spa_misc.c revision 288549
1/* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21/* 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23 * Copyright (c) 2011, 2015 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 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. 27 */ 28 29#include <sys/zfs_context.h> 30#include <sys/spa_impl.h> 31#include <sys/spa_boot.h> 32#include <sys/zio.h> 33#include <sys/zio_checksum.h> 34#include <sys/zio_compress.h> 35#include <sys/dmu.h> 36#include <sys/dmu_tx.h> 37#include <sys/zap.h> 38#include <sys/zil.h> 39#include <sys/vdev_impl.h> 40#include <sys/metaslab.h> 41#include <sys/uberblock_impl.h> 42#include <sys/txg.h> 43#include <sys/avl.h> 44#include <sys/unique.h> 45#include <sys/dsl_pool.h> 46#include <sys/dsl_dir.h> 47#include <sys/dsl_prop.h> 48#include <sys/dsl_scan.h> 49#include <sys/fs/zfs.h> 50#include <sys/metaslab_impl.h> 51#include <sys/arc.h> 52#include <sys/ddt.h> 53#include "zfs_prop.h" 54#include "zfeature_common.h" 55 56/* 57 * SPA locking 58 * 59 * There are four basic locks for managing spa_t structures: 60 * 61 * spa_namespace_lock (global mutex) 62 * 63 * This lock must be acquired to do any of the following: 64 * 65 * - Lookup a spa_t by name 66 * - Add or remove a spa_t from the namespace 67 * - Increase spa_refcount from non-zero 68 * - Check if spa_refcount is zero 69 * - Rename a spa_t 70 * - add/remove/attach/detach devices 71 * - Held for the duration of create/destroy/import/export 72 * 73 * It does not need to handle recursion. A create or destroy may 74 * reference objects (files or zvols) in other pools, but by 75 * definition they must have an existing reference, and will never need 76 * to lookup a spa_t by name. 77 * 78 * spa_refcount (per-spa refcount_t protected by mutex) 79 * 80 * This reference count keep track of any active users of the spa_t. The 81 * spa_t cannot be destroyed or freed while this is non-zero. Internally, 82 * the refcount is never really 'zero' - opening a pool implicitly keeps 83 * some references in the DMU. Internally we check against spa_minref, but 84 * present the image of a zero/non-zero value to consumers. 85 * 86 * spa_config_lock[] (per-spa array of rwlocks) 87 * 88 * This protects the spa_t from config changes, and must be held in 89 * the following circumstances: 90 * 91 * - RW_READER to perform I/O to the spa 92 * - RW_WRITER to change the vdev config 93 * 94 * The locking order is fairly straightforward: 95 * 96 * spa_namespace_lock -> spa_refcount 97 * 98 * The namespace lock must be acquired to increase the refcount from 0 99 * or to check if it is zero. 100 * 101 * spa_refcount -> spa_config_lock[] 102 * 103 * There must be at least one valid reference on the spa_t to acquire 104 * the config lock. 105 * 106 * spa_namespace_lock -> spa_config_lock[] 107 * 108 * The namespace lock must always be taken before the config lock. 109 * 110 * 111 * The spa_namespace_lock can be acquired directly and is globally visible. 112 * 113 * The namespace is manipulated using the following functions, all of which 114 * require the spa_namespace_lock to be held. 115 * 116 * spa_lookup() Lookup a spa_t by name. 117 * 118 * spa_add() Create a new spa_t in the namespace. 119 * 120 * spa_remove() Remove a spa_t from the namespace. This also 121 * frees up any memory associated with the spa_t. 122 * 123 * spa_next() Returns the next spa_t in the system, or the 124 * first if NULL is passed. 125 * 126 * spa_evict_all() Shutdown and remove all spa_t structures in 127 * the system. 128 * 129 * spa_guid_exists() Determine whether a pool/device guid exists. 130 * 131 * The spa_refcount is manipulated using the following functions: 132 * 133 * spa_open_ref() Adds a reference to the given spa_t. Must be 134 * called with spa_namespace_lock held if the 135 * refcount is currently zero. 136 * 137 * spa_close() Remove a reference from the spa_t. This will 138 * not free the spa_t or remove it from the 139 * namespace. No locking is required. 140 * 141 * spa_refcount_zero() Returns true if the refcount is currently 142 * zero. Must be called with spa_namespace_lock 143 * held. 144 * 145 * The spa_config_lock[] is an array of rwlocks, ordered as follows: 146 * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV. 147 * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}(). 148 * 149 * To read the configuration, it suffices to hold one of these locks as reader. 150 * To modify the configuration, you must hold all locks as writer. To modify 151 * vdev state without altering the vdev tree's topology (e.g. online/offline), 152 * you must hold SCL_STATE and SCL_ZIO as writer. 153 * 154 * We use these distinct config locks to avoid recursive lock entry. 155 * For example, spa_sync() (which holds SCL_CONFIG as reader) induces 156 * block allocations (SCL_ALLOC), which may require reading space maps 157 * from disk (dmu_read() -> zio_read() -> SCL_ZIO). 158 * 159 * The spa config locks cannot be normal rwlocks because we need the 160 * ability to hand off ownership. For example, SCL_ZIO is acquired 161 * by the issuing thread and later released by an interrupt thread. 162 * They do, however, obey the usual write-wanted semantics to prevent 163 * writer (i.e. system administrator) starvation. 164 * 165 * The lock acquisition rules are as follows: 166 * 167 * SCL_CONFIG 168 * Protects changes to the vdev tree topology, such as vdev 169 * add/remove/attach/detach. Protects the dirty config list 170 * (spa_config_dirty_list) and the set of spares and l2arc devices. 171 * 172 * SCL_STATE 173 * Protects changes to pool state and vdev state, such as vdev 174 * online/offline/fault/degrade/clear. Protects the dirty state list 175 * (spa_state_dirty_list) and global pool state (spa_state). 176 * 177 * SCL_ALLOC 178 * Protects changes to metaslab groups and classes. 179 * Held as reader by metaslab_alloc() and metaslab_claim(). 180 * 181 * SCL_ZIO 182 * Held by bp-level zios (those which have no io_vd upon entry) 183 * to prevent changes to the vdev tree. The bp-level zio implicitly 184 * protects all of its vdev child zios, which do not hold SCL_ZIO. 185 * 186 * SCL_FREE 187 * Protects changes to metaslab groups and classes. 188 * Held as reader by metaslab_free(). SCL_FREE is distinct from 189 * SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free 190 * blocks in zio_done() while another i/o that holds either 191 * SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete. 192 * 193 * SCL_VDEV 194 * Held as reader to prevent changes to the vdev tree during trivial 195 * inquiries such as bp_get_dsize(). SCL_VDEV is distinct from the 196 * other locks, and lower than all of them, to ensure that it's safe 197 * to acquire regardless of caller context. 198 * 199 * In addition, the following rules apply: 200 * 201 * (a) spa_props_lock protects pool properties, spa_config and spa_config_list. 202 * The lock ordering is SCL_CONFIG > spa_props_lock. 203 * 204 * (b) I/O operations on leaf vdevs. For any zio operation that takes 205 * an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(), 206 * or zio_write_phys() -- the caller must ensure that the config cannot 207 * cannot change in the interim, and that the vdev cannot be reopened. 208 * SCL_STATE as reader suffices for both. 209 * 210 * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit(). 211 * 212 * spa_vdev_enter() Acquire the namespace lock and the config lock 213 * for writing. 