*
4213 4242 * ==========================================================================
4214 4243 * Device manipulation
4215 4244 * ==========================================================================
4216 4245 */
4217 4246
4218 4247 /*
4219 4248 * Add a device to a storage pool.
4220 4249 */
4221 4250 int
4222 4251 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4223 4252 {
4224 4253 uint64_t txg, id;
4225 4254 int error;
4226 4255 vdev_t *rvd = spa->spa_root_vdev;
4227 4256 vdev_t *vd, *tvd;
4228 4257 nvlist_t **spares, **l2cache;
4229 4258 uint_t nspares, nl2cache;
4230 4259
4231 4260 ASSERT(spa_writeable(spa));
4232 4261
4233 4262 txg = spa_vdev_enter(spa);
4234 4263
4235 4264 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4236 4265 VDEV_ALLOC_ADD)) != 0)
4237 4266 return (spa_vdev_exit(spa, NULL, txg, error));
4238 4267
4239 4268 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
4240 4269
4241 4270 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4242 4271 &nspares) != 0)
4243 4272 nspares = 0;
4244 4273
4245 4274 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4246 4275 &nl2cache) != 0)
4247 4276 nl2cache = 0;
4248 4277
4249 4278 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4250 4279 return (spa_vdev_exit(spa, vd, txg, EINVAL));
4251 4280
4252 4281 if (vd->vdev_children != 0 &&
4253 4282 (error = vdev_create(vd, txg, B_FALSE)) != 0)
4254 4283 return (spa_vdev_exit(spa, vd, txg, error));
4255 4284
4256 4285 /*
4257 4286 * We must validate the spares and l2cache devices after checking the
4258 4287 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4259 4288 */
4260 4289 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4261 4290 return (spa_vdev_exit(spa, vd, txg, error));
4262 4291
4263 4292 /*
4264 4293 * Transfer each new top-level vdev from vd to rvd.
4265 4294 */
4266 4295 for (int c = 0; c < vd->vdev_children; c++) {
4267 4296
4268 4297 /*
4269 4298 * Set the vdev id to the first hole, if one exists.
4270 4299 */
4271 4300 for (id = 0; id < rvd->vdev_children; id++) {
4272 4301 if (rvd->vdev_child[id]->vdev_ishole) {
4273 4302 vdev_free(rvd->vdev_child[id]);
4274 4303 break;
4275 4304 }
4276 4305 }
4277 4306 tvd = vd->vdev_child[c];
4278 4307 vdev_remove_child(vd, tvd);
4279 4308 tvd->vdev_id = id;
4280 4309 vdev_add_child(rvd, tvd);
4281 4310 vdev_config_dirty(tvd);
4282 4311 }
4283 4312
4284 4313 if (nspares != 0) {
4285 4314 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4286 4315 ZPOOL_CONFIG_SPARES);
4287 4316 spa_load_spares(spa);
4288 4317 spa->spa_spares.sav_sync = B_TRUE;
4289 4318 }
4290 4319
4291 4320 if (nl2cache != 0) {
4292 4321 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4293 4322 ZPOOL_CONFIG_L2CACHE);
4294 4323 spa_load_l2cache(spa);
4295 4324 spa->spa_l2cache.sav_sync = B_TRUE;
4296 4325 }
4297 4326
4298 4327 /*
4299 4328 * We have to be careful when adding new vdevs to an existing pool.
4300 4329 * If other threads start allocating from these vdevs before we
4301 4330 * sync the config cache, and we lose power, then upon reboot we may
4302 4331 * fail to open the pool because there are DVAs that the config cache
4303 4332 * can't translate. Therefore, we first add the vdevs without
4304 4333 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4305 4334 * and then let spa_config_update() initialize the new metaslabs.
4306 4335 *
4307 4336 * spa_load() checks for added-but-not-initialized vdevs, so that
4308 4337 * if we lose power at any point in this sequence, the remaining
4309 4338 * steps will be completed the next time we load the pool.
4310 4339 */
4311 4340 (void) spa_vdev_exit(spa, vd, txg, 0);
4312 4341
4313 4342 mutex_enter(&spa_namespace_lock);
4314 4343 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4315 4344 mutex_exit(&spa_namespace_lock);
4316 4345
4317 4346 return (0);
4318 4347 }
4319 4348
4320 4349 /*
4321 4350 * Attach a device to a mirror. The arguments are the path to any device
4322 4351 * in the mirror, and the nvroot for the new device. If the path specifies
4323 4352 * a device that is not mirrored, we automatically insert the mirror vdev.
4324 4353 *
4325 4354 * If 'replacing' is specified, the new device is intended to replace the
4326 4355 * existing device; in this case the two devices are made into their own
4327 4356 * mirror using the 'replacing' vdev, which is functionally identical to
4328 4357 * the mirror vdev (it actually reuses all the same ops) but has a few
4329 4358 * extra rules: you can't attach to it after it's been created, and upon
4330 4359 * completion of resilvering, the first disk (the one being replaced)
4331 4360 * is automatically detached.
4332 4361 */
4333 4362 int
4334 4363 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4335 4364 {
4336 4365 uint64_t txg, dtl_max_txg;
4337 4366 vdev_t *rvd = spa->spa_root_vdev;
4338 4367 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4339 4368 vdev_ops_t *pvops;
4340 4369 char *oldvdpath, *newvdpath;
4341 4370 int newvd_isspare;
4342 4371 int error;
4343 4372
4344 4373 ASSERT(spa_writeable(spa));
4345 4374
4346 4375 txg = spa_vdev_enter(spa);
4347 4376
4348 4377 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4349 4378
4350 4379 if (oldvd == NULL)
4351 4380 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4352 4381
4353 4382 if (!oldvd->vdev_ops->vdev_op_leaf)
4354 4383 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4355 4384
4356 4385 pvd = oldvd->vdev_parent;
4357 4386
4358 4387 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4359 4388 VDEV_ALLOC_ATTACH)) != 0)
4360 4389 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4361 4390
4362 4391 if (newrootvd->vdev_children != 1)
4363 4392 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4364 4393
4365 4394 newvd = newrootvd->vdev_child[0];
4366 4395
4367 4396 if (!newvd->vdev_ops->vdev_op_leaf)
4368 4397 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4369 4398
4370 4399 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4371 4400 return (spa_vdev_exit(spa, newrootvd, txg, error));
4372 4401
4373 4402 /*
4374 4403 * Spares can't replace logs
4375 4404 */
4376 4405 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4377 4406 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4378 4407
4379 4408 if (!replacing) {
4380 4409 /*
4381 4410 * For attach, the only allowable parent is a mirror or the root
4382 4411 * vdev.
4383 4412 */
4384 4413 if (pvd->vdev_ops != &vdev_mirror_ops &&
4385 4414 pvd->vdev_ops != &vdev_root_ops)
4386 4415 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4387 4416
4388 4417 pvops = &vdev_mirror_ops;
4389 4418 } else {
4390 4419 /*
4391 4420 * Active hot spares can only be replaced by inactive hot
4392 4421 * spares.
4393 4422 */
4394 4423 if (pvd->vdev_ops == &vdev_spare_ops &&
4395 4424 oldvd->vdev_isspare &&
4396 4425 !spa_has_spare(spa, newvd->vdev_guid))
4397 4426 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4398 4427
4399 4428 /*
4400 4429 * If the source is a hot spare, and the parent isn't already a
4401 4430 * spare, then we want to create a new hot spare. Otherwise, we
4402 4431 * want to create a replacing vdev. The user is not allowed to
4403 4432 * attach to a spared vdev child unless the 'isspare' state is
4404 4433 * the same (spare replaces spare, non-spare replaces
4405 4434 * non-spare).
4406 4435 */
4407 4436 if (pvd->vdev_ops == &vdev_replacing_ops &&
4408 4437 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4409 4438 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4410 4439 } else if (pvd->vdev_ops == &vdev_spare_ops &&
4411 4440 newvd->vdev_isspare != oldvd->vdev_isspare) {
4412 4441 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4413 4442 }
4414 4443
4415 4444 if (newvd->vdev_isspare)
4416 4445 pvops = &vdev_spare_ops;
4417 4446 else
4418 4447 pvops = &vdev_replacing_ops;
4419 4448 }
4420 4449
4421 4450 /*
4422 4451 * Make sure the new device is big enough.
4423 4452 */
4424 4453 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4425 4454 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4426 4455
4427 4456 /*
4428 4457 * The new device cannot have a higher alignment requirement
4429 4458 * than the top-level vdev.
4430 4459 */
4431 4460 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4432 4461 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4433 4462
4434 4463 /*
4435 4464 * If this is an in-place replacement, update oldvd's path and devid
4436 4465 * to make it distinguishable from newvd, and unopenable from now on.
4437 4466 */
4438 4467 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4439 4468 spa_strfree(oldvd->vdev_path);
4440 4469 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4441 4470 KM_SLEEP);
4442 4471 (void) sprintf(oldvd->vdev_path, "%s/%s",
4443 4472 newvd->vdev_path, "old");
4444 4473 if (oldvd->vdev_devid != NULL) {
4445 4474 spa_strfree(oldvd->vdev_devid);
4446 4475 oldvd->vdev_devid = NULL;
4447 4476 }
4448 4477 }
4449 4478
4450 4479 /* mark the device being resilvered */
4451 4480 newvd->vdev_resilver_txg = txg;
4452 4481
4453 4482 /*
4454 4483 * If the parent is not a mirror, or if we're replacing, insert the new
4455 4484 * mirror/replacing/spare vdev above oldvd.
4456 4485 */
4457 4486 if (pvd->vdev_ops != pvops)
4458 4487 pvd = vdev_add_parent(oldvd, pvops);
4459 4488
4460 4489 ASSERT(pvd->vdev_top->vdev_parent == rvd);
4461 4490 ASSERT(pvd->vdev_ops == pvops);
4462 4491 ASSERT(oldvd->vdev_parent == pvd);
4463 4492
4464 4493 /*
4465 4494 * Extract the new device from its root and add it to pvd.
4466 4495 */
4467 4496 vdev_remove_child(newrootvd, newvd);
4468 4497 newvd->vdev_id = pvd->vdev_children;
4469 4498 newvd->vdev_crtxg = oldvd->vdev_crtxg;
4470 4499 vdev_add_child(pvd, newvd);
4471 4500
4472 4501 tvd = newvd->vdev_top;
4473 4502 ASSERT(pvd->vdev_top == tvd);
4474 4503 ASSERT(tvd->vdev_parent == rvd);
4475 4504
4476 4505 vdev_config_dirty(tvd);
4477 4506
4478 4507 /*
4479 4508 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4480 4509 * for any dmu_sync-ed blocks. It will propagate upward when
4481 4510 * spa_vdev_exit() calls vdev_dtl_reassess().
