Print this page
4171 clean up spa_feature_*() interfaces
4172 implement extensible_dataset feature for use by other zpool features
Reviewed by: Max Grossman <max.grossman@delphix.com>
Reviewed by: Christopher Siden <christopher.siden@delphix.com>
Reviewed by: George Wilson <george.wilson@delphix.com>
| Split |
Close |
| Expand all |
| Collapse all |
--- old/usr/src/uts/common/fs/zfs/spa.c
+++ new/usr/src/uts/common/fs/zfs/spa.c
1 1 /*
2 2 * CDDL HEADER START
3 3 *
4 4 * The contents of this file are subject to the terms of the
5 5 * Common Development and Distribution License (the "License").
6 6 * You may not use this file except in compliance with the License.
7 7 *
8 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 9 * or http://www.opensolaris.org/os/licensing.
10 10 * See the License for the specific language governing permissions
11 11 * and limitations under the License.
12 12 *
13 13 * When distributing Covered Code, include this CDDL HEADER in each
14 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 15 * If applicable, add the following below this CDDL HEADER, with the
16 16 * fields enclosed by brackets "[]" replaced with your own identifying
17 17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 18 *
19 19 * CDDL HEADER END
20 20 */
21 21
22 22 /*
23 23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 24 * Copyright (c) 2013 by Delphix. All rights reserved.
25 25 * Copyright 2013 Nexenta Systems, Inc. All rights reserved.
26 26 */
27 27
28 28 /*
29 29 * SPA: Storage Pool Allocator
30 30 *
31 31 * This file contains all the routines used when modifying on-disk SPA state.
32 32 * This includes opening, importing, destroying, exporting a pool, and syncing a
33 33 * pool.
34 34 */
35 35
36 36 #include <sys/zfs_context.h>
37 37 #include <sys/fm/fs/zfs.h>
38 38 #include <sys/spa_impl.h>
39 39 #include <sys/zio.h>
40 40 #include <sys/zio_checksum.h>
41 41 #include <sys/dmu.h>
42 42 #include <sys/dmu_tx.h>
43 43 #include <sys/zap.h>
44 44 #include <sys/zil.h>
45 45 #include <sys/ddt.h>
46 46 #include <sys/vdev_impl.h>
47 47 #include <sys/metaslab.h>
48 48 #include <sys/metaslab_impl.h>
49 49 #include <sys/uberblock_impl.h>
50 50 #include <sys/txg.h>
51 51 #include <sys/avl.h>
52 52 #include <sys/dmu_traverse.h>
53 53 #include <sys/dmu_objset.h>
54 54 #include <sys/unique.h>
55 55 #include <sys/dsl_pool.h>
56 56 #include <sys/dsl_dataset.h>
57 57 #include <sys/dsl_dir.h>
58 58 #include <sys/dsl_prop.h>
59 59 #include <sys/dsl_synctask.h>
60 60 #include <sys/fs/zfs.h>
61 61 #include <sys/arc.h>
62 62 #include <sys/callb.h>
63 63 #include <sys/systeminfo.h>
64 64 #include <sys/spa_boot.h>
65 65 #include <sys/zfs_ioctl.h>
66 66 #include <sys/dsl_scan.h>
67 67 #include <sys/zfeature.h>
68 68 #include <sys/dsl_destroy.h>
69 69
70 70 #ifdef _KERNEL
71 71 #include <sys/bootprops.h>
72 72 #include <sys/callb.h>
73 73 #include <sys/cpupart.h>
74 74 #include <sys/pool.h>
75 75 #include <sys/sysdc.h>
76 76 #include <sys/zone.h>
77 77 #endif /* _KERNEL */
78 78
79 79 #include "zfs_prop.h"
80 80 #include "zfs_comutil.h"
81 81
82 82 /*
83 83 * The interval, in seconds, at which failed configuration cache file writes
84 84 * should be retried.
85 85 */
86 86 static int zfs_ccw_retry_interval = 300;
87 87
88 88 typedef enum zti_modes {
89 89 ZTI_MODE_FIXED, /* value is # of threads (min 1) */
90 90 ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */
91 91 ZTI_MODE_NULL, /* don't create a taskq */
92 92 ZTI_NMODES
93 93 } zti_modes_t;
94 94
95 95 #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) }
96 96 #define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 }
97 97 #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 }
98 98
99 99 #define ZTI_N(n) ZTI_P(n, 1)
100 100 #define ZTI_ONE ZTI_N(1)
101 101
102 102 typedef struct zio_taskq_info {
103 103 zti_modes_t zti_mode;
104 104 uint_t zti_value;
105 105 uint_t zti_count;
106 106 } zio_taskq_info_t;
107 107
108 108 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
109 109 "issue", "issue_high", "intr", "intr_high"
110 110 };
111 111
112 112 /*
113 113 * This table defines the taskq settings for each ZFS I/O type. When
114 114 * initializing a pool, we use this table to create an appropriately sized
115 115 * taskq. Some operations are low volume and therefore have a small, static
116 116 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
117 117 * macros. Other operations process a large amount of data; the ZTI_BATCH
118 118 * macro causes us to create a taskq oriented for throughput. Some operations
119 119 * are so high frequency and short-lived that the taskq itself can become a a
120 120 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
121 121 * additional degree of parallelism specified by the number of threads per-
122 122 * taskq and the number of taskqs; when dispatching an event in this case, the
123 123 * particular taskq is chosen at random.
124 124 *
125 125 * The different taskq priorities are to handle the different contexts (issue
126 126 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
127 127 * need to be handled with minimum delay.
128 128 */
129 129 const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
130 130 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
131 131 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */
132 132 { ZTI_N(8), ZTI_NULL, ZTI_BATCH, ZTI_NULL }, /* READ */
133 133 { ZTI_BATCH, ZTI_N(5), ZTI_N(8), ZTI_N(5) }, /* WRITE */
134 134 { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */
135 135 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */
136 136 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */
137 137 };
138 138
139 139 static void spa_sync_version(void *arg, dmu_tx_t *tx);
140 140 static void spa_sync_props(void *arg, dmu_tx_t *tx);
141 141 static boolean_t spa_has_active_shared_spare(spa_t *spa);
142 142 static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
143 143 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
144 144 char **ereport);
145 145 static void spa_vdev_resilver_done(spa_t *spa);
146 146
147 147 uint_t zio_taskq_batch_pct = 75; /* 1 thread per cpu in pset */
148 148 id_t zio_taskq_psrset_bind = PS_NONE;
149 149 boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
150 150 uint_t zio_taskq_basedc = 80; /* base duty cycle */
151 151
152 152 boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
153 153 extern int zfs_sync_pass_deferred_free;
154 154
155 155 /*
156 156 * This (illegal) pool name is used when temporarily importing a spa_t in order
157 157 * to get the vdev stats associated with the imported devices.
158 158 */
159 159 #define TRYIMPORT_NAME "$import"
160 160
161 161 /*
162 162 * ==========================================================================
163 163 * SPA properties routines
164 164 * ==========================================================================
165 165 */
166 166
167 167 /*
168 168 * Add a (source=src, propname=propval) list to an nvlist.
169 169 */
170 170 static void
171 171 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
172 172 uint64_t intval, zprop_source_t src)
173 173 {
174 174 const char *propname = zpool_prop_to_name(prop);
175 175 nvlist_t *propval;
176 176
177 177 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
178 178 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
179 179
180 180 if (strval != NULL)
181 181 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
182 182 else
183 183 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
184 184
185 185 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
186 186 nvlist_free(propval);
187 187 }
188 188
189 189 /*
190 190 * Get property values from the spa configuration.
191 191 */
192 192 static void
193 193 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
194 194 {
195 195 vdev_t *rvd = spa->spa_root_vdev;
196 196 dsl_pool_t *pool = spa->spa_dsl_pool;
197 197 uint64_t size;
198 198 uint64_t alloc;
199 199 uint64_t space;
200 200 uint64_t cap, version;
201 201 zprop_source_t src = ZPROP_SRC_NONE;
202 202 spa_config_dirent_t *dp;
203 203
204 204 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
205 205
206 206 if (rvd != NULL) {
207 207 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
208 208 size = metaslab_class_get_space(spa_normal_class(spa));
209 209 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
210 210 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
211 211 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
212 212 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
213 213 size - alloc, src);
214 214
215 215 space = 0;
216 216 for (int c = 0; c < rvd->vdev_children; c++) {
217 217 vdev_t *tvd = rvd->vdev_child[c];
218 218 space += tvd->vdev_max_asize - tvd->vdev_asize;
219 219 }
220 220 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, space,
221 221 src);
222 222
223 223 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
224 224 (spa_mode(spa) == FREAD), src);
225 225
226 226 cap = (size == 0) ? 0 : (alloc * 100 / size);
227 227 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
228 228
229 229 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
230 230 ddt_get_pool_dedup_ratio(spa), src);
231 231
232 232 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
233 233 rvd->vdev_state, src);
234 234
235 235 version = spa_version(spa);
236 236 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
237 237 src = ZPROP_SRC_DEFAULT;
238 238 else
239 239 src = ZPROP_SRC_LOCAL;
240 240 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
241 241 }
242 242
243 243 if (pool != NULL) {
244 244 dsl_dir_t *freedir = pool->dp_free_dir;
245 245
246 246 /*
247 247 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
248 248 * when opening pools before this version freedir will be NULL.
249 249 */
250 250 if (freedir != NULL) {
251 251 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
252 252 freedir->dd_phys->dd_used_bytes, src);
253 253 } else {
254 254 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
255 255 NULL, 0, src);
256 256 }
257 257 }
258 258
259 259 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
260 260
261 261 if (spa->spa_comment != NULL) {
262 262 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
263 263 0, ZPROP_SRC_LOCAL);
264 264 }
265 265
266 266 if (spa->spa_root != NULL)
267 267 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
268 268 0, ZPROP_SRC_LOCAL);
269 269
270 270 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
271 271 if (dp->scd_path == NULL) {
272 272 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
273 273 "none", 0, ZPROP_SRC_LOCAL);
274 274 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
275 275 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
276 276 dp->scd_path, 0, ZPROP_SRC_LOCAL);
277 277 }
278 278 }
279 279 }
280 280
281 281 /*
282 282 * Get zpool property values.
283 283 */
284 284 int
285 285 spa_prop_get(spa_t *spa, nvlist_t **nvp)
286 286 {
287 287 objset_t *mos = spa->spa_meta_objset;
288 288 zap_cursor_t zc;
289 289 zap_attribute_t za;
290 290 int err;
291 291
292 292 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
293 293
294 294 mutex_enter(&spa->spa_props_lock);
295 295
296 296 /*
297 297 * Get properties from the spa config.
298 298 */
299 299 spa_prop_get_config(spa, nvp);
300 300
301 301 /* If no pool property object, no more prop to get. */
302 302 if (mos == NULL || spa->spa_pool_props_object == 0) {
303 303 mutex_exit(&spa->spa_props_lock);
304 304 return (0);
305 305 }
306 306
307 307 /*
308 308 * Get properties from the MOS pool property object.
309 309 */
310 310 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
311 311 (err = zap_cursor_retrieve(&zc, &za)) == 0;
312 312 zap_cursor_advance(&zc)) {
313 313 uint64_t intval = 0;
314 314 char *strval = NULL;
315 315 zprop_source_t src = ZPROP_SRC_DEFAULT;
316 316 zpool_prop_t prop;
317 317
318 318 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
319 319 continue;
320 320
321 321 switch (za.za_integer_length) {
322 322 case 8:
323 323 /* integer property */
324 324 if (za.za_first_integer !=
325 325 zpool_prop_default_numeric(prop))
326 326 src = ZPROP_SRC_LOCAL;
327 327
328 328 if (prop == ZPOOL_PROP_BOOTFS) {
329 329 dsl_pool_t *dp;
330 330 dsl_dataset_t *ds = NULL;
331 331
332 332 dp = spa_get_dsl(spa);
333 333 dsl_pool_config_enter(dp, FTAG);
334 334 if (err = dsl_dataset_hold_obj(dp,
335 335 za.za_first_integer, FTAG, &ds)) {
336 336 dsl_pool_config_exit(dp, FTAG);
337 337 break;
338 338 }
339 339
340 340 strval = kmem_alloc(
341 341 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
342 342 KM_SLEEP);
343 343 dsl_dataset_name(ds, strval);
344 344 dsl_dataset_rele(ds, FTAG);
345 345 dsl_pool_config_exit(dp, FTAG);
346 346 } else {
347 347 strval = NULL;
348 348 intval = za.za_first_integer;
349 349 }
350 350
351 351 spa_prop_add_list(*nvp, prop, strval, intval, src);
352 352
353 353 if (strval != NULL)
354 354 kmem_free(strval,
355 355 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
356 356
357 357 break;
358 358
359 359 case 1:
360 360 /* string property */
361 361 strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
362 362 err = zap_lookup(mos, spa->spa_pool_props_object,
363 363 za.za_name, 1, za.za_num_integers, strval);
364 364 if (err) {
365 365 kmem_free(strval, za.za_num_integers);
366 366 break;
367 367 }
368 368 spa_prop_add_list(*nvp, prop, strval, 0, src);
369 369 kmem_free(strval, za.za_num_integers);
370 370 break;
371 371
372 372 default:
373 373 break;
374 374 }
375 375 }
376 376 zap_cursor_fini(&zc);
377 377 mutex_exit(&spa->spa_props_lock);
378 378 out:
379 379 if (err && err != ENOENT) {
380 380 nvlist_free(*nvp);
381 381 *nvp = NULL;
382 382 return (err);
383 383 }
384 384
385 385 return (0);
386 386 }
387 387
388 388 /*
389 389 * Validate the given pool properties nvlist and modify the list
390 390 * for the property values to be set.
391 391 */
392 392 static int
393 393 spa_prop_validate(spa_t *spa, nvlist_t *props)
394 394 {
395 395 nvpair_t *elem;
396 396 int error = 0, reset_bootfs = 0;
397 397 uint64_t objnum = 0;
398 398 boolean_t has_feature = B_FALSE;
399 399
400 400 elem = NULL;
401 401 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
402 402 uint64_t intval;
403 403 char *strval, *slash, *check, *fname;
404 404 const char *propname = nvpair_name(elem);
405 405 zpool_prop_t prop = zpool_name_to_prop(propname);
406 406
407 407 switch (prop) {
408 408 case ZPROP_INVAL:
409 409 if (!zpool_prop_feature(propname)) {
410 410 error = SET_ERROR(EINVAL);
411 411 break;
412 412 }
413 413
414 414 /*
415 415 * Sanitize the input.
416 416 */
417 417 if (nvpair_type(elem) != DATA_TYPE_UINT64) {
418 418 error = SET_ERROR(EINVAL);
419 419 break;
420 420 }
421 421
422 422 if (nvpair_value_uint64(elem, &intval) != 0) {
423 423 error = SET_ERROR(EINVAL);
424 424 break;
425 425 }
426 426
427 427 if (intval != 0) {
428 428 error = SET_ERROR(EINVAL);
429 429 break;
430 430 }
431 431
432 432 fname = strchr(propname, '@') + 1;
433 433 if (zfeature_lookup_name(fname, NULL) != 0) {
434 434 error = SET_ERROR(EINVAL);
435 435 break;
436 436 }
437 437
438 438 has_feature = B_TRUE;
439 439 break;
440 440
441 441 case ZPOOL_PROP_VERSION:
442 442 error = nvpair_value_uint64(elem, &intval);
443 443 if (!error &&
444 444 (intval < spa_version(spa) ||
445 445 intval > SPA_VERSION_BEFORE_FEATURES ||
446 446 has_feature))
447 447 error = SET_ERROR(EINVAL);
448 448 break;
449 449
450 450 case ZPOOL_PROP_DELEGATION:
451 451 case ZPOOL_PROP_AUTOREPLACE:
452 452 case ZPOOL_PROP_LISTSNAPS:
453 453 case ZPOOL_PROP_AUTOEXPAND:
454 454 error = nvpair_value_uint64(elem, &intval);
455 455 if (!error && intval > 1)
456 456 error = SET_ERROR(EINVAL);
457 457 break;
458 458
459 459 case ZPOOL_PROP_BOOTFS:
460 460 /*
461 461 * If the pool version is less than SPA_VERSION_BOOTFS,
462 462 * or the pool is still being created (version == 0),
463 463 * the bootfs property cannot be set.
464 464 */
465 465 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
466 466 error = SET_ERROR(ENOTSUP);
467 467 break;
468 468 }
469 469
470 470 /*
471 471 * Make sure the vdev config is bootable
472 472 */
473 473 if (!vdev_is_bootable(spa->spa_root_vdev)) {
474 474 error = SET_ERROR(ENOTSUP);
475 475 break;
476 476 }
477 477
478 478 reset_bootfs = 1;
479 479
480 480 error = nvpair_value_string(elem, &strval);
481 481
482 482 if (!error) {
483 483 objset_t *os;
484 484 uint64_t compress;
485 485
486 486 if (strval == NULL || strval[0] == '\0') {
487 487 objnum = zpool_prop_default_numeric(
488 488 ZPOOL_PROP_BOOTFS);
489 489 break;
490 490 }
491 491
492 492 if (error = dmu_objset_hold(strval, FTAG, &os))
493 493 break;
494 494
495 495 /* Must be ZPL and not gzip compressed. */
496 496
497 497 if (dmu_objset_type(os) != DMU_OST_ZFS) {
498 498 error = SET_ERROR(ENOTSUP);
499 499 } else if ((error =
500 500 dsl_prop_get_int_ds(dmu_objset_ds(os),
501 501 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
502 502 &compress)) == 0 &&
503 503 !BOOTFS_COMPRESS_VALID(compress)) {
504 504 error = SET_ERROR(ENOTSUP);
505 505 } else {
506 506 objnum = dmu_objset_id(os);
507 507 }
508 508 dmu_objset_rele(os, FTAG);
509 509 }
510 510 break;
511 511
512 512 case ZPOOL_PROP_FAILUREMODE:
513 513 error = nvpair_value_uint64(elem, &intval);
514 514 if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
515 515 intval > ZIO_FAILURE_MODE_PANIC))
516 516 error = SET_ERROR(EINVAL);
517 517
518 518 /*
519 519 * This is a special case which only occurs when
520 520 * the pool has completely failed. This allows
521 521 * the user to change the in-core failmode property
522 522 * without syncing it out to disk (I/Os might
523 523 * currently be blocked). We do this by returning
524 524 * EIO to the caller (spa_prop_set) to trick it
525 525 * into thinking we encountered a property validation
526 526 * error.
527 527 */
528 528 if (!error && spa_suspended(spa)) {
529 529 spa->spa_failmode = intval;
530 530 error = SET_ERROR(EIO);
531 531 }
532 532 break;
533 533
534 534 case ZPOOL_PROP_CACHEFILE:
535 535 if ((error = nvpair_value_string(elem, &strval)) != 0)
536 536 break;
537 537
538 538 if (strval[0] == '\0')
539 539 break;
540 540
541 541 if (strcmp(strval, "none") == 0)
542 542 break;
543 543
544 544 if (strval[0] != '/') {
545 545 error = SET_ERROR(EINVAL);
546 546 break;
547 547 }
548 548
549 549 slash = strrchr(strval, '/');
550 550 ASSERT(slash != NULL);
551 551
552 552 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
553 553 strcmp(slash, "/..") == 0)
554 554 error = SET_ERROR(EINVAL);
555 555 break;
556 556
557 557 case ZPOOL_PROP_COMMENT:
558 558 if ((error = nvpair_value_string(elem, &strval)) != 0)
559 559 break;
560 560 for (check = strval; *check != '\0'; check++) {
561 561 /*
562 562 * The kernel doesn't have an easy isprint()
563 563 * check. For this kernel check, we merely
564 564 * check ASCII apart from DEL. Fix this if
565 565 * there is an easy-to-use kernel isprint().
566 566 */
567 567 if (*check >= 0x7f) {
568 568 error = SET_ERROR(EINVAL);
569 569 break;
570 570 }
571 571 check++;
572 572 }
573 573 if (strlen(strval) > ZPROP_MAX_COMMENT)
574 574 error = E2BIG;
575 575 break;
576 576
577 577 case ZPOOL_PROP_DEDUPDITTO:
578 578 if (spa_version(spa) < SPA_VERSION_DEDUP)
579 579 error = SET_ERROR(ENOTSUP);
580 580 else
581 581 error = nvpair_value_uint64(elem, &intval);
582 582 if (error == 0 &&
583 583 intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
584 584 error = SET_ERROR(EINVAL);
585 585 break;
586 586 }
587 587
588 588 if (error)
589 589 break;
590 590 }
591 591
592 592 if (!error && reset_bootfs) {
593 593 error = nvlist_remove(props,
594 594 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
595 595
596 596 if (!error) {
597 597 error = nvlist_add_uint64(props,
598 598 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
599 599 }
600 600 }
601 601
602 602 return (error);
603 603 }
604 604
605 605 void
606 606 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
607 607 {
608 608 char *cachefile;
609 609 spa_config_dirent_t *dp;
610 610
611 611 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
612 612 &cachefile) != 0)
613 613 return;
614 614
615 615 dp = kmem_alloc(sizeof (spa_config_dirent_t),
616 616 KM_SLEEP);
617 617
618 618 if (cachefile[0] == '\0')
619 619 dp->scd_path = spa_strdup(spa_config_path);
620 620 else if (strcmp(cachefile, "none") == 0)
621 621 dp->scd_path = NULL;
622 622 else
623 623 dp->scd_path = spa_strdup(cachefile);
624 624
625 625 list_insert_head(&spa->spa_config_list, dp);
626 626 if (need_sync)
627 627 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
628 628 }
629 629
630 630 int
631 631 spa_prop_set(spa_t *spa, nvlist_t *nvp)
632 632 {
633 633 int error;
634 634 nvpair_t *elem = NULL;
635 635 boolean_t need_sync = B_FALSE;
636 636
637 637 if ((error = spa_prop_validate(spa, nvp)) != 0)
638 638 return (error);
639 639
640 640 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
641 641 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
642 642
643 643 if (prop == ZPOOL_PROP_CACHEFILE ||
644 644 prop == ZPOOL_PROP_ALTROOT ||
645 645 prop == ZPOOL_PROP_READONLY)
646 646 continue;
647 647
648 648 if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) {
649 649 uint64_t ver;
650 650
651 651 if (prop == ZPOOL_PROP_VERSION) {
652 652 VERIFY(nvpair_value_uint64(elem, &ver) == 0);
653 653 } else {
654 654 ASSERT(zpool_prop_feature(nvpair_name(elem)));
655 655 ver = SPA_VERSION_FEATURES;
656 656 need_sync = B_TRUE;
657 657 }
658 658
659 659 /* Save time if the version is already set. */
660 660 if (ver == spa_version(spa))
661 661 continue;
662 662
663 663 /*
664 664 * In addition to the pool directory object, we might
665 665 * create the pool properties object, the features for
666 666 * read object, the features for write object, or the
667 667 * feature descriptions object.
668 668 */
669 669 error = dsl_sync_task(spa->spa_name, NULL,
670 670 spa_sync_version, &ver, 6);
671 671 if (error)
672 672 return (error);
673 673 continue;
674 674 }
675 675
676 676 need_sync = B_TRUE;
677 677 break;
678 678 }
679 679
680 680 if (need_sync) {
681 681 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props,
682 682 nvp, 6));
683 683 }
684 684
685 685 return (0);
686 686 }
687 687
688 688 /*
689 689 * If the bootfs property value is dsobj, clear it.
690 690 */
691 691 void
692 692 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
693 693 {
694 694 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
695 695 VERIFY(zap_remove(spa->spa_meta_objset,
696 696 spa->spa_pool_props_object,
697 697 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
698 698 spa->spa_bootfs = 0;
699 699 }
700 700 }
701 701
702 702 /*ARGSUSED*/
703 703 static int
704 704 spa_change_guid_check(void *arg, dmu_tx_t *tx)
705 705 {
706 706 uint64_t *newguid = arg;
707 707 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
708 708 vdev_t *rvd = spa->spa_root_vdev;
709 709 uint64_t vdev_state;
710 710
711 711 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
712 712 vdev_state = rvd->vdev_state;
713 713 spa_config_exit(spa, SCL_STATE, FTAG);
714 714
715 715 if (vdev_state != VDEV_STATE_HEALTHY)
716 716 return (SET_ERROR(ENXIO));
717 717
718 718 ASSERT3U(spa_guid(spa), !=, *newguid);
719 719
720 720 return (0);
721 721 }
722 722
723 723 static void
724 724 spa_change_guid_sync(void *arg, dmu_tx_t *tx)
725 725 {
726 726 uint64_t *newguid = arg;
727 727 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
728 728 uint64_t oldguid;
729 729 vdev_t *rvd = spa->spa_root_vdev;
730 730
731 731 oldguid = spa_guid(spa);
732 732
733 733 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
734 734 rvd->vdev_guid = *newguid;
735 735 rvd->vdev_guid_sum += (*newguid - oldguid);
736 736 vdev_config_dirty(rvd);
737 737 spa_config_exit(spa, SCL_STATE, FTAG);
738 738
739 739 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu",
740 740 oldguid, *newguid);
741 741 }
742 742
743 743 /*
744 744 * Change the GUID for the pool. This is done so that we can later
745 745 * re-import a pool built from a clone of our own vdevs. We will modify
746 746 * the root vdev's guid, our own pool guid, and then mark all of our
747 747 * vdevs dirty. Note that we must make sure that all our vdevs are
748 748 * online when we do this, or else any vdevs that weren't present
749 749 * would be orphaned from our pool. We are also going to issue a
750 750 * sysevent to update any watchers.
751 751 */
752 752 int
753 753 spa_change_guid(spa_t *spa)
754 754 {
755 755 int error;
756 756 uint64_t guid;
757 757
758 758 mutex_enter(&spa->spa_vdev_top_lock);
759 759 mutex_enter(&spa_namespace_lock);
760 760 guid = spa_generate_guid(NULL);
761 761
762 762 error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
763 763 spa_change_guid_sync, &guid, 5);
764 764
765 765 if (error == 0) {
766 766 spa_config_sync(spa, B_FALSE, B_TRUE);
767 767 spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
768 768 }
769 769
770 770 mutex_exit(&spa_namespace_lock);
771 771 mutex_exit(&spa->spa_vdev_top_lock);
772 772
773 773 return (error);
774 774 }
775 775
776 776 /*
777 777 * ==========================================================================
778 778 * SPA state manipulation (open/create/destroy/import/export)
779 779 * ==========================================================================
780 780 */
781 781
782 782 static int
783 783 spa_error_entry_compare(const void *a, const void *b)
784 784 {
785 785 spa_error_entry_t *sa = (spa_error_entry_t *)a;
786 786 spa_error_entry_t *sb = (spa_error_entry_t *)b;
787 787 int ret;
788 788
789 789 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
790 790 sizeof (zbookmark_t));
791 791
792 792 if (ret < 0)
793 793 return (-1);
794 794 else if (ret > 0)
795 795 return (1);
796 796 else
797 797 return (0);
798 798 }
799 799
800 800 /*
801 801 * Utility function which retrieves copies of the current logs and
802 802 * re-initializes them in the process.
803 803 */
804 804 void
805 805 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
806 806 {
807 807 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
808 808
809 809 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
810 810 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
811 811
812 812 avl_create(&spa->spa_errlist_scrub,
813 813 spa_error_entry_compare, sizeof (spa_error_entry_t),
814 814 offsetof(spa_error_entry_t, se_avl));
815 815 avl_create(&spa->spa_errlist_last,
816 816 spa_error_entry_compare, sizeof (spa_error_entry_t),
817 817 offsetof(spa_error_entry_t, se_avl));
818 818 }
819 819
820 820 static void
821 821 spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
822 822 {
823 823 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
824 824 enum zti_modes mode = ztip->zti_mode;
825 825 uint_t value = ztip->zti_value;
826 826 uint_t count = ztip->zti_count;
827 827 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
828 828 char name[32];
829 829 uint_t flags = 0;
830 830 boolean_t batch = B_FALSE;
831 831
832 832 if (mode == ZTI_MODE_NULL) {
833 833 tqs->stqs_count = 0;
834 834 tqs->stqs_taskq = NULL;
835 835 return;
836 836 }
837 837
838 838 ASSERT3U(count, >, 0);
839 839
840 840 tqs->stqs_count = count;
841 841 tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP);
842 842
843 843 switch (mode) {
844 844 case ZTI_MODE_FIXED:
845 845 ASSERT3U(value, >=, 1);
846 846 value = MAX(value, 1);
847 847 break;
848 848
849 849 case ZTI_MODE_BATCH:
850 850 batch = B_TRUE;
851 851 flags |= TASKQ_THREADS_CPU_PCT;
852 852 value = zio_taskq_batch_pct;
853 853 break;
854 854
855 855 default:
856 856 panic("unrecognized mode for %s_%s taskq (%u:%u) in "
857 857 "spa_activate()",
858 858 zio_type_name[t], zio_taskq_types[q], mode, value);
859 859 break;
860 860 }
861 861
862 862 for (uint_t i = 0; i < count; i++) {
863 863 taskq_t *tq;
864 864
865 865 if (count > 1) {
866 866 (void) snprintf(name, sizeof (name), "%s_%s_%u",
867 867 zio_type_name[t], zio_taskq_types[q], i);
868 868 } else {
869 869 (void) snprintf(name, sizeof (name), "%s_%s",
870 870 zio_type_name[t], zio_taskq_types[q]);
871 871 }
872 872
873 873 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
874 874 if (batch)
875 875 flags |= TASKQ_DC_BATCH;
876 876
877 877 tq = taskq_create_sysdc(name, value, 50, INT_MAX,
878 878 spa->spa_proc, zio_taskq_basedc, flags);
879 879 } else {
880 880 pri_t pri = maxclsyspri;
881 881 /*
882 882 * The write issue taskq can be extremely CPU
883 883 * intensive. Run it at slightly lower priority
884 884 * than the other taskqs.