214 * 215 * spa_vdev_exit() Release the config lock, wait for all I/O 216 * to complete, sync the updated configs to the 217 * cache, and release the namespace lock. 218 * 219 * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit(). 220 * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual 221 * locking is, always, based on spa_namespace_lock and spa_config_lock[]. 222 * 223 * spa_rename() is also implemented within this file since it requires 224 * manipulation of the namespace. 225 */ 226 227static avl_tree_t spa_namespace_avl; 228kmutex_t spa_namespace_lock; 229static kcondvar_t spa_namespace_cv; 230static int spa_active_count; 231int spa_max_replication_override = SPA_DVAS_PER_BP; 232 233static kmutex_t spa_spare_lock; 234static avl_tree_t spa_spare_avl; 235static kmutex_t spa_l2cache_lock; 236static avl_tree_t spa_l2cache_avl; 237 238kmem_cache_t *spa_buffer_pool; 239int spa_mode_global; 240 241#ifdef ZFS_DEBUG 242/* Everything except dprintf and spa is on by default in debug builds */ 243int zfs_flags = ~(ZFS_DEBUG_DPRINTF | ZFS_DEBUG_SPA); 244#else 245int zfs_flags = 0; 246#endif 247SYSCTL_DECL(_debug); 248TUNABLE_INT("debug.zfs_flags", &zfs_flags); 249SYSCTL_INT(_debug, OID_AUTO, zfs_flags, CTLFLAG_RWTUN, &zfs_flags, 0, 250 "ZFS debug flags."); 251 252/* 253 * zfs_recover can be set to nonzero to attempt to recover from 254 * otherwise-fatal errors, typically caused by on-disk corruption. When 255 * set, calls to zfs_panic_recover() will turn into warning messages. 256 * This should only be used as a last resort, as it typically results 257 * in leaked space, or worse. 258 */ 259boolean_t zfs_recover = B_FALSE; 260SYSCTL_DECL(_vfs_zfs); 261TUNABLE_INT("vfs.zfs.recover", &zfs_recover); 262SYSCTL_INT(_vfs_zfs, OID_AUTO, recover, CTLFLAG_RDTUN, &zfs_recover, 0, 263 "Try to recover from otherwise-fatal errors."); 264 265/* 266 * If destroy encounters an EIO while reading metadata (e.g. indirect 267 * blocks), space referenced by the missing metadata can not be freed. 268 * Normally this causes the background destroy to become "stalled", as 269 * it is unable to make forward progress. While in this stalled state, 270 * all remaining space to free from the error-encountering filesystem is 271 * "temporarily leaked". Set this flag to cause it to ignore the EIO, 272 * permanently leak the space from indirect blocks that can not be read, 273 * and continue to free everything else that it can. 274 * 275 * The default, "stalling" behavior is useful if the storage partially 276 * fails (i.e. some but not all i/os fail), and then later recovers. In 277 * this case, we will be able to continue pool operations while it is 278 * partially failed, and when it recovers, we can continue to free the 279 * space, with no leaks. However, note that this case is actually 280 * fairly rare. 281 * 282 * Typically pools either (a) fail completely (but perhaps temporarily, 283 * e.g. a top-level vdev going offline), or (b) have localized, 284 * permanent errors (e.g. disk returns the wrong data due to bit flip or 285 * firmware bug). In case (a), this setting does not matter because the 286 * pool will be suspended and the sync thread will not be able to make 287 * forward progress regardless. In case (b), because the error is 288 * permanent, the best we can do is leak the minimum amount of space, 289 * which is what setting this flag will do. Therefore, it is reasonable 290 * for this flag to normally be set, but we chose the more conservative 291 * approach of not setting it, so that there is no possibility of 292 * leaking space in the "partial temporary" failure case. 293 */ 294boolean_t zfs_free_leak_on_eio = B_FALSE; 295 296/* 297 * Expiration time in milliseconds. This value has two meanings. First it is 298 * used to determine when the spa_deadman() logic should fire. By default the 299 * spa_deadman() will fire if spa_sync() has not completed in 1000 seconds. 300 * Secondly, the value determines if an I/O is considered "hung". Any I/O that 301 * has not completed in zfs_deadman_synctime_ms is considered "hung" resulting 302 * in a system panic. 303 */ 304uint64_t zfs_deadman_synctime_ms = 1000000ULL; 305TUNABLE_QUAD("vfs.zfs.deadman_synctime_ms", &zfs_deadman_synctime_ms); 306SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, deadman_synctime_ms, CTLFLAG_RDTUN, 307 &zfs_deadman_synctime_ms, 0, 308 "Stalled ZFS I/O expiration time in milliseconds"); 309 310/* 311 * Check time in milliseconds. This defines the frequency at which we check 312 * for hung I/O. 313 */ 314uint64_t zfs_deadman_checktime_ms = 5000ULL; 315TUNABLE_QUAD("vfs.zfs.deadman_checktime_ms", &zfs_deadman_checktime_ms); 316SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, deadman_checktime_ms, CTLFLAG_RDTUN, 317 &zfs_deadman_checktime_ms, 0, 318 "Period of checks for stalled ZFS I/O in milliseconds"); 319 320/* 321 * Default value of -1 for zfs_deadman_enabled is resolved in 322 * zfs_deadman_init() 323 */ 324int zfs_deadman_enabled = -1; 325TUNABLE_INT("vfs.zfs.deadman_enabled", &zfs_deadman_enabled); 326SYSCTL_INT(_vfs_zfs, OID_AUTO, deadman_enabled, CTLFLAG_RDTUN, 327 &zfs_deadman_enabled, 0, "Kernel panic on stalled ZFS I/O"); 328 329/* 330 * The worst case is single-sector max-parity RAID-Z blocks, in which 331 * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1) 332 * times the size; so just assume that. Add to this the fact that 333 * we can have up to 3 DVAs per bp, and one more factor of 2 because 334 * the block may be dittoed with up to 3 DVAs by ddt_sync(). All together, 335 * the worst case is: 336 * (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2 == 24 337 */ 338int spa_asize_inflation = 24; 339TUNABLE_INT("vfs.zfs.spa_asize_inflation", &spa_asize_inflation); 340SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_asize_inflation, CTLFLAG_RWTUN, 341 &spa_asize_inflation, 0, "Worst case inflation factor for single sector writes"); 342 343#ifndef illumos 344#ifdef _KERNEL 345static void 346zfs_deadman_init() 347{ 348 /* 349 * If we are not i386 or amd64 or in a virtual machine, 350 * disable ZFS deadman thread by default 351 */ 352 if (zfs_deadman_enabled == -1) { 353#if defined(__amd64__) || defined(__i386__) 354 zfs_deadman_enabled = (vm_guest == VM_GUEST_NO) ? 1 : 0; 355#else 356 zfs_deadman_enabled = 0; 357#endif 358 } 359} 360#endif /* _KERNEL */ 361#endif /* !illumos */ 362 363/* 364 * Normally, we don't allow the last 3.2% (1/(2^spa_slop_shift)) of space in 365 * the pool to be consumed. This ensures that we don't run the pool 366 * completely out of space, due to unaccounted changes (e.g. to the MOS). 367 * It also limits the worst-case time to allocate space. If we have 368 * less than this amount of free space, most ZPL operations (e.g. write, 369 * create) will return ENOSPC. 