1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 /* 22 * Copyright 2009 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 #include <sys/zfs_context.h> 27 #include <sys/spa_impl.h> 28 #include <sys/zio.h> 29 #include <sys/zio_checksum.h> 30 #include <sys/zio_compress.h> 31 #include <sys/dmu.h> 32 #include <sys/dmu_tx.h> 33 #include <sys/zap.h> 34 #include <sys/zil.h> 35 #include <sys/vdev_impl.h> 36 #include <sys/metaslab.h> 37 #include <sys/uberblock_impl.h> 38 #include <sys/txg.h> 39 #include <sys/avl.h> 40 #include <sys/unique.h> 41 #include <sys/dsl_pool.h> 42 #include <sys/dsl_dir.h> 43 #include <sys/dsl_prop.h> 44 #include <sys/fs/zfs.h> 45 #include <sys/metaslab_impl.h> 46 #include <sys/arc.h> 47 #include <sys/ddt.h> 48 #include "zfs_prop.h" 49 50 /* 51 * SPA locking 52 * 53 * There are four basic locks for managing spa_t structures: 54 * 55 * spa_namespace_lock (global mutex) 56 * 57 * This lock must be acquired to do any of the following: 58 * 59 * - Lookup a spa_t by name 60 * - Add or remove a spa_t from the namespace 61 * - Increase spa_refcount from non-zero 62 * - Check if spa_refcount is zero 63 * - Rename a spa_t 64 * - add/remove/attach/detach devices 65 * - Held for the duration of create/destroy/import/export 66 * 67 * It does not need to handle recursion. A create or destroy may 68 * reference objects (files or zvols) in other pools, but by 69 * definition they must have an existing reference, and will never need 70 * to lookup a spa_t by name. 71 * 72 * spa_refcount (per-spa refcount_t protected by mutex) 73 * 74 * This reference count keep track of any active users of the spa_t. The 75 * spa_t cannot be destroyed or freed while this is non-zero. Internally, 76 * the refcount is never really 'zero' - opening a pool implicitly keeps 77 * some references in the DMU. Internally we check against spa_minref, but 78 * present the image of a zero/non-zero value to consumers. 79 * 80 * spa_config_lock[] (per-spa array of rwlocks) 81 * 82 * This protects the spa_t from config changes, and must be held in 83 * the following circumstances: 84 * 85 * - RW_READER to perform I/O to the spa 86 * - RW_WRITER to change the vdev config 87 * 88 * The locking order is fairly straightforward: 89 * 90 * spa_namespace_lock -> spa_refcount 91 * 92 * The namespace lock must be acquired to increase the refcount from 0 93 * or to check if it is zero. 94 * 95 * spa_refcount -> spa_config_lock[] 96 * 97 * There must be at least one valid reference on the spa_t to acquire 98 * the config lock. 99 * 100 * spa_namespace_lock -> spa_config_lock[] 101 * 102 * The namespace lock must always be taken before the config lock. 103 * 104 * 105 * The spa_namespace_lock can be acquired directly and is globally visible. 106 * 107 * The namespace is manipulated using the following functions, all of which 108 * require the spa_namespace_lock to be held. 109 * 110 * spa_lookup() Lookup a spa_t by name. 111 * 112 * spa_add() Create a new spa_t in the namespace. 113 * 114 * spa_remove() Remove a spa_t from the namespace. This also 115 * frees up any memory associated with the spa_t. 116 * 117 * spa_next() Returns the next spa_t in the system, or the 118 * first if NULL is passed. 119 * 120 * spa_evict_all() Shutdown and remove all spa_t structures in 121 * the system. 122 * 123 * spa_guid_exists() Determine whether a pool/device guid exists. 124 * 125 * The spa_refcount is manipulated using the following functions: 126 * 127 * spa_open_ref() Adds a reference to the given spa_t. Must be 128 * called with spa_namespace_lock held if the 129 * refcount is currently zero. 130 * 131 * spa_close() Remove a reference from the spa_t. This will 132 * not free the spa_t or remove it from the 133 * namespace. No locking is required. 134 * 135 * spa_refcount_zero() Returns true if the refcount is currently 136 * zero. Must be called with spa_namespace_lock 137 * held. 138 * 139 * The spa_config_lock[] is an array of rwlocks, ordered as follows: 140 * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV. 141 * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}(). 142 * 143 * To read the configuration, it suffices to hold one of these locks as reader. 144 * To modify the configuration, you must hold all locks as writer. To modify 145 * vdev state without altering the vdev tree's topology (e.g. online/offline), 146 * you must hold SCL_STATE and SCL_ZIO as writer. 147 * 148 * We use these distinct config locks to avoid recursive lock entry. 149 * For example, spa_sync() (which holds SCL_CONFIG as reader) induces 150 * block allocations (SCL_ALLOC), which may require reading space maps 151 * from disk (dmu_read() -> zio_read() -> SCL_ZIO). 152 * 153 * The spa config locks cannot be normal rwlocks because we need the 154 * ability to hand off ownership. For example, SCL_ZIO is acquired 155 * by the issuing thread and later released by an interrupt thread. 156 * They do, however, obey the usual write-wanted semantics to prevent 157 * writer (i.e. system administrator) starvation. 158 * 159 * The lock acquisition rules are as follows: 160 * 161 * SCL_CONFIG 162 * Protects changes to the vdev tree topology, such as vdev 163 * add/remove/attach/detach. Protects the dirty config list 164 * (spa_config_dirty_list) and the set of spares and l2arc devices. 