885 885 */
886 886 if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE)
887 887 pri--;
888 888
889 889 tq = taskq_create_proc(name, value, pri, 50,
890 890 INT_MAX, spa->spa_proc, flags);
891 891 }
892 892
893 893 tqs->stqs_taskq[i] = tq;
894 894 }
895 895 }
896 896
897 897 static void
898 898 spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
899 899 {
900 900 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
901 901
902 902 if (tqs->stqs_taskq == NULL) {
903 903 ASSERT0(tqs->stqs_count);
904 904 return;
905 905 }
906 906
907 907 for (uint_t i = 0; i < tqs->stqs_count; i++) {
908 908 ASSERT3P(tqs->stqs_taskq[i], !=, NULL);
909 909 taskq_destroy(tqs->stqs_taskq[i]);
910 910 }
911 911
912 912 kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *));
913 913 tqs->stqs_taskq = NULL;
914 914 }
915 915
916 916 /*
917 917 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
918 918 * Note that a type may have multiple discrete taskqs to avoid lock contention
919 919 * on the taskq itself. In that case we choose which taskq at random by using
920 920 * the low bits of gethrtime().
921 921 */
922 922 void
923 923 spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
924 924 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent)
925 925 {
926 926 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
927 927 taskq_t *tq;
928 928
929 929 ASSERT3P(tqs->stqs_taskq, !=, NULL);
930 930 ASSERT3U(tqs->stqs_count, !=, 0);
931 931
932 932 if (tqs->stqs_count == 1) {
933 933 tq = tqs->stqs_taskq[0];
934 934 } else {
935 935 tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count];
936 936 }
937 937
938 938 taskq_dispatch_ent(tq, func, arg, flags, ent);
939 939 }
940 940
941 941 static void
942 942 spa_create_zio_taskqs(spa_t *spa)
943 943 {
944 944 for (int t = 0; t < ZIO_TYPES; t++) {
945 945 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
946 946 spa_taskqs_init(spa, t, q);
947 947 }
948 948 }
949 949 }
950 950
951 951 #ifdef _KERNEL
952 952 static void
953 953 spa_thread(void *arg)
954 954 {
955 955 callb_cpr_t cprinfo;
956 956
957 957 spa_t *spa = arg;
958 958 user_t *pu = PTOU(curproc);
959 959
960 960 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
961 961 spa->spa_name);
962 962
963 963 ASSERT(curproc != &p0);
964 964 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
965 965 "zpool-%s", spa->spa_name);
966 966 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
967 967
968 968 /* bind this thread to the requested psrset */
969 969 if (zio_taskq_psrset_bind != PS_NONE) {
970 970 pool_lock();
971 971 mutex_enter(&cpu_lock);
972 972 mutex_enter(&pidlock);
973 973 mutex_enter(&curproc->p_lock);
974 974
975 975 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
976 976 0, NULL, NULL) == 0) {
977 977 curthread->t_bind_pset = zio_taskq_psrset_bind;
978 978 } else {
979 979 cmn_err(CE_WARN,
980 980 "Couldn't bind process for zfs pool \"%s\" to "
981 981 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
982 982 }
983 983
984 984 mutex_exit(&curproc->p_lock);
985 985 mutex_exit(&pidlock);
986 986 mutex_exit(&cpu_lock);
987 987 pool_unlock();
988 988 }
989 989
990 990 if (zio_taskq_sysdc) {
991 991 sysdc_thread_enter(curthread, 100, 0);
992 992 }
993 993
994 994 spa->spa_proc = curproc;
995 995 spa->spa_did = curthread->t_did;
996 996
997 997 spa_create_zio_taskqs(spa);
998 998
999 999 mutex_enter(&spa->spa_proc_lock);
1000 1000 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
1001 1001
1002 1002 spa->spa_proc_state = SPA_PROC_ACTIVE;
1003 1003 cv_broadcast(&spa->spa_proc_cv);
1004 1004
1005 1005 CALLB_CPR_SAFE_BEGIN(&cprinfo);
1006 1006 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
1007 1007 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1008 1008 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
1009 1009
1010 1010 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
1011 1011 spa->spa_proc_state = SPA_PROC_GONE;
1012 1012 spa->spa_proc = &p0;
1013 1013 cv_broadcast(&spa->spa_proc_cv);
1014 1014 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
1015 1015
1016 1016 mutex_enter(&curproc->p_lock);
1017 1017 lwp_exit();
1018 1018 }
1019 1019 #endif
1020 1020
1021 1021 /*
1022 1022 * Activate an uninitialized pool.
1023 1023 */
1024 1024 static void
1025 1025 spa_activate(spa_t *spa, int mode)
1026 1026 {
1027 1027 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
1028 1028
1029 1029 spa->spa_state = POOL_STATE_ACTIVE;
1030 1030 spa->spa_mode = mode;
1031 1031
1032 1032 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
1033 1033 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
1034 1034
1035 1035 /* Try to create a covering process */
1036 1036 mutex_enter(&spa->spa_proc_lock);
1037 1037 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
1038 1038 ASSERT(spa->spa_proc == &p0);
1039 1039 spa->spa_did = 0;
1040 1040
1041 1041 /* Only create a process if we're going to be around a while. */
1042 1042 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
1043 1043 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
1044 1044 NULL, 0) == 0) {
1045 1045 spa->spa_proc_state = SPA_PROC_CREATED;
1046 1046 while (spa->spa_proc_state == SPA_PROC_CREATED) {
1047 1047 cv_wait(&spa->spa_proc_cv,
1048 1048 &spa->spa_proc_lock);
1049 1049 }
1050 1050 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1051 1051 ASSERT(spa->spa_proc != &p0);
1052 1052 ASSERT(spa->spa_did != 0);
1053 1053 } else {
1054 1054 #ifdef _KERNEL
1055 1055 cmn_err(CE_WARN,
1056 1056 "Couldn't create process for zfs pool \"%s\"\n",
1057 1057 spa->spa_name);
1058 1058 #endif
1059 1059 }
1060 1060 }
1061 1061 mutex_exit(&spa->spa_proc_lock);
1062 1062
1063 1063 /* If we didn't create a process, we need to create our taskqs. */
1064 1064 if (spa->spa_proc == &p0) {
1065 1065 spa_create_zio_taskqs(spa);
1066 1066 }
1067 1067
1068 1068 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1069 1069 offsetof(vdev_t, vdev_config_dirty_node));
1070 1070 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1071 1071 offsetof(vdev_t, vdev_state_dirty_node));
1072 1072
1073 1073 txg_list_create(&spa->spa_vdev_txg_list,
1074 1074 offsetof(struct vdev, vdev_txg_node));
1075 1075
1076 1076 avl_create(&spa->spa_errlist_scrub,
1077 1077 spa_error_entry_compare, sizeof (spa_error_entry_t),
1078 1078 offsetof(spa_error_entry_t, se_avl));
1079 1079 avl_create(&spa->spa_errlist_last,
1080 1080 spa_error_entry_compare, sizeof (spa_error_entry_t),
1081 1081 offsetof(spa_error_entry_t, se_avl));
1082 1082 }
1083 1083
1084 1084 /*
1085 1085 * Opposite of spa_activate().
1086 1086 */
1087 1087 static void
1088 1088 spa_deactivate(spa_t *spa)
1089 1089 {
1090 1090 ASSERT(spa->spa_sync_on == B_FALSE);
1091 1091 ASSERT(spa->spa_dsl_pool == NULL);
1092 1092 ASSERT(spa->spa_root_vdev == NULL);
1093 1093 ASSERT(spa->spa_async_zio_root == NULL);
1094 1094 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1095 1095
1096 1096 txg_list_destroy(&spa->spa_vdev_txg_list);
1097 1097
1098 1098 list_destroy(&spa->spa_config_dirty_list);
1099 1099 list_destroy(&spa->spa_state_dirty_list);
1100 1100
1101 1101 for (int t = 0; t < ZIO_TYPES; t++) {
1102 1102 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1103 1103 spa_taskqs_fini(spa, t, q);
1104 1104 }
1105 1105 }
1106 1106
1107 1107 metaslab_class_destroy(spa->spa_normal_class);
1108 1108 spa->spa_normal_class = NULL;
1109 1109
1110 1110 metaslab_class_destroy(spa->spa_log_class);
1111 1111 spa->spa_log_class = NULL;
1112 1112
1113 1113 /*
1114 1114 * If this was part of an import or the open otherwise failed, we may
1115 1115 * still have errors left in the queues. Empty them just in case.
1116 1116 */
1117 1117 spa_errlog_drain(spa);
1118 1118
1119 1119 avl_destroy(&spa->spa_errlist_scrub);
1120 1120 avl_destroy(&spa->spa_errlist_last);
1121 1121
1122 1122 spa->spa_state = POOL_STATE_UNINITIALIZED;
1123 1123
1124 1124 mutex_enter(&spa->spa_proc_lock);
1125 1125 if (spa->spa_proc_state != SPA_PROC_NONE) {
1126 1126 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1127 1127 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1128 1128 cv_broadcast(&spa->spa_proc_cv);
1129 1129 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1130 1130 ASSERT(spa->spa_proc != &p0);
1131 1131 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1132 1132 }
1133 1133 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1134 1134 spa->spa_proc_state = SPA_PROC_NONE;
1135 1135 }
1136 1136 ASSERT(spa->spa_proc == &p0);
1137 1137 mutex_exit(&spa->spa_proc_lock);
1138 1138
1139 1139 /*
1140 1140 * We want to make sure spa_thread() has actually exited the ZFS
1141 1141 * module, so that the module can't be unloaded out from underneath
1142 1142 * it.
1143 1143 */
1144 1144 if (spa->spa_did != 0) {
1145 1145 thread_join(spa->spa_did);
1146 1146 spa->spa_did = 0;
1147 1147 }
1148 1148 }
1149 1149
1150 1150 /*
1151 1151 * Verify a pool configuration, and construct the vdev tree appropriately. This
1152 1152 * will create all the necessary vdevs in the appropriate layout, with each vdev
1153 1153 * in the CLOSED state. This will prep the pool before open/creation/import.
1154 1154 * All vdev validation is done by the vdev_alloc() routine.
1155 1155 */
1156 1156 static int
1157 1157 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1158 1158 uint_t id, int atype)
1159 1159 {
1160 1160 nvlist_t **child;
1161 1161 uint_t children;
1162 1162 int error;
1163 1163
1164 1164 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1165 1165 return (error);
1166 1166
1167 1167 if ((*vdp)->vdev_ops->vdev_op_leaf)
1168 1168 return (0);
1169 1169
1170 1170 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1171 1171 &child, &children);
1172 1172
1173 1173 if (error == ENOENT)
1174 1174 return (0);
1175 1175
1176 1176 if (error) {
1177 1177 vdev_free(*vdp);
1178 1178 *vdp = NULL;
1179 1179 return (SET_ERROR(EINVAL));
1180 1180 }
1181 1181
1182 1182 for (int c = 0; c < children; c++) {
1183 1183 vdev_t *vd;
1184 1184 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1185 1185 atype)) != 0) {
1186 1186 vdev_free(*vdp);
1187 1187 *vdp = NULL;
1188 1188 return (error);
1189 1189 }
1190 1190 }
1191 1191
1192 1192 ASSERT(*vdp != NULL);
1193 1193
1194 1194 return (0);
1195 1195 }
1196 1196
1197 1197 /*
1198 1198 * Opposite of spa_load().
1199 1199 */
1200 1200 static void
1201 1201 spa_unload(spa_t *spa)
1202 1202 {
1203 1203 int i;
1204 1204
1205 1205 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1206 1206
1207 1207 /*
1208 1208 * Stop async tasks.
1209 1209 */
1210 1210 spa_async_suspend(spa);
1211 1211
1212 1212 /*
1213 1213 * Stop syncing.
1214 1214 */
1215 1215 if (spa->spa_sync_on) {
1216 1216 txg_sync_stop(spa->spa_dsl_pool);
1217 1217 spa->spa_sync_on = B_FALSE;
1218 1218 }
1219 1219
1220 1220 /*
1221 1221 * Wait for any outstanding async I/O to complete.
1222 1222 */
1223 1223 if (spa->spa_async_zio_root != NULL) {
1224 1224 (void) zio_wait(spa->spa_async_zio_root);
1225 1225 spa->spa_async_zio_root = NULL;
1226 1226 }
1227 1227
1228 1228 bpobj_close(&spa->spa_deferred_bpobj);
1229 1229
1230 1230 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1231 1231
1232 1232 /*
1233 1233 * Close all vdevs.
1234 1234 */
1235 1235 if (spa->spa_root_vdev)
1236 1236 vdev_free(spa->spa_root_vdev);
1237 1237 ASSERT(spa->spa_root_vdev == NULL);
1238 1238
1239 1239 /*
1240 1240 * Close the dsl pool.
1241 1241 */
1242 1242 if (spa->spa_dsl_pool) {
1243 1243 dsl_pool_close(spa->spa_dsl_pool);
1244 1244 spa->spa_dsl_pool = NULL;
1245 1245 spa->spa_meta_objset = NULL;
1246 1246 }
1247 1247
1248 1248 ddt_unload(spa);
1249 1249
1250 1250
1251 1251 /*
1252 1252 * Drop and purge level 2 cache
1253 1253 */
1254 1254 spa_l2cache_drop(spa);
1255 1255
1256 1256 for (i = 0; i < spa->spa_spares.sav_count; i++)
1257 1257 vdev_free(spa->spa_spares.sav_vdevs[i]);
1258 1258 if (spa->spa_spares.sav_vdevs) {
1259 1259 kmem_free(spa->spa_spares.sav_vdevs,
1260 1260 spa->spa_spares.sav_count * sizeof (void *));
1261 1261 spa->spa_spares.sav_vdevs = NULL;
1262 1262 }
1263 1263 if (spa->spa_spares.sav_config) {
1264 1264 nvlist_free(spa->spa_spares.sav_config);
1265 1265 spa->spa_spares.sav_config = NULL;
1266 1266 }
1267 1267 spa->spa_spares.sav_count = 0;
1268 1268
1269 1269 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1270 1270 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1271 1271 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1272 1272 }
1273 1273 if (spa->spa_l2cache.sav_vdevs) {
1274 1274 kmem_free(spa->spa_l2cache.sav_vdevs,
1275 1275 spa->spa_l2cache.sav_count * sizeof (void *));
1276 1276 spa->spa_l2cache.sav_vdevs = NULL;
1277 1277 }
1278 1278 if (spa->spa_l2cache.sav_config) {
1279 1279 nvlist_free(spa->spa_l2cache.sav_config);
1280 1280 spa->spa_l2cache.sav_config = NULL;
1281 1281 }
1282 1282 spa->spa_l2cache.sav_count = 0;
1283 1283
1284 1284 spa->spa_async_suspended = 0;
1285 1285
1286 1286 if (spa->spa_comment != NULL) {
1287 1287 spa_strfree(spa->spa_comment);
1288 1288 spa->spa_comment = NULL;
1289 1289 }
1290 1290
1291 1291 spa_config_exit(spa, SCL_ALL, FTAG);
1292 1292 }
1293 1293
1294 1294 /*
1295 1295 * Load (or re-load) the current list of vdevs describing the active spares for
1296 1296 * this pool. When this is called, we have some form of basic information in
1297 1297 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1298 1298 * then re-generate a more complete list including status information.
1299 1299 */
1300 1300 static void
1301 1301 spa_load_spares(spa_t *spa)
1302 1302 {
1303 1303 nvlist_t **spares;
1304 1304 uint_t nspares;
1305 1305 int i;
1306 1306 vdev_t *vd, *tvd;
1307 1307
1308 1308 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1309 1309
1310 1310 /*
1311 1311 * First, close and free any existing spare vdevs.
1312 1312 */
1313 1313 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1314 1314 vd = spa->spa_spares.sav_vdevs[i];
1315 1315
1316 1316 /* Undo the call to spa_activate() below */
1317 1317 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1318 1318 B_FALSE)) != NULL && tvd->vdev_isspare)
1319 1319 spa_spare_remove(tvd);
1320 1320 vdev_close(vd);
1321 1321 vdev_free(vd);
1322 1322 }
1323 1323
1324 1324 if (spa->spa_spares.sav_vdevs)
1325 1325 kmem_free(spa->spa_spares.sav_vdevs,
1326 1326 spa->spa_spares.sav_count * sizeof (void *));
1327 1327
1328 1328 if (spa->spa_spares.sav_config == NULL)
1329 1329 nspares = 0;
1330 1330 else
1331 1331 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1332 1332 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1333 1333
1334 1334 spa->spa_spares.sav_count = (int)nspares;
1335 1335 spa->spa_spares.sav_vdevs = NULL;
1336 1336
1337 1337 if (nspares == 0)
1338 1338 return;
1339 1339
1340 1340 /*
1341 1341 * Construct the array of vdevs, opening them to get status in the
1342 1342 * process. For each spare, there is potentially two different vdev_t
1343 1343 * structures associated with it: one in the list of spares (used only
1344 1344 * for basic validation purposes) and one in the active vdev
1345 1345 * configuration (if it's spared in). During this phase we open and
1346 1346 * validate each vdev on the spare list. If the vdev also exists in the
1347 1347 * active configuration, then we also mark this vdev as an active spare.
1348 1348 */
1349 1349 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1350 1350 KM_SLEEP);
1351 1351 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1352 1352 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1353 1353 VDEV_ALLOC_SPARE) == 0);
1354 1354 ASSERT(vd != NULL);
1355 1355
1356 1356 spa->spa_spares.sav_vdevs[i] = vd;
1357 1357
1358 1358 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1359 1359 B_FALSE)) != NULL) {
1360 1360 if (!tvd->vdev_isspare)
1361 1361 spa_spare_add(tvd);
1362 1362
1363 1363 /*
1364 1364 * We only mark the spare active if we were successfully
1365 1365 * able to load the vdev. Otherwise, importing a pool
1366 1366 * with a bad active spare would result in strange
1367 1367 * behavior, because multiple pool would think the spare
1368 1368 * is actively in use.
1369 1369 *
1370 1370 * There is a vulnerability here to an equally bizarre
1371 1371 * circumstance, where a dead active spare is later
1372 1372 * brought back to life (onlined or otherwise). Given
1373 1373 * the rarity of this scenario, and the extra complexity
1374 1374 * it adds, we ignore the possibility.
1375 1375 */
1376 1376 if (!vdev_is_dead(tvd))
1377 1377 spa_spare_activate(tvd);
1378 1378 }
1379 1379
1380 1380 vd->vdev_top = vd;
1381 1381 vd->vdev_aux = &spa->spa_spares;
1382 1382
1383 1383 if (vdev_open(vd) != 0)
1384 1384 continue;
1385 1385
1386 1386 if (vdev_validate_aux(vd) == 0)
1387 1387 spa_spare_add(vd);
1388 1388 }
1389 1389
1390 1390 /*
1391 1391 * Recompute the stashed list of spares, with status information
1392 1392 * this time.
1393 1393 */
1394 1394 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1395 1395 DATA_TYPE_NVLIST_ARRAY) == 0);
1396 1396
1397 1397 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1398 1398 KM_SLEEP);
1399 1399 for (i = 0; i < spa->spa_spares.sav_count; i++)
1400 1400 spares[i] = vdev_config_generate(spa,
1401 1401 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1402 1402 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1403 1403 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1404 1404 for (i = 0; i < spa->spa_spares.sav_count; i++)
1405 1405 nvlist_free(spares[i]);
1406 1406 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1407 1407 }
1408 1408
1409 1409 /*
1410 1410 * Load (or re-load) the current list of vdevs describing the active l2cache for
1411 1411 * this pool. When this is called, we have some form of basic information in
1412 1412 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1413 1413 * then re-generate a more complete list including status information.
1414 1414 * Devices which are already active have their details maintained, and are
1415 1415 * not re-opened.
1416 1416 */
1417 1417 static void
1418 1418 spa_load_l2cache(spa_t *spa)
1419 1419 {
1420 1420 nvlist_t **l2cache;
1421 1421 uint_t nl2cache;
1422 1422 int i, j, oldnvdevs;
1423 1423 uint64_t guid;
1424 1424 vdev_t *vd, **oldvdevs, **newvdevs;
1425 1425 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1426 1426
1427 1427 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1428 1428
1429 1429 if (sav->sav_config != NULL) {
1430 1430 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1431 1431 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1432 1432 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1433 1433 } else {
1434 1434 nl2cache = 0;
1435 1435 newvdevs = NULL;
1436 1436 }
1437 1437
1438 1438 oldvdevs = sav->sav_vdevs;
1439 1439 oldnvdevs = sav->sav_count;
1440 1440 sav->sav_vdevs = NULL;
1441 1441 sav->sav_count = 0;
1442 1442
1443 1443 /*
1444 1444 * Process new nvlist of vdevs.
1445 1445 */
1446 1446 for (i = 0; i < nl2cache; i++) {
1447 1447 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1448 1448 &guid) == 0);
1449 1449
1450 1450 newvdevs[i] = NULL;
1451 1451 for (j = 0; j < oldnvdevs; j++) {
1452 1452 vd = oldvdevs[j];
1453 1453 if (vd != NULL && guid == vd->vdev_guid) {
1454 1454 /*
1455 1455 * Retain previous vdev for add/remove ops.
1456 1456 */
1457 1457 newvdevs[i] = vd;
1458 1458 oldvdevs[j] = NULL;
1459 1459 break;
1460 1460 }
1461 1461 }
1462 1462
1463 1463 if (newvdevs[i] == NULL) {
1464 1464 /*
1465 1465 * Create new vdev
1466 1466 */
1467 1467 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1468 1468 VDEV_ALLOC_L2CACHE) == 0);
1469 1469 ASSERT(vd != NULL);
1470 1470 newvdevs[i] = vd;
1471 1471
1472 1472 /*
1473 1473 * Commit this vdev as an l2cache device,
1474 1474 * even if it fails to open.
1475 1475 */
1476 1476 spa_l2cache_add(vd);
1477 1477
1478 1478 vd->vdev_top = vd;
1479 1479 vd->vdev_aux = sav;
1480 1480
1481 1481 spa_l2cache_activate(vd);
1482 1482
1483 1483 if (vdev_open(vd) != 0)
1484 1484 continue;
1485 1485
1486 1486 (void) vdev_validate_aux(vd);
1487 1487
1488 1488 if (!vdev_is_dead(vd))
1489 1489 l2arc_add_vdev(spa, vd);
1490 1490 }
1491 1491 }
1492 1492
1493 1493 /*
1494 1494 * Purge vdevs that were dropped
1495 1495 */
1496 1496 for (i = 0; i < oldnvdevs; i++) {
1497 1497 uint64_t pool;
1498 1498
1499 1499 vd = oldvdevs[i];
1500 1500 if (vd != NULL) {
1501 1501 ASSERT(vd->vdev_isl2cache);
1502 1502
1503 1503 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1504 1504 pool != 0ULL && l2arc_vdev_present(vd))
1505 1505 l2arc_remove_vdev(vd);
1506 1506 vdev_clear_stats(vd);
1507 1507 vdev_free(vd);
1508 1508 }
1509 1509 }
1510 1510
1511 1511 if (oldvdevs)
1512 1512 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1513 1513
1514 1514 if (sav->sav_config == NULL)
1515 1515 goto out;
1516 1516
1517 1517 sav->sav_vdevs = newvdevs;
1518 1518 sav->sav_count = (int)nl2cache;
1519 1519
1520 1520 /*
1521 1521 * Recompute the stashed list of l2cache devices, with status
1522 1522 * information this time.
1523 1523 */
1524 1524 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1525 1525 DATA_TYPE_NVLIST_ARRAY) == 0);
1526 1526
1527 1527 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1528 1528 for (i = 0; i < sav->sav_count; i++)
1529 1529 l2cache[i] = vdev_config_generate(spa,
1530 1530 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1531 1531 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1532 1532 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1533 1533 out:
1534 1534 for (i = 0; i < sav->sav_count; i++)
1535 1535 nvlist_free(l2cache[i]);
1536 1536 if (sav->sav_count)
1537 1537 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1538 1538 }
1539 1539
1540 1540 static int
1541 1541 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1542 1542 {
1543 1543 dmu_buf_t *db;
1544 1544 char *packed = NULL;
1545 1545 size_t nvsize = 0;
1546 1546 int error;
1547 1547 *value = NULL;
1548 1548
1549 1549 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1550 1550 nvsize = *(uint64_t *)db->db_data;
1551 1551 dmu_buf_rele(db, FTAG);
1552 1552
1553 1553 packed = kmem_alloc(nvsize, KM_SLEEP);
1554 1554 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1555 1555 DMU_READ_PREFETCH);
1556 1556 if (error == 0)
1557 1557 error = nvlist_unpack(packed, nvsize, value, 0);
1558 1558 kmem_free(packed, nvsize);
1559 1559
1560 1560 return (error);
1561 1561 }
1562 1562
1563 1563 /*
1564 1564 * Checks to see if the given vdev could not be opened, in which case we post a
1565 1565 * sysevent to notify the autoreplace code that the device has been removed.
1566 1566 */
1567 1567 static void
1568 1568 spa_check_removed(vdev_t *vd)
1569 1569 {
1570 1570 for (int c = 0; c < vd->vdev_children; c++)
1571 1571 spa_check_removed(vd->vdev_child[c]);
1572 1572
1573 1573 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1574 1574 !vd->vdev_ishole) {
1575 1575 zfs_post_autoreplace(vd->vdev_spa, vd);
1576 1576 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1577 1577 }
1578 1578 }
1579 1579
1580 1580 /*
1581 1581 * Validate the current config against the MOS config
1582 1582 */
1583 1583 static boolean_t
1584 1584 spa_config_valid(spa_t *spa, nvlist_t *config)
1585 1585 {
1586 1586 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1587 1587 nvlist_t *nv;
1588 1588
1589 1589 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1590 1590
1591 1591 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1592 1592 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1593 1593
1594 1594 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1595 1595
1596 1596 /*
1597 1597 * If we're doing a normal import, then build up any additional
1598 1598 * diagnostic information about missing devices in this config.
1599 1599 * We'll pass this up to the user for further processing.
1600 1600 */
1601 1601 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1602 1602 nvlist_t **child, *nv;
1603 1603 uint64_t idx = 0;
1604 1604
1605 1605 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1606 1606 KM_SLEEP);
1607 1607 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1608 1608
1609 1609 for (int c = 0; c < rvd->vdev_children; c++) {
1610 1610 vdev_t *tvd = rvd->vdev_child[c];
1611 1611 vdev_t *mtvd = mrvd->vdev_child[c];
1612 1612
1613 1613 if (tvd->vdev_ops == &vdev_missing_ops &&
1614 1614 mtvd->vdev_ops != &vdev_missing_ops &&
1615 1615 mtvd->vdev_islog)
1616 1616 child[idx++] = vdev_config_generate(spa, mtvd,
1617 1617 B_FALSE, 0);
1618 1618 }
1619 1619
1620 1620 if (idx) {
1621 1621 VERIFY(nvlist_add_nvlist_array(nv,
1622 1622 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1623 1623 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1624 1624 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1625 1625
1626 1626 for (int i = 0; i < idx; i++)
1627 1627 nvlist_free(child[i]);
1628 1628 }
1629 1629 nvlist_free(nv);
1630 1630 kmem_free(child, rvd->vdev_children * sizeof (char **));
1631 1631 }
1632 1632
1633 1633 /*
1634 1634 * Compare the root vdev tree with the information we have
1635 1635 * from the MOS config (mrvd). Check each top-level vdev
1636 1636 * with the corresponding MOS config top-level (mtvd).
1637 1637 */
1638 1638 for (int c = 0; c < rvd->vdev_children; c++) {
1639 1639 vdev_t *tvd = rvd->vdev_child[c];
1640 1640 vdev_t *mtvd = mrvd->vdev_child[c];
1641 1641
1642 1642 /*
1643 1643 * Resolve any "missing" vdevs in the current configuration.
1644 1644 * If we find that the MOS config has more accurate information
1645 1645 * about the top-level vdev then use that vdev instead.
1646 1646 */
1647 1647 if (tvd->vdev_ops == &vdev_missing_ops &&
1648 1648 mtvd->vdev_ops != &vdev_missing_ops) {
1649 1649
1650 1650 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1651 1651 continue;
1652 1652
1653 1653 /*
1654 1654 * Device specific actions.
1655 1655 */
1656 1656 if (mtvd->vdev_islog) {
1657 1657 spa_set_log_state(spa, SPA_LOG_CLEAR);
1658 1658 } else {
1659 1659 /*
1660 1660 * XXX - once we have 'readonly' pool
1661 1661 * support we should be able to handle
1662 1662 * missing data devices by transitioning
1663 1663 * the pool to readonly.
1664 1664 */
1665 1665 continue;
1666 1666 }
1667 1667
1668 1668 /*
1669 1669 * Swap the missing vdev with the data we were
1670 1670 * able to obtain from the MOS config.
1671 1671 */
1672 1672 vdev_remove_child(rvd, tvd);
1673 1673 vdev_remove_child(mrvd, mtvd);
1674 1674
1675 1675 vdev_add_child(rvd, mtvd);
1676 1676 vdev_add_child(mrvd, tvd);
1677 1677
1678 1678 spa_config_exit(spa, SCL_ALL, FTAG);
1679 1679 vdev_load(mtvd);
1680 1680 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1681 1681
1682 1682 vdev_reopen(rvd);
1683 1683 } else if (mtvd->vdev_islog) {
1684 1684 /*
1685 1685 * Load the slog device's state from the MOS config
1686 1686 * since it's possible that the label does not
1687 1687 * contain the most up-to-date information.
