spa_misc.c revision 260763
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 */ 256int zfs_recover = 0; 257SYSCTL_DECL(_vfs_zfs); 258TUNABLE_INT("vfs.zfs.recover", &zfs_recover); 259SYSCTL_INT(_vfs_zfs, OID_AUTO, recover, CTLFLAG_RDTUN, &zfs_recover, 0, 260 "Try to recover from otherwise-fatal errors."); 261 262/* 263 * Expiration time in milliseconds. This value has two meanings. First it is 264 * used to determine when the spa_deadman() logic should fire. By default the 265 * spa_deadman() will fire if spa_sync() has not completed in 1000 seconds. 266 * Secondly, the value determines if an I/O is considered "hung". Any I/O that 267 * has not completed in zfs_deadman_synctime_ms is considered "hung" resulting 268 * in a system panic. 269 */ 270uint64_t zfs_deadman_synctime_ms = 1000000ULL; 271TUNABLE_QUAD("vfs.zfs.deadman_synctime_ms", &zfs_deadman_synctime_ms); 272SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, deadman_synctime_ms, CTLFLAG_RDTUN, 273 &zfs_deadman_synctime_ms, 0, 274 "Stalled ZFS I/O expiration time in milliseconds"); 275 276/* 277 * Check time in milliseconds. This defines the frequency at which we check 278 * for hung I/O. 279 */ 280uint64_t zfs_deadman_checktime_ms = 5000ULL; 281TUNABLE_QUAD("vfs.zfs.deadman_checktime_ms", &zfs_deadman_checktime_ms); 282SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, deadman_checktime_ms, CTLFLAG_RDTUN, 283 &zfs_deadman_checktime_ms, 0, 284 "Period of checks for stalled ZFS I/O in milliseconds"); 285 286/* 287 * Default value of -1 for zfs_deadman_enabled is resolved in 288 * zfs_deadman_init() 289 */ 290int zfs_deadman_enabled = -1; 291TUNABLE_INT("vfs.zfs.deadman_enabled", &zfs_deadman_enabled); 292SYSCTL_INT(_vfs_zfs, OID_AUTO, deadman_enabled, CTLFLAG_RDTUN, 293 &zfs_deadman_enabled, 0, "Kernel panic on stalled ZFS I/O"); 294 295/* 296 * The worst case is single-sector max-parity RAID-Z blocks, in which 297 * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1) 298 * times the size; so just assume that. Add to this the fact that 299 * we can have up to 3 DVAs per bp, and one more factor of 2 because 300 * the block may be dittoed with up to 3 DVAs by ddt_sync(). All together, 301 * the worst case is: 302 * (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2 == 24 303 */ 304int spa_asize_inflation = 24; 305 306#ifndef illumos 307#ifdef _KERNEL 308static void 309zfs_deadman_init() 310{ 311 /* 312 * If we are not i386 or amd64 or in a virtual machine, 313 * disable ZFS deadman thread by default 314 */ 315 if (zfs_deadman_enabled == -1) { 316#if defined(__amd64__) || defined(__i386__) 317 zfs_deadman_enabled = (vm_guest == VM_GUEST_NO) ? 1 : 0; 318#else 319 zfs_deadman_enabled = 0; 320#endif 321 } 322} 323#endif /* _KERNEL */ 324#endif /* !illumos */ 325 326/* 327 * ========================================================================== 328 * SPA config locking 329 * ========================================================================== 330 */ 331static void 332spa_config_lock_init(spa_t *spa) 333{ 334 for (int i = 0; i < SCL_LOCKS; i++) { 335 spa_config_lock_t *scl = &spa->spa_config_lock[i]; 336 mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL); 337 cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL); 338 refcount_create_untracked(&scl->scl_count); 339 scl->scl_writer = NULL; 340 scl->scl_write_wanted = 0; 341 } 342} 343 344static void 345spa_config_lock_destroy(spa_t *spa) 346{ 347 for (int i = 0; i < SCL_LOCKS; i++) { 348 spa_config_lock_t *scl = &spa->spa_config_lock[i]; 349 mutex_destroy(&scl->scl_lock); 350 cv_destroy(&scl->scl_cv); 351 refcount_destroy(&scl->scl_count); 352 ASSERT(scl->scl_writer == NULL); 353 ASSERT(scl->scl_write_wanted == 0); 354 } 355} 356 357int 358spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw) 359{ 360 for (int i = 0; i < SCL_LOCKS; i++) { 361 spa_config_lock_t *scl = &spa->spa_config_lock[i]; 362 if (!(locks & (1 << i))) 363 continue; 364 mutex_enter(&scl->scl_lock); 365 if (rw == RW_READER) { 366 if (scl->scl_writer || scl->scl_write_wanted) { 367 mutex_exit(&scl->scl_lock); 368 spa_config_exit(spa, locks ^ (1 << i), tag); 369 return (0); 370 } 371 } else { 372 ASSERT(scl->scl_writer != curthread); 373 if (!refcount_is_zero(&scl->scl_count)) { 374 mutex_exit(&scl->scl_lock); 375 spa_config_exit(spa, locks ^ (1 << i), tag); 376 return (0); 377 } 378 scl->scl_writer = curthread; 379 } 380 (void) refcount_add(&scl->scl_count, tag); 381 mutex_exit(&scl->scl_lock); 382 } 383 return (1); 384} 385 386void 387spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw) 388{ 389 int wlocks_held = 0; 390 391 ASSERT3U(SCL_LOCKS, <, sizeof (wlocks_held) * NBBY); 392 393 for (int i = 0; i < SCL_LOCKS; i++) { 394 spa_config_lock_t *scl = &spa->spa_config_lock[i]; 395 if (scl->scl_writer == curthread) 396 wlocks_held |= (1 << i); 397 if (!(locks & (1 << i))) 398 continue; 399 mutex_enter(&scl->scl_lock); 400 if (rw == RW_READER) { 401 while (scl->scl_writer || scl->scl_write_wanted) { 402 cv_wait(&scl->scl_cv, &scl->scl_lock); 403 } 404 } else { 405 ASSERT(scl->scl_writer != curthread); 406 while (!refcount_is_zero(&scl->scl_count)) { 407 scl->scl_write_wanted++; 408 cv_wait(&scl->scl_cv, &scl->scl_lock); 409 scl->scl_write_wanted--; 410 } 411 scl->scl_writer = curthread; 412 } 413 (void) refcount_add(&scl->scl_count, tag); 414 mutex_exit(&scl->scl_lock); 415 } 416 ASSERT(wlocks_held <= locks); 417} 418 419void 420spa_config_exit(spa_t *spa, int locks, void *tag) 421{ 422 for (int i = SCL_LOCKS - 1; i >= 0; 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 ASSERT(!refcount_is_zero(&scl->scl_count)); 428 if (refcount_remove(&scl->scl_count, tag) == 0) { 429 ASSERT(scl->scl_writer == NULL || 430 scl->scl_writer == curthread); 431 scl->scl_writer = NULL; /* OK in either case */ 432 cv_broadcast(&scl->scl_cv); 433 } 434 mutex_exit(&scl->scl_lock); 435 } 436} 437 438int 439spa_config_held(spa_t *spa, int locks, krw_t rw) 440{ 441 int locks_held = 0; 442 443 for (int i = 0; i < SCL_LOCKS; i++) { 444 spa_config_lock_t *scl = &spa->spa_config_lock[i]; 445 if (!