spa_misc.c revision 263407
1/* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21/* 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23 * Copyright (c) 2013 by Delphix. All rights reserved. 24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved. 25 * Copyright 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved. 26 */ 27 28#include <sys/zfs_context.h> 29#include <sys/spa_impl.h> 30#include <sys/spa_boot.h> 31#include <sys/zio.h> 32#include <sys/zio_checksum.h> 33#include <sys/zio_compress.h> 34#include <sys/dmu.h> 35#include <sys/dmu_tx.h> 36#include <sys/zap.h> 37#include <sys/zil.h> 38#include <sys/vdev_impl.h> 39#include <sys/metaslab.h> 40#include <sys/uberblock_impl.h> 41#include <sys/txg.h> 42#include <sys/avl.h> 43#include <sys/unique.h> 44#include <sys/dsl_pool.h> 45#include <sys/dsl_dir.h> 46#include <sys/dsl_prop.h> 47#include <sys/dsl_scan.h> 48#include <sys/fs/zfs.h> 49#include <sys/metaslab_impl.h> 50#include <sys/arc.h> 51#include <sys/ddt.h> 52#include "zfs_prop.h" 53#include "zfeature_common.h" 54 55/* 56 * SPA locking 57 * 58 * There are four basic locks for managing spa_t structures: 59 * 60 * spa_namespace_lock (global mutex) 61 * 62 * This lock must be acquired to do any of the following: 63 * 64 * - Lookup a spa_t by name 65 * - Add or remove a spa_t from the namespace 66 * - Increase spa_refcount from non-zero 67 * - Check if spa_refcount is zero 68 * - Rename a spa_t 69 * - add/remove/attach/detach devices 70 * - Held for the duration of create/destroy/import/export 71 * 72 * It does not need to handle recursion. A create or destroy may 73 * reference objects (files or zvols) in other pools, but by 74 * definition they must have an existing reference, and will never need 75 * to lookup a spa_t by name. 76 * 77 * spa_refcount (per-spa refcount_t protected by mutex) 78 * 79 * This reference count keep track of any active users of the spa_t. The 80 * spa_t cannot be destroyed or freed while this is non-zero. Internally, 81 * the refcount is never really 'zero' - opening a pool implicitly keeps 82 * some references in the DMU. Internally we check against spa_minref, but 83 * present the image of a zero/non-zero value to consumers. 84 * 85 * spa_config_lock[] (per-spa array of rwlocks) 86 * 87 * This protects the spa_t from config changes, and must be held in 88 * the following circumstances: 89 * 90 * - RW_READER to perform I/O to the spa 91 * - RW_WRITER to change the vdev config 92 * 93 * The locking order is fairly straightforward: 94 * 95 * spa_namespace_lock -> spa_refcount 96 * 97 * The namespace lock must be acquired to increase the refcount from 0 98 * or to check if it is zero. 99 * 100 * spa_refcount -> spa_config_lock[] 101 * 102 * There must be at least one valid reference on the spa_t to acquire 103 * the config lock. 104 * 105 * spa_namespace_lock -> spa_config_lock[] 106 * 107 * The namespace lock must always be taken before the config lock. 108 * 109 * 110 * The spa_namespace_lock can be acquired directly and is globally visible. 111 * 112 * The namespace is manipulated using the following functions, all of which 113 * require the spa_namespace_lock to be held. 114 * 115 * spa_lookup() Lookup a spa_t by name. 116 * 117 * spa_add() Create a new spa_t in the namespace. 118 * 119 * spa_remove() Remove a spa_t from the namespace. This also 120 * frees up any memory associated with the spa_t. 121 * 122 * spa_next() Returns the next spa_t in the system, or the 123 * first if NULL is passed. 124 * 125 * spa_evict_all() Shutdown and remove all spa_t structures in 126 * the system. 127 * 128 * spa_guid_exists() Determine whether a pool/device guid exists. 129 * 130 * The spa_refcount is manipulated using the following functions: 131 * 132 * spa_open_ref() Adds a reference to the given spa_t. Must be 133 * called with spa_namespace_lock held if the 134 * refcount is currently zero. 135 * 136 * spa_close() Remove a reference from the spa_t. This will 137 * not free the spa_t or remove it from the 138 * namespace. No locking is required. 139 * 140 * spa_refcount_zero() Returns true if the refcount is currently 141 * zero. Must be called with spa_namespace_lock 142 * held. 143 * 144 * The spa_config_lock[] is an array of rwlocks, ordered as follows: 145 * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV. 146 * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}(). 147 * 148 * To read the configuration, it suffices to hold one of these locks as reader. 149 * To modify the configuration, you must hold all locks as writer. To modify 150 * vdev state without altering the vdev tree's topology (e.g. online/offline), 151 * you must hold SCL_STATE and SCL_ZIO as writer. 152 * 153 * We use these distinct config locks to avoid recursive lock entry. 154 * For example, spa_sync() (which holds SCL_CONFIG as reader) induces 155 * block allocations (SCL_ALLOC), which may require reading space maps 156 * from disk (dmu_read() -> zio_read() -> SCL_ZIO). 157 * 158 * The spa config locks cannot be normal rwlocks because we need the 159 * ability to hand off ownership. For example, SCL_ZIO is acquired 160 * by the issuing thread and later released by an interrupt thread. 161 * They do, however, obey the usual write-wanted semantics to prevent 162 * writer (i.e. system administrator) starvation. 163 * 164 * The lock acquisition rules are as follows: 165 * 166 * SCL_CONFIG 167 * Protects changes to the vdev tree topology, such as vdev 168 * add/remove/attach/detach. Protects the dirty config list 169 * (spa_config_dirty_list) and the set of spares and l2arc devices. 170 * 171 * SCL_STATE 172 * Protects changes to pool state and vdev state, such as vdev 173 * online/offline/fault/degrade/clear. Protects the dirty state list 174 * (spa_state_dirty_list) and global pool state (spa_state). 175 * 176 * SCL_ALLOC 177 * Protects changes to metaslab groups and classes. 178 * Held as reader by metaslab_alloc() and metaslab_claim(). 179 * 180 * SCL_ZIO 181 * Held by bp-level zios (those which have no io_vd upon entry) 182 * to prevent changes to the vdev tree. The bp-level zio implicitly 183 * protects all of its vdev child zios, which do not hold SCL_ZIO. 184 * 185 * SCL_FREE 186 * Protects changes to metaslab groups and classes. 187 * Held as reader by metaslab_free(). SCL_FREE is distinct from 188 * SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free 189 * blocks in zio_done() while another i/o that holds either 190 * SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete. 191 * 192 * SCL_VDEV 193 * Held as reader to prevent changes to the vdev tree during trivial 194 * inquiries such as bp_get_dsize(). SCL_VDEV is distinct from the 195 * other locks, and lower than all of them, to ensure that it's safe 196 * to acquire regardless of caller context. 