1688 1688 */
1689 1689 vdev_load_log_state(tvd, mtvd);
1690 1690 vdev_reopen(tvd);
1691 1691 }
1692 1692 }
1693 1693 vdev_free(mrvd);
1694 1694 spa_config_exit(spa, SCL_ALL, FTAG);
1695 1695
1696 1696 /*
1697 1697 * Ensure we were able to validate the config.
1698 1698 */
1699 1699 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1700 1700 }
1701 1701
1702 1702 /*
1703 1703 * Check for missing log devices
1704 1704 */
1705 1705 static boolean_t
1706 1706 spa_check_logs(spa_t *spa)
1707 1707 {
1708 1708 boolean_t rv = B_FALSE;
1709 1709
1710 1710 switch (spa->spa_log_state) {
1711 1711 case SPA_LOG_MISSING:
1712 1712 /* need to recheck in case slog has been restored */
1713 1713 case SPA_LOG_UNKNOWN:
1714 1714 rv = (dmu_objset_find(spa->spa_name, zil_check_log_chain,
1715 1715 NULL, DS_FIND_CHILDREN) != 0);
1716 1716 if (rv)
1717 1717 spa_set_log_state(spa, SPA_LOG_MISSING);
1718 1718 break;
1719 1719 }
1720 1720 return (rv);
1721 1721 }
1722 1722
1723 1723 static boolean_t
1724 1724 spa_passivate_log(spa_t *spa)
1725 1725 {
1726 1726 vdev_t *rvd = spa->spa_root_vdev;
1727 1727 boolean_t slog_found = B_FALSE;
1728 1728
1729 1729 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1730 1730
1731 1731 if (!spa_has_slogs(spa))
1732 1732 return (B_FALSE);
1733 1733
1734 1734 for (int c = 0; c < rvd->vdev_children; c++) {
1735 1735 vdev_t *tvd = rvd->vdev_child[c];
1736 1736 metaslab_group_t *mg = tvd->vdev_mg;
1737 1737
1738 1738 if (tvd->vdev_islog) {
1739 1739 metaslab_group_passivate(mg);
1740 1740 slog_found = B_TRUE;
1741 1741 }
1742 1742 }
1743 1743
1744 1744 return (slog_found);
1745 1745 }
1746 1746
1747 1747 static void
1748 1748 spa_activate_log(spa_t *spa)
1749 1749 {
1750 1750 vdev_t *rvd = spa->spa_root_vdev;
1751 1751
1752 1752 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1753 1753
1754 1754 for (int c = 0; c < rvd->vdev_children; c++) {
1755 1755 vdev_t *tvd = rvd->vdev_child[c];
1756 1756 metaslab_group_t *mg = tvd->vdev_mg;
1757 1757
1758 1758 if (tvd->vdev_islog)
1759 1759 metaslab_group_activate(mg);
1760 1760 }
1761 1761 }
1762 1762
1763 1763 int
1764 1764 spa_offline_log(spa_t *spa)
1765 1765 {
1766 1766 int error;
1767 1767
1768 1768 error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1769 1769 NULL, DS_FIND_CHILDREN);
1770 1770 if (error == 0) {
1771 1771 /*
1772 1772 * We successfully offlined the log device, sync out the
1773 1773 * current txg so that the "stubby" block can be removed
1774 1774 * by zil_sync().
1775 1775 */
1776 1776 txg_wait_synced(spa->spa_dsl_pool, 0);
1777 1777 }
1778 1778 return (error);
1779 1779 }
1780 1780
1781 1781 static void
1782 1782 spa_aux_check_removed(spa_aux_vdev_t *sav)
1783 1783 {
1784 1784 for (int i = 0; i < sav->sav_count; i++)
1785 1785 spa_check_removed(sav->sav_vdevs[i]);
1786 1786 }
1787 1787
1788 1788 void
1789 1789 spa_claim_notify(zio_t *zio)
1790 1790 {
1791 1791 spa_t *spa = zio->io_spa;
1792 1792
1793 1793 if (zio->io_error)
1794 1794 return;
1795 1795
1796 1796 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1797 1797 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1798 1798 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1799 1799 mutex_exit(&spa->spa_props_lock);
1800 1800 }
1801 1801
1802 1802 typedef struct spa_load_error {
1803 1803 uint64_t sle_meta_count;
1804 1804 uint64_t sle_data_count;
1805 1805 } spa_load_error_t;
1806 1806
1807 1807 static void
1808 1808 spa_load_verify_done(zio_t *zio)
1809 1809 {
1810 1810 blkptr_t *bp = zio->io_bp;
1811 1811 spa_load_error_t *sle = zio->io_private;
1812 1812 dmu_object_type_t type = BP_GET_TYPE(bp);
1813 1813 int error = zio->io_error;
1814 1814
1815 1815 if (error) {
1816 1816 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1817 1817 type != DMU_OT_INTENT_LOG)
1818 1818 atomic_add_64(&sle->sle_meta_count, 1);
1819 1819 else
1820 1820 atomic_add_64(&sle->sle_data_count, 1);
1821 1821 }
1822 1822 zio_data_buf_free(zio->io_data, zio->io_size);
1823 1823 }
1824 1824
1825 1825 /*ARGSUSED*/
1826 1826 static int
1827 1827 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1828 1828 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1829 1829 {
1830 1830 if (bp != NULL) {
1831 1831 zio_t *rio = arg;
1832 1832 size_t size = BP_GET_PSIZE(bp);
1833 1833 void *data = zio_data_buf_alloc(size);
1834 1834
1835 1835 zio_nowait(zio_read(rio, spa, bp, data, size,
1836 1836 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1837 1837 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1838 1838 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1839 1839 }
1840 1840 return (0);
1841 1841 }
1842 1842
1843 1843 static int
1844 1844 spa_load_verify(spa_t *spa)
1845 1845 {
1846 1846 zio_t *rio;
1847 1847 spa_load_error_t sle = { 0 };
1848 1848 zpool_rewind_policy_t policy;
1849 1849 boolean_t verify_ok = B_FALSE;
1850 1850 int error;
1851 1851
1852 1852 zpool_get_rewind_policy(spa->spa_config, &policy);
1853 1853
1854 1854 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1855 1855 return (0);
1856 1856
1857 1857 rio = zio_root(spa, NULL, &sle,
1858 1858 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1859 1859
1860 1860 error = traverse_pool(spa, spa->spa_verify_min_txg,
1861 1861 TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1862 1862
1863 1863 (void) zio_wait(rio);
1864 1864
1865 1865 spa->spa_load_meta_errors = sle.sle_meta_count;
1866 1866 spa->spa_load_data_errors = sle.sle_data_count;
1867 1867
1868 1868 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1869 1869 sle.sle_data_count <= policy.zrp_maxdata) {
1870 1870 int64_t loss = 0;
1871 1871
1872 1872 verify_ok = B_TRUE;
1873 1873 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1874 1874 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1875 1875
1876 1876 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1877 1877 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1878 1878 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1879 1879 VERIFY(nvlist_add_int64(spa->spa_load_info,
1880 1880 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1881 1881 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1882 1882 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1883 1883 } else {
1884 1884 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1885 1885 }
1886 1886
1887 1887 if (error) {
1888 1888 if (error != ENXIO && error != EIO)
1889 1889 error = SET_ERROR(EIO);
1890 1890 return (error);
1891 1891 }
1892 1892
1893 1893 return (verify_ok ? 0 : EIO);
1894 1894 }
1895 1895
1896 1896 /*
1897 1897 * Find a value in the pool props object.
1898 1898 */
1899 1899 static void
1900 1900 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1901 1901 {
1902 1902 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1903 1903 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1904 1904 }
1905 1905
1906 1906 /*
1907 1907 * Find a value in the pool directory object.
1908 1908 */
1909 1909 static int
1910 1910 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1911 1911 {
1912 1912 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1913 1913 name, sizeof (uint64_t), 1, val));
1914 1914 }
1915 1915
1916 1916 static int
1917 1917 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1918 1918 {
1919 1919 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1920 1920 return (err);
1921 1921 }
1922 1922
1923 1923 /*
1924 1924 * Fix up config after a partly-completed split. This is done with the
1925 1925 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
1926 1926 * pool have that entry in their config, but only the splitting one contains
1927 1927 * a list of all the guids of the vdevs that are being split off.
1928 1928 *
1929 1929 * This function determines what to do with that list: either rejoin
1930 1930 * all the disks to the pool, or complete the splitting process. To attempt
1931 1931 * the rejoin, each disk that is offlined is marked online again, and
1932 1932 * we do a reopen() call. If the vdev label for every disk that was
1933 1933 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1934 1934 * then we call vdev_split() on each disk, and complete the split.
1935 1935 *
1936 1936 * Otherwise we leave the config alone, with all the vdevs in place in
1937 1937 * the original pool.
1938 1938 */
1939 1939 static void
1940 1940 spa_try_repair(spa_t *spa, nvlist_t *config)
1941 1941 {
1942 1942 uint_t extracted;
1943 1943 uint64_t *glist;
1944 1944 uint_t i, gcount;
1945 1945 nvlist_t *nvl;
1946 1946 vdev_t **vd;
1947 1947 boolean_t attempt_reopen;
1948 1948
1949 1949 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1950 1950 return;
1951 1951
1952 1952 /* check that the config is complete */
1953 1953 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1954 1954 &glist, &gcount) != 0)
1955 1955 return;
1956 1956
1957 1957 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1958 1958
1959 1959 /* attempt to online all the vdevs & validate */
1960 1960 attempt_reopen = B_TRUE;
1961 1961 for (i = 0; i < gcount; i++) {
1962 1962 if (glist[i] == 0) /* vdev is hole */
1963 1963 continue;
1964 1964
1965 1965 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1966 1966 if (vd[i] == NULL) {
1967 1967 /*
1968 1968 * Don't bother attempting to reopen the disks;
1969 1969 * just do the split.
1970 1970 */
1971 1971 attempt_reopen = B_FALSE;
1972 1972 } else {
1973 1973 /* attempt to re-online it */
1974 1974 vd[i]->vdev_offline = B_FALSE;
1975 1975 }
1976 1976 }
1977 1977
1978 1978 if (attempt_reopen) {
1979 1979 vdev_reopen(spa->spa_root_vdev);
1980 1980
1981 1981 /* check each device to see what state it's in */
1982 1982 for (extracted = 0, i = 0; i < gcount; i++) {
1983 1983 if (vd[i] != NULL &&
1984 1984 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1985 1985 break;
1986 1986 ++extracted;
1987 1987 }
1988 1988 }
1989 1989
1990 1990 /*
1991 1991 * If every disk has been moved to the new pool, or if we never
1992 1992 * even attempted to look at them, then we split them off for
1993 1993 * good.
1994 1994 */
1995 1995 if (!attempt_reopen || gcount == extracted) {
1996 1996 for (i = 0; i < gcount; i++)
1997 1997 if (vd[i] != NULL)
1998 1998 vdev_split(vd[i]);
1999 1999 vdev_reopen(spa->spa_root_vdev);
2000 2000 }
2001 2001
2002 2002 kmem_free(vd, gcount * sizeof (vdev_t *));
2003 2003 }
2004 2004
2005 2005 static int
2006 2006 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
2007 2007 boolean_t mosconfig)
2008 2008 {
2009 2009 nvlist_t *config = spa->spa_config;
2010 2010 char *ereport = FM_EREPORT_ZFS_POOL;
2011 2011 char *comment;
2012 2012 int error;
2013 2013 uint64_t pool_guid;
2014 2014 nvlist_t *nvl;
2015 2015
2016 2016 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
2017 2017 return (SET_ERROR(EINVAL));
2018 2018
2019 2019 ASSERT(spa->spa_comment == NULL);
2020 2020 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2021 2021 spa->spa_comment = spa_strdup(comment);
2022 2022
2023 2023 /*
2024 2024 * Versioning wasn't explicitly added to the label until later, so if
2025 2025 * it's not present treat it as the initial version.
2026 2026 */
2027 2027 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2028 2028 &spa->spa_ubsync.ub_version) != 0)
2029 2029 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2030 2030
2031 2031 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2032 2032 &spa->spa_config_txg);
2033 2033
2034 2034 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
2035 2035 spa_guid_exists(pool_guid, 0)) {
2036 2036 error = SET_ERROR(EEXIST);
2037 2037 } else {
2038 2038 spa->spa_config_guid = pool_guid;
2039 2039
2040 2040 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
2041 2041 &nvl) == 0) {
2042 2042 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
2043 2043 KM_SLEEP) == 0);
2044 2044 }
2045 2045
2046 2046 nvlist_free(spa->spa_load_info);
2047 2047 spa->spa_load_info = fnvlist_alloc();
2048 2048
2049 2049 gethrestime(&spa->spa_loaded_ts);
2050 2050 error = spa_load_impl(spa, pool_guid, config, state, type,
2051 2051 mosconfig, &ereport);
2052 2052 }
2053 2053
2054 2054 spa->spa_minref = refcount_count(&spa->spa_refcount);
2055 2055 if (error) {
2056 2056 if (error != EEXIST) {
2057 2057 spa->spa_loaded_ts.tv_sec = 0;
2058 2058 spa->spa_loaded_ts.tv_nsec = 0;
2059 2059 }
2060 2060 if (error != EBADF) {
2061 2061 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2062 2062 }
2063 2063 }
2064 2064 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2065 2065 spa->spa_ena = 0;
2066 2066
2067 2067 return (error);
2068 2068 }
2069 2069
2070 2070 /*
2071 2071 * Load an existing storage pool, using the pool's builtin spa_config as a
2072 2072 * source of configuration information.
2073 2073 */
2074 2074 static int
2075 2075 spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2076 2076 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2077 2077 char **ereport)
2078 2078 {
2079 2079 int error = 0;
2080 2080 nvlist_t *nvroot = NULL;
2081 2081 nvlist_t *label;
2082 2082 vdev_t *rvd;
2083 2083 uberblock_t *ub = &spa->spa_uberblock;
2084 2084 uint64_t children, config_cache_txg = spa->spa_config_txg;
2085 2085 int orig_mode = spa->spa_mode;
2086 2086 int parse;
2087 2087 uint64_t obj;
2088 2088 boolean_t missing_feat_write = B_FALSE;
2089 2089
2090 2090 /*
2091 2091 * If this is an untrusted config, access the pool in read-only mode.
2092 2092 * This prevents things like resilvering recently removed devices.
2093 2093 */
2094 2094 if (!mosconfig)
2095 2095 spa->spa_mode = FREAD;
2096 2096
2097 2097 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2098 2098
2099 2099 spa->spa_load_state = state;
2100 2100
2101 2101 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2102 2102 return (SET_ERROR(EINVAL));
2103 2103
2104 2104 parse = (type == SPA_IMPORT_EXISTING ?
2105 2105 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2106 2106
2107 2107 /*
2108 2108 * Create "The Godfather" zio to hold all async IOs
2109 2109 */
2110 2110 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2111 2111 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2112 2112
2113 2113 /*
2114 2114 * Parse the configuration into a vdev tree. We explicitly set the
2115 2115 * value that will be returned by spa_version() since parsing the
2116 2116 * configuration requires knowing the version number.
2117 2117 */
2118 2118 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2119 2119 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2120 2120 spa_config_exit(spa, SCL_ALL, FTAG);
2121 2121
2122 2122 if (error != 0)
2123 2123 return (error);
2124 2124
2125 2125 ASSERT(spa->spa_root_vdev == rvd);
2126 2126
2127 2127 if (type != SPA_IMPORT_ASSEMBLE) {
2128 2128 ASSERT(spa_guid(spa) == pool_guid);
2129 2129 }
2130 2130
2131 2131 /*
2132 2132 * Try to open all vdevs, loading each label in the process.
2133 2133 */
2134 2134 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2135 2135 error = vdev_open(rvd);
2136 2136 spa_config_exit(spa, SCL_ALL, FTAG);
2137 2137 if (error != 0)
2138 2138 return (error);
2139 2139
2140 2140 /*
2141 2141 * We need to validate the vdev labels against the configuration that
2142 2142 * we have in hand, which is dependent on the setting of mosconfig. If
2143 2143 * mosconfig is true then we're validating the vdev labels based on
2144 2144 * that config. Otherwise, we're validating against the cached config
2145 2145 * (zpool.cache) that was read when we loaded the zfs module, and then
2146 2146 * later we will recursively call spa_load() and validate against
2147 2147 * the vdev config.
2148 2148 *
2149 2149 * If we're assembling a new pool that's been split off from an
2150 2150 * existing pool, the labels haven't yet been updated so we skip
2151 2151 * validation for now.
2152 2152 */
2153 2153 if (type != SPA_IMPORT_ASSEMBLE) {
2154 2154 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2155 2155 error = vdev_validate(rvd, mosconfig);
2156 2156 spa_config_exit(spa, SCL_ALL, FTAG);
2157 2157
2158 2158 if (error != 0)
2159 2159 return (error);
2160 2160
2161 2161 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2162 2162 return (SET_ERROR(ENXIO));
2163 2163 }
2164 2164
2165 2165 /*
2166 2166 * Find the best uberblock.
2167 2167 */
2168 2168 vdev_uberblock_load(rvd, ub, &label);
2169 2169
2170 2170 /*
2171 2171 * If we weren't able to find a single valid uberblock, return failure.
2172 2172 */
2173 2173 if (ub->ub_txg == 0) {
2174 2174 nvlist_free(label);
2175 2175 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2176 2176 }
2177 2177
2178 2178 /*
2179 2179 * If the pool has an unsupported version we can't open it.
2180 2180 */
2181 2181 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2182 2182 nvlist_free(label);
2183 2183 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2184 2184 }
2185 2185
2186 2186 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2187 2187 nvlist_t *features;
2188 2188
2189 2189 /*
2190 2190 * If we weren't able to find what's necessary for reading the
2191 2191 * MOS in the label, return failure.
2192 2192 */
2193 2193 if (label == NULL || nvlist_lookup_nvlist(label,
2194 2194 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2195 2195 nvlist_free(label);
2196 2196 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2197 2197 ENXIO));
2198 2198 }
2199 2199
2200 2200 /*
2201 2201 * Update our in-core representation with the definitive values
2202 2202 * from the label.
2203 2203 */
2204 2204 nvlist_free(spa->spa_label_features);
2205 2205 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2206 2206 }
2207 2207
2208 2208 nvlist_free(label);
2209 2209
2210 2210 /*
2211 2211 * Look through entries in the label nvlist's features_for_read. If
2212 2212 * there is a feature listed there which we don't understand then we
2213 2213 * cannot open a pool.
2214 2214 */
2215 2215 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2216 2216 nvlist_t *unsup_feat;
2217 2217
2218 2218 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2219 2219 0);
2220 2220
2221 2221 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2222 2222 NULL); nvp != NULL;
2223 2223 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2224 2224 if (!zfeature_is_supported(nvpair_name(nvp))) {
2225 2225 VERIFY(nvlist_add_string(unsup_feat,
2226 2226 nvpair_name(nvp), "") == 0);
2227 2227 }
2228 2228 }
2229 2229
2230 2230 if (!nvlist_empty(unsup_feat)) {
2231 2231 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2232 2232 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2233 2233 nvlist_free(unsup_feat);
2234 2234 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2235 2235 ENOTSUP));
2236 2236 }
2237 2237
2238 2238 nvlist_free(unsup_feat);
2239 2239 }
2240 2240
2241 2241 /*
2242 2242 * If the vdev guid sum doesn't match the uberblock, we have an
2243 2243 * incomplete configuration. We first check to see if the pool
2244 2244 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2245 2245 * If it is, defer the vdev_guid_sum check till later so we
2246 2246 * can handle missing vdevs.
2247 2247 */
2248 2248 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2249 2249 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2250 2250 rvd->vdev_guid_sum != ub->ub_guid_sum)
2251 2251 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2252 2252
2253 2253 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2254 2254 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2255 2255 spa_try_repair(spa, config);
2256 2256 spa_config_exit(spa, SCL_ALL, FTAG);
2257 2257 nvlist_free(spa->spa_config_splitting);
2258 2258 spa->spa_config_splitting = NULL;
2259 2259 }
2260 2260
2261 2261 /*
2262 2262 * Initialize internal SPA structures.
2263 2263 */
2264 2264 spa->spa_state = POOL_STATE_ACTIVE;
2265 2265 spa->spa_ubsync = spa->spa_uberblock;
2266 2266 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2267 2267 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2268 2268 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2269 2269 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2270 2270 spa->spa_claim_max_txg = spa->spa_first_txg;
2271 2271 spa->spa_prev_software_version = ub->ub_software_version;
2272 2272
2273 2273 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2274 2274 if (error)
2275 2275 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2276 2276 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2277 2277
2278 2278 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2279 2279 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2280 2280
2281 2281 if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2282 2282 boolean_t missing_feat_read = B_FALSE;
2283 2283 nvlist_t *unsup_feat, *enabled_feat;
2284 2284
2285 2285 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2286 2286 &spa->spa_feat_for_read_obj) != 0) {
2287 2287 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2288 2288 }
2289 2289
2290 2290 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2291 2291 &spa->spa_feat_for_write_obj) != 0) {
2292 2292 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
|
↓ open down ↓ |
2292 lines elided |
↑ open up ↑ |
2293 2293 }
2294 2294
2295 2295 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2296 2296 &spa->spa_feat_desc_obj) != 0) {
2297 2297 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2298 2298 }
2299 2299
2300 2300 enabled_feat = fnvlist_alloc();
2301 2301 unsup_feat = fnvlist_alloc();
2302 2302
2303 - if (!feature_is_supported(spa->spa_meta_objset,
2304 - spa->spa_feat_for_read_obj, spa->spa_feat_desc_obj,
2303 + if (!spa_features_check(spa, B_FALSE,
2305 2304 unsup_feat, enabled_feat))
2306 2305 missing_feat_read = B_TRUE;
2307 2306
2308 2307 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2309 - if (!feature_is_supported(spa->spa_meta_objset,
2310 - spa->spa_feat_for_write_obj, spa->spa_feat_desc_obj,
2308 + if (!spa_features_check(spa, B_TRUE,
2311 2309 unsup_feat, enabled_feat)) {
2312 2310 missing_feat_write = B_TRUE;
2313 2311 }
2314 2312 }
2315 2313
2316 2314 fnvlist_add_nvlist(spa->spa_load_info,
2317 2315 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2318 2316
2319 2317 if (!nvlist_empty(unsup_feat)) {
2320 2318 fnvlist_add_nvlist(spa->spa_load_info,
2321 2319 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2322 2320 }
2323 2321
2324 2322 fnvlist_free(enabled_feat);
2325 2323 fnvlist_free(unsup_feat);
2326 2324
2327 2325 if (!missing_feat_read) {
2328 2326 fnvlist_add_boolean(spa->spa_load_info,
2329 2327 ZPOOL_CONFIG_CAN_RDONLY);
2330 2328 }
2331 2329
2332 2330 /*
2333 2331 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2334 2332 * twofold: to determine whether the pool is available for
2335 2333 * import in read-write mode and (if it is not) whether the
2336 2334 * pool is available for import in read-only mode. If the pool
2337 2335 * is available for import in read-write mode, it is displayed
2338 2336 * as available in userland; if it is not available for import
2339 2337 * in read-only mode, it is displayed as unavailable in
2340 2338 * userland. If the pool is available for import in read-only
2341 2339 * mode but not read-write mode, it is displayed as unavailable
2342 2340 * in userland with a special note that the pool is actually
2343 2341 * available for open in read-only mode.
2344 2342 *
2345 2343 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2346 2344 * missing a feature for write, we must first determine whether
2347 2345 * the pool can be opened read-only before returning to
2348 2346 * userland in order to know whether to display the
2349 2347 * abovementioned note.
2350 2348 */
2351 2349 if (missing_feat_read || (missing_feat_write &&
2352 2350 spa_writeable(spa))) {
2353 2351 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2354 2352 ENOTSUP));
2355 2353 }
2356 2354 }
2357 2355
2358 2356 spa->spa_is_initializing = B_TRUE;
2359 2357 error = dsl_pool_open(spa->spa_dsl_pool);
2360 2358 spa->spa_is_initializing = B_FALSE;
2361 2359 if (error != 0)
2362 2360 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2363 2361
2364 2362 if (!mosconfig) {
2365 2363 uint64_t hostid;
2366 2364 nvlist_t *policy = NULL, *nvconfig;
2367 2365
2368 2366 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2369 2367 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2370 2368
2371 2369 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2372 2370 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2373 2371 char *hostname;
2374 2372 unsigned long myhostid = 0;
2375 2373
2376 2374 VERIFY(nvlist_lookup_string(nvconfig,
2377 2375 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2378 2376
2379 2377 #ifdef _KERNEL
2380 2378 myhostid = zone_get_hostid(NULL);
2381 2379 #else /* _KERNEL */
2382 2380 /*
2383 2381 * We're emulating the system's hostid in userland, so
2384 2382 * we can't use zone_get_hostid().
2385 2383 */
2386 2384 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2387 2385 #endif /* _KERNEL */
2388 2386 if (hostid != 0 && myhostid != 0 &&
2389 2387 hostid != myhostid) {
2390 2388 nvlist_free(nvconfig);
2391 2389 cmn_err(CE_WARN, "pool '%s' could not be "
2392 2390 "loaded as it was last accessed by "
2393 2391 "another system (host: %s hostid: 0x%lx). "
2394 2392 "See: http://illumos.org/msg/ZFS-8000-EY",
2395 2393 spa_name(spa), hostname,
2396 2394 (unsigned long)hostid);
2397 2395 return (SET_ERROR(EBADF));
2398 2396 }
2399 2397 }
2400 2398 if (nvlist_lookup_nvlist(spa->spa_config,
2401 2399 ZPOOL_REWIND_POLICY, &policy) == 0)
2402 2400 VERIFY(nvlist_add_nvlist(nvconfig,
2403 2401 ZPOOL_REWIND_POLICY, policy) == 0);
2404 2402
2405 2403 spa_config_set(spa, nvconfig);
2406 2404 spa_unload(spa);
2407 2405 spa_deactivate(spa);
2408 2406 spa_activate(spa, orig_mode);
2409 2407
2410 2408 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2411 2409 }
2412 2410
2413 2411 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2414 2412 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2415 2413 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2416 2414 if (error != 0)
2417 2415 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2418 2416
2419 2417 /*
2420 2418 * Load the bit that tells us to use the new accounting function
2421 2419 * (raid-z deflation). If we have an older pool, this will not
2422 2420 * be present.
2423 2421 */
2424 2422 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2425 2423 if (error != 0 && error != ENOENT)
2426 2424 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2427 2425
2428 2426 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2429 2427 &spa->spa_creation_version);
2430 2428 if (error != 0 && error != ENOENT)
2431 2429 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2432 2430
2433 2431 /*
2434 2432 * Load the persistent error log. If we have an older pool, this will
2435 2433 * not be present.
2436 2434 */
2437 2435 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2438 2436 if (error != 0 && error != ENOENT)
2439 2437 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2440 2438
2441 2439 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2442 2440 &spa->spa_errlog_scrub);
2443 2441 if (error != 0 && error != ENOENT)
2444 2442 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2445 2443
2446 2444 /*
2447 2445 * Load the history object. If we have an older pool, this
2448 2446 * will not be present.
2449 2447 */
2450 2448 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2451 2449 if (error != 0 && error != ENOENT)
2452 2450 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2453 2451
2454 2452 /*
2455 2453 * If we're assembling the pool from the split-off vdevs of
2456 2454 * an existing pool, we don't want to attach the spares & cache
2457 2455 * devices.
2458 2456 */
2459 2457
2460 2458 /*
2461 2459 * Load any hot spares for this pool.
2462 2460 */
2463 2461 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2464 2462 if (error != 0 && error != ENOENT)
2465 2463 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2466 2464 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2467 2465 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2468 2466 if (load_nvlist(spa, spa->spa_spares.sav_object,
2469 2467 &spa->spa_spares.sav_config) != 0)
2470 2468 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2471 2469
2472 2470 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2473 2471 spa_load_spares(spa);
2474 2472 spa_config_exit(spa, SCL_ALL, FTAG);
2475 2473 } else if (error == 0) {
2476 2474 spa->spa_spares.sav_sync = B_TRUE;
2477 2475 }
2478 2476
2479 2477 /*
2480 2478 * Load any level 2 ARC devices for this pool.
2481 2479 */
2482 2480 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2483 2481 &spa->spa_l2cache.sav_object);
2484 2482 if (error != 0 && error != ENOENT)
2485 2483 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2486 2484 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2487 2485 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2488 2486 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2489 2487 &spa->spa_l2cache.sav_config) != 0)
2490 2488 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2491 2489
2492 2490 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2493 2491 spa_load_l2cache(spa);
2494 2492 spa_config_exit(spa, SCL_ALL, FTAG);
2495 2493 } else if (error == 0) {
2496 2494 spa->spa_l2cache.sav_sync = B_TRUE;
2497 2495 }
2498 2496
2499 2497 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2500 2498
2501 2499 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2502 2500 if (error && error != ENOENT)
2503 2501 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2504 2502
2505 2503 if (error == 0) {
2506 2504 uint64_t autoreplace;
2507 2505
2508 2506 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2509 2507 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2510 2508 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2511 2509 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2512 2510 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2513 2511 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2514 2512 &spa->spa_dedup_ditto);
2515 2513
2516 2514 spa->spa_autoreplace = (autoreplace != 0);
2517 2515 }
2518 2516
2519 2517 /*
2520 2518 * If the 'autoreplace' property is set, then post a resource notifying
2521 2519 * the ZFS DE that it should not issue any faults for unopenable
2522 2520 * devices. We also iterate over the vdevs, and post a sysevent for any
2523 2521 * unopenable vdevs so that the normal autoreplace handler can take
2524 2522 * over.