370 * 371 * Certain operations (e.g. file removal, most administrative actions) can 372 * use half the slop space. They will only return ENOSPC if less than half 373 * the slop space is free. Typically, once the pool has less than the slop 374 * space free, the user will use these operations to free up space in the pool. 375 * These are the operations that call dsl_pool_adjustedsize() with the netfree 376 * argument set to TRUE. 377 * 378 * A very restricted set of operations are always permitted, regardless of 379 * the amount of free space. These are the operations that call 380 * dsl_sync_task(ZFS_SPACE_CHECK_NONE), e.g. "zfs destroy". If these 381 * operations result in a net increase in the amount of space used, 382 * it is possible to run the pool completely out of space, causing it to 383 * be permanently read-only. 384 * 385 * See also the comments in zfs_space_check_t. 386 */ 387int spa_slop_shift = 5; 388SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_slop_shift, CTLFLAG_RWTUN, 389 &spa_slop_shift, 0, 390 "Shift value of reserved space (1/(2^spa_slop_shift))."); 391 392/* 393 * ========================================================================== 394 * SPA config locking 395 * ========================================================================== 396 */ 397static void 398spa_config_lock_init(spa_t *spa) 399{ 400 for (int i = 0; i < SCL_LOCKS; i++) { 401 spa_config_lock_t *scl = &spa->spa_config_lock[i]; 402 mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL); 403 cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL); 404 refcount_create_untracked(&scl->scl_count); 405 scl->scl_writer = NULL; 406 scl->scl_write_wanted = 0; 407 } 408} 409 410static void 411spa_config_lock_destroy(spa_t *spa) 412{ 413 for (int i = 0; i < SCL_LOCKS; i++) { 414 spa_config_lock_t *scl = &spa->spa_config_lock[i]; 415 mutex_destroy(&scl->scl_lock); 416 cv_destroy(&scl->scl_cv); 417 refcount_destroy(&scl->scl_count); 418 ASSERT(scl->scl_writer == NULL); 419 ASSERT(scl->scl_write_wanted == 0); 420 } 421} 422 423int 424spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw) 425{ 426 for (int i = 0; i < SCL_LOCKS; i++) { 427 spa_config_lock_t *scl = &spa->spa_config_lock[i]; 428 if (!(locks & (1 << i))) 429 continue; 430 mutex_enter(&scl->scl_lock); 431 if (rw == RW_READER) { 432 if (scl->scl_writer || scl->scl_write_wanted) { 433 mutex_exit(&scl->scl_lock); 434 spa_config_exit(spa, locks ^ (1 << i), tag); 435 return (0); 436 } 437 } else { 438 ASSERT(scl->scl_writer != curthread); 439 if (!refcount_is_zero(&scl->scl_count)) { 440 mutex_exit(&scl->scl_lock); 441 spa_config_exit(spa, locks ^ (1 << i), tag); 442 return (0); 443 } 444 scl->scl_writer = curthread; 445 } 446 (void) refcount_add(&scl->scl_count, tag); 447 mutex_exit(&scl->scl_lock); 448 } 449 return (1); 450} 451 452void 453spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw) 454{ 455 int wlocks_held = 0; 456 457 ASSERT3U(SCL_LOCKS, <, sizeof (wlocks_held) * NBBY); 458 459 for (int i = 0; i < SCL_LOCKS; i++) { 460 spa_config_lock_t *scl = &spa->spa_config_lock[i]; 461 if (scl->scl_writer == curthread) 462 wlocks_held |= (1 << i); 463 if (!(locks & (1 << i))) 464 continue; 465 mutex_enter(&scl->scl_lock); 466 if (rw == RW_READER) { 467 while (scl->scl_writer || scl->scl_write_wanted) { 468 cv_wait(&scl->scl_cv, &scl->scl_lock); 469 } 470 } else { 471 ASSERT(scl->scl_writer != curthread); 472 while (!refcount_is_zero(&scl->scl_count)) { 473 scl->scl_write_wanted++; 474 cv_wait(&scl->scl_cv, &scl->scl_lock); 475 scl->scl_write_wanted--; 476 } 477 scl->scl_writer = curthread; 478 } 479 (void) refcount_add(&scl->scl_count, tag); 480 mutex_exit(&scl->scl_lock); 481 } 482 ASSERT(wlocks_held <= locks); 483} 484 485void 486spa_config_exit(spa_t *spa, int locks, void *tag) 487{ 488 for (int i = SCL_LOCKS - 1; i >= 0; i--) { 489 spa_config_lock_t *scl = &spa->spa_config_lock[i]; 490 if (!(locks & (1 << i))) 491 continue; 492 mutex_enter(&scl->scl_lock); 493 ASSERT(!refcount_is_zero(&scl->scl_count)); 494 if (refcount_remove(&scl->scl_count, tag) == 0) { 495 ASSERT(scl->scl_writer == NULL || 496 scl->scl_writer == curthread); 497 scl->scl_writer = NULL; /* OK in either case */ 498 cv_broadcast(&scl->scl_cv); 499 } 500 mutex_exit(&scl->scl_lock); 501 } 502} 503 504int 505spa_config_held(spa_t *spa, int locks, krw_t rw) 506{ 507 int locks_held = 0; 508 509 for (int i = 0; i < SCL_LOCKS; i++) { 510 spa_config_lock_t *scl = &spa->spa_config_lock[i]; 511 if (!(locks & (1 << i))) 512 continue; 513 if ((rw == RW_READER && !refcount_is_zero(&scl->scl_count)) || 514 (rw == RW_WRITER && scl->scl_writer == curthread)) 515 locks_held |= 1 << i; 516 } 517 518 return (locks_held); 519} 520 521/* 522 * ========================================================================== 523 * SPA namespace functions 524 * ========================================================================== 525 */ 526 527/* 528 * Lookup the named spa_t in the AVL tree. The spa_namespace_lock must be held. 529 * Returns NULL if no matching spa_t is found. 530 */ 531spa_t * 532spa_lookup(const char *name) 533{ 534 static spa_t search; /* spa_t is large; don't allocate on stack */ 535 spa_t *spa; 536 avl_index_t where; 537 char *cp; 538 539 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 540 541 (void) strlcpy(search.spa_name, name, sizeof (search.spa_name)); 542 543 /* 544 * If it's a full dataset name, figure out the pool name and 545 * just use that. 546 */ 547 cp = strpbrk(search.spa_name, "/@#"); 548 if (cp != NULL) 549 *cp = '\0'; 550 551 spa = avl_find(&spa_namespace_avl, &search, &where); 552 553 return (spa); 554} 555 556/* 557 * Fires when spa_sync has not completed within zfs_deadman_synctime_ms. 558 * If the zfs_deadman_enabled flag is set then it inspects all vdev queues 559 * looking for potentially hung I/Os. 560 */ 561void 562spa_deadman(void *arg) 563{ 564 spa_t *spa = arg; 565 566 /* 567 * Disable the deadman timer if the pool is suspended. 568 */ 569 if (spa_suspended(spa)) { 570#ifdef illumos 571 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY)); 572#else 573 /* Nothing. just don't schedule any future callouts. */ 574#endif 575 return; 576 } 577 578 zfs_dbgmsg("slow spa_sync: started %llu seconds ago, calls %llu", 579 (gethrtime() - spa->spa_sync_starttime) / NANOSEC, 580 ++spa->spa_deadman_calls); 581 if (zfs_deadman_enabled) 582 vdev_deadman(spa->spa_root_vdev); 583} 584 585/* 586 * Create an uninitialized spa_t with the given name. Requires 587 * spa_namespace_lock. The caller must ensure that the spa_t doesn't already 588 * exist by calling spa_lookup() first. 589 */ 590spa_t * 591spa_add(const char *name, nvlist_t *config, const char *altroot) 592{ 593 spa_t *spa; 594 spa_config_dirent_t *dp; 595#ifdef illumos 596 cyc_handler_t hdlr; 597 cyc_time_t when; 598#endif 599 600 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 601 602 spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP); 603 604 mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL); 605 mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL); 606 mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL); 607 mutex_init(&spa->spa_evicting_os_lock, NULL, MUTEX_DEFAULT, NULL); 608 mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL); 609 mutex_init(&spa->spa_proc_lock, NULL, MUTEX_DEFAULT, NULL); 610 mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL); 611 mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL); 612 mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL); 613 mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL); 614 615 cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL); 616 cv_init(&spa->spa_evicting_os_cv, NULL, CV_DEFAULT, NULL); 617 cv_init(&spa->spa_proc_cv, NULL, CV_DEFAULT, NULL); 618 cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL); 619 cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL); 620 621 for (int t = 0; t < TXG_SIZE; t++) 622 bplist_create(&spa->spa_free_bplist[t]); 623 624 (void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name)); 625 spa->spa_state = POOL_STATE_UNINITIALIZED; 626 spa->spa_freeze_txg = UINT64_MAX; 627 spa->spa_final_txg = UINT64_MAX; 628 spa->spa_load_max_txg = UINT64_MAX; 629 spa->spa_proc = &p0; 630 spa->spa_proc_state = SPA_PROC_NONE; 631 632#ifdef illumos 633 hdlr.cyh_func = spa_deadman; 634 hdlr.cyh_arg = spa; 635 hdlr.cyh_level = CY_LOW_LEVEL; 636#endif 637 638 spa->spa_deadman_synctime = MSEC2NSEC(zfs_deadman_synctime_ms); 639 640#ifdef illumos 641 /* 642 * This determines how often we need to check for hung I/Os after 643 * the cyclic has already fired. Since checking for hung I/Os is 644 * an expensive operation we don't want to check too frequently. 