spa_misc.c revision 297112
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 2015 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 * Copyright (c) 2014 Integros [integros.com] 29 */ 30 31#include <sys/zfs_context.h> 32#include <sys/spa_impl.h> 33#include <sys/spa_boot.h> 34#include <sys/zio.h> 35#include <sys/zio_checksum.h> 36#include <sys/zio_compress.h> 37#include <sys/dmu.h> 38#include <sys/dmu_tx.h> 39#include <sys/zap.h> 40#include <sys/zil.h> 41#include <sys/vdev_impl.h> 42#include <sys/metaslab.h> 43#include <sys/uberblock_impl.h> 44#include <sys/txg.h> 45#include <sys/avl.h> 46#include <sys/unique.h> 47#include <sys/dsl_pool.h> 48#include <sys/dsl_dir.h> 49#include <sys/dsl_prop.h> 50#include <sys/dsl_scan.h> 51#include <sys/fs/zfs.h> 52#include <sys/metaslab_impl.h> 53#include <sys/arc.h> 54#include <sys/ddt.h> 55#include "zfs_prop.h" 56#include <sys/zfeature.h> 57 58/* 59 * SPA locking 60 * 61 * There are four basic locks for managing spa_t structures: 62 * 63 * spa_namespace_lock (global mutex) 64 * 65 * This lock must be acquired to do any of the following: 66 * 67 * - Lookup a spa_t by name 68 * - Add or remove a spa_t from the namespace 69 * - Increase spa_refcount from non-zero 70 * - Check if spa_refcount is zero 71 * - Rename a spa_t 72 * - add/remove/attach/detach devices 73 * - Held for the duration of create/destroy/import/export 74 * 75 * It does not need to handle recursion. A create or destroy may 76 * reference objects (files or zvols) in other pools, but by 77 * definition they must have an existing reference, and will never need 78 * to lookup a spa_t by name. 79 * 80 * spa_refcount (per-spa refcount_t protected by mutex) 81 * 82 * This reference count keep track of any active users of the spa_t. The 83 * spa_t cannot be destroyed or freed while this is non-zero. Internally, 84 * the refcount is never really 'zero' - opening a pool implicitly keeps 85 * some references in the DMU. Internally we check against spa_minref, but 86 * present the image of a zero/non-zero value to consumers. 87 * 88 * spa_config_lock[] (per-spa array of rwlocks) 89 * 90 * This protects the spa_t from config changes, and must be held in 91 * the following circumstances: 92 * 93 * - RW_READER to perform I/O to the spa 94 * - RW_WRITER to change the vdev config 95 * 96 * The locking order is fairly straightforward: 97 * 98 * spa_namespace_lock -> spa_refcount 99 * 100 * The namespace lock must be acquired to increase the refcount from 0 101 * or to check if it is zero. 102 * 103 * spa_refcount -> spa_config_lock[] 104 * 105 * There must be at least one valid reference on the spa_t to acquire 106 * the config lock. 107 * 108 * spa_namespace_lock -> spa_config_lock[] 109 * 110 * The namespace lock must always be taken before the config lock. 111 * 112 * 113 * The spa_namespace_lock can be acquired directly and is globally visible. 114 * 115 * The namespace is manipulated using the following functions, all of which 116 * require the spa_namespace_lock to be held. 117 * 118 * spa_lookup() Lookup a spa_t by name. 119 * 120 * spa_add() Create a new spa_t in the namespace. 121 * 122 * spa_remove() Remove a spa_t from the namespace. This also 123 * frees up any memory associated with the spa_t. 124 * 125 * spa_next() Returns the next spa_t in the system, or the 126 * first if NULL is passed. 127 * 128 * spa_evict_all() Shutdown and remove all spa_t structures in 129 * the system. 130 * 131 * spa_guid_exists() Determine whether a pool/device guid exists. 132 * 133 * The spa_refcount is manipulated using the following functions: 134 * 135 * spa_open_ref() Adds a reference to the given spa_t. Must be 136 * called with spa_namespace_lock held if the 137 * refcount is currently zero. 138 * 139 * spa_close() Remove a reference from the spa_t. This will 140 * not free the spa_t or remove it from the 141 * namespace. No locking is required. 142 * 143 * spa_refcount_zero() Returns true if the refcount is currently 144 * zero. Must be called with spa_namespace_lock 145 * held. 146 * 147 * The spa_config_lock[] is an array of rwlocks, ordered as follows: 148 * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV. 149 * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}(). 150 * 151 * To read the configuration, it suffices to hold one of these locks as reader. 152 * To modify the configuration, you must hold all locks as writer. To modify 153 * vdev state without altering the vdev tree's topology (e.g. online/offline), 154 * you must hold SCL_STATE and SCL_ZIO as writer. 155 * 156 * We use these distinct config locks to avoid recursive lock entry. 157 * For example, spa_sync() (which holds SCL_CONFIG as reader) induces 158 * block allocations (SCL_ALLOC), which may require reading space maps 159 * from disk (dmu_read() -> zio_read() -> SCL_ZIO). 160 * 161 * The spa config locks cannot be normal rwlocks because we need the 162 * ability to hand off ownership. For example, SCL_ZIO is acquired 163 * by the issuing thread and later released by an interrupt thread. 164 * They do, however, obey the usual write-wanted semantics to prevent 165 * writer (i.e. system administrator) starvation. 166 * 167 * The lock acquisition rules are as follows: 168 * 169 * SCL_CONFIG 170 * Protects changes to the vdev tree topology, such as vdev 171 * add/remove/attach/detach. Protects the dirty config list 172 * (spa_config_dirty_list) and the set of spares and l2arc devices. 173 * 174 * SCL_STATE 175 * Protects changes to pool state and vdev state, such as vdev 176 * online/offline/fault/degrade/clear. Protects the dirty state list 177 * (spa_state_dirty_list) and global pool state (spa_state). 178 * 179 * SCL_ALLOC 180 * Protects changes to metaslab groups and classes. 181 * Held as reader by metaslab_alloc() and metaslab_claim(). 182 * 183 * SCL_ZIO 184 * Held by bp-level zios (those which have no io_vd upon entry) 185 * to prevent changes to the vdev tree. The bp-level zio implicitly 186 * protects all of its vdev child zios, which do not hold SCL_ZIO. 187 * 188 * SCL_FREE 189 * Protects changes to metaslab groups and classes. 190 * Held as reader by metaslab_free(). SCL_FREE is distinct from 191 * SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free 192 * blocks in zio_done() while another i/o that holds either 193 * SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete. 194 * 195 * SCL_VDEV 196 * Held as reader to prevent changes to the vdev tree during trivial 197 * inquiries such as bp_get_dsize(). SCL_VDEV is distinct from the 198 * other locks, and lower than all of them, to ensure that it's safe 199 * to acquire regardless of caller context. 200 * 201 * In addition, the following rules apply: 202 * 203 * (a) spa_props_lock protects pool properties, spa_config and spa_config_list. 204 * The lock ordering is SCL_CONFIG > spa_props_lock. 205 * 206 * (b) I/O operations on leaf vdevs. For any zio operation that takes 207 * an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(), 208 * or zio_write_phys() -- the caller must ensure that the config cannot 209 * cannot change in the interim, and that the vdev cannot be reopened. 210 * SCL_STATE as reader suffices for both. 211 * 212 * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit(). 213 * 214 * spa_vdev_enter() Acquire the namespace lock and the config lock 215 * for writing. 216 * 217 * spa_vdev_exit() Release the config lock, wait for all I/O 218 * to complete, sync the updated configs to the 219 * cache, and release the namespace lock. 220 * 221 * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit(). 222 * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual 223 * locking is, always, based on spa_namespace_lock and spa_config_lock[]. 