2525 2523 */
2526 2524 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2527 2525 spa_check_removed(spa->spa_root_vdev);
2528 2526 /*
2529 2527 * For the import case, this is done in spa_import(), because
2530 2528 * at this point we're using the spare definitions from
2531 2529 * the MOS config, not necessarily from the userland config.
2532 2530 */
2533 2531 if (state != SPA_LOAD_IMPORT) {
2534 2532 spa_aux_check_removed(&spa->spa_spares);
2535 2533 spa_aux_check_removed(&spa->spa_l2cache);
2536 2534 }
2537 2535 }
2538 2536
2539 2537 /*
2540 2538 * Load the vdev state for all toplevel vdevs.
2541 2539 */
2542 2540 vdev_load(rvd);
2543 2541
2544 2542 /*
2545 2543 * Propagate the leaf DTLs we just loaded all the way up the tree.
2546 2544 */
2547 2545 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2548 2546 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2549 2547 spa_config_exit(spa, SCL_ALL, FTAG);
2550 2548
2551 2549 /*
2552 2550 * Load the DDTs (dedup tables).
2553 2551 */
2554 2552 error = ddt_load(spa);
2555 2553 if (error != 0)
2556 2554 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2557 2555
2558 2556 spa_update_dspace(spa);
2559 2557
2560 2558 /*
2561 2559 * Validate the config, using the MOS config to fill in any
2562 2560 * information which might be missing. If we fail to validate
2563 2561 * the config then declare the pool unfit for use. If we're
2564 2562 * assembling a pool from a split, the log is not transferred
2565 2563 * over.
2566 2564 */
2567 2565 if (type != SPA_IMPORT_ASSEMBLE) {
2568 2566 nvlist_t *nvconfig;
2569 2567
2570 2568 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2571 2569 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2572 2570
2573 2571 if (!spa_config_valid(spa, nvconfig)) {
2574 2572 nvlist_free(nvconfig);
2575 2573 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2576 2574 ENXIO));
2577 2575 }
2578 2576 nvlist_free(nvconfig);
2579 2577
2580 2578 /*
2581 2579 * Now that we've validated the config, check the state of the
2582 2580 * root vdev. If it can't be opened, it indicates one or
2583 2581 * more toplevel vdevs are faulted.
2584 2582 */
2585 2583 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2586 2584 return (SET_ERROR(ENXIO));
2587 2585
2588 2586 if (spa_check_logs(spa)) {
2589 2587 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2590 2588 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2591 2589 }
2592 2590 }
2593 2591
2594 2592 if (missing_feat_write) {
2595 2593 ASSERT(state == SPA_LOAD_TRYIMPORT);
2596 2594
2597 2595 /*
2598 2596 * At this point, we know that we can open the pool in
2599 2597 * read-only mode but not read-write mode. We now have enough
2600 2598 * information and can return to userland.
2601 2599 */
2602 2600 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2603 2601 }
2604 2602
2605 2603 /*
2606 2604 * We've successfully opened the pool, verify that we're ready
2607 2605 * to start pushing transactions.
2608 2606 */
2609 2607 if (state != SPA_LOAD_TRYIMPORT) {
2610 2608 if (error = spa_load_verify(spa))
2611 2609 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2612 2610 error));
2613 2611 }
2614 2612
2615 2613 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2616 2614 spa->spa_load_max_txg == UINT64_MAX)) {
2617 2615 dmu_tx_t *tx;
2618 2616 int need_update = B_FALSE;
2619 2617
2620 2618 ASSERT(state != SPA_LOAD_TRYIMPORT);
2621 2619
2622 2620 /*
2623 2621 * Claim log blocks that haven't been committed yet.
2624 2622 * This must all happen in a single txg.
2625 2623 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2626 2624 * invoked from zil_claim_log_block()'s i/o done callback.
2627 2625 * Price of rollback is that we abandon the log.
2628 2626 */
2629 2627 spa->spa_claiming = B_TRUE;
2630 2628
2631 2629 tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2632 2630 spa_first_txg(spa));
2633 2631 (void) dmu_objset_find(spa_name(spa),
2634 2632 zil_claim, tx, DS_FIND_CHILDREN);
2635 2633 dmu_tx_commit(tx);
2636 2634
2637 2635 spa->spa_claiming = B_FALSE;
2638 2636
2639 2637 spa_set_log_state(spa, SPA_LOG_GOOD);
2640 2638 spa->spa_sync_on = B_TRUE;
2641 2639 txg_sync_start(spa->spa_dsl_pool);
2642 2640
2643 2641 /*
2644 2642 * Wait for all claims to sync. We sync up to the highest
2645 2643 * claimed log block birth time so that claimed log blocks
2646 2644 * don't appear to be from the future. spa_claim_max_txg
2647 2645 * will have been set for us by either zil_check_log_chain()
2648 2646 * (invoked from spa_check_logs()) or zil_claim() above.
2649 2647 */
2650 2648 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2651 2649
2652 2650 /*
2653 2651 * If the config cache is stale, or we have uninitialized
2654 2652 * metaslabs (see spa_vdev_add()), then update the config.
2655 2653 *
2656 2654 * If this is a verbatim import, trust the current
2657 2655 * in-core spa_config and update the disk labels.
2658 2656 */
2659 2657 if (config_cache_txg != spa->spa_config_txg ||
2660 2658 state == SPA_LOAD_IMPORT ||
2661 2659 state == SPA_LOAD_RECOVER ||
2662 2660 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2663 2661 need_update = B_TRUE;
2664 2662
2665 2663 for (int c = 0; c < rvd->vdev_children; c++)
2666 2664 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2667 2665 need_update = B_TRUE;
2668 2666
2669 2667 /*
2670 2668 * Update the config cache asychronously in case we're the
2671 2669 * root pool, in which case the config cache isn't writable yet.
2672 2670 */
2673 2671 if (need_update)
2674 2672 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2675 2673
2676 2674 /*
2677 2675 * Check all DTLs to see if anything needs resilvering.
2678 2676 */
2679 2677 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2680 2678 vdev_resilver_needed(rvd, NULL, NULL))
2681 2679 spa_async_request(spa, SPA_ASYNC_RESILVER);
2682 2680
2683 2681 /*
2684 2682 * Log the fact that we booted up (so that we can detect if
2685 2683 * we rebooted in the middle of an operation).
2686 2684 */
2687 2685 spa_history_log_version(spa, "open");
2688 2686
2689 2687 /*
2690 2688 * Delete any inconsistent datasets.
2691 2689 */
2692 2690 (void) dmu_objset_find(spa_name(spa),
2693 2691 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2694 2692
2695 2693 /*
2696 2694 * Clean up any stale temporary dataset userrefs.
2697 2695 */
2698 2696 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2699 2697 }
2700 2698
2701 2699 return (0);
2702 2700 }
2703 2701
2704 2702 static int
2705 2703 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2706 2704 {
2707 2705 int mode = spa->spa_mode;
2708 2706
2709 2707 spa_unload(spa);
2710 2708 spa_deactivate(spa);
2711 2709
2712 2710 spa->spa_load_max_txg--;
2713 2711
2714 2712 spa_activate(spa, mode);
2715 2713 spa_async_suspend(spa);
2716 2714
2717 2715 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2718 2716 }
2719 2717
2720 2718 /*
2721 2719 * If spa_load() fails this function will try loading prior txg's. If
2722 2720 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2723 2721 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2724 2722 * function will not rewind the pool and will return the same error as
2725 2723 * spa_load().
2726 2724 */
2727 2725 static int
2728 2726 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2729 2727 uint64_t max_request, int rewind_flags)
2730 2728 {
2731 2729 nvlist_t *loadinfo = NULL;
2732 2730 nvlist_t *config = NULL;
2733 2731 int load_error, rewind_error;
2734 2732 uint64_t safe_rewind_txg;
2735 2733 uint64_t min_txg;
2736 2734
2737 2735 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2738 2736 spa->spa_load_max_txg = spa->spa_load_txg;
2739 2737 spa_set_log_state(spa, SPA_LOG_CLEAR);
2740 2738 } else {
2741 2739 spa->spa_load_max_txg = max_request;
2742 2740 }
2743 2741
2744 2742 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2745 2743 mosconfig);
2746 2744 if (load_error == 0)
2747 2745 return (0);
2748 2746
2749 2747 if (spa->spa_root_vdev != NULL)
2750 2748 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2751 2749
2752 2750 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2753 2751 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2754 2752
2755 2753 if (rewind_flags & ZPOOL_NEVER_REWIND) {
2756 2754 nvlist_free(config);
2757 2755 return (load_error);
2758 2756 }
2759 2757
2760 2758 if (state == SPA_LOAD_RECOVER) {
2761 2759 /* Price of rolling back is discarding txgs, including log */
2762 2760 spa_set_log_state(spa, SPA_LOG_CLEAR);
2763 2761 } else {
2764 2762 /*
2765 2763 * If we aren't rolling back save the load info from our first
2766 2764 * import attempt so that we can restore it after attempting
2767 2765 * to rewind.
2768 2766 */
2769 2767 loadinfo = spa->spa_load_info;
2770 2768 spa->spa_load_info = fnvlist_alloc();
2771 2769 }
2772 2770
2773 2771 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2774 2772 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2775 2773 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2776 2774 TXG_INITIAL : safe_rewind_txg;
2777 2775
2778 2776 /*
2779 2777 * Continue as long as we're finding errors, we're still within
2780 2778 * the acceptable rewind range, and we're still finding uberblocks
2781 2779 */
2782 2780 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2783 2781 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2784 2782 if (spa->spa_load_max_txg < safe_rewind_txg)
2785 2783 spa->spa_extreme_rewind = B_TRUE;
2786 2784 rewind_error = spa_load_retry(spa, state, mosconfig);
2787 2785 }
2788 2786
2789 2787 spa->spa_extreme_rewind = B_FALSE;
2790 2788 spa->spa_load_max_txg = UINT64_MAX;
2791 2789
2792 2790 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2793 2791 spa_config_set(spa, config);
2794 2792
2795 2793 if (state == SPA_LOAD_RECOVER) {
2796 2794 ASSERT3P(loadinfo, ==, NULL);
2797 2795 return (rewind_error);
2798 2796 } else {
2799 2797 /* Store the rewind info as part of the initial load info */
2800 2798 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2801 2799 spa->spa_load_info);
2802 2800
2803 2801 /* Restore the initial load info */
2804 2802 fnvlist_free(spa->spa_load_info);
2805 2803 spa->spa_load_info = loadinfo;
2806 2804
2807 2805 return (load_error);
2808 2806 }
2809 2807 }
2810 2808
2811 2809 /*
2812 2810 * Pool Open/Import
2813 2811 *
2814 2812 * The import case is identical to an open except that the configuration is sent
2815 2813 * down from userland, instead of grabbed from the configuration cache. For the
2816 2814 * case of an open, the pool configuration will exist in the
2817 2815 * POOL_STATE_UNINITIALIZED state.
2818 2816 *
2819 2817 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2820 2818 * the same time open the pool, without having to keep around the spa_t in some
2821 2819 * ambiguous state.
2822 2820 */
2823 2821 static int
2824 2822 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2825 2823 nvlist_t **config)
2826 2824 {
2827 2825 spa_t *spa;
2828 2826 spa_load_state_t state = SPA_LOAD_OPEN;
2829 2827 int error;
2830 2828 int locked = B_FALSE;
2831 2829
2832 2830 *spapp = NULL;
2833 2831
2834 2832 /*
2835 2833 * As disgusting as this is, we need to support recursive calls to this
2836 2834 * function because dsl_dir_open() is called during spa_load(), and ends
2837 2835 * up calling spa_open() again. The real fix is to figure out how to
2838 2836 * avoid dsl_dir_open() calling this in the first place.
2839 2837 */
2840 2838 if (mutex_owner(&spa_namespace_lock) != curthread) {
2841 2839 mutex_enter(&spa_namespace_lock);
2842 2840 locked = B_TRUE;
2843 2841 }
2844 2842
2845 2843 if ((spa = spa_lookup(pool)) == NULL) {
2846 2844 if (locked)
2847 2845 mutex_exit(&spa_namespace_lock);
2848 2846 return (SET_ERROR(ENOENT));
2849 2847 }
2850 2848
2851 2849 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2852 2850 zpool_rewind_policy_t policy;
2853 2851
2854 2852 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2855 2853 &policy);
2856 2854 if (policy.zrp_request & ZPOOL_DO_REWIND)
2857 2855 state = SPA_LOAD_RECOVER;
2858 2856
2859 2857 spa_activate(spa, spa_mode_global);
2860 2858
2861 2859 if (state != SPA_LOAD_RECOVER)
2862 2860 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2863 2861
2864 2862 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2865 2863 policy.zrp_request);
2866 2864
2867 2865 if (error == EBADF) {
2868 2866 /*
2869 2867 * If vdev_validate() returns failure (indicated by
2870 2868 * EBADF), it indicates that one of the vdevs indicates
2871 2869 * that the pool has been exported or destroyed. If
2872 2870 * this is the case, the config cache is out of sync and
2873 2871 * we should remove the pool from the namespace.
2874 2872 */
2875 2873 spa_unload(spa);
2876 2874 spa_deactivate(spa);
2877 2875 spa_config_sync(spa, B_TRUE, B_TRUE);
2878 2876 spa_remove(spa);
2879 2877 if (locked)
2880 2878 mutex_exit(&spa_namespace_lock);
2881 2879 return (SET_ERROR(ENOENT));
2882 2880 }
2883 2881
2884 2882 if (error) {
2885 2883 /*
2886 2884 * We can't open the pool, but we still have useful
2887 2885 * information: the state of each vdev after the
2888 2886 * attempted vdev_open(). Return this to the user.
2889 2887 */
2890 2888 if (config != NULL && spa->spa_config) {
2891 2889 VERIFY(nvlist_dup(spa->spa_config, config,
2892 2890 KM_SLEEP) == 0);
2893 2891 VERIFY(nvlist_add_nvlist(*config,
2894 2892 ZPOOL_CONFIG_LOAD_INFO,
2895 2893 spa->spa_load_info) == 0);
2896 2894 }
2897 2895 spa_unload(spa);
2898 2896 spa_deactivate(spa);
2899 2897 spa->spa_last_open_failed = error;
2900 2898 if (locked)
2901 2899 mutex_exit(&spa_namespace_lock);
2902 2900 *spapp = NULL;
2903 2901 return (error);
2904 2902 }
2905 2903 }
2906 2904
2907 2905 spa_open_ref(spa, tag);
2908 2906
2909 2907 if (config != NULL)
2910 2908 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2911 2909
2912 2910 /*
2913 2911 * If we've recovered the pool, pass back any information we
2914 2912 * gathered while doing the load.
2915 2913 */
2916 2914 if (state == SPA_LOAD_RECOVER) {
2917 2915 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2918 2916 spa->spa_load_info) == 0);
2919 2917 }
2920 2918
2921 2919 if (locked) {
2922 2920 spa->spa_last_open_failed = 0;
2923 2921 spa->spa_last_ubsync_txg = 0;
2924 2922 spa->spa_load_txg = 0;
2925 2923 mutex_exit(&spa_namespace_lock);
2926 2924 }
2927 2925
2928 2926 *spapp = spa;
2929 2927
2930 2928 return (0);
2931 2929 }
2932 2930
2933 2931 int
2934 2932 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2935 2933 nvlist_t **config)
2936 2934 {
2937 2935 return (spa_open_common(name, spapp, tag, policy, config));
2938 2936 }
2939 2937
2940 2938 int
2941 2939 spa_open(const char *name, spa_t **spapp, void *tag)
2942 2940 {
2943 2941 return (spa_open_common(name, spapp, tag, NULL, NULL));
2944 2942 }
2945 2943
2946 2944 /*
2947 2945 * Lookup the given spa_t, incrementing the inject count in the process,
2948 2946 * preventing it from being exported or destroyed.
2949 2947 */
2950 2948 spa_t *
2951 2949 spa_inject_addref(char *name)
2952 2950 {
2953 2951 spa_t *spa;
2954 2952
2955 2953 mutex_enter(&spa_namespace_lock);
2956 2954 if ((spa = spa_lookup(name)) == NULL) {
2957 2955 mutex_exit(&spa_namespace_lock);
2958 2956 return (NULL);
2959 2957 }
2960 2958 spa->spa_inject_ref++;
2961 2959 mutex_exit(&spa_namespace_lock);
2962 2960
2963 2961 return (spa);
2964 2962 }
2965 2963
2966 2964 void
2967 2965 spa_inject_delref(spa_t *spa)
2968 2966 {
2969 2967 mutex_enter(&spa_namespace_lock);
2970 2968 spa->spa_inject_ref--;
2971 2969 mutex_exit(&spa_namespace_lock);
2972 2970 }
2973 2971
2974 2972 /*
2975 2973 * Add spares device information to the nvlist.
2976 2974 */
2977 2975 static void
2978 2976 spa_add_spares(spa_t *spa, nvlist_t *config)
2979 2977 {
2980 2978 nvlist_t **spares;
2981 2979 uint_t i, nspares;
2982 2980 nvlist_t *nvroot;
2983 2981 uint64_t guid;
2984 2982 vdev_stat_t *vs;
2985 2983 uint_t vsc;
2986 2984 uint64_t pool;
2987 2985
2988 2986 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2989 2987
2990 2988 if (spa->spa_spares.sav_count == 0)
2991 2989 return;
2992 2990
2993 2991 VERIFY(nvlist_lookup_nvlist(config,
2994 2992 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2995 2993 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2996 2994 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2997 2995 if (nspares != 0) {
2998 2996 VERIFY(nvlist_add_nvlist_array(nvroot,
2999 2997 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3000 2998 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3001 2999 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3002 3000
3003 3001 /*
3004 3002 * Go through and find any spares which have since been
3005 3003 * repurposed as an active spare. If this is the case, update
3006 3004 * their status appropriately.
3007 3005 */
3008 3006 for (i = 0; i < nspares; i++) {
3009 3007 VERIFY(nvlist_lookup_uint64(spares[i],
3010 3008 ZPOOL_CONFIG_GUID, &guid) == 0);
3011 3009 if (spa_spare_exists(guid, &pool, NULL) &&
3012 3010 pool != 0ULL) {
3013 3011 VERIFY(nvlist_lookup_uint64_array(
3014 3012 spares[i], ZPOOL_CONFIG_VDEV_STATS,
3015 3013 (uint64_t **)&vs, &vsc) == 0);
3016 3014 vs->vs_state = VDEV_STATE_CANT_OPEN;
3017 3015 vs->vs_aux = VDEV_AUX_SPARED;
3018 3016 }
3019 3017 }
3020 3018 }
3021 3019 }
3022 3020
3023 3021 /*
3024 3022 * Add l2cache device information to the nvlist, including vdev stats.
3025 3023 */
3026 3024 static void
3027 3025 spa_add_l2cache(spa_t *spa, nvlist_t *config)
3028 3026 {
3029 3027 nvlist_t **l2cache;
3030 3028 uint_t i, j, nl2cache;
3031 3029 nvlist_t *nvroot;
3032 3030 uint64_t guid;
3033 3031 vdev_t *vd;
3034 3032 vdev_stat_t *vs;
3035 3033 uint_t vsc;
3036 3034
3037 3035 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3038 3036
3039 3037 if (spa->spa_l2cache.sav_count == 0)
3040 3038 return;
3041 3039
3042 3040 VERIFY(nvlist_lookup_nvlist(config,
3043 3041 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3044 3042 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3045 3043 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3046 3044 if (nl2cache != 0) {
3047 3045 VERIFY(nvlist_add_nvlist_array(nvroot,
3048 3046 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3049 3047 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3050 3048 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3051 3049
3052 3050 /*
3053 3051 * Update level 2 cache device stats.
3054 3052 */
3055 3053
3056 3054 for (i = 0; i < nl2cache; i++) {
3057 3055 VERIFY(nvlist_lookup_uint64(l2cache[i],
3058 3056 ZPOOL_CONFIG_GUID, &guid) == 0);
3059 3057
3060 3058 vd = NULL;
3061 3059 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3062 3060 if (guid ==
3063 3061 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3064 3062 vd = spa->spa_l2cache.sav_vdevs[j];
3065 3063 break;
3066 3064 }
3067 3065 }
3068 3066 ASSERT(vd != NULL);
3069 3067
3070 3068 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3071 3069 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3072 3070 == 0);
3073 3071 vdev_get_stats(vd, vs);
3074 3072 }
3075 3073 }
3076 3074 }
3077 3075
3078 3076 static void
3079 3077 spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3080 3078 {
3081 3079 nvlist_t *features;
3082 3080 zap_cursor_t zc;
3083 3081 zap_attribute_t za;
3084 3082
3085 3083 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3086 3084 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3087 3085
3088 3086 if (spa->spa_feat_for_read_obj != 0) {
3089 3087 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3090 3088 spa->spa_feat_for_read_obj);
3091 3089 zap_cursor_retrieve(&zc, &za) == 0;
3092 3090 zap_cursor_advance(&zc)) {
3093 3091 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3094 3092 za.za_num_integers == 1);
3095 3093 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3096 3094 za.za_first_integer));
3097 3095 }
3098 3096 zap_cursor_fini(&zc);
3099 3097 }
3100 3098
3101 3099 if (spa->spa_feat_for_write_obj != 0) {
3102 3100 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3103 3101 spa->spa_feat_for_write_obj);
3104 3102 zap_cursor_retrieve(&zc, &za) == 0;
3105 3103 zap_cursor_advance(&zc)) {
3106 3104 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3107 3105 za.za_num_integers == 1);
3108 3106 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3109 3107 za.za_first_integer));
3110 3108 }
3111 3109 zap_cursor_fini(&zc);
3112 3110 }
3113 3111
3114 3112 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3115 3113 features) == 0);
3116 3114 nvlist_free(features);
3117 3115 }
3118 3116
3119 3117 int
3120 3118 spa_get_stats(const char *name, nvlist_t **config,
3121 3119 char *altroot, size_t buflen)
3122 3120 {
3123 3121 int error;
3124 3122 spa_t *spa;
3125 3123
3126 3124 *config = NULL;
3127 3125 error = spa_open_common(name, &spa, FTAG, NULL, config);
3128 3126
3129 3127 if (spa != NULL) {
3130 3128 /*
3131 3129 * This still leaves a window of inconsistency where the spares
3132 3130 * or l2cache devices could change and the config would be
3133 3131 * self-inconsistent.
3134 3132 */
3135 3133 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3136 3134
3137 3135 if (*config != NULL) {
3138 3136 uint64_t loadtimes[2];
3139 3137
3140 3138 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3141 3139 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3142 3140 VERIFY(nvlist_add_uint64_array(*config,
3143 3141 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3144 3142
3145 3143 VERIFY(nvlist_add_uint64(*config,
3146 3144 ZPOOL_CONFIG_ERRCOUNT,
3147 3145 spa_get_errlog_size(spa)) == 0);
3148 3146
3149 3147 if (spa_suspended(spa))
3150 3148 VERIFY(nvlist_add_uint64(*config,
3151 3149 ZPOOL_CONFIG_SUSPENDED,
3152 3150 spa->spa_failmode) == 0);
3153 3151
3154 3152 spa_add_spares(spa, *config);
3155 3153 spa_add_l2cache(spa, *config);
3156 3154 spa_add_feature_stats(spa, *config);
3157 3155 }
3158 3156 }
3159 3157
3160 3158 /*
3161 3159 * We want to get the alternate root even for faulted pools, so we cheat
3162 3160 * and call spa_lookup() directly.
3163 3161 */
3164 3162 if (altroot) {
3165 3163 if (spa == NULL) {
3166 3164 mutex_enter(&spa_namespace_lock);
3167 3165 spa = spa_lookup(name);
3168 3166 if (spa)
3169 3167 spa_altroot(spa, altroot, buflen);
3170 3168 else
3171 3169 altroot[0] = '\0';
3172 3170 spa = NULL;
3173 3171 mutex_exit(&spa_namespace_lock);
3174 3172 } else {
3175 3173 spa_altroot(spa, altroot, buflen);
3176 3174 }
3177 3175 }
3178 3176
3179 3177 if (spa != NULL) {
3180 3178 spa_config_exit(spa, SCL_CONFIG, FTAG);
3181 3179 spa_close(spa, FTAG);
3182 3180 }
3183 3181
3184 3182 return (error);
3185 3183 }
3186 3184
3187 3185 /*
3188 3186 * Validate that the auxiliary device array is well formed. We must have an
3189 3187 * array of nvlists, each which describes a valid leaf vdev. If this is an
3190 3188 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3191 3189 * specified, as long as they are well-formed.
3192 3190 */
3193 3191 static int
3194 3192 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3195 3193 spa_aux_vdev_t *sav, const char *config, uint64_t version,
3196 3194 vdev_labeltype_t label)
3197 3195 {
3198 3196 nvlist_t **dev;
3199 3197 uint_t i, ndev;
3200 3198 vdev_t *vd;
3201 3199 int error;
3202 3200
3203 3201 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3204 3202
3205 3203 /*
3206 3204 * It's acceptable to have no devs specified.
3207 3205 */
3208 3206 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3209 3207 return (0);
3210 3208
3211 3209 if (ndev == 0)
3212 3210 return (SET_ERROR(EINVAL));
3213 3211
3214 3212 /*
3215 3213 * Make sure the pool is formatted with a version that supports this
3216 3214 * device type.
3217 3215 */
3218 3216 if (spa_version(spa) < version)
3219 3217 return (SET_ERROR(ENOTSUP));
3220 3218
3221 3219 /*
3222 3220 * Set the pending device list so we correctly handle device in-use
3223 3221 * checking.
3224 3222 */
3225 3223 sav->sav_pending = dev;
3226 3224 sav->sav_npending = ndev;
3227 3225
3228 3226 for (i = 0; i < ndev; i++) {
3229 3227 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3230 3228 mode)) != 0)
3231 3229 goto out;
3232 3230
3233 3231 if (!vd->vdev_ops->vdev_op_leaf) {
3234 3232 vdev_free(vd);
3235 3233 error = SET_ERROR(EINVAL);
3236 3234 goto out;
3237 3235 }
3238 3236
3239 3237 /*
3240 3238 * The L2ARC currently only supports disk devices in
3241 3239 * kernel context. For user-level testing, we allow it.
3242 3240 */
3243 3241 #ifdef _KERNEL
3244 3242 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3245 3243 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3246 3244 error = SET_ERROR(ENOTBLK);
3247 3245 vdev_free(vd);
3248 3246 goto out;
3249 3247 }
3250 3248 #endif
3251 3249 vd->vdev_top = vd;
3252 3250
3253 3251 if ((error = vdev_open(vd)) == 0 &&
3254 3252 (error = vdev_label_init(vd, crtxg, label)) == 0) {
3255 3253 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3256 3254 vd->vdev_guid) == 0);
3257 3255 }
3258 3256
3259 3257 vdev_free(vd);
3260 3258
3261 3259 if (error &&
3262 3260 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3263 3261 goto out;
3264 3262 else
3265 3263 error = 0;
3266 3264 }
3267 3265
3268 3266 out:
3269 3267 sav->sav_pending = NULL;
3270 3268 sav->sav_npending = 0;
3271 3269 return (error);
3272 3270 }
3273 3271
3274 3272 static int
3275 3273 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3276 3274 {
3277 3275 int error;
3278 3276
3279 3277 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3280 3278
3281 3279 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3282 3280 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3283 3281 VDEV_LABEL_SPARE)) != 0) {
3284 3282 return (error);
3285 3283 }
3286 3284
3287 3285 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3288 3286 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3289 3287 VDEV_LABEL_L2CACHE));
3290 3288 }
3291 3289
3292 3290 static void
3293 3291 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3294 3292 const char *config)
3295 3293 {
3296 3294 int i;
3297 3295
3298 3296 if (sav->sav_config != NULL) {
3299 3297 nvlist_t **olddevs;
3300 3298 uint_t oldndevs;
3301 3299 nvlist_t **newdevs;
3302 3300
3303 3301 /*
3304 3302 * Generate new dev list by concatentating with the
3305 3303 * current dev list.
3306 3304 */
3307 3305 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3308 3306 &olddevs, &oldndevs) == 0);
3309 3307
3310 3308 newdevs = kmem_alloc(sizeof (void *) *
3311 3309 (ndevs + oldndevs), KM_SLEEP);
3312 3310 for (i = 0; i < oldndevs; i++)
3313 3311 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3314 3312 KM_SLEEP) == 0);
3315 3313 for (i = 0; i < ndevs; i++)
3316 3314 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3317 3315 KM_SLEEP) == 0);
3318 3316
3319 3317 VERIFY(nvlist_remove(sav->sav_config, config,
3320 3318 DATA_TYPE_NVLIST_ARRAY) == 0);
3321 3319
3322 3320 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3323 3321 config, newdevs, ndevs + oldndevs) == 0);
3324 3322 for (i = 0; i < oldndevs + ndevs; i++)
3325 3323 nvlist_free(newdevs[i]);
3326 3324 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3327 3325 } else {
3328 3326 /*
3329 3327 * Generate a new dev list.