(locks & (1 << i))) 446 continue; 447 if ((rw == RW_READER && !refcount_is_zero(&scl->scl_count)) || 448 (rw == RW_WRITER && scl->scl_writer == curthread)) 449 locks_held |= 1 << i; 450 } 451 452 return (locks_held); 453} 454 455/* 456 * ========================================================================== 457 * SPA namespace functions 458 * ========================================================================== 459 */ 460 461/* 462 * Lookup the named spa_t in the AVL tree. The spa_namespace_lock must be held. 463 * Returns NULL if no matching spa_t is found. 464 */ 465spa_t * 466spa_lookup(const char *name) 467{ 468 static spa_t search; /* spa_t is large; don't allocate on stack */ 469 spa_t *spa; 470 avl_index_t where; 471 char *cp; 472 473 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 474 475 (void) strlcpy(search.spa_name, name, sizeof (search.spa_name)); 476 477 /* 478 * If it's a full dataset name, figure out the pool name and 479 * just use that. 480 */ 481 cp = strpbrk(search.spa_name, "/@"); 482 if (cp != NULL) 483 *cp = '\0'; 484 485 spa = avl_find(&spa_namespace_avl, &search, &where); 486 487 return (spa); 488} 489 490/* 491 * Fires when spa_sync has not completed within zfs_deadman_synctime_ms. 492 * If the zfs_deadman_enabled flag is set then it inspects all vdev queues 493 * looking for potentially hung I/Os. 494 */ 495void 496spa_deadman(void *arg) 497{ 498 spa_t *spa = arg; 499 500 zfs_dbgmsg("slow spa_sync: started %llu seconds ago, calls %llu", 501 (gethrtime() - spa->spa_sync_starttime) / NANOSEC, 502 ++spa->spa_deadman_calls); 503 if (zfs_deadman_enabled) 504 vdev_deadman(spa->spa_root_vdev); 505} 506 507/* 508 * Create an uninitialized spa_t with the given name. Requires 509 * spa_namespace_lock. The caller must ensure that the spa_t doesn't already 510 * exist by calling spa_lookup() first. 511 */ 512spa_t * 513spa_add(const char *name, nvlist_t *config, const char *altroot) 514{ 515 spa_t *spa; 516 spa_config_dirent_t *dp; 517#ifdef illumos 518 cyc_handler_t hdlr; 519 cyc_time_t when; 520#endif 521 522 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 523 524 spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP); 525 526 mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL); 527 mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL); 528 mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL); 529 mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL); 530 mutex_init(&spa->spa_proc_lock, NULL, MUTEX_DEFAULT, NULL); 531 mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL); 532 mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL); 533 mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL); 534 mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL); 535 536 cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL); 537 cv_init(&spa->spa_proc_cv, NULL, CV_DEFAULT, NULL); 538 cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL); 539 cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL); 540 541 for (int t = 0; t < TXG_SIZE; t++) 542 bplist_create(&spa->spa_free_bplist[t]); 543 544 (void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name)); 545 spa->spa_state = POOL_STATE_UNINITIALIZED; 546 spa->spa_freeze_txg = UINT64_MAX; 547 spa->spa_final_txg = UINT64_MAX; 548 spa->spa_load_max_txg = UINT64_MAX; 549 spa->spa_proc = &p0; 550 spa->spa_proc_state = SPA_PROC_NONE; 551 552#ifdef illumos 553 hdlr.cyh_func = spa_deadman; 554 hdlr.cyh_arg = spa; 555 hdlr.cyh_level = CY_LOW_LEVEL; 556#endif 557 558 spa->spa_deadman_synctime = MSEC2NSEC(zfs_deadman_synctime_ms); 559 560#ifdef illumos 561 /* 562 * This determines how often we need to check for hung I/Os after 563 * the cyclic has already fired. Since checking for hung I/Os is 564 * an expensive operation we don't want to check too frequently. 565 * Instead wait for 5 seconds before checking again. 566 */ 567 when.cyt_interval = MSEC2NSEC(zfs_deadman_checktime_ms); 568 when.cyt_when = CY_INFINITY; 569 mutex_enter(&cpu_lock); 570 spa->spa_deadman_cycid = cyclic_add(&hdlr, &when); 571 mutex_exit(&cpu_lock); 572#else /* !illumos */ 573#ifdef _KERNEL 574 callout_init(&spa->spa_deadman_cycid, CALLOUT_MPSAFE); 575#endif 576#endif 577 refcount_create(&spa->spa_refcount); 578 spa_config_lock_init(spa); 579 580 avl_add(&spa_namespace_avl, spa); 581 582 /* 583 * Set the alternate root, if there is one. 584 */ 585 if (altroot) { 586 spa->spa_root = spa_strdup(altroot); 587 spa_active_count++; 588 } 589 590 /* 591 * Every pool starts with the default cachefile 592 */ 593 list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t), 594 offsetof(spa_config_dirent_t, scd_link)); 595 596 dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP); 597 dp->scd_path = altroot ? NULL : spa_strdup(spa_config_path); 598 list_insert_head(&spa->spa_config_list, dp); 599 600 VERIFY(nvlist_alloc(&spa->spa_load_info, NV_UNIQUE_NAME, 601 KM_SLEEP) == 0); 602 603 if (config != NULL) { 604 nvlist_t *features; 605 606 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ, 607 &features) == 0) { 608 VERIFY(nvlist_dup(features, &spa->spa_label_features, 609 0) == 0); 610 } 611 612 VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0); 613 } 614 615 if (spa->spa_label_features == NULL) { 616 VERIFY(nvlist_alloc(&spa->spa_label_features, NV_UNIQUE_NAME, 617 KM_SLEEP) == 0); 618 } 619 620 spa->spa_debug = ((zfs_flags & ZFS_DEBUG_SPA) != 0); 621 622 return (spa); 623} 624 625/* 626 * Removes a spa_t from the namespace, freeing up any memory used. Requires 627 * spa_namespace_lock. This is called only after the spa_t has been closed and 628 * deactivated. 