224 * 225 * spa_rename() is also implemented within this file since it requires 226 * manipulation of the namespace. 227 */ 228 229static avl_tree_t spa_namespace_avl; 230kmutex_t spa_namespace_lock; 231static kcondvar_t spa_namespace_cv; 232static int spa_active_count; 233int spa_max_replication_override = SPA_DVAS_PER_BP; 234 235static kmutex_t spa_spare_lock; 236static avl_tree_t spa_spare_avl; 237static kmutex_t spa_l2cache_lock; 238static avl_tree_t spa_l2cache_avl; 239 240kmem_cache_t *spa_buffer_pool; 241int spa_mode_global; 242 243#ifdef ZFS_DEBUG 244/* Everything except dprintf and spa is on by default in debug builds */ 245int zfs_flags = ~(ZFS_DEBUG_DPRINTF | ZFS_DEBUG_SPA); 246#else 247int zfs_flags = 0; 248#endif 249SYSCTL_DECL(_debug); 250TUNABLE_INT("debug.zfs_flags", &zfs_flags); 251SYSCTL_INT(_debug, OID_AUTO, zfs_flags, CTLFLAG_RWTUN, &zfs_flags, 0, 252 "ZFS debug flags."); 253 254/* 255 * zfs_recover can be set to nonzero to attempt to recover from 256 * otherwise-fatal errors, typically caused by on-disk corruption. When 257 * set, calls to zfs_panic_recover() will turn into warning messages. 258 * This should only be used as a last resort, as it typically results 259 * in leaked space, or worse. 260 */ 261boolean_t zfs_recover = B_FALSE; 262SYSCTL_DECL(_vfs_zfs); 263TUNABLE_INT("vfs.zfs.recover", &zfs_recover); 264SYSCTL_INT(_vfs_zfs, OID_AUTO, recover, CTLFLAG_RWTUN, &zfs_recover, 0, 265 "Try to recover from otherwise-fatal errors."); 266 267static int 268sysctl_vfs_zfs_debug_flags(SYSCTL_HANDLER_ARGS) 269{ 270 int err, val; 271 272 val = zfs_flags; 273 err = sysctl_handle_int(oidp, &val, 0, req); 274 if (err != 0 || req->newptr == NULL) 275 return (err); 276 277 /* 278 * ZFS_DEBUG_MODIFY must be enabled prior to boot so all 279 * arc buffers in the system have the necessary additional 280 * checksum data. However, it is safe to disable at any 281 * time. 282 */ 283 if (!(zfs_flags & ZFS_DEBUG_MODIFY)) 284 val &= ~ZFS_DEBUG_MODIFY; 285 zfs_flags = val; 286 287 return (0); 288} 289TUNABLE_INT("vfs.zfs.debug_flags", &zfs_flags); 290SYSCTL_PROC(_vfs_zfs, OID_AUTO, debug_flags, 291 CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(int), 292 sysctl_vfs_zfs_debug_flags, "IU", "Debug flags for ZFS testing."); 293 294/* 295 * If destroy encounters an EIO while reading metadata (e.g. indirect 296 * blocks), space referenced by the missing metadata can not be freed. 297 * Normally this causes the background destroy to become "stalled", as 298 * it is unable to make forward progress. While in this stalled state, 299 * all remaining space to free from the error-encountering filesystem is 300 * "temporarily leaked". Set this flag to cause it to ignore the EIO, 301 * permanently leak the space from indirect blocks that can not be read, 302 * and continue to free everything else that it can. 303 * 304 * The default, "stalling" behavior is useful if the storage partially 305 * fails (i.e. some but not all i/os fail), and then later recovers. In 306 * this case, we will be able to continue pool operations while it is 307 * partially failed, and when it recovers, we can continue to free the 308 * space, with no leaks. However, note that this case is actually 309 * fairly rare. 310 * 311 * Typically pools either (a) fail completely (but perhaps temporarily, 312 * e.g. a top-level vdev going offline), or (b) have localized, 313 * permanent errors (e.g. disk returns the wrong data due to bit flip or 314 * firmware bug). In case (a), this setting does not matter because the 315 * pool will be suspended and the sync thread will not be able to make 316 * forward progress regardless. In case (b), because the error is 317 * permanent, the best we can do is leak the minimum amount of space, 318 * which is what setting this flag will do. Therefore, it is reasonable 319 * for this flag to normally be set, but we chose the more conservative 320 * approach of not setting it, so that there is no possibility of 321 * leaking space in the "partial temporary" failure case. 322 */ 323boolean_t zfs_free_leak_on_eio = B_FALSE; 324 325/* 326 * Expiration time in milliseconds. This value has two meanings. First it is 327 * used to determine when the spa_deadman() logic should fire. By default the 328 * spa_deadman() will fire if spa_sync() has not completed in 1000 seconds. 329 * Secondly, the value determines if an I/O is considered "hung". Any I/O that 330 * has not completed in zfs_deadman_synctime_ms is considered "hung" resulting 331 * in a system panic. 332 */ 333uint64_t zfs_deadman_synctime_ms = 1000000ULL; 334TUNABLE_QUAD("vfs.zfs.deadman_synctime_ms", &zfs_deadman_synctime_ms); 335SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, deadman_synctime_ms, CTLFLAG_RDTUN, 336 &zfs_deadman_synctime_ms, 0, 337 "Stalled ZFS I/O expiration time in milliseconds"); 338 339/* 340 * Check time in milliseconds. This defines the frequency at which we check 341 * for hung I/O. 342 */ 343uint64_t zfs_deadman_checktime_ms = 5000ULL; 344TUNABLE_QUAD("vfs.zfs.deadman_checktime_ms", &zfs_deadman_checktime_ms); 345SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, deadman_checktime_ms, CTLFLAG_RDTUN, 346 &zfs_deadman_checktime_ms, 0, 347 "Period of checks for stalled ZFS I/O in milliseconds"); 348 349/* 350 * Default value of -1 for zfs_deadman_enabled is resolved in 351 * zfs_deadman_init() 352 */ 353int zfs_deadman_enabled = -1; 354TUNABLE_INT("vfs.zfs.deadman_enabled", &zfs_deadman_enabled); 355SYSCTL_INT(_vfs_zfs, OID_AUTO, deadman_enabled, CTLFLAG_RDTUN, 356 &zfs_deadman_enabled, 0, "Kernel panic on stalled ZFS I/O"); 357 358/* 359 * The worst case is single-sector max-parity RAID-Z blocks, in which 360 * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1) 361 * times the size; so just assume that. Add to this the fact that 362 * we can have up to 3 DVAs per bp, and one more factor of 2 because 363 * the block may be dittoed with up to 3 DVAs by ddt_sync(). All together, 364 * the worst case is: 365 * (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2 == 24 366 */ 367int spa_asize_inflation = 24; 368TUNABLE_INT("vfs.zfs.spa_asize_inflation", &spa_asize_inflation); 369SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_asize_inflation, CTLFLAG_RWTUN, 370 &spa_asize_inflation, 0, "Worst case inflation factor for single sector writes"); 371 372#ifndef illumos 373#ifdef _KERNEL 374static void 375zfs_deadman_init() 376{ 377 /* 378 * If we are not i386 or amd64 or in a virtual machine, 379 * disable ZFS deadman thread by default 380 */ 381 if (zfs_deadman_enabled == -1) { 382#if defined(__amd64__) || defined(__i386__) 383 zfs_deadman_enabled = (vm_guest == VM_GUEST_NO) ? 1 : 0; 384#else 385 zfs_deadman_enabled = 0; 386#endif 387 } 388} 389#endif /* _KERNEL */ 390#endif /* !illumos */ 391 392/* 393 * Normally, we don't allow the last 3.2% (1/(2^spa_slop_shift)) of space in 394 * the pool to be consumed. This ensures that we don't run the pool 395 * completely out of space, due to unaccounted changes (e.g. to the MOS). 396 * It also limits the worst-case time to allocate space. If we have 397 * less than this amount of free space, most ZPL operations (e.g. write, 398 * create) will return ENOSPC. 