3330 3328 */
3331 3329 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3332 3330 KM_SLEEP) == 0);
3333 3331 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3334 3332 devs, ndevs) == 0);
3335 3333 }
3336 3334 }
3337 3335
3338 3336 /*
3339 3337 * Stop and drop level 2 ARC devices
3340 3338 */
3341 3339 void
3342 3340 spa_l2cache_drop(spa_t *spa)
3343 3341 {
3344 3342 vdev_t *vd;
3345 3343 int i;
3346 3344 spa_aux_vdev_t *sav = &spa->spa_l2cache;
3347 3345
3348 3346 for (i = 0; i < sav->sav_count; i++) {
3349 3347 uint64_t pool;
3350 3348
3351 3349 vd = sav->sav_vdevs[i];
3352 3350 ASSERT(vd != NULL);
3353 3351
3354 3352 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3355 3353 pool != 0ULL && l2arc_vdev_present(vd))
3356 3354 l2arc_remove_vdev(vd);
3357 3355 }
3358 3356 }
3359 3357
3360 3358 /*
3361 3359 * Pool Creation
3362 3360 */
3363 3361 int
3364 3362 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3365 3363 nvlist_t *zplprops)
3366 3364 {
3367 3365 spa_t *spa;
3368 3366 char *altroot = NULL;
3369 3367 vdev_t *rvd;
3370 3368 dsl_pool_t *dp;
3371 3369 dmu_tx_t *tx;
3372 3370 int error = 0;
3373 3371 uint64_t txg = TXG_INITIAL;
3374 3372 nvlist_t **spares, **l2cache;
3375 3373 uint_t nspares, nl2cache;
3376 3374 uint64_t version, obj;
3377 3375 boolean_t has_features;
3378 3376
3379 3377 /*
3380 3378 * If this pool already exists, return failure.
3381 3379 */
3382 3380 mutex_enter(&spa_namespace_lock);
3383 3381 if (spa_lookup(pool) != NULL) {
3384 3382 mutex_exit(&spa_namespace_lock);
3385 3383 return (SET_ERROR(EEXIST));
3386 3384 }
3387 3385
3388 3386 /*
3389 3387 * Allocate a new spa_t structure.
3390 3388 */
3391 3389 (void) nvlist_lookup_string(props,
3392 3390 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3393 3391 spa = spa_add(pool, NULL, altroot);
3394 3392 spa_activate(spa, spa_mode_global);
3395 3393
3396 3394 if (props && (error = spa_prop_validate(spa, props))) {
3397 3395 spa_deactivate(spa);
3398 3396 spa_remove(spa);
3399 3397 mutex_exit(&spa_namespace_lock);
3400 3398 return (error);
3401 3399 }
3402 3400
3403 3401 has_features = B_FALSE;
3404 3402 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3405 3403 elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3406 3404 if (zpool_prop_feature(nvpair_name(elem)))
3407 3405 has_features = B_TRUE;
3408 3406 }
3409 3407
3410 3408 if (has_features || nvlist_lookup_uint64(props,
3411 3409 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3412 3410 version = SPA_VERSION;
3413 3411 }
3414 3412 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3415 3413
3416 3414 spa->spa_first_txg = txg;
3417 3415 spa->spa_uberblock.ub_txg = txg - 1;
3418 3416 spa->spa_uberblock.ub_version = version;
3419 3417 spa->spa_ubsync = spa->spa_uberblock;
3420 3418
3421 3419 /*
3422 3420 * Create "The Godfather" zio to hold all async IOs
3423 3421 */
3424 3422 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
3425 3423 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
3426 3424
3427 3425 /*
3428 3426 * Create the root vdev.
3429 3427 */
3430 3428 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3431 3429
3432 3430 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3433 3431
3434 3432 ASSERT(error != 0 || rvd != NULL);
3435 3433 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3436 3434
3437 3435 if (error == 0 && !zfs_allocatable_devs(nvroot))
3438 3436 error = SET_ERROR(EINVAL);
3439 3437
3440 3438 if (error == 0 &&
3441 3439 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3442 3440 (error = spa_validate_aux(spa, nvroot, txg,
3443 3441 VDEV_ALLOC_ADD)) == 0) {
3444 3442 for (int c = 0; c < rvd->vdev_children; c++) {
3445 3443 vdev_metaslab_set_size(rvd->vdev_child[c]);
3446 3444 vdev_expand(rvd->vdev_child[c], txg);
3447 3445 }
3448 3446 }
3449 3447
3450 3448 spa_config_exit(spa, SCL_ALL, FTAG);
3451 3449
3452 3450 if (error != 0) {
3453 3451 spa_unload(spa);
3454 3452 spa_deactivate(spa);
3455 3453 spa_remove(spa);
3456 3454 mutex_exit(&spa_namespace_lock);
3457 3455 return (error);
3458 3456 }
3459 3457
3460 3458 /*
3461 3459 * Get the list of spares, if specified.
3462 3460 */
3463 3461 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3464 3462 &spares, &nspares) == 0) {
3465 3463 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3466 3464 KM_SLEEP) == 0);
3467 3465 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3468 3466 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3469 3467 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3470 3468 spa_load_spares(spa);
3471 3469 spa_config_exit(spa, SCL_ALL, FTAG);
3472 3470 spa->spa_spares.sav_sync = B_TRUE;
3473 3471 }
3474 3472
3475 3473 /*
3476 3474 * Get the list of level 2 cache devices, if specified.
3477 3475 */
3478 3476 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3479 3477 &l2cache, &nl2cache) == 0) {
3480 3478 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3481 3479 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3482 3480 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3483 3481 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3484 3482 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3485 3483 spa_load_l2cache(spa);
3486 3484 spa_config_exit(spa, SCL_ALL, FTAG);
3487 3485 spa->spa_l2cache.sav_sync = B_TRUE;
3488 3486 }
3489 3487
3490 3488 spa->spa_is_initializing = B_TRUE;
3491 3489 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3492 3490 spa->spa_meta_objset = dp->dp_meta_objset;
3493 3491 spa->spa_is_initializing = B_FALSE;
3494 3492
3495 3493 /*
3496 3494 * Create DDTs (dedup tables).
3497 3495 */
3498 3496 ddt_create(spa);
3499 3497
3500 3498 spa_update_dspace(spa);
3501 3499
3502 3500 tx = dmu_tx_create_assigned(dp, txg);
3503 3501
3504 3502 /*
3505 3503 * Create the pool config object.
3506 3504 */
3507 3505 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3508 3506 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3509 3507 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3510 3508
3511 3509 if (zap_add(spa->spa_meta_objset,
3512 3510 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3513 3511 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3514 3512 cmn_err(CE_PANIC, "failed to add pool config");
3515 3513 }
3516 3514
3517 3515 if (spa_version(spa) >= SPA_VERSION_FEATURES)
3518 3516 spa_feature_create_zap_objects(spa, tx);
3519 3517
3520 3518 if (zap_add(spa->spa_meta_objset,
3521 3519 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3522 3520 sizeof (uint64_t), 1, &version, tx) != 0) {
3523 3521 cmn_err(CE_PANIC, "failed to add pool version");
3524 3522 }
3525 3523
3526 3524 /* Newly created pools with the right version are always deflated. */
3527 3525 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3528 3526 spa->spa_deflate = TRUE;
3529 3527 if (zap_add(spa->spa_meta_objset,
3530 3528 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3531 3529 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3532 3530 cmn_err(CE_PANIC, "failed to add deflate");
3533 3531 }
3534 3532 }
3535 3533
3536 3534 /*
3537 3535 * Create the deferred-free bpobj. Turn off compression
3538 3536 * because sync-to-convergence takes longer if the blocksize
3539 3537 * keeps changing.
3540 3538 */
3541 3539 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3542 3540 dmu_object_set_compress(spa->spa_meta_objset, obj,
3543 3541 ZIO_COMPRESS_OFF, tx);
3544 3542 if (zap_add(spa->spa_meta_objset,
3545 3543 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3546 3544 sizeof (uint64_t), 1, &obj, tx) != 0) {
3547 3545 cmn_err(CE_PANIC, "failed to add bpobj");
3548 3546 }
3549 3547 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3550 3548 spa->spa_meta_objset, obj));
3551 3549
3552 3550 /*
3553 3551 * Create the pool's history object.
3554 3552 */
3555 3553 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3556 3554 spa_history_create_obj(spa, tx);
3557 3555
3558 3556 /*
3559 3557 * Set pool properties.
3560 3558 */
3561 3559 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3562 3560 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3563 3561 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3564 3562 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3565 3563
3566 3564 if (props != NULL) {
3567 3565 spa_configfile_set(spa, props, B_FALSE);
3568 3566 spa_sync_props(props, tx);
3569 3567 }
3570 3568
3571 3569 dmu_tx_commit(tx);
3572 3570
3573 3571 spa->spa_sync_on = B_TRUE;
3574 3572 txg_sync_start(spa->spa_dsl_pool);
3575 3573
3576 3574 /*
3577 3575 * We explicitly wait for the first transaction to complete so that our
3578 3576 * bean counters are appropriately updated.
3579 3577 */
3580 3578 txg_wait_synced(spa->spa_dsl_pool, txg);
3581 3579
3582 3580 spa_config_sync(spa, B_FALSE, B_TRUE);
3583 3581
3584 3582 spa_history_log_version(spa, "create");
3585 3583
3586 3584 spa->spa_minref = refcount_count(&spa->spa_refcount);
3587 3585
3588 3586 mutex_exit(&spa_namespace_lock);
3589 3587
3590 3588 return (0);
3591 3589 }
3592 3590
3593 3591 #ifdef _KERNEL
3594 3592 /*
3595 3593 * Get the root pool information from the root disk, then import the root pool
3596 3594 * during the system boot up time.
3597 3595 */
3598 3596 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3599 3597
3600 3598 static nvlist_t *
3601 3599 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3602 3600 {
3603 3601 nvlist_t *config;
3604 3602 nvlist_t *nvtop, *nvroot;
3605 3603 uint64_t pgid;
3606 3604
3607 3605 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3608 3606 return (NULL);
3609 3607
3610 3608 /*
3611 3609 * Add this top-level vdev to the child array.
3612 3610 */
3613 3611 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3614 3612 &nvtop) == 0);
3615 3613 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3616 3614 &pgid) == 0);
3617 3615 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3618 3616
3619 3617 /*
3620 3618 * Put this pool's top-level vdevs into a root vdev.
3621 3619 */
3622 3620 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3623 3621 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3624 3622 VDEV_TYPE_ROOT) == 0);
3625 3623 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3626 3624 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3627 3625 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3628 3626 &nvtop, 1) == 0);
3629 3627
3630 3628 /*
3631 3629 * Replace the existing vdev_tree with the new root vdev in
3632 3630 * this pool's configuration (remove the old, add the new).
3633 3631 */
3634 3632 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3635 3633 nvlist_free(nvroot);
3636 3634 return (config);
3637 3635 }
3638 3636
3639 3637 /*
3640 3638 * Walk the vdev tree and see if we can find a device with "better"
3641 3639 * configuration. A configuration is "better" if the label on that
3642 3640 * device has a more recent txg.
3643 3641 */
3644 3642 static void
3645 3643 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3646 3644 {
3647 3645 for (int c = 0; c < vd->vdev_children; c++)
3648 3646 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3649 3647
3650 3648 if (vd->vdev_ops->vdev_op_leaf) {
3651 3649 nvlist_t *label;
3652 3650 uint64_t label_txg;
3653 3651
3654 3652 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3655 3653 &label) != 0)
3656 3654 return;
3657 3655
3658 3656 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3659 3657 &label_txg) == 0);
3660 3658
3661 3659 /*
3662 3660 * Do we have a better boot device?
3663 3661 */
3664 3662 if (label_txg > *txg) {
3665 3663 *txg = label_txg;
3666 3664 *avd = vd;
3667 3665 }
3668 3666 nvlist_free(label);
3669 3667 }
3670 3668 }
3671 3669
3672 3670 /*
3673 3671 * Import a root pool.
3674 3672 *
3675 3673 * For x86. devpath_list will consist of devid and/or physpath name of
3676 3674 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3677 3675 * The GRUB "findroot" command will return the vdev we should boot.
3678 3676 *
3679 3677 * For Sparc, devpath_list consists the physpath name of the booting device
3680 3678 * no matter the rootpool is a single device pool or a mirrored pool.
3681 3679 * e.g.
3682 3680 * "/pci@1f,0/ide@d/disk@0,0:a"
3683 3681 */
3684 3682 int
3685 3683 spa_import_rootpool(char *devpath, char *devid)
3686 3684 {
3687 3685 spa_t *spa;
3688 3686 vdev_t *rvd, *bvd, *avd = NULL;
3689 3687 nvlist_t *config, *nvtop;
3690 3688 uint64_t guid, txg;
3691 3689 char *pname;
3692 3690 int error;
3693 3691
3694 3692 /*
3695 3693 * Read the label from the boot device and generate a configuration.
3696 3694 */
3697 3695 config = spa_generate_rootconf(devpath, devid, &guid);
3698 3696 #if defined(_OBP) && defined(_KERNEL)
3699 3697 if (config == NULL) {
3700 3698 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3701 3699 /* iscsi boot */
3702 3700 get_iscsi_bootpath_phy(devpath);
3703 3701 config = spa_generate_rootconf(devpath, devid, &guid);
3704 3702 }
3705 3703 }
3706 3704 #endif
3707 3705 if (config == NULL) {
3708 3706 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3709 3707 devpath);
3710 3708 return (SET_ERROR(EIO));
3711 3709 }
3712 3710
3713 3711 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3714 3712 &pname) == 0);
3715 3713 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3716 3714
3717 3715 mutex_enter(&spa_namespace_lock);
3718 3716 if ((spa = spa_lookup(pname)) != NULL) {
3719 3717 /*
3720 3718 * Remove the existing root pool from the namespace so that we
3721 3719 * can replace it with the correct config we just read in.
3722 3720 */
3723 3721 spa_remove(spa);
3724 3722 }
3725 3723
3726 3724 spa = spa_add(pname, config, NULL);
3727 3725 spa->spa_is_root = B_TRUE;
3728 3726 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3729 3727
3730 3728 /*
3731 3729 * Build up a vdev tree based on the boot device's label config.
3732 3730 */
3733 3731 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3734 3732 &nvtop) == 0);
3735 3733 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3736 3734 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3737 3735 VDEV_ALLOC_ROOTPOOL);
3738 3736 spa_config_exit(spa, SCL_ALL, FTAG);
3739 3737 if (error) {
3740 3738 mutex_exit(&spa_namespace_lock);
3741 3739 nvlist_free(config);
3742 3740 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3743 3741 pname);
3744 3742 return (error);
3745 3743 }
3746 3744
3747 3745 /*
3748 3746 * Get the boot vdev.
3749 3747 */
3750 3748 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3751 3749 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3752 3750 (u_longlong_t)guid);
3753 3751 error = SET_ERROR(ENOENT);
3754 3752 goto out;
3755 3753 }
3756 3754
3757 3755 /*
3758 3756 * Determine if there is a better boot device.
3759 3757 */
3760 3758 avd = bvd;
3761 3759 spa_alt_rootvdev(rvd, &avd, &txg);
3762 3760 if (avd != bvd) {
3763 3761 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3764 3762 "try booting from '%s'", avd->vdev_path);
3765 3763 error = SET_ERROR(EINVAL);
3766 3764 goto out;
3767 3765 }
3768 3766
3769 3767 /*
3770 3768 * If the boot device is part of a spare vdev then ensure that
3771 3769 * we're booting off the active spare.
3772 3770 */
3773 3771 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3774 3772 !bvd->vdev_isspare) {
3775 3773 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3776 3774 "try booting from '%s'",
3777 3775 bvd->vdev_parent->
3778 3776 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3779 3777 error = SET_ERROR(EINVAL);
3780 3778 goto out;
3781 3779 }
3782 3780
3783 3781 error = 0;
3784 3782 out:
3785 3783 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3786 3784 vdev_free(rvd);
3787 3785 spa_config_exit(spa, SCL_ALL, FTAG);
3788 3786 mutex_exit(&spa_namespace_lock);
3789 3787
3790 3788 nvlist_free(config);
3791 3789 return (error);
3792 3790 }
3793 3791
3794 3792 #endif
3795 3793
3796 3794 /*
3797 3795 * Import a non-root pool into the system.
3798 3796 */
3799 3797 int
3800 3798 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3801 3799 {
3802 3800 spa_t *spa;
3803 3801 char *altroot = NULL;
3804 3802 spa_load_state_t state = SPA_LOAD_IMPORT;
3805 3803 zpool_rewind_policy_t policy;
3806 3804 uint64_t mode = spa_mode_global;
3807 3805 uint64_t readonly = B_FALSE;
3808 3806 int error;
3809 3807 nvlist_t *nvroot;
3810 3808 nvlist_t **spares, **l2cache;
3811 3809 uint_t nspares, nl2cache;
3812 3810
3813 3811 /*
3814 3812 * If a pool with this name exists, return failure.
3815 3813 */
3816 3814 mutex_enter(&spa_namespace_lock);
3817 3815 if (spa_lookup(pool) != NULL) {
3818 3816 mutex_exit(&spa_namespace_lock);
3819 3817 return (SET_ERROR(EEXIST));
3820 3818 }
3821 3819
3822 3820 /*
3823 3821 * Create and initialize the spa structure.
3824 3822 */
3825 3823 (void) nvlist_lookup_string(props,
3826 3824 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3827 3825 (void) nvlist_lookup_uint64(props,
3828 3826 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3829 3827 if (readonly)
3830 3828 mode = FREAD;
3831 3829 spa = spa_add(pool, config, altroot);
3832 3830 spa->spa_import_flags = flags;
3833 3831
3834 3832 /*
3835 3833 * Verbatim import - Take a pool and insert it into the namespace
3836 3834 * as if it had been loaded at boot.
3837 3835 */
3838 3836 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3839 3837 if (props != NULL)
3840 3838 spa_configfile_set(spa, props, B_FALSE);
3841 3839
3842 3840 spa_config_sync(spa, B_FALSE, B_TRUE);
3843 3841
3844 3842 mutex_exit(&spa_namespace_lock);
3845 3843 return (0);
3846 3844 }
3847 3845
3848 3846 spa_activate(spa, mode);
3849 3847
3850 3848 /*
3851 3849 * Don't start async tasks until we know everything is healthy.
3852 3850 */
3853 3851 spa_async_suspend(spa);
3854 3852
3855 3853 zpool_get_rewind_policy(config, &policy);
3856 3854 if (policy.zrp_request & ZPOOL_DO_REWIND)
3857 3855 state = SPA_LOAD_RECOVER;
3858 3856
3859 3857 /*
3860 3858 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
3861 3859 * because the user-supplied config is actually the one to trust when
3862 3860 * doing an import.
3863 3861 */
3864 3862 if (state != SPA_LOAD_RECOVER)
3865 3863 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3866 3864
3867 3865 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3868 3866 policy.zrp_request);
3869 3867
3870 3868 /*
3871 3869 * Propagate anything learned while loading the pool and pass it
3872 3870 * back to caller (i.e. rewind info, missing devices, etc).
3873 3871 */
3874 3872 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3875 3873 spa->spa_load_info) == 0);
3876 3874
3877 3875 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3878 3876 /*
3879 3877 * Toss any existing sparelist, as it doesn't have any validity
3880 3878 * anymore, and conflicts with spa_has_spare().
3881 3879 */
3882 3880 if (spa->spa_spares.sav_config) {
3883 3881 nvlist_free(spa->spa_spares.sav_config);
3884 3882 spa->spa_spares.sav_config = NULL;
3885 3883 spa_load_spares(spa);
3886 3884 }
3887 3885 if (spa->spa_l2cache.sav_config) {
3888 3886 nvlist_free(spa->spa_l2cache.sav_config);
3889 3887 spa->spa_l2cache.sav_config = NULL;
3890 3888 spa_load_l2cache(spa);
3891 3889 }
3892 3890
3893 3891 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3894 3892 &nvroot) == 0);
3895 3893 if (error == 0)
3896 3894 error = spa_validate_aux(spa, nvroot, -1ULL,
3897 3895 VDEV_ALLOC_SPARE);
3898 3896 if (error == 0)
3899 3897 error = spa_validate_aux(spa, nvroot, -1ULL,
3900 3898 VDEV_ALLOC_L2CACHE);
3901 3899 spa_config_exit(spa, SCL_ALL, FTAG);
3902 3900
3903 3901 if (props != NULL)
3904 3902 spa_configfile_set(spa, props, B_FALSE);
3905 3903
3906 3904 if (error != 0 || (props && spa_writeable(spa) &&
3907 3905 (error = spa_prop_set(spa, props)))) {
3908 3906 spa_unload(spa);
3909 3907 spa_deactivate(spa);
3910 3908 spa_remove(spa);
3911 3909 mutex_exit(&spa_namespace_lock);
3912 3910 return (error);
3913 3911 }
3914 3912
3915 3913 spa_async_resume(spa);
3916 3914
3917 3915 /*
3918 3916 * Override any spares and level 2 cache devices as specified by
3919 3917 * the user, as these may have correct device names/devids, etc.
3920 3918 */
3921 3919 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3922 3920 &spares, &nspares) == 0) {
3923 3921 if (spa->spa_spares.sav_config)
3924 3922 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3925 3923 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3926 3924 else
3927 3925 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3928 3926 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3929 3927 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3930 3928 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3931 3929 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3932 3930 spa_load_spares(spa);
3933 3931 spa_config_exit(spa, SCL_ALL, FTAG);
3934 3932 spa->spa_spares.sav_sync = B_TRUE;
3935 3933 }
3936 3934 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3937 3935 &l2cache, &nl2cache) == 0) {
3938 3936 if (spa->spa_l2cache.sav_config)
3939 3937 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3940 3938 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3941 3939 else
3942 3940 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3943 3941 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3944 3942 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3945 3943 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3946 3944 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3947 3945 spa_load_l2cache(spa);
3948 3946 spa_config_exit(spa, SCL_ALL, FTAG);
3949 3947 spa->spa_l2cache.sav_sync = B_TRUE;
3950 3948 }
3951 3949
3952 3950 /*
3953 3951 * Check for any removed devices.
3954 3952 */
3955 3953 if (spa->spa_autoreplace) {
3956 3954 spa_aux_check_removed(&spa->spa_spares);
3957 3955 spa_aux_check_removed(&spa->spa_l2cache);
3958 3956 }
3959 3957
3960 3958 if (spa_writeable(spa)) {
3961 3959 /*
3962 3960 * Update the config cache to include the newly-imported pool.
3963 3961 */
3964 3962 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3965 3963 }
3966 3964
3967 3965 /*
3968 3966 * It's possible that the pool was expanded while it was exported.
3969 3967 * We kick off an async task to handle this for us.
3970 3968 */
3971 3969 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
3972 3970
3973 3971 mutex_exit(&spa_namespace_lock);
3974 3972 spa_history_log_version(spa, "import");
3975 3973
3976 3974 return (0);
3977 3975 }
3978 3976
3979 3977 nvlist_t *
3980 3978 spa_tryimport(nvlist_t *tryconfig)
3981 3979 {
3982 3980 nvlist_t *config = NULL;
3983 3981 char *poolname;
3984 3982 spa_t *spa;
3985 3983 uint64_t state;
3986 3984 int error;
3987 3985
3988 3986 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
3989 3987 return (NULL);
3990 3988
3991 3989 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
3992 3990 return (NULL);
3993 3991
3994 3992 /*
3995 3993 * Create and initialize the spa structure.
3996 3994 */
3997 3995 mutex_enter(&spa_namespace_lock);
3998 3996 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
3999 3997 spa_activate(spa, FREAD);
4000 3998
4001 3999 /*
4002 4000 * Pass off the heavy lifting to spa_load().
4003 4001 * Pass TRUE for mosconfig because the user-supplied config
4004 4002 * is actually the one to trust when doing an import.
4005 4003 */
4006 4004 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
4007 4005
4008 4006 /*
4009 4007 * If 'tryconfig' was at least parsable, return the current config.
4010 4008 */
4011 4009 if (spa->spa_root_vdev != NULL) {
4012 4010 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4013 4011 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4014 4012 poolname) == 0);
4015 4013 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4016 4014 state) == 0);
4017 4015 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4018 4016 spa->spa_uberblock.ub_timestamp) == 0);
4019 4017 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4020 4018 spa->spa_load_info) == 0);
4021 4019
4022 4020 /*
4023 4021 * If the bootfs property exists on this pool then we
4024 4022 * copy it out so that external consumers can tell which
4025 4023 * pools are bootable.
4026 4024 */
4027 4025 if ((!error || error == EEXIST) && spa->spa_bootfs) {
4028 4026 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4029 4027
4030 4028 /*
4031 4029 * We have to play games with the name since the
4032 4030 * pool was opened as TRYIMPORT_NAME.
4033 4031 */
4034 4032 if (dsl_dsobj_to_dsname(spa_name(spa),
4035 4033 spa->spa_bootfs, tmpname) == 0) {
4036 4034 char *cp;
4037 4035 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4038 4036
4039 4037 cp = strchr(tmpname, '/');
4040 4038 if (cp == NULL) {
4041 4039 (void) strlcpy(dsname, tmpname,
4042 4040 MAXPATHLEN);
4043 4041 } else {
4044 4042 (void) snprintf(dsname, MAXPATHLEN,
4045 4043 "%s/%s", poolname, ++cp);
4046 4044 }
4047 4045 VERIFY(nvlist_add_string(config,
4048 4046 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4049 4047 kmem_free(dsname, MAXPATHLEN);
4050 4048 }
4051 4049 kmem_free(tmpname, MAXPATHLEN);
4052 4050 }
4053 4051
4054 4052 /*
4055 4053 * Add the list of hot spares and level 2 cache devices.
4056 4054 */
4057 4055 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4058 4056 spa_add_spares(spa, config);
4059 4057 spa_add_l2cache(spa, config);
4060 4058 spa_config_exit(spa, SCL_CONFIG, FTAG);
4061 4059 }
4062 4060
4063 4061 spa_unload(spa);
4064 4062 spa_deactivate(spa);
4065 4063 spa_remove(spa);
4066 4064 mutex_exit(&spa_namespace_lock);
4067 4065
4068 4066 return (config);
4069 4067 }
4070 4068
4071 4069 /*
4072 4070 * Pool export/destroy
4073 4071 *
4074 4072 * The act of destroying or exporting a pool is very simple. We make sure there
4075 4073 * is no more pending I/O and any references to the pool are gone. Then, we
4076 4074 * update the pool state and sync all the labels to disk, removing the
4077 4075 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4078 4076 * we don't sync the labels or remove the configuration cache.
4079 4077 */
4080 4078 static int
4081 4079 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4082 4080 boolean_t force, boolean_t hardforce)
4083 4081 {
4084 4082 spa_t *spa;
4085 4083
4086 4084 if (oldconfig)
4087 4085 *oldconfig = NULL;
4088 4086
4089 4087 if (!(spa_mode_global & FWRITE))
4090 4088 return (SET_ERROR(EROFS));
4091 4089
4092 4090 mutex_enter(&spa_namespace_lock);
4093 4091 if ((spa = spa_lookup(pool)) == NULL) {
4094 4092 mutex_exit(&spa_namespace_lock);
4095 4093 return (SET_ERROR(ENOENT));
4096 4094 }
4097 4095
4098 4096 /*
4099 4097 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4100 4098 * reacquire the namespace lock, and see if we can export.
4101 4099 */
4102 4100 spa_open_ref(spa, FTAG);
4103 4101 mutex_exit(&spa_namespace_lock);
4104 4102 spa_async_suspend(spa);
4105 4103 mutex_enter(&spa_namespace_lock);
4106 4104 spa_close(spa, FTAG);
4107 4105
4108 4106 /*
4109 4107 * The pool will be in core if it's openable,
4110 4108 * in which case we can modify its state.
4111 4109 */
4112 4110 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4113 4111 /*
4114 4112 * Objsets may be open only because they're dirty, so we
4115 4113 * have to force it to sync before checking spa_refcnt.
4116 4114 */
4117 4115 txg_wait_synced(spa->spa_dsl_pool, 0);
4118 4116
4119 4117 /*
4120 4118 * A pool cannot be exported or destroyed if there are active
4121 4119 * references. If we are resetting a pool, allow references by
4122 4120 * fault injection handlers.
4123 4121 */
4124 4122 if (!spa_refcount_zero(spa) ||
4125 4123 (spa->spa_inject_ref != 0 &&
4126 4124 new_state != POOL_STATE_UNINITIALIZED)) {
4127 4125 spa_async_resume(spa);
4128 4126 mutex_exit(&spa_namespace_lock);
4129 4127 return (SET_ERROR(EBUSY));
4130 4128 }
4131 4129
4132 4130 /*
4133 4131 * A pool cannot be exported if it has an active shared spare.
4134 4132 * This is to prevent other pools stealing the active spare
4135 4133 * from an exported pool. At user's own will, such pool can
4136 4134 * be forcedly exported.
4137 4135 */
4138 4136 if (!force && new_state == POOL_STATE_EXPORTED &&
4139 4137 spa_has_active_shared_spare(spa)) {
4140 4138 spa_async_resume(spa);
4141 4139 mutex_exit(&spa_namespace_lock);
4142 4140 return (SET_ERROR(EXDEV));
4143 4141 }
4144 4142
4145 4143 /*
4146 4144 * We want this to be reflected on every label,
4147 4145 * so mark them all dirty. spa_unload() will do the
4148 4146 * final sync that pushes these changes out.