4482 4511 */
4483 4512 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4484 4513
4485 4514 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4486 4515 dtl_max_txg - TXG_INITIAL);
4487 4516
4488 4517 if (newvd->vdev_isspare) {
4489 4518 spa_spare_activate(newvd);
4490 4519 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4491 4520 }
4492 4521
4493 4522 oldvdpath = spa_strdup(oldvd->vdev_path);
4494 4523 newvdpath = spa_strdup(newvd->vdev_path);
4495 4524 newvd_isspare = newvd->vdev_isspare;
4496 4525
4497 4526 /*
4498 4527 * Mark newvd's DTL dirty in this txg.
4499 4528 */
4500 4529 vdev_dirty(tvd, VDD_DTL, newvd, txg);
4501 4530
4502 4531 /*
4503 4532 * Schedule the resilver to restart in the future. We do this to
4504 4533 * ensure that dmu_sync-ed blocks have been stitched into the
4505 4534 * respective datasets.
4506 4535 */
4507 4536 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4508 4537
4509 4538 /*
4510 4539 * Commit the config
4511 4540 */
4512 4541 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4513 4542
4514 4543 spa_history_log_internal(spa, "vdev attach", NULL,
4515 4544 "%s vdev=%s %s vdev=%s",
4516 4545 replacing && newvd_isspare ? "spare in" :
4517 4546 replacing ? "replace" : "attach", newvdpath,
4518 4547 replacing ? "for" : "to", oldvdpath);
4519 4548
4520 4549 spa_strfree(oldvdpath);
4521 4550 spa_strfree(newvdpath);
4522 4551
4523 4552 if (spa->spa_bootfs)
4524 4553 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4525 4554
4526 4555 return (0);
4527 4556 }
4528 4557
4529 4558 /*
4530 4559 * Detach a device from a mirror or replacing vdev.
4531 4560 *
4532 4561 * If 'replace_done' is specified, only detach if the parent
4533 4562 * is a replacing vdev.
4534 4563 */
4535 4564 int
4536 4565 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4537 4566 {
4538 4567 uint64_t txg;
4539 4568 int error;
4540 4569 vdev_t *rvd = spa->spa_root_vdev;
4541 4570 vdev_t *vd, *pvd, *cvd, *tvd;
4542 4571 boolean_t unspare = B_FALSE;
4543 4572 uint64_t unspare_guid = 0;
4544 4573 char *vdpath;
4545 4574
4546 4575 ASSERT(spa_writeable(spa));
4547 4576
4548 4577 txg = spa_vdev_enter(spa);
4549 4578
4550 4579 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4551 4580
4552 4581 if (vd == NULL)
4553 4582 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4554 4583
4555 4584 if (!vd->vdev_ops->vdev_op_leaf)
4556 4585 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4557 4586
4558 4587 pvd = vd->vdev_parent;
4559 4588
4560 4589 /*
4561 4590 * If the parent/child relationship is not as expected, don't do it.
4562 4591 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4563 4592 * vdev that's replacing B with C. The user's intent in replacing
4564 4593 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4565 4594 * the replace by detaching C, the expected behavior is to end up
4566 4595 * M(A,B). But suppose that right after deciding to detach C,
4567 4596 * the replacement of B completes. We would have M(A,C), and then
4568 4597 * ask to detach C, which would leave us with just A -- not what
4569 4598 * the user wanted. To prevent this, we make sure that the
4570 4599 * parent/child relationship hasn't changed -- in this example,
4571 4600 * that C's parent is still the replacing vdev R.
4572 4601 */
4573 4602 if (pvd->vdev_guid != pguid && pguid != 0)
4574 4603 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4575 4604
4576 4605 /*
4577 4606 * Only 'replacing' or 'spare' vdevs can be replaced.
4578 4607 */
4579 4608 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4580 4609 pvd->vdev_ops != &vdev_spare_ops)
4581 4610 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4582 4611
4583 4612 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4584 4613 spa_version(spa) >= SPA_VERSION_SPARES);
4585 4614
4586 4615 /*
4587 4616 * Only mirror, replacing, and spare vdevs support detach.
4588 4617 */
4589 4618 if (pvd->vdev_ops != &vdev_replacing_ops &&
4590 4619 pvd->vdev_ops != &vdev_mirror_ops &&
4591 4620 pvd->vdev_ops != &vdev_spare_ops)
4592 4621 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4593 4622
4594 4623 /*
4595 4624 * If this device has the only valid copy of some data,
4596 4625 * we cannot safely detach it.
4597 4626 */
4598 4627 if (vdev_dtl_required(vd))
4599 4628 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4600 4629
4601 4630 ASSERT(pvd->vdev_children >= 2);
4602 4631
4603 4632 /*
4604 4633 * If we are detaching the second disk from a replacing vdev, then
4605 4634 * check to see if we changed the original vdev's path to have "/old"
4606 4635 * at the end in spa_vdev_attach(). If so, undo that change now.
4607 4636 */
4608 4637 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4609 4638 vd->vdev_path != NULL) {
4610 4639 size_t len = strlen(vd->vdev_path);
4611 4640
4612 4641 for (int c = 0; c < pvd->vdev_children; c++) {
4613 4642 cvd = pvd->vdev_child[c];
4614 4643
4615 4644 if (cvd == vd || cvd->vdev_path == NULL)
4616 4645 continue;
4617 4646
4618 4647 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4619 4648 strcmp(cvd->vdev_path + len, "/old") == 0) {
4620 4649 spa_strfree(cvd->vdev_path);
4621 4650 cvd->vdev_path = spa_strdup(vd->vdev_path);
4622 4651 break;
4623 4652 }
4624 4653 }
4625 4654 }
4626 4655
4627 4656 /*
4628 4657 * If we are detaching the original disk from a spare, then it implies
4629 4658 * that the spare should become a real disk, and be removed from the
4630 4659 * active spare list for the pool.
4631 4660 */
4632 4661 if (pvd->vdev_ops == &vdev_spare_ops &&
4633 4662 vd->vdev_id == 0 &&
4634 4663 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4635 4664 unspare = B_TRUE;
4636 4665
4637 4666 /*
4638 4667 * Erase the disk labels so the disk can be used for other things.
4639 4668 * This must be done after all other error cases are handled,
4640 4669 * but before we disembowel vd (so we can still do I/O to it).
4641 4670 * But if we can't do it, don't treat the error as fatal --
4642 4671 * it may be that the unwritability of the disk is the reason
4643 4672 * it's being detached!
4644 4673 */
4645 4674 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4646 4675
4647 4676 /*
4648 4677 * Remove vd from its parent and compact the parent's children.
4649 4678 */
4650 4679 vdev_remove_child(pvd, vd);
4651 4680 vdev_compact_children(pvd);
4652 4681
4653 4682 /*
4654 4683 * Remember one of the remaining children so we can get tvd below.
4655 4684 */
4656 4685 cvd = pvd->vdev_child[pvd->vdev_children - 1];
4657 4686
4658 4687 /*
4659 4688 * If we need to remove the remaining child from the list of hot spares,
4660 4689 * do it now, marking the vdev as no longer a spare in the process.
4661 4690 * We must do this before vdev_remove_parent(), because that can
4662 4691 * change the GUID if it creates a new toplevel GUID. For a similar
4663 4692 * reason, we must remove the spare now, in the same txg as the detach;
4664 4693 * otherwise someone could attach a new sibling, change the GUID, and
4665 4694 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4666 4695 */
4667 4696 if (unspare) {
4668 4697 ASSERT(cvd->vdev_isspare);
4669 4698 spa_spare_remove(cvd);
4670 4699 unspare_guid = cvd->vdev_guid;
4671 4700 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4672 4701 cvd->vdev_unspare = B_TRUE;
4673 4702 }
4674 4703
4675 4704 /*
4676 4705 * If the parent mirror/replacing vdev only has one child,
4677 4706 * the parent is no longer needed. Remove it from the tree.
4678 4707 */
4679 4708 if (pvd->vdev_children == 1) {
4680 4709 if (pvd->vdev_ops == &vdev_spare_ops)
4681 4710 cvd->vdev_unspare = B_FALSE;
4682 4711 vdev_remove_parent(cvd);
4683 4712 }
4684 4713
4685 4714
4686 4715 /*
4687 4716 * We don't set tvd until now because the parent we just removed
4688 4717 * may have been the previous top-level vdev.
4689 4718 */
4690 4719 tvd = cvd->vdev_top;
4691 4720 ASSERT(tvd->vdev_parent == rvd);
4692 4721
4693 4722 /*
4694 4723 * Reevaluate the parent vdev state.
4695 4724 */
4696 4725 vdev_propagate_state(cvd);
4697 4726
4698 4727 /*
4699 4728 * If the 'autoexpand' property is set on the pool then automatically
4700 4729 * try to expand the size of the pool. For example if the device we
4701 4730 * just detached was smaller than the others, it may be possible to
4702 4731 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4703 4732 * first so that we can obtain the updated sizes of the leaf vdevs.
4704 4733 */
4705 4734 if (spa->spa_autoexpand) {
4706 4735 vdev_reopen(tvd);
4707 4736 vdev_expand(tvd, txg);
4708 4737 }
4709 4738
4710 4739 vdev_config_dirty(tvd);
4711 4740
4712 4741 /*
4713 4742 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
4714 4743 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4715 4744 * But first make sure we're not on any *other* txg's DTL list, to
4716 4745 * prevent vd from being accessed after it's freed.
4717 4746 */
4718 4747 vdpath = spa_strdup(vd->vdev_path);
4719 4748 for (int t = 0; t < TXG_SIZE; t++)
4720 4749 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4721 4750 vd->vdev_detached = B_TRUE;
4722 4751 vdev_dirty(tvd, VDD_DTL, vd, txg);
4723 4752
4724 4753 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4725 4754
4726 4755 /* hang on to the spa before we release the lock */
4727 4756 spa_open_ref(spa, FTAG);
4728 4757
4729 4758 error = spa_vdev_exit(spa, vd, txg, 0);
4730 4759
4731 4760 spa_history_log_internal(spa, "detach", NULL,
4732 4761 "vdev=%s", vdpath);
4733 4762 spa_strfree(vdpath);
4734 4763
4735 4764 /*
4736 4765 * If this was the removal of the original device in a hot spare vdev,
4737 4766 * then we want to go through and remove the device from the hot spare
4738 4767 * list of every other pool.
4739 4768 */
4740 4769 if (unspare) {
4741 4770 spa_t *altspa = NULL;
4742 4771
4743 4772 mutex_enter(&spa_namespace_lock);
4744 4773 while ((altspa = spa_next(altspa)) != NULL) {
4745 4774 if (altspa->spa_state != POOL_STATE_ACTIVE ||
4746 4775 altspa == spa)
4747 4776 continue;
4748 4777
4749 4778 spa_open_ref(altspa, FTAG);
4750 4779 mutex_exit(&spa_namespace_lock);
4751 4780 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4752 4781 mutex_enter(&spa_namespace_lock);
4753 4782 spa_close(altspa, FTAG);
4754 4783 }
4755 4784 mutex_exit(&spa_namespace_lock);
4756 4785
4757 4786 /* search the rest of the vdevs for spares to remove */
4758 4787 spa_vdev_resilver_done(spa);
4759 4788 }
4760 4789
4761 4790 /* all done with the spa; OK to release */
4762 4791 mutex_enter(&spa_namespace_lock);
4763 4792 spa_close(spa, FTAG);
4764 4793 mutex_exit(&spa_namespace_lock);
4765 4794
4766 4795 return (error);
4767 4796 }
4768 4797
4769 4798 /*
4770 4799 * Split a set of devices from their mirrors, and create a new pool from them.