399 * 400 * Certain operations (e.g. file removal, most administrative actions) can 401 * use half the slop space. They will only return ENOSPC if less than half 402 * the slop space is free. Typically, once the pool has less than the slop 403 * space free, the user will use these operations to free up space in the pool. 404 * These are the operations that call dsl_pool_adjustedsize() with the netfree 405 * argument set to TRUE. 406 * 407 * A very restricted set of operations are always permitted, regardless of 408 * the amount of free space. These are the operations that call 409 * dsl_sync_task(ZFS_SPACE_CHECK_NONE), e.g. "zfs destroy". If these 410 * operations result in a net increase in the amount of space used, 411 * it is possible to run the pool completely out of space, causing it to 412 * be permanently read-only. 413 * 414 * See also the comments in zfs_space_check_t. 415 */ 416int spa_slop_shift = 5; 417SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_slop_shift, CTLFLAG_RWTUN, 418 &spa_slop_shift, 0, 419 "Shift value of reserved space (1/(2^spa_slop_shift))."); 420 421/* 422 * ========================================================================== 423 * SPA config locking 424 * ========================================================================== 425 */ 426static void 427spa_config_lock_init(spa_t *spa) 428{ 429 for (int i = 0; i < SCL_LOCKS; i++) { 430 spa_config_lock_t *scl = &spa->spa_config_lock[i]; 431 mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL); 432 cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL); 433 refcount_create_untracked(&scl->scl_count); 434 scl->scl_writer = NULL; 435 scl->scl_write_wanted = 0; 436 } 437} 438 439static void 440spa_config_lock_destroy(spa_t *spa) 441{ 442 for (int i = 0; i < SCL_LOCKS; i++) { 443 spa_config_lock_t *scl = &spa->spa_config_lock[i]; 444 mutex_destroy(&scl->scl_lock); 445 cv_destroy(&scl->scl_cv); 446 refcount_destroy(&scl->scl_count); 447 ASSERT(scl->scl_writer == NULL); 448 ASSERT(scl->scl_write_wanted == 0); 449 } 450} 451 452int 453spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw) 454{ 455 for (int i = 0; i < SCL_LOCKS; i++) { 456 spa_config_lock_t *scl = &spa->spa_config_lock[i]; 457 if (!(locks & (1 << i))) 458 continue; 459 mutex_enter(&scl->scl_lock); 460 if (rw == RW_READER) { 461 if (scl->scl_writer || scl->scl_write_wanted) { 462 mutex_exit(&scl->scl_lock); 463 spa_config_exit(spa, locks & ((1 << i) - 1), 464 tag); 465 return (0); 466 } 467 } else { 468 ASSERT(scl->scl_writer != curthread); 469 if (!refcount_is_zero(&scl->scl_count)) { 470 mutex_exit(&scl->scl_lock); 471 spa_config_exit(spa, locks & ((1 << i) - 1), 472 tag); 473 return (0); 474 } 475 scl->scl_writer = curthread; 476 } 477 (void) refcount_add(&scl->scl_count, tag); 478 mutex_exit(&scl->scl_lock); 479 } 480 return (1); 481} 482 483void 484spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw) 485{ 486 int wlocks_held = 0; 487 488 ASSERT3U(SCL_LOCKS, <, sizeof (wlocks_held) * NBBY); 489 490 for (int i = 0; i < SCL_LOCKS; i++) { 491 spa_config_lock_t *scl = &spa->spa_config_lock[i]; 492 if (scl->scl_writer == curthread) 493 wlocks_held |= (1 << i); 494 if (!(locks & (1 << i))) 495 continue; 496 mutex_enter(&scl->scl_lock); 497 if (rw == RW_READER) { 498 while (scl->scl_writer || scl->scl_write_wanted) { 499 cv_wait(&scl->scl_cv, &scl->scl_lock); 500 } 501 } else { 502 ASSERT(scl->scl_writer != curthread); 503 while (!refcount_is_zero(&scl->scl_count)) { 504 scl->scl_write_wanted++; 505 cv_wait(&scl->scl_cv, &scl->scl_lock); 506 scl->scl_write_wanted--; 507 } 508 scl->scl_writer = curthread; 509 } 510 (void) refcount_add(&scl->scl_count, tag); 511 mutex_exit(&scl->scl_lock); 512 } 513 ASSERT(wlocks_held <= locks); 514} 515 516void 517spa_config_exit(spa_t *spa, int locks, void *tag) 518{ 519 for (int i = SCL_LOCKS - 1; i >= 0; i--) { 520 spa_config_lock_t *scl = &spa->spa_config_lock[i]; 521 if (!(locks & (1 << i))) 522 continue; 523 mutex_enter(&scl->scl_lock); 524 ASSERT(!refcount_is_zero(&scl->scl_count)); 525 if (refcount_remove(&scl->scl_count, tag) == 0) { 526 ASSERT(scl->scl_writer == NULL || 527 scl->scl_writer == curthread); 528 scl->scl_writer = NULL; /* OK in either case */ 529 cv_broadcast(&scl->scl_cv); 530 } 531 mutex_exit(&scl->scl_lock); 532 } 533} 534 535int 536spa_config_held(spa_t *spa, int locks, krw_t rw) 537{ 538 int locks_held = 0; 539 540 for (int i = 0; i < SCL_LOCKS; i++) { 541 spa_config_lock_t *scl = &spa->spa_config_lock[i]; 542 if (!(locks & (1 << i))) 543 continue; 544 if ((rw == RW_READER && !refcount_is_zero(&scl->scl_count)) || 545 (rw == RW_WRITER && scl->scl_writer == curthread)) 546 locks_held |= 1 << i; 547 } 548 549 return (locks_held); 550} 551 552/* 553 * ========================================================================== 554 * SPA namespace functions 555 * ========================================================================== 556 */ 557 558/* 559 * Lookup the named spa_t in the AVL tree. The spa_namespace_lock must be held. 560 * Returns NULL if no matching spa_t is found. 561 */ 562spa_t * 563spa_lookup(const char *name) 564{ 565 static spa_t search; /* spa_t is large; don't allocate on stack */ 566 spa_t *spa; 567 avl_index_t where; 568 char *cp; 569 570 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 571 572 (void) strlcpy(search.spa_name, name, sizeof (search.spa_name)); 573 574 /* 575 * If it's a full dataset name, figure out the pool name and 576 * just use that. 577 */ 578 cp = strpbrk(search.spa_name, "/@#"); 579 if (cp != NULL) 580 *cp = '\0'; 581 582 spa = avl_find(&spa_namespace_avl, &search, &where); 583 584 return (spa); 585} 586 587/* 588 * Fires when spa_sync has not completed within zfs_deadman_synctime_ms. 589 * If the zfs_deadman_enabled flag is set then it inspects all vdev queues 590 * looking for potentially hung I/Os. 591 */ 592void 593spa_deadman(void *arg) 594{ 595 spa_t *spa = arg; 596 597 /* 598 * Disable the deadman timer if the pool is suspended. 599 */ 600 if (spa_suspended(spa)) { 601#ifdef illumos 602 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY)); 603#else 604 /* Nothing. just don't schedule any future callouts. */ 605#endif 606 return; 607 } 608 609 zfs_dbgmsg("slow spa_sync: started %llu seconds ago, calls %llu", 610 (gethrtime() - spa->spa_sync_starttime) / NANOSEC, 611 ++spa->spa_deadman_calls); 612 if (zfs_deadman_enabled) 613 vdev_deadman(spa->spa_root_vdev); 614#ifdef __FreeBSD__ 615#ifdef _KERNEL 616 callout_schedule(&spa->spa_deadman_cycid, 617 hz * zfs_deadman_checktime_ms / MILLISEC); 618#endif 619#endif 620} 621 622/* 623 * Create an uninitialized spa_t with the given name. Requires 624 * spa_namespace_lock. The caller must ensure that the spa_t doesn't already 625 * exist by calling spa_lookup() first. 626 */ 627spa_t * 628spa_add(const char *name, nvlist_t *config, const char *altroot) 629{ 630 spa_t *spa; 631 spa_config_dirent_t *dp; 632#ifdef illumos 633 cyc_handler_t hdlr; 634 cyc_time_t when; 635#endif 636 637 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 638 639 spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP); 640 641 mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL); 642 mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL); 643 mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL); 644 mutex_init(&spa->spa_evicting_os_lock, NULL, MUTEX_DEFAULT, NULL); 645 mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL); 646 mutex_init(&spa->spa_proc_lock, NULL, MUTEX_DEFAULT, NULL); 647 mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL); 648 mutex_init(&spa->spa_cksum_tmpls_lock, NULL, MUTEX_DEFAULT, NULL); 649 mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL); 650 mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL); 651 mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL); 652 653 cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL); 654 cv_init(&spa->spa_evicting_os_cv, NULL, CV_DEFAULT, NULL); 655 cv_init(&spa->spa_proc_cv, NULL, CV_DEFAULT, NULL); 656 cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL); 657 cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL); 658 659 for (int t = 0; t < TXG_SIZE; t++) 660 bplist_create(&spa->spa_free_bplist[t]); 661 662 (void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name)); 663 spa->spa_state = POOL_STATE_UNINITIALIZED; 664 spa->spa_freeze_txg = UINT64_MAX; 665 spa->spa_final_txg = UINT64_MAX; 666 spa->spa_load_max_txg = UINT64_MAX; 667 spa->spa_proc = &p0; 668 spa->spa_proc_state = SPA_PROC_NONE; 669 670#ifdef illumos 671 hdlr.cyh_func = spa_deadman; 672 hdlr.cyh_arg = spa; 673 hdlr.cyh_level = CY_LOW_LEVEL; 674#endif 675 676 spa->spa_deadman_synctime = MSEC2NSEC(zfs_deadman_synctime_ms); 677 678#ifdef illumos 679 /* 680 * This determines how often we need to check for hung I/Os after 681 * the cyclic has already fired. Since checking for hung I/Os is 682 * an expensive operation we don't want to check too frequently. 