165 * 166 * SCL_STATE 167 * Protects changes to pool state and vdev state, such as vdev 168 * online/offline/fault/degrade/clear. Protects the dirty state list 169 * (spa_state_dirty_list) and global pool state (spa_state). 170 * 171 * SCL_ALLOC 172 * Protects changes to metaslab groups and classes. 173 * Held as reader by metaslab_alloc() and metaslab_claim(). 174 * 175 * SCL_ZIO 176 * Held by bp-level zios (those which have no io_vd upon entry) 177 * to prevent changes to the vdev tree. The bp-level zio implicitly 178 * protects all of its vdev child zios, which do not hold SCL_ZIO. 179 * 180 * SCL_FREE 181 * Protects changes to metaslab groups and classes. 182 * Held as reader by metaslab_free(). SCL_FREE is distinct from 183 * SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free 184 * blocks in zio_done() while another i/o that holds either 185 * SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete. 186 * 187 * SCL_VDEV 188 * Held as reader to prevent changes to the vdev tree during trivial 189 * inquiries such as bp_get_dsize(). SCL_VDEV is distinct from the 190 * other locks, and lower than all of them, to ensure that it's safe 191 * to acquire regardless of caller context. 192 * 193 * In addition, the following rules apply: 194 * 195 * (a) spa_props_lock protects pool properties, spa_config and spa_config_list. 196 * The lock ordering is SCL_CONFIG > spa_props_lock. 197 * 198 * (b) I/O operations on leaf vdevs. For any zio operation that takes 199 * an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(), 200 * or zio_write_phys() -- the caller must ensure that the config cannot 201 * cannot change in the interim, and that the vdev cannot be reopened. 202 * SCL_STATE as reader suffices for both. 203 * 204 * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit(). 205 * 206 * spa_vdev_enter() Acquire the namespace lock and the config lock 207 * for writing. 208 * 209 * spa_vdev_exit() Release the config lock, wait for all I/O 210 * to complete, sync the updated configs to the 211 * cache, and release the namespace lock. 212 * 213 * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit(). 214 * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual 215 * locking is, always, based on spa_namespace_lock and spa_config_lock[]. 216 * 217 * spa_rename() is also implemented within this file since is requires 218 * manipulation of the namespace. 219 */ 220 221 static avl_tree_t spa_namespace_avl; 222 kmutex_t spa_namespace_lock; 223 static kcondvar_t spa_namespace_cv; 224 static int spa_active_count; 225 int spa_max_replication_override = SPA_DVAS_PER_BP; 226 227 static kmutex_t spa_spare_lock; 228 static avl_tree_t spa_spare_avl; 229 static kmutex_t spa_l2cache_lock; 230 static avl_tree_t spa_l2cache_avl; 231 232 kmem_cache_t *spa_buffer_pool; 233 int spa_mode_global; 234 235 #ifdef ZFS_DEBUG 236 /* Everything except dprintf is on by default in debug builds */ 237 int zfs_flags = ~ZFS_DEBUG_DPRINTF; 238 #else 239 int zfs_flags = 0; 240 #endif 241 242 /* 243 * zfs_recover can be set to nonzero to attempt to recover from 244 * otherwise-fatal errors, typically caused by on-disk corruption. When 245 * set, calls to zfs_panic_recover() will turn into warning messages. 246 */ 247 int zfs_recover = 0; 248 249 250 /* 251 * ========================================================================== 252 * SPA config locking 253 * ========================================================================== 254 */ 255 static void 256 spa_config_lock_init(spa_t *spa) 257 { 258 for (int i = 0; i < SCL_LOCKS; i++) { 259 spa_config_lock_t *scl = &spa->spa_config_lock[i]; 260 mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL); 261 cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL); 262 refcount_create(&scl->scl_count); 263 scl->scl_writer = NULL; 264 scl->scl_write_wanted = 0; 265 } 266 } 267 268 static void 269 spa_config_lock_destroy(spa_t *spa) 270 { 271 for (int i = 0; i < SCL_LOCKS; i++) { 272 spa_config_lock_t *scl = &spa->spa_config_lock[i]; 273 mutex_destroy(&scl->scl_lock); 274 cv_destroy(&scl->scl_cv); 275 refcount_destroy(&scl->scl_count); 276 ASSERT(scl->scl_writer == NULL); 277 ASSERT(scl->scl_write_wanted == 0); 278 } 279 } 280 281 int 282 spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw) 283 { 284 for (int i = 0; i < SCL_LOCKS; i++) { 285 spa_config_lock_t *scl = &spa->spa_config_lock[i]; 286 if (!(locks & (1 << i))) 287 continue; 288 mutex_enter(&scl->scl_lock); 289 if (rw == RW_READER) { 290 if (scl->scl_writer || scl->scl_write_wanted) { 291 mutex_exit(&scl->scl_lock); 292 spa_config_exit(spa, locks ^ (1 << i), tag); 293 return (0); 294 } 295 } else { 296 ASSERT(scl->scl_writer != curthread); 297 if (!refcount_is_zero(&scl->scl_count)) { 298 mutex_exit(&scl->scl_lock); 299 spa_config_exit(spa, locks ^ (1 << i), tag); 300 return (0); 301 } 302 scl->scl_writer = curthread; 303 } 304 (void) refcount_add(&scl->scl_count, tag); 305 mutex_exit(&scl->scl_lock); 306 } 307 return (1); 308 } 309 310 void 311 spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw) 312 { 313 int wlocks_held = 0; 314 315 for (int i = 0; i < SCL_LOCKS; i++) { 316 spa_config_lock_t *scl = &spa->spa_config_lock[i]; 317 if (scl->scl_writer == curthread) 318 wlocks_held |= (1 << i); 319 if (!