4771 4800 */
4772 4801 int
4773 4802 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4774 4803 nvlist_t *props, boolean_t exp)
4775 4804 {
4776 4805 int error = 0;
4777 4806 uint64_t txg, *glist;
4778 4807 spa_t *newspa;
4779 4808 uint_t c, children, lastlog;
4780 4809 nvlist_t **child, *nvl, *tmp;
4781 4810 dmu_tx_t *tx;
4782 4811 char *altroot = NULL;
4783 4812 vdev_t *rvd, **vml = NULL; /* vdev modify list */
4784 4813 boolean_t activate_slog;
4785 4814
4786 4815 ASSERT(spa_writeable(spa));
4787 4816
4788 4817 txg = spa_vdev_enter(spa);
4789 4818
4790 4819 /* clear the log and flush everything up to now */
4791 4820 activate_slog = spa_passivate_log(spa);
4792 4821 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4793 4822 error = spa_offline_log(spa);
4794 4823 txg = spa_vdev_config_enter(spa);
4795 4824
4796 4825 if (activate_slog)
4797 4826 spa_activate_log(spa);
4798 4827
4799 4828 if (error != 0)
4800 4829 return (spa_vdev_exit(spa, NULL, txg, error));
4801 4830
4802 4831 /* check new spa name before going any further */
4803 4832 if (spa_lookup(newname) != NULL)
4804 4833 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4805 4834
4806 4835 /*
4807 4836 * scan through all the children to ensure they're all mirrors
4808 4837 */
4809 4838 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4810 4839 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4811 4840 &children) != 0)
4812 4841 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4813 4842
4814 4843 /* first, check to ensure we've got the right child count */
4815 4844 rvd = spa->spa_root_vdev;
4816 4845 lastlog = 0;
4817 4846 for (c = 0; c < rvd->vdev_children; c++) {
4818 4847 vdev_t *vd = rvd->vdev_child[c];
4819 4848
4820 4849 /* don't count the holes & logs as children */
4821 4850 if (vd->vdev_islog || vd->vdev_ishole) {
4822 4851 if (lastlog == 0)
4823 4852 lastlog = c;
4824 4853 continue;
4825 4854 }
4826 4855
4827 4856 lastlog = 0;
4828 4857 }
4829 4858 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4830 4859 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4831 4860
4832 4861 /* next, ensure no spare or cache devices are part of the split */
4833 4862 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4834 4863 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4835 4864 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4836 4865
4837 4866 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4838 4867 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4839 4868
4840 4869 /* then, loop over each vdev and validate it */
4841 4870 for (c = 0; c < children; c++) {
4842 4871 uint64_t is_hole = 0;
4843 4872
4844 4873 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4845 4874 &is_hole);
4846 4875
4847 4876 if (is_hole != 0) {
4848 4877 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4849 4878 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4850 4879 continue;
4851 4880 } else {
4852 4881 error = SET_ERROR(EINVAL);
4853 4882 break;
4854 4883 }
4855 4884 }
4856 4885
4857 4886 /* which disk is going to be split? */
4858 4887 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4859 4888 &glist[c]) != 0) {
4860 4889 error = SET_ERROR(EINVAL);
4861 4890 break;
4862 4891 }
4863 4892
4864 4893 /* look it up in the spa */
4865 4894 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4866 4895 if (vml[c] == NULL) {
4867 4896 error = SET_ERROR(ENODEV);
4868 4897 break;
4869 4898 }
4870 4899
4871 4900 /* make sure there's nothing stopping the split */
4872 4901 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4873 4902 vml[c]->vdev_islog ||
4874 4903 vml[c]->vdev_ishole ||
4875 4904 vml[c]->vdev_isspare ||
4876 4905 vml[c]->vdev_isl2cache ||
4877 4906 !vdev_writeable(vml[c]) ||
4878 4907 vml[c]->vdev_children != 0 ||
4879 4908 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4880 4909 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4881 4910 error = SET_ERROR(EINVAL);
4882 4911 break;
4883 4912 }
4884 4913
4885 4914 if (vdev_dtl_required(vml[c])) {
4886 4915 error = SET_ERROR(EBUSY);
4887 4916 break;
4888 4917 }
4889 4918
4890 4919 /* we need certain info from the top level */
4891 4920 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4892 4921 vml[c]->vdev_top->vdev_ms_array) == 0);
4893 4922 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4894 4923 vml[c]->vdev_top->vdev_ms_shift) == 0);
4895 4924 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4896 4925 vml[c]->vdev_top->vdev_asize) == 0);
4897 4926 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4898 4927 vml[c]->vdev_top->vdev_ashift) == 0);
4899 4928 }
4900 4929
4901 4930 if (error != 0) {
4902 4931 kmem_free(vml, children * sizeof (vdev_t *));
4903 4932 kmem_free(glist, children * sizeof (uint64_t));
4904 4933 return (spa_vdev_exit(spa, NULL, txg, error));
4905 4934 }
4906 4935
4907 4936 /* stop writers from using the disks */
4908 4937 for (c = 0; c < children; c++) {
4909 4938 if (vml[c] != NULL)
4910 4939 vml[c]->vdev_offline = B_TRUE;
4911 4940 }
4912 4941 vdev_reopen(spa->spa_root_vdev);
4913 4942
4914 4943 /*
4915 4944 * Temporarily record the splitting vdevs in the spa config. This
4916 4945 * will disappear once the config is regenerated.
4917 4946 */
4918 4947 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4919 4948 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4920 4949 glist, children) == 0);
4921 4950 kmem_free(glist, children * sizeof (uint64_t));
4922 4951
4923 4952 mutex_enter(&spa->spa_props_lock);
4924 4953 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4925 4954 nvl) == 0);
4926 4955 mutex_exit(&spa->spa_props_lock);
4927 4956 spa->spa_config_splitting = nvl;
4928 4957 vdev_config_dirty(spa->spa_root_vdev);
4929 4958
4930 4959 /* configure and create the new pool */
4931 4960 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4932 4961 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4933 4962 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4934 4963 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4935 4964 spa_version(spa)) == 0);
4936 4965 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4937 4966 spa->spa_config_txg) == 0);
4938 4967 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4939 4968 spa_generate_guid(NULL)) == 0);
4940 4969 (void) nvlist_lookup_string(props,
4941 4970 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4942 4971
4943 4972 /* add the new pool to the namespace */
4944 4973 newspa = spa_add(newname, config, altroot);
4945 4974 newspa->spa_config_txg = spa->spa_config_txg;
4946 4975 spa_set_log_state(newspa, SPA_LOG_CLEAR);
4947 4976
4948 4977 /* release the spa config lock, retaining the namespace lock */
4949 4978 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4950 4979
4951 4980 if (zio_injection_enabled)
4952 4981 zio_handle_panic_injection(spa, FTAG, 1);
4953 4982
4954 4983 spa_activate(newspa, spa_mode_global);
4955 4984 spa_async_suspend(newspa);
4956 4985
4957 4986 /* create the new pool from the disks of the original pool */
4958 4987 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
4959 4988 if (error)
4960 4989 goto out;
4961 4990
4962 4991 /* if that worked, generate a real config for the new pool */
4963 4992 if (newspa->spa_root_vdev != NULL) {
4964 4993 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
4965 4994 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4966 4995 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
4967 4996 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
4968 4997 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
4969 4998 B_TRUE));
4970 4999 }
4971 5000
4972 5001 /* set the props */
4973 5002 if (props != NULL) {
4974 5003 spa_configfile_set(newspa, props, B_FALSE);
4975 5004 error = spa_prop_set(newspa, props);
4976 5005 if (error)
4977 5006 goto out;
4978 5007 }
4979 5008
4980 5009 /* flush everything */
4981 5010 txg = spa_vdev_config_enter(newspa);
4982 5011 vdev_config_dirty(newspa->spa_root_vdev);
4983 5012 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
4984 5013
4985 5014 if (zio_injection_enabled)
4986 5015 zio_handle_panic_injection(spa, FTAG, 2);
4987 5016
4988 5017 spa_async_resume(newspa);
4989 5018
4990 5019 /* finally, update the original pool's config */
4991 5020 txg = spa_vdev_config_enter(spa);
4992 5021 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
4993 5022 error = dmu_tx_assign(tx, TXG_WAIT);
4994 5023 if (error != 0)
4995 5024 dmu_tx_abort(tx);
4996 5025 for (c = 0; c < children; c++) {
4997 5026 if (vml[c] != NULL) {
4998 5027 vdev_split(vml[c]);
4999 5028 if (error == 0)
5000 5029 spa_history_log_internal(spa, "detach", tx,
5001 5030 "vdev=%s", vml[c]->vdev_path);
5002 5031 vdev_free(vml[c]);
5003 5032 }
5004 5033 }
5005 5034 vdev_config_dirty(spa->spa_root_vdev);
5006 5035 spa->spa_config_splitting = NULL;
5007 5036 nvlist_free(nvl);
5008 5037 if (error == 0)
5009 5038 dmu_tx_commit(tx);
5010 5039 (void) spa_vdev_exit(spa, NULL, txg, 0);
5011 5040
5012 5041 if (zio_injection_enabled)
5013 5042 zio_handle_panic_injection(spa, FTAG, 3);
5014 5043
5015 5044 /* split is complete; log a history record */
5016 5045 spa_history_log_internal(newspa, "split", NULL,
5017 5046 "from pool %s", spa_name(spa));
5018 5047
5019 5048 kmem_free(vml, children * sizeof (vdev_t *));
5020 5049
5021 5050 /* if we're not going to mount the filesystems in userland, export */
5022 5051 if (exp)
5023 5052 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5024 5053 B_FALSE, B_FALSE);
5025 5054
5026 5055 return (error);
5027 5056
5028 5057 out:
5029 5058 spa_unload(newspa);
5030 5059 spa_deactivate(newspa);
5031 5060 spa_remove(newspa);
5032 5061
5033 5062 txg = spa_vdev_config_enter(spa);
5034 5063
5035 5064 /* re-online all offlined disks */
5036 5065 for (c = 0; c < children; c++) {
5037 5066 if (vml[c] != NULL)
5038 5067 vml[c]->vdev_offline = B_FALSE;
5039 5068 }
5040 5069 vdev_reopen(spa->spa_root_vdev);
5041 5070
5042 5071 nvlist_free(spa->spa_config_splitting);
5043 5072 spa->spa_config_splitting = NULL;
5044 5073 (void) spa_vdev_exit(spa, NULL, txg, error);
5045 5074
5046 5075 kmem_free(vml, children * sizeof (vdev_t *));
5047 5076 return (error);
5048 5077 }
5049 5078
5050 5079 static nvlist_t *
5051 5080 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5052 5081 {
5053 5082 for (int i = 0; i < count; i++) {
5054 5083 uint64_t guid;
5055 5084
5056 5085 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5057 5086 &guid) == 0);
5058 5087
5059 5088 if (guid == target_guid)
5060 5089 return (nvpp[i]);
5061 5090 }
5062 5091
5063 5092 return (NULL);
5064 5093 }
5065 5094
5066 5095 static void
5067 5096 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5068 5097 nvlist_t *dev_to_remove)
5069 5098 {
5070 5099 nvlist_t **newdev = NULL;
5071 5100
5072 5101 if (count > 1)
5073 5102 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
5074 5103
5075 5104 for (int i = 0, j = 0; i < count; i++) {
5076 5105 if (dev[i] == dev_to_remove)
5077 5106 continue;
5078 5107 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
5079 5108 }
5080 5109
5081 5110 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5082 5111 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5083 5112
5084 5113 for (int i = 0; i < count - 1; i++)
5085 5114 nvlist_free(newdev[i]);
5086 5115
5087 5116 if (count > 1)
5088 5117 kmem_free(newdev, (count - 1) * sizeof (void *));
5089 5118 }
5090 5119
5091 5120 /*
5092 5121 * Evacuate the device.
