spa_misc.c revision 263390
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 return (spa); 640} 641 642/* 643 * Removes a spa_t from the namespace, freeing up any memory used. Requires 644 * spa_namespace_lock. This is called only after the spa_t has been closed and 645 * deactivated. 646 */ 647void 648spa_remove(spa_t *spa) 649{ 650 spa_config_dirent_t *dp; 651 652 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 653 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); 654 655 nvlist_free(spa->spa_config_splitting); 656 657 avl_remove(&spa_namespace_avl, spa); 658 cv_broadcast(&spa_namespace_cv); 659 660 if (spa->spa_root) { 661 spa_strfree(spa->spa_root); 662 spa_active_count--; 663 } 664 665 while ((dp = list_head(&spa->spa_config_list)) != NULL) { 666 list_remove(&spa->spa_config_list, dp); 667 if (dp->scd_path != NULL) 668 spa_strfree(dp->scd_path); 669 kmem_free(dp, sizeof (spa_config_dirent_t)); 670 } 671 672 list_destroy(&spa->spa_config_list); 673 674 nvlist_free(spa->spa_label_features); 675 nvlist_free(spa->spa_load_info); 676 spa_config_set(spa, NULL); 677 678#ifdef illumos 679 mutex_enter(&cpu_lock); 680 if (spa->spa_deadman_cycid != CYCLIC_NONE) 681 cyclic_remove(spa->spa_deadman_cycid); 682 mutex_exit(&cpu_lock); 683 spa->spa_deadman_cycid = CYCLIC_NONE; 684#else /* !illumos */ 685#ifdef _KERNEL 686 callout_drain(&spa->spa_deadman_cycid); 687#endif 688#endif 689 690 refcount_destroy(&spa->spa_refcount); 691 692 spa_config_lock_destroy(spa); 693 694 for (int t = 0; t < TXG_SIZE; t++) 695 bplist_destroy(&spa->spa_free_bplist[t]); 696 697 cv_destroy(&spa->spa_async_cv); 698 cv_destroy(&spa->spa_proc_cv); 699 cv_destroy(&spa->spa_scrub_io_cv); 700 cv_destroy(&spa->spa_suspend_cv); 701 702 mutex_destroy(&spa->spa_async_lock); 703 mutex_destroy(&spa->spa_errlist_lock); 704 mutex_destroy(&spa->spa_errlog_lock); 705 mutex_destroy(&spa->spa_history_lock); 706 mutex_destroy(&spa->spa_proc_lock); 707 mutex_destroy(&spa->spa_props_lock); 708 mutex_destroy(&spa->spa_scrub_lock); 709 mutex_destroy(&spa->spa_suspend_lock); 710 mutex_destroy(&spa->spa_vdev_top_lock); 711 712 kmem_free(spa, sizeof (spa_t)); 713} 714 715/* 716 * Given a pool, return the next pool in the namespace, or NULL if there is 717 * none. If 'prev' is NULL, return the first pool. 718 */ 719spa_t * 720spa_next(spa_t *prev) 721{ 722 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 723 724 if (prev) 725 return (AVL_NEXT(&spa_namespace_avl, prev)); 726 else 727 return (avl_first(&spa_namespace_avl)); 728} 729 730/* 731 * ========================================================================== 732 * SPA refcount functions 733 * ========================================================================== 734 */ 735 736/* 737 * Add a reference to the given spa_t. Must have at least one reference, or 738 * have the namespace lock held. 739 */ 740void 741spa_open_ref(spa_t *spa, void *tag) 742{ 743 ASSERT(refcount_count(&spa->spa_refcount) >= spa->spa_minref || 744 MUTEX_HELD(&spa_namespace_lock)); 745 (void) refcount_add(&spa->spa_refcount, tag); 746} 747 748/* 749 * Remove a reference to the given spa_t. Must have at least one reference, or 750 * have the namespace lock held. 751 */ 752void 753spa_close(spa_t *spa, void *tag) 754{ 755 ASSERT(refcount_count(&spa->spa_refcount) > spa->spa_minref || 756 MUTEX_HELD(&spa_namespace_lock)); 757 (void) refcount_remove(&spa->spa_refcount, tag); 758} 759 760/* 761 * Check to see if the spa refcount is zero. Must be called with 762 * spa_namespace_lock held. We really compare against spa_minref, which is the 763 * number of references acquired when opening a pool 764 */ 765boolean_t 766spa_refcount_zero(spa_t *spa) 767{ 768 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 769 770 return (refcount_count(&spa->spa_refcount) == spa->spa_minref); 771} 772 773/* 774 * ========================================================================== 775 * SPA spare and l2cache tracking 776 * ========================================================================== 777 */ 778 779/* 780 * Hot spares and cache devices are tracked using the same code below, 781 * for 'auxiliary' devices. 