629 */ 630void 631spa_remove(spa_t *spa) 632{ 633 spa_config_dirent_t *dp; 634 635 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 636 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); 637 638 nvlist_free(spa->spa_config_splitting); 639 640 avl_remove(&spa_namespace_avl, spa); 641 cv_broadcast(&spa_namespace_cv); 642 643 if (spa->spa_root) { 644 spa_strfree(spa->spa_root); 645 spa_active_count--; 646 } 647 648 while ((dp = list_head(&spa->spa_config_list)) != NULL) { 649 list_remove(&spa->spa_config_list, dp); 650 if (dp->scd_path != NULL) 651 spa_strfree(dp->scd_path); 652 kmem_free(dp, sizeof (spa_config_dirent_t)); 653 } 654 655 list_destroy(&spa->spa_config_list); 656 657 nvlist_free(spa->spa_label_features); 658 nvlist_free(spa->spa_load_info); 659 spa_config_set(spa, NULL); 660 661#ifdef illumos 662 mutex_enter(&cpu_lock); 663 if (spa->spa_deadman_cycid != CYCLIC_NONE) 664 cyclic_remove(spa->spa_deadman_cycid); 665 mutex_exit(&cpu_lock); 666 spa->spa_deadman_cycid = CYCLIC_NONE; 667#else /* !illumos */ 668#ifdef _KERNEL 669 callout_drain(&spa->spa_deadman_cycid); 670#endif 671#endif 672 673 refcount_destroy(&spa->spa_refcount); 674 675 spa_config_lock_destroy(spa); 676 677 for (int t = 0; t < TXG_SIZE; t++) 678 bplist_destroy(&spa->spa_free_bplist[t]); 679 680 cv_destroy(&spa->spa_async_cv); 681 cv_destroy(&spa->spa_proc_cv); 682 cv_destroy(&spa->spa_scrub_io_cv); 683 cv_destroy(&spa->spa_suspend_cv); 684 685 mutex_destroy(&spa->spa_async_lock); 686 mutex_destroy(&spa->spa_errlist_lock); 687 mutex_destroy(&spa->spa_errlog_lock); 688 mutex_destroy(&spa->spa_history_lock); 689 mutex_destroy(&spa->spa_proc_lock); 690 mutex_destroy(&spa->spa_props_lock); 691 mutex_destroy(&spa->spa_scrub_lock); 692 mutex_destroy(&spa->spa_suspend_lock); 693 mutex_destroy(&spa->spa_vdev_top_lock); 694 695 kmem_free(spa, sizeof (spa_t)); 696} 697 698/* 699 * Given a pool, return the next pool in the namespace, or NULL if there is 700 * none. If 'prev' is NULL, return the first pool. 701 */ 702spa_t * 703spa_next(spa_t *prev) 704{ 705 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 706 707 if (prev) 708 return (AVL_NEXT(&spa_namespace_avl, prev)); 709 else 710 return (avl_first(&spa_namespace_avl)); 711} 712 713/* 714 * ========================================================================== 715 * SPA refcount functions 716 * ========================================================================== 717 */ 718 719/* 720 * Add a reference to the given spa_t. Must have at least one reference, or 721 * have the namespace lock held. 722 */ 723void 724spa_open_ref(spa_t *spa, void *tag) 725{ 726 ASSERT(refcount_count(&spa->spa_refcount) >= spa->spa_minref || 727 MUTEX_HELD(&spa_namespace_lock)); 728 (void) refcount_add(&spa->spa_refcount, tag); 729} 730 731/* 732 * Remove a reference to the given spa_t. Must have at least one reference, or 733 * have the namespace lock held. 734 */ 735void 736spa_close(spa_t *spa, void *tag) 737{ 738 ASSERT(refcount_count(&spa->spa_refcount) > spa->spa_minref || 739 MUTEX_HELD(&spa_namespace_lock)); 740 (void) refcount_remove(&spa->spa_refcount, tag); 741} 742 743/* 744 * Check to see if the spa refcount is zero. Must be called with 745 * spa_namespace_lock held. We really compare against spa_minref, which is the 746 * number of references acquired when opening a pool 747 */ 748boolean_t 749spa_refcount_zero(spa_t *spa) 750{ 751 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 752 753 return (refcount_count(&spa->spa_refcount) == spa->spa_minref); 754} 755 756/* 757 * ========================================================================== 758 * SPA spare and l2cache tracking 759 * ========================================================================== 760 */ 761 762/* 763 * Hot spares and cache devices are tracked using the same code below, 764 * for 'auxiliary' devices. 765 */ 766 767typedef struct spa_aux { 768 uint64_t aux_guid; 769 uint64_t aux_pool; 770 avl_node_t aux_avl; 771 int aux_count; 772} spa_aux_t; 773 774static int 775spa_aux_compare(const void *a, const void *b) 776{ 777 const spa_aux_t *sa = a; 778 const spa_aux_t *sb = b; 779 780 if (sa->aux_guid < sb->aux_guid) 781 return (-1); 782 else if (sa->aux_guid > sb->aux_guid) 783 return (1); 784 else 785 return (0); 786} 787 788void 789spa_aux_add(vdev_t *vd, avl_tree_t *avl) 790{ 791 avl_index_t where; 792 spa_aux_t search; 793 spa_aux_t *aux; 794 795 search.aux_guid = vd->vdev_guid; 796 if ((aux = avl_find(avl, &search, &where)) != NULL) { 797 aux->aux_count++; 798 } else { 799 aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP); 800 aux->aux_guid = vd->vdev_guid; 801 aux->aux_count = 1; 802 avl_insert(avl, aux, where); 