(locks & (1 << i))) 320 continue; 321 mutex_enter(&scl->scl_lock); 322 if (rw == RW_READER) { 323 while (scl->scl_writer || scl->scl_write_wanted) { 324 cv_wait(&scl->scl_cv, &scl->scl_lock); 325 } 326 } else { 327 ASSERT(scl->scl_writer != curthread); 328 while (!refcount_is_zero(&scl->scl_count)) { 329 scl->scl_write_wanted++; 330 cv_wait(&scl->scl_cv, &scl->scl_lock); 331 scl->scl_write_wanted--; 332 } 333 scl->scl_writer = curthread; 334 } 335 (void) refcount_add(&scl->scl_count, tag); 336 mutex_exit(&scl->scl_lock); 337 } 338 ASSERT(wlocks_held <= locks); 339 } 340 341 void 342 spa_config_exit(spa_t *spa, int locks, void *tag) 343 { 344 for (int i = SCL_LOCKS - 1; i >= 0; i--) { 345 spa_config_lock_t *scl = &spa->spa_config_lock[i]; 346 if (!(locks & (1 << i))) 347 continue; 348 mutex_enter(&scl->scl_lock); 349 ASSERT(!refcount_is_zero(&scl->scl_count)); 350 if (refcount_remove(&scl->scl_count, tag) == 0) { 351 ASSERT(scl->scl_writer == NULL || 352 scl->scl_writer == curthread); 353 scl->scl_writer = NULL; /* OK in either case */ 354 cv_broadcast(&scl->scl_cv); 355 } 356 mutex_exit(&scl->scl_lock); 357 } 358 } 359 360 int 361 spa_config_held(spa_t *spa, int locks, krw_t rw) 362 { 363 int locks_held = 0; 364 365 for (int i = 0; i < SCL_LOCKS; i++) { 366 spa_config_lock_t *scl = &spa->spa_config_lock[i]; 367 if (!(locks & (1 << i))) 368 continue; 369 if ((rw == RW_READER && !refcount_is_zero(&scl->scl_count)) || 370 (rw == RW_WRITER && scl->scl_writer == curthread)) 371 locks_held |= 1 << i; 372 } 373 374 return (locks_held); 375 } 376 377 /* 378 * ========================================================================== 379 * SPA namespace functions 380 * ========================================================================== 381 */ 382 383 /* 384 * Lookup the named spa_t in the AVL tree. The spa_namespace_lock must be held. 385 * Returns NULL if no matching spa_t is found. 386 */ 387 spa_t * 388 spa_lookup(const char *name) 389 { 390 static spa_t search; /* spa_t is large; don't allocate on stack */ 391 spa_t *spa; 392 avl_index_t where; 393 char c; 394 char *cp; 395 396 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 397 398 /* 399 * If it's a full dataset name, figure out the pool name and 400 * just use that. 401 */ 402 cp = strpbrk(name, "/@"); 403 if (cp) { 404 c = *cp; 405 *cp = '\0'; 406 } 407 408 (void) strlcpy(search.spa_name, name, sizeof (search.spa_name)); 409 spa = avl_find(&spa_namespace_avl, &search, &where); 410 411 if (cp) 412 *cp = c; 413 414 return (spa); 415 } 416 417 /* 418 * Create an uninitialized spa_t with the given name. Requires 419 * spa_namespace_lock. The caller must ensure that the spa_t doesn't already 420 * exist by calling spa_lookup() first. 421 */ 422 spa_t * 423 spa_add(const char *name, nvlist_t *config, const char *altroot) 424 { 425 spa_t *spa; 426 spa_config_dirent_t *dp; 427 428 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 429 430 spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP); 431 432 mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL); 433 mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL); 434 mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL); 435 mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL); 436 mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL); 437 mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL); 438 mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL); 439 mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL); 440 441 cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL); 442 cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL); 443 cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL); 444 445 for (int t = 0; t < TXG_SIZE; t++) 446 bplist_init(&spa->spa_free_bplist[t]); 447 bplist_init(&spa->spa_deferred_bplist); 448 449 (void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name)); 450 spa->spa_state = POOL_STATE_UNINITIALIZED; 451 spa->spa_freeze_txg = UINT64_MAX; 452 spa->spa_final_txg = UINT64_MAX; 453 spa->spa_load_max_txg = UINT64_MAX; 454 455 refcount_create(&spa->spa_refcount); 456 spa_config_lock_init(spa); 457 458 avl_add(&spa_namespace_avl, spa); 459 460 /* 461 * Set the alternate root, if there is one. 462 */ 463 if (altroot) { 464 spa->spa_root = spa_strdup(altroot); 465 spa_active_count++; 466 } 467 468 /* 469 * Every pool starts with the default cachefile 470 */ 471 list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t), 472 offsetof(spa_config_dirent_t, scd_link)); 473 474 dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP); 475 dp->scd_path = spa_strdup(spa_config_path); 476 list_insert_head(&spa->spa_config_list, dp); 477 478 if (config != NULL) 479 VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0); 480 481 return (spa); 482 } 483 484 /* 485 * Removes a spa_t from the namespace, freeing up any memory used. Requires 486 * spa_namespace_lock. This is called only after the spa_t has been closed and 487 * deactivated. 