645 * Instead wait for 5 seconds before checking again. 646 */ 647 when.cyt_interval = MSEC2NSEC(zfs_deadman_checktime_ms); 648 when.cyt_when = CY_INFINITY; 649 mutex_enter(&cpu_lock); 650 spa->spa_deadman_cycid = cyclic_add(&hdlr, &when); 651 mutex_exit(&cpu_lock); 652#else /* !illumos */ 653#ifdef _KERNEL 654 callout_init(&spa->spa_deadman_cycid, CALLOUT_MPSAFE); 655#endif 656#endif 657 refcount_create(&spa->spa_refcount); 658 spa_config_lock_init(spa); 659 660 avl_add(&spa_namespace_avl, spa); 661 662 /* 663 * Set the alternate root, if there is one. 664 */ 665 if (altroot) { 666 spa->spa_root = spa_strdup(altroot); 667 spa_active_count++; 668 } 669 670 /* 671 * Every pool starts with the default cachefile 672 */ 673 list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t), 674 offsetof(spa_config_dirent_t, scd_link)); 675 676 dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP); 677 dp->scd_path = altroot ? NULL : spa_strdup(spa_config_path); 678 list_insert_head(&spa->spa_config_list, dp); 679 680 VERIFY(nvlist_alloc(&spa->spa_load_info, NV_UNIQUE_NAME, 681 KM_SLEEP) == 0); 682 683 if (config != NULL) { 684 nvlist_t *features; 685 686 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ, 687 &features) == 0) { 688 VERIFY(nvlist_dup(features, &spa->spa_label_features, 689 0) == 0); 690 } 691 692 VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0); 693 } 694 695 if (spa->spa_label_features == NULL) { 696 VERIFY(nvlist_alloc(&spa->spa_label_features, NV_UNIQUE_NAME, 697 KM_SLEEP) == 0); 698 } 699 700 spa->spa_debug = ((zfs_flags & ZFS_DEBUG_SPA) != 0); 701 702 spa->spa_min_ashift = INT_MAX; 703 spa->spa_max_ashift = 0; 704 705 /* 706 * As a pool is being created, treat all features as disabled by 707 * setting SPA_FEATURE_DISABLED for all entries in the feature 708 * refcount cache. 709 */ 710 for (int i = 0; i < SPA_FEATURES; i++) { 711 spa->spa_feat_refcount_cache[i] = SPA_FEATURE_DISABLED; 712 } 713 714 return (spa); 715} 716 717/* 718 * Removes a spa_t from the namespace, freeing up any memory used. Requires 719 * spa_namespace_lock. This is called only after the spa_t has been closed and 720 * deactivated. 721 */ 722void 723spa_remove(spa_t *spa) 724{ 725 spa_config_dirent_t *dp; 726 727 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 728 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); 729 ASSERT3U(refcount_count(&spa->spa_refcount), ==, 0); 730 731 nvlist_free(spa->spa_config_splitting); 732 733 avl_remove(&spa_namespace_avl, spa); 734 cv_broadcast(&spa_namespace_cv); 735 736 if (spa->spa_root) { 737 spa_strfree(spa->spa_root); 738 spa_active_count--; 739 } 740 741 while ((dp = list_head(&spa->spa_config_list)) != NULL) { 742 list_remove(&spa->spa_config_list, dp); 743 if (dp->scd_path != NULL) 744 spa_strfree(dp->scd_path); 745 kmem_free(dp, sizeof (spa_config_dirent_t)); 746 } 747 748 list_destroy(&spa->spa_config_list); 749 750 nvlist_free(spa->spa_label_features); 751 nvlist_free(spa->spa_load_info); 752 spa_config_set(spa, NULL); 753 754#ifdef illumos 755 mutex_enter(&cpu_lock); 756 if (spa->spa_deadman_cycid != CYCLIC_NONE) 757 cyclic_remove(spa->spa_deadman_cycid); 758 mutex_exit(&cpu_lock); 759 spa->spa_deadman_cycid = CYCLIC_NONE; 760#else /* !illumos */ 761#ifdef _KERNEL 762 callout_drain(&spa->spa_deadman_cycid); 763#endif 764#endif 765 766 refcount_destroy(&spa->spa_refcount); 767 768 spa_config_lock_destroy(spa); 769 770 for (int t = 0; t < TXG_SIZE; t++) 771 bplist_destroy(&spa->spa_free_bplist[t]); 772 773 cv_destroy(&spa->spa_async_cv); 774 cv_destroy(&spa->spa_evicting_os_cv); 775 cv_destroy(&spa->spa_proc_cv); 776 cv_destroy(&spa->spa_scrub_io_cv); 777 cv_destroy(&spa->spa_suspend_cv); 778 779 mutex_destroy(&spa->spa_async_lock); 780 mutex_destroy(&spa->spa_errlist_lock); 781 mutex_destroy(&spa->spa_errlog_lock); 782 mutex_destroy(&spa->spa_evicting_os_lock); 783 mutex_destroy(&spa->spa_history_lock); 784 mutex_destroy(&spa->spa_proc_lock); 785 mutex_destroy(&spa->spa_props_lock); 786 mutex_destroy(&spa->spa_scrub_lock); 787 mutex_destroy(&spa->spa_suspend_lock); 788 mutex_destroy(&spa->spa_vdev_top_lock); 789 790 kmem_free(spa, sizeof (spa_t)); 791} 792 793/* 794 * Given a pool, return the next pool in the namespace, or NULL if there is 795 * none. If 'prev' is NULL, return the first pool. 796 */ 797spa_t * 798spa_next(spa_t *prev) 799{ 800 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 801 802 if (prev) 803 return (AVL_NEXT(&spa_namespace_avl, prev)); 804 else 805 return (avl_first(&spa_namespace_avl)); 806} 807 808/* 809 * ========================================================================== 810 * SPA refcount functions 811 * ========================================================================== 812 */ 813 814/* 815 * Add a reference to the given spa_t. Must have at least one reference, or 816 * have the namespace lock held. 817 */ 818void 819spa_open_ref(spa_t *spa, void *tag) 820{ 821 ASSERT(refcount_count(&spa->spa_refcount) >= spa->spa_minref || 822 MUTEX_HELD(&spa_namespace_lock)); 823 (void) refcount_add(&spa->spa_refcount, tag); 824} 825 826/* 827 * Remove a reference to the given spa_t. Must have at least one reference, or 828 * have the namespace lock held. 829 */ 830void 831spa_close(spa_t *spa, void *tag) 832{ 833 ASSERT(refcount_count(&spa->spa_refcount) > spa->spa_minref || 834 MUTEX_HELD(&spa_namespace_lock)); 835 (void) refcount_remove(&spa->spa_refcount, tag); 836} 837 838/* 839 * Remove a reference to the given spa_t held by a dsl dir that is 840 * being asynchronously released. Async releases occur from a taskq 841 * performing eviction of dsl datasets and dirs. The namespace lock 842 * isn't held and the hold by the object being evicted may contribute to 843 * spa_minref (e.g. dataset or directory released during pool export), 844 * so the asserts in spa_close() do not apply. 845 */ 846void 847spa_async_close(spa_t *spa, void *tag) 848{ 849 (void) refcount_remove(&spa->spa_refcount, tag); 850} 851 852/* 853 * Check to see if the spa refcount is zero. Must be called with 854 * spa_namespace_lock held. We really compare against spa_minref, which is the 855 * number of references acquired when opening a pool 856 */ 857boolean_t 858spa_refcount_zero(spa_t *spa) 859{ 860 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 861 862 return (refcount_count(&spa->spa_refcount) == spa->spa_minref); 863} 864 865/* 866 * ========================================================================== 867 * SPA spare and l2cache tracking 868 * ========================================================================== 869 */ 870 871/* 872 * Hot spares and cache devices are tracked using the same code below, 873 * for 'auxiliary' devices. 