803 } 804} 805 806void 807spa_aux_remove(vdev_t *vd, avl_tree_t *avl) 808{ 809 spa_aux_t search; 810 spa_aux_t *aux; 811 avl_index_t where; 812 813 search.aux_guid = vd->vdev_guid; 814 aux = avl_find(avl, &search, &where); 815 816 ASSERT(aux != NULL); 817 818 if (--aux->aux_count == 0) { 819 avl_remove(avl, aux); 820 kmem_free(aux, sizeof (spa_aux_t)); 821 } else if (aux->aux_pool == spa_guid(vd->vdev_spa)) { 822 aux->aux_pool = 0ULL; 823 } 824} 825 826boolean_t 827spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl) 828{ 829 spa_aux_t search, *found; 830 831 search.aux_guid = guid; 832 found = avl_find(avl, &search, NULL); 833 834 if (pool) { 835 if (found) 836 *pool = found->aux_pool; 837 else 838 *pool = 0ULL; 839 } 840 841 if (refcnt) { 842 if (found) 843 *refcnt = found->aux_count; 844 else 845 *refcnt = 0; 846 } 847 848 return (found != NULL); 849} 850 851void 852spa_aux_activate(vdev_t *vd, avl_tree_t *avl) 853{ 854 spa_aux_t search, *found; 855 avl_index_t where; 856 857 search.aux_guid = vd->vdev_guid; 858 found = avl_find(avl, &search, &where); 859 ASSERT(found != NULL); 860 ASSERT(found->aux_pool == 0ULL); 861 862 found->aux_pool = spa_guid(vd->vdev_spa); 863} 864 865/* 866 * Spares are tracked globally due to the following constraints: 867 * 868 * - A spare may be part of multiple pools. 869 * - A spare may be added to a pool even if it's actively in use within 870 * another pool. 871 * - A spare in use in any pool can only be the source of a replacement if 872 * the target is a spare in the same pool. 873 * 874 * We keep track of all spares on the system through the use of a reference 875 * counted AVL tree. When a vdev is added as a spare, or used as a replacement 876 * spare, then we bump the reference count in the AVL tree. In addition, we set 877 * the 'vdev_isspare' member to indicate that the device is a spare (active or 878 * inactive). When a spare is made active (used to replace a device in the 879 * pool), we also keep track of which pool its been made a part of. 880 * 881 * The 'spa_spare_lock' protects the AVL tree. These functions are normally 882 * called under the spa_namespace lock as part of vdev reconfiguration. The 883 * separate spare lock exists for the status query path, which does not need to 884 * be completely consistent with respect to other vdev configuration changes. 885 */ 886 887static int 888spa_spare_compare(const void *a, const void *b) 889{ 890 return (spa_aux_compare(a, b)); 891} 892 893void 894spa_spare_add(vdev_t *vd) 895{ 896 mutex_enter(&spa_spare_lock); 897 ASSERT(!vd->vdev_isspare); 898 spa_aux_add(vd, &spa_spare_avl); 899 vd->vdev_isspare = B_TRUE; 900 mutex_exit(&spa_spare_lock); 901} 902 903void 904spa_spare_remove(vdev_t *vd) 905{ 906 mutex_enter(&spa_spare_lock); 907 ASSERT(vd->vdev_isspare); 908 spa_aux_remove(vd, &spa_spare_avl); 909 vd->vdev_isspare = B_FALSE; 910 mutex_exit(&spa_spare_lock); 911} 912 913boolean_t 914spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt) 915{ 916 boolean_t found; 917 918 mutex_enter(&spa_spare_lock); 919 found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl); 920 mutex_exit(&spa_spare_lock); 921 922 return (found); 923} 924 925void 926spa_spare_activate(vdev_t *vd) 927{ 928 mutex_enter(&spa_spare_lock); 929 ASSERT(vd->vdev_isspare); 930 spa_aux_activate(vd, &spa_spare_avl); 931 mutex_exit(&spa_spare_lock); 932} 933 934/* 935 * Level 2 ARC devices are tracked globally for the same reasons as spares. 936 * Cache devices currently only support one pool per cache device, and so 937 * for these devices the aux reference count is currently unused beyond 1. 938 */ 939 940static int 941spa_l2cache_compare(const void *a, const void *b) 942{ 943 return (spa_aux_compare(a, b)); 944} 945 946void 947spa_l2cache_add(vdev_t *vd) 948{ 949 mutex_enter(&spa_l2cache_lock); 950 ASSERT(!vd->vdev_isl2cache); 951 spa_aux_add(vd, &spa_l2cache_avl); 952 vd->vdev_isl2cache = B_TRUE; 953 mutex_exit(&spa_l2cache_lock); 954} 955 956void 957spa_l2cache_remove(vdev_t *vd) 958{ 959 mutex_enter(&spa_l2cache_lock); 960 ASSERT(vd->vdev_isl2cache); 961 spa_aux_remove(vd, &spa_l2cache_avl); 962 vd->vdev_isl2cache = B_FALSE; 963 mutex_exit(&spa_l2cache_lock); 964} 965 966boolean_t 967spa_l2cache_exists(uint64_t guid, uint64_t *pool) 968{ 969 boolean_t found; 970 971 mutex_enter(&spa_l2cache_lock); 972 found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl); 973 mutex_exit(&spa_l2cache_lock); 974 975 return (found); 976} 977 978void 979spa_l2cache_activate(vdev_t *vd) 980{ 981 mutex_enter(&spa_l2cache_lock); 982 ASSERT(vd->vdev_isl2cache); 983 spa_aux_activate(vd, &spa_l2cache_avl); 984 mutex_exit(&spa_l2cache_lock); 985} 986 987/* 988 * ========================================================================== 989 * SPA vdev locking 990 * ========================================================================== 991 */ 992 993/* 994 * Lock the given spa_t for the purpose of adding or removing a vdev. 995 * Grabs the global spa_namespace_lock plus the spa config lock for writing. 