197 * 198 * In addition, the following rules apply: 199 * 200 * (a) spa_props_lock protects pool properties, spa_config and spa_config_list. 201 * The lock ordering is SCL_CONFIG > spa_props_lock. 202 * 203 * (b) I/O operations on leaf vdevs. For any zio operation that takes 204 * an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(), 205 * or zio_write_phys() -- the caller must ensure that the config cannot 206 * cannot change in the interim, and that the vdev cannot be reopened. 207 * SCL_STATE as reader suffices for both. 208 * 209 * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit(). 210 * 211 * spa_vdev_enter() Acquire the namespace lock and the config lock 212 * for writing. 213 * 214 * spa_vdev_exit() Release the config lock, wait for all I/O 215 * to complete, sync the updated configs to the 216 * cache, and release the namespace lock. 217 * 218 * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit(). 219 * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual 220 * locking is, always, based on spa_namespace_lock and spa_config_lock[]. 221 * 222 * spa_rename() is also implemented within this file since it requires 223 * manipulation of the namespace. 224 */ 225 226static avl_tree_t spa_namespace_avl; 227kmutex_t spa_namespace_lock; 228static kcondvar_t spa_namespace_cv; 229static int spa_active_count; 230int spa_max_replication_override = SPA_DVAS_PER_BP; 231 232static kmutex_t spa_spare_lock; 233static avl_tree_t spa_spare_avl; 234static kmutex_t spa_l2cache_lock; 235static avl_tree_t spa_l2cache_avl; 236 237kmem_cache_t *spa_buffer_pool; 238int spa_mode_global; 239 240#ifdef ZFS_DEBUG 241/* Everything except dprintf and spa is on by default in debug builds */ 242int zfs_flags = ~(ZFS_DEBUG_DPRINTF | ZFS_DEBUG_SPA); 243#else 244int zfs_flags = 0; 245#endif 246SYSCTL_DECL(_debug); 247TUNABLE_INT("debug.zfs_flags", &zfs_flags); 248SYSCTL_INT(_debug, OID_AUTO, zfs_flags, CTLFLAG_RWTUN, &zfs_flags, 0, 249 "ZFS debug flags."); 250 251/* 252 * zfs_recover can be set to nonzero to attempt to recover from 253 * otherwise-fatal errors, typically caused by on-disk corruption. When 254 * set, calls to zfs_panic_recover() will turn into warning messages. 255 * This should only be used as a last resort, as it typically results 256 * in leaked space, or worse. 257 */ 258int zfs_recover = 0; 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 * Expiration time in milliseconds. This value has two meanings. First it is 266 * used to determine when the spa_deadman() logic should fire. By default the 267 * spa_deadman() will fire if spa_sync() has not completed in 1000 seconds. 268 * Secondly, the value determines if an I/O is considered "hung". Any I/O that 269 * has not completed in zfs_deadman_synctime_ms is considered "hung" resulting 270 * in a system panic. 271 */ 272uint64_t zfs_deadman_synctime_ms = 1000000ULL; 273TUNABLE_QUAD("vfs.zfs.deadman_synctime_ms", &zfs_deadman_synctime_ms); 274SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, deadman_synctime_ms, CTLFLAG_RDTUN, 275 &zfs_deadman_synctime_ms, 0, 276 "Stalled ZFS I/O expiration time in milliseconds"); 277 278/* 279 * Check time in milliseconds. This defines the frequency at which we check 280 * for hung I/O. 281 */ 282uint64_t zfs_deadman_checktime_ms = 5000ULL; 283TUNABLE_QUAD("vfs.zfs.deadman_checktime_ms", &zfs_deadman_checktime_ms); 284SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, deadman_checktime_ms, CTLFLAG_RDTUN, 285 &zfs_deadman_checktime_ms, 0, 286 "Period of checks for stalled ZFS I/O in milliseconds"); 287 288/* 289 * Default value of -1 for zfs_deadman_enabled is resolved in 290 * zfs_deadman_init() 291 */ 292int zfs_deadman_enabled = -1; 293TUNABLE_INT("vfs.zfs.deadman_enabled", &zfs_deadman_enabled); 294SYSCTL_INT(_vfs_zfs, OID_AUTO, deadman_enabled, CTLFLAG_RDTUN, 295 &zfs_deadman_enabled, 0, "Kernel panic on stalled ZFS I/O"); 296 297/* 298 * The worst case is single-sector max-parity RAID-Z blocks, in which 299 * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1) 300 * times the size; so just assume that. Add to this the fact that 301 * we can have up to 3 DVAs per bp, and one more factor of 2 because 302 * the block may be dittoed with up to 3 DVAs by ddt_sync(). All together, 303 * the worst case is: 304 * (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2 == 24 305 */ 306int spa_asize_inflation = 24; 307TUNABLE_INT("vfs.zfs.spa_asize_inflation", &spa_asize_inflation); 308SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_asize_inflation, CTLFLAG_RWTUN, 309 &spa_asize_inflation, 0, "Worst case inflation factor for single sector writes"); 310 311#ifndef illumos 312#ifdef _KERNEL 313static void 314zfs_deadman_init() 315{ 316 /* 317 * If we are not i386 or amd64 or in a virtual machine, 318 * disable ZFS deadman thread by default 319 */ 320 if (zfs_deadman_enabled == -1) { 321#if defined(__amd64__) || defined(__i386__) 322 zfs_deadman_enabled = (vm_guest == VM_GUEST_NO) ? 1 : 0; 323#else 324 zfs_deadman_enabled = 0; 325#endif 326 } 327} 328#endif /* _KERNEL */ 329#endif /* !illumos */ 330 331/* 332 * ========================================================================== 333 * SPA config locking 334 * ========================================================================== 335 */ 336static void 337spa_config_lock_init(spa_t *spa) 338{ 339 for (int i = 0; i < SCL_LOCKS; i++) { 340 spa_config_lock_t *scl = &spa->spa_config_lock[i]; 341 mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL); 342 cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL); 343 refcount_create_untracked(&scl->scl_count); 344 scl->scl_writer = NULL; 345 scl->scl_write_wanted = 0; 346 } 347} 348 349static void 350spa_config_lock_destroy(spa_t *spa) 351{ 352 for (int i = 0; i < SCL_LOCKS; i++) { 353 spa_config_lock_t *scl = &spa->spa_config_lock[i]; 354 mutex_destroy(&scl->scl_lock); 355 cv_destroy(&scl->scl_cv); 356 refcount_destroy(&scl->scl_count); 357 ASSERT(scl->scl_writer == NULL); 358 ASSERT(scl->scl_write_wanted == 0); 359 } 360} 361 362int 363spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw) 364{ 365 for (int i = 0; i < SCL_LOCKS; i++) { 366 spa_config_lock_t *scl = &spa->spa_config_lock[i]; 367 if (!(locks & (1 << i))) 368 continue; 369 mutex_enter(&scl->scl_lock); 370 if (rw == RW_READER) { 371 if (scl->scl_writer || scl->scl_write_wanted) { 372 mutex_exit(&scl->scl_lock); 373 spa_config_exit(spa, locks ^ (1 << i), tag); 374 return (0); 375 } 376 } else { 377 ASSERT(scl->scl_writer != curthread); 378 if (!refcount_is_zero(&scl->scl_count)) { 379 mutex_exit(&scl->scl_lock); 380 spa_config_exit(spa, locks ^ (1 << i), tag); 381 return (0); 382 } 383 scl->scl_writer = curthread; 384 } 385 (void) refcount_add(&scl->scl_count, tag); 386 mutex_exit(&scl->scl_lock); 387 } 388 return (1); 389} 390 391void 392spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw) 393{ 394 int wlocks_held = 0; 395 396 ASSERT3U(SCL_LOCKS, <, sizeof (wlocks_held) * NBBY); 397 398 for (int i = 0; i < SCL_LOCKS; i++) { 399 spa_config_lock_t *scl = &spa->spa_config_lock[i]; 400 if (scl->scl_writer == curthread) 401 wlocks_held |= (1 << i); 402 if (!