5093 5122 */
5094 5123 static int
5095 5124 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5096 5125 {
5097 5126 uint64_t txg;
5098 5127 int error = 0;
5099 5128
5100 5129 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5101 5130 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5102 5131 ASSERT(vd == vd->vdev_top);
5103 5132
5104 5133 /*
5105 5134 * Evacuate the device. We don't hold the config lock as writer
5106 5135 * since we need to do I/O but we do keep the
5107 5136 * spa_namespace_lock held. Once this completes the device
5108 5137 * should no longer have any blocks allocated on it.
5109 5138 */
5110 5139 if (vd->vdev_islog) {
5111 5140 if (vd->vdev_stat.vs_alloc != 0)
5112 5141 error = spa_offline_log(spa);
5113 5142 } else {
5114 5143 error = SET_ERROR(ENOTSUP);
5115 5144 }
5116 5145
5117 5146 if (error)
5118 5147 return (error);
5119 5148
5120 5149 /*
5121 5150 * The evacuation succeeded. Remove any remaining MOS metadata
5122 5151 * associated with this vdev, and wait for these changes to sync.
5123 5152 */
5124 5153 ASSERT0(vd->vdev_stat.vs_alloc);
5125 5154 txg = spa_vdev_config_enter(spa);
5126 5155 vd->vdev_removing = B_TRUE;
5127 5156 vdev_dirty_leaves(vd, VDD_DTL, txg);
5128 5157 vdev_config_dirty(vd);
5129 5158 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5130 5159
5131 5160 return (0);
5132 5161 }
5133 5162
5134 5163 /*
5135 5164 * Complete the removal by cleaning up the namespace.
5136 5165 */
5137 5166 static void
5138 5167 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5139 5168 {
5140 5169 vdev_t *rvd = spa->spa_root_vdev;
5141 5170 uint64_t id = vd->vdev_id;
5142 5171 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5143 5172
5144 5173 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5145 5174 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5146 5175 ASSERT(vd == vd->vdev_top);
5147 5176
5148 5177 /*
5149 5178 * Only remove any devices which are empty.
5150 5179 */
5151 5180 if (vd->vdev_stat.vs_alloc != 0)
5152 5181 return;
5153 5182
5154 5183 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5155 5184
5156 5185 if (list_link_active(&vd->vdev_state_dirty_node))
5157 5186 vdev_state_clean(vd);
5158 5187 if (list_link_active(&vd->vdev_config_dirty_node))
5159 5188 vdev_config_clean(vd);
5160 5189
5161 5190 vdev_free(vd);
5162 5191
5163 5192 if (last_vdev) {
5164 5193 vdev_compact_children(rvd);
5165 5194 } else {
5166 5195 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5167 5196 vdev_add_child(rvd, vd);
5168 5197 }
5169 5198 vdev_config_dirty(rvd);
5170 5199
5171 5200 /*
5172 5201 * Reassess the health of our root vdev.
5173 5202 */
5174 5203 vdev_reopen(rvd);
5175 5204 }
5176 5205
5177 5206 /*
5178 5207 * Remove a device from the pool -
5179 5208 *
5180 5209 * Removing a device from the vdev namespace requires several steps
5181 5210 * and can take a significant amount of time. As a result we use
5182 5211 * the spa_vdev_config_[enter/exit] functions which allow us to
5183 5212 * grab and release the spa_config_lock while still holding the namespace
5184 5213 * lock. During each step the configuration is synced out.
5185 5214 *
5186 5215 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5187 5216 * devices.
5188 5217 */
5189 5218 int
5190 5219 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5191 5220 {
5192 5221 vdev_t *vd;
5193 5222 metaslab_group_t *mg;
5194 5223 nvlist_t **spares, **l2cache, *nv;
5195 5224 uint64_t txg = 0;
5196 5225 uint_t nspares, nl2cache;
5197 5226 int error = 0;
5198 5227 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5199 5228
5200 5229 ASSERT(spa_writeable(spa));
5201 5230
5202 5231 if (!locked)
5203 5232 txg = spa_vdev_enter(spa);
5204 5233
5205 5234 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5206 5235
5207 5236 if (spa->spa_spares.sav_vdevs != NULL &&
5208 5237 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5209 5238 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5210 5239 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5211 5240 /*
5212 5241 * Only remove the hot spare if it's not currently in use
5213 5242 * in this pool.
5214 5243 */
5215 5244 if (vd == NULL || unspare) {
5216 5245 spa_vdev_remove_aux(spa->spa_spares.sav_config,
5217 5246 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5218 5247 spa_load_spares(spa);
5219 5248 spa->spa_spares.sav_sync = B_TRUE;
5220 5249 } else {
5221 5250 error = SET_ERROR(EBUSY);
5222 5251 }
5223 5252 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
5224 5253 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5225 5254 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5226 5255 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5227 5256 /*
5228 5257 * Cache devices can always be removed.
5229 5258 */
5230 5259 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5231 5260 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5232 5261 spa_load_l2cache(spa);
5233 5262 spa->spa_l2cache.sav_sync = B_TRUE;
5234 5263 } else if (vd != NULL && vd->vdev_islog) {
5235 5264 ASSERT(!locked);
5236 5265 ASSERT(vd == vd->vdev_top);
5237 5266
5238 5267 /*
5239 5268 * XXX - Once we have bp-rewrite this should
5240 5269 * become the common case.
5241 5270 */
5242 5271
5243 5272 mg = vd->vdev_mg;
5244 5273
5245 5274 /*
5246 5275 * Stop allocating from this vdev.
5247 5276 */
5248 5277 metaslab_group_passivate(mg);
5249 5278
5250 5279 /*
5251 5280 * Wait for the youngest allocations and frees to sync,
5252 5281 * and then wait for the deferral of those frees to finish.
5253 5282 */
5254 5283 spa_vdev_config_exit(spa, NULL,
5255 5284 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5256 5285
5257 5286 /*
5258 5287 * Attempt to evacuate the vdev.
5259 5288 */
5260 5289 error = spa_vdev_remove_evacuate(spa, vd);
5261 5290
5262 5291 txg = spa_vdev_config_enter(spa);
5263 5292
5264 5293 /*
5265 5294 * If we couldn't evacuate the vdev, unwind.
5266 5295 */
5267 5296 if (error) {
5268 5297 metaslab_group_activate(mg);
5269 5298 return (spa_vdev_exit(spa, NULL, txg, error));
5270 5299 }
5271 5300
5272 5301 /*
5273 5302 * Clean up the vdev namespace.
5274 5303 */
5275 5304 spa_vdev_remove_from_namespace(spa, vd);
5276 5305
5277 5306 } else if (vd != NULL) {
5278 5307 /*
5279 5308 * Normal vdevs cannot be removed (yet).
5280 5309 */
5281 5310 error = SET_ERROR(ENOTSUP);
5282 5311 } else {
5283 5312 /*
5284 5313 * There is no vdev of any kind with the specified guid.
5285 5314 */
5286 5315 error = SET_ERROR(ENOENT);
5287 5316 }
5288 5317
5289 5318 if (!locked)
5290 5319 return (spa_vdev_exit(spa, NULL, txg, error));
5291 5320
5292 5321 return (error);
5293 5322 }
5294 5323
5295 5324 /*
5296 5325 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5297 5326 * currently spared, so we can detach it.
5298 5327 */
5299 5328 static vdev_t *
5300 5329 spa_vdev_resilver_done_hunt(vdev_t *vd)
5301 5330 {
5302 5331 vdev_t *newvd, *oldvd;
5303 5332
5304 5333 for (int c = 0; c < vd->vdev_children; c++) {
5305 5334 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5306 5335 if (oldvd != NULL)
5307 5336 return (oldvd);
5308 5337 }
5309 5338
5310 5339 /*
5311 5340 * Check for a completed replacement. We always consider the first
5312 5341 * vdev in the list to be the oldest vdev, and the last one to be
5313 5342 * the newest (see spa_vdev_attach() for how that works). In
5314 5343 * the case where the newest vdev is faulted, we will not automatically
5315 5344 * remove it after a resilver completes. This is OK as it will require
5316 5345 * user intervention to determine which disk the admin wishes to keep.
5317 5346 */
5318 5347 if (vd->vdev_ops == &vdev_replacing_ops) {
5319 5348 ASSERT(vd->vdev_children > 1);
5320 5349
5321 5350 newvd = vd->vdev_child[vd->vdev_children - 1];
5322 5351 oldvd = vd->vdev_child[0];
5323 5352
5324 5353 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5325 5354 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5326 5355 !vdev_dtl_required(oldvd))
5327 5356 return (oldvd);
5328 5357 }
5329 5358
5330 5359 /*
5331 5360 * Check for a completed resilver with the 'unspare' flag set.
5332 5361 */
5333 5362 if (vd->vdev_ops == &vdev_spare_ops) {
5334 5363 vdev_t *first = vd->vdev_child[0];
5335 5364 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5336 5365
5337 5366 if (last->vdev_unspare) {
5338 5367 oldvd = first;
5339 5368 newvd = last;
5340 5369 } else if (first->vdev_unspare) {
5341 5370 oldvd = last;
5342 5371 newvd = first;
5343 5372 } else {
5344 5373 oldvd = NULL;
5345 5374 }
5346 5375
5347 5376 if (oldvd != NULL &&
5348 5377 vdev_dtl_empty(newvd, DTL_MISSING) &&
5349 5378 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5350 5379 !vdev_dtl_required(oldvd))
5351 5380 return (oldvd);
5352 5381
5353 5382 /*
5354 5383 * If there are more than two spares attached to a disk,
5355 5384 * and those spares are not required, then we want to
5356 5385 * attempt to free them up now so that they can be used
5357 5386 * by other pools. Once we're back down to a single
5358 5387 * disk+spare, we stop removing them.