996 * It returns the next transaction group for the spa_t. 997 */ 998uint64_t 999spa_vdev_enter(spa_t *spa) 1000{ 1001 mutex_enter(&spa->spa_vdev_top_lock); 1002 mutex_enter(&spa_namespace_lock); 1003 return (spa_vdev_config_enter(spa)); 1004} 1005 1006/* 1007 * Internal implementation for spa_vdev_enter(). Used when a vdev 1008 * operation requires multiple syncs (i.e. removing a device) while 1009 * keeping the spa_namespace_lock held. 1010 */ 1011uint64_t 1012spa_vdev_config_enter(spa_t *spa) 1013{ 1014 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1015 1016 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER); 1017 1018 return (spa_last_synced_txg(spa) + 1); 1019} 1020 1021/* 1022 * Used in combination with spa_vdev_config_enter() to allow the syncing 1023 * of multiple transactions without releasing the spa_namespace_lock. 1024 */ 1025void 1026spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag) 1027{ 1028 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1029 1030 int config_changed = B_FALSE; 1031 1032 ASSERT(txg > spa_last_synced_txg(spa)); 1033 1034 spa->spa_pending_vdev = NULL; 1035 1036 /* 1037 * Reassess the DTLs. 1038 */ 1039 vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE); 1040 1041 if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) { 1042 config_changed = B_TRUE; 1043 spa->spa_config_generation++; 1044 } 1045 1046 /* 1047 * Verify the metaslab classes. 1048 */ 1049 ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0); 1050 ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0); 1051 1052 spa_config_exit(spa, SCL_ALL, spa); 1053 1054 /* 1055 * Panic the system if the specified tag requires it. This 1056 * is useful for ensuring that configurations are updated 1057 * transactionally. 1058 */ 1059 if (zio_injection_enabled) 1060 zio_handle_panic_injection(spa, tag, 0); 1061 1062 /* 1063 * Note: this txg_wait_synced() is important because it ensures 1064 * that there won't be more than one config change per txg. 1065 * This allows us to use the txg as the generation number. 1066 */ 1067 if (error == 0) 1068 txg_wait_synced(spa->spa_dsl_pool, txg); 1069 1070 if (vd != NULL) { 1071 ASSERT(!vd->vdev_detached || vd->vdev_dtl_smo.smo_object == 0); 1072 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER); 1073 vdev_free(vd); 1074 spa_config_exit(spa, SCL_ALL, spa); 1075 } 1076 1077 /* 1078 * If the config changed, update the config cache. 1079 */ 1080 if (config_changed) 1081 spa_config_sync(spa, B_FALSE, B_TRUE); 1082} 1083 1084/* 1085 * Unlock the spa_t after adding or removing a vdev. Besides undoing the 1086 * locking of spa_vdev_enter(), we also want make sure the transactions have 1087 * synced to disk, and then update the global configuration cache with the new 1088 * information. 1089 */ 1090int 1091spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error) 1092{ 1093 spa_vdev_config_exit(spa, vd, txg, error, FTAG); 1094 mutex_exit(&spa_namespace_lock); 1095 mutex_exit(&spa->spa_vdev_top_lock); 1096 1097 return (error); 1098} 1099 1100/* 1101 * Lock the given spa_t for the purpose of changing vdev state. 1102 */ 1103void 1104spa_vdev_state_enter(spa_t *spa, int oplocks) 1105{ 1106 int locks = SCL_STATE_ALL | oplocks; 1107 1108 /* 1109 * Root pools may need to read of the underlying devfs filesystem 1110 * when opening up a vdev. Unfortunately if we're holding the 1111 * SCL_ZIO lock it will result in a deadlock when we try to issue 1112 * the read from the root filesystem. Instead we "prefetch" 1113 * the associated vnodes that we need prior to opening the 1114 * underlying devices and cache them so that we can prevent 1115 * any I/O when we are doing the actual open. 1116 */ 1117 if (spa_is_root(spa)) { 1118 int low = locks & ~(SCL_ZIO - 1); 1119 int high = locks & ~low; 1120 1121 spa_config_enter(spa, high, spa, RW_WRITER); 1122 vdev_hold(spa->spa_root_vdev); 1123 spa_config_enter(spa, low, spa, RW_WRITER); 1124 } else { 1125 spa_config_enter(spa, locks, spa, RW_WRITER); 1126 } 1127 spa->spa_vdev_locks = locks; 1128} 1129 1130int 1131spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error) 1132{ 1133 boolean_t config_changed = B_FALSE; 1134 1135 if (vd != NULL || error == 0) 1136 vdev_dtl_reassess(vd ? vd->vdev_top : spa->spa_root_vdev, 1137 0, 0, B_FALSE); 1138 1139 if (vd != NULL) { 1140 vdev_state_dirty(vd->vdev_top); 1141 config_changed = B_TRUE; 1142 spa->spa_config_generation++; 1143 } 1144 1145 if (spa_is_root(spa)) 1146 vdev_rele(spa->spa_root_vdev); 1147 1148 ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL); 1149 spa_config_exit(spa, spa->spa_vdev_locks, spa); 1150 1151 /* 1152 * If anything changed, wait for it to sync. This ensures that, 1153 * from the system administrator's perspective, zpool(1M) commands 1154 * are synchronous. This is important for things like zpool offline: 1155 * when the command completes, you expect no further I/O from ZFS. 1156 */ 1157 if (vd != NULL) 1158 txg_wait_synced(spa->spa_dsl_pool, 0); 1159 1160 /* 1161 * If the config changed, update the config cache. 1162 */ 1163 if (config_changed) { 1164 mutex_enter(&spa_namespace_lock); 1165 spa_config_sync(spa, B_FALSE, B_TRUE); 1166 mutex_exit(&spa_namespace_lock); 1167 } 1168 1169 return (error); 1170} 1171 1172/* 1173 * ========================================================================== 1174 * Miscellaneous functions 1175 * ========================================================================== 1176 */ 1177 1178void 1179spa_activate_mos_feature(spa_t *spa, const char *feature) 1180{ 1181 (void) nvlist_add_boolean(spa->spa_label_features, feature); 1182 vdev_config_dirty(spa->spa_root_vdev); 1183} 1184 1185void 1186spa_deactivate_mos_feature(spa_t *spa, const char *feature) 1187{ 1188 (void) nvlist_remove_all(spa->spa_label_features, feature); 1189 vdev_config_dirty(spa->spa_root_vdev); 1190} 1191 1192/* 1193 * Rename a spa_t. 1194 */ 1195int 1196spa_rename(const char *name, const char *newname) 1197{ 1198 spa_t *spa; 1199 int err; 1200 1201 /* 1202 * Lookup the spa_t and grab the config lock for writing. We need to 1203 * actually open the pool so that we can sync out the necessary labels. 