874 */ 875 876typedef struct spa_aux { 877 uint64_t aux_guid; 878 uint64_t aux_pool; 879 avl_node_t aux_avl; 880 int aux_count; 881} spa_aux_t; 882 883static int 884spa_aux_compare(const void *a, const void *b) 885{ 886 const spa_aux_t *sa = a; 887 const spa_aux_t *sb = b; 888 889 if (sa->aux_guid < sb->aux_guid) 890 return (-1); 891 else if (sa->aux_guid > sb->aux_guid) 892 return (1); 893 else 894 return (0); 895} 896 897void 898spa_aux_add(vdev_t *vd, avl_tree_t *avl) 899{ 900 avl_index_t where; 901 spa_aux_t search; 902 spa_aux_t *aux; 903 904 search.aux_guid = vd->vdev_guid; 905 if ((aux = avl_find(avl, &search, &where)) != NULL) { 906 aux->aux_count++; 907 } else { 908 aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP); 909 aux->aux_guid = vd->vdev_guid; 910 aux->aux_count = 1; 911 avl_insert(avl, aux, where); 912 } 913} 914 915void 916spa_aux_remove(vdev_t *vd, avl_tree_t *avl) 917{ 918 spa_aux_t search; 919 spa_aux_t *aux; 920 avl_index_t where; 921 922 search.aux_guid = vd->vdev_guid; 923 aux = avl_find(avl, &search, &where); 924 925 ASSERT(aux != NULL); 926 927 if (--aux->aux_count == 0) { 928 avl_remove(avl, aux); 929 kmem_free(aux, sizeof (spa_aux_t)); 930 } else if (aux->aux_pool == spa_guid(vd->vdev_spa)) { 931 aux->aux_pool = 0ULL; 932 } 933} 934 935boolean_t 936spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl) 937{ 938 spa_aux_t search, *found; 939 940 search.aux_guid = guid; 941 found = avl_find(avl, &search, NULL); 942 943 if (pool) { 944 if (found) 945 *pool = found->aux_pool; 946 else 947 *pool = 0ULL; 948 } 949 950 if (refcnt) { 951 if (found) 952 *refcnt = found->aux_count; 953 else 954 *refcnt = 0; 955 } 956 957 return (found != NULL); 958} 959 960void 961spa_aux_activate(vdev_t *vd, avl_tree_t *avl) 962{ 963 spa_aux_t search, *found; 964 avl_index_t where; 965 966 search.aux_guid = vd->vdev_guid; 967 found = avl_find(avl, &search, &where); 968 ASSERT(found != NULL); 969 ASSERT(found->aux_pool == 0ULL); 970 971 found->aux_pool = spa_guid(vd->vdev_spa); 972} 973 974/* 975 * Spares are tracked globally due to the following constraints: 976 * 977 * - A spare may be part of multiple pools. 978 * - A spare may be added to a pool even if it's actively in use within 979 * another pool. 980 * - A spare in use in any pool can only be the source of a replacement if 981 * the target is a spare in the same pool. 982 * 983 * We keep track of all spares on the system through the use of a reference 984 * counted AVL tree. When a vdev is added as a spare, or used as a replacement 985 * spare, then we bump the reference count in the AVL tree. In addition, we set 986 * the 'vdev_isspare' member to indicate that the device is a spare (active or 987 * inactive). When a spare is made active (used to replace a device in the 988 * pool), we also keep track of which pool its been made a part of. 989 * 990 * The 'spa_spare_lock' protects the AVL tree. These functions are normally 991 * called under the spa_namespace lock as part of vdev reconfiguration. The 992 * separate spare lock exists for the status query path, which does not need to 993 * be completely consistent with respect to other vdev configuration changes. 994 */ 995 996static int 997spa_spare_compare(const void *a, const void *b) 998{ 999 return (spa_aux_compare(a, b)); 1000} 1001 1002void 1003spa_spare_add(vdev_t *vd) 1004{ 1005 mutex_enter(&spa_spare_lock); 1006 ASSERT(!vd->vdev_isspare); 1007 spa_aux_add(vd, &spa_spare_avl); 1008 vd->vdev_isspare = B_TRUE; 1009 mutex_exit(&spa_spare_lock); 1010} 1011 1012void 1013spa_spare_remove(vdev_t *vd) 1014{ 1015 mutex_enter(&spa_spare_lock); 1016 ASSERT(vd->vdev_isspare); 1017 spa_aux_remove(vd, &spa_spare_avl); 1018 vd->vdev_isspare = B_FALSE; 1019 mutex_exit(&spa_spare_lock); 1020} 1021 1022boolean_t 1023spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt) 1024{ 1025 boolean_t found; 1026 1027 mutex_enter(&spa_spare_lock); 1028 found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl); 1029 mutex_exit(&spa_spare_lock); 1030 1031 return (found); 1032} 1033 1034void 1035spa_spare_activate(vdev_t *vd) 1036{ 1037 mutex_enter(&spa_spare_lock); 1038 ASSERT(vd->vdev_isspare); 1039 spa_aux_activate(vd, &spa_spare_avl); 1040 mutex_exit(&spa_spare_lock); 1041} 1042 1043/* 1044 * Level 2 ARC devices are tracked globally for the same reasons as spares. 1045 * Cache devices currently only support one pool per cache device, and so 1046 * for these devices the aux reference count is currently unused beyond 1. 1047 */ 1048 1049static int 1050spa_l2cache_compare(const void *a, const void *b) 1051{ 1052 return (spa_aux_compare(a, b)); 1053} 1054 1055void 1056spa_l2cache_add(vdev_t *vd) 1057{ 1058 mutex_enter(&spa_l2cache_lock); 1059 ASSERT(!vd->vdev_isl2cache); 1060 spa_aux_add(vd, &spa_l2cache_avl); 1061 vd->vdev_isl2cache = B_TRUE; 1062 mutex_exit(&spa_l2cache_lock); 1063} 1064 1065void 1066spa_l2cache_remove(vdev_t *vd) 1067{ 1068 mutex_enter(&spa_l2cache_lock); 1069 ASSERT(vd->vdev_isl2cache); 1070 spa_aux_remove(vd, &spa_l2cache_avl); 1071 vd->vdev_isl2cache = B_FALSE; 1072 mutex_exit(&spa_l2cache_lock); 1073} 1074 1075boolean_t 1076spa_l2cache_exists(uint64_t guid, uint64_t *pool) 1077{ 1078 boolean_t found; 1079 1080 mutex_enter(&spa_l2cache_lock); 1081 found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl); 1082 mutex_exit(&spa_l2cache_lock); 1083 1084 return (found); 1085} 1086 1087void 1088spa_l2cache_activate(vdev_t *vd) 1089{ 1090 mutex_enter(&spa_l2cache_lock); 1091 ASSERT(vd->vdev_isl2cache); 1092 spa_aux_activate(vd, &spa_l2cache_avl); 1093 mutex_exit(&spa_l2cache_lock); 1094} 1095 1096/* 1097 * ========================================================================== 1098 * SPA vdev locking 1099 * ========================================================================== 1100 */ 1101 1102/* 1103 * Lock the given spa_t for the purpose of adding or removing a vdev. 1104 * Grabs the global spa_namespace_lock plus the spa config lock for writing. 1105 * It returns the next transaction group for the spa_t. 1106 */ 1107uint64_t 1108spa_vdev_enter(spa_t *spa) 1109{ 1110 mutex_enter(&spa->spa_vdev_top_lock); 1111 mutex_enter(&spa_namespace_lock); 1112 return (spa_vdev_config_enter(spa)); 1113} 1114 1115/* 1116 * Internal implementation for spa_vdev_enter(). Used when a vdev 1117 * operation requires multiple syncs (i.e. removing a device) while 1118 * keeping the spa_namespace_lock held. 1119 */ 1120uint64_t 1121spa_vdev_config_enter(spa_t *spa) 1122{ 1123 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1124 1125 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER); 1126 1127 return (spa_last_synced_txg(spa) + 1); 1128} 1129 1130/* 1131 * Used in combination with spa_vdev_config_enter() to allow the syncing 1132 * of multiple transactions without releasing the spa_namespace_lock. 1133 */ 1134void 1135spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag) 1136{ 1137 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1138 1139 int config_changed = B_FALSE; 1140 1141 ASSERT(txg > spa_last_synced_txg(spa)); 1142 1143 spa->spa_pending_vdev = NULL; 1144 1145 /* 1146 * Reassess the DTLs. 1147 */ 1148 vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE); 1149 1150 if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) { 1151 config_changed = B_TRUE; 1152 spa->spa_config_generation++; 1153 } 1154 1155 /* 1156 * Verify the metaslab classes. 1157 */ 1158 ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0); 1159 ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0); 1160 1161 spa_config_exit(spa, SCL_ALL, spa); 1162 1163 /* 1164 * Panic the system if the specified tag requires it. This 1165 * is useful for ensuring that configurations are updated 1166 * transactionally. 1167 */ 1168 if (zio_injection_enabled) 1169 zio_handle_panic_injection(spa, tag, 0); 1170 1171 /* 1172 * Note: this txg_wait_synced() is important because it ensures 1173 * that there won't be more than one config change per txg. 1174 * This allows us to use the txg as the generation number. 1175 */ 1176 if (error == 0) 1177 txg_wait_synced(spa->spa_dsl_pool, txg); 1178 1179 if (vd != NULL) { 1180 ASSERT(!vd->vdev_detached || vd->vdev_dtl_sm == NULL); 1181 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER); 1182 vdev_free(vd); 1183 spa_config_exit(spa, SCL_ALL, spa); 1184 } 1185 1186 /* 1187 * If the config changed, update the config cache. 1188 */ 1189 if (config_changed) 1190 spa_config_sync(spa, B_FALSE, B_TRUE); 1191} 1192 1193/* 1194 * Unlock the spa_t after adding or removing a vdev. Besides undoing the 1195 * locking of spa_vdev_enter(), we also want make sure the transactions have 1196 * synced to disk, and then update the global configuration cache with the new 1197 * information. 1198 */ 1199int 1200spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error) 1201{ 1202 spa_vdev_config_exit(spa, vd, txg, error, FTAG); 1203 mutex_exit(&spa_namespace_lock); 1204 mutex_exit(&spa->spa_vdev_top_lock); 1205 1206 return (error); 1207} 1208 1209/* 1210 * Lock the given spa_t for the purpose of changing vdev state. 1211 */ 1212void 1213spa_vdev_state_enter(spa_t *spa, int oplocks) 1214{ 1215 int locks = SCL_STATE_ALL | oplocks; 1216 1217 /* 1218 * Root pools may need to read of the underlying devfs filesystem 1219 * when opening up a vdev. Unfortunately if we're holding the 1220 * SCL_ZIO lock it will result in a deadlock when we try to issue 1221 * the read from the root filesystem. Instead we "prefetch" 1222 * the associated vnodes that we need prior to opening the 1223 * underlying devices and cache them so that we can prevent 1224 * any I/O when we are doing the actual open. 1225 */ 1226 if (spa_is_root(spa)) { 1227 int low = locks & ~(SCL_ZIO - 1); 1228 int high = locks & ~low; 1229 1230 spa_config_enter(spa, high, spa, RW_WRITER); 1231 vdev_hold(spa->spa_root_vdev); 1232 spa_config_enter(spa, low, spa, RW_WRITER); 1233 } else { 1234 spa_config_enter(spa, locks, spa, RW_WRITER); 1235 } 1236 spa->spa_vdev_locks = locks; 1237} 1238 1239int 1240spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error) 1241{ 1242 boolean_t config_changed = B_FALSE; 1243 1244 if (vd != NULL || error == 0) 1245 vdev_dtl_reassess(vd ? vd->vdev_top : spa->spa_root_vdev, 1246 0, 0, B_FALSE); 1247 1248 if (vd != NULL) { 1249 vdev_state_dirty(vd->vdev_top); 1250 config_changed = B_TRUE; 1251 spa->spa_config_generation++; 1252 } 1253 1254 if (spa_is_root(spa)) 1255 vdev_rele(spa->spa_root_vdev); 1256 1257 ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL); 1258 spa_config_exit(spa, spa->spa_vdev_locks, spa); 1259 1260 /* 1261 * If anything changed, wait for it to sync. This ensures that, 1262 * from the system administrator's perspective, zpool(1M) commands 1263 * are synchronous. This is important for things like zpool offline: 1264 * when the command completes, you expect no further I/O from ZFS. 1265 */ 1266 if (vd != NULL) 1267 txg_wait_synced(spa->spa_dsl_pool, 0); 1268 1269 /* 1270 * If the config changed, update the config cache. 1271 */ 1272 if (config_changed) { 1273 mutex_enter(&spa_namespace_lock); 1274 spa_config_sync(spa, B_FALSE, B_TRUE); 1275 mutex_exit(&spa_namespace_lock); 1276 } 1277 1278 return (error); 1279} 1280 1281/* 1282 * ========================================================================== 1283 * Miscellaneous functions 1284 * ========================================================================== 1285 */ 1286 1287void 1288spa_activate_mos_feature(spa_t *spa, const char *feature, dmu_tx_t *tx) 1289{ 1290 if (!nvlist_exists(spa->spa_label_features, feature)) { 1291 fnvlist_add_boolean(spa->spa_label_features, feature); 1292 /* 1293 * When we are creating the pool (tx_txg==TXG_INITIAL), we can't 1294 * dirty the vdev config because lock SCL_CONFIG is not held. 1295 * Thankfully, in this case we don't need to dirty the config 1296 * because it will be written out anyway when we finish 1297 * creating the pool. 1298 */ 1299 if (tx->tx_txg != TXG_INITIAL) 1300 vdev_config_dirty(spa->spa_root_vdev); 1301 } 1302} 1303 1304void 1305spa_deactivate_mos_feature(spa_t *spa, const char *feature) 1306{ 1307 if (nvlist_remove_all(spa->spa_label_features, feature) == 0) 1308 vdev_config_dirty(spa->spa_root_vdev); 1309} 1310 1311/* 1312 * Rename a spa_t. 1313 */ 1314int 1315spa_rename(const char *name, const char *newname) 1316{ 1317 spa_t *spa; 1318 int err; 1319 1320 /* 1321 * Lookup the spa_t and grab the config lock for writing. We need to 1322 * actually open the pool so that we can sync out the necessary labels. 1323 * It's OK to call spa_open() with the namespace lock held because we 1324 * allow recursive calls for other reasons. 1325 */ 1326 mutex_enter(&spa_namespace_lock); 1327 if ((err = spa_open(name, &spa, FTAG)) != 0) { 1328 mutex_exit(&spa_namespace_lock); 1329 return (err); 1330 } 1331 1332 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1333 1334 avl_remove(&spa_namespace_avl, spa); 1335 (void) strlcpy(spa->spa_name, newname, sizeof (spa->spa_name)); 1336 avl_add(&spa_namespace_avl, spa); 1337 1338 /* 1339 * Sync all labels to disk with the new names by marking the root vdev 1340 * dirty and waiting for it to sync. It will pick up the new pool name 1341 * during the sync. 1342 */ 1343 vdev_config_dirty(spa->spa_root_vdev); 1344 1345 spa_config_exit(spa, SCL_ALL, FTAG); 1346 1347 txg_wait_synced(spa->spa_dsl_pool, 0); 1348 1349 /* 1350 * Sync the updated config cache. 1351 */ 1352 spa_config_sync(spa, B_FALSE, B_TRUE); 1353 1354 spa_close(spa, FTAG); 1355 1356 mutex_exit(&spa_namespace_lock); 1357 1358 return (0); 1359} 1360 1361/* 1362 * Return the spa_t associated with given pool_guid, if it exists. If 1363 * device_guid is non-zero, determine whether the pool exists *and* contains 1364 * a device with the specified device_guid. 1365 */ 1366spa_t * 1367spa_by_guid(uint64_t pool_guid, uint64_t device_guid) 1368{ 1369 spa_t *spa; 1370 avl_tree_t *t = &spa_namespace_avl; 1371 1372 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1373 1374 for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) { 1375 if (spa->spa_state == POOL_STATE_UNINITIALIZED) 1376 continue; 1377 if (spa->spa_root_vdev == NULL) 1378 continue; 1379 if (spa_guid(spa) == pool_guid) { 1380 if (device_guid == 0) 1381 break; 1382 1383 if (vdev_lookup_by_guid(spa->spa_root_vdev, 1384 device_guid) != NULL) 1385 break; 1386 1387 /* 1388 * Check any devices we may be in the process of adding. 1389 */ 1390 if (spa->spa_pending_vdev) { 1391 if (vdev_lookup_by_guid(spa->spa_pending_vdev, 1392 device_guid) != NULL) 1393 break; 1394 } 1395 } 1396 } 1397 1398 return (spa); 1399} 1400 1401/* 1402 * Determine whether a pool with the given pool_guid exists. 1403 */ 1404boolean_t 1405spa_guid_exists(uint64_t pool_guid, uint64_t device_guid) 1406{ 1407 return (spa_by_guid(pool_guid, device_guid) != NULL); 1408} 1409 1410char * 1411spa_strdup(const char *s) 1412{ 1413 size_t len; 1414 char *new; 1415 1416 len = strlen(s); 1417 new = kmem_alloc(len + 1, KM_SLEEP); 1418 bcopy(s, new, len); 1419 new[len] = '\0'; 1420 1421 return (new); 1422} 1423 1424void 1425spa_strfree(char *s) 1426{ 1427 kmem_free(s, strlen(s) + 1); 1428} 1429 1430uint64_t 1431spa_get_random(uint64_t range) 1432{ 1433 uint64_t r; 1434 1435 ASSERT(range != 0); 1436 1437 (void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t)); 1438 1439 return (r % range); 1440} 1441 1442uint64_t 1443spa_generate_guid(spa_t *spa) 1444{ 1445 uint64_t guid = spa_get_random(-1ULL); 1446 1447 if (spa != NULL) { 1448 while (guid == 0 || spa_guid_exists(spa_guid(spa), guid)) 1449 guid = spa_get_random(-1ULL); 1450 } else { 1451 while (guid == 0 || spa_guid_exists(guid, 0)) 1452 guid = spa_get_random(-1ULL); 1453 } 1454 1455 return (guid); 1456} 1457 1458void 1459snprintf_blkptr(char *buf, size_t buflen, const blkptr_t *bp) 1460{ 1461 char type[256]; 1462 char *checksum = NULL; 1463 char *compress = NULL; 1464 1465 if (bp != NULL) { 1466 if (BP_GET_TYPE(bp) & DMU_OT_NEWTYPE) { 1467 dmu_object_byteswap_t bswap = 1468 DMU_OT_BYTESWAP(BP_GET_TYPE(bp)); 1469 (void) snprintf(type, sizeof (type), "bswap %s %s", 1470 DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) ? 1471 "metadata" : "data", 1472 dmu_ot_byteswap[bswap].ob_name); 1473 } else { 1474 (void) strlcpy(type, dmu_ot[BP_GET_TYPE(bp)].ot_name, 1475 sizeof (type)); 1476 } 1477 if (!BP_IS_EMBEDDED(bp)) { 1478 checksum = 1479 zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name; 1480 } 1481 compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name; 1482 } 1483 1484 SNPRINTF_BLKPTR(snprintf, ' ', buf, buflen, bp, type, checksum, 1485 compress); 1486} 1487 1488void 1489spa_freeze(spa_t *spa) 1490{ 1491 uint64_t freeze_txg = 0; 1492 1493 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1494 if (spa->spa_freeze_txg == UINT64_MAX) { 1495 freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE; 1496 spa->spa_freeze_txg = freeze_txg; 1497 } 1498 spa_config_exit(spa, SCL_ALL, FTAG); 1499 if (freeze_txg != 0) 1500 txg_wait_synced(spa_get_dsl(spa), freeze_txg); 1501} 1502 1503void 1504zfs_panic_recover(const char *fmt, ...) 1505{ 1506 va_list adx; 1507 1508 va_start(adx, fmt); 1509 vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx); 1510 va_end(adx); 1511} 1512 1513/* 1514 * This is a stripped-down version of strtoull, suitable only for converting 1515 * lowercase hexadecimal numbers that don't overflow. 1516 */ 1517uint64_t 1518zfs_strtonum(const char *str, char **nptr) 1519{ 1520 uint64_t val = 0; 1521 char c; 1522 int digit; 1523 1524 while ((c = *str) != '\0') { 1525 if (c >= '0' && c <= '9') 1526 digit = c - '0'; 1527 else if (c >= 'a' && c <= 'f') 1528 digit = 10 + c - 'a'; 1529 else 1530 break; 1531 1532 val *= 16; 1533 val += digit; 1534 1535 str++; 1536 } 1537 1538 if (nptr) 1539 *nptr = (char *)str; 1540 1541 return (val); 1542} 1543 1544/* 1545 * ========================================================================== 1546 * Accessor functions 1547 * ========================================================================== 1548 */ 1549 1550boolean_t 1551spa_shutting_down(spa_t *spa) 1552{ 1553 return (spa->spa_async_suspended); 1554} 1555 1556dsl_pool_t * 1557spa_get_dsl(spa_t *spa) 1558{ 1559 return (spa->spa_dsl_pool); 1560} 1561 1562boolean_t 1563spa_is_initializing(spa_t *spa) 1564{ 1565 return (spa->spa_is_initializing); 1566} 1567 1568blkptr_t * 1569spa_get_rootblkptr(spa_t *spa) 1570{ 1571 return (&spa->spa_ubsync.ub_rootbp); 1572} 1573 1574void 1575spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp) 1576{ 1577 spa->spa_uberblock.ub_rootbp = *bp; 1578} 1579 1580void 1581spa_altroot(spa_t *spa, char *buf, size_t buflen) 1582{ 1583 if (spa->spa_root == NULL) 1584 buf[0] = '\0'; 1585 else 1586 (void) strncpy(buf, spa->spa_root, buflen); 1587} 1588 1589int 1590spa_sync_pass(spa_t *spa) 1591{ 1592 return (spa->spa_sync_pass); 1593} 1594 1595char * 1596spa_name(spa_t *spa) 1597{ 1598 return (spa->spa_name); 1599} 1600 1601uint64_t 1602spa_guid(spa_t *spa) 1603{ 1604 dsl_pool_t *dp = spa_get_dsl(spa); 1605 uint64_t guid; 1606 1607 /* 1608 * If we fail to parse the config during spa_load(), we can go through 1609 * the error path (which posts an ereport) and end up here with no root 1610 * vdev. We stash the original pool guid in 'spa_config_guid' to handle 1611 * this case. 1612 */ 1613 if (spa->spa_root_vdev == NULL) 1614 return (spa->spa_config_guid); 1615 1616 guid = spa->spa_last_synced_guid != 0 ? 1617 spa->spa_last_synced_guid : spa->spa_root_vdev->vdev_guid; 1618 1619 /* 1620 * Return the most recently synced out guid unless we're 1621 * in syncing context. 1622 */ 1623 if (dp && dsl_pool_sync_context(dp)) 1624 return (spa->spa_root_vdev->vdev_guid); 1625 else 1626 return (guid); 1627} 1628 1629uint64_t 1630spa_load_guid(spa_t *spa) 1631{ 1632 /* 1633 * This is a GUID that exists solely as a reference for the 1634 * purposes of the arc. It is generated at load time, and 1635 * is never written to persistent storage. 1636 */ 1637 return (spa->spa_load_guid); 1638} 1639 1640uint64_t 1641spa_last_synced_txg(spa_t *spa) 1642{ 1643 return (spa->spa_ubsync.ub_txg); 1644} 1645 1646uint64_t 1647spa_first_txg(spa_t *spa) 1648{ 1649 return (spa->spa_first_txg); 1650} 1651 1652uint64_t 1653spa_syncing_txg(spa_t *spa) 1654{ 1655 return (spa->spa_syncing_txg); 1656} 1657 1658pool_state_t 1659spa_state(spa_t *spa) 1660{ 1661 return (spa->spa_state); 1662} 1663 1664spa_load_state_t 1665spa_load_state(spa_t *spa) 1666{ 1667 return (spa->spa_load_state); 1668} 1669 1670uint64_t 1671spa_freeze_txg(spa_t *spa) 1672{ 1673 return (spa->spa_freeze_txg); 1674} 1675 1676/* ARGSUSED */ 1677uint64_t 1678spa_get_asize(spa_t *spa, uint64_t lsize) 1679{ 1680 return (lsize * spa_asize_inflation); 1681} 1682 1683/* 1684 * Return the amount of slop space in bytes. It is 1/32 of the pool (3.2%), 1685 * or at least 32MB. 1686 * 1687 * See the comment above spa_slop_shift for details. 1688 */ 1689uint64_t 1690spa_get_slop_space(spa_t *spa) { 1691 uint64_t space = spa_get_dspace(spa); 1692 return (MAX(space >> spa_slop_shift, SPA_MINDEVSIZE >> 1)); 1693} 1694 1695uint64_t 1696spa_get_dspace(spa_t *spa) 1697{ 1698 return (spa->spa_dspace); 1699} 1700 1701void 1702spa_update_dspace(spa_t *spa) 1703{ 1704 spa->spa_dspace = metaslab_class_get_dspace(spa_normal_class(spa)) + 1705 ddt_get_dedup_dspace(spa); 1706} 1707 1708/* 1709 * Return the failure mode that has been set to this pool. The default 1710 * behavior will be to block all I/Os when a complete failure occurs. 1711 */ 1712uint8_t 1713spa_get_failmode(spa_t *spa) 1714{ 1715 return (spa->spa_failmode); 1716} 1717 1718boolean_t 1719spa_suspended(spa_t *spa) 1720{ 1721 return (spa->spa_suspended); 1722} 1723 1724uint64_t 1725spa_version(spa_t *spa) 1726{ 1727 return (spa->spa_ubsync.ub_version); 1728} 1729 1730boolean_t 1731spa_deflate(spa_t *spa) 1732{ 1733 return (spa->spa_deflate); 1734} 1735 1736metaslab_class_t * 1737spa_normal_class(spa_t *spa) 1738{ 1739 return (spa->spa_normal_class); 1740} 1741 1742metaslab_class_t * 1743spa_log_class(spa_t *spa) 1744{ 1745 return (spa->spa_log_class); 1746} 1747 1748void 1749spa_evicting_os_register(spa_t *spa, objset_t *os) 1750{ 1751 mutex_enter(&spa->spa_evicting_os_lock); 1752 list_insert_head(&spa->spa_evicting_os_list, os); 1753 mutex_exit(&spa->spa_evicting_os_lock); 1754} 1755 1756void 1757spa_evicting_os_deregister(spa_t *spa, objset_t *os) 1758{ 1759 mutex_enter(&spa->spa_evicting_os_lock); 1760 list_remove(&spa->spa_evicting_os_list, os); 1761 cv_broadcast(&spa->spa_evicting_os_cv); 1762 mutex_exit(&spa->spa_evicting_os_lock); 1763} 1764 1765void 1766spa_evicting_os_wait(spa_t *spa) 1767{ 1768 mutex_enter(&spa->spa_evicting_os_lock); 1769 while (!list_is_empty(&spa->spa_evicting_os_list)) 1770 cv_wait(&spa->spa_evicting_os_cv, &spa->spa_evicting_os_lock); 1771 mutex_exit(&spa->spa_evicting_os_lock); 1772 1773 dmu_buf_user_evict_wait(); 1774} 1775 1776int 1777spa_max_replication(spa_t *spa) 1778{ 1779 /* 1780 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to 1781 * handle BPs with more than one DVA allocated. Set our max 1782 * replication level accordingly. 1783 */ 1784 if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS) 1785 return (1); 1786 return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override)); 1787} 1788 1789int 1790spa_prev_software_version(spa_t *spa) 1791{ 1792 return (spa->spa_prev_software_version); 1793} 1794 1795uint64_t 1796spa_deadman_synctime(spa_t *spa) 1797{ 1798 return (spa->spa_deadman_synctime); 1799} 1800 1801uint64_t 1802dva_get_dsize_sync(spa_t *spa, const dva_t *dva) 1803{ 1804 uint64_t asize = DVA_GET_ASIZE(dva); 1805 uint64_t dsize = asize; 1806 1807 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 1808 1809 if (asize != 0 && spa->spa_deflate) { 1810 vdev_t *vd = vdev_lookup_top(spa, DVA_GET_VDEV(dva)); 1811 dsize = (asize >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio; 1812 } 1813 1814 return (dsize); 1815} 1816 1817uint64_t 1818bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp) 1819{ 1820 uint64_t dsize = 0; 1821 1822 for (int d = 0; d < BP_GET_NDVAS(bp); d++) 1823 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]); 1824 1825 return (dsize); 1826} 1827 1828uint64_t 1829bp_get_dsize(spa_t *spa, const blkptr_t *bp) 1830{ 1831 uint64_t dsize = 0; 1832 1833 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); 1834 1835 for (int d = 0; d < BP_GET_NDVAS(bp); d++) 1836 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]); 1837 1838 spa_config_exit(spa, SCL_VDEV, FTAG); 1839 1840 return (dsize); 1841} 1842 1843/* 1844 * ========================================================================== 1845 * Initialization and Termination 1846 * ========================================================================== 1847 */ 1848 1849static int 1850spa_name_compare(const void *a1, const void *a2) 1851{ 1852 const spa_t *s1 = a1; 1853 const spa_t *s2 = a2; 1854 int s; 1855 1856 s = strcmp(s1->spa_name, s2->spa_name); 1857 if (s > 0) 1858 return (1); 1859 if (s < 0) 1860 return (-1); 1861 return (0); 1862} 1863 1864int 1865spa_busy(void) 1866{ 1867 return (spa_active_count); 1868} 1869 1870void 1871spa_boot_init() 1872{ 1873 spa_config_load(); 1874} 1875 1876#ifdef _KERNEL 1877EVENTHANDLER_DEFINE(mountroot, spa_boot_init, NULL, 0); 1878#endif 1879 1880void 1881spa_init(int mode) 1882{ 1883 mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL); 1884 mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL); 1885 mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL); 1886 cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL); 1887 1888 avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t), 1889 offsetof(spa_t, spa_avl)); 1890 1891 avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t), 1892 offsetof(spa_aux_t, aux_avl)); 1893 1894 avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t), 1895 offsetof(spa_aux_t, aux_avl)); 1896 1897 spa_mode_global = mode; 1898 1899#ifdef illumos 1900#ifdef _KERNEL 1901 spa_arch_init(); 1902#else 1903 if (spa_mode_global != FREAD && dprintf_find_string("watch")) { 1904 arc_procfd = open("/proc/self/ctl", O_WRONLY); 1905 if (arc_procfd == -1) { 1906 perror("could not enable watchpoints: " 1907 "opening /proc/self/ctl failed: "); 1908 } else { 1909 arc_watch = B_TRUE; 1910 } 1911 } 1912#endif 1913#endif /* illumos */ 1914 refcount_sysinit(); 1915 unique_init(); 1916 range_tree_init(); 1917 zio_init(); 1918 lz4_init(); 1919 dmu_init(); 1920 zil_init(); 1921 vdev_cache_stat_init(); 1922 zfs_prop_init(); 1923 zpool_prop_init(); 1924 zpool_feature_init(); 1925 spa_config_load(); 1926 l2arc_start(); 1927#ifndef illumos 1928#ifdef _KERNEL 1929 zfs_deadman_init(); 1930#endif 1931#endif /* !illumos */ 1932} 1933 1934void 1935spa_fini(void) 1936{ 1937 l2arc_stop(); 1938 1939 spa_evict_all(); 1940 1941 vdev_cache_stat_fini(); 1942 zil_fini(); 1943 dmu_fini(); 1944 lz4_fini(); 1945 zio_fini(); 1946 range_tree_fini(); 1947 unique_fini(); 1948 refcount_fini(); 1949 1950 avl_destroy(&spa_namespace_avl); 1951 avl_destroy(&spa_spare_avl); 1952 avl_destroy(&spa_l2cache_avl); 1953 1954 cv_destroy(&spa_namespace_cv); 1955 mutex_destroy(&spa_namespace_lock); 1956 mutex_destroy(&spa_spare_lock); 1957 mutex_destroy(&spa_l2cache_lock); 1958} 1959 1960/* 1961 * Return whether this pool has slogs. No locking needed. 1962 * It's not a problem if the wrong answer is returned as it's only for 1963 * performance and not correctness 1964 */ 1965boolean_t 1966spa_has_slogs(spa_t *spa) 1967{ 1968 return (spa->spa_log_class->mc_rotor != NULL); 1969} 1970 1971spa_log_state_t 1972spa_get_log_state(spa_t *spa) 1973{ 1974 return (spa->spa_log_state); 1975} 1976 1977void 1978spa_set_log_state(spa_t *spa, spa_log_state_t state) 1979{ 1980 spa->spa_log_state = state; 1981} 1982 1983boolean_t 1984spa_is_root(spa_t *spa) 1985{ 1986 return (spa->spa_is_root); 1987} 1988 1989boolean_t 1990spa_writeable(spa_t *spa) 1991{ 1992 return (!!(spa->spa_mode & FWRITE)); 1993} 1994 1995/* 1996 * Returns true if there is a pending sync task in any of the current 1997 * syncing txg, the current quiescing txg, or the current open txg. 1998 */ 1999boolean_t 2000spa_has_pending_synctask(spa_t *spa) 2001{ 2002 return (!txg_all_lists_empty(&spa->spa_dsl_pool->dp_sync_tasks)); 2003} 2004 2005int 2006spa_mode(spa_t *spa) 2007{ 2008 return (spa->spa_mode); 2009} 2010 2011uint64_t 2012spa_bootfs(spa_t *spa) 2013{ 2014 return (spa->spa_bootfs); 2015} 2016 2017uint64_t 2018spa_delegation(spa_t *spa) 2019{ 2020 return (spa->spa_delegation); 2021} 2022 2023objset_t * 2024spa_meta_objset(spa_t *spa) 2025{ 2026 return (spa->spa_meta_objset); 2027} 2028 2029enum zio_checksum 2030spa_dedup_checksum(spa_t *spa) 2031{ 2032 return (spa->spa_dedup_checksum); 2033} 2034 2035/* 2036 * Reset pool scan stat per scan pass (or reboot). 2037 */ 2038void 2039spa_scan_stat_init(spa_t *spa) 2040{ 2041 /* data not stored on disk */ 2042 spa->spa_scan_pass_start = gethrestime_sec(); 2043 spa->spa_scan_pass_exam = 0; 2044 vdev_scan_stat_init(spa->spa_root_vdev); 2045} 2046 2047/* 2048 * Get scan stats for zpool status reports 2049 */ 2050int 2051spa_scan_get_stats(spa_t *spa, pool_scan_stat_t *ps) 2052{ 2053 dsl_scan_t *scn = spa->spa_dsl_pool ? spa->spa_dsl_pool->dp_scan : NULL; 2054 2055 if (scn == NULL || scn->scn_phys.scn_func == POOL_SCAN_NONE) 2056 return (SET_ERROR(ENOENT)); 2057 bzero(ps, sizeof (pool_scan_stat_t)); 2058 2059 /* data stored on disk */ 2060 ps->pss_func = scn->scn_phys.scn_func; 2061 ps->pss_start_time = scn->scn_phys.scn_start_time; 2062 ps->pss_end_time = scn->scn_phys.scn_end_time; 2063 ps->pss_to_examine = scn->scn_phys.scn_to_examine; 2064 ps->pss_examined = scn->scn_phys.scn_examined; 2065 ps->pss_to_process = scn->scn_phys.scn_to_process; 2066 ps->pss_processed = scn->scn_phys.scn_processed; 2067 ps->pss_errors = scn->scn_phys.scn_errors; 2068 ps->pss_state = scn->scn_phys.scn_state; 2069 2070 /* data not stored on disk */ 2071 ps->pss_pass_start = spa->spa_scan_pass_start; 2072 ps->pss_pass_exam = spa->spa_scan_pass_exam; 2073 2074 return (0); 2075} 2076 2077boolean_t 2078spa_debug_enabled(spa_t *spa) 2079{ 2080 return (spa->spa_debug); 2081} 2082 2083int 2084spa_maxblocksize(spa_t *spa) 2085{ 2086 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) 2087 return (SPA_MAXBLOCKSIZE); 2088 else 2089 return (SPA_OLD_MAXBLOCKSIZE); 2090} 2091