(locks & (1 << i))) 403 continue; 404 mutex_enter(&scl->scl_lock); 405 if (rw == RW_READER) { 406 while (scl->scl_writer || scl->scl_write_wanted) { 407 cv_wait(&scl->scl_cv, &scl->scl_lock); 408 } 409 } else { 410 ASSERT(scl->scl_writer != curthread); 411 while (!refcount_is_zero(&scl->scl_count)) { 412 scl->scl_write_wanted++; 413 cv_wait(&scl->scl_cv, &scl->scl_lock); 414 scl->scl_write_wanted--; 415 } 416 scl->scl_writer = curthread; 417 } 418 (void) refcount_add(&scl->scl_count, tag); 419 mutex_exit(&scl->scl_lock); 420 } 421 ASSERT(wlocks_held <= locks); 422} 423 424void 425spa_config_exit(spa_t *spa, int locks, void *tag) 426{ 427 for (int i = SCL_LOCKS - 1; i >= 0; i--) { 428 spa_config_lock_t *scl = &spa->spa_config_lock[i]; 429 if (!(locks & (1 << i))) 430 continue; 431 mutex_enter(&scl->scl_lock); 432 ASSERT(!refcount_is_zero(&scl->scl_count)); 433 if (refcount_remove(&scl->scl_count, tag) == 0) { 434 ASSERT(scl->scl_writer == NULL || 435 scl->scl_writer == curthread); 436 scl->scl_writer = NULL; /* OK in either case */ 437 cv_broadcast(&scl->scl_cv); 438 } 439 mutex_exit(&scl->scl_lock); 440 } 441} 442 443int 444spa_config_held(spa_t *spa, int locks, krw_t rw) 445{ 446 int locks_held = 0; 447 448 for (int i = 0; i < SCL_LOCKS; i++) { 449 spa_config_lock_t *scl = &spa->spa_config_lock[i]; 450 if (!(locks & (1 << i))) 451 continue; 452 if ((rw == RW_READER && !refcount_is_zero(&scl->scl_count)) || 453 (rw == RW_WRITER && scl->scl_writer == curthread)) 454 locks_held |= 1 << i; 455 } 456 457 return (locks_held); 458} 459 460/* 461 * ========================================================================== 462 * SPA namespace functions 463 * ========================================================================== 464 */ 465 466/* 467 * Lookup the named spa_t in the AVL tree. The spa_namespace_lock must be held. 468 * Returns NULL if no matching spa_t is found. 469 */ 470spa_t * 471spa_lookup(const char *name) 472{ 473 static spa_t search; /* spa_t is large; don't allocate on stack */ 474 spa_t *spa; 475 avl_index_t where; 476 char *cp; 477 478 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 479 480 (void) strlcpy(search.spa_name, name, sizeof (search.spa_name)); 481 482 /* 483 * If it's a full dataset name, figure out the pool name and 484 * just use that. 485 */ 486 cp = strpbrk(search.spa_name, "/@#"); 487 if (cp != NULL) 488 *cp = '\0'; 489 490 spa = avl_find(&spa_namespace_avl, &search, &where); 491 492 return (spa); 493} 494 495/* 496 * Fires when spa_sync has not completed within zfs_deadman_synctime_ms. 497 * If the zfs_deadman_enabled flag is set then it inspects all vdev queues 498 * looking for potentially hung I/Os. 499 */ 500void 501spa_deadman(void *arg) 502{ 503 spa_t *spa = arg; 504 505 /* 506 * Disable the deadman timer if the pool is suspended. 507 */ 508 if (spa_suspended(spa)) { 509#ifdef illumos 510 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY)); 511#else 512 /* Nothing. just don't schedule any future callouts. */ 513#endif 514 return; 515 } 516 517 zfs_dbgmsg("slow spa_sync: started %llu seconds ago, calls %llu", 518 (gethrtime() - spa->spa_sync_starttime) / NANOSEC, 519 ++spa->spa_deadman_calls); 520 if (zfs_deadman_enabled) 521 vdev_deadman(spa->spa_root_vdev); 522} 523 524/* 525 * Create an uninitialized spa_t with the given name. Requires 526 * spa_namespace_lock. The caller must ensure that the spa_t doesn't already 527 * exist by calling spa_lookup() first. 528 */ 529spa_t * 530spa_add(const char *name, nvlist_t *config, const char *altroot) 531{ 532 spa_t *spa; 533 spa_config_dirent_t *dp; 534#ifdef illumos 535 cyc_handler_t hdlr; 536 cyc_time_t when; 537#endif 538 539 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 540 541 spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP); 542 543 mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL); 544 mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL); 545 mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL); 546 mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL); 547 mutex_init(&spa->spa_proc_lock, NULL, MUTEX_DEFAULT, NULL); 548 mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL); 549 mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL); 550 mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL); 551 mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL); 552 553 cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL); 554 cv_init(&spa->spa_proc_cv, NULL, CV_DEFAULT, NULL); 555 cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL); 556 cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL); 557 558 for (int t = 0; t < TXG_SIZE; t++) 559 bplist_create(&spa->spa_free_bplist[t]); 560 561 (void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name)); 562 spa->spa_state = POOL_STATE_UNINITIALIZED; 563 spa->spa_freeze_txg = UINT64_MAX; 564 spa->spa_final_txg = UINT64_MAX; 565 spa->spa_load_max_txg = UINT64_MAX; 566 spa->spa_proc = &p0; 567 spa->spa_proc_state = SPA_PROC_NONE; 568 569#ifdef illumos 570 hdlr.cyh_func = spa_deadman; 571 hdlr.cyh_arg = spa; 572 hdlr.cyh_level = CY_LOW_LEVEL; 573#endif 574 575 spa->spa_deadman_synctime = MSEC2NSEC(zfs_deadman_synctime_ms); 576 577#ifdef illumos 578 /* 579 * This determines how often we need to check for hung I/Os after 580 * the cyclic has already fired. Since checking for hung I/Os is 581 * an expensive operation we don't want to check too frequently. 582 * Instead wait for 5 seconds before checking again. 