5359 5388 */
5360 5389 if (vd->vdev_children > 2) {
5361 5390 newvd = vd->vdev_child[1];
5362 5391
5363 5392 if (newvd->vdev_isspare && last->vdev_isspare &&
5364 5393 vdev_dtl_empty(last, DTL_MISSING) &&
5365 5394 vdev_dtl_empty(last, DTL_OUTAGE) &&
5366 5395 !vdev_dtl_required(newvd))
5367 5396 return (newvd);
5368 5397 }
5369 5398 }
5370 5399
5371 5400 return (NULL);
5372 5401 }
5373 5402
5374 5403 static void
5375 5404 spa_vdev_resilver_done(spa_t *spa)
5376 5405 {
5377 5406 vdev_t *vd, *pvd, *ppvd;
5378 5407 uint64_t guid, sguid, pguid, ppguid;
5379 5408
5380 5409 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5381 5410
5382 5411 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5383 5412 pvd = vd->vdev_parent;
5384 5413 ppvd = pvd->vdev_parent;
5385 5414 guid = vd->vdev_guid;
5386 5415 pguid = pvd->vdev_guid;
5387 5416 ppguid = ppvd->vdev_guid;
5388 5417 sguid = 0;
5389 5418 /*
5390 5419 * If we have just finished replacing a hot spared device, then
5391 5420 * we need to detach the parent's first child (the original hot
5392 5421 * spare) as well.
5393 5422 */
5394 5423 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5395 5424 ppvd->vdev_children == 2) {
5396 5425 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5397 5426 sguid = ppvd->vdev_child[1]->vdev_guid;
5398 5427 }
5399 5428 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd));
5400 5429
5401 5430 spa_config_exit(spa, SCL_ALL, FTAG);
5402 5431 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5403 5432 return;
5404 5433 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5405 5434 return;
5406 5435 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5407 5436 }
5408 5437
5409 5438 spa_config_exit(spa, SCL_ALL, FTAG);
5410 5439 }
5411 5440
5412 5441 /*
5413 5442 * Update the stored path or FRU for this vdev.
5414 5443 */
5415 5444 int
5416 5445 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5417 5446 boolean_t ispath)
5418 5447 {
5419 5448 vdev_t *vd;
5420 5449 boolean_t sync = B_FALSE;
5421 5450
5422 5451 ASSERT(spa_writeable(spa));
5423 5452
5424 5453 spa_vdev_state_enter(spa, SCL_ALL);
5425 5454
5426 5455 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5427 5456 return (spa_vdev_state_exit(spa, NULL, ENOENT));
5428 5457
5429 5458 if (!vd->vdev_ops->vdev_op_leaf)
5430 5459 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5431 5460
5432 5461 if (ispath) {
5433 5462 if (strcmp(value, vd->vdev_path) != 0) {
5434 5463 spa_strfree(vd->vdev_path);
5435 5464 vd->vdev_path = spa_strdup(value);
5436 5465 sync = B_TRUE;
5437 5466 }
5438 5467 } else {
5439 5468 if (vd->vdev_fru == NULL) {
5440 5469 vd->vdev_fru = spa_strdup(value);
5441 5470 sync = B_TRUE;
5442 5471 } else if (strcmp(value, vd->vdev_fru) != 0) {
5443 5472 spa_strfree(vd->vdev_fru);
5444 5473 vd->vdev_fru = spa_strdup(value);
5445 5474 sync = B_TRUE;
5446 5475 }
5447 5476 }
5448 5477
5449 5478 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5450 5479 }
5451 5480
5452 5481 int
5453 5482 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5454 5483 {
5455 5484 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5456 5485 }
5457 5486
5458 5487 int
5459 5488 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5460 5489 {
5461 5490 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5462 5491 }
5463 5492
5464 5493 /*
5465 5494 * ==========================================================================
5466 5495 * SPA Scanning
5467 5496 * ==========================================================================
5468 5497 */
5469 5498
5470 5499 int
5471 5500 spa_scan_stop(spa_t *spa)
5472 5501 {
5473 5502 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5474 5503 if (dsl_scan_resilvering(spa->spa_dsl_pool))
5475 5504 return (SET_ERROR(EBUSY));
5476 5505 return (dsl_scan_cancel(spa->spa_dsl_pool));
5477 5506 }
5478 5507
5479 5508 int
5480 5509 spa_scan(spa_t *spa, pool_scan_func_t func)
5481 5510 {
5482 5511 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5483 5512
5484 5513 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5485 5514 return (SET_ERROR(ENOTSUP));
5486 5515
5487 5516 /*
5488 5517 * If a resilver was requested, but there is no DTL on a
5489 5518 * writeable leaf device, we have nothing to do.
5490 5519 */
5491 5520 if (func == POOL_SCAN_RESILVER &&
5492 5521 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5493 5522 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5494 5523 return (0);
5495 5524 }
5496 5525
5497 5526 return (dsl_scan(spa->spa_dsl_pool, func));
5498 5527 }
5499 5528
5500 5529 /*
5501 5530 * ==========================================================================
5502 5531 * SPA async task processing
5503 5532 * ==========================================================================
5504 5533 */
5505 5534
5506 5535 static void
5507 5536 spa_async_remove(spa_t *spa, vdev_t *vd)
5508 5537 {
5509 5538 if (vd->vdev_remove_wanted) {
5510 5539 vd->vdev_remove_wanted = B_FALSE;
5511 5540 vd->vdev_delayed_close = B_FALSE;
5512 5541 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5513 5542
5514 5543 /*
5515 5544 * We want to clear the stats, but we don't want to do a full
5516 5545 * vdev_clear() as that will cause us to throw away
5517 5546 * degraded/faulted state as well as attempt to reopen the
5518 5547 * device, all of which is a waste.
5519 5548 */
5520 5549 vd->vdev_stat.vs_read_errors = 0;
5521 5550 vd->vdev_stat.vs_write_errors = 0;
5522 5551 vd->vdev_stat.vs_checksum_errors = 0;
5523 5552
5524 5553 vdev_state_dirty(vd->vdev_top);
5525 5554 }
5526 5555
5527 5556 for (int c = 0; c < vd->vdev_children; c++)
5528 5557 spa_async_remove(spa, vd->vdev_child[c]);
5529 5558 }
5530 5559
5531 5560 static void
5532 5561 spa_async_probe(spa_t *spa, vdev_t *vd)
5533 5562 {
5534 5563 if (vd->vdev_probe_wanted) {
5535 5564 vd->vdev_probe_wanted = B_FALSE;
5536 5565 vdev_reopen(vd); /* vdev_open() does the actual probe */
5537 5566 }
5538 5567
5539 5568 for (int c = 0; c < vd->vdev_children; c++)
5540 5569 spa_async_probe(spa, vd->vdev_child[c]);
5541 5570 }
5542 5571
5543 5572 static void
5544 5573 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5545 5574 {
5546 5575 sysevent_id_t eid;
5547 5576 nvlist_t *attr;
5548 5577 char *physpath;
5549 5578
5550 5579 if (!spa->spa_autoexpand)
5551 5580 return;
5552 5581
5553 5582 for (int c = 0; c < vd->vdev_children; c++) {
5554 5583 vdev_t *cvd = vd->vdev_child[c];
5555 5584 spa_async_autoexpand(spa, cvd);
5556 5585 }
5557 5586
5558 5587 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5559 5588 return;
5560 5589
5561 5590 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5562 5591 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5563 5592
5564 5593 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5565 5594 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5566 5595
5567 5596 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5568 5597 ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
5569 5598
5570 5599 nvlist_free(attr);
5571 5600 kmem_free(physpath, MAXPATHLEN);
5572 5601 }
5573 5602
5574 5603 static void
5575 5604 spa_async_thread(spa_t *spa)
5576 5605 {
5577 5606 int tasks;
5578 5607
5579 5608 ASSERT(spa->spa_sync_on);
5580 5609
5581 5610 mutex_enter(&spa->spa_async_lock);
5582 5611 tasks = spa->spa_async_tasks;
5583 5612 spa->spa_async_tasks = 0;
5584 5613 mutex_exit(&spa->spa_async_lock);
5585 5614
5586 5615 /*
5587 5616 * See if the config needs to be updated.
5588 5617 */
5589 5618 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5590 5619 uint64_t old_space, new_space;
5591 5620
5592 5621 mutex_enter(&spa_namespace_lock);
5593 5622 old_space = metaslab_class_get_space(spa_normal_class(spa));
5594 5623 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5595 5624 new_space = metaslab_class_get_space(spa_normal_class(spa));
5596 5625 mutex_exit(&spa_namespace_lock);
5597 5626
5598 5627 /*
5599 5628 * If the pool grew as a result of the config update,
5600 5629 * then log an internal history event.
5601 5630 */
5602 5631 if (new_space != old_space) {
5603 5632 spa_history_log_internal(spa, "vdev online", NULL,
5604 5633 "pool '%s' size: %llu(+%llu)",
5605 5634 spa_name(spa), new_space, new_space - old_space);
5606 5635 }
5607 5636 }
5608 5637
5609 5638 /*
5610 5639 * See if any devices need to be marked REMOVED.
5611 5640 */
5612 5641 if (tasks & SPA_ASYNC_REMOVE) {
5613 5642 spa_vdev_state_enter(spa, SCL_NONE);
5614 5643 spa_async_remove(spa, spa->spa_root_vdev);
5615 5644 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5616 5645 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5617 5646 for (int i = 0; i < spa->spa_spares.sav_count; i++)
5618 5647 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5619 5648 (void) spa_vdev_state_exit(spa, NULL, 0);
5620 5649 }
5621 5650
5622 5651 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5623 5652 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5624 5653 spa_async_autoexpand(spa, spa->spa_root_vdev);
5625 5654 spa_config_exit(spa, SCL_CONFIG, FTAG);
5626 5655 }
5627 5656
5628 5657 /*
5629 5658 * See if any devices need to be probed.
5630 5659 */
5631 5660 if (tasks & SPA_ASYNC_PROBE) {
5632 5661 spa_vdev_state_enter(spa, SCL_NONE);
5633 5662 spa_async_probe(spa, spa->spa_root_vdev);
5634 5663 (void) spa_vdev_state_exit(spa, NULL, 0);
5635 5664 }
5636 5665
5637 5666 /*
5638 5667 * If any devices are done replacing, detach them.
5639 5668 */
5640 5669 if (tasks & SPA_ASYNC_RESILVER_DONE)
5641 5670 spa_vdev_resilver_done(spa);
5642 5671
5643 5672 /*
5644 5673 * Kick off a resilver.
5645 5674 */
5646 5675 if (tasks & SPA_ASYNC_RESILVER)
5647 5676 dsl_resilver_restart(spa->spa_dsl_pool, 0);
5648 5677
5649 5678 /*
5650 5679 * Let the world know that we're done.