683 * Instead wait for 5 seconds before checking again. 684 */ 685 when.cyt_interval = MSEC2NSEC(zfs_deadman_checktime_ms); 686 when.cyt_when = CY_INFINITY; 687 mutex_enter(&cpu_lock); 688 spa->spa_deadman_cycid = cyclic_add(&hdlr, &when); 689 mutex_exit(&cpu_lock); 690#else /* !illumos */ 691#ifdef _KERNEL 692 callout_init(&spa->spa_deadman_cycid, CALLOUT_MPSAFE); 693#endif 694#endif 695 refcount_create(&spa->spa_refcount); 696 spa_config_lock_init(spa); 697 698 avl_add(&spa_namespace_avl, spa); 699 700 /* 701 * Set the alternate root, if there is one. 702 */ 703 if (altroot) { 704 spa->spa_root = spa_strdup(altroot); 705 spa_active_count++; 706 } 707 708 /* 709 * Every pool starts with the default cachefile 710 */ 711 list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t), 712 offsetof(spa_config_dirent_t, scd_link)); 713 714 dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP); 715 dp->scd_path = altroot ? NULL : spa_strdup(spa_config_path); 716 list_insert_head(&spa->spa_config_list, dp); 717 718 VERIFY(nvlist_alloc(&spa->spa_load_info, NV_UNIQUE_NAME, 719 KM_SLEEP) == 0); 720 721 if (config != NULL) { 722 nvlist_t *features; 723 724 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ, 725 &features) == 0) { 726 VERIFY(nvlist_dup(features, &spa->spa_label_features, 727 0) == 0); 728 } 729 730 VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0); 731 } 732 733 if (spa->spa_label_features == NULL) { 734 VERIFY(nvlist_alloc(&spa->spa_label_features, NV_UNIQUE_NAME, 735 KM_SLEEP) == 0); 736 } 737 738 spa->spa_debug = ((zfs_flags & ZFS_DEBUG_SPA) != 0); 739 740 spa->spa_min_ashift = INT_MAX; 741 spa->spa_max_ashift = 0; 742 743 /* 744 * As a pool is being created, treat all features as disabled by 745 * setting SPA_FEATURE_DISABLED for all entries in the feature 746 * refcount cache. 747 */ 748 for (int i = 0; i < SPA_FEATURES; i++) { 749 spa->spa_feat_refcount_cache[i] = SPA_FEATURE_DISABLED; 750 } 751 752 return (spa); 753} 754 755/* 756 * Removes a spa_t from the namespace, freeing up any memory used. Requires 757 * spa_namespace_lock. This is called only after the spa_t has been closed and 758 * deactivated. 759 */ 760void 761spa_remove(spa_t *spa) 762{ 763 spa_config_dirent_t *dp; 764 765 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 766 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); 767 ASSERT3U(refcount_count(&spa->spa_refcount), ==, 0); 768 769 nvlist_free(spa->spa_config_splitting); 770 771 avl_remove(&spa_namespace_avl, spa); 772 cv_broadcast(&spa_namespace_cv); 773 774 if (spa->spa_root) { 775 spa_strfree(spa->spa_root); 776 spa_active_count--; 777 } 778 779 while ((dp = list_head(&spa->spa_config_list)) != NULL) { 780 list_remove(&spa->spa_config_list, dp); 781 if (dp->scd_path != NULL) 782 spa_strfree(dp->scd_path); 783 kmem_free(dp, sizeof (spa_config_dirent_t)); 784 } 785 786 list_destroy(&spa->spa_config_list); 787 788 nvlist_free(spa->spa_label_features); 789 nvlist_free(spa->spa_load_info); 790 spa_config_set(spa, NULL); 791 792#ifdef illumos 793 mutex_enter(&cpu_lock); 794 if (spa->spa_deadman_cycid != CYCLIC_NONE) 795 cyclic_remove(spa->spa_deadman_cycid); 796 mutex_exit(&cpu_lock); 797 spa->spa_deadman_cycid = CYCLIC_NONE; 798#else /* !illumos */ 799#ifdef _KERNEL 800 callout_drain(&spa->spa_deadman_cycid); 801#endif 802#endif 803 804 refcount_destroy(&spa->spa_refcount); 805 806 spa_config_lock_destroy(spa); 807 808 for (int t = 0; t < TXG_SIZE; t++) 809 bplist_destroy(&spa->spa_free_bplist[t]); 810 811 zio_checksum_templates_free(spa); 812 813 cv_destroy(&spa->spa_async_cv); 814 cv_destroy(&spa->spa_evicting_os_cv); 815 cv_destroy(&spa->spa_proc_cv); 816 cv_destroy(&spa->spa_scrub_io_cv); 817 cv_destroy(&spa->spa_suspend_cv); 818 819 mutex_destroy(&spa->spa_async_lock); 820 mutex_destroy(&spa->spa_errlist_lock); 821 mutex_destroy(&spa->spa_errlog_lock); 822 mutex_destroy(&spa->spa_evicting_os_lock); 823 mutex_destroy(&spa->spa_history_lock); 824 mutex_destroy(&spa->spa_proc_lock); 825 mutex_destroy(&spa->spa_props_lock); 826 mutex_destroy(&spa->spa_cksum_tmpls_lock); 827 mutex_destroy(&spa->spa_scrub_lock); 828 mutex_destroy(&spa->spa_suspend_lock); 829 mutex_destroy(&spa->spa_vdev_top_lock); 830 831 kmem_free(spa, sizeof (spa_t)); 832} 833 834/* 835 * Given a pool, return the next pool in the namespace, or NULL if there is 836 * none. If 'prev' is NULL, return the first pool. 837 */ 838spa_t * 839spa_next(spa_t *prev) 840{ 841 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 842 843 if (prev) 844 return (AVL_NEXT(&spa_namespace_avl, prev)); 845 else 846 return (avl_first(&spa_namespace_avl)); 847} 848 849/* 850 * ========================================================================== 851 * SPA refcount functions 852 * ========================================================================== 853 */ 854 855/* 856 * Add a reference to the given spa_t. Must have at least one reference, or 857 * have the namespace lock held. 858 */ 859void 860spa_open_ref(spa_t *spa, void *tag) 861{ 862 ASSERT(refcount_count(&spa->spa_refcount) >= spa->spa_minref || 863 MUTEX_HELD(&spa_namespace_lock)); 864 (void) refcount_add(&spa->spa_refcount, tag); 865} 866 867/* 868 * Remove a reference to the given spa_t. Must have at least one reference, or 869 * have the namespace lock held. 870 */ 871void 872spa_close(spa_t *spa, void *tag) 873{ 874 ASSERT(refcount_count(&spa->spa_refcount) > spa->spa_minref || 875 MUTEX_HELD(&spa_namespace_lock)); 876 (void) refcount_remove(&spa->spa_refcount, tag); 877} 878 879/* 880 * Remove a reference to the given spa_t held by a dsl dir that is 881 * being asynchronously released. Async releases occur from a taskq 882 * performing eviction of dsl datasets and dirs. The namespace lock 883 * isn't held and the hold by the object being evicted may contribute to 884 * spa_minref (e.g. dataset or directory released during pool export), 885 * so the asserts in spa_close() do not apply. 886 */ 887void 888spa_async_close(spa_t *spa, void *tag) 889{ 890 (void) refcount_remove(&spa->spa_refcount, tag); 891} 892 893/* 894 * Check to see if the spa refcount is zero. Must be called with 895 * spa_namespace_lock held. We really compare against spa_minref, which is the 896 * number of references acquired when opening a pool 897 */ 898boolean_t 899spa_refcount_zero(spa_t *spa) 900{ 901 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 902 903 return (refcount_count(&spa->spa_refcount) == spa->spa_minref); 904} 905 906/* 907 * ========================================================================== 908 * SPA spare and l2cache tracking 909 * ========================================================================== 910 */ 911 912/* 913 * Hot spares and cache devices are tracked using the same code below, 914 * for 'auxiliary' devices. 