1204 * It's OK to call spa_open() with the namespace lock held because we 1205 * allow recursive calls for other reasons. 1206 */ 1207 mutex_enter(&spa_namespace_lock); 1208 if ((err = spa_open(name, &spa, FTAG)) != 0) { 1209 mutex_exit(&spa_namespace_lock); 1210 return (err); 1211 } 1212 1213 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1214 1215 avl_remove(&spa_namespace_avl, spa); 1216 (void) strlcpy(spa->spa_name, newname, sizeof (spa->spa_name)); 1217 avl_add(&spa_namespace_avl, spa); 1218 1219 /* 1220 * Sync all labels to disk with the new names by marking the root vdev 1221 * dirty and waiting for it to sync. It will pick up the new pool name 1222 * during the sync. 1223 */ 1224 vdev_config_dirty(spa->spa_root_vdev); 1225 1226 spa_config_exit(spa, SCL_ALL, FTAG); 1227 1228 txg_wait_synced(spa->spa_dsl_pool, 0); 1229 1230 /* 1231 * Sync the updated config cache. 1232 */ 1233 spa_config_sync(spa, B_FALSE, B_TRUE); 1234 1235 spa_close(spa, FTAG); 1236 1237 mutex_exit(&spa_namespace_lock); 1238 1239 return (0); 1240} 1241 1242/* 1243 * Return the spa_t associated with given pool_guid, if it exists. If 1244 * device_guid is non-zero, determine whether the pool exists *and* contains 1245 * a device with the specified device_guid. 1246 */ 1247spa_t * 1248spa_by_guid(uint64_t pool_guid, uint64_t device_guid) 1249{ 1250 spa_t *spa; 1251 avl_tree_t *t = &spa_namespace_avl; 1252 1253 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1254 1255 for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) { 1256 if (spa->spa_state == POOL_STATE_UNINITIALIZED) 1257 continue; 1258 if (spa->spa_root_vdev == NULL) 1259 continue; 1260 if (spa_guid(spa) == pool_guid) { 1261 if (device_guid == 0) 1262 break; 1263 1264 if (vdev_lookup_by_guid(spa->spa_root_vdev, 1265 device_guid) != NULL) 1266 break; 1267 1268 /* 1269 * Check any devices we may be in the process of adding. 1270 */ 1271 if (spa->spa_pending_vdev) { 1272 if (vdev_lookup_by_guid(spa->spa_pending_vdev, 1273 device_guid) != NULL) 1274 break; 1275 } 1276 } 1277 } 1278 1279 return (spa); 1280} 1281 1282/* 1283 * Determine whether a pool with the given pool_guid exists. 1284 */ 1285boolean_t 1286spa_guid_exists(uint64_t pool_guid, uint64_t device_guid) 1287{ 1288 return (spa_by_guid(pool_guid, device_guid) != NULL); 1289} 1290 1291char * 1292spa_strdup(const char *s) 1293{ 1294 size_t len; 1295 char *new; 1296 1297 len = strlen(s); 1298 new = kmem_alloc(len + 1, KM_SLEEP); 1299 bcopy(s, new, len); 1300 new[len] = '\0'; 1301 1302 return (new); 1303} 1304 1305void 1306spa_strfree(char *s) 1307{ 1308 kmem_free(s, strlen(s) + 1); 1309} 1310 1311uint64_t 1312spa_get_random(uint64_t range) 1313{ 1314 uint64_t r; 1315 1316 ASSERT(range != 0); 1317 1318 (void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t)); 1319 1320 return (r % range); 1321} 1322 1323uint64_t 1324spa_generate_guid(spa_t *spa) 1325{ 1326 uint64_t guid = spa_get_random(-1ULL); 1327 1328 if (spa != NULL) { 1329 while (guid == 0 || spa_guid_exists(spa_guid(spa), guid)) 1330 guid = spa_get_random(-1ULL); 1331 } else { 1332 while (guid == 0 || spa_guid_exists(guid, 0)) 1333 guid = spa_get_random(-1ULL); 1334 } 1335 1336 return (guid); 1337} 1338 1339void 1340sprintf_blkptr(char *buf, const blkptr_t *bp) 1341{ 1342 char type[256]; 1343 char *checksum = NULL; 1344 char *compress = NULL; 1345 1346 if (bp != NULL) { 1347 if (BP_GET_TYPE(bp) & DMU_OT_NEWTYPE) { 1348 dmu_object_byteswap_t bswap = 1349 DMU_OT_BYTESWAP(BP_GET_TYPE(bp)); 1350 (void) snprintf(type, sizeof (type), "bswap %s %s", 1351 DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) ? 1352 "metadata" : "data", 1353 dmu_ot_byteswap[bswap].ob_name); 1354 } else { 1355 (void) strlcpy(type, dmu_ot[BP_GET_TYPE(bp)].ot_name, 1356 sizeof (type)); 1357 } 1358 checksum = zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name; 1359 compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name; 1360 } 1361 1362 SPRINTF_BLKPTR(snprintf, ' ', buf, bp, type, checksum, compress); 1363} 1364 1365void 1366spa_freeze(spa_t *spa) 1367{ 1368 uint64_t freeze_txg = 0; 1369 1370 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1371 if (spa->spa_freeze_txg == UINT64_MAX) { 1372 freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE; 1373 spa->spa_freeze_txg = freeze_txg; 1374 } 1375 spa_config_exit(spa, SCL_ALL, FTAG); 1376 if (freeze_txg != 0) 1377 txg_wait_synced(spa_get_dsl(spa), freeze_txg); 1378} 1379 1380void 1381zfs_panic_recover(const char *fmt, ...) 1382{ 1383 va_list adx; 1384 1385 va_start(adx, fmt); 1386 vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx); 1387 va_end(adx); 1388} 1389 1390/* 1391 * This is a stripped-down version of strtoull, suitable only for converting 1392 * lowercase hexadecimal numbers that don't overflow. 