5651 5680 */
5652 5681 mutex_enter(&spa->spa_async_lock);
5653 5682 spa->spa_async_thread = NULL;
5654 5683 cv_broadcast(&spa->spa_async_cv);
5655 5684 mutex_exit(&spa->spa_async_lock);
5656 5685 thread_exit();
5657 5686 }
5658 5687
5659 5688 void
5660 5689 spa_async_suspend(spa_t *spa)
5661 5690 {
5662 5691 mutex_enter(&spa->spa_async_lock);
5663 5692 spa->spa_async_suspended++;
5664 5693 while (spa->spa_async_thread != NULL)
5665 5694 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5666 5695 mutex_exit(&spa->spa_async_lock);
5667 5696 }
5668 5697
5669 5698 void
5670 5699 spa_async_resume(spa_t *spa)
5671 5700 {
5672 5701 mutex_enter(&spa->spa_async_lock);
5673 5702 ASSERT(spa->spa_async_suspended != 0);
5674 5703 spa->spa_async_suspended--;
5675 5704 mutex_exit(&spa->spa_async_lock);
5676 5705 }
5677 5706
5678 5707 static boolean_t
5679 5708 spa_async_tasks_pending(spa_t *spa)
5680 5709 {
5681 5710 uint_t non_config_tasks;
5682 5711 uint_t config_task;
5683 5712 boolean_t config_task_suspended;
5684 5713
5685 5714 non_config_tasks = spa->spa_async_tasks & ~SPA_ASYNC_CONFIG_UPDATE;
5686 5715 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
5687 5716 if (spa->spa_ccw_fail_time == 0) {
5688 5717 config_task_suspended = B_FALSE;
5689 5718 } else {
5690 5719 config_task_suspended =
5691 5720 (gethrtime() - spa->spa_ccw_fail_time) <
5692 5721 (zfs_ccw_retry_interval * NANOSEC);
5693 5722 }
5694 5723
5695 5724 return (non_config_tasks || (config_task && !config_task_suspended));
5696 5725 }
5697 5726
5698 5727 static void
5699 5728 spa_async_dispatch(spa_t *spa)
5700 5729 {
5701 5730 mutex_enter(&spa->spa_async_lock);
5702 5731 if (spa_async_tasks_pending(spa) &&
5703 5732 !spa->spa_async_suspended &&
5704 5733 spa->spa_async_thread == NULL &&
5705 5734 rootdir != NULL)
5706 5735 spa->spa_async_thread = thread_create(NULL, 0,
5707 5736 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5708 5737 mutex_exit(&spa->spa_async_lock);
5709 5738 }
5710 5739
5711 5740 void
5712 5741 spa_async_request(spa_t *spa, int task)
5713 5742 {
5714 5743 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5715 5744 mutex_enter(&spa->spa_async_lock);
5716 5745 spa->spa_async_tasks |= task;
5717 5746 mutex_exit(&spa->spa_async_lock);
5718 5747 }
5719 5748
5720 5749 /*
5721 5750 * ==========================================================================
5722 5751 * SPA syncing routines
5723 5752 * ==========================================================================
5724 5753 */
5725 5754
5726 5755 static int
5727 5756 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5728 5757 {
5729 5758 bpobj_t *bpo = arg;
5730 5759 bpobj_enqueue(bpo, bp, tx);
5731 5760 return (0);
5732 5761 }
5733 5762
5734 5763 static int
5735 5764 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5736 5765 {
5737 5766 zio_t *zio = arg;
5738 5767
5739 5768 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5740 5769 zio->io_flags));
5741 5770 return (0);
5742 5771 }
5743 5772
5744 5773 /*
5745 5774 * Note: this simple function is not inlined to make it easier to dtrace the
5746 5775 * amount of time spent syncing frees.
5747 5776 */
5748 5777 static void
5749 5778 spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx)
5750 5779 {
5751 5780 zio_t *zio = zio_root(spa, NULL, NULL, 0);
5752 5781 bplist_iterate(bpl, spa_free_sync_cb, zio, tx);
5753 5782 VERIFY(zio_wait(zio) == 0);
5754 5783 }
5755 5784
5756 5785 /*
5757 5786 * Note: this simple function is not inlined to make it easier to dtrace the
5758 5787 * amount of time spent syncing deferred frees.
5759 5788 */
5760 5789 static void
5761 5790 spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx)
5762 5791 {
5763 5792 zio_t *zio = zio_root(spa, NULL, NULL, 0);
5764 5793 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj,
5765 5794 spa_free_sync_cb, zio, tx), ==, 0);
5766 5795 VERIFY0(zio_wait(zio));
5767 5796 }
5768 5797
5769 5798
5770 5799 static void
5771 5800 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5772 5801 {
5773 5802 char *packed = NULL;
5774 5803 size_t bufsize;
5775 5804 size_t nvsize = 0;
5776 5805 dmu_buf_t *db;
5777 5806
5778 5807 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5779 5808
5780 5809 /*
5781 5810 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5782 5811 * information. This avoids the dbuf_will_dirty() path and
5783 5812 * saves us a pre-read to get data we don't actually care about.
5784 5813 */
5785 5814 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
5786 5815 packed = kmem_alloc(bufsize, KM_SLEEP);
5787 5816
5788 5817 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5789 5818 KM_SLEEP) == 0);
5790 5819 bzero(packed + nvsize, bufsize - nvsize);
5791 5820
5792 5821 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5793 5822
5794 5823 kmem_free(packed, bufsize);
5795 5824
5796 5825 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5797 5826 dmu_buf_will_dirty(db, tx);
5798 5827 *(uint64_t *)db->db_data = nvsize;
5799 5828 dmu_buf_rele(db, FTAG);
5800 5829 }
5801 5830
5802 5831 static void
5803 5832 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5804 5833 const char *config, const char *entry)
5805 5834 {
5806 5835 nvlist_t *nvroot;
5807 5836 nvlist_t **list;
5808 5837 int i;
5809 5838
5810 5839 if (!sav->sav_sync)
5811 5840 return;
5812 5841
5813 5842 /*
5814 5843 * Update the MOS nvlist describing the list of available devices.
5815 5844 * spa_validate_aux() will have already made sure this nvlist is
5816 5845 * valid and the vdevs are labeled appropriately.
5817 5846 */
5818 5847 if (sav->sav_object == 0) {
5819 5848 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5820 5849 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5821 5850 sizeof (uint64_t), tx);
5822 5851 VERIFY(zap_update(spa->spa_meta_objset,
5823 5852 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5824 5853 &sav->sav_object, tx) == 0);
5825 5854 }
5826 5855
5827 5856 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5828 5857 if (sav->sav_count == 0) {
5829 5858 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5830 5859 } else {
5831 5860 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5832 5861 for (i = 0; i < sav->sav_count; i++)
5833 5862 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5834 5863 B_FALSE, VDEV_CONFIG_L2CACHE);
5835 5864 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5836 5865 sav->sav_count) == 0);
5837 5866 for (i = 0; i < sav->sav_count; i++)
5838 5867 nvlist_free(list[i]);
5839 5868 kmem_free(list, sav->sav_count * sizeof (void *));
5840 5869 }
5841 5870
5842 5871 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5843 5872 nvlist_free(nvroot);
5844 5873
5845 5874 sav->sav_sync = B_FALSE;
5846 5875 }
5847 5876
5848 5877 static void
5849 5878 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5850 5879 {
5851 5880 nvlist_t *config;
5852 5881
5853 5882 if (list_is_empty(&spa->spa_config_dirty_list))
5854 5883 return;
5855 5884
5856 5885 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5857 5886
5858 5887 config = spa_config_generate(spa, spa->spa_root_vdev,
5859 5888 dmu_tx_get_txg(tx), B_FALSE);
5860 5889
5861 5890 /*
5862 5891 * If we're upgrading the spa version then make sure that
5863 5892 * the config object gets updated with the correct version.
5864 5893 */
5865 5894 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
5866 5895 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5867 5896 spa->spa_uberblock.ub_version);
5868 5897
5869 5898 spa_config_exit(spa, SCL_STATE, FTAG);
5870 5899
5871 5900 if (spa->spa_config_syncing)
5872 5901 nvlist_free(spa->spa_config_syncing);
5873 5902 spa->spa_config_syncing = config;
5874 5903
5875 5904 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5876 5905 }
5877 5906
5878 5907 static void
5879 5908 spa_sync_version(void *arg, dmu_tx_t *tx)
5880 5909 {
5881 5910 uint64_t *versionp = arg;
5882 5911 uint64_t version = *versionp;
5883 5912 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
5884 5913
5885 5914 /*
5886 5915 * Setting the version is special cased when first creating the pool.
5887 5916 */
5888 5917 ASSERT(tx->tx_txg != TXG_INITIAL);
5889 5918
5890 5919 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
5891 5920 ASSERT(version >= spa_version(spa));
5892 5921
5893 5922 spa->spa_uberblock.ub_version = version;
5894 5923 vdev_config_dirty(spa->spa_root_vdev);
5895 5924 spa_history_log_internal(spa, "set", tx, "version=%lld", version);
5896 5925 }
5897 5926
5898 5927 /*
5899 5928 * Set zpool properties.
5900 5929 */
5901 5930 static void
5902 5931 spa_sync_props(void *arg, dmu_tx_t *tx)
5903 5932 {
5904 5933 nvlist_t *nvp = arg;
5905 5934 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
5906 5935 objset_t *mos = spa->spa_meta_objset;
5907 5936 nvpair_t *elem = NULL;
5908 5937
5909 5938 mutex_enter(&spa->spa_props_lock);
5910 5939
5911 5940 while ((elem = nvlist_next_nvpair(nvp, elem))) {
5912 5941 uint64_t intval;
5913 5942 char *strval, *fname;
5914 5943 zpool_prop_t prop;
5915 5944 const char *propname;
5916 5945 zprop_type_t proptype;
5917 5946 zfeature_info_t *feature;
5918 5947
5919 5948 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
5920 5949 case ZPROP_INVAL:
5921 5950 /*
5922 5951 * We checked this earlier in spa_prop_validate().
5923 5952 */
5924 5953 ASSERT(zpool_prop_feature(nvpair_name(elem)));
5925 5954
5926 5955 fname = strchr(nvpair_name(elem), '@') + 1;
5927 5956 VERIFY0(zfeature_lookup_name(fname, &feature));
5928 5957
5929 5958 spa_feature_enable(spa, feature, tx);
5930 5959 spa_history_log_internal(spa, "set", tx,
5931 5960 "%s=enabled", nvpair_name(elem));
5932 5961 break;
5933 5962
5934 5963 case ZPOOL_PROP_VERSION:
5935 5964 intval = fnvpair_value_uint64(elem);
5936 5965 /*
5937 5966 * The version is synced seperatly before other
5938 5967 * properties and should be correct by now.
5939 5968 */
5940 5969 ASSERT3U(spa_version(spa), >=, intval);
5941 5970 break;
5942 5971
5943 5972 case ZPOOL_PROP_ALTROOT:
5944 5973 /*
5945 5974 * 'altroot' is a non-persistent property. It should
5946 5975 * have been set temporarily at creation or import time.
5947 5976 */
5948 5977 ASSERT(spa->spa_root != NULL);
5949 5978 break;
5950 5979
5951 5980 case ZPOOL_PROP_READONLY:
5952 5981 case ZPOOL_PROP_CACHEFILE:
5953 5982 /*
5954 5983 * 'readonly' and 'cachefile' are also non-persisitent
5955 5984 * properties.