4149 4147 */
4150 4148 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4151 4149 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4152 4150 spa->spa_state = new_state;
4153 4151 spa->spa_final_txg = spa_last_synced_txg(spa) +
4154 4152 TXG_DEFER_SIZE + 1;
4155 4153 vdev_config_dirty(spa->spa_root_vdev);
4156 4154 spa_config_exit(spa, SCL_ALL, FTAG);
4157 4155 }
4158 4156 }
4159 4157
4160 4158 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4161 4159
4162 4160 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4163 4161 spa_unload(spa);
4164 4162 spa_deactivate(spa);
4165 4163 }
4166 4164
4167 4165 if (oldconfig && spa->spa_config)
4168 4166 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4169 4167
4170 4168 if (new_state != POOL_STATE_UNINITIALIZED) {
4171 4169 if (!hardforce)
4172 4170 spa_config_sync(spa, B_TRUE, B_TRUE);
4173 4171 spa_remove(spa);
4174 4172 }
4175 4173 mutex_exit(&spa_namespace_lock);
4176 4174
4177 4175 return (0);
4178 4176 }
4179 4177
4180 4178 /*
4181 4179 * Destroy a storage pool.
4182 4180 */
4183 4181 int
4184 4182 spa_destroy(char *pool)
4185 4183 {
4186 4184 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4187 4185 B_FALSE, B_FALSE));
4188 4186 }
4189 4187
4190 4188 /*
4191 4189 * Export a storage pool.
4192 4190 */
4193 4191 int
4194 4192 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4195 4193 boolean_t hardforce)
4196 4194 {
4197 4195 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4198 4196 force, hardforce));
4199 4197 }
4200 4198
4201 4199 /*
4202 4200 * Similar to spa_export(), this unloads the spa_t without actually removing it
4203 4201 * from the namespace in any way.
4204 4202 */
4205 4203 int
4206 4204 spa_reset(char *pool)
4207 4205 {
4208 4206 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4209 4207 B_FALSE, B_FALSE));
4210 4208 }
4211 4209
4212 4210 /*
4213 4211 * ==========================================================================
4214 4212 * Device manipulation
4215 4213 * ==========================================================================
4216 4214 */
4217 4215
4218 4216 /*
4219 4217 * Add a device to a storage pool.
4220 4218 */
4221 4219 int
4222 4220 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4223 4221 {
4224 4222 uint64_t txg, id;
4225 4223 int error;
4226 4224 vdev_t *rvd = spa->spa_root_vdev;
4227 4225 vdev_t *vd, *tvd;
4228 4226 nvlist_t **spares, **l2cache;
4229 4227 uint_t nspares, nl2cache;
4230 4228
4231 4229 ASSERT(spa_writeable(spa));
4232 4230
4233 4231 txg = spa_vdev_enter(spa);
4234 4232
4235 4233 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4236 4234 VDEV_ALLOC_ADD)) != 0)
4237 4235 return (spa_vdev_exit(spa, NULL, txg, error));
4238 4236
4239 4237 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
4240 4238
4241 4239 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4242 4240 &nspares) != 0)
4243 4241 nspares = 0;
4244 4242
4245 4243 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4246 4244 &nl2cache) != 0)
4247 4245 nl2cache = 0;
4248 4246
4249 4247 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4250 4248 return (spa_vdev_exit(spa, vd, txg, EINVAL));
4251 4249
4252 4250 if (vd->vdev_children != 0 &&
4253 4251 (error = vdev_create(vd, txg, B_FALSE)) != 0)
4254 4252 return (spa_vdev_exit(spa, vd, txg, error));
4255 4253
4256 4254 /*
4257 4255 * We must validate the spares and l2cache devices after checking the
4258 4256 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4259 4257 */
4260 4258 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4261 4259 return (spa_vdev_exit(spa, vd, txg, error));
4262 4260
4263 4261 /*
4264 4262 * Transfer each new top-level vdev from vd to rvd.
4265 4263 */
4266 4264 for (int c = 0; c < vd->vdev_children; c++) {
4267 4265
4268 4266 /*
4269 4267 * Set the vdev id to the first hole, if one exists.
4270 4268 */
4271 4269 for (id = 0; id < rvd->vdev_children; id++) {
4272 4270 if (rvd->vdev_child[id]->vdev_ishole) {
4273 4271 vdev_free(rvd->vdev_child[id]);
4274 4272 break;
4275 4273 }
4276 4274 }
4277 4275 tvd = vd->vdev_child[c];
4278 4276 vdev_remove_child(vd, tvd);
4279 4277 tvd->vdev_id = id;
4280 4278 vdev_add_child(rvd, tvd);
4281 4279 vdev_config_dirty(tvd);
4282 4280 }
4283 4281
4284 4282 if (nspares != 0) {
4285 4283 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4286 4284 ZPOOL_CONFIG_SPARES);
4287 4285 spa_load_spares(spa);
4288 4286 spa->spa_spares.sav_sync = B_TRUE;
4289 4287 }
4290 4288
4291 4289 if (nl2cache != 0) {
4292 4290 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4293 4291 ZPOOL_CONFIG_L2CACHE);
4294 4292 spa_load_l2cache(spa);
4295 4293 spa->spa_l2cache.sav_sync = B_TRUE;
4296 4294 }
4297 4295
4298 4296 /*
4299 4297 * We have to be careful when adding new vdevs to an existing pool.
4300 4298 * If other threads start allocating from these vdevs before we
4301 4299 * sync the config cache, and we lose power, then upon reboot we may
4302 4300 * fail to open the pool because there are DVAs that the config cache
4303 4301 * can't translate. Therefore, we first add the vdevs without
4304 4302 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4305 4303 * and then let spa_config_update() initialize the new metaslabs.
4306 4304 *
4307 4305 * spa_load() checks for added-but-not-initialized vdevs, so that
4308 4306 * if we lose power at any point in this sequence, the remaining
4309 4307 * steps will be completed the next time we load the pool.
4310 4308 */
4311 4309 (void) spa_vdev_exit(spa, vd, txg, 0);
4312 4310
4313 4311 mutex_enter(&spa_namespace_lock);
4314 4312 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4315 4313 mutex_exit(&spa_namespace_lock);
4316 4314
4317 4315 return (0);
4318 4316 }
4319 4317
4320 4318 /*
4321 4319 * Attach a device to a mirror. The arguments are the path to any device
4322 4320 * in the mirror, and the nvroot for the new device. If the path specifies
4323 4321 * a device that is not mirrored, we automatically insert the mirror vdev.
4324 4322 *
4325 4323 * If 'replacing' is specified, the new device is intended to replace the
4326 4324 * existing device; in this case the two devices are made into their own
4327 4325 * mirror using the 'replacing' vdev, which is functionally identical to
4328 4326 * the mirror vdev (it actually reuses all the same ops) but has a few
4329 4327 * extra rules: you can't attach to it after it's been created, and upon
4330 4328 * completion of resilvering, the first disk (the one being replaced)
4331 4329 * is automatically detached.
4332 4330 */
4333 4331 int
4334 4332 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4335 4333 {
4336 4334 uint64_t txg, dtl_max_txg;
4337 4335 vdev_t *rvd = spa->spa_root_vdev;
4338 4336 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4339 4337 vdev_ops_t *pvops;
4340 4338 char *oldvdpath, *newvdpath;
4341 4339 int newvd_isspare;
4342 4340 int error;
4343 4341
4344 4342 ASSERT(spa_writeable(spa));
4345 4343
4346 4344 txg = spa_vdev_enter(spa);
4347 4345
4348 4346 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4349 4347
4350 4348 if (oldvd == NULL)
4351 4349 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4352 4350
4353 4351 if (!oldvd->vdev_ops->vdev_op_leaf)
4354 4352 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4355 4353
4356 4354 pvd = oldvd->vdev_parent;
4357 4355
4358 4356 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4359 4357 VDEV_ALLOC_ATTACH)) != 0)
4360 4358 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4361 4359
4362 4360 if (newrootvd->vdev_children != 1)
4363 4361 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4364 4362
4365 4363 newvd = newrootvd->vdev_child[0];
4366 4364
4367 4365 if (!newvd->vdev_ops->vdev_op_leaf)
4368 4366 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4369 4367
4370 4368 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4371 4369 return (spa_vdev_exit(spa, newrootvd, txg, error));
4372 4370
4373 4371 /*
4374 4372 * Spares can't replace logs
4375 4373 */
4376 4374 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4377 4375 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4378 4376
4379 4377 if (!replacing) {
4380 4378 /*
4381 4379 * For attach, the only allowable parent is a mirror or the root
4382 4380 * vdev.
4383 4381 */
4384 4382 if (pvd->vdev_ops != &vdev_mirror_ops &&
4385 4383 pvd->vdev_ops != &vdev_root_ops)
4386 4384 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4387 4385
4388 4386 pvops = &vdev_mirror_ops;
4389 4387 } else {
4390 4388 /*
4391 4389 * Active hot spares can only be replaced by inactive hot
4392 4390 * spares.
4393 4391 */
4394 4392 if (pvd->vdev_ops == &vdev_spare_ops &&
4395 4393 oldvd->vdev_isspare &&
4396 4394 !spa_has_spare(spa, newvd->vdev_guid))
4397 4395 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4398 4396
4399 4397 /*
4400 4398 * If the source is a hot spare, and the parent isn't already a
4401 4399 * spare, then we want to create a new hot spare. Otherwise, we
4402 4400 * want to create a replacing vdev. The user is not allowed to
4403 4401 * attach to a spared vdev child unless the 'isspare' state is
4404 4402 * the same (spare replaces spare, non-spare replaces
4405 4403 * non-spare).
4406 4404 */
4407 4405 if (pvd->vdev_ops == &vdev_replacing_ops &&
4408 4406 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4409 4407 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4410 4408 } else if (pvd->vdev_ops == &vdev_spare_ops &&
4411 4409 newvd->vdev_isspare != oldvd->vdev_isspare) {
4412 4410 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4413 4411 }
4414 4412
4415 4413 if (newvd->vdev_isspare)
4416 4414 pvops = &vdev_spare_ops;
4417 4415 else
4418 4416 pvops = &vdev_replacing_ops;
4419 4417 }
4420 4418
4421 4419 /*
4422 4420 * Make sure the new device is big enough.
4423 4421 */
4424 4422 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4425 4423 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4426 4424
4427 4425 /*
4428 4426 * The new device cannot have a higher alignment requirement
4429 4427 * than the top-level vdev.
4430 4428 */
4431 4429 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4432 4430 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4433 4431
4434 4432 /*
4435 4433 * If this is an in-place replacement, update oldvd's path and devid
4436 4434 * to make it distinguishable from newvd, and unopenable from now on.
4437 4435 */
4438 4436 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4439 4437 spa_strfree(oldvd->vdev_path);
4440 4438 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4441 4439 KM_SLEEP);
4442 4440 (void) sprintf(oldvd->vdev_path, "%s/%s",
4443 4441 newvd->vdev_path, "old");
4444 4442 if (oldvd->vdev_devid != NULL) {
4445 4443 spa_strfree(oldvd->vdev_devid);
4446 4444 oldvd->vdev_devid = NULL;
4447 4445 }
4448 4446 }
4449 4447
4450 4448 /* mark the device being resilvered */
4451 4449 newvd->vdev_resilver_txg = txg;
4452 4450
4453 4451 /*
4454 4452 * If the parent is not a mirror, or if we're replacing, insert the new
4455 4453 * mirror/replacing/spare vdev above oldvd.
4456 4454 */
4457 4455 if (pvd->vdev_ops != pvops)
4458 4456 pvd = vdev_add_parent(oldvd, pvops);
4459 4457
4460 4458 ASSERT(pvd->vdev_top->vdev_parent == rvd);
4461 4459 ASSERT(pvd->vdev_ops == pvops);
4462 4460 ASSERT(oldvd->vdev_parent == pvd);
4463 4461
4464 4462 /*
4465 4463 * Extract the new device from its root and add it to pvd.
4466 4464 */
4467 4465 vdev_remove_child(newrootvd, newvd);
4468 4466 newvd->vdev_id = pvd->vdev_children;
4469 4467 newvd->vdev_crtxg = oldvd->vdev_crtxg;
4470 4468 vdev_add_child(pvd, newvd);
4471 4469
4472 4470 tvd = newvd->vdev_top;
4473 4471 ASSERT(pvd->vdev_top == tvd);
4474 4472 ASSERT(tvd->vdev_parent == rvd);
4475 4473
4476 4474 vdev_config_dirty(tvd);
4477 4475
4478 4476 /*
4479 4477 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4480 4478 * for any dmu_sync-ed blocks. It will propagate upward when
4481 4479 * spa_vdev_exit() calls vdev_dtl_reassess().
4482 4480 */
4483 4481 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4484 4482
4485 4483 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4486 4484 dtl_max_txg - TXG_INITIAL);
4487 4485
4488 4486 if (newvd->vdev_isspare) {
4489 4487 spa_spare_activate(newvd);
4490 4488 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4491 4489 }
4492 4490
4493 4491 oldvdpath = spa_strdup(oldvd->vdev_path);
4494 4492 newvdpath = spa_strdup(newvd->vdev_path);
4495 4493 newvd_isspare = newvd->vdev_isspare;
4496 4494
4497 4495 /*
4498 4496 * Mark newvd's DTL dirty in this txg.
4499 4497 */
4500 4498 vdev_dirty(tvd, VDD_DTL, newvd, txg);
4501 4499
4502 4500 /*
4503 4501 * Schedule the resilver to restart in the future. We do this to
4504 4502 * ensure that dmu_sync-ed blocks have been stitched into the
4505 4503 * respective datasets.
4506 4504 */
4507 4505 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4508 4506
4509 4507 /*
4510 4508 * Commit the config
4511 4509 */
4512 4510 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4513 4511
4514 4512 spa_history_log_internal(spa, "vdev attach", NULL,
4515 4513 "%s vdev=%s %s vdev=%s",
4516 4514 replacing && newvd_isspare ? "spare in" :
4517 4515 replacing ? "replace" : "attach", newvdpath,
4518 4516 replacing ? "for" : "to", oldvdpath);
4519 4517
4520 4518 spa_strfree(oldvdpath);
4521 4519 spa_strfree(newvdpath);
4522 4520
4523 4521 if (spa->spa_bootfs)
4524 4522 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4525 4523
4526 4524 return (0);
4527 4525 }
4528 4526
4529 4527 /*
4530 4528 * Detach a device from a mirror or replacing vdev.
4531 4529 *
4532 4530 * If 'replace_done' is specified, only detach if the parent
4533 4531 * is a replacing vdev.
4534 4532 */
4535 4533 int
4536 4534 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4537 4535 {
4538 4536 uint64_t txg;
4539 4537 int error;
4540 4538 vdev_t *rvd = spa->spa_root_vdev;
4541 4539 vdev_t *vd, *pvd, *cvd, *tvd;
4542 4540 boolean_t unspare = B_FALSE;
4543 4541 uint64_t unspare_guid = 0;
4544 4542 char *vdpath;
4545 4543
4546 4544 ASSERT(spa_writeable(spa));
4547 4545
4548 4546 txg = spa_vdev_enter(spa);
4549 4547
4550 4548 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4551 4549
4552 4550 if (vd == NULL)
4553 4551 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4554 4552
4555 4553 if (!vd->vdev_ops->vdev_op_leaf)
4556 4554 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4557 4555
4558 4556 pvd = vd->vdev_parent;
4559 4557
4560 4558 /*
4561 4559 * If the parent/child relationship is not as expected, don't do it.
4562 4560 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4563 4561 * vdev that's replacing B with C. The user's intent in replacing
4564 4562 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4565 4563 * the replace by detaching C, the expected behavior is to end up
4566 4564 * M(A,B). But suppose that right after deciding to detach C,
4567 4565 * the replacement of B completes. We would have M(A,C), and then
4568 4566 * ask to detach C, which would leave us with just A -- not what
4569 4567 * the user wanted. To prevent this, we make sure that the
4570 4568 * parent/child relationship hasn't changed -- in this example,
4571 4569 * that C's parent is still the replacing vdev R.
4572 4570 */
4573 4571 if (pvd->vdev_guid != pguid && pguid != 0)
4574 4572 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4575 4573
4576 4574 /*
4577 4575 * Only 'replacing' or 'spare' vdevs can be replaced.
4578 4576 */
4579 4577 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4580 4578 pvd->vdev_ops != &vdev_spare_ops)
4581 4579 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4582 4580
4583 4581 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4584 4582 spa_version(spa) >= SPA_VERSION_SPARES);
4585 4583
4586 4584 /*
4587 4585 * Only mirror, replacing, and spare vdevs support detach.
4588 4586 */
4589 4587 if (pvd->vdev_ops != &vdev_replacing_ops &&
4590 4588 pvd->vdev_ops != &vdev_mirror_ops &&
4591 4589 pvd->vdev_ops != &vdev_spare_ops)
4592 4590 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4593 4591
4594 4592 /*
4595 4593 * If this device has the only valid copy of some data,
4596 4594 * we cannot safely detach it.
4597 4595 */
4598 4596 if (vdev_dtl_required(vd))
4599 4597 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4600 4598
4601 4599 ASSERT(pvd->vdev_children >= 2);
4602 4600
4603 4601 /*
4604 4602 * If we are detaching the second disk from a replacing vdev, then
4605 4603 * check to see if we changed the original vdev's path to have "/old"
4606 4604 * at the end in spa_vdev_attach(). If so, undo that change now.
4607 4605 */
4608 4606 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4609 4607 vd->vdev_path != NULL) {
4610 4608 size_t len = strlen(vd->vdev_path);
4611 4609
4612 4610 for (int c = 0; c < pvd->vdev_children; c++) {
4613 4611 cvd = pvd->vdev_child[c];
4614 4612
4615 4613 if (cvd == vd || cvd->vdev_path == NULL)
4616 4614 continue;
4617 4615
4618 4616 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4619 4617 strcmp(cvd->vdev_path + len, "/old") == 0) {
4620 4618 spa_strfree(cvd->vdev_path);
4621 4619 cvd->vdev_path = spa_strdup(vd->vdev_path);
4622 4620 break;
4623 4621 }
4624 4622 }
4625 4623 }
4626 4624
4627 4625 /*
4628 4626 * If we are detaching the original disk from a spare, then it implies
4629 4627 * that the spare should become a real disk, and be removed from the
4630 4628 * active spare list for the pool.
4631 4629 */
4632 4630 if (pvd->vdev_ops == &vdev_spare_ops &&
4633 4631 vd->vdev_id == 0 &&
4634 4632 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4635 4633 unspare = B_TRUE;
4636 4634
4637 4635 /*
4638 4636 * Erase the disk labels so the disk can be used for other things.
4639 4637 * This must be done after all other error cases are handled,
4640 4638 * but before we disembowel vd (so we can still do I/O to it).
4641 4639 * But if we can't do it, don't treat the error as fatal --
4642 4640 * it may be that the unwritability of the disk is the reason
4643 4641 * it's being detached!
4644 4642 */
4645 4643 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4646 4644
4647 4645 /*
4648 4646 * Remove vd from its parent and compact the parent's children.
4649 4647 */
4650 4648 vdev_remove_child(pvd, vd);
4651 4649 vdev_compact_children(pvd);
4652 4650
4653 4651 /*
4654 4652 * Remember one of the remaining children so we can get tvd below.
4655 4653 */
4656 4654 cvd = pvd->vdev_child[pvd->vdev_children - 1];
4657 4655
4658 4656 /*
4659 4657 * If we need to remove the remaining child from the list of hot spares,
4660 4658 * do it now, marking the vdev as no longer a spare in the process.
4661 4659 * We must do this before vdev_remove_parent(), because that can
4662 4660 * change the GUID if it creates a new toplevel GUID. For a similar
4663 4661 * reason, we must remove the spare now, in the same txg as the detach;
4664 4662 * otherwise someone could attach a new sibling, change the GUID, and
4665 4663 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4666 4664 */
4667 4665 if (unspare) {
4668 4666 ASSERT(cvd->vdev_isspare);
4669 4667 spa_spare_remove(cvd);
4670 4668 unspare_guid = cvd->vdev_guid;
4671 4669 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4672 4670 cvd->vdev_unspare = B_TRUE;
4673 4671 }
4674 4672
4675 4673 /*
4676 4674 * If the parent mirror/replacing vdev only has one child,
4677 4675 * the parent is no longer needed. Remove it from the tree.
4678 4676 */
4679 4677 if (pvd->vdev_children == 1) {
4680 4678 if (pvd->vdev_ops == &vdev_spare_ops)
4681 4679 cvd->vdev_unspare = B_FALSE;
4682 4680 vdev_remove_parent(cvd);
4683 4681 }
4684 4682
4685 4683
4686 4684 /*
4687 4685 * We don't set tvd until now because the parent we just removed
4688 4686 * may have been the previous top-level vdev.
4689 4687 */
4690 4688 tvd = cvd->vdev_top;
4691 4689 ASSERT(tvd->vdev_parent == rvd);
4692 4690
4693 4691 /*
4694 4692 * Reevaluate the parent vdev state.
4695 4693 */
4696 4694 vdev_propagate_state(cvd);
4697 4695
4698 4696 /*
4699 4697 * If the 'autoexpand' property is set on the pool then automatically
4700 4698 * try to expand the size of the pool. For example if the device we
4701 4699 * just detached was smaller than the others, it may be possible to
4702 4700 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4703 4701 * first so that we can obtain the updated sizes of the leaf vdevs.
4704 4702 */
4705 4703 if (spa->spa_autoexpand) {
4706 4704 vdev_reopen(tvd);
4707 4705 vdev_expand(tvd, txg);
4708 4706 }
4709 4707
4710 4708 vdev_config_dirty(tvd);
4711 4709
4712 4710 /*
4713 4711 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
4714 4712 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4715 4713 * But first make sure we're not on any *other* txg's DTL list, to
4716 4714 * prevent vd from being accessed after it's freed.
4717 4715 */
4718 4716 vdpath = spa_strdup(vd->vdev_path);
4719 4717 for (int t = 0; t < TXG_SIZE; t++)
4720 4718 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4721 4719 vd->vdev_detached = B_TRUE;
4722 4720 vdev_dirty(tvd, VDD_DTL, vd, txg);
4723 4721
4724 4722 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4725 4723
4726 4724 /* hang on to the spa before we release the lock */
4727 4725 spa_open_ref(spa, FTAG);
4728 4726
4729 4727 error = spa_vdev_exit(spa, vd, txg, 0);
4730 4728
4731 4729 spa_history_log_internal(spa, "detach", NULL,
4732 4730 "vdev=%s", vdpath);
4733 4731 spa_strfree(vdpath);
4734 4732
4735 4733 /*
4736 4734 * If this was the removal of the original device in a hot spare vdev,
4737 4735 * then we want to go through and remove the device from the hot spare
4738 4736 * list of every other pool.
4739 4737 */
4740 4738 if (unspare) {
4741 4739 spa_t *altspa = NULL;
4742 4740
4743 4741 mutex_enter(&spa_namespace_lock);
4744 4742 while ((altspa = spa_next(altspa)) != NULL) {
4745 4743 if (altspa->spa_state != POOL_STATE_ACTIVE ||
4746 4744 altspa == spa)
4747 4745 continue;
4748 4746
4749 4747 spa_open_ref(altspa, FTAG);
4750 4748 mutex_exit(&spa_namespace_lock);
4751 4749 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4752 4750 mutex_enter(&spa_namespace_lock);
4753 4751 spa_close(altspa, FTAG);
4754 4752 }
4755 4753 mutex_exit(&spa_namespace_lock);
4756 4754
4757 4755 /* search the rest of the vdevs for spares to remove */
4758 4756 spa_vdev_resilver_done(spa);
4759 4757 }
4760 4758
4761 4759 /* all done with the spa; OK to release */
4762 4760 mutex_enter(&spa_namespace_lock);
4763 4761 spa_close(spa, FTAG);
4764 4762 mutex_exit(&spa_namespace_lock);
4765 4763
4766 4764 return (error);
4767 4765 }
4768 4766
4769 4767 /*
4770 4768 * Split a set of devices from their mirrors, and create a new pool from them.
4771 4769 */
4772 4770 int
4773 4771 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4774 4772 nvlist_t *props, boolean_t exp)
4775 4773 {
4776 4774 int error = 0;
4777 4775 uint64_t txg, *glist;
4778 4776 spa_t *newspa;
4779 4777 uint_t c, children, lastlog;
4780 4778 nvlist_t **child, *nvl, *tmp;
4781 4779 dmu_tx_t *tx;
4782 4780 char *altroot = NULL;
4783 4781 vdev_t *rvd, **vml = NULL; /* vdev modify list */
4784 4782 boolean_t activate_slog;
4785 4783
4786 4784 ASSERT(spa_writeable(spa));
4787 4785
4788 4786 txg = spa_vdev_enter(spa);
4789 4787
4790 4788 /* clear the log and flush everything up to now */
4791 4789 activate_slog = spa_passivate_log(spa);
4792 4790 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4793 4791 error = spa_offline_log(spa);
4794 4792 txg = spa_vdev_config_enter(spa);
4795 4793
4796 4794 if (activate_slog)
4797 4795 spa_activate_log(spa);
4798 4796
4799 4797 if (error != 0)
4800 4798 return (spa_vdev_exit(spa, NULL, txg, error));
4801 4799
4802 4800 /* check new spa name before going any further */
4803 4801 if (spa_lookup(newname) != NULL)
4804 4802 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4805 4803
4806 4804 /*
4807 4805 * scan through all the children to ensure they're all mirrors
4808 4806 */
4809 4807 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4810 4808 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4811 4809 &children) != 0)
4812 4810 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4813 4811
4814 4812 /* first, check to ensure we've got the right child count */
4815 4813 rvd = spa->spa_root_vdev;
4816 4814 lastlog = 0;
4817 4815 for (c = 0; c < rvd->vdev_children; c++) {
4818 4816 vdev_t *vd = rvd->vdev_child[c];
4819 4817
4820 4818 /* don't count the holes & logs as children */
4821 4819 if (vd->vdev_islog || vd->vdev_ishole) {
4822 4820 if (lastlog == 0)
4823 4821 lastlog = c;
4824 4822 continue;
4825 4823 }
4826 4824
4827 4825 lastlog = 0;
4828 4826 }
4829 4827 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4830 4828 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4831 4829
4832 4830 /* next, ensure no spare or cache devices are part of the split */
4833 4831 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4834 4832 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4835 4833 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4836 4834
4837 4835 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4838 4836 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4839 4837
4840 4838 /* then, loop over each vdev and validate it */
4841 4839 for (c = 0; c < children; c++) {
4842 4840 uint64_t is_hole = 0;
4843 4841
4844 4842 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4845 4843 &is_hole);
4846 4844
4847 4845 if (is_hole != 0) {
4848 4846 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4849 4847 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4850 4848 continue;
4851 4849 } else {
4852 4850 error = SET_ERROR(EINVAL);
4853 4851 break;
4854 4852 }
4855 4853 }
4856 4854
4857 4855 /* which disk is going to be split? */
4858 4856 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4859 4857 &glist[c]) != 0) {
4860 4858 error = SET_ERROR(EINVAL);
4861 4859 break;
4862 4860 }
4863 4861
4864 4862 /* look it up in the spa */
4865 4863 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4866 4864 if (vml[c] == NULL) {
4867 4865 error = SET_ERROR(ENODEV);
4868 4866 break;
4869 4867 }
4870 4868
4871 4869 /* make sure there's nothing stopping the split */
4872 4870 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4873 4871 vml[c]->vdev_islog ||
4874 4872 vml[c]->vdev_ishole ||
4875 4873 vml[c]->vdev_isspare ||
4876 4874 vml[c]->vdev_isl2cache ||
4877 4875 !vdev_writeable(vml[c]) ||
4878 4876 vml[c]->vdev_children != 0 ||
4879 4877 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4880 4878 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4881 4879 error = SET_ERROR(EINVAL);
4882 4880 break;
4883 4881 }
4884 4882
4885 4883 if (vdev_dtl_required(vml[c])) {
4886 4884 error = SET_ERROR(EBUSY);
4887 4885 break;
4888 4886 }
4889 4887
4890 4888 /* we need certain info from the top level */
4891 4889 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4892 4890 vml[c]->vdev_top->vdev_ms_array) == 0);
4893 4891 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4894 4892 vml[c]->vdev_top->vdev_ms_shift) == 0);
4895 4893 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4896 4894 vml[c]->vdev_top->vdev_asize) == 0);
4897 4895 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4898 4896 vml[c]->vdev_top->vdev_ashift) == 0);
4899 4897 }
4900 4898
4901 4899 if (error != 0) {
4902 4900 kmem_free(vml, children * sizeof (vdev_t *));
4903 4901 kmem_free(glist, children * sizeof (uint64_t));
4904 4902 return (spa_vdev_exit(spa, NULL, txg, error));
4905 4903 }
4906 4904
4907 4905 /* stop writers from using the disks */
4908 4906 for (c = 0; c < children; c++) {
4909 4907 if (vml[c] != NULL)
4910 4908 vml[c]->vdev_offline = B_TRUE;
4911 4909 }
4912 4910 vdev_reopen(spa->spa_root_vdev);
4913 4911
4914 4912 /*
4915 4913 * Temporarily record the splitting vdevs in the spa config. This
4916 4914 * will disappear once the config is regenerated.