1393 */ 1394uint64_t 1395zfs_strtonum(const char *str, char **nptr) 1396{ 1397 uint64_t val = 0; 1398 char c; 1399 int digit; 1400 1401 while ((c = *str) != '\0') { 1402 if (c >= '0' && c <= '9') 1403 digit = c - '0'; 1404 else if (c >= 'a' && c <= 'f') 1405 digit = 10 + c - 'a'; 1406 else 1407 break; 1408 1409 val *= 16; 1410 val += digit; 1411 1412 str++; 1413 } 1414 1415 if (nptr) 1416 *nptr = (char *)str; 1417 1418 return (val); 1419} 1420 1421/* 1422 * ========================================================================== 1423 * Accessor functions 1424 * ========================================================================== 1425 */ 1426 1427boolean_t 1428spa_shutting_down(spa_t *spa) 1429{ 1430 return (spa->spa_async_suspended); 1431} 1432 1433dsl_pool_t * 1434spa_get_dsl(spa_t *spa) 1435{ 1436 return (spa->spa_dsl_pool); 1437} 1438 1439boolean_t 1440spa_is_initializing(spa_t *spa) 1441{ 1442 return (spa->spa_is_initializing); 1443} 1444 1445blkptr_t * 1446spa_get_rootblkptr(spa_t *spa) 1447{ 1448 return (&spa->spa_ubsync.ub_rootbp); 1449} 1450 1451void 1452spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp) 1453{ 1454 spa->spa_uberblock.ub_rootbp = *bp; 1455} 1456 1457void 1458spa_altroot(spa_t *spa, char *buf, size_t buflen) 1459{ 1460 if (spa->spa_root == NULL) 1461 buf[0] = '\0'; 1462 else 1463 (void) strncpy(buf, spa->spa_root, buflen); 1464} 1465 1466int 1467spa_sync_pass(spa_t *spa) 1468{ 1469 return (spa->spa_sync_pass); 1470} 1471 1472char * 1473spa_name(spa_t *spa) 1474{ 1475 return (spa->spa_name); 1476} 1477 1478uint64_t 1479spa_guid(spa_t *spa) 1480{ 1481 dsl_pool_t *dp = spa_get_dsl(spa); 1482 uint64_t guid; 1483 1484 /* 1485 * If we fail to parse the config during spa_load(), we can go through 1486 * the error path (which posts an ereport) and end up here with no root 1487 * vdev. We stash the original pool guid in 'spa_config_guid' to handle 1488 * this case. 1489 */ 1490 if (spa->spa_root_vdev == NULL) 1491 return (spa->spa_config_guid); 1492 1493 guid = spa->spa_last_synced_guid != 0 ? 1494 spa->spa_last_synced_guid : spa->spa_root_vdev->vdev_guid; 1495 1496 /* 1497 * Return the most recently synced out guid unless we're 1498 * in syncing context. 1499 */ 1500 if (dp && dsl_pool_sync_context(dp)) 1501 return (spa->spa_root_vdev->vdev_guid); 1502 else 1503 return (guid); 1504} 1505 1506uint64_t 1507spa_load_guid(spa_t *spa) 1508{ 1509 /* 1510 * This is a GUID that exists solely as a reference for the 1511 * purposes of the arc. It is generated at load time, and 1512 * is never written to persistent storage. 1513 */ 1514 return (spa->spa_load_guid); 1515} 1516 1517uint64_t 1518spa_last_synced_txg(spa_t *spa) 1519{ 1520 return (spa->spa_ubsync.ub_txg); 1521} 1522 1523uint64_t 1524spa_first_txg(spa_t *spa) 1525{ 1526 return (spa->spa_first_txg); 1527} 1528 1529uint64_t 1530spa_syncing_txg(spa_t *spa) 1531{ 1532 return (spa->spa_syncing_txg); 1533} 1534 1535pool_state_t 1536spa_state(spa_t *spa) 1537{ 1538 return (spa->spa_state); 1539} 1540 1541spa_load_state_t 1542spa_load_state(spa_t *spa) 1543{ 1544 return (spa->spa_load_state); 1545} 1546 1547uint64_t 1548spa_freeze_txg(spa_t *spa) 1549{ 1550 return (spa->spa_freeze_txg); 1551} 1552 1553/* ARGSUSED */ 1554uint64_t 1555spa_get_asize(spa_t *spa, uint64_t lsize) 1556{ 1557 return (lsize * spa_asize_inflation); 1558} 1559 1560uint64_t 1561spa_get_dspace(spa_t *spa) 1562{ 1563 return (spa->spa_dspace); 1564} 1565 1566void 1567spa_update_dspace(spa_t *spa) 1568{ 1569 spa->spa_dspace = metaslab_class_get_dspace(spa_normal_class(spa)) + 1570 ddt_get_dedup_dspace(spa); 1571} 1572 1573/* 1574 * Return the failure mode that has been set to this pool. The default 1575 * behavior will be to block all I/Os when a complete failure occurs. 1576 */ 1577uint8_t 1578spa_get_failmode(spa_t *spa) 1579{ 1580 return (spa->spa_failmode); 1581} 1582 1583boolean_t 1584spa_suspended(spa_t *spa) 1585{ 1586 return (spa->spa_suspended); 1587} 1588 1589uint64_t 1590spa_version(spa_t *spa) 1591{ 1592 return (spa->spa_ubsync.ub_version); 1593} 1594 1595boolean_t 1596spa_deflate(spa_t *spa) 1597{ 1598 return (spa->spa_deflate); 1599} 1600 1601metaslab_class_t * 1602spa_normal_class(spa_t *spa) 1603{ 1604 return (spa->spa_normal_class); 1605} 1606 1607metaslab_class_t * 1608spa_log_class(spa_t *spa) 1609{ 1610 return (spa->spa_log_class); 1611} 1612 1613int 1614spa_max_replication(spa_t *spa) 1615{ 1616 /* 1617 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to 1618 * handle BPs with more than one DVA allocated. Set our max 1619 * replication level accordingly. 1620 */ 1621 if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS) 1622 return (1); 1623 return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override)); 1624} 1625 1626int 1627spa_prev_software_version(spa_t *spa) 1628{ 1629 return (spa->spa_prev_software_version); 1630} 1631 1632uint64_t 1633spa_deadman_synctime(spa_t *spa) 1634{ 1635 return (spa->spa_deadman_synctime); 1636} 1637 1638uint64_t 1639dva_get_dsize_sync(spa_t *spa, const dva_t *dva) 1640{ 1641 uint64_t asize = DVA_GET_ASIZE(dva); 1642 uint64_t dsize = asize; 1643 1644 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 1645 1646 if (asize != 0 && spa->spa_deflate) { 1647 vdev_t *vd = vdev_lookup_top(spa, DVA_GET_VDEV(dva)); 1648 dsize = (asize >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio; 1649 } 1650 1651 return (dsize); 1652} 1653 1654uint64_t 1655bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp) 1656{ 1657 uint64_t dsize = 0; 1658 1659 for (int d = 0; d < SPA_DVAS_PER_BP; d++) 1660 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]); 1661 1662 return (dsize); 1663} 1664 1665uint64_t 1666bp_get_dsize(spa_t *spa, const blkptr_t *bp) 1667{ 1668 uint64_t dsize = 0; 1669 1670 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); 1671 1672 for (int d = 0; d < SPA_DVAS_PER_BP; d++) 