782 */ 783 784typedef struct spa_aux { 785 uint64_t aux_guid; 786 uint64_t aux_pool; 787 avl_node_t aux_avl; 788 int aux_count; 789} spa_aux_t; 790 791static int 792spa_aux_compare(const void *a, const void *b) 793{ 794 const spa_aux_t *sa = a; 795 const spa_aux_t *sb = b; 796 797 if (sa->aux_guid < sb->aux_guid) 798 return (-1); 799 else if (sa->aux_guid > sb->aux_guid) 800 return (1); 801 else 802 return (0); 803} 804 805void 806spa_aux_add(vdev_t *vd, avl_tree_t *avl) 807{ 808 avl_index_t where; 809 spa_aux_t search; 810 spa_aux_t *aux; 811 812 search.aux_guid = vd->vdev_guid; 813 if ((aux = avl_find(avl, &search, &where)) != NULL) { 814 aux->aux_count++; 815 } else { 816 aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP); 817 aux->aux_guid = vd->vdev_guid; 818 aux->aux_count = 1; 819 avl_insert(avl, aux, where); 820 } 821} 822 823void 824spa_aux_remove(vdev_t *vd, avl_tree_t *avl) 825{ 826 spa_aux_t search; 827 spa_aux_t *aux; 828 avl_index_t where; 829 830 search.aux_guid = vd->vdev_guid; 831 aux = avl_find(avl, &search, &where); 832 833 ASSERT(aux != NULL); 834 835 if (--aux->aux_count == 0) { 836 avl_remove(avl, aux); 837 kmem_free(aux, sizeof (spa_aux_t)); 838 } else if (aux->aux_pool == spa_guid(vd->vdev_spa)) { 839 aux->aux_pool = 0ULL; 840 } 841} 842 843boolean_t 844spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl) 845{ 846 spa_aux_t search, *found; 847 848 search.aux_guid = guid; 849 found = avl_find(avl, &search, NULL); 850 851 if (pool) { 852 if (found) 853 *pool = found->aux_pool; 854 else 855 *pool = 0ULL; 856 } 857 858 if (refcnt) { 859 if (found) 860 *refcnt = found->aux_count; 861 else 862 *refcnt = 0; 863 } 864 865 return (found != NULL); 866} 867 868void 869spa_aux_activate(vdev_t *vd, avl_tree_t *avl) 870{ 871 spa_aux_t search, *found; 872 avl_index_t where; 873 874 search.aux_guid = vd->vdev_guid; 875 found = avl_find(avl, &search, &where); 876 ASSERT(found != NULL); 877 ASSERT(found->aux_pool == 0ULL); 878 879 found->aux_pool = spa_guid(vd->vdev_spa); 880} 881 882/* 883 * Spares are tracked globally due to the following constraints: 884 * 885 * - A spare may be part of multiple pools. 886 * - A spare may be added to a pool even if it's actively in use within 887 * another pool. 888 * - A spare in use in any pool can only be the source of a replacement if 889 * the target is a spare in the same pool. 890 * 891 * We keep track of all spares on the system through the use of a reference 892 * counted AVL tree. When a vdev is added as a spare, or used as a replacement 893 * spare, then we bump the reference count in the AVL tree. In addition, we set 894 * the 'vdev_isspare' member to indicate that the device is a spare (active or 895 * inactive). When a spare is made active (used to replace a device in the 896 * pool), we also keep track of which pool its been made a part of. 897 * 898 * The 'spa_spare_lock' protects the AVL tree. These functions are normally 899 * called under the spa_namespace lock as part of vdev reconfiguration. The 900 * separate spare lock exists for the status query path, which does not need to 901 * be completely consistent with respect to other vdev configuration changes. 902 */ 903 904static int 905spa_spare_compare(const void *a, const void *b) 906{ 907 return (spa_aux_compare(a, b)); 908} 909 910void 911spa_spare_add(vdev_t *vd) 912{ 913 mutex_enter(&spa_spare_lock); 914 ASSERT(!vd->vdev_isspare); 915 spa_aux_add(vd, &spa_spare_avl); 916 vd->vdev_isspare = B_TRUE; 917 mutex_exit(&spa_spare_lock); 918} 919 920void 921spa_spare_remove(vdev_t *vd) 922{ 923 mutex_enter(&spa_spare_lock); 924 ASSERT(vd->vdev_isspare); 925 spa_aux_remove(vd, &spa_spare_avl); 926 vd->vdev_isspare = B_FALSE; 927 mutex_exit(&spa_spare_lock); 928} 929 930boolean_t 931spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt) 932{ 933 boolean_t found; 934 935 mutex_enter(&spa_spare_lock); 936 found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl); 937 mutex_exit(&spa_spare_lock); 938 939 return (found); 940} 941 942void 943spa_spare_activate(vdev_t *vd) 944{ 945 mutex_enter(&spa_spare_lock); 946 ASSERT(vd->vdev_isspare); 947 spa_aux_activate(vd, &spa_spare_avl); 948 mutex_exit(&spa_spare_lock); 949} 950 951/* 952 * Level 2 ARC devices are tracked globally for the same reasons as spares. 