4917 4915 */
4918 4916 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4919 4917 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4920 4918 glist, children) == 0);
4921 4919 kmem_free(glist, children * sizeof (uint64_t));
4922 4920
4923 4921 mutex_enter(&spa->spa_props_lock);
4924 4922 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4925 4923 nvl) == 0);
4926 4924 mutex_exit(&spa->spa_props_lock);
4927 4925 spa->spa_config_splitting = nvl;
4928 4926 vdev_config_dirty(spa->spa_root_vdev);
4929 4927
4930 4928 /* configure and create the new pool */
4931 4929 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4932 4930 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4933 4931 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4934 4932 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4935 4933 spa_version(spa)) == 0);
4936 4934 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4937 4935 spa->spa_config_txg) == 0);
4938 4936 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4939 4937 spa_generate_guid(NULL)) == 0);
4940 4938 (void) nvlist_lookup_string(props,
4941 4939 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4942 4940
4943 4941 /* add the new pool to the namespace */
4944 4942 newspa = spa_add(newname, config, altroot);
4945 4943 newspa->spa_config_txg = spa->spa_config_txg;
4946 4944 spa_set_log_state(newspa, SPA_LOG_CLEAR);
4947 4945
4948 4946 /* release the spa config lock, retaining the namespace lock */
4949 4947 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4950 4948
4951 4949 if (zio_injection_enabled)
4952 4950 zio_handle_panic_injection(spa, FTAG, 1);
4953 4951
4954 4952 spa_activate(newspa, spa_mode_global);
4955 4953 spa_async_suspend(newspa);
4956 4954
4957 4955 /* create the new pool from the disks of the original pool */
4958 4956 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
4959 4957 if (error)
4960 4958 goto out;
4961 4959
4962 4960 /* if that worked, generate a real config for the new pool */
4963 4961 if (newspa->spa_root_vdev != NULL) {
4964 4962 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
4965 4963 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4966 4964 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
4967 4965 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
4968 4966 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
4969 4967 B_TRUE));
4970 4968 }
4971 4969
4972 4970 /* set the props */
4973 4971 if (props != NULL) {
4974 4972 spa_configfile_set(newspa, props, B_FALSE);
4975 4973 error = spa_prop_set(newspa, props);
4976 4974 if (error)
4977 4975 goto out;
4978 4976 }
4979 4977
4980 4978 /* flush everything */
4981 4979 txg = spa_vdev_config_enter(newspa);
4982 4980 vdev_config_dirty(newspa->spa_root_vdev);
4983 4981 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
4984 4982
4985 4983 if (zio_injection_enabled)
4986 4984 zio_handle_panic_injection(spa, FTAG, 2);
4987 4985
4988 4986 spa_async_resume(newspa);
4989 4987
4990 4988 /* finally, update the original pool's config */
4991 4989 txg = spa_vdev_config_enter(spa);
4992 4990 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
4993 4991 error = dmu_tx_assign(tx, TXG_WAIT);
4994 4992 if (error != 0)
4995 4993 dmu_tx_abort(tx);
4996 4994 for (c = 0; c < children; c++) {
4997 4995 if (vml[c] != NULL) {
4998 4996 vdev_split(vml[c]);
4999 4997 if (error == 0)
5000 4998 spa_history_log_internal(spa, "detach", tx,
5001 4999 "vdev=%s", vml[c]->vdev_path);
5002 5000 vdev_free(vml[c]);
5003 5001 }
5004 5002 }
5005 5003 vdev_config_dirty(spa->spa_root_vdev);
5006 5004 spa->spa_config_splitting = NULL;
5007 5005 nvlist_free(nvl);
5008 5006 if (error == 0)
5009 5007 dmu_tx_commit(tx);
5010 5008 (void) spa_vdev_exit(spa, NULL, txg, 0);
5011 5009
5012 5010 if (zio_injection_enabled)
5013 5011 zio_handle_panic_injection(spa, FTAG, 3);
5014 5012
5015 5013 /* split is complete; log a history record */
5016 5014 spa_history_log_internal(newspa, "split", NULL,
5017 5015 "from pool %s", spa_name(spa));
5018 5016
5019 5017 kmem_free(vml, children * sizeof (vdev_t *));
5020 5018
5021 5019 /* if we're not going to mount the filesystems in userland, export */
5022 5020 if (exp)
5023 5021 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5024 5022 B_FALSE, B_FALSE);
5025 5023
5026 5024 return (error);
5027 5025
5028 5026 out:
5029 5027 spa_unload(newspa);
5030 5028 spa_deactivate(newspa);
5031 5029 spa_remove(newspa);
5032 5030
5033 5031 txg = spa_vdev_config_enter(spa);
5034 5032
5035 5033 /* re-online all offlined disks */
5036 5034 for (c = 0; c < children; c++) {
5037 5035 if (vml[c] != NULL)
5038 5036 vml[c]->vdev_offline = B_FALSE;
5039 5037 }
5040 5038 vdev_reopen(spa->spa_root_vdev);
5041 5039
5042 5040 nvlist_free(spa->spa_config_splitting);
5043 5041 spa->spa_config_splitting = NULL;
5044 5042 (void) spa_vdev_exit(spa, NULL, txg, error);
5045 5043
5046 5044 kmem_free(vml, children * sizeof (vdev_t *));
5047 5045 return (error);
5048 5046 }
5049 5047
5050 5048 static nvlist_t *
5051 5049 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5052 5050 {
5053 5051 for (int i = 0; i < count; i++) {
5054 5052 uint64_t guid;
5055 5053
5056 5054 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5057 5055 &guid) == 0);
5058 5056
5059 5057 if (guid == target_guid)
5060 5058 return (nvpp[i]);
5061 5059 }
5062 5060
5063 5061 return (NULL);
5064 5062 }
5065 5063
5066 5064 static void
5067 5065 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5068 5066 nvlist_t *dev_to_remove)
5069 5067 {
5070 5068 nvlist_t **newdev = NULL;
5071 5069
5072 5070 if (count > 1)
5073 5071 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
5074 5072
5075 5073 for (int i = 0, j = 0; i < count; i++) {
5076 5074 if (dev[i] == dev_to_remove)
5077 5075 continue;
5078 5076 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
5079 5077 }
5080 5078
5081 5079 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5082 5080 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5083 5081
5084 5082 for (int i = 0; i < count - 1; i++)
5085 5083 nvlist_free(newdev[i]);
5086 5084
5087 5085 if (count > 1)
5088 5086 kmem_free(newdev, (count - 1) * sizeof (void *));
5089 5087 }
5090 5088
5091 5089 /*
5092 5090 * Evacuate the device.
5093 5091 */
5094 5092 static int
5095 5093 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5096 5094 {
5097 5095 uint64_t txg;
5098 5096 int error = 0;
5099 5097
5100 5098 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5101 5099 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5102 5100 ASSERT(vd == vd->vdev_top);
5103 5101
5104 5102 /*
5105 5103 * Evacuate the device. We don't hold the config lock as writer
5106 5104 * since we need to do I/O but we do keep the
5107 5105 * spa_namespace_lock held. Once this completes the device
5108 5106 * should no longer have any blocks allocated on it.
5109 5107 */
5110 5108 if (vd->vdev_islog) {
5111 5109 if (vd->vdev_stat.vs_alloc != 0)
5112 5110 error = spa_offline_log(spa);
5113 5111 } else {
5114 5112 error = SET_ERROR(ENOTSUP);
5115 5113 }
5116 5114
5117 5115 if (error)
5118 5116 return (error);
5119 5117
5120 5118 /*
5121 5119 * The evacuation succeeded. Remove any remaining MOS metadata
5122 5120 * associated with this vdev, and wait for these changes to sync.
5123 5121 */
5124 5122 ASSERT0(vd->vdev_stat.vs_alloc);
5125 5123 txg = spa_vdev_config_enter(spa);
5126 5124 vd->vdev_removing = B_TRUE;
5127 5125 vdev_dirty_leaves(vd, VDD_DTL, txg);
5128 5126 vdev_config_dirty(vd);
5129 5127 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5130 5128
5131 5129 return (0);
5132 5130 }
5133 5131
5134 5132 /*
5135 5133 * Complete the removal by cleaning up the namespace.
5136 5134 */
5137 5135 static void
5138 5136 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5139 5137 {
5140 5138 vdev_t *rvd = spa->spa_root_vdev;
5141 5139 uint64_t id = vd->vdev_id;
5142 5140 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5143 5141
5144 5142 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5145 5143 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5146 5144 ASSERT(vd == vd->vdev_top);
5147 5145
5148 5146 /*
5149 5147 * Only remove any devices which are empty.
5150 5148 */
5151 5149 if (vd->vdev_stat.vs_alloc != 0)
5152 5150 return;
5153 5151
5154 5152 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5155 5153
5156 5154 if (list_link_active(&vd->vdev_state_dirty_node))
5157 5155 vdev_state_clean(vd);
5158 5156 if (list_link_active(&vd->vdev_config_dirty_node))
5159 5157 vdev_config_clean(vd);
5160 5158
5161 5159 vdev_free(vd);
5162 5160
5163 5161 if (last_vdev) {
5164 5162 vdev_compact_children(rvd);
5165 5163 } else {
5166 5164 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5167 5165 vdev_add_child(rvd, vd);
5168 5166 }
5169 5167 vdev_config_dirty(rvd);
5170 5168
5171 5169 /*
5172 5170 * Reassess the health of our root vdev.
5173 5171 */
5174 5172 vdev_reopen(rvd);
5175 5173 }
5176 5174
5177 5175 /*
5178 5176 * Remove a device from the pool -
5179 5177 *
5180 5178 * Removing a device from the vdev namespace requires several steps
5181 5179 * and can take a significant amount of time. As a result we use
5182 5180 * the spa_vdev_config_[enter/exit] functions which allow us to
5183 5181 * grab and release the spa_config_lock while still holding the namespace
5184 5182 * lock. During each step the configuration is synced out.
5185 5183 *
5186 5184 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5187 5185 * devices.
5188 5186 */
5189 5187 int
5190 5188 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5191 5189 {
5192 5190 vdev_t *vd;
5193 5191 metaslab_group_t *mg;
5194 5192 nvlist_t **spares, **l2cache, *nv;
5195 5193 uint64_t txg = 0;
5196 5194 uint_t nspares, nl2cache;
5197 5195 int error = 0;
5198 5196 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5199 5197
5200 5198 ASSERT(spa_writeable(spa));
5201 5199
5202 5200 if (!locked)
5203 5201 txg = spa_vdev_enter(spa);
5204 5202
5205 5203 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5206 5204
5207 5205 if (spa->spa_spares.sav_vdevs != NULL &&
5208 5206 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5209 5207 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5210 5208 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5211 5209 /*
5212 5210 * Only remove the hot spare if it's not currently in use
5213 5211 * in this pool.
5214 5212 */
5215 5213 if (vd == NULL || unspare) {
5216 5214 spa_vdev_remove_aux(spa->spa_spares.sav_config,
5217 5215 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5218 5216 spa_load_spares(spa);
5219 5217 spa->spa_spares.sav_sync = B_TRUE;
5220 5218 } else {
5221 5219 error = SET_ERROR(EBUSY);
5222 5220 }
5223 5221 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
5224 5222 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5225 5223 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5226 5224 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5227 5225 /*
5228 5226 * Cache devices can always be removed.
5229 5227 */
5230 5228 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5231 5229 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5232 5230 spa_load_l2cache(spa);
5233 5231 spa->spa_l2cache.sav_sync = B_TRUE;
5234 5232 } else if (vd != NULL && vd->vdev_islog) {
5235 5233 ASSERT(!locked);
5236 5234 ASSERT(vd == vd->vdev_top);
5237 5235
5238 5236 /*
5239 5237 * XXX - Once we have bp-rewrite this should
5240 5238 * become the common case.
5241 5239 */
5242 5240
5243 5241 mg = vd->vdev_mg;
5244 5242
5245 5243 /*
5246 5244 * Stop allocating from this vdev.
5247 5245 */
5248 5246 metaslab_group_passivate(mg);
5249 5247
5250 5248 /*
5251 5249 * Wait for the youngest allocations and frees to sync,
5252 5250 * and then wait for the deferral of those frees to finish.
5253 5251 */
5254 5252 spa_vdev_config_exit(spa, NULL,
5255 5253 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5256 5254
5257 5255 /*
5258 5256 * Attempt to evacuate the vdev.
5259 5257 */
5260 5258 error = spa_vdev_remove_evacuate(spa, vd);
5261 5259
5262 5260 txg = spa_vdev_config_enter(spa);
5263 5261
5264 5262 /*
5265 5263 * If we couldn't evacuate the vdev, unwind.
5266 5264 */
5267 5265 if (error) {
5268 5266 metaslab_group_activate(mg);
5269 5267 return (spa_vdev_exit(spa, NULL, txg, error));
5270 5268 }
5271 5269
5272 5270 /*
5273 5271 * Clean up the vdev namespace.
5274 5272 */
5275 5273 spa_vdev_remove_from_namespace(spa, vd);
5276 5274
5277 5275 } else if (vd != NULL) {
5278 5276 /*
5279 5277 * Normal vdevs cannot be removed (yet).
5280 5278 */
5281 5279 error = SET_ERROR(ENOTSUP);
5282 5280 } else {
5283 5281 /*
5284 5282 * There is no vdev of any kind with the specified guid.
5285 5283 */
5286 5284 error = SET_ERROR(ENOENT);
5287 5285 }
5288 5286
5289 5287 if (!locked)
5290 5288 return (spa_vdev_exit(spa, NULL, txg, error));
5291 5289
5292 5290 return (error);
5293 5291 }
5294 5292
5295 5293 /*
5296 5294 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5297 5295 * currently spared, so we can detach it.
5298 5296 */
5299 5297 static vdev_t *
5300 5298 spa_vdev_resilver_done_hunt(vdev_t *vd)
5301 5299 {
5302 5300 vdev_t *newvd, *oldvd;
5303 5301
5304 5302 for (int c = 0; c < vd->vdev_children; c++) {
5305 5303 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5306 5304 if (oldvd != NULL)
5307 5305 return (oldvd);
5308 5306 }
5309 5307
5310 5308 /*
5311 5309 * Check for a completed replacement. We always consider the first
5312 5310 * vdev in the list to be the oldest vdev, and the last one to be
5313 5311 * the newest (see spa_vdev_attach() for how that works). In
5314 5312 * the case where the newest vdev is faulted, we will not automatically
5315 5313 * remove it after a resilver completes. This is OK as it will require
5316 5314 * user intervention to determine which disk the admin wishes to keep.
5317 5315 */
5318 5316 if (vd->vdev_ops == &vdev_replacing_ops) {
5319 5317 ASSERT(vd->vdev_children > 1);
5320 5318
5321 5319 newvd = vd->vdev_child[vd->vdev_children - 1];
5322 5320 oldvd = vd->vdev_child[0];
5323 5321
5324 5322 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5325 5323 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5326 5324 !vdev_dtl_required(oldvd))
5327 5325 return (oldvd);
5328 5326 }
5329 5327
5330 5328 /*
5331 5329 * Check for a completed resilver with the 'unspare' flag set.
5332 5330 */
5333 5331 if (vd->vdev_ops == &vdev_spare_ops) {
5334 5332 vdev_t *first = vd->vdev_child[0];
5335 5333 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5336 5334
5337 5335 if (last->vdev_unspare) {
5338 5336 oldvd = first;
5339 5337 newvd = last;
5340 5338 } else if (first->vdev_unspare) {
5341 5339 oldvd = last;
5342 5340 newvd = first;
5343 5341 } else {
5344 5342 oldvd = NULL;
5345 5343 }
5346 5344
5347 5345 if (oldvd != NULL &&
5348 5346 vdev_dtl_empty(newvd, DTL_MISSING) &&
5349 5347 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5350 5348 !vdev_dtl_required(oldvd))
5351 5349 return (oldvd);
5352 5350
5353 5351 /*
5354 5352 * If there are more than two spares attached to a disk,
5355 5353 * and those spares are not required, then we want to
5356 5354 * attempt to free them up now so that they can be used
5357 5355 * by other pools. Once we're back down to a single
5358 5356 * disk+spare, we stop removing them.
5359 5357 */
5360 5358 if (vd->vdev_children > 2) {
5361 5359 newvd = vd->vdev_child[1];
5362 5360
5363 5361 if (newvd->vdev_isspare && last->vdev_isspare &&
5364 5362 vdev_dtl_empty(last, DTL_MISSING) &&
5365 5363 vdev_dtl_empty(last, DTL_OUTAGE) &&
5366 5364 !vdev_dtl_required(newvd))
5367 5365 return (newvd);
5368 5366 }
5369 5367 }
5370 5368
5371 5369 return (NULL);
5372 5370 }
5373 5371
5374 5372 static void
5375 5373 spa_vdev_resilver_done(spa_t *spa)
5376 5374 {
5377 5375 vdev_t *vd, *pvd, *ppvd;
5378 5376 uint64_t guid, sguid, pguid, ppguid;
5379 5377
5380 5378 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5381 5379
5382 5380 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5383 5381 pvd = vd->vdev_parent;
5384 5382 ppvd = pvd->vdev_parent;
5385 5383 guid = vd->vdev_guid;
5386 5384 pguid = pvd->vdev_guid;
5387 5385 ppguid = ppvd->vdev_guid;
5388 5386 sguid = 0;
5389 5387 /*
5390 5388 * If we have just finished replacing a hot spared device, then
5391 5389 * we need to detach the parent's first child (the original hot
5392 5390 * spare) as well.
5393 5391 */
5394 5392 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5395 5393 ppvd->vdev_children == 2) {
5396 5394 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5397 5395 sguid = ppvd->vdev_child[1]->vdev_guid;
5398 5396 }
5399 5397 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd));
5400 5398
5401 5399 spa_config_exit(spa, SCL_ALL, FTAG);
5402 5400 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5403 5401 return;
5404 5402 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5405 5403 return;
5406 5404 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5407 5405 }
5408 5406
5409 5407 spa_config_exit(spa, SCL_ALL, FTAG);
5410 5408 }
5411 5409
5412 5410 /*
5413 5411 * Update the stored path or FRU for this vdev.
5414 5412 */
5415 5413 int
5416 5414 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5417 5415 boolean_t ispath)
5418 5416 {
5419 5417 vdev_t *vd;
5420 5418 boolean_t sync = B_FALSE;
5421 5419
5422 5420 ASSERT(spa_writeable(spa));
5423 5421
5424 5422 spa_vdev_state_enter(spa, SCL_ALL);
5425 5423
5426 5424 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5427 5425 return (spa_vdev_state_exit(spa, NULL, ENOENT));
5428 5426
5429 5427 if (!vd->vdev_ops->vdev_op_leaf)
5430 5428 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5431 5429
5432 5430 if (ispath) {
5433 5431 if (strcmp(value, vd->vdev_path) != 0) {
5434 5432 spa_strfree(vd->vdev_path);
5435 5433 vd->vdev_path = spa_strdup(value);
5436 5434 sync = B_TRUE;
5437 5435 }
5438 5436 } else {
5439 5437 if (vd->vdev_fru == NULL) {
5440 5438 vd->vdev_fru = spa_strdup(value);
5441 5439 sync = B_TRUE;
5442 5440 } else if (strcmp(value, vd->vdev_fru) != 0) {
5443 5441 spa_strfree(vd->vdev_fru);
5444 5442 vd->vdev_fru = spa_strdup(value);
5445 5443 sync = B_TRUE;
5446 5444 }
5447 5445 }
5448 5446
5449 5447 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5450 5448 }
5451 5449
5452 5450 int
5453 5451 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5454 5452 {
5455 5453 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5456 5454 }
5457 5455
5458 5456 int
5459 5457 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5460 5458 {
5461 5459 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5462 5460 }
5463 5461
5464 5462 /*
5465 5463 * ==========================================================================
5466 5464 * SPA Scanning
5467 5465 * ==========================================================================
5468 5466 */
5469 5467
5470 5468 int
5471 5469 spa_scan_stop(spa_t *spa)
5472 5470 {
5473 5471 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5474 5472 if (dsl_scan_resilvering(spa->spa_dsl_pool))
5475 5473 return (SET_ERROR(EBUSY));
5476 5474 return (dsl_scan_cancel(spa->spa_dsl_pool));
5477 5475 }
5478 5476
5479 5477 int
5480 5478 spa_scan(spa_t *spa, pool_scan_func_t func)
5481 5479 {
5482 5480 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5483 5481
5484 5482 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5485 5483 return (SET_ERROR(ENOTSUP));
5486 5484
5487 5485 /*
5488 5486 * If a resilver was requested, but there is no DTL on a
5489 5487 * writeable leaf device, we have nothing to do.
5490 5488 */
5491 5489 if (func == POOL_SCAN_RESILVER &&
5492 5490 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5493 5491 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5494 5492 return (0);
5495 5493 }
5496 5494
5497 5495 return (dsl_scan(spa->spa_dsl_pool, func));
5498 5496 }
5499 5497
5500 5498 /*
5501 5499 * ==========================================================================
5502 5500 * SPA async task processing
5503 5501 * ==========================================================================
5504 5502 */
5505 5503
5506 5504 static void
5507 5505 spa_async_remove(spa_t *spa, vdev_t *vd)
5508 5506 {
5509 5507 if (vd->vdev_remove_wanted) {
5510 5508 vd->vdev_remove_wanted = B_FALSE;
5511 5509 vd->vdev_delayed_close = B_FALSE;
5512 5510 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5513 5511
5514 5512 /*
5515 5513 * We want to clear the stats, but we don't want to do a full
5516 5514 * vdev_clear() as that will cause us to throw away
5517 5515 * degraded/faulted state as well as attempt to reopen the
5518 5516 * device, all of which is a waste.
5519 5517 */
5520 5518 vd->vdev_stat.vs_read_errors = 0;
5521 5519 vd->vdev_stat.vs_write_errors = 0;
5522 5520 vd->vdev_stat.vs_checksum_errors = 0;
5523 5521
5524 5522 vdev_state_dirty(vd->vdev_top);
5525 5523 }
5526 5524
5527 5525 for (int c = 0; c < vd->vdev_children; c++)
5528 5526 spa_async_remove(spa, vd->vdev_child[c]);
5529 5527 }
5530 5528
5531 5529 static void
5532 5530 spa_async_probe(spa_t *spa, vdev_t *vd)
5533 5531 {
5534 5532 if (vd->vdev_probe_wanted) {
5535 5533 vd->vdev_probe_wanted = B_FALSE;
5536 5534 vdev_reopen(vd); /* vdev_open() does the actual probe */
5537 5535 }
5538 5536
5539 5537 for (int c = 0; c < vd->vdev_children; c++)
5540 5538 spa_async_probe(spa, vd->vdev_child[c]);
5541 5539 }
5542 5540
5543 5541 static void
5544 5542 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5545 5543 {
5546 5544 sysevent_id_t eid;
5547 5545 nvlist_t *attr;
5548 5546 char *physpath;
5549 5547
5550 5548 if (!spa->spa_autoexpand)
5551 5549 return;
5552 5550
5553 5551 for (int c = 0; c < vd->vdev_children; c++) {
5554 5552 vdev_t *cvd = vd->vdev_child[c];
5555 5553 spa_async_autoexpand(spa, cvd);
5556 5554 }
5557 5555
5558 5556 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5559 5557 return;
5560 5558
5561 5559 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5562 5560 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5563 5561
5564 5562 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5565 5563 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5566 5564
5567 5565 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5568 5566 ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
5569 5567
5570 5568 nvlist_free(attr);
5571 5569 kmem_free(physpath, MAXPATHLEN);
5572 5570 }
5573 5571
5574 5572 static void
5575 5573 spa_async_thread(spa_t *spa)
5576 5574 {
5577 5575 int tasks;
5578 5576
5579 5577 ASSERT(spa->spa_sync_on);
5580 5578
5581 5579 mutex_enter(&spa->spa_async_lock);
5582 5580 tasks = spa->spa_async_tasks;
5583 5581 spa->spa_async_tasks = 0;
5584 5582 mutex_exit(&spa->spa_async_lock);
5585 5583
5586 5584 /*
5587 5585 * See if the config needs to be updated.
5588 5586 */
5589 5587 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5590 5588 uint64_t old_space, new_space;
5591 5589
5592 5590 mutex_enter(&spa_namespace_lock);
5593 5591 old_space = metaslab_class_get_space(spa_normal_class(spa));
5594 5592 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5595 5593 new_space = metaslab_class_get_space(spa_normal_class(spa));
5596 5594 mutex_exit(&spa_namespace_lock);
5597 5595
5598 5596 /*
5599 5597 * If the pool grew as a result of the config update,
5600 5598 * then log an internal history event.
5601 5599 */
5602 5600 if (new_space != old_space) {
5603 5601 spa_history_log_internal(spa, "vdev online", NULL,
5604 5602 "pool '%s' size: %llu(+%llu)",
5605 5603 spa_name(spa), new_space, new_space - old_space);
5606 5604 }
5607 5605 }
5608 5606
5609 5607 /*
5610 5608 * See if any devices need to be marked REMOVED.
5611 5609 */
5612 5610 if (tasks & SPA_ASYNC_REMOVE) {
5613 5611 spa_vdev_state_enter(spa, SCL_NONE);
5614 5612 spa_async_remove(spa, spa->spa_root_vdev);
5615 5613 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5616 5614 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5617 5615 for (int i = 0; i < spa->spa_spares.sav_count; i++)
5618 5616 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5619 5617 (void) spa_vdev_state_exit(spa, NULL, 0);
5620 5618 }
5621 5619
5622 5620 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5623 5621 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5624 5622 spa_async_autoexpand(spa, spa->spa_root_vdev);
5625 5623 spa_config_exit(spa, SCL_CONFIG, FTAG);
5626 5624 }
5627 5625
5628 5626 /*
5629 5627 * See if any devices need to be probed.
5630 5628 */
5631 5629 if (tasks & SPA_ASYNC_PROBE) {
5632 5630 spa_vdev_state_enter(spa, SCL_NONE);
5633 5631 spa_async_probe(spa, spa->spa_root_vdev);
5634 5632 (void) spa_vdev_state_exit(spa, NULL, 0);
5635 5633 }
5636 5634
5637 5635 /*
5638 5636 * If any devices are done replacing, detach them.
5639 5637 */
5640 5638 if (tasks & SPA_ASYNC_RESILVER_DONE)
5641 5639 spa_vdev_resilver_done(spa);
5642 5640
5643 5641 /*
5644 5642 * Kick off a resilver.
5645 5643 */
5646 5644 if (tasks & SPA_ASYNC_RESILVER)
5647 5645 dsl_resilver_restart(spa->spa_dsl_pool, 0);
5648 5646
5649 5647 /*
5650 5648 * Let the world know that we're done.
5651 5649 */
5652 5650 mutex_enter(&spa->spa_async_lock);
5653 5651 spa->spa_async_thread = NULL;
5654 5652 cv_broadcast(&spa->spa_async_cv);
5655 5653 mutex_exit(&spa->spa_async_lock);
5656 5654 thread_exit();
5657 5655 }
5658 5656
5659 5657 void
5660 5658 spa_async_suspend(spa_t *spa)
5661 5659 {
5662 5660 mutex_enter(&spa->spa_async_lock);
5663 5661 spa->spa_async_suspended++;
5664 5662 while (spa->spa_async_thread != NULL)
5665 5663 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5666 5664 mutex_exit(&spa->spa_async_lock);
5667 5665 }
5668 5666
5669 5667 void
5670 5668 spa_async_resume(spa_t *spa)
5671 5669 {
5672 5670 mutex_enter(&spa->spa_async_lock);
5673 5671 ASSERT(spa->spa_async_suspended != 0);
5674 5672 spa->spa_async_suspended--;
5675 5673 mutex_exit(&spa->spa_async_lock);
5676 5674 }
5677 5675
5678 5676 static boolean_t
5679 5677 spa_async_tasks_pending(spa_t *spa)
5680 5678 {
5681 5679 uint_t non_config_tasks;
5682 5680 uint_t config_task;
5683 5681 boolean_t config_task_suspended;
5684 5682
5685 5683 non_config_tasks = spa->spa_async_tasks & ~SPA_ASYNC_CONFIG_UPDATE;
5686 5684 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
5687 5685 if (spa->spa_ccw_fail_time == 0) {
5688 5686 config_task_suspended = B_FALSE;
5689 5687 } else {
5690 5688 config_task_suspended =
5691 5689 (gethrtime() - spa->spa_ccw_fail_time) <
5692 5690 (zfs_ccw_retry_interval * NANOSEC);
5693 5691 }
5694 5692
5695 5693 return (non_config_tasks || (config_task && !config_task_suspended));
5696 5694 }
5697 5695
5698 5696 static void
5699 5697 spa_async_dispatch(spa_t *spa)
5700 5698 {
5701 5699 mutex_enter(&spa->spa_async_lock);
5702 5700 if (spa_async_tasks_pending(spa) &&
5703 5701 !spa->spa_async_suspended &&
5704 5702 spa->spa_async_thread == NULL &&
5705 5703 rootdir != NULL)
5706 5704 spa->spa_async_thread = thread_create(NULL, 0,
5707 5705 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5708 5706 mutex_exit(&spa->spa_async_lock);
5709 5707 }
5710 5708
5711 5709 void
5712 5710 spa_async_request(spa_t *spa, int task)
5713 5711 {
5714 5712 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5715 5713 mutex_enter(&spa->spa_async_lock);
5716 5714 spa->spa_async_tasks |= task;
5717 5715 mutex_exit(&spa->spa_async_lock);
5718 5716 }
5719 5717
5720 5718 /*
5721 5719 * ==========================================================================
5722 5720 * SPA syncing routines
5723 5721 * ==========================================================================
5724 5722 */
5725 5723
5726 5724 static int
5727 5725 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5728 5726 {
5729 5727 bpobj_t *bpo = arg;
5730 5728 bpobj_enqueue(bpo, bp, tx);
5731 5729 return (0);
5732 5730 }
5733 5731
5734 5732 static int
5735 5733 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5736 5734 {
5737 5735 zio_t *zio = arg;
5738 5736
5739 5737 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5740 5738 zio->io_flags));
5741 5739 return (0);
5742 5740 }
5743 5741
5744 5742 /*
5745 5743 * Note: this simple function is not inlined to make it easier to dtrace the
5746 5744 * amount of time spent syncing frees.