5956 5985 */
5957 5986 break;
5958 5987 case ZPOOL_PROP_COMMENT:
5959 5988 strval = fnvpair_value_string(elem);
5960 5989 if (spa->spa_comment != NULL)
5961 5990 spa_strfree(spa->spa_comment);
5962 5991 spa->spa_comment = spa_strdup(strval);
5963 5992 /*
5964 5993 * We need to dirty the configuration on all the vdevs
5965 5994 * so that their labels get updated. It's unnecessary
5966 5995 * to do this for pool creation since the vdev's
5967 5996 * configuratoin has already been dirtied.
5968 5997 */
5969 5998 if (tx->tx_txg != TXG_INITIAL)
5970 5999 vdev_config_dirty(spa->spa_root_vdev);
5971 6000 spa_history_log_internal(spa, "set", tx,
5972 6001 "%s=%s", nvpair_name(elem), strval);
5973 6002 break;
5974 6003 default:
5975 6004 /*
5976 6005 * Set pool property values in the poolprops mos object.
5977 6006 */
5978 6007 if (spa->spa_pool_props_object == 0) {
5979 6008 spa->spa_pool_props_object =
5980 6009 zap_create_link(mos, DMU_OT_POOL_PROPS,
5981 6010 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
5982 6011 tx);
5983 6012 }
5984 6013
5985 6014 /* normalize the property name */
5986 6015 propname = zpool_prop_to_name(prop);
5987 6016 proptype = zpool_prop_get_type(prop);
5988 6017
5989 6018 if (nvpair_type(elem) == DATA_TYPE_STRING) {
5990 6019 ASSERT(proptype == PROP_TYPE_STRING);
5991 6020 strval = fnvpair_value_string(elem);
5992 6021 VERIFY0(zap_update(mos,
5993 6022 spa->spa_pool_props_object, propname,
5994 6023 1, strlen(strval) + 1, strval, tx));
5995 6024 spa_history_log_internal(spa, "set", tx,
5996 6025 "%s=%s", nvpair_name(elem), strval);
5997 6026 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
5998 6027 intval = fnvpair_value_uint64(elem);
5999 6028
6000 6029 if (proptype == PROP_TYPE_INDEX) {
6001 6030 const char *unused;
6002 6031 VERIFY0(zpool_prop_index_to_string(
6003 6032 prop, intval, &unused));
6004 6033 }
6005 6034 VERIFY0(zap_update(mos,
6006 6035 spa->spa_pool_props_object, propname,
6007 6036 8, 1, &intval, tx));
6008 6037 spa_history_log_internal(spa, "set", tx,
6009 6038 "%s=%lld", nvpair_name(elem), intval);
6010 6039 } else {
6011 6040 ASSERT(0); /* not allowed */
6012 6041 }
6013 6042
6014 6043 switch (prop) {
6015 6044 case ZPOOL_PROP_DELEGATION:
6016 6045 spa->spa_delegation = intval;
6017 6046 break;
6018 6047 case ZPOOL_PROP_BOOTFS:
6019 6048 spa->spa_bootfs = intval;
6020 6049 break;
6021 6050 case ZPOOL_PROP_FAILUREMODE:
6022 6051 spa->spa_failmode = intval;
6023 6052 break;
6024 6053 case ZPOOL_PROP_AUTOEXPAND:
6025 6054 spa->spa_autoexpand = intval;
6026 6055 if (tx->tx_txg != TXG_INITIAL)
6027 6056 spa_async_request(spa,
6028 6057 SPA_ASYNC_AUTOEXPAND);
6029 6058 break;
6030 6059 case ZPOOL_PROP_DEDUPDITTO:
6031 6060 spa->spa_dedup_ditto = intval;
6032 6061 break;
6033 6062 default:
6034 6063 break;
6035 6064 }
6036 6065 }
6037 6066
6038 6067 }
6039 6068
6040 6069 mutex_exit(&spa->spa_props_lock);
6041 6070 }
6042 6071
6043 6072 /*
6044 6073 * Perform one-time upgrade on-disk changes. spa_version() does not
6045 6074 * reflect the new version this txg, so there must be no changes this
6046 6075 * txg to anything that the upgrade code depends on after it executes.
6047 6076 * Therefore this must be called after dsl_pool_sync() does the sync
6048 6077 * tasks.
6049 6078 */
6050 6079 static void
6051 6080 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
6052 6081 {
6053 6082 dsl_pool_t *dp = spa->spa_dsl_pool;
6054 6083
6055 6084 ASSERT(spa->spa_sync_pass == 1);
6056 6085
6057 6086 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
6058 6087
6059 6088 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
6060 6089 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
6061 6090 dsl_pool_create_origin(dp, tx);
6062 6091
6063 6092 /* Keeping the origin open increases spa_minref */
6064 6093 spa->spa_minref += 3;
6065 6094 }
6066 6095
6067 6096 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
6068 6097 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
6069 6098 dsl_pool_upgrade_clones(dp, tx);
6070 6099 }
6071 6100
6072 6101 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
6073 6102 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
6074 6103 dsl_pool_upgrade_dir_clones(dp, tx);
6075 6104
6076 6105 /* Keeping the freedir open increases spa_minref */
6077 6106 spa->spa_minref += 3;
6078 6107 }
6079 6108
6080 6109 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
6081 6110 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6082 6111 spa_feature_create_zap_objects(spa, tx);
6083 6112 }
6084 6113 rrw_exit(&dp->dp_config_rwlock, FTAG);
6085 6114 }
6086 6115
6087 6116 /*
6088 6117 * Sync the specified transaction group. New blocks may be dirtied as
6089 6118 * part of the process, so we iterate until it converges.
6090 6119 */
6091 6120 void
6092 6121 spa_sync(spa_t *spa, uint64_t txg)
6093 6122 {
6094 6123 dsl_pool_t *dp = spa->spa_dsl_pool;
6095 6124 objset_t *mos = spa->spa_meta_objset;
6096 6125 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
6097 6126 vdev_t *rvd = spa->spa_root_vdev;
6098 6127 vdev_t *vd;
6099 6128 dmu_tx_t *tx;
6100 6129 int error;
6101 6130
6102 6131 VERIFY(spa_writeable(spa));
6103 6132
6104 6133 /*
6105 6134 * Lock out configuration changes.
6106 6135 */
6107 6136 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6108 6137
6109 6138 spa->spa_syncing_txg = txg;
6110 6139 spa->spa_sync_pass = 0;
6111 6140
6112 6141 /*
6113 6142 * If there are any pending vdev state changes, convert them
6114 6143 * into config changes that go out with this transaction group.
6115 6144 */
6116 6145 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6117 6146 while (list_head(&spa->spa_state_dirty_list) != NULL) {
6118 6147 /*
6119 6148 * We need the write lock here because, for aux vdevs,
6120 6149 * calling vdev_config_dirty() modifies sav_config.
6121 6150 * This is ugly and will become unnecessary when we
6122 6151 * eliminate the aux vdev wart by integrating all vdevs
6123 6152 * into the root vdev tree.
6124 6153 */
6125 6154 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6126 6155 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
6127 6156 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
6128 6157 vdev_state_clean(vd);
6129 6158 vdev_config_dirty(vd);
6130 6159 }
6131 6160 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6132 6161 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
6133 6162 }
6134 6163 spa_config_exit(spa, SCL_STATE, FTAG);
6135 6164
6136 6165 tx = dmu_tx_create_assigned(dp, txg);
6137 6166
6138 6167 spa->spa_sync_starttime = gethrtime();
6139 6168 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid,
6140 6169 spa->spa_sync_starttime + spa->spa_deadman_synctime));
6141 6170
6142 6171 /*
6143 6172 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6144 6173 * set spa_deflate if we have no raid-z vdevs.
6145 6174 */
6146 6175 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
6147 6176 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
6148 6177 int i;
6149 6178
6150 6179 for (i = 0; i < rvd->vdev_children; i++) {
6151 6180 vd = rvd->vdev_child[i];
6152 6181 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
6153 6182 break;
6154 6183 }
6155 6184 if (i == rvd->vdev_children) {
6156 6185 spa->spa_deflate = TRUE;
6157 6186 VERIFY(0 == zap_add(spa->spa_meta_objset,
6158 6187 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
6159 6188 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
6160 6189 }
6161 6190 }
6162 6191
6163 6192 /*
6164 6193 * If anything has changed in this txg, or if someone is waiting
6165 6194 * for this txg to sync (eg, spa_vdev_remove()), push the
6166 6195 * deferred frees from the previous txg. If not, leave them
6167 6196 * alone so that we don't generate work on an otherwise idle
6168 6197 * system.
6169 6198 */
6170 6199 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
6171 6200 !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
6172 6201 !txg_list_empty(&dp->dp_sync_tasks, txg) ||
6173 6202 ((dsl_scan_active(dp->dp_scan) ||
6174 6203 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
6175 6204 spa_sync_deferred_frees(spa, tx);
6176 6205 }
6177 6206
6178 6207 /*
6179 6208 * Iterate to convergence.
6180 6209 */
6181 6210 do {
6182 6211 int pass = ++spa->spa_sync_pass;
6183 6212
6184 6213 spa_sync_config_object(spa, tx);
6185 6214 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
6186 6215 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
6187 6216 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
6188 6217 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
6189 6218 spa_errlog_sync(spa, txg);
6190 6219 dsl_pool_sync(dp, txg);
6191 6220
6192 6221 if (pass < zfs_sync_pass_deferred_free) {
6193 6222 spa_sync_frees(spa, free_bpl, tx);
6194 6223 } else {
6195 6224 bplist_iterate(free_bpl, bpobj_enqueue_cb,
6196 6225 &spa->spa_deferred_bpobj, tx);
6197 6226 }
6198 6227
6199 6228 ddt_sync(spa, txg);
6200 6229 dsl_scan_sync(dp, tx);
6201 6230
6202 6231 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
6203 6232 vdev_sync(vd, txg);
6204 6233
6205 6234 if (pass == 1)
6206 6235 spa_sync_upgrades(spa, tx);
6207 6236
6208 6237 } while (dmu_objset_is_dirty(mos, txg));
6209 6238
6210 6239 /*
6211 6240 * Rewrite the vdev configuration (which includes the uberblock)
6212 6241 * to commit the transaction group.
6213 6242 *
6214 6243 * If there are no dirty vdevs, we sync the uberblock to a few
6215 6244 * random top-level vdevs that are known to be visible in the
6216 6245 * config cache (see spa_vdev_add() for a complete description).
6217 6246 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6218 6247 */
6219 6248 for (;;) {
6220 6249 /*
6221 6250 * We hold SCL_STATE to prevent vdev open/close/etc.
6222 6251 * while we're attempting to write the vdev labels.