583 */ 584 when.cyt_interval = MSEC2NSEC(zfs_deadman_checktime_ms); 585 when.cyt_when = CY_INFINITY; 586 mutex_enter(&cpu_lock); 587 spa->spa_deadman_cycid = cyclic_add(&hdlr, &when); 588 mutex_exit(&cpu_lock); 589#else /* !illumos */ 590#ifdef _KERNEL 591 callout_init(&spa->spa_deadman_cycid, CALLOUT_MPSAFE); 592#endif 593#endif 594 refcount_create(&spa->spa_refcount); 595 spa_config_lock_init(spa); 596 597 avl_add(&spa_namespace_avl, spa); 598 599 /* 600 * Set the alternate root, if there is one. 601 */ 602 if (altroot) { 603 spa->spa_root = spa_strdup(altroot); 604 spa_active_count++; 605 } 606 607 /* 608 * Every pool starts with the default cachefile 609 */ 610 list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t), 611 offsetof(spa_config_dirent_t, scd_link)); 612 613 dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP); 614 dp->scd_path = altroot ? NULL : spa_strdup(spa_config_path); 615 list_insert_head(&spa->spa_config_list, dp); 616 617 VERIFY(nvlist_alloc(&spa->spa_load_info, NV_UNIQUE_NAME, 618 KM_SLEEP) == 0); 619 620 if (config != NULL) { 621 nvlist_t *features; 622 623 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ, 624 &features) == 0) { 625 VERIFY(nvlist_dup(features, &spa->spa_label_features, 626 0) == 0); 627 } 628 629 VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0); 630 } 631 632 if (spa->spa_label_features == NULL) { 633 VERIFY(nvlist_alloc(&spa->spa_label_features, NV_UNIQUE_NAME, 634 KM_SLEEP) == 0); 635 } 636 637 spa->spa_debug = ((zfs_flags & ZFS_DEBUG_SPA) != 0); 638 639 /* 640 * As a pool is being created, treat all features as disabled by 641 * setting SPA_FEATURE_DISABLED for all entries in the feature 642 * refcount cache. 643 */ 644 for (int i = 0; i < SPA_FEATURES; i++) { 645 spa->spa_feat_refcount_cache[i] = SPA_FEATURE_DISABLED; 646 } 647 648 return (spa); 649} 650 651/* 652 * Removes a spa_t from the namespace, freeing up any memory used. Requires 653 * spa_namespace_lock. This is called only after the spa_t has been closed and 654 * deactivated. 655 */ 656void 657spa_remove(spa_t *spa) 658{ 659 spa_config_dirent_t *dp; 660 661 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 662 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); 663 664 nvlist_free(spa->spa_config_splitting); 665 666 avl_remove(&spa_namespace_avl, spa); 667 cv_broadcast(&spa_namespace_cv); 668 669 if (spa->spa_root) { 670 spa_strfree(spa->spa_root); 671 spa_active_count--; 672 } 673 674 while ((dp = list_head(&spa->spa_config_list)) != NULL) { 675 list_remove(&spa->spa_config_list, dp); 676 if (dp->scd_path != NULL) 677 spa_strfree(dp->scd_path); 678 kmem_free(dp, sizeof (spa_config_dirent_t)); 679 } 680 681 list_destroy(&spa->spa_config_list); 682 683 nvlist_free(spa->spa_label_features); 684 nvlist_free(spa->spa_load_info); 685 spa_config_set(spa, NULL); 686 687#ifdef illumos 688 mutex_enter(&cpu_lock); 689 if (spa->spa_deadman_cycid != CYCLIC_NONE) 690 cyclic_remove(spa->spa_deadman_cycid); 691 mutex_exit(&cpu_lock); 692 spa->spa_deadman_cycid = CYCLIC_NONE; 693#else /* !illumos */ 694#ifdef _KERNEL 695 callout_drain(&spa->spa_deadman_cycid); 696#endif 697#endif 698 699 refcount_destroy(&spa->spa_refcount); 700 701 spa_config_lock_destroy(spa); 702 703 for (int t = 0; t < TXG_SIZE; t++) 704 bplist_destroy(&spa->spa_free_bplist[t]); 705 706 cv_destroy(&spa->spa_async_cv); 707 cv_destroy(&spa->spa_proc_cv); 708 cv_destroy(&spa->spa_scrub_io_cv); 709 cv_destroy(&spa->spa_suspend_cv); 710 711 mutex_destroy(&spa->spa_async_lock); 712 mutex_destroy(&spa->spa_errlist_lock); 713 mutex_destroy(&spa->spa_errlog_lock); 714 mutex_destroy(&spa->spa_history_lock); 715 mutex_destroy(&spa->spa_proc_lock); 716 mutex_destroy(&spa->spa_props_lock); 717 mutex_destroy(&spa->spa_scrub_lock); 718 mutex_destroy(&spa->spa_suspend_lock); 719 mutex_destroy(&spa->spa_vdev_top_lock); 720 721 kmem_free(spa, sizeof (spa_t)); 722} 723 724/* 725 * Given a pool, return the next pool in the namespace, or NULL if there is 726 * none. If 'prev' is NULL, return the first pool. 727 */ 728spa_t * 729spa_next(spa_t *prev) 730{ 731 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 732 733 if (prev) 734 return (AVL_NEXT(&spa_namespace_avl, prev)); 735 else 736 return (avl_first(&spa_namespace_avl)); 737} 738 739/* 740 * ========================================================================== 741 * SPA refcount functions 742 * ========================================================================== 743 */ 744 745/* 746 * Add a reference to the given spa_t. Must have at least one reference, or 747 * have the namespace lock held. 748 */ 749void 750spa_open_ref(spa_t *spa, void *tag) 751{ 752 ASSERT(refcount_count(&spa->spa_refcount) >= spa->spa_minref || 753 MUTEX_HELD(&spa_namespace_lock)); 754 (void) refcount_add(&spa->spa_refcount, tag); 755} 756 757/* 758 * Remove a reference to the given spa_t. Must have at least one reference, or 759 * have the namespace lock held. 760 */ 761void 762spa_close(spa_t *spa, void *tag) 763{ 764 ASSERT(refcount_count(&spa->spa_refcount) > spa->spa_minref || 765 MUTEX_HELD(&spa_namespace_lock)); 766 (void) refcount_remove(&spa->spa_refcount, tag); 767} 768 769/* 770 * Check to see if the spa refcount is zero. Must be called with 771 * spa_namespace_lock held. We really compare against spa_minref, which is the 772 * number of references acquired when opening a pool 773 */ 774boolean_t 775spa_refcount_zero(spa_t *spa) 776{ 777 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 778 779 return (refcount_count(&spa->spa_refcount) == spa->spa_minref); 780} 781 782/* 783 * ========================================================================== 784 * SPA spare and l2cache tracking 785 * ========================================================================== 786 */ 787 788/* 789 * Hot spares and cache devices are tracked using the same code below, 790 * for 'auxiliary' devices. 791 */ 792 793typedef struct spa_aux { 794 uint64_t aux_guid; 795 uint64_t aux_pool; 796 avl_node_t aux_avl; 797 int aux_count; 798} spa_aux_t; 799 800static int 801spa_aux_compare(const void *a, const void *b) 802{ 803 const spa_aux_t *sa = a; 804 const spa_aux_t *sb = b; 805 806 if (sa->aux_guid < sb->aux_guid) 807 return (-1); 808 else if (sa->aux_guid > sb->aux_guid) 809 return (1); 810 else 811 return (0); 812} 813 814void 815spa_aux_add(vdev_t *vd, avl_tree_t *avl) 816{ 817 avl_index_t where; 818 spa_aux_t search; 819 spa_aux_t *aux; 820 821 search.aux_guid = vd->vdev_guid; 822 if ((aux = avl_find(avl, &search, &where)) != NULL) { 823 aux->aux_count++; 824 } else { 825 aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP); 826 aux->aux_guid = vd->vdev_guid; 827 aux->aux_count = 1; 828 avl_insert(avl, aux, where); 829 } 830} 831 832void 833spa_aux_remove(vdev_t *vd, avl_tree_t *avl) 834{ 835 spa_aux_t search; 836 spa_aux_t *aux; 837 avl_index_t where; 838 839 search.aux_guid = vd->vdev_guid; 840 aux = avl_find(avl, &search, &where); 841 842 ASSERT(aux != NULL); 843 844 if (--aux->aux_count == 0) { 845 avl_remove(avl, aux); 846 kmem_free(aux, sizeof (spa_aux_t)); 847 } else if (aux->aux_pool == spa_guid(vd->vdev_spa)) { 848 aux->aux_pool = 0ULL; 849 } 850} 851 852boolean_t 853spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl) 854{ 855 spa_aux_t search, *found; 856 857 search.aux_guid = guid; 858 found = avl_find(avl, &search, NULL); 859 860 if (pool) { 861 if (found) 862 *pool = found->aux_pool; 863 else 864 *pool = 0ULL; 865 } 866 867 if (refcnt) { 868 if (found) 869 *refcnt = found->aux_count; 870 else 871 *refcnt = 0; 872 } 873 874 return (found != NULL); 875} 876 877void 878spa_aux_activate(vdev_t *vd, avl_tree_t *avl) 879{ 880 spa_aux_t search, *found; 881 avl_index_t where; 882 883 search.aux_guid = vd->vdev_guid; 884 found = avl_find(avl, &search, &where); 885 ASSERT(found != NULL); 886 ASSERT(found->aux_pool == 0ULL); 887 888 found->aux_pool = spa_guid(vd->vdev_spa); 889} 890 891/* 892 * Spares are tracked globally due to the following constraints: 893 * 894 * - A spare may be part of multiple pools. 895 * - A spare may be added to a pool even if it's actively in use within 896 * another pool. 897 * - A spare in use in any pool can only be the source of a replacement if 898 * the target is a spare in the same pool. 899 * 900 * We keep track of all spares on the system through the use of a reference 901 * counted AVL tree. When a vdev is added as a spare, or used as a replacement 902 * spare, then we bump the reference count in the AVL tree. In addition, we set 903 * the 'vdev_isspare' member to indicate that the device is a spare (active or 904 * inactive). When a spare is made active (used to replace a device in the 905 * pool), we also keep track of which pool its been made a part of. 906 * 907 * The 'spa_spare_lock' protects the AVL tree. These functions are normally 908 * called under the spa_namespace lock as part of vdev reconfiguration. The 909 * separate spare lock exists for the status query path, which does not need to 910 * be completely consistent with respect to other vdev configuration changes. 911 */ 912 913static int 914spa_spare_compare(const void *a, const void *b) 915{ 916 return (spa_aux_compare(a, b)); 917} 918 919void 920spa_spare_add(vdev_t *vd) 921{ 922 mutex_enter(&spa_spare_lock); 923 ASSERT(!vd->vdev_isspare); 924 spa_aux_add(vd, &spa_spare_avl); 925 vd->vdev_isspare = B_TRUE; 926 mutex_exit(&spa_spare_lock); 927} 928 929void 930spa_spare_remove(vdev_t *vd) 931{ 932 mutex_enter(&spa_spare_lock); 933 ASSERT(vd->vdev_isspare); 934 spa_aux_remove(vd, &spa_spare_avl); 935 vd->vdev_isspare = B_FALSE; 936 mutex_exit(&spa_spare_lock); 937} 938 939boolean_t 940spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt) 941{ 942 boolean_t found; 943 944 mutex_enter(&spa_spare_lock); 945 found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl); 946 mutex_exit(&spa_spare_lock); 947 948 return (found); 949} 950 951void 952spa_spare_activate(vdev_t *vd) 953{ 954 mutex_enter(&spa_spare_lock); 955 ASSERT(vd->vdev_isspare); 956 spa_aux_activate(vd, &spa_spare_avl); 957 mutex_exit(&spa_spare_lock); 958} 959 960/* 961 * Level 2 ARC devices are tracked globally for the same reasons as spares. 962 * Cache devices currently only support one pool per cache device, and so 963 * for these devices the aux reference count is currently unused beyond 1. 964 */ 965 966static int 967spa_l2cache_compare(const void *a, const void *b) 968{ 969 return (spa_aux_compare(a, b)); 970} 971 972void 973spa_l2cache_add(vdev_t *vd) 974{ 975 mutex_enter(&spa_l2cache_lock); 976 ASSERT(!vd->vdev_isl2cache); 977 spa_aux_add(vd, &spa_l2cache_avl); 978 vd->vdev_isl2cache = B_TRUE; 979 mutex_exit(&spa_l2cache_lock); 980} 981 982void 983spa_l2cache_remove(vdev_t *vd) 984{ 985 mutex_enter(&spa_l2cache_lock); 986 ASSERT(vd->vdev_isl2cache); 987 spa_aux_remove(vd, &spa_l2cache_avl); 988 vd->vdev_isl2cache = B_FALSE; 989 mutex_exit(&spa_l2cache_lock); 990} 991 992boolean_t 993spa_l2cache_exists(uint64_t guid, uint64_t *pool) 994{ 995 boolean_t found; 996 997 mutex_enter(&spa_l2cache_lock); 998 found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl); 999 mutex_exit(&spa_l2cache_lock); 1000 1001 return (found); 1002} 1003 1004void 1005spa_l2cache_activate(vdev_t *vd) 1006{ 1007 mutex_enter(&spa_l2cache_lock); 1008 ASSERT(vd->vdev_isl2cache); 1009 spa_aux_activate(vd, &spa_l2cache_avl); 1010 mutex_exit(&spa_l2cache_lock); 1011} 1012 1013/* 1014 * ========================================================================== 1015 * SPA vdev locking 1016 * ========================================================================== 1017 */ 1018 1019/* 1020 * Lock the given spa_t for the purpose of adding or removing a vdev. 1021 * Grabs the global spa_namespace_lock plus the spa config lock for writing. 1022 * It returns the next transaction group for the spa_t. 1023 */ 1024uint64_t 1025spa_vdev_enter(spa_t *spa) 1026{ 1027 mutex_enter(&spa->spa_vdev_top_lock); 1028 mutex_enter(&spa_namespace_lock); 1029 return (spa_vdev_config_enter(spa)); 1030} 1031 1032/* 1033 * Internal implementation for spa_vdev_enter(). Used when a vdev 1034 * operation requires multiple syncs (i.e. removing a device) while 1035 * keeping the spa_namespace_lock held. 1036 */ 1037uint64_t 1038spa_vdev_config_enter(spa_t *spa) 1039{ 1040 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1041 1042 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER); 1043 1044 return (spa_last_synced_txg(spa) + 1); 1045} 1046 1047/* 1048 * Used in combination with spa_vdev_config_enter() to allow the syncing 1049 * of multiple transactions without releasing the spa_namespace_lock. 1050 */ 1051void 1052spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag) 1053{ 1054 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1055 1056 int config_changed = B_FALSE; 1057 1058 ASSERT(txg > spa_last_synced_txg(spa)); 1059 1060 spa->spa_pending_vdev = NULL; 1061 1062 /* 1063 * Reassess the DTLs. 1064 */ 1065 vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE); 1066 1067 if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) { 1068 config_changed = B_TRUE; 1069 spa->spa_config_generation++; 1070 } 1071 1072 /* 1073 * Verify the metaslab classes. 1074 */ 1075 ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0); 1076 ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0); 1077 1078 spa_config_exit(spa, SCL_ALL, spa); 1079 1080 /* 1081 * Panic the system if the specified tag requires it. This 1082 * is useful for ensuring that configurations are updated 1083 * transactionally. 1084 */ 1085 if (zio_injection_enabled) 1086 zio_handle_panic_injection(spa, tag, 0); 1087 1088 /* 1089 * Note: this txg_wait_synced() is important because it ensures 1090 * that there won't be more than one config change per txg. 1091 * This allows us to use the txg as the generation number. 1092 */ 1093 if (error == 0) 1094 txg_wait_synced(spa->spa_dsl_pool, txg); 1095 1096 if (vd != NULL) { 1097 ASSERT(!vd->vdev_detached || vd->vdev_dtl_sm == NULL); 1098 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER); 1099 vdev_free(vd); 1100 spa_config_exit(spa, SCL_ALL, spa); 1101 } 1102 1103 /* 1104 * If the config changed, update the config cache. 1105 */ 1106 if (config_changed) 1107 spa_config_sync(spa, B_FALSE, B_TRUE); 1108} 1109 1110/* 1111 * Unlock the spa_t after adding or removing a vdev. Besides undoing the 1112 * locking of spa_vdev_enter(), we also want make sure the transactions have 1113 * synced to disk, and then update the global configuration cache with the new 1114 * information. 1115 */ 1116int 1117spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error) 1118{ 1119 spa_vdev_config_exit(spa, vd, txg, error, FTAG); 1120 mutex_exit(&spa_namespace_lock); 1121 mutex_exit(&spa->spa_vdev_top_lock); 1122 1123 return (error); 1124} 1125 1126/* 1127 * Lock the given spa_t for the purpose of changing vdev state. 1128 */ 1129void 1130spa_vdev_state_enter(spa_t *spa, int oplocks) 1131{ 1132 int locks = SCL_STATE_ALL | oplocks; 1133 1134 /* 1135 * Root pools may need to read of the underlying devfs filesystem 1136 * when opening up a vdev. Unfortunately if we're holding the 1137 * SCL_ZIO lock it will result in a deadlock when we try to issue 1138 * the read from the root filesystem. Instead we "prefetch" 1139 * the associated vnodes that we need prior to opening the 1140 * underlying devices and cache them so that we can prevent 1141 * any I/O when we are doing the actual open. 1142 */ 1143 if (spa_is_root(spa)) { 1144 int low = locks & ~(SCL_ZIO - 1); 1145 int high = locks & ~low; 1146 1147 spa_config_enter(spa, high, spa, RW_WRITER); 1148 vdev_hold(spa->spa_root_vdev); 1149 spa_config_enter(spa, low, spa, RW_WRITER); 1150 } else { 1151 spa_config_enter(spa, locks, spa, RW_WRITER); 1152 } 1153 spa->spa_vdev_locks = locks; 1154} 1155 1156int 1157spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error) 1158{ 1159 boolean_t config_changed = B_FALSE; 1160 1161 if (vd != NULL || error == 0) 1162 vdev_dtl_reassess(vd ? vd->vdev_top : spa->spa_root_vdev, 1163 0, 0, B_FALSE); 1164 1165 if (vd != NULL) { 1166 vdev_state_dirty(vd->vdev_top); 1167 config_changed = B_TRUE; 1168 spa->spa_config_generation++; 1169 } 1170 1171 if (spa_is_root(spa)) 1172 vdev_rele(spa->spa_root_vdev); 1173 1174 ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL); 1175 spa_config_exit(spa, spa->spa_vdev_locks, spa); 1176 1177 /* 1178 * If anything changed, wait for it to sync. This ensures that, 1179 * from the system administrator's perspective, zpool(1M) commands 1180 * are synchronous. This is important for things like zpool offline: 1181 * when the command completes, you expect no further I/O from ZFS. 1182 */ 1183 if (vd != NULL) 1184 txg_wait_synced(spa->spa_dsl_pool, 0); 1185 1186 /* 1187 * If the config changed, update the config cache. 1188 */ 1189 if (config_changed) { 1190 mutex_enter(&spa_namespace_lock); 1191 spa_config_sync(spa, B_FALSE, B_TRUE); 1192 mutex_exit(&spa_namespace_lock); 1193 } 1194 1195 return (error); 1196} 1197 1198/* 1199 * ========================================================================== 1200 * Miscellaneous functions 1201 * ========================================================================== 1202 */ 1203 1204void 1205spa_activate_mos_feature(spa_t *spa, const char *feature, dmu_tx_t *tx) 1206{ 1207 if (!nvlist_exists(spa->spa_label_features, feature)) { 1208 fnvlist_add_boolean(spa->spa_label_features, feature); 1209 /* 1210 * When we are creating the pool (tx_txg==TXG_INITIAL), we can't 1211 * dirty the vdev config because lock SCL_CONFIG is not held. 1212 * Thankfully, in this case we don't need to dirty the config 1213 * because it will be written out anyway when we finish 1214 * creating the pool. 1215 */ 1216 if (tx->tx_txg != TXG_INITIAL) 1217 vdev_config_dirty(spa->spa_root_vdev); 1218 } 1219} 1220 1221void 1222spa_deactivate_mos_feature(spa_t *spa, const char *feature) 1223{ 1224 if (nvlist_remove_all(spa->spa_label_features, feature) == 0) 1225 vdev_config_dirty(spa->spa_root_vdev); 1226} 1227 1228/* 1229 * Rename a spa_t. 1230 */ 1231int 1232spa_rename(const char *name, const char *newname) 1233{ 1234 spa_t *spa; 1235 int err; 1236 1237 /* 1238 * Lookup the spa_t and grab the config lock for writing. We need to 1239 * actually open the pool so that we can sync out the necessary labels. 1240 * It's OK to call spa_open() with the namespace lock held because we 1241 * allow recursive calls for other reasons. 1242 */ 1243 mutex_enter(&spa_namespace_lock); 1244 if ((err = spa_open(name, &spa, FTAG)) != 0) { 1245 mutex_exit(&spa_namespace_lock); 1246 return (err); 1247 } 1248 1249 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1250 1251 avl_remove(&spa_namespace_avl, spa); 1252 (void) strlcpy(spa->spa_name, newname, sizeof (spa->spa_name)); 1253 avl_add(&spa_namespace_avl, spa); 1254 1255 /* 1256 * Sync all labels to disk with the new names by marking the root vdev 1257 * dirty and waiting for it to sync. It will pick up the new pool name 1258 * during the sync. 1259 */ 1260 vdev_config_dirty(spa->spa_root_vdev); 1261 1262 spa_config_exit(spa, SCL_ALL, FTAG); 1263 1264 txg_wait_synced(spa->spa_dsl_pool, 0); 1265 1266 /* 1267 * Sync the updated config cache. 1268 */ 1269 spa_config_sync(spa, B_FALSE, B_TRUE); 1270 1271 spa_close(spa, FTAG); 1272 1273 mutex_exit(&spa_namespace_lock); 1274 1275 return (0); 1276} 1277 1278/* 1279 * Return the spa_t associated with given pool_guid, if it exists. If 1280 * device_guid is non-zero, determine whether the pool exists *and* contains 1281 * a device with the specified device_guid. 1282 */ 1283spa_t * 1284spa_by_guid(uint64_t pool_guid, uint64_t device_guid) 1285{ 1286 spa_t *spa; 1287 avl_tree_t *t = &spa_namespace_avl; 1288 1289 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1290 1291 for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) { 1292 if (spa->spa_state == POOL_STATE_UNINITIALIZED) 1293 continue; 1294 if (spa->spa_root_vdev == NULL) 1295 continue; 1296 if (spa_guid(spa) == pool_guid) { 1297 if (device_guid == 0) 1298 break; 1299 1300 if (vdev_lookup_by_guid(spa->spa_root_vdev, 1301 device_guid) != NULL) 1302 break; 1303 1304 /* 1305 * Check any devices we may be in the process of adding. 1306 */ 1307 if (spa->spa_pending_vdev) { 1308 if (vdev_lookup_by_guid(spa->spa_pending_vdev, 1309 device_guid) != NULL) 1310 break; 1311 } 1312 } 1313 } 1314 1315 return (spa); 1316} 1317 1318/* 1319 * Determine whether a pool with the given pool_guid exists. 1320 */ 1321boolean_t 1322spa_guid_exists(uint64_t pool_guid, uint64_t device_guid) 1323{ 1324 return (spa_by_guid(pool_guid, device_guid) != NULL); 1325} 1326 1327char * 1328spa_strdup(const char *s) 1329{ 1330 size_t len; 1331 char *new; 1332 1333 len = strlen(s); 1334 new = kmem_alloc(len + 1, KM_SLEEP); 1335 bcopy(s, new, len); 1336 new[len] = '\0'; 1337 1338 return (new); 1339} 1340 1341void 1342spa_strfree(char *s) 1343{ 1344 kmem_free(s, strlen(s) + 1); 1345} 1346 1347uint64_t 1348spa_get_random(uint64_t range) 1349{ 1350 uint64_t r; 1351 1352 ASSERT(range != 0); 1353 1354 (void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t)); 1355 1356 return (r % range); 1357} 1358 1359uint64_t 1360spa_generate_guid(spa_t *spa) 1361{ 1362 uint64_t guid = spa_get_random(-1ULL); 1363 1364 if (spa != NULL) { 1365 while (guid == 0 || spa_guid_exists(spa_guid(spa), guid)) 1366 guid = spa_get_random(-1ULL); 1367 } else { 1368 while (guid == 0 || spa_guid_exists(guid, 0)) 1369 guid = spa_get_random(-1ULL); 1370 } 1371 1372 return (guid); 1373} 1374 1375void 1376snprintf_blkptr(char *buf, size_t buflen, const blkptr_t *bp) 1377{ 1378 char type[256]; 1379 char *checksum = NULL; 1380 char *compress = NULL; 1381 1382 if (bp != NULL) { 1383 if (BP_GET_TYPE(bp) & DMU_OT_NEWTYPE) { 1384 dmu_object_byteswap_t bswap = 1385 DMU_OT_BYTESWAP(BP_GET_TYPE(bp)); 1386 (void) snprintf(type, sizeof (type), "bswap %s %s", 1387 DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) ? 1388 "metadata" : "data", 1389 dmu_ot_byteswap[bswap].ob_name); 1390 } else { 1391 (void) strlcpy(type, dmu_ot[BP_GET_TYPE(bp)].ot_name, 1392 sizeof (type)); 1393 } 1394 checksum = zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name; 1395 compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name; 1396 } 1397 1398 SNPRINTF_BLKPTR(snprintf, ' ', buf, buflen, bp, type, checksum, 1399 compress); 1400} 1401 1402void 1403spa_freeze(spa_t *spa) 1404{ 1405 uint64_t freeze_txg = 0; 1406 1407 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1408 if (spa->spa_freeze_txg == UINT64_MAX) { 1409 freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE; 1410 spa->spa_freeze_txg = freeze_txg; 1411 } 1412 spa_config_exit(spa, SCL_ALL, FTAG); 1413 if (freeze_txg != 0) 1414 txg_wait_synced(spa_get_dsl(spa), freeze_txg); 1415} 1416 1417void 1418zfs_panic_recover(const char *fmt, ...) 1419{ 1420 va_list adx; 1421 1422 va_start(adx, fmt); 1423 vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx); 1424 va_end(adx); 1425} 1426 1427/* 1428 * This is a stripped-down version of strtoull, suitable only for converting 1429 * lowercase hexadecimal numbers that don't overflow. 1430 */ 1431uint64_t 1432zfs_strtonum(const char *str, char **nptr) 1433{ 1434 uint64_t val = 0; 1435 char c; 1436 int digit; 1437 1438 while ((c = *str) != '\0') { 1439 if (c >= '0' && c <= '9') 1440 digit = c - '0'; 1441 else if (c >= 'a' && c <= 'f') 1442 digit = 10 + c - 'a'; 1443 else 1444 break; 1445 1446 val *= 16; 1447 val += digit; 1448 1449 str++; 1450 } 1451 1452 if (nptr) 1453 *nptr = (char *)str; 1454 1455 return (val); 1456} 1457 1458/* 1459 * ========================================================================== 1460 * Accessor functions 1461 * ========================================================================== 1462 */ 1463 1464boolean_t 1465spa_shutting_down(spa_t *spa) 1466{ 1467 return (spa->spa_async_suspended); 1468} 1469 1470dsl_pool_t * 1471spa_get_dsl(spa_t *spa) 1472{ 1473 return (spa->spa_dsl_pool); 1474} 1475 1476boolean_t 1477spa_is_initializing(spa_t *spa) 1478{ 1479 return (spa->spa_is_initializing); 1480} 1481 1482blkptr_t * 1483spa_get_rootblkptr(spa_t *spa) 1484{ 1485 return (&spa->spa_ubsync.ub_rootbp); 1486} 1487 1488void 1489spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp) 1490{ 1491 spa->spa_uberblock.ub_rootbp = *bp; 1492} 1493 1494void 1495spa_altroot(spa_t *spa, char *buf, size_t buflen) 1496{ 1497 if (spa->spa_root == NULL) 1498 buf[0] = '\0'; 1499 else 1500 (void) strncpy(buf, spa->spa_root, buflen); 1501} 1502 1503int 1504spa_sync_pass(spa_t *spa) 1505{ 1506 return (spa->spa_sync_pass); 1507} 1508 1509char * 1510spa_name(spa_t *spa) 1511{ 1512 return (spa->spa_name); 1513} 1514 1515uint64_t 1516spa_guid(spa_t *spa) 1517{ 1518 dsl_pool_t *dp = spa_get_dsl(spa); 1519 uint64_t guid; 1520 1521 /* 1522 * If we fail to parse the config during spa_load(), we can go through 1523 * the error path (which posts an ereport) and end up here with no root 1524 * vdev. We stash the original pool guid in 'spa_config_guid' to handle 1525 * this case. 1526 */ 1527 if (spa->spa_root_vdev == NULL) 1528 return (spa->spa_config_guid); 1529 1530 guid = spa->spa_last_synced_guid != 0 ? 1531 spa->spa_last_synced_guid : spa->spa_root_vdev->vdev_guid; 1532 1533 /* 1534 * Return the most recently synced out guid unless we're 1535 * in syncing context. 1536 */ 1537 if (dp && dsl_pool_sync_context(dp)) 1538 return (spa->spa_root_vdev->vdev_guid); 1539 else 1540 return (guid); 1541} 1542 1543uint64_t 1544spa_load_guid(spa_t *spa) 1545{ 1546 /* 1547 * This is a GUID that exists solely as a reference for the 1548 * purposes of the arc. It is generated at load time, and 1549 * is never written to persistent storage. 1550 */ 1551 return (spa->spa_load_guid); 1552} 1553 1554uint64_t 1555spa_last_synced_txg(spa_t *spa) 1556{ 1557 return (spa->spa_ubsync.ub_txg); 1558} 1559 1560uint64_t 1561spa_first_txg(spa_t *spa) 1562{ 1563 return (spa->spa_first_txg); 1564} 1565 1566uint64_t 1567spa_syncing_txg(spa_t *spa) 1568{ 1569 return (spa->spa_syncing_txg); 1570} 1571 1572pool_state_t 1573spa_state(spa_t *spa) 1574{ 1575 return (spa->spa_state); 1576} 1577 1578spa_load_state_t 1579spa_load_state(spa_t *spa) 1580{ 1581 return (spa->spa_load_state); 1582} 1583 1584uint64_t 1585spa_freeze_txg(spa_t *spa) 1586{ 1587 return (spa->spa_freeze_txg); 1588} 1589 1590/* ARGSUSED */ 1591uint64_t 1592spa_get_asize(spa_t *spa, uint64_t lsize) 1593{ 1594 return (lsize * spa_asize_inflation); 1595} 1596 1597uint64_t 1598spa_get_dspace(spa_t *spa) 1599{ 1600 return (spa->spa_dspace); 1601} 1602 1603void 1604spa_update_dspace(spa_t *spa) 1605{ 1606 spa->spa_dspace = metaslab_class_get_dspace(spa_normal_class(spa)) + 1607 ddt_get_dedup_dspace(spa); 1608} 1609 1610/* 1611 * Return the failure mode that has been set to this pool. The default 1612 * behavior will be to block all I/Os when a complete failure occurs. 1613 */ 1614uint8_t 1615spa_get_failmode(spa_t *spa) 1616{ 1617 return (spa->spa_failmode); 1618} 1619 1620boolean_t 1621spa_suspended(spa_t *spa) 1622{ 1623 return (spa->spa_suspended); 1624} 1625 1626uint64_t 1627spa_version(spa_t *spa) 1628{ 1629 return (spa->spa_ubsync.ub_version); 1630} 1631 1632boolean_t 1633spa_deflate(spa_t *spa) 1634{ 1635 return (spa->spa_deflate); 1636} 1637 1638metaslab_class_t * 1639spa_normal_class(spa_t *spa) 1640{ 1641 return (spa->spa_normal_class); 1642} 1643 1644metaslab_class_t * 1645spa_log_class(spa_t *spa) 1646{ 1647 return (spa->spa_log_class); 1648} 1649 1650int 1651spa_max_replication(spa_t *spa) 1652{ 1653 /* 1654 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to 1655 * handle BPs with more than one DVA allocated. Set our max 1656 * replication level accordingly. 1657 */ 1658 if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS) 1659 return (1); 1660 return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override)); 1661} 1662 1663int 1664spa_prev_software_version(spa_t *spa) 1665{ 1666 return (spa->spa_prev_software_version); 1667} 1668 1669uint64_t 1670spa_deadman_synctime(spa_t *spa) 1671{ 1672 return (spa->spa_deadman_synctime); 1673} 1674 1675uint64_t 1676dva_get_dsize_sync(spa_t *spa, const dva_t *dva) 1677{ 1678 uint64_t asize = DVA_GET_ASIZE(dva); 1679 uint64_t dsize = asize; 1680 1681 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 1682 1683 if (asize != 0 && spa->spa_deflate) { 1684 vdev_t *vd = vdev_lookup_top(spa, DVA_GET_VDEV(dva)); 1685 dsize = (asize >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio; 1686 } 1687 1688 return (dsize); 1689} 1690 1691uint64_t 1692bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp) 1693{ 1694 uint64_t dsize = 0; 1695 1696 for (int d = 0; d < SPA_DVAS_PER_BP; d++) 1697 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]); 1698 1699 return (dsize); 1700} 1701 1702uint64_t 1703bp_get_dsize(spa_t *spa, const blkptr_t *bp) 1704{ 1705 uint64_t dsize = 0; 1706 1707 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); 1708 1709 for (int d = 0; d < SPA_DVAS_PER_BP; d++) 1710 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]); 1711 1712 spa_config_exit(spa, SCL_VDEV, FTAG); 1713 1714 return (dsize); 1715} 1716 1717/* 1718 * ========================================================================== 1719 * Initialization and Termination 1720 * ========================================================================== 1721 */ 1722 1723static int 1724spa_name_compare(const void *a1, const void *a2) 1725{ 1726 const spa_t *s1 = a1; 1727 const spa_t *s2 = a2; 1728 int s; 1729 1730 s = strcmp(s1->spa_name, s2->spa_name); 1731 if (s > 0) 1732 return (1); 1733 if (s < 0) 1734 return (-1); 1735 return (0); 1736} 1737 1738int 1739spa_busy(void) 1740{ 1741 return (spa_active_count); 1742} 1743 1744void 1745spa_boot_init() 1746{ 1747 spa_config_load(); 1748} 1749 1750#ifdef _KERNEL 1751EVENTHANDLER_DEFINE(mountroot, spa_boot_init, NULL, 0); 1752#endif 1753 1754void 1755spa_init(int mode) 1756{ 1757 mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL); 1758 mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL); 1759 mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL); 1760 cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL); 1761 1762 avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t), 1763 offsetof(spa_t, spa_avl)); 1764 1765 avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t), 1766 offsetof(spa_aux_t, aux_avl)); 1767 1768 avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t), 1769 offsetof(spa_aux_t, aux_avl)); 1770 1771 spa_mode_global = mode; 1772 1773#ifdef illumos 1774#ifdef _KERNEL 1775 spa_arch_init(); 1776#else 1777 if (spa_mode_global != FREAD && dprintf_find_string("watch")) { 1778 arc_procfd = open("/proc/self/ctl", O_WRONLY); 1779 if (arc_procfd == -1) { 1780 perror("could not enable watchpoints: " 1781 "opening /proc/self/ctl failed: "); 1782 } else { 1783 arc_watch = B_TRUE; 1784 } 1785 } 1786#endif 1787#endif /* illumos */ 1788 refcount_sysinit(); 1789 unique_init(); 1790 range_tree_init(); 1791 zio_init(); 1792 lz4_init(); 1793 dmu_init(); 1794 zil_init(); 1795 vdev_cache_stat_init(); 1796 zfs_prop_init(); 1797 zpool_prop_init(); 1798 zpool_feature_init(); 1799 spa_config_load(); 1800 l2arc_start(); 1801#ifndef illumos 1802#ifdef _KERNEL 1803 zfs_deadman_init(); 1804#endif 1805#endif /* !illumos */ 1806} 1807 1808void 1809spa_fini(void) 1810{ 1811 l2arc_stop(); 1812 1813 spa_evict_all(); 1814 1815 vdev_cache_stat_fini(); 1816 zil_fini(); 1817 dmu_fini(); 1818 lz4_fini(); 1819 zio_fini(); 1820 range_tree_fini(); 1821 unique_fini(); 1822 refcount_fini(); 1823 1824 avl_destroy(&spa_namespace_avl); 1825 avl_destroy(&spa_spare_avl); 1826 avl_destroy(&spa_l2cache_avl); 1827 1828 cv_destroy(&spa_namespace_cv); 1829 mutex_destroy(&spa_namespace_lock); 1830 mutex_destroy(&spa_spare_lock); 1831 mutex_destroy(&spa_l2cache_lock); 1832} 1833 1834/* 1835 * Return whether this pool has slogs. No locking needed. 1836 * It's not a problem if the wrong answer is returned as it's only for 1837 * performance and not correctness 1838 */ 1839boolean_t 1840spa_has_slogs(spa_t *spa) 1841{ 1842 return (spa->spa_log_class->mc_rotor != NULL); 1843} 1844 1845spa_log_state_t 1846spa_get_log_state(spa_t *spa) 1847{ 1848 return (spa->spa_log_state); 1849} 1850 1851void 1852spa_set_log_state(spa_t *spa, spa_log_state_t state) 1853{ 1854 spa->spa_log_state = state; 1855} 1856 1857boolean_t 1858spa_is_root(spa_t *spa) 1859{ 1860 return (spa->spa_is_root); 1861} 1862 1863boolean_t 1864spa_writeable(spa_t *spa) 1865{ 1866 return (!!(spa->spa_mode & FWRITE)); 1867} 1868 1869int 1870spa_mode(spa_t *spa) 1871{ 1872 return (spa->spa_mode); 1873} 1874 1875uint64_t 1876spa_bootfs(spa_t *spa) 1877{ 1878 return (spa->spa_bootfs); 1879} 1880 1881uint64_t 1882spa_delegation(spa_t *spa) 1883{ 1884 return (spa->spa_delegation); 1885} 1886 1887objset_t * 1888spa_meta_objset(spa_t *spa) 1889{ 1890 return (spa->spa_meta_objset); 1891} 1892 1893enum zio_checksum 1894spa_dedup_checksum(spa_t *spa) 1895{ 1896 return (spa->spa_dedup_checksum); 1897} 1898 1899/* 1900 * Reset pool scan stat per scan pass (or reboot). 1901 */ 1902void 1903spa_scan_stat_init(spa_t *spa) 1904{ 1905 /* data not stored on disk */ 1906 spa->spa_scan_pass_start = gethrestime_sec(); 1907 spa->spa_scan_pass_exam = 0; 1908 vdev_scan_stat_init(spa->spa_root_vdev); 1909} 1910 1911/* 1912 * Get scan stats for zpool status reports 1913 */ 1914int 1915spa_scan_get_stats(spa_t *spa, pool_scan_stat_t *ps) 1916{ 1917 dsl_scan_t *scn = spa->spa_dsl_pool ? spa->spa_dsl_pool->dp_scan : NULL; 1918 1919 if (scn == NULL || scn->scn_phys.scn_func == POOL_SCAN_NONE) 1920 return (SET_ERROR(ENOENT)); 1921 bzero(ps, sizeof (pool_scan_stat_t)); 1922 1923 /* data stored on disk */ 1924 ps->pss_func = scn->scn_phys.scn_func; 1925 ps->pss_start_time = scn->scn_phys.scn_start_time; 1926 ps->pss_end_time = scn->scn_phys.scn_end_time; 1927 ps->pss_to_examine = scn->scn_phys.scn_to_examine; 1928 ps->pss_examined = scn->scn_phys.scn_examined; 1929 ps->pss_to_process = scn->scn_phys.scn_to_process; 1930 ps->pss_processed = scn->scn_phys.scn_processed; 1931 ps->pss_errors = scn->scn_phys.scn_errors; 1932 ps->pss_state = scn->scn_phys.scn_state; 1933 1934 /* data not stored on disk */ 1935 ps->pss_pass_start = spa->spa_scan_pass_start; 1936 ps->pss_pass_exam = spa->spa_scan_pass_exam; 1937 1938 return (0); 1939} 1940 1941boolean_t 1942spa_debug_enabled(spa_t *spa) 1943{ 1944 return (spa->spa_debug); 1945} 1946