953 * Cache devices currently only support one pool per cache device, and so 954 * for these devices the aux reference count is currently unused beyond 1. 955 */ 956 957static int 958spa_l2cache_compare(const void *a, const void *b) 959{ 960 return (spa_aux_compare(a, b)); 961} 962 963void 964spa_l2cache_add(vdev_t *vd) 965{ 966 mutex_enter(&spa_l2cache_lock); 967 ASSERT(!vd->vdev_isl2cache); 968 spa_aux_add(vd, &spa_l2cache_avl); 969 vd->vdev_isl2cache = B_TRUE; 970 mutex_exit(&spa_l2cache_lock); 971} 972 973void 974spa_l2cache_remove(vdev_t *vd) 975{ 976 mutex_enter(&spa_l2cache_lock); 977 ASSERT(vd->vdev_isl2cache); 978 spa_aux_remove(vd, &spa_l2cache_avl); 979 vd->vdev_isl2cache = B_FALSE; 980 mutex_exit(&spa_l2cache_lock); 981} 982 983boolean_t 984spa_l2cache_exists(uint64_t guid, uint64_t *pool) 985{ 986 boolean_t found; 987 988 mutex_enter(&spa_l2cache_lock); 989 found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl); 990 mutex_exit(&spa_l2cache_lock); 991 992 return (found); 993} 994 995void 996spa_l2cache_activate(vdev_t *vd) 997{ 998 mutex_enter(&spa_l2cache_lock); 999 ASSERT(vd->vdev_isl2cache); 1000 spa_aux_activate(vd, &spa_l2cache_avl); 1001 mutex_exit(&spa_l2cache_lock); 1002} 1003 1004/* 1005 * ========================================================================== 1006 * SPA vdev locking 1007 * ========================================================================== 1008 */ 1009 1010/* 1011 * Lock the given spa_t for the purpose of adding or removing a vdev. 1012 * Grabs the global spa_namespace_lock plus the spa config lock for writing. 1013 * It returns the next transaction group for the spa_t. 1014 */ 1015uint64_t 1016spa_vdev_enter(spa_t *spa) 1017{ 1018 mutex_enter(&spa->spa_vdev_top_lock); 1019 mutex_enter(&spa_namespace_lock); 1020 return (spa_vdev_config_enter(spa)); 1021} 1022 1023/* 1024 * Internal implementation for spa_vdev_enter(). Used when a vdev 1025 * operation requires multiple syncs (i.e. removing a device) while 1026 * keeping the spa_namespace_lock held. 1027 */ 1028uint64_t 1029spa_vdev_config_enter(spa_t *spa) 1030{ 1031 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1032 1033 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER); 1034 1035 return (spa_last_synced_txg(spa) + 1); 1036} 1037 1038/* 1039 * Used in combination with spa_vdev_config_enter() to allow the syncing 1040 * of multiple transactions without releasing the spa_namespace_lock. 1041 */ 1042void 1043spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag) 1044{ 1045 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1046 1047 int config_changed = B_FALSE; 1048 1049 ASSERT(txg > spa_last_synced_txg(spa)); 1050 1051 spa->spa_pending_vdev = NULL; 1052 1053 /* 1054 * Reassess the DTLs. 1055 */ 1056 vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE); 1057 1058 if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) { 1059 config_changed = B_TRUE; 1060 spa->spa_config_generation++; 1061 } 1062 1063 /* 1064 * Verify the metaslab classes. 1065 */ 1066 ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0); 1067 ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0); 1068 1069 spa_config_exit(spa, SCL_ALL, spa); 1070 1071 /* 1072 * Panic the system if the specified tag requires it. This 1073 * is useful for ensuring that configurations are updated 1074 * transactionally. 1075 */ 1076 if (zio_injection_enabled) 1077 zio_handle_panic_injection(spa, tag, 0); 1078 1079 /* 1080 * Note: this txg_wait_synced() is important because it ensures 1081 * that there won't be more than one config change per txg. 1082 * This allows us to use the txg as the generation number. 1083 */ 1084 if (error == 0) 1085 txg_wait_synced(spa->spa_dsl_pool, txg); 1086 1087 if (vd != NULL) { 1088 ASSERT(!vd->vdev_detached || vd->vdev_dtl_sm == NULL); 1089 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER); 1090 vdev_free(vd); 1091 spa_config_exit(spa, SCL_ALL, spa); 1092 } 1093 1094 /* 1095 * If the config changed, update the config cache. 1096 */ 1097 if (config_changed) 1098 spa_config_sync(spa, B_FALSE, B_TRUE); 1099} 1100 1101/* 1102 * Unlock the spa_t after adding or removing a vdev. Besides undoing the 1103 * locking of spa_vdev_enter(), we also want make sure the transactions have 1104 * synced to disk, and then update the global configuration cache with the new 1105 * information. 1106 */ 1107int 1108spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error) 1109{ 1110 spa_vdev_config_exit(spa, vd, txg, error, FTAG); 1111 mutex_exit(&spa_namespace_lock); 1112 mutex_exit(&spa->spa_vdev_top_lock); 1113 1114 return (error); 1115} 1116 1117/* 1118 * Lock the given spa_t for the purpose of changing vdev state. 1119 */ 1120void 1121spa_vdev_state_enter(spa_t *spa, int oplocks) 1122{ 1123 int locks = SCL_STATE_ALL | oplocks; 1124 1125 /* 1126 * Root pools may need to read of the underlying devfs filesystem 1127 * when opening up a vdev. Unfortunately if we're holding the 1128 * SCL_ZIO lock it will result in a deadlock when we try to issue 1129 * the read from the root filesystem. Instead we "prefetch" 1130 * the associated vnodes that we need prior to opening the 1131 * underlying devices and cache them so that we can prevent 1132 * any I/O when we are doing the actual open. 1133 */ 1134 if (spa_is_root(spa)) { 1135 int low = locks & ~(SCL_ZIO - 1); 1136 int high = locks & ~low; 1137 1138 spa_config_enter(spa, high, spa, RW_WRITER); 1139 vdev_hold(spa->spa_root_vdev); 1140 spa_config_enter(spa, low, spa, RW_WRITER); 1141 } else { 1142 spa_config_enter(spa, locks, spa, RW_WRITER); 1143 } 1144 spa->spa_vdev_locks = locks; 1145} 1146 1147int 1148spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error) 1149{ 1150 boolean_t config_changed = B_FALSE; 1151 1152 if (vd != NULL || error == 0) 1153 vdev_dtl_reassess(vd ? vd->vdev_top : spa->spa_root_vdev, 1154 0, 0, B_FALSE); 1155 1156 if (vd != NULL) { 1157 vdev_state_dirty(vd->vdev_top); 1158 config_changed = B_TRUE; 1159 spa->spa_config_generation++; 1160 } 1161 1162 if (spa_is_root(spa)) 1163 vdev_rele(spa->spa_root_vdev); 1164 1165 ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL); 1166 spa_config_exit(spa, spa->spa_vdev_locks, spa); 1167 1168 /* 1169 * If anything changed, wait for it to sync. This ensures that, 1170 * from the system administrator's perspective, zpool(1M) commands 1171 * are synchronous. This is important for things like zpool offline: 1172 * when the command completes, you expect no further I/O from ZFS. 1173 */ 1174 if (vd != NULL) 1175 txg_wait_synced(spa->spa_dsl_pool, 0); 1176 1177 /* 1178 * If the config changed, update the config cache. 1179 */ 1180 if (config_changed) { 1181 mutex_enter(&spa_namespace_lock); 1182 spa_config_sync(spa, B_FALSE, B_TRUE); 1183 mutex_exit(&spa_namespace_lock); 1184 } 1185 1186 return (error); 1187} 1188 1189/* 1190 * ========================================================================== 1191 * Miscellaneous functions 1192 * ========================================================================== 1193 */ 1194 1195void 1196spa_activate_mos_feature(spa_t *spa, const char *feature) 1197{ 1198 if (!nvlist_exists(spa->spa_label_features, feature)) { 1199 fnvlist_add_boolean(spa->spa_label_features, feature); 1200 vdev_config_dirty(spa->spa_root_vdev); 1201 } 1202} 1203 1204void 1205spa_deactivate_mos_feature(spa_t *spa, const char *feature) 1206{ 1207 if (nvlist_remove_all(spa->spa_label_features, feature) == 0) 1208 vdev_config_dirty(spa->spa_root_vdev); 1209} 1210 1211/* 1212 * Rename a spa_t. 1213 */ 1214int 1215spa_rename(const char *name, const char *newname) 1216{ 1217 spa_t *spa; 1218 int err; 1219 1220 /* 1221 * Lookup the spa_t and grab the config lock for writing. We need to 1222 * actually open the pool so that we can sync out the necessary labels. 1223 * It's OK to call spa_open() with the namespace lock held because we 1224 * allow recursive calls for other reasons. 1225 */ 1226 mutex_enter(&spa_namespace_lock); 1227 if ((err = spa_open(name, &spa, FTAG)) != 0) { 1228 mutex_exit(&spa_namespace_lock); 1229 return (err); 1230 } 1231 1232 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1233 1234 avl_remove(&spa_namespace_avl, spa); 1235 (void) strlcpy(spa->spa_name, newname, sizeof (spa->spa_name)); 1236 avl_add(&spa_namespace_avl, spa); 1237 1238 /* 1239 * Sync all labels to disk with the new names by marking the root vdev 1240 * dirty and waiting for it to sync. It will pick up the new pool name 1241 * during the sync. 1242 */ 1243 vdev_config_dirty(spa->spa_root_vdev); 1244 1245 spa_config_exit(spa, SCL_ALL, FTAG); 1246 1247 txg_wait_synced(spa->spa_dsl_pool, 0); 1248 1249 /* 1250 * Sync the updated config cache. 1251 */ 1252 spa_config_sync(spa, B_FALSE, B_TRUE); 1253 1254 spa_close(spa, FTAG); 1255 1256 mutex_exit(&spa_namespace_lock); 1257 1258 return (0); 1259} 1260 1261/* 1262 * Return the spa_t associated with given pool_guid, if it exists. If 1263 * device_guid is non-zero, determine whether the pool exists *and* contains 1264 * a device with the specified device_guid. 1265 */ 1266spa_t * 1267spa_by_guid(uint64_t pool_guid, uint64_t device_guid) 1268{ 1269 spa_t *spa; 1270 avl_tree_t *t = &spa_namespace_avl; 1271 1272 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1273 1274 for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) { 1275 if (spa->spa_state == POOL_STATE_UNINITIALIZED) 1276 continue; 1277 if (spa->spa_root_vdev == NULL) 1278 continue; 1279 if (spa_guid(spa) == pool_guid) { 1280 if (device_guid == 0) 1281 break; 1282 1283 if (vdev_lookup_by_guid(spa->spa_root_vdev, 1284 device_guid) != NULL) 1285 break; 1286 1287 /* 1288 * Check any devices we may be in the process of adding. 1289 */ 1290 if (spa->spa_pending_vdev) { 1291 if (vdev_lookup_by_guid(spa->spa_pending_vdev, 1292 device_guid) != NULL) 1293 break; 1294 } 1295 } 1296 } 1297 1298 return (spa); 1299} 1300 1301/* 1302 * Determine whether a pool with the given pool_guid exists. 1303 */ 1304boolean_t 1305spa_guid_exists(uint64_t pool_guid, uint64_t device_guid) 1306{ 1307 return (spa_by_guid(pool_guid, device_guid) != NULL); 1308} 1309 1310char * 1311spa_strdup(const char *s) 1312{ 1313 size_t len; 1314 char *new; 1315 1316 len = strlen(s); 1317 new = kmem_alloc(len + 1, KM_SLEEP); 1318 bcopy(s, new, len); 1319 new[len] = '\0'; 1320 1321 return (new); 1322} 1323 1324void 1325spa_strfree(char *s) 1326{ 1327 kmem_free(s, strlen(s) + 1); 1328} 1329 1330uint64_t 1331spa_get_random(uint64_t range) 1332{ 1333 uint64_t r; 1334 1335 ASSERT(range != 0); 1336 1337 (void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t)); 1338 1339 return (r % range); 1340} 1341 1342uint64_t 1343spa_generate_guid(spa_t *spa) 1344{ 1345 uint64_t guid = spa_get_random(-1ULL); 1346 1347 if (spa != NULL) { 1348 while (guid == 0 || spa_guid_exists(spa_guid(spa), guid)) 1349 guid = spa_get_random(-1ULL); 1350 } else { 1351 while (guid == 0 || spa_guid_exists(guid, 0)) 1352 guid = spa_get_random(-1ULL); 1353 } 1354 1355 return (guid); 1356} 1357 1358void 1359sprintf_blkptr(char *buf, const blkptr_t *bp) 1360{ 1361 char type[256]; 1362 char *checksum = NULL; 1363 char *compress = NULL; 1364 1365 if (bp != NULL) { 1366 if (BP_GET_TYPE(bp) & DMU_OT_NEWTYPE) { 1367 dmu_object_byteswap_t bswap = 1368 DMU_OT_BYTESWAP(BP_GET_TYPE(bp)); 1369 (void) snprintf(type, sizeof (type), "bswap %s %s", 1370 DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) ? 1371 "metadata" : "data", 1372 dmu_ot_byteswap[bswap].ob_name); 1373 } else { 1374 (void) strlcpy(type, dmu_ot[BP_GET_TYPE(bp)].ot_name, 1375 sizeof (type)); 1376 } 1377 checksum = zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name; 1378 compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name; 1379 } 1380 1381 SPRINTF_BLKPTR(snprintf, ' ', buf, bp, type, checksum, compress); 1382} 1383 1384void 1385spa_freeze(spa_t *spa) 1386{ 1387 uint64_t freeze_txg = 0; 1388 1389 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1390 if (spa->spa_freeze_txg == UINT64_MAX) { 1391 freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE; 1392 spa->spa_freeze_txg = freeze_txg; 1393 } 1394 spa_config_exit(spa, SCL_ALL, FTAG); 1395 if (freeze_txg != 0) 1396 txg_wait_synced(spa_get_dsl(spa), freeze_txg); 1397} 1398 1399void 1400zfs_panic_recover(const char *fmt, ...) 1401{ 1402 va_list adx; 1403 1404 va_start(adx, fmt); 1405 vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx); 1406 va_end(adx); 1407} 1408 1409/* 1410 * This is a stripped-down version of strtoull, suitable only for converting 1411 * lowercase hexadecimal numbers that don't overflow. 1412 */ 1413uint64_t 1414zfs_strtonum(const char *str, char **nptr) 1415{ 1416 uint64_t val = 0; 1417 char c; 1418 int digit; 1419 1420 while ((c = *str) != '\0') { 1421 if (c >= '0' && c <= '9') 1422 digit = c - '0'; 1423 else if (c >= 'a' && c <= 'f') 1424 digit = 10 + c - 'a'; 1425 else 1426 break; 1427 1428 val *= 16; 1429 val += digit; 1430 1431 str++; 1432 } 1433 1434 if (nptr) 1435 *nptr = (char *)str; 1436 1437 return (val); 1438} 1439 1440/* 1441 * ========================================================================== 1442 * Accessor functions 1443 * ========================================================================== 1444 */ 1445 1446boolean_t 1447spa_shutting_down(spa_t *spa) 1448{ 1449 return (spa->spa_async_suspended); 1450} 1451 1452dsl_pool_t * 1453spa_get_dsl(spa_t *spa) 1454{ 1455 return (spa->spa_dsl_pool); 1456} 1457 1458boolean_t 1459spa_is_initializing(spa_t *spa) 1460{ 1461 return (spa->spa_is_initializing); 1462} 1463 1464blkptr_t * 1465spa_get_rootblkptr(spa_t *spa) 1466{ 1467 return (&spa->spa_ubsync.ub_rootbp); 1468} 1469 1470void 1471spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp) 1472{ 1473 spa->spa_uberblock.ub_rootbp = *bp; 1474} 1475 1476void 1477spa_altroot(spa_t *spa, char *buf, size_t buflen) 1478{ 1479 if (spa->spa_root == NULL) 1480 buf[0] = '\0'; 1481 else 1482 (void) strncpy(buf, spa->spa_root, buflen); 1483} 1484 1485int 1486spa_sync_pass(spa_t *spa) 1487{ 1488 return (spa->spa_sync_pass); 1489} 1490 1491char * 1492spa_name(spa_t *spa) 1493{ 1494 return (spa->spa_name); 1495} 1496 1497uint64_t 1498spa_guid(spa_t *spa) 1499{ 1500 dsl_pool_t *dp = spa_get_dsl(spa); 1501 uint64_t guid; 1502 1503 /* 1504 * If we fail to parse the config during spa_load(), we can go through 1505 * the error path (which posts an ereport) and end up here with no root 1506 * vdev. We stash the original pool guid in 'spa_config_guid' to handle 1507 * this case. 1508 */ 1509 if (spa->spa_root_vdev == NULL) 1510 return (spa->spa_config_guid); 1511 1512 guid = spa->spa_last_synced_guid != 0 ? 1513 spa->spa_last_synced_guid : spa->spa_root_vdev->vdev_guid; 1514 1515 /* 1516 * Return the most recently synced out guid unless we're 1517 * in syncing context. 1518 */ 1519 if (dp && dsl_pool_sync_context(dp)) 1520 return (spa->spa_root_vdev->vdev_guid); 1521 else 1522 return (guid); 1523} 1524 1525uint64_t 1526spa_load_guid(spa_t *spa) 1527{ 1528 /* 1529 * This is a GUID that exists solely as a reference for the 1530 * purposes of the arc. It is generated at load time, and 1531 * is never written to persistent storage. 1532 */ 1533 return (spa->spa_load_guid); 1534} 1535 1536uint64_t 1537spa_last_synced_txg(spa_t *spa) 1538{ 1539 return (spa->spa_ubsync.ub_txg); 1540} 1541 1542uint64_t 1543spa_first_txg(spa_t *spa) 1544{ 1545 return (spa->spa_first_txg); 1546} 1547 1548uint64_t 1549spa_syncing_txg(spa_t *spa) 1550{ 1551 return (spa->spa_syncing_txg); 1552} 1553 1554pool_state_t 1555spa_state(spa_t *spa) 1556{ 1557 return (spa->spa_state); 1558} 1559 1560spa_load_state_t 1561spa_load_state(spa_t *spa) 1562{ 1563 return (spa->spa_load_state); 1564} 1565 1566uint64_t 1567spa_freeze_txg(spa_t *spa) 1568{ 1569 return (spa->spa_freeze_txg); 1570} 1571 1572/* ARGSUSED */ 1573uint64_t 1574spa_get_asize(spa_t *spa, uint64_t lsize) 1575{ 1576 return (lsize * spa_asize_inflation); 1577} 1578 1579uint64_t 1580spa_get_dspace(spa_t *spa) 1581{ 1582 return (spa->spa_dspace); 1583} 1584 1585void 1586spa_update_dspace(spa_t *spa) 1587{ 1588 spa->spa_dspace = metaslab_class_get_dspace(spa_normal_class(spa)) + 1589 ddt_get_dedup_dspace(spa); 1590} 1591 1592/* 1593 * Return the failure mode that has been set to this pool. The default 1594 * behavior will be to block all I/Os when a complete failure occurs. 1595 */ 1596uint8_t 1597spa_get_failmode(spa_t *spa) 1598{ 1599 return (spa->spa_failmode); 1600} 1601 1602boolean_t 1603spa_suspended(spa_t *spa) 1604{ 1605 return (spa->spa_suspended); 1606} 1607 1608uint64_t 1609spa_version(spa_t *spa) 1610{ 1611 return (spa->spa_ubsync.ub_version); 1612} 1613 1614boolean_t 1615spa_deflate(spa_t *spa) 1616{ 1617 return (spa->spa_deflate); 1618} 1619 1620metaslab_class_t * 1621spa_normal_class(spa_t *spa) 1622{ 1623 return (spa->spa_normal_class); 1624} 1625 1626metaslab_class_t * 1627spa_log_class(spa_t *spa) 1628{ 1629 return (spa->spa_log_class); 1630} 1631 1632int 1633spa_max_replication(spa_t *spa) 1634{ 1635 /* 1636 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to 1637 * handle BPs with more than one DVA allocated. Set our max 1638 * replication level accordingly. 1639 */ 1640 if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS) 1641 return (1); 1642 return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override)); 1643} 1644 1645int 1646spa_prev_software_version(spa_t *spa) 1647{ 1648 return (spa->spa_prev_software_version); 1649} 1650 1651uint64_t 1652spa_deadman_synctime(spa_t *spa) 1653{ 1654 return (spa->spa_deadman_synctime); 1655} 1656 1657uint64_t 1658dva_get_dsize_sync(spa_t *spa, const dva_t *dva) 1659{ 1660 uint64_t asize = DVA_GET_ASIZE(dva); 1661 uint64_t dsize = asize; 1662 1663 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 1664 1665 if (asize != 0 && spa->spa_deflate) { 1666 vdev_t *vd = vdev_lookup_top(spa, DVA_GET_VDEV(dva)); 1667 dsize = (asize >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio; 1668 } 1669 1670 return (dsize); 1671} 1672 1673uint64_t 1674bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp) 1675{ 1676 uint64_t dsize = 0; 1677 1678 for (int d = 0; d < SPA_DVAS_PER_BP; d++) 1679 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]); 1680 1681 return (dsize); 1682} 1683 1684uint64_t 1685bp_get_dsize(spa_t *spa, const blkptr_t *bp) 1686{ 1687 uint64_t dsize = 0; 1688 1689 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); 1690 1691 for (int d = 0; d < SPA_DVAS_PER_BP; d++) 1692 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]); 1693 1694 spa_config_exit(spa, SCL_VDEV, FTAG); 1695 1696 return (dsize); 1697} 1698 1699/* 1700 * ========================================================================== 1701 * Initialization and Termination 1702 * ========================================================================== 1703 */ 1704 1705static int 1706spa_name_compare(const void *a1, const void *a2) 1707{ 1708 const spa_t *s1 = a1; 1709 const spa_t *s2 = a2; 1710 int s; 1711 1712 s = strcmp(s1->spa_name, s2->spa_name); 1713 if (s > 0) 1714 return (1); 1715 if (s < 0) 1716 return (-1); 1717 return (0); 1718} 1719 1720int 1721spa_busy(void) 1722{ 1723 return (spa_active_count); 1724} 1725 1726void 1727spa_boot_init() 1728{ 1729 spa_config_load(); 1730} 1731 1732#ifdef _KERNEL 1733EVENTHANDLER_DEFINE(mountroot, spa_boot_init, NULL, 0); 1734#endif 1735 1736void 1737spa_init(int mode) 1738{ 1739 mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL); 1740 mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL); 1741 mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL); 1742 cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL); 1743 1744 avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t), 1745 offsetof(spa_t, spa_avl)); 1746 1747 avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t), 1748 offsetof(spa_aux_t, aux_avl)); 1749 1750 avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t), 1751 offsetof(spa_aux_t, aux_avl)); 1752 1753 spa_mode_global = mode; 1754 1755#ifdef illumos 1756#ifdef _KERNEL 1757 spa_arch_init(); 1758#else 1759 if (spa_mode_global != FREAD && dprintf_find_string("watch")) { 1760 arc_procfd = open("/proc/self/ctl", O_WRONLY); 1761 if (arc_procfd == -1) { 1762 perror("could not enable watchpoints: " 1763 "opening /proc/self/ctl failed: "); 1764 } else { 1765 arc_watch = B_TRUE; 1766 } 1767 } 1768#endif 1769#endif /* illumos */ 1770 refcount_sysinit(); 1771 unique_init(); 1772 range_tree_init(); 1773 zio_init(); 1774 lz4_init(); 1775 dmu_init(); 1776 zil_init(); 1777 vdev_cache_stat_init(); 1778 zfs_prop_init(); 1779 zpool_prop_init(); 1780 zpool_feature_init(); 1781 spa_config_load(); 1782 l2arc_start(); 1783#ifndef illumos 1784#ifdef _KERNEL 1785 zfs_deadman_init(); 1786#endif 1787#endif /* !illumos */ 1788} 1789 1790void 1791spa_fini(void) 1792{ 1793 l2arc_stop(); 1794 1795 spa_evict_all(); 1796 1797 vdev_cache_stat_fini(); 1798 zil_fini(); 1799 dmu_fini(); 1800 lz4_fini(); 1801 zio_fini(); 1802 range_tree_fini(); 1803 unique_fini(); 1804 refcount_fini(); 1805 1806 avl_destroy(&spa_namespace_avl); 1807 avl_destroy(&spa_spare_avl); 1808 avl_destroy(&spa_l2cache_avl); 1809 1810 cv_destroy(&spa_namespace_cv); 1811 mutex_destroy(&spa_namespace_lock); 1812 mutex_destroy(&spa_spare_lock); 1813 mutex_destroy(&spa_l2cache_lock); 1814} 1815 1816/* 1817 * Return whether this pool has slogs. No locking needed. 1818 * It's not a problem if the wrong answer is returned as it's only for 1819 * performance and not correctness 1820 */ 1821boolean_t 1822spa_has_slogs(spa_t *spa) 1823{ 1824 return (spa->spa_log_class->mc_rotor != NULL); 1825} 1826 1827spa_log_state_t 1828spa_get_log_state(spa_t *spa) 1829{ 1830 return (spa->spa_log_state); 1831} 1832 1833void 1834spa_set_log_state(spa_t *spa, spa_log_state_t state) 1835{ 1836 spa->spa_log_state = state; 1837} 1838 1839boolean_t 1840spa_is_root(spa_t *spa) 1841{ 1842 return (spa->spa_is_root); 1843} 1844 1845boolean_t 1846spa_writeable(spa_t *spa) 1847{ 1848 return (!!(spa->spa_mode & FWRITE)); 1849} 1850 1851int 1852spa_mode(spa_t *spa) 1853{ 1854 return (spa->spa_mode); 1855} 1856 1857uint64_t 1858spa_bootfs(spa_t *spa) 1859{ 1860 return (spa->spa_bootfs); 1861} 1862 1863uint64_t 1864spa_delegation(spa_t *spa) 1865{ 1866 return (spa->spa_delegation); 1867} 1868 1869objset_t * 1870spa_meta_objset(spa_t *spa) 1871{ 1872 return (spa->spa_meta_objset); 1873} 1874 1875enum zio_checksum 1876spa_dedup_checksum(spa_t *spa) 1877{ 1878 return (spa->spa_dedup_checksum); 1879} 1880 1881/* 1882 * Reset pool scan stat per scan pass (or reboot). 1883 */ 1884void 1885spa_scan_stat_init(spa_t *spa) 1886{ 1887 /* data not stored on disk */ 1888 spa->spa_scan_pass_start = gethrestime_sec(); 1889 spa->spa_scan_pass_exam = 0; 1890 vdev_scan_stat_init(spa->spa_root_vdev); 1891} 1892 1893/* 1894 * Get scan stats for zpool status reports 1895 */ 1896int 1897spa_scan_get_stats(spa_t *spa, pool_scan_stat_t *ps) 1898{ 1899 dsl_scan_t *scn = spa->spa_dsl_pool ? spa->spa_dsl_pool->dp_scan : NULL; 1900 1901 if (scn == NULL || scn->scn_phys.scn_func == POOL_SCAN_NONE) 1902 return (SET_ERROR(ENOENT)); 1903 bzero(ps, sizeof (pool_scan_stat_t)); 1904 1905 /* data stored on disk */ 1906 ps->pss_func = scn->scn_phys.scn_func; 1907 ps->pss_start_time = scn->scn_phys.scn_start_time; 1908 ps->pss_end_time = scn->scn_phys.scn_end_time; 1909 ps->pss_to_examine = scn->scn_phys.scn_to_examine; 1910 ps->pss_examined = scn->scn_phys.scn_examined; 1911 ps->pss_to_process = scn->scn_phys.scn_to_process; 1912 ps->pss_processed = scn->scn_phys.scn_processed; 1913 ps->pss_errors = scn->scn_phys.scn_errors; 1914 ps->pss_state = scn->scn_phys.scn_state; 1915 1916 /* data not stored on disk */ 1917 ps->pss_pass_start = spa->spa_scan_pass_start; 1918 ps->pss_pass_exam = spa->spa_scan_pass_exam; 1919 1920 return (0); 1921} 1922 1923boolean_t 1924spa_debug_enabled(spa_t *spa) 1925{ 1926 return (spa->spa_debug); 1927} 1928