spa_misc.c revision 269006
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 (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23 * Copyright (c) 2013 by Delphix. All rights reserved. 24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved. 25 * Copyright 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved. 26 */ 27 28#include <sys/zfs_context.h> 29#include <sys/spa_impl.h> 30#include <sys/spa_boot.h> 31#include <sys/zio.h> 32#include <sys/zio_checksum.h> 33#include <sys/zio_compress.h> 34#include <sys/dmu.h> 35#include <sys/dmu_tx.h> 36#include <sys/zap.h> 37#include <sys/zil.h> 38#include <sys/vdev_impl.h> 39#include <sys/metaslab.h> 40#include <sys/uberblock_impl.h> 41#include <sys/txg.h> 42#include <sys/avl.h> 43#include <sys/unique.h> 44#include <sys/dsl_pool.h> 45#include <sys/dsl_dir.h> 46#include <sys/dsl_prop.h> 47#include <sys/dsl_scan.h> 48#include <sys/fs/zfs.h> 49#include <sys/metaslab_impl.h> 50#include <sys/arc.h> 51#include <sys/ddt.h> 52#include "zfs_prop.h" 53#include "zfeature_common.h" 54 55/* 56 * SPA locking 57 * 58 * There are four basic locks for managing spa_t structures: 59 * 60 * spa_namespace_lock (global mutex) 61 * 62 * This lock must be acquired to do any of the following: 63 * 64 * - Lookup a spa_t by name 65 * - Add or remove a spa_t from the namespace 66 * - Increase spa_refcount from non-zero 67 * - Check if spa_refcount is zero 68 * - Rename a spa_t 69 * - add/remove/attach/detach devices 70 * - Held for the duration of create/destroy/import/export 71 * 72 * It does not need to handle recursion. A create or destroy may 73 * reference objects (files or zvols) in other pools, but by 74 * definition they must have an existing reference, and will never need 75 * to lookup a spa_t by name. 76 * 77 * spa_refcount (per-spa refcount_t protected by mutex) 78 * 79 * This reference count keep track of any active users of the spa_t. The 80 * spa_t cannot be destroyed or freed while this is non-zero. Internally, 81 * the refcount is never really 'zero' - opening a pool implicitly keeps 82 * some references in the DMU. Internally we check against spa_minref, but 83 * present the image of a zero/non-zero value to consumers. 84 * 85 * spa_config_lock[] (per-spa array of rwlocks) 86 * 87 * This protects the spa_t from config changes, and must be held in 88 * the following circumstances: 89 * 90 * - RW_READER to perform I/O to the spa 91 * - RW_WRITER to change the vdev config 92 * 93 * The locking order is fairly straightforward: 94 * 95 * spa_namespace_lock -> spa_refcount 96 * 97 * The namespace lock must be acquired to increase the refcount from 0 98 * or to check if it is zero. 99 * 100 * spa_refcount -> spa_config_lock[] 101 * 102 * There must be at least one valid reference on the spa_t to acquire 103 * the config lock. 104 * 105 * spa_namespace_lock -> spa_config_lock[] 106 * 107 * The namespace lock must always be taken before the config lock. 108 * 109 * 110 * The spa_namespace_lock can be acquired directly and is globally visible. 111 * 112 * The namespace is manipulated using the following functions, all of which 113 * require the spa_namespace_lock to be held. 114 * 115 * spa_lookup() Lookup a spa_t by name. 116 * 117 * spa_add() Create a new spa_t in the namespace. 118 * 119 * spa_remove() Remove a spa_t from the namespace. This also 120 * frees up any memory associated with the spa_t. 121 * 122 * spa_next() Returns the next spa_t in the system, or the 123 * first if NULL is passed. 124 * 125 * spa_evict_all() Shutdown and remove all spa_t structures in 126 * the system. 127 * 128 * spa_guid_exists() Determine whether a pool/device guid exists. 129 * 130 * The spa_refcount is manipulated using the following functions: 131 * 132 * spa_open_ref() Adds a reference to the given spa_t. Must be 133 * called with spa_namespace_lock held if the 134 * refcount is currently zero. 135 * 136 * spa_close() Remove a reference from the spa_t. This will 137 * not free the spa_t or remove it from the 138 * namespace. No locking is required. 139 * 140 * spa_refcount_zero() Returns true if the refcount is currently 141 * zero. Must be called with spa_namespace_lock 142 * held. 143 * 144 * The spa_config_lock[] is an array of rwlocks, ordered as follows: 145 * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV. 146 * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}(). 147 * 148 * To read the configuration, it suffices to hold one of these locks as reader. 149 * To modify the configuration, you must hold all locks as writer. To modify 150 * vdev state without altering the vdev tree's topology (e.g. online/offline), 151 * you must hold SCL_STATE and SCL_ZIO as writer. 152 * 153 * We use these distinct config locks to avoid recursive lock entry. 154 * For example, spa_sync() (which holds SCL_CONFIG as reader) induces 155 * block allocations (SCL_ALLOC), which may require reading space maps 156 * from disk (dmu_read() -> zio_read() -> SCL_ZIO). 157 * 158 * The spa config locks cannot be normal rwlocks because we need the 159 * ability to hand off ownership. For example, SCL_ZIO is acquired 160 * by the issuing thread and later released by an interrupt thread. 161 * They do, however, obey the usual write-wanted semantics to prevent 162 * writer (i.e. system administrator) starvation. 163 * 164 * The lock acquisition rules are as follows: 165 * 166 * SCL_CONFIG 167 * Protects changes to the vdev tree topology, such as vdev 168 * add/remove/attach/detach. Protects the dirty config list 169 * (spa_config_dirty_list) and the set of spares and l2arc devices. 170 * 171 * SCL_STATE 172 * Protects changes to pool state and vdev state, such as vdev 173 * online/offline/fault/degrade/clear. Protects the dirty state list 174 * (spa_state_dirty_list) and global pool state (spa_state). 175 * 176 * SCL_ALLOC 177 * Protects changes to metaslab groups and classes. 178 * Held as reader by metaslab_alloc() and metaslab_claim(). 179 * 180 * SCL_ZIO 181 * Held by bp-level zios (those which have no io_vd upon entry) 182 * to prevent changes to the vdev tree. The bp-level zio implicitly 183 * protects all of its vdev child zios, which do not hold SCL_ZIO. 184 * 185 * SCL_FREE 186 * Protects changes to metaslab groups and classes. 187 * Held as reader by metaslab_free(). SCL_FREE is distinct from 188 * SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free 189 * blocks in zio_done() while another i/o that holds either 190 * SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete. 191 * 192 * SCL_VDEV 193 * Held as reader to prevent changes to the vdev tree during trivial 194 * inquiries such as bp_get_dsize(). SCL_VDEV is distinct from the 195 * other locks, and lower than all of them, to ensure that it's safe 196 * to acquire regardless of caller context. 197 * 198 * In addition, the following rules apply: 199 * 200 * (a) spa_props_lock protects pool properties, spa_config and spa_config_list. 201 * The lock ordering is SCL_CONFIG > spa_props_lock. 202 * 203 * (b) I/O operations on leaf vdevs. For any zio operation that takes 204 * an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(), 205 * or zio_write_phys() -- the caller must ensure that the config cannot 206 * cannot change in the interim, and that the vdev cannot be reopened. 207 * SCL_STATE as reader suffices for both. 208 * 209 * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit(). 210 * 211 * spa_vdev_enter() Acquire the namespace lock and the config lock 212 * for writing. 213 * 214 * spa_vdev_exit() Release the config lock, wait for all I/O 215 * to complete, sync the updated configs to the 216 * cache, and release the namespace lock. 217 * 218 * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit(). 219 * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual 220 * locking is, always, based on spa_namespace_lock and spa_config_lock[]. 221 * 222 * spa_rename() is also implemented within this file since it requires 223 * manipulation of the namespace. 224 */ 225 226static avl_tree_t spa_namespace_avl; 227kmutex_t spa_namespace_lock; 228static kcondvar_t spa_namespace_cv; 229static int spa_active_count; 230int spa_max_replication_override = SPA_DVAS_PER_BP; 231 232static kmutex_t spa_spare_lock; 233static avl_tree_t spa_spare_avl; 234static kmutex_t spa_l2cache_lock; 235static avl_tree_t spa_l2cache_avl; 236 237kmem_cache_t *spa_buffer_pool; 238int spa_mode_global; 239 240#ifdef ZFS_DEBUG 241/* Everything except dprintf and spa is on by default in debug builds */ 242int zfs_flags = ~(ZFS_DEBUG_DPRINTF | ZFS_DEBUG_SPA); 243#else 244int zfs_flags = 0; 245#endif 246SYSCTL_DECL(_debug); 247TUNABLE_INT("debug.zfs_flags", &zfs_flags); 248SYSCTL_INT(_debug, OID_AUTO, zfs_flags, CTLFLAG_RWTUN, &zfs_flags, 0, 249 "ZFS debug flags."); 250 251/* 252 * zfs_recover can be set to nonzero to attempt to recover from 253 * otherwise-fatal errors, typically caused by on-disk corruption. When 254 * set, calls to zfs_panic_recover() will turn into warning messages. 255 * This should only be used as a last resort, as it typically results 256 * in leaked space, or worse. 257 */ 258boolean_t zfs_recover = B_FALSE; 259SYSCTL_DECL(_vfs_zfs); 260TUNABLE_INT("vfs.zfs.recover", &zfs_recover); 261SYSCTL_INT(_vfs_zfs, OID_AUTO, recover, CTLFLAG_RDTUN, &zfs_recover, 0, 262 "Try to recover from otherwise-fatal errors."); 263 264/* 265 * If destroy encounters an EIO while reading metadata (e.g. indirect 266 * blocks), space referenced by the missing metadata can not be freed. 267 * Normally this causes the background destroy to become "stalled", as 268 * it is unable to make forward progress. While in this stalled state, 269 * all remaining space to free from the error-encountering filesystem is 270 * "temporarily leaked". Set this flag to cause it to ignore the EIO, 271 * permanently leak the space from indirect blocks that can not be read, 272 * and continue to free everything else that it can. 273 * 274 * The default, "stalling" behavior is useful if the storage partially 275 * fails (i.e. some but not all i/os fail), and then later recovers. In 276 * this case, we will be able to continue pool operations while it is 277 * partially failed, and when it recovers, we can continue to free the 278 * space, with no leaks. However, note that this case is actually 279 * fairly rare. 280 * 281 * Typically pools either (a) fail completely (but perhaps temporarily, 282 * e.g. a top-level vdev going offline), or (b) have localized, 283 * permanent errors (e.g. disk returns the wrong data due to bit flip or 284 * firmware bug). In case (a), this setting does not matter because the 285 * pool will be suspended and the sync thread will not be able to make 286 * forward progress regardless. In case (b), because the error is 287 * permanent, the best we can do is leak the minimum amount of space, 288 * which is what setting this flag will do. Therefore, it is reasonable 289 * for this flag to normally be set, but we chose the more conservative 290 * approach of not setting it, so that there is no possibility of 291 * leaking space in the "partial temporary" failure case. 292 */ 293boolean_t zfs_free_leak_on_eio = B_FALSE; 294 295/* 296 * Expiration time in milliseconds. This value has two meanings. First it is 297 * used to determine when the spa_deadman() logic should fire. By default the 298 * spa_deadman() will fire if spa_sync() has not completed in 1000 seconds. 299 * Secondly, the value determines if an I/O is considered "hung". Any I/O that 300 * has not completed in zfs_deadman_synctime_ms is considered "hung" resulting 301 * in a system panic. 302 */ 303uint64_t zfs_deadman_synctime_ms = 1000000ULL; 304TUNABLE_QUAD("vfs.zfs.deadman_synctime_ms", &zfs_deadman_synctime_ms); 305SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, deadman_synctime_ms, CTLFLAG_RDTUN, 306 &zfs_deadman_synctime_ms, 0, 307 "Stalled ZFS I/O expiration time in milliseconds"); 308 309/* 310 * Check time in milliseconds. This defines the frequency at which we check 311 * for hung I/O. 312 */ 313uint64_t zfs_deadman_checktime_ms = 5000ULL; 314TUNABLE_QUAD("vfs.zfs.deadman_checktime_ms", &zfs_deadman_checktime_ms); 315SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, deadman_checktime_ms, CTLFLAG_RDTUN, 316 &zfs_deadman_checktime_ms, 0, 317 "Period of checks for stalled ZFS I/O in milliseconds"); 318 319/* 320 * Default value of -1 for zfs_deadman_enabled is resolved in 321 * zfs_deadman_init() 322 */ 323int zfs_deadman_enabled = -1; 324TUNABLE_INT("vfs.zfs.deadman_enabled", &zfs_deadman_enabled); 325SYSCTL_INT(_vfs_zfs, OID_AUTO, deadman_enabled, CTLFLAG_RDTUN, 326 &zfs_deadman_enabled, 0, "Kernel panic on stalled ZFS I/O"); 327 328/* 329 * The worst case is single-sector max-parity RAID-Z blocks, in which 330 * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1) 331 * times the size; so just assume that. Add to this the fact that 332 * we can have up to 3 DVAs per bp, and one more factor of 2 because 333 * the block may be dittoed with up to 3 DVAs by ddt_sync(). All together, 334 * the worst case is: 335 * (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2 == 24 336 */ 337int spa_asize_inflation = 24; 338TUNABLE_INT("vfs.zfs.spa_asize_inflation", &spa_asize_inflation); 339SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_asize_inflation, CTLFLAG_RWTUN, 340 &spa_asize_inflation, 0, "Worst case inflation factor for single sector writes"); 341 342#ifndef illumos 343#ifdef _KERNEL 344static void 345zfs_deadman_init() 346{ 347 /* 348 * If we are not i386 or amd64 or in a virtual machine, 349 * disable ZFS deadman thread by default 350 */ 351 if (zfs_deadman_enabled == -1) { 352#if defined(__amd64__) || defined(__i386__) 353 zfs_deadman_enabled = (vm_guest == VM_GUEST_NO) ? 1 : 0; 354#else 355 zfs_deadman_enabled = 0; 356#endif 357 } 358} 359#endif /* _KERNEL */ 360#endif /* !illumos */ 361 362/* 363 * Normally, we don't allow the last 3.2% (1/(2^spa_slop_shift)) of space in 364 * the pool to be consumed. This ensures that we don't run the pool 365 * completely out of space, due to unaccounted changes (e.g. to the MOS). 366 * It also limits the worst-case time to allocate space. If we have 367 * less than this amount of free space, most ZPL operations (e.g. write, 368 * create) will return ENOSPC. 369 * 370 * Certain operations (e.g. file removal, most administrative actions) can 371 * use half the slop space. They will only return ENOSPC if less than half 372 * the slop space is free. Typically, once the pool has less than the slop 373 * space free, the user will use these operations to free up space in the pool. 374 * These are the operations that call dsl_pool_adjustedsize() with the netfree 375 * argument set to TRUE. 376 * 377 * A very restricted set of operations are always permitted, regardless of 378 * the amount of free space. These are the operations that call 379 * dsl_sync_task(ZFS_SPACE_CHECK_NONE), e.g. "zfs destroy". If these 380 * operations result in a net increase in the amount of space used, 381 * it is possible to run the pool completely out of space, causing it to 382 * be permanently read-only. 383 * 384 * See also the comments in zfs_space_check_t. 385 */ 386int spa_slop_shift = 5; 387 388/* 389 * ========================================================================== 390 * SPA config locking 391 * ========================================================================== 392 */ 393static void 394spa_config_lock_init(spa_t *spa) 395{ 396 for (int i = 0; i < SCL_LOCKS; i++) { 397 spa_config_lock_t *scl = &spa->spa_config_lock[i]; 398 mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL); 399 cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL); 400 refcount_create_untracked(&scl->scl_count); 401 scl->scl_writer = NULL; 402 scl->scl_write_wanted = 0; 403 } 404} 405 406static void 407spa_config_lock_destroy(spa_t *spa) 408{ 409 for (int i = 0; i < SCL_LOCKS; i++) { 410 spa_config_lock_t *scl = &spa->spa_config_lock[i]; 411 mutex_destroy(&scl->scl_lock); 412 cv_destroy(&scl->scl_cv); 413 refcount_destroy(&scl->scl_count); 414 ASSERT(scl->scl_writer == NULL); 415 ASSERT(scl->scl_write_wanted == 0); 416 } 417} 418 419int 420spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw) 421{ 422 for (int i = 0; i < SCL_LOCKS; i++) { 423 spa_config_lock_t *scl = &spa->spa_config_lock[i]; 424 if (!(locks & (1 << i))) 425 continue; 426 mutex_enter(&scl->scl_lock); 427 if (rw == RW_READER) { 428 if (scl->scl_writer || scl->scl_write_wanted) { 429 mutex_exit(&scl->scl_lock); 430 spa_config_exit(spa, locks ^ (1 << i), tag); 431 return (0); 432 } 433 } else { 434 ASSERT(scl->scl_writer != curthread); 435 if (!refcount_is_zero(&scl->scl_count)) { 436 mutex_exit(&scl->scl_lock); 437 spa_config_exit(spa, locks ^ (1 << i), tag); 438 return (0); 439 } 440 scl->scl_writer = curthread; 441 } 442 (void) refcount_add(&scl->scl_count, tag); 443 mutex_exit(&scl->scl_lock); 444 } 445 return (1); 446} 447 448void 449spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw) 450{ 451 int wlocks_held = 0; 452 453 ASSERT3U(SCL_LOCKS, <, sizeof (wlocks_held) * NBBY); 454 455 for (int i = 0; i < SCL_LOCKS; i++) { 456 spa_config_lock_t *scl = &spa->spa_config_lock[i]; 457 if (scl->scl_writer == curthread) 458 wlocks_held |= (1 << i); 459 if (!(locks & (1 << i))) 460 continue; 461 mutex_enter(&scl->scl_lock); 462 if (rw == RW_READER) { 463 while (scl->scl_writer || scl->scl_write_wanted) { 464 cv_wait(&scl->scl_cv, &scl->scl_lock); 465 } 466 } else { 467 ASSERT(scl->scl_writer != curthread); 468 while (!refcount_is_zero(&scl->scl_count)) { 469 scl->scl_write_wanted++; 470 cv_wait(&scl->scl_cv, &scl->scl_lock); 471 scl->scl_write_wanted--; 472 } 473 scl->scl_writer = curthread; 474 } 475 (void) refcount_add(&scl->scl_count, tag); 476 mutex_exit(&scl->scl_lock); 477 } 478 ASSERT(wlocks_held <= locks); 479} 480 481void 482spa_config_exit(spa_t *spa, int locks, void *tag) 483{ 484 for (int i = SCL_LOCKS - 1; i >= 0; i--) { 485 spa_config_lock_t *scl = &spa->spa_config_lock[i]; 486 if (!(locks & (1 << i))) 487 continue; 488 mutex_enter(&scl->scl_lock); 489 ASSERT(!refcount_is_zero(&scl->scl_count)); 490 if (refcount_remove(&scl->scl_count, tag) == 0) { 491 ASSERT(scl->scl_writer == NULL || 492 scl->scl_writer == curthread); 493 scl->scl_writer = NULL; /* OK in either case */ 494 cv_broadcast(&scl->scl_cv); 495 } 496 mutex_exit(&scl->scl_lock); 497 } 498} 499 500int 501spa_config_held(spa_t *spa, int locks, krw_t rw) 502{ 503 int locks_held = 0; 504 505 for (int i = 0; i < SCL_LOCKS; i++) { 506 spa_config_lock_t *scl = &spa->spa_config_lock[i]; 507 if (!(locks & (1 << i))) 508 continue; 509 if ((rw == RW_READER && !refcount_is_zero(&scl->scl_count)) || 510 (rw == RW_WRITER && scl->scl_writer == curthread)) 511 locks_held |= 1 << i; 512 } 513 514 return (locks_held); 515} 516 517/* 518 * ========================================================================== 519 * SPA namespace functions 520 * ========================================================================== 521 */ 522 523/* 524 * Lookup the named spa_t in the AVL tree. The spa_namespace_lock must be held. 525 * Returns NULL if no matching spa_t is found. 526 */ 527spa_t * 528spa_lookup(const char *name) 529{ 530 static spa_t search; /* spa_t is large; don't allocate on stack */ 531 spa_t *spa; 532 avl_index_t where; 533 char *cp; 534 535 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 536 537 (void) strlcpy(search.spa_name, name, sizeof (search.spa_name)); 538 539 /* 540 * If it's a full dataset name, figure out the pool name and 541 * just use that. 542 */ 543 cp = strpbrk(search.spa_name, "/@#"); 544 if (cp != NULL) 545 *cp = '\0'; 546 547 spa = avl_find(&spa_namespace_avl, &search, &where); 548 549 return (spa); 550} 551 552/* 553 * Fires when spa_sync has not completed within zfs_deadman_synctime_ms. 554 * If the zfs_deadman_enabled flag is set then it inspects all vdev queues 555 * looking for potentially hung I/Os. 556 */ 557void 558spa_deadman(void *arg) 559{ 560 spa_t *spa = arg; 561 562 /* 563 * Disable the deadman timer if the pool is suspended. 564 */ 565 if (spa_suspended(spa)) { 566#ifdef illumos 567 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY)); 568#else 569 /* Nothing. just don't schedule any future callouts. */ 570#endif 571 return; 572 } 573 574 zfs_dbgmsg("slow spa_sync: started %llu seconds ago, calls %llu", 575 (gethrtime() - spa->spa_sync_starttime) / NANOSEC, 576 ++spa->spa_deadman_calls); 577 if (zfs_deadman_enabled) 578 vdev_deadman(spa->spa_root_vdev); 579} 580 581/* 582 * Create an uninitialized spa_t with the given name. Requires 583 * spa_namespace_lock. The caller must ensure that the spa_t doesn't already 584 * exist by calling spa_lookup() first. 585 */ 586spa_t * 587spa_add(const char *name, nvlist_t *config, const char *altroot) 588{ 589 spa_t *spa; 590 spa_config_dirent_t *dp; 591#ifdef illumos 592 cyc_handler_t hdlr; 593 cyc_time_t when; 594#endif 595 596 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 597 598 spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP); 599 600 mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL); 601 mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL); 602 mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL); 603 mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL); 604 mutex_init(&spa->spa_proc_lock, NULL, MUTEX_DEFAULT, NULL); 605 mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL); 606 mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL); 607 mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL); 608 mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL); 609 610 cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL); 611 cv_init(&spa->spa_proc_cv, NULL, CV_DEFAULT, NULL); 612 cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL); 613 cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL); 614 615 for (int t = 0; t < TXG_SIZE; t++) 616 bplist_create(&spa->spa_free_bplist[t]); 617 618 (void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name)); 619 spa->spa_state = POOL_STATE_UNINITIALIZED; 620 spa->spa_freeze_txg = UINT64_MAX; 621 spa->spa_final_txg = UINT64_MAX; 622 spa->spa_load_max_txg = UINT64_MAX; 623 spa->spa_proc = &p0; 624 spa->spa_proc_state = SPA_PROC_NONE; 625 626#ifdef illumos 627 hdlr.cyh_func = spa_deadman; 628 hdlr.cyh_arg = spa; 629 hdlr.cyh_level = CY_LOW_LEVEL; 630#endif 631 632 spa->spa_deadman_synctime = MSEC2NSEC(zfs_deadman_synctime_ms); 633 634#ifdef illumos 635 /* 636 * This determines how often we need to check for hung I/Os after 637 * the cyclic has already fired. Since checking for hung I/Os is 638 * an expensive operation we don't want to check too frequently. 639 * Instead wait for 5 seconds before checking again. 640 */ 641 when.cyt_interval = MSEC2NSEC(zfs_deadman_checktime_ms); 642 when.cyt_when = CY_INFINITY; 643 mutex_enter(&cpu_lock); 644 spa->spa_deadman_cycid = cyclic_add(&hdlr, &when); 645 mutex_exit(&cpu_lock); 646#else /* !illumos */ 647#ifdef _KERNEL 648 callout_init(&spa->spa_deadman_cycid, CALLOUT_MPSAFE); 649#endif 650#endif 651 refcount_create(&spa->spa_refcount); 652 spa_config_lock_init(spa); 653 654 avl_add(&spa_namespace_avl, spa); 655 656 /* 657 * Set the alternate root, if there is one. 658 */ 659 if (altroot) { 660 spa->spa_root = spa_strdup(altroot); 661 spa_active_count++; 662 } 663 664 /* 665 * Every pool starts with the default cachefile 666 */ 667 list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t), 668 offsetof(spa_config_dirent_t, scd_link)); 669 670 dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP); 671 dp->scd_path = altroot ? NULL : spa_strdup(spa_config_path); 672 list_insert_head(&spa->spa_config_list, dp); 673 674 VERIFY(nvlist_alloc(&spa->spa_load_info, NV_UNIQUE_NAME, 675 KM_SLEEP) == 0); 676 677 if (config != NULL) { 678 nvlist_t *features; 679 680 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ, 681 &features) == 0) { 682 VERIFY(nvlist_dup(features, &spa->spa_label_features, 683 0) == 0); 684 } 685 686 VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0); 687 } 688 689 if (spa->spa_label_features == NULL) { 690 VERIFY(nvlist_alloc(&spa->spa_label_features, NV_UNIQUE_NAME, 691 KM_SLEEP) == 0); 692 } 693 694 spa->spa_debug = ((zfs_flags & ZFS_DEBUG_SPA) != 0); 695 696 /* 697 * As a pool is being created, treat all features as disabled by 698 * setting SPA_FEATURE_DISABLED for all entries in the feature 699 * refcount cache. 700 */ 701 for (int i = 0; i < SPA_FEATURES; i++) { 702 spa->spa_feat_refcount_cache[i] = SPA_FEATURE_DISABLED; 703 } 704 705 return (spa); 706} 707 708/* 709 * Removes a spa_t from the namespace, freeing up any memory used. Requires 710 * spa_namespace_lock. This is called only after the spa_t has been closed and 711 * deactivated. 712 */ 713void 714spa_remove(spa_t *spa) 715{ 716 spa_config_dirent_t *dp; 717 718 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 719 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); 720 721 nvlist_free(spa->spa_config_splitting); 722 723 avl_remove(&spa_namespace_avl, spa); 724 cv_broadcast(&spa_namespace_cv); 725 726 if (spa->spa_root) { 727 spa_strfree(spa->spa_root); 728 spa_active_count--; 729 } 730 731 while ((dp = list_head(&spa->spa_config_list)) != NULL) { 732 list_remove(&spa->spa_config_list, dp); 733 if (dp->scd_path != NULL) 734 spa_strfree(dp->scd_path); 735 kmem_free(dp, sizeof (spa_config_dirent_t)); 736 } 737 738 list_destroy(&spa->spa_config_list); 739 740 nvlist_free(spa->spa_label_features); 741 nvlist_free(spa->spa_load_info); 742 spa_config_set(spa, NULL); 743 744#ifdef illumos 745 mutex_enter(&cpu_lock); 746 if (spa->spa_deadman_cycid != CYCLIC_NONE) 747 cyclic_remove(spa->spa_deadman_cycid); 748 mutex_exit(&cpu_lock); 749 spa->spa_deadman_cycid = CYCLIC_NONE; 750#else /* !illumos */ 751#ifdef _KERNEL 752 callout_drain(&spa->spa_deadman_cycid); 753#endif 754#endif 755 756 refcount_destroy(&spa->spa_refcount); 757 758 spa_config_lock_destroy(spa); 759 760 for (int t = 0; t < TXG_SIZE; t++) 761 bplist_destroy(&spa->spa_free_bplist[t]); 762 763 cv_destroy(&spa->spa_async_cv); 764 cv_destroy(&spa->spa_proc_cv); 765 cv_destroy(&spa->spa_scrub_io_cv); 766 cv_destroy(&spa->spa_suspend_cv); 767 768 mutex_destroy(&spa->spa_async_lock); 769 mutex_destroy(&spa->spa_errlist_lock); 770 mutex_destroy(&spa->spa_errlog_lock); 771 mutex_destroy(&spa->spa_history_lock); 772 mutex_destroy(&spa->spa_proc_lock); 773 mutex_destroy(&spa->spa_props_lock); 774 mutex_destroy(&spa->spa_scrub_lock); 775 mutex_destroy(&spa->spa_suspend_lock); 776 mutex_destroy(&spa->spa_vdev_top_lock); 777 778 kmem_free(spa, sizeof (spa_t)); 779} 780 781/* 782 * Given a pool, return the next pool in the namespace, or NULL if there is 783 * none. If 'prev' is NULL, return the first pool. 784 */ 785spa_t * 786spa_next(spa_t *prev) 787{ 788 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 789 790 if (prev) 791 return (AVL_NEXT(&spa_namespace_avl, prev)); 792 else 793 return (avl_first(&spa_namespace_avl)); 794} 795 796/* 797 * ========================================================================== 798 * SPA refcount functions 799 * ========================================================================== 800 */ 801 802/* 803 * Add a reference to the given spa_t. Must have at least one reference, or 804 * have the namespace lock held. 805 */ 806void 807spa_open_ref(spa_t *spa, void *tag) 808{ 809 ASSERT(refcount_count(&spa->spa_refcount) >= spa->spa_minref || 810 MUTEX_HELD(&spa_namespace_lock)); 811 (void) refcount_add(&spa->spa_refcount, tag); 812} 813 814/* 815 * Remove a reference to the given spa_t. Must have at least one reference, or 816 * have the namespace lock held. 817 */ 818void 819spa_close(spa_t *spa, void *tag) 820{ 821 ASSERT(refcount_count(&spa->spa_refcount) > spa->spa_minref || 822 MUTEX_HELD(&spa_namespace_lock)); 823 (void) refcount_remove(&spa->spa_refcount, tag); 824} 825 826/* 827 * Check to see if the spa refcount is zero. Must be called with 828 * spa_namespace_lock held. We really compare against spa_minref, which is the 829 * number of references acquired when opening a pool 830 */ 831boolean_t 832spa_refcount_zero(spa_t *spa) 833{ 834 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 835 836 return (refcount_count(&spa->spa_refcount) == spa->spa_minref); 837} 838 839/* 840 * ========================================================================== 841 * SPA spare and l2cache tracking 842 * ========================================================================== 843 */ 844 845/* 846 * Hot spares and cache devices are tracked using the same code below, 847 * for 'auxiliary' devices. 848 */ 849 850typedef struct spa_aux { 851 uint64_t aux_guid; 852 uint64_t aux_pool; 853 avl_node_t aux_avl; 854 int aux_count; 855} spa_aux_t; 856 857static int 858spa_aux_compare(const void *a, const void *b) 859{ 860 const spa_aux_t *sa = a; 861 const spa_aux_t *sb = b; 862 863 if (sa->aux_guid < sb->aux_guid) 864 return (-1); 865 else if (sa->aux_guid > sb->aux_guid) 866 return (1); 867 else 868 return (0); 869} 870 871void 872spa_aux_add(vdev_t *vd, avl_tree_t *avl) 873{ 874 avl_index_t where; 875 spa_aux_t search; 876 spa_aux_t *aux; 877 878 search.aux_guid = vd->vdev_guid; 879 if ((aux = avl_find(avl, &search, &where)) != NULL) { 880 aux->aux_count++; 881 } else { 882 aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP); 883 aux->aux_guid = vd->vdev_guid; 884 aux->aux_count = 1; 885 avl_insert(avl, aux, where); 886 } 887} 888 889void 890spa_aux_remove(vdev_t *vd, avl_tree_t *avl) 891{ 892 spa_aux_t search; 893 spa_aux_t *aux; 894 avl_index_t where; 895 896 search.aux_guid = vd->vdev_guid; 897 aux = avl_find(avl, &search, &where); 898 899 ASSERT(aux != NULL); 900 901 if (--aux->aux_count == 0) { 902 avl_remove(avl, aux); 903 kmem_free(aux, sizeof (spa_aux_t)); 904 } else if (aux->aux_pool == spa_guid(vd->vdev_spa)) { 905 aux->aux_pool = 0ULL; 906 } 907} 908 909boolean_t 910spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl) 911{ 912 spa_aux_t search, *found; 913 914 search.aux_guid = guid; 915 found = avl_find(avl, &search, NULL); 916 917 if (pool) { 918 if (found) 919 *pool = found->aux_pool; 920 else 921 *pool = 0ULL; 922 } 923 924 if (refcnt) { 925 if (found) 926 *refcnt = found->aux_count; 927 else 928 *refcnt = 0; 929 } 930 931 return (found != NULL); 932} 933 934void 935spa_aux_activate(vdev_t *vd, avl_tree_t *avl) 936{ 937 spa_aux_t search, *found; 938 avl_index_t where; 939 940 search.aux_guid = vd->vdev_guid; 941 found = avl_find(avl, &search, &where); 942 ASSERT(found != NULL); 943 ASSERT(found->aux_pool == 0ULL); 944 945 found->aux_pool = spa_guid(vd->vdev_spa); 946} 947 948/* 949 * Spares are tracked globally due to the following constraints: 950 * 951 * - A spare may be part of multiple pools. 952 * - A spare may be added to a pool even if it's actively in use within 953 * another pool. 954 * - A spare in use in any pool can only be the source of a replacement if 955 * the target is a spare in the same pool. 956 * 957 * We keep track of all spares on the system through the use of a reference 958 * counted AVL tree. When a vdev is added as a spare, or used as a replacement 959 * spare, then we bump the reference count in the AVL tree. In addition, we set 960 * the 'vdev_isspare' member to indicate that the device is a spare (active or 961 * inactive). When a spare is made active (used to replace a device in the 962 * pool), we also keep track of which pool its been made a part of. 963 * 964 * The 'spa_spare_lock' protects the AVL tree. These functions are normally 965 * called under the spa_namespace lock as part of vdev reconfiguration. The 966 * separate spare lock exists for the status query path, which does not need to 967 * be completely consistent with respect to other vdev configuration changes. 968 */ 969 970static int 971spa_spare_compare(const void *a, const void *b) 972{ 973 return (spa_aux_compare(a, b)); 974} 975 976void 977spa_spare_add(vdev_t *vd) 978{ 979 mutex_enter(&spa_spare_lock); 980 ASSERT(!vd->vdev_isspare); 981 spa_aux_add(vd, &spa_spare_avl); 982 vd->vdev_isspare = B_TRUE; 983 mutex_exit(&spa_spare_lock); 984} 985 986void 987spa_spare_remove(vdev_t *vd) 988{ 989 mutex_enter(&spa_spare_lock); 990 ASSERT(vd->vdev_isspare); 991 spa_aux_remove(vd, &spa_spare_avl); 992 vd->vdev_isspare = B_FALSE; 993 mutex_exit(&spa_spare_lock); 994} 995 996boolean_t 997spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt) 998{ 999 boolean_t found; 1000 1001 mutex_enter(&spa_spare_lock); 1002 found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl); 1003 mutex_exit(&spa_spare_lock); 1004 1005 return (found); 1006} 1007 1008void 1009spa_spare_activate(vdev_t *vd) 1010{ 1011 mutex_enter(&spa_spare_lock); 1012 ASSERT(vd->vdev_isspare); 1013 spa_aux_activate(vd, &spa_spare_avl); 1014 mutex_exit(&spa_spare_lock); 1015} 1016 1017/* 1018 * Level 2 ARC devices are tracked globally for the same reasons as spares. 1019 * Cache devices currently only support one pool per cache device, and so 1020 * for these devices the aux reference count is currently unused beyond 1. 1021 */ 1022 1023static int 1024spa_l2cache_compare(const void *a, const void *b) 1025{ 1026 return (spa_aux_compare(a, b)); 1027} 1028 1029void 1030spa_l2cache_add(vdev_t *vd) 1031{ 1032 mutex_enter(&spa_l2cache_lock); 1033 ASSERT(!vd->vdev_isl2cache); 1034 spa_aux_add(vd, &spa_l2cache_avl); 1035 vd->vdev_isl2cache = B_TRUE; 1036 mutex_exit(&spa_l2cache_lock); 1037} 1038 1039void 1040spa_l2cache_remove(vdev_t *vd) 1041{ 1042 mutex_enter(&spa_l2cache_lock); 1043 ASSERT(vd->vdev_isl2cache); 1044 spa_aux_remove(vd, &spa_l2cache_avl); 1045 vd->vdev_isl2cache = B_FALSE; 1046 mutex_exit(&spa_l2cache_lock); 1047} 1048 1049boolean_t 1050spa_l2cache_exists(uint64_t guid, uint64_t *pool) 1051{ 1052 boolean_t found; 1053 1054 mutex_enter(&spa_l2cache_lock); 1055 found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl); 1056 mutex_exit(&spa_l2cache_lock); 1057 1058 return (found); 1059} 1060 1061void 1062spa_l2cache_activate(vdev_t *vd) 1063{ 1064 mutex_enter(&spa_l2cache_lock); 1065 ASSERT(vd->vdev_isl2cache); 1066 spa_aux_activate(vd, &spa_l2cache_avl); 1067 mutex_exit(&spa_l2cache_lock); 1068} 1069 1070/* 1071 * ========================================================================== 1072 * SPA vdev locking 1073 * ========================================================================== 1074 */ 1075 1076/* 1077 * Lock the given spa_t for the purpose of adding or removing a vdev. 1078 * Grabs the global spa_namespace_lock plus the spa config lock for writing. 1079 * It returns the next transaction group for the spa_t. 1080 */ 1081uint64_t 1082spa_vdev_enter(spa_t *spa) 1083{ 1084 mutex_enter(&spa->spa_vdev_top_lock); 1085 mutex_enter(&spa_namespace_lock); 1086 return (spa_vdev_config_enter(spa)); 1087} 1088 1089/* 1090 * Internal implementation for spa_vdev_enter(). Used when a vdev 1091 * operation requires multiple syncs (i.e. removing a device) while 1092 * keeping the spa_namespace_lock held. 1093 */ 1094uint64_t 1095spa_vdev_config_enter(spa_t *spa) 1096{ 1097 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1098 1099 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER); 1100 1101 return (spa_last_synced_txg(spa) + 1); 1102} 1103 1104/* 1105 * Used in combination with spa_vdev_config_enter() to allow the syncing 1106 * of multiple transactions without releasing the spa_namespace_lock. 1107 */ 1108void 1109spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag) 1110{ 1111 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1112 1113 int config_changed = B_FALSE; 1114 1115 ASSERT(txg > spa_last_synced_txg(spa)); 1116 1117 spa->spa_pending_vdev = NULL; 1118 1119 /* 1120 * Reassess the DTLs. 1121 */ 1122 vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE); 1123 1124 if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) { 1125 config_changed = B_TRUE; 1126 spa->spa_config_generation++; 1127 } 1128 1129 /* 1130 * Verify the metaslab classes. 1131 */ 1132 ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0); 1133 ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0); 1134 1135 spa_config_exit(spa, SCL_ALL, spa); 1136 1137 /* 1138 * Panic the system if the specified tag requires it. This 1139 * is useful for ensuring that configurations are updated 1140 * transactionally. 1141 */ 1142 if (zio_injection_enabled) 1143 zio_handle_panic_injection(spa, tag, 0); 1144 1145 /* 1146 * Note: this txg_wait_synced() is important because it ensures 1147 * that there won't be more than one config change per txg. 1148 * This allows us to use the txg as the generation number. 1149 */ 1150 if (error == 0) 1151 txg_wait_synced(spa->spa_dsl_pool, txg); 1152 1153 if (vd != NULL) { 1154 ASSERT(!vd->vdev_detached || vd->vdev_dtl_sm == NULL); 1155 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER); 1156 vdev_free(vd); 1157 spa_config_exit(spa, SCL_ALL, spa); 1158 } 1159 1160 /* 1161 * If the config changed, update the config cache. 1162 */ 1163 if (config_changed) 1164 spa_config_sync(spa, B_FALSE, B_TRUE); 1165} 1166 1167/* 1168 * Unlock the spa_t after adding or removing a vdev. Besides undoing the 1169 * locking of spa_vdev_enter(), we also want make sure the transactions have 1170 * synced to disk, and then update the global configuration cache with the new 1171 * information. 1172 */ 1173int 1174spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error) 1175{ 1176 spa_vdev_config_exit(spa, vd, txg, error, FTAG); 1177 mutex_exit(&spa_namespace_lock); 1178 mutex_exit(&spa->spa_vdev_top_lock); 1179 1180 return (error); 1181} 1182 1183/* 1184 * Lock the given spa_t for the purpose of changing vdev state. 1185 */ 1186void 1187spa_vdev_state_enter(spa_t *spa, int oplocks) 1188{ 1189 int locks = SCL_STATE_ALL | oplocks; 1190 1191 /* 1192 * Root pools may need to read of the underlying devfs filesystem 1193 * when opening up a vdev. Unfortunately if we're holding the 1194 * SCL_ZIO lock it will result in a deadlock when we try to issue 1195 * the read from the root filesystem. Instead we "prefetch" 1196 * the associated vnodes that we need prior to opening the 1197 * underlying devices and cache them so that we can prevent 1198 * any I/O when we are doing the actual open. 1199 */ 1200 if (spa_is_root(spa)) { 1201 int low = locks & ~(SCL_ZIO - 1); 1202 int high = locks & ~low; 1203 1204 spa_config_enter(spa, high, spa, RW_WRITER); 1205 vdev_hold(spa->spa_root_vdev); 1206 spa_config_enter(spa, low, spa, RW_WRITER); 1207 } else { 1208 spa_config_enter(spa, locks, spa, RW_WRITER); 1209 } 1210 spa->spa_vdev_locks = locks; 1211} 1212 1213int 1214spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error) 1215{ 1216 boolean_t config_changed = B_FALSE; 1217 1218 if (vd != NULL || error == 0) 1219 vdev_dtl_reassess(vd ? vd->vdev_top : spa->spa_root_vdev, 1220 0, 0, B_FALSE); 1221 1222 if (vd != NULL) { 1223 vdev_state_dirty(vd->vdev_top); 1224 config_changed = B_TRUE; 1225 spa->spa_config_generation++; 1226 } 1227 1228 if (spa_is_root(spa)) 1229 vdev_rele(spa->spa_root_vdev); 1230 1231 ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL); 1232 spa_config_exit(spa, spa->spa_vdev_locks, spa); 1233 1234 /* 1235 * If anything changed, wait for it to sync. This ensures that, 1236 * from the system administrator's perspective, zpool(1M) commands 1237 * are synchronous. This is important for things like zpool offline: 1238 * when the command completes, you expect no further I/O from ZFS. 1239 */ 1240 if (vd != NULL) 1241 txg_wait_synced(spa->spa_dsl_pool, 0); 1242 1243 /* 1244 * If the config changed, update the config cache. 1245 */ 1246 if (config_changed) { 1247 mutex_enter(&spa_namespace_lock); 1248 spa_config_sync(spa, B_FALSE, B_TRUE); 1249 mutex_exit(&spa_namespace_lock); 1250 } 1251 1252 return (error); 1253} 1254 1255/* 1256 * ========================================================================== 1257 * Miscellaneous functions 1258 * ========================================================================== 1259 */ 1260 1261void 1262spa_activate_mos_feature(spa_t *spa, const char *feature, dmu_tx_t *tx) 1263{ 1264 if (!nvlist_exists(spa->spa_label_features, feature)) { 1265 fnvlist_add_boolean(spa->spa_label_features, feature); 1266 /* 1267 * When we are creating the pool (tx_txg==TXG_INITIAL), we can't 1268 * dirty the vdev config because lock SCL_CONFIG is not held. 1269 * Thankfully, in this case we don't need to dirty the config 1270 * because it will be written out anyway when we finish 1271 * creating the pool. 1272 */ 1273 if (tx->tx_txg != TXG_INITIAL) 1274 vdev_config_dirty(spa->spa_root_vdev); 1275 } 1276} 1277 1278void 1279spa_deactivate_mos_feature(spa_t *spa, const char *feature) 1280{ 1281 if (nvlist_remove_all(spa->spa_label_features, feature) == 0) 1282 vdev_config_dirty(spa->spa_root_vdev); 1283} 1284 1285/* 1286 * Rename a spa_t. 1287 */ 1288int 1289spa_rename(const char *name, const char *newname) 1290{ 1291 spa_t *spa; 1292 int err; 1293 1294 /* 1295 * Lookup the spa_t and grab the config lock for writing. We need to 1296 * actually open the pool so that we can sync out the necessary labels. 1297 * It's OK to call spa_open() with the namespace lock held because we 1298 * allow recursive calls for other reasons. 1299 */ 1300 mutex_enter(&spa_namespace_lock); 1301 if ((err = spa_open(name, &spa, FTAG)) != 0) { 1302 mutex_exit(&spa_namespace_lock); 1303 return (err); 1304 } 1305 1306 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1307 1308 avl_remove(&spa_namespace_avl, spa); 1309 (void) strlcpy(spa->spa_name, newname, sizeof (spa->spa_name)); 1310 avl_add(&spa_namespace_avl, spa); 1311 1312 /* 1313 * Sync all labels to disk with the new names by marking the root vdev 1314 * dirty and waiting for it to sync. It will pick up the new pool name 1315 * during the sync. 1316 */ 1317 vdev_config_dirty(spa->spa_root_vdev); 1318 1319 spa_config_exit(spa, SCL_ALL, FTAG); 1320 1321 txg_wait_synced(spa->spa_dsl_pool, 0); 1322 1323 /* 1324 * Sync the updated config cache. 1325 */ 1326 spa_config_sync(spa, B_FALSE, B_TRUE); 1327 1328 spa_close(spa, FTAG); 1329 1330 mutex_exit(&spa_namespace_lock); 1331 1332 return (0); 1333} 1334 1335/* 1336 * Return the spa_t associated with given pool_guid, if it exists. If 1337 * device_guid is non-zero, determine whether the pool exists *and* contains 1338 * a device with the specified device_guid. 1339 */ 1340spa_t * 1341spa_by_guid(uint64_t pool_guid, uint64_t device_guid) 1342{ 1343 spa_t *spa; 1344 avl_tree_t *t = &spa_namespace_avl; 1345 1346 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1347 1348 for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) { 1349 if (spa->spa_state == POOL_STATE_UNINITIALIZED) 1350 continue; 1351 if (spa->spa_root_vdev == NULL) 1352 continue; 1353 if (spa_guid(spa) == pool_guid) { 1354 if (device_guid == 0) 1355 break; 1356 1357 if (vdev_lookup_by_guid(spa->spa_root_vdev, 1358 device_guid) != NULL) 1359 break; 1360 1361 /* 1362 * Check any devices we may be in the process of adding. 1363 */ 1364 if (spa->spa_pending_vdev) { 1365 if (vdev_lookup_by_guid(spa->spa_pending_vdev, 1366 device_guid) != NULL) 1367 break; 1368 } 1369 } 1370 } 1371 1372 return (spa); 1373} 1374 1375/* 1376 * Determine whether a pool with the given pool_guid exists. 1377 */ 1378boolean_t 1379spa_guid_exists(uint64_t pool_guid, uint64_t device_guid) 1380{ 1381 return (spa_by_guid(pool_guid, device_guid) != NULL); 1382} 1383 1384char * 1385spa_strdup(const char *s) 1386{ 1387 size_t len; 1388 char *new; 1389 1390 len = strlen(s); 1391 new = kmem_alloc(len + 1, KM_SLEEP); 1392 bcopy(s, new, len); 1393 new[len] = '\0'; 1394 1395 return (new); 1396} 1397 1398void 1399spa_strfree(char *s) 1400{ 1401 kmem_free(s, strlen(s) + 1); 1402} 1403 1404uint64_t 1405spa_get_random(uint64_t range) 1406{ 1407 uint64_t r; 1408 1409 ASSERT(range != 0); 1410 1411 (void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t)); 1412 1413 return (r % range); 1414} 1415 1416uint64_t 1417spa_generate_guid(spa_t *spa) 1418{ 1419 uint64_t guid = spa_get_random(-1ULL); 1420 1421 if (spa != NULL) { 1422 while (guid == 0 || spa_guid_exists(spa_guid(spa), guid)) 1423 guid = spa_get_random(-1ULL); 1424 } else { 1425 while (guid == 0 || spa_guid_exists(guid, 0)) 1426 guid = spa_get_random(-1ULL); 1427 } 1428 1429 return (guid); 1430} 1431 1432void 1433snprintf_blkptr(char *buf, size_t buflen, const blkptr_t *bp) 1434{ 1435 char type[256]; 1436 char *checksum = NULL; 1437 char *compress = NULL; 1438 1439 if (bp != NULL) { 1440 if (BP_GET_TYPE(bp) & DMU_OT_NEWTYPE) { 1441 dmu_object_byteswap_t bswap = 1442 DMU_OT_BYTESWAP(BP_GET_TYPE(bp)); 1443 (void) snprintf(type, sizeof (type), "bswap %s %s", 1444 DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) ? 1445 "metadata" : "data", 1446 dmu_ot_byteswap[bswap].ob_name); 1447 } else { 1448 (void) strlcpy(type, dmu_ot[BP_GET_TYPE(bp)].ot_name, 1449 sizeof (type)); 1450 } 1451 if (!BP_IS_EMBEDDED(bp)) { 1452 checksum = 1453 zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name; 1454 } 1455 compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name; 1456 } 1457 1458 SNPRINTF_BLKPTR(snprintf, ' ', buf, buflen, bp, type, checksum, 1459 compress); 1460} 1461 1462void 1463spa_freeze(spa_t *spa) 1464{ 1465 uint64_t freeze_txg = 0; 1466 1467 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1468 if (spa->spa_freeze_txg == UINT64_MAX) { 1469 freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE; 1470 spa->spa_freeze_txg = freeze_txg; 1471 } 1472 spa_config_exit(spa, SCL_ALL, FTAG); 1473 if (freeze_txg != 0) 1474 txg_wait_synced(spa_get_dsl(spa), freeze_txg); 1475} 1476 1477void 1478zfs_panic_recover(const char *fmt, ...) 1479{ 1480 va_list adx; 1481 1482 va_start(adx, fmt); 1483 vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx); 1484 va_end(adx); 1485} 1486 1487/* 1488 * This is a stripped-down version of strtoull, suitable only for converting 1489 * lowercase hexadecimal numbers that don't overflow. 1490 */ 1491uint64_t 1492zfs_strtonum(const char *str, char **nptr) 1493{ 1494 uint64_t val = 0; 1495 char c; 1496 int digit; 1497 1498 while ((c = *str) != '\0') { 1499 if (c >= '0' && c <= '9') 1500 digit = c - '0'; 1501 else if (c >= 'a' && c <= 'f') 1502 digit = 10 + c - 'a'; 1503 else 1504 break; 1505 1506 val *= 16; 1507 val += digit; 1508 1509 str++; 1510 } 1511 1512 if (nptr) 1513 *nptr = (char *)str; 1514 1515 return (val); 1516} 1517 1518/* 1519 * ========================================================================== 1520 * Accessor functions 1521 * ========================================================================== 1522 */ 1523 1524boolean_t 1525spa_shutting_down(spa_t *spa) 1526{ 1527 return (spa->spa_async_suspended); 1528} 1529 1530dsl_pool_t * 1531spa_get_dsl(spa_t *spa) 1532{ 1533 return (spa->spa_dsl_pool); 1534} 1535 1536boolean_t 1537spa_is_initializing(spa_t *spa) 1538{ 1539 return (spa->spa_is_initializing); 1540} 1541 1542blkptr_t * 1543spa_get_rootblkptr(spa_t *spa) 1544{ 1545 return (&spa->spa_ubsync.ub_rootbp); 1546} 1547 1548void 1549spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp) 1550{ 1551 spa->spa_uberblock.ub_rootbp = *bp; 1552} 1553 1554void 1555spa_altroot(spa_t *spa, char *buf, size_t buflen) 1556{ 1557 if (spa->spa_root == NULL) 1558 buf[0] = '\0'; 1559 else 1560 (void) strncpy(buf, spa->spa_root, buflen); 1561} 1562 1563int 1564spa_sync_pass(spa_t *spa) 1565{ 1566 return (spa->spa_sync_pass); 1567} 1568 1569char * 1570spa_name(spa_t *spa) 1571{ 1572 return (spa->spa_name); 1573} 1574 1575uint64_t 1576spa_guid(spa_t *spa) 1577{ 1578 dsl_pool_t *dp = spa_get_dsl(spa); 1579 uint64_t guid; 1580 1581 /* 1582 * If we fail to parse the config during spa_load(), we can go through 1583 * the error path (which posts an ereport) and end up here with no root 1584 * vdev. We stash the original pool guid in 'spa_config_guid' to handle 1585 * this case. 1586 */ 1587 if (spa->spa_root_vdev == NULL) 1588 return (spa->spa_config_guid); 1589 1590 guid = spa->spa_last_synced_guid != 0 ? 1591 spa->spa_last_synced_guid : spa->spa_root_vdev->vdev_guid; 1592 1593 /* 1594 * Return the most recently synced out guid unless we're 1595 * in syncing context. 1596 */ 1597 if (dp && dsl_pool_sync_context(dp)) 1598 return (spa->spa_root_vdev->vdev_guid); 1599 else 1600 return (guid); 1601} 1602 1603uint64_t 1604spa_load_guid(spa_t *spa) 1605{ 1606 /* 1607 * This is a GUID that exists solely as a reference for the 1608 * purposes of the arc. It is generated at load time, and 1609 * is never written to persistent storage. 1610 */ 1611 return (spa->spa_load_guid); 1612} 1613 1614uint64_t 1615spa_last_synced_txg(spa_t *spa) 1616{ 1617 return (spa->spa_ubsync.ub_txg); 1618} 1619 1620uint64_t 1621spa_first_txg(spa_t *spa) 1622{ 1623 return (spa->spa_first_txg); 1624} 1625 1626uint64_t 1627spa_syncing_txg(spa_t *spa) 1628{ 1629 return (spa->spa_syncing_txg); 1630} 1631 1632pool_state_t 1633spa_state(spa_t *spa) 1634{ 1635 return (spa->spa_state); 1636} 1637 1638spa_load_state_t 1639spa_load_state(spa_t *spa) 1640{ 1641 return (spa->spa_load_state); 1642} 1643 1644uint64_t 1645spa_freeze_txg(spa_t *spa) 1646{ 1647 return (spa->spa_freeze_txg); 1648} 1649 1650/* ARGSUSED */ 1651uint64_t 1652spa_get_asize(spa_t *spa, uint64_t lsize) 1653{ 1654 return (lsize * spa_asize_inflation); 1655} 1656 1657/* 1658 * Return the amount of slop space in bytes. It is 1/32 of the pool (3.2%), 1659 * or at least 32MB. 1660 * 1661 * See the comment above spa_slop_shift for details. 1662 */ 1663uint64_t 1664spa_get_slop_space(spa_t *spa) { 1665 uint64_t space = spa_get_dspace(spa); 1666 return (MAX(space >> spa_slop_shift, SPA_MINDEVSIZE >> 1)); 1667} 1668 1669uint64_t 1670spa_get_dspace(spa_t *spa) 1671{ 1672 return (spa->spa_dspace); 1673} 1674 1675void 1676spa_update_dspace(spa_t *spa) 1677{ 1678 spa->spa_dspace = metaslab_class_get_dspace(spa_normal_class(spa)) + 1679 ddt_get_dedup_dspace(spa); 1680} 1681 1682/* 1683 * Return the failure mode that has been set to this pool. The default 1684 * behavior will be to block all I/Os when a complete failure occurs. 1685 */ 1686uint8_t 1687spa_get_failmode(spa_t *spa) 1688{ 1689 return (spa->spa_failmode); 1690} 1691 1692boolean_t 1693spa_suspended(spa_t *spa) 1694{ 1695 return (spa->spa_suspended); 1696} 1697 1698uint64_t 1699spa_version(spa_t *spa) 1700{ 1701 return (spa->spa_ubsync.ub_version); 1702} 1703 1704boolean_t 1705spa_deflate(spa_t *spa) 1706{ 1707 return (spa->spa_deflate); 1708} 1709 1710metaslab_class_t * 1711spa_normal_class(spa_t *spa) 1712{ 1713 return (spa->spa_normal_class); 1714} 1715 1716metaslab_class_t * 1717spa_log_class(spa_t *spa) 1718{ 1719 return (spa->spa_log_class); 1720} 1721 1722int 1723spa_max_replication(spa_t *spa) 1724{ 1725 /* 1726 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to 1727 * handle BPs with more than one DVA allocated. Set our max 1728 * replication level accordingly. 1729 */ 1730 if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS) 1731 return (1); 1732 return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override)); 1733} 1734 1735int 1736spa_prev_software_version(spa_t *spa) 1737{ 1738 return (spa->spa_prev_software_version); 1739} 1740 1741uint64_t 1742spa_deadman_synctime(spa_t *spa) 1743{ 1744 return (spa->spa_deadman_synctime); 1745} 1746 1747uint64_t 1748dva_get_dsize_sync(spa_t *spa, const dva_t *dva) 1749{ 1750 uint64_t asize = DVA_GET_ASIZE(dva); 1751 uint64_t dsize = asize; 1752 1753 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 1754 1755 if (asize != 0 && spa->spa_deflate) { 1756 vdev_t *vd = vdev_lookup_top(spa, DVA_GET_VDEV(dva)); 1757 dsize = (asize >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio; 1758 } 1759 1760 return (dsize); 1761} 1762 1763uint64_t 1764bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp) 1765{ 1766 uint64_t dsize = 0; 1767 1768 for (int d = 0; d < BP_GET_NDVAS(bp); d++) 1769 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]); 1770 1771 return (dsize); 1772} 1773 1774uint64_t 1775bp_get_dsize(spa_t *spa, const blkptr_t *bp) 1776{ 1777 uint64_t dsize = 0; 1778 1779 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); 1780 1781 for (int d = 0; d < BP_GET_NDVAS(bp); d++) 1782 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]); 1783 1784 spa_config_exit(spa, SCL_VDEV, FTAG); 1785 1786 return (dsize); 1787} 1788 1789/* 1790 * ========================================================================== 1791 * Initialization and Termination 1792 * ========================================================================== 1793 */ 1794 1795static int 1796spa_name_compare(const void *a1, const void *a2) 1797{ 1798 const spa_t *s1 = a1; 1799 const spa_t *s2 = a2; 1800 int s; 1801 1802 s = strcmp(s1->spa_name, s2->spa_name); 1803 if (s > 0) 1804 return (1); 1805 if (s < 0) 1806 return (-1); 1807 return (0); 1808} 1809 1810int 1811spa_busy(void) 1812{ 1813 return (spa_active_count); 1814} 1815 1816void 1817spa_boot_init() 1818{ 1819 spa_config_load(); 1820} 1821 1822#ifdef _KERNEL 1823EVENTHANDLER_DEFINE(mountroot, spa_boot_init, NULL, 0); 1824#endif 1825 1826void 1827spa_init(int mode) 1828{ 1829 mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL); 1830 mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL); 1831 mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL); 1832 cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL); 1833 1834 avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t), 1835 offsetof(spa_t, spa_avl)); 1836 1837 avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t), 1838 offsetof(spa_aux_t, aux_avl)); 1839 1840 avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t), 1841 offsetof(spa_aux_t, aux_avl)); 1842 1843 spa_mode_global = mode; 1844 1845#ifdef illumos 1846#ifdef _KERNEL 1847 spa_arch_init(); 1848#else 1849 if (spa_mode_global != FREAD && dprintf_find_string("watch")) { 1850 arc_procfd = open("/proc/self/ctl", O_WRONLY); 1851 if (arc_procfd == -1) { 1852 perror("could not enable watchpoints: " 1853 "opening /proc/self/ctl failed: "); 1854 } else { 1855 arc_watch = B_TRUE; 1856 } 1857 } 1858#endif 1859#endif /* illumos */ 1860 refcount_sysinit(); 1861 unique_init(); 1862 range_tree_init(); 1863 zio_init(); 1864 lz4_init(); 1865 dmu_init(); 1866 zil_init(); 1867 vdev_cache_stat_init(); 1868 zfs_prop_init(); 1869 zpool_prop_init(); 1870 zpool_feature_init(); 1871 spa_config_load(); 1872 l2arc_start(); 1873#ifndef illumos 1874#ifdef _KERNEL 1875 zfs_deadman_init(); 1876#endif 1877#endif /* !illumos */ 1878} 1879 1880void 1881spa_fini(void) 1882{ 1883 l2arc_stop(); 1884 1885 spa_evict_all(); 1886 1887 vdev_cache_stat_fini(); 1888 zil_fini(); 1889 dmu_fini(); 1890 lz4_fini(); 1891 zio_fini(); 1892 range_tree_fini(); 1893 unique_fini(); 1894 refcount_fini(); 1895 1896 avl_destroy(&spa_namespace_avl); 1897 avl_destroy(&spa_spare_avl); 1898 avl_destroy(&spa_l2cache_avl); 1899 1900 cv_destroy(&spa_namespace_cv); 1901 mutex_destroy(&spa_namespace_lock); 1902 mutex_destroy(&spa_spare_lock); 1903 mutex_destroy(&spa_l2cache_lock); 1904} 1905 1906/* 1907 * Return whether this pool has slogs. No locking needed. 1908 * It's not a problem if the wrong answer is returned as it's only for 1909 * performance and not correctness 1910 */ 1911boolean_t 1912spa_has_slogs(spa_t *spa) 1913{ 1914 return (spa->spa_log_class->mc_rotor != NULL); 1915} 1916 1917spa_log_state_t 1918spa_get_log_state(spa_t *spa) 1919{ 1920 return (spa->spa_log_state); 1921} 1922 1923void 1924spa_set_log_state(spa_t *spa, spa_log_state_t state) 1925{ 1926 spa->spa_log_state = state; 1927} 1928 1929boolean_t 1930spa_is_root(spa_t *spa) 1931{ 1932 return (spa->spa_is_root); 1933} 1934 1935boolean_t 1936spa_writeable(spa_t *spa) 1937{ 1938 return (!!(spa->spa_mode & FWRITE)); 1939} 1940 1941int 1942spa_mode(spa_t *spa) 1943{ 1944 return (spa->spa_mode); 1945} 1946 1947uint64_t 1948spa_bootfs(spa_t *spa) 1949{ 1950 return (spa->spa_bootfs); 1951} 1952 1953uint64_t 1954spa_delegation(spa_t *spa) 1955{ 1956 return (spa->spa_delegation); 1957} 1958 1959objset_t * 1960spa_meta_objset(spa_t *spa) 1961{ 1962 return (spa->spa_meta_objset); 1963} 1964 1965enum zio_checksum 1966spa_dedup_checksum(spa_t *spa) 1967{ 1968 return (spa->spa_dedup_checksum); 1969} 1970 1971/* 1972 * Reset pool scan stat per scan pass (or reboot). 1973 */ 1974void 1975spa_scan_stat_init(spa_t *spa) 1976{ 1977 /* data not stored on disk */ 1978 spa->spa_scan_pass_start = gethrestime_sec(); 1979 spa->spa_scan_pass_exam = 0; 1980 vdev_scan_stat_init(spa->spa_root_vdev); 1981} 1982 1983/* 1984 * Get scan stats for zpool status reports 1985 */ 1986int 1987spa_scan_get_stats(spa_t *spa, pool_scan_stat_t *ps) 1988{ 1989 dsl_scan_t *scn = spa->spa_dsl_pool ? spa->spa_dsl_pool->dp_scan : NULL; 1990 1991 if (scn == NULL || scn->scn_phys.scn_func == POOL_SCAN_NONE) 1992 return (SET_ERROR(ENOENT)); 1993 bzero(ps, sizeof (pool_scan_stat_t)); 1994 1995 /* data stored on disk */ 1996 ps->pss_func = scn->scn_phys.scn_func; 1997 ps->pss_start_time = scn->scn_phys.scn_start_time; 1998 ps->pss_end_time = scn->scn_phys.scn_end_time; 1999 ps->pss_to_examine = scn->scn_phys.scn_to_examine; 2000 ps->pss_examined = scn->scn_phys.scn_examined; 2001 ps->pss_to_process = scn->scn_phys.scn_to_process; 2002 ps->pss_processed = scn->scn_phys.scn_processed; 2003 ps->pss_errors = scn->scn_phys.scn_errors; 2004 ps->pss_state = scn->scn_phys.scn_state; 2005 2006 /* data not stored on disk */ 2007 ps->pss_pass_start = spa->spa_scan_pass_start; 2008 ps->pss_pass_exam = spa->spa_scan_pass_exam; 2009 2010 return (0); 2011} 2012 2013boolean_t 2014spa_debug_enabled(spa_t *spa) 2015{ 2016 return (spa->spa_debug); 2017} 2018