1673 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]); 1674 1675 spa_config_exit(spa, SCL_VDEV, FTAG); 1676 1677 return (dsize); 1678} 1679 1680/* 1681 * ========================================================================== 1682 * Initialization and Termination 1683 * ========================================================================== 1684 */ 1685 1686static int 1687spa_name_compare(const void *a1, const void *a2) 1688{ 1689 const spa_t *s1 = a1; 1690 const spa_t *s2 = a2; 1691 int s; 1692 1693 s = strcmp(s1->spa_name, s2->spa_name); 1694 if (s > 0) 1695 return (1); 1696 if (s < 0) 1697 return (-1); 1698 return (0); 1699} 1700 1701int 1702spa_busy(void) 1703{ 1704 return (spa_active_count); 1705} 1706 1707void 1708spa_boot_init() 1709{ 1710 spa_config_load(); 1711} 1712 1713#ifdef _KERNEL 1714EVENTHANDLER_DEFINE(mountroot, spa_boot_init, NULL, 0); 1715#endif 1716 1717void 1718spa_init(int mode) 1719{ 1720 mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL); 1721 mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL); 1722 mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL); 1723 cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL); 1724 1725 avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t), 1726 offsetof(spa_t, spa_avl)); 1727 1728 avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t), 1729 offsetof(spa_aux_t, aux_avl)); 1730 1731 avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t), 1732 offsetof(spa_aux_t, aux_avl)); 1733 1734 spa_mode_global = mode; 1735 1736#ifdef illumos 1737#ifdef _KERNEL 1738 spa_arch_init(); 1739#else 1740 if (spa_mode_global != FREAD && dprintf_find_string("watch")) { 1741 arc_procfd = open("/proc/self/ctl", O_WRONLY); 1742 if (arc_procfd == -1) { 1743 perror("could not enable watchpoints: " 1744 "opening /proc/self/ctl failed: "); 1745 } else { 1746 arc_watch = B_TRUE; 1747 } 1748 } 1749#endif 1750#endif /* illumos */ 1751 refcount_sysinit(); 1752 unique_init(); 1753 space_map_init(); 1754 zio_init(); 1755 lz4_init(); 1756 dmu_init(); 1757 zil_init(); 1758 vdev_cache_stat_init(); 1759 zfs_prop_init(); 1760 zpool_prop_init(); 1761 zpool_feature_init(); 1762 spa_config_load(); 1763 l2arc_start(); 1764#ifndef illumos 1765#ifdef _KERNEL 1766 zfs_deadman_init(); 1767#endif 1768#endif /* !illumos */ 1769} 1770 1771void 1772spa_fini(void) 1773{ 1774 l2arc_stop(); 1775 1776 spa_evict_all(); 1777 1778 vdev_cache_stat_fini(); 1779 zil_fini(); 1780 dmu_fini(); 1781 lz4_fini(); 1782 zio_fini(); 1783 space_map_fini(); 1784 unique_fini(); 1785 refcount_fini(); 1786 1787 avl_destroy(&spa_namespace_avl); 1788 avl_destroy(&spa_spare_avl); 1789 avl_destroy(&spa_l2cache_avl); 1790 1791 cv_destroy(&spa_namespace_cv); 1792 mutex_destroy(&spa_namespace_lock); 1793 mutex_destroy(&spa_spare_lock); 1794 mutex_destroy(&spa_l2cache_lock); 1795} 1796 1797/* 1798 * Return whether this pool has slogs. No locking needed. 1799 * It's not a problem if the wrong answer is returned as it's only for 1800 * performance and not correctness 1801 */ 1802boolean_t 1803spa_has_slogs(spa_t *spa) 1804{ 1805 return (spa->spa_log_class->mc_rotor != NULL); 1806} 1807 1808spa_log_state_t 1809spa_get_log_state(spa_t *spa) 1810{ 1811 return (spa->spa_log_state); 1812} 1813 1814void 1815spa_set_log_state(spa_t *spa, spa_log_state_t state) 1816{ 1817 spa->spa_log_state = state; 1818} 1819 1820boolean_t 1821spa_is_root(spa_t *spa) 1822{ 1823 return (spa->spa_is_root); 1824} 1825 1826boolean_t 1827spa_writeable(spa_t *spa) 1828{ 1829 return (!!(spa->spa_mode & FWRITE)); 1830} 1831 1832int 1833spa_mode(spa_t *spa) 1834{ 1835 return (spa->spa_mode); 1836} 1837 1838uint64_t 1839spa_bootfs(spa_t *spa) 1840{ 1841 return (spa->spa_bootfs); 1842} 1843 1844uint64_t 1845spa_delegation(spa_t *spa) 1846{ 1847 return (spa->spa_delegation); 1848} 1849 1850objset_t * 1851spa_meta_objset(spa_t *spa) 1852{ 1853 return (spa->spa_meta_objset); 1854} 1855 1856enum zio_checksum 1857spa_dedup_checksum(spa_t *spa) 1858{ 1859 return (spa->spa_dedup_checksum); 1860} 1861 1862/* 1863 * Reset pool scan stat per scan pass (or reboot). 1864 */ 1865void 1866spa_scan_stat_init(spa_t *spa) 1867{ 1868 /* data not stored on disk */ 1869 spa->spa_scan_pass_start = gethrestime_sec(); 1870 spa->spa_scan_pass_exam = 0; 1871 vdev_scan_stat_init(spa->spa_root_vdev); 1872} 1873 1874/* 1875 * Get scan stats for zpool status reports 1876 */ 1877int 1878spa_scan_get_stats(spa_t *spa, pool_scan_stat_t *ps) 1879{ 1880 dsl_scan_t *scn = spa->spa_dsl_pool ? spa->spa_dsl_pool->dp_scan : NULL; 1881 1882 if (scn == NULL || scn->scn_phys.scn_func == POOL_SCAN_NONE) 1883 return (SET_ERROR(ENOENT)); 1884 bzero(ps, sizeof (pool_scan_stat_t)); 1885 1886 /* data stored on disk */ 1887 ps->pss_func = scn->scn_phys.scn_func; 1888 ps->pss_start_time = scn->scn_phys.scn_start_time; 1889 ps->pss_end_time = scn->scn_phys.scn_end_time; 1890 ps->pss_to_examine = scn->scn_phys.scn_to_examine; 1891 ps->pss_examined = scn->scn_phys.scn_examined; 1892 ps->pss_to_process = scn->scn_phys.scn_to_process; 1893 ps->pss_processed = scn->scn_phys.scn_processed; 1894 ps->pss_errors = scn->scn_phys.scn_errors; 1895 ps->pss_state = scn->scn_phys.scn_state; 1896 1897 /* data not stored on disk */ 1898 ps->pss_pass_start = spa->spa_scan_pass_start; 1899 ps->pss_pass_exam = spa->spa_scan_pass_exam; 1900 1901 return (0); 1902} 1903 1904boolean_t 1905spa_debug_enabled(spa_t *spa) 1906{ 1907 return (spa->spa_debug); 1908} 1909