915 */ 916 917typedef struct spa_aux { 918 uint64_t aux_guid; 919 uint64_t aux_pool; 920 avl_node_t aux_avl; 921 int aux_count; 922} spa_aux_t; 923 924static int 925spa_aux_compare(const void *a, const void *b) 926{ 927 const spa_aux_t *sa = a; 928 const spa_aux_t *sb = b; 929 930 if (sa->aux_guid < sb->aux_guid) 931 return (-1); 932 else if (sa->aux_guid > sb->aux_guid) 933 return (1); 934 else 935 return (0); 936} 937 938void 939spa_aux_add(vdev_t *vd, avl_tree_t *avl) 940{ 941 avl_index_t where; 942 spa_aux_t search; 943 spa_aux_t *aux; 944 945 search.aux_guid = vd->vdev_guid; 946 if ((aux = avl_find(avl, &search, &where)) != NULL) { 947 aux->aux_count++; 948 } else { 949 aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP); 950 aux->aux_guid = vd->vdev_guid; 951 aux->aux_count = 1; 952 avl_insert(avl, aux, where); 953 } 954} 955 956void 957spa_aux_remove(vdev_t *vd, avl_tree_t *avl) 958{ 959 spa_aux_t search; 960 spa_aux_t *aux; 961 avl_index_t where; 962 963 search.aux_guid = vd->vdev_guid; 964 aux = avl_find(avl, &search, &where); 965 966 ASSERT(aux != NULL); 967 968 if (--aux->aux_count == 0) { 969 avl_remove(avl, aux); 970 kmem_free(aux, sizeof (spa_aux_t)); 971 } else if (aux->aux_pool == spa_guid(vd->vdev_spa)) { 972 aux->aux_pool = 0ULL; 973 } 974} 975 976boolean_t 977spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl) 978{ 979 spa_aux_t search, *found; 980 981 search.aux_guid = guid; 982 found = avl_find(avl, &search, NULL); 983 984 if (pool) { 985 if (found) 986 *pool = found->aux_pool; 987 else 988 *pool = 0ULL; 989 } 990 991 if (refcnt) { 992 if (found) 993 *refcnt = found->aux_count; 994 else 995 *refcnt = 0; 996 } 997 998 return (found != NULL); 999} 1000 1001void 1002spa_aux_activate(vdev_t *vd, avl_tree_t *avl) 1003{ 1004 spa_aux_t search, *found; 1005 avl_index_t where; 1006 1007 search.aux_guid = vd->vdev_guid; 1008 found = avl_find(avl, &search, &where); 1009 ASSERT(found != NULL); 1010 ASSERT(found->aux_pool == 0ULL); 1011 1012 found->aux_pool = spa_guid(vd->vdev_spa); 1013} 1014 1015/* 1016 * Spares are tracked globally due to the following constraints: 1017 * 1018 * - A spare may be part of multiple pools. 1019 * - A spare may be added to a pool even if it's actively in use within 1020 * another pool. 1021 * - A spare in use in any pool can only be the source of a replacement if 1022 * the target is a spare in the same pool. 1023 * 1024 * We keep track of all spares on the system through the use of a reference 1025 * counted AVL tree. When a vdev is added as a spare, or used as a replacement 1026 * spare, then we bump the reference count in the AVL tree. In addition, we set 1027 * the 'vdev_isspare' member to indicate that the device is a spare (active or 1028 * inactive). When a spare is made active (used to replace a device in the 1029 * pool), we also keep track of which pool its been made a part of. 1030 * 1031 * The 'spa_spare_lock' protects the AVL tree. These functions are normally 1032 * called under the spa_namespace lock as part of vdev reconfiguration. The 1033 * separate spare lock exists for the status query path, which does not need to 1034 * be completely consistent with respect to other vdev configuration changes. 1035 */ 1036 1037static int 1038spa_spare_compare(const void *a, const void *b) 1039{ 1040 return (spa_aux_compare(a, b)); 1041} 1042 1043void 1044spa_spare_add(vdev_t *vd) 1045{ 1046 mutex_enter(&spa_spare_lock); 1047 ASSERT(!vd->vdev_isspare); 1048 spa_aux_add(vd, &spa_spare_avl); 1049 vd->vdev_isspare = B_TRUE; 1050 mutex_exit(&spa_spare_lock); 1051} 1052 1053void 1054spa_spare_remove(vdev_t *vd) 1055{ 1056 mutex_enter(&spa_spare_lock); 1057 ASSERT(vd->vdev_isspare); 1058 spa_aux_remove(vd, &spa_spare_avl); 1059 vd->vdev_isspare = B_FALSE; 1060 mutex_exit(&spa_spare_lock); 1061} 1062 1063boolean_t 1064spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt) 1065{ 1066 boolean_t found; 1067 1068 mutex_enter(&spa_spare_lock); 1069 found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl); 1070 mutex_exit(&spa_spare_lock); 1071 1072 return (found); 1073} 1074 1075void 1076spa_spare_activate(vdev_t *vd) 1077{ 1078 mutex_enter(&spa_spare_lock); 1079 ASSERT(vd->vdev_isspare); 1080 spa_aux_activate(vd, &spa_spare_avl); 1081 mutex_exit(&spa_spare_lock); 1082} 1083 1084/* 1085 * Level 2 ARC devices are tracked globally for the same reasons as spares. 1086 * Cache devices currently only support one pool per cache device, and so 1087 * for these devices the aux reference count is currently unused beyond 1. 1088 */ 1089 1090static int 1091spa_l2cache_compare(const void *a, const void *b) 1092{ 1093 return (spa_aux_compare(a, b)); 1094} 1095 1096void 1097spa_l2cache_add(vdev_t *vd) 1098{ 1099 mutex_enter(&spa_l2cache_lock); 1100 ASSERT(!vd->vdev_isl2cache); 1101 spa_aux_add(vd, &spa_l2cache_avl); 1102 vd->vdev_isl2cache = B_TRUE; 1103 mutex_exit(&spa_l2cache_lock); 1104} 1105 1106void 1107spa_l2cache_remove(vdev_t *vd) 1108{ 1109 mutex_enter(&spa_l2cache_lock); 1110 ASSERT(vd->vdev_isl2cache); 1111 spa_aux_remove(vd, &spa_l2cache_avl); 1112 vd->vdev_isl2cache = B_FALSE; 1113 mutex_exit(&spa_l2cache_lock); 1114} 1115 1116boolean_t 1117spa_l2cache_exists(uint64_t guid, uint64_t *pool) 1118{ 1119 boolean_t found; 1120 1121 mutex_enter(&spa_l2cache_lock); 1122 found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl); 1123 mutex_exit(&spa_l2cache_lock); 1124 1125 return (found); 1126} 1127 1128void 1129spa_l2cache_activate(vdev_t *vd) 1130{ 1131 mutex_enter(&spa_l2cache_lock); 1132 ASSERT(vd->vdev_isl2cache); 1133 spa_aux_activate(vd, &spa_l2cache_avl); 1134 mutex_exit(&spa_l2cache_lock); 1135} 1136 1137/* 1138 * ========================================================================== 1139 * SPA vdev locking 1140 * ========================================================================== 1141 */ 1142 1143/* 1144 * Lock the given spa_t for the purpose of adding or removing a vdev. 1145 * Grabs the global spa_namespace_lock plus the spa config lock for writing. 1146 * It returns the next transaction group for the spa_t. 1147 */ 1148uint64_t 1149spa_vdev_enter(spa_t *spa) 1150{ 1151 mutex_enter(&spa->spa_vdev_top_lock); 1152 mutex_enter(&spa_namespace_lock); 1153 return (spa_vdev_config_enter(spa)); 1154} 1155 1156/* 1157 * Internal implementation for spa_vdev_enter(). Used when a vdev 1158 * operation requires multiple syncs (i.e. removing a device) while 1159 * keeping the spa_namespace_lock held. 1160 */ 1161uint64_t 1162spa_vdev_config_enter(spa_t *spa) 1163{ 1164 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1165 1166 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER); 1167 1168 return (spa_last_synced_txg(spa) + 1); 1169} 1170 1171/* 1172 * Used in combination with spa_vdev_config_enter() to allow the syncing 1173 * of multiple transactions without releasing the spa_namespace_lock. 1174 */ 1175void 1176spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag) 1177{ 1178 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1179 1180 int config_changed = B_FALSE; 1181 1182 ASSERT(txg > spa_last_synced_txg(spa)); 1183 1184 spa->spa_pending_vdev = NULL; 1185 1186 /* 1187 * Reassess the DTLs. 1188 */ 1189 vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE); 1190 1191 if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) { 1192 config_changed = B_TRUE; 1193 spa->spa_config_generation++; 1194 } 1195 1196 /* 1197 * Verify the metaslab classes. 1198 */ 1199 ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0); 1200 ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0); 1201 1202 spa_config_exit(spa, SCL_ALL, spa); 1203 1204 /* 1205 * Panic the system if the specified tag requires it. This 1206 * is useful for ensuring that configurations are updated 1207 * transactionally. 1208 */ 1209 if (zio_injection_enabled) 1210 zio_handle_panic_injection(spa, tag, 0); 1211 1212 /* 1213 * Note: this txg_wait_synced() is important because it ensures 1214 * that there won't be more than one config change per txg. 1215 * This allows us to use the txg as the generation number. 1216 */ 1217 if (error == 0) 1218 txg_wait_synced(spa->spa_dsl_pool, txg); 1219 1220 if (vd != NULL) { 1221 ASSERT(!vd->vdev_detached || vd->vdev_dtl_sm == NULL); 1222 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER); 1223 vdev_free(vd); 1224 spa_config_exit(spa, SCL_ALL, spa); 1225 } 1226 1227 /* 1228 * If the config changed, update the config cache. 1229 */ 1230 if (config_changed) 1231 spa_config_sync(spa, B_FALSE, B_TRUE); 1232} 1233 1234/* 1235 * Unlock the spa_t after adding or removing a vdev. Besides undoing the 1236 * locking of spa_vdev_enter(), we also want make sure the transactions have 1237 * synced to disk, and then update the global configuration cache with the new 1238 * information. 1239 */ 1240int 1241spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error) 1242{ 1243 spa_vdev_config_exit(spa, vd, txg, error, FTAG); 1244 mutex_exit(&spa_namespace_lock); 1245 mutex_exit(&spa->spa_vdev_top_lock); 1246 1247 return (error); 1248} 1249 1250/* 1251 * Lock the given spa_t for the purpose of changing vdev state. 1252 */ 1253void 1254spa_vdev_state_enter(spa_t *spa, int oplocks) 1255{ 1256 int locks = SCL_STATE_ALL | oplocks; 1257 1258 /* 1259 * Root pools may need to read of the underlying devfs filesystem 1260 * when opening up a vdev. Unfortunately if we're holding the 1261 * SCL_ZIO lock it will result in a deadlock when we try to issue 1262 * the read from the root filesystem. Instead we "prefetch" 1263 * the associated vnodes that we need prior to opening the 1264 * underlying devices and cache them so that we can prevent 1265 * any I/O when we are doing the actual open. 1266 */ 1267 if (spa_is_root(spa)) { 1268 int low = locks & ~(SCL_ZIO - 1); 1269 int high = locks & ~low; 1270 1271 spa_config_enter(spa, high, spa, RW_WRITER); 1272 vdev_hold(spa->spa_root_vdev); 1273 spa_config_enter(spa, low, spa, RW_WRITER); 1274 } else { 1275 spa_config_enter(spa, locks, spa, RW_WRITER); 1276 } 1277 spa->spa_vdev_locks = locks; 1278} 1279 1280int 1281spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error) 1282{ 1283 boolean_t config_changed = B_FALSE; 1284 1285 if (vd != NULL || error == 0) 1286 vdev_dtl_reassess(vd ? vd->vdev_top : spa->spa_root_vdev, 1287 0, 0, B_FALSE); 1288 1289 if (vd != NULL) { 1290 vdev_state_dirty(vd->vdev_top); 1291 config_changed = B_TRUE; 1292 spa->spa_config_generation++; 1293 } 1294 1295 if (spa_is_root(spa)) 1296 vdev_rele(spa->spa_root_vdev); 1297 1298 ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL); 1299 spa_config_exit(spa, spa->spa_vdev_locks, spa); 1300 1301 /* 1302 * If anything changed, wait for it to sync. This ensures that, 1303 * from the system administrator's perspective, zpool(1M) commands 1304 * are synchronous. This is important for things like zpool offline: 1305 * when the command completes, you expect no further I/O from ZFS. 1306 */ 1307 if (vd != NULL) 1308 txg_wait_synced(spa->spa_dsl_pool, 0); 1309 1310 /* 1311 * If the config changed, update the config cache. 1312 */ 1313 if (config_changed) { 1314 mutex_enter(&spa_namespace_lock); 1315 spa_config_sync(spa, B_FALSE, B_TRUE); 1316 mutex_exit(&spa_namespace_lock); 1317 } 1318 1319 return (error); 1320} 1321 1322/* 1323 * ========================================================================== 1324 * Miscellaneous functions 1325 * ========================================================================== 1326 */ 1327 1328void 1329spa_activate_mos_feature(spa_t *spa, const char *feature, dmu_tx_t *tx) 1330{ 1331 if (!nvlist_exists(spa->spa_label_features, feature)) { 1332 fnvlist_add_boolean(spa->spa_label_features, feature); 1333 /* 1334 * When we are creating the pool (tx_txg==TXG_INITIAL), we can't 1335 * dirty the vdev config because lock SCL_CONFIG is not held. 1336 * Thankfully, in this case we don't need to dirty the config 1337 * because it will be written out anyway when we finish 1338 * creating the pool. 1339 */ 1340 if (tx->tx_txg != TXG_INITIAL) 1341 vdev_config_dirty(spa->spa_root_vdev); 1342 } 1343} 1344 1345void 1346spa_deactivate_mos_feature(spa_t *spa, const char *feature) 1347{ 1348 if (nvlist_remove_all(spa->spa_label_features, feature) == 0) 1349 vdev_config_dirty(spa->spa_root_vdev); 1350} 1351 1352/* 1353 * Rename a spa_t. 1354 */ 1355int 1356spa_rename(const char *name, const char *newname) 1357{ 1358 spa_t *spa; 1359 int err; 1360 1361 /* 1362 * Lookup the spa_t and grab the config lock for writing. We need to 1363 * actually open the pool so that we can sync out the necessary labels. 1364 * It's OK to call spa_open() with the namespace lock held because we 1365 * allow recursive calls for other reasons. 1366 */ 1367 mutex_enter(&spa_namespace_lock); 1368 if ((err = spa_open(name, &spa, FTAG)) != 0) { 1369 mutex_exit(&spa_namespace_lock); 1370 return (err); 1371 } 1372 1373 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1374 1375 avl_remove(&spa_namespace_avl, spa); 1376 (void) strlcpy(spa->spa_name, newname, sizeof (spa->spa_name)); 1377 avl_add(&spa_namespace_avl, spa); 1378 1379 /* 1380 * Sync all labels to disk with the new names by marking the root vdev 1381 * dirty and waiting for it to sync. It will pick up the new pool name 1382 * during the sync. 1383 */ 1384 vdev_config_dirty(spa->spa_root_vdev); 1385 1386 spa_config_exit(spa, SCL_ALL, FTAG); 1387 1388 txg_wait_synced(spa->spa_dsl_pool, 0); 1389 1390 /* 1391 * Sync the updated config cache. 1392 */ 1393 spa_config_sync(spa, B_FALSE, B_TRUE); 1394 1395 spa_close(spa, FTAG); 1396 1397 mutex_exit(&spa_namespace_lock); 1398 1399 return (0); 1400} 1401 1402/* 1403 * Return the spa_t associated with given pool_guid, if it exists. If 1404 * device_guid is non-zero, determine whether the pool exists *and* contains 1405 * a device with the specified device_guid. 1406 */ 1407spa_t * 1408spa_by_guid(uint64_t pool_guid, uint64_t device_guid) 1409{ 1410 spa_t *spa; 1411 avl_tree_t *t = &spa_namespace_avl; 1412 1413 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1414 1415 for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) { 1416 if (spa->spa_state == POOL_STATE_UNINITIALIZED) 1417 continue; 1418 if (spa->spa_root_vdev == NULL) 1419 continue; 1420 if (spa_guid(spa) == pool_guid) { 1421 if (device_guid == 0) 1422 break; 1423 1424 if (vdev_lookup_by_guid(spa->spa_root_vdev, 1425 device_guid) != NULL) 1426 break; 1427 1428 /* 1429 * Check any devices we may be in the process of adding. 1430 */ 1431 if (spa->spa_pending_vdev) { 1432 if (vdev_lookup_by_guid(spa->spa_pending_vdev, 1433 device_guid) != NULL) 1434 break; 1435 } 1436 } 1437 } 1438 1439 return (spa); 1440} 1441 1442/* 1443 * Determine whether a pool with the given pool_guid exists. 1444 */ 1445boolean_t 1446spa_guid_exists(uint64_t pool_guid, uint64_t device_guid) 1447{ 1448 return (spa_by_guid(pool_guid, device_guid) != NULL); 1449} 1450 1451char * 1452spa_strdup(const char *s) 1453{ 1454 size_t len; 1455 char *new; 1456 1457 len = strlen(s); 1458 new = kmem_alloc(len + 1, KM_SLEEP); 1459 bcopy(s, new, len); 1460 new[len] = '\0'; 1461 1462 return (new); 1463} 1464 1465void 1466spa_strfree(char *s) 1467{ 1468 kmem_free(s, strlen(s) + 1); 1469} 1470 1471uint64_t 1472spa_get_random(uint64_t range) 1473{ 1474 uint64_t r; 1475 1476 ASSERT(range != 0); 1477 1478 (void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t)); 1479 1480 return (r % range); 1481} 1482 1483uint64_t 1484spa_generate_guid(spa_t *spa) 1485{ 1486 uint64_t guid = spa_get_random(-1ULL); 1487 1488 if (spa != NULL) { 1489 while (guid == 0 || spa_guid_exists(spa_guid(spa), guid)) 1490 guid = spa_get_random(-1ULL); 1491 } else { 1492 while (guid == 0 || spa_guid_exists(guid, 0)) 1493 guid = spa_get_random(-1ULL); 1494 } 1495 1496 return (guid); 1497} 1498 1499void 1500snprintf_blkptr(char *buf, size_t buflen, const blkptr_t *bp) 1501{ 1502 char type[256]; 1503 char *checksum = NULL; 1504 char *compress = NULL; 1505 1506 if (bp != NULL) { 1507 if (BP_GET_TYPE(bp) & DMU_OT_NEWTYPE) { 1508 dmu_object_byteswap_t bswap = 1509 DMU_OT_BYTESWAP(BP_GET_TYPE(bp)); 1510 (void) snprintf(type, sizeof (type), "bswap %s %s", 1511 DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) ? 1512 "metadata" : "data", 1513 dmu_ot_byteswap[bswap].ob_name); 1514 } else { 1515 (void) strlcpy(type, dmu_ot[BP_GET_TYPE(bp)].ot_name, 1516 sizeof (type)); 1517 } 1518 if (!BP_IS_EMBEDDED(bp)) { 1519 checksum = 1520 zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name; 1521 } 1522 compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name; 1523 } 1524 1525 SNPRINTF_BLKPTR(snprintf, ' ', buf, buflen, bp, type, checksum, 1526 compress); 1527} 1528 1529void 1530spa_freeze(spa_t *spa) 1531{ 1532 uint64_t freeze_txg = 0; 1533 1534 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1535 if (spa->spa_freeze_txg == UINT64_MAX) { 1536 freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE; 1537 spa->spa_freeze_txg = freeze_txg; 1538 } 1539 spa_config_exit(spa, SCL_ALL, FTAG); 1540 if (freeze_txg != 0) 1541 txg_wait_synced(spa_get_dsl(spa), freeze_txg); 1542} 1543 1544void 1545zfs_panic_recover(const char *fmt, ...) 1546{ 1547 va_list adx; 1548 1549 va_start(adx, fmt); 1550 vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx); 1551 va_end(adx); 1552} 1553 1554/* 1555 * This is a stripped-down version of strtoull, suitable only for converting 1556 * lowercase hexadecimal numbers that don't overflow. 1557 */ 1558uint64_t 1559zfs_strtonum(const char *str, char **nptr) 1560{ 1561 uint64_t val = 0; 1562 char c; 1563 int digit; 1564 1565 while ((c = *str) != '\0') { 1566 if (c >= '0' && c <= '9') 1567 digit = c - '0'; 1568 else if (c >= 'a' && c <= 'f') 1569 digit = 10 + c - 'a'; 1570 else 1571 break; 1572 1573 val *= 16; 1574 val += digit; 1575 1576 str++; 1577 } 1578 1579 if (nptr) 1580 *nptr = (char *)str; 1581 1582 return (val); 1583} 1584 1585/* 1586 * ========================================================================== 1587 * Accessor functions 1588 * ========================================================================== 1589 */ 1590 1591boolean_t 1592spa_shutting_down(spa_t *spa) 1593{ 1594 return (spa->spa_async_suspended); 1595} 1596 1597dsl_pool_t * 1598spa_get_dsl(spa_t *spa) 1599{ 1600 return (spa->spa_dsl_pool); 1601} 1602 1603boolean_t 1604spa_is_initializing(spa_t *spa) 1605{ 1606 return (spa->spa_is_initializing); 1607} 1608 1609blkptr_t * 1610spa_get_rootblkptr(spa_t *spa) 1611{ 1612 return (&spa->spa_ubsync.ub_rootbp); 1613} 1614 1615void 1616spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp) 1617{ 1618 spa->spa_uberblock.ub_rootbp = *bp; 1619} 1620 1621void 1622spa_altroot(spa_t *spa, char *buf, size_t buflen) 1623{ 1624 if (spa->spa_root == NULL) 1625 buf[0] = '\0'; 1626 else 1627 (void) strncpy(buf, spa->spa_root, buflen); 1628} 1629 1630int 1631spa_sync_pass(spa_t *spa) 1632{ 1633 return (spa->spa_sync_pass); 1634} 1635 1636char * 1637spa_name(spa_t *spa) 1638{ 1639 return (spa->spa_name); 1640} 1641 1642uint64_t 1643spa_guid(spa_t *spa) 1644{ 1645 dsl_pool_t *dp = spa_get_dsl(spa); 1646 uint64_t guid; 1647 1648 /* 1649 * If we fail to parse the config during spa_load(), we can go through 1650 * the error path (which posts an ereport) and end up here with no root 1651 * vdev. We stash the original pool guid in 'spa_config_guid' to handle 1652 * this case. 1653 */ 1654 if (spa->spa_root_vdev == NULL) 1655 return (spa->spa_config_guid); 1656 1657 guid = spa->spa_last_synced_guid != 0 ? 1658 spa->spa_last_synced_guid : spa->spa_root_vdev->vdev_guid; 1659 1660 /* 1661 * Return the most recently synced out guid unless we're 1662 * in syncing context. 1663 */ 1664 if (dp && dsl_pool_sync_context(dp)) 1665 return (spa->spa_root_vdev->vdev_guid); 1666 else 1667 return (guid); 1668} 1669 1670uint64_t 1671spa_load_guid(spa_t *spa) 1672{ 1673 /* 1674 * This is a GUID that exists solely as a reference for the 1675 * purposes of the arc. It is generated at load time, and 1676 * is never written to persistent storage. 1677 */ 1678 return (spa->spa_load_guid); 1679} 1680 1681uint64_t 1682spa_last_synced_txg(spa_t *spa) 1683{ 1684 return (spa->spa_ubsync.ub_txg); 1685} 1686 1687uint64_t 1688spa_first_txg(spa_t *spa) 1689{ 1690 return (spa->spa_first_txg); 1691} 1692 1693uint64_t 1694spa_syncing_txg(spa_t *spa) 1695{ 1696 return (spa->spa_syncing_txg); 1697} 1698 1699pool_state_t 1700spa_state(spa_t *spa) 1701{ 1702 return (spa->spa_state); 1703} 1704 1705spa_load_state_t 1706spa_load_state(spa_t *spa) 1707{ 1708 return (spa->spa_load_state); 1709} 1710 1711uint64_t 1712spa_freeze_txg(spa_t *spa) 1713{ 1714 return (spa->spa_freeze_txg); 1715} 1716 1717/* ARGSUSED */ 1718uint64_t 1719spa_get_asize(spa_t *spa, uint64_t lsize) 1720{ 1721 return (lsize * spa_asize_inflation); 1722} 1723 1724/* 1725 * Return the amount of slop space in bytes. It is 1/32 of the pool (3.2%), 1726 * or at least 32MB. 1727 * 1728 * See the comment above spa_slop_shift for details. 1729 */ 1730uint64_t 1731spa_get_slop_space(spa_t *spa) { 1732 uint64_t space = spa_get_dspace(spa); 1733 return (MAX(space >> spa_slop_shift, SPA_MINDEVSIZE >> 1)); 1734} 1735 1736uint64_t 1737spa_get_dspace(spa_t *spa) 1738{ 1739 return (spa->spa_dspace); 1740} 1741 1742void 1743spa_update_dspace(spa_t *spa) 1744{ 1745 spa->spa_dspace = metaslab_class_get_dspace(spa_normal_class(spa)) + 1746 ddt_get_dedup_dspace(spa); 1747} 1748 1749/* 1750 * Return the failure mode that has been set to this pool. The default 1751 * behavior will be to block all I/Os when a complete failure occurs. 1752 */ 1753uint8_t 1754spa_get_failmode(spa_t *spa) 1755{ 1756 return (spa->spa_failmode); 1757} 1758 1759boolean_t 1760spa_suspended(spa_t *spa) 1761{ 1762 return (spa->spa_suspended); 1763} 1764 1765uint64_t 1766spa_version(spa_t *spa) 1767{ 1768 return (spa->spa_ubsync.ub_version); 1769} 1770 1771boolean_t 1772spa_deflate(spa_t *spa) 1773{ 1774 return (spa->spa_deflate); 1775} 1776 1777metaslab_class_t * 1778spa_normal_class(spa_t *spa) 1779{ 1780 return (spa->spa_normal_class); 1781} 1782 1783metaslab_class_t * 1784spa_log_class(spa_t *spa) 1785{ 1786 return (spa->spa_log_class); 1787} 1788 1789void 1790spa_evicting_os_register(spa_t *spa, objset_t *os) 1791{ 1792 mutex_enter(&spa->spa_evicting_os_lock); 1793 list_insert_head(&spa->spa_evicting_os_list, os); 1794 mutex_exit(&spa->spa_evicting_os_lock); 1795} 1796 1797void 1798spa_evicting_os_deregister(spa_t *spa, objset_t *os) 1799{ 1800 mutex_enter(&spa->spa_evicting_os_lock); 1801 list_remove(&spa->spa_evicting_os_list, os); 1802 cv_broadcast(&spa->spa_evicting_os_cv); 1803 mutex_exit(&spa->spa_evicting_os_lock); 1804} 1805 1806void 1807spa_evicting_os_wait(spa_t *spa) 1808{ 1809 mutex_enter(&spa->spa_evicting_os_lock); 1810 while (!list_is_empty(&spa->spa_evicting_os_list)) 1811 cv_wait(&spa->spa_evicting_os_cv, &spa->spa_evicting_os_lock); 1812 mutex_exit(&spa->spa_evicting_os_lock); 1813 1814 dmu_buf_user_evict_wait(); 1815} 1816 1817int 1818spa_max_replication(spa_t *spa) 1819{ 1820 /* 1821 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to 1822 * handle BPs with more than one DVA allocated. Set our max 1823 * replication level accordingly. 1824 */ 1825 if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS) 1826 return (1); 1827 return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override)); 1828} 1829 1830int 1831spa_prev_software_version(spa_t *spa) 1832{ 1833 return (spa->spa_prev_software_version); 1834} 1835 1836uint64_t 1837spa_deadman_synctime(spa_t *spa) 1838{ 1839 return (spa->spa_deadman_synctime); 1840} 1841 1842uint64_t 1843dva_get_dsize_sync(spa_t *spa, const dva_t *dva) 1844{ 1845 uint64_t asize = DVA_GET_ASIZE(dva); 1846 uint64_t dsize = asize; 1847 1848 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 1849 1850 if (asize != 0 && spa->spa_deflate) { 1851 uint64_t vdev = DVA_GET_VDEV(dva); 1852 vdev_t *vd = vdev_lookup_top(spa, vdev); 1853 if (vd == NULL) { 1854 panic( 1855 "dva_get_dsize_sync(): bad DVA %llu:%llu", 1856 (u_longlong_t)vdev, (u_longlong_t)asize); 1857 } 1858 dsize = (asize >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio; 1859 } 1860 1861 return (dsize); 1862} 1863 1864uint64_t 1865bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp) 1866{ 1867 uint64_t dsize = 0; 1868 1869 for (int d = 0; d < BP_GET_NDVAS(bp); d++) 1870 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]); 1871 1872 return (dsize); 1873} 1874 1875uint64_t 1876bp_get_dsize(spa_t *spa, const blkptr_t *bp) 1877{ 1878 uint64_t dsize = 0; 1879 1880 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); 1881 1882 for (int d = 0; d < BP_GET_NDVAS(bp); d++) 1883 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]); 1884 1885 spa_config_exit(spa, SCL_VDEV, FTAG); 1886 1887 return (dsize); 1888} 1889 1890/* 1891 * ========================================================================== 1892 * Initialization and Termination 1893 * ========================================================================== 1894 */ 1895 1896static int 1897spa_name_compare(const void *a1, const void *a2) 1898{ 1899 const spa_t *s1 = a1; 1900 const spa_t *s2 = a2; 1901 int s; 1902 1903 s = strcmp(s1->spa_name, s2->spa_name); 1904 if (s > 0) 1905 return (1); 1906 if (s < 0) 1907 return (-1); 1908 return (0); 1909} 1910 1911int 1912spa_busy(void) 1913{ 1914 return (spa_active_count); 1915} 1916 1917void 1918spa_boot_init() 1919{ 1920 spa_config_load(); 1921} 1922 1923#ifdef _KERNEL 1924EVENTHANDLER_DEFINE(mountroot, spa_boot_init, NULL, 0); 1925#endif 1926 1927void 1928spa_init(int mode) 1929{ 1930 mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL); 1931 mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL); 1932 mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL); 1933 cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL); 1934 1935 avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t), 1936 offsetof(spa_t, spa_avl)); 1937 1938 avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t), 1939 offsetof(spa_aux_t, aux_avl)); 1940 1941 avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t), 1942 offsetof(spa_aux_t, aux_avl)); 1943 1944 spa_mode_global = mode; 1945 1946#ifdef illumos 1947#ifdef _KERNEL 1948 spa_arch_init(); 1949#else 1950 if (spa_mode_global != FREAD && dprintf_find_string("watch")) { 1951 arc_procfd = open("/proc/self/ctl", O_WRONLY); 1952 if (arc_procfd == -1) { 1953 perror("could not enable watchpoints: " 1954 "opening /proc/self/ctl failed: "); 1955 } else { 1956 arc_watch = B_TRUE; 1957 } 1958 } 1959#endif 1960#endif /* illumos */ 1961 refcount_sysinit(); 1962 unique_init(); 1963 range_tree_init(); 1964 zio_init(); 1965 lz4_init(); 1966 dmu_init(); 1967 zil_init(); 1968 vdev_cache_stat_init(); 1969 zfs_prop_init(); 1970 zpool_prop_init(); 1971 zpool_feature_init(); 1972 spa_config_load(); 1973 l2arc_start(); 1974#ifndef illumos 1975#ifdef _KERNEL 1976 zfs_deadman_init(); 1977#endif 1978#endif /* !illumos */ 1979} 1980 1981void 1982spa_fini(void) 1983{ 1984 l2arc_stop(); 1985 1986 spa_evict_all(); 1987 1988 vdev_cache_stat_fini(); 1989 zil_fini(); 1990 dmu_fini(); 1991 lz4_fini(); 1992 zio_fini(); 1993 range_tree_fini(); 1994 unique_fini(); 1995 refcount_fini(); 1996 1997 avl_destroy(&spa_namespace_avl); 1998 avl_destroy(&spa_spare_avl); 1999 avl_destroy(&spa_l2cache_avl); 2000 2001 cv_destroy(&spa_namespace_cv); 2002 mutex_destroy(&spa_namespace_lock); 2003 mutex_destroy(&spa_spare_lock); 2004 mutex_destroy(&spa_l2cache_lock); 2005} 2006 2007/* 2008 * Return whether this pool has slogs. No locking needed. 2009 * It's not a problem if the wrong answer is returned as it's only for 2010 * performance and not correctness 2011 */ 2012boolean_t 2013spa_has_slogs(spa_t *spa) 2014{ 2015 return (spa->spa_log_class->mc_rotor != NULL); 2016} 2017 2018spa_log_state_t 2019spa_get_log_state(spa_t *spa) 2020{ 2021 return (spa->spa_log_state); 2022} 2023 2024void 2025spa_set_log_state(spa_t *spa, spa_log_state_t state) 2026{ 2027 spa->spa_log_state = state; 2028} 2029 2030boolean_t 2031spa_is_root(spa_t *spa) 2032{ 2033 return (spa->spa_is_root); 2034} 2035 2036boolean_t 2037spa_writeable(spa_t *spa) 2038{ 2039 return (!!(spa->spa_mode & FWRITE)); 2040} 2041 2042/* 2043 * Returns true if there is a pending sync task in any of the current 2044 * syncing txg, the current quiescing txg, or the current open txg. 2045 */ 2046boolean_t 2047spa_has_pending_synctask(spa_t *spa) 2048{ 2049 return (!txg_all_lists_empty(&spa->spa_dsl_pool->dp_sync_tasks)); 2050} 2051 2052int 2053spa_mode(spa_t *spa) 2054{ 2055 return (spa->spa_mode); 2056} 2057 2058uint64_t 2059spa_bootfs(spa_t *spa) 2060{ 2061 return (spa->spa_bootfs); 2062} 2063 2064uint64_t 2065spa_delegation(spa_t *spa) 2066{ 2067 return (spa->spa_delegation); 2068} 2069 2070objset_t * 2071spa_meta_objset(spa_t *spa) 2072{ 2073 return (spa->spa_meta_objset); 2074} 2075 2076enum zio_checksum 2077spa_dedup_checksum(spa_t *spa) 2078{ 2079 return (spa->spa_dedup_checksum); 2080} 2081 2082/* 2083 * Reset pool scan stat per scan pass (or reboot). 2084 */ 2085void 2086spa_scan_stat_init(spa_t *spa) 2087{ 2088 /* data not stored on disk */ 2089 spa->spa_scan_pass_start = gethrestime_sec(); 2090 spa->spa_scan_pass_exam = 0; 2091 vdev_scan_stat_init(spa->spa_root_vdev); 2092} 2093 2094/* 2095 * Get scan stats for zpool status reports 2096 */ 2097int 2098spa_scan_get_stats(spa_t *spa, pool_scan_stat_t *ps) 2099{ 2100 dsl_scan_t *scn = spa->spa_dsl_pool ? spa->spa_dsl_pool->dp_scan : NULL; 2101 2102 if (scn == NULL || scn->scn_phys.scn_func == POOL_SCAN_NONE) 2103 return (SET_ERROR(ENOENT)); 2104 bzero(ps, sizeof (pool_scan_stat_t)); 2105 2106 /* data stored on disk */ 2107 ps->pss_func = scn->scn_phys.scn_func; 2108 ps->pss_start_time = scn->scn_phys.scn_start_time; 2109 ps->pss_end_time = scn->scn_phys.scn_end_time; 2110 ps->pss_to_examine = scn->scn_phys.scn_to_examine; 2111 ps->pss_examined = scn->scn_phys.scn_examined; 2112 ps->pss_to_process = scn->scn_phys.scn_to_process; 2113 ps->pss_processed = scn->scn_phys.scn_processed; 2114 ps->pss_errors = scn->scn_phys.scn_errors; 2115 ps->pss_state = scn->scn_phys.scn_state; 2116 2117 /* data not stored on disk */ 2118 ps->pss_pass_start = spa->spa_scan_pass_start; 2119 ps->pss_pass_exam = spa->spa_scan_pass_exam; 2120 2121 return (0); 2122} 2123 2124boolean_t 2125spa_debug_enabled(spa_t *spa) 2126{ 2127 return (spa->spa_debug); 2128} 2129 2130int 2131spa_maxblocksize(spa_t *spa) 2132{ 2133 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) 2134 return (SPA_MAXBLOCKSIZE); 2135 else 2136 return (SPA_OLD_MAXBLOCKSIZE); 2137} 2138