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