5747 5745 */
5748 5746 static void
5749 5747 spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx)
5750 5748 {
5751 5749 zio_t *zio = zio_root(spa, NULL, NULL, 0);
5752 5750 bplist_iterate(bpl, spa_free_sync_cb, zio, tx);
5753 5751 VERIFY(zio_wait(zio) == 0);
5754 5752 }
5755 5753
5756 5754 /*
5757 5755 * Note: this simple function is not inlined to make it easier to dtrace the
5758 5756 * amount of time spent syncing deferred frees.
5759 5757 */
5760 5758 static void
5761 5759 spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx)
5762 5760 {
5763 5761 zio_t *zio = zio_root(spa, NULL, NULL, 0);
5764 5762 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj,
5765 5763 spa_free_sync_cb, zio, tx), ==, 0);
5766 5764 VERIFY0(zio_wait(zio));
5767 5765 }
5768 5766
5769 5767
5770 5768 static void
5771 5769 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5772 5770 {
5773 5771 char *packed = NULL;
5774 5772 size_t bufsize;
5775 5773 size_t nvsize = 0;
5776 5774 dmu_buf_t *db;
5777 5775
5778 5776 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5779 5777
5780 5778 /*
5781 5779 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5782 5780 * information. This avoids the dbuf_will_dirty() path and
5783 5781 * saves us a pre-read to get data we don't actually care about.
5784 5782 */
5785 5783 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
5786 5784 packed = kmem_alloc(bufsize, KM_SLEEP);
5787 5785
5788 5786 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5789 5787 KM_SLEEP) == 0);
5790 5788 bzero(packed + nvsize, bufsize - nvsize);
5791 5789
5792 5790 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5793 5791
5794 5792 kmem_free(packed, bufsize);
5795 5793
5796 5794 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5797 5795 dmu_buf_will_dirty(db, tx);
5798 5796 *(uint64_t *)db->db_data = nvsize;
5799 5797 dmu_buf_rele(db, FTAG);
5800 5798 }
5801 5799
5802 5800 static void
5803 5801 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5804 5802 const char *config, const char *entry)
5805 5803 {
5806 5804 nvlist_t *nvroot;
5807 5805 nvlist_t **list;
5808 5806 int i;
5809 5807
5810 5808 if (!sav->sav_sync)
5811 5809 return;
5812 5810
5813 5811 /*
5814 5812 * Update the MOS nvlist describing the list of available devices.
5815 5813 * spa_validate_aux() will have already made sure this nvlist is
5816 5814 * valid and the vdevs are labeled appropriately.
5817 5815 */
5818 5816 if (sav->sav_object == 0) {
5819 5817 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5820 5818 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5821 5819 sizeof (uint64_t), tx);
5822 5820 VERIFY(zap_update(spa->spa_meta_objset,
5823 5821 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5824 5822 &sav->sav_object, tx) == 0);
5825 5823 }
5826 5824
5827 5825 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5828 5826 if (sav->sav_count == 0) {
5829 5827 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5830 5828 } else {
5831 5829 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5832 5830 for (i = 0; i < sav->sav_count; i++)
5833 5831 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5834 5832 B_FALSE, VDEV_CONFIG_L2CACHE);
5835 5833 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5836 5834 sav->sav_count) == 0);
5837 5835 for (i = 0; i < sav->sav_count; i++)
5838 5836 nvlist_free(list[i]);
5839 5837 kmem_free(list, sav->sav_count * sizeof (void *));
5840 5838 }
5841 5839
5842 5840 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5843 5841 nvlist_free(nvroot);
5844 5842
5845 5843 sav->sav_sync = B_FALSE;
5846 5844 }
5847 5845
5848 5846 static void
5849 5847 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5850 5848 {
5851 5849 nvlist_t *config;
5852 5850
5853 5851 if (list_is_empty(&spa->spa_config_dirty_list))
5854 5852 return;
5855 5853
5856 5854 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5857 5855
5858 5856 config = spa_config_generate(spa, spa->spa_root_vdev,
5859 5857 dmu_tx_get_txg(tx), B_FALSE);
5860 5858
5861 5859 /*
5862 5860 * If we're upgrading the spa version then make sure that
5863 5861 * the config object gets updated with the correct version.
5864 5862 */
5865 5863 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
5866 5864 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5867 5865 spa->spa_uberblock.ub_version);
5868 5866
5869 5867 spa_config_exit(spa, SCL_STATE, FTAG);
5870 5868
5871 5869 if (spa->spa_config_syncing)
5872 5870 nvlist_free(spa->spa_config_syncing);
5873 5871 spa->spa_config_syncing = config;
5874 5872
5875 5873 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5876 5874 }
5877 5875
5878 5876 static void
5879 5877 spa_sync_version(void *arg, dmu_tx_t *tx)
5880 5878 {
5881 5879 uint64_t *versionp = arg;
5882 5880 uint64_t version = *versionp;
5883 5881 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
5884 5882
5885 5883 /*
5886 5884 * Setting the version is special cased when first creating the pool.
5887 5885 */
5888 5886 ASSERT(tx->tx_txg != TXG_INITIAL);
5889 5887
5890 5888 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
5891 5889 ASSERT(version >= spa_version(spa));
5892 5890
5893 5891 spa->spa_uberblock.ub_version = version;
5894 5892 vdev_config_dirty(spa->spa_root_vdev);
5895 5893 spa_history_log_internal(spa, "set", tx, "version=%lld", version);
5896 5894 }
5897 5895
5898 5896 /*
5899 5897 * Set zpool properties.
5900 5898 */
5901 5899 static void
5902 5900 spa_sync_props(void *arg, dmu_tx_t *tx)
5903 5901 {
5904 5902 nvlist_t *nvp = arg;
5905 5903 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
5906 5904 objset_t *mos = spa->spa_meta_objset;
|
↓ open down ↓ |
3586 lines elided |
↑ open up ↑ |
5907 5905 nvpair_t *elem = NULL;
5908 5906
5909 5907 mutex_enter(&spa->spa_props_lock);
5910 5908
5911 5909 while ((elem = nvlist_next_nvpair(nvp, elem))) {
5912 5910 uint64_t intval;
5913 5911 char *strval, *fname;
5914 5912 zpool_prop_t prop;
5915 5913 const char *propname;
5916 5914 zprop_type_t proptype;
5917 - zfeature_info_t *feature;
5915 + spa_feature_t fid;
5918 5916
5919 5917 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
5920 5918 case ZPROP_INVAL:
5921 5919 /*
5922 5920 * We checked this earlier in spa_prop_validate().
5923 5921 */
5924 5922 ASSERT(zpool_prop_feature(nvpair_name(elem)));
5925 5923
5926 5924 fname = strchr(nvpair_name(elem), '@') + 1;
5927 - VERIFY0(zfeature_lookup_name(fname, &feature));
5925 + VERIFY0(zfeature_lookup_name(fname, &fid));
5928 5926
5929 - spa_feature_enable(spa, feature, tx);
5927 + spa_feature_enable(spa, fid, tx);
5930 5928 spa_history_log_internal(spa, "set", tx,
5931 5929 "%s=enabled", nvpair_name(elem));
5932 5930 break;
5933 5931
5934 5932 case ZPOOL_PROP_VERSION:
5935 5933 intval = fnvpair_value_uint64(elem);
5936 5934 /*
5937 5935 * The version is synced seperatly before other
5938 5936 * properties and should be correct by now.
5939 5937 */
5940 5938 ASSERT3U(spa_version(spa), >=, intval);
5941 5939 break;
5942 5940
5943 5941 case ZPOOL_PROP_ALTROOT:
5944 5942 /*
5945 5943 * 'altroot' is a non-persistent property. It should
5946 5944 * have been set temporarily at creation or import time.
5947 5945 */
5948 5946 ASSERT(spa->spa_root != NULL);
5949 5947 break;
5950 5948
5951 5949 case ZPOOL_PROP_READONLY:
5952 5950 case ZPOOL_PROP_CACHEFILE:
5953 5951 /*
5954 5952 * 'readonly' and 'cachefile' are also non-persisitent
5955 5953 * properties.
5956 5954 */
5957 5955 break;
5958 5956 case ZPOOL_PROP_COMMENT:
5959 5957 strval = fnvpair_value_string(elem);
5960 5958 if (spa->spa_comment != NULL)
5961 5959 spa_strfree(spa->spa_comment);
5962 5960 spa->spa_comment = spa_strdup(strval);
5963 5961 /*
5964 5962 * We need to dirty the configuration on all the vdevs
5965 5963 * so that their labels get updated. It's unnecessary
5966 5964 * to do this for pool creation since the vdev's
5967 5965 * configuratoin has already been dirtied.
5968 5966 */
5969 5967 if (tx->tx_txg != TXG_INITIAL)
5970 5968 vdev_config_dirty(spa->spa_root_vdev);
5971 5969 spa_history_log_internal(spa, "set", tx,
5972 5970 "%s=%s", nvpair_name(elem), strval);
5973 5971 break;
5974 5972 default:
5975 5973 /*
5976 5974 * Set pool property values in the poolprops mos object.
5977 5975 */
5978 5976 if (spa->spa_pool_props_object == 0) {
5979 5977 spa->spa_pool_props_object =
5980 5978 zap_create_link(mos, DMU_OT_POOL_PROPS,
5981 5979 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
5982 5980 tx);
5983 5981 }
5984 5982
5985 5983 /* normalize the property name */
5986 5984 propname = zpool_prop_to_name(prop);
5987 5985 proptype = zpool_prop_get_type(prop);
5988 5986
5989 5987 if (nvpair_type(elem) == DATA_TYPE_STRING) {
5990 5988 ASSERT(proptype == PROP_TYPE_STRING);
5991 5989 strval = fnvpair_value_string(elem);
5992 5990 VERIFY0(zap_update(mos,
5993 5991 spa->spa_pool_props_object, propname,
5994 5992 1, strlen(strval) + 1, strval, tx));
5995 5993 spa_history_log_internal(spa, "set", tx,
5996 5994 "%s=%s", nvpair_name(elem), strval);
5997 5995 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
5998 5996 intval = fnvpair_value_uint64(elem);
5999 5997
6000 5998 if (proptype == PROP_TYPE_INDEX) {
6001 5999 const char *unused;
6002 6000 VERIFY0(zpool_prop_index_to_string(
6003 6001 prop, intval, &unused));
6004 6002 }
6005 6003 VERIFY0(zap_update(mos,
6006 6004 spa->spa_pool_props_object, propname,
6007 6005 8, 1, &intval, tx));
6008 6006 spa_history_log_internal(spa, "set", tx,
6009 6007 "%s=%lld", nvpair_name(elem), intval);
6010 6008 } else {
6011 6009 ASSERT(0); /* not allowed */
6012 6010 }
6013 6011
6014 6012 switch (prop) {
6015 6013 case ZPOOL_PROP_DELEGATION:
6016 6014 spa->spa_delegation = intval;
6017 6015 break;
6018 6016 case ZPOOL_PROP_BOOTFS:
6019 6017 spa->spa_bootfs = intval;
6020 6018 break;
6021 6019 case ZPOOL_PROP_FAILUREMODE:
6022 6020 spa->spa_failmode = intval;
6023 6021 break;
6024 6022 case ZPOOL_PROP_AUTOEXPAND:
6025 6023 spa->spa_autoexpand = intval;
6026 6024 if (tx->tx_txg != TXG_INITIAL)
6027 6025 spa_async_request(spa,
6028 6026 SPA_ASYNC_AUTOEXPAND);
6029 6027 break;
6030 6028 case ZPOOL_PROP_DEDUPDITTO:
6031 6029 spa->spa_dedup_ditto = intval;
6032 6030 break;
6033 6031 default:
6034 6032 break;
6035 6033 }
6036 6034 }
6037 6035
6038 6036 }
6039 6037
6040 6038 mutex_exit(&spa->spa_props_lock);
6041 6039 }
6042 6040
6043 6041 /*
6044 6042 * Perform one-time upgrade on-disk changes. spa_version() does not
6045 6043 * reflect the new version this txg, so there must be no changes this
6046 6044 * txg to anything that the upgrade code depends on after it executes.
6047 6045 * Therefore this must be called after dsl_pool_sync() does the sync
6048 6046 * tasks.
6049 6047 */
6050 6048 static void
6051 6049 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
6052 6050 {
6053 6051 dsl_pool_t *dp = spa->spa_dsl_pool;
6054 6052
6055 6053 ASSERT(spa->spa_sync_pass == 1);
6056 6054
6057 6055 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
6058 6056
6059 6057 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
6060 6058 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
6061 6059 dsl_pool_create_origin(dp, tx);
6062 6060
6063 6061 /* Keeping the origin open increases spa_minref */
6064 6062 spa->spa_minref += 3;
6065 6063 }
6066 6064
6067 6065 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
6068 6066 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
6069 6067 dsl_pool_upgrade_clones(dp, tx);
6070 6068 }
6071 6069
6072 6070 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
6073 6071 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
6074 6072 dsl_pool_upgrade_dir_clones(dp, tx);
6075 6073
6076 6074 /* Keeping the freedir open increases spa_minref */
6077 6075 spa->spa_minref += 3;
6078 6076 }
6079 6077
6080 6078 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
6081 6079 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6082 6080 spa_feature_create_zap_objects(spa, tx);
6083 6081 }
6084 6082 rrw_exit(&dp->dp_config_rwlock, FTAG);
6085 6083 }
6086 6084
6087 6085 /*
6088 6086 * Sync the specified transaction group. New blocks may be dirtied as
6089 6087 * part of the process, so we iterate until it converges.
6090 6088 */
6091 6089 void
6092 6090 spa_sync(spa_t *spa, uint64_t txg)
6093 6091 {
6094 6092 dsl_pool_t *dp = spa->spa_dsl_pool;
6095 6093 objset_t *mos = spa->spa_meta_objset;
6096 6094 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
6097 6095 vdev_t *rvd = spa->spa_root_vdev;
6098 6096 vdev_t *vd;
6099 6097 dmu_tx_t *tx;
6100 6098 int error;
6101 6099
6102 6100 VERIFY(spa_writeable(spa));
6103 6101
6104 6102 /*
6105 6103 * Lock out configuration changes.
6106 6104 */
6107 6105 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6108 6106
6109 6107 spa->spa_syncing_txg = txg;
6110 6108 spa->spa_sync_pass = 0;
6111 6109
6112 6110 /*
6113 6111 * If there are any pending vdev state changes, convert them
6114 6112 * into config changes that go out with this transaction group.
6115 6113 */
6116 6114 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6117 6115 while (list_head(&spa->spa_state_dirty_list) != NULL) {
6118 6116 /*
6119 6117 * We need the write lock here because, for aux vdevs,
6120 6118 * calling vdev_config_dirty() modifies sav_config.
6121 6119 * This is ugly and will become unnecessary when we
6122 6120 * eliminate the aux vdev wart by integrating all vdevs
6123 6121 * into the root vdev tree.
6124 6122 */
6125 6123 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6126 6124 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
6127 6125 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
6128 6126 vdev_state_clean(vd);
6129 6127 vdev_config_dirty(vd);
6130 6128 }
6131 6129 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6132 6130 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
6133 6131 }
6134 6132 spa_config_exit(spa, SCL_STATE, FTAG);
6135 6133
6136 6134 tx = dmu_tx_create_assigned(dp, txg);
6137 6135
6138 6136 spa->spa_sync_starttime = gethrtime();
6139 6137 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid,
6140 6138 spa->spa_sync_starttime + spa->spa_deadman_synctime));
6141 6139
6142 6140 /*
6143 6141 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6144 6142 * set spa_deflate if we have no raid-z vdevs.
6145 6143 */
6146 6144 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
6147 6145 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
6148 6146 int i;
6149 6147
6150 6148 for (i = 0; i < rvd->vdev_children; i++) {
6151 6149 vd = rvd->vdev_child[i];
6152 6150 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
6153 6151 break;
6154 6152 }
6155 6153 if (i == rvd->vdev_children) {
6156 6154 spa->spa_deflate = TRUE;
6157 6155 VERIFY(0 == zap_add(spa->spa_meta_objset,
6158 6156 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
6159 6157 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
6160 6158 }
6161 6159 }
6162 6160
6163 6161 /*
6164 6162 * If anything has changed in this txg, or if someone is waiting
6165 6163 * for this txg to sync (eg, spa_vdev_remove()), push the
6166 6164 * deferred frees from the previous txg. If not, leave them
6167 6165 * alone so that we don't generate work on an otherwise idle
6168 6166 * system.
6169 6167 */
6170 6168 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
6171 6169 !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
6172 6170 !txg_list_empty(&dp->dp_sync_tasks, txg) ||
6173 6171 ((dsl_scan_active(dp->dp_scan) ||
6174 6172 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
6175 6173 spa_sync_deferred_frees(spa, tx);
6176 6174 }
6177 6175
6178 6176 /*
6179 6177 * Iterate to convergence.
6180 6178 */
6181 6179 do {
6182 6180 int pass = ++spa->spa_sync_pass;
6183 6181
6184 6182 spa_sync_config_object(spa, tx);
6185 6183 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
6186 6184 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
6187 6185 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
6188 6186 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
6189 6187 spa_errlog_sync(spa, txg);
6190 6188 dsl_pool_sync(dp, txg);
6191 6189
6192 6190 if (pass < zfs_sync_pass_deferred_free) {
6193 6191 spa_sync_frees(spa, free_bpl, tx);
6194 6192 } else {
6195 6193 bplist_iterate(free_bpl, bpobj_enqueue_cb,
6196 6194 &spa->spa_deferred_bpobj, tx);
6197 6195 }
6198 6196
6199 6197 ddt_sync(spa, txg);
6200 6198 dsl_scan_sync(dp, tx);
6201 6199
6202 6200 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
6203 6201 vdev_sync(vd, txg);
6204 6202
6205 6203 if (pass == 1)
6206 6204 spa_sync_upgrades(spa, tx);
6207 6205
6208 6206 } while (dmu_objset_is_dirty(mos, txg));
6209 6207
6210 6208 /*
6211 6209 * Rewrite the vdev configuration (which includes the uberblock)
6212 6210 * to commit the transaction group.
6213 6211 *
6214 6212 * If there are no dirty vdevs, we sync the uberblock to a few
6215 6213 * random top-level vdevs that are known to be visible in the
6216 6214 * config cache (see spa_vdev_add() for a complete description).
6217 6215 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6218 6216 */
6219 6217 for (;;) {
6220 6218 /*
6221 6219 * We hold SCL_STATE to prevent vdev open/close/etc.
6222 6220 * while we're attempting to write the vdev labels.
6223 6221 */
6224 6222 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6225 6223
6226 6224 if (list_is_empty(&spa->spa_config_dirty_list)) {
6227 6225 vdev_t *svd[SPA_DVAS_PER_BP];
6228 6226 int svdcount = 0;
6229 6227 int children = rvd->vdev_children;
6230 6228 int c0 = spa_get_random(children);
6231 6229
6232 6230 for (int c = 0; c < children; c++) {
6233 6231 vd = rvd->vdev_child[(c0 + c) % children];
6234 6232 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
6235 6233 continue;
6236 6234 svd[svdcount++] = vd;
6237 6235 if (svdcount == SPA_DVAS_PER_BP)
6238 6236 break;
6239 6237 }
6240 6238 error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
6241 6239 if (error != 0)
6242 6240 error = vdev_config_sync(svd, svdcount, txg,
6243 6241 B_TRUE);
6244 6242 } else {
6245 6243 error = vdev_config_sync(rvd->vdev_child,
6246 6244 rvd->vdev_children, txg, B_FALSE);
6247 6245 if (error != 0)
6248 6246 error = vdev_config_sync(rvd->vdev_child,
6249 6247 rvd->vdev_children, txg, B_TRUE);
6250 6248 }
6251 6249
6252 6250 if (error == 0)
6253 6251 spa->spa_last_synced_guid = rvd->vdev_guid;
6254 6252
6255 6253 spa_config_exit(spa, SCL_STATE, FTAG);
6256 6254
6257 6255 if (error == 0)
6258 6256 break;
6259 6257 zio_suspend(spa, NULL);
6260 6258 zio_resume_wait(spa);
6261 6259 }
6262 6260 dmu_tx_commit(tx);
6263 6261
6264 6262 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
6265 6263
6266 6264 /*
6267 6265 * Clear the dirty config list.
6268 6266 */
6269 6267 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
6270 6268 vdev_config_clean(vd);
6271 6269
6272 6270 /*
6273 6271 * Now that the new config has synced transactionally,
6274 6272 * let it become visible to the config cache.
6275 6273 */
6276 6274 if (spa->spa_config_syncing != NULL) {
6277 6275 spa_config_set(spa, spa->spa_config_syncing);
6278 6276 spa->spa_config_txg = txg;
6279 6277 spa->spa_config_syncing = NULL;
6280 6278 }
6281 6279
6282 6280 spa->spa_ubsync = spa->spa_uberblock;
6283 6281
6284 6282 dsl_pool_sync_done(dp, txg);
6285 6283
6286 6284 /*
6287 6285 * Update usable space statistics.
6288 6286 */
6289 6287 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
6290 6288 vdev_sync_done(vd, txg);
6291 6289
6292 6290 spa_update_dspace(spa);
6293 6291
6294 6292 /*
6295 6293 * It had better be the case that we didn't dirty anything
6296 6294 * since vdev_config_sync().
6297 6295 */
6298 6296 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
6299 6297 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
6300 6298 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
6301 6299
6302 6300 spa->spa_sync_pass = 0;
6303 6301
6304 6302 spa_config_exit(spa, SCL_CONFIG, FTAG);
6305 6303
6306 6304 spa_handle_ignored_writes(spa);
6307 6305
6308 6306 /*
6309 6307 * If any async tasks have been requested, kick them off.
6310 6308 */
6311 6309 spa_async_dispatch(spa);
6312 6310 }
6313 6311
6314 6312 /*
6315 6313 * Sync all pools. We don't want to hold the namespace lock across these
6316 6314 * operations, so we take a reference on the spa_t and drop the lock during the
6317 6315 * sync.
6318 6316 */
6319 6317 void
6320 6318 spa_sync_allpools(void)
6321 6319 {
6322 6320 spa_t *spa = NULL;
6323 6321 mutex_enter(&spa_namespace_lock);
6324 6322 while ((spa = spa_next(spa)) != NULL) {
6325 6323 if (spa_state(spa) != POOL_STATE_ACTIVE ||
6326 6324 !spa_writeable(spa) || spa_suspended(spa))
6327 6325 continue;
6328 6326 spa_open_ref(spa, FTAG);
6329 6327 mutex_exit(&spa_namespace_lock);
6330 6328 txg_wait_synced(spa_get_dsl(spa), 0);
6331 6329 mutex_enter(&spa_namespace_lock);
6332 6330 spa_close(spa, FTAG);
6333 6331 }
6334 6332 mutex_exit(&spa_namespace_lock);
6335 6333 }
6336 6334
6337 6335 /*
6338 6336 * ==========================================================================
6339 6337 * Miscellaneous routines
6340 6338 * ==========================================================================
6341 6339 */
6342 6340
6343 6341 /*
6344 6342 * Remove all pools in the system.
6345 6343 */
6346 6344 void
6347 6345 spa_evict_all(void)
6348 6346 {
6349 6347 spa_t *spa;
6350 6348
6351 6349 /*
6352 6350 * Remove all cached state. All pools should be closed now,
6353 6351 * so every spa in the AVL tree should be unreferenced.
6354 6352 */
6355 6353 mutex_enter(&spa_namespace_lock);
6356 6354 while ((spa = spa_next(NULL)) != NULL) {
6357 6355 /*
6358 6356 * Stop async tasks. The async thread may need to detach
6359 6357 * a device that's been replaced, which requires grabbing
6360 6358 * spa_namespace_lock, so we must drop it here.
6361 6359 */
6362 6360 spa_open_ref(spa, FTAG);
6363 6361 mutex_exit(&spa_namespace_lock);
6364 6362 spa_async_suspend(spa);
6365 6363 mutex_enter(&spa_namespace_lock);
6366 6364 spa_close(spa, FTAG);
6367 6365
6368 6366 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
6369 6367 spa_unload(spa);
6370 6368 spa_deactivate(spa);
6371 6369 }
6372 6370 spa_remove(spa);
6373 6371 }
6374 6372 mutex_exit(&spa_namespace_lock);
6375 6373 }
6376 6374
6377 6375 vdev_t *
6378 6376 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
6379 6377 {
6380 6378 vdev_t *vd;
6381 6379 int i;
6382 6380
6383 6381 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
6384 6382 return (vd);
6385 6383
6386 6384 if (aux) {
6387 6385 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
6388 6386 vd = spa->spa_l2cache.sav_vdevs[i];
6389 6387 if (vd->vdev_guid == guid)
6390 6388 return (vd);
6391 6389 }
6392 6390
6393 6391 for (i = 0; i < spa->spa_spares.sav_count; i++) {
6394 6392 vd = spa->spa_spares.sav_vdevs[i];
6395 6393 if (vd->vdev_guid == guid)
6396 6394 return (vd);
6397 6395 }
6398 6396 }
6399 6397
6400 6398 return (NULL);
6401 6399 }
6402 6400
6403 6401 void
6404 6402 spa_upgrade(spa_t *spa, uint64_t version)
6405 6403 {
6406 6404 ASSERT(spa_writeable(spa));
6407 6405
6408 6406 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6409 6407
6410 6408 /*
6411 6409 * This should only be called for a non-faulted pool, and since a
6412 6410 * future version would result in an unopenable pool, this shouldn't be
6413 6411 * possible.
6414 6412 */
6415 6413 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version));
6416 6414 ASSERT(version >= spa->spa_uberblock.ub_version);
6417 6415
6418 6416 spa->spa_uberblock.ub_version = version;
6419 6417 vdev_config_dirty(spa->spa_root_vdev);
6420 6418
6421 6419 spa_config_exit(spa, SCL_ALL, FTAG);
6422 6420
6423 6421 txg_wait_synced(spa_get_dsl(spa), 0);
6424 6422 }
6425 6423
6426 6424 boolean_t
6427 6425 spa_has_spare(spa_t *spa, uint64_t guid)
6428 6426 {
6429 6427 int i;
6430 6428 uint64_t spareguid;
6431 6429 spa_aux_vdev_t *sav = &spa->spa_spares;
6432 6430
6433 6431 for (i = 0; i < sav->sav_count; i++)
6434 6432 if (sav->sav_vdevs[i]->vdev_guid == guid)
6435 6433 return (B_TRUE);
6436 6434
6437 6435 for (i = 0; i < sav->sav_npending; i++) {
6438 6436 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
6439 6437 &spareguid) == 0 && spareguid == guid)
6440 6438 return (B_TRUE);
6441 6439 }
6442 6440
6443 6441 return (B_FALSE);
6444 6442 }
6445 6443
6446 6444 /*
6447 6445 * Check if a pool has an active shared spare device.
6448 6446 * Note: reference count of an active spare is 2, as a spare and as a replace
6449 6447 */
6450 6448 static boolean_t
6451 6449 spa_has_active_shared_spare(spa_t *spa)
6452 6450 {
6453 6451 int i, refcnt;
6454 6452 uint64_t pool;
6455 6453 spa_aux_vdev_t *sav = &spa->spa_spares;
6456 6454
6457 6455 for (i = 0; i < sav->sav_count; i++) {
6458 6456 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
6459 6457 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
6460 6458 refcnt > 2)
6461 6459 return (B_TRUE);
6462 6460 }
6463 6461
6464 6462 return (B_FALSE);
6465 6463 }
6466 6464
6467 6465 /*
6468 6466 * Post a sysevent corresponding to the given event. The 'name' must be one of
6469 6467 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
6470 6468 * filled in from the spa and (optionally) the vdev. This doesn't do anything
6471 6469 * in the userland libzpool, as we don't want consumers to misinterpret ztest
6472 6470 * or zdb as real changes.
6473 6471 */
6474 6472 void
6475 6473 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
6476 6474 {
6477 6475 #ifdef _KERNEL
6478 6476 sysevent_t *ev;
6479 6477 sysevent_attr_list_t *attr = NULL;
6480 6478 sysevent_value_t value;
6481 6479 sysevent_id_t eid;
6482 6480
6483 6481 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
6484 6482 SE_SLEEP);
6485 6483
6486 6484 value.value_type = SE_DATA_TYPE_STRING;
6487 6485 value.value.sv_string = spa_name(spa);
6488 6486 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
6489 6487 goto done;
6490 6488
6491 6489 value.value_type = SE_DATA_TYPE_UINT64;
6492 6490 value.value.sv_uint64 = spa_guid(spa);
6493 6491 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
6494 6492 goto done;
6495 6493
6496 6494 if (vd) {
6497 6495 value.value_type = SE_DATA_TYPE_UINT64;
6498 6496 value.value.sv_uint64 = vd->vdev_guid;
6499 6497 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
6500 6498 SE_SLEEP) != 0)
6501 6499 goto done;
6502 6500
6503 6501 if (vd->vdev_path) {
6504 6502 value.value_type = SE_DATA_TYPE_STRING;
6505 6503 value.value.sv_string = vd->vdev_path;
6506 6504 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
6507 6505 &value, SE_SLEEP) != 0)
6508 6506 goto done;
6509 6507 }
6510 6508 }
6511 6509
6512 6510 if (sysevent_attach_attributes(ev, attr) != 0)
6513 6511 goto done;
6514 6512 attr = NULL;
6515 6513
6516 6514 (void) log_sysevent(ev, SE_SLEEP, &eid);
6517 6515
6518 6516 done:
6519 6517 if (attr)
6520 6518 sysevent_free_attr(attr);
6521 6519 sysevent_free(ev);
6522 6520 #endif
6523 6521 }
|
↓ open down ↓ |
584 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX