spa_misc.c revision 290757
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 * Copyright 2013 Saso Kiselkov. All rights reserved. 28 */ 29 30#include <sys/zfs_context.h> 31#include <sys/spa_impl.h> 32#include <sys/spa_boot.h> 33#include <sys/zio.h> 34#include <sys/zio_checksum.h> 35#include <sys/zio_compress.h> 36#include <sys/dmu.h> 37#include <sys/dmu_tx.h> 38#include <sys/zap.h> 39#include <sys/zil.h> 40#include <sys/vdev_impl.h> 41#include <sys/metaslab.h> 42#include <sys/uberblock_impl.h> 43#include <sys/txg.h> 44#include <sys/avl.h> 45#include <sys/unique.h> 46#include <sys/dsl_pool.h> 47#include <sys/dsl_dir.h> 48#include <sys/dsl_prop.h> 49#include <sys/dsl_scan.h> 50#include <sys/fs/zfs.h> 51#include <sys/metaslab_impl.h> 52#include <sys/arc.h> 53#include <sys/ddt.h> 54#include "zfs_prop.h" 55#include <sys/zfeature.h> 56 57/* 58 * SPA locking 59 * 60 * There are four basic locks for managing spa_t structures: 61 * 62 * spa_namespace_lock (global mutex) 63 * 64 * This lock must be acquired to do any of the following: 65 * 66 * - Lookup a spa_t by name 67 * - Add or remove a spa_t from the namespace 68 * - Increase spa_refcount from non-zero 69 * - Check if spa_refcount is zero 70 * - Rename a spa_t 71 * - add/remove/attach/detach devices 72 * - Held for the duration of create/destroy/import/export 73 * 74 * It does not need to handle recursion. A create or destroy may 75 * reference objects (files or zvols) in other pools, but by 76 * definition they must have an existing reference, and will never need 77 * to lookup a spa_t by name. 78 * 79 * spa_refcount (per-spa refcount_t protected by mutex) 80 * 81 * This reference count keep track of any active users of the spa_t. The 82 * spa_t cannot be destroyed or freed while this is non-zero. Internally, 83 * the refcount is never really 'zero' - opening a pool implicitly keeps 84 * some references in the DMU. Internally we check against spa_minref, but 85 * present the image of a zero/non-zero value to consumers. 86 * 87 * spa_config_lock[] (per-spa array of rwlocks) 88 * 89 * This protects the spa_t from config changes, and must be held in 90 * the following circumstances: 91 * 92 * - RW_READER to perform I/O to the spa 93 * - RW_WRITER to change the vdev config 94 * 95 * The locking order is fairly straightforward: 96 * 97 * spa_namespace_lock -> spa_refcount 98 * 99 * The namespace lock must be acquired to increase the refcount from 0 100 * or to check if it is zero. 101 * 102 * spa_refcount -> spa_config_lock[] 103 * 104 * There must be at least one valid reference on the spa_t to acquire 105 * the config lock. 106 * 107 * spa_namespace_lock -> spa_config_lock[] 108 * 109 * The namespace lock must always be taken before the config lock. 110 * 111 * 112 * The spa_namespace_lock can be acquired directly and is globally visible. 113 * 114 * The namespace is manipulated using the following functions, all of which 115 * require the spa_namespace_lock to be held. 116 * 117 * spa_lookup() Lookup a spa_t by name. 118 * 119 * spa_add() Create a new spa_t in the namespace. 120 * 121 * spa_remove() Remove a spa_t from the namespace. This also 122 * frees up any memory associated with the spa_t. 123 * 124 * spa_next() Returns the next spa_t in the system, or the 125 * first if NULL is passed. 126 * 127 * spa_evict_all() Shutdown and remove all spa_t structures in 128 * the system. 129 * 130 * spa_guid_exists() Determine whether a pool/device guid exists. 131 * 132 * The spa_refcount is manipulated using the following functions: 133 * 134 * spa_open_ref() Adds a reference to the given spa_t. Must be 135 * called with spa_namespace_lock held if the 136 * refcount is currently zero. 137 * 138 * spa_close() Remove a reference from the spa_t. This will 139 * not free the spa_t or remove it from the 140 * namespace. No locking is required. 141 * 142 * spa_refcount_zero() Returns true if the refcount is currently 143 * zero. Must be called with spa_namespace_lock 144 * held. 145 * 146 * The spa_config_lock[] is an array of rwlocks, ordered as follows: 147 * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV. 148 * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}(). 149 * 150 * To read the configuration, it suffices to hold one of these locks as reader. 151 * To modify the configuration, you must hold all locks as writer. To modify 152 * vdev state without altering the vdev tree's topology (e.g. online/offline), 153 * you must hold SCL_STATE and SCL_ZIO as writer. 154 * 155 * We use these distinct config locks to avoid recursive lock entry. 156 * For example, spa_sync() (which holds SCL_CONFIG as reader) induces 157 * block allocations (SCL_ALLOC), which may require reading space maps 158 * from disk (dmu_read() -> zio_read() -> SCL_ZIO). 159 * 160 * The spa config locks cannot be normal rwlocks because we need the 161 * ability to hand off ownership. For example, SCL_ZIO is acquired 162 * by the issuing thread and later released by an interrupt thread. 163 * They do, however, obey the usual write-wanted semantics to prevent 164 * writer (i.e. system administrator) starvation. 165 * 166 * The lock acquisition rules are as follows: 167 * 168 * SCL_CONFIG 169 * Protects changes to the vdev tree topology, such as vdev 170 * add/remove/attach/detach. Protects the dirty config list 171 * (spa_config_dirty_list) and the set of spares and l2arc devices. 172 * 173 * SCL_STATE 174 * Protects changes to pool state and vdev state, such as vdev 175 * online/offline/fault/degrade/clear. Protects the dirty state list 176 * (spa_state_dirty_list) and global pool state (spa_state). 177 * 178 * SCL_ALLOC 179 * Protects changes to metaslab groups and classes. 180 * Held as reader by metaslab_alloc() and metaslab_claim(). 181 * 182 * SCL_ZIO 183 * Held by bp-level zios (those which have no io_vd upon entry) 184 * to prevent changes to the vdev tree. The bp-level zio implicitly 185 * protects all of its vdev child zios, which do not hold SCL_ZIO. 186 * 187 * SCL_FREE 188 * Protects changes to metaslab groups and classes. 189 * Held as reader by metaslab_free(). SCL_FREE is distinct from 190 * SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free 191 * blocks in zio_done() while another i/o that holds either 192 * SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete. 193 * 194 * SCL_VDEV 195 * Held as reader to prevent changes to the vdev tree during trivial 196 * inquiries such as bp_get_dsize(). SCL_VDEV is distinct from the 197 * other locks, and lower than all of them, to ensure that it's safe 198 * to acquire regardless of caller context. 199 * 200 * In addition, the following rules apply: 201 * 202 * (a) spa_props_lock protects pool properties, spa_config and spa_config_list. 203 * The lock ordering is SCL_CONFIG > spa_props_lock. 204 * 205 * (b) I/O operations on leaf vdevs. For any zio operation that takes 206 * an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(), 207 * or zio_write_phys() -- the caller must ensure that the config cannot 208 * cannot change in the interim, and that the vdev cannot be reopened. 209 * SCL_STATE as reader suffices for both. 210 * 211 * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit(). 212 * 213 * spa_vdev_enter() Acquire the namespace lock and the config lock 214 * for writing. 215 * 216 * spa_vdev_exit() Release the config lock, wait for all I/O 217 * to complete, sync the updated configs to the 218 * cache, and release the namespace lock. 219 * 220 * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit(). 221 * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual 222 * locking is, always, based on spa_namespace_lock and spa_config_lock[]. 223 * 224 * spa_rename() is also implemented within this file since it requires 225 * manipulation of the namespace. 226 */ 227 228static avl_tree_t spa_namespace_avl; 229kmutex_t spa_namespace_lock; 230static kcondvar_t spa_namespace_cv; 231static int spa_active_count; 232int spa_max_replication_override = SPA_DVAS_PER_BP; 233 234static kmutex_t spa_spare_lock; 235static avl_tree_t spa_spare_avl; 236static kmutex_t spa_l2cache_lock; 237static avl_tree_t spa_l2cache_avl; 238 239kmem_cache_t *spa_buffer_pool; 240int spa_mode_global; 241 242#ifdef ZFS_DEBUG 243/* Everything except dprintf and spa is on by default in debug builds */ 244int zfs_flags = ~(ZFS_DEBUG_DPRINTF | ZFS_DEBUG_SPA); 245#else 246int zfs_flags = 0; 247#endif 248SYSCTL_DECL(_debug); 249TUNABLE_INT("debug.zfs_flags", &zfs_flags); 250SYSCTL_INT(_debug, OID_AUTO, zfs_flags, CTLFLAG_RWTUN, &zfs_flags, 0, 251 "ZFS debug flags."); 252 253/* 254 * zfs_recover can be set to nonzero to attempt to recover from 255 * otherwise-fatal errors, typically caused by on-disk corruption. When 256 * set, calls to zfs_panic_recover() will turn into warning messages. 257 * This should only be used as a last resort, as it typically results 258 * in leaked space, or worse. 259 */ 260boolean_t zfs_recover = B_FALSE; 261SYSCTL_DECL(_vfs_zfs); 262TUNABLE_INT("vfs.zfs.recover", &zfs_recover); 263SYSCTL_INT(_vfs_zfs, OID_AUTO, recover, CTLFLAG_RWTUN, &zfs_recover, 0, 264 "Try to recover from otherwise-fatal errors."); 265 266/* 267 * If destroy encounters an EIO while reading metadata (e.g. indirect 268 * blocks), space referenced by the missing metadata can not be freed. 269 * Normally this causes the background destroy to become "stalled", as 270 * it is unable to make forward progress. While in this stalled state, 271 * all remaining space to free from the error-encountering filesystem is 272 * "temporarily leaked". Set this flag to cause it to ignore the EIO, 273 * permanently leak the space from indirect blocks that can not be read, 274 * and continue to free everything else that it can. 275 * 276 * The default, "stalling" behavior is useful if the storage partially 277 * fails (i.e. some but not all i/os fail), and then later recovers. In 278 * this case, we will be able to continue pool operations while it is 279 * partially failed, and when it recovers, we can continue to free the 280 * space, with no leaks. However, note that this case is actually 281 * fairly rare. 282 * 283 * Typically pools either (a) fail completely (but perhaps temporarily, 284 * e.g. a top-level vdev going offline), or (b) have localized, 285 * permanent errors (e.g. disk returns the wrong data due to bit flip or 286 * firmware bug). In case (a), this setting does not matter because the 287 * pool will be suspended and the sync thread will not be able to make 288 * forward progress regardless. In case (b), because the error is 289 * permanent, the best we can do is leak the minimum amount of space, 290 * which is what setting this flag will do. Therefore, it is reasonable 291 * for this flag to normally be set, but we chose the more conservative 292 * approach of not setting it, so that there is no possibility of 293 * leaking space in the "partial temporary" failure case. 294 */ 295boolean_t zfs_free_leak_on_eio = B_FALSE; 296 297/* 298 * Expiration time in milliseconds. This value has two meanings. First it is 299 * used to determine when the spa_deadman() logic should fire. By default the 300 * spa_deadman() will fire if spa_sync() has not completed in 1000 seconds. 301 * Secondly, the value determines if an I/O is considered "hung". Any I/O that 302 * has not completed in zfs_deadman_synctime_ms is considered "hung" resulting 303 * in a system panic. 304 */ 305uint64_t zfs_deadman_synctime_ms = 1000000ULL; 306TUNABLE_QUAD("vfs.zfs.deadman_synctime_ms", &zfs_deadman_synctime_ms); 307SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, deadman_synctime_ms, CTLFLAG_RDTUN, 308 &zfs_deadman_synctime_ms, 0, 309 "Stalled ZFS I/O expiration time in milliseconds"); 310 311/* 312 * Check time in milliseconds. This defines the frequency at which we check 313 * for hung I/O. 314 */ 315uint64_t zfs_deadman_checktime_ms = 5000ULL; 316TUNABLE_QUAD("vfs.zfs.deadman_checktime_ms", &zfs_deadman_checktime_ms); 317SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, deadman_checktime_ms, CTLFLAG_RDTUN, 318 &zfs_deadman_checktime_ms, 0, 319 "Period of checks for stalled ZFS I/O in milliseconds"); 320 321/* 322 * Default value of -1 for zfs_deadman_enabled is resolved in 323 * zfs_deadman_init() 324 */ 325int zfs_deadman_enabled = -1; 326TUNABLE_INT("vfs.zfs.deadman_enabled", &zfs_deadman_enabled); 327SYSCTL_INT(_vfs_zfs, OID_AUTO, deadman_enabled, CTLFLAG_RDTUN, 328 &zfs_deadman_enabled, 0, "Kernel panic on stalled ZFS I/O"); 329 330/* 331 * The worst case is single-sector max-parity RAID-Z blocks, in which 332 * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1) 333 * times the size; so just assume that. Add to this the fact that 334 * we can have up to 3 DVAs per bp, and one more factor of 2 because 335 * the block may be dittoed with up to 3 DVAs by ddt_sync(). All together, 336 * the worst case is: 337 * (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2 == 24 338 */ 339int spa_asize_inflation = 24; 340TUNABLE_INT("vfs.zfs.spa_asize_inflation", &spa_asize_inflation); 341SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_asize_inflation, CTLFLAG_RWTUN, 342 &spa_asize_inflation, 0, "Worst case inflation factor for single sector writes"); 343 344#ifndef illumos 345#ifdef _KERNEL 346static void 347zfs_deadman_init() 348{ 349 /* 350 * If we are not i386 or amd64 or in a virtual machine, 351 * disable ZFS deadman thread by default 352 */ 353 if (zfs_deadman_enabled == -1) { 354#if defined(__amd64__) || defined(__i386__) 355 zfs_deadman_enabled = (vm_guest == VM_GUEST_NO) ? 1 : 0; 356#else 357 zfs_deadman_enabled = 0; 358#endif 359 } 360} 361#endif /* _KERNEL */ 362#endif /* !illumos */ 363 364/* 365 * Normally, we don't allow the last 3.2% (1/(2^spa_slop_shift)) of space in 366 * the pool to be consumed. This ensures that we don't run the pool 367 * completely out of space, due to unaccounted changes (e.g. to the MOS). 368 * It also limits the worst-case time to allocate space. If we have 369 * less than this amount of free space, most ZPL operations (e.g. write, 370 * create) will return ENOSPC. 371 * 372 * Certain operations (e.g. file removal, most administrative actions) can 373 * use half the slop space. They will only return ENOSPC if less than half 374 * the slop space is free. Typically, once the pool has less than the slop 375 * space free, the user will use these operations to free up space in the pool. 376 * These are the operations that call dsl_pool_adjustedsize() with the netfree 377 * argument set to TRUE. 378 * 379 * A very restricted set of operations are always permitted, regardless of 380 * the amount of free space. These are the operations that call 381 * dsl_sync_task(ZFS_SPACE_CHECK_NONE), e.g. "zfs destroy". If these 382 * operations result in a net increase in the amount of space used, 383 * it is possible to run the pool completely out of space, causing it to 384 * be permanently read-only. 385 * 386 * See also the comments in zfs_space_check_t. 387 */ 388int spa_slop_shift = 5; 389SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_slop_shift, CTLFLAG_RWTUN, 390 &spa_slop_shift, 0, 391 "Shift value of reserved space (1/(2^spa_slop_shift))."); 392 393/* 394 * ========================================================================== 395 * SPA config locking 396 * ========================================================================== 397 */ 398static void 399spa_config_lock_init(spa_t *spa) 400{ 401 for (int i = 0; i < SCL_LOCKS; i++) { 402 spa_config_lock_t *scl = &spa->spa_config_lock[i]; 403 mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL); 404 cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL); 405 refcount_create_untracked(&scl->scl_count); 406 scl->scl_writer = NULL; 407 scl->scl_write_wanted = 0; 408 } 409} 410 411static void 412spa_config_lock_destroy(spa_t *spa) 413{ 414 for (int i = 0; i < SCL_LOCKS; i++) { 415 spa_config_lock_t *scl = &spa->spa_config_lock[i]; 416 mutex_destroy(&scl->scl_lock); 417 cv_destroy(&scl->scl_cv); 418 refcount_destroy(&scl->scl_count); 419 ASSERT(scl->scl_writer == NULL); 420 ASSERT(scl->scl_write_wanted == 0); 421 } 422} 423 424int 425spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw) 426{ 427 for (int i = 0; i < SCL_LOCKS; 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 if (rw == RW_READER) { 433 if (scl->scl_writer || scl->scl_write_wanted) { 434 mutex_exit(&scl->scl_lock); 435 spa_config_exit(spa, locks ^ (1 << i), tag); 436 return (0); 437 } 438 } else { 439 ASSERT(scl->scl_writer != curthread); 440 if (!refcount_is_zero(&scl->scl_count)) { 441 mutex_exit(&scl->scl_lock); 442 spa_config_exit(spa, locks ^ (1 << i), tag); 443 return (0); 444 } 445 scl->scl_writer = curthread; 446 } 447 (void) refcount_add(&scl->scl_count, tag); 448 mutex_exit(&scl->scl_lock); 449 } 450 return (1); 451} 452 453void 454spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw) 455{ 456 int wlocks_held = 0; 457 458 ASSERT3U(SCL_LOCKS, <, sizeof (wlocks_held) * NBBY); 459 460 for (int i = 0; i < SCL_LOCKS; i++) { 461 spa_config_lock_t *scl = &spa->spa_config_lock[i]; 462 if (scl->scl_writer == curthread) 463 wlocks_held |= (1 << i); 464 if (!(locks & (1 << i))) 465 continue; 466 mutex_enter(&scl->scl_lock); 467 if (rw == RW_READER) { 468 while (scl->scl_writer || scl->scl_write_wanted) { 469 cv_wait(&scl->scl_cv, &scl->scl_lock); 470 } 471 } else { 472 ASSERT(scl->scl_writer != curthread); 473 while (!refcount_is_zero(&scl->scl_count)) { 474 scl->scl_write_wanted++; 475 cv_wait(&scl->scl_cv, &scl->scl_lock); 476 scl->scl_write_wanted--; 477 } 478 scl->scl_writer = curthread; 479 } 480 (void) refcount_add(&scl->scl_count, tag); 481 mutex_exit(&scl->scl_lock); 482 } 483 ASSERT(wlocks_held <= locks); 484} 485 486void 487spa_config_exit(spa_t *spa, int locks, void *tag) 488{ 489 for (int i = SCL_LOCKS - 1; i >= 0; i--) { 490 spa_config_lock_t *scl = &spa->spa_config_lock[i]; 491 if (!(locks & (1 << i))) 492 continue; 493 mutex_enter(&scl->scl_lock); 494 ASSERT(!refcount_is_zero(&scl->scl_count)); 495 if (refcount_remove(&scl->scl_count, tag) == 0) { 496 ASSERT(scl->scl_writer == NULL || 497 scl->scl_writer == curthread); 498 scl->scl_writer = NULL; /* OK in either case */ 499 cv_broadcast(&scl->scl_cv); 500 } 501 mutex_exit(&scl->scl_lock); 502 } 503} 504 505int 506spa_config_held(spa_t *spa, int locks, krw_t rw) 507{ 508 int locks_held = 0; 509 510 for (int i = 0; i < SCL_LOCKS; i++) { 511 spa_config_lock_t *scl = &spa->spa_config_lock[i]; 512 if (!(locks & (1 << i))) 513 continue; 514 if ((rw == RW_READER && !refcount_is_zero(&scl->scl_count)) || 515 (rw == RW_WRITER && scl->scl_writer == curthread)) 516 locks_held |= 1 << i; 517 } 518 519 return (locks_held); 520} 521 522/* 523 * ========================================================================== 524 * SPA namespace functions 525 * ========================================================================== 526 */ 527 528/* 529 * Lookup the named spa_t in the AVL tree. The spa_namespace_lock must be held. 530 * Returns NULL if no matching spa_t is found. 531 */ 532spa_t * 533spa_lookup(const char *name) 534{ 535 static spa_t search; /* spa_t is large; don't allocate on stack */ 536 spa_t *spa; 537 avl_index_t where; 538 char *cp; 539 540 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 541 542 (void) strlcpy(search.spa_name, name, sizeof (search.spa_name)); 543 544 /* 545 * If it's a full dataset name, figure out the pool name and 546 * just use that. 547 */ 548 cp = strpbrk(search.spa_name, "/@#"); 549 if (cp != NULL) 550 *cp = '\0'; 551 552 spa = avl_find(&spa_namespace_avl, &search, &where); 553 554 return (spa); 555} 556 557/* 558 * Fires when spa_sync has not completed within zfs_deadman_synctime_ms. 559 * If the zfs_deadman_enabled flag is set then it inspects all vdev queues 560 * looking for potentially hung I/Os. 561 */ 562void 563spa_deadman(void *arg) 564{ 565 spa_t *spa = arg; 566 567 /* 568 * Disable the deadman timer if the pool is suspended. 569 */ 570 if (spa_suspended(spa)) { 571#ifdef illumos 572 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY)); 573#else 574 /* Nothing. just don't schedule any future callouts. */ 575#endif 576 return; 577 } 578 579 zfs_dbgmsg("slow spa_sync: started %llu seconds ago, calls %llu", 580 (gethrtime() - spa->spa_sync_starttime) / NANOSEC, 581 ++spa->spa_deadman_calls); 582 if (zfs_deadman_enabled) 583 vdev_deadman(spa->spa_root_vdev); 584} 585 586/* 587 * Create an uninitialized spa_t with the given name. Requires 588 * spa_namespace_lock. The caller must ensure that the spa_t doesn't already 589 * exist by calling spa_lookup() first. 590 */ 591spa_t * 592spa_add(const char *name, nvlist_t *config, const char *altroot) 593{ 594 spa_t *spa; 595 spa_config_dirent_t *dp; 596#ifdef illumos 597 cyc_handler_t hdlr; 598 cyc_time_t when; 599#endif 600 601 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 602 603 spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP); 604 605 mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL); 606 mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL); 607 mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL); 608 mutex_init(&spa->spa_evicting_os_lock, NULL, MUTEX_DEFAULT, NULL); 609 mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL); 610 mutex_init(&spa->spa_proc_lock, NULL, MUTEX_DEFAULT, NULL); 611 mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL); 612 mutex_init(&spa->spa_cksum_tmpls_lock, NULL, MUTEX_DEFAULT, NULL); 613 mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL); 614 mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL); 615 mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL); 616 617 cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL); 618 cv_init(&spa->spa_evicting_os_cv, NULL, CV_DEFAULT, NULL); 619 cv_init(&spa->spa_proc_cv, NULL, CV_DEFAULT, NULL); 620 cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL); 621 cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL); 622 623 for (int t = 0; t < TXG_SIZE; t++) 624 bplist_create(&spa->spa_free_bplist[t]); 625 626 (void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name)); 627 spa->spa_state = POOL_STATE_UNINITIALIZED; 628 spa->spa_freeze_txg = UINT64_MAX; 629 spa->spa_final_txg = UINT64_MAX; 630 spa->spa_load_max_txg = UINT64_MAX; 631 spa->spa_proc = &p0; 632 spa->spa_proc_state = SPA_PROC_NONE; 633 634#ifdef illumos 635 hdlr.cyh_func = spa_deadman; 636 hdlr.cyh_arg = spa; 637 hdlr.cyh_level = CY_LOW_LEVEL; 638#endif 639 640 spa->spa_deadman_synctime = MSEC2NSEC(zfs_deadman_synctime_ms); 641 642#ifdef illumos 643 /* 644 * This determines how often we need to check for hung I/Os after 645 * the cyclic has already fired. Since checking for hung I/Os is 646 * an expensive operation we don't want to check too frequently. 647 * Instead wait for 5 seconds before checking again. 648 */ 649 when.cyt_interval = MSEC2NSEC(zfs_deadman_checktime_ms); 650 when.cyt_when = CY_INFINITY; 651 mutex_enter(&cpu_lock); 652 spa->spa_deadman_cycid = cyclic_add(&hdlr, &when); 653 mutex_exit(&cpu_lock); 654#else /* !illumos */ 655#ifdef _KERNEL 656 callout_init(&spa->spa_deadman_cycid, CALLOUT_MPSAFE); 657#endif 658#endif 659 refcount_create(&spa->spa_refcount); 660 spa_config_lock_init(spa); 661 662 avl_add(&spa_namespace_avl, spa); 663 664 /* 665 * Set the alternate root, if there is one. 666 */ 667 if (altroot) { 668 spa->spa_root = spa_strdup(altroot); 669 spa_active_count++; 670 } 671 672 /* 673 * Every pool starts with the default cachefile 674 */ 675 list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t), 676 offsetof(spa_config_dirent_t, scd_link)); 677 678 dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP); 679 dp->scd_path = altroot ? NULL : spa_strdup(spa_config_path); 680 list_insert_head(&spa->spa_config_list, dp); 681 682 VERIFY(nvlist_alloc(&spa->spa_load_info, NV_UNIQUE_NAME, 683 KM_SLEEP) == 0); 684 685 if (config != NULL) { 686 nvlist_t *features; 687 688 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ, 689 &features) == 0) { 690 VERIFY(nvlist_dup(features, &spa->spa_label_features, 691 0) == 0); 692 } 693 694 VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0); 695 } 696 697 if (spa->spa_label_features == NULL) { 698 VERIFY(nvlist_alloc(&spa->spa_label_features, NV_UNIQUE_NAME, 699 KM_SLEEP) == 0); 700 } 701 702 spa->spa_debug = ((zfs_flags & ZFS_DEBUG_SPA) != 0); 703 704 spa->spa_min_ashift = INT_MAX; 705 spa->spa_max_ashift = 0; 706 707 /* 708 * As a pool is being created, treat all features as disabled by 709 * setting SPA_FEATURE_DISABLED for all entries in the feature 710 * refcount cache. 711 */ 712 for (int i = 0; i < SPA_FEATURES; i++) { 713 spa->spa_feat_refcount_cache[i] = SPA_FEATURE_DISABLED; 714 } 715 716 return (spa); 717} 718 719/* 720 * Removes a spa_t from the namespace, freeing up any memory used. Requires 721 * spa_namespace_lock. This is called only after the spa_t has been closed and 722 * deactivated. 723 */ 724void 725spa_remove(spa_t *spa) 726{ 727 spa_config_dirent_t *dp; 728 729 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 730 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); 731 ASSERT3U(refcount_count(&spa->spa_refcount), ==, 0); 732 733 nvlist_free(spa->spa_config_splitting); 734 735 avl_remove(&spa_namespace_avl, spa); 736 cv_broadcast(&spa_namespace_cv); 737 738 if (spa->spa_root) { 739 spa_strfree(spa->spa_root); 740 spa_active_count--; 741 } 742 743 while ((dp = list_head(&spa->spa_config_list)) != NULL) { 744 list_remove(&spa->spa_config_list, dp); 745 if (dp->scd_path != NULL) 746 spa_strfree(dp->scd_path); 747 kmem_free(dp, sizeof (spa_config_dirent_t)); 748 } 749 750 list_destroy(&spa->spa_config_list); 751 752 nvlist_free(spa->spa_label_features); 753 nvlist_free(spa->spa_load_info); 754 spa_config_set(spa, NULL); 755 756#ifdef illumos 757 mutex_enter(&cpu_lock); 758 if (spa->spa_deadman_cycid != CYCLIC_NONE) 759 cyclic_remove(spa->spa_deadman_cycid); 760 mutex_exit(&cpu_lock); 761 spa->spa_deadman_cycid = CYCLIC_NONE; 762#else /* !illumos */ 763#ifdef _KERNEL 764 callout_drain(&spa->spa_deadman_cycid); 765#endif 766#endif 767 768 refcount_destroy(&spa->spa_refcount); 769 770 spa_config_lock_destroy(spa); 771 772 for (int t = 0; t < TXG_SIZE; t++) 773 bplist_destroy(&spa->spa_free_bplist[t]); 774 775 zio_checksum_templates_free(spa); 776 777 cv_destroy(&spa->spa_async_cv); 778 cv_destroy(&spa->spa_evicting_os_cv); 779 cv_destroy(&spa->spa_proc_cv); 780 cv_destroy(&spa->spa_scrub_io_cv); 781 cv_destroy(&spa->spa_suspend_cv); 782 783 mutex_destroy(&spa->spa_async_lock); 784 mutex_destroy(&spa->spa_errlist_lock); 785 mutex_destroy(&spa->spa_errlog_lock); 786 mutex_destroy(&spa->spa_evicting_os_lock); 787 mutex_destroy(&spa->spa_history_lock); 788 mutex_destroy(&spa->spa_proc_lock); 789 mutex_destroy(&spa->spa_props_lock); 790 mutex_destroy(&spa->spa_cksum_tmpls_lock); 791 mutex_destroy(&spa->spa_scrub_lock); 792 mutex_destroy(&spa->spa_suspend_lock); 793 mutex_destroy(&spa->spa_vdev_top_lock); 794 795 kmem_free(spa, sizeof (spa_t)); 796} 797 798/* 799 * Given a pool, return the next pool in the namespace, or NULL if there is 800 * none. If 'prev' is NULL, return the first pool. 801 */ 802spa_t * 803spa_next(spa_t *prev) 804{ 805 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 806 807 if (prev) 808 return (AVL_NEXT(&spa_namespace_avl, prev)); 809 else 810 return (avl_first(&spa_namespace_avl)); 811} 812 813/* 814 * ========================================================================== 815 * SPA refcount functions 816 * ========================================================================== 817 */ 818 819/* 820 * Add a reference to the given spa_t. Must have at least one reference, or 821 * have the namespace lock held. 822 */ 823void 824spa_open_ref(spa_t *spa, void *tag) 825{ 826 ASSERT(refcount_count(&spa->spa_refcount) >= spa->spa_minref || 827 MUTEX_HELD(&spa_namespace_lock)); 828 (void) refcount_add(&spa->spa_refcount, tag); 829} 830 831/* 832 * Remove a reference to the given spa_t. Must have at least one reference, or 833 * have the namespace lock held. 834 */ 835void 836spa_close(spa_t *spa, void *tag) 837{ 838 ASSERT(refcount_count(&spa->spa_refcount) > spa->spa_minref || 839 MUTEX_HELD(&spa_namespace_lock)); 840 (void) refcount_remove(&spa->spa_refcount, tag); 841} 842 843/* 844 * Remove a reference to the given spa_t held by a dsl dir that is 845 * being asynchronously released. Async releases occur from a taskq 846 * performing eviction of dsl datasets and dirs. The namespace lock 847 * isn't held and the hold by the object being evicted may contribute to 848 * spa_minref (e.g. dataset or directory released during pool export), 849 * so the asserts in spa_close() do not apply. 850 */ 851void 852spa_async_close(spa_t *spa, void *tag) 853{ 854 (void) refcount_remove(&spa->spa_refcount, tag); 855} 856 857/* 858 * Check to see if the spa refcount is zero. Must be called with 859 * spa_namespace_lock held. We really compare against spa_minref, which is the 860 * number of references acquired when opening a pool 861 */ 862boolean_t 863spa_refcount_zero(spa_t *spa) 864{ 865 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 866 867 return (refcount_count(&spa->spa_refcount) == spa->spa_minref); 868} 869 870/* 871 * ========================================================================== 872 * SPA spare and l2cache tracking 873 * ========================================================================== 874 */ 875 876/* 877 * Hot spares and cache devices are tracked using the same code below, 878 * for 'auxiliary' devices. 879 */ 880 881typedef struct spa_aux { 882 uint64_t aux_guid; 883 uint64_t aux_pool; 884 avl_node_t aux_avl; 885 int aux_count; 886} spa_aux_t; 887 888static int 889spa_aux_compare(const void *a, const void *b) 890{ 891 const spa_aux_t *sa = a; 892 const spa_aux_t *sb = b; 893 894 if (sa->aux_guid < sb->aux_guid) 895 return (-1); 896 else if (sa->aux_guid > sb->aux_guid) 897 return (1); 898 else 899 return (0); 900} 901 902void 903spa_aux_add(vdev_t *vd, avl_tree_t *avl) 904{ 905 avl_index_t where; 906 spa_aux_t search; 907 spa_aux_t *aux; 908 909 search.aux_guid = vd->vdev_guid; 910 if ((aux = avl_find(avl, &search, &where)) != NULL) { 911 aux->aux_count++; 912 } else { 913 aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP); 914 aux->aux_guid = vd->vdev_guid; 915 aux->aux_count = 1; 916 avl_insert(avl, aux, where); 917 } 918} 919 920void 921spa_aux_remove(vdev_t *vd, avl_tree_t *avl) 922{ 923 spa_aux_t search; 924 spa_aux_t *aux; 925 avl_index_t where; 926 927 search.aux_guid = vd->vdev_guid; 928 aux = avl_find(avl, &search, &where); 929 930 ASSERT(aux != NULL); 931 932 if (--aux->aux_count == 0) { 933 avl_remove(avl, aux); 934 kmem_free(aux, sizeof (spa_aux_t)); 935 } else if (aux->aux_pool == spa_guid(vd->vdev_spa)) { 936 aux->aux_pool = 0ULL; 937 } 938} 939 940boolean_t 941spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl) 942{ 943 spa_aux_t search, *found; 944 945 search.aux_guid = guid; 946 found = avl_find(avl, &search, NULL); 947 948 if (pool) { 949 if (found) 950 *pool = found->aux_pool; 951 else 952 *pool = 0ULL; 953 } 954 955 if (refcnt) { 956 if (found) 957 *refcnt = found->aux_count; 958 else 959 *refcnt = 0; 960 } 961 962 return (found != NULL); 963} 964 965void 966spa_aux_activate(vdev_t *vd, avl_tree_t *avl) 967{ 968 spa_aux_t search, *found; 969 avl_index_t where; 970 971 search.aux_guid = vd->vdev_guid; 972 found = avl_find(avl, &search, &where); 973 ASSERT(found != NULL); 974 ASSERT(found->aux_pool == 0ULL); 975 976 found->aux_pool = spa_guid(vd->vdev_spa); 977} 978 979/* 980 * Spares are tracked globally due to the following constraints: 981 * 982 * - A spare may be part of multiple pools. 983 * - A spare may be added to a pool even if it's actively in use within 984 * another pool. 985 * - A spare in use in any pool can only be the source of a replacement if 986 * the target is a spare in the same pool. 987 * 988 * We keep track of all spares on the system through the use of a reference 989 * counted AVL tree. When a vdev is added as a spare, or used as a replacement 990 * spare, then we bump the reference count in the AVL tree. In addition, we set 991 * the 'vdev_isspare' member to indicate that the device is a spare (active or 992 * inactive). When a spare is made active (used to replace a device in the 993 * pool), we also keep track of which pool its been made a part of. 994 * 995 * The 'spa_spare_lock' protects the AVL tree. These functions are normally 996 * called under the spa_namespace lock as part of vdev reconfiguration. The 997 * separate spare lock exists for the status query path, which does not need to 998 * be completely consistent with respect to other vdev configuration changes. 999 */ 1000 1001static int 1002spa_spare_compare(const void *a, const void *b) 1003{ 1004 return (spa_aux_compare(a, b)); 1005} 1006 1007void 1008spa_spare_add(vdev_t *vd) 1009{ 1010 mutex_enter(&spa_spare_lock); 1011 ASSERT(!vd->vdev_isspare); 1012 spa_aux_add(vd, &spa_spare_avl); 1013 vd->vdev_isspare = B_TRUE; 1014 mutex_exit(&spa_spare_lock); 1015} 1016 1017void 1018spa_spare_remove(vdev_t *vd) 1019{ 1020 mutex_enter(&spa_spare_lock); 1021 ASSERT(vd->vdev_isspare); 1022 spa_aux_remove(vd, &spa_spare_avl); 1023 vd->vdev_isspare = B_FALSE; 1024 mutex_exit(&spa_spare_lock); 1025} 1026 1027boolean_t 1028spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt) 1029{ 1030 boolean_t found; 1031 1032 mutex_enter(&spa_spare_lock); 1033 found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl); 1034 mutex_exit(&spa_spare_lock); 1035 1036 return (found); 1037} 1038 1039void 1040spa_spare_activate(vdev_t *vd) 1041{ 1042 mutex_enter(&spa_spare_lock); 1043 ASSERT(vd->vdev_isspare); 1044 spa_aux_activate(vd, &spa_spare_avl); 1045 mutex_exit(&spa_spare_lock); 1046} 1047 1048/* 1049 * Level 2 ARC devices are tracked globally for the same reasons as spares. 1050 * Cache devices currently only support one pool per cache device, and so 1051 * for these devices the aux reference count is currently unused beyond 1. 1052 */ 1053 1054static int 1055spa_l2cache_compare(const void *a, const void *b) 1056{ 1057 return (spa_aux_compare(a, b)); 1058} 1059 1060void 1061spa_l2cache_add(vdev_t *vd) 1062{ 1063 mutex_enter(&spa_l2cache_lock); 1064 ASSERT(!vd->vdev_isl2cache); 1065 spa_aux_add(vd, &spa_l2cache_avl); 1066 vd->vdev_isl2cache = B_TRUE; 1067 mutex_exit(&spa_l2cache_lock); 1068} 1069 1070void 1071spa_l2cache_remove(vdev_t *vd) 1072{ 1073 mutex_enter(&spa_l2cache_lock); 1074 ASSERT(vd->vdev_isl2cache); 1075 spa_aux_remove(vd, &spa_l2cache_avl); 1076 vd->vdev_isl2cache = B_FALSE; 1077 mutex_exit(&spa_l2cache_lock); 1078} 1079 1080boolean_t 1081spa_l2cache_exists(uint64_t guid, uint64_t *pool) 1082{ 1083 boolean_t found; 1084 1085 mutex_enter(&spa_l2cache_lock); 1086 found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl); 1087 mutex_exit(&spa_l2cache_lock); 1088 1089 return (found); 1090} 1091 1092void 1093spa_l2cache_activate(vdev_t *vd) 1094{ 1095 mutex_enter(&spa_l2cache_lock); 1096 ASSERT(vd->vdev_isl2cache); 1097 spa_aux_activate(vd, &spa_l2cache_avl); 1098 mutex_exit(&spa_l2cache_lock); 1099} 1100 1101/* 1102 * ========================================================================== 1103 * SPA vdev locking 1104 * ========================================================================== 1105 */ 1106 1107/* 1108 * Lock the given spa_t for the purpose of adding or removing a vdev. 1109 * Grabs the global spa_namespace_lock plus the spa config lock for writing. 1110 * It returns the next transaction group for the spa_t. 1111 */ 1112uint64_t 1113spa_vdev_enter(spa_t *spa) 1114{ 1115 mutex_enter(&spa->spa_vdev_top_lock); 1116 mutex_enter(&spa_namespace_lock); 1117 return (spa_vdev_config_enter(spa)); 1118} 1119 1120/* 1121 * Internal implementation for spa_vdev_enter(). Used when a vdev 1122 * operation requires multiple syncs (i.e. removing a device) while 1123 * keeping the spa_namespace_lock held. 1124 */ 1125uint64_t 1126spa_vdev_config_enter(spa_t *spa) 1127{ 1128 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1129 1130 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER); 1131 1132 return (spa_last_synced_txg(spa) + 1); 1133} 1134 1135/* 1136 * Used in combination with spa_vdev_config_enter() to allow the syncing 1137 * of multiple transactions without releasing the spa_namespace_lock. 1138 */ 1139void 1140spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag) 1141{ 1142 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1143 1144 int config_changed = B_FALSE; 1145 1146 ASSERT(txg > spa_last_synced_txg(spa)); 1147 1148 spa->spa_pending_vdev = NULL; 1149 1150 /* 1151 * Reassess the DTLs. 1152 */ 1153 vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE); 1154 1155 if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) { 1156 config_changed = B_TRUE; 1157 spa->spa_config_generation++; 1158 } 1159 1160 /* 1161 * Verify the metaslab classes. 1162 */ 1163 ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0); 1164 ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0); 1165 1166 spa_config_exit(spa, SCL_ALL, spa); 1167 1168 /* 1169 * Panic the system if the specified tag requires it. This 1170 * is useful for ensuring that configurations are updated 1171 * transactionally. 1172 */ 1173 if (zio_injection_enabled) 1174 zio_handle_panic_injection(spa, tag, 0); 1175 1176 /* 1177 * Note: this txg_wait_synced() is important because it ensures 1178 * that there won't be more than one config change per txg. 1179 * This allows us to use the txg as the generation number. 1180 */ 1181 if (error == 0) 1182 txg_wait_synced(spa->spa_dsl_pool, txg); 1183 1184 if (vd != NULL) { 1185 ASSERT(!vd->vdev_detached || vd->vdev_dtl_sm == NULL); 1186 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER); 1187 vdev_free(vd); 1188 spa_config_exit(spa, SCL_ALL, spa); 1189 } 1190 1191 /* 1192 * If the config changed, update the config cache. 1193 */ 1194 if (config_changed) 1195 spa_config_sync(spa, B_FALSE, B_TRUE); 1196} 1197 1198/* 1199 * Unlock the spa_t after adding or removing a vdev. Besides undoing the 1200 * locking of spa_vdev_enter(), we also want make sure the transactions have 1201 * synced to disk, and then update the global configuration cache with the new 1202 * information. 1203 */ 1204int 1205spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error) 1206{ 1207 spa_vdev_config_exit(spa, vd, txg, error, FTAG); 1208 mutex_exit(&spa_namespace_lock); 1209 mutex_exit(&spa->spa_vdev_top_lock); 1210 1211 return (error); 1212} 1213 1214/* 1215 * Lock the given spa_t for the purpose of changing vdev state. 1216 */ 1217void 1218spa_vdev_state_enter(spa_t *spa, int oplocks) 1219{ 1220 int locks = SCL_STATE_ALL | oplocks; 1221 1222 /* 1223 * Root pools may need to read of the underlying devfs filesystem 1224 * when opening up a vdev. Unfortunately if we're holding the 1225 * SCL_ZIO lock it will result in a deadlock when we try to issue 1226 * the read from the root filesystem. Instead we "prefetch" 1227 * the associated vnodes that we need prior to opening the 1228 * underlying devices and cache them so that we can prevent 1229 * any I/O when we are doing the actual open. 1230 */ 1231 if (spa_is_root(spa)) { 1232 int low = locks & ~(SCL_ZIO - 1); 1233 int high = locks & ~low; 1234 1235 spa_config_enter(spa, high, spa, RW_WRITER); 1236 vdev_hold(spa->spa_root_vdev); 1237 spa_config_enter(spa, low, spa, RW_WRITER); 1238 } else { 1239 spa_config_enter(spa, locks, spa, RW_WRITER); 1240 } 1241 spa->spa_vdev_locks = locks; 1242} 1243 1244int 1245spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error) 1246{ 1247 boolean_t config_changed = B_FALSE; 1248 1249 if (vd != NULL || error == 0) 1250 vdev_dtl_reassess(vd ? vd->vdev_top : spa->spa_root_vdev, 1251 0, 0, B_FALSE); 1252 1253 if (vd != NULL) { 1254 vdev_state_dirty(vd->vdev_top); 1255 config_changed = B_TRUE; 1256 spa->spa_config_generation++; 1257 } 1258 1259 if (spa_is_root(spa)) 1260 vdev_rele(spa->spa_root_vdev); 1261 1262 ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL); 1263 spa_config_exit(spa, spa->spa_vdev_locks, spa); 1264 1265 /* 1266 * If anything changed, wait for it to sync. This ensures that, 1267 * from the system administrator's perspective, zpool(1M) commands 1268 * are synchronous. This is important for things like zpool offline: 1269 * when the command completes, you expect no further I/O from ZFS. 1270 */ 1271 if (vd != NULL) 1272 txg_wait_synced(spa->spa_dsl_pool, 0); 1273 1274 /* 1275 * If the config changed, update the config cache. 1276 */ 1277 if (config_changed) { 1278 mutex_enter(&spa_namespace_lock); 1279 spa_config_sync(spa, B_FALSE, B_TRUE); 1280 mutex_exit(&spa_namespace_lock); 1281 } 1282 1283 return (error); 1284} 1285 1286/* 1287 * ========================================================================== 1288 * Miscellaneous functions 1289 * ========================================================================== 1290 */ 1291 1292void 1293spa_activate_mos_feature(spa_t *spa, const char *feature, dmu_tx_t *tx) 1294{ 1295 if (!nvlist_exists(spa->spa_label_features, feature)) { 1296 fnvlist_add_boolean(spa->spa_label_features, feature); 1297 /* 1298 * When we are creating the pool (tx_txg==TXG_INITIAL), we can't 1299 * dirty the vdev config because lock SCL_CONFIG is not held. 1300 * Thankfully, in this case we don't need to dirty the config 1301 * because it will be written out anyway when we finish 1302 * creating the pool. 1303 */ 1304 if (tx->tx_txg != TXG_INITIAL) 1305 vdev_config_dirty(spa->spa_root_vdev); 1306 } 1307} 1308 1309void 1310spa_deactivate_mos_feature(spa_t *spa, const char *feature) 1311{ 1312 if (nvlist_remove_all(spa->spa_label_features, feature) == 0) 1313 vdev_config_dirty(spa->spa_root_vdev); 1314} 1315 1316/* 1317 * Rename a spa_t. 1318 */ 1319int 1320spa_rename(const char *name, const char *newname) 1321{ 1322 spa_t *spa; 1323 int err; 1324 1325 /* 1326 * Lookup the spa_t and grab the config lock for writing. We need to 1327 * actually open the pool so that we can sync out the necessary labels. 1328 * It's OK to call spa_open() with the namespace lock held because we 1329 * allow recursive calls for other reasons. 1330 */ 1331 mutex_enter(&spa_namespace_lock); 1332 if ((err = spa_open(name, &spa, FTAG)) != 0) { 1333 mutex_exit(&spa_namespace_lock); 1334 return (err); 1335 } 1336 1337 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1338 1339 avl_remove(&spa_namespace_avl, spa); 1340 (void) strlcpy(spa->spa_name, newname, sizeof (spa->spa_name)); 1341 avl_add(&spa_namespace_avl, spa); 1342 1343 /* 1344 * Sync all labels to disk with the new names by marking the root vdev 1345 * dirty and waiting for it to sync. It will pick up the new pool name 1346 * during the sync. 1347 */ 1348 vdev_config_dirty(spa->spa_root_vdev); 1349 1350 spa_config_exit(spa, SCL_ALL, FTAG); 1351 1352 txg_wait_synced(spa->spa_dsl_pool, 0); 1353 1354 /* 1355 * Sync the updated config cache. 1356 */ 1357 spa_config_sync(spa, B_FALSE, B_TRUE); 1358 1359 spa_close(spa, FTAG); 1360 1361 mutex_exit(&spa_namespace_lock); 1362 1363 return (0); 1364} 1365 1366/* 1367 * Return the spa_t associated with given pool_guid, if it exists. If 1368 * device_guid is non-zero, determine whether the pool exists *and* contains 1369 * a device with the specified device_guid. 1370 */ 1371spa_t * 1372spa_by_guid(uint64_t pool_guid, uint64_t device_guid) 1373{ 1374 spa_t *spa; 1375 avl_tree_t *t = &spa_namespace_avl; 1376 1377 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1378 1379 for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) { 1380 if (spa->spa_state == POOL_STATE_UNINITIALIZED) 1381 continue; 1382 if (spa->spa_root_vdev == NULL) 1383 continue; 1384 if (spa_guid(spa) == pool_guid) { 1385 if (device_guid == 0) 1386 break; 1387 1388 if (vdev_lookup_by_guid(spa->spa_root_vdev, 1389 device_guid) != NULL) 1390 break; 1391 1392 /* 1393 * Check any devices we may be in the process of adding. 1394 */ 1395 if (spa->spa_pending_vdev) { 1396 if (vdev_lookup_by_guid(spa->spa_pending_vdev, 1397 device_guid) != NULL) 1398 break; 1399 } 1400 } 1401 } 1402 1403 return (spa); 1404} 1405 1406/* 1407 * Determine whether a pool with the given pool_guid exists. 1408 */ 1409boolean_t 1410spa_guid_exists(uint64_t pool_guid, uint64_t device_guid) 1411{ 1412 return (spa_by_guid(pool_guid, device_guid) != NULL); 1413} 1414 1415char * 1416spa_strdup(const char *s) 1417{ 1418 size_t len; 1419 char *new; 1420 1421 len = strlen(s); 1422 new = kmem_alloc(len + 1, KM_SLEEP); 1423 bcopy(s, new, len); 1424 new[len] = '\0'; 1425 1426 return (new); 1427} 1428 1429void 1430spa_strfree(char *s) 1431{ 1432 kmem_free(s, strlen(s) + 1); 1433} 1434 1435uint64_t 1436spa_get_random(uint64_t range) 1437{ 1438 uint64_t r; 1439 1440 ASSERT(range != 0); 1441 1442 (void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t)); 1443 1444 return (r % range); 1445} 1446 1447uint64_t 1448spa_generate_guid(spa_t *spa) 1449{ 1450 uint64_t guid = spa_get_random(-1ULL); 1451 1452 if (spa != NULL) { 1453 while (guid == 0 || spa_guid_exists(spa_guid(spa), guid)) 1454 guid = spa_get_random(-1ULL); 1455 } else { 1456 while (guid == 0 || spa_guid_exists(guid, 0)) 1457 guid = spa_get_random(-1ULL); 1458 } 1459 1460 return (guid); 1461} 1462 1463void 1464snprintf_blkptr(char *buf, size_t buflen, const blkptr_t *bp) 1465{ 1466 char type[256]; 1467 char *checksum = NULL; 1468 char *compress = NULL; 1469 1470 if (bp != NULL) { 1471 if (BP_GET_TYPE(bp) & DMU_OT_NEWTYPE) { 1472 dmu_object_byteswap_t bswap = 1473 DMU_OT_BYTESWAP(BP_GET_TYPE(bp)); 1474 (void) snprintf(type, sizeof (type), "bswap %s %s", 1475 DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) ? 1476 "metadata" : "data", 1477 dmu_ot_byteswap[bswap].ob_name); 1478 } else { 1479 (void) strlcpy(type, dmu_ot[BP_GET_TYPE(bp)].ot_name, 1480 sizeof (type)); 1481 } 1482 if (!BP_IS_EMBEDDED(bp)) { 1483 checksum = 1484 zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name; 1485 } 1486 compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name; 1487 } 1488 1489 SNPRINTF_BLKPTR(snprintf, ' ', buf, buflen, bp, type, checksum, 1490 compress); 1491} 1492 1493void 1494spa_freeze(spa_t *spa) 1495{ 1496 uint64_t freeze_txg = 0; 1497 1498 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1499 if (spa->spa_freeze_txg == UINT64_MAX) { 1500 freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE; 1501 spa->spa_freeze_txg = freeze_txg; 1502 } 1503 spa_config_exit(spa, SCL_ALL, FTAG); 1504 if (freeze_txg != 0) 1505 txg_wait_synced(spa_get_dsl(spa), freeze_txg); 1506} 1507 1508void 1509zfs_panic_recover(const char *fmt, ...) 1510{ 1511 va_list adx; 1512 1513 va_start(adx, fmt); 1514 vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx); 1515 va_end(adx); 1516} 1517 1518/* 1519 * This is a stripped-down version of strtoull, suitable only for converting 1520 * lowercase hexadecimal numbers that don't overflow. 1521 */ 1522uint64_t 1523zfs_strtonum(const char *str, char **nptr) 1524{ 1525 uint64_t val = 0; 1526 char c; 1527 int digit; 1528 1529 while ((c = *str) != '\0') { 1530 if (c >= '0' && c <= '9') 1531 digit = c - '0'; 1532 else if (c >= 'a' && c <= 'f') 1533 digit = 10 + c - 'a'; 1534 else 1535 break; 1536 1537 val *= 16; 1538 val += digit; 1539 1540 str++; 1541 } 1542 1543 if (nptr) 1544 *nptr = (char *)str; 1545 1546 return (val); 1547} 1548 1549/* 1550 * ========================================================================== 1551 * Accessor functions 1552 * ========================================================================== 1553 */ 1554 1555boolean_t 1556spa_shutting_down(spa_t *spa) 1557{ 1558 return (spa->spa_async_suspended); 1559} 1560 1561dsl_pool_t * 1562spa_get_dsl(spa_t *spa) 1563{ 1564 return (spa->spa_dsl_pool); 1565} 1566 1567boolean_t 1568spa_is_initializing(spa_t *spa) 1569{ 1570 return (spa->spa_is_initializing); 1571} 1572 1573blkptr_t * 1574spa_get_rootblkptr(spa_t *spa) 1575{ 1576 return (&spa->spa_ubsync.ub_rootbp); 1577} 1578 1579void 1580spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp) 1581{ 1582 spa->spa_uberblock.ub_rootbp = *bp; 1583} 1584 1585void 1586spa_altroot(spa_t *spa, char *buf, size_t buflen) 1587{ 1588 if (spa->spa_root == NULL) 1589 buf[0] = '\0'; 1590 else 1591 (void) strncpy(buf, spa->spa_root, buflen); 1592} 1593 1594int 1595spa_sync_pass(spa_t *spa) 1596{ 1597 return (spa->spa_sync_pass); 1598} 1599 1600char * 1601spa_name(spa_t *spa) 1602{ 1603 return (spa->spa_name); 1604} 1605 1606uint64_t 1607spa_guid(spa_t *spa) 1608{ 1609 dsl_pool_t *dp = spa_get_dsl(spa); 1610 uint64_t guid; 1611 1612 /* 1613 * If we fail to parse the config during spa_load(), we can go through 1614 * the error path (which posts an ereport) and end up here with no root 1615 * vdev. We stash the original pool guid in 'spa_config_guid' to handle 1616 * this case. 1617 */ 1618 if (spa->spa_root_vdev == NULL) 1619 return (spa->spa_config_guid); 1620 1621 guid = spa->spa_last_synced_guid != 0 ? 1622 spa->spa_last_synced_guid : spa->spa_root_vdev->vdev_guid; 1623 1624 /* 1625 * Return the most recently synced out guid unless we're 1626 * in syncing context. 1627 */ 1628 if (dp && dsl_pool_sync_context(dp)) 1629 return (spa->spa_root_vdev->vdev_guid); 1630 else 1631 return (guid); 1632} 1633 1634uint64_t 1635spa_load_guid(spa_t *spa) 1636{ 1637 /* 1638 * This is a GUID that exists solely as a reference for the 1639 * purposes of the arc. It is generated at load time, and 1640 * is never written to persistent storage. 1641 */ 1642 return (spa->spa_load_guid); 1643} 1644 1645uint64_t 1646spa_last_synced_txg(spa_t *spa) 1647{ 1648 return (spa->spa_ubsync.ub_txg); 1649} 1650 1651uint64_t 1652spa_first_txg(spa_t *spa) 1653{ 1654 return (spa->spa_first_txg); 1655} 1656 1657uint64_t 1658spa_syncing_txg(spa_t *spa) 1659{ 1660 return (spa->spa_syncing_txg); 1661} 1662 1663pool_state_t 1664spa_state(spa_t *spa) 1665{ 1666 return (spa->spa_state); 1667} 1668 1669spa_load_state_t 1670spa_load_state(spa_t *spa) 1671{ 1672 return (spa->spa_load_state); 1673} 1674 1675uint64_t 1676spa_freeze_txg(spa_t *spa) 1677{ 1678 return (spa->spa_freeze_txg); 1679} 1680 1681/* ARGSUSED */ 1682uint64_t 1683spa_get_asize(spa_t *spa, uint64_t lsize) 1684{ 1685 return (lsize * spa_asize_inflation); 1686} 1687 1688/* 1689 * Return the amount of slop space in bytes. It is 1/32 of the pool (3.2%), 1690 * or at least 32MB. 1691 * 1692 * See the comment above spa_slop_shift for details. 1693 */ 1694uint64_t 1695spa_get_slop_space(spa_t *spa) { 1696 uint64_t space = spa_get_dspace(spa); 1697 return (MAX(space >> spa_slop_shift, SPA_MINDEVSIZE >> 1)); 1698} 1699 1700uint64_t 1701spa_get_dspace(spa_t *spa) 1702{ 1703 return (spa->spa_dspace); 1704} 1705 1706void 1707spa_update_dspace(spa_t *spa) 1708{ 1709 spa->spa_dspace = metaslab_class_get_dspace(spa_normal_class(spa)) + 1710 ddt_get_dedup_dspace(spa); 1711} 1712 1713/* 1714 * Return the failure mode that has been set to this pool. The default 1715 * behavior will be to block all I/Os when a complete failure occurs. 1716 */ 1717uint8_t 1718spa_get_failmode(spa_t *spa) 1719{ 1720 return (spa->spa_failmode); 1721} 1722 1723boolean_t 1724spa_suspended(spa_t *spa) 1725{ 1726 return (spa->spa_suspended); 1727} 1728 1729uint64_t 1730spa_version(spa_t *spa) 1731{ 1732 return (spa->spa_ubsync.ub_version); 1733} 1734 1735boolean_t 1736spa_deflate(spa_t *spa) 1737{ 1738 return (spa->spa_deflate); 1739} 1740 1741metaslab_class_t * 1742spa_normal_class(spa_t *spa) 1743{ 1744 return (spa->spa_normal_class); 1745} 1746 1747metaslab_class_t * 1748spa_log_class(spa_t *spa) 1749{ 1750 return (spa->spa_log_class); 1751} 1752 1753void 1754spa_evicting_os_register(spa_t *spa, objset_t *os) 1755{ 1756 mutex_enter(&spa->spa_evicting_os_lock); 1757 list_insert_head(&spa->spa_evicting_os_list, os); 1758 mutex_exit(&spa->spa_evicting_os_lock); 1759} 1760 1761void 1762spa_evicting_os_deregister(spa_t *spa, objset_t *os) 1763{ 1764 mutex_enter(&spa->spa_evicting_os_lock); 1765 list_remove(&spa->spa_evicting_os_list, os); 1766 cv_broadcast(&spa->spa_evicting_os_cv); 1767 mutex_exit(&spa->spa_evicting_os_lock); 1768} 1769 1770void 1771spa_evicting_os_wait(spa_t *spa) 1772{ 1773 mutex_enter(&spa->spa_evicting_os_lock); 1774 while (!list_is_empty(&spa->spa_evicting_os_list)) 1775 cv_wait(&spa->spa_evicting_os_cv, &spa->spa_evicting_os_lock); 1776 mutex_exit(&spa->spa_evicting_os_lock); 1777 1778 dmu_buf_user_evict_wait(); 1779} 1780 1781int 1782spa_max_replication(spa_t *spa) 1783{ 1784 /* 1785 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to 1786 * handle BPs with more than one DVA allocated. Set our max 1787 * replication level accordingly. 1788 */ 1789 if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS) 1790 return (1); 1791 return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override)); 1792} 1793 1794int 1795spa_prev_software_version(spa_t *spa) 1796{ 1797 return (spa->spa_prev_software_version); 1798} 1799 1800uint64_t 1801spa_deadman_synctime(spa_t *spa) 1802{ 1803 return (spa->spa_deadman_synctime); 1804} 1805 1806uint64_t 1807dva_get_dsize_sync(spa_t *spa, const dva_t *dva) 1808{ 1809 uint64_t asize = DVA_GET_ASIZE(dva); 1810 uint64_t dsize = asize; 1811 1812 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 1813 1814 if (asize != 0 && spa->spa_deflate) { 1815 uint64_t vdev = DVA_GET_VDEV(dva); 1816 vdev_t *vd = vdev_lookup_top(spa, vdev); 1817 if (vd == NULL) { 1818 panic( 1819 "dva_get_dsize_sync(): bad DVA %llu:%llu", 1820 (u_longlong_t)vdev, (u_longlong_t)asize); 1821 } 1822 dsize = (asize >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio; 1823 } 1824 1825 return (dsize); 1826} 1827 1828uint64_t 1829bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp) 1830{ 1831 uint64_t dsize = 0; 1832 1833 for (int d = 0; d < BP_GET_NDVAS(bp); d++) 1834 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]); 1835 1836 return (dsize); 1837} 1838 1839uint64_t 1840bp_get_dsize(spa_t *spa, const blkptr_t *bp) 1841{ 1842 uint64_t dsize = 0; 1843 1844 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); 1845 1846 for (int d = 0; d < BP_GET_NDVAS(bp); d++) 1847 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]); 1848 1849 spa_config_exit(spa, SCL_VDEV, FTAG); 1850 1851 return (dsize); 1852} 1853 1854/* 1855 * ========================================================================== 1856 * Initialization and Termination 1857 * ========================================================================== 1858 */ 1859 1860static int 1861spa_name_compare(const void *a1, const void *a2) 1862{ 1863 const spa_t *s1 = a1; 1864 const spa_t *s2 = a2; 1865 int s; 1866 1867 s = strcmp(s1->spa_name, s2->spa_name); 1868 if (s > 0) 1869 return (1); 1870 if (s < 0) 1871 return (-1); 1872 return (0); 1873} 1874 1875int 1876spa_busy(void) 1877{ 1878 return (spa_active_count); 1879} 1880 1881void 1882spa_boot_init() 1883{ 1884 spa_config_load(); 1885} 1886 1887#ifdef _KERNEL 1888EVENTHANDLER_DEFINE(mountroot, spa_boot_init, NULL, 0); 1889#endif 1890 1891void 1892spa_init(int mode) 1893{ 1894 mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL); 1895 mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL); 1896 mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL); 1897 cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL); 1898 1899 avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t), 1900 offsetof(spa_t, spa_avl)); 1901 1902 avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t), 1903 offsetof(spa_aux_t, aux_avl)); 1904 1905 avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t), 1906 offsetof(spa_aux_t, aux_avl)); 1907 1908 spa_mode_global = mode; 1909 1910#ifdef illumos 1911#ifdef _KERNEL 1912 spa_arch_init(); 1913#else 1914 if (spa_mode_global != FREAD && dprintf_find_string("watch")) { 1915 arc_procfd = open("/proc/self/ctl", O_WRONLY); 1916 if (arc_procfd == -1) { 1917 perror("could not enable watchpoints: " 1918 "opening /proc/self/ctl failed: "); 1919 } else { 1920 arc_watch = B_TRUE; 1921 } 1922 } 1923#endif 1924#endif /* illumos */ 1925 refcount_sysinit(); 1926 unique_init(); 1927 range_tree_init(); 1928 zio_init(); 1929 lz4_init(); 1930 dmu_init(); 1931 zil_init(); 1932 vdev_cache_stat_init(); 1933 zfs_prop_init(); 1934 zpool_prop_init(); 1935 zpool_feature_init(); 1936 spa_config_load(); 1937 l2arc_start(); 1938#ifndef illumos 1939#ifdef _KERNEL 1940 zfs_deadman_init(); 1941#endif 1942#endif /* !illumos */ 1943} 1944 1945void 1946spa_fini(void) 1947{ 1948 l2arc_stop(); 1949 1950 spa_evict_all(); 1951 1952 vdev_cache_stat_fini(); 1953 zil_fini(); 1954 dmu_fini(); 1955 lz4_fini(); 1956 zio_fini(); 1957 range_tree_fini(); 1958 unique_fini(); 1959 refcount_fini(); 1960 1961 avl_destroy(&spa_namespace_avl); 1962 avl_destroy(&spa_spare_avl); 1963 avl_destroy(&spa_l2cache_avl); 1964 1965 cv_destroy(&spa_namespace_cv); 1966 mutex_destroy(&spa_namespace_lock); 1967 mutex_destroy(&spa_spare_lock); 1968 mutex_destroy(&spa_l2cache_lock); 1969} 1970 1971/* 1972 * Return whether this pool has slogs. No locking needed. 1973 * It's not a problem if the wrong answer is returned as it's only for 1974 * performance and not correctness 1975 */ 1976boolean_t 1977spa_has_slogs(spa_t *spa) 1978{ 1979 return (spa->spa_log_class->mc_rotor != NULL); 1980} 1981 1982spa_log_state_t 1983spa_get_log_state(spa_t *spa) 1984{ 1985 return (spa->spa_log_state); 1986} 1987 1988void 1989spa_set_log_state(spa_t *spa, spa_log_state_t state) 1990{ 1991 spa->spa_log_state = state; 1992} 1993 1994boolean_t 1995spa_is_root(spa_t *spa) 1996{ 1997 return (spa->spa_is_root); 1998} 1999 2000boolean_t 2001spa_writeable(spa_t *spa) 2002{ 2003 return (!!(spa->spa_mode & FWRITE)); 2004} 2005 2006/* 2007 * Returns true if there is a pending sync task in any of the current 2008 * syncing txg, the current quiescing txg, or the current open txg. 2009 */ 2010boolean_t 2011spa_has_pending_synctask(spa_t *spa) 2012{ 2013 return (!txg_all_lists_empty(&spa->spa_dsl_pool->dp_sync_tasks)); 2014} 2015 2016int 2017spa_mode(spa_t *spa) 2018{ 2019 return (spa->spa_mode); 2020} 2021 2022uint64_t 2023spa_bootfs(spa_t *spa) 2024{ 2025 return (spa->spa_bootfs); 2026} 2027 2028uint64_t 2029spa_delegation(spa_t *spa) 2030{ 2031 return (spa->spa_delegation); 2032} 2033 2034objset_t * 2035spa_meta_objset(spa_t *spa) 2036{ 2037 return (spa->spa_meta_objset); 2038} 2039 2040enum zio_checksum 2041spa_dedup_checksum(spa_t *spa) 2042{ 2043 return (spa->spa_dedup_checksum); 2044} 2045 2046/* 2047 * Reset pool scan stat per scan pass (or reboot). 2048 */ 2049void 2050spa_scan_stat_init(spa_t *spa) 2051{ 2052 /* data not stored on disk */ 2053 spa->spa_scan_pass_start = gethrestime_sec(); 2054 spa->spa_scan_pass_exam = 0; 2055 vdev_scan_stat_init(spa->spa_root_vdev); 2056} 2057 2058/* 2059 * Get scan stats for zpool status reports 2060 */ 2061int 2062spa_scan_get_stats(spa_t *spa, pool_scan_stat_t *ps) 2063{ 2064 dsl_scan_t *scn = spa->spa_dsl_pool ? spa->spa_dsl_pool->dp_scan : NULL; 2065 2066 if (scn == NULL || scn->scn_phys.scn_func == POOL_SCAN_NONE) 2067 return (SET_ERROR(ENOENT)); 2068 bzero(ps, sizeof (pool_scan_stat_t)); 2069 2070 /* data stored on disk */ 2071 ps->pss_func = scn->scn_phys.scn_func; 2072 ps->pss_start_time = scn->scn_phys.scn_start_time; 2073 ps->pss_end_time = scn->scn_phys.scn_end_time; 2074 ps->pss_to_examine = scn->scn_phys.scn_to_examine; 2075 ps->pss_examined = scn->scn_phys.scn_examined; 2076 ps->pss_to_process = scn->scn_phys.scn_to_process; 2077 ps->pss_processed = scn->scn_phys.scn_processed; 2078 ps->pss_errors = scn->scn_phys.scn_errors; 2079 ps->pss_state = scn->scn_phys.scn_state; 2080 2081 /* data not stored on disk */ 2082 ps->pss_pass_start = spa->spa_scan_pass_start; 2083 ps->pss_pass_exam = spa->spa_scan_pass_exam; 2084 2085 return (0); 2086} 2087 2088boolean_t 2089spa_debug_enabled(spa_t *spa) 2090{ 2091 return (spa->spa_debug); 2092} 2093 2094int 2095spa_maxblocksize(spa_t *spa) 2096{ 2097 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) 2098 return (SPA_MAXBLOCKSIZE); 2099 else 2100 return (SPA_OLD_MAXBLOCKSIZE); 2101} 2102