488 */ 489 void 490 spa_remove(spa_t *spa) 491 { 492 spa_config_dirent_t *dp; 493 494 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 495 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); 496 497 avl_remove(&spa_namespace_avl, spa); 498 cv_broadcast(&spa_namespace_cv); 499 500 if (spa->spa_root) { 501 spa_strfree(spa->spa_root); 502 spa_active_count--; 503 } 504 505 while ((dp = list_head(&spa->spa_config_list)) != NULL) { 506 list_remove(&spa->spa_config_list, dp); 507 if (dp->scd_path != NULL) 508 spa_strfree(dp->scd_path); 509 kmem_free(dp, sizeof (spa_config_dirent_t)); 510 } 511 512 list_destroy(&spa->spa_config_list); 513 514 spa_config_set(spa, NULL); 515 516 refcount_destroy(&spa->spa_refcount); 517 518 spa_config_lock_destroy(spa); 519 520 for (int t = 0; t < TXG_SIZE; t++) 521 bplist_fini(&spa->spa_free_bplist[t]); 522 bplist_fini(&spa->spa_deferred_bplist); 523 524 cv_destroy(&spa->spa_async_cv); 525 cv_destroy(&spa->spa_scrub_io_cv); 526 cv_destroy(&spa->spa_suspend_cv); 527 528 mutex_destroy(&spa->spa_async_lock); 529 mutex_destroy(&spa->spa_scrub_lock); 530 mutex_destroy(&spa->spa_errlog_lock); 531 mutex_destroy(&spa->spa_errlist_lock); 532 mutex_destroy(&spa->spa_history_lock); 533 mutex_destroy(&spa->spa_props_lock); 534 mutex_destroy(&spa->spa_suspend_lock); 535 mutex_destroy(&spa->spa_vdev_top_lock); 536 537 kmem_free(spa, sizeof (spa_t)); 538 } 539 540 /* 541 * Given a pool, return the next pool in the namespace, or NULL if there is 542 * none. If 'prev' is NULL, return the first pool. 543 */ 544 spa_t * 545 spa_next(spa_t *prev) 546 { 547 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 548 549 if (prev) 550 return (AVL_NEXT(&spa_namespace_avl, prev)); 551 else 552 return (avl_first(&spa_namespace_avl)); 553 } 554 555 /* 556 * ========================================================================== 557 * SPA refcount functions 558 * ========================================================================== 559 */ 560 561 /* 562 * Add a reference to the given spa_t. Must have at least one reference, or 563 * have the namespace lock held. 564 */ 565 void 566 spa_open_ref(spa_t *spa, void *tag) 567 { 568 ASSERT(refcount_count(&spa->spa_refcount) >= spa->spa_minref || 569 MUTEX_HELD(&spa_namespace_lock)); 570 (void) refcount_add(&spa->spa_refcount, tag); 571 } 572 573 /* 574 * Remove a reference to the given spa_t. Must have at least one reference, or 575 * have the namespace lock held. 576 */ 577 void 578 spa_close(spa_t *spa, void *tag) 579 { 580 ASSERT(refcount_count(&spa->spa_refcount) > spa->spa_minref || 581 MUTEX_HELD(&spa_namespace_lock)); 582 (void) refcount_remove(&spa->spa_refcount, tag); 583 } 584 585 /* 586 * Check to see if the spa refcount is zero. Must be called with 587 * spa_namespace_lock held. We really compare against spa_minref, which is the 588 * number of references acquired when opening a pool 589 */ 590 boolean_t 591 spa_refcount_zero(spa_t *spa) 592 { 593 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 594 595 return (refcount_count(&spa->spa_refcount) == spa->spa_minref); 596 } 597 598 /* 599 * ========================================================================== 600 * SPA spare and l2cache tracking 601 * ========================================================================== 602 */ 603 604 /* 605 * Hot spares and cache devices are tracked using the same code below, 606 * for 'auxiliary' devices. 607 */ 608 609 typedef struct spa_aux { 610 uint64_t aux_guid; 611 uint64_t aux_pool; 612 avl_node_t aux_avl; 613 int aux_count; 614 } spa_aux_t; 615 616 static int 617 spa_aux_compare(const void *a, const void *b) 618 { 619 const spa_aux_t *sa = a; 620 const spa_aux_t *sb = b; 621 622 if (sa->aux_guid < sb->aux_guid) 623 return (-1); 624 else if (sa->aux_guid > sb->aux_guid) 625 return (1); 626 else 627 return (0); 628 } 629 630 void 631 spa_aux_add(vdev_t *vd, avl_tree_t *avl) 632 { 633 avl_index_t where; 634 spa_aux_t search; 635 spa_aux_t *aux; 636 637 search.aux_guid = vd->vdev_guid; 638 if ((aux = avl_find(avl, &search, &where)) != NULL) { 639 aux->aux_count++; 640 } else { 641 aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP); 642 aux->aux_guid = vd->vdev_guid; 643 aux->aux_count = 1; 644 avl_insert(avl, aux, where); 645 } 646 } 647 648 void 649 spa_aux_remove(vdev_t *vd, avl_tree_t *avl) 650 { 651 spa_aux_t search; 652 spa_aux_t *aux; 653 avl_index_t where; 654 655 search.aux_guid = vd->vdev_guid; 656 aux = avl_find(avl, &search, &where); 657 658 ASSERT(aux != NULL); 659 660 if (--aux->aux_count == 0) { 661 avl_remove(avl, aux); 662 kmem_free(aux, sizeof (spa_aux_t)); 663 } else if (aux->aux_pool == spa_guid(vd->vdev_spa)) { 664 aux->aux_pool = 0ULL; 665 } 666 } 667 668 boolean_t 669 spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl) 670 { 671 spa_aux_t search, *found; 672 673 search.aux_guid = guid; 674 found = avl_find(avl, &search, NULL); 675 676 if (pool) { 677 if (found) 678 *pool = found->aux_pool; 679 else 680 *pool = 0ULL; 681 } 682 683 if (refcnt) { 684 if (found) 685 *refcnt = found->aux_count; 686 else 687 *refcnt = 0; 688 } 689 690 return (found != NULL); 691 } 692 693 void 694 spa_aux_activate(vdev_t *vd, avl_tree_t *avl) 695 { 696 spa_aux_t search, *found; 697 avl_index_t where; 698 699 search.aux_guid = vd->vdev_guid; 700 found = avl_find(avl, &search, &where); 701 ASSERT(found != NULL); 702 ASSERT(found->aux_pool == 0ULL); 703 704 found->aux_pool = spa_guid(vd->vdev_spa); 705 } 706 707 /* 708 * Spares are tracked globally due to the following constraints: 709 * 710 * - A spare may be part of multiple pools. 