6223 6252 */
6224 6253 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6225 6254
6226 6255 if (list_is_empty(&spa->spa_config_dirty_list)) {
6227 6256 vdev_t *svd[SPA_DVAS_PER_BP];
6228 6257 int svdcount = 0;
6229 6258 int children = rvd->vdev_children;
6230 6259 int c0 = spa_get_random(children);
6231 6260
6232 6261 for (int c = 0; c < children; c++) {
6233 6262 vd = rvd->vdev_child[(c0 + c) % children];
6234 6263 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
6235 6264 continue;
6236 6265 svd[svdcount++] = vd;
6237 6266 if (svdcount == SPA_DVAS_PER_BP)
6238 6267 break;
6239 6268 }
6240 6269 error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
6241 6270 if (error != 0)
6242 6271 error = vdev_config_sync(svd, svdcount, txg,
6243 6272 B_TRUE);
6244 6273 } else {
6245 6274 error = vdev_config_sync(rvd->vdev_child,
6246 6275 rvd->vdev_children, txg, B_FALSE);
6247 6276 if (error != 0)
6248 6277 error = vdev_config_sync(rvd->vdev_child,
6249 6278 rvd->vdev_children, txg, B_TRUE);
6250 6279 }
6251 6280
6252 6281 if (error == 0)
6253 6282 spa->spa_last_synced_guid = rvd->vdev_guid;
6254 6283
6255 6284 spa_config_exit(spa, SCL_STATE, FTAG);
6256 6285
6257 6286 if (error == 0)
6258 6287 break;
6259 6288 zio_suspend(spa, NULL);
6260 6289 zio_resume_wait(spa);
6261 6290 }
6262 6291 dmu_tx_commit(tx);
6263 6292
6264 6293 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
6265 6294
6266 6295 /*
6267 6296 * Clear the dirty config list.
6268 6297 */
6269 6298 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
6270 6299 vdev_config_clean(vd);
6271 6300
6272 6301 /*
6273 6302 * Now that the new config has synced transactionally,
6274 6303 * let it become visible to the config cache.
6275 6304 */
6276 6305 if (spa->spa_config_syncing != NULL) {
6277 6306 spa_config_set(spa, spa->spa_config_syncing);
6278 6307 spa->spa_config_txg = txg;
6279 6308 spa->spa_config_syncing = NULL;
6280 6309 }
6281 6310
6282 6311 spa->spa_ubsync = spa->spa_uberblock;
6283 6312
6284 6313 dsl_pool_sync_done(dp, txg);
6285 6314
6286 6315 /*
6287 6316 * Update usable space statistics.
6288 6317 */
6289 6318 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
6290 6319 vdev_sync_done(vd, txg);
6291 6320
6292 6321 spa_update_dspace(spa);
6293 6322
6294 6323 /*
6295 6324 * It had better be the case that we didn't dirty anything
6296 6325 * since vdev_config_sync().
6297 6326 */
6298 6327 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
6299 6328 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
6300 6329 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
6301 6330
6302 6331 spa->spa_sync_pass = 0;
6303 6332
6304 6333 spa_config_exit(spa, SCL_CONFIG, FTAG);
6305 6334
6306 6335 spa_handle_ignored_writes(spa);
6307 6336
6308 6337 /*
6309 6338 * If any async tasks have been requested, kick them off.
6310 6339 */
6311 6340 spa_async_dispatch(spa);
6312 6341 }
6313 6342
6314 6343 /*
6315 6344 * Sync all pools. We don't want to hold the namespace lock across these
6316 6345 * operations, so we take a reference on the spa_t and drop the lock during the
6317 6346 * sync.
6318 6347 */
6319 6348 void
6320 6349 spa_sync_allpools(void)
6321 6350 {
6322 6351 spa_t *spa = NULL;
6323 6352 mutex_enter(&spa_namespace_lock);
6324 6353 while ((spa = spa_next(spa)) != NULL) {
6325 6354 if (spa_state(spa) != POOL_STATE_ACTIVE ||
6326 6355 !spa_writeable(spa) || spa_suspended(spa))
6327 6356 continue;
6328 6357 spa_open_ref(spa, FTAG);
6329 6358 mutex_exit(&spa_namespace_lock);
6330 6359 txg_wait_synced(spa_get_dsl(spa), 0);
6331 6360 mutex_enter(&spa_namespace_lock);
6332 6361 spa_close(spa, FTAG);
6333 6362 }
6334 6363 mutex_exit(&spa_namespace_lock);
6335 6364 }
6336 6365
6337 6366 /*
6338 6367 * ==========================================================================
6339 6368 * Miscellaneous routines
6340 6369 * ==========================================================================
6341 6370 */
6342 6371
6343 6372 /*
6344 6373 * Remove all pools in the system.
6345 6374 */
6346 6375 void
6347 6376 spa_evict_all(void)
6348 6377 {
6349 6378 spa_t *spa;
6350 6379
6351 6380 /*
6352 6381 * Remove all cached state. All pools should be closed now,
6353 6382 * so every spa in the AVL tree should be unreferenced.
6354 6383 */
6355 6384 mutex_enter(&spa_namespace_lock);
6356 6385 while ((spa = spa_next(NULL)) != NULL) {
6357 6386 /*
6358 6387 * Stop async tasks. The async thread may need to detach
6359 6388 * a device that's been replaced, which requires grabbing
6360 6389 * spa_namespace_lock, so we must drop it here.
6361 6390 */
6362 6391 spa_open_ref(spa, FTAG);
6363 6392 mutex_exit(&spa_namespace_lock);
6364 6393 spa_async_suspend(spa);
6365 6394 mutex_enter(&spa_namespace_lock);
6366 6395 spa_close(spa, FTAG);
6367 6396
6368 6397 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
6369 6398 spa_unload(spa);
6370 6399 spa_deactivate(spa);
6371 6400 }
6372 6401 spa_remove(spa);
6373 6402 }
6374 6403 mutex_exit(&spa_namespace_lock);
6375 6404 }
6376 6405
6377 6406 vdev_t *
6378 6407 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
6379 6408 {
6380 6409 vdev_t *vd;
6381 6410 int i;
6382 6411
6383 6412 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
6384 6413 return (vd);
6385 6414
6386 6415 if (aux) {
6387 6416 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
6388 6417 vd = spa->spa_l2cache.sav_vdevs[i];
6389 6418 if (vd->vdev_guid == guid)
6390 6419 return (vd);
6391 6420 }
6392 6421
6393 6422 for (i = 0; i < spa->spa_spares.sav_count; i++) {
6394 6423 vd = spa->spa_spares.sav_vdevs[i];
6395 6424 if (vd->vdev_guid == guid)
6396 6425 return (vd);
6397 6426 }
6398 6427 }
6399 6428
6400 6429 return (NULL);
6401 6430 }
6402 6431
6403 6432 void
6404 6433 spa_upgrade(spa_t *spa, uint64_t version)
6405 6434 {
6406 6435 ASSERT(spa_writeable(spa));
6407 6436
6408 6437 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6409 6438
6410 6439 /*
6411 6440 * This should only be called for a non-faulted pool, and since a
6412 6441 * future version would result in an unopenable pool, this shouldn't be
6413 6442 * possible.
6414 6443 */
6415 6444 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version));
6416 6445 ASSERT(version >= spa->spa_uberblock.ub_version);
6417 6446
6418 6447 spa->spa_uberblock.ub_version = version;
6419 6448 vdev_config_dirty(spa->spa_root_vdev);
6420 6449
6421 6450 spa_config_exit(spa, SCL_ALL, FTAG);
6422 6451
6423 6452 txg_wait_synced(spa_get_dsl(spa), 0);
6424 6453 }
6425 6454
6426 6455 boolean_t
6427 6456 spa_has_spare(spa_t *spa, uint64_t guid)
6428 6457 {
6429 6458 int i;
6430 6459 uint64_t spareguid;
6431 6460 spa_aux_vdev_t *sav = &spa->spa_spares;
6432 6461
6433 6462 for (i = 0; i < sav->sav_count; i++)
6434 6463 if (sav->sav_vdevs[i]->vdev_guid == guid)
6435 6464 return (B_TRUE);
6436 6465
6437 6466 for (i = 0; i < sav->sav_npending; i++) {
6438 6467 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
6439 6468 &spareguid) == 0 && spareguid == guid)
6440 6469 return (B_TRUE);
6441 6470 }
6442 6471
6443 6472 return (B_FALSE);
6444 6473 }
6445 6474
6446 6475 /*
6447 6476 * Check if a pool has an active shared spare device.
6448 6477 * Note: reference count of an active spare is 2, as a spare and as a replace
6449 6478 */
6450 6479 static boolean_t
6451 6480 spa_has_active_shared_spare(spa_t *spa)
6452 6481 {
6453 6482 int i, refcnt;
6454 6483 uint64_t pool;
6455 6484 spa_aux_vdev_t *sav = &spa->spa_spares;
6456 6485
6457 6486 for (i = 0; i < sav->sav_count; i++) {
6458 6487 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
6459 6488 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
6460 6489 refcnt > 2)
6461 6490 return (B_TRUE);
6462 6491 }
6463 6492
6464 6493 return (B_FALSE);
6465 6494 }
6466 6495
6467 6496 /*
6468 6497 * Post a sysevent corresponding to the given event. The 'name' must be one of
6469 6498 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
6470 6499 * filled in from the spa and (optionally) the vdev. This doesn't do anything
6471 6500 * in the userland libzpool, as we don't want consumers to misinterpret ztest
6472 6501 * or zdb as real changes.
6473 6502 */
6474 6503 void
6475 6504 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
6476 6505 {
6477 6506 #ifdef _KERNEL
6478 6507 sysevent_t *ev;
6479 6508 sysevent_attr_list_t *attr = NULL;
6480 6509 sysevent_value_t value;
6481 6510 sysevent_id_t eid;
6482 6511
6483 6512 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
6484 6513 SE_SLEEP);
6485 6514
6486 6515 value.value_type = SE_DATA_TYPE_STRING;
6487 6516 value.value.sv_string = spa_name(spa);
6488 6517 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
6489 6518 goto done;
6490 6519
6491 6520 value.value_type = SE_DATA_TYPE_UINT64;
6492 6521 value.value.sv_uint64 = spa_guid(spa);
6493 6522 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
6494 6523 goto done;
6495 6524
6496 6525 if (vd) {
6497 6526 value.value_type = SE_DATA_TYPE_UINT64;
6498 6527 value.value.sv_uint64 = vd->vdev_guid;
6499 6528 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
6500 6529 SE_SLEEP) != 0)
6501 6530 goto done;
6502 6531
6503 6532 if (vd->vdev_path) {
6504 6533 value.value_type = SE_DATA_TYPE_STRING;
6505 6534 value.value.sv_string = vd->vdev_path;
6506 6535 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
6507 6536 &value, SE_SLEEP) != 0)
6508 6537 goto done;
6509 6538 }
6510 6539 }
6511 6540
6512 6541 if (sysevent_attach_attributes(ev, attr) != 0)
6513 6542 goto done;
6514 6543 attr = NULL;
6515 6544
6516 6545 (void) log_sysevent(ev, SE_SLEEP, &eid);
6517 6546
6518 6547 done:
6519 6548 if (attr)
6520 6549 sysevent_free_attr(attr);
6521 6550 sysevent_free(ev);
6522 6551 #endif
6523 6552 }
|
↓ open down ↓ |
2955 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX