spa_misc.c revision 276081
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) 2011, 2014 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; 387SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_slop_shift, CTLFLAG_RWTUN, 388 &spa_slop_shift, 0, 389 "Shift value of reserved space (1/(2^spa_slop_shift))."); 390 391/* 392 * ========================================================================== 393 * SPA config locking 394 * ========================================================================== 395 */ 396static void 397spa_config_lock_init(spa_t *spa) 398{ 399 for (int i = 0; i < SCL_LOCKS; i++) { 400 spa_config_lock_t *scl = &spa->spa_config_lock[i]; 401 mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL); 402 cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL); 403 refcount_create_untracked(&scl->scl_count); 404 scl->scl_writer = NULL; 405 scl->scl_write_wanted = 0; 406 } 407} 408 409static void 410spa_config_lock_destroy(spa_t *spa) 411{ 412 for (int i = 0; i < SCL_LOCKS; i++) { 413 spa_config_lock_t *scl = &spa->spa_config_lock[i]; 414 mutex_destroy(&scl->scl_lock); 415 cv_destroy(&scl->scl_cv); 416 refcount_destroy(&scl->scl_count); 417 ASSERT(scl->scl_writer == NULL); 418 ASSERT(scl->scl_write_wanted == 0); 419 } 420} 421 422int 423spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw) 424{ 425 for (int i = 0; i < SCL_LOCKS; i++) { 426 spa_config_lock_t *scl = &spa->spa_config_lock[i]; 427 if (!(locks & (1 << i))) 428 continue; 429 mutex_enter(&scl->scl_lock); 430 if (rw == RW_READER) { 431 if (scl->scl_writer || scl->scl_write_wanted) { 432 mutex_exit(&scl->scl_lock); 433 spa_config_exit(spa, locks ^ (1 << i), tag); 434 return (0); 435 } 436 } else { 437 ASSERT(scl->scl_writer != curthread); 438 if (!refcount_is_zero(&scl->scl_count)) { 439 mutex_exit(&scl->scl_lock); 440 spa_config_exit(spa, locks ^ (1 << i), tag); 441 return (0); 442 } 443 scl->scl_writer = curthread; 444 } 445 (void) refcount_add(&scl->scl_count, tag); 446 mutex_exit(&scl->scl_lock); 447 } 448 return (1); 449} 450 451void 452spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw) 453{ 454 int wlocks_held = 0; 455 456 ASSERT3U(SCL_LOCKS, <, sizeof (wlocks_held) * NBBY); 457 458 for (int i = 0; i < SCL_LOCKS; i++) { 459 spa_config_lock_t *scl = &spa->spa_config_lock[i]; 460 if (scl->scl_writer == curthread) 461 wlocks_held |= (1 << i); 462 if (!(locks & (1 << i))) 463 continue; 464 mutex_enter(&scl->scl_lock); 465 if (rw == RW_READER) { 466 while (scl->scl_writer || scl->scl_write_wanted) { 467 cv_wait(&scl->scl_cv, &scl->scl_lock); 468 } 469 } else { 470 ASSERT(scl->scl_writer != curthread); 471 while (!refcount_is_zero(&scl->scl_count)) { 472 scl->scl_write_wanted++; 473 cv_wait(&scl->scl_cv, &scl->scl_lock); 474 scl->scl_write_wanted--; 475 } 476 scl->scl_writer = curthread; 477 } 478 (void) refcount_add(&scl->scl_count, tag); 479 mutex_exit(&scl->scl_lock); 480 } 481 ASSERT(wlocks_held <= locks); 482} 483 484void 485spa_config_exit(spa_t *spa, int locks, void *tag) 486{ 487 for (int i = SCL_LOCKS - 1; i >= 0; i--) { 488 spa_config_lock_t *scl = &spa->spa_config_lock[i]; 489 if (!(locks & (1 << i))) 490 continue; 491 mutex_enter(&scl->scl_lock); 492 ASSERT(!refcount_is_zero(&scl->scl_count)); 493 if (refcount_remove(&scl->scl_count, tag) == 0) { 494 ASSERT(scl->scl_writer == NULL || 495 scl->scl_writer == curthread); 496 scl->scl_writer = NULL; /* OK in either case */ 497 cv_broadcast(&scl->scl_cv); 498 } 499 mutex_exit(&scl->scl_lock); 500 } 501} 502 503int 504spa_config_held(spa_t *spa, int locks, krw_t rw) 505{ 506 int locks_held = 0; 507 508 for (int i = 0; i < SCL_LOCKS; i++) { 509 spa_config_lock_t *scl = &spa->spa_config_lock[i]; 510 if (!(locks & (1 << i))) 511 continue; 512 if ((rw == RW_READER && !refcount_is_zero(&scl->scl_count)) || 513 (rw == RW_WRITER && scl->scl_writer == curthread)) 514 locks_held |= 1 << i; 515 } 516 517 return (locks_held); 518} 519 520/* 521 * ========================================================================== 522 * SPA namespace functions 523 * ========================================================================== 524 */ 525 526/* 527 * Lookup the named spa_t in the AVL tree. The spa_namespace_lock must be held. 528 * Returns NULL if no matching spa_t is found. 529 */ 530spa_t * 531spa_lookup(const char *name) 532{ 533 static spa_t search; /* spa_t is large; don't allocate on stack */ 534 spa_t *spa; 535 avl_index_t where; 536 char *cp; 537 538 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 539 540 (void) strlcpy(search.spa_name, name, sizeof (search.spa_name)); 541 542 /* 543 * If it's a full dataset name, figure out the pool name and 544 * just use that. 545 */ 546 cp = strpbrk(search.spa_name, "/@#"); 547 if (cp != NULL) 548 *cp = '\0'; 549 550 spa = avl_find(&spa_namespace_avl, &search, &where); 551 552 return (spa); 553} 554 555/* 556 * Fires when spa_sync has not completed within zfs_deadman_synctime_ms. 557 * If the zfs_deadman_enabled flag is set then it inspects all vdev queues 558 * looking for potentially hung I/Os. 559 */ 560void 561spa_deadman(void *arg) 562{ 563 spa_t *spa = arg; 564 565 /* 566 * Disable the deadman timer if the pool is suspended. 567 */ 568 if (spa_suspended(spa)) { 569#ifdef illumos 570 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY)); 571#else 572 /* Nothing. just don't schedule any future callouts. */ 573#endif 574 return; 575 } 576 577 zfs_dbgmsg("slow spa_sync: started %llu seconds ago, calls %llu", 578 (gethrtime() - spa->spa_sync_starttime) / NANOSEC, 579 ++spa->spa_deadman_calls); 580 if (zfs_deadman_enabled) 581 vdev_deadman(spa->spa_root_vdev); 582} 583 584/* 585 * Create an uninitialized spa_t with the given name. Requires 586 * spa_namespace_lock. The caller must ensure that the spa_t doesn't already 587 * exist by calling spa_lookup() first. 588 */ 589spa_t * 590spa_add(const char *name, nvlist_t *config, const char *altroot) 591{ 592 spa_t *spa; 593 spa_config_dirent_t *dp; 594#ifdef illumos 595 cyc_handler_t hdlr; 596 cyc_time_t when; 597#endif 598 599 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 600 601 spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP); 602 603 mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL); 604 mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL); 605 mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL); 606 mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL); 607 mutex_init(&spa->spa_proc_lock, NULL, MUTEX_DEFAULT, NULL); 608 mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL); 609 mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL); 610 mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL); 611 mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL); 612 613 cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL); 614 cv_init(&spa->spa_proc_cv, NULL, CV_DEFAULT, NULL); 615 cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL); 616 cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL); 617 618 for (int t = 0; t < TXG_SIZE; t++) 619 bplist_create(&spa->spa_free_bplist[t]); 620 621 (void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name)); 622 spa->spa_state = POOL_STATE_UNINITIALIZED; 623 spa->spa_freeze_txg = UINT64_MAX; 624 spa->spa_final_txg = UINT64_MAX; 625 spa->spa_load_max_txg = UINT64_MAX; 626 spa->spa_proc = &p0; 627 spa->spa_proc_state = SPA_PROC_NONE; 628 629#ifdef illumos 630 hdlr.cyh_func = spa_deadman; 631 hdlr.cyh_arg = spa; 632 hdlr.cyh_level = CY_LOW_LEVEL; 633#endif 634 635 spa->spa_deadman_synctime = MSEC2NSEC(zfs_deadman_synctime_ms); 636 637#ifdef illumos 638 /* 639 * This determines how often we need to check for hung I/Os after 640 * the cyclic has already fired. Since checking for hung I/Os is 641 * an expensive operation we don't want to check too frequently. 642 * Instead wait for 5 seconds before checking again. 643 */ 644 when.cyt_interval = MSEC2NSEC(zfs_deadman_checktime_ms); 645 when.cyt_when = CY_INFINITY; 646 mutex_enter(&cpu_lock); 647 spa->spa_deadman_cycid = cyclic_add(&hdlr, &when); 648 mutex_exit(&cpu_lock); 649#else /* !illumos */ 650#ifdef _KERNEL 651 callout_init(&spa->spa_deadman_cycid, CALLOUT_MPSAFE); 652#endif 653#endif 654 refcount_create(&spa->spa_refcount); 655 spa_config_lock_init(spa); 656 657 avl_add(&spa_namespace_avl, spa); 658 659 /* 660 * Set the alternate root, if there is one. 661 */ 662 if (altroot) { 663 spa->spa_root = spa_strdup(altroot); 664 spa_active_count++; 665 } 666 667 /* 668 * Every pool starts with the default cachefile 669 */ 670 list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t), 671 offsetof(spa_config_dirent_t, scd_link)); 672 673 dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP); 674 dp->scd_path = altroot ? NULL : spa_strdup(spa_config_path); 675 list_insert_head(&spa->spa_config_list, dp); 676 677 VERIFY(nvlist_alloc(&spa->spa_load_info, NV_UNIQUE_NAME, 678 KM_SLEEP) == 0); 679 680 if (config != NULL) { 681 nvlist_t *features; 682 683 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ, 684 &features) == 0) { 685 VERIFY(nvlist_dup(features, &spa->spa_label_features, 686 0) == 0); 687 } 688 689 VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0); 690 } 691 692 if (spa->spa_label_features == NULL) { 693 VERIFY(nvlist_alloc(&spa->spa_label_features, NV_UNIQUE_NAME, 694 KM_SLEEP) == 0); 695 } 696 697 spa->spa_debug = ((zfs_flags & ZFS_DEBUG_SPA) != 0); 698 699 /* 700 * As a pool is being created, treat all features as disabled by 701 * setting SPA_FEATURE_DISABLED for all entries in the feature 702 * refcount cache. 703 */ 704 for (int i = 0; i < SPA_FEATURES; i++) { 705 spa->spa_feat_refcount_cache[i] = SPA_FEATURE_DISABLED; 706 } 707 708 return (spa); 709} 710 711/* 712 * Removes a spa_t from the namespace, freeing up any memory used. Requires 713 * spa_namespace_lock. This is called only after the spa_t has been closed and 714 * deactivated. 715 */ 716void 717spa_remove(spa_t *spa) 718{ 719 spa_config_dirent_t *dp; 720 721 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 722 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); 723 724 nvlist_free(spa->spa_config_splitting); 725 726 avl_remove(&spa_namespace_avl, spa); 727 cv_broadcast(&spa_namespace_cv); 728 729 if (spa->spa_root) { 730 spa_strfree(spa->spa_root); 731 spa_active_count--; 732 } 733 734 while ((dp = list_head(&spa->spa_config_list)) != NULL) { 735 list_remove(&spa->spa_config_list, dp); 736 if (dp->scd_path != NULL) 737 spa_strfree(dp->scd_path); 738 kmem_free(dp, sizeof (spa_config_dirent_t)); 739 } 740 741 list_destroy(&spa->spa_config_list); 742 743 nvlist_free(spa->spa_label_features); 744 nvlist_free(spa->spa_load_info); 745 spa_config_set(spa, NULL); 746 747#ifdef illumos 748 mutex_enter(&cpu_lock); 749 if (spa->spa_deadman_cycid != CYCLIC_NONE) 750 cyclic_remove(spa->spa_deadman_cycid); 751 mutex_exit(&cpu_lock); 752 spa->spa_deadman_cycid = CYCLIC_NONE; 753#else /* !illumos */ 754#ifdef _KERNEL 755 callout_drain(&spa->spa_deadman_cycid); 756#endif 757#endif 758 759 refcount_destroy(&spa->spa_refcount); 760 761 spa_config_lock_destroy(spa); 762 763 for (int t = 0; t < TXG_SIZE; t++) 764 bplist_destroy(&spa->spa_free_bplist[t]); 765 766 cv_destroy(&spa->spa_async_cv); 767 cv_destroy(&spa->spa_proc_cv); 768 cv_destroy(&spa->spa_scrub_io_cv); 769 cv_destroy(&spa->spa_suspend_cv); 770 771 mutex_destroy(&spa->spa_async_lock); 772 mutex_destroy(&spa->spa_errlist_lock); 773 mutex_destroy(&spa->spa_errlog_lock); 774 mutex_destroy(&spa->spa_history_lock); 775 mutex_destroy(&spa->spa_proc_lock); 776 mutex_destroy(&spa->spa_props_lock); 777 mutex_destroy(&spa->spa_scrub_lock); 778 mutex_destroy(&spa->spa_suspend_lock); 779 mutex_destroy(&spa->spa_vdev_top_lock); 780 781 kmem_free(spa, sizeof (spa_t)); 782} 783 784/* 785 * Given a pool, return the next pool in the namespace, or NULL if there is 786 * none. If 'prev' is NULL, return the first pool. 787 */ 788spa_t * 789spa_next(spa_t *prev) 790{ 791 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 792 793 if (prev) 794 return (AVL_NEXT(&spa_namespace_avl, prev)); 795 else 796 return (avl_first(&spa_namespace_avl)); 797} 798 799/* 800 * ========================================================================== 801 * SPA refcount functions 802 * ========================================================================== 803 */ 804 805/* 806 * Add a reference to the given spa_t. Must have at least one reference, or 807 * have the namespace lock held. 808 */ 809void 810spa_open_ref(spa_t *spa, void *tag) 811{ 812 ASSERT(refcount_count(&spa->spa_refcount) >= spa->spa_minref || 813 MUTEX_HELD(&spa_namespace_lock)); 814 (void) refcount_add(&spa->spa_refcount, tag); 815} 816 817/* 818 * Remove a reference to the given spa_t. Must have at least one reference, or 819 * have the namespace lock held. 820 */ 821void 822spa_close(spa_t *spa, void *tag) 823{ 824 ASSERT(refcount_count(&spa->spa_refcount) > spa->spa_minref || 825 MUTEX_HELD(&spa_namespace_lock)); 826 (void) refcount_remove(&spa->spa_refcount, tag); 827} 828 829/* 830 * Check to see if the spa refcount is zero. Must be called with 831 * spa_namespace_lock held. We really compare against spa_minref, which is the 832 * number of references acquired when opening a pool 833 */ 834boolean_t 835spa_refcount_zero(spa_t *spa) 836{ 837 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 838 839 return (refcount_count(&spa->spa_refcount) == spa->spa_minref); 840} 841 842/* 843 * ========================================================================== 844 * SPA spare and l2cache tracking 845 * ========================================================================== 846 */ 847 848/* 849 * Hot spares and cache devices are tracked using the same code below, 850 * for 'auxiliary' devices. 851 */ 852 853typedef struct spa_aux { 854 uint64_t aux_guid; 855 uint64_t aux_pool; 856 avl_node_t aux_avl; 857 int aux_count; 858} spa_aux_t; 859 860static int 861spa_aux_compare(const void *a, const void *b) 862{ 863 const spa_aux_t *sa = a; 864 const spa_aux_t *sb = b; 865 866 if (sa->aux_guid < sb->aux_guid) 867 return (-1); 868 else if (sa->aux_guid > sb->aux_guid) 869 return (1); 870 else 871 return (0); 872} 873 874void 875spa_aux_add(vdev_t *vd, avl_tree_t *avl) 876{ 877 avl_index_t where; 878 spa_aux_t search; 879 spa_aux_t *aux; 880 881 search.aux_guid = vd->vdev_guid; 882 if ((aux = avl_find(avl, &search, &where)) != NULL) { 883 aux->aux_count++; 884 } else { 885 aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP); 886 aux->aux_guid = vd->vdev_guid; 887 aux->aux_count = 1; 888 avl_insert(avl, aux, where); 889 } 890} 891 892void 893spa_aux_remove(vdev_t *vd, avl_tree_t *avl) 894{ 895 spa_aux_t search; 896 spa_aux_t *aux; 897 avl_index_t where; 898 899 search.aux_guid = vd->vdev_guid; 900 aux = avl_find(avl, &search, &where); 901 902 ASSERT(aux != NULL); 903 904 if (--aux->aux_count == 0) { 905 avl_remove(avl, aux); 906 kmem_free(aux, sizeof (spa_aux_t)); 907 } else if (aux->aux_pool == spa_guid(vd->vdev_spa)) { 908 aux->aux_pool = 0ULL; 909 } 910} 911 912boolean_t 913spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl) 914{ 915 spa_aux_t search, *found; 916 917 search.aux_guid = guid; 918 found = avl_find(avl, &search, NULL); 919 920 if (pool) { 921 if (found) 922 *pool = found->aux_pool; 923 else 924 *pool = 0ULL; 925 } 926 927 if (refcnt) { 928 if (found) 929 *refcnt = found->aux_count; 930 else 931 *refcnt = 0; 932 } 933 934 return (found != NULL); 935} 936 937void 938spa_aux_activate(vdev_t *vd, avl_tree_t *avl) 939{ 940 spa_aux_t search, *found; 941 avl_index_t where; 942 943 search.aux_guid = vd->vdev_guid; 944 found = avl_find(avl, &search, &where); 945 ASSERT(found != NULL); 946 ASSERT(found->aux_pool == 0ULL); 947 948 found->aux_pool = spa_guid(vd->vdev_spa); 949} 950 951/* 952 * Spares are tracked globally due to the following constraints: 953 * 954 * - A spare may be part of multiple pools. 955 * - A spare may be added to a pool even if it's actively in use within 956 * another pool. 957 * - A spare in use in any pool can only be the source of a replacement if 958 * the target is a spare in the same pool. 959 * 960 * We keep track of all spares on the system through the use of a reference 961 * counted AVL tree. When a vdev is added as a spare, or used as a replacement 962 * spare, then we bump the reference count in the AVL tree. In addition, we set 963 * the 'vdev_isspare' member to indicate that the device is a spare (active or 964 * inactive). When a spare is made active (used to replace a device in the 965 * pool), we also keep track of which pool its been made a part of. 966 * 967 * The 'spa_spare_lock' protects the AVL tree. These functions are normally 968 * called under the spa_namespace lock as part of vdev reconfiguration. The 969 * separate spare lock exists for the status query path, which does not need to 970 * be completely consistent with respect to other vdev configuration changes. 971 */ 972 973static int 974spa_spare_compare(const void *a, const void *b) 975{ 976 return (spa_aux_compare(a, b)); 977} 978 979void 980spa_spare_add(vdev_t *vd) 981{ 982 mutex_enter(&spa_spare_lock); 983 ASSERT(!vd->vdev_isspare); 984 spa_aux_add(vd, &spa_spare_avl); 985 vd->vdev_isspare = B_TRUE; 986 mutex_exit(&spa_spare_lock); 987} 988 989void 990spa_spare_remove(vdev_t *vd) 991{ 992 mutex_enter(&spa_spare_lock); 993 ASSERT(vd->vdev_isspare); 994 spa_aux_remove(vd, &spa_spare_avl); 995 vd->vdev_isspare = B_FALSE; 996 mutex_exit(&spa_spare_lock); 997} 998 999boolean_t 1000spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt) 1001{ 1002 boolean_t found; 1003 1004 mutex_enter(&spa_spare_lock); 1005 found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl); 1006 mutex_exit(&spa_spare_lock); 1007 1008 return (found); 1009} 1010 1011void 1012spa_spare_activate(vdev_t *vd) 1013{ 1014 mutex_enter(&spa_spare_lock); 1015 ASSERT(vd->vdev_isspare); 1016 spa_aux_activate(vd, &spa_spare_avl); 1017 mutex_exit(&spa_spare_lock); 1018} 1019 1020/* 1021 * Level 2 ARC devices are tracked globally for the same reasons as spares. 1022 * Cache devices currently only support one pool per cache device, and so 1023 * for these devices the aux reference count is currently unused beyond 1. 1024 */ 1025 1026static int 1027spa_l2cache_compare(const void *a, const void *b) 1028{ 1029 return (spa_aux_compare(a, b)); 1030} 1031 1032void 1033spa_l2cache_add(vdev_t *vd) 1034{ 1035 mutex_enter(&spa_l2cache_lock); 1036 ASSERT(!vd->vdev_isl2cache); 1037 spa_aux_add(vd, &spa_l2cache_avl); 1038 vd->vdev_isl2cache = B_TRUE; 1039 mutex_exit(&spa_l2cache_lock); 1040} 1041 1042void 1043spa_l2cache_remove(vdev_t *vd) 1044{ 1045 mutex_enter(&spa_l2cache_lock); 1046 ASSERT(vd->vdev_isl2cache); 1047 spa_aux_remove(vd, &spa_l2cache_avl); 1048 vd->vdev_isl2cache = B_FALSE; 1049 mutex_exit(&spa_l2cache_lock); 1050} 1051 1052boolean_t 1053spa_l2cache_exists(uint64_t guid, uint64_t *pool) 1054{ 1055 boolean_t found; 1056 1057 mutex_enter(&spa_l2cache_lock); 1058 found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl); 1059 mutex_exit(&spa_l2cache_lock); 1060 1061 return (found); 1062} 1063 1064void 1065spa_l2cache_activate(vdev_t *vd) 1066{ 1067 mutex_enter(&spa_l2cache_lock); 1068 ASSERT(vd->vdev_isl2cache); 1069 spa_aux_activate(vd, &spa_l2cache_avl); 1070 mutex_exit(&spa_l2cache_lock); 1071} 1072 1073/* 1074 * ========================================================================== 1075 * SPA vdev locking 1076 * ========================================================================== 1077 */ 1078 1079/* 1080 * Lock the given spa_t for the purpose of adding or removing a vdev. 1081 * Grabs the global spa_namespace_lock plus the spa config lock for writing. 1082 * It returns the next transaction group for the spa_t. 1083 */ 1084uint64_t 1085spa_vdev_enter(spa_t *spa) 1086{ 1087 mutex_enter(&spa->spa_vdev_top_lock); 1088 mutex_enter(&spa_namespace_lock); 1089 return (spa_vdev_config_enter(spa)); 1090} 1091 1092/* 1093 * Internal implementation for spa_vdev_enter(). Used when a vdev 1094 * operation requires multiple syncs (i.e. removing a device) while 1095 * keeping the spa_namespace_lock held. 1096 */ 1097uint64_t 1098spa_vdev_config_enter(spa_t *spa) 1099{ 1100 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1101 1102 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER); 1103 1104 return (spa_last_synced_txg(spa) + 1); 1105} 1106 1107/* 1108 * Used in combination with spa_vdev_config_enter() to allow the syncing 1109 * of multiple transactions without releasing the spa_namespace_lock. 1110 */ 1111void 1112spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag) 1113{ 1114 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1115 1116 int config_changed = B_FALSE; 1117 1118 ASSERT(txg > spa_last_synced_txg(spa)); 1119 1120 spa->spa_pending_vdev = NULL; 1121 1122 /* 1123 * Reassess the DTLs. 1124 */ 1125 vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE); 1126 1127 if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) { 1128 config_changed = B_TRUE; 1129 spa->spa_config_generation++; 1130 } 1131 1132 /* 1133 * Verify the metaslab classes. 1134 */ 1135 ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0); 1136 ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0); 1137 1138 spa_config_exit(spa, SCL_ALL, spa); 1139 1140 /* 1141 * Panic the system if the specified tag requires it. This 1142 * is useful for ensuring that configurations are updated 1143 * transactionally. 1144 */ 1145 if (zio_injection_enabled) 1146 zio_handle_panic_injection(spa, tag, 0); 1147 1148 /* 1149 * Note: this txg_wait_synced() is important because it ensures 1150 * that there won't be more than one config change per txg. 1151 * This allows us to use the txg as the generation number. 1152 */ 1153 if (error == 0) 1154 txg_wait_synced(spa->spa_dsl_pool, txg); 1155 1156 if (vd != NULL) { 1157 ASSERT(!vd->vdev_detached || vd->vdev_dtl_sm == NULL); 1158 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER); 1159 vdev_free(vd); 1160 spa_config_exit(spa, SCL_ALL, spa); 1161 } 1162 1163 /* 1164 * If the config changed, update the config cache. 1165 */ 1166 if (config_changed) 1167 spa_config_sync(spa, B_FALSE, B_TRUE); 1168} 1169 1170/* 1171 * Unlock the spa_t after adding or removing a vdev. Besides undoing the 1172 * locking of spa_vdev_enter(), we also want make sure the transactions have 1173 * synced to disk, and then update the global configuration cache with the new 1174 * information. 1175 */ 1176int 1177spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error) 1178{ 1179 spa_vdev_config_exit(spa, vd, txg, error, FTAG); 1180 mutex_exit(&spa_namespace_lock); 1181 mutex_exit(&spa->spa_vdev_top_lock); 1182 1183 return (error); 1184} 1185 1186/* 1187 * Lock the given spa_t for the purpose of changing vdev state. 1188 */ 1189void 1190spa_vdev_state_enter(spa_t *spa, int oplocks) 1191{ 1192 int locks = SCL_STATE_ALL | oplocks; 1193 1194 /* 1195 * Root pools may need to read of the underlying devfs filesystem 1196 * when opening up a vdev. Unfortunately if we're holding the 1197 * SCL_ZIO lock it will result in a deadlock when we try to issue 1198 * the read from the root filesystem. Instead we "prefetch" 1199 * the associated vnodes that we need prior to opening the 1200 * underlying devices and cache them so that we can prevent 1201 * any I/O when we are doing the actual open. 1202 */ 1203 if (spa_is_root(spa)) { 1204 int low = locks & ~(SCL_ZIO - 1); 1205 int high = locks & ~low; 1206 1207 spa_config_enter(spa, high, spa, RW_WRITER); 1208 vdev_hold(spa->spa_root_vdev); 1209 spa_config_enter(spa, low, spa, RW_WRITER); 1210 } else { 1211 spa_config_enter(spa, locks, spa, RW_WRITER); 1212 } 1213 spa->spa_vdev_locks = locks; 1214} 1215 1216int 1217spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error) 1218{ 1219 boolean_t config_changed = B_FALSE; 1220 1221 if (vd != NULL || error == 0) 1222 vdev_dtl_reassess(vd ? vd->vdev_top : spa->spa_root_vdev, 1223 0, 0, B_FALSE); 1224 1225 if (vd != NULL) { 1226 vdev_state_dirty(vd->vdev_top); 1227 config_changed = B_TRUE; 1228 spa->spa_config_generation++; 1229 } 1230 1231 if (spa_is_root(spa)) 1232 vdev_rele(spa->spa_root_vdev); 1233 1234 ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL); 1235 spa_config_exit(spa, spa->spa_vdev_locks, spa); 1236 1237 /* 1238 * If anything changed, wait for it to sync. This ensures that, 1239 * from the system administrator's perspective, zpool(1M) commands 1240 * are synchronous. This is important for things like zpool offline: 1241 * when the command completes, you expect no further I/O from ZFS. 1242 */ 1243 if (vd != NULL) 1244 txg_wait_synced(spa->spa_dsl_pool, 0); 1245 1246 /* 1247 * If the config changed, update the config cache. 1248 */ 1249 if (config_changed) { 1250 mutex_enter(&spa_namespace_lock); 1251 spa_config_sync(spa, B_FALSE, B_TRUE); 1252 mutex_exit(&spa_namespace_lock); 1253 } 1254 1255 return (error); 1256} 1257 1258/* 1259 * ========================================================================== 1260 * Miscellaneous functions 1261 * ========================================================================== 1262 */ 1263 1264void 1265spa_activate_mos_feature(spa_t *spa, const char *feature, dmu_tx_t *tx) 1266{ 1267 if (!nvlist_exists(spa->spa_label_features, feature)) { 1268 fnvlist_add_boolean(spa->spa_label_features, feature); 1269 /* 1270 * When we are creating the pool (tx_txg==TXG_INITIAL), we can't 1271 * dirty the vdev config because lock SCL_CONFIG is not held. 1272 * Thankfully, in this case we don't need to dirty the config 1273 * because it will be written out anyway when we finish 1274 * creating the pool. 1275 */ 1276 if (tx->tx_txg != TXG_INITIAL) 1277 vdev_config_dirty(spa->spa_root_vdev); 1278 } 1279} 1280 1281void 1282spa_deactivate_mos_feature(spa_t *spa, const char *feature) 1283{ 1284 if (nvlist_remove_all(spa->spa_label_features, feature) == 0) 1285 vdev_config_dirty(spa->spa_root_vdev); 1286} 1287 1288/* 1289 * Rename a spa_t. 1290 */ 1291int 1292spa_rename(const char *name, const char *newname) 1293{ 1294 spa_t *spa; 1295 int err; 1296 1297 /* 1298 * Lookup the spa_t and grab the config lock for writing. We need to 1299 * actually open the pool so that we can sync out the necessary labels. 1300 * It's OK to call spa_open() with the namespace lock held because we 1301 * allow recursive calls for other reasons. 1302 */ 1303 mutex_enter(&spa_namespace_lock); 1304 if ((err = spa_open(name, &spa, FTAG)) != 0) { 1305 mutex_exit(&spa_namespace_lock); 1306 return (err); 1307 } 1308 1309 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1310 1311 avl_remove(&spa_namespace_avl, spa); 1312 (void) strlcpy(spa->spa_name, newname, sizeof (spa->spa_name)); 1313 avl_add(&spa_namespace_avl, spa); 1314 1315 /* 1316 * Sync all labels to disk with the new names by marking the root vdev 1317 * dirty and waiting for it to sync. It will pick up the new pool name 1318 * during the sync. 1319 */ 1320 vdev_config_dirty(spa->spa_root_vdev); 1321 1322 spa_config_exit(spa, SCL_ALL, FTAG); 1323 1324 txg_wait_synced(spa->spa_dsl_pool, 0); 1325 1326 /* 1327 * Sync the updated config cache. 1328 */ 1329 spa_config_sync(spa, B_FALSE, B_TRUE); 1330 1331 spa_close(spa, FTAG); 1332 1333 mutex_exit(&spa_namespace_lock); 1334 1335 return (0); 1336} 1337 1338/* 1339 * Return the spa_t associated with given pool_guid, if it exists. If 1340 * device_guid is non-zero, determine whether the pool exists *and* contains 1341 * a device with the specified device_guid. 1342 */ 1343spa_t * 1344spa_by_guid(uint64_t pool_guid, uint64_t device_guid) 1345{ 1346 spa_t *spa; 1347 avl_tree_t *t = &spa_namespace_avl; 1348 1349 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1350 1351 for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) { 1352 if (spa->spa_state == POOL_STATE_UNINITIALIZED) 1353 continue; 1354 if (spa->spa_root_vdev == NULL) 1355 continue; 1356 if (spa_guid(spa) == pool_guid) { 1357 if (device_guid == 0) 1358 break; 1359 1360 if (vdev_lookup_by_guid(spa->spa_root_vdev, 1361 device_guid) != NULL) 1362 break; 1363 1364 /* 1365 * Check any devices we may be in the process of adding. 1366 */ 1367 if (spa->spa_pending_vdev) { 1368 if (vdev_lookup_by_guid(spa->spa_pending_vdev, 1369 device_guid) != NULL) 1370 break; 1371 } 1372 } 1373 } 1374 1375 return (spa); 1376} 1377 1378/* 1379 * Determine whether a pool with the given pool_guid exists. 1380 */ 1381boolean_t 1382spa_guid_exists(uint64_t pool_guid, uint64_t device_guid) 1383{ 1384 return (spa_by_guid(pool_guid, device_guid) != NULL); 1385} 1386 1387char * 1388spa_strdup(const char *s) 1389{ 1390 size_t len; 1391 char *new; 1392 1393 len = strlen(s); 1394 new = kmem_alloc(len + 1, KM_SLEEP); 1395 bcopy(s, new, len); 1396 new[len] = '\0'; 1397 1398 return (new); 1399} 1400 1401void 1402spa_strfree(char *s) 1403{ 1404 kmem_free(s, strlen(s) + 1); 1405} 1406 1407uint64_t 1408spa_get_random(uint64_t range) 1409{ 1410 uint64_t r; 1411 1412 ASSERT(range != 0); 1413 1414 (void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t)); 1415 1416 return (r % range); 1417} 1418 1419uint64_t 1420spa_generate_guid(spa_t *spa) 1421{ 1422 uint64_t guid = spa_get_random(-1ULL); 1423 1424 if (spa != NULL) { 1425 while (guid == 0 || spa_guid_exists(spa_guid(spa), guid)) 1426 guid = spa_get_random(-1ULL); 1427 } else { 1428 while (guid == 0 || spa_guid_exists(guid, 0)) 1429 guid = spa_get_random(-1ULL); 1430 } 1431 1432 return (guid); 1433} 1434 1435void 1436snprintf_blkptr(char *buf, size_t buflen, const blkptr_t *bp) 1437{ 1438 char type[256]; 1439 char *checksum = NULL; 1440 char *compress = NULL; 1441 1442 if (bp != NULL) { 1443 if (BP_GET_TYPE(bp) & DMU_OT_NEWTYPE) { 1444 dmu_object_byteswap_t bswap = 1445 DMU_OT_BYTESWAP(BP_GET_TYPE(bp)); 1446 (void) snprintf(type, sizeof (type), "bswap %s %s", 1447 DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) ? 1448 "metadata" : "data", 1449 dmu_ot_byteswap[bswap].ob_name); 1450 } else { 1451 (void) strlcpy(type, dmu_ot[BP_GET_TYPE(bp)].ot_name, 1452 sizeof (type)); 1453 } 1454 if (!BP_IS_EMBEDDED(bp)) { 1455 checksum = 1456 zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name; 1457 } 1458 compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name; 1459 } 1460 1461 SNPRINTF_BLKPTR(snprintf, ' ', buf, buflen, bp, type, checksum, 1462 compress); 1463} 1464 1465void 1466spa_freeze(spa_t *spa) 1467{ 1468 uint64_t freeze_txg = 0; 1469 1470 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1471 if (spa->spa_freeze_txg == UINT64_MAX) { 1472 freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE; 1473 spa->spa_freeze_txg = freeze_txg; 1474 } 1475 spa_config_exit(spa, SCL_ALL, FTAG); 1476 if (freeze_txg != 0) 1477 txg_wait_synced(spa_get_dsl(spa), freeze_txg); 1478} 1479 1480void 1481zfs_panic_recover(const char *fmt, ...) 1482{ 1483 va_list adx; 1484 1485 va_start(adx, fmt); 1486 vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx); 1487 va_end(adx); 1488} 1489 1490/* 1491 * This is a stripped-down version of strtoull, suitable only for converting 1492 * lowercase hexadecimal numbers that don't overflow. 1493 */ 1494uint64_t 1495zfs_strtonum(const char *str, char **nptr) 1496{ 1497 uint64_t val = 0; 1498 char c; 1499 int digit; 1500 1501 while ((c = *str) != '\0') { 1502 if (c >= '0' && c <= '9') 1503 digit = c - '0'; 1504 else if (c >= 'a' && c <= 'f') 1505 digit = 10 + c - 'a'; 1506 else 1507 break; 1508 1509 val *= 16; 1510 val += digit; 1511 1512 str++; 1513 } 1514 1515 if (nptr) 1516 *nptr = (char *)str; 1517 1518 return (val); 1519} 1520 1521/* 1522 * ========================================================================== 1523 * Accessor functions 1524 * ========================================================================== 1525 */ 1526 1527boolean_t 1528spa_shutting_down(spa_t *spa) 1529{ 1530 return (spa->spa_async_suspended); 1531} 1532 1533dsl_pool_t * 1534spa_get_dsl(spa_t *spa) 1535{ 1536 return (spa->spa_dsl_pool); 1537} 1538 1539boolean_t 1540spa_is_initializing(spa_t *spa) 1541{ 1542 return (spa->spa_is_initializing); 1543} 1544 1545blkptr_t * 1546spa_get_rootblkptr(spa_t *spa) 1547{ 1548 return (&spa->spa_ubsync.ub_rootbp); 1549} 1550 1551void 1552spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp) 1553{ 1554 spa->spa_uberblock.ub_rootbp = *bp; 1555} 1556 1557void 1558spa_altroot(spa_t *spa, char *buf, size_t buflen) 1559{ 1560 if (spa->spa_root == NULL) 1561 buf[0] = '\0'; 1562 else 1563 (void) strncpy(buf, spa->spa_root, buflen); 1564} 1565 1566int 1567spa_sync_pass(spa_t *spa) 1568{ 1569 return (spa->spa_sync_pass); 1570} 1571 1572char * 1573spa_name(spa_t *spa) 1574{ 1575 return (spa->spa_name); 1576} 1577 1578uint64_t 1579spa_guid(spa_t *spa) 1580{ 1581 dsl_pool_t *dp = spa_get_dsl(spa); 1582 uint64_t guid; 1583 1584 /* 1585 * If we fail to parse the config during spa_load(), we can go through 1586 * the error path (which posts an ereport) and end up here with no root 1587 * vdev. We stash the original pool guid in 'spa_config_guid' to handle 1588 * this case. 1589 */ 1590 if (spa->spa_root_vdev == NULL) 1591 return (spa->spa_config_guid); 1592 1593 guid = spa->spa_last_synced_guid != 0 ? 1594 spa->spa_last_synced_guid : spa->spa_root_vdev->vdev_guid; 1595 1596 /* 1597 * Return the most recently synced out guid unless we're 1598 * in syncing context. 1599 */ 1600 if (dp && dsl_pool_sync_context(dp)) 1601 return (spa->spa_root_vdev->vdev_guid); 1602 else 1603 return (guid); 1604} 1605 1606uint64_t 1607spa_load_guid(spa_t *spa) 1608{ 1609 /* 1610 * This is a GUID that exists solely as a reference for the 1611 * purposes of the arc. It is generated at load time, and 1612 * is never written to persistent storage. 1613 */ 1614 return (spa->spa_load_guid); 1615} 1616 1617uint64_t 1618spa_last_synced_txg(spa_t *spa) 1619{ 1620 return (spa->spa_ubsync.ub_txg); 1621} 1622 1623uint64_t 1624spa_first_txg(spa_t *spa) 1625{ 1626 return (spa->spa_first_txg); 1627} 1628 1629uint64_t 1630spa_syncing_txg(spa_t *spa) 1631{ 1632 return (spa->spa_syncing_txg); 1633} 1634 1635pool_state_t 1636spa_state(spa_t *spa) 1637{ 1638 return (spa->spa_state); 1639} 1640 1641spa_load_state_t 1642spa_load_state(spa_t *spa) 1643{ 1644 return (spa->spa_load_state); 1645} 1646 1647uint64_t 1648spa_freeze_txg(spa_t *spa) 1649{ 1650 return (spa->spa_freeze_txg); 1651} 1652 1653/* ARGSUSED */ 1654uint64_t 1655spa_get_asize(spa_t *spa, uint64_t lsize) 1656{ 1657 return (lsize * spa_asize_inflation); 1658} 1659 1660/* 1661 * Return the amount of slop space in bytes. It is 1/32 of the pool (3.2%), 1662 * or at least 32MB. 1663 * 1664 * See the comment above spa_slop_shift for details. 1665 */ 1666uint64_t 1667spa_get_slop_space(spa_t *spa) { 1668 uint64_t space = spa_get_dspace(spa); 1669 return (MAX(space >> spa_slop_shift, SPA_MINDEVSIZE >> 1)); 1670} 1671 1672uint64_t 1673spa_get_dspace(spa_t *spa) 1674{ 1675 return (spa->spa_dspace); 1676} 1677 1678void 1679spa_update_dspace(spa_t *spa) 1680{ 1681 spa->spa_dspace = metaslab_class_get_dspace(spa_normal_class(spa)) + 1682 ddt_get_dedup_dspace(spa); 1683} 1684 1685/* 1686 * Return the failure mode that has been set to this pool. The default 1687 * behavior will be to block all I/Os when a complete failure occurs. 1688 */ 1689uint8_t 1690spa_get_failmode(spa_t *spa) 1691{ 1692 return (spa->spa_failmode); 1693} 1694 1695boolean_t 1696spa_suspended(spa_t *spa) 1697{ 1698 return (spa->spa_suspended); 1699} 1700 1701uint64_t 1702spa_version(spa_t *spa) 1703{ 1704 return (spa->spa_ubsync.ub_version); 1705} 1706 1707boolean_t 1708spa_deflate(spa_t *spa) 1709{ 1710 return (spa->spa_deflate); 1711} 1712 1713metaslab_class_t * 1714spa_normal_class(spa_t *spa) 1715{ 1716 return (spa->spa_normal_class); 1717} 1718 1719metaslab_class_t * 1720spa_log_class(spa_t *spa) 1721{ 1722 return (spa->spa_log_class); 1723} 1724 1725int 1726spa_max_replication(spa_t *spa) 1727{ 1728 /* 1729 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to 1730 * handle BPs with more than one DVA allocated. Set our max 1731 * replication level accordingly. 1732 */ 1733 if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS) 1734 return (1); 1735 return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override)); 1736} 1737 1738int 1739spa_prev_software_version(spa_t *spa) 1740{ 1741 return (spa->spa_prev_software_version); 1742} 1743 1744uint64_t 1745spa_deadman_synctime(spa_t *spa) 1746{ 1747 return (spa->spa_deadman_synctime); 1748} 1749 1750uint64_t 1751dva_get_dsize_sync(spa_t *spa, const dva_t *dva) 1752{ 1753 uint64_t asize = DVA_GET_ASIZE(dva); 1754 uint64_t dsize = asize; 1755 1756 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 1757 1758 if (asize != 0 && spa->spa_deflate) { 1759 vdev_t *vd = vdev_lookup_top(spa, DVA_GET_VDEV(dva)); 1760 dsize = (asize >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio; 1761 } 1762 1763 return (dsize); 1764} 1765 1766uint64_t 1767bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp) 1768{ 1769 uint64_t dsize = 0; 1770 1771 for (int d = 0; d < BP_GET_NDVAS(bp); d++) 1772 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]); 1773 1774 return (dsize); 1775} 1776 1777uint64_t 1778bp_get_dsize(spa_t *spa, const blkptr_t *bp) 1779{ 1780 uint64_t dsize = 0; 1781 1782 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); 1783 1784 for (int d = 0; d < BP_GET_NDVAS(bp); d++) 1785 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]); 1786 1787 spa_config_exit(spa, SCL_VDEV, FTAG); 1788 1789 return (dsize); 1790} 1791 1792/* 1793 * ========================================================================== 1794 * Initialization and Termination 1795 * ========================================================================== 1796 */ 1797 1798static int 1799spa_name_compare(const void *a1, const void *a2) 1800{ 1801 const spa_t *s1 = a1; 1802 const spa_t *s2 = a2; 1803 int s; 1804 1805 s = strcmp(s1->spa_name, s2->spa_name); 1806 if (s > 0) 1807 return (1); 1808 if (s < 0) 1809 return (-1); 1810 return (0); 1811} 1812 1813int 1814spa_busy(void) 1815{ 1816 return (spa_active_count); 1817} 1818 1819void 1820spa_boot_init() 1821{ 1822 spa_config_load(); 1823} 1824 1825#ifdef _KERNEL 1826EVENTHANDLER_DEFINE(mountroot, spa_boot_init, NULL, 0); 1827#endif 1828 1829void 1830spa_init(int mode) 1831{ 1832 mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL); 1833 mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL); 1834 mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL); 1835 cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL); 1836 1837 avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t), 1838 offsetof(spa_t, spa_avl)); 1839 1840 avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t), 1841 offsetof(spa_aux_t, aux_avl)); 1842 1843 avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t), 1844 offsetof(spa_aux_t, aux_avl)); 1845 1846 spa_mode_global = mode; 1847 1848#ifdef illumos 1849#ifdef _KERNEL 1850 spa_arch_init(); 1851#else 1852 if (spa_mode_global != FREAD && dprintf_find_string("watch")) { 1853 arc_procfd = open("/proc/self/ctl", O_WRONLY); 1854 if (arc_procfd == -1) { 1855 perror("could not enable watchpoints: " 1856 "opening /proc/self/ctl failed: "); 1857 } else { 1858 arc_watch = B_TRUE; 1859 } 1860 } 1861#endif 1862#endif /* illumos */ 1863 refcount_sysinit(); 1864 unique_init(); 1865 range_tree_init(); 1866 zio_init(); 1867 lz4_init(); 1868 dmu_init(); 1869 zil_init(); 1870 vdev_cache_stat_init(); 1871 zfs_prop_init(); 1872 zpool_prop_init(); 1873 zpool_feature_init(); 1874 spa_config_load(); 1875 l2arc_start(); 1876#ifndef illumos 1877#ifdef _KERNEL 1878 zfs_deadman_init(); 1879#endif 1880#endif /* !illumos */ 1881} 1882 1883void 1884spa_fini(void) 1885{ 1886 l2arc_stop(); 1887 1888 spa_evict_all(); 1889 1890 vdev_cache_stat_fini(); 1891 zil_fini(); 1892 dmu_fini(); 1893 lz4_fini(); 1894 zio_fini(); 1895 range_tree_fini(); 1896 unique_fini(); 1897 refcount_fini(); 1898 1899 avl_destroy(&spa_namespace_avl); 1900 avl_destroy(&spa_spare_avl); 1901 avl_destroy(&spa_l2cache_avl); 1902 1903 cv_destroy(&spa_namespace_cv); 1904 mutex_destroy(&spa_namespace_lock); 1905 mutex_destroy(&spa_spare_lock); 1906 mutex_destroy(&spa_l2cache_lock); 1907} 1908 1909/* 1910 * Return whether this pool has slogs. No locking needed. 1911 * It's not a problem if the wrong answer is returned as it's only for 1912 * performance and not correctness 1913 */ 1914boolean_t 1915spa_has_slogs(spa_t *spa) 1916{ 1917 return (spa->spa_log_class->mc_rotor != NULL); 1918} 1919 1920spa_log_state_t 1921spa_get_log_state(spa_t *spa) 1922{ 1923 return (spa->spa_log_state); 1924} 1925 1926void 1927spa_set_log_state(spa_t *spa, spa_log_state_t state) 1928{ 1929 spa->spa_log_state = state; 1930} 1931 1932boolean_t 1933spa_is_root(spa_t *spa) 1934{ 1935 return (spa->spa_is_root); 1936} 1937 1938boolean_t 1939spa_writeable(spa_t *spa) 1940{ 1941 return (!!(spa->spa_mode & FWRITE)); 1942} 1943 1944/* 1945 * Returns true if there is a pending sync task in any of the current 1946 * syncing txg, the current quiescing txg, or the current open txg. 1947 */ 1948boolean_t 1949spa_has_pending_synctask(spa_t *spa) 1950{ 1951 return (!txg_all_lists_empty(&spa->spa_dsl_pool->dp_sync_tasks)); 1952} 1953 1954int 1955spa_mode(spa_t *spa) 1956{ 1957 return (spa->spa_mode); 1958} 1959 1960uint64_t 1961spa_bootfs(spa_t *spa) 1962{ 1963 return (spa->spa_bootfs); 1964} 1965 1966uint64_t 1967spa_delegation(spa_t *spa) 1968{ 1969 return (spa->spa_delegation); 1970} 1971 1972objset_t * 1973spa_meta_objset(spa_t *spa) 1974{ 1975 return (spa->spa_meta_objset); 1976} 1977 1978enum zio_checksum 1979spa_dedup_checksum(spa_t *spa) 1980{ 1981 return (spa->spa_dedup_checksum); 1982} 1983 1984/* 1985 * Reset pool scan stat per scan pass (or reboot). 1986 */ 1987void 1988spa_scan_stat_init(spa_t *spa) 1989{ 1990 /* data not stored on disk */ 1991 spa->spa_scan_pass_start = gethrestime_sec(); 1992 spa->spa_scan_pass_exam = 0; 1993 vdev_scan_stat_init(spa->spa_root_vdev); 1994} 1995 1996/* 1997 * Get scan stats for zpool status reports 1998 */ 1999int 2000spa_scan_get_stats(spa_t *spa, pool_scan_stat_t *ps) 2001{ 2002 dsl_scan_t *scn = spa->spa_dsl_pool ? spa->spa_dsl_pool->dp_scan : NULL; 2003 2004 if (scn == NULL || scn->scn_phys.scn_func == POOL_SCAN_NONE) 2005 return (SET_ERROR(ENOENT)); 2006 bzero(ps, sizeof (pool_scan_stat_t)); 2007 2008 /* data stored on disk */ 2009 ps->pss_func = scn->scn_phys.scn_func; 2010 ps->pss_start_time = scn->scn_phys.scn_start_time; 2011 ps->pss_end_time = scn->scn_phys.scn_end_time; 2012 ps->pss_to_examine = scn->scn_phys.scn_to_examine; 2013 ps->pss_examined = scn->scn_phys.scn_examined; 2014 ps->pss_to_process = scn->scn_phys.scn_to_process; 2015 ps->pss_processed = scn->scn_phys.scn_processed; 2016 ps->pss_errors = scn->scn_phys.scn_errors; 2017 ps->pss_state = scn->scn_phys.scn_state; 2018 2019 /* data not stored on disk */ 2020 ps->pss_pass_start = spa->spa_scan_pass_start; 2021 ps->pss_pass_exam = spa->spa_scan_pass_exam; 2022 2023 return (0); 2024} 2025 2026boolean_t 2027spa_debug_enabled(spa_t *spa) 2028{ 2029 return (spa->spa_debug); 2030} 2031 2032int 2033spa_maxblocksize(spa_t *spa) 2034{ 2035 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) 2036 return (SPA_MAXBLOCKSIZE); 2037 else 2038 return (SPA_OLD_MAXBLOCKSIZE); 2039} 2040