711 * - A spare may be added to a pool even if it's actively in use within 712 * another pool. 713 * - A spare in use in any pool can only be the source of a replacement if 714 * the target is a spare in the same pool. 715 * 716 * We keep track of all spares on the system through the use of a reference 717 * counted AVL tree. When a vdev is added as a spare, or used as a replacement 718 * spare, then we bump the reference count in the AVL tree. In addition, we set 719 * the 'vdev_isspare' member to indicate that the device is a spare (active or 720 * inactive). When a spare is made active (used to replace a device in the 721 * pool), we also keep track of which pool its been made a part of. 722 * 723 * The 'spa_spare_lock' protects the AVL tree. These functions are normally 724 * called under the spa_namespace lock as part of vdev reconfiguration. The 725 * separate spare lock exists for the status query path, which does not need to 726 * be completely consistent with respect to other vdev configuration changes. 727 */ 728 729 static int 730 spa_spare_compare(const void *a, const void *b) 731 { 732 return (spa_aux_compare(a, b)); 733 } 734 735 void 736 spa_spare_add(vdev_t *vd) 737 { 738 mutex_enter(&spa_spare_lock); 739 ASSERT(!vd->vdev_isspare); 740 spa_aux_add(vd, &spa_spare_avl); 741 vd->vdev_isspare = B_TRUE; 742 mutex_exit(&spa_spare_lock); 743 } 744 745 void 746 spa_spare_remove(vdev_t *vd) 747 { 748 mutex_enter(&spa_spare_lock); 749 ASSERT(vd->vdev_isspare); 750 spa_aux_remove(vd, &spa_spare_avl); 751 vd->vdev_isspare = B_FALSE; 752 mutex_exit(&spa_spare_lock); 753 } 754 755 boolean_t 756 spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt) 757 { 758 boolean_t found; 759 760 mutex_enter(&spa_spare_lock); 761 found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl); 762 mutex_exit(&spa_spare_lock); 763 764 return (found); 765 } 766 767 void 768 spa_spare_activate(vdev_t *vd) 769 { 770 mutex_enter(&spa_spare_lock); 771 ASSERT(vd->vdev_isspare); 772 spa_aux_activate(vd, &spa_spare_avl); 773 mutex_exit(&spa_spare_lock); 774 } 775 776 /* 777 * Level 2 ARC devices are tracked globally for the same reasons as spares. 778 * Cache devices currently only support one pool per cache device, and so 779 * for these devices the aux reference count is currently unused beyond 1. 780 */ 781 782 static int 783 spa_l2cache_compare(const void *a, const void *b) 784 { 785 return (spa_aux_compare(a, b)); 786 } 787 788 void 789 spa_l2cache_add(vdev_t *vd) 790 { 791 mutex_enter(&spa_l2cache_lock); 792 ASSERT(!vd->vdev_isl2cache); 793 spa_aux_add(vd, &spa_l2cache_avl); 794 vd->vdev_isl2cache = B_TRUE; 795 mutex_exit(&spa_l2cache_lock); 796 } 797 798 void 799 spa_l2cache_remove(vdev_t *vd) 800 { 801 mutex_enter(&spa_l2cache_lock); 802 ASSERT(vd->vdev_isl2cache); 803 spa_aux_remove(vd, &spa_l2cache_avl); 804 vd->vdev_isl2cache = B_FALSE; 805 mutex_exit(&spa_l2cache_lock); 806 } 807 808 boolean_t 809 spa_l2cache_exists(uint64_t guid, uint64_t *pool) 810 { 811 boolean_t found; 812 813 mutex_enter(&spa_l2cache_lock); 814 found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl); 815 mutex_exit(&spa_l2cache_lock); 816 817 return (found); 818 } 819 820 void 821 spa_l2cache_activate(vdev_t *vd) 822 { 823 mutex_enter(&spa_l2cache_lock); 824 ASSERT(vd->vdev_isl2cache); 825 spa_aux_activate(vd, &spa_l2cache_avl); 826 mutex_exit(&spa_l2cache_lock); 827 } 828 829 /* 830 * ========================================================================== 831 * SPA vdev locking 832 * ========================================================================== 833 */ 834 835 /* 836 * Lock the given spa_t for the purpose of adding or removing a vdev. 837 * Grabs the global spa_namespace_lock plus the spa config lock for writing. 838 * It returns the next transaction group for the spa_t. 839 */ 840 uint64_t 841 spa_vdev_enter(spa_t *spa) 842 { 843 mutex_enter(&spa->spa_vdev_top_lock); 844 mutex_enter(&spa_namespace_lock); 845 return (spa_vdev_config_enter(spa)); 846 } 847 848 /* 849 * Internal implementation for spa_vdev_enter(). Used when a vdev 850 * operation requires multiple syncs (i.e. removing a device) while 851 * keeping the spa_namespace_lock held. 852 */ 853 uint64_t 854 spa_vdev_config_enter(spa_t *spa) 855 { 856 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 857 858 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER); 859 860 return (spa_last_synced_txg(spa) + 1); 861 } 862 863 /* 864 * Used in combination with spa_vdev_config_enter() to allow the syncing 865 * of multiple transactions without releasing the spa_namespace_lock. 866 */ 867 void 868 spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag) 869 { 870 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 871 872 int config_changed = B_FALSE; 873 874 ASSERT(txg > spa_last_synced_txg(spa)); 875 876 spa->spa_pending_vdev = NULL; 877 878 /* 879 * Reassess the DTLs. 880 */ 881 vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE); 882 883 /* 884 * If the config changed, notify the scrub thread that it must restart. 885 */ 886 if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) { 887 dsl_pool_scrub_restart(spa->spa_dsl_pool); 888 config_changed = B_TRUE; 889 spa->spa_config_generation++; 890 } 891 892 /* 893 * Verify the metaslab classes. 894 */ 895 ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0); 896 ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0); 897 898 spa_config_exit(spa, SCL_ALL, spa); 899 900 /* 901 * Panic the system if the specified tag requires it. This 902 * is useful for ensuring that configurations are updated 903 * transactionally. 904 */ 905 if (zio_injection_enabled) 906 zio_handle_panic_injection(spa, tag); 907 908 /* 909 * Note: this txg_wait_synced() is important because it ensures 910 * that there won't be more than one config change per txg. 911 * This allows us to use the txg as the generation number. 912 */ 913 if (error == 0) 914 txg_wait_synced(spa->spa_dsl_pool, txg); 915 916 if (vd != NULL) { 917 ASSERT(!vd->vdev_detached || vd->vdev_dtl_smo.smo_object == 0); 918 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER); 919 vdev_free(vd); 920 spa_config_exit(spa, SCL_ALL, spa); 921 } 922 923 /* 924 * If the config changed, update the config cache. 925 */ 926 if (config_changed) 927 spa_config_sync(spa, B_FALSE, B_TRUE); 928 } 929 930 /* 931 * Unlock the spa_t after adding or removing a vdev. Besides undoing the 932 * locking of spa_vdev_enter(), we also want make sure the transactions have 933 * synced to disk, and then update the global configuration cache with the new 934 * information. 935 */ 936 int 937 spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error) 938 { 939 spa_vdev_config_exit(spa, vd, txg, error, FTAG); 940 mutex_exit(&spa_namespace_lock); 941 mutex_exit(&spa->spa_vdev_top_lock); 942 943 return (error); 944 } 945 946 /* 947 * Lock the given spa_t for the purpose of changing vdev state. 948 */ 949 void 950 spa_vdev_state_enter(spa_t *spa, int oplocks) 951 { 952 int locks = SCL_STATE_ALL | oplocks; 953 954 spa_config_enter(spa, locks, spa, RW_WRITER); 955 spa->spa_vdev_locks = locks; 956 } 957 958 int 959 spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error) 960 { 961 if (vd != NULL || error == 0) 962 vdev_dtl_reassess(vd ? vd->vdev_top : spa->spa_root_vdev, 963 0, 0, B_FALSE); 964 965 if (vd != NULL) { 966 vdev_state_dirty(vd->vdev_top); 967 spa->spa_config_generation++; 968 } 969 970 ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL); 971 spa_config_exit(spa, spa->spa_vdev_locks, spa); 972 973 /* 974 * If anything changed, wait for it to sync. This ensures that, 975 * from the system administrator's perspective, zpool(1M) commands 976 * are synchronous. This is important for things like zpool offline: 977 * when the command completes, you expect no further I/O from ZFS. 978 */ 979 if (vd != NULL) 980 txg_wait_synced(spa->spa_dsl_pool, 0); 981 982 return (error); 983 } 984 985 /* 986 * ========================================================================== 987 * Miscellaneous functions 988 * ========================================================================== 989 */ 990 991 /* 992 * Rename a spa_t. 993 */ 994 int 995 spa_rename(const char *name, const char *newname) 996 { 997 spa_t *spa; 998 int err; 999 1000 /* 1001 * Lookup the spa_t and grab the config lock for writing. We need to 1002 * actually open the pool so that we can sync out the necessary labels. 1003 * It's OK to call spa_open() with the namespace lock held because we 1004 * allow recursive calls for other reasons. 1005 */ 1006 mutex_enter(&spa_namespace_lock); 1007 if ((err = spa_open(name, &spa, FTAG)) != 0) { 1008 mutex_exit(&spa_namespace_lock); 1009 return (err); 1010 } 1011 1012 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1013 1014 avl_remove(&spa_namespace_avl, spa); 1015 (void) strlcpy(spa->spa_name, newname, sizeof (spa->spa_name)); 1016 avl_add(&spa_namespace_avl, spa); 1017 1018 /* 1019 * Sync all labels to disk with the new names by marking the root vdev 1020 * dirty and waiting for it to sync. It will pick up the new pool name 1021 * during the sync. 1022 */ 1023 vdev_config_dirty(spa->spa_root_vdev); 1024 1025 spa_config_exit(spa, SCL_ALL, FTAG); 1026 1027 txg_wait_synced(spa->spa_dsl_pool, 0); 1028 1029 /* 1030 * Sync the updated config cache. 1031 */ 1032 spa_config_sync(spa, B_FALSE, B_TRUE); 1033 1034 spa_close(spa, FTAG); 1035 1036 mutex_exit(&spa_namespace_lock); 1037 1038 return (0); 1039 } 1040 1041 1042 /* 1043 * Determine whether a pool with given pool_guid exists. If device_guid is 1044 * non-zero, determine whether the pool exists *and* contains a device with the 1045 * specified device_guid. 1046 */ 1047 boolean_t 1048 spa_guid_exists(uint64_t pool_guid, uint64_t device_guid) 1049 { 1050 spa_t *spa; 1051 avl_tree_t *t = &spa_namespace_avl; 1052 1053 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1054 1055 for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) { 1056 if (spa->spa_state == POOL_STATE_UNINITIALIZED) 1057 continue; 1058 if (spa->spa_root_vdev == NULL) 1059 continue; 1060 if (spa_guid(spa) == pool_guid) { 1061 if (device_guid == 0) 1062 break; 1063 1064 if (vdev_lookup_by_guid(spa->spa_root_vdev, 1065 device_guid) != NULL) 1066 break; 1067 1068 /* 1069 * Check any devices we may be in the process of adding. 1070 */ 1071 if (spa->spa_pending_vdev) { 1072 if (vdev_lookup_by_guid(spa->spa_pending_vdev, 1073 device_guid) != NULL) 1074 break; 1075 } 1076 } 1077 } 1078 1079 return (spa != NULL); 1080 } 1081 1082 char * 1083 spa_strdup(const char *s) 1084 { 1085 size_t len; 1086 char *new; 1087 1088 len = strlen(s); 1089 new = kmem_alloc(len + 1, KM_SLEEP); 1090 bcopy(s, new, len); 1091 new[len] = '\0'; 1092 1093 return (new); 1094 } 1095 1096 void 1097 spa_strfree(char *s) 1098 { 1099 kmem_free(s, strlen(s) + 1); 1100 } 1101 1102 uint64_t 1103 spa_get_random(uint64_t range) 1104 { 1105 uint64_t r; 1106 1107 ASSERT(range != 0); 1108 1109 (void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t)); 1110 1111 return (r % range); 1112 } 1113 1114 void 1115 sprintf_blkptr(char *buf, const blkptr_t *bp) 1116 { 1117 char *type = dmu_ot[BP_GET_TYPE(bp)].ot_name; 1118 char *checksum = zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name; 1119 char *compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name; 1120 1121 SPRINTF_BLKPTR(snprintf, ' ', buf, bp, type, checksum, compress); 1122 } 1123 1124 void 1125 spa_freeze(spa_t *spa) 1126 { 1127 uint64_t freeze_txg = 0; 1128 1129 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1130 if (spa->spa_freeze_txg == UINT64_MAX) { 1131 freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE; 1132 spa->spa_freeze_txg = freeze_txg; 1133 } 1134 spa_config_exit(spa, SCL_ALL, FTAG); 1135 if (freeze_txg != 0) 1136 txg_wait_synced(spa_get_dsl(spa), freeze_txg); 1137 } 1138 1139 void 1140 zfs_panic_recover(const char *fmt, ...) 1141 { 1142 va_list adx; 1143 1144 va_start(adx, fmt); 1145 vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx); 1146 va_end(adx); 1147 } 1148 1149 /* 1150 * ========================================================================== 1151 * Accessor functions 1152 * ========================================================================== 1153 */ 1154 1155 boolean_t 1156 spa_shutting_down(spa_t *spa) 1157 { 1158 return (spa->spa_async_suspended); 1159 } 1160 1161 dsl_pool_t * 1162 spa_get_dsl(spa_t *spa) 1163 { 1164 return (spa->spa_dsl_pool); 1165 } 1166 1167 blkptr_t * 1168 spa_get_rootblkptr(spa_t *spa) 1169 { 1170 return (&spa->spa_ubsync.ub_rootbp); 1171 } 1172 1173 void 1174 spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp) 1175 { 1176 spa->spa_uberblock.ub_rootbp = *bp; 1177 } 1178 1179 void 1180 spa_altroot(spa_t *spa, char *buf, size_t buflen) 1181 { 1182 if (spa->spa_root == NULL) 1183 buf[0] = '\0'; 1184 else 1185 (void) strncpy(buf, spa->spa_root, buflen); 1186 } 1187 1188 int 1189 spa_sync_pass(spa_t *spa) 1190 { 1191 return (spa->spa_sync_pass); 1192 } 1193 1194 char * 1195 spa_name(spa_t *spa) 1196 { 1197 return (spa->spa_name); 1198 } 1199 1200 uint64_t 1201 spa_guid(spa_t *spa) 1202 { 1203 /* 1204 * If we fail to parse the config during spa_load(), we can go through 1205 * the error path (which posts an ereport) and end up here with no root 1206 * vdev. We stash the original pool guid in 'spa_load_guid' to handle 1207 * this case. 1208 */ 1209 if (spa->spa_root_vdev != NULL) 1210 return (spa->spa_root_vdev->vdev_guid); 1211 else 1212 return (spa->spa_load_guid); 1213 } 1214 1215 uint64_t 1216 spa_last_synced_txg(spa_t *spa) 1217 { 1218 return (spa->spa_ubsync.ub_txg); 1219 } 1220 1221 uint64_t 1222 spa_first_txg(spa_t *spa) 1223 { 1224 return (spa->spa_first_txg); 1225 } 1226 1227 uint64_t 1228 spa_syncing_txg(spa_t *spa) 1229 { 1230 return (spa->spa_syncing_txg); 1231 } 1232 1233 pool_state_t 1234 spa_state(spa_t *spa) 1235 { 1236 return (spa->spa_state); 1237 } 1238 1239 spa_load_state_t 1240 spa_load_state(spa_t *spa) 1241 { 1242 return (spa->spa_load_state); 1243 } 1244 1245 uint64_t 1246 spa_freeze_txg(spa_t *spa) 1247 { 1248 return (spa->spa_freeze_txg); 1249 } 1250 1251 /* ARGSUSED */ 1252 uint64_t 1253 spa_get_asize(spa_t *spa, uint64_t lsize) 1254 { 1255 /* 1256 * The worst case is single-sector max-parity RAID-Z blocks, in which 1257 * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1) 1258 * times the size; so just assume that. Add to this the fact that 1259 * we can have up to 3 DVAs per bp, and one more factor of 2 because 1260 * the block may be dittoed with up to 3 DVAs by ddt_sync(). 1261 */ 1262 return (lsize * (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2); 1263 } 1264 1265 uint64_t 1266 spa_get_dspace(spa_t *spa) 1267 { 1268 return (spa->spa_dspace); 1269 } 1270 1271 void 1272 spa_update_dspace(spa_t *spa) 1273 { 1274 spa->spa_dspace = metaslab_class_get_dspace(spa_normal_class(spa)) + 1275 ddt_get_dedup_dspace(spa); 1276 } 1277 1278 /* 1279 * Return the failure mode that has been set to this pool. The default 1280 * behavior will be to block all I/Os when a complete failure occurs. 1281 */ 1282 uint8_t 1283 spa_get_failmode(spa_t *spa) 1284 { 1285 return (spa->spa_failmode); 1286 } 1287 1288 boolean_t 1289 spa_suspended(spa_t *spa) 1290 { 1291 return (spa->spa_suspended); 1292 } 1293 1294 uint64_t 1295 spa_version(spa_t *spa) 1296 { 1297 return (spa->spa_ubsync.ub_version); 1298 } 1299 1300 boolean_t 1301 spa_deflate(spa_t *spa) 1302 { 1303 return (spa->spa_deflate); 1304 } 1305 1306 metaslab_class_t * 1307 spa_normal_class(spa_t *spa) 1308 { 1309 return (spa->spa_normal_class); 1310 } 1311 1312 metaslab_class_t * 1313 spa_log_class(spa_t *spa) 1314 { 1315 return (spa->spa_log_class); 1316 } 1317 1318 int 1319 spa_max_replication(spa_t *spa) 1320 { 1321 /* 1322 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to 1323 * handle BPs with more than one DVA allocated. Set our max 1324 * replication level accordingly. 1325 */ 1326 if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS) 1327 return (1); 1328 return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override)); 1329 } 1330 1331 uint64_t 1332 dva_get_dsize_sync(spa_t *spa, const dva_t *dva) 1333 { 1334 uint64_t asize = DVA_GET_ASIZE(dva); 1335 uint64_t dsize = asize; 1336 1337 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 1338 1339 if (asize != 0 && spa->spa_deflate) { 1340 vdev_t *vd = vdev_lookup_top(spa, DVA_GET_VDEV(dva)); 1341 dsize = (asize >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio; 1342 } 1343 1344 return (dsize); 1345 } 1346 1347 uint64_t 1348 bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp) 1349 { 1350 uint64_t dsize = 0; 1351 1352 for (int d = 0; d < SPA_DVAS_PER_BP; d++) 1353 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]); 1354 1355 return (dsize); 1356 } 1357 1358 uint64_t 1359 bp_get_dsize(spa_t *spa, const blkptr_t *bp) 1360 { 1361 uint64_t dsize = 0; 1362 1363 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); 1364 1365 for (int d = 0; d < SPA_DVAS_PER_BP; d++) 1366 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]); 1367 1368 spa_config_exit(spa, SCL_VDEV, FTAG); 1369 1370 return (dsize); 1371 } 1372 1373 /* 1374 * ========================================================================== 1375 * Initialization and Termination 1376 * ========================================================================== 1377 */ 1378 1379 static int 1380 spa_name_compare(const void *a1, const void *a2) 1381 { 1382 const spa_t *s1 = a1; 1383 const spa_t *s2 = a2; 1384 int s; 1385 1386 s = strcmp(s1->spa_name, s2->spa_name); 1387 if (s > 0) 1388 return (1); 1389 if (s < 0) 1390 return (-1); 1391 return (0); 1392 } 1393 1394 int 1395 spa_busy(void) 1396 { 1397 return (spa_active_count); 1398 } 1399 1400 void 1401 spa_boot_init() 1402 { 1403 spa_config_load(); 1404 } 1405 1406 void 1407 spa_init(int mode) 1408 { 1409 mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL); 1410 mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL); 1411 mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL); 1412 cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL); 1413 1414 avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t), 1415 offsetof(spa_t, spa_avl)); 1416 1417 avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t), 1418 offsetof(spa_aux_t, aux_avl)); 1419 1420 avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t), 1421 offsetof(spa_aux_t, aux_avl)); 1422 1423 spa_mode_global = mode; 1424 1425 refcount_init(); 1426 unique_init(); 1427 zio_init(); 1428 dmu_init(); 1429 zil_init(); 1430 vdev_cache_stat_init(); 1431 zfs_prop_init(); 1432 zpool_prop_init(); 1433 spa_config_load(); 1434 l2arc_start(); 1435 } 1436 1437 void 1438 spa_fini(void) 1439 { 1440 l2arc_stop(); 1441 1442 spa_evict_all(); 1443 1444 vdev_cache_stat_fini(); 1445 zil_fini(); 1446 dmu_fini(); 1447 zio_fini(); 1448 unique_fini(); 1449 refcount_fini(); 1450 1451 avl_destroy(&spa_namespace_avl); 1452 avl_destroy(&spa_spare_avl); 1453 avl_destroy(&spa_l2cache_avl); 1454 1455 cv_destroy(&spa_namespace_cv); 1456 mutex_destroy(&spa_namespace_lock); 1457 mutex_destroy(&spa_spare_lock); 1458 mutex_destroy(&spa_l2cache_lock); 1459 } 1460 1461 /* 1462 * Return whether this pool has slogs. No locking needed. 1463 * It's not a problem if the wrong answer is returned as it's only for 1464 * performance and not correctness 1465 */ 1466 boolean_t 1467 spa_has_slogs(spa_t *spa) 1468 { 1469 return (spa->spa_log_class->mc_rotor != NULL); 1470 } 1471 1472 spa_log_state_t 1473 spa_get_log_state(spa_t *spa) 1474 { 1475 return (spa->spa_log_state); 1476 } 1477 1478 void 1479 spa_set_log_state(spa_t *spa, spa_log_state_t state) 1480 { 1481 spa->spa_log_state = state; 1482 } 1483 1484 boolean_t 1485 spa_is_root(spa_t *spa) 1486 { 1487 return (spa->spa_is_root); 1488 } 1489 1490 boolean_t 1491 spa_writeable(spa_t *spa) 1492 { 1493 return (!!(spa->spa_mode & FWRITE)); 1494 } 1495 1496 int 1497 spa_mode(spa_t *spa) 1498 { 1499 return (spa->spa_mode); 1500 } 1501 1502 uint64_t 1503 spa_bootfs(spa_t *spa) 1504 { 1505 return (spa->spa_bootfs); 1506 } 1507 1508 uint64_t 1509 spa_delegation(spa_t *spa) 1510 { 1511 return (spa->spa_delegation); 1512 } 1513 1514 objset_t * 1515 spa_meta_objset(spa_t *spa) 1516 { 1517 return (spa->spa_meta_objset); 1518 } 1519 1520 enum zio_checksum 1521 spa_dedup_checksum(spa_t *spa) 1522 { 1523 return (spa->spa_dedup_checksum); 1524 } 1525